commit 2b9661931dfa16dc193102401ce85940322b621d
author William Hesse <whesse@google.com> Mon Aug 08 10:14:47 2016 +0200
committer William Hesse <whesse@google.com> Mon Aug 08 10:14:47 2016 +0200
tree d3463056608337cfe5d710489ca45e3b6ac835a7
parent 12707a48a9d4dd665f189b8bde18ac956b27cd2d
parent 36f52dd43f1111cbc5b51796ab51d8136bac6b9f

Version 1.19.0-dev.4.0

Merge 36f52dd43f1111cbc5b51796ab51d8136bac6b9f into dev

CHANGELOG.md

diff --git a/CHANGELOG.md b/CHANGELOG.md
index 3c62bf5..64a30df 100644
--- a/CHANGELOG.md
+++ b/CHANGELOG.md
@@ -1,5 +1,66 @@
 ## 1.19.0
 
+### Core library changes
+
+* `dart:io`
+  * `Socket.connect` with source-address argument is now non-blocking
+    on Mac. Was already non-blocking on all other platforms.
+  * Report a better error when a bind fails because of a bad source address.
+
+### Strong Mode
+
+*   New feature - an option to disable implicit casts
+    (SDK issue [26583](https://github.com/dart-lang/sdk/issues/26583)),
+    see the [documentation](https://github.com/dart-lang/dev_compiler/blob/master/doc/STATIC_SAFETY.md#disable-implicit-casts)
+    for usage instructions and examples.
+
+*   New feature - an option to disable implicit dynamic
+    (SDK issue [25573](https://github.com/dart-lang/sdk/issues/25573)),
+    see the [documentation](https://github.com/dart-lang/dev_compiler/blob/master/doc/STATIC_SAFETY.md#disable-implicit-dynamic)
+    for usage instructions and examples.
+
+*   Breaking change - infer generic type arguments from the
+    constructor invocation arguments
+    (SDK issue [25220](https://github.com/dart-lang/sdk/issues/25220))
+
+    ```dart
+    var map = new Map<String, String>();
+
+    // infer: Map<String, String>
+    var otherMap = new Map.from(map);
+    ```
+
+*   Breaking change - infer local function return type
+    (SDK issue [26414](https://github.com/dart-lang/sdk/issues/26414))
+
+    ```dart
+    void main() {
+      // infer: return type is int
+      f() { return 40; }
+      int y = f() + 2; // type checks
+      print(y);
+    }
+    ```
+
+*   Breaking change - allow type promotion from a generic type parameter
+    (SDK issue [26414](https://github.com/dart-lang/sdk/issues/26965))
+
+    ```dart
+    void fn/*<T>*/(/*=T*/ object) {
+      if (object is String) {
+        // Treat `object` as `String` inside this block.
+        // But it will require a cast to pass it to something that expects `T`.
+        print(object.substring(1));
+      }
+    }
+    ```
+
+### Dart VM
+
+*   The dependency on BoringSSL has been rolled forward. Going forward, builds
+    of the Dart VM including secure sockets will require a compiler with C11
+    support, and to link against glib 2.16 or newer.
+
 ### Tool Changes
 
 * `dartfmt` - upgraded to v0.2.9
@@ -39,6 +100,9 @@
   * Eliminate some false negatives when determining whether global executables
     are on the user's executable path.
 
+* dart2dart (aka `dart2js --output-type=dart`) has been removed (this was
+  deprecated in Dart 1.11)
+
 [Flutter]: https://flutter.io/
 
 ## 1.18.1 - 2016-08-02

DEPS

diff --git a/DEPS b/DEPS
index 10ac7b8..f53dfc9 100644
--- a/DEPS
+++ b/DEPS
@@ -43,7 +43,7 @@
   "barback_tag" : "@0.15.2+7",
   "bazel_worker_tag": "@v0.1.0",
   "boolean_selector_tag" : "@1.0.0",
-  "boringssl_rev" : "@daeafc22c66ad48f6b32fc8d3362eb9ba31b774e",
+  "boringssl_rev" : "@8d343b44bbab829d1a28fdef650ca95f7db4412e",
   "charcode_tag": "@1.1.0",
   "chrome_rev" : "@19997",
   "cli_util_tag" : "@0.0.1+2",
@@ -56,7 +56,7 @@
   "dart_services_rev" : "@7aea2574e6f3924bf409a80afb8ad52aa2be4f97",
   "dart_style_tag": "@0.2.9",
   "dartdoc_tag" : "@v0.9.6+2",
-  "dev_compiler_rev": "@9cc982b90f57c6132f648b0041c0b03bd23fcbfc",
+  "dev_compiler_rev": "@fa084164b620ea75cd2008c9dc317655a045ad6d",
   "fixnum_tag": "@0.10.5",
   "func_rev": "@8d4aea75c21be2179cb00dc2b94a71414653094e",
   "glob_rev": "@704cf75e4f26b417505c5c611bdaacd8808467dd",

create_sdk.gyp

diff --git a/create_sdk.gyp b/create_sdk.gyp
index 83ddc47..b1b9361 100644
--- a/create_sdk.gyp
+++ b/create_sdk.gyp
@@ -26,7 +26,6 @@
             '<!@(["python", "tools/list_files.py",'
                 '"dart$",'
                 '"sdk/lib"])',
-            'sdk/lib/dart2dart.platform',
             'sdk/lib/dart_client.platform',
             'sdk/lib/dart_server.platform',
             'sdk/lib/dart_shared.platform',

pkg/analysis_server/lib/src/analysis_server.dart

diff --git a/pkg/analysis_server/lib/src/analysis_server.dart b/pkg/analysis_server/lib/src/analysis_server.dart
index ca87546..f72e512 100644
--- a/pkg/analysis_server/lib/src/analysis_server.dart
+++ b/pkg/analysis_server/lib/src/analysis_server.dart
@@ -29,9 +29,10 @@
 import 'package:analyzer/file_system/file_system.dart';
 import 'package:analyzer/instrumentation/instrumentation.dart';
 import 'package:analyzer/plugin/resolver_provider.dart';
-import 'package:analyzer/source/embedder.dart';
 import 'package:analyzer/source/pub_package_map_provider.dart';
+import 'package:analyzer/src/context/builder.dart';
 import 'package:analyzer/src/dart/ast/utilities.dart';
+import 'package:analyzer/src/dart/sdk/sdk.dart';
 import 'package:analyzer/src/generated/engine.dart';
 import 'package:analyzer/src/generated/java_engine.dart';
 import 'package:analyzer/src/generated/java_io.dart';
@@ -1661,7 +1662,7 @@
     EmbedderYamlLocator locator =
         disposition.getEmbedderLocator(resourceProvider);
     Map<Folder, YamlMap> embedderYamls = locator.embedderYamls;
-    EmbedderSdk embedderSdk = new EmbedderSdk(embedderYamls);
+    EmbedderSdk embedderSdk = new EmbedderSdk(resourceProvider, embedderYamls);
     if (embedderSdk.libraryMap.size() == 0) {
       // There was no embedder file, or the file was empty, so used the default
       // SDK.

pkg/analysis_server/lib/src/context_manager.dart

diff --git a/pkg/analysis_server/lib/src/context_manager.dart b/pkg/analysis_server/lib/src/context_manager.dart
index 347788b..86b13a1 100644
--- a/pkg/analysis_server/lib/src/context_manager.dart
+++ b/pkg/analysis_server/lib/src/context_manager.dart
@@ -16,14 +16,15 @@
 import 'package:analyzer/plugin/resolver_provider.dart';
 import 'package:analyzer/source/analysis_options_provider.dart';
 import 'package:analyzer/source/config.dart';
-import 'package:analyzer/source/embedder.dart';
 import 'package:analyzer/source/package_map_provider.dart';
 import 'package:analyzer/source/package_map_resolver.dart';
 import 'package:analyzer/source/path_filter.dart';
 import 'package:analyzer/source/pub_package_map_provider.dart';
 import 'package:analyzer/source/sdk_ext.dart';
+import 'package:analyzer/src/context/builder.dart';
 import 'package:analyzer/src/context/context.dart' as context;
 import 'package:analyzer/src/context/source.dart';
+import 'package:analyzer/src/dart/sdk/sdk.dart';
 import 'package:analyzer/src/generated/engine.dart';
 import 'package:analyzer/src/generated/java_engine.dart';
 import 'package:analyzer/src/generated/java_io.dart';
@@ -391,6 +392,13 @@
   static const String PACKAGE_SPEC_NAME = '.packages';
 
   /**
+   * The name of the key in an embedder file whose value is the list of
+   * libraries in the SDK.
+   * TODO(brianwilkerson) This is also defined in sdk.dart.
+   */
+  static const String _EMBEDDED_LIB_MAP_KEY = 'embedded_libs';
+
+  /**
    * The [ResourceProvider] using which paths are converted into [Resource]s.
    */
   final ResourceProvider resourceProvider;
@@ -515,6 +523,7 @@
       contexts.add(info.context);
       info.children.forEach(addContextAndDescendants);
     }
+
     if (innermostContainingInfo != null) {
       if (analysisRoot == innermostContainingInfo.folder) {
         addContextAndDescendants(innermostContainingInfo);
@@ -529,6 +538,11 @@
     return contexts;
   }
 
+  /**
+   * Check if this map defines embedded libraries.
+   */
+  bool definesEmbeddedLibs(Map map) => map[_EMBEDDED_LIB_MAP_KEY] != null;
+
   @override
   AnalysisContext getContextFor(String path) {
     return _getInnermostContextInfoFor(path)?.context;
@@ -913,7 +927,8 @@
                 .where((r) => r is! DartUriResolver)
                 .toList();
             // Add an embedded URI resolver in its place.
-            resolvers.add(new DartUriResolver(new EmbedderSdk(embedderYamls)));
+            resolvers.add(new DartUriResolver(
+                new EmbedderSdk(resourceProvider, embedderYamls)));
 
             // Set a new source factory.
             SourceFactoryImpl newFactory = sourceFactory.clone();
@@ -1158,7 +1173,7 @@
     EmbedderYamlLocator locator =
         disposition.getEmbedderLocator(resourceProvider);
     Map<Folder, YamlMap> embedderYamls = locator.embedderYamls;
-    EmbedderSdk embedderSdk = new EmbedderSdk(embedderYamls);
+    EmbedderSdk embedderSdk = new EmbedderSdk(resourceProvider, embedderYamls);
     if (embedderSdk.libraryMap.size() == 0) {
       // There was no embedder file, or the file was empty, so used the default
       // SDK.

pkg/analysis_server/lib/src/status/get_handler.dart

diff --git a/pkg/analysis_server/lib/src/status/get_handler.dart b/pkg/analysis_server/lib/src/status/get_handler.dart
index d9486dd..11cccff 100644
--- a/pkg/analysis_server/lib/src/status/get_handler.dart
+++ b/pkg/analysis_server/lib/src/status/get_handler.dart
@@ -28,12 +28,12 @@
 import 'package:analyzer/dart/element/element.dart';
 import 'package:analyzer/dart/element/visitor.dart';
 import 'package:analyzer/file_system/file_system.dart';
-import 'package:analyzer/source/embedder.dart';
 import 'package:analyzer/source/error_processor.dart';
 import 'package:analyzer/source/sdk_ext.dart';
 import 'package:analyzer/src/context/cache.dart';
 import 'package:analyzer/src/context/context.dart' show AnalysisContextImpl;
 import 'package:analyzer/src/context/source.dart';
+import 'package:analyzer/src/dart/sdk/sdk.dart';
 import 'package:analyzer/src/generated/engine.dart';
 import 'package:analyzer/src/generated/error.dart';
 import 'package:analyzer/src/generated/java_engine.dart';
@@ -670,6 +670,7 @@
             _writeRow(buffer, [tag.elapsedMs, percentStr, tag.label],
                 classes: ["right", "right", null]);
           }
+
           tags.forEach(writeRow);
           buffer.write('</table>');
           //
@@ -684,6 +685,7 @@
             }
             counts[key] = count;
           }
+
           Set<AnalysisTarget> countedTargets = new HashSet<AnalysisTarget>();
           Map<String, int> sourceTypeCounts = new HashMap<String, int>();
           Map<String, int> typeCounts = new HashMap<String, int>();
@@ -751,6 +753,7 @@
                 }
               });
             }
+
             explicitSourceCount += explicitSources.length;
             explicitLineCount += lineCount(explicitSources, true);
             implicitSourceCount += implicitSources.length;
@@ -1399,6 +1402,7 @@
         buffer.write('</table></p>');
       }
     }
+
     void writeOptions(StringBuffer buffer, AnalysisOptionsImpl options,
         {void writeAdditionalOptions(StringBuffer buffer)}) {
       if (options == null) {
@@ -1720,6 +1724,7 @@
             });
             buffer.write('</table>');
           }
+
           writeCountMap('Directly Held AST Nodes', data.directNodeCounts);
           writeCountMap('Indirectly Held AST Nodes', data.indirectNodeCounts);
           writeCountMap('Directly Held Elements', data.elementCounts);
@@ -2418,6 +2423,7 @@
       buffer.write(plugin.runtimeType);
       buffer.write(')<br>');
     }
+
     buffer.write('<h3>Plugin Status</h3><p>');
     writePlugin(AnalysisEngine.instance.enginePlugin);
     writePlugin(_server.serverPlugin);

pkg/analysis_server/test/context_manager_test.dart

diff --git a/pkg/analysis_server/test/context_manager_test.dart b/pkg/analysis_server/test/context_manager_test.dart
index d0a2549..41b0983 100644
--- a/pkg/analysis_server/test/context_manager_test.dart
+++ b/pkg/analysis_server/test/context_manager_test.dart
@@ -10,8 +10,9 @@
 import 'package:analyzer/file_system/file_system.dart';
 import 'package:analyzer/file_system/memory_file_system.dart';
 import 'package:analyzer/instrumentation/instrumentation.dart';
-import 'package:analyzer/source/embedder.dart';
 import 'package:analyzer/source/error_processor.dart';
+import 'package:analyzer/src/context/builder.dart';
+import 'package:analyzer/src/dart/sdk/sdk.dart';
 import 'package:analyzer/src/generated/engine.dart';
 import 'package:analyzer/src/generated/error.dart';
 import 'package:analyzer/src/generated/java_io.dart';
@@ -2650,7 +2651,8 @@
     if (currentContext is InternalAnalysisContext) {
       EmbedderYamlLocator embedderYamlLocator =
           disposition.getEmbedderLocator(resourceProvider);
-      EmbedderSdk sdk = new EmbedderSdk(embedderYamlLocator.embedderYamls);
+      EmbedderSdk sdk =
+          new EmbedderSdk(resourceProvider, embedderYamlLocator.embedderYamls);
       if (sdk.libraryMap.size() > 0) {
         // We have some embedder dart: uri mappings, add the resolver
         // to the list.

pkg/analyzer/lib/source/embedder.dart

diff --git a/pkg/analyzer/lib/source/embedder.dart b/pkg/analyzer/lib/source/embedder.dart
index 56abc6a..94a72a8 100644
--- a/pkg/analyzer/lib/source/embedder.dart
+++ b/pkg/analyzer/lib/source/embedder.dart
@@ -2,6 +2,7 @@
 // for details. All rights reserved. Use of this source code is governed by a
 // BSD-style license that can be found in the LICENSE file.
 
+@deprecated
 library analyzer.source.embedder;
 
 import 'dart:collection' show HashMap;
@@ -19,13 +20,17 @@
 import 'package:analyzer/src/summary/idl.dart';
 import 'package:yaml/yaml.dart';
 
+export 'package:analyzer/src/context/builder.dart' show EmbedderYamlLocator;
+
 const String _DART_COLON_PREFIX = 'dart:';
 const String _EMBEDDED_LIB_MAP_KEY = 'embedded_libs';
 
 /// Check if this map defines embedded libraries.
+@deprecated
 bool definesEmbeddedLibs(Map map) => map[_EMBEDDED_LIB_MAP_KEY] != null;
 
 /// An SDK backed by URI mappings derived from an `_embedder.yaml` file.
+@deprecated
 class EmbedderSdk extends AbstractDartSdk {
   final Map<String, String> _urlMappings = new HashMap<String, String>();
 
@@ -175,74 +180,3 @@
     return sdkSource?.uri;
   }
 }
-
-/// Given a packageMap, check in each package's lib directory for the
-/// existence of an `_embedder.yaml` file. If the file contains a top level
-/// YamlMap, it will be added to the [embedderYamls] map.
-class EmbedderYamlLocator {
-  static const String EMBEDDER_FILE_NAME = '_embedder.yaml';
-
-  /// Map from package's library directory to the parsed YamlMap.
-  final Map<Folder, YamlMap> embedderYamls = new HashMap<Folder, YamlMap>();
-
-  EmbedderYamlLocator(Map<String, List<Folder>> packageMap) {
-    if (packageMap != null) {
-      _processPackageMap(packageMap);
-    }
-  }
-
-  /// Programatically add an _embedder.yaml mapping.
-  void addEmbedderYaml(Folder libDir, String embedderYaml) {
-    _processEmbedderYaml(libDir, embedderYaml);
-  }
-
-  void refresh(Map<String, List<Folder>> packageMap) {
-    // Clear existing.
-    embedderYamls.clear();
-    if (packageMap != null) {
-      _processPackageMap(packageMap);
-    }
-  }
-
-  /// Given the yaml for an embedder ([embedderYaml]) and a folder
-  /// ([libDir]), setup the uri mapping.
-  void _processEmbedderYaml(Folder libDir, String embedderYaml) {
-    YamlNode yaml;
-    try {
-      yaml = loadYaml(embedderYaml);
-    } catch (_) {
-      return;
-    }
-    if (yaml is! YamlMap) {
-      return;
-    }
-    embedderYamls[libDir] = yaml;
-  }
-
-  /// Given a package [name] and a list of folders ([libDirs]),
-  /// add any found `_embedder.yaml` files.
-  void _processPackage(String name, List<Folder> libDirs) {
-    for (Folder libDir in libDirs) {
-      String embedderYaml = _readEmbedderYaml(libDir);
-      if (embedderYaml != null) {
-        _processEmbedderYaml(libDir, embedderYaml);
-      }
-    }
-  }
-
-  void _processPackageMap(Map<String, List<Folder>> packageMap) {
-    packageMap.forEach(_processPackage);
-  }
-
-  /// Read the contents of [libDir]/[EMBEDDER_FILE_NAME] as a string.
-  /// Returns null if the file doesn't exist.
-  String _readEmbedderYaml(Folder libDir) {
-    File file = libDir.getChild(EMBEDDER_FILE_NAME);
-    try {
-      return file.readAsStringSync();
-    } on FileSystemException {
-      // File can't be read.
-      return null;
-    }
-  }
-}

pkg/analyzer/lib/src/context/builder.dart

diff --git a/pkg/analyzer/lib/src/context/builder.dart b/pkg/analyzer/lib/src/context/builder.dart
index 377e9b1..729fa65 100644
--- a/pkg/analyzer/lib/src/context/builder.dart
+++ b/pkg/analyzer/lib/src/context/builder.dart
@@ -12,9 +12,9 @@
 import 'package:analyzer/file_system/file_system.dart';
 import 'package:analyzer/plugin/resolver_provider.dart';
 import 'package:analyzer/source/analysis_options_provider.dart';
-import 'package:analyzer/source/embedder.dart';
 import 'package:analyzer/source/package_map_resolver.dart';
 import 'package:analyzer/source/sdk_ext.dart';
+import 'package:analyzer/src/dart/sdk/sdk.dart';
 import 'package:analyzer/src/generated/engine.dart';
 import 'package:analyzer/src/generated/java_io.dart';
 import 'package:analyzer/src/generated/sdk.dart';
@@ -210,6 +210,7 @@
     Folder folder() {
       return _folder ??= resourceProvider.getResource('.');
     }
+
     UriResolver fileResolver = fileResolverProvider == null
         ? new ResourceUriResolver(resourceProvider)
         : fileResolverProvider(folder());
@@ -264,7 +265,8 @@
       List<String> extFilePaths = extResolver.extensionFilePaths;
       EmbedderYamlLocator locator = new EmbedderYamlLocator(packageMap);
       Map<Folder, YamlMap> embedderYamls = locator.embedderYamls;
-      EmbedderSdk embedderSdk = new EmbedderSdk(embedderYamls);
+      EmbedderSdk embedderSdk =
+          new EmbedderSdk(resourceProvider, embedderYamls);
       if (embedderSdk.sdkLibraries.length > 0) {
         //
         // There is an embedder file that defines the content of the SDK and
@@ -357,3 +359,98 @@
     return null;
   }
 }
+
+/**
+ * Given a package map, check in each package's lib directory for the existence
+ * of an `_embedder.yaml` file. If the file contains a top level YamlMap, it
+ * will be added to the [embedderYamls] map.
+ */
+class EmbedderYamlLocator {
+  /**
+   * The name of the embedder files being searched for.
+   */
+  static const String EMBEDDER_FILE_NAME = '_embedder.yaml';
+
+  /**
+   * A mapping from a package's library directory to the parsed YamlMap.
+   */
+  final Map<Folder, YamlMap> embedderYamls = new HashMap<Folder, YamlMap>();
+
+  /**
+   * Initialize a newly created locator by processing the packages in the given
+   * [packageMap].
+   */
+  EmbedderYamlLocator(Map<String, List<Folder>> packageMap) {
+    if (packageMap != null) {
+      _processPackageMap(packageMap);
+    }
+  }
+
+  /**
+   * Programatically add an `_embedder.yaml` mapping.
+   */
+  void addEmbedderYaml(Folder libDir, String embedderYaml) {
+    _processEmbedderYaml(libDir, embedderYaml);
+  }
+
+  /**
+   * Refresh the map of located files to those found by processing the given
+   * [packageMap].
+   */
+  void refresh(Map<String, List<Folder>> packageMap) {
+    // Clear existing.
+    embedderYamls.clear();
+    if (packageMap != null) {
+      _processPackageMap(packageMap);
+    }
+  }
+
+  /**
+   * Given the yaml for an embedder ([embedderYaml]) and a folder ([libDir]),
+   * setup the uri mapping.
+   */
+  void _processEmbedderYaml(Folder libDir, String embedderYaml) {
+    try {
+      YamlNode yaml = loadYaml(embedderYaml);
+      if (yaml is YamlMap) {
+        embedderYamls[libDir] = yaml;
+      }
+    } catch (_) {
+      // Ignored
+    }
+  }
+
+  /**
+   * Given a package [name] and a list of folders ([libDirs]), process any
+   * `_embedder.yaml` files that are found in any of the folders.
+   */
+  void _processPackage(String name, List<Folder> libDirs) {
+    for (Folder libDir in libDirs) {
+      String embedderYaml = _readEmbedderYaml(libDir);
+      if (embedderYaml != null) {
+        _processEmbedderYaml(libDir, embedderYaml);
+      }
+    }
+  }
+
+  /**
+   * Process each of the entries in the [packageMap].
+   */
+  void _processPackageMap(Map<String, List<Folder>> packageMap) {
+    packageMap.forEach(_processPackage);
+  }
+
+  /**
+   * Read and return the contents of [libDir]/[EMBEDDER_FILE_NAME], or `null` if
+   * the file doesn't exist.
+   */
+  String _readEmbedderYaml(Folder libDir) {
+    File file = libDir.getChild(EMBEDDER_FILE_NAME);
+    try {
+      return file.readAsStringSync();
+    } on FileSystemException {
+      // File can't be read.
+      return null;
+    }
+  }
+}

pkg/analyzer/lib/src/context/context.dart

diff --git a/pkg/analyzer/lib/src/context/context.dart b/pkg/analyzer/lib/src/context/context.dart
index fe33e3a..24bf8d5 100644
--- a/pkg/analyzer/lib/src/context/context.dart
+++ b/pkg/analyzer/lib/src/context/context.dart
@@ -13,8 +13,8 @@
 import 'package:analyzer/instrumentation/instrumentation.dart';
 import 'package:analyzer/plugin/resolver_provider.dart';
 import 'package:analyzer/plugin/task.dart';
-import 'package:analyzer/source/embedder.dart';
 import 'package:analyzer/src/cancelable_future.dart';
+import 'package:analyzer/src/context/builder.dart' show EmbedderYamlLocator;
 import 'package:analyzer/src/context/cache.dart';
 import 'package:analyzer/src/dart/element/element.dart';
 import 'package:analyzer/src/dart/element/type.dart';
@@ -1251,6 +1251,7 @@
       setValue(ResultDescriptor result, value) {
         entry.setValue(result, value, TargetedResult.EMPTY_LIST);
       }
+
       setValue(BUILD_DIRECTIVES_ERRORS, AnalysisError.NO_ERRORS);
       setValue(BUILD_LIBRARY_ERRORS, AnalysisError.NO_ERRORS);
       // CLASS_ELEMENTS
@@ -1767,6 +1768,7 @@
       }
       return false;
     }
+
     return _refHtml(librarySource);
   }
 

pkg/analyzer/lib/src/dart/element/element.dart

diff --git a/pkg/analyzer/lib/src/dart/element/element.dart b/pkg/analyzer/lib/src/dart/element/element.dart
index ddb414e..6f1ec8d 100644
--- a/pkg/analyzer/lib/src/dart/element/element.dart
+++ b/pkg/analyzer/lib/src/dart/element/element.dart
@@ -5872,17 +5872,6 @@
    * cached library cycles in the element model which may have been invalidated.
    */
   void invalidateLibraryCycles() {
-    if (_libraryCycle == null) {
-      // We have already invalidated this node, or we have never computed
-      // library cycle information for it.  In the former case, we're done. In
-      // the latter case, this node cannot be reachable from any node for which
-      // we have computed library cycle information.  Therefore, any edges added
-      // or deleted in the update causing this invalidation can only be edges to
-      // nodes which either have no library cycle information (and hence do not
-      // need invalidation), or which do not reach this node by any path.
-      // In either case, no further invalidation is needed.
-      return;
-    }
     // If we have pre-computed library cycle information, then we must
     // invalidate the information both on this element, and on certain
     // other elements.  Edges originating at this node may have been

pkg/analyzer/lib/src/dart/element/type.dart

diff --git a/pkg/analyzer/lib/src/dart/element/type.dart b/pkg/analyzer/lib/src/dart/element/type.dart
index df19008..2c59228 100644
--- a/pkg/analyzer/lib/src/dart/element/type.dart
+++ b/pkg/analyzer/lib/src/dart/element/type.dart
@@ -1957,11 +1957,19 @@
     sj.add(j);
     // compute intersection, reference as set 's'
     List<InterfaceType> s = _intersection(si, sj);
+    return computeTypeAtMaxUniqueDepth(s);
+  }
+  
+  /**
+   * Return the type from the [types] list that has the longest inheritence path
+   * to Object of unique length.
+   */
+  static InterfaceType computeTypeAtMaxUniqueDepth(List<InterfaceType> types) {
     // for each element in Set s, compute the largest inheritance path to Object
-    List<int> depths = new List<int>.filled(s.length, 0);
+    List<int> depths = new List<int>.filled(types.length, 0);
     int maxDepth = 0;
-    for (int n = 0; n < s.length; n++) {
-      depths[n] = computeLongestInheritancePathToObject(s[n]);
+    for (int n = 0; n < types.length; n++) {
+      depths[n] = computeLongestInheritancePathToObject(types[n]);
       if (depths[n] > maxDepth) {
         maxDepth = depths[n];
       }
@@ -1978,7 +1986,7 @@
         }
       }
       if (numberOfTypesAtMaxDepth == 1) {
-        return s[indexOfLeastUpperBound];
+        return types[indexOfLeastUpperBound];
       }
     }
     // Should be impossible--there should always be exactly one type with the

pkg/analyzer/lib/src/dart/sdk/sdk.dart

diff --git a/pkg/analyzer/lib/src/dart/sdk/sdk.dart b/pkg/analyzer/lib/src/dart/sdk/sdk.dart
index 338a560..27e5549 100644
--- a/pkg/analyzer/lib/src/dart/sdk/sdk.dart
+++ b/pkg/analyzer/lib/src/dart/sdk/sdk.dart
@@ -119,6 +119,20 @@
     _useSummary = use;
   }
 
+  /**
+   * Add the extensions from one or more sdk extension files to this sdk. The
+   * [extensions] should be a table mapping the names of extensions to the paths
+   * where those extensions can be found.
+   */
+  void addExtensions(Map<String, String> extensions) {
+    extensions.forEach((String uri, String path) {
+      String shortName = uri.substring(uri.indexOf(':') + 1);
+      SdkLibraryImpl library = new SdkLibraryImpl(shortName);
+      library.path = path;
+      libraryMap.setLibrary(uri, library);
+    });
+  }
+
   @override
   Source fromFileUri(Uri uri) {
     File file =
@@ -284,8 +298,7 @@
       srcPath = library.path;
     } else {
       String libraryPath = library.path;
-      int index =
-          libraryPath.lastIndexOf(resourceProvider.pathContext.separator);
+      int index = libraryPath.lastIndexOf(separator);
       if (index == -1) {
         index = libraryPath.lastIndexOf('/');
         if (index == -1) {
@@ -295,8 +308,7 @@
       String prefix = libraryPath.substring(0, index + 1);
       srcPath = '$prefix$relativePath';
     }
-    String filePath =
-        srcPath.replaceAll('/', resourceProvider.pathContext.separator);
+    String filePath = srcPath.replaceAll('/', separator);
     try {
       File file = resourceProvider.getFile(filePath);
       return file.createSource(parseUriWithException(dartUri));
@@ -522,8 +534,7 @@
   String getRelativePathFromFile(File file) {
     String filePath = file.path;
     String libPath = libraryDirectory.path;
-    if (!filePath
-        .startsWith("$libPath${resourceProvider.pathContext.separator}")) {
+    if (!filePath.startsWith("$libPath$separator")) {
       return null;
     }
     return filePath.substring(libPath.length + 1);

pkg/analyzer/lib/src/generated/engine.dart

diff --git a/pkg/analyzer/lib/src/generated/engine.dart b/pkg/analyzer/lib/src/generated/engine.dart
index 967095a..b229169 100644
--- a/pkg/analyzer/lib/src/generated/engine.dart
+++ b/pkg/analyzer/lib/src/generated/engine.dart
@@ -12,8 +12,8 @@
 import 'package:analyzer/dart/element/element.dart';
 import 'package:analyzer/instrumentation/instrumentation.dart';
 import 'package:analyzer/plugin/resolver_provider.dart';
-import 'package:analyzer/source/embedder.dart';
 import 'package:analyzer/src/cancelable_future.dart';
+import 'package:analyzer/src/context/builder.dart' show EmbedderYamlLocator;
 import 'package:analyzer/src/context/cache.dart';
 import 'package:analyzer/src/context/context.dart';
 import 'package:analyzer/src/generated/constant.dart';
@@ -1456,6 +1456,7 @@
       buffer.write(optionName);
       needsSeparator = true;
     }
+
     if (encoding & ENABLE_ASSERT_FLAG > 0) {
       add('assert');
     }

pkg/analyzer/lib/src/generated/error.dart

diff --git a/pkg/analyzer/lib/src/generated/error.dart b/pkg/analyzer/lib/src/generated/error.dart
index 728435f..db9ad7ec 100644
--- a/pkg/analyzer/lib/src/generated/error.dart
+++ b/pkg/analyzer/lib/src/generated/error.dart
@@ -2743,6 +2743,7 @@
     StaticTypeWarningCode.NON_BOOL_EXPRESSION,
     StaticTypeWarningCode.NON_BOOL_NEGATION_EXPRESSION,
     StaticTypeWarningCode.NON_BOOL_OPERAND,
+    StaticTypeWarningCode.NON_NULLABLE_FIELD_NOT_INITIALIZED,
     StaticTypeWarningCode.NON_TYPE_AS_TYPE_ARGUMENT,
     StaticTypeWarningCode.RETURN_OF_INVALID_TYPE,
     StaticTypeWarningCode.TYPE_ARGUMENT_NOT_MATCHING_BOUNDS,
@@ -4281,6 +4282,14 @@
           "The operands of the '{0}' operator must be assignable to 'bool'");
 
   /**
+   *
+   */
+  static const StaticTypeWarningCode NON_NULLABLE_FIELD_NOT_INITIALIZED =
+      const StaticTypeWarningCode(
+          'NON_NULLABLE_FIELD_NOT_INITIALIZED',
+          "Variable '{0}' of non-nullable type '{1}' must be initialized");
+
+  /**
    * 15.8 Parameterized Types: It is a static type warning if <i>A<sub>i</sub>,
    * 1 &lt;= i &lt;= n</i> does not denote a type in the enclosing lexical scope.
    */

pkg/analyzer/lib/src/generated/incremental_resolver.dart

diff --git a/pkg/analyzer/lib/src/generated/incremental_resolver.dart b/pkg/analyzer/lib/src/generated/incremental_resolver.dart
index 1194f5f..274b7a1 100644
--- a/pkg/analyzer/lib/src/generated/incremental_resolver.dart
+++ b/pkg/analyzer/lib/src/generated/incremental_resolver.dart
@@ -569,41 +569,52 @@
         int endOffsetOld = math.max(firstOffsetOld, lastOffsetOld);
         int beginOffsetNew = math.min(firstOffsetNew, lastOffsetNew);
         int endOffsetNew = math.max(firstOffsetNew, lastOffsetNew);
-        // check for a whitespace only change
-        if (identical(lastPair.oldToken, firstPair.oldToken) &&
-            identical(lastPair.newToken, firstPair.newToken)) {
+        // A pure whitespace change.
+        if (identical(firstPair.oldToken, lastPair.oldToken) &&
+            identical(firstPair.newToken, lastPair.newToken) &&
+            firstPair.kind == _TokenDifferenceKind.OFFSET) {
           _updateOffset = beginOffsetOld - 1;
           _updateEndOld = endOffsetOld;
           _updateEndNew = endOffsetNew;
           _updateDelta = newUnit.length - _oldUnit.length;
-          // A Dart documentation comment change.
-          if (firstPair.kind == _TokenDifferenceKind.COMMENT_DOC) {
-            bool success = _resolveCommentDoc(newUnit, firstPair);
+          logger.log('Whitespace change.');
+          _shiftTokens(firstPair.oldToken, true);
+          IncrementalResolver incrementalResolver = new IncrementalResolver(
+              _cache,
+              _sourceEntry,
+              _unitEntry,
+              _unitElement,
+              _updateOffset,
+              _updateEndOld,
+              _updateEndNew);
+          incrementalResolver._updateCache();
+          incrementalResolver._updateElementNameOffsets();
+          incrementalResolver._shiftEntryErrors();
+          _updateEntry();
+          logger.log('Success.');
+          return true;
+        }
+        // A Dart documentation comment change.
+        {
+          Token firstOldToken = firstPair.oldToken;
+          Token firstNewToken = firstPair.newToken;
+          Token lastOldToken = lastPair.oldToken;
+          Token lastNewToken = lastPair.newToken;
+          if (firstOldToken is DocumentationCommentToken &&
+              firstNewToken is DocumentationCommentToken &&
+              lastOldToken is DocumentationCommentToken &&
+              lastNewToken is DocumentationCommentToken &&
+              identical(firstOldToken.parent, lastOldToken.parent) &&
+              identical(firstNewToken.parent, lastNewToken.parent)) {
+            _updateOffset = beginOffsetOld;
+            _updateEndOld = firstOldToken.parent.offset;
+            _updateEndNew = firstNewToken.parent.offset;
+            _updateDelta = newUnit.length - _oldUnit.length;
+            bool success =
+                _resolveCommentDoc(newUnit, firstOldToken, firstNewToken);
             logger.log('Documentation comment resolved: $success');
             return success;
           }
-          // A pure whitespace change.
-          if (firstPair.kind == _TokenDifferenceKind.OFFSET) {
-            logger.log('Whitespace change.');
-            _shiftTokens(firstPair.oldToken);
-            {
-              IncrementalResolver incrementalResolver = new IncrementalResolver(
-                  _cache,
-                  _sourceEntry,
-                  _unitEntry,
-                  _unitElement,
-                  _updateOffset,
-                  _updateEndOld,
-                  _updateEndNew);
-              incrementalResolver._updateCache();
-              incrementalResolver._updateElementNameOffsets();
-              incrementalResolver._shiftEntryErrors();
-            }
-            _updateEntry();
-            logger.log('Success.');
-            return true;
-          }
-          // fall-through, end-of-line comment
         }
         // Find nodes covering the "old" and "new" token ranges.
         AstNode oldNode =
@@ -677,14 +688,6 @@
             logger.log('Failure: no enclosing function body or executable.');
             return false;
           }
-          // fail if a comment change outside the bodies
-          if (firstPair.kind == _TokenDifferenceKind.COMMENT) {
-            if (beginOffsetOld <= oldNode.offset ||
-                beginOffsetNew <= newNode.offset) {
-              logger.log('Failure: comment outside a function body.');
-              return false;
-            }
-          }
         }
         logger.log(() => 'oldNode: $oldNode');
         logger.log(() => 'newNode: $newNode');
@@ -756,16 +759,13 @@
    * Attempts to resolve a documentation comment change.
    * Returns `true` if success.
    */
-  bool _resolveCommentDoc(CompilationUnit newUnit, _TokenPair firstPair) {
-    Token oldToken = firstPair.oldToken;
-    Token newToken = firstPair.newToken;
-    CommentToken oldComments = oldToken.precedingComments;
-    CommentToken newComments = newToken.precedingComments;
-    if (oldComments == null || newComments == null) {
+  bool _resolveCommentDoc(
+      CompilationUnit newUnit, CommentToken oldToken, CommentToken newToken) {
+    if (oldToken == null || newToken == null) {
       return false;
     }
     // find nodes
-    int offset = oldComments.offset;
+    int offset = oldToken.offset;
     logger.log('offset: $offset');
     AstNode oldNode = _findNodeCovering(_oldUnit, offset, offset);
     AstNode newNode = _findNodeCovering(newUnit, offset, offset);
@@ -776,10 +776,9 @@
     Comment newComment = newNode;
     logger.log('oldComment.beginToken: ${oldComment.beginToken}');
     logger.log('newComment.beginToken: ${newComment.beginToken}');
-    _updateOffset = oldToken.offset - 1;
     // update token references
-    _shiftTokens(firstPair.oldToken);
-    _setPrecedingComments(oldToken, newComment.tokens.first);
+    _shiftTokens(oldToken.parent);
+    _setPrecedingComments(oldToken.parent, newComment.tokens.first);
     // replace node
     NodeReplacer.replace(oldComment, newComment);
     // update elements
@@ -848,8 +847,11 @@
     }
   }
 
-  void _shiftTokens(Token token) {
+  void _shiftTokens(Token token, [bool goUpComment = false]) {
     while (token != null) {
+      if (goUpComment && token is CommentToken) {
+        token = (token as CommentToken).parent;
+      }
       if (token.offset > _updateOffset) {
         token.offset += _updateDelta;
       }
@@ -893,29 +895,21 @@
   }
 
   static _TokenDifferenceKind _compareToken(
-      Token oldToken, Token newToken, int delta, bool forComment) {
-    while (true) {
-      if (oldToken == null && newToken == null) {
-        return null;
-      }
-      if (oldToken == null || newToken == null) {
-        return _TokenDifferenceKind.CONTENT;
-      }
-      if (oldToken.type != newToken.type) {
-        return _TokenDifferenceKind.CONTENT;
-      }
-      if (oldToken.lexeme != newToken.lexeme) {
-        return _TokenDifferenceKind.CONTENT;
-      }
-      if (newToken.offset - oldToken.offset != delta) {
-        return _TokenDifferenceKind.OFFSET;
-      }
-      // continue if comment tokens are being checked
-      if (!forComment) {
-        break;
-      }
-      oldToken = oldToken.next;
-      newToken = newToken.next;
+      Token oldToken, Token newToken, int delta) {
+    if (oldToken == null && newToken == null) {
+      return null;
+    }
+    if (oldToken == null || newToken == null) {
+      return _TokenDifferenceKind.CONTENT;
+    }
+    if (oldToken.type != newToken.type) {
+      return _TokenDifferenceKind.CONTENT;
+    }
+    if (oldToken.lexeme != newToken.lexeme) {
+      return _TokenDifferenceKind.CONTENT;
+    }
+    if (newToken.offset - oldToken.offset != delta) {
+      return _TokenDifferenceKind.OFFSET;
     }
     return null;
   }
@@ -929,18 +923,22 @@
       {
         Token oldComment = oldToken.precedingComments;
         Token newComment = newToken.precedingComments;
-        if (_compareToken(oldComment, newComment, 0, true) != null) {
-          _TokenDifferenceKind diffKind = _TokenDifferenceKind.COMMENT;
-          if (oldComment is DocumentationCommentToken &&
-              newComment is DocumentationCommentToken) {
-            diffKind = _TokenDifferenceKind.COMMENT_DOC;
+        while (true) {
+          _TokenDifferenceKind diffKind =
+              _compareToken(oldComment, newComment, 0);
+          if (diffKind != null) {
+            return new _TokenPair(
+                diffKind, oldComment ?? oldToken, newComment ?? newToken);
           }
-          return new _TokenPair(diffKind, oldToken, newToken);
+          if (oldComment == null && newComment == null) {
+            break;
+          }
+          oldComment = oldComment.next;
+          newComment = newComment.next;
         }
       }
       // compare tokens
-      _TokenDifferenceKind diffKind =
-          _compareToken(oldToken, newToken, 0, false);
+      _TokenDifferenceKind diffKind = _compareToken(oldToken, newToken, 0);
       if (diffKind != null) {
         return new _TokenPair(diffKind, oldToken, newToken);
       }
@@ -954,24 +952,40 @@
 
   static _TokenPair _findLastDifferentToken(Token oldToken, Token newToken) {
     int delta = newToken.offset - oldToken.offset;
+    Token prevOldToken;
+    Token prevNewToken;
     while (oldToken.previous != oldToken && newToken.previous != newToken) {
       // compare tokens
-      _TokenDifferenceKind diffKind =
-          _compareToken(oldToken, newToken, delta, false);
+      _TokenDifferenceKind diffKind = _compareToken(oldToken, newToken, delta);
       if (diffKind != null) {
-        return new _TokenPair(diffKind, oldToken.next, newToken.next);
+        return new _TokenPair(diffKind, prevOldToken, prevNewToken);
       }
+      prevOldToken = oldToken;
+      prevNewToken = newToken;
       // compare comments
       {
         Token oldComment = oldToken.precedingComments;
         Token newComment = newToken.precedingComments;
-        if (_compareToken(oldComment, newComment, delta, true) != null) {
-          _TokenDifferenceKind diffKind = _TokenDifferenceKind.COMMENT;
-          if (oldComment is DocumentationCommentToken &&
-              newComment is DocumentationCommentToken) {
-            diffKind = _TokenDifferenceKind.COMMENT_DOC;
+        while (oldComment?.next != null) {
+          oldComment = oldComment.next;
+        }
+        while (newComment?.next != null) {
+          newComment = newComment.next;
+        }
+        while (true) {
+          _TokenDifferenceKind diffKind =
+              _compareToken(oldComment, newComment, delta);
+          if (diffKind != null) {
+            return new _TokenPair(
+                diffKind, oldComment ?? oldToken, newComment ?? newToken);
           }
-          return new _TokenPair(diffKind, oldToken, newToken);
+          if (oldComment == null && newComment == null) {
+            break;
+          }
+          prevOldToken = oldComment;
+          prevNewToken = newComment;
+          oldComment = oldComment.previous;
+          newComment = newComment.previous;
         }
       }
       // next tokens
@@ -1290,8 +1304,6 @@
  * Describes how two [Token]s are different.
  */
 class _TokenDifferenceKind {
-  static const COMMENT = const _TokenDifferenceKind('COMMENT');
-  static const COMMENT_DOC = const _TokenDifferenceKind('COMMENT_DOC');
   static const CONTENT = const _TokenDifferenceKind('CONTENT');
   static const OFFSET = const _TokenDifferenceKind('OFFSET');
 

pkg/analyzer/lib/src/generated/resolver.dart

diff --git a/pkg/analyzer/lib/src/generated/resolver.dart b/pkg/analyzer/lib/src/generated/resolver.dart
index afa7cce..4a70c66 100644
--- a/pkg/analyzer/lib/src/generated/resolver.dart
+++ b/pkg/analyzer/lib/src/generated/resolver.dart
@@ -4915,6 +4915,8 @@
    * inference.
    */
   static void setType(AstNode node, DartType type) {
+    // TODO(jmesserly): this sets the type even when it's dynamic.
+    // Can we skip that?
     node?.setProperty(_typeProperty, type);
   }
 
@@ -6062,9 +6064,18 @@
       // of the binary operator for other cases.
       if (operatorType == TokenType.QUESTION_QUESTION) {
         InferenceContext.setTypeFromNode(leftOperand, node);
-        InferenceContext.setTypeFromNode(rightOperand, node);
       }
       leftOperand?.accept(this);
+      if (operatorType == TokenType.QUESTION_QUESTION) {
+        // Set the right side, either from the context, or using the information
+        // from the left side if it is more precise.
+        DartType contextType = InferenceContext.getType(node);
+        DartType leftType = leftOperand?.staticType;
+        if (contextType == null || contextType.isDynamic) {
+          contextType = leftType;
+        }
+        InferenceContext.setType(rightOperand, contextType);
+      }
       rightOperand?.accept(this);
     }
     node.accept(elementResolver);

pkg/analyzer/lib/src/generated/static_type_analyzer.dart

diff --git a/pkg/analyzer/lib/src/generated/static_type_analyzer.dart b/pkg/analyzer/lib/src/generated/static_type_analyzer.dart
index f681471..fa82560 100644
--- a/pkg/analyzer/lib/src/generated/static_type_analyzer.dart
+++ b/pkg/analyzer/lib/src/generated/static_type_analyzer.dart
@@ -441,6 +441,10 @@
     if (node.parent is FunctionDeclarationStatement) {
       // TypeResolverVisitor sets the return type for top-level functions, so
       // we only need to handle local functions.
+      if (_strongMode && node.returnType == null) {
+        _inferLocalFunctionReturnType(node.functionExpression);
+        return null;
+      }
       functionElement.returnType =
           _computeStaticReturnTypeOfFunctionDeclaration(node);
       _recordPropagatedTypeOfFunction(functionElement, function.body);
@@ -486,6 +490,15 @@
       // node.
       return null;
     }
+    _inferLocalFunctionReturnType(node);
+    return null;
+  }
+
+  /**
+   * Infers the return type of a local function, either a lambda or
+   * (in strong mode) a local function declaration.
+   */
+  void _inferLocalFunctionReturnType(FunctionExpression node) {
     bool recordInference = false;
     ExecutableElementImpl functionElement =
         node.element as ExecutableElementImpl;
@@ -517,7 +530,6 @@
     if (recordInference) {
       _resolver.inferenceContext.recordInference(node, functionElement.type);
     }
-    return null;
   }
 
   /**

pkg/analyzer/lib/src/generated/type_system.dart

diff --git a/pkg/analyzer/lib/src/generated/type_system.dart b/pkg/analyzer/lib/src/generated/type_system.dart
index 9154132..78d246f 100644
--- a/pkg/analyzer/lib/src/generated/type_system.dart
+++ b/pkg/analyzer/lib/src/generated/type_system.dart
@@ -45,7 +45,38 @@
   }
 
   @override
-  bool canPromoteToType(DartType to, DartType from) => isSubtypeOf(to, from);
+  bool canPromoteToType(DartType to, DartType from) {
+    // Allow promoting to a subtype, for example:
+    //
+    //     f(Base b) {
+    //       if (b is SubTypeOfBase) {
+    //         // promote `b` to SubTypeOfBase for this block
+    //       }
+    //     }
+    //
+    // This allows the variable to be used wherever the supertype (here `Base`)
+    // is expected, while gaining a more precise type.
+    if (isSubtypeOf(to, from)) {
+      return true;
+    }
+    // For a type parameter `T extends U`, allow promoting from the upper bound
+    // `U` to `S` where `S <: U`.
+    //
+    // This does restrict the variable, because `S </: T`, it can no longer be
+    // used as a `T` without another cast.
+    //
+    // However the members you could access from a variable of type `T`, were
+    // already those on the upper bound `U`. So all members on `U` will be
+    // accessible, as well as those on `S`. Pragmatically this feels like a
+    // useful enough trade-off to allow promotion.
+    //
+    // (In general we would need union types to support this feature precisely.)
+    if (from is TypeParameterType) {
+      return isSubtypeOf(to, from.resolveToBound(DynamicTypeImpl.instance));
+    }
+
+    return false;
+  }
 
   @override
   FunctionType functionTypeToConcreteType(
@@ -91,26 +122,24 @@
 
   /// Computes the greatest lower bound of [type1] and [type2].
   DartType getGreatestLowerBound(
-      TypeProvider provider, DartType type1, DartType type2) {
+      TypeProvider provider, DartType type1, DartType type2,
+      {dynamicIsBottom: false}) {
     // The greatest lower bound relation is reflexive.
     if (identical(type1, type2)) {
       return type1;
     }
 
-    // Treat dynamic as top. The GLB of dynamic and any type is just that type
-    // since dynamic permits all values.
-    if (type1.isDynamic) {
+    // The GLB of top and any type is just that type.
+    // Also GLB of bottom and any type is bottom.
+    if (_isTop(type1, dynamicIsBottom: dynamicIsBottom) ||
+        _isBottom(type2, dynamicIsBottom: dynamicIsBottom)) {
       return type2;
     }
-    if (type2.isDynamic) {
+    if (_isTop(type2, dynamicIsBottom: dynamicIsBottom) ||
+        _isBottom(type1, dynamicIsBottom: dynamicIsBottom)) {
       return type1;
     }
 
-    // You can't get any lower than bottom.
-    if (type1.isBottom || type2.isBottom) {
-      return provider.bottomType;
-    }
-
     // Treat void as top-like for GLB. This only comes into play with the
     // return types of two functions whose GLB is being taken. We allow a
     // non-void-returning function to subtype a void-returning one, so match
@@ -142,6 +171,41 @@
   }
 
   /**
+   * Compute the least upper bound of two types.
+   */
+  @override
+  DartType getLeastUpperBound(
+      TypeProvider typeProvider, DartType type1, DartType type2,
+      {bool dynamicIsBottom: false}) {
+    if (isNullableType(type1) && isNonNullableType(type2)) {
+      assert(type2 is InterfaceType);
+      type2 = getLeastNullableSupertype(type2 as InterfaceType);
+    }
+    if (isNullableType(type2) && isNonNullableType(type1)) {
+      assert(type1 is InterfaceType);
+      type1 = getLeastNullableSupertype(type1 as InterfaceType);
+    }
+    return super.getLeastUpperBound(typeProvider, type1, type2,
+        dynamicIsBottom: dynamicIsBottom);
+  }
+
+  /**
+   * Compute the least supertype of [type], which is known to be an interface
+   * type.
+   *
+   * In the event that the algorithm fails (which might occur due to a bug in
+   * the analyzer), `null` is returned.
+   */
+  DartType getLeastNullableSupertype(InterfaceType type) {
+    // compute set of supertypes
+    List<InterfaceType> s = InterfaceTypeImpl
+        .computeSuperinterfaceSet(type)
+        .where(isNullableType)
+        .toList();
+    return InterfaceTypeImpl.computeTypeAtMaxUniqueDepth(s);
+  }
+
+  /**
    * Given a generic function type `F<T0, T1, ... Tn>` and a context type C,
    * infer an instantiation of F, such that `F<S0, S1, ..., Sn>` <: C.
    *
@@ -443,7 +507,8 @@
       DartType paramType;
       if (fType != null && gType != null) {
         // If both functions have this parameter, include both of their types.
-        paramType = getLeastUpperBound(provider, fType, gType);
+        paramType =
+            getLeastUpperBound(provider, fType, gType, dynamicIsBottom: true);
       } else {
         paramType = fType ?? gType;
       }
@@ -527,7 +592,7 @@
   @override
   DartType _functionParameterBound(
           TypeProvider provider, DartType f, DartType g) =>
-      getGreatestLowerBound(provider, f, g);
+      getGreatestLowerBound(provider, f, g, dynamicIsBottom: true);
 
   /**
    * Guard against loops in the class hierarchy
@@ -606,10 +671,6 @@
         provider.dynamicType;
   }
 
-  bool _isBottom(DartType t, {bool dynamicIsBottom: false}) {
-    return (t.isDynamic && dynamicIsBottom) || t.isBottom;
-  }
-
   /**
    * Check that [f1] is a subtype of [f2].
    *
@@ -758,19 +819,22 @@
     return _isFunctionSubtypeOf(t1 as FunctionType, t2 as FunctionType);
   }
 
-  bool _isTop(DartType t, {bool dynamicIsBottom: false}) {
-    // TODO(leafp): Document the rules in play here
-    return (t.isDynamic && !dynamicIsBottom) || t.isObject;
+  /// Check if [type] is in a set of preselected non-nullable types.
+  /// [FunctionType]s are always nullable.
+  bool isNonNullableType(DartType type) {
+    return !isNullableType(type);
   }
 
-  bool isNonNullableType(DartType type) {
-    return nonnullableTypes.contains(_getTypeFullyQualifiedName(type));
+  /// Opposite of [isNonNullableType].
+  bool isNullableType(DartType type) {
+    return type is FunctionType ||
+        !nonnullableTypes.contains(_getTypeFullyQualifiedName(type));
   }
 
   /// Given a type return its name prepended with the URI to its containing
   /// library and separated by a comma.
   String _getTypeFullyQualifiedName(DartType type) {
-    return "${type?.element?.library?.source?.uri},$type";
+    return "${type?.element?.library?.identifier},$type";
   }
 
   /**
@@ -873,16 +937,20 @@
    * Compute the least upper bound of two types.
    */
   DartType getLeastUpperBound(
-      TypeProvider typeProvider, DartType type1, DartType type2) {
+      TypeProvider typeProvider, DartType type1, DartType type2,
+      {bool dynamicIsBottom: false}) {
     // The least upper bound relation is reflexive.
     if (identical(type1, type2)) {
       return type1;
     }
-    // The least upper bound of dynamic and any type T is dynamic.
-    if (type1.isDynamic) {
+    // The least upper bound of top and any type T is top.
+    // The least upper bound of bottom and any type T is T.
+    if (_isTop(type1, dynamicIsBottom: dynamicIsBottom) ||
+        _isBottom(type2, dynamicIsBottom: dynamicIsBottom)) {
       return type1;
     }
-    if (type2.isDynamic) {
+    if (_isTop(type2, dynamicIsBottom: dynamicIsBottom) ||
+        _isBottom(type1, dynamicIsBottom: dynamicIsBottom)) {
       return type2;
     }
     // The least upper bound of void and any type T != dynamic is void.
@@ -892,13 +960,6 @@
     if (type2.isVoid) {
       return type2;
     }
-    // The least upper bound of bottom and any type T is T.
-    if (type1.isBottom) {
-      return type2;
-    }
-    if (type2.isBottom) {
-      return type1;
-    }
 
     if (type1 is TypeParameterType || type2 is TypeParameterType) {
       return _typeParameterLeastUpperBound(typeProvider, type1, type2);
@@ -1516,3 +1577,12 @@
     }
   }
 }
+
+bool _isBottom(DartType t, {bool dynamicIsBottom: false}) {
+  return (t.isDynamic && dynamicIsBottom) || t.isBottom;
+}
+
+bool _isTop(DartType t, {bool dynamicIsBottom: false}) {
+  // TODO(leafp): Document the rules in play here
+  return (t.isDynamic && !dynamicIsBottom) || t.isObject;
+}

pkg/analyzer/lib/src/summary/api_signature.dart

diff --git a/pkg/analyzer/lib/src/summary/api_signature.dart b/pkg/analyzer/lib/src/summary/api_signature.dart
new file mode 100644
index 0000000..f2fd506
--- /dev/null
+++ b/pkg/analyzer/lib/src/summary/api_signature.dart
@@ -0,0 +1,138 @@
+// Copyright (c) 2016, the Dart project authors.  Please see the AUTHORS file
+// for details. All rights reserved. Use of this source code is governed by a
+// BSD-style license that can be found in the LICENSE file.
+
+library analyzer.src.summary.api_signature;
+
+import 'dart:convert';
+import 'dart:typed_data';
+
+import 'package:convert/convert.dart';
+import 'package:crypto/crypto.dart';
+
+/**
+ * An instance of [ApiSignature] collects data in the form of primitive types
+ * (strings, ints, bools, etc.) from a summary "builder" object, and uses them
+ * to generate an MD5 signature of a the non-informative parts of the summary
+ * (i.e. those parts representing the API of the code being summarized).
+ *
+ * Note that the data passed to the MD5 signature algorithm is untyped.  So, for
+ * instance, an API signature built from a sequence of `false` booleans is
+ * likely to match an API signature built from a sequence of zeros.  The caller
+ * should take this into account; e.g. if a data structure may be represented
+ * either by a boolean or an int, the caller should encode a tag distinguishing
+ * the two representations before encoding the data.
+ */
+class ApiSignature {
+  /**
+   * Version number of the code in this class.  Any time this class is changed
+   * in a way that affects the data collected in [_data], this version number
+   * should be incremented, so that a summary signed by a newer version of the
+   * signature algorithm won't accidentally have the same signature as a summary
+   * signed by an older version.
+   */
+  static const int _VERSION = 0;
+
+  /**
+   * Data accumulated so far.
+   */
+  ByteData _data = new ByteData(4096);
+
+  /**
+   * Offset into [_data] where the next byte should be written.
+   */
+  int _offset = 0;
+
+  /**
+   * Create an [ApiSignature] which is ready to accept data.
+   */
+  ApiSignature() {
+    addInt(_VERSION);
+  }
+
+  /**
+   * For testing only: create an [ApiSignature] which doesn't include any
+   * version information.  This makes it easier to unit tests, since the data
+   * is stable even if [_VERSION] is changed.
+   */
+  ApiSignature.unversioned();
+
+  /**
+   * Collect a boolean value.
+   */
+  void addBool(bool b) {
+    _makeRoom(1);
+    _data.setUint8(_offset, b ? 1 : 0);
+    _offset++;
+  }
+
+  /**
+   * Collect a sequence of arbitrary bytes.  Note that the length is not
+   * collected, so for example `addBytes([1, 2]);` will have the same effect as
+   * `addBytes([1]); addBytes([2]);`.
+   */
+  void addBytes(List<int> bytes) {
+    _makeRoom(bytes.length);
+    new Uint8List.view(_data.buffer)
+        .setRange(_offset, _offset + bytes.length, bytes);
+    _offset += bytes.length;
+  }
+
+  /**
+   * Collect a double-precision floating point value.
+   */
+  void addDouble(double d) {
+    _makeRoom(8);
+    _data.setFloat64(_offset, d, Endianness.LITTLE_ENDIAN);
+    _offset += 8;
+  }
+
+  /**
+   * Collect a 32-bit unsigned integer value.
+   */
+  void addInt(int i) {
+    _makeRoom(4);
+    _data.setUint32(_offset, i, Endianness.LITTLE_ENDIAN);
+    _offset += 4;
+  }
+
+  /**
+   * Collect a string.
+   */
+  void addString(String s) {
+    List<int> bytes = UTF8.encode(s);
+    addInt(bytes.length);
+    addBytes(bytes);
+  }
+
+  /**
+   * For testing only: retrieve the internal representation of the data that
+   * has been collected.
+   */
+  List<int> getBytes_forDebug() {
+    return new Uint8List.view(_data.buffer, 0, _offset).toList();
+  }
+
+  /**
+   * Return a hex-encoded MD5 signature of the data collected so far.
+   */
+  String toHex() {
+    return hex.encode(
+        md5.convert(new Uint8List.view(_data.buffer, 0, _offset)).bytes);
+  }
+
+  /**
+   * Ensure that [spaceNeeded] bytes can be added to [_data] at [_offset]
+   * (copying it to a larger object if necessary).
+   */
+  void _makeRoom(int spaceNeeded) {
+    int oldLength = _data.lengthInBytes;
+    if (_offset + spaceNeeded > oldLength) {
+      int newLength = 2 * (_offset + spaceNeeded);
+      ByteData newData = new ByteData(newLength);
+      new Uint8List.view(newData.buffer)
+          .setRange(0, oldLength, new Uint8List.view(_data.buffer));
+      _data = newData;
+    }
+  }
+}

pkg/analyzer/lib/src/summary/format.dart

diff --git a/pkg/analyzer/lib/src/summary/format.dart b/pkg/analyzer/lib/src/summary/format.dart
index bbed56e..bc4d0ab 100644
--- a/pkg/analyzer/lib/src/summary/format.dart
+++ b/pkg/analyzer/lib/src/summary/format.dart
@@ -10,6 +10,7 @@
 import 'flat_buffers.dart' as fb;
 import 'idl.dart' as idl;
 import 'dart:convert' as convert;
+import 'api_signature.dart' as api_sig;
 
 class _CacheSourceKindReader extends fb.Reader<idl.CacheSourceKind> {
   const _CacheSourceKindReader() : super();
@@ -216,6 +217,17 @@
   void flushInformative() {
   }
 
+  /**
+   * Accumulate non-[informative] data into [signature].
+   */
+  void collectApiSignature(api_sig.ApiSignature signature) {
+    signature.addString(this._errorCodeUniqueName ?? '');
+    signature.addInt(this._offset ?? 0);
+    signature.addInt(this._length ?? 0);
+    signature.addString(this._message ?? '');
+    signature.addString(this._correction ?? '');
+  }
+
   fb.Offset finish(fb.Builder fbBuilder) {
     fb.Offset offset_correction;
     fb.Offset offset_errorCodeUniqueName;
@@ -385,6 +397,37 @@
   void flushInformative() {
   }
 
+  /**
+   * Accumulate non-[informative] data into [signature].
+   */
+  void collectApiSignature(api_sig.ApiSignature signature) {
+    signature.addInt(this._kind == null ? 0 : this._kind.index);
+    if (this._importedUris == null) {
+      signature.addInt(0);
+    } else {
+      signature.addInt(this._importedUris.length);
+      for (var x in this._importedUris) {
+        signature.addString(x);
+      }
+    }
+    if (this._exportedUris == null) {
+      signature.addInt(0);
+    } else {
+      signature.addInt(this._exportedUris.length);
+      for (var x in this._exportedUris) {
+        signature.addString(x);
+      }
+    }
+    if (this._partUris == null) {
+      signature.addInt(0);
+    } else {
+      signature.addInt(this._partUris.length);
+      for (var x in this._partUris) {
+        signature.addString(x);
+      }
+    }
+  }
+
   List<int> toBuffer() {
     fb.Builder fbBuilder = new fb.Builder();
     return fbBuilder.finish(finish(fbBuilder), "CaSS");
@@ -514,6 +557,20 @@
     _errors?.forEach((b) => b.flushInformative());
   }
 
+  /**
+   * Accumulate non-[informative] data into [signature].
+   */
+  void collectApiSignature(api_sig.ApiSignature signature) {
+    if (this._errors == null) {
+      signature.addInt(0);
+    } else {
+      signature.addInt(this._errors.length);
+      for (var x in this._errors) {
+        x?.collectApiSignature(signature);
+      }
+    }
+  }
+
   List<int> toBuffer() {
     fb.Builder fbBuilder = new fb.Builder();
     return fbBuilder.finish(finish(fbBuilder), "CSEL");
@@ -612,6 +669,14 @@
   void flushInformative() {
   }
 
+  /**
+   * Accumulate non-[informative] data into [signature].
+   */
+  void collectApiSignature(api_sig.ApiSignature signature) {
+    signature.addInt(this._offset ?? 0);
+    signature.addInt(this._length ?? 0);
+  }
+
   fb.Offset finish(fb.Builder fbBuilder) {
     fbBuilder.startTable();
     if (_length != null && _length != 0) {
@@ -820,6 +885,41 @@
     _typeArguments?.forEach((b) => b.flushInformative());
   }
 
+  /**
+   * Accumulate non-[informative] data into [signature].
+   */
+  void collectApiSignature(api_sig.ApiSignature signature) {
+    signature.addInt(this._reference ?? 0);
+    if (this._typeArguments == null) {
+      signature.addInt(0);
+    } else {
+      signature.addInt(this._typeArguments.length);
+      for (var x in this._typeArguments) {
+        x?.collectApiSignature(signature);
+      }
+    }
+    signature.addInt(this._slot ?? 0);
+    signature.addInt(this._paramReference ?? 0);
+    if (this._implicitFunctionTypeIndices == null) {
+      signature.addInt(0);
+    } else {
+      signature.addInt(this._implicitFunctionTypeIndices.length);
+      for (var x in this._implicitFunctionTypeIndices) {
+        signature.addInt(x);
+      }
+    }
+    signature.addBool(this._syntheticReturnType != null);
+    this._syntheticReturnType?.collectApiSignature(signature);
+    if (this._syntheticParams == null) {
+      signature.addInt(0);
+    } else {
+      signature.addInt(this._syntheticParams.length);
+      for (var x in this._syntheticParams) {
+        x?.collectApiSignature(signature);
+      }
+    }
+  }
+
   fb.Offset finish(fb.Builder fbBuilder) {
     fb.Offset offset_implicitFunctionTypeIndices;
     fb.Offset offset_syntheticParams;
@@ -995,6 +1095,21 @@
   void flushInformative() {
   }
 
+  /**
+   * Accumulate non-[informative] data into [signature].
+   */
+  void collectApiSignature(api_sig.ApiSignature signature) {
+    signature.addString(this._uri ?? '');
+    if (this._parts == null) {
+      signature.addInt(0);
+    } else {
+      signature.addInt(this._parts.length);
+      for (var x in this._parts) {
+        signature.addString(x);
+      }
+    }
+  }
+
   fb.Offset finish(fb.Builder fbBuilder) {
     fb.Offset offset_parts;
     fb.Offset offset_uri;
@@ -1128,6 +1243,16 @@
   void flushInformative() {
   }
 
+  /**
+   * Accumulate non-[informative] data into [signature].
+   */
+  void collectApiSignature(api_sig.ApiSignature signature) {
+    signature.addInt(this._dependency ?? 0);
+    signature.addString(this._name ?? '');
+    signature.addInt(this._unit ?? 0);
+    signature.addInt(this._kind == null ? 0 : this._kind.index);
+  }
+
   fb.Offset finish(fb.Builder fbBuilder) {
     fb.Offset offset_name;
     if (_name != null) {
@@ -1341,6 +1466,54 @@
     _units?.forEach((b) => b.flushInformative());
   }
 
+  /**
+   * Accumulate non-[informative] data into [signature].
+   */
+  void collectApiSignature(api_sig.ApiSignature signature) {
+    if (this._dependencies == null) {
+      signature.addInt(0);
+    } else {
+      signature.addInt(this._dependencies.length);
+      for (var x in this._dependencies) {
+        x?.collectApiSignature(signature);
+      }
+    }
+    if (this._importDependencies == null) {
+      signature.addInt(0);
+    } else {
+      signature.addInt(this._importDependencies.length);
+      for (var x in this._importDependencies) {
+        signature.addInt(x);
+      }
+    }
+    signature.addInt(this._numPrelinkedDependencies ?? 0);
+    if (this._units == null) {
+      signature.addInt(0);
+    } else {
+      signature.addInt(this._units.length);
+      for (var x in this._units) {
+        x?.collectApiSignature(signature);
+      }
+    }
+    if (this._exportNames == null) {
+      signature.addInt(0);
+    } else {
+      signature.addInt(this._exportNames.length);
+      for (var x in this._exportNames) {
+        x?.collectApiSignature(signature);
+      }
+    }
+    signature.addBool(this._fallbackMode == true);
+    if (this._exportDependencies == null) {
+      signature.addInt(0);
+    } else {
+      signature.addInt(this._exportDependencies.length);
+      for (var x in this._exportDependencies) {
+        signature.addInt(x);
+      }
+    }
+  }
+
   List<int> toBuffer() {
     fb.Builder fbBuilder = new fb.Builder();
     return fbBuilder.finish(finish(fbBuilder), "LLib");
@@ -1617,6 +1790,19 @@
   void flushInformative() {
   }
 
+  /**
+   * Accumulate non-[informative] data into [signature].
+   */
+  void collectApiSignature(api_sig.ApiSignature signature) {
+    signature.addInt(this._unit ?? 0);
+    signature.addInt(this._dependency ?? 0);
+    signature.addInt(this._kind == null ? 0 : this._kind.index);
+    signature.addString(this._name ?? '');
+    signature.addInt(this._numTypeParameters ?? 0);
+    signature.addInt(this._containingReference ?? 0);
+    signature.addInt(this._localIndex ?? 0);
+  }
+
   fb.Offset finish(fb.Builder fbBuilder) {
     fb.Offset offset_name;
     if (_name != null) {
@@ -1797,6 +1983,36 @@
     _types?.forEach((b) => b.flushInformative());
   }
 
+  /**
+   * Accumulate non-[informative] data into [signature].
+   */
+  void collectApiSignature(api_sig.ApiSignature signature) {
+    if (this._references == null) {
+      signature.addInt(0);
+    } else {
+      signature.addInt(this._references.length);
+      for (var x in this._references) {
+        x?.collectApiSignature(signature);
+      }
+    }
+    if (this._types == null) {
+      signature.addInt(0);
+    } else {
+      signature.addInt(this._types.length);
+      for (var x in this._types) {
+        x?.collectApiSignature(signature);
+      }
+    }
+    if (this._constCycles == null) {
+      signature.addInt(0);
+    } else {
+      signature.addInt(this._constCycles.length);
+      for (var x in this._constCycles) {
+        signature.addInt(x);
+      }
+    }
+  }
+
   fb.Offset finish(fb.Builder fbBuilder) {
     fb.Offset offset_constCycles;
     fb.Offset offset_references;
@@ -1882,6 +2098,7 @@
 }
 
 class PackageBundleBuilder extends Object with _PackageBundleMixin implements idl.PackageBundle {
+  String _apiSignature;
   List<LinkedLibraryBuilder> _linkedLibraries;
   List<String> _linkedLibraryUris;
   int _majorVersion;
@@ -1891,6 +2108,18 @@
   List<String> _unlinkedUnitUris;
 
   @override
+  String get apiSignature => _apiSignature ??= '';
+
+  /**
+   * MD5 hash of the non-informative fields of the [PackageBundle] (not
+   * including this one).  This can be used to identify when the API of a
+   * package may have changed.
+   */
+  void set apiSignature(String _value) {
+    _apiSignature = _value;
+  }
+
+  @override
   List<LinkedLibraryBuilder> get linkedLibraries => _linkedLibraries ??= <LinkedLibraryBuilder>[];
 
   /**
@@ -1966,8 +2195,9 @@
     _unlinkedUnitUris = _value;
   }
 
-  PackageBundleBuilder({List<LinkedLibraryBuilder> linkedLibraries, List<String> linkedLibraryUris, int majorVersion, int minorVersion, List<String> unlinkedUnitHashes, List<UnlinkedUnitBuilder> unlinkedUnits, List<String> unlinkedUnitUris})
-    : _linkedLibraries = linkedLibraries,
+  PackageBundleBuilder({String apiSignature, List<LinkedLibraryBuilder> linkedLibraries, List<String> linkedLibraryUris, int majorVersion, int minorVersion, List<String> unlinkedUnitHashes, List<UnlinkedUnitBuilder> unlinkedUnits, List<String> unlinkedUnitUris})
+    : _apiSignature = apiSignature,
+      _linkedLibraries = linkedLibraries,
       _linkedLibraryUris = linkedLibraryUris,
       _majorVersion = majorVersion,
       _minorVersion = minorVersion,
@@ -1984,17 +2214,62 @@
     _unlinkedUnits?.forEach((b) => b.flushInformative());
   }
 
+  /**
+   * Accumulate non-[informative] data into [signature].
+   */
+  void collectApiSignature(api_sig.ApiSignature signature) {
+    if (this._linkedLibraries == null) {
+      signature.addInt(0);
+    } else {
+      signature.addInt(this._linkedLibraries.length);
+      for (var x in this._linkedLibraries) {
+        x?.collectApiSignature(signature);
+      }
+    }
+    if (this._linkedLibraryUris == null) {
+      signature.addInt(0);
+    } else {
+      signature.addInt(this._linkedLibraryUris.length);
+      for (var x in this._linkedLibraryUris) {
+        signature.addString(x);
+      }
+    }
+    if (this._unlinkedUnits == null) {
+      signature.addInt(0);
+    } else {
+      signature.addInt(this._unlinkedUnits.length);
+      for (var x in this._unlinkedUnits) {
+        x?.collectApiSignature(signature);
+      }
+    }
+    if (this._unlinkedUnitUris == null) {
+      signature.addInt(0);
+    } else {
+      signature.addInt(this._unlinkedUnitUris.length);
+      for (var x in this._unlinkedUnitUris) {
+        signature.addString(x);
+      }
+    }
+    signature.addInt(this._majorVersion ?? 0);
+    signature.addInt(this._minorVersion ?? 0);
+    signature.addString(this._apiSignature ?? '');
+  }
+
   List<int> toBuffer() {
     fb.Builder fbBuilder = new fb.Builder();
     return fbBuilder.finish(finish(fbBuilder), "PBdl");
   }
 
   fb.Offset finish(fb.Builder fbBuilder) {
+    fb.Offset offset_apiSignature;
     fb.Offset offset_linkedLibraries;
     fb.Offset offset_linkedLibraryUris;
     fb.Offset offset_unlinkedUnitHashes;
     fb.Offset offset_unlinkedUnits;
     fb.Offset offset_unlinkedUnitUris;
+    if (_apiSignature != null) {
+      offset_apiSignature = fbBuilder.writeString(_apiSignature);
+    }
     if (!(_linkedLibraries == null || _linkedLibraries.isEmpty)) {
       offset_linkedLibraries = fbBuilder.writeList(_linkedLibraries.map((b) => b.finish(fbBuilder)).toList());
     }
@@ -2011,6 +2286,9 @@
       offset_unlinkedUnitUris = fbBuilder.writeList(_unlinkedUnitUris.map((b) => fbBuilder.writeString(b)).toList());
     }
     fbBuilder.startTable();
+    if (offset_apiSignature != null) {
+      fbBuilder.addOffset(7, offset_apiSignature);
+    }
     if (offset_linkedLibraries != null) {
       fbBuilder.addOffset(0, offset_linkedLibraries);
     }
@@ -2054,6 +2332,7 @@
 
   _PackageBundleImpl(this._bc, this._bcOffset);
 
+  String _apiSignature;
   List<idl.LinkedLibrary> _linkedLibraries;
   List<String> _linkedLibraryUris;
   int _majorVersion;
@@ -2063,6 +2342,12 @@
   List<String> _unlinkedUnitUris;
 
   @override
+  String get apiSignature {
+    _apiSignature ??= const fb.StringReader().vTableGet(_bc, _bcOffset, 7, '');
+    return _apiSignature;
+  }
+
+  @override
   List<idl.LinkedLibrary> get linkedLibraries {
     _linkedLibraries ??= const fb.ListReader<idl.LinkedLibrary>(const _LinkedLibraryReader()).vTableGet(_bc, _bcOffset, 0, const <idl.LinkedLibrary>[]);
     return _linkedLibraries;
@@ -2109,6 +2394,7 @@
   @override
   Map<String, Object> toJson() {
     Map<String, Object> _result = <String, Object>{};
+    if (apiSignature != '') _result["apiSignature"] = apiSignature;
     if (linkedLibraries.isNotEmpty) _result["linkedLibraries"] = linkedLibraries.map((_value) => _value.toJson()).toList();
     if (linkedLibraryUris.isNotEmpty) _result["linkedLibraryUris"] = linkedLibraryUris;
     if (majorVersion != 0) _result["majorVersion"] = majorVersion;
@@ -2121,6 +2407,7 @@
 
   @override
   Map<String, Object> toMap() => {
+    "apiSignature": apiSignature,
     "linkedLibraries": linkedLibraries,
     "linkedLibraryUris": linkedLibraryUris,
     "majorVersion": majorVersion,
@@ -2279,6 +2566,84 @@
     _units?.forEach((b) => b.flushInformative());
   }
 
+  /**
+   * Accumulate non-[informative] data into [signature].
+   */
+  void collectApiSignature(api_sig.ApiSignature signature) {
+    if (this._elementUnits == null) {
+      signature.addInt(0);
+    } else {
+      signature.addInt(this._elementUnits.length);
+      for (var x in this._elementUnits) {
+        signature.addInt(x);
+      }
+    }
+    if (this._elementNameUnitMemberIds == null) {
+      signature.addInt(0);
+    } else {
+      signature.addInt(this._elementNameUnitMemberIds.length);
+      for (var x in this._elementNameUnitMemberIds) {
+        signature.addInt(x);
+      }
+    }
+    if (this._unitLibraryUris == null) {
+      signature.addInt(0);
+    } else {
+      signature.addInt(this._unitLibraryUris.length);
+      for (var x in this._unitLibraryUris) {
+        signature.addInt(x);
+      }
+    }
+    if (this._unitUnitUris == null) {
+      signature.addInt(0);
+    } else {
+      signature.addInt(this._unitUnitUris.length);
+      for (var x in this._unitUnitUris) {
+        signature.addInt(x);
+      }
+    }
+    if (this._units == null) {
+      signature.addInt(0);
+    } else {
+      signature.addInt(this._units.length);
+      for (var x in this._units) {
+        x?.collectApiSignature(signature);
+      }
+    }
+    if (this._elementKinds == null) {
+      signature.addInt(0);
+    } else {
+      signature.addInt(this._elementKinds.length);
+      for (var x in this._elementKinds) {
+        signature.addInt(x.index);
+      }
+    }
+    if (this._strings == null) {
+      signature.addInt(0);
+    } else {
+      signature.addInt(this._strings.length);
+      for (var x in this._strings) {
+        signature.addString(x);
+      }
+    }
+    if (this._elementNameClassMemberIds == null) {
+      signature.addInt(0);
+    } else {
+      signature.addInt(this._elementNameClassMemberIds.length);
+      for (var x in this._elementNameClassMemberIds) {
+        signature.addInt(x);
+      }
+    }
+    if (this._elementNameParameterIds == null) {
+      signature.addInt(0);
+    } else {
+      signature.addInt(this._elementNameParameterIds.length);
+      for (var x in this._elementNameParameterIds) {
+        signature.addInt(x);
+      }
+    }
+  }
+
   List<int> toBuffer() {
     fb.Builder fbBuilder = new fb.Builder();
     return fbBuilder.finish(finish(fbBuilder), "Indx");
@@ -2656,6 +3021,109 @@
   void flushInformative() {
   }
 
+  /**
+   * Accumulate non-[informative] data into [signature].
+   */
+  void collectApiSignature(api_sig.ApiSignature signature) {
+    signature.addInt(this._unit ?? 0);
+    if (this._usedElementLengths == null) {
+      signature.addInt(0);
+    } else {
+      signature.addInt(this._usedElementLengths.length);
+      for (var x in this._usedElementLengths) {
+        signature.addInt(x);
+      }
+    }
+    if (this._usedElementOffsets == null) {
+      signature.addInt(0);
+    } else {
+      signature.addInt(this._usedElementOffsets.length);
+      for (var x in this._usedElementOffsets) {
+        signature.addInt(x);
+      }
+    }
+    if (this._usedElements == null) {
+      signature.addInt(0);
+    } else {
+      signature.addInt(this._usedElements.length);
+      for (var x in this._usedElements) {
+        signature.addInt(x);
+      }
+    }
+    if (this._usedElementKinds == null) {
+      signature.addInt(0);
+    } else {
+      signature.addInt(this._usedElementKinds.length);
+      for (var x in this._usedElementKinds) {
+        signature.addInt(x.index);
+      }
+    }
+    if (this._definedNames == null) {
+      signature.addInt(0);
+    } else {
+      signature.addInt(this._definedNames.length);
+      for (var x in this._definedNames) {
+        signature.addInt(x);
+      }
+    }
+    if (this._definedNameKinds == null) {
+      signature.addInt(0);
+    } else {
+      signature.addInt(this._definedNameKinds.length);
+      for (var x in this._definedNameKinds) {
+        signature.addInt(x.index);
+      }
+    }
+    if (this._definedNameOffsets == null) {
+      signature.addInt(0);
+    } else {
+      signature.addInt(this._definedNameOffsets.length);
+      for (var x in this._definedNameOffsets) {
+        signature.addInt(x);
+      }
+    }
+    if (this._usedNames == null) {
+      signature.addInt(0);
+    } else {
+      signature.addInt(this._usedNames.length);
+      for (var x in this._usedNames) {
+        signature.addInt(x);
+      }
+    }
+    if (this._usedNameOffsets == null) {
+      signature.addInt(0);
+    } else {
+      signature.addInt(this._usedNameOffsets.length);
+      for (var x in this._usedNameOffsets) {
+        signature.addInt(x);
+      }
+    }
+    if (this._usedNameKinds == null) {
+      signature.addInt(0);
+    } else {
+      signature.addInt(this._usedNameKinds.length);
+      for (var x in this._usedNameKinds) {
+        signature.addInt(x.index);
+      }
+    }
+    if (this._usedElementIsQualifiedFlags == null) {
+      signature.addInt(0);
+    } else {
+      signature.addInt(this._usedElementIsQualifiedFlags.length);
+      for (var x in this._usedElementIsQualifiedFlags) {
+        signature.addBool(x);
+      }
+    }
+    if (this._usedNameIsQualifiedFlags == null) {
+      signature.addInt(0);
+    } else {
+      signature.addInt(this._usedNameIsQualifiedFlags.length);
+      for (var x in this._usedNameIsQualifiedFlags) {
+        signature.addBool(x);
+      }
+    }
+  }
+
   fb.Offset finish(fb.Builder fbBuilder) {
     fb.Offset offset_definedNameKinds;
     fb.Offset offset_definedNameOffsets;
@@ -3089,6 +3557,66 @@
     _typeParameters?.forEach((b) => b.flushInformative());
   }
 
+  /**
+   * Accumulate non-[informative] data into [signature].
+   */
+  void collectApiSignature(api_sig.ApiSignature signature) {
+    signature.addString(this._name ?? '');
+    if (this._executables == null) {
+      signature.addInt(0);
+    } else {
+      signature.addInt(this._executables.length);
+      for (var x in this._executables) {
+        x?.collectApiSignature(signature);
+      }
+    }
+    signature.addBool(this._supertype != null);
+    this._supertype?.collectApiSignature(signature);
+    if (this._fields == null) {
+      signature.addInt(0);
+    } else {
+      signature.addInt(this._fields.length);
+      for (var x in this._fields) {
+        x?.collectApiSignature(signature);
+      }
+    }
+    if (this._annotations == null) {
+      signature.addInt(0);
+    } else {
+      signature.addInt(this._annotations.length);
+      for (var x in this._annotations) {
+        x?.collectApiSignature(signature);
+      }
+    }
+    if (this._interfaces == null) {
+      signature.addInt(0);
+    } else {
+      signature.addInt(this._interfaces.length);
+      for (var x in this._interfaces) {
+        x?.collectApiSignature(signature);
+      }
+    }
+    signature.addBool(this._isAbstract == true);
+    if (this._typeParameters == null) {
+      signature.addInt(0);
+    } else {
+      signature.addInt(this._typeParameters.length);
+      for (var x in this._typeParameters) {
+        x?.collectApiSignature(signature);
+      }
+    }
+    if (this._mixins == null) {
+      signature.addInt(0);
+    } else {
+      signature.addInt(this._mixins.length);
+      for (var x in this._mixins) {
+        x?.collectApiSignature(signature);
+      }
+    }
+    signature.addBool(this._isMixinApplication == true);
+    signature.addBool(this._hasNoSupertype == true);
+  }
+
   fb.Offset finish(fb.Builder fbBuilder) {
     fb.Offset offset_annotations;
     fb.Offset offset_codeRange;
@@ -3397,6 +3925,28 @@
     _offset = null;
   }
 
+  /**
+   * Accumulate non-[informative] data into [signature].
+   */
+  void collectApiSignature(api_sig.ApiSignature signature) {
+    if (this._shows == null) {
+      signature.addInt(0);
+    } else {
+      signature.addInt(this._shows.length);
+      for (var x in this._shows) {
+        signature.addString(x);
+      }
+    }
+    if (this._hides == null) {
+      signature.addInt(0);
+    } else {
+      signature.addInt(this._hides.length);
+      for (var x in this._hides) {
+        signature.addString(x);
+      }
+    }
+  }
+
   fb.Offset finish(fb.Builder fbBuilder) {
     fb.Offset offset_hides;
     fb.Offset offset_shows;
@@ -3593,6 +4143,61 @@
     _references?.forEach((b) => b.flushInformative());
   }
 
+  /**
+   * Accumulate non-[informative] data into [signature].
+   */
+  void collectApiSignature(api_sig.ApiSignature signature) {
+    if (this._operations == null) {
+      signature.addInt(0);
+    } else {
+      signature.addInt(this._operations.length);
+      for (var x in this._operations) {
+        signature.addInt(x.index);
+      }
+    }
+    if (this._ints == null) {
+      signature.addInt(0);
+    } else {
+      signature.addInt(this._ints.length);
+      for (var x in this._ints) {
+        signature.addInt(x);
+      }
+    }
+    if (this._references == null) {
+      signature.addInt(0);
+    } else {
+      signature.addInt(this._references.length);
+      for (var x in this._references) {
+        x?.collectApiSignature(signature);
+      }
+    }
+    if (this._strings == null) {
+      signature.addInt(0);
+    } else {
+      signature.addInt(this._strings.length);
+      for (var x in this._strings) {
+        signature.addString(x);
+      }
+    }
+    if (this._doubles == null) {
+      signature.addInt(0);
+    } else {
+      signature.addInt(this._doubles.length);
+      for (var x in this._doubles) {
+        signature.addDouble(x);
+      }
+    }
+    signature.addBool(this._isValidConst == true);
+    if (this._assignmentOperators == null) {
+      signature.addInt(0);
+    } else {
+      signature.addInt(this._assignmentOperators.length);
+      for (var x in this._assignmentOperators) {
+        signature.addInt(x.index);
+      }
+    }
+  }
+
   fb.Offset finish(fb.Builder fbBuilder) {
     fb.Offset offset_assignmentOperators;
     fb.Offset offset_doubles;
@@ -3816,6 +4421,32 @@
     _expression?.flushInformative();
   }
 
+  /**
+   * Accumulate non-[informative] data into [signature].
+   */
+  void collectApiSignature(api_sig.ApiSignature signature) {
+    signature.addString(this._name ?? '');
+    signature.addBool(this._expression != null);
+    this._expression?.collectApiSignature(signature);
+    signature.addInt(this._kind == null ? 0 : this._kind.index);
+    if (this._arguments == null) {
+      signature.addInt(0);
+    } else {
+      signature.addInt(this._arguments.length);
+      for (var x in this._arguments) {
+        x?.collectApiSignature(signature);
+      }
+    }
+    if (this._argumentNames == null) {
+      signature.addInt(0);
+    } else {
+      signature.addInt(this._argumentNames.length);
+      for (var x in this._argumentNames) {
+        signature.addString(x);
+      }
+    }
+  }
+
   fb.Offset finish(fb.Builder fbBuilder) {
     fb.Offset offset_argumentNames;
     fb.Offset offset_arguments;
@@ -3980,6 +4611,15 @@
   void flushInformative() {
   }
 
+  /**
+   * Accumulate non-[informative] data into [signature].
+   */
+  void collectApiSignature(api_sig.ApiSignature signature) {
+    signature.addInt(this._length ?? 0);
+    signature.addString(this._text ?? '');
+    signature.addInt(this._offset ?? 0);
+  }
+
   fb.Offset finish(fb.Builder fbBuilder) {
     fb.Offset offset_text;
     if (_text != null) {
@@ -4145,6 +4785,29 @@
     _values?.forEach((b) => b.flushInformative());
   }
 
+  /**
+   * Accumulate non-[informative] data into [signature].
+   */
+  void collectApiSignature(api_sig.ApiSignature signature) {
+    signature.addString(this._name ?? '');
+    if (this._values == null) {
+      signature.addInt(0);
+    } else {
+      signature.addInt(this._values.length);
+      for (var x in this._values) {
+        x?.collectApiSignature(signature);
+      }
+    }
+    if (this._annotations == null) {
+      signature.addInt(0);
+    } else {
+      signature.addInt(this._annotations.length);
+      for (var x in this._annotations) {
+        x?.collectApiSignature(signature);
+      }
+    }
+  }
+
   fb.Offset finish(fb.Builder fbBuilder) {
     fb.Offset offset_annotations;
     fb.Offset offset_codeRange;
@@ -4323,6 +4986,13 @@
     _nameOffset = null;
   }
 
+  /**
+   * Accumulate non-[informative] data into [signature].
+   */
+  void collectApiSignature(api_sig.ApiSignature signature) {
+    signature.addString(this._name ?? '');
+  }
+
   fb.Offset finish(fb.Builder fbBuilder) {
     fb.Offset offset_documentationComment;
     fb.Offset offset_name;
@@ -4832,6 +5502,71 @@
     _typeParameters?.forEach((b) => b.flushInformative());
   }
 
+  /**
+   * Accumulate non-[informative] data into [signature].
+   */
+  void collectApiSignature(api_sig.ApiSignature signature) {
+    signature.addString(this._name ?? '');
+    if (this._parameters == null) {
+      signature.addInt(0);
+    } else {
+      signature.addInt(this._parameters.length);
+      for (var x in this._parameters) {
+        x?.collectApiSignature(signature);
+      }
+    }
+    signature.addBool(this._returnType != null);
+    this._returnType?.collectApiSignature(signature);
+    signature.addInt(this._kind == null ? 0 : this._kind.index);
+    signature.addInt(this._inferredReturnTypeSlot ?? 0);
+    if (this._annotations == null) {
+      signature.addInt(0);
+    } else {
+      signature.addInt(this._annotations.length);
+      for (var x in this._annotations) {
+        x?.collectApiSignature(signature);
+      }
+    }
+    signature.addBool(this._isFactory == true);
+    signature.addBool(this._isStatic == true);
+    signature.addBool(this._isAbstract == true);
+    signature.addBool(this._isExternal == true);
+    signature.addBool(this._isConst == true);
+    signature.addBool(this._isRedirectedConstructor == true);
+    if (this._constantInitializers == null) {
+      signature.addInt(0);
+    } else {
+      signature.addInt(this._constantInitializers.length);
+      for (var x in this._constantInitializers) {
+        x?.collectApiSignature(signature);
+      }
+    }
+    signature.addBool(this._redirectedConstructor != null);
+    this._redirectedConstructor?.collectApiSignature(signature);
+    if (this._typeParameters == null) {
+      signature.addInt(0);
+    } else {
+      signature.addInt(this._typeParameters.length);
+      for (var x in this._typeParameters) {
+        x?.collectApiSignature(signature);
+      }
+    }
+    signature.addString(this._redirectedConstructorName ?? '');
+    if (this._localFunctions == null) {
+      signature.addInt(0);
+    } else {
+      signature.addInt(this._localFunctions.length);
+      for (var x in this._localFunctions) {
+        x?.collectApiSignature(signature);
+      }
+    }
+    signature.addInt(this._visibleLength ?? 0);
+    signature.addInt(this._visibleOffset ?? 0);
+    signature.addInt(this._constCycleSlot ?? 0);
+    signature.addBool(this._bodyExpr != null);
+    this._bodyExpr?.collectApiSignature(signature);
+  }
+
   fb.Offset finish(fb.Builder fbBuilder) {
     fb.Offset offset_annotations;
     fb.Offset offset_bodyExpr;
@@ -5351,6 +6086,20 @@
     _uriOffset = null;
   }
 
+  /**
+   * Accumulate non-[informative] data into [signature].
+   */
+  void collectApiSignature(api_sig.ApiSignature signature) {
+    if (this._annotations == null) {
+      signature.addInt(0);
+    } else {
+      signature.addInt(this._annotations.length);
+      for (var x in this._annotations) {
+        x?.collectApiSignature(signature);
+      }
+    }
+  }
+
   fb.Offset finish(fb.Builder fbBuilder) {
     fb.Offset offset_annotations;
     if (!(_annotations == null || _annotations.isEmpty)) {
@@ -5474,6 +6223,21 @@
     _combinators?.forEach((b) => b.flushInformative());
   }
 
+  /**
+   * Accumulate non-[informative] data into [signature].
+   */
+  void collectApiSignature(api_sig.ApiSignature signature) {
+    signature.addString(this._uri ?? '');
+    if (this._combinators == null) {
+      signature.addInt(0);
+    } else {
+      signature.addInt(this._combinators.length);
+      for (var x in this._combinators) {
+        x?.collectApiSignature(signature);
+      }
+    }
+  }
+
   fb.Offset finish(fb.Builder fbBuilder) {
     fb.Offset offset_combinators;
     fb.Offset offset_uri;
@@ -5690,6 +6454,32 @@
     _uriOffset = null;
   }
 
+  /**
+   * Accumulate non-[informative] data into [signature].
+   */
+  void collectApiSignature(api_sig.ApiSignature signature) {
+    signature.addString(this._uri ?? '');
+    if (this._combinators == null) {
+      signature.addInt(0);
+    } else {
+      signature.addInt(this._combinators.length);
+      for (var x in this._combinators) {
+        x?.collectApiSignature(signature);
+      }
+    }
+    signature.addBool(this._isImplicit == true);
+    signature.addInt(this._prefixReference ?? 0);
+    if (this._annotations == null) {
+      signature.addInt(0);
+    } else {
+      signature.addInt(this._annotations.length);
+      for (var x in this._annotations) {
+        x?.collectApiSignature(signature);
+      }
+    }
+    signature.addBool(this._isDeferred == true);
+  }
+
   fb.Offset finish(fb.Builder fbBuilder) {
     fb.Offset offset_annotations;
     fb.Offset offset_combinators;
@@ -5919,6 +6709,15 @@
     _nameOffset = null;
   }
 
+  /**
+   * Accumulate non-[informative] data into [signature].
+   */
+  void collectApiSignature(api_sig.ApiSignature signature) {
+    signature.addString(this._name ?? '');
+    signature.addBool(this._isOnSwitchMember == true);
+    signature.addBool(this._isOnSwitchStatement == true);
+  }
+
   fb.Offset finish(fb.Builder fbBuilder) {
     fb.Offset offset_name;
     if (_name != null) {
@@ -6209,6 +7008,39 @@
     _type?.flushInformative();
   }
 
+  /**
+   * Accumulate non-[informative] data into [signature].
+   */
+  void collectApiSignature(api_sig.ApiSignature signature) {
+    signature.addString(this._name ?? '');
+    signature.addInt(this._inferredTypeSlot ?? 0);
+    signature.addBool(this._type != null);
+    this._type?.collectApiSignature(signature);
+    signature.addInt(this._kind == null ? 0 : this._kind.index);
+    signature.addBool(this._isFunctionTyped == true);
+    signature.addBool(this._isInitializingFormal == true);
+    if (this._parameters == null) {
+      signature.addInt(0);
+    } else {
+      signature.addInt(this._parameters.length);
+      for (var x in this._parameters) {
+        x?.collectApiSignature(signature);
+      }
+    }
+    if (this._annotations == null) {
+      signature.addInt(0);
+    } else {
+      signature.addInt(this._annotations.length);
+      for (var x in this._annotations) {
+        x?.collectApiSignature(signature);
+      }
+    }
+    signature.addInt(this._visibleLength ?? 0);
+    signature.addInt(this._visibleOffset ?? 0);
+    signature.addBool(this._initializer != null);
+    this._initializer?.collectApiSignature(signature);
+  }
+
   fb.Offset finish(fb.Builder fbBuilder) {
     fb.Offset offset_annotations;
     fb.Offset offset_codeRange;
@@ -6494,6 +7326,20 @@
     _uriOffset = null;
   }
 
+  /**
+   * Accumulate non-[informative] data into [signature].
+   */
+  void collectApiSignature(api_sig.ApiSignature signature) {
+    if (this._annotations == null) {
+      signature.addInt(0);
+    } else {
+      signature.addInt(this._annotations.length);
+      for (var x in this._annotations) {
+        x?.collectApiSignature(signature);
+      }
+    }
+  }
+
   fb.Offset finish(fb.Builder fbBuilder) {
     fb.Offset offset_annotations;
     if (!(_annotations == null || _annotations.isEmpty)) {
@@ -6636,6 +7482,23 @@
     _members?.forEach((b) => b.flushInformative());
   }
 
+  /**
+   * Accumulate non-[informative] data into [signature].
+   */
+  void collectApiSignature(api_sig.ApiSignature signature) {
+    signature.addString(this._name ?? '');
+    signature.addInt(this._kind == null ? 0 : this._kind.index);
+    if (this._members == null) {
+      signature.addInt(0);
+    } else {
+      signature.addInt(this._members.length);
+      for (var x in this._members) {
+        x?.collectApiSignature(signature);
+      }
+    }
+    signature.addInt(this._numTypeParameters ?? 0);
+  }
+
   fb.Offset finish(fb.Builder fbBuilder) {
     fb.Offset offset_members;
     fb.Offset offset_name;
@@ -6779,6 +7642,36 @@
     _names?.forEach((b) => b.flushInformative());
   }
 
+  /**
+   * Accumulate non-[informative] data into [signature].
+   */
+  void collectApiSignature(api_sig.ApiSignature signature) {
+    if (this._names == null) {
+      signature.addInt(0);
+    } else {
+      signature.addInt(this._names.length);
+      for (var x in this._names) {
+        x?.collectApiSignature(signature);
+      }
+    }
+    if (this._parts == null) {
+      signature.addInt(0);
+    } else {
+      signature.addInt(this._parts.length);
+      for (var x in this._parts) {
+        signature.addString(x);
+      }
+    }
+    if (this._exports == null) {
+      signature.addInt(0);
+    } else {
+      signature.addInt(this._exports.length);
+      for (var x in this._exports) {
+        x?.collectApiSignature(signature);
+      }
+    }
+  }
+
   List<int> toBuffer() {
     fb.Builder fbBuilder = new fb.Builder();
     return fbBuilder.finish(finish(fbBuilder), "UPNS");
@@ -6915,6 +7808,14 @@
   void flushInformative() {
   }
 
+  /**
+   * Accumulate non-[informative] data into [signature].
+   */
+  void collectApiSignature(api_sig.ApiSignature signature) {
+    signature.addString(this._name ?? '');
+    signature.addInt(this._prefixReference ?? 0);
+  }
+
   fb.Offset finish(fb.Builder fbBuilder) {
     fb.Offset offset_name;
     if (_name != null) {
@@ -7094,6 +7995,39 @@
     _typeParameters?.forEach((b) => b.flushInformative());
   }
 
+  /**
+   * Accumulate non-[informative] data into [signature].
+   */
+  void collectApiSignature(api_sig.ApiSignature signature) {
+    signature.addString(this._name ?? '');
+    signature.addBool(this._returnType != null);
+    this._returnType?.collectApiSignature(signature);
+    if (this._parameters == null) {
+      signature.addInt(0);
+    } else {
+      signature.addInt(this._parameters.length);
+      for (var x in this._parameters) {
+        x?.collectApiSignature(signature);
+      }
+    }
+    if (this._annotations == null) {
+      signature.addInt(0);
+    } else {
+      signature.addInt(this._annotations.length);
+      for (var x in this._annotations) {
+        x?.collectApiSignature(signature);
+      }
+    }
+    if (this._typeParameters == null) {
+      signature.addInt(0);
+    } else {
+      signature.addInt(this._typeParameters.length);
+      for (var x in this._typeParameters) {
+        x?.collectApiSignature(signature);
+      }
+    }
+  }
+
   fb.Offset finish(fb.Builder fbBuilder) {
     fb.Offset offset_annotations;
     fb.Offset offset_codeRange;
@@ -7330,6 +8264,23 @@
     _nameOffset = null;
   }
 
+  /**
+   * Accumulate non-[informative] data into [signature].
+   */
+  void collectApiSignature(api_sig.ApiSignature signature) {
+    signature.addString(this._name ?? '');
+    signature.addBool(this._bound != null);
+    this._bound?.collectApiSignature(signature);
+    if (this._annotations == null) {
+      signature.addInt(0);
+    } else {
+      signature.addInt(this._annotations.length);
+      for (var x in this._annotations) {
+        x?.collectApiSignature(signature);
+      }
+    }
+  }
+
   fb.Offset finish(fb.Builder fbBuilder) {
     fb.Offset offset_annotations;
     fb.Offset offset_bound;
@@ -7686,6 +8637,96 @@
     _variables?.forEach((b) => b.flushInformative());
   }
 
+  /**
+   * Accumulate non-[informative] data into [signature].
+   */
+  void collectApiSignature(api_sig.ApiSignature signature) {
+    signature.addBool(this._publicNamespace != null);
+    this._publicNamespace?.collectApiSignature(signature);
+    if (this._references == null) {
+      signature.addInt(0);
+    } else {
+      signature.addInt(this._references.length);
+      for (var x in this._references) {
+        x?.collectApiSignature(signature);
+      }
+    }
+    if (this._classes == null) {
+      signature.addInt(0);
+    } else {
+      signature.addInt(this._classes.length);
+      for (var x in this._classes) {
+        x?.collectApiSignature(signature);
+      }
+    }
+    if (this._variables == null) {
+      signature.addInt(0);
+    } else {
+      signature.addInt(this._variables.length);
+      for (var x in this._variables) {
+        x?.collectApiSignature(signature);
+      }
+    }
+    if (this._executables == null) {
+      signature.addInt(0);
+    } else {
+      signature.addInt(this._executables.length);
+      for (var x in this._executables) {
+        x?.collectApiSignature(signature);
+      }
+    }
+    if (this._imports == null) {
+      signature.addInt(0);
+    } else {
+      signature.addInt(this._imports.length);
+      for (var x in this._imports) {
+        x?.collectApiSignature(signature);
+      }
+    }
+    signature.addString(this._libraryName ?? '');
+    if (this._typedefs == null) {
+      signature.addInt(0);
+    } else {
+      signature.addInt(this._typedefs.length);
+      for (var x in this._typedefs) {
+        x?.collectApiSignature(signature);
+      }
+    }
+    if (this._parts == null) {
+      signature.addInt(0);
+    } else {
+      signature.addInt(this._parts.length);
+      for (var x in this._parts) {
+        x?.collectApiSignature(signature);
+      }
+    }
+    if (this._enums == null) {
+      signature.addInt(0);
+    } else {
+      signature.addInt(this._enums.length);
+      for (var x in this._enums) {
+        x?.collectApiSignature(signature);
+      }
+    }
+    if (this._exports == null) {
+      signature.addInt(0);
+    } else {
+      signature.addInt(this._exports.length);
+      for (var x in this._exports) {
+        x?.collectApiSignature(signature);
+      }
+    }
+    if (this._libraryAnnotations == null) {
+      signature.addInt(0);
+    } else {
+      signature.addInt(this._libraryAnnotations.length);
+      for (var x in this._libraryAnnotations) {
+        x?.collectApiSignature(signature);
+      }
+    }
+    signature.addString(this._fallbackModePath ?? '');
+  }
+
   List<int> toBuffer() {
     fb.Builder fbBuilder = new fb.Builder();
     return fbBuilder.finish(finish(fbBuilder), "UUnt");
@@ -8199,6 +9240,32 @@
     _type?.flushInformative();
   }
 
+  /**
+   * Accumulate non-[informative] data into [signature].
+   */
+  void collectApiSignature(api_sig.ApiSignature signature) {
+    signature.addString(this._name ?? '');
+    signature.addInt(this._propagatedTypeSlot ?? 0);
+    signature.addBool(this._type != null);
+    this._type?.collectApiSignature(signature);
+    signature.addBool(this._isStatic == true);
+    signature.addBool(this._isConst == true);
+    signature.addBool(this._isFinal == true);
+    if (this._annotations == null) {
+      signature.addInt(0);
+    } else {
+      signature.addInt(this._annotations.length);
+      for (var x in this._annotations) {
+        x?.collectApiSignature(signature);
+      }
+    }
+    signature.addInt(this._inferredTypeSlot ?? 0);
+    signature.addInt(this._visibleLength ?? 0);
+    signature.addInt(this._visibleOffset ?? 0);
+    signature.addBool(this._initializer != null);
+    this._initializer?.collectApiSignature(signature);
+  }
+
   fb.Offset finish(fb.Builder fbBuilder) {
     fb.Offset offset_annotations;
     fb.Offset offset_codeRange;

pkg/analyzer/lib/src/summary/format.fbs

diff --git a/pkg/analyzer/lib/src/summary/format.fbs b/pkg/analyzer/lib/src/summary/format.fbs
index 329fe6b..1449743 100644
--- a/pkg/analyzer/lib/src/summary/format.fbs
+++ b/pkg/analyzer/lib/src/summary/format.fbs
@@ -1181,6 +1181,13 @@
  */
 table PackageBundle {
   /**
+   * MD5 hash of the non-informative fields of the [PackageBundle] (not
+   * including this one).  This can be used to identify when the API of a
+   * package may have changed.
+   */
+  apiSignature:string (id: 7);
+
+  /**
    * Linked libraries.
    */
   linkedLibraries:[LinkedLibrary] (id: 0);

pkg/analyzer/lib/src/summary/idl.dart

diff --git a/pkg/analyzer/lib/src/summary/idl.dart b/pkg/analyzer/lib/src/summary/idl.dart
index 60a0552..15f81a7 100644
--- a/pkg/analyzer/lib/src/summary/idl.dart
+++ b/pkg/analyzer/lib/src/summary/idl.dart
@@ -650,6 +650,14 @@
       generated.readPackageBundle(buffer);
 
   /**
+   * MD5 hash of the non-informative fields of the [PackageBundle] (not
+   * including this one).  This can be used to identify when the API of a
+   * package may have changed.
+   */
+  @Id(7)
+  String get apiSignature;
+
+  /**
    * Linked libraries.
    */
   @Id(0)

pkg/analyzer/lib/src/summary/summarize_elements.dart

diff --git a/pkg/analyzer/lib/src/summary/summarize_elements.dart b/pkg/analyzer/lib/src/summary/summarize_elements.dart
index 98e3de4..e25da27 100644
--- a/pkg/analyzer/lib/src/summary/summarize_elements.dart
+++ b/pkg/analyzer/lib/src/summary/summarize_elements.dart
@@ -15,6 +15,7 @@
 import 'package:analyzer/src/generated/resolver.dart';
 import 'package:analyzer/src/generated/source.dart';
 import 'package:analyzer/src/generated/utilities_dart.dart';
+import 'package:analyzer/src/summary/api_signature.dart';
 import 'package:analyzer/src/summary/format.dart';
 import 'package:analyzer/src/summary/idl.dart';
 import 'package:analyzer/src/summary/name_filter.dart';
@@ -192,7 +193,7 @@
    * Assemble a new [PackageBundleBuilder] using the gathered information.
    */
   PackageBundleBuilder assemble() {
-    return new PackageBundleBuilder(
+    PackageBundleBuilder packageBundle = new PackageBundleBuilder(
         linkedLibraryUris: _linkedLibraryUris,
         linkedLibraries: _linkedLibraries,
         unlinkedUnitUris: _unlinkedUnitUris,
@@ -200,6 +201,10 @@
         unlinkedUnitHashes: _unlinkedUnitHashes,
         majorVersion: currentMajorVersion,
         minorVersion: currentMinorVersion);
+    ApiSignature apiSignature = new ApiSignature();
+    packageBundle.collectApiSignature(apiSignature);
+    packageBundle.apiSignature = apiSignature.toHex();
+    return packageBundle;
   }
 
   /**

pkg/analyzer/lib/src/task/strong/checker.dart

diff --git a/pkg/analyzer/lib/src/task/strong/checker.dart b/pkg/analyzer/lib/src/task/strong/checker.dart
index c074b79..b20d702 100644
--- a/pkg/analyzer/lib/src/task/strong/checker.dart
+++ b/pkg/analyzer/lib/src/task/strong/checker.dart
@@ -21,8 +21,11 @@
 
 import 'ast_properties.dart';
 
-bool isKnownFunction(Expression expression) {
+bool isKnownFunction(Expression expression, {bool instanceMethods: false}) {
   Element element = null;
+  if (expression is ParenthesizedExpression) {
+    expression = (expression as ParenthesizedExpression).expression;
+  }
   if (expression is FunctionExpression) {
     return true;
   } else if (expression is PropertyAccess) {
@@ -30,10 +33,10 @@
   } else if (expression is Identifier) {
     element = expression.staticElement;
   }
-// First class functions and static methods, where we know the original
-// declaration, will have an exact type, so we know a downcast will fail.
+  // First class functions and static methods, where we know the original
+  // declaration, will have an exact type, so we know a downcast will fail.
   return element is FunctionElement ||
-      element is MethodElement && element.isStatic;
+      element is MethodElement && (instanceMethods || element.isStatic);
 }
 
 DartType _elementType(Element e) {
@@ -162,8 +165,9 @@
     if (expr is ParenthesizedExpression) {
       checkAssignment(expr.expression, type);
     } else {
-      _checkDowncast(expr, type);
-      _checkNonNullAssignment(expr, type);
+      if (_checkNonNullAssignment(expr, type)) {
+        _checkDowncast(expr, type);
+      }
     }
   }
 
@@ -625,6 +629,20 @@
   }
 
   @override
+  Object visitVariableDeclaration(VariableDeclaration node) {
+    if (!node.isConst &&
+        !node.isFinal &&
+        node.initializer == null &&
+        rules.isNonNullableType(node?.element?.type)) {
+      _recordMessage(
+          node,
+          StaticTypeWarningCode.NON_NULLABLE_FIELD_NOT_INITIALIZED,
+          [node.name, node?.element?.type]);
+    }
+    return super.visitVariableDeclaration(node);
+  }
+
+  @override
   void visitWhileStatement(WhileStatement node) {
     checkBoolean(node.condition);
     node.visitChildren(this);
@@ -839,12 +857,17 @@
         [declElement.name]);
   }
 
-  void _checkNonNullAssignment(Expression expression, DartType type) {
+  /// Checks if the assignment is valid with respect to non-nullable types.
+  /// Returns `false` if a nullable expression is assigned to a variable of
+  /// non-nullable type and `true` otherwise.
+  bool _checkNonNullAssignment(Expression expression, DartType type) {
     var exprType = expression.staticType;
-    if (rules.isNonNullableType(type) && !rules.isNonNullableType(exprType)) {
+    if (rules.isNonNullableType(type) && rules.isNullableType(exprType)) {
       _recordMessage(expression, StaticTypeWarningCode.INVALID_ASSIGNMENT,
           [exprType, type]);
+      return false;
     }
+    return true;
   }
 
   void _checkReturnOrYield(Expression expression, AstNode node,
@@ -936,10 +959,6 @@
     }
   }
 
-  // Produce a coercion which coerces something of type fromT
-  // to something of type toT.
-  // Returns the error coercion if the types cannot be coerced
-  // according to our current criteria.
   DartType _getStaticType(Expression expr) {
     DartType t = expr.staticType ?? DynamicTypeImpl.instance;
 
@@ -985,14 +1004,9 @@
     // a dynamic parameter type requires a dynamic call in general.
     // However, as an optimization, if we have an original definition, we know
     // dynamic is reified as Object - in this case a regular call is fine.
-    if (call is SimpleIdentifier) {
-      var element = call.staticElement;
-      if (element is FunctionElement || element is MethodElement) {
-        // An original declaration.
-        return false;
-      }
+    if (isKnownFunction(call, instanceMethods: true)) {
+      return false;
     }
-
     return rules.anyParameterType(ft, (pt) => pt.isDynamic);
   }
 

pkg/analyzer/pubspec.yaml

diff --git a/pkg/analyzer/pubspec.yaml b/pkg/analyzer/pubspec.yaml
index 4779ab9..1988ae1 100644
--- a/pkg/analyzer/pubspec.yaml
+++ b/pkg/analyzer/pubspec.yaml
@@ -1,5 +1,5 @@
 name: analyzer
-version: 0.27.4
+version: 0.27.5-alpha.0
 author: Dart Team <misc@dartlang.org>
 description: Static analyzer for Dart.
 homepage: https://github.com/dart-lang/sdk/tree/master/pkg/analyzer

pkg/analyzer/test/generated/engine_test.dart

diff --git a/pkg/analyzer/test/generated/engine_test.dart b/pkg/analyzer/test/generated/engine_test.dart
index 1d0871d..92c7007 100644
--- a/pkg/analyzer/test/generated/engine_test.dart
+++ b/pkg/analyzer/test/generated/engine_test.dart
@@ -9,8 +9,8 @@
 import 'package:analyzer/dart/ast/ast.dart';
 import 'package:analyzer/dart/element/element.dart';
 import 'package:analyzer/plugin/resolver_provider.dart';
-import 'package:analyzer/source/embedder.dart';
 import 'package:analyzer/src/cancelable_future.dart';
+import 'package:analyzer/src/context/builder.dart' show EmbedderYamlLocator;
 import 'package:analyzer/src/context/cache.dart';
 import 'package:analyzer/src/context/context.dart';
 import 'package:analyzer/src/context/source.dart';
@@ -225,6 +225,12 @@
   }
 
   @override
+  CacheConsistencyValidator get cacheConsistencyValidator {
+    fail("Unexpected invocation of cacheConsistencyValidator");
+    return null;
+  }
+
+  @override
   set contentCache(ContentCache value) {
     fail("Unexpected invocation of setContentCache");
   }
@@ -705,10 +711,4 @@
   void visitContentCache(ContentCacheVisitor visitor) {
     fail("Unexpected invocation of visitContentCache");
   }
-
-  @override
-  CacheConsistencyValidator get cacheConsistencyValidator {
-    fail("Unexpected invocation of cacheConsistencyValidator");
-    return null;
-  }
 }

pkg/analyzer/test/generated/incremental_resolver_test.dart

diff --git a/pkg/analyzer/test/generated/incremental_resolver_test.dart b/pkg/analyzer/test/generated/incremental_resolver_test.dart
index fa0ad49..3a06df4 100644
--- a/pkg/analyzer/test/generated/incremental_resolver_test.dart
+++ b/pkg/analyzer/test/generated/incremental_resolver_test.dart
@@ -1348,6 +1348,24 @@
 ''');
   }
 
+  void test_true_emptyLine_betweenClassMembers_insert_beforeComment() {
+    _resolveUnit(r'''
+class A {
+  a() {}
+  /// BBB
+  b() {}
+}
+''');
+    _updateAndValidate(r'''
+class A {
+  a() {}
+
+  /// BBB
+  b() {}
+}
+''');
+  }
+
   void test_true_emptyLine_betweenClassMembers_remove() {
     _resolveUnit(r'''
 class A {
@@ -1364,7 +1382,25 @@
 ''');
   }
 
-  void test_true_emptyLine_betweenCompilationUnitMembers_insert() {
+  void test_true_emptyLine_betweenClassMembers_remove_beforeComment() {
+    _resolveUnit(r'''
+class A {
+  a() {}
+
+  /// BBB
+  b() {}
+}
+''');
+    _updateAndValidate(r'''
+class A {
+  a() {}
+  /// BBB
+  b() {}
+}
+''');
+  }
+
+  void test_true_emptyLine_betweenUnitMembers_insert() {
     _resolveUnit(r'''
 a() {}
 b() {}
@@ -1376,7 +1412,23 @@
 ''');
   }
 
-  void test_true_emptyLine_betweenCompilationUnitMembers_remove() {
+  void test_true_emptyLine_betweenUnitMembers_insert_beforeComment() {
+    _resolveUnit(r'''
+a() {}
+
+// BBB
+b() {}
+''');
+    _updateAndValidate(r'''
+a() {}
+
+
+// BBB
+b() {}
+''');
+  }
+
+  void test_true_emptyLine_betweenUnitMembers_remove() {
     _resolveUnit(r'''
 a() {
   print(1)
@@ -1398,6 +1450,20 @@
 ''');
   }
 
+  void test_true_emptyLine_betweenUnitMembers_remove_beforeComment() {
+    _resolveUnit(r'''
+a() {}
+
+// BBB
+b() {}
+''');
+    _updateAndValidate(r'''
+a() {}
+// BBB
+b() {}
+''');
+  }
+
   void test_true_todoHint() {
     _resolveUnit(r'''
 main() {

pkg/analyzer/test/generated/strong_mode_test.dart

diff --git a/pkg/analyzer/test/generated/strong_mode_test.dart b/pkg/analyzer/test/generated/strong_mode_test.dart
index 75f4ea9..6c5879f 100644
--- a/pkg/analyzer/test/generated/strong_mode_test.dart
+++ b/pkg/analyzer/test/generated/strong_mode_test.dart
@@ -1745,7 +1745,7 @@
 }
 ''');
     expectIdentifierType('f', '<S>(S) → S');
-    expectIdentifierType('g', '<S>(S) → dynamic');
+    expectIdentifierType('g', '<S>(S) → <S>(S) → S');
   }
 
   void test_genericMethod_override() {

pkg/analyzer/test/generated/type_system_test.dart

diff --git a/pkg/analyzer/test/generated/type_system_test.dart b/pkg/analyzer/test/generated/type_system_test.dart
index d82d135..c62f985 100644
--- a/pkg/analyzer/test/generated/type_system_test.dart
+++ b/pkg/analyzer/test/generated/type_system_test.dart
@@ -164,7 +164,6 @@
     FunctionType expected = _functionType([objectType, numType, numType]);
     _checkLeastUpperBound(type1, type2, expected);
   }
-
   void test_nestedNestedFunctionsLubInnermostParamTypes() {
     FunctionType type1 = _functionType([
       _functionType([
@@ -1142,6 +1141,13 @@
     _checkGreatestLowerBound(type1, type2, expected);
   }
 
+  void test_functionsFuzzyArrows() {
+    FunctionType type1 = _functionType([dynamicType]);
+    FunctionType type2 = _functionType([intType]);
+    FunctionType expected = _functionType([intType]);
+    _checkGreatestLowerBound(type1, type2, expected);
+  }
+
   void test_functionsGlbReturnType() {
     FunctionType type1 = _functionType([], returns: intType);
     FunctionType type2 = _functionType([], returns: numType);
@@ -1311,6 +1317,13 @@
     super.setUp();
   }
 
+  void test_functionsFuzzyArrows() {
+    FunctionType type1 = _functionType([dynamicType]);
+    FunctionType type2 = _functionType([intType]);
+    FunctionType expected = _functionType([dynamicType]);
+    _checkLeastUpperBound(type1, type2, expected);
+  }
+
   void test_functionsGlbNamedParams() {
     FunctionType type1 =
         _functionType([], named: {'a': stringType, 'b': intType});

pkg/analyzer/test/source/embedder_test.dart

diff --git a/pkg/analyzer/test/source/embedder_test.dart b/pkg/analyzer/test/source/embedder_test.dart
index c16e6f9..ea20cd7 100644
--- a/pkg/analyzer/test/source/embedder_test.dart
+++ b/pkg/analyzer/test/source/embedder_test.dart
@@ -22,7 +22,6 @@
   runReflectiveTests(DartUriResolverTest);
   runReflectiveTests(EmbedderSdkTest);
   runReflectiveTests(EmbedderUriResolverTest);
-  runReflectiveTests(EmbedderYamlLocatorTest);
 }
 
 @reflectiveTest
@@ -39,6 +38,7 @@
       expect(source, isNotNull, reason: dartUri);
       expect(source.fullName, posixToOSPath(posixPath));
     }
+
     // Check that they map to the correct paths.
     expectResolved('dart:core', '/tmp/core.dart');
     expectResolved('dart:fox', '/tmp/slippy.dart');
@@ -205,26 +205,3 @@
     expectRestore('dart:deep/deep/file.dart');
   }
 }
-
-@reflectiveTest
-class EmbedderYamlLocatorTest extends EmbedderRelatedTest {
-  void test_empty() {
-    EmbedderYamlLocator locator = new EmbedderYamlLocator({
-      'fox': [pathTranslator.getResource('/empty')]
-    });
-    expect(locator.embedderYamls, hasLength(0));
-  }
-
-  void test_invalid() {
-    EmbedderYamlLocator locator = new EmbedderYamlLocator(null);
-    locator.addEmbedderYaml(null, r'''{{{,{{}}},}}''');
-    expect(locator.embedderYamls, hasLength(0));
-  }
-
-  void test_valid() {
-    EmbedderYamlLocator locator = new EmbedderYamlLocator({
-      'fox': [pathTranslator.getResource('/tmp')]
-    });
-    expect(locator.embedderYamls, hasLength(1));
-  }
-}

pkg/analyzer/test/src/context/builder_test.dart

diff --git a/pkg/analyzer/test/src/context/builder_test.dart b/pkg/analyzer/test/src/context/builder_test.dart
index 1f94afa..e6ed76f 100644
--- a/pkg/analyzer/test/src/context/builder_test.dart
+++ b/pkg/analyzer/test/src/context/builder_test.dart
@@ -22,6 +22,7 @@
 
 import '../../generated/test_support.dart';
 import '../../reflective_tests.dart';
+import '../../source/embedder_test.dart';
 import '../../utils.dart';
 import 'mock_sdk.dart';
 
@@ -29,6 +30,7 @@
   initializeTestEnvironment();
   runReflectiveTests(ContextBuilderTest_WithDisk);
   runReflectiveTests(ContextBuilderTest_WithoutDisk);
+  runReflectiveTests(EmbedderYamlLocatorTest);
 }
 
 @reflectiveTest
@@ -695,3 +697,26 @@
     expect(actual.finerGrainedInvalidation, expected.finerGrainedInvalidation);
   }
 }
+
+@reflectiveTest
+class EmbedderYamlLocatorTest extends EmbedderRelatedTest {
+  void test_empty() {
+    EmbedderYamlLocator locator = new EmbedderYamlLocator({
+      'fox': [pathTranslator.getResource('/empty')]
+    });
+    expect(locator.embedderYamls, hasLength(0));
+  }
+
+  void test_invalid() {
+    EmbedderYamlLocator locator = new EmbedderYamlLocator(null);
+    locator.addEmbedderYaml(null, r'''{{{,{{}}},}}''');
+    expect(locator.embedderYamls, hasLength(0));
+  }
+
+  void test_valid() {
+    EmbedderYamlLocator locator = new EmbedderYamlLocator({
+      'fox': [pathTranslator.getResource('/tmp')]
+    });
+    expect(locator.embedderYamls, hasLength(1));
+  }
+}

pkg/analyzer/test/src/context/context_test.dart

diff --git a/pkg/analyzer/test/src/context/context_test.dart b/pkg/analyzer/test/src/context/context_test.dart
index f5b695f..988db56 100644
--- a/pkg/analyzer/test/src/context/context_test.dart
+++ b/pkg/analyzer/test/src/context/context_test.dart
@@ -487,6 +487,43 @@
     expect(context.getErrors(a).errors, hasLength(0));
   }
 
+  void test_applyChanges_addNewImport_invalidateLibraryCycle() {
+    context.analysisOptions =
+        new AnalysisOptionsImpl.from(context.analysisOptions)
+          ..strongMode = true;
+    Source embedder = addSource(
+        '/a.dart',
+        r'''
+library a;
+import 'b.dart';
+//import 'c.dart';
+''');
+    addSource(
+        '/b.dart',
+        r'''
+library b;
+import 'a.dart';
+''');
+    addSource(
+        '/c.dart',
+        r'''
+library c;
+import 'b.dart';
+''');
+    _performPendingAnalysisTasks();
+    // Add a new import into a.dart, this should invalidate its library cycle.
+    // If it doesn't, we will get a task cycle exception.
+    context.setContents(
+        embedder,
+        r'''
+library a;
+import 'b.dart';
+import 'c.dart';
+''');
+    _performPendingAnalysisTasks();
+    expect(context.getCacheEntry(embedder).exception, isNull);
+  }
+
   void test_cacheConsistencyValidator_computed_deleted() {
     CacheConsistencyValidator validator = context.cacheConsistencyValidator;
     var stat = PerformanceStatistics.cacheConsistencyValidationStatistics;

pkg/analyzer/test/src/dart/sdk/sdk_test.dart

diff --git a/pkg/analyzer/test/src/dart/sdk/sdk_test.dart b/pkg/analyzer/test/src/dart/sdk/sdk_test.dart
index e613f96..a212045 100644
--- a/pkg/analyzer/test/src/dart/sdk/sdk_test.dart
+++ b/pkg/analyzer/test/src/dart/sdk/sdk_test.dart
@@ -6,7 +6,7 @@
 
 import 'package:analyzer/file_system/file_system.dart';
 import 'package:analyzer/file_system/memory_file_system.dart';
-import 'package:analyzer/source/embedder.dart' show EmbedderYamlLocator;
+import 'package:analyzer/src/context/builder.dart' show EmbedderYamlLocator;
 import 'package:analyzer/src/dart/sdk/sdk.dart';
 import 'package:analyzer/src/generated/engine.dart';
 import 'package:analyzer/src/generated/java_engine_io.dart';

pkg/analyzer/test/src/summary/api_signature_test.dart

diff --git a/pkg/analyzer/test/src/summary/api_signature_test.dart b/pkg/analyzer/test/src/summary/api_signature_test.dart
new file mode 100644
index 0000000..eabb805
--- /dev/null
+++ b/pkg/analyzer/test/src/summary/api_signature_test.dart
@@ -0,0 +1,112 @@
+// Copyright (c) 2016, the Dart project authors.  Please see the AUTHORS file
+// for details. All rights reserved. Use of this source code is governed by a
+// BSD-style license that can be found in the LICENSE file.
+
+library test.src.summary.api_signature_test;
+
+import 'package:analyzer/src/summary/api_signature.dart';
+import 'package:analyzer/src/summary/format.dart';
+import 'package:convert/convert.dart';
+import 'package:crypto/crypto.dart';
+import 'package:unittest/unittest.dart';
+
+import '../../reflective_tests.dart';
+
+main() {
+  groupSep = ' | ';
+  runReflectiveTests(ApiSignatureTest);
+}
+
+@reflectiveTest
+class ApiSignatureTest {
+  ApiSignature sig = new ApiSignature.unversioned();
+
+  void checkBytes(List<int> bytes) {
+    expect(sig.getBytes_forDebug(), bytes);
+    expect(sig.toHex(), hex.encode(md5.convert(bytes).bytes));
+  }
+
+  String signUnlinkedCombinator(UnlinkedCombinatorBuilder combinator) {
+    ApiSignature sig = new ApiSignature();
+    combinator.collectApiSignature(sig);
+    return sig.toHex();
+  }
+
+  void test_addBool() {
+    sig.addBool(true);
+    sig.addBool(true);
+    sig.addBool(false);
+    sig.addBool(true);
+    sig.addBool(false);
+    sig.addBool(false);
+    sig.addBool(true);
+    sig.addBool(false);
+    checkBytes([1, 1, 0, 1, 0, 0, 1, 0]);
+  }
+
+  void test_addBytes() {
+    // Check that offset works correctly by adding bytes in 2 chunks.
+    sig.addBytes([1, 2, 3, 4, 5]);
+    sig.addBytes([0xff, 0xfe, 0xfd, 0xfc, 0xfb]);
+    checkBytes([1, 2, 3, 4, 5, 0xff, 0xfe, 0xfd, 0xfc, 0xfb]);
+  }
+
+  void test_addDouble() {
+    sig.addDouble(1.0 / 3.0);
+    sig.addDouble(-1.0);
+    checkBytes([85, 85, 85, 85, 85, 85, 213, 63, 0, 0, 0, 0, 0, 0, 240, 191]);
+  }
+
+  void test_addInt() {
+    sig.addInt(1);
+    sig.addInt(1000);
+    sig.addInt(1000000);
+    sig.addInt(1000000000);
+    checkBytes(
+        [1, 0, 0, 0, 0xe8, 3, 0, 0, 0x40, 0x42, 0xf, 0, 0, 0xca, 0x9a, 0x3b]);
+  }
+
+  void test_addString() {
+    sig.addString('abc');
+    sig.addString('\u00f8');
+    checkBytes([3, 0, 0, 0, 0x61, 0x62, 0x63, 2, 0, 0, 0, 0xc3, 0xb8]);
+  }
+
+  void test_excludesInformative() {
+    // Verify that API signatures exclude informative data by checking that two
+    // UnlinkedCombinator instances that differ only in their offset result in
+    // the same signature.
+    UnlinkedCombinatorBuilder combinator1 =
+        new UnlinkedCombinatorBuilder(shows: ['foo'], offset: 1);
+    UnlinkedCombinatorBuilder combinator2 =
+        new UnlinkedCombinatorBuilder(shows: ['foo'], offset: 2);
+    expect(signUnlinkedCombinator(combinator1),
+        signUnlinkedCombinator(combinator2));
+  }
+
+  void test_includesSemantic() {
+    // Verify that API signatures include semantic data by checking that two
+    // UnlinkedCombinator instances that differ only in their "shows" lists
+    // result in different signatures.
+    UnlinkedCombinatorBuilder combinator1 =
+        new UnlinkedCombinatorBuilder(shows: ['foo'], offset: 1);
+    UnlinkedCombinatorBuilder combinator2 =
+        new UnlinkedCombinatorBuilder(shows: ['bar'], offset: 1);
+    expect(signUnlinkedCombinator(combinator1),
+        isNot(signUnlinkedCombinator(combinator2)));
+  }
+
+  void test_manyInts() {
+    // This verifies that the logic to extend the internal buffer works
+    // properly.
+    List<int> expectedResult = [];
+    for (int i = 0; i < 100000; i++) {
+      sig.addInt(i);
+      expectedResult.add(i % 0x100);
+      expectedResult.add((i ~/ 0x100) % 0x100);
+      expectedResult.add((i ~/ 0x10000) % 0x100);
+      expectedResult.add((i ~/ 0x1000000) % 0x100);
+    }
+    checkBytes(expectedResult);
+  }
+}

pkg/analyzer/test/src/summary/linker_test.dart

diff --git a/pkg/analyzer/test/src/summary/linker_test.dart b/pkg/analyzer/test/src/summary/linker_test.dart
index 674ce9b..196fb04 100644
--- a/pkg/analyzer/test/src/summary/linker_test.dart
+++ b/pkg/analyzer/test/src/summary/linker_test.dart
@@ -43,6 +43,24 @@
     return linker.getLibrary(Uri.parse(uri));
   }
 
+  void test_apiSignature_apiChanges() {
+    var bundle0 =
+        createPackageBundle('f(int i) { print(i); }', path: '/test.dart');
+    var bundle1 =
+        createPackageBundle('f(String s) { print(s); }', path: '/test.dart');
+    expect(bundle0.apiSignature, isNotEmpty);
+    expect(bundle1.apiSignature, isNotEmpty);
+    expect(bundle0.apiSignature, isNot(bundle1.apiSignature));
+  }
+
+  void test_apiSignature_localChanges() {
+    var bundle0 = createPackageBundle('f() { print(0); }', path: '/test.dart');
+    var bundle1 = createPackageBundle('f() { print(1); }', path: '/test.dart');
+    expect(bundle0.apiSignature, isNotEmpty);
+    expect(bundle1.apiSignature, isNotEmpty);
+    expect(bundle0.apiSignature, bundle1.apiSignature);
+  }
+
   void test_baseClass_genericWithAccessor() {
     createLinker('''
 class B<T> {

pkg/analyzer/test/src/summary/resynthesize_ast_test.dart

diff --git a/pkg/analyzer/test/src/summary/resynthesize_ast_test.dart b/pkg/analyzer/test/src/summary/resynthesize_ast_test.dart
index 420d515..cb6cfeb 100644
--- a/pkg/analyzer/test/src/summary/resynthesize_ast_test.dart
+++ b/pkg/analyzer/test/src/summary/resynthesize_ast_test.dart
@@ -567,6 +567,12 @@
 
   @override
   @failingTest
+  void test_inferLocalFunctionReturnType() {
+    super.test_inferLocalFunctionReturnType();
+  }
+
+  @override
+  @failingTest
   void test_inferredType_opAssignToProperty_prefixedIdentifier() {
     super.test_inferredType_opAssignToProperty_prefixedIdentifier();
   }
@@ -600,6 +606,12 @@
 
   @override
   @failingTest
+  void test_nullCoalescingOperator() {
+    super.test_nullCoalescingOperator();
+  }
+
+  @override
+  @failingTest
   void test_unsafeBlockClosureInference_closureCall() {
     super.test_unsafeBlockClosureInference_closureCall();
   }

pkg/analyzer/test/src/summary/test_all.dart

diff --git a/pkg/analyzer/test/src/summary/test_all.dart b/pkg/analyzer/test/src/summary/test_all.dart
index b3a0467..6ad0f2f 100644
--- a/pkg/analyzer/test/src/summary/test_all.dart
+++ b/pkg/analyzer/test/src/summary/test_all.dart
@@ -7,6 +7,7 @@
 import 'package:unittest/unittest.dart';
 
 import '../../utils.dart';
+import 'api_signature_test.dart' as api_signature_test;
 import 'flat_buffers_test.dart' as flat_buffers_test;
 import 'in_summary_source_test.dart' as in_summary_source_test;
 import 'incremental_cache_test.dart' as incremental_cache_test;
@@ -27,6 +28,7 @@
 main() {
   initializeTestEnvironment();
   group('summary tests', () {
+    api_signature_test.main();
     flat_buffers_test.main();
     in_summary_source_test.main();
     incremental_cache_test.main();

pkg/analyzer/test/src/task/dart_test.dart

diff --git a/pkg/analyzer/test/src/task/dart_test.dart b/pkg/analyzer/test/src/task/dart_test.dart
index cedcb9b..c274834 100644
--- a/pkg/analyzer/test/src/task/dart_test.dart
+++ b/pkg/analyzer/test/src/task/dart_test.dart
@@ -1556,113 +1556,143 @@
 
   void test_library_cycle_incremental() {
     enableStrongMode();
-    Source lib1Source = newSource(
-        '/my_lib1.dart',
+    Source a = newSource(
+        '/a.dart',
         '''
-library my_lib1;
+library a;
 ''');
-    Source lib2Source = newSource(
-        '/my_lib2.dart',
+    Source b = newSource(
+        '/b.dart',
         '''
-library my_lib2;
-import 'my_lib1.dart';
+library b;
+import 'a.dart';
 ''');
-    Source lib3Source = newSource(
-        '/my_lib3.dart',
+    Source c = newSource(
+        '/c.dart',
         '''
-library my_lib3;
-import 'my_lib2.dart';
+library c;
+import 'b.dart';
 ''');
 
-    computeResult(lib1Source, LIBRARY_CYCLE);
-    expect(outputs[LIBRARY_CYCLE], hasLength(1));
-    computeResult(lib2Source, LIBRARY_CYCLE);
-    expect(outputs[LIBRARY_CYCLE], hasLength(1));
-    computeResult(lib3Source, LIBRARY_CYCLE);
-    expect(outputs[LIBRARY_CYCLE], hasLength(1));
+    _assertLibraryCycle(a, [a]);
+    _assertLibraryCycle(b, [b]);
+    _assertLibraryCycle(c, [c]);
 
-    // create a cycle
+    // Create a cycle.
     context.setContents(
-        lib1Source,
+        a,
         '''
-library my_lib1;
-import 'my_lib3.dart';
+library a;
+import 'c.dart';
 ''');
-    _expectInvalid(lib1Source);
-    _expectInvalid(lib2Source);
-    _expectInvalid(lib3Source);
+    _expectInvalid(a);
+    _expectInvalid(b);
+    _expectInvalid(c);
 
-    computeResult(lib1Source, LIBRARY_CYCLE);
-    expect(outputs[LIBRARY_CYCLE], hasLength(3));
-    computeResult(lib2Source, LIBRARY_CYCLE);
-    expect(outputs[LIBRARY_CYCLE], hasLength(3));
-    computeResult(lib3Source, LIBRARY_CYCLE);
-    expect(outputs[LIBRARY_CYCLE], hasLength(3));
+    _assertLibraryCycle(a, [a, b, c]);
+    _assertLibraryCycle(b, [a, b, c]);
+    _assertLibraryCycle(c, [a, b, c]);
 
-    // break the cycle again
+    // Break the cycle again.
     context.setContents(
-        lib1Source,
+        a,
         '''
-library my_lib1;
+library a;
 ''');
-    _expectInvalid(lib1Source);
-    _expectInvalid(lib2Source);
-    _expectInvalid(lib3Source);
+    _expectInvalid(a);
+    _expectInvalid(b);
+    _expectInvalid(c);
 
-    computeResult(lib1Source, LIBRARY_CYCLE);
-    expect(outputs[LIBRARY_CYCLE], hasLength(1));
-    computeResult(lib2Source, LIBRARY_CYCLE);
-    expect(outputs[LIBRARY_CYCLE], hasLength(1));
-    computeResult(lib3Source, LIBRARY_CYCLE);
-    expect(outputs[LIBRARY_CYCLE], hasLength(1));
+    _assertLibraryCycle(a, [a]);
+    _assertLibraryCycle(b, [b]);
+    _assertLibraryCycle(c, [c]);
   }
 
   void test_library_cycle_incremental_partial() {
     enableStrongMode();
-    Source lib1Source = newSource(
-        '/my_lib1.dart',
-        '''
-library my_lib1;
+    Source a = newSource(
+        '/a.dart',
+        r'''
+library a;
 ''');
-    Source lib2Source = newSource(
-        '/my_lib2.dart',
-        '''
-library my_lib2;
-import 'my_lib1.dart';
+    Source b = newSource(
+        '/b.dart',
+        r'''
+library b;
+import 'a.dart';
 ''');
-    Source lib3Source = newSource(
-        '/my_lib3.dart',
-        '''
-library my_lib3;
-import 'my_lib2.dart';
+    Source c = newSource(
+        '/c.dart',
+        r'''
+library c;
+import 'b.dart';
 ''');
 
-    computeResult(lib1Source, LIBRARY_CYCLE);
-    expect(outputs[LIBRARY_CYCLE], hasLength(1));
-    computeResult(lib2Source, LIBRARY_CYCLE);
-    expect(outputs[LIBRARY_CYCLE], hasLength(1));
-    // lib3 is not reachable, so we have not yet computed its library
-    // cycles
+    _assertLibraryCycle(a, [a]);
+    _assertLibraryCycle(b, [b]);
+    // 'c' is not reachable, so we have not yet computed its library cycles.
 
-    // complete the cycle, via lib3
+    // Complete the cycle, via 'c'.
     context.setContents(
-        lib1Source,
-        '''
-library my_lib1;
-import 'my_lib3.dart';
+        a,
+        r'''
+library a;
+import 'c.dart';
 ''');
-    _expectInvalid(lib1Source);
-    _expectInvalid(lib2Source);
-    _expectInvalid(lib3Source);
+    _expectInvalid(a);
+    _expectInvalid(b);
+    _expectInvalid(c);
 
-    // Ensure that invalidation correctly invalidated everything reachable
-    // through lib3
-    computeResult(lib1Source, LIBRARY_CYCLE);
-    expect(outputs[LIBRARY_CYCLE], hasLength(3));
-    computeResult(lib2Source, LIBRARY_CYCLE);
-    expect(outputs[LIBRARY_CYCLE], hasLength(3));
-    computeResult(lib3Source, LIBRARY_CYCLE);
-    expect(outputs[LIBRARY_CYCLE], hasLength(3));
+    // Ensure that everything reachable through 'c' was invalidated,
+    // and recomputed to include all three sources.
+    _assertLibraryCycle(a, [a, b, c]);
+    _assertLibraryCycle(b, [a, b, c]);
+    _assertLibraryCycle(c, [a, b, c]);
+  }
+
+  void test_library_cycle_incremental_partial2() {
+    enableStrongMode();
+    Source a = newSource(
+        '/a.dart',
+        r'''
+library a;
+import 'b.dart';
+''');
+    Source b = newSource(
+        '/b.dart',
+        r'''
+library b;
+import 'a.dart';
+''');
+    Source c = newSource(
+        '/c.dart',
+        r'''
+library c;
+import 'b.dart';
+''');
+
+    _assertLibraryCycle(a, [a, b]);
+    _assertLibraryCycle(b, [a, b]);
+    _assertLibraryCycle(c, [c]);
+
+    // Include 'c' into the cycle.
+    context.setContents(
+        a,
+        r'''
+library a;
+import 'b.dart';
+import 'c.dart';
+''');
+    _expectInvalid(a);
+    _expectInvalid(b);
+    _expectInvalid(c);
+
+    // Start processing with 'b', so that when we resolve 'b' directives,
+    // and invalidate library cycles, the 'a' directives are not resolved yet,
+    // so we don't know that the cycle must include 'c'.
+    _assertLibraryCycle(b, [a, b, c]);
+    _assertLibraryCycle(a, [a, b, c]);
+    _assertLibraryCycle(c, [a, b, c]);
   }
 
   void test_library_cycle_linear() {
@@ -1974,6 +2004,12 @@
     expect(dep5, hasLength(5)); // dart:core, a.dart, aa.dart, ab.dart, b.dart
   }
 
+  void _assertLibraryCycle(Source source, List<Source> expected) {
+    computeResult(source, LIBRARY_CYCLE);
+    List<LibraryElement> cycle = outputs[LIBRARY_CYCLE] as List<LibraryElement>;
+    expect(cycle.map((e) => e.source), unorderedEquals(expected));
+  }
+
   void _expectInvalid(Source librarySource) {
     CacheEntry entry = context.getCacheEntry(librarySource);
     expect(entry.getState(LIBRARY_CYCLE), CacheState.INVALID);

pkg/analyzer/test/src/task/strong/checker_test.dart

diff --git a/pkg/analyzer/test/src/task/strong/checker_test.dart b/pkg/analyzer/test/src/task/strong/checker_test.dart
index 8bdd4ce..f7ce4b2 100644
--- a/pkg/analyzer/test/src/task/strong/checker_test.dart
+++ b/pkg/analyzer/test/src/task/strong/checker_test.dart
@@ -2612,6 +2612,20 @@
 ''');
   }
 
+  void test_leastUpperBounds_fuzzyArrows() {
+    checkFile(r'''
+typedef String TakesA<T>(T item);
+
+void main() {
+  TakesA<int> f;
+  TakesA<dynamic> g;
+  TakesA<String> h;
+  g = h;
+  f = /*warning:DOWN_CAST_COMPOSITE*/f ?? g;
+}
+''');
+  }
+
   void test_loadLibrary() {
     addFile('''library lib1;''', name: '/lib1.dart');
     checkFile(r'''
@@ -3322,7 +3336,9 @@
                 bool isValidKey(potentialKey)])
     : _comparator = /*warning:DOWN_CAST_COMPOSITE*/(compare == null) ? Comparable.compare : compare,
       _validKey = (isValidKey != null) ? isValidKey : ((v) => true) {
-    _Predicate<Object> v = (isValidKey != null)
+
+    // NOTE: this is a down cast because isValidKey has fuzzy arrow type.
+    _Predicate<Object> v = /*warning:DOWN_CAST_COMPOSITE*/(isValidKey != null)
         ? isValidKey : (/*info:INFERRED_TYPE_CLOSURE*/(_) => true);
 
     v = (isValidKey != null)
@@ -3406,6 +3422,29 @@
 ''');
   }
 
+  void test_typePromotionFromTypeParameter() {
+    // Regression test for https://github.com/dart-lang/sdk/issues/26965
+    checkFile(r'''
+void f/*<T>*/(/*=T*/ object) {
+  if (object is String) print(object.substring(1));
+}
+void g/*<T extends num>*/(/*=T*/ object) {
+  if (object is int) print(object.isEven);
+  if (object is String) print(/*info:DYNAMIC_INVOKE*/object.substring(1));
+}
+class Clonable<T> {}
+class SubClonable<T> extends Clonable<T> {
+  T m(T t) => t;
+}
+void h/*<T extends Clonable<T>>*/(/*=T*/ object) {
+  if (/*info:NON_GROUND_TYPE_CHECK_INFO*/object is SubClonable/*<T>*/) {
+    // Note we need to cast back to T, because promotion lost that type info.
+    print(object.m(object as dynamic/*=T*/));
+  }
+}
+''');
+  }
+
   void test_typeSubtyping_assigningClass() {
     checkFile('''
 class A {}

pkg/analyzer/test/src/task/strong/inferred_type_test.dart

diff --git a/pkg/analyzer/test/src/task/strong/inferred_type_test.dart b/pkg/analyzer/test/src/task/strong/inferred_type_test.dart
index 741659a..4dbcbfe 100644
--- a/pkg/analyzer/test/src/task/strong/inferred_type_test.dart
+++ b/pkg/analyzer/test/src/task/strong/inferred_type_test.dart
@@ -588,6 +588,16 @@
 ''');
   }
 
+  void test_constructors_downwardsWithConstraint() {
+    // Regression test for https://github.com/dart-lang/sdk/issues/26431
+    checkFile(r'''
+class Foo<T extends Iterable> {}
+void main() {
+  Foo<List> foo = /*info:INFERRED_TYPE_ALLOCATION*/new Foo();
+}
+    ''');
+  }
+
   void test_constructors_inferenceFBounded() {
     // Regression for https://github.com/dart-lang/sdk/issues/26990
     var unit = checkFile('''
@@ -2508,6 +2518,42 @@
   ''');
   }
 
+  void test_inferLocalFunctionReturnType() {
+    // Regression test for https://github.com/dart-lang/sdk/issues/26414
+    var unit = checkFile(r'''
+main() {
+  f0() => 42;
+  f1() async => 42;
+
+  f2 /*info:INFERRED_TYPE_CLOSURE*/() { return 42; }
+  f3 /*info:INFERRED_TYPE_CLOSURE*/() async { return 42; }
+  f4 /*info:INFERRED_TYPE_CLOSURE*/() sync* { yield 42; }
+  f5 /*info:INFERRED_TYPE_CLOSURE*/() async* { yield 42; }
+
+  num f6() => 42;
+
+  f7() => f7();
+  f8() => /*error:REFERENCED_BEFORE_DECLARATION*/f9();
+  f9() => f5();
+}
+''');
+    var fns = unit.functions[0].functions;
+    expect(fns[0].type.toString(), '() → int');
+    expect(fns[1].type.toString(), '() → Future<int>');
+
+    expect(fns[2].type.toString(), '() → int');
+    expect(fns[3].type.toString(), '() → Future<int>');
+    expect(fns[4].type.toString(), '() → Iterable<int>');
+    expect(fns[5].type.toString(), '() → Stream<int>');
+
+    expect(fns[6].type.toString(), '() → num');
+
+    // Recursive cases: these infer in declaration order.
+    expect(fns[7].type.toString(), '() → dynamic');
+    expect(fns[8].type.toString(), '() → dynamic');
+    expect(fns[9].type.toString(), '() → Stream<int>');
+  }
+
   void test_inferred_nonstatic_field_depends_on_static_field_complex() {
     var mainUnit = checkFile('''
 class C {
@@ -3616,6 +3662,23 @@
 ''');
   }
 
+  void test_nullCoalescingOperator() {
+    // Regression test for https://github.com/dart-lang/sdk/issues/26552
+    checkFile(r'''
+List<int> x;
+var y = x ?? /*info:INFERRED_TYPE_LITERAL*/[];
+List<int> z = y;
+    ''');
+    // Don't do anything if we already have a context type.
+    var unit = checkFile(r'''
+List<int> x;
+List<num> y = x ?? /*info:INFERRED_TYPE_LITERAL*/[];
+    ''');
+
+    expect(unit.topLevelVariables[1].initializer.returnType.toString(),
+        'List<num>');
+  }
+
   void test_nullLiteralShouldNotInferAsBottom() {
     // Regression test for https://github.com/dart-lang/dev_compiler/issues/47
     checkFile(r'''
@@ -3906,10 +3969,9 @@
   }
 
   void test_unsafeBlockClosureInference_closureCall() {
-    // Note: this is a DYNAMIC_INVOKE due to dartbug.com/26962.
+    // Regression test for https://github.com/dart-lang/sdk/issues/26962
     var mainUnit = checkFile('''
-var v = /*info:DYNAMIC_INVOKE*/((x) => 1.0)(
-  /*info:INFERRED_TYPE_CLOSURE*/() { return 1; });
+var v = ((x) => 1.0)(/*info:INFERRED_TYPE_CLOSURE*/() { return 1; });
 ''');
     var v = mainUnit.topLevelVariables[0];
     expect(v.name, 'v');
@@ -4075,11 +4137,9 @@
   }
 
   void test_unsafeBlockClosureInference_functionCall_noTypeParam_viaExpr() {
-    // TODO(paulberry): why is the call to f() considered a DYNAMIC_INVOKE?
     var mainUnit = checkFile('''
 double f(x) => 1.0;
-var v = /*info:DYNAMIC_INVOKE*/(f)(
-  /*info:INFERRED_TYPE_CLOSURE*/() { return 1; });
+var v = (f)(/*info:INFERRED_TYPE_CLOSURE*/() { return 1; });
 ''');
     var v = mainUnit.topLevelVariables[0];
     expect(v.name, 'v');

pkg/analyzer/test/src/task/strong/non_null_checker_test.dart

diff --git a/pkg/analyzer/test/src/task/strong/non_null_checker_test.dart b/pkg/analyzer/test/src/task/strong/non_null_checker_test.dart
index ac056c2..e5fa7c6 100644
--- a/pkg/analyzer/test/src/task/strong/non_null_checker_test.dart
+++ b/pkg/analyzer/test/src/task/strong/non_null_checker_test.dart
@@ -48,11 +48,17 @@
     checkFile('int x = null;');
   }
 
-  void test_nonnullableTypes() {
-    // If `int`s are non-nullable, then this code should throw an error.
-    addFile('int x;');
+  void test_initialize_nonnullable_with_null() {
     addFile('int x = /*error:INVALID_ASSIGNMENT*/null;');
+    check(nonnullableTypes: <String>['dart:core,int']);
+  }
+
+  void test_initialize_nonnullable_with_valid_value() {
     addFile('int x = 0;');
+    check(nonnullableTypes: <String>['dart:core,int']);
+  }
+
+  void test_assign_null_to_nonnullable() {
     addFile('''
 int x = 0;
 
@@ -63,4 +69,119 @@
 ''');
     check(nonnullableTypes: <String>['dart:core,int']);
   }
+
+  void test_uninitialized_nonnullable_local_variable() {
+    // Ideally, we will do flow analysis and throw an error only if a variable
+    // is used before it has been initialized.
+    addFile('main() { int /*error:NON_NULLABLE_FIELD_NOT_INITIALIZED*/x; }');
+    check(nonnullableTypes: <String>['dart:core,int']);
+  }
+
+  void test_uninitialized_nonnullable_top_level_variable_declaration() {
+    // If `int`s are non-nullable, then this code should throw an error.
+    addFile('int /*error:NON_NULLABLE_FIELD_NOT_INITIALIZED*/x;');
+    check(nonnullableTypes: <String>['dart:core,int']);
+  }
+
+  void test_uninitialized_nonnullable_field_declaration() {
+    addFile('''
+void foo() {}
+
+class A {
+  // Ideally, we should allow x to be init in the constructor, but that requires
+  // too much complication in the checker, so for now we throw a static error at
+  // the declaration site.
+  int /*error:NON_NULLABLE_FIELD_NOT_INITIALIZED*/x;
+
+  A();
+}
+''');
+    check(nonnullableTypes: <String>['dart:core,int']);
+  }
+
+  void test_prefer_final_to_non_nullable_error() {
+    addFile('main() { final int /*error:FINAL_NOT_INITIALIZED*/x; }');
+    addFile('final int /*error:FINAL_NOT_INITIALIZED*/x;');
+    addFile('''
+void foo() {}
+
+class A {
+  final int x;
+
+  /*warning:FINAL_NOT_INITIALIZED_CONSTRUCTOR_1*/A();
+}
+''');
+    check(nonnullableTypes: <String>['dart:core,int']);
+  }
+
+  // Default example from NNBD document.
+  final String defaultNnbdExample = '''
+class Point {
+  final int x, y;
+  Point(this.x, this.y);
+  Point operator +(Point other) => new Point(x + other.x, y + other.y);
+  String toString() => "x: \$x, y: \$y";
+}
+
+void main() {
+  Point p1 = new Point(0, 0);
+  Point p2 = new Point(10, 10);
+  print("p1 + p2 = \${p1 + p2}");
+}
+''';
+
+  final String defaultNnbdExampleMod1 = '''
+class Point {
+  final int x, y;
+  Point(this.x, this.y);
+  Point operator +(Point other) => new Point(x + other.x, y + other.y);
+  String toString() => "x: \$x, y: \$y";
+}
+
+void main() {
+  Point p1 = new Point(0, 0);
+  Point p2 = new Point(10, /*boom*/null); // Change here.
+  print("p1 + p2 = \${p1 + p2}");
+}
+''';
+
+  final String defaultNnbdExampleMod2 = '''
+class Point {
+  final int x, y;
+  Point(this.x, this.y);
+  Point operator +(Point other) => new Point(x + other.x, y + other.y);
+  String toString() => "x: \$x, y: \$y";
+}
+
+void main() {
+  bool f = false; // Necessary, because dead code is otherwise detected.
+  Point p1 = new Point(0, 0);
+  Point p2 = new Point(10, /*boom*/f ? 10 : null); // Change here.
+  print("p1 + p2 = \${p1 + p2}");
+}
+''';
+
+  void test_nullable_fields() {
+    addFile(defaultNnbdExample);
+    // `null` can be passed as an argument to `Point` in default mode.
+    addFile(defaultNnbdExampleMod1);
+    // A nullable expression can be passed as an argument to `Point` in default
+    // mode.
+    addFile(defaultNnbdExampleMod2);
+    check();
+  }
+
+  void test_nonnullable_fields() {
+    addFile(defaultNnbdExample);
+    // `null` can be passed as an argument to `Point` in default mode.
+    addFile(_withError(defaultNnbdExampleMod1, "error:INVALID_ASSIGNMENT"));
+    // A nullable expression can be passed as an argument to `Point` in default
+    // mode.
+    addFile(_withError(defaultNnbdExampleMod2, "error:INVALID_ASSIGNMENT"));
+    check(nonnullableTypes: <String>['dart:core,int']);
+  }
+}
+
+String _withError(String file, String error) {
+  return ("" + file).replaceFirst("boom", error);
 }

pkg/analyzer/tool/summary/generate.dart

diff --git a/pkg/analyzer/tool/summary/generate.dart b/pkg/analyzer/tool/summary/generate.dart
index 869b38d..124a6b8 100644
--- a/pkg/analyzer/tool/summary/generate.dart
+++ b/pkg/analyzer/tool/summary/generate.dart
@@ -451,6 +451,7 @@
     out("import 'flat_buffers.dart' as fb;");
     out("import 'idl.dart' as idl;");
     out("import 'dart:convert' as convert;");
+    out("import 'api_signature.dart' as api_sig;");
     out();
     for (idlModel.EnumDeclaration enm in _idl.enums.values) {
       _generateEnumReader(enm);
@@ -618,6 +619,44 @@
         });
         out('}');
       }
+      // Generate collectApiSignature().
+      {
+        out();
+        out('/**');
+        out(' * Accumulate non-[informative] data into [signature].');
+        out(' */');
+        out('void collectApiSignature(api_sig.ApiSignature signature) {');
+        indent(() {
+          List<idlModel.FieldDeclaration> sortedFields = cls.fields.toList()
+            ..sort((idlModel.FieldDeclaration a, idlModel.FieldDeclaration b) =>
+                a.id.compareTo(b.id));
+          for (idlModel.FieldDeclaration field in sortedFields) {
+            if (field.isInformative) {
+              continue;
+            }
+            String ref = 'this._${field.name}';
+            if (field.type.isList) {
+              out('if ($ref == null) {');
+              indent(() {
+                out('signature.addInt(0);');
+              });
+              out('} else {');
+              indent(() {
+                out('signature.addInt($ref.length);');
+                out('for (var x in $ref) {');
+                indent(() {
+                  _generateSignatureCall(field.type.typeName, 'x', false);
+                });
+                out('}');
+              });
+              out('}');
+            } else {
+              _generateSignatureCall(field.type.typeName, ref, true);
+            }
+          }
+        });
+        out('}');
+      }
       // Generate finish.
       if (cls.isTopLevel) {
         out();
@@ -820,7 +859,8 @@
         } else {
           out('$returnType get $fieldName {');
           indent(() {
-            String readExpr = '$readCode.vTableGet(_bc, _bcOffset, $index, $def)';
+            String readExpr =
+                '$readCode.vTableGet(_bc, _bcOffset, $index, $def)';
             out('_$fieldName ??= $readExpr;');
             out('return _$fieldName;');
           });
@@ -918,6 +958,56 @@
   }
 
   /**
+   * Generate a call to the appropriate method of [ApiSignature] for the type
+   * [typeName], using the data named by [ref].  If [couldBeNull] is `true`,
+   * generate code to handle the possibility that [ref] is `null` (substituting
+   * in the appropriate default value).
+   */
+  void _generateSignatureCall(String typeName, String ref, bool couldBeNull) {
+    if (_idl.enums.containsKey(typeName)) {
+      if (couldBeNull) {
+        out('signature.addInt($ref == null ? 0 : $ref.index);');
+      } else {
+        out('signature.addInt($ref.index);');
+      }
+    } else if (_idl.classes.containsKey(typeName)) {
+      if (couldBeNull) {
+        out('signature.addBool($ref != null);');
+      }
+      out('$ref?.collectApiSignature(signature);');
+    } else {
+      switch (typeName) {
+        case 'String':
+          if (couldBeNull) {
+            ref += " ?? ''";
+          }
+          out("signature.addString($ref);");
+          break;
+        case 'int':
+          if (couldBeNull) {
+            ref += ' ?? 0';
+          }
+          out('signature.addInt($ref);');
+          break;
+        case 'bool':
+          if (couldBeNull) {
+            ref += ' == true';
+          }
+          out('signature.addBool($ref);');
+          break;
+        case 'double':
+          if (couldBeNull) {
+            ref += ' ?? 0.0';
+          }
+          out('signature.addDouble($ref);');
+          break;
+        default:
+          throw "Don't know how to generate signature call for $typeName";
+      }
+    }
+  }
+
+  /**
    * Return the documentation text of the given [node], or `null` if the [node]
    * does not have a comment.  Each line is `\n` separated.
    */

pkg/analyzer/tool/summary/idl_model.dart

diff --git a/pkg/analyzer/tool/summary/idl_model.dart b/pkg/analyzer/tool/summary/idl_model.dart
index 18b981c..375dcb2 100644
--- a/pkg/analyzer/tool/summary/idl_model.dart
+++ b/pkg/analyzer/tool/summary/idl_model.dart
@@ -123,6 +123,9 @@
   final bool isList;
 
   FieldType(this.typeName, this.isList);
+
+  @override
+  String toString() => isList ? 'List<$typeName>' : typeName;
 }
 
 /**

pkg/analyzer_cli/lib/src/build_mode.dart

diff --git a/pkg/analyzer_cli/lib/src/build_mode.dart b/pkg/analyzer_cli/lib/src/build_mode.dart
index 1883318..6d3555f 100644
--- a/pkg/analyzer_cli/lib/src/build_mode.dart
+++ b/pkg/analyzer_cli/lib/src/build_mode.dart
@@ -10,7 +10,6 @@
 import 'package:analyzer/dart/ast/ast.dart' show CompilationUnit;
 import 'package:analyzer/dart/element/element.dart';
 import 'package:analyzer/file_system/file_system.dart';
-import 'package:analyzer/file_system/physical_file_system.dart';
 import 'package:analyzer/src/generated/engine.dart';
 import 'package:analyzer/src/generated/error.dart';
 import 'package:analyzer/src/generated/java_io.dart';
@@ -39,23 +38,26 @@
   final StringBuffer errorBuffer = new StringBuffer();
   final StringBuffer outBuffer = new StringBuffer();
 
+  final ResourceProvider resourceProvider;
   final String dartSdkPath;
 
-  AnalyzerWorkerLoop(SyncWorkerConnection connection, {this.dartSdkPath})
+  AnalyzerWorkerLoop(this.resourceProvider, SyncWorkerConnection connection,
+      {this.dartSdkPath})
       : super(connection: connection);
 
-  factory AnalyzerWorkerLoop.std(
+  factory AnalyzerWorkerLoop.std(ResourceProvider resourceProvider,
       {io.Stdin stdinStream, io.Stdout stdoutStream, String dartSdkPath}) {
     SyncWorkerConnection connection = new StdSyncWorkerConnection(
         stdinStream: stdinStream, stdoutStream: stdoutStream);
-    return new AnalyzerWorkerLoop(connection, dartSdkPath: dartSdkPath);
+    return new AnalyzerWorkerLoop(resourceProvider, connection,
+        dartSdkPath: dartSdkPath);
   }
 
   /**
    * Performs analysis with given [options].
    */
   void analyze(CommandLineOptions options) {
-    new BuildMode(options, new AnalysisStats()).analyze();
+    new BuildMode(resourceProvider, options, new AnalysisStats()).analyze();
     AnalysisEngine.instance.clearCaches();
   }
 
@@ -123,10 +125,10 @@
  * Analyzer used when the "--build-mode" option is supplied.
  */
 class BuildMode {
+  final ResourceProvider resourceProvider;
   final CommandLineOptions options;
   final AnalysisStats stats;
 
-  final ResourceProvider resourceProvider = PhysicalResourceProvider.INSTANCE;
   SummaryDataStore summaryDataStore;
   InternalAnalysisContext context;
   Map<Uri, JavaFile> uriToFileMap;
@@ -136,7 +138,7 @@
   final Set<Source> processedSources = new Set<Source>();
   final Map<Uri, UnlinkedUnit> uriToUnit = <Uri, UnlinkedUnit>{};
 
-  BuildMode(this.options, this.stats);
+  BuildMode(this.resourceProvider, this.options, this.stats);
 
   /**
    * Perform package analysis according to the given [options].
@@ -285,7 +287,7 @@
     ]);
 
     // Set context options.
-    Driver.setAnalysisContextOptions(context, options,
+    Driver.setAnalysisContextOptions(resourceProvider, context, options,
         (AnalysisOptionsImpl contextOptions) {
       if (options.buildSummaryOnlyDiet) {
         contextOptions.analyzeFunctionBodies = false;

pkg/analyzer_cli/lib/src/driver.dart

diff --git a/pkg/analyzer_cli/lib/src/driver.dart b/pkg/analyzer_cli/lib/src/driver.dart
index da11426..9cddf0d 100644
--- a/pkg/analyzer_cli/lib/src/driver.dart
+++ b/pkg/analyzer_cli/lib/src/driver.dart
@@ -13,11 +13,12 @@
 import 'package:analyzer/plugin/options.dart';
 import 'package:analyzer/plugin/resolver_provider.dart';
 import 'package:analyzer/source/analysis_options_provider.dart';
-import 'package:analyzer/source/embedder.dart';
 import 'package:analyzer/source/package_map_provider.dart';
 import 'package:analyzer/source/package_map_resolver.dart';
 import 'package:analyzer/source/pub_package_map_provider.dart';
 import 'package:analyzer/source/sdk_ext.dart';
+import 'package:analyzer/src/context/builder.dart';
+import 'package:analyzer/src/dart/sdk/sdk.dart';
 import 'package:analyzer/src/generated/constant.dart';
 import 'package:analyzer/src/generated/engine.dart';
 import 'package:analyzer/src/generated/error.dart';
@@ -94,6 +95,12 @@
   /// SDK instance.
   DartSdk sdk;
 
+  /**
+   * The resource provider used to access the file system.
+   */
+  file_system.ResourceProvider resourceProvider =
+      PhysicalResourceProvider.INSTANCE;
+
   /// Collected analysis statistics.
   final AnalysisStats stats = new AnalysisStats();
 
@@ -245,9 +252,11 @@
   ErrorSeverity _buildModeAnalyze(CommandLineOptions options) {
     return _analyzeAllTag.makeCurrentWhile(() {
       if (options.buildModePersistentWorker) {
-        new AnalyzerWorkerLoop.std(dartSdkPath: options.dartSdkPath).run();
+        new AnalyzerWorkerLoop.std(resourceProvider,
+                dartSdkPath: options.dartSdkPath)
+            .run();
       } else {
-        return new BuildMode(options, stats).analyze();
+        return new BuildMode(resourceProvider, options, stats).analyze();
       }
     });
   }
@@ -338,8 +347,7 @@
       Map<file_system.Folder, YamlMap> embedderMap, _PackageInfo packageInfo) {
     // Create a custom package resolver if one has been specified.
     if (packageResolverProvider != null) {
-      file_system.Folder folder =
-          PhysicalResourceProvider.INSTANCE.getResource('.');
+      file_system.Folder folder = resourceProvider.getResource('.');
       UriResolver resolver = packageResolverProvider(folder);
       if (resolver != null) {
         UriResolver sdkResolver = new DartUriResolver(sdk);
@@ -348,7 +356,7 @@
         List<UriResolver> resolvers = <UriResolver>[
           sdkResolver,
           resolver,
-          new file_system.ResourceUriResolver(PhysicalResourceProvider.INSTANCE)
+          new file_system.ResourceUriResolver(resourceProvider)
         ];
         return new SourceFactory(resolvers);
       }
@@ -364,9 +372,8 @@
       if (packageInfo.packageMap == null) {
         // Fall back to pub list-package-dirs.
         PubPackageMapProvider pubPackageMapProvider =
-            new PubPackageMapProvider(PhysicalResourceProvider.INSTANCE, sdk);
-        file_system.Resource cwd =
-            PhysicalResourceProvider.INSTANCE.getResource('.');
+            new PubPackageMapProvider(resourceProvider, sdk);
+        file_system.Resource cwd = resourceProvider.getResource('.');
         PackageMapInfo packageMapInfo =
             pubPackageMapProvider.computePackageMap(cwd);
         Map<String, List<file_system.Folder>> packageMap =
@@ -376,8 +383,8 @@
         // If it failed, that's not a problem; it simply means we have no way
         // to resolve packages.
         if (packageMapInfo.packageMap != null) {
-          packageUriResolver = new PackageMapUriResolver(
-              PhysicalResourceProvider.INSTANCE, packageMap);
+          packageUriResolver =
+              new PackageMapUriResolver(resourceProvider, packageMap);
         }
       }
     }
@@ -388,7 +395,7 @@
     // 'dart:' URIs come first.
 
     // Setup embedding.
-    EmbedderSdk embedderSdk = new EmbedderSdk(embedderMap);
+    EmbedderSdk embedderSdk = new EmbedderSdk(resourceProvider, embedderMap);
     if (embedderSdk.libraryMap.size() == 0) {
       // The embedder uri resolver has no mappings. Use the default Dart SDK
       // uri resolver.
@@ -410,8 +417,7 @@
     }
 
     // Finally files.
-    resolvers.add(
-        new file_system.ResourceUriResolver(PhysicalResourceProvider.INSTANCE));
+    resolvers.add(new file_system.ResourceUriResolver(resourceProvider));
 
     return new SourceFactory(resolvers, packageInfo.packages);
   }
@@ -475,7 +481,7 @@
 
     AnalyzeFunctionBodiesPredicate dietParsingPolicy =
         _chooseDietParsingPolicy(options);
-    setAnalysisContextOptions(_context, options,
+    setAnalysisContextOptions(resourceProvider, _context, options,
         (AnalysisOptionsImpl contextOptions) {
       contextOptions.analyzeFunctionBodiesPredicate = dietParsingPolicy;
     });
@@ -546,8 +552,7 @@
       packageMap = _PackageRootPackageMapBuilder
           .buildPackageMap(options.packageRootPath);
     } else {
-      file_system.Resource cwd =
-          PhysicalResourceProvider.INSTANCE.getResource('.');
+      file_system.Resource cwd = resourceProvider.getResource('.');
       // Look for .packages.
       packages = _discoverPackagespec(new Uri.directory(cwd.path));
       packageMap = _getPackageMap(packages);
@@ -564,9 +569,7 @@
     Map<String, List<file_system.Folder>> folderMap =
         new Map<String, List<file_system.Folder>>();
     packages.asMap().forEach((String packagePath, Uri uri) {
-      folderMap[packagePath] = [
-        PhysicalResourceProvider.INSTANCE.getFolder(path.fromUri(uri))
-      ];
+      folderMap[packagePath] = [resourceProvider.getFolder(path.fromUri(uri))];
     });
     return folderMap;
   }
@@ -654,6 +657,7 @@
   }
 
   static void setAnalysisContextOptions(
+      file_system.ResourceProvider resourceProvider,
       AnalysisContext context,
       CommandLineOptions options,
       void configureContextOptions(AnalysisOptionsImpl contextOptions)) {
@@ -678,7 +682,7 @@
     context.analysisOptions = contextOptions;
 
     // Process analysis options file (and notify all interested parties).
-    _processAnalysisOptions(context, options);
+    _processAnalysisOptions(resourceProvider, context, options);
   }
 
   /// Perform a deep comparison of two string maps.
@@ -694,21 +698,23 @@
     return true;
   }
 
-  static file_system.File _getOptionsFile(CommandLineOptions options) {
+  static file_system.File _getOptionsFile(
+      file_system.ResourceProvider resourceProvider,
+      CommandLineOptions options) {
     file_system.File file;
     String filePath = options.analysisOptionsFile;
     if (filePath != null) {
-      file = PhysicalResourceProvider.INSTANCE.getFile(filePath);
+      file = resourceProvider.getFile(filePath);
       if (!file.exists) {
         printAndFail('Options file not found: $filePath',
             exitCode: ErrorSeverity.ERROR.ordinal);
       }
     } else {
       filePath = AnalysisEngine.ANALYSIS_OPTIONS_FILE;
-      file = PhysicalResourceProvider.INSTANCE.getFile(filePath);
+      file = resourceProvider.getFile(filePath);
       if (!file.exists) {
         filePath = AnalysisEngine.ANALYSIS_OPTIONS_YAML_FILE;
-        file = PhysicalResourceProvider.INSTANCE.getFile(filePath);
+        file = resourceProvider.getFile(filePath);
       }
     }
     return file;
@@ -719,8 +725,10 @@
       path.normalize(new File(sourcePath).absolute.path);
 
   static void _processAnalysisOptions(
-      AnalysisContext context, CommandLineOptions options) {
-    file_system.File file = _getOptionsFile(options);
+      file_system.ResourceProvider resourceProvider,
+      AnalysisContext context,
+      CommandLineOptions options) {
+    file_system.File file = _getOptionsFile(resourceProvider, options);
     List<OptionsProcessor> optionsProcessors =
         AnalysisEngine.instance.optionsPlugin.optionsProcessors;
     try {

pkg/analyzer_cli/test/build_mode_test.dart

diff --git a/pkg/analyzer_cli/test/build_mode_test.dart b/pkg/analyzer_cli/test/build_mode_test.dart
index 9c88a0f..5e16d95 100644
--- a/pkg/analyzer_cli/test/build_mode_test.dart
+++ b/pkg/analyzer_cli/test/build_mode_test.dart
@@ -4,6 +4,7 @@
 
 library analyzer_cli.test.built_mode;
 
+import 'package:analyzer/file_system/memory_file_system.dart';
 import 'package:analyzer_cli/src/build_mode.dart';
 import 'package:analyzer_cli/src/driver.dart';
 import 'package:analyzer_cli/src/options.dart';
@@ -17,6 +18,25 @@
   defineReflectiveTests(WorkerLoopTest);
 }
 
+typedef void _TestWorkerLoopAnalyze(CommandLineOptions options);
+
+/**
+ * [AnalyzerWorkerLoop] for testing.
+ */
+class TestAnalyzerWorkerLoop extends AnalyzerWorkerLoop {
+  final _TestWorkerLoopAnalyze _analyze;
+
+  TestAnalyzerWorkerLoop(SyncWorkerConnection connection, [this._analyze])
+      : super(new MemoryResourceProvider(), connection);
+
+  @override
+  void analyze(CommandLineOptions options) {
+    if (_analyze != null) {
+      _analyze(options);
+    }
+  }
+}
+
 @reflectiveTest
 class WorkerLoopTest {
   final TestStdinSync stdinStream = new TestStdinSync();
@@ -30,16 +50,6 @@
 
   void setUp() {}
 
-  List<int> _serializeProto(GeneratedMessage message) {
-    var buffer = message.writeToBuffer();
-
-    var writer = new CodedBufferWriter();
-    writer.writeInt32NoTag(buffer.length);
-    writer.writeRawBytes(buffer);
-
-    return writer.toBuffer();
-  }
-
   test_run() {
     var request = new WorkRequest();
     request.arguments.addAll([
@@ -119,23 +129,14 @@
     stdinStream.close();
     new TestAnalyzerWorkerLoop(connection).run();
   }
-}
 
-typedef void _TestWorkerLoopAnalyze(CommandLineOptions options);
+  List<int> _serializeProto(GeneratedMessage message) {
+    var buffer = message.writeToBuffer();
 
-/**
- * [AnalyzerWorkerLoop] for testing.
- */
-class TestAnalyzerWorkerLoop extends AnalyzerWorkerLoop {
-  final _TestWorkerLoopAnalyze _analyze;
+    var writer = new CodedBufferWriter();
+    writer.writeInt32NoTag(buffer.length);
+    writer.writeRawBytes(buffer);
 
-  TestAnalyzerWorkerLoop(SyncWorkerConnection connection, [this._analyze])
-      : super(connection);
-
-  @override
-  void analyze(CommandLineOptions options) {
-    if (_analyze != null) {
-      _analyze(options);
-    }
+    return writer.toBuffer();
   }
 }

pkg/compiler/lib/src/compiler.dart

diff --git a/pkg/compiler/lib/src/compiler.dart b/pkg/compiler/lib/src/compiler.dart
index 7f73dcc..bdda974 100644
--- a/pkg/compiler/lib/src/compiler.dart
+++ b/pkg/compiler/lib/src/compiler.dart
@@ -27,7 +27,6 @@
 import 'compile_time_constants.dart';
 import 'constants/values.dart';
 import 'core_types.dart' show CoreClasses, CoreTypes;
-import 'dart_backend/dart_backend.dart' as dart_backend;
 import 'dart_types.dart' show DartType, DynamicType, InterfaceType, Types;
 import 'deferred_load.dart' show DeferredLoadTask;
 import 'diagnostics/code_location.dart';
@@ -302,19 +301,13 @@
 
     if (makeBackend != null) {
       backend = makeBackend(this);
-    } else if (options.emitJavaScript) {
+    } else {
       js_backend.JavaScriptBackend jsBackend = new js_backend.JavaScriptBackend(
           this,
           generateSourceMap: options.generateSourceMap,
           useStartupEmitter: options.useStartupEmitter,
           useNewSourceInfo: options.useNewSourceInfo);
       backend = jsBackend;
-    } else {
-      backend = new dart_backend.DartBackend(this, options.strips,
-          multiFile: options.dart2dartMultiFile);
-      if (options.dumpInfo) {
-        throw new ArgumentError('--dump-info is not supported for dart2dart.');
-      }
     }
 
     if (options.dumpInfo && options.useStartupEmitter) {

pkg/compiler/lib/src/dart2js.dart

diff --git a/pkg/compiler/lib/src/dart2js.dart b/pkg/compiler/lib/src/dart2js.dart
index 1b41be1..c4ab29f 100644
--- a/pkg/compiler/lib/src/dart2js.dart
+++ b/pkg/compiler/lib/src/dart2js.dart
@@ -115,8 +115,6 @@
   List<String> explicitOutputArguments = <String>[];
   bool wantHelp = false;
   bool wantVersion = false;
-  String outputLanguage = 'JavaScript';
-  bool stripArgumentSet = false;
   bool analyzeOnly = false;
   bool analyzeAll = false;
   bool resolveOnly = false;
@@ -173,20 +171,10 @@
     optionsImplyCompilation.add(argument);
     if (argument == '--output-type=dart' ||
         argument == '--output-type=dart-multi') {
-      outputLanguage = OUTPUT_LANGUAGE_DART;
-      if (explicitOutputArguments.isNotEmpty) {
-        out = currentDirectory.resolve('out.dart');
-        sourceMapOut = currentDirectory.resolve('out.dart.map');
-      }
-      diagnosticHandler(
-          null,
-          null,
-          null,
-          "--output-type=dart is deprecated. It will remain available "
-          "in Dart 1.11, but will be removed in Dart 1.12.",
-          api.Diagnostic.WARNING);
+      helpAndFail(
+          "--output-type=dart is no longer supported. It was deprecated "
+          "since Dart 1.11 and removed in Dart 1.19.");
     }
-    passThrough(argument);
   }
 
   void setResolutionInput(String argument) {
@@ -208,14 +196,14 @@
     return filenames.join("\n");
   }
 
-  void implyCompilation(String argument) {
+  implyCompilation(String argument) {
     optionsImplyCompilation.add(argument);
     passThrough(argument);
   }
 
-  void setStrip(String argument) {
-    stripArgumentSet = true;
-    implyCompilation(argument);
+  setStrip(String argument) {
+    helpAndFail("Option '--force-strip' is not in use now that"
+        "--output-type=dart is no longer supported.");
   }
 
   void setAnalyzeOnly(String argument) {
@@ -326,6 +314,7 @@
     new OptionHandler(Flags.fatalWarnings, passThrough),
     new OptionHandler(
         Flags.suppressHints, (_) => diagnosticHandler.showHints = false),
+    // TODO(sigmund): remove entirely after Dart 1.20
     new OptionHandler(
         '--output-type=dart|--output-type=dart-multi|--output-type=js',
         setOutputType),
@@ -370,7 +359,7 @@
     new OptionHandler(Flags.disableTypeInference, implyCompilation),
     new OptionHandler(Flags.terse, passThrough),
     new OptionHandler('--deferred-map=.+', implyCompilation),
-    new OptionHandler(Flags.dumpInfo, setDumpInfo),
+    new OptionHandler(Flags.dumpInfo, implyCompilation),
     new OptionHandler(
         '--disallow-unsafe-eval', (_) => hasDisallowUnsafeEval = true),
     new OptionHandler(Option.showPackageWarnings, passThrough),
@@ -429,10 +418,6 @@
         " '$precompiledName'.");
   }
 
-  if (outputLanguage != OUTPUT_LANGUAGE_DART && stripArgumentSet) {
-    helpAndFail("Option '--force-strip' may only be used with "
-        "'--output-type=dart'.");
-  }
   if (arguments.isEmpty) {
     helpAndFail('No Dart file specified.');
   }
@@ -487,10 +472,6 @@
           "in combination with the '${Flags.analyzeOnly}' option.");
     }
   }
-  if (dumpInfo && outputLanguage == OUTPUT_LANGUAGE_DART) {
-    helpAndFail("Option '${Flags.dumpInfo}' is not supported in "
-        "combination with the '--output-type=dart' option.");
-  }
 
   options.add('--out=$out');
   options.add('--source-map=$sourceMapOut');
@@ -511,11 +492,11 @@
     diagnosticHandler
         .info('Compiled ${inputProvider.dartCharactersRead} characters Dart '
             '-> ${outputProvider.totalCharactersWritten} characters '
-            '$outputLanguage in '
+            'JavaScript in '
             '${relativize(currentDirectory, out, Platform.isWindows)}');
     if (diagnosticHandler.verbose) {
       String input = uriPathToNative(arguments[0]);
-      print('Dart file ($input) compiled to $outputLanguage.');
+      print('Dart file ($input) compiled to JavaScript.');
       print('Wrote the following files:');
       for (String filename in outputProvider.allOutputFiles) {
         print("  $filename");
@@ -523,7 +504,7 @@
     } else if (explicitOutputArguments.isNotEmpty) {
       String input = uriPathToNative(arguments[0]);
       String output = relativize(currentDirectory, out, Platform.isWindows);
-      print('Dart file ($input) compiled to $outputLanguage: $output');
+      print('Dart file ($input) compiled to JavaScript: $output');
     }
     return result;
   }
@@ -672,9 +653,6 @@
 The following options are only used for compiler development and may
 be removed in a future version:
 
-  --output-type=dart
-    Output Dart code instead of JavaScript.
-
   --throw-on-error
     Throw an exception if a compile-time error is detected.
 
@@ -703,8 +681,6 @@
     Generates an out.info.json file with information about the generated code.
     You can inspect the generated file with the viewer at:
         https://dart-lang.github.io/dump-info-visualizer/
-    This feature is currently not supported in combination with the
-    '--output-type=dart' option.
 
   --generate-code-with-compile-time-errors
     Generates output even if the program contains compile-time errors. Use the

pkg/compiler/lib/src/dart_backend/backend.dart

diff --git a/pkg/compiler/lib/src/dart_backend/backend.dart b/pkg/compiler/lib/src/dart_backend/backend.dart
deleted file mode 100644
index c9d69ae..0000000
--- a/pkg/compiler/lib/src/dart_backend/backend.dart
+++ /dev/null
@@ -1,546 +0,0 @@
-// Copyright (c) 2012, the Dart project authors.  Please see the AUTHORS file
-// for details. All rights reserved. Use of this source code is governed by a
-// BSD-style license that can be found in the LICENSE file.
-
-part of dart_backend;
-
-// TODO(ahe): This class is simply wrong.  This backend should use
-// elements when it can, not AST nodes.  Perhaps a [Map<Element,
-// TreeElements>] is what is needed.
-class ElementAst {
-  final Node ast;
-  final TreeElements treeElements;
-
-  ElementAst(this.ast, this.treeElements);
-}
-
-class DartBackend extends Backend {
-  final List<CompilerTask> tasks;
-  final bool stripAsserts;
-
-  bool get supportsReflection => true;
-
-  // TODO(zarah) Maybe change this to a command-line option.
-  // Right now, it is set by the tests.
-  bool useMirrorHelperLibrary = false;
-
-  /// Updated to a [MirrorRenamerImpl] instance if the [useMirrorHelperLibrary]
-  /// field is set and mirror are needed.
-  MirrorRenamer mirrorRenamer = const MirrorRenamer();
-
-  final DartOutputter outputter;
-
-  // Used in test.
-  PlaceholderRenamer get placeholderRenamer => outputter.renamer;
-  Map<ClassNode, List<Node>> get memberNodes => outputter.output.memberNodes;
-
-  ConstantSystem get constantSystem {
-    return constantCompilerTask.constantCompiler.constantSystem;
-  }
-
-  BackendConstantEnvironment get constants => constantCompilerTask;
-
-  DartConstantTask constantCompilerTask;
-
-  DartImpactTransformer impactTransformer;
-
-  final Set<ClassElement> usedTypeLiterals = new Set<ClassElement>();
-
-  /// The set of visible platform classes that are implemented by instantiated
-  /// user classes.
-  final Set<ClassElement> _userImplementedPlatformClasses =
-      new Set<ClassElement>();
-
-  bool enableCodegenWithErrorsIfSupported(Spannable node) {
-    reporter.reportHintMessage(node, MessageKind.GENERIC, {
-      'text': "Generation of code with compile time errors is not "
-          "supported for dart2dart."
-    });
-    return false;
-  }
-
-  /**
-   * Tells whether it is safe to remove type declarations from variables,
-   * functions parameters. It becomes not safe if:
-   * 1) TypeError is used somewhere in the code,
-   * 2) The code has typedefs in right hand side of IS checks,
-   * 3) The code has classes which extend typedefs, have type arguments typedefs
-   *    or type variable bounds typedefs.
-   * These restrictions can be less strict.
-   */
-  bool isSafeToRemoveTypeDeclarations(
-      Map<ClassElement, Iterable<Element>> classMembers) {
-    ClassElement typeErrorElement = compiler.coreLibrary.find('TypeError');
-    if (classMembers.containsKey(typeErrorElement) ||
-        compiler.resolverWorld.isChecks
-            .any((DartType type) => type.element == typeErrorElement)) {
-      return false;
-    }
-    Set<DartType> processedTypes = new Set<DartType>();
-    List<DartType> workQueue = new List<DartType>();
-    workQueue
-        .addAll(classMembers.keys.map((classElement) => classElement.thisType));
-    workQueue.addAll(compiler.resolverWorld.isChecks);
-
-    while (!workQueue.isEmpty) {
-      DartType type = workQueue.removeLast();
-      if (processedTypes.contains(type)) continue;
-      processedTypes.add(type);
-      if (type is FunctionType) return false;
-      if (type is TypedefType) return false;
-      if (type is InterfaceType) {
-        InterfaceType interfaceType = type;
-        // Check all type arguments.
-        interfaceType.typeArguments.forEach(workQueue.add);
-        ClassElement element = type.element;
-        // Check all supertypes.
-        if (element.allSupertypes != null) {
-          element.allSupertypes.forEach(workQueue.add);
-        }
-      }
-    }
-    return true;
-  }
-
-  DartBackend(Compiler compiler, List<String> strips, {bool multiFile})
-      : tasks = <CompilerTask>[],
-        stripAsserts = strips.indexOf('asserts') != -1,
-        constantCompilerTask = new DartConstantTask(compiler),
-        outputter = new DartOutputter(
-            compiler.reporter, compiler.outputProvider,
-            forceStripTypes: strips.indexOf('types') != -1,
-            multiFile: multiFile,
-            enableMinification: compiler.options.enableMinification),
-        super(compiler) {
-    impactTransformer = new DartImpactTransformer(this);
-  }
-
-  DiagnosticReporter get reporter => compiler.reporter;
-
-  Resolution get resolution => compiler.resolution;
-
-  bool classNeedsRti(ClassElement cls) => false;
-  bool methodNeedsRti(FunctionElement function) => false;
-
-  void enqueueHelpers(ResolutionEnqueuer world, Registry registry) {
-    // Right now resolver doesn't always resolve interfaces needed
-    // for literals, so force them. TODO(antonm): fix in the resolver.
-    final LITERAL_TYPE_NAMES = const [
-      'Map',
-      'List',
-      'num',
-      'int',
-      'double',
-      'bool'
-    ];
-    final coreLibrary = compiler.coreLibrary;
-    for (final name in LITERAL_TYPE_NAMES) {
-      ClassElement classElement = coreLibrary.findLocal(name);
-      classElement.ensureResolved(resolution);
-    }
-    // Enqueue the methods that the VM might invoke on user objects because
-    // we don't trust the resolution to always get these included.
-    world.registerDynamicUse(new DynamicUse(Selectors.toString_, null));
-    world.registerDynamicUse(new DynamicUse(Selectors.hashCode_, null));
-    world.registerDynamicUse(
-        new DynamicUse(new Selector.binaryOperator('=='), null));
-    world.registerDynamicUse(new DynamicUse(Selectors.compareTo, null));
-  }
-
-  WorldImpact codegen(CodegenWorkItem work) {
-    return const WorldImpact();
-  }
-
-  /**
-   * Tells whether we should output given element. Corelib classes like
-   * Object should not be in the resulting code.
-   */
-  @override
-  bool shouldOutput(Element element) {
-    return (!element.library.isPlatformLibrary &&
-            !element.isSynthesized &&
-            element is! AbstractFieldElement) ||
-        mirrorRenamer.isMirrorHelperLibrary(element.library);
-  }
-
-  int assembleProgram() {
-    ElementAst computeElementAst(AstElement element) {
-      return new ElementAst(
-          element.resolvedAst.node, element.resolvedAst.elements);
-    }
-
-    // TODO(johnniwinther): Remove the need for this method.
-    void postProcessElementAst(AstElement element, ElementAst elementAst,
-        newTypedefElementCallback, newClassElementCallback) {
-      ReferencedElementCollector collector = new ReferencedElementCollector(
-          reporter,
-          element,
-          elementAst,
-          newTypedefElementCallback,
-          newClassElementCallback);
-      collector.collect();
-    }
-
-    int totalSize = outputter.assembleProgram(
-        libraries: compiler.libraryLoader.libraries,
-        instantiatedClasses: compiler.resolverWorld.directlyInstantiatedClasses,
-        resolvedElements: compiler.enqueuer.resolution.processedElements,
-        usedTypeLiterals: usedTypeLiterals,
-        postProcessElementAst: postProcessElementAst,
-        computeElementAst: computeElementAst,
-        shouldOutput: shouldOutput,
-        isSafeToRemoveTypeDeclarations: isSafeToRemoveTypeDeclarations,
-        sortElements: Elements.sortedByPosition,
-        mirrorRenamer: mirrorRenamer,
-        mainFunction: compiler.mainFunction,
-        outputUri: compiler.options.outputUri);
-
-    // Output verbose info about size ratio of resulting bundle to all
-    // referenced non-platform sources.
-    logResultBundleSizeInfo(outputter.libraryInfo.userLibraries,
-        outputter.elementInfo.topLevelElements, totalSize);
-
-    return totalSize;
-  }
-
-  void logResultBundleSizeInfo(Iterable<LibraryElement> userLibraries,
-      Iterable<Element> topLevelElements, int totalOutputSize) {
-    // Sum total size of scripts in each referenced library.
-    int nonPlatformSize = 0;
-    for (LibraryElement lib in userLibraries) {
-      for (CompilationUnitElement compilationUnit in lib.compilationUnits) {
-        nonPlatformSize += compilationUnit.script.file.length;
-      }
-    }
-    int percentage = totalOutputSize * 100 ~/ nonPlatformSize;
-    log('Total used non-platform files size: ${nonPlatformSize} bytes, '
-        'Output total size: $totalOutputSize bytes (${percentage}%)');
-  }
-
-  log(String message) => reporter.log('[DartBackend] $message');
-
-  @override
-  Future onLibrariesLoaded(LoadedLibraries loadedLibraries) {
-    // All platform classes must be resolved to ensure that their member names
-    // are preserved.
-    loadedLibraries.forEachLibrary((LibraryElement library) {
-      if (library.isPlatformLibrary) {
-        library.forEachLocalMember((Element element) {
-          if (element.isClass) {
-            ClassElement classElement = element;
-            classElement.ensureResolved(resolution);
-          }
-        });
-      }
-    });
-    if (useMirrorHelperLibrary &&
-        loadedLibraries.containsLibrary(Uris.dart_mirrors)) {
-      return compiler.libraryLoader
-          .loadLibrary(compiler.resolvedUriTranslator.translate(
-              loadedLibraries.getLibrary(Uris.dart_mirrors),
-              MirrorRenamerImpl.DART_MIRROR_HELPER,
-              null))
-          .then((LibraryElement library) {
-        mirrorRenamer = new MirrorRenamerImpl(compiler, this, library);
-      });
-    }
-    return new Future.value();
-  }
-
-  @override
-  void registerStaticUse(Element element, Enqueuer enqueuer) {
-    if (element == compiler.mirrorSystemGetNameFunction) {
-      FunctionElement getNameFunction = mirrorRenamer.getNameFunction;
-      if (getNameFunction != null) {
-        enqueuer.addToWorkList(getNameFunction);
-      }
-    }
-  }
-
-  @override
-  void registerInstantiatedType(
-      InterfaceType type, Enqueuer enqueuer, Registry registry,
-      {bool mirrorUsage: false}) {
-    registerPlatformMembers(type, registerUse: registry.registerDynamicUse);
-    super.registerInstantiatedType(type, enqueuer, registry,
-        mirrorUsage: mirrorUsage);
-  }
-
-  /// Register dynamic access of members of [type] that implement members
-  /// of types defined in the platform libraries.
-  void registerPlatformMembers(InterfaceType type,
-      {void registerUse(DynamicUse dynamicUse)}) {
-    // Without patching, dart2dart has no way of performing sound tree-shaking
-    // in face external functions. Therefore we employ another scheme:
-    //
-    // Based on the assumption that the platform code only relies on the
-    // interfaces of it's own classes, we can approximate the semantics of
-    // external functions by eagerly registering dynamic invocation of instance
-    // members defined the platform interfaces.
-    //
-    // Since we only need to generate code for non-platform classes we can
-    // restrict this registration to platform interfaces implemented by
-    // instantiated non-platform classes.
-    //
-    // Consider for instance this program:
-    //
-    //     import 'dart:math' show Random;
-    //
-    //     class MyRandom implements Random {
-    //       int nextInt() => 0;
-    //     }
-    //
-    //     main() {
-    //       print([0, 1, 2].shuffle(new MyRandom()));
-    //     }
-    //
-    // Here `MyRandom` is a subtype if `Random` defined in 'dart:math'. By the
-    // assumption, all methods defined `Random` are potentially called, and
-    // therefore, though there are no visible call sites from the user node,
-    // dynamic invocation of for instance `nextInt` should be registered. In
-    // this case, `nextInt` is actually called by the standard implementation of
-    // `shuffle`.
-
-    ClassElement cls = type.element;
-    if (!cls.library.isPlatformLibrary) {
-      for (Link<DartType> link = cls.allSupertypes;
-          !link.isEmpty;
-          link = link.tail) {
-        InterfaceType supertype = link.head;
-        ClassElement superclass = supertype.element;
-        LibraryElement library = superclass.library;
-        if (library.isPlatformLibrary) {
-          if (_userImplementedPlatformClasses.add(superclass)) {
-            // Register selectors for all instance methods since these might
-            // be called on user classes from within the platform
-            // implementation.
-            superclass.forEachLocalMember((MemberElement element) {
-              if (element.isConstructor || element.isStatic) return;
-
-              element.computeType(resolution);
-              Selector selector = new Selector.fromElement(element);
-              registerUse(new DynamicUse(selector, null));
-            });
-          }
-        }
-      }
-    }
-  }
-
-  @override
-  bool enableDeferredLoadingIfSupported(Spannable node, Registry registry) {
-    // TODO(sigurdm): Implement deferred loading for dart2dart.
-    reporter.reportWarningMessage(
-        node, MessageKind.DEFERRED_LIBRARY_DART_2_DART);
-    return false;
-  }
-
-  @override
-  Uri resolvePatchUri(String libraryName, Uri) {
-    // Dart2dart does not use patches.
-    return null;
-  }
-}
-
-class DartImpactTransformer extends ImpactTransformer {
-  final DartBackend backend;
-
-  DartImpactTransformer(this.backend);
-
-  @override
-  WorldImpact transformResolutionImpact(ResolutionImpact worldImpact) {
-    TransformedWorldImpact transformed =
-        new TransformedWorldImpact(worldImpact);
-    for (TypeUse typeUse in worldImpact.typeUses) {
-      if (typeUse.kind == TypeUseKind.TYPE_LITERAL &&
-          typeUse.type.isInterfaceType) {
-        backend.usedTypeLiterals.add(typeUse.type.element);
-      }
-      if (typeUse.kind == TypeUseKind.INSTANTIATION) {
-        backend.registerPlatformMembers(typeUse.type,
-            registerUse: transformed.registerDynamicUse);
-      }
-    }
-    return transformed;
-  }
-}
-
-class EmitterUnparser extends Unparser {
-  final Map<Node, String> renames;
-
-  EmitterUnparser(this.renames, {bool minify, bool stripTypes})
-      : super(minify: minify, stripTypes: stripTypes);
-
-  visit(Node node) {
-    if (node != null && renames.containsKey(node)) {
-      write(renames[node]);
-    } else {
-      super.visit(node);
-    }
-  }
-
-  unparseSendReceiver(Send node, {bool spacesNeeded: false}) {
-    // TODO(smok): Remove ugly hack for library prefices.
-    if (node.receiver != null && renames[node.receiver] == '') return;
-    super.unparseSendReceiver(node, spacesNeeded: spacesNeeded);
-  }
-
-  unparseFunctionName(Node name) {
-    if (name != null && renames.containsKey(name)) {
-      write(renames[name]);
-    } else {
-      super.unparseFunctionName(name);
-    }
-  }
-}
-
-/**
- * Some elements are not recorded by resolver now,
- * for example, typedefs or classes which are only
- * used in signatures, as/is operators or in super clauses
- * (just to name a few).  Retraverse AST to pick those up.
- */
-class ReferencedElementCollector extends Visitor {
-  final DiagnosticReporter reporter;
-  final Element element;
-  final ElementAst elementAst;
-  final newTypedefElementCallback;
-  final newClassElementCallback;
-
-  ReferencedElementCollector(this.reporter, this.element, this.elementAst,
-      this.newTypedefElementCallback, this.newClassElementCallback);
-
-  visitNode(Node node) {
-    node.visitChildren(this);
-  }
-
-  visitTypeAnnotation(TypeAnnotation typeAnnotation) {
-    TreeElements treeElements = elementAst.treeElements;
-    final DartType type = treeElements.getType(typeAnnotation);
-    assert(invariant(typeAnnotation, type != null,
-        message: "Missing type for type annotation: $treeElements."));
-    if (type.isTypedef) newTypedefElementCallback(type.element);
-    if (type.isInterfaceType) newClassElementCallback(type.element);
-    typeAnnotation.visitChildren(this);
-  }
-
-  void collect() {
-    reporter.withCurrentElement(element, () {
-      elementAst.ast.accept(this);
-    });
-  }
-}
-
-Comparator compareBy(f) => (x, y) => f(x).compareTo(f(y));
-
-List sorted(Iterable l, comparison) {
-  final result = new List.from(l);
-  result.sort(comparison);
-  return result;
-}
-
-compareElements(e0, e1) {
-  int result = compareBy((e) => e.library.canonicalUri.toString())(e0, e1);
-  if (result != 0) return result;
-  return compareBy((e) => e.position.charOffset)(e0, e1);
-}
-
-/// [ConstantCompilerTask] for compilation of constants for the Dart backend.
-///
-/// Since this task needs no distinction between frontend and backend constants
-/// it also serves as the [BackendConstantEnvironment].
-class DartConstantTask extends ConstantCompilerTask
-    implements BackendConstantEnvironment {
-  final DartConstantCompiler constantCompiler;
-
-  DartConstantTask(Compiler compiler)
-      : this.constantCompiler = new DartConstantCompiler(compiler),
-        super(compiler.measurer);
-
-  String get name => 'ConstantHandler';
-
-  @override
-  ConstantSystem get constantSystem => constantCompiler.constantSystem;
-
-  @override
-  bool hasConstantValue(ConstantExpression expression) {
-    return constantCompiler.hasConstantValue(expression);
-  }
-
-  @override
-  ConstantValue getConstantValue(ConstantExpression expression) {
-    return constantCompiler.getConstantValue(expression);
-  }
-
-  @override
-  ConstantValue getConstantValueForVariable(VariableElement element) {
-    return constantCompiler.getConstantValueForVariable(element);
-  }
-
-  @override
-  ConstantExpression getConstantForNode(Node node, TreeElements elements) {
-    return constantCompiler.getConstantForNode(node, elements);
-  }
-
-  @override
-  ConstantValue getConstantValueForNode(Node node, TreeElements elements) {
-    return getConstantValue(
-        constantCompiler.getConstantForNode(node, elements));
-  }
-
-  @override
-  ConstantValue getConstantValueForMetadata(MetadataAnnotation metadata) {
-    return getConstantValue(metadata.constant);
-  }
-
-  @override
-  ConstantExpression compileConstant(VariableElement element) {
-    return measure(() {
-      return constantCompiler.compileConstant(element);
-    });
-  }
-
-  @override
-  void evaluate(ConstantExpression constant) {
-    return measure(() {
-      return constantCompiler.evaluate(constant);
-    });
-  }
-
-  @override
-  ConstantExpression compileVariable(VariableElement element) {
-    return measure(() {
-      return constantCompiler.compileVariable(element);
-    });
-  }
-
-  @override
-  ConstantExpression compileNode(Node node, TreeElements elements,
-      {bool enforceConst: true}) {
-    return measure(() {
-      return constantCompiler.compileNodeWithDefinitions(node, elements,
-          isConst: enforceConst);
-    });
-  }
-
-  @override
-  ConstantExpression compileMetadata(
-      MetadataAnnotation metadata, Node node, TreeElements elements) {
-    return measure(() {
-      return constantCompiler.compileMetadata(metadata, node, elements);
-    });
-  }
-
-  // TODO(johnniwinther): Remove this when values are computed from the
-  // expressions.
-  @override
-  void copyConstantValues(DartConstantTask task) {
-    constantCompiler.constantValueMap
-        .addAll(task.constantCompiler.constantValueMap);
-  }
-
-  @override
-  void registerLazyStatic(FieldElement element) {
-    // Do nothing.
-  }
-}

pkg/compiler/lib/src/dart_backend/backend_ast_emitter.dart

diff --git a/pkg/compiler/lib/src/dart_backend/backend_ast_emitter.dart b/pkg/compiler/lib/src/dart_backend/backend_ast_emitter.dart
deleted file mode 100644
index fd7a45f..0000000
--- a/pkg/compiler/lib/src/dart_backend/backend_ast_emitter.dart
+++ /dev/null
@@ -1,93 +0,0 @@
-// Copyright (c) 2014, the Dart project authors.  Please see the AUTHORS file
-// for details. All rights reserved. Use of this source code is governed by a
-// BSD-style license that can be found in the LICENSE file.
-
-library backend_ast_emitter;
-
-import 'backend_ast_nodes.dart';
-import '../dart_types.dart';
-import '../elements/elements.dart';
-
-class TypeGenerator {
-  /// TODO(johnniwinther): Remove this when issue 21283 has been resolved.
-  static int pseudoNameCounter = 0;
-
-  static Parameter emitParameter(DartType type,
-      {String name, Element element}) {
-    if (name == null && element != null) {
-      name = element.name;
-    }
-    if (name == null) {
-      name = '_${pseudoNameCounter++}';
-    }
-    Parameter parameter;
-    if (type.isFunctionType) {
-      FunctionType functionType = type;
-      TypeAnnotation returnType = createOptionalType(functionType.returnType);
-      Parameters innerParameters = createParametersFromType(functionType);
-      parameter = new Parameter.function(name, returnType, innerParameters);
-    } else {
-      TypeAnnotation typeAnnotation = createOptionalType(type);
-      parameter = new Parameter(name, type: typeAnnotation);
-    }
-    parameter.element = element;
-    return parameter;
-  }
-
-  static Parameters createParametersFromType(FunctionType functionType) {
-    pseudoNameCounter = 0;
-    if (functionType.namedParameters.isEmpty) {
-      return new Parameters(createParameters(functionType.parameterTypes),
-          createParameters(functionType.optionalParameterTypes), false);
-    } else {
-      return new Parameters(
-          createParameters(functionType.parameterTypes),
-          createParameters(functionType.namedParameterTypes,
-              names: functionType.namedParameters),
-          true);
-    }
-  }
-
-  static List<Parameter> createParameters(Iterable<DartType> parameterTypes,
-      {Iterable<String> names: const <String>[],
-      Iterable<Element> elements: const <Element>[]}) {
-    Iterator<String> name = names.iterator;
-    Iterator<Element> element = elements.iterator;
-    return parameterTypes.map((DartType type) {
-      name.moveNext();
-      element.moveNext();
-      return emitParameter(type, name: name.current, element: element.current);
-    }).toList();
-  }
-
-  /// Like [createTypeAnnotation] except the dynamic type is converted to null.
-  static TypeAnnotation createOptionalType(DartType type) {
-    if (type.treatAsDynamic) {
-      return null;
-    } else {
-      return createType(type);
-    }
-  }
-
-  /// Creates the [TypeAnnotation] for a [type] that is not function type.
-  static TypeAnnotation createType(DartType type) {
-    if (type is GenericType) {
-      if (type.treatAsRaw) {
-        return new TypeAnnotation(type.element.name)..dartType = type;
-      }
-      return new TypeAnnotation(type.element.name,
-          type.typeArguments.map(createType).toList(growable: false))
-        ..dartType = type;
-    } else if (type is VoidType) {
-      return new TypeAnnotation('void')..dartType = type;
-    } else if (type is TypeVariableType) {
-      return new TypeAnnotation(type.name)..dartType = type;
-    } else if (type is DynamicType) {
-      return new TypeAnnotation("dynamic")..dartType = type;
-    } else if (type is MalformedType) {
-      return new TypeAnnotation(type.name)..dartType = type;
-    } else {
-      throw "Unsupported type annotation: $type";
-    }
-  }
-}

pkg/compiler/lib/src/dart_backend/backend_ast_nodes.dart

diff --git a/pkg/compiler/lib/src/dart_backend/backend_ast_nodes.dart b/pkg/compiler/lib/src/dart_backend/backend_ast_nodes.dart
deleted file mode 100644
index 1c8abd0..0000000
--- a/pkg/compiler/lib/src/dart_backend/backend_ast_nodes.dart
+++ /dev/null
@@ -1,1556 +0,0 @@
-// Copyright (c) 2014, the Dart project authors.  Please see the AUTHORS file
-// for details. All rights reserved. Use of this source code is governed by a
-// BSD-style license that can be found in the LICENSE file.
-
-library backend_ast_nodes;
-
-import '../constants/values.dart' as values;
-import '../dart_types.dart' as types;
-import '../elements/elements.dart' as elements;
-import '../tree/tree.dart' as tree;
-import '../util/characters.dart' as characters;
-
-/// The following nodes correspond to [tree.Send] expressions:
-/// [FieldExpression], [IndexExpression], [Assignment], [Increment],
-/// [CallFunction], [CallMethod], [CallNew], [CallStatic], [UnaryOperator],
-/// [BinaryOperator], and [TypeOperator].
-abstract class Node {}
-
-/// Receiver is an [Expression] or the [SuperReceiver].
-abstract class Receiver extends Node {}
-
-/// Argument is an [Expression] or a [NamedArgument].
-abstract class Argument extends Node {}
-
-abstract class Expression extends Node implements Receiver, Argument {
-  bool get assignable => false;
-}
-
-abstract class RootNode extends Node {
-  elements.Element get element;
-}
-
-class FieldDefinition extends RootNode {
-  final elements.Element element;
-  final Expression initializer;
-  FieldDefinition(this.element, this.initializer);
-}
-
-abstract class Statement extends Node {}
-
-/// Used as receiver in expressions that dispatch to the super class.
-/// For instance, an expression such as `super.f()` is represented
-/// by a [CallMethod] node with [SuperReceiver] as its receiver.
-class SuperReceiver extends Receiver {
-  static final SuperReceiver _instance = new SuperReceiver._create();
-
-  factory SuperReceiver() => _instance;
-  SuperReceiver._create();
-}
-
-/// Named arguments may occur in the argument list of
-/// [CallFunction], [CallMethod], [CallNew], and [CallStatic].
-class NamedArgument extends Argument {
-  final String name;
-  final Expression expression;
-
-  NamedArgument(this.name, this.expression);
-}
-
-class TypeAnnotation extends Node {
-  final String name;
-  final List<TypeAnnotation> typeArguments;
-
-  types.DartType dartType;
-
-  TypeAnnotation(this.name, [this.typeArguments = const <TypeAnnotation>[]]);
-}
-
-// STATEMENTS
-
-class Block extends Statement {
-  final List<Statement> statements;
-
-  Block(this.statements);
-}
-
-class Break extends Statement {
-  final String label;
-
-  Break([this.label]);
-}
-
-class Continue extends Statement {
-  final String label;
-
-  Continue([this.label]);
-}
-
-class EmptyStatement extends Statement {
-  static final EmptyStatement _instance = new EmptyStatement._create();
-
-  factory EmptyStatement() => _instance;
-  EmptyStatement._create();
-}
-
-class ExpressionStatement extends Statement {
-  final Expression expression;
-
-  ExpressionStatement(this.expression);
-}
-
-class For extends Statement {
-  final Node initializer;
-  final Expression condition;
-  final List<Expression> updates;
-  final Statement body;
-
-  /// Initializer must be [VariableDeclarations] or [Expression] or null.
-  For(this.initializer, this.condition, this.updates, this.body) {
-    assert(initializer == null ||
-        initializer is VariableDeclarations ||
-        initializer is Expression);
-  }
-}
-
-class ForIn extends Statement {
-  final Node leftHandValue;
-  final Expression expression;
-  final Statement body;
-
-  /// [leftHandValue] must be [Identifier] or [VariableDeclarations] with
-  /// exactly one definition, and that variable definition must have no
-  /// initializer.
-  ForIn(Node leftHandValue, this.expression, this.body)
-      : this.leftHandValue = leftHandValue {
-    assert(leftHandValue is Identifier ||
-        (leftHandValue is VariableDeclarations &&
-            leftHandValue.declarations.length == 1 &&
-            leftHandValue.declarations[0].initializer == null));
-  }
-}
-
-class While extends Statement {
-  final Expression condition;
-  final Statement body;
-
-  While(this.condition, this.body);
-}
-
-class DoWhile extends Statement {
-  final Statement body;
-  final Expression condition;
-
-  DoWhile(this.body, this.condition);
-}
-
-class If extends Statement {
-  final Expression condition;
-  final Statement thenStatement;
-  final Statement elseStatement;
-
-  If(this.condition, this.thenStatement, [this.elseStatement]);
-}
-
-class LabeledStatement extends Statement {
-  final String label;
-  final Statement statement;
-
-  LabeledStatement(this.label, this.statement);
-}
-
-class Rethrow extends Statement {}
-
-class Return extends Statement {
-  final Expression expression;
-
-  Return([this.expression]);
-}
-
-class Switch extends Statement {
-  final Expression expression;
-  final List<SwitchCase> cases;
-
-  Switch(this.expression, this.cases);
-}
-
-/// A sequence of case clauses followed by a sequence of statements.
-/// Represents the default case if [expressions] is null.
-///
-/// NOTE:
-/// Control will never fall through to the following SwitchCase, even if
-/// the list of statements is empty. An empty list of statements will be
-/// unparsed to a semicolon to guarantee this behaviour.
-class SwitchCase extends Node {
-  final List<Expression> expressions;
-  final List<Statement> statements;
-
-  SwitchCase(this.expressions, this.statements);
-  SwitchCase.defaultCase(this.statements) : expressions = null;
-
-  bool get isDefaultCase => expressions == null;
-}
-
-/// A try statement. The try, catch and finally blocks will automatically
-/// be printed inside a block statement if necessary.
-class Try extends Statement {
-  final Statement tryBlock;
-  final List<CatchBlock> catchBlocks;
-  final Statement finallyBlock;
-
-  Try(this.tryBlock, this.catchBlocks, [this.finallyBlock]) {
-    assert(catchBlocks.length > 0 || finallyBlock != null);
-  }
-}
-
-class CatchBlock extends Node {
-  final TypeAnnotation onType;
-  final VariableDeclaration exceptionVar;
-  final VariableDeclaration stackVar;
-  final Statement body;
-
-  /// At least onType or exceptionVar must be given.
-  /// stackVar may only be given if exceptionVar is also given.
-  CatchBlock(this.body, {this.onType, this.exceptionVar, this.stackVar}) {
-    // Must specify at least a type or an exception binding.
-    assert(onType != null || exceptionVar != null);
-
-    // We cannot bind the stack trace without binding the exception too.
-    assert(stackVar == null || exceptionVar != null);
-  }
-}
-
-class VariableDeclarations extends Statement {
-  final TypeAnnotation type;
-  final bool isFinal;
-  final bool isConst;
-  final List<VariableDeclaration> declarations;
-
-  VariableDeclarations(this.declarations,
-      {this.type, this.isFinal: false, this.isConst: false}) {
-    // Cannot be both final and const.
-    assert(!isFinal || !isConst);
-  }
-}
-
-class VariableDeclaration extends Node {
-  final String name;
-  final Expression initializer;
-
-  elements.Element element;
-
-  VariableDeclaration(this.name, [this.initializer]);
-}
-
-class FunctionDeclaration extends Statement {
-  final FunctionExpression function;
-
-  TypeAnnotation get returnType => function.returnType;
-  Parameters get parameters => function.parameters;
-  String get name => function.name;
-  Statement get body => function.body;
-
-  FunctionDeclaration(this.function);
-}
-
-class Parameters extends Node {
-  final List<Parameter> requiredParameters;
-  final List<Parameter> optionalParameters;
-  final bool hasNamedParameters;
-
-  Parameters(this.requiredParameters,
-      [this.optionalParameters, this.hasNamedParameters = false]);
-
-  Parameters.named(this.requiredParameters, this.optionalParameters)
-      : hasNamedParameters = true;
-
-  Parameters.positional(this.requiredParameters, this.optionalParameters)
-      : hasNamedParameters = false;
-
-  bool get hasOptionalParameters =>
-      optionalParameters != null && optionalParameters.length > 0;
-}
-
-class Parameter extends Node {
-  final String name;
-
-  /// Type of parameter, or return type of function parameter.
-  final TypeAnnotation type;
-
-  Expression defaultValue;
-
-  /// Parameters to function parameter. Null for non-function parameters.
-  final Parameters parameters;
-
-  elements.FormalElement element;
-
-  Parameter(this.name, {this.type, this.defaultValue}) : parameters = null;
-
-  Parameter.function(this.name, TypeAnnotation returnType, this.parameters,
-      [this.defaultValue])
-      : type = returnType {
-    assert(parameters != null);
-  }
-
-  /// True if this is a function parameter.
-  bool get isFunction => parameters != null;
-}
-
-// EXPRESSIONS
-
-abstract class Initializer extends Expression {}
-
-class FieldInitializer extends Initializer {
-  final elements.FieldElement element;
-  final Expression body;
-
-  FieldInitializer(this.element, this.body);
-}
-
-class SuperInitializer extends Initializer {
-  final elements.ConstructorElement target;
-  final List<Argument> arguments;
-
-  SuperInitializer(this.target, this.arguments);
-}
-
-class FunctionExpression extends Expression implements RootNode {
-  final TypeAnnotation returnType;
-  String name;
-  final Parameters parameters;
-  final Statement body;
-  final bool isGetter;
-  final bool isSetter;
-
-  elements.FunctionElement element;
-
-  FunctionExpression(this.parameters, this.body,
-      {this.name,
-      this.returnType,
-      this.isGetter: false,
-      this.isSetter: false}) {
-    // Function must have a name if it has a return type
-    assert(returnType == null || name != null);
-  }
-}
-
-class ConstructorDefinition extends FunctionExpression {
-  final List<Initializer> initializers;
-  final bool isConst;
-
-  ConstructorDefinition(Parameters parameters, Statement body,
-      this.initializers, String name, this.isConst)
-      : super(parameters, body, name: name);
-}
-
-class Conditional extends Expression {
-  final Expression condition;
-  final Expression thenExpression;
-  final Expression elseExpression;
-
-  Conditional(this.condition, this.thenExpression, this.elseExpression);
-}
-
-/// An identifier expression.
-/// The unparser does not concern itself with scoping rules, and it is the
-/// responsibility of the AST creator to ensure that the identifier resolves
-/// to the proper definition.
-/// For the time being, this class is also used to reference static fields and
-/// top-level variables that are qualified with a class and/or library name,
-/// assuming the [element] is set. This is likely to change when the old backend
-/// is replaced.
-class Identifier extends Expression {
-  final String name;
-
-  elements.Element element;
-
-  Identifier(this.name);
-
-  bool get assignable => true;
-}
-
-class Literal extends Expression {
-  final values.PrimitiveConstantValue value;
-
-  Literal(this.value);
-}
-
-class LiteralList extends Expression {
-  final bool isConst;
-  final TypeAnnotation typeArgument;
-  final List<Expression> values;
-
-  LiteralList(this.values, {this.typeArgument, this.isConst: false});
-}
-
-class LiteralMap extends Expression {
-  final bool isConst;
-  final List<TypeAnnotation> typeArguments;
-  final List<LiteralMapEntry> entries;
-
-  LiteralMap(this.entries, {this.typeArguments, this.isConst: false}) {
-    assert(this.typeArguments == null ||
-        this.typeArguments.length == 0 ||
-        this.typeArguments.length == 2);
-  }
-}
-
-class LiteralMapEntry extends Node {
-  final Expression key;
-  final Expression value;
-
-  LiteralMapEntry(this.key, this.value);
-}
-
-class LiteralSymbol extends Expression {
-  final String id;
-
-  /// [id] should not include the # symbol
-  LiteralSymbol(this.id);
-}
-
-/// A type literal. This is distinct from [Identifier] since the unparser
-/// needs to this distinguish a static invocation from a method invocation
-/// on a type literal.
-class LiteralType extends Expression {
-  final String name;
-
-  types.DartType type;
-
-  LiteralType(this.name);
-}
-
-/// Reference to a type variable.
-/// This is distinct from [Identifier] since the unparser needs to this
-/// distinguish a function invocation `T()` from a type variable invocation
-/// `(T)()` (the latter is invalid, but must be generated anyway).
-class ReifyTypeVar extends Expression {
-  final String name;
-
-  elements.TypeVariableElement element;
-
-  ReifyTypeVar(this.name);
-}
-
-/// StringConcat is used in place of string interpolation and juxtaposition.
-/// Semantically, each subexpression is evaluated and converted to a string
-/// by `toString()`. These string are then concatenated and returned.
-/// StringConcat unparses to a string literal, possibly with interpolations.
-/// The unparser will flatten nested StringConcats.
-/// A StringConcat node may have any number of children, including zero and one.
-class StringConcat extends Expression {
-  final List<Expression> expressions;
-
-  StringConcat(this.expressions);
-}
-
-/// Expression of form `e.f`.
-class FieldExpression extends Expression {
-  final Receiver object;
-  final String fieldName;
-
-  FieldExpression(this.object, this.fieldName);
-
-  bool get assignable => true;
-}
-
-/// Expression of form `e1[e2]`.
-class IndexExpression extends Expression {
-  final Receiver object;
-  final Expression index;
-
-  IndexExpression(this.object, this.index);
-
-  bool get assignable => true;
-}
-
-/// Expression of form `e(..)`
-/// Note that if [callee] is a [FieldExpression] this will translate into
-/// `(e.f)(..)` and not `e.f(..)`. Use a [CallMethod] to generate
-/// the latter type of expression.
-class CallFunction extends Expression {
-  final Expression callee;
-  final List<Argument> arguments;
-
-  CallFunction(this.callee, this.arguments);
-}
-
-/// Expression of form `e.f(..)`.
-class CallMethod extends Expression {
-  final Receiver object;
-  final String methodName;
-  final List<Argument> arguments;
-
-  CallMethod(this.object, this.methodName, this.arguments);
-}
-
-/// Expression of form `new T(..)`, `new T.f(..)`, `const T(..)`,
-/// or `const T.f(..)`.
-class CallNew extends Expression {
-  final bool isConst;
-  final TypeAnnotation type;
-  final String constructorName;
-  final List<Argument> arguments;
-
-  elements.FunctionElement constructor;
-  types.DartType dartType;
-
-  CallNew(this.type, this.arguments,
-      {this.constructorName, this.isConst: false});
-}
-
-/// Expression of form `T.f(..)`.
-class CallStatic extends Expression {
-  final String className;
-  final String methodName;
-  final List<Argument> arguments;
-
-  elements.Element element;
-
-  CallStatic(this.className, this.methodName, this.arguments);
-}
-
-/// Expression of form `!e` or `-e` or `~e`.
-class UnaryOperator extends Expression {
-  final String operatorName;
-  final Receiver operand;
-
-  UnaryOperator(this.operatorName, this.operand) {
-    assert(isUnaryOperator(operatorName));
-  }
-}
-
-/// Expression of form `e1 + e2`, `e1 - e2`, etc.
-/// This node also represents application of the logical operators && and ||.
-class BinaryOperator extends Expression {
-  final Receiver left;
-  final String operator;
-  final Expression right;
-
-  BinaryOperator(this.left, this.operator, this.right) {
-    assert(isBinaryOperator(operator));
-  }
-}
-
-/// Expression of form `e is T` or `e is! T` or `e as T`.
-class TypeOperator extends Expression {
-  final Expression expression;
-  final String operator;
-  final TypeAnnotation type;
-
-  TypeOperator(this.expression, this.operator, this.type) {
-    assert(operator == 'is' || operator == 'as' || operator == 'is!');
-  }
-}
-
-class Increment extends Expression {
-  final Expression expression;
-  final String operator;
-  final bool isPrefix;
-
-  Increment(this.expression, this.operator, this.isPrefix) {
-    assert(operator == '++' || operator == '--');
-    assert(expression.assignable);
-  }
-
-  Increment.prefix(Expression expression, String operator)
-      : this(expression, operator, true);
-
-  Increment.postfix(Expression expression, String operator)
-      : this(expression, operator, false);
-}
-
-class Assignment extends Expression {
-  static final _operators = new Set.from([
-    '=',
-    '|=',
-    '^=',
-    '&=',
-    '<<=',
-    '>>=',
-    '+=',
-    '-=',
-    '*=',
-    '/=',
-    '%=',
-    '~/='
-  ]);
-
-  final Expression left;
-  final String operator;
-  final Expression right;
-
-  Assignment(this.left, this.operator, this.right) {
-    assert(_operators.contains(operator));
-    assert(left.assignable);
-  }
-}
-
-class Throw extends Expression {
-  final Expression expression;
-
-  Throw(this.expression);
-}
-
-class This extends Expression {
-  static final This _instance = new This._create();
-
-  factory This() => _instance;
-  This._create();
-}
-
-// UNPARSER
-
-bool isUnaryOperator(String op) {
-  return op == '!' || op == '-' || op == '~';
-}
-
-bool isBinaryOperator(String op) {
-  return BINARY_PRECEDENCE.containsKey(op);
-}
-
-/// True if the given operator can be converted to a compound assignment.
-bool isCompoundableOperator(String op) {
-  switch (BINARY_PRECEDENCE[op]) {
-    case BITWISE_OR:
-    case BITWISE_XOR:
-    case BITWISE_AND:
-    case SHIFT:
-    case ADDITIVE:
-    case MULTIPLICATIVE:
-      return true;
-    default:
-      return false;
-  }
-}
-
-// Precedence levels
-const int EXPRESSION = 1;
-const int CONDITIONAL = 2;
-const int LOGICAL_OR = 3;
-const int LOGICAL_AND = 4;
-const int EQUALITY = 6;
-const int RELATIONAL = 7;
-const int BITWISE_OR = 8;
-const int BITWISE_XOR = 9;
-const int BITWISE_AND = 10;
-const int SHIFT = 11;
-const int ADDITIVE = 12;
-const int MULTIPLICATIVE = 13;
-const int UNARY = 14;
-const int POSTFIX_INCREMENT = 15;
-const int TYPE_LITERAL = 19;
-const int PRIMARY = 20;
-
-/// Precedence level required for the callee in a [FunctionCall].
-const int CALLEE = 21;
-
-const Map<String, int> BINARY_PRECEDENCE = const {
-  '&&': LOGICAL_AND,
-  '||': LOGICAL_OR,
-  '==': EQUALITY,
-  '!=': EQUALITY,
-  '>': RELATIONAL,
-  '>=': RELATIONAL,
-  '<': RELATIONAL,
-  '<=': RELATIONAL,
-  '|': BITWISE_OR,
-  '^': BITWISE_XOR,
-  '&': BITWISE_AND,
-  '>>': SHIFT,
-  '<<': SHIFT,
-  '+': ADDITIVE,
-  '-': ADDITIVE,
-  '*': MULTIPLICATIVE,
-  '%': MULTIPLICATIVE,
-  '/': MULTIPLICATIVE,
-  '~/': MULTIPLICATIVE,
-};
-
-/// Return true if binary operators with the given precedence level are
-/// (left) associative. False if they are non-associative.
-bool isAssociativeBinaryOperator(int precedence) {
-  return precedence != EQUALITY && precedence != RELATIONAL;
-}
-
-/// True if [x] is a letter, digit, or underscore.
-/// Such characters may not follow a shorthand string interpolation.
-bool isIdentifierPartNoDollar(dynamic x) {
-  if (x is! int) {
-    return false;
-  }
-  return (characters.$0 <= x && x <= characters.$9) ||
-      (characters.$A <= x && x <= characters.$Z) ||
-      (characters.$a <= x && x <= characters.$z) ||
-      (x == characters.$_);
-}
-
-/// The unparser will apply the following syntactic rewritings:
-///   Use short-hand function returns:
-///     foo(){return E} ==> foo() => E;
-///   Remove empty else branch:
-///     if (E) S else ; ==> if (E) S
-///   Flatten nested blocks:
-///     {S; {S; S}; S} ==> {S; S; S; S}
-///   Remove empty statements from block:
-///     {S; ; S} ==> {S; S}
-///   Unfold singleton blocks:
-///     {S} ==> S
-///   Empty block to empty statement:
-///     {} ==> ;
-///   Introduce not-equals operator:
-///     !(E == E) ==> E != E
-///   Introduce is-not operator:
-///     !(E is T) ==> E is!T
-///
-/// The following transformations will NOT be applied here:
-///   Use implicit this:
-///     this.foo ==> foo              (preconditions too complex for unparser)
-///   Merge adjacent variable definitions:
-///     var x; var y  ==> var x,y;    (hoisting will be done elsewhere)
-///   Merge adjacent labels:
-///     foo: bar: S ==> foobar: S     (scoping is categorically ignored)
-///
-/// The following transformations might be applied here in the future:
-///   Use implicit dynamic types:
-///     dynamic x = E ==> var x = E
-///     <dynamic>[]   ==> []
-class Unparser {
-  StringSink output;
-
-  Unparser(this.output);
-
-  void write(String s) {
-    output.write(s);
-  }
-
-  /// Outputs each element from [items] separated by [separator].
-  /// The actual printing must be performed by the [callback].
-  void writeEach(String separator, Iterable items, void callback(any)) {
-    bool first = true;
-    for (var x in items) {
-      if (first) {
-        first = false;
-      } else {
-        write(separator);
-      }
-      callback(x);
-    }
-  }
-
-  void writeOperator(String operator) {
-    write(" ");
-    write(operator);
-    write(" ");
-  }
-
-  /// Unfolds singleton blocks and returns the inner statement.
-  /// If an empty block is found, the [EmptyStatement] is returned instead.
-  Statement unfoldBlocks(Statement stmt) {
-    while (stmt is Block && stmt.statements.length == 1) {
-      Statement inner = (stmt as Block).statements[0];
-      if (definesVariable(inner)) {
-        return stmt; // Do not unfold block with lexical scope.
-      }
-      stmt = inner;
-    }
-    if (stmt is Block && stmt.statements.length == 0)
-      return new EmptyStatement();
-    return stmt;
-  }
-
-  void writeArgument(Argument arg) {
-    if (arg is NamedArgument) {
-      write(arg.name);
-      write(':');
-      writeExpression(arg.expression);
-    } else {
-      writeExpression(arg);
-    }
-  }
-
-  /// Prints the expression [e].
-  void writeExpression(Expression e) {
-    writeExp(e, EXPRESSION);
-  }
-
-  /// Prints [e] as an expression with precedence of at least [minPrecedence],
-  /// using parentheses if necessary to raise the precedence level.
-  /// Abusing terminology slightly, the function accepts a [Receiver] which
-  /// may also be the [SuperReceiver] object.
-  void writeExp(Receiver e, int minPrecedence, {beginStmt: false}) {
-    void withPrecedence(int actual, void action()) {
-      if (actual < minPrecedence) {
-        write("(");
-        beginStmt = false;
-        action();
-        write(")");
-      } else {
-        action();
-      }
-    }
-    if (e is SuperReceiver) {
-      write('super');
-    } else if (e is FunctionExpression) {
-      assert(!e.isGetter && !e.isSetter);
-      Statement stmt = unfoldBlocks(e.body);
-      int precedence = stmt is Return ? EXPRESSION : PRIMARY;
-      withPrecedence(precedence, () {
-        // A named function expression at the beginning of a statement
-        // can be mistaken for a function declaration.
-        // (Note: Functions with a return type also have a name)
-        bool needParen = beginStmt && e.name != null;
-        if (needParen) {
-          write('(');
-        }
-        if (e.returnType != null) {
-          writeType(e.returnType);
-          write(' ');
-        }
-        if (e.name != null) {
-          write(e.name);
-        }
-        writeParameters(e.parameters);
-        if (stmt is Return) {
-          write('=> ');
-          writeExp(stmt.expression, EXPRESSION);
-        } else {
-          writeBlock(stmt);
-        }
-        if (needParen) {
-          write(')');
-        }
-      });
-    } else if (e is Conditional) {
-      withPrecedence(CONDITIONAL, () {
-        writeExp(e.condition, LOGICAL_OR, beginStmt: beginStmt);
-        write(' ? ');
-        writeExp(e.thenExpression, EXPRESSION);
-        write(' : ');
-        writeExp(e.elseExpression, EXPRESSION);
-      });
-    } else if (e is Identifier) {
-      write(e.name);
-    } else if (e is Literal) {
-      if (e.value.isString) {
-        writeStringLiteral(e);
-      } else if (e.value.isDouble) {
-        double v = e.value.primitiveValue;
-        if (v == double.INFINITY) {
-          withPrecedence(MULTIPLICATIVE, () {
-            write('1/0.0');
-          });
-        } else if (v == double.NEGATIVE_INFINITY) {
-          withPrecedence(MULTIPLICATIVE, () {
-            write('-1/0.0');
-          });
-        } else if (v.isNaN) {
-          withPrecedence(MULTIPLICATIVE, () {
-            write('0/0.0');
-          });
-        } else {
-          write(v.toString());
-        }
-      } else {
-        write(e.value.toDartText());
-      }
-    } else if (e is LiteralList) {
-      if (e.isConst) {
-        write(' const ');
-      }
-      if (e.typeArgument != null) {
-        write('<');
-        writeType(e.typeArgument);
-        write('>');
-      }
-      write('[');
-      writeEach(',', e.values, writeExpression);
-      write(']');
-    } else if (e is LiteralMap) {
-      // The curly brace can be mistaken for a block statement if we
-      // are at the beginning of a statement.
-      bool needParen = beginStmt;
-      if (e.isConst) {
-        write(' const ');
-        needParen = false;
-      }
-      if (e.typeArguments.length > 0) {
-        write('<');
-        writeEach(',', e.typeArguments, writeType);
-        write('>');
-        needParen = false;
-      }
-      if (needParen) {
-        write('(');
-      }
-      write('{');
-      writeEach(',', e.entries, (LiteralMapEntry en) {
-        writeExp(en.key, EXPRESSION);
-        write(' : ');
-        writeExp(en.value, EXPRESSION);
-      });
-      write('}');
-      if (needParen) {
-        write(')');
-      }
-    } else if (e is LiteralSymbol) {
-      write('#');
-      write(e.id);
-    } else if (e is LiteralType) {
-      withPrecedence(TYPE_LITERAL, () {
-        write(e.name);
-      });
-    } else if (e is ReifyTypeVar) {
-      withPrecedence(PRIMARY, () {
-        write(e.name);
-      });
-    } else if (e is StringConcat) {
-      writeStringLiteral(e);
-    } else if (e is UnaryOperator) {
-      Receiver operand = e.operand;
-      // !(x == y) ==> x != y.
-      if (e.operatorName == '!' &&
-          operand is BinaryOperator &&
-          operand.operator == '==') {
-        withPrecedence(EQUALITY, () {
-          writeExp(operand.left, RELATIONAL);
-          writeOperator('!=');
-          writeExp(operand.right, RELATIONAL);
-        });
-      }
-      // !(x is T) ==> x is!T
-      else if (e.operatorName == '!' &&
-          operand is TypeOperator &&
-          operand.operator == 'is') {
-        withPrecedence(RELATIONAL, () {
-          writeExp(operand.expression, BITWISE_OR, beginStmt: beginStmt);
-          write(' is!');
-          writeType(operand.type);
-        });
-      } else {
-        withPrecedence(UNARY, () {
-          writeOperator(e.operatorName);
-          writeExp(e.operand, UNARY);
-        });
-      }
-    } else if (e is BinaryOperator) {
-      int precedence = BINARY_PRECEDENCE[e.operator];
-      withPrecedence(precedence, () {
-        // All binary operators are left-associative or non-associative.
-        // For each operand, we use either the same precedence level as
-        // the current operator, or one higher.
-        int deltaLeft = isAssociativeBinaryOperator(precedence) ? 0 : 1;
-        writeExp(e.left, precedence + deltaLeft, beginStmt: beginStmt);
-        writeOperator(e.operator);
-        writeExp(e.right, precedence + 1);
-      });
-    } else if (e is TypeOperator) {
-      withPrecedence(RELATIONAL, () {
-        writeExp(e.expression, BITWISE_OR, beginStmt: beginStmt);
-        write(' ');
-        write(e.operator);
-        write(' ');
-        writeType(e.type);
-      });
-    } else if (e is Assignment) {
-      withPrecedence(EXPRESSION, () {
-        writeExp(e.left, PRIMARY, beginStmt: beginStmt);
-        writeOperator(e.operator);
-        writeExp(e.right, EXPRESSION);
-      });
-    } else if (e is FieldExpression) {
-      withPrecedence(PRIMARY, () {
-        writeExp(e.object, PRIMARY, beginStmt: beginStmt);
-        write('.');
-        write(e.fieldName);
-      });
-    } else if (e is IndexExpression) {
-      withPrecedence(CALLEE, () {
-        writeExp(e.object, PRIMARY, beginStmt: beginStmt);
-        write('[');
-        writeExp(e.index, EXPRESSION);
-        write(']');
-      });
-    } else if (e is CallFunction) {
-      withPrecedence(CALLEE, () {
-        writeExp(e.callee, CALLEE, beginStmt: beginStmt);
-        write('(');
-        writeEach(',', e.arguments, writeArgument);
-        write(')');
-      });
-    } else if (e is CallMethod) {
-      withPrecedence(CALLEE, () {
-        writeExp(e.object, PRIMARY, beginStmt: beginStmt);
-        write('.');
-        write(e.methodName);
-        write('(');
-        writeEach(',', e.arguments, writeArgument);
-        write(')');
-      });
-    } else if (e is CallNew) {
-      withPrecedence(CALLEE, () {
-        write(' ');
-        write(e.isConst ? 'const ' : 'new ');
-        writeType(e.type);
-        if (e.constructorName != null) {
-          write('.');
-          write(e.constructorName);
-        }
-        write('(');
-        writeEach(',', e.arguments, writeArgument);
-        write(')');
-      });
-    } else if (e is CallStatic) {
-      withPrecedence(CALLEE, () {
-        write(e.className);
-        write('.');
-        write(e.methodName);
-        write('(');
-        writeEach(',', e.arguments, writeArgument);
-        write(')');
-      });
-    } else if (e is Increment) {
-      int precedence = e.isPrefix ? UNARY : POSTFIX_INCREMENT;
-      withPrecedence(precedence, () {
-        if (e.isPrefix) {
-          write(e.operator);
-          writeExp(e.expression, PRIMARY);
-        } else {
-          writeExp(e.expression, PRIMARY, beginStmt: beginStmt);
-          write(e.operator);
-        }
-      });
-    } else if (e is Throw) {
-      withPrecedence(EXPRESSION, () {
-        write('throw ');
-        writeExp(e.expression, EXPRESSION);
-      });
-    } else if (e is This) {
-      write('this');
-    } else {
-      throw "Unexpected expression: $e";
-    }
-  }
-
-  void writeParameters(Parameters params) {
-    write('(');
-    writeEach(',', params.requiredParameters, (Parameter p) {
-      if (p.type != null) {
-        writeType(p.type);
-        write(' ');
-      }
-      write(p.name);
-      if (p.parameters != null) {
-        writeParameters(p.parameters);
-      }
-    });
-    if (params.hasOptionalParameters) {
-      if (params.requiredParameters.length > 0) {
-        write(',');
-      }
-      write(params.hasNamedParameters ? '{' : '[');
-      writeEach(',', params.optionalParameters, (Parameter p) {
-        if (p.type != null) {
-          writeType(p.type);
-          write(' ');
-        }
-        write(p.name);
-        if (p.parameters != null) {
-          writeParameters(p.parameters);
-        }
-        if (p.defaultValue != null) {
-          write(params.hasNamedParameters ? ':' : '=');
-          writeExp(p.defaultValue, EXPRESSION);
-        }
-      });
-      write(params.hasNamedParameters ? '}' : ']');
-    }
-    write(')');
-  }
-
-  void writeStatement(Statement stmt, {bool shortIf: true}) {
-    stmt = unfoldBlocks(stmt);
-    if (stmt is Block) {
-      write('{');
-      stmt.statements.forEach(writeBlockMember);
-      write('}');
-    } else if (stmt is Break) {
-      write('break');
-      if (stmt.label != null) {
-        write(' ');
-        write(stmt.label);
-      }
-      write(';');
-    } else if (stmt is Continue) {
-      write('continue');
-      if (stmt.label != null) {
-        write(' ');
-        write(stmt.label);
-      }
-      write(';');
-    } else if (stmt is EmptyStatement) {
-      write(';');
-    } else if (stmt is ExpressionStatement) {
-      writeExp(stmt.expression, EXPRESSION, beginStmt: true);
-      write(';');
-    } else if (stmt is For) {
-      write('for(');
-      Node init = stmt.initializer;
-      if (init is Expression) {
-        writeExp(init, EXPRESSION);
-      } else if (init is VariableDeclarations) {
-        writeVariableDefinitions(init);
-      }
-      write(';');
-      if (stmt.condition != null) {
-        writeExp(stmt.condition, EXPRESSION);
-      }
-      write(';');
-      writeEach(',', stmt.updates, writeExpression);
-      write(')');
-      writeStatement(stmt.body, shortIf: shortIf);
-    } else if (stmt is ForIn) {
-      write('for(');
-      Node lhv = stmt.leftHandValue;
-      if (lhv is Identifier) {
-        write(lhv.name);
-      } else {
-        writeVariableDefinitions(lhv as VariableDeclarations);
-      }
-      write(' in ');
-      writeExp(stmt.expression, EXPRESSION);
-      write(')');
-      writeStatement(stmt.body, shortIf: shortIf);
-    } else if (stmt is While) {
-      write('while(');
-      writeExp(stmt.condition, EXPRESSION);
-      write(')');
-      writeStatement(stmt.body, shortIf: shortIf);
-    } else if (stmt is DoWhile) {
-      write('do ');
-      writeStatement(stmt.body);
-      write('while(');
-      writeExp(stmt.condition, EXPRESSION);
-      write(');');
-    } else if (stmt is If) {
-      // if (E) S else ; ==> if (E) S
-      Statement elsePart = unfoldBlocks(stmt.elseStatement);
-      if (elsePart is EmptyStatement) {
-        elsePart = null;
-      }
-      if (!shortIf && elsePart == null) {
-        write('{');
-      }
-      write('if(');
-      writeExp(stmt.condition, EXPRESSION);
-      write(')');
-      writeStatement(stmt.thenStatement, shortIf: elsePart == null);
-      if (elsePart != null) {
-        write('else ');
-        writeStatement(elsePart, shortIf: shortIf);
-      }
-      if (!shortIf && elsePart == null) {
-        write('}');
-      }
-    } else if (stmt is LabeledStatement) {
-      write(stmt.label);
-      write(':');
-      writeStatement(stmt.statement, shortIf: shortIf);
-    } else if (stmt is Rethrow) {
-      write('rethrow;');
-    } else if (stmt is Return) {
-      write('return');
-      if (stmt.expression != null) {
-        write(' ');
-        writeExp(stmt.expression, EXPRESSION);
-      }
-      write(';');
-    } else if (stmt is Switch) {
-      write('switch(');
-      writeExp(stmt.expression, EXPRESSION);
-      write('){');
-      for (SwitchCase caze in stmt.cases) {
-        if (caze.isDefaultCase) {
-          write('default:');
-        } else {
-          for (Expression exp in caze.expressions) {
-            write('case ');
-            writeExp(exp, EXPRESSION);
-            write(':');
-          }
-        }
-        if (caze.statements.isEmpty) {
-          write(';'); // Prevent fall-through.
-        } else {
-          caze.statements.forEach(writeBlockMember);
-        }
-      }
-      write('}');
-    } else if (stmt is Try) {
-      write('try');
-      writeBlock(stmt.tryBlock);
-      for (CatchBlock block in stmt.catchBlocks) {
-        if (block.onType != null) {
-          write('on ');
-          writeType(block.onType);
-        }
-        if (block.exceptionVar != null) {
-          write('catch(');
-          write(block.exceptionVar.name);
-          if (block.stackVar != null) {
-            write(',');
-            write(block.stackVar.name);
-          }
-          write(')');
-        }
-        writeBlock(block.body);
-      }
-      if (stmt.finallyBlock != null) {
-        write('finally');
-        writeBlock(stmt.finallyBlock);
-      }
-    } else if (stmt is VariableDeclarations) {
-      writeVariableDefinitions(stmt);
-      write(';');
-    } else if (stmt is FunctionDeclaration) {
-      assert(!stmt.function.isGetter && !stmt.function.isSetter);
-      if (stmt.returnType != null) {
-        writeType(stmt.returnType);
-        write(' ');
-      }
-      write(stmt.name);
-      writeParameters(stmt.parameters);
-      Statement body = unfoldBlocks(stmt.body);
-      if (body is Return) {
-        write('=> ');
-        writeExp(body.expression, EXPRESSION);
-        write(';');
-      } else {
-        writeBlock(body);
-      }
-    } else {
-      throw "Unexpected statement: $stmt";
-    }
-  }
-
-  /// Writes a variable definition statement without the trailing semicolon
-  void writeVariableDefinitions(VariableDeclarations vds) {
-    if (vds.isConst)
-      write('const ');
-    else if (vds.isFinal) write('final ');
-    if (vds.type != null) {
-      writeType(vds.type);
-      write(' ');
-    }
-    if (!vds.isConst && !vds.isFinal && vds.type == null) {
-      write('var ');
-    }
-    writeEach(',', vds.declarations, (VariableDeclaration vd) {
-      write(vd.name);
-      if (vd.initializer != null) {
-        write('=');
-        writeExp(vd.initializer, EXPRESSION);
-      }
-    });
-  }
-
-  /// True of statements that introduce variables in the scope of their
-  /// surrounding block. Blocks containing such statements cannot be unfolded.
-  static bool definesVariable(Statement s) {
-    return s is VariableDeclarations || s is FunctionDeclaration;
-  }
-
-  /// Writes the given statement in a context where only blocks are allowed.
-  void writeBlock(Statement stmt) {
-    if (stmt is Block) {
-      writeStatement(stmt);
-    } else {
-      write('{');
-      writeBlockMember(stmt);
-      write('}');
-    }
-  }
-
-  /// Outputs a statement that is a member of a block statement (or a similar
-  /// sequence of statements, such as in switch statement).
-  /// This will flatten blocks and skip empty statement.
-  void writeBlockMember(Statement stmt) {
-    if (stmt is Block && !stmt.statements.any(definesVariable)) {
-      stmt.statements.forEach(writeBlockMember);
-    } else if (stmt is EmptyStatement) {
-      // do nothing
-    } else {
-      writeStatement(stmt);
-    }
-  }
-
-  void writeType(TypeAnnotation type) {
-    write(type.name);
-    if (type.typeArguments != null && type.typeArguments.length > 0) {
-      write('<');
-      writeEach(',', type.typeArguments, writeType);
-      write('>');
-    }
-  }
-
-  /// A list of string quotings that the printer may use to quote strings.
-  // Ignore multiline quotings for now. Would need to make sure that no
-  // newline (potentially prefixed by whitespace) follows the quoting.
-  static const _QUOTINGS = const <tree.StringQuoting>[
-    const tree.StringQuoting(characters.$DQ, raw: false, leftQuoteLength: 1),
-    const tree.StringQuoting(characters.$DQ, raw: true, leftQuoteLength: 1),
-    const tree.StringQuoting(characters.$SQ, raw: false, leftQuoteLength: 1),
-    const tree.StringQuoting(characters.$SQ, raw: true, leftQuoteLength: 1),
-  ];
-
-  static StringLiteralOutput analyzeStringLiteral(Expression node) {
-    // Flatten the StringConcat tree.
-    List parts = []; // Expression or int (char node)
-    void collectParts(Expression e) {
-      if (e is StringConcat) {
-        e.expressions.forEach(collectParts);
-      } else if (e is Literal && e.value.isString) {
-        for (int char in e.value.primitiveValue) {
-          parts.add(char);
-        }
-      } else {
-        parts.add(e);
-      }
-    }
-    collectParts(node);
-
-    // We use a dynamic algorithm to compute the optimal way of printing
-    // the string literal.
-    //
-    // Using string juxtapositions, it is possible to switch from one quoting
-    // to another, e.g. the constant "''''" '""""' uses this trick.
-    //
-    // As we move through the string from left to right, we maintain a strategy
-    // for each StringQuoting Q, denoting the best way to print the current
-    // prefix so that we end with a string literal quoted with Q.
-    // At every step, each strategy is either:
-    //  1) Updated to include the cost of printing the next character.
-    //  2) Abandoned because it is cheaper to use another strategy as prefix,
-    //     and then switching quotation using a juxtaposition.
-
-    int getQuoteCost(tree.StringQuoting quot) {
-      return quot.leftQuoteLength + quot.rightQuoteLength;
-    }
-
-    // Create initial scores for each StringQuoting and index them
-    // into raw/non-raw and single-quote/double-quote.
-    List<OpenStringChunk> best = <OpenStringChunk>[];
-    List<int> raws = <int>[];
-    List<int> nonRaws = <int>[];
-    List<int> sqs = <int>[];
-    List<int> dqs = <int>[];
-    for (tree.StringQuoting q in _QUOTINGS) {
-      OpenStringChunk chunk = new OpenStringChunk(null, q, getQuoteCost(q));
-      int index = best.length;
-      best.add(chunk);
-
-      if (q.raw) {
-        raws.add(index);
-      } else {
-        nonRaws.add(index);
-      }
-      if (q.quote == characters.$SQ) {
-        sqs.add(index);
-      } else {
-        dqs.add(index);
-      }
-    }
-
-    /// Applies additional cost to each track in [penalized], and considers
-    /// switching from each [penalized] to a [nonPenalized] track.
-    void penalize(List<int> penalized, List<int> nonPenalized, int endIndex,
-        num cost(tree.StringQuoting q)) {
-      for (int j in penalized) {
-        // Check if another track can benefit from switching from this track.
-        for (int k in nonPenalized) {
-          num newCost = best[j].cost +
-              1 // Whitespace in string juxtaposition
-              +
-              getQuoteCost(best[k].quoting);
-          if (newCost < best[k].cost) {
-            best[k] = new OpenStringChunk(
-                best[j].end(endIndex), best[k].quoting, newCost);
-          }
-        }
-        best[j].cost += cost(best[j].quoting);
-      }
-    }
-
-    // Iterate through the string and update the score for each StringQuoting.
-    for (int i = 0; i < parts.length; i++) {
-      var part = parts[i];
-      if (part is int) {
-        int char = part;
-        switch (char) {
-          case characters.$$:
-          case characters.$BACKSLASH:
-            penalize(nonRaws, raws, i, (q) => 1);
-            break;
-          case characters.$DQ:
-            penalize(dqs, sqs, i, (q) => q.raw ? double.INFINITY : 1);
-            break;
-          case characters.$SQ:
-            penalize(sqs, dqs, i, (q) => q.raw ? double.INFINITY : 1);
-            break;
-          case characters.$LF:
-          case characters.$CR:
-          case characters.$FF:
-          case characters.$BS:
-          case characters.$VTAB:
-          case characters.$TAB:
-          case characters.$EOF:
-            penalize(raws, nonRaws, i, (q) => double.INFINITY);
-            break;
-        }
-      } else {
-        // Penalize raw literals for string interpolation.
-        penalize(raws, nonRaws, i, (q) => double.INFINITY);
-
-        // Splitting a string can sometimes allow us to use a shorthand
-        // string interpolation that would otherwise be illegal.
-        // E.g. "...${foo}x..." -> "...$foo" 'x...'
-        // If are other factors that make splitting advantageous,
-        // we can gain even more by doing the split here.
-        if (part is Identifier &&
-            !part.name.contains(r'$') &&
-            i + 1 < parts.length &&
-            isIdentifierPartNoDollar(parts[i + 1])) {
-          for (int j in nonRaws) {
-            for (int k = 0; k < best.length; k++) {
-              num newCost = best[j].cost +
-                  1 // Whitespace in string juxtaposition
-                  -
-                  2 // Save two curly braces
-                  +
-                  getQuoteCost(best[k].quoting);
-              if (newCost < best[k].cost) {
-                best[k] = new OpenStringChunk(
-                    best[j].end(i + 1), best[k].quoting, newCost);
-              }
-            }
-          }
-        }
-      }
-    }
-
-    // Select the cheapest strategy
-    OpenStringChunk bestChunk = best[0];
-    for (OpenStringChunk chunk in best) {
-      if (chunk.cost < bestChunk.cost) {
-        bestChunk = chunk;
-      }
-    }
-
-    return new StringLiteralOutput(parts, bestChunk.end(parts.length));
-  }
-
-  void writeStringLiteral(Expression node) {
-    StringLiteralOutput output = analyzeStringLiteral(node);
-    List parts = output.parts;
-    void printChunk(StringChunk chunk) {
-      int startIndex;
-      if (chunk.previous != null) {
-        printChunk(chunk.previous);
-        write(' '); // String juxtaposition requires a space between literals.
-        startIndex = chunk.previous.endIndex;
-      } else {
-        startIndex = 0;
-      }
-      if (chunk.quoting.raw) {
-        write('r');
-      }
-      write(chunk.quoting.quoteChar);
-      bool raw = chunk.quoting.raw;
-      int quoteCode = chunk.quoting.quote;
-      for (int i = startIndex; i < chunk.endIndex; i++) {
-        var part = parts[i];
-        if (part is int) {
-          int char = part;
-          write(getEscapedCharacter(char, quoteCode, raw));
-        } else if (part is Identifier &&
-            !part.name.contains(r'$') &&
-            (i == chunk.endIndex - 1 ||
-                !isIdentifierPartNoDollar(parts[i + 1]))) {
-          write(r'$');
-          write(part.name);
-        } else {
-          write(r'${');
-          writeExpression(part);
-          write('}');
-        }
-      }
-      write(chunk.quoting.quoteChar);
-    }
-    printChunk(output.chunk);
-  }
-
-  static String getEscapedCharacter(int char, int quoteCode, bool raw) {
-    switch (char) {
-      case characters.$$:
-        return raw ? r'$' : r'\$';
-      case characters.$BACKSLASH:
-        return raw ? r'\' : r'\\';
-      case characters.$DQ:
-        return quoteCode == char ? r'\"' : r'"';
-      case characters.$SQ:
-        return quoteCode == char ? r"\'" : r"'";
-      case characters.$LF:
-        return r'\n';
-      case characters.$CR:
-        return r'\r';
-      case characters.$FF:
-        return r'\f';
-      case characters.$BS:
-        return r'\b';
-      case characters.$TAB:
-        return r'\t';
-      case characters.$VTAB:
-        return r'\v';
-      case characters.$EOF:
-        return r'\x00';
-      default:
-        return new String.fromCharCode(char);
-    }
-  }
-}
-
-/// The contents of a string literal together with a strategy for printing it.
-class StringLiteralOutput {
-  /// Mix of [Expression] and `int`. Each expression is a string interpolation,
-  /// and each `int` is the character code of a character in a string literal.
-  final List parts;
-  final StringChunk chunk;
-
-  StringLiteralOutput(this.parts, this.chunk);
-}
-
-/// Strategy for printing a prefix of a string literal.
-/// A chunk represents the substring going from [:previous.endIndex:] to
-/// [endIndex] (or from 0 to [endIndex] if [previous] is null).
-class StringChunk {
-  final StringChunk previous;
-  final tree.StringQuoting quoting;
-  final int endIndex;
-
-  StringChunk(this.previous, this.quoting, this.endIndex);
-}
-
-/// [StringChunk] that has not yet been assigned an [endIndex].
-/// It additionally has a [cost] denoting the number of auxilliary characters
-/// (quotes, spaces, etc) needed to print the literal using this strategy
-class OpenStringChunk {
-  final StringChunk previous;
-  final tree.StringQuoting quoting;
-  num cost;
-
-  OpenStringChunk(this.previous, this.quoting, this.cost);
-
-  StringChunk end(int endIndex) {
-    return new StringChunk(previous, quoting, endIndex);
-  }
-}

pkg/compiler/lib/src/dart_backend/backend_ast_to_frontend_ast.dart

diff --git a/pkg/compiler/lib/src/dart_backend/backend_ast_to_frontend_ast.dart b/pkg/compiler/lib/src/dart_backend/backend_ast_to_frontend_ast.dart
deleted file mode 100644
index 469858d..0000000
--- a/pkg/compiler/lib/src/dart_backend/backend_ast_to_frontend_ast.dart
+++ /dev/null
@@ -1,1353 +0,0 @@
-// Copyright (c) 2014, the Dart project authors.  Please see the AUTHORS file
-// for details. All rights reserved. Use of this source code is governed by a
-// BSD-style license that can be found in the LICENSE file.
-
-library dart_tree_printer;
-
-import '../common.dart';
-import '../constants/values.dart' as values;
-import '../dart_types.dart' as types;
-import '../elements/elements.dart' as elements;
-import '../resolution/tree_elements.dart' show TreeElementMapping;
-import '../tokens/token.dart';
-import '../tokens/token_constants.dart';
-import '../tokens/precedence.dart';
-import '../tokens/precedence_constants.dart';
-import '../tree/tree.dart' as tree;
-import '../util/util.dart';
-import 'backend_ast_nodes.dart';
-import 'backend_ast_emitter.dart' show TypeGenerator;
-
-/// Translates the backend AST to Dart frontend AST.
-tree.Node emit(TreeElementMapping treeElements, RootNode root) {
-  return new TreePrinter(treeElements).makeDefinition(root);
-}
-
-/// If true, the unparser will insert a coment in front of every function
-/// it emits. This helps indicate which functions were translated by the new
-/// backend.
-bool INSERT_NEW_BACKEND_COMMENT =
-    const bool.fromEnvironment("INSERT_NEW_BACKEND_COMMENT");
-
-/// Converts backend ASTs to frontend ASTs.
-class TreePrinter {
-  TreeElementMapping treeElements;
-
-  TreePrinter([this.treeElements]);
-
-  tree.Node makeDefinition(RootNode node) {
-    if (node is FieldDefinition) {
-      tree.Node definition;
-      if (node.initializer == null) {
-        definition = makeIdentifier(node.element.name);
-      } else {
-        definition = new tree.SendSet(
-            null,
-            makeIdentifier(node.element.name),
-            new tree.Operator(assignmentToken("=")),
-            singleton(makeExpression(node.initializer)));
-      }
-      setElement(definition, node.element, node);
-      return new tree.VariableDefinitions(
-          null, // TODO(sigurdm): Type
-          makeVarModifiers(
-              useVar: true,
-              isFinal: node.element.isFinal,
-              isStatic: node.element.isStatic,
-              isConst: node.element.isConst),
-          makeList(null, [definition], close: semicolon));
-    } else if (node is FunctionExpression) {
-      return makeExpression(node);
-    } else {
-      assert(false);
-      return null;
-    }
-  }
-
-  void setElement(tree.Node node, elements.Element element, source) {
-    if (treeElements != null) {
-      if (element == null) {
-        throw "Missing element from ${source}";
-      }
-      treeElements[node] = element;
-    }
-  }
-
-  void setType(tree.Node node, types.DartType type, source) {
-    if (treeElements != null) {
-      if (type == null) {
-        throw "Missing type from ${source}";
-      }
-      treeElements.setType(node, type);
-    }
-  }
-
-  // Group tokens: () [] {} <>
-  static BeginGroupToken makeGroup(PrecedenceInfo open, PrecedenceInfo close) {
-    BeginGroupToken openTok = new BeginGroupToken(open, -1);
-    openTok.endGroup = new SymbolToken(close, -1);
-    return openTok;
-  }
-
-  final BeginGroupToken openParen =
-      makeGroup(OPEN_PAREN_INFO, CLOSE_PAREN_INFO);
-  final BeginGroupToken openBrace =
-      makeGroup(OPEN_CURLY_BRACKET_INFO, CLOSE_CURLY_BRACKET_INFO);
-  final BeginGroupToken openBracket =
-      makeGroup(OPEN_SQUARE_BRACKET_INFO, CLOSE_SQUARE_BRACKET_INFO);
-  final BeginGroupToken lt = makeGroup(LT_INFO, GT_INFO);
-
-  Token get closeParen => openParen.endGroup;
-  Token get closeBrace => openBrace.endGroup;
-  Token get closeBracket => openBracket.endGroup;
-  Token get gt => lt.endGroup;
-
-  // Symbol tokens
-  final Token semicolon = new SymbolToken(SEMICOLON_INFO, -1);
-  final Token indexToken = new SymbolToken(INDEX_INFO, -1); // "[]"
-  final Token question = new SymbolToken(QUESTION_INFO, -1);
-  final Token colon = new SymbolToken(COLON_INFO, -1);
-  final Token hash = new SymbolToken(HASH_INFO, -1);
-  final Token bang = new SymbolToken(BANG_INFO, -1);
-  final Token eq = new SymbolToken(EQ_INFO, -1);
-
-  // Keyword tokens
-  static Token makeIdToken(String text) {
-    return new StringToken.fromString(IDENTIFIER_INFO, text, -1);
-  }
-
-  final Token newToken = makeIdToken('new');
-  final Token constToken = makeIdToken('const');
-  final Token throwToken = makeIdToken('throw');
-  final Token rethrowToken = makeIdToken('rethrow');
-  final Token breakToken = makeIdToken('break');
-  final Token continueToken = makeIdToken('continue');
-  final Token doToken = makeIdToken('do');
-  final Token whileToken = makeIdToken('while');
-  final Token ifToken = makeIdToken('if');
-  final Token elseToken = makeIdToken('else');
-  final Token awaitToken = makeIdToken('await');
-  final Token forToken = makeIdToken('for');
-  final Token inToken = makeIdToken('in');
-  final Token returnToken = makeIdToken('return');
-  final Token switchToken = makeIdToken('switch');
-  final Token caseToken = makeIdToken('case');
-  final Token defaultToken = makeIdToken('default');
-  final Token tryToken = makeIdToken('try');
-  final Token catchToken = makeIdToken('catch');
-  final Token onToken = makeIdToken('on');
-  final Token finallyToken = makeIdToken('finally');
-  final Token getToken = makeIdToken('get');
-  final Token setToken = makeIdToken('set');
-  final Token classToken = makeIdToken('class');
-  final Token extendsToken = makeIdToken('extends');
-  final Token withToken = makeIdToken('with');
-  final Token implementsToken = makeIdToken('implements');
-  final Token typedefToken = makeIdToken('typedef');
-  final Token enumToken = makeIdToken('enum');
-
-  static tree.Identifier makeIdentifier(String name) {
-    return new tree.Identifier(
-        new StringToken.fromString(IDENTIFIER_INFO, name, -1));
-  }
-
-  static tree.Operator makeOperator(String name) {
-    return new tree.Operator(
-        new StringToken.fromString(IDENTIFIER_INFO, name, -1));
-  }
-
-  // Utilities for creating NodeLists
-  Link<tree.Node> makeLink(Iterable<tree.Node> nodes) {
-    LinkBuilder builder = new LinkBuilder();
-    for (tree.Node node in nodes) {
-      builder.addLast(node);
-    }
-    return builder.toLink();
-  }
-
-  tree.NodeList blankList() {
-    return new tree.NodeList(null, makeLink([]), null, '');
-  }
-
-  tree.NodeList singleton(tree.Node node) {
-    return new tree.NodeList(null, makeLink([node]), null, '');
-  }
-
-  tree.NodeList makeList(String delimiter, Iterable<tree.Node> nodes,
-      {Token open, Token close}) {
-    return new tree.NodeList(open, makeLink(nodes), close, delimiter);
-  }
-
-  tree.NodeList parenList(String delimiter, Iterable<tree.Node> nodes) {
-    return makeList(delimiter, nodes, open: openParen, close: closeParen);
-  }
-
-  tree.NodeList bracketList(String delimiter, Iterable<tree.Node> nodes) {
-    return makeList(delimiter, nodes, open: openBracket, close: closeBracket);
-  }
-
-  tree.NodeList braceList(String delimiter, Iterable<tree.Node> nodes) {
-    return makeList(delimiter, nodes, open: openBrace, close: closeBrace);
-  }
-
-  tree.NodeList argList(Iterable<tree.Node> nodes) {
-    return parenList(',', nodes);
-  }
-
-  tree.NodeList typeArgList(Iterable<tree.Node> nodes) {
-    return makeList(',', nodes, open: lt, close: gt);
-  }
-
-  /// Converts a qualified name into nested Sends.
-  tree.Node makeName(String name) {
-    if (name == null) {
-      return null;
-    }
-    List<String> names = name.split('.').toList(growable: false);
-    tree.Node node = makeIdentifier(names[0]);
-    for (int i = 1; i < names.length; i++) {
-      node = new tree.Send(node, makeIdentifier(names[i]));
-    }
-    return node;
-  }
-
-  static Token assignmentToken(String operatorName) {
-    switch (operatorName) {
-      case '=':
-        return new SymbolToken(EQ_INFO, -1);
-      case '+=':
-        return new SymbolToken(PLUS_EQ_INFO, -1);
-      case '-=':
-        return new SymbolToken(MINUS_EQ_INFO, -1);
-      case '*=':
-        return new SymbolToken(STAR_EQ_INFO, -1);
-      case '/=':
-        return new SymbolToken(SLASH_EQ_INFO, -1);
-      case '~/=':
-        return new SymbolToken(TILDE_SLASH_EQ_INFO, -1);
-      case '%=':
-        return new SymbolToken(PERCENT_EQ_INFO, -1);
-      case '&=':
-        return new SymbolToken(AMPERSAND_EQ_INFO, -1);
-      case '^=':
-        return new SymbolToken(CARET_EQ_INFO, -1);
-      case '|=':
-        return new SymbolToken(BAR_EQ_INFO, -1);
-      case '>>=':
-        return new SymbolToken(GT_GT_EQ_INFO, -1);
-      case '<<=':
-        return new SymbolToken(LT_LT_EQ_INFO, -1);
-      default:
-        throw "Unrecognized assignment operator: $operatorName";
-    }
-  }
-
-  static Token binopToken(String operatorName) {
-    switch (operatorName) {
-      case '+':
-        return new SymbolToken(PLUS_INFO, -1);
-      case '-':
-        return new SymbolToken(MINUS_INFO, -1);
-      case '*':
-        return new SymbolToken(STAR_INFO, -1);
-      case '/':
-        return new SymbolToken(SLASH_INFO, -1);
-      case '~/':
-        return new SymbolToken(TILDE_SLASH_INFO, -1);
-      case '%':
-        return new SymbolToken(PERCENT_INFO, -1);
-      case '&':
-        return new SymbolToken(AMPERSAND_INFO, -1);
-      case '^':
-        return new SymbolToken(CARET_INFO, -1);
-      case '|':
-        return new SymbolToken(BAR_INFO, -1);
-      case '>>':
-        return new SymbolToken(GT_GT_INFO, -1);
-      case '<<':
-        return new SymbolToken(LT_LT_INFO, -1);
-      case '==':
-        return new SymbolToken(EQ_EQ_INFO, -1);
-      case '!=':
-        return new SymbolToken(BANG_EQ_INFO, -1);
-      case '>':
-        return new SymbolToken(GT_INFO, -1);
-      case '>=':
-        return new SymbolToken(GT_EQ_INFO, -1);
-      case '<':
-        return new SymbolToken(LT_INFO, -1);
-      case '<=':
-        return new SymbolToken(LT_EQ_INFO, -1);
-      case '&&':
-        return new SymbolToken(AMPERSAND_AMPERSAND_INFO, -1);
-      case '||':
-        return new SymbolToken(BAR_BAR_INFO, -1);
-      default:
-        throw "Unrecognized binary operator: $operatorName";
-    }
-  }
-
-  static Token incrementToken(String operatorName) {
-    switch (operatorName) {
-      case '++':
-        return new SymbolToken(PLUS_PLUS_INFO, -1);
-      case '--':
-        return new SymbolToken(MINUS_MINUS_INFO, -1);
-      default:
-        throw "Unrecognized increment operator: $operatorName";
-    }
-  }
-
-  static Token typeOpToken(String operatorName) {
-    switch (operatorName) {
-      // "is!" is not an operator in the frontend AST.
-      case 'is':
-        return new SymbolToken(IS_INFO, -1);
-      case 'as':
-        return new SymbolToken(AS_INFO, -1);
-      default:
-        throw 'Unrecognized type operator: $operatorName';
-    }
-  }
-
-  Token unopToken(String operatorName) {
-    switch (operatorName) {
-      case '-':
-        return new SymbolToken(MINUS_INFO, -1);
-      case '~':
-        return new SymbolToken(TILDE_INFO, -1);
-      case '!':
-        return bang;
-      default:
-        throw "Unrecognized unary operator: $operatorName";
-    }
-  }
-
-  tree.Node makeStaticReceiver(elements.Element element) {
-    if (treeElements == null) return null;
-    if (element.isStatic) {
-      elements.ClassElement enclosingClass = element.enclosingClass;
-      tree.Send send = new tree.Send(null, makeIdentifier(enclosingClass.name));
-      treeElements[send] = enclosingClass;
-      return send;
-    } else {
-      return null;
-    }
-  }
-
-  tree.Node makeArgument(Argument arg) {
-    if (arg is Expression) {
-      return makeExpression(arg);
-    } else if (arg is NamedArgument) {
-      return new tree.NamedArgument(
-          makeIdentifier(arg.name), colon, makeExpression(arg.expression));
-    } else {
-      throw "Unrecognized argument type: ${arg}";
-    }
-  }
-
-  tree.Node makeExpression(Expression exp) {
-    return makeExp(exp, EXPRESSION);
-  }
-
-  /// Converts [exp] to a [tree.Node] that unparses to an expression with
-  /// a precedence level of at least [minPrecedence]. The expression will be
-  /// wrapped in a parenthesis if necessary.
-  tree.Node makeExp(Receiver exp, int minPrecedence, {bool beginStmt: false}) {
-    tree.Node result;
-    int precedence;
-    bool needParen = false;
-    if (exp is SuperReceiver) {
-      precedence = CALLEE;
-      result = makeIdentifier('super');
-    } else if (exp is Assignment) {
-      Expression left = exp.left;
-      tree.Node receiver;
-      tree.Node selector;
-      tree.NodeList arguments;
-      elements.Element element;
-      if (left is Identifier) {
-        receiver = makeStaticReceiver(left.element);
-        selector = makeIdentifier(left.name);
-        arguments = singleton(makeExpression(exp.right));
-        element = left.element;
-      } else if (left is FieldExpression) {
-        receiver = makeExp(left.object, PRIMARY, beginStmt: beginStmt);
-        selector = makeIdentifier(left.fieldName);
-        arguments = singleton(makeExpression(exp.right));
-      } else if (left is IndexExpression) {
-        receiver = makeExp(left.object, PRIMARY, beginStmt: beginStmt);
-        selector = new tree.Operator(indexToken);
-        arguments = bracketList(
-            ',', [makeExpression(left.index), makeExpression(exp.right)]);
-      } else {
-        throw "Unexpected left-hand side of assignment: ${left}";
-      }
-      tree.Operator op = new tree.Operator(assignmentToken(exp.operator));
-      result = new tree.SendSet(receiver, selector, op, arguments);
-      if (left is Identifier) {
-        setElement(result, element, exp);
-      }
-      precedence = EXPRESSION;
-    } else if (exp is FieldInitializer) {
-      precedence = EXPRESSION;
-      tree.Node receiver = makeIdentifier('this');
-      tree.Node selector = makeIdentifier(exp.element.name);
-      tree.Operator op = new tree.Operator(assignmentToken("="));
-      // We pass CALLEE to ensure we write eg.:
-      // class B { var x; B() : x = (() {return a;}) {}}
-      // Not the invalid:
-      // class B { var x; B() : x = () {return a;} {}}
-      result = new tree.SendSet(
-          receiver, selector, op, singleton(makeExp(exp.body, CALLEE)));
-      setElement(result, exp.element, exp);
-    } else if (exp is SuperInitializer) {
-      precedence = EXPRESSION;
-      tree.Node receiver = makeIdentifier('super');
-      tree.NodeList arguments =
-          argList(exp.arguments.map(makeArgument).toList());
-      if (exp.target.name == "") {
-        result = new tree.Send(null, receiver, arguments);
-      } else {
-        result =
-            new tree.Send(receiver, makeIdentifier(exp.target.name), arguments);
-      }
-      setElement(result, exp.target, exp);
-    } else if (exp is BinaryOperator) {
-      precedence = BINARY_PRECEDENCE[exp.operator];
-      int deltaLeft = isAssociativeBinaryOperator(precedence) ? 0 : 1;
-      result = new tree.Send(
-          makeExp(exp.left, precedence + deltaLeft, beginStmt: beginStmt),
-          new tree.Operator(binopToken(exp.operator)),
-          singleton(makeExp(exp.right, precedence + 1)));
-    } else if (exp is CallFunction) {
-      precedence = CALLEE;
-      tree.Node selector;
-      Expression callee = exp.callee;
-      elements.Element element;
-      tree.Node receiver;
-      if (callee is Identifier) {
-        receiver = makeStaticReceiver(callee.element);
-        selector = makeIdentifier(callee.name);
-        element = callee.element;
-      } else {
-        selector = makeExp(callee, CALLEE, beginStmt: beginStmt);
-      }
-      result = new tree.Send(
-          receiver, selector, argList(exp.arguments.map(makeArgument)));
-      if (callee is Identifier) {
-        setElement(result, element, exp);
-      }
-    } else if (exp is CallMethod) {
-      precedence = CALLEE;
-      // TODO(sra): Elide receiver when This, but only if not in a scope that
-      // shadows the method (e.g. constructor body).
-      tree.Node receiver = makeExp(exp.object, PRIMARY, beginStmt: beginStmt);
-      result = new tree.Send(receiver, makeIdentifier(exp.methodName),
-          argList(exp.arguments.map(makeArgument)));
-    } else if (exp is CallNew) {
-      precedence = CALLEE;
-      tree.Node selector = makeName(exp.type.name);
-      if (exp.type.typeArguments.length > 0) {
-        selector = new tree.TypeAnnotation(
-            selector, typeArgList(exp.type.typeArguments.map(makeType)));
-        setType(selector, exp.dartType, exp);
-      }
-      if (exp.constructorName != null) {
-        selector = new tree.Send(selector, makeIdentifier(exp.constructorName));
-      }
-      tree.Send send = new tree.Send(
-          null, selector, argList(exp.arguments.map(makeArgument)));
-      result =
-          new tree.NewExpression(exp.isConst ? constToken : newToken, send);
-      setType(result, exp.dartType, exp);
-      setElement(send, exp.constructor, exp);
-    } else if (exp is CallStatic) {
-      precedence = CALLEE;
-      result = new tree.Send(
-          makeStaticReceiver(exp.element),
-          makeIdentifier(exp.methodName),
-          argList(exp.arguments.map(makeArgument)));
-      setElement(result, exp.element, exp);
-    } else if (exp is Conditional) {
-      precedence = CONDITIONAL;
-      result = new tree.Conditional(
-          makeExp(exp.condition, LOGICAL_OR, beginStmt: beginStmt),
-          makeExp(exp.thenExpression, EXPRESSION),
-          makeExp(exp.elseExpression, EXPRESSION),
-          question,
-          colon);
-    } else if (exp is FieldExpression) {
-      precedence = PRIMARY;
-      // TODO(sra): Elide receiver when This, but only if not in a scope that
-      // shadows the method (e.g. constructor body).
-      tree.Node receiver = makeExp(exp.object, PRIMARY, beginStmt: beginStmt);
-      result = new tree.Send(receiver, makeIdentifier(exp.fieldName));
-    } else if (exp is ConstructorDefinition) {
-      precedence = EXPRESSION;
-      tree.NodeList parameters = makeParameters(exp.parameters);
-      tree.NodeList initializers =
-          exp.initializers == null || exp.initializers.isEmpty
-              ? null
-              : makeList(",", exp.initializers.map(makeExpression).toList());
-      tree.Node body = exp.isConst || exp.body == null
-          ? new tree.EmptyStatement(semicolon)
-          : makeFunctionBody(exp.body);
-      result = new tree.FunctionExpression(
-          constructorName(exp),
-          // GENERIC_METHODS: In order to support generic methods fully,
-          // we must retrieve and pass the actual type variables here.
-          null, // typeVariables
-          parameters,
-          body,
-          null, // return type
-          makeFunctionModifiers(exp),
-          initializers,
-          null, // get/set
-          null); // async modifier
-      setElement(result, exp.element, exp);
-    } else if (exp is FunctionExpression) {
-      precedence = PRIMARY;
-      if (beginStmt && exp.name != null) {
-        needParen = true; // Do not mistake for function declaration.
-      }
-      Token getOrSet = exp.isGetter ? getToken : exp.isSetter ? setToken : null;
-      tree.NodeList parameters =
-          exp.isGetter ? makeList("", []) : makeParameters(exp.parameters);
-      tree.Node body = makeFunctionBody(exp.body);
-      result = new tree.FunctionExpression(
-          functionName(exp),
-          // GENERIC_METHODS: In order to support generic methods fully,
-          // we must retrieve and pass the actual type variables here.
-          null, // typeVariables
-          parameters,
-          body,
-          exp.returnType == null || exp.element.isConstructor
-              ? null
-              : makeType(exp.returnType),
-          makeFunctionModifiers(exp),
-          null, // initializers
-          getOrSet, // get/set
-          null); // async modifier
-      elements.Element element = exp.element;
-      if (element != null) setElement(result, element, exp);
-    } else if (exp is Identifier) {
-      precedence = CALLEE;
-      result = new tree.Send(
-          makeStaticReceiver(exp.element), makeIdentifier(exp.name));
-      setElement(result, exp.element, exp);
-    } else if (exp is Increment) {
-      Expression lvalue = exp.expression;
-      tree.Node receiver;
-      tree.Node selector;
-      tree.Node argument;
-      bool innerBeginStmt = beginStmt && !exp.isPrefix;
-      if (lvalue is Identifier) {
-        receiver = makeStaticReceiver(lvalue.element);
-        selector = makeIdentifier(lvalue.name);
-      } else if (lvalue is FieldExpression) {
-        receiver = makeExp(lvalue.object, PRIMARY, beginStmt: innerBeginStmt);
-        selector = makeIdentifier(lvalue.fieldName);
-      } else if (lvalue is IndexExpression) {
-        receiver = makeExp(lvalue.object, PRIMARY, beginStmt: innerBeginStmt);
-        selector = new tree.Operator(indexToken);
-        argument = makeExpression(lvalue.index);
-      } else {
-        throw "Unrecognized left-hand side: ${lvalue}";
-      }
-      tree.Operator op = new tree.Operator(incrementToken(exp.operator));
-      if (exp.isPrefix) {
-        precedence = UNARY;
-        result = new tree.SendSet.prefix(receiver, selector, op, argument);
-      } else {
-        precedence = POSTFIX_INCREMENT;
-        result = new tree.SendSet.postfix(receiver, selector, op, argument);
-      }
-      if (lvalue is Identifier) {
-        setElement(result, lvalue.element, exp);
-      }
-    } else if (exp is IndexExpression) {
-      precedence = CALLEE;
-      result = new tree.Send(
-          makeExp(exp.object, PRIMARY, beginStmt: beginStmt),
-          new tree.Operator(indexToken),
-          bracketList(',', [makeExpression(exp.index)]));
-    } else if (exp is Literal) {
-      precedence = CALLEE;
-      values.PrimitiveConstantValue value = exp.value;
-      Token tok = new StringToken.fromString(
-          STRING_INFO, '${value.primitiveValue}', -1);
-      if (value.isString) {
-        result = unparseStringLiteral(exp);
-      } else if (value.isInt) {
-        result = new tree.LiteralInt(tok, null);
-      } else if (value.isDouble) {
-        if (value.primitiveValue == double.INFINITY) {
-          precedence = MULTIPLICATIVE;
-          tok = new StringToken.fromString(STRING_INFO, '1/0.0', -1);
-        } else if (value.primitiveValue == double.NEGATIVE_INFINITY) {
-          precedence = MULTIPLICATIVE;
-          tok = new StringToken.fromString(STRING_INFO, '-1/0.0', -1);
-        } else if (value.primitiveValue.isNaN) {
-          precedence = MULTIPLICATIVE;
-          tok = new StringToken.fromString(STRING_INFO, '0/0.0', -1);
-        }
-        result = new tree.LiteralDouble(tok, null);
-      } else if (value.isBool) {
-        result = new tree.LiteralBool(tok, null);
-      } else if (value.isNull) {
-        result = new tree.LiteralNull(tok);
-      } else {
-        throw "Unrecognized constant: ${value}";
-      }
-    } else if (exp is LiteralList) {
-      precedence = PRIMARY;
-      tree.NodeList typeArgs = null;
-      if (exp.typeArgument != null) {
-        typeArgs = typeArgList([makeType(exp.typeArgument)]);
-      }
-      result = new tree.LiteralList(
-          typeArgs,
-          bracketList(',', exp.values.map(makeExpression)),
-          exp.isConst ? constToken : null);
-    } else if (exp is LiteralMap) {
-      precedence = PRIMARY;
-      if (beginStmt) {
-        // The opening brace can be confused with a block statement.
-        needParen = true;
-      }
-      tree.NodeList typeArgs = null;
-      if (exp.typeArguments != null && exp.typeArguments.length > 0) {
-        typeArgs = typeArgList(exp.typeArguments.map(makeType));
-      }
-      result = new tree.LiteralMap(
-          typeArgs,
-          braceList(',', exp.entries.map(makeLiteralMapEntry)),
-          exp.isConst ? constToken : null);
-    } else if (exp is LiteralSymbol) {
-      precedence = PRIMARY;
-      result = new tree.LiteralSymbol(
-          hash, makeList('.', exp.id.split('.').map(makeIdentifier)));
-    } else if (exp is LiteralType) {
-      precedence = TYPE_LITERAL;
-      elements.Element optionalElement = exp.type.element;
-      result = new tree.Send(
-          optionalElement == null ? null : makeStaticReceiver(optionalElement),
-          makeIdentifier(exp.name));
-      treeElements.setType(result, exp.type);
-      if (optionalElement != null) {
-        // dynamic does not have an element
-        setElement(result, optionalElement, exp);
-      }
-    } else if (exp is ReifyTypeVar) {
-      precedence = PRIMARY;
-      result = new tree.Send(null, makeIdentifier(exp.name));
-      setElement(result, exp.element, exp);
-      setType(result, exp.element.type, exp);
-    } else if (exp is StringConcat) {
-      precedence = PRIMARY;
-      result = unparseStringLiteral(exp);
-    } else if (exp is This) {
-      precedence = CALLEE;
-      result = makeIdentifier('this');
-    } else if (exp is Throw) {
-      precedence = EXPRESSION; // ???
-      result = new tree.Throw(makeExpression(exp.expression), throwToken,
-          throwToken); // endToken not used by unparser
-    } else if (exp is TypeOperator) {
-      precedence = RELATIONAL;
-      tree.Operator operator;
-      tree.Node rightOperand = makeType(exp.type);
-      if (exp.operator == 'is!') {
-        operator = new tree.Operator(typeOpToken('is'));
-        rightOperand =
-            new tree.Send(rightOperand, new tree.Operator(bang), blankList());
-      } else {
-        operator = new tree.Operator(typeOpToken(exp.operator));
-      }
-      result = new tree.Send(
-          makeExp(exp.expression, BITWISE_OR, beginStmt: beginStmt),
-          operator,
-          singleton(rightOperand));
-    } else if (exp is UnaryOperator) {
-      precedence = UNARY;
-      result = new tree.Send.prefix(makeExp(exp.operand, UNARY),
-          new tree.Operator(unopToken(exp.operatorName)));
-    } else {
-      throw "Unknown expression type: ${exp}";
-    }
-
-    needParen = needParen || precedence < minPrecedence;
-    if (needParen) {
-      result = parenthesize(result);
-    }
-    return result;
-  }
-
-  /// Creates a LiteralString with [verbatim] as the value.
-  /// No (un)quoting or (un)escaping will be performed by this method.
-  /// The [DartString] inside the literal will be set to null because the
-  /// code emitter does not use it.
-  tree.LiteralString makeVerbatimStringLiteral(String verbatim) {
-    Token tok = new StringToken.fromString(STRING_INFO, verbatim, -1);
-    return new tree.LiteralString(tok, null);
-  }
-
-  tree.LiteralMapEntry makeLiteralMapEntry(LiteralMapEntry en) {
-    return new tree.LiteralMapEntry(
-        makeExpression(en.key), colon, makeExpression(en.value));
-  }
-
-  /// A comment token to be inserted when [INSERT_NEW_BACKEND_COMMENT] is true.
-  final SymbolToken newBackendComment = new SymbolToken(
-      const PrecedenceInfo('/* new backend */ ', 0, OPEN_CURLY_BRACKET_TOKEN),
-      -1);
-
-  tree.Node makeFunctionBody(Statement stmt) {
-    if (INSERT_NEW_BACKEND_COMMENT) {
-      return new tree.Block(
-          makeList('', [makeBlock(stmt)], open: newBackendComment));
-    } else {
-      return makeBlock(stmt);
-    }
-  }
-
-  /// Produces a statement in a context where only blocks are allowed.
-  tree.Node makeBlock(Statement stmt) {
-    if (stmt is Block || stmt is EmptyStatement) {
-      return makeStatement(stmt);
-    } else {
-      return new tree.Block(braceList('', [makeStatement(stmt)]));
-    }
-  }
-
-  /// Adds the given statement to a block. If the statement itself is a block
-  /// it will be flattened (if this does not change lexical scoping).
-  void addBlockMember(Statement stmt, List<tree.Node> accumulator) {
-    if (stmt is Block && !stmt.statements.any(Unparser.definesVariable)) {
-      for (Statement innerStmt in stmt.statements) {
-        addBlockMember(innerStmt, accumulator);
-      }
-    } else if (stmt is EmptyStatement) {
-      // No need to include empty statements inside blocks
-    } else {
-      accumulator.add(makeStatement(stmt));
-    }
-  }
-
-  /// True if [stmt] is equivalent to an empty statement.
-  bool isEmptyStatement(Statement stmt) {
-    return stmt is EmptyStatement ||
-        (stmt is Block && stmt.statements.every(isEmptyStatement));
-  }
-
-  tree.Node makeStatement(Statement stmt, {bool shortIf: true}) {
-    if (stmt is Block) {
-      List<tree.Node> body = <tree.Node>[];
-      for (Statement innerStmt in stmt.statements) {
-        addBlockMember(innerStmt, body);
-      }
-      return new tree.Block(braceList('', body));
-    } else if (stmt is Break) {
-      return new tree.BreakStatement(
-          stmt.label == null ? null : makeIdentifier(stmt.label),
-          breakToken,
-          semicolon);
-    } else if (stmt is Continue) {
-      return new tree.ContinueStatement(
-          stmt.label == null ? null : makeIdentifier(stmt.label),
-          continueToken,
-          semicolon);
-    } else if (stmt is DoWhile) {
-      return new tree.DoWhile(
-          makeStatement(stmt.body, shortIf: shortIf),
-          parenthesize(makeExpression(stmt.condition)),
-          doToken,
-          whileToken,
-          semicolon);
-    } else if (stmt is EmptyStatement) {
-      return new tree.EmptyStatement(semicolon);
-    } else if (stmt is ExpressionStatement) {
-      return new tree.ExpressionStatement(
-          makeExp(stmt.expression, EXPRESSION, beginStmt: true), semicolon);
-    } else if (stmt is For) {
-      tree.Node initializer;
-      if (stmt.initializer is VariableDeclarations) {
-        initializer = makeVariableDeclarations(stmt.initializer);
-      } else if (stmt.initializer is Expression) {
-        initializer = makeExpression(stmt.initializer);
-      } else {
-        initializer = null;
-      }
-      tree.Node condition;
-      if (stmt.condition != null) {
-        condition = new tree.ExpressionStatement(
-            makeExpression(stmt.condition), semicolon);
-      } else {
-        condition = new tree.EmptyStatement(semicolon);
-      }
-      return new tree.For(
-          initializer,
-          condition,
-          makeList(',', stmt.updates.map(makeExpression)),
-          makeStatement(stmt.body, shortIf: shortIf),
-          forToken);
-    } else if (stmt is ForIn) {
-      tree.Node left;
-      if (stmt.leftHandValue is Identifier) {
-        left = makeExpression(stmt.leftHandValue);
-      } else {
-        left = makeVariableDeclarations(stmt.leftHandValue);
-      }
-      return new tree.SyncForIn(left, makeExpression(stmt.expression),
-          makeStatement(stmt.body, shortIf: shortIf), forToken, inToken);
-    } else if (stmt is FunctionDeclaration) {
-      tree.FunctionExpression function = new tree.FunctionExpression(
-          stmt.name != null ? makeIdentifier(stmt.name) : null,
-          // GENERIC_METHODS: In order to support generic methods fully,
-          // we must retrieve and pass the actual type variables here.
-          null, // typeVariables
-          makeParameters(stmt.parameters),
-          makeFunctionBody(stmt.body),
-          stmt.returnType != null ? makeType(stmt.returnType) : null,
-          makeEmptyModifiers(),
-          null, // initializers
-          null, // get/set
-          null); // async modifier
-      setElement(function, stmt.function.element, stmt);
-      return new tree.FunctionDeclaration(function);
-    } else if (stmt is If) {
-      if (stmt.elseStatement == null || isEmptyStatement(stmt.elseStatement)) {
-        tree.Node node = new tree.If(
-            parenthesize(makeExpression(stmt.condition)),
-            makeStatement(stmt.thenStatement),
-            null, // else statement
-            ifToken,
-            null); // else token
-        if (shortIf)
-          return node;
-        else
-          return new tree.Block(braceList('', [node]));
-      } else {
-        return new tree.If(
-            parenthesize(makeExpression(stmt.condition)),
-            makeStatement(stmt.thenStatement, shortIf: false),
-            makeStatement(stmt.elseStatement, shortIf: shortIf),
-            ifToken,
-            elseToken); // else token
-      }
-    } else if (stmt is LabeledStatement) {
-      List<tree.Label> labels = [];
-      Statement inner = stmt;
-      while (inner is LabeledStatement) {
-        LabeledStatement lbl = inner as LabeledStatement;
-        labels.add(new tree.Label(makeIdentifier(lbl.label), colon));
-        inner = lbl.statement;
-      }
-      return new tree.LabeledStatement(
-          makeList('', labels), makeStatement(inner, shortIf: shortIf));
-    } else if (stmt is Rethrow) {
-      return new tree.Rethrow(rethrowToken, semicolon);
-    } else if (stmt is Return) {
-      return new tree.Return(returnToken, semicolon,
-          stmt.expression == null ? null : makeExpression(stmt.expression));
-    } else if (stmt is Switch) {
-      return new tree.SwitchStatement(
-          parenthesize(makeExpression(stmt.expression)),
-          braceList('', stmt.cases.map(makeSwitchCase)),
-          switchToken);
-    } else if (stmt is Try) {
-      return new tree.TryStatement(
-          makeBlock(stmt.tryBlock),
-          makeList(null, stmt.catchBlocks.map(makeCatchBlock)),
-          stmt.finallyBlock == null ? null : makeBlock(stmt.finallyBlock),
-          tryToken,
-          stmt.finallyBlock == null ? null : finallyToken);
-    } else if (stmt is VariableDeclarations) {
-      return makeVariableDeclarations(stmt, useVar: true, endToken: semicolon);
-    } else if (stmt is While) {
-      return new tree.While(parenthesize(makeExpression(stmt.condition)),
-          makeStatement(stmt.body, shortIf: shortIf), whileToken);
-    } else {
-      throw "Unrecognized statement: ${stmt}";
-    }
-  }
-
-  tree.Node makeVariableDeclaration(VariableDeclaration vd) {
-    tree.Node id = makeIdentifier(vd.name);
-    setElement(id, vd.element, vd);
-    if (vd.initializer == null) {
-      return id;
-    }
-    tree.Node send = new tree.SendSet(null, id, new tree.Operator(eq),
-        singleton(makeExpression(vd.initializer)));
-    setElement(send, vd.element, vd);
-    return send;
-  }
-
-  /// If [useVar] is true, the variable definition will use `var` as modifier
-  /// if no other modifiers are present.
-  /// [endToken] will be used to terminate the declaration list.
-  tree.Node makeVariableDeclarations(VariableDeclarations decl,
-      {bool useVar: false, Token endToken: null}) {
-    return new tree.VariableDefinitions(
-        decl.type == null ? null : makeType(decl.type),
-        makeVarModifiers(
-            isConst: decl.isConst,
-            isFinal: decl.isFinal,
-            useVar: useVar && decl.type == null),
-        makeList(',', decl.declarations.map(makeVariableDeclaration),
-            close: endToken));
-  }
-
-  tree.CatchBlock makeCatchBlock(CatchBlock block) {
-    List<tree.VariableDefinitions> formals = [];
-    if (block.exceptionVar != null) {
-      tree.Node exceptionName = makeIdentifier(block.exceptionVar.name);
-      setElement(exceptionName, block.exceptionVar.element, block.exceptionVar);
-      formals.add(new tree.VariableDefinitions(
-          null, makeEmptyModifiers(), singleton(exceptionName)));
-    }
-    if (block.stackVar != null) {
-      tree.Node stackTraceName = makeIdentifier(block.stackVar.name);
-      setElement(stackTraceName, block.stackVar.element, block.stackVar);
-      formals.add(new tree.VariableDefinitions(
-          null, makeEmptyModifiers(), singleton(stackTraceName)));
-    }
-    return new tree.CatchBlock(
-        block.onType == null ? null : makeType(block.onType),
-        block.exceptionVar == null ? null : argList(formals),
-        makeBlock(block.body),
-        block.onType == null ? null : onToken,
-        block.exceptionVar == null ? null : catchToken);
-  }
-
-  tree.SwitchCase makeSwitchCase(SwitchCase caze) {
-    if (caze.isDefaultCase) {
-      return new tree.SwitchCase(
-          blankList(),
-          defaultToken,
-          makeList('', caze.statements.map(makeStatement)),
-          null); // startToken unused by unparser
-    } else {
-      return new tree.SwitchCase(
-          makeList('', caze.expressions.map(makeCaseMatch)),
-          null, // defaultKeyword,
-          makeList('', caze.statements.map(makeStatement)),
-          null); // startToken unused by unparser
-    }
-  }
-
-  tree.CaseMatch makeCaseMatch(Expression exp) {
-    return new tree.CaseMatch(caseToken, makeExpression(exp), colon);
-  }
-
-  tree.TypeAnnotation makeType(TypeAnnotation type) {
-    tree.NodeList typeArgs;
-    if (type.typeArguments.length > 0) {
-      typeArgs = typeArgList(type.typeArguments.map(makeType));
-    } else {
-      typeArgs = null;
-    }
-    tree.TypeAnnotation result =
-        new tree.TypeAnnotation(makeIdentifier(type.name), typeArgs);
-    setType(result, type.dartType, type);
-    return result;
-  }
-
-  tree.NodeList makeParameters(Parameters params) {
-    List<tree.Node> nodes =
-        params.requiredParameters.map(makeParameter).toList();
-    if (params.hasOptionalParameters) {
-      Token assign = params.hasNamedParameters ? colon : eq;
-      Token open = params.hasNamedParameters ? openBrace : openBracket;
-      Token close = params.hasNamedParameters ? closeBrace : closeBracket;
-      Iterable<tree.Node> opt =
-          params.optionalParameters.map((p) => makeParameter(p, assign));
-      nodes.add(new tree.NodeList(open, makeLink(opt), close, ','));
-    }
-    return argList(nodes);
-  }
-
-  /// [assignOperator] is used for writing the default value.
-  tree.Node makeParameter(Parameter param, [Token assignOperator]) {
-    if (param.isFunction) {
-      tree.Node definition = new tree.FunctionExpression(
-          makeIdentifier(param.name),
-          // GENERIC_METHODS: In order to support generic methods fully,
-          // we must retrieve and pass the actual type variables here.
-          null, // typeVariables
-          makeParameters(param.parameters),
-          null, // body
-              param.type == null ? null : makeType(param.type),
-          makeEmptyModifiers(), // TODO: Function parameter modifiers?
-          null, // initializers
-          null, // get/set
-          null); // async modifier
-      if (param.element != null) {
-        setElement(definition, param.element, param);
-      }
-      if (param.defaultValue != null) {
-        definition = new tree.SendSet(
-            null,
-            definition,
-            new tree.Operator(assignOperator),
-            singleton(makeExpression(param.defaultValue)));
-      }
-      return new tree.VariableDefinitions(
-          null, makeEmptyModifiers(), singleton(definition));
-    } else {
-      tree.Node definition;
-      if (param.defaultValue != null) {
-        definition = new tree.SendSet(
-            null,
-            makeIdentifier(param.name),
-            new tree.Operator(assignOperator),
-            singleton(makeExpression(param.defaultValue)));
-      } else {
-        definition = makeIdentifier(param.name);
-      }
-      if (param.element != null) {
-        setElement(definition, param.element, param);
-      }
-      return new tree.VariableDefinitions(
-          param.type == null ? null : makeType(param.type),
-          makeEmptyModifiers(), // TODO: Parameter modifiers?
-          singleton(definition));
-    }
-  }
-
-  tree.Modifiers makeEmptyModifiers() {
-    return new tree.Modifiers(blankList());
-  }
-
-  tree.Modifiers makeModifiers(
-      {bool isExternal: false,
-      bool isStatic: false,
-      bool isAbstract: false,
-      bool isFactory: false,
-      bool isConst: false,
-      bool isFinal: false,
-      bool isVar: false}) {
-    List<tree.Node> nodes = [];
-    if (isExternal) {
-      nodes.add(makeIdentifier('external'));
-    }
-    if (isStatic) {
-      nodes.add(makeIdentifier('static'));
-    }
-    if (isAbstract) {
-      nodes.add(makeIdentifier('abstract'));
-    }
-    if (isFactory) {
-      nodes.add(makeIdentifier('factory'));
-    }
-    if (isConst) {
-      nodes.add(makeIdentifier('const'));
-    }
-    if (isFinal) {
-      nodes.add(makeIdentifier('final'));
-    }
-    if (isVar) {
-      nodes.add(makeIdentifier('var'));
-    }
-    return new tree.Modifiers(makeList(' ', nodes));
-  }
-
-  tree.Modifiers makeVarModifiers(
-      {bool isConst: false,
-      bool isFinal: false,
-      bool useVar: false,
-      bool isStatic: false}) {
-    return makeModifiers(
-        isStatic: isStatic,
-        isConst: isConst,
-        isFinal: isFinal,
-        isVar: useVar && !(isConst || isFinal));
-  }
-
-  tree.Modifiers makeFunctionModifiers(FunctionExpression exp) {
-    if (exp.element == null) return makeEmptyModifiers();
-    return makeModifiers(
-        isExternal: exp.element.isExternal,
-        isStatic: exp.element.isStatic,
-        isFactory: exp.element.isFactoryConstructor,
-        isConst: exp.element.isConst);
-  }
-
-  tree.Node makeNodeForClassElement(elements.ClassElement cls) {
-    if (cls.isMixinApplication) {
-      return makeNamedMixinApplication(cls);
-    } else if (cls.isEnumClass) {
-      return makeEnum(cls);
-    } else {
-      return makeClassNode(cls);
-    }
-  }
-
-  tree.Typedef makeTypedef(elements.TypedefElement typdef) {
-    types.FunctionType functionType = typdef.alias;
-    final tree.TypeAnnotation returnType =
-        makeType(TypeGenerator.createType(functionType.returnType));
-
-    final tree.Identifier name = makeIdentifier(typdef.name);
-    final tree.NodeList typeParameters =
-        makeTypeParameters(typdef.typeVariables);
-    final tree.NodeList formals =
-        makeParameters(TypeGenerator.createParametersFromType(functionType));
-
-    final Token typedefKeyword = typedefToken;
-    final Token endToken = semicolon;
-
-    return new tree.Typedef(
-        returnType, name, typeParameters, formals, typedefKeyword, endToken);
-  }
-
-  /// Create a [tree.NodeList] containing the type variable declarations in
-  /// [typeVaraiables.
-  tree.NodeList makeTypeParameters(List<types.DartType> typeVariables) {
-    if (typeVariables.isEmpty) {
-      return new tree.NodeList.empty();
-    } else {
-      List<tree.Node> typeVariableList = <tree.Node>[];
-      for (types.TypeVariableType typeVariable in typeVariables) {
-        tree.Node id = makeIdentifier(typeVariable.name);
-        treeElements[id] = typeVariable.element;
-        tree.Node bound;
-        if (!typeVariable.element.bound.isObject) {
-          bound =
-              makeType(TypeGenerator.createType(typeVariable.element.bound));
-        }
-        tree.TypeVariable node = new tree.TypeVariable(id, extendsToken, bound);
-        treeElements.setType(node, typeVariable);
-        typeVariableList.add(node);
-      }
-      return makeList(',', typeVariableList, open: lt, close: gt);
-    }
-  }
-
-  /// Create a [tree.NodeList] containing the declared interfaces.
-  ///
-  /// [interfaces] is from [elements.ClassElement] in reverse declaration order
-  /// and it contains mixins. To produce a list of the declared interfaces only,
-  /// interfaces in [mixinTypes] are omitted.
-  ///
-  /// [forNamedMixinApplication] is because the structure of the [tree.NodeList]
-  /// differs between [tree.NamedMixinApplication] and [tree.ClassNode].
-  // TODO(johnniwinther): Normalize interfaces on[tree.NamedMixinApplication]
-  // and [tree.ClassNode].
-  tree.NodeList makeInterfaces(
-      Link<types.DartType> interfaces, Set<types.DartType> mixinTypes,
-      {bool forNamedMixinApplication: false}) {
-    Link<tree.Node> typeAnnotations = const Link<tree.Node>();
-    for (Link<types.DartType> link = interfaces;
-        !link.isEmpty;
-        link = link.tail) {
-      types.DartType interface = link.head;
-      if (!mixinTypes.contains(interface)) {
-        typeAnnotations = typeAnnotations
-            .prepend(makeType(TypeGenerator.createType(interface)));
-      }
-    }
-    if (typeAnnotations.isEmpty) {
-      return forNamedMixinApplication ? null : new tree.NodeList.empty();
-    } else {
-      return new tree.NodeList(
-          forNamedMixinApplication ? null : implementsToken,
-          typeAnnotations,
-          null,
-          ',');
-    }
-  }
-
-  /// Creates a [tree.NamedMixinApplication] node for [cls].
-  // TODO(johnniwinther): Unify creation of mixin lists between
-  // [NamedMixinApplicationElement] and [ClassElement].
-  tree.NamedMixinApplication makeNamedMixinApplication(
-      elements.MixinApplicationElement cls) {
-    assert(invariant(cls, !cls.isUnnamedMixinApplication,
-        message: "Cannot create ClassNode for unnamed mixin application "
-            "$cls."));
-    tree.Modifiers modifiers = makeModifiers(isAbstract: cls.isAbstract);
-    tree.Identifier name = makeIdentifier(cls.name);
-    tree.NodeList typeParameters = makeTypeParameters(cls.typeVariables);
-
-    Set<types.DartType> mixinTypes = new Set<types.DartType>();
-    Link<tree.Node> mixins = const Link<tree.Node>();
-
-    void addMixin(types.DartType mixinType) {
-      mixinTypes.add(mixinType);
-      mixins = mixins.prepend(makeType(TypeGenerator.createType(mixinType)));
-    }
-
-    addMixin(cls.mixinType);
-
-    tree.Node superclass;
-    types.InterfaceType supertype = cls.supertype;
-    while (supertype.element.isUnnamedMixinApplication) {
-      elements.MixinApplicationElement mixinApplication = supertype.element;
-      addMixin(cls.asInstanceOf(mixinApplication.mixin));
-      supertype = mixinApplication.supertype;
-    }
-    superclass =
-        makeType(TypeGenerator.createType(cls.asInstanceOf(supertype.element)));
-    tree.Node supernode = new tree.MixinApplication(
-        superclass, new tree.NodeList(null, mixins, null, ','));
-
-    tree.NodeList interfaces = makeInterfaces(cls.interfaces, mixinTypes,
-        forNamedMixinApplication: true);
-
-    return new tree.NamedMixinApplication(name, typeParameters, modifiers,
-        supernode, interfaces, classToken, semicolon);
-  }
-
-  tree.Enum makeEnum(elements.EnumClassElement cls) {
-    return new tree.Enum(
-        enumToken,
-        makeIdentifier(cls.name),
-        makeList(',', cls.enumValues.map((e) => makeIdentifier(e.name)),
-            open: openBrace, close: closeBrace));
-  }
-
-  /// Creates a [tree.ClassNode] node for [cls].
-  tree.ClassNode makeClassNode(elements.ClassElement cls) {
-    assert(invariant(cls, !cls.isUnnamedMixinApplication,
-        message: "Cannot create ClassNode for unnamed mixin application "
-            "$cls."));
-    tree.Modifiers modifiers = makeModifiers(isAbstract: cls.isAbstract);
-    tree.Identifier name = makeIdentifier(cls.name);
-    tree.NodeList typeParameters = makeTypeParameters(cls.typeVariables);
-    tree.Node supernode;
-    types.InterfaceType supertype = cls.supertype;
-    Set<types.DartType> mixinTypes = new Set<types.DartType>();
-    Link<tree.Node> mixins = const Link<tree.Node>();
-
-    void addMixin(types.DartType mixinType) {
-      mixinTypes.add(mixinType);
-      mixins = mixins.prepend(makeType(TypeGenerator.createType(mixinType)));
-    }
-
-    if (supertype != null) {
-      if (supertype.element.isUnnamedMixinApplication) {
-        while (supertype.element.isUnnamedMixinApplication) {
-          elements.MixinApplicationElement mixinApplication = supertype.element;
-          addMixin(cls.asInstanceOf(mixinApplication.mixin));
-          supertype = mixinApplication.supertype;
-        }
-        tree.Node superclass = makeType(
-            TypeGenerator.createType(cls.asInstanceOf(supertype.element)));
-        supernode = new tree.MixinApplication(
-            superclass, new tree.NodeList(null, mixins, null, ','));
-      } else if (!supertype.isObject) {
-        supernode = makeType(TypeGenerator.createType(supertype));
-      }
-    }
-    tree.NodeList interfaces = makeInterfaces(cls.interfaces, mixinTypes);
-
-    Token extendsKeyword = supernode != null ? extendsToken : null;
-    return new tree.ClassNode(
-        modifiers,
-        name,
-        typeParameters,
-        supernode,
-        interfaces,
-        openBrace,
-        extendsKeyword,
-        null, // No body.
-        closeBrace);
-  }
-
-  tree.Node constructorName(ConstructorDefinition exp) {
-    String name = exp.name;
-    tree.Identifier className = makeIdentifier(exp.element.enclosingClass.name);
-    tree.Node result =
-        name == "" ? className : new tree.Send(className, makeIdentifier(name));
-    setElement(result, exp.element, exp);
-    return result;
-  }
-
-  tree.Node functionName(FunctionExpression exp) {
-    String name = exp.name;
-    if (name == null) return null;
-    if (isUserDefinableOperator(name)) {
-      return makeOperator("operator$name");
-    } else if (name == "unary-") {
-      return makeOperator("operator-");
-    }
-    return makeIdentifier(name);
-  }
-
-  tree.Node parenthesize(tree.Node node) {
-    return new tree.ParenthesizedExpression(node, openParen);
-  }
-
-  tree.Node unparseStringLiteral(Expression exp) {
-    StringLiteralOutput output = Unparser.analyzeStringLiteral(exp);
-    List parts = output.parts;
-    tree.Node printStringChunk(StringChunk chunk) {
-      bool raw = chunk.quoting.raw;
-      int quoteCode = chunk.quoting.quote;
-
-      List<tree.StringInterpolationPart> literalParts = [];
-      tree.LiteralString firstLiteral;
-      tree.Node currentInterpolation;
-
-      // sb contains the current unfinished LiteralString
-      StringBuffer sb = new StringBuffer();
-      if (raw) {
-        sb.write('r');
-      }
-      for (int i = 0; i < chunk.quoting.leftQuoteCharCount; i++) {
-        sb.write(chunk.quoting.quoteChar);
-      }
-
-      // Print every character and string interpolation
-      int startIndex = chunk.previous != null ? chunk.previous.endIndex : 0;
-      for (int i = startIndex; i < chunk.endIndex; i++) {
-        var part = parts[i];
-        if (part is Expression) {
-          // Finish the previous string interpolation, if there is one.
-          tree.LiteralString lit = makeVerbatimStringLiteral(sb.toString());
-          if (currentInterpolation != null) {
-            literalParts.add(
-                new tree.StringInterpolationPart(currentInterpolation, lit));
-          } else {
-            firstLiteral = lit;
-          }
-          sb.clear();
-          currentInterpolation = makeExpression(part);
-        } else {
-          int char = part;
-          sb.write(Unparser.getEscapedCharacter(char, quoteCode, raw));
-        }
-      }
-
-      // Print ending quotes
-      for (int i = 0; i < chunk.quoting.rightQuoteLength; i++) {
-        sb.write(chunk.quoting.quoteChar);
-      }
-
-      // Finish the previous string interpolation, if there is one.
-      // Then wrap everything in a StringInterpolation, if relevant.
-      tree.LiteralString lit = makeVerbatimStringLiteral(sb.toString());
-      tree.Node node;
-      if (firstLiteral == null) {
-        node = lit;
-      } else {
-        literalParts
-            .add(new tree.StringInterpolationPart(currentInterpolation, lit));
-        node = new tree.StringInterpolation(
-            firstLiteral, makeList('', literalParts));
-      }
-
-      // Juxtapose with the previous string chunks, if any.
-      if (chunk.previous != null) {
-        return new tree.StringJuxtaposition(
-            printStringChunk(chunk.previous), node);
-      } else {
-        return node;
-      }
-    }
-    return printStringChunk(output.chunk);
-  }
-}

pkg/compiler/lib/src/dart_backend/dart_backend.dart

diff --git a/pkg/compiler/lib/src/dart_backend/dart_backend.dart b/pkg/compiler/lib/src/dart_backend/dart_backend.dart
deleted file mode 100644
index 729091f..0000000
--- a/pkg/compiler/lib/src/dart_backend/dart_backend.dart
+++ /dev/null
@@ -1,40 +0,0 @@
-// Copyright (c) 2012, the Dart project authors.  Please see the AUTHORS file
-// for details. All rights reserved. Use of this source code is governed by a
-// BSD-style license that can be found in the LICENSE file.
-
-library dart_backend;
-
-import 'dart:async' show Future;
-import 'dart:math' show max;
-
-import '../../compiler.dart' show CompilerOutputProvider;
-import '../common.dart';
-import '../common/backend_api.dart' show Backend, ImpactTransformer;
-import '../common/codegen.dart' show CodegenWorkItem;
-import '../common/names.dart' show Selectors, Uris;
-import '../common/registry.dart' show Registry;
-import '../common/resolution.dart' show Resolution, ResolutionImpact;
-import '../common/tasks.dart' show CompilerTask;
-import '../compiler.dart' show Compiler;
-import '../compile_time_constants.dart';
-import '../constants/constant_system.dart';
-import '../constants/expressions.dart';
-import '../constants/values.dart';
-import '../dart_types.dart';
-import '../elements/elements.dart';
-import '../enqueue.dart' show Enqueuer, ResolutionEnqueuer;
-import '../library_loader.dart' show LoadedLibraries;
-import '../mirror_renamer/mirror_renamer.dart';
-import '../resolution/tree_elements.dart' show TreeElements, TreeElementMapping;
-import '../tokens/keyword.dart' show Keyword;
-import '../tree/tree.dart';
-import '../universe/selector.dart' show Selector;
-import '../universe/use.dart' show DynamicUse, TypeUse, TypeUseKind;
-import '../universe/world_impact.dart' show WorldImpact, TransformedWorldImpact;
-import '../util/util.dart';
-import 'backend_ast_to_frontend_ast.dart' as backend2frontend;
-
-part 'backend.dart';
-part 'renamer.dart';
-part 'placeholder_collector.dart';
-part 'outputter.dart';

pkg/compiler/lib/src/dart_backend/outputter.dart

diff --git a/pkg/compiler/lib/src/dart_backend/outputter.dart b/pkg/compiler/lib/src/dart_backend/outputter.dart
deleted file mode 100644
index ce9022a..0000000
--- a/pkg/compiler/lib/src/dart_backend/outputter.dart
+++ /dev/null
@@ -1,576 +0,0 @@
-// Copyright (c) 2014, the Dart project authors.  Please see the AUTHORS file
-// for details. All rights reserved. Use of this source code is governed by a
-// BSD-style license that can be found in the LICENSE file.
-
-part of dart_backend;
-
-typedef bool IsSafeToRemoveTypeDeclarations(
-    Map<ClassElement, Iterable<Element>> classMembers);
-typedef void ElementCallback<E>(E element);
-typedef void ElementPostProcessFunction(
-    AstElement element,
-    ElementAst elementAst,
-    ElementCallback<TypedefElement> typedefCallback,
-    ElementCallback<ClassElement> classCallback);
-typedef ElementAst ComputeElementAstFunction(AstElement element);
-typedef bool ElementFilter(Element element);
-typedef List<Element> ElementSorter(Iterable<Element> elements);
-
-/// Output engine for dart2dart that is shared between the dart2js and the
-/// analyzer implementations of dart2dart.
-class DartOutputter {
-  final DiagnosticReporter reporter;
-  final CompilerOutputProvider outputProvider;
-  final bool forceStripTypes;
-
-  // TODO(antonm): make available from command-line options.
-  final bool outputAst = false;
-  final bool enableMinification;
-
-  /// If `true`, libraries are generated into separate files.
-  final bool multiFile;
-
-  /// Internal structures accessible for tests and logging.
-  // TODO(johnniwinther): Clean this up.
-  PlaceholderRenamer renamer;
-  MainOutputGenerator output;
-  LibraryInfo libraryInfo;
-  ElementInfo elementInfo;
-
-  // TODO(johnniwinther): Support recompilation.
-  DartOutputter(this.reporter, this.outputProvider,
-      {bool this.forceStripTypes: false,
-      bool this.enableMinification: false,
-      bool this.multiFile: false});
-
-  /// Generate Dart code for the program starting at [mainFunction].
-  ///
-  /// [libraries] is the set of all libraries (user/package/sdk) that are
-  /// referenced in the program.
-  ///
-  /// [instantiatedClasses] is the set of classes that are potentially
-  /// instantiated in the program.
-  ///
-  /// [resolvedElements] is the set of methods, constructors, and fields that
-  /// are potentially accessed/called in the program.
-  ///
-  /// The [sortElements] function is used to sort [instantiatedClasses] and
-  /// [resolvedElements] in the generated output.
-  ///
-  /// Returns the total size of the generated code.
-  int assembleProgram(
-      {MirrorRenamer mirrorRenamer: const MirrorRenamer(),
-      Iterable<LibraryElement> libraries,
-      Iterable<Element> instantiatedClasses,
-      Iterable<Element> resolvedElements,
-      Iterable<ClassElement> usedTypeLiterals: const <ClassElement>[],
-      FunctionElement mainFunction,
-      Uri outputUri,
-      ElementPostProcessFunction postProcessElementAst,
-      ComputeElementAstFunction computeElementAst,
-      ElementFilter shouldOutput,
-      IsSafeToRemoveTypeDeclarations isSafeToRemoveTypeDeclarations,
-      ElementSorter sortElements}) {
-    assert(invariant(NO_LOCATION_SPANNABLE, libraries != null,
-        message: "'libraries' must be non-null."));
-    assert(invariant(NO_LOCATION_SPANNABLE, instantiatedClasses != null,
-        message: "'instantiatedClasses' must be non-null."));
-    assert(invariant(NO_LOCATION_SPANNABLE, resolvedElements != null,
-        message: "'resolvedElements' must be non-null."));
-    assert(invariant(NO_LOCATION_SPANNABLE, mainFunction != null,
-        message: "'mainFunction' must be non-null."));
-    assert(invariant(NO_LOCATION_SPANNABLE, computeElementAst != null,
-        message: "'computeElementAst' must be non-null."));
-    assert(invariant(NO_LOCATION_SPANNABLE, shouldOutput != null,
-        message: "'shouldOutput' must be non-null."));
-    assert(invariant(
-        NO_LOCATION_SPANNABLE, isSafeToRemoveTypeDeclarations != null,
-        message: "'isSafeToRemoveTypeDeclarations' must be non-null."));
-
-    if (sortElements == null) {
-      // Ensure deterministic output order.
-      sortElements = (Iterable<Element> elements) {
-        List<Element> list = elements.toList();
-        list.sort((Element a, Element b) => a.name.compareTo(b.name));
-        return list;
-      };
-    }
-
-    libraryInfo =
-        LibraryInfo.processLibraries(reporter, libraries, resolvedElements);
-
-    elementInfo = ElementInfoProcessor.createElementInfo(
-        instantiatedClasses, resolvedElements, usedTypeLiterals,
-        postProcessElementAst: postProcessElementAst,
-        parseElementAst: computeElementAst,
-        shouldOutput: shouldOutput,
-        sortElements: sortElements);
-
-    PlaceholderCollector collector = collectPlaceholders(
-        reporter, mirrorRenamer, mainFunction, libraryInfo, elementInfo);
-
-    renamer = createRenamer(collector, libraryInfo, elementInfo,
-        enableMinification: enableMinification,
-        forceStripTypes: forceStripTypes,
-        isSafeToRemoveTypeDeclarations: isSafeToRemoveTypeDeclarations);
-
-    if (outputAst) {
-      String code = astOutput(reporter, elementInfo);
-      outputProvider("", "dart")
-        ..add(code)
-        ..close();
-      return code.length;
-    } else {
-      output = new MainOutputGenerator();
-      return output.generateCode(libraryInfo, elementInfo, collector, renamer,
-          mainFunction, outputUri, outputProvider, mirrorRenamer,
-          multiFile: multiFile,
-          forceStripTypes: forceStripTypes,
-          enableMinification: enableMinification);
-    }
-  }
-
-  static PlaceholderCollector collectPlaceholders(
-      DiagnosticReporter reporter,
-      MirrorRenamer mirrorRenamer,
-      FunctionElement mainFunction,
-      LibraryInfo libraryInfo,
-      ElementInfo elementInfo) {
-    // Create all necessary placeholders.
-    PlaceholderCollector collector = new PlaceholderCollector(
-        reporter,
-        mirrorRenamer,
-        libraryInfo.fixedDynamicNames,
-        elementInfo.elementAsts,
-        mainFunction);
-
-    makePlaceholders(element) {
-      collector.collect(element);
-
-      if (element.isClass && !element.isEnumClass) {
-        elementInfo.classMembers[element].forEach(makePlaceholders);
-      }
-    }
-    elementInfo.topLevelElements.forEach(makePlaceholders);
-    return collector;
-  }
-
-  static PlaceholderRenamer createRenamer(PlaceholderCollector collector,
-      LibraryInfo libraryInfo, ElementInfo elementInfo,
-      {bool enableMinification: false,
-      bool forceStripTypes: false,
-      isSafeToRemoveTypeDeclarations}) {
-    // Create renames.
-    bool shouldCutDeclarationTypes = forceStripTypes ||
-        (enableMinification &&
-            isSafeToRemoveTypeDeclarations(elementInfo.classMembers));
-
-    PlaceholderRenamer placeholderRenamer = new PlaceholderRenamer(
-        libraryInfo.fixedDynamicNames,
-        libraryInfo.fixedStaticNames,
-        libraryInfo.reexportingLibraries,
-        cutDeclarationTypes: shouldCutDeclarationTypes,
-        enableMinification: enableMinification);
-
-    placeholderRenamer.computeRenames(collector);
-    return placeholderRenamer;
-  }
-
-  static String astOutput(
-      DiagnosticReporter reporter, ElementInfo elementInfo) {
-    // TODO(antonm): Ideally XML should be a separate backend.
-    // TODO(antonm): obey renames and minification, at least as an option.
-    StringBuffer sb = new StringBuffer();
-    outputElement(element) {
-      sb.write(element.parseNode(reporter).toDebugString());
-    }
-
-    // Emit XML for AST instead of the program.
-    for (Element topLevel in elementInfo.topLevelElements) {
-      if (topLevel.isClass &&
-          !elementInfo.emitNoMembersFor.contains(topLevel)) {
-        // TODO(antonm): add some class info.
-        elementInfo.classMembers[topLevel].forEach(outputElement);
-      } else {
-        outputElement(topLevel);
-      }
-    }
-    return '<Program>\n$sb</Program>\n';
-  }
-}
-
-class LibraryInfo {
-  final Set<String> fixedStaticNames;
-  final Set<String> fixedDynamicNames;
-  final Map<Element, LibraryElement> reexportingLibraries;
-  final List<LibraryElement> userLibraries;
-
-  LibraryInfo(this.fixedStaticNames, this.fixedDynamicNames,
-      this.reexportingLibraries, this.userLibraries);
-
-  static LibraryInfo processLibraries(
-      DiagnosticReporter reporter,
-      Iterable<LibraryElement> libraries,
-      Iterable<AstElement> resolvedElements) {
-    Set<String> fixedStaticNames = new Set<String>();
-    Set<String> fixedDynamicNames = new Set<String>();
-    Map<Element, LibraryElement> reexportingLibraries =
-        <Element, LibraryElement>{};
-    List<LibraryElement> userLibraries = <LibraryElement>[];
-    // Conservatively traverse all platform libraries and collect member names.
-    // TODO(antonm): ideally we should only collect names of used members,
-    // however as of today there are problems with names of some core library
-    // interfaces, most probably for interfaces of literals.
-
-    for (LibraryElement library in libraries) {
-      if (!library.isPlatformLibrary) {
-        userLibraries.add(library);
-        continue;
-      }
-      library.forEachLocalMember((Element element) {
-        if (element.isClass) {
-          ClassElement classElement = element;
-          assert(invariant(classElement, classElement.isResolved,
-              message: "Unresolved platform class."));
-          classElement.forEachLocalMember((member) {
-            if (member.isInstanceMember) {
-              fixedDynamicNames.add(member.name);
-            } else {
-              fixedStaticNames.add(member.name);
-            }
-          });
-        }
-        // Even class names are added due to a delicate problem we have:
-        // if one imports dart:core with a prefix, we cannot tell prefix.name
-        // from dynamic invocation (alas!).  So we'd better err on preserving
-        // those names.
-        fixedStaticNames.add(element.name);
-      });
-
-      library.forEachExport((Element export) {
-        if (!library.isInternalLibrary && export.library.isInternalLibrary) {
-          // If an element of an internal library is reexported by a platform
-          // library, we have to import the reexporting library instead of the
-          // internal library, because the internal library is an
-          // implementation detail of dart2js.
-          reexportingLibraries[export] = library;
-        }
-      });
-    }
-
-    // Map to keep track of names of enum classes. Since these cannot be renamed
-    // we ensure that they are unique.
-    Map<String, ClassElement> enumClassMap = <String, ClassElement>{};
-
-    // As of now names of named optionals are not renamed. Therefore add all
-    // field names used as named optionals into [fixedMemberNames].
-    for (final element in resolvedElements) {
-      if (!element.isConstructor) continue;
-      for (ParameterElement parameter in element.parameters) {
-        if (parameter.isInitializingFormal && parameter.isNamed) {
-          fixedDynamicNames.add(parameter.name);
-        }
-      }
-      ClassElement cls = element.enclosingClass;
-      if (cls != null && cls.isEnumClass) {
-        fixedDynamicNames.add('index');
-
-        ClassElement existingEnumClass =
-            enumClassMap.putIfAbsent(cls.name, () => cls);
-        if (existingEnumClass != cls) {
-          reporter.reportError(
-              reporter.createMessage(cls, MessageKind.GENERIC, {
-                'text': "Duplicate enum names are not supported "
-                    "in dart2dart."
-              }),
-              <DiagnosticMessage>[
-                reporter.createMessage(existingEnumClass, MessageKind.GENERIC, {
-                  'text': "This is the other declaration of '${cls.name}'."
-                }),
-              ]);
-        }
-      }
-    }
-
-    fixedStaticNames.addAll(enumClassMap.keys);
-
-    // The VM will automatically invoke the call method of objects
-    // that are invoked as functions. Make sure to not rename that.
-    fixedDynamicNames.add('call');
-
-    return new LibraryInfo(fixedStaticNames, fixedDynamicNames,
-        reexportingLibraries, userLibraries);
-  }
-}
-
-class ElementInfo {
-  final Map<Element, ElementAst> elementAsts;
-  final Iterable<Element> topLevelElements;
-  final Map<ClassElement, Iterable<Element>> classMembers;
-  final Iterable<ClassElement> emitNoMembersFor;
-
-  ElementInfo(this.elementAsts, this.topLevelElements, this.classMembers,
-      this.emitNoMembersFor);
-}
-
-class ElementInfoProcessor implements ElementInfo {
-  final Map<Element, ElementAst> elementAsts = new Map<Element, ElementAst>();
-  final Set<Element> topLevelElements = new Set<Element>();
-  final Map<ClassElement, Set<Element>> classMembers =
-      new Map<ClassElement, Set<Element>>();
-  final Set<ClassElement> emitNoMembersFor = new Set<ClassElement>();
-  final ElementPostProcessFunction postProcessElementAst;
-  final ComputeElementAstFunction parseElementAst;
-  final ElementFilter shouldOutput;
-
-  ElementInfoProcessor(
-      {this.postProcessElementAst, this.parseElementAst, this.shouldOutput});
-
-  static ElementInfo createElementInfo(
-      Iterable<ClassElement> instantiatedClasses,
-      Iterable<AstElement> resolvedElements,
-      Iterable<ClassElement> usedTypeLiterals,
-      {ElementPostProcessFunction postProcessElementAst,
-      ComputeElementAstFunction parseElementAst,
-      ElementFilter shouldOutput,
-      ElementSorter sortElements}) {
-    ElementInfoProcessor processor = new ElementInfoProcessor(
-        postProcessElementAst: postProcessElementAst,
-        parseElementAst: parseElementAst,
-        shouldOutput: shouldOutput);
-    return processor.process(
-        instantiatedClasses, resolvedElements, usedTypeLiterals,
-        sortElements: sortElements);
-  }
-
-  ElementInfo process(
-      Iterable<ClassElement> instantiatedClasses,
-      Iterable<AstElement> resolvedElements,
-      Iterable<ClassElement> usedTypeLiterals,
-      {ElementSorter sortElements}) {
-    // Build all top level elements to emit and necessary class members.
-    instantiatedClasses.where(shouldOutput).forEach(addClass);
-    resolvedElements.where(shouldOutput).forEach(addMember);
-    usedTypeLiterals.forEach((ClassElement element) {
-      if (shouldOutput(element)) {
-        if (!topLevelElements.contains(element)) {
-          // The class is only referenced by type literals.
-          emitNoMembersFor.add(element);
-        }
-        addClass(element);
-      }
-    });
-
-    // Sort elements.
-    List<Element> sortedTopLevels = sortElements(topLevelElements);
-    Map<ClassElement, List<Element>> sortedClassMembers =
-        new Map<ClassElement, List<Element>>();
-    classMembers.forEach((classElement, members) {
-      sortedClassMembers[classElement] = sortElements(members);
-    });
-
-    return new ElementInfo(
-        elementAsts, sortedTopLevels, sortedClassMembers, emitNoMembersFor);
-  }
-
-  void processElement(Element element, ElementAst elementAst) {
-    if (postProcessElementAst != null) {
-      postProcessElementAst(element, elementAst, newTypedefElementCallback,
-          newClassElementCallback);
-    }
-    elementAsts[element] = elementAst;
-  }
-
-  void addTopLevel(AstElement element, ElementAst elementAst) {
-    if (topLevelElements.contains(element)) return;
-    topLevelElements.add(element);
-    processElement(element, elementAst);
-  }
-
-  void addClass(ClassElement classElement) {
-    TreeElements treeElements = new TreeElementMapping(classElement);
-    backend2frontend.TreePrinter treePrinter =
-        new backend2frontend.TreePrinter(treeElements);
-    Node node = treePrinter.makeNodeForClassElement(classElement);
-    addTopLevel(classElement, new ElementAst(node, treeElements));
-    classMembers.putIfAbsent(classElement, () => new Set());
-  }
-
-  void newTypedefElementCallback(TypedefElement element) {
-    if (!shouldOutput(element)) return;
-    TreeElements treeElements = new TreeElementMapping(element);
-    backend2frontend.TreePrinter treePrinter =
-        new backend2frontend.TreePrinter(treeElements);
-    Node node = treePrinter.makeTypedef(element);
-    addTopLevel(element, new ElementAst(node, treeElements));
-  }
-
-  void newClassElementCallback(ClassElement classElement) {
-    if (!shouldOutput(classElement)) return;
-    addClass(classElement);
-  }
-
-  void addMember(element) {
-    if (element.isClassMember) {
-      ClassElement enclosingClass = element.enclosingClass;
-      assert(enclosingClass.isClass);
-      assert(shouldOutput(enclosingClass));
-      addClass(enclosingClass);
-      classMembers[enclosingClass].add(element);
-      if (enclosingClass.isEnumClass) return;
-      processElement(element, parseElementAst(element));
-    } else {
-      if (element.isTopLevel) {
-        addTopLevel(element, parseElementAst(element));
-      }
-    }
-  }
-}
-
-/// Main output generator for [DartOutputter] that emits dart code through a
-/// [CompilerOutputProvider].
-class MainOutputGenerator {
-  final Map<ClassNode, List<Node>> memberNodes =
-      new Map<ClassNode, List<Node>>();
-  final List<Node> topLevelNodes = <Node>[];
-
-  /// Generates the code and returns the total size.
-  int generateCode(
-      LibraryInfo libraryInfo,
-      ElementInfo elementInfo,
-      PlaceholderCollector collector,
-      PlaceholderRenamer placeholderRenamer,
-      FunctionElement mainFunction,
-      Uri outputUri,
-      CompilerOutputProvider outputProvider,
-      MirrorRenamer mirrorRenamer,
-      {bool multiFile: false,
-      bool forceStripTypes: false,
-      bool enableMinification: false}) {
-    for (Element element in elementInfo.topLevelElements) {
-      topLevelNodes.add(elementInfo.elementAsts[element].ast);
-      if (element.isClass) {
-        ClassElement cls = element;
-        if (cls.isMixinApplication || cls.isEnumClass) {
-          continue;
-        }
-        final members = <Node>[];
-        for (Element member in elementInfo.classMembers[cls]) {
-          members.add(elementInfo.elementAsts[member].ast);
-        }
-        memberNodes[elementInfo.elementAsts[cls].ast] = members;
-      }
-    }
-
-    mirrorRenamer.addRenames(
-        placeholderRenamer.renames, topLevelNodes, collector);
-
-    Map<LibraryElement, String> outputPaths = new Map<LibraryElement, String>();
-    Map<LibraryElement, EmitterUnparser> unparsers =
-        new Map<LibraryElement, EmitterUnparser>();
-
-    // The single unparser used if we collect all the output in one file.
-    EmitterUnparser mainUnparser = multiFile
-        ? null
-        : new EmitterUnparser(placeholderRenamer.renames,
-            stripTypes: forceStripTypes, minify: enableMinification);
-
-    if (multiFile) {
-      // TODO(sigurdm): Factor handling of library-paths out from emitting.
-      String mainName = outputUri.pathSegments.last;
-      String mainBaseName = mainName.endsWith(".dart")
-          ? mainName.substring(0, mainName.length - 5)
-          : mainName;
-      // Map each library to a path based on the uri of the original
-      // library and [compiler.options.outputUri].
-      Set<String> usedLibraryPaths = new Set<String>();
-      for (LibraryElement library in libraryInfo.userLibraries) {
-        if (library == mainFunction.library) {
-          outputPaths[library] = mainBaseName;
-        } else {
-          List<String> names =
-              library.canonicalUri.pathSegments.last.split(".");
-          if (names.last == "dart") {
-            names = names.sublist(0, names.length - 1);
-          }
-          outputPaths[library] =
-              "$mainBaseName.${makeUnique(names.join("."), usedLibraryPaths)}";
-        }
-      }
-
-      /// Rewrites imports/exports to refer to the paths given in [outputPaths].
-      for (LibraryElement outputLibrary in libraryInfo.userLibraries) {
-        EmitterUnparser unparser = new EmitterUnparser(
-            placeholderRenamer.renames,
-            stripTypes: forceStripTypes,
-            minify: enableMinification);
-        unparsers[outputLibrary] = unparser;
-        if (outputLibrary.hasLibraryName) {
-          unparser.unparseLibraryName(outputLibrary.libraryName);
-        }
-        for (ImportElement import in outputLibrary.imports) {
-          LibraryElement libraryElement = import.importedLibrary;
-          String uri = outputPaths.containsKey(libraryElement)
-              ? "${outputPaths[libraryElement]}.dart"
-              : libraryElement.canonicalUri.toString();
-          unparser.unparseImportTag(uri);
-        }
-        for (ExportElement export in outputLibrary.exports) {
-          LibraryElement libraryElement = export.exportedLibrary;
-          String uri = outputPaths.containsKey(libraryElement)
-              ? "${outputPaths[libraryElement]}.dart"
-              : libraryElement.canonicalUri.toString();
-          unparser.unparseExportTag(uri);
-        }
-      }
-    } else {
-      placeholderRenamer.platformImports
-          .forEach((LibraryElement library, String prefix) {
-        assert(library.isPlatformLibrary && !library.isInternalLibrary);
-        mainUnparser.unparseImportTag(library.canonicalUri.toString());
-        if (prefix != null) {
-          // Adding a prefixed import because (some) top-level access need
-          // it to avoid shadowing.
-          // TODO(johnniwinther): Avoid prefix-less import if not needed.
-          mainUnparser.unparseImportTag(library.canonicalUri.toString(),
-              prefix: prefix);
-        }
-      });
-    }
-
-    for (int i = 0; i < elementInfo.topLevelElements.length; i++) {
-      Element element = elementInfo.topLevelElements.elementAt(i);
-      Node node = topLevelNodes[i];
-      Unparser unparser = multiFile ? unparsers[element.library] : mainUnparser;
-      if (node is ClassNode) {
-        // TODO(smok): Filter out default constructors here.
-        unparser.unparseClassWithBody(node, memberNodes[node]);
-      } else {
-        unparser.unparse(node);
-      }
-      unparser.newline();
-    }
-
-    int totalSize = 0;
-    if (multiFile) {
-      for (LibraryElement outputLibrary in libraryInfo.userLibraries) {
-        // TODO(sigurdm): Make the unparser output directly into the buffer
-        // instead of caching in `.result`.
-        String code = unparsers[outputLibrary].result;
-        totalSize += code.length;
-        outputProvider(outputPaths[outputLibrary], "dart")
-          ..add(code)
-          ..close();
-      }
-    } else {
-      String code = mainUnparser.result;
-      outputProvider("", "dart")
-        ..add(code)
-        ..close();
-
-      totalSize = code.length;
-    }
-
-    return totalSize;
-  }
-}

pkg/compiler/lib/src/dart_backend/placeholder_collector.dart

diff --git a/pkg/compiler/lib/src/dart_backend/placeholder_collector.dart b/pkg/compiler/lib/src/dart_backend/placeholder_collector.dart
deleted file mode 100644
index 65f2576..0000000
--- a/pkg/compiler/lib/src/dart_backend/placeholder_collector.dart
+++ /dev/null
@@ -1,724 +0,0 @@
-// Copyright (c) 2012, the Dart project authors.  Please see the AUTHORS file
-// for details. All rights reserved. Use of this source code is governed by a
-// BSD-style license that can be found in the LICENSE file.
-
-part of dart_backend;
-
-class LocalPlaceholder {
-  final String identifier;
-  final Set<Node> nodes;
-  LocalPlaceholder(this.identifier) : nodes = new Set<Node>();
-  int get hashCode => identifier.hashCode;
-  String toString() => 'local_placeholder[id($identifier), nodes($nodes)]';
-}
-
-class FunctionScope {
-  final Set<String> parameterIdentifiers;
-  final Set<LocalPlaceholder> localPlaceholders;
-  FunctionScope()
-      : parameterIdentifiers = new Set<String>(),
-        localPlaceholders = new Set<LocalPlaceholder>();
-  void registerParameter(Identifier node) {
-    parameterIdentifiers.add(node.source);
-  }
-}
-
-class ConstructorPlaceholder {
-  final Identifier node;
-  final ConstructorElement element;
-
-  ConstructorPlaceholder(this.node, this.element);
-}
-
-class DeclarationTypePlaceholder {
-  final TypeAnnotation typeNode;
-  final bool requiresVar;
-  DeclarationTypePlaceholder(this.typeNode, this.requiresVar);
-}
-
-class SendVisitor extends Visitor {
-  final TreeElements elements;
-  final PlaceholderCollector collector;
-
-  SendVisitor(this.collector, this.elements);
-
-  visitSend(Send node) {
-    Element element = elements[node];
-    if (elements.isTypeLiteral(node)) {
-      DartType type = elements.getTypeLiteralType(node);
-      if (!type.isDynamic) {
-        if (type is TypeVariableType) {
-          collector.makeTypeVariablePlaceholder(node.selector, type);
-        } else {
-          collector.makeTypePlaceholder(node.selector, type);
-        }
-      }
-    } else if (node.isSuperCall) {
-      if (element != null && element.isConstructor) {
-        collector.tryMakeConstructorPlaceholder(node, element);
-      } else {
-        collector.tryMakeMemberPlaceholder(node.selector);
-      }
-    } else if (node.isOperator) {
-      return;
-    } else if (node.isPropertyAccess) {
-      if (!Elements.isUnresolved(element) && element.impliesType) {
-        collector.makeElementPlaceholder(node, element);
-      } else {
-        visitGetterSend(node);
-      }
-    } else if (element != null && Initializers.isConstructorRedirect(node)) {
-      visitStaticSend(node);
-    } else if (Elements.isClosureSend(node, element)) {
-      if (element != null) {
-        collector.tryMakeLocalPlaceholder(element, node.selector);
-      }
-    } else {
-      if (Elements.isUnresolved(element)) {
-        if (element == null) {
-          // Example: f() with 'f' unbound.
-          // This can only happen inside an instance method.
-          visitDynamicSend(node);
-        } else {
-          visitStaticSend(node);
-        }
-      } else if (element.isInstanceMember) {
-        // Example: f() with 'f' bound to instance method.
-        visitDynamicSend(node);
-      } else if (!element.isInstanceMember) {
-        // Example: A.f() or f() with 'f' bound to a static function.
-        // Also includes new A() or new A.named() which is treated like a
-        // static call to a factory.
-        visitStaticSend(node);
-      }
-    }
-  }
-
-  visitDynamicSend(Send node) {
-    final element = elements[node];
-    if (element == null || !element.isMalformed) {
-      collector.tryMakeMemberPlaceholder(node.selector);
-    }
-  }
-
-  visitGetterSend(Send node) {
-    final element = elements[node];
-    // element == null means dynamic property access.
-    if (element == null) {
-      collector.tryMakeMemberPlaceholder(node.selector);
-    } else if (element.isMalformed) {
-      collector.makeUnresolvedPlaceholder(node);
-      return;
-    } else if (element.isPrefix) {
-      // Node is prefix part in case of source 'lib.somesetter = 5;'
-      collector.makeErasePrefixPlaceholder(node);
-    } else if (Elements.isStaticOrTopLevel(element)) {
-      // Unqualified or prefixed top level or static.
-      collector.makeElementPlaceholder(node.selector, element);
-    } else if (!element.isTopLevel) {
-      if (element.isInstanceMember) {
-        collector.tryMakeMemberPlaceholder(node.selector);
-      } else {
-        // May get FunctionExpression here in selector
-        // in case of A(int this.f());
-        if (node.selector is Identifier) {
-          collector.tryMakeLocalPlaceholder(element, node.selector);
-        } else {
-          assert(node.selector is FunctionExpression);
-        }
-      }
-    }
-  }
-
-  visitStaticSend(Send node) {
-    Element element = elements[node];
-    collector.mirrorRenamer
-        .registerStaticSend(collector.currentElement, element, node);
-
-    if (Elements.isUnresolved(element) || element.isDeferredLoaderGetter) {
-      return;
-    }
-    if (element.isConstructor || element.isFactoryConstructor) {
-      // Rename named constructor in redirection position:
-      // class C { C.named(); C.redirecting() : this.named(); }
-      if (node.receiver is Identifier &&
-          node.receiver.asIdentifier().isThis()) {
-        assert(node.selector is Identifier);
-        collector.tryMakeConstructorPlaceholder(node, element);
-      }
-      return;
-    }
-    collector.makeElementPlaceholder(node.selector, element);
-    // Another ugly case: <lib prefix>.<top level> is represented as
-    // receiver: lib prefix, selector: top level.
-    if (element.isTopLevel && node.receiver != null) {
-      assert(elements[node.receiver].isPrefix);
-      // Hack: putting null into map overrides receiver of original node.
-      collector.makeErasePrefixPlaceholder(node.receiver);
-    }
-  }
-
-  internalError(Spannable node, String reason) {
-    collector.internalError(reason, node: node);
-  }
-
-  visitNode(Node node) {
-    internalError(node, "Unhandled node");
-  }
-}
-
-class PlaceholderCollector extends Visitor {
-  final DiagnosticReporter reporter;
-  final MirrorRenamer mirrorRenamer;
-  final FunctionElement mainFunction;
-  final Set<String> fixedMemberNames; // member names which cannot be renamed.
-  final Map<Element, ElementAst> elementAsts;
-  final Set<Node> prefixNodesToErase = new Set<Node>();
-  final Set<Node> unresolvedNodes = new Set<Node>();
-  final Map<Element, Set<Node>> elementNodes = new Map<Element, Set<Node>>();
-  final Map<FunctionElement, FunctionScope> functionScopes =
-      new Map<FunctionElement, FunctionScope>();
-  final Map<LibraryElement, Set<Identifier>> privateNodes =
-      new Map<LibraryElement, Set<Identifier>>();
-  final List<DeclarationTypePlaceholder> declarationTypePlaceholders =
-      new List<DeclarationTypePlaceholder>();
-  final Map<String, Set<Identifier>> memberPlaceholders =
-      new Map<String, Set<Identifier>>();
-  final List<ConstructorPlaceholder> constructorPlaceholders =
-      new List<ConstructorPlaceholder>();
-  Map<String, LocalPlaceholder> currentLocalPlaceholders;
-  Element currentElement;
-  FunctionElement topmostEnclosingFunction;
-  TreeElements treeElements;
-
-  get currentFunctionScope => functionScopes.putIfAbsent(
-      topmostEnclosingFunction, () => new FunctionScope());
-
-  PlaceholderCollector(this.reporter, this.mirrorRenamer, this.fixedMemberNames,
-      this.elementAsts, this.mainFunction);
-
-  void collectFunctionDeclarationPlaceholders(
-      FunctionElement element, FunctionExpression node) {
-    if (element.isConstructor) {
-      ConstructorElement constructor = element;
-      tryMakeConstructorPlaceholder(node.name, element);
-      RedirectingFactoryBody bodyAsRedirectingFactoryBody =
-          node.body.asRedirectingFactoryBody();
-      if (bodyAsRedirectingFactoryBody != null) {
-        // Factory redirection.
-        FunctionElement redirectTarget = constructor.immediateRedirectionTarget;
-        assert(redirectTarget != null && redirectTarget != element);
-        tryMakeConstructorPlaceholder(
-            bodyAsRedirectingFactoryBody.constructorReference, redirectTarget);
-      }
-    } else if (Elements.isStaticOrTopLevel(element)) {
-      // Note: this code should only rename private identifiers for class'
-      // fields/getters/setters/methods.  Top-level identifiers are renamed
-      // just to escape conflicts and that should be enough as we shouldn't
-      // be able to resolve private identifiers for other libraries.
-      makeElementPlaceholder(node.name, element);
-    } else if (element.isClassMember) {
-      if (node.name is Identifier) {
-        tryMakeMemberPlaceholder(node.name);
-      } else {
-        assert(node.name.asSend().isOperator);
-      }
-    }
-  }
-
-  void collectFieldDeclarationPlaceholders(Element element, Node node) {
-    Identifier name = node is Identifier ? node : node.asSend().selector;
-    if (Elements.isStaticOrTopLevel(element)) {
-      makeElementPlaceholder(name, element);
-    } else if (Elements.isInstanceField(element)) {
-      tryMakeMemberPlaceholder(name);
-    }
-  }
-
-  void collect(Element element) {
-    this.currentElement = element;
-    this.topmostEnclosingFunction = null;
-    final ElementAst elementAst = elementAsts[element];
-    this.treeElements = elementAst.treeElements;
-    Node elementNode = elementAst.ast;
-    if (element is FunctionElement) {
-      collectFunctionDeclarationPlaceholders(element, elementNode);
-    } else if (element is VariableElement) {
-      VariableDefinitions definitions = elementNode;
-      Node definition = definitions.definitions.nodes.head;
-      collectFieldDeclarationPlaceholders(element, definition);
-      makeVarDeclarationTypePlaceholder(definitions);
-    } else {
-      assert(element is ClassElement || element is TypedefElement);
-    }
-    currentLocalPlaceholders = new Map<String, LocalPlaceholder>();
-    if (!(element is ConstructorElement && element.isRedirectingFactory)) {
-      // Do not visit the body of redirecting factories.
-      reporter.withCurrentElement(element, () {
-        elementNode.accept(this);
-      });
-    }
-  }
-
-  // TODO(karlklose): should we create placeholders for these?
-  bool isTypedefParameter(Element element) {
-    return element != null &&
-        element.enclosingElement != null &&
-        element.enclosingElement.isTypedef;
-  }
-
-  void tryMakeLocalPlaceholder(Element element, Identifier node) {
-    bool isNamedOptionalParameter() {
-      FunctionTypedElement function = element.enclosingElement;
-      FunctionSignature signature = function.functionSignature;
-      if (!signature.optionalParametersAreNamed) return false;
-      for (Element parameter in signature.optionalParameters) {
-        if (identical(parameter, element)) return true;
-      }
-      return false;
-    }
-    if (element.isRegularParameter &&
-        !isTypedefParameter(element) &&
-        isNamedOptionalParameter()) {
-      currentFunctionScope.registerParameter(node);
-    } else if (Elements.isLocal(element) && !isTypedefParameter(element)) {
-      makeLocalPlaceholder(node);
-    }
-  }
-
-  void tryMakeMemberPlaceholder(Identifier node) {
-    assert(node != null);
-    if (node is Operator) return;
-    String identifier = node.source;
-    if (fixedMemberNames.contains(identifier)) return;
-    memberPlaceholders
-        .putIfAbsent(identifier, () => new Set<Identifier>())
-        .add(node);
-  }
-
-  void makeTypePlaceholder(Node node, DartType type) {
-    Send send = node.asSend();
-    if (send != null) {
-      // Prefix.
-      assert(send.receiver is Identifier);
-      assert(send.selector is Identifier);
-      makeErasePrefixPlaceholder(send.receiver);
-      node = send.selector;
-    }
-    makeElementPlaceholder(node, type.element);
-  }
-
-  void makeTypeVariablePlaceholder(Node node, TypeVariableType type) {
-    Send send = node.asSend();
-    if (send != null) {
-      // Prefix.
-      assert(send.receiver is Identifier);
-      assert(send.selector is Identifier);
-      makeErasePrefixPlaceholder(send.receiver);
-      node = send.selector;
-    }
-    tryMakeMemberPlaceholder(node);
-  }
-
-  void makeOmitDeclarationTypePlaceholder(TypeAnnotation type) {
-    if (type == null) return;
-    declarationTypePlaceholders
-        .add(new DeclarationTypePlaceholder(type, false));
-  }
-
-  void makeVarDeclarationTypePlaceholder(VariableDefinitions node) {
-    // TODO(smok): Maybe instead of calling this method and
-    // makeDeclaratioTypePlaceholder have type declaration placeholder
-    // collector logic in visitVariableDefinitions when resolver becomes better
-    // and/or catch syntax changes.
-    if (node.type == null) return;
-    bool requiresVar = !node.modifiers.isFinalOrConst;
-    declarationTypePlaceholders
-        .add(new DeclarationTypePlaceholder(node.type, requiresVar));
-  }
-
-  /// Marks [node] to be erased in the output.
-  /// This is done for library prefixes because they are not used in the output
-  /// because all imports are flattened and conflicts are renamed away.
-  void makeErasePrefixPlaceholder(Node node) {
-    assert(node is Identifier || node is Send);
-    prefixNodesToErase.add(node);
-  }
-
-  void makeElementPlaceholder(Node node, Element element) {
-    assert(node != null);
-    assert(element != null);
-    LibraryElement library = element.library;
-    if (identical(element, mainFunction)) return;
-    if (library.isDartCore) return;
-
-    if (library.isPlatformLibrary && !element.isTopLevel) {
-      return;
-    }
-
-    ClassElement cls = element.enclosingClass;
-    if (cls != null && cls.isEnumClass) {
-      // Enums and enum values cannot be changed, since the semantics of
-      // `toString` is defined by the names of the declarations.
-      return;
-    }
-
-    if (element.isAccessor) {
-      element = (element as AccessorElement).abstractField;
-    }
-    elementNodes.putIfAbsent(element, () => new Set<Node>()).add(node);
-  }
-
-  /// Marks [node] to be renamed per-library if it names an instance member
-  /// and has a private name.
-  void tryMakePrivateIdentifier(Node node, Element element) {
-    if (node is Identifier &&
-        !Elements.isStaticOrTopLevel(element) &&
-        !Elements.isLocal(element) &&
-        Name.isPrivateName(node.source)) {
-      privateNodes
-          .putIfAbsent(currentElement.library, () => new Set<Identifier>())
-          .add(node);
-    }
-  }
-
-  void makeUnresolvedPlaceholder(Node node) {
-    unresolvedNodes.add(node);
-  }
-
-  void makeLocalPlaceholder(Identifier identifier) {
-    LocalPlaceholder getLocalPlaceholder() {
-      String name = identifier.source;
-      return currentLocalPlaceholders.putIfAbsent(name, () {
-        LocalPlaceholder localPlaceholder = new LocalPlaceholder(name);
-        currentFunctionScope.localPlaceholders.add(localPlaceholder);
-        return localPlaceholder;
-      });
-    }
-    getLocalPlaceholder().nodes.add(identifier);
-  }
-
-  /// Finds the first constructor on the chain of definingConstructor from
-  /// [element] that is not in a synthetic class.
-  Element findDefiningConstructor(ConstructorElement element) {
-    while (element.definingConstructor != null) {
-      element = element.definingConstructor;
-    }
-    return element;
-  }
-
-  void tryMakeConstructorPlaceholder(Node node, ConstructorElement element) {
-    if (Elements.isUnresolved(element)) {
-      makeUnresolvedPlaceholder(node);
-      return;
-    }
-    // A library prefix.
-    Node prefix;
-    // The name of the class with the constructor.
-    Node className;
-    // Will be null for unnamed constructors.
-    Identifier constructorName;
-    // First deconstruct the constructor, there are 4 possibilities:
-    //  ClassName()
-    //  prefix.ClassName()
-    //  ClassName.constructorName()
-    //  prefix.ClassName.constructorName()
-    if (node is Send) {
-      if (node.receiver is Send) {
-        Send receiver = node.receiver;
-        // prefix.ClassName.constructorName()
-        assert(treeElements[receiver.receiver] != null &&
-            treeElements[receiver.receiver].isPrefix);
-        prefix = receiver.receiver;
-        className = receiver.selector;
-        constructorName = node.selector;
-      } else {
-        Element receiverElement = treeElements[node.receiver];
-        if (receiverElement != null && receiverElement.isPrefix) {
-          // prefix.ClassName()
-          prefix = node.receiver;
-          className = node.selector;
-        } else {
-          // ClassName.constructorName()
-          className = node.receiver;
-          constructorName = node.selector;
-        }
-      }
-    } else {
-      // ClassName()
-      className = node;
-    }
-
-    if (prefix != null) {
-      makeErasePrefixPlaceholder(prefix);
-    }
-
-    if (className is TypeAnnotation) {
-      visitTypeAnnotation(className);
-    } else if (Elements.isUnresolved(element)) {
-      // We handle unresolved nodes elsewhere.
-    } else if (className.isThis() || className.isSuper()) {
-      // Do not rename super and this.
-    } else if (className is Identifier) {
-      makeElementPlaceholder(className, element.contextClass);
-    } else {
-      throw "Bad type of constructor name $className";
-    }
-
-    if (constructorName != null) {
-      Element definingConstructor = findDefiningConstructor(element);
-      constructorPlaceholders.add(
-          new ConstructorPlaceholder(constructorName, definingConstructor));
-      tryMakePrivateIdentifier(constructorName, element);
-    }
-  }
-
-  void internalError(String reason, {Node node}) {
-    reporter.internalError(node, reason);
-  }
-
-  visit(Node node) => (node == null) ? null : node.accept(this);
-
-  visitNode(Node node) {
-    node.visitChildren(this);
-  } // We must go deeper.
-
-  visitNewExpression(NewExpression node) {
-    Send send = node.send;
-    DartType type = treeElements.getType(node);
-    assert(type != null);
-    Element constructor = treeElements[send];
-    assert(constructor != null);
-    assert(send.receiver == null);
-    if (!Elements.isMalformed(constructor)) {
-      tryMakeConstructorPlaceholder(node.send.selector, constructor);
-      // TODO(smok): Should this be in visitNamedArgument?
-      // Field names can be exposed as names of optional arguments, e.g.
-      // class C {
-      //   final field;
-      //   C([this.field]);
-      // }
-      // Do not forget to rename them as well.
-      FunctionElement constructorFunction = constructor;
-      List<Element> optionalParameters =
-          constructorFunction.functionSignature.optionalParameters;
-      for (final argument in send.argumentsNode) {
-        NamedArgument named = argument.asNamedArgument();
-        if (named == null) continue;
-        Identifier name = named.name;
-        String nameAsString = name.source;
-        for (final parameter in optionalParameters) {
-          if (parameter.isInitializingFormal) {
-            if (parameter.name == nameAsString) {
-              tryMakeMemberPlaceholder(name);
-              break;
-            }
-          }
-        }
-      }
-    } else {
-      makeUnresolvedPlaceholder(node.send.selector);
-    }
-    visit(node.send.argumentsNode);
-  }
-
-  visitSend(Send send) {
-    Element element = treeElements[send];
-    tryMakePrivateIdentifier(send.selector, element);
-    new SendVisitor(this, treeElements).visitSend(send);
-    send.visitChildren(this);
-  }
-
-  visitSendSet(SendSet send) {
-    Element element = treeElements[send];
-    if (Elements.isMalformed(element)) {
-      // Complicated case: constructs like receiver.selector++ can resolve
-      // to ErroneousElement.  Fortunately, receiver.selector still
-      // can be resoved via treeElements[send.selector], that's all
-      // that is needed to rename the construct properly.
-      element = treeElements[send.selector];
-    }
-    tryMakePrivateIdentifier(send.selector, element);
-    if (element == null) {
-      if (send.receiver != null) tryMakeMemberPlaceholder(send.selector);
-    } else if (!element.isMalformed) {
-      if (Elements.isStaticOrTopLevel(element)) {
-        // TODO(smok): Worth investigating why sometimes we get getter/setter
-        // here and sometimes abstract field.
-        assert(element.isClass ||
-            element is VariableElement ||
-            element.isAccessor ||
-            element.isAbstractField ||
-            element.isFunction ||
-            element.isTypedef ||
-            element is TypeVariableElement);
-        makeElementPlaceholder(send.selector, element);
-      } else {
-        Identifier identifier = send.selector.asIdentifier();
-        if (identifier == null) {
-          // Handle optional function expression parameters with default values.
-          identifier = send.selector.asFunctionExpression().name;
-        }
-        if (Elements.isInstanceField(element)) {
-          tryMakeMemberPlaceholder(identifier);
-        } else {
-          tryMakeLocalPlaceholder(element, identifier);
-        }
-      }
-    }
-    send.visitChildren(this);
-  }
-
-  visitTypeAnnotation(TypeAnnotation node) {
-    final type = treeElements.getType(node);
-    assert(invariant(node, type != null,
-        message: "Missing type for type annotation: $treeElements"));
-    if (!type.isVoid) {
-      if (!type.treatAsDynamic) {
-        if (type is TypeVariableType) {
-          makeTypeVariablePlaceholder(node.typeName, type);
-        } else {
-          makeTypePlaceholder(node.typeName, type);
-        }
-      } else if (!type.isDynamic) {
-        makeUnresolvedPlaceholder(node.typeName);
-      }
-    }
-    // Visit only type arguments, otherwise in case of lib.Class type
-    // annotation typeName is Send and we go to visitGetterSend, as a result
-    // "Class" is added to member placeholders.
-    visit(node.typeArguments);
-  }
-
-  visitVariableDefinitions(VariableDefinitions node) {
-    // Collect only local placeholders.
-    for (Node definition in node.definitions.nodes) {
-      Element definitionElement = treeElements[definition];
-      // definitionElement may be null if we're inside variable definitions
-      // of a function that is a parameter of another function.
-      // TODO(smok): Fix this when resolver correctly deals with
-      // such cases.
-      if (definitionElement == null) continue;
-
-      Send send = definition.asSend();
-      Identifier identifier = definition is Identifier
-          ? definition
-          : definition is Send
-              ? (send.selector is Identifier ? send.selector : null)
-              : null;
-
-      tryMakePrivateIdentifier(identifier, definitionElement);
-
-      if (send != null) {
-        // May get FunctionExpression here in definition.selector
-        // in case of A(int this.f());
-        if (send.selector is Identifier) {
-          if (definitionElement.isInitializingFormal) {
-            tryMakeMemberPlaceholder(send.selector);
-          } else {
-            tryMakeLocalPlaceholder(definitionElement, send.selector);
-          }
-        } else {
-          assert(send.selector is FunctionExpression);
-          if (definitionElement.isInitializingFormal) {
-            tryMakeMemberPlaceholder(send.selector.asFunctionExpression().name);
-          }
-        }
-      } else if (definition is Identifier) {
-        tryMakeLocalPlaceholder(definitionElement, definition);
-      } else if (definition is FunctionExpression) {
-        // Skip, it will be processed in visitFunctionExpression.
-      } else {
-        internalError('Unexpected definition structure $definition');
-      }
-    }
-    node.visitChildren(this);
-  }
-
-  visitFunctionExpression(FunctionExpression node) {
-    bool isKeyword(Identifier id) =>
-        id != null && Keyword.keywords[id.source] != null;
-
-    Element element = treeElements[node];
-    // May get null;
-    if (element != null) {
-      tryMakePrivateIdentifier(node.name, element);
-
-      // Rename only local functions.
-      if (topmostEnclosingFunction == null &&
-          element is! LocalParameterElement &&
-          element is! InitializingFormalElement) {
-        topmostEnclosingFunction = element;
-      }
-      if (!identical(element, currentElement)) {
-        if (node.name != null) {
-          assert(node.name is Identifier);
-          tryMakeLocalPlaceholder(element, node.name);
-        }
-      }
-    }
-
-    node.visitChildren(this);
-
-    // Make sure we don't omit return type of methods which names are
-    // identifiers, because the following works fine:
-    // int interface() => 1;
-    // But omitting 'int' makes VM unhappy.
-    // TODO(smok): Remove it when http://dartbug.com/5278 is fixed.
-    if (node.name == null || !isKeyword(node.name.asIdentifier())) {
-      makeOmitDeclarationTypePlaceholder(node.returnType);
-    }
-    collectFunctionParameters(node.parameters);
-  }
-
-  void collectFunctionParameters(NodeList parameters) {
-    if (parameters == null) return;
-    for (Node parameter in parameters.nodes) {
-      if (parameter is NodeList) {
-        // Optional parameter list.
-        collectFunctionParameters(parameter);
-      } else {
-        assert(parameter is VariableDefinitions);
-        makeOmitDeclarationTypePlaceholder(
-            parameter.asVariableDefinitions().type);
-      }
-    }
-  }
-
-  visitClassNode(ClassNode node) {
-    ClassElement classElement = currentElement;
-    makeElementPlaceholder(node.name, classElement);
-    node.visitChildren(this);
-  }
-
-  visitNamedMixinApplication(NamedMixinApplication node) {
-    ClassElement classElement = currentElement;
-    makeElementPlaceholder(node.name, classElement);
-    node.visitChildren(this);
-  }
-
-  visitTypeVariable(TypeVariable node) {
-    DartType type = treeElements.getType(node);
-    assert(invariant(node, type != null,
-        message: "Missing type for type variable: $treeElements"));
-    makeTypeVariablePlaceholder(node.name, type);
-    node.visitChildren(this);
-  }
-
-  visitTypedef(Typedef node) {
-    assert(currentElement is TypedefElement);
-    makeElementPlaceholder(node.name, currentElement);
-    node.visitChildren(this);
-    makeOmitDeclarationTypePlaceholder(node.returnType);
-    collectFunctionParameters(node.formals);
-  }
-
-  visitBlock(Block node) {
-    for (Node statement in node.statements.nodes) {
-      if (statement is VariableDefinitions) {
-        makeVarDeclarationTypePlaceholder(statement);
-      }
-    }
-    node.visitChildren(this);
-  }
-}

pkg/compiler/lib/src/dart_backend/renamer.dart

diff --git a/pkg/compiler/lib/src/dart_backend/renamer.dart b/pkg/compiler/lib/src/dart_backend/renamer.dart
deleted file mode 100644
index 29b26d7..0000000
--- a/pkg/compiler/lib/src/dart_backend/renamer.dart
+++ /dev/null
@@ -1,372 +0,0 @@
-// Copyright (c) 2012, the Dart project authors.  Please see the AUTHORS file
-// for details. All rights reserved. Use of this source code is governed by a
-// BSD-style license that can be found in the LICENSE file.
-
-part of dart_backend;
-
-Comparator get _compareNodes => compareBy((n) => n.getBeginToken().charOffset);
-
-abstract class Renamable implements Comparable {
-  final int RENAMABLE_TYPE_ELEMENT = 1;
-  final int RENAMABLE_TYPE_MEMBER = 2;
-  final int RENAMABLE_TYPE_LOCAL = 3;
-
-  final Set<Node> nodes;
-
-  Renamable(this.nodes);
-  int compareTo(Renamable other) {
-    int nodesDiff = other.nodes.length.compareTo(this.nodes.length);
-    if (nodesDiff != 0) return nodesDiff;
-    int typeDiff = this.kind.compareTo(other.kind);
-    return typeDiff != 0 ? typeDiff : compareInternals(other);
-  }
-
-  int compareInternals(Renamable other);
-  int get kind;
-
-  String createNewName(PlaceholderRenamer placeholderRenamer);
-}
-
-class GlobalRenamable extends Renamable {
-  final Entity entity;
-
-  GlobalRenamable(this.entity, Set<Node> nodes) : super(nodes);
-
-  int compareInternals(GlobalRenamable other) =>
-      compareElements(this.entity, other.entity);
-  int get kind => RENAMABLE_TYPE_ELEMENT;
-  String createNewName(PlaceholderRenamer placeholderRenamer) {
-    return placeholderRenamer._renameGlobal(entity);
-  }
-}
-
-class MemberRenamable extends Renamable {
-  final String identifier;
-  MemberRenamable(this.identifier, Set<Node> nodes) : super(nodes);
-  int compareInternals(MemberRenamable other) =>
-      this.identifier.compareTo(other.identifier);
-  int get kind => RENAMABLE_TYPE_MEMBER;
-  String createNewName(PlaceholderRenamer placeholderRenamer) {
-    return placeholderRenamer._generateMemberName(identifier);
-  }
-}
-
-class LocalRenamable extends Renamable {
-  LocalRenamable(Set<Node> nodes) : super(nodes);
-  int compareInternals(LocalRenamable other) => _compareNodes(
-      sorted(this.nodes, _compareNodes)[0],
-      sorted(other.nodes, _compareNodes)[0]);
-  int get kind => RENAMABLE_TYPE_LOCAL;
-  String createNewName(PlaceholderRenamer placeholderRenamer) {
-    return placeholderRenamer._generateUniqueTopLevelName("");
-  }
-}
-
-/**
- * Renames only top-level elements that would lead to ambiguity if not renamed.
- */
-class PlaceholderRenamer {
-  /// After running [computeRenames] this will contain the computed renames.
-  final Map<Node, String> renames = new Map<Node, String>();
-
-  /// After running [computeRenames] this will map the used platform
-  /// libraries to their respective prefixes.
-  final Map<LibraryElement, String> platformImports =
-      <LibraryElement, String>{};
-
-  final bool enableMinification;
-  final Set<String> fixedDynamicNames;
-  final Set<String> fixedStaticNames;
-  final Map<Element, LibraryElement> reexportingLibraries;
-  final bool cutDeclarationTypes;
-
-  final Map<Entity, String> _renamedCache = new Map<Entity, String>();
-  final Map<Entity, Map<String, String>> _privateCache =
-      new Map<Entity, Map<String, String>>();
-
-  // Identifiers that has already been used, or are reserved by the
-  // language/platform.
-  Set<String> _forbiddenIdentifiers;
-  Set<String> _allNamedParameterIdentifiers;
-
-  Generator _generator;
-
-  PlaceholderRenamer(
-      this.fixedDynamicNames, this.fixedStaticNames, this.reexportingLibraries,
-      {this.enableMinification, this.cutDeclarationTypes});
-
-  void _renameNodes(Iterable<Node> nodes, String renamer(Node node)) {
-    for (Node node in sorted(nodes, _compareNodes)) {
-      renames[node] = renamer(node);
-    }
-  }
-
-  String _generateUniqueTopLevelName(String originalName) {
-    String newName = _generator.generate(originalName, (name) {
-      return _forbiddenIdentifiers.contains(name) ||
-          _allNamedParameterIdentifiers.contains(name);
-    });
-    _forbiddenIdentifiers.add(newName);
-    return newName;
-  }
-
-  String _generateMemberName(String original) {
-    return _generator.generate(original, _forbiddenIdentifiers.contains);
-  }
-
-  /// Looks up [originalName] in the [_privateCache] cache of [library].
-  /// If [originalName] was not renamed before, generate a new name.
-  String _getPrivateName(LibraryElement library, String originalName) {
-    return _privateCache
-        .putIfAbsent(library, () => new Map<String, String>())
-        .putIfAbsent(
-            originalName, () => _generateUniqueTopLevelName(originalName));
-  }
-
-  String _renameConstructor(ConstructorPlaceholder placeholder) {
-    String name = placeholder.element.name;
-    if (name == '') return "";
-    String result = _renameGlobal(placeholder.element);
-    return result;
-  }
-
-  String _renameGlobal(Entity entity) {
-    assert(entity is! Element ||
-        Elements.isMalformed(entity) ||
-        Elements.isStaticOrTopLevel(entity) ||
-        entity is TypeVariableElement);
-    // TODO(smok): We may want to reuse class static field and method names.
-    if (entity is Element) {
-      LibraryElement library = entity.library;
-      if (reexportingLibraries.containsKey(entity)) {
-        library = reexportingLibraries[entity];
-      }
-      if (library.isPlatformLibrary) {
-        // TODO(johnniwinther): Handle prefixes for dart:core.
-        if (library.canonicalUri == Uris.dart_core) return entity.name;
-        if (library.isInternalLibrary) {
-          throw new SpannableAssertionFailure(
-              entity,
-              "Internal library $library should never have been imported from "
-              "the code compiled by dart2dart.");
-        }
-
-        String prefix = platformImports.putIfAbsent(library, () => null);
-        if (entity.isTopLevel && fixedDynamicNames.contains(entity.name)) {
-          if (prefix == null) {
-            prefix = _generateUniqueTopLevelName('');
-            platformImports[library] = prefix;
-          }
-          return '$prefix.${entity.name}';
-        }
-        return entity.name;
-      }
-    }
-    String name = _renamedCache.putIfAbsent(
-        entity, () => _generateUniqueTopLevelName(entity.name));
-    // Look up in [_renamedCache] for a name for [entity] .
-    // If it was not renamed before, generate a new name.
-    return name;
-  }
-
-  void _computeMinifiedRenames(PlaceholderCollector placeholderCollector) {
-    _generator = new MinifyingGenerator();
-
-    // Build a list sorted by usage of local nodes that will be renamed to
-    // the same identifier. So the top-used local variables in all functions
-    // will be renamed first and will all share the same new identifier.
-    int maxLength = placeholderCollector.functionScopes.values
-        .fold(0, (a, b) => max(a, b.localPlaceholders.length));
-
-    List<Set<Node>> allLocals =
-        new List<Set<Node>>.generate(maxLength, (_) => new Set<Node>());
-
-    for (FunctionScope functionScope
-        in placeholderCollector.functionScopes.values) {
-      // Add current sorted local identifiers to the whole sorted list
-      // of all local identifiers for all functions.
-      List<LocalPlaceholder> currentSortedPlaceholders = sorted(
-          functionScope.localPlaceholders,
-          compareBy((LocalPlaceholder ph) => -ph.nodes.length));
-
-      List<Set<Node>> currentSortedNodes = currentSortedPlaceholders
-          .map((LocalPlaceholder ph) => ph.nodes)
-          .toList();
-
-      for (int i = 0; i < currentSortedNodes.length; i++) {
-        allLocals[i].addAll(currentSortedNodes[i]);
-      }
-    }
-
-    // Rename elements, members and locals together based on their usage
-    // count, otherwise when we rename elements first there will be no good
-    // identifiers left for members even if they are used often.
-    List<Renamable> renamables = new List<Renamable>();
-    placeholderCollector.elementNodes
-        .forEach((Element element, Set<Node> nodes) {
-      renamables.add(new GlobalRenamable(element, nodes));
-    });
-    placeholderCollector.memberPlaceholders
-        .forEach((String memberName, Set<Identifier> identifiers) {
-      renamables.add(new MemberRenamable(memberName, identifiers));
-    });
-    for (Set<Node> localIdentifiers in allLocals) {
-      renamables.add(new LocalRenamable(localIdentifiers));
-    }
-    renamables.sort();
-    for (Renamable renamable in renamables) {
-      String newName = renamable.createNewName(this);
-      _renameNodes(renamable.nodes, (_) => newName);
-    }
-  }
-
-  void _computeNonMinifiedRenames(PlaceholderCollector placeholderCollector) {
-    _generator = new ConservativeGenerator();
-    // Rename elements.
-    placeholderCollector.elementNodes
-        .forEach((Element element, Set<Node> nodes) {
-      _renameNodes(nodes, (_) => _renameGlobal(element));
-    });
-
-    // Rename locals.
-    placeholderCollector.functionScopes
-        .forEach((functionElement, functionScope) {
-      Set<String> memberIdentifiers = new Set<String>();
-      Set<LocalPlaceholder> placeholders = functionScope.localPlaceholders;
-      if (functionElement != null && functionElement.enclosingClass != null) {
-        functionElement.enclosingClass.forEachMember((enclosingClass, member) {
-          memberIdentifiers.add(member.name);
-        });
-      }
-      Set<String> usedLocalIdentifiers = new Set<String>();
-      for (LocalPlaceholder placeholder in placeholders) {
-        String nextId = _generator.generate(placeholder.identifier, (name) {
-          return functionScope.parameterIdentifiers.contains(name) ||
-              _forbiddenIdentifiers.contains(name) ||
-              usedLocalIdentifiers.contains(name) ||
-              memberIdentifiers.contains(name);
-        });
-        usedLocalIdentifiers.add(nextId);
-        _renameNodes(placeholder.nodes, (_) => nextId);
-      }
-    });
-
-    // Do not rename members to top-levels, that allows to avoid renaming
-    // members to constructors.
-    placeholderCollector.memberPlaceholders.forEach((identifier, nodes) {
-      String newIdentifier = _generateMemberName(identifier);
-      _renameNodes(nodes, (_) => newIdentifier);
-    });
-  }
-
-  /// Finds renamings for all the placeholders in [placeholderCollector] and
-  /// stores them in [renames].
-  /// Also adds to [platformImports] all the platform-libraries that are used.
-  void computeRenames(PlaceholderCollector placeholderCollector) {
-    _allNamedParameterIdentifiers = new Set<String>();
-    for (FunctionScope functionScope
-        in placeholderCollector.functionScopes.values) {
-      _allNamedParameterIdentifiers.addAll(functionScope.parameterIdentifiers);
-    }
-
-    _forbiddenIdentifiers = new Set<String>.from(fixedDynamicNames);
-    _forbiddenIdentifiers.addAll(fixedStaticNames);
-    _forbiddenIdentifiers.addAll(Keyword.keywords.keys);
-    _forbiddenIdentifiers.add('main');
-
-    if (enableMinification) {
-      _computeMinifiedRenames(placeholderCollector);
-    } else {
-      _computeNonMinifiedRenames(placeholderCollector);
-    }
-
-    // Rename constructors.
-    for (ConstructorPlaceholder placeholder
-        in placeholderCollector.constructorPlaceholders) {
-      renames[placeholder.node] = _renameConstructor(placeholder);
-    }
-    ;
-
-    // Rename private identifiers uniquely for each library.
-    placeholderCollector.privateNodes
-        .forEach((LibraryElement library, Set<Identifier> identifiers) {
-      for (Identifier identifier in identifiers) {
-        renames[identifier] = _getPrivateName(library, identifier.source);
-      }
-    });
-
-    // Rename unresolved nodes, to make sure they still do not resolve.
-    for (Node node in placeholderCollector.unresolvedNodes) {
-      renames[node] = _generateUniqueTopLevelName('Unresolved');
-    }
-
-    // Erase prefixes that are now not needed.
-    for (Node node in placeholderCollector.prefixNodesToErase) {
-      renames[node] = '';
-    }
-
-    if (cutDeclarationTypes) {
-      for (DeclarationTypePlaceholder placeholder
-          in placeholderCollector.declarationTypePlaceholders) {
-        renames[placeholder.typeNode] = placeholder.requiresVar ? 'var' : '';
-      }
-    }
-  }
-}
-
-/**
- * Generates mini ID based on index.
- * In other words, it converts index to visual representation
- * as if digits are given characters.
- */
-String generateMiniId(int index) {
-  const String firstCharAlphabet =
-      r'ABCDEFGHIJKLMNOPQRSTUVWXYZabcdefghijklmnopqrstuvwxyz';
-  const String otherCharsAlphabet =
-      r'ABCDEFGHIJKLMNOPQRSTUVWXYZabcdefghijklmnopqrstuvwxyz0123456789_$';
-  // It's like converting index in decimal to [chars] radix.
-  if (index < firstCharAlphabet.length) return firstCharAlphabet[index];
-  StringBuffer resultBuilder = new StringBuffer();
-  resultBuilder.writeCharCode(
-      firstCharAlphabet.codeUnitAt(index % firstCharAlphabet.length));
-  index ~/= firstCharAlphabet.length;
-  int length = otherCharsAlphabet.length;
-  while (index >= length) {
-    resultBuilder.writeCharCode(otherCharsAlphabet.codeUnitAt(index % length));
-    index ~/= length;
-  }
-  resultBuilder.write(otherCharsAlphabet[index]);
-  return resultBuilder.toString();
-}
-
-abstract class Generator {
-  String generate(String originalName, bool isForbidden(String name));
-}
-
-/// Always tries to return original identifier name unless it is forbidden.
-class ConservativeGenerator implements Generator {
-  String generate(String originalName, bool isForbidden(String name)) {
-    String result = originalName;
-    int index = 0;
-    while (isForbidden(result)) {
-      //|| result == originalName) {
-      result = '${originalName}_${generateMiniId(index++)}';
-    }
-    return result;
-  }
-}
-
-/// Always tries to generate the most compact identifier.
-class MinifyingGenerator implements Generator {
-  int index = 0;
-
-  MinifyingGenerator();
-
-  String generate(String originalName, bool isForbidden(String name)) {
-    String result;
-    do {
-      result = generateMiniId(index++);
-    } while (isForbidden(result));
-    return result;
-  }
-}

pkg/compiler/lib/src/diagnostics/messages.dart

diff --git a/pkg/compiler/lib/src/diagnostics/messages.dart b/pkg/compiler/lib/src/diagnostics/messages.dart
index 11c0000..472e016 100644
--- a/pkg/compiler/lib/src/diagnostics/messages.dart
+++ b/pkg/compiler/lib/src/diagnostics/messages.dart
@@ -3626,15 +3626,18 @@
       MessageKind.EXTERNAL_WITH_BODY: const MessageTemplate(
           MessageKind.EXTERNAL_WITH_BODY,
           "External function '#{functionName}' cannot have a function body.",
-          options: const ["--output-type=dart"],
           howToFix:
               "Try removing the 'external' modifier or the function body.",
           examples: const [
             """
+import 'package:js/js.dart';
+@JS()
 external foo() => 0;
 main() => foo();
 """,
             """
+import 'package:js/js.dart';
+@JS()
 external foo() {}
 main() => foo();
 """

pkg/compiler/lib/src/mirror_renamer/mirror_renamer.dart

diff --git a/pkg/compiler/lib/src/mirror_renamer/mirror_renamer.dart b/pkg/compiler/lib/src/mirror_renamer/mirror_renamer.dart
deleted file mode 100644
index e838f8e..0000000
--- a/pkg/compiler/lib/src/mirror_renamer/mirror_renamer.dart
+++ /dev/null
@@ -1,29 +0,0 @@
-// Copyright (c) 2013, the Dart project authors.  Please see the AUTHORS file
-// for details. All rights reserved. Use of this source code is governed by a
-// BSD-style license that can be found in the LICENSE file.
-
-library mirror_renamer;
-
-import '../compiler.dart' show Compiler;
-import '../dart_backend/dart_backend.dart'
-    show DartBackend, PlaceholderCollector;
-import '../elements/elements.dart';
-import '../tokens/token.dart' show Token;
-import '../tree/tree.dart';
-
-part 'renamer.dart';
-
-class MirrorRenamer {
-  const MirrorRenamer();
-
-  LibraryElement get helperLibrary => null;
-
-  FunctionElement get getNameFunction => null;
-
-  bool isMirrorHelperLibrary(LibraryElement element) => false;
-
-  void registerStaticSend(Element currentElement, Element target, Node node) {}
-
-  void addRenames(Map<Node, String> renames, List<Node> topLevelNodes,
-      PlaceholderCollector placeholderCollector) {}
-}

pkg/compiler/lib/src/mirror_renamer/renamer.dart

diff --git a/pkg/compiler/lib/src/mirror_renamer/renamer.dart b/pkg/compiler/lib/src/mirror_renamer/renamer.dart
deleted file mode 100644
index 5b2daa1..0000000
--- a/pkg/compiler/lib/src/mirror_renamer/renamer.dart
+++ /dev/null
@@ -1,118 +0,0 @@
-// Copyright (c) 2013, the Dart project authors.  Please see the AUTHORS file
-// for details. All rights reserved. Use of this source code is governed by a
-// BSD-style license that can be found in the LICENSE file.
-
-part of mirror_renamer;
-
-class MirrorRenamerImpl implements MirrorRenamer {
-  static const String MIRROR_HELPER_GET_NAME_FUNCTION = 'helperGetName';
-  static final Uri DART_MIRROR_HELPER =
-      new Uri(scheme: 'dart', path: '_mirror_helper');
-  static const String MIRROR_HELPER_SYMBOLS_MAP_NAME = '_SYMBOLS';
-
-  /// Initialized when dart:mirrors is loaded if the useMirrorHelperLibrary
-  /// field is set.
-  final LibraryElement helperLibrary;
-
-  /// Initialized when dart:mirrors is loaded if the useMirrorHelperLibrary
-  /// field is set.
-  final FunctionElement getNameFunction;
-
-  /// Initialized when dart:mirrors is loaded if the useMirrorHelperLibrary
-  /// field is set.
-  final FieldElement symbolsMapVariable;
-
-  /// Maps mangled name to original name.
-  Map<String, String> symbols = new Map<String, String>();
-
-  /// Contains all occurrencs of MirrorSystem.getName() calls in the user code.
-  List<Node> mirrorSystemGetNameNodes = <Node>[];
-
-  /**
-   *  Initialized when the placeholderCollector collects the FunctionElement
-   *  backend.mirrorHelperGetNameFunction which represents the helperGetName
-   *  function in _mirror_helper.
-   */
-  FunctionExpression get getNameFunctionNode => getNameFunction.node;
-  VariableDefinitions get symbolsMapNode => symbolsMapVariable.node;
-  Compiler compiler;
-  DartBackend backend;
-
-  MirrorRenamerImpl(this.compiler, this.backend, LibraryElement library)
-      : this.helperLibrary = library,
-        getNameFunction =
-            library.find(MirrorRenamerImpl.MIRROR_HELPER_GET_NAME_FUNCTION),
-        symbolsMapVariable =
-            library.find(MirrorRenamerImpl.MIRROR_HELPER_SYMBOLS_MAP_NAME);
-
-  bool isMirrorHelperLibrary(LibraryElement element) {
-    return element == helperLibrary;
-  }
-
-  void registerStaticSend(Element currentElement, Element target, Send node) {
-    if (target == compiler.mirrorSystemGetNameFunction &&
-        currentElement.library != helperLibrary) {
-      // Access to `MirrorSystem.getName` that needs to be redirected to the
-      // [getNameFunction].
-      mirrorSystemGetNameNodes.add(node);
-    }
-  }
-
-  /**
-   * Adds a toplevel node to the output containing a map from the mangled
-   * to the unmangled names and replaces calls to MirrorSystem.getName()
-   * with calls to the corresponding wrapper from _mirror_helper which has
-   * been added during resolution. [renames] is assumed to map nodes in user
-   * code to mangled names appearing in output code, and [topLevelNodes] should
-   * contain all the toplevel ast nodes that will be emitted in the output.
-   */
-  void addRenames(Map<Node, String> renames, List<Node> topLevelNodes,
-      PlaceholderCollector placeholderCollector) {
-    // Right now we only support instances of MirrorSystem.getName,
-    // hence if there are no occurence of these we don't do anything.
-    if (mirrorSystemGetNameNodes.isEmpty) {
-      return;
-    }
-
-    Node parse(String text) {
-      Token tokens = compiler.scanner.tokenize(text);
-      return compiler.parser.parseCompilationUnit(tokens);
-    }
-
-    // Add toplevel map containing all renames of members.
-    symbols = new Map<String, String>();
-    for (Set<Identifier> s in placeholderCollector.memberPlaceholders.values) {
-      // All members in a set have the same name so we only need to look at one.
-      Identifier sampleNode = s.first;
-      symbols.putIfAbsent(renames[sampleNode], () => sampleNode.source);
-    }
-
-    Identifier symbolsMapIdentifier =
-        symbolsMapNode.definitions.nodes.head.asSend().selector;
-    assert(symbolsMapIdentifier != null);
-    topLevelNodes.remove(symbolsMapNode);
-
-    StringBuffer sb = new StringBuffer(
-        'const ${renames[symbolsMapIdentifier]} = const<String,String>{');
-    bool first = true;
-    for (String mangledName in symbols.keys) {
-      if (!first) {
-        sb.write(',');
-      } else {
-        first = false;
-      }
-      sb.write("'$mangledName' : '");
-      sb.write(symbols[mangledName]);
-      sb.write("'");
-    }
-    sb.write('};');
-    sb.writeCharCode(0); // Terminate the string with '0', see [StringScanner].
-    topLevelNodes.add(parse(sb.toString()));
-
-    // Replace calls to Mirrorsystem.getName with calls to helper function.
-    mirrorSystemGetNameNodes.forEach((node) {
-      renames[node.selector] = renames[getNameFunctionNode.name];
-      renames[node.receiver] = '';
-    });
-  }
-}

pkg/compiler/lib/src/options.dart

diff --git a/pkg/compiler/lib/src/options.dart b/pkg/compiler/lib/src/options.dart
index c5eefce..98b88e0 100644
--- a/pkg/compiler/lib/src/options.dart
+++ b/pkg/compiler/lib/src/options.dart
@@ -245,12 +245,6 @@
   /// Whether to preserve comments while scanning (only use for dart:mirrors).
   final bool preserveComments;
 
-  /// Whether to emit JavaScript (false enables dart2dart).
-  final bool emitJavaScript;
-
-  /// When using dart2dart, whether to use the multi file format.
-  final bool dart2dartMultiFile;
-
   /// Strip option used by dart2dart.
   final List<String> strips;
 
@@ -280,7 +274,6 @@
         analyzeSignaturesOnly: _hasOption(options, Flags.analyzeSignaturesOnly),
         buildId: _extractStringOption(
             options, '--build-id=', _UNDETERMINED_BUILD_ID),
-        dart2dartMultiFile: _hasOption(options, '--output-type=dart-multi'),
         deferredMapUri: _extractUriOption(options, '--deferred-map='),
         fatalWarnings: _hasOption(options, Flags.fatalWarnings),
         terseDiagnostics: _hasOption(options, Flags.terse),
@@ -291,8 +284,6 @@
         disableInlining: _hasOption(options, Flags.disableInlining),
         disableTypeInference: _hasOption(options, Flags.disableTypeInference),
         dumpInfo: _hasOption(options, Flags.dumpInfo),
-        emitJavaScript: !(_hasOption(options, '--output-type=dart') ||
-            _hasOption(options, '--output-type=dart-multi')),
         enableAssertMessage: _hasOption(options, Flags.enableAssertMessage),
         enableGenericMethodSyntax:
             _hasOption(options, Flags.genericMethodSyntax),
@@ -354,7 +345,6 @@
       bool analyzeOnly: false,
       bool analyzeSignaturesOnly: false,
       String buildId: _UNDETERMINED_BUILD_ID,
-      bool dart2dartMultiFile: false,
       Uri deferredMapUri: null,
       bool fatalWarnings: false,
       bool terseDiagnostics: false,
@@ -364,7 +354,6 @@
       bool disableInlining: false,
       bool disableTypeInference: false,
       bool dumpInfo: false,
-      bool emitJavaScript: true,
       bool enableAssertMessage: false,
       bool enableGenericMethodSyntax: false,
       bool enableInitializingFormalAccess: false,
@@ -428,7 +417,6 @@
             analyzeOnly || analyzeSignaturesOnly || analyzeAll || resolveOnly,
         analyzeSignaturesOnly: analyzeSignaturesOnly,
         buildId: buildId,
-        dart2dartMultiFile: dart2dartMultiFile,
         deferredMapUri: deferredMapUri,
         fatalWarnings: fatalWarnings,
         terseDiagnostics: terseDiagnostics,
@@ -436,9 +424,8 @@
         suppressHints: suppressHints,
         shownPackageWarnings: shownPackageWarnings,
         disableInlining: disableInlining || hasIncrementalSupport,
-        disableTypeInference: disableTypeInference || !emitJavaScript,
+        disableTypeInference: disableTypeInference,
         dumpInfo: dumpInfo,
-        emitJavaScript: emitJavaScript,
         enableAssertMessage: enableAssertMessage,
         enableGenericMethodSyntax: enableGenericMethodSyntax,
         enableInitializingFormalAccess: enableInitializingFormalAccess,
@@ -452,8 +439,7 @@
         hasIncrementalSupport: hasIncrementalSupport,
         outputUri: outputUri,
         platformConfigUri: platformConfigUri ??
-            _resolvePlatformConfig(
-                libraryRoot, null, !emitJavaScript, const []),
+            _resolvePlatformConfig(libraryRoot, null, const []),
         preserveComments: preserveComments,
         preserveUris: preserveUris,
         resolutionInputs: resolutionInputs,
@@ -482,7 +468,6 @@
       this.analyzeOnly: false,
       this.analyzeSignaturesOnly: false,
       this.buildId: _UNDETERMINED_BUILD_ID,
-      this.dart2dartMultiFile: false,
       this.deferredMapUri: null,
       this.fatalWarnings: false,
       this.terseDiagnostics: false,
@@ -492,7 +477,6 @@
       this.disableInlining: false,
       this.disableTypeInference: false,
       this.dumpInfo: false,
-      this.emitJavaScript: true,
       this.enableAssertMessage: false,
       this.enableGenericMethodSyntax: false,
       this.enableInitializingFormalAccess: false,
@@ -541,7 +525,6 @@
       analyzeOnly,
       analyzeSignaturesOnly,
       buildId,
-      dart2dartMultiFile,
       deferredMapUri,
       fatalWarnings,
       terseDiagnostics,
@@ -551,7 +534,6 @@
       disableInlining,
       disableTypeInference,
       dumpInfo,
-      emitJavaScript,
       enableAssertMessage,
       enableGenericMethodSyntax,
       enableInitializingFormalAccess,
@@ -599,7 +581,6 @@
         analyzeSignaturesOnly:
             analyzeSignaturesOnly ?? options.analyzeSignaturesOnly,
         buildId: buildId ?? options.buildId,
-        dart2dartMultiFile: dart2dartMultiFile ?? options.dart2dartMultiFile,
         deferredMapUri: deferredMapUri ?? options.deferredMapUri,
         fatalWarnings: fatalWarnings ?? options.fatalWarnings,
         terseDiagnostics: terseDiagnostics ?? options.terseDiagnostics,
@@ -611,7 +592,6 @@
         disableTypeInference:
             disableTypeInference ?? options.disableTypeInference,
         dumpInfo: dumpInfo ?? options.dumpInfo,
-        emitJavaScript: emitJavaScript ?? options.emitJavaScript,
         enableAssertMessage: enableAssertMessage ?? options.enableAssertMessage,
         enableGenericMethodSyntax:
             enableGenericMethodSyntax ?? options.enableGenericMethodSyntax,
@@ -722,11 +702,9 @@
 }
 
 Uri _resolvePlatformConfig(Uri libraryRoot, String platformConfigPath,
-    bool isDart2Dart, Iterable<String> categories) {
+    Iterable<String> categories) {
   if (platformConfigPath != null) {
     return libraryRoot.resolve(platformConfigPath);
-  } else if (isDart2Dart) {
-    return libraryRoot.resolve(_dart2dartPlatform);
   } else {
     if (categories.length == 0) {
       return libraryRoot.resolve(_clientPlatform);
@@ -747,7 +725,6 @@
   return _resolvePlatformConfig(
       libraryRoot,
       _extractStringOption(options, "--platform-config=", null),
-      _hasOption(options, '--output-type=dart'),
       _extractCsvOption(options, '--categories='));
 }
 
@@ -755,7 +732,6 @@
 const String _clientPlatform = "lib/dart_client.platform";
 const String _serverPlatform = "lib/dart_server.platform";
 const String _sharedPlatform = "lib/dart_shared.platform";
-const String _dart2dartPlatform = "lib/dart2dart.platform";
 
 const String _UNDETERMINED_BUILD_ID = "build number could not be determined";
 const bool _forceIncrementalSupport =

pkg/compiler/lib/src/ssa/interceptor_simplifier.dart

diff --git a/pkg/compiler/lib/src/ssa/interceptor_simplifier.dart b/pkg/compiler/lib/src/ssa/interceptor_simplifier.dart
index 3249c31..6a5c82d 100644
--- a/pkg/compiler/lib/src/ssa/interceptor_simplifier.dart
+++ b/pkg/compiler/lib/src/ssa/interceptor_simplifier.dart
@@ -232,17 +232,24 @@
       if (interceptedClasses.contains(helpers.jsNumberClass) &&
           !(interceptedClasses.contains(helpers.jsDoubleClass) ||
               interceptedClasses.contains(helpers.jsIntClass))) {
+        Set<ClassElement> required;
         for (HInstruction user in node.usedBy) {
           if (user is! HInvoke) continue;
           Set<ClassElement> intercepted =
               backend.getInterceptedClassesOn(user.selector.name);
           if (intercepted.contains(helpers.jsIntClass)) {
-            interceptedClasses.add(helpers.jsIntClass);
+            required ??= new Set<ClassElement>();
+            required.add(helpers.jsIntClass);
           }
           if (intercepted.contains(helpers.jsDoubleClass)) {
-            interceptedClasses.add(helpers.jsDoubleClass);
+            required ??= new Set<ClassElement>();
+            required.add(helpers.jsDoubleClass);
           }
         }
+        // Don't modify the result of [backend.getInterceptedClassesOn].
+        if (required != null) {
+          interceptedClasses = interceptedClasses.union(required);
+        }
       }
     } else {
       interceptedClasses = new Set<ClassElement>();

runtime/bin/io_natives.cc

diff --git a/runtime/bin/io_natives.cc b/runtime/bin/io_natives.cc
index 609bbef..8c9c8c8 100644
--- a/runtime/bin/io_natives.cc
+++ b/runtime/bin/io_natives.cc
@@ -121,6 +121,7 @@
   V(Socket_CreateConnect, 3)                                                   \
   V(Socket_CreateBindConnect, 4)                                               \
   V(Socket_CreateBindDatagram, 4)                                              \
+  V(Socket_IsBindError, 2)                                                     \
   V(Socket_Available, 1)                                                       \
   V(Socket_Read, 2)                                                            \
   V(Socket_RecvFrom, 1)                                                        \

runtime/bin/secure_socket_boringssl.cc

diff --git a/runtime/bin/secure_socket_boringssl.cc b/runtime/bin/secure_socket_boringssl.cc
index def3aae..e881815 100644
--- a/runtime/bin/secure_socket_boringssl.cc
+++ b/runtime/bin/secure_socket_boringssl.cc
@@ -94,11 +94,14 @@
                              const char* message) {
   char error_string[SSL_ERROR_MESSAGE_BUFFER_SIZE];
   FetchErrorString(error_string, SSL_ERROR_MESSAGE_BUFFER_SIZE);
-  OSError os_error_struct(status, error_string, OSError::kBoringSSL);
-  Dart_Handle os_error = DartUtils::NewDartOSError(&os_error_struct);
-  Dart_Handle exception =
-      DartUtils::NewDartIOException(exception_type, message, os_error);
-  ASSERT(!Dart_IsError(exception));
+  Dart_Handle exception;
+  {
+    OSError os_error_struct(status, error_string, OSError::kBoringSSL);
+    Dart_Handle os_error = DartUtils::NewDartOSError(&os_error_struct);
+    exception =
+        DartUtils::NewDartIOException(exception_type, message, os_error);
+    ASSERT(!Dart_IsError(exception));
+  }
   Dart_ThrowException(exception);
   UNREACHABLE();
 }
@@ -143,8 +146,8 @@
 }
 
 
-static SSL_CTX* GetSecurityContext(Dart_NativeArguments args) {
-  SSL_CTX* context;
+static SSLContext* GetSecurityContext(Dart_NativeArguments args) {
+  SSLContext* context;
   Dart_Handle dart_this = ThrowIfError(Dart_GetNativeArgument(args, 0));
   ASSERT(Dart_IsInstance(dart_this));
   ThrowIfError(Dart_GetNativeInstanceField(
@@ -155,17 +158,17 @@
 }
 
 
-static void FreeSecurityContext(
+static void DeleteSecurityContext(
     void* isolate_data,
     Dart_WeakPersistentHandle handle,
     void* context_pointer) {
-  SSL_CTX* context = static_cast<SSL_CTX*>(context_pointer);
-  SSL_CTX_free(context);
+  SSLContext* context = static_cast<SSLContext*>(context_pointer);
+  delete context;
 }
 
 
 static Dart_Handle SetSecurityContext(Dart_NativeArguments args,
-                                      SSL_CTX* context) {
+                                      SSLContext* context) {
   const int approximate_size_of_context = 1500;
   Dart_Handle dart_this = Dart_GetNativeArgument(args, 0);
   RETURN_IF_ERROR(dart_this);
@@ -178,7 +181,7 @@
   Dart_NewWeakPersistentHandle(dart_this,
                                context,
                                approximate_size_of_context,
-                               FreeSecurityContext);
+                               DeleteSecurityContext);
   return Dart_Null();
 }
 
@@ -198,7 +201,7 @@
 // Forward declaration.
 static void SetAlpnProtocolList(Dart_Handle protocols_handle,
                                 SSL* ssl,
-                                SSL_CTX* context,
+                                SSLContext* context,
                                 bool is_server);
 
 
@@ -235,7 +238,7 @@
   // TODO(whesse): Is truncating a Dart string containing \0 what we want?
   ThrowIfError(Dart_StringToCString(host_name_object, &host_name));
 
-  SSL_CTX* context = NULL;
+  SSLContext* context = NULL;
   if (!Dart_IsNull(context_object)) {
     ThrowIfError(Dart_GetNativeInstanceField(
         context_object,
@@ -248,7 +251,7 @@
   ASSERT(!Dart_IsNull(protocols_handle));
 
   GetFilter(args)->Connect(host_name,
-                           context,
+                           context->context(),
                            is_server,
                            request_client_certificate,
                            require_client_certificate,
@@ -426,14 +429,15 @@
 
 void FUNCTION_NAME(SecurityContext_Allocate)(Dart_NativeArguments args) {
   SSLFilter::InitializeLibrary();
-  SSL_CTX* context = SSL_CTX_new(TLS_method());
-  SSL_CTX_set_verify(context, SSL_VERIFY_PEER, CertificateCallback);
-  SSL_CTX_set_min_version(context, TLS1_VERSION);
-  SSL_CTX_set_cipher_list(context, "HIGH:MEDIUM");
-  SSL_CTX_set_cipher_list_tls11(context, "HIGH:MEDIUM");
+  SSL_CTX* ctx = SSL_CTX_new(TLS_method());
+  SSL_CTX_set_verify(ctx, SSL_VERIFY_PEER, CertificateCallback);
+  SSL_CTX_set_min_version(ctx, TLS1_VERSION);
+  SSL_CTX_set_cipher_list(ctx, "HIGH:MEDIUM");
+  SSL_CTX_set_cipher_list_tls11(ctx, "HIGH:MEDIUM");
+  SSLContext* context = new SSLContext(ctx);
   Dart_Handle err = SetSecurityContext(args, context);
   if (Dart_IsError(err)) {
-    SSL_CTX_free(context);
+    delete context;
     Dart_PropagateError(err);
   }
 }
@@ -663,14 +667,17 @@
 
 void FUNCTION_NAME(SecurityContext_UsePrivateKeyBytes)(
     Dart_NativeArguments args) {
-  SSL_CTX* context = GetSecurityContext(args);
+  SSLContext* context = GetSecurityContext(args);
   const char* password = GetPasswordArgument(args, 2);
 
   int status;
   {
     ScopedMemBIO bio(ThrowIfError(Dart_GetNativeArgument(args, 1)));
     EVP_PKEY *key = GetPrivateKey(bio.bio(), password);
-    status = SSL_CTX_use_PrivateKey(context, key);
+    status = SSL_CTX_use_PrivateKey(context->context(), key);
+    // SSL_CTX_use_PrivateKey increments the reference count of key on success,
+    // so we have to call EVP_PKEY_free on both success and failure.
+    EVP_PKEY_free(key);
   }
 
   // TODO(24184): Handle different expected errors here - file missing,
@@ -699,16 +706,18 @@
   ScopedX509Stack cert_stack(ca_certs);
   X509_STORE* store = SSL_CTX_get_cert_store(context);
   status = X509_STORE_add_cert(store, cert);
+  // X509_STORE_add_cert increments the reference count of cert on success.
+  X509_free(cert);
   if (status == 0) {
-    X509_free(cert);
     return status;
   }
 
   X509* ca;
   while ((ca = sk_X509_shift(cert_stack.get())) != NULL) {
     status = X509_STORE_add_cert(store, ca);
+    // X509_STORE_add_cert increments the reference count of cert on success.
+    X509_free(ca);
     if (status == 0) {
-      X509_free(ca);
       return status;
     }
   }
@@ -724,8 +733,9 @@
   X509* cert = NULL;
   while ((cert = PEM_read_bio_X509(bio, NULL, NULL, NULL)) != NULL) {
     status = X509_STORE_add_cert(store, cert);
+    // X509_STORE_add_cert increments the reference count of cert on success.
+    X509_free(cert);
     if (status == 0) {
-      X509_free(cert);
       return status;
     }
   }
@@ -759,12 +769,13 @@
 
 void FUNCTION_NAME(SecurityContext_SetTrustedCertificatesBytes)(
     Dart_NativeArguments args) {
-  SSL_CTX* context = GetSecurityContext(args);
+  SSLContext* context = GetSecurityContext(args);
   const char* password = GetPasswordArgument(args, 2);
   int status;
   {
     ScopedMemBIO bio(ThrowIfError(Dart_GetNativeArgument(args, 1)));
-    status = SetTrustedCertificatesBytes(context, bio.bio(), password);
+    status = SetTrustedCertificatesBytes(
+        context->context(), bio.bio(), password);
   }
   CheckStatus(status,
               "TlsException",
@@ -779,7 +790,7 @@
 
 void FUNCTION_NAME(SecurityContext_TrustBuiltinRoots)(
     Dart_NativeArguments args) {
-  SSL_CTX* context = GetSecurityContext(args);
+  SSLContext* context = GetSecurityContext(args);
 #if defined(TARGET_OS_ANDROID)
   // On Android, we don't compile in the trusted root certificates. Insead,
   // we use the directory of trusted certificates already present on the device.
@@ -789,10 +800,11 @@
   // the Dart thread so that Dart code can be invoked from the "bad certificate"
   // callback called by SSL_do_handshake.
   const char* android_cacerts = "/system/etc/security/cacerts";
-  int status = SSL_CTX_load_verify_locations(context, NULL, android_cacerts);
+  int status = SSL_CTX_load_verify_locations(
+      context->context(), NULL, android_cacerts);
   CheckStatus(status, "TlsException", "Failure trusting builtint roots");
 #else
-  X509_STORE* store = SSL_CTX_get_cert_store(context);
+  X509_STORE* store = SSL_CTX_get_cert_store(context->context());
   BIO* roots_bio =
       BIO_new_mem_buf(const_cast<unsigned char*>(root_certificates_pem),
                       root_certificates_pem_length);
@@ -801,7 +813,12 @@
   // backed by a memory buffer), and returns X509 objects, one by one.
   // When the end of the bio is reached, it returns null.
   while ((root_cert = PEM_read_bio_X509(roots_bio, NULL, NULL, NULL)) != NULL) {
-    X509_STORE_add_cert(store, root_cert);
+    int status = X509_STORE_add_cert(store, root_cert);
+    // X509_STORE_add_cert increments the reference count of cert on success.
+    X509_free(root_cert);
+    if (status == 0) {
+      break;
+    }
   }
   BIO_free(roots_bio);
   // If there is an error here, it must be the error indicating that we are done
@@ -844,6 +861,8 @@
   X509* ca;
   while ((ca = sk_X509_shift(certs.get())) != NULL) {
     status = SSL_CTX_add0_chain_cert(context, ca);
+    // SSL_CTX_add0_chain_cert does not inc ref count, so don't free unless the
+    // call fails.
     if (status == 0) {
       X509_free(ca);
       return status;
@@ -875,6 +894,8 @@
   X509* ca;
   while ((ca = PEM_read_bio_X509(bio, NULL, NULL, NULL)) != NULL) {
     status = SSL_CTX_add0_chain_cert(context, ca);
+    // SSL_CTX_add0_chain_cert does not inc ref count, so don't free unless the
+    // call fails.
     if (status == 0) {
       X509_free(ca);
       return status;
@@ -906,12 +927,12 @@
 
 void FUNCTION_NAME(SecurityContext_UseCertificateChainBytes)(
     Dart_NativeArguments args) {
-  SSL_CTX* context = GetSecurityContext(args);
+  SSLContext* context = GetSecurityContext(args);
   const char* password = GetPasswordArgument(args, 2);
   int status;
   {
     ScopedMemBIO bio(ThrowIfError(Dart_GetNativeArgument(args, 1)));
-    status = UseChainBytes(context, bio.bio(), password);
+    status = UseChainBytes(context->context(), bio.bio(), password);
   }
   CheckStatus(status,
               "TlsException",
@@ -937,14 +958,16 @@
 
   ScopedX509Stack cert_stack(ca_certs);
   status = SSL_CTX_add_client_CA(context, cert);
+  // SSL_CTX_add_client_CA increments the reference count of cert on success.
+  X509_free(cert);
   if (status == 0) {
-    X509_free(cert);
     return status;
   }
 
   X509* ca;
   while ((ca = sk_X509_shift(cert_stack.get())) != NULL) {
     status = SSL_CTX_add_client_CA(context, ca);
+    // SSL_CTX_add_client_CA increments the reference count of ca on success.
     X509_free(ca);  // The name has been extracted.
     if (status == 0) {
       return status;
@@ -989,13 +1012,13 @@
 
 void FUNCTION_NAME(SecurityContext_SetClientAuthoritiesBytes)(
     Dart_NativeArguments args) {
-  SSL_CTX* context = GetSecurityContext(args);
+  SSLContext* context = GetSecurityContext(args);
   const char* password = GetPasswordArgument(args, 2);
 
   int status;
   {
     ScopedMemBIO bio(ThrowIfError(Dart_GetNativeArgument(args, 1)));
-    status = SetClientAuthorities(context, bio.bio(), password);
+    status = SetClientAuthorities(context->context(), bio.bio(), password);
   }
 
   CheckStatus(status,
@@ -1006,7 +1029,7 @@
 
 void FUNCTION_NAME(SecurityContext_SetAlpnProtocols)(
     Dart_NativeArguments args) {
-  SSL_CTX* context = GetSecurityContext(args);
+  SSLContext* context = GetSecurityContext(args);
   Dart_Handle protocols_handle =
       ThrowIfError(Dart_GetNativeArgument(args, 1));
   Dart_Handle is_server_handle =
@@ -1363,7 +1386,7 @@
 // Sets the protocol list for ALPN on a SSL object or a context.
 static void SetAlpnProtocolList(Dart_Handle protocols_handle,
                                 SSL* ssl,
-                                SSL_CTX* context,
+                                SSLContext* context,
                                 bool is_server) {
   // Enable ALPN (application layer protocol negotiation) if the caller provides
   // a valid list of supported protocols.
@@ -1400,13 +1423,9 @@
           static_cast<uint8_t*>(malloc(protocol_string_len + 1));
       memmove(protocol_string_copy, protocol_string, protocol_string_len);
       protocol_string_copy[protocol_string_len] = '\0';
-      SSL_CTX_set_alpn_select_cb(context, AlpnCallback, protocol_string_copy);
-      // TODO(whesse): If this function is called again, free the previous
-      // protocol_string_copy.  It may be better to keep this as a native
-      // field on the Dart object, since fetching it from the structure is
-      // not in the public api.
-      // Also free protocol_string_copy when the context is destroyed,
-      // in FreeSecurityContext()
+      SSL_CTX_set_alpn_select_cb(
+          context->context(), AlpnCallback, protocol_string_copy);
+      context->set_alpn_protocol_string(protocol_string_copy);
     } else {
       // The function makes a local copy of protocol_string, which it owns.
       if (ssl != NULL) {
@@ -1416,7 +1435,7 @@
         ASSERT(context != NULL);
         ASSERT(ssl == NULL);
         status = SSL_CTX_set_alpn_protos(
-            context, protocol_string, protocol_string_len);
+            context->context(), protocol_string, protocol_string_len);
       }
       ASSERT(status == 0);  // The function returns a non-standard status.
     }

runtime/bin/secure_socket_boringssl.h

diff --git a/runtime/bin/secure_socket_boringssl.h b/runtime/bin/secure_socket_boringssl.h
index d5449a4..788b257 100644
--- a/runtime/bin/secure_socket_boringssl.h
+++ b/runtime/bin/secure_socket_boringssl.h
@@ -33,8 +33,39 @@
 extern const unsigned char* root_certificates_pem;
 extern unsigned int root_certificates_pem_length;
 
+class SSLContext {
+ public:
+  explicit SSLContext(SSL_CTX* context) :
+      context_(context),
+      alpn_protocol_string_(NULL) {
+  }
+
+  ~SSLContext() {
+    SSL_CTX_free(context_);
+    if (alpn_protocol_string_ != NULL) {
+      free(alpn_protocol_string_);
+    }
+  }
+
+  SSL_CTX* context() const { return context_; }
+
+  uint8_t* alpn_protocol_string() const { return alpn_protocol_string_; }
+  void set_alpn_protocol_string(uint8_t* protocol_string) {
+    if (alpn_protocol_string_ != NULL) {
+      free(alpn_protocol_string_);
+    }
+    alpn_protocol_string_ = protocol_string;
+  }
+
+ private:
+  SSL_CTX* context_;
+  uint8_t* alpn_protocol_string_;
+
+  DISALLOW_COPY_AND_ASSIGN(SSLContext);
+};
+
 /*
- * SSLFilter encapsulates the NSS SSL(TLS) code in a filter, that communicates
+ * SSLFilter encapsulates the SSL(TLS) code in a filter, that communicates
  * with the containing _SecureFilterImpl Dart object through four shared
  * ExternalByteArray buffers, for reading and writing plaintext, and
  * reading and writing encrypted text.  The filter handles handshaking

runtime/bin/socket.cc

diff --git a/runtime/bin/socket.cc b/runtime/bin/socket.cc
index 7b6f95a..745306d 100644
--- a/runtime/bin/socket.cc
+++ b/runtime/bin/socket.cc
@@ -234,6 +234,12 @@
   }
 }
 
+void FUNCTION_NAME(Socket_IsBindError)(Dart_NativeArguments args) {
+  intptr_t error_number =
+      DartUtils::GetIntptrValue(Dart_GetNativeArgument(args, 1));
+  bool is_bind_error = Socket::IsBindError(error_number);
+  Dart_SetReturnValue(args, is_bind_error ? Dart_True() : Dart_False());
+}
 
 void FUNCTION_NAME(Socket_CreateBindDatagram)(Dart_NativeArguments args) {
   RawAddr addr;

runtime/bin/socket.h

diff --git a/runtime/bin/socket.h b/runtime/bin/socket.h
index 0b17706..6a50d79 100644
--- a/runtime/bin/socket.h
+++ b/runtime/bin/socket.h
@@ -278,6 +278,9 @@
   // specified as the port component of the passed RawAddr structure.
   static intptr_t CreateBindConnect(const RawAddr& addr,
                                     const RawAddr& source_addr);
+  // Returns true if the given error-number is because the system was not able
+  // to bind the socket to a specific IP.
+  static bool IsBindError(intptr_t error_number);
   // Creates a datagram socket which is bound. The port to bind
   // to is specified as the port component of the RawAddr structure.
   static intptr_t CreateBindDatagram(const RawAddr& addr, bool reuseAddress);

runtime/bin/socket_android.cc

diff --git a/runtime/bin/socket_android.cc b/runtime/bin/socket_android.cc
index b3e3fc4..caddb3a 100644
--- a/runtime/bin/socket_android.cc
+++ b/runtime/bin/socket_android.cc
@@ -101,6 +101,12 @@
 }
 
 
+bool Socket::IsBindError(intptr_t error_number) {
+  return error_number == EADDRINUSE || error_number == EADDRNOTAVAIL ||
+      error_number == EINVAL;
+}
+
+
 intptr_t Socket::Available(intptr_t fd) {
   return FDUtils::AvailableBytes(fd);
 }

runtime/bin/socket_fuchsia.cc

diff --git a/runtime/bin/socket_fuchsia.cc b/runtime/bin/socket_fuchsia.cc
index 78b5071..b79d1c0 100644
--- a/runtime/bin/socket_fuchsia.cc
+++ b/runtime/bin/socket_fuchsia.cc
@@ -45,6 +45,12 @@
 }
 
 
+bool Socket::IsBindError(intptr_t error_number) {
+  UNIMPLEMENTED();
+  return false;
+}
+
+
 intptr_t Socket::Available(intptr_t fd) {
   UNIMPLEMENTED();
   return -1;

runtime/bin/socket_linux.cc

diff --git a/runtime/bin/socket_linux.cc b/runtime/bin/socket_linux.cc
index 6393567..fc0f1e4 100644
--- a/runtime/bin/socket_linux.cc
+++ b/runtime/bin/socket_linux.cc
@@ -101,6 +101,12 @@
 }
 
 
+bool Socket::IsBindError(intptr_t error_number) {
+  return error_number == EADDRINUSE || error_number == EADDRNOTAVAIL ||
+      error_number == EINVAL;
+}
+
+
 intptr_t Socket::Available(intptr_t fd) {
   return FDUtils::AvailableBytes(fd);
 }

runtime/bin/socket_macos.cc

diff --git a/runtime/bin/socket_macos.cc b/runtime/bin/socket_macos.cc
index 708cd3c..e54cc59 100644
--- a/runtime/bin/socket_macos.cc
+++ b/runtime/bin/socket_macos.cc
@@ -58,6 +58,7 @@
     return -1;
   }
   FDUtils::SetCloseOnExec(fd);
+  FDUtils::SetNonBlocking(fd);
   return fd;
 }
 
@@ -79,8 +80,6 @@
     return fd;
   }
 
-  FDUtils::SetNonBlocking(fd);
-
   return Connect(fd, addr);
 }
 
@@ -103,6 +102,12 @@
 }
 
 
+bool Socket::IsBindError(intptr_t error_number) {
+  return error_number == EADDRINUSE || error_number == EADDRNOTAVAIL ||
+      error_number == EINVAL;
+}
+
+
 intptr_t Socket::Available(intptr_t fd) {
   return FDUtils::AvailableBytes(fd);
 }

runtime/bin/socket_patch.dart

diff --git a/runtime/bin/socket_patch.dart b/runtime/bin/socket_patch.dart
index 2d3ba73..a8b60c3 100644
--- a/runtime/bin/socket_patch.dart
+++ b/runtime/bin/socket_patch.dart
@@ -431,7 +431,13 @@
             if (result is OSError) {
               // Keep first error, if present.
               if (error == null) {
-                error = createError(result, "Connection failed", address, port);
+                int errorCode = result.errorCode;
+                if (errorCode != null && socket.isBindError(errorCode)) {
+                  error = createError(result, "Bind failed", sourceAddress);
+                } else {
+                  error =
+                      createError(result, "Connection failed", address, port);
+                }
               }
               connectNext();
             } else {
@@ -1075,6 +1081,7 @@
   nativeCreateBindConnect(
       List<int> addr, int port, List<int> sourceAddr)
       native "Socket_CreateBindConnect";
+  bool isBindError(int errorNumber) native "Socket_IsBindError";
   nativeCreateBindListen(List<int> addr, int port, int backlog, bool v6Only,
                          bool shared)
       native "ServerSocket_CreateBindListen";

runtime/bin/socket_unsupported.cc

diff --git a/runtime/bin/socket_unsupported.cc b/runtime/bin/socket_unsupported.cc
index 0ea27f1..089e48e 100644
--- a/runtime/bin/socket_unsupported.cc
+++ b/runtime/bin/socket_unsupported.cc
@@ -35,6 +35,12 @@
 }
 
 
+void FUNCTION_NAME(Socket_IsBindError)(Dart_NativeArguments args) {
+  Dart_ThrowException(DartUtils::NewDartArgumentError(
+      "Sockets unsupported on this platform"));
+}
+
+
 void FUNCTION_NAME(Socket_CreateBindDatagram)(Dart_NativeArguments args) {
   Dart_ThrowException(DartUtils::NewDartArgumentError(
       "Sockets unsupported on this platform"));

runtime/bin/socket_win.cc

diff --git a/runtime/bin/socket_win.cc b/runtime/bin/socket_win.cc
index e3d366d..f76f764 100644
--- a/runtime/bin/socket_win.cc
+++ b/runtime/bin/socket_win.cc
@@ -250,6 +250,12 @@
 }
 
 
+bool Socket::IsBindError(intptr_t error_number) {
+  return error_number == WSAEADDRINUSE || error_number == WSAEADDRNOTAVAIL ||
+      error_number == WSAEINVAL;
+}
+
+
 void Socket::GetError(intptr_t fd, OSError* os_error) {
   Handle* handle = reinterpret_cast<Handle*>(fd);
   os_error->SetCodeAndMessage(OSError::kSystem, handle->last_error());

runtime/lib/double.dart

diff --git a/runtime/lib/double.dart b/runtime/lib/double.dart
index ade4b93..34d7803 100644
--- a/runtime/lib/double.dart
+++ b/runtime/lib/double.dart
@@ -277,4 +277,11 @@
       return LESS;
     }
   }
+
+  static const int _FRACTIONAL_BITS = // Bits to keep after the decimal point.
+      const int.fromEnvironment("doubleFractionalBits", defaultValue: 20);
+  static const double _BIAS = 1.5 * (1 << (52 - _FRACTIONAL_BITS));
+
+  // Returns this with only _FRACTIONAL_BITS bits after the decimal point.
+  double get p => this + _BIAS - _BIAS;
 }

runtime/observatory/lib/app.dart

diff --git a/runtime/observatory/lib/app.dart b/runtime/observatory/lib/app.dart
index 371c266..4856448 100644
--- a/runtime/observatory/lib/app.dart
+++ b/runtime/observatory/lib/app.dart
@@ -12,7 +12,6 @@
 import 'package:logging/logging.dart';
 import 'package:observatory/service_html.dart';
 import 'package:observatory/elements.dart';
-import 'package:observatory/mocks.dart';
 import 'package:observatory/models.dart' as M;
 import 'package:observatory/repositories.dart';
 import 'package:observatory/tracer.dart';
@@ -23,6 +22,7 @@
 export 'package:observatory/utils.dart';
 
 part 'src/app/application.dart';
+part 'src/app/event.dart';
 part 'src/app/location_manager.dart';
 part 'src/app/notification.dart';
 part 'src/app/page.dart';

runtime/observatory/lib/elements.dart

diff --git a/runtime/observatory/lib/elements.dart b/runtime/observatory/lib/elements.dart
index 6e2f4f8..ecc32ac 100644
--- a/runtime/observatory/lib/elements.dart
+++ b/runtime/observatory/lib/elements.dart
@@ -3,7 +3,6 @@
 // Export elements.
 export 'package:observatory/src/elements/action_link.dart';
 export 'package:observatory/src/elements/class_ref_as_value.dart';
-export 'package:observatory/src/elements/class_tree.dart';
 export 'package:observatory/src/elements/class_view.dart';
 export 'package:observatory/src/elements/code_view.dart';
 export 'package:observatory/src/elements/context_ref.dart';
@@ -53,6 +52,7 @@
 
 import 'package:observatory/src/elements/class_ref.dart';
 import 'package:observatory/src/elements/class_ref_wrapper.dart';
+import 'package:observatory/src/elements/class_tree.dart';
 import 'package:observatory/src/elements/code_ref.dart';
 import 'package:observatory/src/elements/code_ref_wrapper.dart';
 import 'package:observatory/src/elements/containers/virtual_collection.dart';
@@ -101,6 +101,7 @@
 export 'package:observatory/src/elements/helpers/rendering_queue.dart';
 
 export 'package:observatory/src/elements/class_ref.dart';
+export 'package:observatory/src/elements/class_tree.dart';
 export 'package:observatory/src/elements/code_ref.dart';
 export 'package:observatory/src/elements/containers/virtual_collection.dart';
 export 'package:observatory/src/elements/containers/virtual_tree.dart';
@@ -134,6 +135,7 @@
 Future initElements() async {
   ClassRefElement.tag.ensureRegistration();
   ClassRefElementWrapper.tag.ensureRegistration();
+  ClassTreeElement.tag.ensureRegistration();
   CodeRefElement.tag.ensureRegistration();
   CodeRefElementWrapper.tag.ensureRegistration();
   CurlyBlockElement.tag.ensureRegistration();

runtime/observatory/lib/elements.html

diff --git a/runtime/observatory/lib/elements.html b/runtime/observatory/lib/elements.html
index ed5c73c..2b42d47 100644
--- a/runtime/observatory/lib/elements.html
+++ b/runtime/observatory/lib/elements.html
@@ -1,6 +1,5 @@
 <link rel="import" href="src/elements/action_link.html">
 <link rel="import" href="src/elements/class_ref_as_value.html">
-<link rel="import" href="src/elements/class_tree.html">
 <link rel="import" href="src/elements/class_view.html">
 <link rel="import" href="src/elements/code_view.html">
 <link rel="import" href="src/elements/cpu_profile.html">

runtime/observatory/lib/mocks.dart

diff --git a/runtime/observatory/lib/mocks.dart b/runtime/observatory/lib/mocks.dart
deleted file mode 100644
index 2535fa0..0000000
--- a/runtime/observatory/lib/mocks.dart
+++ /dev/null
@@ -1,31 +0,0 @@
-// Copyright (c) 2016, the Dart project authors.  Please see the AUTHORS file
-// for details. All rights reserved. Use of this source code is governed by a
-// BSD-style license that can be found in the LICENSE file.
-
-library mocks;
-
-import 'dart:async';
-
-import 'package:observatory/models.dart' as M;
-
-part 'src/mocks/exceptions/connection_exception.dart';
-
-part 'src/mocks/objects/error.dart';
-part 'src/mocks/objects/code.dart';
-part 'src/mocks/objects/event.dart';
-part 'src/mocks/objects/class.dart';
-part 'src/mocks/objects/flag.dart';
-part 'src/mocks/objects/function.dart';
-part 'src/mocks/objects/isolate.dart';
-part 'src/mocks/objects/library.dart';
-part 'src/mocks/objects/notification.dart';
-part 'src/mocks/objects/script.dart';
-part 'src/mocks/objects/source_location.dart';
-part 'src/mocks/objects/target.dart';
-part 'src/mocks/objects/vm.dart';
-
-part 'src/mocks/repositories/crash_dump.dart';
-part 'src/mocks/repositories/flag.dart';
-part 'src/mocks/repositories/notification.dart';
-part 'src/mocks/repositories/script.dart';
-part 'src/mocks/repositories/target.dart';

runtime/observatory/lib/models.dart

diff --git a/runtime/observatory/lib/models.dart b/runtime/observatory/lib/models.dart
index 6fa8853..ded6d78 100644
--- a/runtime/observatory/lib/models.dart
+++ b/runtime/observatory/lib/models.dart
@@ -28,9 +28,10 @@
 part 'src/models/objects/timeline_event.dart';
 part 'src/models/objects/vm.dart';
 
-part 'src/models/repository.dart';
-part 'src/models/repositories/crash_dump.dart';
+part 'src/models/repositories/class.dart';
+part 'src/models/repositories/event.dart';
 part 'src/models/repositories/flag.dart';
+part 'src/models/repositories/instance.dart';
 part 'src/models/repositories/notification.dart';
 part 'src/models/repositories/script.dart';
 part 'src/models/repositories/target.dart';

runtime/observatory/lib/repositories.dart

diff --git a/runtime/observatory/lib/repositories.dart b/runtime/observatory/lib/repositories.dart
index feeb8f2..9f3c340 100644
--- a/runtime/observatory/lib/repositories.dart
+++ b/runtime/observatory/lib/repositories.dart
@@ -12,7 +12,10 @@
 import 'package:observatory/service_common.dart' as SC;
 import 'package:observatory/utils.dart';
 
+part 'src/repositories/class.dart';
+part 'src/repositories/event.dart';
 part 'src/repositories/flag.dart';
+part 'src/repositories/instance.dart';
 part 'src/repositories/notification.dart';
 part 'src/repositories/script.dart';
 part 'src/repositories/settings.dart';

runtime/observatory/lib/service_common.dart

diff --git a/runtime/observatory/lib/service_common.dart b/runtime/observatory/lib/service_common.dart
index 8e7b5c0..cd146ee 100644
--- a/runtime/observatory/lib/service_common.dart
+++ b/runtime/observatory/lib/service_common.dart
@@ -91,6 +91,8 @@
   bool _hasFinishedConnect = false;
   Utf8Decoder _utf8Decoder = const Utf8Decoder();
 
+  String get displayName => '${name}@${target.name}';
+
   CommonWebSocket _webSocket;
 
   CommonWebSocketVM(this.target, this._webSocket) {

runtime/observatory/lib/src/app/application.dart

diff --git a/runtime/observatory/lib/src/app/application.dart b/runtime/observatory/lib/src/app/application.dart
index a8bfabb..d72994d 100644
--- a/runtime/observatory/lib/src/app/application.dart
+++ b/runtime/observatory/lib/src/app/application.dart
@@ -10,6 +10,7 @@
   static ObservatoryApplication app;
   final RenderingQueue queue = new RenderingQueue();
   final TargetRepository targets = new TargetRepository();
+  final EventRepository events = new EventRepository();
   final NotificationRepository notifications = new NotificationRepository();
   final _pageRegistry = new List<Page>();
   LocationManager _locationManager;
@@ -40,11 +41,11 @@
           // This disconnect event occured *after* a new VM was installed.
           return;
         }
-        notifications.add(
-            new EventNotification.fromServiceEvent(
-              new ServiceEvent.connectionClosed(reason)));
+        events.add(new ConnectionClosedEvent(new DateTime.now(), reason));
       });
 
+      // TODO(cbernaschina) smart connection of streams in the events object.
+      vm.listenEventStream(VM.kVMStream, _onEvent);
       vm.listenEventStream(VM.kIsolateStream, _onEvent);
       vm.listenEventStream(VM.kDebugStream, _onEvent);
     }
@@ -65,48 +66,84 @@
     locationManager._visit();
   }
 
+  void _deletePauseEvents(e) {
+    notifications.deletePauseEvents(isolate: e.isolate);
+  }
+  void _addNotification(M.Event e) {
+    notifications.add(new EventNotification(e));
+  }
+
   void _onEvent(ServiceEvent event) {
     assert(event.kind != ServiceEvent.kNone);
 
+    M.Event e;
+
     switch(event.kind) {
       case ServiceEvent.kVMUpdate:
+        e = new VMUpdateEvent(event.timestamp, event.vm);
+        break;
       case ServiceEvent.kIsolateStart:
+        e = new IsolateStartEvent(event.timestamp, event.isolate);
+        break;
       case ServiceEvent.kIsolateRunnable:
+        e = new IsolateRunnableEvent(event.timestamp, event.isolate);
+        break;
       case ServiceEvent.kIsolateUpdate:
-      case ServiceEvent.kBreakpointAdded:
-      case ServiceEvent.kBreakpointResolved:
-      case ServiceEvent.kBreakpointRemoved:
-      case ServiceEvent.kDebuggerSettingsUpdate:
-        // Ignore for now.
+        e = new IsolateUpdateEvent(event.timestamp, event.isolate);
         break;
-
       case ServiceEvent.kIsolateReload:
-        notifications.add(new EventNotification.fromServiceEvent(event));
+        e = new IsolateReloadEvent(event.timestamp, event.isolate, event.error);
         break;
-
       case ServiceEvent.kIsolateExit:
+        e = new IsolateExitEvent(event.timestamp, event.isolate);
+        break;
+      case ServiceEvent.kBreakpointAdded:
+        e = new BreakpointAddedEvent(event.timestamp, event.isolate,
+            event.breakpoint);
+        break;
+      case ServiceEvent.kBreakpointResolved:
+        e = new BreakpointResolvedEvent(event.timestamp, event.isolate,
+            event.breakpoint);
+        break;
+      case ServiceEvent.kBreakpointRemoved:
+        e = new BreakpointRemovedEvent(event.timestamp, event.isolate,
+          event.breakpoint);
+        break;
+      case ServiceEvent.kDebuggerSettingsUpdate:
+        e = new DebuggerSettingsUpdateEvent(event.timestamp, event.isolate);
+        break;
       case ServiceEvent.kResume:
-        notifications.deletePauseEvents(isolate: event.isolate);
+        e = new ResumeEvent(event.timestamp, event.isolate, event.topFrame);
         break;
-
       case ServiceEvent.kPauseStart:
+        e = new PauseStartEvent(event.timestamp, event.isolate);
+        break;
       case ServiceEvent.kPauseExit:
+        e = new PauseExitEvent(event.timestamp, event.isolate);
+        break;
       case ServiceEvent.kPauseBreakpoint:
+        e = new PauseBreakpointEvent(event.timestamp, event.isolate,
+            event.pauseBreakpoints, event.topFrame, event.atAsyncSuspension,
+            event.breakpoint);
+        break;
       case ServiceEvent.kPauseInterrupted:
+        e = new PauseInterruptedEvent(event.timestamp, event.isolate,
+            event.topFrame, event.atAsyncSuspension);
+        break;
       case ServiceEvent.kPauseException:
-        notifications.deletePauseEvents(isolate: event.isolate);
-        notifications.add(new EventNotification.fromServiceEvent(event));
+        e = new PauseExceptionEvent(event.timestamp, event.isolate,
+            event.topFrame, event.exception);
         break;
-
       case ServiceEvent.kInspect:
-        notifications.add(new EventNotification.fromServiceEvent(event));
+        e = new InspectEvent(event.timestamp, event.isolate,
+            event.inspectee);
         break;
-
       default:
         // Ignore unrecognized events.
         Logger.root.severe('Unrecognized event: $event');
-        break;
+        return;
     }
+    events.add(e);
   }
 
   void _registerPages() {
@@ -201,6 +238,23 @@
     });
     _setVM(new WebSocketVM(targets.current));
     _initOnce();
+
+    // delete pause events.
+    events.onIsolateExit.listen(_deletePauseEvents);
+    events.onResume.listen(_deletePauseEvents);
+    events.onPauseStart.listen(_deletePauseEvents);
+    events.onPauseExit.listen(_deletePauseEvents);
+    events.onPauseBreakpoint.listen(_deletePauseEvents);
+    events.onPauseInterrupted.listen(_deletePauseEvents);
+    events.onPauseException.listen(_deletePauseEvents);
+
+    // show notification for an event.
+    events.onIsolateReload.listen(_addNotification);
+    events.onPauseExit.listen(_addNotification);
+    events.onPauseBreakpoint.listen(_addNotification);
+    events.onPauseInterrupted.listen(_addNotification);
+    events.onPauseException.listen(_addNotification);
+    events.onInspect.listen(_addNotification);
   }
 
   loadCrashDump(Map crashDump) {

runtime/observatory/lib/src/app/event.dart

diff --git a/runtime/observatory/lib/src/app/event.dart b/runtime/observatory/lib/src/app/event.dart
new file mode 100644
index 0000000..9e6f9e9
--- /dev/null
+++ b/runtime/observatory/lib/src/app/event.dart
@@ -0,0 +1,254 @@
+// Copyright (c) 2016, the Dart project authors.  Please see the AUTHORS file
+// for details. All rights reserved. Use of this source code is governed by a
+// BSD-style license that can be found in the LICENSE file.
+
+part of app;
+
+class VMUpdateEvent implements M.VMUpdateEvent {
+  final DateTime timestamp;
+  final M.VMRef vm;
+  VMUpdateEvent(this.timestamp, this.vm) {
+    assert(timestamp != null);
+    assert(vm != null);
+  }
+}
+
+class IsolateStartEvent implements M.IsolateStartEvent {
+  final DateTime timestamp;
+  final M.IsolateRef isolate;
+  IsolateStartEvent(this.timestamp, this.isolate) {
+    assert(timestamp != null);
+    assert(isolate != null);
+  }
+}
+
+class IsolateRunnableEvent implements M.IsolateRunnableEvent {
+  final DateTime timestamp;
+  final M.IsolateRef isolate;
+  IsolateRunnableEvent(this.timestamp, this.isolate) {
+    assert(timestamp != null);
+    assert(isolate != null);
+  }
+}
+
+class IsolateExitEvent implements M.IsolateExitEvent {
+  final DateTime timestamp;
+  final M.IsolateRef isolate;
+  IsolateExitEvent(this.timestamp, this.isolate) {
+    assert(timestamp != null);
+    assert(isolate != null);
+  }
+}
+
+class IsolateUpdateEvent implements M.IsolateUpdateEvent {
+  final DateTime timestamp;
+  final M.IsolateRef isolate;
+  IsolateUpdateEvent(this.timestamp, this.isolate) {
+    assert(timestamp != null);
+    assert(isolate != null);
+  }
+}
+
+class IsolateReloadEvent implements M.IsolateReloadEvent {
+  final DateTime timestamp;
+  final M.IsolateRef isolate;
+  final M.ErrorRef error;
+  IsolateReloadEvent(this.timestamp, this.isolate, this.error) {
+    assert(timestamp != null);
+    assert(isolate != null);
+    assert(error != null);
+  }
+}
+
+class ServiceExtensionAddedEvent implements M.ServiceExtensionAddedEvent {
+  final DateTime timestamp;
+  final M.IsolateRef isolate;
+  final String extensionRPC;
+  ServiceExtensionAddedEvent(this.timestamp, this.isolate, this.extensionRPC) {
+    assert(timestamp != null);
+    assert(isolate != null);
+    assert(extensionRPC != null);
+  }
+}
+
+class DebuggerSettingsUpdateEvent implements M.DebuggerSettingsUpdateEvent {
+  final DateTime timestamp;
+  final M.IsolateRef isolate;
+  DebuggerSettingsUpdateEvent(this.timestamp, this.isolate) {
+    assert(timestamp != null);
+    assert(isolate != null);
+  }
+}
+
+class PauseStartEvent implements M.PauseStartEvent {
+  final DateTime timestamp;
+  final M.IsolateRef isolate;
+  PauseStartEvent(this.timestamp, this.isolate) {
+    assert(timestamp != null);
+    assert(isolate != null);
+  }
+}
+
+class PauseExitEvent implements M.PauseExitEvent {
+  final DateTime timestamp;
+  final M.IsolateRef isolate;
+  PauseExitEvent(this.timestamp, this.isolate) {
+    assert(timestamp != null);
+    assert(isolate != null);
+  }
+}
+
+class PauseBreakpointEvent implements M.PauseBreakpointEvent {
+  final DateTime timestamp;
+  final M.IsolateRef isolate;
+  final Iterable<M.Breakpoint> pauseBreakpoints;
+  final M.Frame topFrame;
+  final bool atAsyncSuspension;
+  /// [optional]
+  final M.Breakpoint breakpoint;
+  PauseBreakpointEvent(this.timestamp, this.isolate,
+      Iterable<M.Breakpoint> pauseBreakpoints, this.topFrame,
+      this.atAsyncSuspension, [this.breakpoint])
+    : pauseBreakpoints = new List.unmodifiable(pauseBreakpoints){
+    assert(timestamp != null);
+    assert(isolate != null);
+    assert(pauseBreakpoints != null);
+    assert(topFrame != null);
+    assert(atAsyncSuspension != null);
+  }
+}
+
+class PauseInterruptedEvent implements M.PauseInterruptedEvent {
+  final DateTime timestamp;
+  final M.IsolateRef isolate;
+  final M.Frame topFrame;
+  final bool atAsyncSuspension;
+  PauseInterruptedEvent(this.timestamp, this.isolate, this.topFrame,
+      this.atAsyncSuspension) {
+    assert(timestamp != null);
+    assert(isolate != null);
+    assert(atAsyncSuspension != null);
+  }
+}
+
+class PauseExceptionEvent implements M.PauseExceptionEvent {
+  final DateTime timestamp;
+  final M.IsolateRef isolate;
+  final M.Frame topFrame;
+  final M.InstanceRef exception;
+  PauseExceptionEvent(this.timestamp, this.isolate, this.topFrame,
+      this.exception) {
+    assert(timestamp != null);
+    assert(isolate != null);
+    assert(topFrame != null);
+    assert(exception != null);
+  }
+}
+
+class ResumeEvent implements M.ResumeEvent {
+  final DateTime timestamp;
+  final M.IsolateRef isolate;
+  final M.Frame topFrame;
+  ResumeEvent(this.timestamp, this.isolate, this.topFrame) {
+    assert(timestamp != null);
+    assert(isolate != null);
+  }
+}
+
+class BreakpointAddedEvent implements M.BreakpointAddedEvent {
+  final DateTime timestamp;
+  final M.IsolateRef isolate;
+  final M.Breakpoint breakpoint;
+  BreakpointAddedEvent(this.timestamp, this.isolate, this.breakpoint) {
+    assert(timestamp != null);
+    assert(isolate != null);
+    assert(breakpoint != null);
+  }
+}
+
+class BreakpointResolvedEvent implements M.BreakpointResolvedEvent {
+  final DateTime timestamp;
+  final M.IsolateRef isolate;
+  final M.Breakpoint breakpoint;
+  BreakpointResolvedEvent(this.timestamp, this.isolate, this.breakpoint) {
+    assert(timestamp != null);
+    assert(isolate != null);
+    assert(breakpoint != null);
+  }
+}
+
+class BreakpointRemovedEvent implements M.BreakpointRemovedEvent {
+  final DateTime timestamp;
+  final M.IsolateRef isolate;
+  final M.Breakpoint breakpoint;
+  BreakpointRemovedEvent(this.timestamp, this.isolate, this.breakpoint) {
+    assert(timestamp != null);
+    assert(isolate != null);
+    assert(breakpoint != null);
+  }
+}
+
+class InspectEvent implements M.InspectEvent {
+  final DateTime timestamp;
+  final M.IsolateRef isolate;
+  final M.InstanceRef inspectee;
+  InspectEvent(this.timestamp, this.isolate, this.inspectee) {
+    assert(timestamp != null);
+    assert(isolate != null);
+    assert(inspectee != null);
+  }
+}
+
+class NoneEvent implements M.NoneEvent {
+  final DateTime timestamp;
+  final M.IsolateRef isolate;
+  NoneEvent(this.timestamp, this.isolate) {
+    assert(timestamp != null);
+    assert(isolate != null);
+  }
+}
+
+class GCEvent implements M.GCEvent {
+  final DateTime timestamp;
+  final M.IsolateRef isolate;
+  GCEvent(this.timestamp, this.isolate) {
+    assert(timestamp != null);
+    assert(isolate != null);
+  }
+}
+
+class ExtensionEvent implements M.ExtensionEvent {
+  final DateTime timestamp;
+  final M.IsolateRef isolate;
+  final String extensionKind;
+  final M.ExtensionData extensionData;
+  ExtensionEvent(this.timestamp, this.isolate, this.extensionKind,
+      this.extensionData) {
+    assert(timestamp != null);
+    assert(isolate != null);
+    assert(extensionKind != null);
+    assert(extensionData != null);
+  }
+}
+
+class TimelineEventsEvent implements M.TimelineEventsEvent {
+  final DateTime timestamp;
+  final M.IsolateRef isolate;
+  final Iterable<M.TimelineEvent> timelineEvents;
+  TimelineEventsEvent(this.timestamp, this.isolate,
+      Iterable<M.TimelineEvent> timelineEvents)
+    : timelineEvents = new List.unmodifiable(timelineEvents){
+    assert(timestamp != null);
+    assert(isolate != null);
+    assert(timelineEvents != null);
+  }
+}
+
+class ConnectionClosedEvent implements M.ConnectionClosedEvent {
+  final DateTime timestamp;
+  final String reason;
+  ConnectionClosedEvent(this.timestamp, this.reason) {
+    assert(timestamp != null);
+    assert(reason != null);
+  }
+}

runtime/observatory/lib/src/app/notification.dart

diff --git a/runtime/observatory/lib/src/app/notification.dart b/runtime/observatory/lib/src/app/notification.dart
index b95c34e..140e0dc 100644
--- a/runtime/observatory/lib/src/app/notification.dart
+++ b/runtime/observatory/lib/src/app/notification.dart
@@ -14,102 +14,4 @@
 class EventNotification implements M.EventNotification {
   final M.Event event;
   EventNotification(this.event);
-  factory EventNotification.fromServiceEvent(ServiceEvent event) {
-    M.Event e;
-    switch(event.kind) {
-      case ServiceEvent.kVMUpdate:
-        e = new VMUpdateEventMock(timestamp: event.timestamp, vm: event.vm);
-        break;
-      case ServiceEvent.kIsolateStart:
-        e = new IsolateStartEventMock(timestamp: event.timestamp,
-            isolate: event.isolate);
-        break;
-      case ServiceEvent.kIsolateRunnable:
-        e = new IsolateRunnableEventMock(timestamp: event.timestamp,
-            isolate: event.isolate);
-        break;
-      case ServiceEvent.kIsolateExit:
-        e = new IsolateExitEventMock(timestamp: event.timestamp,
-            isolate: event.isolate);
-        break;
-      case ServiceEvent.kIsolateUpdate:
-        e = new IsolateUpdateEventMock(timestamp: event.timestamp,
-            isolate: event.isolate);
-        break;
-      case ServiceEvent.kIsolateReload:
-        // TODO(bernaschina) add error: realoadError.
-        e = new IsolateRealodEventMock(timestamp: event.timestamp,
-            isolate: event.isolate);
-        break;
-      case ServiceEvent.kServiceExtensionAdded:
-        e = new ServiceExtensionAddedEventMock(timestamp: event.timestamp,
-            isolate: event.isolate, extensionRPC: event.extensionRPC);
-        break;
-      case ServiceEvent.kPauseStart:
-        e = new PauseStartEventMock(timestamp: event.timestamp,
-            isolate: event.isolate);
-        break;
-      case ServiceEvent.kPauseExit:
-        e = new PauseExitEventMock(timestamp: event.timestamp,
-            isolate: event.isolate);
-        break;
-      case ServiceEvent.kPauseBreakpoint:
-        // TODO(cbernaschina) add pauseBreakpoints.
-        e = new PauseBreakpointEventMock(timestamp: event.timestamp,
-            isolate: event.isolate, breakpoint: event.breakpoint,
-            pauseBreakpoints: const <M.Breakpoint>[],
-            topFrame: event.topFrame,
-            atAsyncSuspension: event.atAsyncSuspension);
-        break;
-      case ServiceEvent.kPauseInterrupted:
-        e = new PauseInterruptedEventMock(timestamp: event.timestamp,
-            isolate: event.isolate, topFrame: event.topFrame,
-            atAsyncSuspension: event.atAsyncSuspension);
-        break;
-      case ServiceEvent.kPauseException:
-        // TODO(cbernaschina) add exception.
-        e = new PauseExceptionEventMock(timestamp: event.timestamp,
-            isolate: event.isolate, topFrame: event.topFrame);
-        break;
-      case ServiceEvent.kNone:
-        e = new NoneEventMock(timestamp: event.timestamp,
-            isolate: event.isolate);
-        break;
-      case ServiceEvent.kResume:
-        e = new ResumeEventMock(timestamp: event.timestamp,
-            isolate: event.isolate);
-        break;
-      case ServiceEvent.kBreakpointAdded:
-        e = new BreakpointAddedEventMock(timestamp: event.timestamp,
-            isolate: event.isolate, breakpoint: event.breakpoint);
-        break;
-      case ServiceEvent.kBreakpointResolved:
-        e = new BreakpointResolvedEventMock(timestamp: event.timestamp,
-            isolate: event.isolate, breakpoint: event.breakpoint);
-        break;
-      case ServiceEvent.kBreakpointRemoved:
-        e = new BreakpointRemovedEventMock(timestamp: event.timestamp,
-            isolate: event.isolate, breakpoint: event.breakpoint);
-        break;
-      case ServiceEvent.kGC:
-        e = new GCEventMock(timestamp: event.timestamp,
-            isolate: event.isolate);
-        break;
-      case ServiceEvent.kInspect:
-       // TODO(cbernaschina) add inspectee: event.inspectee.
-        e = new InspectEventMock(timestamp: event.timestamp,
-            isolate: event.isolate);
-        break;
-      case ServiceEvent.kConnectionClosed:
-        e = new ConnectionClockedEventMock(timestamp: event.timestamp,
-            reason: event.reason);
-        break;
-      case ServiceEvent.kExtension:
-        e = new ExtensionEventMock(timestamp: event.timestamp,
-            isolate: event.isolate, extensionKind: event.extensionKind,
-            extensionData: event.extensionData);
-        break;
-    }
-    return new EventNotification(e);
-  }
 }

runtime/observatory/lib/src/app/page.dart

diff --git a/runtime/observatory/lib/src/app/page.dart b/runtime/observatory/lib/src/app/page.dart
index 980fcfa..9d624c1 100644
--- a/runtime/observatory/lib/src/app/page.dart
+++ b/runtime/observatory/lib/src/app/page.dart
@@ -124,9 +124,9 @@
   @override
   onInstall() {
     element = new FlagListElement(app.vm,
-        app.vm.changes.map((_) => new VMUpdateEventMock(vm: app.vm)),
-        new FlagsRepository(),
-        app.notifications);
+                                  app.events,
+                                  new FlagsRepository(app.vm),
+                                  app.notifications);
   }
 
   void _visit(Uri uri) {
@@ -162,14 +162,23 @@
 class ClassTreePage extends SimplePage {
   ClassTreePage(app) : super('class-tree', 'class-tree', app);
 
+  final DivElement container = new DivElement();
+
+  @override
+  void onInstall() {
+    element = container;
+  }
+
   void _visit(Uri uri) {
     super._visit(uri);
     getIsolate(uri).then((isolate) {
-      if (element != null) {
-        /// Update the page.
-        ClassTreeElement page = element;
-        page.isolate = isolate;
-      }
+      container.children = [
+        new ClassTreeElement(app.vm,
+                             isolate,
+                             app.events,
+                             app.notifications,
+                             new ClassRepository(isolate))
+      ];
     });
   }
 }
@@ -362,8 +371,7 @@
     if (element == null) {
       element = new VMConnectElement(
             ObservatoryApplication.app.targets,
-            new CrashDumpRepositoryMock(
-                load: ObservatoryApplication.app.loadCrashDump),
+            ObservatoryApplication.app.loadCrashDump,
             ObservatoryApplication.app.notifications,
             queue: ObservatoryApplication.app.queue);
     }

runtime/observatory/lib/src/elements/class_tree.dart

diff --git a/runtime/observatory/lib/src/elements/class_tree.dart b/runtime/observatory/lib/src/elements/class_tree.dart
index 16f6928..f17f27b 100644
--- a/runtime/observatory/lib/src/elements/class_tree.dart
+++ b/runtime/observatory/lib/src/elements/class_tree.dart
@@ -4,150 +4,182 @@
 
 library class_tree_element;
 
-import 'observatory_element.dart';
-import 'dart:async';
 import 'dart:html';
-import 'package:logging/logging.dart';
-import 'package:observatory/app.dart';
-import 'package:observatory/service.dart';
-import 'package:polymer/polymer.dart';
+import 'dart:async';
+import 'package:observatory/models.dart' as M;
+import 'package:observatory/src/elements/class_ref.dart';
+import 'package:observatory/src/elements/containers/virtual_tree.dart';
+import 'package:observatory/src/elements/helpers/rendering_scheduler.dart';
+import 'package:observatory/src/elements/helpers/tag.dart';
+import 'package:observatory/src/elements/helpers/uris.dart';
+import 'package:observatory/src/elements/nav/bar.dart';
+import 'package:observatory/src/elements/nav/isolate_menu.dart';
+import 'package:observatory/src/elements/nav/menu.dart';
+import 'package:observatory/src/elements/nav/notify.dart';
+import 'package:observatory/src/elements/nav/top_menu.dart';
+import 'package:observatory/src/elements/nav/vm_menu.dart';
 
-class ClassTreeRow extends TableTreeRow {
-  @reflectable final Isolate isolate;
-  @reflectable final Class cls;
-  ClassTreeRow(this.isolate, this.cls, TableTree tree, ClassTreeRow parent)
-      : super(tree, parent) {
+
+class ClassTreeElement extends HtmlElement implements Renderable{
+  static const tag = const Tag<ClassTreeElement>('class-tree',
+                     dependencies: const [ClassRefElement.tag,
+                                          NavBarElement.tag,
+                                          NavIsolateMenuElement.tag,
+                                          NavMenuElement.tag,
+                                          NavNotifyElement.tag,
+                                          NavTopMenuElement.tag,
+                                          NavVMMenuElement.tag,
+                                          VirtualTreeElement.tag]);
+
+  RenderingScheduler _r;
+
+  Stream<RenderedEvent<ClassTreeElement>> get onRendered => _r.onRendered;
+
+  M.VMRef _vm;
+  M.IsolateRef _isolate;
+  M.EventRepository _events;
+  M.NotificationRepository _notifications;
+  M.ClassRepository _classes;
+  M.Class _object;
+  final _subclasses = <String, Iterable<M.Class>>{};
+  final _mixins = <String, List<M.Instance>>{};
+
+  factory ClassTreeElement(M.VMRef vm, M.IsolateRef isolate,
+                           M.EventRepository events,
+                           M.NotificationRepository notifications,
+                           M.ClassRepository classes,
+                           {RenderingQueue queue}) {
+    assert(vm != null);
     assert(isolate != null);
-    assert(cls != null);
+    assert(events != null);
+    assert(notifications != null);
+    assert(classes != null);
+    ClassTreeElement e = document.createElement(tag.name);
+    e._r = new RenderingScheduler(e, queue: queue);
+    e._vm = vm;
+    e._isolate = isolate;
+    e._events = events;
+    e._notifications = notifications;
+    e._classes = classes;
+    return e;
   }
 
-  void _addChildren(List<Class> subclasses) {
-    for (var subclass in subclasses) {
-      if (subclass.isPatch) {
-        continue;
-      }
-      if (subclass.mixin != null) {
-        _addChildren(subclass.subclasses);
-      } else {
-        var row = new ClassTreeRow(isolate, subclass, tree, this);
-        children.add(row);
-      }
-    }
-  }
-
-  Future _addMixins(Class cls) async {
-    var classCell = flexColumns[0];
-    if (cls.superclass == null) {
-      return;
-    }
-    bool first = true;
-    while (cls.superclass != null && cls.superclass.mixin != null) {
-      cls = cls.superclass;
-      await cls.mixin.load();
-      var span = new SpanElement();
-      span.style.alignSelf = 'center';
-      span.style.whiteSpace = 'pre';
-      if (first) {
-        span.text = ' with ';
-      } else {
-        span.text = ', ';
-      }
-      classCell.children.add(span);
-      var mixinRef = new Element.tag('class-ref');
-      mixinRef.ref = cls.mixin.typeClass;
-      mixinRef.style.alignSelf = 'center';
-      classCell.children.add(mixinRef);
-      first = false;
-    }
-  }
-
-  Future _addClass(Class cls) async {
-    var classCell = flexColumns[0];
-    classCell.style.justifyContent = 'flex-start';
-    var classRef = new Element.tag('class-ref');
-    classRef.ref = cls;
-    classRef.style.alignSelf = 'center';
-    classCell.children.add(classRef);
-    if (cls.superclass != null && cls.superclass.mixin != null) {
-      await _addMixins(cls);
-    }
-    if (cls.subclasses.isNotEmpty) {
-      var span = new SpanElement();
-      span.style.paddingLeft = '.5em';
-      span.style.alignSelf = 'center';
-      int subclassCount = _indirectSubclassCount(cls) - 1;
-      if (subclassCount > 1) {
-        span.text = '($subclassCount subclasses)';
-      } else {
-        span.text = '($subclassCount subclass)';
-      }
-      classCell.children.add(span);
-    }
-  }
-
-  void onShow() {
-    super.onShow();
-    if (children.length == 0) {
-      _addChildren(cls.subclasses);
-    }
-    _addClass(cls);
-  }
-
-  static int _indirectSubclassCount(var cls) {
-    int count = 0;
-    if (cls.mixin == null) {
-      // Don't count synthetic mixin classes in subclass count.
-      count++;
-    }
-    for (var subclass in cls.subclasses) {
-      count += _indirectSubclassCount(subclass);
-    }
-    return count;
-  }
-
-  bool hasChildren() {
-    return cls.subclasses.isNotEmpty;
-  }
-}
-
-
-@CustomTag('class-tree')
-class ClassTreeElement extends ObservatoryElement {
-  @observable Isolate isolate;
-
-  TableTree tree;
-
   ClassTreeElement.created() : super.created();
 
   @override
   void attached() {
     super.attached();
-    var tableBody = shadowRoot.querySelector('#tableTreeBody');
-    assert(tableBody != null);
-    tree = new TableTree(tableBody, 1);
-    if (isolate != null) {
-      _update(isolate.objectClass);
+    _refresh();
+    _r.enable();
+  }
+
+  @override
+  void detached() {
+    super.detached();
+    children = [];
+    _r.disable(notify: true);
+  }
+
+  VirtualTreeElement _tree;
+
+  void render() {
+    children = [
+      new NavBarElement(queue: _r.queue)
+        ..children = [
+          new NavTopMenuElement(queue: _r.queue),
+          new NavVMMenuElement(_vm, _events, queue: _r.queue),
+          new NavIsolateMenuElement(_isolate, _events, queue: _r.queue),
+          new NavMenuElement('class hierarchy', link: Uris.classTree(_isolate),
+                             last: true, queue: _r.queue),
+          new NavNotifyElement(_notifications, queue: _r.queue)
+        ],
+      new DivElement()
+        ..classes = ['content-centered']
+        ..children = [
+          new HeadingElement.h1()..text = 'Class Hierarchy',
+          new BRElement(), new HRElement(),
+          _object == null ? (new HeadingElement.h2()..text = 'Loading...')
+                          : _createTree()
+        ]
+    ];
+  }
+
+  Element _createTree() {
+    _tree = new VirtualTreeElement(_create, _update, _children,
+                                   items: [_object], queue: _r.queue);
+    _tree.expand(_object, autoExpandSingleChildNodes: true);
+    return _tree;
+  }
+
+  Future _refresh() async {
+    _object = null;
+    _subclasses.clear();
+    _mixins.clear();
+    _object = await _register(await _classes.getObject());
+    _r.dirty();
+  }
+
+  Future<M.Class> _register(M.Class cls) async {
+    _subclasses[cls.id] = await Future.wait(
+      (await Future.wait(cls.subclasses.map(_getActualChildrens)))
+      .expand((f) => f)
+      .map(_register)
+    );
+    return cls;
+  }
+
+  Future<Iterable<M.Class>> _getActualChildrens(M.ClassRef ref) async {
+    var cls = await _classes.get(ref.id);
+    if (cls.isPatch) {
+      return const [];
+    }
+    if (cls.mixin == null) {
+      return [cls];
+    }
+    return (await Future.wait(cls.subclasses.map(_getActualChildrens)))
+      .expand((f) => f)
+      ..forEach((subcls) {
+        _mixins[subcls.id] = (_mixins[subcls.id] ?? [])..add(cls.mixin);
+      });
+  }
+
+  static Element _create(toggle) {
+    return new DivElement()..classes = ['class-tree-item']
+        ..children = [
+          new SpanElement()..classes = ['lines'],
+          new SpanElement()..classes = ['expander']
+            ..onClick.listen((_) => toggle(autoToggleSingleChildNodes: true)),
+          new SpanElement()..classes = ['name']
+        ];
+  }
+
+  void _update(HtmlElement el, M.Class cls, int index) {
+    virtualTreeUpdateLines(el.children[0], index);
+    if (cls.subclasses.isEmpty) {
+      el.children[1].text = '';
+    } else {
+      el.children[1].text = _tree.isExpanded(cls) ? '▼' : '►';
+    }
+    el.children[2].children = [
+      new ClassRefElement(_isolate, cls, queue: _r.queue)
+    ];
+    if (_mixins[cls.id] != null) {
+      el.children[2].children.addAll(_createMixins(_mixins[cls.id]));
     }
   }
 
-  isolateChanged(oldValue) {
-    isolate.getClassHierarchy().then((objectClass) {
-      _update(objectClass);
-    });
+  List<Element> _createMixins(List<M.Instance> types) {
+    final  children = types.expand((type) => [
+      new SpanElement()..text = ', ',
+      type.typeClass == null
+          ? (new SpanElement()..text = type.name.split('<').first)
+          : new ClassRefElement(_isolate, type.typeClass, queue: _r.queue)
+    ]).toList();
+    children.first.text = ' with ';
+    return children;
   }
 
-  void _update(Class root) {
-    try {
-      var rootRow = new ClassTreeRow(isolate, root, tree, null);
-      rootRow.children.add(new ClassTreeRow(isolate, root, tree, rootRow));
-      tree.initialize(rootRow);
-    } catch (e, stackTrace) {
-      Logger.root.warning('_update', e, stackTrace);
-    }
-    // Check if we only have one node at the root and expand it.
-    if (tree.rows.length == 1) {
-      tree.toggle(tree.rows[0]);
-    }
-    notifyPropertyChange(#tree, null, tree);
+  Iterable<M.Class> _children(M.Class cls) {
+    return _subclasses[cls.id];
   }
 }

runtime/observatory/lib/src/elements/class_tree.html

diff --git a/runtime/observatory/lib/src/elements/class_tree.html b/runtime/observatory/lib/src/elements/class_tree.html
deleted file mode 100644
index e2fbb32..0000000
--- a/runtime/observatory/lib/src/elements/class_tree.html
+++ /dev/null
@@ -1,42 +0,0 @@
-<link rel="import" href="../../../../packages/polymer/polymer.html">
-
-<polymer-element name="class-tree">
-  <template>
-    <link rel="stylesheet" href="css/shared.css">
-    <style>
-      .table {
-        border-spacing: 0px;
-        width: 100%;
-        margin-bottom: 20px
-        vertical-align: middle;
-      }
-
-      tr {
-        background-color: #FFFFFF;
-      }
-
-      tr:hover {
-        background-color: #FAFAFA;
-      }
-
-      th {
-        text-align: left;
-      }
-    </style>
-    <nav-bar>
-      <top-nav-menu></top-nav-menu>
-      <vm-nav-menu vm="{{ isolate.vm }}"></vm-nav-menu>
-      <isolate-nav-menu isolate="{{ isolate }}"></isolate-nav-menu>
-      <nav-menu link="{{ makeLink('/class-tree', isolate) }}" anchor="class hierarchy" last="{{ true }}"></nav-menu>
-      <nav-notify notifications="{{ app.notifications }}"></nav-notify>
-    </nav-bar>
-    <div class="content-centered">
-      <h1>Class Hierarchy</h1>
-      <table id="tableTree" class="table">
-        <tbody id="tableTreeBody"></tbody>
-      </table>
-    </div>
-  </template>
-</polymer-element>
-
-<script type="application/dart" src="class_tree.dart"></script>

runtime/observatory/lib/src/elements/code_ref.dart

diff --git a/runtime/observatory/lib/src/elements/code_ref.dart b/runtime/observatory/lib/src/elements/code_ref.dart
index bead036..61917a8 100644
--- a/runtime/observatory/lib/src/elements/code_ref.dart
+++ b/runtime/observatory/lib/src/elements/code_ref.dart
@@ -47,8 +47,8 @@
   @override
   void detached() {
     super.detached();
-    _r.disable(notify: true);
     children = [];
+    _r.disable(notify: true);
   }
 
   void render() {

runtime/observatory/lib/src/elements/code_ref_wrapper.dart

diff --git a/runtime/observatory/lib/src/elements/code_ref_wrapper.dart b/runtime/observatory/lib/src/elements/code_ref_wrapper.dart
index 503a784..f630f33 100644
--- a/runtime/observatory/lib/src/elements/code_ref_wrapper.dart
+++ b/runtime/observatory/lib/src/elements/code_ref_wrapper.dart
@@ -20,8 +20,13 @@
   static const tag = const Tag<CodeRefElementWrapper>('code-ref');
 
   Code _code;
+
   Code get ref => _code;
-  void set ref(Code ref) { _code = ref; render(); }
+
+  void set ref(Code value) {
+    _code = value;
+    render();
+  }
 
   CodeRefElementWrapper.created() : super.created() {
     binder.registerCallback(this);
@@ -37,7 +42,9 @@
 
   void render() {
     shadowRoot.children = [];
-    if (ref == null) return;
+    if (ref == null) {
+      return;
+    }
 
     shadowRoot.children = [
       new StyleElement()

runtime/observatory/lib/src/elements/containers/virtual_collection.dart

diff --git a/runtime/observatory/lib/src/elements/containers/virtual_collection.dart b/runtime/observatory/lib/src/elements/containers/virtual_collection.dart
index 36e682c..e45a143 100644
--- a/runtime/observatory/lib/src/elements/containers/virtual_collection.dart
+++ b/runtime/observatory/lib/src/elements/containers/virtual_collection.dart
@@ -24,7 +24,7 @@
   VirtualCollectionUpdateCallback _update;
   double _itemHeight;
   int _top;
-  int _height;
+  double _height;
   List _items;
   StreamSubscription _onScrollSubscription;
   StreamSubscription _onResizeSubscription;

runtime/observatory/lib/src/elements/containers/virtual_tree.dart

diff --git a/runtime/observatory/lib/src/elements/containers/virtual_tree.dart b/runtime/observatory/lib/src/elements/containers/virtual_tree.dart
index 78ea96b..4df0dd7 100644
--- a/runtime/observatory/lib/src/elements/containers/virtual_tree.dart
+++ b/runtime/observatory/lib/src/elements/containers/virtual_tree.dart
@@ -4,6 +4,7 @@
 
 import 'dart:async';
 import 'dart:html';
+import 'dart:math' as Math;
 import 'package:observatory/src/elements/containers/virtual_collection.dart';
 import 'package:observatory/src/elements/helpers/rendering_scheduler.dart';
 import 'package:observatory/src/elements/helpers/tag.dart';
@@ -13,6 +14,16 @@
 typedef void VirtualTreeUpdateCallback(HtmlElement el, dynamic item, int depth);
 typedef Iterable<dynamic> VritualTreeGetChildrenCallback(dynamic value);
 
+void virtualTreeUpdateLines(SpanElement element, int n) {
+  n = Math.max(0, n);
+  while (element.children.length > n) {
+    element.children.removeLast();
+  }
+  while (element.children.length < n) {
+    element.children.add(new SpanElement());
+  }
+}
+
 class VirtualTreeElement extends HtmlElement implements Renderable {
   static const tag =
       const Tag<VirtualTreeElement>('virtual-tree', dependencies: const [

runtime/observatory/lib/src/elements/css/shared.css

diff --git a/runtime/observatory/lib/src/elements/css/shared.css b/runtime/observatory/lib/src/elements/css/shared.css
index bd15c93..e468fc8 100644
--- a/runtime/observatory/lib/src/elements/css/shared.css
+++ b/runtime/observatory/lib/src/elements/css/shared.css
@@ -388,6 +388,34 @@
   text-decoration: none;
 }
 
+/* class-tree */
+
+class-tree {
+  position: relative;
+  display: block;
+  height: 100%;
+}
+
+class-tree virtual-tree {
+  position: absolute;
+  height: auto;
+  top: 60px;
+  bottom: 0;
+  left: 0;
+  right: 0;
+}
+
+class-tree virtual-tree .class-tree-item {
+  line-height: 25px;
+  height: 25px;
+  padding-left: 10%;
+  padding-right: 10%;
+}
+
+class-tree virtual-tree .class-tree-item .name {
+  margin-left: 0.5em;
+}
+
 /* error-ref */
 /* TODO(cbernaschina) fix error-ref-wrapped to error-ref when wrapper
 removed */
@@ -542,7 +570,8 @@
   padding: 8px;
   border-width: 2px;
   line-height: 13px;
-  font: 400 13px 'Montserrat', sans-serif;
+  font-size: 13px;
+  font: 400 'Montserrat', sans-serif;
 }
 nav-refresh-wrapped > li > button[disabled] {
   color: #aaa;
@@ -553,6 +582,18 @@
   margin: 0;
 }
 
+/* observatory-application */
+
+observatory-application {
+  display: block;
+  height: 100%;
+}
+
+observatory-application > div {
+  display: block;
+  height: 100%;
+}
+
 /* script-ref */
 /* TODO(cbernaschina) fix script-ref-wrapped to script-ref when wrapper
 removed */
@@ -622,8 +663,64 @@
 }
 
 virtual-collection .shifter > * {
+  display: block;
   position: relative;
   top: -25%;
+  width: 100%;
+  white-space: nowrap;
+  text-overflow: ellipsis;
+  overflow-x: hidden;
+  white-space: nowrap;
+}
+
+/* virtual-tree */
+
+virtual-tree .expander {
+  display: inline-block;
+  text-align: center;
+  font-weight: bold;
+  font-size: 18px;
+  width: 19px;
+}
+
+virtual-tree .lines,
+virtual-tree .lines > span {
+  display: inline-block;
+  vertical-align: top;
+  height: 100%;
+  line-height: 100%;
+}
+
+virtual-tree .lines:before,
+virtual-tree .lines > span:before {
+  display: inline-block;
+  content: '';
+}
+
+virtual-tree .lines > span {
+  width: 3px;
+  margin-left: 8px;
+  margin-right: 8px;
+}
+
+virtual-tree .lines > span:nth-child(5n) {
+  background-color: #673AB7;
+}
+
+virtual-tree .lines > span:nth-child(5n+1) {
+  background-color: #F44336;
+}
+
+virtual-tree .lines > span:nth-child(5n+2) {
+  background-color: #4CAF50;
+}
+
+virtual-tree .lines > span:nth-child(5n+3) {
+  background-color: #3F51B5;
+}
+
+virtual-tree .lines > span:nth-child(5n+4) {
+  background-color: #FF9800;
 }
 
 /* vm-connect-target */
@@ -638,3 +735,9 @@
 vm-connect-target > button.delete-button:hover {
   background: #ff0000;
 }
+
+/* vm-connect */
+
+vm-connect ul {
+  list-style-type: none;
+}

runtime/observatory/lib/src/elements/error_ref.dart

diff --git a/runtime/observatory/lib/src/elements/error_ref.dart b/runtime/observatory/lib/src/elements/error_ref.dart
index b30dd90..55685ed 100644
--- a/runtime/observatory/lib/src/elements/error_ref.dart
+++ b/runtime/observatory/lib/src/elements/error_ref.dart
@@ -19,6 +19,7 @@
   Stream<RenderedEvent<ErrorRefElement>> get onRendered => _r.onRendered;
 
   ErrorRef _error;
+  
   ErrorRef get error => _error;
 
   factory ErrorRefElement(ErrorRef error, {RenderingQueue queue}) {

runtime/observatory/lib/src/elements/error_ref_wrapper.dart

diff --git a/runtime/observatory/lib/src/elements/error_ref_wrapper.dart b/runtime/observatory/lib/src/elements/error_ref_wrapper.dart
index 4efe939..50f8b9a 100644
--- a/runtime/observatory/lib/src/elements/error_ref_wrapper.dart
+++ b/runtime/observatory/lib/src/elements/error_ref_wrapper.dart
@@ -5,8 +5,7 @@
 import 'dart:html';
 
 import 'package:observatory/app.dart';
-import 'package:observatory/mocks.dart' show ErrorRefMock;
-import 'package:observatory/service_html.dart' show ServiceMap, DartError;
+import 'package:observatory/service.dart';
 import 'package:observatory/src/elements/error_ref.dart';
 import 'package:observatory/src/elements/helpers/tag.dart';
 import 'package:observatory/src/elements/shims/binding.dart';
@@ -19,9 +18,14 @@
 
   static const tag = const Tag<ErrorRefElementWrapper>('error-ref');
 
-  ServiceMap _error;
-  ServiceMap get ref => _error;
-  void set ref(ServiceMap ref) { _error = ref; render(); }
+  DartError _error;
+
+  DartError get ref => _error;
+
+  void set ref(DartError value) {
+    _error = value;
+    render();
+  }
 
   ErrorRefElementWrapper.created() : super.created() {
     binder.registerCallback(this);
@@ -37,42 +41,24 @@
 
   void render() {
     shadowRoot.children = [];
-    if (_error == null) return;
-
-    if (ref is Map) {
-      shadowRoot.children = [
-        new StyleElement()
-          ..text = '''
-          error-ref-wrapped > pre {
-            background-color: #f5f5f5;
-            border: 1px solid #ccc;
-            padding: 10px;
-            font-family: consolas, courier, monospace;
-            font-size: 1em;
-            line-height: 1.2em;
-            white-space: pre;
-          }
-          ''',
-        new ErrorRefElement(new ErrorRefMock(message: ref['message']))
-      ];
-    } else {
-      shadowRoot.children = [
-        new StyleElement()
-          ..text = '''
-          error-ref-wrapped > pre {
-            background-color: #f5f5f5;
-            border: 1px solid #ccc;
-            padding: 10px;
-            font-family: consolas, courier, monospace;
-            font-size: 1em;
-            line-height: 1.2em;
-            white-space: pre;
-          }
-          ''',
-        new ErrorRefElement(
-                        new ErrorRefMock(message: (ref as DartError).message),
-                        queue: ObservatoryApplication.app.queue)
-      ];
+    if (_error == null) {
+      return;
     }
+
+    shadowRoot.children = [
+      new StyleElement()
+        ..text = '''
+        error-ref-wrapped > pre {
+          background-color: #f5f5f5;
+          border: 1px solid #ccc;
+          padding: 10px;
+          font-family: consolas, courier, monospace;
+          font-size: 1em;
+          line-height: 1.2em;
+          white-space: pre;
+        }
+        ''',
+      new ErrorRefElement(_error, queue: ObservatoryApplication.app.queue)
+    ];
   }
 }

runtime/observatory/lib/src/elements/flag_list.dart

diff --git a/runtime/observatory/lib/src/elements/flag_list.dart b/runtime/observatory/lib/src/elements/flag_list.dart
index 19d7816..1fdb3d5 100644
--- a/runtime/observatory/lib/src/elements/flag_list.dart
+++ b/runtime/observatory/lib/src/elements/flag_list.dart
@@ -33,7 +33,7 @@
   Stream<RenderedEvent<FlagListElement>> get onRendered => _r.onRendered;
 
   M.VMRef _vm;
-  Stream<M.VMUpdateEvent> _vmUpdates;
+  M.EventRepository _events;
   M.FlagsRepository _repository;
   M.NotificationRepository _notifications;
   Iterable<M.Flag> _flags;
@@ -41,18 +41,18 @@
   M.VMRef get vm => _vm;
 
   factory FlagListElement(M.VMRef vm,
-                          Stream<M.VMUpdateEvent> vmUpdates,
+                          M.EventRepository events,
                           M.FlagsRepository repository,
                           M.NotificationRepository notifications,
                           {RenderingQueue queue}) {
     assert(vm != null);
-    assert(vmUpdates != null);
+    assert(events != null);
     assert(repository != null);
     assert(notifications != null);
     FlagListElement e = document.createElement(tag.name);
     e._r = new RenderingScheduler(e, queue: queue);
     e._vm = vm;
-    e._vmUpdates = vmUpdates;
+    e._events = events;
     e._repository = repository;
     e._notifications = notifications;
     return e;
@@ -63,8 +63,8 @@
   @override
   void attached() {
     super.attached();
-    _r.enable();
     _refresh();
+    _r.enable();
   }
 
   @override
@@ -103,7 +103,7 @@
       new NavBarElement(queue: _r.queue)
         ..children = [
           new NavTopMenuElement(queue: _r.queue),
-          new NavVMMenuElement(_vm, _vmUpdates, queue: _r.queue),
+          new NavVMMenuElement(_vm, _events, queue: _r.queue),
           new NavMenuElement('flags', link: Uris.flags(), last: true,
                              queue: _r.queue),
           new NavRefreshElement(queue: _r.queue)
@@ -124,9 +124,11 @@
     ];
   }
 
-  Future _refresh() async {
-    _flags = await _repository.list(_vm);
-    _r.dirty();
+  Future _refresh() {
+    return _repository.list().then((flags) {
+      _flags = flags;
+      _r.dirty();
+    });
   }
 
   static bool _isModified(M.Flag flag) => flag.modified;

runtime/observatory/lib/src/elements/function_ref.dart

diff --git a/runtime/observatory/lib/src/elements/function_ref.dart b/runtime/observatory/lib/src/elements/function_ref.dart
index f6b2ec3..204169d 100644
--- a/runtime/observatory/lib/src/elements/function_ref.dart
+++ b/runtime/observatory/lib/src/elements/function_ref.dart
@@ -52,8 +52,8 @@
   @override
   void detached() {
     super.detached();
-    _r.disable(notify: true);
     children = [];
+    _r.disable(notify: true);
   }
 
   void render() {

runtime/observatory/lib/src/elements/function_ref_wrapper.dart

diff --git a/runtime/observatory/lib/src/elements/function_ref_wrapper.dart b/runtime/observatory/lib/src/elements/function_ref_wrapper.dart
index e41e43a..dbde827 100644
--- a/runtime/observatory/lib/src/elements/function_ref_wrapper.dart
+++ b/runtime/observatory/lib/src/elements/function_ref_wrapper.dart
@@ -21,10 +21,18 @@
 
   bool _qualified = true;
   ServiceFunction _function;
+
   bool get qualified => _qualified;
   ServiceFunction get ref => _function;
-  void set qualified(bool qualified) { _qualified = qualified; render(); }
-  void set ref(ServiceFunction ref) { _function = ref; render(); }
+
+  void set qualified(bool value) {
+    _qualified = value;
+    render();
+  }
+  void set ref(ServiceFunction value) {
+    _function = value;
+    render();
+  }
 
   FunctionRefElementWrapper.created() : super.created() {
     binder.registerCallback(this);
@@ -40,7 +48,9 @@
 
   void render() {
     shadowRoot.children = [];
-    if (ref == null) return;
+    if (ref == null) {
+      return;
+    }
 
     shadowRoot.children = [
       new StyleElement()

runtime/observatory/lib/src/elements/isolate_ref.dart

diff --git a/runtime/observatory/lib/src/elements/isolate_ref.dart b/runtime/observatory/lib/src/elements/isolate_ref.dart
index f27c8fb..4440001 100644
--- a/runtime/observatory/lib/src/elements/isolate_ref.dart
+++ b/runtime/observatory/lib/src/elements/isolate_ref.dart
@@ -7,7 +7,7 @@
 import 'dart:html';
 import 'dart:async';
 import 'package:observatory/models.dart' as M
-  show IsolateRef, IsolateUpdateEvent;
+  show IsolateRef, EventRepository;
 import 'package:observatory/src/elements/helpers/rendering_scheduler.dart';
 import 'package:observatory/src/elements/helpers/tag.dart';
 import 'package:observatory/src/elements/helpers/uris.dart';
@@ -19,20 +19,21 @@
 
   Stream<RenderedEvent<IsolateRefElement>> get onRendered => _r.onRendered;
 
-  Stream<M.IsolateUpdateEvent> _updates;
-  StreamSubscription _updatesSubscription;
+
   M.IsolateRef _isolate;
+  M.EventRepository _events;
+  StreamSubscription _updatesSubscription;
 
   M.IsolateRef get isolate => _isolate;
 
-  factory IsolateRefElement(M.IsolateRef isolate,
-      Stream<M.IsolateUpdateEvent> updates, {RenderingQueue queue}) {
+  factory IsolateRefElement(M.IsolateRef isolate, M.EventRepository events,
+      {RenderingQueue queue}) {
     assert(isolate != null);
-    assert(updates != null);
+    assert(events != null);
     IsolateRefElement e = document.createElement(tag.name);
     e._r = new RenderingScheduler(e, queue: queue);
     e._isolate = isolate;
-    e._updates = updates;
+    e._events = events;
     return e;
   }
 
@@ -41,21 +42,18 @@
   @override
   void attached() {
     super.attached();
-    assert(_isolate != null);
-    assert(_updates != null);
+    _updatesSubscription = _events.onIsolateUpdate
+      .where((e) => e.isolate.id == isolate.id)
+      .listen((e) { _isolate = e.isolate; _r.dirty(); });
     _r.enable();
-    _updatesSubscription = _updates
-      .where((M.IsolateUpdateEvent e) => e.isolate.id == isolate.id)
-      .listen((M.IsolateUpdateEvent e) { _isolate = e.isolate; _r.dirty(); });
   }
 
   @override
   void detached() {
-    super.detached(); _r.disable(notify: true);
+    super.detached();
     children = [];
-    assert(_updatesSubscription != null);
+    _r.disable(notify: true);
     _updatesSubscription.cancel();
-    _updatesSubscription = null;
   }
 
   void render() {

runtime/observatory/lib/src/elements/isolate_ref_wrapper.dart

diff --git a/runtime/observatory/lib/src/elements/isolate_ref_wrapper.dart b/runtime/observatory/lib/src/elements/isolate_ref_wrapper.dart
index 369e68a..6c9547b 100644
--- a/runtime/observatory/lib/src/elements/isolate_ref_wrapper.dart
+++ b/runtime/observatory/lib/src/elements/isolate_ref_wrapper.dart
@@ -3,11 +3,8 @@
 // BSD-style license that can be found in the LICENSE file.
 
 import 'dart:html';
-import 'dart:async';
 
 import 'package:observatory/app.dart';
-import 'package:observatory/models.dart' show IsolateUpdateEvent;
-import 'package:observatory/mocks.dart' show IsolateUpdateEventMock;
 import 'package:observatory/service_html.dart' show Isolate;
 import 'package:observatory/src/elements/isolate_ref.dart';
 import 'package:observatory/src/elements/helpers/tag.dart';
@@ -22,17 +19,16 @@
 
   static const tag = const Tag<IsolateRefElementWrapper>('isolate-ref');
 
-  final StreamController<IsolateUpdateEvent> _updatesController =
-    new StreamController<IsolateUpdateEvent>();
-  Stream<IsolateUpdateEvent> _updates;
-  StreamSubscription _subscription;
-
   Isolate _isolate;
+
   Isolate get ref => _isolate;
-  void set ref(Isolate ref) { _isolate = ref; _detached(); _attached(); }
+
+  void set ref(Isolate value) {
+    _isolate = value;
+    render();
+  }
 
   IsolateRefElementWrapper.created() : super.created() {
-    _updates = _updatesController.stream.asBroadcastStream();
     binder.registerCallback(this);
     createShadowRoot();
     render();
@@ -41,34 +37,14 @@
   @override
   void attached() {
     super.attached();
-    _attached();
-  }
-
-  void _attached() {
-    if (ref != null) {
-      _subscription = ref.changes.listen((_) {
-        _updatesController.add(new IsolateUpdateEventMock(isolate: ref));
-      });
-    }
     render();
   }
 
-  @override
-  void detached() {
-    super.detached();
-    _detached();
-  }
-
-  void _detached() {
-    if (_subscription != null) {
-      _subscription.cancel();
-      _subscription = null;
-    }
-  }
-
   void render() {
     shadowRoot.children = [];
-    if (ref == null) return;
+    if (ref == null) {
+      return;
+    }
 
     shadowRoot.children = [
       new StyleElement()
@@ -80,8 +56,9 @@
             color: #0489c3;
             text-decoration: none;
         }''',
-      new IsolateRefElement(_isolate, _updates,
-                                 queue: ObservatoryApplication.app.queue)
+      new IsolateRefElement(_isolate, app.events, queue: app.queue)
     ];
   }
+
+  ObservatoryApplication get app => ObservatoryApplication.app;
 }

runtime/observatory/lib/src/elements/nav/bar.dart

diff --git a/runtime/observatory/lib/src/elements/nav/bar.dart b/runtime/observatory/lib/src/elements/nav/bar.dart
index 8bb115e..b175ba6 100644
--- a/runtime/observatory/lib/src/elements/nav/bar.dart
+++ b/runtime/observatory/lib/src/elements/nav/bar.dart
@@ -11,8 +11,9 @@
   static final StyleElement _style = () {
       var style = new StyleElement();
       style.text = 'nav.nav-bar {'
+                     'line-height: normal;'
                      'position: fixed;'
-                     'top: -56px;'
+                     'top: 0;'
                      'width: 100%;'
                      'z-index: 1000;'
                      '}'
@@ -21,20 +22,12 @@
                      'position: relative;'
                      'list-style: none;'
                      'padding-left: 0;'
-                     'margin-left: 0;'
+                     'margin: 0;'
                      'width: 100%;'
                      'z-index: 1000;'
                      'font: 400 16px \'Montserrat\', sans-serif;'
                      'color: white;'
                      'background-color: #0489c3;'
-                   '}'
-                   'nav.nav-bar:after {'
-                     'content: ""; clear: both; display: block;'
-                   '}'
-                   'nav.nav-bar:before {'
-                     'height: 40px;'
-                     'background-color: #0489c3;'
-                     'content: ""; display: block;'
                    '}';
       return style;
   }();

runtime/observatory/lib/src/elements/nav/class_menu_wrapper.dart

diff --git a/runtime/observatory/lib/src/elements/nav/class_menu_wrapper.dart b/runtime/observatory/lib/src/elements/nav/class_menu_wrapper.dart
index 35fc448..42a65de 100644
--- a/runtime/observatory/lib/src/elements/nav/class_menu_wrapper.dart
+++ b/runtime/observatory/lib/src/elements/nav/class_menu_wrapper.dart
@@ -21,13 +21,17 @@
 
   bool _last = false;
   Class _cls;
+
   bool get last => _last;
   Class get cls => _cls;
+  
   set last(bool value) {
-    _last = value; render();
+    _last = value;
+    render();
   }
   set cls(Class value) {
-    _cls = value; render();
+    _cls = value;
+    render();
   }
 
   NavClassMenuElementWrapper.created() : super.created() {
@@ -45,7 +49,9 @@
 
   void render() {
     shadowRoot.children = [];
-    if (_cls == null || _last == null) return;
+    if (_cls == null || _last == null) {
+      return;
+    }
 
     shadowRoot.children = [
       new NavClassMenuElement(cls.isolate, cls, last: last,

runtime/observatory/lib/src/elements/nav/isolate_menu.dart

diff --git a/runtime/observatory/lib/src/elements/nav/isolate_menu.dart b/runtime/observatory/lib/src/elements/nav/isolate_menu.dart
index 3f0089a..d9f82c9 100644
--- a/runtime/observatory/lib/src/elements/nav/isolate_menu.dart
+++ b/runtime/observatory/lib/src/elements/nav/isolate_menu.dart
@@ -5,7 +5,7 @@
 import 'dart:html';
 import 'dart:async';
 import 'package:observatory/models.dart' as M
-  show IsolateRef, IsolateUpdateEvent;
+  show IsolateRef, EventRepository;
 import 'package:observatory/src/elements/helpers/rendering_scheduler.dart';
 import 'package:observatory/src/elements/helpers/tag.dart';
 import 'package:observatory/src/elements/helpers/uris.dart';
@@ -21,25 +21,27 @@
 
   Stream<RenderedEvent<NavIsolateMenuElement>> get onRendered => _r.onRendered;
 
-  Stream<M.IsolateUpdateEvent> _updates;
-  StreamSubscription _updatesSubscription;
-
   bool _last;
   M.IsolateRef _isolate;
+  M.EventRepository _events;
+  StreamSubscription _updatesSubscription;
+
   bool get last => _last;
   M.IsolateRef get isolate => _isolate;
+
   set last(bool value) => _last = _r.checkAndReact(_last, value);
 
   factory NavIsolateMenuElement(M.IsolateRef isolate,
-      Stream<M.IsolateUpdateEvent> updates, {bool last: false,
+      M.EventRepository events, {bool last: false,
       RenderingQueue queue}) {
     assert(isolate != null);
+    assert(events != null);
     assert(last != null);
     NavIsolateMenuElement e = document.createElement(tag.name);
     e._r = new RenderingScheduler(e, queue: queue);
     e._isolate = isolate;
     e._last = last;
-    e._updates = updates;
+    e._events = events;
     return e;
   }
 
@@ -51,12 +53,10 @@
   @override
   void attached() {
     super.attached();
-    assert(_isolate != null);
-    assert(_updates != null);
+    _updatesSubscription = _events.onIsolateUpdate
+      .where((e) => e.isolate.id == isolate.id)
+      .listen((e) { _isolate = e.isolate; _r.dirty(); });
     _r.enable();
-    _updatesSubscription = _updates
-      .where((M.IsolateUpdateEvent e) => e.isolate.id == isolate.id)
-      .listen((M.IsolateUpdateEvent e) { _isolate = e.isolate; _r.dirty(); });
   }
 
   @override

runtime/observatory/lib/src/elements/nav/isolate_menu_wrapper.dart

diff --git a/runtime/observatory/lib/src/elements/nav/isolate_menu_wrapper.dart b/runtime/observatory/lib/src/elements/nav/isolate_menu_wrapper.dart
index 3af85d2..ff4e1ed 100644
--- a/runtime/observatory/lib/src/elements/nav/isolate_menu_wrapper.dart
+++ b/runtime/observatory/lib/src/elements/nav/isolate_menu_wrapper.dart
@@ -3,12 +3,9 @@
 // BSD-style license that can be found in the LICENSE file.
 
 import 'dart:html';
-import 'dart:async';
 
 import 'package:observatory/app.dart';
 import 'package:observatory/service.dart';
-import 'package:observatory/mocks.dart';
-import 'package:observatory/models.dart' as M;
 import 'package:observatory/src/elements/helpers/tag.dart';
 import 'package:observatory/src/elements/shims/binding.dart';
 import 'package:observatory/src/elements/nav/isolate_menu.dart';
@@ -22,24 +19,22 @@
   static const tag =
     const Tag<NavIsolateMenuElementWrapper>('isolate-nav-menu');
 
-  final StreamController<M.IsolateUpdateEvent> _updatesController =
-    new StreamController<M.IsolateUpdateEvent>();
-  Stream<M.IsolateUpdateEvent> _updates;
-  StreamSubscription _subscription;
-
   bool _last = false;
   Isolate _isolate;
+  
   bool get last => _last;
   Isolate get isolate => _isolate;
+
   set last(bool value) {
-    _last = value; render();
+    _last = value;
+    render();
   }
   set isolate(Isolate value) {
-    _isolate = value;  _detached(); _attached();
+    _isolate = value;
+    render();
   }
 
   NavIsolateMenuElementWrapper.created() : super.created() {
-    _updates = _updatesController.stream.asBroadcastStream();
     binder.registerCallback(this);
     _last = _getBoolAttribute('last');
     createShadowRoot();
@@ -49,42 +44,24 @@
   @override
   void attached() {
     super.attached();
-    _attached();
-  }
-
-  @override
-  void detached() {
-    super.detached();
-    _detached();
-  }
-
-  void _attached() {
-    if (_isolate != null) {
-      _subscription = _isolate.changes.listen((_) {
-        _updatesController.add(new IsolateUpdateEventMock(isolate: isolate));
-      });
-    }
     render();
   }
 
-  void _detached() {
-    if (_subscription != null) {
-      _subscription.cancel();
-      _subscription = null;
-    }
-  }
-
   void render() {
     shadowRoot.children = [];
-    if (_isolate == null || _last == null) return;
+    if (_isolate == null || _last == null) {
+      return;
+    }
 
     shadowRoot.children = [
-      new NavIsolateMenuElement(isolate, _updates, last: last,
-                                 queue: ObservatoryApplication.app.queue)
+      new NavIsolateMenuElement(_isolate, app.events, last: _last,
+          queue: app.queue)
         ..children = [new ContentElement()]
     ];
   }
 
+  ObservatoryApplication get app => ObservatoryApplication.app;
+
   bool _getBoolAttribute(String name) {
     final String value = getAttribute(name);
     return !(value == null || value == 'false');

runtime/observatory/lib/src/elements/nav/library_menu.dart

diff --git a/runtime/observatory/lib/src/elements/nav/library_menu.dart b/runtime/observatory/lib/src/elements/nav/library_menu.dart
index 0403ac3..36160d6 100644
--- a/runtime/observatory/lib/src/elements/nav/library_menu.dart
+++ b/runtime/observatory/lib/src/elements/nav/library_menu.dart
@@ -21,9 +21,11 @@
   bool _last;
   M.IsolateRef _isolate;
   M.LibraryRef _library;
+
   bool get last => _last;
   M.IsolateRef get isolate => _isolate;
   M.LibraryRef get library => _library;
+  
   set last(bool value) => _last = _r.checkAndReact(_last, value);
 
   factory NavLibraryMenuElement(M.IsolateRef isolate, M.LibraryRef library,
@@ -42,11 +44,15 @@
   NavLibraryMenuElement.created() : super.created() { createShadowRoot(); }
 
   @override
-  void attached() { super.attached(); _r.enable(); }
+  void attached() {
+    super.attached();
+    _r.enable();
+  }
 
   @override
   void detached() {
-    super.detached(); _r.disable(notify: true);
+    super.detached();
+    _r.disable(notify: true);
     shadowRoot.children = [];
   }
 

runtime/observatory/lib/src/elements/nav/library_menu_wrapper.dart

diff --git a/runtime/observatory/lib/src/elements/nav/library_menu_wrapper.dart b/runtime/observatory/lib/src/elements/nav/library_menu_wrapper.dart
index 1097bdc..8db0347 100644
--- a/runtime/observatory/lib/src/elements/nav/library_menu_wrapper.dart
+++ b/runtime/observatory/lib/src/elements/nav/library_menu_wrapper.dart
@@ -21,13 +21,17 @@
 
   bool _last = false;
   Library _library;
+
   bool get last => _last;
   Library get library => _library;
+  
   set last(bool value) {
-    _last = value; render();
+    _last = value;
+    render();
   }
   set library(Library value) {
-    _library = value; render();
+    _library = value;
+    render();
   }
 
   NavLibraryMenuElementWrapper.created() : super.created() {
@@ -45,7 +49,9 @@
 
   void render() {
     shadowRoot.children = [];
-    if (_library == null || _last == null) return;
+    if (_library == null || _last == null) {
+      return;
+    }
 
     shadowRoot.children = [
       new NavLibraryMenuElement(library.isolate, library, last: last,

runtime/observatory/lib/src/elements/nav/menu.dart

diff --git a/runtime/observatory/lib/src/elements/nav/menu.dart b/runtime/observatory/lib/src/elements/nav/menu.dart
index bcd9c90..2624ad7 100644
--- a/runtime/observatory/lib/src/elements/nav/menu.dart
+++ b/runtime/observatory/lib/src/elements/nav/menu.dart
@@ -53,9 +53,11 @@
   String _label;
   String _link;
   bool _last;
+
   String get label => _label;
   String get link => _link;
   bool get last => _last;
+  
   set label(String value) => _label = _r.checkAndReact(_label, value);
   set link(String value) => _link = _r.checkAndReact(_link, value);
   set last(bool value) => _last = _r.checkAndReact(_link, value);
@@ -75,11 +77,15 @@
   NavMenuElement.created() : super.created() { createShadowRoot(); }
 
   @override
-  void attached() { super.attached(); _r.enable(); }
+  void attached() {
+    super.attached();
+    _r.enable();
+  }
 
   @override
   void detached() {
-    super.detached(); _r.disable(notify: true);
+    super.detached();
+    _r.disable(notify: true);
     shadowRoot.children = [];
   }
 

runtime/observatory/lib/src/elements/nav/menu_item.dart

diff --git a/runtime/observatory/lib/src/elements/nav/menu_item.dart b/runtime/observatory/lib/src/elements/nav/menu_item.dart
index 2411ee5..03ced3d 100644
--- a/runtime/observatory/lib/src/elements/nav/menu_item.dart
+++ b/runtime/observatory/lib/src/elements/nav/menu_item.dart
@@ -55,8 +55,10 @@
 
   String _label;
   String _link;
+
   String get label => _label;
   String get link => _link;
+  
   set label(String value) => _label = _r.checkAndReact(_label, value);
   set link(String value) => _link = _r.checkAndReact(_link, value);
 
@@ -74,11 +76,15 @@
   NavMenuItemElement.created() : super.created() { createShadowRoot(); }
 
   @override
-  void attached() { super.attached(); _r.enable(); }
+  void attached() {
+    super.attached();
+    _r.enable();
+  }
 
   @override
   void detached() {
-    super.detached(); _r.disable(notify: true);
+    super.detached();
+    _r.disable(notify: true);
     shadowRoot.children = [];
   }
 

runtime/observatory/lib/src/elements/nav/menu_item_wrapper.dart

diff --git a/runtime/observatory/lib/src/elements/nav/menu_item_wrapper.dart b/runtime/observatory/lib/src/elements/nav/menu_item_wrapper.dart
index 7888745..e00699c 100644
--- a/runtime/observatory/lib/src/elements/nav/menu_item_wrapper.dart
+++ b/runtime/observatory/lib/src/elements/nav/menu_item_wrapper.dart
@@ -20,13 +20,17 @@
 
   String _anchor;
   String _link;
+
   String get anchor => _anchor;
   String get link => _link;
+  
   set anchor(String value) {
-    _anchor = value; render();
+    _anchor = value;
+    render();
   }
   set link(String value) {
-    _link = value; render();
+    _link = value;
+    render();
   }
 
   NavMenuItemElementWrapper.created() : super.created() {
@@ -45,7 +49,9 @@
 
   void render() {
     shadowRoot.children = [];
-    if (_anchor == null) return;
+    if (_anchor == null) {
+      return;
+    }
 
     shadowRoot.children = [
       new NavMenuItemElement(_anchor, link: '#$link',

runtime/observatory/lib/src/elements/nav/menu_wrapper.dart

diff --git a/runtime/observatory/lib/src/elements/nav/menu_wrapper.dart b/runtime/observatory/lib/src/elements/nav/menu_wrapper.dart
index 04577c6..3b287fe 100644
--- a/runtime/observatory/lib/src/elements/nav/menu_wrapper.dart
+++ b/runtime/observatory/lib/src/elements/nav/menu_wrapper.dart
@@ -21,17 +21,22 @@
   String _anchor = '---';
   String _link;
   bool _last = false;
+
   String get anchor => _anchor;
   String get link => _link;
   bool get last => _last;
+  
   set anchor(String value) {
-    _anchor = value; render();
+    _anchor = value;
+    render();
   }
   set link(String value) {
-    _link = value; render();
+    _link = value;
+    render();
   }
   set last(bool value) {
-    _last = value; render();
+    _last = value;
+    render();
   }
 
   NavMenuElementWrapper.created() : super.created() {
@@ -51,7 +56,9 @@
 
   void render() {
     shadowRoot.children = [];
-    if (_anchor == null || _last == null) return;
+    if (_anchor == null || _last == null) {
+      return;
+    }
 
     shadowRoot.children = [
       new NavMenuElement(_anchor, link: '#$_link', last: last,

runtime/observatory/lib/src/elements/nav/notify.dart

diff --git a/runtime/observatory/lib/src/elements/nav/notify.dart b/runtime/observatory/lib/src/elements/nav/notify.dart
index ba99f01..1566a04 100644
--- a/runtime/observatory/lib/src/elements/nav/notify.dart
+++ b/runtime/observatory/lib/src/elements/nav/notify.dart
@@ -23,7 +23,9 @@
   StreamSubscription _subscription;
 
   bool _notifyOnPause;
+
   bool get notifyOnPause => _notifyOnPause;
+  
   set notifyOnPause(bool value) =>
       _notifyOnPause = _r.checkAndReact(_notifyOnPause, value);
 
@@ -43,8 +45,8 @@
   @override
   void attached() {
     super.attached();
-    _r.enable();
     _subscription = _repository.onChange.listen((_) => _r.dirty());
+    _r.enable();
   }
 
   @override

runtime/observatory/lib/src/elements/nav/notify_event.dart

diff --git a/runtime/observatory/lib/src/elements/nav/notify_event.dart b/runtime/observatory/lib/src/elements/nav/notify_event.dart
index 0f75062..b46236e 100644
--- a/runtime/observatory/lib/src/elements/nav/notify_event.dart
+++ b/runtime/observatory/lib/src/elements/nav/notify_event.dart
@@ -26,6 +26,7 @@
   Stream<EventDeleteEvent> get onDelete => _onDelete.stream;
 
   M.Event _event;
+  
   M.Event get event => _event;
 
   factory NavNotifyEventElement(M.Event event, {RenderingQueue queue}) {

runtime/observatory/lib/src/elements/nav/notify_exception.dart

diff --git a/runtime/observatory/lib/src/elements/nav/notify_exception.dart b/runtime/observatory/lib/src/elements/nav/notify_exception.dart
index cf0a463..0e92bda 100644
--- a/runtime/observatory/lib/src/elements/nav/notify_exception.dart
+++ b/runtime/observatory/lib/src/elements/nav/notify_exception.dart
@@ -31,6 +31,7 @@
 
   Exception _exception;
   StackTrace _stacktrace;
+  
   Exception get exception => _exception;
   StackTrace get stacktrace => _stacktrace;
 

runtime/observatory/lib/src/elements/nav/notify_wrapper.dart

diff --git a/runtime/observatory/lib/src/elements/nav/notify_wrapper.dart b/runtime/observatory/lib/src/elements/nav/notify_wrapper.dart
index 7676727..58dbf39 100644
--- a/runtime/observatory/lib/src/elements/nav/notify_wrapper.dart
+++ b/runtime/observatory/lib/src/elements/nav/notify_wrapper.dart
@@ -20,13 +20,17 @@
 
   NotificationRepository _notifications;
   bool _notifyOnPause = true;
+
   NotificationRepository get notifications => _notifications;
   bool get notifyOnPause => _notifyOnPause;
+  
   set notifications(NotificationRepository value) {
-    _notifications = value; render();
+    _notifications = value;
+    render();
   }
   set notifyOnPause(bool value) {
-    _notifyOnPause = value; render();
+    _notifyOnPause = value;
+    render();
   }
 
   NavNotifyElementWrapper.created() : super.created() {
@@ -43,7 +47,9 @@
 
   void render() {
     shadowRoot.children = [];
-    if (_notifications == null) return;
+    if (_notifications == null) {
+      return;
+    }
 
     shadowRoot.children = [
       new StyleElement()

runtime/observatory/lib/src/elements/nav/refresh.dart

diff --git a/runtime/observatory/lib/src/elements/nav/refresh.dart b/runtime/observatory/lib/src/elements/nav/refresh.dart
index 795c7bd..da40abd 100644
--- a/runtime/observatory/lib/src/elements/nav/refresh.dart
+++ b/runtime/observatory/lib/src/elements/nav/refresh.dart
@@ -25,8 +25,10 @@
 
   bool _disabled;
   String _label;
+
   bool get disabled => _disabled;
   String get label => _label;
+  
   set disabled(bool value) => _disabled = _r.checkAndReact(_disabled, value);
   set label(String value) => _label = _r.checkAndReact(_label, value);
 
@@ -44,10 +46,17 @@
   NavRefreshElement.created() : super.created();
 
   @override
-  void attached() { super.attached(); _r.enable(); }
+  void attached() {
+    super.attached();
+    _r.enable();
+  }
 
   @override
-  void detached() { super.detached(); children = []; _r.disable(notify: true); }
+  void detached() {
+    super.detached();
+    children = [];
+    _r.disable(notify: true);
+  }
 
   void render() {
     children = [

runtime/observatory/lib/src/elements/nav/refresh_wrapper.dart

diff --git a/runtime/observatory/lib/src/elements/nav/refresh_wrapper.dart b/runtime/observatory/lib/src/elements/nav/refresh_wrapper.dart
index 19e7cce..1989a10 100644
--- a/runtime/observatory/lib/src/elements/nav/refresh_wrapper.dart
+++ b/runtime/observatory/lib/src/elements/nav/refresh_wrapper.dart
@@ -20,13 +20,17 @@
 
   Function _callback;
   String _label;
+  
   Function get callback => _callback;
   String get label => _label;
+
   set callback(Function value) {
-    _callback = value; render();
+    _callback = value;
+    render();
   }
   set label(String value) {
-    _label = value; render();
+    _label = value;
+    render();
   }
 
   NavRefreshElementWrapper.created() : super.created() {
@@ -44,7 +48,9 @@
 
   void render() {
     shadowRoot.children = [];
-    if (_callback == null || _label == null) return;
+    if (_callback == null || _label == null) {
+      return;
+    }
 
     shadowRoot.children = [
       new StyleElement()

runtime/observatory/lib/src/elements/nav/top_menu.dart

diff --git a/runtime/observatory/lib/src/elements/nav/top_menu.dart b/runtime/observatory/lib/src/elements/nav/top_menu.dart
index a289277..f4c6fa1 100644
--- a/runtime/observatory/lib/src/elements/nav/top_menu.dart
+++ b/runtime/observatory/lib/src/elements/nav/top_menu.dart
@@ -20,7 +20,9 @@
   Stream<RenderedEvent<NavTopMenuElement>> get onRendered => _r.onRendered;
 
   bool _last;
+  
   bool get last => _last;
+
   set last(bool value) => _last = _r.checkAndReact(_last, value);
 
   factory NavTopMenuElement({bool last: false, RenderingQueue queue}) {

runtime/observatory/lib/src/elements/nav/top_menu_wrapper.dart

diff --git a/runtime/observatory/lib/src/elements/nav/top_menu_wrapper.dart b/runtime/observatory/lib/src/elements/nav/top_menu_wrapper.dart
index edc5e27..db99fff 100644
--- a/runtime/observatory/lib/src/elements/nav/top_menu_wrapper.dart
+++ b/runtime/observatory/lib/src/elements/nav/top_menu_wrapper.dart
@@ -18,9 +18,12 @@
   static const tag = const Tag<NavTopMenuElementWrapper>('top-nav-menu');
 
   bool _last = false;
+  
   bool get last => _last;
+
   set last(bool value) {
-    _last = value; render();
+    _last = value;
+    render();
   }
 
   NavTopMenuElementWrapper.created() : super.created() {
@@ -38,7 +41,9 @@
 
   void render() {
     shadowRoot.children = [];
-    if (_last == null) return;
+    if (_last == null) {
+      return;
+    }
 
     shadowRoot.children = [
       new NavTopMenuElement(last: last, queue: ObservatoryApplication.app.queue)

runtime/observatory/lib/src/elements/nav/vm_menu.dart

diff --git a/runtime/observatory/lib/src/elements/nav/vm_menu.dart b/runtime/observatory/lib/src/elements/nav/vm_menu.dart
index a46d4ca..e4351af 100644
--- a/runtime/observatory/lib/src/elements/nav/vm_menu.dart
+++ b/runtime/observatory/lib/src/elements/nav/vm_menu.dart
@@ -5,7 +5,7 @@
 import 'dart:html';
 import 'dart:async';
 import 'package:observatory/models.dart' as M
-  show VM, IsolateRef, Target, VMUpdateEvent;
+  show VM, EventRepository;
 import 'package:observatory/src/elements/helpers/rendering_scheduler.dart';
 import 'package:observatory/src/elements/helpers/tag.dart';
 import 'package:observatory/src/elements/helpers/uris.dart';
@@ -21,28 +21,27 @@
 
   Stream<RenderedEvent<NavVMMenuElement>> get onRendered => _r.onRendered;
 
-  Stream<M.VMUpdateEvent> _updates;
-  StreamSubscription _updatesSubscription;
-
   bool _last;
   M.VM _vm;
-  M.Target _target;
+  M.EventRepository _events;
+  StreamSubscription _updatesSubscription;
+
+
   bool get last => _last;
   M.VM get vm => _vm;
-  M.Target get target => _target;
+
   set last(bool value) => _last = _r.checkAndReact(_last, value);
 
-  factory NavVMMenuElement(M.VM vm, Stream<M.VMUpdateEvent> updates,
-      {bool last: false, M.Target target, RenderingQueue queue}) {
+  factory NavVMMenuElement(M.VM vm, M.EventRepository events, {bool last: false,
+    RenderingQueue queue}) {
     assert(vm != null);
-    assert(updates != null);
+    assert(events != null);
     assert(last != null);
     NavVMMenuElement e = document.createElement(tag.name);
     e._r = new RenderingScheduler(e, queue: queue);
     e._vm = vm;
-    e._updates = updates;
+    e._events = events;
     e._last = last;
-    e._target = target;
     return e;
   }
 
@@ -51,29 +50,25 @@
   @override
   void attached() {
     super.attached();
+    _updatesSubscription = _events.onVMUpdate
+        .listen((e) { _vm = e.vm; _r.dirty(); });
     _r.enable();
-    _updatesSubscription = _updates
-      .listen((M.VMUpdateEvent e) { _vm = e.vm; _r.dirty(); });
   }
 
   @override
   void detached() {
     super.detached();
-    _r.disable(notify: true);
     shadowRoot.children = [];
-    assert(_updatesSubscription != null);
+    _r.disable(notify: true);
     _updatesSubscription.cancel();
-    _updatesSubscription = null;
   }
 
   void render() {
-    final String name = (target == null) ? vm.name
-                                         : '${vm.name}@${target.name}';
-    /// TODO(cbernaschina) use the isolate repository.
     shadowRoot.children = [
-      new NavMenuElement(name, link: Uris.vm(), last: last, queue: _r.queue)
+      new NavMenuElement(vm.displayName, link: Uris.vm(), last: last,
+          queue: _r.queue)
         ..children = (
-          _vm.isolates.map((M.IsolateRef isolate) {
+          _vm.isolates.map((isolate) {
             return new NavMenuItemElement(isolate.name, queue: _r.queue,
                 link: Uris.inspect(isolate));
           }).toList()

runtime/observatory/lib/src/elements/nav/vm_menu_wrapper.dart

diff --git a/runtime/observatory/lib/src/elements/nav/vm_menu_wrapper.dart b/runtime/observatory/lib/src/elements/nav/vm_menu_wrapper.dart
index 9327e19..ef8bf3d 100644
--- a/runtime/observatory/lib/src/elements/nav/vm_menu_wrapper.dart
+++ b/runtime/observatory/lib/src/elements/nav/vm_menu_wrapper.dart
@@ -3,11 +3,8 @@
 // BSD-style license that can be found in the LICENSE file.
 
 import 'dart:html';
-import 'dart:async';
 
 import 'package:observatory/app.dart';
-import 'package:observatory/mocks.dart';
-import 'package:observatory/models.dart' as M;
 import 'package:observatory/service.dart';
 import 'package:observatory/service_common.dart';
 import 'package:observatory/src/elements/helpers/tag.dart';
@@ -22,19 +19,18 @@
 
   static const tag = const Tag<NavVMMenuElementWrapper>('vm-nav-menu');
 
-  StreamSubscription _subscription;
-  StreamController<M.VMUpdateEvent> _updatesController =
-      new StreamController<M.VMUpdateEvent>.broadcast();
-
   bool _last = false;
   VM _vm;
+
   bool get last => _last;
   VM get vm => _vm;
+
   set last(bool value) {
-    _last = value; render();
-  }
+    _last = value;
+    render(); }
   set vm(VM value) {
-    _vm = value; _detached(); _attached();
+    _vm = value;
+    render();
   }
 
   NavVMMenuElementWrapper.created() : super.created() {
@@ -47,43 +43,24 @@
   @override
   void attached() {
     super.attached();
-    _attached();
-  }
-
-  @override
-  void detached() {
-    super.detached();
-    _detached();
-  }
-
-  void _attached() {
-    if (_vm != null) {
-      _subscription = _vm.changes.listen((_) {
-        _updatesController.add(new VMUpdateEventMock(vm: _vm));
-      });
-    }
     render();
   }
 
-  void _detached() {
-    if (_subscription != null) {
-      _subscription.cancel();
-      _subscription = null;
-    }
-  }
-
   void render() {
     shadowRoot.children = [];
-    if (_vm == null || _last == null) return;
+    if (_vm == null || _last == null) {
+      return;
+    }
 
     shadowRoot.children = [
-      new NavVMMenuElement(vm, _updatesController.stream, last: last,
-          target: (vm as CommonWebSocketVM)?.target,
-          queue: ObservatoryApplication.app.queue)
+      new NavVMMenuElement(vm, app.events, last: last,
+          queue: app.queue)
         ..children = [new ContentElement()]
     ];
   }
 
+  ObservatoryApplication get app => ObservatoryApplication.app;
+
   bool _getBoolAttribute(String name) {
     final String value = getAttribute(name);
     return !(value == null || value == 'false');

runtime/observatory/lib/src/elements/script_ref.dart

diff --git a/runtime/observatory/lib/src/elements/script_ref.dart b/runtime/observatory/lib/src/elements/script_ref.dart
index 8cc2f68..67ae388 100644
--- a/runtime/observatory/lib/src/elements/script_ref.dart
+++ b/runtime/observatory/lib/src/elements/script_ref.dart
@@ -41,7 +41,6 @@
   @override
   void attached() {
     super.attached();
-    assert(script != null);
     _r.enable();
   }
 

runtime/observatory/lib/src/elements/script_ref_wrapper.dart

diff --git a/runtime/observatory/lib/src/elements/script_ref_wrapper.dart b/runtime/observatory/lib/src/elements/script_ref_wrapper.dart
index 01f6940..8f0e343 100644
--- a/runtime/observatory/lib/src/elements/script_ref_wrapper.dart
+++ b/runtime/observatory/lib/src/elements/script_ref_wrapper.dart
@@ -20,8 +20,13 @@
   static const tag = const Tag<ScriptRefElementWrapper>('script-ref');
 
   Script _script;
+
   Script get ref => _script;
-  set ref(Script script) { _script = script; render(); }
+
+  set ref(Script value) {
+    _script = value;
+    render();
+  }
 
   ScriptRefElementWrapper.created() : super.created() {
     binder.registerCallback(this);
@@ -37,7 +42,9 @@
 
   Future render() async {
     shadowRoot.children = [];
-    if (_script == null) return;
+    if (_script == null) {
+      return;
+    }
 
     shadowRoot.children = [
       new StyleElement()

runtime/observatory/lib/src/elements/source_link.dart

diff --git a/runtime/observatory/lib/src/elements/source_link.dart b/runtime/observatory/lib/src/elements/source_link.dart
index 02eec26..110b8e1 100644
--- a/runtime/observatory/lib/src/elements/source_link.dart
+++ b/runtime/observatory/lib/src/elements/source_link.dart
@@ -44,16 +44,19 @@
   @override
   void attached() {
     super.attached();
-    assert(location != null);
-    _r.enable();
     _repository.get(_location.script.id).then((script) {
       _script = script;
       _r.dirty();
     });
+    _r.enable();
   }
 
   @override
-  void detached() { super.detached(); children = []; _r.disable(notify: true); }
+  void detached() {
+    super.detached();
+    children = [];
+    _r.disable(notify: true);
+  }
 
   Future render() async {
     if (_script == null) {

runtime/observatory/lib/src/elements/source_link_wrapper.dart

diff --git a/runtime/observatory/lib/src/elements/source_link_wrapper.dart b/runtime/observatory/lib/src/elements/source_link_wrapper.dart
index d38d70f..20221ea 100644
--- a/runtime/observatory/lib/src/elements/source_link_wrapper.dart
+++ b/runtime/observatory/lib/src/elements/source_link_wrapper.dart
@@ -22,7 +22,10 @@
 
   SourceLocation _location;
   SourceLocation get location => location;
-  set location(SourceLocation location) { _location = location; render(); }
+  set location(SourceLocation value) {
+    _location = value;
+    render();
+  }
 
   SourceLinkElementWrapper.created() : super.created() {
     binder.registerCallback(this);
@@ -38,7 +41,9 @@
 
   Future render() async {
     shadowRoot.children = [];
-    if (_location == null) return;
+    if (_location == null) {
+      return;
+    }
 
     ScriptRepository repository = new ScriptRepository(_location.isolate);
 

runtime/observatory/lib/src/elements/view_footer.dart

diff --git a/runtime/observatory/lib/src/elements/view_footer.dart b/runtime/observatory/lib/src/elements/view_footer.dart
index 7493a7a..3e7dcd1 100644
--- a/runtime/observatory/lib/src/elements/view_footer.dart
+++ b/runtime/observatory/lib/src/elements/view_footer.dart
@@ -23,7 +23,7 @@
   }
 
   ViewFooterElement.created() : super.created() {
-    // TODO(cbernaschina) remove this when no more needed.
+    // TODO(cbernaschina) remove this when polymer is removed.
     _r = new RenderingScheduler(this);
   }
 
@@ -36,8 +36,8 @@
   @override
   void detached() {
     super.detached();
-    _r.disable(notify: true);
     children = [];
+    _r.disable(notify: true);
   }
 
   void render() {

runtime/observatory/lib/src/elements/vm_connect.dart

diff --git a/runtime/observatory/lib/src/elements/vm_connect.dart b/runtime/observatory/lib/src/elements/vm_connect.dart
index 65487cf..a322e91 100644
--- a/runtime/observatory/lib/src/elements/vm_connect.dart
+++ b/runtime/observatory/lib/src/elements/vm_connect.dart
@@ -16,6 +16,8 @@
 import 'package:observatory/src/elements/view_footer.dart';
 import 'package:observatory/src/elements/vm_connect_target.dart';
 
+typedef void CrashDumpLoadCallback(Map dump);
+
 class VMConnectElement extends HtmlElement implements Renderable {
   static const tag = const Tag<VMConnectElement>('vm-connect',
                      dependencies: const [NavBarElement.tag,
@@ -28,7 +30,7 @@
 
   Stream<RenderedEvent<VMConnectElement>> get onRendered => _r.onRendered;
 
-  M.CrashDumpRepository _dump;
+  CrashDumpLoadCallback _loadDump;
   M.NotificationRepository _notifications;
   M.TargetRepository _targets;
   StreamSubscription _targetsSubscription;
@@ -36,17 +38,17 @@
   String _address;
 
   factory VMConnectElement(M.TargetRepository targets,
-                           M.CrashDumpRepository dump,
+                           CrashDumpLoadCallback loadDump,
                            M.NotificationRepository notifications,
                            {String address: '', RenderingQueue queue}) {
     assert(address != null);
-    assert(dump != null);
+    assert(loadDump != null);
     assert(notifications != null);
     assert(targets != null);
     VMConnectElement e = document.createElement(tag.name);
     e._r = new RenderingScheduler(e, queue: queue);
     e._address = address;
-    e._dump = dump;
+    e._loadDump = loadDump;
     e._notifications = notifications;
     e._targets = targets;
     return e;
@@ -56,14 +58,17 @@
 
   @override
   void attached() {
-    super.attached(); _r.enable();
+    super.attached();
     _targetsSubscription = _targets.onChange.listen((_) => _r.dirty());
+    _r.enable();
   }
 
   @override
   void detached() {
-    super.detached(); children = []; _r.disable(notify: true);
-    _targetsSubscription.cancel(); _targetsSubscription = null;
+    super.detached();
+    children = [];
+    _r.disable(notify: true);
+    _targetsSubscription.cancel();
   }
 
   void render() {
@@ -158,7 +163,7 @@
       reader.readAsText(e.files[0]);
       reader.onLoad.listen((_) {
         var crashDump = JSON.decode(reader.result);
-        _dump.load(crashDump);
+        _loadDump(crashDump);
       });
     });
     return e;

runtime/observatory/lib/src/mocks/objects/class.dart

diff --git a/runtime/observatory/lib/src/mocks/objects/class.dart b/runtime/observatory/lib/src/mocks/objects/class.dart
deleted file mode 100644
index 33eb764..0000000
--- a/runtime/observatory/lib/src/mocks/objects/class.dart
+++ /dev/null
@@ -1,22 +0,0 @@
-// Copyright (c) 2016, the Dart project authors.  Please see the AUTHORS file
-// for details. All rights reserved. Use of this source code is governed by a
-// BSD-style license that can be found in the LICENSE file
-
-part of mocks;
-
-class ClassRefMock implements M.ClassRef {
-  final String id;
-  final String name;
-  const ClassRefMock({this.id, this.name});
-}
-
-class ClassMock implements M.ClassRef {
-  final String id;
-  final String name;
-  final bool isAbstract;
-  final bool isConst;
-  final M.ClassRef superclass;
-  final Iterable<M.ClassRef> subclasses;
-  const ClassMock({this.id, this.name, this.isAbstract, this.isConst,
-      this.superclass, this.subclasses: const []});
-}

runtime/observatory/lib/src/mocks/objects/vm.dart

diff --git a/runtime/observatory/lib/src/mocks/objects/vm.dart b/runtime/observatory/lib/src/mocks/objects/vm.dart
deleted file mode 100644
index a58272e..0000000
--- a/runtime/observatory/lib/src/mocks/objects/vm.dart
+++ /dev/null
@@ -1,23 +0,0 @@
-// Copyright (c) 2016, the Dart project authors.  Please see the AUTHORS file
-// for details. All rights reserved. Use of this source code is governed by a
-// BSD-style license that can be found in the LICENSE file
-
-part of mocks;
-
-class VMRefMock implements M.VMRef {
-  final String name;
-  const VMRefMock({this.name});
-}
-
-class VMMock implements M.VM {
-  final String name;
-  final int architectureBits;
-  final String targetCPU;
-  final String hostCPU;
-  final String version;
-  final int pid;
-  final DateTime startTime;
-  final Iterable<M.IsolateRef> isolates;
-  const VMMock({this.name, this.architectureBits, this.targetCPU, this.hostCPU,
-      this.version, this.pid, this.startTime, this.isolates : const []});
-}

runtime/observatory/lib/src/mocks/repositories/crash_dump.dart

diff --git a/runtime/observatory/lib/src/mocks/repositories/crash_dump.dart b/runtime/observatory/lib/src/mocks/repositories/crash_dump.dart
deleted file mode 100644
index 68488e9..0000000
--- a/runtime/observatory/lib/src/mocks/repositories/crash_dump.dart
+++ /dev/null
@@ -1,23 +0,0 @@
-// Copyright (c) 2016, the Dart project authors.  Please see the AUTHORS file
-// for details. All rights reserved. Use of this source code is governed by a
-// BSD-style license that can be found in the LICENSE file
-
-part of mocks;
-
-typedef void CrashDumpRepositoryMockCallback(Map);
-
-class CrashDumpRepositoryMock implements M.CrashDumpRepository {
-  final CrashDumpRepositoryMockCallback _load;
-
-  bool loadInvoked = false;
-
-  void load(Map dump) {
-    loadInvoked = true;
-    if (_load != null) {
-      _load(dump);
-    }
-  }
-
-  CrashDumpRepositoryMock({CrashDumpRepositoryMockCallback load})
-    : _load = load;
-}

runtime/observatory/lib/src/mocks/repositories/flag.dart

diff --git a/runtime/observatory/lib/src/mocks/repositories/flag.dart b/runtime/observatory/lib/src/mocks/repositories/flag.dart
deleted file mode 100644
index 1c9c11d..0000000
--- a/runtime/observatory/lib/src/mocks/repositories/flag.dart
+++ /dev/null
@@ -1,21 +0,0 @@
-// Copyright (c) 2016, the Dart project authors.  Please see the AUTHORS file
-// for details. All rights reserved. Use of this source code is governed by a
-// BSD-style license that can be found in the LICENSE file
-
-part of mocks;
-
-typedef Future<Iterable<M.Flag>> FlagsRepositoryMockCallback(M.VMRef vm);
-
-class FlagsRepositoryMock implements M.FlagsRepository {
-  final FlagsRepositoryMockCallback _list;
-
-  Future<Iterable<M.Flag>> list(M.VMRef vm) {
-    if (_list != null) {
-      return _list(vm);
-    }
-    return new Future.value(const []);
-  }
-
-  FlagsRepositoryMock({FlagsRepositoryMockCallback list})
-    : _list = list;
-}

runtime/observatory/lib/src/mocks/repositories/script.dart

diff --git a/runtime/observatory/lib/src/mocks/repositories/script.dart b/runtime/observatory/lib/src/mocks/repositories/script.dart
deleted file mode 100644
index 7ab9114..0000000
--- a/runtime/observatory/lib/src/mocks/repositories/script.dart
+++ /dev/null
@@ -1,19 +0,0 @@
-// Copyright (c) 2016, the Dart project authors.  Please see the AUTHORS file
-// for details. All rights reserved. Use of this source code is governed by a
-// BSD-style license that can be found in the LICENSE file.
-
-part of mocks;
-
-class ScriptRepositoryMock implements M.ScriptRepository {
-  final Map<String, M.Script> scripts;
-
-  bool _invoked = false;
-  bool get invoked => _invoked;
-
-  ScriptRepositoryMock(this.scripts);
-
-  Future<M.Script> get(String id) async {
-    _invoked = true;
-    return scripts[id];
-  }
-}

runtime/observatory/lib/src/models/exceptions.dart

diff --git a/runtime/observatory/lib/src/models/exceptions.dart b/runtime/observatory/lib/src/models/exceptions.dart
index 871c095..81a7ba7 100644
--- a/runtime/observatory/lib/src/models/exceptions.dart
+++ b/runtime/observatory/lib/src/models/exceptions.dart
@@ -1,6 +1,6 @@
 // Copyright (c) 2016, the Dart project authors.  Please see the AUTHORS file
 // for details. All rights reserved. Use of this source code is governed by a
-// BSD-style license that can be found in the LICENSE file
+// BSD-style license that can be found in the LICENSE file.
 
 part of models;
 

runtime/observatory/lib/src/models/objects/breakpoint.dart

diff --git a/runtime/observatory/lib/src/models/objects/breakpoint.dart b/runtime/observatory/lib/src/models/objects/breakpoint.dart
index e5e9324..daf1277 100644
--- a/runtime/observatory/lib/src/models/objects/breakpoint.dart
+++ b/runtime/observatory/lib/src/models/objects/breakpoint.dart
@@ -1,6 +1,6 @@
 // Copyright (c) 2016, the Dart project authors.  Please see the AUTHORS file
 // for details. All rights reserved. Use of this source code is governed by a
-// BSD-style license that can be found in the LICENSE file
+// BSD-style license that can be found in the LICENSE file.
 
 part of models;
 

runtime/observatory/lib/src/models/objects/class.dart

diff --git a/runtime/observatory/lib/src/models/objects/class.dart b/runtime/observatory/lib/src/models/objects/class.dart
index b9e5f50..ae8e241 100644
--- a/runtime/observatory/lib/src/models/objects/class.dart
+++ b/runtime/observatory/lib/src/models/objects/class.dart
@@ -1,6 +1,6 @@
 // Copyright (c) 2016, the Dart project authors.  Please see the AUTHORS file
 // for details. All rights reserved. Use of this source code is governed by a
-// BSD-style license that can be found in the LICENSE file
+// BSD-style license that can be found in the LICENSE file.
 
 part of models;
 
@@ -20,6 +20,9 @@
   /// Is this a const class?
   bool get isConst;
 
+  /// [internal]
+  bool get isPatch;
+
   /// The library which contains this class.
   LibraryRef get library;
 

runtime/observatory/lib/src/models/objects/code.dart

diff --git a/runtime/observatory/lib/src/models/objects/code.dart b/runtime/observatory/lib/src/models/objects/code.dart
index 1a8758e..ddc759d 100644
--- a/runtime/observatory/lib/src/models/objects/code.dart
+++ b/runtime/observatory/lib/src/models/objects/code.dart
@@ -1,6 +1,6 @@
 // Copyright (c) 2016, the Dart project authors.  Please see the AUTHORS file
 // for details. All rights reserved. Use of this source code is governed by a
-// BSD-style license that can be found in the LICENSE file
+// BSD-style license that can be found in the LICENSE file.
 
 part of models;
 

runtime/observatory/lib/src/models/objects/error.dart

diff --git a/runtime/observatory/lib/src/models/objects/error.dart b/runtime/observatory/lib/src/models/objects/error.dart
index b5d1bbd..23de456 100644
--- a/runtime/observatory/lib/src/models/objects/error.dart
+++ b/runtime/observatory/lib/src/models/objects/error.dart
@@ -1,6 +1,6 @@
 // Copyright (c) 2016, the Dart project authors.  Please see the AUTHORS file
 // for details. All rights reserved. Use of this source code is governed by a
-// BSD-style license that can be found in the LICENSE file
+// BSD-style license that can be found in the LICENSE file.
 
 part of models;
 

runtime/observatory/lib/src/models/objects/event.dart

diff --git a/runtime/observatory/lib/src/models/objects/event.dart b/runtime/observatory/lib/src/models/objects/event.dart
index aff1946..33dfc63 100644
--- a/runtime/observatory/lib/src/models/objects/event.dart
+++ b/runtime/observatory/lib/src/models/objects/event.dart
@@ -1,10 +1,14 @@
 // Copyright (c) 2016, the Dart project authors.  Please see the AUTHORS file
 // for details. All rights reserved. Use of this source code is governed by a
-// BSD-style license that can be found in the LICENSE file
+// BSD-style license that can be found in the LICENSE file.
 
 part of models;
 
 abstract class Event {
+  /// The timestamp (in milliseconds since the epoch) associated with this
+  /// event. For some isolate pause events, the timestamp is from when the
+  /// isolate was paused. For other events, the timestamp is from when the
+  /// event was created.
   DateTime get timestamp;
   static bool isPauseEvent(Event event) {
     return event is PauseStartEvent || event is PauseExitEvent ||
@@ -12,64 +16,130 @@
         event is PauseExceptionEvent || event is NoneEvent;
   }
 }
+
 abstract class VMEvent extends Event {
+  /// The vm with which this event is associated.
   VMRef get vm;
 }
+
 abstract class VMUpdateEvent extends VMEvent {}
+
 abstract class IsolateEvent extends Event {
+  /// The isolate with which this event is associated.
   IsolateRef get isolate;
 }
+
 abstract class IsolateStartEvent extends IsolateEvent {}
+
 abstract class IsolateRunnableEvent extends IsolateEvent {}
+
 abstract class IsolateExitEvent extends IsolateEvent {}
+
 abstract class IsolateUpdateEvent extends IsolateEvent {}
+
 abstract class IsolateReloadEvent extends IsolateEvent {
   ErrorRef get error;
 }
+
 abstract class ServiceExtensionAddedEvent extends IsolateEvent {
+  /// The RPC name of the extension that was added.
   String get extensionRPC;
 }
-abstract class DebugEvent extends IsolateEvent {}
+
+abstract class DebugEvent extends Event {
+  /// The isolate with which this event is associated.
+  IsolateRef get isolate;
+}
+
+abstract class DebuggerSettingsUpdateEvent extends DebugEvent {}
+
 abstract class PauseStartEvent extends DebugEvent {}
+
 abstract class PauseExitEvent extends DebugEvent {}
+
 abstract class PauseBreakpointEvent extends DebugEvent {
-  /// [optional]
+  /// [optional] The breakpoint at which we are currently paused.
   Breakpoint get breakpoint;
+  /// The list of breakpoints at which we are currently paused
+  /// for a PauseBreakpoint event.
+  ///
+  /// This list may be empty. For example, while single-stepping, the
+  /// VM sends a PauseBreakpoint event with no breakpoints.
+  ///
+  /// If there is more than one breakpoint set at the program position,
+  /// then all of them will be provided.
   Iterable<Breakpoint> get pauseBreakpoints;
+  /// The top stack frame associated with this event.
   Frame get topFrame;
   bool get atAsyncSuspension;
 }
+
 abstract class PauseInterruptedEvent extends DebugEvent {
+  /// [optional] The top stack frame associated with this event. There will be
+  /// no top frame if the isolate is idle (waiting in the message loop).
   Frame get topFrame;
+  /// Is the isolate paused at an await, yield, or yield* statement?
   bool get atAsyncSuspension;
 }
+
 abstract class PauseExceptionEvent extends DebugEvent {
+  /// The top stack frame associated with this event.
   Frame get topFrame;
+  /// The exception associated with this event
   InstanceRef get exception;
 }
-abstract class ResumeEvent extends DebugEvent {}
+
+abstract class ResumeEvent extends DebugEvent {
+  /// [optional] The top stack frame associated with this event. It is provided
+  /// at all times except for the initial resume event that is delivered when an
+  /// isolate begins execution.
+  Frame get topFrame;
+}
+
 abstract class BreakpointAddedEvent extends DebugEvent {
+  /// The breakpoint which was added.
   Breakpoint get breakpoint;
 }
+
 abstract class BreakpointResolvedEvent extends DebugEvent {
+  /// The breakpoint which was resolved.
   Breakpoint get breakpoint;
 }
+
 abstract class BreakpointRemovedEvent extends DebugEvent {
+  /// The breakpoint which was removed.
   Breakpoint get breakpoint;
 }
+
 abstract class InspectEvent extends DebugEvent {
+  /// The argument passed to dart:developer.inspect.
   InstanceRef get inspectee;
 }
+
 abstract class NoneEvent extends DebugEvent {}
-abstract class GCEvent extends IsolateEvent {}
-abstract class ExtensionEvent extends Event {
+
+abstract class GCEvent extends Event {
+  /// The isolate with which this event is associated.
   IsolateRef get isolate;
+}
+
+abstract class ExtensionEvent extends Event {
+  /// The isolate with which this event is associated.
+  IsolateRef get isolate;
+  /// The extension event kind.
   String get extensionKind;
+  /// The extension event data.
   ExtensionData get extensionData;
 }
-abstract class TimelineEventsEvent extends IsolateEvent {
+
+abstract class TimelineEventsEvent extends Event {
+  /// The isolate with which this event is associated.
+  IsolateRef get isolate;
+  /// An array of TimelineEvents
   Iterable<TimelineEvent> get timelineEvents;
 }
+
 abstract class ConnectionClosedEvent extends Event {
+  /// The reason of the closed connection
   String get reason;
 }

runtime/observatory/lib/src/models/objects/extension_data.dart

diff --git a/runtime/observatory/lib/src/models/objects/extension_data.dart b/runtime/observatory/lib/src/models/objects/extension_data.dart
index 5000a9b..80a0012 100644
--- a/runtime/observatory/lib/src/models/objects/extension_data.dart
+++ b/runtime/observatory/lib/src/models/objects/extension_data.dart
@@ -1,6 +1,6 @@
 // Copyright (c) 2016, the Dart project authors.  Please see the AUTHORS file
 // for details. All rights reserved. Use of this source code is governed by a
-// BSD-style license that can be found in the LICENSE file
+// BSD-style license that can be found in the LICENSE file.
 
 part of models;
 

runtime/observatory/lib/src/models/objects/frame.dart

diff --git a/runtime/observatory/lib/src/models/objects/frame.dart b/runtime/observatory/lib/src/models/objects/frame.dart
index d8e87cd..c0e40c3 100644
--- a/runtime/observatory/lib/src/models/objects/frame.dart
+++ b/runtime/observatory/lib/src/models/objects/frame.dart
@@ -1,6 +1,6 @@
 // Copyright (c) 2016, the Dart project authors.  Please see the AUTHORS file
 // for details. All rights reserved. Use of this source code is governed by a
-// BSD-style license that can be found in the LICENSE file
+// BSD-style license that can be found in the LICENSE file.
 
 part of models;
 

runtime/observatory/lib/src/models/objects/function.dart

diff --git a/runtime/observatory/lib/src/models/objects/function.dart b/runtime/observatory/lib/src/models/objects/function.dart
index bb763af..e52c564 100644
--- a/runtime/observatory/lib/src/models/objects/function.dart
+++ b/runtime/observatory/lib/src/models/objects/function.dart
@@ -1,6 +1,6 @@
 // Copyright (c) 2016, the Dart project authors.  Please see the AUTHORS file
 // for details. All rights reserved. Use of this source code is governed by a
-// BSD-style license that can be found in the LICENSE file
+// BSD-style license that can be found in the LICENSE file.
 
 part of models;
 

runtime/observatory/lib/src/models/objects/instance.dart

diff --git a/runtime/observatory/lib/src/models/objects/instance.dart b/runtime/observatory/lib/src/models/objects/instance.dart
index b337f79..eb65ea5 100644
--- a/runtime/observatory/lib/src/models/objects/instance.dart
+++ b/runtime/observatory/lib/src/models/objects/instance.dart
@@ -1,7 +1,81 @@
 // Copyright (c) 2016, the Dart project authors.  Please see the AUTHORS file
 // for details. All rights reserved. Use of this source code is governed by a
-// BSD-style license that can be found in the LICENSE file
+// BSD-style license that can be found in the LICENSE file.
 
 part of models;
 
-abstract class InstanceRef extends ObjectRef {}
+abstract class InstanceRef extends ObjectRef {
+  /// Instance references always include their class.
+  ClassRef get clazz;
+
+  /// [optional] The value of this instance as a string.
+  ///
+  /// Provided for the instance kinds:
+  ///   Null (null)
+  ///   Bool (true or false)
+  ///   Double (suitable for passing to Double.parse())
+  ///   Int (suitable for passing to int.parse())
+  ///   String (value may be truncated)
+  ///   Float32x4
+  ///   Float64x2
+  ///   Int32x4
+  ///   StackTrace
+  String get valueAsString;
+
+  /// [optional] The valueAsString for String references may be truncated. If so,
+  /// this property is added with the value 'true'.
+  ///
+  /// New code should use 'length' and 'count' instead.
+  bool get valueAsStringIsTruncated;
+
+  /// [optional] The length of a List or the number of associations in a Map or
+  /// the number of codeunits in a String.
+  ///
+  /// Provided for instance kinds:
+  ///   String
+  ///   List
+  ///   Map
+  ///   Uint8ClampedList
+  ///   Uint8List
+  ///   Uint16List
+  ///   Uint32List
+  ///   Uint64List
+  ///   Int8List
+  ///   Int16List
+  ///   Int32List
+  ///   Int64List
+  ///   Float32List
+  ///   Float64List
+  ///   Int32x4List
+  ///   Float32x4List
+  ///   Float64x2List
+  int get length;
+
+  /// [optional] The name of a Type instance.
+  ///
+  /// Provided for instance kinds:
+  ///   Type
+  String get name;
+
+  /// [optional] The corresponding Class if this Type is canonical.
+  ///
+  /// Provided for instance kinds:
+  ///   Type
+  ClassRef get typeClass;
+
+  /// [optional] The parameterized class of a type parameter:
+  ///
+  /// Provided for instance kinds:
+  ///   TypeParameter
+  ClassRef get parameterizedClass;
+
+  /// [optional] The pattern of a RegExp instance.
+  ///
+  /// The pattern is always an instance of kind String.
+  ///
+  /// Provided for instance kinds:
+  ///   RegExp
+  InstanceRef get pattern;
+}
+
+abstract class Instance extends Object implements InstanceRef {}

runtime/observatory/lib/src/models/objects/isolate.dart

diff --git a/runtime/observatory/lib/src/models/objects/isolate.dart b/runtime/observatory/lib/src/models/objects/isolate.dart
index 007a6f4..81d54c5 100644
--- a/runtime/observatory/lib/src/models/objects/isolate.dart
+++ b/runtime/observatory/lib/src/models/objects/isolate.dart
@@ -1,6 +1,6 @@
 // Copyright (c) 2016, the Dart project authors.  Please see the AUTHORS file
 // for details. All rights reserved. Use of this source code is governed by a
-// BSD-style license that can be found in the LICENSE file
+// BSD-style license that can be found in the LICENSE file.
 
 part of models;
 

runtime/observatory/lib/src/models/objects/library.dart

diff --git a/runtime/observatory/lib/src/models/objects/library.dart b/runtime/observatory/lib/src/models/objects/library.dart
index 6543ef9..281242a 100644
--- a/runtime/observatory/lib/src/models/objects/library.dart
+++ b/runtime/observatory/lib/src/models/objects/library.dart
@@ -1,6 +1,6 @@
 // Copyright (c) 2016, the Dart project authors.  Please see the AUTHORS file
 // for details. All rights reserved. Use of this source code is governed by a
-// BSD-style license that can be found in the LICENSE file
+// BSD-style license that can be found in the LICENSE file.
 
 part of models;
 

runtime/observatory/lib/src/models/objects/notification.dart

diff --git a/runtime/observatory/lib/src/models/objects/notification.dart b/runtime/observatory/lib/src/models/objects/notification.dart
index 9a9f8af..3a2ae9a 100644
--- a/runtime/observatory/lib/src/models/objects/notification.dart
+++ b/runtime/observatory/lib/src/models/objects/notification.dart
@@ -1,6 +1,6 @@
 // Copyright (c) 2016, the Dart project authors.  Please see the AUTHORS file
 // for details. All rights reserved. Use of this source code is governed by a
-// BSD-style license that can be found in the LICENSE file
+// BSD-style license that can be found in the LICENSE file.
 
 part of models;
 

runtime/observatory/lib/src/models/objects/object.dart

diff --git a/runtime/observatory/lib/src/models/objects/object.dart b/runtime/observatory/lib/src/models/objects/object.dart
index 76b2393..829a60d2 100644
--- a/runtime/observatory/lib/src/models/objects/object.dart
+++ b/runtime/observatory/lib/src/models/objects/object.dart
@@ -1,6 +1,6 @@
 // Copyright (c) 2016, the Dart project authors.  Please see the AUTHORS file
 // for details. All rights reserved. Use of this source code is governed by a
-// BSD-style license that can be found in the LICENSE file
+// BSD-style license that can be found in the LICENSE file.
 
 part of models;
 

runtime/observatory/lib/src/models/objects/script.dart

diff --git a/runtime/observatory/lib/src/models/objects/script.dart b/runtime/observatory/lib/src/models/objects/script.dart
index 274fb08..3736651 100644
--- a/runtime/observatory/lib/src/models/objects/script.dart
+++ b/runtime/observatory/lib/src/models/objects/script.dart
@@ -1,6 +1,6 @@
 // Copyright (c) 2016, the Dart project authors.  Please see the AUTHORS file
 // for details. All rights reserved. Use of this source code is governed by a
-// BSD-style license that can be found in the LICENSE file
+// BSD-style license that can be found in the LICENSE file.
 
 part of models;
 

runtime/observatory/lib/src/models/objects/source_location.dart

diff --git a/runtime/observatory/lib/src/models/objects/source_location.dart b/runtime/observatory/lib/src/models/objects/source_location.dart
index 5d35d57..dd9f663 100644
--- a/runtime/observatory/lib/src/models/objects/source_location.dart
+++ b/runtime/observatory/lib/src/models/objects/source_location.dart
@@ -1,6 +1,6 @@
 // Copyright (c) 2016, the Dart project authors.  Please see the AUTHORS file
 // for details. All rights reserved. Use of this source code is governed by a
-// BSD-style license that can be found in the LICENSE file
+// BSD-style license that can be found in the LICENSE file.
 
 part of models;
 

runtime/observatory/lib/src/models/objects/target.dart

diff --git a/runtime/observatory/lib/src/models/objects/target.dart b/runtime/observatory/lib/src/models/objects/target.dart
index f2a1658..a58673f 100644
--- a/runtime/observatory/lib/src/models/objects/target.dart
+++ b/runtime/observatory/lib/src/models/objects/target.dart
@@ -1,6 +1,6 @@
 // Copyright (c) 2016, the Dart project authors.  Please see the AUTHORS file
 // for details. All rights reserved. Use of this source code is governed by a
-// BSD-style license that can be found in the LICENSE file
+// BSD-style license that can be found in the LICENSE file.
 
 part of models;
 

runtime/observatory/lib/src/models/objects/timeline_event.dart

diff --git a/runtime/observatory/lib/src/models/objects/timeline_event.dart b/runtime/observatory/lib/src/models/objects/timeline_event.dart
index 6cac6d1..61bf444 100644
--- a/runtime/observatory/lib/src/models/objects/timeline_event.dart
+++ b/runtime/observatory/lib/src/models/objects/timeline_event.dart
@@ -1,6 +1,6 @@
 // Copyright (c) 2016, the Dart project authors.  Please see the AUTHORS file
 // for details. All rights reserved. Use of this source code is governed by a
-// BSD-style license that can be found in the LICENSE file
+// BSD-style license that can be found in the LICENSE file.
 
 part of models;
 

runtime/observatory/lib/src/models/objects/vm.dart

diff --git a/runtime/observatory/lib/src/models/objects/vm.dart b/runtime/observatory/lib/src/models/objects/vm.dart
index 051a9fa..11e3af5 100644
--- a/runtime/observatory/lib/src/models/objects/vm.dart
+++ b/runtime/observatory/lib/src/models/objects/vm.dart
@@ -1,11 +1,16 @@
 // Copyright (c) 2016, the Dart project authors.  Please see the AUTHORS file
 // for details. All rights reserved. Use of this source code is governed by a
-// BSD-style license that can be found in the LICENSE file
+// BSD-style license that can be found in the LICENSE file.
 
 part of models;
 
 abstract class VMRef {
+  /// A name identifying this vm. Not guaranteed to be unique.
   String get name;
+
+  /// [Not actually from the apis]
+  /// A name used to identify the VM in the UI.
+  String get displayName;
 }
 
 abstract class VM implements VMRef {

runtime/observatory/lib/src/models/repositories/class.dart

diff --git a/runtime/observatory/lib/src/models/repositories/crash_dump.dart b/runtime/observatory/lib/src/models/repositories/class.dart
similarity index 71%
copy from runtime/observatory/lib/src/models/repositories/crash_dump.dart
copy to runtime/observatory/lib/src/models/repositories/class.dart
index 8eb64c1..388256e 100644
--- a/runtime/observatory/lib/src/models/repositories/crash_dump.dart
+++ b/runtime/observatory/lib/src/models/repositories/class.dart
@@ -4,6 +4,7 @@
 
 part of models;
 
-abstract class CrashDumpRepository {
-  void load(Map);
+abstract class ClassRepository{
+  Future<Class> getObject();
+  Future<Class> get(String id);
 }

runtime/observatory/lib/src/models/repositories/event.dart

diff --git a/runtime/observatory/lib/src/models/repositories/event.dart b/runtime/observatory/lib/src/models/repositories/event.dart
new file mode 100644
index 0000000..3e00359
--- /dev/null
+++ b/runtime/observatory/lib/src/models/repositories/event.dart
@@ -0,0 +1,33 @@
+// Copyright (c) 2016, the Dart project authors.  Please see the AUTHORS file
+// for details. All rights reserved. Use of this source code is governed by a
+// BSD-style license that can be found in the LICENSE file.
+
+part of models;
+
+abstract class EventRepository {
+  Stream<Event> get onEvent;
+  Stream<VMEvent> get onVMEvent;
+  Stream<VMUpdateEvent> get onVMUpdate;
+  Stream<IsolateEvent> get onIsolateEvent;
+  Stream<IsolateStartEvent> get onIsolateStart;
+  Stream<IsolateRunnableEvent> get onIsolateRunnable;
+  Stream<IsolateExitEvent> get onIsolateExit;
+  Stream<IsolateUpdateEvent> get onIsolateUpdate;
+  Stream<IsolateReloadEvent> get onIsolateReload;
+  Stream<ServiceExtensionAddedEvent> get onServiceExtensionAdded;
+  Stream<DebugEvent> get onDebugEvent;
+  Stream<PauseStartEvent> get onPauseStart;
+  Stream<PauseExitEvent> get onPauseExit;
+  Stream<PauseBreakpointEvent> get onPauseBreakpoint;
+  Stream<PauseInterruptedEvent> get onPauseInterrupted;
+  Stream<PauseExceptionEvent> get onPauseException;
+  Stream<ResumeEvent> get onResume;
+  Stream<BreakpointAddedEvent> get onBreakpointAdded;
+  Stream<BreakpointResolvedEvent> get onBreakpointResolved;
+  Stream<BreakpointRemovedEvent> get onBreakpointRemoved;
+  Stream<InspectEvent> get onInspect;
+  Stream<GCEvent> get onGCEvent;
+  Stream<ExtensionEvent> get onExtensionEvent;
+  Stream<TimelineEventsEvent> get onTimelineEvents;
+  Stream<ConnectionClosedEvent> get onConnectionClosed;
+}

runtime/observatory/lib/src/models/repositories/flag.dart

diff --git a/runtime/observatory/lib/src/models/repositories/flag.dart b/runtime/observatory/lib/src/models/repositories/flag.dart
index 2bdc018..358b933 100644
--- a/runtime/observatory/lib/src/models/repositories/flag.dart
+++ b/runtime/observatory/lib/src/models/repositories/flag.dart
@@ -5,5 +5,5 @@
 part of models;
 
 abstract class FlagsRepository {
-  Future<Iterable<Flag>> list(VMRef vm);
+  Future<Iterable<Flag>> list();
 }

runtime/observatory/lib/src/models/repositories/instance.dart

diff --git a/runtime/observatory/lib/src/models/repositories/crash_dump.dart b/runtime/observatory/lib/src/models/repositories/instance.dart
similarity index 76%
rename from runtime/observatory/lib/src/models/repositories/crash_dump.dart
rename to runtime/observatory/lib/src/models/repositories/instance.dart
index 8eb64c1..3f98018 100644
--- a/runtime/observatory/lib/src/models/repositories/crash_dump.dart
+++ b/runtime/observatory/lib/src/models/repositories/instance.dart
@@ -4,6 +4,6 @@
 
 part of models;
 
-abstract class CrashDumpRepository {
-  void load(Map);
+abstract class InstanceRepository{
+  Future<Instance> get(String id);
 }

runtime/observatory/lib/src/models/repositories/notification.dart

diff --git a/runtime/observatory/lib/src/models/repositories/notification.dart b/runtime/observatory/lib/src/models/repositories/notification.dart
index 9ab3543..816bf04 100644
--- a/runtime/observatory/lib/src/models/repositories/notification.dart
+++ b/runtime/observatory/lib/src/models/repositories/notification.dart
@@ -1,6 +1,6 @@
 // Copyright (c) 2016, the Dart project authors.  Please see the AUTHORS file
 // for details. All rights reserved. Use of this source code is governed by a
-// BSD-style license that can be found in the LICENSE file
+// BSD-style license that can be found in the LICENSE file.
 
 part of models;
 

runtime/observatory/lib/src/models/repositories/script.dart

diff --git a/runtime/observatory/lib/src/models/repositories/script.dart b/runtime/observatory/lib/src/models/repositories/script.dart
index e3b54ad..3073c2e 100644
--- a/runtime/observatory/lib/src/models/repositories/script.dart
+++ b/runtime/observatory/lib/src/models/repositories/script.dart
@@ -4,4 +4,6 @@
 
 part of models;
 
-abstract class ScriptRepository extends Repository<Script, String> {}
+abstract class ScriptRepository {
+  Future<Script> get(String id);
+}

runtime/observatory/lib/src/models/repositories/target.dart

diff --git a/runtime/observatory/lib/src/models/repositories/target.dart b/runtime/observatory/lib/src/models/repositories/target.dart
index 8df29ca..4d276f6 100644
--- a/runtime/observatory/lib/src/models/repositories/target.dart
+++ b/runtime/observatory/lib/src/models/repositories/target.dart
@@ -1,6 +1,6 @@
 // Copyright (c) 2016, the Dart project authors.  Please see the AUTHORS file
 // for details. All rights reserved. Use of this source code is governed by a
-// BSD-style license that can be found in the LICENSE file
+// BSD-style license that can be found in the LICENSE file.
 
 part of models;
 

runtime/observatory/lib/src/models/repository.dart

diff --git a/runtime/observatory/lib/src/models/repository.dart b/runtime/observatory/lib/src/models/repository.dart
deleted file mode 100644
index 9aec1fd..0000000
--- a/runtime/observatory/lib/src/models/repository.dart
+++ /dev/null
@@ -1,9 +0,0 @@
-// Copyright (c) 2016, the Dart project authors.  Please see the AUTHORS file
-// for details. All rights reserved. Use of this source code is governed by a
-// BSD-style license that can be found in the LICENSE file.
-
-part of models;
-
-abstract class Repository<ObjectT, DescriptorT> {
-  Future<ObjectT> get(DescriptorT id);
-}

runtime/observatory/lib/src/repositories/class.dart

diff --git a/runtime/observatory/lib/src/repositories/class.dart b/runtime/observatory/lib/src/repositories/class.dart
new file mode 100644
index 0000000..911c363
--- /dev/null
+++ b/runtime/observatory/lib/src/repositories/class.dart
@@ -0,0 +1,18 @@
+// Copyright (c) 2016, the Dart project authors.  Please see the AUTHORS file
+// for details. All rights reserved. Use of this source code is governed by a
+// BSD-style license that can be found in the LICENSE file
+
+part of repositories;
+
+class ClassRepository extends M.ClassRepository {
+  final S.Isolate isolate;
+
+  ClassRepository(this.isolate);
+
+  Future<M.Class> getObject() {
+    return isolate.getClassHierarchy();
+  }
+  Future<M.Class> get(String id) async {
+    return (await isolate.getObject(id)) as S.Class;
+  }
+}

runtime/observatory/lib/src/repositories/event.dart

diff --git a/runtime/observatory/lib/src/repositories/event.dart b/runtime/observatory/lib/src/repositories/event.dart
new file mode 100644
index 0000000..d6300e9
--- /dev/null
+++ b/runtime/observatory/lib/src/repositories/event.dart
@@ -0,0 +1,89 @@
+// Copyright (c) 2016, the Dart project authors.  Please see the AUTHORS file
+// for details. All rights reserved. Use of this source code is governed by a
+// BSD-style license that can be found in the LICENSE file.
+
+part of repositories;
+
+class EventRepository implements M.EventRepository {
+  final StreamController<M.Event> _onEvent;
+  Stream<M.Event> get onEvent => _onEvent.stream;
+
+  final Stream<M.VMEvent> onVMEvent;
+  final Stream<M.VMUpdateEvent> onVMUpdate;
+  final Stream<M.IsolateEvent> onIsolateEvent;
+  final Stream<M.IsolateStartEvent> onIsolateStart;
+  final Stream<M.IsolateRunnableEvent> onIsolateRunnable;
+  final Stream<M.IsolateExitEvent> onIsolateExit;
+  final Stream<M.IsolateUpdateEvent> onIsolateUpdate;
+  final Stream<M.IsolateReloadEvent> onIsolateReload;
+  final Stream<M.ServiceExtensionAddedEvent> onServiceExtensionAdded;
+  final Stream<M.DebugEvent> onDebugEvent;
+  final Stream<M.PauseStartEvent> onPauseStart;
+  final Stream<M.PauseExitEvent> onPauseExit;
+  final Stream<M.PauseBreakpointEvent> onPauseBreakpoint;
+  final Stream<M.PauseInterruptedEvent> onPauseInterrupted;
+  final Stream<M.PauseExceptionEvent> onPauseException;
+  final Stream<M.ResumeEvent> onResume;
+  final Stream<M.BreakpointAddedEvent> onBreakpointAdded;
+  final Stream<M.BreakpointResolvedEvent> onBreakpointResolved;
+  final Stream<M.BreakpointRemovedEvent> onBreakpointRemoved;
+  final Stream<M.InspectEvent> onInspect;
+  final Stream<M.GCEvent> onGCEvent;
+  final Stream<M.ExtensionEvent> onExtensionEvent;
+  final Stream<M.TimelineEventsEvent> onTimelineEvents;
+  final Stream<M.ConnectionClosedEvent> onConnectionClosed;
+
+  EventRepository() : this._(new StreamController.broadcast());
+
+  EventRepository._(StreamController controller) : this.__(controller,
+      controller.stream.where((e) => e is M.VMEvent),
+      controller.stream.where((e) => e is M.IsolateEvent),
+      controller.stream.where((e) => e is M.DebugEvent),
+      controller.stream.where((e) => e is M.GCEvent),
+      controller.stream.where((e) => e is M.ExtensionEvent),
+      controller.stream.where((e) => e is M.TimelineEventsEvent),
+      controller.stream.where((e) => e is M.ConnectionClosedEvent));
+
+  EventRepository.__(StreamController controller,
+    Stream<M.VMEvent> onVMEvent, Stream<M.IsolateEvent> onIsolateEvent,
+    Stream<M.DebugEvent> onDebugEvent, Stream<M.GCEvent> onGCEvent,
+    Stream<M.ExtensionEvent> onExtensionEvent,
+    Stream<M.TimelineEventsEvent> onTimelineEvents,
+    Stream<M.ConnectionClosedEvent> onConnectionClosed)
+    : _onEvent = controller,
+      onVMEvent = onVMEvent,
+      onVMUpdate = onVMEvent.where((e) => e is M.VMUpdateEvent),
+      onIsolateEvent = onIsolateEvent,
+      onIsolateStart = onIsolateEvent.where((e) => e is M.IsolateStartEvent),
+      onIsolateRunnable =
+          onIsolateEvent.where((e) => e is M.IsolateRunnableEvent),
+      onIsolateExit = onIsolateEvent.where((e) => e is M.IsolateExitEvent),
+      onIsolateUpdate = onIsolateEvent.where((e) => e is M.IsolateUpdateEvent),
+      onIsolateReload = onIsolateEvent.where((e) => e is M.IsolateReloadEvent),
+      onServiceExtensionAdded =
+          onIsolateEvent.where((e) => e is M.IsolateReloadEvent),
+      onDebugEvent = onDebugEvent,
+      onPauseStart = onDebugEvent.where((e) => e is M.PauseStartEvent),
+      onPauseExit = onDebugEvent.where((e) => e is M.PauseExitEvent),
+      onPauseBreakpoint =
+          onDebugEvent.where((e) => e is M.PauseBreakpointEvent),
+      onPauseInterrupted =
+          onDebugEvent.where((e) => e is M.PauseInterruptedEvent),
+      onPauseException = onDebugEvent.where((e) => e is M.PauseExceptionEvent),
+      onResume = onDebugEvent.where((e) => e is M.ResumeEvent),
+      onBreakpointAdded =
+          onDebugEvent.where((e) => e is M.BreakpointAddedEvent),
+      onBreakpointResolved =
+          onDebugEvent.where((e) => e is M.BreakpointResolvedEvent),
+      onBreakpointRemoved =
+          onDebugEvent.where((e) => e is M.BreakpointRemovedEvent),
+      onInspect = onDebugEvent.where((e) => e is M.InspectEvent),
+      onGCEvent = onGCEvent,
+      onExtensionEvent = onExtensionEvent,
+      onTimelineEvents = onTimelineEvents,
+      onConnectionClosed = onConnectionClosed;
+
+  void add(M.Event e) {
+    _onEvent.add(e);
+  }
+}

runtime/observatory/lib/src/repositories/flag.dart

diff --git a/runtime/observatory/lib/src/repositories/flag.dart b/runtime/observatory/lib/src/repositories/flag.dart
index 9a644e6..0072c0e 100644
--- a/runtime/observatory/lib/src/repositories/flag.dart
+++ b/runtime/observatory/lib/src/repositories/flag.dart
@@ -17,12 +17,13 @@
 }
 
 class FlagsRepository implements M.FlagsRepository {
-  Future<Iterable<Flag>> list(M.VM vm) async{
-    if (vm is S.VM) {
-      List<Map> flags = ((await vm.getFlagList()) as S.ServiceMap)['flags'];
-      return flags.map(_toFlag);
-    }
-    return const [];
+  final S.VM vm;
+
+  FlagsRepository(this.vm);
+
+  Future<Iterable<Flag>> list() async{
+    List<Map> flags = ((await vm.getFlagList()) as S.ServiceMap)['flags'];
+    return flags.map(_toFlag);
   }
 
   static _toFlag(Map map){

runtime/observatory/lib/src/repositories/instance.dart

diff --git a/runtime/observatory/lib/src/repositories/instance.dart b/runtime/observatory/lib/src/repositories/instance.dart
new file mode 100644
index 0000000..6a3fe2b
--- /dev/null
+++ b/runtime/observatory/lib/src/repositories/instance.dart
@@ -0,0 +1,15 @@
+// Copyright (c) 2016, the Dart project authors.  Please see the AUTHORS file
+// for details. All rights reserved. Use of this source code is governed by a
+// BSD-style license that can be found in the LICENSE file
+
+part of repositories;
+
+class InstanceRepository extends M.InstanceRepository {
+  final S.Isolate isolate;
+
+  InstanceRepository(this.isolate);
+
+  Future<M.Instance> get(String id) async{
+    return (await isolate.getObject(id)) as S.Instance;
+  }
+}

runtime/observatory/lib/src/repositories/notification.dart

diff --git a/runtime/observatory/lib/src/repositories/notification.dart b/runtime/observatory/lib/src/repositories/notification.dart
index 348dbca..80c25f2 100644
--- a/runtime/observatory/lib/src/repositories/notification.dart
+++ b/runtime/observatory/lib/src/repositories/notification.dart
@@ -1,6 +1,6 @@
 // Copyright (c) 2016, the Dart project authors.  Please see the AUTHORS file
 // for details. All rights reserved. Use of this source code is governed by a
-// BSD-style license that can be found in the LICENSE file
+// BSD-style license that can be found in the LICENSE file.
 
 part of repositories;
 

runtime/observatory/lib/src/repositories/settings.dart

diff --git a/runtime/observatory/lib/src/repositories/settings.dart b/runtime/observatory/lib/src/repositories/settings.dart
index 771c48e..4121d93 100644
--- a/runtime/observatory/lib/src/repositories/settings.dart
+++ b/runtime/observatory/lib/src/repositories/settings.dart
@@ -1,6 +1,6 @@
 // Copyright (c) 2016, the Dart project authors.  Please see the AUTHORS file
 // for details. All rights reserved. Use of this source code is governed by a
-// BSD-style license that can be found in the LICENSE file
+// BSD-style license that can be found in the LICENSE file.
 
 part of repositories;
 

runtime/observatory/lib/src/repositories/target.dart

diff --git a/runtime/observatory/lib/src/repositories/target.dart b/runtime/observatory/lib/src/repositories/target.dart
index 69ba8c8..b9b2e69 100644
--- a/runtime/observatory/lib/src/repositories/target.dart
+++ b/runtime/observatory/lib/src/repositories/target.dart
@@ -1,6 +1,6 @@
 // Copyright (c) 2016, the Dart project authors.  Please see the AUTHORS file
 // for details. All rights reserved. Use of this source code is governed by a
-// BSD-style license that can be found in the LICENSE file
+// BSD-style license that can be found in the LICENSE file.
 
 part of repositories;
 

runtime/observatory/lib/src/service/object.dart

diff --git a/runtime/observatory/lib/src/service/object.dart b/runtime/observatory/lib/src/service/object.dart
index d2f44d9..afd9240 100644
--- a/runtime/observatory/lib/src/service/object.dart
+++ b/runtime/observatory/lib/src/service/object.dart
@@ -927,6 +927,8 @@
 }
 
 class FakeVM extends VM {
+  String get displayName => name;
+
   final Map _responses = {};
   FakeVM(Map responses) {
     if (responses == null) {
@@ -1851,16 +1853,29 @@
   String toString() => "ServiceMap($_map)";
 }
 
+M.ErrorKind stringToErrorKind(String value) {
+  switch(value) {
+    case 'UnhandledException': return M.ErrorKind.UnhandledException;
+    case 'LanguageError': return M.ErrorKind.UnhandledException;
+    case 'InternalError': return M.ErrorKind.InternalError;
+    case 'TerminationError': return M.ErrorKind.TerminationError;
+  }
+  Logger.root.severe('Unrecognized error kind: $value');
+  throw new FallThroughError();
+}
+
 /// A [DartError] is peered to a Dart Error object.
-class DartError extends ServiceObject {
+class DartError extends ServiceObject implements M.Error {
   DartError._empty(ServiceObject owner) : super._empty(owner);
 
+  M.ErrorKind kind;
   @observable String message;
   @observable Instance exception;
   @observable Instance stacktrace;
 
   void _update(ObservableMap map, bool mapIsRef) {
     message = map['message'];
+    kind = stringToErrorKind(map['kind']);
     exception = new ServiceObject._fromMap(owner, map['exception']);
     stacktrace = new ServiceObject._fromMap(owner, map['stacktrace']);
     name = 'DartError($message)';
@@ -1916,13 +1931,15 @@
 
   @observable String kind;
   @observable DateTime timestamp;
+  List<M.Breakpoint> pauseBreakpoints;
   @observable Breakpoint breakpoint;
   @observable Frame topFrame;
+  @observable DartError error;
   @observable String extensionRPC;
   @observable Instance exception;
   @observable Instance reloadError;
   @observable bool atAsyncSuspension;
-  @observable ServiceObject inspectee;
+  @observable Instance inspectee;
   @observable ByteData data;
   @observable int count;
   @observable String reason;
@@ -1960,13 +1977,16 @@
     if (map['breakpoint'] != null) {
       breakpoint = map['breakpoint'];
     }
-    // TODO(turnidge): Expose the full list of breakpoints.  For now
-    // we just pretend like there is only one active breakpoint.
     if (map['pauseBreakpoints'] != null) {
-      var pauseBpts = map['pauseBreakpoints'];
-      if (pauseBpts.length > 0) {
-        breakpoint = pauseBpts[0];
+      pauseBreakpoints = map['pauseBreakpoints'];
+      if (pauseBreakpoints.length > 0) {
+        breakpoint = pauseBreakpoints[0];
       }
+    } else {
+      pauseBreakpoints = const [];
+    }
+    if (map['error'] != null) {
+      error = map['error'];
     }
     if (map['extensionRPC'] != null) {
       extensionRPC = map['extensionRPC'];
@@ -2232,7 +2252,7 @@
   bool get empty => accumulated.empty && current.empty;
 }
 
-class Class extends HeapObject implements M.ClassRef {
+class Class extends HeapObject implements M.Class {
   @observable Library library;
 
   @observable bool isAbstract;
@@ -2243,7 +2263,7 @@
 
   @observable SourceLocation location;
 
-  @observable ServiceMap error;
+  @observable DartError error;
   @observable int vmCid;
 
   final Allocations newSpace = new Allocations();
@@ -2370,7 +2390,7 @@
   String toString() => 'Class($vmName)';
 }
 
-class Instance extends HeapObject {
+class Instance extends HeapObject implements M.Instance {
   @observable String kind;
   @observable String valueAsString;  // If primitive.
   @observable bool valueAsStringIsTruncated;
@@ -2469,7 +2489,7 @@
   Instance._empty(ServiceObjectOwner owner) : super._empty(owner);
 
   void _update(ObservableMap map, bool mapIsRef) {
-    // Extract full properties.
+    // Extract full properties.1
     _upgradeCollection(map, isolate);
     super._update(map, mapIsRef);
 
@@ -2482,6 +2502,7 @@
     name = map['name'];
     length = map['length'];
     pattern = map['pattern'];
+    typeClass = map['typeClass'];
 
     if (mapIsRef) {
       return;
@@ -2534,7 +2555,6 @@
           typedElements = bytes.buffer.asFloat64x2List(); break;
       }
     }
-    typeClass = map['typeClass'];
     parameterizedClass = map['parameterizedClass'];
     typeArguments = map['typeArguments'];
     parameterIndex = map['parameterIndex'];

runtime/observatory/observatory_sources.gypi

diff --git a/runtime/observatory/observatory_sources.gypi b/runtime/observatory/observatory_sources.gypi
index adf1c26..fe89855 100644
--- a/runtime/observatory/observatory_sources.gypi
+++ b/runtime/observatory/observatory_sources.gypi
@@ -11,7 +11,6 @@
     'lib/debugger.dart',
     'lib/elements.dart',
     'lib/elements.html',
-    'lib/mocks.dart',
     'lib/models.dart',
     'lib/object_graph.dart',
     'lib/repositories.dart',
@@ -21,6 +20,7 @@
     'lib/service_io.dart',
     'lib/src/app/analytics.dart',
     'lib/src/app/application.dart',
+    'lib/src/app/event.dart',
     'lib/src/app/location_manager.dart',
     'lib/src/app/notification.dart',
     'lib/src/app/page.dart',
@@ -37,7 +37,6 @@
     'lib/src/elements/class_ref_as_value.dart',
     'lib/src/elements/class_ref_as_value.html',
     'lib/src/elements/class_tree.dart',
-    'lib/src/elements/class_tree.html',
     'lib/src/elements/class_view.dart',
     'lib/src/elements/class_view.html',
     'lib/src/elements/code_ref.dart',
@@ -178,25 +177,6 @@
     'lib/src/elements/vm_connect.dart',
     'lib/src/elements/vm_view.dart',
     'lib/src/elements/vm_view.html',
-    'lib/src/mocks/exceptions/connection_exception.dart',
-    'lib/src/mocks/objects/class.dart',
-    'lib/src/mocks/objects/code.dart',
-    'lib/src/mocks/objects/error.dart',
-    'lib/src/mocks/objects/event.dart',
-    'lib/src/mocks/objects/flag.dart',
-    'lib/src/mocks/objects/function.dart',
-    'lib/src/mocks/objects/isolate.dart',
-    'lib/src/mocks/objects/library.dart',
-    'lib/src/mocks/objects/notification.dart',
-    'lib/src/mocks/objects/script.dart',
-    'lib/src/mocks/objects/source_location.dart',
-    'lib/src/mocks/objects/target.dart',
-    'lib/src/mocks/objects/vm.dart',
-    'lib/src/mocks/repositories/crash_dump.dart',
-    'lib/src/mocks/repositories/flag.dart',
-    'lib/src/mocks/repositories/script.dart',
-    'lib/src/mocks/repositories/notification.dart',
-    'lib/src/mocks/repositories/target.dart',
     'lib/src/models/exceptions.dart',
     'lib/src/models/objects/breakpoint.dart',
     'lib/src/models/objects/class.dart',
@@ -217,13 +197,17 @@
     'lib/src/models/objects/target.dart',
     'lib/src/models/objects/timeline_event.dart',
     'lib/src/models/objects/vm.dart',
-    'lib/src/models/repositories/crash_dump.dart',
+    'lib/src/models/repositories/class.dart',
+    'lib/src/models/repositories/event.dart',
     'lib/src/models/repositories/flag.dart',
+    'lib/src/models/repositories/instance.dart',
     'lib/src/models/repositories/notification.dart',
     'lib/src/models/repositories/script.dart',
     'lib/src/models/repositories/target.dart',
-    'lib/src/models/repository.dart',
+    'lib/src/repositories/class.dart',
+    'lib/src/repositories/event.dart',
     'lib/src/repositories/flag.dart',
+    'lib/src/repositories/instance.dart',
     'lib/src/repositories/notification.dart',
     'lib/src/repositories/script.dart',
     'lib/src/repositories/settings.dart',

runtime/observatory/tests/observatory_ui/class_ref/element_test.dart

diff --git a/runtime/observatory/tests/observatory_ui/class_ref/element_test.dart b/runtime/observatory/tests/observatory_ui/class_ref/element_test.dart
index ef812fd..0f0eed2 100644
--- a/runtime/observatory/tests/observatory_ui/class_ref/element_test.dart
+++ b/runtime/observatory/tests/observatory_ui/class_ref/element_test.dart
@@ -3,8 +3,8 @@
 // BSD-style license that can be found in the LICENSE file.
 import 'dart:html';
 import 'package:unittest/unittest.dart';
-import 'package:observatory/mocks.dart';
 import 'package:observatory/src/elements/class_ref.dart';
+import '../mocks.dart';
 
 main(){
   ClassRefElement.tag.ensureRegistration();

runtime/observatory/tests/observatory_ui/class_tree/element_test.dart

diff --git a/runtime/observatory/tests/observatory_ui/class_tree/element_test.dart b/runtime/observatory/tests/observatory_ui/class_tree/element_test.dart
new file mode 100644
index 0000000..518613f
--- /dev/null
+++ b/runtime/observatory/tests/observatory_ui/class_tree/element_test.dart
@@ -0,0 +1,64 @@
+// Copyright (c) 2016, the Dart project authors.  Please see the AUTHORS file
+// for details. All rights reserved. Use of this source code is governed by a
+// BSD-style license that can be found in the LICENSE file.
+import 'dart:html';
+import 'package:unittest/unittest.dart';
+import 'package:observatory/src/elements/class_tree.dart';
+import 'package:observatory/src/elements/nav/notify.dart';
+import '../mocks.dart';
+
+main() {
+  ClassTreeElement.tag.ensureRegistration();
+
+  final nTag = NavNotifyElement.tag.name;
+  const vm = const VMMock();
+  const isolate = const IsolateRefMock();
+  final events = new EventRepositoryMock();
+  final notifications = new NotificationRepositoryMock();
+
+  group('instantiation', () {
+    test('default', () {
+      final e = new ClassTreeElement(vm, isolate, events, notifications,
+          new ClassRepositoryMock());
+      expect(e, isNotNull, reason: 'element correctly created');
+    });
+  });
+  group('elements', () {
+    test('created after attachment', () async {
+      const child2_id = 'c2-id';
+      const child1_1_id = 'c1_1-id';
+      const child1_id = 'c1-id';
+      const child2 = const ClassMock(id: child2_id);
+      const child1_1 = const ClassMock(id: child1_1_id);
+      const child1 = const ClassMock(id: child1_id,
+                                     subclasses: const [child1_1]);
+      const object = const ClassMock(id: 'o-id',
+                                     subclasses: const [child1, child2]);
+      const ids = const [child1_id, child1_1_id, child2_id];
+      bool rendered = false;
+      final e = new ClassTreeElement(vm, isolate, events, notifications,
+          new ClassRepositoryMock(
+              object: expectAsync(() async {
+                expect(rendered, isFalse);
+                return object;
+              }, count: 1),
+              getter: expectAsync((id) async {
+                expect(ids.contains(id), isTrue);
+                switch (id) {
+                  case child1_id: return child1;
+                  case child1_1_id: return child1_1;
+                  case child2_id: return child2;
+                  default: return null;
+                }
+              }, count: 3)));
+      document.body.append(e);
+      await e.onRendered.first;
+      rendered = true;
+      expect(e.children.length, isNonZero, reason: 'has elements');
+      expect(e.querySelectorAll(nTag).length, equals(1));
+      e.remove();
+      await e.onRendered.first;
+      expect(e.children.length, isZero, reason: 'is empty');
+    });
+  });
+}

runtime/observatory/tests/observatory_ui/class_tree/element_test.html

diff --git a/runtime/observatory/tests/observatory_ui/class_tree/element_test.html b/runtime/observatory/tests/observatory_ui/class_tree/element_test.html
new file mode 100644
index 0000000..00d59bc
--- /dev/null
+++ b/runtime/observatory/tests/observatory_ui/class_tree/element_test.html
@@ -0,0 +1,26 @@
+<!DOCTYPE html>
+<html>
+<head>
+  <meta http-equiv="X-UA-Compatible" content="IE=edge">
+  <meta name="dart.unittest" content="full-stack-traces">
+  <style>
+     .unittest-table { font-family:monospace; border:1px; }
+     .unittest-pass { background: #6b3;}
+     .unittest-fail { background: #d55;}
+     .unittest-error { background: #a11;}
+     class-tree virtual-tree .class-tree-item {
+       line-height: 25px;
+       height: 25px;
+       padding-left: 10%;
+       padding-right: 10%;
+     }
+  </style>
+  <script src="/packages/web_components/webcomponents.js"></script>
+  <script src="/packages/web_components/dart_support.js"></script>
+</head>
+<body>
+  <script type="text/javascript"
+      src="/root_dart/tools/testing/dart/test_controller.js"></script>
+  %TEST_SCRIPTS%
+</body>
+</html>

runtime/observatory/tests/observatory_ui/code_ref/element_test.dart

diff --git a/runtime/observatory/tests/observatory_ui/code_ref/element_test.dart b/runtime/observatory/tests/observatory_ui/code_ref/element_test.dart
index 0e2c9ee..694d81b 100644
--- a/runtime/observatory/tests/observatory_ui/code_ref/element_test.dart
+++ b/runtime/observatory/tests/observatory_ui/code_ref/element_test.dart
@@ -3,22 +3,22 @@
 // BSD-style license that can be found in the LICENSE file.
 import 'dart:html';
 import 'package:unittest/unittest.dart';
-import 'package:observatory/mocks.dart';
 import 'package:observatory/src/elements/code_ref.dart';
+import '../mocks.dart';
 
 main(){
   CodeRefElement.tag.ensureRegistration();
 
-  final IsolateRefMock isolate = new IsolateRefMock(id: 'i-id', name: 'i-name');
-  final CodeRefMock code = new CodeRefMock(id: 'c-id', name: 'c-name');
+  final isolate = new IsolateRefMock(id: 'i-id', name: 'i-name');
+  final code = new CodeRefMock(id: 'c-id', name: 'c-name');
   test('instantiation', () {
-    final CodeRefElement e = new CodeRefElement(isolate, code);
+    final e = new CodeRefElement(isolate, code);
     expect(e, isNotNull, reason: 'element correctly created');
     expect(e.isolate, equals(isolate));
     expect(e.code, equals(code));
   });
   test('elements created after attachment', () async {
-    final CodeRefElement e = new CodeRefElement(isolate, code);
+    final e = new CodeRefElement(isolate, code);
     document.body.append(e);
     await e.onRendered.first;
     expect(e.children.length, isNonZero, reason: 'has elements');

runtime/observatory/tests/observatory_ui/curly_block/element_test.dart

diff --git a/runtime/observatory/tests/observatory_ui/curly_block/element_test.dart b/runtime/observatory/tests/observatory_ui/curly_block/element_test.dart
index fe1ca8f..6010bc6 100644
--- a/runtime/observatory/tests/observatory_ui/curly_block/element_test.dart
+++ b/runtime/observatory/tests/observatory_ui/curly_block/element_test.dart
@@ -10,66 +10,64 @@
 
   group('instantiation', () {
     test('default', () {
-      final CurlyBlockElement e = new CurlyBlockElement();
+      final e = new CurlyBlockElement();
       expect(e, isNotNull, reason: 'element correctly created');
       expect(e.expanded, isFalse);
       expect(e.disabled, isFalse);
     });
     test('not expanded', () {
-      final CurlyBlockElement e = new CurlyBlockElement(expanded: false);
+      final e = new CurlyBlockElement(expanded: false);
       expect(e, isNotNull, reason: 'element correctly created');
       expect(e.expanded, isFalse);
       expect(e.disabled, isFalse);
     });
     test('not expanded / not disabled', () {
-      final CurlyBlockElement e = new CurlyBlockElement(expanded: false,
+      final e = new CurlyBlockElement(expanded: false,
                                                         disabled: false);
       expect(e, isNotNull, reason: 'element correctly created');
       expect(e.expanded, isFalse);
       expect(e.disabled, isFalse);
     });
     test('not expanded / disabled', () {
-      final CurlyBlockElement e = new CurlyBlockElement(expanded: false,
+      final e = new CurlyBlockElement(expanded: false,
                                                         disabled: true);
       expect(e, isNotNull, reason: 'element correctly created');
       expect(e.expanded, isFalse);
       expect(e.disabled, isTrue);
     });
     test('expanded', () {
-      final CurlyBlockElement e = new CurlyBlockElement(expanded: true);
+      final e = new CurlyBlockElement(expanded: true);
       expect(e, isNotNull, reason: 'element correctly created');
       expect(e.expanded, isTrue);
       expect(e.disabled, isFalse);
     });
     test('expanded / not disabled', () {
-      final CurlyBlockElement e = new CurlyBlockElement(expanded: true,
-                                                        disabled: false);
+      final e = new CurlyBlockElement(expanded: true, disabled: false);
       expect(e, isNotNull, reason: 'element correctly created');
       expect(e.expanded, isTrue);
       expect(e.disabled, isFalse);
     });
     test('expanded / disabled', () {
-      final CurlyBlockElement e = new CurlyBlockElement(expanded: true,
-                                                        disabled: true);
+      final e = new CurlyBlockElement(expanded: true, disabled: true);
       expect(e, isNotNull, reason: 'element correctly created');
       expect(e.expanded, isTrue);
       expect(e.disabled, isTrue);
     });
     test('not disabled', () {
-      final CurlyBlockElement e = new CurlyBlockElement(disabled: false);
+      final e = new CurlyBlockElement(disabled: false);
       expect(e, isNotNull, reason: 'element correctly created');
       expect(e.expanded, isFalse);
       expect(e.disabled, isFalse);
     });
     test('disabled', () {
-      final CurlyBlockElement e = new CurlyBlockElement(disabled: true);
+      final e = new CurlyBlockElement(disabled: true);
       expect(e, isNotNull, reason: 'element correctly created');
       expect(e.expanded, isFalse);
       expect(e.disabled, isTrue);
     });
   });
   test('elements created', () async {
-    final CurlyBlockElement e = new CurlyBlockElement();
+    final e = new CurlyBlockElement();
     expect(e.shadowRoot, isNotNull, reason: 'shadowRoot is created');
     document.body.append(e);
     await e.onRendered.first;

runtime/observatory/tests/observatory_ui/error_ref/element_test.dart

diff --git a/runtime/observatory/tests/observatory_ui/error_ref/element_test.dart b/runtime/observatory/tests/observatory_ui/error_ref/element_test.dart
index 0495656..b131472 100644
--- a/runtime/observatory/tests/observatory_ui/error_ref/element_test.dart
+++ b/runtime/observatory/tests/observatory_ui/error_ref/element_test.dart
@@ -3,14 +3,13 @@
 // BSD-style license that can be found in the LICENSE file.
 import 'dart:html';
 import 'package:unittest/unittest.dart';
-import 'package:observatory/models.dart';
-import 'package:observatory/mocks.dart';
 import 'package:observatory/src/elements/error_ref.dart';
+import '../mocks.dart';
 
 main() {
   ErrorRefElement.tag.ensureRegistration();
 
-  final ErrorRef ref = new ErrorRefMock(id: 'id', message: 'fixed-error-m');
+  final ref = new ErrorRefMock(id: 'id', message: 'fixed-error-m');
   test('instantiation', () {
     final ErrorRefElement e = new ErrorRefElement(ref);
     expect(e, isNotNull, reason: 'element correctly created');

runtime/observatory/tests/observatory_ui/flag_list/element_test.dart

diff --git a/runtime/observatory/tests/observatory_ui/flag_list/element_test.dart b/runtime/observatory/tests/observatory_ui/flag_list/element_test.dart
index e88dea6..d23fffa 100644
--- a/runtime/observatory/tests/observatory_ui/flag_list/element_test.dart
+++ b/runtime/observatory/tests/observatory_ui/flag_list/element_test.dart
@@ -2,24 +2,24 @@
 // for details. All rights reserved. Use of this source code is governed by a
 // BSD-style license that can be found in the LICENSE file.
 import 'dart:html';
-import 'dart:async';
 import 'package:unittest/unittest.dart';
-import 'package:observatory/mocks.dart';
 import 'package:observatory/src/elements/flag_list.dart';
 import 'package:observatory/src/elements/nav/notify.dart';
+import '../mocks.dart';
 
 main() {
   FlagListElement.tag.ensureRegistration();
 
   final nTag = NavNotifyElement.tag.name;
-  const vm = const VMMock(name: 'vm');
-  final stream = new StreamController().stream;
+  const vm = const VMMock();
+  final events = new EventRepositoryMock();
+  final notifications = new NotificationRepositoryMock();
 
   group('instantiation', () {
     test('default', () {
-      final FlagListElement e = new FlagListElement(vm, stream,
-          new FlagsRepositoryMock(),
-          new NotificationRepositoryMock());
+      final e = new FlagListElement(vm, events,
+                                    new FlagsRepositoryMock(),
+                                    notifications);
       expect(e, isNotNull, reason: 'element correctly created');
     });
   });
@@ -32,15 +32,13 @@
         const FlagMock(name: 'f2', comment: 'c2', modified: false),
         const FlagMock(name: 'f3', comment: 'c3', modified: false),
       ];
-      final flags = new List.unmodifiable([]..addAll(modified)
-                                            ..addAll(unmodifed));
-      final FlagListElement e = new FlagListElement(vm, stream,
-          new FlagsRepositoryMock(list: expectAsync((input) async {
-            expect(input, equals(vm));
-            return flags;
-          }, count: 1)), new NotificationRepositoryMock());
+      final flags = []..addAll(modified)..addAll(unmodifed);
+      final repository = new FlagsRepositoryMock(list: flags);
+      final e = new FlagListElement(vm, events, repository, notifications);
       document.body.append(e);
+      expect(repository.isListInvoked, isFalse);
       await e.onRendered.first;
+      expect(repository.isListInvoked, isTrue);
       expect(e.children.length, isNonZero, reason: 'has elements');
       expect(e.querySelectorAll(nTag).length, equals(1));
       expect(e.querySelectorAll('.flag').length, equals(flags.length));

runtime/observatory/tests/observatory_ui/function_ref/element_test.dart

diff --git a/runtime/observatory/tests/observatory_ui/function_ref/element_test.dart b/runtime/observatory/tests/observatory_ui/function_ref/element_test.dart
index a647d3d..cd6b8af 100644
--- a/runtime/observatory/tests/observatory_ui/function_ref/element_test.dart
+++ b/runtime/observatory/tests/observatory_ui/function_ref/element_test.dart
@@ -3,23 +3,22 @@
 // BSD-style license that can be found in the LICENSE file.
 import 'dart:html';
 import 'package:unittest/unittest.dart';
-import 'package:observatory/mocks.dart';
 import 'package:observatory/src/elements/function_ref.dart';
+import '../mocks.dart';
 
 main(){
   FunctionRefElement.tag.ensureRegistration();
 
-  final IsolateRefMock isolate = new IsolateRefMock(id: 'i-id', name: 'i-name');
-  final FunctionRefMock function = new FunctionRefMock(id: 'f-id',
-      name: 'f-name');
+  final isolate = new IsolateRefMock(id: 'i-id', name: 'i-name');
+  final function = new FunctionRefMock(id: 'f-id', name: 'f-name');
   test('instantiation', () {
-    final FunctionRefElement e = new FunctionRefElement(isolate, function);
+    final e = new FunctionRefElement(isolate, function);
     expect(e, isNotNull, reason: 'element correctly created');
     expect(e.isolate, equals(isolate));
     expect(e.function, equals(function));
   });
   test('elements created after attachment', () async {
-    final FunctionRefElement e = new FunctionRefElement(isolate, function);
+    final e = new FunctionRefElement(isolate, function);
     document.body.append(e);
     await e.onRendered.first;
     expect(e.children.length, isNonZero, reason: 'has elements');

runtime/observatory/tests/observatory_ui/general_error/element_test.dart

diff --git a/runtime/observatory/tests/observatory_ui/general_error/element_test.dart b/runtime/observatory/tests/observatory_ui/general_error/element_test.dart
index 31a9031..e056bac 100644
--- a/runtime/observatory/tests/observatory_ui/general_error/element_test.dart
+++ b/runtime/observatory/tests/observatory_ui/general_error/element_test.dart
@@ -3,9 +3,9 @@
 // BSD-style license that can be found in the LICENSE file.
 import 'dart:html';
 import 'package:unittest/unittest.dart';
-import 'package:observatory/mocks.dart';
 import 'package:observatory/src/elements/general_error.dart';
 import 'package:observatory/src/elements/nav/notify.dart';
+import '../mocks.dart';
 
 main() {
   GeneralErrorElement.tag.ensureRegistration();

runtime/observatory/tests/observatory_ui/isolate_ref/element_test.dart

diff --git a/runtime/observatory/tests/observatory_ui/isolate_ref/element_test.dart b/runtime/observatory/tests/observatory_ui/isolate_ref/element_test.dart
index c7d64a4..2cfaafc 100644
--- a/runtime/observatory/tests/observatory_ui/isolate_ref/element_test.dart
+++ b/runtime/observatory/tests/observatory_ui/isolate_ref/element_test.dart
@@ -2,38 +2,30 @@
 // for details. All rights reserved. Use of this source code is governed by a
 // BSD-style license that can be found in the LICENSE file.
 import 'dart:html';
-import 'dart:async';
 import 'package:unittest/unittest.dart';
-import 'package:observatory/models.dart';
-import 'package:observatory/mocks.dart';
 import 'package:observatory/src/elements/isolate_ref.dart';
+import '../mocks.dart';
 
 main(){
   IsolateRefElement.tag.ensureRegistration();
 
-  StreamController<IsolateUpdateEvent> updatesController;
-  final IsolateRefMock ref = new IsolateRefMock(id: 'id', name: 'old-name');
-  final IsolateMock obj = new IsolateMock(id: 'id', name: 'new-name');
-  setUp(() {
-    updatesController = new StreamController<IsolateUpdateEvent>();
-  });
+  final ref = new IsolateRefMock(id: 'id', name: 'old-name');
+  final events = new EventRepositoryMock();
+  final obj = new IsolateMock(id: 'id', name: 'new-name');
   group('instantiation', () {
     test('IsolateRef', () {
-      final IsolateRefElement e = new IsolateRefElement(ref,
-                                      updatesController.stream);
+      final e = new IsolateRefElement(ref, events);
       expect(e, isNotNull, reason: 'element correctly created');
       expect(e.isolate, equals(ref));
     });
     test('Isolate', () {
-      final IsolateRefElement e = new IsolateRefElement(obj,
-                                      updatesController.stream);
+      final e = new IsolateRefElement(obj, events);
       expect(e, isNotNull, reason: 'element correctly created');
       expect(e.isolate, equals(obj));
     });
   });
   test('elements created after attachment', () async {
-    final IsolateRefElement e = new IsolateRefElement(ref,
-                                    updatesController.stream);
+    final e = new IsolateRefElement(ref, events);
     document.body.append(e);
     await e.onRendered.first;
     expect(e.children.length, isNonZero, reason: 'has elements');
@@ -43,23 +35,21 @@
   });
   group('updates', () {
     test('are correctly listen', () async {
-      final IsolateRefElement e = new IsolateRefElement(ref,
-                                      updatesController.stream);
-      expect(updatesController.hasListener, isFalse);
+      final e = new IsolateRefElement(ref, events);
+      expect(events.onIsolateUpdateHasListener, isFalse);
       document.body.append(e);
       await e.onRendered.first;
-      expect(updatesController.hasListener, isTrue);
+      expect(events.onIsolateUpdateHasListener, isTrue);
       e.remove();
       await e.onRendered.first;
-      expect(updatesController.hasListener, isFalse);
+      expect(events.onIsolateUpdateHasListener, isFalse);
     });
     test('have effects', () async {
-      final IsolateRefElement e = new IsolateRefElement(ref,
-                                      updatesController.stream);
+      final e = new IsolateRefElement(ref, events);
       document.body.append(e);
       await e.onRendered.first;
       expect(e.innerHtml.contains(ref.id), isTrue);
-      updatesController.add(new IsolateUpdateEventMock(isolate: obj));
+      events.add(new IsolateUpdateEventMock(isolate: obj));
       await e.onRendered.first;
       expect(e.innerHtml.contains(ref.name), isFalse);
       expect(e.innerHtml.contains(obj.name), isTrue);

runtime/observatory/tests/observatory_ui/library_ref/element_test.dart

diff --git a/runtime/observatory/tests/observatory_ui/library_ref/element_test.dart b/runtime/observatory/tests/observatory_ui/library_ref/element_test.dart
index cb33f6b..48291b5 100644
--- a/runtime/observatory/tests/observatory_ui/library_ref/element_test.dart
+++ b/runtime/observatory/tests/observatory_ui/library_ref/element_test.dart
@@ -3,22 +3,22 @@
 // BSD-style license that can be found in the LICENSE file.
 import 'dart:html';
 import 'package:unittest/unittest.dart';
-import 'package:observatory/mocks.dart';
 import 'package:observatory/src/elements/library_ref.dart';
+import '../mocks.dart';
 
 main(){
   LibraryRefElement.tag.ensureRegistration();
 
-  final IsolateRefMock isolate = new IsolateRefMock(id: 'i-id', name: 'i-name');
-  final LibraryRefMock library = new LibraryRefMock(id: 'c-id', name: 'c-name');
+  const isolate = const IsolateRefMock(id: 'i-id', name: 'i-name');
+  const library = const LibraryRefMock(id: 'c-id', name: 'c-name');
   test('instantiation', () {
-    final LibraryRefElement e = new LibraryRefElement(isolate, library);
+    final e = new LibraryRefElement(isolate, library);
     expect(e, isNotNull, reason: 'element correctly created');
     expect(e.isolate, equals(isolate));
     expect(e.library, equals(library));
   });
   test('elements created after attachment', () async {
-    final LibraryRefElement e = new LibraryRefElement(isolate, library);
+    final e = new LibraryRefElement(isolate, library);
     document.body.append(e);
     await e.onRendered.first;
     expect(e.children.length, isNonZero, reason: 'has elements');

runtime/observatory/tests/observatory_ui/mocks.dart

diff --git a/runtime/observatory/tests/observatory_ui/mocks.dart b/runtime/observatory/tests/observatory_ui/mocks.dart
new file mode 100644
index 0000000..b52b93b
--- /dev/null
+++ b/runtime/observatory/tests/observatory_ui/mocks.dart
@@ -0,0 +1,33 @@
+// Copyright (c) 2016, the Dart project authors.  Please see the AUTHORS file
+// for details. All rights reserved. Use of this source code is governed by a
+// BSD-style license that can be found in the LICENSE file.
+
+library mocks;
+
+import 'dart:async';
+
+import 'package:observatory/models.dart' as M;
+
+part 'mocks/exceptions/connection_exception.dart';
+
+part 'mocks/objects/error.dart';
+part 'mocks/objects/code.dart';
+part 'mocks/objects/event.dart';
+part 'mocks/objects/class.dart';
+part 'mocks/objects/flag.dart';
+part 'mocks/objects/function.dart';
+part 'mocks/objects/isolate.dart';
+part 'mocks/objects/library.dart';
+part 'mocks/objects/notification.dart';
+part 'mocks/objects/script.dart';
+part 'mocks/objects/source_location.dart';
+part 'mocks/objects/target.dart';
+part 'mocks/objects/vm.dart';
+
+part 'mocks/repositories/class.dart';
+part 'mocks/repositories/event.dart';
+part 'mocks/repositories/flag.dart';
+part 'mocks/repositories/instance.dart';
+part 'mocks/repositories/notification.dart';
+part 'mocks/repositories/script.dart';
+part 'mocks/repositories/target.dart';

runtime/observatory/tests/observatory_ui/mocks/exceptions/connection_exception.dart

diff --git a/runtime/observatory/lib/src/mocks/exceptions/connection_exception.dart b/runtime/observatory/tests/observatory_ui/mocks/exceptions/connection_exception.dart
similarity index 83%
rename from runtime/observatory/lib/src/mocks/exceptions/connection_exception.dart
rename to runtime/observatory/tests/observatory_ui/mocks/exceptions/connection_exception.dart
index 127cb20..d036951 100644
--- a/runtime/observatory/lib/src/mocks/exceptions/connection_exception.dart
+++ b/runtime/observatory/tests/observatory_ui/mocks/exceptions/connection_exception.dart
@@ -1,6 +1,6 @@
 // Copyright (c) 2016, the Dart project authors.  Please see the AUTHORS file
 // for details. All rights reserved. Use of this source code is governed by a
-// BSD-style license that can be found in the LICENSE file
+// BSD-style license that can be found in the LICENSE file.
 
 part of mocks;
 

runtime/observatory/tests/observatory_ui/mocks/objects/class.dart

diff --git a/runtime/observatory/tests/observatory_ui/mocks/objects/class.dart b/runtime/observatory/tests/observatory_ui/mocks/objects/class.dart
new file mode 100644
index 0000000..bc290af
--- /dev/null
+++ b/runtime/observatory/tests/observatory_ui/mocks/objects/class.dart
@@ -0,0 +1,31 @@
+// Copyright (c) 2016, the Dart project authors.  Please see the AUTHORS file
+// for details. All rights reserved. Use of this source code is governed by a
+// BSD-style license that can be found in the LICENSE file.
+
+part of mocks;
+
+class ClassRefMock implements M.ClassRef {
+  final String id;
+  final String name;
+  const ClassRefMock({this.id, this.name});
+}
+
+class ClassMock implements M.Class {
+  final String id;
+  final String name;
+  final M.ErrorRef error;
+  final bool isAbstract;
+  final bool isConst;
+  final bool isPatch;
+  final M.LibraryRef library;
+  final M.SourceLocation location;
+  final M.ClassRef superclass;
+  final M.InstanceRef superType;
+  final Iterable<M.InstanceRef> interfaces;
+  final M.InstanceRef mixin;
+  final Iterable<M.ClassRef> subclasses;
+  const ClassMock({this.id: 'c-id', this.name: 'c-name', this.error,
+      this.isAbstract: false, this.isConst: false, this.isPatch: false,
+      this.library, this.location, this.superclass, this.superType,
+      this.interfaces: const [], this.mixin, this.subclasses: const []});
+}

runtime/observatory/tests/observatory_ui/mocks/objects/code.dart

diff --git a/runtime/observatory/lib/src/mocks/objects/code.dart b/runtime/observatory/tests/observatory_ui/mocks/objects/code.dart
similarity index 90%
rename from runtime/observatory/lib/src/mocks/objects/code.dart
rename to runtime/observatory/tests/observatory_ui/mocks/objects/code.dart
index d0529ae..7d2f725 100644
--- a/runtime/observatory/lib/src/mocks/objects/code.dart
+++ b/runtime/observatory/tests/observatory_ui/mocks/objects/code.dart
@@ -1,6 +1,6 @@
 // Copyright (c) 2016, the Dart project authors.  Please see the AUTHORS file
 // for details. All rights reserved. Use of this source code is governed by a
-// BSD-style license that can be found in the LICENSE file
+// BSD-style license that can be found in the LICENSE file.
 
 part of mocks;
 

runtime/observatory/tests/observatory_ui/mocks/objects/error.dart

diff --git a/runtime/observatory/lib/src/mocks/objects/error.dart b/runtime/observatory/tests/observatory_ui/mocks/objects/error.dart
similarity index 89%
rename from runtime/observatory/lib/src/mocks/objects/error.dart
rename to runtime/observatory/tests/observatory_ui/mocks/objects/error.dart
index 8b47830..4a2b755 100644
--- a/runtime/observatory/lib/src/mocks/objects/error.dart
+++ b/runtime/observatory/tests/observatory_ui/mocks/objects/error.dart
@@ -1,6 +1,6 @@
 // Copyright (c) 2016, the Dart project authors.  Please see the AUTHORS file
 // for details. All rights reserved. Use of this source code is governed by a
-// BSD-style license that can be found in the LICENSE file
+// BSD-style license that can be found in the LICENSE file.
 
 part of mocks;
 

runtime/observatory/tests/observatory_ui/mocks/objects/event.dart

diff --git a/runtime/observatory/lib/src/mocks/objects/event.dart b/runtime/observatory/tests/observatory_ui/mocks/objects/event.dart
similarity index 93%
rename from runtime/observatory/lib/src/mocks/objects/event.dart
rename to runtime/observatory/tests/observatory_ui/mocks/objects/event.dart
index 6a646f3..7e4aa17 100644
--- a/runtime/observatory/lib/src/mocks/objects/event.dart
+++ b/runtime/observatory/tests/observatory_ui/mocks/objects/event.dart
@@ -1,6 +1,6 @@
 // Copyright (c) 2016, the Dart project authors.  Please see the AUTHORS file
 // for details. All rights reserved. Use of this source code is governed by a
-// BSD-style license that can be found in the LICENSE file
+// BSD-style license that can be found in the LICENSE file.
 
 part of mocks;
 
@@ -42,6 +42,11 @@
   const ServiceExtensionAddedEventMock({this.extensionRPC, this.isolate,
       this.timestamp});
 }
+class DebuggerSettingsUpdateEventMock implements M.PauseStartEvent {
+  final DateTime timestamp;
+  final M.IsolateRef isolate;
+  const DebuggerSettingsUpdateEventMock({this.isolate, this.timestamp});
+}
 class PauseStartEventMock implements M.PauseStartEvent {
   final DateTime timestamp;
   final M.IsolateRef isolate;
@@ -81,7 +86,8 @@
 class ResumeEventMock implements M.ResumeEvent {
   final DateTime timestamp;
   final M.IsolateRef isolate;
-  const ResumeEventMock({this.timestamp, this.isolate});
+  final M.Frame topFrame;
+  const ResumeEventMock({this.timestamp, this.isolate, this.topFrame});
 }
 class BreakpointAddedEventMock implements M.BreakpointAddedEvent {
   final DateTime timestamp;

runtime/observatory/tests/observatory_ui/mocks/objects/flag.dart

diff --git a/runtime/observatory/lib/src/mocks/objects/flag.dart b/runtime/observatory/tests/observatory_ui/mocks/objects/flag.dart
similarity index 100%
rename from runtime/observatory/lib/src/mocks/objects/flag.dart
rename to runtime/observatory/tests/observatory_ui/mocks/objects/flag.dart

runtime/observatory/tests/observatory_ui/mocks/objects/function.dart

diff --git a/runtime/observatory/lib/src/mocks/objects/function.dart b/runtime/observatory/tests/observatory_ui/mocks/objects/function.dart
similarity index 93%
rename from runtime/observatory/lib/src/mocks/objects/function.dart
rename to runtime/observatory/tests/observatory_ui/mocks/objects/function.dart
index 8a60adf..3458f97 100644
--- a/runtime/observatory/lib/src/mocks/objects/function.dart
+++ b/runtime/observatory/tests/observatory_ui/mocks/objects/function.dart
@@ -1,6 +1,6 @@
 // Copyright (c) 2016, the Dart project authors.  Please see the AUTHORS file
 // for details. All rights reserved. Use of this source code is governed by a
-// BSD-style license that can be found in the LICENSE file
+// BSD-style license that can be found in the LICENSE file.
 
 part of mocks;
 

runtime/observatory/tests/observatory_ui/mocks/objects/isolate.dart

diff --git a/runtime/observatory/lib/src/mocks/objects/isolate.dart b/runtime/observatory/tests/observatory_ui/mocks/objects/isolate.dart
similarity index 64%
rename from runtime/observatory/lib/src/mocks/objects/isolate.dart
rename to runtime/observatory/tests/observatory_ui/mocks/objects/isolate.dart
index 9f5ba80..d65b840 100644
--- a/runtime/observatory/lib/src/mocks/objects/isolate.dart
+++ b/runtime/observatory/tests/observatory_ui/mocks/objects/isolate.dart
@@ -1,6 +1,6 @@
 // Copyright (c) 2016, the Dart project authors.  Please see the AUTHORS file
 // for details. All rights reserved. Use of this source code is governed by a
-// BSD-style license that can be found in the LICENSE file
+// BSD-style license that can be found in the LICENSE file.
 
 part of mocks;
 
@@ -9,7 +9,7 @@
   final int number;
   final String name;
 
-  const IsolateRefMock({this.id, this.number, this.name});
+  const IsolateRefMock({this.id: 'i-id', this.number, this.name: 'i-name'});
 }
 
 class IsolateMock implements M.Isolate {
@@ -19,7 +19,7 @@
   final DateTime startTime;
   final bool runnable;
 
-  const IsolateMock({this.id, this.number, this.name, this.startTime,
-      this.runnable});
+  const IsolateMock({this.id: 'i-id', this.number, this.name: 'i-name',
+                     this.startTime, this.runnable: true});
   // TODO(cbernaschina) add other properties.
 }

runtime/observatory/tests/observatory_ui/mocks/objects/library.dart

diff --git a/runtime/observatory/lib/src/mocks/objects/library.dart b/runtime/observatory/tests/observatory_ui/mocks/objects/library.dart
similarity index 91%
rename from runtime/observatory/lib/src/mocks/objects/library.dart
rename to runtime/observatory/tests/observatory_ui/mocks/objects/library.dart
index 2932efa..8022904 100644
--- a/runtime/observatory/lib/src/mocks/objects/library.dart
+++ b/runtime/observatory/tests/observatory_ui/mocks/objects/library.dart
@@ -1,6 +1,6 @@
 // Copyright (c) 2016, the Dart project authors.  Please see the AUTHORS file
 // for details. All rights reserved. Use of this source code is governed by a
-// BSD-style license that can be found in the LICENSE file
+// BSD-style license that can be found in the LICENSE file.
 
 part of mocks;
 

runtime/observatory/tests/observatory_ui/mocks/objects/notification.dart

diff --git a/runtime/observatory/lib/src/mocks/objects/notification.dart b/runtime/observatory/tests/observatory_ui/mocks/objects/notification.dart
similarity index 89%
rename from runtime/observatory/lib/src/mocks/objects/notification.dart
rename to runtime/observatory/tests/observatory_ui/mocks/objects/notification.dart
index 68bfd75..d452cee 100644
--- a/runtime/observatory/lib/src/mocks/objects/notification.dart
+++ b/runtime/observatory/tests/observatory_ui/mocks/objects/notification.dart
@@ -1,6 +1,6 @@
 // Copyright (c) 2016, the Dart project authors.  Please see the AUTHORS file
 // for details. All rights reserved. Use of this source code is governed by a
-// BSD-style license that can be found in the LICENSE file
+// BSD-style license that can be found in the LICENSE file.
 
 part of mocks;
 

runtime/observatory/tests/observatory_ui/mocks/objects/script.dart

diff --git a/runtime/observatory/lib/src/mocks/objects/script.dart b/runtime/observatory/tests/observatory_ui/mocks/objects/script.dart
similarity index 92%
rename from runtime/observatory/lib/src/mocks/objects/script.dart
rename to runtime/observatory/tests/observatory_ui/mocks/objects/script.dart
index 7ce801a..895b1c9 100644
--- a/runtime/observatory/lib/src/mocks/objects/script.dart
+++ b/runtime/observatory/tests/observatory_ui/mocks/objects/script.dart
@@ -1,6 +1,6 @@
 // Copyright (c) 2016, the Dart project authors.  Please see the AUTHORS file
 // for details. All rights reserved. Use of this source code is governed by a
-// BSD-style license that can be found in the LICENSE file
+// BSD-style license that can be found in the LICENSE file.
 
 part of mocks;
 

runtime/observatory/tests/observatory_ui/mocks/objects/source_location.dart

diff --git a/runtime/observatory/lib/src/mocks/objects/source_location.dart b/runtime/observatory/tests/observatory_ui/mocks/objects/source_location.dart
similarity index 86%
rename from runtime/observatory/lib/src/mocks/objects/source_location.dart
rename to runtime/observatory/tests/observatory_ui/mocks/objects/source_location.dart
index 1c9c0f2..a795589 100644
--- a/runtime/observatory/lib/src/mocks/objects/source_location.dart
+++ b/runtime/observatory/tests/observatory_ui/mocks/objects/source_location.dart
@@ -1,6 +1,6 @@
 // Copyright (c) 2016, the Dart project authors.  Please see the AUTHORS file
 // for details. All rights reserved. Use of this source code is governed by a
-// BSD-style license that can be found in the LICENSE file
+// BSD-style license that can be found in the LICENSE file.
 
 part of mocks;
 

runtime/observatory/tests/observatory_ui/mocks/objects/target.dart

diff --git a/runtime/observatory/lib/src/mocks/objects/target.dart b/runtime/observatory/tests/observatory_ui/mocks/objects/target.dart
similarity index 81%
rename from runtime/observatory/lib/src/mocks/objects/target.dart
rename to runtime/observatory/tests/observatory_ui/mocks/objects/target.dart
index c5ae189..0d15e5c 100644
--- a/runtime/observatory/lib/src/mocks/objects/target.dart
+++ b/runtime/observatory/tests/observatory_ui/mocks/objects/target.dart
@@ -1,6 +1,6 @@
 // Copyright (c) 2016, the Dart project authors.  Please see the AUTHORS file
 // for details. All rights reserved. Use of this source code is governed by a
-// BSD-style license that can be found in the LICENSE file
+// BSD-style license that can be found in the LICENSE file.
 
 part of mocks;
 

runtime/observatory/tests/observatory_ui/mocks/objects/vm.dart

diff --git a/runtime/observatory/tests/observatory_ui/mocks/objects/vm.dart b/runtime/observatory/tests/observatory_ui/mocks/objects/vm.dart
new file mode 100644
index 0000000..2740b83
--- /dev/null
+++ b/runtime/observatory/tests/observatory_ui/mocks/objects/vm.dart
@@ -0,0 +1,27 @@
+// Copyright (c) 2016, the Dart project authors.  Please see the AUTHORS file
+// for details. All rights reserved. Use of this source code is governed by a
+// BSD-style license that can be found in the LICENSE file.
+
+part of mocks;
+
+class VMRefMock implements M.VMRef {
+  final String name;
+  final String displayName;
+  const VMRefMock({this.name: 'vm-name', this.displayName: 'vm-display-name'});
+}
+
+class VMMock implements M.VM {
+  final String name;
+  final String displayName;
+  final int architectureBits;
+  final String targetCPU;
+  final String hostCPU;
+  final String version;
+  final int pid;
+  final DateTime startTime;
+  final Iterable<M.IsolateRef> isolates;
+
+  const VMMock({this.name: 'vm-name', this.displayName: 'vm-display-name',
+      this.architectureBits, this.targetCPU, this.hostCPU, this.version,
+      this.pid, this.startTime, this.isolates : const []});
+}

runtime/observatory/tests/observatory_ui/mocks/repositories/class.dart

diff --git a/runtime/observatory/tests/observatory_ui/mocks/repositories/class.dart b/runtime/observatory/tests/observatory_ui/mocks/repositories/class.dart
new file mode 100644
index 0000000..3c5fe7b
--- /dev/null
+++ b/runtime/observatory/tests/observatory_ui/mocks/repositories/class.dart
@@ -0,0 +1,32 @@
+// Copyright (c) 2016, the Dart project authors.  Please see the AUTHORS file
+// for details. All rights reserved. Use of this source code is governed by a
+// BSD-style license that can be found in the LICENSE file.
+
+part of mocks;
+
+typedef Future<M.Class> ClassRepositoryMockObjectCallback();
+typedef Future<M.Class> ClassRepositoryMockGetterCallback(String id);
+
+class ClassRepositoryMock implements M.ClassRepository {
+  final ClassRepositoryMockObjectCallback _object;
+  final ClassRepositoryMockGetterCallback _get;
+
+  ClassRepositoryMock({ClassRepositoryMockObjectCallback object,
+                       ClassRepositoryMockGetterCallback getter})
+    : _object = object,
+      _get = getter;
+
+  Future<M.Class> getObject(){
+    if (_object != null) {
+      return _object();
+    }
+    return new Future.value(null);
+  }
+
+  Future<M.Class> get(String id){
+    if (_get != null) {
+      return _get(id);
+    }
+    return new Future.value(null);
+  }
+}

runtime/observatory/tests/observatory_ui/mocks/repositories/event.dart

diff --git a/runtime/observatory/tests/observatory_ui/mocks/repositories/event.dart b/runtime/observatory/tests/observatory_ui/mocks/repositories/event.dart
new file mode 100644
index 0000000..7748786
--- /dev/null
+++ b/runtime/observatory/tests/observatory_ui/mocks/repositories/event.dart
@@ -0,0 +1,171 @@
+// Copyright (c) 2016, the Dart project authors.  Please see the AUTHORS file
+// for details. All rights reserved. Use of this source code is governed by a
+// BSD-style license that can be found in the LICENSE file.
+
+part of mocks;
+
+class EventRepositoryMock implements M.EventRepository {
+  final _onEvent = new StreamController<M.Event>.broadcast();
+  get onEvent => _onEvent.stream;
+  get onEventHasListener => _onEvent.hasListener;
+
+  final _onVMEvent = new StreamController<M.VMEvent>.broadcast();
+  get onVMEvent => _onVMEvent.stream;
+  get onVMEventHasListener => _onVMEvent.hasListener;
+
+  final _onVMUpdate = new StreamController<M.Event>.broadcast();
+  get onVMUpdate => _onVMUpdate.stream;
+  get onVMUpdateHasListener => _onVMUpdate.hasListener;
+
+  final _onIsolateEvent = new StreamController<M.IsolateEvent>.broadcast();
+  get onIsolateEvent => _onIsolateEvent.stream;
+  get onIsolateEventHasListener => _onIsolateEvent.hasListener;
+
+  final _onIsolateStart = new StreamController<M.IsolateStartEvent>.broadcast();
+  get onIsolateStart => _onIsolateStart.stream;
+  get onIsolateStartHasListener => _onIsolateStart.hasListener;
+
+  final _onIsolateRunnable =
+      new StreamController<M.IsolateRunnableEvent>.broadcast();
+  get onIsolateRunnable => _onIsolateRunnable.stream;
+  get onIsolateRunnableHasListener => _onIsolateRunnable.hasListener;
+
+  final _onIsolateExit = new StreamController<M.IsolateExitEvent>.broadcast();
+  get onIsolateExit => _onIsolateExit.stream;
+  get onIsolateExitHasListener => _onIsolateExit.hasListener;
+
+  final _onIsolateUpdate =
+      new StreamController<M.IsolateUpdateEvent>.broadcast();
+  get onIsolateUpdate => _onIsolateUpdate.stream;
+  get onIsolateUpdateHasListener => _onIsolateUpdate.hasListener;
+
+  final _onIsolateReload = new StreamController<M.IsolateReloadEvent>.broadcast();
+  get onIsolateReload => _onIsolateReload.stream;
+  get onIsolateReloadHasListener => _onIsolateReload.hasListener;
+
+  final _onServiceExtensionAdded =
+      new StreamController<M.ServiceExtensionAddedEvent>.broadcast();
+  get onServiceExtensionAdded => _onServiceExtensionAdded.stream;
+  get onServiceExtensionAddedHasListener =>
+      _onServiceExtensionAdded.hasListener;
+
+  final _onDebugEvent = new StreamController<M.DebugEvent>.broadcast();
+  get onDebugEvent => _onDebugEvent.stream;
+  get onDebugEventHasListener => _onDebugEvent.hasListener;
+
+  final _onPauseStart = new StreamController<M.PauseStartEvent>.broadcast();
+  get onPauseStart => _onPauseStart.stream;
+  get onPauseStartHasListener => _onPauseStart.hasListener;
+
+  final _onPauseExit = new StreamController<M.PauseExitEvent>.broadcast();
+  get onPauseExit => _onPauseExit.stream;
+  get onPauseExitHasListener => _onPauseExit.hasListener;
+
+  final _onPauseBreakpoint =
+      new StreamController<M.PauseBreakpointEvent>.broadcast();
+  get onPauseBreakpoint => _onPauseBreakpoint.stream;
+  get onPauseBreakpointHasListener => _onPauseBreakpoint.hasListener;
+
+  final _onPauseInterrupted =
+      new StreamController<M.PauseInterruptedEvent>.broadcast();
+  get onPauseInterrupted => _onPauseInterrupted.stream;
+  get onPauseInterruptedHasListener => _onPauseInterrupted.hasListener;
+
+  final _onPauseException =
+      new StreamController<M.PauseExceptionEvent>.broadcast();
+  get onPauseException => _onPauseException.stream;
+  get onPauseExceptionHasListener => _onPauseException.hasListener;
+
+  final _onResume = new StreamController<M.ResumeEvent>.broadcast();
+  get onResume => _onResume.stream;
+  get onResumeHasListener => _onResume.hasListener;
+
+  final _onBreakpointAdded =
+      new StreamController<M.BreakpointAddedEvent>.broadcast();
+  get onBreakpointAdded => _onBreakpointAdded.stream;
+  get onBreakpointAddedHasListener => _onBreakpointAdded.hasListener;
+
+  final _onBreakpointResolved =
+    new StreamController<M.BreakpointResolvedEvent>.broadcast();
+  get onBreakpointResolved => _onBreakpointResolved.stream;
+  get onBreakpointResolvedHasListener => _onBreakpointResolved.hasListener;
+
+  final _onBreakpointRemoved =
+      new StreamController<M.BreakpointRemovedEvent>.broadcast();
+  get onBreakpointRemoved => _onBreakpointRemoved.stream;
+  get onBreakpointRemovedHasListener => _onBreakpointRemoved.hasListener;
+
+  final _onInspect = new StreamController<M.InspectEvent>.broadcast();
+  get onInspect => _onInspect.stream;
+  get onInspectHasListener => _onInspect.hasListener;
+
+  final _onGCEvent = new StreamController<M.GCEvent>.broadcast();
+  get onGCEvent => _onGCEvent.stream;
+  get onGCEventHasListener => _onGCEvent.hasListener;
+
+  final _onExtensionEvent = new StreamController<M.ExtensionEvent>.broadcast();
+  get onExtensionEvent => _onExtensionEvent.stream;
+  get onExtensionEventHasListener => _onExtensionEvent.hasListener;
+
+  final _onTimelineEvents =
+      new StreamController<M.TimelineEventsEvent>.broadcast();
+  get onTimelineEvents => _onTimelineEvents.stream;
+  get onTimelineEventsEventHasListener => _onTimelineEvents.hasListener;
+
+  final _onConnectionClosed =
+      new StreamController<M.ConnectionClosedEvent>.broadcast();
+  get onConnectionClosed => _onConnectionClosed.stream;
+  get onConnectionClosedHasListener => _onConnectionClosed.hasListener;
+
+  void add(M.Event event) {
+    _onEvent.add(event);
+    if (event is M.VMEvent) {
+      _onVMEvent.add(event);
+      if (event is M.VMUpdateEvent) {
+        _onVMUpdate.add(event);
+      }
+    } else if (event is M.IsolateEvent) {
+      _onIsolateEvent.add(event);
+      if (event is M.IsolateStartEvent) {
+        _onIsolateStart.add(event);
+      } else if (event is M.IsolateRunnableEvent) {
+        _onIsolateRunnable.add(event);
+      } else if (event is M.IsolateExitEvent) {
+        _onIsolateExit.add(event);
+      } else if (event is M.IsolateUpdateEvent) {
+        _onIsolateUpdate.add(event);
+      } else if (event is M.ServiceExtensionAddedEvent) {
+        _onServiceExtensionAdded.add(event);
+      }
+    } else if (event is M.DebugEvent) {
+      _onDebugEvent.add(event);
+      if (event is M.PauseStartEvent) {
+        _onPauseStart.add(event);
+      } else if (event is M.PauseExitEvent) {
+        _onPauseExit.add(event);
+      } else if (event is M.PauseBreakpointEvent) {
+        _onPauseBreakpoint.add(event);
+      } else if (event is M.PauseInterruptedEvent) {
+        _onPauseInterrupted.add(event);
+      } else if (event is M.PauseExceptionEvent) {
+        _onPauseException.add(event);
+      } else if (event is M.ResumeEvent) {
+        _onResume.add(event);
+      } else if (event is M.BreakpointAddedEvent) {
+        _onBreakpointAdded.add(event);
+      } else if (event is M.BreakpointResolvedEvent) {
+        _onBreakpointResolved.add(event);
+      } else if (event is M.BreakpointRemovedEvent) {
+        _onBreakpointRemoved.add(event);
+      } else if (event is M.InspectEvent) {
+        _onInspect.add(event);
+      }
+    } else if (event is M.GCEvent) {
+      _onGCEvent.add(event);
+    } else if (event is M.ExtensionEvent) {
+      _onExtensionEvent.add(event);
+    } else if (event is M.TimelineEventsEvent) {
+      _onTimelineEvents.add(event);
+    }
+  }
+}

runtime/observatory/tests/observatory_ui/mocks/repositories/flag.dart

diff --git a/runtime/observatory/tests/observatory_ui/mocks/repositories/flag.dart b/runtime/observatory/tests/observatory_ui/mocks/repositories/flag.dart
new file mode 100644
index 0000000..e782b30
--- /dev/null
+++ b/runtime/observatory/tests/observatory_ui/mocks/repositories/flag.dart
@@ -0,0 +1,18 @@
+// Copyright (c) 2016, the Dart project authors.  Please see the AUTHORS file
+// for details. All rights reserved. Use of this source code is governed by a
+// BSD-style license that can be found in the LICENSE file
+
+part of mocks;
+
+class FlagsRepositoryMock implements M.FlagsRepository {
+  final Iterable<M.Flag> _list;
+  bool isListInvoked = false;
+
+  Future<Iterable<M.Flag>> list() async {
+    isListInvoked = true;
+    return _list;
+  }
+
+  FlagsRepositoryMock({Iterable<M.Flag> list: const []})
+    : _list = new List.unmodifiable(list);
+}

runtime/observatory/tests/observatory_ui/mocks/repositories/instance.dart

diff --git a/runtime/observatory/tests/observatory_ui/mocks/repositories/instance.dart b/runtime/observatory/tests/observatory_ui/mocks/repositories/instance.dart
new file mode 100644
index 0000000..299dcfc
--- /dev/null
+++ b/runtime/observatory/tests/observatory_ui/mocks/repositories/instance.dart
@@ -0,0 +1,21 @@
+// Copyright (c) 2016, the Dart project authors.  Please see the AUTHORS file
+// for details. All rights reserved. Use of this source code is governed by a
+// BSD-style license that can be found in the LICENSE file.
+
+part of mocks;
+
+typedef Future<M.Instance> InstanceRepositoryMockCallback(String id);
+
+class InstanceRepositoryMock implements M.InstanceRepository {
+  final InstanceRepositoryMockCallback _get;
+
+  InstanceRepositoryMock({InstanceRepositoryMockCallback getter})
+    : _get = getter;
+
+  Future<M.Instance> get(String id){
+    if (_get != null) {
+      return _get(id);
+    }
+    return new Future.value(null);
+  }
+}

runtime/observatory/tests/observatory_ui/mocks/repositories/notification.dart

diff --git a/runtime/observatory/lib/src/mocks/repositories/notification.dart b/runtime/observatory/tests/observatory_ui/mocks/repositories/notification.dart
similarity index 96%
rename from runtime/observatory/lib/src/mocks/repositories/notification.dart
rename to runtime/observatory/tests/observatory_ui/mocks/repositories/notification.dart
index cf7f2b9..f03cd27 100644
--- a/runtime/observatory/lib/src/mocks/repositories/notification.dart
+++ b/runtime/observatory/tests/observatory_ui/mocks/repositories/notification.dart
@@ -1,6 +1,6 @@
 // Copyright (c) 2016, the Dart project authors.  Please see the AUTHORS file
 // for details. All rights reserved. Use of this source code is governed by a
-// BSD-style license that can be found in the LICENSE file
+// BSD-style license that can be found in the LICENSE file.
 
 part of mocks;
 

runtime/observatory/tests/observatory_ui/mocks/repositories/script.dart

diff --git a/runtime/observatory/tests/observatory_ui/mocks/repositories/script.dart b/runtime/observatory/tests/observatory_ui/mocks/repositories/script.dart
new file mode 100644
index 0000000..78b9610
--- /dev/null
+++ b/runtime/observatory/tests/observatory_ui/mocks/repositories/script.dart
@@ -0,0 +1,21 @@
+// Copyright (c) 2016, the Dart project authors.  Please see the AUTHORS file
+// for details. All rights reserved. Use of this source code is governed by a
+// BSD-style license that can be found in the LICENSE file.
+
+part of mocks;
+
+typedef Future<M.Script> ScriptRepositoryMockCallback(String id);
+
+class ScriptRepositoryMock implements M.ScriptRepository {
+  final ScriptRepositoryMockCallback _get;
+
+  ScriptRepositoryMock({ScriptRepositoryMockCallback getter})
+    : _get = getter;
+
+  Future<M.Script> get(String id){
+    if (_get != null) {
+      return _get(id);
+    }
+    return new Future.value(null);
+  }
+}

runtime/observatory/tests/observatory_ui/mocks/repositories/target.dart

diff --git a/runtime/observatory/lib/src/mocks/repositories/target.dart b/runtime/observatory/tests/observatory_ui/mocks/repositories/target.dart
similarity index 96%
rename from runtime/observatory/lib/src/mocks/repositories/target.dart
rename to runtime/observatory/tests/observatory_ui/mocks/repositories/target.dart
index 8321f86..1c7f3ed 100644
--- a/runtime/observatory/lib/src/mocks/repositories/target.dart
+++ b/runtime/observatory/tests/observatory_ui/mocks/repositories/target.dart
@@ -1,6 +1,6 @@
 // Copyright (c) 2016, the Dart project authors.  Please see the AUTHORS file
 // for details. All rights reserved. Use of this source code is governed by a
-// BSD-style license that can be found in the LICENSE file
+// BSD-style license that can be found in the LICENSE file.
 
 part of mocks;
 

runtime/observatory/tests/observatory_ui/nav/bar/element_test.dart

diff --git a/runtime/observatory/tests/observatory_ui/nav/bar/element_test.dart b/runtime/observatory/tests/observatory_ui/nav/bar/element_test.dart
index fe54785..98b9653 100644
--- a/runtime/observatory/tests/observatory_ui/nav/bar/element_test.dart
+++ b/runtime/observatory/tests/observatory_ui/nav/bar/element_test.dart
@@ -9,11 +9,11 @@
   NavBarElement.tag.ensureRegistration();
 
   test('instantiation', () {
-    final NavBarElement e = new NavBarElement();
+    final e = new NavBarElement();
     expect(e, isNotNull, reason: 'element correctly created');
   });
   test('elements created', () async {
-    final NavBarElement e = new NavBarElement();
+    final e = new NavBarElement();
     document.body.append(e);
     await e.onRendered.first;
     expect(e.shadowRoot.children.length, isNonZero, reason: 'has elements');

runtime/observatory/tests/observatory_ui/nav/class-menu/element_test.dart

diff --git a/runtime/observatory/tests/observatory_ui/nav/class-menu/element_test.dart b/runtime/observatory/tests/observatory_ui/nav/class-menu/element_test.dart
index df83ea4..1b3766d 100644
--- a/runtime/observatory/tests/observatory_ui/nav/class-menu/element_test.dart
+++ b/runtime/observatory/tests/observatory_ui/nav/class-menu/element_test.dart
@@ -3,22 +3,22 @@
 // BSD-style license that can be found in the LICENSE file.
 import 'dart:html';
 import 'package:unittest/unittest.dart';
-import 'package:observatory/mocks.dart';
 import 'package:observatory/src/elements/nav/class_menu.dart';
+import '../../mocks.dart';
 
 main(){
   NavClassMenuElement.tag.ensureRegistration();
 
-  final IsolateRefMock i_ref = const IsolateRefMock(id: 'i-id', name: 'i-name');
-  final ClassRefMock c_ref = const ClassRefMock(id: 'c-id', name: 'c-name');
+  final i_ref = const IsolateRefMock(id: 'i-id', name: 'i-name');
+  final c_ref = const ClassRefMock(id: 'c-id', name: 'c-name');
   test('instantiation', () {
-    final NavClassMenuElement e = new NavClassMenuElement(i_ref, c_ref);
+    final e = new NavClassMenuElement(i_ref, c_ref);
     expect(e, isNotNull, reason: 'element correctly created');
     expect(e.isolate, equals(i_ref));
     expect(e.cls, equals(c_ref));
   });
   test('elements created after attachment', () async {
-    final NavClassMenuElement e = new NavClassMenuElement(i_ref, c_ref);
+    final e = new NavClassMenuElement(i_ref, c_ref);
     document.body.append(e);
     await e.onRendered.first;
     expect(e.shadowRoot.children.length, isNonZero, reason: 'has elements');

runtime/observatory/tests/observatory_ui/nav/isolate-menu/element_test.dart

diff --git a/runtime/observatory/tests/observatory_ui/nav/isolate-menu/element_test.dart b/runtime/observatory/tests/observatory_ui/nav/isolate-menu/element_test.dart
index 27bf79c..235975f 100644
--- a/runtime/observatory/tests/observatory_ui/nav/isolate-menu/element_test.dart
+++ b/runtime/observatory/tests/observatory_ui/nav/isolate-menu/element_test.dart
@@ -2,41 +2,36 @@
 // for details. All rights reserved. Use of this source code is governed by a
 // BSD-style license that can be found in the LICENSE file.
 import 'dart:html';
-import 'dart:async';
 import 'package:unittest/unittest.dart';
-import 'package:observatory/models.dart';
-import 'package:observatory/mocks.dart';
 import 'package:observatory/src/elements/nav/menu.dart';
 import 'package:observatory/src/elements/nav/isolate_menu.dart';
+import '../../mocks.dart';
 
 main(){
   NavIsolateMenuElement.tag.ensureRegistration();
 
-  final String tag = NavMenuElement.tag.name;
+  final tag = NavMenuElement.tag.name;
 
-  StreamController<IsolateUpdateEvent> updatesController;
-  final IsolateRefMock ref = const IsolateRefMock(id: 'i-id', name: 'old-name');
-  final IsolateMock obj = const IsolateMock(id: 'i-id', name: 'new-name');
+  EventRepositoryMock events;
+  final ref = const IsolateRefMock(id: 'i-id', name: 'old-name');
+  final obj = const IsolateMock(id: 'i-id', name: 'new-name');
   setUp(() {
-    updatesController = new StreamController<IsolateUpdateEvent>();
+    events = new EventRepositoryMock();
   });
   group('instantiation', () {
     test('IsolateRef', () {
-      final NavIsolateMenuElement e = new NavIsolateMenuElement(ref,
-                                      updatesController.stream);
+      final e = new NavIsolateMenuElement(ref, events);
       expect(e, isNotNull, reason: 'element correctly created');
       expect(e.isolate, equals(ref));
     });
     test('Isolate', () {
-      final NavIsolateMenuElement e = new NavIsolateMenuElement(obj,
-                                      updatesController.stream);
+      final e = new NavIsolateMenuElement(obj, events);
       expect(e, isNotNull, reason: 'element correctly created');
       expect(e.isolate, equals(obj));
     });
   });
   test('elements created after attachment', () async {
-    final NavIsolateMenuElement e = new NavIsolateMenuElement(ref,
-                                    updatesController.stream);
+    final e = new NavIsolateMenuElement(ref, events);
     document.body.append(e);
     await e.onRendered.first;
     expect(e.shadowRoot.children.length, isNonZero, reason: 'has elements');
@@ -46,24 +41,22 @@
   });
   group('updates', () {
     test('are correctly listen', () async {
-      final NavIsolateMenuElement e = new NavIsolateMenuElement(ref,
-                                      updatesController.stream);
-      expect(updatesController.hasListener, isFalse);
+      final e = new NavIsolateMenuElement(ref, events);
+      expect(events.onIsolateUpdateHasListener, isFalse);
       document.body.append(e);
       await e.onRendered.first;
-      expect(updatesController.hasListener, isTrue);
+      expect(events.onIsolateUpdateHasListener, isTrue);
       e.remove();
       await e.onRendered.first;
-      expect(updatesController.hasListener, isFalse);
+      expect(events.onIsolateUpdateHasListener, isFalse);
     });
     test('have effects', () async {
-      final NavIsolateMenuElement e = new NavIsolateMenuElement(ref,
-                                      updatesController.stream);
+      final e = new NavIsolateMenuElement(ref, events);
       document.body.append(e);
       await e.onRendered.first;
       expect((e.shadowRoot.querySelector(tag) as NavMenuElement)
              .label.contains(ref.name), isTrue);
-      updatesController.add(new IsolateUpdateEventMock(isolate: obj));
+      events.add(new IsolateUpdateEventMock(isolate: obj));
       await e.onRendered.first;
       expect((e.shadowRoot.querySelector(tag) as NavMenuElement)
              .label.contains(ref.name), isFalse);

runtime/observatory/tests/observatory_ui/nav/library-menu/element_test.dart

diff --git a/runtime/observatory/tests/observatory_ui/nav/library-menu/element_test.dart b/runtime/observatory/tests/observatory_ui/nav/library-menu/element_test.dart
index f6967c9..2144520 100644
--- a/runtime/observatory/tests/observatory_ui/nav/library-menu/element_test.dart
+++ b/runtime/observatory/tests/observatory_ui/nav/library-menu/element_test.dart
@@ -3,22 +3,22 @@
 // BSD-style license that can be found in the LICENSE file.
 import 'dart:html';
 import 'package:unittest/unittest.dart';
-import 'package:observatory/mocks.dart';
 import 'package:observatory/src/elements/nav/library_menu.dart';
+import '../../mocks.dart';
 
 main(){
   NavLibraryMenuElement.tag.ensureRegistration();
 
-  final IsolateRefMock i_ref = const IsolateRefMock(id: 'i-id', name: 'i-name');
-  final LibraryRefMock l_ref = const LibraryRefMock(id: 'l-id', name: 'l-name');
+  final i_ref = const IsolateRefMock(id: 'i-id', name: 'i-name');
+  final l_ref = const LibraryRefMock(id: 'l-id', name: 'l-name');
   test('instantiation', () {
-    final NavLibraryMenuElement e = new NavLibraryMenuElement(i_ref, l_ref);
+    final e = new NavLibraryMenuElement(i_ref, l_ref);
     expect(e, isNotNull, reason: 'element correctly created');
     expect(e.isolate, equals(i_ref));
     expect(e.library, equals(l_ref));
   });
   test('elements created after attachment', () async {
-    final NavLibraryMenuElement e = new NavLibraryMenuElement(i_ref, l_ref);
+    final e = new NavLibraryMenuElement(i_ref, l_ref);
     document.body.append(e);
     await e.onRendered.first;
     expect(e.shadowRoot.children.length, isNonZero, reason: 'has elements');

runtime/observatory/tests/observatory_ui/nav/menu-item/element_test.dart

diff --git a/runtime/observatory/tests/observatory_ui/nav/menu-item/element_test.dart b/runtime/observatory/tests/observatory_ui/nav/menu-item/element_test.dart
index f4ee551..d439a60 100644
--- a/runtime/observatory/tests/observatory_ui/nav/menu-item/element_test.dart
+++ b/runtime/observatory/tests/observatory_ui/nav/menu-item/element_test.dart
@@ -12,12 +12,12 @@
     final label = 'custom-label';
     final link = 'link-to-target';
     test('label', () {
-      final NavMenuItemElement e = new NavMenuItemElement(label);
+      final e = new NavMenuItemElement(label);
       expect(e, isNotNull, reason: 'element correctly created');
       expect(e.label, equals(label), reason: 'element correctly created');
     });
     test('label', () {
-      final NavMenuItemElement e = new NavMenuItemElement(label, link: link);
+      final e = new NavMenuItemElement(label, link: link);
       expect(e, isNotNull, reason: 'element correctly created');
       expect(e.link, equals(link), reason: 'element correctly created');
     });
@@ -25,7 +25,7 @@
   group('elements', () {
     test('created', () async {
       final label = 'custom-label';
-      final NavMenuItemElement e = new NavMenuItemElement(label);
+      final e = new NavMenuItemElement(label);
       document.body.append(e);
       await e.onRendered.first;
       expect(e.shadowRoot.children.length, isNonZero, reason: 'has elements');
@@ -38,7 +38,7 @@
     test('react to label change', () async {
       final label1 = 'custom-label-1';
       final label2 = 'custom-label-2';
-      final NavMenuItemElement e = new NavMenuItemElement(label1);
+      final e = new NavMenuItemElement(label1);
       document.body.append(e);
       await e.onRendered.first;
       expect(e.shadowRoot.innerHtml.contains(label1), isTrue);
@@ -54,7 +54,7 @@
       final label = 'custom-label';
       final link1 = 'custom-label-1';
       final link2 = 'custom-label-2';
-      final NavMenuItemElement e = new NavMenuItemElement(label, link: link1);
+      final e = new NavMenuItemElement(label, link: link1);
       document.body.append(e);
       await e.onRendered.first;
       expect(e.shadowRoot.innerHtml.contains(link1), isTrue);

runtime/observatory/tests/observatory_ui/nav/menu/element_test.dart

diff --git a/runtime/observatory/tests/observatory_ui/nav/menu/element_test.dart b/runtime/observatory/tests/observatory_ui/nav/menu/element_test.dart
index 0cabc3c..7e32086 100644
--- a/runtime/observatory/tests/observatory_ui/nav/menu/element_test.dart
+++ b/runtime/observatory/tests/observatory_ui/nav/menu/element_test.dart
@@ -11,17 +11,17 @@
   group('instantiation', () {
     final label = 'custom-label';
     test('label', () {
-      final NavMenuElement e = new NavMenuElement(label);
+      final e = new NavMenuElement(label);
       expect(e, isNotNull, reason: 'element correctly created');
       expect(e.label, equals(label), reason: 'element correctly created');
     });
     test('not last', () {
-      final NavMenuElement e = new NavMenuElement(label, last: false);
+      final e = new NavMenuElement(label, last: false);
       expect(e, isNotNull, reason: 'element correctly created');
       expect(e.last, isFalse, reason: 'element correctly created');
     });
     test('last', () {
-      final NavMenuElement e = new NavMenuElement(label, last: true);
+      final e = new NavMenuElement(label, last: true);
       expect(e, isNotNull, reason: 'element correctly created');
       expect(e.last, isTrue, reason: 'element correctly created');
     });
@@ -29,7 +29,7 @@
   group('elements', () {
     test('created', () async {
       final label = 'custom-label';
-      final NavMenuElement e = new NavMenuElement(label);
+      final e = new NavMenuElement(label);
       document.body.append(e);
       await e.onRendered.first;
       expect(e.shadowRoot.children.length, isNonZero, reason: 'has elements');
@@ -42,7 +42,7 @@
     test('react to label change', () async {
       final label1 = 'custom-label-1';
       final label2 = 'custom-label-2';
-      final NavMenuElement e = new NavMenuElement(label1);
+      final e = new NavMenuElement(label1);
       document.body.append(e);
       await e.onRendered.first;
       expect(e.shadowRoot.innerHtml.contains(label1), isTrue);
@@ -56,7 +56,7 @@
     });
     test('react to last change', () async {
       final label = 'custom-label';
-      final NavMenuElement e = new NavMenuElement(label, last: false);
+      final e = new NavMenuElement(label, last: false);
       document.body.append(e);
       await e.onRendered.first;
       expect(e.shadowRoot.innerHtml.contains('>'), isTrue);

runtime/observatory/tests/observatory_ui/nav/notify/element_test.dart

diff --git a/runtime/observatory/tests/observatory_ui/nav/notify/element_test.dart b/runtime/observatory/tests/observatory_ui/nav/notify/element_test.dart
index 64adcfb..9cee6cf 100644
--- a/runtime/observatory/tests/observatory_ui/nav/notify/element_test.dart
+++ b/runtime/observatory/tests/observatory_ui/nav/notify/element_test.dart
@@ -3,11 +3,11 @@
 // BSD-style license that can be found in the LICENSE file.
 import 'dart:html' hide Notification, NotificationEvent;
 import 'package:unittest/unittest.dart';
-import 'package:observatory/mocks.dart';
 import 'package:observatory/models.dart' as M;
 import 'package:observatory/src/elements/nav/notify.dart';
 import 'package:observatory/src/elements/nav/notify_event.dart';
 import 'package:observatory/src/elements/nav/notify_exception.dart';
+import '../../mocks.dart';
 
 main() {
   NavNotifyElement.tag.ensureRegistration();
@@ -19,29 +19,29 @@
   const isolate = const IsolateRefMock(id: 'i-id', name: 'i-name');
 
   group('instantiation', () {
+    NotificationRepositoryMock repository;
+    setUp(() {
+      repository = new NotificationRepositoryMock();
+    });
     test('default', () {
-      final NavNotifyElement e = new NavNotifyElement(
-          new NotificationRepositoryMock());
+      final e = new NavNotifyElement(repository);
       expect(e, isNotNull, reason: 'element correctly created');
       expect(e.notifyOnPause, isTrue, reason: 'notifyOnPause is default');
     });
     test('notify on pause', () {
-      final NavNotifyElement e = new NavNotifyElement(
-          new NotificationRepositoryMock(), notifyOnPause: true);
+      final e = new NavNotifyElement(repository, notifyOnPause: true);
       expect(e, isNotNull, reason: 'element correctly created');
       expect(e.notifyOnPause, isTrue, reason: 'notifyOnPause is the same');
     });
     test('do not notify on pause', () {
-      final NavNotifyElement e = new NavNotifyElement(
-          new NotificationRepositoryMock(), notifyOnPause: false);
+      final e = new NavNotifyElement(repository, notifyOnPause: false);
       expect(e, isNotNull, reason: 'element correctly created');
       expect(e.notifyOnPause, isFalse, reason: 'notifyOnPause is the same');
     });
   });
   test('is correctly listening', () async {
-    final NotificationRepositoryMock repository =
-        new NotificationRepositoryMock();
-    final NavNotifyElement e = new NavNotifyElement(repository);
+    final repository = new NotificationRepositoryMock();
+    final e = new NavNotifyElement(repository);
     document.body.append(e);
     await e.onRendered.first;
     expect(repository.hasListeners, isTrue, reason: 'is listening');
@@ -51,13 +51,13 @@
   });
   group('elements', () {
     test('created after attachment', () async {
-      final NotificationRepositoryMock repository =
+      final repository =
           new NotificationRepositoryMock(list: [
             new ExceptionNotificationMock(exception: new Exception("ex")),
             const EventNotificationMock(event: const VMUpdateEventMock(vm: vm)),
             const EventNotificationMock(event: const VMUpdateEventMock(vm: vm))
           ]);
-      final NavNotifyElement e = new NavNotifyElement(repository);
+      final e = new NavNotifyElement(repository);
       document.body.append(e);
       await e.onRendered.first;
       expect(repository.listInvoked, isTrue, reason: 'should invoke list()');
@@ -76,8 +76,7 @@
             const EventNotificationMock(
                 event: const PauseStartEventMock(isolate: isolate))
           ]);
-      final NavNotifyElement e = new NavNotifyElement(repository,
-          notifyOnPause: true);
+      final e = new NavNotifyElement(repository, notifyOnPause: true);
       document.body.append(e);
       await e.onRendered.first;
       expect(e.querySelectorAll(evTag).length, equals(2));
@@ -99,10 +98,8 @@
         new ExceptionNotificationMock(exception: new Exception("ex")),
         const EventNotificationMock(event: const VMUpdateEventMock()),
       ];
-      final NotificationRepositoryMock repository =
-          new NotificationRepositoryMock(list: list);
-      final NavNotifyElement e = new NavNotifyElement(repository,
-          notifyOnPause: true);
+      final repository = new NotificationRepositoryMock(list: list);
+      final e = new NavNotifyElement(repository, notifyOnPause: true);
       document.body.append(e);
       await e.onRendered.first;
       expect(e.querySelectorAll(evTag).length, equals(1));

runtime/observatory/tests/observatory_ui/nav/notify_event/element_test.dart

diff --git a/runtime/observatory/tests/observatory_ui/nav/notify_event/element_test.dart b/runtime/observatory/tests/observatory_ui/nav/notify_event/element_test.dart
index 5d5de5c..9cea3f5 100644
--- a/runtime/observatory/tests/observatory_ui/nav/notify_event/element_test.dart
+++ b/runtime/observatory/tests/observatory_ui/nav/notify_event/element_test.dart
@@ -4,22 +4,22 @@
 import 'dart:html';
 import 'dart:async';
 import 'package:unittest/unittest.dart';
-import 'package:observatory/mocks.dart';
 import 'package:observatory/src/elements/nav/notify_event.dart';
+import '../../mocks.dart';
 
 main() {
   NavNotifyEventElement.tag.ensureRegistration();
 
-  final PauseStartEventMock event = new PauseStartEventMock(
+  final event = new PauseStartEventMock(
               isolate: new IsolateMock(id: 'isolate-id', name: 'isolate-name'));
   group('instantiation', () {
-    final NavNotifyEventElement e = new NavNotifyEventElement(event);
+    final e = new NavNotifyEventElement(event);
     expect(e, isNotNull, reason: 'element correctly created');
     expect(e.event, equals(event));
   });
   group('elements', () {
     test('created after attachment', () async {
-      final NavNotifyEventElement e = new NavNotifyEventElement(event);
+      final e = new NavNotifyEventElement(event);
       document.body.append(e);
       await e.onRendered.first;
       expect(e.children.length, isNonZero, reason: 'has elements');

runtime/observatory/tests/observatory_ui/nav/notify_exception/connection_exception_test.dart

diff --git a/runtime/observatory/tests/observatory_ui/nav/notify_exception/connection_exception_test.dart b/runtime/observatory/tests/observatory_ui/nav/notify_exception/connection_exception_test.dart
index e6b3bc2..f09ed32 100644
--- a/runtime/observatory/tests/observatory_ui/nav/notify_exception/connection_exception_test.dart
+++ b/runtime/observatory/tests/observatory_ui/nav/notify_exception/connection_exception_test.dart
@@ -4,23 +4,19 @@
 import 'dart:html';
 import 'dart:async';
 import 'package:unittest/unittest.dart';
-import 'package:observatory/models.dart';
-import 'package:observatory/mocks.dart';
 import 'package:observatory/src/elements/nav/notify_exception.dart';
+import '../../mocks.dart';
 
 main() {
   NavNotifyExceptionElement.tag.ensureRegistration();
 
-  final ConnectionException exception =
-      new ConnectionExceptionMock(message: 'message');
+  final exception = new ConnectionExceptionMock(message: 'message');
   test('instantiation', () {
-    final NavNotifyExceptionElement e =
-                                new NavNotifyExceptionElement(exception);
+    final e = new NavNotifyExceptionElement(exception);
     expect(e, isNotNull, reason: 'element correctly created');
   });
   test('elements created after attachment', () async {
-    final NavNotifyExceptionElement e =
-                                new NavNotifyExceptionElement(exception);
+    final e = new NavNotifyExceptionElement(exception);
     document.body.append(e);
     await e.onRendered.first;
     expect(e.children.length, isNonZero, reason: 'has elements');

runtime/observatory/tests/observatory_ui/nav/notify_exception/exception_test.dart

diff --git a/runtime/observatory/tests/observatory_ui/nav/notify_exception/exception_test.dart b/runtime/observatory/tests/observatory_ui/nav/notify_exception/exception_test.dart
index be9fdc5..4696ee8 100644
--- a/runtime/observatory/tests/observatory_ui/nav/notify_exception/exception_test.dart
+++ b/runtime/observatory/tests/observatory_ui/nav/notify_exception/exception_test.dart
@@ -9,46 +9,42 @@
 main() {
   NavNotifyExceptionElement.tag.ensureRegistration();
 
-  final StackTrace stacktrace = new StackTrace.fromString('stacktrace string');
+  final st = new StackTrace.fromString('stacktrace string');
   group('normal exception', () {
-    final Exception exception = new Exception('exception message');
+    final ex = new Exception('exception message');
     group('instantiation', () {
       test('no stacktrace', () {
-        final NavNotifyExceptionElement e =
-                                      new NavNotifyExceptionElement(exception);
+        final e = new NavNotifyExceptionElement(ex);
         expect(e, isNotNull, reason: 'element correctly created');
-        expect(e.exception, equals(exception));
+        expect(e.exception, equals(ex));
         expect(e.stacktrace, isNull);
       });
       test('with stacktrace', () {
-        final NavNotifyExceptionElement e =
-              new NavNotifyExceptionElement(exception, stacktrace: stacktrace);
+        final e = new NavNotifyExceptionElement(ex, stacktrace: st);
         expect(e, isNotNull, reason: 'element correctly created');
-        expect(e.exception, equals(exception));
-        expect(e.stacktrace, equals(stacktrace));
+        expect(e.exception, equals(ex));
+        expect(e.stacktrace, equals(st));
       });
     });
     group('elements', () {
       test('created after attachment (no stacktrace)', () async {
-        final NavNotifyExceptionElement e =
-                                      new NavNotifyExceptionElement(exception);
+        final e = new NavNotifyExceptionElement(ex);
         document.body.append(e);
         await e.onRendered.first;
         expect(e.children.length, isNonZero, reason: 'has elements');
-        expect(e.innerHtml.contains(exception.toString()), isTrue);
-        expect(e.innerHtml.contains(stacktrace.toString()), isFalse);
+        expect(e.innerHtml.contains(ex.toString()), isTrue);
+        expect(e.innerHtml.contains(st.toString()), isFalse);
         e.remove();
         await e.onRendered.first;
         expect(e.children.length, isZero, reason: 'is empty');
       });
       test('created after attachment (with stacktrace)', () async {
-        final NavNotifyExceptionElement e =
-              new NavNotifyExceptionElement(exception, stacktrace: stacktrace);
+        final e = new NavNotifyExceptionElement(ex, stacktrace: st);
         document.body.append(e);
         await e.onRendered.first;
         expect(e.children.length, isNonZero, reason: 'has elements');
-        expect(e.innerHtml.contains(exception.toString()), isTrue);
-        expect(e.innerHtml.contains(stacktrace.toString()), isTrue);
+        expect(e.innerHtml.contains(ex.toString()), isTrue);
+        expect(e.innerHtml.contains(st.toString()), isTrue);
         e.remove();
         await e.onRendered.first;
         expect(e.children.length, isZero, reason: 'is empty');
@@ -58,7 +54,7 @@
       NavNotifyExceptionElement e;
       StreamSubscription sub;
       setUp(() async {
-        e = new NavNotifyExceptionElement(exception, stacktrace: stacktrace);
+        e = new NavNotifyExceptionElement(ex, stacktrace: st);
         document.body.append(e);
         await e.onRendered.first;
       });
@@ -74,9 +70,8 @@
       test('onDelete events (DOM)', () async {
         sub = e.onDelete.listen(expectAsync((ExceptionDeleteEvent event) {
           expect(event, isNotNull, reason: 'event is passed');
-          expect(event.exception, equals(exception),
-                                              reason: 'exception is the same');
-          expect(event.stacktrace, equals(stacktrace),
+          expect(event.exception, equals(ex), reason: 'exception is the same');
+          expect(event.stacktrace, equals(st),
                                             reason: 'stacktrace is the same');
         }, count: 1, reason: 'event is fired'));
         e.querySelector('button').click();
@@ -84,9 +79,8 @@
       test('onDelete events (code)', () async {
         sub = e.onDelete.listen(expectAsync((ExceptionDeleteEvent event) {
           expect(event, isNotNull, reason: 'event is passed');
-          expect(event.exception, equals(exception),
-                                              reason: 'exception is the same');
-          expect(event.stacktrace, equals(stacktrace),
+          expect(event.exception, equals(ex), reason: 'exception is the same');
+          expect(event.stacktrace, equals(st),
                                             reason: 'stacktrace is the same');
         }, count: 1, reason: 'event is fired'));
         e.delete();

runtime/observatory/tests/observatory_ui/nav/refresh/element_test.dart

diff --git a/runtime/observatory/tests/observatory_ui/nav/refresh/element_test.dart b/runtime/observatory/tests/observatory_ui/nav/refresh/element_test.dart
index 26c3ada..05b3370 100644
--- a/runtime/observatory/tests/observatory_ui/nav/refresh/element_test.dart
+++ b/runtime/observatory/tests/observatory_ui/nav/refresh/element_test.dart
@@ -8,34 +8,35 @@
 
 main() {
   NavRefreshElement.tag.ensureRegistration();
+
   group('instantiation', () {
     test('no parameters', () {
-      final NavRefreshElement e = new NavRefreshElement();
+      final e = new NavRefreshElement();
       expect(e, isNotNull, reason: 'element correctly created');
       expect(e.label, isNotNull, reason: 'label is set to default');
       expect(e.disabled, isFalse, reason: 'element correctly created');
     });
     test('label', () {
       final label = 'custom-label';
-      final NavRefreshElement e = new NavRefreshElement(label: label);
+      final e = new NavRefreshElement(label: label);
       expect(e, isNotNull, reason: 'element correctly created');
       expect(e.label, isNotNull, reason: 'label is set');
       expect(e.label, equals(label), reason: 'label is set to value');
     });
     test('not disabled', () {
-      final NavRefreshElement e = new NavRefreshElement(disabled: false);
+      final e = new NavRefreshElement(disabled: false);
       expect(e, isNotNull, reason: 'element correctly created');
       expect(e.disabled, isFalse, reason: 'element correctly created');
     });
     test('disabled', () {
-      final NavRefreshElement e = new NavRefreshElement(disabled: true);
+      final e = new NavRefreshElement(disabled: true);
       expect(e, isNotNull, reason: 'element correctly created');
       expect(e.disabled, isTrue, reason: 'element correctly created');
     });
   });
   group('elements', () {
     test('created after attachment', () async {
-      final NavRefreshElement e = new NavRefreshElement();
+      final e = new NavRefreshElement();
       document.body.append(e);
       await e.onRendered.first;
       expect(e.children.length, isNonZero, reason: 'has elements');
@@ -45,7 +46,7 @@
     });
     test('contain custom label', () async {
       final label = 'custom-label';
-      final NavRefreshElement e = new NavRefreshElement(label: label);
+      final e = new NavRefreshElement(label: label);
       document.body.append(e);
       await e.onRendered.first;
       expect(e.innerHtml.contains(label), isTrue);
@@ -55,7 +56,7 @@
     test('react to label change', () async {
       final label1 = 'custom-label-1';
       final label2 = 'custom-label-2';
-      final NavRefreshElement e = new NavRefreshElement(label: label1);
+      final e = new NavRefreshElement(label: label1);
       document.body.append(e);
       await e.onRendered.first;
       expect(e.innerHtml.contains(label1), isTrue);
@@ -68,7 +69,7 @@
       await e.onRendered.first;
     });
     test('react to disabled change', () async {
-      final NavRefreshElement e = new NavRefreshElement(disabled: false);
+      final e = new NavRefreshElement(disabled: false);
       document.body.append(e);
       await e.onRendered.first;
       expect(e.disabled, isFalse);

runtime/observatory/tests/observatory_ui/nav/top-menu/element_test.dart

diff --git a/runtime/observatory/tests/observatory_ui/nav/top-menu/element_test.dart b/runtime/observatory/tests/observatory_ui/nav/top-menu/element_test.dart
index df9f88c..5e62069 100644
--- a/runtime/observatory/tests/observatory_ui/nav/top-menu/element_test.dart
+++ b/runtime/observatory/tests/observatory_ui/nav/top-menu/element_test.dart
@@ -9,27 +9,27 @@
 main() {
   NavTopMenuElement.tag.ensureRegistration();
 
-  final String tag = NavMenuElement.tag.name;
+  final tag = NavMenuElement.tag.name;
 
   group('instantiation', () {
     test('default', () {
-      final NavTopMenuElement e = new NavTopMenuElement();
+      final e = new NavTopMenuElement();
       expect(e, isNotNull, reason: 'element correctly created');
     });
     test('not last', () {
-      final NavTopMenuElement e = new NavTopMenuElement(last: false);
+      final e = new NavTopMenuElement(last: false);
       expect(e, isNotNull, reason: 'element correctly created');
       expect(e.last, isFalse, reason: 'element correctly created');
     });
     test('last', () {
-      final NavTopMenuElement e = new NavTopMenuElement(last: true);
+      final e = new NavTopMenuElement(last: true);
       expect(e, isNotNull, reason: 'element correctly created');
       expect(e.last, isTrue, reason: 'element correctly created');
     });
   });
   group('elements', () {
     test('created', () async {
-      final NavTopMenuElement e = new NavTopMenuElement();
+      final e = new NavTopMenuElement();
       document.body.append(e);
       await e.onRendered.first;
       expect(e.shadowRoot.children.length, isNonZero, reason: 'has elements');
@@ -40,7 +40,7 @@
       expect(e.shadowRoot.children.length, isZero, reason: 'is empty');
     });
     test('react to last change', () async {
-      final NavTopMenuElement e = new NavTopMenuElement(last: false);
+      final e = new NavTopMenuElement(last: false);
       document.body.append(e);
       await e.onRendered.first;
       expect((e.shadowRoot.querySelector(tag) as NavMenuElement).last, isFalse);

runtime/observatory/tests/observatory_ui/nav/vm_menu/element_test.dart

diff --git a/runtime/observatory/tests/observatory_ui/nav/vm_menu/element_test.dart b/runtime/observatory/tests/observatory_ui/nav/vm_menu/element_test.dart
index b0539ff..7aba8ed 100644
--- a/runtime/observatory/tests/observatory_ui/nav/vm_menu/element_test.dart
+++ b/runtime/observatory/tests/observatory_ui/nav/vm_menu/element_test.dart
@@ -2,13 +2,11 @@
 // for details. All rights reserved. Use of this source code is governed by a
 // BSD-style license that can be found in the LICENSE file.
 import 'dart:html';
-import 'dart:async';
 import 'package:unittest/unittest.dart';
-import 'package:observatory/mocks.dart';
-import 'package:observatory/models.dart' as M;
 import 'package:observatory/src/elements/nav/menu.dart';
 import 'package:observatory/src/elements/nav/menu_item.dart';
 import 'package:observatory/src/elements/nav/vm_menu.dart';
+import '../../mocks.dart';
 
 main(){
   NavVMMenuElement.tag.ensureRegistration();
@@ -16,35 +14,22 @@
   final mTag = NavMenuElement.tag.name;
   final miTag = NavMenuItemElement.tag.name;
 
-  StreamController<M.VMUpdateEvent> updatesController;
-  final TargetMock target = new TargetMock(name: 'target-name');
-  final VMMock vm1 = const VMMock(name: 'vm-name-1',
+  EventRepositoryMock events;
+  final vm1 = const VMMock(name: 'vm-name-1', displayName: 'display-name-1',
       isolates: const [const IsolateRefMock(id: 'i-id-1', name: 'i-name-1')]);
-  final VMMock vm2 = const VMMock(name: 'vm-name-2',
+  final vm2 = const VMMock(name: 'vm-name-2', displayName: 'display-name-2',
       isolates: const [const IsolateRefMock(id: 'i-id-1', name: 'i-name-1'),
                        const IsolateRefMock(id: 'i-id-2', name: 'i-name-2')]);
   setUp(() {
-    updatesController = new StreamController<M.VMUpdateEvent>.broadcast();
+    events = new EventRepositoryMock();
   });
-  group('instantiation', () {
-    test('no target', () {
-      final NavVMMenuElement e = new NavVMMenuElement(vm1,
-          updatesController.stream);
-      expect(e, isNotNull, reason: 'element correctly created');
-      expect(e.vm, equals(vm1));
-      expect(e.target, isNull);
-    });
-    test('target', () {
-      final NavVMMenuElement e = new NavVMMenuElement(vm1,
-          updatesController.stream, target: target);
-      expect(e, isNotNull, reason: 'element correctly created');
-      expect(e.vm, equals(vm1));
-      expect(e.target, equals(target));
-    });
+  test('instantiation', () {
+    final e = new NavVMMenuElement(vm1, events);
+    expect(e, isNotNull, reason: 'element correctly created');
+    expect(e.vm, equals(vm1));
   });
   test('elements created after attachment', () async {
-    final NavVMMenuElement e = new NavVMMenuElement(vm1,
-        updatesController.stream);
+    final e = new NavVMMenuElement(vm1, events);
     document.body.append(e);
     await e.onRendered.first;
     expect(e.shadowRoot.children.length, isNonZero, reason: 'has elements');
@@ -54,29 +39,27 @@
   });
   group('updates', () {
     test('are correctly listen', () async {
-      final NavVMMenuElement e = new NavVMMenuElement(vm1,
-          updatesController.stream);
-      expect(updatesController.hasListener, isFalse);
+      final e = new NavVMMenuElement(vm1, events);
+      expect(events.onVMUpdateHasListener, isFalse);
       document.body.append(e);
       await e.onRendered.first;
-      expect(updatesController.hasListener, isTrue);
+      expect(events.onVMUpdateHasListener, isTrue);
       e.remove();
       await e.onRendered.first;
-      expect(updatesController.hasListener, isFalse);
+      expect(events.onVMUpdateHasListener, isFalse);
     });
     test('have effects', () async {
-      final NavVMMenuElement e = new NavVMMenuElement(vm1,
-          updatesController.stream);
+      final e = new NavVMMenuElement(vm1, events);
       document.body.append(e);
       await e.onRendered.first;
       expect((e.shadowRoot.querySelector(mTag) as NavMenuElement).label,
-          equals(vm1.name));
+          equals(vm1.displayName));
       expect(e.shadowRoot.querySelectorAll(miTag).length,
           equals(vm1.isolates.length));
-      updatesController.add(new VMUpdateEventMock(vm: vm2));
+      events.add(new VMUpdateEventMock(vm: vm2));
       await e.onRendered.first;
       expect((e.shadowRoot.querySelector(mTag) as NavMenuElement).label,
-          equals(vm2.name));
+          equals(vm2.displayName));
       expect(e.shadowRoot.querySelectorAll(miTag).length,
           equals(vm2.isolates.length));
       e.remove();

runtime/observatory/tests/observatory_ui/script_ref/element_test.dart

diff --git a/runtime/observatory/tests/observatory_ui/script_ref/element_test.dart b/runtime/observatory/tests/observatory_ui/script_ref/element_test.dart
index eda7d49..6dc2b84 100644
--- a/runtime/observatory/tests/observatory_ui/script_ref/element_test.dart
+++ b/runtime/observatory/tests/observatory_ui/script_ref/element_test.dart
@@ -3,32 +3,31 @@
 // BSD-style license that can be found in the LICENSE file.
 import 'dart:html';
 import 'package:unittest/unittest.dart';
-import 'package:observatory/mocks.dart';
-import 'package:observatory/models.dart' as M;
 import 'package:observatory/src/elements/script_ref.dart';
+import '../mocks.dart';
 
 main() {
   ScriptRefElement.tag.ensureRegistration();
 
-  final M.IsolateRef isolate = const IsolateRefMock(id: 'isolate-id');
-  final M.ScriptRef ref = const ScriptRefMock(id: 'script-id',
-                                  uri: 'package/filename.dart');
+  const isolate = const IsolateRefMock(id: 'isolate-id');
+  const file = 'filename.dart';
+  const ref = const ScriptRefMock(id: 'script-id', uri: 'package/$file');
   group('instantiation', () {
     test('no position', () {
-      final ScriptRefElement e = new ScriptRefElement(isolate, ref);
+      final e = new ScriptRefElement(isolate, ref);
       expect(e, isNotNull, reason: 'element correctly created');
       expect(e.isolate, equals(isolate));
       expect(e.script, equals(ref));
     });
   });
   test('elements created after attachment', () async {
-    final ScriptRefElement e = new ScriptRefElement(isolate, ref);
+    final e = new ScriptRefElement(isolate, ref);
     document.body.append(e);
     await e.onRendered.first;
     expect(e.children.length, isNonZero, reason: 'has elements');
     expect(e.innerHtml.contains(isolate.id), isTrue,
       reason: 'no message in the component');
-    expect(e.innerHtml.contains('filename.dart'), isTrue,
+    expect(e.innerHtml.contains(file), isTrue,
       reason: 'no message in the component');
     e.remove();
     await e.onRendered.first;

runtime/observatory/tests/observatory_ui/source_link/element_test.dart

diff --git a/runtime/observatory/tests/observatory_ui/source_link/element_test.dart b/runtime/observatory/tests/observatory_ui/source_link/element_test.dart
index 2d4cfba..3dcb5dd 100644
--- a/runtime/observatory/tests/observatory_ui/source_link/element_test.dart
+++ b/runtime/observatory/tests/observatory_ui/source_link/element_test.dart
@@ -3,40 +3,41 @@
 // BSD-style license that can be found in the LICENSE file.
 import 'dart:html';
 import 'package:unittest/unittest.dart';
-import 'package:observatory/mocks.dart';
-import 'package:observatory/models.dart' as M;
 import 'package:observatory/src/elements/source_link.dart';
+import '../mocks.dart';
 
 main() {
   SourceLinkElement.tag.ensureRegistration();
 
-  final M.IsolateRef isolate = const IsolateRefMock(id: 'isolate-id');
-  final M.Script script = new  ScriptMock(id: 'script-id',
-                                  uri: 'package/filename.dart',
-                                  tokenToLine: (int token) => 1,
-                                  tokenToCol: (int token) => 2);
-  final M.SourceLocation location = new SourceLocationMock(script: script,
-                                  tokenPos: 0, endTokenPos: 1);
-  M.ScriptRepository repository;
-  setUp(() {
-    repository = new ScriptRepositoryMock({ 'script-id': script });
-  });
+  const isolate = const IsolateRefMock(id: 'isolate-id');
+  const script_id = 'script-id';
+  const file = 'filename.dart';
+  final script = new ScriptMock(id: script_id, uri: 'package/$file',
+      tokenToLine: (int token) => 1, tokenToCol: (int token) => 2);
+  final location = new SourceLocationMock(script: script, tokenPos: 0,
+      endTokenPos: 1);
   test('instantiation', () {
-    final SourceLinkElement e = new SourceLinkElement(isolate, location,
-                                                               repository);
+    final e = new SourceLinkElement(isolate, location,
+        new ScriptRepositoryMock());
     expect(e, isNotNull, reason: 'element correctly created');
     expect(e.isolate, equals(isolate));
     expect(e.location, equals(location));
   });
   test('elements created after attachment', () async {
-    final SourceLinkElement e = new SourceLinkElement(isolate, location,
-                                                               repository);
+    bool rendered = false;
+    final e = new SourceLinkElement(isolate, location,
+        new ScriptRepositoryMock(getter: expectAsync((String id) async {
+          expect(rendered, isFalse);
+          expect(id, equals(script_id));
+          return script;
+        }, count: 1)));
     document.body.append(e);
     await e.onRendered.first;
+    rendered = true;
     expect(e.children.length, isNonZero, reason: 'has elements');
     expect(e.innerHtml.contains(isolate.id), isTrue,
       reason: 'no message in the component');
-    expect(e.innerHtml.contains('filename.dart'), isTrue,
+    expect(e.innerHtml.contains(file), isTrue,
       reason: 'no message in the component');
     e.remove();
     await e.onRendered.first;

runtime/observatory/tests/observatory_ui/view_footer/element_test.dart

diff --git a/runtime/observatory/tests/observatory_ui/view_footer/element_test.dart b/runtime/observatory/tests/observatory_ui/view_footer/element_test.dart
index f3ab727..143464e 100644
--- a/runtime/observatory/tests/observatory_ui/view_footer/element_test.dart
+++ b/runtime/observatory/tests/observatory_ui/view_footer/element_test.dart
@@ -9,11 +9,11 @@
   ViewFooterElement.tag.ensureRegistration();
 
   test('instantiation', () {
-    final ViewFooterElement e = new ViewFooterElement();
+    final e = new ViewFooterElement();
     expect(e, isNotNull, reason: 'element correctly created');
   });
   test('elements created', () async {
-    final ViewFooterElement e = new ViewFooterElement();
+    final e = new ViewFooterElement();
     document.body.append(e);
     await e.onRendered.first;
     expect(e.children.length, isNonZero, reason: 'has elements');

runtime/observatory/tests/observatory_ui/vm_connect/element_test.dart

diff --git a/runtime/observatory/tests/observatory_ui/vm_connect/element_test.dart b/runtime/observatory/tests/observatory_ui/vm_connect/element_test.dart
index 15a853ca..270e0d5 100644
--- a/runtime/observatory/tests/observatory_ui/vm_connect/element_test.dart
+++ b/runtime/observatory/tests/observatory_ui/vm_connect/element_test.dart
@@ -3,31 +3,30 @@
 // BSD-style license that can be found in the LICENSE file.
 import 'dart:html';
 import 'package:unittest/unittest.dart';
-import 'package:observatory/mocks.dart';
-import 'package:observatory/models.dart' as M;
 import 'package:observatory/src/elements/nav/notify.dart';
 import 'package:observatory/src/elements/vm_connect_target.dart';
 import 'package:observatory/src/elements/vm_connect.dart';
+import '../mocks.dart';
+
+void load(_) {}
 
 main() {
   VMConnectElement.tag.ensureRegistration();
 
-  final String nTag = NavNotifyElement.tag.name;
-  final String tTag = VMConnectTargetElement.tag.name;
+  final nTag = NavNotifyElement.tag.name;
+  final tTag = VMConnectTargetElement.tag.name;
 
   group('instantiation', () {
     test('default', () {
-      final VMConnectElement e = new VMConnectElement(
-          new TargetRepositoryMock(),
-          new CrashDumpRepositoryMock(),
-          new NotificationRepositoryMock());
+      final e = new VMConnectElement(new TargetRepositoryMock(),
+          load, new NotificationRepositoryMock());
       expect(e, isNotNull, reason: 'element correctly created');
     });
   });
   test('is correctly listening', () async {
     final targets = new TargetRepositoryMock();
-    final VMConnectElement e = new VMConnectElement(targets,
-        new CrashDumpRepositoryMock(), new NotificationRepositoryMock());
+    final e = new VMConnectElement(targets, load,
+        new NotificationRepositoryMock());
     document.body.append(e);
     await e.onRendered.first;
     expect(targets.hasListeners, isTrue, reason: 'is listening');
@@ -40,8 +39,8 @@
       final targets = new TargetRepositoryMock(list: const [
           const TargetMock(name: 't-1'), const TargetMock(name: 't-2'),
           ]);
-      final VMConnectElement e = new VMConnectElement(targets,
-          new CrashDumpRepositoryMock(), new NotificationRepositoryMock());
+      final e = new VMConnectElement(targets, load,
+          new NotificationRepositoryMock());
       document.body.append(e);
       await e.onRendered.first;
       expect(targets.listInvoked, isTrue, reason: 'should invoke list()');
@@ -54,10 +53,10 @@
       expect(e.children.length, isZero, reason: 'is empty');
     });
     test('react to update event', () async {
-      final list = <M.Target>[const TargetMock(name: 't-1')];
+      final list = <TargetMock>[const TargetMock(name: 't-1')];
       final targets = new TargetRepositoryMock(list: list);
-      final VMConnectElement e = new VMConnectElement(targets,
-          new CrashDumpRepositoryMock(), new NotificationRepositoryMock());
+      final e = new VMConnectElement(targets, load,
+          new NotificationRepositoryMock());
       document.body.append(e);
       await e.onRendered.first;
       expect(e.querySelectorAll(tTag).length, equals(1));
@@ -73,14 +72,13 @@
   group('invokes', () {
     test('add on click', () async {
       final address = 'ws://host:1234';
-      final list = <M.Target>[const TargetMock(name: 't-1')];
+      final list = <TargetMock>[const TargetMock(name: 't-1')];
       final targets = new TargetRepositoryMock(list: list,
           add: expectAsync((String val) {
             expect(val, equals(address));
           }, count: 1, reason: 'should be invoked'));
-      final VMConnectElement e = new VMConnectElement(targets,
-          new CrashDumpRepositoryMock(), new NotificationRepositoryMock(),
-          address: address);
+      final e = new VMConnectElement(targets, load,
+          new NotificationRepositoryMock(), address: address);
       document.body.append(e);
       await e.onRendered.first;
       (e.querySelector('button.vm_connect') as ButtonElement).click();
@@ -88,13 +86,13 @@
       await e.onRendered.first;
     });
     test('connect', () async {
-      final list = <M.Target>[const TargetMock(name: 't-1')];
+      final list = <TargetMock>[const TargetMock(name: 't-1')];
       final targets = new TargetRepositoryMock(list: list,
-          setCurrent: expectAsync((M.Target t) {
+          setCurrent: expectAsync((TargetMock t) {
             expect(t, equals(list[0]));
           }, count: 1, reason: 'should be invoked'));
-      final VMConnectElement e = new VMConnectElement(targets,
-          new CrashDumpRepositoryMock(), new NotificationRepositoryMock());
+      final e = new VMConnectElement(targets, load,
+          new NotificationRepositoryMock());
       document.body.append(e);
       await e.onRendered.first;
       (e.querySelector(tTag) as VMConnectTargetElement).connect();
@@ -102,13 +100,13 @@
       await e.onRendered.first;
     });
     test('delete', () async {
-      final list = <M.Target>[const TargetMock(name: 't-1')];
+      final list = <TargetMock>[const TargetMock(name: 't-1')];
       final targets = new TargetRepositoryMock(list: list,
-          delete: expectAsync((M.Target t) {
+          delete: expectAsync((TargetMock t) {
             expect(t, equals(list[0]));
           }, count: 1, reason: 'should be invoked'));
-      final VMConnectElement e = new VMConnectElement(targets,
-          new CrashDumpRepositoryMock(), new NotificationRepositoryMock());
+      final e = new VMConnectElement(targets, load,
+          new NotificationRepositoryMock());
       document.body.append(e);
       await e.onRendered.first;
       (e.querySelector(tTag) as VMConnectTargetElement).delete();

runtime/observatory/tests/observatory_ui/vm_connect_target/element_test.dart

diff --git a/runtime/observatory/tests/observatory_ui/vm_connect_target/element_test.dart b/runtime/observatory/tests/observatory_ui/vm_connect_target/element_test.dart
index a4340e8..45efab6 100644
--- a/runtime/observatory/tests/observatory_ui/vm_connect_target/element_test.dart
+++ b/runtime/observatory/tests/observatory_ui/vm_connect_target/element_test.dart
@@ -3,8 +3,8 @@
 // BSD-style license that can be found in the LICENSE file.
 import 'dart:html';
 import 'package:unittest/unittest.dart';
-import 'package:observatory/mocks.dart';
 import 'package:observatory/src/elements/vm_connect_target.dart';
+import '../mocks.dart';
 
 main() {
   VMConnectTargetElement.tag.ensureRegistration();
@@ -15,28 +15,26 @@
   });
   group('instantiation', () {
     test('no other parameters', () {
-      final VMConnectTargetElement e = new VMConnectTargetElement(t);
+      final e = new VMConnectTargetElement(t);
       expect(e, isNotNull, reason: 'element correctly created');
       expect(e.target, t, reason: 'target not setted');
       expect(e.current, isFalse, reason: 'default to not current');
     });
     test('isCurrent: false', () {
-      final VMConnectTargetElement e = new VMConnectTargetElement(t,
-                                            current:false);
+      final e = new VMConnectTargetElement(t, current:false);
       expect(e, isNotNull, reason: 'element correctly created');
       expect(e.target, t, reason: 'target not setted');
       expect(e.current, isFalse, reason: 'default to not current');
     });
     test('isCurrent: true', () {
-      final VMConnectTargetElement e = new VMConnectTargetElement(t,
-                                            current:true);
+      final e = new VMConnectTargetElement(t, current:true);
       expect(e, isNotNull, reason: 'element correctly created');
       expect(e.target, t, reason: 'target not setted');
       expect(e.current, isTrue, reason: 'default to not current');
     });
   });
   test('elements created after attachment', () async {
-    final VMConnectTargetElement e = new VMConnectTargetElement(t);
+    final e = new VMConnectTargetElement(t);
     document.body.append(e);
     await e.onRendered.first;
     expect(e.children.length, isNonZero, reason: 'has elements');

runtime/observatory/tests/service/debugger_inspect_test.dart

diff --git a/runtime/observatory/tests/service/debugger_inspect_test.dart b/runtime/observatory/tests/service/debugger_inspect_test.dart
index 1f7fe31..269a123 100644
--- a/runtime/observatory/tests/service/debugger_inspect_test.dart
+++ b/runtime/observatory/tests/service/debugger_inspect_test.dart
@@ -26,7 +26,7 @@
   var subscription;
   subscription = stream.listen((ServiceEvent event) {
     if (event.kind == ServiceEvent.kInspect) {
-      expect((event.inspectee as Instance).clazz.name, equals('Point'));
+      expect(event.inspectee.clazz.name, equals('Point'));
       subscription.cancel();
       completer.complete();
     }

runtime/vm/isolate.cc

diff --git a/runtime/vm/isolate.cc b/runtime/vm/isolate.cc
index 32dc967..52aaa44 100644
--- a/runtime/vm/isolate.cc
+++ b/runtime/vm/isolate.cc
@@ -2595,7 +2595,8 @@
     ASSERT(heap() != NULL);
     thread->heap_ = heap();
     thread->set_os_thread(os_thread);
-    ASSERT(thread->execution_state() == Thread::kThreadInVM);
+    ASSERT(thread->execution_state() == Thread::kThreadInNative);
+    thread->set_execution_state(Thread::kThreadInVM);
     thread->set_safepoint_state(0);
     thread->set_vm_tag(VMTag::kVMTagId);
     ASSERT(thread->no_safepoint_scope_depth() == 0);
@@ -2646,8 +2647,8 @@
   thread->isolate_ = NULL;
   thread->heap_ = NULL;
   thread->set_os_thread(NULL);
-  thread->set_execution_state(Thread::kThreadInVM);
-  thread->set_safepoint_state(0);
+  thread->set_execution_state(Thread::kThreadInNative);
+  thread->set_safepoint_state(Thread::SetAtSafepoint(true, 0));
   thread->clear_pending_functions();
   ASSERT(thread->no_safepoint_scope_depth() == 0);
   // Return thread structure.

runtime/vm/isolate_reload.cc

diff --git a/runtime/vm/isolate_reload.cc b/runtime/vm/isolate_reload.cc
index e635a7b..0957f6f 100644
--- a/runtime/vm/isolate_reload.cc
+++ b/runtime/vm/isolate_reload.cc
@@ -442,6 +442,7 @@
   if (!modified_libs_->Contains(root_lib.index())) {
     ASSERT(modified_libs_->IsEmpty());
     reload_skipped_ = true;
+    ReportOnJSON(js_);
     TIR_Print("---- SKIPPING RELOAD (No libraries were modified)\n");
     return;
   }
@@ -596,14 +597,20 @@
 
 
 void IsolateReloadContext::ReportOnJSON(JSONStream* stream) {
-  // Clear the buffer.
-  stream->buffer()->Clear();
   JSONObject jsobj(stream);
   jsobj.AddProperty("type", "ReloadReport");
-  jsobj.AddProperty("success", !HasReasonsForCancelling());
+  jsobj.AddProperty("success", reload_skipped_ || !HasReasonsForCancelling());
   {
     JSONObject details(&jsobj, "details");
-    if (HasReasonsForCancelling()) {
+    if (reload_skipped_) {
+      // Reload was skipped.
+      const GrowableObjectArray& libs =
+          GrowableObjectArray::Handle(object_store()->libraries());
+      const intptr_t final_library_count = libs.Length();
+      details.AddProperty("savedLibraryCount", final_library_count);
+      details.AddProperty("loadedLibraryCount", static_cast<intptr_t>(0));
+      details.AddProperty("finalLibraryCount", final_library_count);
+    } else if (HasReasonsForCancelling()) {
       // Reload was rejected.
       JSONArray array(&jsobj, "notices");
       for (intptr_t i = 0; i < reasons_to_cancel_reload_.length(); i++) {
@@ -1359,8 +1366,12 @@
 
 
 void IsolateReloadContext::ResetUnoptimizedICsOnStack() {
-  Code& code = Code::Handle();
-  Function& function = Function::Handle();
+  Thread* thread = Thread::Current();
+  StackZone stack_zone(thread);
+  Zone* zone = stack_zone.GetZone();
+
+  Code& code = Code::Handle(zone);
+  Function& function = Function::Handle(zone);
   DartFrameIterator iterator;
   StackFrame* frame = iterator.NextFrame();
   while (frame != NULL) {
@@ -1372,9 +1383,9 @@
       function = code.function();
       code = function.unoptimized_code();
       ASSERT(!code.IsNull());
-      code.ResetICDatas();
+      code.ResetICDatas(zone);
     } else {
-      code.ResetICDatas();
+      code.ResetICDatas(zone);
     }
     frame = iterator.NextFrame();
   }
@@ -1393,13 +1404,15 @@
 class MarkFunctionsForRecompilation : public ObjectVisitor {
  public:
   MarkFunctionsForRecompilation(Isolate* isolate,
-                                IsolateReloadContext* reload_context)
+                                IsolateReloadContext* reload_context,
+                                Zone* zone)
     : ObjectVisitor(),
-      handle_(Object::Handle()),
-      owning_class_(Class::Handle()),
-      owning_lib_(Library::Handle()),
-      code_(Code::Handle()),
-      reload_context_(reload_context) {
+      handle_(Object::Handle(zone)),
+      owning_class_(Class::Handle(zone)),
+      owning_lib_(Library::Handle(zone)),
+      code_(Code::Handle(zone)),
+      reload_context_(reload_context),
+      zone_(zone) {
   }
 
   virtual void VisitObject(RawObject* obj) {
@@ -1453,7 +1466,7 @@
     ASSERT(!code_.IsNull());
     // We are preserving the unoptimized code, fill all ICData arrays with
     // the sentinel values so that we have no stale type feedback.
-    code_.ResetICDatas();
+    code_.ResetICDatas(zone_);
   }
 
   bool IsFromDirtyLibrary(const Function& func) {
@@ -1467,15 +1480,19 @@
   Library& owning_lib_;
   Code& code_;
   IsolateReloadContext* reload_context_;
+  Zone* zone_;
 };
 
 
 void IsolateReloadContext::MarkAllFunctionsForRecompilation() {
   TIMELINE_SCOPE(MarkAllFunctionsForRecompilation);
   TIR_Print("---- MARKING ALL FUNCTIONS FOR RECOMPILATION\n");
+  Thread* thread = Thread::Current();
+  StackZone stack_zone(thread);
+  Zone* zone = stack_zone.GetZone();
   NoSafepointScope no_safepoint;
   HeapIterationScope heap_iteration_scope;
-  MarkFunctionsForRecompilation visitor(isolate_, this);
+  MarkFunctionsForRecompilation visitor(isolate_, this, zone);
   isolate_->heap()->VisitObjects(&visitor);
 }
 

runtime/vm/isolate_reload_test.cc

diff --git a/runtime/vm/isolate_reload_test.cc b/runtime/vm/isolate_reload_test.cc
index 070897e..b6a6c98 100644
--- a/runtime/vm/isolate_reload_test.cc
+++ b/runtime/vm/isolate_reload_test.cc
@@ -1487,6 +1487,354 @@
 }
 
 
+TEST_CASE(IsolateReload_DanglingGetter_Instance) {
+  const char* kScript =
+      "import 'test:isolate_reload_helper';\n"
+      "class C {\n"
+      "  var x = 3;\n"
+      "  var y = 4;\n"
+      "}\n"
+      "invoke(f) {\n"
+      "  try {\n"
+      "    return f();\n"
+      "  } catch (e) {\n"
+      "    return e.toString().split('\\n').first;\n"
+      "  }\n"
+      "}\n"
+      "main() {\n"
+      "  var c = new C();\n"
+      "  var f = c#y;\n"
+      "  var r1 = invoke(f);\n"
+      "  reloadTest();\n"
+      "  var r2 = invoke(f);\n"
+      "  return '$r1 $r2';\n"
+      "}\n";
+
+  Dart_Handle lib = TestCase::LoadTestScript(kScript, NULL);
+  EXPECT_VALID(lib);
+
+  const char* kReloadScript =
+      "import 'test:isolate_reload_helper';\n"
+      "class C {\n"
+      "  var x = 3;\n"
+      "}\n"
+      "invoke(f) {\n"
+      "  try {\n"
+      "    return f();\n"
+      "  } catch (e) {\n"
+      "    return e.toString().split('\\n').first;\n"
+      "  }\n"
+      "}\n"
+      "main() {\n"
+      "  var c = new C();\n"
+      "  var f = c#y;\n"
+      "  var r1 = invoke(f);\n"
+      "  reloadTest();\n"
+      "  var r2 = invoke(f);\n"
+      "  return '$r1 $r2';\n"
+      "}\n";
+
+  TestCase::SetReloadTestScript(kReloadScript);
+
+  EXPECT_STREQ("4 Class 'C' has no instance getter 'y'.",
+               SimpleInvokeStr(lib, "main"));
+
+  lib = TestCase::GetReloadErrorOrRootLibrary();
+  EXPECT_VALID(lib);
+}
+
+
+TEST_CASE(IsolateReload_DanglingGetter_Class) {
+  const char* kScript =
+      "import 'test:isolate_reload_helper';\n"
+      "class C {\n"
+      "  static var x;\n"
+      "  static var y;\n"
+      "}\n"
+      "invoke(f) {\n"
+      "  try {\n"
+      "    return f();\n"
+      "  } catch (e) {\n"
+      "    return e.toString().split('\\n').first;\n"
+      "  }\n"
+      "}\n"
+      "main() {\n"
+      "  C.x = 3;\n"
+      "  C.y = 4;\n"
+      "  var f = C#y;\n"
+      "  var r1 = invoke(f);\n"
+      "  reloadTest();\n"
+      "  var r2 = invoke(f);\n"
+      "  return '$r1 $r2';\n"
+      "}\n";
+
+  Dart_Handle lib = TestCase::LoadTestScript(kScript, NULL);
+  EXPECT_VALID(lib);
+
+  const char* kReloadScript =
+      "import 'test:isolate_reload_helper';\n"
+      "class C {\n"
+      "  static var x;\n"
+      "}\n"
+      "invoke(f) {\n"
+      "  try {\n"
+      "    return f();\n"
+      "  } catch (e) {\n"
+      "    return e.toString().split('\\n').first;\n"
+      "  }\n"
+      "}\n"
+      "main() {\n"
+      "  C.x = 3;\n"
+      "  C.y = 4;\n"
+      "  var f = C#y;\n"
+      "  var r1 = invoke(f);\n"
+      "  reloadTest();\n"
+      "  var r2 = invoke(f);\n"
+      "  return '$r1 $r2';\n"
+      "}\n";
+
+  TestCase::SetReloadTestScript(kReloadScript);
+
+  EXPECT_STREQ("4 No static getter 'y' declared in class 'C'.",
+               SimpleInvokeStr(lib, "main"));
+
+  lib = TestCase::GetReloadErrorOrRootLibrary();
+  EXPECT_VALID(lib);
+}
+
+
+static void IsolateReload_DanlingGetter_LibraryReload(
+    Dart_NativeArguments native_args) {
+  const char* kImportScript2 =
+    "var x;\n";
+  TestCase::AddTestLib("test:other", kImportScript2);
+
+  DART_CHECK_VALID(TestCase::TriggerReload());
+}
+
+
+static Dart_NativeFunction IsolateReload_DanlingGetter_LibraryNativeResolver(
+    Dart_Handle name,
+    int num_of_arguments,
+    bool* auto_setup_scope) {
+  return IsolateReload_DanlingGetter_LibraryReload;
+}
+
+
+TEST_CASE(IsolateReload_DanglingGetter_Library) {
+  const char* kImportScript =
+      "var x;\n"
+      "var y;\n";
+  TestCase::AddTestLib("test:other", kImportScript);
+
+  const char* kScript =
+      "import 'test:other' as prefix;\n"
+      "reloadTest() native 'ReloadTest';\n"
+      "invoke(f) {\n"
+      "  try {\n"
+      "    return f();\n"
+      "  } catch (e) {\n"
+      "    return e.toString().split('\\n').first;\n"
+      "  }\n"
+      "}\n"
+      "main() {\n"
+      "  prefix.x = 3;\n"
+      "  prefix.y = 4;\n"
+      "  var f = prefix#y;\n"
+      "  var r1 = invoke(f);\n"
+      "  reloadTest();\n"
+      "  var r2 = invoke(f);\n"
+      "  return '$r1 $r2';\n"
+      "}\n";
+
+  Dart_Handle lib = TestCase::LoadTestScript(
+      kScript, IsolateReload_DanlingGetter_LibraryNativeResolver);
+  EXPECT_VALID(lib);
+
+  TestCase::SetReloadTestScript(kScript);  // Root library does not change.
+
+  EXPECT_STREQ("4 No top-level getter 'y' declared.",
+               SimpleInvokeStr(lib, "main"));
+
+  lib = TestCase::GetReloadErrorOrRootLibrary();
+  EXPECT_VALID(lib);
+}
+
+
+TEST_CASE(IsolateReload_DanglingSetter_Instance) {
+  const char* kScript =
+      "import 'test:isolate_reload_helper';\n"
+      "class C {\n"
+      "  var x = 3;\n"
+      "  var y = 4;\n"
+      "}\n"
+      "invoke(f, a) {\n"
+      "  try {\n"
+      "    return f(a);\n"
+      "  } catch (e) {\n"
+      "    return e.toString().split('\\n').first;\n"
+      "  }\n"
+      "}\n"
+      "main() {\n"
+      "  var c = new C();\n"
+      "  var f = c#y=;\n"
+      "  var r1 = invoke(f, 5);\n"
+      "  reloadTest();\n"
+      "  var r2 = invoke(f, 6);\n"
+      "  return '$r1 $r2';\n"
+      "}\n";
+
+  Dart_Handle lib = TestCase::LoadTestScript(kScript, NULL);
+  EXPECT_VALID(lib);
+
+  const char* kReloadScript =
+      "import 'test:isolate_reload_helper';\n"
+      "class C {\n"
+      "  var x = 3;\n"
+      "}\n"
+      "invoke(f, a) {\n"
+      "  try {\n"
+      "    return f(a);\n"
+      "  } catch (e) {\n"
+      "    return e.toString().split('\\n').first;\n"
+      "  }\n"
+      "}\n"
+      "main() {\n"
+      "  var c = new C();\n"
+      "  var f = c#y=;\n"
+      "  var r1 = invoke(f, 5);\n"
+      "  reloadTest();\n"
+      "  var r2 = invoke(f, 6);\n"
+      "  return '$r1 $r2';\n"
+      "}\n";
+
+  TestCase::SetReloadTestScript(kReloadScript);
+
+  EXPECT_STREQ("null Class 'C' has no instance setter 'y='.",
+               SimpleInvokeStr(lib, "main"));
+
+  lib = TestCase::GetReloadErrorOrRootLibrary();
+  EXPECT_VALID(lib);
+}
+
+
+TEST_CASE(IsolateReload_DanglingSetter_Class) {
+  const char* kScript =
+      "import 'test:isolate_reload_helper';\n"
+      "class C {\n"
+      "  static var x;\n"
+      "  static var y;\n"
+      "}\n"
+      "invoke(f, a) {\n"
+      "  try {\n"
+      "    return f(a);\n"
+      "  } catch (e) {\n"
+      "    return e.toString().split('\\n').first;\n"
+      "  }\n"
+      "}\n"
+      "main() {\n"
+      "  C.x = 3;\n"
+      "  C.y = 4;\n"
+      "  var f = C#y=;\n"
+      "  var r1 = invoke(f, 5);\n"
+      "  reloadTest();\n"
+      "  var r2 = invoke(f, 6);\n"
+      "  return '$r1 $r2';\n"
+      "}\n";
+
+  Dart_Handle lib = TestCase::LoadTestScript(kScript, NULL);
+  EXPECT_VALID(lib);
+
+  const char* kReloadScript =
+      "import 'test:isolate_reload_helper';\n"
+      "class C {\n"
+      "  static var x;\n"
+      "}\n"
+      "invoke(f, a) {\n"
+      "  try {\n"
+      "    return f(a);\n"
+      "  } catch (e) {\n"
+      "    return e.toString().split('\\n').first;\n"
+      "  }\n"
+      "}\n"
+      "main() {\n"
+      "  C.x = 3;\n"
+      "  C.y = 4;\n"
+      "  var f = C#y=;\n"
+      "  var r1 = invoke(f, 5);\n"
+      "  reloadTest();\n"
+      "  var r2 = invoke(f, 6);\n"
+      "  return '$r1 $r2';\n"
+      "}\n";
+
+  TestCase::SetReloadTestScript(kReloadScript);
+
+  EXPECT_STREQ("5 No static setter 'y=' declared in class 'C'.",
+               SimpleInvokeStr(lib, "main"));
+
+  lib = TestCase::GetReloadErrorOrRootLibrary();
+  EXPECT_VALID(lib);
+}
+
+
+static void IsolateReload_DanlingSetter_LibraryReload(
+    Dart_NativeArguments native_args) {
+  const char* kImportScript2 =
+    "var x;\n";
+  TestCase::AddTestLib("test:other", kImportScript2);
+
+  DART_CHECK_VALID(TestCase::TriggerReload());
+}
+
+
+static Dart_NativeFunction IsolateReload_DanlingSetter_LibraryNativeResolver(
+    Dart_Handle name,
+    int num_of_arguments,
+    bool* auto_setup_scope) {
+  return IsolateReload_DanlingSetter_LibraryReload;
+}
+
+
+TEST_CASE(IsolateReload_DanglingSetter_Library) {
+  const char* kImportScript =
+      "var x;\n"
+      "var y;\n";
+  TestCase::AddTestLib("test:other", kImportScript);
+
+  const char* kScript =
+      "import 'test:other' as prefix;\n"
+      "reloadTest() native 'ReloadTest';\n"
+      "invoke(f, a) {\n"
+      "  try {\n"
+      "    return f(a);\n"
+      "  } catch (e) {\n"
+      "    return e.toString().split('\\n').first;\n"
+      "  }\n"
+      "}\n"
+      "main() {\n"
+      "  prefix.x = 3;\n"
+      "  prefix.y = 4;\n"
+      "  var f = prefix#y=;\n"
+      "  var r1 = invoke(f, 5);\n"
+      "  reloadTest();\n"
+      "  var r2 = invoke(f, 6);\n"
+      "  return '$r1 $r2';\n"
+      "}\n";
+
+  Dart_Handle lib = TestCase::LoadTestScript(
+      kScript, IsolateReload_DanlingSetter_LibraryNativeResolver);
+  EXPECT_VALID(lib);
+
+  TestCase::SetReloadTestScript(kScript);  // Root library does not change.
+
+  EXPECT_STREQ("5 No top-level setter 'y=' declared.",
+               SimpleInvokeStr(lib, "main"));
+
+  lib = TestCase::GetReloadErrorOrRootLibrary();
+  EXPECT_VALID(lib);
+}
+
+
 TEST_CASE(IsolateReload_EnumEquality) {
   const char* kScript =
       "enum Fruit {\n"

runtime/vm/message_handler.cc

diff --git a/runtime/vm/message_handler.cc b/runtime/vm/message_handler.cc
index 5e5f685..80cfbbf 100644
--- a/runtime/vm/message_handler.cc
+++ b/runtime/vm/message_handler.cc
@@ -64,6 +64,7 @@
       should_pause_on_exit_(false),
       is_paused_on_start_(false),
       is_paused_on_exit_(false),
+      delete_me_(false),
       paused_timestamp_(-1),
       pool_(NULL),
       task_(NULL),
@@ -78,6 +79,10 @@
 MessageHandler::~MessageHandler() {
   delete queue_;
   delete oob_queue_;
+  queue_ = NULL;
+  oob_queue_ = NULL;
+  pool_ = NULL;
+  task_ = NULL;
 }
 
 
@@ -110,6 +115,7 @@
               name());
   }
   ASSERT(pool_ == NULL);
+  ASSERT(!delete_me_);
   pool_ = pool;
   start_callback_ = start_callback;
   end_callback_ = end_callback;
@@ -148,6 +154,7 @@
     message = NULL;  // Do not access message.  May have been deleted.
 
     if ((pool_ != NULL) && (task_ == NULL)) {
+      ASSERT(!delete_me_);
       task_ = new MessageHandlerTask(this);
       task_running = pool_->Run(task_);
     }
@@ -251,6 +258,7 @@
   // assigned to a thread pool.
   MonitorLocker ml(&monitor_);
   ASSERT(pool_ == NULL);
+  ASSERT(!delete_me_);
 #if defined(DEBUG)
   CheckAccess();
 #endif
@@ -263,6 +271,7 @@
   // assigned to a thread pool.
   MonitorLocker ml(&monitor_);
   ASSERT(pool_ == NULL);
+  ASSERT(!delete_me_);
 #if defined(DEBUG)
   CheckAccess();
 #endif
@@ -275,6 +284,7 @@
     return kOK;
   }
   MonitorLocker ml(&monitor_);
+  ASSERT(!delete_me_);
 #if defined(DEBUG)
   CheckAccess();
 #endif
@@ -316,6 +326,7 @@
   ASSERT(Isolate::Current() == NULL);
   MessageStatus status = kOK;
   bool run_end_callback = false;
+  bool delete_me = false;
   {
     // We will occasionally release and reacquire this monitor in this
     // function. Whenever we reacquire the monitor we *must* process
@@ -395,6 +406,7 @@
       }
       pool_ = NULL;
       run_end_callback = true;
+      delete_me = delete_me_;
     }
 
     // Clear the task_ last.  This allows other tasks to potentially start
@@ -402,10 +414,17 @@
     ASSERT(oob_queue_->IsEmpty());
     task_ = NULL;
   }
+
+  // Message handlers either use delete_me or end_callback but not both.
+  ASSERT(!delete_me || end_callback_ == NULL);
+
   if (run_end_callback && end_callback_ != NULL) {
     end_callback_(callback_data_);
     // The handler may have been deleted after this point.
   }
+  if (delete_me) {
+    delete this;
+  }
 }
 
 
@@ -433,6 +452,22 @@
 }
 
 
+void MessageHandler::RequestDeletion() {
+  ASSERT(OwnedByPortMap());
+  {
+    MonitorLocker ml(&monitor_);
+    if (task_ != NULL) {
+      // This message handler currently has a task running on the thread pool.
+      delete_me_ = true;
+      return;
+    }
+  }
+
+  // This message handler has no current task.  Delete it.
+  delete this;
+}
+
+
 void MessageHandler::increment_live_ports() {
   MonitorLocker ml(&monitor_);
 #if defined(DEBUG)

runtime/vm/message_handler.h

diff --git a/runtime/vm/message_handler.h b/runtime/vm/message_handler.h
index c52431d..825e3d3 100644
--- a/runtime/vm/message_handler.h
+++ b/runtime/vm/message_handler.h
@@ -187,6 +187,10 @@
   // This is used to delete handlers when their last live port is closed.
   virtual bool OwnedByPortMap() const { return false; }
 
+  // Requests deletion of this message handler when the next task
+  // completes.
+  void RequestDeletion();
+
   void increment_live_ports();
   void decrement_live_ports();
   // ------------ END PortMap API ------------
@@ -241,6 +245,7 @@
   bool should_pause_on_exit_;
   bool is_paused_on_start_;
   bool is_paused_on_exit_;
+  bool delete_me_;
   int64_t paused_timestamp_;
   ThreadPool* pool_;
   ThreadPool::Task* task_;

runtime/vm/native_api_impl.cc

diff --git a/runtime/vm/native_api_impl.cc b/runtime/vm/native_api_impl.cc
index 1bc5d4c..7abd52e 100644
--- a/runtime/vm/native_api_impl.cc
+++ b/runtime/vm/native_api_impl.cc
@@ -90,6 +90,7 @@
 
   NativeMessageHandler* nmh = new NativeMessageHandler(name, handler);
   Dart_Port port_id = PortMap::CreatePort(nmh);
+  PortMap::SetPortState(port_id, PortMap::kLivePort);
   nmh->Run(Dart::thread_pool(), NULL, NULL, 0);
   return port_id;
 }

runtime/vm/native_message_handler.cc

diff --git a/runtime/vm/native_message_handler.cc b/runtime/vm/native_message_handler.cc
index 7779e95..1763341 100644
--- a/runtime/vm/native_message_handler.cc
+++ b/runtime/vm/native_message_handler.cc
@@ -15,8 +15,6 @@
                                            Dart_NativeMessageHandler func)
     : name_(strdup(name)),
       func_(func) {
-  // A NativeMessageHandler always has one live port.
-  increment_live_ports();
 }
 
 

runtime/vm/object.cc

diff --git a/runtime/vm/object.cc b/runtime/vm/object.cc
index 0506fa7..0886a30 100644
--- a/runtime/vm/object.cc
+++ b/runtime/vm/object.cc
@@ -12973,10 +12973,6 @@
 }
 
 
-void ICData::ValidateSentinelLocations() const {
-}
-
-
 void ICData::AddReceiverCheck(intptr_t receiver_class_id,
                               const Function& target,
                               intptr_t count) const {
@@ -13416,12 +13412,10 @@
 }
 
 
-RawArray* ICData::NewEmptyICDataArray(intptr_t num_args_tested) {
+RawArray* ICData::CachedEmptyICDataArray(intptr_t num_args_tested) {
   ASSERT(num_args_tested >= 0);
-  if (num_args_tested < kCachedICDataArrayCount) {
-    return cached_icdata_arrays_[num_args_tested];
-  }
-  return NewNonCachedEmptyICDataArray(num_args_tested);
+  ASSERT(num_args_tested < kCachedICDataArrayCount);
+  return cached_icdata_arrays_[num_args_tested];
 }
 
 
@@ -13468,16 +13462,6 @@
 }
 
 
-void ICData::ResetData() const {
-  // Number of array elements in one test entry.
-  intptr_t len = TestEntryLength();
-  // IC data array must be null terminated (sentinel entry).
-  const Array& ic_data = Array::Handle(Array::New(len, Heap::kOld));
-  set_ic_data_array(ic_data);
-  WriteSentinel(ic_data, len);
-}
-
-
 RawICData* ICData::New() {
   ICData& result = ICData::Handle();
   {
@@ -13513,7 +13497,7 @@
                                                       num_args_tested,
                                                       is_static_call));
   result.set_ic_data_array(
-      Array::Handle(zone, NewEmptyICDataArray(num_args_tested)));
+      Array::Handle(zone, CachedEmptyICDataArray(num_args_tested)));
   return result.raw();
 }
 

runtime/vm/object.h

diff --git a/runtime/vm/object.h b/runtime/vm/object.h
index e9c9280..6be932d 100644
--- a/runtime/vm/object.h
+++ b/runtime/vm/object.h
@@ -1853,8 +1853,7 @@
 
   bool IsImmutable() const;
 
-  void Reset() const;
-  void ResetData() const;
+  void Reset(Zone* zone) const;
 
   // Note: only deopts with reasons before Unknown in this list are recorded in
   // the ICData. All other reasons are used purely for informational messages
@@ -1960,7 +1959,6 @@
   RawArray* FindFreeIndex(intptr_t* index) const;
 
   void DebugDump() const;
-  void ValidateSentinelLocations() const;
 
   // Returns true if this is a two arg smi operation.
   bool AddSmiSmiCheckForFastSmiStubs() const;
@@ -2133,7 +2131,7 @@
 
   intptr_t TestEntryLength() const;
   static RawArray* NewNonCachedEmptyICDataArray(intptr_t num_args_tested);
-  static RawArray* NewEmptyICDataArray(intptr_t num_args_tested);
+  static RawArray* CachedEmptyICDataArray(intptr_t num_args_tested);
   static RawICData* NewDescriptor(Zone* zone,
                                   const Function& owner,
                                   const String& target_name,
@@ -4504,7 +4502,7 @@
 
   // Used during reloading (see object_reload.cc). Calls Reset on all ICDatas
   // that are embedded inside the Code object.
-  void ResetICDatas() const;
+  void ResetICDatas(Zone* zone) const;
 
   TokenPosition GetTokenPositionAt(intptr_t offset) const;
 

runtime/vm/object_reload.cc

diff --git a/runtime/vm/object_reload.cc b/runtime/vm/object_reload.cc
index ec79c6b..31fdfed 100644
--- a/runtime/vm/object_reload.cc
+++ b/runtime/vm/object_reload.cc
@@ -57,7 +57,7 @@
 }
 
 
-void Code::ResetICDatas() const {
+void Code::ResetICDatas(Zone* zone) const {
   // Iterate over the Code's object pool and reset all ICDatas.
 #ifdef TARGET_ARCH_IA32
   // IA32 does not have an object pool, but, we can iterate over all
@@ -65,10 +65,10 @@
   if (!is_alive()) {
     return;
   }
-  const Instructions& instrs = Instructions::Handle(instructions());
+  const Instructions& instrs = Instructions::Handle(zone, instructions());
   ASSERT(!instrs.IsNull());
   uword base_address = instrs.EntryPoint();
-  Object& object = Object::Handle();
+  Object& object = Object::Handle(zone);
   intptr_t offsets_length = pointer_offsets_length();
   const int32_t* offsets = raw_ptr()->data();
   for (intptr_t i = 0; i < offsets_length; i++) {
@@ -81,12 +81,12 @@
     }
     object = raw_object;
     if (object.IsICData()) {
-      ICData::Cast(object).Reset();
+      ICData::Cast(object).Reset(zone);
     }
   }
 #else
-  const ObjectPool& pool = ObjectPool::Handle(object_pool());
-  Object& object = Object::Handle();
+  const ObjectPool& pool = ObjectPool::Handle(zone, object_pool());
+  Object& object = Object::Handle(zone);
   ASSERT(!pool.IsNull());
   for (intptr_t i = 0; i < pool.Length(); i++) {
     ObjectPool::EntryType entry_type = pool.InfoAt(i);
@@ -95,7 +95,7 @@
     }
     object = pool.ObjectAt(i);
     if (object.IsICData()) {
-      ICData::Cast(object).Reset();
+      ICData::Cast(object).Reset(zone);
     }
   }
 #endif
@@ -679,24 +679,24 @@
 static const Function* static_call_target = NULL;
 
 
-void ICData::Reset() const {
+void ICData::Reset(Zone* zone) const {
   if (is_static_call()) {
-    const Function& old_target = Function::Handle(GetTargetAt(0));
+    const Function& old_target = Function::Handle(zone, GetTargetAt(0));
     if (old_target.IsNull()) {
       FATAL("old_target is NULL.\n");
     }
     static_call_target = &old_target;
 
-    const String& selector = String::Handle(old_target.name());
-    Function& new_target = Function::Handle();
+    const String& selector = String::Handle(zone, old_target.name());
+    Function& new_target = Function::Handle(zone);
     if (!old_target.is_static()) {
       if (old_target.kind() == RawFunction::kConstructor) {
         return;  // Super constructor call.
       }
-      Function& caller = Function::Handle();
+      Function& caller = Function::Handle(zone);
       caller ^= Owner();
       ASSERT(!caller.is_static());
-      Class& cls = Class::Handle(caller.Owner());
+      Class& cls = Class::Handle(zone, caller.Owner());
       if (cls.raw() == old_target.Owner()) {
         // Dispatcher.
         if (caller.IsImplicitClosureFunction()) {
@@ -707,7 +707,7 @@
           return;
         }
         const Function& caller_parent =
-            Function::Handle(caller.parent_function());
+            Function::Handle(zone, caller.parent_function());
         if (!caller_parent.IsNull()) {
           if (caller_parent.kind() == RawFunction::kInvokeFieldDispatcher) {
             return;  // Call-through-getter.
@@ -730,23 +730,30 @@
       }
     } else {
       // This can be incorrect if the call site was an unqualified invocation.
-      const Class& cls = Class::Handle(old_target.Owner());
+      const Class& cls = Class::Handle(zone, old_target.Owner());
       new_target = cls.LookupStaticFunction(selector);
     }
 
-    const Array& args_desc_array = Array::Handle(arguments_descriptor());
+    const Array& args_desc_array = Array::Handle(zone, arguments_descriptor());
     ArgumentsDescriptor args_desc(args_desc_array);
     if (new_target.IsNull() ||
         !new_target.AreValidArguments(args_desc, NULL)) {
       // TODO(rmacnak): Patch to a NSME stub.
       VTIR_Print("Cannot rebind static call to %s from %s\n",
                  old_target.ToCString(),
-                 Object::Handle(Owner()).ToCString());
+                 Object::Handle(zone, Owner()).ToCString());
       return;
     }
     ClearAndSetStaticTarget(new_target);
   } else {
-    ClearWithSentinel();
+    intptr_t num_args = NumArgsTested();
+    if (num_args == 2) {
+      ClearWithSentinel();
+    } else {
+      const Array& data_array =
+          Array::Handle(zone, CachedEmptyICDataArray(num_args));
+      set_ic_data_array(data_array);
+    }
   }
 }
 

runtime/vm/object_service.cc

diff --git a/runtime/vm/object_service.cc b/runtime/vm/object_service.cc
index 7eab800..84708d7 100644
--- a/runtime/vm/object_service.cc
+++ b/runtime/vm/object_service.cc
@@ -1135,15 +1135,15 @@
   JSONObject jsobj(stream);
   PrintSharedInstanceJSON(&jsobj, ref);
   jsobj.AddProperty("kind", "Type");
-  if (IsCanonical()) {
+  if (HasResolvedTypeClass()) {
     const Class& type_cls = Class::Handle(type_class());
     if (type_cls.CanonicalType() == raw()) {
       intptr_t cid = type_cls.id();
       jsobj.AddFixedServiceId("classes/%" Pd "/types/%d", cid, 0);
-      jsobj.AddProperty("typeClass", type_cls);
     } else {
       jsobj.AddServiceId(*this);
     }
+    jsobj.AddProperty("typeClass", type_cls);
   } else {
     jsobj.AddServiceId(*this);
   }

runtime/vm/parser.cc

diff --git a/runtime/vm/parser.cc b/runtime/vm/parser.cc
index fc3e0f9..8626a5f 100644
--- a/runtime/vm/parser.cc
+++ b/runtime/vm/parser.cc
@@ -48,6 +48,8 @@
 // committed to the current version.
 DEFINE_FLAG(bool, conditional_directives, true,
     "Enable conditional directives");
+DEFINE_FLAG(bool, initializing_formal_access, false,
+    "Make initializing formal parameters visible in initializer list.");
 DEFINE_FLAG(bool, warn_super, false,
     "Warning if super initializer not last in initializer list.");
 DEFINE_FLAG(bool, await_is_keyword, false,
@@ -697,12 +699,23 @@
     for (int i = 0; i < num_params; i++) {
       ParamDesc& param = (*parameters)[i];
       ASSERT(param.var != NULL);
-      if (!param.is_field_initializer) {
+      if (FLAG_initializing_formal_access || !param.is_field_initializer) {
         param.var->set_invisible(invisible);
       }
     }
   }
 
+  void HideInitFormals() {
+    const intptr_t num_params = parameters->length();
+    for (int i = 0; i < num_params; i++) {
+      ParamDesc& param = (*parameters)[i];
+      if (param.is_field_initializer) {
+        ASSERT(param.var != NULL);
+        param.var->set_invisible(true);
+      }
+    }
+  }
+
   void SetImplicitlyFinal() {
     implicitly_final = true;
   }
@@ -1548,32 +1561,14 @@
       &Object::dynamic_type());
 
   const Function& parent = Function::Handle(func.parent_function());
-  const String& target_name = String::Handle(parent.name());
-  const Class& owner = Class::Handle(parent.Owner());
-  Function& target = Function::ZoneHandle(owner.LookupFunction(target_name));
-  if (target.raw() != parent.raw()) {
-    ASSERT(Isolate::Current()->HasAttemptedReload());
-    if (target.IsNull() ||
-        (target.is_static() != parent.is_static()) ||
-        (target.kind() != parent.kind())) {
-      // TODO(26977): call noSuchMethod/throw NSME instead.
-      target = parent.raw();
-    }
-  }
-
-  if (target.IsImplicitSetterFunction()) {
+  if (parent.IsImplicitSetterFunction()) {
     const TokenPosition ident_pos = func.token_pos();
     ASSERT(IsIdentifier());
-    const String& field_name = *CurrentLiteral();
-    const Class& field_class = Class::ZoneHandle(Z, target.Owner());
-    const Field& field =
-        Field::ZoneHandle(Z, field_class.LookupInstanceField(field_name));
-    const AbstractType& field_type = AbstractType::ZoneHandle(Z, field.type());
     params.AddFinalParameter(ident_pos,
                              &Symbols::Value(),
-                             &field_type);
+                             &Object::dynamic_type());
     ASSERT(func.num_fixed_parameters() == 2);  // closure, value.
-  } else if (!target.IsGetterFunction() && !target.IsImplicitGetterFunction()) {
+  } else if (!parent.IsGetterFunction() && !parent.IsImplicitGetterFunction()) {
     const bool allow_explicit_default_values = true;
     SkipFunctionPreamble();
     ParseFormalParameterList(allow_explicit_default_values, false, &params);
@@ -1604,7 +1599,71 @@
     }
     func_args->set_names(arg_names);
   }
-  StaticCallNode* call = new StaticCallNode(token_pos, target, func_args);
+
+  const String& func_name = String::ZoneHandle(parent.name());
+  const Class& owner = Class::Handle(parent.Owner());
+  Function& target = Function::ZoneHandle(owner.LookupFunction(func_name));
+  if (target.raw() != parent.raw()) {
+    ASSERT(Isolate::Current()->HasAttemptedReload());
+    if (target.IsNull() ||
+        (target.is_static() != parent.is_static()) ||
+        (target.kind() != parent.kind())) {
+      target = Function::null();
+    }
+  }
+
+  AstNode* call = NULL;
+  if (!target.IsNull()) {
+    call = new StaticCallNode(token_pos, target, func_args);
+  } else if (!parent.is_static()) {
+    ASSERT(Isolate::Current()->HasAttemptedReload());
+    // If a subsequent reload reintroduces the target in the middle of the
+    // Invocation object being constructed, we won't be able to successfully
+    // deopt because the generated AST will change.
+    current_function().SetIsOptimizable(false);
+
+    ArgumentListNode* arguments = BuildNoSuchMethodArguments(
+        token_pos, func_name, *func_args, NULL, false);
+    const intptr_t kNumArguments = 2;  // Receiver, InvocationMirror.
+    ArgumentsDescriptor args_desc(
+        Array::Handle(Z, ArgumentsDescriptor::New(kNumArguments)));
+    Function& no_such_method = Function::ZoneHandle(Z,
+    Resolver::ResolveDynamicForReceiverClass(owner,
+                                             Symbols::NoSuchMethod(),
+                                             args_desc));
+    if (no_such_method.IsNull()) {
+      // If noSuchMethod(i) is not found, call Object:noSuchMethod.
+      no_such_method ^= Resolver::ResolveDynamicForReceiverClass(
+          Class::Handle(Z, I->object_store()->object_class()),
+          Symbols::NoSuchMethod(),
+          args_desc);
+    }
+    call = new StaticCallNode(token_pos, no_such_method, arguments);
+  } else {
+    ASSERT(Isolate::Current()->HasAttemptedReload());
+    // If a subsequent reload reintroduces the target in the middle of the
+    // arguments array being constructed, we won't be able to successfully
+    // deopt because the generated AST will change.
+    current_function().SetIsOptimizable(false);
+
+    InvocationMirror::Type im_type;
+    if (parent.IsImplicitGetterFunction()) {
+      im_type = InvocationMirror::kGetter;
+    } else if (parent.IsImplicitSetterFunction()) {
+      im_type = InvocationMirror::kSetter;
+    } else {
+      im_type = InvocationMirror::kMethod;
+    }
+    call = ThrowNoSuchMethodError(TokenPos(),
+                                  owner,
+                                  func_name,
+                                  func_args,
+                                  InvocationMirror::kStatic,
+                                  im_type,
+                                  NULL);  // No existing function.
+  }
+
+  ASSERT(call != NULL);
   ReturnNode* return_node = new ReturnNode(token_pos, call);
   current_block_->statements->Add(return_node);
   return CloseBlock();
@@ -1902,6 +1961,7 @@
   TRACE_PARSER("ParseFormalParameter");
   ParamDesc parameter;
   bool var_seen = false;
+  bool final_seen = false;
   bool this_seen = false;
 
   if (evaluate_metadata && (CurrentToken() == Token::kAT)) {
@@ -1912,6 +1972,7 @@
 
   if (CurrentToken() == Token::kFINAL) {
     ConsumeToken();
+    final_seen = true;
     parameter.is_final = true;
   } else if (CurrentToken() == Token::kVAR) {
     ConsumeToken();
@@ -1926,6 +1987,9 @@
     ExpectToken(Token::kPERIOD);
     this_seen = true;
     parameter.is_field_initializer = true;
+    if (FLAG_initializing_formal_access) {
+      parameter.is_final = true;
+    }
   }
   if ((parameter.type == NULL) && (CurrentToken() == Token::kVOID)) {
     ConsumeToken();
@@ -1967,6 +2031,9 @@
     ExpectToken(Token::kPERIOD);
     this_seen = true;
     parameter.is_field_initializer = true;
+    if (FLAG_initializing_formal_access) {
+      parameter.is_final = true;
+    }
   }
 
   // At this point, we must see an identifier for the parameter name.
@@ -1996,7 +2063,11 @@
     // This parameter is probably a closure. If we saw the keyword 'var'
     // or 'final', a closure is not legal here and we ignore the
     // opening parens.
-    if (!var_seen && !parameter.is_final) {
+    // TODO(hausner): The language spec appears to allow var and final
+    // in signature types when used with initializing formals:
+    // fieldFormalParameter:
+    // metadata finalConstVarOrType? this ‘.’ identifier formalParameterList? ;
+    if (!var_seen && !final_seen) {
       // The parsed parameter type is actually the function result type.
       const AbstractType& result_type =
           AbstractType::Handle(Z, parameter.type->raw());
@@ -3202,7 +3273,6 @@
         // Thus, they are set to be invisible when added to the scope.
         LocalVariable* p = param.var;
         ASSERT(p != NULL);
-        ASSERT(p->is_invisible());
         AstNode* value = new LoadLocalNode(param.name_pos, p);
         EnsureExpressionTemp();
         AstNode* initializer =
@@ -3232,6 +3302,9 @@
 
   // Parsing of initializers done. Now we parse the constructor body.
   OpenBlock();  // Block to collect constructor body nodes.
+  if (FLAG_initializing_formal_access) {
+    params.HideInitFormals();
+  }
   if (CurrentToken() == Token::kLBRACE) {
     // We checked in the top-level parse phase that a redirecting
     // constructor does not have a body.
@@ -7422,7 +7495,10 @@
     if (param_desc.is_final) {
       parameter->set_is_final();
     }
-    if (param_desc.is_field_initializer) {
+    if (FLAG_initializing_formal_access) {
+      // Field initializer parameters are implicitly final.
+      ASSERT(!param_desc.is_field_initializer || param_desc.is_final);
+    } else if (param_desc.is_field_initializer) {
       parameter->set_invisible(true);
     }
   }

runtime/vm/port.cc

diff --git a/runtime/vm/port.cc b/runtime/vm/port.cc
index 2541b38..d5513d5 100644
--- a/runtime/vm/port.cc
+++ b/runtime/vm/port.cc
@@ -216,7 +216,8 @@
   }
   handler->ClosePort(port);
   if (!handler->HasLivePorts() && handler->OwnedByPortMap()) {
-    delete handler;
+    // Delete handler as soon as it isn't busy with a task.
+    handler->RequestDeletion();
   }
   return true;
 }

runtime/vm/profiler.cc

diff --git a/runtime/vm/profiler.cc b/runtime/vm/profiler.cc
index 21b8bf8..b8827c58 100644
--- a/runtime/vm/profiler.cc
+++ b/runtime/vm/profiler.cc
@@ -949,7 +949,9 @@
 
 void Profiler::DumpStackTrace(bool native_stack_trace) {
   Thread* thread = Thread::Current();
-  ASSERT(thread != NULL);
+  if (thread == NULL) {
+    return;
+  }
   OSThread* os_thread = thread->os_thread();
   ASSERT(os_thread != NULL);
   Isolate* isolate = thread->isolate();

runtime/vm/service/service.md

diff --git a/runtime/vm/service/service.md b/runtime/vm/service/service.md
index 6e3c180..7aab061 100644
--- a/runtime/vm/service/service.md
+++ b/runtime/vm/service/service.md
@@ -1535,7 +1535,7 @@
   //   Type
   string name [optional];
 
-  // The corresponding Class if this Type is canonical.
+  // The corresponding Class if this Type has a resolved typeClass.
   //
   // Provided for instance kinds:
   //   Type

runtime/vm/thread.cc

diff --git a/runtime/vm/thread.cc b/runtime/vm/thread.cc
index 69e7820..406f0bc 100644
--- a/runtime/vm/thread.cc
+++ b/runtime/vm/thread.cc
@@ -97,7 +97,7 @@
       REUSABLE_HANDLE_LIST(REUSABLE_HANDLE_INITIALIZERS)
       REUSABLE_HANDLE_LIST(REUSABLE_HANDLE_SCOPE_INIT)
       safepoint_state_(0),
-      execution_state_(kThreadInVM),
+      execution_state_(kThreadInNative),
       next_(NULL) {
 NOT_IN_PRODUCT(
   dart_stream_ = Timeline::GetDartStream();

samples/samples.status

diff --git a/samples/samples.status b/samples/samples.status
index 1fa1d28..c01c0d9 100644
--- a/samples/samples.status
+++ b/samples/samples.status
@@ -18,10 +18,6 @@
 [ $compiler == dart2analyzer ]
 build_dart: Skip
 
-[ $compiler == dart2dart ]
-# Skip until we stabilize language tests.
-*: Skip
-
 [ $arch == arm ]
 sample_extension/test/sample_extension_test: Skip # Issue 14705
 

sdk/lib/_internal/js_runtime/lib/mirror_helper.dart

diff --git a/sdk/lib/_internal/js_runtime/lib/mirror_helper.dart b/sdk/lib/_internal/js_runtime/lib/mirror_helper.dart
deleted file mode 100644
index 9c3024f..0000000
--- a/sdk/lib/_internal/js_runtime/lib/mirror_helper.dart
+++ /dev/null
@@ -1,25 +0,0 @@
-// Copyright (c) 2013, the Dart project authors.  Please see the AUTHORS file
-// for details. All rights reserved. Use of this source code is governed by a
-// BSD-style license that can be found in the LICENSE file.
-/**
- * Helps dealing with reflection in the case that the source code has been
- * changed as a result of compiling with dart2dart.
- */
-library _mirror_helper;
-
-import 'dart:mirrors';
-
-/// The compiler will replace this variable with a map containing all the
-/// renames made in dart2dart.
-const Map<String, String> _SYMBOLS = null;
-
-/// This method is a wrapper for MirrorSystem.getName() and will be inlined and
-/// called in the generated output Dart code.
-String helperGetName(Symbol sym) {
-  var name = MirrorSystem.getName(sym);
-  if (_SYMBOLS.containsKey(name)) {
-    return _SYMBOLS[name];
-  } else {
-    return name;
-  }
-}

sdk/lib/_internal/sdk_library_metadata/lib/libraries.dart

diff --git a/sdk/lib/_internal/sdk_library_metadata/lib/libraries.dart b/sdk/lib/_internal/sdk_library_metadata/lib/libraries.dart
index 0ecb55805..b439338 100644
--- a/sdk/lib/_internal/sdk_library_metadata/lib/libraries.dart
+++ b/sdk/lib/_internal/sdk_library_metadata/lib/libraries.dart
@@ -227,14 +227,6 @@
       documented: false,
       platforms: DART2JS_PLATFORM),
 
-  // TODO(ahe): This library is only for dart2dart, perhaps it should use a
-  // different platform.
-  "_mirror_helper": const LibraryInfo(
-      "_internal/js_runtime/lib/mirror_helper.dart",
-      categories: "",
-      documented: false,
-      platforms: DART2JS_PLATFORM),
-
   "_js_embedded_names": const LibraryInfo(
       "_internal/js_runtime/lib/shared/embedded_names.dart",
       categories: "",

sdk/lib/core/iterable.dart

diff --git a/sdk/lib/core/iterable.dart b/sdk/lib/core/iterable.dart
index 3e7c1cb..aa9292e 100644
--- a/sdk/lib/core/iterable.dart
+++ b/sdk/lib/core/iterable.dart
@@ -181,6 +181,17 @@
    *
    * The returned [Iterable] is lazy, and calls [f] for each element
    * of this every time it's iterated.
+   *
+   * Example:
+   *
+   *     var pairs = [[1, 2], [3, 4]];
+   *     var flattened = pairs.expand((pair) => pair).toList();
+   *     print(flattened); // => [1, 2, 3, 4];
+   *
+   *     var input = [1, 2, 3];
+   *     var duplicated = input.expand((i) => [i, i]).toList();
+   *     print(duplicated); // => [1, 1, 2, 2, 3, 3]
+   *
    */
   Iterable/*<T>*/ expand/*<T>*/(Iterable/*<T>*/ f(E element)) =>
       new ExpandIterable<E, dynamic/*=T*/>(this, f);
@@ -208,7 +219,6 @@
     return false;
   }
 
-
   /**
    * Applies the function [f] to each element of this collection in iteration
    * order.
@@ -316,7 +326,6 @@
     return buffer.toString();
   }
 
-
   /**
    * Checks whether any element of this iterable satisfies [test].
    *
@@ -382,7 +391,7 @@
    */
   bool get isNotEmpty => !isEmpty;
 
-   /**
+  /**
    * Returns a lazy iterable of the [count] first elements of this iterable.
    *
    * The returned `Iterable` may contain fewer than `count` elements, if `this`
@@ -480,7 +489,7 @@
     E result;
     do {
       result = it.current;
-    } while(it.moveNext());
+    } while (it.moveNext());
     return result;
   }
 
@@ -506,7 +515,7 @@
    * function is returned.
    * If [orElse] is omitted, it defaults to throwing a [StateError].
    */
-  E firstWhere(bool test(E element), { E orElse() }) {
+  E firstWhere(bool test(E element), {E orElse()}) {
     for (E element in this) {
       if (test(element)) return element;
     }

sdk/lib/dart2dart.platform

diff --git a/sdk/lib/dart2dart.platform b/sdk/lib/dart2dart.platform
deleted file mode 100644
index d9aaa19..0000000
--- a/sdk/lib/dart2dart.platform
+++ /dev/null
@@ -1,49 +0,0 @@
-# Copyright (c) 2015, the Dart project authors.  Please see the AUTHORS file
-# for details. All rights reserved. Use of this source code is governed by a
-# BSD-style license that can be found in the LICENSE file.
-
-# The platform when compiling with dart2js for dart2dart.
-#
-# Includes the _mirror_helpers private libraries
-
-[dart-spec]
-spec:3rd edition.
-
-[features]
-# No extra features.
-
-[libraries]
-async: async/async.dart
-_chrome: _chrome/dart2js/chrome_dart2js.dart
-collection: collection/collection.dart
-convert: convert/convert.dart
-core: core/core.dart
-developer: developer/developer.dart
-html: html/dart2js/html_dart2js.dart
-html_common: html/html_common/html_common_dart2js.dart
-indexed_db: indexed_db/dart2js/indexed_db_dart2js.dart
-io: io/io.dart
-isolate: isolate/isolate.dart
-js: js/dart2js/js_dart2js.dart
-js_util: js_util/dart2js/js_util_dart2js.dart
-math: math/math.dart
-mirrors: mirrors/mirrors.dart
-nativewrappers: html/dart2js/nativewrappers.dart
-typed_data: typed_data/typed_data.dart
-_native_typed_data: _internal/js_runtime/lib/native_typed_data.dart
-svg: svg/dart2js/svg_dart2js.dart
-web_audio: web_audio/dart2js/web_audio_dart2js.dart
-web_gl: web_gl/dart2js/web_gl_dart2js.dart
-web_sql: web_sql/dart2js/web_sql_dart2js.dart
-_internal: internal/internal.dart
-_js_helper: _internal/js_runtime/lib/js_helper.dart
-_interceptors: _internal/js_runtime/lib/interceptors.dart
-_foreign_helper: _internal/js_runtime/lib/foreign_helper.dart
-_isolate_helper: _internal/js_runtime/lib/isolate_helper.dart
-_js_mirrors: _internal/js_runtime/lib/js_mirrors.dart
-_js_names: _internal/js_runtime/lib/js_names.dart
-_js_primitives: _internal/js_runtime/lib/js_primitives.dart
-_mirror_helper: _internal/js_runtime/lib/mirror_helper.dart
-_js_embedded_names: _internal/js_runtime/lib/shared/embedded_names.dart
-_async_await_error_codes: _internal/js_runtime/lib/shared/async_await_error_codes.dart
-_metadata: html/html_common/metadata.dart

tests/co19/co19-dart2js.status

diff --git a/tests/co19/co19-dart2js.status b/tests/co19/co19-dart2js.status
index d1662c9..a0dd69f 100644
--- a/tests/co19/co19-dart2js.status
+++ b/tests/co19/co19-dart2js.status
@@ -1677,6 +1677,7 @@
 [ $compiler == dart2js && $runtime == chrome && $system == macos ]
 Language/Expressions/Function_Invocation/async_invokation_t04: RuntimeError # Please triage this failure
 LayoutTests/fast/canvas/webgl/canvas-test_t01: Skip # Times out. Please triage this failure
+LayoutTests/fast/canvas/webgl/tex-image-and-sub-image-2d-with-video-rgba5551_t01: Skip # Times out flakily.
 LayoutTests/fast/canvas/webgl/context-lost-restored_t01: Skip # Times out. Please triage this failure.
 LayoutTests/fast/canvas/webgl/draw-webgl-to-canvas-2d_t01: Pass, RuntimeError # Please triage this failure
 LayoutTests/fast/canvas/webgl/oes-vertex-array-object_t01: RuntimeError # Please triage this failure
@@ -1842,10 +1843,7 @@
 LayoutTests/fast/canvas/canvas-currentColor_t01: RuntimeError # Please triage this failure
 LayoutTests/fast/canvas/canvas-currentTransform_t01: RuntimeError # Feature is not implemented.
 LayoutTests/fast/canvas/canvas-drawImage-incomplete_t01: RuntimeError # Please triage this failure
-LayoutTests/fast/canvas/canvas-ellipse-360-winding_t01: RuntimeError # Please triage this failure
-LayoutTests/fast/canvas/canvas-ellipse-negative-radius_t01: RuntimeError # Please triage this failure
 LayoutTests/fast/canvas/canvas-ellipse-zero-lineto_t01: RuntimeError # Please triage this failure
-LayoutTests/fast/canvas/canvas-ellipse_t01: RuntimeError # Please triage this failure
 LayoutTests/fast/canvas/canvas-empty-image-pattern_t01: RuntimeError # Please triage this failure
 LayoutTests/fast/canvas/canvas-fill-zeroSizeGradient_t01: RuntimeError # Please triage this failure
 LayoutTests/fast/canvas/canvas-fillRect-zeroSizeGradient_t01: RuntimeError # Please triage this failure
@@ -2057,7 +2055,6 @@
 LayoutTests/fast/css/parsing-css-allowed-string-characters_t01: RuntimeError # Please triage this failure
 LayoutTests/fast/css/parsing-css-escapes_t01: RuntimeError # Please triage this failure
 LayoutTests/fast/css/parsing-css-nonascii_t01: RuntimeError # Please triage this failure
-LayoutTests/fast/css/parsing-css-nth-child_t01: RuntimeError # Please triage this failure
 LayoutTests/fast/css/parsing-page-rule_t01: RuntimeError # Please triage this failure
 LayoutTests/fast/css/parsing-selector-error-recovery_t01: RuntimeError # Please triage this failure
 LayoutTests/fast/css/parsing-unexpected-eof_t01: RuntimeError # Please triage this failure
@@ -2814,7 +2811,6 @@
 LayoutTests/fast/writing-mode/positionForPoint_t01: RuntimeError # Please triage this failure
 LayoutTests/fast/writing-mode/table-hit-test_t01: RuntimeError # Please triage this failure
 LayoutTests/fast/writing-mode/vertical-inline-block-hittest_t01: RuntimeError # Please triage this failure
-LayoutTests/fast/xmlhttprequest/xmlhttprequest-invalid-values_t01: RuntimeError # Please triage this failure
 LayoutTests/fast/xmlhttprequest/xmlhttprequest-responseXML-invalid-xml_t01: RuntimeError # Please triage this failure
 LayoutTests/fast/xmlhttprequest/xmlhttprequest-responseXML-xml-text-responsetype_t01: RuntimeError # Please triage this failure
 LayoutTests/fast/xmlhttprequest/xmlhttprequest-responsetype-abort_t01: RuntimeError # Please triage this failure
@@ -2944,6 +2940,7 @@
 LibTest/html/Window/requestFileSystem_A02_t01: RuntimeError # Please triage this failure
 LibTest/html/Window/resizeBy_A01_t01: RuntimeError # Please triage this failure
 LibTest/html/Window/resizeTo_A01_t01: RuntimeError # Please triage this failure
+LibTest/math/log_A01_t01: Fail # Please triage this failure.
 LibTest/typed_data/Float32List/Float32List.view_A06_t01: RuntimeError # Please triage this failure
 LibTest/typed_data/Float32x4List/Float32x4List.view_A06_t01: RuntimeError # Please triage this failure
 LibTest/typed_data/Float64List/Float64List.view_A06_t01: RuntimeError # Please triage this failure

tests/compiler/dart2js/analyze_unused_dart2js_test.dart

diff --git a/tests/compiler/dart2js/analyze_unused_dart2js_test.dart b/tests/compiler/dart2js/analyze_unused_dart2js_test.dart
index 8106a06..a17914e 100644
--- a/tests/compiler/dart2js/analyze_unused_dart2js_test.dart
+++ b/tests/compiler/dart2js/analyze_unused_dart2js_test.dart
@@ -23,9 +23,6 @@
       "The method 'asAssert' is never called.",
       "The method 'asLiteralBool' is never called."],
 
-  // Some things in dart_printer are not yet used
-  "lib/src/dart_backend/backend_ast_nodes.dart": const [" is never "],
-
   // Uncalled methods in SemanticSendVisitor and subclasses.
   "lib/src/resolution/semantic_visitor.dart": const [
       "The method 'error"],
@@ -56,10 +53,6 @@
     "accept", "CreateFunction",
   ],
 
-  "lib/src/dart_backend/backend_ast_to_frontend_ast.dart": const [
-    " is never "
-  ],
-
   // Useful utility functions that are not currently used.
   "lib/src/cps_ir/cps_fragment.dart": const [
     "The method 'beginLoop' is never called.",

tests/compiler/dart2js/backend_dart/dart_backend_test.dart

diff --git a/tests/compiler/dart2js/backend_dart/dart_backend_test.dart b/tests/compiler/dart2js/backend_dart/dart_backend_test.dart
deleted file mode 100644
index 5db4160..0000000
--- a/tests/compiler/dart2js/backend_dart/dart_backend_test.dart
+++ /dev/null
@@ -1,1119 +0,0 @@
-// Copyright (c) 2012, the Dart project authors.  Please see the AUTHORS file
-// for details. All rights reserved. Use of this source code is governed by a
-// BSD-style license that can be found in the LICENSE file.
-
-import "package:expect/expect.dart";
-import 'dart:async';
-import 'dart:io' as io;
-import "package:async_helper/async_helper.dart";
-import '../mock_compiler.dart';
-import '../mock_libraries.dart';
-import '../output_collector.dart';
-import 'package:compiler/compiler.dart';
-import 'package:compiler/src/common/names.dart' show Identifiers;
-import 'package:compiler/src/dart_backend/dart_backend.dart';
-import 'package:compiler/src/elements/elements.dart';
-import 'package:compiler/src/tree/tree.dart';
-
-const ioLib = r'''
-library io;
-class Platform {
-  static int operatingSystem;
-}
-''';
-
-const htmlLib = r'''
-library html;
-Window __window;
-Window get window => __window;
-abstract class Window {
-  Navigator get navigator;
-}
-abstract class Navigator {
-  String get userAgent;
-}
-''';
-
-/**
- * Library name is assumed to be 'mylib' in 'mylib.dart' file.
- */
-testDart2Dart(String mainSrc, {String librarySrc,
-                               String expectedResult,
-                               bool minify: false,
-                               bool stripTypes: false}) {
-
-  // If expectedResult is not provided, check that source string remains the
-  // same.
-  if (expectedResult == null) {
-    Expect.equals(null, librarySrc);
-    expectedResult = mainSrc;
-  }
-
-  fileUri(path) => new Uri(scheme: 'file', path: path);
-
-  final scriptUri = fileUri('script.dart');
-  final libUri = fileUri('mylib.dart');
-
-  provider(uri) {
-    if (uri == scriptUri) return new Future.value(mainSrc);
-    if (uri.toString() == libUri.toString()) {
-      return new Future.value(librarySrc);
-    }
-    if (uri.path.endsWith('/dart2dart.platform')) {
-      return new io.File.fromUri(uri).readAsBytes();
-    } else if (uri.path.endsWith('/core.dart')) {
-      return new Future.value(buildLibrarySource(DEFAULT_CORE_LIBRARY));
-    } else if (uri.path.endsWith('/core_patch.dart')) {
-      return new Future.value(DEFAULT_PATCH_CORE_SOURCE);
-    } else if (uri.path.endsWith('/io.dart')) {
-      return new Future.value(ioLib);
-    } else if (uri.path.endsWith('/js_helper.dart')) {
-      return new Future.value(buildLibrarySource(DEFAULT_JS_HELPER_LIBRARY));
-    } else if (uri.path.endsWith('/html_dart2js.dart')) {
-      // TODO(smok): The file should change to html_dartium at some point.
-      return new Future.value(htmlLib);
-    } else if (uri.path.endsWith('/foreign_helper.dart')) {
-      return new Future.value(
-          buildLibrarySource(DEFAULT_FOREIGN_HELPER_LIBRARY));
-    } else if (uri.path.endsWith('/isolate_helper.dart')) {
-      return new Future.value(
-          buildLibrarySource(DEFAULT_ISOLATE_HELPER_LIBRARY));
-    }
-    return new Future.value('');
-  }
-
-  handler(uri, begin, end, message, kind) {
-    if (identical(kind, Diagnostic.ERROR) || identical(kind, Diagnostic.CRASH)) {
-      Expect.fail('$uri: $begin-$end: $message [$kind]');
-    }
-  }
-
-  final options = <String>['--output-type=dart'];
-  // Some tests below are using dart:io.
-  if (minify) options.add('--minify');
-  if (stripTypes) options.add('--force-strip=types');
-
-  asyncTest(() {
-    OutputCollector outputCollector = new OutputCollector();
-    return compile(
-        scriptUri,
-        Uri.base.resolve('sdk/'),
-        fileUri('packageRoot/'),
-        provider,
-        handler,
-        options,
-        outputCollector).then((_) {
-      String code = outputCollector.getOutput('', 'dart');
-      Expect.equals(expectedResult, code,
-          'expected:\n$expectedResult\nactual:\n$code');
-    });
-  });
-}
-
-testSimpleFileUnparse() {
-  final src = '''
-should_be_dropped() {
-}
-
-should_be_kept() {
-}
-
-main() {
-  should_be_kept();
-}
-''';
-  testDart2Dart(src, expectedResult: '''
-should_be_kept() {}
-main() {
-  should_be_kept();
-}
-''');
-}
-testTopLevelField() {
-  testDart2Dart('''
-final String x = "asd";
-main() {
-  x;
-}
-''');
-}
-
-testSimpleTopLevelClass() {
-  testDart2Dart('''
-main() {
-  new A();
-}
-class A {
-  A() {}
-}
-''');
-}
-
-testClassWithSynthesizedConstructor() {
-  testDart2Dart('''
-main() {
-  new A();
-}
-class A {}
-''');
-}
-
-testClassWithMethod() {
-  testDart2Dart(r'''
-main() {
-  var a = new A();
-  a.foo();
-}
-class A {
-  void foo() {}
-}
-''');
-}
-
-testExtendsImplements() {
-  testDart2Dart('''
-main() {
-  new B<Object>();
-}
-class A<T> {}
-class B<T> extends A<T> {}
-''');
-}
-
-testVariableDefinitions() {
-  testDart2Dart('''
-main() {
-  var x, y;
-  final String s = null;
-}
-''');
-  testDart2Dart('''
-main() {
-  final int x = 0, y = 0;
-  final String s = null;
-}
-''');
-  testDart2Dart('''
-foo(f, g) {}
-main() {
-  foo(1, 2);
-}
-''');
-  testDart2Dart('''
-foo(f(arg)) {}
-main() {
-  foo(main);
-}
-''');
-  // A couple of static/finals inside a class.
-  testDart2Dart('''
-main() {
-  A.a;
-  A.b;
-}
-class A {
-  static const String a = "5";
-  static const String b = "4";
-}
-''');
-  // Class member of typedef-ed function type.
-  // Maybe typedef should be included in the result too, but it
-  // works fine without it.
-  testDart2Dart('''
-typedef void foofunc(_0);
-main() {
-  new A((arg) {});
-}
-class A {
-  A(foofunc this.handler);
-  final foofunc handler;
-}
-''');
-}
-
-testGetSet() {
-  // Top-level get/set.
-  testDart2Dart('''
-set foo(arg) {}
-get foo {
-  return 5;
-}
-main() {
-  foo;
-  foo = 5;
-}
-''');
-  // Field get/set.
-  testDart2Dart('''
-main() {
-  var a = new A();
-  a.foo;
-  a.foo = 5;
-}
-class A {
-  set foo(a) {}
-  get foo {
-    return 5;
-  }
-}
-''');
-  // Typed get/set.
-  testDart2Dart('''
-String get foo {
-  return "a";
-}
-main() {
-  foo;
-}
-''');
-}
-
-testAbstractClass() {
-  testDart2Dart('''
-main() {
-  A.foo;
-}
-abstract class A {
-  static final num foo = 0;
-}
-''');
-}
-
-testConflictSendsRename() {
-  // Various Send-s to current library and external library. Verify that
-  // everything is renamed correctly in conflicting class names and global
-  // functions.
-  var librarySrc = '''
-library mylib;
-
-globalfoo() {}
-var globalVar;
-var globalVarInitialized = 6, globalVarInitialized2 = 7;
-
-class A {
-  A(){}
-  A.fromFoo(){}
-  static staticfoo(){}
-  foo(){}
-  static const field = 5;
-}
-''';
-  var mainSrc = '''
-import 'mylib.dart' as mylib;
-
-globalfoo() {}
-var globalVar;
-var globalVarInitialized = 6, globalVarInitialized2 = 7;
-
-class A {
-  A(){}
-  A.fromFoo(){}
-  static staticfoo(){}
-  foo(){}
-  static const field = 5;
-}
-
-main() {
-  globalVar;
-  globalVarInitialized;
-  globalVarInitialized2;
-  globalfoo();
-  A.field;
-  A.staticfoo();
-  new A();
-  new A.fromFoo();
-  new A().foo();
-
-  mylib.globalVar;
-  mylib.globalVarInitialized;
-  mylib.globalVarInitialized2;
-  mylib.globalfoo();
-  mylib.A.field;
-  mylib.A.staticfoo();
-  new mylib.A();
-  new mylib.A.fromFoo();
-  new mylib.A().foo();
-}
-''';
-  var expectedResult = '''
-globalfoo() {}
-var globalVar;
-var globalVarInitialized = 6;
-var globalVarInitialized2 = 7;
-class A {
-  A() {}
-  A.fromFoo() {}
-  static staticfoo() {}
-  foo() {}
-  static const field = 5;
-}
-globalfoo_A() {}
-var globalVar_A;
-var globalVarInitialized_A = 6;
-var globalVarInitialized2_A = 7;
-class A_A {
-  A_A() {}
-  A_A.fromFoo_A() {}
-  static staticfoo_A() {}
-  foo() {}
-  static const field_A = 5;
-}
-main() {
-  globalVar_A;
-  globalVarInitialized_A;
-  globalVarInitialized2_A;
-  globalfoo_A();
-  A_A.field_A;
-  A_A.staticfoo_A();
-  new A_A();
-  new A_A.fromFoo_A();
-  new A_A().foo();
-  globalVar;
-  globalVarInitialized;
-  globalVarInitialized2;
-  globalfoo();
-  A.field;
-  A.staticfoo();
-  new A();
-  new A.fromFoo();
-  new A().foo();
-}
-''';
-  testDart2Dart(mainSrc, librarySrc: librarySrc,
-      expectedResult: expectedResult);
-}
-
-testNoConflictSendsRename() {
-  // Various Send-s to current library and external library. Nothing should be
-  // renamed here, only library prefixes must be cut.
-  var librarySrc = '''
-library mylib;
-
-globalfoo() {}
-
-class A {
-  A(){}
-  A.fromFoo(){}
-  static staticfoo(){}
-  foo(){}
-  static const field = 5;
-}
-''';
-  var mainSrc = '''
-import 'mylib.dart' as mylib;
-
-myglobalfoo() {}
-
-class MyA {
-  MyA(){}
-  MyA.myfromFoo(){}
-  static mystaticfoo(){}
-  myfoo(){}
-  static const myfield = 5;
-}
-
-main() {
-  myglobalfoo();
-  MyA.myfield;
-  MyA.mystaticfoo();
-  new MyA();
-  new MyA.myfromFoo();
-  new MyA().myfoo();
-
-  mylib.globalfoo();
-  mylib.A.field;
-  mylib.A.staticfoo();
-  new mylib.A();
-  new mylib.A.fromFoo();
-  new mylib.A().foo();
-}
-''';
-  var expectedResult = '''
-globalfoo() {}
-class A {
-  A() {}
-  A.fromFoo() {}
-  static staticfoo() {}
-  foo() {}
-  static const field = 5;
-}
-myglobalfoo() {}
-class MyA {
-  MyA() {}
-  MyA.myfromFoo() {}
-  static mystaticfoo() {}
-  myfoo() {}
-  static const myfield = 5;
-}
-main() {
-  myglobalfoo();
-  MyA.myfield;
-  MyA.mystaticfoo();
-  new MyA();
-  new MyA.myfromFoo();
-  new MyA().myfoo();
-  globalfoo();
-  A.field;
-  A.staticfoo();
-  new A();
-  new A.fromFoo();
-  new A().foo();
-}
-''';
-  testDart2Dart(mainSrc, librarySrc: librarySrc,
-      expectedResult: expectedResult);
-}
-
-testConflictLibraryClassRename() {
-  var librarySrc = '''
-library mylib;
-
-topfoo() {}
-
-class A {
-  foo() {}
-}
-''';
-  var mainSrc = '''
-import 'mylib.dart' as mylib;
-topfoo() {
-  var x = 5;
-}
-class A {
-  num foo() {}
-  A.fromFoo() {}
-  mylib.A myliba;
-  List<A> mylist;
-}
-mylib.A getA() => null;
-main() {
-  var a = new mylib.A();
-  a.foo();
-  var b = new A.fromFoo();
-  b.foo();
-  var GREATVAR = b.myliba;
-  b.mylist;
-  a = getA();
-  topfoo();
-  mylib.topfoo();
-}
-''';
-  var expectedResult = '''
-topfoo() {}
-class A {
-  foo() {}
-}
-topfoo_A() {
-  var x = 5;
-}
-class A_A {
-  num foo() {}
-  A_A.fromFoo() {}
-  A myliba;
-  List<A_A> mylist;
-}
-A getA() => null;
-main() {
-  var a = new A();
-  a.foo();
-  var b = new A_A.fromFoo();
-  b.foo();
-  var GREATVAR = b.myliba;
-  b.mylist;
-  a = getA();
-  topfoo_A();
-  topfoo();
-}
-''';
-  testDart2Dart(mainSrc, librarySrc: librarySrc,
-      expectedResult: expectedResult);
-}
-
-testClassExtendsWithArgs() {
-  testDart2Dart('''
-main() {
-  new B<Object>();
-}
-class A<T extends Object> {}
-class B<T extends Object> extends A<T> {}
-''', expectedResult: '''
-main() {
-  new B<Object>();
-}
-class A<T> {}
-class B<T> extends A<T> {}
-''');
-}
-
-testStaticInvocation() {
-  testDart2Dart('''
-main() {
-  var x = double.parseDouble("1");
-}
-''');
-}
-
-testLibraryGetSet() {
-  var librarySrc = '''
-library mylib;
-
-get topgetset => 5;
-set topgetset(arg) {}
-''';
-  var mainSrc = '''
-import 'mylib.dart' as mylib;
-
-get topgetset => 6;
-set topgetset(arg) {}
-
-main() {
-  topgetset;
-  topgetset = 6;
-
-  mylib.topgetset;
-  mylib.topgetset = 5;
-}
-''';
-  var expectedResult = '''
-get topgetset => 5;
-set topgetset(arg) {}
-get topgetset_A => 6;
-set topgetset_A(arg) {}
-main() {
-  topgetset_A;
-  topgetset_A = 6;
-  topgetset;
-  topgetset = 5;
-}
-''';
-  testDart2Dart(mainSrc, librarySrc: librarySrc,
-      expectedResult: expectedResult);
-}
-
-testFieldTypeOutput() {
-  testDart2Dart('''
-main() {
-  new A().field;
-}
-class B {}
-class A {
-  B field;
-}
-''');
-}
-
-class DynoMap implements Map<Element, ElementAst> {
-  final compiler;
-  DynoMap(this.compiler);
-
-  ElementAst operator[](AstElement element) {
-    return new ElementAst(element.resolvedAst.node,
-                          element.resolvedAst.elements);
-  }
-
-  noSuchMethod(Invocation invocation) => throw 'unimplemented method';
-}
-
-PlaceholderCollector collectPlaceholders(compiler, element) {
-  DartBackend backend = compiler.backend;
-  return new PlaceholderCollector(
-      compiler.reporter,
-      backend.mirrorRenamer,
-      new Set<String>(),
-      new DynoMap(compiler),
-      compiler.mainFunction)
-    ..collect(element);
-}
-
-testLocalFunctionPlaceholder() {
-  var src = '''
-main() {
-  function localfoo() {}
-  localfoo();
-}
-''';
-  MockCompiler compiler = new MockCompiler.internal(emitJavaScript: false);
-  asyncTest(() => compiler.init().then((_) {
-    assert(compiler.backend is DartBackend);
-    compiler.parseScript(src);
-    FunctionElement mainElement = compiler.mainApp.find(Identifiers.main);
-    compiler.processQueue(compiler.enqueuer.resolution, mainElement);
-    PlaceholderCollector collector = collectPlaceholders(compiler, mainElement);
-    FunctionExpression mainNode = mainElement.node;
-    Block body = mainNode.body;
-    FunctionDeclaration functionDeclaration = body.statements.nodes.head;
-    FunctionExpression fooNode = functionDeclaration.function;
-    LocalPlaceholder fooPlaceholder =
-        collector.functionScopes[mainElement].localPlaceholders.first;
-    Expect.isTrue(fooPlaceholder.nodes.contains(fooNode.name));
-  }));
-}
-
-testTypeVariablesAreRenamed() {
-  // Somewhat a hack: we require all the references of the identifier
-  // to be renamed in the same way for the whole library. Hence
-  // if we have a class and type variable with the same name, they
-  // both should be renamed.
-  var librarySrc = '''
-library mylib;
-typedef void MyFunction<T extends num>(T arg);
-class T {}
-class B<T> {}
-class A<T> extends B<T> { T f; }
-''';
-  var mainSrc = '''
-import 'mylib.dart' as mylib;
-typedef void MyFunction<T extends num>(T arg);
-class T {}
-class B<T> {}
-class A<T> extends B<T> { T f; }
-
-main() {
-  MyFunction myf1;
-  mylib.MyFunction myf2;
-  new A<int>().f;
-  new T();
-
-  new mylib.A<int>().f;
-  new mylib.T();
-}
-''';
-  var expectedResult = '''
-typedef void MyFunction<T_B extends num>(T_B _0);
-class T {}
-class B<T_B> {}
-class A<T_B> extends B<T_B> {
-  T_B f;
-}
-typedef void MyFunction_A<T_B extends num>(T_B _0);
-class T_A {}
-class B_A<T_B> {}
-class A_A<T_B> extends B_A<T_B> {
-  T_B f;
-}
-main() {
-  MyFunction_A myf1;
-  MyFunction myf2;
-  new A_A<int>().f;
-  new T_A();
-  new A<int>().f;
-  new T();
-}
-''';
-  testDart2Dart(mainSrc, librarySrc: librarySrc,
-      expectedResult: expectedResult);
-}
-
-testClassTypeArgumentBound() {
-  var librarySrc = '''
-library mylib;
-
-class I {}
-class A<T extends I> {}
-
-''';
-  var mainSrc = '''
-import 'mylib.dart' as mylib;
-
-class I {}
-class A<T extends I> {}
-
-main() {
-  new A();
-  new mylib.A();
-}
-''';
-  var expectedResult = '''
-class I {}
-class A<T extends I> {}
-class I_A {}
-class A_A<T extends I_A> {}
-main() {
-  new A_A();
-  new A();
-}
-''';
-  testDart2Dart(mainSrc, librarySrc: librarySrc,
-      expectedResult: expectedResult);
-  }
-
-testDoubleMains() {
-  var librarySrc = '''
-library mylib;
-main() {}
-''';
-  var mainSrc = '''
-import 'mylib.dart' as mylib;
-main() {
-  mylib.main();
-}
-''';
-  var expectedResult = '''
-main_A() {}
-main() {
-  main_A();
-}
-''';
-  testDart2Dart(mainSrc, librarySrc: librarySrc,
-      expectedResult: expectedResult);
-}
-
-testStaticAccessIoLib() {
-  var src = '''
-import 'dart:io';
-
-main() {
-  Platform.operatingSystem;
-}
-''';
-  var expectedResult = '''
-import "dart:io";
-main() {
-  Platform.operatingSystem;
-}
-''';
-  testDart2Dart(src, expectedResult: expectedResult);
-}
-
-testMinification() {
-  var src = '''
-class ClassWithVeryVeryLongName {}
-main() {
-  new ClassWithVeryVeryLongName();
-}
-''';
-  var expectedResult =
-      'class A{}'
-      'main(){new A();}';
-  testDart2Dart(src, expectedResult: expectedResult, minify: true);
-}
-
-testClosureLocalsMinified() {
-  var src = '''
-main() {
-  var a = 7;
-  void foo1(a,b) {
-    void foo2(c,d) {
-       var E = a;
-    }
-    foo2(b, a);
-  }
-  foo1(a, 8);
-}
-''';
-  var expectedResult =
-      'main(){var A=7; B(A,C){ D(E,F){var G=A;}D(C,A);}B(A,8);}';
-  testDart2Dart(src, expectedResult: expectedResult, minify: true);
-}
-
-testParametersMinified() {
-  var src = '''
-class A {
-  var a;
-  static foo(arg1) {
-    // Should not rename arg1 to a.
-    arg1 = 5;
-  }
-}
-
-fooglobal(arg,{optionalarg: 7}) {
-  arg = 6;
-}
-
-main() {
-  new A().a;
-  A.foo(8);
-  fooglobal(8);
-}
-''';
-  var expectedResult =
-      'class B{var E;static C(A){A=5;}}D(A,{optionalarg: 7}){A=6;}'
-      'main(){new B().E;B.C(8);D(8);}';
-  testDart2Dart(src, expectedResult: expectedResult, minify: true);
-}
-
-testDeclarationTypePlaceholders() {
-  var src = '''
-String globalfield;
-const String globalconstfield;
-
-void foo(String arg) {}
-
-main() {
-  String localvar;
-  foo("5");
-}
-''';
-  var expectedResult = '''
-foo( arg) {}
-main() {
-  var localvar;
-  foo("5");
-}
-''';
-  testDart2Dart(src, expectedResult: expectedResult, stripTypes: true);
-}
-
-testPlatformLibraryMemberNamesAreFixed() {
-  var src = '''
-import 'dart:html';
-
-class A {
-  static String get userAgent => window.navigator.userAgent;
-}
-
-main() {
-  A.userAgent;
-}
-''';
-  var expectedResult = '''
-import "dart:html";
-class A {
-  static String get userAgent_A => window.navigator.userAgent;
-}
-main() {
-  A.userAgent_A;
-}
-''';
-  testDart2Dart(src, expectedResult: expectedResult);
-}
-
-testConflictsWithCoreLib() {
-  var src = '''
-import 'dart:core' as fisk;
-
-print(x) { throw 'fisk'; }
-
-main() {
-  fisk.print('corelib');
-  print('local');
-}
-''';
-  var expectedResult = """
-print_A(x) {
-  throw 'fisk';
-}
-main() {
-  print('corelib');
-  print_A('local');
-}
-""";
-  testDart2Dart(src, expectedResult: expectedResult);
-}
-
-testUnresolvedNamedConstructor1() {
-  var src = '''
-class A {
-}
-
-main() {
-  new A.named();
-}
-''';
-  var expectedResult = """
-main() {
-  new Unresolved();
-}
-""";
-  testDart2Dart(src, expectedResult: expectedResult);
-}
-
-testUnresolvedNamedConstructor2() {
-  var src = '''
-class A {
-  A() {}
-}
-
-main() {
-  new A();
-  new A.named();
-}
-''';
-  var expectedResult = """
-class A {
-  A() {}
-}
-main() {
-  new A();
-  new Unresolved();
-}
-""";
-  testDart2Dart(src, expectedResult: expectedResult);
-}
-
-testUnresolvedNamedConstructor3() {
-  var src = '''
-class A {
-  static method() {}
-}
-
-main() {
-  A.method();
-  new A.named();
-}
-''';
-  var expectedResult = """
-class A {
-  static method() {}
-}
-main() {
-  A.method();
-  new Unresolved();
-}
-""";
-  testDart2Dart(src, expectedResult: expectedResult);
-}
-
-testClassAndNamedMixinDeclarations() {
-  test(String declarations, {String expectedDeclarations}) {
-    const String mainSource = 'main() => new A();';
-    if (expectedDeclarations == null) {
-      expectedDeclarations = declarations;
-    }
-    testDart2Dart('$declarations\n$mainSource\n',
-                  expectedResult: '$expectedDeclarations\n$mainSource\n');
-  }
-
-  test('class A {}');
-  test('class A<T> {}');
-  test('class A<T extends num> {}');
-  test('class A<T extends Object> {}', expectedDeclarations: 'class A<T> {}');
-  test('class A extends Object {}', expectedDeclarations: 'class A {}');
-
-  test('''
-class S1 {}
-class A extends S1 {}''');
-
-  test('''
-class S1 {}
-class A implements S1 {}''');
-
-  test('''
-class S1 {}
-class S2 {}
-class A extends S1 implements S2 {}''');
-
-  test('''
-class S1 {}
-class S2 {}
-class S3 {}
-class A extends S1 implements S2, S3 {}''');
-
-  test('''
-class S1 {}
-class S2 {}
-class A implements S1, S2 {}''');
-
-  test('''
-class S1 {}
-class S2 {}
-class A extends Object implements S1, S2 {}''',
-       expectedDeclarations: '''
-class S1 {}
-class S2 {}
-class A implements S1, S2 {}''');
-
-  test('''
-class S1 {}
-class A extends Object with S1 {}''');
-
-  test('''
-class S1 {}
-class A = Object with S1;''');
-
-  test('''
-class S1 {}
-class S2 {}
-class A extends S1 with S2 {}''');
-
-  test('''
-class S1 {}
-class S2 {}
-class A = S1 with S2;''');
-
-  test('''
-class S1 {}
-class S2 {}
-class S3 {}
-class A extends S1 with S2, S3 {}''');
-
-  test('''
-class S1 {}
-class S2 {}
-class S3 {}
-class A = S1 with S2, S3;''');
-
-  test('''
-class S1 {}
-class S2 {}
-class S3 {}
-class S4 {}
-class S5 {}
-class A extends S1 with S2, S3 implements S4, S5 {}''');
-
-  test('''
-class S1 {}
-class S2 {}
-class S3 {}
-class S4 {}
-class S5 {}
-class A = S1 with S2, S3 implements S4, S5;''');
-
-  test('''
-class S1 {}
-class A extends Object with S1 implements S1 {}''',
-       expectedDeclarations: '''
-class S1 {}
-class A extends Object with S1 {}''');
-
-  test('''
-class S1 {}
-class A = Object with S1 implements S1;''',
-       expectedDeclarations: '''
-class S1 {}
-class A = Object with S1;''');
-
-  test('''
-class S1<T1> {}
-class S2<T2> {}
-class S3<T3> {}
-class S4<T4> {}
-class S5<T5, T6> {}
-class A<U1, U2, U3, U4, U5> extends S1<U1> with S2<U2>, S3<U3> '''
- '''implements S4<U4>, S5<U5, S5<U5, int>> {}''');
-}
-
-main() {
-  testSimpleFileUnparse();
-  testTopLevelField();
-  testSimpleTopLevelClass();
-  testClassWithSynthesizedConstructor();
-  testClassWithMethod();
-  testExtendsImplements();
-  testVariableDefinitions();
-  testGetSet();
-  testAbstractClass();
-  testConflictSendsRename();
-  testNoConflictSendsRename();
-  testConflictLibraryClassRename();
-  testClassExtendsWithArgs();
-  testStaticInvocation();
-  testLibraryGetSet();
-  testFieldTypeOutput();
-  testTypeVariablesAreRenamed();
-  testClassTypeArgumentBound();
-  testDoubleMains();
-  testStaticAccessIoLib();
-  testLocalFunctionPlaceholder();
-  testMinification();
-  testClosureLocalsMinified();
-  testParametersMinified();
-  testDeclarationTypePlaceholders();
-  testPlatformLibraryMemberNamesAreFixed();
-  testConflictsWithCoreLib();
-  testUnresolvedNamedConstructor1();
-  testUnresolvedNamedConstructor2();
-  testUnresolvedNamedConstructor3();
-  testClassAndNamedMixinDeclarations();
-}
-

tests/compiler/dart2js/backend_dart/dart_printer_test.dart

diff --git a/tests/compiler/dart2js/backend_dart/dart_printer_test.dart b/tests/compiler/dart2js/backend_dart/dart_printer_test.dart
deleted file mode 100644
index 4d331b3..0000000
--- a/tests/compiler/dart2js/backend_dart/dart_printer_test.dart
+++ /dev/null
@@ -1,990 +0,0 @@
-// Copyright (c) 2014, the Dart project authors.  Please see the AUTHORS file
-// for details. All rights reserved. Use of this source code is governed by a
-// BSD-style license that can be found in the LICENSE file.
-
-library dart_printer_test;
-
-import 'dart:mirrors';
-import 'package:expect/expect.dart';
-import 'package:compiler/src/constants/values.dart';
-import 'package:compiler/src/dart_backend/backend_ast_nodes.dart';
-import 'package:compiler/src/dart_backend/backend_ast_to_frontend_ast.dart'
-    show TreePrinter;
-import 'package:compiler/src/diagnostics/diagnostic_listener.dart';
-import 'package:compiler/src/diagnostics/messages.dart';
-import 'package:compiler/src/diagnostics/spannable.dart' show Spannable;
-import 'package:compiler/src/parser/listener.dart';
-import 'package:compiler/src/parser/parser.dart';
-import 'package:compiler/src/scanner/scanner.dart';
-import 'package:compiler/src/tokens/token.dart';
-import 'package:compiler/src/tokens/token_constants.dart';
-import 'package:compiler/src/io/source_file.dart';
-import 'package:compiler/src/string_validator.dart';
-import 'package:compiler/src/tree/tree.dart' show DartString;
-import 'package:compiler/src/tree/tree.dart' as tree;
-import '../options_helper.dart';
-
-/// For debugging the [AstBuilder] stack. Prints information about [x].
-void show(x) {
-  StringBuffer buf = new StringBuffer();
-  Unparser unparser = new Unparser(buf);
-  void unparse(x) {
-    if (x is Expression)
-      unparser.writeExpression(x);
-    else if (x is TypeAnnotation)
-      unparser.writeType(x);
-    else if (x is Statement)
-      unparser.writeStatement(x);
-    else if (x is List) {
-      buf.write('[');
-      bool first = true;
-      for (var y in x) {
-        if (first)
-          first = false;
-        else
-          buf.write(', ');
-        unparse(y);
-      }
-      buf.write(']');
-    }
-  }
-  unparse(x);
-  print("${x.runtimeType}: ${buf.toString()}");
-}
-
-class PrintDiagnosticListener implements DiagnosticReporter {
-  void log(message) {
-    print(message);
-  }
-
-  void internalError(Spannable spannable, message) {
-    print(message);
-  }
-
-  SourceSpan spanFromSpannable(Spannable node) {
-    return new SourceSpan(null, 0, 0);
-  }
-
-  void reportFatalError(Spannable node, MessageKind errorCode,
-                        [Map arguments = const {}]) {
-    print(errorCode);
-    throw new Error();
-  }
-
-  void reportError(Spannable node, MessageKind errorCode,
-                   [Map arguments = const {}]) {
-    print(errorCode);
-  }
-
-  void reportWarning(Spannable node, MessageKind errorCode,
-                     [Map arguments = const {}]) {
-    print(errorCode);
-  }
-
-  void reportHint(Spannable node, MessageKind errorCode,
-                  [Map arguments = const {}]) {
-    print(errorCode);
-  }
-
-  void reportInfo(Spannable node, MessageKind errorCode,
-                  [Map arguments = const {}]) {
-    print(errorCode);
-  }
-
-  withCurrentElement(element, f()) {
-    f();
-  }
-}
-
-class AstBuilder extends Listener {
-  final List stack = [];
-  final StringValidator stringValidator
-         = new StringValidator(new PrintDiagnosticListener());
-
-  String asName(e) {
-    if (e is Identifier)
-      return e.name;
-    else if (e == null)
-      return null;
-    else
-      throw 'Expression is not a name: ${e.runtimeType}';
-  }
-
-  TypeAnnotation asType(x) {
-    if (x is TypeAnnotation)
-      return x;
-    if (x is Identifier)
-      return new TypeAnnotation(x.name);
-    if (x == null)
-      return null;
-    else
-      throw "Not a type: ${x.runtimeType}";
-  }
-
-  Parameter asParameter(x) {
-    if (x is Parameter)
-      return x;
-    if (x is Identifier)
-      return new Parameter(x.name);
-    else
-      throw "Not a parameter: ${x.runtimeType}";
-  }
-
-  void push(node) {
-    stack.add(node);
-  }
-  dynamic peek() {
-    return stack.last;
-  }
-  dynamic pop([coerce(x) = null]) {
-    var x = stack.removeLast();
-    if (coerce != null)
-      return coerce(x);
-    else
-      return x;
-  }
-  List popList(int count, [List result, coerce(x) = null]) {
-    if (result == null)
-      result = <Node>[];
-    for (int i=0; i<count; i++) {
-      var x = stack[stack.length-count+i];
-      if (coerce != null) {
-        x = coerce(x);
-      }
-      result.add(x);
-    }
-    stack.removeRange(stack.length-count, stack.length);
-    return result;
-  }
-  popTypeAnnotation() {
-    List<TypeAnnotation> args = pop();
-    if (args == null)
-      return null;
-    String name = pop(asName);
-    return new TypeAnnotation(name, args);
-  }
-
-  // EXPRESSIONS
-  endCascade() {
-    throw "Cascade not supported yet";
-  }
-  endIdentifierList(int count) {
-    push(popList(count, <Identifier>[]));
-  }
-  endTypeList(int count) {
-    push(popList(count, <TypeAnnotation>[], asType));
-  }
-  beginLiteralString(Token token) {
-    String source = token.value;
-    tree.StringQuoting quoting = StringValidator.quotingFromString(source);
-    push(quoting);
-    push(token); // collect token at the end
-  }
-  handleStringPart(Token token) {
-    push(token); // collect token at the end
-  }
-  endLiteralString(int interpCount) {
-    List parts = popList(2 * interpCount + 1, []);
-    tree.StringQuoting quoting = pop();
-    List<Expression> members = <Expression>[];
-    for (var i=0; i<parts.length; i++) {
-      var part = parts[i];
-      if (part is Expression) {
-        members.add(part);
-      } else {
-        assert(part is Token);
-        DartString str = stringValidator.validateInterpolationPart(
-            part as Token,
-            quoting,
-            isFirst: i == 0,
-            isLast: i == parts.length - 1);
-        members.add(new Literal(new StringConstantValue(str)));
-      }
-    }
-    push(new StringConcat(members));
-  }
-  handleStringJuxtaposition(int litCount) {
-    push(new StringConcat(popList(litCount, <Expression>[])));
-  }
-  endArguments(int count, begin, end) {
-    push(popList(count, <Argument>[]));
-  }
-  handleNoArguments(token) {
-    push(null);
-  }
-  handleNoTypeArguments(token) {
-    push(<TypeAnnotation>[]);
-  }
-  endTypeArguments(int count, t, y) {
-    List<TypeAnnotation> args = <TypeAnnotation>[];
-    for (var i=0; i<count; i++) {
-      args.add(popTypeAnnotation());
-    }
-    push(args.reversed.toList(growable:false));
-  }
-  handleVoidKeyword(token) {
-    push(new Identifier("void"));
-    push(<TypeAnnotation>[]); // prepare for popTypeAnnotation
-  }
-  handleQualified(Token period) {
-    String last = pop(asName);
-    String first = pop(asName);
-    push(new Identifier('$first.$last'));
-  }
-  endSend(t) {
-    List<Argument> arguments = pop();
-    pop(); // typeArguments
-    if (arguments == null)
-      return; // not a function call
-    Expression selector = pop();
-    push(new CallFunction(selector, arguments));
-  }
-  endThrowExpression(t, tt) {
-    push(new Throw(pop()));
-  }
-  handleAssignmentExpression(Token token) {
-    Expression right = pop();
-    Expression left = pop();
-    push(new Assignment(left, token.value, right));
-  }
-  handleBinaryExpression(Token token) {
-    Expression right = pop();
-    Receiver left = pop();
-    String tokenString = token.stringValue;
-    if (tokenString == '.') {
-      if (right is CallFunction) {
-        String name = (right.callee as Identifier).name;
-        push(new CallMethod(left, name, right.arguments));
-      } else {
-        push(new FieldExpression(left, (right as Identifier).name));
-      }
-    } else {
-      push(new BinaryOperator(left, tokenString, right));
-    }
-  }
-  handleConditionalExpression(question, colon) {
-    Expression elseExpression = pop();
-    Expression thenExpression = pop();
-    Expression condition = pop();
-    push(new Conditional(condition, thenExpression, elseExpression));
-  }
-  handleIdentifier(Token t) {
-    push(new Identifier(t.value));
-  }
-  handleOperator(t) {
-    push(new Identifier(t.value));
-  }
-  handleIndexedExpression(open, close) {
-    Expression index = pop();
-    Receiver object = pop();
-    push(new IndexExpression(object, index));
-  }
-  handleIsOperator(operathor, not, endToken) {
-    TypeAnnotation type = popTypeAnnotation();
-    Expression exp = pop();
-    TypeOperator r = new TypeOperator(exp, 'is', type);
-    if (not != null) {
-      push(new UnaryOperator('!', r));
-    } else {
-      push(r);
-    }
-  }
-  handleAsOperator(operathor, endToken) {
-    TypeAnnotation type = popTypeAnnotation();
-    Expression exp = pop();
-    push(new TypeOperator(exp, 'as', type));
-  }
-  handleLiteralBool(Token t) {
-    bool value = t.value == 'true';
-    push(new Literal(
-        value ? new TrueConstantValue() : new FalseConstantValue()));
-  }
-  handleLiteralDouble(t) {
-    push(new Literal(new DoubleConstantValue(double.parse(t.value))));
-  }
-  handleLiteralInt(Token t) {
-    push(new Literal(new IntConstantValue(int.parse(t.value))));
-  }
-  handleLiteralNull(t) {
-    push(new Literal(new NullConstantValue()));
-  }
-  endLiteralSymbol(Token hash, int idCount) {
-    List<Identifier> ids = popList(idCount, <Identifier>[]);
-    push(new LiteralSymbol(ids.map((id) => id.name).join('.')));
-  }
-  handleLiteralList(int count, begin, constKeyword, end) {
-    List<Expression> exps = popList(count, <Expression>[]);
-    List<TypeAnnotation> types = pop();
-    assert(types.length <= 1);
-    push(new LiteralList(exps,
-      isConst: constKeyword != null,
-      typeArgument: types.length == 0 ? null : types[0]
-    ));
-  }
-  handleLiteralMap(int count, begin, constKeyword, end) {
-    List<LiteralMapEntry> entries = popList(count, <LiteralMapEntry>[]);
-    List<TypeAnnotation> types = pop();
-    assert(types.length == 0 || types.length == 2);
-    push(new LiteralMap(entries,
-        isConst: constKeyword != null,
-        typeArguments: types
-    ));
-  }
-  endLiteralMapEntry(colon, endToken) {
-    Expression value = pop();
-    Expression key = pop();
-    push(new LiteralMapEntry(key,value));
-  }
-  handleNamedArgument(colon) {
-    Expression exp = pop();
-    Identifier name = pop();
-    push(new NamedArgument(name.name, exp));
-  }
-  endConstructorReference(Token start, Token period, Token end) {
-    if (period == null) {
-      push(null); // indicate missing constructor name
-    }
-  }
-  handleNewExpression(t) {
-    List<Argument> args = pop();
-    String constructorName = pop(asName);
-    TypeAnnotation type = popTypeAnnotation();
-    push(new CallNew(type, args, constructorName: constructorName));
-  }
-  handleConstExpression(t) {
-    List<Argument> args = pop();
-    String constructorName = pop(asName);
-    TypeAnnotation type = popTypeAnnotation();
-    push(new CallNew(type, args, constructorName: constructorName,
-                     isConst:true));
-  }
-  handleParenthesizedExpression(t) {
-    // do nothing, just leave expression on top of stack
-  }
-  handleSuperExpression(t) {
-    push(new SuperReceiver());
-  }
-  handleThisExpression(t) {
-    push(new This());
-  }
-  handleUnaryPostfixAssignmentExpression(Token t) {
-    push(new Increment.postfix(pop(), t.value));
-  }
-  handleUnaryPrefixAssignmentExpression(Token t) {
-    push(new Increment.prefix(pop(), t.value));
-  }
-  handleUnaryPrefixExpression(Token t) {
-    push(new UnaryOperator(t.value, pop()));
-  }
-
-  handleFunctionTypedFormalParameter(tok) {
-    // handled in endFormalParameter
-  }
-  endFormalParameter(thisKeyword) {
-    Expression defaultValue = null;
-    var x = pop();
-    if (x is DefaultValue) {
-      defaultValue = x.expression;
-      x = pop();
-    }
-    if (x is Parameters) {
-      String name = pop(asName);
-      TypeAnnotation returnType = popTypeAnnotation();
-      push(new Parameter.function(name, returnType, x, defaultValue));
-    } else {
-      String name = asName(x);
-      TypeAnnotation type = popTypeAnnotation();
-      push(new Parameter(name, type:type, defaultValue:defaultValue));
-    }
-  }
-  handleValuedFormalParameter(eq, tok) {
-    push(new DefaultValue(pop()));
-  }
-  endOptionalFormalParameters(int count, begin, end) {
-    bool isNamed = end.value == '}';
-    push(popList(count, <Parameter>[], asParameter));
-    push(isNamed); // Indicate optional parameters to endFormalParameters.
-  }
-  endFormalParameters(count, begin, end) {
-    if (count == 0) {
-      push(new Parameters([]));
-      return;
-    }
-    var last = pop();   // Detect if optional parameters are present.
-    if (last is bool) { // See endOptionalFormalParameters.
-      List<Parameter> optional = pop();
-      List<Parameter> required = popList(count-1, <Parameter>[], asParameter);
-      push(new Parameters(required, optional, last));
-    } else {
-      // No optional parameters.
-      List<Parameter> required = popList(count-1, <Parameter>[], asParameter);
-      required.add(last);
-      push(new Parameters(required));
-    }
-  }
-  handleNoFormalParameters(tok) {
-    push(new Parameters([]));
-  }
-
-  endUnnamedFunction(t) {
-    Statement body = pop();
-    Parameters parameters = pop();
-    push(new FunctionExpression(parameters, body));
-  }
-
-  handleNoType(Token token) {
-    push(null);
-  }
-
-  endReturnStatement(bool hasExpression, begin, end) {
-    // This is also called for functions whose body is "=> expression"
-    if (hasExpression) {
-      push(new Return(pop()));
-    } else {
-      push(new Return());
-    }
-  }
-
-  endExpressionStatement(Token token) {
-    push(new ExpressionStatement(pop()));
-  }
-
-  endDoWhileStatement(Token doKeyword, Token whileKeyword, Token end) {
-    Expression condition = pop();
-    Statement body = pop();
-    push(new DoWhile(body, condition));
-  }
-
-  endWhileStatement(Token whileKeyword, Token end) {
-    Statement body = pop();
-    Expression condition = pop();
-    push(new While(condition, body));
-  }
-
-  endBlock(int count, Token begin, Token end) {
-    push(new Block(popList(count, <Statement>[])));
-  }
-
-  endRethrowStatement(Token throwToken, Token endToken) {
-    push(new Rethrow());
-  }
-
-  endTryStatement(int catchCount, Token tryKeyword, Token finallyKeyword) {
-    Statement finallyBlock = null;
-    if (finallyKeyword != null) {
-      finallyBlock = pop();
-    }
-    List<CatchBlock> catchBlocks = popList(catchCount, <CatchBlock>[]);
-    Statement tryBlock = pop();
-    push(new Try(tryBlock, catchBlocks, finallyBlock));
-  }
-
-  void handleCatchBlock(Token onKeyword, Token catchKeyword) {
-    Statement block = pop();
-    String exceptionVar = null;
-    String stackVar = null;
-    if (catchKeyword != null) {
-      Parameters params = pop();
-      exceptionVar = params.requiredParameters[0].name;
-      if (params.requiredParameters.length > 1) {
-        stackVar = params.requiredParameters[1].name;
-      }
-    }
-    TypeAnnotation type = onKeyword == null ? null : pop();
-    push(new CatchBlock(block,
-      onType: type,
-      exceptionVar: exceptionVar,
-      stackVar: stackVar
-    ));
-  }
-
-  endSwitchStatement(Token switchKeyword, Token end) {
-    List<SwitchCase> cases = pop();
-    Expression expression = pop();
-    push(new Switch(expression, cases));
-  }
-
-  endSwitchBlock(int caseCount, Token begin, Token end) {
-    push(popList(caseCount, <SwitchCase>[]));
-  }
-
-  handleSwitchCase(int labelCount, int caseCount, Token defaultKeyword,
-                   int statementCount, Token first, Token end) {
-    List<Statement> statements = popList(statementCount, <Statement>[]);
-    List<Expression> cases = popList(caseCount, <Expression>[]);
-    if (defaultKeyword != null) {
-      cases = null;
-    }
-    push(new SwitchCase(cases, statements));
-  }
-
-  handleCaseMatch(Token caseKeyword, Token colon) {
-    // do nothing, leave case expression on stack
-  }
-
-  handleBreakStatement(bool hasTarget, Token breakKeyword, Token end) {
-    String target = hasTarget ? pop(asName) : null;
-    push(new Break(target));
-  }
-
-  handleContinueStatement(bool hasTarget, Token continueKeyword, Token end) {
-    String target = hasTarget ? pop(asName) : null;
-    push(new Continue(target));
-  }
-
-  handleEmptyStatement(Token token) {
-    push(new EmptyStatement());
-  }
-
-
-  VariableDeclaration asVariableDeclaration(x) {
-    if (x is VariableDeclaration)
-      return x;
-    if (x is Identifier)
-      return new VariableDeclaration(x.name);
-    throw "Not a variable definition: ${x.runtimeType}";
-  }
-
-  endVariablesDeclaration(int count, Token end) {
-    List<VariableDeclaration> variables =
-        popList(count, <VariableDeclaration>[], asVariableDeclaration);
-    TypeAnnotation type = popTypeAnnotation();
-    push(new VariableDeclarations(variables,
-      type: type,
-      isFinal: false, // TODO(asgerf): Parse modifiers.
-      isConst: false
-    ));
-  }
-
-  endInitializer(Token assign) {
-    Expression init = pop();
-    String name = pop(asName);
-    push(new VariableDeclaration(name, init));
-  }
-
-  endIfStatement(Token ifToken, Token elseToken) {
-    Statement elsePart = (elseToken == null) ? null : pop();
-    Statement thenPart = pop();
-    Expression condition = pop();
-    push(new If(condition, thenPart, elsePart));
-  }
-
-  endForStatement(int updateCount, Token begin, Token end) {
-    Statement body = pop();
-    List<Expression> updates = popList(updateCount, <Expression>[]);
-    ExpressionStatement condition = pop(); // parsed as expression statement
-    Expression exp = condition == null ? null : condition.expression;
-    Node initializer = pop();
-    push(new For(initializer, exp, updates, body));
-  }
-
-  handleNoExpression(Token token) {
-    push(null);
-  }
-
-  endForIn(Token await, Token begin, Token inKeyword, Token end) {
-    Statement body = pop();
-    Expression exp = pop();
-    Node declaredIdentifier = pop();
-    push(new ForIn(declaredIdentifier, exp, body));
-  }
-
-  handleAssertStatement(Token assertKeyword,
-                        Token commaToken, Token semicolonToken) {
-    Expression message;
-    if (commaToken != null) message = pop();
-    Expression exp = pop();
-    var arguments = [exp];
-    if (message != null) arguments.add(message);
-    Expression call = new CallFunction(new Identifier("assert"), arguments);
-    push(new ExpressionStatement(call));
-  }
-
-  endLabeledStatement(int labelCount) {
-    Statement statement = pop();
-    for (int i=0; i<labelCount; i++) {
-      String label = pop(asName);
-      statement = new LabeledStatement(label, statement);
-    }
-    push(statement);
-  }
-
-  endFunctionDeclaration(Token end) {
-    Statement body = pop();
-    Parameters parameters = pop();
-    String name = pop(asName);
-    TypeAnnotation returnType = popTypeAnnotation();
-    push(new FunctionDeclaration(new FunctionExpression(parameters, body,
-        name: name,
-        returnType: returnType)));
-  }
-
-  endFunctionBody(int count, Token begin, Token end) {
-    push(new Block(popList(count, <Statement>[])));
-  }
-}
-
-class DefaultValue {
-  final Expression expression;
-  DefaultValue(this.expression);
-}
-
-/// Compares ASTs for structural equality.
-void checkDeepEqual(x, y) {
-  if (x is List && y is List) {
-    if (x.length != y.length)
-      return;
-    for (var i=0; i<x.length; i++) {
-      checkDeepEqual(x[i], y[i]);
-    }
-  }
-  else if (x is Node && y is Node) {
-    if (x.runtimeType != y.runtimeType)
-      throw new Error();
-    InstanceMirror xm = reflect(x);
-    InstanceMirror ym = reflect(y);
-    for (Symbol name in xm.type.instanceMembers.keys) {
-      if (reflectClass(Object).declarations.containsKey(name)) {
-        continue; // do not check things from Object, such as hashCode
-      }
-      MethodMirror mm = xm.type.instanceMembers[name];
-      if (mm.isGetter) {
-        var xv = xm.getField(name).reflectee;
-        var yv = ym.getField(name).reflectee;
-        checkDeepEqual(xv,yv);
-      }
-    }
-  }
-  else if (x is PrimitiveConstantValue && y is PrimitiveConstantValue) {
-    checkDeepEqual(x.primitiveValue, y.primitiveValue);
-  }
-  else if (x is DartString && y is DartString) {
-    if (x.slowToString() != y.slowToString()) {
-      throw new Error();
-    }
-  }
-  else {
-    if (x != y) {
-      throw new Error();
-    }
-  }
-}
-
-Expression parseExpression(String code) {
-  SourceFile file = new StringSourceFile.fromName('', code);
-  Scanner scan = new Scanner(file);
-  Token tok = scan.tokenize();
-  AstBuilder builder = new AstBuilder();
-  Parser parser = new Parser(builder, new MockParserOptions());
-  tok = parser.parseExpression(tok);
-  if (builder.stack.length != 1 || tok.kind != EOF_TOKEN) {
-    throw "Parse error in $code";
-  }
-  return builder.pop();
-}
-Statement parseStatement(String code) {
-  SourceFile file = new StringSourceFile.fromName('', code);
-  Scanner scan = new Scanner(file);
-  Token tok = scan.tokenize();
-  AstBuilder builder = new AstBuilder();
-  Parser parser = new Parser(builder, new MockParserOptions());
-  tok = parser.parseStatement(tok);
-  if (builder.stack.length != 1 || tok.kind != EOF_TOKEN) {
-    throw "Parse error in $code";
-  }
-  return builder.pop();
-}
-
-String unparseExpression(Expression exp) {
-  StringBuffer buf = new StringBuffer();
-  new Unparser(buf).writeExpression(exp);
-  return buf.toString();
-}
-String unparseStatement(Statement stmt) {
-  StringBuffer buf = new StringBuffer();
-  new Unparser(buf).writeStatement(stmt);
-  return buf.toString();
-}
-
-/// Converts [exp] to an instance of the frontend AST and unparses that.
-String frontUnparseExpression(Expression exp) {
-  tree.Node node = new TreePrinter().makeExpression(exp);
-  return tree.unparse(node);
-}
-/// Converts [stmt] to an instance of the frontend AST and unparses that.
-String frontUnparseStatement(Statement stmt) {
-  tree.Node node = new TreePrinter().makeStatement(stmt);
-  return tree.unparse(node);
-}
-
-/// Parses [code], unparses the resulting AST, then parses the unparsed text.
-/// The ASTs from the first and second parse are then compared for structural
-/// equality. Alternatively, if [expected] is not an empty string, the second
-/// parse must match the AST of parsing [expected].
-void checkFn(String code, String expected, Function parse, Function unparse) {
-  var firstParse = parse(code);
-  String unparsed = unparse(firstParse);
-  try {
-    var secondParse = parse(unparsed);
-    var baseline = expected == "" ? firstParse : parse(expected);
-    checkDeepEqual(baseline, secondParse);
-  } catch (e, stack) {
-    Expect.fail('"$code" was unparsed as "$unparsed"');
-  }
-}
-
-void checkExpression(String code, [String expected="", String expected2=""]) {
-  checkFn(code, expected, parseExpression, unparseExpression);
-  checkFn(code, expected2, parseExpression, frontUnparseExpression);
-}
-void checkStatement(String code, [String expected="", String expected2=""]) {
-  checkFn(code, expected, parseStatement, unparseStatement);
-  checkFn(code, expected2, parseStatement, frontUnparseStatement);
-}
-
-void debugTokens(String code) {
-  SourceFile file = new StringSourceFile.fromName('', code);
-  Scanner scan = new Scanner(file);
-  Token tok = scan.tokenize();
-  while (tok.next != tok) {
-    print(tok.toString());
-    tok = tok.next;
-  }
-}
-
-void main() {
-  // To check if these tests are effective, one should manually alter
-  // something in [Unparser] and see if a test fails.
-
-  checkExpression(" a +  b  + c");
-  checkExpression("(a +  b) + c");
-  checkExpression(" a + (b  + c)");
-
-  checkExpression(" a +  b  - c");
-  checkExpression("(a +  b) - c");
-  checkExpression(" a + (b  - c)");
-
-  checkExpression(" a -  b  + c");
-  checkExpression("(a -  b) + c");
-  checkExpression(" a - (b  + c)");
-
-  checkExpression(" a *  b  + c");
-  checkExpression("(a *  b) + c");
-  checkExpression(" a * (b  + c)");
-
-  checkExpression(" a +  b  * c");
-  checkExpression("(a +  b) * c");
-  checkExpression(" a + (b  * c)");
-
-  checkExpression(" a *  b  * c");
-  checkExpression("(a *  b) * c");
-  checkExpression(" a * (b  * c)");
-
-  checkExpression("a is T");
-  checkExpression("a is! T");
-  checkExpression("!(a is T)");
-
-  checkExpression("a is T.x");
-  checkExpression("a is! T.x");
-  checkExpression("!(a is T.x)");
-  checkExpression("!(a is T).x");
-
-  checkExpression("a as T.x");
-  checkExpression("(a as T).x");
-
-  checkExpression("a == b");
-  checkExpression("a != b");
-  checkExpression("!(a == b)", "a != b");
-
-  checkExpression("a && b ? c : d");
-  checkExpression("(a && b) ? c : d");
-  checkExpression("a && (b ? c : d)");
-
-  checkExpression("a || b ? c : d");
-  checkExpression("(a || b) ? c : d");
-  checkExpression("a || (b ? c : d)");
-
-  checkExpression(" a ? b :  c && d");
-  checkExpression(" a ? b : (c && d)");
-  checkExpression("(a ? b :  c) && d");
-
-  checkExpression(" a ? b : c = d");
-  checkExpression(" a ? b : (c = d)");
-
-  checkExpression("(a == b) == c");
-  checkExpression("a == (b == c)");
-
-  checkExpression(" a <  b  == c");
-  checkExpression("(a <  b) == c");
-  checkExpression(" a < (b  == c)");
-
-  checkExpression(" a ==  b  < c");
-  checkExpression("(a ==  b) < c");
-  checkExpression(" a == (b  < c)");
-
-  checkExpression("x.f()");
-  checkExpression("(x.f)()");
-
-  checkExpression("x.f()()");
-  checkExpression("(x.f)()()");
-
-  checkExpression("x.f().g()");
-  checkExpression("(x.f)().g()");
-
-  checkExpression("x.f()");
-  checkExpression("x.f(1 + 2)");
-  checkExpression("x.f(1 + 2, 3 + 4)");
-  checkExpression("x.f(1 + 2, foo:3 + 4)");
-  checkExpression("x.f(1 + 2, foo:3 + 4, bar: 5)");
-  checkExpression("x.f(foo:3 + 4)");
-  checkExpression("x.f(foo:3 + 4, bar: 5)");
-
-  checkExpression("x.f.g.h");
-  checkExpression("(x.f).g.h");
-  checkExpression("(x.f.g).h");
-
-  checkExpression(" a =  b  + c");
-  checkExpression(" a = (b  + c)");
-  checkExpression("(a =  b) + c");
-
-  checkExpression("a + (b = c)");
-
-  checkExpression("dx * dx + dy * dy < r * r",
-                  "((dx * dx) + (dy * dy)) < (r * r)");
-  checkExpression("mid = left + right << 1",
-                  "mid = ((left + right) << 1)");
-  checkExpression("a + b % c * -d ^  e - f  ~/ x & ++y / z++ | w > a ? b : c");
-  checkExpression("a + b % c * -d ^ (e - f) ~/ x & ++y / z++ | w > a ? b : c");
-
-  checkExpression("'foo'");
-  checkExpression("'foo' 'bar'", "'foobar'");
-
-  checkExpression("{}.length");
-  checkExpression("{x: 1+2}.length");
-  checkExpression("<String,int>{}.length");
-  checkExpression("<String,int>{x: 1+2}.length");
-
-  checkExpression("[].length");
-  checkExpression("[1+2].length");
-  checkExpression("<num>[].length");
-  checkExpression("<num>[1+2].length");
-
-  checkExpression("x + -y");
-  checkExpression("x + --y");
-  checkExpression("x++ + y");
-  checkExpression("x + ++y");
-  checkExpression("x-- - y");
-  checkExpression("x-- - -y");
-  checkExpression("x - --y");
-
-  checkExpression("x && !y");
-  checkExpression("!x && y");
-  checkExpression("!(x && y)");
-
-  checkExpression(" super +  1  * 2");
-  checkExpression("(super +  1) * 2");
-  checkExpression(" super + (1  * 2)");
-  checkExpression("x + -super");
-  checkExpression("x-- - -super");
-  checkExpression("x - -super");
-  checkExpression("x && !super");
-
-  checkExpression("super.f(1, 2) + 3");
-  checkExpression("super.f + 3");
-
-  checkExpression(r"'foo\nbar'");
-  checkExpression(r"'foo\r\nbar'");
-  checkExpression(r"'foo\rbar'");
-  checkExpression(r"'foo\'bar'");
-  checkExpression(r"""'foo"bar'""");
-  checkExpression(r"r'foo\nbar'");
-  checkExpression("''");
-  checkExpression("r''");
-
-  var sq = "'";
-  var dq = '"';
-  checkExpression("'$dq$dq' \"$sq$sq\"");
-  checkExpression("'$dq$dq$dq$dq' \"$sq$sq$sq$sq\"");
-  checkExpression(r"'\$\$\$\$\$\$\$\$\$'");
-  checkExpression("'$dq$dq$dq' '\\n\\n\\n\\n\\n\\n\\n\\n\\n\\n' \"$sq$sq$sq\"");
-  checkExpression("'$dq$dq$dq' '\\r\\r\\r\\r\\r\\r\\r\\r\\r\\r' \"$sq$sq$sq\"");
-  checkExpression("'$dq$dq$dq' '\\r\\n\\r\\n\\r\\n\\r\\n\\r\\n' \"$sq$sq$sq\"");
-
-  checkExpression(r"'$foo'");
-  checkExpression(r"'${foo}x'");
-  checkExpression(r"'${foo}x\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\'");
-  checkExpression(r"'abc' '${foo}' r'\\\\\\\'");
-
-  checkExpression(r"'${$x}'");
-  checkExpression(r"'${$x}y'");
-  checkExpression("null + null");
-
-  checkExpression("(x) => x",
-                  '',
-                  '(x){return x;}');
-  checkStatement("fn(x) => x;",
-                  '',
-                  'fn(x){return x;}');
-
-  checkExpression("throw x");
-  checkStatement("throw x;");
-
-  checkStatement("var x, y, z;");
-  checkStatement("final x, y, z;");
-  checkStatement("dynamic x, y, z;");
-  checkStatement("String x, y, z;");
-  checkStatement("List<int> x, y, z;");
-  checkStatement("final dynamic x, y, z;");
-  checkStatement("final String x, y, z;");
-  checkStatement("final List<int> x, y, z;");
-
-  checkStatement("var x = y, z;");
-  checkStatement("var x, y = z;");
-  checkStatement("var x = y = z;");
-
-  // Note: We sometimes have to pass an expected string to account for
-  //       block flattening which does not preserve structural AST equality
-  checkStatement("if (x)   if (y) foo();   else bar();  ");
-  checkStatement("if (x) { if (y) foo(); } else bar();  ");
-  checkStatement("if (x) { if (y) foo();   else bar(); }",
-                 "if (x)   if (y) foo();   else bar();  ");
-
-  checkStatement("if (x) while (y)   if (z) foo();   else bar();  ");
-  checkStatement("if (x) while (y) { if (z) foo(); } else bar();  ");
-  checkStatement("if (x) while (y) { if (z) foo();   else bar(); }",
-                 "if (x) while (y)   if (z) foo();   else bar();  ");
-
-  checkStatement("{var x = 1; {var x = 2;} return x;}");
-  checkStatement("{var x = 1; {x = 2;} return x;}",
-                 "{var x = 1;  x = 2;  return x;}",
-                 "{var x = 1;  x = 2;  return x;}");
-
-  checkStatement("if (x) {var x = 1;}");
-
-  checkStatement("({'foo': 1}).bar();");
-  checkStatement("({'foo': 1}).length;");
-  checkStatement("({'foo': 1}).length + 1;");
-  checkStatement("({'foo': 1})['foo'].toString();");
-  checkStatement("({'foo': 1})['foo'] = 3;");
-  checkStatement("({'foo': 1}['foo']());");
-  checkStatement("({'foo': 1}['foo'])();");
-  checkStatement("({'foo': 1})['foo'].x++;");
-  checkStatement("({'foo': 1}) is Map;");
-  checkStatement("({'foo': 1}) as Map;");
-  checkStatement("({'foo': 1}) is util.Map;");
-  checkStatement("({'foo': 1}) + 1;");
-
-  checkStatement("[1].bar();");
-  checkStatement("1.bar();");
-  checkStatement("'foo'.bar();");
-
-  checkStatement("do while(x); while (y);");
-  checkStatement("{do; while(x); while (y);}");
-
-  checkStatement('switch(x) { case 1: case 2: return y; }');
-  checkStatement('switch(x) { default: return y; }');
-  checkStatement('switch(x) { case 1: x=y; default: return y; }');
-  checkStatement('switch(x) { case 1: x=y; y=z; break; default: return y; }');
-
-}
-

tests/compiler/dart2js/backend_dart/end2end_test.dart

diff --git a/tests/compiler/dart2js/backend_dart/end2end_test.dart b/tests/compiler/dart2js/backend_dart/end2end_test.dart
deleted file mode 100644
index 60156c7..0000000
--- a/tests/compiler/dart2js/backend_dart/end2end_test.dart
+++ /dev/null
@@ -1,37 +0,0 @@
-// Copyright (c) 2014, the Dart project authors.  Please see the AUTHORS file
-// for details. All rights reserved. Use of this source code is governed by a
-// BSD-style license that can be found in the LICENSE file.
-
-/// End-to-end test of the dart2dart compiler.
-library dart_backend.end2end_test;
-
-import 'package:async_helper/async_helper.dart';
-import 'package:compiler/src/compiler.dart';
-import 'package:compiler/src/dart_backend/dart_backend.dart';
-import 'package:expect/expect.dart';
-
-import '../../../../pkg/analyzer2dart/test/test_helper.dart' hide TestSpec;
-import '../../../../pkg/analyzer2dart/test/end2end_data.dart';
-
-import 'test_helper.dart';
-import '../output_collector.dart';
-
-main(List<String> args) {
-  performTests(TEST_DATA, asyncTester, runTest, args);
-}
-
-runTest(TestSpec result) {
-  OutputCollector outputCollector = new OutputCollector();
-  asyncTest(() => compilerFor(result.input, outputProvider: outputCollector)
-      .then((Compiler compiler) {
-    String expectedOutput = result.output.trim();
-    compiler.phase = Compiler.PHASE_COMPILING;
-    DartBackend backend = compiler.backend;
-    backend.assembleProgram();
-    String output = outputCollector.getOutput('', 'dart').trim();
-    Expect.equals(expectedOutput, output,
-        '\nInput:\n${result.input}\n'
-        'Expected:\n$expectedOutput\n'
-        'Actual:\n$output\n');
-  }));
-}
\ No newline at end of file

tests/compiler/dart2js/backend_dart/opt_constprop_test.dart

diff --git a/tests/compiler/dart2js/backend_dart/opt_constprop_test.dart b/tests/compiler/dart2js/backend_dart/opt_constprop_test.dart
deleted file mode 100644
index 2ee3890..0000000
--- a/tests/compiler/dart2js/backend_dart/opt_constprop_test.dart
+++ /dev/null
@@ -1,528 +0,0 @@
-// Copyright (c) 2014, the Dart project authors.  Please see the AUTHORS file
-// for details. All rights reserved. Use of this source code is governed by a
-// BSD-style license that can be found in the LICENSE file.
-
-import 'dart:async';
-import '../mock_compiler.dart';
-import 'sexpr_unstringifier.dart';
-import 'package:async_helper/async_helper.dart';
-import "package:expect/expect.dart";
-import 'package:compiler/src/cps_ir/cps_ir_nodes_sexpr.dart';
-import 'package:compiler/src/cps_ir/optimizers.dart';
-import 'package:compiler/src/constant_system_dart.dart';
-
-// The tests in this file that ensure that sparse constant propagation on the
-// CPS IR works as expected.
-
-// CP1 represents the following incoming dart code:
-//
-//  int main() {
-//    int i = 1;
-//    int j;
-//    if (i == 1) {
-//      j = 2;
-//    } else {
-//      j = 3;
-//    }
-//    return j;
-//  }
-
-String CP1_IN = """
-(FunctionDefinition main () () return
-  (LetPrim (v0 (Constant (Int 1)))
-    (LetPrim (v1 (Constant (Int 1)))
-      (LetCont
-        ((k0 (v2)
-           (LetCont
-             ((k1 ()
-                (LetPrim (v3 (Constant (Int 2)))
-                  (InvokeContinuation return (v3))))
-              (k2 ()
-                (LetPrim (v4 (Constant (Int 3)))
-                  (InvokeContinuation return (v4)))))
-             (Branch (IsTrue v2) k1 k2))))
-        (InvokeMethod v0 == (v1) k0)))))
-""";
-String CP1_OUT = """
-(FunctionDefinition main () () return
-  (LetPrim (v0 (Constant (Int 1)))
-    (LetPrim (v1 (Constant (Int 1)))
-      (LetCont
-        ((k0 (v2)
-           (LetCont
-             ((k1 ()
-                (LetPrim (v3 (Constant (Int 2)))
-                  (InvokeContinuation return (v3))))
-              (k2 ()
-                (LetPrim (v4 (Constant (Int 3)))
-                  (InvokeContinuation return (v4)))))
-             (InvokeContinuation k1 ()))))
-        (LetPrim (v5 (Constant (Bool true)))
-          (InvokeContinuation k0 (v5)))))))
-""";
-
-// CP2 represents the following incoming dart code:
-//
-//  int main() {
-//    int i = 1;
-//    while (true) {
-//      if (false || false) {
-//        return i;
-//      }
-//      if (true && i == 1) {
-//        return i;
-//      }
-//    }
-//    return 42;
-//  }
-
-String CP2_IN = """
-(FunctionDefinition main () () return
-  (LetPrim (v0 (Constant (Int 1)))
-    (LetCont
-        ((rec k0 ()
-           (LetCont
-               ((k1 ()
-                  (LetPrim (v1 (Constant (Int 42)))
-                    (InvokeContinuation return (v1))))
-                (k2 ()
-                  (LetPrim (v2 (Constant (Bool false)))
-                    (LetCont
-                        ((k3 (v3)
-                           (LetCont
-                               ((k4 ()
-                                  (InvokeContinuation return (v0)))
-                                (k5 ()
-                                  (LetPrim (v4 (Constant (Bool true)))
-                                    (LetCont
-                                        ((k6 (v5)
-                                           (LetCont
-                                               ((k7 ()
-                                                  (InvokeContinuation return (v0)))
-                                                (k8 ()
-                                                  (InvokeContinuation rec k0 ())))
-                                             (Branch (IsTrue v5) k7 k8))))
-                                      (LetCont
-                                          ((k9 ()
-                                             (LetPrim (v6 (Constant (Int 1)))
-                                               (LetCont
-                                                   ((k10 (v7)
-                                                      (LetCont
-                                                          ((k11 ()
-                                                             (LetPrim (v8 (Constant (Bool true)))
-                                                               (InvokeContinuation k6 (v8))))
-                                                           (k12 ()
-                                                             (LetPrim (v9 (Constant (Bool false)))
-                                                               (InvokeContinuation k6 (v9)))))
-                                                        (Branch (IsTrue v7) k11 k12))))
-                                                 (InvokeMethod v0 == (v6) k10))))
-                                           (k13 ()
-                                             (LetPrim (v10 (Constant (Bool false)))
-                                               (InvokeContinuation k6 (v10)))))
-                                        (Branch (IsTrue v4) k9 k13))))))
-                             (Branch (IsTrue v3) k4 k5))))
-                      (LetCont
-                          ((k14 ()
-                             (LetPrim (v11 (Constant (Bool true)))
-                               (InvokeContinuation k3 (v11))))
-                           (k15 ()
-                             (LetPrim (v12 (Constant (Bool false)))
-                               (LetCont
-                                   ((k16 ()
-                                      (LetPrim (v13 (Constant (Bool true)))
-                                        (InvokeContinuation k3 (v13))))
-                                    (k17 ()
-                                      (LetPrim (v14 (Constant (Bool false)))
-                                        (InvokeContinuation k3 (v14)))))
-                                 (Branch (IsTrue v12) k16 k17)))))
-                        (Branch (IsTrue v2) k14 k15))))))
-             (LetPrim (v15 (Constant (Bool true)))
-               (Branch (IsTrue v15) k2 k1)))))
-      (InvokeContinuation k0 ()))))
-""";
-String CP2_OUT = """
-(FunctionDefinition main () () return
-  (LetPrim (v0 (Constant (Int 1)))
-    (LetCont
-        ((rec k0 ()
-           (LetCont
-               ((k1 ()
-                  (LetPrim (v1 (Constant (Int 42)))
-                    (InvokeContinuation return (v1))))
-                (k2 ()
-                  (LetPrim (v2 (Constant (Bool false)))
-                    (LetCont
-                        ((k3 (v3)
-                           (LetCont
-                               ((k4 ()
-                                  (InvokeContinuation return (v0)))
-                                (k5 ()
-                                  (LetPrim (v4 (Constant (Bool true)))
-                                    (LetCont
-                                        ((k6 (v5)
-                                           (LetCont
-                                               ((k7 ()
-                                                  (InvokeContinuation return (v0)))
-                                                (k8 ()
-                                                  (InvokeContinuation rec k0 ())))
-                                             (InvokeContinuation k7 ()))))
-                                      (LetCont
-                                          ((k9 ()
-                                             (LetPrim (v6 (Constant (Int 1)))
-                                               (LetCont
-                                                   ((k10 (v7)
-                                                      (LetCont
-                                                          ((k11 ()
-                                                             (LetPrim (v8 (Constant (Bool true)))
-                                                               (InvokeContinuation k6 (v8))))
-                                                           (k12 ()
-                                                             (LetPrim (v9 (Constant (Bool false)))
-                                                               (InvokeContinuation k6 (v9)))))
-                                                        (InvokeContinuation k11 ()))))
-                                                 (LetPrim (v10 (Constant (Bool true)))
-                                                   (InvokeContinuation k10 (v10))))))
-                                           (k13 ()
-                                             (LetPrim (v11 (Constant (Bool false)))
-                                               (InvokeContinuation k6 (v11)))))
-                                        (InvokeContinuation k9 ()))))))
-                             (InvokeContinuation k5 ()))))
-                      (LetCont
-                          ((k14 ()
-                             (LetPrim (v12 (Constant (Bool true)))
-                               (InvokeContinuation k3 (v12))))
-                           (k15 ()
-                             (LetPrim (v13 (Constant (Bool false)))
-                               (LetCont
-                                   ((k16 ()
-                                      (LetPrim (v14 (Constant (Bool true)))
-                                        (InvokeContinuation k3 (v14))))
-                                    (k17 ()
-                                      (LetPrim (v15 (Constant (Bool false)))
-                                        (InvokeContinuation k3 (v15)))))
-                                 (InvokeContinuation k17 ())))))
-                        (InvokeContinuation k15 ()))))))
-             (LetPrim (v16 (Constant (Bool true)))
-               (InvokeContinuation k2 ())))))
-      (InvokeContinuation k0 ()))))
-""";
-
-// CP3 represents the following incoming dart code:
-//
-//  int main() {
-//    int i = 1;
-//    i = f();
-//    if (i == 1) {
-//      return 42;
-//    }
-//    return i;
-//  }
-
-String CP3_IN = """
-(FunctionDefinition main () () return
-  (LetPrim (v0 (Constant (Int 1)))
-    (LetCont
-        ((k0 (v1)
-           (LetPrim (v2 (Constant (Int 1)))
-             (LetCont
-                 ((k1 (v3)
-                    (LetCont
-                        ((k2 ()
-                           (LetPrim (v4 (Constant (Int 42)))
-                             (InvokeContinuation return (v4))))
-                         (k3 ()
-                           (InvokeContinuation return (v1))))
-                      (Branch (IsTrue v3) k2 k3))))
-               (InvokeMethod v1 == (v2) k1)))))
-      (InvokeStatic f () k0))))
-""";
-String CP3_OUT = CP3_IN;
-
-// Addition.
-
-String CP4_IN = """
-(FunctionDefinition main () () return
-  (LetPrim (v0 (Constant (Int 1)))
-    (LetPrim (v1 (Constant (Int 2)))
-      (LetCont
-          ((k0 (v2)
-             (InvokeContinuation return (v2))))
-        (InvokeMethod v0 + (v1) k0)))))
-""";
-String CP4_OUT = """
-(FunctionDefinition main () () return
-  (LetPrim (v0 (Constant (Int 1)))
-    (LetPrim (v1 (Constant (Int 2)))
-      (LetCont
-          ((k0 (v2)
-             (InvokeContinuation return (v2))))
-        (LetPrim (v3 (Constant (Int 3)))
-          (InvokeContinuation k0 (v3)))))))
-""";
-
-// Array access operator (no optimization).
-
-String CP5_IN = """
-(FunctionDefinition main () () return
-  (LetPrim (v0 (Constant (Int 1)))
-    (LetPrim (v1 (Constant (Int 2)))
-      (LetCont
-          ((k0 (v2)
-             (InvokeContinuation return (v2))))
-        (InvokeMethod v0 [] (v1) k0)))))
-""";
-String CP5_OUT = CP5_IN;
-
-// Division by 0.
-
-String CP6_IN = """
-(FunctionDefinition main () () return
-  (LetPrim (v0 (Constant (Int 1)))
-    (LetPrim (v1 (Constant (Int 0)))
-      (LetCont
-          ((k0 (v2)
-             (InvokeContinuation return (v2))))
-        (InvokeMethod v0 / (v1) k0)))))
-""";
-String CP6_OUT = """
-(FunctionDefinition main () () return
-  (LetPrim (v0 (Constant (Int 1)))
-    (LetPrim (v1 (Constant (Int 0)))
-      (LetCont
-          ((k0 (v2)
-             (InvokeContinuation return (v2))))
-        (LetPrim (v3 (Constant (Double Infinity)))
-          (InvokeContinuation k0 (v3)))))))
-""";
-
-// Concatenate strings.
-
-String CP7_IN = """
-(FunctionDefinition main () () return
-  (LetPrim (v0 (Constant (String "b")))
-    (LetPrim (v1 (Constant (String "d")))
-      (LetPrim (v2 (Constant (String "a")))
-        (LetPrim (v3 (Constant (String "c")))
-          (LetPrim (v4 (Constant (String "")))
-            (LetCont
-                ((k0 (v5)
-                   (LetCont
-                       ((k1 (v6)
-                          (InvokeContinuation return (v6))))
-                     (InvokeMethod v5 length () k1))))
-              (ConcatenateStrings (v2 v0 v3 v1 v4) k0))))))))
-""";
-String CP7_OUT = """
-(FunctionDefinition main () () return
-  (LetPrim (v0 (Constant (String "b")))
-    (LetPrim (v1 (Constant (String "d")))
-      (LetPrim (v2 (Constant (String "a")))
-        (LetPrim (v3 (Constant (String "c")))
-          (LetPrim (v4 (Constant (String "")))
-            (LetCont
-                ((k0 (v5)
-                   (LetCont
-                       ((k1 (v6)
-                          (InvokeContinuation return (v6))))
-                     (InvokeMethod v5 length () k1))))
-              (LetPrim (v7 (Constant (String "abcd")))
-                (InvokeContinuation k0 (v7))))))))))
-""";
-
-// TODO(jgruber): We can't test is-check optimization because the unstringifier
-// does not recreate accurate types for the TypeOperator node.
-
-// Simple branch removal.
-
-String CP8_IN = """
-(FunctionDefinition main () () return
-  (LetPrim (v0 (Constant (Int 1)))
-    (LetPrim (v1 (Constant (Int 1)))
-      (LetCont
-          ((k0 (v2)
-             (LetCont
-                 ((k1 ()
-                    (LetPrim (v3 (Constant (Int 42)))
-                      (InvokeContinuation return (v3))))
-                  (k2 ()
-                    (InvokeContinuation return (v0))))
-               (Branch (IsTrue v2) k1 k2))))
-        (InvokeMethod v0 == (v1) k0)))))
-""";
-String CP8_OUT = """
-(FunctionDefinition main () () return
-  (LetPrim (v0 (Constant (Int 1)))
-    (LetPrim (v1 (Constant (Int 1)))
-      (LetCont
-          ((k0 (v2)
-             (LetCont
-                 ((k1 ()
-                    (LetPrim (v3 (Constant (Int 42)))
-                      (InvokeContinuation return (v3))))
-                  (k2 ()
-                    (InvokeContinuation return (v0))))
-               (InvokeContinuation k1 ()))))
-        (LetPrim (v4 (Constant (Bool true)))
-          (InvokeContinuation k0 (v4)))))))
-""";
-
-// While loop.
-
-String CP9_IN = """
-(FunctionDefinition main () () return
-  (LetPrim (v0 (Constant (Int 1)))
-    (LetCont
-        ((rec k0 (v1)
-           (LetCont
-               ((k1 ()
-                  (InvokeContinuation return (v1)))
-                (k2 ()
-                  (LetPrim (v2 (Constant (Int 1)))
-                    (LetCont
-                        ((k3 (v3)
-                           (LetCont
-                               ((k4 (v4)
-                                  (LetCont
-                                      ((k5 ()
-                                         (LetPrim (v5 (Constant (Int 42)))
-                                           (InvokeContinuation return (v5))))
-                                       (k6 ()
-                                         (LetPrim (v6 (Constant (Int 1)))
-                                           (LetCont
-                                               ((k7 (v7)
-                                                  (InvokeContinuation rec k0 (v7))))
-                                             (InvokeMethod v1 + (v6) k7)))))
-                                    (Branch (IsTrue v4) k5 k6))))
-                             (LetCont
-                                 ((k8 ()
-                                    (LetPrim (v8 (Constant (Bool false)))
-                                      (InvokeContinuation k4 (v8))))
-                                  (k9 ()
-                                    (LetPrim (v9 (Constant (Bool true)))
-                                      (InvokeContinuation k4 (v9)))))
-                               (Branch (IsTrue v3) k8 k9)))))
-                      (InvokeMethod v1 == (v2) k3)))))
-             (LetPrim (v10 (Constant (Bool true)))
-               (Branch (IsTrue v10) k2 k1)))))
-      (InvokeContinuation k0 (v0)))))
-""";
-String CP9_OUT = """
-(FunctionDefinition main () () return
-  (LetPrim (v0 (Constant (Int 1)))
-    (LetCont
-        ((rec k0 (v1)
-           (LetCont
-               ((k1 ()
-                  (InvokeContinuation return (v1)))
-                (k2 ()
-                  (LetPrim (v2 (Constant (Int 1)))
-                    (LetCont
-                        ((k3 (v3)
-                           (LetCont
-                              ((k4 (v4)
-                                 (LetCont
-                                     ((k5 ()
-                                        (LetPrim (v5 (Constant (Int 42)))
-                                          (InvokeContinuation return (v5))))
-                                      (k6 ()
-                                        (LetPrim (v6 (Constant (Int 1)))
-                                          (LetCont
-                                              ((k7 (v7)
-                                                 (InvokeContinuation rec k0 (v7))))
-                                            (InvokeMethod v1 + (v6) k7)))))
-                                   (Branch (IsTrue v4) k5 k6))))
-                             (LetCont
-                                 ((k8 ()
-                                    (LetPrim (v8 (Constant (Bool false)))
-                                      (InvokeContinuation k4 (v8))))
-                                  (k9 ()
-                                    (LetPrim (v9 (Constant (Bool true)))
-                                      (InvokeContinuation k4 (v9)))))
-                               (Branch (IsTrue v3) k8 k9)))))
-                      (InvokeMethod v1 == (v2) k3)))))
-             (LetPrim (v10 (Constant (Bool true)))
-               (InvokeContinuation k2 ())))))
-      (InvokeContinuation k0 (v0)))))
-""";
-
-// While loop, from:
-//
-//  int main() {
-//    for (int i = 0; i < 2; i++) {
-//      print(42 + i);
-//    }
-//  }
-
-String CP10_IN = """
-(FunctionDefinition main () () return
-  (LetPrim (v0 (Constant (Int 0)))
-    (LetCont
-        ((rec k0 (v1)
-           (LetCont
-               ((k1 ()
-                  (LetPrim (v2 (Constant (Null)))
-                    (InvokeContinuation return (v2))))
-                (k2 ()
-                  (LetPrim (v3 (Constant (Int 42)))
-                    (LetCont
-                        ((k3 (v4)
-                           (LetCont
-                               ((k4 (v5)
-                                  (LetPrim (v6 (Constant (Int 1)))
-                                    (LetCont
-                                        ((k5 (v7)
-                                           (InvokeContinuation rec k0 (v7))))
-                                      (InvokeMethod v1 + (v6) k5)))))
-                             (InvokeStatic print (v4) k4))))
-                      (InvokeMethod v3 + (v1) k3)))))
-             (LetPrim (v8 (Constant (Int 2)))
-               (LetCont
-                   ((k6 (v9)
-                      (Branch (IsTrue v9) k2 k1)))
-                 (InvokeMethod v1 < (v8) k6))))))
-      (InvokeContinuation k0 (v0)))))
-""";
-String CP10_OUT = CP10_IN;
-
-/// Normalizes whitespace by replacing all whitespace sequences by a single
-/// space and trimming leading and trailing whitespace.
-String normalizeSExpr(String input) {
-  return input.replaceAll(new RegExp(r'[ \n\t]+'), ' ').trim();
-}
-
-/// Parses the given input IR, runs an optimization pass over it, and compares
-/// the stringification of the result against the expected output.
-Future testConstantPropagator(String input, String expectedOutput) {
-  final compiler = new MockCompiler.internal(
-      emitJavaScript: false,
-      enableMinification: false);
-  return compiler.init().then((_) {
-    final unstringifier = new SExpressionUnstringifier();
-    final stringifier   = new SExpressionStringifier();
-    final optimizer     = new TypePropagator(
-        compiler.types,
-        DART_CONSTANT_SYSTEM,
-        new UnitTypeSystem(),
-        compiler.internalError);
-
-    final f = unstringifier.unstringify(input);
-    optimizer.rewrite(f);
-
-    String expected = normalizeSExpr(expectedOutput);
-    String actual   = normalizeSExpr(stringifier.visit(f));
-
-    Expect.equals(expected, actual);
-  });
-}
-
-void main() {
-  asyncTest(() => testConstantPropagator(CP1_IN, CP1_OUT));
-  asyncTest(() => testConstantPropagator(CP2_IN, CP2_OUT));
-  asyncTest(() => testConstantPropagator(CP3_IN, CP3_OUT));
-  asyncTest(() => testConstantPropagator(CP4_IN, CP4_OUT));
-  asyncTest(() => testConstantPropagator(CP5_IN, CP5_OUT));
-  asyncTest(() => testConstantPropagator(CP6_IN, CP6_OUT));
-  asyncTest(() => testConstantPropagator(CP7_IN, CP7_OUT));
-  asyncTest(() => testConstantPropagator(CP8_IN, CP8_OUT));
-  asyncTest(() => testConstantPropagator(CP9_IN, CP9_OUT));
-  asyncTest(() => testConstantPropagator(CP10_IN, CP10_OUT));
-}

tests/compiler/dart2js/backend_dart/opt_cyclic_redundant_phi_test.dart

diff --git a/tests/compiler/dart2js/backend_dart/opt_cyclic_redundant_phi_test.dart b/tests/compiler/dart2js/backend_dart/opt_cyclic_redundant_phi_test.dart
deleted file mode 100644
index 7132824..0000000
--- a/tests/compiler/dart2js/backend_dart/opt_cyclic_redundant_phi_test.dart
+++ /dev/null
@@ -1,97 +0,0 @@
-// Copyright (c) 2014, the Dart project authors.  Please see the AUTHORS file
-// for details. All rights reserved. Use of this source code is governed by a
-// BSD-style license that can be found in the LICENSE file.
-
-import 'opt_redundant_phi_test.dart';
-
-// The 'cyclic deps' IR tests removal of redundant phis with cyclic
-// dependencies.
-//
-//  void main() {
-//    var x = 0;
-//    var y = x;
-//    for (int i = 0; i < 10; i++) {
-//      if (i == -1) x = y;
-//      if (i == -1) y = x;
-//    }
-//    print(x);
-//    print(y);
-//  }
-
-String CYCLIC_DEPS_IN = """
-(FunctionDefinition main (return) (LetPrim v0 (Constant 0))
-  (LetPrim v1 (Constant 0))
-  (LetCont* (k0 v2 v3 v4)
-    (LetCont (k1)
-      (LetCont (k2 v5)
-        (LetCont (k3 v6) (LetPrim v7 (Constant null))
-          (InvokeContinuation return v7))
-        (InvokeStatic print v3 k3))
-      (InvokeStatic print v2 k2))
-    (LetCont (k4) (LetPrim v8 (Constant 1))
-      (LetCont (k5 v9)
-        (LetCont (k6 v10)
-          (LetCont (k7 v11) (LetPrim v12 (Constant 1))
-            (LetCont (k8 v13)
-              (LetCont (k9 v14)
-                (LetCont (k10 v15)
-                  (LetPrim v16 (Constant 1))
-                  (LetCont (k11 v17)
-                    (InvokeContinuation* k0 v11 v15 v17))
-                  (InvokeMethod v4 + v16 k11))
-                (LetCont (k12) (InvokeContinuation k10 v11))
-                (LetCont (k13) (InvokeContinuation k10 v3))
-                (Branch (IsTrue v14) k12 k13))
-              (InvokeMethod v4 == v13 k9))
-            (InvokeMethod v12 unary- k8))
-          (LetCont (k14) (InvokeContinuation k7 v3))
-          (LetCont (k15) (InvokeContinuation k7 v2))
-          (Branch (IsTrue v10) k14 k15))
-        (InvokeMethod v4 == v9 k6))
-      (InvokeMethod v8 unary- k5))
-    (LetPrim v18 (Constant 10))
-    (LetCont (k16 v19) (Branch (IsTrue v19) k4 k1))
-    (InvokeMethod v4 < v18 k16))
-  (InvokeContinuation k0 v0 v0 v1))
-""";
-
-String CYCLIC_DEPS_OUT = """
-(FunctionDefinition main (return) (LetPrim v0 (Constant 0))
-  (LetPrim v1 (Constant 0))
-  (LetCont* (k0 v2)
-    (LetCont (k1)
-      (LetCont (k2 v3)
-        (LetCont (k3 v4) (LetPrim v5 (Constant null))
-          (InvokeContinuation return v5))
-        (InvokeStatic print v0 k3))
-      (InvokeStatic print v0 k2))
-    (LetCont (k4) (LetPrim v6 (Constant 1))
-      (LetCont (k5 v7)
-        (LetCont (k6 v8)
-          (LetCont (k7) (LetPrim v9 (Constant 1))
-            (LetCont (k8 v10)
-              (LetCont (k9 v11)
-                (LetCont (k10)
-                  (LetPrim v12 (Constant 1))
-                  (LetCont (k11 v13)
-                    (InvokeContinuation* k0 v13))
-                  (InvokeMethod v2 + v12 k11))
-                (LetCont (k12) (InvokeContinuation k10))
-                (LetCont (k13) (InvokeContinuation k10))
-                (Branch (IsTrue v11) k12 k13))
-              (InvokeMethod v2 == v10 k9))
-            (InvokeMethod v9 unary- k8))
-          (LetCont (k14) (InvokeContinuation k7))
-          (LetCont (k15) (InvokeContinuation k7))
-          (Branch (IsTrue v8) k14 k15))
-        (InvokeMethod v2 == v7 k6))
-      (InvokeMethod v6 unary- k5))
-    (LetPrim v14 (Constant 10))
-    (LetCont (k16 v15) (Branch (IsTrue v15) k4 k1))
-    (InvokeMethod v2 < v14 k16))
-  (InvokeContinuation k0 v1))
-""";
-
-void main() {
-  testRedundantPhi(CYCLIC_DEPS_IN, CYCLIC_DEPS_OUT);
-}

tests/compiler/dart2js/backend_dart/opt_redundant_phi_test.dart

diff --git a/tests/compiler/dart2js/backend_dart/opt_redundant_phi_test.dart b/tests/compiler/dart2js/backend_dart/opt_redundant_phi_test.dart
deleted file mode 100644
index 50422e2..0000000
--- a/tests/compiler/dart2js/backend_dart/opt_redundant_phi_test.dart
+++ /dev/null
@@ -1,309 +0,0 @@
-// Copyright (c) 2014, the Dart project authors.  Please see the AUTHORS file
-// for details. All rights reserved. Use of this source code is governed by a
-// BSD-style license that can be found in the LICENSE file.
-
-import 'sexpr_unstringifier.dart';
-import "package:expect/expect.dart";
-import 'package:compiler/src/cps_ir/cps_ir_nodes.dart';
-import 'package:compiler/src/cps_ir/cps_ir_nodes_sexpr.dart';
-import 'package:compiler/src/cps_ir/optimizers.dart';
-
-// The 'read in loop' IR tests the most basic case of redundant phi removal
-// and represents the following source code:
-//
-// void main() {
-//   int j = 42;
-//   for (int i = 0; i < 2; i++) {
-//     print(j.toString());
-//   }
-// }
-
-String READ_IN_LOOP_IN = """
-(FunctionDefinition main () () return
-  (LetPrim (v0 (Constant (Int 42)))
-    (LetPrim (v1 (Constant (Int 0)))
-      (LetCont
-          ((rec k0 (v2 v3)
-             (LetCont
-                 ((k1 ()
-                    (LetPrim (v4 (Constant (Null)))
-                      (InvokeContinuation return (v4))))
-                  (k2 ()
-                    (LetCont
-                        ((k3 (v5)
-                           (LetCont
-                               ((k4 (v6)
-                                  (LetPrim (v7 (Constant (Int 1)))
-                                    (LetCont
-                                        ((k5 (v8)
-                                           (InvokeContinuation rec k0 (v2 v8))))
-                                      (InvokeMethod v3 + (v7) k5)))))
-                             (InvokeStatic print (v5) k4))))
-                      (InvokeMethod v2 toString () k3))))
-               (LetPrim (v9 (Constant (Int 2)))
-                 (LetCont
-                     ((k6 (v10)
-                        (Branch (IsTrue v10) k2 k1)))
-                   (InvokeMethod v3 < (v9) k6))))))
-        (InvokeContinuation k0 (v0 v1))))))
-""";
-
-String READ_IN_LOOP_OUT = """
-(FunctionDefinition main () () return
-  (LetPrim (v0 (Constant (Int 42)))
-    (LetPrim (v1 (Constant (Int 0)))
-      (LetCont
-          ((rec k0 (v2)
-             (LetCont
-                 ((k1 ()
-                    (LetPrim (v3 (Constant (Null)))
-                      (InvokeContinuation return (v3))))
-                  (k2 ()
-                    (LetCont
-                        ((k3 (v4)
-                           (LetCont
-                               ((k4 (v5)
-                                  (LetPrim (v6 (Constant (Int 1)))
-                                    (LetCont
-                                        ((k5 (v7)
-                                           (InvokeContinuation rec k0 (v7))))
-                                      (InvokeMethod v2 + (v6) k5)))))
-                             (InvokeStatic print (v4) k4))))
-                      (InvokeMethod v0 toString () k3))))
-               (LetPrim (v8 (Constant (Int 2)))
-                 (LetCont
-                     ((k6 (v9)
-                        (Branch (IsTrue v9) k2 k1)))
-                   (InvokeMethod v2 < (v8) k6))))))
-        (InvokeContinuation k0 (v1))))))
-""";
-
-// The 'inner loop' IR represents the following source code:
-//
-// void main() {
-//   int j = 42;
-//   for (int i = 0; i < 2; i++) {
-//     for (int k = 0; k < 2; k++) {
-//       print(i.toString());
-//     }
-//   }
-//   print(j.toString());
-// }
-//
-// This test case ensures that iterative optimization works: first, v8 and v9
-// are removed from k5, and only then can k0 be optimized as well.
-
-const String INNER_LOOP_IN = """
-(FunctionDefinition main () () return
-  (LetPrim (v0 (Constant (Int 42)))
-    (LetPrim (v1 (Constant (Int 0)))
-      (LetCont
-          ((rec k0 (v2 v3)
-             (LetCont
-                 ((k1 ()
-                    (LetCont
-                        ((k2 (v4)
-                           (LetCont
-                               ((k3 (v5)
-                                  (LetPrim (v6 (Constant (Null)))
-                                    (InvokeContinuation return (v6)))))
-                             (InvokeStatic print (v4) k3))))
-                      (InvokeMethod v2 toString () k2)))
-                  (k4 ()
-                    (LetPrim (v7 (Constant (Int 0)))
-                      (LetCont
-                          ((rec k5 (v8 v9 v10)
-                             (LetCont
-                                 ((k6 ()
-                                    (LetPrim (v11 (Constant (Int 1)))
-                                      (LetCont
-                                          ((k7 (v12)
-                                             (InvokeContinuation rec k0 (v8 v12))))
-                                        (InvokeMethod v9 + (v11) k7))))
-                                  (k8 ()
-                                    (LetCont
-                                        ((k9 (v13)
-                                           (LetCont
-                                               ((k10 (v14)
-                                                  (LetPrim (v15 (Constant (Int 1)))
-                                                    (LetCont
-                                                        ((k11 (v16)
-                                                           (InvokeContinuation rec k5 (v8 v9 v16))))
-                                                      (InvokeMethod v10 + (v15) k11)))))
-                                             (InvokeStatic print (v13) k10))))
-                                      (InvokeMethod v9 toString () k9))))
-                               (LetPrim (v17 (Constant (Int 2)))
-                                 (LetCont
-                                     ((k12 (v18)
-                                        (Branch (IsTrue v18) k8 k6)))
-                                   (InvokeMethod v10 < (v17) k12))))))
-                        (InvokeContinuation k5 (v2 v3 v7))))))
-               (LetPrim (v19 (Constant (Int 2)))
-                 (LetCont
-                     ((k13 (v20)
-                        (Branch (IsTrue v20) k4 k1)))
-                   (InvokeMethod v3 < (v19) k13))))))
-        (InvokeContinuation k0 (v0 v1))))))
-""";
-
-const String INNER_LOOP_OUT = """
-(FunctionDefinition main () () return
-  (LetPrim (v0 (Constant (Int 42)))
-    (LetPrim (v1 (Constant (Int 0)))
-      (LetCont
-          ((rec k0 (v2)
-             (LetCont
-                 ((k1 ()
-                    (LetCont
-                        ((k2 (v3)
-                           (LetCont
-                               ((k3 (v4)
-                                  (LetPrim (v5 (Constant (Null)))
-                                    (InvokeContinuation return (v5)))))
-                             (InvokeStatic print (v3) k3))))
-                      (InvokeMethod v0 toString () k2)))
-                  (k4 ()
-                    (LetPrim (v6 (Constant (Int 0)))
-                      (LetCont
-                          ((rec k5 (v7)
-                             (LetCont
-                                 ((k6 ()
-                                    (LetPrim (v8 (Constant (Int 1)))
-                                      (LetCont
-                                          ((k7 (v9)
-                                             (InvokeContinuation rec k0 (v9))))
-                                        (InvokeMethod v2 + (v8) k7))))
-                                  (k8 ()
-                                    (LetCont
-                                        ((k9 (v10)
-                                           (LetCont
-                                               ((k10 (v11)
-                                                  (LetPrim (v12 (Constant (Int 1)))
-                                                    (LetCont
-                                                        ((k11 (v13)
-                                                           (InvokeContinuation rec k5 (v13))))
-                                                      (InvokeMethod v7 + (v12) k11)))))
-                                             (InvokeStatic print (v10) k10))))
-                                      (InvokeMethod v2 toString () k9))))
-                               (LetPrim (v14 (Constant (Int 2)))
-                                 (LetCont
-                                     ((k12 (v15)
-                                        (Branch (IsTrue v15) k8 k6)))
-                                   (InvokeMethod v7 < (v14) k12))))))
-                        (InvokeContinuation k5 (v6))))))
-               (LetPrim (v16 (Constant (Int 2)))
-                 (LetCont
-                     ((k13 (v17)
-                        (Branch (IsTrue v17) k4 k1)))
-                   (InvokeMethod v2 < (v16) k13))))))
-        (InvokeContinuation k0 (v1))))))
-""";
-
-// There are no redundant phis in the 'basic loop' IR, and this test ensures
-// simply that the optimization does not alter the IR. It represents the
-// following program:
-//
-// void main() {
-//   for (int i = 0; i < 2; i++) {
-//     print(i.toString());
-//   }
-// }
-
-String BASIC_LOOP_IN = """
-(FunctionDefinition main () () return
-  (LetPrim (v0 (Constant (Int 0)))
-    (LetCont
-       ((rec k0 (v1)
-          (LetCont
-              ((k1 ()
-                 (LetPrim (v2 (Constant (Null)))
-                   (InvokeContinuation return (v2))))
-               (k2 ()
-                 (LetCont
-                     ((k3 (v3)
-                        (LetCont
-                            ((k4 (v4)
-                               (LetPrim (v5 (Constant (Int 1)))
-                                 (LetCont
-                                     ((k5 (v6)
-                                        (InvokeContinuation rec k0 (v6))))
-                                   (InvokeMethod v1 + (v5) k5)))))
-                          (InvokeStatic print (v3) k4))))
-                   (InvokeMethod v1 toString () k3))))
-            (LetPrim (v7 (Constant (Int 2)))
-              (LetCont
-                  ((k6 (v8)
-                     (Branch (IsTrue v8) k2 k1)))
-                (InvokeMethod v1 < (v7) k6))))))
-      (InvokeContinuation k0 (v0)))))
-""";
-
-String BASIC_LOOP_OUT = BASIC_LOOP_IN;
-
-// Ensures that proper scoping is preserved, i.e. that the optimized
-// continuation body does reference out of scope primitives.
-// IR written by hand since this case is currently not being generated.
-
-String SCOPING_IN = """
-(FunctionDefinition main () () return
-  (LetCont
-      ((k0 (v1)
-         (InvokeStatic print (v1) return)))
-    (LetPrim (v0 (Constant (Int 0)))
-      (LetPrim (v2 (Constant (Null)))
-        (InvokeContinuation k0 (v0))))))
-""";
-
-String SCOPING_OUT = """
-(FunctionDefinition main () () return
-  (LetPrim (v0 (Constant (Int 0)))
-    (LetCont
-        ((k0 ()
-           (InvokeStatic print (v0) return)))
-      (LetPrim (v1 (Constant (Null)))
-        (InvokeContinuation k0 ())))))
-""";
-
-// Ensures that continuations which are never invoked are not optimized.
-// IR written by hand.
-
-String NEVER_INVOKED_IN = """
-(FunctionDefinition main () () return
-  (LetPrim (v0 (Constant (Int 0)))
-    (LetCont
-        ((k0 (v1)
-           (InvokeStatic print (v1) return)))
-      (InvokeContinuation return (v0)))))
-""";
-
-String NEVER_INVOKED_OUT = NEVER_INVOKED_IN;
-
-/// Normalizes whitespace by replacing all whitespace sequences by a single
-/// space and trimming leading and trailing whitespace.
-String normalizeSExpr(String input) {
-  return input.replaceAll(new RegExp(r'[ \n\t]+'), ' ').trim();
-}
-
-/// Parses the given input IR, runs a redundant phi pass over it, and compares
-/// the stringification of the result against the expected output.
-void testRedundantPhi(String input, String expectedOutput) {
-  final unstringifier = new SExpressionUnstringifier();
-  final stringifier   = new SExpressionStringifier();
-  final optimizer     = new RedundantPhiEliminator();
-
-  FunctionDefinition f = unstringifier.unstringify(input);
-  optimizer.rewrite(f);
-
-  String expected = normalizeSExpr(expectedOutput);
-  String actual   = normalizeSExpr(stringifier.visit(f));
-
-  Expect.equals(expected, actual, "Actual:\n$actual");
-}
-
-void main() {
-  testRedundantPhi(READ_IN_LOOP_IN, READ_IN_LOOP_OUT);
-  testRedundantPhi(INNER_LOOP_IN, INNER_LOOP_OUT);
-  testRedundantPhi(BASIC_LOOP_IN, BASIC_LOOP_OUT);
-  testRedundantPhi(SCOPING_IN, SCOPING_OUT);
-  testRedundantPhi(NEVER_INVOKED_IN, NEVER_INVOKED_OUT);
-}

tests/compiler/dart2js/backend_dart/opt_shrinking_test.dart

diff --git a/tests/compiler/dart2js/backend_dart/opt_shrinking_test.dart b/tests/compiler/dart2js/backend_dart/opt_shrinking_test.dart
deleted file mode 100644
index 906a6cf..0000000
--- a/tests/compiler/dart2js/backend_dart/opt_shrinking_test.dart
+++ /dev/null
@@ -1,331 +0,0 @@
-// Copyright (c) 2014, the Dart project authors.  Please see the AUTHORS file
-// for details. All rights reserved. Use of this source code is governed by a
-// BSD-style license that can be found in the LICENSE file.
-
-import 'sexpr_unstringifier.dart';
-import "package:expect/expect.dart";
-import 'package:compiler/src/cps_ir/cps_ir_nodes.dart';
-import 'package:compiler/src/cps_ir/cps_ir_nodes_sexpr.dart';
-import 'package:compiler/src/cps_ir/optimizers.dart';
-
-// The tests in this file that ensure shrinking reductions work as expected.
-// Reductions and their corresponding names are taken from
-// 'Compiling with Continuations, Continued' by Andrew Kennedy.
-
-// Basic dead-val: letprim x = V in K -> K (x not free in K).
-//
-//  int main() {
-//    int i = 42;
-//    return 0;
-//  }
-
-String DEAD_VAL_IN = """
-(FunctionDefinition main () () return
-  (LetPrim (v0 (Constant (Int 42)))
-    (LetPrim (v1 (Constant (Int 0)))
-      (InvokeContinuation return (v1)))))
-""";
-String DEAD_VAL_OUT = """
-(FunctionDefinition main () () return
-  (LetPrim (v0 (Constant (Int 0)))
-    (InvokeContinuation return (v0))))
-""";
-
-// Iterative dead-val. No optimizations possible since the continuation to
-// InvokeMethod must have one argument, even if it is unused.
-//
-//  int main() {
-//    int i = 42;
-//    int j = i + 1;
-//    return 0;
-//  }
-
-String ITERATIVE_DEAD_VAL1_IN = """
-(FunctionDefinition main () () return
-  (LetPrim (v0 (Constant (Int 42)))
-    (LetPrim (v1 (Constant (Int 1)))
-      (LetCont ((k0 (v2)
-                  (LetPrim (v3 (Constant (Int 0)))
-                    (InvokeContinuation return (v3)))))
-        (InvokeMethod v0 + (v1) k0)))))
-""";
-String ITERATIVE_DEAD_VAL1_OUT = ITERATIVE_DEAD_VAL1_IN;
-
-// Iterative dead-val. IR written by hand.
-
-String ITERATIVE_DEAD_VAL2_IN = """
-(FunctionDefinition main () () return
-  (LetPrim (v0 (Constant (Int 42)))
-    (LetPrim (v1
-        (CreateFunction
-          (FunctionDefinition f () (i) return
-            (InvokeContinuation return (v0)))))
-      (LetPrim (v2 (Constant (Int 0)))
-        (InvokeContinuation return (v2))))))
-""";
-String ITERATIVE_DEAD_VAL2_OUT = """
-(FunctionDefinition main () () return
-  (LetPrim (v0 (Constant (Int 0)))
-    (InvokeContinuation return (v0))))
-""";
-
-// Basic dead-cont: letcont k x = L in K -> K (k not free in K).
-// IR written by hand.
-
-String DEAD_CONT_IN = """
-(FunctionDefinition main () () return
-  (LetPrim (v4 (Constant (Int 0)))
-    (LetCont ((k0 (v0)
-                (InvokeConstructor List () return)))
-      (LetCont ((k1 (v1)
-                  (LetCont ((k2 (v2)
-                              (LetPrim (v3 (Constant (Int 0)))
-                                (InvokeContinuation return (v3)))))
-                    (InvokeStatic print (v4) k2))))
-        (InvokeStatic print (v4) k1)))))
-""";
-String DEAD_CONT_OUT = """
-(FunctionDefinition main () () return
-  (LetPrim (v0 (Constant (Int 0)))
-    (LetCont ((k0 (v1)
-                (LetCont ((k1 (v2)
-                            (LetPrim (v3 (Constant (Int 0)))
-                              (InvokeContinuation return (v3)))))
-                  (InvokeStatic print (v0) k1))))
-      (InvokeStatic print (v0) k0))))
-""";
-
-// Iterative dead-cont. IR written by hand.
-
-String ITERATIVE_DEAD_CONT_IN = """
-(FunctionDefinition main () () return
-  (LetPrim (v4 (Constant (Int 0)))
-    (LetCont ((k0 (v0)
-                (InvokeConstructor List () return)))
-      (LetCont ((k3 (v5)
-                  (InvokeContinuation k0 (v5))))
-        (LetCont ((k1 (v1)
-                    (LetCont ((k2 (v2)
-                                (LetPrim (v3 (Constant (Int 0)))
-                                  (InvokeContinuation return (v3)))))
-                      (InvokeStatic print (v4) k2))))
-          (InvokeStatic print (v4) k1))))))
-""";
-String ITERATIVE_DEAD_CONT_OUT = """
-(FunctionDefinition main () () return
-  (LetPrim (v0 (Constant (Int 0)))
-    (LetCont ((k0 (v1)
-                (LetCont ((k1 (v2)
-                            (LetPrim (v3 (Constant (Int 0)))
-                              (InvokeContinuation return (v3)))))
-                  (InvokeStatic print (v0) k1))))
-      (InvokeStatic print (v0) k0))))
-""";
-
-// Beta-cont-lin: letcont k x = K in C[k y] -> C[K[y/x]] (k not free in C).
-// IR written by hand.
-
-String BETA_CONT_LIN_IN = """
-(FunctionDefinition main () () return
-  (LetCont ((k0 (v0)
-              (LetCont ((k1 (v1)
-                          (LetCont ((k2 (v2)
-                                      (LetPrim (v3 (Constant (Int 0)))
-                                        (InvokeContinuation return (v3)))))
-                            (InvokeStatic print (v0) k2))))
-                (InvokeStatic print (v0) k1))))
-    (LetPrim (v4 (Constant (Int 0)))
-      (InvokeContinuation k0 (v4)))))
-""";
-String BETA_CONT_LIN_OUT = """
-(FunctionDefinition main () () return
-  (LetPrim (v0 (Constant (Int 0)))
-    (LetCont ((k0 (v1)
-                (LetCont ((k1 (v2)
-                            (LetPrim (v3 (Constant (Int 0)))
-                              (InvokeContinuation return (v3)))))
-                  (InvokeStatic print (v0) k1))))
-      (InvokeStatic print (v0) k0))))
-""";
-
-// Beta-cont-lin with recursive continuation. IR written by hand.
-
-String RECURSIVE_BETA_CONT_LIN_IN = """
-(FunctionDefinition main () () return
-  (LetCont ((rec k0 (v0)
-              (InvokeContinuation rec k0 (v0))))
-    (LetPrim (v1 (Constant (Int 0)))
-      (InvokeContinuation k0 (v1)))))
-""";
-String RECURSIVE_BETA_CONT_LIN_OUT = RECURSIVE_BETA_CONT_LIN_IN;
-
-// Beta-cont-lin used inside body. IR written by hand.
-
-String USED_BETA_CONT_LIN_IN = """
-(FunctionDefinition main () () return
-  (LetPrim (v0 (Constant (Int 0)))
-    (LetCont ((k0 (v1)
-                (LetCont ((k1 (v2)
-                            (LetCont ((k2 (v3)
-                                        (LetPrim (v4 (Constant (Int 0)))
-                                          (InvokeContinuation return (v4)))))
-                              (InvokeStatic print (v1) k2))))
-                  (InvokeStatic print (v1) k1))))
-      (LetPrim (v5
-                 (CreateFunction
-                   (FunctionDefinition f () () return
-                     (InvokeContinuation return (v1)))))
-        (InvokeContinuation k0 (v0))))))
-""";
-String USED_BETA_CONT_LIN_OUT = """
-(FunctionDefinition main () () return
-  (LetPrim (v0 (Constant (Int 0)))
-    (LetCont ((k0 (v1)
-                (LetCont ((k1 (v2)
-                            (LetPrim (v3 (Constant (Int 0)))
-                              (InvokeContinuation return (v3)))))
-                  (InvokeStatic print (v0) k1))))
-      (InvokeStatic print (v0) k0))))
-""";
-
-// Eta-cont: letcont k x = j x in K -> K[j/k].
-// IR written by hand.
-//
-// This test is incorrectly named: with the current implementation, there is no
-// eta reduction.  Instead, dead-parameter, beta-cont-lin, and dead-val
-// reductions are performed, which in turn creates a second beta-cont-lin
-// reduction.
-//
-// TODO(kmillikin): To test continuation eta reduction, use eta redexes that are
-// not overlapping beta redexes.
-String ETA_CONT_IN = """
-(FunctionDefinition main () () return
-  (LetPrim (v0 (Constant (Int 0)))
-    (LetCont ((rec k0 (v1)
-                (InvokeContinuation return (v0))))
-      (LetCont ((k1 (v2)
-                  (InvokeContinuation k0 (v2))))
-        (LetPrim (v3
-                   (CreateFunction
-                     (FunctionDefinition f () () return
-                       (InvokeContinuation k0 (v0)))))
-          (InvokeContinuation k1 (v0)))))))
-""";
-String ETA_CONT_OUT = """
-(FunctionDefinition main () () return
-  (LetPrim (v0 (Constant (Int 0)))
-    (InvokeContinuation return (v0))))
-""";
-
-// Dead-parameter:
-// letcont k x = E0 in E1 -> letcont k () = E0 in E1,
-//    if x does not occur free in E0.
-
-// Parameter v1 is unused in k0.
-String DEAD_PARAMETER_IN = """
-(FunctionDefinition main () (x) return
-  (LetCont ((k0 (v0 v1 v2)
-              (InvokeStatic foo (v0 v2) return)))
-    (LetCont ((k1 ()
-                (LetPrim (v3 (Constant (Int 0)))
-                  (LetPrim (v4 (Constant (Int 1)))
-                    (LetPrim (v5 (Constant (Int 2)))
-                      (InvokeContinuation k0 (v3 v4 v5))))))
-              (k2 ()
-                (LetPrim (v6 (Constant (Int 3)))
-                  (LetPrim (v7 (Constant (Int 4)))
-                    (LetPrim (v8 (Constant (Int 5)))
-                      (InvokeContinuation k0 (v6 v7 v8)))))))
-      (Branch (IsTrue x) k1 k2))))
-""";
-String DEAD_PARAMETER_OUT = """
-(FunctionDefinition main () (x) return
-  (LetCont ((k0 (v0 v1)
-              (InvokeStatic foo (v0 v1) return)))
-    (LetCont ((k1 ()
-                (LetPrim (v2 (Constant (Int 0)))
-                  (LetPrim (v3 (Constant (Int 2)))
-                    (InvokeContinuation k0 (v2 v3)))))
-              (k2 ()
-                (LetPrim (v4 (Constant (Int 3)))
-                  (LetPrim (v5 (Constant (Int 5)))
-                    (InvokeContinuation k0 (v4 v5))))))
-      (Branch (IsTrue x) k1 k2))))
-""";
-
-// Create an eta-cont redex:
-// Dead parameter reductions can create an eta-cont redex by removing unused
-// continuation parameters and thus creating the eta redex.
-String CREATE_ETA_CONT_IN = """
-(FunctionDefinition main () (x) return
-  (LetCont ((rec loop (v0)
-              (InvokeContinuation rec loop (v0))))
-    (LetCont ((created (v1 v2 v3)
-                (InvokeContinuation loop (v2))))
-      (LetCont ((then ()
-                  (LetPrim (v4 (Constant (Int 0)))
-                    (LetPrim (v5 (Constant (Int 1)))
-                      (LetPrim (v6 (Constant (Int 2)))
-                        (InvokeContinuation created (v4 v5 v6))))))
-                (else ()
-                  (LetPrim (v6 (Constant (Int 3)))
-                    (LetPrim (v7 (Constant (Int 4)))
-                      (LetPrim (v8 (Constant (Int 5)))
-                        (InvokeContinuation created (v6 v7 v8)))))))
-        (Branch (IsTrue x) then else)))))
-""";
-String CREATE_ETA_CONT_OUT = """
-(FunctionDefinition main () (x) return
-  (LetCont ((rec k0 (v0)
-              (InvokeContinuation rec k0 (v0))))
-    (LetCont ((k1 ()
-                (LetPrim (v1 (Constant (Int 1)))
-                  (InvokeContinuation k0 (v1))))
-              (k2 ()
-                (LetPrim (v2 (Constant (Int 4)))
-                  (InvokeContinuation k0 (v2)))))
-      (Branch (IsTrue x) k1 k2))))
-""";
-
-
-
-// Beta-fun-lin and eta-fun might not apply to us, since
-// a. in (InvokeMethod v0 call k0), v0 might carry state, and
-// b. there is no way to generate static nested functions that we could
-//    use InvokeStatic on.
-
-/// Normalizes whitespace by replacing all whitespace sequences by a single
-/// space and trimming leading and trailing whitespace.
-String normalizeSExpr(String input) {
-  return input.replaceAll(new RegExp(r'[ \n\t]+'), ' ').trim();
-}
-
-/// Parses the given input IR, runs an optimization pass over it, and compares
-/// the stringification of the result against the expected output.
-void testShrinkingReducer(String input, String expectedOutput) {
-  final unstringifier = new SExpressionUnstringifier();
-  final stringifier   = new SExpressionStringifier();
-  final optimizer     = new ShrinkingReducer();
-
-  FunctionDefinition f = unstringifier.unstringify(input);
-  optimizer.rewrite(f);
-
-  String expected = normalizeSExpr(expectedOutput);
-  String actual = normalizeSExpr(stringifier.visit(f));
-
-  Expect.equals(expected, actual);
-}
-
-void main() {
-  testShrinkingReducer(DEAD_VAL_IN, DEAD_VAL_OUT);
-  testShrinkingReducer(ITERATIVE_DEAD_VAL1_IN, ITERATIVE_DEAD_VAL1_OUT);
-  testShrinkingReducer(ITERATIVE_DEAD_VAL2_IN, ITERATIVE_DEAD_VAL2_OUT);
-  testShrinkingReducer(DEAD_CONT_IN, DEAD_CONT_OUT);
-  testShrinkingReducer(ITERATIVE_DEAD_CONT_IN, ITERATIVE_DEAD_CONT_OUT);
-  testShrinkingReducer(BETA_CONT_LIN_IN, BETA_CONT_LIN_OUT);
-  testShrinkingReducer(RECURSIVE_BETA_CONT_LIN_IN, RECURSIVE_BETA_CONT_LIN_OUT);
-  testShrinkingReducer(USED_BETA_CONT_LIN_IN, USED_BETA_CONT_LIN_OUT);
-  testShrinkingReducer(ETA_CONT_IN, ETA_CONT_OUT);
-  testShrinkingReducer(DEAD_PARAMETER_IN, DEAD_PARAMETER_OUT);
-  testShrinkingReducer(CREATE_ETA_CONT_IN, CREATE_ETA_CONT_OUT);
-}

tests/compiler/dart2js/backend_dart/sexpr2_test.dart

diff --git a/tests/compiler/dart2js/backend_dart/sexpr2_test.dart b/tests/compiler/dart2js/backend_dart/sexpr2_test.dart
deleted file mode 100644
index 03803bd..0000000
--- a/tests/compiler/dart2js/backend_dart/sexpr2_test.dart
+++ /dev/null
@@ -1,62 +0,0 @@
-// Copyright (c) 2014, the Dart project authors.  Please see the AUTHORS file
-// for details. All rights reserved. Use of this source code is governed by a
-// BSD-style license that can be found in the LICENSE file.
-
-/// Unittest test of the CPS ir generated by the dart2dart compiler.
-library dart_backend.sexpr2_test;
-
-import 'package:compiler/src/compiler.dart';
-import 'package:compiler/src/cps_ir/cps_ir_nodes.dart';
-import 'package:compiler/src/cps_ir/cps_ir_nodes_sexpr.dart';
-import 'package:compiler/src/elements/elements.dart';
-import 'package:expect/expect.dart';
-
-import '../../../../pkg/analyzer2dart/test/test_helper.dart';
-import '../../../../pkg/analyzer2dart/test/sexpr_data.dart';
-
-import 'test_helper.dart';
-
-main(List<String> args) {
-  performTests(TEST_DATA, asyncTester, runTest, args);
-}
-
-runTest(TestSpec result) {
-  return compilerFor(result.input).then((Compiler compiler) {
-    void checkOutput(String elementName,
-                     Element element,
-                     String expectedOutput) {
-      FunctionDefinition ir = compiler.irBuilder.getIr(element);
-      if (expectedOutput == null) {
-        Expect.isNull(ir, "\nInput:\n${result.input}\n"
-                          "No CPS IR expected for $element");
-      } else {
-        Expect.isNotNull(ir, "\nInput:\n${result.input}\n"
-                             "No CPS IR for $element");
-        expectedOutput = expectedOutput.trim();
-        String output = ir.accept(new SExpressionStringifier()).trim();
-        Expect.equals(expectedOutput, output,
-            "\nInput:\n${result.input}\n"
-            "Expected for '$elementName':\n$expectedOutput\n"
-            "Actual for '$elementName':\n$output\n");
-      }
-    }
-
-    if (result.output is String) {
-      checkOutput('main', compiler.mainFunction, result.output);
-    } else {
-      assert(result.output is Map<String, String>);
-      result.output.forEach((String elementName, String output) {
-        Element element;
-        if (elementName.contains('.')) {
-          ClassElement cls = compiler.mainApp.localLookup(
-              elementName.substring(0, elementName.indexOf('.')));
-          element = cls.localLookup(
-              elementName.substring(elementName.indexOf('.') + 1));
-        } else {
-          element = compiler.mainApp.localLookup(elementName);
-        }
-        checkOutput(elementName, element, output);
-      });
-    }
-  });
-}
\ No newline at end of file

tests/compiler/dart2js/backend_dart/sexpr_test.dart

diff --git a/tests/compiler/dart2js/backend_dart/sexpr_test.dart b/tests/compiler/dart2js/backend_dart/sexpr_test.dart
deleted file mode 100644
index 2293cdb..0000000
--- a/tests/compiler/dart2js/backend_dart/sexpr_test.dart
+++ /dev/null
@@ -1,177 +0,0 @@
-// Copyright (c) 2014, the Dart project authors.  Please see the AUTHORS file
-// for details. All rights reserved. Use of this source code is governed by a
-// BSD-style license that can be found in the LICENSE file.
-
-library dart_backend.sexpr_test;
-
-import 'dart:async';
-
-import 'package:async_helper/async_helper.dart';
-import 'package:compiler/src/compiler.dart';
-import 'package:compiler/src/cps_ir/cps_ir_nodes.dart';
-import 'package:compiler/src/cps_ir/cps_ir_nodes_sexpr.dart';
-import 'package:expect/expect.dart';
-
-import '../compiler_helper.dart' hide compilerFor;
-import 'sexpr_unstringifier.dart';
-import 'test_helper.dart';
-
-const String CODE = """
-class Foo {
-  static int x = 1;
-  foo() => 42;
-}
-
-class Bar extends Foo {
-  int y = 2;
-  Bar() {
-    Foo.x = 2;
-  }
-  foo() =>  this.y + super.foo();
-}
-
-class FooBar<T> {
-  bool foobar() => T is int;
-}
-
-main() {
-  Foo foo;
-  if (foo is Foo) {
-    print("Surprise");
-  }
-
-  String inner() => "Inner function";
-  print(inner());
-
-  int j = 42;
-  for (int i = 0; i < 2; i++) {
-    print(i.toString());
-  }
-
-  print(Foo.x);
-
-  String recursive(int h) {
-    if (h < 1) {
-      return j.toString();
-    }
-    j++;
-    return "h\$\{recursive(h - 1)\}";
-  }
-  print(recursive(5));
-
-  Bar bar = new Bar();
-  print(bar.foo().toString());
-
-  bar = foo as Bar;
-  if (bar == null) {
-    var list = [1, 2, 3, 4];
-    var map  = { 1: "one"
-               , 2: "two"
-               , 3: "three"
-               };
-    var double = 3.14;
-    list.forEach((i) => print(i.toString()));
-    print("\$list \$map \$double");
-  }
-
-  print(new FooBar<int>().foobar());
-}
-
-""";
-
-bool shouldOutput(Element element) {
-  return (!element.library.isPlatformLibrary &&
-          !element.isSynthesized &&
-          element.kind == ElementKind.FUNCTION);
-}
-
-/// Compiles the given dart code (which must include a 'main' function) and
-/// returns a list of all generated CPS IR definitions.
-Future<List<FunctionDefinition>> compile(String code) {
-  return compilerFor(code).then((Compiler compiler) {
-    return compiler.enqueuer.resolution.resolvedElements
-              .where(shouldOutput)
-              .map(compiler.irBuilder.getIr)
-              .toList();
-  });
-}
-
-/// Returns an S-expression string for each compiled function.
-List<String> stringifyAll(Iterable<FunctionDefinition> functions) {
-  final stringifier = new SExpressionStringifier();
-  return functions.map((f) => stringifier.visitFunctionDefinition(f)).toList();
-}
-
-Future<List<String>> testStringifier(String code,
-                                     Iterable<String> expectedTokens) {
-  return compile(code)
-      .then((List<FunctionDefinition> functions) {
-        List<String> sexprs = stringifyAll(functions);
-        String combined = sexprs.join();
-        String withoutNullConstants = combined.replaceAll("Constant null", "");
-        Expect.isFalse(withoutNullConstants.contains("null"));
-        for (String token in expectedTokens) {
-          Expect.isTrue(combined.contains(token),
-              "'$combined' doesn't contain '$token' in test:\n$code");
-        }
-
-        return sexprs;
-      });
-}
-
-/// Checks if the generated S-expressions can be processed by the unstringifier,
-/// returns the resulting definitions.
-List<FunctionDefinition> testUnstringifier(List<String> sexprs) {
-  return sexprs.map((String sexpr) {
-    try {
-      final function = new SExpressionUnstringifier().unstringify(sexpr);
-      Expect.isNotNull(function, "Unstringification failed:\n\n$sexpr");
-      return function;
-    } catch (e, s) {
-      print('$e\n$s');
-      Expect.fail('Error unstringifying "$sexpr": $e');
-    }
-  }).toList();
-}
-
-void main() {
-  final tokens =
-          [ "FunctionDefinition"
-          , "IsTrue"
-
-          // Expressions
-          , "Branch"
-          , "ConcatenateStrings"
-          , "DeclareFunction"
-          , "InvokeConstructor"
-          , "InvokeContinuation"
-          , "InvokeMethod"
-          , "InvokeStatic"
-          , "InvokeMethodDirectly"
-          , "LetCont"
-          , "LetPrim"
-          , "SetMutableVariable"
-          , "TypeOperator"
-
-          // Primitives
-          , "Constant"
-          , "CreateFunction"
-          , "GetMutableVariable"
-          , "LiteralList"
-          , "LiteralMap"
-          // Parameters are encoded by name only and hence are not in this list.
-          , "ReifyTypeVar"
-
-          , "(this)"   // 'this' Parameter declarations
-          , "this"     // 'this' Parameter uses
-
-          ];
-
-  asyncTest(() => testStringifier(CODE, tokens).then((List<String> sexprs) {
-    final functions = testUnstringifier(sexprs);
-
-    // Ensure that
-    // stringified(CODE) == stringified(unstringified(stringified(CODE)))
-    Expect.listEquals(sexprs, stringifyAll(functions));
-  }));
-}

tests/compiler/dart2js/backend_dart/sexpr_unstringifier.dart

diff --git a/tests/compiler/dart2js/backend_dart/sexpr_unstringifier.dart b/tests/compiler/dart2js/backend_dart/sexpr_unstringifier.dart
deleted file mode 100644
index e0252e7..0000000
--- a/tests/compiler/dart2js/backend_dart/sexpr_unstringifier.dart
+++ /dev/null
@@ -1,885 +0,0 @@
-// Copyright (c) 2014, the Dart project authors.  Please see the AUTHORS file
-// for details. All rights reserved. Use of this source code is governed by a
-// BSD-style license that can be found in the LICENSE file.
-
-// SExpressionUnstringifier implements the inverse operation to
-// [SExpressionStringifier].
-
-library sexpr_unstringifier;
-
-import 'package:compiler/src/constants/expressions.dart';
-import 'package:compiler/src/constants/values.dart';
-import 'package:compiler/src/dart_types.dart' as dart_types
-    show DartType;
-import 'package:compiler/src/diagnostics/messages.dart'
-    show MessageKind;
-import 'package:compiler/src/elements/elements.dart';
-import 'package:compiler/src/elements/modelx.dart'
-    show ErroneousElementX, TypeVariableElementX;
-import 'package:compiler/src/tree/tree.dart' show LiteralDartString;
-import 'package:compiler/src/universe/call_structure.dart'
-    show CallStructure;
-import 'package:compiler/src/universe/selector.dart'
-    show Selector, SelectorKind;
-import 'package:compiler/src/cps_ir/cps_ir_nodes.dart';
-
-/// Used whenever a node constructed by [SExpressionUnstringifier] needs a
-/// named entity.
-class DummyEntity extends Entity {
-  final String name;
-  DummyEntity(this.name);
-}
-
-/// Used whenever a node constructed by [SExpressionUnstringifier] needs a
-/// local.
-class DummyLocal extends DummyEntity implements Local {
-  DummyLocal(String name) : super(name);
-
-  ExecutableElement get executableContext => null;
-}
-
-// TODO(karlklose): we should remove all references to [ErroneousElement] from
-// the CPS IR.  Instead, the builder must construct appropriate terms for ASTs
-// that could not be resolved correctly.  Perhaps the IR should not rely on
-// elements at all for naming.
-/// Used whenever a node constructed by [SExpressionUnstringifier] requires
-/// an [Element] or [FunctionElement]. Extends [ErroneousElementX] since there
-/// is currently a large amount of overhead when extending the base abstract
-/// classes, and erroneous elements conveniently also skip several assertion
-/// checks in CPS IR nodes that are irrelevant to us.
-class DummyElement extends ErroneousElementX
-    implements TypeVariableElement, FieldElement {
-  DummyElement(String name)
-      : super(MessageKind.GENERIC, {}, name, null);
-
-  final dart_types.DartType bound = null;
-  final TypeDeclarationElement typeDeclaration = null;
-
-  noSuchMethod(inv) => super.noSuchMethod(inv);
-}
-
-/// Used whenever a node constructed by [SExpressionUnstringifier] requires
-/// a named type.
-class DummyNamedType extends dart_types.DartType {
-  final String name;
-
-  final kind = null;
-  final element = null;
-
-  DummyNamedType(this.name);
-
-  subst(arguments, parameters) => null;
-  unalias(compiler) => null;
-  accept(visitor, argument) => null;
-
-  String toString() => name;
-}
-
-/// Represents a list of tokens, but is basically a partial view into a list
-/// with appropriate convenience methods.
-class Tokens {
-  final List<String> _list;
-  int _index; // Current index into the list.
-
-  Tokens(List<String> this._list) : _index = 0;
-
-  String get current => _list[_index];
-  String get next    => _list[_index + 1];
-
-  String read([String expected]) {
-    if (expected != null) {
-      if (current != expected) {
-        print('expected "$expected", found "$current"');
-        int start = _index - 15;
-        String dotdotdot = '... ';
-        if (start < 0) {
-          start = 0;
-          dotdotdot = '';
-        }
-        print('${dotdotdot}${_list.sublist(start, _index + 1).join(' ')}');
-        assert(current == expected);
-      }
-    }
-    return _list[_index++];
-  }
-
-  /// Consumes the preamble to a new node, consisting of an opening parenthesis
-  /// and a tag.
-  void consumeStart([String tag]) {
-    read("(");
-    if (tag != null) {
-      read(tag);
-    }
-  }
-
-  void consumeEnd() {
-    read(")");
-  }
-
-  bool get hasNext  => _index < _list.length;
-  String toString() => _list.sublist(_index).toString();
-}
-
-/// Constructs a minimal in-memory representation of the IR represented
-/// by the given string. Many fields are currently simply set to null.
-class SExpressionUnstringifier {
-
-  // Expressions
-  static const String BRANCH = "Branch";
-  static const String CONCATENATE_STRINGS = "ConcatenateStrings";
-  static const String DECLARE_FUNCTION = "DeclareFunction";
-  static const String INVOKE_CONSTRUCTOR = "InvokeConstructor";
-  static const String INVOKE_CONTINUATION = "InvokeContinuation";
-  static const String INVOKE_STATIC = "InvokeStatic";
-  static const String INVOKE_METHOD_DIRECTLY = "InvokeMethodDirectly";
-  static const String INVOKE_METHOD = "InvokeMethod";
-  static const String LET_PRIM = "LetPrim";
-  static const String LET_CONT = "LetCont";
-  static const String LET_MUTABLE = "LetMutable";
-  static const String TYPE_CAST = "TypeCast";
-  static const String GET_LAZY_STATIC = "GetLazyStatic";
-  static const String UNREACHABLE = "Unreachable";
-
-  // Primitives
-  static const String CONSTANT = "Constant";
-  static const String CREATE_FUNCTION = "CreateFunction";
-  static const String GET_MUTABLE = "GetMutable";
-  static const String SET_MUTABLE = "SetMutable";
-  static const String LITERAL_LIST = "LiteralList";
-  static const String LITERAL_MAP = "LiteralMap";
-  static const String REIFY_TYPE_VAR = "ReifyTypeVar";
-  static const String GET_STATIC = "GetStatic";
-  static const String SET_STATIC = "SetStatic";
-  static const String TYPE_TEST = "TypeTest";
-  static const String APPLY_BUILTIN_OPERATOR = "ApplyBuiltinOperator";
-  static const String GET_LENGTH = "GetLength";
-  static const String GET_INDEX = "GetIndex";
-  static const String SET_INDEX = "SetIndex";
-  static const String GET_FIELD = "GetField";
-  static const String SET_FIELD = "SetField";
-
-  // Other
-  static const String FUNCTION_DEFINITION = "FunctionDefinition";
-  static const String IS_TRUE = "IsTrue";
-
-  // Constants
-  static const String BOOL = "Bool";
-  static const String DOUBLE = "Double";
-  static const String INT = "Int";
-  static const String NULL = "Null";
-  static const String STRING = "String";
-
-  final Map<String, Definition> name2variable =
-      <String, Definition>{ "return": new Continuation.retrn() };
-
-  // Operator names used for canonicalization. In theory, we could simply use
-  // Elements.isOperatorName() on the parsed tokens; however, comparisons are
-  // done using identical() for performance reasons, which are reliable only for
-  // compile-time literal strings.
-  static Set<String> OPERATORS = new Set<String>.from(
-      [ '~', '==', '[]', '*', '/', '%', '~/', '+', '<<', 'unary-'
-      , '>>', '>=', '>', '<=', '<', '&', '^', '|', '[]=', '-'
-      ]);
-
-  // The tokens currently being parsed.
-  Tokens tokens;
-
-  FunctionDefinition unstringify(String s) {
-    tokens = tokenize(s);
-    FunctionDefinition def = parseFunctionDefinition();
-    assert(!tokens.hasNext);
-    return def;
-  }
-
-  /// Returns a new named dummy selector with a roughly appropriate kind.
-  Selector dummySelector(String name, int argumentCount) {
-    SelectorKind kind;
-    if (name == "[]") {
-      kind = SelectorKind.INDEX;
-    } else if (Elements.isOperatorName(name)) {
-      kind = SelectorKind.OPERATOR;
-    } else {
-      kind = SelectorKind.CALL;
-    }
-    return new Selector(kind, new PublicName(name),
-        new CallStructure.unnamed(argumentCount));
-  }
-
-  /// Returns the tokens in s. Note that string literals are not necessarily
-  /// preserved; for instance, "(literalString)" is transformed to
-  /// " ( literalString ) ".
-  Tokens tokenize(String s) =>
-      new Tokens(
-          s.replaceAll("(", " ( ")
-           .replaceAll(")", " ) ")
-           .replaceAll("{", " { ")
-           .replaceAll("}", " } ")
-           .replaceAll(new RegExp(r"[ \t\n]+"), " ")
-           .trim()
-           .split(" ")
-           .map(canonicalizeOperators)
-           .toList());
-
-  /// Canonicalizes strings containing operator names.
-  String canonicalizeOperators(String token) {
-    String opname = OPERATORS.lookup(token);
-    if (opname != null) {
-      return opname;
-    }
-    return token;
-  }
-
-  Expression parseExpression() {
-    assert(tokens.current == "(");
-
-    switch (tokens.next) {
-      case BRANCH:
-        return parseBranch();
-      case CONCATENATE_STRINGS:
-        return parseConcatenateStrings();
-      case DECLARE_FUNCTION:
-        return parseDeclareFunction();
-      case INVOKE_CONSTRUCTOR:
-        return parseInvokeConstructor();
-      case INVOKE_CONTINUATION:
-        return parseInvokeContinuation();
-      case INVOKE_METHOD:
-        return parseInvokeMethod();
-      case INVOKE_STATIC:
-        return parseInvokeStatic();
-      case INVOKE_METHOD_DIRECTLY:
-        return parseInvokeMethodDirectly();
-      case LET_PRIM:
-        return parseLetPrim();
-      case LET_CONT:
-        return parseLetCont();
-      case LET_MUTABLE:
-        return parseLetMutable();
-      case TYPE_CAST:
-        return parseTypeCast();
-      case GET_LAZY_STATIC:
-        return parseGetLazyStatic();
-      case UNREACHABLE:
-        return parseUnreachable();
-      default:
-        assert(false);
-    }
-
-    return null;
-  }
-
-  /// (prim1 prim2 ... primn)
-  List<Primitive> parsePrimitiveList() {
-    tokens.consumeStart();
-    List<Primitive> prims = <Primitive>[];
-    while (tokens.current != ")") {
-      Primitive prim = name2variable[tokens.read()];
-      assert(prim != null);
-      prims.add(prim);
-    }
-    tokens.consumeEnd();
-    return prims;
-  }
-
-  /// (FunctionDefinition name (parameters) continuation body)
-  FunctionDefinition parseFunctionDefinition() {
-    tokens.consumeStart(FUNCTION_DEFINITION);
-
-    // name
-    Element element = new DummyElement("");
-    if (tokens.current != '(') {
-      // This is a named function.
-      element = new DummyElement(tokens.read());
-    }
-
-    // (this) or ()
-    Definition thisParameter = null;
-    tokens.consumeStart();
-    if (tokens.current != ')') {
-      String thisName = tokens.read();
-      if (name2variable.containsKey(thisName)) {
-        thisParameter = name2variable[thisName];
-      } else {
-        thisParameter = new Parameter(new DummyElement(thisName));
-        name2variable[thisName] = thisParameter;
-      }
-    }
-    tokens.consumeEnd();
-
-    // (parameters)
-    List<Definition> parameters = <Definition>[];
-    tokens.consumeStart();
-    while (tokens.current != ")") {
-      String paramName = tokens.read();
-      if (name2variable.containsKey(paramName)) {
-        parameters.add(name2variable[paramName]);
-      } else {
-        Parameter param = new Parameter(new DummyElement(paramName));
-        name2variable[paramName] = param;
-        parameters.add(param);
-      }
-    }
-    tokens.consumeEnd();
-
-    // continuation
-    String contName = tokens.read("return");
-    Continuation cont = name2variable[contName];
-    assert(cont != null);
-
-    // body
-    Expression body = parseExpression();
-
-    tokens.consumeEnd();
-    return new FunctionDefinition(element, thisParameter, parameters,
-        new Body(body, cont), null, null);
-  }
-
-  /// (IsTrue arg)
-  Condition parseCondition() {
-    // Handles IsTrue only for now.
-    tokens.consumeStart(IS_TRUE);
-
-    Definition value = name2variable[tokens.read()];
-    assert(value != null);
-
-    tokens.consumeEnd();
-    return new IsTrue(value);
-  }
-
-  /// (Branch condition cont cont)
-  Branch parseBranch() {
-    tokens.consumeStart(BRANCH);
-
-    Condition cond = parseCondition();
-    Continuation trueCont = name2variable[tokens.read()];
-    Continuation falseCont = name2variable[tokens.read()];
-    assert(trueCont != null && falseCont != null);
-
-    tokens.consumeEnd();
-    return new Branch(cond, trueCont, falseCont);
-  }
-
-  /// (ConcatenateStrings (args) cont)
-  ConcatenateStrings parseConcatenateStrings() {
-    tokens.consumeStart(CONCATENATE_STRINGS);
-
-    List<Primitive> args = parsePrimitiveList();
-
-    Continuation cont = name2variable[tokens.read()];
-    assert(cont != null);
-
-    tokens.consumeEnd();
-    return new ConcatenateStrings(args, cont);
-  }
-
-  /// (DeclareFunction name = function in body)
-  DeclareFunction parseDeclareFunction() {
-    tokens.consumeStart(DECLARE_FUNCTION);
-
-    // name =
-    MutableVariable local = addMutableVariable(tokens.read());
-    tokens.read("=");
-
-    // function in
-    FunctionDefinition def = parseFunctionDefinition();
-    tokens.read("in");
-
-    // body
-    Expression body = parseExpression();
-
-    tokens.consumeEnd();
-    return new DeclareFunction(local, def)..plug(body);
-  }
-
-  /// (InvokeConstructor name (args) cont)
-  InvokeConstructor parseInvokeConstructor() {
-    tokens.consumeStart(INVOKE_CONSTRUCTOR);
-
-    String constructorName = tokens.read();
-    List<String> split = constructorName.split(".");
-    assert(split.length < 3);
-
-    dart_types.DartType type = new DummyNamedType(split[0]);
-    Element element = new DummyElement((split.length == 1) ? "" : split[1]);
-
-    List<Primitive> args = parsePrimitiveList();
-
-    Continuation cont = name2variable[tokens.read()];
-    assert(cont != null);
-
-    tokens.consumeEnd();
-    Selector selector = dummySelector(constructorName, args.length);
-    return new InvokeConstructor(type, element, selector, args, cont);
-  }
-
-  /// (InvokeContinuation rec? name (args))
-  InvokeContinuation parseInvokeContinuation() {
-    tokens.consumeStart(INVOKE_CONTINUATION);
-    String name = tokens.read();
-    bool isRecursive = name == "rec";
-    if (isRecursive) name = tokens.read();
-
-    Continuation cont = name2variable[name];
-    assert(cont != null);
-
-    List<Primitive> args = parsePrimitiveList();
-
-    tokens.consumeEnd();
-    return new InvokeContinuation(cont, args, isRecursive: isRecursive);
-  }
-
-  /// (InvokeMethod receiver method (args) cont)
-  InvokeMethod parseInvokeMethod() {
-    tokens.consumeStart(INVOKE_METHOD);
-
-    Definition receiver = name2variable[tokens.read()];
-    assert(receiver != null);
-
-    String methodName = tokens.read();
-
-    List<Primitive> args = parsePrimitiveList();
-
-    Continuation cont = name2variable[tokens.read()];
-    assert(cont != null);
-
-    tokens.consumeEnd();
-    Selector selector = dummySelector(methodName, args.length);
-    return new InvokeMethod(receiver, selector, args, cont);
-  }
-
-  /// (InvokeStatic method (args) cont)
-  InvokeStatic parseInvokeStatic() {
-    tokens.consumeStart(INVOKE_STATIC);
-
-    String methodName = tokens.read();
-
-    List<Primitive> args = parsePrimitiveList();
-
-    Continuation cont = name2variable[tokens.read()];
-    assert(cont != null);
-
-    Entity entity = new DummyEntity(methodName);
-    Selector selector = dummySelector(methodName, args.length);
-
-    tokens.consumeEnd();
-    return new InvokeStatic(entity, selector, args, cont, null);
-  }
-
-  /// (InvokeMethodDirectly receiver method (args) cont)
-  InvokeMethodDirectly parseInvokeMethodDirectly() {
-    tokens.consumeStart(INVOKE_METHOD_DIRECTLY);
-
-    Definition receiver = name2variable[tokens.read()];
-    assert(receiver != null);
-
-    String methodName = tokens.read();
-
-    List<Primitive> args = parsePrimitiveList();
-
-    Continuation cont = name2variable[tokens.read()];
-    assert(cont != null);
-
-    tokens.consumeEnd();
-    Element element = new DummyElement(methodName);
-    Selector selector = dummySelector(methodName, args.length);
-    return new InvokeMethodDirectly(receiver, element, selector, args, cont);
-  }
-
-  // (rec? name (args) body)
-  Continuation parseContinuation() {
-    // (rec? name
-    tokens.consumeStart();
-    String name = tokens.read();
-    bool isRecursive = name == "rec";
-    if (isRecursive) name = tokens.read();
-
-    // (args)
-    tokens.consumeStart();
-    List<Parameter> params = <Parameter>[];
-    while (tokens.current != ")") {
-      String paramName = tokens.read();
-      Parameter param = new Parameter(new DummyElement(paramName));
-      name2variable[paramName] = param;
-      params.add(param);
-    }
-    tokens.consumeEnd();
-
-    Continuation cont = new Continuation(params);
-    name2variable[name] = cont;
-
-    cont.isRecursive = isRecursive;
-    // cont_body
-    cont.body = parseExpression();
-    tokens.consumeEnd();
-    return cont;
-  }
-
-  /// (LetCont (continuations) body)
-  LetCont parseLetCont() {
-    tokens.consumeStart(LET_CONT);
-    tokens.consumeStart();
-    List<Continuation> continuations = <Continuation>[];
-    while (tokens.current != ")") {
-      continuations.add(parseContinuation());
-    }
-    tokens.consumeEnd();
-
-    // body)
-    Expression body = parseExpression();
-    tokens.consumeEnd();
-
-    return new LetCont.many(continuations, body);
-  }
-
-  /// (LetMutable (name value) body)
-  LetMutable parseLetMutable() {
-    tokens.consumeStart(LET_MUTABLE);
-
-    tokens.consumeStart();
-    String name = tokens.read();
-    MutableVariable local = addMutableVariable(name);
-    Primitive value = name2variable[tokens.read()];
-    tokens.consumeEnd();
-
-    Expression body = parseExpression();
-    tokens.consumeEnd();
-    return new LetMutable(local, value)..plug(body);
-  }
-
-  /// (SetMutable name value)
-  SetMutable parseSetMutable() {
-    tokens.consumeStart(SET_MUTABLE);
-
-    MutableVariable local = name2variable[tokens.read()];
-    Primitive value = name2variable[tokens.read()];
-    assert(value != null);
-
-    tokens.consumeEnd();
-    return new SetMutable(local, value);
-  }
-
-  /// (TypeCast value type args cont)
-  TypeCast parseTypeCast() {
-    tokens.consumeStart(TYPE_CAST);
-
-    Primitive value = name2variable[tokens.read()];
-    assert(value != null);
-
-    dart_types.DartType type = new DummyNamedType(tokens.read());
-
-    List<ir.Primitive> typeArguments = parsePrimitiveList();
-
-    Continuation cont = name2variable[tokens.read()];
-    assert(cont != null);
-
-    tokens.consumeEnd();
-    return new TypeCast(value, type, typeArguments, cont);
-  }
-
-  /// (TypeTest value type args)
-  TypeTest parseTypeTest() {
-    tokens.consumeStart(TYPE_TEST);
-
-    Primitive value = name2variable[tokens.read()];
-    assert(value != null);
-
-    dart_types.DartType type = new DummyNamedType(tokens.read());
-
-    List<ir.Primitive> typeArguments = parsePrimitiveList();
-
-    tokens.consumeEnd();
-    return new TypeTest(value, type, typeArguments);
-  }
-
-  /// (ApplyBuiltinOperator operator args)
-  ApplyBuiltinOperator parseApplyBuiltinOperator() {
-    tokens.consumeStart(APPLY_BUILTIN_OPERATOR);
-
-    String operatorName = tokens.read();
-    BuiltinOperator operator;
-    for (BuiltinOperator op in BuiltinOperator.values) {
-      if (op.toString() == operatorName) {
-        operator = op;
-        break;
-      }
-    }
-    assert(operator != null);
-    List<ir.Primitive> arguments = parsePrimitiveList();
-
-    tokens.consumeEnd();
-    return new ApplyBuiltinOperator(operator, arguments);
-  }
-
-  /// (GetLength object)
-  GetLength parseGetLength() {
-    tokens.consumeStart(GET_LENGTH);
-    Primitive object = name2variable[tokens.read()];
-    tokens.consumeEnd();
-    return new GetLength(object);
-  }
-
-  /// (GetIndex object index)
-  GetIndex parseGetIndex() {
-    tokens.consumeStart(GET_INDEX);
-    Primitive object = name2variable[tokens.read()];
-    Primitive index = name2variable[tokens.read()];
-    tokens.consumeEnd();
-    return new GetIndex(object, index);
-  }
-
-  /// (SetIndex object index value)
-  SetIndex parseSetIndex() {
-    tokens.consumeStart(SET_INDEX);
-    Primitive object = name2variable[tokens.read()];
-    Primitive index = name2variable[tokens.read()];
-    Primitive value = name2variable[tokens.read()];
-    tokens.consumeEnd();
-    return new SetIndex(object, index, value);
-  }
-
-  /// (SetStatic field value)
-  SetStatic parseSetStatic() {
-    tokens.consumeStart(SET_STATIC);
-
-    Element fieldElement = new DummyElement(tokens.read());
-    Primitive value = name2variable[tokens.read()];
-    assert(value != null);
-
-    tokens.consumeEnd();
-    return new SetStatic(fieldElement, value, null);
-  }
-
-  /// (GetLazyStatic field cont)
-  GetLazyStatic parseGetLazyStatic() {
-    tokens.consumeStart(GET_LAZY_STATIC);
-
-    Element fieldElement = new DummyElement(tokens.read());
-    Continuation cont = name2variable[tokens.read()];
-    assert(cont != null);
-
-    tokens.consumeEnd();
-    return new GetLazyStatic(fieldElement, cont, null);
-  }
-
-  /// (Unreachable)
-  Unreachable parseUnreachable() {
-    tokens.consumeStart(UNREACHABLE);
-    tokens.consumeEnd();
-    return new Unreachable();
-  }
-
-  /// (LetPrim (name primitive) body)
-  LetPrim parseLetPrim() {
-    tokens.consumeStart(LET_PRIM);
-
-    // (name
-    tokens.consumeStart();
-    String name = tokens.read();
-
-    // primitive)
-    Primitive primitive = parsePrimitive();
-    name2variable[name] = primitive;
-    tokens.consumeEnd();
-
-    // body)
-    Expression body = parseExpression();
-    tokens.consumeEnd();
-
-    return new LetPrim(primitive)..plug(body);
-  }
-
-  Primitive parsePrimitive() {
-    assert(tokens.current == "(");
-
-    switch (tokens.next) {
-      case CONSTANT:
-        return parseConstant();
-      case CREATE_FUNCTION:
-        return parseCreateFunction();
-      case GET_MUTABLE:
-        return parseGetMutable();
-      case SET_MUTABLE:
-        return parseSetMutable();
-      case LITERAL_LIST:
-        return parseLiteralList();
-      case LITERAL_MAP:
-        return parseLiteralMap();
-      case REIFY_TYPE_VAR:
-        return parseReifyTypeVar();
-      case GET_STATIC:
-        return parseGetStatic();
-      case SET_STATIC:
-        return parseSetStatic();
-      case TYPE_TEST:
-        return parseTypeTest();
-      case APPLY_BUILTIN_OPERATOR:
-        return parseApplyBuiltinOperator();
-      case GET_LENGTH:
-        return parseGetLength();
-      case GET_INDEX:
-        return parseGetIndex();
-      case SET_INDEX:
-        return parseSetIndex();
-      case GET_FIELD:
-        return parseGetField();
-      case SET_FIELD:
-        return parseSetField();
-      default:
-        assert(false);
-    }
-
-    return null;
-  }
-
-  /// (Constant (constant))
-  Constant parseConstant() {
-    tokens.consumeStart(CONSTANT);
-    tokens.consumeStart();
-    Constant result;
-    String tag = tokens.read();
-    switch (tag) {
-      case NULL:
-        result = new Constant(
-            new NullConstantExpression(new NullConstantValue()));
-        break;
-      case BOOL:
-        String value = tokens.read();
-        if (value == "true") {
-          result = new Constant(
-              new BoolConstantExpression(true, new TrueConstantValue()));
-        } else if (value == "false") {
-          result = new Constant(
-              new BoolConstantExpression(false, new FalseConstantValue()));
-        } else {
-          throw "Invalid Boolean value '$value'.";
-        }
-        break;
-      case STRING:
-        List<String> strings = <String>[];
-        do {
-          strings.add(tokens.read());
-        } while (tokens.current != ")");
-        String string = strings.join(" ");
-        assert(string.startsWith('"') && string.endsWith('"'));
-        String text = string.substring(1, string.length - 1);
-        StringConstantValue value = new StringConstantValue(
-            new LiteralDartString(text));
-        result = new Constant(new StringConstantExpression(text, value));
-        break;
-      case INT:
-        String value = tokens.read();
-        int intValue = int.parse(value, onError: (_) => null);
-        if (intValue == null) {
-          throw "Invalid int value 'value'.";
-        }
-        result = new Constant(new IntConstantExpression(
-            intValue, new IntConstantValue(intValue)));
-        break;
-      case DOUBLE:
-        String value = tokens.read();
-        double doubleValue = double.parse(value, (_) => null);
-        if (doubleValue == null) {
-          throw "Invalid double value '$value'.";
-        }
-        result = new Constant(new DoubleConstantExpression(
-            doubleValue, new DoubleConstantValue(doubleValue)));
-        break;
-      default:
-        throw "Unexpected constant tag '$tag'.";
-    }
-    tokens.consumeEnd();
-    tokens.consumeEnd();
-    return result;
-  }
-
-  /// (CreateFunction (definition))
-  CreateFunction parseCreateFunction() {
-    tokens.consumeStart(CREATE_FUNCTION);
-    FunctionDefinition def = parseFunctionDefinition();
-    tokens.consumeEnd();
-    return new CreateFunction(def);
-  }
-
-  MutableVariable addMutableVariable(String name) {
-    assert(!name2variable.containsKey(name));
-    MutableVariable variable = new MutableVariable(new DummyElement(name));
-    name2variable[name] = variable;
-    return variable;
-  }
-
-  /// (GetMutable name)
-  GetMutable parseGetMutable() {
-    tokens.consumeStart(GET_MUTABLE);
-    MutableVariable local = name2variable[tokens.read()];
-    tokens.consumeEnd();
-
-    return new GetMutable(local);
-  }
-
-  /// (LiteralList (values))
-  LiteralList parseLiteralList() {
-    tokens.consumeStart(LITERAL_LIST);
-    List<Primitive> values = parsePrimitiveList();
-    tokens.consumeEnd();
-    return new LiteralList(null, values);
-  }
-
-  /// (LiteralMap (keys) (values))
-  LiteralMap parseLiteralMap() {
-    tokens.consumeStart(LITERAL_MAP);
-
-    List<Primitive> keys   = parsePrimitiveList();
-    List<Primitive> values = parsePrimitiveList();
-
-    List<LiteralMapEntry> entries = <LiteralMapEntry>[];
-    for (int i = 0; i < keys.length; i++) {
-      entries.add(new LiteralMapEntry(keys[i], values[i]));
-    }
-
-    tokens.consumeEnd();
-    return new LiteralMap(null, entries);
-  }
-
-  /// (ReifyTypeVar type)
-  ReifyTypeVar parseReifyTypeVar() {
-    tokens.consumeStart(REIFY_TYPE_VAR);
-
-    TypeVariableElement type = new DummyElement(tokens.read());
-
-    tokens.consumeEnd();
-    return new ReifyTypeVar(type);
-  }
-
-  /// (GetStatic field)
-  GetStatic parseGetStatic() {
-    tokens.consumeStart(GET_STATIC);
-
-    Element field = new DummyElement(tokens.read());
-
-    tokens.consumeEnd();
-    return new GetStatic(field, null);
-  }
-
-  /// (GetField object field)
-  GetField parseGetField() {
-    tokens.consumeStart(GET_FIELD);
-
-    Primitive object = name2variable[tokens.read()];
-    Element field = new DummyElement(tokens.read());
-
-    tokens.consumeEnd();
-    return new GetField(object, field);
-  }
-
-  /// (SetField object field value)
-  SetField parseSetField() {
-    tokens.consumeStart(SET_FIELD);
-
-    Primitive object = name2variable[tokens.read()];
-    Element field = new DummyElement(tokens.read());
-    Primitive value = name2variable[tokens.read()];
-
-    tokens.consumeEnd();
-    return new SetField(object, field, value);
-  }
-}

tests/compiler/dart2js/backend_dart/test_helper.dart

diff --git a/tests/compiler/dart2js/backend_dart/test_helper.dart b/tests/compiler/dart2js/backend_dart/test_helper.dart
deleted file mode 100644
index 0eeb656..0000000
--- a/tests/compiler/dart2js/backend_dart/test_helper.dart
+++ /dev/null
@@ -1,46 +0,0 @@
-// Copyright (c) 2014, the Dart project authors.  Please see the AUTHORS file
-// for details. All rights reserved. Use of this source code is governed by a
-// BSD-style license that can be found in the LICENSE file.
-
-library dart_backend.test_helper;
-
-import 'dart:async';
-import 'package:async_helper/async_helper.dart';
-import 'package:compiler/compiler.dart' as api;
-import 'package:compiler/src/compiler.dart';
-import '../../../../pkg/analyzer2dart/test/test_helper.dart';
-import '../compiler_helper.dart';
-
-/// Compiles the given dart code (which must include a 'main' function) and
-/// returns the compiler.
-Future<Compiler> compilerFor(String code,
-                             {api.CompilerOutputProvider outputProvider}) {
-  MockCompiler compiler = new MockCompiler.internal(
-      emitJavaScript: false,
-      enableMinification: false,
-      outputProvider: outputProvider);
-  compiler.diagnosticHandler = createHandler(compiler, code);
-  return compiler.init().then((_) {
-    compiler.parseScript(code);
-
-    Element element = compiler.mainApp.find('main');
-    if (element == null) return null;
-
-    compiler.mainFunction = element;
-    compiler.phase = Compiler.PHASE_RESOLVING;
-    compiler.backend.enqueueHelpers(compiler.enqueuer.resolution,
-                                    compiler.globalDependencies);
-    compiler.processQueue(compiler.enqueuer.resolution, element);
-    compiler.world.populate();
-    compiler.backend.onResolutionComplete();
-
-    compiler.irBuilder.buildNodes();
-
-    return compiler;
-  });
-}
-
-/// Test group using async_helper.
-asyncTester(Group group, RunTest runTest) {
-  asyncTest(() => Future.forEach(group.results, runTest));
-}

tests/compiler/dart2js/generate_code_with_compile_time_errors_test.dart

diff --git a/tests/compiler/dart2js/generate_code_with_compile_time_errors_test.dart b/tests/compiler/dart2js/generate_code_with_compile_time_errors_test.dart
index d2b29b0..92910dc 100644
--- a/tests/compiler/dart2js/generate_code_with_compile_time_errors_test.dart
+++ b/tests/compiler/dart2js/generate_code_with_compile_time_errors_test.dart
@@ -10,7 +10,6 @@
 import 'package:expect/expect.dart';
 import 'package:async_helper/async_helper.dart';
 import 'package:compiler/src/compiler.dart';
-import 'package:compiler/src/dart_backend/dart_backend.dart';
 import 'package:compiler/src/js_backend/js_backend.dart';
 import 'memory_compiler.dart';
 import 'output_collector.dart';
@@ -53,14 +52,8 @@
       collector.hints.isNotEmpty,
       "Unexpected hints: ${collector.warnings}");
 
-  bool isCodeGenerated;
-  if (options.contains('--output-type=dart')) {
-    DartBackend backend = compiler.backend;
-    isCodeGenerated = backend.outputter.libraryInfo != null;
-  } else {
-    JavaScriptBackend backend = compiler.backend;
-    isCodeGenerated = backend.generatedCode.isNotEmpty;
-  }
+  JavaScriptBackend backend = compiler.backend;
+  bool isCodeGenerated = backend.generatedCode.isNotEmpty;
   Expect.equals(
       expectedCodeGenerated,
       isCodeGenerated,
@@ -114,26 +107,5 @@
        expectedCodeGenerated: false,
        expectedOutput: false,
        expectHint: true);
-
-    await test(
-       ['--output-type=dart'],
-       expectedCodeGenerated: false,
-       expectedOutput: false);
-    await test(
-       ['--output-type=dart', '--test-mode'],
-       expectedCodeGenerated: false,
-       expectedOutput: false);
-    await test(
-       ['--output-type=dart', '--generate-code-with-compile-time-errors'],
-       expectedCodeGenerated: false,
-       expectedOutput: false,
-       expectHint: true);
-    await test(
-       ['--output-type=dart',
-        '--generate-code-with-compile-time-errors',
-        '--test-mode'],
-       expectedCodeGenerated: false,
-       expectedOutput: false,
-       expectHint: true);
   });
 }

tests/compiler/dart2js/message_kind_helper.dart

diff --git a/tests/compiler/dart2js/message_kind_helper.dart b/tests/compiler/dart2js/message_kind_helper.dart
index 34e7efe..31094e4 100644
--- a/tests/compiler/dart2js/message_kind_helper.dart
+++ b/tests/compiler/dart2js/message_kind_helper.dart
@@ -10,8 +10,6 @@
 import 'package:compiler/src/commandline_options.dart';
 import 'package:compiler/src/compiler.dart' show
     Compiler;
-import 'package:compiler/src/dart_backend/dart_backend.dart' show
-    DartBackend;
 import 'package:compiler/src/diagnostics/messages.dart' show
     MessageKind,
     MessageTemplate;
@@ -71,21 +69,12 @@
     }
     DiagnosticCollector collector = new DiagnosticCollector();
 
-    bool oldBackendIsDart;
-    if (cachedCompiler != null) {
-      oldBackendIsDart = cachedCompiler.backend is DartBackend;
-    }
-    bool newBackendIsDart = template.options.contains('--output-type=dart');
-
     Compiler compiler = compilerFor(
         memorySourceFiles: example,
         diagnosticHandler: collector,
         options: [Flags.analyzeOnly,
                   Flags.enableExperimentalMirrors]..addAll(template.options),
-        cachedCompiler:
-             // TODO(johnniwinther): Remove this restriction when constant
-             // values can be computed directly from the expressions.
-             oldBackendIsDart == newBackendIsDart ? cachedCompiler : null);
+        cachedCompiler: cachedCompiler);
 
     return compiler.run(Uri.parse('memory:main.dart')).then((_) {
       Iterable<CollectedMessage> messages = collector.filterMessagesByKinds(

tests/compiler/dart2js/mirror_helper_rename_test.dart

diff --git a/tests/compiler/dart2js/mirror_helper_rename_test.dart b/tests/compiler/dart2js/mirror_helper_rename_test.dart
deleted file mode 100644
index 80edc15..0000000
--- a/tests/compiler/dart2js/mirror_helper_rename_test.dart
+++ /dev/null
@@ -1,108 +0,0 @@
-// Copyright (c) 2013, the Dart project authors.  Please see the AUTHORS file
-// for details. All rights reserved. Use of this source code is governed by a
-// BSD-style license that can be found in the LICENSE file.
-
-import "package:expect/expect.dart";
-import 'dart:async';
-import "package:async_helper/async_helper.dart";
-import 'memory_compiler.dart' show runCompiler, OutputCollector;
-import 'package:compiler/src/apiimpl.dart' show
-    CompilerImpl;
-import 'package:compiler/src/tree/tree.dart' show
-    Node;
-import 'package:compiler/src/dart_backend/dart_backend.dart';
-import 'package:compiler/src/mirror_renamer/mirror_renamer.dart';
-
-main() {
-  asyncTest(() async {
-    await testWithMirrorHelperLibrary(minify: true);
-    await testWithMirrorHelperLibrary(minify: false);
-    await testWithoutMirrorHelperLibrary(minify: true);
-    await testWithoutMirrorHelperLibrary(minify: false);
-  });
-}
-
-Future<CompilerImpl> run({OutputCollector outputCollector,
-                      bool useMirrorHelperLibrary: false,
-                      bool minify: false}) async {
-  List<String> options = ['--output-type=dart'];
-  if (minify) {
-    options.add('--minify');
-  }
-  var result = await runCompiler(
-      memorySourceFiles: MEMORY_SOURCE_FILES,
-      outputProvider: outputCollector,
-      options: options,
-      beforeRun: (CompilerImpl compiler) {
-        DartBackend backend = compiler.backend;
-        backend.useMirrorHelperLibrary = useMirrorHelperLibrary;
-      });
-  return result.compiler;
-}
-
-Future testWithMirrorHelperLibrary({bool minify}) async {
-  OutputCollector outputCollector = new OutputCollector();
-  CompilerImpl compiler = await run(
-      outputCollector: outputCollector,
-      useMirrorHelperLibrary: true,
-      minify: minify);
-  DartBackend backend = compiler.backend;
-  MirrorRenamerImpl mirrorRenamer = backend.mirrorRenamer;
-  Map<Node, String> renames = backend.placeholderRenamer.renames;
-  Map<String, String> symbols = mirrorRenamer.symbols;
-
-  Expect.isFalse(null == mirrorRenamer.helperLibrary);
-  Expect.isFalse(null == mirrorRenamer.getNameFunction);
-
-  for (Node n in renames.keys) {
-    if (symbols.containsKey(renames[n])) {
-      if(n.toString() == 'getName') {
-        Expect.equals(
-            MirrorRenamerImpl.MIRROR_HELPER_GET_NAME_FUNCTION,
-            symbols[renames[n]]);
-      } else {
-        Expect.equals(n.toString(), symbols[renames[n]]);
-      }
-    }
-  }
-
-  String output = outputCollector.getOutput('', 'dart');
-  String getNameMatch = MirrorRenamerImpl.MIRROR_HELPER_GET_NAME_FUNCTION;
-  Iterable i = getNameMatch.allMatches(output);
-  print(output);
-  if (minify) {
-    Expect.equals(0, i.length);
-  } else {
-    // Appears twice in code (defined & called).
-    Expect.equals(2, i.length);
-  }
-
-  RegExp mapMatch = new RegExp('const<String,( )?String>');
-  i = mapMatch.allMatches(output);
-  Expect.equals(1, i.length);
-}
-
-Future testWithoutMirrorHelperLibrary({bool minify}) async {
-  CompilerImpl compiler =
-      await run(useMirrorHelperLibrary: false, minify: minify);
-  DartBackend backend = compiler.backend;
-  MirrorRenamer mirrorRenamer = backend.mirrorRenamer;
-
-  Expect.equals(null, mirrorRenamer.helperLibrary);
-  Expect.equals(null, mirrorRenamer.getNameFunction);
-}
-
-const MEMORY_SOURCE_FILES = const <String, String> {
-  'main.dart': """
-import 'dart:mirrors';
-
-class Foo {
-  noSuchMethod(Invocation invocation) {
-    MirrorSystem.getName(invocation.memberName);
-  }
-}
-
-void main() {
-  new Foo().fisk();
-}
-"""};
\ No newline at end of file

tests/compiler/dart2js/mirror_helper_test.dart

diff --git a/tests/compiler/dart2js/mirror_helper_test.dart b/tests/compiler/dart2js/mirror_helper_test.dart
deleted file mode 100644
index ce257c7..0000000
--- a/tests/compiler/dart2js/mirror_helper_test.dart
+++ /dev/null
@@ -1,94 +0,0 @@
-// Copyright (c) 2013, the Dart project authors.  Please see the AUTHORS file
-// for details. All rights reserved. Use of this source code is governed by a
-// BSD-style license that can be found in the LICENSE file.
-
-import "package:expect/expect.dart";
-import 'dart:async';
-import "package:async_helper/async_helper.dart";
-import 'memory_compiler.dart' show runCompiler;
-import 'package:compiler/src/apiimpl.dart' show
-    CompilerImpl;
-import 'package:compiler/src/elements/elements.dart' show
-    Element, LibraryElement, ClassElement;
-import 'package:compiler/src/tree/tree.dart' show
-    Block, ExpressionStatement, FunctionExpression, Node, Send;
-import 'package:compiler/src/dart_backend/dart_backend.dart' show
-    DartBackend, ElementAst;
-import 'package:compiler/src/mirror_renamer/mirror_renamer.dart' show
-    MirrorRenamerImpl;
-
-main() {
-  asyncTest(() async {
-    await testWithMirrorRenaming(minify: true);
-    await testWithMirrorRenaming(minify: false);
-    await testWithoutMirrorRenaming(minify: true);
-    await testWithoutMirrorRenaming(minify: false);
-  });
-}
-
-Future<CompilerImpl> run({useMirrorHelperLibrary: false, minify: false}) async {
-  List<String> options = ['--output-type=dart'];
-  if (minify) {
-    options.add('--minify');
-  }
-  var result = await runCompiler(
-      memorySourceFiles: MEMORY_SOURCE_FILES,
-      options: options,
-      beforeRun: (CompilerImpl compiler) {
-        DartBackend backend = compiler.backend;
-        backend.useMirrorHelperLibrary = useMirrorHelperLibrary;
-      });
-  return result.compiler;
-}
-
-Future testWithMirrorRenaming({bool minify}) async {
-  CompilerImpl compiler =
-      await run(useMirrorHelperLibrary: true, minify: minify);
-  DartBackend backend = compiler.backend;
-  MirrorRenamerImpl mirrorRenamer = backend.mirrorRenamer;
-  Map<Node, String> renames = backend.placeholderRenamer.renames;
-  Iterable<LibraryElement> imports =
-      backend.placeholderRenamer.platformImports.keys;
-
-  FunctionExpression node = backend.memberNodes.values.first.first;
-  Block block = node.body;
-  ExpressionStatement getNameFunctionNode = block.statements.nodes.head;
-  Send send = getNameFunctionNode.expression;
-
-  Expect.equals(renames[mirrorRenamer.getNameFunctionNode.name],
-                renames[send.selector]);
-  Expect.equals("",
-                renames[send.receiver]);
-  Expect.equals(1, imports.length);
-}
-
-Future testWithoutMirrorRenaming({bool minify}) async {
-  CompilerImpl compiler =
-      await run(useMirrorHelperLibrary: false, minify: minify);
-  DartBackend backend = compiler.backend;
-  Map<Node, String> renames = backend.placeholderRenamer.renames;
-  Iterable<LibraryElement> imports =
-      backend.placeholderRenamer.platformImports.keys;
-  FunctionExpression node = backend.memberNodes.values.first.first;
-  Block block = node.body;
-  ExpressionStatement getNameFunctionNode = block.statements.nodes.head;
-  Send send = getNameFunctionNode.expression;
-
-  Expect.isFalse(renames.containsKey(send.selector));
-  Expect.equals(1, imports.length);
-}
-
-const MEMORY_SOURCE_FILES = const <String, String> {
-  'main.dart': """
-import 'dart:mirrors';
-
-class Foo {
-  noSuchMethod(Invocation invocation) {
-    MirrorSystem.getName(invocation.memberName);
-  }
-}
-
-void main() {
-  new Foo().fisk();
-}
-"""};

tests/compiler/dart2js/mirror_helper_unique_minification_test.dart

diff --git a/tests/compiler/dart2js/mirror_helper_unique_minification_test.dart b/tests/compiler/dart2js/mirror_helper_unique_minification_test.dart
deleted file mode 100644
index bec8f9f..0000000
--- a/tests/compiler/dart2js/mirror_helper_unique_minification_test.dart
+++ /dev/null
@@ -1,89 +0,0 @@
-// Copyright (c) 2013, the Dart project authors.  Please see the AUTHORS file
-// for details. All rights reserved. Use of this source code is governed by a
-// BSD-style license that can be found in the LICENSE file.
-
-import "package:expect/expect.dart";
-import 'dart:async';
-import "package:async_helper/async_helper.dart";
-import 'memory_compiler.dart' show runCompiler;
-import 'package:compiler/src/apiimpl.dart' show
-    CompilerImpl;
-import 'package:compiler/src/dart_backend/dart_backend.dart' show
-    DartBackend;
-import 'package:compiler/src/tree/tree.dart' show
-    Identifier, Node, Send;
-import 'package:compiler/src/mirror_renamer/mirror_renamer.dart' show
-    MirrorRenamerImpl;
-
-main() {
-  asyncTest(() async {
-    await testUniqueMinification();
-    await testNoUniqueMinification();
-  });
-}
-
-Future<CompilerImpl> run({useMirrorHelperLibrary: false, minify: false}) async {
-  List<String> options = ['--output-type=dart'];
-  if (minify) {
-    options.add('--minify');
-  }
-  var result = await runCompiler(
-      memorySourceFiles: MEMORY_SOURCE_FILES,
-      options: options,
-      beforeRun: (CompilerImpl compiler) {
-        DartBackend backend = compiler.backend;
-        backend.useMirrorHelperLibrary = useMirrorHelperLibrary;
-      });
-  return result.compiler;
-}
-
-Future testUniqueMinification() async {
-  CompilerImpl compiler = await run(useMirrorHelperLibrary: true, minify: true);
-  DartBackend backend = compiler.backend;
-  MirrorRenamerImpl mirrorRenamer = backend.mirrorRenamer;
-  Map<Node, String> renames = backend.placeholderRenamer.renames;
-  Map<String, String> symbols = mirrorRenamer.symbols;
-
-  // Check that no two different source code names get the same mangled name,
-  // with the exception of MirrorSystem.getName that gets renamed to the same
-  // mangled name as the getNameHelper from _mirror_helper.dart.
-  for (Node node in renames.keys) {
-    Identifier identifier = node.asIdentifier();
-    if (identifier != null) {
-      String source = identifier.source;
-      Send send = mirrorRenamer.mirrorSystemGetNameNodes.first;
-      if (send.selector == node)
-        continue;
-      if (symbols.containsKey(renames[node])) {
-        print(node);
-        Expect.equals(source, symbols[renames[node]]);
-      }
-    }
-  }
-}
-
-Future testNoUniqueMinification() async {
-  CompilerImpl compiler =
-      await run(useMirrorHelperLibrary: false, minify: true);
-  DartBackend backend = compiler.backend;
-  Map<Node, String> renames = backend.placeholderRenamer.renames;
-
-  // 'Foo' appears twice and 'invocation' and 'hest' get the same mangled
-  // name.
-  Expect.equals(renames.values.toSet().length, renames.values.length - 2);
-}
-
-const MEMORY_SOURCE_FILES = const <String, String> {
-  'main.dart': """
-import 'dart:mirrors';
-
-class Foo {
-  noSuchMethod(invocation) {
-    MirrorSystem.getName(null);
-  }
-}
-
-main(hest) {
-  new Foo().fisk();
-}
-"""};

tests/compiler/dart2js/mock_compiler.dart

diff --git a/tests/compiler/dart2js/mock_compiler.dart b/tests/compiler/dart2js/mock_compiler.dart
index cf612bf..5d83c5d 100644
--- a/tests/compiler/dart2js/mock_compiler.dart
+++ b/tests/compiler/dart2js/mock_compiler.dart
@@ -80,7 +80,6 @@
        bool disableTypeInference: false,
        bool analyzeAll: false,
        bool analyzeOnly: false,
-       bool emitJavaScript: true,
        bool preserveComments: false,
        // Our unit tests check code generation output that is
        // affected by inlining support.
@@ -106,7 +105,6 @@
               disableTypeInference: disableTypeInference,
               analyzeAll: analyzeAll,
               analyzeOnly: analyzeOnly,
-              emitJavaScript: emitJavaScript,
               preserveComments: preserveComments,
               trustTypeAnnotations: trustTypeAnnotations,
               trustJSInteropTypeAnnotations: trustJSInteropTypeAnnotations,

tests/compiler/dart2js/platform_consistency_test.dart

diff --git a/tests/compiler/dart2js/platform_consistency_test.dart b/tests/compiler/dart2js/platform_consistency_test.dart
index e946c16..fa50b29 100644
--- a/tests/compiler/dart2js/platform_consistency_test.dart
+++ b/tests/compiler/dart2js/platform_consistency_test.dart
@@ -20,12 +20,8 @@
   Map<String, Uri> shared = await load(
       Uri.base.resolve("sdk/lib/dart_shared.platform"),
       input);
-  Map<String, Uri> dart2dart = await load(
-      Uri.base.resolve("sdk/lib/dart2dart.platform"),
-      input);
   Expect.setEquals(new Set.from(shared.keys), new Set.from(client.keys));
   Expect.setEquals(new Set.from(shared.keys), new Set.from(server.keys));
-  Expect.setEquals(new Set.from(shared.keys), new Set.from(dart2dart.keys));
 
   for (String libraryName in shared.keys) {
     test(Map<String, Uri> m) {
@@ -36,6 +32,5 @@
     }
     test(client);
     test(server);
-    test(dart2dart);
   }
  }

tests/html/html.status

diff --git a/tests/html/html.status b/tests/html/html.status
index ed15788..2612dd1 100644
--- a/tests/html/html.status
+++ b/tests/html/html.status
@@ -50,7 +50,6 @@
 custom/entered_left_view_test/viewless_document: Fail # Polyfill does not handle this
 fontface_test: Fail # Fontface not supported on these.
 custom/attribute_changed_callback_test/unsupported_on_polyfill: Fail # Polyfill does not support
-element_animate_test: Fail # Element.animate not supported on these browsers.
 
 [ $compiler == none && $runtime == dartium && $system == macos]
 canvasrenderingcontext2d_test/drawImage_video_element_dataUrl: Pass,Fail # Issue 11834
@@ -115,6 +114,9 @@
 transition_event_test/functional: Skip # Times out. Issue 22167
 request_animation_frame_test: Skip # Times out. Issue 22167
 custom/*: Pass, Timeout # Issue 26789
+custom_elements_test: Pass, Timeout # Issue 26789
+custom_element_method_clash_test: Pass, Timeout # Issue 26789
+custom_element_name_clash_test: Pass, Timeout # Issue 26789
 
 [$runtime == drt || $runtime == dartium || $runtime == chrome || $runtime == chromeOnAndroid ]
 webgl_1_test: Pass, Fail # Issue 8219
@@ -148,6 +150,7 @@
 transition_event_test/functional: Skip # Times out. Issue 22167
 request_animation_frame_test: Skip # Times out. Issue 22167
 js_util_test/callConstructor: RuntimeError # Issue 26978
+element_animate_test: Fail # Element.animate not supported on these browsers.
 
 [$runtime == ie10 ]
 # IE10 Feature support statuses-
@@ -266,6 +269,7 @@
 canvasrenderingcontext2d_test/drawImage_video_element_dataUrl: Fail # Safari does not support drawImage w/ video element
 element_test: Pass, Fail # Issue 21434
 mediasource_test: Pass, Fail # MediaSource only available on Safari 8 desktop, we can't express that.
+element_animate_test: Fail # Element.animate not supported on these browsers.
 
 # Safari Feature support statuses-
 # All changes should be accompanied by platform support annotation changes.

tests/language/dangling_else_test.dart

diff --git a/tests/language/dangling_else_test.dart b/tests/language/dangling_else_test.dart
index 84bdec8..cad2fc7 100644
--- a/tests/language/dangling_else_test.dart
+++ b/tests/language/dangling_else_test.dart
@@ -1,8 +1,8 @@
 // Copyright (c) 2012, the Dart project authors.  Please see the AUTHORS file
 // for details. All rights reserved. Use of this source code is governed by a
 // BSD-style license that can be found in the LICENSE file.
-// Tests dangling else. The VM should not have any problems, but dart2js or
-// dart2dart could get this wrong.
+// Tests dangling else. The VM should not have any problems, but dart2js could
+// get this wrong.
 
 import "package:expect/expect.dart";
 

tests/language/initializing_formal_access_test.dart

diff --git a/tests/language/initializing_formal_access_test.dart b/tests/language/initializing_formal_access_test.dart
index 3d3ceed..71f445e 100644
--- a/tests/language/initializing_formal_access_test.dart
+++ b/tests/language/initializing_formal_access_test.dart
@@ -3,6 +3,7 @@
 // BSD-style license that can be found in the LICENSE file.
 //
 // DartOptions=--initializing-formal-access
+// VMOptions=--initializing-formal-access
 
 import "package:expect/expect.dart";
 

tests/language/initializing_formal_capture_test.dart

diff --git a/tests/language/initializing_formal_capture_test.dart b/tests/language/initializing_formal_capture_test.dart
index 7ad3f08..8257d4f 100644
--- a/tests/language/initializing_formal_capture_test.dart
+++ b/tests/language/initializing_formal_capture_test.dart
@@ -3,6 +3,7 @@
 // BSD-style license that can be found in the LICENSE file.
 //
 // DartOptions=--initializing-formal-access
+// VMOptions=--initializing-formal-access
 
 import "package:expect/expect.dart";
 

tests/language/initializing_formal_final_test.dart

diff --git a/tests/language/initializing_formal_final_test.dart b/tests/language/initializing_formal_final_test.dart
index 0832ce0..3df4ba5 100644
--- a/tests/language/initializing_formal_final_test.dart
+++ b/tests/language/initializing_formal_final_test.dart
@@ -3,6 +3,7 @@
 // BSD-style license that can be found in the LICENSE file.
 //
 // DartOptions=--initializing-formal-access
+// VMOptions=--initializing-formal-access
 
 import "package:expect/expect.dart";
 

tests/language/initializing_formal_promotion_test.dart

diff --git a/tests/language/initializing_formal_promotion_test.dart b/tests/language/initializing_formal_promotion_test.dart
index b0314e8..5f0a3b6 100644
--- a/tests/language/initializing_formal_promotion_test.dart
+++ b/tests/language/initializing_formal_promotion_test.dart
@@ -3,6 +3,7 @@
 // BSD-style license that can be found in the LICENSE file.
 //
 // DartOptions=--initializing-formal-access
+// VMOptions=--initializing-formal-access
 
 import "package:expect/expect.dart";
 

tests/language/initializing_formal_scope_test.dart

diff --git a/tests/language/initializing_formal_scope_test.dart b/tests/language/initializing_formal_scope_test.dart
index dd85fad..5544a8e 100644
--- a/tests/language/initializing_formal_scope_test.dart
+++ b/tests/language/initializing_formal_scope_test.dart
@@ -3,6 +3,7 @@
 // BSD-style license that can be found in the LICENSE file.
 //
 // DartOptions=--initializing-formal-access
+// VMOptions=--initializing-formal-access
 
 import "package:expect/expect.dart";
 

tests/language/initializing_formal_type_test.dart

diff --git a/tests/language/initializing_formal_type_test.dart b/tests/language/initializing_formal_type_test.dart
index 352eecf..7544f9d 100644
--- a/tests/language/initializing_formal_type_test.dart
+++ b/tests/language/initializing_formal_type_test.dart
@@ -3,6 +3,7 @@
 // BSD-style license that can be found in the LICENSE file.
 //
 // DartOptions=--initializing-formal-access
+// VMOptions=--initializing-formal-access
 
 import "package:expect/expect.dart";
 

tests/language/language.status

diff --git a/tests/language/language.status b/tests/language/language.status
index 64c1c4a..bd95b58 100644
--- a/tests/language/language.status
+++ b/tests/language/language.status
@@ -53,12 +53,6 @@
 generic_methods_function_type_test: CompiletimeError # Issue 25869
 generic_methods_type_expression_test: CompiletimeError # Issue 25869
 
-# Experimental feature: Use initializing formals in initializers and constructor body.
-initializing_formal_access_test: CompiletimeError # Issue 26656
-initializing_formal_capture_test: CompiletimeError # Issue 26656
-initializing_formal_final_test: CompiletimeError # Issue 26656
-initializing_formal_type_test: CompiletimeError # Issue 26656
-
 [ ($compiler == none || $compiler == precompiler || $compiler == dart2app || $compiler == dart2appjit) && ($runtime == vm || $runtime == dart_precompiled || $runtime == dart_app) ]
 
 class_keyword_test/02: MissingCompileTimeError # Issue 13627
@@ -99,6 +93,12 @@
 mirror_in_static_init_test: Fail # Issue 22071
 vm/debug_break_enabled_vm_test/*: Skip # Issue 14651.
 
+# These tests need the flag --initializing-formal-access to pass:
+initializing_formal_access_test: Skip
+initializing_formal_capture_test: Skip
+initializing_formal_final_test: Skip
+initializing_formal_type_test: Skip
+
 # Experimental feature: Syntactic support for generic methods.
 generic_methods_test: RuntimeError # Issue 25869
 generic_functions_test: RuntimeError # Issue 25869
@@ -108,12 +108,6 @@
 generic_methods_function_type_test: RuntimeError # Issue 25869
 generic_methods_type_expression_test: RuntimeError # Issue 25869
 
-# Experimental feature: Use initializing formals in initializers and constructor body.
-initializing_formal_access_test: RuntimeError # Issue 26656
-initializing_formal_capture_test: RuntimeError # Issue 26656
-initializing_formal_final_test: RuntimeError # Issue 26656
-initializing_formal_type_test: RuntimeError # Issue 26656
-
 config_import_test: Skip  # Issue 26250
 
 [ $compiler == none && $runtime == dartium && $system == linux && $arch != x64 ]

tests/standalone/fixed_precision_double_test.dart

diff --git a/tests/standalone/fixed_precision_double_test.dart b/tests/standalone/fixed_precision_double_test.dart
new file mode 100644
index 0000000..350bbc3
--- /dev/null
+++ b/tests/standalone/fixed_precision_double_test.dart
@@ -0,0 +1,82 @@
+// Copyright (c) 2016, the Dart project authors.  Please see the AUTHORS file
+// for details. All rights reserved. Use of this source code is governed by a
+// BSD-style license that can be found in the LICENSE file.
+
+import "package:expect/expect.dart";
+
+bool testAssociativity(Function f) {
+  // Example from https://en.wikipedia.org/wiki/Floating_point
+  // Test that (a + b) + c == a + (b + c).
+  double a = f(1234.567);  // Chop literals.
+  double b = f(45.67834);
+  double c = f(0.0004);
+  double x = (a + b) + c;  // Chop result of multiplication or division only.
+  double y = a + (b + c);
+  print("x: $x");
+  print("y: $y");
+  return x == y;
+}
+
+bool testDistributivity(Function f) {
+  // Example from https://en.wikipedia.org/wiki/Floating_point
+  // Test that (a + b)*c == a*c + b*c.
+  double a = f(1234.567);  // Chop literals.
+  double b = f(1.234567);
+  double c = f(3.333333);
+  double x = f((a + b)*c);  // Chop result of multiplication.
+  double y = f(a*c) + f(b*c);
+  print("x: $x");
+  print("y: $y");
+  return x == y;
+}
+
+// Simulate precision checking with assert.
+assertP(double d) {
+  assert(d == d.p);
+}
+
+bool assertionsEnabled() {
+  try {
+    assert(false);
+    return false;
+  } on AssertionError catch (e) {
+    return true;
+  }
+  return false;
+}
+
+main() {
+  // The getter p keeps only 20 (by default) bits after the decimal point.
+  Expect.equals(0.0, 0.0.p);  // 0.0 has no 1-bit after the decimal point.
+  Expect.equals(1.5, 1.5.p);  // 1.5 has a single 1-bit after the decimal point.
+  Expect.notEquals(1.1, 1.1.p);  // 1.1 has many 1-bits after the decimal point.
+  Expect.notEquals(1/3, (1/3).p);  // 0.33333333... ditto.
+
+  Expect.equals(1.1 + 1/3, 1/3 + 1.1);  // Test addition commutativity.
+  Expect.equals(1.1.p + (1/3).p, (1/3).p + 1.1.p);
+  Expect.equals(1.1 * 1/3, 1/3 * 1.1);  // Test multiplication commutativity.
+  Expect.equals(1.1.p * (1/3).p, (1/3).p * 1.1.p);
+
+  print("Without chopping fractional bits:");
+  Expect.isFalse(testAssociativity((x) => x));
+  Expect.isFalse(testDistributivity((x) => x));
+  print("With chopping fractional bits:");
+  Expect.isTrue(testAssociativity((x) => x.p));
+  Expect.isTrue(testDistributivity((x) => x.p));
+
+  // Check that p works with NaN and Infinity.
+  Expect.isTrue(double.NAN.p.isNaN);
+  Expect.isTrue(double.INFINITY.p.isInfinite);
+  Expect.isFalse(double.INFINITY.p.isNegative);
+  Expect.isTrue(double.NEGATIVE_INFINITY.p.isInfinite);
+  Expect.isTrue(double.NEGATIVE_INFINITY.p.isNegative);
+
+  // Check use of assert to verify precision.
+  if (assertionsEnabled()) {
+    assertP(1.5);
+    assertP(1.1.p);
+    Expect.throws(() => assertP(1.1), (e) => e is AssertionError);
+    assertP(1.23456789.p);
+    Expect.throws(() => assertP(1.23456789), (e) => e is AssertionError);
+  }
+}

tests/standalone/io/socket_source_address_test.dart

diff --git a/tests/standalone/io/socket_source_address_test.dart b/tests/standalone/io/socket_source_address_test.dart
index 4b95433..64dc490 100644
--- a/tests/standalone/io/socket_source_address_test.dart
+++ b/tests/standalone/io/socket_source_address_test.dart
@@ -43,7 +43,8 @@
     await throws(() => connectFunction('127.0.0.1',
                                        server.port,
                                        sourceAddress: sourceAddress),
-                 (e) => e is SocketException);
+                 (e) => e is SocketException &&
+                     e.address == new InternetAddress('8.8.8.8'));
   }
   // Address family mismatch.
   for (sourceAddress in ['::1', InternetAddress.LOOPBACK_IP_V6]) {

tests/standalone/standalone.status

diff --git a/tests/standalone/standalone.status b/tests/standalone/standalone.status
index 3476560..8163f89 100644
--- a/tests/standalone/standalone.status
+++ b/tests/standalone/standalone.status
@@ -78,6 +78,7 @@
 
 [ $compiler == dartanalyzer || $compiler == dart2analyzer ]
 issue14236_test: Skip # Analyzer can't handle Script snapshots.
+fixed_precision_double_test: Skip # double p getter only implemented in VM.
 
 # test issue https://code.google.com/p/dart/issues/detail?id=11518
 io/file_constructor_test: fail
@@ -114,6 +115,7 @@
 pair_location_remapping_test: Skip
 regress_25335_test: Skip # Int64List not supported.
 deferred_transitive_import_error_test: Skip # Contains intentional errors.
+fixed_precision_double_test: Skip # double p getter only implemented in VM.
 
 [ $compiler == dart2js && $cps_ir && $checked ]
 *: Skip # `assert` not implemented

third_party/boringssl/BUILD.generated.gni

diff --git a/third_party/boringssl/BUILD.generated.gni b/third_party/boringssl/BUILD.generated.gni
new file mode 100644
index 0000000..d1ac9be
--- /dev/null
+++ b/third_party/boringssl/BUILD.generated.gni
@@ -0,0 +1,458 @@
+# Copyright (c) 2016 The Chromium Authors. All rights reserved.
+# Use of this source code is governed by a BSD-style license that can be
+# found in the LICENSE file.
+
+# This file is created by generate_build_files.py. Do not edit manually.
+
+crypto_sources = [
+  "err_data.c",
+  "src/crypto/aes/aes.c",
+  "src/crypto/aes/mode_wrappers.c",
+  "src/crypto/asn1/a_bitstr.c",
+  "src/crypto/asn1/a_bool.c",
+  "src/crypto/asn1/a_bytes.c",
+  "src/crypto/asn1/a_d2i_fp.c",
+  "src/crypto/asn1/a_dup.c",
+  "src/crypto/asn1/a_enum.c",
+  "src/crypto/asn1/a_gentm.c",
+  "src/crypto/asn1/a_i2d_fp.c",
+  "src/crypto/asn1/a_int.c",
+  "src/crypto/asn1/a_mbstr.c",
+  "src/crypto/asn1/a_object.c",
+  "src/crypto/asn1/a_octet.c",
+  "src/crypto/asn1/a_print.c",
+  "src/crypto/asn1/a_strnid.c",
+  "src/crypto/asn1/a_time.c",
+  "src/crypto/asn1/a_type.c",
+  "src/crypto/asn1/a_utctm.c",
+  "src/crypto/asn1/a_utf8.c",
+  "src/crypto/asn1/asn1_lib.c",
+  "src/crypto/asn1/asn1_par.c",
+  "src/crypto/asn1/asn_pack.c",
+  "src/crypto/asn1/f_enum.c",
+  "src/crypto/asn1/f_int.c",
+  "src/crypto/asn1/f_string.c",
+  "src/crypto/asn1/t_bitst.c",
+  "src/crypto/asn1/tasn_dec.c",
+  "src/crypto/asn1/tasn_enc.c",
+  "src/crypto/asn1/tasn_fre.c",
+  "src/crypto/asn1/tasn_new.c",
+  "src/crypto/asn1/tasn_typ.c",
+  "src/crypto/asn1/tasn_utl.c",
+  "src/crypto/asn1/x_bignum.c",
+  "src/crypto/asn1/x_long.c",
+  "src/crypto/base64/base64.c",
+  "src/crypto/bio/bio.c",
+  "src/crypto/bio/bio_mem.c",
+  "src/crypto/bio/buffer.c",
+  "src/crypto/bio/connect.c",
+  "src/crypto/bio/fd.c",
+  "src/crypto/bio/file.c",
+  "src/crypto/bio/hexdump.c",
+  "src/crypto/bio/pair.c",
+  "src/crypto/bio/printf.c",
+  "src/crypto/bio/socket.c",
+  "src/crypto/bio/socket_helper.c",
+  "src/crypto/bn/add.c",
+  "src/crypto/bn/asm/x86_64-gcc.c",
+  "src/crypto/bn/bn.c",
+  "src/crypto/bn/bn_asn1.c",
+  "src/crypto/bn/cmp.c",
+  "src/crypto/bn/convert.c",
+  "src/crypto/bn/ctx.c",
+  "src/crypto/bn/div.c",
+  "src/crypto/bn/exponentiation.c",
+  "src/crypto/bn/gcd.c",
+  "src/crypto/bn/generic.c",
+  "src/crypto/bn/kronecker.c",
+  "src/crypto/bn/montgomery.c",
+  "src/crypto/bn/mul.c",
+  "src/crypto/bn/prime.c",
+  "src/crypto/bn/random.c",
+  "src/crypto/bn/rsaz_exp.c",
+  "src/crypto/bn/shift.c",
+  "src/crypto/bn/sqrt.c",
+  "src/crypto/buf/buf.c",
+  "src/crypto/bytestring/asn1_compat.c",
+  "src/crypto/bytestring/ber.c",
+  "src/crypto/bytestring/cbb.c",
+  "src/crypto/bytestring/cbs.c",
+  "src/crypto/chacha/chacha.c",
+  "src/crypto/cipher/aead.c",
+  "src/crypto/cipher/cipher.c",
+  "src/crypto/cipher/derive_key.c",
+  "src/crypto/cipher/e_aes.c",
+  "src/crypto/cipher/e_chacha20poly1305.c",
+  "src/crypto/cipher/e_des.c",
+  "src/crypto/cipher/e_null.c",
+  "src/crypto/cipher/e_rc2.c",
+  "src/crypto/cipher/e_rc4.c",
+  "src/crypto/cipher/e_ssl3.c",
+  "src/crypto/cipher/e_tls.c",
+  "src/crypto/cipher/tls_cbc.c",
+  "src/crypto/cmac/cmac.c",
+  "src/crypto/conf/conf.c",
+  "src/crypto/cpu-aarch64-linux.c",
+  "src/crypto/cpu-arm-linux.c",
+  "src/crypto/cpu-arm.c",
+  "src/crypto/cpu-intel.c",
+  "src/crypto/crypto.c",
+  "src/crypto/curve25519/curve25519.c",
+  "src/crypto/curve25519/spake25519.c",
+  "src/crypto/curve25519/x25519-x86_64.c",
+  "src/crypto/des/des.c",
+  "src/crypto/dh/check.c",
+  "src/crypto/dh/dh.c",
+  "src/crypto/dh/dh_asn1.c",
+  "src/crypto/dh/params.c",
+  "src/crypto/digest/digest.c",
+  "src/crypto/digest/digests.c",
+  "src/crypto/dsa/dsa.c",
+  "src/crypto/dsa/dsa_asn1.c",
+  "src/crypto/ec/ec.c",
+  "src/crypto/ec/ec_asn1.c",
+  "src/crypto/ec/ec_key.c",
+  "src/crypto/ec/ec_montgomery.c",
+  "src/crypto/ec/oct.c",
+  "src/crypto/ec/p224-64.c",
+  "src/crypto/ec/p256-64.c",
+  "src/crypto/ec/p256-x86_64.c",
+  "src/crypto/ec/simple.c",
+  "src/crypto/ec/util-64.c",
+  "src/crypto/ec/wnaf.c",
+  "src/crypto/ecdh/ecdh.c",
+  "src/crypto/ecdsa/ecdsa.c",
+  "src/crypto/ecdsa/ecdsa_asn1.c",
+  "src/crypto/engine/engine.c",
+  "src/crypto/err/err.c",
+  "src/crypto/evp/digestsign.c",
+  "src/crypto/evp/evp.c",
+  "src/crypto/evp/evp_asn1.c",
+  "src/crypto/evp/evp_ctx.c",
+  "src/crypto/evp/p_dsa_asn1.c",
+  "src/crypto/evp/p_ec.c",
+  "src/crypto/evp/p_ec_asn1.c",
+  "src/crypto/evp/p_rsa.c",
+  "src/crypto/evp/p_rsa_asn1.c",
+  "src/crypto/evp/pbkdf.c",
+  "src/crypto/evp/print.c",
+  "src/crypto/evp/sign.c",
+  "src/crypto/ex_data.c",
+  "src/crypto/hkdf/hkdf.c",
+  "src/crypto/hmac/hmac.c",
+  "src/crypto/lhash/lhash.c",
+  "src/crypto/md4/md4.c",
+  "src/crypto/md5/md5.c",
+  "src/crypto/mem.c",
+  "src/crypto/modes/cbc.c",
+  "src/crypto/modes/cfb.c",
+  "src/crypto/modes/ctr.c",
+  "src/crypto/modes/gcm.c",
+  "src/crypto/modes/ofb.c",
+  "src/crypto/newhope/error_correction.c",
+  "src/crypto/newhope/newhope.c",
+  "src/crypto/newhope/ntt.c",
+  "src/crypto/newhope/poly.c",
+  "src/crypto/newhope/precomp.c",
+  "src/crypto/newhope/reduce.c",
+  "src/crypto/obj/obj.c",
+  "src/crypto/obj/obj_xref.c",
+  "src/crypto/pem/pem_all.c",
+  "src/crypto/pem/pem_info.c",
+  "src/crypto/pem/pem_lib.c",
+  "src/crypto/pem/pem_oth.c",
+  "src/crypto/pem/pem_pk8.c",
+  "src/crypto/pem/pem_pkey.c",
+  "src/crypto/pem/pem_x509.c",
+  "src/crypto/pem/pem_xaux.c",
+  "src/crypto/pkcs8/p5_pbe.c",
+  "src/crypto/pkcs8/p5_pbev2.c",
+  "src/crypto/pkcs8/p8_pkey.c",
+  "src/crypto/pkcs8/pkcs8.c",
+  "src/crypto/poly1305/poly1305.c",
+  "src/crypto/poly1305/poly1305_arm.c",
+  "src/crypto/poly1305/poly1305_vec.c",
+  "src/crypto/rand/deterministic.c",
+  "src/crypto/rand/rand.c",
+  "src/crypto/rand/urandom.c",
+  "src/crypto/rand/windows.c",
+  "src/crypto/rc4/rc4.c",
+  "src/crypto/refcount_c11.c",
+  "src/crypto/refcount_lock.c",
+  "src/crypto/rsa/blinding.c",
+  "src/crypto/rsa/padding.c",
+  "src/crypto/rsa/rsa.c",
+  "src/crypto/rsa/rsa_asn1.c",
+  "src/crypto/rsa/rsa_impl.c",
+  "src/crypto/sha/sha1.c",
+  "src/crypto/sha/sha256.c",
+  "src/crypto/sha/sha512.c",
+  "src/crypto/stack/stack.c",
+  "src/crypto/thread.c",
+  "src/crypto/thread_none.c",
+  "src/crypto/thread_pthread.c",
+  "src/crypto/thread_win.c",
+  "src/crypto/time_support.c",
+  "src/crypto/x509/a_digest.c",
+  "src/crypto/x509/a_sign.c",
+  "src/crypto/x509/a_strex.c",
+  "src/crypto/x509/a_verify.c",
+  "src/crypto/x509/algorithm.c",
+  "src/crypto/x509/asn1_gen.c",
+  "src/crypto/x509/by_dir.c",
+  "src/crypto/x509/by_file.c",
+  "src/crypto/x509/i2d_pr.c",
+  "src/crypto/x509/pkcs7.c",
+  "src/crypto/x509/rsa_pss.c",
+  "src/crypto/x509/t_crl.c",
+  "src/crypto/x509/t_req.c",
+  "src/crypto/x509/t_x509.c",
+  "src/crypto/x509/t_x509a.c",
+  "src/crypto/x509/x509.c",
+  "src/crypto/x509/x509_att.c",
+  "src/crypto/x509/x509_cmp.c",
+  "src/crypto/x509/x509_d2.c",
+  "src/crypto/x509/x509_def.c",
+  "src/crypto/x509/x509_ext.c",
+  "src/crypto/x509/x509_lu.c",
+  "src/crypto/x509/x509_obj.c",
+  "src/crypto/x509/x509_r2x.c",
+  "src/crypto/x509/x509_req.c",
+  "src/crypto/x509/x509_set.c",
+  "src/crypto/x509/x509_trs.c",
+  "src/crypto/x509/x509_txt.c",
+  "src/crypto/x509/x509_v3.c",
+  "src/crypto/x509/x509_vfy.c",
+  "src/crypto/x509/x509_vpm.c",
+  "src/crypto/x509/x509cset.c",
+  "src/crypto/x509/x509name.c",
+  "src/crypto/x509/x509rset.c",
+  "src/crypto/x509/x509spki.c",
+  "src/crypto/x509/x509type.c",
+  "src/crypto/x509/x_algor.c",
+  "src/crypto/x509/x_all.c",
+  "src/crypto/x509/x_attrib.c",
+  "src/crypto/x509/x_crl.c",
+  "src/crypto/x509/x_exten.c",
+  "src/crypto/x509/x_info.c",
+  "src/crypto/x509/x_name.c",
+  "src/crypto/x509/x_pkey.c",
+  "src/crypto/x509/x_pubkey.c",
+  "src/crypto/x509/x_req.c",
+  "src/crypto/x509/x_sig.c",
+  "src/crypto/x509/x_spki.c",
+  "src/crypto/x509/x_val.c",
+  "src/crypto/x509/x_x509.c",
+  "src/crypto/x509/x_x509a.c",
+  "src/crypto/x509v3/pcy_cache.c",
+  "src/crypto/x509v3/pcy_data.c",
+  "src/crypto/x509v3/pcy_lib.c",
+  "src/crypto/x509v3/pcy_map.c",
+  "src/crypto/x509v3/pcy_node.c",
+  "src/crypto/x509v3/pcy_tree.c",
+  "src/crypto/x509v3/v3_akey.c",
+  "src/crypto/x509v3/v3_akeya.c",
+  "src/crypto/x509v3/v3_alt.c",
+  "src/crypto/x509v3/v3_bcons.c",
+  "src/crypto/x509v3/v3_bitst.c",
+  "src/crypto/x509v3/v3_conf.c",
+  "src/crypto/x509v3/v3_cpols.c",
+  "src/crypto/x509v3/v3_crld.c",
+  "src/crypto/x509v3/v3_enum.c",
+  "src/crypto/x509v3/v3_extku.c",
+  "src/crypto/x509v3/v3_genn.c",
+  "src/crypto/x509v3/v3_ia5.c",
+  "src/crypto/x509v3/v3_info.c",
+  "src/crypto/x509v3/v3_int.c",
+  "src/crypto/x509v3/v3_lib.c",
+  "src/crypto/x509v3/v3_ncons.c",
+  "src/crypto/x509v3/v3_pci.c",
+  "src/crypto/x509v3/v3_pcia.c",
+  "src/crypto/x509v3/v3_pcons.c",
+  "src/crypto/x509v3/v3_pku.c",
+  "src/crypto/x509v3/v3_pmaps.c",
+  "src/crypto/x509v3/v3_prn.c",
+  "src/crypto/x509v3/v3_purp.c",
+  "src/crypto/x509v3/v3_skey.c",
+  "src/crypto/x509v3/v3_sxnet.c",
+  "src/crypto/x509v3/v3_utl.c",
+]
+
+ssl_sources = [
+  "src/ssl/custom_extensions.c",
+  "src/ssl/d1_both.c",
+  "src/ssl/d1_lib.c",
+  "src/ssl/d1_pkt.c",
+  "src/ssl/d1_srtp.c",
+  "src/ssl/dtls_method.c",
+  "src/ssl/dtls_record.c",
+  "src/ssl/handshake_client.c",
+  "src/ssl/handshake_server.c",
+  "src/ssl/s3_both.c",
+  "src/ssl/s3_enc.c",
+  "src/ssl/s3_lib.c",
+  "src/ssl/s3_pkt.c",
+  "src/ssl/ssl_aead_ctx.c",
+  "src/ssl/ssl_asn1.c",
+  "src/ssl/ssl_buffer.c",
+  "src/ssl/ssl_cert.c",
+  "src/ssl/ssl_cipher.c",
+  "src/ssl/ssl_ecdh.c",
+  "src/ssl/ssl_file.c",
+  "src/ssl/ssl_lib.c",
+  "src/ssl/ssl_rsa.c",
+  "src/ssl/ssl_session.c",
+  "src/ssl/ssl_stat.c",
+  "src/ssl/t1_enc.c",
+  "src/ssl/t1_lib.c",
+  "src/ssl/tls_method.c",
+  "src/ssl/tls_record.c",
+]
+
+crypto_sources_linux_aarch64 = [
+  "linux-aarch64/crypto/aes/aesv8-armx64.S",
+  "linux-aarch64/crypto/bn/armv8-mont.S",
+  "linux-aarch64/crypto/chacha/chacha-armv8.S",
+  "linux-aarch64/crypto/modes/ghashv8-armx64.S",
+  "linux-aarch64/crypto/sha/sha1-armv8.S",
+  "linux-aarch64/crypto/sha/sha256-armv8.S",
+  "linux-aarch64/crypto/sha/sha512-armv8.S",
+]
+
+crypto_sources_linux_arm = [
+  "linux-arm/crypto/aes/aes-armv4.S",
+  "linux-arm/crypto/aes/aesv8-armx32.S",
+  "linux-arm/crypto/aes/bsaes-armv7.S",
+  "linux-arm/crypto/bn/armv4-mont.S",
+  "linux-arm/crypto/chacha/chacha-armv4.S",
+  "linux-arm/crypto/modes/ghash-armv4.S",
+  "linux-arm/crypto/modes/ghashv8-armx32.S",
+  "linux-arm/crypto/sha/sha1-armv4-large.S",
+  "linux-arm/crypto/sha/sha256-armv4.S",
+  "linux-arm/crypto/sha/sha512-armv4.S",
+  "src/crypto/curve25519/asm/x25519-asm-arm.S",
+  "src/crypto/poly1305/poly1305_arm_asm.S",
+]
+
+crypto_sources_linux_x86 = [
+  "linux-x86/crypto/aes/aes-586.S",
+  "linux-x86/crypto/aes/aesni-x86.S",
+  "linux-x86/crypto/aes/vpaes-x86.S",
+  "linux-x86/crypto/bn/bn-586.S",
+  "linux-x86/crypto/bn/co-586.S",
+  "linux-x86/crypto/bn/x86-mont.S",
+  "linux-x86/crypto/chacha/chacha-x86.S",
+  "linux-x86/crypto/md5/md5-586.S",
+  "linux-x86/crypto/modes/ghash-x86.S",
+  "linux-x86/crypto/rc4/rc4-586.S",
+  "linux-x86/crypto/sha/sha1-586.S",
+  "linux-x86/crypto/sha/sha256-586.S",
+  "linux-x86/crypto/sha/sha512-586.S",
+]
+
+crypto_sources_linux_x86_64 = [
+  "linux-x86_64/crypto/aes/aes-x86_64.S",
+  "linux-x86_64/crypto/aes/aesni-x86_64.S",
+  "linux-x86_64/crypto/aes/bsaes-x86_64.S",
+  "linux-x86_64/crypto/aes/vpaes-x86_64.S",
+  "linux-x86_64/crypto/bn/rsaz-avx2.S",
+  "linux-x86_64/crypto/bn/rsaz-x86_64.S",
+  "linux-x86_64/crypto/bn/x86_64-mont.S",
+  "linux-x86_64/crypto/bn/x86_64-mont5.S",
+  "linux-x86_64/crypto/chacha/chacha-x86_64.S",
+  "linux-x86_64/crypto/ec/p256-x86_64-asm.S",
+  "linux-x86_64/crypto/md5/md5-x86_64.S",
+  "linux-x86_64/crypto/modes/aesni-gcm-x86_64.S",
+  "linux-x86_64/crypto/modes/ghash-x86_64.S",
+  "linux-x86_64/crypto/rand/rdrand-x86_64.S",
+  "linux-x86_64/crypto/rc4/rc4-x86_64.S",
+  "linux-x86_64/crypto/sha/sha1-x86_64.S",
+  "linux-x86_64/crypto/sha/sha256-x86_64.S",
+  "linux-x86_64/crypto/sha/sha512-x86_64.S",
+  "src/crypto/curve25519/asm/x25519-asm-x86_64.S",
+]
+
+crypto_sources_mac_x86 = [
+  "mac-x86/crypto/aes/aes-586.S",
+  "mac-x86/crypto/aes/aesni-x86.S",
+  "mac-x86/crypto/aes/vpaes-x86.S",
+  "mac-x86/crypto/bn/bn-586.S",
+  "mac-x86/crypto/bn/co-586.S",
+  "mac-x86/crypto/bn/x86-mont.S",
+  "mac-x86/crypto/chacha/chacha-x86.S",
+  "mac-x86/crypto/md5/md5-586.S",
+  "mac-x86/crypto/modes/ghash-x86.S",
+  "mac-x86/crypto/rc4/rc4-586.S",
+  "mac-x86/crypto/sha/sha1-586.S",
+  "mac-x86/crypto/sha/sha256-586.S",
+  "mac-x86/crypto/sha/sha512-586.S",
+]
+
+crypto_sources_mac_x86_64 = [
+  "mac-x86_64/crypto/aes/aes-x86_64.S",
+  "mac-x86_64/crypto/aes/aesni-x86_64.S",
+  "mac-x86_64/crypto/aes/bsaes-x86_64.S",
+  "mac-x86_64/crypto/aes/vpaes-x86_64.S",
+  "mac-x86_64/crypto/bn/rsaz-avx2.S",
+  "mac-x86_64/crypto/bn/rsaz-x86_64.S",
+  "mac-x86_64/crypto/bn/x86_64-mont.S",
+  "mac-x86_64/crypto/bn/x86_64-mont5.S",
+  "mac-x86_64/crypto/chacha/chacha-x86_64.S",
+  "mac-x86_64/crypto/ec/p256-x86_64-asm.S",
+  "mac-x86_64/crypto/md5/md5-x86_64.S",
+  "mac-x86_64/crypto/modes/aesni-gcm-x86_64.S",
+  "mac-x86_64/crypto/modes/ghash-x86_64.S",
+  "mac-x86_64/crypto/rand/rdrand-x86_64.S",
+  "mac-x86_64/crypto/rc4/rc4-x86_64.S",
+  "mac-x86_64/crypto/sha/sha1-x86_64.S",
+  "mac-x86_64/crypto/sha/sha256-x86_64.S",
+  "mac-x86_64/crypto/sha/sha512-x86_64.S",
+  "src/crypto/curve25519/asm/x25519-asm-x86_64.S",
+]
+
+crypto_sources_win_x86 = [
+  "win-x86/crypto/aes/aes-586.asm",
+  "win-x86/crypto/aes/aesni-x86.asm",
+  "win-x86/crypto/aes/vpaes-x86.asm",
+  "win-x86/crypto/bn/bn-586.asm",
+  "win-x86/crypto/bn/co-586.asm",
+  "win-x86/crypto/bn/x86-mont.asm",
+  "win-x86/crypto/chacha/chacha-x86.asm",
+  "win-x86/crypto/md5/md5-586.asm",
+  "win-x86/crypto/modes/ghash-x86.asm",
+  "win-x86/crypto/rc4/rc4-586.asm",
+  "win-x86/crypto/sha/sha1-586.asm",
+  "win-x86/crypto/sha/sha256-586.asm",
+  "win-x86/crypto/sha/sha512-586.asm",
+]
+
+crypto_sources_win_x86_64 = [
+  "win-x86_64/crypto/aes/aes-x86_64.asm",
+  "win-x86_64/crypto/aes/aesni-x86_64.asm",
+  "win-x86_64/crypto/aes/bsaes-x86_64.asm",
+  "win-x86_64/crypto/aes/vpaes-x86_64.asm",
+  "win-x86_64/crypto/bn/rsaz-avx2.asm",
+  "win-x86_64/crypto/bn/rsaz-x86_64.asm",
+  "win-x86_64/crypto/bn/x86_64-mont.asm",
+  "win-x86_64/crypto/bn/x86_64-mont5.asm",
+  "win-x86_64/crypto/chacha/chacha-x86_64.asm",
+  "win-x86_64/crypto/ec/p256-x86_64-asm.asm",
+  "win-x86_64/crypto/md5/md5-x86_64.asm",
+  "win-x86_64/crypto/modes/aesni-gcm-x86_64.asm",
+  "win-x86_64/crypto/modes/ghash-x86_64.asm",
+  "win-x86_64/crypto/rand/rdrand-x86_64.asm",
+  "win-x86_64/crypto/rc4/rc4-x86_64.asm",
+  "win-x86_64/crypto/sha/sha1-x86_64.asm",
+  "win-x86_64/crypto/sha/sha256-x86_64.asm",
+  "win-x86_64/crypto/sha/sha512-x86_64.asm",
+]
+
+fuzzers = [
+  "cert",
+  "client",
+  "pkcs8",
+  "privkey",
+  "read_pem",
+  "server",
+  "spki",
+]

third_party/boringssl/BUILD.generated_tests.gni

diff --git a/third_party/boringssl/BUILD.generated_tests.gni b/third_party/boringssl/BUILD.generated_tests.gni
new file mode 100644
index 0000000..16bddff
--- /dev/null
+++ b/third_party/boringssl/BUILD.generated_tests.gni
@@ -0,0 +1,584 @@
+# Copyright (c) 2016 The Chromium Authors. All rights reserved.
+# Use of this source code is governed by a BSD-style license that can be
+# found in the LICENSE file.
+
+# This file is created by generate_build_files.py. Do not edit manually.
+
+_test_support_sources = [
+  "src/crypto/test/file_test.cc",
+  "src/crypto/test/file_test.h",
+  "src/crypto/test/malloc.cc",
+  "src/crypto/test/scoped_types.h",
+  "src/crypto/test/test_util.cc",
+  "src/crypto/test/test_util.h",
+  "src/ssl/test/async_bio.h",
+  "src/ssl/test/packeted_bio.h",
+  "src/ssl/test/scoped_types.h",
+  "src/ssl/test/test_config.h",
+]
+
+template("create_tests") {
+  executable("boringssl_aes_test") {
+    sources = [
+      "src/crypto/aes/aes_test.cc",
+    ]
+    sources += _test_support_sources
+    if (defined(invoker.configs_exclude)) {
+      configs -= invoker.configs_exclude
+    }
+    configs += invoker.configs
+    deps = invoker.deps
+  }
+
+  executable("boringssl_asn1_test") {
+    sources = [
+      "src/crypto/asn1/asn1_test.cc",
+    ]
+    sources += _test_support_sources
+    if (defined(invoker.configs_exclude)) {
+      configs -= invoker.configs_exclude
+    }
+    configs += invoker.configs
+    deps = invoker.deps
+  }
+
+  executable("boringssl_base64_test") {
+    sources = [
+      "src/crypto/base64/base64_test.cc",
+    ]
+    sources += _test_support_sources
+    if (defined(invoker.configs_exclude)) {
+      configs -= invoker.configs_exclude
+    }
+    configs += invoker.configs
+    deps = invoker.deps
+  }
+
+  executable("boringssl_bio_test") {
+    sources = [
+      "src/crypto/bio/bio_test.cc",
+    ]
+    sources += _test_support_sources
+    if (defined(invoker.configs_exclude)) {
+      configs -= invoker.configs_exclude
+    }
+    configs += invoker.configs
+    deps = invoker.deps
+  }
+
+  executable("boringssl_bn_test") {
+    sources = [
+      "src/crypto/bn/bn_test.cc",
+    ]
+    sources += _test_support_sources
+    if (defined(invoker.configs_exclude)) {
+      configs -= invoker.configs_exclude
+    }
+    configs += invoker.configs
+    deps = invoker.deps
+  }
+
+  executable("boringssl_bytestring_test") {
+    sources = [
+      "src/crypto/bytestring/bytestring_test.cc",
+    ]
+    sources += _test_support_sources
+    if (defined(invoker.configs_exclude)) {
+      configs -= invoker.configs_exclude
+    }
+    configs += invoker.configs
+    deps = invoker.deps
+  }
+
+  executable("boringssl_chacha_test") {
+    sources = [
+      "src/crypto/chacha/chacha_test.cc",
+    ]
+    sources += _test_support_sources
+    if (defined(invoker.configs_exclude)) {
+      configs -= invoker.configs_exclude
+    }
+    configs += invoker.configs
+    deps = invoker.deps
+  }
+
+  executable("boringssl_aead_test") {
+    sources = [
+      "src/crypto/cipher/aead_test.cc",
+    ]
+    sources += _test_support_sources
+    if (defined(invoker.configs_exclude)) {
+      configs -= invoker.configs_exclude
+    }
+    configs += invoker.configs
+    deps = invoker.deps
+  }
+
+  executable("boringssl_cipher_test") {
+    sources = [
+      "src/crypto/cipher/cipher_test.cc",
+    ]
+    sources += _test_support_sources
+    if (defined(invoker.configs_exclude)) {
+      configs -= invoker.configs_exclude
+    }
+    configs += invoker.configs
+    deps = invoker.deps
+  }
+
+  executable("boringssl_cmac_test") {
+    sources = [
+      "src/crypto/cmac/cmac_test.cc",
+    ]
+    sources += _test_support_sources
+    if (defined(invoker.configs_exclude)) {
+      configs -= invoker.configs_exclude
+    }
+    configs += invoker.configs
+    deps = invoker.deps
+  }
+
+  executable("boringssl_constant_time_test") {
+    sources = [
+      "src/crypto/constant_time_test.c",
+    ]
+    sources += _test_support_sources
+    if (defined(invoker.configs_exclude)) {
+      configs -= invoker.configs_exclude
+    }
+    configs += invoker.configs
+    deps = invoker.deps
+  }
+
+  executable("boringssl_ed25519_test") {
+    sources = [
+      "src/crypto/curve25519/ed25519_test.cc",
+    ]
+    sources += _test_support_sources
+    if (defined(invoker.configs_exclude)) {
+      configs -= invoker.configs_exclude
+    }
+    configs += invoker.configs
+    deps = invoker.deps
+  }
+
+  executable("boringssl_spake25519_test") {
+    sources = [
+      "src/crypto/curve25519/spake25519_test.cc",
+    ]
+    sources += _test_support_sources
+    if (defined(invoker.configs_exclude)) {
+      configs -= invoker.configs_exclude
+    }
+    configs += invoker.configs
+    deps = invoker.deps
+  }
+
+  executable("boringssl_x25519_test") {
+    sources = [
+      "src/crypto/curve25519/x25519_test.cc",
+    ]
+    sources += _test_support_sources
+    if (defined(invoker.configs_exclude)) {
+      configs -= invoker.configs_exclude
+    }
+    configs += invoker.configs
+    deps = invoker.deps
+  }
+
+  executable("boringssl_dh_test") {
+    sources = [
+      "src/crypto/dh/dh_test.cc",
+    ]
+    sources += _test_support_sources
+    if (defined(invoker.configs_exclude)) {
+      configs -= invoker.configs_exclude
+    }
+    configs += invoker.configs
+    deps = invoker.deps
+  }
+
+  executable("boringssl_digest_test") {
+    sources = [
+      "src/crypto/digest/digest_test.cc",
+    ]
+    sources += _test_support_sources
+    if (defined(invoker.configs_exclude)) {
+      configs -= invoker.configs_exclude
+    }
+    configs += invoker.configs
+    deps = invoker.deps
+  }
+
+  executable("boringssl_dsa_test") {
+    sources = [
+      "src/crypto/dsa/dsa_test.c",
+    ]
+    sources += _test_support_sources
+    if (defined(invoker.configs_exclude)) {
+      configs -= invoker.configs_exclude
+    }
+    configs += invoker.configs
+    deps = invoker.deps
+  }
+
+  executable("boringssl_ec_test") {
+    sources = [
+      "src/crypto/ec/ec_test.cc",
+    ]
+    sources += _test_support_sources
+    if (defined(invoker.configs_exclude)) {
+      configs -= invoker.configs_exclude
+    }
+    configs += invoker.configs
+    deps = invoker.deps
+  }
+
+  executable("boringssl_example_mul") {
+    sources = [
+      "src/crypto/ec/example_mul.c",
+    ]
+    sources += _test_support_sources
+    if (defined(invoker.configs_exclude)) {
+      configs -= invoker.configs_exclude
+    }
+    configs += invoker.configs
+    deps = invoker.deps
+  }
+
+  executable("boringssl_ecdsa_test") {
+    sources = [
+      "src/crypto/ecdsa/ecdsa_test.cc",
+    ]
+    sources += _test_support_sources
+    if (defined(invoker.configs_exclude)) {
+      configs -= invoker.configs_exclude
+    }
+    configs += invoker.configs
+    deps = invoker.deps
+  }
+
+  executable("boringssl_err_test") {
+    sources = [
+      "src/crypto/err/err_test.cc",
+    ]
+    sources += _test_support_sources
+    if (defined(invoker.configs_exclude)) {
+      configs -= invoker.configs_exclude
+    }
+    configs += invoker.configs
+    deps = invoker.deps
+  }
+
+  executable("boringssl_evp_extra_test") {
+    sources = [
+      "src/crypto/evp/evp_extra_test.cc",
+    ]
+    sources += _test_support_sources
+    if (defined(invoker.configs_exclude)) {
+      configs -= invoker.configs_exclude
+    }
+    configs += invoker.configs
+    deps = invoker.deps
+  }
+
+  executable("boringssl_evp_test") {
+    sources = [
+      "src/crypto/evp/evp_test.cc",
+    ]
+    sources += _test_support_sources
+    if (defined(invoker.configs_exclude)) {
+      configs -= invoker.configs_exclude
+    }
+    configs += invoker.configs
+    deps = invoker.deps
+  }
+
+  executable("boringssl_pbkdf_test") {
+    sources = [
+      "src/crypto/evp/pbkdf_test.cc",
+    ]
+    sources += _test_support_sources
+    if (defined(invoker.configs_exclude)) {
+      configs -= invoker.configs_exclude
+    }
+    configs += invoker.configs
+    deps = invoker.deps
+  }
+
+  executable("boringssl_hkdf_test") {
+    sources = [
+      "src/crypto/hkdf/hkdf_test.c",
+    ]
+    sources += _test_support_sources
+    if (defined(invoker.configs_exclude)) {
+      configs -= invoker.configs_exclude
+    }
+    configs += invoker.configs
+    deps = invoker.deps
+  }
+
+  executable("boringssl_hmac_test") {
+    sources = [
+      "src/crypto/hmac/hmac_test.cc",
+    ]
+    sources += _test_support_sources
+    if (defined(invoker.configs_exclude)) {
+      configs -= invoker.configs_exclude
+    }
+    configs += invoker.configs
+    deps = invoker.deps
+  }
+
+  executable("boringssl_lhash_test") {
+    sources = [
+      "src/crypto/lhash/lhash_test.c",
+    ]
+    sources += _test_support_sources
+    if (defined(invoker.configs_exclude)) {
+      configs -= invoker.configs_exclude
+    }
+    configs += invoker.configs
+    deps = invoker.deps
+  }
+
+  executable("boringssl_gcm_test") {
+    sources = [
+      "src/crypto/modes/gcm_test.c",
+    ]
+    sources += _test_support_sources
+    if (defined(invoker.configs_exclude)) {
+      configs -= invoker.configs_exclude
+    }
+    configs += invoker.configs
+    deps = invoker.deps
+  }
+
+  executable("boringssl_newhope_statistical_test") {
+    sources = [
+      "src/crypto/newhope/newhope_statistical_test.cc",
+    ]
+    sources += _test_support_sources
+    if (defined(invoker.configs_exclude)) {
+      configs -= invoker.configs_exclude
+    }
+    configs += invoker.configs
+    deps = invoker.deps
+  }
+
+  executable("boringssl_newhope_test") {
+    sources = [
+      "src/crypto/newhope/newhope_test.cc",
+    ]
+    sources += _test_support_sources
+    if (defined(invoker.configs_exclude)) {
+      configs -= invoker.configs_exclude
+    }
+    configs += invoker.configs
+    deps = invoker.deps
+  }
+
+  executable("boringssl_newhope_vectors_test") {
+    sources = [
+      "src/crypto/newhope/newhope_vectors_test.cc",
+    ]
+    sources += _test_support_sources
+    if (defined(invoker.configs_exclude)) {
+      configs -= invoker.configs_exclude
+    }
+    configs += invoker.configs
+    deps = invoker.deps
+  }
+
+  executable("boringssl_obj_test") {
+    sources = [
+      "src/crypto/obj/obj_test.cc",
+    ]
+    sources += _test_support_sources
+    if (defined(invoker.configs_exclude)) {
+      configs -= invoker.configs_exclude
+    }
+    configs += invoker.configs
+    deps = invoker.deps
+  }
+
+  executable("boringssl_pkcs12_test") {
+    sources = [
+      "src/crypto/pkcs8/pkcs12_test.cc",
+    ]
+    sources += _test_support_sources
+    if (defined(invoker.configs_exclude)) {
+      configs -= invoker.configs_exclude
+    }
+    configs += invoker.configs
+    deps = invoker.deps
+  }
+
+  executable("boringssl_pkcs8_test") {
+    sources = [
+      "src/crypto/pkcs8/pkcs8_test.cc",
+    ]
+    sources += _test_support_sources
+    if (defined(invoker.configs_exclude)) {
+      configs -= invoker.configs_exclude
+    }
+    configs += invoker.configs
+    deps = invoker.deps
+  }
+
+  executable("boringssl_poly1305_test") {
+    sources = [
+      "src/crypto/poly1305/poly1305_test.cc",
+    ]
+    sources += _test_support_sources
+    if (defined(invoker.configs_exclude)) {
+      configs -= invoker.configs_exclude
+    }
+    configs += invoker.configs
+    deps = invoker.deps
+  }
+
+  executable("boringssl_refcount_test") {
+    sources = [
+      "src/crypto/refcount_test.c",
+    ]
+    sources += _test_support_sources
+    if (defined(invoker.configs_exclude)) {
+      configs -= invoker.configs_exclude
+    }
+    configs += invoker.configs
+    deps = invoker.deps
+  }
+
+  executable("boringssl_rsa_test") {
+    sources = [
+      "src/crypto/rsa/rsa_test.cc",
+    ]
+    sources += _test_support_sources
+    if (defined(invoker.configs_exclude)) {
+      configs -= invoker.configs_exclude
+    }
+    configs += invoker.configs
+    deps = invoker.deps
+  }
+
+  executable("boringssl_thread_test") {
+    sources = [
+      "src/crypto/thread_test.c",
+    ]
+    sources += _test_support_sources
+    if (defined(invoker.configs_exclude)) {
+      configs -= invoker.configs_exclude
+    }
+    configs += invoker.configs
+    deps = invoker.deps
+  }
+
+  executable("boringssl_pkcs7_test") {
+    sources = [
+      "src/crypto/x509/pkcs7_test.c",
+    ]
+    sources += _test_support_sources
+    if (defined(invoker.configs_exclude)) {
+      configs -= invoker.configs_exclude
+    }
+    configs += invoker.configs
+    deps = invoker.deps
+  }
+
+  executable("boringssl_x509_test") {
+    sources = [
+      "src/crypto/x509/x509_test.cc",
+    ]
+    sources += _test_support_sources
+    if (defined(invoker.configs_exclude)) {
+      configs -= invoker.configs_exclude
+    }
+    configs += invoker.configs
+    deps = invoker.deps
+  }
+
+  executable("boringssl_tab_test") {
+    sources = [
+      "src/crypto/x509v3/tab_test.c",
+    ]
+    sources += _test_support_sources
+    if (defined(invoker.configs_exclude)) {
+      configs -= invoker.configs_exclude
+    }
+    configs += invoker.configs
+    deps = invoker.deps
+  }
+
+  executable("boringssl_v3name_test") {
+    sources = [
+      "src/crypto/x509v3/v3name_test.c",
+    ]
+    sources += _test_support_sources
+    if (defined(invoker.configs_exclude)) {
+      configs -= invoker.configs_exclude
+    }
+    configs += invoker.configs
+    deps = invoker.deps
+  }
+
+  executable("boringssl_ssl_test") {
+    sources = [
+      "src/ssl/ssl_test.cc",
+    ]
+    sources += _test_support_sources
+    if (defined(invoker.configs_exclude)) {
+      configs -= invoker.configs_exclude
+    }
+    configs += invoker.configs
+    deps = invoker.deps
+  }
+
+  group(target_name) {
+    deps = [
+      ":boringssl_aead_test",
+      ":boringssl_aes_test",
+      ":boringssl_asn1_test",
+      ":boringssl_base64_test",
+      ":boringssl_bio_test",
+      ":boringssl_bn_test",
+      ":boringssl_bytestring_test",
+      ":boringssl_chacha_test",
+      ":boringssl_cipher_test",
+      ":boringssl_cmac_test",
+      ":boringssl_constant_time_test",
+      ":boringssl_dh_test",
+      ":boringssl_digest_test",
+      ":boringssl_dsa_test",
+      ":boringssl_ec_test",
+      ":boringssl_ecdsa_test",
+      ":boringssl_ed25519_test",
+      ":boringssl_err_test",
+      ":boringssl_evp_extra_test",
+      ":boringssl_evp_test",
+      ":boringssl_example_mul",
+      ":boringssl_gcm_test",
+      ":boringssl_hkdf_test",
+      ":boringssl_hmac_test",
+      ":boringssl_lhash_test",
+      ":boringssl_newhope_statistical_test",
+      ":boringssl_newhope_test",
+      ":boringssl_newhope_vectors_test",
+      ":boringssl_obj_test",
+      ":boringssl_pbkdf_test",
+      ":boringssl_pkcs12_test",
+      ":boringssl_pkcs7_test",
+      ":boringssl_pkcs8_test",
+      ":boringssl_poly1305_test",
+      ":boringssl_refcount_test",
+      ":boringssl_rsa_test",
+      ":boringssl_spake25519_test",
+      ":boringssl_ssl_test",
+      ":boringssl_tab_test",
+      ":boringssl_thread_test",
+      ":boringssl_v3name_test",
+      ":boringssl_x25519_test",
+      ":boringssl_x509_test",
+    ]
+  }
+}

third_party/boringssl/BUILD.gn

diff --git a/third_party/boringssl/BUILD.gn b/third_party/boringssl/BUILD.gn
index dc26330..a71a735 100644
--- a/third_party/boringssl/BUILD.gn
+++ b/third_party/boringssl/BUILD.gn
@@ -4,6 +4,7 @@
 
 import("//build/config/android/config.gni")
 import("//build/config/sanitizers/sanitizers.gni")
+import("BUILD.generated.gni")
 
 # Config for us and everybody else depending on BoringSSL.
 config("external_config") {
@@ -13,83 +14,107 @@
   }
 }
 
-
 # Config internal to this build file, shared by boringssl and boringssl_fuzzer.
 config("internal_config") {
   visibility = [ ":*" ]  # Only targets in this file can depend on this.
   defines = [
     "BORINGSSL_IMPLEMENTATION",
     "BORINGSSL_NO_STATIC_INITIALIZER",
-    "OPENSSL_SMALL_FOOTPRINT",
+    "OPENSSL_SMALL",
   ]
+  # configs = [
+  #   # TODO(davidben): Fix size_t truncations in BoringSSL.
+  #   # https://crbug.com/429039
+  #   "//build/config/compiler:no_size_t_to_int_warning",
+  # ]
+  if (is_posix) {
+    cflags_c = [ "-std=c99" ]
+    defines += [ "_XOPEN_SOURCE=700" ]
+  }
 }
 
-
 config("no_asm_config") {
   visibility = [ ":*" ]  # Only targets in this file can depend on this.
   defines = [ "OPENSSL_NO_ASM" ]
 }
 
+all_sources = crypto_sources + ssl_sources
 
-# The list of BoringSSL files is kept in boringssl.gypi.
-gypi_values =
-    exec_script("../../tools/gypi_to_gn.py",
-                [ rebase_path("boringssl.gypi") ],
-                "scope",
-                [ "boringssl.gypi" ])
-boringssl_sources =
-    gypi_values.boringssl_crypto_sources + gypi_values.boringssl_ssl_sources
+# Windows' assembly is built with Yasm. The other platforms use the platform
+# assembler.
+if (is_win && !is_msan) {
+  import("//third_party/yasm/yasm_assemble.gni")
+  yasm_assemble("boringssl_asm") {
+    if (current_cpu == "x64") {
+      sources = crypto_sources_win_x86_64
+    } else if (current_cpu == "x86") {
+      sources = crypto_sources_win_x86
+    }
+  }
+} else {
+  # This has no sources on some platforms so must be a source_set.
+  source_set("boringssl_asm") {
+    visibility = [ ":*" ]  # Only targets in this file can depend on this.
 
+    defines = []
+    sources = []
+    include_dirs = [ "src/include" ]
 
-source_set("boringssl_asm") {
-  visibility = [ ":*" ]  # Only targets in this file can depend on this.
-  sources = []
-  #asmflags = []
-  include_dirs = [
-    "src/include",
-    # This is for arm_arch.h, which is needed by some asm files. Since the
-    # asm files are generated and kept in a different directory, they
-    # cannot use relative paths to find this file.
-    "src/crypto",
-  ]
+    if ((current_cpu == "arm" || current_cpu == "arm64") && is_clang) {
+      if (current_cpu == "arm") {
+        # TODO(hans) Enable integrated-as (crbug.com/124610).
+        asmflags += [ "-fno-integrated-as" ]
+      }
+      if (is_android) {
+        rebased_android_toolchain_root =
+            rebase_path(android_toolchain_root, root_build_dir)
 
-  if (current_cpu == "x64") {
-    if (is_ios) {
-      defines += [ "OPENSSL_NO_ASM" ]
-    } else if (is_mac) {
-      sources += gypi_values.boringssl_mac_x86_64_sources
-    } else if (is_linux || is_android) {
-      sources += gypi_values.boringssl_linux_x86_64_sources
+        # Else /usr/bin/as gets picked up.
+        asmflags += [ "-B${rebased_android_toolchain_root}/bin" ]
+      }
+    }
+
+    if (is_msan) {
+      public_configs = [ ":no_asm_config" ]
+    } else if (current_cpu == "x64") {
+      if (is_mac) {
+        sources += crypto_sources_mac_x86_64
+      } else if (is_linux || is_android) {
+        sources += crypto_sources_linux_x86_64
+      } else {
+        public_configs = [ ":no_asm_config" ]
+      }
+    } else if (current_cpu == "x86") {
+      if (is_mac) {
+        sources += crypto_sources_mac_x86
+      } else if (is_linux || is_android) {
+        sources += crypto_sources_linux_x86
+      } else {
+        public_configs = [ ":no_asm_config" ]
+      }
+    } else if (current_cpu == "arm" && (is_linux || is_android)) {
+      sources += crypto_sources_linux_arm
+    } else if (current_cpu == "arm64" && (is_linux || is_android)) {
+      sources += crypto_sources_linux_aarch64
+
+      # TODO(davidben): Remove explicit arch flag once https://crbug.com/576858
+      # is fixed.
+      asmflags += [ "-march=armv8-a+crypto" ]
     } else {
       public_configs = [ ":no_asm_config" ]
     }
-  } else if (current_cpu == "x86") {
-    if (is_ios) {
-      defines += [ "OPENSSL_NO_ASM" ]
-    } else if (is_mac) {
-      sources += gypi_values.boringssl_mac_x86_sources
-    } else if (is_linux || is_android) {
-      sources += gypi_values.boringssl_linux_x86_sources
-    } else {
-      public_configs = [ ":no_asm_config" ]
-    }
-  } else if (current_cpu == "arm" && (is_linux || is_android)) {
-    sources += gypi_values.boringssl_linux_arm_sources
-  } else if (current_cpu == "arm64" && (is_linux || is_android)) {
-    sources += gypi_values.boringssl_linux_aarch64_sources
-  } else {
-    public_configs = [ ":no_asm_config" ]
   }
 }
 
-
 component("boringssl") {
-  sources = boringssl_sources
+  sources = all_sources
   deps = [
     ":boringssl_asm",
   ]
+
   public_configs = [ ":external_config" ]
   configs += [ ":internal_config" ]
+
   configs -= [ "//build/config/compiler:chromium_code" ]
   configs += [ "//build/config/compiler:no_chromium_code" ]
 }

third_party/boringssl/boringssl.gypi

diff --git a/third_party/boringssl/boringssl.gypi b/third_party/boringssl/boringssl.gypi
index fb8e643..2c7e7c7 100644
--- a/third_party/boringssl/boringssl.gypi
+++ b/third_party/boringssl/boringssl.gypi
@@ -1,4 +1,4 @@
-# Copyright (c) 2014 The Chromium Authors. All rights reserved.
+# Copyright (c) 2016 The Chromium Authors. All rights reserved.
 # Use of this source code is governed by a BSD-style license that can be
 # found in the LICENSE file.
 
@@ -7,34 +7,34 @@
 {
   'variables': {
     'boringssl_ssl_sources': [
+      'src/ssl/custom_extensions.c',
       'src/ssl/d1_both.c',
-      'src/ssl/d1_clnt.c',
       'src/ssl/d1_lib.c',
-      'src/ssl/d1_meth.c',
       'src/ssl/d1_pkt.c',
       'src/ssl/d1_srtp.c',
-      'src/ssl/d1_srvr.c',
-      'src/ssl/pqueue/pqueue.c',
+      'src/ssl/dtls_method.c',
+      'src/ssl/dtls_record.c',
+      'src/ssl/handshake_client.c',
+      'src/ssl/handshake_server.c',
       'src/ssl/s3_both.c',
-      'src/ssl/s3_clnt.c',
       'src/ssl/s3_enc.c',
       'src/ssl/s3_lib.c',
-      'src/ssl/s3_meth.c',
       'src/ssl/s3_pkt.c',
-      'src/ssl/s3_srvr.c',
       'src/ssl/ssl_aead_ctx.c',
-      'src/ssl/ssl_algs.c',
       'src/ssl/ssl_asn1.c',
+      'src/ssl/ssl_buffer.c',
       'src/ssl/ssl_cert.c',
       'src/ssl/ssl_cipher.c',
+      'src/ssl/ssl_ecdh.c',
+      'src/ssl/ssl_file.c',
       'src/ssl/ssl_lib.c',
       'src/ssl/ssl_rsa.c',
-      'src/ssl/ssl_sess.c',
+      'src/ssl/ssl_session.c',
       'src/ssl/ssl_stat.c',
-      'src/ssl/ssl_txt.c',
       'src/ssl/t1_enc.c',
       'src/ssl/t1_lib.c',
-      'src/ssl/t1_reneg.c',
+      'src/ssl/tls_method.c',
+      'src/ssl/tls_record.c',
     ],
     'boringssl_crypto_sources': [
       'err_data.c',
@@ -61,18 +61,14 @@
       'src/crypto/asn1/asn1_lib.c',
       'src/crypto/asn1/asn1_par.c',
       'src/crypto/asn1/asn_pack.c',
-      'src/crypto/asn1/bio_asn1.c',
-      'src/crypto/asn1/bio_ndef.c',
       'src/crypto/asn1/f_enum.c',
       'src/crypto/asn1/f_int.c',
       'src/crypto/asn1/f_string.c',
       'src/crypto/asn1/t_bitst.c',
-      'src/crypto/asn1/t_pkey.c',
       'src/crypto/asn1/tasn_dec.c',
       'src/crypto/asn1/tasn_enc.c',
       'src/crypto/asn1/tasn_fre.c',
       'src/crypto/asn1/tasn_new.c',
-      'src/crypto/asn1/tasn_prn.c',
       'src/crypto/asn1/tasn_typ.c',
       'src/crypto/asn1/tasn_utl.c',
       'src/crypto/asn1/x_bignum.c',
@@ -92,6 +88,7 @@
       'src/crypto/bn/add.c',
       'src/crypto/bn/asm/x86_64-gcc.c',
       'src/crypto/bn/bn.c',
+      'src/crypto/bn/bn_asn1.c',
       'src/crypto/bn/cmp.c',
       'src/crypto/bn/convert.c',
       'src/crypto/bn/ctx.c',
@@ -108,11 +105,11 @@
       'src/crypto/bn/shift.c',
       'src/crypto/bn/sqrt.c',
       'src/crypto/buf/buf.c',
+      'src/crypto/bytestring/asn1_compat.c',
       'src/crypto/bytestring/ber.c',
       'src/crypto/bytestring/cbb.c',
       'src/crypto/bytestring/cbs.c',
-      'src/crypto/chacha/chacha_generic.c',
-      'src/crypto/chacha/chacha_vec.c',
+      'src/crypto/chacha/chacha.c',
       'src/crypto/cipher/aead.c',
       'src/crypto/cipher/cipher.c',
       'src/crypto/cipher/derive_key.c',
@@ -127,28 +124,31 @@
       'src/crypto/cipher/tls_cbc.c',
       'src/crypto/cmac/cmac.c',
       'src/crypto/conf/conf.c',
+      'src/crypto/cpu-aarch64-linux.c',
+      'src/crypto/cpu-arm-linux.c',
       'src/crypto/cpu-arm.c',
       'src/crypto/cpu-intel.c',
       'src/crypto/crypto.c',
+      'src/crypto/curve25519/curve25519.c',
+      'src/crypto/curve25519/spake25519.c',
+      'src/crypto/curve25519/x25519-x86_64.c',
       'src/crypto/des/des.c',
       'src/crypto/dh/check.c',
       'src/crypto/dh/dh.c',
       'src/crypto/dh/dh_asn1.c',
-      'src/crypto/dh/dh_impl.c',
       'src/crypto/dh/params.c',
       'src/crypto/digest/digest.c',
       'src/crypto/digest/digests.c',
-      'src/crypto/directory_posix.c',
-      'src/crypto/directory_win.c',
       'src/crypto/dsa/dsa.c',
       'src/crypto/dsa/dsa_asn1.c',
-      'src/crypto/dsa/dsa_impl.c',
       'src/crypto/ec/ec.c',
       'src/crypto/ec/ec_asn1.c',
       'src/crypto/ec/ec_key.c',
       'src/crypto/ec/ec_montgomery.c',
       'src/crypto/ec/oct.c',
+      'src/crypto/ec/p224-64.c',
       'src/crypto/ec/p256-64.c',
+      'src/crypto/ec/p256-x86_64.c',
       'src/crypto/ec/simple.c',
       'src/crypto/ec/util-64.c',
       'src/crypto/ec/wnaf.c',
@@ -157,10 +157,9 @@
       'src/crypto/ecdsa/ecdsa_asn1.c',
       'src/crypto/engine/engine.c',
       'src/crypto/err/err.c',
-      'src/crypto/evp/algorithm.c',
-      'src/crypto/evp/asn1.c',
       'src/crypto/evp/digestsign.c',
       'src/crypto/evp/evp.c',
+      'src/crypto/evp/evp_asn1.c',
       'src/crypto/evp/evp_ctx.c',
       'src/crypto/evp/p_dsa_asn1.c',
       'src/crypto/evp/p_ec.c',
@@ -168,6 +167,7 @@
       'src/crypto/evp/p_rsa.c',
       'src/crypto/evp/p_rsa_asn1.c',
       'src/crypto/evp/pbkdf.c',
+      'src/crypto/evp/print.c',
       'src/crypto/evp/sign.c',
       'src/crypto/ex_data.c',
       'src/crypto/hkdf/hkdf.c',
@@ -181,6 +181,12 @@
       'src/crypto/modes/ctr.c',
       'src/crypto/modes/gcm.c',
       'src/crypto/modes/ofb.c',
+      'src/crypto/newhope/error_correction.c',
+      'src/crypto/newhope/newhope.c',
+      'src/crypto/newhope/ntt.c',
+      'src/crypto/newhope/poly.c',
+      'src/crypto/newhope/precomp.c',
+      'src/crypto/newhope/reduce.c',
       'src/crypto/obj/obj.c',
       'src/crypto/obj/obj_xref.c',
       'src/crypto/pem/pem_all.c',
@@ -198,7 +204,7 @@
       'src/crypto/poly1305/poly1305.c',
       'src/crypto/poly1305/poly1305_arm.c',
       'src/crypto/poly1305/poly1305_vec.c',
-      'src/crypto/rand/hwrand.c',
+      'src/crypto/rand/deterministic.c',
       'src/crypto/rand/rand.c',
       'src/crypto/rand/urandom.c',
       'src/crypto/rand/windows.c',
@@ -223,11 +229,13 @@
       'src/crypto/x509/a_sign.c',
       'src/crypto/x509/a_strex.c',
       'src/crypto/x509/a_verify.c',
+      'src/crypto/x509/algorithm.c',
       'src/crypto/x509/asn1_gen.c',
       'src/crypto/x509/by_dir.c',
       'src/crypto/x509/by_file.c',
       'src/crypto/x509/i2d_pr.c',
       'src/crypto/x509/pkcs7.c',
+      'src/crypto/x509/rsa_pss.c',
       'src/crypto/x509/t_crl.c',
       'src/crypto/x509/t_req.c',
       'src/crypto/x509/t_x509.c',
@@ -303,6 +311,8 @@
     ],
     'boringssl_linux_aarch64_sources': [
       'linux-aarch64/crypto/aes/aesv8-armx64.S',
+      'linux-aarch64/crypto/bn/armv8-mont.S',
+      'linux-aarch64/crypto/chacha/chacha-armv8.S',
       'linux-aarch64/crypto/modes/ghashv8-armx64.S',
       'linux-aarch64/crypto/sha/sha1-armv8.S',
       'linux-aarch64/crypto/sha/sha256-armv8.S',
@@ -313,13 +323,13 @@
       'linux-arm/crypto/aes/aesv8-armx32.S',
       'linux-arm/crypto/aes/bsaes-armv7.S',
       'linux-arm/crypto/bn/armv4-mont.S',
+      'linux-arm/crypto/chacha/chacha-armv4.S',
       'linux-arm/crypto/modes/ghash-armv4.S',
       'linux-arm/crypto/modes/ghashv8-armx32.S',
       'linux-arm/crypto/sha/sha1-armv4-large.S',
       'linux-arm/crypto/sha/sha256-armv4.S',
       'linux-arm/crypto/sha/sha512-armv4.S',
-      'src/crypto/chacha/chacha_vec_arm.S',
-      'src/crypto/cpu-arm-asm.S',
+      'src/crypto/curve25519/asm/x25519-asm-arm.S',
       'src/crypto/poly1305/poly1305_arm_asm.S',
     ],
     'boringssl_linux_x86_sources': [
@@ -329,7 +339,7 @@
       'linux-x86/crypto/bn/bn-586.S',
       'linux-x86/crypto/bn/co-586.S',
       'linux-x86/crypto/bn/x86-mont.S',
-      'linux-x86/crypto/cpu-x86-asm.S',
+      'linux-x86/crypto/chacha/chacha-x86.S',
       'linux-x86/crypto/md5/md5-586.S',
       'linux-x86/crypto/modes/ghash-x86.S',
       'linux-x86/crypto/rc4/rc4-586.S',
@@ -346,16 +356,17 @@
       'linux-x86_64/crypto/bn/rsaz-x86_64.S',
       'linux-x86_64/crypto/bn/x86_64-mont.S',
       'linux-x86_64/crypto/bn/x86_64-mont5.S',
-      'linux-x86_64/crypto/cpu-x86_64-asm.S',
+      'linux-x86_64/crypto/chacha/chacha-x86_64.S',
+      'linux-x86_64/crypto/ec/p256-x86_64-asm.S',
       'linux-x86_64/crypto/md5/md5-x86_64.S',
       'linux-x86_64/crypto/modes/aesni-gcm-x86_64.S',
       'linux-x86_64/crypto/modes/ghash-x86_64.S',
       'linux-x86_64/crypto/rand/rdrand-x86_64.S',
-      'linux-x86_64/crypto/rc4/rc4-md5-x86_64.S',
       'linux-x86_64/crypto/rc4/rc4-x86_64.S',
       'linux-x86_64/crypto/sha/sha1-x86_64.S',
       'linux-x86_64/crypto/sha/sha256-x86_64.S',
       'linux-x86_64/crypto/sha/sha512-x86_64.S',
+      'src/crypto/curve25519/asm/x25519-asm-x86_64.S',
     ],
     'boringssl_mac_x86_sources': [
       'mac-x86/crypto/aes/aes-586.S',
@@ -364,7 +375,7 @@
       'mac-x86/crypto/bn/bn-586.S',
       'mac-x86/crypto/bn/co-586.S',
       'mac-x86/crypto/bn/x86-mont.S',
-      'mac-x86/crypto/cpu-x86-asm.S',
+      'mac-x86/crypto/chacha/chacha-x86.S',
       'mac-x86/crypto/md5/md5-586.S',
       'mac-x86/crypto/modes/ghash-x86.S',
       'mac-x86/crypto/rc4/rc4-586.S',
@@ -381,16 +392,17 @@
       'mac-x86_64/crypto/bn/rsaz-x86_64.S',
       'mac-x86_64/crypto/bn/x86_64-mont.S',
       'mac-x86_64/crypto/bn/x86_64-mont5.S',
-      'mac-x86_64/crypto/cpu-x86_64-asm.S',
+      'mac-x86_64/crypto/chacha/chacha-x86_64.S',
+      'mac-x86_64/crypto/ec/p256-x86_64-asm.S',
       'mac-x86_64/crypto/md5/md5-x86_64.S',
       'mac-x86_64/crypto/modes/aesni-gcm-x86_64.S',
       'mac-x86_64/crypto/modes/ghash-x86_64.S',
       'mac-x86_64/crypto/rand/rdrand-x86_64.S',
-      'mac-x86_64/crypto/rc4/rc4-md5-x86_64.S',
       'mac-x86_64/crypto/rc4/rc4-x86_64.S',
       'mac-x86_64/crypto/sha/sha1-x86_64.S',
       'mac-x86_64/crypto/sha/sha256-x86_64.S',
       'mac-x86_64/crypto/sha/sha512-x86_64.S',
+      'src/crypto/curve25519/asm/x25519-asm-x86_64.S',
     ],
     'boringssl_win_x86_sources': [
       'win-x86/crypto/aes/aes-586.asm',
@@ -399,7 +411,7 @@
       'win-x86/crypto/bn/bn-586.asm',
       'win-x86/crypto/bn/co-586.asm',
       'win-x86/crypto/bn/x86-mont.asm',
-      'win-x86/crypto/cpu-x86-asm.asm',
+      'win-x86/crypto/chacha/chacha-x86.asm',
       'win-x86/crypto/md5/md5-586.asm',
       'win-x86/crypto/modes/ghash-x86.asm',
       'win-x86/crypto/rc4/rc4-586.asm',
@@ -416,12 +428,12 @@
       'win-x86_64/crypto/bn/rsaz-x86_64.asm',
       'win-x86_64/crypto/bn/x86_64-mont.asm',
       'win-x86_64/crypto/bn/x86_64-mont5.asm',
-      'win-x86_64/crypto/cpu-x86_64-asm.asm',
+      'win-x86_64/crypto/chacha/chacha-x86_64.asm',
+      'win-x86_64/crypto/ec/p256-x86_64-asm.asm',
       'win-x86_64/crypto/md5/md5-x86_64.asm',
       'win-x86_64/crypto/modes/aesni-gcm-x86_64.asm',
       'win-x86_64/crypto/modes/ghash-x86_64.asm',
       'win-x86_64/crypto/rand/rdrand-x86_64.asm',
-      'win-x86_64/crypto/rc4/rc4-md5-x86_64.asm',
       'win-x86_64/crypto/rc4/rc4-x86_64.asm',
       'win-x86_64/crypto/sha/sha1-x86_64.asm',
       'win-x86_64/crypto/sha/sha256-x86_64.asm',

third_party/boringssl/boringssl_configurations.gypi

diff --git a/third_party/boringssl/boringssl_configurations.gypi b/third_party/boringssl/boringssl_configurations.gypi
index b963505..3162d79 100644
--- a/third_party/boringssl/boringssl_configurations.gypi
+++ b/third_party/boringssl/boringssl_configurations.gypi
@@ -4,29 +4,20 @@
 
 # This file is included to modify the configurations to build third-party
 # code from BoringSSL.
-# This code is C code, not C++, and is not warning-free, so we need to remove
-# C++-specific flags, and add flags to supress the warnings in the code.
 {
-  'variables': {
-    # Used by third_party/nss, which is from Chromium.
-    # Include the built-in set of root certificate authorities.
-    'exclude_nss_root_certs': 0,
-    'os_posix%': 1,
-    'os_bsd%': 0,
-    'chromeos%': 0,
-    'clang%': 0,
-  },
   'target_defaults': {
-    'cflags': [
-      '-w',
-      '-UHAVE_CVAR_BUILT_ON_SEM',
+    'conditions': [
+      ['OS == "linux" or OS == "android"', {
+        'cflags_c': [
+          '-std=c99',
+        ],
+        'defines': [
+          '_XOPEN_SOURCE=700',
+        ],
+      }],
     ],
     # Removes these flags from the list cflags.
     'cflags!': [
-      # NSS code from upstream mozilla builds with warnings,
-      # so we must allow warnings without failing.
-      '-Werror',
-      '-Wall',
       '-ansi',
       # Not supported for C, only for C++.
       '-Wnon-virtual-dtor',
@@ -35,93 +26,5 @@
       '-fvisibility-inlines-hidden',
       '-Woverloaded-virtual',
     ],
-    'configurations': {
-      'Dart_Base': {
-        'xcode_settings': {
-          'WARNING_CFLAGS': [
-            '-w',
-          ],
-          'WARNING_CFLAGS!': [
-            '-Wall',
-            '-Wextra',
-          ],
-        },
-      },
-      # Dart_Macos_Debug and Dart_Macos_Release are merged after
-      # Dart_Macos_Base, so we can override the 'ansi' and '-Werror' flags set
-      # at the global level in tools/gyp/configurations_xcode.gypi.
-      'Dart_Macos_Debug': {
-        'abstract': 1,
-        'xcode_settings': {
-          # Remove 'ansi' setting.
-          'GCC_C_LANGUAGE_STANDARD': 'c99',
-          'GCC_TREAT_WARNINGS_AS_ERRORS': 'NO', # -Werror off
-        },
-      },
-      'Dart_Macos_Release': {
-        'abstract': 1,
-        'xcode_settings': {
-          # Remove 'ansi' setting.
-          'GCC_C_LANGUAGE_STANDARD': 'c99',
-          'GCC_TREAT_WARNINGS_AS_ERRORS': 'NO', # -Werror off
-        },
-      },
-      # Disable hand-coded assembly routines on ARMv6 and ARMv5TE.
-      'Dart_armv6_Base': {
-        'abstract': 1,
-        'defines': [
-          'OPENSSL_NO_ASM',
-        ],
-      },
-      'Dart_armv5te_Base': {
-        'abstract': 1,
-        'defines': [
-          'OPENSSL_NO_ASM',
-        ],
-      },
-      # TODO(24321): Also disable temporarily on arm64. Reenable after the next
-      # roll.
-      'Dart_arm64_Base': {
-        'abstract': 1,
-        'defines': [
-          'OPENSSL_NO_ASM',
-        ],
-      },
-      # Android 64-bit dbc build is for arm64, disable temporarily as well.
-      'Dart_Android_arm64_Base': {
-        'abstract': 1,
-        'defines': [
-          'OPENSSL_NO_ASM',
-        ],
-      },
-      # When being built for Android nss expects __linux__ to be defined.
-      'Dart_Android_Base': {
-        'target_conditions': [
-          ['_toolset=="host"', {
-            'defines!': [
-              'ANDROID',
-            ],
-            # Define __linux__ on Android build for NSS.
-            'defines': [
-              '__linux__',
-            ],
-            'cflags!': [
-              '-U__linux__',
-            ],
-          }],
-          ['_toolset=="target"', {
-            'defines': [
-              '__linux__',
-              'CHECK_FORK_GETPID',  # Android does not provide pthread_atfork.
-              '__USE_LARGEFILE64',
-            ],
-            # Define __linux__ on Android build for NSS.
-            'cflags!': [
-              '-U__linux__',
-            ],
-          }]
-        ],
-      },
-    },
   },
 }

third_party/boringssl/boringssl_dart.gyp

diff --git a/third_party/boringssl/boringssl_dart.gyp b/third_party/boringssl/boringssl_dart.gyp
index 30270fa..1dc460f 100644
--- a/third_party/boringssl/boringssl_dart.gyp
+++ b/third_party/boringssl/boringssl_dart.gyp
@@ -22,6 +22,7 @@
       'defines': [
         'BORINGSSL_IMPLEMENTATION',
         'BORINGSSL_NO_STATIC_INITIALIZER',
+        'OPENSSL_SMALL'
       ],
       # TODO(davidben): Fix size_t truncations in BoringSSL.
       # https://crbug.com/429039
@@ -47,10 +48,6 @@
       ],
       'include_dirs': [
         'src/include',
-        # This is for arm_arch.h, which is needed by some asm files. Since the
-        # asm files are generated and kept in a different directory, they
-        # cannot use relative paths to find this file.
-        'src/crypto',
       ],
       'direct_dependent_settings': {
         'include_dirs': [

third_party/boringssl/boringssl_tests.gypi

diff --git a/third_party/boringssl/boringssl_tests.gypi b/third_party/boringssl/boringssl_tests.gypi
index 9413d8c..1076214 100644
--- a/third_party/boringssl/boringssl_tests.gypi
+++ b/third_party/boringssl/boringssl_tests.gypi
@@ -1,4 +1,4 @@
-# Copyright (c) 2014 The Chromium Authors. All rights reserved.
+# Copyright (c) 2016 The Chromium Authors. All rights reserved.
 # Use of this source code is governed by a BSD-style license that can be
 # found in the LICENSE file.
 
@@ -7,6 +7,34 @@
 {
   'targets': [
     {
+      'target_name': 'boringssl_aes_test',
+      'type': 'executable',
+      'dependencies': [
+        'boringssl.gyp:boringssl',
+      ],
+      'sources': [
+        'src/crypto/aes/aes_test.cc',
+        '<@(boringssl_test_support_sources)',
+      ],
+      # TODO(davidben): Fix size_t truncations in BoringSSL.
+      # https://crbug.com/429039
+      'msvs_disabled_warnings': [ 4267, ],
+    },
+    {
+      'target_name': 'boringssl_asn1_test',
+      'type': 'executable',
+      'dependencies': [
+        'boringssl.gyp:boringssl',
+      ],
+      'sources': [
+        'src/crypto/asn1/asn1_test.cc',
+        '<@(boringssl_test_support_sources)',
+      ],
+      # TODO(davidben): Fix size_t truncations in BoringSSL.
+      # https://crbug.com/429039
+      'msvs_disabled_warnings': [ 4267, ],
+    },
+    {
       'target_name': 'boringssl_base64_test',
       'type': 'executable',
       'dependencies': [
@@ -63,6 +91,20 @@
       'msvs_disabled_warnings': [ 4267, ],
     },
     {
+      'target_name': 'boringssl_chacha_test',
+      'type': 'executable',
+      'dependencies': [
+        'boringssl.gyp:boringssl',
+      ],
+      'sources': [
+        'src/crypto/chacha/chacha_test.cc',
+        '<@(boringssl_test_support_sources)',
+      ],
+      # TODO(davidben): Fix size_t truncations in BoringSSL.
+      # https://crbug.com/429039
+      'msvs_disabled_warnings': [ 4267, ],
+    },
+    {
       'target_name': 'boringssl_aead_test',
       'type': 'executable',
       'dependencies': [
@@ -119,6 +161,48 @@
       'msvs_disabled_warnings': [ 4267, ],
     },
     {
+      'target_name': 'boringssl_ed25519_test',
+      'type': 'executable',
+      'dependencies': [
+        'boringssl.gyp:boringssl',
+      ],
+      'sources': [
+        'src/crypto/curve25519/ed25519_test.cc',
+        '<@(boringssl_test_support_sources)',
+      ],
+      # TODO(davidben): Fix size_t truncations in BoringSSL.
+      # https://crbug.com/429039
+      'msvs_disabled_warnings': [ 4267, ],
+    },
+    {
+      'target_name': 'boringssl_spake25519_test',
+      'type': 'executable',
+      'dependencies': [
+        'boringssl.gyp:boringssl',
+      ],
+      'sources': [
+        'src/crypto/curve25519/spake25519_test.cc',
+        '<@(boringssl_test_support_sources)',
+      ],
+      # TODO(davidben): Fix size_t truncations in BoringSSL.
+      # https://crbug.com/429039
+      'msvs_disabled_warnings': [ 4267, ],
+    },
+    {
+      'target_name': 'boringssl_x25519_test',
+      'type': 'executable',
+      'dependencies': [
+        'boringssl.gyp:boringssl',
+      ],
+      'sources': [
+        'src/crypto/curve25519/x25519_test.cc',
+        '<@(boringssl_test_support_sources)',
+      ],
+      # TODO(davidben): Fix size_t truncations in BoringSSL.
+      # https://crbug.com/429039
+      'msvs_disabled_warnings': [ 4267, ],
+    },
+    {
       'target_name': 'boringssl_dh_test',
       'type': 'executable',
       'dependencies': [
@@ -315,6 +399,62 @@
       'msvs_disabled_warnings': [ 4267, ],
     },
     {
+      'target_name': 'boringssl_newhope_statistical_test',
+      'type': 'executable',
+      'dependencies': [
+        'boringssl.gyp:boringssl',
+      ],
+      'sources': [
+        'src/crypto/newhope/newhope_statistical_test.cc',
+        '<@(boringssl_test_support_sources)',
+      ],
+      # TODO(davidben): Fix size_t truncations in BoringSSL.
+      # https://crbug.com/429039
+      'msvs_disabled_warnings': [ 4267, ],
+    },
+    {
+      'target_name': 'boringssl_newhope_test',
+      'type': 'executable',
+      'dependencies': [
+        'boringssl.gyp:boringssl',
+      ],
+      'sources': [
+        'src/crypto/newhope/newhope_test.cc',
+        '<@(boringssl_test_support_sources)',
+      ],
+      # TODO(davidben): Fix size_t truncations in BoringSSL.
+      # https://crbug.com/429039
+      'msvs_disabled_warnings': [ 4267, ],
+    },
+    {
+      'target_name': 'boringssl_newhope_vectors_test',
+      'type': 'executable',
+      'dependencies': [
+        'boringssl.gyp:boringssl',
+      ],
+      'sources': [
+        'src/crypto/newhope/newhope_vectors_test.cc',
+        '<@(boringssl_test_support_sources)',
+      ],
+      # TODO(davidben): Fix size_t truncations in BoringSSL.
+      # https://crbug.com/429039
+      'msvs_disabled_warnings': [ 4267, ],
+    },
+    {
+      'target_name': 'boringssl_obj_test',
+      'type': 'executable',
+      'dependencies': [
+        'boringssl.gyp:boringssl',
+      ],
+      'sources': [
+        'src/crypto/obj/obj_test.cc',
+        '<@(boringssl_test_support_sources)',
+      ],
+      # TODO(davidben): Fix size_t truncations in BoringSSL.
+      # https://crbug.com/429039
+      'msvs_disabled_warnings': [ 4267, ],
+    },
+    {
       'target_name': 'boringssl_pkcs12_test',
       'type': 'executable',
       'dependencies': [
@@ -329,6 +469,34 @@
       'msvs_disabled_warnings': [ 4267, ],
     },
     {
+      'target_name': 'boringssl_pkcs8_test',
+      'type': 'executable',
+      'dependencies': [
+        'boringssl.gyp:boringssl',
+      ],
+      'sources': [
+        'src/crypto/pkcs8/pkcs8_test.cc',
+        '<@(boringssl_test_support_sources)',
+      ],
+      # TODO(davidben): Fix size_t truncations in BoringSSL.
+      # https://crbug.com/429039
+      'msvs_disabled_warnings': [ 4267, ],
+    },
+    {
+      'target_name': 'boringssl_poly1305_test',
+      'type': 'executable',
+      'dependencies': [
+        'boringssl.gyp:boringssl',
+      ],
+      'sources': [
+        'src/crypto/poly1305/poly1305_test.cc',
+        '<@(boringssl_test_support_sources)',
+      ],
+      # TODO(davidben): Fix size_t truncations in BoringSSL.
+      # https://crbug.com/429039
+      'msvs_disabled_warnings': [ 4267, ],
+    },
+    {
       'target_name': 'boringssl_refcount_test',
       'type': 'executable',
       'dependencies': [
@@ -385,6 +553,20 @@
       'msvs_disabled_warnings': [ 4267, ],
     },
     {
+      'target_name': 'boringssl_x509_test',
+      'type': 'executable',
+      'dependencies': [
+        'boringssl.gyp:boringssl',
+      ],
+      'sources': [
+        'src/crypto/x509/x509_test.cc',
+        '<@(boringssl_test_support_sources)',
+      ],
+      # TODO(davidben): Fix size_t truncations in BoringSSL.
+      # https://crbug.com/429039
+      'msvs_disabled_warnings': [ 4267, ],
+    },
+    {
       'target_name': 'boringssl_tab_test',
       'type': 'executable',
       'dependencies': [
@@ -413,20 +595,6 @@
       'msvs_disabled_warnings': [ 4267, ],
     },
     {
-      'target_name': 'boringssl_pqueue_test',
-      'type': 'executable',
-      'dependencies': [
-        'boringssl.gyp:boringssl',
-      ],
-      'sources': [
-        'src/ssl/pqueue/pqueue_test.c',
-        '<@(boringssl_test_support_sources)',
-      ],
-      # TODO(davidben): Fix size_t truncations in BoringSSL.
-      # https://crbug.com/429039
-      'msvs_disabled_warnings': [ 4267, ],
-    },
-    {
       'target_name': 'boringssl_ssl_test',
       'type': 'executable',
       'dependencies': [
@@ -444,14 +612,25 @@
   'variables': {
     'boringssl_test_support_sources': [
       'src/crypto/test/file_test.cc',
+      'src/crypto/test/file_test.h',
       'src/crypto/test/malloc.cc',
+      'src/crypto/test/scoped_types.h',
+      'src/crypto/test/test_util.cc',
+      'src/crypto/test/test_util.h',
+      'src/ssl/test/async_bio.h',
+      'src/ssl/test/packeted_bio.h',
+      'src/ssl/test/scoped_types.h',
+      'src/ssl/test/test_config.h',
     ],
     'boringssl_test_targets': [
       'boringssl_aead_test',
+      'boringssl_aes_test',
+      'boringssl_asn1_test',
       'boringssl_base64_test',
       'boringssl_bio_test',
       'boringssl_bn_test',
       'boringssl_bytestring_test',
+      'boringssl_chacha_test',
       'boringssl_cipher_test',
       'boringssl_cmac_test',
       'boringssl_constant_time_test',
@@ -460,6 +639,7 @@
       'boringssl_dsa_test',
       'boringssl_ec_test',
       'boringssl_ecdsa_test',
+      'boringssl_ed25519_test',
       'boringssl_err_test',
       'boringssl_evp_extra_test',
       'boringssl_evp_test',
@@ -468,16 +648,24 @@
       'boringssl_hkdf_test',
       'boringssl_hmac_test',
       'boringssl_lhash_test',
+      'boringssl_newhope_statistical_test',
+      'boringssl_newhope_test',
+      'boringssl_newhope_vectors_test',
+      'boringssl_obj_test',
       'boringssl_pbkdf_test',
       'boringssl_pkcs12_test',
       'boringssl_pkcs7_test',
-      'boringssl_pqueue_test',
+      'boringssl_pkcs8_test',
+      'boringssl_poly1305_test',
       'boringssl_refcount_test',
       'boringssl_rsa_test',
+      'boringssl_spake25519_test',
       'boringssl_ssl_test',
       'boringssl_tab_test',
       'boringssl_thread_test',
       'boringssl_v3name_test',
+      'boringssl_x25519_test',
+      'boringssl_x509_test',
     ],
   }
 }

third_party/boringssl/err_data.c

diff --git a/third_party/boringssl/err_data.c b/third_party/boringssl/err_data.c
index 27acc89..d685679 100644
--- a/third_party/boringssl/err_data.c
+++ b/third_party/boringssl/err_data.c
@@ -49,1554 +49,171 @@
 OPENSSL_COMPILE_ASSERT(ERR_LIB_HMAC == 28, library_values_changed_28);
 OPENSSL_COMPILE_ASSERT(ERR_LIB_DIGEST == 29, library_values_changed_29);
 OPENSSL_COMPILE_ASSERT(ERR_LIB_CIPHER == 30, library_values_changed_30);
-OPENSSL_COMPILE_ASSERT(ERR_LIB_USER == 31, library_values_changed_31);
-OPENSSL_COMPILE_ASSERT(ERR_LIB_HKDF == 32, library_values_changed_32);
+OPENSSL_COMPILE_ASSERT(ERR_LIB_HKDF == 31, library_values_changed_31);
+OPENSSL_COMPILE_ASSERT(ERR_LIB_USER == 32, library_values_changed_32);
 OPENSSL_COMPILE_ASSERT(ERR_NUM_LIBS == 33, library_values_changed_num);
 
-const uint32_t kOpenSSLFunctionValues[] = {
-    0xc32054b,
-    0xc328556,
-    0xc330561,
-    0xc33856e,
-    0xc340578,
-    0xc348582,
-    0xc350589,
-    0xc358595,
-    0xc36059c,
-    0xc3685b2,
-    0xc3705d1,
-    0xc3785e2,
-    0xc3805f2,
-    0xc38860c,
-    0xc390621,
-    0xc398630,
-    0xc3a0649,
-    0xc3a865d,
-    0xc3b0669,
-    0xc3b8670,
-    0xc3c0678,
-    0xc3c8690,
-    0xc3d0698,
-    0xc3d86a0,
-    0xc3e06ab,
-    0xc3e85c7,
-    0xc3f0686,
-    0x1032193d,
-    0x10329954,
-    0x1033196d,
-    0x10339983,
-    0x10341993,
-    0x103499bb,
-    0x103519c9,
-    0x103599d8,
-    0x103619f8,
-    0x10369a17,
-    0x10371a34,
-    0x10379a51,
-    0x10381a66,
-    0x10389a88,
-    0x10391aa7,
-    0x10399ac6,
-    0x103a1add,
-    0x103a9af4,
-    0x103b1afd,
-    0x103b9b08,
-    0x103c1b22,
-    0x103c9b2a,
-    0x103d1b32,
-    0x103d99a6,
-    0x103e1b4b,
-    0x103e9b5d,
-    0x103f1b70,
-    0x103f9b79,
-    0x10401b39,
-    0x14320a4e,
-    0x14328a5c,
-    0x14330a68,
-    0x14338a75,
-    0x18361224,
-    0x18371252,
-    0x18379263,
-    0x18381279,
-    0x1839129c,
-    0x183992b1,
-    0x183a12c3,
-    0x183c1307,
-    0x183c9315,
-    0x183d1328,
-    0x183d9338,
-    0x183e935e,
-    0x183f1371,
-    0x183f9380,
-    0x184093aa,
-    0x18411416,
-    0x18419427,
-    0x1842143a,
-    0x1842944c,
-    0x1843145e,
-    0x1843946f,
-    0x18441480,
-    0x18449491,
-    0x184514a2,
-    0x184594af,
-    0x184614d1,
-    0x184694e4,
-    0x184714f8,
-    0x18479505,
-    0x18481514,
-    0x18489523,
-    0x18491534,
-    0x18499550,
-    0x184a155e,
-    0x184a956f,
-    0x184b1580,
-    0x184b958e,
-    0x184c159e,
-    0x184c95c4,
-    0x184d15d3,
-    0x184d95e3,
-    0x184e15f3,
-    0x184e9602,
-    0x184f1541,
-    0x184f91b3,
-    0x18501156,
-    0x1850916e,
-    0x18511190,
-    0x185191a2,
-    0x185211d4,
-    0x185291ed,
-    0x185311fe,
-    0x18539214,
-    0x18541239,
-    0x1854928a,
-    0x185512d3,
-    0x185592e8,
-    0x185612f5,
-    0x1856934d,
-    0x18571390,
-    0x1857939d,
-    0x185813b9,
-    0x185893ca,
-    0x185913da,
-    0x185993ea,
-    0x185a13f9,
-    0x185a9408,
-    0x185b14bd,
-    0x1c3206b8,
-    0x1c3286c4,
-    0x1c3306cf,
-    0x1c3386db,
-    0x20321616,
-    0x20329621,
-    0x20331629,
-    0x20339635,
-    0x24321641,
-    0x2432964f,
-    0x24331661,
-    0x24339670,
-    0x24341683,
-    0x24349696,
-    0x243516ad,
-    0x243596c5,
-    0x243616d3,
-    0x243696eb,
-    0x243716f4,
-    0x24379706,
-    0x2438171a,
-    0x24389727,
-    0x2439173d,
-    0x24399755,
-    0x243a176d,
-    0x243a9777,
-    0x243b178c,
-    0x243b979a,
-    0x243c17b2,
-    0x243c97c9,
-    0x243d17d4,
-    0x243d97e2,
-    0x28320aae,
-    0x28328abd,
-    0x28330ac8,
-    0x28338acd,
-    0x28340ad8,
-    0x2c322b99,
-    0x2c32aba5,
-    0x2c332bb8,
-    0x2c33abc9,
-    0x2c342be2,
-    0x2c34ac0a,
-    0x2c352c21,
-    0x2c35ac3e,
-    0x2c362c5b,
-    0x2c36ac78,
-    0x2c372c91,
-    0x2c37acaa,
-    0x2c382cc0,
-    0x2c38acce,
-    0x2c392ce0,
-    0x2c39acfd,
-    0x2c3a2d1a,
-    0x2c3aad28,
-    0x2c3b2d46,
-    0x2c3bad64,
-    0x2c3c2d7f,
-    0x2c3cad93,
-    0x2c3d2da5,
-    0x2c3dadb5,
-    0x2c3e2dc3,
-    0x2c3eadd3,
-    0x2c3f2de3,
-    0x2c3fae22,
-    0x2c402e33,
-    0x2c40ae4e,
-    0x2c412e62,
-    0x2c41ae75,
-    0x2c422e94,
-    0x2c42aea8,
-    0x2c432ebb,
-    0x2c43aeca,
-    0x2c442ed9,
-    0x2c44aef0,
-    0x2c452f0b,
-    0x2c45af23,
-    0x2c462f37,
-    0x2c46af4a,
-    0x2c472f5b,
-    0x2c47af6c,
-    0x2c482f7d,
-    0x2c48af8e,
-    0x2c492f9d,
-    0x2c49afaa,
-    0x2c4a2fb7,
-    0x2c4aafc4,
-    0x2c4b2fcd,
-    0x2c4bafe1,
-    0x2c4c2ff0,
-    0x2c4caffe,
-    0x2c4d3020,
-    0x2c4db031,
-    0x2c4e3042,
-    0x2c4eb00d,
-    0x2c4f2bfb,
-    0x2c4fadfe,
-    0x2c502e10,
-    0x30320000,
-    0x30328018,
-    0x3033002c,
-    0x30338042,
-    0x3034005b,
-    0x3034806c,
-    0x3035007f,
-    0x3035808f,
-    0x3036009d,
-    0x303680b3,
-    0x303700c3,
-    0x303780d8,
-    0x303800e6,
-    0x303880f7,
-    0x30390103,
-    0x3039810c,
-    0x303a011d,
-    0x303a812d,
-    0x303b013a,
-    0x303b8146,
-    0x303c0157,
-    0x303c8165,
-    0x303d0176,
-    0x303d8188,
-    0x303e0199,
-    0x303e81a8,
-    0x303f01b9,
-    0x303f81cd,
-    0x304001df,
-    0x304081ec,
-    0x30410202,
-    0x30418215,
-    0x30420225,
-    0x30428239,
-    0x3043024a,
-    0x3043825a,
-    0x30440265,
-    0x3044826d,
-    0x3045027d,
-    0x30458294,
-    0x304602a1,
-    0x304682b7,
-    0x304702c9,
-    0x304782d5,
-    0x304802e1,
-    0x304882ef,
-    0x30490308,
-    0x30498316,
-    0x304a032b,
-    0x304a8343,
-    0x304b034d,
-    0x304b8361,
-    0x304c0372,
-    0x304c8382,
-    0x304d038f,
-    0x304d83a0,
-    0x304e03b0,
-    0x304e83c2,
-    0x304f03d3,
-    0x304f83e2,
-    0x305003f6,
-    0x30508404,
-    0x30510413,
-    0x3051841c,
-    0x343209d6,
-    0x343289e6,
-    0x343309f1,
-    0x343389fe,
-    0x38320a07,
-    0x38328a1f,
-    0x38330a32,
-    0x38338a3c,
-    0x3c320aeb,
-    0x3c328af9,
-    0x3c330b10,
-    0x3c338b24,
-    0x3c340b56,
-    0x3c348b67,
-    0x3c350b73,
-    0x3c358ba0,
-    0x3c360bb2,
-    0x3c368bdb,
-    0x3c370be8,
-    0x3c378bf5,
-    0x3c380c03,
-    0x3c388c10,
-    0x3c390c1d,
-    0x3c398c41,
-    0x3c3a0c51,
-    0x3c3a8c69,
-    0x3c3b0c7e,
-    0x3c3b8c93,
-    0x3c3c0ca0,
-    0x3c3c8cb3,
-    0x3c3d0cc6,
-    0x3c3d8cea,
-    0x3c3e0d12,
-    0x3c3e8d2b,
-    0x3c3f0d41,
-    0x3c3f8d4e,
-    0x3c400d61,
-    0x3c408d72,
-    0x3c410d83,
-    0x3c418d9c,
-    0x3c420db5,
-    0x3c428dcb,
-    0x3c430de8,
-    0x3c438dfe,
-    0x3c440e82,
-    0x3c448ea9,
-    0x3c450ec7,
-    0x3c458ee1,
-    0x3c460ef9,
-    0x3c468f11,
-    0x3c470f3c,
-    0x3c478f67,
-    0x3c480f88,
-    0x3c488fb1,
-    0x3c490fcc,
-    0x3c498ff5,
-    0x3c4a1002,
-    0x3c4a9019,
-    0x3c4b1030,
-    0x3c4b9059,
-    0x3c4c1069,
-    0x3c4c9075,
-    0x3c4d108d,
-    0x3c4d90a0,
-    0x3c4e10b1,
-    0x3c4e90c2,
-    0x3c4f10e8,
-    0x3c4f8adf,
-    0x3c500e1a,
-    0x3c508e3a,
-    0x3c510e67,
-    0x3c518fe7,
-    0x3c5210d2,
-    0x3c528b87,
-    0x3c530b3f,
-    0x40321bb9,
-    0x40329bf2,
-    0x40331c1a,
-    0x40339c32,
-    0x40341c50,
-    0x40349cb6,
-    0x40351ccd,
-    0x40359ce9,
-    0x40361d05,
-    0x40369d1f,
-    0x40371d3e,
-    0x40379d5d,
-    0x40381d75,
-    0x40389d92,
-    0x40391db5,
-    0x40399dd2,
-    0x403a1e07,
-    0x403a9e61,
-    0x403b1e76,
-    0x403b9e92,
-    0x403c1eac,
-    0x403c9eb7,
-    0x403d1eda,
-    0x403d9efe,
-    0x403e1f14,
-    0x403e9f1e,
-    0x403f1f2a,
-    0x403f9f3b,
-    0x40401f53,
-    0x40409f5b,
-    0x40411f64,
-    0x40419f6d,
-    0x40421f95,
-    0x40429fa9,
-    0x40431fb4,
-    0x40439fc0,
-    0x40442014,
-    0x4044a020,
-    0x4045202d,
-    0x4045a040,
-    0x40462058,
-    0x4046a070,
-    0x40472086,
-    0x4047a0a1,
-    0x404820bc,
-    0x4048a0d0,
-    0x404920e9,
-    0x4049a102,
-    0x404a211c,
-    0x404aa126,
-    0x404b1e29,
-    0x404b9e48,
-    0x404c2136,
-    0x404ca144,
-    0x404d2151,
-    0x404da165,
-    0x404e217d,
-    0x404ea18b,
-    0x404f21b5,
-    0x404fa1cc,
-    0x405021de,
-    0x4050a20f,
-    0x40512240,
-    0x4051a255,
-    0x40522278,
-    0x4052a298,
-    0x405322ad,
-    0x4053a2bd,
-    0x4054a2c9,
-    0x405522df,
-    0x4055a31f,
-    0x4056232c,
-    0x4056a336,
-    0x40572344,
-    0x4057a35f,
-    0x4058237a,
-    0x4058a399,
-    0x405923ae,
-    0x4059a3c3,
-    0x405a23e0,
-    0x405aa3f4,
-    0x405b2410,
-    0x405ba426,
-    0x405c2443,
-    0x405ca455,
-    0x405d246c,
-    0x405da47d,
-    0x405e2499,
-    0x405ea4ad,
-    0x405f24bd,
-    0x405fa4d9,
-    0x406024ee,
-    0x4060a504,
-    0x40612521,
-    0x4061a53a,
-    0x40622564,
-    0x4062a56d,
-    0x4063257d,
-    0x4063a5b6,
-    0x406425cc,
-    0x4064a5ea,
-    0x406525ff,
-    0x4065a61c,
-    0x40662633,
-    0x4066a651,
-    0x4067266e,
-    0x4067a685,
-    0x406826a3,
-    0x4068a6ba,
-    0x406926d2,
-    0x4069a6e3,
-    0x406a26f6,
-    0x406aa709,
-    0x406b271d,
-    0x406ba741,
-    0x406c275c,
-    0x406ca77d,
-    0x406d27a1,
-    0x406da7bc,
-    0x406e27dd,
-    0x406ea7f2,
-    0x406f280b,
-    0x406fa818,
-    0x40702826,
-    0x4070a833,
-    0x40712850,
-    0x4071a870,
-    0x4072288b,
-    0x4072a8a4,
-    0x407328bb,
-    0x4073a8d5,
-    0x407428f9,
-    0x4074a90f,
-    0x40752923,
-    0x4075a938,
-    0x40762952,
-    0x4076a964,
-    0x40772979,
-    0x4077a99f,
-    0x407829bc,
-    0x4078a9df,
-    0x40792a05,
-    0x4079aa22,
-    0x407a2a45,
-    0x407aaa61,
-    0x407b2a7d,
-    0x407baa8f,
-    0x407c2a9c,
-    0x407e2aa9,
-    0x407eaabf,
-    0x407f2ad7,
-    0x407faaea,
-    0x40802aff,
-    0x4080ab18,
-    0x40812b36,
-    0x4081ab56,
-    0x40822b5f,
-    0x4082ab7b,
-    0x40832b84,
-    0x4083a19a,
-    0x40842229,
-    0x4084a1f9,
-    0x408525a5,
-    0x4085a589,
-    0x40861c8e,
-    0x40869ca1,
-    0x40871ff4,
-    0x4087a003,
-    0x40881bfe,
-    0x40889f7d,
-    0x40891fdb,
-    0x4089a54d,
-    0x408a1b84,
-    0x408a9b95,
-    0x408b1ba7,
-    0x408ba266,
-    0x408c1df0,
-    0x408c9e17,
-    0x408d22fd,
-    0x408d9bd3,
-    0x408e1c6f,
-    0x4432042a,
-    0x4432843c,
-    0x44330445,
-    0x4433844d,
-    0x4434045a,
-    0x44348475,
-    0x44350490,
-    0x443584b0,
-    0x443604cc,
-    0x443684ed,
-    0x443704f4,
-    0x44378502,
-    0x4438050c,
-    0x44388518,
-    0x44390522,
-    0x4439852d,
-    0x443a0537,
-    0x443a8541,
-    0x443b046a,
-    0x4c3217ea,
-    0x4c3297f9,
-    0x4c331808,
-    0x4c339821,
-    0x4c34183c,
-    0x4c349858,
-    0x4c35186a,
-    0x4c359878,
-    0x4c36188d,
-    0x4c36989e,
-    0x4c3718ac,
-    0x4c3798ba,
-    0x4c3818cc,
-    0x4c3898dc,
-    0x4c3918e6,
-    0x4c3998fe,
-    0x4c3a1916,
-    0x4c3a9929,
-    0x50323053,
-    0x5032b068,
-    0x50333079,
-    0x5033b08c,
-    0x5034309d,
-    0x5034b0b0,
-    0x503530bf,
-    0x5035b0d4,
-    0x503630e4,
-    0x5036b0f3,
-    0x50373104,
-    0x5037b114,
-    0x50383125,
-    0x5038b138,
-    0x5039314a,
-    0x5039b160,
-    0x503a3172,
-    0x503ab183,
-    0x503b3194,
-    0x503bb1a5,
-    0x503c31b0,
-    0x503cb1bc,
-    0x503d31c7,
-    0x503db1d2,
-    0x503e31df,
-    0x503eb1f4,
-    0x503f3202,
-    0x503fb216,
-    0x50403229,
-    0x5040b23a,
-    0x50413254,
-    0x5041b263,
-    0x5042326c,
-    0x5042b27b,
-    0x5043328d,
-    0x5043b299,
-    0x504432a1,
-    0x5044b2b4,
-    0x504532c5,
-    0x5045b2db,
-    0x504632e7,
-    0x5046b2fb,
-    0x50473309,
-    0x5047b31d,
-    0x50483337,
-    0x5048b34b,
-    0x50493361,
-    0x5049b378,
-    0x504a338a,
-    0x504ab39e,
-    0x504b33b3,
-    0x504bb3ca,
-    0x504c33de,
-    0x504cb3e7,
-    0x504d33ef,
-    0x504db3fe,
-    0x504e340e,
-    0x68321109,
-    0x6832911a,
-    0x6833112a,
-    0x68339138,
-    0x68341145,
-    0x6c3210f8,
-    0x74320a89,
-    0x74328a9b,
-    0x783206e8,
-    0x7832871b,
-    0x7833072d,
-    0x7833873f,
-    0x78340753,
-    0x78348767,
-    0x78350785,
-    0x78358797,
-    0x783607ab,
-    0x78368819,
-    0x7837082b,
-    0x7837883d,
-    0x7838084f,
-    0x78388866,
-    0x7839087d,
-    0x78398894,
-    0x783a08b0,
-    0x783a88cc,
-    0x783b08e8,
-    0x783b88fe,
-    0x783c0914,
-    0x783c892a,
-    0x783d0947,
-    0x783d8956,
-    0x783e0965,
-    0x783e8974,
-    0x783f0990,
-    0x783f899e,
-    0x784009ac,
-    0x784089ba,
-    0x784109c7,
-    0x784186fa,
-    0x784207bf,
-    0x784287dd,
-    0x784307fb,
-    0x80321611,
-};
-
-const size_t kOpenSSLFunctionValuesLen = sizeof(kOpenSSLFunctionValues) / sizeof(kOpenSSLFunctionValues[0]);
-
-const char kOpenSSLFunctionStringData[] =
-    "ASN1_BIT_STRING_set_bit\0"
-    "ASN1_ENUMERATED_set\0"
-    "ASN1_ENUMERATED_to_BN\0"
-    "ASN1_GENERALIZEDTIME_adj\0"
-    "ASN1_INTEGER_set\0"
-    "ASN1_INTEGER_to_BN\0"
-    "ASN1_OBJECT_new\0"
-    "ASN1_PCTX_new\0"
-    "ASN1_STRING_TABLE_add\0"
-    "ASN1_STRING_set\0"
-    "ASN1_STRING_type_new\0"
-    "ASN1_TIME_adj\0"
-    "ASN1_UTCTIME_adj\0"
-    "ASN1_d2i_fp\0"
-    "ASN1_dup\0"
-    "ASN1_generate_v3\0"
-    "ASN1_get_object\0"
-    "ASN1_i2d_bio\0"
-    "ASN1_i2d_fp\0"
-    "ASN1_item_d2i_fp\0"
-    "ASN1_item_dup\0"
-    "ASN1_item_ex_d2i\0"
-    "ASN1_item_i2d_bio\0"
-    "ASN1_item_i2d_fp\0"
-    "ASN1_item_pack\0"
-    "ASN1_item_unpack\0"
-    "ASN1_mbstring_ncopy\0"
-    "ASN1_template_new\0"
-    "BIO_new_NDEF\0"
-    "BN_to_ASN1_ENUMERATED\0"
-    "BN_to_ASN1_INTEGER\0"
-    "a2d_ASN1_OBJECT\0"
-    "a2i_ASN1_ENUMERATED\0"
-    "a2i_ASN1_INTEGER\0"
-    "a2i_ASN1_STRING\0"
-    "append_exp\0"
-    "asn1_cb\0"
-    "asn1_check_tlen\0"
-    "asn1_collate_primitive\0"
-    "asn1_collect\0"
-    "asn1_d2i_ex_primitive\0"
-    "asn1_d2i_read_bio\0"
-    "asn1_do_adb\0"
-    "asn1_ex_c2i\0"
-    "asn1_find_end\0"
-    "asn1_item_ex_combine_new\0"
-    "asn1_str2type\0"
-    "asn1_template_ex_d2i\0"
-    "asn1_template_noexp_d2i\0"
-    "bitstr_cb\0"
-    "c2i_ASN1_BIT_STRING\0"
-    "c2i_ASN1_INTEGER\0"
-    "c2i_ASN1_OBJECT\0"
-    "collect_data\0"
-    "d2i_ASN1_BOOLEAN\0"
-    "d2i_ASN1_OBJECT\0"
-    "d2i_ASN1_UINTEGER\0"
-    "d2i_ASN1_UTCTIME\0"
-    "d2i_ASN1_bytes\0"
-    "d2i_ASN1_type_bytes\0"
-    "i2d_ASN1_TIME\0"
-    "i2d_PrivateKey\0"
-    "long_c2i\0"
-    "parse_tagging\0"
-    "BIO_callback_ctrl\0"
-    "BIO_ctrl\0"
-    "BIO_new\0"
-    "BIO_new_file\0"
-    "BIO_new_mem_buf\0"
-    "BIO_printf\0"
-    "BIO_zero_copy_get_read_buf\0"
-    "BIO_zero_copy_get_read_buf_done\0"
-    "BIO_zero_copy_get_write_buf\0"
-    "BIO_zero_copy_get_write_buf_done\0"
-    "bio_io\0"
-    "bio_make_pair\0"
-    "bio_write\0"
-    "buffer_ctrl\0"
-    "conn_ctrl\0"
-    "conn_state\0"
-    "file_ctrl\0"
-    "file_read\0"
-    "mem_write\0"
-    "BN_CTX_get\0"
-    "BN_CTX_new\0"
-    "BN_CTX_start\0"
-    "BN_bn2dec\0"
-    "BN_bn2hex\0"
-    "BN_div\0"
-    "BN_div_recp\0"
-    "BN_exp\0"
-    "BN_generate_dsa_nonce\0"
-    "BN_generate_prime_ex\0"
-    "BN_lshift\0"
-    "BN_mod_exp2_mont\0"
-    "BN_mod_exp_mont\0"
-    "BN_mod_exp_mont_consttime\0"
-    "BN_mod_exp_mont_word\0"
-    "BN_mod_inverse\0"
-    "BN_mod_inverse_no_branch\0"
-    "BN_mod_lshift_quick\0"
-    "BN_mod_sqrt\0"
-    "BN_new\0"
-    "BN_rand\0"
-    "BN_rand_range\0"
-    "BN_rshift\0"
-    "BN_sqrt\0"
-    "BN_usub\0"
-    "bn_wexpand\0"
-    "mod_exp_recp\0"
-    "BUF_MEM_new\0"
-    "BUF_memdup\0"
-    "BUF_strndup\0"
-    "buf_mem_grow\0"
-    "EVP_AEAD_CTX_init\0"
-    "EVP_AEAD_CTX_init_with_direction\0"
-    "EVP_AEAD_CTX_open\0"
-    "EVP_AEAD_CTX_seal\0"
-    "EVP_CIPHER_CTX_copy\0"
-    "EVP_CIPHER_CTX_ctrl\0"
-    "EVP_CIPHER_CTX_set_key_length\0"
-    "EVP_CipherInit_ex\0"
-    "EVP_DecryptFinal_ex\0"
-    "EVP_EncryptFinal_ex\0"
-    "aead_aes_ctr_hmac_sha256_init\0"
-    "aead_aes_ctr_hmac_sha256_open\0"
-    "aead_aes_ctr_hmac_sha256_seal\0"
-    "aead_aes_gcm_init\0"
-    "aead_aes_gcm_open\0"
-    "aead_aes_gcm_seal\0"
-    "aead_aes_key_wrap_init\0"
-    "aead_aes_key_wrap_open\0"
-    "aead_aes_key_wrap_seal\0"
-    "aead_chacha20_poly1305_init\0"
-    "aead_chacha20_poly1305_open\0"
-    "aead_chacha20_poly1305_seal\0"
-    "aead_rc4_md5_tls_init\0"
-    "aead_rc4_md5_tls_open\0"
-    "aead_rc4_md5_tls_seal\0"
-    "aead_ssl3_ensure_cipher_init\0"
-    "aead_ssl3_init\0"
-    "aead_ssl3_open\0"
-    "aead_ssl3_seal\0"
-    "aead_tls_ensure_cipher_init\0"
-    "aead_tls_init\0"
-    "aead_tls_open\0"
-    "aead_tls_seal\0"
-    "aes_init_key\0"
-    "aesni_init_key\0"
-    "CONF_parse_list\0"
-    "NCONF_load\0"
-    "def_load_bio\0"
-    "str_copy\0"
-    "CRYPTO_get_ex_new_index\0"
-    "CRYPTO_set_ex_data\0"
-    "get_class\0"
-    "get_func_pointers\0"
-    "DH_new_method\0"
-    "compute_key\0"
-    "generate_key\0"
-    "generate_parameters\0"
-    "EVP_DigestInit_ex\0"
-    "EVP_MD_CTX_copy_ex\0"
-    "DSA_new_method\0"
-    "dsa_sig_cb\0"
-    "sign\0"
-    "sign_setup\0"
-    "verify\0"
-    "BN_to_felem\0"
-    "EC_GROUP_copy\0"
-    "EC_GROUP_get_curve_GFp\0"
-    "EC_GROUP_get_degree\0"
-    "EC_GROUP_new_by_curve_name\0"
-    "EC_GROUP_new_curve_GFp\0"
-    "EC_KEY_check_key\0"
-    "EC_KEY_copy\0"
-    "EC_KEY_generate_key\0"
-    "EC_KEY_new_by_curve_name\0"
-    "EC_KEY_new_method\0"
-    "EC_KEY_set_public_key_affine_coordinates\0"
-    "EC_POINT_add\0"
-    "EC_POINT_cmp\0"
-    "EC_POINT_copy\0"
-    "EC_POINT_dbl\0"
-    "EC_POINT_dup\0"
-    "EC_POINT_get_affine_coordinates_GFp\0"
-    "EC_POINT_invert\0"
-    "EC_POINT_is_at_infinity\0"
-    "EC_POINT_is_on_curve\0"
-    "EC_POINT_make_affine\0"
-    "EC_POINT_new\0"
-    "EC_POINT_oct2point\0"
-    "EC_POINT_point2oct\0"
-    "EC_POINT_set_affine_coordinates_GFp\0"
-    "EC_POINT_set_compressed_coordinates_GFp\0"
-    "EC_POINT_set_to_infinity\0"
-    "EC_POINTs_make_affine\0"
-    "compute_wNAF\0"
-    "d2i_ECPKParameters\0"
-    "d2i_ECParameters\0"
-    "d2i_ECPrivateKey\0"
-    "ec_GFp_mont_field_decode\0"
-    "ec_GFp_mont_field_encode\0"
-    "ec_GFp_mont_field_mul\0"
-    "ec_GFp_mont_field_set_to_one\0"
-    "ec_GFp_mont_field_sqr\0"
-    "ec_GFp_mont_group_set_curve\0"
-    "ec_GFp_nistp256_group_set_curve\0"
-    "ec_GFp_nistp256_point_get_affine_coordinates\0"
-    "ec_GFp_nistp256_points_mul\0"
-    "ec_GFp_simple_group_check_discriminant\0"
-    "ec_GFp_simple_group_set_curve\0"
-    "ec_GFp_simple_make_affine\0"
-    "ec_GFp_simple_oct2point\0"
-    "ec_GFp_simple_point2oct\0"
-    "ec_GFp_simple_point_get_affine_coordinates\0"
-    "ec_GFp_simple_point_set_affine_coordinates\0"
-    "ec_GFp_simple_points_make_affine\0"
-    "ec_GFp_simple_set_compressed_coordinates\0"
-    "ec_asn1_group2pkparameters\0"
-    "ec_asn1_pkparameters2group\0"
-    "ec_group_copy\0"
-    "ec_group_new\0"
-    "ec_group_new_curve_GFp\0"
-    "ec_group_new_from_data\0"
-    "ec_point_set_Jprojective_coordinates_GFp\0"
-    "ec_pre_comp_new\0"
-    "ec_wNAF_mul\0"
-    "ec_wNAF_precompute_mult\0"
-    "i2d_ECPKParameters\0"
-    "i2d_ECParameters\0"
-    "i2d_ECPrivateKey\0"
-    "i2o_ECPublicKey\0"
-    "nistp256_pre_comp_new\0"
-    "o2i_ECPublicKey\0"
-    "ECDH_compute_key\0"
-    "ECDSA_do_sign_ex\0"
-    "ECDSA_do_verify\0"
-    "ECDSA_sign_ex\0"
-    "digest_to_bn\0"
-    "ecdsa_sign_setup\0"
-    "EVP_DigestSignAlgorithm\0"
-    "EVP_DigestVerifyInitFromAlgorithm\0"
-    "EVP_PKEY_CTX_ctrl\0"
-    "EVP_PKEY_CTX_dup\0"
-    "EVP_PKEY_CTX_get0_rsa_oaep_label\0"
-    "EVP_PKEY_copy_parameters\0"
-    "EVP_PKEY_decrypt\0"
-    "EVP_PKEY_decrypt_init\0"
-    "EVP_PKEY_derive\0"
-    "EVP_PKEY_derive_init\0"
-    "EVP_PKEY_derive_set_peer\0"
-    "EVP_PKEY_encrypt\0"
-    "EVP_PKEY_encrypt_init\0"
-    "EVP_PKEY_get1_DH\0"
-    "EVP_PKEY_get1_DSA\0"
-    "EVP_PKEY_get1_EC_KEY\0"
-    "EVP_PKEY_get1_RSA\0"
-    "EVP_PKEY_keygen\0"
-    "EVP_PKEY_keygen_init\0"
-    "EVP_PKEY_new\0"
-    "EVP_PKEY_set_type\0"
-    "EVP_PKEY_sign\0"
-    "EVP_PKEY_sign_init\0"
-    "EVP_PKEY_verify\0"
-    "EVP_PKEY_verify_init\0"
-    "check_padding_md\0"
-    "d2i_AutoPrivateKey\0"
-    "d2i_PrivateKey\0"
-    "do_EC_KEY_print\0"
-    "do_dsa_print\0"
-    "do_rsa_print\0"
-    "do_sigver_init\0"
-    "dsa_param_decode\0"
-    "dsa_priv_decode\0"
-    "dsa_priv_encode\0"
-    "dsa_pub_decode\0"
-    "dsa_pub_encode\0"
-    "dsa_sig_print\0"
-    "eckey_param2type\0"
-    "eckey_param_decode\0"
-    "eckey_priv_decode\0"
-    "eckey_priv_encode\0"
-    "eckey_pub_decode\0"
-    "eckey_pub_encode\0"
-    "eckey_type2param\0"
-    "evp_pkey_ctx_new\0"
-    "hmac_signctx\0"
-    "i2d_PublicKey\0"
-    "old_dsa_priv_decode\0"
-    "old_ec_priv_decode\0"
-    "old_rsa_priv_decode\0"
-    "pkey_ec_ctrl\0"
-    "pkey_ec_derive\0"
-    "pkey_ec_keygen\0"
-    "pkey_ec_paramgen\0"
-    "pkey_ec_sign\0"
-    "pkey_hmac_ctrl\0"
-    "pkey_rsa_ctrl\0"
-    "pkey_rsa_decrypt\0"
-    "pkey_rsa_encrypt\0"
-    "pkey_rsa_sign\0"
-    "rsa_algor_to_md\0"
-    "rsa_digest_verify_init_from_algorithm\0"
-    "rsa_mgf1_to_md\0"
-    "rsa_priv_decode\0"
-    "rsa_priv_encode\0"
-    "rsa_pss_to_ctx\0"
-    "rsa_pub_decode\0"
-    "HKDF\0"
-    "OBJ_create\0"
-    "OBJ_dup\0"
-    "OBJ_nid2obj\0"
-    "OBJ_txt2obj\0"
-    "PEM_ASN1_read\0"
-    "PEM_ASN1_read_bio\0"
-    "PEM_ASN1_write\0"
-    "PEM_ASN1_write_bio\0"
-    "PEM_X509_INFO_read\0"
-    "PEM_X509_INFO_read_bio\0"
-    "PEM_X509_INFO_write_bio\0"
-    "PEM_do_header\0"
-    "PEM_get_EVP_CIPHER_INFO\0"
-    "PEM_read\0"
-    "PEM_read_DHparams\0"
-    "PEM_read_PrivateKey\0"
-    "PEM_read_bio\0"
-    "PEM_read_bio_DHparams\0"
-    "PEM_read_bio_Parameters\0"
-    "PEM_read_bio_PrivateKey\0"
-    "PEM_write\0"
-    "PEM_write_PrivateKey\0"
-    "PEM_write_bio\0"
-    "d2i_PKCS8PrivateKey_bio\0"
-    "d2i_PKCS8PrivateKey_fp\0"
-    "do_pk8pkey\0"
-    "do_pk8pkey_fp\0"
-    "load_iv\0"
-    "EVP_PKCS82PKEY\0"
-    "EVP_PKEY2PKCS8\0"
-    "PKCS12_get_key_and_certs\0"
-    "PKCS12_handle_content_info\0"
-    "PKCS12_handle_content_infos\0"
-    "PKCS5_pbe2_set_iv\0"
-    "PKCS5_pbe_set\0"
-    "PKCS5_pbe_set0_algor\0"
-    "PKCS5_pbkdf2_set\0"
-    "PKCS8_decrypt\0"
-    "PKCS8_encrypt\0"
-    "PKCS8_encrypt_pbe\0"
-    "pbe_cipher_init\0"
-    "pbe_crypt\0"
-    "pkcs12_item_decrypt_d2i\0"
-    "pkcs12_item_i2d_encrypt\0"
-    "pkcs12_key_gen_raw\0"
-    "pkcs12_pbe_keyivgen\0"
-    "BN_BLINDING_convert_ex\0"
-    "BN_BLINDING_create_param\0"
-    "BN_BLINDING_invert_ex\0"
-    "BN_BLINDING_new\0"
-    "BN_BLINDING_update\0"
-    "RSA_add_pkcs1_prefix\0"
-    "RSA_check_key\0"
-    "RSA_new_method\0"
-    "RSA_padding_add_PKCS1_OAEP_mgf1\0"
-    "RSA_padding_add_PKCS1_PSS_mgf1\0"
-    "RSA_padding_add_PKCS1_type_1\0"
-    "RSA_padding_add_PKCS1_type_2\0"
-    "RSA_padding_add_none\0"
-    "RSA_padding_check_PKCS1_OAEP_mgf1\0"
-    "RSA_padding_check_PKCS1_type_1\0"
-    "RSA_padding_check_PKCS1_type_2\0"
-    "RSA_padding_check_none\0"
-    "RSA_recover_crt_params\0"
-    "RSA_sign\0"
-    "RSA_verify\0"
-    "RSA_verify_PKCS1_PSS_mgf1\0"
-    "decrypt\0"
-    "encrypt\0"
-    "keygen\0"
-    "keygen_multiprime\0"
-    "private_transform\0"
-    "rsa_setup_blinding\0"
-    "sign_raw\0"
-    "verify_raw\0"
-    "SSL_AEAD_CTX_new\0"
-    "SSL_AEAD_CTX_open\0"
-    "SSL_AEAD_CTX_seal\0"
-    "SSL_CTX_check_private_key\0"
-    "SSL_CTX_get_tlsext_ticket_keys\0"
-    "SSL_CTX_new\0"
-    "SSL_CTX_set1_tls_channel_id\0"
-    "SSL_CTX_set_cipher_list\0"
-    "SSL_CTX_set_cipher_list_tls11\0"
-    "SSL_CTX_set_session_id_context\0"
-    "SSL_CTX_set_tlsext_ticket_keys\0"
-    "SSL_CTX_set_tmp_dh\0"
-    "SSL_CTX_set_tmp_ecdh\0"
-    "SSL_CTX_use_PrivateKey\0"
-    "SSL_CTX_use_PrivateKey_ASN1\0"
-    "SSL_CTX_use_PrivateKey_file\0"
-    "SSL_CTX_use_RSAPrivateKey\0"
-    "SSL_CTX_use_RSAPrivateKey_ASN1\0"
-    "SSL_CTX_use_RSAPrivateKey_file\0"
-    "SSL_CTX_use_certificate\0"
-    "SSL_CTX_use_certificate_ASN1\0"
-    "SSL_CTX_use_certificate_chain_file\0"
-    "SSL_CTX_use_certificate_file\0"
-    "SSL_CTX_use_psk_identity_hint\0"
-    "SSL_SESSION_from_bytes\0"
-    "SSL_SESSION_new\0"
-    "SSL_SESSION_parse\0"
-    "SSL_SESSION_parse_octet_string\0"
-    "SSL_SESSION_parse_string\0"
-    "SSL_SESSION_print_fp\0"
-    "SSL_SESSION_set1_id_context\0"
-    "SSL_SESSION_to_bytes_full\0"
-    "SSL_accept\0"
-    "SSL_add_dir_cert_subjects_to_stack\0"
-    "SSL_add_file_cert_subjects_to_stack\0"
-    "SSL_check_private_key\0"
-    "SSL_clear\0"
-    "SSL_connect\0"
-    "SSL_do_handshake\0"
-    "SSL_load_client_CA_file\0"
-    "SSL_new\0"
-    "SSL_peek\0"
-    "SSL_read\0"
-    "SSL_renegotiate\0"
-    "SSL_set1_tls_channel_id\0"
-    "SSL_set_cipher_list\0"
-    "SSL_set_fd\0"
-    "SSL_set_rfd\0"
-    "SSL_set_session_id_context\0"
-    "SSL_set_tlsext_host_name\0"
-    "SSL_set_tmp_dh\0"
-    "SSL_set_tmp_ecdh\0"
-    "SSL_set_wfd\0"
-    "SSL_shutdown\0"
-    "SSL_use_PrivateKey\0"
-    "SSL_use_PrivateKey_ASN1\0"
-    "SSL_use_PrivateKey_file\0"
-    "SSL_use_RSAPrivateKey\0"
-    "SSL_use_RSAPrivateKey_ASN1\0"
-    "SSL_use_RSAPrivateKey_file\0"
-    "SSL_use_certificate\0"
-    "SSL_use_certificate_ASN1\0"
-    "SSL_use_certificate_file\0"
-    "SSL_use_psk_identity_hint\0"
-    "SSL_write\0"
-    "d2i_SSL_SESSION\0"
-    "do_ssl3_write\0"
-    "dtls1_accept\0"
-    "dtls1_buffer_record\0"
-    "dtls1_check_timeout_num\0"
-    "dtls1_connect\0"
-    "dtls1_do_write\0"
-    "dtls1_get_buffered_message\0"
-    "dtls1_get_hello_verify\0"
-    "dtls1_get_message\0"
-    "dtls1_get_message_fragment\0"
-    "dtls1_hm_fragment_new\0"
-    "dtls1_preprocess_fragment\0"
-    "dtls1_process_fragment\0"
-    "dtls1_process_record\0"
-    "dtls1_read_bytes\0"
-    "dtls1_seal_record\0"
-    "dtls1_send_hello_verify_request\0"
-    "dtls1_write_app_data\0"
-    "i2d_SSL_SESSION\0"
-    "ssl3_accept\0"
-    "ssl3_cert_verify_hash\0"
-    "ssl3_check_cert_and_algorithm\0"
-    "ssl3_check_certificate_for_cipher\0"
-    "ssl3_connect\0"
-    "ssl3_ctrl\0"
-    "ssl3_ctx_ctrl\0"
-    "ssl3_digest_cached_records\0"
-    "ssl3_do_change_cipher_spec\0"
-    "ssl3_expect_change_cipher_spec\0"
-    "ssl3_get_cert_status\0"
-    "ssl3_get_cert_verify\0"
-    "ssl3_get_certificate_request\0"
-    "ssl3_get_channel_id\0"
-    "ssl3_get_client_certificate\0"
-    "ssl3_get_client_hello\0"
-    "ssl3_get_client_key_exchange\0"
-    "ssl3_get_finished\0"
-    "ssl3_get_initial_bytes\0"
-    "ssl3_get_message\0"
-    "ssl3_get_new_session_ticket\0"
-    "ssl3_get_next_proto\0"
-    "ssl3_get_record\0"
-    "ssl3_get_server_certificate\0"
-    "ssl3_get_server_done\0"
-    "ssl3_get_server_hello\0"
-    "ssl3_get_server_key_exchange\0"
-    "ssl3_get_v2_client_hello\0"
-    "ssl3_handshake_mac\0"
-    "ssl3_output_cert_chain\0"
-    "ssl3_prf\0"
-    "ssl3_read_bytes\0"
-    "ssl3_read_n\0"
-    "ssl3_record_sequence_update\0"
-    "ssl3_seal_record\0"
-    "ssl3_send_cert_verify\0"
-    "ssl3_send_certificate_request\0"
-    "ssl3_send_channel_id\0"
-    "ssl3_send_client_certificate\0"
-    "ssl3_send_client_hello\0"
-    "ssl3_send_client_key_exchange\0"
-    "ssl3_send_server_certificate\0"
-    "ssl3_send_server_hello\0"
-    "ssl3_send_server_key_exchange\0"
-    "ssl3_setup_read_buffer\0"
-    "ssl3_setup_write_buffer\0"
-    "ssl3_write_bytes\0"
-    "ssl3_write_pending\0"
-    "ssl_add_cert_chain\0"
-    "ssl_add_cert_to_buf\0"
-    "ssl_add_clienthello_renegotiate_ext\0"
-    "ssl_add_clienthello_tlsext\0"
-    "ssl_add_clienthello_use_srtp_ext\0"
-    "ssl_add_serverhello_renegotiate_ext\0"
-    "ssl_add_serverhello_tlsext\0"
-    "ssl_add_serverhello_use_srtp_ext\0"
-    "ssl_build_cert_chain\0"
-    "ssl_bytes_to_cipher_list\0"
-    "ssl_cert_dup\0"
-    "ssl_cert_inst\0"
-    "ssl_cert_new\0"
-    "ssl_check_serverhello_tlsext\0"
-    "ssl_check_srvr_ecc_cert_and_alg\0"
-    "ssl_cipher_process_rulestr\0"
-    "ssl_cipher_strength_sort\0"
-    "ssl_create_cipher_list\0"
-    "ssl_ctx_log_master_secret\0"
-    "ssl_ctx_log_rsa_client_key_exchange\0"
-    "ssl_ctx_make_profiles\0"
-    "ssl_get_new_session\0"
-    "ssl_get_prev_session\0"
-    "ssl_get_server_cert_index\0"
-    "ssl_get_sign_pkey\0"
-    "ssl_init_wbio_buffer\0"
-    "ssl_parse_clienthello_renegotiate_ext\0"
-    "ssl_parse_clienthello_tlsext\0"
-    "ssl_parse_clienthello_use_srtp_ext\0"
-    "ssl_parse_serverhello_renegotiate_ext\0"
-    "ssl_parse_serverhello_tlsext\0"
-    "ssl_parse_serverhello_use_srtp_ext\0"
-    "ssl_scan_clienthello_tlsext\0"
-    "ssl_scan_serverhello_tlsext\0"
-    "ssl_sess_cert_new\0"
-    "ssl_set_cert\0"
-    "ssl_set_pkey\0"
-    "ssl_verify_cert_chain\0"
-    "tls12_check_peer_sigalg\0"
-    "tls1_aead_ctx_init\0"
-    "tls1_cert_verify_mac\0"
-    "tls1_change_cipher_state\0"
-    "tls1_change_cipher_state_aead\0"
-    "tls1_check_duplicate_extensions\0"
-    "tls1_enc\0"
-    "tls1_export_keying_material\0"
-    "tls1_prf\0"
-    "tls1_setup_key_block\0"
-    "ASN1_digest\0"
-    "ASN1_item_sign_ctx\0"
-    "ASN1_item_verify\0"
-    "NETSCAPE_SPKI_b64_decode\0"
-    "NETSCAPE_SPKI_b64_encode\0"
-    "PKCS7_get_CRLs\0"
-    "PKCS7_get_certificates\0"
-    "X509_ATTRIBUTE_create_by_NID\0"
-    "X509_ATTRIBUTE_create_by_OBJ\0"
-    "X509_ATTRIBUTE_create_by_txt\0"
-    "X509_ATTRIBUTE_get0_data\0"
-    "X509_ATTRIBUTE_set1_data\0"
-    "X509_CRL_add0_revoked\0"
-    "X509_CRL_diff\0"
-    "X509_CRL_print_fp\0"
-    "X509_EXTENSION_create_by_NID\0"
-    "X509_EXTENSION_create_by_OBJ\0"
-    "X509_INFO_new\0"
-    "X509_NAME_ENTRY_create_by_NID\0"
-    "X509_NAME_ENTRY_create_by_txt\0"
-    "X509_NAME_ENTRY_set_object\0"
-    "X509_NAME_add_entry\0"
-    "X509_NAME_oneline\0"
-    "X509_NAME_print\0"
-    "X509_PKEY_new\0"
-    "X509_PUBKEY_get\0"
-    "X509_PUBKEY_set\0"
-    "X509_REQ_check_private_key\0"
-    "X509_REQ_print_ex\0"
-    "X509_REQ_print_fp\0"
-    "X509_REQ_to_X509\0"
-    "X509_STORE_CTX_get1_issuer\0"
-    "X509_STORE_CTX_init\0"
-    "X509_STORE_CTX_new\0"
-    "X509_STORE_CTX_purpose_inherit\0"
-    "X509_STORE_add_cert\0"
-    "X509_STORE_add_crl\0"
-    "X509_TRUST_add\0"
-    "X509_TRUST_set\0"
-    "X509_check_private_key\0"
-    "X509_get_pubkey_parameters\0"
-    "X509_load_cert_crl_file\0"
-    "X509_load_cert_file\0"
-    "X509_load_crl_file\0"
-    "X509_print_ex_fp\0"
-    "X509_to_X509_REQ\0"
-    "X509_verify_cert\0"
-    "X509at_add1_attr\0"
-    "X509v3_add_ext\0"
-    "add_cert_dir\0"
-    "by_file_ctrl\0"
-    "check_policy\0"
-    "dir_ctrl\0"
-    "get_cert_by_subject\0"
-    "i2d_DSA_PUBKEY\0"
-    "i2d_EC_PUBKEY\0"
-    "i2d_RSA_PUBKEY\0"
-    "pkcs7_parse_header\0"
-    "x509_name_encode\0"
-    "x509_name_ex_d2i\0"
-    "x509_name_ex_new\0"
-    "SXNET_add_id_INTEGER\0"
-    "SXNET_add_id_asc\0"
-    "SXNET_add_id_ulong\0"
-    "SXNET_get_id_asc\0"
-    "SXNET_get_id_ulong\0"
-    "X509V3_EXT_add\0"
-    "X509V3_EXT_add_alias\0"
-    "X509V3_EXT_free\0"
-    "X509V3_EXT_i2d\0"
-    "X509V3_EXT_nconf\0"
-    "X509V3_add1_i2d\0"
-    "X509V3_add_value\0"
-    "X509V3_get_section\0"
-    "X509V3_get_string\0"
-    "X509V3_get_value_bool\0"
-    "X509V3_parse_list\0"
-    "X509_PURPOSE_add\0"
-    "X509_PURPOSE_set\0"
-    "a2i_GENERAL_NAME\0"
-    "copy_email\0"
-    "copy_issuer\0"
-    "do_dirname\0"
-    "do_ext_i2d\0"
-    "do_ext_nconf\0"
-    "gnames_from_sectname\0"
-    "hex_to_string\0"
-    "i2s_ASN1_ENUMERATED\0"
-    "i2s_ASN1_IA5STRING\0"
-    "i2s_ASN1_INTEGER\0"
-    "i2v_AUTHORITY_INFO_ACCESS\0"
-    "notice_section\0"
-    "nref_nos\0"
-    "policy_section\0"
-    "process_pci_value\0"
-    "r2i_certpol\0"
-    "r2i_pci\0"
-    "s2i_ASN1_IA5STRING\0"
-    "s2i_ASN1_INTEGER\0"
-    "s2i_ASN1_OCTET_STRING\0"
-    "s2i_skey_id\0"
-    "set_dist_point_name\0"
-    "string_to_hex\0"
-    "v2i_ASN1_BIT_STRING\0"
-    "v2i_AUTHORITY_INFO_ACCESS\0"
-    "v2i_AUTHORITY_KEYID\0"
-    "v2i_BASIC_CONSTRAINTS\0"
-    "v2i_EXTENDED_KEY_USAGE\0"
-    "v2i_GENERAL_NAMES\0"
-    "v2i_GENERAL_NAME_ex\0"
-    "v2i_NAME_CONSTRAINTS\0"
-    "v2i_POLICY_CONSTRAINTS\0"
-    "v2i_POLICY_MAPPINGS\0"
-    "v2i_crld\0"
-    "v2i_idp\0"
-    "v2i_issuer_alt\0"
-    "v2i_subject_alt\0"
-    "v3_generic_extension\0"
-    "";
-
 const uint32_t kOpenSSLReasonValues[] = {
-    0xc3207ba,
-    0xc3287c7,
-    0xc3307d6,
-    0xc3387e6,
-    0xc3407f5,
-    0xc34880e,
-    0xc35081a,
-    0xc358837,
-    0xc360849,
-    0xc368857,
-    0xc370867,
-    0xc378874,
-    0xc380884,
-    0xc38888f,
-    0xc3908a5,
-    0xc3988b4,
-    0xc3a08c8,
-    0x1032146b,
-    0x10329477,
-    0x10331490,
-    0x103394a3,
-    0x10340dd4,
-    0x103494b6,
-    0x103514cb,
-    0x103594fd,
-    0x10361516,
-    0x1036952b,
-    0x10371549,
-    0x10379558,
-    0x10381574,
-    0x1038958f,
-    0x1039159e,
-    0x103995ba,
-    0x103a15d5,
-    0x103a95ec,
-    0x103b15fd,
-    0x103b9611,
-    0x103c1630,
-    0x103c963f,
-    0x103d1656,
-    0x103d9669,
-    0x103e0b5f,
-    0x103e969a,
-    0x103f16ad,
-    0x103f96c7,
-    0x104016d7,
-    0x104096eb,
-    0x10411701,
-    0x10419719,
-    0x1042172e,
-    0x10429742,
-    0x10431754,
-    0x104385d0,
-    0x104408b4,
-    0x10449769,
-    0x10451780,
-    0x10459795,
-    0x104617a3,
-    0x1046967c,
-    0x104714de,
-    0x14320b42,
-    0x14328b50,
-    0x14330b5f,
-    0x14338b71,
+    0xc320838,
+    0xc328852,
+    0xc330861,
+    0xc338871,
+    0xc340880,
+    0xc348899,
+    0xc3508a5,
+    0xc3588c2,
+    0xc3608d4,
+    0xc3688e2,
+    0xc3708f2,
+    0xc3788ff,
+    0xc38090f,
+    0xc38891a,
+    0xc390930,
+    0xc39893f,
+    0xc3a0953,
+    0xc3a8845,
+    0xc3b00ea,
+    0x10320845,
+    0x103293ab,
+    0x103313b7,
+    0x103393d0,
+    0x103413e3,
+    0x10348e8b,
+    0x10350c19,
+    0x103593f6,
+    0x1036140b,
+    0x1036941e,
+    0x1037143d,
+    0x10379456,
+    0x1038146b,
+    0x10389489,
+    0x10391498,
+    0x103994b4,
+    0x103a14cf,
+    0x103a94de,
+    0x103b14fa,
+    0x103b9515,
+    0x103c152c,
+    0x103c80ea,
+    0x103d153d,
+    0x103d9551,
+    0x103e1570,
+    0x103e957f,
+    0x103f1596,
+    0x103f95a9,
+    0x10400bea,
+    0x104095bc,
+    0x104115da,
+    0x104195ed,
+    0x10421607,
+    0x10429617,
+    0x1043162b,
+    0x10439641,
+    0x10441659,
+    0x1044966e,
+    0x10451682,
+    0x10459694,
+    0x104605fb,
+    0x1046893f,
+    0x104716a9,
+    0x104796c0,
+    0x104816d5,
+    0x104896e3,
+    0x14320bcd,
+    0x14328bdb,
+    0x14330bea,
+    0x14338bfc,
+    0x143400ac,
+    0x143480ea,
     0x18320083,
-    0x18328e3a,
-    0x18340e68,
-    0x18348e7c,
-    0x18358eb3,
-    0x18368ee0,
-    0x18370ef3,
-    0x18378f07,
-    0x18380f2b,
-    0x18388f39,
-    0x18390f4f,
-    0x18398f63,
-    0x183a0f73,
-    0x183b0f83,
-    0x183b8f98,
-    0x183c8fc3,
-    0x183d0fd7,
-    0x183d8fe7,
-    0x183e0b8e,
-    0x183e8ff4,
-    0x183f1006,
-    0x183f9011,
-    0x18401021,
-    0x18409032,
-    0x18411043,
-    0x18419055,
-    0x1842107e,
-    0x184290b0,
-    0x184310bf,
-    0x18451128,
-    0x1845913e,
-    0x18461159,
-    0x18468ecb,
-    0x184709cc,
-    0x18478094,
-    0x18480faf,
-    0x184890f4,
-    0x18490e50,
-    0x18498e91,
-    0x184a118f,
-    0x184a910c,
-    0x184b10d3,
-    0x184b8e2a,
-    0x184c1097,
-    0x184c866b,
-    0x184d1174,
-    0x203211b6,
-    0x243211c2,
-    0x243288fa,
-    0x243311d4,
-    0x243391e1,
-    0x243411ee,
-    0x24349200,
-    0x2435120f,
-    0x2435922c,
-    0x24361239,
-    0x24369247,
-    0x24371255,
-    0x24379263,
-    0x2438126c,
-    0x24389279,
-    0x2439128c,
-    0x28320b82,
-    0x28328b8e,
-    0x28330b5f,
-    0x28338ba1,
-    0x2c322b08,
-    0x2c32ab16,
-    0x2c332b28,
-    0x2c33ab3a,
-    0x2c342b4e,
-    0x2c34ab60,
-    0x2c352b7b,
-    0x2c35ab8d,
-    0x2c362ba0,
-    0x2c3682f3,
-    0x2c372bad,
-    0x2c37abbf,
-    0x2c382bd2,
-    0x2c38abe0,
-    0x2c392bf0,
-    0x2c39ac02,
-    0x2c3a2c16,
-    0x2c3aac27,
-    0x2c3b134c,
-    0x2c3bac38,
-    0x2c3c2c4c,
-    0x2c3cac62,
-    0x2c3d2c7b,
-    0x2c3daca9,
-    0x2c3e2cb7,
-    0x2c3eaccf,
-    0x2c3f2ce7,
-    0x2c3facf4,
-    0x2c402d17,
-    0x2c40ad36,
-    0x2c4111b6,
-    0x2c41ad47,
-    0x2c422d5a,
-    0x2c429128,
-    0x2c432d6b,
-    0x2c4386a2,
-    0x2c442c98,
+    0x18328ee1,
+    0x183300ac,
+    0x18338ef7,
+    0x18340f0b,
+    0x183480ea,
+    0x18350f20,
+    0x18358f38,
+    0x18360f4d,
+    0x18368f61,
+    0x18370f85,
+    0x18378f9b,
+    0x18380faf,
+    0x18388fbf,
+    0x18390a57,
+    0x18398fcf,
+    0x183a0fe4,
+    0x183a8ff8,
+    0x183b0c25,
+    0x183b9005,
+    0x183c1017,
+    0x183c9022,
+    0x183d1032,
+    0x183d9043,
+    0x183e1054,
+    0x183e9066,
+    0x183f108f,
+    0x183f90a8,
+    0x184010c0,
+    0x184086d3,
+    0x203210e7,
+    0x243210f3,
+    0x24328985,
+    0x24331105,
+    0x24339112,
+    0x2434111f,
+    0x24349131,
+    0x24351140,
+    0x2435915d,
+    0x2436116a,
+    0x24369178,
+    0x24371186,
+    0x24379194,
+    0x2438119d,
+    0x243891aa,
+    0x243911bd,
+    0x28320c0d,
+    0x28328c25,
+    0x28330bea,
+    0x28338c38,
+    0x28340c19,
+    0x283480ac,
+    0x283500ea,
+    0x2c3227cb,
+    0x2c32a7d9,
+    0x2c3327eb,
+    0x2c33a7fd,
+    0x2c342811,
+    0x2c34a823,
+    0x2c35283e,
+    0x2c35a850,
+    0x2c362863,
+    0x2c36832d,
+    0x2c372870,
+    0x2c37a882,
+    0x2c382895,
+    0x2c38a8ac,
+    0x2c3928ba,
+    0x2c39a8ca,
+    0x2c3a28dc,
+    0x2c3aa8f0,
+    0x2c3b2901,
+    0x2c3ba920,
+    0x2c3c2934,
+    0x2c3ca94a,
+    0x2c3d2963,
+    0x2c3da980,
+    0x2c3e2991,
+    0x2c3ea99f,
+    0x2c3f29b7,
+    0x2c3fa9cf,
+    0x2c4029dc,
+    0x2c4090e7,
+    0x2c4129ed,
+    0x2c41aa00,
+    0x2c4210c0,
+    0x2c42aa11,
+    0x2c430720,
+    0x2c43a912,
     0x30320000,
     0x30328015,
     0x3033001f,
@@ -1606,469 +223,454 @@
     0x3035006b,
     0x30358083,
     0x30360094,
-    0x303680a1,
-    0x303700b0,
-    0x303780bd,
-    0x303800d0,
-    0x303880eb,
-    0x30390100,
-    0x30398114,
-    0x303a0128,
-    0x303a8139,
-    0x303b0152,
-    0x303b816f,
-    0x303c017d,
-    0x303c8191,
-    0x303d01a1,
-    0x303d81ba,
-    0x303e01ca,
-    0x303e81dd,
-    0x303f01ec,
-    0x303f81f8,
-    0x3040020d,
-    0x3040821d,
-    0x30410234,
-    0x30418241,
-    0x30420254,
-    0x30428263,
-    0x30430278,
-    0x30438299,
-    0x304402ac,
-    0x304482bf,
-    0x304502d8,
-    0x304582f3,
-    0x30460310,
-    0x30468329,
-    0x30470337,
-    0x30478348,
-    0x30480357,
-    0x3048836f,
-    0x30490381,
-    0x30498395,
-    0x304a03b4,
-    0x304a83c7,
-    0x304b03d2,
-    0x304b83e1,
-    0x304c03f2,
-    0x304c83fe,
-    0x304d0414,
-    0x304d8422,
-    0x304e0438,
-    0x304e844a,
-    0x304f045c,
-    0x304f846f,
-    0x30500482,
-    0x30508493,
-    0x305104a3,
-    0x305184bb,
-    0x305204d0,
-    0x305284e8,
-    0x305304fc,
-    0x30538514,
-    0x3054052d,
-    0x30548546,
-    0x30550563,
-    0x3055856e,
-    0x30560586,
-    0x30568596,
-    0x305705a7,
-    0x305785ba,
-    0x305805d0,
-    0x305885d9,
-    0x305905ee,
-    0x30598601,
-    0x305a0610,
-    0x305a8630,
-    0x305b063f,
-    0x305b864b,
-    0x305c066b,
-    0x305c8687,
-    0x305d0698,
-    0x305d86a2,
-    0x34320abc,
-    0x34328ad0,
-    0x34330aed,
-    0x34338b00,
-    0x34340b0f,
-    0x34348b2c,
+    0x303680ac,
+    0x303700b9,
+    0x303780c8,
+    0x303800ea,
+    0x303880f7,
+    0x3039010a,
+    0x30398125,
+    0x303a013a,
+    0x303a814e,
+    0x303b0162,
+    0x303b8173,
+    0x303c018c,
+    0x303c81a9,
+    0x303d01b7,
+    0x303d81cb,
+    0x303e01db,
+    0x303e81f4,
+    0x303f0204,
+    0x303f8217,
+    0x30400226,
+    0x30408232,
+    0x30410247,
+    0x30418257,
+    0x3042026e,
+    0x3042827b,
+    0x3043028e,
+    0x3043829d,
+    0x304402b2,
+    0x304482d3,
+    0x304502e6,
+    0x304582f9,
+    0x30460312,
+    0x3046832d,
+    0x3047034a,
+    0x30478363,
+    0x30480371,
+    0x30488382,
+    0x30490391,
+    0x304983a9,
+    0x304a03bb,
+    0x304a83cf,
+    0x304b03ee,
+    0x304b8401,
+    0x304c040c,
+    0x304c841d,
+    0x304d0429,
+    0x304d843f,
+    0x304e044d,
+    0x304e8463,
+    0x304f0475,
+    0x304f8487,
+    0x3050049a,
+    0x305084ad,
+    0x305104be,
+    0x305184ce,
+    0x305204e6,
+    0x305284fb,
+    0x30530513,
+    0x30538527,
+    0x3054053f,
+    0x30548558,
+    0x30550571,
+    0x3055858e,
+    0x30560599,
+    0x305685b1,
+    0x305705c1,
+    0x305785d2,
+    0x305805e5,
+    0x305885fb,
+    0x30590604,
+    0x30598619,
+    0x305a062c,
+    0x305a863b,
+    0x305b065b,
+    0x305b866a,
+    0x305c068b,
+    0x305c86a7,
+    0x305d06b3,
+    0x305d86d3,
+    0x305e06ef,
+    0x305e8700,
+    0x305f0716,
+    0x305f8720,
+    0x34320b47,
+    0x34328b5b,
+    0x34330b78,
+    0x34338b8b,
+    0x34340b9a,
+    0x34348bb7,
     0x3c320083,
-    0x3c328bcb,
-    0x3c330be4,
-    0x3c338bff,
-    0x3c340c1c,
-    0x3c348c37,
-    0x3c350c52,
-    0x3c358c67,
-    0x3c360c80,
-    0x3c368c98,
-    0x3c370ca9,
-    0x3c378cb7,
-    0x3c380cc4,
-    0x3c388cd8,
-    0x3c390b8e,
-    0x3c398cec,
-    0x3c3a0d00,
-    0x3c3a8874,
-    0x3c3b0d10,
-    0x3c3b8d2b,
-    0x3c3c0d3d,
-    0x3c3c8d53,
-    0x3c3d0d5d,
-    0x3c3d8d71,
-    0x3c3e0d7f,
-    0x3c3e8da4,
-    0x3c3f0bb7,
-    0x3c3f8d8d,
-    0x403217ba,
-    0x403297d0,
-    0x403317fe,
-    0x40339808,
-    0x4034181f,
-    0x4034983d,
-    0x4035184d,
-    0x4035985f,
-    0x4036186c,
-    0x40369878,
-    0x4037188d,
-    0x403798a2,
-    0x403818b4,
-    0x403898bf,
-    0x403918d1,
-    0x40398dd4,
-    0x403a18e1,
-    0x403a98f4,
-    0x403b1915,
-    0x403b9926,
-    0x403c1936,
-    0x403c8064,
-    0x403d1942,
-    0x403d995e,
-    0x403e1974,
-    0x403e9983,
-    0x403f1996,
-    0x403f99b0,
-    0x404019be,
-    0x404099d3,
-    0x404119e7,
-    0x40419a04,
-    0x40421a1d,
-    0x40429a38,
-    0x40431a51,
-    0x40439a64,
-    0x40441a78,
-    0x40449a90,
-    0x40451aa0,
-    0x40459aae,
-    0x40461acc,
-    0x40468094,
-    0x40471ae1,
-    0x40479af3,
-    0x40481b17,
-    0x40489b37,
-    0x40491b4b,
-    0x40499b60,
-    0x404a1b79,
-    0x404a9bb3,
-    0x404b1bcd,
-    0x404b9beb,
-    0x404c1c06,
-    0x404c9c20,
-    0x404d1c37,
-    0x404d9c5f,
-    0x404e1c76,
-    0x404e9c92,
-    0x404f1cae,
-    0x404f9ccf,
-    0x40501cf1,
-    0x40509d0d,
-    0x40511d21,
-    0x40519d2e,
-    0x40521d45,
-    0x40529d55,
-    0x40531d65,
-    0x40539d79,
-    0x40541d94,
-    0x40549da4,
-    0x40551dbb,
-    0x40559dca,
-    0x40561de5,
-    0x40569dfd,
-    0x40571e19,
-    0x40579e32,
-    0x40581e45,
-    0x40589e5a,
-    0x40591e7d,
-    0x40599e8b,
-    0x405a1e98,
-    0x405a9eb1,
-    0x405b1ec9,
-    0x405b9edc,
-    0x405c1ef1,
-    0x405c9f03,
-    0x405d1f18,
-    0x405d9f28,
-    0x405e1f41,
-    0x405e9f55,
-    0x405f1f65,
-    0x405f9f7d,
-    0x40601f8e,
-    0x40609fa1,
-    0x40611fb2,
-    0x40619fd0,
-    0x40621fe1,
-    0x40629fee,
-    0x40632005,
-    0x4063a046,
-    0x4064205d,
-    0x4064a06a,
-    0x40652078,
-    0x4065a09a,
-    0x406620c2,
-    0x4066a0d7,
-    0x406720ee,
-    0x4067a0ff,
-    0x40682110,
-    0x4068a121,
-    0x40692136,
-    0x4069a14d,
-    0x406a215e,
-    0x406aa177,
-    0x406b2192,
-    0x406ba1a9,
-    0x406c2216,
-    0x406ca237,
-    0x406d224a,
-    0x406da26b,
-    0x406e2286,
-    0x406ea2a1,
-    0x406f22c2,
-    0x406fa2e8,
-    0x40702308,
-    0x4070a324,
-    0x407124b1,
-    0x4071a4d4,
-    0x407224ea,
-    0x4072a509,
-    0x40732521,
-    0x4073a541,
-    0x4074276b,
-    0x4074a790,
-    0x407527ab,
-    0x4075a7ca,
-    0x407627f9,
-    0x4076a821,
-    0x40772852,
-    0x4077a871,
-    0x40782896,
-    0x4078a8ad,
-    0x407928c0,
-    0x4079a8dd,
-    0x407a0782,
-    0x407aa8ef,
-    0x407b2902,
-    0x407ba91b,
-    0x407c2933,
-    0x407c90b0,
-    0x407d2947,
-    0x407da961,
-    0x407e2972,
-    0x407ea986,
-    0x407f2994,
-    0x407fa9af,
-    0x40801279,
-    0x4080a9d4,
-    0x408129f6,
-    0x4081aa11,
-    0x40822a26,
-    0x4082aa3e,
-    0x40832a56,
-    0x4083aa6d,
-    0x40842a83,
-    0x4084aa8f,
-    0x40852aa2,
-    0x4085aab7,
-    0x40862ac9,
-    0x4086aade,
-    0x40872ae7,
-    0x40879c4d,
-    0x40880083,
-    0x4088a025,
-    0x40890a0a,
-    0x4089a1c1,
-    0x408a1b9c,
-    0x408aa1eb,
-    0x408b283a,
-    0x41f423dc,
-    0x41f9246e,
-    0x41fe2361,
-    0x41fea592,
-    0x41ff2683,
-    0x420323f5,
-    0x42082417,
-    0x4208a453,
-    0x42092345,
-    0x4209a48d,
-    0x420a239c,
-    0x420aa37c,
-    0x420b23bc,
-    0x420ba435,
-    0x420c269f,
-    0x420ca55f,
-    0x420d2579,
-    0x420da5b0,
-    0x421225ca,
-    0x42172666,
-    0x4217a60c,
-    0x421c262e,
-    0x421f25e9,
-    0x422126b6,
-    0x42262649,
-    0x422b274f,
-    0x422ba718,
-    0x422c2737,
-    0x422ca6f2,
-    0x422d26d1,
-    0x443206ad,
-    0x443286bc,
-    0x443306c8,
-    0x443386d6,
-    0x443406e9,
-    0x443486fa,
-    0x44350701,
-    0x4435870b,
-    0x4436071e,
-    0x44368734,
-    0x44370746,
-    0x44378753,
-    0x44380762,
-    0x4438876a,
-    0x44390782,
-    0x44398790,
-    0x443a07a3,
-    0x4c3212a3,
-    0x4c3292b3,
-    0x4c3312c6,
-    0x4c3392e6,
-    0x4c340094,
-    0x4c3480b0,
-    0x4c3512f2,
-    0x4c359300,
-    0x4c36131c,
-    0x4c36932f,
-    0x4c37133e,
-    0x4c37934c,
-    0x4c381361,
-    0x4c38936d,
-    0x4c39138d,
-    0x4c3993b7,
-    0x4c3a13d0,
-    0x4c3a93e9,
-    0x4c3b05d0,
-    0x4c3b9402,
-    0x4c3c1414,
-    0x4c3c9423,
-    0x4c3d10b0,
-    0x4c3d943c,
-    0x4c3e1449,
-    0x50322d7d,
-    0x5032ad8c,
-    0x50332d97,
-    0x5033ada7,
-    0x50342dc0,
-    0x5034adda,
-    0x50352de8,
-    0x5035adfe,
-    0x50362e10,
-    0x5036ae26,
-    0x50372e3f,
-    0x5037ae52,
-    0x50382e6a,
-    0x5038ae7b,
-    0x50392e90,
-    0x5039aea4,
-    0x503a2ec4,
-    0x503aaeda,
-    0x503b2ef2,
-    0x503baf04,
-    0x503c2f20,
-    0x503caf37,
-    0x503d2f50,
-    0x503daf66,
-    0x503e2f73,
-    0x503eaf89,
-    0x503f2f9b,
-    0x503f8348,
-    0x50402fae,
-    0x5040afbe,
-    0x50412fd8,
-    0x5041afe7,
-    0x50423001,
-    0x5042b01e,
-    0x5043302e,
-    0x5043b03e,
-    0x5044304d,
-    0x50448414,
-    0x50453061,
-    0x5045b07f,
-    0x50463092,
-    0x5046b0a8,
-    0x504730ba,
-    0x5047b0cf,
-    0x504830f5,
-    0x5048b103,
-    0x50493116,
-    0x5049b12b,
-    0x504a3141,
-    0x504ab151,
-    0x504b3171,
-    0x504bb184,
-    0x504c31a7,
-    0x504cb1d5,
-    0x504d31e7,
-    0x504db204,
-    0x504e321f,
-    0x504eb23b,
-    0x504f324d,
-    0x504fb264,
-    0x50503273,
-    0x50508687,
-    0x50513286,
-    0x58320e12,
-    0x68320dd4,
-    0x68328b8e,
-    0x68330ba1,
-    0x68338de2,
-    0x68340df2,
-    0x6c320db0,
-    0x6c328b71,
-    0x6c330dbb,
-    0x74320980,
-    0x783208e5,
-    0x783288fa,
-    0x78330906,
+    0x3c328c62,
+    0x3c330c7b,
+    0x3c338c96,
+    0x3c340cb3,
+    0x3c348cdd,
+    0x3c350cf8,
+    0x3c358d1e,
+    0x3c360d37,
+    0x3c368d4f,
+    0x3c370d60,
+    0x3c378d6e,
+    0x3c380d7b,
+    0x3c388d8f,
+    0x3c390c25,
+    0x3c398da3,
+    0x3c3a0db7,
+    0x3c3a88ff,
+    0x3c3b0dc7,
+    0x3c3b8de2,
+    0x3c3c0df4,
+    0x3c3c8e0a,
+    0x3c3d0e14,
+    0x3c3d8e28,
+    0x3c3e0e36,
+    0x3c3e8e5b,
+    0x3c3f0c4e,
+    0x3c3f8e44,
+    0x3c4000ac,
+    0x3c4080ea,
+    0x3c410cce,
+    0x3c418d0d,
+    0x403216fa,
+    0x40329710,
+    0x4033173e,
+    0x40339748,
+    0x4034175f,
+    0x4034977d,
+    0x4035178d,
+    0x4035979f,
+    0x403617ac,
+    0x403697b8,
+    0x403717cd,
+    0x403797df,
+    0x403817ea,
+    0x403897fc,
+    0x40390e8b,
+    0x4039980c,
+    0x403a181f,
+    0x403a9840,
+    0x403b1851,
+    0x403b9861,
+    0x403c0064,
+    0x403c8083,
+    0x403d186d,
+    0x403d9883,
+    0x403e1892,
+    0x403e98a5,
+    0x403f18bf,
+    0x403f98cd,
+    0x404018e2,
+    0x404098f6,
+    0x40411913,
+    0x4041992e,
+    0x40421947,
+    0x4042995a,
+    0x4043196e,
+    0x40439986,
+    0x4044199d,
+    0x404480ac,
+    0x404519b2,
+    0x404599c4,
+    0x404619e8,
+    0x40469a08,
+    0x40471a16,
+    0x40479a3d,
+    0x40481a52,
+    0x40489a6b,
+    0x40491a82,
+    0x40499a9c,
+    0x404a1ab3,
+    0x404a9ad1,
+    0x404b1ae9,
+    0x404b9b00,
+    0x404c1b16,
+    0x404c9b28,
+    0x404d1b49,
+    0x404d9b6b,
+    0x404e1b7f,
+    0x404e9b8c,
+    0x404f1ba3,
+    0x404f9bb3,
+    0x40501bdd,
+    0x40509bf1,
+    0x40511c0c,
+    0x40519c1c,
+    0x40521c33,
+    0x40529c45,
+    0x40531c5d,
+    0x40539c70,
+    0x40541c85,
+    0x40549ca8,
+    0x40551cb6,
+    0x40559cd3,
+    0x40561ce0,
+    0x40569cf9,
+    0x40571d11,
+    0x40579d24,
+    0x40581d39,
+    0x40589d4b,
+    0x40591d7a,
+    0x40599d93,
+    0x405a1da7,
+    0x405a9db7,
+    0x405b1dcf,
+    0x405b9de0,
+    0x405c1df3,
+    0x405c9e04,
+    0x405d1e11,
+    0x405d9e28,
+    0x405e1e48,
+    0x405e8a95,
+    0x405f1e69,
+    0x405f9e76,
+    0x40601e84,
+    0x40609ea6,
+    0x40611ece,
+    0x40619ee3,
+    0x40621efa,
+    0x40629f0b,
+    0x40631f1c,
+    0x40639f31,
+    0x40641f48,
+    0x40649f59,
+    0x40651f74,
+    0x40659f8b,
+    0x40661fa3,
+    0x40669fcd,
+    0x40671ff8,
+    0x4067a019,
+    0x4068202c,
+    0x4068a04d,
+    0x4069207f,
+    0x4069a0ad,
+    0x406a20ce,
+    0x406aa0ee,
+    0x406b2276,
+    0x406ba299,
+    0x406c22af,
+    0x406ca4db,
+    0x406d250a,
+    0x406da532,
+    0x406e254b,
+    0x406ea563,
+    0x406f2582,
+    0x406fa597,
+    0x407025aa,
+    0x4070a5c7,
+    0x40710800,
+    0x4071a5d9,
+    0x407225ec,
+    0x4072a605,
+    0x4073261d,
+    0x4073936d,
+    0x40742631,
+    0x4074a64b,
+    0x4075265c,
+    0x4075a670,
+    0x4076267e,
+    0x407691aa,
+    0x407726a3,
+    0x4077a6c5,
+    0x407826e0,
+    0x4078a719,
+    0x40792730,
+    0x4079a746,
+    0x407a2752,
+    0x407aa765,
+    0x407b277a,
+    0x407ba78c,
+    0x407c27a1,
+    0x407ca7aa,
+    0x407d2068,
+    0x407d9bc3,
+    0x407e26f5,
+    0x407e9d5b,
+    0x407f1a2a,
+    0x41f421a1,
+    0x41f92233,
+    0x41fe2126,
+    0x41fea302,
+    0x41ff23f3,
+    0x420321ba,
+    0x420821dc,
+    0x4208a218,
+    0x4209210a,
+    0x4209a252,
+    0x420a2161,
+    0x420aa141,
+    0x420b2181,
+    0x420ba1fa,
+    0x420c240f,
+    0x420ca2cf,
+    0x420d22e9,
+    0x420da320,
+    0x4212233a,
+    0x421723d6,
+    0x4217a37c,
+    0x421c239e,
+    0x421f2359,
+    0x42212426,
+    0x422623b9,
+    0x422b24bf,
+    0x422ba488,
+    0x422c24a7,
+    0x422ca462,
+    0x422d2441,
+    0x4432072b,
+    0x4432873a,
+    0x44330746,
+    0x44338754,
+    0x44340767,
+    0x44348778,
+    0x4435077f,
+    0x44358789,
+    0x4436079c,
+    0x443687b2,
+    0x443707c4,
+    0x443787d1,
+    0x443807e0,
+    0x443887e8,
+    0x44390800,
+    0x4439880e,
+    0x443a0821,
+    0x4c3211d4,
+    0x4c3291e4,
+    0x4c3311f7,
+    0x4c339217,
+    0x4c3400ac,
+    0x4c3480ea,
+    0x4c351223,
+    0x4c359231,
+    0x4c36124d,
+    0x4c369260,
+    0x4c37126f,
+    0x4c37927d,
+    0x4c381292,
+    0x4c38929e,
+    0x4c3912be,
+    0x4c3992e8,
+    0x4c3a1301,
+    0x4c3a931a,
+    0x4c3b05fb,
+    0x4c3b9333,
+    0x4c3c1345,
+    0x4c3c9354,
+    0x4c3d136d,
+    0x4c3d937c,
+    0x4c3e1389,
+    0x50322a23,
+    0x5032aa32,
+    0x50332a3d,
+    0x5033aa4d,
+    0x50342a66,
+    0x5034aa80,
+    0x50352a8e,
+    0x5035aaa4,
+    0x50362ab6,
+    0x5036aacc,
+    0x50372ae5,
+    0x5037aaf8,
+    0x50382b10,
+    0x5038ab21,
+    0x50392b36,
+    0x5039ab4a,
+    0x503a2b6a,
+    0x503aab80,
+    0x503b2b98,
+    0x503babaa,
+    0x503c2bc6,
+    0x503cabdd,
+    0x503d2bf6,
+    0x503dac0c,
+    0x503e2c19,
+    0x503eac2f,
+    0x503f2c41,
+    0x503f8382,
+    0x50402c54,
+    0x5040ac64,
+    0x50412c7e,
+    0x5041ac8d,
+    0x50422ca7,
+    0x5042acc4,
+    0x50432cd4,
+    0x5043ace4,
+    0x50442cf3,
+    0x5044843f,
+    0x50452d07,
+    0x5045ad25,
+    0x50462d38,
+    0x5046ad4e,
+    0x50472d60,
+    0x5047ad75,
+    0x50482d9b,
+    0x5048ada9,
+    0x50492dbc,
+    0x5049add1,
+    0x504a2de7,
+    0x504aadf7,
+    0x504b2e17,
+    0x504bae2a,
+    0x504c2e4d,
+    0x504cae7b,
+    0x504d2e8d,
+    0x504daeaa,
+    0x504e2ec5,
+    0x504eaee1,
+    0x504f2ef3,
+    0x504faf0a,
+    0x50502f19,
+    0x505086ef,
+    0x50512f2c,
+    0x58320ec9,
+    0x68320e8b,
+    0x68328c25,
+    0x68330c38,
+    0x68338e99,
+    0x68340ea9,
+    0x683480ea,
+    0x6c320e67,
+    0x6c328bfc,
+    0x6c330e72,
+    0x74320a0b,
+    0x78320970,
+    0x78328985,
+    0x78330991,
     0x78338083,
-    0x78340915,
-    0x7834892a,
-    0x78350949,
-    0x7835896b,
-    0x78360980,
-    0x78368996,
-    0x783709a6,
-    0x783789b9,
-    0x783809cc,
-    0x783889de,
-    0x783909eb,
-    0x78398a0a,
-    0x783a0a1f,
-    0x783a8a2d,
-    0x783b0a37,
-    0x783b8a4b,
-    0x783c0a62,
-    0x783c8a77,
-    0x783d0a8e,
-    0x783d8aa3,
-    0x783e09f9,
-    0x803211a5,
+    0x783409a0,
+    0x783489b5,
+    0x783509d4,
+    0x783589f6,
+    0x78360a0b,
+    0x78368a21,
+    0x78370a31,
+    0x78378a44,
+    0x78380a57,
+    0x78388a69,
+    0x78390a76,
+    0x78398a95,
+    0x783a0aaa,
+    0x783a8ab8,
+    0x783b0ac2,
+    0x783b8ad6,
+    0x783c0aed,
+    0x783c8b02,
+    0x783d0b19,
+    0x783d8b2e,
+    0x783e0a84,
+    0x7c3210d6,
 };
 
 const size_t kOpenSSLReasonValuesLen = sizeof(kOpenSSLReasonValues) / sizeof(kOpenSSLReasonValues[0]);
@@ -2082,8 +684,10 @@
     "BN_LIB\0"
     "BOOLEAN_IS_WRONG_LENGTH\0"
     "BUFFER_TOO_SMALL\0"
+    "CONTEXT_NOT_INITIALISED\0"
     "DECODE_ERROR\0"
     "DEPTH_EXCEEDED\0"
+    "DIGEST_AND_KEY_TYPE_NOT_SUPPORTED\0"
     "ENCODE_ERROR\0"
     "ERROR_GETTING_TIME\0"
     "EXPECTING_AN_ASN1_SEQUENCE\0"
@@ -2124,7 +728,6 @@
     "INVALID_UNIVERSALSTRING_LENGTH\0"
     "INVALID_UTF8STRING\0"
     "LIST_ERROR\0"
-    "MALLOC_FAILURE\0"
     "MISSING_ASN1_EOS\0"
     "MISSING_EOC\0"
     "MISSING_SECOND_NUMBER\0"
@@ -2156,10 +759,13 @@
     "UNEXPECTED_EOC\0"
     "UNIVERSALSTRING_IS_WRONG_LENGTH\0"
     "UNKNOWN_FORMAT\0"
+    "UNKNOWN_MESSAGE_DIGEST_ALGORITHM\0"
+    "UNKNOWN_SIGNATURE_ALGORITHM\0"
     "UNKNOWN_TAG\0"
     "UNSUPPORTED_ANY_DEFINED_BY_TYPE\0"
     "UNSUPPORTED_PUBLIC_KEY_TYPE\0"
     "UNSUPPORTED_TYPE\0"
+    "WRONG_PUBLIC_KEY_TYPE\0"
     "WRONG_TAG\0"
     "WRONG_TYPE\0"
     "BAD_FOPEN_MODE\0"
@@ -2180,6 +786,7 @@
     "UNSUPPORTED_METHOD\0"
     "WRITE_TO_READ_ONLY_BIO\0"
     "ARG2_LT_ARG3\0"
+    "BAD_ENCODING\0"
     "BAD_RECIPROCAL\0"
     "BIGNUM_TOO_LONG\0"
     "BITS_TOO_SMALL\0"
@@ -2231,6 +838,7 @@
     "MODULUS_TOO_LARGE\0"
     "NO_PRIVATE_VALUE\0"
     "BAD_Q_VALUE\0"
+    "BAD_VERSION\0"
     "MISSING_PARAMETERS\0"
     "NEED_NEW_SETUP_VALUES\0"
     "BIGNUM_OUT_OF_RANGE\0"
@@ -2238,8 +846,10 @@
     "D2I_ECPKPARAMETERS_FAILURE\0"
     "EC_GROUP_NEW_BY_NAME_FAILURE\0"
     "GROUP2PKPARAMETERS_FAILURE\0"
+    "GROUP_MISMATCH\0"
     "I2D_ECPKPARAMETERS_FAILURE\0"
     "INCOMPATIBLE_OBJECTS\0"
+    "INVALID_COFACTOR\0"
     "INVALID_COMPRESSED_POINT\0"
     "INVALID_COMPRESSION_BIT\0"
     "INVALID_ENCODING\0"
@@ -2264,27 +874,19 @@
     "NOT_IMPLEMENTED\0"
     "RANDOM_NUMBER_GENERATION_FAILED\0"
     "OPERATION_NOT_SUPPORTED\0"
-    "BN_DECODE_ERROR\0"
     "COMMAND_NOT_SUPPORTED\0"
-    "CONTEXT_NOT_INITIALISED\0"
     "DIFFERENT_KEY_TYPES\0"
     "DIFFERENT_PARAMETERS\0"
-    "DIGEST_AND_KEY_TYPE_NOT_SUPPORTED\0"
     "EXPECTING_AN_EC_KEY_KEY\0"
     "EXPECTING_AN_RSA_KEY\0"
-    "EXPECTING_A_DH_KEY\0"
     "EXPECTING_A_DSA_KEY\0"
     "ILLEGAL_OR_UNSUPPORTED_PADDING_MODE\0"
-    "INVALID_CURVE\0"
     "INVALID_DIGEST_LENGTH\0"
     "INVALID_DIGEST_TYPE\0"
     "INVALID_KEYBITS\0"
     "INVALID_MGF1_MD\0"
     "INVALID_PADDING_MODE\0"
-    "INVALID_PSS_PARAMETERS\0"
     "INVALID_PSS_SALTLEN\0"
-    "INVALID_SALT_LENGTH\0"
-    "INVALID_TRAILER\0"
     "KEYS_NOT_SET\0"
     "NO_DEFAULT_DIGEST\0"
     "NO_KEY_SET\0"
@@ -2294,17 +896,8 @@
     "NO_PARAMETERS_SET\0"
     "OPERATION_NOT_SUPPORTED_FOR_THIS_KEYTYPE\0"
     "OPERATON_NOT_INITIALIZED\0"
-    "PARAMETER_ENCODING_ERROR\0"
-    "UNKNOWN_DIGEST\0"
-    "UNKNOWN_MASK_DIGEST\0"
-    "UNKNOWN_MESSAGE_DIGEST_ALGORITHM\0"
     "UNKNOWN_PUBLIC_KEY_TYPE\0"
-    "UNKNOWN_SIGNATURE_ALGORITHM\0"
     "UNSUPPORTED_ALGORITHM\0"
-    "UNSUPPORTED_MASK_ALGORITHM\0"
-    "UNSUPPORTED_MASK_PARAMETER\0"
-    "UNSUPPORTED_SIGNATURE_TYPE\0"
-    "WRONG_PUBLIC_KEY_TYPE\0"
     "OUTPUT_TOO_LARGE\0"
     "UNKNOWN_NID\0"
     "BAD_BASE64_DECODE\0"
@@ -2340,6 +933,7 @@
     "UNKNOWN_ALGORITHM\0"
     "UNKNOWN_CIPHER\0"
     "UNKNOWN_CIPHER_ALGORITHM\0"
+    "UNKNOWN_DIGEST\0"
     "UNKNOWN_HASH\0"
     "UNSUPPORTED_PRIVATE_KEY_ALGORITHM\0"
     "BAD_E_VALUE\0"
@@ -2390,7 +984,6 @@
     "BAD_DIGEST_LENGTH\0"
     "BAD_ECC_CERT\0"
     "BAD_ECPOINT\0"
-    "BAD_HANDSHAKE_LENGTH\0"
     "BAD_HANDSHAKE_RECORD\0"
     "BAD_HELLO_REQUEST\0"
     "BAD_LENGTH\0"
@@ -2401,7 +994,6 @@
     "BAD_SSL_FILETYPE\0"
     "BAD_WRITE_RETRY\0"
     "BIO_NOT_SET\0"
-    "CANNOT_SERIALIZE_PUBLIC_KEY\0"
     "CA_DN_LENGTH_MISMATCH\0"
     "CA_DN_TOO_LONG\0"
     "CCS_RECEIVED_EARLY\0"
@@ -2410,63 +1002,56 @@
     "CERT_LENGTH_MISMATCH\0"
     "CHANNEL_ID_NOT_P256\0"
     "CHANNEL_ID_SIGNATURE_INVALID\0"
-    "CIPHER_CODE_WRONG_LENGTH\0"
     "CIPHER_OR_HASH_UNAVAILABLE\0"
     "CLIENTHELLO_PARSE_FAILED\0"
     "CLIENTHELLO_TLSEXT\0"
     "CONNECTION_REJECTED\0"
     "CONNECTION_TYPE_NOT_SET\0"
-    "COOKIE_MISMATCH\0"
-    "D2I_ECDSA_SIG\0"
-    "DATA_BETWEEN_CCS_AND_FINISHED\0"
+    "CUSTOM_EXTENSION_ERROR\0"
     "DATA_LENGTH_TOO_LONG\0"
     "DECRYPTION_FAILED\0"
     "DECRYPTION_FAILED_OR_BAD_RECORD_MAC\0"
     "DH_PUBLIC_VALUE_LENGTH_IS_WRONG\0"
+    "DH_P_TOO_LONG\0"
     "DIGEST_CHECK_FAILED\0"
+    "DOWNGRADE_DETECTED\0"
     "DTLS_MESSAGE_TOO_BIG\0"
     "ECC_CERT_NOT_FOR_SIGNING\0"
-    "EMPTY_SRTP_PROTECTION_PROFILE_LIST\0"
     "EMS_STATE_INCONSISTENT\0"
     "ENCRYPTED_LENGTH_TOO_LONG\0"
+    "ERROR_ADDING_EXTENSION\0"
     "ERROR_IN_RECEIVED_CIPHER_LIST\0"
-    "EVP_DIGESTSIGNFINAL_FAILED\0"
-    "EVP_DIGESTSIGNINIT_FAILED\0"
+    "ERROR_PARSING_EXTENSION\0"
     "EXCESSIVE_MESSAGE_SIZE\0"
     "EXTRA_DATA_IN_MESSAGE\0"
     "FRAGMENT_MISMATCH\0"
-    "GOT_A_FIN_BEFORE_A_CCS\0"
-    "GOT_CHANNEL_ID_BEFORE_A_CCS\0"
-    "GOT_NEXT_PROTO_BEFORE_A_CCS\0"
     "GOT_NEXT_PROTO_WITHOUT_EXTENSION\0"
     "HANDSHAKE_FAILURE_ON_CLIENT_HELLO\0"
-    "HANDSHAKE_RECORD_BEFORE_CCS\0"
     "HTTPS_PROXY_REQUEST\0"
     "HTTP_REQUEST\0"
     "INAPPROPRIATE_FALLBACK\0"
     "INVALID_COMMAND\0"
     "INVALID_MESSAGE\0"
+    "INVALID_OUTER_RECORD_TYPE\0"
     "INVALID_SSL_SESSION\0"
     "INVALID_TICKET_KEYS_LENGTH\0"
     "LENGTH_MISMATCH\0"
     "LIBRARY_HAS_NO_CIPHERS\0"
-    "MISSING_DH_KEY\0"
-    "MISSING_ECDSA_SIGNING_CERT\0"
+    "MISSING_EXTENSION\0"
     "MISSING_RSA_CERTIFICATE\0"
-    "MISSING_RSA_ENCRYPTING_CERT\0"
-    "MISSING_RSA_SIGNING_CERT\0"
     "MISSING_TMP_DH_KEY\0"
     "MISSING_TMP_ECDH_KEY\0"
     "MIXED_SPECIAL_OPERATOR_WITH_GROUPS\0"
     "MTU_TOO_SMALL\0"
+    "NEGOTIATED_BOTH_NPN_AND_ALPN\0"
     "NESTED_GROUP\0"
     "NO_CERTIFICATES_RETURNED\0"
     "NO_CERTIFICATE_ASSIGNED\0"
     "NO_CERTIFICATE_SET\0"
     "NO_CIPHERS_AVAILABLE\0"
     "NO_CIPHERS_PASSED\0"
-    "NO_CIPHERS_SPECIFIED\0"
     "NO_CIPHER_MATCH\0"
+    "NO_COMMON_SIGNATURE_ALGORITHMS\0"
     "NO_COMPRESSION_SPECIFIED\0"
     "NO_METHOD_SPECIFIED\0"
     "NO_P256_SUPPORT\0"
@@ -2474,13 +1059,10 @@
     "NO_RENEGOTIATION\0"
     "NO_REQUIRED_DIGEST\0"
     "NO_SHARED_CIPHER\0"
-    "NO_SHARED_SIGATURE_ALGORITHMS\0"
-    "NO_SRTP_PROFILES\0"
     "NULL_SSL_CTX\0"
     "NULL_SSL_METHOD_PASSED\0"
     "OLD_SESSION_CIPHER_NOT_RETURNED\0"
     "OLD_SESSION_VERSION_NOT_RETURNED\0"
-    "PACKET_LENGTH_TOO_LONG\0"
     "PARSE_TLSEXT\0"
     "PATH_TOO_LONG\0"
     "PEER_DID_NOT_RETURN_A_CERTIFICATE\0"
@@ -2489,11 +1071,9 @@
     "PSK_IDENTITY_NOT_FOUND\0"
     "PSK_NO_CLIENT_CB\0"
     "PSK_NO_SERVER_CB\0"
-    "READ_BIO_NOT_SET\0"
     "READ_TIMEOUT_EXPIRED\0"
     "RECORD_LENGTH_MISMATCH\0"
     "RECORD_TOO_LARGE\0"
-    "RENEGOTIATE_EXT_TOO_LONG\0"
     "RENEGOTIATION_ENCODING_ERR\0"
     "RENEGOTIATION_MISMATCH\0"
     "REQUIRED_CIPHER_MISSING\0"
@@ -2503,12 +1083,11 @@
     "SERVERHELLO_TLSEXT\0"
     "SESSION_ID_CONTEXT_UNINITIALIZED\0"
     "SESSION_MAY_NOT_BE_CREATED\0"
-    "SIGNATURE_ALGORITHMS_ERROR\0"
+    "SHUTDOWN_WHILE_IN_INIT\0"
+    "SIGNATURE_ALGORITHMS_EXTENSION_SENT_BY_SERVER\0"
     "SRTP_COULD_NOT_ALLOCATE_PROFILES\0"
-    "SRTP_PROTECTION_PROFILE_LIST_TOO_LONG\0"
     "SRTP_UNKNOWN_PROTECTION_PROFILE\0"
     "SSL3_EXT_INVALID_SERVERNAME\0"
-    "SSL3_EXT_INVALID_SERVERNAME_TYPE\0"
     "SSLV3_ALERT_BAD_CERTIFICATE\0"
     "SSLV3_ALERT_BAD_RECORD_MAC\0"
     "SSLV3_ALERT_CERTIFICATE_EXPIRED\0"
@@ -2523,10 +1102,7 @@
     "SSLV3_ALERT_UNSUPPORTED_CERTIFICATE\0"
     "SSL_CTX_HAS_NO_DEFAULT_SSL_VERSION\0"
     "SSL_HANDSHAKE_FAILURE\0"
-    "SSL_SESSION_ID_CALLBACK_FAILED\0"
-    "SSL_SESSION_ID_CONFLICT\0"
     "SSL_SESSION_ID_CONTEXT_TOO_LONG\0"
-    "SSL_SESSION_ID_HAS_BAD_LENGTH\0"
     "TLSV1_ALERT_ACCESS_DENIED\0"
     "TLSV1_ALERT_DECODE_ERROR\0"
     "TLSV1_ALERT_DECRYPTION_FAILED\0"
@@ -2545,16 +1121,12 @@
     "TLSV1_CERTIFICATE_UNOBTAINABLE\0"
     "TLSV1_UNRECOGNIZED_NAME\0"
     "TLSV1_UNSUPPORTED_EXTENSION\0"
-    "TLS_CLIENT_CERT_REQ_WITH_ANON_CIPHER\0"
-    "TLS_ILLEGAL_EXPORTER_LABEL\0"
-    "TLS_INVALID_ECPOINTFORMAT_LIST\0"
     "TLS_PEER_DID_NOT_RESPOND_WITH_CERTIFICATE_LIST\0"
     "TLS_RSA_ENCRYPTED_VALUE_LENGTH_IS_WRONG\0"
     "TOO_MANY_EMPTY_FRAGMENTS\0"
     "TOO_MANY_WARNING_ALERTS\0"
     "UNABLE_TO_FIND_ECDH_PARAMETERS\0"
-    "UNABLE_TO_FIND_PUBLIC_KEY_PARAMETERS\0"
-    "UNEXPECTED_GROUP_CLOSE\0"
+    "UNEXPECTED_EXTENSION\0"
     "UNEXPECTED_MESSAGE\0"
     "UNEXPECTED_OPERATOR_IN_GROUP\0"
     "UNEXPECTED_RECORD\0"
@@ -2566,13 +1138,11 @@
     "UNKNOWN_PROTOCOL\0"
     "UNKNOWN_SSL_VERSION\0"
     "UNKNOWN_STATE\0"
-    "UNPROCESSED_HANDSHAKE_DATA\0"
     "UNSAFE_LEGACY_RENEGOTIATION_DISABLED\0"
     "UNSUPPORTED_COMPRESSION_ALGORITHM\0"
     "UNSUPPORTED_ELLIPTIC_CURVE\0"
     "UNSUPPORTED_PROTOCOL\0"
-    "UNSUPPORTED_SSL_VERSION\0"
-    "USE_SRTP_NOT_NEGOTIATED\0"
+    "UNSUPPORTED_PROTOCOL_FOR_CUSTOM_KEY\0"
     "WRONG_CERTIFICATE_TYPE\0"
     "WRONG_CIPHER_RETURNED\0"
     "WRONG_CURVE\0"
@@ -2593,12 +1163,14 @@
     "IDP_MISMATCH\0"
     "INVALID_DIRECTORY\0"
     "INVALID_FIELD_NAME\0"
+    "INVALID_PSS_PARAMETERS\0"
     "INVALID_TRUST\0"
     "ISSUER_MISMATCH\0"
     "KEY_TYPE_MISMATCH\0"
     "KEY_VALUES_MISMATCH\0"
     "LOADING_CERT_DIR\0"
     "LOADING_DEFAULTS\0"
+    "NAME_TOO_LONG\0"
     "NEWER_CRL_NOT_NEWER\0"
     "NOT_PKCS7_SIGNED_DATA\0"
     "NO_CERTIFICATES_INCLUDED\0"
@@ -2608,8 +1180,6 @@
     "PUBLIC_KEY_DECODE_ERROR\0"
     "PUBLIC_KEY_ENCODE_ERROR\0"
     "SHOULD_RETRY\0"
-    "UNABLE_TO_FIND_PARAMETERS_IN_CHAIN\0"
-    "UNABLE_TO_GET_CERTS_PUBLIC_KEY\0"
     "UNKNOWN_KEY_TYPE\0"
     "UNKNOWN_PURPOSE_ID\0"
     "UNKNOWN_TRUST_ID\0"

third_party/boringssl/linux-aarch64/crypto/aes/aesv8-armx64.S

diff --git a/third_party/boringssl/linux-aarch64/crypto/aes/aesv8-armx64.S b/third_party/boringssl/linux-aarch64/crypto/aes/aesv8-armx64.S
index c414476..3e8cb16 100644
--- a/third_party/boringssl/linux-aarch64/crypto/aes/aesv8-armx64.S
+++ b/third_party/boringssl/linux-aarch64/crypto/aes/aesv8-armx64.S
@@ -1,5 +1,5 @@
 #if defined(__aarch64__)
-#include "arm_arch.h"
+#include <openssl/arm_arch.h>
 
 #if __ARM_MAX_ARCH__>=7
 .text
@@ -13,6 +13,7 @@
 .long	0x1b,0x1b,0x1b,0x1b
 
 .globl	aes_v8_set_encrypt_key
+.hidden	aes_v8_set_encrypt_key
 .type	aes_v8_set_encrypt_key,%function
 .align	5
 aes_v8_set_encrypt_key:
@@ -180,6 +181,7 @@
 .size	aes_v8_set_encrypt_key,.-aes_v8_set_encrypt_key
 
 .globl	aes_v8_set_decrypt_key
+.hidden	aes_v8_set_decrypt_key
 .type	aes_v8_set_decrypt_key,%function
 .align	5
 aes_v8_set_decrypt_key:
@@ -219,6 +221,7 @@
 	ret
 .size	aes_v8_set_decrypt_key,.-aes_v8_set_decrypt_key
 .globl	aes_v8_encrypt
+.hidden	aes_v8_encrypt
 .type	aes_v8_encrypt,%function
 .align	5
 aes_v8_encrypt:
@@ -248,6 +251,7 @@
 	ret
 .size	aes_v8_encrypt,.-aes_v8_encrypt
 .globl	aes_v8_decrypt
+.hidden	aes_v8_decrypt
 .type	aes_v8_decrypt,%function
 .align	5
 aes_v8_decrypt:
@@ -277,6 +281,7 @@
 	ret
 .size	aes_v8_decrypt,.-aes_v8_decrypt
 .globl	aes_v8_cbc_encrypt
+.hidden	aes_v8_cbc_encrypt
 .type	aes_v8_cbc_encrypt,%function
 .align	5
 aes_v8_cbc_encrypt:
@@ -567,6 +572,7 @@
 	ret
 .size	aes_v8_cbc_encrypt,.-aes_v8_cbc_encrypt
 .globl	aes_v8_ctr32_encrypt_blocks
+.hidden	aes_v8_ctr32_encrypt_blocks
 .type	aes_v8_ctr32_encrypt_blocks,%function
 .align	5
 aes_v8_ctr32_encrypt_blocks:
@@ -748,4 +754,4 @@
 	ret
 .size	aes_v8_ctr32_encrypt_blocks,.-aes_v8_ctr32_encrypt_blocks
 #endif
-#endif
\ No newline at end of file
+#endif

third_party/boringssl/linux-aarch64/crypto/bn/armv8-mont.S

diff --git a/third_party/boringssl/linux-aarch64/crypto/bn/armv8-mont.S b/third_party/boringssl/linux-aarch64/crypto/bn/armv8-mont.S
new file mode 100644
index 0000000..74702db
--- /dev/null
+++ b/third_party/boringssl/linux-aarch64/crypto/bn/armv8-mont.S
@@ -0,0 +1,1407 @@
+#if defined(__aarch64__)
+.text
+
+.globl	bn_mul_mont
+.hidden	bn_mul_mont
+.type	bn_mul_mont,%function
+.align	5
+bn_mul_mont:
+	tst	x5,#7
+	b.eq	__bn_sqr8x_mont
+	tst	x5,#3
+	b.eq	__bn_mul4x_mont
+.Lmul_mont:
+	stp	x29,x30,[sp,#-64]!
+	add	x29,sp,#0
+	stp	x19,x20,[sp,#16]
+	stp	x21,x22,[sp,#32]
+	stp	x23,x24,[sp,#48]
+
+	ldr	x9,[x2],#8		// bp[0]
+	sub	x22,sp,x5,lsl#3
+	ldp	x7,x8,[x1],#16	// ap[0..1]
+	lsl	x5,x5,#3
+	ldr	x4,[x4]		// *n0
+	and	x22,x22,#-16		// ABI says so
+	ldp	x13,x14,[x3],#16	// np[0..1]
+
+	mul	x6,x7,x9		// ap[0]*bp[0]
+	sub	x21,x5,#16		// j=num-2
+	umulh	x7,x7,x9
+	mul	x10,x8,x9		// ap[1]*bp[0]
+	umulh	x11,x8,x9
+
+	mul	x15,x6,x4		// "tp[0]"*n0
+	mov	sp,x22			// alloca
+
+	// (*)	mul	x12,x13,x15	// np[0]*m1
+	umulh	x13,x13,x15
+	mul	x16,x14,x15		// np[1]*m1
+	// (*)	adds	x12,x12,x6	// discarded
+	// (*)	As for removal of first multiplication and addition
+	//	instructions. The outcome of first addition is
+	//	guaranteed to be zero, which leaves two computationally
+	//	significant outcomes: it either carries or not. Then
+	//	question is when does it carry? Is there alternative
+	//	way to deduce it? If you follow operations, you can
+	//	observe that condition for carry is quite simple:
+	//	x6 being non-zero. So that carry can be calculated
+	//	by adding -1 to x6. That's what next instruction does.
+	subs	xzr,x6,#1		// (*)
+	umulh	x17,x14,x15
+	adc	x13,x13,xzr
+	cbz	x21,.L1st_skip
+
+.L1st:
+	ldr	x8,[x1],#8
+	adds	x6,x10,x7
+	sub	x21,x21,#8		// j--
+	adc	x7,x11,xzr
+
+	ldr	x14,[x3],#8
+	adds	x12,x16,x13
+	mul	x10,x8,x9		// ap[j]*bp[0]
+	adc	x13,x17,xzr
+	umulh	x11,x8,x9
+
+	adds	x12,x12,x6
+	mul	x16,x14,x15		// np[j]*m1
+	adc	x13,x13,xzr
+	umulh	x17,x14,x15
+	str	x12,[x22],#8		// tp[j-1]
+	cbnz	x21,.L1st
+
+.L1st_skip:
+	adds	x6,x10,x7
+	sub	x1,x1,x5		// rewind x1
+	adc	x7,x11,xzr
+
+	adds	x12,x16,x13
+	sub	x3,x3,x5		// rewind x3
+	adc	x13,x17,xzr
+
+	adds	x12,x12,x6
+	sub	x20,x5,#8		// i=num-1
+	adcs	x13,x13,x7
+
+	adc	x19,xzr,xzr		// upmost overflow bit
+	stp	x12,x13,[x22]
+
+.Louter:
+	ldr	x9,[x2],#8		// bp[i]
+	ldp	x7,x8,[x1],#16
+	ldr	x23,[sp]		// tp[0]
+	add	x22,sp,#8
+
+	mul	x6,x7,x9		// ap[0]*bp[i]
+	sub	x21,x5,#16		// j=num-2
+	umulh	x7,x7,x9
+	ldp	x13,x14,[x3],#16
+	mul	x10,x8,x9		// ap[1]*bp[i]
+	adds	x6,x6,x23
+	umulh	x11,x8,x9
+	adc	x7,x7,xzr
+
+	mul	x15,x6,x4
+	sub	x20,x20,#8		// i--
+
+	// (*)	mul	x12,x13,x15	// np[0]*m1
+	umulh	x13,x13,x15
+	mul	x16,x14,x15		// np[1]*m1
+	// (*)	adds	x12,x12,x6
+	subs	xzr,x6,#1		// (*)
+	umulh	x17,x14,x15
+	cbz	x21,.Linner_skip
+
+.Linner:
+	ldr	x8,[x1],#8
+	adc	x13,x13,xzr
+	ldr	x23,[x22],#8		// tp[j]
+	adds	x6,x10,x7
+	sub	x21,x21,#8		// j--
+	adc	x7,x11,xzr
+
+	adds	x12,x16,x13
+	ldr	x14,[x3],#8
+	adc	x13,x17,xzr
+
+	mul	x10,x8,x9		// ap[j]*bp[i]
+	adds	x6,x6,x23
+	umulh	x11,x8,x9
+	adc	x7,x7,xzr
+
+	mul	x16,x14,x15		// np[j]*m1
+	adds	x12,x12,x6
+	umulh	x17,x14,x15
+	str	x12,[x22,#-16]		// tp[j-1]
+	cbnz	x21,.Linner
+
+.Linner_skip:
+	ldr	x23,[x22],#8		// tp[j]
+	adc	x13,x13,xzr
+	adds	x6,x10,x7
+	sub	x1,x1,x5		// rewind x1
+	adc	x7,x11,xzr
+
+	adds	x12,x16,x13
+	sub	x3,x3,x5		// rewind x3
+	adcs	x13,x17,x19
+	adc	x19,xzr,xzr
+
+	adds	x6,x6,x23
+	adc	x7,x7,xzr
+
+	adds	x12,x12,x6
+	adcs	x13,x13,x7
+	adc	x19,x19,xzr		// upmost overflow bit
+	stp	x12,x13,[x22,#-16]
+
+	cbnz	x20,.Louter
+
+	// Final step. We see if result is larger than modulus, and
+	// if it is, subtract the modulus. But comparison implies
+	// subtraction. So we subtract modulus, see if it borrowed,
+	// and conditionally copy original value.
+	ldr	x23,[sp]		// tp[0]
+	add	x22,sp,#8
+	ldr	x14,[x3],#8		// np[0]
+	subs	x21,x5,#8		// j=num-1 and clear borrow
+	mov	x1,x0
+.Lsub:
+	sbcs	x8,x23,x14		// tp[j]-np[j]
+	ldr	x23,[x22],#8
+	sub	x21,x21,#8		// j--
+	ldr	x14,[x3],#8
+	str	x8,[x1],#8		// rp[j]=tp[j]-np[j]
+	cbnz	x21,.Lsub
+
+	sbcs	x8,x23,x14
+	sbcs	x19,x19,xzr		// did it borrow?
+	str	x8,[x1],#8		// rp[num-1]
+
+	ldr	x23,[sp]		// tp[0]
+	add	x22,sp,#8
+	ldr	x8,[x0],#8		// rp[0]
+	sub	x5,x5,#8		// num--
+	nop
+.Lcond_copy:
+	sub	x5,x5,#8		// num--
+	csel	x14,x23,x8,lo		// did it borrow?
+	ldr	x23,[x22],#8
+	ldr	x8,[x0],#8
+	str	xzr,[x22,#-16]		// wipe tp
+	str	x14,[x0,#-16]
+	cbnz	x5,.Lcond_copy
+
+	csel	x14,x23,x8,lo
+	str	xzr,[x22,#-8]		// wipe tp
+	str	x14,[x0,#-8]
+
+	ldp	x19,x20,[x29,#16]
+	mov	sp,x29
+	ldp	x21,x22,[x29,#32]
+	mov	x0,#1
+	ldp	x23,x24,[x29,#48]
+	ldr	x29,[sp],#64
+	ret
+.size	bn_mul_mont,.-bn_mul_mont
+.type	__bn_sqr8x_mont,%function
+.align	5
+__bn_sqr8x_mont:
+	cmp	x1,x2
+	b.ne	__bn_mul4x_mont
+.Lsqr8x_mont:
+	stp	x29,x30,[sp,#-128]!
+	add	x29,sp,#0
+	stp	x19,x20,[sp,#16]
+	stp	x21,x22,[sp,#32]
+	stp	x23,x24,[sp,#48]
+	stp	x25,x26,[sp,#64]
+	stp	x27,x28,[sp,#80]
+	stp	x0,x3,[sp,#96]	// offload rp and np
+
+	ldp	x6,x7,[x1,#8*0]
+	ldp	x8,x9,[x1,#8*2]
+	ldp	x10,x11,[x1,#8*4]
+	ldp	x12,x13,[x1,#8*6]
+
+	sub	x2,sp,x5,lsl#4
+	lsl	x5,x5,#3
+	ldr	x4,[x4]		// *n0
+	mov	sp,x2			// alloca
+	sub	x27,x5,#8*8
+	b	.Lsqr8x_zero_start
+
+.Lsqr8x_zero:
+	sub	x27,x27,#8*8
+	stp	xzr,xzr,[x2,#8*0]
+	stp	xzr,xzr,[x2,#8*2]
+	stp	xzr,xzr,[x2,#8*4]
+	stp	xzr,xzr,[x2,#8*6]
+.Lsqr8x_zero_start:
+	stp	xzr,xzr,[x2,#8*8]
+	stp	xzr,xzr,[x2,#8*10]
+	stp	xzr,xzr,[x2,#8*12]
+	stp	xzr,xzr,[x2,#8*14]
+	add	x2,x2,#8*16
+	cbnz	x27,.Lsqr8x_zero
+
+	add	x3,x1,x5
+	add	x1,x1,#8*8
+	mov	x19,xzr
+	mov	x20,xzr
+	mov	x21,xzr
+	mov	x22,xzr
+	mov	x23,xzr
+	mov	x24,xzr
+	mov	x25,xzr
+	mov	x26,xzr
+	mov	x2,sp
+	str	x4,[x29,#112]		// offload n0
+
+	// Multiply everything but a[i]*a[i]
+.align	4
+.Lsqr8x_outer_loop:
+        //                                                 a[1]a[0]	(i)
+        //                                             a[2]a[0]
+        //                                         a[3]a[0]
+        //                                     a[4]a[0]
+        //                                 a[5]a[0]
+        //                             a[6]a[0]
+        //                         a[7]a[0]
+        //                                         a[2]a[1]		(ii)
+        //                                     a[3]a[1]
+        //                                 a[4]a[1]
+        //                             a[5]a[1]
+        //                         a[6]a[1]
+        //                     a[7]a[1]
+        //                                 a[3]a[2]			(iii)
+        //                             a[4]a[2]
+        //                         a[5]a[2]
+        //                     a[6]a[2]
+        //                 a[7]a[2]
+        //                         a[4]a[3]				(iv)
+        //                     a[5]a[3]
+        //                 a[6]a[3]
+        //             a[7]a[3]
+        //                 a[5]a[4]					(v)
+        //             a[6]a[4]
+        //         a[7]a[4]
+        //         a[6]a[5]						(vi)
+        //     a[7]a[5]
+        // a[7]a[6]							(vii)
+
+	mul	x14,x7,x6		// lo(a[1..7]*a[0])		(i)
+	mul	x15,x8,x6
+	mul	x16,x9,x6
+	mul	x17,x10,x6
+	adds	x20,x20,x14		// t[1]+lo(a[1]*a[0])
+	mul	x14,x11,x6
+	adcs	x21,x21,x15
+	mul	x15,x12,x6
+	adcs	x22,x22,x16
+	mul	x16,x13,x6
+	adcs	x23,x23,x17
+	umulh	x17,x7,x6		// hi(a[1..7]*a[0])
+	adcs	x24,x24,x14
+	umulh	x14,x8,x6
+	adcs	x25,x25,x15
+	umulh	x15,x9,x6
+	adcs	x26,x26,x16
+	umulh	x16,x10,x6
+	stp	x19,x20,[x2],#8*2	// t[0..1]
+	adc	x19,xzr,xzr		// t[8]
+	adds	x21,x21,x17		// t[2]+lo(a[1]*a[0])
+	umulh	x17,x11,x6
+	adcs	x22,x22,x14
+	umulh	x14,x12,x6
+	adcs	x23,x23,x15
+	umulh	x15,x13,x6
+	adcs	x24,x24,x16
+	mul	x16,x8,x7		// lo(a[2..7]*a[1])		(ii)
+	adcs	x25,x25,x17
+	mul	x17,x9,x7
+	adcs	x26,x26,x14
+	mul	x14,x10,x7
+	adc	x19,x19,x15
+
+	mul	x15,x11,x7
+	adds	x22,x22,x16
+	mul	x16,x12,x7
+	adcs	x23,x23,x17
+	mul	x17,x13,x7
+	adcs	x24,x24,x14
+	umulh	x14,x8,x7		// hi(a[2..7]*a[1])
+	adcs	x25,x25,x15
+	umulh	x15,x9,x7
+	adcs	x26,x26,x16
+	umulh	x16,x10,x7
+	adcs	x19,x19,x17
+	umulh	x17,x11,x7
+	stp	x21,x22,[x2],#8*2	// t[2..3]
+	adc	x20,xzr,xzr		// t[9]
+	adds	x23,x23,x14
+	umulh	x14,x12,x7
+	adcs	x24,x24,x15
+	umulh	x15,x13,x7
+	adcs	x25,x25,x16
+	mul	x16,x9,x8		// lo(a[3..7]*a[2])		(iii)
+	adcs	x26,x26,x17
+	mul	x17,x10,x8
+	adcs	x19,x19,x14
+	mul	x14,x11,x8
+	adc	x20,x20,x15
+
+	mul	x15,x12,x8
+	adds	x24,x24,x16
+	mul	x16,x13,x8
+	adcs	x25,x25,x17
+	umulh	x17,x9,x8		// hi(a[3..7]*a[2])
+	adcs	x26,x26,x14
+	umulh	x14,x10,x8
+	adcs	x19,x19,x15
+	umulh	x15,x11,x8
+	adcs	x20,x20,x16
+	umulh	x16,x12,x8
+	stp	x23,x24,[x2],#8*2	// t[4..5]
+	adc	x21,xzr,xzr		// t[10]
+	adds	x25,x25,x17
+	umulh	x17,x13,x8
+	adcs	x26,x26,x14
+	mul	x14,x10,x9		// lo(a[4..7]*a[3])		(iv)
+	adcs	x19,x19,x15
+	mul	x15,x11,x9
+	adcs	x20,x20,x16
+	mul	x16,x12,x9
+	adc	x21,x21,x17
+
+	mul	x17,x13,x9
+	adds	x26,x26,x14
+	umulh	x14,x10,x9		// hi(a[4..7]*a[3])
+	adcs	x19,x19,x15
+	umulh	x15,x11,x9
+	adcs	x20,x20,x16
+	umulh	x16,x12,x9
+	adcs	x21,x21,x17
+	umulh	x17,x13,x9
+	stp	x25,x26,[x2],#8*2	// t[6..7]
+	adc	x22,xzr,xzr		// t[11]
+	adds	x19,x19,x14
+	mul	x14,x11,x10		// lo(a[5..7]*a[4])		(v)
+	adcs	x20,x20,x15
+	mul	x15,x12,x10
+	adcs	x21,x21,x16
+	mul	x16,x13,x10
+	adc	x22,x22,x17
+
+	umulh	x17,x11,x10		// hi(a[5..7]*a[4])
+	adds	x20,x20,x14
+	umulh	x14,x12,x10
+	adcs	x21,x21,x15
+	umulh	x15,x13,x10
+	adcs	x22,x22,x16
+	mul	x16,x12,x11		// lo(a[6..7]*a[5])		(vi)
+	adc	x23,xzr,xzr		// t[12]
+	adds	x21,x21,x17
+	mul	x17,x13,x11
+	adcs	x22,x22,x14
+	umulh	x14,x12,x11		// hi(a[6..7]*a[5])
+	adc	x23,x23,x15
+
+	umulh	x15,x13,x11
+	adds	x22,x22,x16
+	mul	x16,x13,x12		// lo(a[7]*a[6])		(vii)
+	adcs	x23,x23,x17
+	umulh	x17,x13,x12		// hi(a[7]*a[6])
+	adc	x24,xzr,xzr		// t[13]
+	adds	x23,x23,x14
+	sub	x27,x3,x1	// done yet?
+	adc	x24,x24,x15
+
+	adds	x24,x24,x16
+	sub	x14,x3,x5	// rewinded ap
+	adc	x25,xzr,xzr		// t[14]
+	add	x25,x25,x17
+
+	cbz	x27,.Lsqr8x_outer_break
+
+	mov	x4,x6
+	ldp	x6,x7,[x2,#8*0]
+	ldp	x8,x9,[x2,#8*2]
+	ldp	x10,x11,[x2,#8*4]
+	ldp	x12,x13,[x2,#8*6]
+	adds	x19,x19,x6
+	adcs	x20,x20,x7
+	ldp	x6,x7,[x1,#8*0]
+	adcs	x21,x21,x8
+	adcs	x22,x22,x9
+	ldp	x8,x9,[x1,#8*2]
+	adcs	x23,x23,x10
+	adcs	x24,x24,x11
+	ldp	x10,x11,[x1,#8*4]
+	adcs	x25,x25,x12
+	mov	x0,x1
+	adcs	x26,xzr,x13
+	ldp	x12,x13,[x1,#8*6]
+	add	x1,x1,#8*8
+	//adc	x28,xzr,xzr		// moved below
+	mov	x27,#-8*8
+
+	//                                                         a[8]a[0]
+	//                                                     a[9]a[0]
+	//                                                 a[a]a[0]
+	//                                             a[b]a[0]
+	//                                         a[c]a[0]
+	//                                     a[d]a[0]
+	//                                 a[e]a[0]
+	//                             a[f]a[0]
+	//                                                     a[8]a[1]
+	//                         a[f]a[1]........................
+	//                                                 a[8]a[2]
+	//                     a[f]a[2]........................
+	//                                             a[8]a[3]
+	//                 a[f]a[3]........................
+	//                                         a[8]a[4]
+	//             a[f]a[4]........................
+	//                                     a[8]a[5]
+	//         a[f]a[5]........................
+	//                                 a[8]a[6]
+	//     a[f]a[6]........................
+	//                             a[8]a[7]
+	// a[f]a[7]........................
+.Lsqr8x_mul:
+	mul	x14,x6,x4
+	adc	x28,xzr,xzr		// carry bit, modulo-scheduled
+	mul	x15,x7,x4
+	add	x27,x27,#8
+	mul	x16,x8,x4
+	mul	x17,x9,x4
+	adds	x19,x19,x14
+	mul	x14,x10,x4
+	adcs	x20,x20,x15
+	mul	x15,x11,x4
+	adcs	x21,x21,x16
+	mul	x16,x12,x4
+	adcs	x22,x22,x17
+	mul	x17,x13,x4
+	adcs	x23,x23,x14
+	umulh	x14,x6,x4
+	adcs	x24,x24,x15
+	umulh	x15,x7,x4
+	adcs	x25,x25,x16
+	umulh	x16,x8,x4
+	adcs	x26,x26,x17
+	umulh	x17,x9,x4
+	adc	x28,x28,xzr
+	str	x19,[x2],#8
+	adds	x19,x20,x14
+	umulh	x14,x10,x4
+	adcs	x20,x21,x15
+	umulh	x15,x11,x4
+	adcs	x21,x22,x16
+	umulh	x16,x12,x4
+	adcs	x22,x23,x17
+	umulh	x17,x13,x4
+	ldr	x4,[x0,x27]
+	adcs	x23,x24,x14
+	adcs	x24,x25,x15
+	adcs	x25,x26,x16
+	adcs	x26,x28,x17
+	//adc	x28,xzr,xzr		// moved above
+	cbnz	x27,.Lsqr8x_mul
+					// note that carry flag is guaranteed
+					// to be zero at this point
+	cmp	x1,x3		// done yet?
+	b.eq	.Lsqr8x_break
+
+	ldp	x6,x7,[x2,#8*0]
+	ldp	x8,x9,[x2,#8*2]
+	ldp	x10,x11,[x2,#8*4]
+	ldp	x12,x13,[x2,#8*6]
+	adds	x19,x19,x6
+	ldr	x4,[x0,#-8*8]
+	adcs	x20,x20,x7
+	ldp	x6,x7,[x1,#8*0]
+	adcs	x21,x21,x8
+	adcs	x22,x22,x9
+	ldp	x8,x9,[x1,#8*2]
+	adcs	x23,x23,x10
+	adcs	x24,x24,x11
+	ldp	x10,x11,[x1,#8*4]
+	adcs	x25,x25,x12
+	mov	x27,#-8*8
+	adcs	x26,x26,x13
+	ldp	x12,x13,[x1,#8*6]
+	add	x1,x1,#8*8
+	//adc	x28,xzr,xzr		// moved above
+	b	.Lsqr8x_mul
+
+.align	4
+.Lsqr8x_break:
+	ldp	x6,x7,[x0,#8*0]
+	add	x1,x0,#8*8
+	ldp	x8,x9,[x0,#8*2]
+	sub	x14,x3,x1		// is it last iteration?
+	ldp	x10,x11,[x0,#8*4]
+	sub	x15,x2,x14
+	ldp	x12,x13,[x0,#8*6]
+	cbz	x14,.Lsqr8x_outer_loop
+
+	stp	x19,x20,[x2,#8*0]
+	ldp	x19,x20,[x15,#8*0]
+	stp	x21,x22,[x2,#8*2]
+	ldp	x21,x22,[x15,#8*2]
+	stp	x23,x24,[x2,#8*4]
+	ldp	x23,x24,[x15,#8*4]
+	stp	x25,x26,[x2,#8*6]
+	mov	x2,x15
+	ldp	x25,x26,[x15,#8*6]
+	b	.Lsqr8x_outer_loop
+
+.align	4
+.Lsqr8x_outer_break:
+	// Now multiply above result by 2 and add a[n-1]*a[n-1]|...|a[0]*a[0]
+	ldp	x7,x9,[x14,#8*0]	// recall that x14 is &a[0]
+	ldp	x15,x16,[sp,#8*1]
+	ldp	x11,x13,[x14,#8*2]
+	add	x1,x14,#8*4
+	ldp	x17,x14,[sp,#8*3]
+
+	stp	x19,x20,[x2,#8*0]
+	mul	x19,x7,x7
+	stp	x21,x22,[x2,#8*2]
+	umulh	x7,x7,x7
+	stp	x23,x24,[x2,#8*4]
+	mul	x8,x9,x9
+	stp	x25,x26,[x2,#8*6]
+	mov	x2,sp
+	umulh	x9,x9,x9
+	adds	x20,x7,x15,lsl#1
+	extr	x15,x16,x15,#63
+	sub	x27,x5,#8*4
+
+.Lsqr4x_shift_n_add:
+	adcs	x21,x8,x15
+	extr	x16,x17,x16,#63
+	sub	x27,x27,#8*4
+	adcs	x22,x9,x16
+	ldp	x15,x16,[x2,#8*5]
+	mul	x10,x11,x11
+	ldp	x7,x9,[x1],#8*2
+	umulh	x11,x11,x11
+	mul	x12,x13,x13
+	umulh	x13,x13,x13
+	extr	x17,x14,x17,#63
+	stp	x19,x20,[x2,#8*0]
+	adcs	x23,x10,x17
+	extr	x14,x15,x14,#63
+	stp	x21,x22,[x2,#8*2]
+	adcs	x24,x11,x14
+	ldp	x17,x14,[x2,#8*7]
+	extr	x15,x16,x15,#63
+	adcs	x25,x12,x15
+	extr	x16,x17,x16,#63
+	adcs	x26,x13,x16
+	ldp	x15,x16,[x2,#8*9]
+	mul	x6,x7,x7
+	ldp	x11,x13,[x1],#8*2
+	umulh	x7,x7,x7
+	mul	x8,x9,x9
+	umulh	x9,x9,x9
+	stp	x23,x24,[x2,#8*4]
+	extr	x17,x14,x17,#63
+	stp	x25,x26,[x2,#8*6]
+	add	x2,x2,#8*8
+	adcs	x19,x6,x17
+	extr	x14,x15,x14,#63
+	adcs	x20,x7,x14
+	ldp	x17,x14,[x2,#8*3]
+	extr	x15,x16,x15,#63
+	cbnz	x27,.Lsqr4x_shift_n_add
+	ldp	x1,x4,[x29,#104]	// pull np and n0
+
+	adcs	x21,x8,x15
+	extr	x16,x17,x16,#63
+	adcs	x22,x9,x16
+	ldp	x15,x16,[x2,#8*5]
+	mul	x10,x11,x11
+	umulh	x11,x11,x11
+	stp	x19,x20,[x2,#8*0]
+	mul	x12,x13,x13
+	umulh	x13,x13,x13
+	stp	x21,x22,[x2,#8*2]
+	extr	x17,x14,x17,#63
+	adcs	x23,x10,x17
+	extr	x14,x15,x14,#63
+	ldp	x19,x20,[sp,#8*0]
+	adcs	x24,x11,x14
+	extr	x15,x16,x15,#63
+	ldp	x6,x7,[x1,#8*0]
+	adcs	x25,x12,x15
+	extr	x16,xzr,x16,#63
+	ldp	x8,x9,[x1,#8*2]
+	adc	x26,x13,x16
+	ldp	x10,x11,[x1,#8*4]
+
+	// Reduce by 512 bits per iteration
+	mul	x28,x4,x19		// t[0]*n0
+	ldp	x12,x13,[x1,#8*6]
+	add	x3,x1,x5
+	ldp	x21,x22,[sp,#8*2]
+	stp	x23,x24,[x2,#8*4]
+	ldp	x23,x24,[sp,#8*4]
+	stp	x25,x26,[x2,#8*6]
+	ldp	x25,x26,[sp,#8*6]
+	add	x1,x1,#8*8
+	mov	x30,xzr		// initial top-most carry
+	mov	x2,sp
+	mov	x27,#8
+
+.Lsqr8x_reduction:
+	// (*)	mul	x14,x6,x28	// lo(n[0-7])*lo(t[0]*n0)
+	mul	x15,x7,x28
+	sub	x27,x27,#1
+	mul	x16,x8,x28
+	str	x28,[x2],#8		// put aside t[0]*n0 for tail processing
+	mul	x17,x9,x28
+	// (*)	adds	xzr,x19,x14
+	subs	xzr,x19,#1		// (*)
+	mul	x14,x10,x28
+	adcs	x19,x20,x15
+	mul	x15,x11,x28
+	adcs	x20,x21,x16
+	mul	x16,x12,x28
+	adcs	x21,x22,x17
+	mul	x17,x13,x28
+	adcs	x22,x23,x14
+	umulh	x14,x6,x28		// hi(n[0-7])*lo(t[0]*n0)
+	adcs	x23,x24,x15
+	umulh	x15,x7,x28
+	adcs	x24,x25,x16
+	umulh	x16,x8,x28
+	adcs	x25,x26,x17
+	umulh	x17,x9,x28
+	adc	x26,xzr,xzr
+	adds	x19,x19,x14
+	umulh	x14,x10,x28
+	adcs	x20,x20,x15
+	umulh	x15,x11,x28
+	adcs	x21,x21,x16
+	umulh	x16,x12,x28
+	adcs	x22,x22,x17
+	umulh	x17,x13,x28
+	mul	x28,x4,x19		// next t[0]*n0
+	adcs	x23,x23,x14
+	adcs	x24,x24,x15
+	adcs	x25,x25,x16
+	adc	x26,x26,x17
+	cbnz	x27,.Lsqr8x_reduction
+
+	ldp	x14,x15,[x2,#8*0]
+	ldp	x16,x17,[x2,#8*2]
+	mov	x0,x2
+	sub	x27,x3,x1	// done yet?
+	adds	x19,x19,x14
+	adcs	x20,x20,x15
+	ldp	x14,x15,[x2,#8*4]
+	adcs	x21,x21,x16
+	adcs	x22,x22,x17
+	ldp	x16,x17,[x2,#8*6]
+	adcs	x23,x23,x14
+	adcs	x24,x24,x15
+	adcs	x25,x25,x16
+	adcs	x26,x26,x17
+	//adc	x28,xzr,xzr		// moved below
+	cbz	x27,.Lsqr8x8_post_condition
+
+	ldr	x4,[x2,#-8*8]
+	ldp	x6,x7,[x1,#8*0]
+	ldp	x8,x9,[x1,#8*2]
+	ldp	x10,x11,[x1,#8*4]
+	mov	x27,#-8*8
+	ldp	x12,x13,[x1,#8*6]
+	add	x1,x1,#8*8
+
+.Lsqr8x_tail:
+	mul	x14,x6,x4
+	adc	x28,xzr,xzr		// carry bit, modulo-scheduled
+	mul	x15,x7,x4
+	add	x27,x27,#8
+	mul	x16,x8,x4
+	mul	x17,x9,x4
+	adds	x19,x19,x14
+	mul	x14,x10,x4
+	adcs	x20,x20,x15
+	mul	x15,x11,x4
+	adcs	x21,x21,x16
+	mul	x16,x12,x4
+	adcs	x22,x22,x17
+	mul	x17,x13,x4
+	adcs	x23,x23,x14
+	umulh	x14,x6,x4
+	adcs	x24,x24,x15
+	umulh	x15,x7,x4
+	adcs	x25,x25,x16
+	umulh	x16,x8,x4
+	adcs	x26,x26,x17
+	umulh	x17,x9,x4
+	adc	x28,x28,xzr
+	str	x19,[x2],#8
+	adds	x19,x20,x14
+	umulh	x14,x10,x4
+	adcs	x20,x21,x15
+	umulh	x15,x11,x4
+	adcs	x21,x22,x16
+	umulh	x16,x12,x4
+	adcs	x22,x23,x17
+	umulh	x17,x13,x4
+	ldr	x4,[x0,x27]
+	adcs	x23,x24,x14
+	adcs	x24,x25,x15
+	adcs	x25,x26,x16
+	adcs	x26,x28,x17
+	//adc	x28,xzr,xzr		// moved above
+	cbnz	x27,.Lsqr8x_tail
+					// note that carry flag is guaranteed
+					// to be zero at this point
+	ldp	x6,x7,[x2,#8*0]
+	sub	x27,x3,x1	// done yet?
+	sub	x16,x3,x5	// rewinded np
+	ldp	x8,x9,[x2,#8*2]
+	ldp	x10,x11,[x2,#8*4]
+	ldp	x12,x13,[x2,#8*6]
+	cbz	x27,.Lsqr8x_tail_break
+
+	ldr	x4,[x0,#-8*8]
+	adds	x19,x19,x6
+	adcs	x20,x20,x7
+	ldp	x6,x7,[x1,#8*0]
+	adcs	x21,x21,x8
+	adcs	x22,x22,x9
+	ldp	x8,x9,[x1,#8*2]
+	adcs	x23,x23,x10
+	adcs	x24,x24,x11
+	ldp	x10,x11,[x1,#8*4]
+	adcs	x25,x25,x12
+	mov	x27,#-8*8
+	adcs	x26,x26,x13
+	ldp	x12,x13,[x1,#8*6]
+	add	x1,x1,#8*8
+	//adc	x28,xzr,xzr		// moved above
+	b	.Lsqr8x_tail
+
+.align	4
+.Lsqr8x_tail_break:
+	ldr	x4,[x29,#112]		// pull n0
+	add	x27,x2,#8*8		// end of current t[num] window
+
+	subs	xzr,x30,#1		// "move" top-most carry to carry bit
+	adcs	x14,x19,x6
+	adcs	x15,x20,x7
+	ldp	x19,x20,[x0,#8*0]
+	adcs	x21,x21,x8
+	ldp	x6,x7,[x16,#8*0]	// recall that x16 is &n[0]
+	adcs	x22,x22,x9
+	ldp	x8,x9,[x16,#8*2]
+	adcs	x23,x23,x10
+	adcs	x24,x24,x11
+	ldp	x10,x11,[x16,#8*4]
+	adcs	x25,x25,x12
+	adcs	x26,x26,x13
+	ldp	x12,x13,[x16,#8*6]
+	add	x1,x16,#8*8
+	adc	x30,xzr,xzr	// top-most carry
+	mul	x28,x4,x19
+	stp	x14,x15,[x2,#8*0]
+	stp	x21,x22,[x2,#8*2]
+	ldp	x21,x22,[x0,#8*2]
+	stp	x23,x24,[x2,#8*4]
+	ldp	x23,x24,[x0,#8*4]
+	cmp	x27,x29		// did we hit the bottom?
+	stp	x25,x26,[x2,#8*6]
+	mov	x2,x0			// slide the window
+	ldp	x25,x26,[x0,#8*6]
+	mov	x27,#8
+	b.ne	.Lsqr8x_reduction
+
+	// Final step. We see if result is larger than modulus, and
+	// if it is, subtract the modulus. But comparison implies
+	// subtraction. So we subtract modulus, see if it borrowed,
+	// and conditionally copy original value.
+	ldr	x0,[x29,#96]		// pull rp
+	add	x2,x2,#8*8
+	subs	x14,x19,x6
+	sbcs	x15,x20,x7
+	sub	x27,x5,#8*8
+	mov	x3,x0		// x0 copy
+
+.Lsqr8x_sub:
+	sbcs	x16,x21,x8
+	ldp	x6,x7,[x1,#8*0]
+	sbcs	x17,x22,x9
+	stp	x14,x15,[x0,#8*0]
+	sbcs	x14,x23,x10
+	ldp	x8,x9,[x1,#8*2]
+	sbcs	x15,x24,x11
+	stp	x16,x17,[x0,#8*2]
+	sbcs	x16,x25,x12
+	ldp	x10,x11,[x1,#8*4]
+	sbcs	x17,x26,x13
+	ldp	x12,x13,[x1,#8*6]
+	add	x1,x1,#8*8
+	ldp	x19,x20,[x2,#8*0]
+	sub	x27,x27,#8*8
+	ldp	x21,x22,[x2,#8*2]
+	ldp	x23,x24,[x2,#8*4]
+	ldp	x25,x26,[x2,#8*6]
+	add	x2,x2,#8*8
+	stp	x14,x15,[x0,#8*4]
+	sbcs	x14,x19,x6
+	stp	x16,x17,[x0,#8*6]
+	add	x0,x0,#8*8
+	sbcs	x15,x20,x7
+	cbnz	x27,.Lsqr8x_sub
+
+	sbcs	x16,x21,x8
+	mov	x2,sp
+	add	x1,sp,x5
+	ldp	x6,x7,[x3,#8*0]
+	sbcs	x17,x22,x9
+	stp	x14,x15,[x0,#8*0]
+	sbcs	x14,x23,x10
+	ldp	x8,x9,[x3,#8*2]
+	sbcs	x15,x24,x11
+	stp	x16,x17,[x0,#8*2]
+	sbcs	x16,x25,x12
+	ldp	x19,x20,[x1,#8*0]
+	sbcs	x17,x26,x13
+	ldp	x21,x22,[x1,#8*2]
+	sbcs	xzr,x30,xzr	// did it borrow?
+	ldr	x30,[x29,#8]		// pull return address
+	stp	x14,x15,[x0,#8*4]
+	stp	x16,x17,[x0,#8*6]
+
+	sub	x27,x5,#8*4
+.Lsqr4x_cond_copy:
+	sub	x27,x27,#8*4
+	csel	x14,x19,x6,lo
+	stp	xzr,xzr,[x2,#8*0]
+	csel	x15,x20,x7,lo
+	ldp	x6,x7,[x3,#8*4]
+	ldp	x19,x20,[x1,#8*4]
+	csel	x16,x21,x8,lo
+	stp	xzr,xzr,[x2,#8*2]
+	add	x2,x2,#8*4
+	csel	x17,x22,x9,lo
+	ldp	x8,x9,[x3,#8*6]
+	ldp	x21,x22,[x1,#8*6]
+	add	x1,x1,#8*4
+	stp	x14,x15,[x3,#8*0]
+	stp	x16,x17,[x3,#8*2]
+	add	x3,x3,#8*4
+	stp	xzr,xzr,[x1,#8*0]
+	stp	xzr,xzr,[x1,#8*2]
+	cbnz	x27,.Lsqr4x_cond_copy
+
+	csel	x14,x19,x6,lo
+	stp	xzr,xzr,[x2,#8*0]
+	csel	x15,x20,x7,lo
+	stp	xzr,xzr,[x2,#8*2]
+	csel	x16,x21,x8,lo
+	csel	x17,x22,x9,lo
+	stp	x14,x15,[x3,#8*0]
+	stp	x16,x17,[x3,#8*2]
+
+	b	.Lsqr8x_done
+
+.align	4
+.Lsqr8x8_post_condition:
+	adc	x28,xzr,xzr
+	ldr	x30,[x29,#8]		// pull return address
+	// x19-7,x28 hold result, x6-7 hold modulus
+	subs	x6,x19,x6
+	ldr	x1,[x29,#96]		// pull rp
+	sbcs	x7,x20,x7
+	stp	xzr,xzr,[sp,#8*0]
+	sbcs	x8,x21,x8
+	stp	xzr,xzr,[sp,#8*2]
+	sbcs	x9,x22,x9
+	stp	xzr,xzr,[sp,#8*4]
+	sbcs	x10,x23,x10
+	stp	xzr,xzr,[sp,#8*6]
+	sbcs	x11,x24,x11
+	stp	xzr,xzr,[sp,#8*8]
+	sbcs	x12,x25,x12
+	stp	xzr,xzr,[sp,#8*10]
+	sbcs	x13,x26,x13
+	stp	xzr,xzr,[sp,#8*12]
+	sbcs	x28,x28,xzr	// did it borrow?
+	stp	xzr,xzr,[sp,#8*14]
+
+	// x6-7 hold result-modulus
+	csel	x6,x19,x6,lo
+	csel	x7,x20,x7,lo
+	csel	x8,x21,x8,lo
+	csel	x9,x22,x9,lo
+	stp	x6,x7,[x1,#8*0]
+	csel	x10,x23,x10,lo
+	csel	x11,x24,x11,lo
+	stp	x8,x9,[x1,#8*2]
+	csel	x12,x25,x12,lo
+	csel	x13,x26,x13,lo
+	stp	x10,x11,[x1,#8*4]
+	stp	x12,x13,[x1,#8*6]
+
+.Lsqr8x_done:
+	ldp	x19,x20,[x29,#16]
+	mov	sp,x29
+	ldp	x21,x22,[x29,#32]
+	mov	x0,#1
+	ldp	x23,x24,[x29,#48]
+	ldp	x25,x26,[x29,#64]
+	ldp	x27,x28,[x29,#80]
+	ldr	x29,[sp],#128
+	ret
+.size	__bn_sqr8x_mont,.-__bn_sqr8x_mont
+.type	__bn_mul4x_mont,%function
+.align	5
+__bn_mul4x_mont:
+	stp	x29,x30,[sp,#-128]!
+	add	x29,sp,#0
+	stp	x19,x20,[sp,#16]
+	stp	x21,x22,[sp,#32]
+	stp	x23,x24,[sp,#48]
+	stp	x25,x26,[sp,#64]
+	stp	x27,x28,[sp,#80]
+
+	sub	x26,sp,x5,lsl#3
+	lsl	x5,x5,#3
+	ldr	x4,[x4]		// *n0
+	sub	sp,x26,#8*4		// alloca
+
+	add	x10,x2,x5
+	add	x27,x1,x5
+	stp	x0,x10,[x29,#96]	// offload rp and &b[num]
+
+	ldr	x24,[x2,#8*0]		// b[0]
+	ldp	x6,x7,[x1,#8*0]	// a[0..3]
+	ldp	x8,x9,[x1,#8*2]
+	add	x1,x1,#8*4
+	mov	x19,xzr
+	mov	x20,xzr
+	mov	x21,xzr
+	mov	x22,xzr
+	ldp	x14,x15,[x3,#8*0]	// n[0..3]
+	ldp	x16,x17,[x3,#8*2]
+	adds	x3,x3,#8*4		// clear carry bit
+	mov	x0,xzr
+	mov	x28,#0
+	mov	x26,sp
+
+.Loop_mul4x_1st_reduction:
+	mul	x10,x6,x24		// lo(a[0..3]*b[0])
+	adc	x0,x0,xzr	// modulo-scheduled
+	mul	x11,x7,x24
+	add	x28,x28,#8
+	mul	x12,x8,x24
+	and	x28,x28,#31
+	mul	x13,x9,x24
+	adds	x19,x19,x10
+	umulh	x10,x6,x24		// hi(a[0..3]*b[0])
+	adcs	x20,x20,x11
+	mul	x25,x19,x4		// t[0]*n0
+	adcs	x21,x21,x12
+	umulh	x11,x7,x24
+	adcs	x22,x22,x13
+	umulh	x12,x8,x24
+	adc	x23,xzr,xzr
+	umulh	x13,x9,x24
+	ldr	x24,[x2,x28]		// next b[i] (or b[0])
+	adds	x20,x20,x10
+	// (*)	mul	x10,x14,x25	// lo(n[0..3]*t[0]*n0)
+	str	x25,[x26],#8		// put aside t[0]*n0 for tail processing
+	adcs	x21,x21,x11
+	mul	x11,x15,x25
+	adcs	x22,x22,x12
+	mul	x12,x16,x25
+	adc	x23,x23,x13		// can't overflow
+	mul	x13,x17,x25
+	// (*)	adds	xzr,x19,x10
+	subs	xzr,x19,#1		// (*)
+	umulh	x10,x14,x25		// hi(n[0..3]*t[0]*n0)
+	adcs	x19,x20,x11
+	umulh	x11,x15,x25
+	adcs	x20,x21,x12
+	umulh	x12,x16,x25
+	adcs	x21,x22,x13
+	umulh	x13,x17,x25
+	adcs	x22,x23,x0
+	adc	x0,xzr,xzr
+	adds	x19,x19,x10
+	sub	x10,x27,x1
+	adcs	x20,x20,x11
+	adcs	x21,x21,x12
+	adcs	x22,x22,x13
+	//adc	x0,x0,xzr
+	cbnz	x28,.Loop_mul4x_1st_reduction
+
+	cbz	x10,.Lmul4x4_post_condition
+
+	ldp	x6,x7,[x1,#8*0]	// a[4..7]
+	ldp	x8,x9,[x1,#8*2]
+	add	x1,x1,#8*4
+	ldr	x25,[sp]		// a[0]*n0
+	ldp	x14,x15,[x3,#8*0]	// n[4..7]
+	ldp	x16,x17,[x3,#8*2]
+	add	x3,x3,#8*4
+
+.Loop_mul4x_1st_tail:
+	mul	x10,x6,x24		// lo(a[4..7]*b[i])
+	adc	x0,x0,xzr	// modulo-scheduled
+	mul	x11,x7,x24
+	add	x28,x28,#8
+	mul	x12,x8,x24
+	and	x28,x28,#31
+	mul	x13,x9,x24
+	adds	x19,x19,x10
+	umulh	x10,x6,x24		// hi(a[4..7]*b[i])
+	adcs	x20,x20,x11
+	umulh	x11,x7,x24
+	adcs	x21,x21,x12
+	umulh	x12,x8,x24
+	adcs	x22,x22,x13
+	umulh	x13,x9,x24
+	adc	x23,xzr,xzr
+	ldr	x24,[x2,x28]		// next b[i] (or b[0])
+	adds	x20,x20,x10
+	mul	x10,x14,x25		// lo(n[4..7]*a[0]*n0)
+	adcs	x21,x21,x11
+	mul	x11,x15,x25
+	adcs	x22,x22,x12
+	mul	x12,x16,x25
+	adc	x23,x23,x13		// can't overflow
+	mul	x13,x17,x25
+	adds	x19,x19,x10
+	umulh	x10,x14,x25		// hi(n[4..7]*a[0]*n0)
+	adcs	x20,x20,x11
+	umulh	x11,x15,x25
+	adcs	x21,x21,x12
+	umulh	x12,x16,x25
+	adcs	x22,x22,x13
+	adcs	x23,x23,x0
+	umulh	x13,x17,x25
+	adc	x0,xzr,xzr
+	ldr	x25,[sp,x28]		// next t[0]*n0
+	str	x19,[x26],#8		// result!!!
+	adds	x19,x20,x10
+	sub	x10,x27,x1		// done yet?
+	adcs	x20,x21,x11
+	adcs	x21,x22,x12
+	adcs	x22,x23,x13
+	//adc	x0,x0,xzr
+	cbnz	x28,.Loop_mul4x_1st_tail
+
+	sub	x11,x27,x5	// rewinded x1
+	cbz	x10,.Lmul4x_proceed
+
+	ldp	x6,x7,[x1,#8*0]
+	ldp	x8,x9,[x1,#8*2]
+	add	x1,x1,#8*4
+	ldp	x14,x15,[x3,#8*0]
+	ldp	x16,x17,[x3,#8*2]
+	add	x3,x3,#8*4
+	b	.Loop_mul4x_1st_tail
+
+.align	5
+.Lmul4x_proceed:
+	ldr	x24,[x2,#8*4]!		// *++b
+	adc	x30,x0,xzr
+	ldp	x6,x7,[x11,#8*0]	// a[0..3]
+	sub	x3,x3,x5		// rewind np
+	ldp	x8,x9,[x11,#8*2]
+	add	x1,x11,#8*4
+
+	stp	x19,x20,[x26,#8*0]	// result!!!
+	ldp	x19,x20,[sp,#8*4]	// t[0..3]
+	stp	x21,x22,[x26,#8*2]	// result!!!
+	ldp	x21,x22,[sp,#8*6]
+
+	ldp	x14,x15,[x3,#8*0]	// n[0..3]
+	mov	x26,sp
+	ldp	x16,x17,[x3,#8*2]
+	adds	x3,x3,#8*4		// clear carry bit
+	mov	x0,xzr
+
+.align	4
+.Loop_mul4x_reduction:
+	mul	x10,x6,x24		// lo(a[0..3]*b[4])
+	adc	x0,x0,xzr	// modulo-scheduled
+	mul	x11,x7,x24
+	add	x28,x28,#8
+	mul	x12,x8,x24
+	and	x28,x28,#31
+	mul	x13,x9,x24
+	adds	x19,x19,x10
+	umulh	x10,x6,x24		// hi(a[0..3]*b[4])
+	adcs	x20,x20,x11
+	mul	x25,x19,x4		// t[0]*n0
+	adcs	x21,x21,x12
+	umulh	x11,x7,x24
+	adcs	x22,x22,x13
+	umulh	x12,x8,x24
+	adc	x23,xzr,xzr
+	umulh	x13,x9,x24
+	ldr	x24,[x2,x28]		// next b[i]
+	adds	x20,x20,x10
+	// (*)	mul	x10,x14,x25
+	str	x25,[x26],#8		// put aside t[0]*n0 for tail processing
+	adcs	x21,x21,x11
+	mul	x11,x15,x25		// lo(n[0..3]*t[0]*n0
+	adcs	x22,x22,x12
+	mul	x12,x16,x25
+	adc	x23,x23,x13		// can't overflow
+	mul	x13,x17,x25
+	// (*)	adds	xzr,x19,x10
+	subs	xzr,x19,#1		// (*)
+	umulh	x10,x14,x25		// hi(n[0..3]*t[0]*n0
+	adcs	x19,x20,x11
+	umulh	x11,x15,x25
+	adcs	x20,x21,x12
+	umulh	x12,x16,x25
+	adcs	x21,x22,x13
+	umulh	x13,x17,x25
+	adcs	x22,x23,x0
+	adc	x0,xzr,xzr
+	adds	x19,x19,x10
+	adcs	x20,x20,x11
+	adcs	x21,x21,x12
+	adcs	x22,x22,x13
+	//adc	x0,x0,xzr
+	cbnz	x28,.Loop_mul4x_reduction
+
+	adc	x0,x0,xzr
+	ldp	x10,x11,[x26,#8*4]	// t[4..7]
+	ldp	x12,x13,[x26,#8*6]
+	ldp	x6,x7,[x1,#8*0]	// a[4..7]
+	ldp	x8,x9,[x1,#8*2]
+	add	x1,x1,#8*4
+	adds	x19,x19,x10
+	adcs	x20,x20,x11
+	adcs	x21,x21,x12
+	adcs	x22,x22,x13
+	//adc	x0,x0,xzr
+
+	ldr	x25,[sp]		// t[0]*n0
+	ldp	x14,x15,[x3,#8*0]	// n[4..7]
+	ldp	x16,x17,[x3,#8*2]
+	add	x3,x3,#8*4
+
+.align	4
+.Loop_mul4x_tail:
+	mul	x10,x6,x24		// lo(a[4..7]*b[4])
+	adc	x0,x0,xzr	// modulo-scheduled
+	mul	x11,x7,x24
+	add	x28,x28,#8
+	mul	x12,x8,x24
+	and	x28,x28,#31
+	mul	x13,x9,x24
+	adds	x19,x19,x10
+	umulh	x10,x6,x24		// hi(a[4..7]*b[4])
+	adcs	x20,x20,x11
+	umulh	x11,x7,x24
+	adcs	x21,x21,x12
+	umulh	x12,x8,x24
+	adcs	x22,x22,x13
+	umulh	x13,x9,x24
+	adc	x23,xzr,xzr
+	ldr	x24,[x2,x28]		// next b[i]
+	adds	x20,x20,x10
+	mul	x10,x14,x25		// lo(n[4..7]*t[0]*n0)
+	adcs	x21,x21,x11
+	mul	x11,x15,x25
+	adcs	x22,x22,x12
+	mul	x12,x16,x25
+	adc	x23,x23,x13		// can't overflow
+	mul	x13,x17,x25
+	adds	x19,x19,x10
+	umulh	x10,x14,x25		// hi(n[4..7]*t[0]*n0)
+	adcs	x20,x20,x11
+	umulh	x11,x15,x25
+	adcs	x21,x21,x12
+	umulh	x12,x16,x25
+	adcs	x22,x22,x13
+	umulh	x13,x17,x25
+	adcs	x23,x23,x0
+	ldr	x25,[sp,x28]		// next a[0]*n0
+	adc	x0,xzr,xzr
+	str	x19,[x26],#8		// result!!!
+	adds	x19,x20,x10
+	sub	x10,x27,x1		// done yet?
+	adcs	x20,x21,x11
+	adcs	x21,x22,x12
+	adcs	x22,x23,x13
+	//adc	x0,x0,xzr
+	cbnz	x28,.Loop_mul4x_tail
+
+	sub	x11,x3,x5		// rewinded np?
+	adc	x0,x0,xzr
+	cbz	x10,.Loop_mul4x_break
+
+	ldp	x10,x11,[x26,#8*4]
+	ldp	x12,x13,[x26,#8*6]
+	ldp	x6,x7,[x1,#8*0]
+	ldp	x8,x9,[x1,#8*2]
+	add	x1,x1,#8*4
+	adds	x19,x19,x10
+	adcs	x20,x20,x11
+	adcs	x21,x21,x12
+	adcs	x22,x22,x13
+	//adc	x0,x0,xzr
+	ldp	x14,x15,[x3,#8*0]
+	ldp	x16,x17,[x3,#8*2]
+	add	x3,x3,#8*4
+	b	.Loop_mul4x_tail
+
+.align	4
+.Loop_mul4x_break:
+	ldp	x12,x13,[x29,#96]	// pull rp and &b[num]
+	adds	x19,x19,x30
+	add	x2,x2,#8*4		// bp++
+	adcs	x20,x20,xzr
+	sub	x1,x1,x5		// rewind ap
+	adcs	x21,x21,xzr
+	stp	x19,x20,[x26,#8*0]	// result!!!
+	adcs	x22,x22,xzr
+	ldp	x19,x20,[sp,#8*4]	// t[0..3]
+	adc	x30,x0,xzr
+	stp	x21,x22,[x26,#8*2]	// result!!!
+	cmp	x2,x13			// done yet?
+	ldp	x21,x22,[sp,#8*6]
+	ldp	x14,x15,[x11,#8*0]	// n[0..3]
+	ldp	x16,x17,[x11,#8*2]
+	add	x3,x11,#8*4
+	b.eq	.Lmul4x_post
+
+	ldr	x24,[x2]
+	ldp	x6,x7,[x1,#8*0]	// a[0..3]
+	ldp	x8,x9,[x1,#8*2]
+	adds	x1,x1,#8*4		// clear carry bit
+	mov	x0,xzr
+	mov	x26,sp
+	b	.Loop_mul4x_reduction
+
+.align	4
+.Lmul4x_post:
+	// Final step. We see if result is larger than modulus, and
+	// if it is, subtract the modulus. But comparison implies
+	// subtraction. So we subtract modulus, see if it borrowed,
+	// and conditionally copy original value.
+	mov	x0,x12
+	mov	x27,x12		// x0 copy
+	subs	x10,x19,x14
+	add	x26,sp,#8*8
+	sbcs	x11,x20,x15
+	sub	x28,x5,#8*4
+
+.Lmul4x_sub:
+	sbcs	x12,x21,x16
+	ldp	x14,x15,[x3,#8*0]
+	sub	x28,x28,#8*4
+	ldp	x19,x20,[x26,#8*0]
+	sbcs	x13,x22,x17
+	ldp	x16,x17,[x3,#8*2]
+	add	x3,x3,#8*4
+	ldp	x21,x22,[x26,#8*2]
+	add	x26,x26,#8*4
+	stp	x10,x11,[x0,#8*0]
+	sbcs	x10,x19,x14
+	stp	x12,x13,[x0,#8*2]
+	add	x0,x0,#8*4
+	sbcs	x11,x20,x15
+	cbnz	x28,.Lmul4x_sub
+
+	sbcs	x12,x21,x16
+	mov	x26,sp
+	add	x1,sp,#8*4
+	ldp	x6,x7,[x27,#8*0]
+	sbcs	x13,x22,x17
+	stp	x10,x11,[x0,#8*0]
+	ldp	x8,x9,[x27,#8*2]
+	stp	x12,x13,[x0,#8*2]
+	ldp	x19,x20,[x1,#8*0]
+	ldp	x21,x22,[x1,#8*2]
+	sbcs	xzr,x30,xzr	// did it borrow?
+	ldr	x30,[x29,#8]		// pull return address
+
+	sub	x28,x5,#8*4
+.Lmul4x_cond_copy:
+	sub	x28,x28,#8*4
+	csel	x10,x19,x6,lo
+	stp	xzr,xzr,[x26,#8*0]
+	csel	x11,x20,x7,lo
+	ldp	x6,x7,[x27,#8*4]
+	ldp	x19,x20,[x1,#8*4]
+	csel	x12,x21,x8,lo
+	stp	xzr,xzr,[x26,#8*2]
+	add	x26,x26,#8*4
+	csel	x13,x22,x9,lo
+	ldp	x8,x9,[x27,#8*6]
+	ldp	x21,x22,[x1,#8*6]
+	add	x1,x1,#8*4
+	stp	x10,x11,[x27,#8*0]
+	stp	x12,x13,[x27,#8*2]
+	add	x27,x27,#8*4
+	cbnz	x28,.Lmul4x_cond_copy
+
+	csel	x10,x19,x6,lo
+	stp	xzr,xzr,[x26,#8*0]
+	csel	x11,x20,x7,lo
+	stp	xzr,xzr,[x26,#8*2]
+	csel	x12,x21,x8,lo
+	stp	xzr,xzr,[x26,#8*3]
+	csel	x13,x22,x9,lo
+	stp	xzr,xzr,[x26,#8*4]
+	stp	x10,x11,[x27,#8*0]
+	stp	x12,x13,[x27,#8*2]
+
+	b	.Lmul4x_done
+
+.align	4
+.Lmul4x4_post_condition:
+	adc	x0,x0,xzr
+	ldr	x1,[x29,#96]		// pull rp
+	// x19-3,x0 hold result, x14-7 hold modulus
+	subs	x6,x19,x14
+	ldr	x30,[x29,#8]		// pull return address
+	sbcs	x7,x20,x15
+	stp	xzr,xzr,[sp,#8*0]
+	sbcs	x8,x21,x16
+	stp	xzr,xzr,[sp,#8*2]
+	sbcs	x9,x22,x17
+	stp	xzr,xzr,[sp,#8*4]
+	sbcs	xzr,x0,xzr		// did it borrow?
+	stp	xzr,xzr,[sp,#8*6]
+
+	// x6-3 hold result-modulus
+	csel	x6,x19,x6,lo
+	csel	x7,x20,x7,lo
+	csel	x8,x21,x8,lo
+	csel	x9,x22,x9,lo
+	stp	x6,x7,[x1,#8*0]
+	stp	x8,x9,[x1,#8*2]
+
+.Lmul4x_done:
+	ldp	x19,x20,[x29,#16]
+	mov	sp,x29
+	ldp	x21,x22,[x29,#32]
+	mov	x0,#1
+	ldp	x23,x24,[x29,#48]
+	ldp	x25,x26,[x29,#64]
+	ldp	x27,x28,[x29,#80]
+	ldr	x29,[sp],#128
+	ret
+.size	__bn_mul4x_mont,.-__bn_mul4x_mont
+.byte	77,111,110,116,103,111,109,101,114,121,32,77,117,108,116,105,112,108,105,99,97,116,105,111,110,32,102,111,114,32,65,82,77,118,56,44,32,67,82,89,80,84,79,71,65,77,83,32,98,121,32,60,97,112,112,114,111,64,111,112,101,110,115,115,108,46,111,114,103,62,0
+.align	2
+.align	4
+#endif

third_party/boringssl/linux-aarch64/crypto/chacha/chacha-armv8.S

diff --git a/third_party/boringssl/linux-aarch64/crypto/chacha/chacha-armv8.S b/third_party/boringssl/linux-aarch64/crypto/chacha/chacha-armv8.S
new file mode 100644
index 0000000..6ff6bff
--- /dev/null
+++ b/third_party/boringssl/linux-aarch64/crypto/chacha/chacha-armv8.S
@@ -0,0 +1,1971 @@
+#if defined(__aarch64__)
+#include <openssl/arm_arch.h>
+
+.text
+
+
+
+.align	5
+.Lsigma:
+.quad	0x3320646e61707865,0x6b20657479622d32		// endian-neutral
+.Lone:
+.long	1,0,0,0
+.LOPENSSL_armcap_P:
+#ifdef	__ILP32__
+.long	OPENSSL_armcap_P-.
+#else
+.quad	OPENSSL_armcap_P-.
+#endif
+.byte	67,104,97,67,104,97,50,48,32,102,111,114,32,65,82,77,118,56,44,32,67,82,89,80,84,79,71,65,77,83,32,98,121,32,60,97,112,112,114,111,64,111,112,101,110,115,115,108,46,111,114,103,62,0
+.align	2
+
+.globl	ChaCha20_ctr32
+.hidden	ChaCha20_ctr32
+.type	ChaCha20_ctr32,%function
+.align	5
+ChaCha20_ctr32:
+	cbz	x2,.Labort
+	adr	x5,.LOPENSSL_armcap_P
+	cmp	x2,#192
+	b.lo	.Lshort
+#ifdef	__ILP32__
+	ldrsw	x6,[x5]
+#else
+	ldr	x6,[x5]
+#endif
+	ldr	w17,[x6,x5]
+	tst	w17,#ARMV7_NEON
+	b.ne	ChaCha20_neon
+
+.Lshort:
+	stp	x29,x30,[sp,#-96]!
+	add	x29,sp,#0
+
+	adr	x5,.Lsigma
+	stp	x19,x20,[sp,#16]
+	stp	x21,x22,[sp,#32]
+	stp	x23,x24,[sp,#48]
+	stp	x25,x26,[sp,#64]
+	stp	x27,x28,[sp,#80]
+	sub	sp,sp,#64
+
+	ldp	x22,x23,[x5]		// load sigma
+	ldp	x24,x25,[x3]		// load key
+	ldp	x26,x27,[x3,#16]
+	ldp	x28,x30,[x4]		// load counter
+#ifdef	__ARMEB__
+	ror	x24,x24,#32
+	ror	x25,x25,#32
+	ror	x26,x26,#32
+	ror	x27,x27,#32
+	ror	x28,x28,#32
+	ror	x30,x30,#32
+#endif
+
+.Loop_outer:
+	mov	w5,w22			// unpack key block
+	lsr	x6,x22,#32
+	mov	w7,w23
+	lsr	x8,x23,#32
+	mov	w9,w24
+	lsr	x10,x24,#32
+	mov	w11,w25
+	lsr	x12,x25,#32
+	mov	w13,w26
+	lsr	x14,x26,#32
+	mov	w15,w27
+	lsr	x16,x27,#32
+	mov	w17,w28
+	lsr	x19,x28,#32
+	mov	w20,w30
+	lsr	x21,x30,#32
+
+	mov	x4,#10
+	subs	x2,x2,#64
+.Loop:
+	sub	x4,x4,#1
+	add	w5,w5,w9
+	add	w6,w6,w10
+	add	w7,w7,w11
+	add	w8,w8,w12
+	eor	w17,w17,w5
+	eor	w19,w19,w6
+	eor	w20,w20,w7
+	eor	w21,w21,w8
+	ror	w17,w17,#16
+	ror	w19,w19,#16
+	ror	w20,w20,#16
+	ror	w21,w21,#16
+	add	w13,w13,w17
+	add	w14,w14,w19
+	add	w15,w15,w20
+	add	w16,w16,w21
+	eor	w9,w9,w13
+	eor	w10,w10,w14
+	eor	w11,w11,w15
+	eor	w12,w12,w16
+	ror	w9,w9,#20
+	ror	w10,w10,#20
+	ror	w11,w11,#20
+	ror	w12,w12,#20
+	add	w5,w5,w9
+	add	w6,w6,w10
+	add	w7,w7,w11
+	add	w8,w8,w12
+	eor	w17,w17,w5
+	eor	w19,w19,w6
+	eor	w20,w20,w7
+	eor	w21,w21,w8
+	ror	w17,w17,#24
+	ror	w19,w19,#24
+	ror	w20,w20,#24
+	ror	w21,w21,#24
+	add	w13,w13,w17
+	add	w14,w14,w19
+	add	w15,w15,w20
+	add	w16,w16,w21
+	eor	w9,w9,w13
+	eor	w10,w10,w14
+	eor	w11,w11,w15
+	eor	w12,w12,w16
+	ror	w9,w9,#25
+	ror	w10,w10,#25
+	ror	w11,w11,#25
+	ror	w12,w12,#25
+	add	w5,w5,w10
+	add	w6,w6,w11
+	add	w7,w7,w12
+	add	w8,w8,w9
+	eor	w21,w21,w5
+	eor	w17,w17,w6
+	eor	w19,w19,w7
+	eor	w20,w20,w8
+	ror	w21,w21,#16
+	ror	w17,w17,#16
+	ror	w19,w19,#16
+	ror	w20,w20,#16
+	add	w15,w15,w21
+	add	w16,w16,w17
+	add	w13,w13,w19
+	add	w14,w14,w20
+	eor	w10,w10,w15
+	eor	w11,w11,w16
+	eor	w12,w12,w13
+	eor	w9,w9,w14
+	ror	w10,w10,#20
+	ror	w11,w11,#20
+	ror	w12,w12,#20
+	ror	w9,w9,#20
+	add	w5,w5,w10
+	add	w6,w6,w11
+	add	w7,w7,w12
+	add	w8,w8,w9
+	eor	w21,w21,w5
+	eor	w17,w17,w6
+	eor	w19,w19,w7
+	eor	w20,w20,w8
+	ror	w21,w21,#24
+	ror	w17,w17,#24
+	ror	w19,w19,#24
+	ror	w20,w20,#24
+	add	w15,w15,w21
+	add	w16,w16,w17
+	add	w13,w13,w19
+	add	w14,w14,w20
+	eor	w10,w10,w15
+	eor	w11,w11,w16
+	eor	w12,w12,w13
+	eor	w9,w9,w14
+	ror	w10,w10,#25
+	ror	w11,w11,#25
+	ror	w12,w12,#25
+	ror	w9,w9,#25
+	cbnz	x4,.Loop
+
+	add	w5,w5,w22		// accumulate key block
+	add	x6,x6,x22,lsr#32
+	add	w7,w7,w23
+	add	x8,x8,x23,lsr#32
+	add	w9,w9,w24
+	add	x10,x10,x24,lsr#32
+	add	w11,w11,w25
+	add	x12,x12,x25,lsr#32
+	add	w13,w13,w26
+	add	x14,x14,x26,lsr#32
+	add	w15,w15,w27
+	add	x16,x16,x27,lsr#32
+	add	w17,w17,w28
+	add	x19,x19,x28,lsr#32
+	add	w20,w20,w30
+	add	x21,x21,x30,lsr#32
+
+	b.lo	.Ltail
+
+	add	x5,x5,x6,lsl#32	// pack
+	add	x7,x7,x8,lsl#32
+	ldp	x6,x8,[x1,#0]		// load input
+	add	x9,x9,x10,lsl#32
+	add	x11,x11,x12,lsl#32
+	ldp	x10,x12,[x1,#16]
+	add	x13,x13,x14,lsl#32
+	add	x15,x15,x16,lsl#32
+	ldp	x14,x16,[x1,#32]
+	add	x17,x17,x19,lsl#32
+	add	x20,x20,x21,lsl#32
+	ldp	x19,x21,[x1,#48]
+	add	x1,x1,#64
+#ifdef	__ARMEB__
+	rev	x5,x5
+	rev	x7,x7
+	rev	x9,x9
+	rev	x11,x11
+	rev	x13,x13
+	rev	x15,x15
+	rev	x17,x17
+	rev	x20,x20
+#endif
+	eor	x5,x5,x6
+	eor	x7,x7,x8
+	eor	x9,x9,x10
+	eor	x11,x11,x12
+	eor	x13,x13,x14
+	eor	x15,x15,x16
+	eor	x17,x17,x19
+	eor	x20,x20,x21
+
+	stp	x5,x7,[x0,#0]		// store output
+	add	x28,x28,#1			// increment counter
+	stp	x9,x11,[x0,#16]
+	stp	x13,x15,[x0,#32]
+	stp	x17,x20,[x0,#48]
+	add	x0,x0,#64
+
+	b.hi	.Loop_outer
+
+	ldp	x19,x20,[x29,#16]
+	add	sp,sp,#64
+	ldp	x21,x22,[x29,#32]
+	ldp	x23,x24,[x29,#48]
+	ldp	x25,x26,[x29,#64]
+	ldp	x27,x28,[x29,#80]
+	ldp	x29,x30,[sp],#96
+.Labort:
+	ret
+
+.align	4
+.Ltail:
+	add	x2,x2,#64
+.Less_than_64:
+	sub	x0,x0,#1
+	add	x1,x1,x2
+	add	x0,x0,x2
+	add	x4,sp,x2
+	neg	x2,x2
+
+	add	x5,x5,x6,lsl#32	// pack
+	add	x7,x7,x8,lsl#32
+	add	x9,x9,x10,lsl#32
+	add	x11,x11,x12,lsl#32
+	add	x13,x13,x14,lsl#32
+	add	x15,x15,x16,lsl#32
+	add	x17,x17,x19,lsl#32
+	add	x20,x20,x21,lsl#32
+#ifdef	__ARMEB__
+	rev	x5,x5
+	rev	x7,x7
+	rev	x9,x9
+	rev	x11,x11
+	rev	x13,x13
+	rev	x15,x15
+	rev	x17,x17
+	rev	x20,x20
+#endif
+	stp	x5,x7,[sp,#0]
+	stp	x9,x11,[sp,#16]
+	stp	x13,x15,[sp,#32]
+	stp	x17,x20,[sp,#48]
+
+.Loop_tail:
+	ldrb	w10,[x1,x2]
+	ldrb	w11,[x4,x2]
+	add	x2,x2,#1
+	eor	w10,w10,w11
+	strb	w10,[x0,x2]
+	cbnz	x2,.Loop_tail
+
+	stp	xzr,xzr,[sp,#0]
+	stp	xzr,xzr,[sp,#16]
+	stp	xzr,xzr,[sp,#32]
+	stp	xzr,xzr,[sp,#48]
+
+	ldp	x19,x20,[x29,#16]
+	add	sp,sp,#64
+	ldp	x21,x22,[x29,#32]
+	ldp	x23,x24,[x29,#48]
+	ldp	x25,x26,[x29,#64]
+	ldp	x27,x28,[x29,#80]
+	ldp	x29,x30,[sp],#96
+	ret
+.size	ChaCha20_ctr32,.-ChaCha20_ctr32
+
+.type	ChaCha20_neon,%function
+.align	5
+ChaCha20_neon:
+	stp	x29,x30,[sp,#-96]!
+	add	x29,sp,#0
+
+	adr	x5,.Lsigma
+	stp	x19,x20,[sp,#16]
+	stp	x21,x22,[sp,#32]
+	stp	x23,x24,[sp,#48]
+	stp	x25,x26,[sp,#64]
+	stp	x27,x28,[sp,#80]
+	cmp	x2,#512
+	b.hs	.L512_or_more_neon
+
+	sub	sp,sp,#64
+
+	ldp	x22,x23,[x5]		// load sigma
+	ld1	{v24.4s},[x5],#16
+	ldp	x24,x25,[x3]		// load key
+	ldp	x26,x27,[x3,#16]
+	ld1	{v25.4s,v26.4s},[x3]
+	ldp	x28,x30,[x4]		// load counter
+	ld1	{v27.4s},[x4]
+	ld1	{v31.4s},[x5]
+#ifdef	__ARMEB__
+	rev64	v24.4s,v24.4s
+	ror	x24,x24,#32
+	ror	x25,x25,#32
+	ror	x26,x26,#32
+	ror	x27,x27,#32
+	ror	x28,x28,#32
+	ror	x30,x30,#32
+#endif
+	add	v27.4s,v27.4s,v31.4s		// += 1
+	add	v28.4s,v27.4s,v31.4s
+	add	v29.4s,v28.4s,v31.4s
+	shl	v31.4s,v31.4s,#2			// 1 -> 4
+
+.Loop_outer_neon:
+	mov	w5,w22			// unpack key block
+	lsr	x6,x22,#32
+	mov	v0.16b,v24.16b
+	mov	w7,w23
+	lsr	x8,x23,#32
+	mov	v4.16b,v24.16b
+	mov	w9,w24
+	lsr	x10,x24,#32
+	mov	v16.16b,v24.16b
+	mov	w11,w25
+	mov	v1.16b,v25.16b
+	lsr	x12,x25,#32
+	mov	v5.16b,v25.16b
+	mov	w13,w26
+	mov	v17.16b,v25.16b
+	lsr	x14,x26,#32
+	mov	v3.16b,v27.16b
+	mov	w15,w27
+	mov	v7.16b,v28.16b
+	lsr	x16,x27,#32
+	mov	v19.16b,v29.16b
+	mov	w17,w28
+	mov	v2.16b,v26.16b
+	lsr	x19,x28,#32
+	mov	v6.16b,v26.16b
+	mov	w20,w30
+	mov	v18.16b,v26.16b
+	lsr	x21,x30,#32
+
+	mov	x4,#10
+	subs	x2,x2,#256
+.Loop_neon:
+	sub	x4,x4,#1
+	add	v0.4s,v0.4s,v1.4s
+	add	w5,w5,w9
+	add	v4.4s,v4.4s,v5.4s
+	add	w6,w6,w10
+	add	v16.4s,v16.4s,v17.4s
+	add	w7,w7,w11
+	eor	v3.16b,v3.16b,v0.16b
+	add	w8,w8,w12
+	eor	v7.16b,v7.16b,v4.16b
+	eor	w17,w17,w5
+	eor	v19.16b,v19.16b,v16.16b
+	eor	w19,w19,w6
+	rev32	v3.8h,v3.8h
+	eor	w20,w20,w7
+	rev32	v7.8h,v7.8h
+	eor	w21,w21,w8
+	rev32	v19.8h,v19.8h
+	ror	w17,w17,#16
+	add	v2.4s,v2.4s,v3.4s
+	ror	w19,w19,#16
+	add	v6.4s,v6.4s,v7.4s
+	ror	w20,w20,#16
+	add	v18.4s,v18.4s,v19.4s
+	ror	w21,w21,#16
+	eor	v20.16b,v1.16b,v2.16b
+	add	w13,w13,w17
+	eor	v21.16b,v5.16b,v6.16b
+	add	w14,w14,w19
+	eor	v22.16b,v17.16b,v18.16b
+	add	w15,w15,w20
+	ushr	v1.4s,v20.4s,#20
+	add	w16,w16,w21
+	ushr	v5.4s,v21.4s,#20
+	eor	w9,w9,w13
+	ushr	v17.4s,v22.4s,#20
+	eor	w10,w10,w14
+	sli	v1.4s,v20.4s,#12
+	eor	w11,w11,w15
+	sli	v5.4s,v21.4s,#12
+	eor	w12,w12,w16
+	sli	v17.4s,v22.4s,#12
+	ror	w9,w9,#20
+	add	v0.4s,v0.4s,v1.4s
+	ror	w10,w10,#20
+	add	v4.4s,v4.4s,v5.4s
+	ror	w11,w11,#20
+	add	v16.4s,v16.4s,v17.4s
+	ror	w12,w12,#20
+	eor	v20.16b,v3.16b,v0.16b
+	add	w5,w5,w9
+	eor	v21.16b,v7.16b,v4.16b
+	add	w6,w6,w10
+	eor	v22.16b,v19.16b,v16.16b
+	add	w7,w7,w11
+	ushr	v3.4s,v20.4s,#24
+	add	w8,w8,w12
+	ushr	v7.4s,v21.4s,#24
+	eor	w17,w17,w5
+	ushr	v19.4s,v22.4s,#24
+	eor	w19,w19,w6
+	sli	v3.4s,v20.4s,#8
+	eor	w20,w20,w7
+	sli	v7.4s,v21.4s,#8
+	eor	w21,w21,w8
+	sli	v19.4s,v22.4s,#8
+	ror	w17,w17,#24
+	add	v2.4s,v2.4s,v3.4s
+	ror	w19,w19,#24
+	add	v6.4s,v6.4s,v7.4s
+	ror	w20,w20,#24
+	add	v18.4s,v18.4s,v19.4s
+	ror	w21,w21,#24
+	eor	v20.16b,v1.16b,v2.16b
+	add	w13,w13,w17
+	eor	v21.16b,v5.16b,v6.16b
+	add	w14,w14,w19
+	eor	v22.16b,v17.16b,v18.16b
+	add	w15,w15,w20
+	ushr	v1.4s,v20.4s,#25
+	add	w16,w16,w21
+	ushr	v5.4s,v21.4s,#25
+	eor	w9,w9,w13
+	ushr	v17.4s,v22.4s,#25
+	eor	w10,w10,w14
+	sli	v1.4s,v20.4s,#7
+	eor	w11,w11,w15
+	sli	v5.4s,v21.4s,#7
+	eor	w12,w12,w16
+	sli	v17.4s,v22.4s,#7
+	ror	w9,w9,#25
+	ext	v2.16b,v2.16b,v2.16b,#8
+	ror	w10,w10,#25
+	ext	v6.16b,v6.16b,v6.16b,#8
+	ror	w11,w11,#25
+	ext	v18.16b,v18.16b,v18.16b,#8
+	ror	w12,w12,#25
+	ext	v3.16b,v3.16b,v3.16b,#12
+	ext	v7.16b,v7.16b,v7.16b,#12
+	ext	v19.16b,v19.16b,v19.16b,#12
+	ext	v1.16b,v1.16b,v1.16b,#4
+	ext	v5.16b,v5.16b,v5.16b,#4
+	ext	v17.16b,v17.16b,v17.16b,#4
+	add	v0.4s,v0.4s,v1.4s
+	add	w5,w5,w10
+	add	v4.4s,v4.4s,v5.4s
+	add	w6,w6,w11
+	add	v16.4s,v16.4s,v17.4s
+	add	w7,w7,w12
+	eor	v3.16b,v3.16b,v0.16b
+	add	w8,w8,w9
+	eor	v7.16b,v7.16b,v4.16b
+	eor	w21,w21,w5
+	eor	v19.16b,v19.16b,v16.16b
+	eor	w17,w17,w6
+	rev32	v3.8h,v3.8h
+	eor	w19,w19,w7
+	rev32	v7.8h,v7.8h
+	eor	w20,w20,w8
+	rev32	v19.8h,v19.8h
+	ror	w21,w21,#16
+	add	v2.4s,v2.4s,v3.4s
+	ror	w17,w17,#16
+	add	v6.4s,v6.4s,v7.4s
+	ror	w19,w19,#16
+	add	v18.4s,v18.4s,v19.4s
+	ror	w20,w20,#16
+	eor	v20.16b,v1.16b,v2.16b
+	add	w15,w15,w21
+	eor	v21.16b,v5.16b,v6.16b
+	add	w16,w16,w17
+	eor	v22.16b,v17.16b,v18.16b
+	add	w13,w13,w19
+	ushr	v1.4s,v20.4s,#20
+	add	w14,w14,w20
+	ushr	v5.4s,v21.4s,#20
+	eor	w10,w10,w15
+	ushr	v17.4s,v22.4s,#20
+	eor	w11,w11,w16
+	sli	v1.4s,v20.4s,#12
+	eor	w12,w12,w13
+	sli	v5.4s,v21.4s,#12
+	eor	w9,w9,w14
+	sli	v17.4s,v22.4s,#12
+	ror	w10,w10,#20
+	add	v0.4s,v0.4s,v1.4s
+	ror	w11,w11,#20
+	add	v4.4s,v4.4s,v5.4s
+	ror	w12,w12,#20
+	add	v16.4s,v16.4s,v17.4s
+	ror	w9,w9,#20
+	eor	v20.16b,v3.16b,v0.16b
+	add	w5,w5,w10
+	eor	v21.16b,v7.16b,v4.16b
+	add	w6,w6,w11
+	eor	v22.16b,v19.16b,v16.16b
+	add	w7,w7,w12
+	ushr	v3.4s,v20.4s,#24
+	add	w8,w8,w9
+	ushr	v7.4s,v21.4s,#24
+	eor	w21,w21,w5
+	ushr	v19.4s,v22.4s,#24
+	eor	w17,w17,w6
+	sli	v3.4s,v20.4s,#8
+	eor	w19,w19,w7
+	sli	v7.4s,v21.4s,#8
+	eor	w20,w20,w8
+	sli	v19.4s,v22.4s,#8
+	ror	w21,w21,#24
+	add	v2.4s,v2.4s,v3.4s
+	ror	w17,w17,#24
+	add	v6.4s,v6.4s,v7.4s
+	ror	w19,w19,#24
+	add	v18.4s,v18.4s,v19.4s
+	ror	w20,w20,#24
+	eor	v20.16b,v1.16b,v2.16b
+	add	w15,w15,w21
+	eor	v21.16b,v5.16b,v6.16b
+	add	w16,w16,w17
+	eor	v22.16b,v17.16b,v18.16b
+	add	w13,w13,w19
+	ushr	v1.4s,v20.4s,#25
+	add	w14,w14,w20
+	ushr	v5.4s,v21.4s,#25
+	eor	w10,w10,w15
+	ushr	v17.4s,v22.4s,#25
+	eor	w11,w11,w16
+	sli	v1.4s,v20.4s,#7
+	eor	w12,w12,w13
+	sli	v5.4s,v21.4s,#7
+	eor	w9,w9,w14
+	sli	v17.4s,v22.4s,#7
+	ror	w10,w10,#25
+	ext	v2.16b,v2.16b,v2.16b,#8
+	ror	w11,w11,#25
+	ext	v6.16b,v6.16b,v6.16b,#8
+	ror	w12,w12,#25
+	ext	v18.16b,v18.16b,v18.16b,#8
+	ror	w9,w9,#25
+	ext	v3.16b,v3.16b,v3.16b,#4
+	ext	v7.16b,v7.16b,v7.16b,#4
+	ext	v19.16b,v19.16b,v19.16b,#4
+	ext	v1.16b,v1.16b,v1.16b,#12
+	ext	v5.16b,v5.16b,v5.16b,#12
+	ext	v17.16b,v17.16b,v17.16b,#12
+	cbnz	x4,.Loop_neon
+
+	add	w5,w5,w22		// accumulate key block
+	add	v0.4s,v0.4s,v24.4s
+	add	x6,x6,x22,lsr#32
+	add	v4.4s,v4.4s,v24.4s
+	add	w7,w7,w23
+	add	v16.4s,v16.4s,v24.4s
+	add	x8,x8,x23,lsr#32
+	add	v2.4s,v2.4s,v26.4s
+	add	w9,w9,w24
+	add	v6.4s,v6.4s,v26.4s
+	add	x10,x10,x24,lsr#32
+	add	v18.4s,v18.4s,v26.4s
+	add	w11,w11,w25
+	add	v3.4s,v3.4s,v27.4s
+	add	x12,x12,x25,lsr#32
+	add	w13,w13,w26
+	add	v7.4s,v7.4s,v28.4s
+	add	x14,x14,x26,lsr#32
+	add	w15,w15,w27
+	add	v19.4s,v19.4s,v29.4s
+	add	x16,x16,x27,lsr#32
+	add	w17,w17,w28
+	add	v1.4s,v1.4s,v25.4s
+	add	x19,x19,x28,lsr#32
+	add	w20,w20,w30
+	add	v5.4s,v5.4s,v25.4s
+	add	x21,x21,x30,lsr#32
+	add	v17.4s,v17.4s,v25.4s
+
+	b.lo	.Ltail_neon
+
+	add	x5,x5,x6,lsl#32	// pack
+	add	x7,x7,x8,lsl#32
+	ldp	x6,x8,[x1,#0]		// load input
+	add	x9,x9,x10,lsl#32
+	add	x11,x11,x12,lsl#32
+	ldp	x10,x12,[x1,#16]
+	add	x13,x13,x14,lsl#32
+	add	x15,x15,x16,lsl#32
+	ldp	x14,x16,[x1,#32]
+	add	x17,x17,x19,lsl#32
+	add	x20,x20,x21,lsl#32
+	ldp	x19,x21,[x1,#48]
+	add	x1,x1,#64
+#ifdef	__ARMEB__
+	rev	x5,x5
+	rev	x7,x7
+	rev	x9,x9
+	rev	x11,x11
+	rev	x13,x13
+	rev	x15,x15
+	rev	x17,x17
+	rev	x20,x20
+#endif
+	ld1	{v20.16b,v21.16b,v22.16b,v23.16b},[x1],#64
+	eor	x5,x5,x6
+	eor	x7,x7,x8
+	eor	x9,x9,x10
+	eor	x11,x11,x12
+	eor	x13,x13,x14
+	eor	v0.16b,v0.16b,v20.16b
+	eor	x15,x15,x16
+	eor	v1.16b,v1.16b,v21.16b
+	eor	x17,x17,x19
+	eor	v2.16b,v2.16b,v22.16b
+	eor	x20,x20,x21
+	eor	v3.16b,v3.16b,v23.16b
+	ld1	{v20.16b,v21.16b,v22.16b,v23.16b},[x1],#64
+
+	stp	x5,x7,[x0,#0]		// store output
+	add	x28,x28,#4			// increment counter
+	stp	x9,x11,[x0,#16]
+	add	v27.4s,v27.4s,v31.4s		// += 4
+	stp	x13,x15,[x0,#32]
+	add	v28.4s,v28.4s,v31.4s
+	stp	x17,x20,[x0,#48]
+	add	v29.4s,v29.4s,v31.4s
+	add	x0,x0,#64
+
+	st1	{v0.16b,v1.16b,v2.16b,v3.16b},[x0],#64
+	ld1	{v0.16b,v1.16b,v2.16b,v3.16b},[x1],#64
+
+	eor	v4.16b,v4.16b,v20.16b
+	eor	v5.16b,v5.16b,v21.16b
+	eor	v6.16b,v6.16b,v22.16b
+	eor	v7.16b,v7.16b,v23.16b
+	st1	{v4.16b,v5.16b,v6.16b,v7.16b},[x0],#64
+
+	eor	v16.16b,v16.16b,v0.16b
+	eor	v17.16b,v17.16b,v1.16b
+	eor	v18.16b,v18.16b,v2.16b
+	eor	v19.16b,v19.16b,v3.16b
+	st1	{v16.16b,v17.16b,v18.16b,v19.16b},[x0],#64
+
+	b.hi	.Loop_outer_neon
+
+	ldp	x19,x20,[x29,#16]
+	add	sp,sp,#64
+	ldp	x21,x22,[x29,#32]
+	ldp	x23,x24,[x29,#48]
+	ldp	x25,x26,[x29,#64]
+	ldp	x27,x28,[x29,#80]
+	ldp	x29,x30,[sp],#96
+	ret
+
+.Ltail_neon:
+	add	x2,x2,#256
+	cmp	x2,#64
+	b.lo	.Less_than_64
+
+	add	x5,x5,x6,lsl#32	// pack
+	add	x7,x7,x8,lsl#32
+	ldp	x6,x8,[x1,#0]		// load input
+	add	x9,x9,x10,lsl#32
+	add	x11,x11,x12,lsl#32
+	ldp	x10,x12,[x1,#16]
+	add	x13,x13,x14,lsl#32
+	add	x15,x15,x16,lsl#32
+	ldp	x14,x16,[x1,#32]
+	add	x17,x17,x19,lsl#32
+	add	x20,x20,x21,lsl#32
+	ldp	x19,x21,[x1,#48]
+	add	x1,x1,#64
+#ifdef	__ARMEB__
+	rev	x5,x5
+	rev	x7,x7
+	rev	x9,x9
+	rev	x11,x11
+	rev	x13,x13
+	rev	x15,x15
+	rev	x17,x17
+	rev	x20,x20
+#endif
+	eor	x5,x5,x6
+	eor	x7,x7,x8
+	eor	x9,x9,x10
+	eor	x11,x11,x12
+	eor	x13,x13,x14
+	eor	x15,x15,x16
+	eor	x17,x17,x19
+	eor	x20,x20,x21
+
+	stp	x5,x7,[x0,#0]		// store output
+	add	x28,x28,#4			// increment counter
+	stp	x9,x11,[x0,#16]
+	stp	x13,x15,[x0,#32]
+	stp	x17,x20,[x0,#48]
+	add	x0,x0,#64
+	b.eq	.Ldone_neon
+	sub	x2,x2,#64
+	cmp	x2,#64
+	b.lo	.Less_than_128
+
+	ld1	{v20.16b,v21.16b,v22.16b,v23.16b},[x1],#64
+	eor	v0.16b,v0.16b,v20.16b
+	eor	v1.16b,v1.16b,v21.16b
+	eor	v2.16b,v2.16b,v22.16b
+	eor	v3.16b,v3.16b,v23.16b
+	st1	{v0.16b,v1.16b,v2.16b,v3.16b},[x0],#64
+	b.eq	.Ldone_neon
+	sub	x2,x2,#64
+	cmp	x2,#64
+	b.lo	.Less_than_192
+
+	ld1	{v20.16b,v21.16b,v22.16b,v23.16b},[x1],#64
+	eor	v4.16b,v4.16b,v20.16b
+	eor	v5.16b,v5.16b,v21.16b
+	eor	v6.16b,v6.16b,v22.16b
+	eor	v7.16b,v7.16b,v23.16b
+	st1	{v4.16b,v5.16b,v6.16b,v7.16b},[x0],#64
+	b.eq	.Ldone_neon
+	sub	x2,x2,#64
+
+	st1	{v16.16b,v17.16b,v18.16b,v19.16b},[sp]
+	b	.Last_neon
+
+.Less_than_128:
+	st1	{v0.16b,v1.16b,v2.16b,v3.16b},[sp]
+	b	.Last_neon
+.Less_than_192:
+	st1	{v4.16b,v5.16b,v6.16b,v7.16b},[sp]
+	b	.Last_neon
+
+.align	4
+.Last_neon:
+	sub	x0,x0,#1
+	add	x1,x1,x2
+	add	x0,x0,x2
+	add	x4,sp,x2
+	neg	x2,x2
+
+.Loop_tail_neon:
+	ldrb	w10,[x1,x2]
+	ldrb	w11,[x4,x2]
+	add	x2,x2,#1
+	eor	w10,w10,w11
+	strb	w10,[x0,x2]
+	cbnz	x2,.Loop_tail_neon
+
+	stp	xzr,xzr,[sp,#0]
+	stp	xzr,xzr,[sp,#16]
+	stp	xzr,xzr,[sp,#32]
+	stp	xzr,xzr,[sp,#48]
+
+.Ldone_neon:
+	ldp	x19,x20,[x29,#16]
+	add	sp,sp,#64
+	ldp	x21,x22,[x29,#32]
+	ldp	x23,x24,[x29,#48]
+	ldp	x25,x26,[x29,#64]
+	ldp	x27,x28,[x29,#80]
+	ldp	x29,x30,[sp],#96
+	ret
+.size	ChaCha20_neon,.-ChaCha20_neon
+.type	ChaCha20_512_neon,%function
+.align	5
+ChaCha20_512_neon:
+	stp	x29,x30,[sp,#-96]!
+	add	x29,sp,#0
+
+	adr	x5,.Lsigma
+	stp	x19,x20,[sp,#16]
+	stp	x21,x22,[sp,#32]
+	stp	x23,x24,[sp,#48]
+	stp	x25,x26,[sp,#64]
+	stp	x27,x28,[sp,#80]
+
+.L512_or_more_neon:
+	sub	sp,sp,#128+64
+
+	ldp	x22,x23,[x5]		// load sigma
+	ld1	{v24.4s},[x5],#16
+	ldp	x24,x25,[x3]		// load key
+	ldp	x26,x27,[x3,#16]
+	ld1	{v25.4s,v26.4s},[x3]
+	ldp	x28,x30,[x4]		// load counter
+	ld1	{v27.4s},[x4]
+	ld1	{v31.4s},[x5]
+#ifdef	__ARMEB__
+	rev64	v24.4s,v24.4s
+	ror	x24,x24,#32
+	ror	x25,x25,#32
+	ror	x26,x26,#32
+	ror	x27,x27,#32
+	ror	x28,x28,#32
+	ror	x30,x30,#32
+#endif
+	add	v27.4s,v27.4s,v31.4s		// += 1
+	stp	q24,q25,[sp,#0]		// off-load key block, invariant part
+	add	v27.4s,v27.4s,v31.4s		// not typo
+	str	q26,[sp,#32]
+	add	v28.4s,v27.4s,v31.4s
+	add	v29.4s,v28.4s,v31.4s
+	add	v30.4s,v29.4s,v31.4s
+	shl	v31.4s,v31.4s,#2			// 1 -> 4
+
+	stp	d8,d9,[sp,#128+0]		// meet ABI requirements
+	stp	d10,d11,[sp,#128+16]
+	stp	d12,d13,[sp,#128+32]
+	stp	d14,d15,[sp,#128+48]
+
+	sub	x2,x2,#512			// not typo
+
+.Loop_outer_512_neon:
+	mov	v0.16b,v24.16b
+	mov	v4.16b,v24.16b
+	mov	v8.16b,v24.16b
+	mov	v12.16b,v24.16b
+	mov	v16.16b,v24.16b
+	mov	v20.16b,v24.16b
+	mov	v1.16b,v25.16b
+	mov	w5,w22			// unpack key block
+	mov	v5.16b,v25.16b
+	lsr	x6,x22,#32
+	mov	v9.16b,v25.16b
+	mov	w7,w23
+	mov	v13.16b,v25.16b
+	lsr	x8,x23,#32
+	mov	v17.16b,v25.16b
+	mov	w9,w24
+	mov	v21.16b,v25.16b
+	lsr	x10,x24,#32
+	mov	v3.16b,v27.16b
+	mov	w11,w25
+	mov	v7.16b,v28.16b
+	lsr	x12,x25,#32
+	mov	v11.16b,v29.16b
+	mov	w13,w26
+	mov	v15.16b,v30.16b
+	lsr	x14,x26,#32
+	mov	v2.16b,v26.16b
+	mov	w15,w27
+	mov	v6.16b,v26.16b
+	lsr	x16,x27,#32
+	add	v19.4s,v3.4s,v31.4s			// +4
+	mov	w17,w28
+	add	v23.4s,v7.4s,v31.4s			// +4
+	lsr	x19,x28,#32
+	mov	v10.16b,v26.16b
+	mov	w20,w30
+	mov	v14.16b,v26.16b
+	lsr	x21,x30,#32
+	mov	v18.16b,v26.16b
+	stp	q27,q28,[sp,#48]		// off-load key block, variable part
+	mov	v22.16b,v26.16b
+	str	q29,[sp,#80]
+
+	mov	x4,#5
+	subs	x2,x2,#512
+.Loop_upper_neon:
+	sub	x4,x4,#1
+	add	v0.4s,v0.4s,v1.4s
+	add	w5,w5,w9
+	add	v4.4s,v4.4s,v5.4s
+	add	w6,w6,w10
+	add	v8.4s,v8.4s,v9.4s
+	add	w7,w7,w11
+	add	v12.4s,v12.4s,v13.4s
+	add	w8,w8,w12
+	add	v16.4s,v16.4s,v17.4s
+	eor	w17,w17,w5
+	add	v20.4s,v20.4s,v21.4s
+	eor	w19,w19,w6
+	eor	v3.16b,v3.16b,v0.16b
+	eor	w20,w20,w7
+	eor	v7.16b,v7.16b,v4.16b
+	eor	w21,w21,w8
+	eor	v11.16b,v11.16b,v8.16b
+	ror	w17,w17,#16
+	eor	v15.16b,v15.16b,v12.16b
+	ror	w19,w19,#16
+	eor	v19.16b,v19.16b,v16.16b
+	ror	w20,w20,#16
+	eor	v23.16b,v23.16b,v20.16b
+	ror	w21,w21,#16
+	rev32	v3.8h,v3.8h
+	add	w13,w13,w17
+	rev32	v7.8h,v7.8h
+	add	w14,w14,w19
+	rev32	v11.8h,v11.8h
+	add	w15,w15,w20
+	rev32	v15.8h,v15.8h
+	add	w16,w16,w21
+	rev32	v19.8h,v19.8h
+	eor	w9,w9,w13
+	rev32	v23.8h,v23.8h
+	eor	w10,w10,w14
+	add	v2.4s,v2.4s,v3.4s
+	eor	w11,w11,w15
+	add	v6.4s,v6.4s,v7.4s
+	eor	w12,w12,w16
+	add	v10.4s,v10.4s,v11.4s
+	ror	w9,w9,#20
+	add	v14.4s,v14.4s,v15.4s
+	ror	w10,w10,#20
+	add	v18.4s,v18.4s,v19.4s
+	ror	w11,w11,#20
+	add	v22.4s,v22.4s,v23.4s
+	ror	w12,w12,#20
+	eor	v24.16b,v1.16b,v2.16b
+	add	w5,w5,w9
+	eor	v25.16b,v5.16b,v6.16b
+	add	w6,w6,w10
+	eor	v26.16b,v9.16b,v10.16b
+	add	w7,w7,w11
+	eor	v27.16b,v13.16b,v14.16b
+	add	w8,w8,w12
+	eor	v28.16b,v17.16b,v18.16b
+	eor	w17,w17,w5
+	eor	v29.16b,v21.16b,v22.16b
+	eor	w19,w19,w6
+	ushr	v1.4s,v24.4s,#20
+	eor	w20,w20,w7
+	ushr	v5.4s,v25.4s,#20
+	eor	w21,w21,w8
+	ushr	v9.4s,v26.4s,#20
+	ror	w17,w17,#24
+	ushr	v13.4s,v27.4s,#20
+	ror	w19,w19,#24
+	ushr	v17.4s,v28.4s,#20
+	ror	w20,w20,#24
+	ushr	v21.4s,v29.4s,#20
+	ror	w21,w21,#24
+	sli	v1.4s,v24.4s,#12
+	add	w13,w13,w17
+	sli	v5.4s,v25.4s,#12
+	add	w14,w14,w19
+	sli	v9.4s,v26.4s,#12
+	add	w15,w15,w20
+	sli	v13.4s,v27.4s,#12
+	add	w16,w16,w21
+	sli	v17.4s,v28.4s,#12
+	eor	w9,w9,w13
+	sli	v21.4s,v29.4s,#12
+	eor	w10,w10,w14
+	add	v0.4s,v0.4s,v1.4s
+	eor	w11,w11,w15
+	add	v4.4s,v4.4s,v5.4s
+	eor	w12,w12,w16
+	add	v8.4s,v8.4s,v9.4s
+	ror	w9,w9,#25
+	add	v12.4s,v12.4s,v13.4s
+	ror	w10,w10,#25
+	add	v16.4s,v16.4s,v17.4s
+	ror	w11,w11,#25
+	add	v20.4s,v20.4s,v21.4s
+	ror	w12,w12,#25
+	eor	v24.16b,v3.16b,v0.16b
+	add	w5,w5,w10
+	eor	v25.16b,v7.16b,v4.16b
+	add	w6,w6,w11
+	eor	v26.16b,v11.16b,v8.16b
+	add	w7,w7,w12
+	eor	v27.16b,v15.16b,v12.16b
+	add	w8,w8,w9
+	eor	v28.16b,v19.16b,v16.16b
+	eor	w21,w21,w5
+	eor	v29.16b,v23.16b,v20.16b
+	eor	w17,w17,w6
+	ushr	v3.4s,v24.4s,#24
+	eor	w19,w19,w7
+	ushr	v7.4s,v25.4s,#24
+	eor	w20,w20,w8
+	ushr	v11.4s,v26.4s,#24
+	ror	w21,w21,#16
+	ushr	v15.4s,v27.4s,#24
+	ror	w17,w17,#16
+	ushr	v19.4s,v28.4s,#24
+	ror	w19,w19,#16
+	ushr	v23.4s,v29.4s,#24
+	ror	w20,w20,#16
+	sli	v3.4s,v24.4s,#8
+	add	w15,w15,w21
+	sli	v7.4s,v25.4s,#8
+	add	w16,w16,w17
+	sli	v11.4s,v26.4s,#8
+	add	w13,w13,w19
+	sli	v15.4s,v27.4s,#8
+	add	w14,w14,w20
+	sli	v19.4s,v28.4s,#8
+	eor	w10,w10,w15
+	sli	v23.4s,v29.4s,#8
+	eor	w11,w11,w16
+	add	v2.4s,v2.4s,v3.4s
+	eor	w12,w12,w13
+	add	v6.4s,v6.4s,v7.4s
+	eor	w9,w9,w14
+	add	v10.4s,v10.4s,v11.4s
+	ror	w10,w10,#20
+	add	v14.4s,v14.4s,v15.4s
+	ror	w11,w11,#20
+	add	v18.4s,v18.4s,v19.4s
+	ror	w12,w12,#20
+	add	v22.4s,v22.4s,v23.4s
+	ror	w9,w9,#20
+	eor	v24.16b,v1.16b,v2.16b
+	add	w5,w5,w10
+	eor	v25.16b,v5.16b,v6.16b
+	add	w6,w6,w11
+	eor	v26.16b,v9.16b,v10.16b
+	add	w7,w7,w12
+	eor	v27.16b,v13.16b,v14.16b
+	add	w8,w8,w9
+	eor	v28.16b,v17.16b,v18.16b
+	eor	w21,w21,w5
+	eor	v29.16b,v21.16b,v22.16b
+	eor	w17,w17,w6
+	ushr	v1.4s,v24.4s,#25
+	eor	w19,w19,w7
+	ushr	v5.4s,v25.4s,#25
+	eor	w20,w20,w8
+	ushr	v9.4s,v26.4s,#25
+	ror	w21,w21,#24
+	ushr	v13.4s,v27.4s,#25
+	ror	w17,w17,#24
+	ushr	v17.4s,v28.4s,#25
+	ror	w19,w19,#24
+	ushr	v21.4s,v29.4s,#25
+	ror	w20,w20,#24
+	sli	v1.4s,v24.4s,#7
+	add	w15,w15,w21
+	sli	v5.4s,v25.4s,#7
+	add	w16,w16,w17
+	sli	v9.4s,v26.4s,#7
+	add	w13,w13,w19
+	sli	v13.4s,v27.4s,#7
+	add	w14,w14,w20
+	sli	v17.4s,v28.4s,#7
+	eor	w10,w10,w15
+	sli	v21.4s,v29.4s,#7
+	eor	w11,w11,w16
+	ext	v2.16b,v2.16b,v2.16b,#8
+	eor	w12,w12,w13
+	ext	v6.16b,v6.16b,v6.16b,#8
+	eor	w9,w9,w14
+	ext	v10.16b,v10.16b,v10.16b,#8
+	ror	w10,w10,#25
+	ext	v14.16b,v14.16b,v14.16b,#8
+	ror	w11,w11,#25
+	ext	v18.16b,v18.16b,v18.16b,#8
+	ror	w12,w12,#25
+	ext	v22.16b,v22.16b,v22.16b,#8
+	ror	w9,w9,#25
+	ext	v3.16b,v3.16b,v3.16b,#12
+	ext	v7.16b,v7.16b,v7.16b,#12
+	ext	v11.16b,v11.16b,v11.16b,#12
+	ext	v15.16b,v15.16b,v15.16b,#12
+	ext	v19.16b,v19.16b,v19.16b,#12
+	ext	v23.16b,v23.16b,v23.16b,#12
+	ext	v1.16b,v1.16b,v1.16b,#4
+	ext	v5.16b,v5.16b,v5.16b,#4
+	ext	v9.16b,v9.16b,v9.16b,#4
+	ext	v13.16b,v13.16b,v13.16b,#4
+	ext	v17.16b,v17.16b,v17.16b,#4
+	ext	v21.16b,v21.16b,v21.16b,#4
+	add	v0.4s,v0.4s,v1.4s
+	add	w5,w5,w9
+	add	v4.4s,v4.4s,v5.4s
+	add	w6,w6,w10
+	add	v8.4s,v8.4s,v9.4s
+	add	w7,w7,w11
+	add	v12.4s,v12.4s,v13.4s
+	add	w8,w8,w12
+	add	v16.4s,v16.4s,v17.4s
+	eor	w17,w17,w5
+	add	v20.4s,v20.4s,v21.4s
+	eor	w19,w19,w6
+	eor	v3.16b,v3.16b,v0.16b
+	eor	w20,w20,w7
+	eor	v7.16b,v7.16b,v4.16b
+	eor	w21,w21,w8
+	eor	v11.16b,v11.16b,v8.16b
+	ror	w17,w17,#16
+	eor	v15.16b,v15.16b,v12.16b
+	ror	w19,w19,#16
+	eor	v19.16b,v19.16b,v16.16b
+	ror	w20,w20,#16
+	eor	v23.16b,v23.16b,v20.16b
+	ror	w21,w21,#16
+	rev32	v3.8h,v3.8h
+	add	w13,w13,w17
+	rev32	v7.8h,v7.8h
+	add	w14,w14,w19
+	rev32	v11.8h,v11.8h
+	add	w15,w15,w20
+	rev32	v15.8h,v15.8h
+	add	w16,w16,w21
+	rev32	v19.8h,v19.8h
+	eor	w9,w9,w13
+	rev32	v23.8h,v23.8h
+	eor	w10,w10,w14
+	add	v2.4s,v2.4s,v3.4s
+	eor	w11,w11,w15
+	add	v6.4s,v6.4s,v7.4s
+	eor	w12,w12,w16
+	add	v10.4s,v10.4s,v11.4s
+	ror	w9,w9,#20
+	add	v14.4s,v14.4s,v15.4s
+	ror	w10,w10,#20
+	add	v18.4s,v18.4s,v19.4s
+	ror	w11,w11,#20
+	add	v22.4s,v22.4s,v23.4s
+	ror	w12,w12,#20
+	eor	v24.16b,v1.16b,v2.16b
+	add	w5,w5,w9
+	eor	v25.16b,v5.16b,v6.16b
+	add	w6,w6,w10
+	eor	v26.16b,v9.16b,v10.16b
+	add	w7,w7,w11
+	eor	v27.16b,v13.16b,v14.16b
+	add	w8,w8,w12
+	eor	v28.16b,v17.16b,v18.16b
+	eor	w17,w17,w5
+	eor	v29.16b,v21.16b,v22.16b
+	eor	w19,w19,w6
+	ushr	v1.4s,v24.4s,#20
+	eor	w20,w20,w7
+	ushr	v5.4s,v25.4s,#20
+	eor	w21,w21,w8
+	ushr	v9.4s,v26.4s,#20
+	ror	w17,w17,#24
+	ushr	v13.4s,v27.4s,#20
+	ror	w19,w19,#24
+	ushr	v17.4s,v28.4s,#20
+	ror	w20,w20,#24
+	ushr	v21.4s,v29.4s,#20
+	ror	w21,w21,#24
+	sli	v1.4s,v24.4s,#12
+	add	w13,w13,w17
+	sli	v5.4s,v25.4s,#12
+	add	w14,w14,w19
+	sli	v9.4s,v26.4s,#12
+	add	w15,w15,w20
+	sli	v13.4s,v27.4s,#12
+	add	w16,w16,w21
+	sli	v17.4s,v28.4s,#12
+	eor	w9,w9,w13
+	sli	v21.4s,v29.4s,#12
+	eor	w10,w10,w14
+	add	v0.4s,v0.4s,v1.4s
+	eor	w11,w11,w15
+	add	v4.4s,v4.4s,v5.4s
+	eor	w12,w12,w16
+	add	v8.4s,v8.4s,v9.4s
+	ror	w9,w9,#25
+	add	v12.4s,v12.4s,v13.4s
+	ror	w10,w10,#25
+	add	v16.4s,v16.4s,v17.4s
+	ror	w11,w11,#25
+	add	v20.4s,v20.4s,v21.4s
+	ror	w12,w12,#25
+	eor	v24.16b,v3.16b,v0.16b
+	add	w5,w5,w10
+	eor	v25.16b,v7.16b,v4.16b
+	add	w6,w6,w11
+	eor	v26.16b,v11.16b,v8.16b
+	add	w7,w7,w12
+	eor	v27.16b,v15.16b,v12.16b
+	add	w8,w8,w9
+	eor	v28.16b,v19.16b,v16.16b
+	eor	w21,w21,w5
+	eor	v29.16b,v23.16b,v20.16b
+	eor	w17,w17,w6
+	ushr	v3.4s,v24.4s,#24
+	eor	w19,w19,w7
+	ushr	v7.4s,v25.4s,#24
+	eor	w20,w20,w8
+	ushr	v11.4s,v26.4s,#24
+	ror	w21,w21,#16
+	ushr	v15.4s,v27.4s,#24
+	ror	w17,w17,#16
+	ushr	v19.4s,v28.4s,#24
+	ror	w19,w19,#16
+	ushr	v23.4s,v29.4s,#24
+	ror	w20,w20,#16
+	sli	v3.4s,v24.4s,#8
+	add	w15,w15,w21
+	sli	v7.4s,v25.4s,#8
+	add	w16,w16,w17
+	sli	v11.4s,v26.4s,#8
+	add	w13,w13,w19
+	sli	v15.4s,v27.4s,#8
+	add	w14,w14,w20
+	sli	v19.4s,v28.4s,#8
+	eor	w10,w10,w15
+	sli	v23.4s,v29.4s,#8
+	eor	w11,w11,w16
+	add	v2.4s,v2.4s,v3.4s
+	eor	w12,w12,w13
+	add	v6.4s,v6.4s,v7.4s
+	eor	w9,w9,w14
+	add	v10.4s,v10.4s,v11.4s
+	ror	w10,w10,#20
+	add	v14.4s,v14.4s,v15.4s
+	ror	w11,w11,#20
+	add	v18.4s,v18.4s,v19.4s
+	ror	w12,w12,#20
+	add	v22.4s,v22.4s,v23.4s
+	ror	w9,w9,#20
+	eor	v24.16b,v1.16b,v2.16b
+	add	w5,w5,w10
+	eor	v25.16b,v5.16b,v6.16b
+	add	w6,w6,w11
+	eor	v26.16b,v9.16b,v10.16b
+	add	w7,w7,w12
+	eor	v27.16b,v13.16b,v14.16b
+	add	w8,w8,w9
+	eor	v28.16b,v17.16b,v18.16b
+	eor	w21,w21,w5
+	eor	v29.16b,v21.16b,v22.16b
+	eor	w17,w17,w6
+	ushr	v1.4s,v24.4s,#25
+	eor	w19,w19,w7
+	ushr	v5.4s,v25.4s,#25
+	eor	w20,w20,w8
+	ushr	v9.4s,v26.4s,#25
+	ror	w21,w21,#24
+	ushr	v13.4s,v27.4s,#25
+	ror	w17,w17,#24
+	ushr	v17.4s,v28.4s,#25
+	ror	w19,w19,#24
+	ushr	v21.4s,v29.4s,#25
+	ror	w20,w20,#24
+	sli	v1.4s,v24.4s,#7
+	add	w15,w15,w21
+	sli	v5.4s,v25.4s,#7
+	add	w16,w16,w17
+	sli	v9.4s,v26.4s,#7
+	add	w13,w13,w19
+	sli	v13.4s,v27.4s,#7
+	add	w14,w14,w20
+	sli	v17.4s,v28.4s,#7
+	eor	w10,w10,w15
+	sli	v21.4s,v29.4s,#7
+	eor	w11,w11,w16
+	ext	v2.16b,v2.16b,v2.16b,#8
+	eor	w12,w12,w13
+	ext	v6.16b,v6.16b,v6.16b,#8
+	eor	w9,w9,w14
+	ext	v10.16b,v10.16b,v10.16b,#8
+	ror	w10,w10,#25
+	ext	v14.16b,v14.16b,v14.16b,#8
+	ror	w11,w11,#25
+	ext	v18.16b,v18.16b,v18.16b,#8
+	ror	w12,w12,#25
+	ext	v22.16b,v22.16b,v22.16b,#8
+	ror	w9,w9,#25
+	ext	v3.16b,v3.16b,v3.16b,#4
+	ext	v7.16b,v7.16b,v7.16b,#4
+	ext	v11.16b,v11.16b,v11.16b,#4
+	ext	v15.16b,v15.16b,v15.16b,#4
+	ext	v19.16b,v19.16b,v19.16b,#4
+	ext	v23.16b,v23.16b,v23.16b,#4
+	ext	v1.16b,v1.16b,v1.16b,#12
+	ext	v5.16b,v5.16b,v5.16b,#12
+	ext	v9.16b,v9.16b,v9.16b,#12
+	ext	v13.16b,v13.16b,v13.16b,#12
+	ext	v17.16b,v17.16b,v17.16b,#12
+	ext	v21.16b,v21.16b,v21.16b,#12
+	cbnz	x4,.Loop_upper_neon
+
+	add	w5,w5,w22		// accumulate key block
+	add	x6,x6,x22,lsr#32
+	add	w7,w7,w23
+	add	x8,x8,x23,lsr#32
+	add	w9,w9,w24
+	add	x10,x10,x24,lsr#32
+	add	w11,w11,w25
+	add	x12,x12,x25,lsr#32
+	add	w13,w13,w26
+	add	x14,x14,x26,lsr#32
+	add	w15,w15,w27
+	add	x16,x16,x27,lsr#32
+	add	w17,w17,w28
+	add	x19,x19,x28,lsr#32
+	add	w20,w20,w30
+	add	x21,x21,x30,lsr#32
+
+	add	x5,x5,x6,lsl#32	// pack
+	add	x7,x7,x8,lsl#32
+	ldp	x6,x8,[x1,#0]		// load input
+	add	x9,x9,x10,lsl#32
+	add	x11,x11,x12,lsl#32
+	ldp	x10,x12,[x1,#16]
+	add	x13,x13,x14,lsl#32
+	add	x15,x15,x16,lsl#32
+	ldp	x14,x16,[x1,#32]
+	add	x17,x17,x19,lsl#32
+	add	x20,x20,x21,lsl#32
+	ldp	x19,x21,[x1,#48]
+	add	x1,x1,#64
+#ifdef	__ARMEB__
+	rev	x5,x5
+	rev	x7,x7
+	rev	x9,x9
+	rev	x11,x11
+	rev	x13,x13
+	rev	x15,x15
+	rev	x17,x17
+	rev	x20,x20
+#endif
+	eor	x5,x5,x6
+	eor	x7,x7,x8
+	eor	x9,x9,x10
+	eor	x11,x11,x12
+	eor	x13,x13,x14
+	eor	x15,x15,x16
+	eor	x17,x17,x19
+	eor	x20,x20,x21
+
+	stp	x5,x7,[x0,#0]		// store output
+	add	x28,x28,#1			// increment counter
+	mov	w5,w22			// unpack key block
+	lsr	x6,x22,#32
+	stp	x9,x11,[x0,#16]
+	mov	w7,w23
+	lsr	x8,x23,#32
+	stp	x13,x15,[x0,#32]
+	mov	w9,w24
+	lsr	x10,x24,#32
+	stp	x17,x20,[x0,#48]
+	add	x0,x0,#64
+	mov	w11,w25
+	lsr	x12,x25,#32
+	mov	w13,w26
+	lsr	x14,x26,#32
+	mov	w15,w27
+	lsr	x16,x27,#32
+	mov	w17,w28
+	lsr	x19,x28,#32
+	mov	w20,w30
+	lsr	x21,x30,#32
+
+	mov	x4,#5
+.Loop_lower_neon:
+	sub	x4,x4,#1
+	add	v0.4s,v0.4s,v1.4s
+	add	w5,w5,w9
+	add	v4.4s,v4.4s,v5.4s
+	add	w6,w6,w10
+	add	v8.4s,v8.4s,v9.4s
+	add	w7,w7,w11
+	add	v12.4s,v12.4s,v13.4s
+	add	w8,w8,w12
+	add	v16.4s,v16.4s,v17.4s
+	eor	w17,w17,w5
+	add	v20.4s,v20.4s,v21.4s
+	eor	w19,w19,w6
+	eor	v3.16b,v3.16b,v0.16b
+	eor	w20,w20,w7
+	eor	v7.16b,v7.16b,v4.16b
+	eor	w21,w21,w8
+	eor	v11.16b,v11.16b,v8.16b
+	ror	w17,w17,#16
+	eor	v15.16b,v15.16b,v12.16b
+	ror	w19,w19,#16
+	eor	v19.16b,v19.16b,v16.16b
+	ror	w20,w20,#16
+	eor	v23.16b,v23.16b,v20.16b
+	ror	w21,w21,#16
+	rev32	v3.8h,v3.8h
+	add	w13,w13,w17
+	rev32	v7.8h,v7.8h
+	add	w14,w14,w19
+	rev32	v11.8h,v11.8h
+	add	w15,w15,w20
+	rev32	v15.8h,v15.8h
+	add	w16,w16,w21
+	rev32	v19.8h,v19.8h
+	eor	w9,w9,w13
+	rev32	v23.8h,v23.8h
+	eor	w10,w10,w14
+	add	v2.4s,v2.4s,v3.4s
+	eor	w11,w11,w15
+	add	v6.4s,v6.4s,v7.4s
+	eor	w12,w12,w16
+	add	v10.4s,v10.4s,v11.4s
+	ror	w9,w9,#20
+	add	v14.4s,v14.4s,v15.4s
+	ror	w10,w10,#20
+	add	v18.4s,v18.4s,v19.4s
+	ror	w11,w11,#20
+	add	v22.4s,v22.4s,v23.4s
+	ror	w12,w12,#20
+	eor	v24.16b,v1.16b,v2.16b
+	add	w5,w5,w9
+	eor	v25.16b,v5.16b,v6.16b
+	add	w6,w6,w10
+	eor	v26.16b,v9.16b,v10.16b
+	add	w7,w7,w11
+	eor	v27.16b,v13.16b,v14.16b
+	add	w8,w8,w12
+	eor	v28.16b,v17.16b,v18.16b
+	eor	w17,w17,w5
+	eor	v29.16b,v21.16b,v22.16b
+	eor	w19,w19,w6
+	ushr	v1.4s,v24.4s,#20
+	eor	w20,w20,w7
+	ushr	v5.4s,v25.4s,#20
+	eor	w21,w21,w8
+	ushr	v9.4s,v26.4s,#20
+	ror	w17,w17,#24
+	ushr	v13.4s,v27.4s,#20
+	ror	w19,w19,#24
+	ushr	v17.4s,v28.4s,#20
+	ror	w20,w20,#24
+	ushr	v21.4s,v29.4s,#20
+	ror	w21,w21,#24
+	sli	v1.4s,v24.4s,#12
+	add	w13,w13,w17
+	sli	v5.4s,v25.4s,#12
+	add	w14,w14,w19
+	sli	v9.4s,v26.4s,#12
+	add	w15,w15,w20
+	sli	v13.4s,v27.4s,#12
+	add	w16,w16,w21
+	sli	v17.4s,v28.4s,#12
+	eor	w9,w9,w13
+	sli	v21.4s,v29.4s,#12
+	eor	w10,w10,w14
+	add	v0.4s,v0.4s,v1.4s
+	eor	w11,w11,w15
+	add	v4.4s,v4.4s,v5.4s
+	eor	w12,w12,w16
+	add	v8.4s,v8.4s,v9.4s
+	ror	w9,w9,#25
+	add	v12.4s,v12.4s,v13.4s
+	ror	w10,w10,#25
+	add	v16.4s,v16.4s,v17.4s
+	ror	w11,w11,#25
+	add	v20.4s,v20.4s,v21.4s
+	ror	w12,w12,#25
+	eor	v24.16b,v3.16b,v0.16b
+	add	w5,w5,w10
+	eor	v25.16b,v7.16b,v4.16b
+	add	w6,w6,w11
+	eor	v26.16b,v11.16b,v8.16b
+	add	w7,w7,w12
+	eor	v27.16b,v15.16b,v12.16b
+	add	w8,w8,w9
+	eor	v28.16b,v19.16b,v16.16b
+	eor	w21,w21,w5
+	eor	v29.16b,v23.16b,v20.16b
+	eor	w17,w17,w6
+	ushr	v3.4s,v24.4s,#24
+	eor	w19,w19,w7
+	ushr	v7.4s,v25.4s,#24
+	eor	w20,w20,w8
+	ushr	v11.4s,v26.4s,#24
+	ror	w21,w21,#16
+	ushr	v15.4s,v27.4s,#24
+	ror	w17,w17,#16
+	ushr	v19.4s,v28.4s,#24
+	ror	w19,w19,#16
+	ushr	v23.4s,v29.4s,#24
+	ror	w20,w20,#16
+	sli	v3.4s,v24.4s,#8
+	add	w15,w15,w21
+	sli	v7.4s,v25.4s,#8
+	add	w16,w16,w17
+	sli	v11.4s,v26.4s,#8
+	add	w13,w13,w19
+	sli	v15.4s,v27.4s,#8
+	add	w14,w14,w20
+	sli	v19.4s,v28.4s,#8
+	eor	w10,w10,w15
+	sli	v23.4s,v29.4s,#8
+	eor	w11,w11,w16
+	add	v2.4s,v2.4s,v3.4s
+	eor	w12,w12,w13
+	add	v6.4s,v6.4s,v7.4s
+	eor	w9,w9,w14
+	add	v10.4s,v10.4s,v11.4s
+	ror	w10,w10,#20
+	add	v14.4s,v14.4s,v15.4s
+	ror	w11,w11,#20
+	add	v18.4s,v18.4s,v19.4s
+	ror	w12,w12,#20
+	add	v22.4s,v22.4s,v23.4s
+	ror	w9,w9,#20
+	eor	v24.16b,v1.16b,v2.16b
+	add	w5,w5,w10
+	eor	v25.16b,v5.16b,v6.16b
+	add	w6,w6,w11
+	eor	v26.16b,v9.16b,v10.16b
+	add	w7,w7,w12
+	eor	v27.16b,v13.16b,v14.16b
+	add	w8,w8,w9
+	eor	v28.16b,v17.16b,v18.16b
+	eor	w21,w21,w5
+	eor	v29.16b,v21.16b,v22.16b
+	eor	w17,w17,w6
+	ushr	v1.4s,v24.4s,#25
+	eor	w19,w19,w7
+	ushr	v5.4s,v25.4s,#25
+	eor	w20,w20,w8
+	ushr	v9.4s,v26.4s,#25
+	ror	w21,w21,#24
+	ushr	v13.4s,v27.4s,#25
+	ror	w17,w17,#24
+	ushr	v17.4s,v28.4s,#25
+	ror	w19,w19,#24
+	ushr	v21.4s,v29.4s,#25
+	ror	w20,w20,#24
+	sli	v1.4s,v24.4s,#7
+	add	w15,w15,w21
+	sli	v5.4s,v25.4s,#7
+	add	w16,w16,w17
+	sli	v9.4s,v26.4s,#7
+	add	w13,w13,w19
+	sli	v13.4s,v27.4s,#7
+	add	w14,w14,w20
+	sli	v17.4s,v28.4s,#7
+	eor	w10,w10,w15
+	sli	v21.4s,v29.4s,#7
+	eor	w11,w11,w16
+	ext	v2.16b,v2.16b,v2.16b,#8
+	eor	w12,w12,w13
+	ext	v6.16b,v6.16b,v6.16b,#8
+	eor	w9,w9,w14
+	ext	v10.16b,v10.16b,v10.16b,#8
+	ror	w10,w10,#25
+	ext	v14.16b,v14.16b,v14.16b,#8
+	ror	w11,w11,#25
+	ext	v18.16b,v18.16b,v18.16b,#8
+	ror	w12,w12,#25
+	ext	v22.16b,v22.16b,v22.16b,#8
+	ror	w9,w9,#25
+	ext	v3.16b,v3.16b,v3.16b,#12
+	ext	v7.16b,v7.16b,v7.16b,#12
+	ext	v11.16b,v11.16b,v11.16b,#12
+	ext	v15.16b,v15.16b,v15.16b,#12
+	ext	v19.16b,v19.16b,v19.16b,#12
+	ext	v23.16b,v23.16b,v23.16b,#12
+	ext	v1.16b,v1.16b,v1.16b,#4
+	ext	v5.16b,v5.16b,v5.16b,#4
+	ext	v9.16b,v9.16b,v9.16b,#4
+	ext	v13.16b,v13.16b,v13.16b,#4
+	ext	v17.16b,v17.16b,v17.16b,#4
+	ext	v21.16b,v21.16b,v21.16b,#4
+	add	v0.4s,v0.4s,v1.4s
+	add	w5,w5,w9
+	add	v4.4s,v4.4s,v5.4s
+	add	w6,w6,w10
+	add	v8.4s,v8.4s,v9.4s
+	add	w7,w7,w11
+	add	v12.4s,v12.4s,v13.4s
+	add	w8,w8,w12
+	add	v16.4s,v16.4s,v17.4s
+	eor	w17,w17,w5
+	add	v20.4s,v20.4s,v21.4s
+	eor	w19,w19,w6
+	eor	v3.16b,v3.16b,v0.16b
+	eor	w20,w20,w7
+	eor	v7.16b,v7.16b,v4.16b
+	eor	w21,w21,w8
+	eor	v11.16b,v11.16b,v8.16b
+	ror	w17,w17,#16
+	eor	v15.16b,v15.16b,v12.16b
+	ror	w19,w19,#16
+	eor	v19.16b,v19.16b,v16.16b
+	ror	w20,w20,#16
+	eor	v23.16b,v23.16b,v20.16b
+	ror	w21,w21,#16
+	rev32	v3.8h,v3.8h
+	add	w13,w13,w17
+	rev32	v7.8h,v7.8h
+	add	w14,w14,w19
+	rev32	v11.8h,v11.8h
+	add	w15,w15,w20
+	rev32	v15.8h,v15.8h
+	add	w16,w16,w21
+	rev32	v19.8h,v19.8h
+	eor	w9,w9,w13
+	rev32	v23.8h,v23.8h
+	eor	w10,w10,w14
+	add	v2.4s,v2.4s,v3.4s
+	eor	w11,w11,w15
+	add	v6.4s,v6.4s,v7.4s
+	eor	w12,w12,w16
+	add	v10.4s,v10.4s,v11.4s
+	ror	w9,w9,#20
+	add	v14.4s,v14.4s,v15.4s
+	ror	w10,w10,#20
+	add	v18.4s,v18.4s,v19.4s
+	ror	w11,w11,#20
+	add	v22.4s,v22.4s,v23.4s
+	ror	w12,w12,#20
+	eor	v24.16b,v1.16b,v2.16b
+	add	w5,w5,w9
+	eor	v25.16b,v5.16b,v6.16b
+	add	w6,w6,w10
+	eor	v26.16b,v9.16b,v10.16b
+	add	w7,w7,w11
+	eor	v27.16b,v13.16b,v14.16b
+	add	w8,w8,w12
+	eor	v28.16b,v17.16b,v18.16b
+	eor	w17,w17,w5
+	eor	v29.16b,v21.16b,v22.16b
+	eor	w19,w19,w6
+	ushr	v1.4s,v24.4s,#20
+	eor	w20,w20,w7
+	ushr	v5.4s,v25.4s,#20
+	eor	w21,w21,w8
+	ushr	v9.4s,v26.4s,#20
+	ror	w17,w17,#24
+	ushr	v13.4s,v27.4s,#20
+	ror	w19,w19,#24
+	ushr	v17.4s,v28.4s,#20
+	ror	w20,w20,#24
+	ushr	v21.4s,v29.4s,#20
+	ror	w21,w21,#24
+	sli	v1.4s,v24.4s,#12
+	add	w13,w13,w17
+	sli	v5.4s,v25.4s,#12
+	add	w14,w14,w19
+	sli	v9.4s,v26.4s,#12
+	add	w15,w15,w20
+	sli	v13.4s,v27.4s,#12
+	add	w16,w16,w21
+	sli	v17.4s,v28.4s,#12
+	eor	w9,w9,w13
+	sli	v21.4s,v29.4s,#12
+	eor	w10,w10,w14
+	add	v0.4s,v0.4s,v1.4s
+	eor	w11,w11,w15
+	add	v4.4s,v4.4s,v5.4s
+	eor	w12,w12,w16
+	add	v8.4s,v8.4s,v9.4s
+	ror	w9,w9,#25
+	add	v12.4s,v12.4s,v13.4s
+	ror	w10,w10,#25
+	add	v16.4s,v16.4s,v17.4s
+	ror	w11,w11,#25
+	add	v20.4s,v20.4s,v21.4s
+	ror	w12,w12,#25
+	eor	v24.16b,v3.16b,v0.16b
+	add	w5,w5,w10
+	eor	v25.16b,v7.16b,v4.16b
+	add	w6,w6,w11
+	eor	v26.16b,v11.16b,v8.16b
+	add	w7,w7,w12
+	eor	v27.16b,v15.16b,v12.16b
+	add	w8,w8,w9
+	eor	v28.16b,v19.16b,v16.16b
+	eor	w21,w21,w5
+	eor	v29.16b,v23.16b,v20.16b
+	eor	w17,w17,w6
+	ushr	v3.4s,v24.4s,#24
+	eor	w19,w19,w7
+	ushr	v7.4s,v25.4s,#24
+	eor	w20,w20,w8
+	ushr	v11.4s,v26.4s,#24
+	ror	w21,w21,#16
+	ushr	v15.4s,v27.4s,#24
+	ror	w17,w17,#16
+	ushr	v19.4s,v28.4s,#24
+	ror	w19,w19,#16
+	ushr	v23.4s,v29.4s,#24
+	ror	w20,w20,#16
+	sli	v3.4s,v24.4s,#8
+	add	w15,w15,w21
+	sli	v7.4s,v25.4s,#8
+	add	w16,w16,w17
+	sli	v11.4s,v26.4s,#8
+	add	w13,w13,w19
+	sli	v15.4s,v27.4s,#8
+	add	w14,w14,w20
+	sli	v19.4s,v28.4s,#8
+	eor	w10,w10,w15
+	sli	v23.4s,v29.4s,#8
+	eor	w11,w11,w16
+	add	v2.4s,v2.4s,v3.4s
+	eor	w12,w12,w13
+	add	v6.4s,v6.4s,v7.4s
+	eor	w9,w9,w14
+	add	v10.4s,v10.4s,v11.4s
+	ror	w10,w10,#20
+	add	v14.4s,v14.4s,v15.4s
+	ror	w11,w11,#20
+	add	v18.4s,v18.4s,v19.4s
+	ror	w12,w12,#20
+	add	v22.4s,v22.4s,v23.4s
+	ror	w9,w9,#20
+	eor	v24.16b,v1.16b,v2.16b
+	add	w5,w5,w10
+	eor	v25.16b,v5.16b,v6.16b
+	add	w6,w6,w11
+	eor	v26.16b,v9.16b,v10.16b
+	add	w7,w7,w12
+	eor	v27.16b,v13.16b,v14.16b
+	add	w8,w8,w9
+	eor	v28.16b,v17.16b,v18.16b
+	eor	w21,w21,w5
+	eor	v29.16b,v21.16b,v22.16b
+	eor	w17,w17,w6
+	ushr	v1.4s,v24.4s,#25
+	eor	w19,w19,w7
+	ushr	v5.4s,v25.4s,#25
+	eor	w20,w20,w8
+	ushr	v9.4s,v26.4s,#25
+	ror	w21,w21,#24
+	ushr	v13.4s,v27.4s,#25
+	ror	w17,w17,#24
+	ushr	v17.4s,v28.4s,#25
+	ror	w19,w19,#24
+	ushr	v21.4s,v29.4s,#25
+	ror	w20,w20,#24
+	sli	v1.4s,v24.4s,#7
+	add	w15,w15,w21
+	sli	v5.4s,v25.4s,#7
+	add	w16,w16,w17
+	sli	v9.4s,v26.4s,#7
+	add	w13,w13,w19
+	sli	v13.4s,v27.4s,#7
+	add	w14,w14,w20
+	sli	v17.4s,v28.4s,#7
+	eor	w10,w10,w15
+	sli	v21.4s,v29.4s,#7
+	eor	w11,w11,w16
+	ext	v2.16b,v2.16b,v2.16b,#8
+	eor	w12,w12,w13
+	ext	v6.16b,v6.16b,v6.16b,#8
+	eor	w9,w9,w14
+	ext	v10.16b,v10.16b,v10.16b,#8
+	ror	w10,w10,#25
+	ext	v14.16b,v14.16b,v14.16b,#8
+	ror	w11,w11,#25
+	ext	v18.16b,v18.16b,v18.16b,#8
+	ror	w12,w12,#25
+	ext	v22.16b,v22.16b,v22.16b,#8
+	ror	w9,w9,#25
+	ext	v3.16b,v3.16b,v3.16b,#4
+	ext	v7.16b,v7.16b,v7.16b,#4
+	ext	v11.16b,v11.16b,v11.16b,#4
+	ext	v15.16b,v15.16b,v15.16b,#4
+	ext	v19.16b,v19.16b,v19.16b,#4
+	ext	v23.16b,v23.16b,v23.16b,#4
+	ext	v1.16b,v1.16b,v1.16b,#12
+	ext	v5.16b,v5.16b,v5.16b,#12
+	ext	v9.16b,v9.16b,v9.16b,#12
+	ext	v13.16b,v13.16b,v13.16b,#12
+	ext	v17.16b,v17.16b,v17.16b,#12
+	ext	v21.16b,v21.16b,v21.16b,#12
+	cbnz	x4,.Loop_lower_neon
+
+	add	w5,w5,w22		// accumulate key block
+	ldp	q24,q25,[sp,#0]
+	add	x6,x6,x22,lsr#32
+	ldp	q26,q27,[sp,#32]
+	add	w7,w7,w23
+	ldp	q28,q29,[sp,#64]
+	add	x8,x8,x23,lsr#32
+	add	v0.4s,v0.4s,v24.4s
+	add	w9,w9,w24
+	add	v4.4s,v4.4s,v24.4s
+	add	x10,x10,x24,lsr#32
+	add	v8.4s,v8.4s,v24.4s
+	add	w11,w11,w25
+	add	v12.4s,v12.4s,v24.4s
+	add	x12,x12,x25,lsr#32
+	add	v16.4s,v16.4s,v24.4s
+	add	w13,w13,w26
+	add	v20.4s,v20.4s,v24.4s
+	add	x14,x14,x26,lsr#32
+	add	v2.4s,v2.4s,v26.4s
+	add	w15,w15,w27
+	add	v6.4s,v6.4s,v26.4s
+	add	x16,x16,x27,lsr#32
+	add	v10.4s,v10.4s,v26.4s
+	add	w17,w17,w28
+	add	v14.4s,v14.4s,v26.4s
+	add	x19,x19,x28,lsr#32
+	add	v18.4s,v18.4s,v26.4s
+	add	w20,w20,w30
+	add	v22.4s,v22.4s,v26.4s
+	add	x21,x21,x30,lsr#32
+	add	v19.4s,v19.4s,v31.4s			// +4
+	add	x5,x5,x6,lsl#32	// pack
+	add	v23.4s,v23.4s,v31.4s			// +4
+	add	x7,x7,x8,lsl#32
+	add	v3.4s,v3.4s,v27.4s
+	ldp	x6,x8,[x1,#0]		// load input
+	add	v7.4s,v7.4s,v28.4s
+	add	x9,x9,x10,lsl#32
+	add	v11.4s,v11.4s,v29.4s
+	add	x11,x11,x12,lsl#32
+	add	v15.4s,v15.4s,v30.4s
+	ldp	x10,x12,[x1,#16]
+	add	v19.4s,v19.4s,v27.4s
+	add	x13,x13,x14,lsl#32
+	add	v23.4s,v23.4s,v28.4s
+	add	x15,x15,x16,lsl#32
+	add	v1.4s,v1.4s,v25.4s
+	ldp	x14,x16,[x1,#32]
+	add	v5.4s,v5.4s,v25.4s
+	add	x17,x17,x19,lsl#32
+	add	v9.4s,v9.4s,v25.4s
+	add	x20,x20,x21,lsl#32
+	add	v13.4s,v13.4s,v25.4s
+	ldp	x19,x21,[x1,#48]
+	add	v17.4s,v17.4s,v25.4s
+	add	x1,x1,#64
+	add	v21.4s,v21.4s,v25.4s
+
+#ifdef	__ARMEB__
+	rev	x5,x5
+	rev	x7,x7
+	rev	x9,x9
+	rev	x11,x11
+	rev	x13,x13
+	rev	x15,x15
+	rev	x17,x17
+	rev	x20,x20
+#endif
+	ld1	{v24.16b,v25.16b,v26.16b,v27.16b},[x1],#64
+	eor	x5,x5,x6
+	eor	x7,x7,x8
+	eor	x9,x9,x10
+	eor	x11,x11,x12
+	eor	x13,x13,x14
+	eor	v0.16b,v0.16b,v24.16b
+	eor	x15,x15,x16
+	eor	v1.16b,v1.16b,v25.16b
+	eor	x17,x17,x19
+	eor	v2.16b,v2.16b,v26.16b
+	eor	x20,x20,x21
+	eor	v3.16b,v3.16b,v27.16b
+	ld1	{v24.16b,v25.16b,v26.16b,v27.16b},[x1],#64
+
+	stp	x5,x7,[x0,#0]		// store output
+	add	x28,x28,#7			// increment counter
+	stp	x9,x11,[x0,#16]
+	stp	x13,x15,[x0,#32]
+	stp	x17,x20,[x0,#48]
+	add	x0,x0,#64
+	st1	{v0.16b,v1.16b,v2.16b,v3.16b},[x0],#64
+
+	ld1	{v0.16b,v1.16b,v2.16b,v3.16b},[x1],#64
+	eor	v4.16b,v4.16b,v24.16b
+	eor	v5.16b,v5.16b,v25.16b
+	eor	v6.16b,v6.16b,v26.16b
+	eor	v7.16b,v7.16b,v27.16b
+	st1	{v4.16b,v5.16b,v6.16b,v7.16b},[x0],#64
+
+	ld1	{v4.16b,v5.16b,v6.16b,v7.16b},[x1],#64
+	eor	v8.16b,v8.16b,v0.16b
+	ldp	q24,q25,[sp,#0]
+	eor	v9.16b,v9.16b,v1.16b
+	ldp	q26,q27,[sp,#32]
+	eor	v10.16b,v10.16b,v2.16b
+	eor	v11.16b,v11.16b,v3.16b
+	st1	{v8.16b,v9.16b,v10.16b,v11.16b},[x0],#64
+
+	ld1	{v8.16b,v9.16b,v10.16b,v11.16b},[x1],#64
+	eor	v12.16b,v12.16b,v4.16b
+	eor	v13.16b,v13.16b,v5.16b
+	eor	v14.16b,v14.16b,v6.16b
+	eor	v15.16b,v15.16b,v7.16b
+	st1	{v12.16b,v13.16b,v14.16b,v15.16b},[x0],#64
+
+	ld1	{v12.16b,v13.16b,v14.16b,v15.16b},[x1],#64
+	eor	v16.16b,v16.16b,v8.16b
+	eor	v17.16b,v17.16b,v9.16b
+	eor	v18.16b,v18.16b,v10.16b
+	eor	v19.16b,v19.16b,v11.16b
+	st1	{v16.16b,v17.16b,v18.16b,v19.16b},[x0],#64
+
+	shl	v0.4s,v31.4s,#1			// 4 -> 8
+	eor	v20.16b,v20.16b,v12.16b
+	eor	v21.16b,v21.16b,v13.16b
+	eor	v22.16b,v22.16b,v14.16b
+	eor	v23.16b,v23.16b,v15.16b
+	st1	{v20.16b,v21.16b,v22.16b,v23.16b},[x0],#64
+
+	add	v27.4s,v27.4s,v0.4s			// += 8
+	add	v28.4s,v28.4s,v0.4s
+	add	v29.4s,v29.4s,v0.4s
+	add	v30.4s,v30.4s,v0.4s
+
+	b.hs	.Loop_outer_512_neon
+
+	adds	x2,x2,#512
+	ushr	v0.4s,v31.4s,#2			// 4 -> 1
+
+	ldp	d8,d9,[sp,#128+0]		// meet ABI requirements
+	ldp	d10,d11,[sp,#128+16]
+	ldp	d12,d13,[sp,#128+32]
+	ldp	d14,d15,[sp,#128+48]
+
+	stp	q24,q31,[sp,#0]		// wipe off-load area
+	stp	q24,q31,[sp,#32]
+	stp	q24,q31,[sp,#64]
+
+	b.eq	.Ldone_512_neon
+
+	cmp	x2,#192
+	sub	v27.4s,v27.4s,v0.4s			// -= 1
+	sub	v28.4s,v28.4s,v0.4s
+	sub	v29.4s,v29.4s,v0.4s
+	add	sp,sp,#128
+	b.hs	.Loop_outer_neon
+
+	eor	v25.16b,v25.16b,v25.16b
+	eor	v26.16b,v26.16b,v26.16b
+	eor	v27.16b,v27.16b,v27.16b
+	eor	v28.16b,v28.16b,v28.16b
+	eor	v29.16b,v29.16b,v29.16b
+	eor	v30.16b,v30.16b,v30.16b
+	b	.Loop_outer
+
+.Ldone_512_neon:
+	ldp	x19,x20,[x29,#16]
+	add	sp,sp,#128+64
+	ldp	x21,x22,[x29,#32]
+	ldp	x23,x24,[x29,#48]
+	ldp	x25,x26,[x29,#64]
+	ldp	x27,x28,[x29,#80]
+	ldp	x29,x30,[sp],#96
+	ret
+.size	ChaCha20_512_neon,.-ChaCha20_512_neon
+#endif

third_party/boringssl/linux-aarch64/crypto/modes/ghashv8-armx64.S

diff --git a/third_party/boringssl/linux-aarch64/crypto/modes/ghashv8-armx64.S b/third_party/boringssl/linux-aarch64/crypto/modes/ghashv8-armx64.S
index a0a9b680..f39f3ba 100644
--- a/third_party/boringssl/linux-aarch64/crypto/modes/ghashv8-armx64.S
+++ b/third_party/boringssl/linux-aarch64/crypto/modes/ghashv8-armx64.S
@@ -1,11 +1,12 @@
 #if defined(__aarch64__)
-#include "arm_arch.h"
+#include <openssl/arm_arch.h>
 
 .text
 #if !defined(__clang__)
 .arch	armv8-a+crypto
 #endif
 .globl	gcm_init_v8
+.hidden	gcm_init_v8
 .type	gcm_init_v8,%function
 .align	4
 gcm_init_v8:
@@ -56,6 +57,7 @@
 	ret
 .size	gcm_init_v8,.-gcm_init_v8
 .globl	gcm_gmult_v8
+.hidden	gcm_gmult_v8
 .type	gcm_gmult_v8,%function
 .align	4
 gcm_gmult_v8:
@@ -68,10 +70,10 @@
 #endif
 	ext	v3.16b,v17.16b,v17.16b,#8
 
-	pmull	v0.1q,v20.1d,v3.1d		//H.lo·Xi.lo
+	pmull	v0.1q,v20.1d,v3.1d		//H.lo·Xi.lo
 	eor	v17.16b,v17.16b,v3.16b		//Karatsuba pre-processing
-	pmull2	v2.1q,v20.2d,v3.2d		//H.hi·Xi.hi
-	pmull	v1.1q,v21.1d,v17.1d		//(H.lo+H.hi)·(Xi.lo+Xi.hi)
+	pmull2	v2.1q,v20.2d,v3.2d		//H.hi·Xi.hi
+	pmull	v1.1q,v21.1d,v17.1d		//(H.lo+H.hi)·(Xi.lo+Xi.hi)
 
 	ext	v17.16b,v0.16b,v2.16b,#8		//Karatsuba post-processing
 	eor	v18.16b,v0.16b,v2.16b
@@ -97,6 +99,7 @@
 	ret
 .size	gcm_gmult_v8,.-gcm_gmult_v8
 .globl	gcm_ghash_v8
+.hidden	gcm_ghash_v8
 .type	gcm_ghash_v8,%function
 .align	4
 gcm_ghash_v8:
@@ -135,7 +138,7 @@
 #endif
 	ext	v7.16b,v17.16b,v17.16b,#8
 	eor	v3.16b,v3.16b,v0.16b		//I[i]^=Xi
-	pmull	v4.1q,v20.1d,v7.1d		//H·Ii+1
+	pmull	v4.1q,v20.1d,v7.1d		//H·Ii+1
 	eor	v17.16b,v17.16b,v7.16b		//Karatsuba pre-processing
 	pmull2	v6.1q,v20.2d,v7.2d
 	b	.Loop_mod2x_v8
@@ -144,14 +147,14 @@
 .Loop_mod2x_v8:
 	ext	v18.16b,v3.16b,v3.16b,#8
 	subs	x3,x3,#32		//is there more data?
-	pmull	v0.1q,v22.1d,v3.1d		//H^2.lo·Xi.lo
+	pmull	v0.1q,v22.1d,v3.1d		//H^2.lo·Xi.lo
 	csel	x12,xzr,x12,lo			//is it time to zero x12?
 
 	pmull	v5.1q,v21.1d,v17.1d
 	eor	v18.16b,v18.16b,v3.16b		//Karatsuba pre-processing
-	pmull2	v2.1q,v22.2d,v3.2d		//H^2.hi·Xi.hi
+	pmull2	v2.1q,v22.2d,v3.2d		//H^2.hi·Xi.hi
 	eor	v0.16b,v0.16b,v4.16b		//accumulate
-	pmull2	v1.1q,v21.2d,v18.2d		//(H^2.lo+H^2.hi)·(Xi.lo+Xi.hi)
+	pmull2	v1.1q,v21.2d,v18.2d		//(H^2.lo+H^2.hi)·(Xi.lo+Xi.hi)
 	ld1	{v16.2d},[x2],x12	//load [rotated] I[i+2]
 
 	eor	v2.16b,v2.16b,v6.16b
@@ -176,7 +179,7 @@
 	ext	v7.16b,v17.16b,v17.16b,#8
 	ext	v3.16b,v16.16b,v16.16b,#8
 	eor	v0.16b,v1.16b,v18.16b
-	pmull	v4.1q,v20.1d,v7.1d		//H·Ii+1
+	pmull	v4.1q,v20.1d,v7.1d		//H·Ii+1
 	eor	v3.16b,v3.16b,v2.16b		//accumulate v3.16b early
 
 	ext	v18.16b,v0.16b,v0.16b,#8		//2nd phase of reduction
@@ -197,10 +200,10 @@
 	eor	v3.16b,v3.16b,v0.16b		//inp^=Xi
 	eor	v17.16b,v16.16b,v18.16b		//v17.16b is rotated inp^Xi
 
-	pmull	v0.1q,v20.1d,v3.1d		//H.lo·Xi.lo
+	pmull	v0.1q,v20.1d,v3.1d		//H.lo·Xi.lo
 	eor	v17.16b,v17.16b,v3.16b		//Karatsuba pre-processing
-	pmull2	v2.1q,v20.2d,v3.2d		//H.hi·Xi.hi
-	pmull	v1.1q,v21.1d,v17.1d		//(H.lo+H.hi)·(Xi.lo+Xi.hi)
+	pmull2	v2.1q,v20.2d,v3.2d		//H.hi·Xi.hi
+	pmull	v1.1q,v21.1d,v17.1d		//(H.lo+H.hi)·(Xi.lo+Xi.hi)
 
 	ext	v17.16b,v0.16b,v2.16b,#8		//Karatsuba post-processing
 	eor	v18.16b,v0.16b,v2.16b
@@ -229,4 +232,4 @@
 .byte	71,72,65,83,72,32,102,111,114,32,65,82,77,118,56,44,32,67,82,89,80,84,79,71,65,77,83,32,98,121,32,60,97,112,112,114,111,64,111,112,101,110,115,115,108,46,111,114,103,62,0
 .align	2
 .align	2
-#endif
\ No newline at end of file
+#endif

third_party/boringssl/linux-aarch64/crypto/sha/sha1-armv8.S

diff --git a/third_party/boringssl/linux-aarch64/crypto/sha/sha1-armv8.S b/third_party/boringssl/linux-aarch64/crypto/sha/sha1-armv8.S
index 487a497..cfb4aa02 100644
--- a/third_party/boringssl/linux-aarch64/crypto/sha/sha1-armv8.S
+++ b/third_party/boringssl/linux-aarch64/crypto/sha/sha1-armv8.S
@@ -1,10 +1,11 @@
 #if defined(__aarch64__)
-#include "arm_arch.h"
+#include <openssl/arm_arch.h>
 
 .text
 
 
 .globl	sha1_block_data_order
+.hidden	sha1_block_data_order
 .type	sha1_block_data_order,%function
 .align	6
 sha1_block_data_order:
@@ -1212,4 +1213,4 @@
 .align	2
 .align	2
 .comm	OPENSSL_armcap_P,4,4
-#endif
\ No newline at end of file
+#endif

third_party/boringssl/linux-aarch64/crypto/sha/sha256-armv8.S

diff --git a/third_party/boringssl/linux-aarch64/crypto/sha/sha256-armv8.S b/third_party/boringssl/linux-aarch64/crypto/sha/sha256-armv8.S
index 4834553..bfc552c 100644
--- a/third_party/boringssl/linux-aarch64/crypto/sha/sha256-armv8.S
+++ b/third_party/boringssl/linux-aarch64/crypto/sha/sha256-armv8.S
@@ -1,10 +1,11 @@
 #if defined(__aarch64__)
-#include "arm_arch.h"
+#include <openssl/arm_arch.h>
 
 .text
 
 
 .globl	sha256_block_data_order
+.hidden	sha256_block_data_order
 .type	sha256_block_data_order,%function
 .align	6
 sha256_block_data_order:
@@ -1142,4 +1143,4 @@
 	ret
 .size	sha256_block_armv8,.-sha256_block_armv8
 .comm	OPENSSL_armcap_P,4,4
-#endif
\ No newline at end of file
+#endif

third_party/boringssl/linux-aarch64/crypto/sha/sha512-armv8.S

diff --git a/third_party/boringssl/linux-aarch64/crypto/sha/sha512-armv8.S b/third_party/boringssl/linux-aarch64/crypto/sha/sha512-armv8.S
index 654c473..4645722 100644
--- a/third_party/boringssl/linux-aarch64/crypto/sha/sha512-armv8.S
+++ b/third_party/boringssl/linux-aarch64/crypto/sha/sha512-armv8.S
@@ -1,10 +1,11 @@
 #if defined(__aarch64__)
-#include "arm_arch.h"
+#include <openssl/arm_arch.h>
 
 .text
 
 
 .globl	sha512_block_data_order
+.hidden	sha512_block_data_order
 .type	sha512_block_data_order,%function
 .align	6
 sha512_block_data_order:
@@ -1022,4 +1023,4 @@
 .align	2
 .align	2
 .comm	OPENSSL_armcap_P,4,4
-#endif
\ No newline at end of file
+#endif

third_party/boringssl/linux-arm/crypto/aes/aes-armv4.S

diff --git a/third_party/boringssl/linux-arm/crypto/aes/aes-armv4.S b/third_party/boringssl/linux-arm/crypto/aes/aes-armv4.S
index cb94841..bc11e3f 100644
--- a/third_party/boringssl/linux-arm/crypto/aes/aes-armv4.S
+++ b/third_party/boringssl/linux-arm/crypto/aes/aes-armv4.S
@@ -34,7 +34,7 @@
 
 #if defined(__arm__)
 #ifndef __KERNEL__
-# include "arm_arch.h"
+# include <openssl/arm_arch.h>
 #else
 # define __ARM_ARCH__ __LINUX_ARM_ARCH__
 #endif
@@ -1197,4 +1197,4 @@
 .align	2
 
 #endif
-#endif
\ No newline at end of file
+#endif

third_party/boringssl/linux-arm/crypto/aes/aesv8-armx32.S

diff --git a/third_party/boringssl/linux-arm/crypto/aes/aesv8-armx32.S b/third_party/boringssl/linux-arm/crypto/aes/aesv8-armx32.S
index 6f0ee7d..95a2ea4 100644
--- a/third_party/boringssl/linux-arm/crypto/aes/aesv8-armx32.S
+++ b/third_party/boringssl/linux-arm/crypto/aes/aesv8-armx32.S
@@ -1,5 +1,5 @@
 #if defined(__arm__)
-#include "arm_arch.h"
+#include <openssl/arm_arch.h>
 
 #if __ARM_MAX_ARCH__>=7
 .text
@@ -13,6 +13,7 @@
 .long	0x1b,0x1b,0x1b,0x1b
 
 .globl	aes_v8_set_encrypt_key
+.hidden	aes_v8_set_encrypt_key
 .type	aes_v8_set_encrypt_key,%function
 .align	5
 aes_v8_set_encrypt_key:
@@ -183,6 +184,7 @@
 .size	aes_v8_set_encrypt_key,.-aes_v8_set_encrypt_key
 
 .globl	aes_v8_set_decrypt_key
+.hidden	aes_v8_set_decrypt_key
 .type	aes_v8_set_decrypt_key,%function
 .align	5
 aes_v8_set_decrypt_key:
@@ -220,6 +222,7 @@
 	ldmia	sp!,{r4,pc}
 .size	aes_v8_set_decrypt_key,.-aes_v8_set_decrypt_key
 .globl	aes_v8_encrypt
+.hidden	aes_v8_encrypt
 .type	aes_v8_encrypt,%function
 .align	5
 aes_v8_encrypt:
@@ -249,6 +252,7 @@
 	bx	lr
 .size	aes_v8_encrypt,.-aes_v8_encrypt
 .globl	aes_v8_decrypt
+.hidden	aes_v8_decrypt
 .type	aes_v8_decrypt,%function
 .align	5
 aes_v8_decrypt:
@@ -278,6 +282,7 @@
 	bx	lr
 .size	aes_v8_decrypt,.-aes_v8_decrypt
 .globl	aes_v8_cbc_encrypt
+.hidden	aes_v8_cbc_encrypt
 .type	aes_v8_cbc_encrypt,%function
 .align	5
 aes_v8_cbc_encrypt:
@@ -570,6 +575,7 @@
 	ldmia	sp!,{r4,r5,r6,r7,r8,pc}
 .size	aes_v8_cbc_encrypt,.-aes_v8_cbc_encrypt
 .globl	aes_v8_ctr32_encrypt_blocks
+.hidden	aes_v8_ctr32_encrypt_blocks
 .type	aes_v8_ctr32_encrypt_blocks,%function
 .align	5
 aes_v8_ctr32_encrypt_blocks:
@@ -753,4 +759,4 @@
 	ldmia	sp!,{r4,r5,r6,r7,r8,r9,r10,pc}
 .size	aes_v8_ctr32_encrypt_blocks,.-aes_v8_ctr32_encrypt_blocks
 #endif
-#endif
\ No newline at end of file
+#endif

third_party/boringssl/linux-arm/crypto/aes/bsaes-armv7.S

diff --git a/third_party/boringssl/linux-arm/crypto/aes/bsaes-armv7.S b/third_party/boringssl/linux-arm/crypto/aes/bsaes-armv7.S
index dd84f35..abb414d 100644
--- a/third_party/boringssl/linux-arm/crypto/aes/bsaes-armv7.S
+++ b/third_party/boringssl/linux-arm/crypto/aes/bsaes-armv7.S
@@ -47,9 +47,8 @@
 @
 @					<ard.biesheuvel@linaro.org>
 
-#if defined(__arm__)
 #ifndef __KERNEL__
-# include "arm_arch.h"
+# include <openssl/arm_arch.h>
 
 # define VFP_ABI_PUSH	vstmdb	sp!,{d8-d15}
 # define VFP_ABI_POP	vldmia	sp!,{d8-d15}
@@ -2576,4 +2575,3 @@
 .size	bsaes_xts_decrypt,.-bsaes_xts_decrypt
 #endif
 #endif
-#endif
\ No newline at end of file

third_party/boringssl/linux-arm/crypto/bn/armv4-mont.S

diff --git a/third_party/boringssl/linux-arm/crypto/bn/armv4-mont.S b/third_party/boringssl/linux-arm/crypto/bn/armv4-mont.S
index 68dfb2c..e59599f 100644
--- a/third_party/boringssl/linux-arm/crypto/bn/armv4-mont.S
+++ b/third_party/boringssl/linux-arm/crypto/bn/armv4-mont.S
@@ -1,5 +1,5 @@
 #if defined(__arm__)
-#include "arm_arch.h"
+#include <openssl/arm_arch.h>
 
 .text
 .code	32
@@ -28,7 +28,7 @@
 #ifdef	__APPLE__
 	ldr	r0,[r0]
 #endif
-	tst	r0,#1			@ NEON available?
+	tst	r0,#ARMV7_NEON		@ NEON available?
 	ldmia	sp, {r0,r2}
 	beq	.Lialu
 	add	sp,sp,#8
@@ -586,4 +586,4 @@
 .comm	OPENSSL_armcap_P,4,4
 .hidden	OPENSSL_armcap_P
 #endif
-#endif
\ No newline at end of file
+#endif

third_party/boringssl/linux-arm/crypto/chacha/chacha-armv4.S

diff --git a/third_party/boringssl/linux-arm/crypto/chacha/chacha-armv4.S b/third_party/boringssl/linux-arm/crypto/chacha/chacha-armv4.S
new file mode 100644
index 0000000..19a4d2c
--- /dev/null
+++ b/third_party/boringssl/linux-arm/crypto/chacha/chacha-armv4.S
@@ -0,0 +1,1471 @@
+#if defined(__arm__)
+#include <openssl/arm_arch.h>
+
+.text
+#if defined(__thumb2__)
+.syntax	unified
+.thumb
+#else
+.code	32
+#endif
+
+#if defined(__thumb2__) || defined(__clang__)
+#define ldrhsb	ldrbhs
+#endif
+
+.align	5
+.Lsigma:
+.long	0x61707865,0x3320646e,0x79622d32,0x6b206574	@ endian-neutral
+.Lone:
+.long	1,0,0,0
+#if __ARM_MAX_ARCH__>=7
+.LOPENSSL_armcap:
+.word	OPENSSL_armcap_P-.LChaCha20_ctr32
+#else
+.word	-1
+#endif
+
+.globl	ChaCha20_ctr32
+.hidden	ChaCha20_ctr32
+.type	ChaCha20_ctr32,%function
+.align	5
+ChaCha20_ctr32:
+.LChaCha20_ctr32:
+	ldr	r12,[sp,#0]		@ pull pointer to counter and nonce
+	stmdb	sp!,{r0,r1,r2,r4-r11,lr}
+#if __ARM_ARCH__<7 && !defined(__thumb2__)
+	sub	r14,pc,#16		@ ChaCha20_ctr32
+#else
+	adr	r14,.LChaCha20_ctr32
+#endif
+	cmp	r2,#0			@ len==0?
+#ifdef	__thumb2__
+	itt	eq
+#endif
+	addeq	sp,sp,#4*3
+	beq	.Lno_data
+#if __ARM_MAX_ARCH__>=7
+	cmp	r2,#192			@ test len
+	bls	.Lshort
+	ldr	r4,[r14,#-32]
+	ldr	r4,[r14,r4]
+# ifdef	__APPLE__
+	ldr	r4,[r4]
+# endif
+	tst	r4,#ARMV7_NEON
+	bne	.LChaCha20_neon
+.Lshort:
+#endif
+	ldmia	r12,{r4,r5,r6,r7}		@ load counter and nonce
+	sub	sp,sp,#4*(16)		@ off-load area
+	sub	r14,r14,#64		@ .Lsigma
+	stmdb	sp!,{r4,r5,r6,r7}		@ copy counter and nonce
+	ldmia	r3,{r4,r5,r6,r7,r8,r9,r10,r11}		@ load key
+	ldmia	r14,{r0,r1,r2,r3}		@ load sigma
+	stmdb	sp!,{r4,r5,r6,r7,r8,r9,r10,r11}		@ copy key
+	stmdb	sp!,{r0,r1,r2,r3}		@ copy sigma
+	str	r10,[sp,#4*(16+10)]	@ off-load "rx"
+	str	r11,[sp,#4*(16+11)]	@ off-load "rx"
+	b	.Loop_outer_enter
+
+.align	4
+.Loop_outer:
+	ldmia	sp,{r0,r1,r2,r3,r4,r5,r6,r7,r8,r9}		@ load key material
+	str	r11,[sp,#4*(32+2)]	@ save len
+	str	r12,  [sp,#4*(32+1)]	@ save inp
+	str	r14,  [sp,#4*(32+0)]	@ save out
+.Loop_outer_enter:
+	ldr	r11, [sp,#4*(15)]
+	ldr	r12,[sp,#4*(12)]	@ modulo-scheduled load
+	ldr	r10, [sp,#4*(13)]
+	ldr	r14,[sp,#4*(14)]
+	str	r11, [sp,#4*(16+15)]
+	mov	r11,#10
+	b	.Loop
+
+.align	4
+.Loop:
+	subs	r11,r11,#1
+	add	r0,r0,r4
+	mov	r12,r12,ror#16
+	add	r1,r1,r5
+	mov	r10,r10,ror#16
+	eor	r12,r12,r0,ror#16
+	eor	r10,r10,r1,ror#16
+	add	r8,r8,r12
+	mov	r4,r4,ror#20
+	add	r9,r9,r10
+	mov	r5,r5,ror#20
+	eor	r4,r4,r8,ror#20
+	eor	r5,r5,r9,ror#20
+	add	r0,r0,r4
+	mov	r12,r12,ror#24
+	add	r1,r1,r5
+	mov	r10,r10,ror#24
+	eor	r12,r12,r0,ror#24
+	eor	r10,r10,r1,ror#24
+	add	r8,r8,r12
+	mov	r4,r4,ror#25
+	add	r9,r9,r10
+	mov	r5,r5,ror#25
+	str	r10,[sp,#4*(16+13)]
+	ldr	r10,[sp,#4*(16+15)]
+	eor	r4,r4,r8,ror#25
+	eor	r5,r5,r9,ror#25
+	str	r8,[sp,#4*(16+8)]
+	ldr	r8,[sp,#4*(16+10)]
+	add	r2,r2,r6
+	mov	r14,r14,ror#16
+	str	r9,[sp,#4*(16+9)]
+	ldr	r9,[sp,#4*(16+11)]
+	add	r3,r3,r7
+	mov	r10,r10,ror#16
+	eor	r14,r14,r2,ror#16
+	eor	r10,r10,r3,ror#16
+	add	r8,r8,r14
+	mov	r6,r6,ror#20
+	add	r9,r9,r10
+	mov	r7,r7,ror#20
+	eor	r6,r6,r8,ror#20
+	eor	r7,r7,r9,ror#20
+	add	r2,r2,r6
+	mov	r14,r14,ror#24
+	add	r3,r3,r7
+	mov	r10,r10,ror#24
+	eor	r14,r14,r2,ror#24
+	eor	r10,r10,r3,ror#24
+	add	r8,r8,r14
+	mov	r6,r6,ror#25
+	add	r9,r9,r10
+	mov	r7,r7,ror#25
+	eor	r6,r6,r8,ror#25
+	eor	r7,r7,r9,ror#25
+	add	r0,r0,r5
+	mov	r10,r10,ror#16
+	add	r1,r1,r6
+	mov	r12,r12,ror#16
+	eor	r10,r10,r0,ror#16
+	eor	r12,r12,r1,ror#16
+	add	r8,r8,r10
+	mov	r5,r5,ror#20
+	add	r9,r9,r12
+	mov	r6,r6,ror#20
+	eor	r5,r5,r8,ror#20
+	eor	r6,r6,r9,ror#20
+	add	r0,r0,r5
+	mov	r10,r10,ror#24
+	add	r1,r1,r6
+	mov	r12,r12,ror#24
+	eor	r10,r10,r0,ror#24
+	eor	r12,r12,r1,ror#24
+	add	r8,r8,r10
+	mov	r5,r5,ror#25
+	str	r10,[sp,#4*(16+15)]
+	ldr	r10,[sp,#4*(16+13)]
+	add	r9,r9,r12
+	mov	r6,r6,ror#25
+	eor	r5,r5,r8,ror#25
+	eor	r6,r6,r9,ror#25
+	str	r8,[sp,#4*(16+10)]
+	ldr	r8,[sp,#4*(16+8)]
+	add	r2,r2,r7
+	mov	r10,r10,ror#16
+	str	r9,[sp,#4*(16+11)]
+	ldr	r9,[sp,#4*(16+9)]
+	add	r3,r3,r4
+	mov	r14,r14,ror#16
+	eor	r10,r10,r2,ror#16
+	eor	r14,r14,r3,ror#16
+	add	r8,r8,r10
+	mov	r7,r7,ror#20
+	add	r9,r9,r14
+	mov	r4,r4,ror#20
+	eor	r7,r7,r8,ror#20
+	eor	r4,r4,r9,ror#20
+	add	r2,r2,r7
+	mov	r10,r10,ror#24
+	add	r3,r3,r4
+	mov	r14,r14,ror#24
+	eor	r10,r10,r2,ror#24
+	eor	r14,r14,r3,ror#24
+	add	r8,r8,r10
+	mov	r7,r7,ror#25
+	add	r9,r9,r14
+	mov	r4,r4,ror#25
+	eor	r7,r7,r8,ror#25
+	eor	r4,r4,r9,ror#25
+	bne	.Loop
+
+	ldr	r11,[sp,#4*(32+2)]	@ load len
+
+	str	r8, [sp,#4*(16+8)]	@ modulo-scheduled store
+	str	r9, [sp,#4*(16+9)]
+	str	r12,[sp,#4*(16+12)]
+	str	r10, [sp,#4*(16+13)]
+	str	r14,[sp,#4*(16+14)]
+
+	@ at this point we have first half of 512-bit result in
+	@ rx and second half at sp+4*(16+8)
+
+	cmp	r11,#64		@ done yet?
+#ifdef	__thumb2__
+	itete	lo
+#endif
+	addlo	r12,sp,#4*(0)		@ shortcut or ...
+	ldrhs	r12,[sp,#4*(32+1)]	@ ... load inp
+	addlo	r14,sp,#4*(0)		@ shortcut or ...
+	ldrhs	r14,[sp,#4*(32+0)]	@ ... load out
+
+	ldr	r8,[sp,#4*(0)]	@ load key material
+	ldr	r9,[sp,#4*(1)]
+
+#if __ARM_ARCH__>=6 || !defined(__ARMEB__)
+# if __ARM_ARCH__<7
+	orr	r10,r12,r14
+	tst	r10,#3		@ are input and output aligned?
+	ldr	r10,[sp,#4*(2)]
+	bne	.Lunaligned
+	cmp	r11,#64		@ restore flags
+# else
+	ldr	r10,[sp,#4*(2)]
+# endif
+	ldr	r11,[sp,#4*(3)]
+
+	add	r0,r0,r8	@ accumulate key material
+	add	r1,r1,r9
+# ifdef	__thumb2__
+	itt	hs
+# endif
+	ldrhs	r8,[r12],#16		@ load input
+	ldrhs	r9,[r12,#-12]
+
+	add	r2,r2,r10
+	add	r3,r3,r11
+# ifdef	__thumb2__
+	itt	hs
+# endif
+	ldrhs	r10,[r12,#-8]
+	ldrhs	r11,[r12,#-4]
+# if __ARM_ARCH__>=6 && defined(__ARMEB__)
+	rev	r0,r0
+	rev	r1,r1
+	rev	r2,r2
+	rev	r3,r3
+# endif
+# ifdef	__thumb2__
+	itt	hs
+# endif
+	eorhs	r0,r0,r8	@ xor with input
+	eorhs	r1,r1,r9
+	add	r8,sp,#4*(4)
+	str	r0,[r14],#16		@ store output
+# ifdef	__thumb2__
+	itt	hs
+# endif
+	eorhs	r2,r2,r10
+	eorhs	r3,r3,r11
+	ldmia	r8,{r8,r9,r10,r11}	@ load key material
+	str	r1,[r14,#-12]
+	str	r2,[r14,#-8]
+	str	r3,[r14,#-4]
+
+	add	r4,r4,r8	@ accumulate key material
+	add	r5,r5,r9
+# ifdef	__thumb2__
+	itt	hs
+# endif
+	ldrhs	r8,[r12],#16		@ load input
+	ldrhs	r9,[r12,#-12]
+	add	r6,r6,r10
+	add	r7,r7,r11
+# ifdef	__thumb2__
+	itt	hs
+# endif
+	ldrhs	r10,[r12,#-8]
+	ldrhs	r11,[r12,#-4]
+# if __ARM_ARCH__>=6 && defined(__ARMEB__)
+	rev	r4,r4
+	rev	r5,r5
+	rev	r6,r6
+	rev	r7,r7
+# endif
+# ifdef	__thumb2__
+	itt	hs
+# endif
+	eorhs	r4,r4,r8
+	eorhs	r5,r5,r9
+	add	r8,sp,#4*(8)
+	str	r4,[r14],#16		@ store output
+# ifdef	__thumb2__
+	itt	hs
+# endif
+	eorhs	r6,r6,r10
+	eorhs	r7,r7,r11
+	str	r5,[r14,#-12]
+	ldmia	r8,{r8,r9,r10,r11}	@ load key material
+	str	r6,[r14,#-8]
+	add	r0,sp,#4*(16+8)
+	str	r7,[r14,#-4]
+
+	ldmia	r0,{r0,r1,r2,r3,r4,r5,r6,r7}	@ load second half
+
+	add	r0,r0,r8	@ accumulate key material
+	add	r1,r1,r9
+# ifdef	__thumb2__
+	itt	hs
+# endif
+	ldrhs	r8,[r12],#16		@ load input
+	ldrhs	r9,[r12,#-12]
+# ifdef	__thumb2__
+	itt	hi
+# endif
+	strhi	r10,[sp,#4*(16+10)]	@ copy "rx" while at it
+	strhi	r11,[sp,#4*(16+11)]	@ copy "rx" while at it
+	add	r2,r2,r10
+	add	r3,r3,r11
+# ifdef	__thumb2__
+	itt	hs
+# endif
+	ldrhs	r10,[r12,#-8]
+	ldrhs	r11,[r12,#-4]
+# if __ARM_ARCH__>=6 && defined(__ARMEB__)
+	rev	r0,r0
+	rev	r1,r1
+	rev	r2,r2
+	rev	r3,r3
+# endif
+# ifdef	__thumb2__
+	itt	hs
+# endif
+	eorhs	r0,r0,r8
+	eorhs	r1,r1,r9
+	add	r8,sp,#4*(12)
+	str	r0,[r14],#16		@ store output
+# ifdef	__thumb2__
+	itt	hs
+# endif
+	eorhs	r2,r2,r10
+	eorhs	r3,r3,r11
+	str	r1,[r14,#-12]
+	ldmia	r8,{r8,r9,r10,r11}	@ load key material
+	str	r2,[r14,#-8]
+	str	r3,[r14,#-4]
+
+	add	r4,r4,r8	@ accumulate key material
+	add	r5,r5,r9
+# ifdef	__thumb2__
+	itt	hi
+# endif
+	addhi	r8,r8,#1		@ next counter value
+	strhi	r8,[sp,#4*(12)]	@ save next counter value
+# ifdef	__thumb2__
+	itt	hs
+# endif
+	ldrhs	r8,[r12],#16		@ load input
+	ldrhs	r9,[r12,#-12]
+	add	r6,r6,r10
+	add	r7,r7,r11
+# ifdef	__thumb2__
+	itt	hs
+# endif
+	ldrhs	r10,[r12,#-8]
+	ldrhs	r11,[r12,#-4]
+# if __ARM_ARCH__>=6 && defined(__ARMEB__)
+	rev	r4,r4
+	rev	r5,r5
+	rev	r6,r6
+	rev	r7,r7
+# endif
+# ifdef	__thumb2__
+	itt	hs
+# endif
+	eorhs	r4,r4,r8
+	eorhs	r5,r5,r9
+# ifdef	__thumb2__
+	it	ne
+# endif
+	ldrne	r8,[sp,#4*(32+2)]	@ re-load len
+# ifdef	__thumb2__
+	itt	hs
+# endif
+	eorhs	r6,r6,r10
+	eorhs	r7,r7,r11
+	str	r4,[r14],#16		@ store output
+	str	r5,[r14,#-12]
+# ifdef	__thumb2__
+	it	hs
+# endif
+	subhs	r11,r8,#64		@ len-=64
+	str	r6,[r14,#-8]
+	str	r7,[r14,#-4]
+	bhi	.Loop_outer
+
+	beq	.Ldone
+# if __ARM_ARCH__<7
+	b	.Ltail
+
+.align	4
+.Lunaligned:@ unaligned endian-neutral path
+	cmp	r11,#64		@ restore flags
+# endif
+#endif
+#if __ARM_ARCH__<7
+	ldr	r11,[sp,#4*(3)]
+	add	r0,r0,r8		@ accumulate key material
+	add	r1,r1,r9
+	add	r2,r2,r10
+# ifdef	__thumb2__
+	itete	lo
+# endif
+	eorlo	r8,r8,r8		@ zero or ...
+	ldrhsb	r8,[r12],#16			@ ... load input
+	eorlo	r9,r9,r9
+	ldrhsb	r9,[r12,#-12]
+
+	add	r3,r3,r11
+# ifdef	__thumb2__
+	itete	lo
+# endif
+	eorlo	r10,r10,r10
+	ldrhsb	r10,[r12,#-8]
+	eorlo	r11,r11,r11
+	ldrhsb	r11,[r12,#-4]
+
+	eor	r0,r8,r0		@ xor with input (or zero)
+	eor	r1,r9,r1
+# ifdef	__thumb2__
+	itt	hs
+# endif
+	ldrhsb	r8,[r12,#-15]		@ load more input
+	ldrhsb	r9,[r12,#-11]
+	eor	r2,r10,r2
+	strb	r0,[r14],#16		@ store output
+	eor	r3,r11,r3
+# ifdef	__thumb2__
+	itt	hs
+# endif
+	ldrhsb	r10,[r12,#-7]
+	ldrhsb	r11,[r12,#-3]
+	strb	r1,[r14,#-12]
+	eor	r0,r8,r0,lsr#8
+	strb	r2,[r14,#-8]
+	eor	r1,r9,r1,lsr#8
+# ifdef	__thumb2__
+	itt	hs
+# endif
+	ldrhsb	r8,[r12,#-14]		@ load more input
+	ldrhsb	r9,[r12,#-10]
+	strb	r3,[r14,#-4]
+	eor	r2,r10,r2,lsr#8
+	strb	r0,[r14,#-15]
+	eor	r3,r11,r3,lsr#8
+# ifdef	__thumb2__
+	itt	hs
+# endif
+	ldrhsb	r10,[r12,#-6]
+	ldrhsb	r11,[r12,#-2]
+	strb	r1,[r14,#-11]
+	eor	r0,r8,r0,lsr#8
+	strb	r2,[r14,#-7]
+	eor	r1,r9,r1,lsr#8
+# ifdef	__thumb2__
+	itt	hs
+# endif
+	ldrhsb	r8,[r12,#-13]		@ load more input
+	ldrhsb	r9,[r12,#-9]
+	strb	r3,[r14,#-3]
+	eor	r2,r10,r2,lsr#8
+	strb	r0,[r14,#-14]
+	eor	r3,r11,r3,lsr#8
+# ifdef	__thumb2__
+	itt	hs
+# endif
+	ldrhsb	r10,[r12,#-5]
+	ldrhsb	r11,[r12,#-1]
+	strb	r1,[r14,#-10]
+	strb	r2,[r14,#-6]
+	eor	r0,r8,r0,lsr#8
+	strb	r3,[r14,#-2]
+	eor	r1,r9,r1,lsr#8
+	strb	r0,[r14,#-13]
+	eor	r2,r10,r2,lsr#8
+	strb	r1,[r14,#-9]
+	eor	r3,r11,r3,lsr#8
+	strb	r2,[r14,#-5]
+	strb	r3,[r14,#-1]
+	add	r8,sp,#4*(4+0)
+	ldmia	r8,{r8,r9,r10,r11}		@ load key material
+	add	r0,sp,#4*(16+8)
+	add	r4,r4,r8		@ accumulate key material
+	add	r5,r5,r9
+	add	r6,r6,r10
+# ifdef	__thumb2__
+	itete	lo
+# endif
+	eorlo	r8,r8,r8		@ zero or ...
+	ldrhsb	r8,[r12],#16			@ ... load input
+	eorlo	r9,r9,r9
+	ldrhsb	r9,[r12,#-12]
+
+	add	r7,r7,r11
+# ifdef	__thumb2__
+	itete	lo
+# endif
+	eorlo	r10,r10,r10
+	ldrhsb	r10,[r12,#-8]
+	eorlo	r11,r11,r11
+	ldrhsb	r11,[r12,#-4]
+
+	eor	r4,r8,r4		@ xor with input (or zero)
+	eor	r5,r9,r5
+# ifdef	__thumb2__
+	itt	hs
+# endif
+	ldrhsb	r8,[r12,#-15]		@ load more input
+	ldrhsb	r9,[r12,#-11]
+	eor	r6,r10,r6
+	strb	r4,[r14],#16		@ store output
+	eor	r7,r11,r7
+# ifdef	__thumb2__
+	itt	hs
+# endif
+	ldrhsb	r10,[r12,#-7]
+	ldrhsb	r11,[r12,#-3]
+	strb	r5,[r14,#-12]
+	eor	r4,r8,r4,lsr#8
+	strb	r6,[r14,#-8]
+	eor	r5,r9,r5,lsr#8
+# ifdef	__thumb2__
+	itt	hs
+# endif
+	ldrhsb	r8,[r12,#-14]		@ load more input
+	ldrhsb	r9,[r12,#-10]
+	strb	r7,[r14,#-4]
+	eor	r6,r10,r6,lsr#8
+	strb	r4,[r14,#-15]
+	eor	r7,r11,r7,lsr#8
+# ifdef	__thumb2__
+	itt	hs
+# endif
+	ldrhsb	r10,[r12,#-6]
+	ldrhsb	r11,[r12,#-2]
+	strb	r5,[r14,#-11]
+	eor	r4,r8,r4,lsr#8
+	strb	r6,[r14,#-7]
+	eor	r5,r9,r5,lsr#8
+# ifdef	__thumb2__
+	itt	hs
+# endif
+	ldrhsb	r8,[r12,#-13]		@ load more input
+	ldrhsb	r9,[r12,#-9]
+	strb	r7,[r14,#-3]
+	eor	r6,r10,r6,lsr#8
+	strb	r4,[r14,#-14]
+	eor	r7,r11,r7,lsr#8
+# ifdef	__thumb2__
+	itt	hs
+# endif
+	ldrhsb	r10,[r12,#-5]
+	ldrhsb	r11,[r12,#-1]
+	strb	r5,[r14,#-10]
+	strb	r6,[r14,#-6]
+	eor	r4,r8,r4,lsr#8
+	strb	r7,[r14,#-2]
+	eor	r5,r9,r5,lsr#8
+	strb	r4,[r14,#-13]
+	eor	r6,r10,r6,lsr#8
+	strb	r5,[r14,#-9]
+	eor	r7,r11,r7,lsr#8
+	strb	r6,[r14,#-5]
+	strb	r7,[r14,#-1]
+	add	r8,sp,#4*(4+4)
+	ldmia	r8,{r8,r9,r10,r11}		@ load key material
+	ldmia	r0,{r0,r1,r2,r3,r4,r5,r6,r7}		@ load second half
+# ifdef	__thumb2__
+	itt	hi
+# endif
+	strhi	r10,[sp,#4*(16+10)]		@ copy "rx"
+	strhi	r11,[sp,#4*(16+11)]		@ copy "rx"
+	add	r0,r0,r8		@ accumulate key material
+	add	r1,r1,r9
+	add	r2,r2,r10
+# ifdef	__thumb2__
+	itete	lo
+# endif
+	eorlo	r8,r8,r8		@ zero or ...
+	ldrhsb	r8,[r12],#16			@ ... load input
+	eorlo	r9,r9,r9
+	ldrhsb	r9,[r12,#-12]
+
+	add	r3,r3,r11
+# ifdef	__thumb2__
+	itete	lo
+# endif
+	eorlo	r10,r10,r10
+	ldrhsb	r10,[r12,#-8]
+	eorlo	r11,r11,r11
+	ldrhsb	r11,[r12,#-4]
+
+	eor	r0,r8,r0		@ xor with input (or zero)
+	eor	r1,r9,r1
+# ifdef	__thumb2__
+	itt	hs
+# endif
+	ldrhsb	r8,[r12,#-15]		@ load more input
+	ldrhsb	r9,[r12,#-11]
+	eor	r2,r10,r2
+	strb	r0,[r14],#16		@ store output
+	eor	r3,r11,r3
+# ifdef	__thumb2__
+	itt	hs
+# endif
+	ldrhsb	r10,[r12,#-7]
+	ldrhsb	r11,[r12,#-3]
+	strb	r1,[r14,#-12]
+	eor	r0,r8,r0,lsr#8
+	strb	r2,[r14,#-8]
+	eor	r1,r9,r1,lsr#8
+# ifdef	__thumb2__
+	itt	hs
+# endif
+	ldrhsb	r8,[r12,#-14]		@ load more input
+	ldrhsb	r9,[r12,#-10]
+	strb	r3,[r14,#-4]
+	eor	r2,r10,r2,lsr#8
+	strb	r0,[r14,#-15]
+	eor	r3,r11,r3,lsr#8
+# ifdef	__thumb2__
+	itt	hs
+# endif
+	ldrhsb	r10,[r12,#-6]
+	ldrhsb	r11,[r12,#-2]
+	strb	r1,[r14,#-11]
+	eor	r0,r8,r0,lsr#8
+	strb	r2,[r14,#-7]
+	eor	r1,r9,r1,lsr#8
+# ifdef	__thumb2__
+	itt	hs
+# endif
+	ldrhsb	r8,[r12,#-13]		@ load more input
+	ldrhsb	r9,[r12,#-9]
+	strb	r3,[r14,#-3]
+	eor	r2,r10,r2,lsr#8
+	strb	r0,[r14,#-14]
+	eor	r3,r11,r3,lsr#8
+# ifdef	__thumb2__
+	itt	hs
+# endif
+	ldrhsb	r10,[r12,#-5]
+	ldrhsb	r11,[r12,#-1]
+	strb	r1,[r14,#-10]
+	strb	r2,[r14,#-6]
+	eor	r0,r8,r0,lsr#8
+	strb	r3,[r14,#-2]
+	eor	r1,r9,r1,lsr#8
+	strb	r0,[r14,#-13]
+	eor	r2,r10,r2,lsr#8
+	strb	r1,[r14,#-9]
+	eor	r3,r11,r3,lsr#8
+	strb	r2,[r14,#-5]
+	strb	r3,[r14,#-1]
+	add	r8,sp,#4*(4+8)
+	ldmia	r8,{r8,r9,r10,r11}		@ load key material
+	add	r4,r4,r8		@ accumulate key material
+# ifdef	__thumb2__
+	itt	hi
+# endif
+	addhi	r8,r8,#1			@ next counter value
+	strhi	r8,[sp,#4*(12)]		@ save next counter value
+	add	r5,r5,r9
+	add	r6,r6,r10
+# ifdef	__thumb2__
+	itete	lo
+# endif
+	eorlo	r8,r8,r8		@ zero or ...
+	ldrhsb	r8,[r12],#16			@ ... load input
+	eorlo	r9,r9,r9
+	ldrhsb	r9,[r12,#-12]
+
+	add	r7,r7,r11
+# ifdef	__thumb2__
+	itete	lo
+# endif
+	eorlo	r10,r10,r10
+	ldrhsb	r10,[r12,#-8]
+	eorlo	r11,r11,r11
+	ldrhsb	r11,[r12,#-4]
+
+	eor	r4,r8,r4		@ xor with input (or zero)
+	eor	r5,r9,r5
+# ifdef	__thumb2__
+	itt	hs
+# endif
+	ldrhsb	r8,[r12,#-15]		@ load more input
+	ldrhsb	r9,[r12,#-11]
+	eor	r6,r10,r6
+	strb	r4,[r14],#16		@ store output
+	eor	r7,r11,r7
+# ifdef	__thumb2__
+	itt	hs
+# endif
+	ldrhsb	r10,[r12,#-7]
+	ldrhsb	r11,[r12,#-3]
+	strb	r5,[r14,#-12]
+	eor	r4,r8,r4,lsr#8
+	strb	r6,[r14,#-8]
+	eor	r5,r9,r5,lsr#8
+# ifdef	__thumb2__
+	itt	hs
+# endif
+	ldrhsb	r8,[r12,#-14]		@ load more input
+	ldrhsb	r9,[r12,#-10]
+	strb	r7,[r14,#-4]
+	eor	r6,r10,r6,lsr#8
+	strb	r4,[r14,#-15]
+	eor	r7,r11,r7,lsr#8
+# ifdef	__thumb2__
+	itt	hs
+# endif
+	ldrhsb	r10,[r12,#-6]
+	ldrhsb	r11,[r12,#-2]
+	strb	r5,[r14,#-11]
+	eor	r4,r8,r4,lsr#8
+	strb	r6,[r14,#-7]
+	eor	r5,r9,r5,lsr#8
+# ifdef	__thumb2__
+	itt	hs
+# endif
+	ldrhsb	r8,[r12,#-13]		@ load more input
+	ldrhsb	r9,[r12,#-9]
+	strb	r7,[r14,#-3]
+	eor	r6,r10,r6,lsr#8
+	strb	r4,[r14,#-14]
+	eor	r7,r11,r7,lsr#8
+# ifdef	__thumb2__
+	itt	hs
+# endif
+	ldrhsb	r10,[r12,#-5]
+	ldrhsb	r11,[r12,#-1]
+	strb	r5,[r14,#-10]
+	strb	r6,[r14,#-6]
+	eor	r4,r8,r4,lsr#8
+	strb	r7,[r14,#-2]
+	eor	r5,r9,r5,lsr#8
+	strb	r4,[r14,#-13]
+	eor	r6,r10,r6,lsr#8
+	strb	r5,[r14,#-9]
+	eor	r7,r11,r7,lsr#8
+	strb	r6,[r14,#-5]
+	strb	r7,[r14,#-1]
+# ifdef	__thumb2__
+	it	ne
+# endif
+	ldrne	r8,[sp,#4*(32+2)]		@ re-load len
+# ifdef	__thumb2__
+	it	hs
+# endif
+	subhs	r11,r8,#64			@ len-=64
+	bhi	.Loop_outer
+
+	beq	.Ldone
+#endif
+
+.Ltail:
+	ldr	r12,[sp,#4*(32+1)]	@ load inp
+	add	r9,sp,#4*(0)
+	ldr	r14,[sp,#4*(32+0)]	@ load out
+
+.Loop_tail:
+	ldrb	r10,[r9],#1	@ read buffer on stack
+	ldrb	r11,[r12],#1		@ read input
+	subs	r8,r8,#1
+	eor	r11,r11,r10
+	strb	r11,[r14],#1		@ store output
+	bne	.Loop_tail
+
+.Ldone:
+	add	sp,sp,#4*(32+3)
+.Lno_data:
+	ldmia	sp!,{r4,r5,r6,r7,r8,r9,r10,r11,pc}
+.size	ChaCha20_ctr32,.-ChaCha20_ctr32
+#if __ARM_MAX_ARCH__>=7
+.arch	armv7-a
+.fpu	neon
+
+.type	ChaCha20_neon,%function
+.align	5
+ChaCha20_neon:
+	ldr	r12,[sp,#0]		@ pull pointer to counter and nonce
+	stmdb	sp!,{r0,r1,r2,r4-r11,lr}
+.LChaCha20_neon:
+	adr	r14,.Lsigma
+	vstmdb	sp!,{d8,d9,d10,d11,d12,d13,d14,d15}		@ ABI spec says so
+	stmdb	sp!,{r0,r1,r2,r3}
+
+	vld1.32	{q1,q2},[r3]		@ load key
+	ldmia	r3,{r4,r5,r6,r7,r8,r9,r10,r11}		@ load key
+
+	sub	sp,sp,#4*(16+16)
+	vld1.32	{q3},[r12]		@ load counter and nonce
+	add	r12,sp,#4*8
+	ldmia	r14,{r0,r1,r2,r3}		@ load sigma
+	vld1.32	{q0},[r14]!		@ load sigma
+	vld1.32	{q12},[r14]		@ one
+	vst1.32	{q2,q3},[r12]		@ copy 1/2key|counter|nonce
+	vst1.32	{q0,q1},[sp]		@ copy sigma|1/2key
+
+	str	r10,[sp,#4*(16+10)]	@ off-load "rx"
+	str	r11,[sp,#4*(16+11)]	@ off-load "rx"
+	vshl.i32	d26,d24,#1	@ two
+	vstr	d24,[sp,#4*(16+0)]
+	vshl.i32	d28,d24,#2	@ four
+	vstr	d26,[sp,#4*(16+2)]
+	vmov	q4,q0
+	vstr	d28,[sp,#4*(16+4)]
+	vmov	q8,q0
+	vmov	q5,q1
+	vmov	q9,q1
+	b	.Loop_neon_enter
+
+.align	4
+.Loop_neon_outer:
+	ldmia	sp,{r0,r1,r2,r3,r4,r5,r6,r7,r8,r9}		@ load key material
+	cmp	r11,#64*2		@ if len<=64*2
+	bls	.Lbreak_neon		@ switch to integer-only
+	vmov	q4,q0
+	str	r11,[sp,#4*(32+2)]	@ save len
+	vmov	q8,q0
+	str	r12,  [sp,#4*(32+1)]	@ save inp
+	vmov	q5,q1
+	str	r14,  [sp,#4*(32+0)]	@ save out
+	vmov	q9,q1
+.Loop_neon_enter:
+	ldr	r11, [sp,#4*(15)]
+	vadd.i32	q7,q3,q12		@ counter+1
+	ldr	r12,[sp,#4*(12)]	@ modulo-scheduled load
+	vmov	q6,q2
+	ldr	r10, [sp,#4*(13)]
+	vmov	q10,q2
+	ldr	r14,[sp,#4*(14)]
+	vadd.i32	q11,q7,q12		@ counter+2
+	str	r11, [sp,#4*(16+15)]
+	mov	r11,#10
+	add	r12,r12,#3	@ counter+3
+	b	.Loop_neon
+
+.align	4
+.Loop_neon:
+	subs	r11,r11,#1
+	vadd.i32	q0,q0,q1
+	add	r0,r0,r4
+	vadd.i32	q4,q4,q5
+	mov	r12,r12,ror#16
+	vadd.i32	q8,q8,q9
+	add	r1,r1,r5
+	veor	q3,q3,q0
+	mov	r10,r10,ror#16
+	veor	q7,q7,q4
+	eor	r12,r12,r0,ror#16
+	veor	q11,q11,q8
+	eor	r10,r10,r1,ror#16
+	vrev32.16	q3,q3
+	add	r8,r8,r12
+	vrev32.16	q7,q7
+	mov	r4,r4,ror#20
+	vrev32.16	q11,q11
+	add	r9,r9,r10
+	vadd.i32	q2,q2,q3
+	mov	r5,r5,ror#20
+	vadd.i32	q6,q6,q7
+	eor	r4,r4,r8,ror#20
+	vadd.i32	q10,q10,q11
+	eor	r5,r5,r9,ror#20
+	veor	q12,q1,q2
+	add	r0,r0,r4
+	veor	q13,q5,q6
+	mov	r12,r12,ror#24
+	veor	q14,q9,q10
+	add	r1,r1,r5
+	vshr.u32	q1,q12,#20
+	mov	r10,r10,ror#24
+	vshr.u32	q5,q13,#20
+	eor	r12,r12,r0,ror#24
+	vshr.u32	q9,q14,#20
+	eor	r10,r10,r1,ror#24
+	vsli.32	q1,q12,#12
+	add	r8,r8,r12
+	vsli.32	q5,q13,#12
+	mov	r4,r4,ror#25
+	vsli.32	q9,q14,#12
+	add	r9,r9,r10
+	vadd.i32	q0,q0,q1
+	mov	r5,r5,ror#25
+	vadd.i32	q4,q4,q5
+	str	r10,[sp,#4*(16+13)]
+	vadd.i32	q8,q8,q9
+	ldr	r10,[sp,#4*(16+15)]
+	veor	q12,q3,q0
+	eor	r4,r4,r8,ror#25
+	veor	q13,q7,q4
+	eor	r5,r5,r9,ror#25
+	veor	q14,q11,q8
+	str	r8,[sp,#4*(16+8)]
+	vshr.u32	q3,q12,#24
+	ldr	r8,[sp,#4*(16+10)]
+	vshr.u32	q7,q13,#24
+	add	r2,r2,r6
+	vshr.u32	q11,q14,#24
+	mov	r14,r14,ror#16
+	vsli.32	q3,q12,#8
+	str	r9,[sp,#4*(16+9)]
+	vsli.32	q7,q13,#8
+	ldr	r9,[sp,#4*(16+11)]
+	vsli.32	q11,q14,#8
+	add	r3,r3,r7
+	vadd.i32	q2,q2,q3
+	mov	r10,r10,ror#16
+	vadd.i32	q6,q6,q7
+	eor	r14,r14,r2,ror#16
+	vadd.i32	q10,q10,q11
+	eor	r10,r10,r3,ror#16
+	veor	q12,q1,q2
+	add	r8,r8,r14
+	veor	q13,q5,q6
+	mov	r6,r6,ror#20
+	veor	q14,q9,q10
+	add	r9,r9,r10
+	vshr.u32	q1,q12,#25
+	mov	r7,r7,ror#20
+	vshr.u32	q5,q13,#25
+	eor	r6,r6,r8,ror#20
+	vshr.u32	q9,q14,#25
+	eor	r7,r7,r9,ror#20
+	vsli.32	q1,q12,#7
+	add	r2,r2,r6
+	vsli.32	q5,q13,#7
+	mov	r14,r14,ror#24
+	vsli.32	q9,q14,#7
+	add	r3,r3,r7
+	vext.8	q2,q2,q2,#8
+	mov	r10,r10,ror#24
+	vext.8	q6,q6,q6,#8
+	eor	r14,r14,r2,ror#24
+	vext.8	q10,q10,q10,#8
+	eor	r10,r10,r3,ror#24
+	vext.8	q1,q1,q1,#4
+	add	r8,r8,r14
+	vext.8	q5,q5,q5,#4
+	mov	r6,r6,ror#25
+	vext.8	q9,q9,q9,#4
+	add	r9,r9,r10
+	vext.8	q3,q3,q3,#12
+	mov	r7,r7,ror#25
+	vext.8	q7,q7,q7,#12
+	eor	r6,r6,r8,ror#25
+	vext.8	q11,q11,q11,#12
+	eor	r7,r7,r9,ror#25
+	vadd.i32	q0,q0,q1
+	add	r0,r0,r5
+	vadd.i32	q4,q4,q5
+	mov	r10,r10,ror#16
+	vadd.i32	q8,q8,q9
+	add	r1,r1,r6
+	veor	q3,q3,q0
+	mov	r12,r12,ror#16
+	veor	q7,q7,q4
+	eor	r10,r10,r0,ror#16
+	veor	q11,q11,q8
+	eor	r12,r12,r1,ror#16
+	vrev32.16	q3,q3
+	add	r8,r8,r10
+	vrev32.16	q7,q7
+	mov	r5,r5,ror#20
+	vrev32.16	q11,q11
+	add	r9,r9,r12
+	vadd.i32	q2,q2,q3
+	mov	r6,r6,ror#20
+	vadd.i32	q6,q6,q7
+	eor	r5,r5,r8,ror#20
+	vadd.i32	q10,q10,q11
+	eor	r6,r6,r9,ror#20
+	veor	q12,q1,q2
+	add	r0,r0,r5
+	veor	q13,q5,q6
+	mov	r10,r10,ror#24
+	veor	q14,q9,q10
+	add	r1,r1,r6
+	vshr.u32	q1,q12,#20
+	mov	r12,r12,ror#24
+	vshr.u32	q5,q13,#20
+	eor	r10,r10,r0,ror#24
+	vshr.u32	q9,q14,#20
+	eor	r12,r12,r1,ror#24
+	vsli.32	q1,q12,#12
+	add	r8,r8,r10
+	vsli.32	q5,q13,#12
+	mov	r5,r5,ror#25
+	vsli.32	q9,q14,#12
+	str	r10,[sp,#4*(16+15)]
+	vadd.i32	q0,q0,q1
+	ldr	r10,[sp,#4*(16+13)]
+	vadd.i32	q4,q4,q5
+	add	r9,r9,r12
+	vadd.i32	q8,q8,q9
+	mov	r6,r6,ror#25
+	veor	q12,q3,q0
+	eor	r5,r5,r8,ror#25
+	veor	q13,q7,q4
+	eor	r6,r6,r9,ror#25
+	veor	q14,q11,q8
+	str	r8,[sp,#4*(16+10)]
+	vshr.u32	q3,q12,#24
+	ldr	r8,[sp,#4*(16+8)]
+	vshr.u32	q7,q13,#24
+	add	r2,r2,r7
+	vshr.u32	q11,q14,#24
+	mov	r10,r10,ror#16
+	vsli.32	q3,q12,#8
+	str	r9,[sp,#4*(16+11)]
+	vsli.32	q7,q13,#8
+	ldr	r9,[sp,#4*(16+9)]
+	vsli.32	q11,q14,#8
+	add	r3,r3,r4
+	vadd.i32	q2,q2,q3
+	mov	r14,r14,ror#16
+	vadd.i32	q6,q6,q7
+	eor	r10,r10,r2,ror#16
+	vadd.i32	q10,q10,q11
+	eor	r14,r14,r3,ror#16
+	veor	q12,q1,q2
+	add	r8,r8,r10
+	veor	q13,q5,q6
+	mov	r7,r7,ror#20
+	veor	q14,q9,q10
+	add	r9,r9,r14
+	vshr.u32	q1,q12,#25
+	mov	r4,r4,ror#20
+	vshr.u32	q5,q13,#25
+	eor	r7,r7,r8,ror#20
+	vshr.u32	q9,q14,#25
+	eor	r4,r4,r9,ror#20
+	vsli.32	q1,q12,#7
+	add	r2,r2,r7
+	vsli.32	q5,q13,#7
+	mov	r10,r10,ror#24
+	vsli.32	q9,q14,#7
+	add	r3,r3,r4
+	vext.8	q2,q2,q2,#8
+	mov	r14,r14,ror#24
+	vext.8	q6,q6,q6,#8
+	eor	r10,r10,r2,ror#24
+	vext.8	q10,q10,q10,#8
+	eor	r14,r14,r3,ror#24
+	vext.8	q1,q1,q1,#12
+	add	r8,r8,r10
+	vext.8	q5,q5,q5,#12
+	mov	r7,r7,ror#25
+	vext.8	q9,q9,q9,#12
+	add	r9,r9,r14
+	vext.8	q3,q3,q3,#4
+	mov	r4,r4,ror#25
+	vext.8	q7,q7,q7,#4
+	eor	r7,r7,r8,ror#25
+	vext.8	q11,q11,q11,#4
+	eor	r4,r4,r9,ror#25
+	bne	.Loop_neon
+
+	add	r11,sp,#32
+	vld1.32	{q12,q13},[sp]		@ load key material
+	vld1.32	{q14,q15},[r11]
+
+	ldr	r11,[sp,#4*(32+2)]	@ load len
+
+	str	r8, [sp,#4*(16+8)]	@ modulo-scheduled store
+	str	r9, [sp,#4*(16+9)]
+	str	r12,[sp,#4*(16+12)]
+	str	r10, [sp,#4*(16+13)]
+	str	r14,[sp,#4*(16+14)]
+
+	@ at this point we have first half of 512-bit result in
+	@ rx and second half at sp+4*(16+8)
+
+	ldr	r12,[sp,#4*(32+1)]	@ load inp
+	ldr	r14,[sp,#4*(32+0)]	@ load out
+
+	vadd.i32	q0,q0,q12		@ accumulate key material
+	vadd.i32	q4,q4,q12
+	vadd.i32	q8,q8,q12
+	vldr	d24,[sp,#4*(16+0)]	@ one
+
+	vadd.i32	q1,q1,q13
+	vadd.i32	q5,q5,q13
+	vadd.i32	q9,q9,q13
+	vldr	d26,[sp,#4*(16+2)]	@ two
+
+	vadd.i32	q2,q2,q14
+	vadd.i32	q6,q6,q14
+	vadd.i32	q10,q10,q14
+	vadd.i32	d14,d14,d24	@ counter+1
+	vadd.i32	d22,d22,d26	@ counter+2
+
+	vadd.i32	q3,q3,q15
+	vadd.i32	q7,q7,q15
+	vadd.i32	q11,q11,q15
+
+	cmp	r11,#64*4
+	blo	.Ltail_neon
+
+	vld1.8	{q12,q13},[r12]!	@ load input
+	mov	r11,sp
+	vld1.8	{q14,q15},[r12]!
+	veor	q0,q0,q12		@ xor with input
+	veor	q1,q1,q13
+	vld1.8	{q12,q13},[r12]!
+	veor	q2,q2,q14
+	veor	q3,q3,q15
+	vld1.8	{q14,q15},[r12]!
+
+	veor	q4,q4,q12
+	vst1.8	{q0,q1},[r14]!	@ store output
+	veor	q5,q5,q13
+	vld1.8	{q12,q13},[r12]!
+	veor	q6,q6,q14
+	vst1.8	{q2,q3},[r14]!
+	veor	q7,q7,q15
+	vld1.8	{q14,q15},[r12]!
+
+	veor	q8,q8,q12
+	vld1.32	{q0,q1},[r11]!	@ load for next iteration
+	veor	d25,d25,d25
+	vldr	d24,[sp,#4*(16+4)]	@ four
+	veor	q9,q9,q13
+	vld1.32	{q2,q3},[r11]
+	veor	q10,q10,q14
+	vst1.8	{q4,q5},[r14]!
+	veor	q11,q11,q15
+	vst1.8	{q6,q7},[r14]!
+
+	vadd.i32	d6,d6,d24	@ next counter value
+	vldr	d24,[sp,#4*(16+0)]	@ one
+
+	ldmia	sp,{r8,r9,r10,r11}	@ load key material
+	add	r0,r0,r8	@ accumulate key material
+	ldr	r8,[r12],#16		@ load input
+	vst1.8	{q8,q9},[r14]!
+	add	r1,r1,r9
+	ldr	r9,[r12,#-12]
+	vst1.8	{q10,q11},[r14]!
+	add	r2,r2,r10
+	ldr	r10,[r12,#-8]
+	add	r3,r3,r11
+	ldr	r11,[r12,#-4]
+# ifdef	__ARMEB__
+	rev	r0,r0
+	rev	r1,r1
+	rev	r2,r2
+	rev	r3,r3
+# endif
+	eor	r0,r0,r8	@ xor with input
+	add	r8,sp,#4*(4)
+	eor	r1,r1,r9
+	str	r0,[r14],#16		@ store output
+	eor	r2,r2,r10
+	str	r1,[r14,#-12]
+	eor	r3,r3,r11
+	ldmia	r8,{r8,r9,r10,r11}	@ load key material
+	str	r2,[r14,#-8]
+	str	r3,[r14,#-4]
+
+	add	r4,r4,r8	@ accumulate key material
+	ldr	r8,[r12],#16		@ load input
+	add	r5,r5,r9
+	ldr	r9,[r12,#-12]
+	add	r6,r6,r10
+	ldr	r10,[r12,#-8]
+	add	r7,r7,r11
+	ldr	r11,[r12,#-4]
+# ifdef	__ARMEB__
+	rev	r4,r4
+	rev	r5,r5
+	rev	r6,r6
+	rev	r7,r7
+# endif
+	eor	r4,r4,r8
+	add	r8,sp,#4*(8)
+	eor	r5,r5,r9
+	str	r4,[r14],#16		@ store output
+	eor	r6,r6,r10
+	str	r5,[r14,#-12]
+	eor	r7,r7,r11
+	ldmia	r8,{r8,r9,r10,r11}	@ load key material
+	str	r6,[r14,#-8]
+	add	r0,sp,#4*(16+8)
+	str	r7,[r14,#-4]
+
+	ldmia	r0,{r0,r1,r2,r3,r4,r5,r6,r7}	@ load second half
+
+	add	r0,r0,r8	@ accumulate key material
+	ldr	r8,[r12],#16		@ load input
+	add	r1,r1,r9
+	ldr	r9,[r12,#-12]
+# ifdef	__thumb2__
+	it	hi
+# endif
+	strhi	r10,[sp,#4*(16+10)]	@ copy "rx" while at it
+	add	r2,r2,r10
+	ldr	r10,[r12,#-8]
+# ifdef	__thumb2__
+	it	hi
+# endif
+	strhi	r11,[sp,#4*(16+11)]	@ copy "rx" while at it
+	add	r3,r3,r11
+	ldr	r11,[r12,#-4]
+# ifdef	__ARMEB__
+	rev	r0,r0
+	rev	r1,r1
+	rev	r2,r2
+	rev	r3,r3
+# endif
+	eor	r0,r0,r8
+	add	r8,sp,#4*(12)
+	eor	r1,r1,r9
+	str	r0,[r14],#16		@ store output
+	eor	r2,r2,r10
+	str	r1,[r14,#-12]
+	eor	r3,r3,r11
+	ldmia	r8,{r8,r9,r10,r11}	@ load key material
+	str	r2,[r14,#-8]
+	str	r3,[r14,#-4]
+
+	add	r4,r4,r8	@ accumulate key material
+	add	r8,r8,#4		@ next counter value
+	add	r5,r5,r9
+	str	r8,[sp,#4*(12)]	@ save next counter value
+	ldr	r8,[r12],#16		@ load input
+	add	r6,r6,r10
+	add	r4,r4,#3		@ counter+3
+	ldr	r9,[r12,#-12]
+	add	r7,r7,r11
+	ldr	r10,[r12,#-8]
+	ldr	r11,[r12,#-4]
+# ifdef	__ARMEB__
+	rev	r4,r4
+	rev	r5,r5
+	rev	r6,r6
+	rev	r7,r7
+# endif
+	eor	r4,r4,r8
+# ifdef	__thumb2__
+	it	hi
+# endif
+	ldrhi	r8,[sp,#4*(32+2)]	@ re-load len
+	eor	r5,r5,r9
+	eor	r6,r6,r10
+	str	r4,[r14],#16		@ store output
+	eor	r7,r7,r11
+	str	r5,[r14,#-12]
+	sub	r11,r8,#64*4	@ len-=64*4
+	str	r6,[r14,#-8]
+	str	r7,[r14,#-4]
+	bhi	.Loop_neon_outer
+
+	b	.Ldone_neon
+
+.align	4
+.Lbreak_neon:
+	@ harmonize NEON and integer-only stack frames: load data
+	@ from NEON frame, but save to integer-only one; distance
+	@ between the two is 4*(32+4+16-32)=4*(20).
+
+	str	r11, [sp,#4*(20+32+2)]	@ save len
+	add	r11,sp,#4*(32+4)
+	str	r12,   [sp,#4*(20+32+1)]	@ save inp
+	str	r14,   [sp,#4*(20+32+0)]	@ save out
+
+	ldr	r12,[sp,#4*(16+10)]
+	ldr	r14,[sp,#4*(16+11)]
+	vldmia	r11,{d8,d9,d10,d11,d12,d13,d14,d15}			@ fulfill ABI requirement
+	str	r12,[sp,#4*(20+16+10)]	@ copy "rx"
+	str	r14,[sp,#4*(20+16+11)]	@ copy "rx"
+
+	ldr	r11, [sp,#4*(15)]
+	ldr	r12,[sp,#4*(12)]		@ modulo-scheduled load
+	ldr	r10, [sp,#4*(13)]
+	ldr	r14,[sp,#4*(14)]
+	str	r11, [sp,#4*(20+16+15)]
+	add	r11,sp,#4*(20)
+	vst1.32	{q0,q1},[r11]!		@ copy key
+	add	sp,sp,#4*(20)			@ switch frame
+	vst1.32	{q2,q3},[r11]
+	mov	r11,#10
+	b	.Loop				@ go integer-only
+
+.align	4
+.Ltail_neon:
+	cmp	r11,#64*3
+	bhs	.L192_or_more_neon
+	cmp	r11,#64*2
+	bhs	.L128_or_more_neon
+	cmp	r11,#64*1
+	bhs	.L64_or_more_neon
+
+	add	r8,sp,#4*(8)
+	vst1.8	{q0,q1},[sp]
+	add	r10,sp,#4*(0)
+	vst1.8	{q2,q3},[r8]
+	b	.Loop_tail_neon
+
+.align	4
+.L64_or_more_neon:
+	vld1.8	{q12,q13},[r12]!
+	vld1.8	{q14,q15},[r12]!
+	veor	q0,q0,q12
+	veor	q1,q1,q13
+	veor	q2,q2,q14
+	veor	q3,q3,q15
+	vst1.8	{q0,q1},[r14]!
+	vst1.8	{q2,q3},[r14]!
+
+	beq	.Ldone_neon
+
+	add	r8,sp,#4*(8)
+	vst1.8	{q4,q5},[sp]
+	add	r10,sp,#4*(0)
+	vst1.8	{q6,q7},[r8]
+	sub	r11,r11,#64*1	@ len-=64*1
+	b	.Loop_tail_neon
+
+.align	4
+.L128_or_more_neon:
+	vld1.8	{q12,q13},[r12]!
+	vld1.8	{q14,q15},[r12]!
+	veor	q0,q0,q12
+	veor	q1,q1,q13
+	vld1.8	{q12,q13},[r12]!
+	veor	q2,q2,q14
+	veor	q3,q3,q15
+	vld1.8	{q14,q15},[r12]!
+
+	veor	q4,q4,q12
+	veor	q5,q5,q13
+	vst1.8	{q0,q1},[r14]!
+	veor	q6,q6,q14
+	vst1.8	{q2,q3},[r14]!
+	veor	q7,q7,q15
+	vst1.8	{q4,q5},[r14]!
+	vst1.8	{q6,q7},[r14]!
+
+	beq	.Ldone_neon
+
+	add	r8,sp,#4*(8)
+	vst1.8	{q8,q9},[sp]
+	add	r10,sp,#4*(0)
+	vst1.8	{q10,q11},[r8]
+	sub	r11,r11,#64*2	@ len-=64*2
+	b	.Loop_tail_neon
+
+.align	4
+.L192_or_more_neon:
+	vld1.8	{q12,q13},[r12]!
+	vld1.8	{q14,q15},[r12]!
+	veor	q0,q0,q12
+	veor	q1,q1,q13
+	vld1.8	{q12,q13},[r12]!
+	veor	q2,q2,q14
+	veor	q3,q3,q15
+	vld1.8	{q14,q15},[r12]!
+
+	veor	q4,q4,q12
+	veor	q5,q5,q13
+	vld1.8	{q12,q13},[r12]!
+	veor	q6,q6,q14
+	vst1.8	{q0,q1},[r14]!
+	veor	q7,q7,q15
+	vld1.8	{q14,q15},[r12]!
+
+	veor	q8,q8,q12
+	vst1.8	{q2,q3},[r14]!
+	veor	q9,q9,q13
+	vst1.8	{q4,q5},[r14]!
+	veor	q10,q10,q14
+	vst1.8	{q6,q7},[r14]!
+	veor	q11,q11,q15
+	vst1.8	{q8,q9},[r14]!
+	vst1.8	{q10,q11},[r14]!
+
+	beq	.Ldone_neon
+
+	ldmia	sp,{r8,r9,r10,r11}	@ load key material
+	add	r0,r0,r8	@ accumulate key material
+	add	r8,sp,#4*(4)
+	add	r1,r1,r9
+	add	r2,r2,r10
+	add	r3,r3,r11
+	ldmia	r8,{r8,r9,r10,r11}	@ load key material
+
+	add	r4,r4,r8	@ accumulate key material
+	add	r8,sp,#4*(8)
+	add	r5,r5,r9
+	add	r6,r6,r10
+	add	r7,r7,r11
+	ldmia	r8,{r8,r9,r10,r11}	@ load key material
+# ifdef	__ARMEB__
+	rev	r0,r0
+	rev	r1,r1
+	rev	r2,r2
+	rev	r3,r3
+	rev	r4,r4
+	rev	r5,r5
+	rev	r6,r6
+	rev	r7,r7
+# endif
+	stmia	sp,{r0,r1,r2,r3,r4,r5,r6,r7}
+	add	r0,sp,#4*(16+8)
+
+	ldmia	r0,{r0,r1,r2,r3,r4,r5,r6,r7}	@ load second half
+
+	add	r0,r0,r8	@ accumulate key material
+	add	r8,sp,#4*(12)
+	add	r1,r1,r9
+	add	r2,r2,r10
+	add	r3,r3,r11
+	ldmia	r8,{r8,r9,r10,r11}	@ load key material
+
+	add	r4,r4,r8	@ accumulate key material
+	add	r8,sp,#4*(8)
+	add	r5,r5,r9
+	add	r4,r4,#3		@ counter+3
+	add	r6,r6,r10
+	add	r7,r7,r11
+	ldr	r11,[sp,#4*(32+2)]	@ re-load len
+# ifdef	__ARMEB__
+	rev	r0,r0
+	rev	r1,r1
+	rev	r2,r2
+	rev	r3,r3
+	rev	r4,r4
+	rev	r5,r5
+	rev	r6,r6
+	rev	r7,r7
+# endif
+	stmia	r8,{r0,r1,r2,r3,r4,r5,r6,r7}
+	add	r10,sp,#4*(0)
+	sub	r11,r11,#64*3	@ len-=64*3
+
+.Loop_tail_neon:
+	ldrb	r8,[r10],#1	@ read buffer on stack
+	ldrb	r9,[r12],#1		@ read input
+	subs	r11,r11,#1
+	eor	r8,r8,r9
+	strb	r8,[r14],#1		@ store ouput
+	bne	.Loop_tail_neon
+
+.Ldone_neon:
+	add	sp,sp,#4*(32+4)
+	vldmia	sp,{d8,d9,d10,d11,d12,d13,d14,d15}
+	add	sp,sp,#4*(16+3)
+	ldmia	sp!,{r4,r5,r6,r7,r8,r9,r10,r11,pc}
+.size	ChaCha20_neon,.-ChaCha20_neon
+.comm	OPENSSL_armcap_P,4,4
+#endif
+#endif

third_party/boringssl/linux-arm/crypto/modes/ghash-armv4.S

diff --git a/third_party/boringssl/linux-arm/crypto/modes/ghash-armv4.S b/third_party/boringssl/linux-arm/crypto/modes/ghash-armv4.S
index c6f025d..791b289 100644
--- a/third_party/boringssl/linux-arm/crypto/modes/ghash-armv4.S
+++ b/third_party/boringssl/linux-arm/crypto/modes/ghash-armv4.S
@@ -1,13 +1,12 @@
 #if defined(__arm__)
-#if defined(__arm__)
-#include "arm_arch.h"
+#include <openssl/arm_arch.h>
 
 .syntax	unified
 
 .text
 .code	32
 
-#ifdef  __APPLE__
+#ifdef  __clang__
 #define ldrplb  ldrbpl
 #define ldrneb  ldrbne
 #endif
@@ -536,6 +535,4 @@
 .byte	71,72,65,83,72,32,102,111,114,32,65,82,77,118,52,47,78,69,79,78,44,32,67,82,89,80,84,79,71,65,77,83,32,98,121,32,60,97,112,112,114,111,64,111,112,101,110,115,115,108,46,111,114,103,62,0
 .align	2
 .align	2
-
 #endif
-#endif
\ No newline at end of file

third_party/boringssl/linux-arm/crypto/modes/ghashv8-armx32.S

diff --git a/third_party/boringssl/linux-arm/crypto/modes/ghashv8-armx32.S b/third_party/boringssl/linux-arm/crypto/modes/ghashv8-armx32.S
index bdbbae9..0e1e631 100644
--- a/third_party/boringssl/linux-arm/crypto/modes/ghashv8-armx32.S
+++ b/third_party/boringssl/linux-arm/crypto/modes/ghashv8-armx32.S
@@ -1,10 +1,11 @@
 #if defined(__arm__)
-#include "arm_arch.h"
+#include <openssl/arm_arch.h>
 
 .text
 .fpu	neon
 .code	32
 .globl	gcm_init_v8
+.hidden	gcm_init_v8
 .type	gcm_init_v8,%function
 .align	4
 gcm_init_v8:
@@ -55,6 +56,7 @@
 	bx	lr
 .size	gcm_init_v8,.-gcm_init_v8
 .globl	gcm_gmult_v8
+.hidden	gcm_gmult_v8
 .type	gcm_gmult_v8,%function
 .align	4
 gcm_gmult_v8:
@@ -67,10 +69,10 @@
 #endif
 	vext.8	q3,q9,q9,#8
 
-.byte	0x86,0x0e,0xa8,0xf2	@ pmull q0,q12,q3		@ H.lo·Xi.lo
+.byte	0x86,0x0e,0xa8,0xf2	@ pmull q0,q12,q3		@ H.lo·Xi.lo
 	veor	q9,q9,q3		@ Karatsuba pre-processing
-.byte	0x87,0x4e,0xa9,0xf2	@ pmull2 q2,q12,q3		@ H.hi·Xi.hi
-.byte	0xa2,0x2e,0xaa,0xf2	@ pmull q1,q13,q9		@ (H.lo+H.hi)·(Xi.lo+Xi.hi)
+.byte	0x87,0x4e,0xa9,0xf2	@ pmull2 q2,q12,q3		@ H.hi·Xi.hi
+.byte	0xa2,0x2e,0xaa,0xf2	@ pmull q1,q13,q9		@ (H.lo+H.hi)·(Xi.lo+Xi.hi)
 
 	vext.8	q9,q0,q2,#8		@ Karatsuba post-processing
 	veor	q10,q0,q2
@@ -96,6 +98,7 @@
 	bx	lr
 .size	gcm_gmult_v8,.-gcm_gmult_v8
 .globl	gcm_ghash_v8
+.hidden	gcm_ghash_v8
 .type	gcm_ghash_v8,%function
 .align	4
 gcm_ghash_v8:
@@ -135,7 +138,7 @@
 #endif
 	vext.8	q7,q9,q9,#8
 	veor	q3,q3,q0		@ I[i]^=Xi
-.byte	0x8e,0x8e,0xa8,0xf2	@ pmull q4,q12,q7		@ H·Ii+1
+.byte	0x8e,0x8e,0xa8,0xf2	@ pmull q4,q12,q7		@ H·Ii+1
 	veor	q9,q9,q7		@ Karatsuba pre-processing
 .byte	0x8f,0xce,0xa9,0xf2	@ pmull2 q6,q12,q7
 	b	.Loop_mod2x_v8
@@ -144,14 +147,14 @@
 .Loop_mod2x_v8:
 	vext.8	q10,q3,q3,#8
 	subs	r3,r3,#32		@ is there more data?
-.byte	0x86,0x0e,0xac,0xf2	@ pmull q0,q14,q3		@ H^2.lo·Xi.lo
+.byte	0x86,0x0e,0xac,0xf2	@ pmull q0,q14,q3		@ H^2.lo·Xi.lo
 	movlo	r12,#0			@ is it time to zero r12?
 
 .byte	0xa2,0xae,0xaa,0xf2	@ pmull q5,q13,q9
 	veor	q10,q10,q3		@ Karatsuba pre-processing
-.byte	0x87,0x4e,0xad,0xf2	@ pmull2 q2,q14,q3		@ H^2.hi·Xi.hi
+.byte	0x87,0x4e,0xad,0xf2	@ pmull2 q2,q14,q3		@ H^2.hi·Xi.hi
 	veor	q0,q0,q4		@ accumulate
-.byte	0xa5,0x2e,0xab,0xf2	@ pmull2 q1,q13,q10		@ (H^2.lo+H^2.hi)·(Xi.lo+Xi.hi)
+.byte	0xa5,0x2e,0xab,0xf2	@ pmull2 q1,q13,q10		@ (H^2.lo+H^2.hi)·(Xi.lo+Xi.hi)
 	vld1.64	{q8},[r2],r12	@ load [rotated] I[i+2]
 
 	veor	q2,q2,q6
@@ -176,7 +179,7 @@
 	vext.8	q7,q9,q9,#8
 	vext.8	q3,q8,q8,#8
 	veor	q0,q1,q10
-.byte	0x8e,0x8e,0xa8,0xf2	@ pmull q4,q12,q7		@ H·Ii+1
+.byte	0x8e,0x8e,0xa8,0xf2	@ pmull q4,q12,q7		@ H·Ii+1
 	veor	q3,q3,q2		@ accumulate q3 early
 
 	vext.8	q10,q0,q0,#8		@ 2nd phase of reduction
@@ -197,10 +200,10 @@
 	veor	q3,q3,q0		@ inp^=Xi
 	veor	q9,q8,q10		@ q9 is rotated inp^Xi
 
-.byte	0x86,0x0e,0xa8,0xf2	@ pmull q0,q12,q3		@ H.lo·Xi.lo
+.byte	0x86,0x0e,0xa8,0xf2	@ pmull q0,q12,q3		@ H.lo·Xi.lo
 	veor	q9,q9,q3		@ Karatsuba pre-processing
-.byte	0x87,0x4e,0xa9,0xf2	@ pmull2 q2,q12,q3		@ H.hi·Xi.hi
-.byte	0xa2,0x2e,0xaa,0xf2	@ pmull q1,q13,q9		@ (H.lo+H.hi)·(Xi.lo+Xi.hi)
+.byte	0x87,0x4e,0xa9,0xf2	@ pmull2 q2,q12,q3		@ H.hi·Xi.hi
+.byte	0xa2,0x2e,0xaa,0xf2	@ pmull q1,q13,q9		@ (H.lo+H.hi)·(Xi.lo+Xi.hi)
 
 	vext.8	q9,q0,q2,#8		@ Karatsuba post-processing
 	veor	q10,q0,q2
@@ -230,4 +233,4 @@
 .byte	71,72,65,83,72,32,102,111,114,32,65,82,77,118,56,44,32,67,82,89,80,84,79,71,65,77,83,32,98,121,32,60,97,112,112,114,111,64,111,112,101,110,115,115,108,46,111,114,103,62,0
 .align	2
 .align	2
-#endif
\ No newline at end of file
+#endif

third_party/boringssl/linux-arm/crypto/sha/sha1-armv4-large.S

diff --git a/third_party/boringssl/linux-arm/crypto/sha/sha1-armv4-large.S b/third_party/boringssl/linux-arm/crypto/sha/sha1-armv4-large.S
index 4911458..36955fa 100644
--- a/third_party/boringssl/linux-arm/crypto/sha/sha1-armv4-large.S
+++ b/third_party/boringssl/linux-arm/crypto/sha/sha1-armv4-large.S
@@ -1,10 +1,11 @@
 #if defined(__arm__)
-#include "arm_arch.h"
+#include <openssl/arm_arch.h>
 
 .text
 .code	32
 
 .globl	sha1_block_data_order
+.hidden	sha1_block_data_order
 .type	sha1_block_data_order,%function
 
 .align	5
@@ -1459,4 +1460,4 @@
 .comm	OPENSSL_armcap_P,4,4
 .hidden	OPENSSL_armcap_P
 #endif
-#endif
\ No newline at end of file
+#endif

third_party/boringssl/linux-arm/crypto/sha/sha256-armv4.S

diff --git a/third_party/boringssl/linux-arm/crypto/sha/sha256-armv4.S b/third_party/boringssl/linux-arm/crypto/sha/sha256-armv4.S
index ac9f2f1..6040041 100644
--- a/third_party/boringssl/linux-arm/crypto/sha/sha256-armv4.S
+++ b/third_party/boringssl/linux-arm/crypto/sha/sha256-armv4.S
@@ -38,7 +38,7 @@
 @ Add ARMv8 code path performing at 2.0 cpb on Apple A7.
 
 #ifndef __KERNEL__
-# include "arm_arch.h"
+# include <openssl/arm_arch.h>
 #else
 # define __ARM_ARCH__ __LINUX_ARM_ARCH__
 # define __ARM_MAX_ARCH__ 7
@@ -85,6 +85,7 @@
 .align	5
 
 .globl	sha256_block_data_order
+.hidden	sha256_block_data_order
 .type	sha256_block_data_order,%function
 sha256_block_data_order:
 .Lsha256_block_data_order:
@@ -1875,6 +1876,7 @@
 .fpu	neon
 
 .globl	sha256_block_data_order_neon
+.hidden	sha256_block_data_order_neon
 .type	sha256_block_data_order_neon,%function
 .align	4
 sha256_block_data_order_neon:
@@ -2815,4 +2817,4 @@
 .comm	OPENSSL_armcap_P,4,4
 .hidden	OPENSSL_armcap_P
 #endif
-#endif
\ No newline at end of file
+#endif

third_party/boringssl/linux-arm/crypto/sha/sha512-armv4.S

diff --git a/third_party/boringssl/linux-arm/crypto/sha/sha512-armv4.S b/third_party/boringssl/linux-arm/crypto/sha/sha512-armv4.S
index c794f87..93a7bf8 100644
--- a/third_party/boringssl/linux-arm/crypto/sha/sha512-armv4.S
+++ b/third_party/boringssl/linux-arm/crypto/sha/sha512-armv4.S
@@ -47,7 +47,7 @@
 @ was reflected in below two parameters as 0 and 4. Now caller is
 @ expected to maintain native byte order for whole 64-bit values.
 #ifndef __KERNEL__
-# include "arm_arch.h"
+# include <openssl/arm_arch.h>
 # define VFP_ABI_PUSH	vstmdb	sp!,{d8-d15}
 # define VFP_ABI_POP	vldmia	sp!,{d8-d15}
 #else
@@ -133,6 +133,7 @@
 #endif
 
 .globl	sha512_block_data_order
+.hidden	sha512_block_data_order
 .type	sha512_block_data_order,%function
 sha512_block_data_order:
 .Lsha512_block_data_order:
@@ -147,7 +148,7 @@
 #ifdef	__APPLE__
 	ldr	r12,[r12]
 #endif
-	tst	r12,#1
+	tst	r12,#ARMV7_NEON
 	bne	.LNEON
 #endif
 	add	r2,r1,r2,lsl#7	@ len to point at the end of inp
@@ -533,6 +534,7 @@
 .fpu	neon
 
 .globl	sha512_block_data_order_neon
+.hidden	sha512_block_data_order_neon
 .type	sha512_block_data_order_neon,%function
 .align	4
 sha512_block_data_order_neon:
@@ -1866,4 +1868,4 @@
 .comm	OPENSSL_armcap_P,4,4
 .hidden	OPENSSL_armcap_P
 #endif
-#endif
\ No newline at end of file
+#endif

third_party/boringssl/linux-x86/crypto/chacha/chacha-x86.S

diff --git a/third_party/boringssl/linux-x86/crypto/chacha/chacha-x86.S b/third_party/boringssl/linux-x86/crypto/chacha/chacha-x86.S
new file mode 100644
index 0000000..d3c39ac
--- /dev/null
+++ b/third_party/boringssl/linux-x86/crypto/chacha/chacha-x86.S
@@ -0,0 +1,969 @@
+#if defined(__i386__)
+.file	"chacha-x86.S"
+.text
+.globl	ChaCha20_ctr32
+.hidden	ChaCha20_ctr32
+.type	ChaCha20_ctr32,@function
+.align	16
+ChaCha20_ctr32:
+.L_ChaCha20_ctr32_begin:
+	pushl	%ebp
+	pushl	%ebx
+	pushl	%esi
+	pushl	%edi
+	xorl	%eax,%eax
+	cmpl	28(%esp),%eax
+	je	.L000no_data
+	call	.Lpic_point
+.Lpic_point:
+	popl	%eax
+	leal	OPENSSL_ia32cap_P-.Lpic_point(%eax),%ebp
+	testl	$16777216,(%ebp)
+	jz	.L001x86
+	testl	$512,4(%ebp)
+	jz	.L001x86
+	jmp	.Lssse3_shortcut
+.L001x86:
+	movl	32(%esp),%esi
+	movl	36(%esp),%edi
+	subl	$132,%esp
+	movl	(%esi),%eax
+	movl	4(%esi),%ebx
+	movl	8(%esi),%ecx
+	movl	12(%esi),%edx
+	movl	%eax,80(%esp)
+	movl	%ebx,84(%esp)
+	movl	%ecx,88(%esp)
+	movl	%edx,92(%esp)
+	movl	16(%esi),%eax
+	movl	20(%esi),%ebx
+	movl	24(%esi),%ecx
+	movl	28(%esi),%edx
+	movl	%eax,96(%esp)
+	movl	%ebx,100(%esp)
+	movl	%ecx,104(%esp)
+	movl	%edx,108(%esp)
+	movl	(%edi),%eax
+	movl	4(%edi),%ebx
+	movl	8(%edi),%ecx
+	movl	12(%edi),%edx
+	subl	$1,%eax
+	movl	%eax,112(%esp)
+	movl	%ebx,116(%esp)
+	movl	%ecx,120(%esp)
+	movl	%edx,124(%esp)
+	jmp	.L002entry
+.align	16
+.L003outer_loop:
+	movl	%ebx,156(%esp)
+	movl	%eax,152(%esp)
+	movl	%ecx,160(%esp)
+.L002entry:
+	movl	$1634760805,%eax
+	movl	$857760878,4(%esp)
+	movl	$2036477234,8(%esp)
+	movl	$1797285236,12(%esp)
+	movl	84(%esp),%ebx
+	movl	88(%esp),%ebp
+	movl	104(%esp),%ecx
+	movl	108(%esp),%esi
+	movl	116(%esp),%edx
+	movl	120(%esp),%edi
+	movl	%ebx,20(%esp)
+	movl	%ebp,24(%esp)
+	movl	%ecx,40(%esp)
+	movl	%esi,44(%esp)
+	movl	%edx,52(%esp)
+	movl	%edi,56(%esp)
+	movl	92(%esp),%ebx
+	movl	124(%esp),%edi
+	movl	112(%esp),%edx
+	movl	80(%esp),%ebp
+	movl	96(%esp),%ecx
+	movl	100(%esp),%esi
+	addl	$1,%edx
+	movl	%ebx,28(%esp)
+	movl	%edi,60(%esp)
+	movl	%edx,112(%esp)
+	movl	$10,%ebx
+	jmp	.L004loop
+.align	16
+.L004loop:
+	addl	%ebp,%eax
+	movl	%ebx,128(%esp)
+	movl	%ebp,%ebx
+	xorl	%eax,%edx
+	roll	$16,%edx
+	addl	%edx,%ecx
+	xorl	%ecx,%ebx
+	movl	52(%esp),%edi
+	roll	$12,%ebx
+	movl	20(%esp),%ebp
+	addl	%ebx,%eax
+	xorl	%eax,%edx
+	movl	%eax,(%esp)
+	roll	$8,%edx
+	movl	4(%esp),%eax
+	addl	%edx,%ecx
+	movl	%edx,48(%esp)
+	xorl	%ecx,%ebx
+	addl	%ebp,%eax
+	roll	$7,%ebx
+	xorl	%eax,%edi
+	movl	%ecx,32(%esp)
+	roll	$16,%edi
+	movl	%ebx,16(%esp)
+	addl	%edi,%esi
+	movl	40(%esp),%ecx
+	xorl	%esi,%ebp
+	movl	56(%esp),%edx
+	roll	$12,%ebp
+	movl	24(%esp),%ebx
+	addl	%ebp,%eax
+	xorl	%eax,%edi
+	movl	%eax,4(%esp)
+	roll	$8,%edi
+	movl	8(%esp),%eax
+	addl	%edi,%esi
+	movl	%edi,52(%esp)
+	xorl	%esi,%ebp
+	addl	%ebx,%eax
+	roll	$7,%ebp
+	xorl	%eax,%edx
+	movl	%esi,36(%esp)
+	roll	$16,%edx
+	movl	%ebp,20(%esp)
+	addl	%edx,%ecx
+	movl	44(%esp),%esi
+	xorl	%ecx,%ebx
+	movl	60(%esp),%edi
+	roll	$12,%ebx
+	movl	28(%esp),%ebp
+	addl	%ebx,%eax
+	xorl	%eax,%edx
+	movl	%eax,8(%esp)
+	roll	$8,%edx
+	movl	12(%esp),%eax
+	addl	%edx,%ecx
+	movl	%edx,56(%esp)
+	xorl	%ecx,%ebx
+	addl	%ebp,%eax
+	roll	$7,%ebx
+	xorl	%eax,%edi
+	roll	$16,%edi
+	movl	%ebx,24(%esp)
+	addl	%edi,%esi
+	xorl	%esi,%ebp
+	roll	$12,%ebp
+	movl	20(%esp),%ebx
+	addl	%ebp,%eax
+	xorl	%eax,%edi
+	movl	%eax,12(%esp)
+	roll	$8,%edi
+	movl	(%esp),%eax
+	addl	%edi,%esi
+	movl	%edi,%edx
+	xorl	%esi,%ebp
+	addl	%ebx,%eax
+	roll	$7,%ebp
+	xorl	%eax,%edx
+	roll	$16,%edx
+	movl	%ebp,28(%esp)
+	addl	%edx,%ecx
+	xorl	%ecx,%ebx
+	movl	48(%esp),%edi
+	roll	$12,%ebx
+	movl	24(%esp),%ebp
+	addl	%ebx,%eax
+	xorl	%eax,%edx
+	movl	%eax,(%esp)
+	roll	$8,%edx
+	movl	4(%esp),%eax
+	addl	%edx,%ecx
+	movl	%edx,60(%esp)
+	xorl	%ecx,%ebx
+	addl	%ebp,%eax
+	roll	$7,%ebx
+	xorl	%eax,%edi
+	movl	%ecx,40(%esp)
+	roll	$16,%edi
+	movl	%ebx,20(%esp)
+	addl	%edi,%esi
+	movl	32(%esp),%ecx
+	xorl	%esi,%ebp
+	movl	52(%esp),%edx
+	roll	$12,%ebp
+	movl	28(%esp),%ebx
+	addl	%ebp,%eax
+	xorl	%eax,%edi
+	movl	%eax,4(%esp)
+	roll	$8,%edi
+	movl	8(%esp),%eax
+	addl	%edi,%esi
+	movl	%edi,48(%esp)
+	xorl	%esi,%ebp
+	addl	%ebx,%eax
+	roll	$7,%ebp
+	xorl	%eax,%edx
+	movl	%esi,44(%esp)
+	roll	$16,%edx
+	movl	%ebp,24(%esp)
+	addl	%edx,%ecx
+	movl	36(%esp),%esi
+	xorl	%ecx,%ebx
+	movl	56(%esp),%edi
+	roll	$12,%ebx
+	movl	16(%esp),%ebp
+	addl	%ebx,%eax
+	xorl	%eax,%edx
+	movl	%eax,8(%esp)
+	roll	$8,%edx
+	movl	12(%esp),%eax
+	addl	%edx,%ecx
+	movl	%edx,52(%esp)
+	xorl	%ecx,%ebx
+	addl	%ebp,%eax
+	roll	$7,%ebx
+	xorl	%eax,%edi
+	roll	$16,%edi
+	movl	%ebx,28(%esp)
+	addl	%edi,%esi
+	xorl	%esi,%ebp
+	movl	48(%esp),%edx
+	roll	$12,%ebp
+	movl	128(%esp),%ebx
+	addl	%ebp,%eax
+	xorl	%eax,%edi
+	movl	%eax,12(%esp)
+	roll	$8,%edi
+	movl	(%esp),%eax
+	addl	%edi,%esi
+	movl	%edi,56(%esp)
+	xorl	%esi,%ebp
+	roll	$7,%ebp
+	decl	%ebx
+	jnz	.L004loop
+	movl	160(%esp),%ebx
+	addl	$1634760805,%eax
+	addl	80(%esp),%ebp
+	addl	96(%esp),%ecx
+	addl	100(%esp),%esi
+	cmpl	$64,%ebx
+	jb	.L005tail
+	movl	156(%esp),%ebx
+	addl	112(%esp),%edx
+	addl	120(%esp),%edi
+	xorl	(%ebx),%eax
+	xorl	16(%ebx),%ebp
+	movl	%eax,(%esp)
+	movl	152(%esp),%eax
+	xorl	32(%ebx),%ecx
+	xorl	36(%ebx),%esi
+	xorl	48(%ebx),%edx
+	xorl	56(%ebx),%edi
+	movl	%ebp,16(%eax)
+	movl	%ecx,32(%eax)
+	movl	%esi,36(%eax)
+	movl	%edx,48(%eax)
+	movl	%edi,56(%eax)
+	movl	4(%esp),%ebp
+	movl	8(%esp),%ecx
+	movl	12(%esp),%esi
+	movl	20(%esp),%edx
+	movl	24(%esp),%edi
+	addl	$857760878,%ebp
+	addl	$2036477234,%ecx
+	addl	$1797285236,%esi
+	addl	84(%esp),%edx
+	addl	88(%esp),%edi
+	xorl	4(%ebx),%ebp
+	xorl	8(%ebx),%ecx
+	xorl	12(%ebx),%esi
+	xorl	20(%ebx),%edx
+	xorl	24(%ebx),%edi
+	movl	%ebp,4(%eax)
+	movl	%ecx,8(%eax)
+	movl	%esi,12(%eax)
+	movl	%edx,20(%eax)
+	movl	%edi,24(%eax)
+	movl	28(%esp),%ebp
+	movl	40(%esp),%ecx
+	movl	44(%esp),%esi
+	movl	52(%esp),%edx
+	movl	60(%esp),%edi
+	addl	92(%esp),%ebp
+	addl	104(%esp),%ecx
+	addl	108(%esp),%esi
+	addl	116(%esp),%edx
+	addl	124(%esp),%edi
+	xorl	28(%ebx),%ebp
+	xorl	40(%ebx),%ecx
+	xorl	44(%ebx),%esi
+	xorl	52(%ebx),%edx
+	xorl	60(%ebx),%edi
+	leal	64(%ebx),%ebx
+	movl	%ebp,28(%eax)
+	movl	(%esp),%ebp
+	movl	%ecx,40(%eax)
+	movl	160(%esp),%ecx
+	movl	%esi,44(%eax)
+	movl	%edx,52(%eax)
+	movl	%edi,60(%eax)
+	movl	%ebp,(%eax)
+	leal	64(%eax),%eax
+	subl	$64,%ecx
+	jnz	.L003outer_loop
+	jmp	.L006done
+.L005tail:
+	addl	112(%esp),%edx
+	addl	120(%esp),%edi
+	movl	%eax,(%esp)
+	movl	%ebp,16(%esp)
+	movl	%ecx,32(%esp)
+	movl	%esi,36(%esp)
+	movl	%edx,48(%esp)
+	movl	%edi,56(%esp)
+	movl	4(%esp),%ebp
+	movl	8(%esp),%ecx
+	movl	12(%esp),%esi
+	movl	20(%esp),%edx
+	movl	24(%esp),%edi
+	addl	$857760878,%ebp
+	addl	$2036477234,%ecx
+	addl	$1797285236,%esi
+	addl	84(%esp),%edx
+	addl	88(%esp),%edi
+	movl	%ebp,4(%esp)
+	movl	%ecx,8(%esp)
+	movl	%esi,12(%esp)
+	movl	%edx,20(%esp)
+	movl	%edi,24(%esp)
+	movl	28(%esp),%ebp
+	movl	40(%esp),%ecx
+	movl	44(%esp),%esi
+	movl	52(%esp),%edx
+	movl	60(%esp),%edi
+	addl	92(%esp),%ebp
+	addl	104(%esp),%ecx
+	addl	108(%esp),%esi
+	addl	116(%esp),%edx
+	addl	124(%esp),%edi
+	movl	%ebp,28(%esp)
+	movl	156(%esp),%ebp
+	movl	%ecx,40(%esp)
+	movl	152(%esp),%ecx
+	movl	%esi,44(%esp)
+	xorl	%esi,%esi
+	movl	%edx,52(%esp)
+	movl	%edi,60(%esp)
+	xorl	%eax,%eax
+	xorl	%edx,%edx
+.L007tail_loop:
+	movb	(%esi,%ebp,1),%al
+	movb	(%esp,%esi,1),%dl
+	leal	1(%esi),%esi
+	xorb	%dl,%al
+	movb	%al,-1(%ecx,%esi,1)
+	decl	%ebx
+	jnz	.L007tail_loop
+.L006done:
+	addl	$132,%esp
+.L000no_data:
+	popl	%edi
+	popl	%esi
+	popl	%ebx
+	popl	%ebp
+	ret
+.size	ChaCha20_ctr32,.-.L_ChaCha20_ctr32_begin
+.globl	ChaCha20_ssse3
+.hidden	ChaCha20_ssse3
+.type	ChaCha20_ssse3,@function
+.align	16
+ChaCha20_ssse3:
+.L_ChaCha20_ssse3_begin:
+	pushl	%ebp
+	pushl	%ebx
+	pushl	%esi
+	pushl	%edi
+.Lssse3_shortcut:
+	movl	20(%esp),%edi
+	movl	24(%esp),%esi
+	movl	28(%esp),%ecx
+	movl	32(%esp),%edx
+	movl	36(%esp),%ebx
+	movl	%esp,%ebp
+	subl	$524,%esp
+	andl	$-64,%esp
+	movl	%ebp,512(%esp)
+	leal	.Lssse3_data-.Lpic_point(%eax),%eax
+	movdqu	(%ebx),%xmm3
+	cmpl	$256,%ecx
+	jb	.L0081x
+	movl	%edx,516(%esp)
+	movl	%ebx,520(%esp)
+	subl	$256,%ecx
+	leal	384(%esp),%ebp
+	movdqu	(%edx),%xmm7
+	pshufd	$0,%xmm3,%xmm0
+	pshufd	$85,%xmm3,%xmm1
+	pshufd	$170,%xmm3,%xmm2
+	pshufd	$255,%xmm3,%xmm3
+	paddd	48(%eax),%xmm0
+	pshufd	$0,%xmm7,%xmm4
+	pshufd	$85,%xmm7,%xmm5
+	psubd	64(%eax),%xmm0
+	pshufd	$170,%xmm7,%xmm6
+	pshufd	$255,%xmm7,%xmm7
+	movdqa	%xmm0,64(%ebp)
+	movdqa	%xmm1,80(%ebp)
+	movdqa	%xmm2,96(%ebp)
+	movdqa	%xmm3,112(%ebp)
+	movdqu	16(%edx),%xmm3
+	movdqa	%xmm4,-64(%ebp)
+	movdqa	%xmm5,-48(%ebp)
+	movdqa	%xmm6,-32(%ebp)
+	movdqa	%xmm7,-16(%ebp)
+	movdqa	32(%eax),%xmm7
+	leal	128(%esp),%ebx
+	pshufd	$0,%xmm3,%xmm0
+	pshufd	$85,%xmm3,%xmm1
+	pshufd	$170,%xmm3,%xmm2
+	pshufd	$255,%xmm3,%xmm3
+	pshufd	$0,%xmm7,%xmm4
+	pshufd	$85,%xmm7,%xmm5
+	pshufd	$170,%xmm7,%xmm6
+	pshufd	$255,%xmm7,%xmm7
+	movdqa	%xmm0,(%ebp)
+	movdqa	%xmm1,16(%ebp)
+	movdqa	%xmm2,32(%ebp)
+	movdqa	%xmm3,48(%ebp)
+	movdqa	%xmm4,-128(%ebp)
+	movdqa	%xmm5,-112(%ebp)
+	movdqa	%xmm6,-96(%ebp)
+	movdqa	%xmm7,-80(%ebp)
+	leal	128(%esi),%esi
+	leal	128(%edi),%edi
+	jmp	.L009outer_loop
+.align	16
+.L009outer_loop:
+	movdqa	-112(%ebp),%xmm1
+	movdqa	-96(%ebp),%xmm2
+	movdqa	-80(%ebp),%xmm3
+	movdqa	-48(%ebp),%xmm5
+	movdqa	-32(%ebp),%xmm6
+	movdqa	-16(%ebp),%xmm7
+	movdqa	%xmm1,-112(%ebx)
+	movdqa	%xmm2,-96(%ebx)
+	movdqa	%xmm3,-80(%ebx)
+	movdqa	%xmm5,-48(%ebx)
+	movdqa	%xmm6,-32(%ebx)
+	movdqa	%xmm7,-16(%ebx)
+	movdqa	32(%ebp),%xmm2
+	movdqa	48(%ebp),%xmm3
+	movdqa	64(%ebp),%xmm4
+	movdqa	80(%ebp),%xmm5
+	movdqa	96(%ebp),%xmm6
+	movdqa	112(%ebp),%xmm7
+	paddd	64(%eax),%xmm4
+	movdqa	%xmm2,32(%ebx)
+	movdqa	%xmm3,48(%ebx)
+	movdqa	%xmm4,64(%ebx)
+	movdqa	%xmm5,80(%ebx)
+	movdqa	%xmm6,96(%ebx)
+	movdqa	%xmm7,112(%ebx)
+	movdqa	%xmm4,64(%ebp)
+	movdqa	-128(%ebp),%xmm0
+	movdqa	%xmm4,%xmm6
+	movdqa	-64(%ebp),%xmm3
+	movdqa	(%ebp),%xmm4
+	movdqa	16(%ebp),%xmm5
+	movl	$10,%edx
+	nop
+.align	16
+.L010loop:
+	paddd	%xmm3,%xmm0
+	movdqa	%xmm3,%xmm2
+	pxor	%xmm0,%xmm6
+	pshufb	(%eax),%xmm6
+	paddd	%xmm6,%xmm4
+	pxor	%xmm4,%xmm2
+	movdqa	-48(%ebx),%xmm3
+	movdqa	%xmm2,%xmm1
+	pslld	$12,%xmm2
+	psrld	$20,%xmm1
+	por	%xmm1,%xmm2
+	movdqa	-112(%ebx),%xmm1
+	paddd	%xmm2,%xmm0
+	movdqa	80(%ebx),%xmm7
+	pxor	%xmm0,%xmm6
+	movdqa	%xmm0,-128(%ebx)
+	pshufb	16(%eax),%xmm6
+	paddd	%xmm6,%xmm4
+	movdqa	%xmm6,64(%ebx)
+	pxor	%xmm4,%xmm2
+	paddd	%xmm3,%xmm1
+	movdqa	%xmm2,%xmm0
+	pslld	$7,%xmm2
+	psrld	$25,%xmm0
+	pxor	%xmm1,%xmm7
+	por	%xmm0,%xmm2
+	movdqa	%xmm4,(%ebx)
+	pshufb	(%eax),%xmm7
+	movdqa	%xmm2,-64(%ebx)
+	paddd	%xmm7,%xmm5
+	movdqa	32(%ebx),%xmm4
+	pxor	%xmm5,%xmm3
+	movdqa	-32(%ebx),%xmm2
+	movdqa	%xmm3,%xmm0
+	pslld	$12,%xmm3
+	psrld	$20,%xmm0
+	por	%xmm0,%xmm3
+	movdqa	-96(%ebx),%xmm0
+	paddd	%xmm3,%xmm1
+	movdqa	96(%ebx),%xmm6
+	pxor	%xmm1,%xmm7
+	movdqa	%xmm1,-112(%ebx)
+	pshufb	16(%eax),%xmm7
+	paddd	%xmm7,%xmm5
+	movdqa	%xmm7,80(%ebx)
+	pxor	%xmm5,%xmm3
+	paddd	%xmm2,%xmm0
+	movdqa	%xmm3,%xmm1
+	pslld	$7,%xmm3
+	psrld	$25,%xmm1
+	pxor	%xmm0,%xmm6
+	por	%xmm1,%xmm3
+	movdqa	%xmm5,16(%ebx)
+	pshufb	(%eax),%xmm6
+	movdqa	%xmm3,-48(%ebx)
+	paddd	%xmm6,%xmm4
+	movdqa	48(%ebx),%xmm5
+	pxor	%xmm4,%xmm2
+	movdqa	-16(%ebx),%xmm3
+	movdqa	%xmm2,%xmm1
+	pslld	$12,%xmm2
+	psrld	$20,%xmm1
+	por	%xmm1,%xmm2
+	movdqa	-80(%ebx),%xmm1
+	paddd	%xmm2,%xmm0
+	movdqa	112(%ebx),%xmm7
+	pxor	%xmm0,%xmm6
+	movdqa	%xmm0,-96(%ebx)
+	pshufb	16(%eax),%xmm6
+	paddd	%xmm6,%xmm4
+	movdqa	%xmm6,96(%ebx)
+	pxor	%xmm4,%xmm2
+	paddd	%xmm3,%xmm1
+	movdqa	%xmm2,%xmm0
+	pslld	$7,%xmm2
+	psrld	$25,%xmm0
+	pxor	%xmm1,%xmm7
+	por	%xmm0,%xmm2
+	pshufb	(%eax),%xmm7
+	movdqa	%xmm2,-32(%ebx)
+	paddd	%xmm7,%xmm5
+	pxor	%xmm5,%xmm3
+	movdqa	-48(%ebx),%xmm2
+	movdqa	%xmm3,%xmm0
+	pslld	$12,%xmm3
+	psrld	$20,%xmm0
+	por	%xmm0,%xmm3
+	movdqa	-128(%ebx),%xmm0
+	paddd	%xmm3,%xmm1
+	pxor	%xmm1,%xmm7
+	movdqa	%xmm1,-80(%ebx)
+	pshufb	16(%eax),%xmm7
+	paddd	%xmm7,%xmm5
+	movdqa	%xmm7,%xmm6
+	pxor	%xmm5,%xmm3
+	paddd	%xmm2,%xmm0
+	movdqa	%xmm3,%xmm1
+	pslld	$7,%xmm3
+	psrld	$25,%xmm1
+	pxor	%xmm0,%xmm6
+	por	%xmm1,%xmm3
+	pshufb	(%eax),%xmm6
+	movdqa	%xmm3,-16(%ebx)
+	paddd	%xmm6,%xmm4
+	pxor	%xmm4,%xmm2
+	movdqa	-32(%ebx),%xmm3
+	movdqa	%xmm2,%xmm1
+	pslld	$12,%xmm2
+	psrld	$20,%xmm1
+	por	%xmm1,%xmm2
+	movdqa	-112(%ebx),%xmm1
+	paddd	%xmm2,%xmm0
+	movdqa	64(%ebx),%xmm7
+	pxor	%xmm0,%xmm6
+	movdqa	%xmm0,-128(%ebx)
+	pshufb	16(%eax),%xmm6
+	paddd	%xmm6,%xmm4
+	movdqa	%xmm6,112(%ebx)
+	pxor	%xmm4,%xmm2
+	paddd	%xmm3,%xmm1
+	movdqa	%xmm2,%xmm0
+	pslld	$7,%xmm2
+	psrld	$25,%xmm0
+	pxor	%xmm1,%xmm7
+	por	%xmm0,%xmm2
+	movdqa	%xmm4,32(%ebx)
+	pshufb	(%eax),%xmm7
+	movdqa	%xmm2,-48(%ebx)
+	paddd	%xmm7,%xmm5
+	movdqa	(%ebx),%xmm4
+	pxor	%xmm5,%xmm3
+	movdqa	-16(%ebx),%xmm2
+	movdqa	%xmm3,%xmm0
+	pslld	$12,%xmm3
+	psrld	$20,%xmm0
+	por	%xmm0,%xmm3
+	movdqa	-96(%ebx),%xmm0
+	paddd	%xmm3,%xmm1
+	movdqa	80(%ebx),%xmm6
+	pxor	%xmm1,%xmm7
+	movdqa	%xmm1,-112(%ebx)
+	pshufb	16(%eax),%xmm7
+	paddd	%xmm7,%xmm5
+	movdqa	%xmm7,64(%ebx)
+	pxor	%xmm5,%xmm3
+	paddd	%xmm2,%xmm0
+	movdqa	%xmm3,%xmm1
+	pslld	$7,%xmm3
+	psrld	$25,%xmm1
+	pxor	%xmm0,%xmm6
+	por	%xmm1,%xmm3
+	movdqa	%xmm5,48(%ebx)
+	pshufb	(%eax),%xmm6
+	movdqa	%xmm3,-32(%ebx)
+	paddd	%xmm6,%xmm4
+	movdqa	16(%ebx),%xmm5
+	pxor	%xmm4,%xmm2
+	movdqa	-64(%ebx),%xmm3
+	movdqa	%xmm2,%xmm1
+	pslld	$12,%xmm2
+	psrld	$20,%xmm1
+	por	%xmm1,%xmm2
+	movdqa	-80(%ebx),%xmm1
+	paddd	%xmm2,%xmm0
+	movdqa	96(%ebx),%xmm7
+	pxor	%xmm0,%xmm6
+	movdqa	%xmm0,-96(%ebx)
+	pshufb	16(%eax),%xmm6
+	paddd	%xmm6,%xmm4
+	movdqa	%xmm6,80(%ebx)
+	pxor	%xmm4,%xmm2
+	paddd	%xmm3,%xmm1
+	movdqa	%xmm2,%xmm0
+	pslld	$7,%xmm2
+	psrld	$25,%xmm0
+	pxor	%xmm1,%xmm7
+	por	%xmm0,%xmm2
+	pshufb	(%eax),%xmm7
+	movdqa	%xmm2,-16(%ebx)
+	paddd	%xmm7,%xmm5
+	pxor	%xmm5,%xmm3
+	movdqa	%xmm3,%xmm0
+	pslld	$12,%xmm3
+	psrld	$20,%xmm0
+	por	%xmm0,%xmm3
+	movdqa	-128(%ebx),%xmm0
+	paddd	%xmm3,%xmm1
+	movdqa	64(%ebx),%xmm6
+	pxor	%xmm1,%xmm7
+	movdqa	%xmm1,-80(%ebx)
+	pshufb	16(%eax),%xmm7
+	paddd	%xmm7,%xmm5
+	movdqa	%xmm7,96(%ebx)
+	pxor	%xmm5,%xmm3
+	movdqa	%xmm3,%xmm1
+	pslld	$7,%xmm3
+	psrld	$25,%xmm1
+	por	%xmm1,%xmm3
+	decl	%edx
+	jnz	.L010loop
+	movdqa	%xmm3,-64(%ebx)
+	movdqa	%xmm4,(%ebx)
+	movdqa	%xmm5,16(%ebx)
+	movdqa	%xmm6,64(%ebx)
+	movdqa	%xmm7,96(%ebx)
+	movdqa	-112(%ebx),%xmm1
+	movdqa	-96(%ebx),%xmm2
+	movdqa	-80(%ebx),%xmm3
+	paddd	-128(%ebp),%xmm0
+	paddd	-112(%ebp),%xmm1
+	paddd	-96(%ebp),%xmm2
+	paddd	-80(%ebp),%xmm3
+	movdqa	%xmm0,%xmm6
+	punpckldq	%xmm1,%xmm0
+	movdqa	%xmm2,%xmm7
+	punpckldq	%xmm3,%xmm2
+	punpckhdq	%xmm1,%xmm6
+	punpckhdq	%xmm3,%xmm7
+	movdqa	%xmm0,%xmm1
+	punpcklqdq	%xmm2,%xmm0
+	movdqa	%xmm6,%xmm3
+	punpcklqdq	%xmm7,%xmm6
+	punpckhqdq	%xmm2,%xmm1
+	punpckhqdq	%xmm7,%xmm3
+	movdqu	-128(%esi),%xmm4
+	movdqu	-64(%esi),%xmm5
+	movdqu	(%esi),%xmm2
+	movdqu	64(%esi),%xmm7
+	leal	16(%esi),%esi
+	pxor	%xmm0,%xmm4
+	movdqa	-64(%ebx),%xmm0
+	pxor	%xmm1,%xmm5
+	movdqa	-48(%ebx),%xmm1
+	pxor	%xmm2,%xmm6
+	movdqa	-32(%ebx),%xmm2
+	pxor	%xmm3,%xmm7
+	movdqa	-16(%ebx),%xmm3
+	movdqu	%xmm4,-128(%edi)
+	movdqu	%xmm5,-64(%edi)
+	movdqu	%xmm6,(%edi)
+	movdqu	%xmm7,64(%edi)
+	leal	16(%edi),%edi
+	paddd	-64(%ebp),%xmm0
+	paddd	-48(%ebp),%xmm1
+	paddd	-32(%ebp),%xmm2
+	paddd	-16(%ebp),%xmm3
+	movdqa	%xmm0,%xmm6
+	punpckldq	%xmm1,%xmm0
+	movdqa	%xmm2,%xmm7
+	punpckldq	%xmm3,%xmm2
+	punpckhdq	%xmm1,%xmm6
+	punpckhdq	%xmm3,%xmm7
+	movdqa	%xmm0,%xmm1
+	punpcklqdq	%xmm2,%xmm0
+	movdqa	%xmm6,%xmm3
+	punpcklqdq	%xmm7,%xmm6
+	punpckhqdq	%xmm2,%xmm1
+	punpckhqdq	%xmm7,%xmm3
+	movdqu	-128(%esi),%xmm4
+	movdqu	-64(%esi),%xmm5
+	movdqu	(%esi),%xmm2
+	movdqu	64(%esi),%xmm7
+	leal	16(%esi),%esi
+	pxor	%xmm0,%xmm4
+	movdqa	(%ebx),%xmm0
+	pxor	%xmm1,%xmm5
+	movdqa	16(%ebx),%xmm1
+	pxor	%xmm2,%xmm6
+	movdqa	32(%ebx),%xmm2
+	pxor	%xmm3,%xmm7
+	movdqa	48(%ebx),%xmm3
+	movdqu	%xmm4,-128(%edi)
+	movdqu	%xmm5,-64(%edi)
+	movdqu	%xmm6,(%edi)
+	movdqu	%xmm7,64(%edi)
+	leal	16(%edi),%edi
+	paddd	(%ebp),%xmm0
+	paddd	16(%ebp),%xmm1
+	paddd	32(%ebp),%xmm2
+	paddd	48(%ebp),%xmm3
+	movdqa	%xmm0,%xmm6
+	punpckldq	%xmm1,%xmm0
+	movdqa	%xmm2,%xmm7
+	punpckldq	%xmm3,%xmm2
+	punpckhdq	%xmm1,%xmm6
+	punpckhdq	%xmm3,%xmm7
+	movdqa	%xmm0,%xmm1
+	punpcklqdq	%xmm2,%xmm0
+	movdqa	%xmm6,%xmm3
+	punpcklqdq	%xmm7,%xmm6
+	punpckhqdq	%xmm2,%xmm1
+	punpckhqdq	%xmm7,%xmm3
+	movdqu	-128(%esi),%xmm4
+	movdqu	-64(%esi),%xmm5
+	movdqu	(%esi),%xmm2
+	movdqu	64(%esi),%xmm7
+	leal	16(%esi),%esi
+	pxor	%xmm0,%xmm4
+	movdqa	64(%ebx),%xmm0
+	pxor	%xmm1,%xmm5
+	movdqa	80(%ebx),%xmm1
+	pxor	%xmm2,%xmm6
+	movdqa	96(%ebx),%xmm2
+	pxor	%xmm3,%xmm7
+	movdqa	112(%ebx),%xmm3
+	movdqu	%xmm4,-128(%edi)
+	movdqu	%xmm5,-64(%edi)
+	movdqu	%xmm6,(%edi)
+	movdqu	%xmm7,64(%edi)
+	leal	16(%edi),%edi
+	paddd	64(%ebp),%xmm0
+	paddd	80(%ebp),%xmm1
+	paddd	96(%ebp),%xmm2
+	paddd	112(%ebp),%xmm3
+	movdqa	%xmm0,%xmm6
+	punpckldq	%xmm1,%xmm0
+	movdqa	%xmm2,%xmm7
+	punpckldq	%xmm3,%xmm2
+	punpckhdq	%xmm1,%xmm6
+	punpckhdq	%xmm3,%xmm7
+	movdqa	%xmm0,%xmm1
+	punpcklqdq	%xmm2,%xmm0
+	movdqa	%xmm6,%xmm3
+	punpcklqdq	%xmm7,%xmm6
+	punpckhqdq	%xmm2,%xmm1
+	punpckhqdq	%xmm7,%xmm3
+	movdqu	-128(%esi),%xmm4
+	movdqu	-64(%esi),%xmm5
+	movdqu	(%esi),%xmm2
+	movdqu	64(%esi),%xmm7
+	leal	208(%esi),%esi
+	pxor	%xmm0,%xmm4
+	pxor	%xmm1,%xmm5
+	pxor	%xmm2,%xmm6
+	pxor	%xmm3,%xmm7
+	movdqu	%xmm4,-128(%edi)
+	movdqu	%xmm5,-64(%edi)
+	movdqu	%xmm6,(%edi)
+	movdqu	%xmm7,64(%edi)
+	leal	208(%edi),%edi
+	subl	$256,%ecx
+	jnc	.L009outer_loop
+	addl	$256,%ecx
+	jz	.L011done
+	movl	520(%esp),%ebx
+	leal	-128(%esi),%esi
+	movl	516(%esp),%edx
+	leal	-128(%edi),%edi
+	movd	64(%ebp),%xmm2
+	movdqu	(%ebx),%xmm3
+	paddd	96(%eax),%xmm2
+	pand	112(%eax),%xmm3
+	por	%xmm2,%xmm3
+.L0081x:
+	movdqa	32(%eax),%xmm0
+	movdqu	(%edx),%xmm1
+	movdqu	16(%edx),%xmm2
+	movdqa	(%eax),%xmm6
+	movdqa	16(%eax),%xmm7
+	movl	%ebp,48(%esp)
+	movdqa	%xmm0,(%esp)
+	movdqa	%xmm1,16(%esp)
+	movdqa	%xmm2,32(%esp)
+	movdqa	%xmm3,48(%esp)
+	movl	$10,%edx
+	jmp	.L012loop1x
+.align	16
+.L013outer1x:
+	movdqa	80(%eax),%xmm3
+	movdqa	(%esp),%xmm0
+	movdqa	16(%esp),%xmm1
+	movdqa	32(%esp),%xmm2
+	paddd	48(%esp),%xmm3
+	movl	$10,%edx
+	movdqa	%xmm3,48(%esp)
+	jmp	.L012loop1x
+.align	16
+.L012loop1x:
+	paddd	%xmm1,%xmm0
+	pxor	%xmm0,%xmm3
+.byte	102,15,56,0,222
+	paddd	%xmm3,%xmm2
+	pxor	%xmm2,%xmm1
+	movdqa	%xmm1,%xmm4
+	psrld	$20,%xmm1
+	pslld	$12,%xmm4
+	por	%xmm4,%xmm1
+	paddd	%xmm1,%xmm0
+	pxor	%xmm0,%xmm3
+.byte	102,15,56,0,223
+	paddd	%xmm3,%xmm2
+	pxor	%xmm2,%xmm1
+	movdqa	%xmm1,%xmm4
+	psrld	$25,%xmm1
+	pslld	$7,%xmm4
+	por	%xmm4,%xmm1
+	pshufd	$78,%xmm2,%xmm2
+	pshufd	$57,%xmm1,%xmm1
+	pshufd	$147,%xmm3,%xmm3
+	nop
+	paddd	%xmm1,%xmm0
+	pxor	%xmm0,%xmm3
+.byte	102,15,56,0,222
+	paddd	%xmm3,%xmm2
+	pxor	%xmm2,%xmm1
+	movdqa	%xmm1,%xmm4
+	psrld	$20,%xmm1
+	pslld	$12,%xmm4
+	por	%xmm4,%xmm1
+	paddd	%xmm1,%xmm0
+	pxor	%xmm0,%xmm3
+.byte	102,15,56,0,223
+	paddd	%xmm3,%xmm2
+	pxor	%xmm2,%xmm1
+	movdqa	%xmm1,%xmm4
+	psrld	$25,%xmm1
+	pslld	$7,%xmm4
+	por	%xmm4,%xmm1
+	pshufd	$78,%xmm2,%xmm2
+	pshufd	$147,%xmm1,%xmm1
+	pshufd	$57,%xmm3,%xmm3
+	decl	%edx
+	jnz	.L012loop1x
+	paddd	(%esp),%xmm0
+	paddd	16(%esp),%xmm1
+	paddd	32(%esp),%xmm2
+	paddd	48(%esp),%xmm3
+	cmpl	$64,%ecx
+	jb	.L014tail
+	movdqu	(%esi),%xmm4
+	movdqu	16(%esi),%xmm5
+	pxor	%xmm4,%xmm0
+	movdqu	32(%esi),%xmm4
+	pxor	%xmm5,%xmm1
+	movdqu	48(%esi),%xmm5
+	pxor	%xmm4,%xmm2
+	pxor	%xmm5,%xmm3
+	leal	64(%esi),%esi
+	movdqu	%xmm0,(%edi)
+	movdqu	%xmm1,16(%edi)
+	movdqu	%xmm2,32(%edi)
+	movdqu	%xmm3,48(%edi)
+	leal	64(%edi),%edi
+	subl	$64,%ecx
+	jnz	.L013outer1x
+	jmp	.L011done
+.L014tail:
+	movdqa	%xmm0,(%esp)
+	movdqa	%xmm1,16(%esp)
+	movdqa	%xmm2,32(%esp)
+	movdqa	%xmm3,48(%esp)
+	xorl	%eax,%eax
+	xorl	%edx,%edx
+	xorl	%ebp,%ebp
+.L015tail_loop:
+	movb	(%esp,%ebp,1),%al
+	movb	(%esi,%ebp,1),%dl
+	leal	1(%ebp),%ebp
+	xorb	%dl,%al
+	movb	%al,-1(%edi,%ebp,1)
+	decl	%ecx
+	jnz	.L015tail_loop
+.L011done:
+	movl	512(%esp),%esp
+	popl	%edi
+	popl	%esi
+	popl	%ebx
+	popl	%ebp
+	ret
+.size	ChaCha20_ssse3,.-.L_ChaCha20_ssse3_begin
+.align	64
+.Lssse3_data:
+.byte	2,3,0,1,6,7,4,5,10,11,8,9,14,15,12,13
+.byte	3,0,1,2,7,4,5,6,11,8,9,10,15,12,13,14
+.long	1634760805,857760878,2036477234,1797285236
+.long	0,1,2,3
+.long	4,4,4,4
+.long	1,0,0,0
+.long	4,0,0,0
+.long	0,-1,-1,-1
+.align	64
+.byte	67,104,97,67,104,97,50,48,32,102,111,114,32,120,56,54
+.byte	44,32,67,82,89,80,84,79,71,65,77,83,32,98,121,32
+.byte	60,97,112,112,114,111,64,111,112,101,110,115,115,108,46,111
+.byte	114,103,62,0
+#endif

third_party/boringssl/linux-x86/crypto/cpu-x86-asm.S

diff --git a/third_party/boringssl/linux-x86/crypto/cpu-x86-asm.S b/third_party/boringssl/linux-x86/crypto/cpu-x86-asm.S
deleted file mode 100644
index 24a8dd4..0000000
--- a/third_party/boringssl/linux-x86/crypto/cpu-x86-asm.S
+++ /dev/null
@@ -1,322 +0,0 @@
-#if defined(__i386__)
-.file	"crypto/cpu-x86-asm.S"
-.text
-.globl	OPENSSL_ia32_cpuid
-.hidden	OPENSSL_ia32_cpuid
-.type	OPENSSL_ia32_cpuid,@function
-.align	16
-OPENSSL_ia32_cpuid:
-.L_OPENSSL_ia32_cpuid_begin:
-	pushl	%ebp
-	pushl	%ebx
-	pushl	%esi
-	pushl	%edi
-	xorl	%edx,%edx
-	pushfl
-	popl	%eax
-	movl	%eax,%ecx
-	xorl	$2097152,%eax
-	pushl	%eax
-	popfl
-	pushfl
-	popl	%eax
-	xorl	%eax,%ecx
-	xorl	%eax,%eax
-	btl	$21,%ecx
-	jnc	.L000nocpuid
-	movl	20(%esp),%esi
-	movl	%eax,8(%esi)
-	.byte	0x0f,0xa2
-	movl	%eax,%edi
-	xorl	%eax,%eax
-	cmpl	$1970169159,%ebx
-	setne	%al
-	movl	%eax,%ebp
-	cmpl	$1231384169,%edx
-	setne	%al
-	orl	%eax,%ebp
-	cmpl	$1818588270,%ecx
-	setne	%al
-	orl	%eax,%ebp
-	jz	.L001intel
-	cmpl	$1752462657,%ebx
-	setne	%al
-	movl	%eax,%esi
-	cmpl	$1769238117,%edx
-	setne	%al
-	orl	%eax,%esi
-	cmpl	$1145913699,%ecx
-	setne	%al
-	orl	%eax,%esi
-	jnz	.L001intel
-	movl	$2147483648,%eax
-	.byte	0x0f,0xa2
-	cmpl	$2147483649,%eax
-	jb	.L001intel
-	movl	%eax,%esi
-	movl	$2147483649,%eax
-	.byte	0x0f,0xa2
-	orl	%ecx,%ebp
-	andl	$2049,%ebp
-	cmpl	$2147483656,%esi
-	jb	.L001intel
-	movl	$2147483656,%eax
-	.byte	0x0f,0xa2
-	movzbl	%cl,%esi
-	incl	%esi
-	movl	$1,%eax
-	xorl	%ecx,%ecx
-	.byte	0x0f,0xa2
-	btl	$28,%edx
-	jnc	.L002generic
-	shrl	$16,%ebx
-	andl	$255,%ebx
-	cmpl	%esi,%ebx
-	ja	.L002generic
-	andl	$4026531839,%edx
-	jmp	.L002generic
-.L001intel:
-	cmpl	$7,%edi
-	jb	.L003cacheinfo
-	movl	20(%esp),%esi
-	movl	$7,%eax
-	xorl	%ecx,%ecx
-	.byte	0x0f,0xa2
-	movl	%ebx,8(%esi)
-.L003cacheinfo:
-	cmpl	$4,%edi
-	movl	$-1,%edi
-	jb	.L004nocacheinfo
-	movl	$4,%eax
-	movl	$0,%ecx
-	.byte	0x0f,0xa2
-	movl	%eax,%edi
-	shrl	$14,%edi
-	andl	$4095,%edi
-.L004nocacheinfo:
-	movl	$1,%eax
-	xorl	%ecx,%ecx
-	.byte	0x0f,0xa2
-	andl	$3220176895,%edx
-	cmpl	$0,%ebp
-	jne	.L005notintel
-	orl	$1073741824,%edx
-.L005notintel:
-	btl	$28,%edx
-	jnc	.L002generic
-	andl	$4026531839,%edx
-	cmpl	$0,%edi
-	je	.L002generic
-	orl	$268435456,%edx
-	shrl	$16,%ebx
-	cmpb	$1,%bl
-	ja	.L002generic
-	andl	$4026531839,%edx
-.L002generic:
-	andl	$2048,%ebp
-	andl	$4294965247,%ecx
-	movl	%edx,%esi
-	orl	%ecx,%ebp
-	btl	$27,%ecx
-	jnc	.L006clear_avx
-	xorl	%ecx,%ecx
-.byte	15,1,208
-	andl	$6,%eax
-	cmpl	$6,%eax
-	je	.L007done
-	cmpl	$2,%eax
-	je	.L006clear_avx
-.L008clear_xmm:
-	andl	$4261412861,%ebp
-	andl	$4278190079,%esi
-.L006clear_avx:
-	andl	$4026525695,%ebp
-	movl	20(%esp),%edi
-	andl	$4294967263,8(%edi)
-.L007done:
-	movl	%esi,%eax
-	movl	%ebp,%edx
-.L000nocpuid:
-	popl	%edi
-	popl	%esi
-	popl	%ebx
-	popl	%ebp
-	ret
-.size	OPENSSL_ia32_cpuid,.-.L_OPENSSL_ia32_cpuid_begin
-.globl	OPENSSL_rdtsc
-.hidden	OPENSSL_rdtsc
-.type	OPENSSL_rdtsc,@function
-.align	16
-OPENSSL_rdtsc:
-.L_OPENSSL_rdtsc_begin:
-	xorl	%eax,%eax
-	xorl	%edx,%edx
-	call	.L009PIC_me_up
-.L009PIC_me_up:
-	popl	%ecx
-	leal	OPENSSL_ia32cap_P-.L009PIC_me_up(%ecx),%ecx
-	btl	$4,(%ecx)
-	jnc	.L010notsc
-	.byte	0x0f,0x31
-.L010notsc:
-	ret
-.size	OPENSSL_rdtsc,.-.L_OPENSSL_rdtsc_begin
-.globl	OPENSSL_instrument_halt
-.hidden	OPENSSL_instrument_halt
-.type	OPENSSL_instrument_halt,@function
-.align	16
-OPENSSL_instrument_halt:
-.L_OPENSSL_instrument_halt_begin:
-	call	.L011PIC_me_up
-.L011PIC_me_up:
-	popl	%ecx
-	leal	OPENSSL_ia32cap_P-.L011PIC_me_up(%ecx),%ecx
-	btl	$4,(%ecx)
-	jnc	.L012nohalt
-.long	2421723150
-	andl	$3,%eax
-	jnz	.L012nohalt
-	pushfl
-	popl	%eax
-	btl	$9,%eax
-	jnc	.L012nohalt
-	.byte	0x0f,0x31
-	pushl	%edx
-	pushl	%eax
-	hlt
-	.byte	0x0f,0x31
-	subl	(%esp),%eax
-	sbbl	4(%esp),%edx
-	addl	$8,%esp
-	ret
-.L012nohalt:
-	xorl	%eax,%eax
-	xorl	%edx,%edx
-	ret
-.size	OPENSSL_instrument_halt,.-.L_OPENSSL_instrument_halt_begin
-.globl	OPENSSL_far_spin
-.hidden	OPENSSL_far_spin
-.type	OPENSSL_far_spin,@function
-.align	16
-OPENSSL_far_spin:
-.L_OPENSSL_far_spin_begin:
-	pushfl
-	popl	%eax
-	btl	$9,%eax
-	jnc	.L013nospin
-	movl	4(%esp),%eax
-	movl	8(%esp),%ecx
-.long	2430111262
-	xorl	%eax,%eax
-	movl	(%ecx),%edx
-	jmp	.L014spin
-.align	16
-.L014spin:
-	incl	%eax
-	cmpl	(%ecx),%edx
-	je	.L014spin
-.long	529567888
-	ret
-.L013nospin:
-	xorl	%eax,%eax
-	xorl	%edx,%edx
-	ret
-.size	OPENSSL_far_spin,.-.L_OPENSSL_far_spin_begin
-.globl	OPENSSL_wipe_cpu
-.hidden	OPENSSL_wipe_cpu
-.type	OPENSSL_wipe_cpu,@function
-.align	16
-OPENSSL_wipe_cpu:
-.L_OPENSSL_wipe_cpu_begin:
-	xorl	%eax,%eax
-	xorl	%edx,%edx
-	call	.L015PIC_me_up
-.L015PIC_me_up:
-	popl	%ecx
-	leal	OPENSSL_ia32cap_P-.L015PIC_me_up(%ecx),%ecx
-	movl	(%ecx),%ecx
-	btl	$1,(%ecx)
-	jnc	.L016no_x87
-	andl	$83886080,%ecx
-	cmpl	$83886080,%ecx
-	jne	.L017no_sse2
-	pxor	%xmm0,%xmm0
-	pxor	%xmm1,%xmm1
-	pxor	%xmm2,%xmm2
-	pxor	%xmm3,%xmm3
-	pxor	%xmm4,%xmm4
-	pxor	%xmm5,%xmm5
-	pxor	%xmm6,%xmm6
-	pxor	%xmm7,%xmm7
-.L017no_sse2:
-.long	4007259865,4007259865,4007259865,4007259865,2430851995
-.L016no_x87:
-	leal	4(%esp),%eax
-	ret
-.size	OPENSSL_wipe_cpu,.-.L_OPENSSL_wipe_cpu_begin
-.globl	OPENSSL_atomic_add
-.hidden	OPENSSL_atomic_add
-.type	OPENSSL_atomic_add,@function
-.align	16
-OPENSSL_atomic_add:
-.L_OPENSSL_atomic_add_begin:
-	movl	4(%esp),%edx
-	movl	8(%esp),%ecx
-	pushl	%ebx
-	nop
-	movl	(%edx),%eax
-.L018spin:
-	leal	(%eax,%ecx,1),%ebx
-	nop
-.long	447811568
-	jne	.L018spin
-	movl	%ebx,%eax
-	popl	%ebx
-	ret
-.size	OPENSSL_atomic_add,.-.L_OPENSSL_atomic_add_begin
-.globl	OPENSSL_indirect_call
-.hidden	OPENSSL_indirect_call
-.type	OPENSSL_indirect_call,@function
-.align	16
-OPENSSL_indirect_call:
-.L_OPENSSL_indirect_call_begin:
-	pushl	%ebp
-	movl	%esp,%ebp
-	subl	$28,%esp
-	movl	12(%ebp),%ecx
-	movl	%ecx,(%esp)
-	movl	16(%ebp),%edx
-	movl	%edx,4(%esp)
-	movl	20(%ebp),%eax
-	movl	%eax,8(%esp)
-	movl	24(%ebp),%eax
-	movl	%eax,12(%esp)
-	movl	28(%ebp),%eax
-	movl	%eax,16(%esp)
-	movl	32(%ebp),%eax
-	movl	%eax,20(%esp)
-	movl	36(%ebp),%eax
-	movl	%eax,24(%esp)
-	call	*8(%ebp)
-	movl	%ebp,%esp
-	popl	%ebp
-	ret
-.size	OPENSSL_indirect_call,.-.L_OPENSSL_indirect_call_begin
-.globl	OPENSSL_ia32_rdrand
-.hidden	OPENSSL_ia32_rdrand
-.type	OPENSSL_ia32_rdrand,@function
-.align	16
-OPENSSL_ia32_rdrand:
-.L_OPENSSL_ia32_rdrand_begin:
-	movl	$8,%ecx
-.L019loop:
-.byte	15,199,240
-	jc	.L020break
-	loop	.L019loop
-.L020break:
-	cmpl	$0,%eax
-	cmovel	%ecx,%eax
-	ret
-.size	OPENSSL_ia32_rdrand,.-.L_OPENSSL_ia32_rdrand_begin
-.hidden	OPENSSL_ia32cap_P
-#endif

third_party/boringssl/linux-x86/crypto/rc4/rc4-586.S

diff --git a/third_party/boringssl/linux-x86/crypto/rc4/rc4-586.S b/third_party/boringssl/linux-x86/crypto/rc4/rc4-586.S
index a5cce47..d245589 100644
--- a/third_party/boringssl/linux-x86/crypto/rc4/rc4-586.S
+++ b/third_party/boringssl/linux-x86/crypto/rc4/rc4-586.S
@@ -347,39 +347,4 @@
 	popl	%ebp
 	ret
 .size	asm_RC4_set_key,.-.L_asm_RC4_set_key_begin
-.globl	RC4_options
-.hidden	RC4_options
-.type	RC4_options,@function
-.align	16
-RC4_options:
-.L_RC4_options_begin:
-	call	.L018pic_point
-.L018pic_point:
-	popl	%eax
-	leal	.L019opts-.L018pic_point(%eax),%eax
-	call	.L020PIC_me_up
-.L020PIC_me_up:
-	popl	%edx
-	leal	OPENSSL_ia32cap_P-.L020PIC_me_up(%edx),%edx
-	movl	(%edx),%edx
-	btl	$20,%edx
-	jc	.L0211xchar
-	btl	$26,%edx
-	jnc	.L022ret
-	addl	$25,%eax
-	ret
-.L0211xchar:
-	addl	$12,%eax
-.L022ret:
-	ret
-.align	64
-.L019opts:
-.byte	114,99,52,40,52,120,44,105,110,116,41,0
-.byte	114,99,52,40,49,120,44,99,104,97,114,41,0
-.byte	114,99,52,40,56,120,44,109,109,120,41,0
-.byte	82,67,52,32,102,111,114,32,120,56,54,44,32,67,82,89
-.byte	80,84,79,71,65,77,83,32,98,121,32,60,97,112,112,114
-.byte	111,64,111,112,101,110,115,115,108,46,111,114,103,62,0
-.align	64
-.size	RC4_options,.-.L_RC4_options_begin
 #endif

third_party/boringssl/linux-x86/crypto/sha/sha1-586.S

diff --git a/third_party/boringssl/linux-x86/crypto/sha/sha1-586.S b/third_party/boringssl/linux-x86/crypto/sha/sha1-586.S
index 808ccac..58d0bc1 100644
--- a/third_party/boringssl/linux-x86/crypto/sha/sha1-586.S
+++ b/third_party/boringssl/linux-x86/crypto/sha/sha1-586.S
@@ -23,8 +23,11 @@
 	movl	8(%esi),%ecx
 	testl	$16777216,%eax
 	jz	.L001x86
-	testl	$536870912,%ecx
-	jnz	.Lshaext_shortcut
+	andl	$268435456,%edx
+	andl	$1073741824,%eax
+	orl	%edx,%eax
+	cmpl	$1342177280,%eax
+	je	.Lavx_shortcut
 	jmp	.Lssse3_shortcut
 .align	16
 .L001x86:
@@ -1393,177 +1396,6 @@
 	popl	%ebp
 	ret
 .size	sha1_block_data_order,.-.L_sha1_block_data_order_begin
-.hidden	_sha1_block_data_order_shaext
-.type	_sha1_block_data_order_shaext,@function
-.align	16
-_sha1_block_data_order_shaext:
-	pushl	%ebp
-	pushl	%ebx
-	pushl	%esi
-	pushl	%edi
-	call	.L003pic_point
-.L003pic_point:
-	popl	%ebp
-	leal	.LK_XX_XX-.L003pic_point(%ebp),%ebp
-.Lshaext_shortcut:
-	movl	20(%esp),%edi
-	movl	%esp,%ebx
-	movl	24(%esp),%esi
-	movl	28(%esp),%ecx
-	subl	$32,%esp
-	movdqu	(%edi),%xmm0
-	movd	16(%edi),%xmm1
-	andl	$-32,%esp
-	movdqa	80(%ebp),%xmm3
-	movdqu	(%esi),%xmm4
-	pshufd	$27,%xmm0,%xmm0
-	movdqu	16(%esi),%xmm5
-	pshufd	$27,%xmm1,%xmm1
-	movdqu	32(%esi),%xmm6
-.byte	102,15,56,0,227
-	movdqu	48(%esi),%xmm7
-.byte	102,15,56,0,235
-.byte	102,15,56,0,243
-.byte	102,15,56,0,251
-	jmp	.L004loop_shaext
-.align	16
-.L004loop_shaext:
-	decl	%ecx
-	leal	64(%esi),%eax
-	movdqa	%xmm1,(%esp)
-	paddd	%xmm4,%xmm1
-	cmovnel	%eax,%esi
-	movdqa	%xmm0,16(%esp)
-.byte	15,56,201,229
-	movdqa	%xmm0,%xmm2
-.byte	15,58,204,193,0
-.byte	15,56,200,213
-	pxor	%xmm6,%xmm4
-.byte	15,56,201,238
-.byte	15,56,202,231
-	movdqa	%xmm0,%xmm1
-.byte	15,58,204,194,0
-.byte	15,56,200,206
-	pxor	%xmm7,%xmm5
-.byte	15,56,202,236
-.byte	15,56,201,247
-	movdqa	%xmm0,%xmm2
-.byte	15,58,204,193,0
-.byte	15,56,200,215
-	pxor	%xmm4,%xmm6
-.byte	15,56,201,252
-.byte	15,56,202,245
-	movdqa	%xmm0,%xmm1
-.byte	15,58,204,194,0
-.byte	15,56,200,204
-	pxor	%xmm5,%xmm7
-.byte	15,56,202,254
-.byte	15,56,201,229
-	movdqa	%xmm0,%xmm2
-.byte	15,58,204,193,0
-.byte	15,56,200,213
-	pxor	%xmm6,%xmm4
-.byte	15,56,201,238
-.byte	15,56,202,231
-	movdqa	%xmm0,%xmm1
-.byte	15,58,204,194,1
-.byte	15,56,200,206
-	pxor	%xmm7,%xmm5
-.byte	15,56,202,236
-.byte	15,56,201,247
-	movdqa	%xmm0,%xmm2
-.byte	15,58,204,193,1
-.byte	15,56,200,215
-	pxor	%xmm4,%xmm6
-.byte	15,56,201,252
-.byte	15,56,202,245
-	movdqa	%xmm0,%xmm1
-.byte	15,58,204,194,1
-.byte	15,56,200,204
-	pxor	%xmm5,%xmm7
-.byte	15,56,202,254
-.byte	15,56,201,229
-	movdqa	%xmm0,%xmm2
-.byte	15,58,204,193,1
-.byte	15,56,200,213
-	pxor	%xmm6,%xmm4
-.byte	15,56,201,238
-.byte	15,56,202,231
-	movdqa	%xmm0,%xmm1
-.byte	15,58,204,194,1
-.byte	15,56,200,206
-	pxor	%xmm7,%xmm5
-.byte	15,56,202,236
-.byte	15,56,201,247
-	movdqa	%xmm0,%xmm2
-.byte	15,58,204,193,2
-.byte	15,56,200,215
-	pxor	%xmm4,%xmm6
-.byte	15,56,201,252
-.byte	15,56,202,245
-	movdqa	%xmm0,%xmm1
-.byte	15,58,204,194,2
-.byte	15,56,200,204
-	pxor	%xmm5,%xmm7
-.byte	15,56,202,254
-.byte	15,56,201,229
-	movdqa	%xmm0,%xmm2
-.byte	15,58,204,193,2
-.byte	15,56,200,213
-	pxor	%xmm6,%xmm4
-.byte	15,56,201,238
-.byte	15,56,202,231
-	movdqa	%xmm0,%xmm1
-.byte	15,58,204,194,2
-.byte	15,56,200,206
-	pxor	%xmm7,%xmm5
-.byte	15,56,202,236
-.byte	15,56,201,247
-	movdqa	%xmm0,%xmm2
-.byte	15,58,204,193,2
-.byte	15,56,200,215
-	pxor	%xmm4,%xmm6
-.byte	15,56,201,252
-.byte	15,56,202,245
-	movdqa	%xmm0,%xmm1
-.byte	15,58,204,194,3
-.byte	15,56,200,204
-	pxor	%xmm5,%xmm7
-.byte	15,56,202,254
-	movdqu	(%esi),%xmm4
-	movdqa	%xmm0,%xmm2
-.byte	15,58,204,193,3
-.byte	15,56,200,213
-	movdqu	16(%esi),%xmm5
-.byte	102,15,56,0,227
-	movdqa	%xmm0,%xmm1
-.byte	15,58,204,194,3
-.byte	15,56,200,206
-	movdqu	32(%esi),%xmm6
-.byte	102,15,56,0,235
-	movdqa	%xmm0,%xmm2
-.byte	15,58,204,193,3
-.byte	15,56,200,215
-	movdqu	48(%esi),%xmm7
-.byte	102,15,56,0,243
-	movdqa	%xmm0,%xmm1
-.byte	15,58,204,194,3
-	movdqa	(%esp),%xmm2
-.byte	102,15,56,0,251
-.byte	15,56,200,202
-	paddd	16(%esp),%xmm0
-	jnz	.L004loop_shaext
-	pshufd	$27,%xmm0,%xmm0
-	pshufd	$27,%xmm1,%xmm1
-	movdqu	%xmm0,(%edi)
-	movd	%xmm1,16(%edi)
-	movl	%ebx,%esp
-	popl	%edi
-	popl	%esi
-	popl	%ebx
-	popl	%ebp
-	ret
-.size	_sha1_block_data_order_shaext,.-_sha1_block_data_order_shaext
 .hidden	_sha1_block_data_order_ssse3
 .type	_sha1_block_data_order_ssse3,@function
 .align	16
@@ -1572,10 +1404,10 @@
 	pushl	%ebx
 	pushl	%esi
 	pushl	%edi
-	call	.L005pic_point
-.L005pic_point:
+	call	.L003pic_point
+.L003pic_point:
 	popl	%ebp
-	leal	.LK_XX_XX-.L005pic_point(%ebp),%ebp
+	leal	.LK_XX_XX-.L003pic_point(%ebp),%ebp
 .Lssse3_shortcut:
 	movdqa	(%ebp),%xmm7
 	movdqa	16(%ebp),%xmm0
@@ -1628,9 +1460,9 @@
 	xorl	%edx,%ebp
 	pshufd	$238,%xmm0,%xmm4
 	andl	%ebp,%esi
-	jmp	.L006loop
+	jmp	.L004loop
 .align	16
-.L006loop:
+.L004loop:
 	rorl	$2,%ebx
 	xorl	%edx,%esi
 	movl	%eax,%ebp
@@ -2533,7 +2365,7 @@
 	addl	%edx,%ecx
 	movl	196(%esp),%ebp
 	cmpl	200(%esp),%ebp
-	je	.L007done
+	je	.L005done
 	movdqa	160(%esp),%xmm7
 	movdqa	176(%esp),%xmm6
 	movdqu	(%ebp),%xmm0
@@ -2668,9 +2500,9 @@
 	pshufd	$238,%xmm0,%xmm4
 	andl	%ebx,%esi
 	movl	%ebp,%ebx
-	jmp	.L006loop
+	jmp	.L004loop
 .align	16
-.L007done:
+.L005done:
 	addl	16(%esp),%ebx
 	xorl	%edi,%esi
 	movl	%ecx,%ebp
@@ -2784,6 +2616,1177 @@
 	popl	%ebp
 	ret
 .size	_sha1_block_data_order_ssse3,.-_sha1_block_data_order_ssse3
+.hidden	_sha1_block_data_order_avx
+.type	_sha1_block_data_order_avx,@function
+.align	16
+_sha1_block_data_order_avx:
+	pushl	%ebp
+	pushl	%ebx
+	pushl	%esi
+	pushl	%edi
+	call	.L006pic_point
+.L006pic_point:
+	popl	%ebp
+	leal	.LK_XX_XX-.L006pic_point(%ebp),%ebp
+.Lavx_shortcut:
+	vzeroall
+	vmovdqa	(%ebp),%xmm7
+	vmovdqa	16(%ebp),%xmm0
+	vmovdqa	32(%ebp),%xmm1
+	vmovdqa	48(%ebp),%xmm2
+	vmovdqa	64(%ebp),%xmm6
+	movl	20(%esp),%edi
+	movl	24(%esp),%ebp
+	movl	28(%esp),%edx
+	movl	%esp,%esi
+	subl	$208,%esp
+	andl	$-64,%esp
+	vmovdqa	%xmm0,112(%esp)
+	vmovdqa	%xmm1,128(%esp)
+	vmovdqa	%xmm2,144(%esp)
+	shll	$6,%edx
+	vmovdqa	%xmm7,160(%esp)
+	addl	%ebp,%edx
+	vmovdqa	%xmm6,176(%esp)
+	addl	$64,%ebp
+	movl	%edi,192(%esp)
+	movl	%ebp,196(%esp)
+	movl	%edx,200(%esp)
+	movl	%esi,204(%esp)
+	movl	(%edi),%eax
+	movl	4(%edi),%ebx
+	movl	8(%edi),%ecx
+	movl	12(%edi),%edx
+	movl	16(%edi),%edi
+	movl	%ebx,%esi
+	vmovdqu	-64(%ebp),%xmm0
+	vmovdqu	-48(%ebp),%xmm1
+	vmovdqu	-32(%ebp),%xmm2
+	vmovdqu	-16(%ebp),%xmm3
+	vpshufb	%xmm6,%xmm0,%xmm0
+	vpshufb	%xmm6,%xmm1,%xmm1
+	vpshufb	%xmm6,%xmm2,%xmm2
+	vmovdqa	%xmm7,96(%esp)
+	vpshufb	%xmm6,%xmm3,%xmm3
+	vpaddd	%xmm7,%xmm0,%xmm4
+	vpaddd	%xmm7,%xmm1,%xmm5
+	vpaddd	%xmm7,%xmm2,%xmm6
+	vmovdqa	%xmm4,(%esp)
+	movl	%ecx,%ebp
+	vmovdqa	%xmm5,16(%esp)
+	xorl	%edx,%ebp
+	vmovdqa	%xmm6,32(%esp)
+	andl	%ebp,%esi
+	jmp	.L007loop
+.align	16
+.L007loop:
+	shrdl	$2,%ebx,%ebx
+	xorl	%edx,%esi
+	vpalignr	$8,%xmm0,%xmm1,%xmm4
+	movl	%eax,%ebp
+	addl	(%esp),%edi
+	vpaddd	%xmm3,%xmm7,%xmm7
+	vmovdqa	%xmm0,64(%esp)
+	xorl	%ecx,%ebx
+	shldl	$5,%eax,%eax
+	vpsrldq	$4,%xmm3,%xmm6
+	addl	%esi,%edi
+	andl	%ebx,%ebp
+	vpxor	%xmm0,%xmm4,%xmm4
+	xorl	%ecx,%ebx
+	addl	%eax,%edi
+	vpxor	%xmm2,%xmm6,%xmm6
+	shrdl	$7,%eax,%eax
+	xorl	%ecx,%ebp
+	vmovdqa	%xmm7,48(%esp)
+	movl	%edi,%esi
+	addl	4(%esp),%edx
+	vpxor	%xmm6,%xmm4,%xmm4
+	xorl	%ebx,%eax
+	shldl	$5,%edi,%edi
+	addl	%ebp,%edx
+	andl	%eax,%esi
+	vpsrld	$31,%xmm4,%xmm6
+	xorl	%ebx,%eax
+	addl	%edi,%edx
+	shrdl	$7,%edi,%edi
+	xorl	%ebx,%esi
+	vpslldq	$12,%xmm4,%xmm0
+	vpaddd	%xmm4,%xmm4,%xmm4
+	movl	%edx,%ebp
+	addl	8(%esp),%ecx
+	xorl	%eax,%edi
+	shldl	$5,%edx,%edx
+	vpsrld	$30,%xmm0,%xmm7
+	vpor	%xmm6,%xmm4,%xmm4
+	addl	%esi,%ecx
+	andl	%edi,%ebp
+	xorl	%eax,%edi
+	addl	%edx,%ecx
+	vpslld	$2,%xmm0,%xmm0
+	shrdl	$7,%edx,%edx
+	xorl	%eax,%ebp
+	vpxor	%xmm7,%xmm4,%xmm4
+	movl	%ecx,%esi
+	addl	12(%esp),%ebx
+	xorl	%edi,%edx
+	shldl	$5,%ecx,%ecx
+	vpxor	%xmm0,%xmm4,%xmm4
+	addl	%ebp,%ebx
+	andl	%edx,%esi
+	vmovdqa	96(%esp),%xmm0
+	xorl	%edi,%edx
+	addl	%ecx,%ebx
+	shrdl	$7,%ecx,%ecx
+	xorl	%edi,%esi
+	vpalignr	$8,%xmm1,%xmm2,%xmm5
+	movl	%ebx,%ebp
+	addl	16(%esp),%eax
+	vpaddd	%xmm4,%xmm0,%xmm0
+	vmovdqa	%xmm1,80(%esp)
+	xorl	%edx,%ecx
+	shldl	$5,%ebx,%ebx
+	vpsrldq	$4,%xmm4,%xmm7
+	addl	%esi,%eax
+	andl	%ecx,%ebp
+	vpxor	%xmm1,%xmm5,%xmm5
+	xorl	%edx,%ecx
+	addl	%ebx,%eax
+	vpxor	%xmm3,%xmm7,%xmm7
+	shrdl	$7,%ebx,%ebx
+	xorl	%edx,%ebp
+	vmovdqa	%xmm0,(%esp)
+	movl	%eax,%esi
+	addl	20(%esp),%edi
+	vpxor	%xmm7,%xmm5,%xmm5
+	xorl	%ecx,%ebx
+	shldl	$5,%eax,%eax
+	addl	%ebp,%edi
+	andl	%ebx,%esi
+	vpsrld	$31,%xmm5,%xmm7
+	xorl	%ecx,%ebx
+	addl	%eax,%edi
+	shrdl	$7,%eax,%eax
+	xorl	%ecx,%esi
+	vpslldq	$12,%xmm5,%xmm1
+	vpaddd	%xmm5,%xmm5,%xmm5
+	movl	%edi,%ebp
+	addl	24(%esp),%edx
+	xorl	%ebx,%eax
+	shldl	$5,%edi,%edi
+	vpsrld	$30,%xmm1,%xmm0
+	vpor	%xmm7,%xmm5,%xmm5
+	addl	%esi,%edx
+	andl	%eax,%ebp
+	xorl	%ebx,%eax
+	addl	%edi,%edx
+	vpslld	$2,%xmm1,%xmm1
+	shrdl	$7,%edi,%edi
+	xorl	%ebx,%ebp
+	vpxor	%xmm0,%xmm5,%xmm5
+	movl	%edx,%esi
+	addl	28(%esp),%ecx
+	xorl	%eax,%edi
+	shldl	$5,%edx,%edx
+	vpxor	%xmm1,%xmm5,%xmm5
+	addl	%ebp,%ecx
+	andl	%edi,%esi
+	vmovdqa	112(%esp),%xmm1
+	xorl	%eax,%edi
+	addl	%edx,%ecx
+	shrdl	$7,%edx,%edx
+	xorl	%eax,%esi
+	vpalignr	$8,%xmm2,%xmm3,%xmm6
+	movl	%ecx,%ebp
+	addl	32(%esp),%ebx
+	vpaddd	%xmm5,%xmm1,%xmm1
+	vmovdqa	%xmm2,96(%esp)
+	xorl	%edi,%edx
+	shldl	$5,%ecx,%ecx
+	vpsrldq	$4,%xmm5,%xmm0
+	addl	%esi,%ebx
+	andl	%edx,%ebp
+	vpxor	%xmm2,%xmm6,%xmm6
+	xorl	%edi,%edx
+	addl	%ecx,%ebx
+	vpxor	%xmm4,%xmm0,%xmm0
+	shrdl	$7,%ecx,%ecx
+	xorl	%edi,%ebp
+	vmovdqa	%xmm1,16(%esp)
+	movl	%ebx,%esi
+	addl	36(%esp),%eax
+	vpxor	%xmm0,%xmm6,%xmm6
+	xorl	%edx,%ecx
+	shldl	$5,%ebx,%ebx
+	addl	%ebp,%eax
+	andl	%ecx,%esi
+	vpsrld	$31,%xmm6,%xmm0
+	xorl	%edx,%ecx
+	addl	%ebx,%eax
+	shrdl	$7,%ebx,%ebx
+	xorl	%edx,%esi
+	vpslldq	$12,%xmm6,%xmm2
+	vpaddd	%xmm6,%xmm6,%xmm6
+	movl	%eax,%ebp
+	addl	40(%esp),%edi
+	xorl	%ecx,%ebx
+	shldl	$5,%eax,%eax
+	vpsrld	$30,%xmm2,%xmm1
+	vpor	%xmm0,%xmm6,%xmm6
+	addl	%esi,%edi
+	andl	%ebx,%ebp
+	xorl	%ecx,%ebx
+	addl	%eax,%edi
+	vpslld	$2,%xmm2,%xmm2
+	vmovdqa	64(%esp),%xmm0
+	shrdl	$7,%eax,%eax
+	xorl	%ecx,%ebp
+	vpxor	%xmm1,%xmm6,%xmm6
+	movl	%edi,%esi
+	addl	44(%esp),%edx
+	xorl	%ebx,%eax
+	shldl	$5,%edi,%edi
+	vpxor	%xmm2,%xmm6,%xmm6
+	addl	%ebp,%edx
+	andl	%eax,%esi
+	vmovdqa	112(%esp),%xmm2
+	xorl	%ebx,%eax
+	addl	%edi,%edx
+	shrdl	$7,%edi,%edi
+	xorl	%ebx,%esi
+	vpalignr	$8,%xmm3,%xmm4,%xmm7
+	movl	%edx,%ebp
+	addl	48(%esp),%ecx
+	vpaddd	%xmm6,%xmm2,%xmm2
+	vmovdqa	%xmm3,64(%esp)
+	xorl	%eax,%edi
+	shldl	$5,%edx,%edx
+	vpsrldq	$4,%xmm6,%xmm1
+	addl	%esi,%ecx
+	andl	%edi,%ebp
+	vpxor	%xmm3,%xmm7,%xmm7
+	xorl	%eax,%edi
+	addl	%edx,%ecx
+	vpxor	%xmm5,%xmm1,%xmm1
+	shrdl	$7,%edx,%edx
+	xorl	%eax,%ebp
+	vmovdqa	%xmm2,32(%esp)
+	movl	%ecx,%esi
+	addl	52(%esp),%ebx
+	vpxor	%xmm1,%xmm7,%xmm7
+	xorl	%edi,%edx
+	shldl	$5,%ecx,%ecx
+	addl	%ebp,%ebx
+	andl	%edx,%esi
+	vpsrld	$31,%xmm7,%xmm1
+	xorl	%edi,%edx
+	addl	%ecx,%ebx
+	shrdl	$7,%ecx,%ecx
+	xorl	%edi,%esi
+	vpslldq	$12,%xmm7,%xmm3
+	vpaddd	%xmm7,%xmm7,%xmm7
+	movl	%ebx,%ebp
+	addl	56(%esp),%eax
+	xorl	%edx,%ecx
+	shldl	$5,%ebx,%ebx
+	vpsrld	$30,%xmm3,%xmm2
+	vpor	%xmm1,%xmm7,%xmm7
+	addl	%esi,%eax
+	andl	%ecx,%ebp
+	xorl	%edx,%ecx
+	addl	%ebx,%eax
+	vpslld	$2,%xmm3,%xmm3
+	vmovdqa	80(%esp),%xmm1
+	shrdl	$7,%ebx,%ebx
+	xorl	%edx,%ebp
+	vpxor	%xmm2,%xmm7,%xmm7
+	movl	%eax,%esi
+	addl	60(%esp),%edi
+	xorl	%ecx,%ebx
+	shldl	$5,%eax,%eax
+	vpxor	%xmm3,%xmm7,%xmm7
+	addl	%ebp,%edi
+	andl	%ebx,%esi
+	vmovdqa	112(%esp),%xmm3
+	xorl	%ecx,%ebx
+	addl	%eax,%edi
+	vpalignr	$8,%xmm6,%xmm7,%xmm2
+	vpxor	%xmm4,%xmm0,%xmm0
+	shrdl	$7,%eax,%eax
+	xorl	%ecx,%esi
+	movl	%edi,%ebp
+	addl	(%esp),%edx
+	vpxor	%xmm1,%xmm0,%xmm0
+	vmovdqa	%xmm4,80(%esp)
+	xorl	%ebx,%eax
+	shldl	$5,%edi,%edi
+	vmovdqa	%xmm3,%xmm4
+	vpaddd	%xmm7,%xmm3,%xmm3
+	addl	%esi,%edx
+	andl	%eax,%ebp
+	vpxor	%xmm2,%xmm0,%xmm0
+	xorl	%ebx,%eax
+	addl	%edi,%edx
+	shrdl	$7,%edi,%edi
+	xorl	%ebx,%ebp
+	vpsrld	$30,%xmm0,%xmm2
+	vmovdqa	%xmm3,48(%esp)
+	movl	%edx,%esi
+	addl	4(%esp),%ecx
+	xorl	%eax,%edi
+	shldl	$5,%edx,%edx
+	vpslld	$2,%xmm0,%xmm0
+	addl	%ebp,%ecx
+	andl	%edi,%esi
+	xorl	%eax,%edi
+	addl	%edx,%ecx
+	shrdl	$7,%edx,%edx
+	xorl	%eax,%esi
+	movl	%ecx,%ebp
+	addl	8(%esp),%ebx
+	vpor	%xmm2,%xmm0,%xmm0
+	xorl	%edi,%edx
+	shldl	$5,%ecx,%ecx
+	vmovdqa	96(%esp),%xmm2
+	addl	%esi,%ebx
+	andl	%edx,%ebp
+	xorl	%edi,%edx
+	addl	%ecx,%ebx
+	addl	12(%esp),%eax
+	xorl	%edi,%ebp
+	movl	%ebx,%esi
+	shldl	$5,%ebx,%ebx
+	addl	%ebp,%eax
+	xorl	%edx,%esi
+	shrdl	$7,%ecx,%ecx
+	addl	%ebx,%eax
+	vpalignr	$8,%xmm7,%xmm0,%xmm3
+	vpxor	%xmm5,%xmm1,%xmm1
+	addl	16(%esp),%edi
+	xorl	%ecx,%esi
+	movl	%eax,%ebp
+	shldl	$5,%eax,%eax
+	vpxor	%xmm2,%xmm1,%xmm1
+	vmovdqa	%xmm5,96(%esp)
+	addl	%esi,%edi
+	xorl	%ecx,%ebp
+	vmovdqa	%xmm4,%xmm5
+	vpaddd	%xmm0,%xmm4,%xmm4
+	shrdl	$7,%ebx,%ebx
+	addl	%eax,%edi
+	vpxor	%xmm3,%xmm1,%xmm1
+	addl	20(%esp),%edx
+	xorl	%ebx,%ebp
+	movl	%edi,%esi
+	shldl	$5,%edi,%edi
+	vpsrld	$30,%xmm1,%xmm3
+	vmovdqa	%xmm4,(%esp)
+	addl	%ebp,%edx
+	xorl	%ebx,%esi
+	shrdl	$7,%eax,%eax
+	addl	%edi,%edx
+	vpslld	$2,%xmm1,%xmm1
+	addl	24(%esp),%ecx
+	xorl	%eax,%esi
+	movl	%edx,%ebp
+	shldl	$5,%edx,%edx
+	addl	%esi,%ecx
+	xorl	%eax,%ebp
+	shrdl	$7,%edi,%edi
+	addl	%edx,%ecx
+	vpor	%xmm3,%xmm1,%xmm1
+	addl	28(%esp),%ebx
+	xorl	%edi,%ebp
+	vmovdqa	64(%esp),%xmm3
+	movl	%ecx,%esi
+	shldl	$5,%ecx,%ecx
+	addl	%ebp,%ebx
+	xorl	%edi,%esi
+	shrdl	$7,%edx,%edx
+	addl	%ecx,%ebx
+	vpalignr	$8,%xmm0,%xmm1,%xmm4
+	vpxor	%xmm6,%xmm2,%xmm2
+	addl	32(%esp),%eax
+	xorl	%edx,%esi
+	movl	%ebx,%ebp
+	shldl	$5,%ebx,%ebx
+	vpxor	%xmm3,%xmm2,%xmm2
+	vmovdqa	%xmm6,64(%esp)
+	addl	%esi,%eax
+	xorl	%edx,%ebp
+	vmovdqa	128(%esp),%xmm6
+	vpaddd	%xmm1,%xmm5,%xmm5
+	shrdl	$7,%ecx,%ecx
+	addl	%ebx,%eax
+	vpxor	%xmm4,%xmm2,%xmm2
+	addl	36(%esp),%edi
+	xorl	%ecx,%ebp
+	movl	%eax,%esi
+	shldl	$5,%eax,%eax
+	vpsrld	$30,%xmm2,%xmm4
+	vmovdqa	%xmm5,16(%esp)
+	addl	%ebp,%edi
+	xorl	%ecx,%esi
+	shrdl	$7,%ebx,%ebx
+	addl	%eax,%edi
+	vpslld	$2,%xmm2,%xmm2
+	addl	40(%esp),%edx
+	xorl	%ebx,%esi
+	movl	%edi,%ebp
+	shldl	$5,%edi,%edi
+	addl	%esi,%edx
+	xorl	%ebx,%ebp
+	shrdl	$7,%eax,%eax
+	addl	%edi,%edx
+	vpor	%xmm4,%xmm2,%xmm2
+	addl	44(%esp),%ecx
+	xorl	%eax,%ebp
+	vmovdqa	80(%esp),%xmm4
+	movl	%edx,%esi
+	shldl	$5,%edx,%edx
+	addl	%ebp,%ecx
+	xorl	%eax,%esi
+	shrdl	$7,%edi,%edi
+	addl	%edx,%ecx
+	vpalignr	$8,%xmm1,%xmm2,%xmm5
+	vpxor	%xmm7,%xmm3,%xmm3
+	addl	48(%esp),%ebx
+	xorl	%edi,%esi
+	movl	%ecx,%ebp
+	shldl	$5,%ecx,%ecx
+	vpxor	%xmm4,%xmm3,%xmm3
+	vmovdqa	%xmm7,80(%esp)
+	addl	%esi,%ebx
+	xorl	%edi,%ebp
+	vmovdqa	%xmm6,%xmm7
+	vpaddd	%xmm2,%xmm6,%xmm6
+	shrdl	$7,%edx,%edx
+	addl	%ecx,%ebx
+	vpxor	%xmm5,%xmm3,%xmm3
+	addl	52(%esp),%eax
+	xorl	%edx,%ebp
+	movl	%ebx,%esi
+	shldl	$5,%ebx,%ebx
+	vpsrld	$30,%xmm3,%xmm5
+	vmovdqa	%xmm6,32(%esp)
+	addl	%ebp,%eax
+	xorl	%edx,%esi
+	shrdl	$7,%ecx,%ecx
+	addl	%ebx,%eax
+	vpslld	$2,%xmm3,%xmm3
+	addl	56(%esp),%edi
+	xorl	%ecx,%esi
+	movl	%eax,%ebp
+	shldl	$5,%eax,%eax
+	addl	%esi,%edi
+	xorl	%ecx,%ebp
+	shrdl	$7,%ebx,%ebx
+	addl	%eax,%edi
+	vpor	%xmm5,%xmm3,%xmm3
+	addl	60(%esp),%edx
+	xorl	%ebx,%ebp
+	vmovdqa	96(%esp),%xmm5
+	movl	%edi,%esi
+	shldl	$5,%edi,%edi
+	addl	%ebp,%edx
+	xorl	%ebx,%esi
+	shrdl	$7,%eax,%eax
+	addl	%edi,%edx
+	vpalignr	$8,%xmm2,%xmm3,%xmm6
+	vpxor	%xmm0,%xmm4,%xmm4
+	addl	(%esp),%ecx
+	xorl	%eax,%esi
+	movl	%edx,%ebp
+	shldl	$5,%edx,%edx
+	vpxor	%xmm5,%xmm4,%xmm4
+	vmovdqa	%xmm0,96(%esp)
+	addl	%esi,%ecx
+	xorl	%eax,%ebp
+	vmovdqa	%xmm7,%xmm0
+	vpaddd	%xmm3,%xmm7,%xmm7
+	shrdl	$7,%edi,%edi
+	addl	%edx,%ecx
+	vpxor	%xmm6,%xmm4,%xmm4
+	addl	4(%esp),%ebx
+	xorl	%edi,%ebp
+	movl	%ecx,%esi
+	shldl	$5,%ecx,%ecx
+	vpsrld	$30,%xmm4,%xmm6
+	vmovdqa	%xmm7,48(%esp)
+	addl	%ebp,%ebx
+	xorl	%edi,%esi
+	shrdl	$7,%edx,%edx
+	addl	%ecx,%ebx
+	vpslld	$2,%xmm4,%xmm4
+	addl	8(%esp),%eax
+	xorl	%edx,%esi
+	movl	%ebx,%ebp
+	shldl	$5,%ebx,%ebx
+	addl	%esi,%eax
+	xorl	%edx,%ebp
+	shrdl	$7,%ecx,%ecx
+	addl	%ebx,%eax
+	vpor	%xmm6,%xmm4,%xmm4
+	addl	12(%esp),%edi
+	xorl	%ecx,%ebp
+	vmovdqa	64(%esp),%xmm6
+	movl	%eax,%esi
+	shldl	$5,%eax,%eax
+	addl	%ebp,%edi
+	xorl	%ecx,%esi
+	shrdl	$7,%ebx,%ebx
+	addl	%eax,%edi
+	vpalignr	$8,%xmm3,%xmm4,%xmm7
+	vpxor	%xmm1,%xmm5,%xmm5
+	addl	16(%esp),%edx
+	xorl	%ebx,%esi
+	movl	%edi,%ebp
+	shldl	$5,%edi,%edi
+	vpxor	%xmm6,%xmm5,%xmm5
+	vmovdqa	%xmm1,64(%esp)
+	addl	%esi,%edx
+	xorl	%ebx,%ebp
+	vmovdqa	%xmm0,%xmm1
+	vpaddd	%xmm4,%xmm0,%xmm0
+	shrdl	$7,%eax,%eax
+	addl	%edi,%edx
+	vpxor	%xmm7,%xmm5,%xmm5
+	addl	20(%esp),%ecx
+	xorl	%eax,%ebp
+	movl	%edx,%esi
+	shldl	$5,%edx,%edx
+	vpsrld	$30,%xmm5,%xmm7
+	vmovdqa	%xmm0,(%esp)
+	addl	%ebp,%ecx
+	xorl	%eax,%esi
+	shrdl	$7,%edi,%edi
+	addl	%edx,%ecx
+	vpslld	$2,%xmm5,%xmm5
+	addl	24(%esp),%ebx
+	xorl	%edi,%esi
+	movl	%ecx,%ebp
+	shldl	$5,%ecx,%ecx
+	addl	%esi,%ebx
+	xorl	%edi,%ebp
+	shrdl	$7,%edx,%edx
+	addl	%ecx,%ebx
+	vpor	%xmm7,%xmm5,%xmm5
+	addl	28(%esp),%eax
+	vmovdqa	80(%esp),%xmm7
+	shrdl	$7,%ecx,%ecx
+	movl	%ebx,%esi
+	xorl	%edx,%ebp
+	shldl	$5,%ebx,%ebx
+	addl	%ebp,%eax
+	xorl	%ecx,%esi
+	xorl	%edx,%ecx
+	addl	%ebx,%eax
+	vpalignr	$8,%xmm4,%xmm5,%xmm0
+	vpxor	%xmm2,%xmm6,%xmm6
+	addl	32(%esp),%edi
+	andl	%ecx,%esi
+	xorl	%edx,%ecx
+	shrdl	$7,%ebx,%ebx
+	vpxor	%xmm7,%xmm6,%xmm6
+	vmovdqa	%xmm2,80(%esp)
+	movl	%eax,%ebp
+	xorl	%ecx,%esi
+	vmovdqa	%xmm1,%xmm2
+	vpaddd	%xmm5,%xmm1,%xmm1
+	shldl	$5,%eax,%eax
+	addl	%esi,%edi
+	vpxor	%xmm0,%xmm6,%xmm6
+	xorl	%ebx,%ebp
+	xorl	%ecx,%ebx
+	addl	%eax,%edi
+	addl	36(%esp),%edx
+	vpsrld	$30,%xmm6,%xmm0
+	vmovdqa	%xmm1,16(%esp)
+	andl	%ebx,%ebp
+	xorl	%ecx,%ebx
+	shrdl	$7,%eax,%eax
+	movl	%edi,%esi
+	vpslld	$2,%xmm6,%xmm6
+	xorl	%ebx,%ebp
+	shldl	$5,%edi,%edi
+	addl	%ebp,%edx
+	xorl	%eax,%esi
+	xorl	%ebx,%eax
+	addl	%edi,%edx
+	addl	40(%esp),%ecx
+	andl	%eax,%esi
+	vpor	%xmm0,%xmm6,%xmm6
+	xorl	%ebx,%eax
+	shrdl	$7,%edi,%edi
+	vmovdqa	96(%esp),%xmm0
+	movl	%edx,%ebp
+	xorl	%eax,%esi
+	shldl	$5,%edx,%edx
+	addl	%esi,%ecx
+	xorl	%edi,%ebp
+	xorl	%eax,%edi
+	addl	%edx,%ecx
+	addl	44(%esp),%ebx
+	andl	%edi,%ebp
+	xorl	%eax,%edi
+	shrdl	$7,%edx,%edx
+	movl	%ecx,%esi
+	xorl	%edi,%ebp
+	shldl	$5,%ecx,%ecx
+	addl	%ebp,%ebx
+	xorl	%edx,%esi
+	xorl	%edi,%edx
+	addl	%ecx,%ebx
+	vpalignr	$8,%xmm5,%xmm6,%xmm1
+	vpxor	%xmm3,%xmm7,%xmm7
+	addl	48(%esp),%eax
+	andl	%edx,%esi
+	xorl	%edi,%edx
+	shrdl	$7,%ecx,%ecx
+	vpxor	%xmm0,%xmm7,%xmm7
+	vmovdqa	%xmm3,96(%esp)
+	movl	%ebx,%ebp
+	xorl	%edx,%esi
+	vmovdqa	144(%esp),%xmm3
+	vpaddd	%xmm6,%xmm2,%xmm2
+	shldl	$5,%ebx,%ebx
+	addl	%esi,%eax
+	vpxor	%xmm1,%xmm7,%xmm7
+	xorl	%ecx,%ebp
+	xorl	%edx,%ecx
+	addl	%ebx,%eax
+	addl	52(%esp),%edi
+	vpsrld	$30,%xmm7,%xmm1
+	vmovdqa	%xmm2,32(%esp)
+	andl	%ecx,%ebp
+	xorl	%edx,%ecx
+	shrdl	$7,%ebx,%ebx
+	movl	%eax,%esi
+	vpslld	$2,%xmm7,%xmm7
+	xorl	%ecx,%ebp
+	shldl	$5,%eax,%eax
+	addl	%ebp,%edi
+	xorl	%ebx,%esi
+	xorl	%ecx,%ebx
+	addl	%eax,%edi
+	addl	56(%esp),%edx
+	andl	%ebx,%esi
+	vpor	%xmm1,%xmm7,%xmm7
+	xorl	%ecx,%ebx
+	shrdl	$7,%eax,%eax
+	vmovdqa	64(%esp),%xmm1
+	movl	%edi,%ebp
+	xorl	%ebx,%esi
+	shldl	$5,%edi,%edi
+	addl	%esi,%edx
+	xorl	%eax,%ebp
+	xorl	%ebx,%eax
+	addl	%edi,%edx
+	addl	60(%esp),%ecx
+	andl	%eax,%ebp
+	xorl	%ebx,%eax
+	shrdl	$7,%edi,%edi
+	movl	%edx,%esi
+	xorl	%eax,%ebp
+	shldl	$5,%edx,%edx
+	addl	%ebp,%ecx
+	xorl	%edi,%esi
+	xorl	%eax,%edi
+	addl	%edx,%ecx
+	vpalignr	$8,%xmm6,%xmm7,%xmm2
+	vpxor	%xmm4,%xmm0,%xmm0
+	addl	(%esp),%ebx
+	andl	%edi,%esi
+	xorl	%eax,%edi
+	shrdl	$7,%edx,%edx
+	vpxor	%xmm1,%xmm0,%xmm0
+	vmovdqa	%xmm4,64(%esp)
+	movl	%ecx,%ebp
+	xorl	%edi,%esi
+	vmovdqa	%xmm3,%xmm4
+	vpaddd	%xmm7,%xmm3,%xmm3
+	shldl	$5,%ecx,%ecx
+	addl	%esi,%ebx
+	vpxor	%xmm2,%xmm0,%xmm0
+	xorl	%edx,%ebp
+	xorl	%edi,%edx
+	addl	%ecx,%ebx
+	addl	4(%esp),%eax
+	vpsrld	$30,%xmm0,%xmm2
+	vmovdqa	%xmm3,48(%esp)
+	andl	%edx,%ebp
+	xorl	%edi,%edx
+	shrdl	$7,%ecx,%ecx
+	movl	%ebx,%esi
+	vpslld	$2,%xmm0,%xmm0
+	xorl	%edx,%ebp
+	shldl	$5,%ebx,%ebx
+	addl	%ebp,%eax
+	xorl	%ecx,%esi
+	xorl	%edx,%ecx
+	addl	%ebx,%eax
+	addl	8(%esp),%edi
+	andl	%ecx,%esi
+	vpor	%xmm2,%xmm0,%xmm0
+	xorl	%edx,%ecx
+	shrdl	$7,%ebx,%ebx
+	vmovdqa	80(%esp),%xmm2
+	movl	%eax,%ebp
+	xorl	%ecx,%esi
+	shldl	$5,%eax,%eax
+	addl	%esi,%edi
+	xorl	%ebx,%ebp
+	xorl	%ecx,%ebx
+	addl	%eax,%edi
+	addl	12(%esp),%edx
+	andl	%ebx,%ebp
+	xorl	%ecx,%ebx
+	shrdl	$7,%eax,%eax
+	movl	%edi,%esi
+	xorl	%ebx,%ebp
+	shldl	$5,%edi,%edi
+	addl	%ebp,%edx
+	xorl	%eax,%esi
+	xorl	%ebx,%eax
+	addl	%edi,%edx
+	vpalignr	$8,%xmm7,%xmm0,%xmm3
+	vpxor	%xmm5,%xmm1,%xmm1
+	addl	16(%esp),%ecx
+	andl	%eax,%esi
+	xorl	%ebx,%eax
+	shrdl	$7,%edi,%edi
+	vpxor	%xmm2,%xmm1,%xmm1
+	vmovdqa	%xmm5,80(%esp)
+	movl	%edx,%ebp
+	xorl	%eax,%esi
+	vmovdqa	%xmm4,%xmm5
+	vpaddd	%xmm0,%xmm4,%xmm4
+	shldl	$5,%edx,%edx
+	addl	%esi,%ecx
+	vpxor	%xmm3,%xmm1,%xmm1
+	xorl	%edi,%ebp
+	xorl	%eax,%edi
+	addl	%edx,%ecx
+	addl	20(%esp),%ebx
+	vpsrld	$30,%xmm1,%xmm3
+	vmovdqa	%xmm4,(%esp)
+	andl	%edi,%ebp
+	xorl	%eax,%edi
+	shrdl	$7,%edx,%edx
+	movl	%ecx,%esi
+	vpslld	$2,%xmm1,%xmm1
+	xorl	%edi,%ebp
+	shldl	$5,%ecx,%ecx
+	addl	%ebp,%ebx
+	xorl	%edx,%esi
+	xorl	%edi,%edx
+	addl	%ecx,%ebx
+	addl	24(%esp),%eax
+	andl	%edx,%esi
+	vpor	%xmm3,%xmm1,%xmm1
+	xorl	%edi,%edx
+	shrdl	$7,%ecx,%ecx
+	vmovdqa	96(%esp),%xmm3
+	movl	%ebx,%ebp
+	xorl	%edx,%esi
+	shldl	$5,%ebx,%ebx
+	addl	%esi,%eax
+	xorl	%ecx,%ebp
+	xorl	%edx,%ecx
+	addl	%ebx,%eax
+	addl	28(%esp),%edi
+	andl	%ecx,%ebp
+	xorl	%edx,%ecx
+	shrdl	$7,%ebx,%ebx
+	movl	%eax,%esi
+	xorl	%ecx,%ebp
+	shldl	$5,%eax,%eax
+	addl	%ebp,%edi
+	xorl	%ebx,%esi
+	xorl	%ecx,%ebx
+	addl	%eax,%edi
+	vpalignr	$8,%xmm0,%xmm1,%xmm4
+	vpxor	%xmm6,%xmm2,%xmm2
+	addl	32(%esp),%edx
+	andl	%ebx,%esi
+	xorl	%ecx,%ebx
+	shrdl	$7,%eax,%eax
+	vpxor	%xmm3,%xmm2,%xmm2
+	vmovdqa	%xmm6,96(%esp)
+	movl	%edi,%ebp
+	xorl	%ebx,%esi
+	vmovdqa	%xmm5,%xmm6
+	vpaddd	%xmm1,%xmm5,%xmm5
+	shldl	$5,%edi,%edi
+	addl	%esi,%edx
+	vpxor	%xmm4,%xmm2,%xmm2
+	xorl	%eax,%ebp
+	xorl	%ebx,%eax
+	addl	%edi,%edx
+	addl	36(%esp),%ecx
+	vpsrld	$30,%xmm2,%xmm4
+	vmovdqa	%xmm5,16(%esp)
+	andl	%eax,%ebp
+	xorl	%ebx,%eax
+	shrdl	$7,%edi,%edi
+	movl	%edx,%esi
+	vpslld	$2,%xmm2,%xmm2
+	xorl	%eax,%ebp
+	shldl	$5,%edx,%edx
+	addl	%ebp,%ecx
+	xorl	%edi,%esi
+	xorl	%eax,%edi
+	addl	%edx,%ecx
+	addl	40(%esp),%ebx
+	andl	%edi,%esi
+	vpor	%xmm4,%xmm2,%xmm2
+	xorl	%eax,%edi
+	shrdl	$7,%edx,%edx
+	vmovdqa	64(%esp),%xmm4
+	movl	%ecx,%ebp
+	xorl	%edi,%esi
+	shldl	$5,%ecx,%ecx
+	addl	%esi,%ebx
+	xorl	%edx,%ebp
+	xorl	%edi,%edx
+	addl	%ecx,%ebx
+	addl	44(%esp),%eax
+	andl	%edx,%ebp
+	xorl	%edi,%edx
+	shrdl	$7,%ecx,%ecx
+	movl	%ebx,%esi
+	xorl	%edx,%ebp
+	shldl	$5,%ebx,%ebx
+	addl	%ebp,%eax
+	xorl	%edx,%esi
+	addl	%ebx,%eax
+	vpalignr	$8,%xmm1,%xmm2,%xmm5
+	vpxor	%xmm7,%xmm3,%xmm3
+	addl	48(%esp),%edi
+	xorl	%ecx,%esi
+	movl	%eax,%ebp
+	shldl	$5,%eax,%eax
+	vpxor	%xmm4,%xmm3,%xmm3
+	vmovdqa	%xmm7,64(%esp)
+	addl	%esi,%edi
+	xorl	%ecx,%ebp
+	vmovdqa	%xmm6,%xmm7
+	vpaddd	%xmm2,%xmm6,%xmm6
+	shrdl	$7,%ebx,%ebx
+	addl	%eax,%edi
+	vpxor	%xmm5,%xmm3,%xmm3
+	addl	52(%esp),%edx
+	xorl	%ebx,%ebp
+	movl	%edi,%esi
+	shldl	$5,%edi,%edi
+	vpsrld	$30,%xmm3,%xmm5
+	vmovdqa	%xmm6,32(%esp)
+	addl	%ebp,%edx
+	xorl	%ebx,%esi
+	shrdl	$7,%eax,%eax
+	addl	%edi,%edx
+	vpslld	$2,%xmm3,%xmm3
+	addl	56(%esp),%ecx
+	xorl	%eax,%esi
+	movl	%edx,%ebp
+	shldl	$5,%edx,%edx
+	addl	%esi,%ecx
+	xorl	%eax,%ebp
+	shrdl	$7,%edi,%edi
+	addl	%edx,%ecx
+	vpor	%xmm5,%xmm3,%xmm3
+	addl	60(%esp),%ebx
+	xorl	%edi,%ebp
+	movl	%ecx,%esi
+	shldl	$5,%ecx,%ecx
+	addl	%ebp,%ebx
+	xorl	%edi,%esi
+	shrdl	$7,%edx,%edx
+	addl	%ecx,%ebx
+	addl	(%esp),%eax
+	vpaddd	%xmm3,%xmm7,%xmm7
+	xorl	%edx,%esi
+	movl	%ebx,%ebp
+	shldl	$5,%ebx,%ebx
+	addl	%esi,%eax
+	vmovdqa	%xmm7,48(%esp)
+	xorl	%edx,%ebp
+	shrdl	$7,%ecx,%ecx
+	addl	%ebx,%eax
+	addl	4(%esp),%edi
+	xorl	%ecx,%ebp
+	movl	%eax,%esi
+	shldl	$5,%eax,%eax
+	addl	%ebp,%edi
+	xorl	%ecx,%esi
+	shrdl	$7,%ebx,%ebx
+	addl	%eax,%edi
+	addl	8(%esp),%edx
+	xorl	%ebx,%esi
+	movl	%edi,%ebp
+	shldl	$5,%edi,%edi
+	addl	%esi,%edx
+	xorl	%ebx,%ebp
+	shrdl	$7,%eax,%eax
+	addl	%edi,%edx
+	addl	12(%esp),%ecx
+	xorl	%eax,%ebp
+	movl	%edx,%esi
+	shldl	$5,%edx,%edx
+	addl	%ebp,%ecx
+	xorl	%eax,%esi
+	shrdl	$7,%edi,%edi
+	addl	%edx,%ecx
+	movl	196(%esp),%ebp
+	cmpl	200(%esp),%ebp
+	je	.L008done
+	vmovdqa	160(%esp),%xmm7
+	vmovdqa	176(%esp),%xmm6
+	vmovdqu	(%ebp),%xmm0
+	vmovdqu	16(%ebp),%xmm1
+	vmovdqu	32(%ebp),%xmm2
+	vmovdqu	48(%ebp),%xmm3
+	addl	$64,%ebp
+	vpshufb	%xmm6,%xmm0,%xmm0
+	movl	%ebp,196(%esp)
+	vmovdqa	%xmm7,96(%esp)
+	addl	16(%esp),%ebx
+	xorl	%edi,%esi
+	vpshufb	%xmm6,%xmm1,%xmm1
+	movl	%ecx,%ebp
+	shldl	$5,%ecx,%ecx
+	vpaddd	%xmm7,%xmm0,%xmm4
+	addl	%esi,%ebx
+	xorl	%edi,%ebp
+	shrdl	$7,%edx,%edx
+	addl	%ecx,%ebx
+	vmovdqa	%xmm4,(%esp)
+	addl	20(%esp),%eax
+	xorl	%edx,%ebp
+	movl	%ebx,%esi
+	shldl	$5,%ebx,%ebx
+	addl	%ebp,%eax
+	xorl	%edx,%esi
+	shrdl	$7,%ecx,%ecx
+	addl	%ebx,%eax
+	addl	24(%esp),%edi
+	xorl	%ecx,%esi
+	movl	%eax,%ebp
+	shldl	$5,%eax,%eax
+	addl	%esi,%edi
+	xorl	%ecx,%ebp
+	shrdl	$7,%ebx,%ebx
+	addl	%eax,%edi
+	addl	28(%esp),%edx
+	xorl	%ebx,%ebp
+	movl	%edi,%esi
+	shldl	$5,%edi,%edi
+	addl	%ebp,%edx
+	xorl	%ebx,%esi
+	shrdl	$7,%eax,%eax
+	addl	%edi,%edx
+	addl	32(%esp),%ecx
+	xorl	%eax,%esi
+	vpshufb	%xmm6,%xmm2,%xmm2
+	movl	%edx,%ebp
+	shldl	$5,%edx,%edx
+	vpaddd	%xmm7,%xmm1,%xmm5
+	addl	%esi,%ecx
+	xorl	%eax,%ebp
+	shrdl	$7,%edi,%edi
+	addl	%edx,%ecx
+	vmovdqa	%xmm5,16(%esp)
+	addl	36(%esp),%ebx
+	xorl	%edi,%ebp
+	movl	%ecx,%esi
+	shldl	$5,%ecx,%ecx
+	addl	%ebp,%ebx
+	xorl	%edi,%esi
+	shrdl	$7,%edx,%edx
+	addl	%ecx,%ebx
+	addl	40(%esp),%eax
+	xorl	%edx,%esi
+	movl	%ebx,%ebp
+	shldl	$5,%ebx,%ebx
+	addl	%esi,%eax
+	xorl	%edx,%ebp
+	shrdl	$7,%ecx,%ecx
+	addl	%ebx,%eax
+	addl	44(%esp),%edi
+	xorl	%ecx,%ebp
+	movl	%eax,%esi
+	shldl	$5,%eax,%eax
+	addl	%ebp,%edi
+	xorl	%ecx,%esi
+	shrdl	$7,%ebx,%ebx
+	addl	%eax,%edi
+	addl	48(%esp),%edx
+	xorl	%ebx,%esi
+	vpshufb	%xmm6,%xmm3,%xmm3
+	movl	%edi,%ebp
+	shldl	$5,%edi,%edi
+	vpaddd	%xmm7,%xmm2,%xmm6
+	addl	%esi,%edx
+	xorl	%ebx,%ebp
+	shrdl	$7,%eax,%eax
+	addl	%edi,%edx
+	vmovdqa	%xmm6,32(%esp)
+	addl	52(%esp),%ecx
+	xorl	%eax,%ebp
+	movl	%edx,%esi
+	shldl	$5,%edx,%edx
+	addl	%ebp,%ecx
+	xorl	%eax,%esi
+	shrdl	$7,%edi,%edi
+	addl	%edx,%ecx
+	addl	56(%esp),%ebx
+	xorl	%edi,%esi
+	movl	%ecx,%ebp
+	shldl	$5,%ecx,%ecx
+	addl	%esi,%ebx
+	xorl	%edi,%ebp
+	shrdl	$7,%edx,%edx
+	addl	%ecx,%ebx
+	addl	60(%esp),%eax
+	xorl	%edx,%ebp
+	movl	%ebx,%esi
+	shldl	$5,%ebx,%ebx
+	addl	%ebp,%eax
+	shrdl	$7,%ecx,%ecx
+	addl	%ebx,%eax
+	movl	192(%esp),%ebp
+	addl	(%ebp),%eax
+	addl	4(%ebp),%esi
+	addl	8(%ebp),%ecx
+	movl	%eax,(%ebp)
+	addl	12(%ebp),%edx
+	movl	%esi,4(%ebp)
+	addl	16(%ebp),%edi
+	movl	%ecx,%ebx
+	movl	%ecx,8(%ebp)
+	xorl	%edx,%ebx
+	movl	%edx,12(%ebp)
+	movl	%edi,16(%ebp)
+	movl	%esi,%ebp
+	andl	%ebx,%esi
+	movl	%ebp,%ebx
+	jmp	.L007loop
+.align	16
+.L008done:
+	addl	16(%esp),%ebx
+	xorl	%edi,%esi
+	movl	%ecx,%ebp
+	shldl	$5,%ecx,%ecx
+	addl	%esi,%ebx
+	xorl	%edi,%ebp
+	shrdl	$7,%edx,%edx
+	addl	%ecx,%ebx
+	addl	20(%esp),%eax
+	xorl	%edx,%ebp
+	movl	%ebx,%esi
+	shldl	$5,%ebx,%ebx
+	addl	%ebp,%eax
+	xorl	%edx,%esi
+	shrdl	$7,%ecx,%ecx
+	addl	%ebx,%eax
+	addl	24(%esp),%edi
+	xorl	%ecx,%esi
+	movl	%eax,%ebp
+	shldl	$5,%eax,%eax
+	addl	%esi,%edi
+	xorl	%ecx,%ebp
+	shrdl	$7,%ebx,%ebx
+	addl	%eax,%edi
+	addl	28(%esp),%edx
+	xorl	%ebx,%ebp
+	movl	%edi,%esi
+	shldl	$5,%edi,%edi
+	addl	%ebp,%edx
+	xorl	%ebx,%esi
+	shrdl	$7,%eax,%eax
+	addl	%edi,%edx
+	addl	32(%esp),%ecx
+	xorl	%eax,%esi
+	movl	%edx,%ebp
+	shldl	$5,%edx,%edx
+	addl	%esi,%ecx
+	xorl	%eax,%ebp
+	shrdl	$7,%edi,%edi
+	addl	%edx,%ecx
+	addl	36(%esp),%ebx
+	xorl	%edi,%ebp
+	movl	%ecx,%esi
+	shldl	$5,%ecx,%ecx
+	addl	%ebp,%ebx
+	xorl	%edi,%esi
+	shrdl	$7,%edx,%edx
+	addl	%ecx,%ebx
+	addl	40(%esp),%eax
+	xorl	%edx,%esi
+	movl	%ebx,%ebp
+	shldl	$5,%ebx,%ebx
+	addl	%esi,%eax
+	xorl	%edx,%ebp
+	shrdl	$7,%ecx,%ecx
+	addl	%ebx,%eax
+	addl	44(%esp),%edi
+	xorl	%ecx,%ebp
+	movl	%eax,%esi
+	shldl	$5,%eax,%eax
+	addl	%ebp,%edi
+	xorl	%ecx,%esi
+	shrdl	$7,%ebx,%ebx
+	addl	%eax,%edi
+	addl	48(%esp),%edx
+	xorl	%ebx,%esi
+	movl	%edi,%ebp
+	shldl	$5,%edi,%edi
+	addl	%esi,%edx
+	xorl	%ebx,%ebp
+	shrdl	$7,%eax,%eax
+	addl	%edi,%edx
+	addl	52(%esp),%ecx
+	xorl	%eax,%ebp
+	movl	%edx,%esi
+	shldl	$5,%edx,%edx
+	addl	%ebp,%ecx
+	xorl	%eax,%esi
+	shrdl	$7,%edi,%edi
+	addl	%edx,%ecx
+	addl	56(%esp),%ebx
+	xorl	%edi,%esi
+	movl	%ecx,%ebp
+	shldl	$5,%ecx,%ecx
+	addl	%esi,%ebx
+	xorl	%edi,%ebp
+	shrdl	$7,%edx,%edx
+	addl	%ecx,%ebx
+	addl	60(%esp),%eax
+	xorl	%edx,%ebp
+	movl	%ebx,%esi
+	shldl	$5,%ebx,%ebx
+	addl	%ebp,%eax
+	shrdl	$7,%ecx,%ecx
+	addl	%ebx,%eax
+	vzeroall
+	movl	192(%esp),%ebp
+	addl	(%ebp),%eax
+	movl	204(%esp),%esp
+	addl	4(%ebp),%esi
+	addl	8(%ebp),%ecx
+	movl	%eax,(%ebp)
+	addl	12(%ebp),%edx
+	movl	%esi,4(%ebp)
+	addl	16(%ebp),%edi
+	movl	%ecx,8(%ebp)
+	movl	%edx,12(%ebp)
+	movl	%edi,16(%ebp)
+	popl	%edi
+	popl	%esi
+	popl	%ebx
+	popl	%ebp
+	ret
+.size	_sha1_block_data_order_avx,.-_sha1_block_data_order_avx
 .align	64
 .LK_XX_XX:
 .long	1518500249,1518500249,1518500249,1518500249

third_party/boringssl/linux-x86/crypto/sha/sha256-586.S

diff --git a/third_party/boringssl/linux-x86/crypto/sha/sha256-586.S b/third_party/boringssl/linux-x86/crypto/sha/sha256-586.S
index 08d9484..38acbd8 100644
--- a/third_party/boringssl/linux-x86/crypto/sha/sha256-586.S
+++ b/third_party/boringssl/linux-x86/crypto/sha/sha256-586.S
@@ -37,11 +37,10 @@
 	jz	.L003no_xmm
 	andl	$1073741824,%ecx
 	andl	$268435968,%ebx
-	testl	$536870912,%edx
-	jnz	.L004shaext
 	orl	%ebx,%ecx
 	andl	$1342177280,%ecx
 	cmpl	$1342177280,%ecx
+	je	.L004AVX
 	testl	$512,%ebx
 	jnz	.L005SSSE3
 .L003no_xmm:
@@ -3167,204 +3166,6 @@
 	popl	%ebp
 	ret
 .align	32
-.L004shaext:
-	subl	$32,%esp
-	movdqu	(%esi),%xmm1
-	leal	128(%ebp),%ebp
-	movdqu	16(%esi),%xmm2
-	movdqa	128(%ebp),%xmm7
-	pshufd	$27,%xmm1,%xmm0
-	pshufd	$177,%xmm1,%xmm1
-	pshufd	$27,%xmm2,%xmm2
-.byte	102,15,58,15,202,8
-	punpcklqdq	%xmm0,%xmm2
-	jmp	.L010loop_shaext
-.align	16
-.L010loop_shaext:
-	movdqu	(%edi),%xmm3
-	movdqu	16(%edi),%xmm4
-	movdqu	32(%edi),%xmm5
-.byte	102,15,56,0,223
-	movdqu	48(%edi),%xmm6
-	movdqa	%xmm2,16(%esp)
-	movdqa	-128(%ebp),%xmm0
-	paddd	%xmm3,%xmm0
-.byte	102,15,56,0,231
-.byte	15,56,203,209
-	pshufd	$14,%xmm0,%xmm0
-	nop
-	movdqa	%xmm1,(%esp)
-.byte	15,56,203,202
-	movdqa	-112(%ebp),%xmm0
-	paddd	%xmm4,%xmm0
-.byte	102,15,56,0,239
-.byte	15,56,203,209
-	pshufd	$14,%xmm0,%xmm0
-	leal	64(%edi),%edi
-.byte	15,56,204,220
-.byte	15,56,203,202
-	movdqa	-96(%ebp),%xmm0
-	paddd	%xmm5,%xmm0
-.byte	102,15,56,0,247
-.byte	15,56,203,209
-	pshufd	$14,%xmm0,%xmm0
-	movdqa	%xmm6,%xmm7
-.byte	102,15,58,15,253,4
-	nop
-	paddd	%xmm7,%xmm3
-.byte	15,56,204,229
-.byte	15,56,203,202
-	movdqa	-80(%ebp),%xmm0
-	paddd	%xmm6,%xmm0
-.byte	15,56,205,222
-.byte	15,56,203,209
-	pshufd	$14,%xmm0,%xmm0
-	movdqa	%xmm3,%xmm7
-.byte	102,15,58,15,254,4
-	nop
-	paddd	%xmm7,%xmm4
-.byte	15,56,204,238
-.byte	15,56,203,202
-	movdqa	-64(%ebp),%xmm0
-	paddd	%xmm3,%xmm0
-.byte	15,56,205,227
-.byte	15,56,203,209
-	pshufd	$14,%xmm0,%xmm0
-	movdqa	%xmm4,%xmm7
-.byte	102,15,58,15,251,4
-	nop
-	paddd	%xmm7,%xmm5
-.byte	15,56,204,243
-.byte	15,56,203,202
-	movdqa	-48(%ebp),%xmm0
-	paddd	%xmm4,%xmm0
-.byte	15,56,205,236
-.byte	15,56,203,209
-	pshufd	$14,%xmm0,%xmm0
-	movdqa	%xmm5,%xmm7
-.byte	102,15,58,15,252,4
-	nop
-	paddd	%xmm7,%xmm6
-.byte	15,56,204,220
-.byte	15,56,203,202
-	movdqa	-32(%ebp),%xmm0
-	paddd	%xmm5,%xmm0
-.byte	15,56,205,245
-.byte	15,56,203,209
-	pshufd	$14,%xmm0,%xmm0
-	movdqa	%xmm6,%xmm7
-.byte	102,15,58,15,253,4
-	nop
-	paddd	%xmm7,%xmm3
-.byte	15,56,204,229
-.byte	15,56,203,202
-	movdqa	-16(%ebp),%xmm0
-	paddd	%xmm6,%xmm0
-.byte	15,56,205,222
-.byte	15,56,203,209
-	pshufd	$14,%xmm0,%xmm0
-	movdqa	%xmm3,%xmm7
-.byte	102,15,58,15,254,4
-	nop
-	paddd	%xmm7,%xmm4
-.byte	15,56,204,238
-.byte	15,56,203,202
-	movdqa	(%ebp),%xmm0
-	paddd	%xmm3,%xmm0
-.byte	15,56,205,227
-.byte	15,56,203,209
-	pshufd	$14,%xmm0,%xmm0
-	movdqa	%xmm4,%xmm7
-.byte	102,15,58,15,251,4
-	nop
-	paddd	%xmm7,%xmm5
-.byte	15,56,204,243
-.byte	15,56,203,202
-	movdqa	16(%ebp),%xmm0
-	paddd	%xmm4,%xmm0
-.byte	15,56,205,236
-.byte	15,56,203,209
-	pshufd	$14,%xmm0,%xmm0
-	movdqa	%xmm5,%xmm7
-.byte	102,15,58,15,252,4
-	nop
-	paddd	%xmm7,%xmm6
-.byte	15,56,204,220
-.byte	15,56,203,202
-	movdqa	32(%ebp),%xmm0
-	paddd	%xmm5,%xmm0
-.byte	15,56,205,245
-.byte	15,56,203,209
-	pshufd	$14,%xmm0,%xmm0
-	movdqa	%xmm6,%xmm7
-.byte	102,15,58,15,253,4
-	nop
-	paddd	%xmm7,%xmm3
-.byte	15,56,204,229
-.byte	15,56,203,202
-	movdqa	48(%ebp),%xmm0
-	paddd	%xmm6,%xmm0
-.byte	15,56,205,222
-.byte	15,56,203,209
-	pshufd	$14,%xmm0,%xmm0
-	movdqa	%xmm3,%xmm7
-.byte	102,15,58,15,254,4
-	nop
-	paddd	%xmm7,%xmm4
-.byte	15,56,204,238
-.byte	15,56,203,202
-	movdqa	64(%ebp),%xmm0
-	paddd	%xmm3,%xmm0
-.byte	15,56,205,227
-.byte	15,56,203,209
-	pshufd	$14,%xmm0,%xmm0
-	movdqa	%xmm4,%xmm7
-.byte	102,15,58,15,251,4
-	nop
-	paddd	%xmm7,%xmm5
-.byte	15,56,204,243
-.byte	15,56,203,202
-	movdqa	80(%ebp),%xmm0
-	paddd	%xmm4,%xmm0
-.byte	15,56,205,236
-.byte	15,56,203,209
-	pshufd	$14,%xmm0,%xmm0
-	movdqa	%xmm5,%xmm7
-.byte	102,15,58,15,252,4
-.byte	15,56,203,202
-	paddd	%xmm7,%xmm6
-	movdqa	96(%ebp),%xmm0
-	paddd	%xmm5,%xmm0
-.byte	15,56,203,209
-	pshufd	$14,%xmm0,%xmm0
-.byte	15,56,205,245
-	movdqa	128(%ebp),%xmm7
-.byte	15,56,203,202
-	movdqa	112(%ebp),%xmm0
-	paddd	%xmm6,%xmm0
-	nop
-.byte	15,56,203,209
-	pshufd	$14,%xmm0,%xmm0
-	cmpl	%edi,%eax
-	nop
-.byte	15,56,203,202
-	paddd	16(%esp),%xmm2
-	paddd	(%esp),%xmm1
-	jnz	.L010loop_shaext
-	pshufd	$177,%xmm2,%xmm2
-	pshufd	$27,%xmm1,%xmm7
-	pshufd	$177,%xmm1,%xmm1
-	punpckhqdq	%xmm2,%xmm1
-.byte	102,15,58,15,215,8
-	movl	44(%esp),%esp
-	movdqu	%xmm1,(%esi)
-	movdqu	%xmm2,16(%esi)
-	popl	%edi
-	popl	%esi
-	popl	%ebx
-	popl	%ebp
-	ret
-.align	32
 .L005SSSE3:
 	leal	-96(%esp),%esp
 	movl	(%esi),%eax
@@ -3384,9 +3185,9 @@
 	movl	%ecx,24(%esp)
 	movl	%esi,28(%esp)
 	movdqa	256(%ebp),%xmm7
-	jmp	.L011grand_ssse3
+	jmp	.L010grand_ssse3
 .align	16
-.L011grand_ssse3:
+.L010grand_ssse3:
 	movdqu	(%edi),%xmm0
 	movdqu	16(%edi),%xmm1
 	movdqu	32(%edi),%xmm2
@@ -3409,9 +3210,9 @@
 	paddd	%xmm3,%xmm7
 	movdqa	%xmm6,64(%esp)
 	movdqa	%xmm7,80(%esp)
-	jmp	.L012ssse3_00_47
+	jmp	.L011ssse3_00_47
 .align	16
-.L012ssse3_00_47:
+.L011ssse3_00_47:
 	addl	$64,%ebp
 	movl	%edx,%ecx
 	movdqa	%xmm1,%xmm4
@@ -4054,7 +3855,7 @@
 	addl	%ecx,%eax
 	movdqa	%xmm6,80(%esp)
 	cmpl	$66051,64(%ebp)
-	jne	.L012ssse3_00_47
+	jne	.L011ssse3_00_47
 	movl	%edx,%ecx
 	rorl	$14,%edx
 	movl	20(%esp),%esi
@@ -4568,12 +4369,1193 @@
 	movdqa	64(%ebp),%xmm7
 	subl	$192,%ebp
 	cmpl	104(%esp),%edi
-	jb	.L011grand_ssse3
+	jb	.L010grand_ssse3
 	movl	108(%esp),%esp
 	popl	%edi
 	popl	%esi
 	popl	%ebx
 	popl	%ebp
 	ret
+.align	32
+.L004AVX:
+	leal	-96(%esp),%esp
+	vzeroall
+	movl	(%esi),%eax
+	movl	4(%esi),%ebx
+	movl	8(%esi),%ecx
+	movl	12(%esi),%edi
+	movl	%ebx,4(%esp)
+	xorl	%ecx,%ebx
+	movl	%ecx,8(%esp)
+	movl	%edi,12(%esp)
+	movl	16(%esi),%edx
+	movl	20(%esi),%edi
+	movl	24(%esi),%ecx
+	movl	28(%esi),%esi
+	movl	%edi,20(%esp)
+	movl	100(%esp),%edi
+	movl	%ecx,24(%esp)
+	movl	%esi,28(%esp)
+	vmovdqa	256(%ebp),%xmm7
+	jmp	.L012grand_avx
+.align	32
+.L012grand_avx:
+	vmovdqu	(%edi),%xmm0
+	vmovdqu	16(%edi),%xmm1
+	vmovdqu	32(%edi),%xmm2
+	vmovdqu	48(%edi),%xmm3
+	addl	$64,%edi
+	vpshufb	%xmm7,%xmm0,%xmm0
+	movl	%edi,100(%esp)
+	vpshufb	%xmm7,%xmm1,%xmm1
+	vpshufb	%xmm7,%xmm2,%xmm2
+	vpaddd	(%ebp),%xmm0,%xmm4
+	vpshufb	%xmm7,%xmm3,%xmm3
+	vpaddd	16(%ebp),%xmm1,%xmm5
+	vpaddd	32(%ebp),%xmm2,%xmm6
+	vpaddd	48(%ebp),%xmm3,%xmm7
+	vmovdqa	%xmm4,32(%esp)
+	vmovdqa	%xmm5,48(%esp)
+	vmovdqa	%xmm6,64(%esp)
+	vmovdqa	%xmm7,80(%esp)
+	jmp	.L013avx_00_47
+.align	16
+.L013avx_00_47:
+	addl	$64,%ebp
+	vpalignr	$4,%xmm0,%xmm1,%xmm4
+	movl	%edx,%ecx
+	shrdl	$14,%edx,%edx
+	movl	20(%esp),%esi
+	vpalignr	$4,%xmm2,%xmm3,%xmm7
+	xorl	%ecx,%edx
+	movl	24(%esp),%edi
+	xorl	%edi,%esi
+	vpsrld	$7,%xmm4,%xmm6
+	shrdl	$5,%edx,%edx
+	andl	%ecx,%esi
+	movl	%ecx,16(%esp)
+	vpaddd	%xmm7,%xmm0,%xmm0
+	xorl	%ecx,%edx
+	xorl	%esi,%edi
+	shrdl	$6,%edx,%edx
+	vpsrld	$3,%xmm4,%xmm7
+	movl	%eax,%ecx
+	addl	%edi,%edx
+	movl	4(%esp),%edi
+	vpslld	$14,%xmm4,%xmm5
+	movl	%eax,%esi
+	shrdl	$9,%ecx,%ecx
+	movl	%eax,(%esp)
+	vpxor	%xmm6,%xmm7,%xmm4
+	xorl	%eax,%ecx
+	xorl	%edi,%eax
+	addl	28(%esp),%edx
+	vpshufd	$250,%xmm3,%xmm7
+	shrdl	$11,%ecx,%ecx
+	andl	%eax,%ebx
+	xorl	%esi,%ecx
+	vpsrld	$11,%xmm6,%xmm6
+	addl	32(%esp),%edx
+	xorl	%edi,%ebx
+	shrdl	$2,%ecx,%ecx
+	vpxor	%xmm5,%xmm4,%xmm4
+	addl	%edx,%ebx
+	addl	12(%esp),%edx
+	addl	%ecx,%ebx
+	vpslld	$11,%xmm5,%xmm5
+	movl	%edx,%ecx
+	shrdl	$14,%edx,%edx
+	movl	16(%esp),%esi
+	vpxor	%xmm6,%xmm4,%xmm4
+	xorl	%ecx,%edx
+	movl	20(%esp),%edi
+	xorl	%edi,%esi
+	vpsrld	$10,%xmm7,%xmm6
+	shrdl	$5,%edx,%edx
+	andl	%ecx,%esi
+	movl	%ecx,12(%esp)
+	vpxor	%xmm5,%xmm4,%xmm4
+	xorl	%ecx,%edx
+	xorl	%esi,%edi
+	shrdl	$6,%edx,%edx
+	vpsrlq	$17,%xmm7,%xmm5
+	movl	%ebx,%ecx
+	addl	%edi,%edx
+	movl	(%esp),%edi
+	vpaddd	%xmm4,%xmm0,%xmm0
+	movl	%ebx,%esi
+	shrdl	$9,%ecx,%ecx
+	movl	%ebx,28(%esp)
+	vpxor	%xmm5,%xmm6,%xmm6
+	xorl	%ebx,%ecx
+	xorl	%edi,%ebx
+	addl	24(%esp),%edx
+	vpsrlq	$19,%xmm7,%xmm7
+	shrdl	$11,%ecx,%ecx
+	andl	%ebx,%eax
+	xorl	%esi,%ecx
+	vpxor	%xmm7,%xmm6,%xmm6
+	addl	36(%esp),%edx
+	xorl	%edi,%eax
+	shrdl	$2,%ecx,%ecx
+	vpshufd	$132,%xmm6,%xmm7
+	addl	%edx,%eax
+	addl	8(%esp),%edx
+	addl	%ecx,%eax
+	vpsrldq	$8,%xmm7,%xmm7
+	movl	%edx,%ecx
+	shrdl	$14,%edx,%edx
+	movl	12(%esp),%esi
+	vpaddd	%xmm7,%xmm0,%xmm0
+	xorl	%ecx,%edx
+	movl	16(%esp),%edi
+	xorl	%edi,%esi
+	vpshufd	$80,%xmm0,%xmm7
+	shrdl	$5,%edx,%edx
+	andl	%ecx,%esi
+	movl	%ecx,8(%esp)
+	vpsrld	$10,%xmm7,%xmm6
+	xorl	%ecx,%edx
+	xorl	%esi,%edi
+	shrdl	$6,%edx,%edx
+	vpsrlq	$17,%xmm7,%xmm5
+	movl	%eax,%ecx
+	addl	%edi,%edx
+	movl	28(%esp),%edi
+	vpxor	%xmm5,%xmm6,%xmm6
+	movl	%eax,%esi
+	shrdl	$9,%ecx,%ecx
+	movl	%eax,24(%esp)
+	vpsrlq	$19,%xmm7,%xmm7
+	xorl	%eax,%ecx
+	xorl	%edi,%eax
+	addl	20(%esp),%edx
+	vpxor	%xmm7,%xmm6,%xmm6
+	shrdl	$11,%ecx,%ecx
+	andl	%eax,%ebx
+	xorl	%esi,%ecx
+	vpshufd	$232,%xmm6,%xmm7
+	addl	40(%esp),%edx
+	xorl	%edi,%ebx
+	shrdl	$2,%ecx,%ecx
+	vpslldq	$8,%xmm7,%xmm7
+	addl	%edx,%ebx
+	addl	4(%esp),%edx
+	addl	%ecx,%ebx
+	vpaddd	%xmm7,%xmm0,%xmm0
+	movl	%edx,%ecx
+	shrdl	$14,%edx,%edx
+	movl	8(%esp),%esi
+	vpaddd	(%ebp),%xmm0,%xmm6
+	xorl	%ecx,%edx
+	movl	12(%esp),%edi
+	xorl	%edi,%esi
+	shrdl	$5,%edx,%edx
+	andl	%ecx,%esi
+	movl	%ecx,4(%esp)
+	xorl	%ecx,%edx
+	xorl	%esi,%edi
+	shrdl	$6,%edx,%edx
+	movl	%ebx,%ecx
+	addl	%edi,%edx
+	movl	24(%esp),%edi
+	movl	%ebx,%esi
+	shrdl	$9,%ecx,%ecx
+	movl	%ebx,20(%esp)
+	xorl	%ebx,%ecx
+	xorl	%edi,%ebx
+	addl	16(%esp),%edx
+	shrdl	$11,%ecx,%ecx
+	andl	%ebx,%eax
+	xorl	%esi,%ecx
+	addl	44(%esp),%edx
+	xorl	%edi,%eax
+	shrdl	$2,%ecx,%ecx
+	addl	%edx,%eax
+	addl	(%esp),%edx
+	addl	%ecx,%eax
+	vmovdqa	%xmm6,32(%esp)
+	vpalignr	$4,%xmm1,%xmm2,%xmm4
+	movl	%edx,%ecx
+	shrdl	$14,%edx,%edx
+	movl	4(%esp),%esi
+	vpalignr	$4,%xmm3,%xmm0,%xmm7
+	xorl	%ecx,%edx
+	movl	8(%esp),%edi
+	xorl	%edi,%esi
+	vpsrld	$7,%xmm4,%xmm6
+	shrdl	$5,%edx,%edx
+	andl	%ecx,%esi
+	movl	%ecx,(%esp)
+	vpaddd	%xmm7,%xmm1,%xmm1
+	xorl	%ecx,%edx
+	xorl	%esi,%edi
+	shrdl	$6,%edx,%edx
+	vpsrld	$3,%xmm4,%xmm7
+	movl	%eax,%ecx
+	addl	%edi,%edx
+	movl	20(%esp),%edi
+	vpslld	$14,%xmm4,%xmm5
+	movl	%eax,%esi
+	shrdl	$9,%ecx,%ecx
+	movl	%eax,16(%esp)
+	vpxor	%xmm6,%xmm7,%xmm4
+	xorl	%eax,%ecx
+	xorl	%edi,%eax
+	addl	12(%esp),%edx
+	vpshufd	$250,%xmm0,%xmm7
+	shrdl	$11,%ecx,%ecx
+	andl	%eax,%ebx
+	xorl	%esi,%ecx
+	vpsrld	$11,%xmm6,%xmm6
+	addl	48(%esp),%edx
+	xorl	%edi,%ebx
+	shrdl	$2,%ecx,%ecx
+	vpxor	%xmm5,%xmm4,%xmm4
+	addl	%edx,%ebx
+	addl	28(%esp),%edx
+	addl	%ecx,%ebx
+	vpslld	$11,%xmm5,%xmm5
+	movl	%edx,%ecx
+	shrdl	$14,%edx,%edx
+	movl	(%esp),%esi
+	vpxor	%xmm6,%xmm4,%xmm4
+	xorl	%ecx,%edx
+	movl	4(%esp),%edi
+	xorl	%edi,%esi
+	vpsrld	$10,%xmm7,%xmm6
+	shrdl	$5,%edx,%edx
+	andl	%ecx,%esi
+	movl	%ecx,28(%esp)
+	vpxor	%xmm5,%xmm4,%xmm4
+	xorl	%ecx,%edx
+	xorl	%esi,%edi
+	shrdl	$6,%edx,%edx
+	vpsrlq	$17,%xmm7,%xmm5
+	movl	%ebx,%ecx
+	addl	%edi,%edx
+	movl	16(%esp),%edi
+	vpaddd	%xmm4,%xmm1,%xmm1
+	movl	%ebx,%esi
+	shrdl	$9,%ecx,%ecx
+	movl	%ebx,12(%esp)
+	vpxor	%xmm5,%xmm6,%xmm6
+	xorl	%ebx,%ecx
+	xorl	%edi,%ebx
+	addl	8(%esp),%edx
+	vpsrlq	$19,%xmm7,%xmm7
+	shrdl	$11,%ecx,%ecx
+	andl	%ebx,%eax
+	xorl	%esi,%ecx
+	vpxor	%xmm7,%xmm6,%xmm6
+	addl	52(%esp),%edx
+	xorl	%edi,%eax
+	shrdl	$2,%ecx,%ecx
+	vpshufd	$132,%xmm6,%xmm7
+	addl	%edx,%eax
+	addl	24(%esp),%edx
+	addl	%ecx,%eax
+	vpsrldq	$8,%xmm7,%xmm7
+	movl	%edx,%ecx
+	shrdl	$14,%edx,%edx
+	movl	28(%esp),%esi
+	vpaddd	%xmm7,%xmm1,%xmm1
+	xorl	%ecx,%edx
+	movl	(%esp),%edi
+	xorl	%edi,%esi
+	vpshufd	$80,%xmm1,%xmm7
+	shrdl	$5,%edx,%edx
+	andl	%ecx,%esi
+	movl	%ecx,24(%esp)
+	vpsrld	$10,%xmm7,%xmm6
+	xorl	%ecx,%edx
+	xorl	%esi,%edi
+	shrdl	$6,%edx,%edx
+	vpsrlq	$17,%xmm7,%xmm5
+	movl	%eax,%ecx
+	addl	%edi,%edx
+	movl	12(%esp),%edi
+	vpxor	%xmm5,%xmm6,%xmm6
+	movl	%eax,%esi
+	shrdl	$9,%ecx,%ecx
+	movl	%eax,8(%esp)
+	vpsrlq	$19,%xmm7,%xmm7
+	xorl	%eax,%ecx
+	xorl	%edi,%eax
+	addl	4(%esp),%edx
+	vpxor	%xmm7,%xmm6,%xmm6
+	shrdl	$11,%ecx,%ecx
+	andl	%eax,%ebx
+	xorl	%esi,%ecx
+	vpshufd	$232,%xmm6,%xmm7
+	addl	56(%esp),%edx
+	xorl	%edi,%ebx
+	shrdl	$2,%ecx,%ecx
+	vpslldq	$8,%xmm7,%xmm7
+	addl	%edx,%ebx
+	addl	20(%esp),%edx
+	addl	%ecx,%ebx
+	vpaddd	%xmm7,%xmm1,%xmm1
+	movl	%edx,%ecx
+	shrdl	$14,%edx,%edx
+	movl	24(%esp),%esi
+	vpaddd	16(%ebp),%xmm1,%xmm6
+	xorl	%ecx,%edx
+	movl	28(%esp),%edi
+	xorl	%edi,%esi
+	shrdl	$5,%edx,%edx
+	andl	%ecx,%esi
+	movl	%ecx,20(%esp)
+	xorl	%ecx,%edx
+	xorl	%esi,%edi
+	shrdl	$6,%edx,%edx
+	movl	%ebx,%ecx
+	addl	%edi,%edx
+	movl	8(%esp),%edi
+	movl	%ebx,%esi
+	shrdl	$9,%ecx,%ecx
+	movl	%ebx,4(%esp)
+	xorl	%ebx,%ecx
+	xorl	%edi,%ebx
+	addl	(%esp),%edx
+	shrdl	$11,%ecx,%ecx
+	andl	%ebx,%eax
+	xorl	%esi,%ecx
+	addl	60(%esp),%edx
+	xorl	%edi,%eax
+	shrdl	$2,%ecx,%ecx
+	addl	%edx,%eax
+	addl	16(%esp),%edx
+	addl	%ecx,%eax
+	vmovdqa	%xmm6,48(%esp)
+	vpalignr	$4,%xmm2,%xmm3,%xmm4
+	movl	%edx,%ecx
+	shrdl	$14,%edx,%edx
+	movl	20(%esp),%esi
+	vpalignr	$4,%xmm0,%xmm1,%xmm7
+	xorl	%ecx,%edx
+	movl	24(%esp),%edi
+	xorl	%edi,%esi
+	vpsrld	$7,%xmm4,%xmm6
+	shrdl	$5,%edx,%edx
+	andl	%ecx,%esi
+	movl	%ecx,16(%esp)
+	vpaddd	%xmm7,%xmm2,%xmm2
+	xorl	%ecx,%edx
+	xorl	%esi,%edi
+	shrdl	$6,%edx,%edx
+	vpsrld	$3,%xmm4,%xmm7
+	movl	%eax,%ecx
+	addl	%edi,%edx
+	movl	4(%esp),%edi
+	vpslld	$14,%xmm4,%xmm5
+	movl	%eax,%esi
+	shrdl	$9,%ecx,%ecx
+	movl	%eax,(%esp)
+	vpxor	%xmm6,%xmm7,%xmm4
+	xorl	%eax,%ecx
+	xorl	%edi,%eax
+	addl	28(%esp),%edx
+	vpshufd	$250,%xmm1,%xmm7
+	shrdl	$11,%ecx,%ecx
+	andl	%eax,%ebx
+	xorl	%esi,%ecx
+	vpsrld	$11,%xmm6,%xmm6
+	addl	64(%esp),%edx
+	xorl	%edi,%ebx
+	shrdl	$2,%ecx,%ecx
+	vpxor	%xmm5,%xmm4,%xmm4
+	addl	%edx,%ebx
+	addl	12(%esp),%edx
+	addl	%ecx,%ebx
+	vpslld	$11,%xmm5,%xmm5
+	movl	%edx,%ecx
+	shrdl	$14,%edx,%edx
+	movl	16(%esp),%esi
+	vpxor	%xmm6,%xmm4,%xmm4
+	xorl	%ecx,%edx
+	movl	20(%esp),%edi
+	xorl	%edi,%esi
+	vpsrld	$10,%xmm7,%xmm6
+	shrdl	$5,%edx,%edx
+	andl	%ecx,%esi
+	movl	%ecx,12(%esp)
+	vpxor	%xmm5,%xmm4,%xmm4
+	xorl	%ecx,%edx
+	xorl	%esi,%edi
+	shrdl	$6,%edx,%edx
+	vpsrlq	$17,%xmm7,%xmm5
+	movl	%ebx,%ecx
+	addl	%edi,%edx
+	movl	(%esp),%edi
+	vpaddd	%xmm4,%xmm2,%xmm2
+	movl	%ebx,%esi
+	shrdl	$9,%ecx,%ecx
+	movl	%ebx,28(%esp)
+	vpxor	%xmm5,%xmm6,%xmm6
+	xorl	%ebx,%ecx
+	xorl	%edi,%ebx
+	addl	24(%esp),%edx
+	vpsrlq	$19,%xmm7,%xmm7
+	shrdl	$11,%ecx,%ecx
+	andl	%ebx,%eax
+	xorl	%esi,%ecx
+	vpxor	%xmm7,%xmm6,%xmm6
+	addl	68(%esp),%edx
+	xorl	%edi,%eax
+	shrdl	$2,%ecx,%ecx
+	vpshufd	$132,%xmm6,%xmm7
+	addl	%edx,%eax
+	addl	8(%esp),%edx
+	addl	%ecx,%eax
+	vpsrldq	$8,%xmm7,%xmm7
+	movl	%edx,%ecx
+	shrdl	$14,%edx,%edx
+	movl	12(%esp),%esi
+	vpaddd	%xmm7,%xmm2,%xmm2
+	xorl	%ecx,%edx
+	movl	16(%esp),%edi
+	xorl	%edi,%esi
+	vpshufd	$80,%xmm2,%xmm7
+	shrdl	$5,%edx,%edx
+	andl	%ecx,%esi
+	movl	%ecx,8(%esp)
+	vpsrld	$10,%xmm7,%xmm6
+	xorl	%ecx,%edx
+	xorl	%esi,%edi
+	shrdl	$6,%edx,%edx
+	vpsrlq	$17,%xmm7,%xmm5
+	movl	%eax,%ecx
+	addl	%edi,%edx
+	movl	28(%esp),%edi
+	vpxor	%xmm5,%xmm6,%xmm6
+	movl	%eax,%esi
+	shrdl	$9,%ecx,%ecx
+	movl	%eax,24(%esp)
+	vpsrlq	$19,%xmm7,%xmm7
+	xorl	%eax,%ecx
+	xorl	%edi,%eax
+	addl	20(%esp),%edx
+	vpxor	%xmm7,%xmm6,%xmm6
+	shrdl	$11,%ecx,%ecx
+	andl	%eax,%ebx
+	xorl	%esi,%ecx
+	vpshufd	$232,%xmm6,%xmm7
+	addl	72(%esp),%edx
+	xorl	%edi,%ebx
+	shrdl	$2,%ecx,%ecx
+	vpslldq	$8,%xmm7,%xmm7
+	addl	%edx,%ebx
+	addl	4(%esp),%edx
+	addl	%ecx,%ebx
+	vpaddd	%xmm7,%xmm2,%xmm2
+	movl	%edx,%ecx
+	shrdl	$14,%edx,%edx
+	movl	8(%esp),%esi
+	vpaddd	32(%ebp),%xmm2,%xmm6
+	xorl	%ecx,%edx
+	movl	12(%esp),%edi
+	xorl	%edi,%esi
+	shrdl	$5,%edx,%edx
+	andl	%ecx,%esi
+	movl	%ecx,4(%esp)
+	xorl	%ecx,%edx
+	xorl	%esi,%edi
+	shrdl	$6,%edx,%edx
+	movl	%ebx,%ecx
+	addl	%edi,%edx
+	movl	24(%esp),%edi
+	movl	%ebx,%esi
+	shrdl	$9,%ecx,%ecx
+	movl	%ebx,20(%esp)
+	xorl	%ebx,%ecx
+	xorl	%edi,%ebx
+	addl	16(%esp),%edx
+	shrdl	$11,%ecx,%ecx
+	andl	%ebx,%eax
+	xorl	%esi,%ecx
+	addl	76(%esp),%edx
+	xorl	%edi,%eax
+	shrdl	$2,%ecx,%ecx
+	addl	%edx,%eax
+	addl	(%esp),%edx
+	addl	%ecx,%eax
+	vmovdqa	%xmm6,64(%esp)
+	vpalignr	$4,%xmm3,%xmm0,%xmm4
+	movl	%edx,%ecx
+	shrdl	$14,%edx,%edx
+	movl	4(%esp),%esi
+	vpalignr	$4,%xmm1,%xmm2,%xmm7
+	xorl	%ecx,%edx
+	movl	8(%esp),%edi
+	xorl	%edi,%esi
+	vpsrld	$7,%xmm4,%xmm6
+	shrdl	$5,%edx,%edx
+	andl	%ecx,%esi
+	movl	%ecx,(%esp)
+	vpaddd	%xmm7,%xmm3,%xmm3
+	xorl	%ecx,%edx
+	xorl	%esi,%edi
+	shrdl	$6,%edx,%edx
+	vpsrld	$3,%xmm4,%xmm7
+	movl	%eax,%ecx
+	addl	%edi,%edx
+	movl	20(%esp),%edi
+	vpslld	$14,%xmm4,%xmm5
+	movl	%eax,%esi
+	shrdl	$9,%ecx,%ecx
+	movl	%eax,16(%esp)
+	vpxor	%xmm6,%xmm7,%xmm4
+	xorl	%eax,%ecx
+	xorl	%edi,%eax
+	addl	12(%esp),%edx
+	vpshufd	$250,%xmm2,%xmm7
+	shrdl	$11,%ecx,%ecx
+	andl	%eax,%ebx
+	xorl	%esi,%ecx
+	vpsrld	$11,%xmm6,%xmm6
+	addl	80(%esp),%edx
+	xorl	%edi,%ebx
+	shrdl	$2,%ecx,%ecx
+	vpxor	%xmm5,%xmm4,%xmm4
+	addl	%edx,%ebx
+	addl	28(%esp),%edx
+	addl	%ecx,%ebx
+	vpslld	$11,%xmm5,%xmm5
+	movl	%edx,%ecx
+	shrdl	$14,%edx,%edx
+	movl	(%esp),%esi
+	vpxor	%xmm6,%xmm4,%xmm4
+	xorl	%ecx,%edx
+	movl	4(%esp),%edi
+	xorl	%edi,%esi
+	vpsrld	$10,%xmm7,%xmm6
+	shrdl	$5,%edx,%edx
+	andl	%ecx,%esi
+	movl	%ecx,28(%esp)
+	vpxor	%xmm5,%xmm4,%xmm4
+	xorl	%ecx,%edx
+	xorl	%esi,%edi
+	shrdl	$6,%edx,%edx
+	vpsrlq	$17,%xmm7,%xmm5
+	movl	%ebx,%ecx
+	addl	%edi,%edx
+	movl	16(%esp),%edi
+	vpaddd	%xmm4,%xmm3,%xmm3
+	movl	%ebx,%esi
+	shrdl	$9,%ecx,%ecx
+	movl	%ebx,12(%esp)
+	vpxor	%xmm5,%xmm6,%xmm6
+	xorl	%ebx,%ecx
+	xorl	%edi,%ebx
+	addl	8(%esp),%edx
+	vpsrlq	$19,%xmm7,%xmm7
+	shrdl	$11,%ecx,%ecx
+	andl	%ebx,%eax
+	xorl	%esi,%ecx
+	vpxor	%xmm7,%xmm6,%xmm6
+	addl	84(%esp),%edx
+	xorl	%edi,%eax
+	shrdl	$2,%ecx,%ecx
+	vpshufd	$132,%xmm6,%xmm7
+	addl	%edx,%eax
+	addl	24(%esp),%edx
+	addl	%ecx,%eax
+	vpsrldq	$8,%xmm7,%xmm7
+	movl	%edx,%ecx
+	shrdl	$14,%edx,%edx
+	movl	28(%esp),%esi
+	vpaddd	%xmm7,%xmm3,%xmm3
+	xorl	%ecx,%edx
+	movl	(%esp),%edi
+	xorl	%edi,%esi
+	vpshufd	$80,%xmm3,%xmm7
+	shrdl	$5,%edx,%edx
+	andl	%ecx,%esi
+	movl	%ecx,24(%esp)
+	vpsrld	$10,%xmm7,%xmm6
+	xorl	%ecx,%edx
+	xorl	%esi,%edi
+	shrdl	$6,%edx,%edx
+	vpsrlq	$17,%xmm7,%xmm5
+	movl	%eax,%ecx
+	addl	%edi,%edx
+	movl	12(%esp),%edi
+	vpxor	%xmm5,%xmm6,%xmm6
+	movl	%eax,%esi
+	shrdl	$9,%ecx,%ecx
+	movl	%eax,8(%esp)
+	vpsrlq	$19,%xmm7,%xmm7
+	xorl	%eax,%ecx
+	xorl	%edi,%eax
+	addl	4(%esp),%edx
+	vpxor	%xmm7,%xmm6,%xmm6
+	shrdl	$11,%ecx,%ecx
+	andl	%eax,%ebx
+	xorl	%esi,%ecx
+	vpshufd	$232,%xmm6,%xmm7
+	addl	88(%esp),%edx
+	xorl	%edi,%ebx
+	shrdl	$2,%ecx,%ecx
+	vpslldq	$8,%xmm7,%xmm7
+	addl	%edx,%ebx
+	addl	20(%esp),%edx
+	addl	%ecx,%ebx
+	vpaddd	%xmm7,%xmm3,%xmm3
+	movl	%edx,%ecx
+	shrdl	$14,%edx,%edx
+	movl	24(%esp),%esi
+	vpaddd	48(%ebp),%xmm3,%xmm6
+	xorl	%ecx,%edx
+	movl	28(%esp),%edi
+	xorl	%edi,%esi
+	shrdl	$5,%edx,%edx
+	andl	%ecx,%esi
+	movl	%ecx,20(%esp)
+	xorl	%ecx,%edx
+	xorl	%esi,%edi
+	shrdl	$6,%edx,%edx
+	movl	%ebx,%ecx
+	addl	%edi,%edx
+	movl	8(%esp),%edi
+	movl	%ebx,%esi
+	shrdl	$9,%ecx,%ecx
+	movl	%ebx,4(%esp)
+	xorl	%ebx,%ecx
+	xorl	%edi,%ebx
+	addl	(%esp),%edx
+	shrdl	$11,%ecx,%ecx
+	andl	%ebx,%eax
+	xorl	%esi,%ecx
+	addl	92(%esp),%edx
+	xorl	%edi,%eax
+	shrdl	$2,%ecx,%ecx
+	addl	%edx,%eax
+	addl	16(%esp),%edx
+	addl	%ecx,%eax
+	vmovdqa	%xmm6,80(%esp)
+	cmpl	$66051,64(%ebp)
+	jne	.L013avx_00_47
+	movl	%edx,%ecx
+	shrdl	$14,%edx,%edx
+	movl	20(%esp),%esi
+	xorl	%ecx,%edx
+	movl	24(%esp),%edi
+	xorl	%edi,%esi
+	shrdl	$5,%edx,%edx
+	andl	%ecx,%esi
+	movl	%ecx,16(%esp)
+	xorl	%ecx,%edx
+	xorl	%esi,%edi
+	shrdl	$6,%edx,%edx
+	movl	%eax,%ecx
+	addl	%edi,%edx
+	movl	4(%esp),%edi
+	movl	%eax,%esi
+	shrdl	$9,%ecx,%ecx
+	movl	%eax,(%esp)
+	xorl	%eax,%ecx
+	xorl	%edi,%eax
+	addl	28(%esp),%edx
+	shrdl	$11,%ecx,%ecx
+	andl	%eax,%ebx
+	xorl	%esi,%ecx
+	addl	32(%esp),%edx
+	xorl	%edi,%ebx
+	shrdl	$2,%ecx,%ecx
+	addl	%edx,%ebx
+	addl	12(%esp),%edx
+	addl	%ecx,%ebx
+	movl	%edx,%ecx
+	shrdl	$14,%edx,%edx
+	movl	16(%esp),%esi
+	xorl	%ecx,%edx
+	movl	20(%esp),%edi
+	xorl	%edi,%esi
+	shrdl	$5,%edx,%edx
+	andl	%ecx,%esi
+	movl	%ecx,12(%esp)
+	xorl	%ecx,%edx
+	xorl	%esi,%edi
+	shrdl	$6,%edx,%edx
+	movl	%ebx,%ecx
+	addl	%edi,%edx
+	movl	(%esp),%edi
+	movl	%ebx,%esi
+	shrdl	$9,%ecx,%ecx
+	movl	%ebx,28(%esp)
+	xorl	%ebx,%ecx
+	xorl	%edi,%ebx
+	addl	24(%esp),%edx
+	shrdl	$11,%ecx,%ecx
+	andl	%ebx,%eax
+	xorl	%esi,%ecx
+	addl	36(%esp),%edx
+	xorl	%edi,%eax
+	shrdl	$2,%ecx,%ecx
+	addl	%edx,%eax
+	addl	8(%esp),%edx
+	addl	%ecx,%eax
+	movl	%edx,%ecx
+	shrdl	$14,%edx,%edx
+	movl	12(%esp),%esi
+	xorl	%ecx,%edx
+	movl	16(%esp),%edi
+	xorl	%edi,%esi
+	shrdl	$5,%edx,%edx
+	andl	%ecx,%esi
+	movl	%ecx,8(%esp)
+	xorl	%ecx,%edx
+	xorl	%esi,%edi
+	shrdl	$6,%edx,%edx
+	movl	%eax,%ecx
+	addl	%edi,%edx
+	movl	28(%esp),%edi
+	movl	%eax,%esi
+	shrdl	$9,%ecx,%ecx
+	movl	%eax,24(%esp)
+	xorl	%eax,%ecx
+	xorl	%edi,%eax
+	addl	20(%esp),%edx
+	shrdl	$11,%ecx,%ecx
+	andl	%eax,%ebx
+	xorl	%esi,%ecx
+	addl	40(%esp),%edx
+	xorl	%edi,%ebx
+	shrdl	$2,%ecx,%ecx
+	addl	%edx,%ebx
+	addl	4(%esp),%edx
+	addl	%ecx,%ebx
+	movl	%edx,%ecx
+	shrdl	$14,%edx,%edx
+	movl	8(%esp),%esi
+	xorl	%ecx,%edx
+	movl	12(%esp),%edi
+	xorl	%edi,%esi
+	shrdl	$5,%edx,%edx
+	andl	%ecx,%esi
+	movl	%ecx,4(%esp)
+	xorl	%ecx,%edx
+	xorl	%esi,%edi
+	shrdl	$6,%edx,%edx
+	movl	%ebx,%ecx
+	addl	%edi,%edx
+	movl	24(%esp),%edi
+	movl	%ebx,%esi
+	shrdl	$9,%ecx,%ecx
+	movl	%ebx,20(%esp)
+	xorl	%ebx,%ecx
+	xorl	%edi,%ebx
+	addl	16(%esp),%edx
+	shrdl	$11,%ecx,%ecx
+	andl	%ebx,%eax
+	xorl	%esi,%ecx
+	addl	44(%esp),%edx
+	xorl	%edi,%eax
+	shrdl	$2,%ecx,%ecx
+	addl	%edx,%eax
+	addl	(%esp),%edx
+	addl	%ecx,%eax
+	movl	%edx,%ecx
+	shrdl	$14,%edx,%edx
+	movl	4(%esp),%esi
+	xorl	%ecx,%edx
+	movl	8(%esp),%edi
+	xorl	%edi,%esi
+	shrdl	$5,%edx,%edx
+	andl	%ecx,%esi
+	movl	%ecx,(%esp)
+	xorl	%ecx,%edx
+	xorl	%esi,%edi
+	shrdl	$6,%edx,%edx
+	movl	%eax,%ecx
+	addl	%edi,%edx
+	movl	20(%esp),%edi
+	movl	%eax,%esi
+	shrdl	$9,%ecx,%ecx
+	movl	%eax,16(%esp)
+	xorl	%eax,%ecx
+	xorl	%edi,%eax
+	addl	12(%esp),%edx
+	shrdl	$11,%ecx,%ecx
+	andl	%eax,%ebx
+	xorl	%esi,%ecx
+	addl	48(%esp),%edx
+	xorl	%edi,%ebx
+	shrdl	$2,%ecx,%ecx
+	addl	%edx,%ebx
+	addl	28(%esp),%edx
+	addl	%ecx,%ebx
+	movl	%edx,%ecx
+	shrdl	$14,%edx,%edx
+	movl	(%esp),%esi
+	xorl	%ecx,%edx
+	movl	4(%esp),%edi
+	xorl	%edi,%esi
+	shrdl	$5,%edx,%edx
+	andl	%ecx,%esi
+	movl	%ecx,28(%esp)
+	xorl	%ecx,%edx
+	xorl	%esi,%edi
+	shrdl	$6,%edx,%edx
+	movl	%ebx,%ecx
+	addl	%edi,%edx
+	movl	16(%esp),%edi
+	movl	%ebx,%esi
+	shrdl	$9,%ecx,%ecx
+	movl	%ebx,12(%esp)
+	xorl	%ebx,%ecx
+	xorl	%edi,%ebx
+	addl	8(%esp),%edx
+	shrdl	$11,%ecx,%ecx
+	andl	%ebx,%eax
+	xorl	%esi,%ecx
+	addl	52(%esp),%edx
+	xorl	%edi,%eax
+	shrdl	$2,%ecx,%ecx
+	addl	%edx,%eax
+	addl	24(%esp),%edx
+	addl	%ecx,%eax
+	movl	%edx,%ecx
+	shrdl	$14,%edx,%edx
+	movl	28(%esp),%esi
+	xorl	%ecx,%edx
+	movl	(%esp),%edi
+	xorl	%edi,%esi
+	shrdl	$5,%edx,%edx
+	andl	%ecx,%esi
+	movl	%ecx,24(%esp)
+	xorl	%ecx,%edx
+	xorl	%esi,%edi
+	shrdl	$6,%edx,%edx
+	movl	%eax,%ecx
+	addl	%edi,%edx
+	movl	12(%esp),%edi
+	movl	%eax,%esi
+	shrdl	$9,%ecx,%ecx
+	movl	%eax,8(%esp)
+	xorl	%eax,%ecx
+	xorl	%edi,%eax
+	addl	4(%esp),%edx
+	shrdl	$11,%ecx,%ecx
+	andl	%eax,%ebx
+	xorl	%esi,%ecx
+	addl	56(%esp),%edx
+	xorl	%edi,%ebx
+	shrdl	$2,%ecx,%ecx
+	addl	%edx,%ebx
+	addl	20(%esp),%edx
+	addl	%ecx,%ebx
+	movl	%edx,%ecx
+	shrdl	$14,%edx,%edx
+	movl	24(%esp),%esi
+	xorl	%ecx,%edx
+	movl	28(%esp),%edi
+	xorl	%edi,%esi
+	shrdl	$5,%edx,%edx
+	andl	%ecx,%esi
+	movl	%ecx,20(%esp)
+	xorl	%ecx,%edx
+	xorl	%esi,%edi
+	shrdl	$6,%edx,%edx
+	movl	%ebx,%ecx
+	addl	%edi,%edx
+	movl	8(%esp),%edi
+	movl	%ebx,%esi
+	shrdl	$9,%ecx,%ecx
+	movl	%ebx,4(%esp)
+	xorl	%ebx,%ecx
+	xorl	%edi,%ebx
+	addl	(%esp),%edx
+	shrdl	$11,%ecx,%ecx
+	andl	%ebx,%eax
+	xorl	%esi,%ecx
+	addl	60(%esp),%edx
+	xorl	%edi,%eax
+	shrdl	$2,%ecx,%ecx
+	addl	%edx,%eax
+	addl	16(%esp),%edx
+	addl	%ecx,%eax
+	movl	%edx,%ecx
+	shrdl	$14,%edx,%edx
+	movl	20(%esp),%esi
+	xorl	%ecx,%edx
+	movl	24(%esp),%edi
+	xorl	%edi,%esi
+	shrdl	$5,%edx,%edx
+	andl	%ecx,%esi
+	movl	%ecx,16(%esp)
+	xorl	%ecx,%edx
+	xorl	%esi,%edi
+	shrdl	$6,%edx,%edx
+	movl	%eax,%ecx
+	addl	%edi,%edx
+	movl	4(%esp),%edi
+	movl	%eax,%esi
+	shrdl	$9,%ecx,%ecx
+	movl	%eax,(%esp)
+	xorl	%eax,%ecx
+	xorl	%edi,%eax
+	addl	28(%esp),%edx
+	shrdl	$11,%ecx,%ecx
+	andl	%eax,%ebx
+	xorl	%esi,%ecx
+	addl	64(%esp),%edx
+	xorl	%edi,%ebx
+	shrdl	$2,%ecx,%ecx
+	addl	%edx,%ebx
+	addl	12(%esp),%edx
+	addl	%ecx,%ebx
+	movl	%edx,%ecx
+	shrdl	$14,%edx,%edx
+	movl	16(%esp),%esi
+	xorl	%ecx,%edx
+	movl	20(%esp),%edi
+	xorl	%edi,%esi
+	shrdl	$5,%edx,%edx
+	andl	%ecx,%esi
+	movl	%ecx,12(%esp)
+	xorl	%ecx,%edx
+	xorl	%esi,%edi
+	shrdl	$6,%edx,%edx
+	movl	%ebx,%ecx
+	addl	%edi,%edx
+	movl	(%esp),%edi
+	movl	%ebx,%esi
+	shrdl	$9,%ecx,%ecx
+	movl	%ebx,28(%esp)
+	xorl	%ebx,%ecx
+	xorl	%edi,%ebx
+	addl	24(%esp),%edx
+	shrdl	$11,%ecx,%ecx
+	andl	%ebx,%eax
+	xorl	%esi,%ecx
+	addl	68(%esp),%edx
+	xorl	%edi,%eax
+	shrdl	$2,%ecx,%ecx
+	addl	%edx,%eax
+	addl	8(%esp),%edx
+	addl	%ecx,%eax
+	movl	%edx,%ecx
+	shrdl	$14,%edx,%edx
+	movl	12(%esp),%esi
+	xorl	%ecx,%edx
+	movl	16(%esp),%edi
+	xorl	%edi,%esi
+	shrdl	$5,%edx,%edx
+	andl	%ecx,%esi
+	movl	%ecx,8(%esp)
+	xorl	%ecx,%edx
+	xorl	%esi,%edi
+	shrdl	$6,%edx,%edx
+	movl	%eax,%ecx
+	addl	%edi,%edx
+	movl	28(%esp),%edi
+	movl	%eax,%esi
+	shrdl	$9,%ecx,%ecx
+	movl	%eax,24(%esp)
+	xorl	%eax,%ecx
+	xorl	%edi,%eax
+	addl	20(%esp),%edx
+	shrdl	$11,%ecx,%ecx
+	andl	%eax,%ebx
+	xorl	%esi,%ecx
+	addl	72(%esp),%edx
+	xorl	%edi,%ebx
+	shrdl	$2,%ecx,%ecx
+	addl	%edx,%ebx
+	addl	4(%esp),%edx
+	addl	%ecx,%ebx
+	movl	%edx,%ecx
+	shrdl	$14,%edx,%edx
+	movl	8(%esp),%esi
+	xorl	%ecx,%edx
+	movl	12(%esp),%edi
+	xorl	%edi,%esi
+	shrdl	$5,%edx,%edx
+	andl	%ecx,%esi
+	movl	%ecx,4(%esp)
+	xorl	%ecx,%edx
+	xorl	%esi,%edi
+	shrdl	$6,%edx,%edx
+	movl	%ebx,%ecx
+	addl	%edi,%edx
+	movl	24(%esp),%edi
+	movl	%ebx,%esi
+	shrdl	$9,%ecx,%ecx
+	movl	%ebx,20(%esp)
+	xorl	%ebx,%ecx
+	xorl	%edi,%ebx
+	addl	16(%esp),%edx
+	shrdl	$11,%ecx,%ecx
+	andl	%ebx,%eax
+	xorl	%esi,%ecx
+	addl	76(%esp),%edx
+	xorl	%edi,%eax
+	shrdl	$2,%ecx,%ecx
+	addl	%edx,%eax
+	addl	(%esp),%edx
+	addl	%ecx,%eax
+	movl	%edx,%ecx
+	shrdl	$14,%edx,%edx
+	movl	4(%esp),%esi
+	xorl	%ecx,%edx
+	movl	8(%esp),%edi
+	xorl	%edi,%esi
+	shrdl	$5,%edx,%edx
+	andl	%ecx,%esi
+	movl	%ecx,(%esp)
+	xorl	%ecx,%edx
+	xorl	%esi,%edi
+	shrdl	$6,%edx,%edx
+	movl	%eax,%ecx
+	addl	%edi,%edx
+	movl	20(%esp),%edi
+	movl	%eax,%esi
+	shrdl	$9,%ecx,%ecx
+	movl	%eax,16(%esp)
+	xorl	%eax,%ecx
+	xorl	%edi,%eax
+	addl	12(%esp),%edx
+	shrdl	$11,%ecx,%ecx
+	andl	%eax,%ebx
+	xorl	%esi,%ecx
+	addl	80(%esp),%edx
+	xorl	%edi,%ebx
+	shrdl	$2,%ecx,%ecx
+	addl	%edx,%ebx
+	addl	28(%esp),%edx
+	addl	%ecx,%ebx
+	movl	%edx,%ecx
+	shrdl	$14,%edx,%edx
+	movl	(%esp),%esi
+	xorl	%ecx,%edx
+	movl	4(%esp),%edi
+	xorl	%edi,%esi
+	shrdl	$5,%edx,%edx
+	andl	%ecx,%esi
+	movl	%ecx,28(%esp)
+	xorl	%ecx,%edx
+	xorl	%esi,%edi
+	shrdl	$6,%edx,%edx
+	movl	%ebx,%ecx
+	addl	%edi,%edx
+	movl	16(%esp),%edi
+	movl	%ebx,%esi
+	shrdl	$9,%ecx,%ecx
+	movl	%ebx,12(%esp)
+	xorl	%ebx,%ecx
+	xorl	%edi,%ebx
+	addl	8(%esp),%edx
+	shrdl	$11,%ecx,%ecx
+	andl	%ebx,%eax
+	xorl	%esi,%ecx
+	addl	84(%esp),%edx
+	xorl	%edi,%eax
+	shrdl	$2,%ecx,%ecx
+	addl	%edx,%eax
+	addl	24(%esp),%edx
+	addl	%ecx,%eax
+	movl	%edx,%ecx
+	shrdl	$14,%edx,%edx
+	movl	28(%esp),%esi
+	xorl	%ecx,%edx
+	movl	(%esp),%edi
+	xorl	%edi,%esi
+	shrdl	$5,%edx,%edx
+	andl	%ecx,%esi
+	movl	%ecx,24(%esp)
+	xorl	%ecx,%edx
+	xorl	%esi,%edi
+	shrdl	$6,%edx,%edx
+	movl	%eax,%ecx
+	addl	%edi,%edx
+	movl	12(%esp),%edi
+	movl	%eax,%esi
+	shrdl	$9,%ecx,%ecx
+	movl	%eax,8(%esp)
+	xorl	%eax,%ecx
+	xorl	%edi,%eax
+	addl	4(%esp),%edx
+	shrdl	$11,%ecx,%ecx
+	andl	%eax,%ebx
+	xorl	%esi,%ecx
+	addl	88(%esp),%edx
+	xorl	%edi,%ebx
+	shrdl	$2,%ecx,%ecx
+	addl	%edx,%ebx
+	addl	20(%esp),%edx
+	addl	%ecx,%ebx
+	movl	%edx,%ecx
+	shrdl	$14,%edx,%edx
+	movl	24(%esp),%esi
+	xorl	%ecx,%edx
+	movl	28(%esp),%edi
+	xorl	%edi,%esi
+	shrdl	$5,%edx,%edx
+	andl	%ecx,%esi
+	movl	%ecx,20(%esp)
+	xorl	%ecx,%edx
+	xorl	%esi,%edi
+	shrdl	$6,%edx,%edx
+	movl	%ebx,%ecx
+	addl	%edi,%edx
+	movl	8(%esp),%edi
+	movl	%ebx,%esi
+	shrdl	$9,%ecx,%ecx
+	movl	%ebx,4(%esp)
+	xorl	%ebx,%ecx
+	xorl	%edi,%ebx
+	addl	(%esp),%edx
+	shrdl	$11,%ecx,%ecx
+	andl	%ebx,%eax
+	xorl	%esi,%ecx
+	addl	92(%esp),%edx
+	xorl	%edi,%eax
+	shrdl	$2,%ecx,%ecx
+	addl	%edx,%eax
+	addl	16(%esp),%edx
+	addl	%ecx,%eax
+	movl	96(%esp),%esi
+	xorl	%edi,%ebx
+	movl	12(%esp),%ecx
+	addl	(%esi),%eax
+	addl	4(%esi),%ebx
+	addl	8(%esi),%edi
+	addl	12(%esi),%ecx
+	movl	%eax,(%esi)
+	movl	%ebx,4(%esi)
+	movl	%edi,8(%esi)
+	movl	%ecx,12(%esi)
+	movl	%ebx,4(%esp)
+	xorl	%edi,%ebx
+	movl	%edi,8(%esp)
+	movl	%ecx,12(%esp)
+	movl	20(%esp),%edi
+	movl	24(%esp),%ecx
+	addl	16(%esi),%edx
+	addl	20(%esi),%edi
+	addl	24(%esi),%ecx
+	movl	%edx,16(%esi)
+	movl	%edi,20(%esi)
+	movl	%edi,20(%esp)
+	movl	28(%esp),%edi
+	movl	%ecx,24(%esi)
+	addl	28(%esi),%edi
+	movl	%ecx,24(%esp)
+	movl	%edi,28(%esi)
+	movl	%edi,28(%esp)
+	movl	100(%esp),%edi
+	vmovdqa	64(%ebp),%xmm7
+	subl	$192,%ebp
+	cmpl	104(%esp),%edi
+	jb	.L012grand_avx
+	movl	108(%esp),%esp
+	vzeroall
+	popl	%edi
+	popl	%esi
+	popl	%ebx
+	popl	%ebp
+	ret
 .size	sha256_block_data_order,.-.L_sha256_block_data_order_begin
 #endif

third_party/boringssl/linux-x86_64/crypto/aes/aes-x86_64.S

diff --git a/third_party/boringssl/linux-x86_64/crypto/aes/aes-x86_64.S b/third_party/boringssl/linux-x86_64/crypto/aes/aes-x86_64.S
index 5f4b057..361e84c 100644
--- a/third_party/boringssl/linux-x86_64/crypto/aes/aes-x86_64.S
+++ b/third_party/boringssl/linux-x86_64/crypto/aes/aes-x86_64.S
@@ -82,8 +82,8 @@
 	movl	0(%r14,%rdi,8),%edi
 	movl	0(%r14,%rbp,8),%ebp
 
-	andl	$65280,%edi
-	andl	$65280,%ebp
+	andl	$0x0000ff00,%edi
+	andl	$0x0000ff00,%ebp
 
 	xorl	%edi,%r10d
 	xorl	%ebp,%r11d
@@ -95,8 +95,8 @@
 	movl	0(%r14,%rsi,8),%esi
 	movl	0(%r14,%rdi,8),%edi
 
-	andl	$65280,%esi
-	andl	$65280,%edi
+	andl	$0x0000ff00,%esi
+	andl	$0x0000ff00,%edi
 	shrl	$16,%ebx
 	xorl	%esi,%r12d
 	xorl	%edi,%r8d
@@ -109,9 +109,9 @@
 	movl	0(%r14,%rdi,8),%edi
 	movl	0(%r14,%rbp,8),%ebp
 
-	andl	$16711680,%esi
-	andl	$16711680,%edi
-	andl	$16711680,%ebp
+	andl	$0x00ff0000,%esi
+	andl	$0x00ff0000,%edi
+	andl	$0x00ff0000,%ebp
 
 	xorl	%esi,%r10d
 	xorl	%edi,%r11d
@@ -124,9 +124,9 @@
 	movl	2(%r14,%rdi,8),%edi
 	movl	2(%r14,%rbp,8),%ebp
 
-	andl	$16711680,%esi
-	andl	$4278190080,%edi
-	andl	$4278190080,%ebp
+	andl	$0x00ff0000,%esi
+	andl	$0xff000000,%edi
+	andl	$0xff000000,%ebp
 
 	xorl	%esi,%r8d
 	xorl	%edi,%r10d
@@ -139,8 +139,8 @@
 	movl	2(%r14,%rdi,8),%edi
 	movl	16+0(%r15),%eax
 
-	andl	$4278190080,%esi
-	andl	$4278190080,%edi
+	andl	$0xff000000,%esi
+	andl	$0xff000000,%edi
 
 	xorl	%esi,%r12d
 	xorl	%edi,%r8d
@@ -242,8 +242,8 @@
 	xorl	%r8d,%edx
 	cmpq	16(%rsp),%r15
 	je	.Lenc_compact_done
-	movl	$2155905152,%r10d
-	movl	$2155905152,%r11d
+	movl	$0x80808080,%r10d
+	movl	$0x80808080,%r11d
 	andl	%eax,%r10d
 	andl	%ebx,%r11d
 	movl	%r10d,%esi
@@ -254,10 +254,10 @@
 	leal	(%rbx,%rbx,1),%r9d
 	subl	%r10d,%esi
 	subl	%r11d,%edi
-	andl	$4278124286,%r8d
-	andl	$4278124286,%r9d
-	andl	$454761243,%esi
-	andl	$454761243,%edi
+	andl	$0xfefefefe,%r8d
+	andl	$0xfefefefe,%r9d
+	andl	$0x1b1b1b1b,%esi
+	andl	$0x1b1b1b1b,%edi
 	movl	%eax,%r10d
 	movl	%ebx,%r11d
 	xorl	%esi,%r8d
@@ -265,9 +265,9 @@
 
 	xorl	%r8d,%eax
 	xorl	%r9d,%ebx
-	movl	$2155905152,%r12d
+	movl	$0x80808080,%r12d
 	roll	$24,%eax
-	movl	$2155905152,%ebp
+	movl	$0x80808080,%ebp
 	roll	$24,%ebx
 	andl	%ecx,%r12d
 	andl	%edx,%ebp
@@ -290,10 +290,10 @@
 	xorl	%r10d,%eax
 	xorl	%r11d,%ebx
 
-	andl	$4278124286,%r8d
-	andl	$4278124286,%r9d
-	andl	$454761243,%esi
-	andl	$454761243,%edi
+	andl	$0xfefefefe,%r8d
+	andl	$0xfefefefe,%r9d
+	andl	$0x1b1b1b1b,%esi
+	andl	$0x1b1b1b1b,%edi
 	movl	%ecx,%r12d
 	movl	%edx,%ebp
 	xorl	%esi,%r8d
@@ -345,7 +345,7 @@
 	andq	$-64,%rsp
 	subq	%rsp,%rcx
 	negq	%rcx
-	andq	$960,%rcx
+	andq	$0x3c0,%rcx
 	subq	%rcx,%rsp
 	subq	$32,%rsp
 
@@ -370,7 +370,7 @@
 	leaq	.LAES_Te+2048(%rip),%r14
 	leaq	768(%rsp),%rbp
 	subq	%r14,%rbp
-	andq	$768,%rbp
+	andq	$0x300,%rbp
 	leaq	(%r14,%rbp,1),%r14
 
 	call	_x86_64_AES_encrypt_compact
@@ -791,7 +791,7 @@
 	andq	$-64,%rsp
 	subq	%rsp,%rcx
 	negq	%rcx
-	andq	$960,%rcx
+	andq	$0x3c0,%rcx
 	subq	%rcx,%rsp
 	subq	$32,%rsp
 
@@ -816,7 +816,7 @@
 	leaq	.LAES_Td+2048(%rip),%r14
 	leaq	768(%rsp),%rbp
 	subq	%r14,%rbp
-	andq	$768,%rbp
+	andq	$0x300,%rbp
 	leaq	(%r14,%rbp,1),%r14
 	shrq	$3,%rbp
 	addq	%rbp,%r14
@@ -1334,9 +1334,9 @@
 	movq	%r14,%r10
 	leaq	2304(%r14),%r11
 	movq	%r15,%r12
-	andq	$4095,%r10
-	andq	$4095,%r11
-	andq	$4095,%r12
+	andq	$0xFFF,%r10
+	andq	$0xFFF,%r11
+	andq	$0xFFF,%r12
 
 	cmpq	%r11,%r12
 	jb	.Lcbc_te_break_out
@@ -1345,7 +1345,7 @@
 	jmp	.Lcbc_te_ok
 .Lcbc_te_break_out:
 	subq	%r10,%r12
-	andq	$4095,%r12
+	andq	$0xFFF,%r12
 	addq	$320,%r12
 	subq	%r12,%r15
 .align	4
@@ -1371,7 +1371,7 @@
 
 	movq	%r15,%r10
 	subq	%r14,%r10
-	andq	$4095,%r10
+	andq	$0xfff,%r10
 	cmpq	$2304,%r10
 	jb	.Lcbc_do_ecopy
 	cmpq	$4096-248,%r10
@@ -1558,7 +1558,7 @@
 	leaq	-88-63(%rcx),%r10
 	subq	%rbp,%r10
 	negq	%r10
-	andq	$960,%r10
+	andq	$0x3c0,%r10
 	subq	%r10,%rbp
 
 	xchgq	%rsp,%rbp
@@ -1587,7 +1587,7 @@
 	leaq	2048(%r14),%r14
 	leaq	768-8(%rsp),%rax
 	subq	%r14,%rax
-	andq	$768,%rax
+	andq	$0x300,%rax
 	leaq	(%r14,%rax,1),%r14
 
 	cmpq	$0,%rbx

third_party/boringssl/linux-x86_64/crypto/aes/aesni-x86_64.S

diff --git a/third_party/boringssl/linux-x86_64/crypto/aes/aesni-x86_64.S b/third_party/boringssl/linux-x86_64/crypto/aes/aesni-x86_64.S
index 1d51d5b..5709a2d 100644
--- a/third_party/boringssl/linux-x86_64/crypto/aes/aesni-x86_64.S
+++ b/third_party/boringssl/linux-x86_64/crypto/aes/aesni-x86_64.S
@@ -508,7 +508,7 @@
 	testl	%r8d,%r8d
 	jz	.Lecb_decrypt
 
-	cmpq	$128,%rdx
+	cmpq	$0x80,%rdx
 	jb	.Lecb_enc_tail
 
 	movdqu	(%rdi),%xmm2
@@ -520,7 +520,7 @@
 	movdqu	96(%rdi),%xmm8
 	movdqu	112(%rdi),%xmm9
 	leaq	128(%rdi),%rdi
-	subq	$128,%rdx
+	subq	$0x80,%rdx
 	jmp	.Lecb_enc_loop8_enter
 .align	16
 .Lecb_enc_loop8:
@@ -548,7 +548,7 @@
 
 	call	_aesni_encrypt8
 
-	subq	$128,%rdx
+	subq	$0x80,%rdx
 	jnc	.Lecb_enc_loop8
 
 	movups	%xmm2,(%rsi)
@@ -562,22 +562,22 @@
 	movups	%xmm8,96(%rsi)
 	movups	%xmm9,112(%rsi)
 	leaq	128(%rsi),%rsi
-	addq	$128,%rdx
+	addq	$0x80,%rdx
 	jz	.Lecb_ret
 
 .Lecb_enc_tail:
 	movups	(%rdi),%xmm2
-	cmpq	$32,%rdx
+	cmpq	$0x20,%rdx
 	jb	.Lecb_enc_one
 	movups	16(%rdi),%xmm3
 	je	.Lecb_enc_two
 	movups	32(%rdi),%xmm4
-	cmpq	$64,%rdx
+	cmpq	$0x40,%rdx
 	jb	.Lecb_enc_three
 	movups	48(%rdi),%xmm5
 	je	.Lecb_enc_four
 	movups	64(%rdi),%xmm6
-	cmpq	$96,%rdx
+	cmpq	$0x60,%rdx
 	jb	.Lecb_enc_five
 	movups	80(%rdi),%xmm7
 	je	.Lecb_enc_six
@@ -651,7 +651,7 @@
 
 .align	16
 .Lecb_decrypt:
-	cmpq	$128,%rdx
+	cmpq	$0x80,%rdx
 	jb	.Lecb_dec_tail
 
 	movdqu	(%rdi),%xmm2
@@ -663,7 +663,7 @@
 	movdqu	96(%rdi),%xmm8
 	movdqu	112(%rdi),%xmm9
 	leaq	128(%rdi),%rdi
-	subq	$128,%rdx
+	subq	$0x80,%rdx
 	jmp	.Lecb_dec_loop8_enter
 .align	16
 .Lecb_dec_loop8:
@@ -692,7 +692,7 @@
 	call	_aesni_decrypt8
 
 	movups	(%r11),%xmm0
-	subq	$128,%rdx
+	subq	$0x80,%rdx
 	jnc	.Lecb_dec_loop8
 
 	movups	%xmm2,(%rsi)
@@ -714,22 +714,22 @@
 	movups	%xmm9,112(%rsi)
 	pxor	%xmm9,%xmm9
 	leaq	128(%rsi),%rsi
-	addq	$128,%rdx
+	addq	$0x80,%rdx
 	jz	.Lecb_ret
 
 .Lecb_dec_tail:
 	movups	(%rdi),%xmm2
-	cmpq	$32,%rdx
+	cmpq	$0x20,%rdx
 	jb	.Lecb_dec_one
 	movups	16(%rdi),%xmm3
 	je	.Lecb_dec_two
 	movups	32(%rdi),%xmm4
-	cmpq	$64,%rdx
+	cmpq	$0x40,%rdx
 	jb	.Lecb_dec_three
 	movups	48(%rdi),%xmm5
 	je	.Lecb_dec_four
 	movups	64(%rdi),%xmm6
-	cmpq	$96,%rdx
+	cmpq	$0x60,%rdx
 	jb	.Lecb_dec_five
 	movups	80(%rdi),%xmm7
 	je	.Lecb_dec_six
@@ -1607,7 +1607,7 @@
 
 	movdqa	.Lxts_magic(%rip),%xmm8
 	movdqa	%xmm2,%xmm15
-	pshufd	$95,%xmm2,%xmm9
+	pshufd	$0x5f,%xmm2,%xmm9
 	pxor	%xmm0,%xmm1
 	movdqa	%xmm9,%xmm14
 	paddd	%xmm9,%xmm9
@@ -1706,7 +1706,7 @@
 .byte	102,15,56,220,248
 	movups	64(%r11),%xmm0
 	movdqa	%xmm8,80(%rsp)
-	pshufd	$95,%xmm15,%xmm9
+	pshufd	$0x5f,%xmm15,%xmm9
 	jmp	.Lxts_enc_loop6
 .align	32
 .Lxts_enc_loop6:
@@ -1845,13 +1845,13 @@
 	jz	.Lxts_enc_done
 
 	pxor	%xmm0,%xmm11
-	cmpq	$32,%rdx
+	cmpq	$0x20,%rdx
 	jb	.Lxts_enc_one
 	pxor	%xmm0,%xmm12
 	je	.Lxts_enc_two
 
 	pxor	%xmm0,%xmm13
-	cmpq	$64,%rdx
+	cmpq	$0x40,%rdx
 	jb	.Lxts_enc_three
 	pxor	%xmm0,%xmm14
 	je	.Lxts_enc_four
@@ -2079,7 +2079,7 @@
 
 	movdqa	.Lxts_magic(%rip),%xmm8
 	movdqa	%xmm2,%xmm15
-	pshufd	$95,%xmm2,%xmm9
+	pshufd	$0x5f,%xmm2,%xmm9
 	pxor	%xmm0,%xmm1
 	movdqa	%xmm9,%xmm14
 	paddd	%xmm9,%xmm9
@@ -2178,7 +2178,7 @@
 .byte	102,15,56,222,248
 	movups	64(%r11),%xmm0
 	movdqa	%xmm8,80(%rsp)
-	pshufd	$95,%xmm15,%xmm9
+	pshufd	$0x5f,%xmm15,%xmm9
 	jmp	.Lxts_dec_loop6
 .align	32
 .Lxts_dec_loop6:
@@ -2318,13 +2318,13 @@
 	jz	.Lxts_dec_done
 
 	pxor	%xmm0,%xmm12
-	cmpq	$32,%rdx
+	cmpq	$0x20,%rdx
 	jb	.Lxts_dec_one
 	pxor	%xmm0,%xmm13
 	je	.Lxts_dec_two
 
 	pxor	%xmm0,%xmm14
-	cmpq	$64,%rdx
+	cmpq	$0x40,%rdx
 	jb	.Lxts_dec_three
 	je	.Lxts_dec_four
 
@@ -2355,7 +2355,7 @@
 	pcmpgtd	%xmm15,%xmm14
 	movdqu	%xmm6,64(%rsi)
 	leaq	80(%rsi),%rsi
-	pshufd	$19,%xmm14,%xmm11
+	pshufd	$0x13,%xmm14,%xmm11
 	andq	$15,%r9
 	jz	.Lxts_dec_ret
 
@@ -2645,7 +2645,7 @@
 	leaq	-8(%rax),%rbp
 	movups	(%r8),%xmm10
 	movl	%r10d,%eax
-	cmpq	$80,%rdx
+	cmpq	$0x50,%rdx
 	jbe	.Lcbc_dec_tail
 
 	movups	(%rcx),%xmm0
@@ -2661,14 +2661,14 @@
 	movdqu	80(%rdi),%xmm7
 	movdqa	%xmm6,%xmm15
 	movl	OPENSSL_ia32cap_P+4(%rip),%r9d
-	cmpq	$112,%rdx
+	cmpq	$0x70,%rdx
 	jbe	.Lcbc_dec_six_or_seven
 
 	andl	$71303168,%r9d
-	subq	$80,%rdx
+	subq	$0x50,%rdx
 	cmpl	$4194304,%r9d
 	je	.Lcbc_dec_loop6_enter
-	subq	$32,%rdx
+	subq	$0x20,%rdx
 	leaq	112(%rcx),%rcx
 	jmp	.Lcbc_dec_loop8_enter
 .align	16
@@ -2683,7 +2683,7 @@
 	movups	16-112(%rcx),%xmm1
 	pxor	%xmm0,%xmm4
 	xorq	%r11,%r11
-	cmpq	$112,%rdx
+	cmpq	$0x70,%rdx
 	pxor	%xmm0,%xmm5
 	pxor	%xmm0,%xmm6
 	pxor	%xmm0,%xmm7
@@ -2868,21 +2868,21 @@
 	movups	%xmm8,96(%rsi)
 	leaq	112(%rsi),%rsi
 
-	subq	$128,%rdx
+	subq	$0x80,%rdx
 	ja	.Lcbc_dec_loop8
 
 	movaps	%xmm9,%xmm2
 	leaq	-112(%rcx),%rcx
-	addq	$112,%rdx
+	addq	$0x70,%rdx
 	jle	.Lcbc_dec_clear_tail_collected
 	movups	%xmm9,(%rsi)
 	leaq	16(%rsi),%rsi
-	cmpq	$80,%rdx
+	cmpq	$0x50,%rdx
 	jbe	.Lcbc_dec_tail
 
 	movaps	%xmm11,%xmm2
 .Lcbc_dec_six_or_seven:
-	cmpq	$96,%rdx
+	cmpq	$0x60,%rdx
 	ja	.Lcbc_dec_seven
 
 	movaps	%xmm7,%xmm8
@@ -2975,33 +2975,33 @@
 	movl	%r10d,%eax
 	movdqu	%xmm6,64(%rsi)
 	leaq	80(%rsi),%rsi
-	subq	$96,%rdx
+	subq	$0x60,%rdx
 	ja	.Lcbc_dec_loop6
 
 	movdqa	%xmm7,%xmm2
-	addq	$80,%rdx
+	addq	$0x50,%rdx
 	jle	.Lcbc_dec_clear_tail_collected
 	movups	%xmm7,(%rsi)
 	leaq	16(%rsi),%rsi
 
 .Lcbc_dec_tail:
 	movups	(%rdi),%xmm2
-	subq	$16,%rdx
+	subq	$0x10,%rdx
 	jbe	.Lcbc_dec_one
 
 	movups	16(%rdi),%xmm3
 	movaps	%xmm2,%xmm11
-	subq	$16,%rdx
+	subq	$0x10,%rdx
 	jbe	.Lcbc_dec_two
 
 	movups	32(%rdi),%xmm4
 	movaps	%xmm3,%xmm12
-	subq	$16,%rdx
+	subq	$0x10,%rdx
 	jbe	.Lcbc_dec_three
 
 	movups	48(%rdi),%xmm5
 	movaps	%xmm4,%xmm13
-	subq	$16,%rdx
+	subq	$0x10,%rdx
 	jbe	.Lcbc_dec_four
 
 	movups	64(%rdi),%xmm6
@@ -3026,7 +3026,7 @@
 	movdqa	%xmm6,%xmm2
 	pxor	%xmm6,%xmm6
 	pxor	%xmm7,%xmm7
-	subq	$16,%rdx
+	subq	$0x10,%rdx
 	jmp	.Lcbc_dec_tail_collected
 
 .align	16
@@ -3345,7 +3345,7 @@
 	pslldq	$4,%xmm0
 	pxor	%xmm3,%xmm0
 
-	pshufd	$255,%xmm0,%xmm3
+	pshufd	$0xff,%xmm0,%xmm3
 	pxor	%xmm1,%xmm3
 	pslldq	$4,%xmm1
 	pxor	%xmm1,%xmm3
@@ -3432,7 +3432,7 @@
 	decl	%r10d
 	jz	.Ldone_key256
 
-	pshufd	$255,%xmm0,%xmm2
+	pshufd	$0xff,%xmm0,%xmm2
 	pxor	%xmm3,%xmm3
 .byte	102,15,56,221,211
 

third_party/boringssl/linux-x86_64/crypto/aes/bsaes-x86_64.S

diff --git a/third_party/boringssl/linux-x86_64/crypto/aes/bsaes-x86_64.S b/third_party/boringssl/linux-x86_64/crypto/aes/bsaes-x86_64.S
index 8cfa4df..c5491ce4 100644
--- a/third_party/boringssl/linux-x86_64/crypto/aes/bsaes-x86_64.S
+++ b/third_party/boringssl/linux-x86_64/crypto/aes/bsaes-x86_64.S
@@ -327,45 +327,45 @@
 	pxor	%xmm2,%xmm5
 	decl	%r10d
 	jl	.Lenc_done
-	pshufd	$147,%xmm15,%xmm7
-	pshufd	$147,%xmm0,%xmm8
+	pshufd	$0x93,%xmm15,%xmm7
+	pshufd	$0x93,%xmm0,%xmm8
 	pxor	%xmm7,%xmm15
-	pshufd	$147,%xmm3,%xmm9
+	pshufd	$0x93,%xmm3,%xmm9
 	pxor	%xmm8,%xmm0
-	pshufd	$147,%xmm5,%xmm10
+	pshufd	$0x93,%xmm5,%xmm10
 	pxor	%xmm9,%xmm3
-	pshufd	$147,%xmm2,%xmm11
+	pshufd	$0x93,%xmm2,%xmm11
 	pxor	%xmm10,%xmm5
-	pshufd	$147,%xmm6,%xmm12
+	pshufd	$0x93,%xmm6,%xmm12
 	pxor	%xmm11,%xmm2
-	pshufd	$147,%xmm1,%xmm13
+	pshufd	$0x93,%xmm1,%xmm13
 	pxor	%xmm12,%xmm6
-	pshufd	$147,%xmm4,%xmm14
+	pshufd	$0x93,%xmm4,%xmm14
 	pxor	%xmm13,%xmm1
 	pxor	%xmm14,%xmm4
 
 	pxor	%xmm15,%xmm8
 	pxor	%xmm4,%xmm7
 	pxor	%xmm4,%xmm8
-	pshufd	$78,%xmm15,%xmm15
+	pshufd	$0x4E,%xmm15,%xmm15
 	pxor	%xmm0,%xmm9
-	pshufd	$78,%xmm0,%xmm0
+	pshufd	$0x4E,%xmm0,%xmm0
 	pxor	%xmm2,%xmm12
 	pxor	%xmm7,%xmm15
 	pxor	%xmm6,%xmm13
 	pxor	%xmm8,%xmm0
 	pxor	%xmm5,%xmm11
-	pshufd	$78,%xmm2,%xmm7
+	pshufd	$0x4E,%xmm2,%xmm7
 	pxor	%xmm1,%xmm14
-	pshufd	$78,%xmm6,%xmm8
+	pshufd	$0x4E,%xmm6,%xmm8
 	pxor	%xmm3,%xmm10
-	pshufd	$78,%xmm5,%xmm2
+	pshufd	$0x4E,%xmm5,%xmm2
 	pxor	%xmm4,%xmm10
-	pshufd	$78,%xmm4,%xmm6
+	pshufd	$0x4E,%xmm4,%xmm6
 	pxor	%xmm4,%xmm11
-	pshufd	$78,%xmm1,%xmm5
+	pshufd	$0x4E,%xmm1,%xmm5
 	pxor	%xmm11,%xmm7
-	pshufd	$78,%xmm3,%xmm1
+	pshufd	$0x4E,%xmm3,%xmm1
 	pxor	%xmm12,%xmm8
 	pxor	%xmm10,%xmm2
 	pxor	%xmm14,%xmm6
@@ -799,24 +799,24 @@
 	decl	%r10d
 	jl	.Ldec_done
 
-	pshufd	$78,%xmm15,%xmm7
-	pshufd	$78,%xmm2,%xmm13
+	pshufd	$0x4E,%xmm15,%xmm7
+	pshufd	$0x4E,%xmm2,%xmm13
 	pxor	%xmm15,%xmm7
-	pshufd	$78,%xmm4,%xmm14
+	pshufd	$0x4E,%xmm4,%xmm14
 	pxor	%xmm2,%xmm13
-	pshufd	$78,%xmm0,%xmm8
+	pshufd	$0x4E,%xmm0,%xmm8
 	pxor	%xmm4,%xmm14
-	pshufd	$78,%xmm5,%xmm9
+	pshufd	$0x4E,%xmm5,%xmm9
 	pxor	%xmm0,%xmm8
-	pshufd	$78,%xmm3,%xmm10
+	pshufd	$0x4E,%xmm3,%xmm10
 	pxor	%xmm5,%xmm9
 	pxor	%xmm13,%xmm15
 	pxor	%xmm13,%xmm0
-	pshufd	$78,%xmm1,%xmm11
+	pshufd	$0x4E,%xmm1,%xmm11
 	pxor	%xmm3,%xmm10
 	pxor	%xmm7,%xmm5
 	pxor	%xmm8,%xmm3
-	pshufd	$78,%xmm6,%xmm12
+	pshufd	$0x4E,%xmm6,%xmm12
 	pxor	%xmm1,%xmm11
 	pxor	%xmm14,%xmm0
 	pxor	%xmm9,%xmm1
@@ -830,45 +830,45 @@
 	pxor	%xmm14,%xmm1
 	pxor	%xmm14,%xmm6
 	pxor	%xmm12,%xmm4
-	pshufd	$147,%xmm15,%xmm7
-	pshufd	$147,%xmm0,%xmm8
+	pshufd	$0x93,%xmm15,%xmm7
+	pshufd	$0x93,%xmm0,%xmm8
 	pxor	%xmm7,%xmm15
-	pshufd	$147,%xmm5,%xmm9
+	pshufd	$0x93,%xmm5,%xmm9
 	pxor	%xmm8,%xmm0
-	pshufd	$147,%xmm3,%xmm10
+	pshufd	$0x93,%xmm3,%xmm10
 	pxor	%xmm9,%xmm5
-	pshufd	$147,%xmm1,%xmm11
+	pshufd	$0x93,%xmm1,%xmm11
 	pxor	%xmm10,%xmm3
-	pshufd	$147,%xmm6,%xmm12
+	pshufd	$0x93,%xmm6,%xmm12
 	pxor	%xmm11,%xmm1
-	pshufd	$147,%xmm2,%xmm13
+	pshufd	$0x93,%xmm2,%xmm13
 	pxor	%xmm12,%xmm6
-	pshufd	$147,%xmm4,%xmm14
+	pshufd	$0x93,%xmm4,%xmm14
 	pxor	%xmm13,%xmm2
 	pxor	%xmm14,%xmm4
 
 	pxor	%xmm15,%xmm8
 	pxor	%xmm4,%xmm7
 	pxor	%xmm4,%xmm8
-	pshufd	$78,%xmm15,%xmm15
+	pshufd	$0x4E,%xmm15,%xmm15
 	pxor	%xmm0,%xmm9
-	pshufd	$78,%xmm0,%xmm0
+	pshufd	$0x4E,%xmm0,%xmm0
 	pxor	%xmm1,%xmm12
 	pxor	%xmm7,%xmm15
 	pxor	%xmm6,%xmm13
 	pxor	%xmm8,%xmm0
 	pxor	%xmm3,%xmm11
-	pshufd	$78,%xmm1,%xmm7
+	pshufd	$0x4E,%xmm1,%xmm7
 	pxor	%xmm2,%xmm14
-	pshufd	$78,%xmm6,%xmm8
+	pshufd	$0x4E,%xmm6,%xmm8
 	pxor	%xmm5,%xmm10
-	pshufd	$78,%xmm3,%xmm1
+	pshufd	$0x4E,%xmm3,%xmm1
 	pxor	%xmm4,%xmm10
-	pshufd	$78,%xmm4,%xmm6
+	pshufd	$0x4E,%xmm4,%xmm6
 	pxor	%xmm4,%xmm11
-	pshufd	$78,%xmm2,%xmm3
+	pshufd	$0x4E,%xmm2,%xmm3
 	pxor	%xmm11,%xmm7
-	pshufd	$78,%xmm5,%xmm2
+	pshufd	$0x4E,%xmm5,%xmm2
 	pxor	%xmm12,%xmm8
 	pxor	%xmm1,%xmm10
 	pxor	%xmm14,%xmm6
@@ -1559,20 +1559,20 @@
 	movdqa	%xmm7,(%rax)
 
 	andq	$-16,%r14
-	subq	$128,%rsp
+	subq	$0x80,%rsp
 	movdqa	32(%rbp),%xmm6
 
 	pxor	%xmm14,%xmm14
 	movdqa	.Lxts_magic(%rip),%xmm12
 	pcmpgtd	%xmm6,%xmm14
 
-	subq	$128,%r14
+	subq	$0x80,%r14
 	jc	.Lxts_enc_short
 	jmp	.Lxts_enc_loop
 
 .align	16
 .Lxts_enc_loop:
-	pshufd	$19,%xmm14,%xmm13
+	pshufd	$0x13,%xmm14,%xmm13
 	pxor	%xmm14,%xmm14
 	movdqa	%xmm6,%xmm15
 	movdqa	%xmm6,0(%rsp)
@@ -1580,7 +1580,7 @@
 	pand	%xmm12,%xmm13
 	pcmpgtd	%xmm6,%xmm14
 	pxor	%xmm13,%xmm6
-	pshufd	$19,%xmm14,%xmm13
+	pshufd	$0x13,%xmm14,%xmm13
 	pxor	%xmm14,%xmm14
 	movdqa	%xmm6,%xmm0
 	movdqa	%xmm6,16(%rsp)
@@ -1589,7 +1589,7 @@
 	pcmpgtd	%xmm6,%xmm14
 	pxor	%xmm13,%xmm6
 	movdqu	0(%r12),%xmm7
-	pshufd	$19,%xmm14,%xmm13
+	pshufd	$0x13,%xmm14,%xmm13
 	pxor	%xmm14,%xmm14
 	movdqa	%xmm6,%xmm1
 	movdqa	%xmm6,32(%rsp)
@@ -1599,7 +1599,7 @@
 	pxor	%xmm13,%xmm6
 	movdqu	16(%r12),%xmm8
 	pxor	%xmm7,%xmm15
-	pshufd	$19,%xmm14,%xmm13
+	pshufd	$0x13,%xmm14,%xmm13
 	pxor	%xmm14,%xmm14
 	movdqa	%xmm6,%xmm2
 	movdqa	%xmm6,48(%rsp)
@@ -1609,7 +1609,7 @@
 	pxor	%xmm13,%xmm6
 	movdqu	32(%r12),%xmm9
 	pxor	%xmm8,%xmm0
-	pshufd	$19,%xmm14,%xmm13
+	pshufd	$0x13,%xmm14,%xmm13
 	pxor	%xmm14,%xmm14
 	movdqa	%xmm6,%xmm3
 	movdqa	%xmm6,64(%rsp)
@@ -1619,7 +1619,7 @@
 	pxor	%xmm13,%xmm6
 	movdqu	48(%r12),%xmm10
 	pxor	%xmm9,%xmm1
-	pshufd	$19,%xmm14,%xmm13
+	pshufd	$0x13,%xmm14,%xmm13
 	pxor	%xmm14,%xmm14
 	movdqa	%xmm6,%xmm4
 	movdqa	%xmm6,80(%rsp)
@@ -1629,7 +1629,7 @@
 	pxor	%xmm13,%xmm6
 	movdqu	64(%r12),%xmm11
 	pxor	%xmm10,%xmm2
-	pshufd	$19,%xmm14,%xmm13
+	pshufd	$0x13,%xmm14,%xmm13
 	pxor	%xmm14,%xmm14
 	movdqa	%xmm6,%xmm5
 	movdqa	%xmm6,96(%rsp)
@@ -1673,20 +1673,20 @@
 	pxor	%xmm14,%xmm14
 	movdqa	.Lxts_magic(%rip),%xmm12
 	pcmpgtd	%xmm6,%xmm14
-	pshufd	$19,%xmm14,%xmm13
+	pshufd	$0x13,%xmm14,%xmm13
 	pxor	%xmm14,%xmm14
 	paddq	%xmm6,%xmm6
 	pand	%xmm12,%xmm13
 	pcmpgtd	%xmm6,%xmm14
 	pxor	%xmm13,%xmm6
 
-	subq	$128,%r14
+	subq	$0x80,%r14
 	jnc	.Lxts_enc_loop
 
 .Lxts_enc_short:
-	addq	$128,%r14
+	addq	$0x80,%r14
 	jz	.Lxts_enc_done
-	pshufd	$19,%xmm14,%xmm13
+	pshufd	$0x13,%xmm14,%xmm13
 	pxor	%xmm14,%xmm14
 	movdqa	%xmm6,%xmm15
 	movdqa	%xmm6,0(%rsp)
@@ -1694,7 +1694,7 @@
 	pand	%xmm12,%xmm13
 	pcmpgtd	%xmm6,%xmm14
 	pxor	%xmm13,%xmm6
-	pshufd	$19,%xmm14,%xmm13
+	pshufd	$0x13,%xmm14,%xmm13
 	pxor	%xmm14,%xmm14
 	movdqa	%xmm6,%xmm0
 	movdqa	%xmm6,16(%rsp)
@@ -1705,7 +1705,7 @@
 	movdqu	0(%r12),%xmm7
 	cmpq	$16,%r14
 	je	.Lxts_enc_1
-	pshufd	$19,%xmm14,%xmm13
+	pshufd	$0x13,%xmm14,%xmm13
 	pxor	%xmm14,%xmm14
 	movdqa	%xmm6,%xmm1
 	movdqa	%xmm6,32(%rsp)
@@ -1717,7 +1717,7 @@
 	cmpq	$32,%r14
 	je	.Lxts_enc_2
 	pxor	%xmm7,%xmm15
-	pshufd	$19,%xmm14,%xmm13
+	pshufd	$0x13,%xmm14,%xmm13
 	pxor	%xmm14,%xmm14
 	movdqa	%xmm6,%xmm2
 	movdqa	%xmm6,48(%rsp)
@@ -1729,7 +1729,7 @@
 	cmpq	$48,%r14
 	je	.Lxts_enc_3
 	pxor	%xmm8,%xmm0
-	pshufd	$19,%xmm14,%xmm13
+	pshufd	$0x13,%xmm14,%xmm13
 	pxor	%xmm14,%xmm14
 	movdqa	%xmm6,%xmm3
 	movdqa	%xmm6,64(%rsp)
@@ -1741,7 +1741,7 @@
 	cmpq	$64,%r14
 	je	.Lxts_enc_4
 	pxor	%xmm9,%xmm1
-	pshufd	$19,%xmm14,%xmm13
+	pshufd	$0x13,%xmm14,%xmm13
 	pxor	%xmm14,%xmm14
 	movdqa	%xmm6,%xmm4
 	movdqa	%xmm6,80(%rsp)
@@ -1753,7 +1753,7 @@
 	cmpq	$80,%r14
 	je	.Lxts_enc_5
 	pxor	%xmm10,%xmm2
-	pshufd	$19,%xmm14,%xmm13
+	pshufd	$0x13,%xmm14,%xmm13
 	pxor	%xmm14,%xmm14
 	movdqa	%xmm6,%xmm5
 	movdqa	%xmm6,96(%rsp)
@@ -2019,20 +2019,20 @@
 	shlq	$4,%rax
 	subq	%rax,%r14
 
-	subq	$128,%rsp
+	subq	$0x80,%rsp
 	movdqa	32(%rbp),%xmm6
 
 	pxor	%xmm14,%xmm14
 	movdqa	.Lxts_magic(%rip),%xmm12
 	pcmpgtd	%xmm6,%xmm14
 
-	subq	$128,%r14
+	subq	$0x80,%r14
 	jc	.Lxts_dec_short
 	jmp	.Lxts_dec_loop
 
 .align	16
 .Lxts_dec_loop:
-	pshufd	$19,%xmm14,%xmm13
+	pshufd	$0x13,%xmm14,%xmm13
 	pxor	%xmm14,%xmm14
 	movdqa	%xmm6,%xmm15
 	movdqa	%xmm6,0(%rsp)
@@ -2040,7 +2040,7 @@
 	pand	%xmm12,%xmm13
 	pcmpgtd	%xmm6,%xmm14
 	pxor	%xmm13,%xmm6
-	pshufd	$19,%xmm14,%xmm13
+	pshufd	$0x13,%xmm14,%xmm13
 	pxor	%xmm14,%xmm14
 	movdqa	%xmm6,%xmm0
 	movdqa	%xmm6,16(%rsp)
@@ -2049,7 +2049,7 @@
 	pcmpgtd	%xmm6,%xmm14
 	pxor	%xmm13,%xmm6
 	movdqu	0(%r12),%xmm7
-	pshufd	$19,%xmm14,%xmm13
+	pshufd	$0x13,%xmm14,%xmm13
 	pxor	%xmm14,%xmm14
 	movdqa	%xmm6,%xmm1
 	movdqa	%xmm6,32(%rsp)
@@ -2059,7 +2059,7 @@
 	pxor	%xmm13,%xmm6
 	movdqu	16(%r12),%xmm8
 	pxor	%xmm7,%xmm15
-	pshufd	$19,%xmm14,%xmm13
+	pshufd	$0x13,%xmm14,%xmm13
 	pxor	%xmm14,%xmm14
 	movdqa	%xmm6,%xmm2
 	movdqa	%xmm6,48(%rsp)
@@ -2069,7 +2069,7 @@
 	pxor	%xmm13,%xmm6
 	movdqu	32(%r12),%xmm9
 	pxor	%xmm8,%xmm0
-	pshufd	$19,%xmm14,%xmm13
+	pshufd	$0x13,%xmm14,%xmm13
 	pxor	%xmm14,%xmm14
 	movdqa	%xmm6,%xmm3
 	movdqa	%xmm6,64(%rsp)
@@ -2079,7 +2079,7 @@
 	pxor	%xmm13,%xmm6
 	movdqu	48(%r12),%xmm10
 	pxor	%xmm9,%xmm1
-	pshufd	$19,%xmm14,%xmm13
+	pshufd	$0x13,%xmm14,%xmm13
 	pxor	%xmm14,%xmm14
 	movdqa	%xmm6,%xmm4
 	movdqa	%xmm6,80(%rsp)
@@ -2089,7 +2089,7 @@
 	pxor	%xmm13,%xmm6
 	movdqu	64(%r12),%xmm11
 	pxor	%xmm10,%xmm2
-	pshufd	$19,%xmm14,%xmm13
+	pshufd	$0x13,%xmm14,%xmm13
 	pxor	%xmm14,%xmm14
 	movdqa	%xmm6,%xmm5
 	movdqa	%xmm6,96(%rsp)
@@ -2133,20 +2133,20 @@
 	pxor	%xmm14,%xmm14
 	movdqa	.Lxts_magic(%rip),%xmm12
 	pcmpgtd	%xmm6,%xmm14
-	pshufd	$19,%xmm14,%xmm13
+	pshufd	$0x13,%xmm14,%xmm13
 	pxor	%xmm14,%xmm14
 	paddq	%xmm6,%xmm6
 	pand	%xmm12,%xmm13
 	pcmpgtd	%xmm6,%xmm14
 	pxor	%xmm13,%xmm6
 
-	subq	$128,%r14
+	subq	$0x80,%r14
 	jnc	.Lxts_dec_loop
 
 .Lxts_dec_short:
-	addq	$128,%r14
+	addq	$0x80,%r14
 	jz	.Lxts_dec_done
-	pshufd	$19,%xmm14,%xmm13
+	pshufd	$0x13,%xmm14,%xmm13
 	pxor	%xmm14,%xmm14
 	movdqa	%xmm6,%xmm15
 	movdqa	%xmm6,0(%rsp)
@@ -2154,7 +2154,7 @@
 	pand	%xmm12,%xmm13
 	pcmpgtd	%xmm6,%xmm14
 	pxor	%xmm13,%xmm6
-	pshufd	$19,%xmm14,%xmm13
+	pshufd	$0x13,%xmm14,%xmm13
 	pxor	%xmm14,%xmm14
 	movdqa	%xmm6,%xmm0
 	movdqa	%xmm6,16(%rsp)
@@ -2165,7 +2165,7 @@
 	movdqu	0(%r12),%xmm7
 	cmpq	$16,%r14
 	je	.Lxts_dec_1
-	pshufd	$19,%xmm14,%xmm13
+	pshufd	$0x13,%xmm14,%xmm13
 	pxor	%xmm14,%xmm14
 	movdqa	%xmm6,%xmm1
 	movdqa	%xmm6,32(%rsp)
@@ -2177,7 +2177,7 @@
 	cmpq	$32,%r14
 	je	.Lxts_dec_2
 	pxor	%xmm7,%xmm15
-	pshufd	$19,%xmm14,%xmm13
+	pshufd	$0x13,%xmm14,%xmm13
 	pxor	%xmm14,%xmm14
 	movdqa	%xmm6,%xmm2
 	movdqa	%xmm6,48(%rsp)
@@ -2189,7 +2189,7 @@
 	cmpq	$48,%r14
 	je	.Lxts_dec_3
 	pxor	%xmm8,%xmm0
-	pshufd	$19,%xmm14,%xmm13
+	pshufd	$0x13,%xmm14,%xmm13
 	pxor	%xmm14,%xmm14
 	movdqa	%xmm6,%xmm3
 	movdqa	%xmm6,64(%rsp)
@@ -2201,7 +2201,7 @@
 	cmpq	$64,%r14
 	je	.Lxts_dec_4
 	pxor	%xmm9,%xmm1
-	pshufd	$19,%xmm14,%xmm13
+	pshufd	$0x13,%xmm14,%xmm13
 	pxor	%xmm14,%xmm14
 	movdqa	%xmm6,%xmm4
 	movdqa	%xmm6,80(%rsp)
@@ -2213,7 +2213,7 @@
 	cmpq	$80,%r14
 	je	.Lxts_dec_5
 	pxor	%xmm10,%xmm2
-	pshufd	$19,%xmm14,%xmm13
+	pshufd	$0x13,%xmm14,%xmm13
 	pxor	%xmm14,%xmm14
 	movdqa	%xmm6,%xmm5
 	movdqa	%xmm6,96(%rsp)
@@ -2390,7 +2390,7 @@
 	pxor	%xmm14,%xmm14
 	movdqa	.Lxts_magic(%rip),%xmm12
 	pcmpgtd	%xmm6,%xmm14
-	pshufd	$19,%xmm14,%xmm13
+	pshufd	$0x13,%xmm14,%xmm13
 	movdqa	%xmm6,%xmm5
 	paddq	%xmm6,%xmm6
 	pand	%xmm12,%xmm13

third_party/boringssl/linux-x86_64/crypto/aes/vpaes-x86_64.S

diff --git a/third_party/boringssl/linux-x86_64/crypto/aes/vpaes-x86_64.S b/third_party/boringssl/linux-x86_64/crypto/aes/vpaes-x86_64.S
index 1d12424..4dfafa9 100644
--- a/third_party/boringssl/linux-x86_64/crypto/aes/vpaes-x86_64.S
+++ b/third_party/boringssl/linux-x86_64/crypto/aes/vpaes-x86_64.S
@@ -61,7 +61,7 @@
 	addq	$16,%r11
 	pxor	%xmm0,%xmm3
 .byte	102,15,56,0,193
-	andq	$48,%r11
+	andq	$0x30,%r11
 	subq	$1,%rax
 	pxor	%xmm3,%xmm0
 
@@ -121,10 +121,10 @@
 	pand	%xmm9,%xmm0
 .byte	102,15,56,0,208
 	movdqa	.Lk_dipt+16(%rip),%xmm0
-	xorq	$48,%r11
+	xorq	$0x30,%r11
 	leaq	.Lk_dsbd(%rip),%r10
 .byte	102,15,56,0,193
-	andq	$48,%r11
+	andq	$0x30,%r11
 	pxor	%xmm5,%xmm2
 	movdqa	.Lk_mc_forward+48(%rip),%xmm5
 	pxor	%xmm2,%xmm0
@@ -243,7 +243,7 @@
 	movdqa	(%r8,%r10,1),%xmm1
 .byte	102,15,56,0,217
 	movdqu	%xmm3,(%rdx)
-	xorq	$48,%r8
+	xorq	$0x30,%r8
 
 .Lschedule_go:
 	cmpl	$192,%esi
@@ -333,7 +333,7 @@
 	call	_vpaes_schedule_mangle
 
 
-	pshufd	$255,%xmm0,%xmm0
+	pshufd	$0xFF,%xmm0,%xmm0
 	movdqa	%xmm7,%xmm5
 	movdqa	%xmm6,%xmm7
 	call	_vpaes_schedule_low_round
@@ -400,8 +400,8 @@
 .type	_vpaes_schedule_192_smear,@function
 .align	16
 _vpaes_schedule_192_smear:
-	pshufd	$128,%xmm6,%xmm1
-	pshufd	$254,%xmm7,%xmm0
+	pshufd	$0x80,%xmm6,%xmm1
+	pshufd	$0xFE,%xmm7,%xmm0
 	pxor	%xmm1,%xmm6
 	pxor	%xmm1,%xmm1
 	pxor	%xmm0,%xmm6
@@ -438,7 +438,7 @@
 	pxor	%xmm1,%xmm7
 
 
-	pshufd	$255,%xmm0,%xmm0
+	pshufd	$0xFF,%xmm0,%xmm0
 .byte	102,15,58,15,192,1
 
 
@@ -597,7 +597,7 @@
 	movdqa	(%r8,%r10,1),%xmm1
 .byte	102,15,56,0,217
 	addq	$-16,%r8
-	andq	$48,%r8
+	andq	$0x30,%r8
 	movdqu	%xmm3,(%rdx)
 	.byte	0xf3,0xc3
 .size	_vpaes_schedule_mangle,.-_vpaes_schedule_mangle
@@ -616,7 +616,7 @@
 	movl	%eax,240(%rdx)
 
 	movl	$0,%ecx
-	movl	$48,%r8d
+	movl	$0x30,%r8d
 	call	_vpaes_schedule_core
 	xorl	%eax,%eax
 	.byte	0xf3,0xc3

third_party/boringssl/linux-x86_64/crypto/bn/rsaz-x86_64.S

diff --git a/third_party/boringssl/linux-x86_64/crypto/bn/rsaz-x86_64.S b/third_party/boringssl/linux-x86_64/crypto/bn/rsaz-x86_64.S
index dd3d310..21531d1 100644
--- a/third_party/boringssl/linux-x86_64/crypto/bn/rsaz-x86_64.S
+++ b/third_party/boringssl/linux-x86_64/crypto/bn/rsaz-x86_64.S
@@ -466,48 +466,94 @@
 	pushq	%r14
 	pushq	%r15
 
-	movl	%r9d,%r9d
-	subq	$128+24,%rsp
+	subq	$152,%rsp
 .Lmul_gather4_body:
-	movl	64(%rdx,%r9,4),%eax
-.byte	102,72,15,110,199
-	movl	(%rdx,%r9,4),%ebx
-.byte	102,72,15,110,201
-	movq	%r8,128(%rsp)
+	movd	%r9d,%xmm8
+	movdqa	.Linc+16(%rip),%xmm1
+	movdqa	.Linc(%rip),%xmm0
 
-	shlq	$32,%rax
-	orq	%rax,%rbx
+	pshufd	$0,%xmm8,%xmm8
+	movdqa	%xmm1,%xmm7
+	movdqa	%xmm1,%xmm2
+	paddd	%xmm0,%xmm1
+	pcmpeqd	%xmm8,%xmm0
+	movdqa	%xmm7,%xmm3
+	paddd	%xmm1,%xmm2
+	pcmpeqd	%xmm8,%xmm1
+	movdqa	%xmm7,%xmm4
+	paddd	%xmm2,%xmm3
+	pcmpeqd	%xmm8,%xmm2
+	movdqa	%xmm7,%xmm5
+	paddd	%xmm3,%xmm4
+	pcmpeqd	%xmm8,%xmm3
+	movdqa	%xmm7,%xmm6
+	paddd	%xmm4,%xmm5
+	pcmpeqd	%xmm8,%xmm4
+	paddd	%xmm5,%xmm6
+	pcmpeqd	%xmm8,%xmm5
+	paddd	%xmm6,%xmm7
+	pcmpeqd	%xmm8,%xmm6
+	pcmpeqd	%xmm8,%xmm7
+
+	movdqa	0(%rdx),%xmm8
+	movdqa	16(%rdx),%xmm9
+	movdqa	32(%rdx),%xmm10
+	movdqa	48(%rdx),%xmm11
+	pand	%xmm0,%xmm8
+	movdqa	64(%rdx),%xmm12
+	pand	%xmm1,%xmm9
+	movdqa	80(%rdx),%xmm13
+	pand	%xmm2,%xmm10
+	movdqa	96(%rdx),%xmm14
+	pand	%xmm3,%xmm11
+	movdqa	112(%rdx),%xmm15
+	leaq	128(%rdx),%rbp
+	pand	%xmm4,%xmm12
+	pand	%xmm5,%xmm13
+	pand	%xmm6,%xmm14
+	pand	%xmm7,%xmm15
+	por	%xmm10,%xmm8
+	por	%xmm11,%xmm9
+	por	%xmm12,%xmm8
+	por	%xmm13,%xmm9
+	por	%xmm14,%xmm8
+	por	%xmm15,%xmm9
+
+	por	%xmm9,%xmm8
+	pshufd	$0x4e,%xmm8,%xmm9
+	por	%xmm9,%xmm8
+.byte	102,76,15,126,195
+
+	movq	%r8,128(%rsp)
+	movq	%rdi,128+8(%rsp)
+	movq	%rcx,128+16(%rsp)
+
 	movq	(%rsi),%rax
 	movq	8(%rsi),%rcx
-	leaq	128(%rdx,%r9,4),%rbp
 	mulq	%rbx
 	movq	%rax,(%rsp)
 	movq	%rcx,%rax
 	movq	%rdx,%r8
 
 	mulq	%rbx
-	movd	(%rbp),%xmm4
 	addq	%rax,%r8
 	movq	16(%rsi),%rax
 	movq	%rdx,%r9
 	adcq	$0,%r9
 
 	mulq	%rbx
-	movd	64(%rbp),%xmm5
 	addq	%rax,%r9
 	movq	24(%rsi),%rax
 	movq	%rdx,%r10
 	adcq	$0,%r10
 
 	mulq	%rbx
-	pslldq	$4,%xmm5
 	addq	%rax,%r10
 	movq	32(%rsi),%rax
 	movq	%rdx,%r11
 	adcq	$0,%r11
 
 	mulq	%rbx
-	por	%xmm5,%xmm4
 	addq	%rax,%r11
 	movq	40(%rsi),%rax
 	movq	%rdx,%r12
@@ -520,14 +566,12 @@
 	adcq	$0,%r13
 
 	mulq	%rbx
-	leaq	128(%rbp),%rbp
 	addq	%rax,%r13
 	movq	56(%rsi),%rax
 	movq	%rdx,%r14
 	adcq	$0,%r14
 
 	mulq	%rbx
-.byte	102,72,15,126,227
 	addq	%rax,%r14
 	movq	(%rsi),%rax
 	movq	%rdx,%r15
@@ -539,6 +583,35 @@
 
 .align	32
 .Loop_mul_gather:
+	movdqa	0(%rbp),%xmm8
+	movdqa	16(%rbp),%xmm9
+	movdqa	32(%rbp),%xmm10
+	movdqa	48(%rbp),%xmm11
+	pand	%xmm0,%xmm8
+	movdqa	64(%rbp),%xmm12
+	pand	%xmm1,%xmm9
+	movdqa	80(%rbp),%xmm13
+	pand	%xmm2,%xmm10
+	movdqa	96(%rbp),%xmm14
+	pand	%xmm3,%xmm11
+	movdqa	112(%rbp),%xmm15
+	leaq	128(%rbp),%rbp
+	pand	%xmm4,%xmm12
+	pand	%xmm5,%xmm13
+	pand	%xmm6,%xmm14
+	pand	%xmm7,%xmm15
+	por	%xmm10,%xmm8
+	por	%xmm11,%xmm9
+	por	%xmm12,%xmm8
+	por	%xmm13,%xmm9
+	por	%xmm14,%xmm8
+	por	%xmm15,%xmm9
+
+	por	%xmm9,%xmm8
+	pshufd	$0x4e,%xmm8,%xmm9
+	por	%xmm9,%xmm8
+.byte	102,76,15,126,195
+
 	mulq	%rbx
 	addq	%rax,%r8
 	movq	8(%rsi),%rax
@@ -547,7 +620,6 @@
 	adcq	$0,%r8
 
 	mulq	%rbx
-	movd	(%rbp),%xmm4
 	addq	%rax,%r9
 	movq	16(%rsi),%rax
 	adcq	$0,%rdx
@@ -556,7 +628,6 @@
 	adcq	$0,%r9
 
 	mulq	%rbx
-	movd	64(%rbp),%xmm5
 	addq	%rax,%r10
 	movq	24(%rsi),%rax
 	adcq	$0,%rdx
@@ -565,7 +636,6 @@
 	adcq	$0,%r10
 
 	mulq	%rbx
-	pslldq	$4,%xmm5
 	addq	%rax,%r11
 	movq	32(%rsi),%rax
 	adcq	$0,%rdx
@@ -574,7 +644,6 @@
 	adcq	$0,%r11
 
 	mulq	%rbx
-	por	%xmm5,%xmm4
 	addq	%rax,%r12
 	movq	40(%rsi),%rax
 	adcq	$0,%rdx
@@ -599,7 +668,6 @@
 	adcq	$0,%r14
 
 	mulq	%rbx
-.byte	102,72,15,126,227
 	addq	%rax,%r15
 	movq	(%rsi),%rax
 	adcq	$0,%rdx
@@ -607,7 +675,6 @@
 	movq	%rdx,%r15
 	adcq	$0,%r15
 
-	leaq	128(%rbp),%rbp
 	leaq	8(%rdi),%rdi
 
 	decl	%ecx
@@ -622,8 +689,8 @@
 	movq	%r14,48(%rdi)
 	movq	%r15,56(%rdi)
 
-.byte	102,72,15,126,199
-.byte	102,72,15,126,205
+	movq	128+8(%rsp),%rdi
+	movq	128+16(%rsp),%rbp
 
 	movq	(%rsp),%r8
 	movq	8(%rsp),%r9
@@ -673,7 +740,7 @@
 	movl	%r9d,%r9d
 	subq	$128+24,%rsp
 .Lmul_scatter4_body:
-	leaq	(%r8,%r9,4),%r8
+	leaq	(%r8,%r9,8),%r8
 .byte	102,72,15,110,199
 .byte	102,72,15,110,202
 .byte	102,73,15,110,208
@@ -709,30 +776,14 @@
 
 	call	__rsaz_512_subtract
 
-	movl	%r8d,0(%rsi)
-	shrq	$32,%r8
-	movl	%r9d,128(%rsi)
-	shrq	$32,%r9
-	movl	%r10d,256(%rsi)
-	shrq	$32,%r10
-	movl	%r11d,384(%rsi)
-	shrq	$32,%r11
-	movl	%r12d,512(%rsi)
-	shrq	$32,%r12
-	movl	%r13d,640(%rsi)
-	shrq	$32,%r13
-	movl	%r14d,768(%rsi)
-	shrq	$32,%r14
-	movl	%r15d,896(%rsi)
-	shrq	$32,%r15
-	movl	%r8d,64(%rsi)
-	movl	%r9d,192(%rsi)
-	movl	%r10d,320(%rsi)
-	movl	%r11d,448(%rsi)
-	movl	%r12d,576(%rsi)
-	movl	%r13d,704(%rsi)
-	movl	%r14d,832(%rsi)
-	movl	%r15d,960(%rsi)
+	movq	%r8,0(%rsi)
+	movq	%r9,128(%rsi)
+	movq	%r10,256(%rsi)
+	movq	%r11,384(%rsi)
+	movq	%r12,512(%rsi)
+	movq	%r13,640(%rsi)
+	movq	%r14,768(%rsi)
+	movq	%r15,896(%rsi)
 
 	leaq	128+24+48(%rsp),%rax
 	movq	-48(%rax),%r15
@@ -1087,16 +1138,14 @@
 .type	rsaz_512_scatter4,@function
 .align	16
 rsaz_512_scatter4:
-	leaq	(%rdi,%rdx,4),%rdi
+	leaq	(%rdi,%rdx,8),%rdi
 	movl	$8,%r9d
 	jmp	.Loop_scatter
 .align	16
 .Loop_scatter:
 	movq	(%rsi),%rax
 	leaq	8(%rsi),%rsi
-	movl	%eax,(%rdi)
-	shrq	$32,%rax
-	movl	%eax,64(%rdi)
+	movq	%rax,(%rdi)
 	leaq	128(%rdi),%rdi
 	decl	%r9d
 	jnz	.Loop_scatter
@@ -1108,20 +1157,73 @@
 .type	rsaz_512_gather4,@function
 .align	16
 rsaz_512_gather4:
-	leaq	(%rsi,%rdx,4),%rsi
+	movd	%edx,%xmm8
+	movdqa	.Linc+16(%rip),%xmm1
+	movdqa	.Linc(%rip),%xmm0
+
+	pshufd	$0,%xmm8,%xmm8
+	movdqa	%xmm1,%xmm7
+	movdqa	%xmm1,%xmm2
+	paddd	%xmm0,%xmm1
+	pcmpeqd	%xmm8,%xmm0
+	movdqa	%xmm7,%xmm3
+	paddd	%xmm1,%xmm2
+	pcmpeqd	%xmm8,%xmm1
+	movdqa	%xmm7,%xmm4
+	paddd	%xmm2,%xmm3
+	pcmpeqd	%xmm8,%xmm2
+	movdqa	%xmm7,%xmm5
+	paddd	%xmm3,%xmm4
+	pcmpeqd	%xmm8,%xmm3
+	movdqa	%xmm7,%xmm6
+	paddd	%xmm4,%xmm5
+	pcmpeqd	%xmm8,%xmm4
+	paddd	%xmm5,%xmm6
+	pcmpeqd	%xmm8,%xmm5
+	paddd	%xmm6,%xmm7
+	pcmpeqd	%xmm8,%xmm6
+	pcmpeqd	%xmm8,%xmm7
 	movl	$8,%r9d
 	jmp	.Loop_gather
 .align	16
 .Loop_gather:
-	movl	(%rsi),%eax
-	movl	64(%rsi),%r8d
+	movdqa	0(%rsi),%xmm8
+	movdqa	16(%rsi),%xmm9
+	movdqa	32(%rsi),%xmm10
+	movdqa	48(%rsi),%xmm11
+	pand	%xmm0,%xmm8
+	movdqa	64(%rsi),%xmm12
+	pand	%xmm1,%xmm9
+	movdqa	80(%rsi),%xmm13
+	pand	%xmm2,%xmm10
+	movdqa	96(%rsi),%xmm14
+	pand	%xmm3,%xmm11
+	movdqa	112(%rsi),%xmm15
 	leaq	128(%rsi),%rsi
-	shlq	$32,%r8
-	orq	%r8,%rax
-	movq	%rax,(%rdi)
+	pand	%xmm4,%xmm12
+	pand	%xmm5,%xmm13
+	pand	%xmm6,%xmm14
+	pand	%xmm7,%xmm15
+	por	%xmm10,%xmm8
+	por	%xmm11,%xmm9
+	por	%xmm12,%xmm8
+	por	%xmm13,%xmm9
+	por	%xmm14,%xmm8
+	por	%xmm15,%xmm9
+
+	por	%xmm9,%xmm8
+	pshufd	$0x4e,%xmm8,%xmm9
+	por	%xmm9,%xmm8
+	movq	%xmm8,(%rdi)
 	leaq	8(%rdi),%rdi
 	decl	%r9d
 	jnz	.Loop_gather
 	.byte	0xf3,0xc3
+.LSEH_end_rsaz_512_gather4:
 .size	rsaz_512_gather4,.-rsaz_512_gather4
+
+.align	64
+.Linc:
+.long	0,0, 1,1
+.long	2,2, 2,2
 #endif

third_party/boringssl/linux-x86_64/crypto/bn/x86_64-mont.S

diff --git a/third_party/boringssl/linux-x86_64/crypto/bn/x86_64-mont.S b/third_party/boringssl/linux-x86_64/crypto/bn/x86_64-mont.S
index 4d401c6..83926ad 100644
--- a/third_party/boringssl/linux-x86_64/crypto/bn/x86_64-mont.S
+++ b/third_party/boringssl/linux-x86_64/crypto/bn/x86_64-mont.S
@@ -636,20 +636,20 @@
 
 
 
-	leaq	-64(%rsp,%r9,4),%r11
+	leaq	-64(%rsp,%r9,2),%r11
 	movq	(%r8),%r8
 	subq	%rsi,%r11
 	andq	$4095,%r11
 	cmpq	%r11,%r10
 	jb	.Lsqr8x_sp_alt
 	subq	%r11,%rsp
-	leaq	-64(%rsp,%r9,4),%rsp
+	leaq	-64(%rsp,%r9,2),%rsp
 	jmp	.Lsqr8x_sp_done
 
 .align	32
 .Lsqr8x_sp_alt:
-	leaq	4096-64(,%r9,4),%r10
-	leaq	-64(%rsp,%r9,4),%rsp
+	leaq	4096-64(,%r9,2),%r10
+	leaq	-64(%rsp,%r9,2),%rsp
 	subq	%r10,%r11
 	movq	$0,%r10
 	cmovcq	%r10,%r11
@@ -659,58 +659,80 @@
 	movq	%r9,%r10
 	negq	%r9
 
-	leaq	64(%rsp,%r9,2),%r11
 	movq	%r8,32(%rsp)
 	movq	%rax,40(%rsp)
 .Lsqr8x_body:
 
-	movq	%r9,%rbp
-.byte	102,73,15,110,211
-	shrq	$3+2,%rbp
-	movl	OPENSSL_ia32cap_P+8(%rip),%eax
-	jmp	.Lsqr8x_copy_n
-
-.align	32
-.Lsqr8x_copy_n:
-	movq	0(%rcx),%xmm0
-	movq	8(%rcx),%xmm1
-	movq	16(%rcx),%xmm3
-	movq	24(%rcx),%xmm4
-	leaq	32(%rcx),%rcx
-	movdqa	%xmm0,0(%r11)
-	movdqa	%xmm1,16(%r11)
-	movdqa	%xmm3,32(%r11)
-	movdqa	%xmm4,48(%r11)
-	leaq	64(%r11),%r11
-	decq	%rbp
-	jnz	.Lsqr8x_copy_n
-
+.byte	102,72,15,110,209
 	pxor	%xmm0,%xmm0
 .byte	102,72,15,110,207
 .byte	102,73,15,110,218
 	call	bn_sqr8x_internal
 
-	pxor	%xmm0,%xmm0
-	leaq	48(%rsp),%rax
-	leaq	64(%rsp,%r9,2),%rdx
-	shrq	$3+2,%r9
-	movq	40(%rsp),%rsi
-	jmp	.Lsqr8x_zero
+
+
+
+	leaq	(%rdi,%r9,1),%rbx
+	movq	%r9,%rcx
+	movq	%r9,%rdx
+.byte	102,72,15,126,207
+	sarq	$3+2,%rcx
+	jmp	.Lsqr8x_sub
 
 .align	32
-.Lsqr8x_zero:
-	movdqa	%xmm0,0(%rax)
-	movdqa	%xmm0,16(%rax)
-	movdqa	%xmm0,32(%rax)
-	movdqa	%xmm0,48(%rax)
-	leaq	64(%rax),%rax
-	movdqa	%xmm0,0(%rdx)
-	movdqa	%xmm0,16(%rdx)
-	movdqa	%xmm0,32(%rdx)
-	movdqa	%xmm0,48(%rdx)
-	leaq	64(%rdx),%rdx
-	decq	%r9
-	jnz	.Lsqr8x_zero
+.Lsqr8x_sub:
+	movq	0(%rbx),%r12
+	movq	8(%rbx),%r13
+	movq	16(%rbx),%r14
+	movq	24(%rbx),%r15
+	leaq	32(%rbx),%rbx
+	sbbq	0(%rbp),%r12
+	sbbq	8(%rbp),%r13
+	sbbq	16(%rbp),%r14
+	sbbq	24(%rbp),%r15
+	leaq	32(%rbp),%rbp
+	movq	%r12,0(%rdi)
+	movq	%r13,8(%rdi)
+	movq	%r14,16(%rdi)
+	movq	%r15,24(%rdi)
+	leaq	32(%rdi),%rdi
+	incq	%rcx
+	jnz	.Lsqr8x_sub
+
+	sbbq	$0,%rax
+	leaq	(%rbx,%r9,1),%rbx
+	leaq	(%rdi,%r9,1),%rdi
+
+.byte	102,72,15,110,200
+	pxor	%xmm0,%xmm0
+	pshufd	$0,%xmm1,%xmm1
+	movq	40(%rsp),%rsi
+	jmp	.Lsqr8x_cond_copy
+
+.align	32
+.Lsqr8x_cond_copy:
+	movdqa	0(%rbx),%xmm2
+	movdqa	16(%rbx),%xmm3
+	leaq	32(%rbx),%rbx
+	movdqu	0(%rdi),%xmm4
+	movdqu	16(%rdi),%xmm5
+	leaq	32(%rdi),%rdi
+	movdqa	%xmm0,-32(%rbx)
+	movdqa	%xmm0,-16(%rbx)
+	movdqa	%xmm0,-32(%rbx,%rdx,1)
+	movdqa	%xmm0,-16(%rbx,%rdx,1)
+	pcmpeqd	%xmm1,%xmm0
+	pand	%xmm1,%xmm2
+	pand	%xmm1,%xmm3
+	pand	%xmm0,%xmm4
+	pand	%xmm0,%xmm5
+	pxor	%xmm0,%xmm0
+	por	%xmm2,%xmm4
+	por	%xmm3,%xmm5
+	movdqu	%xmm4,-32(%rdi)
+	movdqu	%xmm5,-16(%rdi)
+	addq	$32,%r9
+	jnz	.Lsqr8x_cond_copy
 
 	movq	$1,%rax
 	movq	-48(%rsi),%r15

third_party/boringssl/linux-x86_64/crypto/bn/x86_64-mont5.S

diff --git a/third_party/boringssl/linux-x86_64/crypto/bn/x86_64-mont5.S b/third_party/boringssl/linux-x86_64/crypto/bn/x86_64-mont5.S
index 02edc69..554df1f 100644
--- a/third_party/boringssl/linux-x86_64/crypto/bn/x86_64-mont5.S
+++ b/third_party/boringssl/linux-x86_64/crypto/bn/x86_64-mont5.S
@@ -17,46 +17,151 @@
 .Lmul_enter:
 	movl	%r9d,%r9d
 	movq	%rsp,%rax
-	movl	8(%rsp),%r10d
+	movd	8(%rsp),%xmm5
+	leaq	.Linc(%rip),%r10
 	pushq	%rbx
 	pushq	%rbp
 	pushq	%r12
 	pushq	%r13
 	pushq	%r14
 	pushq	%r15
+
 	leaq	2(%r9),%r11
 	negq	%r11
-	leaq	(%rsp,%r11,8),%rsp
+	leaq	-264(%rsp,%r11,8),%rsp
 	andq	$-1024,%rsp
 
 	movq	%rax,8(%rsp,%r9,8)
 .Lmul_body:
-	movq	%rdx,%r12
-	movq	%r10,%r11
-	shrq	$3,%r10
-	andq	$7,%r11
-	notq	%r10
-	leaq	.Lmagic_masks(%rip),%rax
-	andq	$3,%r10
-	leaq	96(%r12,%r11,8),%r12
-	movq	0(%rax,%r10,8),%xmm4
-	movq	8(%rax,%r10,8),%xmm5
-	movq	16(%rax,%r10,8),%xmm6
-	movq	24(%rax,%r10,8),%xmm7
+	leaq	128(%rdx),%r12
+	movdqa	0(%r10),%xmm0
+	movdqa	16(%r10),%xmm1
+	leaq	24-112(%rsp,%r9,8),%r10
+	andq	$-16,%r10
 
-	movq	-96(%r12),%xmm0
-	movq	-32(%r12),%xmm1
-	pand	%xmm4,%xmm0
-	movq	32(%r12),%xmm2
-	pand	%xmm5,%xmm1
-	movq	96(%r12),%xmm3
-	pand	%xmm6,%xmm2
-	por	%xmm1,%xmm0
-	pand	%xmm7,%xmm3
+	pshufd	$0,%xmm5,%xmm5
+	movdqa	%xmm1,%xmm4
+	movdqa	%xmm1,%xmm2
+	paddd	%xmm0,%xmm1
+	pcmpeqd	%xmm5,%xmm0
+.byte	0x67
+	movdqa	%xmm4,%xmm3
+	paddd	%xmm1,%xmm2
+	pcmpeqd	%xmm5,%xmm1
+	movdqa	%xmm0,112(%r10)
+	movdqa	%xmm4,%xmm0
+
+	paddd	%xmm2,%xmm3
+	pcmpeqd	%xmm5,%xmm2
+	movdqa	%xmm1,128(%r10)
+	movdqa	%xmm4,%xmm1
+
+	paddd	%xmm3,%xmm0
+	pcmpeqd	%xmm5,%xmm3
+	movdqa	%xmm2,144(%r10)
+	movdqa	%xmm4,%xmm2
+
+	paddd	%xmm0,%xmm1
+	pcmpeqd	%xmm5,%xmm0
+	movdqa	%xmm3,160(%r10)
+	movdqa	%xmm4,%xmm3
+	paddd	%xmm1,%xmm2
+	pcmpeqd	%xmm5,%xmm1
+	movdqa	%xmm0,176(%r10)
+	movdqa	%xmm4,%xmm0
+
+	paddd	%xmm2,%xmm3
+	pcmpeqd	%xmm5,%xmm2
+	movdqa	%xmm1,192(%r10)
+	movdqa	%xmm4,%xmm1
+
+	paddd	%xmm3,%xmm0
+	pcmpeqd	%xmm5,%xmm3
+	movdqa	%xmm2,208(%r10)
+	movdqa	%xmm4,%xmm2
+
+	paddd	%xmm0,%xmm1
+	pcmpeqd	%xmm5,%xmm0
+	movdqa	%xmm3,224(%r10)
+	movdqa	%xmm4,%xmm3
+	paddd	%xmm1,%xmm2
+	pcmpeqd	%xmm5,%xmm1
+	movdqa	%xmm0,240(%r10)
+	movdqa	%xmm4,%xmm0
+
+	paddd	%xmm2,%xmm3
+	pcmpeqd	%xmm5,%xmm2
+	movdqa	%xmm1,256(%r10)
+	movdqa	%xmm4,%xmm1
+
+	paddd	%xmm3,%xmm0
+	pcmpeqd	%xmm5,%xmm3
+	movdqa	%xmm2,272(%r10)
+	movdqa	%xmm4,%xmm2
+
+	paddd	%xmm0,%xmm1
+	pcmpeqd	%xmm5,%xmm0
+	movdqa	%xmm3,288(%r10)
+	movdqa	%xmm4,%xmm3
+	paddd	%xmm1,%xmm2
+	pcmpeqd	%xmm5,%xmm1
+	movdqa	%xmm0,304(%r10)
+
+	paddd	%xmm2,%xmm3
+.byte	0x67
+	pcmpeqd	%xmm5,%xmm2
+	movdqa	%xmm1,320(%r10)
+
+	pcmpeqd	%xmm5,%xmm3
+	movdqa	%xmm2,336(%r10)
+	pand	64(%r12),%xmm0
+
+	pand	80(%r12),%xmm1
+	pand	96(%r12),%xmm2
+	movdqa	%xmm3,352(%r10)
+	pand	112(%r12),%xmm3
 	por	%xmm2,%xmm0
+	por	%xmm3,%xmm1
+	movdqa	-128(%r12),%xmm4
+	movdqa	-112(%r12),%xmm5
+	movdqa	-96(%r12),%xmm2
+	pand	112(%r10),%xmm4
+	movdqa	-80(%r12),%xmm3
+	pand	128(%r10),%xmm5
+	por	%xmm4,%xmm0
+	pand	144(%r10),%xmm2
+	por	%xmm5,%xmm1
+	pand	160(%r10),%xmm3
+	por	%xmm2,%xmm0
+	por	%xmm3,%xmm1
+	movdqa	-64(%r12),%xmm4
+	movdqa	-48(%r12),%xmm5
+	movdqa	-32(%r12),%xmm2
+	pand	176(%r10),%xmm4
+	movdqa	-16(%r12),%xmm3
+	pand	192(%r10),%xmm5
+	por	%xmm4,%xmm0
+	pand	208(%r10),%xmm2
+	por	%xmm5,%xmm1
+	pand	224(%r10),%xmm3
+	por	%xmm2,%xmm0
+	por	%xmm3,%xmm1
+	movdqa	0(%r12),%xmm4
+	movdqa	16(%r12),%xmm5
+	movdqa	32(%r12),%xmm2
+	pand	240(%r10),%xmm4
+	movdqa	48(%r12),%xmm3
+	pand	256(%r10),%xmm5
+	por	%xmm4,%xmm0
+	pand	272(%r10),%xmm2
+	por	%xmm5,%xmm1
+	pand	288(%r10),%xmm3
+	por	%xmm2,%xmm0
+	por	%xmm3,%xmm1
+	por	%xmm1,%xmm0
+	pshufd	$0x4e,%xmm0,%xmm1
+	por	%xmm1,%xmm0
 	leaq	256(%r12),%r12
-	por	%xmm3,%xmm0
-
 .byte	102,72,15,126,195
 
 	movq	(%r8),%r8
@@ -65,29 +170,14 @@
 	xorq	%r14,%r14
 	xorq	%r15,%r15
 
-	movq	-96(%r12),%xmm0
-	movq	-32(%r12),%xmm1
-	pand	%xmm4,%xmm0
-	movq	32(%r12),%xmm2
-	pand	%xmm5,%xmm1
-
 	movq	%r8,%rbp
 	mulq	%rbx
 	movq	%rax,%r10
 	movq	(%rcx),%rax
 
-	movq	96(%r12),%xmm3
-	pand	%xmm6,%xmm2
-	por	%xmm1,%xmm0
-	pand	%xmm7,%xmm3
-
 	imulq	%r10,%rbp
 	movq	%rdx,%r11
 
-	por	%xmm2,%xmm0
-	leaq	256(%r12),%r12
-	por	%xmm3,%xmm0
-
 	mulq	%rbp
 	addq	%rax,%r10
 	movq	8(%rsi),%rax
@@ -120,14 +210,12 @@
 	cmpq	%r9,%r15
 	jne	.L1st
 
-.byte	102,72,15,126,195
 
 	addq	%rax,%r13
-	movq	(%rsi),%rax
 	adcq	$0,%rdx
 	addq	%r11,%r13
 	adcq	$0,%rdx
-	movq	%r13,-16(%rsp,%r15,8)
+	movq	%r13,-16(%rsp,%r9,8)
 	movq	%rdx,%r13
 	movq	%r10,%r11
 
@@ -141,33 +229,78 @@
 	jmp	.Louter
 .align	16
 .Louter:
+	leaq	24+128(%rsp,%r9,8),%rdx
+	andq	$-16,%rdx
+	pxor	%xmm4,%xmm4
+	pxor	%xmm5,%xmm5
+	movdqa	-128(%r12),%xmm0
+	movdqa	-112(%r12),%xmm1
+	movdqa	-96(%r12),%xmm2
+	movdqa	-80(%r12),%xmm3
+	pand	-128(%rdx),%xmm0
+	pand	-112(%rdx),%xmm1
+	por	%xmm0,%xmm4
+	pand	-96(%rdx),%xmm2
+	por	%xmm1,%xmm5
+	pand	-80(%rdx),%xmm3
+	por	%xmm2,%xmm4
+	por	%xmm3,%xmm5
+	movdqa	-64(%r12),%xmm0
+	movdqa	-48(%r12),%xmm1
+	movdqa	-32(%r12),%xmm2
+	movdqa	-16(%r12),%xmm3
+	pand	-64(%rdx),%xmm0
+	pand	-48(%rdx),%xmm1
+	por	%xmm0,%xmm4
+	pand	-32(%rdx),%xmm2
+	por	%xmm1,%xmm5
+	pand	-16(%rdx),%xmm3
+	por	%xmm2,%xmm4
+	por	%xmm3,%xmm5
+	movdqa	0(%r12),%xmm0
+	movdqa	16(%r12),%xmm1
+	movdqa	32(%r12),%xmm2
+	movdqa	48(%r12),%xmm3
+	pand	0(%rdx),%xmm0
+	pand	16(%rdx),%xmm1
+	por	%xmm0,%xmm4
+	pand	32(%rdx),%xmm2
+	por	%xmm1,%xmm5
+	pand	48(%rdx),%xmm3
+	por	%xmm2,%xmm4
+	por	%xmm3,%xmm5
+	movdqa	64(%r12),%xmm0
+	movdqa	80(%r12),%xmm1
+	movdqa	96(%r12),%xmm2
+	movdqa	112(%r12),%xmm3
+	pand	64(%rdx),%xmm0
+	pand	80(%rdx),%xmm1
+	por	%xmm0,%xmm4
+	pand	96(%rdx),%xmm2
+	por	%xmm1,%xmm5
+	pand	112(%rdx),%xmm3
+	por	%xmm2,%xmm4
+	por	%xmm3,%xmm5
+	por	%xmm5,%xmm4
+	pshufd	$0x4e,%xmm4,%xmm0
+	por	%xmm4,%xmm0
+	leaq	256(%r12),%r12
+
+	movq	(%rsi),%rax
+.byte	102,72,15,126,195
+
 	xorq	%r15,%r15
 	movq	%r8,%rbp
 	movq	(%rsp),%r10
 
-	movq	-96(%r12),%xmm0
-	movq	-32(%r12),%xmm1
-	pand	%xmm4,%xmm0
-	movq	32(%r12),%xmm2
-	pand	%xmm5,%xmm1
-
 	mulq	%rbx
 	addq	%rax,%r10
 	movq	(%rcx),%rax
 	adcq	$0,%rdx
 
-	movq	96(%r12),%xmm3
-	pand	%xmm6,%xmm2
-	por	%xmm1,%xmm0
-	pand	%xmm7,%xmm3
-
 	imulq	%r10,%rbp
 	movq	%rdx,%r11
 
-	por	%xmm2,%xmm0
-	leaq	256(%r12),%r12
-	por	%xmm3,%xmm0
-
 	mulq	%rbp
 	addq	%rax,%r10
 	movq	8(%rsi),%rax
@@ -203,15 +336,12 @@
 	cmpq	%r9,%r15
 	jne	.Linner
 
-.byte	102,72,15,126,195
-
 	addq	%rax,%r13
-	movq	(%rsi),%rax
 	adcq	$0,%rdx
 	addq	%r10,%r13
-	movq	(%rsp,%r15,8),%r10
+	movq	(%rsp,%r9,8),%r10
 	adcq	$0,%rdx
-	movq	%r13,-16(%rsp,%r15,8)
+	movq	%r13,-16(%rsp,%r9,8)
 	movq	%rdx,%r13
 
 	xorq	%rdx,%rdx
@@ -257,6 +387,7 @@
 
 	movq	8(%rsp,%r9,8),%rsi
 	movq	$1,%rax
+
 	movq	-48(%rsi),%r15
 	movq	-40(%rsi),%r14
 	movq	-32(%rsi),%r13
@@ -279,10 +410,10 @@
 	pushq	%r13
 	pushq	%r14
 	pushq	%r15
+
 .byte	0x67
-	movl	%r9d,%r10d
 	shll	$3,%r9d
-	shll	$3+2,%r10d
+	leaq	(%r9,%r9,2),%r10
 	negq	%r9
 
 
@@ -292,19 +423,21 @@
 
 
 
-	leaq	-64(%rsp,%r9,2),%r11
-	subq	%rsi,%r11
+
+
+	leaq	-320(%rsp,%r9,2),%r11
+	subq	%rdi,%r11
 	andq	$4095,%r11
 	cmpq	%r11,%r10
 	jb	.Lmul4xsp_alt
 	subq	%r11,%rsp
-	leaq	-64(%rsp,%r9,2),%rsp
+	leaq	-320(%rsp,%r9,2),%rsp
 	jmp	.Lmul4xsp_done
 
 .align	32
 .Lmul4xsp_alt:
-	leaq	4096-64(,%r9,2),%r10
-	leaq	-64(%rsp,%r9,2),%rsp
+	leaq	4096-320(,%r9,2),%r10
+	leaq	-320(%rsp,%r9,2),%rsp
 	subq	%r10,%r11
 	movq	$0,%r10
 	cmovcq	%r10,%r11
@@ -320,6 +453,7 @@
 
 	movq	40(%rsp),%rsi
 	movq	$1,%rax
+
 	movq	-48(%rsi),%r15
 	movq	-40(%rsi),%r14
 	movq	-32(%rsi),%r13
@@ -335,47 +469,141 @@
 .align	32
 mul4x_internal:
 	shlq	$5,%r9
-	movl	8(%rax),%r10d
-	leaq	256(%rdx,%r9,1),%r13
+	movd	8(%rax),%xmm5
+	leaq	.Linc(%rip),%rax
+	leaq	128(%rdx,%r9,1),%r13
 	shrq	$5,%r9
-	movq	%r10,%r11
-	shrq	$3,%r10
-	andq	$7,%r11
-	notq	%r10
-	leaq	.Lmagic_masks(%rip),%rax
-	andq	$3,%r10
-	leaq	96(%rdx,%r11,8),%r12
-	movq	0(%rax,%r10,8),%xmm4
-	movq	8(%rax,%r10,8),%xmm5
-	addq	$7,%r11
-	movq	16(%rax,%r10,8),%xmm6
-	movq	24(%rax,%r10,8),%xmm7
-	andq	$7,%r11
+	movdqa	0(%rax),%xmm0
+	movdqa	16(%rax),%xmm1
+	leaq	88-112(%rsp,%r9,1),%r10
+	leaq	128(%rdx),%r12
 
-	movq	-96(%r12),%xmm0
-	leaq	256(%r12),%r14
-	movq	-32(%r12),%xmm1
-	pand	%xmm4,%xmm0
-	movq	32(%r12),%xmm2
-	pand	%xmm5,%xmm1
-	movq	96(%r12),%xmm3
-	pand	%xmm6,%xmm2
+	pshufd	$0,%xmm5,%xmm5
+	movdqa	%xmm1,%xmm4
+.byte	0x67,0x67
+	movdqa	%xmm1,%xmm2
+	paddd	%xmm0,%xmm1
+	pcmpeqd	%xmm5,%xmm0
 .byte	0x67
-	por	%xmm1,%xmm0
-	movq	-96(%r14),%xmm1
+	movdqa	%xmm4,%xmm3
+	paddd	%xmm1,%xmm2
+	pcmpeqd	%xmm5,%xmm1
+	movdqa	%xmm0,112(%r10)
+	movdqa	%xmm4,%xmm0
+
+	paddd	%xmm2,%xmm3
+	pcmpeqd	%xmm5,%xmm2
+	movdqa	%xmm1,128(%r10)
+	movdqa	%xmm4,%xmm1
+
+	paddd	%xmm3,%xmm0
+	pcmpeqd	%xmm5,%xmm3
+	movdqa	%xmm2,144(%r10)
+	movdqa	%xmm4,%xmm2
+
+	paddd	%xmm0,%xmm1
+	pcmpeqd	%xmm5,%xmm0
+	movdqa	%xmm3,160(%r10)
+	movdqa	%xmm4,%xmm3
+	paddd	%xmm1,%xmm2
+	pcmpeqd	%xmm5,%xmm1
+	movdqa	%xmm0,176(%r10)
+	movdqa	%xmm4,%xmm0
+
+	paddd	%xmm2,%xmm3
+	pcmpeqd	%xmm5,%xmm2
+	movdqa	%xmm1,192(%r10)
+	movdqa	%xmm4,%xmm1
+
+	paddd	%xmm3,%xmm0
+	pcmpeqd	%xmm5,%xmm3
+	movdqa	%xmm2,208(%r10)
+	movdqa	%xmm4,%xmm2
+
+	paddd	%xmm0,%xmm1
+	pcmpeqd	%xmm5,%xmm0
+	movdqa	%xmm3,224(%r10)
+	movdqa	%xmm4,%xmm3
+	paddd	%xmm1,%xmm2
+	pcmpeqd	%xmm5,%xmm1
+	movdqa	%xmm0,240(%r10)
+	movdqa	%xmm4,%xmm0
+
+	paddd	%xmm2,%xmm3
+	pcmpeqd	%xmm5,%xmm2
+	movdqa	%xmm1,256(%r10)
+	movdqa	%xmm4,%xmm1
+
+	paddd	%xmm3,%xmm0
+	pcmpeqd	%xmm5,%xmm3
+	movdqa	%xmm2,272(%r10)
+	movdqa	%xmm4,%xmm2
+
+	paddd	%xmm0,%xmm1
+	pcmpeqd	%xmm5,%xmm0
+	movdqa	%xmm3,288(%r10)
+	movdqa	%xmm4,%xmm3
+	paddd	%xmm1,%xmm2
+	pcmpeqd	%xmm5,%xmm1
+	movdqa	%xmm0,304(%r10)
+
+	paddd	%xmm2,%xmm3
 .byte	0x67
-	pand	%xmm7,%xmm3
-.byte	0x67
+	pcmpeqd	%xmm5,%xmm2
+	movdqa	%xmm1,320(%r10)
+
+	pcmpeqd	%xmm5,%xmm3
+	movdqa	%xmm2,336(%r10)
+	pand	64(%r12),%xmm0
+
+	pand	80(%r12),%xmm1
+	pand	96(%r12),%xmm2
+	movdqa	%xmm3,352(%r10)
+	pand	112(%r12),%xmm3
 	por	%xmm2,%xmm0
-	movq	-32(%r14),%xmm2
-.byte	0x67
-	pand	%xmm4,%xmm1
-.byte	0x67
-	por	%xmm3,%xmm0
-	movq	32(%r14),%xmm3
-
+	por	%xmm3,%xmm1
+	movdqa	-128(%r12),%xmm4
+	movdqa	-112(%r12),%xmm5
+	movdqa	-96(%r12),%xmm2
+	pand	112(%r10),%xmm4
+	movdqa	-80(%r12),%xmm3
+	pand	128(%r10),%xmm5
+	por	%xmm4,%xmm0
+	pand	144(%r10),%xmm2
+	por	%xmm5,%xmm1
+	pand	160(%r10),%xmm3
+	por	%xmm2,%xmm0
+	por	%xmm3,%xmm1
+	movdqa	-64(%r12),%xmm4
+	movdqa	-48(%r12),%xmm5
+	movdqa	-32(%r12),%xmm2
+	pand	176(%r10),%xmm4
+	movdqa	-16(%r12),%xmm3
+	pand	192(%r10),%xmm5
+	por	%xmm4,%xmm0
+	pand	208(%r10),%xmm2
+	por	%xmm5,%xmm1
+	pand	224(%r10),%xmm3
+	por	%xmm2,%xmm0
+	por	%xmm3,%xmm1
+	movdqa	0(%r12),%xmm4
+	movdqa	16(%r12),%xmm5
+	movdqa	32(%r12),%xmm2
+	pand	240(%r10),%xmm4
+	movdqa	48(%r12),%xmm3
+	pand	256(%r10),%xmm5
+	por	%xmm4,%xmm0
+	pand	272(%r10),%xmm2
+	por	%xmm5,%xmm1
+	pand	288(%r10),%xmm3
+	por	%xmm2,%xmm0
+	por	%xmm3,%xmm1
+	por	%xmm1,%xmm0
+	pshufd	$0x4e,%xmm0,%xmm1
+	por	%xmm1,%xmm0
+	leaq	256(%r12),%r12
 .byte	102,72,15,126,195
-	movq	96(%r14),%xmm0
+
 	movq	%r13,16+8(%rsp)
 	movq	%rdi,56+8(%rsp)
 
@@ -389,26 +617,10 @@
 	movq	%rax,%r10
 	movq	(%rcx),%rax
 
-	pand	%xmm5,%xmm2
-	pand	%xmm6,%xmm3
-	por	%xmm2,%xmm1
-
 	imulq	%r10,%rbp
-
-
-
-
-
-
-
-	leaq	64+8(%rsp,%r11,8),%r14
+	leaq	64+8(%rsp),%r14
 	movq	%rdx,%r11
 
-	pand	%xmm7,%xmm0
-	por	%xmm3,%xmm1
-	leaq	512(%r12),%r12
-	por	%xmm1,%xmm0
-
 	mulq	%rbp
 	addq	%rax,%r10
 	movq	8(%rsi,%r9,1),%rax
@@ -417,7 +629,7 @@
 
 	mulq	%rbx
 	addq	%rax,%r11
-	movq	16(%rcx),%rax
+	movq	8(%rcx),%rax
 	adcq	$0,%rdx
 	movq	%rdx,%r10
 
@@ -427,7 +639,7 @@
 	adcq	$0,%rdx
 	addq	%r11,%rdi
 	leaq	32(%r9),%r15
-	leaq	64(%rcx),%rcx
+	leaq	32(%rcx),%rcx
 	adcq	$0,%rdx
 	movq	%rdi,(%r14)
 	movq	%rdx,%r13
@@ -437,7 +649,7 @@
 .L1st4x:
 	mulq	%rbx
 	addq	%rax,%r10
-	movq	-32(%rcx),%rax
+	movq	-16(%rcx),%rax
 	leaq	32(%r14),%r14
 	adcq	$0,%rdx
 	movq	%rdx,%r11
@@ -453,7 +665,7 @@
 
 	mulq	%rbx
 	addq	%rax,%r11
-	movq	-16(%rcx),%rax
+	movq	-8(%rcx),%rax
 	adcq	$0,%rdx
 	movq	%rdx,%r10
 
@@ -483,7 +695,7 @@
 
 	mulq	%rbx
 	addq	%rax,%r11
-	movq	16(%rcx),%rax
+	movq	8(%rcx),%rax
 	adcq	$0,%rdx
 	movq	%rdx,%r10
 
@@ -492,7 +704,7 @@
 	movq	16(%rsi,%r15,1),%rax
 	adcq	$0,%rdx
 	addq	%r11,%rdi
-	leaq	64(%rcx),%rcx
+	leaq	32(%rcx),%rcx
 	adcq	$0,%rdx
 	movq	%rdi,(%r14)
 	movq	%rdx,%r13
@@ -502,7 +714,7 @@
 
 	mulq	%rbx
 	addq	%rax,%r10
-	movq	-32(%rcx),%rax
+	movq	-16(%rcx),%rax
 	leaq	32(%r14),%r14
 	adcq	$0,%rdx
 	movq	%rdx,%r11
@@ -518,7 +730,7 @@
 
 	mulq	%rbx
 	addq	%rax,%r11
-	movq	-16(%rcx),%rax
+	movq	-8(%rcx),%rax
 	adcq	$0,%rdx
 	movq	%rdx,%r10
 
@@ -531,8 +743,7 @@
 	movq	%rdi,-16(%r14)
 	movq	%rdx,%r13
 
-.byte	102,72,15,126,195
-	leaq	(%rcx,%r9,2),%rcx
+	leaq	(%rcx,%r9,1),%rcx
 
 	xorq	%rdi,%rdi
 	addq	%r10,%r13
@@ -543,6 +754,63 @@
 
 .align	32
 .Louter4x:
+	leaq	16+128(%r14),%rdx
+	pxor	%xmm4,%xmm4
+	pxor	%xmm5,%xmm5
+	movdqa	-128(%r12),%xmm0
+	movdqa	-112(%r12),%xmm1
+	movdqa	-96(%r12),%xmm2
+	movdqa	-80(%r12),%xmm3
+	pand	-128(%rdx),%xmm0
+	pand	-112(%rdx),%xmm1
+	por	%xmm0,%xmm4
+	pand	-96(%rdx),%xmm2
+	por	%xmm1,%xmm5
+	pand	-80(%rdx),%xmm3
+	por	%xmm2,%xmm4
+	por	%xmm3,%xmm5
+	movdqa	-64(%r12),%xmm0
+	movdqa	-48(%r12),%xmm1
+	movdqa	-32(%r12),%xmm2
+	movdqa	-16(%r12),%xmm3
+	pand	-64(%rdx),%xmm0
+	pand	-48(%rdx),%xmm1
+	por	%xmm0,%xmm4
+	pand	-32(%rdx),%xmm2
+	por	%xmm1,%xmm5
+	pand	-16(%rdx),%xmm3
+	por	%xmm2,%xmm4
+	por	%xmm3,%xmm5
+	movdqa	0(%r12),%xmm0
+	movdqa	16(%r12),%xmm1
+	movdqa	32(%r12),%xmm2
+	movdqa	48(%r12),%xmm3
+	pand	0(%rdx),%xmm0
+	pand	16(%rdx),%xmm1
+	por	%xmm0,%xmm4
+	pand	32(%rdx),%xmm2
+	por	%xmm1,%xmm5
+	pand	48(%rdx),%xmm3
+	por	%xmm2,%xmm4
+	por	%xmm3,%xmm5
+	movdqa	64(%r12),%xmm0
+	movdqa	80(%r12),%xmm1
+	movdqa	96(%r12),%xmm2
+	movdqa	112(%r12),%xmm3
+	pand	64(%rdx),%xmm0
+	pand	80(%rdx),%xmm1
+	por	%xmm0,%xmm4
+	pand	96(%rdx),%xmm2
+	por	%xmm1,%xmm5
+	pand	112(%rdx),%xmm3
+	por	%xmm2,%xmm4
+	por	%xmm3,%xmm5
+	por	%xmm5,%xmm4
+	pshufd	$0x4e,%xmm4,%xmm0
+	por	%xmm4,%xmm0
+	leaq	256(%r12),%r12
+.byte	102,72,15,126,195
+
 	movq	(%r14,%r9,1),%r10
 	movq	%r8,%rbp
 	mulq	%rbx
@@ -550,25 +818,11 @@
 	movq	(%rcx),%rax
 	adcq	$0,%rdx
 
-	movq	-96(%r12),%xmm0
-	movq	-32(%r12),%xmm1
-	pand	%xmm4,%xmm0
-	movq	32(%r12),%xmm2
-	pand	%xmm5,%xmm1
-	movq	96(%r12),%xmm3
-
 	imulq	%r10,%rbp
-.byte	0x67
 	movq	%rdx,%r11
 	movq	%rdi,(%r14)
 
-	pand	%xmm6,%xmm2
-	por	%xmm1,%xmm0
-	pand	%xmm7,%xmm3
-	por	%xmm2,%xmm0
 	leaq	(%r14,%r9,1),%r14
-	leaq	256(%r12),%r12
-	por	%xmm3,%xmm0
 
 	mulq	%rbp
 	addq	%rax,%r10
@@ -578,7 +832,7 @@
 
 	mulq	%rbx
 	addq	%rax,%r11
-	movq	16(%rcx),%rax
+	movq	8(%rcx),%rax
 	adcq	$0,%rdx
 	addq	8(%r14),%r11
 	adcq	$0,%rdx
@@ -590,7 +844,7 @@
 	adcq	$0,%rdx
 	addq	%r11,%rdi
 	leaq	32(%r9),%r15
-	leaq	64(%rcx),%rcx
+	leaq	32(%rcx),%rcx
 	adcq	$0,%rdx
 	movq	%rdx,%r13
 	jmp	.Linner4x
@@ -599,7 +853,7 @@
 .Linner4x:
 	mulq	%rbx
 	addq	%rax,%r10
-	movq	-32(%rcx),%rax
+	movq	-16(%rcx),%rax
 	adcq	$0,%rdx
 	addq	16(%r14),%r10
 	leaq	32(%r14),%r14
@@ -617,7 +871,7 @@
 
 	mulq	%rbx
 	addq	%rax,%r11
-	movq	-16(%rcx),%rax
+	movq	-8(%rcx),%rax
 	adcq	$0,%rdx
 	addq	-8(%r14),%r11
 	adcq	$0,%rdx
@@ -651,7 +905,7 @@
 
 	mulq	%rbx
 	addq	%rax,%r11
-	movq	16(%rcx),%rax
+	movq	8(%rcx),%rax
 	adcq	$0,%rdx
 	addq	8(%r14),%r11
 	adcq	$0,%rdx
@@ -662,7 +916,7 @@
 	movq	16(%rsi,%r15,1),%rax
 	adcq	$0,%rdx
 	addq	%r11,%rdi
-	leaq	64(%rcx),%rcx
+	leaq	32(%rcx),%rcx
 	adcq	$0,%rdx
 	movq	%r13,-8(%r14)
 	movq	%rdx,%r13
@@ -672,7 +926,7 @@
 
 	mulq	%rbx
 	addq	%rax,%r10
-	movq	-32(%rcx),%rax
+	movq	-16(%rcx),%rax
 	adcq	$0,%rdx
 	addq	16(%r14),%r10
 	leaq	32(%r14),%r14
@@ -691,7 +945,7 @@
 	mulq	%rbx
 	addq	%rax,%r11
 	movq	%rbp,%rax
-	movq	-16(%rcx),%rbp
+	movq	-8(%rcx),%rbp
 	adcq	$0,%rdx
 	addq	-8(%r14),%r11
 	adcq	$0,%rdx
@@ -706,9 +960,8 @@
 	movq	%r13,-24(%r14)
 	movq	%rdx,%r13
 
-.byte	102,72,15,126,195
 	movq	%rdi,-16(%r14)
-	leaq	(%rcx,%r9,2),%rcx
+	leaq	(%rcx,%r9,1),%rcx
 
 	xorq	%rdi,%rdi
 	addq	%r10,%r13
@@ -719,16 +972,23 @@
 
 	cmpq	16+8(%rsp),%r12
 	jb	.Louter4x
+	xorq	%rax,%rax
 	subq	%r13,%rbp
 	adcq	%r15,%r15
 	orq	%r15,%rdi
-	xorq	$1,%rdi
+	subq	%rdi,%rax
 	leaq	(%r14,%r9,1),%rbx
-	leaq	(%rcx,%rdi,8),%rbp
+	movq	(%rcx),%r12
+	leaq	(%rcx),%rbp
 	movq	%r9,%rcx
 	sarq	$3+2,%rcx
 	movq	56+8(%rsp),%rdi
-	jmp	.Lsqr4x_sub
+	decq	%r12
+	xorq	%r10,%r10
+	movq	8(%rbp),%r13
+	movq	16(%rbp),%r14
+	movq	24(%rbp),%r15
+	jmp	.Lsqr4x_sub_entry
 .size	mul4x_internal,.-mul4x_internal
 .globl	bn_power5
 .hidden bn_power5
@@ -742,9 +1002,9 @@
 	pushq	%r13
 	pushq	%r14
 	pushq	%r15
-	movl	%r9d,%r10d
+
 	shll	$3,%r9d
-	shll	$3+2,%r10d
+	leal	(%r9,%r9,2),%r10d
 	negq	%r9
 	movq	(%r8),%r8
 
@@ -754,19 +1014,20 @@
 
 
 
-	leaq	-64(%rsp,%r9,2),%r11
-	subq	%rsi,%r11
+
+	leaq	-320(%rsp,%r9,2),%r11
+	subq	%rdi,%r11
 	andq	$4095,%r11
 	cmpq	%r11,%r10
 	jb	.Lpwr_sp_alt
 	subq	%r11,%rsp
-	leaq	-64(%rsp,%r9,2),%rsp
+	leaq	-320(%rsp,%r9,2),%rsp
 	jmp	.Lpwr_sp_done
 
 .align	32
 .Lpwr_sp_alt:
-	leaq	4096-64(,%r9,2),%r10
-	leaq	-64(%rsp,%r9,2),%rsp
+	leaq	4096-320(,%r9,2),%r10
+	leaq	-320(%rsp,%r9,2),%rsp
 	subq	%r10,%r11
 	movq	$0,%r10
 	cmovcq	%r10,%r11
@@ -794,10 +1055,15 @@
 .byte	102,72,15,110,226
 
 	call	__bn_sqr8x_internal
+	call	__bn_post4x_internal
 	call	__bn_sqr8x_internal
+	call	__bn_post4x_internal
 	call	__bn_sqr8x_internal
+	call	__bn_post4x_internal
 	call	__bn_sqr8x_internal
+	call	__bn_post4x_internal
 	call	__bn_sqr8x_internal
+	call	__bn_post4x_internal
 
 .byte	102,72,15,126,209
 .byte	102,72,15,126,226
@@ -1342,9 +1608,9 @@
 	movq	%rbx,-16(%rdi)
 	movq	%r8,-8(%rdi)
 .byte	102,72,15,126,213
-sqr8x_reduction:
+__bn_sqr8x_reduction:
 	xorq	%rax,%rax
-	leaq	(%rbp,%r9,2),%rcx
+	leaq	(%r9,%rbp,1),%rcx
 	leaq	48+8(%rsp,%r9,2),%rdx
 	movq	%rcx,0+8(%rsp)
 	leaq	48+8(%rsp,%r9,1),%rdi
@@ -1377,14 +1643,14 @@
 .align	32
 .L8x_reduce:
 	mulq	%rbx
-	movq	16(%rbp),%rax
+	movq	8(%rbp),%rax
 	negq	%r8
 	movq	%rdx,%r8
 	adcq	$0,%r8
 
 	mulq	%rbx
 	addq	%rax,%r9
-	movq	32(%rbp),%rax
+	movq	16(%rbp),%rax
 	adcq	$0,%rdx
 	addq	%r9,%r8
 	movq	%rbx,48-8+8(%rsp,%rcx,8)
@@ -1393,7 +1659,7 @@
 
 	mulq	%rbx
 	addq	%rax,%r10
-	movq	48(%rbp),%rax
+	movq	24(%rbp),%rax
 	adcq	$0,%rdx
 	addq	%r10,%r9
 	movq	32+8(%rsp),%rsi
@@ -1402,7 +1668,7 @@
 
 	mulq	%rbx
 	addq	%rax,%r11
-	movq	64(%rbp),%rax
+	movq	32(%rbp),%rax
 	adcq	$0,%rdx
 	imulq	%r8,%rsi
 	addq	%r11,%r10
@@ -1411,7 +1677,7 @@
 
 	mulq	%rbx
 	addq	%rax,%r12
-	movq	80(%rbp),%rax
+	movq	40(%rbp),%rax
 	adcq	$0,%rdx
 	addq	%r12,%r11
 	movq	%rdx,%r12
@@ -1419,7 +1685,7 @@
 
 	mulq	%rbx
 	addq	%rax,%r13
-	movq	96(%rbp),%rax
+	movq	48(%rbp),%rax
 	adcq	$0,%rdx
 	addq	%r13,%r12
 	movq	%rdx,%r13
@@ -1427,7 +1693,7 @@
 
 	mulq	%rbx
 	addq	%rax,%r14
-	movq	112(%rbp),%rax
+	movq	56(%rbp),%rax
 	adcq	$0,%rdx
 	addq	%r14,%r13
 	movq	%rdx,%r14
@@ -1445,7 +1711,7 @@
 	decl	%ecx
 	jnz	.L8x_reduce
 
-	leaq	128(%rbp),%rbp
+	leaq	64(%rbp),%rbp
 	xorq	%rax,%rax
 	movq	8+8(%rsp),%rdx
 	cmpq	0+8(%rsp),%rbp
@@ -1471,14 +1737,14 @@
 .L8x_tail:
 	mulq	%rbx
 	addq	%rax,%r8
-	movq	16(%rbp),%rax
+	movq	8(%rbp),%rax
 	movq	%r8,(%rdi)
 	movq	%rdx,%r8
 	adcq	$0,%r8
 
 	mulq	%rbx
 	addq	%rax,%r9
-	movq	32(%rbp),%rax
+	movq	16(%rbp),%rax
 	adcq	$0,%rdx
 	addq	%r9,%r8
 	leaq	8(%rdi),%rdi
@@ -1487,7 +1753,7 @@
 
 	mulq	%rbx
 	addq	%rax,%r10
-	movq	48(%rbp),%rax
+	movq	24(%rbp),%rax
 	adcq	$0,%rdx
 	addq	%r10,%r9
 	movq	%rdx,%r10
@@ -1495,7 +1761,7 @@
 
 	mulq	%rbx
 	addq	%rax,%r11
-	movq	64(%rbp),%rax
+	movq	32(%rbp),%rax
 	adcq	$0,%rdx
 	addq	%r11,%r10
 	movq	%rdx,%r11
@@ -1503,7 +1769,7 @@
 
 	mulq	%rbx
 	addq	%rax,%r12
-	movq	80(%rbp),%rax
+	movq	40(%rbp),%rax
 	adcq	$0,%rdx
 	addq	%r12,%r11
 	movq	%rdx,%r12
@@ -1511,7 +1777,7 @@
 
 	mulq	%rbx
 	addq	%rax,%r13
-	movq	96(%rbp),%rax
+	movq	48(%rbp),%rax
 	adcq	$0,%rdx
 	addq	%r13,%r12
 	movq	%rdx,%r13
@@ -1519,7 +1785,7 @@
 
 	mulq	%rbx
 	addq	%rax,%r14
-	movq	112(%rbp),%rax
+	movq	56(%rbp),%rax
 	adcq	$0,%rdx
 	addq	%r14,%r13
 	movq	%rdx,%r14
@@ -1537,7 +1803,7 @@
 	decl	%ecx
 	jnz	.L8x_tail
 
-	leaq	128(%rbp),%rbp
+	leaq	64(%rbp),%rbp
 	movq	8+8(%rsp),%rdx
 	cmpq	0+8(%rsp),%rbp
 	jae	.L8x_tail_done
@@ -1561,6 +1827,15 @@
 .align	32
 .L8x_tail_done:
 	addq	(%rdx),%r8
+	adcq	$0,%r9
+	adcq	$0,%r10
+	adcq	$0,%r11
+	adcq	$0,%r12
+	adcq	$0,%r13
+	adcq	$0,%r14
+	adcq	$0,%r15
+
+
 	xorq	%rax,%rax
 
 	negq	%rsi
@@ -1574,7 +1849,7 @@
 	adcq	48(%rdi),%r14
 	adcq	56(%rdi),%r15
 	adcq	$0,%rax
-	movq	-16(%rbp),%rcx
+	movq	-8(%rbp),%rcx
 	xorq	%rsi,%rsi
 
 .byte	102,72,15,126,213
@@ -1592,44 +1867,62 @@
 
 	cmpq	%rdx,%rdi
 	jb	.L8x_reduction_loop
-
-	subq	%r15,%rcx
-	leaq	(%rdi,%r9,1),%rbx
-	adcq	%rsi,%rsi
-	movq	%r9,%rcx
-	orq	%rsi,%rax
-.byte	102,72,15,126,207
-	xorq	$1,%rax
-.byte	102,72,15,126,206
-	leaq	(%rbp,%rax,8),%rbp
-	sarq	$3+2,%rcx
-	jmp	.Lsqr4x_sub
-
+	.byte	0xf3,0xc3
+.size	bn_sqr8x_internal,.-bn_sqr8x_internal
+.type	__bn_post4x_internal,@function
 .align	32
+__bn_post4x_internal:
+	movq	0(%rbp),%r12
+	leaq	(%rdi,%r9,1),%rbx
+	movq	%r9,%rcx
+.byte	102,72,15,126,207
+	negq	%rax
+.byte	102,72,15,126,206
+	sarq	$3+2,%rcx
+	decq	%r12
+	xorq	%r10,%r10
+	movq	8(%rbp),%r13
+	movq	16(%rbp),%r14
+	movq	24(%rbp),%r15
+	jmp	.Lsqr4x_sub_entry
+
+.align	16
 .Lsqr4x_sub:
-.byte	0x66
-	movq	0(%rbx),%r12
-	movq	8(%rbx),%r13
-	sbbq	0(%rbp),%r12
-	movq	16(%rbx),%r14
-	sbbq	16(%rbp),%r13
-	movq	24(%rbx),%r15
-	leaq	32(%rbx),%rbx
-	sbbq	32(%rbp),%r14
+	movq	0(%rbp),%r12
+	movq	8(%rbp),%r13
+	movq	16(%rbp),%r14
+	movq	24(%rbp),%r15
+.Lsqr4x_sub_entry:
+	leaq	32(%rbp),%rbp
+	notq	%r12
+	notq	%r13
+	notq	%r14
+	notq	%r15
+	andq	%rax,%r12
+	andq	%rax,%r13
+	andq	%rax,%r14
+	andq	%rax,%r15
+
+	negq	%r10
+	adcq	0(%rbx),%r12
+	adcq	8(%rbx),%r13
+	adcq	16(%rbx),%r14
+	adcq	24(%rbx),%r15
 	movq	%r12,0(%rdi)
-	sbbq	48(%rbp),%r15
-	leaq	64(%rbp),%rbp
+	leaq	32(%rbx),%rbx
 	movq	%r13,8(%rdi)
+	sbbq	%r10,%r10
 	movq	%r14,16(%rdi)
 	movq	%r15,24(%rdi)
 	leaq	32(%rdi),%rdi
 
 	incq	%rcx
 	jnz	.Lsqr4x_sub
+
 	movq	%r9,%r10
 	negq	%r9
 	.byte	0xf3,0xc3
-.size	bn_sqr8x_internal,.-bn_sqr8x_internal
+.size	__bn_post4x_internal,.-__bn_post4x_internal
 .globl	bn_from_montgomery
 .hidden bn_from_montgomery
 .type	bn_from_montgomery,@function
@@ -1652,10 +1945,9 @@
 	pushq	%r13
 	pushq	%r14
 	pushq	%r15
-.byte	0x67
-	movl	%r9d,%r10d
+
 	shll	$3,%r9d
-	shll	$3+2,%r10d
+	leaq	(%r9,%r9,2),%r10
 	negq	%r9
 	movq	(%r8),%r8
 
@@ -1665,19 +1957,20 @@
 
 
 
-	leaq	-64(%rsp,%r9,2),%r11
-	subq	%rsi,%r11
+
+	leaq	-320(%rsp,%r9,2),%r11
+	subq	%rdi,%r11
 	andq	$4095,%r11
 	cmpq	%r11,%r10
 	jb	.Lfrom_sp_alt
 	subq	%r11,%rsp
-	leaq	-64(%rsp,%r9,2),%rsp
+	leaq	-320(%rsp,%r9,2),%rsp
 	jmp	.Lfrom_sp_done
 
 .align	32
 .Lfrom_sp_alt:
-	leaq	4096-64(,%r9,2),%r10
-	leaq	-64(%rsp,%r9,2),%rsp
+	leaq	4096-320(,%r9,2),%r10
+	leaq	-320(%rsp,%r9,2),%rsp
 	subq	%r10,%r11
 	movq	$0,%r10
 	cmovcq	%r10,%r11
@@ -1728,7 +2021,8 @@
 .byte	0x67
 	movq	%rcx,%rbp
 .byte	102,73,15,110,218
-	call	sqr8x_reduction
+	call	__bn_sqr8x_reduction
+	call	__bn_post4x_internal
 
 	pxor	%xmm0,%xmm0
 	leaq	48(%rsp),%rax
@@ -1778,46 +2072,170 @@
 .globl	bn_gather5
 .hidden bn_gather5
 .type	bn_gather5,@function
-.align	16
+.align	32
 bn_gather5:
-	movl	%ecx,%r11d
-	shrl	$3,%ecx
-	andq	$7,%r11
-	notl	%ecx
-	leaq	.Lmagic_masks(%rip),%rax
-	andl	$3,%ecx
-	leaq	128(%rdx,%r11,8),%rdx
-	movq	0(%rax,%rcx,8),%xmm4
-	movq	8(%rax,%rcx,8),%xmm5
-	movq	16(%rax,%rcx,8),%xmm6
-	movq	24(%rax,%rcx,8),%xmm7
-	jmp	.Lgather
-.align	16
-.Lgather:
-	movq	-128(%rdx),%xmm0
-	movq	-64(%rdx),%xmm1
-	pand	%xmm4,%xmm0
-	movq	0(%rdx),%xmm2
-	pand	%xmm5,%xmm1
-	movq	64(%rdx),%xmm3
-	pand	%xmm6,%xmm2
-	por	%xmm1,%xmm0
-	pand	%xmm7,%xmm3
-.byte	0x67,0x67
-	por	%xmm2,%xmm0
-	leaq	256(%rdx),%rdx
-	por	%xmm3,%xmm0
+.LSEH_begin_bn_gather5:
 
+.byte	0x4c,0x8d,0x14,0x24
+.byte	0x48,0x81,0xec,0x08,0x01,0x00,0x00
+	leaq	.Linc(%rip),%rax
+	andq	$-16,%rsp
+
+	movd	%ecx,%xmm5
+	movdqa	0(%rax),%xmm0
+	movdqa	16(%rax),%xmm1
+	leaq	128(%rdx),%r11
+	leaq	128(%rsp),%rax
+
+	pshufd	$0,%xmm5,%xmm5
+	movdqa	%xmm1,%xmm4
+	movdqa	%xmm1,%xmm2
+	paddd	%xmm0,%xmm1
+	pcmpeqd	%xmm5,%xmm0
+	movdqa	%xmm4,%xmm3
+
+	paddd	%xmm1,%xmm2
+	pcmpeqd	%xmm5,%xmm1
+	movdqa	%xmm0,-128(%rax)
+	movdqa	%xmm4,%xmm0
+
+	paddd	%xmm2,%xmm3
+	pcmpeqd	%xmm5,%xmm2
+	movdqa	%xmm1,-112(%rax)
+	movdqa	%xmm4,%xmm1
+
+	paddd	%xmm3,%xmm0
+	pcmpeqd	%xmm5,%xmm3
+	movdqa	%xmm2,-96(%rax)
+	movdqa	%xmm4,%xmm2
+	paddd	%xmm0,%xmm1
+	pcmpeqd	%xmm5,%xmm0
+	movdqa	%xmm3,-80(%rax)
+	movdqa	%xmm4,%xmm3
+
+	paddd	%xmm1,%xmm2
+	pcmpeqd	%xmm5,%xmm1
+	movdqa	%xmm0,-64(%rax)
+	movdqa	%xmm4,%xmm0
+
+	paddd	%xmm2,%xmm3
+	pcmpeqd	%xmm5,%xmm2
+	movdqa	%xmm1,-48(%rax)
+	movdqa	%xmm4,%xmm1
+
+	paddd	%xmm3,%xmm0
+	pcmpeqd	%xmm5,%xmm3
+	movdqa	%xmm2,-32(%rax)
+	movdqa	%xmm4,%xmm2
+	paddd	%xmm0,%xmm1
+	pcmpeqd	%xmm5,%xmm0
+	movdqa	%xmm3,-16(%rax)
+	movdqa	%xmm4,%xmm3
+
+	paddd	%xmm1,%xmm2
+	pcmpeqd	%xmm5,%xmm1
+	movdqa	%xmm0,0(%rax)
+	movdqa	%xmm4,%xmm0
+
+	paddd	%xmm2,%xmm3
+	pcmpeqd	%xmm5,%xmm2
+	movdqa	%xmm1,16(%rax)
+	movdqa	%xmm4,%xmm1
+
+	paddd	%xmm3,%xmm0
+	pcmpeqd	%xmm5,%xmm3
+	movdqa	%xmm2,32(%rax)
+	movdqa	%xmm4,%xmm2
+	paddd	%xmm0,%xmm1
+	pcmpeqd	%xmm5,%xmm0
+	movdqa	%xmm3,48(%rax)
+	movdqa	%xmm4,%xmm3
+
+	paddd	%xmm1,%xmm2
+	pcmpeqd	%xmm5,%xmm1
+	movdqa	%xmm0,64(%rax)
+	movdqa	%xmm4,%xmm0
+
+	paddd	%xmm2,%xmm3
+	pcmpeqd	%xmm5,%xmm2
+	movdqa	%xmm1,80(%rax)
+	movdqa	%xmm4,%xmm1
+
+	paddd	%xmm3,%xmm0
+	pcmpeqd	%xmm5,%xmm3
+	movdqa	%xmm2,96(%rax)
+	movdqa	%xmm4,%xmm2
+	movdqa	%xmm3,112(%rax)
+	jmp	.Lgather
+
+.align	32
+.Lgather:
+	pxor	%xmm4,%xmm4
+	pxor	%xmm5,%xmm5
+	movdqa	-128(%r11),%xmm0
+	movdqa	-112(%r11),%xmm1
+	movdqa	-96(%r11),%xmm2
+	pand	-128(%rax),%xmm0
+	movdqa	-80(%r11),%xmm3
+	pand	-112(%rax),%xmm1
+	por	%xmm0,%xmm4
+	pand	-96(%rax),%xmm2
+	por	%xmm1,%xmm5
+	pand	-80(%rax),%xmm3
+	por	%xmm2,%xmm4
+	por	%xmm3,%xmm5
+	movdqa	-64(%r11),%xmm0
+	movdqa	-48(%r11),%xmm1
+	movdqa	-32(%r11),%xmm2
+	pand	-64(%rax),%xmm0
+	movdqa	-16(%r11),%xmm3
+	pand	-48(%rax),%xmm1
+	por	%xmm0,%xmm4
+	pand	-32(%rax),%xmm2
+	por	%xmm1,%xmm5
+	pand	-16(%rax),%xmm3
+	por	%xmm2,%xmm4
+	por	%xmm3,%xmm5
+	movdqa	0(%r11),%xmm0
+	movdqa	16(%r11),%xmm1
+	movdqa	32(%r11),%xmm2
+	pand	0(%rax),%xmm0
+	movdqa	48(%r11),%xmm3
+	pand	16(%rax),%xmm1
+	por	%xmm0,%xmm4
+	pand	32(%rax),%xmm2
+	por	%xmm1,%xmm5
+	pand	48(%rax),%xmm3
+	por	%xmm2,%xmm4
+	por	%xmm3,%xmm5
+	movdqa	64(%r11),%xmm0
+	movdqa	80(%r11),%xmm1
+	movdqa	96(%r11),%xmm2
+	pand	64(%rax),%xmm0
+	movdqa	112(%r11),%xmm3
+	pand	80(%rax),%xmm1
+	por	%xmm0,%xmm4
+	pand	96(%rax),%xmm2
+	por	%xmm1,%xmm5
+	pand	112(%rax),%xmm3
+	por	%xmm2,%xmm4
+	por	%xmm3,%xmm5
+	por	%xmm5,%xmm4
+	leaq	256(%r11),%r11
+	pshufd	$0x4e,%xmm4,%xmm0
+	por	%xmm4,%xmm0
 	movq	%xmm0,(%rdi)
 	leaq	8(%rdi),%rdi
 	subl	$1,%esi
 	jnz	.Lgather
+
+	leaq	(%r10),%rsp
 	.byte	0xf3,0xc3
 .LSEH_end_bn_gather5:
 .size	bn_gather5,.-bn_gather5
 .align	64
-.Lmagic_masks:
-.long	0,0, 0,0, 0,0, -1,-1
-.long	0,0, 0,0, 0,0,  0,0
+.Linc:
+.long	0,0, 1,1
+.long	2,2, 2,2
 .byte	77,111,110,116,103,111,109,101,114,121,32,77,117,108,116,105,112,108,105,99,97,116,105,111,110,32,119,105,116,104,32,115,99,97,116,116,101,114,47,103,97,116,104,101,114,32,102,111,114,32,120,56,54,95,54,52,44,32,67,82,89,80,84,79,71,65,77,83,32,98,121,32,60,97,112,112,114,111,64,111,112,101,110,115,115,108,46,111,114,103,62,0
 #endif

third_party/boringssl/linux-x86_64/crypto/chacha/chacha-x86_64.S

diff --git a/third_party/boringssl/linux-x86_64/crypto/chacha/chacha-x86_64.S b/third_party/boringssl/linux-x86_64/crypto/chacha/chacha-x86_64.S
new file mode 100644
index 0000000..e994940
--- /dev/null
+++ b/third_party/boringssl/linux-x86_64/crypto/chacha/chacha-x86_64.S
@@ -0,0 +1,1585 @@
+#if defined(__x86_64__)
+.text	
+
+.extern	OPENSSL_ia32cap_P
+.hidden OPENSSL_ia32cap_P
+
+.align	64
+.Lzero:
+.long	0,0,0,0
+.Lone:
+.long	1,0,0,0
+.Linc:
+.long	0,1,2,3
+.Lfour:
+.long	4,4,4,4
+.Lincy:
+.long	0,2,4,6,1,3,5,7
+.Leight:
+.long	8,8,8,8,8,8,8,8
+.Lrot16:
+.byte	0x2,0x3,0x0,0x1, 0x6,0x7,0x4,0x5, 0xa,0xb,0x8,0x9, 0xe,0xf,0xc,0xd
+.Lrot24:
+.byte	0x3,0x0,0x1,0x2, 0x7,0x4,0x5,0x6, 0xb,0x8,0x9,0xa, 0xf,0xc,0xd,0xe
+.Lsigma:
+.byte	101,120,112,97,110,100,32,51,50,45,98,121,116,101,32,107,0
+.byte	67,104,97,67,104,97,50,48,32,102,111,114,32,120,56,54,95,54,52,44,32,67,82,89,80,84,79,71,65,77,83,32,98,121,32,60,97,112,112,114,111,64,111,112,101,110,115,115,108,46,111,114,103,62,0
+.globl	ChaCha20_ctr32
+.hidden ChaCha20_ctr32
+.type	ChaCha20_ctr32,@function
+.align	64
+ChaCha20_ctr32:
+	cmpq	$0,%rdx
+	je	.Lno_data
+	movq	OPENSSL_ia32cap_P+4(%rip),%r10
+	testl	$512,%r10d
+	jnz	.LChaCha20_ssse3
+
+	pushq	%rbx
+	pushq	%rbp
+	pushq	%r12
+	pushq	%r13
+	pushq	%r14
+	pushq	%r15
+	subq	$64+24,%rsp
+
+
+	movdqu	(%rcx),%xmm1
+	movdqu	16(%rcx),%xmm2
+	movdqu	(%r8),%xmm3
+	movdqa	.Lone(%rip),%xmm4
+
+
+	movdqa	%xmm1,16(%rsp)
+	movdqa	%xmm2,32(%rsp)
+	movdqa	%xmm3,48(%rsp)
+	movq	%rdx,%rbp
+	jmp	.Loop_outer
+
+.align	32
+.Loop_outer:
+	movl	$0x61707865,%eax
+	movl	$0x3320646e,%ebx
+	movl	$0x79622d32,%ecx
+	movl	$0x6b206574,%edx
+	movl	16(%rsp),%r8d
+	movl	20(%rsp),%r9d
+	movl	24(%rsp),%r10d
+	movl	28(%rsp),%r11d
+	movd	%xmm3,%r12d
+	movl	52(%rsp),%r13d
+	movl	56(%rsp),%r14d
+	movl	60(%rsp),%r15d
+
+	movq	%rbp,64+0(%rsp)
+	movl	$10,%ebp
+	movq	%rsi,64+8(%rsp)
+.byte	102,72,15,126,214
+	movq	%rdi,64+16(%rsp)
+	movq	%rsi,%rdi
+	shrq	$32,%rdi
+	jmp	.Loop
+
+.align	32
+.Loop:
+	addl	%r8d,%eax
+	xorl	%eax,%r12d
+	roll	$16,%r12d
+	addl	%r9d,%ebx
+	xorl	%ebx,%r13d
+	roll	$16,%r13d
+	addl	%r12d,%esi
+	xorl	%esi,%r8d
+	roll	$12,%r8d
+	addl	%r13d,%edi
+	xorl	%edi,%r9d
+	roll	$12,%r9d
+	addl	%r8d,%eax
+	xorl	%eax,%r12d
+	roll	$8,%r12d
+	addl	%r9d,%ebx
+	xorl	%ebx,%r13d
+	roll	$8,%r13d
+	addl	%r12d,%esi
+	xorl	%esi,%r8d
+	roll	$7,%r8d
+	addl	%r13d,%edi
+	xorl	%edi,%r9d
+	roll	$7,%r9d
+	movl	%esi,32(%rsp)
+	movl	%edi,36(%rsp)
+	movl	40(%rsp),%esi
+	movl	44(%rsp),%edi
+	addl	%r10d,%ecx
+	xorl	%ecx,%r14d
+	roll	$16,%r14d
+	addl	%r11d,%edx
+	xorl	%edx,%r15d
+	roll	$16,%r15d
+	addl	%r14d,%esi
+	xorl	%esi,%r10d
+	roll	$12,%r10d
+	addl	%r15d,%edi
+	xorl	%edi,%r11d
+	roll	$12,%r11d
+	addl	%r10d,%ecx
+	xorl	%ecx,%r14d
+	roll	$8,%r14d
+	addl	%r11d,%edx
+	xorl	%edx,%r15d
+	roll	$8,%r15d
+	addl	%r14d,%esi
+	xorl	%esi,%r10d
+	roll	$7,%r10d
+	addl	%r15d,%edi
+	xorl	%edi,%r11d
+	roll	$7,%r11d
+	addl	%r9d,%eax
+	xorl	%eax,%r15d
+	roll	$16,%r15d
+	addl	%r10d,%ebx
+	xorl	%ebx,%r12d
+	roll	$16,%r12d
+	addl	%r15d,%esi
+	xorl	%esi,%r9d
+	roll	$12,%r9d
+	addl	%r12d,%edi
+	xorl	%edi,%r10d
+	roll	$12,%r10d
+	addl	%r9d,%eax
+	xorl	%eax,%r15d
+	roll	$8,%r15d
+	addl	%r10d,%ebx
+	xorl	%ebx,%r12d
+	roll	$8,%r12d
+	addl	%r15d,%esi
+	xorl	%esi,%r9d
+	roll	$7,%r9d
+	addl	%r12d,%edi
+	xorl	%edi,%r10d
+	roll	$7,%r10d
+	movl	%esi,40(%rsp)
+	movl	%edi,44(%rsp)
+	movl	32(%rsp),%esi
+	movl	36(%rsp),%edi
+	addl	%r11d,%ecx
+	xorl	%ecx,%r13d
+	roll	$16,%r13d
+	addl	%r8d,%edx
+	xorl	%edx,%r14d
+	roll	$16,%r14d
+	addl	%r13d,%esi
+	xorl	%esi,%r11d
+	roll	$12,%r11d
+	addl	%r14d,%edi
+	xorl	%edi,%r8d
+	roll	$12,%r8d
+	addl	%r11d,%ecx
+	xorl	%ecx,%r13d
+	roll	$8,%r13d
+	addl	%r8d,%edx
+	xorl	%edx,%r14d
+	roll	$8,%r14d
+	addl	%r13d,%esi
+	xorl	%esi,%r11d
+	roll	$7,%r11d
+	addl	%r14d,%edi
+	xorl	%edi,%r8d
+	roll	$7,%r8d
+	decl	%ebp
+	jnz	.Loop
+	movl	%edi,36(%rsp)
+	movl	%esi,32(%rsp)
+	movq	64(%rsp),%rbp
+	movdqa	%xmm2,%xmm1
+	movq	64+8(%rsp),%rsi
+	paddd	%xmm4,%xmm3
+	movq	64+16(%rsp),%rdi
+
+	addl	$0x61707865,%eax
+	addl	$0x3320646e,%ebx
+	addl	$0x79622d32,%ecx
+	addl	$0x6b206574,%edx
+	addl	16(%rsp),%r8d
+	addl	20(%rsp),%r9d
+	addl	24(%rsp),%r10d
+	addl	28(%rsp),%r11d
+	addl	48(%rsp),%r12d
+	addl	52(%rsp),%r13d
+	addl	56(%rsp),%r14d
+	addl	60(%rsp),%r15d
+	paddd	32(%rsp),%xmm1
+
+	cmpq	$64,%rbp
+	jb	.Ltail
+
+	xorl	0(%rsi),%eax
+	xorl	4(%rsi),%ebx
+	xorl	8(%rsi),%ecx
+	xorl	12(%rsi),%edx
+	xorl	16(%rsi),%r8d
+	xorl	20(%rsi),%r9d
+	xorl	24(%rsi),%r10d
+	xorl	28(%rsi),%r11d
+	movdqu	32(%rsi),%xmm0
+	xorl	48(%rsi),%r12d
+	xorl	52(%rsi),%r13d
+	xorl	56(%rsi),%r14d
+	xorl	60(%rsi),%r15d
+	leaq	64(%rsi),%rsi
+	pxor	%xmm1,%xmm0
+
+	movdqa	%xmm2,32(%rsp)
+	movd	%xmm3,48(%rsp)
+
+	movl	%eax,0(%rdi)
+	movl	%ebx,4(%rdi)
+	movl	%ecx,8(%rdi)
+	movl	%edx,12(%rdi)
+	movl	%r8d,16(%rdi)
+	movl	%r9d,20(%rdi)
+	movl	%r10d,24(%rdi)
+	movl	%r11d,28(%rdi)
+	movdqu	%xmm0,32(%rdi)
+	movl	%r12d,48(%rdi)
+	movl	%r13d,52(%rdi)
+	movl	%r14d,56(%rdi)
+	movl	%r15d,60(%rdi)
+	leaq	64(%rdi),%rdi
+
+	subq	$64,%rbp
+	jnz	.Loop_outer
+
+	jmp	.Ldone
+
+.align	16
+.Ltail:
+	movl	%eax,0(%rsp)
+	movl	%ebx,4(%rsp)
+	xorq	%rbx,%rbx
+	movl	%ecx,8(%rsp)
+	movl	%edx,12(%rsp)
+	movl	%r8d,16(%rsp)
+	movl	%r9d,20(%rsp)
+	movl	%r10d,24(%rsp)
+	movl	%r11d,28(%rsp)
+	movdqa	%xmm1,32(%rsp)
+	movl	%r12d,48(%rsp)
+	movl	%r13d,52(%rsp)
+	movl	%r14d,56(%rsp)
+	movl	%r15d,60(%rsp)
+
+.Loop_tail:
+	movzbl	(%rsi,%rbx,1),%eax
+	movzbl	(%rsp,%rbx,1),%edx
+	leaq	1(%rbx),%rbx
+	xorl	%edx,%eax
+	movb	%al,-1(%rdi,%rbx,1)
+	decq	%rbp
+	jnz	.Loop_tail
+
+.Ldone:
+	addq	$64+24,%rsp
+	popq	%r15
+	popq	%r14
+	popq	%r13
+	popq	%r12
+	popq	%rbp
+	popq	%rbx
+.Lno_data:
+	.byte	0xf3,0xc3
+.size	ChaCha20_ctr32,.-ChaCha20_ctr32
+.type	ChaCha20_ssse3,@function
+.align	32
+ChaCha20_ssse3:
+.LChaCha20_ssse3:
+	cmpq	$128,%rdx
+	ja	.LChaCha20_4x
+
+.Ldo_sse3_after_all:
+	pushq	%rbx
+	pushq	%rbp
+	pushq	%r12
+	pushq	%r13
+	pushq	%r14
+	pushq	%r15
+
+	subq	$64+24,%rsp
+	movdqa	.Lsigma(%rip),%xmm0
+	movdqu	(%rcx),%xmm1
+	movdqu	16(%rcx),%xmm2
+	movdqu	(%r8),%xmm3
+	movdqa	.Lrot16(%rip),%xmm6
+	movdqa	.Lrot24(%rip),%xmm7
+
+	movdqa	%xmm0,0(%rsp)
+	movdqa	%xmm1,16(%rsp)
+	movdqa	%xmm2,32(%rsp)
+	movdqa	%xmm3,48(%rsp)
+	movl	$10,%ebp
+	jmp	.Loop_ssse3
+
+.align	32
+.Loop_outer_ssse3:
+	movdqa	.Lone(%rip),%xmm3
+	movdqa	0(%rsp),%xmm0
+	movdqa	16(%rsp),%xmm1
+	movdqa	32(%rsp),%xmm2
+	paddd	48(%rsp),%xmm3
+	movl	$10,%ebp
+	movdqa	%xmm3,48(%rsp)
+	jmp	.Loop_ssse3
+
+.align	32
+.Loop_ssse3:
+	paddd	%xmm1,%xmm0
+	pxor	%xmm0,%xmm3
+.byte	102,15,56,0,222
+	paddd	%xmm3,%xmm2
+	pxor	%xmm2,%xmm1
+	movdqa	%xmm1,%xmm4
+	psrld	$20,%xmm1
+	pslld	$12,%xmm4
+	por	%xmm4,%xmm1
+	paddd	%xmm1,%xmm0
+	pxor	%xmm0,%xmm3
+.byte	102,15,56,0,223
+	paddd	%xmm3,%xmm2
+	pxor	%xmm2,%xmm1
+	movdqa	%xmm1,%xmm4
+	psrld	$25,%xmm1
+	pslld	$7,%xmm4
+	por	%xmm4,%xmm1
+	pshufd	$78,%xmm2,%xmm2
+	pshufd	$57,%xmm1,%xmm1
+	pshufd	$147,%xmm3,%xmm3
+	nop
+	paddd	%xmm1,%xmm0
+	pxor	%xmm0,%xmm3
+.byte	102,15,56,0,222
+	paddd	%xmm3,%xmm2
+	pxor	%xmm2,%xmm1
+	movdqa	%xmm1,%xmm4
+	psrld	$20,%xmm1
+	pslld	$12,%xmm4
+	por	%xmm4,%xmm1
+	paddd	%xmm1,%xmm0
+	pxor	%xmm0,%xmm3
+.byte	102,15,56,0,223
+	paddd	%xmm3,%xmm2
+	pxor	%xmm2,%xmm1
+	movdqa	%xmm1,%xmm4
+	psrld	$25,%xmm1
+	pslld	$7,%xmm4
+	por	%xmm4,%xmm1
+	pshufd	$78,%xmm2,%xmm2
+	pshufd	$147,%xmm1,%xmm1
+	pshufd	$57,%xmm3,%xmm3
+	decl	%ebp
+	jnz	.Loop_ssse3
+	paddd	0(%rsp),%xmm0
+	paddd	16(%rsp),%xmm1
+	paddd	32(%rsp),%xmm2
+	paddd	48(%rsp),%xmm3
+
+	cmpq	$64,%rdx
+	jb	.Ltail_ssse3
+
+	movdqu	0(%rsi),%xmm4
+	movdqu	16(%rsi),%xmm5
+	pxor	%xmm4,%xmm0
+	movdqu	32(%rsi),%xmm4
+	pxor	%xmm5,%xmm1
+	movdqu	48(%rsi),%xmm5
+	leaq	64(%rsi),%rsi
+	pxor	%xmm4,%xmm2
+	pxor	%xmm5,%xmm3
+
+	movdqu	%xmm0,0(%rdi)
+	movdqu	%xmm1,16(%rdi)
+	movdqu	%xmm2,32(%rdi)
+	movdqu	%xmm3,48(%rdi)
+	leaq	64(%rdi),%rdi
+
+	subq	$64,%rdx
+	jnz	.Loop_outer_ssse3
+
+	jmp	.Ldone_ssse3
+
+.align	16
+.Ltail_ssse3:
+	movdqa	%xmm0,0(%rsp)
+	movdqa	%xmm1,16(%rsp)
+	movdqa	%xmm2,32(%rsp)
+	movdqa	%xmm3,48(%rsp)
+	xorq	%rbx,%rbx
+
+.Loop_tail_ssse3:
+	movzbl	(%rsi,%rbx,1),%eax
+	movzbl	(%rsp,%rbx,1),%ecx
+	leaq	1(%rbx),%rbx
+	xorl	%ecx,%eax
+	movb	%al,-1(%rdi,%rbx,1)
+	decq	%rdx
+	jnz	.Loop_tail_ssse3
+
+.Ldone_ssse3:
+	addq	$64+24,%rsp
+	popq	%r15
+	popq	%r14
+	popq	%r13
+	popq	%r12
+	popq	%rbp
+	popq	%rbx
+	.byte	0xf3,0xc3
+.size	ChaCha20_ssse3,.-ChaCha20_ssse3
+.type	ChaCha20_4x,@function
+.align	32
+ChaCha20_4x:
+.LChaCha20_4x:
+	movq	%r10,%r11
+	shrq	$32,%r10
+	testq	$32,%r10
+	jnz	.LChaCha20_8x
+	cmpq	$192,%rdx
+	ja	.Lproceed4x
+
+	andq	$71303168,%r11
+	cmpq	$4194304,%r11
+	je	.Ldo_sse3_after_all
+
+.Lproceed4x:
+	leaq	-120(%rsp),%r11
+	subq	$0x148+0,%rsp
+	movdqa	.Lsigma(%rip),%xmm11
+	movdqu	(%rcx),%xmm15
+	movdqu	16(%rcx),%xmm7
+	movdqu	(%r8),%xmm3
+	leaq	256(%rsp),%rcx
+	leaq	.Lrot16(%rip),%r10
+	leaq	.Lrot24(%rip),%r11
+
+	pshufd	$0x00,%xmm11,%xmm8
+	pshufd	$0x55,%xmm11,%xmm9
+	movdqa	%xmm8,64(%rsp)
+	pshufd	$0xaa,%xmm11,%xmm10
+	movdqa	%xmm9,80(%rsp)
+	pshufd	$0xff,%xmm11,%xmm11
+	movdqa	%xmm10,96(%rsp)
+	movdqa	%xmm11,112(%rsp)
+
+	pshufd	$0x00,%xmm15,%xmm12
+	pshufd	$0x55,%xmm15,%xmm13
+	movdqa	%xmm12,128-256(%rcx)
+	pshufd	$0xaa,%xmm15,%xmm14
+	movdqa	%xmm13,144-256(%rcx)
+	pshufd	$0xff,%xmm15,%xmm15
+	movdqa	%xmm14,160-256(%rcx)
+	movdqa	%xmm15,176-256(%rcx)
+
+	pshufd	$0x00,%xmm7,%xmm4
+	pshufd	$0x55,%xmm7,%xmm5
+	movdqa	%xmm4,192-256(%rcx)
+	pshufd	$0xaa,%xmm7,%xmm6
+	movdqa	%xmm5,208-256(%rcx)
+	pshufd	$0xff,%xmm7,%xmm7
+	movdqa	%xmm6,224-256(%rcx)
+	movdqa	%xmm7,240-256(%rcx)
+
+	pshufd	$0x00,%xmm3,%xmm0
+	pshufd	$0x55,%xmm3,%xmm1
+	paddd	.Linc(%rip),%xmm0
+	pshufd	$0xaa,%xmm3,%xmm2
+	movdqa	%xmm1,272-256(%rcx)
+	pshufd	$0xff,%xmm3,%xmm3
+	movdqa	%xmm2,288-256(%rcx)
+	movdqa	%xmm3,304-256(%rcx)
+
+	jmp	.Loop_enter4x
+
+.align	32
+.Loop_outer4x:
+	movdqa	64(%rsp),%xmm8
+	movdqa	80(%rsp),%xmm9
+	movdqa	96(%rsp),%xmm10
+	movdqa	112(%rsp),%xmm11
+	movdqa	128-256(%rcx),%xmm12
+	movdqa	144-256(%rcx),%xmm13
+	movdqa	160-256(%rcx),%xmm14
+	movdqa	176-256(%rcx),%xmm15
+	movdqa	192-256(%rcx),%xmm4
+	movdqa	208-256(%rcx),%xmm5
+	movdqa	224-256(%rcx),%xmm6
+	movdqa	240-256(%rcx),%xmm7
+	movdqa	256-256(%rcx),%xmm0
+	movdqa	272-256(%rcx),%xmm1
+	movdqa	288-256(%rcx),%xmm2
+	movdqa	304-256(%rcx),%xmm3
+	paddd	.Lfour(%rip),%xmm0
+
+.Loop_enter4x:
+	movdqa	%xmm6,32(%rsp)
+	movdqa	%xmm7,48(%rsp)
+	movdqa	(%r10),%xmm7
+	movl	$10,%eax
+	movdqa	%xmm0,256-256(%rcx)
+	jmp	.Loop4x
+
+.align	32
+.Loop4x:
+	paddd	%xmm12,%xmm8
+	paddd	%xmm13,%xmm9
+	pxor	%xmm8,%xmm0
+	pxor	%xmm9,%xmm1
+.byte	102,15,56,0,199
+.byte	102,15,56,0,207
+	paddd	%xmm0,%xmm4
+	paddd	%xmm1,%xmm5
+	pxor	%xmm4,%xmm12
+	pxor	%xmm5,%xmm13
+	movdqa	%xmm12,%xmm6
+	pslld	$12,%xmm12
+	psrld	$20,%xmm6
+	movdqa	%xmm13,%xmm7
+	pslld	$12,%xmm13
+	por	%xmm6,%xmm12
+	psrld	$20,%xmm7
+	movdqa	(%r11),%xmm6
+	por	%xmm7,%xmm13
+	paddd	%xmm12,%xmm8
+	paddd	%xmm13,%xmm9
+	pxor	%xmm8,%xmm0
+	pxor	%xmm9,%xmm1
+.byte	102,15,56,0,198
+.byte	102,15,56,0,206
+	paddd	%xmm0,%xmm4
+	paddd	%xmm1,%xmm5
+	pxor	%xmm4,%xmm12
+	pxor	%xmm5,%xmm13
+	movdqa	%xmm12,%xmm7
+	pslld	$7,%xmm12
+	psrld	$25,%xmm7
+	movdqa	%xmm13,%xmm6
+	pslld	$7,%xmm13
+	por	%xmm7,%xmm12
+	psrld	$25,%xmm6
+	movdqa	(%r10),%xmm7
+	por	%xmm6,%xmm13
+	movdqa	%xmm4,0(%rsp)
+	movdqa	%xmm5,16(%rsp)
+	movdqa	32(%rsp),%xmm4
+	movdqa	48(%rsp),%xmm5
+	paddd	%xmm14,%xmm10
+	paddd	%xmm15,%xmm11
+	pxor	%xmm10,%xmm2
+	pxor	%xmm11,%xmm3
+.byte	102,15,56,0,215
+.byte	102,15,56,0,223
+	paddd	%xmm2,%xmm4
+	paddd	%xmm3,%xmm5
+	pxor	%xmm4,%xmm14
+	pxor	%xmm5,%xmm15
+	movdqa	%xmm14,%xmm6
+	pslld	$12,%xmm14
+	psrld	$20,%xmm6
+	movdqa	%xmm15,%xmm7
+	pslld	$12,%xmm15
+	por	%xmm6,%xmm14
+	psrld	$20,%xmm7
+	movdqa	(%r11),%xmm6
+	por	%xmm7,%xmm15
+	paddd	%xmm14,%xmm10
+	paddd	%xmm15,%xmm11
+	pxor	%xmm10,%xmm2
+	pxor	%xmm11,%xmm3
+.byte	102,15,56,0,214
+.byte	102,15,56,0,222
+	paddd	%xmm2,%xmm4
+	paddd	%xmm3,%xmm5
+	pxor	%xmm4,%xmm14
+	pxor	%xmm5,%xmm15
+	movdqa	%xmm14,%xmm7
+	pslld	$7,%xmm14
+	psrld	$25,%xmm7
+	movdqa	%xmm15,%xmm6
+	pslld	$7,%xmm15
+	por	%xmm7,%xmm14
+	psrld	$25,%xmm6
+	movdqa	(%r10),%xmm7
+	por	%xmm6,%xmm15
+	paddd	%xmm13,%xmm8
+	paddd	%xmm14,%xmm9
+	pxor	%xmm8,%xmm3
+	pxor	%xmm9,%xmm0
+.byte	102,15,56,0,223
+.byte	102,15,56,0,199
+	paddd	%xmm3,%xmm4
+	paddd	%xmm0,%xmm5
+	pxor	%xmm4,%xmm13
+	pxor	%xmm5,%xmm14
+	movdqa	%xmm13,%xmm6
+	pslld	$12,%xmm13
+	psrld	$20,%xmm6
+	movdqa	%xmm14,%xmm7
+	pslld	$12,%xmm14
+	por	%xmm6,%xmm13
+	psrld	$20,%xmm7
+	movdqa	(%r11),%xmm6
+	por	%xmm7,%xmm14
+	paddd	%xmm13,%xmm8
+	paddd	%xmm14,%xmm9
+	pxor	%xmm8,%xmm3
+	pxor	%xmm9,%xmm0
+.byte	102,15,56,0,222
+.byte	102,15,56,0,198
+	paddd	%xmm3,%xmm4
+	paddd	%xmm0,%xmm5
+	pxor	%xmm4,%xmm13
+	pxor	%xmm5,%xmm14
+	movdqa	%xmm13,%xmm7
+	pslld	$7,%xmm13
+	psrld	$25,%xmm7
+	movdqa	%xmm14,%xmm6
+	pslld	$7,%xmm14
+	por	%xmm7,%xmm13
+	psrld	$25,%xmm6
+	movdqa	(%r10),%xmm7
+	por	%xmm6,%xmm14
+	movdqa	%xmm4,32(%rsp)
+	movdqa	%xmm5,48(%rsp)
+	movdqa	0(%rsp),%xmm4
+	movdqa	16(%rsp),%xmm5
+	paddd	%xmm15,%xmm10
+	paddd	%xmm12,%xmm11
+	pxor	%xmm10,%xmm1
+	pxor	%xmm11,%xmm2
+.byte	102,15,56,0,207
+.byte	102,15,56,0,215
+	paddd	%xmm1,%xmm4
+	paddd	%xmm2,%xmm5
+	pxor	%xmm4,%xmm15
+	pxor	%xmm5,%xmm12
+	movdqa	%xmm15,%xmm6
+	pslld	$12,%xmm15
+	psrld	$20,%xmm6
+	movdqa	%xmm12,%xmm7
+	pslld	$12,%xmm12
+	por	%xmm6,%xmm15
+	psrld	$20,%xmm7
+	movdqa	(%r11),%xmm6
+	por	%xmm7,%xmm12
+	paddd	%xmm15,%xmm10
+	paddd	%xmm12,%xmm11
+	pxor	%xmm10,%xmm1
+	pxor	%xmm11,%xmm2
+.byte	102,15,56,0,206
+.byte	102,15,56,0,214
+	paddd	%xmm1,%xmm4
+	paddd	%xmm2,%xmm5
+	pxor	%xmm4,%xmm15
+	pxor	%xmm5,%xmm12
+	movdqa	%xmm15,%xmm7
+	pslld	$7,%xmm15
+	psrld	$25,%xmm7
+	movdqa	%xmm12,%xmm6
+	pslld	$7,%xmm12
+	por	%xmm7,%xmm15
+	psrld	$25,%xmm6
+	movdqa	(%r10),%xmm7
+	por	%xmm6,%xmm12
+	decl	%eax
+	jnz	.Loop4x
+
+	paddd	64(%rsp),%xmm8
+	paddd	80(%rsp),%xmm9
+	paddd	96(%rsp),%xmm10
+	paddd	112(%rsp),%xmm11
+
+	movdqa	%xmm8,%xmm6
+	punpckldq	%xmm9,%xmm8
+	movdqa	%xmm10,%xmm7
+	punpckldq	%xmm11,%xmm10
+	punpckhdq	%xmm9,%xmm6
+	punpckhdq	%xmm11,%xmm7
+	movdqa	%xmm8,%xmm9
+	punpcklqdq	%xmm10,%xmm8
+	movdqa	%xmm6,%xmm11
+	punpcklqdq	%xmm7,%xmm6
+	punpckhqdq	%xmm10,%xmm9
+	punpckhqdq	%xmm7,%xmm11
+	paddd	128-256(%rcx),%xmm12
+	paddd	144-256(%rcx),%xmm13
+	paddd	160-256(%rcx),%xmm14
+	paddd	176-256(%rcx),%xmm15
+
+	movdqa	%xmm8,0(%rsp)
+	movdqa	%xmm9,16(%rsp)
+	movdqa	32(%rsp),%xmm8
+	movdqa	48(%rsp),%xmm9
+
+	movdqa	%xmm12,%xmm10
+	punpckldq	%xmm13,%xmm12
+	movdqa	%xmm14,%xmm7
+	punpckldq	%xmm15,%xmm14
+	punpckhdq	%xmm13,%xmm10
+	punpckhdq	%xmm15,%xmm7
+	movdqa	%xmm12,%xmm13
+	punpcklqdq	%xmm14,%xmm12
+	movdqa	%xmm10,%xmm15
+	punpcklqdq	%xmm7,%xmm10
+	punpckhqdq	%xmm14,%xmm13
+	punpckhqdq	%xmm7,%xmm15
+	paddd	192-256(%rcx),%xmm4
+	paddd	208-256(%rcx),%xmm5
+	paddd	224-256(%rcx),%xmm8
+	paddd	240-256(%rcx),%xmm9
+
+	movdqa	%xmm6,32(%rsp)
+	movdqa	%xmm11,48(%rsp)
+
+	movdqa	%xmm4,%xmm14
+	punpckldq	%xmm5,%xmm4
+	movdqa	%xmm8,%xmm7
+	punpckldq	%xmm9,%xmm8
+	punpckhdq	%xmm5,%xmm14
+	punpckhdq	%xmm9,%xmm7
+	movdqa	%xmm4,%xmm5
+	punpcklqdq	%xmm8,%xmm4
+	movdqa	%xmm14,%xmm9
+	punpcklqdq	%xmm7,%xmm14
+	punpckhqdq	%xmm8,%xmm5
+	punpckhqdq	%xmm7,%xmm9
+	paddd	256-256(%rcx),%xmm0
+	paddd	272-256(%rcx),%xmm1
+	paddd	288-256(%rcx),%xmm2
+	paddd	304-256(%rcx),%xmm3
+
+	movdqa	%xmm0,%xmm8
+	punpckldq	%xmm1,%xmm0
+	movdqa	%xmm2,%xmm7
+	punpckldq	%xmm3,%xmm2
+	punpckhdq	%xmm1,%xmm8
+	punpckhdq	%xmm3,%xmm7
+	movdqa	%xmm0,%xmm1
+	punpcklqdq	%xmm2,%xmm0
+	movdqa	%xmm8,%xmm3
+	punpcklqdq	%xmm7,%xmm8
+	punpckhqdq	%xmm2,%xmm1
+	punpckhqdq	%xmm7,%xmm3
+	cmpq	$256,%rdx
+	jb	.Ltail4x
+
+	movdqu	0(%rsi),%xmm6
+	movdqu	16(%rsi),%xmm11
+	movdqu	32(%rsi),%xmm2
+	movdqu	48(%rsi),%xmm7
+	pxor	0(%rsp),%xmm6
+	pxor	%xmm12,%xmm11
+	pxor	%xmm4,%xmm2
+	pxor	%xmm0,%xmm7
+
+	movdqu	%xmm6,0(%rdi)
+	movdqu	64(%rsi),%xmm6
+	movdqu	%xmm11,16(%rdi)
+	movdqu	80(%rsi),%xmm11
+	movdqu	%xmm2,32(%rdi)
+	movdqu	96(%rsi),%xmm2
+	movdqu	%xmm7,48(%rdi)
+	movdqu	112(%rsi),%xmm7
+	leaq	128(%rsi),%rsi
+	pxor	16(%rsp),%xmm6
+	pxor	%xmm13,%xmm11
+	pxor	%xmm5,%xmm2
+	pxor	%xmm1,%xmm7
+
+	movdqu	%xmm6,64(%rdi)
+	movdqu	0(%rsi),%xmm6
+	movdqu	%xmm11,80(%rdi)
+	movdqu	16(%rsi),%xmm11
+	movdqu	%xmm2,96(%rdi)
+	movdqu	32(%rsi),%xmm2
+	movdqu	%xmm7,112(%rdi)
+	leaq	128(%rdi),%rdi
+	movdqu	48(%rsi),%xmm7
+	pxor	32(%rsp),%xmm6
+	pxor	%xmm10,%xmm11
+	pxor	%xmm14,%xmm2
+	pxor	%xmm8,%xmm7
+
+	movdqu	%xmm6,0(%rdi)
+	movdqu	64(%rsi),%xmm6
+	movdqu	%xmm11,16(%rdi)
+	movdqu	80(%rsi),%xmm11
+	movdqu	%xmm2,32(%rdi)
+	movdqu	96(%rsi),%xmm2
+	movdqu	%xmm7,48(%rdi)
+	movdqu	112(%rsi),%xmm7
+	leaq	128(%rsi),%rsi
+	pxor	48(%rsp),%xmm6
+	pxor	%xmm15,%xmm11
+	pxor	%xmm9,%xmm2
+	pxor	%xmm3,%xmm7
+	movdqu	%xmm6,64(%rdi)
+	movdqu	%xmm11,80(%rdi)
+	movdqu	%xmm2,96(%rdi)
+	movdqu	%xmm7,112(%rdi)
+	leaq	128(%rdi),%rdi
+
+	subq	$256,%rdx
+	jnz	.Loop_outer4x
+
+	jmp	.Ldone4x
+
+.Ltail4x:
+	cmpq	$192,%rdx
+	jae	.L192_or_more4x
+	cmpq	$128,%rdx
+	jae	.L128_or_more4x
+	cmpq	$64,%rdx
+	jae	.L64_or_more4x
+
+
+	xorq	%r10,%r10
+
+	movdqa	%xmm12,16(%rsp)
+	movdqa	%xmm4,32(%rsp)
+	movdqa	%xmm0,48(%rsp)
+	jmp	.Loop_tail4x
+
+.align	32
+.L64_or_more4x:
+	movdqu	0(%rsi),%xmm6
+	movdqu	16(%rsi),%xmm11
+	movdqu	32(%rsi),%xmm2
+	movdqu	48(%rsi),%xmm7
+	pxor	0(%rsp),%xmm6
+	pxor	%xmm12,%xmm11
+	pxor	%xmm4,%xmm2
+	pxor	%xmm0,%xmm7
+	movdqu	%xmm6,0(%rdi)
+	movdqu	%xmm11,16(%rdi)
+	movdqu	%xmm2,32(%rdi)
+	movdqu	%xmm7,48(%rdi)
+	je	.Ldone4x
+
+	movdqa	16(%rsp),%xmm6
+	leaq	64(%rsi),%rsi
+	xorq	%r10,%r10
+	movdqa	%xmm6,0(%rsp)
+	movdqa	%xmm13,16(%rsp)
+	leaq	64(%rdi),%rdi
+	movdqa	%xmm5,32(%rsp)
+	subq	$64,%rdx
+	movdqa	%xmm1,48(%rsp)
+	jmp	.Loop_tail4x
+
+.align	32
+.L128_or_more4x:
+	movdqu	0(%rsi),%xmm6
+	movdqu	16(%rsi),%xmm11
+	movdqu	32(%rsi),%xmm2
+	movdqu	48(%rsi),%xmm7
+	pxor	0(%rsp),%xmm6
+	pxor	%xmm12,%xmm11
+	pxor	%xmm4,%xmm2
+	pxor	%xmm0,%xmm7
+
+	movdqu	%xmm6,0(%rdi)
+	movdqu	64(%rsi),%xmm6
+	movdqu	%xmm11,16(%rdi)
+	movdqu	80(%rsi),%xmm11
+	movdqu	%xmm2,32(%rdi)
+	movdqu	96(%rsi),%xmm2
+	movdqu	%xmm7,48(%rdi)
+	movdqu	112(%rsi),%xmm7
+	pxor	16(%rsp),%xmm6
+	pxor	%xmm13,%xmm11
+	pxor	%xmm5,%xmm2
+	pxor	%xmm1,%xmm7
+	movdqu	%xmm6,64(%rdi)
+	movdqu	%xmm11,80(%rdi)
+	movdqu	%xmm2,96(%rdi)
+	movdqu	%xmm7,112(%rdi)
+	je	.Ldone4x
+
+	movdqa	32(%rsp),%xmm6
+	leaq	128(%rsi),%rsi
+	xorq	%r10,%r10
+	movdqa	%xmm6,0(%rsp)
+	movdqa	%xmm10,16(%rsp)
+	leaq	128(%rdi),%rdi
+	movdqa	%xmm14,32(%rsp)
+	subq	$128,%rdx
+	movdqa	%xmm8,48(%rsp)
+	jmp	.Loop_tail4x
+
+.align	32
+.L192_or_more4x:
+	movdqu	0(%rsi),%xmm6
+	movdqu	16(%rsi),%xmm11
+	movdqu	32(%rsi),%xmm2
+	movdqu	48(%rsi),%xmm7
+	pxor	0(%rsp),%xmm6
+	pxor	%xmm12,%xmm11
+	pxor	%xmm4,%xmm2
+	pxor	%xmm0,%xmm7
+
+	movdqu	%xmm6,0(%rdi)
+	movdqu	64(%rsi),%xmm6
+	movdqu	%xmm11,16(%rdi)
+	movdqu	80(%rsi),%xmm11
+	movdqu	%xmm2,32(%rdi)
+	movdqu	96(%rsi),%xmm2
+	movdqu	%xmm7,48(%rdi)
+	movdqu	112(%rsi),%xmm7
+	leaq	128(%rsi),%rsi
+	pxor	16(%rsp),%xmm6
+	pxor	%xmm13,%xmm11
+	pxor	%xmm5,%xmm2
+	pxor	%xmm1,%xmm7
+
+	movdqu	%xmm6,64(%rdi)
+	movdqu	0(%rsi),%xmm6
+	movdqu	%xmm11,80(%rdi)
+	movdqu	16(%rsi),%xmm11
+	movdqu	%xmm2,96(%rdi)
+	movdqu	32(%rsi),%xmm2
+	movdqu	%xmm7,112(%rdi)
+	leaq	128(%rdi),%rdi
+	movdqu	48(%rsi),%xmm7
+	pxor	32(%rsp),%xmm6
+	pxor	%xmm10,%xmm11
+	pxor	%xmm14,%xmm2
+	pxor	%xmm8,%xmm7
+	movdqu	%xmm6,0(%rdi)
+	movdqu	%xmm11,16(%rdi)
+	movdqu	%xmm2,32(%rdi)
+	movdqu	%xmm7,48(%rdi)
+	je	.Ldone4x
+
+	movdqa	48(%rsp),%xmm6
+	leaq	64(%rsi),%rsi
+	xorq	%r10,%r10
+	movdqa	%xmm6,0(%rsp)
+	movdqa	%xmm15,16(%rsp)
+	leaq	64(%rdi),%rdi
+	movdqa	%xmm9,32(%rsp)
+	subq	$192,%rdx
+	movdqa	%xmm3,48(%rsp)
+
+.Loop_tail4x:
+	movzbl	(%rsi,%r10,1),%eax
+	movzbl	(%rsp,%r10,1),%ecx
+	leaq	1(%r10),%r10
+	xorl	%ecx,%eax
+	movb	%al,-1(%rdi,%r10,1)
+	decq	%rdx
+	jnz	.Loop_tail4x
+
+.Ldone4x:
+	addq	$0x148+0,%rsp
+	.byte	0xf3,0xc3
+.size	ChaCha20_4x,.-ChaCha20_4x
+.type	ChaCha20_8x,@function
+.align	32
+ChaCha20_8x:
+.LChaCha20_8x:
+	movq	%rsp,%r10
+	subq	$0x280+8,%rsp
+	andq	$-32,%rsp
+	vzeroupper
+	movq	%r10,640(%rsp)
+
+
+
+
+
+
+
+
+
+
+	vbroadcasti128	.Lsigma(%rip),%ymm11
+	vbroadcasti128	(%rcx),%ymm3
+	vbroadcasti128	16(%rcx),%ymm15
+	vbroadcasti128	(%r8),%ymm7
+	leaq	256(%rsp),%rcx
+	leaq	512(%rsp),%rax
+	leaq	.Lrot16(%rip),%r10
+	leaq	.Lrot24(%rip),%r11
+
+	vpshufd	$0x00,%ymm11,%ymm8
+	vpshufd	$0x55,%ymm11,%ymm9
+	vmovdqa	%ymm8,128-256(%rcx)
+	vpshufd	$0xaa,%ymm11,%ymm10
+	vmovdqa	%ymm9,160-256(%rcx)
+	vpshufd	$0xff,%ymm11,%ymm11
+	vmovdqa	%ymm10,192-256(%rcx)
+	vmovdqa	%ymm11,224-256(%rcx)
+
+	vpshufd	$0x00,%ymm3,%ymm0
+	vpshufd	$0x55,%ymm3,%ymm1
+	vmovdqa	%ymm0,256-256(%rcx)
+	vpshufd	$0xaa,%ymm3,%ymm2
+	vmovdqa	%ymm1,288-256(%rcx)
+	vpshufd	$0xff,%ymm3,%ymm3
+	vmovdqa	%ymm2,320-256(%rcx)
+	vmovdqa	%ymm3,352-256(%rcx)
+
+	vpshufd	$0x00,%ymm15,%ymm12
+	vpshufd	$0x55,%ymm15,%ymm13
+	vmovdqa	%ymm12,384-512(%rax)
+	vpshufd	$0xaa,%ymm15,%ymm14
+	vmovdqa	%ymm13,416-512(%rax)
+	vpshufd	$0xff,%ymm15,%ymm15
+	vmovdqa	%ymm14,448-512(%rax)
+	vmovdqa	%ymm15,480-512(%rax)
+
+	vpshufd	$0x00,%ymm7,%ymm4
+	vpshufd	$0x55,%ymm7,%ymm5
+	vpaddd	.Lincy(%rip),%ymm4,%ymm4
+	vpshufd	$0xaa,%ymm7,%ymm6
+	vmovdqa	%ymm5,544-512(%rax)
+	vpshufd	$0xff,%ymm7,%ymm7
+	vmovdqa	%ymm6,576-512(%rax)
+	vmovdqa	%ymm7,608-512(%rax)
+
+	jmp	.Loop_enter8x
+
+.align	32
+.Loop_outer8x:
+	vmovdqa	128-256(%rcx),%ymm8
+	vmovdqa	160-256(%rcx),%ymm9
+	vmovdqa	192-256(%rcx),%ymm10
+	vmovdqa	224-256(%rcx),%ymm11
+	vmovdqa	256-256(%rcx),%ymm0
+	vmovdqa	288-256(%rcx),%ymm1
+	vmovdqa	320-256(%rcx),%ymm2
+	vmovdqa	352-256(%rcx),%ymm3
+	vmovdqa	384-512(%rax),%ymm12
+	vmovdqa	416-512(%rax),%ymm13
+	vmovdqa	448-512(%rax),%ymm14
+	vmovdqa	480-512(%rax),%ymm15
+	vmovdqa	512-512(%rax),%ymm4
+	vmovdqa	544-512(%rax),%ymm5
+	vmovdqa	576-512(%rax),%ymm6
+	vmovdqa	608-512(%rax),%ymm7
+	vpaddd	.Leight(%rip),%ymm4,%ymm4
+
+.Loop_enter8x:
+	vmovdqa	%ymm14,64(%rsp)
+	vmovdqa	%ymm15,96(%rsp)
+	vbroadcasti128	(%r10),%ymm15
+	vmovdqa	%ymm4,512-512(%rax)
+	movl	$10,%eax
+	jmp	.Loop8x
+
+.align	32
+.Loop8x:
+	vpaddd	%ymm0,%ymm8,%ymm8
+	vpxor	%ymm4,%ymm8,%ymm4
+	vpshufb	%ymm15,%ymm4,%ymm4
+	vpaddd	%ymm1,%ymm9,%ymm9
+	vpxor	%ymm5,%ymm9,%ymm5
+	vpshufb	%ymm15,%ymm5,%ymm5
+	vpaddd	%ymm4,%ymm12,%ymm12
+	vpxor	%ymm0,%ymm12,%ymm0
+	vpslld	$12,%ymm0,%ymm14
+	vpsrld	$20,%ymm0,%ymm0
+	vpor	%ymm0,%ymm14,%ymm0
+	vbroadcasti128	(%r11),%ymm14
+	vpaddd	%ymm5,%ymm13,%ymm13
+	vpxor	%ymm1,%ymm13,%ymm1
+	vpslld	$12,%ymm1,%ymm15
+	vpsrld	$20,%ymm1,%ymm1
+	vpor	%ymm1,%ymm15,%ymm1
+	vpaddd	%ymm0,%ymm8,%ymm8
+	vpxor	%ymm4,%ymm8,%ymm4
+	vpshufb	%ymm14,%ymm4,%ymm4
+	vpaddd	%ymm1,%ymm9,%ymm9
+	vpxor	%ymm5,%ymm9,%ymm5
+	vpshufb	%ymm14,%ymm5,%ymm5
+	vpaddd	%ymm4,%ymm12,%ymm12
+	vpxor	%ymm0,%ymm12,%ymm0
+	vpslld	$7,%ymm0,%ymm15
+	vpsrld	$25,%ymm0,%ymm0
+	vpor	%ymm0,%ymm15,%ymm0
+	vbroadcasti128	(%r10),%ymm15
+	vpaddd	%ymm5,%ymm13,%ymm13
+	vpxor	%ymm1,%ymm13,%ymm1
+	vpslld	$7,%ymm1,%ymm14
+	vpsrld	$25,%ymm1,%ymm1
+	vpor	%ymm1,%ymm14,%ymm1
+	vmovdqa	%ymm12,0(%rsp)
+	vmovdqa	%ymm13,32(%rsp)
+	vmovdqa	64(%rsp),%ymm12
+	vmovdqa	96(%rsp),%ymm13
+	vpaddd	%ymm2,%ymm10,%ymm10
+	vpxor	%ymm6,%ymm10,%ymm6
+	vpshufb	%ymm15,%ymm6,%ymm6
+	vpaddd	%ymm3,%ymm11,%ymm11
+	vpxor	%ymm7,%ymm11,%ymm7
+	vpshufb	%ymm15,%ymm7,%ymm7
+	vpaddd	%ymm6,%ymm12,%ymm12
+	vpxor	%ymm2,%ymm12,%ymm2
+	vpslld	$12,%ymm2,%ymm14
+	vpsrld	$20,%ymm2,%ymm2
+	vpor	%ymm2,%ymm14,%ymm2
+	vbroadcasti128	(%r11),%ymm14
+	vpaddd	%ymm7,%ymm13,%ymm13
+	vpxor	%ymm3,%ymm13,%ymm3
+	vpslld	$12,%ymm3,%ymm15
+	vpsrld	$20,%ymm3,%ymm3
+	vpor	%ymm3,%ymm15,%ymm3
+	vpaddd	%ymm2,%ymm10,%ymm10
+	vpxor	%ymm6,%ymm10,%ymm6
+	vpshufb	%ymm14,%ymm6,%ymm6
+	vpaddd	%ymm3,%ymm11,%ymm11
+	vpxor	%ymm7,%ymm11,%ymm7
+	vpshufb	%ymm14,%ymm7,%ymm7
+	vpaddd	%ymm6,%ymm12,%ymm12
+	vpxor	%ymm2,%ymm12,%ymm2
+	vpslld	$7,%ymm2,%ymm15
+	vpsrld	$25,%ymm2,%ymm2
+	vpor	%ymm2,%ymm15,%ymm2
+	vbroadcasti128	(%r10),%ymm15
+	vpaddd	%ymm7,%ymm13,%ymm13
+	vpxor	%ymm3,%ymm13,%ymm3
+	vpslld	$7,%ymm3,%ymm14
+	vpsrld	$25,%ymm3,%ymm3
+	vpor	%ymm3,%ymm14,%ymm3
+	vpaddd	%ymm1,%ymm8,%ymm8
+	vpxor	%ymm7,%ymm8,%ymm7
+	vpshufb	%ymm15,%ymm7,%ymm7
+	vpaddd	%ymm2,%ymm9,%ymm9
+	vpxor	%ymm4,%ymm9,%ymm4
+	vpshufb	%ymm15,%ymm4,%ymm4
+	vpaddd	%ymm7,%ymm12,%ymm12
+	vpxor	%ymm1,%ymm12,%ymm1
+	vpslld	$12,%ymm1,%ymm14
+	vpsrld	$20,%ymm1,%ymm1
+	vpor	%ymm1,%ymm14,%ymm1
+	vbroadcasti128	(%r11),%ymm14
+	vpaddd	%ymm4,%ymm13,%ymm13
+	vpxor	%ymm2,%ymm13,%ymm2
+	vpslld	$12,%ymm2,%ymm15
+	vpsrld	$20,%ymm2,%ymm2
+	vpor	%ymm2,%ymm15,%ymm2
+	vpaddd	%ymm1,%ymm8,%ymm8
+	vpxor	%ymm7,%ymm8,%ymm7
+	vpshufb	%ymm14,%ymm7,%ymm7
+	vpaddd	%ymm2,%ymm9,%ymm9
+	vpxor	%ymm4,%ymm9,%ymm4
+	vpshufb	%ymm14,%ymm4,%ymm4
+	vpaddd	%ymm7,%ymm12,%ymm12
+	vpxor	%ymm1,%ymm12,%ymm1
+	vpslld	$7,%ymm1,%ymm15
+	vpsrld	$25,%ymm1,%ymm1
+	vpor	%ymm1,%ymm15,%ymm1
+	vbroadcasti128	(%r10),%ymm15
+	vpaddd	%ymm4,%ymm13,%ymm13
+	vpxor	%ymm2,%ymm13,%ymm2
+	vpslld	$7,%ymm2,%ymm14
+	vpsrld	$25,%ymm2,%ymm2
+	vpor	%ymm2,%ymm14,%ymm2
+	vmovdqa	%ymm12,64(%rsp)
+	vmovdqa	%ymm13,96(%rsp)
+	vmovdqa	0(%rsp),%ymm12
+	vmovdqa	32(%rsp),%ymm13
+	vpaddd	%ymm3,%ymm10,%ymm10
+	vpxor	%ymm5,%ymm10,%ymm5
+	vpshufb	%ymm15,%ymm5,%ymm5
+	vpaddd	%ymm0,%ymm11,%ymm11
+	vpxor	%ymm6,%ymm11,%ymm6
+	vpshufb	%ymm15,%ymm6,%ymm6
+	vpaddd	%ymm5,%ymm12,%ymm12
+	vpxor	%ymm3,%ymm12,%ymm3
+	vpslld	$12,%ymm3,%ymm14
+	vpsrld	$20,%ymm3,%ymm3
+	vpor	%ymm3,%ymm14,%ymm3
+	vbroadcasti128	(%r11),%ymm14
+	vpaddd	%ymm6,%ymm13,%ymm13
+	vpxor	%ymm0,%ymm13,%ymm0
+	vpslld	$12,%ymm0,%ymm15
+	vpsrld	$20,%ymm0,%ymm0
+	vpor	%ymm0,%ymm15,%ymm0
+	vpaddd	%ymm3,%ymm10,%ymm10
+	vpxor	%ymm5,%ymm10,%ymm5
+	vpshufb	%ymm14,%ymm5,%ymm5
+	vpaddd	%ymm0,%ymm11,%ymm11
+	vpxor	%ymm6,%ymm11,%ymm6
+	vpshufb	%ymm14,%ymm6,%ymm6
+	vpaddd	%ymm5,%ymm12,%ymm12
+	vpxor	%ymm3,%ymm12,%ymm3
+	vpslld	$7,%ymm3,%ymm15
+	vpsrld	$25,%ymm3,%ymm3
+	vpor	%ymm3,%ymm15,%ymm3
+	vbroadcasti128	(%r10),%ymm15
+	vpaddd	%ymm6,%ymm13,%ymm13
+	vpxor	%ymm0,%ymm13,%ymm0
+	vpslld	$7,%ymm0,%ymm14
+	vpsrld	$25,%ymm0,%ymm0
+	vpor	%ymm0,%ymm14,%ymm0
+	decl	%eax
+	jnz	.Loop8x
+
+	leaq	512(%rsp),%rax
+	vpaddd	128-256(%rcx),%ymm8,%ymm8
+	vpaddd	160-256(%rcx),%ymm9,%ymm9
+	vpaddd	192-256(%rcx),%ymm10,%ymm10
+	vpaddd	224-256(%rcx),%ymm11,%ymm11
+
+	vpunpckldq	%ymm9,%ymm8,%ymm14
+	vpunpckldq	%ymm11,%ymm10,%ymm15
+	vpunpckhdq	%ymm9,%ymm8,%ymm8
+	vpunpckhdq	%ymm11,%ymm10,%ymm10
+	vpunpcklqdq	%ymm15,%ymm14,%ymm9
+	vpunpckhqdq	%ymm15,%ymm14,%ymm14
+	vpunpcklqdq	%ymm10,%ymm8,%ymm11
+	vpunpckhqdq	%ymm10,%ymm8,%ymm8
+	vpaddd	256-256(%rcx),%ymm0,%ymm0
+	vpaddd	288-256(%rcx),%ymm1,%ymm1
+	vpaddd	320-256(%rcx),%ymm2,%ymm2
+	vpaddd	352-256(%rcx),%ymm3,%ymm3
+
+	vpunpckldq	%ymm1,%ymm0,%ymm10
+	vpunpckldq	%ymm3,%ymm2,%ymm15
+	vpunpckhdq	%ymm1,%ymm0,%ymm0
+	vpunpckhdq	%ymm3,%ymm2,%ymm2
+	vpunpcklqdq	%ymm15,%ymm10,%ymm1
+	vpunpckhqdq	%ymm15,%ymm10,%ymm10
+	vpunpcklqdq	%ymm2,%ymm0,%ymm3
+	vpunpckhqdq	%ymm2,%ymm0,%ymm0
+	vperm2i128	$0x20,%ymm1,%ymm9,%ymm15
+	vperm2i128	$0x31,%ymm1,%ymm9,%ymm1
+	vperm2i128	$0x20,%ymm10,%ymm14,%ymm9
+	vperm2i128	$0x31,%ymm10,%ymm14,%ymm10
+	vperm2i128	$0x20,%ymm3,%ymm11,%ymm14
+	vperm2i128	$0x31,%ymm3,%ymm11,%ymm3
+	vperm2i128	$0x20,%ymm0,%ymm8,%ymm11
+	vperm2i128	$0x31,%ymm0,%ymm8,%ymm0
+	vmovdqa	%ymm15,0(%rsp)
+	vmovdqa	%ymm9,32(%rsp)
+	vmovdqa	64(%rsp),%ymm15
+	vmovdqa	96(%rsp),%ymm9
+
+	vpaddd	384-512(%rax),%ymm12,%ymm12
+	vpaddd	416-512(%rax),%ymm13,%ymm13
+	vpaddd	448-512(%rax),%ymm15,%ymm15
+	vpaddd	480-512(%rax),%ymm9,%ymm9
+
+	vpunpckldq	%ymm13,%ymm12,%ymm2
+	vpunpckldq	%ymm9,%ymm15,%ymm8
+	vpunpckhdq	%ymm13,%ymm12,%ymm12
+	vpunpckhdq	%ymm9,%ymm15,%ymm15
+	vpunpcklqdq	%ymm8,%ymm2,%ymm13
+	vpunpckhqdq	%ymm8,%ymm2,%ymm2
+	vpunpcklqdq	%ymm15,%ymm12,%ymm9
+	vpunpckhqdq	%ymm15,%ymm12,%ymm12
+	vpaddd	512-512(%rax),%ymm4,%ymm4
+	vpaddd	544-512(%rax),%ymm5,%ymm5
+	vpaddd	576-512(%rax),%ymm6,%ymm6
+	vpaddd	608-512(%rax),%ymm7,%ymm7
+
+	vpunpckldq	%ymm5,%ymm4,%ymm15
+	vpunpckldq	%ymm7,%ymm6,%ymm8
+	vpunpckhdq	%ymm5,%ymm4,%ymm4
+	vpunpckhdq	%ymm7,%ymm6,%ymm6
+	vpunpcklqdq	%ymm8,%ymm15,%ymm5
+	vpunpckhqdq	%ymm8,%ymm15,%ymm15
+	vpunpcklqdq	%ymm6,%ymm4,%ymm7
+	vpunpckhqdq	%ymm6,%ymm4,%ymm4
+	vperm2i128	$0x20,%ymm5,%ymm13,%ymm8
+	vperm2i128	$0x31,%ymm5,%ymm13,%ymm5
+	vperm2i128	$0x20,%ymm15,%ymm2,%ymm13
+	vperm2i128	$0x31,%ymm15,%ymm2,%ymm15
+	vperm2i128	$0x20,%ymm7,%ymm9,%ymm2
+	vperm2i128	$0x31,%ymm7,%ymm9,%ymm7
+	vperm2i128	$0x20,%ymm4,%ymm12,%ymm9
+	vperm2i128	$0x31,%ymm4,%ymm12,%ymm4
+	vmovdqa	0(%rsp),%ymm6
+	vmovdqa	32(%rsp),%ymm12
+
+	cmpq	$512,%rdx
+	jb	.Ltail8x
+
+	vpxor	0(%rsi),%ymm6,%ymm6
+	vpxor	32(%rsi),%ymm8,%ymm8
+	vpxor	64(%rsi),%ymm1,%ymm1
+	vpxor	96(%rsi),%ymm5,%ymm5
+	leaq	128(%rsi),%rsi
+	vmovdqu	%ymm6,0(%rdi)
+	vmovdqu	%ymm8,32(%rdi)
+	vmovdqu	%ymm1,64(%rdi)
+	vmovdqu	%ymm5,96(%rdi)
+	leaq	128(%rdi),%rdi
+
+	vpxor	0(%rsi),%ymm12,%ymm12
+	vpxor	32(%rsi),%ymm13,%ymm13
+	vpxor	64(%rsi),%ymm10,%ymm10
+	vpxor	96(%rsi),%ymm15,%ymm15
+	leaq	128(%rsi),%rsi
+	vmovdqu	%ymm12,0(%rdi)
+	vmovdqu	%ymm13,32(%rdi)
+	vmovdqu	%ymm10,64(%rdi)
+	vmovdqu	%ymm15,96(%rdi)
+	leaq	128(%rdi),%rdi
+
+	vpxor	0(%rsi),%ymm14,%ymm14
+	vpxor	32(%rsi),%ymm2,%ymm2
+	vpxor	64(%rsi),%ymm3,%ymm3
+	vpxor	96(%rsi),%ymm7,%ymm7
+	leaq	128(%rsi),%rsi
+	vmovdqu	%ymm14,0(%rdi)
+	vmovdqu	%ymm2,32(%rdi)
+	vmovdqu	%ymm3,64(%rdi)
+	vmovdqu	%ymm7,96(%rdi)
+	leaq	128(%rdi),%rdi
+
+	vpxor	0(%rsi),%ymm11,%ymm11
+	vpxor	32(%rsi),%ymm9,%ymm9
+	vpxor	64(%rsi),%ymm0,%ymm0
+	vpxor	96(%rsi),%ymm4,%ymm4
+	leaq	128(%rsi),%rsi
+	vmovdqu	%ymm11,0(%rdi)
+	vmovdqu	%ymm9,32(%rdi)
+	vmovdqu	%ymm0,64(%rdi)
+	vmovdqu	%ymm4,96(%rdi)
+	leaq	128(%rdi),%rdi
+
+	subq	$512,%rdx
+	jnz	.Loop_outer8x
+
+	jmp	.Ldone8x
+
+.Ltail8x:
+	cmpq	$448,%rdx
+	jae	.L448_or_more8x
+	cmpq	$384,%rdx
+	jae	.L384_or_more8x
+	cmpq	$320,%rdx
+	jae	.L320_or_more8x
+	cmpq	$256,%rdx
+	jae	.L256_or_more8x
+	cmpq	$192,%rdx
+	jae	.L192_or_more8x
+	cmpq	$128,%rdx
+	jae	.L128_or_more8x
+	cmpq	$64,%rdx
+	jae	.L64_or_more8x
+
+	xorq	%r10,%r10
+	vmovdqa	%ymm6,0(%rsp)
+	vmovdqa	%ymm8,32(%rsp)
+	jmp	.Loop_tail8x
+
+.align	32
+.L64_or_more8x:
+	vpxor	0(%rsi),%ymm6,%ymm6
+	vpxor	32(%rsi),%ymm8,%ymm8
+	vmovdqu	%ymm6,0(%rdi)
+	vmovdqu	%ymm8,32(%rdi)
+	je	.Ldone8x
+
+	leaq	64(%rsi),%rsi
+	xorq	%r10,%r10
+	vmovdqa	%ymm1,0(%rsp)
+	leaq	64(%rdi),%rdi
+	subq	$64,%rdx
+	vmovdqa	%ymm5,32(%rsp)
+	jmp	.Loop_tail8x
+
+.align	32
+.L128_or_more8x:
+	vpxor	0(%rsi),%ymm6,%ymm6
+	vpxor	32(%rsi),%ymm8,%ymm8
+	vpxor	64(%rsi),%ymm1,%ymm1
+	vpxor	96(%rsi),%ymm5,%ymm5
+	vmovdqu	%ymm6,0(%rdi)
+	vmovdqu	%ymm8,32(%rdi)
+	vmovdqu	%ymm1,64(%rdi)
+	vmovdqu	%ymm5,96(%rdi)
+	je	.Ldone8x
+
+	leaq	128(%rsi),%rsi
+	xorq	%r10,%r10
+	vmovdqa	%ymm12,0(%rsp)
+	leaq	128(%rdi),%rdi
+	subq	$128,%rdx
+	vmovdqa	%ymm13,32(%rsp)
+	jmp	.Loop_tail8x
+
+.align	32
+.L192_or_more8x:
+	vpxor	0(%rsi),%ymm6,%ymm6
+	vpxor	32(%rsi),%ymm8,%ymm8
+	vpxor	64(%rsi),%ymm1,%ymm1
+	vpxor	96(%rsi),%ymm5,%ymm5
+	vpxor	128(%rsi),%ymm12,%ymm12
+	vpxor	160(%rsi),%ymm13,%ymm13
+	vmovdqu	%ymm6,0(%rdi)
+	vmovdqu	%ymm8,32(%rdi)
+	vmovdqu	%ymm1,64(%rdi)
+	vmovdqu	%ymm5,96(%rdi)
+	vmovdqu	%ymm12,128(%rdi)
+	vmovdqu	%ymm13,160(%rdi)
+	je	.Ldone8x
+
+	leaq	192(%rsi),%rsi
+	xorq	%r10,%r10
+	vmovdqa	%ymm10,0(%rsp)
+	leaq	192(%rdi),%rdi
+	subq	$192,%rdx
+	vmovdqa	%ymm15,32(%rsp)
+	jmp	.Loop_tail8x
+
+.align	32
+.L256_or_more8x:
+	vpxor	0(%rsi),%ymm6,%ymm6
+	vpxor	32(%rsi),%ymm8,%ymm8
+	vpxor	64(%rsi),%ymm1,%ymm1
+	vpxor	96(%rsi),%ymm5,%ymm5
+	vpxor	128(%rsi),%ymm12,%ymm12
+	vpxor	160(%rsi),%ymm13,%ymm13
+	vpxor	192(%rsi),%ymm10,%ymm10
+	vpxor	224(%rsi),%ymm15,%ymm15
+	vmovdqu	%ymm6,0(%rdi)
+	vmovdqu	%ymm8,32(%rdi)
+	vmovdqu	%ymm1,64(%rdi)
+	vmovdqu	%ymm5,96(%rdi)
+	vmovdqu	%ymm12,128(%rdi)
+	vmovdqu	%ymm13,160(%rdi)
+	vmovdqu	%ymm10,192(%rdi)
+	vmovdqu	%ymm15,224(%rdi)
+	je	.Ldone8x
+
+	leaq	256(%rsi),%rsi
+	xorq	%r10,%r10
+	vmovdqa	%ymm14,0(%rsp)
+	leaq	256(%rdi),%rdi
+	subq	$256,%rdx
+	vmovdqa	%ymm2,32(%rsp)
+	jmp	.Loop_tail8x
+
+.align	32
+.L320_or_more8x:
+	vpxor	0(%rsi),%ymm6,%ymm6
+	vpxor	32(%rsi),%ymm8,%ymm8
+	vpxor	64(%rsi),%ymm1,%ymm1
+	vpxor	96(%rsi),%ymm5,%ymm5
+	vpxor	128(%rsi),%ymm12,%ymm12
+	vpxor	160(%rsi),%ymm13,%ymm13
+	vpxor	192(%rsi),%ymm10,%ymm10
+	vpxor	224(%rsi),%ymm15,%ymm15
+	vpxor	256(%rsi),%ymm14,%ymm14
+	vpxor	288(%rsi),%ymm2,%ymm2
+	vmovdqu	%ymm6,0(%rdi)
+	vmovdqu	%ymm8,32(%rdi)
+	vmovdqu	%ymm1,64(%rdi)
+	vmovdqu	%ymm5,96(%rdi)
+	vmovdqu	%ymm12,128(%rdi)
+	vmovdqu	%ymm13,160(%rdi)
+	vmovdqu	%ymm10,192(%rdi)
+	vmovdqu	%ymm15,224(%rdi)
+	vmovdqu	%ymm14,256(%rdi)
+	vmovdqu	%ymm2,288(%rdi)
+	je	.Ldone8x
+
+	leaq	320(%rsi),%rsi
+	xorq	%r10,%r10
+	vmovdqa	%ymm3,0(%rsp)
+	leaq	320(%rdi),%rdi
+	subq	$320,%rdx
+	vmovdqa	%ymm7,32(%rsp)
+	jmp	.Loop_tail8x
+
+.align	32
+.L384_or_more8x:
+	vpxor	0(%rsi),%ymm6,%ymm6
+	vpxor	32(%rsi),%ymm8,%ymm8
+	vpxor	64(%rsi),%ymm1,%ymm1
+	vpxor	96(%rsi),%ymm5,%ymm5
+	vpxor	128(%rsi),%ymm12,%ymm12
+	vpxor	160(%rsi),%ymm13,%ymm13
+	vpxor	192(%rsi),%ymm10,%ymm10
+	vpxor	224(%rsi),%ymm15,%ymm15
+	vpxor	256(%rsi),%ymm14,%ymm14
+	vpxor	288(%rsi),%ymm2,%ymm2
+	vpxor	320(%rsi),%ymm3,%ymm3
+	vpxor	352(%rsi),%ymm7,%ymm7
+	vmovdqu	%ymm6,0(%rdi)
+	vmovdqu	%ymm8,32(%rdi)
+	vmovdqu	%ymm1,64(%rdi)
+	vmovdqu	%ymm5,96(%rdi)
+	vmovdqu	%ymm12,128(%rdi)
+	vmovdqu	%ymm13,160(%rdi)
+	vmovdqu	%ymm10,192(%rdi)
+	vmovdqu	%ymm15,224(%rdi)
+	vmovdqu	%ymm14,256(%rdi)
+	vmovdqu	%ymm2,288(%rdi)
+	vmovdqu	%ymm3,320(%rdi)
+	vmovdqu	%ymm7,352(%rdi)
+	je	.Ldone8x
+
+	leaq	384(%rsi),%rsi
+	xorq	%r10,%r10
+	vmovdqa	%ymm11,0(%rsp)
+	leaq	384(%rdi),%rdi
+	subq	$384,%rdx
+	vmovdqa	%ymm9,32(%rsp)
+	jmp	.Loop_tail8x
+
+.align	32
+.L448_or_more8x:
+	vpxor	0(%rsi),%ymm6,%ymm6
+	vpxor	32(%rsi),%ymm8,%ymm8
+	vpxor	64(%rsi),%ymm1,%ymm1
+	vpxor	96(%rsi),%ymm5,%ymm5
+	vpxor	128(%rsi),%ymm12,%ymm12
+	vpxor	160(%rsi),%ymm13,%ymm13
+	vpxor	192(%rsi),%ymm10,%ymm10
+	vpxor	224(%rsi),%ymm15,%ymm15
+	vpxor	256(%rsi),%ymm14,%ymm14
+	vpxor	288(%rsi),%ymm2,%ymm2
+	vpxor	320(%rsi),%ymm3,%ymm3
+	vpxor	352(%rsi),%ymm7,%ymm7
+	vpxor	384(%rsi),%ymm11,%ymm11
+	vpxor	416(%rsi),%ymm9,%ymm9
+	vmovdqu	%ymm6,0(%rdi)
+	vmovdqu	%ymm8,32(%rdi)
+	vmovdqu	%ymm1,64(%rdi)
+	vmovdqu	%ymm5,96(%rdi)
+	vmovdqu	%ymm12,128(%rdi)
+	vmovdqu	%ymm13,160(%rdi)
+	vmovdqu	%ymm10,192(%rdi)
+	vmovdqu	%ymm15,224(%rdi)
+	vmovdqu	%ymm14,256(%rdi)
+	vmovdqu	%ymm2,288(%rdi)
+	vmovdqu	%ymm3,320(%rdi)
+	vmovdqu	%ymm7,352(%rdi)
+	vmovdqu	%ymm11,384(%rdi)
+	vmovdqu	%ymm9,416(%rdi)
+	je	.Ldone8x
+
+	leaq	448(%rsi),%rsi
+	xorq	%r10,%r10
+	vmovdqa	%ymm0,0(%rsp)
+	leaq	448(%rdi),%rdi
+	subq	$448,%rdx
+	vmovdqa	%ymm4,32(%rsp)
+
+.Loop_tail8x:
+	movzbl	(%rsi,%r10,1),%eax
+	movzbl	(%rsp,%r10,1),%ecx
+	leaq	1(%r10),%r10
+	xorl	%ecx,%eax
+	movb	%al,-1(%rdi,%r10,1)
+	decq	%rdx
+	jnz	.Loop_tail8x
+
+.Ldone8x:
+	vzeroall
+	movq	640(%rsp),%rsp
+	.byte	0xf3,0xc3
+.size	ChaCha20_8x,.-ChaCha20_8x
+#endif

third_party/boringssl/linux-x86_64/crypto/cpu-x86_64-asm.S

diff --git a/third_party/boringssl/linux-x86_64/crypto/cpu-x86_64-asm.S b/third_party/boringssl/linux-x86_64/crypto/cpu-x86_64-asm.S
deleted file mode 100644
index 9eef154..0000000
--- a/third_party/boringssl/linux-x86_64/crypto/cpu-x86_64-asm.S
+++ /dev/null
@@ -1,143 +0,0 @@
-#if defined(__x86_64__)
-.text	
-
-.globl	OPENSSL_ia32_cpuid
-.hidden OPENSSL_ia32_cpuid
-.type	OPENSSL_ia32_cpuid,@function
-.align	16
-OPENSSL_ia32_cpuid:
-
-
-	movq	%rdi,%rdi
-	movq	%rbx,%r8
-
-	xorl	%eax,%eax
-	movl	%eax,8(%rdi)
-	cpuid
-	movl	%eax,%r11d
-
-	xorl	%eax,%eax
-	cmpl	$1970169159,%ebx
-	setne	%al
-	movl	%eax,%r9d
-	cmpl	$1231384169,%edx
-	setne	%al
-	orl	%eax,%r9d
-	cmpl	$1818588270,%ecx
-	setne	%al
-	orl	%eax,%r9d
-	jz	.Lintel
-
-	cmpl	$1752462657,%ebx
-	setne	%al
-	movl	%eax,%r10d
-	cmpl	$1769238117,%edx
-	setne	%al
-	orl	%eax,%r10d
-	cmpl	$1145913699,%ecx
-	setne	%al
-	orl	%eax,%r10d
-	jnz	.Lintel
-
-
-
-
-	movl	$2147483648,%eax
-	cpuid
-
-
-	cmpl	$2147483649,%eax
-	jb	.Lintel
-	movl	%eax,%r10d
-	movl	$2147483649,%eax
-	cpuid
-
-
-	orl	%ecx,%r9d
-	andl	$2049,%r9d
-
-	cmpl	$2147483656,%r10d
-	jb	.Lintel
-
-	movl	$2147483656,%eax
-	cpuid
-
-	movzbq	%cl,%r10
-	incq	%r10
-
-	movl	$1,%eax
-	cpuid
-
-	btl	$28,%edx
-	jnc	.Lgeneric
-	shrl	$16,%ebx
-	cmpb	%r10b,%bl
-	ja	.Lgeneric
-	andl	$4026531839,%edx
-	jmp	.Lgeneric
-
-.Lintel:
-	cmpl	$4,%r11d
-	movl	$-1,%r10d
-	jb	.Lnocacheinfo
-
-	movl	$4,%eax
-	movl	$0,%ecx
-	cpuid
-	movl	%eax,%r10d
-	shrl	$14,%r10d
-	andl	$4095,%r10d
-
-	cmpl	$7,%r11d
-	jb	.Lnocacheinfo
-
-	movl	$7,%eax
-	xorl	%ecx,%ecx
-	cpuid
-	movl	%ebx,8(%rdi)
-
-.Lnocacheinfo:
-	movl	$1,%eax
-	cpuid
-
-	andl	$3220176895,%edx
-	cmpl	$0,%r9d
-	jne	.Lnotintel
-	orl	$1073741824,%edx
-.Lnotintel:
-	btl	$28,%edx
-	jnc	.Lgeneric
-	andl	$4026531839,%edx
-	cmpl	$0,%r10d
-	je	.Lgeneric
-
-	orl	$268435456,%edx
-	shrl	$16,%ebx
-	cmpb	$1,%bl
-	ja	.Lgeneric
-	andl	$4026531839,%edx
-.Lgeneric:
-	andl	$2048,%r9d
-	andl	$4294965247,%ecx
-	orl	%ecx,%r9d
-
-	movl	%edx,%r10d
-	btl	$27,%r9d
-	jnc	.Lclear_avx
-	xorl	%ecx,%ecx
-.byte	0x0f,0x01,0xd0
-	andl	$6,%eax
-	cmpl	$6,%eax
-	je	.Ldone
-.Lclear_avx:
-	movl	$4026525695,%eax
-	andl	%eax,%r9d
-	andl	$4294967263,8(%rdi)
-.Ldone:
-	movl	%r9d,4(%rdi)
-	movl	%r10d,0(%rdi)
-	movq	%r8,%rbx
-	.byte	0xf3,0xc3
-.size	OPENSSL_ia32_cpuid,.-OPENSSL_ia32_cpuid
-
-#endif

third_party/boringssl/linux-x86_64/crypto/ec/p256-x86_64-asm.S

diff --git a/third_party/boringssl/linux-x86_64/crypto/ec/p256-x86_64-asm.S b/third_party/boringssl/linux-x86_64/crypto/ec/p256-x86_64-asm.S
new file mode 100644
index 0000000..4abce6f
--- /dev/null
+++ b/third_party/boringssl/linux-x86_64/crypto/ec/p256-x86_64-asm.S
@@ -0,0 +1,1789 @@
+#if defined(__x86_64__)
+.text	
+.extern	OPENSSL_ia32cap_P
+.hidden OPENSSL_ia32cap_P
+
+
+.align	64
+.Lpoly:
+.quad	0xffffffffffffffff, 0x00000000ffffffff, 0x0000000000000000, 0xffffffff00000001
+
+.LOne:
+.long	1,1,1,1,1,1,1,1
+.LTwo:
+.long	2,2,2,2,2,2,2,2
+.LThree:
+.long	3,3,3,3,3,3,3,3
+.LONE_mont:
+.quad	0x0000000000000001, 0xffffffff00000000, 0xffffffffffffffff, 0x00000000fffffffe
+
+.type	ecp_nistz256_mul_by_2,@function
+.align	64
+ecp_nistz256_mul_by_2:
+	pushq	%r12
+	pushq	%r13
+
+	movq	0(%rsi),%r8
+	movq	8(%rsi),%r9
+	addq	%r8,%r8
+	movq	16(%rsi),%r10
+	adcq	%r9,%r9
+	movq	24(%rsi),%r11
+	leaq	.Lpoly(%rip),%rsi
+	movq	%r8,%rax
+	adcq	%r10,%r10
+	adcq	%r11,%r11
+	movq	%r9,%rdx
+	sbbq	%r13,%r13
+
+	subq	0(%rsi),%r8
+	movq	%r10,%rcx
+	sbbq	8(%rsi),%r9
+	sbbq	16(%rsi),%r10
+	movq	%r11,%r12
+	sbbq	24(%rsi),%r11
+	testq	%r13,%r13
+
+	cmovzq	%rax,%r8
+	cmovzq	%rdx,%r9
+	movq	%r8,0(%rdi)
+	cmovzq	%rcx,%r10
+	movq	%r9,8(%rdi)
+	cmovzq	%r12,%r11
+	movq	%r10,16(%rdi)
+	movq	%r11,24(%rdi)
+
+	popq	%r13
+	popq	%r12
+	.byte	0xf3,0xc3
+.size	ecp_nistz256_mul_by_2,.-ecp_nistz256_mul_by_2
+
+
+
+.globl	ecp_nistz256_neg
+.hidden ecp_nistz256_neg
+.type	ecp_nistz256_neg,@function
+.align	32
+ecp_nistz256_neg:
+	pushq	%r12
+	pushq	%r13
+
+	xorq	%r8,%r8
+	xorq	%r9,%r9
+	xorq	%r10,%r10
+	xorq	%r11,%r11
+	xorq	%r13,%r13
+
+	subq	0(%rsi),%r8
+	sbbq	8(%rsi),%r9
+	sbbq	16(%rsi),%r10
+	movq	%r8,%rax
+	sbbq	24(%rsi),%r11
+	leaq	.Lpoly(%rip),%rsi
+	movq	%r9,%rdx
+	sbbq	$0,%r13
+
+	addq	0(%rsi),%r8
+	movq	%r10,%rcx
+	adcq	8(%rsi),%r9
+	adcq	16(%rsi),%r10
+	movq	%r11,%r12
+	adcq	24(%rsi),%r11
+	testq	%r13,%r13
+
+	cmovzq	%rax,%r8
+	cmovzq	%rdx,%r9
+	movq	%r8,0(%rdi)
+	cmovzq	%rcx,%r10
+	movq	%r9,8(%rdi)
+	cmovzq	%r12,%r11
+	movq	%r10,16(%rdi)
+	movq	%r11,24(%rdi)
+
+	popq	%r13
+	popq	%r12
+	.byte	0xf3,0xc3
+.size	ecp_nistz256_neg,.-ecp_nistz256_neg
+
+
+
+
+
+
+.globl	ecp_nistz256_mul_mont
+.hidden ecp_nistz256_mul_mont
+.type	ecp_nistz256_mul_mont,@function
+.align	32
+ecp_nistz256_mul_mont:
+.Lmul_mont:
+	pushq	%rbp
+	pushq	%rbx
+	pushq	%r12
+	pushq	%r13
+	pushq	%r14
+	pushq	%r15
+	movq	%rdx,%rbx
+	movq	0(%rdx),%rax
+	movq	0(%rsi),%r9
+	movq	8(%rsi),%r10
+	movq	16(%rsi),%r11
+	movq	24(%rsi),%r12
+
+	call	__ecp_nistz256_mul_montq
+.Lmul_mont_done:
+	popq	%r15
+	popq	%r14
+	popq	%r13
+	popq	%r12
+	popq	%rbx
+	popq	%rbp
+	.byte	0xf3,0xc3
+.size	ecp_nistz256_mul_mont,.-ecp_nistz256_mul_mont
+
+.type	__ecp_nistz256_mul_montq,@function
+.align	32
+__ecp_nistz256_mul_montq:
+
+
+	movq	%rax,%rbp
+	mulq	%r9
+	movq	.Lpoly+8(%rip),%r14
+	movq	%rax,%r8
+	movq	%rbp,%rax
+	movq	%rdx,%r9
+
+	mulq	%r10
+	movq	.Lpoly+24(%rip),%r15
+	addq	%rax,%r9
+	movq	%rbp,%rax
+	adcq	$0,%rdx
+	movq	%rdx,%r10
+
+	mulq	%r11
+	addq	%rax,%r10
+	movq	%rbp,%rax
+	adcq	$0,%rdx
+	movq	%rdx,%r11
+
+	mulq	%r12
+	addq	%rax,%r11
+	movq	%r8,%rax
+	adcq	$0,%rdx
+	xorq	%r13,%r13
+	movq	%rdx,%r12
+
+
+
+
+
+
+
+
+
+
+	movq	%r8,%rbp
+	shlq	$32,%r8
+	mulq	%r15
+	shrq	$32,%rbp
+	addq	%r8,%r9
+	adcq	%rbp,%r10
+	adcq	%rax,%r11
+	movq	8(%rbx),%rax
+	adcq	%rdx,%r12
+	adcq	$0,%r13
+	xorq	%r8,%r8
+
+
+
+	movq	%rax,%rbp
+	mulq	0(%rsi)
+	addq	%rax,%r9
+	movq	%rbp,%rax
+	adcq	$0,%rdx
+	movq	%rdx,%rcx
+
+	mulq	8(%rsi)
+	addq	%rcx,%r10
+	adcq	$0,%rdx
+	addq	%rax,%r10
+	movq	%rbp,%rax
+	adcq	$0,%rdx
+	movq	%rdx,%rcx
+
+	mulq	16(%rsi)
+	addq	%rcx,%r11
+	adcq	$0,%rdx
+	addq	%rax,%r11
+	movq	%rbp,%rax
+	adcq	$0,%rdx
+	movq	%rdx,%rcx
+
+	mulq	24(%rsi)
+	addq	%rcx,%r12
+	adcq	$0,%rdx
+	addq	%rax,%r12
+	movq	%r9,%rax
+	adcq	%rdx,%r13
+	adcq	$0,%r8
+
+
+
+	movq	%r9,%rbp
+	shlq	$32,%r9
+	mulq	%r15
+	shrq	$32,%rbp
+	addq	%r9,%r10
+	adcq	%rbp,%r11
+	adcq	%rax,%r12
+	movq	16(%rbx),%rax
+	adcq	%rdx,%r13
+	adcq	$0,%r8
+	xorq	%r9,%r9
+
+
+
+	movq	%rax,%rbp
+	mulq	0(%rsi)
+	addq	%rax,%r10
+	movq	%rbp,%rax
+	adcq	$0,%rdx
+	movq	%rdx,%rcx
+
+	mulq	8(%rsi)
+	addq	%rcx,%r11
+	adcq	$0,%rdx
+	addq	%rax,%r11
+	movq	%rbp,%rax
+	adcq	$0,%rdx
+	movq	%rdx,%rcx
+
+	mulq	16(%rsi)
+	addq	%rcx,%r12
+	adcq	$0,%rdx
+	addq	%rax,%r12
+	movq	%rbp,%rax
+	adcq	$0,%rdx
+	movq	%rdx,%rcx
+
+	mulq	24(%rsi)
+	addq	%rcx,%r13
+	adcq	$0,%rdx
+	addq	%rax,%r13
+	movq	%r10,%rax
+	adcq	%rdx,%r8
+	adcq	$0,%r9
+
+
+
+	movq	%r10,%rbp
+	shlq	$32,%r10
+	mulq	%r15
+	shrq	$32,%rbp
+	addq	%r10,%r11
+	adcq	%rbp,%r12
+	adcq	%rax,%r13
+	movq	24(%rbx),%rax
+	adcq	%rdx,%r8
+	adcq	$0,%r9
+	xorq	%r10,%r10
+
+
+
+	movq	%rax,%rbp
+	mulq	0(%rsi)
+	addq	%rax,%r11
+	movq	%rbp,%rax
+	adcq	$0,%rdx
+	movq	%rdx,%rcx
+
+	mulq	8(%rsi)
+	addq	%rcx,%r12
+	adcq	$0,%rdx
+	addq	%rax,%r12
+	movq	%rbp,%rax
+	adcq	$0,%rdx
+	movq	%rdx,%rcx
+
+	mulq	16(%rsi)
+	addq	%rcx,%r13
+	adcq	$0,%rdx
+	addq	%rax,%r13
+	movq	%rbp,%rax
+	adcq	$0,%rdx
+	movq	%rdx,%rcx
+
+	mulq	24(%rsi)
+	addq	%rcx,%r8
+	adcq	$0,%rdx
+	addq	%rax,%r8
+	movq	%r11,%rax
+	adcq	%rdx,%r9
+	adcq	$0,%r10
+
+
+
+	movq	%r11,%rbp
+	shlq	$32,%r11
+	mulq	%r15
+	shrq	$32,%rbp
+	addq	%r11,%r12
+	adcq	%rbp,%r13
+	movq	%r12,%rcx
+	adcq	%rax,%r8
+	adcq	%rdx,%r9
+	movq	%r13,%rbp
+	adcq	$0,%r10
+
+
+
+	subq	$-1,%r12
+	movq	%r8,%rbx
+	sbbq	%r14,%r13
+	sbbq	$0,%r8
+	movq	%r9,%rdx
+	sbbq	%r15,%r9
+	sbbq	$0,%r10
+
+	cmovcq	%rcx,%r12
+	cmovcq	%rbp,%r13
+	movq	%r12,0(%rdi)
+	cmovcq	%rbx,%r8
+	movq	%r13,8(%rdi)
+	cmovcq	%rdx,%r9
+	movq	%r8,16(%rdi)
+	movq	%r9,24(%rdi)
+
+	.byte	0xf3,0xc3
+.size	__ecp_nistz256_mul_montq,.-__ecp_nistz256_mul_montq
+
+
+
+
+
+
+
+
+.globl	ecp_nistz256_sqr_mont
+.hidden ecp_nistz256_sqr_mont
+.type	ecp_nistz256_sqr_mont,@function
+.align	32
+ecp_nistz256_sqr_mont:
+	pushq	%rbp
+	pushq	%rbx
+	pushq	%r12
+	pushq	%r13
+	pushq	%r14
+	pushq	%r15
+	movq	0(%rsi),%rax
+	movq	8(%rsi),%r14
+	movq	16(%rsi),%r15
+	movq	24(%rsi),%r8
+
+	call	__ecp_nistz256_sqr_montq
+.Lsqr_mont_done:
+	popq	%r15
+	popq	%r14
+	popq	%r13
+	popq	%r12
+	popq	%rbx
+	popq	%rbp
+	.byte	0xf3,0xc3
+.size	ecp_nistz256_sqr_mont,.-ecp_nistz256_sqr_mont
+
+.type	__ecp_nistz256_sqr_montq,@function
+.align	32
+__ecp_nistz256_sqr_montq:
+	movq	%rax,%r13
+	mulq	%r14
+	movq	%rax,%r9
+	movq	%r15,%rax
+	movq	%rdx,%r10
+
+	mulq	%r13
+	addq	%rax,%r10
+	movq	%r8,%rax
+	adcq	$0,%rdx
+	movq	%rdx,%r11
+
+	mulq	%r13
+	addq	%rax,%r11
+	movq	%r15,%rax
+	adcq	$0,%rdx
+	movq	%rdx,%r12
+
+
+	mulq	%r14
+	addq	%rax,%r11
+	movq	%r8,%rax
+	adcq	$0,%rdx
+	movq	%rdx,%rbp
+
+	mulq	%r14
+	addq	%rax,%r12
+	movq	%r8,%rax
+	adcq	$0,%rdx
+	addq	%rbp,%r12
+	movq	%rdx,%r13
+	adcq	$0,%r13
+
+
+	mulq	%r15
+	xorq	%r15,%r15
+	addq	%rax,%r13
+	movq	0(%rsi),%rax
+	movq	%rdx,%r14
+	adcq	$0,%r14
+
+	addq	%r9,%r9
+	adcq	%r10,%r10
+	adcq	%r11,%r11
+	adcq	%r12,%r12
+	adcq	%r13,%r13
+	adcq	%r14,%r14
+	adcq	$0,%r15
+
+	mulq	%rax
+	movq	%rax,%r8
+	movq	8(%rsi),%rax
+	movq	%rdx,%rcx
+
+	mulq	%rax
+	addq	%rcx,%r9
+	adcq	%rax,%r10
+	movq	16(%rsi),%rax
+	adcq	$0,%rdx
+	movq	%rdx,%rcx
+
+	mulq	%rax
+	addq	%rcx,%r11
+	adcq	%rax,%r12
+	movq	24(%rsi),%rax
+	adcq	$0,%rdx
+	movq	%rdx,%rcx
+
+	mulq	%rax
+	addq	%rcx,%r13
+	adcq	%rax,%r14
+	movq	%r8,%rax
+	adcq	%rdx,%r15
+
+	movq	.Lpoly+8(%rip),%rsi
+	movq	.Lpoly+24(%rip),%rbp
+
+
+
+
+	movq	%r8,%rcx
+	shlq	$32,%r8
+	mulq	%rbp
+	shrq	$32,%rcx
+	addq	%r8,%r9
+	adcq	%rcx,%r10
+	adcq	%rax,%r11
+	movq	%r9,%rax
+	adcq	$0,%rdx
+
+
+
+	movq	%r9,%rcx
+	shlq	$32,%r9
+	movq	%rdx,%r8
+	mulq	%rbp
+	shrq	$32,%rcx
+	addq	%r9,%r10
+	adcq	%rcx,%r11
+	adcq	%rax,%r8
+	movq	%r10,%rax
+	adcq	$0,%rdx
+
+
+
+	movq	%r10,%rcx
+	shlq	$32,%r10
+	movq	%rdx,%r9
+	mulq	%rbp
+	shrq	$32,%rcx
+	addq	%r10,%r11
+	adcq	%rcx,%r8
+	adcq	%rax,%r9
+	movq	%r11,%rax
+	adcq	$0,%rdx
+
+
+
+	movq	%r11,%rcx
+	shlq	$32,%r11
+	movq	%rdx,%r10
+	mulq	%rbp
+	shrq	$32,%rcx
+	addq	%r11,%r8
+	adcq	%rcx,%r9
+	adcq	%rax,%r10
+	adcq	$0,%rdx
+	xorq	%r11,%r11
+
+
+
+	addq	%r8,%r12
+	adcq	%r9,%r13
+	movq	%r12,%r8
+	adcq	%r10,%r14
+	adcq	%rdx,%r15
+	movq	%r13,%r9
+	adcq	$0,%r11
+
+	subq	$-1,%r12
+	movq	%r14,%r10
+	sbbq	%rsi,%r13
+	sbbq	$0,%r14
+	movq	%r15,%rcx
+	sbbq	%rbp,%r15
+	sbbq	$0,%r11
+
+	cmovcq	%r8,%r12
+	cmovcq	%r9,%r13
+	movq	%r12,0(%rdi)
+	cmovcq	%r10,%r14
+	movq	%r13,8(%rdi)
+	cmovcq	%rcx,%r15
+	movq	%r14,16(%rdi)
+	movq	%r15,24(%rdi)
+
+	.byte	0xf3,0xc3
+.size	__ecp_nistz256_sqr_montq,.-__ecp_nistz256_sqr_montq
+
+
+
+
+
+
+.globl	ecp_nistz256_from_mont
+.hidden ecp_nistz256_from_mont
+.type	ecp_nistz256_from_mont,@function
+.align	32
+ecp_nistz256_from_mont:
+	pushq	%r12
+	pushq	%r13
+
+	movq	0(%rsi),%rax
+	movq	.Lpoly+24(%rip),%r13
+	movq	8(%rsi),%r9
+	movq	16(%rsi),%r10
+	movq	24(%rsi),%r11
+	movq	%rax,%r8
+	movq	.Lpoly+8(%rip),%r12
+
+
+
+	movq	%rax,%rcx
+	shlq	$32,%r8
+	mulq	%r13
+	shrq	$32,%rcx
+	addq	%r8,%r9
+	adcq	%rcx,%r10
+	adcq	%rax,%r11
+	movq	%r9,%rax
+	adcq	$0,%rdx
+
+
+
+	movq	%r9,%rcx
+	shlq	$32,%r9
+	movq	%rdx,%r8
+	mulq	%r13
+	shrq	$32,%rcx
+	addq	%r9,%r10
+	adcq	%rcx,%r11
+	adcq	%rax,%r8
+	movq	%r10,%rax
+	adcq	$0,%rdx
+
+
+
+	movq	%r10,%rcx
+	shlq	$32,%r10
+	movq	%rdx,%r9
+	mulq	%r13
+	shrq	$32,%rcx
+	addq	%r10,%r11
+	adcq	%rcx,%r8
+	adcq	%rax,%r9
+	movq	%r11,%rax
+	adcq	$0,%rdx
+
+
+
+	movq	%r11,%rcx
+	shlq	$32,%r11
+	movq	%rdx,%r10
+	mulq	%r13
+	shrq	$32,%rcx
+	addq	%r11,%r8
+	adcq	%rcx,%r9
+	movq	%r8,%rcx
+	adcq	%rax,%r10
+	movq	%r9,%rsi
+	adcq	$0,%rdx
+
+	subq	$-1,%r8
+	movq	%r10,%rax
+	sbbq	%r12,%r9
+	sbbq	$0,%r10
+	movq	%rdx,%r11
+	sbbq	%r13,%rdx
+	sbbq	%r13,%r13
+
+	cmovnzq	%rcx,%r8
+	cmovnzq	%rsi,%r9
+	movq	%r8,0(%rdi)
+	cmovnzq	%rax,%r10
+	movq	%r9,8(%rdi)
+	cmovzq	%rdx,%r11
+	movq	%r10,16(%rdi)
+	movq	%r11,24(%rdi)
+
+	popq	%r13
+	popq	%r12
+	.byte	0xf3,0xc3
+.size	ecp_nistz256_from_mont,.-ecp_nistz256_from_mont
+
+
+.globl	ecp_nistz256_select_w5
+.hidden ecp_nistz256_select_w5
+.type	ecp_nistz256_select_w5,@function
+.align	32
+ecp_nistz256_select_w5:
+	movdqa	.LOne(%rip),%xmm0
+	movd	%edx,%xmm1
+
+	pxor	%xmm2,%xmm2
+	pxor	%xmm3,%xmm3
+	pxor	%xmm4,%xmm4
+	pxor	%xmm5,%xmm5
+	pxor	%xmm6,%xmm6
+	pxor	%xmm7,%xmm7
+
+	movdqa	%xmm0,%xmm8
+	pshufd	$0,%xmm1,%xmm1
+
+	movq	$16,%rax
+.Lselect_loop_sse_w5:
+
+	movdqa	%xmm8,%xmm15
+	paddd	%xmm0,%xmm8
+	pcmpeqd	%xmm1,%xmm15
+
+	movdqa	0(%rsi),%xmm9
+	movdqa	16(%rsi),%xmm10
+	movdqa	32(%rsi),%xmm11
+	movdqa	48(%rsi),%xmm12
+	movdqa	64(%rsi),%xmm13
+	movdqa	80(%rsi),%xmm14
+	leaq	96(%rsi),%rsi
+
+	pand	%xmm15,%xmm9
+	pand	%xmm15,%xmm10
+	por	%xmm9,%xmm2
+	pand	%xmm15,%xmm11
+	por	%xmm10,%xmm3
+	pand	%xmm15,%xmm12
+	por	%xmm11,%xmm4
+	pand	%xmm15,%xmm13
+	por	%xmm12,%xmm5
+	pand	%xmm15,%xmm14
+	por	%xmm13,%xmm6
+	por	%xmm14,%xmm7
+
+	decq	%rax
+	jnz	.Lselect_loop_sse_w5
+
+	movdqu	%xmm2,0(%rdi)
+	movdqu	%xmm3,16(%rdi)
+	movdqu	%xmm4,32(%rdi)
+	movdqu	%xmm5,48(%rdi)
+	movdqu	%xmm6,64(%rdi)
+	movdqu	%xmm7,80(%rdi)
+	.byte	0xf3,0xc3
+.size	ecp_nistz256_select_w5,.-ecp_nistz256_select_w5
+
+
+
+.globl	ecp_nistz256_select_w7
+.hidden ecp_nistz256_select_w7
+.type	ecp_nistz256_select_w7,@function
+.align	32
+ecp_nistz256_select_w7:
+	movdqa	.LOne(%rip),%xmm8
+	movd	%edx,%xmm1
+
+	pxor	%xmm2,%xmm2
+	pxor	%xmm3,%xmm3
+	pxor	%xmm4,%xmm4
+	pxor	%xmm5,%xmm5
+
+	movdqa	%xmm8,%xmm0
+	pshufd	$0,%xmm1,%xmm1
+	movq	$64,%rax
+
+.Lselect_loop_sse_w7:
+	movdqa	%xmm8,%xmm15
+	paddd	%xmm0,%xmm8
+	movdqa	0(%rsi),%xmm9
+	movdqa	16(%rsi),%xmm10
+	pcmpeqd	%xmm1,%xmm15
+	movdqa	32(%rsi),%xmm11
+	movdqa	48(%rsi),%xmm12
+	leaq	64(%rsi),%rsi
+
+	pand	%xmm15,%xmm9
+	pand	%xmm15,%xmm10
+	por	%xmm9,%xmm2
+	pand	%xmm15,%xmm11
+	por	%xmm10,%xmm3
+	pand	%xmm15,%xmm12
+	por	%xmm11,%xmm4
+	prefetcht0	255(%rsi)
+	por	%xmm12,%xmm5
+
+	decq	%rax
+	jnz	.Lselect_loop_sse_w7
+
+	movdqu	%xmm2,0(%rdi)
+	movdqu	%xmm3,16(%rdi)
+	movdqu	%xmm4,32(%rdi)
+	movdqu	%xmm5,48(%rdi)
+	.byte	0xf3,0xc3
+.size	ecp_nistz256_select_w7,.-ecp_nistz256_select_w7
+.globl	ecp_nistz256_avx2_select_w7
+.hidden ecp_nistz256_avx2_select_w7
+.type	ecp_nistz256_avx2_select_w7,@function
+.align	32
+ecp_nistz256_avx2_select_w7:
+.byte	0x0f,0x0b
+	.byte	0xf3,0xc3
+.size	ecp_nistz256_avx2_select_w7,.-ecp_nistz256_avx2_select_w7
+.type	__ecp_nistz256_add_toq,@function
+.align	32
+__ecp_nistz256_add_toq:
+	addq	0(%rbx),%r12
+	adcq	8(%rbx),%r13
+	movq	%r12,%rax
+	adcq	16(%rbx),%r8
+	adcq	24(%rbx),%r9
+	movq	%r13,%rbp
+	sbbq	%r11,%r11
+
+	subq	$-1,%r12
+	movq	%r8,%rcx
+	sbbq	%r14,%r13
+	sbbq	$0,%r8
+	movq	%r9,%r10
+	sbbq	%r15,%r9
+	testq	%r11,%r11
+
+	cmovzq	%rax,%r12
+	cmovzq	%rbp,%r13
+	movq	%r12,0(%rdi)
+	cmovzq	%rcx,%r8
+	movq	%r13,8(%rdi)
+	cmovzq	%r10,%r9
+	movq	%r8,16(%rdi)
+	movq	%r9,24(%rdi)
+
+	.byte	0xf3,0xc3
+.size	__ecp_nistz256_add_toq,.-__ecp_nistz256_add_toq
+
+.type	__ecp_nistz256_sub_fromq,@function
+.align	32
+__ecp_nistz256_sub_fromq:
+	subq	0(%rbx),%r12
+	sbbq	8(%rbx),%r13
+	movq	%r12,%rax
+	sbbq	16(%rbx),%r8
+	sbbq	24(%rbx),%r9
+	movq	%r13,%rbp
+	sbbq	%r11,%r11
+
+	addq	$-1,%r12
+	movq	%r8,%rcx
+	adcq	%r14,%r13
+	adcq	$0,%r8
+	movq	%r9,%r10
+	adcq	%r15,%r9
+	testq	%r11,%r11
+
+	cmovzq	%rax,%r12
+	cmovzq	%rbp,%r13
+	movq	%r12,0(%rdi)
+	cmovzq	%rcx,%r8
+	movq	%r13,8(%rdi)
+	cmovzq	%r10,%r9
+	movq	%r8,16(%rdi)
+	movq	%r9,24(%rdi)
+
+	.byte	0xf3,0xc3
+.size	__ecp_nistz256_sub_fromq,.-__ecp_nistz256_sub_fromq
+
+.type	__ecp_nistz256_subq,@function
+.align	32
+__ecp_nistz256_subq:
+	subq	%r12,%rax
+	sbbq	%r13,%rbp
+	movq	%rax,%r12
+	sbbq	%r8,%rcx
+	sbbq	%r9,%r10
+	movq	%rbp,%r13
+	sbbq	%r11,%r11
+
+	addq	$-1,%rax
+	movq	%rcx,%r8
+	adcq	%r14,%rbp
+	adcq	$0,%rcx
+	movq	%r10,%r9
+	adcq	%r15,%r10
+	testq	%r11,%r11
+
+	cmovnzq	%rax,%r12
+	cmovnzq	%rbp,%r13
+	cmovnzq	%rcx,%r8
+	cmovnzq	%r10,%r9
+
+	.byte	0xf3,0xc3
+.size	__ecp_nistz256_subq,.-__ecp_nistz256_subq
+
+.type	__ecp_nistz256_mul_by_2q,@function
+.align	32
+__ecp_nistz256_mul_by_2q:
+	addq	%r12,%r12
+	adcq	%r13,%r13
+	movq	%r12,%rax
+	adcq	%r8,%r8
+	adcq	%r9,%r9
+	movq	%r13,%rbp
+	sbbq	%r11,%r11
+
+	subq	$-1,%r12
+	movq	%r8,%rcx
+	sbbq	%r14,%r13
+	sbbq	$0,%r8
+	movq	%r9,%r10
+	sbbq	%r15,%r9
+	testq	%r11,%r11
+
+	cmovzq	%rax,%r12
+	cmovzq	%rbp,%r13
+	movq	%r12,0(%rdi)
+	cmovzq	%rcx,%r8
+	movq	%r13,8(%rdi)
+	cmovzq	%r10,%r9
+	movq	%r8,16(%rdi)
+	movq	%r9,24(%rdi)
+
+	.byte	0xf3,0xc3
+.size	__ecp_nistz256_mul_by_2q,.-__ecp_nistz256_mul_by_2q
+.globl	ecp_nistz256_point_double
+.hidden ecp_nistz256_point_double
+.type	ecp_nistz256_point_double,@function
+.align	32
+ecp_nistz256_point_double:
+	pushq	%rbp
+	pushq	%rbx
+	pushq	%r12
+	pushq	%r13
+	pushq	%r14
+	pushq	%r15
+	subq	$160+8,%rsp
+
+.Lpoint_double_shortcutq:
+	movdqu	0(%rsi),%xmm0
+	movq	%rsi,%rbx
+	movdqu	16(%rsi),%xmm1
+	movq	32+0(%rsi),%r12
+	movq	32+8(%rsi),%r13
+	movq	32+16(%rsi),%r8
+	movq	32+24(%rsi),%r9
+	movq	.Lpoly+8(%rip),%r14
+	movq	.Lpoly+24(%rip),%r15
+	movdqa	%xmm0,96(%rsp)
+	movdqa	%xmm1,96+16(%rsp)
+	leaq	32(%rdi),%r10
+	leaq	64(%rdi),%r11
+.byte	102,72,15,110,199
+.byte	102,73,15,110,202
+.byte	102,73,15,110,211
+
+	leaq	0(%rsp),%rdi
+	call	__ecp_nistz256_mul_by_2q
+
+	movq	64+0(%rsi),%rax
+	movq	64+8(%rsi),%r14
+	movq	64+16(%rsi),%r15
+	movq	64+24(%rsi),%r8
+	leaq	64-0(%rsi),%rsi
+	leaq	64(%rsp),%rdi
+	call	__ecp_nistz256_sqr_montq
+
+	movq	0+0(%rsp),%rax
+	movq	8+0(%rsp),%r14
+	leaq	0+0(%rsp),%rsi
+	movq	16+0(%rsp),%r15
+	movq	24+0(%rsp),%r8
+	leaq	0(%rsp),%rdi
+	call	__ecp_nistz256_sqr_montq
+
+	movq	32(%rbx),%rax
+	movq	64+0(%rbx),%r9
+	movq	64+8(%rbx),%r10
+	movq	64+16(%rbx),%r11
+	movq	64+24(%rbx),%r12
+	leaq	64-0(%rbx),%rsi
+	leaq	32(%rbx),%rbx
+.byte	102,72,15,126,215
+	call	__ecp_nistz256_mul_montq
+	call	__ecp_nistz256_mul_by_2q
+
+	movq	96+0(%rsp),%r12
+	movq	96+8(%rsp),%r13
+	leaq	64(%rsp),%rbx
+	movq	96+16(%rsp),%r8
+	movq	96+24(%rsp),%r9
+	leaq	32(%rsp),%rdi
+	call	__ecp_nistz256_add_toq
+
+	movq	96+0(%rsp),%r12
+	movq	96+8(%rsp),%r13
+	leaq	64(%rsp),%rbx
+	movq	96+16(%rsp),%r8
+	movq	96+24(%rsp),%r9
+	leaq	64(%rsp),%rdi
+	call	__ecp_nistz256_sub_fromq
+
+	movq	0+0(%rsp),%rax
+	movq	8+0(%rsp),%r14
+	leaq	0+0(%rsp),%rsi
+	movq	16+0(%rsp),%r15
+	movq	24+0(%rsp),%r8
+.byte	102,72,15,126,207
+	call	__ecp_nistz256_sqr_montq
+	xorq	%r9,%r9
+	movq	%r12,%rax
+	addq	$-1,%r12
+	movq	%r13,%r10
+	adcq	%rsi,%r13
+	movq	%r14,%rcx
+	adcq	$0,%r14
+	movq	%r15,%r8
+	adcq	%rbp,%r15
+	adcq	$0,%r9
+	xorq	%rsi,%rsi
+	testq	$1,%rax
+
+	cmovzq	%rax,%r12
+	cmovzq	%r10,%r13
+	cmovzq	%rcx,%r14
+	cmovzq	%r8,%r15
+	cmovzq	%rsi,%r9
+
+	movq	%r13,%rax
+	shrq	$1,%r12
+	shlq	$63,%rax
+	movq	%r14,%r10
+	shrq	$1,%r13
+	orq	%rax,%r12
+	shlq	$63,%r10
+	movq	%r15,%rcx
+	shrq	$1,%r14
+	orq	%r10,%r13
+	shlq	$63,%rcx
+	movq	%r12,0(%rdi)
+	shrq	$1,%r15
+	movq	%r13,8(%rdi)
+	shlq	$63,%r9
+	orq	%rcx,%r14
+	orq	%r9,%r15
+	movq	%r14,16(%rdi)
+	movq	%r15,24(%rdi)
+	movq	64(%rsp),%rax
+	leaq	64(%rsp),%rbx
+	movq	0+32(%rsp),%r9
+	movq	8+32(%rsp),%r10
+	leaq	0+32(%rsp),%rsi
+	movq	16+32(%rsp),%r11
+	movq	24+32(%rsp),%r12
+	leaq	32(%rsp),%rdi
+	call	__ecp_nistz256_mul_montq
+
+	leaq	128(%rsp),%rdi
+	call	__ecp_nistz256_mul_by_2q
+
+	leaq	32(%rsp),%rbx
+	leaq	32(%rsp),%rdi
+	call	__ecp_nistz256_add_toq
+
+	movq	96(%rsp),%rax
+	leaq	96(%rsp),%rbx
+	movq	0+0(%rsp),%r9
+	movq	8+0(%rsp),%r10
+	leaq	0+0(%rsp),%rsi
+	movq	16+0(%rsp),%r11
+	movq	24+0(%rsp),%r12
+	leaq	0(%rsp),%rdi
+	call	__ecp_nistz256_mul_montq
+
+	leaq	128(%rsp),%rdi
+	call	__ecp_nistz256_mul_by_2q
+
+	movq	0+32(%rsp),%rax
+	movq	8+32(%rsp),%r14
+	leaq	0+32(%rsp),%rsi
+	movq	16+32(%rsp),%r15
+	movq	24+32(%rsp),%r8
+.byte	102,72,15,126,199
+	call	__ecp_nistz256_sqr_montq
+
+	leaq	128(%rsp),%rbx
+	movq	%r14,%r8
+	movq	%r15,%r9
+	movq	%rsi,%r14
+	movq	%rbp,%r15
+	call	__ecp_nistz256_sub_fromq
+
+	movq	0+0(%rsp),%rax
+	movq	0+8(%rsp),%rbp
+	movq	0+16(%rsp),%rcx
+	movq	0+24(%rsp),%r10
+	leaq	0(%rsp),%rdi
+	call	__ecp_nistz256_subq
+
+	movq	32(%rsp),%rax
+	leaq	32(%rsp),%rbx
+	movq	%r12,%r14
+	xorl	%ecx,%ecx
+	movq	%r12,0+0(%rsp)
+	movq	%r13,%r10
+	movq	%r13,0+8(%rsp)
+	cmovzq	%r8,%r11
+	movq	%r8,0+16(%rsp)
+	leaq	0-0(%rsp),%rsi
+	cmovzq	%r9,%r12
+	movq	%r9,0+24(%rsp)
+	movq	%r14,%r9
+	leaq	0(%rsp),%rdi
+	call	__ecp_nistz256_mul_montq
+
+.byte	102,72,15,126,203
+.byte	102,72,15,126,207
+	call	__ecp_nistz256_sub_fromq
+
+	addq	$160+8,%rsp
+	popq	%r15
+	popq	%r14
+	popq	%r13
+	popq	%r12
+	popq	%rbx
+	popq	%rbp
+	.byte	0xf3,0xc3
+.size	ecp_nistz256_point_double,.-ecp_nistz256_point_double
+.globl	ecp_nistz256_point_add
+.hidden ecp_nistz256_point_add
+.type	ecp_nistz256_point_add,@function
+.align	32
+ecp_nistz256_point_add:
+	pushq	%rbp
+	pushq	%rbx
+	pushq	%r12
+	pushq	%r13
+	pushq	%r14
+	pushq	%r15
+	subq	$576+8,%rsp
+
+	movdqu	0(%rsi),%xmm0
+	movdqu	16(%rsi),%xmm1
+	movdqu	32(%rsi),%xmm2
+	movdqu	48(%rsi),%xmm3
+	movdqu	64(%rsi),%xmm4
+	movdqu	80(%rsi),%xmm5
+	movq	%rsi,%rbx
+	movq	%rdx,%rsi
+	movdqa	%xmm0,384(%rsp)
+	movdqa	%xmm1,384+16(%rsp)
+	por	%xmm0,%xmm1
+	movdqa	%xmm2,416(%rsp)
+	movdqa	%xmm3,416+16(%rsp)
+	por	%xmm2,%xmm3
+	movdqa	%xmm4,448(%rsp)
+	movdqa	%xmm5,448+16(%rsp)
+	por	%xmm1,%xmm3
+
+	movdqu	0(%rsi),%xmm0
+	pshufd	$0xb1,%xmm3,%xmm5
+	movdqu	16(%rsi),%xmm1
+	movdqu	32(%rsi),%xmm2
+	por	%xmm3,%xmm5
+	movdqu	48(%rsi),%xmm3
+	movq	64+0(%rsi),%rax
+	movq	64+8(%rsi),%r14
+	movq	64+16(%rsi),%r15
+	movq	64+24(%rsi),%r8
+	movdqa	%xmm0,480(%rsp)
+	pshufd	$0x1e,%xmm5,%xmm4
+	movdqa	%xmm1,480+16(%rsp)
+	por	%xmm0,%xmm1
+.byte	102,72,15,110,199
+	movdqa	%xmm2,512(%rsp)
+	movdqa	%xmm3,512+16(%rsp)
+	por	%xmm2,%xmm3
+	por	%xmm4,%xmm5
+	pxor	%xmm4,%xmm4
+	por	%xmm1,%xmm3
+
+	leaq	64-0(%rsi),%rsi
+	movq	%rax,544+0(%rsp)
+	movq	%r14,544+8(%rsp)
+	movq	%r15,544+16(%rsp)
+	movq	%r8,544+24(%rsp)
+	leaq	96(%rsp),%rdi
+	call	__ecp_nistz256_sqr_montq
+
+	pcmpeqd	%xmm4,%xmm5
+	pshufd	$0xb1,%xmm3,%xmm4
+	por	%xmm3,%xmm4
+	pshufd	$0,%xmm5,%xmm5
+	pshufd	$0x1e,%xmm4,%xmm3
+	por	%xmm3,%xmm4
+	pxor	%xmm3,%xmm3
+	pcmpeqd	%xmm3,%xmm4
+	pshufd	$0,%xmm4,%xmm4
+	movq	64+0(%rbx),%rax
+	movq	64+8(%rbx),%r14
+	movq	64+16(%rbx),%r15
+	movq	64+24(%rbx),%r8
+.byte	102,72,15,110,203
+
+	leaq	64-0(%rbx),%rsi
+	leaq	32(%rsp),%rdi
+	call	__ecp_nistz256_sqr_montq
+
+	movq	544(%rsp),%rax
+	leaq	544(%rsp),%rbx
+	movq	0+96(%rsp),%r9
+	movq	8+96(%rsp),%r10
+	leaq	0+96(%rsp),%rsi
+	movq	16+96(%rsp),%r11
+	movq	24+96(%rsp),%r12
+	leaq	224(%rsp),%rdi
+	call	__ecp_nistz256_mul_montq
+
+	movq	448(%rsp),%rax
+	leaq	448(%rsp),%rbx
+	movq	0+32(%rsp),%r9
+	movq	8+32(%rsp),%r10
+	leaq	0+32(%rsp),%rsi
+	movq	16+32(%rsp),%r11
+	movq	24+32(%rsp),%r12
+	leaq	256(%rsp),%rdi
+	call	__ecp_nistz256_mul_montq
+
+	movq	416(%rsp),%rax
+	leaq	416(%rsp),%rbx
+	movq	0+224(%rsp),%r9
+	movq	8+224(%rsp),%r10
+	leaq	0+224(%rsp),%rsi
+	movq	16+224(%rsp),%r11
+	movq	24+224(%rsp),%r12
+	leaq	224(%rsp),%rdi
+	call	__ecp_nistz256_mul_montq
+
+	movq	512(%rsp),%rax
+	leaq	512(%rsp),%rbx
+	movq	0+256(%rsp),%r9
+	movq	8+256(%rsp),%r10
+	leaq	0+256(%rsp),%rsi
+	movq	16+256(%rsp),%r11
+	movq	24+256(%rsp),%r12
+	leaq	256(%rsp),%rdi
+	call	__ecp_nistz256_mul_montq
+
+	leaq	224(%rsp),%rbx
+	leaq	64(%rsp),%rdi
+	call	__ecp_nistz256_sub_fromq
+
+	orq	%r13,%r12
+	movdqa	%xmm4,%xmm2
+	orq	%r8,%r12
+	orq	%r9,%r12
+	por	%xmm5,%xmm2
+.byte	102,73,15,110,220
+
+	movq	384(%rsp),%rax
+	leaq	384(%rsp),%rbx
+	movq	0+96(%rsp),%r9
+	movq	8+96(%rsp),%r10
+	leaq	0+96(%rsp),%rsi
+	movq	16+96(%rsp),%r11
+	movq	24+96(%rsp),%r12
+	leaq	160(%rsp),%rdi
+	call	__ecp_nistz256_mul_montq
+
+	movq	480(%rsp),%rax
+	leaq	480(%rsp),%rbx
+	movq	0+32(%rsp),%r9
+	movq	8+32(%rsp),%r10
+	leaq	0+32(%rsp),%rsi
+	movq	16+32(%rsp),%r11
+	movq	24+32(%rsp),%r12
+	leaq	192(%rsp),%rdi
+	call	__ecp_nistz256_mul_montq
+
+	leaq	160(%rsp),%rbx
+	leaq	0(%rsp),%rdi
+	call	__ecp_nistz256_sub_fromq
+
+	orq	%r13,%r12
+	orq	%r8,%r12
+	orq	%r9,%r12
+
+.byte	0x3e
+	jnz	.Ladd_proceedq
+.byte	102,73,15,126,208
+.byte	102,73,15,126,217
+	testq	%r8,%r8
+	jnz	.Ladd_proceedq
+	testq	%r9,%r9
+	jz	.Ladd_doubleq
+
+.byte	102,72,15,126,199
+	pxor	%xmm0,%xmm0
+	movdqu	%xmm0,0(%rdi)
+	movdqu	%xmm0,16(%rdi)
+	movdqu	%xmm0,32(%rdi)
+	movdqu	%xmm0,48(%rdi)
+	movdqu	%xmm0,64(%rdi)
+	movdqu	%xmm0,80(%rdi)
+	jmp	.Ladd_doneq
+
+.align	32
+.Ladd_doubleq:
+.byte	102,72,15,126,206
+.byte	102,72,15,126,199
+	addq	$416,%rsp
+	jmp	.Lpoint_double_shortcutq
+
+.align	32
+.Ladd_proceedq:
+	movq	0+64(%rsp),%rax
+	movq	8+64(%rsp),%r14
+	leaq	0+64(%rsp),%rsi
+	movq	16+64(%rsp),%r15
+	movq	24+64(%rsp),%r8
+	leaq	96(%rsp),%rdi
+	call	__ecp_nistz256_sqr_montq
+
+	movq	448(%rsp),%rax
+	leaq	448(%rsp),%rbx
+	movq	0+0(%rsp),%r9
+	movq	8+0(%rsp),%r10
+	leaq	0+0(%rsp),%rsi
+	movq	16+0(%rsp),%r11
+	movq	24+0(%rsp),%r12
+	leaq	352(%rsp),%rdi
+	call	__ecp_nistz256_mul_montq
+
+	movq	0+0(%rsp),%rax
+	movq	8+0(%rsp),%r14
+	leaq	0+0(%rsp),%rsi
+	movq	16+0(%rsp),%r15
+	movq	24+0(%rsp),%r8
+	leaq	32(%rsp),%rdi
+	call	__ecp_nistz256_sqr_montq
+
+	movq	544(%rsp),%rax
+	leaq	544(%rsp),%rbx
+	movq	0+352(%rsp),%r9
+	movq	8+352(%rsp),%r10
+	leaq	0+352(%rsp),%rsi
+	movq	16+352(%rsp),%r11
+	movq	24+352(%rsp),%r12
+	leaq	352(%rsp),%rdi
+	call	__ecp_nistz256_mul_montq
+
+	movq	0(%rsp),%rax
+	leaq	0(%rsp),%rbx
+	movq	0+32(%rsp),%r9
+	movq	8+32(%rsp),%r10
+	leaq	0+32(%rsp),%rsi
+	movq	16+32(%rsp),%r11
+	movq	24+32(%rsp),%r12
+	leaq	128(%rsp),%rdi
+	call	__ecp_nistz256_mul_montq
+
+	movq	160(%rsp),%rax
+	leaq	160(%rsp),%rbx
+	movq	0+32(%rsp),%r9
+	movq	8+32(%rsp),%r10
+	leaq	0+32(%rsp),%rsi
+	movq	16+32(%rsp),%r11
+	movq	24+32(%rsp),%r12
+	leaq	192(%rsp),%rdi
+	call	__ecp_nistz256_mul_montq
+
+
+
+
+	addq	%r12,%r12
+	leaq	96(%rsp),%rsi
+	adcq	%r13,%r13
+	movq	%r12,%rax
+	adcq	%r8,%r8
+	adcq	%r9,%r9
+	movq	%r13,%rbp
+	sbbq	%r11,%r11
+
+	subq	$-1,%r12
+	movq	%r8,%rcx
+	sbbq	%r14,%r13
+	sbbq	$0,%r8
+	movq	%r9,%r10
+	sbbq	%r15,%r9
+	testq	%r11,%r11
+
+	cmovzq	%rax,%r12
+	movq	0(%rsi),%rax
+	cmovzq	%rbp,%r13
+	movq	8(%rsi),%rbp
+	cmovzq	%rcx,%r8
+	movq	16(%rsi),%rcx
+	cmovzq	%r10,%r9
+	movq	24(%rsi),%r10
+
+	call	__ecp_nistz256_subq
+
+	leaq	128(%rsp),%rbx
+	leaq	288(%rsp),%rdi
+	call	__ecp_nistz256_sub_fromq
+
+	movq	192+0(%rsp),%rax
+	movq	192+8(%rsp),%rbp
+	movq	192+16(%rsp),%rcx
+	movq	192+24(%rsp),%r10
+	leaq	320(%rsp),%rdi
+
+	call	__ecp_nistz256_subq
+
+	movq	%r12,0(%rdi)
+	movq	%r13,8(%rdi)
+	movq	%r8,16(%rdi)
+	movq	%r9,24(%rdi)
+	movq	128(%rsp),%rax
+	leaq	128(%rsp),%rbx
+	movq	0+224(%rsp),%r9
+	movq	8+224(%rsp),%r10
+	leaq	0+224(%rsp),%rsi
+	movq	16+224(%rsp),%r11
+	movq	24+224(%rsp),%r12
+	leaq	256(%rsp),%rdi
+	call	__ecp_nistz256_mul_montq
+
+	movq	320(%rsp),%rax
+	leaq	320(%rsp),%rbx
+	movq	0+64(%rsp),%r9
+	movq	8+64(%rsp),%r10
+	leaq	0+64(%rsp),%rsi
+	movq	16+64(%rsp),%r11
+	movq	24+64(%rsp),%r12
+	leaq	320(%rsp),%rdi
+	call	__ecp_nistz256_mul_montq
+
+	leaq	256(%rsp),%rbx
+	leaq	320(%rsp),%rdi
+	call	__ecp_nistz256_sub_fromq
+
+.byte	102,72,15,126,199
+
+	movdqa	%xmm5,%xmm0
+	movdqa	%xmm5,%xmm1
+	pandn	352(%rsp),%xmm0
+	movdqa	%xmm5,%xmm2
+	pandn	352+16(%rsp),%xmm1
+	movdqa	%xmm5,%xmm3
+	pand	544(%rsp),%xmm2
+	pand	544+16(%rsp),%xmm3
+	por	%xmm0,%xmm2
+	por	%xmm1,%xmm3
+
+	movdqa	%xmm4,%xmm0
+	movdqa	%xmm4,%xmm1
+	pandn	%xmm2,%xmm0
+	movdqa	%xmm4,%xmm2
+	pandn	%xmm3,%xmm1
+	movdqa	%xmm4,%xmm3
+	pand	448(%rsp),%xmm2
+	pand	448+16(%rsp),%xmm3
+	por	%xmm0,%xmm2
+	por	%xmm1,%xmm3
+	movdqu	%xmm2,64(%rdi)
+	movdqu	%xmm3,80(%rdi)
+
+	movdqa	%xmm5,%xmm0
+	movdqa	%xmm5,%xmm1
+	pandn	288(%rsp),%xmm0
+	movdqa	%xmm5,%xmm2
+	pandn	288+16(%rsp),%xmm1
+	movdqa	%xmm5,%xmm3
+	pand	480(%rsp),%xmm2
+	pand	480+16(%rsp),%xmm3
+	por	%xmm0,%xmm2
+	por	%xmm1,%xmm3
+
+	movdqa	%xmm4,%xmm0
+	movdqa	%xmm4,%xmm1
+	pandn	%xmm2,%xmm0
+	movdqa	%xmm4,%xmm2
+	pandn	%xmm3,%xmm1
+	movdqa	%xmm4,%xmm3
+	pand	384(%rsp),%xmm2
+	pand	384+16(%rsp),%xmm3
+	por	%xmm0,%xmm2
+	por	%xmm1,%xmm3
+	movdqu	%xmm2,0(%rdi)
+	movdqu	%xmm3,16(%rdi)
+
+	movdqa	%xmm5,%xmm0
+	movdqa	%xmm5,%xmm1
+	pandn	320(%rsp),%xmm0
+	movdqa	%xmm5,%xmm2
+	pandn	320+16(%rsp),%xmm1
+	movdqa	%xmm5,%xmm3
+	pand	512(%rsp),%xmm2
+	pand	512+16(%rsp),%xmm3
+	por	%xmm0,%xmm2
+	por	%xmm1,%xmm3
+
+	movdqa	%xmm4,%xmm0
+	movdqa	%xmm4,%xmm1
+	pandn	%xmm2,%xmm0
+	movdqa	%xmm4,%xmm2
+	pandn	%xmm3,%xmm1
+	movdqa	%xmm4,%xmm3
+	pand	416(%rsp),%xmm2
+	pand	416+16(%rsp),%xmm3
+	por	%xmm0,%xmm2
+	por	%xmm1,%xmm3
+	movdqu	%xmm2,32(%rdi)
+	movdqu	%xmm3,48(%rdi)
+
+.Ladd_doneq:
+	addq	$576+8,%rsp
+	popq	%r15
+	popq	%r14
+	popq	%r13
+	popq	%r12
+	popq	%rbx
+	popq	%rbp
+	.byte	0xf3,0xc3
+.size	ecp_nistz256_point_add,.-ecp_nistz256_point_add
+.globl	ecp_nistz256_point_add_affine
+.hidden ecp_nistz256_point_add_affine
+.type	ecp_nistz256_point_add_affine,@function
+.align	32
+ecp_nistz256_point_add_affine:
+	pushq	%rbp
+	pushq	%rbx
+	pushq	%r12
+	pushq	%r13
+	pushq	%r14
+	pushq	%r15
+	subq	$480+8,%rsp
+
+	movdqu	0(%rsi),%xmm0
+	movq	%rdx,%rbx
+	movdqu	16(%rsi),%xmm1
+	movdqu	32(%rsi),%xmm2
+	movdqu	48(%rsi),%xmm3
+	movdqu	64(%rsi),%xmm4
+	movdqu	80(%rsi),%xmm5
+	movq	64+0(%rsi),%rax
+	movq	64+8(%rsi),%r14
+	movq	64+16(%rsi),%r15
+	movq	64+24(%rsi),%r8
+	movdqa	%xmm0,320(%rsp)
+	movdqa	%xmm1,320+16(%rsp)
+	por	%xmm0,%xmm1
+	movdqa	%xmm2,352(%rsp)
+	movdqa	%xmm3,352+16(%rsp)
+	por	%xmm2,%xmm3
+	movdqa	%xmm4,384(%rsp)
+	movdqa	%xmm5,384+16(%rsp)
+	por	%xmm1,%xmm3
+
+	movdqu	0(%rbx),%xmm0
+	pshufd	$0xb1,%xmm3,%xmm5
+	movdqu	16(%rbx),%xmm1
+	movdqu	32(%rbx),%xmm2
+	por	%xmm3,%xmm5
+	movdqu	48(%rbx),%xmm3
+	movdqa	%xmm0,416(%rsp)
+	pshufd	$0x1e,%xmm5,%xmm4
+	movdqa	%xmm1,416+16(%rsp)
+	por	%xmm0,%xmm1
+.byte	102,72,15,110,199
+	movdqa	%xmm2,448(%rsp)
+	movdqa	%xmm3,448+16(%rsp)
+	por	%xmm2,%xmm3
+	por	%xmm4,%xmm5
+	pxor	%xmm4,%xmm4
+	por	%xmm1,%xmm3
+
+	leaq	64-0(%rsi),%rsi
+	leaq	32(%rsp),%rdi
+	call	__ecp_nistz256_sqr_montq
+
+	pcmpeqd	%xmm4,%xmm5
+	pshufd	$0xb1,%xmm3,%xmm4
+	movq	0(%rbx),%rax
+
+	movq	%r12,%r9
+	por	%xmm3,%xmm4
+	pshufd	$0,%xmm5,%xmm5
+	pshufd	$0x1e,%xmm4,%xmm3
+	movq	%r13,%r10
+	por	%xmm3,%xmm4
+	pxor	%xmm3,%xmm3
+	movq	%r14,%r11
+	pcmpeqd	%xmm3,%xmm4
+	pshufd	$0,%xmm4,%xmm4
+
+	leaq	32-0(%rsp),%rsi
+	movq	%r15,%r12
+	leaq	0(%rsp),%rdi
+	call	__ecp_nistz256_mul_montq
+
+	leaq	320(%rsp),%rbx
+	leaq	64(%rsp),%rdi
+	call	__ecp_nistz256_sub_fromq
+
+	movq	384(%rsp),%rax
+	leaq	384(%rsp),%rbx
+	movq	0+32(%rsp),%r9
+	movq	8+32(%rsp),%r10
+	leaq	0+32(%rsp),%rsi
+	movq	16+32(%rsp),%r11
+	movq	24+32(%rsp),%r12
+	leaq	32(%rsp),%rdi
+	call	__ecp_nistz256_mul_montq
+
+	movq	384(%rsp),%rax
+	leaq	384(%rsp),%rbx
+	movq	0+64(%rsp),%r9
+	movq	8+64(%rsp),%r10
+	leaq	0+64(%rsp),%rsi
+	movq	16+64(%rsp),%r11
+	movq	24+64(%rsp),%r12
+	leaq	288(%rsp),%rdi
+	call	__ecp_nistz256_mul_montq
+
+	movq	448(%rsp),%rax
+	leaq	448(%rsp),%rbx
+	movq	0+32(%rsp),%r9
+	movq	8+32(%rsp),%r10
+	leaq	0+32(%rsp),%rsi
+	movq	16+32(%rsp),%r11
+	movq	24+32(%rsp),%r12
+	leaq	32(%rsp),%rdi
+	call	__ecp_nistz256_mul_montq
+
+	leaq	352(%rsp),%rbx
+	leaq	96(%rsp),%rdi
+	call	__ecp_nistz256_sub_fromq
+
+	movq	0+64(%rsp),%rax
+	movq	8+64(%rsp),%r14
+	leaq	0+64(%rsp),%rsi
+	movq	16+64(%rsp),%r15
+	movq	24+64(%rsp),%r8
+	leaq	128(%rsp),%rdi
+	call	__ecp_nistz256_sqr_montq
+
+	movq	0+96(%rsp),%rax
+	movq	8+96(%rsp),%r14
+	leaq	0+96(%rsp),%rsi
+	movq	16+96(%rsp),%r15
+	movq	24+96(%rsp),%r8
+	leaq	192(%rsp),%rdi
+	call	__ecp_nistz256_sqr_montq
+
+	movq	128(%rsp),%rax
+	leaq	128(%rsp),%rbx
+	movq	0+64(%rsp),%r9
+	movq	8+64(%rsp),%r10
+	leaq	0+64(%rsp),%rsi
+	movq	16+64(%rsp),%r11
+	movq	24+64(%rsp),%r12
+	leaq	160(%rsp),%rdi
+	call	__ecp_nistz256_mul_montq
+
+	movq	320(%rsp),%rax
+	leaq	320(%rsp),%rbx
+	movq	0+128(%rsp),%r9
+	movq	8+128(%rsp),%r10
+	leaq	0+128(%rsp),%rsi
+	movq	16+128(%rsp),%r11
+	movq	24+128(%rsp),%r12
+	leaq	0(%rsp),%rdi
+	call	__ecp_nistz256_mul_montq
+
+
+
+
+	addq	%r12,%r12
+	leaq	192(%rsp),%rsi
+	adcq	%r13,%r13
+	movq	%r12,%rax
+	adcq	%r8,%r8
+	adcq	%r9,%r9
+	movq	%r13,%rbp
+	sbbq	%r11,%r11
+
+	subq	$-1,%r12
+	movq	%r8,%rcx
+	sbbq	%r14,%r13
+	sbbq	$0,%r8
+	movq	%r9,%r10
+	sbbq	%r15,%r9
+	testq	%r11,%r11
+
+	cmovzq	%rax,%r12
+	movq	0(%rsi),%rax
+	cmovzq	%rbp,%r13
+	movq	8(%rsi),%rbp
+	cmovzq	%rcx,%r8
+	movq	16(%rsi),%rcx
+	cmovzq	%r10,%r9
+	movq	24(%rsi),%r10
+
+	call	__ecp_nistz256_subq
+
+	leaq	160(%rsp),%rbx
+	leaq	224(%rsp),%rdi
+	call	__ecp_nistz256_sub_fromq
+
+	movq	0+0(%rsp),%rax
+	movq	0+8(%rsp),%rbp
+	movq	0+16(%rsp),%rcx
+	movq	0+24(%rsp),%r10
+	leaq	64(%rsp),%rdi
+
+	call	__ecp_nistz256_subq
+
+	movq	%r12,0(%rdi)
+	movq	%r13,8(%rdi)
+	movq	%r8,16(%rdi)
+	movq	%r9,24(%rdi)
+	movq	352(%rsp),%rax
+	leaq	352(%rsp),%rbx
+	movq	0+160(%rsp),%r9
+	movq	8+160(%rsp),%r10
+	leaq	0+160(%rsp),%rsi
+	movq	16+160(%rsp),%r11
+	movq	24+160(%rsp),%r12
+	leaq	32(%rsp),%rdi
+	call	__ecp_nistz256_mul_montq
+
+	movq	96(%rsp),%rax
+	leaq	96(%rsp),%rbx
+	movq	0+64(%rsp),%r9
+	movq	8+64(%rsp),%r10
+	leaq	0+64(%rsp),%rsi
+	movq	16+64(%rsp),%r11
+	movq	24+64(%rsp),%r12
+	leaq	64(%rsp),%rdi
+	call	__ecp_nistz256_mul_montq
+
+	leaq	32(%rsp),%rbx
+	leaq	256(%rsp),%rdi
+	call	__ecp_nistz256_sub_fromq
+
+.byte	102,72,15,126,199
+
+	movdqa	%xmm5,%xmm0
+	movdqa	%xmm5,%xmm1
+	pandn	288(%rsp),%xmm0
+	movdqa	%xmm5,%xmm2
+	pandn	288+16(%rsp),%xmm1
+	movdqa	%xmm5,%xmm3
+	pand	.LONE_mont(%rip),%xmm2
+	pand	.LONE_mont+16(%rip),%xmm3
+	por	%xmm0,%xmm2
+	por	%xmm1,%xmm3
+
+	movdqa	%xmm4,%xmm0
+	movdqa	%xmm4,%xmm1
+	pandn	%xmm2,%xmm0
+	movdqa	%xmm4,%xmm2
+	pandn	%xmm3,%xmm1
+	movdqa	%xmm4,%xmm3
+	pand	384(%rsp),%xmm2
+	pand	384+16(%rsp),%xmm3
+	por	%xmm0,%xmm2
+	por	%xmm1,%xmm3
+	movdqu	%xmm2,64(%rdi)
+	movdqu	%xmm3,80(%rdi)
+
+	movdqa	%xmm5,%xmm0
+	movdqa	%xmm5,%xmm1
+	pandn	224(%rsp),%xmm0
+	movdqa	%xmm5,%xmm2
+	pandn	224+16(%rsp),%xmm1
+	movdqa	%xmm5,%xmm3
+	pand	416(%rsp),%xmm2
+	pand	416+16(%rsp),%xmm3
+	por	%xmm0,%xmm2
+	por	%xmm1,%xmm3
+
+	movdqa	%xmm4,%xmm0
+	movdqa	%xmm4,%xmm1
+	pandn	%xmm2,%xmm0
+	movdqa	%xmm4,%xmm2
+	pandn	%xmm3,%xmm1
+	movdqa	%xmm4,%xmm3
+	pand	320(%rsp),%xmm2
+	pand	320+16(%rsp),%xmm3
+	por	%xmm0,%xmm2
+	por	%xmm1,%xmm3
+	movdqu	%xmm2,0(%rdi)
+	movdqu	%xmm3,16(%rdi)
+
+	movdqa	%xmm5,%xmm0
+	movdqa	%xmm5,%xmm1
+	pandn	256(%rsp),%xmm0
+	movdqa	%xmm5,%xmm2
+	pandn	256+16(%rsp),%xmm1
+	movdqa	%xmm5,%xmm3
+	pand	448(%rsp),%xmm2
+	pand	448+16(%rsp),%xmm3
+	por	%xmm0,%xmm2
+	por	%xmm1,%xmm3
+
+	movdqa	%xmm4,%xmm0
+	movdqa	%xmm4,%xmm1
+	pandn	%xmm2,%xmm0
+	movdqa	%xmm4,%xmm2
+	pandn	%xmm3,%xmm1
+	movdqa	%xmm4,%xmm3
+	pand	352(%rsp),%xmm2
+	pand	352+16(%rsp),%xmm3
+	por	%xmm0,%xmm2
+	por	%xmm1,%xmm3
+	movdqu	%xmm2,32(%rdi)
+	movdqu	%xmm3,48(%rdi)
+
+	addq	$480+8,%rsp
+	popq	%r15
+	popq	%r14
+	popq	%r13
+	popq	%r12
+	popq	%rbx
+	popq	%rbp
+	.byte	0xf3,0xc3
+.size	ecp_nistz256_point_add_affine,.-ecp_nistz256_point_add_affine
+#endif

third_party/boringssl/linux-x86_64/crypto/md5/md5-x86_64.S

diff --git a/third_party/boringssl/linux-x86_64/crypto/md5/md5-x86_64.S b/third_party/boringssl/linux-x86_64/crypto/md5/md5-x86_64.S
index 7644689..05369e2 100644
--- a/third_party/boringssl/linux-x86_64/crypto/md5/md5-x86_64.S
+++ b/third_party/boringssl/linux-x86_64/crypto/md5/md5-x86_64.S
@@ -495,14 +495,14 @@
 	movl	%ecx,%r11d
 	addl	%ecx,%ebx
 	movl	0(%rsi),%r10d
-	movl	$4294967295,%r11d
+	movl	$0xffffffff,%r11d
 	xorl	%edx,%r11d
 	leal	-198630844(%rax,%r10,1),%eax
 	orl	%ebx,%r11d
 	xorl	%ecx,%r11d
 	addl	%r11d,%eax
 	movl	28(%rsi),%r10d
-	movl	$4294967295,%r11d
+	movl	$0xffffffff,%r11d
 	roll	$6,%eax
 	xorl	%ecx,%r11d
 	addl	%ebx,%eax
@@ -511,7 +511,7 @@
 	xorl	%ebx,%r11d
 	addl	%r11d,%edx
 	movl	56(%rsi),%r10d
-	movl	$4294967295,%r11d
+	movl	$0xffffffff,%r11d
 	roll	$10,%edx
 	xorl	%ebx,%r11d
 	addl	%eax,%edx
@@ -520,7 +520,7 @@
 	xorl	%eax,%r11d
 	addl	%r11d,%ecx
 	movl	20(%rsi),%r10d
-	movl	$4294967295,%r11d
+	movl	$0xffffffff,%r11d
 	roll	$15,%ecx
 	xorl	%eax,%r11d
 	addl	%edx,%ecx
@@ -529,7 +529,7 @@
 	xorl	%edx,%r11d
 	addl	%r11d,%ebx
 	movl	48(%rsi),%r10d
-	movl	$4294967295,%r11d
+	movl	$0xffffffff,%r11d
 	roll	$21,%ebx
 	xorl	%edx,%r11d
 	addl	%ecx,%ebx
@@ -538,7 +538,7 @@
 	xorl	%ecx,%r11d
 	addl	%r11d,%eax
 	movl	12(%rsi),%r10d
-	movl	$4294967295,%r11d
+	movl	$0xffffffff,%r11d
 	roll	$6,%eax
 	xorl	%ecx,%r11d
 	addl	%ebx,%eax
@@ -547,7 +547,7 @@
 	xorl	%ebx,%r11d
 	addl	%r11d,%edx
 	movl	40(%rsi),%r10d
-	movl	$4294967295,%r11d
+	movl	$0xffffffff,%r11d
 	roll	$10,%edx
 	xorl	%ebx,%r11d
 	addl	%eax,%edx
@@ -556,7 +556,7 @@
 	xorl	%eax,%r11d
 	addl	%r11d,%ecx
 	movl	4(%rsi),%r10d
-	movl	$4294967295,%r11d
+	movl	$0xffffffff,%r11d
 	roll	$15,%ecx
 	xorl	%eax,%r11d
 	addl	%edx,%ecx
@@ -565,7 +565,7 @@
 	xorl	%edx,%r11d
 	addl	%r11d,%ebx
 	movl	32(%rsi),%r10d
-	movl	$4294967295,%r11d
+	movl	$0xffffffff,%r11d
 	roll	$21,%ebx
 	xorl	%edx,%r11d
 	addl	%ecx,%ebx
@@ -574,7 +574,7 @@
 	xorl	%ecx,%r11d
 	addl	%r11d,%eax
 	movl	60(%rsi),%r10d
-	movl	$4294967295,%r11d
+	movl	$0xffffffff,%r11d
 	roll	$6,%eax
 	xorl	%ecx,%r11d
 	addl	%ebx,%eax
@@ -583,7 +583,7 @@
 	xorl	%ebx,%r11d
 	addl	%r11d,%edx
 	movl	24(%rsi),%r10d
-	movl	$4294967295,%r11d
+	movl	$0xffffffff,%r11d
 	roll	$10,%edx
 	xorl	%ebx,%r11d
 	addl	%eax,%edx
@@ -592,7 +592,7 @@
 	xorl	%eax,%r11d
 	addl	%r11d,%ecx
 	movl	52(%rsi),%r10d
-	movl	$4294967295,%r11d
+	movl	$0xffffffff,%r11d
 	roll	$15,%ecx
 	xorl	%eax,%r11d
 	addl	%edx,%ecx
@@ -601,7 +601,7 @@
 	xorl	%edx,%r11d
 	addl	%r11d,%ebx
 	movl	16(%rsi),%r10d
-	movl	$4294967295,%r11d
+	movl	$0xffffffff,%r11d
 	roll	$21,%ebx
 	xorl	%edx,%r11d
 	addl	%ecx,%ebx
@@ -610,7 +610,7 @@
 	xorl	%ecx,%r11d
 	addl	%r11d,%eax
 	movl	44(%rsi),%r10d
-	movl	$4294967295,%r11d
+	movl	$0xffffffff,%r11d
 	roll	$6,%eax
 	xorl	%ecx,%r11d
 	addl	%ebx,%eax
@@ -619,7 +619,7 @@
 	xorl	%ebx,%r11d
 	addl	%r11d,%edx
 	movl	8(%rsi),%r10d
-	movl	$4294967295,%r11d
+	movl	$0xffffffff,%r11d
 	roll	$10,%edx
 	xorl	%ebx,%r11d
 	addl	%eax,%edx
@@ -628,7 +628,7 @@
 	xorl	%eax,%r11d
 	addl	%r11d,%ecx
 	movl	36(%rsi),%r10d
-	movl	$4294967295,%r11d
+	movl	$0xffffffff,%r11d
 	roll	$15,%ecx
 	xorl	%eax,%r11d
 	addl	%edx,%ecx
@@ -637,7 +637,7 @@
 	xorl	%edx,%r11d
 	addl	%r11d,%ebx
 	movl	0(%rsi),%r10d
-	movl	$4294967295,%r11d
+	movl	$0xffffffff,%r11d
 	roll	$21,%ebx
 	xorl	%edx,%r11d
 	addl	%ecx,%ebx

third_party/boringssl/linux-x86_64/crypto/modes/ghash-x86_64.S

diff --git a/third_party/boringssl/linux-x86_64/crypto/modes/ghash-x86_64.S b/third_party/boringssl/linux-x86_64/crypto/modes/ghash-x86_64.S
index 1db7d69..b47bdc9b 100644
--- a/third_party/boringssl/linux-x86_64/crypto/modes/ghash-x86_64.S
+++ b/third_party/boringssl/linux-x86_64/crypto/modes/ghash-x86_64.S
@@ -23,14 +23,14 @@
 	movq	$14,%rcx
 	movq	8(%rsi,%rax,1),%r8
 	movq	(%rsi,%rax,1),%r9
-	andb	$240,%bl
+	andb	$0xf0,%bl
 	movq	%r8,%rdx
 	jmp	.Loop1
 
 .align	16
 .Loop1:
 	shrq	$4,%r8
-	andq	$15,%rdx
+	andq	$0xf,%rdx
 	movq	%r9,%r10
 	movb	(%rdi,%rcx,1),%al
 	shrq	$4,%r9
@@ -46,13 +46,13 @@
 	js	.Lbreak1
 
 	shrq	$4,%r8
-	andq	$15,%rdx
+	andq	$0xf,%rdx
 	movq	%r9,%r10
 	shrq	$4,%r9
 	xorq	8(%rsi,%rax,1),%r8
 	shlq	$60,%r10
 	xorq	(%rsi,%rax,1),%r9
-	andb	$240,%bl
+	andb	$0xf0,%bl
 	xorq	(%r11,%rdx,8),%r9
 	movq	%r8,%rdx
 	xorq	%r10,%r8
@@ -61,19 +61,19 @@
 .align	16
 .Lbreak1:
 	shrq	$4,%r8
-	andq	$15,%rdx
+	andq	$0xf,%rdx
 	movq	%r9,%r10
 	shrq	$4,%r9
 	xorq	8(%rsi,%rax,1),%r8
 	shlq	$60,%r10
 	xorq	(%rsi,%rax,1),%r9
-	andb	$240,%bl
+	andb	$0xf0,%bl
 	xorq	(%r11,%rdx,8),%r9
 	movq	%r8,%rdx
 	xorq	%r10,%r8
 
 	shrq	$4,%r8
-	andq	$15,%rdx
+	andq	$0xf,%rdx
 	movq	%r9,%r10
 	shrq	$4,%r9
 	xorq	8(%rsi,%rbx,1),%r8
@@ -881,20 +881,20 @@
 	movdqu	32(%rsi),%xmm7
 .byte	102,65,15,56,0,194
 
-	subq	$16,%rcx
+	subq	$0x10,%rcx
 	jz	.Lodd_tail
 
 	movdqu	16(%rsi),%xmm6
 	movl	OPENSSL_ia32cap_P+4(%rip),%eax
-	cmpq	$48,%rcx
+	cmpq	$0x30,%rcx
 	jb	.Lskip4x
 
 	andl	$71303168,%eax
 	cmpl	$4194304,%eax
 	je	.Lskip4x
 
-	subq	$48,%rcx
-	movq	$11547335547999543296,%rax
+	subq	$0x30,%rcx
+	movq	$0xA040608020C0E000,%rax
 	movdqu	48(%rsi),%xmm14
 	movdqu	64(%rsi),%xmm15
 
@@ -941,7 +941,7 @@
 	xorps	%xmm13,%xmm5
 
 	leaq	64(%rdx),%rdx
-	subq	$64,%rcx
+	subq	$0x40,%rcx
 	jc	.Ltail4x
 
 	jmp	.Lmod4_loop
@@ -1024,7 +1024,7 @@
 	xorps	%xmm13,%xmm5
 
 	leaq	64(%rdx),%rdx
-	subq	$64,%rcx
+	subq	$0x40,%rcx
 	jnc	.Lmod4_loop
 
 .Ltail4x:
@@ -1068,10 +1068,10 @@
 	pxor	%xmm4,%xmm0
 	psrlq	$1,%xmm0
 	pxor	%xmm1,%xmm0
-	addq	$64,%rcx
+	addq	$0x40,%rcx
 	jz	.Ldone
 	movdqu	32(%rsi),%xmm7
-	subq	$16,%rcx
+	subq	$0x10,%rcx
 	jz	.Lodd_tail
 .Lskip4x:
 
@@ -1094,7 +1094,7 @@
 
 	leaq	32(%rdx),%rdx
 	nop
-	subq	$32,%rcx
+	subq	$0x20,%rcx
 	jbe	.Leven_tail
 	nop
 	jmp	.Lmod_loop
@@ -1157,7 +1157,7 @@
 .byte	102,15,58,68,231,0
 	pxor	%xmm1,%xmm0
 
-	subq	$32,%rcx
+	subq	$0x20,%rcx
 	ja	.Lmod_loop
 
 .Leven_tail:

third_party/boringssl/linux-x86_64/crypto/rc4/rc4-md5-x86_64.S

diff --git a/third_party/boringssl/linux-x86_64/crypto/rc4/rc4-md5-x86_64.S b/third_party/boringssl/linux-x86_64/crypto/rc4/rc4-md5-x86_64.S
deleted file mode 100644
index 06c8d67..0000000
--- a/third_party/boringssl/linux-x86_64/crypto/rc4/rc4-md5-x86_64.S
+++ /dev/null
@@ -1,1262 +0,0 @@
-#if defined(__x86_64__)
-.text	
-.align	16
-
-.globl	rc4_md5_enc
-.hidden rc4_md5_enc
-.type	rc4_md5_enc,@function
-rc4_md5_enc:
-	cmpq	$0,%r9
-	je	.Labort
-	pushq	%rbx
-	pushq	%rbp
-	pushq	%r12
-	pushq	%r13
-	pushq	%r14
-	pushq	%r15
-	subq	$40,%rsp
-.Lbody:
-	movq	%rcx,%r11
-	movq	%r9,%r12
-	movq	%rsi,%r13
-	movq	%rdx,%r14
-	movq	%r8,%r15
-	xorq	%rbp,%rbp
-	xorq	%rcx,%rcx
-
-	leaq	8(%rdi),%rdi
-	movb	-8(%rdi),%bpl
-	movb	-4(%rdi),%cl
-
-	incb	%bpl
-	subq	%r13,%r14
-	movl	(%rdi,%rbp,4),%eax
-	addb	%al,%cl
-	leaq	(%rdi,%rbp,4),%rsi
-	shlq	$6,%r12
-	addq	%r15,%r12
-	movq	%r12,16(%rsp)
-
-	movq	%r11,24(%rsp)
-	movl	0(%r11),%r8d
-	movl	4(%r11),%r9d
-	movl	8(%r11),%r10d
-	movl	12(%r11),%r11d
-	jmp	.Loop
-
-.align	16
-.Loop:
-	movl	%r8d,0(%rsp)
-	movl	%r9d,4(%rsp)
-	movl	%r10d,8(%rsp)
-	movl	%r11d,%r12d
-	movl	%r11d,12(%rsp)
-	pxor	%xmm0,%xmm0
-	movl	(%rdi,%rcx,4),%edx
-	xorl	%r10d,%r12d
-	movl	%eax,(%rdi,%rcx,4)
-	andl	%r9d,%r12d
-	addl	0(%r15),%r8d
-	addb	%dl,%al
-	movl	4(%rsi),%ebx
-	addl	$3614090360,%r8d
-	xorl	%r11d,%r12d
-	movzbl	%al,%eax
-	movl	%edx,0(%rsi)
-	addl	%r12d,%r8d
-	addb	%bl,%cl
-	roll	$7,%r8d
-	movl	%r10d,%r12d
-	movd	(%rdi,%rax,4),%xmm0
-
-	addl	%r9d,%r8d
-	pxor	%xmm1,%xmm1
-	movl	(%rdi,%rcx,4),%edx
-	xorl	%r9d,%r12d
-	movl	%ebx,(%rdi,%rcx,4)
-	andl	%r8d,%r12d
-	addl	4(%r15),%r11d
-	addb	%dl,%bl
-	movl	8(%rsi),%eax
-	addl	$3905402710,%r11d
-	xorl	%r10d,%r12d
-	movzbl	%bl,%ebx
-	movl	%edx,4(%rsi)
-	addl	%r12d,%r11d
-	addb	%al,%cl
-	roll	$12,%r11d
-	movl	%r9d,%r12d
-	movd	(%rdi,%rbx,4),%xmm1
-
-	addl	%r8d,%r11d
-	movl	(%rdi,%rcx,4),%edx
-	xorl	%r8d,%r12d
-	movl	%eax,(%rdi,%rcx,4)
-	andl	%r11d,%r12d
-	addl	8(%r15),%r10d
-	addb	%dl,%al
-	movl	12(%rsi),%ebx
-	addl	$606105819,%r10d
-	xorl	%r9d,%r12d
-	movzbl	%al,%eax
-	movl	%edx,8(%rsi)
-	addl	%r12d,%r10d
-	addb	%bl,%cl
-	roll	$17,%r10d
-	movl	%r8d,%r12d
-	pinsrw	$1,(%rdi,%rax,4),%xmm0
-
-	addl	%r11d,%r10d
-	movl	(%rdi,%rcx,4),%edx
-	xorl	%r11d,%r12d
-	movl	%ebx,(%rdi,%rcx,4)
-	andl	%r10d,%r12d
-	addl	12(%r15),%r9d
-	addb	%dl,%bl
-	movl	16(%rsi),%eax
-	addl	$3250441966,%r9d
-	xorl	%r8d,%r12d
-	movzbl	%bl,%ebx
-	movl	%edx,12(%rsi)
-	addl	%r12d,%r9d
-	addb	%al,%cl
-	roll	$22,%r9d
-	movl	%r11d,%r12d
-	pinsrw	$1,(%rdi,%rbx,4),%xmm1
-
-	addl	%r10d,%r9d
-	movl	(%rdi,%rcx,4),%edx
-	xorl	%r10d,%r12d
-	movl	%eax,(%rdi,%rcx,4)
-	andl	%r9d,%r12d
-	addl	16(%r15),%r8d
-	addb	%dl,%al
-	movl	20(%rsi),%ebx
-	addl	$4118548399,%r8d
-	xorl	%r11d,%r12d
-	movzbl	%al,%eax
-	movl	%edx,16(%rsi)
-	addl	%r12d,%r8d
-	addb	%bl,%cl
-	roll	$7,%r8d
-	movl	%r10d,%r12d
-	pinsrw	$2,(%rdi,%rax,4),%xmm0
-
-	addl	%r9d,%r8d
-	movl	(%rdi,%rcx,4),%edx
-	xorl	%r9d,%r12d
-	movl	%ebx,(%rdi,%rcx,4)
-	andl	%r8d,%r12d
-	addl	20(%r15),%r11d
-	addb	%dl,%bl
-	movl	24(%rsi),%eax
-	addl	$1200080426,%r11d
-	xorl	%r10d,%r12d
-	movzbl	%bl,%ebx
-	movl	%edx,20(%rsi)
-	addl	%r12d,%r11d
-	addb	%al,%cl
-	roll	$12,%r11d
-	movl	%r9d,%r12d
-	pinsrw	$2,(%rdi,%rbx,4),%xmm1
-
-	addl	%r8d,%r11d
-	movl	(%rdi,%rcx,4),%edx
-	xorl	%r8d,%r12d
-	movl	%eax,(%rdi,%rcx,4)
-	andl	%r11d,%r12d
-	addl	24(%r15),%r10d
-	addb	%dl,%al
-	movl	28(%rsi),%ebx
-	addl	$2821735955,%r10d
-	xorl	%r9d,%r12d
-	movzbl	%al,%eax
-	movl	%edx,24(%rsi)
-	addl	%r12d,%r10d
-	addb	%bl,%cl
-	roll	$17,%r10d
-	movl	%r8d,%r12d
-	pinsrw	$3,(%rdi,%rax,4),%xmm0
-
-	addl	%r11d,%r10d
-	movl	(%rdi,%rcx,4),%edx
-	xorl	%r11d,%r12d
-	movl	%ebx,(%rdi,%rcx,4)
-	andl	%r10d,%r12d
-	addl	28(%r15),%r9d
-	addb	%dl,%bl
-	movl	32(%rsi),%eax
-	addl	$4249261313,%r9d
-	xorl	%r8d,%r12d
-	movzbl	%bl,%ebx
-	movl	%edx,28(%rsi)
-	addl	%r12d,%r9d
-	addb	%al,%cl
-	roll	$22,%r9d
-	movl	%r11d,%r12d
-	pinsrw	$3,(%rdi,%rbx,4),%xmm1
-
-	addl	%r10d,%r9d
-	movl	(%rdi,%rcx,4),%edx
-	xorl	%r10d,%r12d
-	movl	%eax,(%rdi,%rcx,4)
-	andl	%r9d,%r12d
-	addl	32(%r15),%r8d
-	addb	%dl,%al
-	movl	36(%rsi),%ebx
-	addl	$1770035416,%r8d
-	xorl	%r11d,%r12d
-	movzbl	%al,%eax
-	movl	%edx,32(%rsi)
-	addl	%r12d,%r8d
-	addb	%bl,%cl
-	roll	$7,%r8d
-	movl	%r10d,%r12d
-	pinsrw	$4,(%rdi,%rax,4),%xmm0
-
-	addl	%r9d,%r8d
-	movl	(%rdi,%rcx,4),%edx
-	xorl	%r9d,%r12d
-	movl	%ebx,(%rdi,%rcx,4)
-	andl	%r8d,%r12d
-	addl	36(%r15),%r11d
-	addb	%dl,%bl
-	movl	40(%rsi),%eax
-	addl	$2336552879,%r11d
-	xorl	%r10d,%r12d
-	movzbl	%bl,%ebx
-	movl	%edx,36(%rsi)
-	addl	%r12d,%r11d
-	addb	%al,%cl
-	roll	$12,%r11d
-	movl	%r9d,%r12d
-	pinsrw	$4,(%rdi,%rbx,4),%xmm1
-
-	addl	%r8d,%r11d
-	movl	(%rdi,%rcx,4),%edx
-	xorl	%r8d,%r12d
-	movl	%eax,(%rdi,%rcx,4)
-	andl	%r11d,%r12d
-	addl	40(%r15),%r10d
-	addb	%dl,%al
-	movl	44(%rsi),%ebx
-	addl	$4294925233,%r10d
-	xorl	%r9d,%r12d
-	movzbl	%al,%eax
-	movl	%edx,40(%rsi)
-	addl	%r12d,%r10d
-	addb	%bl,%cl
-	roll	$17,%r10d
-	movl	%r8d,%r12d
-	pinsrw	$5,(%rdi,%rax,4),%xmm0
-
-	addl	%r11d,%r10d
-	movl	(%rdi,%rcx,4),%edx
-	xorl	%r11d,%r12d
-	movl	%ebx,(%rdi,%rcx,4)
-	andl	%r10d,%r12d
-	addl	44(%r15),%r9d
-	addb	%dl,%bl
-	movl	48(%rsi),%eax
-	addl	$2304563134,%r9d
-	xorl	%r8d,%r12d
-	movzbl	%bl,%ebx
-	movl	%edx,44(%rsi)
-	addl	%r12d,%r9d
-	addb	%al,%cl
-	roll	$22,%r9d
-	movl	%r11d,%r12d
-	pinsrw	$5,(%rdi,%rbx,4),%xmm1
-
-	addl	%r10d,%r9d
-	movl	(%rdi,%rcx,4),%edx
-	xorl	%r10d,%r12d
-	movl	%eax,(%rdi,%rcx,4)
-	andl	%r9d,%r12d
-	addl	48(%r15),%r8d
-	addb	%dl,%al
-	movl	52(%rsi),%ebx
-	addl	$1804603682,%r8d
-	xorl	%r11d,%r12d
-	movzbl	%al,%eax
-	movl	%edx,48(%rsi)
-	addl	%r12d,%r8d
-	addb	%bl,%cl
-	roll	$7,%r8d
-	movl	%r10d,%r12d
-	pinsrw	$6,(%rdi,%rax,4),%xmm0
-
-	addl	%r9d,%r8d
-	movl	(%rdi,%rcx,4),%edx
-	xorl	%r9d,%r12d
-	movl	%ebx,(%rdi,%rcx,4)
-	andl	%r8d,%r12d
-	addl	52(%r15),%r11d
-	addb	%dl,%bl
-	movl	56(%rsi),%eax
-	addl	$4254626195,%r11d
-	xorl	%r10d,%r12d
-	movzbl	%bl,%ebx
-	movl	%edx,52(%rsi)
-	addl	%r12d,%r11d
-	addb	%al,%cl
-	roll	$12,%r11d
-	movl	%r9d,%r12d
-	pinsrw	$6,(%rdi,%rbx,4),%xmm1
-
-	addl	%r8d,%r11d
-	movl	(%rdi,%rcx,4),%edx
-	xorl	%r8d,%r12d
-	movl	%eax,(%rdi,%rcx,4)
-	andl	%r11d,%r12d
-	addl	56(%r15),%r10d
-	addb	%dl,%al
-	movl	60(%rsi),%ebx
-	addl	$2792965006,%r10d
-	xorl	%r9d,%r12d
-	movzbl	%al,%eax
-	movl	%edx,56(%rsi)
-	addl	%r12d,%r10d
-	addb	%bl,%cl
-	roll	$17,%r10d
-	movl	%r8d,%r12d
-	pinsrw	$7,(%rdi,%rax,4),%xmm0
-
-	addl	%r11d,%r10d
-	movdqu	(%r13),%xmm2
-	movl	(%rdi,%rcx,4),%edx
-	xorl	%r11d,%r12d
-	movl	%ebx,(%rdi,%rcx,4)
-	andl	%r10d,%r12d
-	addl	60(%r15),%r9d
-	addb	%dl,%bl
-	movl	64(%rsi),%eax
-	addl	$1236535329,%r9d
-	xorl	%r8d,%r12d
-	movzbl	%bl,%ebx
-	movl	%edx,60(%rsi)
-	addl	%r12d,%r9d
-	addb	%al,%cl
-	roll	$22,%r9d
-	movl	%r10d,%r12d
-	pinsrw	$7,(%rdi,%rbx,4),%xmm1
-
-	addl	%r10d,%r9d
-	psllq	$8,%xmm1
-	pxor	%xmm0,%xmm2
-	pxor	%xmm1,%xmm2
-	pxor	%xmm0,%xmm0
-	movl	(%rdi,%rcx,4),%edx
-	xorl	%r9d,%r12d
-	movl	%eax,(%rdi,%rcx,4)
-	andl	%r11d,%r12d
-	addl	4(%r15),%r8d
-	addb	%dl,%al
-	movl	68(%rsi),%ebx
-	addl	$4129170786,%r8d
-	xorl	%r10d,%r12d
-	movzbl	%al,%eax
-	movl	%edx,64(%rsi)
-	addl	%r12d,%r8d
-	addb	%bl,%cl
-	roll	$5,%r8d
-	movl	%r9d,%r12d
-	movd	(%rdi,%rax,4),%xmm0
-
-	addl	%r9d,%r8d
-	pxor	%xmm1,%xmm1
-	movl	(%rdi,%rcx,4),%edx
-	xorl	%r8d,%r12d
-	movl	%ebx,(%rdi,%rcx,4)
-	andl	%r10d,%r12d
-	addl	24(%r15),%r11d
-	addb	%dl,%bl
-	movl	72(%rsi),%eax
-	addl	$3225465664,%r11d
-	xorl	%r9d,%r12d
-	movzbl	%bl,%ebx
-	movl	%edx,68(%rsi)
-	addl	%r12d,%r11d
-	addb	%al,%cl
-	roll	$9,%r11d
-	movl	%r8d,%r12d
-	movd	(%rdi,%rbx,4),%xmm1
-
-	addl	%r8d,%r11d
-	movl	(%rdi,%rcx,4),%edx
-	xorl	%r11d,%r12d
-	movl	%eax,(%rdi,%rcx,4)
-	andl	%r9d,%r12d
-	addl	44(%r15),%r10d
-	addb	%dl,%al
-	movl	76(%rsi),%ebx
-	addl	$643717713,%r10d
-	xorl	%r8d,%r12d
-	movzbl	%al,%eax
-	movl	%edx,72(%rsi)
-	addl	%r12d,%r10d
-	addb	%bl,%cl
-	roll	$14,%r10d
-	movl	%r11d,%r12d
-	pinsrw	$1,(%rdi,%rax,4),%xmm0
-
-	addl	%r11d,%r10d
-	movl	(%rdi,%rcx,4),%edx
-	xorl	%r10d,%r12d
-	movl	%ebx,(%rdi,%rcx,4)
-	andl	%r8d,%r12d
-	addl	0(%r15),%r9d
-	addb	%dl,%bl
-	movl	80(%rsi),%eax
-	addl	$3921069994,%r9d
-	xorl	%r11d,%r12d
-	movzbl	%bl,%ebx
-	movl	%edx,76(%rsi)
-	addl	%r12d,%r9d
-	addb	%al,%cl
-	roll	$20,%r9d
-	movl	%r10d,%r12d
-	pinsrw	$1,(%rdi,%rbx,4),%xmm1
-
-	addl	%r10d,%r9d
-	movl	(%rdi,%rcx,4),%edx
-	xorl	%r9d,%r12d
-	movl	%eax,(%rdi,%rcx,4)
-	andl	%r11d,%r12d
-	addl	20(%r15),%r8d
-	addb	%dl,%al
-	movl	84(%rsi),%ebx
-	addl	$3593408605,%r8d
-	xorl	%r10d,%r12d
-	movzbl	%al,%eax
-	movl	%edx,80(%rsi)
-	addl	%r12d,%r8d
-	addb	%bl,%cl
-	roll	$5,%r8d
-	movl	%r9d,%r12d
-	pinsrw	$2,(%rdi,%rax,4),%xmm0
-
-	addl	%r9d,%r8d
-	movl	(%rdi,%rcx,4),%edx
-	xorl	%r8d,%r12d
-	movl	%ebx,(%rdi,%rcx,4)
-	andl	%r10d,%r12d
-	addl	40(%r15),%r11d
-	addb	%dl,%bl
-	movl	88(%rsi),%eax
-	addl	$38016083,%r11d
-	xorl	%r9d,%r12d
-	movzbl	%bl,%ebx
-	movl	%edx,84(%rsi)
-	addl	%r12d,%r11d
-	addb	%al,%cl
-	roll	$9,%r11d
-	movl	%r8d,%r12d
-	pinsrw	$2,(%rdi,%rbx,4),%xmm1
-
-	addl	%r8d,%r11d
-	movl	(%rdi,%rcx,4),%edx
-	xorl	%r11d,%r12d
-	movl	%eax,(%rdi,%rcx,4)
-	andl	%r9d,%r12d
-	addl	60(%r15),%r10d
-	addb	%dl,%al
-	movl	92(%rsi),%ebx
-	addl	$3634488961,%r10d
-	xorl	%r8d,%r12d
-	movzbl	%al,%eax
-	movl	%edx,88(%rsi)
-	addl	%r12d,%r10d
-	addb	%bl,%cl
-	roll	$14,%r10d
-	movl	%r11d,%r12d
-	pinsrw	$3,(%rdi,%rax,4),%xmm0
-
-	addl	%r11d,%r10d
-	movl	(%rdi,%rcx,4),%edx
-	xorl	%r10d,%r12d
-	movl	%ebx,(%rdi,%rcx,4)
-	andl	%r8d,%r12d
-	addl	16(%r15),%r9d
-	addb	%dl,%bl
-	movl	96(%rsi),%eax
-	addl	$3889429448,%r9d
-	xorl	%r11d,%r12d
-	movzbl	%bl,%ebx
-	movl	%edx,92(%rsi)
-	addl	%r12d,%r9d
-	addb	%al,%cl
-	roll	$20,%r9d
-	movl	%r10d,%r12d
-	pinsrw	$3,(%rdi,%rbx,4),%xmm1
-
-	addl	%r10d,%r9d
-	movl	(%rdi,%rcx,4),%edx
-	xorl	%r9d,%r12d
-	movl	%eax,(%rdi,%rcx,4)
-	andl	%r11d,%r12d
-	addl	36(%r15),%r8d
-	addb	%dl,%al
-	movl	100(%rsi),%ebx
-	addl	$568446438,%r8d
-	xorl	%r10d,%r12d
-	movzbl	%al,%eax
-	movl	%edx,96(%rsi)
-	addl	%r12d,%r8d
-	addb	%bl,%cl
-	roll	$5,%r8d
-	movl	%r9d,%r12d
-	pinsrw	$4,(%rdi,%rax,4),%xmm0
-
-	addl	%r9d,%r8d
-	movl	(%rdi,%rcx,4),%edx
-	xorl	%r8d,%r12d
-	movl	%ebx,(%rdi,%rcx,4)
-	andl	%r10d,%r12d
-	addl	56(%r15),%r11d
-	addb	%dl,%bl
-	movl	104(%rsi),%eax
-	addl	$3275163606,%r11d
-	xorl	%r9d,%r12d
-	movzbl	%bl,%ebx
-	movl	%edx,100(%rsi)
-	addl	%r12d,%r11d
-	addb	%al,%cl
-	roll	$9,%r11d
-	movl	%r8d,%r12d
-	pinsrw	$4,(%rdi,%rbx,4),%xmm1
-
-	addl	%r8d,%r11d
-	movl	(%rdi,%rcx,4),%edx
-	xorl	%r11d,%r12d
-	movl	%eax,(%rdi,%rcx,4)
-	andl	%r9d,%r12d
-	addl	12(%r15),%r10d
-	addb	%dl,%al
-	movl	108(%rsi),%ebx
-	addl	$4107603335,%r10d
-	xorl	%r8d,%r12d
-	movzbl	%al,%eax
-	movl	%edx,104(%rsi)
-	addl	%r12d,%r10d
-	addb	%bl,%cl
-	roll	$14,%r10d
-	movl	%r11d,%r12d
-	pinsrw	$5,(%rdi,%rax,4),%xmm0
-
-	addl	%r11d,%r10d
-	movl	(%rdi,%rcx,4),%edx
-	xorl	%r10d,%r12d
-	movl	%ebx,(%rdi,%rcx,4)
-	andl	%r8d,%r12d
-	addl	32(%r15),%r9d
-	addb	%dl,%bl
-	movl	112(%rsi),%eax
-	addl	$1163531501,%r9d
-	xorl	%r11d,%r12d
-	movzbl	%bl,%ebx
-	movl	%edx,108(%rsi)
-	addl	%r12d,%r9d
-	addb	%al,%cl
-	roll	$20,%r9d
-	movl	%r10d,%r12d
-	pinsrw	$5,(%rdi,%rbx,4),%xmm1
-
-	addl	%r10d,%r9d
-	movl	(%rdi,%rcx,4),%edx
-	xorl	%r9d,%r12d
-	movl	%eax,(%rdi,%rcx,4)
-	andl	%r11d,%r12d
-	addl	52(%r15),%r8d
-	addb	%dl,%al
-	movl	116(%rsi),%ebx
-	addl	$2850285829,%r8d
-	xorl	%r10d,%r12d
-	movzbl	%al,%eax
-	movl	%edx,112(%rsi)
-	addl	%r12d,%r8d
-	addb	%bl,%cl
-	roll	$5,%r8d
-	movl	%r9d,%r12d
-	pinsrw	$6,(%rdi,%rax,4),%xmm0
-
-	addl	%r9d,%r8d
-	movl	(%rdi,%rcx,4),%edx
-	xorl	%r8d,%r12d
-	movl	%ebx,(%rdi,%rcx,4)
-	andl	%r10d,%r12d
-	addl	8(%r15),%r11d
-	addb	%dl,%bl
-	movl	120(%rsi),%eax
-	addl	$4243563512,%r11d
-	xorl	%r9d,%r12d
-	movzbl	%bl,%ebx
-	movl	%edx,116(%rsi)
-	addl	%r12d,%r11d
-	addb	%al,%cl
-	roll	$9,%r11d
-	movl	%r8d,%r12d
-	pinsrw	$6,(%rdi,%rbx,4),%xmm1
-
-	addl	%r8d,%r11d
-	movl	(%rdi,%rcx,4),%edx
-	xorl	%r11d,%r12d
-	movl	%eax,(%rdi,%rcx,4)
-	andl	%r9d,%r12d
-	addl	28(%r15),%r10d
-	addb	%dl,%al
-	movl	124(%rsi),%ebx
-	addl	$1735328473,%r10d
-	xorl	%r8d,%r12d
-	movzbl	%al,%eax
-	movl	%edx,120(%rsi)
-	addl	%r12d,%r10d
-	addb	%bl,%cl
-	roll	$14,%r10d
-	movl	%r11d,%r12d
-	pinsrw	$7,(%rdi,%rax,4),%xmm0
-
-	addl	%r11d,%r10d
-	movdqu	16(%r13),%xmm3
-	addb	$32,%bpl
-	movl	(%rdi,%rcx,4),%edx
-	xorl	%r10d,%r12d
-	movl	%ebx,(%rdi,%rcx,4)
-	andl	%r8d,%r12d
-	addl	48(%r15),%r9d
-	addb	%dl,%bl
-	movl	0(%rdi,%rbp,4),%eax
-	addl	$2368359562,%r9d
-	xorl	%r11d,%r12d
-	movzbl	%bl,%ebx
-	movl	%edx,124(%rsi)
-	addl	%r12d,%r9d
-	addb	%al,%cl
-	roll	$20,%r9d
-	movl	%r11d,%r12d
-	pinsrw	$7,(%rdi,%rbx,4),%xmm1
-
-	addl	%r10d,%r9d
-	movq	%rcx,%rsi
-	xorq	%rcx,%rcx
-	movb	%sil,%cl
-	leaq	(%rdi,%rbp,4),%rsi
-	psllq	$8,%xmm1
-	pxor	%xmm0,%xmm3
-	pxor	%xmm1,%xmm3
-	pxor	%xmm0,%xmm0
-	movl	(%rdi,%rcx,4),%edx
-	xorl	%r10d,%r12d
-	movl	%eax,(%rdi,%rcx,4)
-	xorl	%r9d,%r12d
-	addl	20(%r15),%r8d
-	addb	%dl,%al
-	movl	4(%rsi),%ebx
-	addl	$4294588738,%r8d
-	movzbl	%al,%eax
-	addl	%r12d,%r8d
-	movl	%edx,0(%rsi)
-	addb	%bl,%cl
-	roll	$4,%r8d
-	movl	%r10d,%r12d
-	movd	(%rdi,%rax,4),%xmm0
-
-	addl	%r9d,%r8d
-	pxor	%xmm1,%xmm1
-	movl	(%rdi,%rcx,4),%edx
-	xorl	%r9d,%r12d
-	movl	%ebx,(%rdi,%rcx,4)
-	xorl	%r8d,%r12d
-	addl	32(%r15),%r11d
-	addb	%dl,%bl
-	movl	8(%rsi),%eax
-	addl	$2272392833,%r11d
-	movzbl	%bl,%ebx
-	addl	%r12d,%r11d
-	movl	%edx,4(%rsi)
-	addb	%al,%cl
-	roll	$11,%r11d
-	movl	%r9d,%r12d
-	movd	(%rdi,%rbx,4),%xmm1
-
-	addl	%r8d,%r11d
-	movl	(%rdi,%rcx,4),%edx
-	xorl	%r8d,%r12d
-	movl	%eax,(%rdi,%rcx,4)
-	xorl	%r11d,%r12d
-	addl	44(%r15),%r10d
-	addb	%dl,%al
-	movl	12(%rsi),%ebx
-	addl	$1839030562,%r10d
-	movzbl	%al,%eax
-	addl	%r12d,%r10d
-	movl	%edx,8(%rsi)
-	addb	%bl,%cl
-	roll	$16,%r10d
-	movl	%r8d,%r12d
-	pinsrw	$1,(%rdi,%rax,4),%xmm0
-
-	addl	%r11d,%r10d
-	movl	(%rdi,%rcx,4),%edx
-	xorl	%r11d,%r12d
-	movl	%ebx,(%rdi,%rcx,4)
-	xorl	%r10d,%r12d
-	addl	56(%r15),%r9d
-	addb	%dl,%bl
-	movl	16(%rsi),%eax
-	addl	$4259657740,%r9d
-	movzbl	%bl,%ebx
-	addl	%r12d,%r9d
-	movl	%edx,12(%rsi)
-	addb	%al,%cl
-	roll	$23,%r9d
-	movl	%r11d,%r12d
-	pinsrw	$1,(%rdi,%rbx,4),%xmm1
-
-	addl	%r10d,%r9d
-	movl	(%rdi,%rcx,4),%edx
-	xorl	%r10d,%r12d
-	movl	%eax,(%rdi,%rcx,4)
-	xorl	%r9d,%r12d
-	addl	4(%r15),%r8d
-	addb	%dl,%al
-	movl	20(%rsi),%ebx
-	addl	$2763975236,%r8d
-	movzbl	%al,%eax
-	addl	%r12d,%r8d
-	movl	%edx,16(%rsi)
-	addb	%bl,%cl
-	roll	$4,%r8d
-	movl	%r10d,%r12d
-	pinsrw	$2,(%rdi,%rax,4),%xmm0
-
-	addl	%r9d,%r8d
-	movl	(%rdi,%rcx,4),%edx
-	xorl	%r9d,%r12d
-	movl	%ebx,(%rdi,%rcx,4)
-	xorl	%r8d,%r12d
-	addl	16(%r15),%r11d
-	addb	%dl,%bl
-	movl	24(%rsi),%eax
-	addl	$1272893353,%r11d
-	movzbl	%bl,%ebx
-	addl	%r12d,%r11d
-	movl	%edx,20(%rsi)
-	addb	%al,%cl
-	roll	$11,%r11d
-	movl	%r9d,%r12d
-	pinsrw	$2,(%rdi,%rbx,4),%xmm1
-
-	addl	%r8d,%r11d
-	movl	(%rdi,%rcx,4),%edx
-	xorl	%r8d,%r12d
-	movl	%eax,(%rdi,%rcx,4)
-	xorl	%r11d,%r12d
-	addl	28(%r15),%r10d
-	addb	%dl,%al
-	movl	28(%rsi),%ebx
-	addl	$4139469664,%r10d
-	movzbl	%al,%eax
-	addl	%r12d,%r10d
-	movl	%edx,24(%rsi)
-	addb	%bl,%cl
-	roll	$16,%r10d
-	movl	%r8d,%r12d
-	pinsrw	$3,(%rdi,%rax,4),%xmm0
-
-	addl	%r11d,%r10d
-	movl	(%rdi,%rcx,4),%edx
-	xorl	%r11d,%r12d
-	movl	%ebx,(%rdi,%rcx,4)
-	xorl	%r10d,%r12d
-	addl	40(%r15),%r9d
-	addb	%dl,%bl
-	movl	32(%rsi),%eax
-	addl	$3200236656,%r9d
-	movzbl	%bl,%ebx
-	addl	%r12d,%r9d
-	movl	%edx,28(%rsi)
-	addb	%al,%cl
-	roll	$23,%r9d
-	movl	%r11d,%r12d
-	pinsrw	$3,(%rdi,%rbx,4),%xmm1
-
-	addl	%r10d,%r9d
-	movl	(%rdi,%rcx,4),%edx
-	xorl	%r10d,%r12d
-	movl	%eax,(%rdi,%rcx,4)
-	xorl	%r9d,%r12d
-	addl	52(%r15),%r8d
-	addb	%dl,%al
-	movl	36(%rsi),%ebx
-	addl	$681279174,%r8d
-	movzbl	%al,%eax
-	addl	%r12d,%r8d
-	movl	%edx,32(%rsi)
-	addb	%bl,%cl
-	roll	$4,%r8d
-	movl	%r10d,%r12d
-	pinsrw	$4,(%rdi,%rax,4),%xmm0
-
-	addl	%r9d,%r8d
-	movl	(%rdi,%rcx,4),%edx
-	xorl	%r9d,%r12d
-	movl	%ebx,(%rdi,%rcx,4)
-	xorl	%r8d,%r12d
-	addl	0(%r15),%r11d
-	addb	%dl,%bl
-	movl	40(%rsi),%eax
-	addl	$3936430074,%r11d
-	movzbl	%bl,%ebx
-	addl	%r12d,%r11d
-	movl	%edx,36(%rsi)
-	addb	%al,%cl
-	roll	$11,%r11d
-	movl	%r9d,%r12d
-	pinsrw	$4,(%rdi,%rbx,4),%xmm1
-
-	addl	%r8d,%r11d
-	movl	(%rdi,%rcx,4),%edx
-	xorl	%r8d,%r12d
-	movl	%eax,(%rdi,%rcx,4)
-	xorl	%r11d,%r12d
-	addl	12(%r15),%r10d
-	addb	%dl,%al
-	movl	44(%rsi),%ebx
-	addl	$3572445317,%r10d
-	movzbl	%al,%eax
-	addl	%r12d,%r10d
-	movl	%edx,40(%rsi)
-	addb	%bl,%cl
-	roll	$16,%r10d
-	movl	%r8d,%r12d
-	pinsrw	$5,(%rdi,%rax,4),%xmm0
-
-	addl	%r11d,%r10d
-	movl	(%rdi,%rcx,4),%edx
-	xorl	%r11d,%r12d
-	movl	%ebx,(%rdi,%rcx,4)
-	xorl	%r10d,%r12d
-	addl	24(%r15),%r9d
-	addb	%dl,%bl
-	movl	48(%rsi),%eax
-	addl	$76029189,%r9d
-	movzbl	%bl,%ebx
-	addl	%r12d,%r9d
-	movl	%edx,44(%rsi)
-	addb	%al,%cl
-	roll	$23,%r9d
-	movl	%r11d,%r12d
-	pinsrw	$5,(%rdi,%rbx,4),%xmm1
-
-	addl	%r10d,%r9d
-	movl	(%rdi,%rcx,4),%edx
-	xorl	%r10d,%r12d
-	movl	%eax,(%rdi,%rcx,4)
-	xorl	%r9d,%r12d
-	addl	36(%r15),%r8d
-	addb	%dl,%al
-	movl	52(%rsi),%ebx
-	addl	$3654602809,%r8d
-	movzbl	%al,%eax
-	addl	%r12d,%r8d
-	movl	%edx,48(%rsi)
-	addb	%bl,%cl
-	roll	$4,%r8d
-	movl	%r10d,%r12d
-	pinsrw	$6,(%rdi,%rax,4),%xmm0
-
-	addl	%r9d,%r8d
-	movl	(%rdi,%rcx,4),%edx
-	xorl	%r9d,%r12d
-	movl	%ebx,(%rdi,%rcx,4)
-	xorl	%r8d,%r12d
-	addl	48(%r15),%r11d
-	addb	%dl,%bl
-	movl	56(%rsi),%eax
-	addl	$3873151461,%r11d
-	movzbl	%bl,%ebx
-	addl	%r12d,%r11d
-	movl	%edx,52(%rsi)
-	addb	%al,%cl
-	roll	$11,%r11d
-	movl	%r9d,%r12d
-	pinsrw	$6,(%rdi,%rbx,4),%xmm1
-
-	addl	%r8d,%r11d
-	movl	(%rdi,%rcx,4),%edx
-	xorl	%r8d,%r12d
-	movl	%eax,(%rdi,%rcx,4)
-	xorl	%r11d,%r12d
-	addl	60(%r15),%r10d
-	addb	%dl,%al
-	movl	60(%rsi),%ebx
-	addl	$530742520,%r10d
-	movzbl	%al,%eax
-	addl	%r12d,%r10d
-	movl	%edx,56(%rsi)
-	addb	%bl,%cl
-	roll	$16,%r10d
-	movl	%r8d,%r12d
-	pinsrw	$7,(%rdi,%rax,4),%xmm0
-
-	addl	%r11d,%r10d
-	movdqu	32(%r13),%xmm4
-	movl	(%rdi,%rcx,4),%edx
-	xorl	%r11d,%r12d
-	movl	%ebx,(%rdi,%rcx,4)
-	xorl	%r10d,%r12d
-	addl	8(%r15),%r9d
-	addb	%dl,%bl
-	movl	64(%rsi),%eax
-	addl	$3299628645,%r9d
-	movzbl	%bl,%ebx
-	addl	%r12d,%r9d
-	movl	%edx,60(%rsi)
-	addb	%al,%cl
-	roll	$23,%r9d
-	movl	$-1,%r12d
-	pinsrw	$7,(%rdi,%rbx,4),%xmm1
-
-	addl	%r10d,%r9d
-	psllq	$8,%xmm1
-	pxor	%xmm0,%xmm4
-	pxor	%xmm1,%xmm4
-	pxor	%xmm0,%xmm0
-	movl	(%rdi,%rcx,4),%edx
-	xorl	%r11d,%r12d
-	movl	%eax,(%rdi,%rcx,4)
-	orl	%r9d,%r12d
-	addl	0(%r15),%r8d
-	addb	%dl,%al
-	movl	68(%rsi),%ebx
-	addl	$4096336452,%r8d
-	movzbl	%al,%eax
-	xorl	%r10d,%r12d
-	movl	%edx,64(%rsi)
-	addl	%r12d,%r8d
-	addb	%bl,%cl
-	roll	$6,%r8d
-	movl	$-1,%r12d
-	movd	(%rdi,%rax,4),%xmm0
-
-	addl	%r9d,%r8d
-	pxor	%xmm1,%xmm1
-	movl	(%rdi,%rcx,4),%edx
-	xorl	%r10d,%r12d
-	movl	%ebx,(%rdi,%rcx,4)
-	orl	%r8d,%r12d
-	addl	28(%r15),%r11d
-	addb	%dl,%bl
-	movl	72(%rsi),%eax
-	addl	$1126891415,%r11d
-	movzbl	%bl,%ebx
-	xorl	%r9d,%r12d
-	movl	%edx,68(%rsi)
-	addl	%r12d,%r11d
-	addb	%al,%cl
-	roll	$10,%r11d
-	movl	$-1,%r12d
-	movd	(%rdi,%rbx,4),%xmm1
-
-	addl	%r8d,%r11d
-	movl	(%rdi,%rcx,4),%edx
-	xorl	%r9d,%r12d
-	movl	%eax,(%rdi,%rcx,4)
-	orl	%r11d,%r12d
-	addl	56(%r15),%r10d
-	addb	%dl,%al
-	movl	76(%rsi),%ebx
-	addl	$2878612391,%r10d
-	movzbl	%al,%eax
-	xorl	%r8d,%r12d
-	movl	%edx,72(%rsi)
-	addl	%r12d,%r10d
-	addb	%bl,%cl
-	roll	$15,%r10d
-	movl	$-1,%r12d
-	pinsrw	$1,(%rdi,%rax,4),%xmm0
-
-	addl	%r11d,%r10d
-	movl	(%rdi,%rcx,4),%edx
-	xorl	%r8d,%r12d
-	movl	%ebx,(%rdi,%rcx,4)
-	orl	%r10d,%r12d
-	addl	20(%r15),%r9d
-	addb	%dl,%bl
-	movl	80(%rsi),%eax
-	addl	$4237533241,%r9d
-	movzbl	%bl,%ebx
-	xorl	%r11d,%r12d
-	movl	%edx,76(%rsi)
-	addl	%r12d,%r9d
-	addb	%al,%cl
-	roll	$21,%r9d
-	movl	$-1,%r12d
-	pinsrw	$1,(%rdi,%rbx,4),%xmm1
-
-	addl	%r10d,%r9d
-	movl	(%rdi,%rcx,4),%edx
-	xorl	%r11d,%r12d
-	movl	%eax,(%rdi,%rcx,4)
-	orl	%r9d,%r12d
-	addl	48(%r15),%r8d
-	addb	%dl,%al
-	movl	84(%rsi),%ebx
-	addl	$1700485571,%r8d
-	movzbl	%al,%eax
-	xorl	%r10d,%r12d
-	movl	%edx,80(%rsi)
-	addl	%r12d,%r8d
-	addb	%bl,%cl
-	roll	$6,%r8d
-	movl	$-1,%r12d
-	pinsrw	$2,(%rdi,%rax,4),%xmm0
-
-	addl	%r9d,%r8d
-	movl	(%rdi,%rcx,4),%edx
-	xorl	%r10d,%r12d
-	movl	%ebx,(%rdi,%rcx,4)
-	orl	%r8d,%r12d
-	addl	12(%r15),%r11d
-	addb	%dl,%bl
-	movl	88(%rsi),%eax
-	addl	$2399980690,%r11d
-	movzbl	%bl,%ebx
-	xorl	%r9d,%r12d
-	movl	%edx,84(%rsi)
-	addl	%r12d,%r11d
-	addb	%al,%cl
-	roll	$10,%r11d
-	movl	$-1,%r12d
-	pinsrw	$2,(%rdi,%rbx,4),%xmm1
-
-	addl	%r8d,%r11d
-	movl	(%rdi,%rcx,4),%edx
-	xorl	%r9d,%r12d
-	movl	%eax,(%rdi,%rcx,4)
-	orl	%r11d,%r12d
-	addl	40(%r15),%r10d
-	addb	%dl,%al
-	movl	92(%rsi),%ebx
-	addl	$4293915773,%r10d
-	movzbl	%al,%eax
-	xorl	%r8d,%r12d
-	movl	%edx,88(%rsi)
-	addl	%r12d,%r10d
-	addb	%bl,%cl
-	roll	$15,%r10d
-	movl	$-1,%r12d
-	pinsrw	$3,(%rdi,%rax,4),%xmm0
-
-	addl	%r11d,%r10d
-	movl	(%rdi,%rcx,4),%edx
-	xorl	%r8d,%r12d
-	movl	%ebx,(%rdi,%rcx,4)
-	orl	%r10d,%r12d
-	addl	4(%r15),%r9d
-	addb	%dl,%bl
-	movl	96(%rsi),%eax
-	addl	$2240044497,%r9d
-	movzbl	%bl,%ebx
-	xorl	%r11d,%r12d
-	movl	%edx,92(%rsi)
-	addl	%r12d,%r9d
-	addb	%al,%cl
-	roll	$21,%r9d
-	movl	$-1,%r12d
-	pinsrw	$3,(%rdi,%rbx,4),%xmm1
-
-	addl	%r10d,%r9d
-	movl	(%rdi,%rcx,4),%edx
-	xorl	%r11d,%r12d
-	movl	%eax,(%rdi,%rcx,4)
-	orl	%r9d,%r12d
-	addl	32(%r15),%r8d
-	addb	%dl,%al
-	movl	100(%rsi),%ebx
-	addl	$1873313359,%r8d
-	movzbl	%al,%eax
-	xorl	%r10d,%r12d
-	movl	%edx,96(%rsi)
-	addl	%r12d,%r8d
-	addb	%bl,%cl
-	roll	$6,%r8d
-	movl	$-1,%r12d
-	pinsrw	$4,(%rdi,%rax,4),%xmm0
-
-	addl	%r9d,%r8d
-	movl	(%rdi,%rcx,4),%edx
-	xorl	%r10d,%r12d
-	movl	%ebx,(%rdi,%rcx,4)
-	orl	%r8d,%r12d
-	addl	60(%r15),%r11d
-	addb	%dl,%bl
-	movl	104(%rsi),%eax
-	addl	$4264355552,%r11d
-	movzbl	%bl,%ebx
-	xorl	%r9d,%r12d
-	movl	%edx,100(%rsi)
-	addl	%r12d,%r11d
-	addb	%al,%cl
-	roll	$10,%r11d
-	movl	$-1,%r12d
-	pinsrw	$4,(%rdi,%rbx,4),%xmm1
-
-	addl	%r8d,%r11d
-	movl	(%rdi,%rcx,4),%edx
-	xorl	%r9d,%r12d
-	movl	%eax,(%rdi,%rcx,4)
-	orl	%r11d,%r12d
-	addl	24(%r15),%r10d
-	addb	%dl,%al
-	movl	108(%rsi),%ebx
-	addl	$2734768916,%r10d
-	movzbl	%al,%eax
-	xorl	%r8d,%r12d
-	movl	%edx,104(%rsi)
-	addl	%r12d,%r10d
-	addb	%bl,%cl
-	roll	$15,%r10d
-	movl	$-1,%r12d
-	pinsrw	$5,(%rdi,%rax,4),%xmm0
-
-	addl	%r11d,%r10d
-	movl	(%rdi,%rcx,4),%edx
-	xorl	%r8d,%r12d
-	movl	%ebx,(%rdi,%rcx,4)
-	orl	%r10d,%r12d
-	addl	52(%r15),%r9d
-	addb	%dl,%bl
-	movl	112(%rsi),%eax
-	addl	$1309151649,%r9d
-	movzbl	%bl,%ebx
-	xorl	%r11d,%r12d
-	movl	%edx,108(%rsi)
-	addl	%r12d,%r9d
-	addb	%al,%cl
-	roll	$21,%r9d
-	movl	$-1,%r12d
-	pinsrw	$5,(%rdi,%rbx,4),%xmm1
-
-	addl	%r10d,%r9d
-	movl	(%rdi,%rcx,4),%edx
-	xorl	%r11d,%r12d
-	movl	%eax,(%rdi,%rcx,4)
-	orl	%r9d,%r12d
-	addl	16(%r15),%r8d
-	addb	%dl,%al
-	movl	116(%rsi),%ebx
-	addl	$4149444226,%r8d
-	movzbl	%al,%eax
-	xorl	%r10d,%r12d
-	movl	%edx,112(%rsi)
-	addl	%r12d,%r8d
-	addb	%bl,%cl
-	roll	$6,%r8d
-	movl	$-1,%r12d
-	pinsrw	$6,(%rdi,%rax,4),%xmm0
-
-	addl	%r9d,%r8d
-	movl	(%rdi,%rcx,4),%edx
-	xorl	%r10d,%r12d
-	movl	%ebx,(%rdi,%rcx,4)
-	orl	%r8d,%r12d
-	addl	44(%r15),%r11d
-	addb	%dl,%bl
-	movl	120(%rsi),%eax
-	addl	$3174756917,%r11d
-	movzbl	%bl,%ebx
-	xorl	%r9d,%r12d
-	movl	%edx,116(%rsi)
-	addl	%r12d,%r11d
-	addb	%al,%cl
-	roll	$10,%r11d
-	movl	$-1,%r12d
-	pinsrw	$6,(%rdi,%rbx,4),%xmm1
-
-	addl	%r8d,%r11d
-	movl	(%rdi,%rcx,4),%edx
-	xorl	%r9d,%r12d
-	movl	%eax,(%rdi,%rcx,4)
-	orl	%r11d,%r12d
-	addl	8(%r15),%r10d
-	addb	%dl,%al
-	movl	124(%rsi),%ebx
-	addl	$718787259,%r10d
-	movzbl	%al,%eax
-	xorl	%r8d,%r12d
-	movl	%edx,120(%rsi)
-	addl	%r12d,%r10d
-	addb	%bl,%cl
-	roll	$15,%r10d
-	movl	$-1,%r12d
-	pinsrw	$7,(%rdi,%rax,4),%xmm0
-
-	addl	%r11d,%r10d
-	movdqu	48(%r13),%xmm5
-	addb	$32,%bpl
-	movl	(%rdi,%rcx,4),%edx
-	xorl	%r8d,%r12d
-	movl	%ebx,(%rdi,%rcx,4)
-	orl	%r10d,%r12d
-	addl	36(%r15),%r9d
-	addb	%dl,%bl
-	movl	0(%rdi,%rbp,4),%eax
-	addl	$3951481745,%r9d
-	movzbl	%bl,%ebx
-	xorl	%r11d,%r12d
-	movl	%edx,124(%rsi)
-	addl	%r12d,%r9d
-	addb	%al,%cl
-	roll	$21,%r9d
-	movl	$-1,%r12d
-	pinsrw	$7,(%rdi,%rbx,4),%xmm1
-
-	addl	%r10d,%r9d
-	movq	%rbp,%rsi
-	xorq	%rbp,%rbp
-	movb	%sil,%bpl
-	movq	%rcx,%rsi
-	xorq	%rcx,%rcx
-	movb	%sil,%cl
-	leaq	(%rdi,%rbp,4),%rsi
-	psllq	$8,%xmm1
-	pxor	%xmm0,%xmm5
-	pxor	%xmm1,%xmm5
-	addl	0(%rsp),%r8d
-	addl	4(%rsp),%r9d
-	addl	8(%rsp),%r10d
-	addl	12(%rsp),%r11d
-
-	movdqu	%xmm2,(%r14,%r13,1)
-	movdqu	%xmm3,16(%r14,%r13,1)
-	movdqu	%xmm4,32(%r14,%r13,1)
-	movdqu	%xmm5,48(%r14,%r13,1)
-	leaq	64(%r15),%r15
-	leaq	64(%r13),%r13
-	cmpq	16(%rsp),%r15
-	jb	.Loop
-
-	movq	24(%rsp),%r12
-	subb	%al,%cl
-	movl	%r8d,0(%r12)
-	movl	%r9d,4(%r12)
-	movl	%r10d,8(%r12)
-	movl	%r11d,12(%r12)
-	subb	$1,%bpl
-	movl	%ebp,-8(%rdi)
-	movl	%ecx,-4(%rdi)
-
-	movq	40(%rsp),%r15
-	movq	48(%rsp),%r14
-	movq	56(%rsp),%r13
-	movq	64(%rsp),%r12
-	movq	72(%rsp),%rbp
-	movq	80(%rsp),%rbx
-	leaq	88(%rsp),%rsp
-.Lepilogue:
-.Labort:
-	.byte	0xf3,0xc3
-.size	rc4_md5_enc,.-rc4_md5_enc
-#endif

third_party/boringssl/linux-x86_64/crypto/sha/sha1-x86_64.S

diff --git a/third_party/boringssl/linux-x86_64/crypto/sha/sha1-x86_64.S b/third_party/boringssl/linux-x86_64/crypto/sha/sha1-x86_64.S
index 7668c2b..d830b53 100644
--- a/third_party/boringssl/linux-x86_64/crypto/sha/sha1-x86_64.S
+++ b/third_party/boringssl/linux-x86_64/crypto/sha/sha1-x86_64.S
@@ -13,6 +13,11 @@
 	movl	OPENSSL_ia32cap_P+8(%rip),%r10d
 	testl	$512,%r8d
 	jz	.Lialu
+	andl	$268435456,%r8d
+	andl	$1073741824,%r9d
+	orl	%r9d,%r8d
+	cmpl	$1342177280,%r8d
+	je	_avx_shortcut
 	jmp	_ssse3_shortcut
 
 .align	16
@@ -2408,6 +2413,1122 @@
 .Lepilogue_ssse3:
 	.byte	0xf3,0xc3
 .size	sha1_block_data_order_ssse3,.-sha1_block_data_order_ssse3
+.type	sha1_block_data_order_avx,@function
+.align	16
+sha1_block_data_order_avx:
+_avx_shortcut:
+	movq	%rsp,%rax
+	pushq	%rbx
+	pushq	%rbp
+	pushq	%r12
+	pushq	%r13
+	pushq	%r14
+	leaq	-64(%rsp),%rsp
+	vzeroupper
+	movq	%rax,%r14
+	andq	$-64,%rsp
+	movq	%rdi,%r8
+	movq	%rsi,%r9
+	movq	%rdx,%r10
+
+	shlq	$6,%r10
+	addq	%r9,%r10
+	leaq	K_XX_XX+64(%rip),%r11
+
+	movl	0(%r8),%eax
+	movl	4(%r8),%ebx
+	movl	8(%r8),%ecx
+	movl	12(%r8),%edx
+	movl	%ebx,%esi
+	movl	16(%r8),%ebp
+	movl	%ecx,%edi
+	xorl	%edx,%edi
+	andl	%edi,%esi
+
+	vmovdqa	64(%r11),%xmm6
+	vmovdqa	-64(%r11),%xmm11
+	vmovdqu	0(%r9),%xmm0
+	vmovdqu	16(%r9),%xmm1
+	vmovdqu	32(%r9),%xmm2
+	vmovdqu	48(%r9),%xmm3
+	vpshufb	%xmm6,%xmm0,%xmm0
+	addq	$64,%r9
+	vpshufb	%xmm6,%xmm1,%xmm1
+	vpshufb	%xmm6,%xmm2,%xmm2
+	vpshufb	%xmm6,%xmm3,%xmm3
+	vpaddd	%xmm11,%xmm0,%xmm4
+	vpaddd	%xmm11,%xmm1,%xmm5
+	vpaddd	%xmm11,%xmm2,%xmm6
+	vmovdqa	%xmm4,0(%rsp)
+	vmovdqa	%xmm5,16(%rsp)
+	vmovdqa	%xmm6,32(%rsp)
+	jmp	.Loop_avx
+.align	16
+.Loop_avx:
+	shrdl	$2,%ebx,%ebx
+	xorl	%edx,%esi
+	vpalignr	$8,%xmm0,%xmm1,%xmm4
+	movl	%eax,%edi
+	addl	0(%rsp),%ebp
+	vpaddd	%xmm3,%xmm11,%xmm9
+	xorl	%ecx,%ebx
+	shldl	$5,%eax,%eax
+	vpsrldq	$4,%xmm3,%xmm8
+	addl	%esi,%ebp
+	andl	%ebx,%edi
+	vpxor	%xmm0,%xmm4,%xmm4
+	xorl	%ecx,%ebx
+	addl	%eax,%ebp
+	vpxor	%xmm2,%xmm8,%xmm8
+	shrdl	$7,%eax,%eax
+	xorl	%ecx,%edi
+	movl	%ebp,%esi
+	addl	4(%rsp),%edx
+	vpxor	%xmm8,%xmm4,%xmm4
+	xorl	%ebx,%eax
+	shldl	$5,%ebp,%ebp
+	vmovdqa	%xmm9,48(%rsp)
+	addl	%edi,%edx
+	andl	%eax,%esi
+	vpsrld	$31,%xmm4,%xmm8
+	xorl	%ebx,%eax
+	addl	%ebp,%edx
+	shrdl	$7,%ebp,%ebp
+	xorl	%ebx,%esi
+	vpslldq	$12,%xmm4,%xmm10
+	vpaddd	%xmm4,%xmm4,%xmm4
+	movl	%edx,%edi
+	addl	8(%rsp),%ecx
+	xorl	%eax,%ebp
+	shldl	$5,%edx,%edx
+	vpsrld	$30,%xmm10,%xmm9
+	vpor	%xmm8,%xmm4,%xmm4
+	addl	%esi,%ecx
+	andl	%ebp,%edi
+	xorl	%eax,%ebp
+	addl	%edx,%ecx
+	vpslld	$2,%xmm10,%xmm10
+	vpxor	%xmm9,%xmm4,%xmm4
+	shrdl	$7,%edx,%edx
+	xorl	%eax,%edi
+	movl	%ecx,%esi
+	addl	12(%rsp),%ebx
+	vpxor	%xmm10,%xmm4,%xmm4
+	xorl	%ebp,%edx
+	shldl	$5,%ecx,%ecx
+	addl	%edi,%ebx
+	andl	%edx,%esi
+	xorl	%ebp,%edx
+	addl	%ecx,%ebx
+	shrdl	$7,%ecx,%ecx
+	xorl	%ebp,%esi
+	vpalignr	$8,%xmm1,%xmm2,%xmm5
+	movl	%ebx,%edi
+	addl	16(%rsp),%eax
+	vpaddd	%xmm4,%xmm11,%xmm9
+	xorl	%edx,%ecx
+	shldl	$5,%ebx,%ebx
+	vpsrldq	$4,%xmm4,%xmm8
+	addl	%esi,%eax
+	andl	%ecx,%edi
+	vpxor	%xmm1,%xmm5,%xmm5
+	xorl	%edx,%ecx
+	addl	%ebx,%eax
+	vpxor	%xmm3,%xmm8,%xmm8
+	shrdl	$7,%ebx,%ebx
+	xorl	%edx,%edi
+	movl	%eax,%esi
+	addl	20(%rsp),%ebp
+	vpxor	%xmm8,%xmm5,%xmm5
+	xorl	%ecx,%ebx
+	shldl	$5,%eax,%eax
+	vmovdqa	%xmm9,0(%rsp)
+	addl	%edi,%ebp
+	andl	%ebx,%esi
+	vpsrld	$31,%xmm5,%xmm8
+	xorl	%ecx,%ebx
+	addl	%eax,%ebp
+	shrdl	$7,%eax,%eax
+	xorl	%ecx,%esi
+	vpslldq	$12,%xmm5,%xmm10
+	vpaddd	%xmm5,%xmm5,%xmm5
+	movl	%ebp,%edi
+	addl	24(%rsp),%edx
+	xorl	%ebx,%eax
+	shldl	$5,%ebp,%ebp
+	vpsrld	$30,%xmm10,%xmm9
+	vpor	%xmm8,%xmm5,%xmm5
+	addl	%esi,%edx
+	andl	%eax,%edi
+	xorl	%ebx,%eax
+	addl	%ebp,%edx
+	vpslld	$2,%xmm10,%xmm10
+	vpxor	%xmm9,%xmm5,%xmm5
+	shrdl	$7,%ebp,%ebp
+	xorl	%ebx,%edi
+	movl	%edx,%esi
+	addl	28(%rsp),%ecx
+	vpxor	%xmm10,%xmm5,%xmm5
+	xorl	%eax,%ebp
+	shldl	$5,%edx,%edx
+	vmovdqa	-32(%r11),%xmm11
+	addl	%edi,%ecx
+	andl	%ebp,%esi
+	xorl	%eax,%ebp
+	addl	%edx,%ecx
+	shrdl	$7,%edx,%edx
+	xorl	%eax,%esi
+	vpalignr	$8,%xmm2,%xmm3,%xmm6
+	movl	%ecx,%edi
+	addl	32(%rsp),%ebx
+	vpaddd	%xmm5,%xmm11,%xmm9
+	xorl	%ebp,%edx
+	shldl	$5,%ecx,%ecx
+	vpsrldq	$4,%xmm5,%xmm8
+	addl	%esi,%ebx
+	andl	%edx,%edi
+	vpxor	%xmm2,%xmm6,%xmm6
+	xorl	%ebp,%edx
+	addl	%ecx,%ebx
+	vpxor	%xmm4,%xmm8,%xmm8
+	shrdl	$7,%ecx,%ecx
+	xorl	%ebp,%edi
+	movl	%ebx,%esi
+	addl	36(%rsp),%eax
+	vpxor	%xmm8,%xmm6,%xmm6
+	xorl	%edx,%ecx
+	shldl	$5,%ebx,%ebx
+	vmovdqa	%xmm9,16(%rsp)
+	addl	%edi,%eax
+	andl	%ecx,%esi
+	vpsrld	$31,%xmm6,%xmm8
+	xorl	%edx,%ecx
+	addl	%ebx,%eax
+	shrdl	$7,%ebx,%ebx
+	xorl	%edx,%esi
+	vpslldq	$12,%xmm6,%xmm10
+	vpaddd	%xmm6,%xmm6,%xmm6
+	movl	%eax,%edi
+	addl	40(%rsp),%ebp
+	xorl	%ecx,%ebx
+	shldl	$5,%eax,%eax
+	vpsrld	$30,%xmm10,%xmm9
+	vpor	%xmm8,%xmm6,%xmm6
+	addl	%esi,%ebp
+	andl	%ebx,%edi
+	xorl	%ecx,%ebx
+	addl	%eax,%ebp
+	vpslld	$2,%xmm10,%xmm10
+	vpxor	%xmm9,%xmm6,%xmm6
+	shrdl	$7,%eax,%eax
+	xorl	%ecx,%edi
+	movl	%ebp,%esi
+	addl	44(%rsp),%edx
+	vpxor	%xmm10,%xmm6,%xmm6
+	xorl	%ebx,%eax
+	shldl	$5,%ebp,%ebp
+	addl	%edi,%edx
+	andl	%eax,%esi
+	xorl	%ebx,%eax
+	addl	%ebp,%edx
+	shrdl	$7,%ebp,%ebp
+	xorl	%ebx,%esi
+	vpalignr	$8,%xmm3,%xmm4,%xmm7
+	movl	%edx,%edi
+	addl	48(%rsp),%ecx
+	vpaddd	%xmm6,%xmm11,%xmm9
+	xorl	%eax,%ebp
+	shldl	$5,%edx,%edx
+	vpsrldq	$4,%xmm6,%xmm8
+	addl	%esi,%ecx
+	andl	%ebp,%edi
+	vpxor	%xmm3,%xmm7,%xmm7
+	xorl	%eax,%ebp
+	addl	%edx,%ecx
+	vpxor	%xmm5,%xmm8,%xmm8
+	shrdl	$7,%edx,%edx
+	xorl	%eax,%edi
+	movl	%ecx,%esi
+	addl	52(%rsp),%ebx
+	vpxor	%xmm8,%xmm7,%xmm7
+	xorl	%ebp,%edx
+	shldl	$5,%ecx,%ecx
+	vmovdqa	%xmm9,32(%rsp)
+	addl	%edi,%ebx
+	andl	%edx,%esi
+	vpsrld	$31,%xmm7,%xmm8
+	xorl	%ebp,%edx
+	addl	%ecx,%ebx
+	shrdl	$7,%ecx,%ecx
+	xorl	%ebp,%esi
+	vpslldq	$12,%xmm7,%xmm10
+	vpaddd	%xmm7,%xmm7,%xmm7
+	movl	%ebx,%edi
+	addl	56(%rsp),%eax
+	xorl	%edx,%ecx
+	shldl	$5,%ebx,%ebx
+	vpsrld	$30,%xmm10,%xmm9
+	vpor	%xmm8,%xmm7,%xmm7
+	addl	%esi,%eax
+	andl	%ecx,%edi
+	xorl	%edx,%ecx
+	addl	%ebx,%eax
+	vpslld	$2,%xmm10,%xmm10
+	vpxor	%xmm9,%xmm7,%xmm7
+	shrdl	$7,%ebx,%ebx
+	xorl	%edx,%edi
+	movl	%eax,%esi
+	addl	60(%rsp),%ebp
+	vpxor	%xmm10,%xmm7,%xmm7
+	xorl	%ecx,%ebx
+	shldl	$5,%eax,%eax
+	addl	%edi,%ebp
+	andl	%ebx,%esi
+	xorl	%ecx,%ebx
+	addl	%eax,%ebp
+	vpalignr	$8,%xmm6,%xmm7,%xmm8
+	vpxor	%xmm4,%xmm0,%xmm0
+	shrdl	$7,%eax,%eax
+	xorl	%ecx,%esi
+	movl	%ebp,%edi
+	addl	0(%rsp),%edx
+	vpxor	%xmm1,%xmm0,%xmm0
+	xorl	%ebx,%eax
+	shldl	$5,%ebp,%ebp
+	vpaddd	%xmm7,%xmm11,%xmm9
+	addl	%esi,%edx
+	andl	%eax,%edi
+	vpxor	%xmm8,%xmm0,%xmm0
+	xorl	%ebx,%eax
+	addl	%ebp,%edx
+	shrdl	$7,%ebp,%ebp
+	xorl	%ebx,%edi
+	vpsrld	$30,%xmm0,%xmm8
+	vmovdqa	%xmm9,48(%rsp)
+	movl	%edx,%esi
+	addl	4(%rsp),%ecx
+	xorl	%eax,%ebp
+	shldl	$5,%edx,%edx
+	vpslld	$2,%xmm0,%xmm0
+	addl	%edi,%ecx
+	andl	%ebp,%esi
+	xorl	%eax,%ebp
+	addl	%edx,%ecx
+	shrdl	$7,%edx,%edx
+	xorl	%eax,%esi
+	movl	%ecx,%edi
+	addl	8(%rsp),%ebx
+	vpor	%xmm8,%xmm0,%xmm0
+	xorl	%ebp,%edx
+	shldl	$5,%ecx,%ecx
+	addl	%esi,%ebx
+	andl	%edx,%edi
+	xorl	%ebp,%edx
+	addl	%ecx,%ebx
+	addl	12(%rsp),%eax
+	xorl	%ebp,%edi
+	movl	%ebx,%esi
+	shldl	$5,%ebx,%ebx
+	addl	%edi,%eax
+	xorl	%edx,%esi
+	shrdl	$7,%ecx,%ecx
+	addl	%ebx,%eax
+	vpalignr	$8,%xmm7,%xmm0,%xmm8
+	vpxor	%xmm5,%xmm1,%xmm1
+	addl	16(%rsp),%ebp
+	xorl	%ecx,%esi
+	movl	%eax,%edi
+	shldl	$5,%eax,%eax
+	vpxor	%xmm2,%xmm1,%xmm1
+	addl	%esi,%ebp
+	xorl	%ecx,%edi
+	vpaddd	%xmm0,%xmm11,%xmm9
+	shrdl	$7,%ebx,%ebx
+	addl	%eax,%ebp
+	vpxor	%xmm8,%xmm1,%xmm1
+	addl	20(%rsp),%edx
+	xorl	%ebx,%edi
+	movl	%ebp,%esi
+	shldl	$5,%ebp,%ebp
+	vpsrld	$30,%xmm1,%xmm8
+	vmovdqa	%xmm9,0(%rsp)
+	addl	%edi,%edx
+	xorl	%ebx,%esi
+	shrdl	$7,%eax,%eax
+	addl	%ebp,%edx
+	vpslld	$2,%xmm1,%xmm1
+	addl	24(%rsp),%ecx
+	xorl	%eax,%esi
+	movl	%edx,%edi
+	shldl	$5,%edx,%edx
+	addl	%esi,%ecx
+	xorl	%eax,%edi
+	shrdl	$7,%ebp,%ebp
+	addl	%edx,%ecx
+	vpor	%xmm8,%xmm1,%xmm1
+	addl	28(%rsp),%ebx
+	xorl	%ebp,%edi
+	movl	%ecx,%esi
+	shldl	$5,%ecx,%ecx
+	addl	%edi,%ebx
+	xorl	%ebp,%esi
+	shrdl	$7,%edx,%edx
+	addl	%ecx,%ebx
+	vpalignr	$8,%xmm0,%xmm1,%xmm8
+	vpxor	%xmm6,%xmm2,%xmm2
+	addl	32(%rsp),%eax
+	xorl	%edx,%esi
+	movl	%ebx,%edi
+	shldl	$5,%ebx,%ebx
+	vpxor	%xmm3,%xmm2,%xmm2
+	addl	%esi,%eax
+	xorl	%edx,%edi
+	vpaddd	%xmm1,%xmm11,%xmm9
+	vmovdqa	0(%r11),%xmm11
+	shrdl	$7,%ecx,%ecx
+	addl	%ebx,%eax
+	vpxor	%xmm8,%xmm2,%xmm2
+	addl	36(%rsp),%ebp
+	xorl	%ecx,%edi
+	movl	%eax,%esi
+	shldl	$5,%eax,%eax
+	vpsrld	$30,%xmm2,%xmm8
+	vmovdqa	%xmm9,16(%rsp)
+	addl	%edi,%ebp
+	xorl	%ecx,%esi
+	shrdl	$7,%ebx,%ebx
+	addl	%eax,%ebp
+	vpslld	$2,%xmm2,%xmm2
+	addl	40(%rsp),%edx
+	xorl	%ebx,%esi
+	movl	%ebp,%edi
+	shldl	$5,%ebp,%ebp
+	addl	%esi,%edx
+	xorl	%ebx,%edi
+	shrdl	$7,%eax,%eax
+	addl	%ebp,%edx
+	vpor	%xmm8,%xmm2,%xmm2
+	addl	44(%rsp),%ecx
+	xorl	%eax,%edi
+	movl	%edx,%esi
+	shldl	$5,%edx,%edx
+	addl	%edi,%ecx
+	xorl	%eax,%esi
+	shrdl	$7,%ebp,%ebp
+	addl	%edx,%ecx
+	vpalignr	$8,%xmm1,%xmm2,%xmm8
+	vpxor	%xmm7,%xmm3,%xmm3
+	addl	48(%rsp),%ebx
+	xorl	%ebp,%esi
+	movl	%ecx,%edi
+	shldl	$5,%ecx,%ecx
+	vpxor	%xmm4,%xmm3,%xmm3
+	addl	%esi,%ebx
+	xorl	%ebp,%edi
+	vpaddd	%xmm2,%xmm11,%xmm9
+	shrdl	$7,%edx,%edx
+	addl	%ecx,%ebx
+	vpxor	%xmm8,%xmm3,%xmm3
+	addl	52(%rsp),%eax
+	xorl	%edx,%edi
+	movl	%ebx,%esi
+	shldl	$5,%ebx,%ebx
+	vpsrld	$30,%xmm3,%xmm8
+	vmovdqa	%xmm9,32(%rsp)
+	addl	%edi,%eax
+	xorl	%edx,%esi
+	shrdl	$7,%ecx,%ecx
+	addl	%ebx,%eax
+	vpslld	$2,%xmm3,%xmm3
+	addl	56(%rsp),%ebp
+	xorl	%ecx,%esi
+	movl	%eax,%edi
+	shldl	$5,%eax,%eax
+	addl	%esi,%ebp
+	xorl	%ecx,%edi
+	shrdl	$7,%ebx,%ebx
+	addl	%eax,%ebp
+	vpor	%xmm8,%xmm3,%xmm3
+	addl	60(%rsp),%edx
+	xorl	%ebx,%edi
+	movl	%ebp,%esi
+	shldl	$5,%ebp,%ebp
+	addl	%edi,%edx
+	xorl	%ebx,%esi
+	shrdl	$7,%eax,%eax
+	addl	%ebp,%edx
+	vpalignr	$8,%xmm2,%xmm3,%xmm8
+	vpxor	%xmm0,%xmm4,%xmm4
+	addl	0(%rsp),%ecx
+	xorl	%eax,%esi
+	movl	%edx,%edi
+	shldl	$5,%edx,%edx
+	vpxor	%xmm5,%xmm4,%xmm4
+	addl	%esi,%ecx
+	xorl	%eax,%edi
+	vpaddd	%xmm3,%xmm11,%xmm9
+	shrdl	$7,%ebp,%ebp
+	addl	%edx,%ecx
+	vpxor	%xmm8,%xmm4,%xmm4
+	addl	4(%rsp),%ebx
+	xorl	%ebp,%edi
+	movl	%ecx,%esi
+	shldl	$5,%ecx,%ecx
+	vpsrld	$30,%xmm4,%xmm8
+	vmovdqa	%xmm9,48(%rsp)
+	addl	%edi,%ebx
+	xorl	%ebp,%esi
+	shrdl	$7,%edx,%edx
+	addl	%ecx,%ebx
+	vpslld	$2,%xmm4,%xmm4
+	addl	8(%rsp),%eax
+	xorl	%edx,%esi
+	movl	%ebx,%edi
+	shldl	$5,%ebx,%ebx
+	addl	%esi,%eax
+	xorl	%edx,%edi
+	shrdl	$7,%ecx,%ecx
+	addl	%ebx,%eax
+	vpor	%xmm8,%xmm4,%xmm4
+	addl	12(%rsp),%ebp
+	xorl	%ecx,%edi
+	movl	%eax,%esi
+	shldl	$5,%eax,%eax
+	addl	%edi,%ebp
+	xorl	%ecx,%esi
+	shrdl	$7,%ebx,%ebx
+	addl	%eax,%ebp
+	vpalignr	$8,%xmm3,%xmm4,%xmm8
+	vpxor	%xmm1,%xmm5,%xmm5
+	addl	16(%rsp),%edx
+	xorl	%ebx,%esi
+	movl	%ebp,%edi
+	shldl	$5,%ebp,%ebp
+	vpxor	%xmm6,%xmm5,%xmm5
+	addl	%esi,%edx
+	xorl	%ebx,%edi
+	vpaddd	%xmm4,%xmm11,%xmm9
+	shrdl	$7,%eax,%eax
+	addl	%ebp,%edx
+	vpxor	%xmm8,%xmm5,%xmm5
+	addl	20(%rsp),%ecx
+	xorl	%eax,%edi
+	movl	%edx,%esi
+	shldl	$5,%edx,%edx
+	vpsrld	$30,%xmm5,%xmm8
+	vmovdqa	%xmm9,0(%rsp)
+	addl	%edi,%ecx
+	xorl	%eax,%esi
+	shrdl	$7,%ebp,%ebp
+	addl	%edx,%ecx
+	vpslld	$2,%xmm5,%xmm5
+	addl	24(%rsp),%ebx
+	xorl	%ebp,%esi
+	movl	%ecx,%edi
+	shldl	$5,%ecx,%ecx
+	addl	%esi,%ebx
+	xorl	%ebp,%edi
+	shrdl	$7,%edx,%edx
+	addl	%ecx,%ebx
+	vpor	%xmm8,%xmm5,%xmm5
+	addl	28(%rsp),%eax
+	shrdl	$7,%ecx,%ecx
+	movl	%ebx,%esi
+	xorl	%edx,%edi
+	shldl	$5,%ebx,%ebx
+	addl	%edi,%eax
+	xorl	%ecx,%esi
+	xorl	%edx,%ecx
+	addl	%ebx,%eax
+	vpalignr	$8,%xmm4,%xmm5,%xmm8
+	vpxor	%xmm2,%xmm6,%xmm6
+	addl	32(%rsp),%ebp
+	andl	%ecx,%esi
+	xorl	%edx,%ecx
+	shrdl	$7,%ebx,%ebx
+	vpxor	%xmm7,%xmm6,%xmm6
+	movl	%eax,%edi
+	xorl	%ecx,%esi
+	vpaddd	%xmm5,%xmm11,%xmm9
+	shldl	$5,%eax,%eax
+	addl	%esi,%ebp
+	vpxor	%xmm8,%xmm6,%xmm6
+	xorl	%ebx,%edi
+	xorl	%ecx,%ebx
+	addl	%eax,%ebp
+	addl	36(%rsp),%edx
+	vpsrld	$30,%xmm6,%xmm8
+	vmovdqa	%xmm9,16(%rsp)
+	andl	%ebx,%edi
+	xorl	%ecx,%ebx
+	shrdl	$7,%eax,%eax
+	movl	%ebp,%esi
+	vpslld	$2,%xmm6,%xmm6
+	xorl	%ebx,%edi
+	shldl	$5,%ebp,%ebp
+	addl	%edi,%edx
+	xorl	%eax,%esi
+	xorl	%ebx,%eax
+	addl	%ebp,%edx
+	addl	40(%rsp),%ecx
+	andl	%eax,%esi
+	vpor	%xmm8,%xmm6,%xmm6
+	xorl	%ebx,%eax
+	shrdl	$7,%ebp,%ebp
+	movl	%edx,%edi
+	xorl	%eax,%esi
+	shldl	$5,%edx,%edx
+	addl	%esi,%ecx
+	xorl	%ebp,%edi
+	xorl	%eax,%ebp
+	addl	%edx,%ecx
+	addl	44(%rsp),%ebx
+	andl	%ebp,%edi
+	xorl	%eax,%ebp
+	shrdl	$7,%edx,%edx
+	movl	%ecx,%esi
+	xorl	%ebp,%edi
+	shldl	$5,%ecx,%ecx
+	addl	%edi,%ebx
+	xorl	%edx,%esi
+	xorl	%ebp,%edx
+	addl	%ecx,%ebx
+	vpalignr	$8,%xmm5,%xmm6,%xmm8
+	vpxor	%xmm3,%xmm7,%xmm7
+	addl	48(%rsp),%eax
+	andl	%edx,%esi
+	xorl	%ebp,%edx
+	shrdl	$7,%ecx,%ecx
+	vpxor	%xmm0,%xmm7,%xmm7
+	movl	%ebx,%edi
+	xorl	%edx,%esi
+	vpaddd	%xmm6,%xmm11,%xmm9
+	vmovdqa	32(%r11),%xmm11
+	shldl	$5,%ebx,%ebx
+	addl	%esi,%eax
+	vpxor	%xmm8,%xmm7,%xmm7
+	xorl	%ecx,%edi
+	xorl	%edx,%ecx
+	addl	%ebx,%eax
+	addl	52(%rsp),%ebp
+	vpsrld	$30,%xmm7,%xmm8
+	vmovdqa	%xmm9,32(%rsp)
+	andl	%ecx,%edi
+	xorl	%edx,%ecx
+	shrdl	$7,%ebx,%ebx
+	movl	%eax,%esi
+	vpslld	$2,%xmm7,%xmm7
+	xorl	%ecx,%edi
+	shldl	$5,%eax,%eax
+	addl	%edi,%ebp
+	xorl	%ebx,%esi
+	xorl	%ecx,%ebx
+	addl	%eax,%ebp
+	addl	56(%rsp),%edx
+	andl	%ebx,%esi
+	vpor	%xmm8,%xmm7,%xmm7
+	xorl	%ecx,%ebx
+	shrdl	$7,%eax,%eax
+	movl	%ebp,%edi
+	xorl	%ebx,%esi
+	shldl	$5,%ebp,%ebp
+	addl	%esi,%edx
+	xorl	%eax,%edi
+	xorl	%ebx,%eax
+	addl	%ebp,%edx
+	addl	60(%rsp),%ecx
+	andl	%eax,%edi
+	xorl	%ebx,%eax
+	shrdl	$7,%ebp,%ebp
+	movl	%edx,%esi
+	xorl	%eax,%edi
+	shldl	$5,%edx,%edx
+	addl	%edi,%ecx
+	xorl	%ebp,%esi
+	xorl	%eax,%ebp
+	addl	%edx,%ecx
+	vpalignr	$8,%xmm6,%xmm7,%xmm8
+	vpxor	%xmm4,%xmm0,%xmm0
+	addl	0(%rsp),%ebx
+	andl	%ebp,%esi
+	xorl	%eax,%ebp
+	shrdl	$7,%edx,%edx
+	vpxor	%xmm1,%xmm0,%xmm0
+	movl	%ecx,%edi
+	xorl	%ebp,%esi
+	vpaddd	%xmm7,%xmm11,%xmm9
+	shldl	$5,%ecx,%ecx
+	addl	%esi,%ebx
+	vpxor	%xmm8,%xmm0,%xmm0
+	xorl	%edx,%edi
+	xorl	%ebp,%edx
+	addl	%ecx,%ebx
+	addl	4(%rsp),%eax
+	vpsrld	$30,%xmm0,%xmm8
+	vmovdqa	%xmm9,48(%rsp)
+	andl	%edx,%edi
+	xorl	%ebp,%edx
+	shrdl	$7,%ecx,%ecx
+	movl	%ebx,%esi
+	vpslld	$2,%xmm0,%xmm0
+	xorl	%edx,%edi
+	shldl	$5,%ebx,%ebx
+	addl	%edi,%eax
+	xorl	%ecx,%esi
+	xorl	%edx,%ecx
+	addl	%ebx,%eax
+	addl	8(%rsp),%ebp
+	andl	%ecx,%esi
+	vpor	%xmm8,%xmm0,%xmm0
+	xorl	%edx,%ecx
+	shrdl	$7,%ebx,%ebx
+	movl	%eax,%edi
+	xorl	%ecx,%esi
+	shldl	$5,%eax,%eax
+	addl	%esi,%ebp
+	xorl	%ebx,%edi
+	xorl	%ecx,%ebx
+	addl	%eax,%ebp
+	addl	12(%rsp),%edx
+	andl	%ebx,%edi
+	xorl	%ecx,%ebx
+	shrdl	$7,%eax,%eax
+	movl	%ebp,%esi
+	xorl	%ebx,%edi
+	shldl	$5,%ebp,%ebp
+	addl	%edi,%edx
+	xorl	%eax,%esi
+	xorl	%ebx,%eax
+	addl	%ebp,%edx
+	vpalignr	$8,%xmm7,%xmm0,%xmm8
+	vpxor	%xmm5,%xmm1,%xmm1
+	addl	16(%rsp),%ecx
+	andl	%eax,%esi
+	xorl	%ebx,%eax
+	shrdl	$7,%ebp,%ebp
+	vpxor	%xmm2,%xmm1,%xmm1
+	movl	%edx,%edi
+	xorl	%eax,%esi
+	vpaddd	%xmm0,%xmm11,%xmm9
+	shldl	$5,%edx,%edx
+	addl	%esi,%ecx
+	vpxor	%xmm8,%xmm1,%xmm1
+	xorl	%ebp,%edi
+	xorl	%eax,%ebp
+	addl	%edx,%ecx
+	addl	20(%rsp),%ebx
+	vpsrld	$30,%xmm1,%xmm8
+	vmovdqa	%xmm9,0(%rsp)
+	andl	%ebp,%edi
+	xorl	%eax,%ebp
+	shrdl	$7,%edx,%edx
+	movl	%ecx,%esi
+	vpslld	$2,%xmm1,%xmm1
+	xorl	%ebp,%edi
+	shldl	$5,%ecx,%ecx
+	addl	%edi,%ebx
+	xorl	%edx,%esi
+	xorl	%ebp,%edx
+	addl	%ecx,%ebx
+	addl	24(%rsp),%eax
+	andl	%edx,%esi
+	vpor	%xmm8,%xmm1,%xmm1
+	xorl	%ebp,%edx
+	shrdl	$7,%ecx,%ecx
+	movl	%ebx,%edi
+	xorl	%edx,%esi
+	shldl	$5,%ebx,%ebx
+	addl	%esi,%eax
+	xorl	%ecx,%edi
+	xorl	%edx,%ecx
+	addl	%ebx,%eax
+	addl	28(%rsp),%ebp
+	andl	%ecx,%edi
+	xorl	%edx,%ecx
+	shrdl	$7,%ebx,%ebx
+	movl	%eax,%esi
+	xorl	%ecx,%edi
+	shldl	$5,%eax,%eax
+	addl	%edi,%ebp
+	xorl	%ebx,%esi
+	xorl	%ecx,%ebx
+	addl	%eax,%ebp
+	vpalignr	$8,%xmm0,%xmm1,%xmm8
+	vpxor	%xmm6,%xmm2,%xmm2
+	addl	32(%rsp),%edx
+	andl	%ebx,%esi
+	xorl	%ecx,%ebx
+	shrdl	$7,%eax,%eax
+	vpxor	%xmm3,%xmm2,%xmm2
+	movl	%ebp,%edi
+	xorl	%ebx,%esi
+	vpaddd	%xmm1,%xmm11,%xmm9
+	shldl	$5,%ebp,%ebp
+	addl	%esi,%edx
+	vpxor	%xmm8,%xmm2,%xmm2
+	xorl	%eax,%edi
+	xorl	%ebx,%eax
+	addl	%ebp,%edx
+	addl	36(%rsp),%ecx
+	vpsrld	$30,%xmm2,%xmm8
+	vmovdqa	%xmm9,16(%rsp)
+	andl	%eax,%edi
+	xorl	%ebx,%eax
+	shrdl	$7,%ebp,%ebp
+	movl	%edx,%esi
+	vpslld	$2,%xmm2,%xmm2
+	xorl	%eax,%edi
+	shldl	$5,%edx,%edx
+	addl	%edi,%ecx
+	xorl	%ebp,%esi
+	xorl	%eax,%ebp
+	addl	%edx,%ecx
+	addl	40(%rsp),%ebx
+	andl	%ebp,%esi
+	vpor	%xmm8,%xmm2,%xmm2
+	xorl	%eax,%ebp
+	shrdl	$7,%edx,%edx
+	movl	%ecx,%edi
+	xorl	%ebp,%esi
+	shldl	$5,%ecx,%ecx
+	addl	%esi,%ebx
+	xorl	%edx,%edi
+	xorl	%ebp,%edx
+	addl	%ecx,%ebx
+	addl	44(%rsp),%eax
+	andl	%edx,%edi
+	xorl	%ebp,%edx
+	shrdl	$7,%ecx,%ecx
+	movl	%ebx,%esi
+	xorl	%edx,%edi
+	shldl	$5,%ebx,%ebx
+	addl	%edi,%eax
+	xorl	%edx,%esi
+	addl	%ebx,%eax
+	vpalignr	$8,%xmm1,%xmm2,%xmm8
+	vpxor	%xmm7,%xmm3,%xmm3
+	addl	48(%rsp),%ebp
+	xorl	%ecx,%esi
+	movl	%eax,%edi
+	shldl	$5,%eax,%eax
+	vpxor	%xmm4,%xmm3,%xmm3
+	addl	%esi,%ebp
+	xorl	%ecx,%edi
+	vpaddd	%xmm2,%xmm11,%xmm9
+	shrdl	$7,%ebx,%ebx
+	addl	%eax,%ebp
+	vpxor	%xmm8,%xmm3,%xmm3
+	addl	52(%rsp),%edx
+	xorl	%ebx,%edi
+	movl	%ebp,%esi
+	shldl	$5,%ebp,%ebp
+	vpsrld	$30,%xmm3,%xmm8
+	vmovdqa	%xmm9,32(%rsp)
+	addl	%edi,%edx
+	xorl	%ebx,%esi
+	shrdl	$7,%eax,%eax
+	addl	%ebp,%edx
+	vpslld	$2,%xmm3,%xmm3
+	addl	56(%rsp),%ecx
+	xorl	%eax,%esi
+	movl	%edx,%edi
+	shldl	$5,%edx,%edx
+	addl	%esi,%ecx
+	xorl	%eax,%edi
+	shrdl	$7,%ebp,%ebp
+	addl	%edx,%ecx
+	vpor	%xmm8,%xmm3,%xmm3
+	addl	60(%rsp),%ebx
+	xorl	%ebp,%edi
+	movl	%ecx,%esi
+	shldl	$5,%ecx,%ecx
+	addl	%edi,%ebx
+	xorl	%ebp,%esi
+	shrdl	$7,%edx,%edx
+	addl	%ecx,%ebx
+	addl	0(%rsp),%eax
+	vpaddd	%xmm3,%xmm11,%xmm9
+	xorl	%edx,%esi
+	movl	%ebx,%edi
+	shldl	$5,%ebx,%ebx
+	addl	%esi,%eax
+	vmovdqa	%xmm9,48(%rsp)
+	xorl	%edx,%edi
+	shrdl	$7,%ecx,%ecx
+	addl	%ebx,%eax
+	addl	4(%rsp),%ebp
+	xorl	%ecx,%edi
+	movl	%eax,%esi
+	shldl	$5,%eax,%eax
+	addl	%edi,%ebp
+	xorl	%ecx,%esi
+	shrdl	$7,%ebx,%ebx
+	addl	%eax,%ebp
+	addl	8(%rsp),%edx
+	xorl	%ebx,%esi
+	movl	%ebp,%edi
+	shldl	$5,%ebp,%ebp
+	addl	%esi,%edx
+	xorl	%ebx,%edi
+	shrdl	$7,%eax,%eax
+	addl	%ebp,%edx
+	addl	12(%rsp),%ecx
+	xorl	%eax,%edi
+	movl	%edx,%esi
+	shldl	$5,%edx,%edx
+	addl	%edi,%ecx
+	xorl	%eax,%esi
+	shrdl	$7,%ebp,%ebp
+	addl	%edx,%ecx
+	cmpq	%r10,%r9
+	je	.Ldone_avx
+	vmovdqa	64(%r11),%xmm6
+	vmovdqa	-64(%r11),%xmm11
+	vmovdqu	0(%r9),%xmm0
+	vmovdqu	16(%r9),%xmm1
+	vmovdqu	32(%r9),%xmm2
+	vmovdqu	48(%r9),%xmm3
+	vpshufb	%xmm6,%xmm0,%xmm0
+	addq	$64,%r9
+	addl	16(%rsp),%ebx
+	xorl	%ebp,%esi
+	vpshufb	%xmm6,%xmm1,%xmm1
+	movl	%ecx,%edi
+	shldl	$5,%ecx,%ecx
+	vpaddd	%xmm11,%xmm0,%xmm4
+	addl	%esi,%ebx
+	xorl	%ebp,%edi
+	shrdl	$7,%edx,%edx
+	addl	%ecx,%ebx
+	vmovdqa	%xmm4,0(%rsp)
+	addl	20(%rsp),%eax
+	xorl	%edx,%edi
+	movl	%ebx,%esi
+	shldl	$5,%ebx,%ebx
+	addl	%edi,%eax
+	xorl	%edx,%esi
+	shrdl	$7,%ecx,%ecx
+	addl	%ebx,%eax
+	addl	24(%rsp),%ebp
+	xorl	%ecx,%esi
+	movl	%eax,%edi
+	shldl	$5,%eax,%eax
+	addl	%esi,%ebp
+	xorl	%ecx,%edi
+	shrdl	$7,%ebx,%ebx
+	addl	%eax,%ebp
+	addl	28(%rsp),%edx
+	xorl	%ebx,%edi
+	movl	%ebp,%esi
+	shldl	$5,%ebp,%ebp
+	addl	%edi,%edx
+	xorl	%ebx,%esi
+	shrdl	$7,%eax,%eax
+	addl	%ebp,%edx
+	addl	32(%rsp),%ecx
+	xorl	%eax,%esi
+	vpshufb	%xmm6,%xmm2,%xmm2
+	movl	%edx,%edi
+	shldl	$5,%edx,%edx
+	vpaddd	%xmm11,%xmm1,%xmm5
+	addl	%esi,%ecx
+	xorl	%eax,%edi
+	shrdl	$7,%ebp,%ebp
+	addl	%edx,%ecx
+	vmovdqa	%xmm5,16(%rsp)
+	addl	36(%rsp),%ebx
+	xorl	%ebp,%edi
+	movl	%ecx,%esi
+	shldl	$5,%ecx,%ecx
+	addl	%edi,%ebx
+	xorl	%ebp,%esi
+	shrdl	$7,%edx,%edx
+	addl	%ecx,%ebx
+	addl	40(%rsp),%eax
+	xorl	%edx,%esi
+	movl	%ebx,%edi
+	shldl	$5,%ebx,%ebx
+	addl	%esi,%eax
+	xorl	%edx,%edi
+	shrdl	$7,%ecx,%ecx
+	addl	%ebx,%eax
+	addl	44(%rsp),%ebp
+	xorl	%ecx,%edi
+	movl	%eax,%esi
+	shldl	$5,%eax,%eax
+	addl	%edi,%ebp
+	xorl	%ecx,%esi
+	shrdl	$7,%ebx,%ebx
+	addl	%eax,%ebp
+	addl	48(%rsp),%edx
+	xorl	%ebx,%esi
+	vpshufb	%xmm6,%xmm3,%xmm3
+	movl	%ebp,%edi
+	shldl	$5,%ebp,%ebp
+	vpaddd	%xmm11,%xmm2,%xmm6
+	addl	%esi,%edx
+	xorl	%ebx,%edi
+	shrdl	$7,%eax,%eax
+	addl	%ebp,%edx
+	vmovdqa	%xmm6,32(%rsp)
+	addl	52(%rsp),%ecx
+	xorl	%eax,%edi
+	movl	%edx,%esi
+	shldl	$5,%edx,%edx
+	addl	%edi,%ecx
+	xorl	%eax,%esi
+	shrdl	$7,%ebp,%ebp
+	addl	%edx,%ecx
+	addl	56(%rsp),%ebx
+	xorl	%ebp,%esi
+	movl	%ecx,%edi
+	shldl	$5,%ecx,%ecx
+	addl	%esi,%ebx
+	xorl	%ebp,%edi
+	shrdl	$7,%edx,%edx
+	addl	%ecx,%ebx
+	addl	60(%rsp),%eax
+	xorl	%edx,%edi
+	movl	%ebx,%esi
+	shldl	$5,%ebx,%ebx
+	addl	%edi,%eax
+	shrdl	$7,%ecx,%ecx
+	addl	%ebx,%eax
+	addl	0(%r8),%eax
+	addl	4(%r8),%esi
+	addl	8(%r8),%ecx
+	addl	12(%r8),%edx
+	movl	%eax,0(%r8)
+	addl	16(%r8),%ebp
+	movl	%esi,4(%r8)
+	movl	%esi,%ebx
+	movl	%ecx,8(%r8)
+	movl	%ecx,%edi
+	movl	%edx,12(%r8)
+	xorl	%edx,%edi
+	movl	%ebp,16(%r8)
+	andl	%edi,%esi
+	jmp	.Loop_avx
+
+.align	16
+.Ldone_avx:
+	addl	16(%rsp),%ebx
+	xorl	%ebp,%esi
+	movl	%ecx,%edi
+	shldl	$5,%ecx,%ecx
+	addl	%esi,%ebx
+	xorl	%ebp,%edi
+	shrdl	$7,%edx,%edx
+	addl	%ecx,%ebx
+	addl	20(%rsp),%eax
+	xorl	%edx,%edi
+	movl	%ebx,%esi
+	shldl	$5,%ebx,%ebx
+	addl	%edi,%eax
+	xorl	%edx,%esi
+	shrdl	$7,%ecx,%ecx
+	addl	%ebx,%eax
+	addl	24(%rsp),%ebp
+	xorl	%ecx,%esi
+	movl	%eax,%edi
+	shldl	$5,%eax,%eax
+	addl	%esi,%ebp
+	xorl	%ecx,%edi
+	shrdl	$7,%ebx,%ebx
+	addl	%eax,%ebp
+	addl	28(%rsp),%edx
+	xorl	%ebx,%edi
+	movl	%ebp,%esi
+	shldl	$5,%ebp,%ebp
+	addl	%edi,%edx
+	xorl	%ebx,%esi
+	shrdl	$7,%eax,%eax
+	addl	%ebp,%edx
+	addl	32(%rsp),%ecx
+	xorl	%eax,%esi
+	movl	%edx,%edi
+	shldl	$5,%edx,%edx
+	addl	%esi,%ecx
+	xorl	%eax,%edi
+	shrdl	$7,%ebp,%ebp
+	addl	%edx,%ecx
+	addl	36(%rsp),%ebx
+	xorl	%ebp,%edi
+	movl	%ecx,%esi
+	shldl	$5,%ecx,%ecx
+	addl	%edi,%ebx
+	xorl	%ebp,%esi
+	shrdl	$7,%edx,%edx
+	addl	%ecx,%ebx
+	addl	40(%rsp),%eax
+	xorl	%edx,%esi
+	movl	%ebx,%edi
+	shldl	$5,%ebx,%ebx
+	addl	%esi,%eax
+	xorl	%edx,%edi
+	shrdl	$7,%ecx,%ecx
+	addl	%ebx,%eax
+	addl	44(%rsp),%ebp
+	xorl	%ecx,%edi
+	movl	%eax,%esi
+	shldl	$5,%eax,%eax
+	addl	%edi,%ebp
+	xorl	%ecx,%esi
+	shrdl	$7,%ebx,%ebx
+	addl	%eax,%ebp
+	addl	48(%rsp),%edx
+	xorl	%ebx,%esi
+	movl	%ebp,%edi
+	shldl	$5,%ebp,%ebp
+	addl	%esi,%edx
+	xorl	%ebx,%edi
+	shrdl	$7,%eax,%eax
+	addl	%ebp,%edx
+	addl	52(%rsp),%ecx
+	xorl	%eax,%edi
+	movl	%edx,%esi
+	shldl	$5,%edx,%edx
+	addl	%edi,%ecx
+	xorl	%eax,%esi
+	shrdl	$7,%ebp,%ebp
+	addl	%edx,%ecx
+	addl	56(%rsp),%ebx
+	xorl	%ebp,%esi
+	movl	%ecx,%edi
+	shldl	$5,%ecx,%ecx
+	addl	%esi,%ebx
+	xorl	%ebp,%edi
+	shrdl	$7,%edx,%edx
+	addl	%ecx,%ebx
+	addl	60(%rsp),%eax
+	xorl	%edx,%edi
+	movl	%ebx,%esi
+	shldl	$5,%ebx,%ebx
+	addl	%edi,%eax
+	shrdl	$7,%ecx,%ecx
+	addl	%ebx,%eax
+	vzeroupper
+
+	addl	0(%r8),%eax
+	addl	4(%r8),%esi
+	addl	8(%r8),%ecx
+	movl	%eax,0(%r8)
+	addl	12(%r8),%edx
+	movl	%esi,4(%r8)
+	addl	16(%r8),%ebp
+	movl	%ecx,8(%r8)
+	movl	%edx,12(%r8)
+	movl	%ebp,16(%r8)
+	leaq	(%r14),%rsi
+	movq	-40(%rsi),%r14
+	movq	-32(%rsi),%r13
+	movq	-24(%rsi),%r12
+	movq	-16(%rsi),%rbp
+	movq	-8(%rsi),%rbx
+	leaq	(%rsi),%rsp
+.Lepilogue_avx:
+	.byte	0xf3,0xc3
+.size	sha1_block_data_order_avx,.-sha1_block_data_order_avx
 .align	64
 K_XX_XX:
 .long	0x5a827999,0x5a827999,0x5a827999,0x5a827999

third_party/boringssl/linux-x86_64/crypto/sha/sha256-x86_64.S

diff --git a/third_party/boringssl/linux-x86_64/crypto/sha/sha256-x86_64.S b/third_party/boringssl/linux-x86_64/crypto/sha/sha256-x86_64.S
index f526de5..445b497 100644
--- a/third_party/boringssl/linux-x86_64/crypto/sha/sha256-x86_64.S
+++ b/third_party/boringssl/linux-x86_64/crypto/sha/sha256-x86_64.S
@@ -12,6 +12,11 @@
 	movl	0(%r11),%r9d
 	movl	4(%r11),%r10d
 	movl	8(%r11),%r11d
+	andl	$1073741824,%r9d
+	andl	$268435968,%r10d
+	orl	%r9d,%r10d
+	cmpl	$1342177792,%r10d
+	je	.Lavx_shortcut
 	testl	$512,%r10d
 	jnz	.Lssse3_shortcut
 	pushq	%rbx
@@ -2841,4 +2846,1061 @@
 .Lepilogue_ssse3:
 	.byte	0xf3,0xc3
 .size	sha256_block_data_order_ssse3,.-sha256_block_data_order_ssse3
+.type	sha256_block_data_order_avx,@function
+.align	64
+sha256_block_data_order_avx:
+.Lavx_shortcut:
+	pushq	%rbx
+	pushq	%rbp
+	pushq	%r12
+	pushq	%r13
+	pushq	%r14
+	pushq	%r15
+	movq	%rsp,%r11
+	shlq	$4,%rdx
+	subq	$96,%rsp
+	leaq	(%rsi,%rdx,4),%rdx
+	andq	$-64,%rsp
+	movq	%rdi,64+0(%rsp)
+	movq	%rsi,64+8(%rsp)
+	movq	%rdx,64+16(%rsp)
+	movq	%r11,64+24(%rsp)
+.Lprologue_avx:
+
+	vzeroupper
+	movl	0(%rdi),%eax
+	movl	4(%rdi),%ebx
+	movl	8(%rdi),%ecx
+	movl	12(%rdi),%edx
+	movl	16(%rdi),%r8d
+	movl	20(%rdi),%r9d
+	movl	24(%rdi),%r10d
+	movl	28(%rdi),%r11d
+	vmovdqa	K256+512+32(%rip),%xmm8
+	vmovdqa	K256+512+64(%rip),%xmm9
+	jmp	.Lloop_avx
+.align	16
+.Lloop_avx:
+	vmovdqa	K256+512(%rip),%xmm7
+	vmovdqu	0(%rsi),%xmm0
+	vmovdqu	16(%rsi),%xmm1
+	vmovdqu	32(%rsi),%xmm2
+	vmovdqu	48(%rsi),%xmm3
+	vpshufb	%xmm7,%xmm0,%xmm0
+	leaq	K256(%rip),%rbp
+	vpshufb	%xmm7,%xmm1,%xmm1
+	vpshufb	%xmm7,%xmm2,%xmm2
+	vpaddd	0(%rbp),%xmm0,%xmm4
+	vpshufb	%xmm7,%xmm3,%xmm3
+	vpaddd	32(%rbp),%xmm1,%xmm5
+	vpaddd	64(%rbp),%xmm2,%xmm6
+	vpaddd	96(%rbp),%xmm3,%xmm7
+	vmovdqa	%xmm4,0(%rsp)
+	movl	%eax,%r14d
+	vmovdqa	%xmm5,16(%rsp)
+	movl	%ebx,%edi
+	vmovdqa	%xmm6,32(%rsp)
+	xorl	%ecx,%edi
+	vmovdqa	%xmm7,48(%rsp)
+	movl	%r8d,%r13d
+	jmp	.Lavx_00_47
+
+.align	16
+.Lavx_00_47:
+	subq	$-128,%rbp
+	vpalignr	$4,%xmm0,%xmm1,%xmm4
+	shrdl	$14,%r13d,%r13d
+	movl	%r14d,%eax
+	movl	%r9d,%r12d
+	vpalignr	$4,%xmm2,%xmm3,%xmm7
+	shrdl	$9,%r14d,%r14d
+	xorl	%r8d,%r13d
+	xorl	%r10d,%r12d
+	vpsrld	$7,%xmm4,%xmm6
+	shrdl	$5,%r13d,%r13d
+	xorl	%eax,%r14d
+	andl	%r8d,%r12d
+	vpaddd	%xmm7,%xmm0,%xmm0
+	xorl	%r8d,%r13d
+	addl	0(%rsp),%r11d
+	movl	%eax,%r15d
+	vpsrld	$3,%xmm4,%xmm7
+	xorl	%r10d,%r12d
+	shrdl	$11,%r14d,%r14d
+	xorl	%ebx,%r15d
+	vpslld	$14,%xmm4,%xmm5
+	addl	%r12d,%r11d
+	shrdl	$6,%r13d,%r13d
+	andl	%r15d,%edi
+	vpxor	%xmm6,%xmm7,%xmm4
+	xorl	%eax,%r14d
+	addl	%r13d,%r11d
+	xorl	%ebx,%edi
+	vpshufd	$250,%xmm3,%xmm7
+	shrdl	$2,%r14d,%r14d
+	addl	%r11d,%edx
+	addl	%edi,%r11d
+	vpsrld	$11,%xmm6,%xmm6
+	movl	%edx,%r13d
+	addl	%r11d,%r14d
+	shrdl	$14,%r13d,%r13d
+	vpxor	%xmm5,%xmm4,%xmm4
+	movl	%r14d,%r11d
+	movl	%r8d,%r12d
+	shrdl	$9,%r14d,%r14d
+	vpslld	$11,%xmm5,%xmm5
+	xorl	%edx,%r13d
+	xorl	%r9d,%r12d
+	shrdl	$5,%r13d,%r13d
+	vpxor	%xmm6,%xmm4,%xmm4
+	xorl	%r11d,%r14d
+	andl	%edx,%r12d
+	xorl	%edx,%r13d
+	vpsrld	$10,%xmm7,%xmm6
+	addl	4(%rsp),%r10d
+	movl	%r11d,%edi
+	xorl	%r9d,%r12d
+	vpxor	%xmm5,%xmm4,%xmm4
+	shrdl	$11,%r14d,%r14d
+	xorl	%eax,%edi
+	addl	%r12d,%r10d
+	vpsrlq	$17,%xmm7,%xmm7
+	shrdl	$6,%r13d,%r13d
+	andl	%edi,%r15d
+	xorl	%r11d,%r14d
+	vpaddd	%xmm4,%xmm0,%xmm0
+	addl	%r13d,%r10d
+	xorl	%eax,%r15d
+	shrdl	$2,%r14d,%r14d
+	vpxor	%xmm7,%xmm6,%xmm6
+	addl	%r10d,%ecx
+	addl	%r15d,%r10d
+	movl	%ecx,%r13d
+	vpsrlq	$2,%xmm7,%xmm7
+	addl	%r10d,%r14d
+	shrdl	$14,%r13d,%r13d
+	movl	%r14d,%r10d
+	vpxor	%xmm7,%xmm6,%xmm6
+	movl	%edx,%r12d
+	shrdl	$9,%r14d,%r14d
+	xorl	%ecx,%r13d
+	vpshufb	%xmm8,%xmm6,%xmm6
+	xorl	%r8d,%r12d
+	shrdl	$5,%r13d,%r13d
+	xorl	%r10d,%r14d
+	vpaddd	%xmm6,%xmm0,%xmm0
+	andl	%ecx,%r12d
+	xorl	%ecx,%r13d
+	addl	8(%rsp),%r9d
+	vpshufd	$80,%xmm0,%xmm7
+	movl	%r10d,%r15d
+	xorl	%r8d,%r12d
+	shrdl	$11,%r14d,%r14d
+	vpsrld	$10,%xmm7,%xmm6
+	xorl	%r11d,%r15d
+	addl	%r12d,%r9d
+	shrdl	$6,%r13d,%r13d
+	vpsrlq	$17,%xmm7,%xmm7
+	andl	%r15d,%edi
+	xorl	%r10d,%r14d
+	addl	%r13d,%r9d
+	vpxor	%xmm7,%xmm6,%xmm6
+	xorl	%r11d,%edi
+	shrdl	$2,%r14d,%r14d
+	addl	%r9d,%ebx
+	vpsrlq	$2,%xmm7,%xmm7
+	addl	%edi,%r9d
+	movl	%ebx,%r13d
+	addl	%r9d,%r14d
+	vpxor	%xmm7,%xmm6,%xmm6
+	shrdl	$14,%r13d,%r13d
+	movl	%r14d,%r9d
+	movl	%ecx,%r12d
+	vpshufb	%xmm9,%xmm6,%xmm6
+	shrdl	$9,%r14d,%r14d
+	xorl	%ebx,%r13d
+	xorl	%edx,%r12d
+	vpaddd	%xmm6,%xmm0,%xmm0
+	shrdl	$5,%r13d,%r13d
+	xorl	%r9d,%r14d
+	andl	%ebx,%r12d
+	vpaddd	0(%rbp),%xmm0,%xmm6
+	xorl	%ebx,%r13d
+	addl	12(%rsp),%r8d
+	movl	%r9d,%edi
+	xorl	%edx,%r12d
+	shrdl	$11,%r14d,%r14d
+	xorl	%r10d,%edi
+	addl	%r12d,%r8d
+	shrdl	$6,%r13d,%r13d
+	andl	%edi,%r15d
+	xorl	%r9d,%r14d
+	addl	%r13d,%r8d
+	xorl	%r10d,%r15d
+	shrdl	$2,%r14d,%r14d
+	addl	%r8d,%eax
+	addl	%r15d,%r8d
+	movl	%eax,%r13d
+	addl	%r8d,%r14d
+	vmovdqa	%xmm6,0(%rsp)
+	vpalignr	$4,%xmm1,%xmm2,%xmm4
+	shrdl	$14,%r13d,%r13d
+	movl	%r14d,%r8d
+	movl	%ebx,%r12d
+	vpalignr	$4,%xmm3,%xmm0,%xmm7
+	shrdl	$9,%r14d,%r14d
+	xorl	%eax,%r13d
+	xorl	%ecx,%r12d
+	vpsrld	$7,%xmm4,%xmm6
+	shrdl	$5,%r13d,%r13d
+	xorl	%r8d,%r14d
+	andl	%eax,%r12d
+	vpaddd	%xmm7,%xmm1,%xmm1
+	xorl	%eax,%r13d
+	addl	16(%rsp),%edx
+	movl	%r8d,%r15d
+	vpsrld	$3,%xmm4,%xmm7
+	xorl	%ecx,%r12d
+	shrdl	$11,%r14d,%r14d
+	xorl	%r9d,%r15d
+	vpslld	$14,%xmm4,%xmm5
+	addl	%r12d,%edx
+	shrdl	$6,%r13d,%r13d
+	andl	%r15d,%edi
+	vpxor	%xmm6,%xmm7,%xmm4
+	xorl	%r8d,%r14d
+	addl	%r13d,%edx
+	xorl	%r9d,%edi
+	vpshufd	$250,%xmm0,%xmm7
+	shrdl	$2,%r14d,%r14d
+	addl	%edx,%r11d
+	addl	%edi,%edx
+	vpsrld	$11,%xmm6,%xmm6
+	movl	%r11d,%r13d
+	addl	%edx,%r14d
+	shrdl	$14,%r13d,%r13d
+	vpxor	%xmm5,%xmm4,%xmm4
+	movl	%r14d,%edx
+	movl	%eax,%r12d
+	shrdl	$9,%r14d,%r14d
+	vpslld	$11,%xmm5,%xmm5
+	xorl	%r11d,%r13d
+	xorl	%ebx,%r12d
+	shrdl	$5,%r13d,%r13d
+	vpxor	%xmm6,%xmm4,%xmm4
+	xorl	%edx,%r14d
+	andl	%r11d,%r12d
+	xorl	%r11d,%r13d
+	vpsrld	$10,%xmm7,%xmm6
+	addl	20(%rsp),%ecx
+	movl	%edx,%edi
+	xorl	%ebx,%r12d
+	vpxor	%xmm5,%xmm4,%xmm4
+	shrdl	$11,%r14d,%r14d
+	xorl	%r8d,%edi
+	addl	%r12d,%ecx
+	vpsrlq	$17,%xmm7,%xmm7
+	shrdl	$6,%r13d,%r13d
+	andl	%edi,%r15d
+	xorl	%edx,%r14d
+	vpaddd	%xmm4,%xmm1,%xmm1
+	addl	%r13d,%ecx
+	xorl	%r8d,%r15d
+	shrdl	$2,%r14d,%r14d
+	vpxor	%xmm7,%xmm6,%xmm6
+	addl	%ecx,%r10d
+	addl	%r15d,%ecx
+	movl	%r10d,%r13d
+	vpsrlq	$2,%xmm7,%xmm7
+	addl	%ecx,%r14d
+	shrdl	$14,%r13d,%r13d
+	movl	%r14d,%ecx
+	vpxor	%xmm7,%xmm6,%xmm6
+	movl	%r11d,%r12d
+	shrdl	$9,%r14d,%r14d
+	xorl	%r10d,%r13d
+	vpshufb	%xmm8,%xmm6,%xmm6
+	xorl	%eax,%r12d
+	shrdl	$5,%r13d,%r13d
+	xorl	%ecx,%r14d
+	vpaddd	%xmm6,%xmm1,%xmm1
+	andl	%r10d,%r12d
+	xorl	%r10d,%r13d
+	addl	24(%rsp),%ebx
+	vpshufd	$80,%xmm1,%xmm7
+	movl	%ecx,%r15d
+	xorl	%eax,%r12d
+	shrdl	$11,%r14d,%r14d
+	vpsrld	$10,%xmm7,%xmm6
+	xorl	%edx,%r15d
+	addl	%r12d,%ebx
+	shrdl	$6,%r13d,%r13d
+	vpsrlq	$17,%xmm7,%xmm7
+	andl	%r15d,%edi
+	xorl	%ecx,%r14d
+	addl	%r13d,%ebx
+	vpxor	%xmm7,%xmm6,%xmm6
+	xorl	%edx,%edi
+	shrdl	$2,%r14d,%r14d
+	addl	%ebx,%r9d
+	vpsrlq	$2,%xmm7,%xmm7
+	addl	%edi,%ebx
+	movl	%r9d,%r13d
+	addl	%ebx,%r14d
+	vpxor	%xmm7,%xmm6,%xmm6
+	shrdl	$14,%r13d,%r13d
+	movl	%r14d,%ebx
+	movl	%r10d,%r12d
+	vpshufb	%xmm9,%xmm6,%xmm6
+	shrdl	$9,%r14d,%r14d
+	xorl	%r9d,%r13d
+	xorl	%r11d,%r12d
+	vpaddd	%xmm6,%xmm1,%xmm1
+	shrdl	$5,%r13d,%r13d
+	xorl	%ebx,%r14d
+	andl	%r9d,%r12d
+	vpaddd	32(%rbp),%xmm1,%xmm6
+	xorl	%r9d,%r13d
+	addl	28(%rsp),%eax
+	movl	%ebx,%edi
+	xorl	%r11d,%r12d
+	shrdl	$11,%r14d,%r14d
+	xorl	%ecx,%edi
+	addl	%r12d,%eax
+	shrdl	$6,%r13d,%r13d
+	andl	%edi,%r15d
+	xorl	%ebx,%r14d
+	addl	%r13d,%eax
+	xorl	%ecx,%r15d
+	shrdl	$2,%r14d,%r14d
+	addl	%eax,%r8d
+	addl	%r15d,%eax
+	movl	%r8d,%r13d
+	addl	%eax,%r14d
+	vmovdqa	%xmm6,16(%rsp)
+	vpalignr	$4,%xmm2,%xmm3,%xmm4
+	shrdl	$14,%r13d,%r13d
+	movl	%r14d,%eax
+	movl	%r9d,%r12d
+	vpalignr	$4,%xmm0,%xmm1,%xmm7
+	shrdl	$9,%r14d,%r14d
+	xorl	%r8d,%r13d
+	xorl	%r10d,%r12d
+	vpsrld	$7,%xmm4,%xmm6
+	shrdl	$5,%r13d,%r13d
+	xorl	%eax,%r14d
+	andl	%r8d,%r12d
+	vpaddd	%xmm7,%xmm2,%xmm2
+	xorl	%r8d,%r13d
+	addl	32(%rsp),%r11d
+	movl	%eax,%r15d
+	vpsrld	$3,%xmm4,%xmm7
+	xorl	%r10d,%r12d
+	shrdl	$11,%r14d,%r14d
+	xorl	%ebx,%r15d
+	vpslld	$14,%xmm4,%xmm5
+	addl	%r12d,%r11d
+	shrdl	$6,%r13d,%r13d
+	andl	%r15d,%edi
+	vpxor	%xmm6,%xmm7,%xmm4
+	xorl	%eax,%r14d
+	addl	%r13d,%r11d
+	xorl	%ebx,%edi
+	vpshufd	$250,%xmm1,%xmm7
+	shrdl	$2,%r14d,%r14d
+	addl	%r11d,%edx
+	addl	%edi,%r11d
+	vpsrld	$11,%xmm6,%xmm6
+	movl	%edx,%r13d
+	addl	%r11d,%r14d
+	shrdl	$14,%r13d,%r13d
+	vpxor	%xmm5,%xmm4,%xmm4
+	movl	%r14d,%r11d
+	movl	%r8d,%r12d
+	shrdl	$9,%r14d,%r14d
+	vpslld	$11,%xmm5,%xmm5
+	xorl	%edx,%r13d
+	xorl	%r9d,%r12d
+	shrdl	$5,%r13d,%r13d
+	vpxor	%xmm6,%xmm4,%xmm4
+	xorl	%r11d,%r14d
+	andl	%edx,%r12d
+	xorl	%edx,%r13d
+	vpsrld	$10,%xmm7,%xmm6
+	addl	36(%rsp),%r10d
+	movl	%r11d,%edi
+	xorl	%r9d,%r12d
+	vpxor	%xmm5,%xmm4,%xmm4
+	shrdl	$11,%r14d,%r14d
+	xorl	%eax,%edi
+	addl	%r12d,%r10d
+	vpsrlq	$17,%xmm7,%xmm7
+	shrdl	$6,%r13d,%r13d
+	andl	%edi,%r15d
+	xorl	%r11d,%r14d
+	vpaddd	%xmm4,%xmm2,%xmm2
+	addl	%r13d,%r10d
+	xorl	%eax,%r15d
+	shrdl	$2,%r14d,%r14d
+	vpxor	%xmm7,%xmm6,%xmm6
+	addl	%r10d,%ecx
+	addl	%r15d,%r10d
+	movl	%ecx,%r13d
+	vpsrlq	$2,%xmm7,%xmm7
+	addl	%r10d,%r14d
+	shrdl	$14,%r13d,%r13d
+	movl	%r14d,%r10d
+	vpxor	%xmm7,%xmm6,%xmm6
+	movl	%edx,%r12d
+	shrdl	$9,%r14d,%r14d
+	xorl	%ecx,%r13d
+	vpshufb	%xmm8,%xmm6,%xmm6
+	xorl	%r8d,%r12d
+	shrdl	$5,%r13d,%r13d
+	xorl	%r10d,%r14d
+	vpaddd	%xmm6,%xmm2,%xmm2
+	andl	%ecx,%r12d
+	xorl	%ecx,%r13d
+	addl	40(%rsp),%r9d
+	vpshufd	$80,%xmm2,%xmm7
+	movl	%r10d,%r15d
+	xorl	%r8d,%r12d
+	shrdl	$11,%r14d,%r14d
+	vpsrld	$10,%xmm7,%xmm6
+	xorl	%r11d,%r15d
+	addl	%r12d,%r9d
+	shrdl	$6,%r13d,%r13d
+	vpsrlq	$17,%xmm7,%xmm7
+	andl	%r15d,%edi
+	xorl	%r10d,%r14d
+	addl	%r13d,%r9d
+	vpxor	%xmm7,%xmm6,%xmm6
+	xorl	%r11d,%edi
+	shrdl	$2,%r14d,%r14d
+	addl	%r9d,%ebx
+	vpsrlq	$2,%xmm7,%xmm7
+	addl	%edi,%r9d
+	movl	%ebx,%r13d
+	addl	%r9d,%r14d
+	vpxor	%xmm7,%xmm6,%xmm6
+	shrdl	$14,%r13d,%r13d
+	movl	%r14d,%r9d
+	movl	%ecx,%r12d
+	vpshufb	%xmm9,%xmm6,%xmm6
+	shrdl	$9,%r14d,%r14d
+	xorl	%ebx,%r13d
+	xorl	%edx,%r12d
+	vpaddd	%xmm6,%xmm2,%xmm2
+	shrdl	$5,%r13d,%r13d
+	xorl	%r9d,%r14d
+	andl	%ebx,%r12d
+	vpaddd	64(%rbp),%xmm2,%xmm6
+	xorl	%ebx,%r13d
+	addl	44(%rsp),%r8d
+	movl	%r9d,%edi
+	xorl	%edx,%r12d
+	shrdl	$11,%r14d,%r14d
+	xorl	%r10d,%edi
+	addl	%r12d,%r8d
+	shrdl	$6,%r13d,%r13d
+	andl	%edi,%r15d
+	xorl	%r9d,%r14d
+	addl	%r13d,%r8d
+	xorl	%r10d,%r15d
+	shrdl	$2,%r14d,%r14d
+	addl	%r8d,%eax
+	addl	%r15d,%r8d
+	movl	%eax,%r13d
+	addl	%r8d,%r14d
+	vmovdqa	%xmm6,32(%rsp)
+	vpalignr	$4,%xmm3,%xmm0,%xmm4
+	shrdl	$14,%r13d,%r13d
+	movl	%r14d,%r8d
+	movl	%ebx,%r12d
+	vpalignr	$4,%xmm1,%xmm2,%xmm7
+	shrdl	$9,%r14d,%r14d
+	xorl	%eax,%r13d
+	xorl	%ecx,%r12d
+	vpsrld	$7,%xmm4,%xmm6
+	shrdl	$5,%r13d,%r13d
+	xorl	%r8d,%r14d
+	andl	%eax,%r12d
+	vpaddd	%xmm7,%xmm3,%xmm3
+	xorl	%eax,%r13d
+	addl	48(%rsp),%edx
+	movl	%r8d,%r15d
+	vpsrld	$3,%xmm4,%xmm7
+	xorl	%ecx,%r12d
+	shrdl	$11,%r14d,%r14d
+	xorl	%r9d,%r15d
+	vpslld	$14,%xmm4,%xmm5
+	addl	%r12d,%edx
+	shrdl	$6,%r13d,%r13d
+	andl	%r15d,%edi
+	vpxor	%xmm6,%xmm7,%xmm4
+	xorl	%r8d,%r14d
+	addl	%r13d,%edx
+	xorl	%r9d,%edi
+	vpshufd	$250,%xmm2,%xmm7
+	shrdl	$2,%r14d,%r14d
+	addl	%edx,%r11d
+	addl	%edi,%edx
+	vpsrld	$11,%xmm6,%xmm6
+	movl	%r11d,%r13d
+	addl	%edx,%r14d
+	shrdl	$14,%r13d,%r13d
+	vpxor	%xmm5,%xmm4,%xmm4
+	movl	%r14d,%edx
+	movl	%eax,%r12d
+	shrdl	$9,%r14d,%r14d
+	vpslld	$11,%xmm5,%xmm5
+	xorl	%r11d,%r13d
+	xorl	%ebx,%r12d
+	shrdl	$5,%r13d,%r13d
+	vpxor	%xmm6,%xmm4,%xmm4
+	xorl	%edx,%r14d
+	andl	%r11d,%r12d
+	xorl	%r11d,%r13d
+	vpsrld	$10,%xmm7,%xmm6
+	addl	52(%rsp),%ecx
+	movl	%edx,%edi
+	xorl	%ebx,%r12d
+	vpxor	%xmm5,%xmm4,%xmm4
+	shrdl	$11,%r14d,%r14d
+	xorl	%r8d,%edi
+	addl	%r12d,%ecx
+	vpsrlq	$17,%xmm7,%xmm7
+	shrdl	$6,%r13d,%r13d
+	andl	%edi,%r15d
+	xorl	%edx,%r14d
+	vpaddd	%xmm4,%xmm3,%xmm3
+	addl	%r13d,%ecx
+	xorl	%r8d,%r15d
+	shrdl	$2,%r14d,%r14d
+	vpxor	%xmm7,%xmm6,%xmm6
+	addl	%ecx,%r10d
+	addl	%r15d,%ecx
+	movl	%r10d,%r13d
+	vpsrlq	$2,%xmm7,%xmm7
+	addl	%ecx,%r14d
+	shrdl	$14,%r13d,%r13d
+	movl	%r14d,%ecx
+	vpxor	%xmm7,%xmm6,%xmm6
+	movl	%r11d,%r12d
+	shrdl	$9,%r14d,%r14d
+	xorl	%r10d,%r13d
+	vpshufb	%xmm8,%xmm6,%xmm6
+	xorl	%eax,%r12d
+	shrdl	$5,%r13d,%r13d
+	xorl	%ecx,%r14d
+	vpaddd	%xmm6,%xmm3,%xmm3
+	andl	%r10d,%r12d
+	xorl	%r10d,%r13d
+	addl	56(%rsp),%ebx
+	vpshufd	$80,%xmm3,%xmm7
+	movl	%ecx,%r15d
+	xorl	%eax,%r12d
+	shrdl	$11,%r14d,%r14d
+	vpsrld	$10,%xmm7,%xmm6
+	xorl	%edx,%r15d
+	addl	%r12d,%ebx
+	shrdl	$6,%r13d,%r13d
+	vpsrlq	$17,%xmm7,%xmm7
+	andl	%r15d,%edi
+	xorl	%ecx,%r14d
+	addl	%r13d,%ebx
+	vpxor	%xmm7,%xmm6,%xmm6
+	xorl	%edx,%edi
+	shrdl	$2,%r14d,%r14d
+	addl	%ebx,%r9d
+	vpsrlq	$2,%xmm7,%xmm7
+	addl	%edi,%ebx
+	movl	%r9d,%r13d
+	addl	%ebx,%r14d
+	vpxor	%xmm7,%xmm6,%xmm6
+	shrdl	$14,%r13d,%r13d
+	movl	%r14d,%ebx
+	movl	%r10d,%r12d
+	vpshufb	%xmm9,%xmm6,%xmm6
+	shrdl	$9,%r14d,%r14d
+	xorl	%r9d,%r13d
+	xorl	%r11d,%r12d
+	vpaddd	%xmm6,%xmm3,%xmm3
+	shrdl	$5,%r13d,%r13d
+	xorl	%ebx,%r14d
+	andl	%r9d,%r12d
+	vpaddd	96(%rbp),%xmm3,%xmm6
+	xorl	%r9d,%r13d
+	addl	60(%rsp),%eax
+	movl	%ebx,%edi
+	xorl	%r11d,%r12d
+	shrdl	$11,%r14d,%r14d
+	xorl	%ecx,%edi
+	addl	%r12d,%eax
+	shrdl	$6,%r13d,%r13d
+	andl	%edi,%r15d
+	xorl	%ebx,%r14d
+	addl	%r13d,%eax
+	xorl	%ecx,%r15d
+	shrdl	$2,%r14d,%r14d
+	addl	%eax,%r8d
+	addl	%r15d,%eax
+	movl	%r8d,%r13d
+	addl	%eax,%r14d
+	vmovdqa	%xmm6,48(%rsp)
+	cmpb	$0,131(%rbp)
+	jne	.Lavx_00_47
+	shrdl	$14,%r13d,%r13d
+	movl	%r14d,%eax
+	movl	%r9d,%r12d
+	shrdl	$9,%r14d,%r14d
+	xorl	%r8d,%r13d
+	xorl	%r10d,%r12d
+	shrdl	$5,%r13d,%r13d
+	xorl	%eax,%r14d
+	andl	%r8d,%r12d
+	xorl	%r8d,%r13d
+	addl	0(%rsp),%r11d
+	movl	%eax,%r15d
+	xorl	%r10d,%r12d
+	shrdl	$11,%r14d,%r14d
+	xorl	%ebx,%r15d
+	addl	%r12d,%r11d
+	shrdl	$6,%r13d,%r13d
+	andl	%r15d,%edi
+	xorl	%eax,%r14d
+	addl	%r13d,%r11d
+	xorl	%ebx,%edi
+	shrdl	$2,%r14d,%r14d
+	addl	%r11d,%edx
+	addl	%edi,%r11d
+	movl	%edx,%r13d
+	addl	%r11d,%r14d
+	shrdl	$14,%r13d,%r13d
+	movl	%r14d,%r11d
+	movl	%r8d,%r12d
+	shrdl	$9,%r14d,%r14d
+	xorl	%edx,%r13d
+	xorl	%r9d,%r12d
+	shrdl	$5,%r13d,%r13d
+	xorl	%r11d,%r14d
+	andl	%edx,%r12d
+	xorl	%edx,%r13d
+	addl	4(%rsp),%r10d
+	movl	%r11d,%edi
+	xorl	%r9d,%r12d
+	shrdl	$11,%r14d,%r14d
+	xorl	%eax,%edi
+	addl	%r12d,%r10d
+	shrdl	$6,%r13d,%r13d
+	andl	%edi,%r15d
+	xorl	%r11d,%r14d
+	addl	%r13d,%r10d
+	xorl	%eax,%r15d
+	shrdl	$2,%r14d,%r14d
+	addl	%r10d,%ecx
+	addl	%r15d,%r10d
+	movl	%ecx,%r13d
+	addl	%r10d,%r14d
+	shrdl	$14,%r13d,%r13d
+	movl	%r14d,%r10d
+	movl	%edx,%r12d
+	shrdl	$9,%r14d,%r14d
+	xorl	%ecx,%r13d
+	xorl	%r8d,%r12d
+	shrdl	$5,%r13d,%r13d
+	xorl	%r10d,%r14d
+	andl	%ecx,%r12d
+	xorl	%ecx,%r13d
+	addl	8(%rsp),%r9d
+	movl	%r10d,%r15d
+	xorl	%r8d,%r12d
+	shrdl	$11,%r14d,%r14d
+	xorl	%r11d,%r15d
+	addl	%r12d,%r9d
+	shrdl	$6,%r13d,%r13d
+	andl	%r15d,%edi
+	xorl	%r10d,%r14d
+	addl	%r13d,%r9d
+	xorl	%r11d,%edi
+	shrdl	$2,%r14d,%r14d
+	addl	%r9d,%ebx
+	addl	%edi,%r9d
+	movl	%ebx,%r13d
+	addl	%r9d,%r14d
+	shrdl	$14,%r13d,%r13d
+	movl	%r14d,%r9d
+	movl	%ecx,%r12d
+	shrdl	$9,%r14d,%r14d
+	xorl	%ebx,%r13d
+	xorl	%edx,%r12d
+	shrdl	$5,%r13d,%r13d
+	xorl	%r9d,%r14d
+	andl	%ebx,%r12d
+	xorl	%ebx,%r13d
+	addl	12(%rsp),%r8d
+	movl	%r9d,%edi
+	xorl	%edx,%r12d
+	shrdl	$11,%r14d,%r14d
+	xorl	%r10d,%edi
+	addl	%r12d,%r8d
+	shrdl	$6,%r13d,%r13d
+	andl	%edi,%r15d
+	xorl	%r9d,%r14d
+	addl	%r13d,%r8d
+	xorl	%r10d,%r15d
+	shrdl	$2,%r14d,%r14d
+	addl	%r8d,%eax
+	addl	%r15d,%r8d
+	movl	%eax,%r13d
+	addl	%r8d,%r14d
+	shrdl	$14,%r13d,%r13d
+	movl	%r14d,%r8d
+	movl	%ebx,%r12d
+	shrdl	$9,%r14d,%r14d
+	xorl	%eax,%r13d
+	xorl	%ecx,%r12d
+	shrdl	$5,%r13d,%r13d
+	xorl	%r8d,%r14d
+	andl	%eax,%r12d
+	xorl	%eax,%r13d
+	addl	16(%rsp),%edx
+	movl	%r8d,%r15d
+	xorl	%ecx,%r12d
+	shrdl	$11,%r14d,%r14d
+	xorl	%r9d,%r15d
+	addl	%r12d,%edx
+	shrdl	$6,%r13d,%r13d
+	andl	%r15d,%edi
+	xorl	%r8d,%r14d
+	addl	%r13d,%edx
+	xorl	%r9d,%edi
+	shrdl	$2,%r14d,%r14d
+	addl	%edx,%r11d
+	addl	%edi,%edx
+	movl	%r11d,%r13d
+	addl	%edx,%r14d
+	shrdl	$14,%r13d,%r13d
+	movl	%r14d,%edx
+	movl	%eax,%r12d
+	shrdl	$9,%r14d,%r14d
+	xorl	%r11d,%r13d
+	xorl	%ebx,%r12d
+	shrdl	$5,%r13d,%r13d
+	xorl	%edx,%r14d
+	andl	%r11d,%r12d
+	xorl	%r11d,%r13d
+	addl	20(%rsp),%ecx
+	movl	%edx,%edi
+	xorl	%ebx,%r12d
+	shrdl	$11,%r14d,%r14d
+	xorl	%r8d,%edi
+	addl	%r12d,%ecx
+	shrdl	$6,%r13d,%r13d
+	andl	%edi,%r15d
+	xorl	%edx,%r14d
+	addl	%r13d,%ecx
+	xorl	%r8d,%r15d
+	shrdl	$2,%r14d,%r14d
+	addl	%ecx,%r10d
+	addl	%r15d,%ecx
+	movl	%r10d,%r13d
+	addl	%ecx,%r14d
+	shrdl	$14,%r13d,%r13d
+	movl	%r14d,%ecx
+	movl	%r11d,%r12d
+	shrdl	$9,%r14d,%r14d
+	xorl	%r10d,%r13d
+	xorl	%eax,%r12d
+	shrdl	$5,%r13d,%r13d
+	xorl	%ecx,%r14d
+	andl	%r10d,%r12d
+	xorl	%r10d,%r13d
+	addl	24(%rsp),%ebx
+	movl	%ecx,%r15d
+	xorl	%eax,%r12d
+	shrdl	$11,%r14d,%r14d
+	xorl	%edx,%r15d
+	addl	%r12d,%ebx
+	shrdl	$6,%r13d,%r13d
+	andl	%r15d,%edi
+	xorl	%ecx,%r14d
+	addl	%r13d,%ebx
+	xorl	%edx,%edi
+	shrdl	$2,%r14d,%r14d
+	addl	%ebx,%r9d
+	addl	%edi,%ebx
+	movl	%r9d,%r13d
+	addl	%ebx,%r14d
+	shrdl	$14,%r13d,%r13d
+	movl	%r14d,%ebx
+	movl	%r10d,%r12d
+	shrdl	$9,%r14d,%r14d
+	xorl	%r9d,%r13d
+	xorl	%r11d,%r12d
+	shrdl	$5,%r13d,%r13d
+	xorl	%ebx,%r14d
+	andl	%r9d,%r12d
+	xorl	%r9d,%r13d
+	addl	28(%rsp),%eax
+	movl	%ebx,%edi
+	xorl	%r11d,%r12d
+	shrdl	$11,%r14d,%r14d
+	xorl	%ecx,%edi
+	addl	%r12d,%eax
+	shrdl	$6,%r13d,%r13d
+	andl	%edi,%r15d
+	xorl	%ebx,%r14d
+	addl	%r13d,%eax
+	xorl	%ecx,%r15d
+	shrdl	$2,%r14d,%r14d
+	addl	%eax,%r8d
+	addl	%r15d,%eax
+	movl	%r8d,%r13d
+	addl	%eax,%r14d
+	shrdl	$14,%r13d,%r13d
+	movl	%r14d,%eax
+	movl	%r9d,%r12d
+	shrdl	$9,%r14d,%r14d
+	xorl	%r8d,%r13d
+	xorl	%r10d,%r12d
+	shrdl	$5,%r13d,%r13d
+	xorl	%eax,%r14d
+	andl	%r8d,%r12d
+	xorl	%r8d,%r13d
+	addl	32(%rsp),%r11d
+	movl	%eax,%r15d
+	xorl	%r10d,%r12d
+	shrdl	$11,%r14d,%r14d
+	xorl	%ebx,%r15d
+	addl	%r12d,%r11d
+	shrdl	$6,%r13d,%r13d
+	andl	%r15d,%edi
+	xorl	%eax,%r14d
+	addl	%r13d,%r11d
+	xorl	%ebx,%edi
+	shrdl	$2,%r14d,%r14d
+	addl	%r11d,%edx
+	addl	%edi,%r11d
+	movl	%edx,%r13d
+	addl	%r11d,%r14d
+	shrdl	$14,%r13d,%r13d
+	movl	%r14d,%r11d
+	movl	%r8d,%r12d
+	shrdl	$9,%r14d,%r14d
+	xorl	%edx,%r13d
+	xorl	%r9d,%r12d
+	shrdl	$5,%r13d,%r13d
+	xorl	%r11d,%r14d
+	andl	%edx,%r12d
+	xorl	%edx,%r13d
+	addl	36(%rsp),%r10d
+	movl	%r11d,%edi
+	xorl	%r9d,%r12d
+	shrdl	$11,%r14d,%r14d
+	xorl	%eax,%edi
+	addl	%r12d,%r10d
+	shrdl	$6,%r13d,%r13d
+	andl	%edi,%r15d
+	xorl	%r11d,%r14d
+	addl	%r13d,%r10d
+	xorl	%eax,%r15d
+	shrdl	$2,%r14d,%r14d
+	addl	%r10d,%ecx
+	addl	%r15d,%r10d
+	movl	%ecx,%r13d
+	addl	%r10d,%r14d
+	shrdl	$14,%r13d,%r13d
+	movl	%r14d,%r10d
+	movl	%edx,%r12d
+	shrdl	$9,%r14d,%r14d
+	xorl	%ecx,%r13d
+	xorl	%r8d,%r12d
+	shrdl	$5,%r13d,%r13d
+	xorl	%r10d,%r14d
+	andl	%ecx,%r12d
+	xorl	%ecx,%r13d
+	addl	40(%rsp),%r9d
+	movl	%r10d,%r15d
+	xorl	%r8d,%r12d
+	shrdl	$11,%r14d,%r14d
+	xorl	%r11d,%r15d
+	addl	%r12d,%r9d
+	shrdl	$6,%r13d,%r13d
+	andl	%r15d,%edi
+	xorl	%r10d,%r14d
+	addl	%r13d,%r9d
+	xorl	%r11d,%edi
+	shrdl	$2,%r14d,%r14d
+	addl	%r9d,%ebx
+	addl	%edi,%r9d
+	movl	%ebx,%r13d
+	addl	%r9d,%r14d
+	shrdl	$14,%r13d,%r13d
+	movl	%r14d,%r9d
+	movl	%ecx,%r12d
+	shrdl	$9,%r14d,%r14d
+	xorl	%ebx,%r13d
+	xorl	%edx,%r12d
+	shrdl	$5,%r13d,%r13d
+	xorl	%r9d,%r14d
+	andl	%ebx,%r12d
+	xorl	%ebx,%r13d
+	addl	44(%rsp),%r8d
+	movl	%r9d,%edi
+	xorl	%edx,%r12d
+	shrdl	$11,%r14d,%r14d
+	xorl	%r10d,%edi
+	addl	%r12d,%r8d
+	shrdl	$6,%r13d,%r13d
+	andl	%edi,%r15d
+	xorl	%r9d,%r14d
+	addl	%r13d,%r8d
+	xorl	%r10d,%r15d
+	shrdl	$2,%r14d,%r14d
+	addl	%r8d,%eax
+	addl	%r15d,%r8d
+	movl	%eax,%r13d
+	addl	%r8d,%r14d
+	shrdl	$14,%r13d,%r13d
+	movl	%r14d,%r8d
+	movl	%ebx,%r12d
+	shrdl	$9,%r14d,%r14d
+	xorl	%eax,%r13d
+	xorl	%ecx,%r12d
+	shrdl	$5,%r13d,%r13d
+	xorl	%r8d,%r14d
+	andl	%eax,%r12d
+	xorl	%eax,%r13d
+	addl	48(%rsp),%edx
+	movl	%r8d,%r15d
+	xorl	%ecx,%r12d
+	shrdl	$11,%r14d,%r14d
+	xorl	%r9d,%r15d
+	addl	%r12d,%edx
+	shrdl	$6,%r13d,%r13d
+	andl	%r15d,%edi
+	xorl	%r8d,%r14d
+	addl	%r13d,%edx
+	xorl	%r9d,%edi
+	shrdl	$2,%r14d,%r14d
+	addl	%edx,%r11d
+	addl	%edi,%edx
+	movl	%r11d,%r13d
+	addl	%edx,%r14d
+	shrdl	$14,%r13d,%r13d
+	movl	%r14d,%edx
+	movl	%eax,%r12d
+	shrdl	$9,%r14d,%r14d
+	xorl	%r11d,%r13d
+	xorl	%ebx,%r12d
+	shrdl	$5,%r13d,%r13d
+	xorl	%edx,%r14d
+	andl	%r11d,%r12d
+	xorl	%r11d,%r13d
+	addl	52(%rsp),%ecx
+	movl	%edx,%edi
+	xorl	%ebx,%r12d
+	shrdl	$11,%r14d,%r14d
+	xorl	%r8d,%edi
+	addl	%r12d,%ecx
+	shrdl	$6,%r13d,%r13d
+	andl	%edi,%r15d
+	xorl	%edx,%r14d
+	addl	%r13d,%ecx
+	xorl	%r8d,%r15d
+	shrdl	$2,%r14d,%r14d
+	addl	%ecx,%r10d
+	addl	%r15d,%ecx
+	movl	%r10d,%r13d
+	addl	%ecx,%r14d
+	shrdl	$14,%r13d,%r13d
+	movl	%r14d,%ecx
+	movl	%r11d,%r12d
+	shrdl	$9,%r14d,%r14d
+	xorl	%r10d,%r13d
+	xorl	%eax,%r12d
+	shrdl	$5,%r13d,%r13d
+	xorl	%ecx,%r14d
+	andl	%r10d,%r12d
+	xorl	%r10d,%r13d
+	addl	56(%rsp),%ebx
+	movl	%ecx,%r15d
+	xorl	%eax,%r12d
+	shrdl	$11,%r14d,%r14d
+	xorl	%edx,%r15d
+	addl	%r12d,%ebx
+	shrdl	$6,%r13d,%r13d
+	andl	%r15d,%edi
+	xorl	%ecx,%r14d
+	addl	%r13d,%ebx
+	xorl	%edx,%edi
+	shrdl	$2,%r14d,%r14d
+	addl	%ebx,%r9d
+	addl	%edi,%ebx
+	movl	%r9d,%r13d
+	addl	%ebx,%r14d
+	shrdl	$14,%r13d,%r13d
+	movl	%r14d,%ebx
+	movl	%r10d,%r12d
+	shrdl	$9,%r14d,%r14d
+	xorl	%r9d,%r13d
+	xorl	%r11d,%r12d
+	shrdl	$5,%r13d,%r13d
+	xorl	%ebx,%r14d
+	andl	%r9d,%r12d
+	xorl	%r9d,%r13d
+	addl	60(%rsp),%eax
+	movl	%ebx,%edi
+	xorl	%r11d,%r12d
+	shrdl	$11,%r14d,%r14d
+	xorl	%ecx,%edi
+	addl	%r12d,%eax
+	shrdl	$6,%r13d,%r13d
+	andl	%edi,%r15d
+	xorl	%ebx,%r14d
+	addl	%r13d,%eax
+	xorl	%ecx,%r15d
+	shrdl	$2,%r14d,%r14d
+	addl	%eax,%r8d
+	addl	%r15d,%eax
+	movl	%r8d,%r13d
+	addl	%eax,%r14d
+	movq	64+0(%rsp),%rdi
+	movl	%r14d,%eax
+
+	addl	0(%rdi),%eax
+	leaq	64(%rsi),%rsi
+	addl	4(%rdi),%ebx
+	addl	8(%rdi),%ecx
+	addl	12(%rdi),%edx
+	addl	16(%rdi),%r8d
+	addl	20(%rdi),%r9d
+	addl	24(%rdi),%r10d
+	addl	28(%rdi),%r11d
+
+	cmpq	64+16(%rsp),%rsi
+
+	movl	%eax,0(%rdi)
+	movl	%ebx,4(%rdi)
+	movl	%ecx,8(%rdi)
+	movl	%edx,12(%rdi)
+	movl	%r8d,16(%rdi)
+	movl	%r9d,20(%rdi)
+	movl	%r10d,24(%rdi)
+	movl	%r11d,28(%rdi)
+	jb	.Lloop_avx
+
+	movq	64+24(%rsp),%rsi
+	vzeroupper
+	movq	(%rsi),%r15
+	movq	8(%rsi),%r14
+	movq	16(%rsi),%r13
+	movq	24(%rsi),%r12
+	movq	32(%rsi),%rbp
+	movq	40(%rsi),%rbx
+	leaq	48(%rsi),%rsp
+.Lepilogue_avx:
+	.byte	0xf3,0xc3
+.size	sha256_block_data_order_avx,.-sha256_block_data_order_avx
 #endif

third_party/boringssl/linux-x86_64/crypto/sha/sha512-x86_64.S

diff --git a/third_party/boringssl/linux-x86_64/crypto/sha/sha512-x86_64.S b/third_party/boringssl/linux-x86_64/crypto/sha/sha512-x86_64.S
index ca3a3a1..d65743f 100644
--- a/third_party/boringssl/linux-x86_64/crypto/sha/sha512-x86_64.S
+++ b/third_party/boringssl/linux-x86_64/crypto/sha/sha512-x86_64.S
@@ -8,6 +8,17 @@
 .type	sha512_block_data_order,@function
 .align	16
 sha512_block_data_order:
+	leaq	OPENSSL_ia32cap_P(%rip),%r11
+	movl	0(%r11),%r9d
+	movl	4(%r11),%r10d
+	movl	8(%r11),%r11d
+	testl	$2048,%r10d
+	jnz	.Lxop_shortcut
+	andl	$1073741824,%r9d
+	andl	$268435968,%r10d
+	orl	%r9d,%r10d
+	cmpl	$1342177792,%r10d
+	je	.Lavx_shortcut
 	pushq	%rbx
 	pushq	%rbp
 	pushq	%r12
@@ -1784,4 +1795,2234 @@
 .quad	0x0001020304050607,0x08090a0b0c0d0e0f
 .quad	0x0001020304050607,0x08090a0b0c0d0e0f
 .byte	83,72,65,53,49,50,32,98,108,111,99,107,32,116,114,97,110,115,102,111,114,109,32,102,111,114,32,120,56,54,95,54,52,44,32,67,82,89,80,84,79,71,65,77,83,32,98,121,32,60,97,112,112,114,111,64,111,112,101,110,115,115,108,46,111,114,103,62,0
+.type	sha512_block_data_order_xop,@function
+.align	64
+sha512_block_data_order_xop:
+.Lxop_shortcut:
+	pushq	%rbx
+	pushq	%rbp
+	pushq	%r12
+	pushq	%r13
+	pushq	%r14
+	pushq	%r15
+	movq	%rsp,%r11
+	shlq	$4,%rdx
+	subq	$160,%rsp
+	leaq	(%rsi,%rdx,8),%rdx
+	andq	$-64,%rsp
+	movq	%rdi,128+0(%rsp)
+	movq	%rsi,128+8(%rsp)
+	movq	%rdx,128+16(%rsp)
+	movq	%r11,128+24(%rsp)
+.Lprologue_xop:
+
+	vzeroupper
+	movq	0(%rdi),%rax
+	movq	8(%rdi),%rbx
+	movq	16(%rdi),%rcx
+	movq	24(%rdi),%rdx
+	movq	32(%rdi),%r8
+	movq	40(%rdi),%r9
+	movq	48(%rdi),%r10
+	movq	56(%rdi),%r11
+	jmp	.Lloop_xop
+.align	16
+.Lloop_xop:
+	vmovdqa	K512+1280(%rip),%xmm11
+	vmovdqu	0(%rsi),%xmm0
+	leaq	K512+128(%rip),%rbp
+	vmovdqu	16(%rsi),%xmm1
+	vmovdqu	32(%rsi),%xmm2
+	vpshufb	%xmm11,%xmm0,%xmm0
+	vmovdqu	48(%rsi),%xmm3
+	vpshufb	%xmm11,%xmm1,%xmm1
+	vmovdqu	64(%rsi),%xmm4
+	vpshufb	%xmm11,%xmm2,%xmm2
+	vmovdqu	80(%rsi),%xmm5
+	vpshufb	%xmm11,%xmm3,%xmm3
+	vmovdqu	96(%rsi),%xmm6
+	vpshufb	%xmm11,%xmm4,%xmm4
+	vmovdqu	112(%rsi),%xmm7
+	vpshufb	%xmm11,%xmm5,%xmm5
+	vpaddq	-128(%rbp),%xmm0,%xmm8
+	vpshufb	%xmm11,%xmm6,%xmm6
+	vpaddq	-96(%rbp),%xmm1,%xmm9
+	vpshufb	%xmm11,%xmm7,%xmm7
+	vpaddq	-64(%rbp),%xmm2,%xmm10
+	vpaddq	-32(%rbp),%xmm3,%xmm11
+	vmovdqa	%xmm8,0(%rsp)
+	vpaddq	0(%rbp),%xmm4,%xmm8
+	vmovdqa	%xmm9,16(%rsp)
+	vpaddq	32(%rbp),%xmm5,%xmm9
+	vmovdqa	%xmm10,32(%rsp)
+	vpaddq	64(%rbp),%xmm6,%xmm10
+	vmovdqa	%xmm11,48(%rsp)
+	vpaddq	96(%rbp),%xmm7,%xmm11
+	vmovdqa	%xmm8,64(%rsp)
+	movq	%rax,%r14
+	vmovdqa	%xmm9,80(%rsp)
+	movq	%rbx,%rdi
+	vmovdqa	%xmm10,96(%rsp)
+	xorq	%rcx,%rdi
+	vmovdqa	%xmm11,112(%rsp)
+	movq	%r8,%r13
+	jmp	.Lxop_00_47
+
+.align	16
+.Lxop_00_47:
+	addq	$256,%rbp
+	vpalignr	$8,%xmm0,%xmm1,%xmm8
+	rorq	$23,%r13
+	movq	%r14,%rax
+	vpalignr	$8,%xmm4,%xmm5,%xmm11
+	movq	%r9,%r12
+	rorq	$5,%r14
+.byte	143,72,120,195,200,56
+	xorq	%r8,%r13
+	xorq	%r10,%r12
+	vpsrlq	$7,%xmm8,%xmm8
+	rorq	$4,%r13
+	xorq	%rax,%r14
+	vpaddq	%xmm11,%xmm0,%xmm0
+	andq	%r8,%r12
+	xorq	%r8,%r13
+	addq	0(%rsp),%r11
+	movq	%rax,%r15
+.byte	143,72,120,195,209,7
+	xorq	%r10,%r12
+	rorq	$6,%r14
+	vpxor	%xmm9,%xmm8,%xmm8
+	xorq	%rbx,%r15
+	addq	%r12,%r11
+	rorq	$14,%r13
+	andq	%r15,%rdi
+.byte	143,104,120,195,223,3
+	xorq	%rax,%r14
+	addq	%r13,%r11
+	vpxor	%xmm10,%xmm8,%xmm8
+	xorq	%rbx,%rdi
+	rorq	$28,%r14
+	vpsrlq	$6,%xmm7,%xmm10
+	addq	%r11,%rdx
+	addq	%rdi,%r11
+	vpaddq	%xmm8,%xmm0,%xmm0
+	movq	%rdx,%r13
+	addq	%r11,%r14
+.byte	143,72,120,195,203,42
+	rorq	$23,%r13
+	movq	%r14,%r11
+	vpxor	%xmm10,%xmm11,%xmm11
+	movq	%r8,%r12
+	rorq	$5,%r14
+	xorq	%rdx,%r13
+	xorq	%r9,%r12
+	vpxor	%xmm9,%xmm11,%xmm11
+	rorq	$4,%r13
+	xorq	%r11,%r14
+	andq	%rdx,%r12
+	xorq	%rdx,%r13
+	vpaddq	%xmm11,%xmm0,%xmm0
+	addq	8(%rsp),%r10
+	movq	%r11,%rdi
+	xorq	%r9,%r12
+	rorq	$6,%r14
+	vpaddq	-128(%rbp),%xmm0,%xmm10
+	xorq	%rax,%rdi
+	addq	%r12,%r10
+	rorq	$14,%r13
+	andq	%rdi,%r15
+	xorq	%r11,%r14
+	addq	%r13,%r10
+	xorq	%rax,%r15
+	rorq	$28,%r14
+	addq	%r10,%rcx
+	addq	%r15,%r10
+	movq	%rcx,%r13
+	addq	%r10,%r14
+	vmovdqa	%xmm10,0(%rsp)
+	vpalignr	$8,%xmm1,%xmm2,%xmm8
+	rorq	$23,%r13
+	movq	%r14,%r10
+	vpalignr	$8,%xmm5,%xmm6,%xmm11
+	movq	%rdx,%r12
+	rorq	$5,%r14
+.byte	143,72,120,195,200,56
+	xorq	%rcx,%r13
+	xorq	%r8,%r12
+	vpsrlq	$7,%xmm8,%xmm8
+	rorq	$4,%r13
+	xorq	%r10,%r14
+	vpaddq	%xmm11,%xmm1,%xmm1
+	andq	%rcx,%r12
+	xorq	%rcx,%r13
+	addq	16(%rsp),%r9
+	movq	%r10,%r15
+.byte	143,72,120,195,209,7
+	xorq	%r8,%r12
+	rorq	$6,%r14
+	vpxor	%xmm9,%xmm8,%xmm8
+	xorq	%r11,%r15
+	addq	%r12,%r9
+	rorq	$14,%r13
+	andq	%r15,%rdi
+.byte	143,104,120,195,216,3
+	xorq	%r10,%r14
+	addq	%r13,%r9
+	vpxor	%xmm10,%xmm8,%xmm8
+	xorq	%r11,%rdi
+	rorq	$28,%r14
+	vpsrlq	$6,%xmm0,%xmm10
+	addq	%r9,%rbx
+	addq	%rdi,%r9
+	vpaddq	%xmm8,%xmm1,%xmm1
+	movq	%rbx,%r13
+	addq	%r9,%r14
+.byte	143,72,120,195,203,42
+	rorq	$23,%r13
+	movq	%r14,%r9
+	vpxor	%xmm10,%xmm11,%xmm11
+	movq	%rcx,%r12
+	rorq	$5,%r14
+	xorq	%rbx,%r13
+	xorq	%rdx,%r12
+	vpxor	%xmm9,%xmm11,%xmm11
+	rorq	$4,%r13
+	xorq	%r9,%r14
+	andq	%rbx,%r12
+	xorq	%rbx,%r13
+	vpaddq	%xmm11,%xmm1,%xmm1
+	addq	24(%rsp),%r8
+	movq	%r9,%rdi
+	xorq	%rdx,%r12
+	rorq	$6,%r14
+	vpaddq	-96(%rbp),%xmm1,%xmm10
+	xorq	%r10,%rdi
+	addq	%r12,%r8
+	rorq	$14,%r13
+	andq	%rdi,%r15
+	xorq	%r9,%r14
+	addq	%r13,%r8
+	xorq	%r10,%r15
+	rorq	$28,%r14
+	addq	%r8,%rax
+	addq	%r15,%r8
+	movq	%rax,%r13
+	addq	%r8,%r14
+	vmovdqa	%xmm10,16(%rsp)
+	vpalignr	$8,%xmm2,%xmm3,%xmm8
+	rorq	$23,%r13
+	movq	%r14,%r8
+	vpalignr	$8,%xmm6,%xmm7,%xmm11
+	movq	%rbx,%r12
+	rorq	$5,%r14
+.byte	143,72,120,195,200,56
+	xorq	%rax,%r13
+	xorq	%rcx,%r12
+	vpsrlq	$7,%xmm8,%xmm8
+	rorq	$4,%r13
+	xorq	%r8,%r14
+	vpaddq	%xmm11,%xmm2,%xmm2
+	andq	%rax,%r12
+	xorq	%rax,%r13
+	addq	32(%rsp),%rdx
+	movq	%r8,%r15
+.byte	143,72,120,195,209,7
+	xorq	%rcx,%r12
+	rorq	$6,%r14
+	vpxor	%xmm9,%xmm8,%xmm8
+	xorq	%r9,%r15
+	addq	%r12,%rdx
+	rorq	$14,%r13
+	andq	%r15,%rdi
+.byte	143,104,120,195,217,3
+	xorq	%r8,%r14
+	addq	%r13,%rdx
+	vpxor	%xmm10,%xmm8,%xmm8
+	xorq	%r9,%rdi
+	rorq	$28,%r14
+	vpsrlq	$6,%xmm1,%xmm10
+	addq	%rdx,%r11
+	addq	%rdi,%rdx
+	vpaddq	%xmm8,%xmm2,%xmm2
+	movq	%r11,%r13
+	addq	%rdx,%r14
+.byte	143,72,120,195,203,42
+	rorq	$23,%r13
+	movq	%r14,%rdx
+	vpxor	%xmm10,%xmm11,%xmm11
+	movq	%rax,%r12
+	rorq	$5,%r14
+	xorq	%r11,%r13
+	xorq	%rbx,%r12
+	vpxor	%xmm9,%xmm11,%xmm11
+	rorq	$4,%r13
+	xorq	%rdx,%r14
+	andq	%r11,%r12
+	xorq	%r11,%r13
+	vpaddq	%xmm11,%xmm2,%xmm2
+	addq	40(%rsp),%rcx
+	movq	%rdx,%rdi
+	xorq	%rbx,%r12
+	rorq	$6,%r14
+	vpaddq	-64(%rbp),%xmm2,%xmm10
+	xorq	%r8,%rdi
+	addq	%r12,%rcx
+	rorq	$14,%r13
+	andq	%rdi,%r15
+	xorq	%rdx,%r14
+	addq	%r13,%rcx
+	xorq	%r8,%r15
+	rorq	$28,%r14
+	addq	%rcx,%r10
+	addq	%r15,%rcx
+	movq	%r10,%r13
+	addq	%rcx,%r14
+	vmovdqa	%xmm10,32(%rsp)
+	vpalignr	$8,%xmm3,%xmm4,%xmm8
+	rorq	$23,%r13
+	movq	%r14,%rcx
+	vpalignr	$8,%xmm7,%xmm0,%xmm11
+	movq	%r11,%r12
+	rorq	$5,%r14
+.byte	143,72,120,195,200,56
+	xorq	%r10,%r13
+	xorq	%rax,%r12
+	vpsrlq	$7,%xmm8,%xmm8
+	rorq	$4,%r13
+	xorq	%rcx,%r14
+	vpaddq	%xmm11,%xmm3,%xmm3
+	andq	%r10,%r12
+	xorq	%r10,%r13
+	addq	48(%rsp),%rbx
+	movq	%rcx,%r15
+.byte	143,72,120,195,209,7
+	xorq	%rax,%r12
+	rorq	$6,%r14
+	vpxor	%xmm9,%xmm8,%xmm8
+	xorq	%rdx,%r15
+	addq	%r12,%rbx
+	rorq	$14,%r13
+	andq	%r15,%rdi
+.byte	143,104,120,195,218,3
+	xorq	%rcx,%r14
+	addq	%r13,%rbx
+	vpxor	%xmm10,%xmm8,%xmm8
+	xorq	%rdx,%rdi
+	rorq	$28,%r14
+	vpsrlq	$6,%xmm2,%xmm10
+	addq	%rbx,%r9
+	addq	%rdi,%rbx
+	vpaddq	%xmm8,%xmm3,%xmm3
+	movq	%r9,%r13
+	addq	%rbx,%r14
+.byte	143,72,120,195,203,42
+	rorq	$23,%r13
+	movq	%r14,%rbx
+	vpxor	%xmm10,%xmm11,%xmm11
+	movq	%r10,%r12
+	rorq	$5,%r14
+	xorq	%r9,%r13
+	xorq	%r11,%r12
+	vpxor	%xmm9,%xmm11,%xmm11
+	rorq	$4,%r13
+	xorq	%rbx,%r14
+	andq	%r9,%r12
+	xorq	%r9,%r13
+	vpaddq	%xmm11,%xmm3,%xmm3
+	addq	56(%rsp),%rax
+	movq	%rbx,%rdi
+	xorq	%r11,%r12
+	rorq	$6,%r14
+	vpaddq	-32(%rbp),%xmm3,%xmm10
+	xorq	%rcx,%rdi
+	addq	%r12,%rax
+	rorq	$14,%r13
+	andq	%rdi,%r15
+	xorq	%rbx,%r14
+	addq	%r13,%rax
+	xorq	%rcx,%r15
+	rorq	$28,%r14
+	addq	%rax,%r8
+	addq	%r15,%rax
+	movq	%r8,%r13
+	addq	%rax,%r14
+	vmovdqa	%xmm10,48(%rsp)
+	vpalignr	$8,%xmm4,%xmm5,%xmm8
+	rorq	$23,%r13
+	movq	%r14,%rax
+	vpalignr	$8,%xmm0,%xmm1,%xmm11
+	movq	%r9,%r12
+	rorq	$5,%r14
+.byte	143,72,120,195,200,56
+	xorq	%r8,%r13
+	xorq	%r10,%r12
+	vpsrlq	$7,%xmm8,%xmm8
+	rorq	$4,%r13
+	xorq	%rax,%r14
+	vpaddq	%xmm11,%xmm4,%xmm4
+	andq	%r8,%r12
+	xorq	%r8,%r13
+	addq	64(%rsp),%r11
+	movq	%rax,%r15
+.byte	143,72,120,195,209,7
+	xorq	%r10,%r12
+	rorq	$6,%r14
+	vpxor	%xmm9,%xmm8,%xmm8
+	xorq	%rbx,%r15
+	addq	%r12,%r11
+	rorq	$14,%r13
+	andq	%r15,%rdi
+.byte	143,104,120,195,219,3
+	xorq	%rax,%r14
+	addq	%r13,%r11
+	vpxor	%xmm10,%xmm8,%xmm8
+	xorq	%rbx,%rdi
+	rorq	$28,%r14
+	vpsrlq	$6,%xmm3,%xmm10
+	addq	%r11,%rdx
+	addq	%rdi,%r11
+	vpaddq	%xmm8,%xmm4,%xmm4
+	movq	%rdx,%r13
+	addq	%r11,%r14
+.byte	143,72,120,195,203,42
+	rorq	$23,%r13
+	movq	%r14,%r11
+	vpxor	%xmm10,%xmm11,%xmm11
+	movq	%r8,%r12
+	rorq	$5,%r14
+	xorq	%rdx,%r13
+	xorq	%r9,%r12
+	vpxor	%xmm9,%xmm11,%xmm11
+	rorq	$4,%r13
+	xorq	%r11,%r14
+	andq	%rdx,%r12
+	xorq	%rdx,%r13
+	vpaddq	%xmm11,%xmm4,%xmm4
+	addq	72(%rsp),%r10
+	movq	%r11,%rdi
+	xorq	%r9,%r12
+	rorq	$6,%r14
+	vpaddq	0(%rbp),%xmm4,%xmm10
+	xorq	%rax,%rdi
+	addq	%r12,%r10
+	rorq	$14,%r13
+	andq	%rdi,%r15
+	xorq	%r11,%r14
+	addq	%r13,%r10
+	xorq	%rax,%r15
+	rorq	$28,%r14
+	addq	%r10,%rcx
+	addq	%r15,%r10
+	movq	%rcx,%r13
+	addq	%r10,%r14
+	vmovdqa	%xmm10,64(%rsp)
+	vpalignr	$8,%xmm5,%xmm6,%xmm8
+	rorq	$23,%r13
+	movq	%r14,%r10
+	vpalignr	$8,%xmm1,%xmm2,%xmm11
+	movq	%rdx,%r12
+	rorq	$5,%r14
+.byte	143,72,120,195,200,56
+	xorq	%rcx,%r13
+	xorq	%r8,%r12
+	vpsrlq	$7,%xmm8,%xmm8
+	rorq	$4,%r13
+	xorq	%r10,%r14
+	vpaddq	%xmm11,%xmm5,%xmm5
+	andq	%rcx,%r12
+	xorq	%rcx,%r13
+	addq	80(%rsp),%r9
+	movq	%r10,%r15
+.byte	143,72,120,195,209,7
+	xorq	%r8,%r12
+	rorq	$6,%r14
+	vpxor	%xmm9,%xmm8,%xmm8
+	xorq	%r11,%r15
+	addq	%r12,%r9
+	rorq	$14,%r13
+	andq	%r15,%rdi
+.byte	143,104,120,195,220,3
+	xorq	%r10,%r14
+	addq	%r13,%r9
+	vpxor	%xmm10,%xmm8,%xmm8
+	xorq	%r11,%rdi
+	rorq	$28,%r14
+	vpsrlq	$6,%xmm4,%xmm10
+	addq	%r9,%rbx
+	addq	%rdi,%r9
+	vpaddq	%xmm8,%xmm5,%xmm5
+	movq	%rbx,%r13
+	addq	%r9,%r14
+.byte	143,72,120,195,203,42
+	rorq	$23,%r13
+	movq	%r14,%r9
+	vpxor	%xmm10,%xmm11,%xmm11
+	movq	%rcx,%r12
+	rorq	$5,%r14
+	xorq	%rbx,%r13
+	xorq	%rdx,%r12
+	vpxor	%xmm9,%xmm11,%xmm11
+	rorq	$4,%r13
+	xorq	%r9,%r14
+	andq	%rbx,%r12
+	xorq	%rbx,%r13
+	vpaddq	%xmm11,%xmm5,%xmm5
+	addq	88(%rsp),%r8
+	movq	%r9,%rdi
+	xorq	%rdx,%r12
+	rorq	$6,%r14
+	vpaddq	32(%rbp),%xmm5,%xmm10
+	xorq	%r10,%rdi
+	addq	%r12,%r8
+	rorq	$14,%r13
+	andq	%rdi,%r15
+	xorq	%r9,%r14
+	addq	%r13,%r8
+	xorq	%r10,%r15
+	rorq	$28,%r14
+	addq	%r8,%rax
+	addq	%r15,%r8
+	movq	%rax,%r13
+	addq	%r8,%r14
+	vmovdqa	%xmm10,80(%rsp)
+	vpalignr	$8,%xmm6,%xmm7,%xmm8
+	rorq	$23,%r13
+	movq	%r14,%r8
+	vpalignr	$8,%xmm2,%xmm3,%xmm11
+	movq	%rbx,%r12
+	rorq	$5,%r14
+.byte	143,72,120,195,200,56
+	xorq	%rax,%r13
+	xorq	%rcx,%r12
+	vpsrlq	$7,%xmm8,%xmm8
+	rorq	$4,%r13
+	xorq	%r8,%r14
+	vpaddq	%xmm11,%xmm6,%xmm6
+	andq	%rax,%r12
+	xorq	%rax,%r13
+	addq	96(%rsp),%rdx
+	movq	%r8,%r15
+.byte	143,72,120,195,209,7
+	xorq	%rcx,%r12
+	rorq	$6,%r14
+	vpxor	%xmm9,%xmm8,%xmm8
+	xorq	%r9,%r15
+	addq	%r12,%rdx
+	rorq	$14,%r13
+	andq	%r15,%rdi
+.byte	143,104,120,195,221,3
+	xorq	%r8,%r14
+	addq	%r13,%rdx
+	vpxor	%xmm10,%xmm8,%xmm8
+	xorq	%r9,%rdi
+	rorq	$28,%r14
+	vpsrlq	$6,%xmm5,%xmm10
+	addq	%rdx,%r11
+	addq	%rdi,%rdx
+	vpaddq	%xmm8,%xmm6,%xmm6
+	movq	%r11,%r13
+	addq	%rdx,%r14
+.byte	143,72,120,195,203,42
+	rorq	$23,%r13
+	movq	%r14,%rdx
+	vpxor	%xmm10,%xmm11,%xmm11
+	movq	%rax,%r12
+	rorq	$5,%r14
+	xorq	%r11,%r13
+	xorq	%rbx,%r12
+	vpxor	%xmm9,%xmm11,%xmm11
+	rorq	$4,%r13
+	xorq	%rdx,%r14
+	andq	%r11,%r12
+	xorq	%r11,%r13
+	vpaddq	%xmm11,%xmm6,%xmm6
+	addq	104(%rsp),%rcx
+	movq	%rdx,%rdi
+	xorq	%rbx,%r12
+	rorq	$6,%r14
+	vpaddq	64(%rbp),%xmm6,%xmm10
+	xorq	%r8,%rdi
+	addq	%r12,%rcx
+	rorq	$14,%r13
+	andq	%rdi,%r15
+	xorq	%rdx,%r14
+	addq	%r13,%rcx
+	xorq	%r8,%r15
+	rorq	$28,%r14
+	addq	%rcx,%r10
+	addq	%r15,%rcx
+	movq	%r10,%r13
+	addq	%rcx,%r14
+	vmovdqa	%xmm10,96(%rsp)
+	vpalignr	$8,%xmm7,%xmm0,%xmm8
+	rorq	$23,%r13
+	movq	%r14,%rcx
+	vpalignr	$8,%xmm3,%xmm4,%xmm11
+	movq	%r11,%r12
+	rorq	$5,%r14
+.byte	143,72,120,195,200,56
+	xorq	%r10,%r13
+	xorq	%rax,%r12
+	vpsrlq	$7,%xmm8,%xmm8
+	rorq	$4,%r13
+	xorq	%rcx,%r14
+	vpaddq	%xmm11,%xmm7,%xmm7
+	andq	%r10,%r12
+	xorq	%r10,%r13
+	addq	112(%rsp),%rbx
+	movq	%rcx,%r15
+.byte	143,72,120,195,209,7
+	xorq	%rax,%r12
+	rorq	$6,%r14
+	vpxor	%xmm9,%xmm8,%xmm8
+	xorq	%rdx,%r15
+	addq	%r12,%rbx
+	rorq	$14,%r13
+	andq	%r15,%rdi
+.byte	143,104,120,195,222,3
+	xorq	%rcx,%r14
+	addq	%r13,%rbx
+	vpxor	%xmm10,%xmm8,%xmm8
+	xorq	%rdx,%rdi
+	rorq	$28,%r14
+	vpsrlq	$6,%xmm6,%xmm10
+	addq	%rbx,%r9
+	addq	%rdi,%rbx
+	vpaddq	%xmm8,%xmm7,%xmm7
+	movq	%r9,%r13
+	addq	%rbx,%r14
+.byte	143,72,120,195,203,42
+	rorq	$23,%r13
+	movq	%r14,%rbx
+	vpxor	%xmm10,%xmm11,%xmm11
+	movq	%r10,%r12
+	rorq	$5,%r14
+	xorq	%r9,%r13
+	xorq	%r11,%r12
+	vpxor	%xmm9,%xmm11,%xmm11
+	rorq	$4,%r13
+	xorq	%rbx,%r14
+	andq	%r9,%r12
+	xorq	%r9,%r13
+	vpaddq	%xmm11,%xmm7,%xmm7
+	addq	120(%rsp),%rax
+	movq	%rbx,%rdi
+	xorq	%r11,%r12
+	rorq	$6,%r14
+	vpaddq	96(%rbp),%xmm7,%xmm10
+	xorq	%rcx,%rdi
+	addq	%r12,%rax
+	rorq	$14,%r13
+	andq	%rdi,%r15
+	xorq	%rbx,%r14
+	addq	%r13,%rax
+	xorq	%rcx,%r15
+	rorq	$28,%r14
+	addq	%rax,%r8
+	addq	%r15,%rax
+	movq	%r8,%r13
+	addq	%rax,%r14
+	vmovdqa	%xmm10,112(%rsp)
+	cmpb	$0,135(%rbp)
+	jne	.Lxop_00_47
+	rorq	$23,%r13
+	movq	%r14,%rax
+	movq	%r9,%r12
+	rorq	$5,%r14
+	xorq	%r8,%r13
+	xorq	%r10,%r12
+	rorq	$4,%r13
+	xorq	%rax,%r14
+	andq	%r8,%r12
+	xorq	%r8,%r13
+	addq	0(%rsp),%r11
+	movq	%rax,%r15
+	xorq	%r10,%r12
+	rorq	$6,%r14
+	xorq	%rbx,%r15
+	addq	%r12,%r11
+	rorq	$14,%r13
+	andq	%r15,%rdi
+	xorq	%rax,%r14
+	addq	%r13,%r11
+	xorq	%rbx,%rdi
+	rorq	$28,%r14
+	addq	%r11,%rdx
+	addq	%rdi,%r11
+	movq	%rdx,%r13
+	addq	%r11,%r14
+	rorq	$23,%r13
+	movq	%r14,%r11
+	movq	%r8,%r12
+	rorq	$5,%r14
+	xorq	%rdx,%r13
+	xorq	%r9,%r12
+	rorq	$4,%r13
+	xorq	%r11,%r14
+	andq	%rdx,%r12
+	xorq	%rdx,%r13
+	addq	8(%rsp),%r10
+	movq	%r11,%rdi
+	xorq	%r9,%r12
+	rorq	$6,%r14
+	xorq	%rax,%rdi
+	addq	%r12,%r10
+	rorq	$14,%r13
+	andq	%rdi,%r15
+	xorq	%r11,%r14
+	addq	%r13,%r10
+	xorq	%rax,%r15
+	rorq	$28,%r14
+	addq	%r10,%rcx
+	addq	%r15,%r10
+	movq	%rcx,%r13
+	addq	%r10,%r14
+	rorq	$23,%r13
+	movq	%r14,%r10
+	movq	%rdx,%r12
+	rorq	$5,%r14
+	xorq	%rcx,%r13
+	xorq	%r8,%r12
+	rorq	$4,%r13
+	xorq	%r10,%r14
+	andq	%rcx,%r12
+	xorq	%rcx,%r13
+	addq	16(%rsp),%r9
+	movq	%r10,%r15
+	xorq	%r8,%r12
+	rorq	$6,%r14
+	xorq	%r11,%r15
+	addq	%r12,%r9
+	rorq	$14,%r13
+	andq	%r15,%rdi
+	xorq	%r10,%r14
+	addq	%r13,%r9
+	xorq	%r11,%rdi
+	rorq	$28,%r14
+	addq	%r9,%rbx
+	addq	%rdi,%r9
+	movq	%rbx,%r13
+	addq	%r9,%r14
+	rorq	$23,%r13
+	movq	%r14,%r9
+	movq	%rcx,%r12
+	rorq	$5,%r14
+	xorq	%rbx,%r13
+	xorq	%rdx,%r12
+	rorq	$4,%r13
+	xorq	%r9,%r14
+	andq	%rbx,%r12
+	xorq	%rbx,%r13
+	addq	24(%rsp),%r8
+	movq	%r9,%rdi
+	xorq	%rdx,%r12
+	rorq	$6,%r14
+	xorq	%r10,%rdi
+	addq	%r12,%r8
+	rorq	$14,%r13
+	andq	%rdi,%r15
+	xorq	%r9,%r14
+	addq	%r13,%r8
+	xorq	%r10,%r15
+	rorq	$28,%r14
+	addq	%r8,%rax
+	addq	%r15,%r8
+	movq	%rax,%r13
+	addq	%r8,%r14
+	rorq	$23,%r13
+	movq	%r14,%r8
+	movq	%rbx,%r12
+	rorq	$5,%r14
+	xorq	%rax,%r13
+	xorq	%rcx,%r12
+	rorq	$4,%r13
+	xorq	%r8,%r14
+	andq	%rax,%r12
+	xorq	%rax,%r13
+	addq	32(%rsp),%rdx
+	movq	%r8,%r15
+	xorq	%rcx,%r12
+	rorq	$6,%r14
+	xorq	%r9,%r15
+	addq	%r12,%rdx
+	rorq	$14,%r13
+	andq	%r15,%rdi
+	xorq	%r8,%r14
+	addq	%r13,%rdx
+	xorq	%r9,%rdi
+	rorq	$28,%r14
+	addq	%rdx,%r11
+	addq	%rdi,%rdx
+	movq	%r11,%r13
+	addq	%rdx,%r14
+	rorq	$23,%r13
+	movq	%r14,%rdx
+	movq	%rax,%r12
+	rorq	$5,%r14
+	xorq	%r11,%r13
+	xorq	%rbx,%r12
+	rorq	$4,%r13
+	xorq	%rdx,%r14
+	andq	%r11,%r12
+	xorq	%r11,%r13
+	addq	40(%rsp),%rcx
+	movq	%rdx,%rdi
+	xorq	%rbx,%r12
+	rorq	$6,%r14
+	xorq	%r8,%rdi
+	addq	%r12,%rcx
+	rorq	$14,%r13
+	andq	%rdi,%r15
+	xorq	%rdx,%r14
+	addq	%r13,%rcx
+	xorq	%r8,%r15
+	rorq	$28,%r14
+	addq	%rcx,%r10
+	addq	%r15,%rcx
+	movq	%r10,%r13
+	addq	%rcx,%r14
+	rorq	$23,%r13
+	movq	%r14,%rcx
+	movq	%r11,%r12
+	rorq	$5,%r14
+	xorq	%r10,%r13
+	xorq	%rax,%r12
+	rorq	$4,%r13
+	xorq	%rcx,%r14
+	andq	%r10,%r12
+	xorq	%r10,%r13
+	addq	48(%rsp),%rbx
+	movq	%rcx,%r15
+	xorq	%rax,%r12
+	rorq	$6,%r14
+	xorq	%rdx,%r15
+	addq	%r12,%rbx
+	rorq	$14,%r13
+	andq	%r15,%rdi
+	xorq	%rcx,%r14
+	addq	%r13,%rbx
+	xorq	%rdx,%rdi
+	rorq	$28,%r14
+	addq	%rbx,%r9
+	addq	%rdi,%rbx
+	movq	%r9,%r13
+	addq	%rbx,%r14
+	rorq	$23,%r13
+	movq	%r14,%rbx
+	movq	%r10,%r12
+	rorq	$5,%r14
+	xorq	%r9,%r13
+	xorq	%r11,%r12
+	rorq	$4,%r13
+	xorq	%rbx,%r14
+	andq	%r9,%r12
+	xorq	%r9,%r13
+	addq	56(%rsp),%rax
+	movq	%rbx,%rdi
+	xorq	%r11,%r12
+	rorq	$6,%r14
+	xorq	%rcx,%rdi
+	addq	%r12,%rax
+	rorq	$14,%r13
+	andq	%rdi,%r15
+	xorq	%rbx,%r14
+	addq	%r13,%rax
+	xorq	%rcx,%r15
+	rorq	$28,%r14
+	addq	%rax,%r8
+	addq	%r15,%rax
+	movq	%r8,%r13
+	addq	%rax,%r14
+	rorq	$23,%r13
+	movq	%r14,%rax
+	movq	%r9,%r12
+	rorq	$5,%r14
+	xorq	%r8,%r13
+	xorq	%r10,%r12
+	rorq	$4,%r13
+	xorq	%rax,%r14
+	andq	%r8,%r12
+	xorq	%r8,%r13
+	addq	64(%rsp),%r11
+	movq	%rax,%r15
+	xorq	%r10,%r12
+	rorq	$6,%r14
+	xorq	%rbx,%r15
+	addq	%r12,%r11
+	rorq	$14,%r13
+	andq	%r15,%rdi
+	xorq	%rax,%r14
+	addq	%r13,%r11
+	xorq	%rbx,%rdi
+	rorq	$28,%r14
+	addq	%r11,%rdx
+	addq	%rdi,%r11
+	movq	%rdx,%r13
+	addq	%r11,%r14
+	rorq	$23,%r13
+	movq	%r14,%r11
+	movq	%r8,%r12
+	rorq	$5,%r14
+	xorq	%rdx,%r13
+	xorq	%r9,%r12
+	rorq	$4,%r13
+	xorq	%r11,%r14
+	andq	%rdx,%r12
+	xorq	%rdx,%r13
+	addq	72(%rsp),%r10
+	movq	%r11,%rdi
+	xorq	%r9,%r12
+	rorq	$6,%r14
+	xorq	%rax,%rdi
+	addq	%r12,%r10
+	rorq	$14,%r13
+	andq	%rdi,%r15
+	xorq	%r11,%r14
+	addq	%r13,%r10
+	xorq	%rax,%r15
+	rorq	$28,%r14
+	addq	%r10,%rcx
+	addq	%r15,%r10
+	movq	%rcx,%r13
+	addq	%r10,%r14
+	rorq	$23,%r13
+	movq	%r14,%r10
+	movq	%rdx,%r12
+	rorq	$5,%r14
+	xorq	%rcx,%r13
+	xorq	%r8,%r12
+	rorq	$4,%r13
+	xorq	%r10,%r14
+	andq	%rcx,%r12
+	xorq	%rcx,%r13
+	addq	80(%rsp),%r9
+	movq	%r10,%r15
+	xorq	%r8,%r12
+	rorq	$6,%r14
+	xorq	%r11,%r15
+	addq	%r12,%r9
+	rorq	$14,%r13
+	andq	%r15,%rdi
+	xorq	%r10,%r14
+	addq	%r13,%r9
+	xorq	%r11,%rdi
+	rorq	$28,%r14
+	addq	%r9,%rbx
+	addq	%rdi,%r9
+	movq	%rbx,%r13
+	addq	%r9,%r14
+	rorq	$23,%r13
+	movq	%r14,%r9
+	movq	%rcx,%r12
+	rorq	$5,%r14
+	xorq	%rbx,%r13
+	xorq	%rdx,%r12
+	rorq	$4,%r13
+	xorq	%r9,%r14
+	andq	%rbx,%r12
+	xorq	%rbx,%r13
+	addq	88(%rsp),%r8
+	movq	%r9,%rdi
+	xorq	%rdx,%r12
+	rorq	$6,%r14
+	xorq	%r10,%rdi
+	addq	%r12,%r8
+	rorq	$14,%r13
+	andq	%rdi,%r15
+	xorq	%r9,%r14
+	addq	%r13,%r8
+	xorq	%r10,%r15
+	rorq	$28,%r14
+	addq	%r8,%rax
+	addq	%r15,%r8
+	movq	%rax,%r13
+	addq	%r8,%r14
+	rorq	$23,%r13
+	movq	%r14,%r8
+	movq	%rbx,%r12
+	rorq	$5,%r14
+	xorq	%rax,%r13
+	xorq	%rcx,%r12
+	rorq	$4,%r13
+	xorq	%r8,%r14
+	andq	%rax,%r12
+	xorq	%rax,%r13
+	addq	96(%rsp),%rdx
+	movq	%r8,%r15
+	xorq	%rcx,%r12
+	rorq	$6,%r14
+	xorq	%r9,%r15
+	addq	%r12,%rdx
+	rorq	$14,%r13
+	andq	%r15,%rdi
+	xorq	%r8,%r14
+	addq	%r13,%rdx
+	xorq	%r9,%rdi
+	rorq	$28,%r14
+	addq	%rdx,%r11
+	addq	%rdi,%rdx
+	movq	%r11,%r13
+	addq	%rdx,%r14
+	rorq	$23,%r13
+	movq	%r14,%rdx
+	movq	%rax,%r12
+	rorq	$5,%r14
+	xorq	%r11,%r13
+	xorq	%rbx,%r12
+	rorq	$4,%r13
+	xorq	%rdx,%r14
+	andq	%r11,%r12
+	xorq	%r11,%r13
+	addq	104(%rsp),%rcx
+	movq	%rdx,%rdi
+	xorq	%rbx,%r12
+	rorq	$6,%r14
+	xorq	%r8,%rdi
+	addq	%r12,%rcx
+	rorq	$14,%r13
+	andq	%rdi,%r15
+	xorq	%rdx,%r14
+	addq	%r13,%rcx
+	xorq	%r8,%r15
+	rorq	$28,%r14
+	addq	%rcx,%r10
+	addq	%r15,%rcx
+	movq	%r10,%r13
+	addq	%rcx,%r14
+	rorq	$23,%r13
+	movq	%r14,%rcx
+	movq	%r11,%r12
+	rorq	$5,%r14
+	xorq	%r10,%r13
+	xorq	%rax,%r12
+	rorq	$4,%r13
+	xorq	%rcx,%r14
+	andq	%r10,%r12
+	xorq	%r10,%r13
+	addq	112(%rsp),%rbx
+	movq	%rcx,%r15
+	xorq	%rax,%r12
+	rorq	$6,%r14
+	xorq	%rdx,%r15
+	addq	%r12,%rbx
+	rorq	$14,%r13
+	andq	%r15,%rdi
+	xorq	%rcx,%r14
+	addq	%r13,%rbx
+	xorq	%rdx,%rdi
+	rorq	$28,%r14
+	addq	%rbx,%r9
+	addq	%rdi,%rbx
+	movq	%r9,%r13
+	addq	%rbx,%r14
+	rorq	$23,%r13
+	movq	%r14,%rbx
+	movq	%r10,%r12
+	rorq	$5,%r14
+	xorq	%r9,%r13
+	xorq	%r11,%r12
+	rorq	$4,%r13
+	xorq	%rbx,%r14
+	andq	%r9,%r12
+	xorq	%r9,%r13
+	addq	120(%rsp),%rax
+	movq	%rbx,%rdi
+	xorq	%r11,%r12
+	rorq	$6,%r14
+	xorq	%rcx,%rdi
+	addq	%r12,%rax
+	rorq	$14,%r13
+	andq	%rdi,%r15
+	xorq	%rbx,%r14
+	addq	%r13,%rax
+	xorq	%rcx,%r15
+	rorq	$28,%r14
+	addq	%rax,%r8
+	addq	%r15,%rax
+	movq	%r8,%r13
+	addq	%rax,%r14
+	movq	128+0(%rsp),%rdi
+	movq	%r14,%rax
+
+	addq	0(%rdi),%rax
+	leaq	128(%rsi),%rsi
+	addq	8(%rdi),%rbx
+	addq	16(%rdi),%rcx
+	addq	24(%rdi),%rdx
+	addq	32(%rdi),%r8
+	addq	40(%rdi),%r9
+	addq	48(%rdi),%r10
+	addq	56(%rdi),%r11
+
+	cmpq	128+16(%rsp),%rsi
+
+	movq	%rax,0(%rdi)
+	movq	%rbx,8(%rdi)
+	movq	%rcx,16(%rdi)
+	movq	%rdx,24(%rdi)
+	movq	%r8,32(%rdi)
+	movq	%r9,40(%rdi)
+	movq	%r10,48(%rdi)
+	movq	%r11,56(%rdi)
+	jb	.Lloop_xop
+
+	movq	128+24(%rsp),%rsi
+	vzeroupper
+	movq	(%rsi),%r15
+	movq	8(%rsi),%r14
+	movq	16(%rsi),%r13
+	movq	24(%rsi),%r12
+	movq	32(%rsi),%rbp
+	movq	40(%rsi),%rbx
+	leaq	48(%rsi),%rsp
+.Lepilogue_xop:
+	.byte	0xf3,0xc3
+.size	sha512_block_data_order_xop,.-sha512_block_data_order_xop
+.type	sha512_block_data_order_avx,@function
+.align	64
+sha512_block_data_order_avx:
+.Lavx_shortcut:
+	pushq	%rbx
+	pushq	%rbp
+	pushq	%r12
+	pushq	%r13
+	pushq	%r14
+	pushq	%r15
+	movq	%rsp,%r11
+	shlq	$4,%rdx
+	subq	$160,%rsp
+	leaq	(%rsi,%rdx,8),%rdx
+	andq	$-64,%rsp
+	movq	%rdi,128+0(%rsp)
+	movq	%rsi,128+8(%rsp)
+	movq	%rdx,128+16(%rsp)
+	movq	%r11,128+24(%rsp)
+.Lprologue_avx:
+
+	vzeroupper
+	movq	0(%rdi),%rax
+	movq	8(%rdi),%rbx
+	movq	16(%rdi),%rcx
+	movq	24(%rdi),%rdx
+	movq	32(%rdi),%r8
+	movq	40(%rdi),%r9
+	movq	48(%rdi),%r10
+	movq	56(%rdi),%r11
+	jmp	.Lloop_avx
+.align	16
+.Lloop_avx:
+	vmovdqa	K512+1280(%rip),%xmm11
+	vmovdqu	0(%rsi),%xmm0
+	leaq	K512+128(%rip),%rbp
+	vmovdqu	16(%rsi),%xmm1
+	vmovdqu	32(%rsi),%xmm2
+	vpshufb	%xmm11,%xmm0,%xmm0
+	vmovdqu	48(%rsi),%xmm3
+	vpshufb	%xmm11,%xmm1,%xmm1
+	vmovdqu	64(%rsi),%xmm4
+	vpshufb	%xmm11,%xmm2,%xmm2
+	vmovdqu	80(%rsi),%xmm5
+	vpshufb	%xmm11,%xmm3,%xmm3
+	vmovdqu	96(%rsi),%xmm6
+	vpshufb	%xmm11,%xmm4,%xmm4
+	vmovdqu	112(%rsi),%xmm7
+	vpshufb	%xmm11,%xmm5,%xmm5
+	vpaddq	-128(%rbp),%xmm0,%xmm8
+	vpshufb	%xmm11,%xmm6,%xmm6
+	vpaddq	-96(%rbp),%xmm1,%xmm9
+	vpshufb	%xmm11,%xmm7,%xmm7
+	vpaddq	-64(%rbp),%xmm2,%xmm10
+	vpaddq	-32(%rbp),%xmm3,%xmm11
+	vmovdqa	%xmm8,0(%rsp)
+	vpaddq	0(%rbp),%xmm4,%xmm8
+	vmovdqa	%xmm9,16(%rsp)
+	vpaddq	32(%rbp),%xmm5,%xmm9
+	vmovdqa	%xmm10,32(%rsp)
+	vpaddq	64(%rbp),%xmm6,%xmm10
+	vmovdqa	%xmm11,48(%rsp)
+	vpaddq	96(%rbp),%xmm7,%xmm11
+	vmovdqa	%xmm8,64(%rsp)
+	movq	%rax,%r14
+	vmovdqa	%xmm9,80(%rsp)
+	movq	%rbx,%rdi
+	vmovdqa	%xmm10,96(%rsp)
+	xorq	%rcx,%rdi
+	vmovdqa	%xmm11,112(%rsp)
+	movq	%r8,%r13
+	jmp	.Lavx_00_47
+
+.align	16
+.Lavx_00_47:
+	addq	$256,%rbp
+	vpalignr	$8,%xmm0,%xmm1,%xmm8
+	shrdq	$23,%r13,%r13
+	movq	%r14,%rax
+	vpalignr	$8,%xmm4,%xmm5,%xmm11
+	movq	%r9,%r12
+	shrdq	$5,%r14,%r14
+	vpsrlq	$1,%xmm8,%xmm10
+	xorq	%r8,%r13
+	xorq	%r10,%r12
+	vpaddq	%xmm11,%xmm0,%xmm0
+	shrdq	$4,%r13,%r13
+	xorq	%rax,%r14
+	vpsrlq	$7,%xmm8,%xmm11
+	andq	%r8,%r12
+	xorq	%r8,%r13
+	vpsllq	$56,%xmm8,%xmm9
+	addq	0(%rsp),%r11
+	movq	%rax,%r15
+	vpxor	%xmm10,%xmm11,%xmm8
+	xorq	%r10,%r12
+	shrdq	$6,%r14,%r14
+	vpsrlq	$7,%xmm10,%xmm10
+	xorq	%rbx,%r15
+	addq	%r12,%r11
+	vpxor	%xmm9,%xmm8,%xmm8
+	shrdq	$14,%r13,%r13
+	andq	%r15,%rdi
+	vpsllq	$7,%xmm9,%xmm9
+	xorq	%rax,%r14
+	addq	%r13,%r11
+	vpxor	%xmm10,%xmm8,%xmm8
+	xorq	%rbx,%rdi
+	shrdq	$28,%r14,%r14
+	vpsrlq	$6,%xmm7,%xmm11
+	addq	%r11,%rdx
+	addq	%rdi,%r11
+	vpxor	%xmm9,%xmm8,%xmm8
+	movq	%rdx,%r13
+	addq	%r11,%r14
+	vpsllq	$3,%xmm7,%xmm10
+	shrdq	$23,%r13,%r13
+	movq	%r14,%r11
+	vpaddq	%xmm8,%xmm0,%xmm0
+	movq	%r8,%r12
+	shrdq	$5,%r14,%r14
+	vpsrlq	$19,%xmm7,%xmm9
+	xorq	%rdx,%r13
+	xorq	%r9,%r12
+	vpxor	%xmm10,%xmm11,%xmm11
+	shrdq	$4,%r13,%r13
+	xorq	%r11,%r14
+	vpsllq	$42,%xmm10,%xmm10
+	andq	%rdx,%r12
+	xorq	%rdx,%r13
+	vpxor	%xmm9,%xmm11,%xmm11
+	addq	8(%rsp),%r10
+	movq	%r11,%rdi
+	vpsrlq	$42,%xmm9,%xmm9
+	xorq	%r9,%r12
+	shrdq	$6,%r14,%r14
+	vpxor	%xmm10,%xmm11,%xmm11
+	xorq	%rax,%rdi
+	addq	%r12,%r10
+	vpxor	%xmm9,%xmm11,%xmm11
+	shrdq	$14,%r13,%r13
+	andq	%rdi,%r15
+	vpaddq	%xmm11,%xmm0,%xmm0
+	xorq	%r11,%r14
+	addq	%r13,%r10
+	vpaddq	-128(%rbp),%xmm0,%xmm10
+	xorq	%rax,%r15
+	shrdq	$28,%r14,%r14
+	addq	%r10,%rcx
+	addq	%r15,%r10
+	movq	%rcx,%r13
+	addq	%r10,%r14
+	vmovdqa	%xmm10,0(%rsp)
+	vpalignr	$8,%xmm1,%xmm2,%xmm8
+	shrdq	$23,%r13,%r13
+	movq	%r14,%r10
+	vpalignr	$8,%xmm5,%xmm6,%xmm11
+	movq	%rdx,%r12
+	shrdq	$5,%r14,%r14
+	vpsrlq	$1,%xmm8,%xmm10
+	xorq	%rcx,%r13
+	xorq	%r8,%r12
+	vpaddq	%xmm11,%xmm1,%xmm1
+	shrdq	$4,%r13,%r13
+	xorq	%r10,%r14
+	vpsrlq	$7,%xmm8,%xmm11
+	andq	%rcx,%r12
+	xorq	%rcx,%r13
+	vpsllq	$56,%xmm8,%xmm9
+	addq	16(%rsp),%r9
+	movq	%r10,%r15
+	vpxor	%xmm10,%xmm11,%xmm8
+	xorq	%r8,%r12
+	shrdq	$6,%r14,%r14
+	vpsrlq	$7,%xmm10,%xmm10
+	xorq	%r11,%r15
+	addq	%r12,%r9
+	vpxor	%xmm9,%xmm8,%xmm8
+	shrdq	$14,%r13,%r13
+	andq	%r15,%rdi
+	vpsllq	$7,%xmm9,%xmm9
+	xorq	%r10,%r14
+	addq	%r13,%r9
+	vpxor	%xmm10,%xmm8,%xmm8
+	xorq	%r11,%rdi
+	shrdq	$28,%r14,%r14
+	vpsrlq	$6,%xmm0,%xmm11
+	addq	%r9,%rbx
+	addq	%rdi,%r9
+	vpxor	%xmm9,%xmm8,%xmm8
+	movq	%rbx,%r13
+	addq	%r9,%r14
+	vpsllq	$3,%xmm0,%xmm10
+	shrdq	$23,%r13,%r13
+	movq	%r14,%r9
+	vpaddq	%xmm8,%xmm1,%xmm1
+	movq	%rcx,%r12
+	shrdq	$5,%r14,%r14
+	vpsrlq	$19,%xmm0,%xmm9
+	xorq	%rbx,%r13
+	xorq	%rdx,%r12
+	vpxor	%xmm10,%xmm11,%xmm11
+	shrdq	$4,%r13,%r13
+	xorq	%r9,%r14
+	vpsllq	$42,%xmm10,%xmm10
+	andq	%rbx,%r12
+	xorq	%rbx,%r13
+	vpxor	%xmm9,%xmm11,%xmm11
+	addq	24(%rsp),%r8
+	movq	%r9,%rdi
+	vpsrlq	$42,%xmm9,%xmm9
+	xorq	%rdx,%r12
+	shrdq	$6,%r14,%r14
+	vpxor	%xmm10,%xmm11,%xmm11
+	xorq	%r10,%rdi
+	addq	%r12,%r8
+	vpxor	%xmm9,%xmm11,%xmm11
+	shrdq	$14,%r13,%r13
+	andq	%rdi,%r15
+	vpaddq	%xmm11,%xmm1,%xmm1
+	xorq	%r9,%r14
+	addq	%r13,%r8
+	vpaddq	-96(%rbp),%xmm1,%xmm10
+	xorq	%r10,%r15
+	shrdq	$28,%r14,%r14
+	addq	%r8,%rax
+	addq	%r15,%r8
+	movq	%rax,%r13
+	addq	%r8,%r14
+	vmovdqa	%xmm10,16(%rsp)
+	vpalignr	$8,%xmm2,%xmm3,%xmm8
+	shrdq	$23,%r13,%r13
+	movq	%r14,%r8
+	vpalignr	$8,%xmm6,%xmm7,%xmm11
+	movq	%rbx,%r12
+	shrdq	$5,%r14,%r14
+	vpsrlq	$1,%xmm8,%xmm10
+	xorq	%rax,%r13
+	xorq	%rcx,%r12
+	vpaddq	%xmm11,%xmm2,%xmm2
+	shrdq	$4,%r13,%r13
+	xorq	%r8,%r14
+	vpsrlq	$7,%xmm8,%xmm11
+	andq	%rax,%r12
+	xorq	%rax,%r13
+	vpsllq	$56,%xmm8,%xmm9
+	addq	32(%rsp),%rdx
+	movq	%r8,%r15
+	vpxor	%xmm10,%xmm11,%xmm8
+	xorq	%rcx,%r12
+	shrdq	$6,%r14,%r14
+	vpsrlq	$7,%xmm10,%xmm10
+	xorq	%r9,%r15
+	addq	%r12,%rdx
+	vpxor	%xmm9,%xmm8,%xmm8
+	shrdq	$14,%r13,%r13
+	andq	%r15,%rdi
+	vpsllq	$7,%xmm9,%xmm9
+	xorq	%r8,%r14
+	addq	%r13,%rdx
+	vpxor	%xmm10,%xmm8,%xmm8
+	xorq	%r9,%rdi
+	shrdq	$28,%r14,%r14
+	vpsrlq	$6,%xmm1,%xmm11
+	addq	%rdx,%r11
+	addq	%rdi,%rdx
+	vpxor	%xmm9,%xmm8,%xmm8
+	movq	%r11,%r13
+	addq	%rdx,%r14
+	vpsllq	$3,%xmm1,%xmm10
+	shrdq	$23,%r13,%r13
+	movq	%r14,%rdx
+	vpaddq	%xmm8,%xmm2,%xmm2
+	movq	%rax,%r12
+	shrdq	$5,%r14,%r14
+	vpsrlq	$19,%xmm1,%xmm9
+	xorq	%r11,%r13
+	xorq	%rbx,%r12
+	vpxor	%xmm10,%xmm11,%xmm11
+	shrdq	$4,%r13,%r13
+	xorq	%rdx,%r14
+	vpsllq	$42,%xmm10,%xmm10
+	andq	%r11,%r12
+	xorq	%r11,%r13
+	vpxor	%xmm9,%xmm11,%xmm11
+	addq	40(%rsp),%rcx
+	movq	%rdx,%rdi
+	vpsrlq	$42,%xmm9,%xmm9
+	xorq	%rbx,%r12
+	shrdq	$6,%r14,%r14
+	vpxor	%xmm10,%xmm11,%xmm11
+	xorq	%r8,%rdi
+	addq	%r12,%rcx
+	vpxor	%xmm9,%xmm11,%xmm11
+	shrdq	$14,%r13,%r13
+	andq	%rdi,%r15
+	vpaddq	%xmm11,%xmm2,%xmm2
+	xorq	%rdx,%r14
+	addq	%r13,%rcx
+	vpaddq	-64(%rbp),%xmm2,%xmm10
+	xorq	%r8,%r15
+	shrdq	$28,%r14,%r14
+	addq	%rcx,%r10
+	addq	%r15,%rcx
+	movq	%r10,%r13
+	addq	%rcx,%r14
+	vmovdqa	%xmm10,32(%rsp)
+	vpalignr	$8,%xmm3,%xmm4,%xmm8
+	shrdq	$23,%r13,%r13
+	movq	%r14,%rcx
+	vpalignr	$8,%xmm7,%xmm0,%xmm11
+	movq	%r11,%r12
+	shrdq	$5,%r14,%r14
+	vpsrlq	$1,%xmm8,%xmm10
+	xorq	%r10,%r13
+	xorq	%rax,%r12
+	vpaddq	%xmm11,%xmm3,%xmm3
+	shrdq	$4,%r13,%r13
+	xorq	%rcx,%r14
+	vpsrlq	$7,%xmm8,%xmm11
+	andq	%r10,%r12
+	xorq	%r10,%r13
+	vpsllq	$56,%xmm8,%xmm9
+	addq	48(%rsp),%rbx
+	movq	%rcx,%r15
+	vpxor	%xmm10,%xmm11,%xmm8
+	xorq	%rax,%r12
+	shrdq	$6,%r14,%r14
+	vpsrlq	$7,%xmm10,%xmm10
+	xorq	%rdx,%r15
+	addq	%r12,%rbx
+	vpxor	%xmm9,%xmm8,%xmm8
+	shrdq	$14,%r13,%r13
+	andq	%r15,%rdi
+	vpsllq	$7,%xmm9,%xmm9
+	xorq	%rcx,%r14
+	addq	%r13,%rbx
+	vpxor	%xmm10,%xmm8,%xmm8
+	xorq	%rdx,%rdi
+	shrdq	$28,%r14,%r14
+	vpsrlq	$6,%xmm2,%xmm11
+	addq	%rbx,%r9
+	addq	%rdi,%rbx
+	vpxor	%xmm9,%xmm8,%xmm8
+	movq	%r9,%r13
+	addq	%rbx,%r14
+	vpsllq	$3,%xmm2,%xmm10
+	shrdq	$23,%r13,%r13
+	movq	%r14,%rbx
+	vpaddq	%xmm8,%xmm3,%xmm3
+	movq	%r10,%r12
+	shrdq	$5,%r14,%r14
+	vpsrlq	$19,%xmm2,%xmm9
+	xorq	%r9,%r13
+	xorq	%r11,%r12
+	vpxor	%xmm10,%xmm11,%xmm11
+	shrdq	$4,%r13,%r13
+	xorq	%rbx,%r14
+	vpsllq	$42,%xmm10,%xmm10
+	andq	%r9,%r12
+	xorq	%r9,%r13
+	vpxor	%xmm9,%xmm11,%xmm11
+	addq	56(%rsp),%rax
+	movq	%rbx,%rdi
+	vpsrlq	$42,%xmm9,%xmm9
+	xorq	%r11,%r12
+	shrdq	$6,%r14,%r14
+	vpxor	%xmm10,%xmm11,%xmm11
+	xorq	%rcx,%rdi
+	addq	%r12,%rax
+	vpxor	%xmm9,%xmm11,%xmm11
+	shrdq	$14,%r13,%r13
+	andq	%rdi,%r15
+	vpaddq	%xmm11,%xmm3,%xmm3
+	xorq	%rbx,%r14
+	addq	%r13,%rax
+	vpaddq	-32(%rbp),%xmm3,%xmm10
+	xorq	%rcx,%r15
+	shrdq	$28,%r14,%r14
+	addq	%rax,%r8
+	addq	%r15,%rax
+	movq	%r8,%r13
+	addq	%rax,%r14
+	vmovdqa	%xmm10,48(%rsp)
+	vpalignr	$8,%xmm4,%xmm5,%xmm8
+	shrdq	$23,%r13,%r13
+	movq	%r14,%rax
+	vpalignr	$8,%xmm0,%xmm1,%xmm11
+	movq	%r9,%r12
+	shrdq	$5,%r14,%r14
+	vpsrlq	$1,%xmm8,%xmm10
+	xorq	%r8,%r13
+	xorq	%r10,%r12
+	vpaddq	%xmm11,%xmm4,%xmm4
+	shrdq	$4,%r13,%r13
+	xorq	%rax,%r14
+	vpsrlq	$7,%xmm8,%xmm11
+	andq	%r8,%r12
+	xorq	%r8,%r13
+	vpsllq	$56,%xmm8,%xmm9
+	addq	64(%rsp),%r11
+	movq	%rax,%r15
+	vpxor	%xmm10,%xmm11,%xmm8
+	xorq	%r10,%r12
+	shrdq	$6,%r14,%r14
+	vpsrlq	$7,%xmm10,%xmm10
+	xorq	%rbx,%r15
+	addq	%r12,%r11
+	vpxor	%xmm9,%xmm8,%xmm8
+	shrdq	$14,%r13,%r13
+	andq	%r15,%rdi
+	vpsllq	$7,%xmm9,%xmm9
+	xorq	%rax,%r14
+	addq	%r13,%r11
+	vpxor	%xmm10,%xmm8,%xmm8
+	xorq	%rbx,%rdi
+	shrdq	$28,%r14,%r14
+	vpsrlq	$6,%xmm3,%xmm11
+	addq	%r11,%rdx
+	addq	%rdi,%r11
+	vpxor	%xmm9,%xmm8,%xmm8
+	movq	%rdx,%r13
+	addq	%r11,%r14
+	vpsllq	$3,%xmm3,%xmm10
+	shrdq	$23,%r13,%r13
+	movq	%r14,%r11
+	vpaddq	%xmm8,%xmm4,%xmm4
+	movq	%r8,%r12
+	shrdq	$5,%r14,%r14
+	vpsrlq	$19,%xmm3,%xmm9
+	xorq	%rdx,%r13
+	xorq	%r9,%r12
+	vpxor	%xmm10,%xmm11,%xmm11
+	shrdq	$4,%r13,%r13
+	xorq	%r11,%r14
+	vpsllq	$42,%xmm10,%xmm10
+	andq	%rdx,%r12
+	xorq	%rdx,%r13
+	vpxor	%xmm9,%xmm11,%xmm11
+	addq	72(%rsp),%r10
+	movq	%r11,%rdi
+	vpsrlq	$42,%xmm9,%xmm9
+	xorq	%r9,%r12
+	shrdq	$6,%r14,%r14
+	vpxor	%xmm10,%xmm11,%xmm11
+	xorq	%rax,%rdi
+	addq	%r12,%r10
+	vpxor	%xmm9,%xmm11,%xmm11
+	shrdq	$14,%r13,%r13
+	andq	%rdi,%r15
+	vpaddq	%xmm11,%xmm4,%xmm4
+	xorq	%r11,%r14
+	addq	%r13,%r10
+	vpaddq	0(%rbp),%xmm4,%xmm10
+	xorq	%rax,%r15
+	shrdq	$28,%r14,%r14
+	addq	%r10,%rcx
+	addq	%r15,%r10
+	movq	%rcx,%r13
+	addq	%r10,%r14
+	vmovdqa	%xmm10,64(%rsp)
+	vpalignr	$8,%xmm5,%xmm6,%xmm8
+	shrdq	$23,%r13,%r13
+	movq	%r14,%r10
+	vpalignr	$8,%xmm1,%xmm2,%xmm11
+	movq	%rdx,%r12
+	shrdq	$5,%r14,%r14
+	vpsrlq	$1,%xmm8,%xmm10
+	xorq	%rcx,%r13
+	xorq	%r8,%r12
+	vpaddq	%xmm11,%xmm5,%xmm5
+	shrdq	$4,%r13,%r13
+	xorq	%r10,%r14
+	vpsrlq	$7,%xmm8,%xmm11
+	andq	%rcx,%r12
+	xorq	%rcx,%r13
+	vpsllq	$56,%xmm8,%xmm9
+	addq	80(%rsp),%r9
+	movq	%r10,%r15
+	vpxor	%xmm10,%xmm11,%xmm8
+	xorq	%r8,%r12
+	shrdq	$6,%r14,%r14
+	vpsrlq	$7,%xmm10,%xmm10
+	xorq	%r11,%r15
+	addq	%r12,%r9
+	vpxor	%xmm9,%xmm8,%xmm8
+	shrdq	$14,%r13,%r13
+	andq	%r15,%rdi
+	vpsllq	$7,%xmm9,%xmm9
+	xorq	%r10,%r14
+	addq	%r13,%r9
+	vpxor	%xmm10,%xmm8,%xmm8
+	xorq	%r11,%rdi
+	shrdq	$28,%r14,%r14
+	vpsrlq	$6,%xmm4,%xmm11
+	addq	%r9,%rbx
+	addq	%rdi,%r9
+	vpxor	%xmm9,%xmm8,%xmm8
+	movq	%rbx,%r13
+	addq	%r9,%r14
+	vpsllq	$3,%xmm4,%xmm10
+	shrdq	$23,%r13,%r13
+	movq	%r14,%r9
+	vpaddq	%xmm8,%xmm5,%xmm5
+	movq	%rcx,%r12
+	shrdq	$5,%r14,%r14
+	vpsrlq	$19,%xmm4,%xmm9
+	xorq	%rbx,%r13
+	xorq	%rdx,%r12
+	vpxor	%xmm10,%xmm11,%xmm11
+	shrdq	$4,%r13,%r13
+	xorq	%r9,%r14
+	vpsllq	$42,%xmm10,%xmm10
+	andq	%rbx,%r12
+	xorq	%rbx,%r13
+	vpxor	%xmm9,%xmm11,%xmm11
+	addq	88(%rsp),%r8
+	movq	%r9,%rdi
+	vpsrlq	$42,%xmm9,%xmm9
+	xorq	%rdx,%r12
+	shrdq	$6,%r14,%r14
+	vpxor	%xmm10,%xmm11,%xmm11
+	xorq	%r10,%rdi
+	addq	%r12,%r8
+	vpxor	%xmm9,%xmm11,%xmm11
+	shrdq	$14,%r13,%r13
+	andq	%rdi,%r15
+	vpaddq	%xmm11,%xmm5,%xmm5
+	xorq	%r9,%r14
+	addq	%r13,%r8
+	vpaddq	32(%rbp),%xmm5,%xmm10
+	xorq	%r10,%r15
+	shrdq	$28,%r14,%r14
+	addq	%r8,%rax
+	addq	%r15,%r8
+	movq	%rax,%r13
+	addq	%r8,%r14
+	vmovdqa	%xmm10,80(%rsp)
+	vpalignr	$8,%xmm6,%xmm7,%xmm8
+	shrdq	$23,%r13,%r13
+	movq	%r14,%r8
+	vpalignr	$8,%xmm2,%xmm3,%xmm11
+	movq	%rbx,%r12
+	shrdq	$5,%r14,%r14
+	vpsrlq	$1,%xmm8,%xmm10
+	xorq	%rax,%r13
+	xorq	%rcx,%r12
+	vpaddq	%xmm11,%xmm6,%xmm6
+	shrdq	$4,%r13,%r13
+	xorq	%r8,%r14
+	vpsrlq	$7,%xmm8,%xmm11
+	andq	%rax,%r12
+	xorq	%rax,%r13
+	vpsllq	$56,%xmm8,%xmm9
+	addq	96(%rsp),%rdx
+	movq	%r8,%r15
+	vpxor	%xmm10,%xmm11,%xmm8
+	xorq	%rcx,%r12
+	shrdq	$6,%r14,%r14
+	vpsrlq	$7,%xmm10,%xmm10
+	xorq	%r9,%r15
+	addq	%r12,%rdx
+	vpxor	%xmm9,%xmm8,%xmm8
+	shrdq	$14,%r13,%r13
+	andq	%r15,%rdi
+	vpsllq	$7,%xmm9,%xmm9
+	xorq	%r8,%r14
+	addq	%r13,%rdx
+	vpxor	%xmm10,%xmm8,%xmm8
+	xorq	%r9,%rdi
+	shrdq	$28,%r14,%r14
+	vpsrlq	$6,%xmm5,%xmm11
+	addq	%rdx,%r11
+	addq	%rdi,%rdx
+	vpxor	%xmm9,%xmm8,%xmm8
+	movq	%r11,%r13
+	addq	%rdx,%r14
+	vpsllq	$3,%xmm5,%xmm10
+	shrdq	$23,%r13,%r13
+	movq	%r14,%rdx
+	vpaddq	%xmm8,%xmm6,%xmm6
+	movq	%rax,%r12
+	shrdq	$5,%r14,%r14
+	vpsrlq	$19,%xmm5,%xmm9
+	xorq	%r11,%r13
+	xorq	%rbx,%r12
+	vpxor	%xmm10,%xmm11,%xmm11
+	shrdq	$4,%r13,%r13
+	xorq	%rdx,%r14
+	vpsllq	$42,%xmm10,%xmm10
+	andq	%r11,%r12
+	xorq	%r11,%r13
+	vpxor	%xmm9,%xmm11,%xmm11
+	addq	104(%rsp),%rcx
+	movq	%rdx,%rdi
+	vpsrlq	$42,%xmm9,%xmm9
+	xorq	%rbx,%r12
+	shrdq	$6,%r14,%r14
+	vpxor	%xmm10,%xmm11,%xmm11
+	xorq	%r8,%rdi
+	addq	%r12,%rcx
+	vpxor	%xmm9,%xmm11,%xmm11
+	shrdq	$14,%r13,%r13
+	andq	%rdi,%r15
+	vpaddq	%xmm11,%xmm6,%xmm6
+	xorq	%rdx,%r14
+	addq	%r13,%rcx
+	vpaddq	64(%rbp),%xmm6,%xmm10
+	xorq	%r8,%r15
+	shrdq	$28,%r14,%r14
+	addq	%rcx,%r10
+	addq	%r15,%rcx
+	movq	%r10,%r13
+	addq	%rcx,%r14
+	vmovdqa	%xmm10,96(%rsp)
+	vpalignr	$8,%xmm7,%xmm0,%xmm8
+	shrdq	$23,%r13,%r13
+	movq	%r14,%rcx
+	vpalignr	$8,%xmm3,%xmm4,%xmm11
+	movq	%r11,%r12
+	shrdq	$5,%r14,%r14
+	vpsrlq	$1,%xmm8,%xmm10
+	xorq	%r10,%r13
+	xorq	%rax,%r12
+	vpaddq	%xmm11,%xmm7,%xmm7
+	shrdq	$4,%r13,%r13
+	xorq	%rcx,%r14
+	vpsrlq	$7,%xmm8,%xmm11
+	andq	%r10,%r12
+	xorq	%r10,%r13
+	vpsllq	$56,%xmm8,%xmm9
+	addq	112(%rsp),%rbx
+	movq	%rcx,%r15
+	vpxor	%xmm10,%xmm11,%xmm8
+	xorq	%rax,%r12
+	shrdq	$6,%r14,%r14
+	vpsrlq	$7,%xmm10,%xmm10
+	xorq	%rdx,%r15
+	addq	%r12,%rbx
+	vpxor	%xmm9,%xmm8,%xmm8
+	shrdq	$14,%r13,%r13
+	andq	%r15,%rdi
+	vpsllq	$7,%xmm9,%xmm9
+	xorq	%rcx,%r14
+	addq	%r13,%rbx
+	vpxor	%xmm10,%xmm8,%xmm8
+	xorq	%rdx,%rdi
+	shrdq	$28,%r14,%r14
+	vpsrlq	$6,%xmm6,%xmm11
+	addq	%rbx,%r9
+	addq	%rdi,%rbx
+	vpxor	%xmm9,%xmm8,%xmm8
+	movq	%r9,%r13
+	addq	%rbx,%r14
+	vpsllq	$3,%xmm6,%xmm10
+	shrdq	$23,%r13,%r13
+	movq	%r14,%rbx
+	vpaddq	%xmm8,%xmm7,%xmm7
+	movq	%r10,%r12
+	shrdq	$5,%r14,%r14
+	vpsrlq	$19,%xmm6,%xmm9
+	xorq	%r9,%r13
+	xorq	%r11,%r12
+	vpxor	%xmm10,%xmm11,%xmm11
+	shrdq	$4,%r13,%r13
+	xorq	%rbx,%r14
+	vpsllq	$42,%xmm10,%xmm10
+	andq	%r9,%r12
+	xorq	%r9,%r13
+	vpxor	%xmm9,%xmm11,%xmm11
+	addq	120(%rsp),%rax
+	movq	%rbx,%rdi
+	vpsrlq	$42,%xmm9,%xmm9
+	xorq	%r11,%r12
+	shrdq	$6,%r14,%r14
+	vpxor	%xmm10,%xmm11,%xmm11
+	xorq	%rcx,%rdi
+	addq	%r12,%rax
+	vpxor	%xmm9,%xmm11,%xmm11
+	shrdq	$14,%r13,%r13
+	andq	%rdi,%r15
+	vpaddq	%xmm11,%xmm7,%xmm7
+	xorq	%rbx,%r14
+	addq	%r13,%rax
+	vpaddq	96(%rbp),%xmm7,%xmm10
+	xorq	%rcx,%r15
+	shrdq	$28,%r14,%r14
+	addq	%rax,%r8
+	addq	%r15,%rax
+	movq	%r8,%r13
+	addq	%rax,%r14
+	vmovdqa	%xmm10,112(%rsp)
+	cmpb	$0,135(%rbp)
+	jne	.Lavx_00_47
+	shrdq	$23,%r13,%r13
+	movq	%r14,%rax
+	movq	%r9,%r12
+	shrdq	$5,%r14,%r14
+	xorq	%r8,%r13
+	xorq	%r10,%r12
+	shrdq	$4,%r13,%r13
+	xorq	%rax,%r14
+	andq	%r8,%r12
+	xorq	%r8,%r13
+	addq	0(%rsp),%r11
+	movq	%rax,%r15
+	xorq	%r10,%r12
+	shrdq	$6,%r14,%r14
+	xorq	%rbx,%r15
+	addq	%r12,%r11
+	shrdq	$14,%r13,%r13
+	andq	%r15,%rdi
+	xorq	%rax,%r14
+	addq	%r13,%r11
+	xorq	%rbx,%rdi
+	shrdq	$28,%r14,%r14
+	addq	%r11,%rdx
+	addq	%rdi,%r11
+	movq	%rdx,%r13
+	addq	%r11,%r14
+	shrdq	$23,%r13,%r13
+	movq	%r14,%r11
+	movq	%r8,%r12
+	shrdq	$5,%r14,%r14
+	xorq	%rdx,%r13
+	xorq	%r9,%r12
+	shrdq	$4,%r13,%r13
+	xorq	%r11,%r14
+	andq	%rdx,%r12
+	xorq	%rdx,%r13
+	addq	8(%rsp),%r10
+	movq	%r11,%rdi
+	xorq	%r9,%r12
+	shrdq	$6,%r14,%r14
+	xorq	%rax,%rdi
+	addq	%r12,%r10
+	shrdq	$14,%r13,%r13
+	andq	%rdi,%r15
+	xorq	%r11,%r14
+	addq	%r13,%r10
+	xorq	%rax,%r15
+	shrdq	$28,%r14,%r14
+	addq	%r10,%rcx
+	addq	%r15,%r10
+	movq	%rcx,%r13
+	addq	%r10,%r14
+	shrdq	$23,%r13,%r13
+	movq	%r14,%r10
+	movq	%rdx,%r12
+	shrdq	$5,%r14,%r14
+	xorq	%rcx,%r13
+	xorq	%r8,%r12
+	shrdq	$4,%r13,%r13
+	xorq	%r10,%r14
+	andq	%rcx,%r12
+	xorq	%rcx,%r13
+	addq	16(%rsp),%r9
+	movq	%r10,%r15
+	xorq	%r8,%r12
+	shrdq	$6,%r14,%r14
+	xorq	%r11,%r15
+	addq	%r12,%r9
+	shrdq	$14,%r13,%r13
+	andq	%r15,%rdi
+	xorq	%r10,%r14
+	addq	%r13,%r9
+	xorq	%r11,%rdi
+	shrdq	$28,%r14,%r14
+	addq	%r9,%rbx
+	addq	%rdi,%r9
+	movq	%rbx,%r13
+	addq	%r9,%r14
+	shrdq	$23,%r13,%r13
+	movq	%r14,%r9
+	movq	%rcx,%r12
+	shrdq	$5,%r14,%r14
+	xorq	%rbx,%r13
+	xorq	%rdx,%r12
+	shrdq	$4,%r13,%r13
+	xorq	%r9,%r14
+	andq	%rbx,%r12
+	xorq	%rbx,%r13
+	addq	24(%rsp),%r8
+	movq	%r9,%rdi
+	xorq	%rdx,%r12
+	shrdq	$6,%r14,%r14
+	xorq	%r10,%rdi
+	addq	%r12,%r8
+	shrdq	$14,%r13,%r13
+	andq	%rdi,%r15
+	xorq	%r9,%r14
+	addq	%r13,%r8
+	xorq	%r10,%r15
+	shrdq	$28,%r14,%r14
+	addq	%r8,%rax
+	addq	%r15,%r8
+	movq	%rax,%r13
+	addq	%r8,%r14
+	shrdq	$23,%r13,%r13
+	movq	%r14,%r8
+	movq	%rbx,%r12
+	shrdq	$5,%r14,%r14
+	xorq	%rax,%r13
+	xorq	%rcx,%r12
+	shrdq	$4,%r13,%r13
+	xorq	%r8,%r14
+	andq	%rax,%r12
+	xorq	%rax,%r13
+	addq	32(%rsp),%rdx
+	movq	%r8,%r15
+	xorq	%rcx,%r12
+	shrdq	$6,%r14,%r14
+	xorq	%r9,%r15
+	addq	%r12,%rdx
+	shrdq	$14,%r13,%r13
+	andq	%r15,%rdi
+	xorq	%r8,%r14
+	addq	%r13,%rdx
+	xorq	%r9,%rdi
+	shrdq	$28,%r14,%r14
+	addq	%rdx,%r11
+	addq	%rdi,%rdx
+	movq	%r11,%r13
+	addq	%rdx,%r14
+	shrdq	$23,%r13,%r13
+	movq	%r14,%rdx
+	movq	%rax,%r12
+	shrdq	$5,%r14,%r14
+	xorq	%r11,%r13
+	xorq	%rbx,%r12
+	shrdq	$4,%r13,%r13
+	xorq	%rdx,%r14
+	andq	%r11,%r12
+	xorq	%r11,%r13
+	addq	40(%rsp),%rcx
+	movq	%rdx,%rdi
+	xorq	%rbx,%r12
+	shrdq	$6,%r14,%r14
+	xorq	%r8,%rdi
+	addq	%r12,%rcx
+	shrdq	$14,%r13,%r13
+	andq	%rdi,%r15
+	xorq	%rdx,%r14
+	addq	%r13,%rcx
+	xorq	%r8,%r15
+	shrdq	$28,%r14,%r14
+	addq	%rcx,%r10
+	addq	%r15,%rcx
+	movq	%r10,%r13
+	addq	%rcx,%r14
+	shrdq	$23,%r13,%r13
+	movq	%r14,%rcx
+	movq	%r11,%r12
+	shrdq	$5,%r14,%r14
+	xorq	%r10,%r13
+	xorq	%rax,%r12
+	shrdq	$4,%r13,%r13
+	xorq	%rcx,%r14
+	andq	%r10,%r12
+	xorq	%r10,%r13
+	addq	48(%rsp),%rbx
+	movq	%rcx,%r15
+	xorq	%rax,%r12
+	shrdq	$6,%r14,%r14
+	xorq	%rdx,%r15
+	addq	%r12,%rbx
+	shrdq	$14,%r13,%r13
+	andq	%r15,%rdi
+	xorq	%rcx,%r14
+	addq	%r13,%rbx
+	xorq	%rdx,%rdi
+	shrdq	$28,%r14,%r14
+	addq	%rbx,%r9
+	addq	%rdi,%rbx
+	movq	%r9,%r13
+	addq	%rbx,%r14
+	shrdq	$23,%r13,%r13
+	movq	%r14,%rbx
+	movq	%r10,%r12
+	shrdq	$5,%r14,%r14
+	xorq	%r9,%r13
+	xorq	%r11,%r12
+	shrdq	$4,%r13,%r13
+	xorq	%rbx,%r14
+	andq	%r9,%r12
+	xorq	%r9,%r13
+	addq	56(%rsp),%rax
+	movq	%rbx,%rdi
+	xorq	%r11,%r12
+	shrdq	$6,%r14,%r14
+	xorq	%rcx,%rdi
+	addq	%r12,%rax
+	shrdq	$14,%r13,%r13
+	andq	%rdi,%r15
+	xorq	%rbx,%r14
+	addq	%r13,%rax
+	xorq	%rcx,%r15
+	shrdq	$28,%r14,%r14
+	addq	%rax,%r8
+	addq	%r15,%rax
+	movq	%r8,%r13
+	addq	%rax,%r14
+	shrdq	$23,%r13,%r13
+	movq	%r14,%rax
+	movq	%r9,%r12
+	shrdq	$5,%r14,%r14
+	xorq	%r8,%r13
+	xorq	%r10,%r12
+	shrdq	$4,%r13,%r13
+	xorq	%rax,%r14
+	andq	%r8,%r12
+	xorq	%r8,%r13
+	addq	64(%rsp),%r11
+	movq	%rax,%r15
+	xorq	%r10,%r12
+	shrdq	$6,%r14,%r14
+	xorq	%rbx,%r15
+	addq	%r12,%r11
+	shrdq	$14,%r13,%r13
+	andq	%r15,%rdi
+	xorq	%rax,%r14
+	addq	%r13,%r11
+	xorq	%rbx,%rdi
+	shrdq	$28,%r14,%r14
+	addq	%r11,%rdx
+	addq	%rdi,%r11
+	movq	%rdx,%r13
+	addq	%r11,%r14
+	shrdq	$23,%r13,%r13
+	movq	%r14,%r11
+	movq	%r8,%r12
+	shrdq	$5,%r14,%r14
+	xorq	%rdx,%r13
+	xorq	%r9,%r12
+	shrdq	$4,%r13,%r13
+	xorq	%r11,%r14
+	andq	%rdx,%r12
+	xorq	%rdx,%r13
+	addq	72(%rsp),%r10
+	movq	%r11,%rdi
+	xorq	%r9,%r12
+	shrdq	$6,%r14,%r14
+	xorq	%rax,%rdi
+	addq	%r12,%r10
+	shrdq	$14,%r13,%r13
+	andq	%rdi,%r15
+	xorq	%r11,%r14
+	addq	%r13,%r10
+	xorq	%rax,%r15
+	shrdq	$28,%r14,%r14
+	addq	%r10,%rcx
+	addq	%r15,%r10
+	movq	%rcx,%r13
+	addq	%r10,%r14
+	shrdq	$23,%r13,%r13
+	movq	%r14,%r10
+	movq	%rdx,%r12
+	shrdq	$5,%r14,%r14
+	xorq	%rcx,%r13
+	xorq	%r8,%r12
+	shrdq	$4,%r13,%r13
+	xorq	%r10,%r14
+	andq	%rcx,%r12
+	xorq	%rcx,%r13
+	addq	80(%rsp),%r9
+	movq	%r10,%r15
+	xorq	%r8,%r12
+	shrdq	$6,%r14,%r14
+	xorq	%r11,%r15
+	addq	%r12,%r9
+	shrdq	$14,%r13,%r13
+	andq	%r15,%rdi
+	xorq	%r10,%r14
+	addq	%r13,%r9
+	xorq	%r11,%rdi
+	shrdq	$28,%r14,%r14
+	addq	%r9,%rbx
+	addq	%rdi,%r9
+	movq	%rbx,%r13
+	addq	%r9,%r14
+	shrdq	$23,%r13,%r13
+	movq	%r14,%r9
+	movq	%rcx,%r12
+	shrdq	$5,%r14,%r14
+	xorq	%rbx,%r13
+	xorq	%rdx,%r12
+	shrdq	$4,%r13,%r13
+	xorq	%r9,%r14
+	andq	%rbx,%r12
+	xorq	%rbx,%r13
+	addq	88(%rsp),%r8
+	movq	%r9,%rdi
+	xorq	%rdx,%r12
+	shrdq	$6,%r14,%r14
+	xorq	%r10,%rdi
+	addq	%r12,%r8
+	shrdq	$14,%r13,%r13
+	andq	%rdi,%r15
+	xorq	%r9,%r14
+	addq	%r13,%r8
+	xorq	%r10,%r15
+	shrdq	$28,%r14,%r14
+	addq	%r8,%rax
+	addq	%r15,%r8
+	movq	%rax,%r13
+	addq	%r8,%r14
+	shrdq	$23,%r13,%r13
+	movq	%r14,%r8
+	movq	%rbx,%r12
+	shrdq	$5,%r14,%r14
+	xorq	%rax,%r13
+	xorq	%rcx,%r12
+	shrdq	$4,%r13,%r13
+	xorq	%r8,%r14
+	andq	%rax,%r12
+	xorq	%rax,%r13
+	addq	96(%rsp),%rdx
+	movq	%r8,%r15
+	xorq	%rcx,%r12
+	shrdq	$6,%r14,%r14
+	xorq	%r9,%r15
+	addq	%r12,%rdx
+	shrdq	$14,%r13,%r13
+	andq	%r15,%rdi
+	xorq	%r8,%r14
+	addq	%r13,%rdx
+	xorq	%r9,%rdi
+	shrdq	$28,%r14,%r14
+	addq	%rdx,%r11
+	addq	%rdi,%rdx
+	movq	%r11,%r13
+	addq	%rdx,%r14
+	shrdq	$23,%r13,%r13
+	movq	%r14,%rdx
+	movq	%rax,%r12
+	shrdq	$5,%r14,%r14
+	xorq	%r11,%r13
+	xorq	%rbx,%r12
+	shrdq	$4,%r13,%r13
+	xorq	%rdx,%r14
+	andq	%r11,%r12
+	xorq	%r11,%r13
+	addq	104(%rsp),%rcx
+	movq	%rdx,%rdi
+	xorq	%rbx,%r12
+	shrdq	$6,%r14,%r14
+	xorq	%r8,%rdi
+	addq	%r12,%rcx
+	shrdq	$14,%r13,%r13
+	andq	%rdi,%r15
+	xorq	%rdx,%r14
+	addq	%r13,%rcx
+	xorq	%r8,%r15
+	shrdq	$28,%r14,%r14
+	addq	%rcx,%r10
+	addq	%r15,%rcx
+	movq	%r10,%r13
+	addq	%rcx,%r14
+	shrdq	$23,%r13,%r13
+	movq	%r14,%rcx
+	movq	%r11,%r12
+	shrdq	$5,%r14,%r14
+	xorq	%r10,%r13
+	xorq	%rax,%r12
+	shrdq	$4,%r13,%r13
+	xorq	%rcx,%r14
+	andq	%r10,%r12
+	xorq	%r10,%r13
+	addq	112(%rsp),%rbx
+	movq	%rcx,%r15
+	xorq	%rax,%r12
+	shrdq	$6,%r14,%r14
+	xorq	%rdx,%r15
+	addq	%r12,%rbx
+	shrdq	$14,%r13,%r13
+	andq	%r15,%rdi
+	xorq	%rcx,%r14
+	addq	%r13,%rbx
+	xorq	%rdx,%rdi
+	shrdq	$28,%r14,%r14
+	addq	%rbx,%r9
+	addq	%rdi,%rbx
+	movq	%r9,%r13
+	addq	%rbx,%r14
+	shrdq	$23,%r13,%r13
+	movq	%r14,%rbx
+	movq	%r10,%r12
+	shrdq	$5,%r14,%r14
+	xorq	%r9,%r13
+	xorq	%r11,%r12
+	shrdq	$4,%r13,%r13
+	xorq	%rbx,%r14
+	andq	%r9,%r12
+	xorq	%r9,%r13
+	addq	120(%rsp),%rax
+	movq	%rbx,%rdi
+	xorq	%r11,%r12
+	shrdq	$6,%r14,%r14
+	xorq	%rcx,%rdi
+	addq	%r12,%rax
+	shrdq	$14,%r13,%r13
+	andq	%rdi,%r15
+	xorq	%rbx,%r14
+	addq	%r13,%rax
+	xorq	%rcx,%r15
+	shrdq	$28,%r14,%r14
+	addq	%rax,%r8
+	addq	%r15,%rax
+	movq	%r8,%r13
+	addq	%rax,%r14
+	movq	128+0(%rsp),%rdi
+	movq	%r14,%rax
+
+	addq	0(%rdi),%rax
+	leaq	128(%rsi),%rsi
+	addq	8(%rdi),%rbx
+	addq	16(%rdi),%rcx
+	addq	24(%rdi),%rdx
+	addq	32(%rdi),%r8
+	addq	40(%rdi),%r9
+	addq	48(%rdi),%r10
+	addq	56(%rdi),%r11
+
+	cmpq	128+16(%rsp),%rsi
+
+	movq	%rax,0(%rdi)
+	movq	%rbx,8(%rdi)
+	movq	%rcx,16(%rdi)
+	movq	%rdx,24(%rdi)
+	movq	%r8,32(%rdi)
+	movq	%r9,40(%rdi)
+	movq	%r10,48(%rdi)
+	movq	%r11,56(%rdi)
+	jb	.Lloop_avx
+
+	movq	128+24(%rsp),%rsi
+	vzeroupper
+	movq	(%rsi),%r15
+	movq	8(%rsi),%r14
+	movq	16(%rsi),%r13
+	movq	24(%rsi),%r12
+	movq	32(%rsi),%rbp
+	movq	40(%rsi),%rbx
+	leaq	48(%rsi),%rsp
+.Lepilogue_avx:
+	.byte	0xf3,0xc3
+.size	sha512_block_data_order_avx,.-sha512_block_data_order_avx
 #endif

third_party/boringssl/mac-x86/crypto/chacha/chacha-x86.S

diff --git a/third_party/boringssl/mac-x86/crypto/chacha/chacha-x86.S b/third_party/boringssl/mac-x86/crypto/chacha/chacha-x86.S
new file mode 100644
index 0000000..5de98a3
--- /dev/null
+++ b/third_party/boringssl/mac-x86/crypto/chacha/chacha-x86.S
@@ -0,0 +1,969 @@
+#if defined(__i386__)
+.file	"chacha-x86.S"
+.text
+.globl	_ChaCha20_ctr32
+.private_extern	_ChaCha20_ctr32
+.align	4
+_ChaCha20_ctr32:
+L_ChaCha20_ctr32_begin:
+	pushl	%ebp
+	pushl	%ebx
+	pushl	%esi
+	pushl	%edi
+	xorl	%eax,%eax
+	cmpl	28(%esp),%eax
+	je	L000no_data
+	call	Lpic_point
+Lpic_point:
+	popl	%eax
+	movl	L_OPENSSL_ia32cap_P$non_lazy_ptr-Lpic_point(%eax),%ebp
+	testl	$16777216,(%ebp)
+	jz	L001x86
+	testl	$512,4(%ebp)
+	jz	L001x86
+	jmp	Lssse3_shortcut
+L001x86:
+	movl	32(%esp),%esi
+	movl	36(%esp),%edi
+	subl	$132,%esp
+	movl	(%esi),%eax
+	movl	4(%esi),%ebx
+	movl	8(%esi),%ecx
+	movl	12(%esi),%edx
+	movl	%eax,80(%esp)
+	movl	%ebx,84(%esp)
+	movl	%ecx,88(%esp)
+	movl	%edx,92(%esp)
+	movl	16(%esi),%eax
+	movl	20(%esi),%ebx
+	movl	24(%esi),%ecx
+	movl	28(%esi),%edx
+	movl	%eax,96(%esp)
+	movl	%ebx,100(%esp)
+	movl	%ecx,104(%esp)
+	movl	%edx,108(%esp)
+	movl	(%edi),%eax
+	movl	4(%edi),%ebx
+	movl	8(%edi),%ecx
+	movl	12(%edi),%edx
+	subl	$1,%eax
+	movl	%eax,112(%esp)
+	movl	%ebx,116(%esp)
+	movl	%ecx,120(%esp)
+	movl	%edx,124(%esp)
+	jmp	L002entry
+.align	4,0x90
+L003outer_loop:
+	movl	%ebx,156(%esp)
+	movl	%eax,152(%esp)
+	movl	%ecx,160(%esp)
+L002entry:
+	movl	$1634760805,%eax
+	movl	$857760878,4(%esp)
+	movl	$2036477234,8(%esp)
+	movl	$1797285236,12(%esp)
+	movl	84(%esp),%ebx
+	movl	88(%esp),%ebp
+	movl	104(%esp),%ecx
+	movl	108(%esp),%esi
+	movl	116(%esp),%edx
+	movl	120(%esp),%edi
+	movl	%ebx,20(%esp)
+	movl	%ebp,24(%esp)
+	movl	%ecx,40(%esp)
+	movl	%esi,44(%esp)
+	movl	%edx,52(%esp)
+	movl	%edi,56(%esp)
+	movl	92(%esp),%ebx
+	movl	124(%esp),%edi
+	movl	112(%esp),%edx
+	movl	80(%esp),%ebp
+	movl	96(%esp),%ecx
+	movl	100(%esp),%esi
+	addl	$1,%edx
+	movl	%ebx,28(%esp)
+	movl	%edi,60(%esp)
+	movl	%edx,112(%esp)
+	movl	$10,%ebx
+	jmp	L004loop
+.align	4,0x90
+L004loop:
+	addl	%ebp,%eax
+	movl	%ebx,128(%esp)
+	movl	%ebp,%ebx
+	xorl	%eax,%edx
+	roll	$16,%edx
+	addl	%edx,%ecx
+	xorl	%ecx,%ebx
+	movl	52(%esp),%edi
+	roll	$12,%ebx
+	movl	20(%esp),%ebp
+	addl	%ebx,%eax
+	xorl	%eax,%edx
+	movl	%eax,(%esp)
+	roll	$8,%edx
+	movl	4(%esp),%eax
+	addl	%edx,%ecx
+	movl	%edx,48(%esp)
+	xorl	%ecx,%ebx
+	addl	%ebp,%eax
+	roll	$7,%ebx
+	xorl	%eax,%edi
+	movl	%ecx,32(%esp)
+	roll	$16,%edi
+	movl	%ebx,16(%esp)
+	addl	%edi,%esi
+	movl	40(%esp),%ecx
+	xorl	%esi,%ebp
+	movl	56(%esp),%edx
+	roll	$12,%ebp
+	movl	24(%esp),%ebx
+	addl	%ebp,%eax
+	xorl	%eax,%edi
+	movl	%eax,4(%esp)
+	roll	$8,%edi
+	movl	8(%esp),%eax
+	addl	%edi,%esi
+	movl	%edi,52(%esp)
+	xorl	%esi,%ebp
+	addl	%ebx,%eax
+	roll	$7,%ebp
+	xorl	%eax,%edx
+	movl	%esi,36(%esp)
+	roll	$16,%edx
+	movl	%ebp,20(%esp)
+	addl	%edx,%ecx
+	movl	44(%esp),%esi
+	xorl	%ecx,%ebx
+	movl	60(%esp),%edi
+	roll	$12,%ebx
+	movl	28(%esp),%ebp
+	addl	%ebx,%eax
+	xorl	%eax,%edx
+	movl	%eax,8(%esp)
+	roll	$8,%edx
+	movl	12(%esp),%eax
+	addl	%edx,%ecx
+	movl	%edx,56(%esp)
+	xorl	%ecx,%ebx
+	addl	%ebp,%eax
+	roll	$7,%ebx
+	xorl	%eax,%edi
+	roll	$16,%edi
+	movl	%ebx,24(%esp)
+	addl	%edi,%esi
+	xorl	%esi,%ebp
+	roll	$12,%ebp
+	movl	20(%esp),%ebx
+	addl	%ebp,%eax
+	xorl	%eax,%edi
+	movl	%eax,12(%esp)
+	roll	$8,%edi
+	movl	(%esp),%eax
+	addl	%edi,%esi
+	movl	%edi,%edx
+	xorl	%esi,%ebp
+	addl	%ebx,%eax
+	roll	$7,%ebp
+	xorl	%eax,%edx
+	roll	$16,%edx
+	movl	%ebp,28(%esp)
+	addl	%edx,%ecx
+	xorl	%ecx,%ebx
+	movl	48(%esp),%edi
+	roll	$12,%ebx
+	movl	24(%esp),%ebp
+	addl	%ebx,%eax
+	xorl	%eax,%edx
+	movl	%eax,(%esp)
+	roll	$8,%edx
+	movl	4(%esp),%eax
+	addl	%edx,%ecx
+	movl	%edx,60(%esp)
+	xorl	%ecx,%ebx
+	addl	%ebp,%eax
+	roll	$7,%ebx
+	xorl	%eax,%edi
+	movl	%ecx,40(%esp)
+	roll	$16,%edi
+	movl	%ebx,20(%esp)
+	addl	%edi,%esi
+	movl	32(%esp),%ecx
+	xorl	%esi,%ebp
+	movl	52(%esp),%edx
+	roll	$12,%ebp
+	movl	28(%esp),%ebx
+	addl	%ebp,%eax
+	xorl	%eax,%edi
+	movl	%eax,4(%esp)
+	roll	$8,%edi
+	movl	8(%esp),%eax
+	addl	%edi,%esi
+	movl	%edi,48(%esp)
+	xorl	%esi,%ebp
+	addl	%ebx,%eax
+	roll	$7,%ebp
+	xorl	%eax,%edx
+	movl	%esi,44(%esp)
+	roll	$16,%edx
+	movl	%ebp,24(%esp)
+	addl	%edx,%ecx
+	movl	36(%esp),%esi
+	xorl	%ecx,%ebx
+	movl	56(%esp),%edi
+	roll	$12,%ebx
+	movl	16(%esp),%ebp
+	addl	%ebx,%eax
+	xorl	%eax,%edx
+	movl	%eax,8(%esp)
+	roll	$8,%edx
+	movl	12(%esp),%eax
+	addl	%edx,%ecx
+	movl	%edx,52(%esp)
+	xorl	%ecx,%ebx
+	addl	%ebp,%eax
+	roll	$7,%ebx
+	xorl	%eax,%edi
+	roll	$16,%edi
+	movl	%ebx,28(%esp)
+	addl	%edi,%esi
+	xorl	%esi,%ebp
+	movl	48(%esp),%edx
+	roll	$12,%ebp
+	movl	128(%esp),%ebx
+	addl	%ebp,%eax
+	xorl	%eax,%edi
+	movl	%eax,12(%esp)
+	roll	$8,%edi
+	movl	(%esp),%eax
+	addl	%edi,%esi
+	movl	%edi,56(%esp)
+	xorl	%esi,%ebp
+	roll	$7,%ebp
+	decl	%ebx
+	jnz	L004loop
+	movl	160(%esp),%ebx
+	addl	$1634760805,%eax
+	addl	80(%esp),%ebp
+	addl	96(%esp),%ecx
+	addl	100(%esp),%esi
+	cmpl	$64,%ebx
+	jb	L005tail
+	movl	156(%esp),%ebx
+	addl	112(%esp),%edx
+	addl	120(%esp),%edi
+	xorl	(%ebx),%eax
+	xorl	16(%ebx),%ebp
+	movl	%eax,(%esp)
+	movl	152(%esp),%eax
+	xorl	32(%ebx),%ecx
+	xorl	36(%ebx),%esi
+	xorl	48(%ebx),%edx
+	xorl	56(%ebx),%edi
+	movl	%ebp,16(%eax)
+	movl	%ecx,32(%eax)
+	movl	%esi,36(%eax)
+	movl	%edx,48(%eax)
+	movl	%edi,56(%eax)
+	movl	4(%esp),%ebp
+	movl	8(%esp),%ecx
+	movl	12(%esp),%esi
+	movl	20(%esp),%edx
+	movl	24(%esp),%edi
+	addl	$857760878,%ebp
+	addl	$2036477234,%ecx
+	addl	$1797285236,%esi
+	addl	84(%esp),%edx
+	addl	88(%esp),%edi
+	xorl	4(%ebx),%ebp
+	xorl	8(%ebx),%ecx
+	xorl	12(%ebx),%esi
+	xorl	20(%ebx),%edx
+	xorl	24(%ebx),%edi
+	movl	%ebp,4(%eax)
+	movl	%ecx,8(%eax)
+	movl	%esi,12(%eax)
+	movl	%edx,20(%eax)
+	movl	%edi,24(%eax)
+	movl	28(%esp),%ebp
+	movl	40(%esp),%ecx
+	movl	44(%esp),%esi
+	movl	52(%esp),%edx
+	movl	60(%esp),%edi
+	addl	92(%esp),%ebp
+	addl	104(%esp),%ecx
+	addl	108(%esp),%esi
+	addl	116(%esp),%edx
+	addl	124(%esp),%edi
+	xorl	28(%ebx),%ebp
+	xorl	40(%ebx),%ecx
+	xorl	44(%ebx),%esi
+	xorl	52(%ebx),%edx
+	xorl	60(%ebx),%edi
+	leal	64(%ebx),%ebx
+	movl	%ebp,28(%eax)
+	movl	(%esp),%ebp
+	movl	%ecx,40(%eax)
+	movl	160(%esp),%ecx
+	movl	%esi,44(%eax)
+	movl	%edx,52(%eax)
+	movl	%edi,60(%eax)
+	movl	%ebp,(%eax)
+	leal	64(%eax),%eax
+	subl	$64,%ecx
+	jnz	L003outer_loop
+	jmp	L006done
+L005tail:
+	addl	112(%esp),%edx
+	addl	120(%esp),%edi
+	movl	%eax,(%esp)
+	movl	%ebp,16(%esp)
+	movl	%ecx,32(%esp)
+	movl	%esi,36(%esp)
+	movl	%edx,48(%esp)
+	movl	%edi,56(%esp)
+	movl	4(%esp),%ebp
+	movl	8(%esp),%ecx
+	movl	12(%esp),%esi
+	movl	20(%esp),%edx
+	movl	24(%esp),%edi
+	addl	$857760878,%ebp
+	addl	$2036477234,%ecx
+	addl	$1797285236,%esi
+	addl	84(%esp),%edx
+	addl	88(%esp),%edi
+	movl	%ebp,4(%esp)
+	movl	%ecx,8(%esp)
+	movl	%esi,12(%esp)
+	movl	%edx,20(%esp)
+	movl	%edi,24(%esp)
+	movl	28(%esp),%ebp
+	movl	40(%esp),%ecx
+	movl	44(%esp),%esi
+	movl	52(%esp),%edx
+	movl	60(%esp),%edi
+	addl	92(%esp),%ebp
+	addl	104(%esp),%ecx
+	addl	108(%esp),%esi
+	addl	116(%esp),%edx
+	addl	124(%esp),%edi
+	movl	%ebp,28(%esp)
+	movl	156(%esp),%ebp
+	movl	%ecx,40(%esp)
+	movl	152(%esp),%ecx
+	movl	%esi,44(%esp)
+	xorl	%esi,%esi
+	movl	%edx,52(%esp)
+	movl	%edi,60(%esp)
+	xorl	%eax,%eax
+	xorl	%edx,%edx
+L007tail_loop:
+	movb	(%esi,%ebp,1),%al
+	movb	(%esp,%esi,1),%dl
+	leal	1(%esi),%esi
+	xorb	%dl,%al
+	movb	%al,-1(%ecx,%esi,1)
+	decl	%ebx
+	jnz	L007tail_loop
+L006done:
+	addl	$132,%esp
+L000no_data:
+	popl	%edi
+	popl	%esi
+	popl	%ebx
+	popl	%ebp
+	ret
+.globl	_ChaCha20_ssse3
+.private_extern	_ChaCha20_ssse3
+.align	4
+_ChaCha20_ssse3:
+L_ChaCha20_ssse3_begin:
+	pushl	%ebp
+	pushl	%ebx
+	pushl	%esi
+	pushl	%edi
+Lssse3_shortcut:
+	movl	20(%esp),%edi
+	movl	24(%esp),%esi
+	movl	28(%esp),%ecx
+	movl	32(%esp),%edx
+	movl	36(%esp),%ebx
+	movl	%esp,%ebp
+	subl	$524,%esp
+	andl	$-64,%esp
+	movl	%ebp,512(%esp)
+	leal	Lssse3_data-Lpic_point(%eax),%eax
+	movdqu	(%ebx),%xmm3
+	cmpl	$256,%ecx
+	jb	L0081x
+	movl	%edx,516(%esp)
+	movl	%ebx,520(%esp)
+	subl	$256,%ecx
+	leal	384(%esp),%ebp
+	movdqu	(%edx),%xmm7
+	pshufd	$0,%xmm3,%xmm0
+	pshufd	$85,%xmm3,%xmm1
+	pshufd	$170,%xmm3,%xmm2
+	pshufd	$255,%xmm3,%xmm3
+	paddd	48(%eax),%xmm0
+	pshufd	$0,%xmm7,%xmm4
+	pshufd	$85,%xmm7,%xmm5
+	psubd	64(%eax),%xmm0
+	pshufd	$170,%xmm7,%xmm6
+	pshufd	$255,%xmm7,%xmm7
+	movdqa	%xmm0,64(%ebp)
+	movdqa	%xmm1,80(%ebp)
+	movdqa	%xmm2,96(%ebp)
+	movdqa	%xmm3,112(%ebp)
+	movdqu	16(%edx),%xmm3
+	movdqa	%xmm4,-64(%ebp)
+	movdqa	%xmm5,-48(%ebp)
+	movdqa	%xmm6,-32(%ebp)
+	movdqa	%xmm7,-16(%ebp)
+	movdqa	32(%eax),%xmm7
+	leal	128(%esp),%ebx
+	pshufd	$0,%xmm3,%xmm0
+	pshufd	$85,%xmm3,%xmm1
+	pshufd	$170,%xmm3,%xmm2
+	pshufd	$255,%xmm3,%xmm3
+	pshufd	$0,%xmm7,%xmm4
+	pshufd	$85,%xmm7,%xmm5
+	pshufd	$170,%xmm7,%xmm6
+	pshufd	$255,%xmm7,%xmm7
+	movdqa	%xmm0,(%ebp)
+	movdqa	%xmm1,16(%ebp)
+	movdqa	%xmm2,32(%ebp)
+	movdqa	%xmm3,48(%ebp)
+	movdqa	%xmm4,-128(%ebp)
+	movdqa	%xmm5,-112(%ebp)
+	movdqa	%xmm6,-96(%ebp)
+	movdqa	%xmm7,-80(%ebp)
+	leal	128(%esi),%esi
+	leal	128(%edi),%edi
+	jmp	L009outer_loop
+.align	4,0x90
+L009outer_loop:
+	movdqa	-112(%ebp),%xmm1
+	movdqa	-96(%ebp),%xmm2
+	movdqa	-80(%ebp),%xmm3
+	movdqa	-48(%ebp),%xmm5
+	movdqa	-32(%ebp),%xmm6
+	movdqa	-16(%ebp),%xmm7
+	movdqa	%xmm1,-112(%ebx)
+	movdqa	%xmm2,-96(%ebx)
+	movdqa	%xmm3,-80(%ebx)
+	movdqa	%xmm5,-48(%ebx)
+	movdqa	%xmm6,-32(%ebx)
+	movdqa	%xmm7,-16(%ebx)
+	movdqa	32(%ebp),%xmm2
+	movdqa	48(%ebp),%xmm3
+	movdqa	64(%ebp),%xmm4
+	movdqa	80(%ebp),%xmm5
+	movdqa	96(%ebp),%xmm6
+	movdqa	112(%ebp),%xmm7
+	paddd	64(%eax),%xmm4
+	movdqa	%xmm2,32(%ebx)
+	movdqa	%xmm3,48(%ebx)
+	movdqa	%xmm4,64(%ebx)
+	movdqa	%xmm5,80(%ebx)
+	movdqa	%xmm6,96(%ebx)
+	movdqa	%xmm7,112(%ebx)
+	movdqa	%xmm4,64(%ebp)
+	movdqa	-128(%ebp),%xmm0
+	movdqa	%xmm4,%xmm6
+	movdqa	-64(%ebp),%xmm3
+	movdqa	(%ebp),%xmm4
+	movdqa	16(%ebp),%xmm5
+	movl	$10,%edx
+	nop
+.align	4,0x90
+L010loop:
+	paddd	%xmm3,%xmm0
+	movdqa	%xmm3,%xmm2
+	pxor	%xmm0,%xmm6
+	pshufb	(%eax),%xmm6
+	paddd	%xmm6,%xmm4
+	pxor	%xmm4,%xmm2
+	movdqa	-48(%ebx),%xmm3
+	movdqa	%xmm2,%xmm1
+	pslld	$12,%xmm2
+	psrld	$20,%xmm1
+	por	%xmm1,%xmm2
+	movdqa	-112(%ebx),%xmm1
+	paddd	%xmm2,%xmm0
+	movdqa	80(%ebx),%xmm7
+	pxor	%xmm0,%xmm6
+	movdqa	%xmm0,-128(%ebx)
+	pshufb	16(%eax),%xmm6
+	paddd	%xmm6,%xmm4
+	movdqa	%xmm6,64(%ebx)
+	pxor	%xmm4,%xmm2
+	paddd	%xmm3,%xmm1
+	movdqa	%xmm2,%xmm0
+	pslld	$7,%xmm2
+	psrld	$25,%xmm0
+	pxor	%xmm1,%xmm7
+	por	%xmm0,%xmm2
+	movdqa	%xmm4,(%ebx)
+	pshufb	(%eax),%xmm7
+	movdqa	%xmm2,-64(%ebx)
+	paddd	%xmm7,%xmm5
+	movdqa	32(%ebx),%xmm4
+	pxor	%xmm5,%xmm3
+	movdqa	-32(%ebx),%xmm2
+	movdqa	%xmm3,%xmm0
+	pslld	$12,%xmm3
+	psrld	$20,%xmm0
+	por	%xmm0,%xmm3
+	movdqa	-96(%ebx),%xmm0
+	paddd	%xmm3,%xmm1
+	movdqa	96(%ebx),%xmm6
+	pxor	%xmm1,%xmm7
+	movdqa	%xmm1,-112(%ebx)
+	pshufb	16(%eax),%xmm7
+	paddd	%xmm7,%xmm5
+	movdqa	%xmm7,80(%ebx)
+	pxor	%xmm5,%xmm3
+	paddd	%xmm2,%xmm0
+	movdqa	%xmm3,%xmm1
+	pslld	$7,%xmm3
+	psrld	$25,%xmm1
+	pxor	%xmm0,%xmm6
+	por	%xmm1,%xmm3
+	movdqa	%xmm5,16(%ebx)
+	pshufb	(%eax),%xmm6
+	movdqa	%xmm3,-48(%ebx)
+	paddd	%xmm6,%xmm4
+	movdqa	48(%ebx),%xmm5
+	pxor	%xmm4,%xmm2
+	movdqa	-16(%ebx),%xmm3
+	movdqa	%xmm2,%xmm1
+	pslld	$12,%xmm2
+	psrld	$20,%xmm1
+	por	%xmm1,%xmm2
+	movdqa	-80(%ebx),%xmm1
+	paddd	%xmm2,%xmm0
+	movdqa	112(%ebx),%xmm7
+	pxor	%xmm0,%xmm6
+	movdqa	%xmm0,-96(%ebx)
+	pshufb	16(%eax),%xmm6
+	paddd	%xmm6,%xmm4
+	movdqa	%xmm6,96(%ebx)
+	pxor	%xmm4,%xmm2
+	paddd	%xmm3,%xmm1
+	movdqa	%xmm2,%xmm0
+	pslld	$7,%xmm2
+	psrld	$25,%xmm0
+	pxor	%xmm1,%xmm7
+	por	%xmm0,%xmm2
+	pshufb	(%eax),%xmm7
+	movdqa	%xmm2,-32(%ebx)
+	paddd	%xmm7,%xmm5
+	pxor	%xmm5,%xmm3
+	movdqa	-48(%ebx),%xmm2
+	movdqa	%xmm3,%xmm0
+	pslld	$12,%xmm3
+	psrld	$20,%xmm0
+	por	%xmm0,%xmm3
+	movdqa	-128(%ebx),%xmm0
+	paddd	%xmm3,%xmm1
+	pxor	%xmm1,%xmm7
+	movdqa	%xmm1,-80(%ebx)
+	pshufb	16(%eax),%xmm7
+	paddd	%xmm7,%xmm5
+	movdqa	%xmm7,%xmm6
+	pxor	%xmm5,%xmm3
+	paddd	%xmm2,%xmm0
+	movdqa	%xmm3,%xmm1
+	pslld	$7,%xmm3
+	psrld	$25,%xmm1
+	pxor	%xmm0,%xmm6
+	por	%xmm1,%xmm3
+	pshufb	(%eax),%xmm6
+	movdqa	%xmm3,-16(%ebx)
+	paddd	%xmm6,%xmm4
+	pxor	%xmm4,%xmm2
+	movdqa	-32(%ebx),%xmm3
+	movdqa	%xmm2,%xmm1
+	pslld	$12,%xmm2
+	psrld	$20,%xmm1
+	por	%xmm1,%xmm2
+	movdqa	-112(%ebx),%xmm1
+	paddd	%xmm2,%xmm0
+	movdqa	64(%ebx),%xmm7
+	pxor	%xmm0,%xmm6
+	movdqa	%xmm0,-128(%ebx)
+	pshufb	16(%eax),%xmm6
+	paddd	%xmm6,%xmm4
+	movdqa	%xmm6,112(%ebx)
+	pxor	%xmm4,%xmm2
+	paddd	%xmm3,%xmm1
+	movdqa	%xmm2,%xmm0
+	pslld	$7,%xmm2
+	psrld	$25,%xmm0
+	pxor	%xmm1,%xmm7
+	por	%xmm0,%xmm2
+	movdqa	%xmm4,32(%ebx)
+	pshufb	(%eax),%xmm7
+	movdqa	%xmm2,-48(%ebx)
+	paddd	%xmm7,%xmm5
+	movdqa	(%ebx),%xmm4
+	pxor	%xmm5,%xmm3
+	movdqa	-16(%ebx),%xmm2
+	movdqa	%xmm3,%xmm0
+	pslld	$12,%xmm3
+	psrld	$20,%xmm0
+	por	%xmm0,%xmm3
+	movdqa	-96(%ebx),%xmm0
+	paddd	%xmm3,%xmm1
+	movdqa	80(%ebx),%xmm6
+	pxor	%xmm1,%xmm7
+	movdqa	%xmm1,-112(%ebx)
+	pshufb	16(%eax),%xmm7
+	paddd	%xmm7,%xmm5
+	movdqa	%xmm7,64(%ebx)
+	pxor	%xmm5,%xmm3
+	paddd	%xmm2,%xmm0
+	movdqa	%xmm3,%xmm1
+	pslld	$7,%xmm3
+	psrld	$25,%xmm1
+	pxor	%xmm0,%xmm6
+	por	%xmm1,%xmm3
+	movdqa	%xmm5,48(%ebx)
+	pshufb	(%eax),%xmm6
+	movdqa	%xmm3,-32(%ebx)
+	paddd	%xmm6,%xmm4
+	movdqa	16(%ebx),%xmm5
+	pxor	%xmm4,%xmm2
+	movdqa	-64(%ebx),%xmm3
+	movdqa	%xmm2,%xmm1
+	pslld	$12,%xmm2
+	psrld	$20,%xmm1
+	por	%xmm1,%xmm2
+	movdqa	-80(%ebx),%xmm1
+	paddd	%xmm2,%xmm0
+	movdqa	96(%ebx),%xmm7
+	pxor	%xmm0,%xmm6
+	movdqa	%xmm0,-96(%ebx)
+	pshufb	16(%eax),%xmm6
+	paddd	%xmm6,%xmm4
+	movdqa	%xmm6,80(%ebx)
+	pxor	%xmm4,%xmm2
+	paddd	%xmm3,%xmm1
+	movdqa	%xmm2,%xmm0
+	pslld	$7,%xmm2
+	psrld	$25,%xmm0
+	pxor	%xmm1,%xmm7
+	por	%xmm0,%xmm2
+	pshufb	(%eax),%xmm7
+	movdqa	%xmm2,-16(%ebx)
+	paddd	%xmm7,%xmm5
+	pxor	%xmm5,%xmm3
+	movdqa	%xmm3,%xmm0
+	pslld	$12,%xmm3
+	psrld	$20,%xmm0
+	por	%xmm0,%xmm3
+	movdqa	-128(%ebx),%xmm0
+	paddd	%xmm3,%xmm1
+	movdqa	64(%ebx),%xmm6
+	pxor	%xmm1,%xmm7
+	movdqa	%xmm1,-80(%ebx)
+	pshufb	16(%eax),%xmm7
+	paddd	%xmm7,%xmm5
+	movdqa	%xmm7,96(%ebx)
+	pxor	%xmm5,%xmm3
+	movdqa	%xmm3,%xmm1
+	pslld	$7,%xmm3
+	psrld	$25,%xmm1
+	por	%xmm1,%xmm3
+	decl	%edx
+	jnz	L010loop
+	movdqa	%xmm3,-64(%ebx)
+	movdqa	%xmm4,(%ebx)
+	movdqa	%xmm5,16(%ebx)
+	movdqa	%xmm6,64(%ebx)
+	movdqa	%xmm7,96(%ebx)
+	movdqa	-112(%ebx),%xmm1
+	movdqa	-96(%ebx),%xmm2
+	movdqa	-80(%ebx),%xmm3
+	paddd	-128(%ebp),%xmm0
+	paddd	-112(%ebp),%xmm1
+	paddd	-96(%ebp),%xmm2
+	paddd	-80(%ebp),%xmm3
+	movdqa	%xmm0,%xmm6
+	punpckldq	%xmm1,%xmm0
+	movdqa	%xmm2,%xmm7
+	punpckldq	%xmm3,%xmm2
+	punpckhdq	%xmm1,%xmm6
+	punpckhdq	%xmm3,%xmm7
+	movdqa	%xmm0,%xmm1
+	punpcklqdq	%xmm2,%xmm0
+	movdqa	%xmm6,%xmm3
+	punpcklqdq	%xmm7,%xmm6
+	punpckhqdq	%xmm2,%xmm1
+	punpckhqdq	%xmm7,%xmm3
+	movdqu	-128(%esi),%xmm4
+	movdqu	-64(%esi),%xmm5
+	movdqu	(%esi),%xmm2
+	movdqu	64(%esi),%xmm7
+	leal	16(%esi),%esi
+	pxor	%xmm0,%xmm4
+	movdqa	-64(%ebx),%xmm0
+	pxor	%xmm1,%xmm5
+	movdqa	-48(%ebx),%xmm1
+	pxor	%xmm2,%xmm6
+	movdqa	-32(%ebx),%xmm2
+	pxor	%xmm3,%xmm7
+	movdqa	-16(%ebx),%xmm3
+	movdqu	%xmm4,-128(%edi)
+	movdqu	%xmm5,-64(%edi)
+	movdqu	%xmm6,(%edi)
+	movdqu	%xmm7,64(%edi)
+	leal	16(%edi),%edi
+	paddd	-64(%ebp),%xmm0
+	paddd	-48(%ebp),%xmm1
+	paddd	-32(%ebp),%xmm2
+	paddd	-16(%ebp),%xmm3
+	movdqa	%xmm0,%xmm6
+	punpckldq	%xmm1,%xmm0
+	movdqa	%xmm2,%xmm7
+	punpckldq	%xmm3,%xmm2
+	punpckhdq	%xmm1,%xmm6
+	punpckhdq	%xmm3,%xmm7
+	movdqa	%xmm0,%xmm1
+	punpcklqdq	%xmm2,%xmm0
+	movdqa	%xmm6,%xmm3
+	punpcklqdq	%xmm7,%xmm6
+	punpckhqdq	%xmm2,%xmm1
+	punpckhqdq	%xmm7,%xmm3
+	movdqu	-128(%esi),%xmm4
+	movdqu	-64(%esi),%xmm5
+	movdqu	(%esi),%xmm2
+	movdqu	64(%esi),%xmm7
+	leal	16(%esi),%esi
+	pxor	%xmm0,%xmm4
+	movdqa	(%ebx),%xmm0
+	pxor	%xmm1,%xmm5
+	movdqa	16(%ebx),%xmm1
+	pxor	%xmm2,%xmm6
+	movdqa	32(%ebx),%xmm2
+	pxor	%xmm3,%xmm7
+	movdqa	48(%ebx),%xmm3
+	movdqu	%xmm4,-128(%edi)
+	movdqu	%xmm5,-64(%edi)
+	movdqu	%xmm6,(%edi)
+	movdqu	%xmm7,64(%edi)
+	leal	16(%edi),%edi
+	paddd	(%ebp),%xmm0
+	paddd	16(%ebp),%xmm1
+	paddd	32(%ebp),%xmm2
+	paddd	48(%ebp),%xmm3
+	movdqa	%xmm0,%xmm6
+	punpckldq	%xmm1,%xmm0
+	movdqa	%xmm2,%xmm7
+	punpckldq	%xmm3,%xmm2
+	punpckhdq	%xmm1,%xmm6
+	punpckhdq	%xmm3,%xmm7
+	movdqa	%xmm0,%xmm1
+	punpcklqdq	%xmm2,%xmm0
+	movdqa	%xmm6,%xmm3
+	punpcklqdq	%xmm7,%xmm6
+	punpckhqdq	%xmm2,%xmm1
+	punpckhqdq	%xmm7,%xmm3
+	movdqu	-128(%esi),%xmm4
+	movdqu	-64(%esi),%xmm5
+	movdqu	(%esi),%xmm2
+	movdqu	64(%esi),%xmm7
+	leal	16(%esi),%esi
+	pxor	%xmm0,%xmm4
+	movdqa	64(%ebx),%xmm0
+	pxor	%xmm1,%xmm5
+	movdqa	80(%ebx),%xmm1
+	pxor	%xmm2,%xmm6
+	movdqa	96(%ebx),%xmm2
+	pxor	%xmm3,%xmm7
+	movdqa	112(%ebx),%xmm3
+	movdqu	%xmm4,-128(%edi)
+	movdqu	%xmm5,-64(%edi)
+	movdqu	%xmm6,(%edi)
+	movdqu	%xmm7,64(%edi)
+	leal	16(%edi),%edi
+	paddd	64(%ebp),%xmm0
+	paddd	80(%ebp),%xmm1
+	paddd	96(%ebp),%xmm2
+	paddd	112(%ebp),%xmm3
+	movdqa	%xmm0,%xmm6
+	punpckldq	%xmm1,%xmm0
+	movdqa	%xmm2,%xmm7
+	punpckldq	%xmm3,%xmm2
+	punpckhdq	%xmm1,%xmm6
+	punpckhdq	%xmm3,%xmm7
+	movdqa	%xmm0,%xmm1
+	punpcklqdq	%xmm2,%xmm0
+	movdqa	%xmm6,%xmm3
+	punpcklqdq	%xmm7,%xmm6
+	punpckhqdq	%xmm2,%xmm1
+	punpckhqdq	%xmm7,%xmm3
+	movdqu	-128(%esi),%xmm4
+	movdqu	-64(%esi),%xmm5
+	movdqu	(%esi),%xmm2
+	movdqu	64(%esi),%xmm7
+	leal	208(%esi),%esi
+	pxor	%xmm0,%xmm4
+	pxor	%xmm1,%xmm5
+	pxor	%xmm2,%xmm6
+	pxor	%xmm3,%xmm7
+	movdqu	%xmm4,-128(%edi)
+	movdqu	%xmm5,-64(%edi)
+	movdqu	%xmm6,(%edi)
+	movdqu	%xmm7,64(%edi)
+	leal	208(%edi),%edi
+	subl	$256,%ecx
+	jnc	L009outer_loop
+	addl	$256,%ecx
+	jz	L011done
+	movl	520(%esp),%ebx
+	leal	-128(%esi),%esi
+	movl	516(%esp),%edx
+	leal	-128(%edi),%edi
+	movd	64(%ebp),%xmm2
+	movdqu	(%ebx),%xmm3
+	paddd	96(%eax),%xmm2
+	pand	112(%eax),%xmm3
+	por	%xmm2,%xmm3
+L0081x:
+	movdqa	32(%eax),%xmm0
+	movdqu	(%edx),%xmm1
+	movdqu	16(%edx),%xmm2
+	movdqa	(%eax),%xmm6
+	movdqa	16(%eax),%xmm7
+	movl	%ebp,48(%esp)
+	movdqa	%xmm0,(%esp)
+	movdqa	%xmm1,16(%esp)
+	movdqa	%xmm2,32(%esp)
+	movdqa	%xmm3,48(%esp)
+	movl	$10,%edx
+	jmp	L012loop1x
+.align	4,0x90
+L013outer1x:
+	movdqa	80(%eax),%xmm3
+	movdqa	(%esp),%xmm0
+	movdqa	16(%esp),%xmm1
+	movdqa	32(%esp),%xmm2
+	paddd	48(%esp),%xmm3
+	movl	$10,%edx
+	movdqa	%xmm3,48(%esp)
+	jmp	L012loop1x
+.align	4,0x90
+L012loop1x:
+	paddd	%xmm1,%xmm0
+	pxor	%xmm0,%xmm3
+.byte	102,15,56,0,222
+	paddd	%xmm3,%xmm2
+	pxor	%xmm2,%xmm1
+	movdqa	%xmm1,%xmm4
+	psrld	$20,%xmm1
+	pslld	$12,%xmm4
+	por	%xmm4,%xmm1
+	paddd	%xmm1,%xmm0
+	pxor	%xmm0,%xmm3
+.byte	102,15,56,0,223
+	paddd	%xmm3,%xmm2
+	pxor	%xmm2,%xmm1
+	movdqa	%xmm1,%xmm4
+	psrld	$25,%xmm1
+	pslld	$7,%xmm4
+	por	%xmm4,%xmm1
+	pshufd	$78,%xmm2,%xmm2
+	pshufd	$57,%xmm1,%xmm1
+	pshufd	$147,%xmm3,%xmm3
+	nop
+	paddd	%xmm1,%xmm0
+	pxor	%xmm0,%xmm3
+.byte	102,15,56,0,222
+	paddd	%xmm3,%xmm2
+	pxor	%xmm2,%xmm1
+	movdqa	%xmm1,%xmm4
+	psrld	$20,%xmm1
+	pslld	$12,%xmm4
+	por	%xmm4,%xmm1
+	paddd	%xmm1,%xmm0
+	pxor	%xmm0,%xmm3
+.byte	102,15,56,0,223
+	paddd	%xmm3,%xmm2
+	pxor	%xmm2,%xmm1
+	movdqa	%xmm1,%xmm4
+	psrld	$25,%xmm1
+	pslld	$7,%xmm4
+	por	%xmm4,%xmm1
+	pshufd	$78,%xmm2,%xmm2
+	pshufd	$147,%xmm1,%xmm1
+	pshufd	$57,%xmm3,%xmm3
+	decl	%edx
+	jnz	L012loop1x
+	paddd	(%esp),%xmm0
+	paddd	16(%esp),%xmm1
+	paddd	32(%esp),%xmm2
+	paddd	48(%esp),%xmm3
+	cmpl	$64,%ecx
+	jb	L014tail
+	movdqu	(%esi),%xmm4
+	movdqu	16(%esi),%xmm5
+	pxor	%xmm4,%xmm0
+	movdqu	32(%esi),%xmm4
+	pxor	%xmm5,%xmm1
+	movdqu	48(%esi),%xmm5
+	pxor	%xmm4,%xmm2
+	pxor	%xmm5,%xmm3
+	leal	64(%esi),%esi
+	movdqu	%xmm0,(%edi)
+	movdqu	%xmm1,16(%edi)
+	movdqu	%xmm2,32(%edi)
+	movdqu	%xmm3,48(%edi)
+	leal	64(%edi),%edi
+	subl	$64,%ecx
+	jnz	L013outer1x
+	jmp	L011done
+L014tail:
+	movdqa	%xmm0,(%esp)
+	movdqa	%xmm1,16(%esp)
+	movdqa	%xmm2,32(%esp)
+	movdqa	%xmm3,48(%esp)
+	xorl	%eax,%eax
+	xorl	%edx,%edx
+	xorl	%ebp,%ebp
+L015tail_loop:
+	movb	(%esp,%ebp,1),%al
+	movb	(%esi,%ebp,1),%dl
+	leal	1(%ebp),%ebp
+	xorb	%dl,%al
+	movb	%al,-1(%edi,%ebp,1)
+	decl	%ecx
+	jnz	L015tail_loop
+L011done:
+	movl	512(%esp),%esp
+	popl	%edi
+	popl	%esi
+	popl	%ebx
+	popl	%ebp
+	ret
+.align	6,0x90
+Lssse3_data:
+.byte	2,3,0,1,6,7,4,5,10,11,8,9,14,15,12,13
+.byte	3,0,1,2,7,4,5,6,11,8,9,10,15,12,13,14
+.long	1634760805,857760878,2036477234,1797285236
+.long	0,1,2,3
+.long	4,4,4,4
+.long	1,0,0,0
+.long	4,0,0,0
+.long	0,-1,-1,-1
+.align	6,0x90
+.byte	67,104,97,67,104,97,50,48,32,102,111,114,32,120,56,54
+.byte	44,32,67,82,89,80,84,79,71,65,77,83,32,98,121,32
+.byte	60,97,112,112,114,111,64,111,112,101,110,115,115,108,46,111
+.byte	114,103,62,0
+.section __IMPORT,__pointers,non_lazy_symbol_pointers
+L_OPENSSL_ia32cap_P$non_lazy_ptr:
+.indirect_symbol	_OPENSSL_ia32cap_P
+.long	0
+#endif

third_party/boringssl/mac-x86/crypto/cpu-x86-asm.S

diff --git a/third_party/boringssl/mac-x86/crypto/cpu-x86-asm.S b/third_party/boringssl/mac-x86/crypto/cpu-x86-asm.S
deleted file mode 100644
index bfb292c..0000000
--- a/third_party/boringssl/mac-x86/crypto/cpu-x86-asm.S
+++ /dev/null
@@ -1,309 +0,0 @@
-#if defined(__i386__)
-.file	"crypto/cpu-x86-asm.S"
-.text
-.globl	_OPENSSL_ia32_cpuid
-.private_extern	_OPENSSL_ia32_cpuid
-.align	4
-_OPENSSL_ia32_cpuid:
-L_OPENSSL_ia32_cpuid_begin:
-	pushl	%ebp
-	pushl	%ebx
-	pushl	%esi
-	pushl	%edi
-	xorl	%edx,%edx
-	pushfl
-	popl	%eax
-	movl	%eax,%ecx
-	xorl	$2097152,%eax
-	pushl	%eax
-	popfl
-	pushfl
-	popl	%eax
-	xorl	%eax,%ecx
-	xorl	%eax,%eax
-	btl	$21,%ecx
-	jnc	L000nocpuid
-	movl	20(%esp),%esi
-	movl	%eax,8(%esi)
-	.byte	0x0f,0xa2
-	movl	%eax,%edi
-	xorl	%eax,%eax
-	cmpl	$1970169159,%ebx
-	setne	%al
-	movl	%eax,%ebp
-	cmpl	$1231384169,%edx
-	setne	%al
-	orl	%eax,%ebp
-	cmpl	$1818588270,%ecx
-	setne	%al
-	orl	%eax,%ebp
-	jz	L001intel
-	cmpl	$1752462657,%ebx
-	setne	%al
-	movl	%eax,%esi
-	cmpl	$1769238117,%edx
-	setne	%al
-	orl	%eax,%esi
-	cmpl	$1145913699,%ecx
-	setne	%al
-	orl	%eax,%esi
-	jnz	L001intel
-	movl	$2147483648,%eax
-	.byte	0x0f,0xa2
-	cmpl	$2147483649,%eax
-	jb	L001intel
-	movl	%eax,%esi
-	movl	$2147483649,%eax
-	.byte	0x0f,0xa2
-	orl	%ecx,%ebp
-	andl	$2049,%ebp
-	cmpl	$2147483656,%esi
-	jb	L001intel
-	movl	$2147483656,%eax
-	.byte	0x0f,0xa2
-	movzbl	%cl,%esi
-	incl	%esi
-	movl	$1,%eax
-	xorl	%ecx,%ecx
-	.byte	0x0f,0xa2
-	btl	$28,%edx
-	jnc	L002generic
-	shrl	$16,%ebx
-	andl	$255,%ebx
-	cmpl	%esi,%ebx
-	ja	L002generic
-	andl	$4026531839,%edx
-	jmp	L002generic
-L001intel:
-	cmpl	$7,%edi
-	jb	L003cacheinfo
-	movl	20(%esp),%esi
-	movl	$7,%eax
-	xorl	%ecx,%ecx
-	.byte	0x0f,0xa2
-	movl	%ebx,8(%esi)
-L003cacheinfo:
-	cmpl	$4,%edi
-	movl	$-1,%edi
-	jb	L004nocacheinfo
-	movl	$4,%eax
-	movl	$0,%ecx
-	.byte	0x0f,0xa2
-	movl	%eax,%edi
-	shrl	$14,%edi
-	andl	$4095,%edi
-L004nocacheinfo:
-	movl	$1,%eax
-	xorl	%ecx,%ecx
-	.byte	0x0f,0xa2
-	andl	$3220176895,%edx
-	cmpl	$0,%ebp
-	jne	L005notintel
-	orl	$1073741824,%edx
-L005notintel:
-	btl	$28,%edx
-	jnc	L002generic
-	andl	$4026531839,%edx
-	cmpl	$0,%edi
-	je	L002generic
-	orl	$268435456,%edx
-	shrl	$16,%ebx
-	cmpb	$1,%bl
-	ja	L002generic
-	andl	$4026531839,%edx
-L002generic:
-	andl	$2048,%ebp
-	andl	$4294965247,%ecx
-	movl	%edx,%esi
-	orl	%ecx,%ebp
-	btl	$27,%ecx
-	jnc	L006clear_avx
-	xorl	%ecx,%ecx
-.byte	15,1,208
-	andl	$6,%eax
-	cmpl	$6,%eax
-	je	L007done
-	cmpl	$2,%eax
-	je	L006clear_avx
-L008clear_xmm:
-	andl	$4261412861,%ebp
-	andl	$4278190079,%esi
-L006clear_avx:
-	andl	$4026525695,%ebp
-	movl	20(%esp),%edi
-	andl	$4294967263,8(%edi)
-L007done:
-	movl	%esi,%eax
-	movl	%ebp,%edx
-L000nocpuid:
-	popl	%edi
-	popl	%esi
-	popl	%ebx
-	popl	%ebp
-	ret
-.globl	_OPENSSL_rdtsc
-.private_extern	_OPENSSL_rdtsc
-.align	4
-_OPENSSL_rdtsc:
-L_OPENSSL_rdtsc_begin:
-	xorl	%eax,%eax
-	xorl	%edx,%edx
-	call	L009PIC_me_up
-L009PIC_me_up:
-	popl	%ecx
-	movl	L_OPENSSL_ia32cap_P$non_lazy_ptr-L009PIC_me_up(%ecx),%ecx
-	btl	$4,(%ecx)
-	jnc	L010notsc
-	.byte	0x0f,0x31
-L010notsc:
-	ret
-.globl	_OPENSSL_instrument_halt
-.private_extern	_OPENSSL_instrument_halt
-.align	4
-_OPENSSL_instrument_halt:
-L_OPENSSL_instrument_halt_begin:
-	call	L011PIC_me_up
-L011PIC_me_up:
-	popl	%ecx
-	movl	L_OPENSSL_ia32cap_P$non_lazy_ptr-L011PIC_me_up(%ecx),%ecx
-	btl	$4,(%ecx)
-	jnc	L012nohalt
-.long	2421723150
-	andl	$3,%eax
-	jnz	L012nohalt
-	pushfl
-	popl	%eax
-	btl	$9,%eax
-	jnc	L012nohalt
-	.byte	0x0f,0x31
-	pushl	%edx
-	pushl	%eax
-	hlt
-	.byte	0x0f,0x31
-	subl	(%esp),%eax
-	sbbl	4(%esp),%edx
-	addl	$8,%esp
-	ret
-L012nohalt:
-	xorl	%eax,%eax
-	xorl	%edx,%edx
-	ret
-.globl	_OPENSSL_far_spin
-.private_extern	_OPENSSL_far_spin
-.align	4
-_OPENSSL_far_spin:
-L_OPENSSL_far_spin_begin:
-	pushfl
-	popl	%eax
-	btl	$9,%eax
-	jnc	L013nospin
-	movl	4(%esp),%eax
-	movl	8(%esp),%ecx
-.long	2430111262
-	xorl	%eax,%eax
-	movl	(%ecx),%edx
-	jmp	L014spin
-.align	4,0x90
-L014spin:
-	incl	%eax
-	cmpl	(%ecx),%edx
-	je	L014spin
-.long	529567888
-	ret
-L013nospin:
-	xorl	%eax,%eax
-	xorl	%edx,%edx
-	ret
-.globl	_OPENSSL_wipe_cpu
-.private_extern	_OPENSSL_wipe_cpu
-.align	4
-_OPENSSL_wipe_cpu:
-L_OPENSSL_wipe_cpu_begin:
-	xorl	%eax,%eax
-	xorl	%edx,%edx
-	call	L015PIC_me_up
-L015PIC_me_up:
-	popl	%ecx
-	movl	L_OPENSSL_ia32cap_P$non_lazy_ptr-L015PIC_me_up(%ecx),%ecx
-	movl	(%ecx),%ecx
-	btl	$1,(%ecx)
-	jnc	L016no_x87
-	andl	$83886080,%ecx
-	cmpl	$83886080,%ecx
-	jne	L017no_sse2
-	pxor	%xmm0,%xmm0
-	pxor	%xmm1,%xmm1
-	pxor	%xmm2,%xmm2
-	pxor	%xmm3,%xmm3
-	pxor	%xmm4,%xmm4
-	pxor	%xmm5,%xmm5
-	pxor	%xmm6,%xmm6
-	pxor	%xmm7,%xmm7
-L017no_sse2:
-.long	4007259865,4007259865,4007259865,4007259865,2430851995
-L016no_x87:
-	leal	4(%esp),%eax
-	ret
-.globl	_OPENSSL_atomic_add
-.private_extern	_OPENSSL_atomic_add
-.align	4
-_OPENSSL_atomic_add:
-L_OPENSSL_atomic_add_begin:
-	movl	4(%esp),%edx
-	movl	8(%esp),%ecx
-	pushl	%ebx
-	nop
-	movl	(%edx),%eax
-L018spin:
-	leal	(%eax,%ecx,1),%ebx
-	nop
-.long	447811568
-	jne	L018spin
-	movl	%ebx,%eax
-	popl	%ebx
-	ret
-.globl	_OPENSSL_indirect_call
-.private_extern	_OPENSSL_indirect_call
-.align	4
-_OPENSSL_indirect_call:
-L_OPENSSL_indirect_call_begin:
-	pushl	%ebp
-	movl	%esp,%ebp
-	subl	$28,%esp
-	movl	12(%ebp),%ecx
-	movl	%ecx,(%esp)
-	movl	16(%ebp),%edx
-	movl	%edx,4(%esp)
-	movl	20(%ebp),%eax
-	movl	%eax,8(%esp)
-	movl	24(%ebp),%eax
-	movl	%eax,12(%esp)
-	movl	28(%ebp),%eax
-	movl	%eax,16(%esp)
-	movl	32(%ebp),%eax
-	movl	%eax,20(%esp)
-	movl	36(%ebp),%eax
-	movl	%eax,24(%esp)
-	call	*8(%ebp)
-	movl	%ebp,%esp
-	popl	%ebp
-	ret
-.globl	_OPENSSL_ia32_rdrand
-.private_extern	_OPENSSL_ia32_rdrand
-.align	4
-_OPENSSL_ia32_rdrand:
-L_OPENSSL_ia32_rdrand_begin:
-	movl	$8,%ecx
-L019loop:
-.byte	15,199,240
-	jc	L020break
-	loop	L019loop
-L020break:
-	cmpl	$0,%eax
-	cmovel	%ecx,%eax
-	ret
-.section __IMPORT,__pointers,non_lazy_symbol_pointers
-L_OPENSSL_ia32cap_P$non_lazy_ptr:
-.indirect_symbol	_OPENSSL_ia32cap_P
-.long	0
-#endif

third_party/boringssl/mac-x86/crypto/rc4/rc4-586.S

diff --git a/third_party/boringssl/mac-x86/crypto/rc4/rc4-586.S b/third_party/boringssl/mac-x86/crypto/rc4/rc4-586.S
index faecdfa..dcddc58 100644
--- a/third_party/boringssl/mac-x86/crypto/rc4/rc4-586.S
+++ b/third_party/boringssl/mac-x86/crypto/rc4/rc4-586.S
@@ -343,39 +343,6 @@
 	popl	%ebx
 	popl	%ebp
 	ret
-.globl	_RC4_options
-.private_extern	_RC4_options
-.align	4
-_RC4_options:
-L_RC4_options_begin:
-	call	L018pic_point
-L018pic_point:
-	popl	%eax
-	leal	L019opts-L018pic_point(%eax),%eax
-	call	L020PIC_me_up
-L020PIC_me_up:
-	popl	%edx
-	movl	L_OPENSSL_ia32cap_P$non_lazy_ptr-L020PIC_me_up(%edx),%edx
-	movl	(%edx),%edx
-	btl	$20,%edx
-	jc	L0211xchar
-	btl	$26,%edx
-	jnc	L022ret
-	addl	$25,%eax
-	ret
-L0211xchar:
-	addl	$12,%eax
-L022ret:
-	ret
-.align	6,0x90
-L019opts:
-.byte	114,99,52,40,52,120,44,105,110,116,41,0
-.byte	114,99,52,40,49,120,44,99,104,97,114,41,0
-.byte	114,99,52,40,56,120,44,109,109,120,41,0
-.byte	82,67,52,32,102,111,114,32,120,56,54,44,32,67,82,89
-.byte	80,84,79,71,65,77,83,32,98,121,32,60,97,112,112,114
-.byte	111,64,111,112,101,110,115,115,108,46,111,114,103,62,0
-.align	6,0x90
 .section __IMPORT,__pointers,non_lazy_symbol_pointers
 L_OPENSSL_ia32cap_P$non_lazy_ptr:
 .indirect_symbol	_OPENSSL_ia32cap_P

third_party/boringssl/mac-x86/crypto/sha/sha1-586.S

diff --git a/third_party/boringssl/mac-x86/crypto/sha/sha1-586.S b/third_party/boringssl/mac-x86/crypto/sha/sha1-586.S
index 97aafbf..72a7205 100644
--- a/third_party/boringssl/mac-x86/crypto/sha/sha1-586.S
+++ b/third_party/boringssl/mac-x86/crypto/sha/sha1-586.S
@@ -22,8 +22,11 @@
 	movl	8(%esi),%ecx
 	testl	$16777216,%eax
 	jz	L001x86
-	testl	$536870912,%ecx
-	jnz	Lshaext_shortcut
+	andl	$268435456,%edx
+	andl	$1073741824,%eax
+	orl	%edx,%eax
+	cmpl	$1342177280,%eax
+	je	Lavx_shortcut
 	jmp	Lssse3_shortcut
 .align	4,0x90
 L001x86:
@@ -1391,9 +1394,9 @@
 	popl	%ebx
 	popl	%ebp
 	ret
-.private_extern	__sha1_block_data_order_shaext
+.private_extern	__sha1_block_data_order_ssse3
 .align	4
-__sha1_block_data_order_shaext:
+__sha1_block_data_order_ssse3:
 	pushl	%ebp
 	pushl	%ebx
 	pushl	%esi
@@ -1402,175 +1405,6 @@
 L003pic_point:
 	popl	%ebp
 	leal	LK_XX_XX-L003pic_point(%ebp),%ebp
-Lshaext_shortcut:
-	movl	20(%esp),%edi
-	movl	%esp,%ebx
-	movl	24(%esp),%esi
-	movl	28(%esp),%ecx
-	subl	$32,%esp
-	movdqu	(%edi),%xmm0
-	movd	16(%edi),%xmm1
-	andl	$-32,%esp
-	movdqa	80(%ebp),%xmm3
-	movdqu	(%esi),%xmm4
-	pshufd	$27,%xmm0,%xmm0
-	movdqu	16(%esi),%xmm5
-	pshufd	$27,%xmm1,%xmm1
-	movdqu	32(%esi),%xmm6
-.byte	102,15,56,0,227
-	movdqu	48(%esi),%xmm7
-.byte	102,15,56,0,235
-.byte	102,15,56,0,243
-.byte	102,15,56,0,251
-	jmp	L004loop_shaext
-.align	4,0x90
-L004loop_shaext:
-	decl	%ecx
-	leal	64(%esi),%eax
-	movdqa	%xmm1,(%esp)
-	paddd	%xmm4,%xmm1
-	cmovnel	%eax,%esi
-	movdqa	%xmm0,16(%esp)
-.byte	15,56,201,229
-	movdqa	%xmm0,%xmm2
-.byte	15,58,204,193,0
-.byte	15,56,200,213
-	pxor	%xmm6,%xmm4
-.byte	15,56,201,238
-.byte	15,56,202,231
-	movdqa	%xmm0,%xmm1
-.byte	15,58,204,194,0
-.byte	15,56,200,206
-	pxor	%xmm7,%xmm5
-.byte	15,56,202,236
-.byte	15,56,201,247
-	movdqa	%xmm0,%xmm2
-.byte	15,58,204,193,0
-.byte	15,56,200,215
-	pxor	%xmm4,%xmm6
-.byte	15,56,201,252
-.byte	15,56,202,245
-	movdqa	%xmm0,%xmm1
-.byte	15,58,204,194,0
-.byte	15,56,200,204
-	pxor	%xmm5,%xmm7
-.byte	15,56,202,254
-.byte	15,56,201,229
-	movdqa	%xmm0,%xmm2
-.byte	15,58,204,193,0
-.byte	15,56,200,213
-	pxor	%xmm6,%xmm4
-.byte	15,56,201,238
-.byte	15,56,202,231
-	movdqa	%xmm0,%xmm1
-.byte	15,58,204,194,1
-.byte	15,56,200,206
-	pxor	%xmm7,%xmm5
-.byte	15,56,202,236
-.byte	15,56,201,247
-	movdqa	%xmm0,%xmm2
-.byte	15,58,204,193,1
-.byte	15,56,200,215
-	pxor	%xmm4,%xmm6
-.byte	15,56,201,252
-.byte	15,56,202,245
-	movdqa	%xmm0,%xmm1
-.byte	15,58,204,194,1
-.byte	15,56,200,204
-	pxor	%xmm5,%xmm7
-.byte	15,56,202,254
-.byte	15,56,201,229
-	movdqa	%xmm0,%xmm2
-.byte	15,58,204,193,1
-.byte	15,56,200,213
-	pxor	%xmm6,%xmm4
-.byte	15,56,201,238
-.byte	15,56,202,231
-	movdqa	%xmm0,%xmm1
-.byte	15,58,204,194,1
-.byte	15,56,200,206
-	pxor	%xmm7,%xmm5
-.byte	15,56,202,236
-.byte	15,56,201,247
-	movdqa	%xmm0,%xmm2
-.byte	15,58,204,193,2
-.byte	15,56,200,215
-	pxor	%xmm4,%xmm6
-.byte	15,56,201,252
-.byte	15,56,202,245
-	movdqa	%xmm0,%xmm1
-.byte	15,58,204,194,2
-.byte	15,56,200,204
-	pxor	%xmm5,%xmm7
-.byte	15,56,202,254
-.byte	15,56,201,229
-	movdqa	%xmm0,%xmm2
-.byte	15,58,204,193,2
-.byte	15,56,200,213
-	pxor	%xmm6,%xmm4
-.byte	15,56,201,238
-.byte	15,56,202,231
-	movdqa	%xmm0,%xmm1
-.byte	15,58,204,194,2
-.byte	15,56,200,206
-	pxor	%xmm7,%xmm5
-.byte	15,56,202,236
-.byte	15,56,201,247
-	movdqa	%xmm0,%xmm2
-.byte	15,58,204,193,2
-.byte	15,56,200,215
-	pxor	%xmm4,%xmm6
-.byte	15,56,201,252
-.byte	15,56,202,245
-	movdqa	%xmm0,%xmm1
-.byte	15,58,204,194,3
-.byte	15,56,200,204
-	pxor	%xmm5,%xmm7
-.byte	15,56,202,254
-	movdqu	(%esi),%xmm4
-	movdqa	%xmm0,%xmm2
-.byte	15,58,204,193,3
-.byte	15,56,200,213
-	movdqu	16(%esi),%xmm5
-.byte	102,15,56,0,227
-	movdqa	%xmm0,%xmm1
-.byte	15,58,204,194,3
-.byte	15,56,200,206
-	movdqu	32(%esi),%xmm6
-.byte	102,15,56,0,235
-	movdqa	%xmm0,%xmm2
-.byte	15,58,204,193,3
-.byte	15,56,200,215
-	movdqu	48(%esi),%xmm7
-.byte	102,15,56,0,243
-	movdqa	%xmm0,%xmm1
-.byte	15,58,204,194,3
-	movdqa	(%esp),%xmm2
-.byte	102,15,56,0,251
-.byte	15,56,200,202
-	paddd	16(%esp),%xmm0
-	jnz	L004loop_shaext
-	pshufd	$27,%xmm0,%xmm0
-	pshufd	$27,%xmm1,%xmm1
-	movdqu	%xmm0,(%edi)
-	movd	%xmm1,16(%edi)
-	movl	%ebx,%esp
-	popl	%edi
-	popl	%esi
-	popl	%ebx
-	popl	%ebp
-	ret
-.private_extern	__sha1_block_data_order_ssse3
-.align	4
-__sha1_block_data_order_ssse3:
-	pushl	%ebp
-	pushl	%ebx
-	pushl	%esi
-	pushl	%edi
-	call	L005pic_point
-L005pic_point:
-	popl	%ebp
-	leal	LK_XX_XX-L005pic_point(%ebp),%ebp
 Lssse3_shortcut:
 	movdqa	(%ebp),%xmm7
 	movdqa	16(%ebp),%xmm0
@@ -1623,9 +1457,9 @@
 	xorl	%edx,%ebp
 	pshufd	$238,%xmm0,%xmm4
 	andl	%ebp,%esi
-	jmp	L006loop
+	jmp	L004loop
 .align	4,0x90
-L006loop:
+L004loop:
 	rorl	$2,%ebx
 	xorl	%edx,%esi
 	movl	%eax,%ebp
@@ -2528,7 +2362,7 @@
 	addl	%edx,%ecx
 	movl	196(%esp),%ebp
 	cmpl	200(%esp),%ebp
-	je	L007done
+	je	L005done
 	movdqa	160(%esp),%xmm7
 	movdqa	176(%esp),%xmm6
 	movdqu	(%ebp),%xmm0
@@ -2663,9 +2497,9 @@
 	pshufd	$238,%xmm0,%xmm4
 	andl	%ebx,%esi
 	movl	%ebp,%ebx
-	jmp	L006loop
+	jmp	L004loop
 .align	4,0x90
-L007done:
+L005done:
 	addl	16(%esp),%ebx
 	xorl	%edi,%esi
 	movl	%ecx,%ebp
@@ -2778,6 +2612,1175 @@
 	popl	%ebx
 	popl	%ebp
 	ret
+.private_extern	__sha1_block_data_order_avx
+.align	4
+__sha1_block_data_order_avx:
+	pushl	%ebp
+	pushl	%ebx
+	pushl	%esi
+	pushl	%edi
+	call	L006pic_point
+L006pic_point:
+	popl	%ebp
+	leal	LK_XX_XX-L006pic_point(%ebp),%ebp
+Lavx_shortcut:
+	vzeroall
+	vmovdqa	(%ebp),%xmm7
+	vmovdqa	16(%ebp),%xmm0
+	vmovdqa	32(%ebp),%xmm1
+	vmovdqa	48(%ebp),%xmm2
+	vmovdqa	64(%ebp),%xmm6
+	movl	20(%esp),%edi
+	movl	24(%esp),%ebp
+	movl	28(%esp),%edx
+	movl	%esp,%esi
+	subl	$208,%esp
+	andl	$-64,%esp
+	vmovdqa	%xmm0,112(%esp)
+	vmovdqa	%xmm1,128(%esp)
+	vmovdqa	%xmm2,144(%esp)
+	shll	$6,%edx
+	vmovdqa	%xmm7,160(%esp)
+	addl	%ebp,%edx
+	vmovdqa	%xmm6,176(%esp)
+	addl	$64,%ebp
+	movl	%edi,192(%esp)
+	movl	%ebp,196(%esp)
+	movl	%edx,200(%esp)
+	movl	%esi,204(%esp)
+	movl	(%edi),%eax
+	movl	4(%edi),%ebx
+	movl	8(%edi),%ecx
+	movl	12(%edi),%edx
+	movl	16(%edi),%edi
+	movl	%ebx,%esi
+	vmovdqu	-64(%ebp),%xmm0
+	vmovdqu	-48(%ebp),%xmm1
+	vmovdqu	-32(%ebp),%xmm2
+	vmovdqu	-16(%ebp),%xmm3
+	vpshufb	%xmm6,%xmm0,%xmm0
+	vpshufb	%xmm6,%xmm1,%xmm1
+	vpshufb	%xmm6,%xmm2,%xmm2
+	vmovdqa	%xmm7,96(%esp)
+	vpshufb	%xmm6,%xmm3,%xmm3
+	vpaddd	%xmm7,%xmm0,%xmm4
+	vpaddd	%xmm7,%xmm1,%xmm5
+	vpaddd	%xmm7,%xmm2,%xmm6
+	vmovdqa	%xmm4,(%esp)
+	movl	%ecx,%ebp
+	vmovdqa	%xmm5,16(%esp)
+	xorl	%edx,%ebp
+	vmovdqa	%xmm6,32(%esp)
+	andl	%ebp,%esi
+	jmp	L007loop
+.align	4,0x90
+L007loop:
+	shrdl	$2,%ebx,%ebx
+	xorl	%edx,%esi
+	vpalignr	$8,%xmm0,%xmm1,%xmm4
+	movl	%eax,%ebp
+	addl	(%esp),%edi
+	vpaddd	%xmm3,%xmm7,%xmm7
+	vmovdqa	%xmm0,64(%esp)
+	xorl	%ecx,%ebx
+	shldl	$5,%eax,%eax
+	vpsrldq	$4,%xmm3,%xmm6
+	addl	%esi,%edi
+	andl	%ebx,%ebp
+	vpxor	%xmm0,%xmm4,%xmm4
+	xorl	%ecx,%ebx
+	addl	%eax,%edi
+	vpxor	%xmm2,%xmm6,%xmm6
+	shrdl	$7,%eax,%eax
+	xorl	%ecx,%ebp
+	vmovdqa	%xmm7,48(%esp)
+	movl	%edi,%esi
+	addl	4(%esp),%edx
+	vpxor	%xmm6,%xmm4,%xmm4
+	xorl	%ebx,%eax
+	shldl	$5,%edi,%edi
+	addl	%ebp,%edx
+	andl	%eax,%esi
+	vpsrld	$31,%xmm4,%xmm6
+	xorl	%ebx,%eax
+	addl	%edi,%edx
+	shrdl	$7,%edi,%edi
+	xorl	%ebx,%esi
+	vpslldq	$12,%xmm4,%xmm0
+	vpaddd	%xmm4,%xmm4,%xmm4
+	movl	%edx,%ebp
+	addl	8(%esp),%ecx
+	xorl	%eax,%edi
+	shldl	$5,%edx,%edx
+	vpsrld	$30,%xmm0,%xmm7
+	vpor	%xmm6,%xmm4,%xmm4
+	addl	%esi,%ecx
+	andl	%edi,%ebp
+	xorl	%eax,%edi
+	addl	%edx,%ecx
+	vpslld	$2,%xmm0,%xmm0
+	shrdl	$7,%edx,%edx
+	xorl	%eax,%ebp
+	vpxor	%xmm7,%xmm4,%xmm4
+	movl	%ecx,%esi
+	addl	12(%esp),%ebx
+	xorl	%edi,%edx
+	shldl	$5,%ecx,%ecx
+	vpxor	%xmm0,%xmm4,%xmm4
+	addl	%ebp,%ebx
+	andl	%edx,%esi
+	vmovdqa	96(%esp),%xmm0
+	xorl	%edi,%edx
+	addl	%ecx,%ebx
+	shrdl	$7,%ecx,%ecx
+	xorl	%edi,%esi
+	vpalignr	$8,%xmm1,%xmm2,%xmm5
+	movl	%ebx,%ebp
+	addl	16(%esp),%eax
+	vpaddd	%xmm4,%xmm0,%xmm0
+	vmovdqa	%xmm1,80(%esp)
+	xorl	%edx,%ecx
+	shldl	$5,%ebx,%ebx
+	vpsrldq	$4,%xmm4,%xmm7
+	addl	%esi,%eax
+	andl	%ecx,%ebp
+	vpxor	%xmm1,%xmm5,%xmm5
+	xorl	%edx,%ecx
+	addl	%ebx,%eax
+	vpxor	%xmm3,%xmm7,%xmm7
+	shrdl	$7,%ebx,%ebx
+	xorl	%edx,%ebp
+	vmovdqa	%xmm0,(%esp)
+	movl	%eax,%esi
+	addl	20(%esp),%edi
+	vpxor	%xmm7,%xmm5,%xmm5
+	xorl	%ecx,%ebx
+	shldl	$5,%eax,%eax
+	addl	%ebp,%edi
+	andl	%ebx,%esi
+	vpsrld	$31,%xmm5,%xmm7
+	xorl	%ecx,%ebx
+	addl	%eax,%edi
+	shrdl	$7,%eax,%eax
+	xorl	%ecx,%esi
+	vpslldq	$12,%xmm5,%xmm1
+	vpaddd	%xmm5,%xmm5,%xmm5
+	movl	%edi,%ebp
+	addl	24(%esp),%edx
+	xorl	%ebx,%eax
+	shldl	$5,%edi,%edi
+	vpsrld	$30,%xmm1,%xmm0
+	vpor	%xmm7,%xmm5,%xmm5
+	addl	%esi,%edx
+	andl	%eax,%ebp
+	xorl	%ebx,%eax
+	addl	%edi,%edx
+	vpslld	$2,%xmm1,%xmm1
+	shrdl	$7,%edi,%edi
+	xorl	%ebx,%ebp
+	vpxor	%xmm0,%xmm5,%xmm5
+	movl	%edx,%esi
+	addl	28(%esp),%ecx
+	xorl	%eax,%edi
+	shldl	$5,%edx,%edx
+	vpxor	%xmm1,%xmm5,%xmm5
+	addl	%ebp,%ecx
+	andl	%edi,%esi
+	vmovdqa	112(%esp),%xmm1
+	xorl	%eax,%edi
+	addl	%edx,%ecx
+	shrdl	$7,%edx,%edx
+	xorl	%eax,%esi
+	vpalignr	$8,%xmm2,%xmm3,%xmm6
+	movl	%ecx,%ebp
+	addl	32(%esp),%ebx
+	vpaddd	%xmm5,%xmm1,%xmm1
+	vmovdqa	%xmm2,96(%esp)
+	xorl	%edi,%edx
+	shldl	$5,%ecx,%ecx
+	vpsrldq	$4,%xmm5,%xmm0
+	addl	%esi,%ebx
+	andl	%edx,%ebp
+	vpxor	%xmm2,%xmm6,%xmm6
+	xorl	%edi,%edx
+	addl	%ecx,%ebx
+	vpxor	%xmm4,%xmm0,%xmm0
+	shrdl	$7,%ecx,%ecx
+	xorl	%edi,%ebp
+	vmovdqa	%xmm1,16(%esp)
+	movl	%ebx,%esi
+	addl	36(%esp),%eax
+	vpxor	%xmm0,%xmm6,%xmm6
+	xorl	%edx,%ecx
+	shldl	$5,%ebx,%ebx
+	addl	%ebp,%eax
+	andl	%ecx,%esi
+	vpsrld	$31,%xmm6,%xmm0
+	xorl	%edx,%ecx
+	addl	%ebx,%eax
+	shrdl	$7,%ebx,%ebx
+	xorl	%edx,%esi
+	vpslldq	$12,%xmm6,%xmm2
+	vpaddd	%xmm6,%xmm6,%xmm6
+	movl	%eax,%ebp
+	addl	40(%esp),%edi
+	xorl	%ecx,%ebx
+	shldl	$5,%eax,%eax
+	vpsrld	$30,%xmm2,%xmm1
+	vpor	%xmm0,%xmm6,%xmm6
+	addl	%esi,%edi
+	andl	%ebx,%ebp
+	xorl	%ecx,%ebx
+	addl	%eax,%edi
+	vpslld	$2,%xmm2,%xmm2
+	vmovdqa	64(%esp),%xmm0
+	shrdl	$7,%eax,%eax
+	xorl	%ecx,%ebp
+	vpxor	%xmm1,%xmm6,%xmm6
+	movl	%edi,%esi
+	addl	44(%esp),%edx
+	xorl	%ebx,%eax
+	shldl	$5,%edi,%edi
+	vpxor	%xmm2,%xmm6,%xmm6
+	addl	%ebp,%edx
+	andl	%eax,%esi
+	vmovdqa	112(%esp),%xmm2
+	xorl	%ebx,%eax
+	addl	%edi,%edx
+	shrdl	$7,%edi,%edi
+	xorl	%ebx,%esi
+	vpalignr	$8,%xmm3,%xmm4,%xmm7
+	movl	%edx,%ebp
+	addl	48(%esp),%ecx
+	vpaddd	%xmm6,%xmm2,%xmm2
+	vmovdqa	%xmm3,64(%esp)
+	xorl	%eax,%edi
+	shldl	$5,%edx,%edx
+	vpsrldq	$4,%xmm6,%xmm1
+	addl	%esi,%ecx
+	andl	%edi,%ebp
+	vpxor	%xmm3,%xmm7,%xmm7
+	xorl	%eax,%edi
+	addl	%edx,%ecx
+	vpxor	%xmm5,%xmm1,%xmm1
+	shrdl	$7,%edx,%edx
+	xorl	%eax,%ebp
+	vmovdqa	%xmm2,32(%esp)
+	movl	%ecx,%esi
+	addl	52(%esp),%ebx
+	vpxor	%xmm1,%xmm7,%xmm7
+	xorl	%edi,%edx
+	shldl	$5,%ecx,%ecx
+	addl	%ebp,%ebx
+	andl	%edx,%esi
+	vpsrld	$31,%xmm7,%xmm1
+	xorl	%edi,%edx
+	addl	%ecx,%ebx
+	shrdl	$7,%ecx,%ecx
+	xorl	%edi,%esi
+	vpslldq	$12,%xmm7,%xmm3
+	vpaddd	%xmm7,%xmm7,%xmm7
+	movl	%ebx,%ebp
+	addl	56(%esp),%eax
+	xorl	%edx,%ecx
+	shldl	$5,%ebx,%ebx
+	vpsrld	$30,%xmm3,%xmm2
+	vpor	%xmm1,%xmm7,%xmm7
+	addl	%esi,%eax
+	andl	%ecx,%ebp
+	xorl	%edx,%ecx
+	addl	%ebx,%eax
+	vpslld	$2,%xmm3,%xmm3
+	vmovdqa	80(%esp),%xmm1
+	shrdl	$7,%ebx,%ebx
+	xorl	%edx,%ebp
+	vpxor	%xmm2,%xmm7,%xmm7
+	movl	%eax,%esi
+	addl	60(%esp),%edi
+	xorl	%ecx,%ebx
+	shldl	$5,%eax,%eax
+	vpxor	%xmm3,%xmm7,%xmm7
+	addl	%ebp,%edi
+	andl	%ebx,%esi
+	vmovdqa	112(%esp),%xmm3
+	xorl	%ecx,%ebx
+	addl	%eax,%edi
+	vpalignr	$8,%xmm6,%xmm7,%xmm2
+	vpxor	%xmm4,%xmm0,%xmm0
+	shrdl	$7,%eax,%eax
+	xorl	%ecx,%esi
+	movl	%edi,%ebp
+	addl	(%esp),%edx
+	vpxor	%xmm1,%xmm0,%xmm0
+	vmovdqa	%xmm4,80(%esp)
+	xorl	%ebx,%eax
+	shldl	$5,%edi,%edi
+	vmovdqa	%xmm3,%xmm4
+	vpaddd	%xmm7,%xmm3,%xmm3
+	addl	%esi,%edx
+	andl	%eax,%ebp
+	vpxor	%xmm2,%xmm0,%xmm0
+	xorl	%ebx,%eax
+	addl	%edi,%edx
+	shrdl	$7,%edi,%edi
+	xorl	%ebx,%ebp
+	vpsrld	$30,%xmm0,%xmm2
+	vmovdqa	%xmm3,48(%esp)
+	movl	%edx,%esi
+	addl	4(%esp),%ecx
+	xorl	%eax,%edi
+	shldl	$5,%edx,%edx
+	vpslld	$2,%xmm0,%xmm0
+	addl	%ebp,%ecx
+	andl	%edi,%esi
+	xorl	%eax,%edi
+	addl	%edx,%ecx
+	shrdl	$7,%edx,%edx
+	xorl	%eax,%esi
+	movl	%ecx,%ebp
+	addl	8(%esp),%ebx
+	vpor	%xmm2,%xmm0,%xmm0
+	xorl	%edi,%edx
+	shldl	$5,%ecx,%ecx
+	vmovdqa	96(%esp),%xmm2
+	addl	%esi,%ebx
+	andl	%edx,%ebp
+	xorl	%edi,%edx
+	addl	%ecx,%ebx
+	addl	12(%esp),%eax
+	xorl	%edi,%ebp
+	movl	%ebx,%esi
+	shldl	$5,%ebx,%ebx
+	addl	%ebp,%eax
+	xorl	%edx,%esi
+	shrdl	$7,%ecx,%ecx
+	addl	%ebx,%eax
+	vpalignr	$8,%xmm7,%xmm0,%xmm3
+	vpxor	%xmm5,%xmm1,%xmm1
+	addl	16(%esp),%edi
+	xorl	%ecx,%esi
+	movl	%eax,%ebp
+	shldl	$5,%eax,%eax
+	vpxor	%xmm2,%xmm1,%xmm1
+	vmovdqa	%xmm5,96(%esp)
+	addl	%esi,%edi
+	xorl	%ecx,%ebp
+	vmovdqa	%xmm4,%xmm5
+	vpaddd	%xmm0,%xmm4,%xmm4
+	shrdl	$7,%ebx,%ebx
+	addl	%eax,%edi
+	vpxor	%xmm3,%xmm1,%xmm1
+	addl	20(%esp),%edx
+	xorl	%ebx,%ebp
+	movl	%edi,%esi
+	shldl	$5,%edi,%edi
+	vpsrld	$30,%xmm1,%xmm3
+	vmovdqa	%xmm4,(%esp)
+	addl	%ebp,%edx
+	xorl	%ebx,%esi
+	shrdl	$7,%eax,%eax
+	addl	%edi,%edx
+	vpslld	$2,%xmm1,%xmm1
+	addl	24(%esp),%ecx
+	xorl	%eax,%esi
+	movl	%edx,%ebp
+	shldl	$5,%edx,%edx
+	addl	%esi,%ecx
+	xorl	%eax,%ebp
+	shrdl	$7,%edi,%edi
+	addl	%edx,%ecx
+	vpor	%xmm3,%xmm1,%xmm1
+	addl	28(%esp),%ebx
+	xorl	%edi,%ebp
+	vmovdqa	64(%esp),%xmm3
+	movl	%ecx,%esi
+	shldl	$5,%ecx,%ecx
+	addl	%ebp,%ebx
+	xorl	%edi,%esi
+	shrdl	$7,%edx,%edx
+	addl	%ecx,%ebx
+	vpalignr	$8,%xmm0,%xmm1,%xmm4
+	vpxor	%xmm6,%xmm2,%xmm2
+	addl	32(%esp),%eax
+	xorl	%edx,%esi
+	movl	%ebx,%ebp
+	shldl	$5,%ebx,%ebx
+	vpxor	%xmm3,%xmm2,%xmm2
+	vmovdqa	%xmm6,64(%esp)
+	addl	%esi,%eax
+	xorl	%edx,%ebp
+	vmovdqa	128(%esp),%xmm6
+	vpaddd	%xmm1,%xmm5,%xmm5
+	shrdl	$7,%ecx,%ecx
+	addl	%ebx,%eax
+	vpxor	%xmm4,%xmm2,%xmm2
+	addl	36(%esp),%edi
+	xorl	%ecx,%ebp
+	movl	%eax,%esi
+	shldl	$5,%eax,%eax
+	vpsrld	$30,%xmm2,%xmm4
+	vmovdqa	%xmm5,16(%esp)
+	addl	%ebp,%edi
+	xorl	%ecx,%esi
+	shrdl	$7,%ebx,%ebx
+	addl	%eax,%edi
+	vpslld	$2,%xmm2,%xmm2
+	addl	40(%esp),%edx
+	xorl	%ebx,%esi
+	movl	%edi,%ebp
+	shldl	$5,%edi,%edi
+	addl	%esi,%edx
+	xorl	%ebx,%ebp
+	shrdl	$7,%eax,%eax
+	addl	%edi,%edx
+	vpor	%xmm4,%xmm2,%xmm2
+	addl	44(%esp),%ecx
+	xorl	%eax,%ebp
+	vmovdqa	80(%esp),%xmm4
+	movl	%edx,%esi
+	shldl	$5,%edx,%edx
+	addl	%ebp,%ecx
+	xorl	%eax,%esi
+	shrdl	$7,%edi,%edi
+	addl	%edx,%ecx
+	vpalignr	$8,%xmm1,%xmm2,%xmm5
+	vpxor	%xmm7,%xmm3,%xmm3
+	addl	48(%esp),%ebx
+	xorl	%edi,%esi
+	movl	%ecx,%ebp
+	shldl	$5,%ecx,%ecx
+	vpxor	%xmm4,%xmm3,%xmm3
+	vmovdqa	%xmm7,80(%esp)
+	addl	%esi,%ebx
+	xorl	%edi,%ebp
+	vmovdqa	%xmm6,%xmm7
+	vpaddd	%xmm2,%xmm6,%xmm6
+	shrdl	$7,%edx,%edx
+	addl	%ecx,%ebx
+	vpxor	%xmm5,%xmm3,%xmm3
+	addl	52(%esp),%eax
+	xorl	%edx,%ebp
+	movl	%ebx,%esi
+	shldl	$5,%ebx,%ebx
+	vpsrld	$30,%xmm3,%xmm5
+	vmovdqa	%xmm6,32(%esp)
+	addl	%ebp,%eax
+	xorl	%edx,%esi
+	shrdl	$7,%ecx,%ecx
+	addl	%ebx,%eax
+	vpslld	$2,%xmm3,%xmm3
+	addl	56(%esp),%edi
+	xorl	%ecx,%esi
+	movl	%eax,%ebp
+	shldl	$5,%eax,%eax
+	addl	%esi,%edi
+	xorl	%ecx,%ebp
+	shrdl	$7,%ebx,%ebx
+	addl	%eax,%edi
+	vpor	%xmm5,%xmm3,%xmm3
+	addl	60(%esp),%edx
+	xorl	%ebx,%ebp
+	vmovdqa	96(%esp),%xmm5
+	movl	%edi,%esi
+	shldl	$5,%edi,%edi
+	addl	%ebp,%edx
+	xorl	%ebx,%esi
+	shrdl	$7,%eax,%eax
+	addl	%edi,%edx
+	vpalignr	$8,%xmm2,%xmm3,%xmm6
+	vpxor	%xmm0,%xmm4,%xmm4
+	addl	(%esp),%ecx
+	xorl	%eax,%esi
+	movl	%edx,%ebp
+	shldl	$5,%edx,%edx
+	vpxor	%xmm5,%xmm4,%xmm4
+	vmovdqa	%xmm0,96(%esp)
+	addl	%esi,%ecx
+	xorl	%eax,%ebp
+	vmovdqa	%xmm7,%xmm0
+	vpaddd	%xmm3,%xmm7,%xmm7
+	shrdl	$7,%edi,%edi
+	addl	%edx,%ecx
+	vpxor	%xmm6,%xmm4,%xmm4
+	addl	4(%esp),%ebx
+	xorl	%edi,%ebp
+	movl	%ecx,%esi
+	shldl	$5,%ecx,%ecx
+	vpsrld	$30,%xmm4,%xmm6
+	vmovdqa	%xmm7,48(%esp)
+	addl	%ebp,%ebx
+	xorl	%edi,%esi
+	shrdl	$7,%edx,%edx
+	addl	%ecx,%ebx
+	vpslld	$2,%xmm4,%xmm4
+	addl	8(%esp),%eax
+	xorl	%edx,%esi
+	movl	%ebx,%ebp
+	shldl	$5,%ebx,%ebx
+	addl	%esi,%eax
+	xorl	%edx,%ebp
+	shrdl	$7,%ecx,%ecx
+	addl	%ebx,%eax
+	vpor	%xmm6,%xmm4,%xmm4
+	addl	12(%esp),%edi
+	xorl	%ecx,%ebp
+	vmovdqa	64(%esp),%xmm6
+	movl	%eax,%esi
+	shldl	$5,%eax,%eax
+	addl	%ebp,%edi
+	xorl	%ecx,%esi
+	shrdl	$7,%ebx,%ebx
+	addl	%eax,%edi
+	vpalignr	$8,%xmm3,%xmm4,%xmm7
+	vpxor	%xmm1,%xmm5,%xmm5
+	addl	16(%esp),%edx
+	xorl	%ebx,%esi
+	movl	%edi,%ebp
+	shldl	$5,%edi,%edi
+	vpxor	%xmm6,%xmm5,%xmm5
+	vmovdqa	%xmm1,64(%esp)
+	addl	%esi,%edx
+	xorl	%ebx,%ebp
+	vmovdqa	%xmm0,%xmm1
+	vpaddd	%xmm4,%xmm0,%xmm0
+	shrdl	$7,%eax,%eax
+	addl	%edi,%edx
+	vpxor	%xmm7,%xmm5,%xmm5
+	addl	20(%esp),%ecx
+	xorl	%eax,%ebp
+	movl	%edx,%esi
+	shldl	$5,%edx,%edx
+	vpsrld	$30,%xmm5,%xmm7
+	vmovdqa	%xmm0,(%esp)
+	addl	%ebp,%ecx
+	xorl	%eax,%esi
+	shrdl	$7,%edi,%edi
+	addl	%edx,%ecx
+	vpslld	$2,%xmm5,%xmm5
+	addl	24(%esp),%ebx
+	xorl	%edi,%esi
+	movl	%ecx,%ebp
+	shldl	$5,%ecx,%ecx
+	addl	%esi,%ebx
+	xorl	%edi,%ebp
+	shrdl	$7,%edx,%edx
+	addl	%ecx,%ebx
+	vpor	%xmm7,%xmm5,%xmm5
+	addl	28(%esp),%eax
+	vmovdqa	80(%esp),%xmm7
+	shrdl	$7,%ecx,%ecx
+	movl	%ebx,%esi
+	xorl	%edx,%ebp
+	shldl	$5,%ebx,%ebx
+	addl	%ebp,%eax
+	xorl	%ecx,%esi
+	xorl	%edx,%ecx
+	addl	%ebx,%eax
+	vpalignr	$8,%xmm4,%xmm5,%xmm0
+	vpxor	%xmm2,%xmm6,%xmm6
+	addl	32(%esp),%edi
+	andl	%ecx,%esi
+	xorl	%edx,%ecx
+	shrdl	$7,%ebx,%ebx
+	vpxor	%xmm7,%xmm6,%xmm6
+	vmovdqa	%xmm2,80(%esp)
+	movl	%eax,%ebp
+	xorl	%ecx,%esi
+	vmovdqa	%xmm1,%xmm2
+	vpaddd	%xmm5,%xmm1,%xmm1
+	shldl	$5,%eax,%eax
+	addl	%esi,%edi
+	vpxor	%xmm0,%xmm6,%xmm6
+	xorl	%ebx,%ebp
+	xorl	%ecx,%ebx
+	addl	%eax,%edi
+	addl	36(%esp),%edx
+	vpsrld	$30,%xmm6,%xmm0
+	vmovdqa	%xmm1,16(%esp)
+	andl	%ebx,%ebp
+	xorl	%ecx,%ebx
+	shrdl	$7,%eax,%eax
+	movl	%edi,%esi
+	vpslld	$2,%xmm6,%xmm6
+	xorl	%ebx,%ebp
+	shldl	$5,%edi,%edi
+	addl	%ebp,%edx
+	xorl	%eax,%esi
+	xorl	%ebx,%eax
+	addl	%edi,%edx
+	addl	40(%esp),%ecx
+	andl	%eax,%esi
+	vpor	%xmm0,%xmm6,%xmm6
+	xorl	%ebx,%eax
+	shrdl	$7,%edi,%edi
+	vmovdqa	96(%esp),%xmm0
+	movl	%edx,%ebp
+	xorl	%eax,%esi
+	shldl	$5,%edx,%edx
+	addl	%esi,%ecx
+	xorl	%edi,%ebp
+	xorl	%eax,%edi
+	addl	%edx,%ecx
+	addl	44(%esp),%ebx
+	andl	%edi,%ebp
+	xorl	%eax,%edi
+	shrdl	$7,%edx,%edx
+	movl	%ecx,%esi
+	xorl	%edi,%ebp
+	shldl	$5,%ecx,%ecx
+	addl	%ebp,%ebx
+	xorl	%edx,%esi
+	xorl	%edi,%edx
+	addl	%ecx,%ebx
+	vpalignr	$8,%xmm5,%xmm6,%xmm1
+	vpxor	%xmm3,%xmm7,%xmm7
+	addl	48(%esp),%eax
+	andl	%edx,%esi
+	xorl	%edi,%edx
+	shrdl	$7,%ecx,%ecx
+	vpxor	%xmm0,%xmm7,%xmm7
+	vmovdqa	%xmm3,96(%esp)
+	movl	%ebx,%ebp
+	xorl	%edx,%esi
+	vmovdqa	144(%esp),%xmm3
+	vpaddd	%xmm6,%xmm2,%xmm2
+	shldl	$5,%ebx,%ebx
+	addl	%esi,%eax
+	vpxor	%xmm1,%xmm7,%xmm7
+	xorl	%ecx,%ebp
+	xorl	%edx,%ecx
+	addl	%ebx,%eax
+	addl	52(%esp),%edi
+	vpsrld	$30,%xmm7,%xmm1
+	vmovdqa	%xmm2,32(%esp)
+	andl	%ecx,%ebp
+	xorl	%edx,%ecx
+	shrdl	$7,%ebx,%ebx
+	movl	%eax,%esi
+	vpslld	$2,%xmm7,%xmm7
+	xorl	%ecx,%ebp
+	shldl	$5,%eax,%eax
+	addl	%ebp,%edi
+	xorl	%ebx,%esi
+	xorl	%ecx,%ebx
+	addl	%eax,%edi
+	addl	56(%esp),%edx
+	andl	%ebx,%esi
+	vpor	%xmm1,%xmm7,%xmm7
+	xorl	%ecx,%ebx
+	shrdl	$7,%eax,%eax
+	vmovdqa	64(%esp),%xmm1
+	movl	%edi,%ebp
+	xorl	%ebx,%esi
+	shldl	$5,%edi,%edi
+	addl	%esi,%edx
+	xorl	%eax,%ebp
+	xorl	%ebx,%eax
+	addl	%edi,%edx
+	addl	60(%esp),%ecx
+	andl	%eax,%ebp
+	xorl	%ebx,%eax
+	shrdl	$7,%edi,%edi
+	movl	%edx,%esi
+	xorl	%eax,%ebp
+	shldl	$5,%edx,%edx
+	addl	%ebp,%ecx
+	xorl	%edi,%esi
+	xorl	%eax,%edi
+	addl	%edx,%ecx
+	vpalignr	$8,%xmm6,%xmm7,%xmm2
+	vpxor	%xmm4,%xmm0,%xmm0
+	addl	(%esp),%ebx
+	andl	%edi,%esi
+	xorl	%eax,%edi
+	shrdl	$7,%edx,%edx
+	vpxor	%xmm1,%xmm0,%xmm0
+	vmovdqa	%xmm4,64(%esp)
+	movl	%ecx,%ebp
+	xorl	%edi,%esi
+	vmovdqa	%xmm3,%xmm4
+	vpaddd	%xmm7,%xmm3,%xmm3
+	shldl	$5,%ecx,%ecx
+	addl	%esi,%ebx
+	vpxor	%xmm2,%xmm0,%xmm0
+	xorl	%edx,%ebp
+	xorl	%edi,%edx
+	addl	%ecx,%ebx
+	addl	4(%esp),%eax
+	vpsrld	$30,%xmm0,%xmm2
+	vmovdqa	%xmm3,48(%esp)
+	andl	%edx,%ebp
+	xorl	%edi,%edx
+	shrdl	$7,%ecx,%ecx
+	movl	%ebx,%esi
+	vpslld	$2,%xmm0,%xmm0
+	xorl	%edx,%ebp
+	shldl	$5,%ebx,%ebx
+	addl	%ebp,%eax
+	xorl	%ecx,%esi
+	xorl	%edx,%ecx
+	addl	%ebx,%eax
+	addl	8(%esp),%edi
+	andl	%ecx,%esi
+	vpor	%xmm2,%xmm0,%xmm0
+	xorl	%edx,%ecx
+	shrdl	$7,%ebx,%ebx
+	vmovdqa	80(%esp),%xmm2
+	movl	%eax,%ebp
+	xorl	%ecx,%esi
+	shldl	$5,%eax,%eax
+	addl	%esi,%edi
+	xorl	%ebx,%ebp
+	xorl	%ecx,%ebx
+	addl	%eax,%edi
+	addl	12(%esp),%edx
+	andl	%ebx,%ebp
+	xorl	%ecx,%ebx
+	shrdl	$7,%eax,%eax
+	movl	%edi,%esi
+	xorl	%ebx,%ebp
+	shldl	$5,%edi,%edi
+	addl	%ebp,%edx
+	xorl	%eax,%esi
+	xorl	%ebx,%eax
+	addl	%edi,%edx
+	vpalignr	$8,%xmm7,%xmm0,%xmm3
+	vpxor	%xmm5,%xmm1,%xmm1
+	addl	16(%esp),%ecx
+	andl	%eax,%esi
+	xorl	%ebx,%eax
+	shrdl	$7,%edi,%edi
+	vpxor	%xmm2,%xmm1,%xmm1
+	vmovdqa	%xmm5,80(%esp)
+	movl	%edx,%ebp
+	xorl	%eax,%esi
+	vmovdqa	%xmm4,%xmm5
+	vpaddd	%xmm0,%xmm4,%xmm4
+	shldl	$5,%edx,%edx
+	addl	%esi,%ecx
+	vpxor	%xmm3,%xmm1,%xmm1
+	xorl	%edi,%ebp
+	xorl	%eax,%edi
+	addl	%edx,%ecx
+	addl	20(%esp),%ebx
+	vpsrld	$30,%xmm1,%xmm3
+	vmovdqa	%xmm4,(%esp)
+	andl	%edi,%ebp
+	xorl	%eax,%edi
+	shrdl	$7,%edx,%edx
+	movl	%ecx,%esi
+	vpslld	$2,%xmm1,%xmm1
+	xorl	%edi,%ebp
+	shldl	$5,%ecx,%ecx
+	addl	%ebp,%ebx
+	xorl	%edx,%esi
+	xorl	%edi,%edx
+	addl	%ecx,%ebx
+	addl	24(%esp),%eax
+	andl	%edx,%esi
+	vpor	%xmm3,%xmm1,%xmm1
+	xorl	%edi,%edx
+	shrdl	$7,%ecx,%ecx
+	vmovdqa	96(%esp),%xmm3
+	movl	%ebx,%ebp
+	xorl	%edx,%esi
+	shldl	$5,%ebx,%ebx
+	addl	%esi,%eax
+	xorl	%ecx,%ebp
+	xorl	%edx,%ecx
+	addl	%ebx,%eax
+	addl	28(%esp),%edi
+	andl	%ecx,%ebp
+	xorl	%edx,%ecx
+	shrdl	$7,%ebx,%ebx
+	movl	%eax,%esi
+	xorl	%ecx,%ebp
+	shldl	$5,%eax,%eax
+	addl	%ebp,%edi
+	xorl	%ebx,%esi
+	xorl	%ecx,%ebx
+	addl	%eax,%edi
+	vpalignr	$8,%xmm0,%xmm1,%xmm4
+	vpxor	%xmm6,%xmm2,%xmm2
+	addl	32(%esp),%edx
+	andl	%ebx,%esi
+	xorl	%ecx,%ebx
+	shrdl	$7,%eax,%eax
+	vpxor	%xmm3,%xmm2,%xmm2
+	vmovdqa	%xmm6,96(%esp)
+	movl	%edi,%ebp
+	xorl	%ebx,%esi
+	vmovdqa	%xmm5,%xmm6
+	vpaddd	%xmm1,%xmm5,%xmm5
+	shldl	$5,%edi,%edi
+	addl	%esi,%edx
+	vpxor	%xmm4,%xmm2,%xmm2
+	xorl	%eax,%ebp
+	xorl	%ebx,%eax
+	addl	%edi,%edx
+	addl	36(%esp),%ecx
+	vpsrld	$30,%xmm2,%xmm4
+	vmovdqa	%xmm5,16(%esp)
+	andl	%eax,%ebp
+	xorl	%ebx,%eax
+	shrdl	$7,%edi,%edi
+	movl	%edx,%esi
+	vpslld	$2,%xmm2,%xmm2
+	xorl	%eax,%ebp
+	shldl	$5,%edx,%edx
+	addl	%ebp,%ecx
+	xorl	%edi,%esi
+	xorl	%eax,%edi
+	addl	%edx,%ecx
+	addl	40(%esp),%ebx
+	andl	%edi,%esi
+	vpor	%xmm4,%xmm2,%xmm2
+	xorl	%eax,%edi
+	shrdl	$7,%edx,%edx
+	vmovdqa	64(%esp),%xmm4
+	movl	%ecx,%ebp
+	xorl	%edi,%esi
+	shldl	$5,%ecx,%ecx
+	addl	%esi,%ebx
+	xorl	%edx,%ebp
+	xorl	%edi,%edx
+	addl	%ecx,%ebx
+	addl	44(%esp),%eax
+	andl	%edx,%ebp
+	xorl	%edi,%edx
+	shrdl	$7,%ecx,%ecx
+	movl	%ebx,%esi
+	xorl	%edx,%ebp
+	shldl	$5,%ebx,%ebx
+	addl	%ebp,%eax
+	xorl	%edx,%esi
+	addl	%ebx,%eax
+	vpalignr	$8,%xmm1,%xmm2,%xmm5
+	vpxor	%xmm7,%xmm3,%xmm3
+	addl	48(%esp),%edi
+	xorl	%ecx,%esi
+	movl	%eax,%ebp
+	shldl	$5,%eax,%eax
+	vpxor	%xmm4,%xmm3,%xmm3
+	vmovdqa	%xmm7,64(%esp)
+	addl	%esi,%edi
+	xorl	%ecx,%ebp
+	vmovdqa	%xmm6,%xmm7
+	vpaddd	%xmm2,%xmm6,%xmm6
+	shrdl	$7,%ebx,%ebx
+	addl	%eax,%edi
+	vpxor	%xmm5,%xmm3,%xmm3
+	addl	52(%esp),%edx
+	xorl	%ebx,%ebp
+	movl	%edi,%esi
+	shldl	$5,%edi,%edi
+	vpsrld	$30,%xmm3,%xmm5
+	vmovdqa	%xmm6,32(%esp)
+	addl	%ebp,%edx
+	xorl	%ebx,%esi
+	shrdl	$7,%eax,%eax
+	addl	%edi,%edx
+	vpslld	$2,%xmm3,%xmm3
+	addl	56(%esp),%ecx
+	xorl	%eax,%esi
+	movl	%edx,%ebp
+	shldl	$5,%edx,%edx
+	addl	%esi,%ecx
+	xorl	%eax,%ebp
+	shrdl	$7,%edi,%edi
+	addl	%edx,%ecx
+	vpor	%xmm5,%xmm3,%xmm3
+	addl	60(%esp),%ebx
+	xorl	%edi,%ebp
+	movl	%ecx,%esi
+	shldl	$5,%ecx,%ecx
+	addl	%ebp,%ebx
+	xorl	%edi,%esi
+	shrdl	$7,%edx,%edx
+	addl	%ecx,%ebx
+	addl	(%esp),%eax
+	vpaddd	%xmm3,%xmm7,%xmm7
+	xorl	%edx,%esi
+	movl	%ebx,%ebp
+	shldl	$5,%ebx,%ebx
+	addl	%esi,%eax
+	vmovdqa	%xmm7,48(%esp)
+	xorl	%edx,%ebp
+	shrdl	$7,%ecx,%ecx
+	addl	%ebx,%eax
+	addl	4(%esp),%edi
+	xorl	%ecx,%ebp
+	movl	%eax,%esi
+	shldl	$5,%eax,%eax
+	addl	%ebp,%edi
+	xorl	%ecx,%esi
+	shrdl	$7,%ebx,%ebx
+	addl	%eax,%edi
+	addl	8(%esp),%edx
+	xorl	%ebx,%esi
+	movl	%edi,%ebp
+	shldl	$5,%edi,%edi
+	addl	%esi,%edx
+	xorl	%ebx,%ebp
+	shrdl	$7,%eax,%eax
+	addl	%edi,%edx
+	addl	12(%esp),%ecx
+	xorl	%eax,%ebp
+	movl	%edx,%esi
+	shldl	$5,%edx,%edx
+	addl	%ebp,%ecx
+	xorl	%eax,%esi
+	shrdl	$7,%edi,%edi
+	addl	%edx,%ecx
+	movl	196(%esp),%ebp
+	cmpl	200(%esp),%ebp
+	je	L008done
+	vmovdqa	160(%esp),%xmm7
+	vmovdqa	176(%esp),%xmm6
+	vmovdqu	(%ebp),%xmm0
+	vmovdqu	16(%ebp),%xmm1
+	vmovdqu	32(%ebp),%xmm2
+	vmovdqu	48(%ebp),%xmm3
+	addl	$64,%ebp
+	vpshufb	%xmm6,%xmm0,%xmm0
+	movl	%ebp,196(%esp)
+	vmovdqa	%xmm7,96(%esp)
+	addl	16(%esp),%ebx
+	xorl	%edi,%esi
+	vpshufb	%xmm6,%xmm1,%xmm1
+	movl	%ecx,%ebp
+	shldl	$5,%ecx,%ecx
+	vpaddd	%xmm7,%xmm0,%xmm4
+	addl	%esi,%ebx
+	xorl	%edi,%ebp
+	shrdl	$7,%edx,%edx
+	addl	%ecx,%ebx
+	vmovdqa	%xmm4,(%esp)
+	addl	20(%esp),%eax
+	xorl	%edx,%ebp
+	movl	%ebx,%esi
+	shldl	$5,%ebx,%ebx
+	addl	%ebp,%eax
+	xorl	%edx,%esi
+	shrdl	$7,%ecx,%ecx
+	addl	%ebx,%eax
+	addl	24(%esp),%edi
+	xorl	%ecx,%esi
+	movl	%eax,%ebp
+	shldl	$5,%eax,%eax
+	addl	%esi,%edi
+	xorl	%ecx,%ebp
+	shrdl	$7,%ebx,%ebx
+	addl	%eax,%edi
+	addl	28(%esp),%edx
+	xorl	%ebx,%ebp
+	movl	%edi,%esi
+	shldl	$5,%edi,%edi
+	addl	%ebp,%edx
+	xorl	%ebx,%esi
+	shrdl	$7,%eax,%eax
+	addl	%edi,%edx
+	addl	32(%esp),%ecx
+	xorl	%eax,%esi
+	vpshufb	%xmm6,%xmm2,%xmm2
+	movl	%edx,%ebp
+	shldl	$5,%edx,%edx
+	vpaddd	%xmm7,%xmm1,%xmm5
+	addl	%esi,%ecx
+	xorl	%eax,%ebp
+	shrdl	$7,%edi,%edi
+	addl	%edx,%ecx
+	vmovdqa	%xmm5,16(%esp)
+	addl	36(%esp),%ebx
+	xorl	%edi,%ebp
+	movl	%ecx,%esi
+	shldl	$5,%ecx,%ecx
+	addl	%ebp,%ebx
+	xorl	%edi,%esi
+	shrdl	$7,%edx,%edx
+	addl	%ecx,%ebx
+	addl	40(%esp),%eax
+	xorl	%edx,%esi
+	movl	%ebx,%ebp
+	shldl	$5,%ebx,%ebx
+	addl	%esi,%eax
+	xorl	%edx,%ebp
+	shrdl	$7,%ecx,%ecx
+	addl	%ebx,%eax
+	addl	44(%esp),%edi
+	xorl	%ecx,%ebp
+	movl	%eax,%esi
+	shldl	$5,%eax,%eax
+	addl	%ebp,%edi
+	xorl	%ecx,%esi
+	shrdl	$7,%ebx,%ebx
+	addl	%eax,%edi
+	addl	48(%esp),%edx
+	xorl	%ebx,%esi
+	vpshufb	%xmm6,%xmm3,%xmm3
+	movl	%edi,%ebp
+	shldl	$5,%edi,%edi
+	vpaddd	%xmm7,%xmm2,%xmm6
+	addl	%esi,%edx
+	xorl	%ebx,%ebp
+	shrdl	$7,%eax,%eax
+	addl	%edi,%edx
+	vmovdqa	%xmm6,32(%esp)
+	addl	52(%esp),%ecx
+	xorl	%eax,%ebp
+	movl	%edx,%esi
+	shldl	$5,%edx,%edx
+	addl	%ebp,%ecx
+	xorl	%eax,%esi
+	shrdl	$7,%edi,%edi
+	addl	%edx,%ecx
+	addl	56(%esp),%ebx
+	xorl	%edi,%esi
+	movl	%ecx,%ebp
+	shldl	$5,%ecx,%ecx
+	addl	%esi,%ebx
+	xorl	%edi,%ebp
+	shrdl	$7,%edx,%edx
+	addl	%ecx,%ebx
+	addl	60(%esp),%eax
+	xorl	%edx,%ebp
+	movl	%ebx,%esi
+	shldl	$5,%ebx,%ebx
+	addl	%ebp,%eax
+	shrdl	$7,%ecx,%ecx
+	addl	%ebx,%eax
+	movl	192(%esp),%ebp
+	addl	(%ebp),%eax
+	addl	4(%ebp),%esi
+	addl	8(%ebp),%ecx
+	movl	%eax,(%ebp)
+	addl	12(%ebp),%edx
+	movl	%esi,4(%ebp)
+	addl	16(%ebp),%edi
+	movl	%ecx,%ebx
+	movl	%ecx,8(%ebp)
+	xorl	%edx,%ebx
+	movl	%edx,12(%ebp)
+	movl	%edi,16(%ebp)
+	movl	%esi,%ebp
+	andl	%ebx,%esi
+	movl	%ebp,%ebx
+	jmp	L007loop
+.align	4,0x90
+L008done:
+	addl	16(%esp),%ebx
+	xorl	%edi,%esi
+	movl	%ecx,%ebp
+	shldl	$5,%ecx,%ecx
+	addl	%esi,%ebx
+	xorl	%edi,%ebp
+	shrdl	$7,%edx,%edx
+	addl	%ecx,%ebx
+	addl	20(%esp),%eax
+	xorl	%edx,%ebp
+	movl	%ebx,%esi
+	shldl	$5,%ebx,%ebx
+	addl	%ebp,%eax
+	xorl	%edx,%esi
+	shrdl	$7,%ecx,%ecx
+	addl	%ebx,%eax
+	addl	24(%esp),%edi
+	xorl	%ecx,%esi
+	movl	%eax,%ebp
+	shldl	$5,%eax,%eax
+	addl	%esi,%edi
+	xorl	%ecx,%ebp
+	shrdl	$7,%ebx,%ebx
+	addl	%eax,%edi
+	addl	28(%esp),%edx
+	xorl	%ebx,%ebp
+	movl	%edi,%esi
+	shldl	$5,%edi,%edi
+	addl	%ebp,%edx
+	xorl	%ebx,%esi
+	shrdl	$7,%eax,%eax
+	addl	%edi,%edx
+	addl	32(%esp),%ecx
+	xorl	%eax,%esi
+	movl	%edx,%ebp
+	shldl	$5,%edx,%edx
+	addl	%esi,%ecx
+	xorl	%eax,%ebp
+	shrdl	$7,%edi,%edi
+	addl	%edx,%ecx
+	addl	36(%esp),%ebx
+	xorl	%edi,%ebp
+	movl	%ecx,%esi
+	shldl	$5,%ecx,%ecx
+	addl	%ebp,%ebx
+	xorl	%edi,%esi
+	shrdl	$7,%edx,%edx
+	addl	%ecx,%ebx
+	addl	40(%esp),%eax
+	xorl	%edx,%esi
+	movl	%ebx,%ebp
+	shldl	$5,%ebx,%ebx
+	addl	%esi,%eax
+	xorl	%edx,%ebp
+	shrdl	$7,%ecx,%ecx
+	addl	%ebx,%eax
+	addl	44(%esp),%edi
+	xorl	%ecx,%ebp
+	movl	%eax,%esi
+	shldl	$5,%eax,%eax
+	addl	%ebp,%edi
+	xorl	%ecx,%esi
+	shrdl	$7,%ebx,%ebx
+	addl	%eax,%edi
+	addl	48(%esp),%edx
+	xorl	%ebx,%esi
+	movl	%edi,%ebp
+	shldl	$5,%edi,%edi
+	addl	%esi,%edx
+	xorl	%ebx,%ebp
+	shrdl	$7,%eax,%eax
+	addl	%edi,%edx
+	addl	52(%esp),%ecx
+	xorl	%eax,%ebp
+	movl	%edx,%esi
+	shldl	$5,%edx,%edx
+	addl	%ebp,%ecx
+	xorl	%eax,%esi
+	shrdl	$7,%edi,%edi
+	addl	%edx,%ecx
+	addl	56(%esp),%ebx
+	xorl	%edi,%esi
+	movl	%ecx,%ebp
+	shldl	$5,%ecx,%ecx
+	addl	%esi,%ebx
+	xorl	%edi,%ebp
+	shrdl	$7,%edx,%edx
+	addl	%ecx,%ebx
+	addl	60(%esp),%eax
+	xorl	%edx,%ebp
+	movl	%ebx,%esi
+	shldl	$5,%ebx,%ebx
+	addl	%ebp,%eax
+	shrdl	$7,%ecx,%ecx
+	addl	%ebx,%eax
+	vzeroall
+	movl	192(%esp),%ebp
+	addl	(%ebp),%eax
+	movl	204(%esp),%esp
+	addl	4(%ebp),%esi
+	addl	8(%ebp),%ecx
+	movl	%eax,(%ebp)
+	addl	12(%ebp),%edx
+	movl	%esi,4(%ebp)
+	addl	16(%ebp),%edi
+	movl	%ecx,8(%ebp)
+	movl	%edx,12(%ebp)
+	movl	%edi,16(%ebp)
+	popl	%edi
+	popl	%esi
+	popl	%ebx
+	popl	%ebp
+	ret
 .align	6,0x90
 LK_XX_XX:
 .long	1518500249,1518500249,1518500249,1518500249

third_party/boringssl/mac-x86/crypto/sha/sha256-586.S

diff --git a/third_party/boringssl/mac-x86/crypto/sha/sha256-586.S b/third_party/boringssl/mac-x86/crypto/sha/sha256-586.S
index f0ba612..841854f 100644
--- a/third_party/boringssl/mac-x86/crypto/sha/sha256-586.S
+++ b/third_party/boringssl/mac-x86/crypto/sha/sha256-586.S
@@ -36,11 +36,10 @@
 	jz	L003no_xmm
 	andl	$1073741824,%ecx
 	andl	$268435968,%ebx
-	testl	$536870912,%edx
-	jnz	L004shaext
 	orl	%ebx,%ecx
 	andl	$1342177280,%ecx
 	cmpl	$1342177280,%ecx
+	je	L004AVX
 	testl	$512,%ebx
 	jnz	L005SSSE3
 L003no_xmm:
@@ -3166,204 +3165,6 @@
 	popl	%ebp
 	ret
 .align	5,0x90
-L004shaext:
-	subl	$32,%esp
-	movdqu	(%esi),%xmm1
-	leal	128(%ebp),%ebp
-	movdqu	16(%esi),%xmm2
-	movdqa	128(%ebp),%xmm7
-	pshufd	$27,%xmm1,%xmm0
-	pshufd	$177,%xmm1,%xmm1
-	pshufd	$27,%xmm2,%xmm2
-.byte	102,15,58,15,202,8
-	punpcklqdq	%xmm0,%xmm2
-	jmp	L010loop_shaext
-.align	4,0x90
-L010loop_shaext:
-	movdqu	(%edi),%xmm3
-	movdqu	16(%edi),%xmm4
-	movdqu	32(%edi),%xmm5
-.byte	102,15,56,0,223
-	movdqu	48(%edi),%xmm6
-	movdqa	%xmm2,16(%esp)
-	movdqa	-128(%ebp),%xmm0
-	paddd	%xmm3,%xmm0
-.byte	102,15,56,0,231
-.byte	15,56,203,209
-	pshufd	$14,%xmm0,%xmm0
-	nop
-	movdqa	%xmm1,(%esp)
-.byte	15,56,203,202
-	movdqa	-112(%ebp),%xmm0
-	paddd	%xmm4,%xmm0
-.byte	102,15,56,0,239
-.byte	15,56,203,209
-	pshufd	$14,%xmm0,%xmm0
-	leal	64(%edi),%edi
-.byte	15,56,204,220
-.byte	15,56,203,202
-	movdqa	-96(%ebp),%xmm0
-	paddd	%xmm5,%xmm0
-.byte	102,15,56,0,247
-.byte	15,56,203,209
-	pshufd	$14,%xmm0,%xmm0
-	movdqa	%xmm6,%xmm7
-.byte	102,15,58,15,253,4
-	nop
-	paddd	%xmm7,%xmm3
-.byte	15,56,204,229
-.byte	15,56,203,202
-	movdqa	-80(%ebp),%xmm0
-	paddd	%xmm6,%xmm0
-.byte	15,56,205,222
-.byte	15,56,203,209
-	pshufd	$14,%xmm0,%xmm0
-	movdqa	%xmm3,%xmm7
-.byte	102,15,58,15,254,4
-	nop
-	paddd	%xmm7,%xmm4
-.byte	15,56,204,238
-.byte	15,56,203,202
-	movdqa	-64(%ebp),%xmm0
-	paddd	%xmm3,%xmm0
-.byte	15,56,205,227
-.byte	15,56,203,209
-	pshufd	$14,%xmm0,%xmm0
-	movdqa	%xmm4,%xmm7
-.byte	102,15,58,15,251,4
-	nop
-	paddd	%xmm7,%xmm5
-.byte	15,56,204,243
-.byte	15,56,203,202
-	movdqa	-48(%ebp),%xmm0
-	paddd	%xmm4,%xmm0
-.byte	15,56,205,236
-.byte	15,56,203,209
-	pshufd	$14,%xmm0,%xmm0
-	movdqa	%xmm5,%xmm7
-.byte	102,15,58,15,252,4
-	nop
-	paddd	%xmm7,%xmm6
-.byte	15,56,204,220
-.byte	15,56,203,202
-	movdqa	-32(%ebp),%xmm0
-	paddd	%xmm5,%xmm0
-.byte	15,56,205,245
-.byte	15,56,203,209
-	pshufd	$14,%xmm0,%xmm0
-	movdqa	%xmm6,%xmm7
-.byte	102,15,58,15,253,4
-	nop
-	paddd	%xmm7,%xmm3
-.byte	15,56,204,229
-.byte	15,56,203,202
-	movdqa	-16(%ebp),%xmm0
-	paddd	%xmm6,%xmm0
-.byte	15,56,205,222
-.byte	15,56,203,209
-	pshufd	$14,%xmm0,%xmm0
-	movdqa	%xmm3,%xmm7
-.byte	102,15,58,15,254,4
-	nop
-	paddd	%xmm7,%xmm4
-.byte	15,56,204,238
-.byte	15,56,203,202
-	movdqa	(%ebp),%xmm0
-	paddd	%xmm3,%xmm0
-.byte	15,56,205,227
-.byte	15,56,203,209
-	pshufd	$14,%xmm0,%xmm0
-	movdqa	%xmm4,%xmm7
-.byte	102,15,58,15,251,4
-	nop
-	paddd	%xmm7,%xmm5
-.byte	15,56,204,243
-.byte	15,56,203,202
-	movdqa	16(%ebp),%xmm0
-	paddd	%xmm4,%xmm0
-.byte	15,56,205,236
-.byte	15,56,203,209
-	pshufd	$14,%xmm0,%xmm0
-	movdqa	%xmm5,%xmm7
-.byte	102,15,58,15,252,4
-	nop
-	paddd	%xmm7,%xmm6
-.byte	15,56,204,220
-.byte	15,56,203,202
-	movdqa	32(%ebp),%xmm0
-	paddd	%xmm5,%xmm0
-.byte	15,56,205,245
-.byte	15,56,203,209
-	pshufd	$14,%xmm0,%xmm0
-	movdqa	%xmm6,%xmm7
-.byte	102,15,58,15,253,4
-	nop
-	paddd	%xmm7,%xmm3
-.byte	15,56,204,229
-.byte	15,56,203,202
-	movdqa	48(%ebp),%xmm0
-	paddd	%xmm6,%xmm0
-.byte	15,56,205,222
-.byte	15,56,203,209
-	pshufd	$14,%xmm0,%xmm0
-	movdqa	%xmm3,%xmm7
-.byte	102,15,58,15,254,4
-	nop
-	paddd	%xmm7,%xmm4
-.byte	15,56,204,238
-.byte	15,56,203,202
-	movdqa	64(%ebp),%xmm0
-	paddd	%xmm3,%xmm0
-.byte	15,56,205,227
-.byte	15,56,203,209
-	pshufd	$14,%xmm0,%xmm0
-	movdqa	%xmm4,%xmm7
-.byte	102,15,58,15,251,4
-	nop
-	paddd	%xmm7,%xmm5
-.byte	15,56,204,243
-.byte	15,56,203,202
-	movdqa	80(%ebp),%xmm0
-	paddd	%xmm4,%xmm0
-.byte	15,56,205,236
-.byte	15,56,203,209
-	pshufd	$14,%xmm0,%xmm0
-	movdqa	%xmm5,%xmm7
-.byte	102,15,58,15,252,4
-.byte	15,56,203,202
-	paddd	%xmm7,%xmm6
-	movdqa	96(%ebp),%xmm0
-	paddd	%xmm5,%xmm0
-.byte	15,56,203,209
-	pshufd	$14,%xmm0,%xmm0
-.byte	15,56,205,245
-	movdqa	128(%ebp),%xmm7
-.byte	15,56,203,202
-	movdqa	112(%ebp),%xmm0
-	paddd	%xmm6,%xmm0
-	nop
-.byte	15,56,203,209
-	pshufd	$14,%xmm0,%xmm0
-	cmpl	%edi,%eax
-	nop
-.byte	15,56,203,202
-	paddd	16(%esp),%xmm2
-	paddd	(%esp),%xmm1
-	jnz	L010loop_shaext
-	pshufd	$177,%xmm2,%xmm2
-	pshufd	$27,%xmm1,%xmm7
-	pshufd	$177,%xmm1,%xmm1
-	punpckhqdq	%xmm2,%xmm1
-.byte	102,15,58,15,215,8
-	movl	44(%esp),%esp
-	movdqu	%xmm1,(%esi)
-	movdqu	%xmm2,16(%esi)
-	popl	%edi
-	popl	%esi
-	popl	%ebx
-	popl	%ebp
-	ret
-.align	5,0x90
 L005SSSE3:
 	leal	-96(%esp),%esp
 	movl	(%esi),%eax
@@ -3383,9 +3184,9 @@
 	movl	%ecx,24(%esp)
 	movl	%esi,28(%esp)
 	movdqa	256(%ebp),%xmm7
-	jmp	L011grand_ssse3
+	jmp	L010grand_ssse3
 .align	4,0x90
-L011grand_ssse3:
+L010grand_ssse3:
 	movdqu	(%edi),%xmm0
 	movdqu	16(%edi),%xmm1
 	movdqu	32(%edi),%xmm2
@@ -3408,9 +3209,9 @@
 	paddd	%xmm3,%xmm7
 	movdqa	%xmm6,64(%esp)
 	movdqa	%xmm7,80(%esp)
-	jmp	L012ssse3_00_47
+	jmp	L011ssse3_00_47
 .align	4,0x90
-L012ssse3_00_47:
+L011ssse3_00_47:
 	addl	$64,%ebp
 	movl	%edx,%ecx
 	movdqa	%xmm1,%xmm4
@@ -4053,7 +3854,7 @@
 	addl	%ecx,%eax
 	movdqa	%xmm6,80(%esp)
 	cmpl	$66051,64(%ebp)
-	jne	L012ssse3_00_47
+	jne	L011ssse3_00_47
 	movl	%edx,%ecx
 	rorl	$14,%edx
 	movl	20(%esp),%esi
@@ -4567,13 +4368,1194 @@
 	movdqa	64(%ebp),%xmm7
 	subl	$192,%ebp
 	cmpl	104(%esp),%edi
-	jb	L011grand_ssse3
+	jb	L010grand_ssse3
 	movl	108(%esp),%esp
 	popl	%edi
 	popl	%esi
 	popl	%ebx
 	popl	%ebp
 	ret
+.align	5,0x90
+L004AVX:
+	leal	-96(%esp),%esp
+	vzeroall
+	movl	(%esi),%eax
+	movl	4(%esi),%ebx
+	movl	8(%esi),%ecx
+	movl	12(%esi),%edi
+	movl	%ebx,4(%esp)
+	xorl	%ecx,%ebx
+	movl	%ecx,8(%esp)
+	movl	%edi,12(%esp)
+	movl	16(%esi),%edx
+	movl	20(%esi),%edi
+	movl	24(%esi),%ecx
+	movl	28(%esi),%esi
+	movl	%edi,20(%esp)
+	movl	100(%esp),%edi
+	movl	%ecx,24(%esp)
+	movl	%esi,28(%esp)
+	vmovdqa	256(%ebp),%xmm7
+	jmp	L012grand_avx
+.align	5,0x90
+L012grand_avx:
+	vmovdqu	(%edi),%xmm0
+	vmovdqu	16(%edi),%xmm1
+	vmovdqu	32(%edi),%xmm2
+	vmovdqu	48(%edi),%xmm3
+	addl	$64,%edi
+	vpshufb	%xmm7,%xmm0,%xmm0
+	movl	%edi,100(%esp)
+	vpshufb	%xmm7,%xmm1,%xmm1
+	vpshufb	%xmm7,%xmm2,%xmm2
+	vpaddd	(%ebp),%xmm0,%xmm4
+	vpshufb	%xmm7,%xmm3,%xmm3
+	vpaddd	16(%ebp),%xmm1,%xmm5
+	vpaddd	32(%ebp),%xmm2,%xmm6
+	vpaddd	48(%ebp),%xmm3,%xmm7
+	vmovdqa	%xmm4,32(%esp)
+	vmovdqa	%xmm5,48(%esp)
+	vmovdqa	%xmm6,64(%esp)
+	vmovdqa	%xmm7,80(%esp)
+	jmp	L013avx_00_47
+.align	4,0x90
+L013avx_00_47:
+	addl	$64,%ebp
+	vpalignr	$4,%xmm0,%xmm1,%xmm4
+	movl	%edx,%ecx
+	shrdl	$14,%edx,%edx
+	movl	20(%esp),%esi
+	vpalignr	$4,%xmm2,%xmm3,%xmm7
+	xorl	%ecx,%edx
+	movl	24(%esp),%edi
+	xorl	%edi,%esi
+	vpsrld	$7,%xmm4,%xmm6
+	shrdl	$5,%edx,%edx
+	andl	%ecx,%esi
+	movl	%ecx,16(%esp)
+	vpaddd	%xmm7,%xmm0,%xmm0
+	xorl	%ecx,%edx
+	xorl	%esi,%edi
+	shrdl	$6,%edx,%edx
+	vpsrld	$3,%xmm4,%xmm7
+	movl	%eax,%ecx
+	addl	%edi,%edx
+	movl	4(%esp),%edi
+	vpslld	$14,%xmm4,%xmm5
+	movl	%eax,%esi
+	shrdl	$9,%ecx,%ecx
+	movl	%eax,(%esp)
+	vpxor	%xmm6,%xmm7,%xmm4
+	xorl	%eax,%ecx
+	xorl	%edi,%eax
+	addl	28(%esp),%edx
+	vpshufd	$250,%xmm3,%xmm7
+	shrdl	$11,%ecx,%ecx
+	andl	%eax,%ebx
+	xorl	%esi,%ecx
+	vpsrld	$11,%xmm6,%xmm6
+	addl	32(%esp),%edx
+	xorl	%edi,%ebx
+	shrdl	$2,%ecx,%ecx
+	vpxor	%xmm5,%xmm4,%xmm4
+	addl	%edx,%ebx
+	addl	12(%esp),%edx
+	addl	%ecx,%ebx
+	vpslld	$11,%xmm5,%xmm5
+	movl	%edx,%ecx
+	shrdl	$14,%edx,%edx
+	movl	16(%esp),%esi
+	vpxor	%xmm6,%xmm4,%xmm4
+	xorl	%ecx,%edx
+	movl	20(%esp),%edi
+	xorl	%edi,%esi
+	vpsrld	$10,%xmm7,%xmm6
+	shrdl	$5,%edx,%edx
+	andl	%ecx,%esi
+	movl	%ecx,12(%esp)
+	vpxor	%xmm5,%xmm4,%xmm4
+	xorl	%ecx,%edx
+	xorl	%esi,%edi
+	shrdl	$6,%edx,%edx
+	vpsrlq	$17,%xmm7,%xmm5
+	movl	%ebx,%ecx
+	addl	%edi,%edx
+	movl	(%esp),%edi
+	vpaddd	%xmm4,%xmm0,%xmm0
+	movl	%ebx,%esi
+	shrdl	$9,%ecx,%ecx
+	movl	%ebx,28(%esp)
+	vpxor	%xmm5,%xmm6,%xmm6
+	xorl	%ebx,%ecx
+	xorl	%edi,%ebx
+	addl	24(%esp),%edx
+	vpsrlq	$19,%xmm7,%xmm7
+	shrdl	$11,%ecx,%ecx
+	andl	%ebx,%eax
+	xorl	%esi,%ecx
+	vpxor	%xmm7,%xmm6,%xmm6
+	addl	36(%esp),%edx
+	xorl	%edi,%eax
+	shrdl	$2,%ecx,%ecx
+	vpshufd	$132,%xmm6,%xmm7
+	addl	%edx,%eax
+	addl	8(%esp),%edx
+	addl	%ecx,%eax
+	vpsrldq	$8,%xmm7,%xmm7
+	movl	%edx,%ecx
+	shrdl	$14,%edx,%edx
+	movl	12(%esp),%esi
+	vpaddd	%xmm7,%xmm0,%xmm0
+	xorl	%ecx,%edx
+	movl	16(%esp),%edi
+	xorl	%edi,%esi
+	vpshufd	$80,%xmm0,%xmm7
+	shrdl	$5,%edx,%edx
+	andl	%ecx,%esi
+	movl	%ecx,8(%esp)
+	vpsrld	$10,%xmm7,%xmm6
+	xorl	%ecx,%edx
+	xorl	%esi,%edi
+	shrdl	$6,%edx,%edx
+	vpsrlq	$17,%xmm7,%xmm5
+	movl	%eax,%ecx
+	addl	%edi,%edx
+	movl	28(%esp),%edi
+	vpxor	%xmm5,%xmm6,%xmm6
+	movl	%eax,%esi
+	shrdl	$9,%ecx,%ecx
+	movl	%eax,24(%esp)
+	vpsrlq	$19,%xmm7,%xmm7
+	xorl	%eax,%ecx
+	xorl	%edi,%eax
+	addl	20(%esp),%edx
+	vpxor	%xmm7,%xmm6,%xmm6
+	shrdl	$11,%ecx,%ecx
+	andl	%eax,%ebx
+	xorl	%esi,%ecx
+	vpshufd	$232,%xmm6,%xmm7
+	addl	40(%esp),%edx
+	xorl	%edi,%ebx
+	shrdl	$2,%ecx,%ecx
+	vpslldq	$8,%xmm7,%xmm7
+	addl	%edx,%ebx
+	addl	4(%esp),%edx
+	addl	%ecx,%ebx
+	vpaddd	%xmm7,%xmm0,%xmm0
+	movl	%edx,%ecx
+	shrdl	$14,%edx,%edx
+	movl	8(%esp),%esi
+	vpaddd	(%ebp),%xmm0,%xmm6
+	xorl	%ecx,%edx
+	movl	12(%esp),%edi
+	xorl	%edi,%esi
+	shrdl	$5,%edx,%edx
+	andl	%ecx,%esi
+	movl	%ecx,4(%esp)
+	xorl	%ecx,%edx
+	xorl	%esi,%edi
+	shrdl	$6,%edx,%edx
+	movl	%ebx,%ecx
+	addl	%edi,%edx
+	movl	24(%esp),%edi
+	movl	%ebx,%esi
+	shrdl	$9,%ecx,%ecx
+	movl	%ebx,20(%esp)
+	xorl	%ebx,%ecx
+	xorl	%edi,%ebx
+	addl	16(%esp),%edx
+	shrdl	$11,%ecx,%ecx
+	andl	%ebx,%eax
+	xorl	%esi,%ecx
+	addl	44(%esp),%edx
+	xorl	%edi,%eax
+	shrdl	$2,%ecx,%ecx
+	addl	%edx,%eax
+	addl	(%esp),%edx
+	addl	%ecx,%eax
+	vmovdqa	%xmm6,32(%esp)
+	vpalignr	$4,%xmm1,%xmm2,%xmm4
+	movl	%edx,%ecx
+	shrdl	$14,%edx,%edx
+	movl	4(%esp),%esi
+	vpalignr	$4,%xmm3,%xmm0,%xmm7
+	xorl	%ecx,%edx
+	movl	8(%esp),%edi
+	xorl	%edi,%esi
+	vpsrld	$7,%xmm4,%xmm6
+	shrdl	$5,%edx,%edx
+	andl	%ecx,%esi
+	movl	%ecx,(%esp)
+	vpaddd	%xmm7,%xmm1,%xmm1
+	xorl	%ecx,%edx
+	xorl	%esi,%edi
+	shrdl	$6,%edx,%edx
+	vpsrld	$3,%xmm4,%xmm7
+	movl	%eax,%ecx
+	addl	%edi,%edx
+	movl	20(%esp),%edi
+	vpslld	$14,%xmm4,%xmm5
+	movl	%eax,%esi
+	shrdl	$9,%ecx,%ecx
+	movl	%eax,16(%esp)
+	vpxor	%xmm6,%xmm7,%xmm4
+	xorl	%eax,%ecx
+	xorl	%edi,%eax
+	addl	12(%esp),%edx
+	vpshufd	$250,%xmm0,%xmm7
+	shrdl	$11,%ecx,%ecx
+	andl	%eax,%ebx
+	xorl	%esi,%ecx
+	vpsrld	$11,%xmm6,%xmm6
+	addl	48(%esp),%edx
+	xorl	%edi,%ebx
+	shrdl	$2,%ecx,%ecx
+	vpxor	%xmm5,%xmm4,%xmm4
+	addl	%edx,%ebx
+	addl	28(%esp),%edx
+	addl	%ecx,%ebx
+	vpslld	$11,%xmm5,%xmm5
+	movl	%edx,%ecx
+	shrdl	$14,%edx,%edx
+	movl	(%esp),%esi
+	vpxor	%xmm6,%xmm4,%xmm4
+	xorl	%ecx,%edx
+	movl	4(%esp),%edi
+	xorl	%edi,%esi
+	vpsrld	$10,%xmm7,%xmm6
+	shrdl	$5,%edx,%edx
+	andl	%ecx,%esi
+	movl	%ecx,28(%esp)
+	vpxor	%xmm5,%xmm4,%xmm4
+	xorl	%ecx,%edx
+	xorl	%esi,%edi
+	shrdl	$6,%edx,%edx
+	vpsrlq	$17,%xmm7,%xmm5
+	movl	%ebx,%ecx
+	addl	%edi,%edx
+	movl	16(%esp),%edi
+	vpaddd	%xmm4,%xmm1,%xmm1
+	movl	%ebx,%esi
+	shrdl	$9,%ecx,%ecx
+	movl	%ebx,12(%esp)
+	vpxor	%xmm5,%xmm6,%xmm6
+	xorl	%ebx,%ecx
+	xorl	%edi,%ebx
+	addl	8(%esp),%edx
+	vpsrlq	$19,%xmm7,%xmm7
+	shrdl	$11,%ecx,%ecx
+	andl	%ebx,%eax
+	xorl	%esi,%ecx
+	vpxor	%xmm7,%xmm6,%xmm6
+	addl	52(%esp),%edx
+	xorl	%edi,%eax
+	shrdl	$2,%ecx,%ecx
+	vpshufd	$132,%xmm6,%xmm7
+	addl	%edx,%eax
+	addl	24(%esp),%edx
+	addl	%ecx,%eax
+	vpsrldq	$8,%xmm7,%xmm7
+	movl	%edx,%ecx
+	shrdl	$14,%edx,%edx
+	movl	28(%esp),%esi
+	vpaddd	%xmm7,%xmm1,%xmm1
+	xorl	%ecx,%edx
+	movl	(%esp),%edi
+	xorl	%edi,%esi
+	vpshufd	$80,%xmm1,%xmm7
+	shrdl	$5,%edx,%edx
+	andl	%ecx,%esi
+	movl	%ecx,24(%esp)
+	vpsrld	$10,%xmm7,%xmm6
+	xorl	%ecx,%edx
+	xorl	%esi,%edi
+	shrdl	$6,%edx,%edx
+	vpsrlq	$17,%xmm7,%xmm5
+	movl	%eax,%ecx
+	addl	%edi,%edx
+	movl	12(%esp),%edi
+	vpxor	%xmm5,%xmm6,%xmm6
+	movl	%eax,%esi
+	shrdl	$9,%ecx,%ecx
+	movl	%eax,8(%esp)
+	vpsrlq	$19,%xmm7,%xmm7
+	xorl	%eax,%ecx
+	xorl	%edi,%eax
+	addl	4(%esp),%edx
+	vpxor	%xmm7,%xmm6,%xmm6
+	shrdl	$11,%ecx,%ecx
+	andl	%eax,%ebx
+	xorl	%esi,%ecx
+	vpshufd	$232,%xmm6,%xmm7
+	addl	56(%esp),%edx
+	xorl	%edi,%ebx
+	shrdl	$2,%ecx,%ecx
+	vpslldq	$8,%xmm7,%xmm7
+	addl	%edx,%ebx
+	addl	20(%esp),%edx
+	addl	%ecx,%ebx
+	vpaddd	%xmm7,%xmm1,%xmm1
+	movl	%edx,%ecx
+	shrdl	$14,%edx,%edx
+	movl	24(%esp),%esi
+	vpaddd	16(%ebp),%xmm1,%xmm6
+	xorl	%ecx,%edx
+	movl	28(%esp),%edi
+	xorl	%edi,%esi
+	shrdl	$5,%edx,%edx
+	andl	%ecx,%esi
+	movl	%ecx,20(%esp)
+	xorl	%ecx,%edx
+	xorl	%esi,%edi
+	shrdl	$6,%edx,%edx
+	movl	%ebx,%ecx
+	addl	%edi,%edx
+	movl	8(%esp),%edi
+	movl	%ebx,%esi
+	shrdl	$9,%ecx,%ecx
+	movl	%ebx,4(%esp)
+	xorl	%ebx,%ecx
+	xorl	%edi,%ebx
+	addl	(%esp),%edx
+	shrdl	$11,%ecx,%ecx
+	andl	%ebx,%eax
+	xorl	%esi,%ecx
+	addl	60(%esp),%edx
+	xorl	%edi,%eax
+	shrdl	$2,%ecx,%ecx
+	addl	%edx,%eax
+	addl	16(%esp),%edx
+	addl	%ecx,%eax
+	vmovdqa	%xmm6,48(%esp)
+	vpalignr	$4,%xmm2,%xmm3,%xmm4
+	movl	%edx,%ecx
+	shrdl	$14,%edx,%edx
+	movl	20(%esp),%esi
+	vpalignr	$4,%xmm0,%xmm1,%xmm7
+	xorl	%ecx,%edx
+	movl	24(%esp),%edi
+	xorl	%edi,%esi
+	vpsrld	$7,%xmm4,%xmm6
+	shrdl	$5,%edx,%edx
+	andl	%ecx,%esi
+	movl	%ecx,16(%esp)
+	vpaddd	%xmm7,%xmm2,%xmm2
+	xorl	%ecx,%edx
+	xorl	%esi,%edi
+	shrdl	$6,%edx,%edx
+	vpsrld	$3,%xmm4,%xmm7
+	movl	%eax,%ecx
+	addl	%edi,%edx
+	movl	4(%esp),%edi
+	vpslld	$14,%xmm4,%xmm5
+	movl	%eax,%esi
+	shrdl	$9,%ecx,%ecx
+	movl	%eax,(%esp)
+	vpxor	%xmm6,%xmm7,%xmm4
+	xorl	%eax,%ecx
+	xorl	%edi,%eax
+	addl	28(%esp),%edx
+	vpshufd	$250,%xmm1,%xmm7
+	shrdl	$11,%ecx,%ecx
+	andl	%eax,%ebx
+	xorl	%esi,%ecx
+	vpsrld	$11,%xmm6,%xmm6
+	addl	64(%esp),%edx
+	xorl	%edi,%ebx
+	shrdl	$2,%ecx,%ecx
+	vpxor	%xmm5,%xmm4,%xmm4
+	addl	%edx,%ebx
+	addl	12(%esp),%edx
+	addl	%ecx,%ebx
+	vpslld	$11,%xmm5,%xmm5
+	movl	%edx,%ecx
+	shrdl	$14,%edx,%edx
+	movl	16(%esp),%esi
+	vpxor	%xmm6,%xmm4,%xmm4
+	xorl	%ecx,%edx
+	movl	20(%esp),%edi
+	xorl	%edi,%esi
+	vpsrld	$10,%xmm7,%xmm6
+	shrdl	$5,%edx,%edx
+	andl	%ecx,%esi
+	movl	%ecx,12(%esp)
+	vpxor	%xmm5,%xmm4,%xmm4
+	xorl	%ecx,%edx
+	xorl	%esi,%edi
+	shrdl	$6,%edx,%edx
+	vpsrlq	$17,%xmm7,%xmm5
+	movl	%ebx,%ecx
+	addl	%edi,%edx
+	movl	(%esp),%edi
+	vpaddd	%xmm4,%xmm2,%xmm2
+	movl	%ebx,%esi
+	shrdl	$9,%ecx,%ecx
+	movl	%ebx,28(%esp)
+	vpxor	%xmm5,%xmm6,%xmm6
+	xorl	%ebx,%ecx
+	xorl	%edi,%ebx
+	addl	24(%esp),%edx
+	vpsrlq	$19,%xmm7,%xmm7
+	shrdl	$11,%ecx,%ecx
+	andl	%ebx,%eax
+	xorl	%esi,%ecx
+	vpxor	%xmm7,%xmm6,%xmm6
+	addl	68(%esp),%edx
+	xorl	%edi,%eax
+	shrdl	$2,%ecx,%ecx
+	vpshufd	$132,%xmm6,%xmm7
+	addl	%edx,%eax
+	addl	8(%esp),%edx
+	addl	%ecx,%eax
+	vpsrldq	$8,%xmm7,%xmm7
+	movl	%edx,%ecx
+	shrdl	$14,%edx,%edx
+	movl	12(%esp),%esi
+	vpaddd	%xmm7,%xmm2,%xmm2
+	xorl	%ecx,%edx
+	movl	16(%esp),%edi
+	xorl	%edi,%esi
+	vpshufd	$80,%xmm2,%xmm7
+	shrdl	$5,%edx,%edx
+	andl	%ecx,%esi
+	movl	%ecx,8(%esp)
+	vpsrld	$10,%xmm7,%xmm6
+	xorl	%ecx,%edx
+	xorl	%esi,%edi
+	shrdl	$6,%edx,%edx
+	vpsrlq	$17,%xmm7,%xmm5
+	movl	%eax,%ecx
+	addl	%edi,%edx
+	movl	28(%esp),%edi
+	vpxor	%xmm5,%xmm6,%xmm6
+	movl	%eax,%esi
+	shrdl	$9,%ecx,%ecx
+	movl	%eax,24(%esp)
+	vpsrlq	$19,%xmm7,%xmm7
+	xorl	%eax,%ecx
+	xorl	%edi,%eax
+	addl	20(%esp),%edx
+	vpxor	%xmm7,%xmm6,%xmm6
+	shrdl	$11,%ecx,%ecx
+	andl	%eax,%ebx
+	xorl	%esi,%ecx
+	vpshufd	$232,%xmm6,%xmm7
+	addl	72(%esp),%edx
+	xorl	%edi,%ebx
+	shrdl	$2,%ecx,%ecx
+	vpslldq	$8,%xmm7,%xmm7
+	addl	%edx,%ebx
+	addl	4(%esp),%edx
+	addl	%ecx,%ebx
+	vpaddd	%xmm7,%xmm2,%xmm2
+	movl	%edx,%ecx
+	shrdl	$14,%edx,%edx
+	movl	8(%esp),%esi
+	vpaddd	32(%ebp),%xmm2,%xmm6
+	xorl	%ecx,%edx
+	movl	12(%esp),%edi
+	xorl	%edi,%esi
+	shrdl	$5,%edx,%edx
+	andl	%ecx,%esi
+	movl	%ecx,4(%esp)
+	xorl	%ecx,%edx
+	xorl	%esi,%edi
+	shrdl	$6,%edx,%edx
+	movl	%ebx,%ecx
+	addl	%edi,%edx
+	movl	24(%esp),%edi
+	movl	%ebx,%esi
+	shrdl	$9,%ecx,%ecx
+	movl	%ebx,20(%esp)
+	xorl	%ebx,%ecx
+	xorl	%edi,%ebx
+	addl	16(%esp),%edx
+	shrdl	$11,%ecx,%ecx
+	andl	%ebx,%eax
+	xorl	%esi,%ecx
+	addl	76(%esp),%edx
+	xorl	%edi,%eax
+	shrdl	$2,%ecx,%ecx
+	addl	%edx,%eax
+	addl	(%esp),%edx
+	addl	%ecx,%eax
+	vmovdqa	%xmm6,64(%esp)
+	vpalignr	$4,%xmm3,%xmm0,%xmm4
+	movl	%edx,%ecx
+	shrdl	$14,%edx,%edx
+	movl	4(%esp),%esi
+	vpalignr	$4,%xmm1,%xmm2,%xmm7
+	xorl	%ecx,%edx
+	movl	8(%esp),%edi
+	xorl	%edi,%esi
+	vpsrld	$7,%xmm4,%xmm6
+	shrdl	$5,%edx,%edx
+	andl	%ecx,%esi
+	movl	%ecx,(%esp)
+	vpaddd	%xmm7,%xmm3,%xmm3
+	xorl	%ecx,%edx
+	xorl	%esi,%edi
+	shrdl	$6,%edx,%edx
+	vpsrld	$3,%xmm4,%xmm7
+	movl	%eax,%ecx
+	addl	%edi,%edx
+	movl	20(%esp),%edi
+	vpslld	$14,%xmm4,%xmm5
+	movl	%eax,%esi
+	shrdl	$9,%ecx,%ecx
+	movl	%eax,16(%esp)
+	vpxor	%xmm6,%xmm7,%xmm4
+	xorl	%eax,%ecx
+	xorl	%edi,%eax
+	addl	12(%esp),%edx
+	vpshufd	$250,%xmm2,%xmm7
+	shrdl	$11,%ecx,%ecx
+	andl	%eax,%ebx
+	xorl	%esi,%ecx
+	vpsrld	$11,%xmm6,%xmm6
+	addl	80(%esp),%edx
+	xorl	%edi,%ebx
+	shrdl	$2,%ecx,%ecx
+	vpxor	%xmm5,%xmm4,%xmm4
+	addl	%edx,%ebx
+	addl	28(%esp),%edx
+	addl	%ecx,%ebx
+	vpslld	$11,%xmm5,%xmm5
+	movl	%edx,%ecx
+	shrdl	$14,%edx,%edx
+	movl	(%esp),%esi
+	vpxor	%xmm6,%xmm4,%xmm4
+	xorl	%ecx,%edx
+	movl	4(%esp),%edi
+	xorl	%edi,%esi
+	vpsrld	$10,%xmm7,%xmm6
+	shrdl	$5,%edx,%edx
+	andl	%ecx,%esi
+	movl	%ecx,28(%esp)
+	vpxor	%xmm5,%xmm4,%xmm4
+	xorl	%ecx,%edx
+	xorl	%esi,%edi
+	shrdl	$6,%edx,%edx
+	vpsrlq	$17,%xmm7,%xmm5
+	movl	%ebx,%ecx
+	addl	%edi,%edx
+	movl	16(%esp),%edi
+	vpaddd	%xmm4,%xmm3,%xmm3
+	movl	%ebx,%esi
+	shrdl	$9,%ecx,%ecx
+	movl	%ebx,12(%esp)
+	vpxor	%xmm5,%xmm6,%xmm6
+	xorl	%ebx,%ecx
+	xorl	%edi,%ebx
+	addl	8(%esp),%edx
+	vpsrlq	$19,%xmm7,%xmm7
+	shrdl	$11,%ecx,%ecx
+	andl	%ebx,%eax
+	xorl	%esi,%ecx
+	vpxor	%xmm7,%xmm6,%xmm6
+	addl	84(%esp),%edx
+	xorl	%edi,%eax
+	shrdl	$2,%ecx,%ecx
+	vpshufd	$132,%xmm6,%xmm7
+	addl	%edx,%eax
+	addl	24(%esp),%edx
+	addl	%ecx,%eax
+	vpsrldq	$8,%xmm7,%xmm7
+	movl	%edx,%ecx
+	shrdl	$14,%edx,%edx
+	movl	28(%esp),%esi
+	vpaddd	%xmm7,%xmm3,%xmm3
+	xorl	%ecx,%edx
+	movl	(%esp),%edi
+	xorl	%edi,%esi
+	vpshufd	$80,%xmm3,%xmm7
+	shrdl	$5,%edx,%edx
+	andl	%ecx,%esi
+	movl	%ecx,24(%esp)
+	vpsrld	$10,%xmm7,%xmm6
+	xorl	%ecx,%edx
+	xorl	%esi,%edi
+	shrdl	$6,%edx,%edx
+	vpsrlq	$17,%xmm7,%xmm5
+	movl	%eax,%ecx
+	addl	%edi,%edx
+	movl	12(%esp),%edi
+	vpxor	%xmm5,%xmm6,%xmm6
+	movl	%eax,%esi
+	shrdl	$9,%ecx,%ecx
+	movl	%eax,8(%esp)
+	vpsrlq	$19,%xmm7,%xmm7
+	xorl	%eax,%ecx
+	xorl	%edi,%eax
+	addl	4(%esp),%edx
+	vpxor	%xmm7,%xmm6,%xmm6
+	shrdl	$11,%ecx,%ecx
+	andl	%eax,%ebx
+	xorl	%esi,%ecx
+	vpshufd	$232,%xmm6,%xmm7
+	addl	88(%esp),%edx
+	xorl	%edi,%ebx
+	shrdl	$2,%ecx,%ecx
+	vpslldq	$8,%xmm7,%xmm7
+	addl	%edx,%ebx
+	addl	20(%esp),%edx
+	addl	%ecx,%ebx
+	vpaddd	%xmm7,%xmm3,%xmm3
+	movl	%edx,%ecx
+	shrdl	$14,%edx,%edx
+	movl	24(%esp),%esi
+	vpaddd	48(%ebp),%xmm3,%xmm6
+	xorl	%ecx,%edx
+	movl	28(%esp),%edi
+	xorl	%edi,%esi
+	shrdl	$5,%edx,%edx
+	andl	%ecx,%esi
+	movl	%ecx,20(%esp)
+	xorl	%ecx,%edx
+	xorl	%esi,%edi
+	shrdl	$6,%edx,%edx
+	movl	%ebx,%ecx
+	addl	%edi,%edx
+	movl	8(%esp),%edi
+	movl	%ebx,%esi
+	shrdl	$9,%ecx,%ecx
+	movl	%ebx,4(%esp)
+	xorl	%ebx,%ecx
+	xorl	%edi,%ebx
+	addl	(%esp),%edx
+	shrdl	$11,%ecx,%ecx
+	andl	%ebx,%eax
+	xorl	%esi,%ecx
+	addl	92(%esp),%edx
+	xorl	%edi,%eax
+	shrdl	$2,%ecx,%ecx
+	addl	%edx,%eax
+	addl	16(%esp),%edx
+	addl	%ecx,%eax
+	vmovdqa	%xmm6,80(%esp)
+	cmpl	$66051,64(%ebp)
+	jne	L013avx_00_47
+	movl	%edx,%ecx
+	shrdl	$14,%edx,%edx
+	movl	20(%esp),%esi
+	xorl	%ecx,%edx
+	movl	24(%esp),%edi
+	xorl	%edi,%esi
+	shrdl	$5,%edx,%edx
+	andl	%ecx,%esi
+	movl	%ecx,16(%esp)
+	xorl	%ecx,%edx
+	xorl	%esi,%edi
+	shrdl	$6,%edx,%edx
+	movl	%eax,%ecx
+	addl	%edi,%edx
+	movl	4(%esp),%edi
+	movl	%eax,%esi
+	shrdl	$9,%ecx,%ecx
+	movl	%eax,(%esp)
+	xorl	%eax,%ecx
+	xorl	%edi,%eax
+	addl	28(%esp),%edx
+	shrdl	$11,%ecx,%ecx
+	andl	%eax,%ebx
+	xorl	%esi,%ecx
+	addl	32(%esp),%edx
+	xorl	%edi,%ebx
+	shrdl	$2,%ecx,%ecx
+	addl	%edx,%ebx
+	addl	12(%esp),%edx
+	addl	%ecx,%ebx
+	movl	%edx,%ecx
+	shrdl	$14,%edx,%edx
+	movl	16(%esp),%esi
+	xorl	%ecx,%edx
+	movl	20(%esp),%edi
+	xorl	%edi,%esi
+	shrdl	$5,%edx,%edx
+	andl	%ecx,%esi
+	movl	%ecx,12(%esp)
+	xorl	%ecx,%edx
+	xorl	%esi,%edi
+	shrdl	$6,%edx,%edx
+	movl	%ebx,%ecx
+	addl	%edi,%edx
+	movl	(%esp),%edi
+	movl	%ebx,%esi
+	shrdl	$9,%ecx,%ecx
+	movl	%ebx,28(%esp)
+	xorl	%ebx,%ecx
+	xorl	%edi,%ebx
+	addl	24(%esp),%edx
+	shrdl	$11,%ecx,%ecx
+	andl	%ebx,%eax
+	xorl	%esi,%ecx
+	addl	36(%esp),%edx
+	xorl	%edi,%eax
+	shrdl	$2,%ecx,%ecx
+	addl	%edx,%eax
+	addl	8(%esp),%edx
+	addl	%ecx,%eax
+	movl	%edx,%ecx
+	shrdl	$14,%edx,%edx
+	movl	12(%esp),%esi
+	xorl	%ecx,%edx
+	movl	16(%esp),%edi
+	xorl	%edi,%esi
+	shrdl	$5,%edx,%edx
+	andl	%ecx,%esi
+	movl	%ecx,8(%esp)
+	xorl	%ecx,%edx
+	xorl	%esi,%edi
+	shrdl	$6,%edx,%edx
+	movl	%eax,%ecx
+	addl	%edi,%edx
+	movl	28(%esp),%edi
+	movl	%eax,%esi
+	shrdl	$9,%ecx,%ecx
+	movl	%eax,24(%esp)
+	xorl	%eax,%ecx
+	xorl	%edi,%eax
+	addl	20(%esp),%edx
+	shrdl	$11,%ecx,%ecx
+	andl	%eax,%ebx
+	xorl	%esi,%ecx
+	addl	40(%esp),%edx
+	xorl	%edi,%ebx
+	shrdl	$2,%ecx,%ecx
+	addl	%edx,%ebx
+	addl	4(%esp),%edx
+	addl	%ecx,%ebx
+	movl	%edx,%ecx
+	shrdl	$14,%edx,%edx
+	movl	8(%esp),%esi
+	xorl	%ecx,%edx
+	movl	12(%esp),%edi
+	xorl	%edi,%esi
+	shrdl	$5,%edx,%edx
+	andl	%ecx,%esi
+	movl	%ecx,4(%esp)
+	xorl	%ecx,%edx
+	xorl	%esi,%edi
+	shrdl	$6,%edx,%edx
+	movl	%ebx,%ecx
+	addl	%edi,%edx
+	movl	24(%esp),%edi
+	movl	%ebx,%esi
+	shrdl	$9,%ecx,%ecx
+	movl	%ebx,20(%esp)
+	xorl	%ebx,%ecx
+	xorl	%edi,%ebx
+	addl	16(%esp),%edx
+	shrdl	$11,%ecx,%ecx
+	andl	%ebx,%eax
+	xorl	%esi,%ecx
+	addl	44(%esp),%edx
+	xorl	%edi,%eax
+	shrdl	$2,%ecx,%ecx
+	addl	%edx,%eax
+	addl	(%esp),%edx
+	addl	%ecx,%eax
+	movl	%edx,%ecx
+	shrdl	$14,%edx,%edx
+	movl	4(%esp),%esi
+	xorl	%ecx,%edx
+	movl	8(%esp),%edi
+	xorl	%edi,%esi
+	shrdl	$5,%edx,%edx
+	andl	%ecx,%esi
+	movl	%ecx,(%esp)
+	xorl	%ecx,%edx
+	xorl	%esi,%edi
+	shrdl	$6,%edx,%edx
+	movl	%eax,%ecx
+	addl	%edi,%edx
+	movl	20(%esp),%edi
+	movl	%eax,%esi
+	shrdl	$9,%ecx,%ecx
+	movl	%eax,16(%esp)
+	xorl	%eax,%ecx
+	xorl	%edi,%eax
+	addl	12(%esp),%edx
+	shrdl	$11,%ecx,%ecx
+	andl	%eax,%ebx
+	xorl	%esi,%ecx
+	addl	48(%esp),%edx
+	xorl	%edi,%ebx
+	shrdl	$2,%ecx,%ecx
+	addl	%edx,%ebx
+	addl	28(%esp),%edx
+	addl	%ecx,%ebx
+	movl	%edx,%ecx
+	shrdl	$14,%edx,%edx
+	movl	(%esp),%esi
+	xorl	%ecx,%edx
+	movl	4(%esp),%edi
+	xorl	%edi,%esi
+	shrdl	$5,%edx,%edx
+	andl	%ecx,%esi
+	movl	%ecx,28(%esp)
+	xorl	%ecx,%edx
+	xorl	%esi,%edi
+	shrdl	$6,%edx,%edx
+	movl	%ebx,%ecx
+	addl	%edi,%edx
+	movl	16(%esp),%edi
+	movl	%ebx,%esi
+	shrdl	$9,%ecx,%ecx
+	movl	%ebx,12(%esp)
+	xorl	%ebx,%ecx
+	xorl	%edi,%ebx
+	addl	8(%esp),%edx
+	shrdl	$11,%ecx,%ecx
+	andl	%ebx,%eax
+	xorl	%esi,%ecx
+	addl	52(%esp),%edx
+	xorl	%edi,%eax
+	shrdl	$2,%ecx,%ecx
+	addl	%edx,%eax
+	addl	24(%esp),%edx
+	addl	%ecx,%eax
+	movl	%edx,%ecx
+	shrdl	$14,%edx,%edx
+	movl	28(%esp),%esi
+	xorl	%ecx,%edx
+	movl	(%esp),%edi
+	xorl	%edi,%esi
+	shrdl	$5,%edx,%edx
+	andl	%ecx,%esi
+	movl	%ecx,24(%esp)
+	xorl	%ecx,%edx
+	xorl	%esi,%edi
+	shrdl	$6,%edx,%edx
+	movl	%eax,%ecx
+	addl	%edi,%edx
+	movl	12(%esp),%edi
+	movl	%eax,%esi
+	shrdl	$9,%ecx,%ecx
+	movl	%eax,8(%esp)
+	xorl	%eax,%ecx
+	xorl	%edi,%eax
+	addl	4(%esp),%edx
+	shrdl	$11,%ecx,%ecx
+	andl	%eax,%ebx
+	xorl	%esi,%ecx
+	addl	56(%esp),%edx
+	xorl	%edi,%ebx
+	shrdl	$2,%ecx,%ecx
+	addl	%edx,%ebx
+	addl	20(%esp),%edx
+	addl	%ecx,%ebx
+	movl	%edx,%ecx
+	shrdl	$14,%edx,%edx
+	movl	24(%esp),%esi
+	xorl	%ecx,%edx
+	movl	28(%esp),%edi
+	xorl	%edi,%esi
+	shrdl	$5,%edx,%edx
+	andl	%ecx,%esi
+	movl	%ecx,20(%esp)
+	xorl	%ecx,%edx
+	xorl	%esi,%edi
+	shrdl	$6,%edx,%edx
+	movl	%ebx,%ecx
+	addl	%edi,%edx
+	movl	8(%esp),%edi
+	movl	%ebx,%esi
+	shrdl	$9,%ecx,%ecx
+	movl	%ebx,4(%esp)
+	xorl	%ebx,%ecx
+	xorl	%edi,%ebx
+	addl	(%esp),%edx
+	shrdl	$11,%ecx,%ecx
+	andl	%ebx,%eax
+	xorl	%esi,%ecx
+	addl	60(%esp),%edx
+	xorl	%edi,%eax
+	shrdl	$2,%ecx,%ecx
+	addl	%edx,%eax
+	addl	16(%esp),%edx
+	addl	%ecx,%eax
+	movl	%edx,%ecx
+	shrdl	$14,%edx,%edx
+	movl	20(%esp),%esi
+	xorl	%ecx,%edx
+	movl	24(%esp),%edi
+	xorl	%edi,%esi
+	shrdl	$5,%edx,%edx
+	andl	%ecx,%esi
+	movl	%ecx,16(%esp)
+	xorl	%ecx,%edx
+	xorl	%esi,%edi
+	shrdl	$6,%edx,%edx
+	movl	%eax,%ecx
+	addl	%edi,%edx
+	movl	4(%esp),%edi
+	movl	%eax,%esi
+	shrdl	$9,%ecx,%ecx
+	movl	%eax,(%esp)
+	xorl	%eax,%ecx
+	xorl	%edi,%eax
+	addl	28(%esp),%edx
+	shrdl	$11,%ecx,%ecx
+	andl	%eax,%ebx
+	xorl	%esi,%ecx
+	addl	64(%esp),%edx
+	xorl	%edi,%ebx
+	shrdl	$2,%ecx,%ecx
+	addl	%edx,%ebx
+	addl	12(%esp),%edx
+	addl	%ecx,%ebx
+	movl	%edx,%ecx
+	shrdl	$14,%edx,%edx
+	movl	16(%esp),%esi
+	xorl	%ecx,%edx
+	movl	20(%esp),%edi
+	xorl	%edi,%esi
+	shrdl	$5,%edx,%edx
+	andl	%ecx,%esi
+	movl	%ecx,12(%esp)
+	xorl	%ecx,%edx
+	xorl	%esi,%edi
+	shrdl	$6,%edx,%edx
+	movl	%ebx,%ecx
+	addl	%edi,%edx
+	movl	(%esp),%edi
+	movl	%ebx,%esi
+	shrdl	$9,%ecx,%ecx
+	movl	%ebx,28(%esp)
+	xorl	%ebx,%ecx
+	xorl	%edi,%ebx
+	addl	24(%esp),%edx
+	shrdl	$11,%ecx,%ecx
+	andl	%ebx,%eax
+	xorl	%esi,%ecx
+	addl	68(%esp),%edx
+	xorl	%edi,%eax
+	shrdl	$2,%ecx,%ecx
+	addl	%edx,%eax
+	addl	8(%esp),%edx
+	addl	%ecx,%eax
+	movl	%edx,%ecx
+	shrdl	$14,%edx,%edx
+	movl	12(%esp),%esi
+	xorl	%ecx,%edx
+	movl	16(%esp),%edi
+	xorl	%edi,%esi
+	shrdl	$5,%edx,%edx
+	andl	%ecx,%esi
+	movl	%ecx,8(%esp)
+	xorl	%ecx,%edx
+	xorl	%esi,%edi
+	shrdl	$6,%edx,%edx
+	movl	%eax,%ecx
+	addl	%edi,%edx
+	movl	28(%esp),%edi
+	movl	%eax,%esi
+	shrdl	$9,%ecx,%ecx
+	movl	%eax,24(%esp)
+	xorl	%eax,%ecx
+	xorl	%edi,%eax
+	addl	20(%esp),%edx
+	shrdl	$11,%ecx,%ecx
+	andl	%eax,%ebx
+	xorl	%esi,%ecx
+	addl	72(%esp),%edx
+	xorl	%edi,%ebx
+	shrdl	$2,%ecx,%ecx
+	addl	%edx,%ebx
+	addl	4(%esp),%edx
+	addl	%ecx,%ebx
+	movl	%edx,%ecx
+	shrdl	$14,%edx,%edx
+	movl	8(%esp),%esi
+	xorl	%ecx,%edx
+	movl	12(%esp),%edi
+	xorl	%edi,%esi
+	shrdl	$5,%edx,%edx
+	andl	%ecx,%esi
+	movl	%ecx,4(%esp)
+	xorl	%ecx,%edx
+	xorl	%esi,%edi
+	shrdl	$6,%edx,%edx
+	movl	%ebx,%ecx
+	addl	%edi,%edx
+	movl	24(%esp),%edi
+	movl	%ebx,%esi
+	shrdl	$9,%ecx,%ecx
+	movl	%ebx,20(%esp)
+	xorl	%ebx,%ecx
+	xorl	%edi,%ebx
+	addl	16(%esp),%edx
+	shrdl	$11,%ecx,%ecx
+	andl	%ebx,%eax
+	xorl	%esi,%ecx
+	addl	76(%esp),%edx
+	xorl	%edi,%eax
+	shrdl	$2,%ecx,%ecx
+	addl	%edx,%eax
+	addl	(%esp),%edx
+	addl	%ecx,%eax
+	movl	%edx,%ecx
+	shrdl	$14,%edx,%edx
+	movl	4(%esp),%esi
+	xorl	%ecx,%edx
+	movl	8(%esp),%edi
+	xorl	%edi,%esi
+	shrdl	$5,%edx,%edx
+	andl	%ecx,%esi
+	movl	%ecx,(%esp)
+	xorl	%ecx,%edx
+	xorl	%esi,%edi
+	shrdl	$6,%edx,%edx
+	movl	%eax,%ecx
+	addl	%edi,%edx
+	movl	20(%esp),%edi
+	movl	%eax,%esi
+	shrdl	$9,%ecx,%ecx
+	movl	%eax,16(%esp)
+	xorl	%eax,%ecx
+	xorl	%edi,%eax
+	addl	12(%esp),%edx
+	shrdl	$11,%ecx,%ecx
+	andl	%eax,%ebx
+	xorl	%esi,%ecx
+	addl	80(%esp),%edx
+	xorl	%edi,%ebx
+	shrdl	$2,%ecx,%ecx
+	addl	%edx,%ebx
+	addl	28(%esp),%edx
+	addl	%ecx,%ebx
+	movl	%edx,%ecx
+	shrdl	$14,%edx,%edx
+	movl	(%esp),%esi
+	xorl	%ecx,%edx
+	movl	4(%esp),%edi
+	xorl	%edi,%esi
+	shrdl	$5,%edx,%edx
+	andl	%ecx,%esi
+	movl	%ecx,28(%esp)
+	xorl	%ecx,%edx
+	xorl	%esi,%edi
+	shrdl	$6,%edx,%edx
+	movl	%ebx,%ecx
+	addl	%edi,%edx
+	movl	16(%esp),%edi
+	movl	%ebx,%esi
+	shrdl	$9,%ecx,%ecx
+	movl	%ebx,12(%esp)
+	xorl	%ebx,%ecx
+	xorl	%edi,%ebx
+	addl	8(%esp),%edx
+	shrdl	$11,%ecx,%ecx
+	andl	%ebx,%eax
+	xorl	%esi,%ecx
+	addl	84(%esp),%edx
+	xorl	%edi,%eax
+	shrdl	$2,%ecx,%ecx
+	addl	%edx,%eax
+	addl	24(%esp),%edx
+	addl	%ecx,%eax
+	movl	%edx,%ecx
+	shrdl	$14,%edx,%edx
+	movl	28(%esp),%esi
+	xorl	%ecx,%edx
+	movl	(%esp),%edi
+	xorl	%edi,%esi
+	shrdl	$5,%edx,%edx
+	andl	%ecx,%esi
+	movl	%ecx,24(%esp)
+	xorl	%ecx,%edx
+	xorl	%esi,%edi
+	shrdl	$6,%edx,%edx
+	movl	%eax,%ecx
+	addl	%edi,%edx
+	movl	12(%esp),%edi
+	movl	%eax,%esi
+	shrdl	$9,%ecx,%ecx
+	movl	%eax,8(%esp)
+	xorl	%eax,%ecx
+	xorl	%edi,%eax
+	addl	4(%esp),%edx
+	shrdl	$11,%ecx,%ecx
+	andl	%eax,%ebx
+	xorl	%esi,%ecx
+	addl	88(%esp),%edx
+	xorl	%edi,%ebx
+	shrdl	$2,%ecx,%ecx
+	addl	%edx,%ebx
+	addl	20(%esp),%edx
+	addl	%ecx,%ebx
+	movl	%edx,%ecx
+	shrdl	$14,%edx,%edx
+	movl	24(%esp),%esi
+	xorl	%ecx,%edx
+	movl	28(%esp),%edi
+	xorl	%edi,%esi
+	shrdl	$5,%edx,%edx
+	andl	%ecx,%esi
+	movl	%ecx,20(%esp)
+	xorl	%ecx,%edx
+	xorl	%esi,%edi
+	shrdl	$6,%edx,%edx
+	movl	%ebx,%ecx
+	addl	%edi,%edx
+	movl	8(%esp),%edi
+	movl	%ebx,%esi
+	shrdl	$9,%ecx,%ecx
+	movl	%ebx,4(%esp)
+	xorl	%ebx,%ecx
+	xorl	%edi,%ebx
+	addl	(%esp),%edx
+	shrdl	$11,%ecx,%ecx
+	andl	%ebx,%eax
+	xorl	%esi,%ecx
+	addl	92(%esp),%edx
+	xorl	%edi,%eax
+	shrdl	$2,%ecx,%ecx
+	addl	%edx,%eax
+	addl	16(%esp),%edx
+	addl	%ecx,%eax
+	movl	96(%esp),%esi
+	xorl	%edi,%ebx
+	movl	12(%esp),%ecx
+	addl	(%esi),%eax
+	addl	4(%esi),%ebx
+	addl	8(%esi),%edi
+	addl	12(%esi),%ecx
+	movl	%eax,(%esi)
+	movl	%ebx,4(%esi)
+	movl	%edi,8(%esi)
+	movl	%ecx,12(%esi)
+	movl	%ebx,4(%esp)
+	xorl	%edi,%ebx
+	movl	%edi,8(%esp)
+	movl	%ecx,12(%esp)
+	movl	20(%esp),%edi
+	movl	24(%esp),%ecx
+	addl	16(%esi),%edx
+	addl	20(%esi),%edi
+	addl	24(%esi),%ecx
+	movl	%edx,16(%esi)
+	movl	%edi,20(%esi)
+	movl	%edi,20(%esp)
+	movl	28(%esp),%edi
+	movl	%ecx,24(%esi)
+	addl	28(%esi),%edi
+	movl	%ecx,24(%esp)
+	movl	%edi,28(%esi)
+	movl	%edi,28(%esp)
+	movl	100(%esp),%edi
+	vmovdqa	64(%ebp),%xmm7
+	subl	$192,%ebp
+	cmpl	104(%esp),%edi
+	jb	L012grand_avx
+	movl	108(%esp),%esp
+	vzeroall
+	popl	%edi
+	popl	%esi
+	popl	%ebx
+	popl	%ebp
+	ret
 .section __IMPORT,__pointers,non_lazy_symbol_pointers
 L_OPENSSL_ia32cap_P$non_lazy_ptr:
 .indirect_symbol	_OPENSSL_ia32cap_P

third_party/boringssl/mac-x86_64/crypto/aes/aes-x86_64.S

diff --git a/third_party/boringssl/mac-x86_64/crypto/aes/aes-x86_64.S b/third_party/boringssl/mac-x86_64/crypto/aes/aes-x86_64.S
index 02f378b..b5d188a 100644
--- a/third_party/boringssl/mac-x86_64/crypto/aes/aes-x86_64.S
+++ b/third_party/boringssl/mac-x86_64/crypto/aes/aes-x86_64.S
@@ -82,8 +82,8 @@
 	movl	0(%r14,%rdi,8),%edi
 	movl	0(%r14,%rbp,8),%ebp
 
-	andl	$65280,%edi
-	andl	$65280,%ebp
+	andl	$0x0000ff00,%edi
+	andl	$0x0000ff00,%ebp
 
 	xorl	%edi,%r10d
 	xorl	%ebp,%r11d
@@ -95,8 +95,8 @@
 	movl	0(%r14,%rsi,8),%esi
 	movl	0(%r14,%rdi,8),%edi
 
-	andl	$65280,%esi
-	andl	$65280,%edi
+	andl	$0x0000ff00,%esi
+	andl	$0x0000ff00,%edi
 	shrl	$16,%ebx
 	xorl	%esi,%r12d
 	xorl	%edi,%r8d
@@ -109,9 +109,9 @@
 	movl	0(%r14,%rdi,8),%edi
 	movl	0(%r14,%rbp,8),%ebp
 
-	andl	$16711680,%esi
-	andl	$16711680,%edi
-	andl	$16711680,%ebp
+	andl	$0x00ff0000,%esi
+	andl	$0x00ff0000,%edi
+	andl	$0x00ff0000,%ebp
 
 	xorl	%esi,%r10d
 	xorl	%edi,%r11d
@@ -124,9 +124,9 @@
 	movl	2(%r14,%rdi,8),%edi
 	movl	2(%r14,%rbp,8),%ebp
 
-	andl	$16711680,%esi
-	andl	$4278190080,%edi
-	andl	$4278190080,%ebp
+	andl	$0x00ff0000,%esi
+	andl	$0xff000000,%edi
+	andl	$0xff000000,%ebp
 
 	xorl	%esi,%r8d
 	xorl	%edi,%r10d
@@ -139,8 +139,8 @@
 	movl	2(%r14,%rdi,8),%edi
 	movl	16+0(%r15),%eax
 
-	andl	$4278190080,%esi
-	andl	$4278190080,%edi
+	andl	$0xff000000,%esi
+	andl	$0xff000000,%edi
 
 	xorl	%esi,%r12d
 	xorl	%edi,%r8d
@@ -242,8 +242,8 @@
 	xorl	%r8d,%edx
 	cmpq	16(%rsp),%r15
 	je	L$enc_compact_done
-	movl	$2155905152,%r10d
-	movl	$2155905152,%r11d
+	movl	$0x80808080,%r10d
+	movl	$0x80808080,%r11d
 	andl	%eax,%r10d
 	andl	%ebx,%r11d
 	movl	%r10d,%esi
@@ -254,10 +254,10 @@
 	leal	(%rbx,%rbx,1),%r9d
 	subl	%r10d,%esi
 	subl	%r11d,%edi
-	andl	$4278124286,%r8d
-	andl	$4278124286,%r9d
-	andl	$454761243,%esi
-	andl	$454761243,%edi
+	andl	$0xfefefefe,%r8d
+	andl	$0xfefefefe,%r9d
+	andl	$0x1b1b1b1b,%esi
+	andl	$0x1b1b1b1b,%edi
 	movl	%eax,%r10d
 	movl	%ebx,%r11d
 	xorl	%esi,%r8d
@@ -265,9 +265,9 @@
 
 	xorl	%r8d,%eax
 	xorl	%r9d,%ebx
-	movl	$2155905152,%r12d
+	movl	$0x80808080,%r12d
 	roll	$24,%eax
-	movl	$2155905152,%ebp
+	movl	$0x80808080,%ebp
 	roll	$24,%ebx
 	andl	%ecx,%r12d
 	andl	%edx,%ebp
@@ -290,10 +290,10 @@
 	xorl	%r10d,%eax
 	xorl	%r11d,%ebx
 
-	andl	$4278124286,%r8d
-	andl	$4278124286,%r9d
-	andl	$454761243,%esi
-	andl	$454761243,%edi
+	andl	$0xfefefefe,%r8d
+	andl	$0xfefefefe,%r9d
+	andl	$0x1b1b1b1b,%esi
+	andl	$0x1b1b1b1b,%edi
 	movl	%ecx,%r12d
 	movl	%edx,%ebp
 	xorl	%esi,%r8d
@@ -345,7 +345,7 @@
 	andq	$-64,%rsp
 	subq	%rsp,%rcx
 	negq	%rcx
-	andq	$960,%rcx
+	andq	$0x3c0,%rcx
 	subq	%rcx,%rsp
 	subq	$32,%rsp
 
@@ -370,7 +370,7 @@
 	leaq	L$AES_Te+2048(%rip),%r14
 	leaq	768(%rsp),%rbp
 	subq	%r14,%rbp
-	andq	$768,%rbp
+	andq	$0x300,%rbp
 	leaq	(%r14,%rbp,1),%r14
 
 	call	_x86_64_AES_encrypt_compact
@@ -791,7 +791,7 @@
 	andq	$-64,%rsp
 	subq	%rsp,%rcx
 	negq	%rcx
-	andq	$960,%rcx
+	andq	$0x3c0,%rcx
 	subq	%rcx,%rsp
 	subq	$32,%rsp
 
@@ -816,7 +816,7 @@
 	leaq	L$AES_Td+2048(%rip),%r14
 	leaq	768(%rsp),%rbp
 	subq	%r14,%rbp
-	andq	$768,%rbp
+	andq	$0x300,%rbp
 	leaq	(%r14,%rbp,1),%r14
 	shrq	$3,%rbp
 	addq	%rbp,%r14
@@ -1333,9 +1333,9 @@
 	movq	%r14,%r10
 	leaq	2304(%r14),%r11
 	movq	%r15,%r12
-	andq	$4095,%r10
-	andq	$4095,%r11
-	andq	$4095,%r12
+	andq	$0xFFF,%r10
+	andq	$0xFFF,%r11
+	andq	$0xFFF,%r12
 
 	cmpq	%r11,%r12
 	jb	L$cbc_te_break_out
@@ -1344,7 +1344,7 @@
 	jmp	L$cbc_te_ok
 L$cbc_te_break_out:
 	subq	%r10,%r12
-	andq	$4095,%r12
+	andq	$0xFFF,%r12
 	addq	$320,%r12
 	subq	%r12,%r15
 .p2align	2
@@ -1370,7 +1370,7 @@
 
 	movq	%r15,%r10
 	subq	%r14,%r10
-	andq	$4095,%r10
+	andq	$0xfff,%r10
 	cmpq	$2304,%r10
 	jb	L$cbc_do_ecopy
 	cmpq	$4096-248,%r10
@@ -1557,7 +1557,7 @@
 	leaq	-88-63(%rcx),%r10
 	subq	%rbp,%r10
 	negq	%r10
-	andq	$960,%r10
+	andq	$0x3c0,%r10
 	subq	%r10,%rbp
 
 	xchgq	%rsp,%rbp
@@ -1586,7 +1586,7 @@
 	leaq	2048(%r14),%r14
 	leaq	768-8(%rsp),%rax
 	subq	%r14,%rax
-	andq	$768,%rax
+	andq	$0x300,%rax
 	leaq	(%r14,%rax,1),%r14
 
 	cmpq	$0,%rbx

third_party/boringssl/mac-x86_64/crypto/aes/aesni-x86_64.S

diff --git a/third_party/boringssl/mac-x86_64/crypto/aes/aesni-x86_64.S b/third_party/boringssl/mac-x86_64/crypto/aes/aesni-x86_64.S
index 69b22c2..3d98fa1 100644
--- a/third_party/boringssl/mac-x86_64/crypto/aes/aesni-x86_64.S
+++ b/third_party/boringssl/mac-x86_64/crypto/aes/aesni-x86_64.S
@@ -507,7 +507,7 @@
 	testl	%r8d,%r8d
 	jz	L$ecb_decrypt
 
-	cmpq	$128,%rdx
+	cmpq	$0x80,%rdx
 	jb	L$ecb_enc_tail
 
 	movdqu	(%rdi),%xmm2
@@ -519,7 +519,7 @@
 	movdqu	96(%rdi),%xmm8
 	movdqu	112(%rdi),%xmm9
 	leaq	128(%rdi),%rdi
-	subq	$128,%rdx
+	subq	$0x80,%rdx
 	jmp	L$ecb_enc_loop8_enter
 .p2align	4
 L$ecb_enc_loop8:
@@ -547,7 +547,7 @@
 
 	call	_aesni_encrypt8
 
-	subq	$128,%rdx
+	subq	$0x80,%rdx
 	jnc	L$ecb_enc_loop8
 
 	movups	%xmm2,(%rsi)
@@ -561,22 +561,22 @@
 	movups	%xmm8,96(%rsi)
 	movups	%xmm9,112(%rsi)
 	leaq	128(%rsi),%rsi
-	addq	$128,%rdx
+	addq	$0x80,%rdx
 	jz	L$ecb_ret
 
 L$ecb_enc_tail:
 	movups	(%rdi),%xmm2
-	cmpq	$32,%rdx
+	cmpq	$0x20,%rdx
 	jb	L$ecb_enc_one
 	movups	16(%rdi),%xmm3
 	je	L$ecb_enc_two
 	movups	32(%rdi),%xmm4
-	cmpq	$64,%rdx
+	cmpq	$0x40,%rdx
 	jb	L$ecb_enc_three
 	movups	48(%rdi),%xmm5
 	je	L$ecb_enc_four
 	movups	64(%rdi),%xmm6
-	cmpq	$96,%rdx
+	cmpq	$0x60,%rdx
 	jb	L$ecb_enc_five
 	movups	80(%rdi),%xmm7
 	je	L$ecb_enc_six
@@ -650,7 +650,7 @@
 
 .p2align	4
 L$ecb_decrypt:
-	cmpq	$128,%rdx
+	cmpq	$0x80,%rdx
 	jb	L$ecb_dec_tail
 
 	movdqu	(%rdi),%xmm2
@@ -662,7 +662,7 @@
 	movdqu	96(%rdi),%xmm8
 	movdqu	112(%rdi),%xmm9
 	leaq	128(%rdi),%rdi
-	subq	$128,%rdx
+	subq	$0x80,%rdx
 	jmp	L$ecb_dec_loop8_enter
 .p2align	4
 L$ecb_dec_loop8:
@@ -691,7 +691,7 @@
 	call	_aesni_decrypt8
 
 	movups	(%r11),%xmm0
-	subq	$128,%rdx
+	subq	$0x80,%rdx
 	jnc	L$ecb_dec_loop8
 
 	movups	%xmm2,(%rsi)
@@ -713,22 +713,22 @@
 	movups	%xmm9,112(%rsi)
 	pxor	%xmm9,%xmm9
 	leaq	128(%rsi),%rsi
-	addq	$128,%rdx
+	addq	$0x80,%rdx
 	jz	L$ecb_ret
 
 L$ecb_dec_tail:
 	movups	(%rdi),%xmm2
-	cmpq	$32,%rdx
+	cmpq	$0x20,%rdx
 	jb	L$ecb_dec_one
 	movups	16(%rdi),%xmm3
 	je	L$ecb_dec_two
 	movups	32(%rdi),%xmm4
-	cmpq	$64,%rdx
+	cmpq	$0x40,%rdx
 	jb	L$ecb_dec_three
 	movups	48(%rdi),%xmm5
 	je	L$ecb_dec_four
 	movups	64(%rdi),%xmm6
-	cmpq	$96,%rdx
+	cmpq	$0x60,%rdx
 	jb	L$ecb_dec_five
 	movups	80(%rdi),%xmm7
 	je	L$ecb_dec_six
@@ -1606,7 +1606,7 @@
 
 	movdqa	L$xts_magic(%rip),%xmm8
 	movdqa	%xmm2,%xmm15
-	pshufd	$95,%xmm2,%xmm9
+	pshufd	$0x5f,%xmm2,%xmm9
 	pxor	%xmm0,%xmm1
 	movdqa	%xmm9,%xmm14
 	paddd	%xmm9,%xmm9
@@ -1705,7 +1705,7 @@
 .byte	102,15,56,220,248
 	movups	64(%r11),%xmm0
 	movdqa	%xmm8,80(%rsp)
-	pshufd	$95,%xmm15,%xmm9
+	pshufd	$0x5f,%xmm15,%xmm9
 	jmp	L$xts_enc_loop6
 .p2align	5
 L$xts_enc_loop6:
@@ -1844,13 +1844,13 @@
 	jz	L$xts_enc_done
 
 	pxor	%xmm0,%xmm11
-	cmpq	$32,%rdx
+	cmpq	$0x20,%rdx
 	jb	L$xts_enc_one
 	pxor	%xmm0,%xmm12
 	je	L$xts_enc_two
 
 	pxor	%xmm0,%xmm13
-	cmpq	$64,%rdx
+	cmpq	$0x40,%rdx
 	jb	L$xts_enc_three
 	pxor	%xmm0,%xmm14
 	je	L$xts_enc_four
@@ -2078,7 +2078,7 @@
 
 	movdqa	L$xts_magic(%rip),%xmm8
 	movdqa	%xmm2,%xmm15
-	pshufd	$95,%xmm2,%xmm9
+	pshufd	$0x5f,%xmm2,%xmm9
 	pxor	%xmm0,%xmm1
 	movdqa	%xmm9,%xmm14
 	paddd	%xmm9,%xmm9
@@ -2177,7 +2177,7 @@
 .byte	102,15,56,222,248
 	movups	64(%r11),%xmm0
 	movdqa	%xmm8,80(%rsp)
-	pshufd	$95,%xmm15,%xmm9
+	pshufd	$0x5f,%xmm15,%xmm9
 	jmp	L$xts_dec_loop6
 .p2align	5
 L$xts_dec_loop6:
@@ -2317,13 +2317,13 @@
 	jz	L$xts_dec_done
 
 	pxor	%xmm0,%xmm12
-	cmpq	$32,%rdx
+	cmpq	$0x20,%rdx
 	jb	L$xts_dec_one
 	pxor	%xmm0,%xmm13
 	je	L$xts_dec_two
 
 	pxor	%xmm0,%xmm14
-	cmpq	$64,%rdx
+	cmpq	$0x40,%rdx
 	jb	L$xts_dec_three
 	je	L$xts_dec_four
 
@@ -2354,7 +2354,7 @@
 	pcmpgtd	%xmm15,%xmm14
 	movdqu	%xmm6,64(%rsi)
 	leaq	80(%rsi),%rsi
-	pshufd	$19,%xmm14,%xmm11
+	pshufd	$0x13,%xmm14,%xmm11
 	andq	$15,%r9
 	jz	L$xts_dec_ret
 
@@ -2644,7 +2644,7 @@
 	leaq	-8(%rax),%rbp
 	movups	(%r8),%xmm10
 	movl	%r10d,%eax
-	cmpq	$80,%rdx
+	cmpq	$0x50,%rdx
 	jbe	L$cbc_dec_tail
 
 	movups	(%rcx),%xmm0
@@ -2660,14 +2660,14 @@
 	movdqu	80(%rdi),%xmm7
 	movdqa	%xmm6,%xmm15
 	movl	_OPENSSL_ia32cap_P+4(%rip),%r9d
-	cmpq	$112,%rdx
+	cmpq	$0x70,%rdx
 	jbe	L$cbc_dec_six_or_seven
 
 	andl	$71303168,%r9d
-	subq	$80,%rdx
+	subq	$0x50,%rdx
 	cmpl	$4194304,%r9d
 	je	L$cbc_dec_loop6_enter
-	subq	$32,%rdx
+	subq	$0x20,%rdx
 	leaq	112(%rcx),%rcx
 	jmp	L$cbc_dec_loop8_enter
 .p2align	4
@@ -2682,7 +2682,7 @@
 	movups	16-112(%rcx),%xmm1
 	pxor	%xmm0,%xmm4
 	xorq	%r11,%r11
-	cmpq	$112,%rdx
+	cmpq	$0x70,%rdx
 	pxor	%xmm0,%xmm5
 	pxor	%xmm0,%xmm6
 	pxor	%xmm0,%xmm7
@@ -2867,21 +2867,21 @@
 	movups	%xmm8,96(%rsi)
 	leaq	112(%rsi),%rsi
 
-	subq	$128,%rdx
+	subq	$0x80,%rdx
 	ja	L$cbc_dec_loop8
 
 	movaps	%xmm9,%xmm2
 	leaq	-112(%rcx),%rcx
-	addq	$112,%rdx
+	addq	$0x70,%rdx
 	jle	L$cbc_dec_clear_tail_collected
 	movups	%xmm9,(%rsi)
 	leaq	16(%rsi),%rsi
-	cmpq	$80,%rdx
+	cmpq	$0x50,%rdx
 	jbe	L$cbc_dec_tail
 
 	movaps	%xmm11,%xmm2
 L$cbc_dec_six_or_seven:
-	cmpq	$96,%rdx
+	cmpq	$0x60,%rdx
 	ja	L$cbc_dec_seven
 
 	movaps	%xmm7,%xmm8
@@ -2974,33 +2974,33 @@
 	movl	%r10d,%eax
 	movdqu	%xmm6,64(%rsi)
 	leaq	80(%rsi),%rsi
-	subq	$96,%rdx
+	subq	$0x60,%rdx
 	ja	L$cbc_dec_loop6
 
 	movdqa	%xmm7,%xmm2
-	addq	$80,%rdx
+	addq	$0x50,%rdx
 	jle	L$cbc_dec_clear_tail_collected
 	movups	%xmm7,(%rsi)
 	leaq	16(%rsi),%rsi
 
 L$cbc_dec_tail:
 	movups	(%rdi),%xmm2
-	subq	$16,%rdx
+	subq	$0x10,%rdx
 	jbe	L$cbc_dec_one
 
 	movups	16(%rdi),%xmm3
 	movaps	%xmm2,%xmm11
-	subq	$16,%rdx
+	subq	$0x10,%rdx
 	jbe	L$cbc_dec_two
 
 	movups	32(%rdi),%xmm4
 	movaps	%xmm3,%xmm12
-	subq	$16,%rdx
+	subq	$0x10,%rdx
 	jbe	L$cbc_dec_three
 
 	movups	48(%rdi),%xmm5
 	movaps	%xmm4,%xmm13
-	subq	$16,%rdx
+	subq	$0x10,%rdx
 	jbe	L$cbc_dec_four
 
 	movups	64(%rdi),%xmm6
@@ -3025,7 +3025,7 @@
 	movdqa	%xmm6,%xmm2
 	pxor	%xmm6,%xmm6
 	pxor	%xmm7,%xmm7
-	subq	$16,%rdx
+	subq	$0x10,%rdx
 	jmp	L$cbc_dec_tail_collected
 
 .p2align	4
@@ -3344,7 +3344,7 @@
 	pslldq	$4,%xmm0
 	pxor	%xmm3,%xmm0
 
-	pshufd	$255,%xmm0,%xmm3
+	pshufd	$0xff,%xmm0,%xmm3
 	pxor	%xmm1,%xmm3
 	pslldq	$4,%xmm1
 	pxor	%xmm1,%xmm3
@@ -3431,7 +3431,7 @@
 	decl	%r10d
 	jz	L$done_key256
 
-	pshufd	$255,%xmm0,%xmm2
+	pshufd	$0xff,%xmm0,%xmm2
 	pxor	%xmm3,%xmm3
 .byte	102,15,56,221,211
 

third_party/boringssl/mac-x86_64/crypto/aes/bsaes-x86_64.S

diff --git a/third_party/boringssl/mac-x86_64/crypto/aes/bsaes-x86_64.S b/third_party/boringssl/mac-x86_64/crypto/aes/bsaes-x86_64.S
index c2d0477..ad802e3 100644
--- a/third_party/boringssl/mac-x86_64/crypto/aes/bsaes-x86_64.S
+++ b/third_party/boringssl/mac-x86_64/crypto/aes/bsaes-x86_64.S
@@ -325,45 +325,45 @@
 	pxor	%xmm2,%xmm5
 	decl	%r10d
 	jl	L$enc_done
-	pshufd	$147,%xmm15,%xmm7
-	pshufd	$147,%xmm0,%xmm8
+	pshufd	$0x93,%xmm15,%xmm7
+	pshufd	$0x93,%xmm0,%xmm8
 	pxor	%xmm7,%xmm15
-	pshufd	$147,%xmm3,%xmm9
+	pshufd	$0x93,%xmm3,%xmm9
 	pxor	%xmm8,%xmm0
-	pshufd	$147,%xmm5,%xmm10
+	pshufd	$0x93,%xmm5,%xmm10
 	pxor	%xmm9,%xmm3
-	pshufd	$147,%xmm2,%xmm11
+	pshufd	$0x93,%xmm2,%xmm11
 	pxor	%xmm10,%xmm5
-	pshufd	$147,%xmm6,%xmm12
+	pshufd	$0x93,%xmm6,%xmm12
 	pxor	%xmm11,%xmm2
-	pshufd	$147,%xmm1,%xmm13
+	pshufd	$0x93,%xmm1,%xmm13
 	pxor	%xmm12,%xmm6
-	pshufd	$147,%xmm4,%xmm14
+	pshufd	$0x93,%xmm4,%xmm14
 	pxor	%xmm13,%xmm1
 	pxor	%xmm14,%xmm4
 
 	pxor	%xmm15,%xmm8
 	pxor	%xmm4,%xmm7
 	pxor	%xmm4,%xmm8
-	pshufd	$78,%xmm15,%xmm15
+	pshufd	$0x4E,%xmm15,%xmm15
 	pxor	%xmm0,%xmm9
-	pshufd	$78,%xmm0,%xmm0
+	pshufd	$0x4E,%xmm0,%xmm0
 	pxor	%xmm2,%xmm12
 	pxor	%xmm7,%xmm15
 	pxor	%xmm6,%xmm13
 	pxor	%xmm8,%xmm0
 	pxor	%xmm5,%xmm11
-	pshufd	$78,%xmm2,%xmm7
+	pshufd	$0x4E,%xmm2,%xmm7
 	pxor	%xmm1,%xmm14
-	pshufd	$78,%xmm6,%xmm8
+	pshufd	$0x4E,%xmm6,%xmm8
 	pxor	%xmm3,%xmm10
-	pshufd	$78,%xmm5,%xmm2
+	pshufd	$0x4E,%xmm5,%xmm2
 	pxor	%xmm4,%xmm10
-	pshufd	$78,%xmm4,%xmm6
+	pshufd	$0x4E,%xmm4,%xmm6
 	pxor	%xmm4,%xmm11
-	pshufd	$78,%xmm1,%xmm5
+	pshufd	$0x4E,%xmm1,%xmm5
 	pxor	%xmm11,%xmm7
-	pshufd	$78,%xmm3,%xmm1
+	pshufd	$0x4E,%xmm3,%xmm1
 	pxor	%xmm12,%xmm8
 	pxor	%xmm10,%xmm2
 	pxor	%xmm14,%xmm6
@@ -797,24 +797,24 @@
 	decl	%r10d
 	jl	L$dec_done
 
-	pshufd	$78,%xmm15,%xmm7
-	pshufd	$78,%xmm2,%xmm13
+	pshufd	$0x4E,%xmm15,%xmm7
+	pshufd	$0x4E,%xmm2,%xmm13
 	pxor	%xmm15,%xmm7
-	pshufd	$78,%xmm4,%xmm14
+	pshufd	$0x4E,%xmm4,%xmm14
 	pxor	%xmm2,%xmm13
-	pshufd	$78,%xmm0,%xmm8
+	pshufd	$0x4E,%xmm0,%xmm8
 	pxor	%xmm4,%xmm14
-	pshufd	$78,%xmm5,%xmm9
+	pshufd	$0x4E,%xmm5,%xmm9
 	pxor	%xmm0,%xmm8
-	pshufd	$78,%xmm3,%xmm10
+	pshufd	$0x4E,%xmm3,%xmm10
 	pxor	%xmm5,%xmm9
 	pxor	%xmm13,%xmm15
 	pxor	%xmm13,%xmm0
-	pshufd	$78,%xmm1,%xmm11
+	pshufd	$0x4E,%xmm1,%xmm11
 	pxor	%xmm3,%xmm10
 	pxor	%xmm7,%xmm5
 	pxor	%xmm8,%xmm3
-	pshufd	$78,%xmm6,%xmm12
+	pshufd	$0x4E,%xmm6,%xmm12
 	pxor	%xmm1,%xmm11
 	pxor	%xmm14,%xmm0
 	pxor	%xmm9,%xmm1
@@ -828,45 +828,45 @@
 	pxor	%xmm14,%xmm1
 	pxor	%xmm14,%xmm6
 	pxor	%xmm12,%xmm4
-	pshufd	$147,%xmm15,%xmm7
-	pshufd	$147,%xmm0,%xmm8
+	pshufd	$0x93,%xmm15,%xmm7
+	pshufd	$0x93,%xmm0,%xmm8
 	pxor	%xmm7,%xmm15
-	pshufd	$147,%xmm5,%xmm9
+	pshufd	$0x93,%xmm5,%xmm9
 	pxor	%xmm8,%xmm0
-	pshufd	$147,%xmm3,%xmm10
+	pshufd	$0x93,%xmm3,%xmm10
 	pxor	%xmm9,%xmm5
-	pshufd	$147,%xmm1,%xmm11
+	pshufd	$0x93,%xmm1,%xmm11
 	pxor	%xmm10,%xmm3
-	pshufd	$147,%xmm6,%xmm12
+	pshufd	$0x93,%xmm6,%xmm12
 	pxor	%xmm11,%xmm1
-	pshufd	$147,%xmm2,%xmm13
+	pshufd	$0x93,%xmm2,%xmm13
 	pxor	%xmm12,%xmm6
-	pshufd	$147,%xmm4,%xmm14
+	pshufd	$0x93,%xmm4,%xmm14
 	pxor	%xmm13,%xmm2
 	pxor	%xmm14,%xmm4
 
 	pxor	%xmm15,%xmm8
 	pxor	%xmm4,%xmm7
 	pxor	%xmm4,%xmm8
-	pshufd	$78,%xmm15,%xmm15
+	pshufd	$0x4E,%xmm15,%xmm15
 	pxor	%xmm0,%xmm9
-	pshufd	$78,%xmm0,%xmm0
+	pshufd	$0x4E,%xmm0,%xmm0
 	pxor	%xmm1,%xmm12
 	pxor	%xmm7,%xmm15
 	pxor	%xmm6,%xmm13
 	pxor	%xmm8,%xmm0
 	pxor	%xmm3,%xmm11
-	pshufd	$78,%xmm1,%xmm7
+	pshufd	$0x4E,%xmm1,%xmm7
 	pxor	%xmm2,%xmm14
-	pshufd	$78,%xmm6,%xmm8
+	pshufd	$0x4E,%xmm6,%xmm8
 	pxor	%xmm5,%xmm10
-	pshufd	$78,%xmm3,%xmm1
+	pshufd	$0x4E,%xmm3,%xmm1
 	pxor	%xmm4,%xmm10
-	pshufd	$78,%xmm4,%xmm6
+	pshufd	$0x4E,%xmm4,%xmm6
 	pxor	%xmm4,%xmm11
-	pshufd	$78,%xmm2,%xmm3
+	pshufd	$0x4E,%xmm2,%xmm3
 	pxor	%xmm11,%xmm7
-	pshufd	$78,%xmm5,%xmm2
+	pshufd	$0x4E,%xmm5,%xmm2
 	pxor	%xmm12,%xmm8
 	pxor	%xmm1,%xmm10
 	pxor	%xmm14,%xmm6
@@ -1556,20 +1556,20 @@
 	movdqa	%xmm7,(%rax)
 
 	andq	$-16,%r14
-	subq	$128,%rsp
+	subq	$0x80,%rsp
 	movdqa	32(%rbp),%xmm6
 
 	pxor	%xmm14,%xmm14
 	movdqa	L$xts_magic(%rip),%xmm12
 	pcmpgtd	%xmm6,%xmm14
 
-	subq	$128,%r14
+	subq	$0x80,%r14
 	jc	L$xts_enc_short
 	jmp	L$xts_enc_loop
 
 .p2align	4
 L$xts_enc_loop:
-	pshufd	$19,%xmm14,%xmm13
+	pshufd	$0x13,%xmm14,%xmm13
 	pxor	%xmm14,%xmm14
 	movdqa	%xmm6,%xmm15
 	movdqa	%xmm6,0(%rsp)
@@ -1577,7 +1577,7 @@
 	pand	%xmm12,%xmm13
 	pcmpgtd	%xmm6,%xmm14
 	pxor	%xmm13,%xmm6
-	pshufd	$19,%xmm14,%xmm13
+	pshufd	$0x13,%xmm14,%xmm13
 	pxor	%xmm14,%xmm14
 	movdqa	%xmm6,%xmm0
 	movdqa	%xmm6,16(%rsp)
@@ -1586,7 +1586,7 @@
 	pcmpgtd	%xmm6,%xmm14
 	pxor	%xmm13,%xmm6
 	movdqu	0(%r12),%xmm7
-	pshufd	$19,%xmm14,%xmm13
+	pshufd	$0x13,%xmm14,%xmm13
 	pxor	%xmm14,%xmm14
 	movdqa	%xmm6,%xmm1
 	movdqa	%xmm6,32(%rsp)
@@ -1596,7 +1596,7 @@
 	pxor	%xmm13,%xmm6
 	movdqu	16(%r12),%xmm8
 	pxor	%xmm7,%xmm15
-	pshufd	$19,%xmm14,%xmm13
+	pshufd	$0x13,%xmm14,%xmm13
 	pxor	%xmm14,%xmm14
 	movdqa	%xmm6,%xmm2
 	movdqa	%xmm6,48(%rsp)
@@ -1606,7 +1606,7 @@
 	pxor	%xmm13,%xmm6
 	movdqu	32(%r12),%xmm9
 	pxor	%xmm8,%xmm0
-	pshufd	$19,%xmm14,%xmm13
+	pshufd	$0x13,%xmm14,%xmm13
 	pxor	%xmm14,%xmm14
 	movdqa	%xmm6,%xmm3
 	movdqa	%xmm6,64(%rsp)
@@ -1616,7 +1616,7 @@
 	pxor	%xmm13,%xmm6
 	movdqu	48(%r12),%xmm10
 	pxor	%xmm9,%xmm1
-	pshufd	$19,%xmm14,%xmm13
+	pshufd	$0x13,%xmm14,%xmm13
 	pxor	%xmm14,%xmm14
 	movdqa	%xmm6,%xmm4
 	movdqa	%xmm6,80(%rsp)
@@ -1626,7 +1626,7 @@
 	pxor	%xmm13,%xmm6
 	movdqu	64(%r12),%xmm11
 	pxor	%xmm10,%xmm2
-	pshufd	$19,%xmm14,%xmm13
+	pshufd	$0x13,%xmm14,%xmm13
 	pxor	%xmm14,%xmm14
 	movdqa	%xmm6,%xmm5
 	movdqa	%xmm6,96(%rsp)
@@ -1670,20 +1670,20 @@
 	pxor	%xmm14,%xmm14
 	movdqa	L$xts_magic(%rip),%xmm12
 	pcmpgtd	%xmm6,%xmm14
-	pshufd	$19,%xmm14,%xmm13
+	pshufd	$0x13,%xmm14,%xmm13
 	pxor	%xmm14,%xmm14
 	paddq	%xmm6,%xmm6
 	pand	%xmm12,%xmm13
 	pcmpgtd	%xmm6,%xmm14
 	pxor	%xmm13,%xmm6
 
-	subq	$128,%r14
+	subq	$0x80,%r14
 	jnc	L$xts_enc_loop
 
 L$xts_enc_short:
-	addq	$128,%r14
+	addq	$0x80,%r14
 	jz	L$xts_enc_done
-	pshufd	$19,%xmm14,%xmm13
+	pshufd	$0x13,%xmm14,%xmm13
 	pxor	%xmm14,%xmm14
 	movdqa	%xmm6,%xmm15
 	movdqa	%xmm6,0(%rsp)
@@ -1691,7 +1691,7 @@
 	pand	%xmm12,%xmm13
 	pcmpgtd	%xmm6,%xmm14
 	pxor	%xmm13,%xmm6
-	pshufd	$19,%xmm14,%xmm13
+	pshufd	$0x13,%xmm14,%xmm13
 	pxor	%xmm14,%xmm14
 	movdqa	%xmm6,%xmm0
 	movdqa	%xmm6,16(%rsp)
@@ -1702,7 +1702,7 @@
 	movdqu	0(%r12),%xmm7
 	cmpq	$16,%r14
 	je	L$xts_enc_1
-	pshufd	$19,%xmm14,%xmm13
+	pshufd	$0x13,%xmm14,%xmm13
 	pxor	%xmm14,%xmm14
 	movdqa	%xmm6,%xmm1
 	movdqa	%xmm6,32(%rsp)
@@ -1714,7 +1714,7 @@
 	cmpq	$32,%r14
 	je	L$xts_enc_2
 	pxor	%xmm7,%xmm15
-	pshufd	$19,%xmm14,%xmm13
+	pshufd	$0x13,%xmm14,%xmm13
 	pxor	%xmm14,%xmm14
 	movdqa	%xmm6,%xmm2
 	movdqa	%xmm6,48(%rsp)
@@ -1726,7 +1726,7 @@
 	cmpq	$48,%r14
 	je	L$xts_enc_3
 	pxor	%xmm8,%xmm0
-	pshufd	$19,%xmm14,%xmm13
+	pshufd	$0x13,%xmm14,%xmm13
 	pxor	%xmm14,%xmm14
 	movdqa	%xmm6,%xmm3
 	movdqa	%xmm6,64(%rsp)
@@ -1738,7 +1738,7 @@
 	cmpq	$64,%r14
 	je	L$xts_enc_4
 	pxor	%xmm9,%xmm1
-	pshufd	$19,%xmm14,%xmm13
+	pshufd	$0x13,%xmm14,%xmm13
 	pxor	%xmm14,%xmm14
 	movdqa	%xmm6,%xmm4
 	movdqa	%xmm6,80(%rsp)
@@ -1750,7 +1750,7 @@
 	cmpq	$80,%r14
 	je	L$xts_enc_5
 	pxor	%xmm10,%xmm2
-	pshufd	$19,%xmm14,%xmm13
+	pshufd	$0x13,%xmm14,%xmm13
 	pxor	%xmm14,%xmm14
 	movdqa	%xmm6,%xmm5
 	movdqa	%xmm6,96(%rsp)
@@ -2016,20 +2016,20 @@
 	shlq	$4,%rax
 	subq	%rax,%r14
 
-	subq	$128,%rsp
+	subq	$0x80,%rsp
 	movdqa	32(%rbp),%xmm6
 
 	pxor	%xmm14,%xmm14
 	movdqa	L$xts_magic(%rip),%xmm12
 	pcmpgtd	%xmm6,%xmm14
 
-	subq	$128,%r14
+	subq	$0x80,%r14
 	jc	L$xts_dec_short
 	jmp	L$xts_dec_loop
 
 .p2align	4
 L$xts_dec_loop:
-	pshufd	$19,%xmm14,%xmm13
+	pshufd	$0x13,%xmm14,%xmm13
 	pxor	%xmm14,%xmm14
 	movdqa	%xmm6,%xmm15
 	movdqa	%xmm6,0(%rsp)
@@ -2037,7 +2037,7 @@
 	pand	%xmm12,%xmm13
 	pcmpgtd	%xmm6,%xmm14
 	pxor	%xmm13,%xmm6
-	pshufd	$19,%xmm14,%xmm13
+	pshufd	$0x13,%xmm14,%xmm13
 	pxor	%xmm14,%xmm14
 	movdqa	%xmm6,%xmm0
 	movdqa	%xmm6,16(%rsp)
@@ -2046,7 +2046,7 @@
 	pcmpgtd	%xmm6,%xmm14
 	pxor	%xmm13,%xmm6
 	movdqu	0(%r12),%xmm7
-	pshufd	$19,%xmm14,%xmm13
+	pshufd	$0x13,%xmm14,%xmm13
 	pxor	%xmm14,%xmm14
 	movdqa	%xmm6,%xmm1
 	movdqa	%xmm6,32(%rsp)
@@ -2056,7 +2056,7 @@
 	pxor	%xmm13,%xmm6
 	movdqu	16(%r12),%xmm8
 	pxor	%xmm7,%xmm15
-	pshufd	$19,%xmm14,%xmm13
+	pshufd	$0x13,%xmm14,%xmm13
 	pxor	%xmm14,%xmm14
 	movdqa	%xmm6,%xmm2
 	movdqa	%xmm6,48(%rsp)
@@ -2066,7 +2066,7 @@
 	pxor	%xmm13,%xmm6
 	movdqu	32(%r12),%xmm9
 	pxor	%xmm8,%xmm0
-	pshufd	$19,%xmm14,%xmm13
+	pshufd	$0x13,%xmm14,%xmm13
 	pxor	%xmm14,%xmm14
 	movdqa	%xmm6,%xmm3
 	movdqa	%xmm6,64(%rsp)
@@ -2076,7 +2076,7 @@
 	pxor	%xmm13,%xmm6
 	movdqu	48(%r12),%xmm10
 	pxor	%xmm9,%xmm1
-	pshufd	$19,%xmm14,%xmm13
+	pshufd	$0x13,%xmm14,%xmm13
 	pxor	%xmm14,%xmm14
 	movdqa	%xmm6,%xmm4
 	movdqa	%xmm6,80(%rsp)
@@ -2086,7 +2086,7 @@
 	pxor	%xmm13,%xmm6
 	movdqu	64(%r12),%xmm11
 	pxor	%xmm10,%xmm2
-	pshufd	$19,%xmm14,%xmm13
+	pshufd	$0x13,%xmm14,%xmm13
 	pxor	%xmm14,%xmm14
 	movdqa	%xmm6,%xmm5
 	movdqa	%xmm6,96(%rsp)
@@ -2130,20 +2130,20 @@
 	pxor	%xmm14,%xmm14
 	movdqa	L$xts_magic(%rip),%xmm12
 	pcmpgtd	%xmm6,%xmm14
-	pshufd	$19,%xmm14,%xmm13
+	pshufd	$0x13,%xmm14,%xmm13
 	pxor	%xmm14,%xmm14
 	paddq	%xmm6,%xmm6
 	pand	%xmm12,%xmm13
 	pcmpgtd	%xmm6,%xmm14
 	pxor	%xmm13,%xmm6
 
-	subq	$128,%r14
+	subq	$0x80,%r14
 	jnc	L$xts_dec_loop
 
 L$xts_dec_short:
-	addq	$128,%r14
+	addq	$0x80,%r14
 	jz	L$xts_dec_done
-	pshufd	$19,%xmm14,%xmm13
+	pshufd	$0x13,%xmm14,%xmm13
 	pxor	%xmm14,%xmm14
 	movdqa	%xmm6,%xmm15
 	movdqa	%xmm6,0(%rsp)
@@ -2151,7 +2151,7 @@
 	pand	%xmm12,%xmm13
 	pcmpgtd	%xmm6,%xmm14
 	pxor	%xmm13,%xmm6
-	pshufd	$19,%xmm14,%xmm13
+	pshufd	$0x13,%xmm14,%xmm13
 	pxor	%xmm14,%xmm14
 	movdqa	%xmm6,%xmm0
 	movdqa	%xmm6,16(%rsp)
@@ -2162,7 +2162,7 @@
 	movdqu	0(%r12),%xmm7
 	cmpq	$16,%r14
 	je	L$xts_dec_1
-	pshufd	$19,%xmm14,%xmm13
+	pshufd	$0x13,%xmm14,%xmm13
 	pxor	%xmm14,%xmm14
 	movdqa	%xmm6,%xmm1
 	movdqa	%xmm6,32(%rsp)
@@ -2174,7 +2174,7 @@
 	cmpq	$32,%r14
 	je	L$xts_dec_2
 	pxor	%xmm7,%xmm15
-	pshufd	$19,%xmm14,%xmm13
+	pshufd	$0x13,%xmm14,%xmm13
 	pxor	%xmm14,%xmm14
 	movdqa	%xmm6,%xmm2
 	movdqa	%xmm6,48(%rsp)
@@ -2186,7 +2186,7 @@
 	cmpq	$48,%r14
 	je	L$xts_dec_3
 	pxor	%xmm8,%xmm0
-	pshufd	$19,%xmm14,%xmm13
+	pshufd	$0x13,%xmm14,%xmm13
 	pxor	%xmm14,%xmm14
 	movdqa	%xmm6,%xmm3
 	movdqa	%xmm6,64(%rsp)
@@ -2198,7 +2198,7 @@
 	cmpq	$64,%r14
 	je	L$xts_dec_4
 	pxor	%xmm9,%xmm1
-	pshufd	$19,%xmm14,%xmm13
+	pshufd	$0x13,%xmm14,%xmm13
 	pxor	%xmm14,%xmm14
 	movdqa	%xmm6,%xmm4
 	movdqa	%xmm6,80(%rsp)
@@ -2210,7 +2210,7 @@
 	cmpq	$80,%r14
 	je	L$xts_dec_5
 	pxor	%xmm10,%xmm2
-	pshufd	$19,%xmm14,%xmm13
+	pshufd	$0x13,%xmm14,%xmm13
 	pxor	%xmm14,%xmm14
 	movdqa	%xmm6,%xmm5
 	movdqa	%xmm6,96(%rsp)
@@ -2387,7 +2387,7 @@
 	pxor	%xmm14,%xmm14
 	movdqa	L$xts_magic(%rip),%xmm12
 	pcmpgtd	%xmm6,%xmm14
-	pshufd	$19,%xmm14,%xmm13
+	pshufd	$0x13,%xmm14,%xmm13
 	movdqa	%xmm6,%xmm5
 	paddq	%xmm6,%xmm6
 	pand	%xmm12,%xmm13

third_party/boringssl/mac-x86_64/crypto/aes/vpaes-x86_64.S

diff --git a/third_party/boringssl/mac-x86_64/crypto/aes/vpaes-x86_64.S b/third_party/boringssl/mac-x86_64/crypto/aes/vpaes-x86_64.S
index 711ea43..997cde8 100644
--- a/third_party/boringssl/mac-x86_64/crypto/aes/vpaes-x86_64.S
+++ b/third_party/boringssl/mac-x86_64/crypto/aes/vpaes-x86_64.S
@@ -61,7 +61,7 @@
 	addq	$16,%r11
 	pxor	%xmm0,%xmm3
 .byte	102,15,56,0,193
-	andq	$48,%r11
+	andq	$0x30,%r11
 	subq	$1,%rax
 	pxor	%xmm3,%xmm0
 
@@ -121,10 +121,10 @@
 	pand	%xmm9,%xmm0
 .byte	102,15,56,0,208
 	movdqa	L$k_dipt+16(%rip),%xmm0
-	xorq	$48,%r11
+	xorq	$0x30,%r11
 	leaq	L$k_dsbd(%rip),%r10
 .byte	102,15,56,0,193
-	andq	$48,%r11
+	andq	$0x30,%r11
 	pxor	%xmm5,%xmm2
 	movdqa	L$k_mc_forward+48(%rip),%xmm5
 	pxor	%xmm2,%xmm0
@@ -243,7 +243,7 @@
 	movdqa	(%r8,%r10,1),%xmm1
 .byte	102,15,56,0,217
 	movdqu	%xmm3,(%rdx)
-	xorq	$48,%r8
+	xorq	$0x30,%r8
 
 L$schedule_go:
 	cmpl	$192,%esi
@@ -333,7 +333,7 @@
 	call	_vpaes_schedule_mangle
 
 
-	pshufd	$255,%xmm0,%xmm0
+	pshufd	$0xFF,%xmm0,%xmm0
 	movdqa	%xmm7,%xmm5
 	movdqa	%xmm6,%xmm7
 	call	_vpaes_schedule_low_round
@@ -400,8 +400,8 @@
 
 .p2align	4
 _vpaes_schedule_192_smear:
-	pshufd	$128,%xmm6,%xmm1
-	pshufd	$254,%xmm7,%xmm0
+	pshufd	$0x80,%xmm6,%xmm1
+	pshufd	$0xFE,%xmm7,%xmm0
 	pxor	%xmm1,%xmm6
 	pxor	%xmm1,%xmm1
 	pxor	%xmm0,%xmm6
@@ -438,7 +438,7 @@
 	pxor	%xmm1,%xmm7
 
 
-	pshufd	$255,%xmm0,%xmm0
+	pshufd	$0xFF,%xmm0,%xmm0
 .byte	102,15,58,15,192,1
 
 
@@ -597,7 +597,7 @@
 	movdqa	(%r8,%r10,1),%xmm1
 .byte	102,15,56,0,217
 	addq	$-16,%r8
-	andq	$48,%r8
+	andq	$0x30,%r8
 	movdqu	%xmm3,(%rdx)
 	.byte	0xf3,0xc3
 
@@ -616,7 +616,7 @@
 	movl	%eax,240(%rdx)
 
 	movl	$0,%ecx
-	movl	$48,%r8d
+	movl	$0x30,%r8d
 	call	_vpaes_schedule_core
 	xorl	%eax,%eax
 	.byte	0xf3,0xc3

third_party/boringssl/mac-x86_64/crypto/bn/rsaz-x86_64.S

diff --git a/third_party/boringssl/mac-x86_64/crypto/bn/rsaz-x86_64.S b/third_party/boringssl/mac-x86_64/crypto/bn/rsaz-x86_64.S
index 5e9e82f..337276f 100644
--- a/third_party/boringssl/mac-x86_64/crypto/bn/rsaz-x86_64.S
+++ b/third_party/boringssl/mac-x86_64/crypto/bn/rsaz-x86_64.S
@@ -465,48 +465,94 @@
 	pushq	%r14
 	pushq	%r15
 
-	movl	%r9d,%r9d
-	subq	$128+24,%rsp
+	subq	$152,%rsp
 L$mul_gather4_body:
-	movl	64(%rdx,%r9,4),%eax
-.byte	102,72,15,110,199
-	movl	(%rdx,%r9,4),%ebx
-.byte	102,72,15,110,201
-	movq	%r8,128(%rsp)
+	movd	%r9d,%xmm8
+	movdqa	L$inc+16(%rip),%xmm1
+	movdqa	L$inc(%rip),%xmm0
 
-	shlq	$32,%rax
-	orq	%rax,%rbx
+	pshufd	$0,%xmm8,%xmm8
+	movdqa	%xmm1,%xmm7
+	movdqa	%xmm1,%xmm2
+	paddd	%xmm0,%xmm1
+	pcmpeqd	%xmm8,%xmm0
+	movdqa	%xmm7,%xmm3
+	paddd	%xmm1,%xmm2
+	pcmpeqd	%xmm8,%xmm1
+	movdqa	%xmm7,%xmm4
+	paddd	%xmm2,%xmm3
+	pcmpeqd	%xmm8,%xmm2
+	movdqa	%xmm7,%xmm5
+	paddd	%xmm3,%xmm4
+	pcmpeqd	%xmm8,%xmm3
+	movdqa	%xmm7,%xmm6
+	paddd	%xmm4,%xmm5
+	pcmpeqd	%xmm8,%xmm4
+	paddd	%xmm5,%xmm6
+	pcmpeqd	%xmm8,%xmm5
+	paddd	%xmm6,%xmm7
+	pcmpeqd	%xmm8,%xmm6
+	pcmpeqd	%xmm8,%xmm7
+
+	movdqa	0(%rdx),%xmm8
+	movdqa	16(%rdx),%xmm9
+	movdqa	32(%rdx),%xmm10
+	movdqa	48(%rdx),%xmm11
+	pand	%xmm0,%xmm8
+	movdqa	64(%rdx),%xmm12
+	pand	%xmm1,%xmm9
+	movdqa	80(%rdx),%xmm13
+	pand	%xmm2,%xmm10
+	movdqa	96(%rdx),%xmm14
+	pand	%xmm3,%xmm11
+	movdqa	112(%rdx),%xmm15
+	leaq	128(%rdx),%rbp
+	pand	%xmm4,%xmm12
+	pand	%xmm5,%xmm13
+	pand	%xmm6,%xmm14
+	pand	%xmm7,%xmm15
+	por	%xmm10,%xmm8
+	por	%xmm11,%xmm9
+	por	%xmm12,%xmm8
+	por	%xmm13,%xmm9
+	por	%xmm14,%xmm8
+	por	%xmm15,%xmm9
+
+	por	%xmm9,%xmm8
+	pshufd	$0x4e,%xmm8,%xmm9
+	por	%xmm9,%xmm8
+.byte	102,76,15,126,195
+
+	movq	%r8,128(%rsp)
+	movq	%rdi,128+8(%rsp)
+	movq	%rcx,128+16(%rsp)
+
 	movq	(%rsi),%rax
 	movq	8(%rsi),%rcx
-	leaq	128(%rdx,%r9,4),%rbp
 	mulq	%rbx
 	movq	%rax,(%rsp)
 	movq	%rcx,%rax
 	movq	%rdx,%r8
 
 	mulq	%rbx
-	movd	(%rbp),%xmm4
 	addq	%rax,%r8
 	movq	16(%rsi),%rax
 	movq	%rdx,%r9
 	adcq	$0,%r9
 
 	mulq	%rbx
-	movd	64(%rbp),%xmm5
 	addq	%rax,%r9
 	movq	24(%rsi),%rax
 	movq	%rdx,%r10
 	adcq	$0,%r10
 
 	mulq	%rbx
-	pslldq	$4,%xmm5
 	addq	%rax,%r10
 	movq	32(%rsi),%rax
 	movq	%rdx,%r11
 	adcq	$0,%r11
 
 	mulq	%rbx
-	por	%xmm5,%xmm4
 	addq	%rax,%r11
 	movq	40(%rsi),%rax
 	movq	%rdx,%r12
@@ -519,14 +565,12 @@
 	adcq	$0,%r13
 
 	mulq	%rbx
-	leaq	128(%rbp),%rbp
 	addq	%rax,%r13
 	movq	56(%rsi),%rax
 	movq	%rdx,%r14
 	adcq	$0,%r14
 
 	mulq	%rbx
-.byte	102,72,15,126,227
 	addq	%rax,%r14
 	movq	(%rsi),%rax
 	movq	%rdx,%r15
@@ -538,6 +582,35 @@
 
 .p2align	5
 L$oop_mul_gather:
+	movdqa	0(%rbp),%xmm8
+	movdqa	16(%rbp),%xmm9
+	movdqa	32(%rbp),%xmm10
+	movdqa	48(%rbp),%xmm11
+	pand	%xmm0,%xmm8
+	movdqa	64(%rbp),%xmm12
+	pand	%xmm1,%xmm9
+	movdqa	80(%rbp),%xmm13
+	pand	%xmm2,%xmm10
+	movdqa	96(%rbp),%xmm14
+	pand	%xmm3,%xmm11
+	movdqa	112(%rbp),%xmm15
+	leaq	128(%rbp),%rbp
+	pand	%xmm4,%xmm12
+	pand	%xmm5,%xmm13
+	pand	%xmm6,%xmm14
+	pand	%xmm7,%xmm15
+	por	%xmm10,%xmm8
+	por	%xmm11,%xmm9
+	por	%xmm12,%xmm8
+	por	%xmm13,%xmm9
+	por	%xmm14,%xmm8
+	por	%xmm15,%xmm9
+
+	por	%xmm9,%xmm8
+	pshufd	$0x4e,%xmm8,%xmm9
+	por	%xmm9,%xmm8
+.byte	102,76,15,126,195
+
 	mulq	%rbx
 	addq	%rax,%r8
 	movq	8(%rsi),%rax
@@ -546,7 +619,6 @@
 	adcq	$0,%r8
 
 	mulq	%rbx
-	movd	(%rbp),%xmm4
 	addq	%rax,%r9
 	movq	16(%rsi),%rax
 	adcq	$0,%rdx
@@ -555,7 +627,6 @@
 	adcq	$0,%r9
 
 	mulq	%rbx
-	movd	64(%rbp),%xmm5
 	addq	%rax,%r10
 	movq	24(%rsi),%rax
 	adcq	$0,%rdx
@@ -564,7 +635,6 @@
 	adcq	$0,%r10
 
 	mulq	%rbx
-	pslldq	$4,%xmm5
 	addq	%rax,%r11
 	movq	32(%rsi),%rax
 	adcq	$0,%rdx
@@ -573,7 +643,6 @@
 	adcq	$0,%r11
 
 	mulq	%rbx
-	por	%xmm5,%xmm4
 	addq	%rax,%r12
 	movq	40(%rsi),%rax
 	adcq	$0,%rdx
@@ -598,7 +667,6 @@
 	adcq	$0,%r14
 
 	mulq	%rbx
-.byte	102,72,15,126,227
 	addq	%rax,%r15
 	movq	(%rsi),%rax
 	adcq	$0,%rdx
@@ -606,7 +674,6 @@
 	movq	%rdx,%r15
 	adcq	$0,%r15
 
-	leaq	128(%rbp),%rbp
 	leaq	8(%rdi),%rdi
 
 	decl	%ecx
@@ -621,8 +688,8 @@
 	movq	%r14,48(%rdi)
 	movq	%r15,56(%rdi)
 
-.byte	102,72,15,126,199
-.byte	102,72,15,126,205
+	movq	128+8(%rsp),%rdi
+	movq	128+16(%rsp),%rbp
 
 	movq	(%rsp),%r8
 	movq	8(%rsp),%r9
@@ -672,7 +739,7 @@
 	movl	%r9d,%r9d
 	subq	$128+24,%rsp
 L$mul_scatter4_body:
-	leaq	(%r8,%r9,4),%r8
+	leaq	(%r8,%r9,8),%r8
 .byte	102,72,15,110,199
 .byte	102,72,15,110,202
 .byte	102,73,15,110,208
@@ -708,30 +775,14 @@
 
 	call	__rsaz_512_subtract
 
-	movl	%r8d,0(%rsi)
-	shrq	$32,%r8
-	movl	%r9d,128(%rsi)
-	shrq	$32,%r9
-	movl	%r10d,256(%rsi)
-	shrq	$32,%r10
-	movl	%r11d,384(%rsi)
-	shrq	$32,%r11
-	movl	%r12d,512(%rsi)
-	shrq	$32,%r12
-	movl	%r13d,640(%rsi)
-	shrq	$32,%r13
-	movl	%r14d,768(%rsi)
-	shrq	$32,%r14
-	movl	%r15d,896(%rsi)
-	shrq	$32,%r15
-	movl	%r8d,64(%rsi)
-	movl	%r9d,192(%rsi)
-	movl	%r10d,320(%rsi)
-	movl	%r11d,448(%rsi)
-	movl	%r12d,576(%rsi)
-	movl	%r13d,704(%rsi)
-	movl	%r14d,832(%rsi)
-	movl	%r15d,960(%rsi)
+	movq	%r8,0(%rsi)
+	movq	%r9,128(%rsi)
+	movq	%r10,256(%rsi)
+	movq	%r11,384(%rsi)
+	movq	%r12,512(%rsi)
+	movq	%r13,640(%rsi)
+	movq	%r14,768(%rsi)
+	movq	%r15,896(%rsi)
 
 	leaq	128+24+48(%rsp),%rax
 	movq	-48(%rax),%r15
@@ -1086,16 +1137,14 @@
 
 .p2align	4
 _rsaz_512_scatter4:
-	leaq	(%rdi,%rdx,4),%rdi
+	leaq	(%rdi,%rdx,8),%rdi
 	movl	$8,%r9d
 	jmp	L$oop_scatter
 .p2align	4
 L$oop_scatter:
 	movq	(%rsi),%rax
 	leaq	8(%rsi),%rsi
-	movl	%eax,(%rdi)
-	shrq	$32,%rax
-	movl	%eax,64(%rdi)
+	movq	%rax,(%rdi)
 	leaq	128(%rdi),%rdi
 	decl	%r9d
 	jnz	L$oop_scatter
@@ -1107,20 +1156,73 @@
 
 .p2align	4
 _rsaz_512_gather4:
-	leaq	(%rsi,%rdx,4),%rsi
+	movd	%edx,%xmm8
+	movdqa	L$inc+16(%rip),%xmm1
+	movdqa	L$inc(%rip),%xmm0
+
+	pshufd	$0,%xmm8,%xmm8
+	movdqa	%xmm1,%xmm7
+	movdqa	%xmm1,%xmm2
+	paddd	%xmm0,%xmm1
+	pcmpeqd	%xmm8,%xmm0
+	movdqa	%xmm7,%xmm3
+	paddd	%xmm1,%xmm2
+	pcmpeqd	%xmm8,%xmm1
+	movdqa	%xmm7,%xmm4
+	paddd	%xmm2,%xmm3
+	pcmpeqd	%xmm8,%xmm2
+	movdqa	%xmm7,%xmm5
+	paddd	%xmm3,%xmm4
+	pcmpeqd	%xmm8,%xmm3
+	movdqa	%xmm7,%xmm6
+	paddd	%xmm4,%xmm5
+	pcmpeqd	%xmm8,%xmm4
+	paddd	%xmm5,%xmm6
+	pcmpeqd	%xmm8,%xmm5
+	paddd	%xmm6,%xmm7
+	pcmpeqd	%xmm8,%xmm6
+	pcmpeqd	%xmm8,%xmm7
 	movl	$8,%r9d
 	jmp	L$oop_gather
 .p2align	4
 L$oop_gather:
-	movl	(%rsi),%eax
-	movl	64(%rsi),%r8d
+	movdqa	0(%rsi),%xmm8
+	movdqa	16(%rsi),%xmm9
+	movdqa	32(%rsi),%xmm10
+	movdqa	48(%rsi),%xmm11
+	pand	%xmm0,%xmm8
+	movdqa	64(%rsi),%xmm12
+	pand	%xmm1,%xmm9
+	movdqa	80(%rsi),%xmm13
+	pand	%xmm2,%xmm10
+	movdqa	96(%rsi),%xmm14
+	pand	%xmm3,%xmm11
+	movdqa	112(%rsi),%xmm15
 	leaq	128(%rsi),%rsi
-	shlq	$32,%r8
-	orq	%r8,%rax
-	movq	%rax,(%rdi)
+	pand	%xmm4,%xmm12
+	pand	%xmm5,%xmm13
+	pand	%xmm6,%xmm14
+	pand	%xmm7,%xmm15
+	por	%xmm10,%xmm8
+	por	%xmm11,%xmm9
+	por	%xmm12,%xmm8
+	por	%xmm13,%xmm9
+	por	%xmm14,%xmm8
+	por	%xmm15,%xmm9
+
+	por	%xmm9,%xmm8
+	pshufd	$0x4e,%xmm8,%xmm9
+	por	%xmm9,%xmm8
+	movq	%xmm8,(%rdi)
 	leaq	8(%rdi),%rdi
 	decl	%r9d
 	jnz	L$oop_gather
 	.byte	0xf3,0xc3
+L$SEH_end_rsaz_512_gather4:
 
+
+.p2align	6
+L$inc:
+.long	0,0, 1,1
+.long	2,2, 2,2
 #endif

third_party/boringssl/mac-x86_64/crypto/bn/x86_64-mont.S

diff --git a/third_party/boringssl/mac-x86_64/crypto/bn/x86_64-mont.S b/third_party/boringssl/mac-x86_64/crypto/bn/x86_64-mont.S
index 6b9bc05..51e5d19 100644
--- a/third_party/boringssl/mac-x86_64/crypto/bn/x86_64-mont.S
+++ b/third_party/boringssl/mac-x86_64/crypto/bn/x86_64-mont.S
@@ -634,20 +634,20 @@
 
 
 
-	leaq	-64(%rsp,%r9,4),%r11
+	leaq	-64(%rsp,%r9,2),%r11
 	movq	(%r8),%r8
 	subq	%rsi,%r11
 	andq	$4095,%r11
 	cmpq	%r11,%r10
 	jb	L$sqr8x_sp_alt
 	subq	%r11,%rsp
-	leaq	-64(%rsp,%r9,4),%rsp
+	leaq	-64(%rsp,%r9,2),%rsp
 	jmp	L$sqr8x_sp_done
 
 .p2align	5
 L$sqr8x_sp_alt:
-	leaq	4096-64(,%r9,4),%r10
-	leaq	-64(%rsp,%r9,4),%rsp
+	leaq	4096-64(,%r9,2),%r10
+	leaq	-64(%rsp,%r9,2),%rsp
 	subq	%r10,%r11
 	movq	$0,%r10
 	cmovcq	%r10,%r11
@@ -657,58 +657,80 @@
 	movq	%r9,%r10
 	negq	%r9
 
-	leaq	64(%rsp,%r9,2),%r11
 	movq	%r8,32(%rsp)
 	movq	%rax,40(%rsp)
 L$sqr8x_body:
 
-	movq	%r9,%rbp
-.byte	102,73,15,110,211
-	shrq	$3+2,%rbp
-	movl	_OPENSSL_ia32cap_P+8(%rip),%eax
-	jmp	L$sqr8x_copy_n
-
-.p2align	5
-L$sqr8x_copy_n:
-	movq	0(%rcx),%xmm0
-	movq	8(%rcx),%xmm1
-	movq	16(%rcx),%xmm3
-	movq	24(%rcx),%xmm4
-	leaq	32(%rcx),%rcx
-	movdqa	%xmm0,0(%r11)
-	movdqa	%xmm1,16(%r11)
-	movdqa	%xmm3,32(%r11)
-	movdqa	%xmm4,48(%r11)
-	leaq	64(%r11),%r11
-	decq	%rbp
-	jnz	L$sqr8x_copy_n
-
+.byte	102,72,15,110,209
 	pxor	%xmm0,%xmm0
 .byte	102,72,15,110,207
 .byte	102,73,15,110,218
 	call	_bn_sqr8x_internal
 
-	pxor	%xmm0,%xmm0
-	leaq	48(%rsp),%rax
-	leaq	64(%rsp,%r9,2),%rdx
-	shrq	$3+2,%r9
-	movq	40(%rsp),%rsi
-	jmp	L$sqr8x_zero
+
+
+
+	leaq	(%rdi,%r9,1),%rbx
+	movq	%r9,%rcx
+	movq	%r9,%rdx
+.byte	102,72,15,126,207
+	sarq	$3+2,%rcx
+	jmp	L$sqr8x_sub
 
 .p2align	5
-L$sqr8x_zero:
-	movdqa	%xmm0,0(%rax)
-	movdqa	%xmm0,16(%rax)
-	movdqa	%xmm0,32(%rax)
-	movdqa	%xmm0,48(%rax)
-	leaq	64(%rax),%rax
-	movdqa	%xmm0,0(%rdx)
-	movdqa	%xmm0,16(%rdx)
-	movdqa	%xmm0,32(%rdx)
-	movdqa	%xmm0,48(%rdx)
-	leaq	64(%rdx),%rdx
-	decq	%r9
-	jnz	L$sqr8x_zero
+L$sqr8x_sub:
+	movq	0(%rbx),%r12
+	movq	8(%rbx),%r13
+	movq	16(%rbx),%r14
+	movq	24(%rbx),%r15
+	leaq	32(%rbx),%rbx
+	sbbq	0(%rbp),%r12
+	sbbq	8(%rbp),%r13
+	sbbq	16(%rbp),%r14
+	sbbq	24(%rbp),%r15
+	leaq	32(%rbp),%rbp
+	movq	%r12,0(%rdi)
+	movq	%r13,8(%rdi)
+	movq	%r14,16(%rdi)
+	movq	%r15,24(%rdi)
+	leaq	32(%rdi),%rdi
+	incq	%rcx
+	jnz	L$sqr8x_sub
+
+	sbbq	$0,%rax
+	leaq	(%rbx,%r9,1),%rbx
+	leaq	(%rdi,%r9,1),%rdi
+
+.byte	102,72,15,110,200
+	pxor	%xmm0,%xmm0
+	pshufd	$0,%xmm1,%xmm1
+	movq	40(%rsp),%rsi
+	jmp	L$sqr8x_cond_copy
+
+.p2align	5
+L$sqr8x_cond_copy:
+	movdqa	0(%rbx),%xmm2
+	movdqa	16(%rbx),%xmm3
+	leaq	32(%rbx),%rbx
+	movdqu	0(%rdi),%xmm4
+	movdqu	16(%rdi),%xmm5
+	leaq	32(%rdi),%rdi
+	movdqa	%xmm0,-32(%rbx)
+	movdqa	%xmm0,-16(%rbx)
+	movdqa	%xmm0,-32(%rbx,%rdx,1)
+	movdqa	%xmm0,-16(%rbx,%rdx,1)
+	pcmpeqd	%xmm1,%xmm0
+	pand	%xmm1,%xmm2
+	pand	%xmm1,%xmm3
+	pand	%xmm0,%xmm4
+	pand	%xmm0,%xmm5
+	pxor	%xmm0,%xmm0
+	por	%xmm2,%xmm4
+	por	%xmm3,%xmm5
+	movdqu	%xmm4,-32(%rdi)
+	movdqu	%xmm5,-16(%rdi)
+	addq	$32,%r9
+	jnz	L$sqr8x_cond_copy
 
 	movq	$1,%rax
 	movq	-48(%rsi),%r15

third_party/boringssl/mac-x86_64/crypto/bn/x86_64-mont5.S

diff --git a/third_party/boringssl/mac-x86_64/crypto/bn/x86_64-mont5.S b/third_party/boringssl/mac-x86_64/crypto/bn/x86_64-mont5.S
index 2e8f469..f3ad8d7 100644
--- a/third_party/boringssl/mac-x86_64/crypto/bn/x86_64-mont5.S
+++ b/third_party/boringssl/mac-x86_64/crypto/bn/x86_64-mont5.S
@@ -16,46 +16,151 @@
 L$mul_enter:
 	movl	%r9d,%r9d
 	movq	%rsp,%rax
-	movl	8(%rsp),%r10d
+	movd	8(%rsp),%xmm5
+	leaq	L$inc(%rip),%r10
 	pushq	%rbx
 	pushq	%rbp
 	pushq	%r12
 	pushq	%r13
 	pushq	%r14
 	pushq	%r15
+
 	leaq	2(%r9),%r11
 	negq	%r11
-	leaq	(%rsp,%r11,8),%rsp
+	leaq	-264(%rsp,%r11,8),%rsp
 	andq	$-1024,%rsp
 
 	movq	%rax,8(%rsp,%r9,8)
 L$mul_body:
-	movq	%rdx,%r12
-	movq	%r10,%r11
-	shrq	$3,%r10
-	andq	$7,%r11
-	notq	%r10
-	leaq	L$magic_masks(%rip),%rax
-	andq	$3,%r10
-	leaq	96(%r12,%r11,8),%r12
-	movq	0(%rax,%r10,8),%xmm4
-	movq	8(%rax,%r10,8),%xmm5
-	movq	16(%rax,%r10,8),%xmm6
-	movq	24(%rax,%r10,8),%xmm7
+	leaq	128(%rdx),%r12
+	movdqa	0(%r10),%xmm0
+	movdqa	16(%r10),%xmm1
+	leaq	24-112(%rsp,%r9,8),%r10
+	andq	$-16,%r10
 
-	movq	-96(%r12),%xmm0
-	movq	-32(%r12),%xmm1
-	pand	%xmm4,%xmm0
-	movq	32(%r12),%xmm2
-	pand	%xmm5,%xmm1
-	movq	96(%r12),%xmm3
-	pand	%xmm6,%xmm2
-	por	%xmm1,%xmm0
-	pand	%xmm7,%xmm3
+	pshufd	$0,%xmm5,%xmm5
+	movdqa	%xmm1,%xmm4
+	movdqa	%xmm1,%xmm2
+	paddd	%xmm0,%xmm1
+	pcmpeqd	%xmm5,%xmm0
+.byte	0x67
+	movdqa	%xmm4,%xmm3
+	paddd	%xmm1,%xmm2
+	pcmpeqd	%xmm5,%xmm1
+	movdqa	%xmm0,112(%r10)
+	movdqa	%xmm4,%xmm0
+
+	paddd	%xmm2,%xmm3
+	pcmpeqd	%xmm5,%xmm2
+	movdqa	%xmm1,128(%r10)
+	movdqa	%xmm4,%xmm1
+
+	paddd	%xmm3,%xmm0
+	pcmpeqd	%xmm5,%xmm3
+	movdqa	%xmm2,144(%r10)
+	movdqa	%xmm4,%xmm2
+
+	paddd	%xmm0,%xmm1
+	pcmpeqd	%xmm5,%xmm0
+	movdqa	%xmm3,160(%r10)
+	movdqa	%xmm4,%xmm3
+	paddd	%xmm1,%xmm2
+	pcmpeqd	%xmm5,%xmm1
+	movdqa	%xmm0,176(%r10)
+	movdqa	%xmm4,%xmm0
+
+	paddd	%xmm2,%xmm3
+	pcmpeqd	%xmm5,%xmm2
+	movdqa	%xmm1,192(%r10)
+	movdqa	%xmm4,%xmm1
+
+	paddd	%xmm3,%xmm0
+	pcmpeqd	%xmm5,%xmm3
+	movdqa	%xmm2,208(%r10)
+	movdqa	%xmm4,%xmm2
+
+	paddd	%xmm0,%xmm1
+	pcmpeqd	%xmm5,%xmm0
+	movdqa	%xmm3,224(%r10)
+	movdqa	%xmm4,%xmm3
+	paddd	%xmm1,%xmm2
+	pcmpeqd	%xmm5,%xmm1
+	movdqa	%xmm0,240(%r10)
+	movdqa	%xmm4,%xmm0
+
+	paddd	%xmm2,%xmm3
+	pcmpeqd	%xmm5,%xmm2
+	movdqa	%xmm1,256(%r10)
+	movdqa	%xmm4,%xmm1
+
+	paddd	%xmm3,%xmm0
+	pcmpeqd	%xmm5,%xmm3
+	movdqa	%xmm2,272(%r10)
+	movdqa	%xmm4,%xmm2
+
+	paddd	%xmm0,%xmm1
+	pcmpeqd	%xmm5,%xmm0
+	movdqa	%xmm3,288(%r10)
+	movdqa	%xmm4,%xmm3
+	paddd	%xmm1,%xmm2
+	pcmpeqd	%xmm5,%xmm1
+	movdqa	%xmm0,304(%r10)
+
+	paddd	%xmm2,%xmm3
+.byte	0x67
+	pcmpeqd	%xmm5,%xmm2
+	movdqa	%xmm1,320(%r10)
+
+	pcmpeqd	%xmm5,%xmm3
+	movdqa	%xmm2,336(%r10)
+	pand	64(%r12),%xmm0
+
+	pand	80(%r12),%xmm1
+	pand	96(%r12),%xmm2
+	movdqa	%xmm3,352(%r10)
+	pand	112(%r12),%xmm3
 	por	%xmm2,%xmm0
+	por	%xmm3,%xmm1
+	movdqa	-128(%r12),%xmm4
+	movdqa	-112(%r12),%xmm5
+	movdqa	-96(%r12),%xmm2
+	pand	112(%r10),%xmm4
+	movdqa	-80(%r12),%xmm3
+	pand	128(%r10),%xmm5
+	por	%xmm4,%xmm0
+	pand	144(%r10),%xmm2
+	por	%xmm5,%xmm1
+	pand	160(%r10),%xmm3
+	por	%xmm2,%xmm0
+	por	%xmm3,%xmm1
+	movdqa	-64(%r12),%xmm4
+	movdqa	-48(%r12),%xmm5
+	movdqa	-32(%r12),%xmm2
+	pand	176(%r10),%xmm4
+	movdqa	-16(%r12),%xmm3
+	pand	192(%r10),%xmm5
+	por	%xmm4,%xmm0
+	pand	208(%r10),%xmm2
+	por	%xmm5,%xmm1
+	pand	224(%r10),%xmm3
+	por	%xmm2,%xmm0
+	por	%xmm3,%xmm1
+	movdqa	0(%r12),%xmm4
+	movdqa	16(%r12),%xmm5
+	movdqa	32(%r12),%xmm2
+	pand	240(%r10),%xmm4
+	movdqa	48(%r12),%xmm3
+	pand	256(%r10),%xmm5
+	por	%xmm4,%xmm0
+	pand	272(%r10),%xmm2
+	por	%xmm5,%xmm1
+	pand	288(%r10),%xmm3
+	por	%xmm2,%xmm0
+	por	%xmm3,%xmm1
+	por	%xmm1,%xmm0
+	pshufd	$0x4e,%xmm0,%xmm1
+	por	%xmm1,%xmm0
 	leaq	256(%r12),%r12
-	por	%xmm3,%xmm0
-
 .byte	102,72,15,126,195
 
 	movq	(%r8),%r8
@@ -64,29 +169,14 @@
 	xorq	%r14,%r14
 	xorq	%r15,%r15
 
-	movq	-96(%r12),%xmm0
-	movq	-32(%r12),%xmm1
-	pand	%xmm4,%xmm0
-	movq	32(%r12),%xmm2
-	pand	%xmm5,%xmm1
-
 	movq	%r8,%rbp
 	mulq	%rbx
 	movq	%rax,%r10
 	movq	(%rcx),%rax
 
-	movq	96(%r12),%xmm3
-	pand	%xmm6,%xmm2
-	por	%xmm1,%xmm0
-	pand	%xmm7,%xmm3
-
 	imulq	%r10,%rbp
 	movq	%rdx,%r11
 
-	por	%xmm2,%xmm0
-	leaq	256(%r12),%r12
-	por	%xmm3,%xmm0
-
 	mulq	%rbp
 	addq	%rax,%r10
 	movq	8(%rsi),%rax
@@ -119,14 +209,12 @@
 	cmpq	%r9,%r15
 	jne	L$1st
 
-.byte	102,72,15,126,195
 
 	addq	%rax,%r13
-	movq	(%rsi),%rax
 	adcq	$0,%rdx
 	addq	%r11,%r13
 	adcq	$0,%rdx
-	movq	%r13,-16(%rsp,%r15,8)
+	movq	%r13,-16(%rsp,%r9,8)
 	movq	%rdx,%r13
 	movq	%r10,%r11
 
@@ -140,33 +228,78 @@
 	jmp	L$outer
 .p2align	4
 L$outer:
+	leaq	24+128(%rsp,%r9,8),%rdx
+	andq	$-16,%rdx
+	pxor	%xmm4,%xmm4
+	pxor	%xmm5,%xmm5
+	movdqa	-128(%r12),%xmm0
+	movdqa	-112(%r12),%xmm1
+	movdqa	-96(%r12),%xmm2
+	movdqa	-80(%r12),%xmm3
+	pand	-128(%rdx),%xmm0
+	pand	-112(%rdx),%xmm1
+	por	%xmm0,%xmm4
+	pand	-96(%rdx),%xmm2
+	por	%xmm1,%xmm5
+	pand	-80(%rdx),%xmm3
+	por	%xmm2,%xmm4
+	por	%xmm3,%xmm5
+	movdqa	-64(%r12),%xmm0
+	movdqa	-48(%r12),%xmm1
+	movdqa	-32(%r12),%xmm2
+	movdqa	-16(%r12),%xmm3
+	pand	-64(%rdx),%xmm0
+	pand	-48(%rdx),%xmm1
+	por	%xmm0,%xmm4
+	pand	-32(%rdx),%xmm2
+	por	%xmm1,%xmm5
+	pand	-16(%rdx),%xmm3
+	por	%xmm2,%xmm4
+	por	%xmm3,%xmm5
+	movdqa	0(%r12),%xmm0
+	movdqa	16(%r12),%xmm1
+	movdqa	32(%r12),%xmm2
+	movdqa	48(%r12),%xmm3
+	pand	0(%rdx),%xmm0
+	pand	16(%rdx),%xmm1
+	por	%xmm0,%xmm4
+	pand	32(%rdx),%xmm2
+	por	%xmm1,%xmm5
+	pand	48(%rdx),%xmm3
+	por	%xmm2,%xmm4
+	por	%xmm3,%xmm5
+	movdqa	64(%r12),%xmm0
+	movdqa	80(%r12),%xmm1
+	movdqa	96(%r12),%xmm2
+	movdqa	112(%r12),%xmm3
+	pand	64(%rdx),%xmm0
+	pand	80(%rdx),%xmm1
+	por	%xmm0,%xmm4
+	pand	96(%rdx),%xmm2
+	por	%xmm1,%xmm5
+	pand	112(%rdx),%xmm3
+	por	%xmm2,%xmm4
+	por	%xmm3,%xmm5
+	por	%xmm5,%xmm4
+	pshufd	$0x4e,%xmm4,%xmm0
+	por	%xmm4,%xmm0
+	leaq	256(%r12),%r12
+
+	movq	(%rsi),%rax
+.byte	102,72,15,126,195
+
 	xorq	%r15,%r15
 	movq	%r8,%rbp
 	movq	(%rsp),%r10
 
-	movq	-96(%r12),%xmm0
-	movq	-32(%r12),%xmm1
-	pand	%xmm4,%xmm0
-	movq	32(%r12),%xmm2
-	pand	%xmm5,%xmm1
-
 	mulq	%rbx
 	addq	%rax,%r10
 	movq	(%rcx),%rax
 	adcq	$0,%rdx
 
-	movq	96(%r12),%xmm3
-	pand	%xmm6,%xmm2
-	por	%xmm1,%xmm0
-	pand	%xmm7,%xmm3
-
 	imulq	%r10,%rbp
 	movq	%rdx,%r11
 
-	por	%xmm2,%xmm0
-	leaq	256(%r12),%r12
-	por	%xmm3,%xmm0
-
 	mulq	%rbp
 	addq	%rax,%r10
 	movq	8(%rsi),%rax
@@ -202,15 +335,12 @@
 	cmpq	%r9,%r15
 	jne	L$inner
 
-.byte	102,72,15,126,195
-
 	addq	%rax,%r13
-	movq	(%rsi),%rax
 	adcq	$0,%rdx
 	addq	%r10,%r13
-	movq	(%rsp,%r15,8),%r10
+	movq	(%rsp,%r9,8),%r10
 	adcq	$0,%rdx
-	movq	%r13,-16(%rsp,%r15,8)
+	movq	%r13,-16(%rsp,%r9,8)
 	movq	%rdx,%r13
 
 	xorq	%rdx,%rdx
@@ -256,6 +386,7 @@
 
 	movq	8(%rsp,%r9,8),%rsi
 	movq	$1,%rax
+
 	movq	-48(%rsi),%r15
 	movq	-40(%rsi),%r14
 	movq	-32(%rsi),%r13
@@ -278,10 +409,10 @@
 	pushq	%r13
 	pushq	%r14
 	pushq	%r15
+
 .byte	0x67
-	movl	%r9d,%r10d
 	shll	$3,%r9d
-	shll	$3+2,%r10d
+	leaq	(%r9,%r9,2),%r10
 	negq	%r9
 
 
@@ -291,19 +422,21 @@
 
 
 
-	leaq	-64(%rsp,%r9,2),%r11
-	subq	%rsi,%r11
+
+
+	leaq	-320(%rsp,%r9,2),%r11
+	subq	%rdi,%r11
 	andq	$4095,%r11
 	cmpq	%r11,%r10
 	jb	L$mul4xsp_alt
 	subq	%r11,%rsp
-	leaq	-64(%rsp,%r9,2),%rsp
+	leaq	-320(%rsp,%r9,2),%rsp
 	jmp	L$mul4xsp_done
 
 .p2align	5
 L$mul4xsp_alt:
-	leaq	4096-64(,%r9,2),%r10
-	leaq	-64(%rsp,%r9,2),%rsp
+	leaq	4096-320(,%r9,2),%r10
+	leaq	-320(%rsp,%r9,2),%rsp
 	subq	%r10,%r11
 	movq	$0,%r10
 	cmovcq	%r10,%r11
@@ -319,6 +452,7 @@
 
 	movq	40(%rsp),%rsi
 	movq	$1,%rax
+
 	movq	-48(%rsi),%r15
 	movq	-40(%rsi),%r14
 	movq	-32(%rsi),%r13
@@ -334,47 +468,141 @@
 .p2align	5
 mul4x_internal:
 	shlq	$5,%r9
-	movl	8(%rax),%r10d
-	leaq	256(%rdx,%r9,1),%r13
+	movd	8(%rax),%xmm5
+	leaq	L$inc(%rip),%rax
+	leaq	128(%rdx,%r9,1),%r13
 	shrq	$5,%r9
-	movq	%r10,%r11
-	shrq	$3,%r10
-	andq	$7,%r11
-	notq	%r10
-	leaq	L$magic_masks(%rip),%rax
-	andq	$3,%r10
-	leaq	96(%rdx,%r11,8),%r12
-	movq	0(%rax,%r10,8),%xmm4
-	movq	8(%rax,%r10,8),%xmm5
-	addq	$7,%r11
-	movq	16(%rax,%r10,8),%xmm6
-	movq	24(%rax,%r10,8),%xmm7
-	andq	$7,%r11
+	movdqa	0(%rax),%xmm0
+	movdqa	16(%rax),%xmm1
+	leaq	88-112(%rsp,%r9,1),%r10
+	leaq	128(%rdx),%r12
 
-	movq	-96(%r12),%xmm0
-	leaq	256(%r12),%r14
-	movq	-32(%r12),%xmm1
-	pand	%xmm4,%xmm0
-	movq	32(%r12),%xmm2
-	pand	%xmm5,%xmm1
-	movq	96(%r12),%xmm3
-	pand	%xmm6,%xmm2
+	pshufd	$0,%xmm5,%xmm5
+	movdqa	%xmm1,%xmm4
+.byte	0x67,0x67
+	movdqa	%xmm1,%xmm2
+	paddd	%xmm0,%xmm1
+	pcmpeqd	%xmm5,%xmm0
 .byte	0x67
-	por	%xmm1,%xmm0
-	movq	-96(%r14),%xmm1
+	movdqa	%xmm4,%xmm3
+	paddd	%xmm1,%xmm2
+	pcmpeqd	%xmm5,%xmm1
+	movdqa	%xmm0,112(%r10)
+	movdqa	%xmm4,%xmm0
+
+	paddd	%xmm2,%xmm3
+	pcmpeqd	%xmm5,%xmm2
+	movdqa	%xmm1,128(%r10)
+	movdqa	%xmm4,%xmm1
+
+	paddd	%xmm3,%xmm0
+	pcmpeqd	%xmm5,%xmm3
+	movdqa	%xmm2,144(%r10)
+	movdqa	%xmm4,%xmm2
+
+	paddd	%xmm0,%xmm1
+	pcmpeqd	%xmm5,%xmm0
+	movdqa	%xmm3,160(%r10)
+	movdqa	%xmm4,%xmm3
+	paddd	%xmm1,%xmm2
+	pcmpeqd	%xmm5,%xmm1
+	movdqa	%xmm0,176(%r10)
+	movdqa	%xmm4,%xmm0
+
+	paddd	%xmm2,%xmm3
+	pcmpeqd	%xmm5,%xmm2
+	movdqa	%xmm1,192(%r10)
+	movdqa	%xmm4,%xmm1
+
+	paddd	%xmm3,%xmm0
+	pcmpeqd	%xmm5,%xmm3
+	movdqa	%xmm2,208(%r10)
+	movdqa	%xmm4,%xmm2
+
+	paddd	%xmm0,%xmm1
+	pcmpeqd	%xmm5,%xmm0
+	movdqa	%xmm3,224(%r10)
+	movdqa	%xmm4,%xmm3
+	paddd	%xmm1,%xmm2
+	pcmpeqd	%xmm5,%xmm1
+	movdqa	%xmm0,240(%r10)
+	movdqa	%xmm4,%xmm0
+
+	paddd	%xmm2,%xmm3
+	pcmpeqd	%xmm5,%xmm2
+	movdqa	%xmm1,256(%r10)
+	movdqa	%xmm4,%xmm1
+
+	paddd	%xmm3,%xmm0
+	pcmpeqd	%xmm5,%xmm3
+	movdqa	%xmm2,272(%r10)
+	movdqa	%xmm4,%xmm2
+
+	paddd	%xmm0,%xmm1
+	pcmpeqd	%xmm5,%xmm0
+	movdqa	%xmm3,288(%r10)
+	movdqa	%xmm4,%xmm3
+	paddd	%xmm1,%xmm2
+	pcmpeqd	%xmm5,%xmm1
+	movdqa	%xmm0,304(%r10)
+
+	paddd	%xmm2,%xmm3
 .byte	0x67
-	pand	%xmm7,%xmm3
-.byte	0x67
+	pcmpeqd	%xmm5,%xmm2
+	movdqa	%xmm1,320(%r10)
+
+	pcmpeqd	%xmm5,%xmm3
+	movdqa	%xmm2,336(%r10)
+	pand	64(%r12),%xmm0
+
+	pand	80(%r12),%xmm1
+	pand	96(%r12),%xmm2
+	movdqa	%xmm3,352(%r10)
+	pand	112(%r12),%xmm3
 	por	%xmm2,%xmm0
-	movq	-32(%r14),%xmm2
-.byte	0x67
-	pand	%xmm4,%xmm1
-.byte	0x67
-	por	%xmm3,%xmm0
-	movq	32(%r14),%xmm3
-
+	por	%xmm3,%xmm1
+	movdqa	-128(%r12),%xmm4
+	movdqa	-112(%r12),%xmm5
+	movdqa	-96(%r12),%xmm2
+	pand	112(%r10),%xmm4
+	movdqa	-80(%r12),%xmm3
+	pand	128(%r10),%xmm5
+	por	%xmm4,%xmm0
+	pand	144(%r10),%xmm2
+	por	%xmm5,%xmm1
+	pand	160(%r10),%xmm3
+	por	%xmm2,%xmm0
+	por	%xmm3,%xmm1
+	movdqa	-64(%r12),%xmm4
+	movdqa	-48(%r12),%xmm5
+	movdqa	-32(%r12),%xmm2
+	pand	176(%r10),%xmm4
+	movdqa	-16(%r12),%xmm3
+	pand	192(%r10),%xmm5
+	por	%xmm4,%xmm0
+	pand	208(%r10),%xmm2
+	por	%xmm5,%xmm1
+	pand	224(%r10),%xmm3
+	por	%xmm2,%xmm0
+	por	%xmm3,%xmm1
+	movdqa	0(%r12),%xmm4
+	movdqa	16(%r12),%xmm5
+	movdqa	32(%r12),%xmm2
+	pand	240(%r10),%xmm4
+	movdqa	48(%r12),%xmm3
+	pand	256(%r10),%xmm5
+	por	%xmm4,%xmm0
+	pand	272(%r10),%xmm2
+	por	%xmm5,%xmm1
+	pand	288(%r10),%xmm3
+	por	%xmm2,%xmm0
+	por	%xmm3,%xmm1
+	por	%xmm1,%xmm0
+	pshufd	$0x4e,%xmm0,%xmm1
+	por	%xmm1,%xmm0
+	leaq	256(%r12),%r12
 .byte	102,72,15,126,195
-	movq	96(%r14),%xmm0
+
 	movq	%r13,16+8(%rsp)
 	movq	%rdi,56+8(%rsp)
 
@@ -388,26 +616,10 @@
 	movq	%rax,%r10
 	movq	(%rcx),%rax
 
-	pand	%xmm5,%xmm2
-	pand	%xmm6,%xmm3
-	por	%xmm2,%xmm1
-
 	imulq	%r10,%rbp
-
-
-
-
-
-
-
-	leaq	64+8(%rsp,%r11,8),%r14
+	leaq	64+8(%rsp),%r14
 	movq	%rdx,%r11
 
-	pand	%xmm7,%xmm0
-	por	%xmm3,%xmm1
-	leaq	512(%r12),%r12
-	por	%xmm1,%xmm0
-
 	mulq	%rbp
 	addq	%rax,%r10
 	movq	8(%rsi,%r9,1),%rax
@@ -416,7 +628,7 @@
 
 	mulq	%rbx
 	addq	%rax,%r11
-	movq	16(%rcx),%rax
+	movq	8(%rcx),%rax
 	adcq	$0,%rdx
 	movq	%rdx,%r10
 
@@ -426,7 +638,7 @@
 	adcq	$0,%rdx
 	addq	%r11,%rdi
 	leaq	32(%r9),%r15
-	leaq	64(%rcx),%rcx
+	leaq	32(%rcx),%rcx
 	adcq	$0,%rdx
 	movq	%rdi,(%r14)
 	movq	%rdx,%r13
@@ -436,7 +648,7 @@
 L$1st4x:
 	mulq	%rbx
 	addq	%rax,%r10
-	movq	-32(%rcx),%rax
+	movq	-16(%rcx),%rax
 	leaq	32(%r14),%r14
 	adcq	$0,%rdx
 	movq	%rdx,%r11
@@ -452,7 +664,7 @@
 
 	mulq	%rbx
 	addq	%rax,%r11
-	movq	-16(%rcx),%rax
+	movq	-8(%rcx),%rax
 	adcq	$0,%rdx
 	movq	%rdx,%r10
 
@@ -482,7 +694,7 @@
 
 	mulq	%rbx
 	addq	%rax,%r11
-	movq	16(%rcx),%rax
+	movq	8(%rcx),%rax
 	adcq	$0,%rdx
 	movq	%rdx,%r10
 
@@ -491,7 +703,7 @@
 	movq	16(%rsi,%r15,1),%rax
 	adcq	$0,%rdx
 	addq	%r11,%rdi
-	leaq	64(%rcx),%rcx
+	leaq	32(%rcx),%rcx
 	adcq	$0,%rdx
 	movq	%rdi,(%r14)
 	movq	%rdx,%r13
@@ -501,7 +713,7 @@
 
 	mulq	%rbx
 	addq	%rax,%r10
-	movq	-32(%rcx),%rax
+	movq	-16(%rcx),%rax
 	leaq	32(%r14),%r14
 	adcq	$0,%rdx
 	movq	%rdx,%r11
@@ -517,7 +729,7 @@
 
 	mulq	%rbx
 	addq	%rax,%r11
-	movq	-16(%rcx),%rax
+	movq	-8(%rcx),%rax
 	adcq	$0,%rdx
 	movq	%rdx,%r10
 
@@ -530,8 +742,7 @@
 	movq	%rdi,-16(%r14)
 	movq	%rdx,%r13
 
-.byte	102,72,15,126,195
-	leaq	(%rcx,%r9,2),%rcx
+	leaq	(%rcx,%r9,1),%rcx
 
 	xorq	%rdi,%rdi
 	addq	%r10,%r13
@@ -542,6 +753,63 @@
 
 .p2align	5
 L$outer4x:
+	leaq	16+128(%r14),%rdx
+	pxor	%xmm4,%xmm4
+	pxor	%xmm5,%xmm5
+	movdqa	-128(%r12),%xmm0
+	movdqa	-112(%r12),%xmm1
+	movdqa	-96(%r12),%xmm2
+	movdqa	-80(%r12),%xmm3
+	pand	-128(%rdx),%xmm0
+	pand	-112(%rdx),%xmm1
+	por	%xmm0,%xmm4
+	pand	-96(%rdx),%xmm2
+	por	%xmm1,%xmm5
+	pand	-80(%rdx),%xmm3
+	por	%xmm2,%xmm4
+	por	%xmm3,%xmm5
+	movdqa	-64(%r12),%xmm0
+	movdqa	-48(%r12),%xmm1
+	movdqa	-32(%r12),%xmm2
+	movdqa	-16(%r12),%xmm3
+	pand	-64(%rdx),%xmm0
+	pand	-48(%rdx),%xmm1
+	por	%xmm0,%xmm4
+	pand	-32(%rdx),%xmm2
+	por	%xmm1,%xmm5
+	pand	-16(%rdx),%xmm3
+	por	%xmm2,%xmm4
+	por	%xmm3,%xmm5
+	movdqa	0(%r12),%xmm0
+	movdqa	16(%r12),%xmm1
+	movdqa	32(%r12),%xmm2
+	movdqa	48(%r12),%xmm3
+	pand	0(%rdx),%xmm0
+	pand	16(%rdx),%xmm1
+	por	%xmm0,%xmm4
+	pand	32(%rdx),%xmm2
+	por	%xmm1,%xmm5
+	pand	48(%rdx),%xmm3
+	por	%xmm2,%xmm4
+	por	%xmm3,%xmm5
+	movdqa	64(%r12),%xmm0
+	movdqa	80(%r12),%xmm1
+	movdqa	96(%r12),%xmm2
+	movdqa	112(%r12),%xmm3
+	pand	64(%rdx),%xmm0
+	pand	80(%rdx),%xmm1
+	por	%xmm0,%xmm4
+	pand	96(%rdx),%xmm2
+	por	%xmm1,%xmm5
+	pand	112(%rdx),%xmm3
+	por	%xmm2,%xmm4
+	por	%xmm3,%xmm5
+	por	%xmm5,%xmm4
+	pshufd	$0x4e,%xmm4,%xmm0
+	por	%xmm4,%xmm0
+	leaq	256(%r12),%r12
+.byte	102,72,15,126,195
+
 	movq	(%r14,%r9,1),%r10
 	movq	%r8,%rbp
 	mulq	%rbx
@@ -549,25 +817,11 @@
 	movq	(%rcx),%rax
 	adcq	$0,%rdx
 
-	movq	-96(%r12),%xmm0
-	movq	-32(%r12),%xmm1
-	pand	%xmm4,%xmm0
-	movq	32(%r12),%xmm2
-	pand	%xmm5,%xmm1
-	movq	96(%r12),%xmm3
-
 	imulq	%r10,%rbp
-.byte	0x67
 	movq	%rdx,%r11
 	movq	%rdi,(%r14)
 
-	pand	%xmm6,%xmm2
-	por	%xmm1,%xmm0
-	pand	%xmm7,%xmm3
-	por	%xmm2,%xmm0
 	leaq	(%r14,%r9,1),%r14
-	leaq	256(%r12),%r12
-	por	%xmm3,%xmm0
 
 	mulq	%rbp
 	addq	%rax,%r10
@@ -577,7 +831,7 @@
 
 	mulq	%rbx
 	addq	%rax,%r11
-	movq	16(%rcx),%rax
+	movq	8(%rcx),%rax
 	adcq	$0,%rdx
 	addq	8(%r14),%r11
 	adcq	$0,%rdx
@@ -589,7 +843,7 @@
 	adcq	$0,%rdx
 	addq	%r11,%rdi
 	leaq	32(%r9),%r15
-	leaq	64(%rcx),%rcx
+	leaq	32(%rcx),%rcx
 	adcq	$0,%rdx
 	movq	%rdx,%r13
 	jmp	L$inner4x
@@ -598,7 +852,7 @@
 L$inner4x:
 	mulq	%rbx
 	addq	%rax,%r10
-	movq	-32(%rcx),%rax
+	movq	-16(%rcx),%rax
 	adcq	$0,%rdx
 	addq	16(%r14),%r10
 	leaq	32(%r14),%r14
@@ -616,7 +870,7 @@
 
 	mulq	%rbx
 	addq	%rax,%r11
-	movq	-16(%rcx),%rax
+	movq	-8(%rcx),%rax
 	adcq	$0,%rdx
 	addq	-8(%r14),%r11
 	adcq	$0,%rdx
@@ -650,7 +904,7 @@
 
 	mulq	%rbx
 	addq	%rax,%r11
-	movq	16(%rcx),%rax
+	movq	8(%rcx),%rax
 	adcq	$0,%rdx
 	addq	8(%r14),%r11
 	adcq	$0,%rdx
@@ -661,7 +915,7 @@
 	movq	16(%rsi,%r15,1),%rax
 	adcq	$0,%rdx
 	addq	%r11,%rdi
-	leaq	64(%rcx),%rcx
+	leaq	32(%rcx),%rcx
 	adcq	$0,%rdx
 	movq	%r13,-8(%r14)
 	movq	%rdx,%r13
@@ -671,7 +925,7 @@
 
 	mulq	%rbx
 	addq	%rax,%r10
-	movq	-32(%rcx),%rax
+	movq	-16(%rcx),%rax
 	adcq	$0,%rdx
 	addq	16(%r14),%r10
 	leaq	32(%r14),%r14
@@ -690,7 +944,7 @@
 	mulq	%rbx
 	addq	%rax,%r11
 	movq	%rbp,%rax
-	movq	-16(%rcx),%rbp
+	movq	-8(%rcx),%rbp
 	adcq	$0,%rdx
 	addq	-8(%r14),%r11
 	adcq	$0,%rdx
@@ -705,9 +959,8 @@
 	movq	%r13,-24(%r14)
 	movq	%rdx,%r13
 
-.byte	102,72,15,126,195
 	movq	%rdi,-16(%r14)
-	leaq	(%rcx,%r9,2),%rcx
+	leaq	(%rcx,%r9,1),%rcx
 
 	xorq	%rdi,%rdi
 	addq	%r10,%r13
@@ -718,16 +971,23 @@
 
 	cmpq	16+8(%rsp),%r12
 	jb	L$outer4x
+	xorq	%rax,%rax
 	subq	%r13,%rbp
 	adcq	%r15,%r15
 	orq	%r15,%rdi
-	xorq	$1,%rdi
+	subq	%rdi,%rax
 	leaq	(%r14,%r9,1),%rbx
-	leaq	(%rcx,%rdi,8),%rbp
+	movq	(%rcx),%r12
+	leaq	(%rcx),%rbp
 	movq	%r9,%rcx
 	sarq	$3+2,%rcx
 	movq	56+8(%rsp),%rdi
-	jmp	L$sqr4x_sub
+	decq	%r12
+	xorq	%r10,%r10
+	movq	8(%rbp),%r13
+	movq	16(%rbp),%r14
+	movq	24(%rbp),%r15
+	jmp	L$sqr4x_sub_entry
 
 .globl	_bn_power5
 .private_extern _bn_power5
@@ -741,9 +1001,9 @@
 	pushq	%r13
 	pushq	%r14
 	pushq	%r15
-	movl	%r9d,%r10d
+
 	shll	$3,%r9d
-	shll	$3+2,%r10d
+	leal	(%r9,%r9,2),%r10d
 	negq	%r9
 	movq	(%r8),%r8
 
@@ -753,19 +1013,20 @@
 
 
 
-	leaq	-64(%rsp,%r9,2),%r11
-	subq	%rsi,%r11
+
+	leaq	-320(%rsp,%r9,2),%r11
+	subq	%rdi,%r11
 	andq	$4095,%r11
 	cmpq	%r11,%r10
 	jb	L$pwr_sp_alt
 	subq	%r11,%rsp
-	leaq	-64(%rsp,%r9,2),%rsp
+	leaq	-320(%rsp,%r9,2),%rsp
 	jmp	L$pwr_sp_done
 
 .p2align	5
 L$pwr_sp_alt:
-	leaq	4096-64(,%r9,2),%r10
-	leaq	-64(%rsp,%r9,2),%rsp
+	leaq	4096-320(,%r9,2),%r10
+	leaq	-320(%rsp,%r9,2),%rsp
 	subq	%r10,%r11
 	movq	$0,%r10
 	cmovcq	%r10,%r11
@@ -793,10 +1054,15 @@
 .byte	102,72,15,110,226
 
 	call	__bn_sqr8x_internal
+	call	__bn_post4x_internal
 	call	__bn_sqr8x_internal
+	call	__bn_post4x_internal
 	call	__bn_sqr8x_internal
+	call	__bn_post4x_internal
 	call	__bn_sqr8x_internal
+	call	__bn_post4x_internal
 	call	__bn_sqr8x_internal
+	call	__bn_post4x_internal
 
 .byte	102,72,15,126,209
 .byte	102,72,15,126,226
@@ -1341,9 +1607,9 @@
 	movq	%rbx,-16(%rdi)
 	movq	%r8,-8(%rdi)
 .byte	102,72,15,126,213
-sqr8x_reduction:
+__bn_sqr8x_reduction:
 	xorq	%rax,%rax
-	leaq	(%rbp,%r9,2),%rcx
+	leaq	(%r9,%rbp,1),%rcx
 	leaq	48+8(%rsp,%r9,2),%rdx
 	movq	%rcx,0+8(%rsp)
 	leaq	48+8(%rsp,%r9,1),%rdi
@@ -1376,14 +1642,14 @@
 .p2align	5
 L$8x_reduce:
 	mulq	%rbx
-	movq	16(%rbp),%rax
+	movq	8(%rbp),%rax
 	negq	%r8
 	movq	%rdx,%r8
 	adcq	$0,%r8
 
 	mulq	%rbx
 	addq	%rax,%r9
-	movq	32(%rbp),%rax
+	movq	16(%rbp),%rax
 	adcq	$0,%rdx
 	addq	%r9,%r8
 	movq	%rbx,48-8+8(%rsp,%rcx,8)
@@ -1392,7 +1658,7 @@
 
 	mulq	%rbx
 	addq	%rax,%r10
-	movq	48(%rbp),%rax
+	movq	24(%rbp),%rax
 	adcq	$0,%rdx
 	addq	%r10,%r9
 	movq	32+8(%rsp),%rsi
@@ -1401,7 +1667,7 @@
 
 	mulq	%rbx
 	addq	%rax,%r11
-	movq	64(%rbp),%rax
+	movq	32(%rbp),%rax
 	adcq	$0,%rdx
 	imulq	%r8,%rsi
 	addq	%r11,%r10
@@ -1410,7 +1676,7 @@
 
 	mulq	%rbx
 	addq	%rax,%r12
-	movq	80(%rbp),%rax
+	movq	40(%rbp),%rax
 	adcq	$0,%rdx
 	addq	%r12,%r11
 	movq	%rdx,%r12
@@ -1418,7 +1684,7 @@
 
 	mulq	%rbx
 	addq	%rax,%r13
-	movq	96(%rbp),%rax
+	movq	48(%rbp),%rax
 	adcq	$0,%rdx
 	addq	%r13,%r12
 	movq	%rdx,%r13
@@ -1426,7 +1692,7 @@
 
 	mulq	%rbx
 	addq	%rax,%r14
-	movq	112(%rbp),%rax
+	movq	56(%rbp),%rax
 	adcq	$0,%rdx
 	addq	%r14,%r13
 	movq	%rdx,%r14
@@ -1444,7 +1710,7 @@
 	decl	%ecx
 	jnz	L$8x_reduce
 
-	leaq	128(%rbp),%rbp
+	leaq	64(%rbp),%rbp
 	xorq	%rax,%rax
 	movq	8+8(%rsp),%rdx
 	cmpq	0+8(%rsp),%rbp
@@ -1470,14 +1736,14 @@
 L$8x_tail:
 	mulq	%rbx
 	addq	%rax,%r8
-	movq	16(%rbp),%rax
+	movq	8(%rbp),%rax
 	movq	%r8,(%rdi)
 	movq	%rdx,%r8
 	adcq	$0,%r8
 
 	mulq	%rbx
 	addq	%rax,%r9
-	movq	32(%rbp),%rax
+	movq	16(%rbp),%rax
 	adcq	$0,%rdx
 	addq	%r9,%r8
 	leaq	8(%rdi),%rdi
@@ -1486,7 +1752,7 @@
 
 	mulq	%rbx
 	addq	%rax,%r10
-	movq	48(%rbp),%rax
+	movq	24(%rbp),%rax
 	adcq	$0,%rdx
 	addq	%r10,%r9
 	movq	%rdx,%r10
@@ -1494,7 +1760,7 @@
 
 	mulq	%rbx
 	addq	%rax,%r11
-	movq	64(%rbp),%rax
+	movq	32(%rbp),%rax
 	adcq	$0,%rdx
 	addq	%r11,%r10
 	movq	%rdx,%r11
@@ -1502,7 +1768,7 @@
 
 	mulq	%rbx
 	addq	%rax,%r12
-	movq	80(%rbp),%rax
+	movq	40(%rbp),%rax
 	adcq	$0,%rdx
 	addq	%r12,%r11
 	movq	%rdx,%r12
@@ -1510,7 +1776,7 @@
 
 	mulq	%rbx
 	addq	%rax,%r13
-	movq	96(%rbp),%rax
+	movq	48(%rbp),%rax
 	adcq	$0,%rdx
 	addq	%r13,%r12
 	movq	%rdx,%r13
@@ -1518,7 +1784,7 @@
 
 	mulq	%rbx
 	addq	%rax,%r14
-	movq	112(%rbp),%rax
+	movq	56(%rbp),%rax
 	adcq	$0,%rdx
 	addq	%r14,%r13
 	movq	%rdx,%r14
@@ -1536,7 +1802,7 @@
 	decl	%ecx
 	jnz	L$8x_tail
 
-	leaq	128(%rbp),%rbp
+	leaq	64(%rbp),%rbp
 	movq	8+8(%rsp),%rdx
 	cmpq	0+8(%rsp),%rbp
 	jae	L$8x_tail_done
@@ -1560,6 +1826,15 @@
 .p2align	5
 L$8x_tail_done:
 	addq	(%rdx),%r8
+	adcq	$0,%r9
+	adcq	$0,%r10
+	adcq	$0,%r11
+	adcq	$0,%r12
+	adcq	$0,%r13
+	adcq	$0,%r14
+	adcq	$0,%r15
+
+
 	xorq	%rax,%rax
 
 	negq	%rsi
@@ -1573,7 +1848,7 @@
 	adcq	48(%rdi),%r14
 	adcq	56(%rdi),%r15
 	adcq	$0,%rax
-	movq	-16(%rbp),%rcx
+	movq	-8(%rbp),%rcx
 	xorq	%rsi,%rsi
 
 .byte	102,72,15,126,213
@@ -1591,40 +1866,58 @@
 
 	cmpq	%rdx,%rdi
 	jb	L$8x_reduction_loop
+	.byte	0xf3,0xc3
 
-	subq	%r15,%rcx
-	leaq	(%rdi,%r9,1),%rbx
-	adcq	%rsi,%rsi
-	movq	%r9,%rcx
-	orq	%rsi,%rax
-.byte	102,72,15,126,207
-	xorq	$1,%rax
-.byte	102,72,15,126,206
-	leaq	(%rbp,%rax,8),%rbp
-	sarq	$3+2,%rcx
-	jmp	L$sqr4x_sub
 
 .p2align	5
+__bn_post4x_internal:
+	movq	0(%rbp),%r12
+	leaq	(%rdi,%r9,1),%rbx
+	movq	%r9,%rcx
+.byte	102,72,15,126,207
+	negq	%rax
+.byte	102,72,15,126,206
+	sarq	$3+2,%rcx
+	decq	%r12
+	xorq	%r10,%r10
+	movq	8(%rbp),%r13
+	movq	16(%rbp),%r14
+	movq	24(%rbp),%r15
+	jmp	L$sqr4x_sub_entry
+
+.p2align	4
 L$sqr4x_sub:
-.byte	0x66
-	movq	0(%rbx),%r12
-	movq	8(%rbx),%r13
-	sbbq	0(%rbp),%r12
-	movq	16(%rbx),%r14
-	sbbq	16(%rbp),%r13
-	movq	24(%rbx),%r15
-	leaq	32(%rbx),%rbx
-	sbbq	32(%rbp),%r14
+	movq	0(%rbp),%r12
+	movq	8(%rbp),%r13
+	movq	16(%rbp),%r14
+	movq	24(%rbp),%r15
+L$sqr4x_sub_entry:
+	leaq	32(%rbp),%rbp
+	notq	%r12
+	notq	%r13
+	notq	%r14
+	notq	%r15
+	andq	%rax,%r12
+	andq	%rax,%r13
+	andq	%rax,%r14
+	andq	%rax,%r15
+
+	negq	%r10
+	adcq	0(%rbx),%r12
+	adcq	8(%rbx),%r13
+	adcq	16(%rbx),%r14
+	adcq	24(%rbx),%r15
 	movq	%r12,0(%rdi)
-	sbbq	48(%rbp),%r15
-	leaq	64(%rbp),%rbp
+	leaq	32(%rbx),%rbx
 	movq	%r13,8(%rdi)
+	sbbq	%r10,%r10
 	movq	%r14,16(%rdi)
 	movq	%r15,24(%rdi)
 	leaq	32(%rdi),%rdi
 
 	incq	%rcx
 	jnz	L$sqr4x_sub
+
 	movq	%r9,%r10
 	negq	%r9
 	.byte	0xf3,0xc3
@@ -1651,10 +1944,9 @@
 	pushq	%r13
 	pushq	%r14
 	pushq	%r15
-.byte	0x67
-	movl	%r9d,%r10d
+
 	shll	$3,%r9d
-	shll	$3+2,%r10d
+	leaq	(%r9,%r9,2),%r10
 	negq	%r9
 	movq	(%r8),%r8
 
@@ -1664,19 +1956,20 @@
 
 
 
-	leaq	-64(%rsp,%r9,2),%r11
-	subq	%rsi,%r11
+
+	leaq	-320(%rsp,%r9,2),%r11
+	subq	%rdi,%r11
 	andq	$4095,%r11
 	cmpq	%r11,%r10
 	jb	L$from_sp_alt
 	subq	%r11,%rsp
-	leaq	-64(%rsp,%r9,2),%rsp
+	leaq	-320(%rsp,%r9,2),%rsp
 	jmp	L$from_sp_done
 
 .p2align	5
 L$from_sp_alt:
-	leaq	4096-64(,%r9,2),%r10
-	leaq	-64(%rsp,%r9,2),%rsp
+	leaq	4096-320(,%r9,2),%r10
+	leaq	-320(%rsp,%r9,2),%rsp
 	subq	%r10,%r11
 	movq	$0,%r10
 	cmovcq	%r10,%r11
@@ -1727,7 +2020,8 @@
 .byte	0x67
 	movq	%rcx,%rbp
 .byte	102,73,15,110,218
-	call	sqr8x_reduction
+	call	__bn_sqr8x_reduction
+	call	__bn_post4x_internal
 
 	pxor	%xmm0,%xmm0
 	leaq	48(%rsp),%rax
@@ -1777,46 +2071,170 @@
 .globl	_bn_gather5
 .private_extern _bn_gather5
 
-.p2align	4
+.p2align	5
 _bn_gather5:
-	movl	%ecx,%r11d
-	shrl	$3,%ecx
-	andq	$7,%r11
-	notl	%ecx
-	leaq	L$magic_masks(%rip),%rax
-	andl	$3,%ecx
-	leaq	128(%rdx,%r11,8),%rdx
-	movq	0(%rax,%rcx,8),%xmm4
-	movq	8(%rax,%rcx,8),%xmm5
-	movq	16(%rax,%rcx,8),%xmm6
-	movq	24(%rax,%rcx,8),%xmm7
-	jmp	L$gather
-.p2align	4
-L$gather:
-	movq	-128(%rdx),%xmm0
-	movq	-64(%rdx),%xmm1
-	pand	%xmm4,%xmm0
-	movq	0(%rdx),%xmm2
-	pand	%xmm5,%xmm1
-	movq	64(%rdx),%xmm3
-	pand	%xmm6,%xmm2
-	por	%xmm1,%xmm0
-	pand	%xmm7,%xmm3
-.byte	0x67,0x67
-	por	%xmm2,%xmm0
-	leaq	256(%rdx),%rdx
-	por	%xmm3,%xmm0
+L$SEH_begin_bn_gather5:
 
+.byte	0x4c,0x8d,0x14,0x24
+.byte	0x48,0x81,0xec,0x08,0x01,0x00,0x00
+	leaq	L$inc(%rip),%rax
+	andq	$-16,%rsp
+
+	movd	%ecx,%xmm5
+	movdqa	0(%rax),%xmm0
+	movdqa	16(%rax),%xmm1
+	leaq	128(%rdx),%r11
+	leaq	128(%rsp),%rax
+
+	pshufd	$0,%xmm5,%xmm5
+	movdqa	%xmm1,%xmm4
+	movdqa	%xmm1,%xmm2
+	paddd	%xmm0,%xmm1
+	pcmpeqd	%xmm5,%xmm0
+	movdqa	%xmm4,%xmm3
+
+	paddd	%xmm1,%xmm2
+	pcmpeqd	%xmm5,%xmm1
+	movdqa	%xmm0,-128(%rax)
+	movdqa	%xmm4,%xmm0
+
+	paddd	%xmm2,%xmm3
+	pcmpeqd	%xmm5,%xmm2
+	movdqa	%xmm1,-112(%rax)
+	movdqa	%xmm4,%xmm1
+
+	paddd	%xmm3,%xmm0
+	pcmpeqd	%xmm5,%xmm3
+	movdqa	%xmm2,-96(%rax)
+	movdqa	%xmm4,%xmm2
+	paddd	%xmm0,%xmm1
+	pcmpeqd	%xmm5,%xmm0
+	movdqa	%xmm3,-80(%rax)
+	movdqa	%xmm4,%xmm3
+
+	paddd	%xmm1,%xmm2
+	pcmpeqd	%xmm5,%xmm1
+	movdqa	%xmm0,-64(%rax)
+	movdqa	%xmm4,%xmm0
+
+	paddd	%xmm2,%xmm3
+	pcmpeqd	%xmm5,%xmm2
+	movdqa	%xmm1,-48(%rax)
+	movdqa	%xmm4,%xmm1
+
+	paddd	%xmm3,%xmm0
+	pcmpeqd	%xmm5,%xmm3
+	movdqa	%xmm2,-32(%rax)
+	movdqa	%xmm4,%xmm2
+	paddd	%xmm0,%xmm1
+	pcmpeqd	%xmm5,%xmm0
+	movdqa	%xmm3,-16(%rax)
+	movdqa	%xmm4,%xmm3
+
+	paddd	%xmm1,%xmm2
+	pcmpeqd	%xmm5,%xmm1
+	movdqa	%xmm0,0(%rax)
+	movdqa	%xmm4,%xmm0
+
+	paddd	%xmm2,%xmm3
+	pcmpeqd	%xmm5,%xmm2
+	movdqa	%xmm1,16(%rax)
+	movdqa	%xmm4,%xmm1
+
+	paddd	%xmm3,%xmm0
+	pcmpeqd	%xmm5,%xmm3
+	movdqa	%xmm2,32(%rax)
+	movdqa	%xmm4,%xmm2
+	paddd	%xmm0,%xmm1
+	pcmpeqd	%xmm5,%xmm0
+	movdqa	%xmm3,48(%rax)
+	movdqa	%xmm4,%xmm3
+
+	paddd	%xmm1,%xmm2
+	pcmpeqd	%xmm5,%xmm1
+	movdqa	%xmm0,64(%rax)
+	movdqa	%xmm4,%xmm0
+
+	paddd	%xmm2,%xmm3
+	pcmpeqd	%xmm5,%xmm2
+	movdqa	%xmm1,80(%rax)
+	movdqa	%xmm4,%xmm1
+
+	paddd	%xmm3,%xmm0
+	pcmpeqd	%xmm5,%xmm3
+	movdqa	%xmm2,96(%rax)
+	movdqa	%xmm4,%xmm2
+	movdqa	%xmm3,112(%rax)
+	jmp	L$gather
+
+.p2align	5
+L$gather:
+	pxor	%xmm4,%xmm4
+	pxor	%xmm5,%xmm5
+	movdqa	-128(%r11),%xmm0
+	movdqa	-112(%r11),%xmm1
+	movdqa	-96(%r11),%xmm2
+	pand	-128(%rax),%xmm0
+	movdqa	-80(%r11),%xmm3
+	pand	-112(%rax),%xmm1
+	por	%xmm0,%xmm4
+	pand	-96(%rax),%xmm2
+	por	%xmm1,%xmm5
+	pand	-80(%rax),%xmm3
+	por	%xmm2,%xmm4
+	por	%xmm3,%xmm5
+	movdqa	-64(%r11),%xmm0
+	movdqa	-48(%r11),%xmm1
+	movdqa	-32(%r11),%xmm2
+	pand	-64(%rax),%xmm0
+	movdqa	-16(%r11),%xmm3
+	pand	-48(%rax),%xmm1
+	por	%xmm0,%xmm4
+	pand	-32(%rax),%xmm2
+	por	%xmm1,%xmm5
+	pand	-16(%rax),%xmm3
+	por	%xmm2,%xmm4
+	por	%xmm3,%xmm5
+	movdqa	0(%r11),%xmm0
+	movdqa	16(%r11),%xmm1
+	movdqa	32(%r11),%xmm2
+	pand	0(%rax),%xmm0
+	movdqa	48(%r11),%xmm3
+	pand	16(%rax),%xmm1
+	por	%xmm0,%xmm4
+	pand	32(%rax),%xmm2
+	por	%xmm1,%xmm5
+	pand	48(%rax),%xmm3
+	por	%xmm2,%xmm4
+	por	%xmm3,%xmm5
+	movdqa	64(%r11),%xmm0
+	movdqa	80(%r11),%xmm1
+	movdqa	96(%r11),%xmm2
+	pand	64(%rax),%xmm0
+	movdqa	112(%r11),%xmm3
+	pand	80(%rax),%xmm1
+	por	%xmm0,%xmm4
+	pand	96(%rax),%xmm2
+	por	%xmm1,%xmm5
+	pand	112(%rax),%xmm3
+	por	%xmm2,%xmm4
+	por	%xmm3,%xmm5
+	por	%xmm5,%xmm4
+	leaq	256(%r11),%r11
+	pshufd	$0x4e,%xmm4,%xmm0
+	por	%xmm4,%xmm0
 	movq	%xmm0,(%rdi)
 	leaq	8(%rdi),%rdi
 	subl	$1,%esi
 	jnz	L$gather
+
+	leaq	(%r10),%rsp
 	.byte	0xf3,0xc3
 L$SEH_end_bn_gather5:
 
 .p2align	6
-L$magic_masks:
-.long	0,0, 0,0, 0,0, -1,-1
-.long	0,0, 0,0, 0,0,  0,0
+L$inc:
+.long	0,0, 1,1
+.long	2,2, 2,2
 .byte	77,111,110,116,103,111,109,101,114,121,32,77,117,108,116,105,112,108,105,99,97,116,105,111,110,32,119,105,116,104,32,115,99,97,116,116,101,114,47,103,97,116,104,101,114,32,102,111,114,32,120,56,54,95,54,52,44,32,67,82,89,80,84,79,71,65,77,83,32,98,121,32,60,97,112,112,114,111,64,111,112,101,110,115,115,108,46,111,114,103,62,0
 #endif

third_party/boringssl/mac-x86_64/crypto/chacha/chacha-x86_64.S

diff --git a/third_party/boringssl/mac-x86_64/crypto/chacha/chacha-x86_64.S b/third_party/boringssl/mac-x86_64/crypto/chacha/chacha-x86_64.S
new file mode 100644
index 0000000..c3554c8
--- /dev/null
+++ b/third_party/boringssl/mac-x86_64/crypto/chacha/chacha-x86_64.S
@@ -0,0 +1,1584 @@
+#if defined(__x86_64__)
+.text	
+
+
+
+.p2align	6
+L$zero:
+.long	0,0,0,0
+L$one:
+.long	1,0,0,0
+L$inc:
+.long	0,1,2,3
+L$four:
+.long	4,4,4,4
+L$incy:
+.long	0,2,4,6,1,3,5,7
+L$eight:
+.long	8,8,8,8,8,8,8,8
+L$rot16:
+.byte	0x2,0x3,0x0,0x1, 0x6,0x7,0x4,0x5, 0xa,0xb,0x8,0x9, 0xe,0xf,0xc,0xd
+L$rot24:
+.byte	0x3,0x0,0x1,0x2, 0x7,0x4,0x5,0x6, 0xb,0x8,0x9,0xa, 0xf,0xc,0xd,0xe
+L$sigma:
+.byte	101,120,112,97,110,100,32,51,50,45,98,121,116,101,32,107,0
+.byte	67,104,97,67,104,97,50,48,32,102,111,114,32,120,56,54,95,54,52,44,32,67,82,89,80,84,79,71,65,77,83,32,98,121,32,60,97,112,112,114,111,64,111,112,101,110,115,115,108,46,111,114,103,62,0
+.globl	_ChaCha20_ctr32
+.private_extern _ChaCha20_ctr32
+
+.p2align	6
+_ChaCha20_ctr32:
+	cmpq	$0,%rdx
+	je	L$no_data
+	movq	_OPENSSL_ia32cap_P+4(%rip),%r10
+	testl	$512,%r10d
+	jnz	L$ChaCha20_ssse3
+
+	pushq	%rbx
+	pushq	%rbp
+	pushq	%r12
+	pushq	%r13
+	pushq	%r14
+	pushq	%r15
+	subq	$64+24,%rsp
+
+
+	movdqu	(%rcx),%xmm1
+	movdqu	16(%rcx),%xmm2
+	movdqu	(%r8),%xmm3
+	movdqa	L$one(%rip),%xmm4
+
+
+	movdqa	%xmm1,16(%rsp)
+	movdqa	%xmm2,32(%rsp)
+	movdqa	%xmm3,48(%rsp)
+	movq	%rdx,%rbp
+	jmp	L$oop_outer
+
+.p2align	5
+L$oop_outer:
+	movl	$0x61707865,%eax
+	movl	$0x3320646e,%ebx
+	movl	$0x79622d32,%ecx
+	movl	$0x6b206574,%edx
+	movl	16(%rsp),%r8d
+	movl	20(%rsp),%r9d
+	movl	24(%rsp),%r10d
+	movl	28(%rsp),%r11d
+	movd	%xmm3,%r12d
+	movl	52(%rsp),%r13d
+	movl	56(%rsp),%r14d
+	movl	60(%rsp),%r15d
+
+	movq	%rbp,64+0(%rsp)
+	movl	$10,%ebp
+	movq	%rsi,64+8(%rsp)
+.byte	102,72,15,126,214
+	movq	%rdi,64+16(%rsp)
+	movq	%rsi,%rdi
+	shrq	$32,%rdi
+	jmp	L$oop
+
+.p2align	5
+L$oop:
+	addl	%r8d,%eax
+	xorl	%eax,%r12d
+	roll	$16,%r12d
+	addl	%r9d,%ebx
+	xorl	%ebx,%r13d
+	roll	$16,%r13d
+	addl	%r12d,%esi
+	xorl	%esi,%r8d
+	roll	$12,%r8d
+	addl	%r13d,%edi
+	xorl	%edi,%r9d
+	roll	$12,%r9d
+	addl	%r8d,%eax
+	xorl	%eax,%r12d
+	roll	$8,%r12d
+	addl	%r9d,%ebx
+	xorl	%ebx,%r13d
+	roll	$8,%r13d
+	addl	%r12d,%esi
+	xorl	%esi,%r8d
+	roll	$7,%r8d
+	addl	%r13d,%edi
+	xorl	%edi,%r9d
+	roll	$7,%r9d
+	movl	%esi,32(%rsp)
+	movl	%edi,36(%rsp)
+	movl	40(%rsp),%esi
+	movl	44(%rsp),%edi
+	addl	%r10d,%ecx
+	xorl	%ecx,%r14d
+	roll	$16,%r14d
+	addl	%r11d,%edx
+	xorl	%edx,%r15d
+	roll	$16,%r15d
+	addl	%r14d,%esi
+	xorl	%esi,%r10d
+	roll	$12,%r10d
+	addl	%r15d,%edi
+	xorl	%edi,%r11d
+	roll	$12,%r11d
+	addl	%r10d,%ecx
+	xorl	%ecx,%r14d
+	roll	$8,%r14d
+	addl	%r11d,%edx
+	xorl	%edx,%r15d
+	roll	$8,%r15d
+	addl	%r14d,%esi
+	xorl	%esi,%r10d
+	roll	$7,%r10d
+	addl	%r15d,%edi
+	xorl	%edi,%r11d
+	roll	$7,%r11d
+	addl	%r9d,%eax
+	xorl	%eax,%r15d
+	roll	$16,%r15d
+	addl	%r10d,%ebx
+	xorl	%ebx,%r12d
+	roll	$16,%r12d
+	addl	%r15d,%esi
+	xorl	%esi,%r9d
+	roll	$12,%r9d
+	addl	%r12d,%edi
+	xorl	%edi,%r10d
+	roll	$12,%r10d
+	addl	%r9d,%eax
+	xorl	%eax,%r15d
+	roll	$8,%r15d
+	addl	%r10d,%ebx
+	xorl	%ebx,%r12d
+	roll	$8,%r12d
+	addl	%r15d,%esi
+	xorl	%esi,%r9d
+	roll	$7,%r9d
+	addl	%r12d,%edi
+	xorl	%edi,%r10d
+	roll	$7,%r10d
+	movl	%esi,40(%rsp)
+	movl	%edi,44(%rsp)
+	movl	32(%rsp),%esi
+	movl	36(%rsp),%edi
+	addl	%r11d,%ecx
+	xorl	%ecx,%r13d
+	roll	$16,%r13d
+	addl	%r8d,%edx
+	xorl	%edx,%r14d
+	roll	$16,%r14d
+	addl	%r13d,%esi
+	xorl	%esi,%r11d
+	roll	$12,%r11d
+	addl	%r14d,%edi
+	xorl	%edi,%r8d
+	roll	$12,%r8d
+	addl	%r11d,%ecx
+	xorl	%ecx,%r13d
+	roll	$8,%r13d
+	addl	%r8d,%edx
+	xorl	%edx,%r14d
+	roll	$8,%r14d
+	addl	%r13d,%esi
+	xorl	%esi,%r11d
+	roll	$7,%r11d
+	addl	%r14d,%edi
+	xorl	%edi,%r8d
+	roll	$7,%r8d
+	decl	%ebp
+	jnz	L$oop
+	movl	%edi,36(%rsp)
+	movl	%esi,32(%rsp)
+	movq	64(%rsp),%rbp
+	movdqa	%xmm2,%xmm1
+	movq	64+8(%rsp),%rsi
+	paddd	%xmm4,%xmm3
+	movq	64+16(%rsp),%rdi
+
+	addl	$0x61707865,%eax
+	addl	$0x3320646e,%ebx
+	addl	$0x79622d32,%ecx
+	addl	$0x6b206574,%edx
+	addl	16(%rsp),%r8d
+	addl	20(%rsp),%r9d
+	addl	24(%rsp),%r10d
+	addl	28(%rsp),%r11d
+	addl	48(%rsp),%r12d
+	addl	52(%rsp),%r13d
+	addl	56(%rsp),%r14d
+	addl	60(%rsp),%r15d
+	paddd	32(%rsp),%xmm1
+
+	cmpq	$64,%rbp
+	jb	L$tail
+
+	xorl	0(%rsi),%eax
+	xorl	4(%rsi),%ebx
+	xorl	8(%rsi),%ecx
+	xorl	12(%rsi),%edx
+	xorl	16(%rsi),%r8d
+	xorl	20(%rsi),%r9d
+	xorl	24(%rsi),%r10d
+	xorl	28(%rsi),%r11d
+	movdqu	32(%rsi),%xmm0
+	xorl	48(%rsi),%r12d
+	xorl	52(%rsi),%r13d
+	xorl	56(%rsi),%r14d
+	xorl	60(%rsi),%r15d
+	leaq	64(%rsi),%rsi
+	pxor	%xmm1,%xmm0
+
+	movdqa	%xmm2,32(%rsp)
+	movd	%xmm3,48(%rsp)
+
+	movl	%eax,0(%rdi)
+	movl	%ebx,4(%rdi)
+	movl	%ecx,8(%rdi)
+	movl	%edx,12(%rdi)
+	movl	%r8d,16(%rdi)
+	movl	%r9d,20(%rdi)
+	movl	%r10d,24(%rdi)
+	movl	%r11d,28(%rdi)
+	movdqu	%xmm0,32(%rdi)
+	movl	%r12d,48(%rdi)
+	movl	%r13d,52(%rdi)
+	movl	%r14d,56(%rdi)
+	movl	%r15d,60(%rdi)
+	leaq	64(%rdi),%rdi
+
+	subq	$64,%rbp
+	jnz	L$oop_outer
+
+	jmp	L$done
+
+.p2align	4
+L$tail:
+	movl	%eax,0(%rsp)
+	movl	%ebx,4(%rsp)
+	xorq	%rbx,%rbx
+	movl	%ecx,8(%rsp)
+	movl	%edx,12(%rsp)
+	movl	%r8d,16(%rsp)
+	movl	%r9d,20(%rsp)
+	movl	%r10d,24(%rsp)
+	movl	%r11d,28(%rsp)
+	movdqa	%xmm1,32(%rsp)
+	movl	%r12d,48(%rsp)
+	movl	%r13d,52(%rsp)
+	movl	%r14d,56(%rsp)
+	movl	%r15d,60(%rsp)
+
+L$oop_tail:
+	movzbl	(%rsi,%rbx,1),%eax
+	movzbl	(%rsp,%rbx,1),%edx
+	leaq	1(%rbx),%rbx
+	xorl	%edx,%eax
+	movb	%al,-1(%rdi,%rbx,1)
+	decq	%rbp
+	jnz	L$oop_tail
+
+L$done:
+	addq	$64+24,%rsp
+	popq	%r15
+	popq	%r14
+	popq	%r13
+	popq	%r12
+	popq	%rbp
+	popq	%rbx
+L$no_data:
+	.byte	0xf3,0xc3
+
+
+.p2align	5
+ChaCha20_ssse3:
+L$ChaCha20_ssse3:
+	cmpq	$128,%rdx
+	ja	L$ChaCha20_4x
+
+L$do_sse3_after_all:
+	pushq	%rbx
+	pushq	%rbp
+	pushq	%r12
+	pushq	%r13
+	pushq	%r14
+	pushq	%r15
+
+	subq	$64+24,%rsp
+	movdqa	L$sigma(%rip),%xmm0
+	movdqu	(%rcx),%xmm1
+	movdqu	16(%rcx),%xmm2
+	movdqu	(%r8),%xmm3
+	movdqa	L$rot16(%rip),%xmm6
+	movdqa	L$rot24(%rip),%xmm7
+
+	movdqa	%xmm0,0(%rsp)
+	movdqa	%xmm1,16(%rsp)
+	movdqa	%xmm2,32(%rsp)
+	movdqa	%xmm3,48(%rsp)
+	movl	$10,%ebp
+	jmp	L$oop_ssse3
+
+.p2align	5
+L$oop_outer_ssse3:
+	movdqa	L$one(%rip),%xmm3
+	movdqa	0(%rsp),%xmm0
+	movdqa	16(%rsp),%xmm1
+	movdqa	32(%rsp),%xmm2
+	paddd	48(%rsp),%xmm3
+	movl	$10,%ebp
+	movdqa	%xmm3,48(%rsp)
+	jmp	L$oop_ssse3
+
+.p2align	5
+L$oop_ssse3:
+	paddd	%xmm1,%xmm0
+	pxor	%xmm0,%xmm3
+.byte	102,15,56,0,222
+	paddd	%xmm3,%xmm2
+	pxor	%xmm2,%xmm1
+	movdqa	%xmm1,%xmm4
+	psrld	$20,%xmm1
+	pslld	$12,%xmm4
+	por	%xmm4,%xmm1
+	paddd	%xmm1,%xmm0
+	pxor	%xmm0,%xmm3
+.byte	102,15,56,0,223
+	paddd	%xmm3,%xmm2
+	pxor	%xmm2,%xmm1
+	movdqa	%xmm1,%xmm4
+	psrld	$25,%xmm1
+	pslld	$7,%xmm4
+	por	%xmm4,%xmm1
+	pshufd	$78,%xmm2,%xmm2
+	pshufd	$57,%xmm1,%xmm1
+	pshufd	$147,%xmm3,%xmm3
+	nop
+	paddd	%xmm1,%xmm0
+	pxor	%xmm0,%xmm3
+.byte	102,15,56,0,222
+	paddd	%xmm3,%xmm2
+	pxor	%xmm2,%xmm1
+	movdqa	%xmm1,%xmm4
+	psrld	$20,%xmm1
+	pslld	$12,%xmm4
+	por	%xmm4,%xmm1
+	paddd	%xmm1,%xmm0
+	pxor	%xmm0,%xmm3
+.byte	102,15,56,0,223
+	paddd	%xmm3,%xmm2
+	pxor	%xmm2,%xmm1
+	movdqa	%xmm1,%xmm4
+	psrld	$25,%xmm1
+	pslld	$7,%xmm4
+	por	%xmm4,%xmm1
+	pshufd	$78,%xmm2,%xmm2
+	pshufd	$147,%xmm1,%xmm1
+	pshufd	$57,%xmm3,%xmm3
+	decl	%ebp
+	jnz	L$oop_ssse3
+	paddd	0(%rsp),%xmm0
+	paddd	16(%rsp),%xmm1
+	paddd	32(%rsp),%xmm2
+	paddd	48(%rsp),%xmm3
+
+	cmpq	$64,%rdx
+	jb	L$tail_ssse3
+
+	movdqu	0(%rsi),%xmm4
+	movdqu	16(%rsi),%xmm5
+	pxor	%xmm4,%xmm0
+	movdqu	32(%rsi),%xmm4
+	pxor	%xmm5,%xmm1
+	movdqu	48(%rsi),%xmm5
+	leaq	64(%rsi),%rsi
+	pxor	%xmm4,%xmm2
+	pxor	%xmm5,%xmm3
+
+	movdqu	%xmm0,0(%rdi)
+	movdqu	%xmm1,16(%rdi)
+	movdqu	%xmm2,32(%rdi)
+	movdqu	%xmm3,48(%rdi)
+	leaq	64(%rdi),%rdi
+
+	subq	$64,%rdx
+	jnz	L$oop_outer_ssse3
+
+	jmp	L$done_ssse3
+
+.p2align	4
+L$tail_ssse3:
+	movdqa	%xmm0,0(%rsp)
+	movdqa	%xmm1,16(%rsp)
+	movdqa	%xmm2,32(%rsp)
+	movdqa	%xmm3,48(%rsp)
+	xorq	%rbx,%rbx
+
+L$oop_tail_ssse3:
+	movzbl	(%rsi,%rbx,1),%eax
+	movzbl	(%rsp,%rbx,1),%ecx
+	leaq	1(%rbx),%rbx
+	xorl	%ecx,%eax
+	movb	%al,-1(%rdi,%rbx,1)
+	decq	%rdx
+	jnz	L$oop_tail_ssse3
+
+L$done_ssse3:
+	addq	$64+24,%rsp
+	popq	%r15
+	popq	%r14
+	popq	%r13
+	popq	%r12
+	popq	%rbp
+	popq	%rbx
+	.byte	0xf3,0xc3
+
+
+.p2align	5
+ChaCha20_4x:
+L$ChaCha20_4x:
+	movq	%r10,%r11
+	shrq	$32,%r10
+	testq	$32,%r10
+	jnz	L$ChaCha20_8x
+	cmpq	$192,%rdx
+	ja	L$proceed4x
+
+	andq	$71303168,%r11
+	cmpq	$4194304,%r11
+	je	L$do_sse3_after_all
+
+L$proceed4x:
+	leaq	-120(%rsp),%r11
+	subq	$0x148+0,%rsp
+	movdqa	L$sigma(%rip),%xmm11
+	movdqu	(%rcx),%xmm15
+	movdqu	16(%rcx),%xmm7
+	movdqu	(%r8),%xmm3
+	leaq	256(%rsp),%rcx
+	leaq	L$rot16(%rip),%r10
+	leaq	L$rot24(%rip),%r11
+
+	pshufd	$0x00,%xmm11,%xmm8
+	pshufd	$0x55,%xmm11,%xmm9
+	movdqa	%xmm8,64(%rsp)
+	pshufd	$0xaa,%xmm11,%xmm10
+	movdqa	%xmm9,80(%rsp)
+	pshufd	$0xff,%xmm11,%xmm11
+	movdqa	%xmm10,96(%rsp)
+	movdqa	%xmm11,112(%rsp)
+
+	pshufd	$0x00,%xmm15,%xmm12
+	pshufd	$0x55,%xmm15,%xmm13
+	movdqa	%xmm12,128-256(%rcx)
+	pshufd	$0xaa,%xmm15,%xmm14
+	movdqa	%xmm13,144-256(%rcx)
+	pshufd	$0xff,%xmm15,%xmm15
+	movdqa	%xmm14,160-256(%rcx)
+	movdqa	%xmm15,176-256(%rcx)
+
+	pshufd	$0x00,%xmm7,%xmm4
+	pshufd	$0x55,%xmm7,%xmm5
+	movdqa	%xmm4,192-256(%rcx)
+	pshufd	$0xaa,%xmm7,%xmm6
+	movdqa	%xmm5,208-256(%rcx)
+	pshufd	$0xff,%xmm7,%xmm7
+	movdqa	%xmm6,224-256(%rcx)
+	movdqa	%xmm7,240-256(%rcx)
+
+	pshufd	$0x00,%xmm3,%xmm0
+	pshufd	$0x55,%xmm3,%xmm1
+	paddd	L$inc(%rip),%xmm0
+	pshufd	$0xaa,%xmm3,%xmm2
+	movdqa	%xmm1,272-256(%rcx)
+	pshufd	$0xff,%xmm3,%xmm3
+	movdqa	%xmm2,288-256(%rcx)
+	movdqa	%xmm3,304-256(%rcx)
+
+	jmp	L$oop_enter4x
+
+.p2align	5
+L$oop_outer4x:
+	movdqa	64(%rsp),%xmm8
+	movdqa	80(%rsp),%xmm9
+	movdqa	96(%rsp),%xmm10
+	movdqa	112(%rsp),%xmm11
+	movdqa	128-256(%rcx),%xmm12
+	movdqa	144-256(%rcx),%xmm13
+	movdqa	160-256(%rcx),%xmm14
+	movdqa	176-256(%rcx),%xmm15
+	movdqa	192-256(%rcx),%xmm4
+	movdqa	208-256(%rcx),%xmm5
+	movdqa	224-256(%rcx),%xmm6
+	movdqa	240-256(%rcx),%xmm7
+	movdqa	256-256(%rcx),%xmm0
+	movdqa	272-256(%rcx),%xmm1
+	movdqa	288-256(%rcx),%xmm2
+	movdqa	304-256(%rcx),%xmm3
+	paddd	L$four(%rip),%xmm0
+
+L$oop_enter4x:
+	movdqa	%xmm6,32(%rsp)
+	movdqa	%xmm7,48(%rsp)
+	movdqa	(%r10),%xmm7
+	movl	$10,%eax
+	movdqa	%xmm0,256-256(%rcx)
+	jmp	L$oop4x
+
+.p2align	5
+L$oop4x:
+	paddd	%xmm12,%xmm8
+	paddd	%xmm13,%xmm9
+	pxor	%xmm8,%xmm0
+	pxor	%xmm9,%xmm1
+.byte	102,15,56,0,199
+.byte	102,15,56,0,207
+	paddd	%xmm0,%xmm4
+	paddd	%xmm1,%xmm5
+	pxor	%xmm4,%xmm12
+	pxor	%xmm5,%xmm13
+	movdqa	%xmm12,%xmm6
+	pslld	$12,%xmm12
+	psrld	$20,%xmm6
+	movdqa	%xmm13,%xmm7
+	pslld	$12,%xmm13
+	por	%xmm6,%xmm12
+	psrld	$20,%xmm7
+	movdqa	(%r11),%xmm6
+	por	%xmm7,%xmm13
+	paddd	%xmm12,%xmm8
+	paddd	%xmm13,%xmm9
+	pxor	%xmm8,%xmm0
+	pxor	%xmm9,%xmm1
+.byte	102,15,56,0,198
+.byte	102,15,56,0,206
+	paddd	%xmm0,%xmm4
+	paddd	%xmm1,%xmm5
+	pxor	%xmm4,%xmm12
+	pxor	%xmm5,%xmm13
+	movdqa	%xmm12,%xmm7
+	pslld	$7,%xmm12
+	psrld	$25,%xmm7
+	movdqa	%xmm13,%xmm6
+	pslld	$7,%xmm13
+	por	%xmm7,%xmm12
+	psrld	$25,%xmm6
+	movdqa	(%r10),%xmm7
+	por	%xmm6,%xmm13
+	movdqa	%xmm4,0(%rsp)
+	movdqa	%xmm5,16(%rsp)
+	movdqa	32(%rsp),%xmm4
+	movdqa	48(%rsp),%xmm5
+	paddd	%xmm14,%xmm10
+	paddd	%xmm15,%xmm11
+	pxor	%xmm10,%xmm2
+	pxor	%xmm11,%xmm3
+.byte	102,15,56,0,215
+.byte	102,15,56,0,223
+	paddd	%xmm2,%xmm4
+	paddd	%xmm3,%xmm5
+	pxor	%xmm4,%xmm14
+	pxor	%xmm5,%xmm15
+	movdqa	%xmm14,%xmm6
+	pslld	$12,%xmm14
+	psrld	$20,%xmm6
+	movdqa	%xmm15,%xmm7
+	pslld	$12,%xmm15
+	por	%xmm6,%xmm14
+	psrld	$20,%xmm7
+	movdqa	(%r11),%xmm6
+	por	%xmm7,%xmm15
+	paddd	%xmm14,%xmm10
+	paddd	%xmm15,%xmm11
+	pxor	%xmm10,%xmm2
+	pxor	%xmm11,%xmm3
+.byte	102,15,56,0,214
+.byte	102,15,56,0,222
+	paddd	%xmm2,%xmm4
+	paddd	%xmm3,%xmm5
+	pxor	%xmm4,%xmm14
+	pxor	%xmm5,%xmm15
+	movdqa	%xmm14,%xmm7
+	pslld	$7,%xmm14
+	psrld	$25,%xmm7
+	movdqa	%xmm15,%xmm6
+	pslld	$7,%xmm15
+	por	%xmm7,%xmm14
+	psrld	$25,%xmm6
+	movdqa	(%r10),%xmm7
+	por	%xmm6,%xmm15
+	paddd	%xmm13,%xmm8
+	paddd	%xmm14,%xmm9
+	pxor	%xmm8,%xmm3
+	pxor	%xmm9,%xmm0
+.byte	102,15,56,0,223
+.byte	102,15,56,0,199
+	paddd	%xmm3,%xmm4
+	paddd	%xmm0,%xmm5
+	pxor	%xmm4,%xmm13
+	pxor	%xmm5,%xmm14
+	movdqa	%xmm13,%xmm6
+	pslld	$12,%xmm13
+	psrld	$20,%xmm6
+	movdqa	%xmm14,%xmm7
+	pslld	$12,%xmm14
+	por	%xmm6,%xmm13
+	psrld	$20,%xmm7
+	movdqa	(%r11),%xmm6
+	por	%xmm7,%xmm14
+	paddd	%xmm13,%xmm8
+	paddd	%xmm14,%xmm9
+	pxor	%xmm8,%xmm3
+	pxor	%xmm9,%xmm0
+.byte	102,15,56,0,222
+.byte	102,15,56,0,198
+	paddd	%xmm3,%xmm4
+	paddd	%xmm0,%xmm5
+	pxor	%xmm4,%xmm13
+	pxor	%xmm5,%xmm14
+	movdqa	%xmm13,%xmm7
+	pslld	$7,%xmm13
+	psrld	$25,%xmm7
+	movdqa	%xmm14,%xmm6
+	pslld	$7,%xmm14
+	por	%xmm7,%xmm13
+	psrld	$25,%xmm6
+	movdqa	(%r10),%xmm7
+	por	%xmm6,%xmm14
+	movdqa	%xmm4,32(%rsp)
+	movdqa	%xmm5,48(%rsp)
+	movdqa	0(%rsp),%xmm4
+	movdqa	16(%rsp),%xmm5
+	paddd	%xmm15,%xmm10
+	paddd	%xmm12,%xmm11
+	pxor	%xmm10,%xmm1
+	pxor	%xmm11,%xmm2
+.byte	102,15,56,0,207
+.byte	102,15,56,0,215
+	paddd	%xmm1,%xmm4
+	paddd	%xmm2,%xmm5
+	pxor	%xmm4,%xmm15
+	pxor	%xmm5,%xmm12
+	movdqa	%xmm15,%xmm6
+	pslld	$12,%xmm15
+	psrld	$20,%xmm6
+	movdqa	%xmm12,%xmm7
+	pslld	$12,%xmm12
+	por	%xmm6,%xmm15
+	psrld	$20,%xmm7
+	movdqa	(%r11),%xmm6
+	por	%xmm7,%xmm12
+	paddd	%xmm15,%xmm10
+	paddd	%xmm12,%xmm11
+	pxor	%xmm10,%xmm1
+	pxor	%xmm11,%xmm2
+.byte	102,15,56,0,206
+.byte	102,15,56,0,214
+	paddd	%xmm1,%xmm4
+	paddd	%xmm2,%xmm5
+	pxor	%xmm4,%xmm15
+	pxor	%xmm5,%xmm12
+	movdqa	%xmm15,%xmm7
+	pslld	$7,%xmm15
+	psrld	$25,%xmm7
+	movdqa	%xmm12,%xmm6
+	pslld	$7,%xmm12
+	por	%xmm7,%xmm15
+	psrld	$25,%xmm6
+	movdqa	(%r10),%xmm7
+	por	%xmm6,%xmm12
+	decl	%eax
+	jnz	L$oop4x
+
+	paddd	64(%rsp),%xmm8
+	paddd	80(%rsp),%xmm9
+	paddd	96(%rsp),%xmm10
+	paddd	112(%rsp),%xmm11
+
+	movdqa	%xmm8,%xmm6
+	punpckldq	%xmm9,%xmm8
+	movdqa	%xmm10,%xmm7
+	punpckldq	%xmm11,%xmm10
+	punpckhdq	%xmm9,%xmm6
+	punpckhdq	%xmm11,%xmm7
+	movdqa	%xmm8,%xmm9
+	punpcklqdq	%xmm10,%xmm8
+	movdqa	%xmm6,%xmm11
+	punpcklqdq	%xmm7,%xmm6
+	punpckhqdq	%xmm10,%xmm9
+	punpckhqdq	%xmm7,%xmm11
+	paddd	128-256(%rcx),%xmm12
+	paddd	144-256(%rcx),%xmm13
+	paddd	160-256(%rcx),%xmm14
+	paddd	176-256(%rcx),%xmm15
+
+	movdqa	%xmm8,0(%rsp)
+	movdqa	%xmm9,16(%rsp)
+	movdqa	32(%rsp),%xmm8
+	movdqa	48(%rsp),%xmm9
+
+	movdqa	%xmm12,%xmm10
+	punpckldq	%xmm13,%xmm12
+	movdqa	%xmm14,%xmm7
+	punpckldq	%xmm15,%xmm14
+	punpckhdq	%xmm13,%xmm10
+	punpckhdq	%xmm15,%xmm7
+	movdqa	%xmm12,%xmm13
+	punpcklqdq	%xmm14,%xmm12
+	movdqa	%xmm10,%xmm15
+	punpcklqdq	%xmm7,%xmm10
+	punpckhqdq	%xmm14,%xmm13
+	punpckhqdq	%xmm7,%xmm15
+	paddd	192-256(%rcx),%xmm4
+	paddd	208-256(%rcx),%xmm5
+	paddd	224-256(%rcx),%xmm8
+	paddd	240-256(%rcx),%xmm9
+
+	movdqa	%xmm6,32(%rsp)
+	movdqa	%xmm11,48(%rsp)
+
+	movdqa	%xmm4,%xmm14
+	punpckldq	%xmm5,%xmm4
+	movdqa	%xmm8,%xmm7
+	punpckldq	%xmm9,%xmm8
+	punpckhdq	%xmm5,%xmm14
+	punpckhdq	%xmm9,%xmm7
+	movdqa	%xmm4,%xmm5
+	punpcklqdq	%xmm8,%xmm4
+	movdqa	%xmm14,%xmm9
+	punpcklqdq	%xmm7,%xmm14
+	punpckhqdq	%xmm8,%xmm5
+	punpckhqdq	%xmm7,%xmm9
+	paddd	256-256(%rcx),%xmm0
+	paddd	272-256(%rcx),%xmm1
+	paddd	288-256(%rcx),%xmm2
+	paddd	304-256(%rcx),%xmm3
+
+	movdqa	%xmm0,%xmm8
+	punpckldq	%xmm1,%xmm0
+	movdqa	%xmm2,%xmm7
+	punpckldq	%xmm3,%xmm2
+	punpckhdq	%xmm1,%xmm8
+	punpckhdq	%xmm3,%xmm7
+	movdqa	%xmm0,%xmm1
+	punpcklqdq	%xmm2,%xmm0
+	movdqa	%xmm8,%xmm3
+	punpcklqdq	%xmm7,%xmm8
+	punpckhqdq	%xmm2,%xmm1
+	punpckhqdq	%xmm7,%xmm3
+	cmpq	$256,%rdx
+	jb	L$tail4x
+
+	movdqu	0(%rsi),%xmm6
+	movdqu	16(%rsi),%xmm11
+	movdqu	32(%rsi),%xmm2
+	movdqu	48(%rsi),%xmm7
+	pxor	0(%rsp),%xmm6
+	pxor	%xmm12,%xmm11
+	pxor	%xmm4,%xmm2
+	pxor	%xmm0,%xmm7
+
+	movdqu	%xmm6,0(%rdi)
+	movdqu	64(%rsi),%xmm6
+	movdqu	%xmm11,16(%rdi)
+	movdqu	80(%rsi),%xmm11
+	movdqu	%xmm2,32(%rdi)
+	movdqu	96(%rsi),%xmm2
+	movdqu	%xmm7,48(%rdi)
+	movdqu	112(%rsi),%xmm7
+	leaq	128(%rsi),%rsi
+	pxor	16(%rsp),%xmm6
+	pxor	%xmm13,%xmm11
+	pxor	%xmm5,%xmm2
+	pxor	%xmm1,%xmm7
+
+	movdqu	%xmm6,64(%rdi)
+	movdqu	0(%rsi),%xmm6
+	movdqu	%xmm11,80(%rdi)
+	movdqu	16(%rsi),%xmm11
+	movdqu	%xmm2,96(%rdi)
+	movdqu	32(%rsi),%xmm2
+	movdqu	%xmm7,112(%rdi)
+	leaq	128(%rdi),%rdi
+	movdqu	48(%rsi),%xmm7
+	pxor	32(%rsp),%xmm6
+	pxor	%xmm10,%xmm11
+	pxor	%xmm14,%xmm2
+	pxor	%xmm8,%xmm7
+
+	movdqu	%xmm6,0(%rdi)
+	movdqu	64(%rsi),%xmm6
+	movdqu	%xmm11,16(%rdi)
+	movdqu	80(%rsi),%xmm11
+	movdqu	%xmm2,32(%rdi)
+	movdqu	96(%rsi),%xmm2
+	movdqu	%xmm7,48(%rdi)
+	movdqu	112(%rsi),%xmm7
+	leaq	128(%rsi),%rsi
+	pxor	48(%rsp),%xmm6
+	pxor	%xmm15,%xmm11
+	pxor	%xmm9,%xmm2
+	pxor	%xmm3,%xmm7
+	movdqu	%xmm6,64(%rdi)
+	movdqu	%xmm11,80(%rdi)
+	movdqu	%xmm2,96(%rdi)
+	movdqu	%xmm7,112(%rdi)
+	leaq	128(%rdi),%rdi
+
+	subq	$256,%rdx
+	jnz	L$oop_outer4x
+
+	jmp	L$done4x
+
+L$tail4x:
+	cmpq	$192,%rdx
+	jae	L$192_or_more4x
+	cmpq	$128,%rdx
+	jae	L$128_or_more4x
+	cmpq	$64,%rdx
+	jae	L$64_or_more4x
+
+
+	xorq	%r10,%r10
+
+	movdqa	%xmm12,16(%rsp)
+	movdqa	%xmm4,32(%rsp)
+	movdqa	%xmm0,48(%rsp)
+	jmp	L$oop_tail4x
+
+.p2align	5
+L$64_or_more4x:
+	movdqu	0(%rsi),%xmm6
+	movdqu	16(%rsi),%xmm11
+	movdqu	32(%rsi),%xmm2
+	movdqu	48(%rsi),%xmm7
+	pxor	0(%rsp),%xmm6
+	pxor	%xmm12,%xmm11
+	pxor	%xmm4,%xmm2
+	pxor	%xmm0,%xmm7
+	movdqu	%xmm6,0(%rdi)
+	movdqu	%xmm11,16(%rdi)
+	movdqu	%xmm2,32(%rdi)
+	movdqu	%xmm7,48(%rdi)
+	je	L$done4x
+
+	movdqa	16(%rsp),%xmm6
+	leaq	64(%rsi),%rsi
+	xorq	%r10,%r10
+	movdqa	%xmm6,0(%rsp)
+	movdqa	%xmm13,16(%rsp)
+	leaq	64(%rdi),%rdi
+	movdqa	%xmm5,32(%rsp)
+	subq	$64,%rdx
+	movdqa	%xmm1,48(%rsp)
+	jmp	L$oop_tail4x
+
+.p2align	5
+L$128_or_more4x:
+	movdqu	0(%rsi),%xmm6
+	movdqu	16(%rsi),%xmm11
+	movdqu	32(%rsi),%xmm2
+	movdqu	48(%rsi),%xmm7
+	pxor	0(%rsp),%xmm6
+	pxor	%xmm12,%xmm11
+	pxor	%xmm4,%xmm2
+	pxor	%xmm0,%xmm7
+
+	movdqu	%xmm6,0(%rdi)
+	movdqu	64(%rsi),%xmm6
+	movdqu	%xmm11,16(%rdi)
+	movdqu	80(%rsi),%xmm11
+	movdqu	%xmm2,32(%rdi)
+	movdqu	96(%rsi),%xmm2
+	movdqu	%xmm7,48(%rdi)
+	movdqu	112(%rsi),%xmm7
+	pxor	16(%rsp),%xmm6
+	pxor	%xmm13,%xmm11
+	pxor	%xmm5,%xmm2
+	pxor	%xmm1,%xmm7
+	movdqu	%xmm6,64(%rdi)
+	movdqu	%xmm11,80(%rdi)
+	movdqu	%xmm2,96(%rdi)
+	movdqu	%xmm7,112(%rdi)
+	je	L$done4x
+
+	movdqa	32(%rsp),%xmm6
+	leaq	128(%rsi),%rsi
+	xorq	%r10,%r10
+	movdqa	%xmm6,0(%rsp)
+	movdqa	%xmm10,16(%rsp)
+	leaq	128(%rdi),%rdi
+	movdqa	%xmm14,32(%rsp)
+	subq	$128,%rdx
+	movdqa	%xmm8,48(%rsp)
+	jmp	L$oop_tail4x
+
+.p2align	5
+L$192_or_more4x:
+	movdqu	0(%rsi),%xmm6
+	movdqu	16(%rsi),%xmm11
+	movdqu	32(%rsi),%xmm2
+	movdqu	48(%rsi),%xmm7
+	pxor	0(%rsp),%xmm6
+	pxor	%xmm12,%xmm11
+	pxor	%xmm4,%xmm2
+	pxor	%xmm0,%xmm7
+
+	movdqu	%xmm6,0(%rdi)
+	movdqu	64(%rsi),%xmm6
+	movdqu	%xmm11,16(%rdi)
+	movdqu	80(%rsi),%xmm11
+	movdqu	%xmm2,32(%rdi)
+	movdqu	96(%rsi),%xmm2
+	movdqu	%xmm7,48(%rdi)
+	movdqu	112(%rsi),%xmm7
+	leaq	128(%rsi),%rsi
+	pxor	16(%rsp),%xmm6
+	pxor	%xmm13,%xmm11
+	pxor	%xmm5,%xmm2
+	pxor	%xmm1,%xmm7
+
+	movdqu	%xmm6,64(%rdi)
+	movdqu	0(%rsi),%xmm6
+	movdqu	%xmm11,80(%rdi)
+	movdqu	16(%rsi),%xmm11
+	movdqu	%xmm2,96(%rdi)
+	movdqu	32(%rsi),%xmm2
+	movdqu	%xmm7,112(%rdi)
+	leaq	128(%rdi),%rdi
+	movdqu	48(%rsi),%xmm7
+	pxor	32(%rsp),%xmm6
+	pxor	%xmm10,%xmm11
+	pxor	%xmm14,%xmm2
+	pxor	%xmm8,%xmm7
+	movdqu	%xmm6,0(%rdi)
+	movdqu	%xmm11,16(%rdi)
+	movdqu	%xmm2,32(%rdi)
+	movdqu	%xmm7,48(%rdi)
+	je	L$done4x
+
+	movdqa	48(%rsp),%xmm6
+	leaq	64(%rsi),%rsi
+	xorq	%r10,%r10
+	movdqa	%xmm6,0(%rsp)
+	movdqa	%xmm15,16(%rsp)
+	leaq	64(%rdi),%rdi
+	movdqa	%xmm9,32(%rsp)
+	subq	$192,%rdx
+	movdqa	%xmm3,48(%rsp)
+
+L$oop_tail4x:
+	movzbl	(%rsi,%r10,1),%eax
+	movzbl	(%rsp,%r10,1),%ecx
+	leaq	1(%r10),%r10
+	xorl	%ecx,%eax
+	movb	%al,-1(%rdi,%r10,1)
+	decq	%rdx
+	jnz	L$oop_tail4x
+
+L$done4x:
+	addq	$0x148+0,%rsp
+	.byte	0xf3,0xc3
+
+
+.p2align	5
+ChaCha20_8x:
+L$ChaCha20_8x:
+	movq	%rsp,%r10
+	subq	$0x280+8,%rsp
+	andq	$-32,%rsp
+	vzeroupper
+	movq	%r10,640(%rsp)
+
+
+
+
+
+
+
+
+
+
+	vbroadcasti128	L$sigma(%rip),%ymm11
+	vbroadcasti128	(%rcx),%ymm3
+	vbroadcasti128	16(%rcx),%ymm15
+	vbroadcasti128	(%r8),%ymm7
+	leaq	256(%rsp),%rcx
+	leaq	512(%rsp),%rax
+	leaq	L$rot16(%rip),%r10
+	leaq	L$rot24(%rip),%r11
+
+	vpshufd	$0x00,%ymm11,%ymm8
+	vpshufd	$0x55,%ymm11,%ymm9
+	vmovdqa	%ymm8,128-256(%rcx)
+	vpshufd	$0xaa,%ymm11,%ymm10
+	vmovdqa	%ymm9,160-256(%rcx)
+	vpshufd	$0xff,%ymm11,%ymm11
+	vmovdqa	%ymm10,192-256(%rcx)
+	vmovdqa	%ymm11,224-256(%rcx)
+
+	vpshufd	$0x00,%ymm3,%ymm0
+	vpshufd	$0x55,%ymm3,%ymm1
+	vmovdqa	%ymm0,256-256(%rcx)
+	vpshufd	$0xaa,%ymm3,%ymm2
+	vmovdqa	%ymm1,288-256(%rcx)
+	vpshufd	$0xff,%ymm3,%ymm3
+	vmovdqa	%ymm2,320-256(%rcx)
+	vmovdqa	%ymm3,352-256(%rcx)
+
+	vpshufd	$0x00,%ymm15,%ymm12
+	vpshufd	$0x55,%ymm15,%ymm13
+	vmovdqa	%ymm12,384-512(%rax)
+	vpshufd	$0xaa,%ymm15,%ymm14
+	vmovdqa	%ymm13,416-512(%rax)
+	vpshufd	$0xff,%ymm15,%ymm15
+	vmovdqa	%ymm14,448-512(%rax)
+	vmovdqa	%ymm15,480-512(%rax)
+
+	vpshufd	$0x00,%ymm7,%ymm4
+	vpshufd	$0x55,%ymm7,%ymm5
+	vpaddd	L$incy(%rip),%ymm4,%ymm4
+	vpshufd	$0xaa,%ymm7,%ymm6
+	vmovdqa	%ymm5,544-512(%rax)
+	vpshufd	$0xff,%ymm7,%ymm7
+	vmovdqa	%ymm6,576-512(%rax)
+	vmovdqa	%ymm7,608-512(%rax)
+
+	jmp	L$oop_enter8x
+
+.p2align	5
+L$oop_outer8x:
+	vmovdqa	128-256(%rcx),%ymm8
+	vmovdqa	160-256(%rcx),%ymm9
+	vmovdqa	192-256(%rcx),%ymm10
+	vmovdqa	224-256(%rcx),%ymm11
+	vmovdqa	256-256(%rcx),%ymm0
+	vmovdqa	288-256(%rcx),%ymm1
+	vmovdqa	320-256(%rcx),%ymm2
+	vmovdqa	352-256(%rcx),%ymm3
+	vmovdqa	384-512(%rax),%ymm12
+	vmovdqa	416-512(%rax),%ymm13
+	vmovdqa	448-512(%rax),%ymm14
+	vmovdqa	480-512(%rax),%ymm15
+	vmovdqa	512-512(%rax),%ymm4
+	vmovdqa	544-512(%rax),%ymm5
+	vmovdqa	576-512(%rax),%ymm6
+	vmovdqa	608-512(%rax),%ymm7
+	vpaddd	L$eight(%rip),%ymm4,%ymm4
+
+L$oop_enter8x:
+	vmovdqa	%ymm14,64(%rsp)
+	vmovdqa	%ymm15,96(%rsp)
+	vbroadcasti128	(%r10),%ymm15
+	vmovdqa	%ymm4,512-512(%rax)
+	movl	$10,%eax
+	jmp	L$oop8x
+
+.p2align	5
+L$oop8x:
+	vpaddd	%ymm0,%ymm8,%ymm8
+	vpxor	%ymm4,%ymm8,%ymm4
+	vpshufb	%ymm15,%ymm4,%ymm4
+	vpaddd	%ymm1,%ymm9,%ymm9
+	vpxor	%ymm5,%ymm9,%ymm5
+	vpshufb	%ymm15,%ymm5,%ymm5
+	vpaddd	%ymm4,%ymm12,%ymm12
+	vpxor	%ymm0,%ymm12,%ymm0
+	vpslld	$12,%ymm0,%ymm14
+	vpsrld	$20,%ymm0,%ymm0
+	vpor	%ymm0,%ymm14,%ymm0
+	vbroadcasti128	(%r11),%ymm14
+	vpaddd	%ymm5,%ymm13,%ymm13
+	vpxor	%ymm1,%ymm13,%ymm1
+	vpslld	$12,%ymm1,%ymm15
+	vpsrld	$20,%ymm1,%ymm1
+	vpor	%ymm1,%ymm15,%ymm1
+	vpaddd	%ymm0,%ymm8,%ymm8
+	vpxor	%ymm4,%ymm8,%ymm4
+	vpshufb	%ymm14,%ymm4,%ymm4
+	vpaddd	%ymm1,%ymm9,%ymm9
+	vpxor	%ymm5,%ymm9,%ymm5
+	vpshufb	%ymm14,%ymm5,%ymm5
+	vpaddd	%ymm4,%ymm12,%ymm12
+	vpxor	%ymm0,%ymm12,%ymm0
+	vpslld	$7,%ymm0,%ymm15
+	vpsrld	$25,%ymm0,%ymm0
+	vpor	%ymm0,%ymm15,%ymm0
+	vbroadcasti128	(%r10),%ymm15
+	vpaddd	%ymm5,%ymm13,%ymm13
+	vpxor	%ymm1,%ymm13,%ymm1
+	vpslld	$7,%ymm1,%ymm14
+	vpsrld	$25,%ymm1,%ymm1
+	vpor	%ymm1,%ymm14,%ymm1
+	vmovdqa	%ymm12,0(%rsp)
+	vmovdqa	%ymm13,32(%rsp)
+	vmovdqa	64(%rsp),%ymm12
+	vmovdqa	96(%rsp),%ymm13
+	vpaddd	%ymm2,%ymm10,%ymm10
+	vpxor	%ymm6,%ymm10,%ymm6
+	vpshufb	%ymm15,%ymm6,%ymm6
+	vpaddd	%ymm3,%ymm11,%ymm11
+	vpxor	%ymm7,%ymm11,%ymm7
+	vpshufb	%ymm15,%ymm7,%ymm7
+	vpaddd	%ymm6,%ymm12,%ymm12
+	vpxor	%ymm2,%ymm12,%ymm2
+	vpslld	$12,%ymm2,%ymm14
+	vpsrld	$20,%ymm2,%ymm2
+	vpor	%ymm2,%ymm14,%ymm2
+	vbroadcasti128	(%r11),%ymm14
+	vpaddd	%ymm7,%ymm13,%ymm13
+	vpxor	%ymm3,%ymm13,%ymm3
+	vpslld	$12,%ymm3,%ymm15
+	vpsrld	$20,%ymm3,%ymm3
+	vpor	%ymm3,%ymm15,%ymm3
+	vpaddd	%ymm2,%ymm10,%ymm10
+	vpxor	%ymm6,%ymm10,%ymm6
+	vpshufb	%ymm14,%ymm6,%ymm6
+	vpaddd	%ymm3,%ymm11,%ymm11
+	vpxor	%ymm7,%ymm11,%ymm7
+	vpshufb	%ymm14,%ymm7,%ymm7
+	vpaddd	%ymm6,%ymm12,%ymm12
+	vpxor	%ymm2,%ymm12,%ymm2
+	vpslld	$7,%ymm2,%ymm15
+	vpsrld	$25,%ymm2,%ymm2
+	vpor	%ymm2,%ymm15,%ymm2
+	vbroadcasti128	(%r10),%ymm15
+	vpaddd	%ymm7,%ymm13,%ymm13
+	vpxor	%ymm3,%ymm13,%ymm3
+	vpslld	$7,%ymm3,%ymm14
+	vpsrld	$25,%ymm3,%ymm3
+	vpor	%ymm3,%ymm14,%ymm3
+	vpaddd	%ymm1,%ymm8,%ymm8
+	vpxor	%ymm7,%ymm8,%ymm7
+	vpshufb	%ymm15,%ymm7,%ymm7
+	vpaddd	%ymm2,%ymm9,%ymm9
+	vpxor	%ymm4,%ymm9,%ymm4
+	vpshufb	%ymm15,%ymm4,%ymm4
+	vpaddd	%ymm7,%ymm12,%ymm12
+	vpxor	%ymm1,%ymm12,%ymm1
+	vpslld	$12,%ymm1,%ymm14
+	vpsrld	$20,%ymm1,%ymm1
+	vpor	%ymm1,%ymm14,%ymm1
+	vbroadcasti128	(%r11),%ymm14
+	vpaddd	%ymm4,%ymm13,%ymm13
+	vpxor	%ymm2,%ymm13,%ymm2
+	vpslld	$12,%ymm2,%ymm15
+	vpsrld	$20,%ymm2,%ymm2
+	vpor	%ymm2,%ymm15,%ymm2
+	vpaddd	%ymm1,%ymm8,%ymm8
+	vpxor	%ymm7,%ymm8,%ymm7
+	vpshufb	%ymm14,%ymm7,%ymm7
+	vpaddd	%ymm2,%ymm9,%ymm9
+	vpxor	%ymm4,%ymm9,%ymm4
+	vpshufb	%ymm14,%ymm4,%ymm4
+	vpaddd	%ymm7,%ymm12,%ymm12
+	vpxor	%ymm1,%ymm12,%ymm1
+	vpslld	$7,%ymm1,%ymm15
+	vpsrld	$25,%ymm1,%ymm1
+	vpor	%ymm1,%ymm15,%ymm1
+	vbroadcasti128	(%r10),%ymm15
+	vpaddd	%ymm4,%ymm13,%ymm13
+	vpxor	%ymm2,%ymm13,%ymm2
+	vpslld	$7,%ymm2,%ymm14
+	vpsrld	$25,%ymm2,%ymm2
+	vpor	%ymm2,%ymm14,%ymm2
+	vmovdqa	%ymm12,64(%rsp)
+	vmovdqa	%ymm13,96(%rsp)
+	vmovdqa	0(%rsp),%ymm12
+	vmovdqa	32(%rsp),%ymm13
+	vpaddd	%ymm3,%ymm10,%ymm10
+	vpxor	%ymm5,%ymm10,%ymm5
+	vpshufb	%ymm15,%ymm5,%ymm5
+	vpaddd	%ymm0,%ymm11,%ymm11
+	vpxor	%ymm6,%ymm11,%ymm6
+	vpshufb	%ymm15,%ymm6,%ymm6
+	vpaddd	%ymm5,%ymm12,%ymm12
+	vpxor	%ymm3,%ymm12,%ymm3
+	vpslld	$12,%ymm3,%ymm14
+	vpsrld	$20,%ymm3,%ymm3
+	vpor	%ymm3,%ymm14,%ymm3
+	vbroadcasti128	(%r11),%ymm14
+	vpaddd	%ymm6,%ymm13,%ymm13
+	vpxor	%ymm0,%ymm13,%ymm0
+	vpslld	$12,%ymm0,%ymm15
+	vpsrld	$20,%ymm0,%ymm0
+	vpor	%ymm0,%ymm15,%ymm0
+	vpaddd	%ymm3,%ymm10,%ymm10
+	vpxor	%ymm5,%ymm10,%ymm5
+	vpshufb	%ymm14,%ymm5,%ymm5
+	vpaddd	%ymm0,%ymm11,%ymm11
+	vpxor	%ymm6,%ymm11,%ymm6
+	vpshufb	%ymm14,%ymm6,%ymm6
+	vpaddd	%ymm5,%ymm12,%ymm12
+	vpxor	%ymm3,%ymm12,%ymm3
+	vpslld	$7,%ymm3,%ymm15
+	vpsrld	$25,%ymm3,%ymm3
+	vpor	%ymm3,%ymm15,%ymm3
+	vbroadcasti128	(%r10),%ymm15
+	vpaddd	%ymm6,%ymm13,%ymm13
+	vpxor	%ymm0,%ymm13,%ymm0
+	vpslld	$7,%ymm0,%ymm14
+	vpsrld	$25,%ymm0,%ymm0
+	vpor	%ymm0,%ymm14,%ymm0
+	decl	%eax
+	jnz	L$oop8x
+
+	leaq	512(%rsp),%rax
+	vpaddd	128-256(%rcx),%ymm8,%ymm8
+	vpaddd	160-256(%rcx),%ymm9,%ymm9
+	vpaddd	192-256(%rcx),%ymm10,%ymm10
+	vpaddd	224-256(%rcx),%ymm11,%ymm11
+
+	vpunpckldq	%ymm9,%ymm8,%ymm14
+	vpunpckldq	%ymm11,%ymm10,%ymm15
+	vpunpckhdq	%ymm9,%ymm8,%ymm8
+	vpunpckhdq	%ymm11,%ymm10,%ymm10
+	vpunpcklqdq	%ymm15,%ymm14,%ymm9
+	vpunpckhqdq	%ymm15,%ymm14,%ymm14
+	vpunpcklqdq	%ymm10,%ymm8,%ymm11
+	vpunpckhqdq	%ymm10,%ymm8,%ymm8
+	vpaddd	256-256(%rcx),%ymm0,%ymm0
+	vpaddd	288-256(%rcx),%ymm1,%ymm1
+	vpaddd	320-256(%rcx),%ymm2,%ymm2
+	vpaddd	352-256(%rcx),%ymm3,%ymm3
+
+	vpunpckldq	%ymm1,%ymm0,%ymm10
+	vpunpckldq	%ymm3,%ymm2,%ymm15
+	vpunpckhdq	%ymm1,%ymm0,%ymm0
+	vpunpckhdq	%ymm3,%ymm2,%ymm2
+	vpunpcklqdq	%ymm15,%ymm10,%ymm1
+	vpunpckhqdq	%ymm15,%ymm10,%ymm10
+	vpunpcklqdq	%ymm2,%ymm0,%ymm3
+	vpunpckhqdq	%ymm2,%ymm0,%ymm0
+	vperm2i128	$0x20,%ymm1,%ymm9,%ymm15
+	vperm2i128	$0x31,%ymm1,%ymm9,%ymm1
+	vperm2i128	$0x20,%ymm10,%ymm14,%ymm9
+	vperm2i128	$0x31,%ymm10,%ymm14,%ymm10
+	vperm2i128	$0x20,%ymm3,%ymm11,%ymm14
+	vperm2i128	$0x31,%ymm3,%ymm11,%ymm3
+	vperm2i128	$0x20,%ymm0,%ymm8,%ymm11
+	vperm2i128	$0x31,%ymm0,%ymm8,%ymm0
+	vmovdqa	%ymm15,0(%rsp)
+	vmovdqa	%ymm9,32(%rsp)
+	vmovdqa	64(%rsp),%ymm15
+	vmovdqa	96(%rsp),%ymm9
+
+	vpaddd	384-512(%rax),%ymm12,%ymm12
+	vpaddd	416-512(%rax),%ymm13,%ymm13
+	vpaddd	448-512(%rax),%ymm15,%ymm15
+	vpaddd	480-512(%rax),%ymm9,%ymm9
+
+	vpunpckldq	%ymm13,%ymm12,%ymm2
+	vpunpckldq	%ymm9,%ymm15,%ymm8
+	vpunpckhdq	%ymm13,%ymm12,%ymm12
+	vpunpckhdq	%ymm9,%ymm15,%ymm15
+	vpunpcklqdq	%ymm8,%ymm2,%ymm13
+	vpunpckhqdq	%ymm8,%ymm2,%ymm2
+	vpunpcklqdq	%ymm15,%ymm12,%ymm9
+	vpunpckhqdq	%ymm15,%ymm12,%ymm12
+	vpaddd	512-512(%rax),%ymm4,%ymm4
+	vpaddd	544-512(%rax),%ymm5,%ymm5
+	vpaddd	576-512(%rax),%ymm6,%ymm6
+	vpaddd	608-512(%rax),%ymm7,%ymm7
+
+	vpunpckldq	%ymm5,%ymm4,%ymm15
+	vpunpckldq	%ymm7,%ymm6,%ymm8
+	vpunpckhdq	%ymm5,%ymm4,%ymm4
+	vpunpckhdq	%ymm7,%ymm6,%ymm6
+	vpunpcklqdq	%ymm8,%ymm15,%ymm5
+	vpunpckhqdq	%ymm8,%ymm15,%ymm15
+	vpunpcklqdq	%ymm6,%ymm4,%ymm7
+	vpunpckhqdq	%ymm6,%ymm4,%ymm4
+	vperm2i128	$0x20,%ymm5,%ymm13,%ymm8
+	vperm2i128	$0x31,%ymm5,%ymm13,%ymm5
+	vperm2i128	$0x20,%ymm15,%ymm2,%ymm13
+	vperm2i128	$0x31,%ymm15,%ymm2,%ymm15
+	vperm2i128	$0x20,%ymm7,%ymm9,%ymm2
+	vperm2i128	$0x31,%ymm7,%ymm9,%ymm7
+	vperm2i128	$0x20,%ymm4,%ymm12,%ymm9
+	vperm2i128	$0x31,%ymm4,%ymm12,%ymm4
+	vmovdqa	0(%rsp),%ymm6
+	vmovdqa	32(%rsp),%ymm12
+
+	cmpq	$512,%rdx
+	jb	L$tail8x
+
+	vpxor	0(%rsi),%ymm6,%ymm6
+	vpxor	32(%rsi),%ymm8,%ymm8
+	vpxor	64(%rsi),%ymm1,%ymm1
+	vpxor	96(%rsi),%ymm5,%ymm5
+	leaq	128(%rsi),%rsi
+	vmovdqu	%ymm6,0(%rdi)
+	vmovdqu	%ymm8,32(%rdi)
+	vmovdqu	%ymm1,64(%rdi)
+	vmovdqu	%ymm5,96(%rdi)
+	leaq	128(%rdi),%rdi
+
+	vpxor	0(%rsi),%ymm12,%ymm12
+	vpxor	32(%rsi),%ymm13,%ymm13
+	vpxor	64(%rsi),%ymm10,%ymm10
+	vpxor	96(%rsi),%ymm15,%ymm15
+	leaq	128(%rsi),%rsi
+	vmovdqu	%ymm12,0(%rdi)
+	vmovdqu	%ymm13,32(%rdi)
+	vmovdqu	%ymm10,64(%rdi)
+	vmovdqu	%ymm15,96(%rdi)
+	leaq	128(%rdi),%rdi
+
+	vpxor	0(%rsi),%ymm14,%ymm14
+	vpxor	32(%rsi),%ymm2,%ymm2
+	vpxor	64(%rsi),%ymm3,%ymm3
+	vpxor	96(%rsi),%ymm7,%ymm7
+	leaq	128(%rsi),%rsi
+	vmovdqu	%ymm14,0(%rdi)
+	vmovdqu	%ymm2,32(%rdi)
+	vmovdqu	%ymm3,64(%rdi)
+	vmovdqu	%ymm7,96(%rdi)
+	leaq	128(%rdi),%rdi
+
+	vpxor	0(%rsi),%ymm11,%ymm11
+	vpxor	32(%rsi),%ymm9,%ymm9
+	vpxor	64(%rsi),%ymm0,%ymm0
+	vpxor	96(%rsi),%ymm4,%ymm4
+	leaq	128(%rsi),%rsi
+	vmovdqu	%ymm11,0(%rdi)
+	vmovdqu	%ymm9,32(%rdi)
+	vmovdqu	%ymm0,64(%rdi)
+	vmovdqu	%ymm4,96(%rdi)
+	leaq	128(%rdi),%rdi
+
+	subq	$512,%rdx
+	jnz	L$oop_outer8x
+
+	jmp	L$done8x
+
+L$tail8x:
+	cmpq	$448,%rdx
+	jae	L$448_or_more8x
+	cmpq	$384,%rdx
+	jae	L$384_or_more8x
+	cmpq	$320,%rdx
+	jae	L$320_or_more8x
+	cmpq	$256,%rdx
+	jae	L$256_or_more8x
+	cmpq	$192,%rdx
+	jae	L$192_or_more8x
+	cmpq	$128,%rdx
+	jae	L$128_or_more8x
+	cmpq	$64,%rdx
+	jae	L$64_or_more8x
+
+	xorq	%r10,%r10
+	vmovdqa	%ymm6,0(%rsp)
+	vmovdqa	%ymm8,32(%rsp)
+	jmp	L$oop_tail8x
+
+.p2align	5
+L$64_or_more8x:
+	vpxor	0(%rsi),%ymm6,%ymm6
+	vpxor	32(%rsi),%ymm8,%ymm8
+	vmovdqu	%ymm6,0(%rdi)
+	vmovdqu	%ymm8,32(%rdi)
+	je	L$done8x
+
+	leaq	64(%rsi),%rsi
+	xorq	%r10,%r10
+	vmovdqa	%ymm1,0(%rsp)
+	leaq	64(%rdi),%rdi
+	subq	$64,%rdx
+	vmovdqa	%ymm5,32(%rsp)
+	jmp	L$oop_tail8x
+
+.p2align	5
+L$128_or_more8x:
+	vpxor	0(%rsi),%ymm6,%ymm6
+	vpxor	32(%rsi),%ymm8,%ymm8
+	vpxor	64(%rsi),%ymm1,%ymm1
+	vpxor	96(%rsi),%ymm5,%ymm5
+	vmovdqu	%ymm6,0(%rdi)
+	vmovdqu	%ymm8,32(%rdi)
+	vmovdqu	%ymm1,64(%rdi)
+	vmovdqu	%ymm5,96(%rdi)
+	je	L$done8x
+
+	leaq	128(%rsi),%rsi
+	xorq	%r10,%r10
+	vmovdqa	%ymm12,0(%rsp)
+	leaq	128(%rdi),%rdi
+	subq	$128,%rdx
+	vmovdqa	%ymm13,32(%rsp)
+	jmp	L$oop_tail8x
+
+.p2align	5
+L$192_or_more8x:
+	vpxor	0(%rsi),%ymm6,%ymm6
+	vpxor	32(%rsi),%ymm8,%ymm8
+	vpxor	64(%rsi),%ymm1,%ymm1
+	vpxor	96(%rsi),%ymm5,%ymm5
+	vpxor	128(%rsi),%ymm12,%ymm12
+	vpxor	160(%rsi),%ymm13,%ymm13
+	vmovdqu	%ymm6,0(%rdi)
+	vmovdqu	%ymm8,32(%rdi)
+	vmovdqu	%ymm1,64(%rdi)
+	vmovdqu	%ymm5,96(%rdi)
+	vmovdqu	%ymm12,128(%rdi)
+	vmovdqu	%ymm13,160(%rdi)
+	je	L$done8x
+
+	leaq	192(%rsi),%rsi
+	xorq	%r10,%r10
+	vmovdqa	%ymm10,0(%rsp)
+	leaq	192(%rdi),%rdi
+	subq	$192,%rdx
+	vmovdqa	%ymm15,32(%rsp)
+	jmp	L$oop_tail8x
+
+.p2align	5
+L$256_or_more8x:
+	vpxor	0(%rsi),%ymm6,%ymm6
+	vpxor	32(%rsi),%ymm8,%ymm8
+	vpxor	64(%rsi),%ymm1,%ymm1
+	vpxor	96(%rsi),%ymm5,%ymm5
+	vpxor	128(%rsi),%ymm12,%ymm12
+	vpxor	160(%rsi),%ymm13,%ymm13
+	vpxor	192(%rsi),%ymm10,%ymm10
+	vpxor	224(%rsi),%ymm15,%ymm15
+	vmovdqu	%ymm6,0(%rdi)
+	vmovdqu	%ymm8,32(%rdi)
+	vmovdqu	%ymm1,64(%rdi)
+	vmovdqu	%ymm5,96(%rdi)
+	vmovdqu	%ymm12,128(%rdi)
+	vmovdqu	%ymm13,160(%rdi)
+	vmovdqu	%ymm10,192(%rdi)
+	vmovdqu	%ymm15,224(%rdi)
+	je	L$done8x
+
+	leaq	256(%rsi),%rsi
+	xorq	%r10,%r10
+	vmovdqa	%ymm14,0(%rsp)
+	leaq	256(%rdi),%rdi
+	subq	$256,%rdx
+	vmovdqa	%ymm2,32(%rsp)
+	jmp	L$oop_tail8x
+
+.p2align	5
+L$320_or_more8x:
+	vpxor	0(%rsi),%ymm6,%ymm6
+	vpxor	32(%rsi),%ymm8,%ymm8
+	vpxor	64(%rsi),%ymm1,%ymm1
+	vpxor	96(%rsi),%ymm5,%ymm5
+	vpxor	128(%rsi),%ymm12,%ymm12
+	vpxor	160(%rsi),%ymm13,%ymm13
+	vpxor	192(%rsi),%ymm10,%ymm10
+	vpxor	224(%rsi),%ymm15,%ymm15
+	vpxor	256(%rsi),%ymm14,%ymm14
+	vpxor	288(%rsi),%ymm2,%ymm2
+	vmovdqu	%ymm6,0(%rdi)
+	vmovdqu	%ymm8,32(%rdi)
+	vmovdqu	%ymm1,64(%rdi)
+	vmovdqu	%ymm5,96(%rdi)
+	vmovdqu	%ymm12,128(%rdi)
+	vmovdqu	%ymm13,160(%rdi)
+	vmovdqu	%ymm10,192(%rdi)
+	vmovdqu	%ymm15,224(%rdi)
+	vmovdqu	%ymm14,256(%rdi)
+	vmovdqu	%ymm2,288(%rdi)
+	je	L$done8x
+
+	leaq	320(%rsi),%rsi
+	xorq	%r10,%r10
+	vmovdqa	%ymm3,0(%rsp)
+	leaq	320(%rdi),%rdi
+	subq	$320,%rdx
+	vmovdqa	%ymm7,32(%rsp)
+	jmp	L$oop_tail8x
+
+.p2align	5
+L$384_or_more8x:
+	vpxor	0(%rsi),%ymm6,%ymm6
+	vpxor	32(%rsi),%ymm8,%ymm8
+	vpxor	64(%rsi),%ymm1,%ymm1
+	vpxor	96(%rsi),%ymm5,%ymm5
+	vpxor	128(%rsi),%ymm12,%ymm12
+	vpxor	160(%rsi),%ymm13,%ymm13
+	vpxor	192(%rsi),%ymm10,%ymm10
+	vpxor	224(%rsi),%ymm15,%ymm15
+	vpxor	256(%rsi),%ymm14,%ymm14
+	vpxor	288(%rsi),%ymm2,%ymm2
+	vpxor	320(%rsi),%ymm3,%ymm3
+	vpxor	352(%rsi),%ymm7,%ymm7
+	vmovdqu	%ymm6,0(%rdi)
+	vmovdqu	%ymm8,32(%rdi)
+	vmovdqu	%ymm1,64(%rdi)
+	vmovdqu	%ymm5,96(%rdi)
+	vmovdqu	%ymm12,128(%rdi)
+	vmovdqu	%ymm13,160(%rdi)
+	vmovdqu	%ymm10,192(%rdi)
+	vmovdqu	%ymm15,224(%rdi)
+	vmovdqu	%ymm14,256(%rdi)
+	vmovdqu	%ymm2,288(%rdi)
+	vmovdqu	%ymm3,320(%rdi)
+	vmovdqu	%ymm7,352(%rdi)
+	je	L$done8x
+
+	leaq	384(%rsi),%rsi
+	xorq	%r10,%r10
+	vmovdqa	%ymm11,0(%rsp)
+	leaq	384(%rdi),%rdi
+	subq	$384,%rdx
+	vmovdqa	%ymm9,32(%rsp)
+	jmp	L$oop_tail8x
+
+.p2align	5
+L$448_or_more8x:
+	vpxor	0(%rsi),%ymm6,%ymm6
+	vpxor	32(%rsi),%ymm8,%ymm8
+	vpxor	64(%rsi),%ymm1,%ymm1
+	vpxor	96(%rsi),%ymm5,%ymm5
+	vpxor	128(%rsi),%ymm12,%ymm12
+	vpxor	160(%rsi),%ymm13,%ymm13
+	vpxor	192(%rsi),%ymm10,%ymm10
+	vpxor	224(%rsi),%ymm15,%ymm15
+	vpxor	256(%rsi),%ymm14,%ymm14
+	vpxor	288(%rsi),%ymm2,%ymm2
+	vpxor	320(%rsi),%ymm3,%ymm3
+	vpxor	352(%rsi),%ymm7,%ymm7
+	vpxor	384(%rsi),%ymm11,%ymm11
+	vpxor	416(%rsi),%ymm9,%ymm9
+	vmovdqu	%ymm6,0(%rdi)
+	vmovdqu	%ymm8,32(%rdi)
+	vmovdqu	%ymm1,64(%rdi)
+	vmovdqu	%ymm5,96(%rdi)
+	vmovdqu	%ymm12,128(%rdi)
+	vmovdqu	%ymm13,160(%rdi)
+	vmovdqu	%ymm10,192(%rdi)
+	vmovdqu	%ymm15,224(%rdi)
+	vmovdqu	%ymm14,256(%rdi)
+	vmovdqu	%ymm2,288(%rdi)
+	vmovdqu	%ymm3,320(%rdi)
+	vmovdqu	%ymm7,352(%rdi)
+	vmovdqu	%ymm11,384(%rdi)
+	vmovdqu	%ymm9,416(%rdi)
+	je	L$done8x
+
+	leaq	448(%rsi),%rsi
+	xorq	%r10,%r10
+	vmovdqa	%ymm0,0(%rsp)
+	leaq	448(%rdi),%rdi
+	subq	$448,%rdx
+	vmovdqa	%ymm4,32(%rsp)
+
+L$oop_tail8x:
+	movzbl	(%rsi,%r10,1),%eax
+	movzbl	(%rsp,%r10,1),%ecx
+	leaq	1(%r10),%r10
+	xorl	%ecx,%eax
+	movb	%al,-1(%rdi,%r10,1)
+	decq	%rdx
+	jnz	L$oop_tail8x
+
+L$done8x:
+	vzeroall
+	movq	640(%rsp),%rsp
+	.byte	0xf3,0xc3
+
+#endif

third_party/boringssl/mac-x86_64/crypto/cpu-x86_64-asm.S

diff --git a/third_party/boringssl/mac-x86_64/crypto/cpu-x86_64-asm.S b/third_party/boringssl/mac-x86_64/crypto/cpu-x86_64-asm.S
deleted file mode 100644
index 0dde04d..0000000
--- a/third_party/boringssl/mac-x86_64/crypto/cpu-x86_64-asm.S
+++ /dev/null
@@ -1,143 +0,0 @@
-#if defined(__x86_64__)
-.text	
-
-.globl	_OPENSSL_ia32_cpuid
-.private_extern _OPENSSL_ia32_cpuid
-
-.p2align	4
-_OPENSSL_ia32_cpuid:
-
-
-	movq	%rdi,%rdi
-	movq	%rbx,%r8
-
-	xorl	%eax,%eax
-	movl	%eax,8(%rdi)
-	cpuid
-	movl	%eax,%r11d
-
-	xorl	%eax,%eax
-	cmpl	$1970169159,%ebx
-	setne	%al
-	movl	%eax,%r9d
-	cmpl	$1231384169,%edx
-	setne	%al
-	orl	%eax,%r9d
-	cmpl	$1818588270,%ecx
-	setne	%al
-	orl	%eax,%r9d
-	jz	L$intel
-
-	cmpl	$1752462657,%ebx
-	setne	%al
-	movl	%eax,%r10d
-	cmpl	$1769238117,%edx
-	setne	%al
-	orl	%eax,%r10d
-	cmpl	$1145913699,%ecx
-	setne	%al
-	orl	%eax,%r10d
-	jnz	L$intel
-
-
-
-
-	movl	$2147483648,%eax
-	cpuid
-
-
-	cmpl	$2147483649,%eax
-	jb	L$intel
-	movl	%eax,%r10d
-	movl	$2147483649,%eax
-	cpuid
-
-
-	orl	%ecx,%r9d
-	andl	$2049,%r9d
-
-	cmpl	$2147483656,%r10d
-	jb	L$intel
-
-	movl	$2147483656,%eax
-	cpuid
-
-	movzbq	%cl,%r10
-	incq	%r10
-
-	movl	$1,%eax
-	cpuid
-
-	btl	$28,%edx
-	jnc	L$generic
-	shrl	$16,%ebx
-	cmpb	%r10b,%bl
-	ja	L$generic
-	andl	$4026531839,%edx
-	jmp	L$generic
-
-L$intel:
-	cmpl	$4,%r11d
-	movl	$-1,%r10d
-	jb	L$nocacheinfo
-
-	movl	$4,%eax
-	movl	$0,%ecx
-	cpuid
-	movl	%eax,%r10d
-	shrl	$14,%r10d
-	andl	$4095,%r10d
-
-	cmpl	$7,%r11d
-	jb	L$nocacheinfo
-
-	movl	$7,%eax
-	xorl	%ecx,%ecx
-	cpuid
-	movl	%ebx,8(%rdi)
-
-L$nocacheinfo:
-	movl	$1,%eax
-	cpuid
-
-	andl	$3220176895,%edx
-	cmpl	$0,%r9d
-	jne	L$notintel
-	orl	$1073741824,%edx
-L$notintel:
-	btl	$28,%edx
-	jnc	L$generic
-	andl	$4026531839,%edx
-	cmpl	$0,%r10d
-	je	L$generic
-
-	orl	$268435456,%edx
-	shrl	$16,%ebx
-	cmpb	$1,%bl
-	ja	L$generic
-	andl	$4026531839,%edx
-L$generic:
-	andl	$2048,%r9d
-	andl	$4294965247,%ecx
-	orl	%ecx,%r9d
-
-	movl	%edx,%r10d
-	btl	$27,%r9d
-	jnc	L$clear_avx
-	xorl	%ecx,%ecx
-.byte	0x0f,0x01,0xd0
-	andl	$6,%eax
-	cmpl	$6,%eax
-	je	L$done
-L$clear_avx:
-	movl	$4026525695,%eax
-	andl	%eax,%r9d
-	andl	$4294967263,8(%rdi)
-L$done:
-	movl	%r9d,4(%rdi)
-	movl	%r10d,0(%rdi)
-	movq	%r8,%rbx
-	.byte	0xf3,0xc3
-
-
-#endif

third_party/boringssl/mac-x86_64/crypto/ec/p256-x86_64-asm.S

diff --git a/third_party/boringssl/mac-x86_64/crypto/ec/p256-x86_64-asm.S b/third_party/boringssl/mac-x86_64/crypto/ec/p256-x86_64-asm.S
new file mode 100644
index 0000000..1cd0cc3
--- /dev/null
+++ b/third_party/boringssl/mac-x86_64/crypto/ec/p256-x86_64-asm.S
@@ -0,0 +1,1788 @@
+#if defined(__x86_64__)
+.text	
+
+
+
+.p2align	6
+L$poly:
+.quad	0xffffffffffffffff, 0x00000000ffffffff, 0x0000000000000000, 0xffffffff00000001
+
+L$One:
+.long	1,1,1,1,1,1,1,1
+L$Two:
+.long	2,2,2,2,2,2,2,2
+L$Three:
+.long	3,3,3,3,3,3,3,3
+L$ONE_mont:
+.quad	0x0000000000000001, 0xffffffff00000000, 0xffffffffffffffff, 0x00000000fffffffe
+
+
+.p2align	6
+ecp_nistz256_mul_by_2:
+	pushq	%r12
+	pushq	%r13
+
+	movq	0(%rsi),%r8
+	movq	8(%rsi),%r9
+	addq	%r8,%r8
+	movq	16(%rsi),%r10
+	adcq	%r9,%r9
+	movq	24(%rsi),%r11
+	leaq	L$poly(%rip),%rsi
+	movq	%r8,%rax
+	adcq	%r10,%r10
+	adcq	%r11,%r11
+	movq	%r9,%rdx
+	sbbq	%r13,%r13
+
+	subq	0(%rsi),%r8
+	movq	%r10,%rcx
+	sbbq	8(%rsi),%r9
+	sbbq	16(%rsi),%r10
+	movq	%r11,%r12
+	sbbq	24(%rsi),%r11
+	testq	%r13,%r13
+
+	cmovzq	%rax,%r8
+	cmovzq	%rdx,%r9
+	movq	%r8,0(%rdi)
+	cmovzq	%rcx,%r10
+	movq	%r9,8(%rdi)
+	cmovzq	%r12,%r11
+	movq	%r10,16(%rdi)
+	movq	%r11,24(%rdi)
+
+	popq	%r13
+	popq	%r12
+	.byte	0xf3,0xc3
+
+
+
+
+.globl	_ecp_nistz256_neg
+.private_extern _ecp_nistz256_neg
+
+.p2align	5
+_ecp_nistz256_neg:
+	pushq	%r12
+	pushq	%r13
+
+	xorq	%r8,%r8
+	xorq	%r9,%r9
+	xorq	%r10,%r10
+	xorq	%r11,%r11
+	xorq	%r13,%r13
+
+	subq	0(%rsi),%r8
+	sbbq	8(%rsi),%r9
+	sbbq	16(%rsi),%r10
+	movq	%r8,%rax
+	sbbq	24(%rsi),%r11
+	leaq	L$poly(%rip),%rsi
+	movq	%r9,%rdx
+	sbbq	$0,%r13
+
+	addq	0(%rsi),%r8
+	movq	%r10,%rcx
+	adcq	8(%rsi),%r9
+	adcq	16(%rsi),%r10
+	movq	%r11,%r12
+	adcq	24(%rsi),%r11
+	testq	%r13,%r13
+
+	cmovzq	%rax,%r8
+	cmovzq	%rdx,%r9
+	movq	%r8,0(%rdi)
+	cmovzq	%rcx,%r10
+	movq	%r9,8(%rdi)
+	cmovzq	%r12,%r11
+	movq	%r10,16(%rdi)
+	movq	%r11,24(%rdi)
+
+	popq	%r13
+	popq	%r12
+	.byte	0xf3,0xc3
+
+
+
+
+
+
+
+.globl	_ecp_nistz256_mul_mont
+.private_extern _ecp_nistz256_mul_mont
+
+.p2align	5
+_ecp_nistz256_mul_mont:
+L$mul_mont:
+	pushq	%rbp
+	pushq	%rbx
+	pushq	%r12
+	pushq	%r13
+	pushq	%r14
+	pushq	%r15
+	movq	%rdx,%rbx
+	movq	0(%rdx),%rax
+	movq	0(%rsi),%r9
+	movq	8(%rsi),%r10
+	movq	16(%rsi),%r11
+	movq	24(%rsi),%r12
+
+	call	__ecp_nistz256_mul_montq
+L$mul_mont_done:
+	popq	%r15
+	popq	%r14
+	popq	%r13
+	popq	%r12
+	popq	%rbx
+	popq	%rbp
+	.byte	0xf3,0xc3
+
+
+
+.p2align	5
+__ecp_nistz256_mul_montq:
+
+
+	movq	%rax,%rbp
+	mulq	%r9
+	movq	L$poly+8(%rip),%r14
+	movq	%rax,%r8
+	movq	%rbp,%rax
+	movq	%rdx,%r9
+
+	mulq	%r10
+	movq	L$poly+24(%rip),%r15
+	addq	%rax,%r9
+	movq	%rbp,%rax
+	adcq	$0,%rdx
+	movq	%rdx,%r10
+
+	mulq	%r11
+	addq	%rax,%r10
+	movq	%rbp,%rax
+	adcq	$0,%rdx
+	movq	%rdx,%r11
+
+	mulq	%r12
+	addq	%rax,%r11
+	movq	%r8,%rax
+	adcq	$0,%rdx
+	xorq	%r13,%r13
+	movq	%rdx,%r12
+
+
+
+
+
+
+
+
+
+
+	movq	%r8,%rbp
+	shlq	$32,%r8
+	mulq	%r15
+	shrq	$32,%rbp
+	addq	%r8,%r9
+	adcq	%rbp,%r10
+	adcq	%rax,%r11
+	movq	8(%rbx),%rax
+	adcq	%rdx,%r12
+	adcq	$0,%r13
+	xorq	%r8,%r8
+
+
+
+	movq	%rax,%rbp
+	mulq	0(%rsi)
+	addq	%rax,%r9
+	movq	%rbp,%rax
+	adcq	$0,%rdx
+	movq	%rdx,%rcx
+
+	mulq	8(%rsi)
+	addq	%rcx,%r10
+	adcq	$0,%rdx
+	addq	%rax,%r10
+	movq	%rbp,%rax
+	adcq	$0,%rdx
+	movq	%rdx,%rcx
+
+	mulq	16(%rsi)
+	addq	%rcx,%r11
+	adcq	$0,%rdx
+	addq	%rax,%r11
+	movq	%rbp,%rax
+	adcq	$0,%rdx
+	movq	%rdx,%rcx
+
+	mulq	24(%rsi)
+	addq	%rcx,%r12
+	adcq	$0,%rdx
+	addq	%rax,%r12
+	movq	%r9,%rax
+	adcq	%rdx,%r13
+	adcq	$0,%r8
+
+
+
+	movq	%r9,%rbp
+	shlq	$32,%r9
+	mulq	%r15
+	shrq	$32,%rbp
+	addq	%r9,%r10
+	adcq	%rbp,%r11
+	adcq	%rax,%r12
+	movq	16(%rbx),%rax
+	adcq	%rdx,%r13
+	adcq	$0,%r8
+	xorq	%r9,%r9
+
+
+
+	movq	%rax,%rbp
+	mulq	0(%rsi)
+	addq	%rax,%r10
+	movq	%rbp,%rax
+	adcq	$0,%rdx
+	movq	%rdx,%rcx
+
+	mulq	8(%rsi)
+	addq	%rcx,%r11
+	adcq	$0,%rdx
+	addq	%rax,%r11
+	movq	%rbp,%rax
+	adcq	$0,%rdx
+	movq	%rdx,%rcx
+
+	mulq	16(%rsi)
+	addq	%rcx,%r12
+	adcq	$0,%rdx
+	addq	%rax,%r12
+	movq	%rbp,%rax
+	adcq	$0,%rdx
+	movq	%rdx,%rcx
+
+	mulq	24(%rsi)
+	addq	%rcx,%r13
+	adcq	$0,%rdx
+	addq	%rax,%r13
+	movq	%r10,%rax
+	adcq	%rdx,%r8
+	adcq	$0,%r9
+
+
+
+	movq	%r10,%rbp
+	shlq	$32,%r10
+	mulq	%r15
+	shrq	$32,%rbp
+	addq	%r10,%r11
+	adcq	%rbp,%r12
+	adcq	%rax,%r13
+	movq	24(%rbx),%rax
+	adcq	%rdx,%r8
+	adcq	$0,%r9
+	xorq	%r10,%r10
+
+
+
+	movq	%rax,%rbp
+	mulq	0(%rsi)
+	addq	%rax,%r11
+	movq	%rbp,%rax
+	adcq	$0,%rdx
+	movq	%rdx,%rcx
+
+	mulq	8(%rsi)
+	addq	%rcx,%r12
+	adcq	$0,%rdx
+	addq	%rax,%r12
+	movq	%rbp,%rax
+	adcq	$0,%rdx
+	movq	%rdx,%rcx
+
+	mulq	16(%rsi)
+	addq	%rcx,%r13
+	adcq	$0,%rdx
+	addq	%rax,%r13
+	movq	%rbp,%rax
+	adcq	$0,%rdx
+	movq	%rdx,%rcx
+
+	mulq	24(%rsi)
+	addq	%rcx,%r8
+	adcq	$0,%rdx
+	addq	%rax,%r8
+	movq	%r11,%rax
+	adcq	%rdx,%r9
+	adcq	$0,%r10
+
+
+
+	movq	%r11,%rbp
+	shlq	$32,%r11
+	mulq	%r15
+	shrq	$32,%rbp
+	addq	%r11,%r12
+	adcq	%rbp,%r13
+	movq	%r12,%rcx
+	adcq	%rax,%r8
+	adcq	%rdx,%r9
+	movq	%r13,%rbp
+	adcq	$0,%r10
+
+
+
+	subq	$-1,%r12
+	movq	%r8,%rbx
+	sbbq	%r14,%r13
+	sbbq	$0,%r8
+	movq	%r9,%rdx
+	sbbq	%r15,%r9
+	sbbq	$0,%r10
+
+	cmovcq	%rcx,%r12
+	cmovcq	%rbp,%r13
+	movq	%r12,0(%rdi)
+	cmovcq	%rbx,%r8
+	movq	%r13,8(%rdi)
+	cmovcq	%rdx,%r9
+	movq	%r8,16(%rdi)
+	movq	%r9,24(%rdi)
+
+	.byte	0xf3,0xc3
+
+
+
+
+
+
+
+
+
+.globl	_ecp_nistz256_sqr_mont
+.private_extern _ecp_nistz256_sqr_mont
+
+.p2align	5
+_ecp_nistz256_sqr_mont:
+	pushq	%rbp
+	pushq	%rbx
+	pushq	%r12
+	pushq	%r13
+	pushq	%r14
+	pushq	%r15
+	movq	0(%rsi),%rax
+	movq	8(%rsi),%r14
+	movq	16(%rsi),%r15
+	movq	24(%rsi),%r8
+
+	call	__ecp_nistz256_sqr_montq
+L$sqr_mont_done:
+	popq	%r15
+	popq	%r14
+	popq	%r13
+	popq	%r12
+	popq	%rbx
+	popq	%rbp
+	.byte	0xf3,0xc3
+
+
+
+.p2align	5
+__ecp_nistz256_sqr_montq:
+	movq	%rax,%r13
+	mulq	%r14
+	movq	%rax,%r9
+	movq	%r15,%rax
+	movq	%rdx,%r10
+
+	mulq	%r13
+	addq	%rax,%r10
+	movq	%r8,%rax
+	adcq	$0,%rdx
+	movq	%rdx,%r11
+
+	mulq	%r13
+	addq	%rax,%r11
+	movq	%r15,%rax
+	adcq	$0,%rdx
+	movq	%rdx,%r12
+
+
+	mulq	%r14
+	addq	%rax,%r11
+	movq	%r8,%rax
+	adcq	$0,%rdx
+	movq	%rdx,%rbp
+
+	mulq	%r14
+	addq	%rax,%r12
+	movq	%r8,%rax
+	adcq	$0,%rdx
+	addq	%rbp,%r12
+	movq	%rdx,%r13
+	adcq	$0,%r13
+
+
+	mulq	%r15
+	xorq	%r15,%r15
+	addq	%rax,%r13
+	movq	0(%rsi),%rax
+	movq	%rdx,%r14
+	adcq	$0,%r14
+
+	addq	%r9,%r9
+	adcq	%r10,%r10
+	adcq	%r11,%r11
+	adcq	%r12,%r12
+	adcq	%r13,%r13
+	adcq	%r14,%r14
+	adcq	$0,%r15
+
+	mulq	%rax
+	movq	%rax,%r8
+	movq	8(%rsi),%rax
+	movq	%rdx,%rcx
+
+	mulq	%rax
+	addq	%rcx,%r9
+	adcq	%rax,%r10
+	movq	16(%rsi),%rax
+	adcq	$0,%rdx
+	movq	%rdx,%rcx
+
+	mulq	%rax
+	addq	%rcx,%r11
+	adcq	%rax,%r12
+	movq	24(%rsi),%rax
+	adcq	$0,%rdx
+	movq	%rdx,%rcx
+
+	mulq	%rax
+	addq	%rcx,%r13
+	adcq	%rax,%r14
+	movq	%r8,%rax
+	adcq	%rdx,%r15
+
+	movq	L$poly+8(%rip),%rsi
+	movq	L$poly+24(%rip),%rbp
+
+
+
+
+	movq	%r8,%rcx
+	shlq	$32,%r8
+	mulq	%rbp
+	shrq	$32,%rcx
+	addq	%r8,%r9
+	adcq	%rcx,%r10
+	adcq	%rax,%r11
+	movq	%r9,%rax
+	adcq	$0,%rdx
+
+
+
+	movq	%r9,%rcx
+	shlq	$32,%r9
+	movq	%rdx,%r8
+	mulq	%rbp
+	shrq	$32,%rcx
+	addq	%r9,%r10
+	adcq	%rcx,%r11
+	adcq	%rax,%r8
+	movq	%r10,%rax
+	adcq	$0,%rdx
+
+
+
+	movq	%r10,%rcx
+	shlq	$32,%r10
+	movq	%rdx,%r9
+	mulq	%rbp
+	shrq	$32,%rcx
+	addq	%r10,%r11
+	adcq	%rcx,%r8
+	adcq	%rax,%r9
+	movq	%r11,%rax
+	adcq	$0,%rdx
+
+
+
+	movq	%r11,%rcx
+	shlq	$32,%r11
+	movq	%rdx,%r10
+	mulq	%rbp
+	shrq	$32,%rcx
+	addq	%r11,%r8
+	adcq	%rcx,%r9
+	adcq	%rax,%r10
+	adcq	$0,%rdx
+	xorq	%r11,%r11
+
+
+
+	addq	%r8,%r12
+	adcq	%r9,%r13
+	movq	%r12,%r8
+	adcq	%r10,%r14
+	adcq	%rdx,%r15
+	movq	%r13,%r9
+	adcq	$0,%r11
+
+	subq	$-1,%r12
+	movq	%r14,%r10
+	sbbq	%rsi,%r13
+	sbbq	$0,%r14
+	movq	%r15,%rcx
+	sbbq	%rbp,%r15
+	sbbq	$0,%r11
+
+	cmovcq	%r8,%r12
+	cmovcq	%r9,%r13
+	movq	%r12,0(%rdi)
+	cmovcq	%r10,%r14
+	movq	%r13,8(%rdi)
+	cmovcq	%rcx,%r15
+	movq	%r14,16(%rdi)
+	movq	%r15,24(%rdi)
+
+	.byte	0xf3,0xc3
+
+
+
+
+
+
+
+.globl	_ecp_nistz256_from_mont
+.private_extern _ecp_nistz256_from_mont
+
+.p2align	5
+_ecp_nistz256_from_mont:
+	pushq	%r12
+	pushq	%r13
+
+	movq	0(%rsi),%rax
+	movq	L$poly+24(%rip),%r13
+	movq	8(%rsi),%r9
+	movq	16(%rsi),%r10
+	movq	24(%rsi),%r11
+	movq	%rax,%r8
+	movq	L$poly+8(%rip),%r12
+
+
+
+	movq	%rax,%rcx
+	shlq	$32,%r8
+	mulq	%r13
+	shrq	$32,%rcx
+	addq	%r8,%r9
+	adcq	%rcx,%r10
+	adcq	%rax,%r11
+	movq	%r9,%rax
+	adcq	$0,%rdx
+
+
+
+	movq	%r9,%rcx
+	shlq	$32,%r9
+	movq	%rdx,%r8
+	mulq	%r13
+	shrq	$32,%rcx
+	addq	%r9,%r10
+	adcq	%rcx,%r11
+	adcq	%rax,%r8
+	movq	%r10,%rax
+	adcq	$0,%rdx
+
+
+
+	movq	%r10,%rcx
+	shlq	$32,%r10
+	movq	%rdx,%r9
+	mulq	%r13
+	shrq	$32,%rcx
+	addq	%r10,%r11
+	adcq	%rcx,%r8
+	adcq	%rax,%r9
+	movq	%r11,%rax
+	adcq	$0,%rdx
+
+
+
+	movq	%r11,%rcx
+	shlq	$32,%r11
+	movq	%rdx,%r10
+	mulq	%r13
+	shrq	$32,%rcx
+	addq	%r11,%r8
+	adcq	%rcx,%r9
+	movq	%r8,%rcx
+	adcq	%rax,%r10
+	movq	%r9,%rsi
+	adcq	$0,%rdx
+
+	subq	$-1,%r8
+	movq	%r10,%rax
+	sbbq	%r12,%r9
+	sbbq	$0,%r10
+	movq	%rdx,%r11
+	sbbq	%r13,%rdx
+	sbbq	%r13,%r13
+
+	cmovnzq	%rcx,%r8
+	cmovnzq	%rsi,%r9
+	movq	%r8,0(%rdi)
+	cmovnzq	%rax,%r10
+	movq	%r9,8(%rdi)
+	cmovzq	%rdx,%r11
+	movq	%r10,16(%rdi)
+	movq	%r11,24(%rdi)
+
+	popq	%r13
+	popq	%r12
+	.byte	0xf3,0xc3
+
+
+
+.globl	_ecp_nistz256_select_w5
+.private_extern _ecp_nistz256_select_w5
+
+.p2align	5
+_ecp_nistz256_select_w5:
+	movdqa	L$One(%rip),%xmm0
+	movd	%edx,%xmm1
+
+	pxor	%xmm2,%xmm2
+	pxor	%xmm3,%xmm3
+	pxor	%xmm4,%xmm4
+	pxor	%xmm5,%xmm5
+	pxor	%xmm6,%xmm6
+	pxor	%xmm7,%xmm7
+
+	movdqa	%xmm0,%xmm8
+	pshufd	$0,%xmm1,%xmm1
+
+	movq	$16,%rax
+L$select_loop_sse_w5:
+
+	movdqa	%xmm8,%xmm15
+	paddd	%xmm0,%xmm8
+	pcmpeqd	%xmm1,%xmm15
+
+	movdqa	0(%rsi),%xmm9
+	movdqa	16(%rsi),%xmm10
+	movdqa	32(%rsi),%xmm11
+	movdqa	48(%rsi),%xmm12
+	movdqa	64(%rsi),%xmm13
+	movdqa	80(%rsi),%xmm14
+	leaq	96(%rsi),%rsi
+
+	pand	%xmm15,%xmm9
+	pand	%xmm15,%xmm10
+	por	%xmm9,%xmm2
+	pand	%xmm15,%xmm11
+	por	%xmm10,%xmm3
+	pand	%xmm15,%xmm12
+	por	%xmm11,%xmm4
+	pand	%xmm15,%xmm13
+	por	%xmm12,%xmm5
+	pand	%xmm15,%xmm14
+	por	%xmm13,%xmm6
+	por	%xmm14,%xmm7
+
+	decq	%rax
+	jnz	L$select_loop_sse_w5
+
+	movdqu	%xmm2,0(%rdi)
+	movdqu	%xmm3,16(%rdi)
+	movdqu	%xmm4,32(%rdi)
+	movdqu	%xmm5,48(%rdi)
+	movdqu	%xmm6,64(%rdi)
+	movdqu	%xmm7,80(%rdi)
+	.byte	0xf3,0xc3
+
+
+
+
+.globl	_ecp_nistz256_select_w7
+.private_extern _ecp_nistz256_select_w7
+
+.p2align	5
+_ecp_nistz256_select_w7:
+	movdqa	L$One(%rip),%xmm8
+	movd	%edx,%xmm1
+
+	pxor	%xmm2,%xmm2
+	pxor	%xmm3,%xmm3
+	pxor	%xmm4,%xmm4
+	pxor	%xmm5,%xmm5
+
+	movdqa	%xmm8,%xmm0
+	pshufd	$0,%xmm1,%xmm1
+	movq	$64,%rax
+
+L$select_loop_sse_w7:
+	movdqa	%xmm8,%xmm15
+	paddd	%xmm0,%xmm8
+	movdqa	0(%rsi),%xmm9
+	movdqa	16(%rsi),%xmm10
+	pcmpeqd	%xmm1,%xmm15
+	movdqa	32(%rsi),%xmm11
+	movdqa	48(%rsi),%xmm12
+	leaq	64(%rsi),%rsi
+
+	pand	%xmm15,%xmm9
+	pand	%xmm15,%xmm10
+	por	%xmm9,%xmm2
+	pand	%xmm15,%xmm11
+	por	%xmm10,%xmm3
+	pand	%xmm15,%xmm12
+	por	%xmm11,%xmm4
+	prefetcht0	255(%rsi)
+	por	%xmm12,%xmm5
+
+	decq	%rax
+	jnz	L$select_loop_sse_w7
+
+	movdqu	%xmm2,0(%rdi)
+	movdqu	%xmm3,16(%rdi)
+	movdqu	%xmm4,32(%rdi)
+	movdqu	%xmm5,48(%rdi)
+	.byte	0xf3,0xc3
+
+.globl	_ecp_nistz256_avx2_select_w7
+.private_extern _ecp_nistz256_avx2_select_w7
+
+.p2align	5
+_ecp_nistz256_avx2_select_w7:
+.byte	0x0f,0x0b
+	.byte	0xf3,0xc3
+
+
+.p2align	5
+__ecp_nistz256_add_toq:
+	addq	0(%rbx),%r12
+	adcq	8(%rbx),%r13
+	movq	%r12,%rax
+	adcq	16(%rbx),%r8
+	adcq	24(%rbx),%r9
+	movq	%r13,%rbp
+	sbbq	%r11,%r11
+
+	subq	$-1,%r12
+	movq	%r8,%rcx
+	sbbq	%r14,%r13
+	sbbq	$0,%r8
+	movq	%r9,%r10
+	sbbq	%r15,%r9
+	testq	%r11,%r11
+
+	cmovzq	%rax,%r12
+	cmovzq	%rbp,%r13
+	movq	%r12,0(%rdi)
+	cmovzq	%rcx,%r8
+	movq	%r13,8(%rdi)
+	cmovzq	%r10,%r9
+	movq	%r8,16(%rdi)
+	movq	%r9,24(%rdi)
+
+	.byte	0xf3,0xc3
+
+
+
+.p2align	5
+__ecp_nistz256_sub_fromq:
+	subq	0(%rbx),%r12
+	sbbq	8(%rbx),%r13
+	movq	%r12,%rax
+	sbbq	16(%rbx),%r8
+	sbbq	24(%rbx),%r9
+	movq	%r13,%rbp
+	sbbq	%r11,%r11
+
+	addq	$-1,%r12
+	movq	%r8,%rcx
+	adcq	%r14,%r13
+	adcq	$0,%r8
+	movq	%r9,%r10
+	adcq	%r15,%r9
+	testq	%r11,%r11
+
+	cmovzq	%rax,%r12
+	cmovzq	%rbp,%r13
+	movq	%r12,0(%rdi)
+	cmovzq	%rcx,%r8
+	movq	%r13,8(%rdi)
+	cmovzq	%r10,%r9
+	movq	%r8,16(%rdi)
+	movq	%r9,24(%rdi)
+
+	.byte	0xf3,0xc3
+
+
+
+.p2align	5
+__ecp_nistz256_subq:
+	subq	%r12,%rax
+	sbbq	%r13,%rbp
+	movq	%rax,%r12
+	sbbq	%r8,%rcx
+	sbbq	%r9,%r10
+	movq	%rbp,%r13
+	sbbq	%r11,%r11
+
+	addq	$-1,%rax
+	movq	%rcx,%r8
+	adcq	%r14,%rbp
+	adcq	$0,%rcx
+	movq	%r10,%r9
+	adcq	%r15,%r10
+	testq	%r11,%r11
+
+	cmovnzq	%rax,%r12
+	cmovnzq	%rbp,%r13
+	cmovnzq	%rcx,%r8
+	cmovnzq	%r10,%r9
+
+	.byte	0xf3,0xc3
+
+
+
+.p2align	5
+__ecp_nistz256_mul_by_2q:
+	addq	%r12,%r12
+	adcq	%r13,%r13
+	movq	%r12,%rax
+	adcq	%r8,%r8
+	adcq	%r9,%r9
+	movq	%r13,%rbp
+	sbbq	%r11,%r11
+
+	subq	$-1,%r12
+	movq	%r8,%rcx
+	sbbq	%r14,%r13
+	sbbq	$0,%r8
+	movq	%r9,%r10
+	sbbq	%r15,%r9
+	testq	%r11,%r11
+
+	cmovzq	%rax,%r12
+	cmovzq	%rbp,%r13
+	movq	%r12,0(%rdi)
+	cmovzq	%rcx,%r8
+	movq	%r13,8(%rdi)
+	cmovzq	%r10,%r9
+	movq	%r8,16(%rdi)
+	movq	%r9,24(%rdi)
+
+	.byte	0xf3,0xc3
+
+.globl	_ecp_nistz256_point_double
+.private_extern _ecp_nistz256_point_double
+
+.p2align	5
+_ecp_nistz256_point_double:
+	pushq	%rbp
+	pushq	%rbx
+	pushq	%r12
+	pushq	%r13
+	pushq	%r14
+	pushq	%r15
+	subq	$160+8,%rsp
+
+L$point_double_shortcutq:
+	movdqu	0(%rsi),%xmm0
+	movq	%rsi,%rbx
+	movdqu	16(%rsi),%xmm1
+	movq	32+0(%rsi),%r12
+	movq	32+8(%rsi),%r13
+	movq	32+16(%rsi),%r8
+	movq	32+24(%rsi),%r9
+	movq	L$poly+8(%rip),%r14
+	movq	L$poly+24(%rip),%r15
+	movdqa	%xmm0,96(%rsp)
+	movdqa	%xmm1,96+16(%rsp)
+	leaq	32(%rdi),%r10
+	leaq	64(%rdi),%r11
+.byte	102,72,15,110,199
+.byte	102,73,15,110,202
+.byte	102,73,15,110,211
+
+	leaq	0(%rsp),%rdi
+	call	__ecp_nistz256_mul_by_2q
+
+	movq	64+0(%rsi),%rax
+	movq	64+8(%rsi),%r14
+	movq	64+16(%rsi),%r15
+	movq	64+24(%rsi),%r8
+	leaq	64-0(%rsi),%rsi
+	leaq	64(%rsp),%rdi
+	call	__ecp_nistz256_sqr_montq
+
+	movq	0+0(%rsp),%rax
+	movq	8+0(%rsp),%r14
+	leaq	0+0(%rsp),%rsi
+	movq	16+0(%rsp),%r15
+	movq	24+0(%rsp),%r8
+	leaq	0(%rsp),%rdi
+	call	__ecp_nistz256_sqr_montq
+
+	movq	32(%rbx),%rax
+	movq	64+0(%rbx),%r9
+	movq	64+8(%rbx),%r10
+	movq	64+16(%rbx),%r11
+	movq	64+24(%rbx),%r12
+	leaq	64-0(%rbx),%rsi
+	leaq	32(%rbx),%rbx
+.byte	102,72,15,126,215
+	call	__ecp_nistz256_mul_montq
+	call	__ecp_nistz256_mul_by_2q
+
+	movq	96+0(%rsp),%r12
+	movq	96+8(%rsp),%r13
+	leaq	64(%rsp),%rbx
+	movq	96+16(%rsp),%r8
+	movq	96+24(%rsp),%r9
+	leaq	32(%rsp),%rdi
+	call	__ecp_nistz256_add_toq
+
+	movq	96+0(%rsp),%r12
+	movq	96+8(%rsp),%r13
+	leaq	64(%rsp),%rbx
+	movq	96+16(%rsp),%r8
+	movq	96+24(%rsp),%r9
+	leaq	64(%rsp),%rdi
+	call	__ecp_nistz256_sub_fromq
+
+	movq	0+0(%rsp),%rax
+	movq	8+0(%rsp),%r14
+	leaq	0+0(%rsp),%rsi
+	movq	16+0(%rsp),%r15
+	movq	24+0(%rsp),%r8
+.byte	102,72,15,126,207
+	call	__ecp_nistz256_sqr_montq
+	xorq	%r9,%r9
+	movq	%r12,%rax
+	addq	$-1,%r12
+	movq	%r13,%r10
+	adcq	%rsi,%r13
+	movq	%r14,%rcx
+	adcq	$0,%r14
+	movq	%r15,%r8
+	adcq	%rbp,%r15
+	adcq	$0,%r9
+	xorq	%rsi,%rsi
+	testq	$1,%rax
+
+	cmovzq	%rax,%r12
+	cmovzq	%r10,%r13
+	cmovzq	%rcx,%r14
+	cmovzq	%r8,%r15
+	cmovzq	%rsi,%r9
+
+	movq	%r13,%rax
+	shrq	$1,%r12
+	shlq	$63,%rax
+	movq	%r14,%r10
+	shrq	$1,%r13
+	orq	%rax,%r12
+	shlq	$63,%r10
+	movq	%r15,%rcx
+	shrq	$1,%r14
+	orq	%r10,%r13
+	shlq	$63,%rcx
+	movq	%r12,0(%rdi)
+	shrq	$1,%r15
+	movq	%r13,8(%rdi)
+	shlq	$63,%r9
+	orq	%rcx,%r14
+	orq	%r9,%r15
+	movq	%r14,16(%rdi)
+	movq	%r15,24(%rdi)
+	movq	64(%rsp),%rax
+	leaq	64(%rsp),%rbx
+	movq	0+32(%rsp),%r9
+	movq	8+32(%rsp),%r10
+	leaq	0+32(%rsp),%rsi
+	movq	16+32(%rsp),%r11
+	movq	24+32(%rsp),%r12
+	leaq	32(%rsp),%rdi
+	call	__ecp_nistz256_mul_montq
+
+	leaq	128(%rsp),%rdi
+	call	__ecp_nistz256_mul_by_2q
+
+	leaq	32(%rsp),%rbx
+	leaq	32(%rsp),%rdi
+	call	__ecp_nistz256_add_toq
+
+	movq	96(%rsp),%rax
+	leaq	96(%rsp),%rbx
+	movq	0+0(%rsp),%r9
+	movq	8+0(%rsp),%r10
+	leaq	0+0(%rsp),%rsi
+	movq	16+0(%rsp),%r11
+	movq	24+0(%rsp),%r12
+	leaq	0(%rsp),%rdi
+	call	__ecp_nistz256_mul_montq
+
+	leaq	128(%rsp),%rdi
+	call	__ecp_nistz256_mul_by_2q
+
+	movq	0+32(%rsp),%rax
+	movq	8+32(%rsp),%r14
+	leaq	0+32(%rsp),%rsi
+	movq	16+32(%rsp),%r15
+	movq	24+32(%rsp),%r8
+.byte	102,72,15,126,199
+	call	__ecp_nistz256_sqr_montq
+
+	leaq	128(%rsp),%rbx
+	movq	%r14,%r8
+	movq	%r15,%r9
+	movq	%rsi,%r14
+	movq	%rbp,%r15
+	call	__ecp_nistz256_sub_fromq
+
+	movq	0+0(%rsp),%rax
+	movq	0+8(%rsp),%rbp
+	movq	0+16(%rsp),%rcx
+	movq	0+24(%rsp),%r10
+	leaq	0(%rsp),%rdi
+	call	__ecp_nistz256_subq
+
+	movq	32(%rsp),%rax
+	leaq	32(%rsp),%rbx
+	movq	%r12,%r14
+	xorl	%ecx,%ecx
+	movq	%r12,0+0(%rsp)
+	movq	%r13,%r10
+	movq	%r13,0+8(%rsp)
+	cmovzq	%r8,%r11
+	movq	%r8,0+16(%rsp)
+	leaq	0-0(%rsp),%rsi
+	cmovzq	%r9,%r12
+	movq	%r9,0+24(%rsp)
+	movq	%r14,%r9
+	leaq	0(%rsp),%rdi
+	call	__ecp_nistz256_mul_montq
+
+.byte	102,72,15,126,203
+.byte	102,72,15,126,207
+	call	__ecp_nistz256_sub_fromq
+
+	addq	$160+8,%rsp
+	popq	%r15
+	popq	%r14
+	popq	%r13
+	popq	%r12
+	popq	%rbx
+	popq	%rbp
+	.byte	0xf3,0xc3
+
+.globl	_ecp_nistz256_point_add
+.private_extern _ecp_nistz256_point_add
+
+.p2align	5
+_ecp_nistz256_point_add:
+	pushq	%rbp
+	pushq	%rbx
+	pushq	%r12
+	pushq	%r13
+	pushq	%r14
+	pushq	%r15
+	subq	$576+8,%rsp
+
+	movdqu	0(%rsi),%xmm0
+	movdqu	16(%rsi),%xmm1
+	movdqu	32(%rsi),%xmm2
+	movdqu	48(%rsi),%xmm3
+	movdqu	64(%rsi),%xmm4
+	movdqu	80(%rsi),%xmm5
+	movq	%rsi,%rbx
+	movq	%rdx,%rsi
+	movdqa	%xmm0,384(%rsp)
+	movdqa	%xmm1,384+16(%rsp)
+	por	%xmm0,%xmm1
+	movdqa	%xmm2,416(%rsp)
+	movdqa	%xmm3,416+16(%rsp)
+	por	%xmm2,%xmm3
+	movdqa	%xmm4,448(%rsp)
+	movdqa	%xmm5,448+16(%rsp)
+	por	%xmm1,%xmm3
+
+	movdqu	0(%rsi),%xmm0
+	pshufd	$0xb1,%xmm3,%xmm5
+	movdqu	16(%rsi),%xmm1
+	movdqu	32(%rsi),%xmm2
+	por	%xmm3,%xmm5
+	movdqu	48(%rsi),%xmm3
+	movq	64+0(%rsi),%rax
+	movq	64+8(%rsi),%r14
+	movq	64+16(%rsi),%r15
+	movq	64+24(%rsi),%r8
+	movdqa	%xmm0,480(%rsp)
+	pshufd	$0x1e,%xmm5,%xmm4
+	movdqa	%xmm1,480+16(%rsp)
+	por	%xmm0,%xmm1
+.byte	102,72,15,110,199
+	movdqa	%xmm2,512(%rsp)
+	movdqa	%xmm3,512+16(%rsp)
+	por	%xmm2,%xmm3
+	por	%xmm4,%xmm5
+	pxor	%xmm4,%xmm4
+	por	%xmm1,%xmm3
+
+	leaq	64-0(%rsi),%rsi
+	movq	%rax,544+0(%rsp)
+	movq	%r14,544+8(%rsp)
+	movq	%r15,544+16(%rsp)
+	movq	%r8,544+24(%rsp)
+	leaq	96(%rsp),%rdi
+	call	__ecp_nistz256_sqr_montq
+
+	pcmpeqd	%xmm4,%xmm5
+	pshufd	$0xb1,%xmm3,%xmm4
+	por	%xmm3,%xmm4
+	pshufd	$0,%xmm5,%xmm5
+	pshufd	$0x1e,%xmm4,%xmm3
+	por	%xmm3,%xmm4
+	pxor	%xmm3,%xmm3
+	pcmpeqd	%xmm3,%xmm4
+	pshufd	$0,%xmm4,%xmm4
+	movq	64+0(%rbx),%rax
+	movq	64+8(%rbx),%r14
+	movq	64+16(%rbx),%r15
+	movq	64+24(%rbx),%r8
+.byte	102,72,15,110,203
+
+	leaq	64-0(%rbx),%rsi
+	leaq	32(%rsp),%rdi
+	call	__ecp_nistz256_sqr_montq
+
+	movq	544(%rsp),%rax
+	leaq	544(%rsp),%rbx
+	movq	0+96(%rsp),%r9
+	movq	8+96(%rsp),%r10
+	leaq	0+96(%rsp),%rsi
+	movq	16+96(%rsp),%r11
+	movq	24+96(%rsp),%r12
+	leaq	224(%rsp),%rdi
+	call	__ecp_nistz256_mul_montq
+
+	movq	448(%rsp),%rax
+	leaq	448(%rsp),%rbx
+	movq	0+32(%rsp),%r9
+	movq	8+32(%rsp),%r10
+	leaq	0+32(%rsp),%rsi
+	movq	16+32(%rsp),%r11
+	movq	24+32(%rsp),%r12
+	leaq	256(%rsp),%rdi
+	call	__ecp_nistz256_mul_montq
+
+	movq	416(%rsp),%rax
+	leaq	416(%rsp),%rbx
+	movq	0+224(%rsp),%r9
+	movq	8+224(%rsp),%r10
+	leaq	0+224(%rsp),%rsi
+	movq	16+224(%rsp),%r11
+	movq	24+224(%rsp),%r12
+	leaq	224(%rsp),%rdi
+	call	__ecp_nistz256_mul_montq
+
+	movq	512(%rsp),%rax
+	leaq	512(%rsp),%rbx
+	movq	0+256(%rsp),%r9
+	movq	8+256(%rsp),%r10
+	leaq	0+256(%rsp),%rsi
+	movq	16+256(%rsp),%r11
+	movq	24+256(%rsp),%r12
+	leaq	256(%rsp),%rdi
+	call	__ecp_nistz256_mul_montq
+
+	leaq	224(%rsp),%rbx
+	leaq	64(%rsp),%rdi
+	call	__ecp_nistz256_sub_fromq
+
+	orq	%r13,%r12
+	movdqa	%xmm4,%xmm2
+	orq	%r8,%r12
+	orq	%r9,%r12
+	por	%xmm5,%xmm2
+.byte	102,73,15,110,220
+
+	movq	384(%rsp),%rax
+	leaq	384(%rsp),%rbx
+	movq	0+96(%rsp),%r9
+	movq	8+96(%rsp),%r10
+	leaq	0+96(%rsp),%rsi
+	movq	16+96(%rsp),%r11
+	movq	24+96(%rsp),%r12
+	leaq	160(%rsp),%rdi
+	call	__ecp_nistz256_mul_montq
+
+	movq	480(%rsp),%rax
+	leaq	480(%rsp),%rbx
+	movq	0+32(%rsp),%r9
+	movq	8+32(%rsp),%r10
+	leaq	0+32(%rsp),%rsi
+	movq	16+32(%rsp),%r11
+	movq	24+32(%rsp),%r12
+	leaq	192(%rsp),%rdi
+	call	__ecp_nistz256_mul_montq
+
+	leaq	160(%rsp),%rbx
+	leaq	0(%rsp),%rdi
+	call	__ecp_nistz256_sub_fromq
+
+	orq	%r13,%r12
+	orq	%r8,%r12
+	orq	%r9,%r12
+
+.byte	0x3e
+	jnz	L$add_proceedq
+.byte	102,73,15,126,208
+.byte	102,73,15,126,217
+	testq	%r8,%r8
+	jnz	L$add_proceedq
+	testq	%r9,%r9
+	jz	L$add_doubleq
+
+.byte	102,72,15,126,199
+	pxor	%xmm0,%xmm0
+	movdqu	%xmm0,0(%rdi)
+	movdqu	%xmm0,16(%rdi)
+	movdqu	%xmm0,32(%rdi)
+	movdqu	%xmm0,48(%rdi)
+	movdqu	%xmm0,64(%rdi)
+	movdqu	%xmm0,80(%rdi)
+	jmp	L$add_doneq
+
+.p2align	5
+L$add_doubleq:
+.byte	102,72,15,126,206
+.byte	102,72,15,126,199
+	addq	$416,%rsp
+	jmp	L$point_double_shortcutq
+
+.p2align	5
+L$add_proceedq:
+	movq	0+64(%rsp),%rax
+	movq	8+64(%rsp),%r14
+	leaq	0+64(%rsp),%rsi
+	movq	16+64(%rsp),%r15
+	movq	24+64(%rsp),%r8
+	leaq	96(%rsp),%rdi
+	call	__ecp_nistz256_sqr_montq
+
+	movq	448(%rsp),%rax
+	leaq	448(%rsp),%rbx
+	movq	0+0(%rsp),%r9
+	movq	8+0(%rsp),%r10
+	leaq	0+0(%rsp),%rsi
+	movq	16+0(%rsp),%r11
+	movq	24+0(%rsp),%r12
+	leaq	352(%rsp),%rdi
+	call	__ecp_nistz256_mul_montq
+
+	movq	0+0(%rsp),%rax
+	movq	8+0(%rsp),%r14
+	leaq	0+0(%rsp),%rsi
+	movq	16+0(%rsp),%r15
+	movq	24+0(%rsp),%r8
+	leaq	32(%rsp),%rdi
+	call	__ecp_nistz256_sqr_montq
+
+	movq	544(%rsp),%rax
+	leaq	544(%rsp),%rbx
+	movq	0+352(%rsp),%r9
+	movq	8+352(%rsp),%r10
+	leaq	0+352(%rsp),%rsi
+	movq	16+352(%rsp),%r11
+	movq	24+352(%rsp),%r12
+	leaq	352(%rsp),%rdi
+	call	__ecp_nistz256_mul_montq
+
+	movq	0(%rsp),%rax
+	leaq	0(%rsp),%rbx
+	movq	0+32(%rsp),%r9
+	movq	8+32(%rsp),%r10
+	leaq	0+32(%rsp),%rsi
+	movq	16+32(%rsp),%r11
+	movq	24+32(%rsp),%r12
+	leaq	128(%rsp),%rdi
+	call	__ecp_nistz256_mul_montq
+
+	movq	160(%rsp),%rax
+	leaq	160(%rsp),%rbx
+	movq	0+32(%rsp),%r9
+	movq	8+32(%rsp),%r10
+	leaq	0+32(%rsp),%rsi
+	movq	16+32(%rsp),%r11
+	movq	24+32(%rsp),%r12
+	leaq	192(%rsp),%rdi
+	call	__ecp_nistz256_mul_montq
+
+
+
+
+	addq	%r12,%r12
+	leaq	96(%rsp),%rsi
+	adcq	%r13,%r13
+	movq	%r12,%rax
+	adcq	%r8,%r8
+	adcq	%r9,%r9
+	movq	%r13,%rbp
+	sbbq	%r11,%r11
+
+	subq	$-1,%r12
+	movq	%r8,%rcx
+	sbbq	%r14,%r13
+	sbbq	$0,%r8
+	movq	%r9,%r10
+	sbbq	%r15,%r9
+	testq	%r11,%r11
+
+	cmovzq	%rax,%r12
+	movq	0(%rsi),%rax
+	cmovzq	%rbp,%r13
+	movq	8(%rsi),%rbp
+	cmovzq	%rcx,%r8
+	movq	16(%rsi),%rcx
+	cmovzq	%r10,%r9
+	movq	24(%rsi),%r10
+
+	call	__ecp_nistz256_subq
+
+	leaq	128(%rsp),%rbx
+	leaq	288(%rsp),%rdi
+	call	__ecp_nistz256_sub_fromq
+
+	movq	192+0(%rsp),%rax
+	movq	192+8(%rsp),%rbp
+	movq	192+16(%rsp),%rcx
+	movq	192+24(%rsp),%r10
+	leaq	320(%rsp),%rdi
+
+	call	__ecp_nistz256_subq
+
+	movq	%r12,0(%rdi)
+	movq	%r13,8(%rdi)
+	movq	%r8,16(%rdi)
+	movq	%r9,24(%rdi)
+	movq	128(%rsp),%rax
+	leaq	128(%rsp),%rbx
+	movq	0+224(%rsp),%r9
+	movq	8+224(%rsp),%r10
+	leaq	0+224(%rsp),%rsi
+	movq	16+224(%rsp),%r11
+	movq	24+224(%rsp),%r12
+	leaq	256(%rsp),%rdi
+	call	__ecp_nistz256_mul_montq
+
+	movq	320(%rsp),%rax
+	leaq	320(%rsp),%rbx
+	movq	0+64(%rsp),%r9
+	movq	8+64(%rsp),%r10
+	leaq	0+64(%rsp),%rsi
+	movq	16+64(%rsp),%r11
+	movq	24+64(%rsp),%r12
+	leaq	320(%rsp),%rdi
+	call	__ecp_nistz256_mul_montq
+
+	leaq	256(%rsp),%rbx
+	leaq	320(%rsp),%rdi
+	call	__ecp_nistz256_sub_fromq
+
+.byte	102,72,15,126,199
+
+	movdqa	%xmm5,%xmm0
+	movdqa	%xmm5,%xmm1
+	pandn	352(%rsp),%xmm0
+	movdqa	%xmm5,%xmm2
+	pandn	352+16(%rsp),%xmm1
+	movdqa	%xmm5,%xmm3
+	pand	544(%rsp),%xmm2
+	pand	544+16(%rsp),%xmm3
+	por	%xmm0,%xmm2
+	por	%xmm1,%xmm3
+
+	movdqa	%xmm4,%xmm0
+	movdqa	%xmm4,%xmm1
+	pandn	%xmm2,%xmm0
+	movdqa	%xmm4,%xmm2
+	pandn	%xmm3,%xmm1
+	movdqa	%xmm4,%xmm3
+	pand	448(%rsp),%xmm2
+	pand	448+16(%rsp),%xmm3
+	por	%xmm0,%xmm2
+	por	%xmm1,%xmm3
+	movdqu	%xmm2,64(%rdi)
+	movdqu	%xmm3,80(%rdi)
+
+	movdqa	%xmm5,%xmm0
+	movdqa	%xmm5,%xmm1
+	pandn	288(%rsp),%xmm0
+	movdqa	%xmm5,%xmm2
+	pandn	288+16(%rsp),%xmm1
+	movdqa	%xmm5,%xmm3
+	pand	480(%rsp),%xmm2
+	pand	480+16(%rsp),%xmm3
+	por	%xmm0,%xmm2
+	por	%xmm1,%xmm3
+
+	movdqa	%xmm4,%xmm0
+	movdqa	%xmm4,%xmm1
+	pandn	%xmm2,%xmm0
+	movdqa	%xmm4,%xmm2
+	pandn	%xmm3,%xmm1
+	movdqa	%xmm4,%xmm3
+	pand	384(%rsp),%xmm2
+	pand	384+16(%rsp),%xmm3
+	por	%xmm0,%xmm2
+	por	%xmm1,%xmm3
+	movdqu	%xmm2,0(%rdi)
+	movdqu	%xmm3,16(%rdi)
+
+	movdqa	%xmm5,%xmm0
+	movdqa	%xmm5,%xmm1
+	pandn	320(%rsp),%xmm0
+	movdqa	%xmm5,%xmm2
+	pandn	320+16(%rsp),%xmm1
+	movdqa	%xmm5,%xmm3
+	pand	512(%rsp),%xmm2
+	pand	512+16(%rsp),%xmm3
+	por	%xmm0,%xmm2
+	por	%xmm1,%xmm3
+
+	movdqa	%xmm4,%xmm0
+	movdqa	%xmm4,%xmm1
+	pandn	%xmm2,%xmm0
+	movdqa	%xmm4,%xmm2
+	pandn	%xmm3,%xmm1
+	movdqa	%xmm4,%xmm3
+	pand	416(%rsp),%xmm2
+	pand	416+16(%rsp),%xmm3
+	por	%xmm0,%xmm2
+	por	%xmm1,%xmm3
+	movdqu	%xmm2,32(%rdi)
+	movdqu	%xmm3,48(%rdi)
+
+L$add_doneq:
+	addq	$576+8,%rsp
+	popq	%r15
+	popq	%r14
+	popq	%r13
+	popq	%r12
+	popq	%rbx
+	popq	%rbp
+	.byte	0xf3,0xc3
+
+.globl	_ecp_nistz256_point_add_affine
+.private_extern _ecp_nistz256_point_add_affine
+
+.p2align	5
+_ecp_nistz256_point_add_affine:
+	pushq	%rbp
+	pushq	%rbx
+	pushq	%r12
+	pushq	%r13
+	pushq	%r14
+	pushq	%r15
+	subq	$480+8,%rsp
+
+	movdqu	0(%rsi),%xmm0
+	movq	%rdx,%rbx
+	movdqu	16(%rsi),%xmm1
+	movdqu	32(%rsi),%xmm2
+	movdqu	48(%rsi),%xmm3
+	movdqu	64(%rsi),%xmm4
+	movdqu	80(%rsi),%xmm5
+	movq	64+0(%rsi),%rax
+	movq	64+8(%rsi),%r14
+	movq	64+16(%rsi),%r15
+	movq	64+24(%rsi),%r8
+	movdqa	%xmm0,320(%rsp)
+	movdqa	%xmm1,320+16(%rsp)
+	por	%xmm0,%xmm1
+	movdqa	%xmm2,352(%rsp)
+	movdqa	%xmm3,352+16(%rsp)
+	por	%xmm2,%xmm3
+	movdqa	%xmm4,384(%rsp)
+	movdqa	%xmm5,384+16(%rsp)
+	por	%xmm1,%xmm3
+
+	movdqu	0(%rbx),%xmm0
+	pshufd	$0xb1,%xmm3,%xmm5
+	movdqu	16(%rbx),%xmm1
+	movdqu	32(%rbx),%xmm2
+	por	%xmm3,%xmm5
+	movdqu	48(%rbx),%xmm3
+	movdqa	%xmm0,416(%rsp)
+	pshufd	$0x1e,%xmm5,%xmm4
+	movdqa	%xmm1,416+16(%rsp)
+	por	%xmm0,%xmm1
+.byte	102,72,15,110,199
+	movdqa	%xmm2,448(%rsp)
+	movdqa	%xmm3,448+16(%rsp)
+	por	%xmm2,%xmm3
+	por	%xmm4,%xmm5
+	pxor	%xmm4,%xmm4
+	por	%xmm1,%xmm3
+
+	leaq	64-0(%rsi),%rsi
+	leaq	32(%rsp),%rdi
+	call	__ecp_nistz256_sqr_montq
+
+	pcmpeqd	%xmm4,%xmm5
+	pshufd	$0xb1,%xmm3,%xmm4
+	movq	0(%rbx),%rax
+
+	movq	%r12,%r9
+	por	%xmm3,%xmm4
+	pshufd	$0,%xmm5,%xmm5
+	pshufd	$0x1e,%xmm4,%xmm3
+	movq	%r13,%r10
+	por	%xmm3,%xmm4
+	pxor	%xmm3,%xmm3
+	movq	%r14,%r11
+	pcmpeqd	%xmm3,%xmm4
+	pshufd	$0,%xmm4,%xmm4
+
+	leaq	32-0(%rsp),%rsi
+	movq	%r15,%r12
+	leaq	0(%rsp),%rdi
+	call	__ecp_nistz256_mul_montq
+
+	leaq	320(%rsp),%rbx
+	leaq	64(%rsp),%rdi
+	call	__ecp_nistz256_sub_fromq
+
+	movq	384(%rsp),%rax
+	leaq	384(%rsp),%rbx
+	movq	0+32(%rsp),%r9
+	movq	8+32(%rsp),%r10
+	leaq	0+32(%rsp),%rsi
+	movq	16+32(%rsp),%r11
+	movq	24+32(%rsp),%r12
+	leaq	32(%rsp),%rdi
+	call	__ecp_nistz256_mul_montq
+
+	movq	384(%rsp),%rax
+	leaq	384(%rsp),%rbx
+	movq	0+64(%rsp),%r9
+	movq	8+64(%rsp),%r10
+	leaq	0+64(%rsp),%rsi
+	movq	16+64(%rsp),%r11
+	movq	24+64(%rsp),%r12
+	leaq	288(%rsp),%rdi
+	call	__ecp_nistz256_mul_montq
+
+	movq	448(%rsp),%rax
+	leaq	448(%rsp),%rbx
+	movq	0+32(%rsp),%r9
+	movq	8+32(%rsp),%r10
+	leaq	0+32(%rsp),%rsi
+	movq	16+32(%rsp),%r11
+	movq	24+32(%rsp),%r12
+	leaq	32(%rsp),%rdi
+	call	__ecp_nistz256_mul_montq
+
+	leaq	352(%rsp),%rbx
+	leaq	96(%rsp),%rdi
+	call	__ecp_nistz256_sub_fromq
+
+	movq	0+64(%rsp),%rax
+	movq	8+64(%rsp),%r14
+	leaq	0+64(%rsp),%rsi
+	movq	16+64(%rsp),%r15
+	movq	24+64(%rsp),%r8
+	leaq	128(%rsp),%rdi
+	call	__ecp_nistz256_sqr_montq
+
+	movq	0+96(%rsp),%rax
+	movq	8+96(%rsp),%r14
+	leaq	0+96(%rsp),%rsi
+	movq	16+96(%rsp),%r15
+	movq	24+96(%rsp),%r8
+	leaq	192(%rsp),%rdi
+	call	__ecp_nistz256_sqr_montq
+
+	movq	128(%rsp),%rax
+	leaq	128(%rsp),%rbx
+	movq	0+64(%rsp),%r9
+	movq	8+64(%rsp),%r10
+	leaq	0+64(%rsp),%rsi
+	movq	16+64(%rsp),%r11
+	movq	24+64(%rsp),%r12
+	leaq	160(%rsp),%rdi
+	call	__ecp_nistz256_mul_montq
+
+	movq	320(%rsp),%rax
+	leaq	320(%rsp),%rbx
+	movq	0+128(%rsp),%r9
+	movq	8+128(%rsp),%r10
+	leaq	0+128(%rsp),%rsi
+	movq	16+128(%rsp),%r11
+	movq	24+128(%rsp),%r12
+	leaq	0(%rsp),%rdi
+	call	__ecp_nistz256_mul_montq
+
+
+
+
+	addq	%r12,%r12
+	leaq	192(%rsp),%rsi
+	adcq	%r13,%r13
+	movq	%r12,%rax
+	adcq	%r8,%r8
+	adcq	%r9,%r9
+	movq	%r13,%rbp
+	sbbq	%r11,%r11
+
+	subq	$-1,%r12
+	movq	%r8,%rcx
+	sbbq	%r14,%r13
+	sbbq	$0,%r8
+	movq	%r9,%r10
+	sbbq	%r15,%r9
+	testq	%r11,%r11
+
+	cmovzq	%rax,%r12
+	movq	0(%rsi),%rax
+	cmovzq	%rbp,%r13
+	movq	8(%rsi),%rbp
+	cmovzq	%rcx,%r8
+	movq	16(%rsi),%rcx
+	cmovzq	%r10,%r9
+	movq	24(%rsi),%r10
+
+	call	__ecp_nistz256_subq
+
+	leaq	160(%rsp),%rbx
+	leaq	224(%rsp),%rdi
+	call	__ecp_nistz256_sub_fromq
+
+	movq	0+0(%rsp),%rax
+	movq	0+8(%rsp),%rbp
+	movq	0+16(%rsp),%rcx
+	movq	0+24(%rsp),%r10
+	leaq	64(%rsp),%rdi
+
+	call	__ecp_nistz256_subq
+
+	movq	%r12,0(%rdi)
+	movq	%r13,8(%rdi)
+	movq	%r8,16(%rdi)
+	movq	%r9,24(%rdi)
+	movq	352(%rsp),%rax
+	leaq	352(%rsp),%rbx
+	movq	0+160(%rsp),%r9
+	movq	8+160(%rsp),%r10
+	leaq	0+160(%rsp),%rsi
+	movq	16+160(%rsp),%r11
+	movq	24+160(%rsp),%r12
+	leaq	32(%rsp),%rdi
+	call	__ecp_nistz256_mul_montq
+
+	movq	96(%rsp),%rax
+	leaq	96(%rsp),%rbx
+	movq	0+64(%rsp),%r9
+	movq	8+64(%rsp),%r10
+	leaq	0+64(%rsp),%rsi
+	movq	16+64(%rsp),%r11
+	movq	24+64(%rsp),%r12
+	leaq	64(%rsp),%rdi
+	call	__ecp_nistz256_mul_montq
+
+	leaq	32(%rsp),%rbx
+	leaq	256(%rsp),%rdi
+	call	__ecp_nistz256_sub_fromq
+
+.byte	102,72,15,126,199
+
+	movdqa	%xmm5,%xmm0
+	movdqa	%xmm5,%xmm1
+	pandn	288(%rsp),%xmm0
+	movdqa	%xmm5,%xmm2
+	pandn	288+16(%rsp),%xmm1
+	movdqa	%xmm5,%xmm3
+	pand	L$ONE_mont(%rip),%xmm2
+	pand	L$ONE_mont+16(%rip),%xmm3
+	por	%xmm0,%xmm2
+	por	%xmm1,%xmm3
+
+	movdqa	%xmm4,%xmm0
+	movdqa	%xmm4,%xmm1
+	pandn	%xmm2,%xmm0
+	movdqa	%xmm4,%xmm2
+	pandn	%xmm3,%xmm1
+	movdqa	%xmm4,%xmm3
+	pand	384(%rsp),%xmm2
+	pand	384+16(%rsp),%xmm3
+	por	%xmm0,%xmm2
+	por	%xmm1,%xmm3
+	movdqu	%xmm2,64(%rdi)
+	movdqu	%xmm3,80(%rdi)
+
+	movdqa	%xmm5,%xmm0
+	movdqa	%xmm5,%xmm1
+	pandn	224(%rsp),%xmm0
+	movdqa	%xmm5,%xmm2
+	pandn	224+16(%rsp),%xmm1
+	movdqa	%xmm5,%xmm3
+	pand	416(%rsp),%xmm2
+	pand	416+16(%rsp),%xmm3
+	por	%xmm0,%xmm2
+	por	%xmm1,%xmm3
+
+	movdqa	%xmm4,%xmm0
+	movdqa	%xmm4,%xmm1
+	pandn	%xmm2,%xmm0
+	movdqa	%xmm4,%xmm2
+	pandn	%xmm3,%xmm1
+	movdqa	%xmm4,%xmm3
+	pand	320(%rsp),%xmm2
+	pand	320+16(%rsp),%xmm3
+	por	%xmm0,%xmm2
+	por	%xmm1,%xmm3
+	movdqu	%xmm2,0(%rdi)
+	movdqu	%xmm3,16(%rdi)
+
+	movdqa	%xmm5,%xmm0
+	movdqa	%xmm5,%xmm1
+	pandn	256(%rsp),%xmm0
+	movdqa	%xmm5,%xmm2
+	pandn	256+16(%rsp),%xmm1
+	movdqa	%xmm5,%xmm3
+	pand	448(%rsp),%xmm2
+	pand	448+16(%rsp),%xmm3
+	por	%xmm0,%xmm2
+	por	%xmm1,%xmm3
+
+	movdqa	%xmm4,%xmm0
+	movdqa	%xmm4,%xmm1
+	pandn	%xmm2,%xmm0
+	movdqa	%xmm4,%xmm2
+	pandn	%xmm3,%xmm1
+	movdqa	%xmm4,%xmm3
+	pand	352(%rsp),%xmm2
+	pand	352+16(%rsp),%xmm3
+	por	%xmm0,%xmm2
+	por	%xmm1,%xmm3
+	movdqu	%xmm2,32(%rdi)
+	movdqu	%xmm3,48(%rdi)
+
+	addq	$480+8,%rsp
+	popq	%r15
+	popq	%r14
+	popq	%r13
+	popq	%r12
+	popq	%rbx
+	popq	%rbp
+	.byte	0xf3,0xc3
+
+#endif

third_party/boringssl/mac-x86_64/crypto/md5/md5-x86_64.S

diff --git a/third_party/boringssl/mac-x86_64/crypto/md5/md5-x86_64.S b/third_party/boringssl/mac-x86_64/crypto/md5/md5-x86_64.S
index 1e61479..16fd2cc 100644
--- a/third_party/boringssl/mac-x86_64/crypto/md5/md5-x86_64.S
+++ b/third_party/boringssl/mac-x86_64/crypto/md5/md5-x86_64.S
@@ -495,14 +495,14 @@
 	movl	%ecx,%r11d
 	addl	%ecx,%ebx
 	movl	0(%rsi),%r10d
-	movl	$4294967295,%r11d
+	movl	$0xffffffff,%r11d
 	xorl	%edx,%r11d
 	leal	-198630844(%rax,%r10,1),%eax
 	orl	%ebx,%r11d
 	xorl	%ecx,%r11d
 	addl	%r11d,%eax
 	movl	28(%rsi),%r10d
-	movl	$4294967295,%r11d
+	movl	$0xffffffff,%r11d
 	roll	$6,%eax
 	xorl	%ecx,%r11d
 	addl	%ebx,%eax
@@ -511,7 +511,7 @@
 	xorl	%ebx,%r11d
 	addl	%r11d,%edx
 	movl	56(%rsi),%r10d
-	movl	$4294967295,%r11d
+	movl	$0xffffffff,%r11d
 	roll	$10,%edx
 	xorl	%ebx,%r11d
 	addl	%eax,%edx
@@ -520,7 +520,7 @@
 	xorl	%eax,%r11d
 	addl	%r11d,%ecx
 	movl	20(%rsi),%r10d
-	movl	$4294967295,%r11d
+	movl	$0xffffffff,%r11d
 	roll	$15,%ecx
 	xorl	%eax,%r11d
 	addl	%edx,%ecx
@@ -529,7 +529,7 @@
 	xorl	%edx,%r11d
 	addl	%r11d,%ebx
 	movl	48(%rsi),%r10d
-	movl	$4294967295,%r11d
+	movl	$0xffffffff,%r11d
 	roll	$21,%ebx
 	xorl	%edx,%r11d
 	addl	%ecx,%ebx
@@ -538,7 +538,7 @@
 	xorl	%ecx,%r11d
 	addl	%r11d,%eax
 	movl	12(%rsi),%r10d
-	movl	$4294967295,%r11d
+	movl	$0xffffffff,%r11d
 	roll	$6,%eax
 	xorl	%ecx,%r11d
 	addl	%ebx,%eax
@@ -547,7 +547,7 @@
 	xorl	%ebx,%r11d
 	addl	%r11d,%edx
 	movl	40(%rsi),%r10d
-	movl	$4294967295,%r11d
+	movl	$0xffffffff,%r11d
 	roll	$10,%edx
 	xorl	%ebx,%r11d
 	addl	%eax,%edx
@@ -556,7 +556,7 @@
 	xorl	%eax,%r11d
 	addl	%r11d,%ecx
 	movl	4(%rsi),%r10d
-	movl	$4294967295,%r11d
+	movl	$0xffffffff,%r11d
 	roll	$15,%ecx
 	xorl	%eax,%r11d
 	addl	%edx,%ecx
@@ -565,7 +565,7 @@
 	xorl	%edx,%r11d
 	addl	%r11d,%ebx
 	movl	32(%rsi),%r10d
-	movl	$4294967295,%r11d
+	movl	$0xffffffff,%r11d
 	roll	$21,%ebx
 	xorl	%edx,%r11d
 	addl	%ecx,%ebx
@@ -574,7 +574,7 @@
 	xorl	%ecx,%r11d
 	addl	%r11d,%eax
 	movl	60(%rsi),%r10d
-	movl	$4294967295,%r11d
+	movl	$0xffffffff,%r11d
 	roll	$6,%eax
 	xorl	%ecx,%r11d
 	addl	%ebx,%eax
@@ -583,7 +583,7 @@
 	xorl	%ebx,%r11d
 	addl	%r11d,%edx
 	movl	24(%rsi),%r10d
-	movl	$4294967295,%r11d
+	movl	$0xffffffff,%r11d
 	roll	$10,%edx
 	xorl	%ebx,%r11d
 	addl	%eax,%edx
@@ -592,7 +592,7 @@
 	xorl	%eax,%r11d
 	addl	%r11d,%ecx
 	movl	52(%rsi),%r10d
-	movl	$4294967295,%r11d
+	movl	$0xffffffff,%r11d
 	roll	$15,%ecx
 	xorl	%eax,%r11d
 	addl	%edx,%ecx
@@ -601,7 +601,7 @@
 	xorl	%edx,%r11d
 	addl	%r11d,%ebx
 	movl	16(%rsi),%r10d
-	movl	$4294967295,%r11d
+	movl	$0xffffffff,%r11d
 	roll	$21,%ebx
 	xorl	%edx,%r11d
 	addl	%ecx,%ebx
@@ -610,7 +610,7 @@
 	xorl	%ecx,%r11d
 	addl	%r11d,%eax
 	movl	44(%rsi),%r10d
-	movl	$4294967295,%r11d
+	movl	$0xffffffff,%r11d
 	roll	$6,%eax
 	xorl	%ecx,%r11d
 	addl	%ebx,%eax
@@ -619,7 +619,7 @@
 	xorl	%ebx,%r11d
 	addl	%r11d,%edx
 	movl	8(%rsi),%r10d
-	movl	$4294967295,%r11d
+	movl	$0xffffffff,%r11d
 	roll	$10,%edx
 	xorl	%ebx,%r11d
 	addl	%eax,%edx
@@ -628,7 +628,7 @@
 	xorl	%eax,%r11d
 	addl	%r11d,%ecx
 	movl	36(%rsi),%r10d
-	movl	$4294967295,%r11d
+	movl	$0xffffffff,%r11d
 	roll	$15,%ecx
 	xorl	%eax,%r11d
 	addl	%edx,%ecx
@@ -637,7 +637,7 @@
 	xorl	%edx,%r11d
 	addl	%r11d,%ebx
 	movl	0(%rsi),%r10d
-	movl	$4294967295,%r11d
+	movl	$0xffffffff,%r11d
 	roll	$21,%ebx
 	xorl	%edx,%r11d
 	addl	%ecx,%ebx

third_party/boringssl/mac-x86_64/crypto/modes/ghash-x86_64.S

diff --git a/third_party/boringssl/mac-x86_64/crypto/modes/ghash-x86_64.S b/third_party/boringssl/mac-x86_64/crypto/modes/ghash-x86_64.S
index 305a91c..1072c7f 100644
--- a/third_party/boringssl/mac-x86_64/crypto/modes/ghash-x86_64.S
+++ b/third_party/boringssl/mac-x86_64/crypto/modes/ghash-x86_64.S
@@ -22,14 +22,14 @@
 	movq	$14,%rcx
 	movq	8(%rsi,%rax,1),%r8
 	movq	(%rsi,%rax,1),%r9
-	andb	$240,%bl
+	andb	$0xf0,%bl
 	movq	%r8,%rdx
 	jmp	L$oop1
 
 .p2align	4
 L$oop1:
 	shrq	$4,%r8
-	andq	$15,%rdx
+	andq	$0xf,%rdx
 	movq	%r9,%r10
 	movb	(%rdi,%rcx,1),%al
 	shrq	$4,%r9
@@ -45,13 +45,13 @@
 	js	L$break1
 
 	shrq	$4,%r8
-	andq	$15,%rdx
+	andq	$0xf,%rdx
 	movq	%r9,%r10
 	shrq	$4,%r9
 	xorq	8(%rsi,%rax,1),%r8
 	shlq	$60,%r10
 	xorq	(%rsi,%rax,1),%r9
-	andb	$240,%bl
+	andb	$0xf0,%bl
 	xorq	(%r11,%rdx,8),%r9
 	movq	%r8,%rdx
 	xorq	%r10,%r8
@@ -60,19 +60,19 @@
 .p2align	4
 L$break1:
 	shrq	$4,%r8
-	andq	$15,%rdx
+	andq	$0xf,%rdx
 	movq	%r9,%r10
 	shrq	$4,%r9
 	xorq	8(%rsi,%rax,1),%r8
 	shlq	$60,%r10
 	xorq	(%rsi,%rax,1),%r9
-	andb	$240,%bl
+	andb	$0xf0,%bl
 	xorq	(%r11,%rdx,8),%r9
 	movq	%r8,%rdx
 	xorq	%r10,%r8
 
 	shrq	$4,%r8
-	andq	$15,%rdx
+	andq	$0xf,%rdx
 	movq	%r9,%r10
 	shrq	$4,%r9
 	xorq	8(%rsi,%rbx,1),%r8
@@ -880,20 +880,20 @@
 	movdqu	32(%rsi),%xmm7
 .byte	102,65,15,56,0,194
 
-	subq	$16,%rcx
+	subq	$0x10,%rcx
 	jz	L$odd_tail
 
 	movdqu	16(%rsi),%xmm6
 	movl	_OPENSSL_ia32cap_P+4(%rip),%eax
-	cmpq	$48,%rcx
+	cmpq	$0x30,%rcx
 	jb	L$skip4x
 
 	andl	$71303168,%eax
 	cmpl	$4194304,%eax
 	je	L$skip4x
 
-	subq	$48,%rcx
-	movq	$11547335547999543296,%rax
+	subq	$0x30,%rcx
+	movq	$0xA040608020C0E000,%rax
 	movdqu	48(%rsi),%xmm14
 	movdqu	64(%rsi),%xmm15
 
@@ -940,7 +940,7 @@
 	xorps	%xmm13,%xmm5
 
 	leaq	64(%rdx),%rdx
-	subq	$64,%rcx
+	subq	$0x40,%rcx
 	jc	L$tail4x
 
 	jmp	L$mod4_loop
@@ -1023,7 +1023,7 @@
 	xorps	%xmm13,%xmm5
 
 	leaq	64(%rdx),%rdx
-	subq	$64,%rcx
+	subq	$0x40,%rcx
 	jnc	L$mod4_loop
 
 L$tail4x:
@@ -1067,10 +1067,10 @@
 	pxor	%xmm4,%xmm0
 	psrlq	$1,%xmm0
 	pxor	%xmm1,%xmm0
-	addq	$64,%rcx
+	addq	$0x40,%rcx
 	jz	L$done
 	movdqu	32(%rsi),%xmm7
-	subq	$16,%rcx
+	subq	$0x10,%rcx
 	jz	L$odd_tail
 L$skip4x:
 
@@ -1093,7 +1093,7 @@
 
 	leaq	32(%rdx),%rdx
 	nop
-	subq	$32,%rcx
+	subq	$0x20,%rcx
 	jbe	L$even_tail
 	nop
 	jmp	L$mod_loop
@@ -1156,7 +1156,7 @@
 .byte	102,15,58,68,231,0
 	pxor	%xmm1,%xmm0
 
-	subq	$32,%rcx
+	subq	$0x20,%rcx
 	ja	L$mod_loop
 
 L$even_tail:

third_party/boringssl/mac-x86_64/crypto/rc4/rc4-md5-x86_64.S

diff --git a/third_party/boringssl/mac-x86_64/crypto/rc4/rc4-md5-x86_64.S b/third_party/boringssl/mac-x86_64/crypto/rc4/rc4-md5-x86_64.S
deleted file mode 100644
index 31ee7d2..0000000
--- a/third_party/boringssl/mac-x86_64/crypto/rc4/rc4-md5-x86_64.S
+++ /dev/null
@@ -1,1262 +0,0 @@
-#if defined(__x86_64__)
-.text	
-.p2align	4
-
-.globl	_rc4_md5_enc
-.private_extern _rc4_md5_enc
-
-_rc4_md5_enc:
-	cmpq	$0,%r9
-	je	L$abort
-	pushq	%rbx
-	pushq	%rbp
-	pushq	%r12
-	pushq	%r13
-	pushq	%r14
-	pushq	%r15
-	subq	$40,%rsp
-L$body:
-	movq	%rcx,%r11
-	movq	%r9,%r12
-	movq	%rsi,%r13
-	movq	%rdx,%r14
-	movq	%r8,%r15
-	xorq	%rbp,%rbp
-	xorq	%rcx,%rcx
-
-	leaq	8(%rdi),%rdi
-	movb	-8(%rdi),%bpl
-	movb	-4(%rdi),%cl
-
-	incb	%bpl
-	subq	%r13,%r14
-	movl	(%rdi,%rbp,4),%eax
-	addb	%al,%cl
-	leaq	(%rdi,%rbp,4),%rsi
-	shlq	$6,%r12
-	addq	%r15,%r12
-	movq	%r12,16(%rsp)
-
-	movq	%r11,24(%rsp)
-	movl	0(%r11),%r8d
-	movl	4(%r11),%r9d
-	movl	8(%r11),%r10d
-	movl	12(%r11),%r11d
-	jmp	L$oop
-
-.p2align	4
-L$oop:
-	movl	%r8d,0(%rsp)
-	movl	%r9d,4(%rsp)
-	movl	%r10d,8(%rsp)
-	movl	%r11d,%r12d
-	movl	%r11d,12(%rsp)
-	pxor	%xmm0,%xmm0
-	movl	(%rdi,%rcx,4),%edx
-	xorl	%r10d,%r12d
-	movl	%eax,(%rdi,%rcx,4)
-	andl	%r9d,%r12d
-	addl	0(%r15),%r8d
-	addb	%dl,%al
-	movl	4(%rsi),%ebx
-	addl	$3614090360,%r8d
-	xorl	%r11d,%r12d
-	movzbl	%al,%eax
-	movl	%edx,0(%rsi)
-	addl	%r12d,%r8d
-	addb	%bl,%cl
-	roll	$7,%r8d
-	movl	%r10d,%r12d
-	movd	(%rdi,%rax,4),%xmm0
-
-	addl	%r9d,%r8d
-	pxor	%xmm1,%xmm1
-	movl	(%rdi,%rcx,4),%edx
-	xorl	%r9d,%r12d
-	movl	%ebx,(%rdi,%rcx,4)
-	andl	%r8d,%r12d
-	addl	4(%r15),%r11d
-	addb	%dl,%bl
-	movl	8(%rsi),%eax
-	addl	$3905402710,%r11d
-	xorl	%r10d,%r12d
-	movzbl	%bl,%ebx
-	movl	%edx,4(%rsi)
-	addl	%r12d,%r11d
-	addb	%al,%cl
-	roll	$12,%r11d
-	movl	%r9d,%r12d
-	movd	(%rdi,%rbx,4),%xmm1
-
-	addl	%r8d,%r11d
-	movl	(%rdi,%rcx,4),%edx
-	xorl	%r8d,%r12d
-	movl	%eax,(%rdi,%rcx,4)
-	andl	%r11d,%r12d
-	addl	8(%r15),%r10d
-	addb	%dl,%al
-	movl	12(%rsi),%ebx
-	addl	$606105819,%r10d
-	xorl	%r9d,%r12d
-	movzbl	%al,%eax
-	movl	%edx,8(%rsi)
-	addl	%r12d,%r10d
-	addb	%bl,%cl
-	roll	$17,%r10d
-	movl	%r8d,%r12d
-	pinsrw	$1,(%rdi,%rax,4),%xmm0
-
-	addl	%r11d,%r10d
-	movl	(%rdi,%rcx,4),%edx
-	xorl	%r11d,%r12d
-	movl	%ebx,(%rdi,%rcx,4)
-	andl	%r10d,%r12d
-	addl	12(%r15),%r9d
-	addb	%dl,%bl
-	movl	16(%rsi),%eax
-	addl	$3250441966,%r9d
-	xorl	%r8d,%r12d
-	movzbl	%bl,%ebx
-	movl	%edx,12(%rsi)
-	addl	%r12d,%r9d
-	addb	%al,%cl
-	roll	$22,%r9d
-	movl	%r11d,%r12d
-	pinsrw	$1,(%rdi,%rbx,4),%xmm1
-
-	addl	%r10d,%r9d
-	movl	(%rdi,%rcx,4),%edx
-	xorl	%r10d,%r12d
-	movl	%eax,(%rdi,%rcx,4)
-	andl	%r9d,%r12d
-	addl	16(%r15),%r8d
-	addb	%dl,%al
-	movl	20(%rsi),%ebx
-	addl	$4118548399,%r8d
-	xorl	%r11d,%r12d
-	movzbl	%al,%eax
-	movl	%edx,16(%rsi)
-	addl	%r12d,%r8d
-	addb	%bl,%cl
-	roll	$7,%r8d
-	movl	%r10d,%r12d
-	pinsrw	$2,(%rdi,%rax,4),%xmm0
-
-	addl	%r9d,%r8d
-	movl	(%rdi,%rcx,4),%edx
-	xorl	%r9d,%r12d
-	movl	%ebx,(%rdi,%rcx,4)
-	andl	%r8d,%r12d
-	addl	20(%r15),%r11d
-	addb	%dl,%bl
-	movl	24(%rsi),%eax
-	addl	$1200080426,%r11d
-	xorl	%r10d,%r12d
-	movzbl	%bl,%ebx
-	movl	%edx,20(%rsi)
-	addl	%r12d,%r11d
-	addb	%al,%cl
-	roll	$12,%r11d
-	movl	%r9d,%r12d
-	pinsrw	$2,(%rdi,%rbx,4),%xmm1
-
-	addl	%r8d,%r11d
-	movl	(%rdi,%rcx,4),%edx
-	xorl	%r8d,%r12d
-	movl	%eax,(%rdi,%rcx,4)
-	andl	%r11d,%r12d
-	addl	24(%r15),%r10d
-	addb	%dl,%al
-	movl	28(%rsi),%ebx
-	addl	$2821735955,%r10d
-	xorl	%r9d,%r12d
-	movzbl	%al,%eax
-	movl	%edx,24(%rsi)
-	addl	%r12d,%r10d
-	addb	%bl,%cl
-	roll	$17,%r10d
-	movl	%r8d,%r12d
-	pinsrw	$3,(%rdi,%rax,4),%xmm0
-
-	addl	%r11d,%r10d
-	movl	(%rdi,%rcx,4),%edx
-	xorl	%r11d,%r12d
-	movl	%ebx,(%rdi,%rcx,4)
-	andl	%r10d,%r12d
-	addl	28(%r15),%r9d
-	addb	%dl,%bl
-	movl	32(%rsi),%eax
-	addl	$4249261313,%r9d
-	xorl	%r8d,%r12d
-	movzbl	%bl,%ebx
-	movl	%edx,28(%rsi)
-	addl	%r12d,%r9d
-	addb	%al,%cl
-	roll	$22,%r9d
-	movl	%r11d,%r12d
-	pinsrw	$3,(%rdi,%rbx,4),%xmm1
-
-	addl	%r10d,%r9d
-	movl	(%rdi,%rcx,4),%edx
-	xorl	%r10d,%r12d
-	movl	%eax,(%rdi,%rcx,4)
-	andl	%r9d,%r12d
-	addl	32(%r15),%r8d
-	addb	%dl,%al
-	movl	36(%rsi),%ebx
-	addl	$1770035416,%r8d
-	xorl	%r11d,%r12d
-	movzbl	%al,%eax
-	movl	%edx,32(%rsi)
-	addl	%r12d,%r8d
-	addb	%bl,%cl
-	roll	$7,%r8d
-	movl	%r10d,%r12d
-	pinsrw	$4,(%rdi,%rax,4),%xmm0
-
-	addl	%r9d,%r8d
-	movl	(%rdi,%rcx,4),%edx
-	xorl	%r9d,%r12d
-	movl	%ebx,(%rdi,%rcx,4)
-	andl	%r8d,%r12d
-	addl	36(%r15),%r11d
-	addb	%dl,%bl
-	movl	40(%rsi),%eax
-	addl	$2336552879,%r11d
-	xorl	%r10d,%r12d
-	movzbl	%bl,%ebx
-	movl	%edx,36(%rsi)
-	addl	%r12d,%r11d
-	addb	%al,%cl
-	roll	$12,%r11d
-	movl	%r9d,%r12d
-	pinsrw	$4,(%rdi,%rbx,4),%xmm1
-
-	addl	%r8d,%r11d
-	movl	(%rdi,%rcx,4),%edx
-	xorl	%r8d,%r12d
-	movl	%eax,(%rdi,%rcx,4)
-	andl	%r11d,%r12d
-	addl	40(%r15),%r10d
-	addb	%dl,%al
-	movl	44(%rsi),%ebx
-	addl	$4294925233,%r10d
-	xorl	%r9d,%r12d
-	movzbl	%al,%eax
-	movl	%edx,40(%rsi)
-	addl	%r12d,%r10d
-	addb	%bl,%cl
-	roll	$17,%r10d
-	movl	%r8d,%r12d
-	pinsrw	$5,(%rdi,%rax,4),%xmm0
-
-	addl	%r11d,%r10d
-	movl	(%rdi,%rcx,4),%edx
-	xorl	%r11d,%r12d
-	movl	%ebx,(%rdi,%rcx,4)
-	andl	%r10d,%r12d
-	addl	44(%r15),%r9d
-	addb	%dl,%bl
-	movl	48(%rsi),%eax
-	addl	$2304563134,%r9d
-	xorl	%r8d,%r12d
-	movzbl	%bl,%ebx
-	movl	%edx,44(%rsi)
-	addl	%r12d,%r9d
-	addb	%al,%cl
-	roll	$22,%r9d
-	movl	%r11d,%r12d
-	pinsrw	$5,(%rdi,%rbx,4),%xmm1
-
-	addl	%r10d,%r9d
-	movl	(%rdi,%rcx,4),%edx
-	xorl	%r10d,%r12d
-	movl	%eax,(%rdi,%rcx,4)
-	andl	%r9d,%r12d
-	addl	48(%r15),%r8d
-	addb	%dl,%al
-	movl	52(%rsi),%ebx
-	addl	$1804603682,%r8d
-	xorl	%r11d,%r12d
-	movzbl	%al,%eax
-	movl	%edx,48(%rsi)
-	addl	%r12d,%r8d
-	addb	%bl,%cl
-	roll	$7,%r8d
-	movl	%r10d,%r12d
-	pinsrw	$6,(%rdi,%rax,4),%xmm0
-
-	addl	%r9d,%r8d
-	movl	(%rdi,%rcx,4),%edx
-	xorl	%r9d,%r12d
-	movl	%ebx,(%rdi,%rcx,4)
-	andl	%r8d,%r12d
-	addl	52(%r15),%r11d
-	addb	%dl,%bl
-	movl	56(%rsi),%eax
-	addl	$4254626195,%r11d
-	xorl	%r10d,%r12d
-	movzbl	%bl,%ebx
-	movl	%edx,52(%rsi)
-	addl	%r12d,%r11d
-	addb	%al,%cl
-	roll	$12,%r11d
-	movl	%r9d,%r12d
-	pinsrw	$6,(%rdi,%rbx,4),%xmm1
-
-	addl	%r8d,%r11d
-	movl	(%rdi,%rcx,4),%edx
-	xorl	%r8d,%r12d
-	movl	%eax,(%rdi,%rcx,4)
-	andl	%r11d,%r12d
-	addl	56(%r15),%r10d
-	addb	%dl,%al
-	movl	60(%rsi),%ebx
-	addl	$2792965006,%r10d
-	xorl	%r9d,%r12d
-	movzbl	%al,%eax
-	movl	%edx,56(%rsi)
-	addl	%r12d,%r10d
-	addb	%bl,%cl
-	roll	$17,%r10d
-	movl	%r8d,%r12d
-	pinsrw	$7,(%rdi,%rax,4),%xmm0
-
-	addl	%r11d,%r10d
-	movdqu	(%r13),%xmm2
-	movl	(%rdi,%rcx,4),%edx
-	xorl	%r11d,%r12d
-	movl	%ebx,(%rdi,%rcx,4)
-	andl	%r10d,%r12d
-	addl	60(%r15),%r9d
-	addb	%dl,%bl
-	movl	64(%rsi),%eax
-	addl	$1236535329,%r9d
-	xorl	%r8d,%r12d
-	movzbl	%bl,%ebx
-	movl	%edx,60(%rsi)
-	addl	%r12d,%r9d
-	addb	%al,%cl
-	roll	$22,%r9d
-	movl	%r10d,%r12d
-	pinsrw	$7,(%rdi,%rbx,4),%xmm1
-
-	addl	%r10d,%r9d
-	psllq	$8,%xmm1
-	pxor	%xmm0,%xmm2
-	pxor	%xmm1,%xmm2
-	pxor	%xmm0,%xmm0
-	movl	(%rdi,%rcx,4),%edx
-	xorl	%r9d,%r12d
-	movl	%eax,(%rdi,%rcx,4)
-	andl	%r11d,%r12d
-	addl	4(%r15),%r8d
-	addb	%dl,%al
-	movl	68(%rsi),%ebx
-	addl	$4129170786,%r8d
-	xorl	%r10d,%r12d
-	movzbl	%al,%eax
-	movl	%edx,64(%rsi)
-	addl	%r12d,%r8d
-	addb	%bl,%cl
-	roll	$5,%r8d
-	movl	%r9d,%r12d
-	movd	(%rdi,%rax,4),%xmm0
-
-	addl	%r9d,%r8d
-	pxor	%xmm1,%xmm1
-	movl	(%rdi,%rcx,4),%edx
-	xorl	%r8d,%r12d
-	movl	%ebx,(%rdi,%rcx,4)
-	andl	%r10d,%r12d
-	addl	24(%r15),%r11d
-	addb	%dl,%bl
-	movl	72(%rsi),%eax
-	addl	$3225465664,%r11d
-	xorl	%r9d,%r12d
-	movzbl	%bl,%ebx
-	movl	%edx,68(%rsi)
-	addl	%r12d,%r11d
-	addb	%al,%cl
-	roll	$9,%r11d
-	movl	%r8d,%r12d
-	movd	(%rdi,%rbx,4),%xmm1
-
-	addl	%r8d,%r11d
-	movl	(%rdi,%rcx,4),%edx
-	xorl	%r11d,%r12d
-	movl	%eax,(%rdi,%rcx,4)
-	andl	%r9d,%r12d
-	addl	44(%r15),%r10d
-	addb	%dl,%al
-	movl	76(%rsi),%ebx
-	addl	$643717713,%r10d
-	xorl	%r8d,%r12d
-	movzbl	%al,%eax
-	movl	%edx,72(%rsi)
-	addl	%r12d,%r10d
-	addb	%bl,%cl
-	roll	$14,%r10d
-	movl	%r11d,%r12d
-	pinsrw	$1,(%rdi,%rax,4),%xmm0
-
-	addl	%r11d,%r10d
-	movl	(%rdi,%rcx,4),%edx
-	xorl	%r10d,%r12d
-	movl	%ebx,(%rdi,%rcx,4)
-	andl	%r8d,%r12d
-	addl	0(%r15),%r9d
-	addb	%dl,%bl
-	movl	80(%rsi),%eax
-	addl	$3921069994,%r9d
-	xorl	%r11d,%r12d
-	movzbl	%bl,%ebx
-	movl	%edx,76(%rsi)
-	addl	%r12d,%r9d
-	addb	%al,%cl
-	roll	$20,%r9d
-	movl	%r10d,%r12d
-	pinsrw	$1,(%rdi,%rbx,4),%xmm1
-
-	addl	%r10d,%r9d
-	movl	(%rdi,%rcx,4),%edx
-	xorl	%r9d,%r12d
-	movl	%eax,(%rdi,%rcx,4)
-	andl	%r11d,%r12d
-	addl	20(%r15),%r8d
-	addb	%dl,%al
-	movl	84(%rsi),%ebx
-	addl	$3593408605,%r8d
-	xorl	%r10d,%r12d
-	movzbl	%al,%eax
-	movl	%edx,80(%rsi)
-	addl	%r12d,%r8d
-	addb	%bl,%cl
-	roll	$5,%r8d
-	movl	%r9d,%r12d
-	pinsrw	$2,(%rdi,%rax,4),%xmm0
-
-	addl	%r9d,%r8d
-	movl	(%rdi,%rcx,4),%edx
-	xorl	%r8d,%r12d
-	movl	%ebx,(%rdi,%rcx,4)
-	andl	%r10d,%r12d
-	addl	40(%r15),%r11d
-	addb	%dl,%bl
-	movl	88(%rsi),%eax
-	addl	$38016083,%r11d
-	xorl	%r9d,%r12d
-	movzbl	%bl,%ebx
-	movl	%edx,84(%rsi)
-	addl	%r12d,%r11d
-	addb	%al,%cl
-	roll	$9,%r11d
-	movl	%r8d,%r12d
-	pinsrw	$2,(%rdi,%rbx,4),%xmm1
-
-	addl	%r8d,%r11d
-	movl	(%rdi,%rcx,4),%edx
-	xorl	%r11d,%r12d
-	movl	%eax,(%rdi,%rcx,4)
-	andl	%r9d,%r12d
-	addl	60(%r15),%r10d
-	addb	%dl,%al
-	movl	92(%rsi),%ebx
-	addl	$3634488961,%r10d
-	xorl	%r8d,%r12d
-	movzbl	%al,%eax
-	movl	%edx,88(%rsi)
-	addl	%r12d,%r10d
-	addb	%bl,%cl
-	roll	$14,%r10d
-	movl	%r11d,%r12d
-	pinsrw	$3,(%rdi,%rax,4),%xmm0
-
-	addl	%r11d,%r10d
-	movl	(%rdi,%rcx,4),%edx
-	xorl	%r10d,%r12d
-	movl	%ebx,(%rdi,%rcx,4)
-	andl	%r8d,%r12d
-	addl	16(%r15),%r9d
-	addb	%dl,%bl
-	movl	96(%rsi),%eax
-	addl	$3889429448,%r9d
-	xorl	%r11d,%r12d
-	movzbl	%bl,%ebx
-	movl	%edx,92(%rsi)
-	addl	%r12d,%r9d
-	addb	%al,%cl
-	roll	$20,%r9d
-	movl	%r10d,%r12d
-	pinsrw	$3,(%rdi,%rbx,4),%xmm1
-
-	addl	%r10d,%r9d
-	movl	(%rdi,%rcx,4),%edx
-	xorl	%r9d,%r12d
-	movl	%eax,(%rdi,%rcx,4)
-	andl	%r11d,%r12d
-	addl	36(%r15),%r8d
-	addb	%dl,%al
-	movl	100(%rsi),%ebx
-	addl	$568446438,%r8d
-	xorl	%r10d,%r12d
-	movzbl	%al,%eax
-	movl	%edx,96(%rsi)
-	addl	%r12d,%r8d
-	addb	%bl,%cl
-	roll	$5,%r8d
-	movl	%r9d,%r12d
-	pinsrw	$4,(%rdi,%rax,4),%xmm0
-
-	addl	%r9d,%r8d
-	movl	(%rdi,%rcx,4),%edx
-	xorl	%r8d,%r12d
-	movl	%ebx,(%rdi,%rcx,4)
-	andl	%r10d,%r12d
-	addl	56(%r15),%r11d
-	addb	%dl,%bl
-	movl	104(%rsi),%eax
-	addl	$3275163606,%r11d
-	xorl	%r9d,%r12d
-	movzbl	%bl,%ebx
-	movl	%edx,100(%rsi)
-	addl	%r12d,%r11d
-	addb	%al,%cl
-	roll	$9,%r11d
-	movl	%r8d,%r12d
-	pinsrw	$4,(%rdi,%rbx,4),%xmm1
-
-	addl	%r8d,%r11d
-	movl	(%rdi,%rcx,4),%edx
-	xorl	%r11d,%r12d
-	movl	%eax,(%rdi,%rcx,4)
-	andl	%r9d,%r12d
-	addl	12(%r15),%r10d
-	addb	%dl,%al
-	movl	108(%rsi),%ebx
-	addl	$4107603335,%r10d
-	xorl	%r8d,%r12d
-	movzbl	%al,%eax
-	movl	%edx,104(%rsi)
-	addl	%r12d,%r10d
-	addb	%bl,%cl
-	roll	$14,%r10d
-	movl	%r11d,%r12d
-	pinsrw	$5,(%rdi,%rax,4),%xmm0
-
-	addl	%r11d,%r10d
-	movl	(%rdi,%rcx,4),%edx
-	xorl	%r10d,%r12d
-	movl	%ebx,(%rdi,%rcx,4)
-	andl	%r8d,%r12d
-	addl	32(%r15),%r9d
-	addb	%dl,%bl
-	movl	112(%rsi),%eax
-	addl	$1163531501,%r9d
-	xorl	%r11d,%r12d
-	movzbl	%bl,%ebx
-	movl	%edx,108(%rsi)
-	addl	%r12d,%r9d
-	addb	%al,%cl
-	roll	$20,%r9d
-	movl	%r10d,%r12d
-	pinsrw	$5,(%rdi,%rbx,4),%xmm1
-
-	addl	%r10d,%r9d
-	movl	(%rdi,%rcx,4),%edx
-	xorl	%r9d,%r12d
-	movl	%eax,(%rdi,%rcx,4)
-	andl	%r11d,%r12d
-	addl	52(%r15),%r8d
-	addb	%dl,%al
-	movl	116(%rsi),%ebx
-	addl	$2850285829,%r8d
-	xorl	%r10d,%r12d
-	movzbl	%al,%eax
-	movl	%edx,112(%rsi)
-	addl	%r12d,%r8d
-	addb	%bl,%cl
-	roll	$5,%r8d
-	movl	%r9d,%r12d
-	pinsrw	$6,(%rdi,%rax,4),%xmm0
-
-	addl	%r9d,%r8d
-	movl	(%rdi,%rcx,4),%edx
-	xorl	%r8d,%r12d
-	movl	%ebx,(%rdi,%rcx,4)
-	andl	%r10d,%r12d
-	addl	8(%r15),%r11d
-	addb	%dl,%bl
-	movl	120(%rsi),%eax
-	addl	$4243563512,%r11d
-	xorl	%r9d,%r12d
-	movzbl	%bl,%ebx
-	movl	%edx,116(%rsi)
-	addl	%r12d,%r11d
-	addb	%al,%cl
-	roll	$9,%r11d
-	movl	%r8d,%r12d
-	pinsrw	$6,(%rdi,%rbx,4),%xmm1
-
-	addl	%r8d,%r11d
-	movl	(%rdi,%rcx,4),%edx
-	xorl	%r11d,%r12d
-	movl	%eax,(%rdi,%rcx,4)
-	andl	%r9d,%r12d
-	addl	28(%r15),%r10d
-	addb	%dl,%al
-	movl	124(%rsi),%ebx
-	addl	$1735328473,%r10d
-	xorl	%r8d,%r12d
-	movzbl	%al,%eax
-	movl	%edx,120(%rsi)
-	addl	%r12d,%r10d
-	addb	%bl,%cl
-	roll	$14,%r10d
-	movl	%r11d,%r12d
-	pinsrw	$7,(%rdi,%rax,4),%xmm0
-
-	addl	%r11d,%r10d
-	movdqu	16(%r13),%xmm3
-	addb	$32,%bpl
-	movl	(%rdi,%rcx,4),%edx
-	xorl	%r10d,%r12d
-	movl	%ebx,(%rdi,%rcx,4)
-	andl	%r8d,%r12d
-	addl	48(%r15),%r9d
-	addb	%dl,%bl
-	movl	0(%rdi,%rbp,4),%eax
-	addl	$2368359562,%r9d
-	xorl	%r11d,%r12d
-	movzbl	%bl,%ebx
-	movl	%edx,124(%rsi)
-	addl	%r12d,%r9d
-	addb	%al,%cl
-	roll	$20,%r9d
-	movl	%r11d,%r12d
-	pinsrw	$7,(%rdi,%rbx,4),%xmm1
-
-	addl	%r10d,%r9d
-	movq	%rcx,%rsi
-	xorq	%rcx,%rcx
-	movb	%sil,%cl
-	leaq	(%rdi,%rbp,4),%rsi
-	psllq	$8,%xmm1
-	pxor	%xmm0,%xmm3
-	pxor	%xmm1,%xmm3
-	pxor	%xmm0,%xmm0
-	movl	(%rdi,%rcx,4),%edx
-	xorl	%r10d,%r12d
-	movl	%eax,(%rdi,%rcx,4)
-	xorl	%r9d,%r12d
-	addl	20(%r15),%r8d
-	addb	%dl,%al
-	movl	4(%rsi),%ebx
-	addl	$4294588738,%r8d
-	movzbl	%al,%eax
-	addl	%r12d,%r8d
-	movl	%edx,0(%rsi)
-	addb	%bl,%cl
-	roll	$4,%r8d
-	movl	%r10d,%r12d
-	movd	(%rdi,%rax,4),%xmm0
-
-	addl	%r9d,%r8d
-	pxor	%xmm1,%xmm1
-	movl	(%rdi,%rcx,4),%edx
-	xorl	%r9d,%r12d
-	movl	%ebx,(%rdi,%rcx,4)
-	xorl	%r8d,%r12d
-	addl	32(%r15),%r11d
-	addb	%dl,%bl
-	movl	8(%rsi),%eax
-	addl	$2272392833,%r11d
-	movzbl	%bl,%ebx
-	addl	%r12d,%r11d
-	movl	%edx,4(%rsi)
-	addb	%al,%cl
-	roll	$11,%r11d
-	movl	%r9d,%r12d
-	movd	(%rdi,%rbx,4),%xmm1
-
-	addl	%r8d,%r11d
-	movl	(%rdi,%rcx,4),%edx
-	xorl	%r8d,%r12d
-	movl	%eax,(%rdi,%rcx,4)
-	xorl	%r11d,%r12d
-	addl	44(%r15),%r10d
-	addb	%dl,%al
-	movl	12(%rsi),%ebx
-	addl	$1839030562,%r10d
-	movzbl	%al,%eax
-	addl	%r12d,%r10d
-	movl	%edx,8(%rsi)
-	addb	%bl,%cl
-	roll	$16,%r10d
-	movl	%r8d,%r12d
-	pinsrw	$1,(%rdi,%rax,4),%xmm0
-
-	addl	%r11d,%r10d
-	movl	(%rdi,%rcx,4),%edx
-	xorl	%r11d,%r12d
-	movl	%ebx,(%rdi,%rcx,4)
-	xorl	%r10d,%r12d
-	addl	56(%r15),%r9d
-	addb	%dl,%bl
-	movl	16(%rsi),%eax
-	addl	$4259657740,%r9d
-	movzbl	%bl,%ebx
-	addl	%r12d,%r9d
-	movl	%edx,12(%rsi)
-	addb	%al,%cl
-	roll	$23,%r9d
-	movl	%r11d,%r12d
-	pinsrw	$1,(%rdi,%rbx,4),%xmm1
-
-	addl	%r10d,%r9d
-	movl	(%rdi,%rcx,4),%edx
-	xorl	%r10d,%r12d
-	movl	%eax,(%rdi,%rcx,4)
-	xorl	%r9d,%r12d
-	addl	4(%r15),%r8d
-	addb	%dl,%al
-	movl	20(%rsi),%ebx
-	addl	$2763975236,%r8d
-	movzbl	%al,%eax
-	addl	%r12d,%r8d
-	movl	%edx,16(%rsi)
-	addb	%bl,%cl
-	roll	$4,%r8d
-	movl	%r10d,%r12d
-	pinsrw	$2,(%rdi,%rax,4),%xmm0
-
-	addl	%r9d,%r8d
-	movl	(%rdi,%rcx,4),%edx
-	xorl	%r9d,%r12d
-	movl	%ebx,(%rdi,%rcx,4)
-	xorl	%r8d,%r12d
-	addl	16(%r15),%r11d
-	addb	%dl,%bl
-	movl	24(%rsi),%eax
-	addl	$1272893353,%r11d
-	movzbl	%bl,%ebx
-	addl	%r12d,%r11d
-	movl	%edx,20(%rsi)
-	addb	%al,%cl
-	roll	$11,%r11d
-	movl	%r9d,%r12d
-	pinsrw	$2,(%rdi,%rbx,4),%xmm1
-
-	addl	%r8d,%r11d
-	movl	(%rdi,%rcx,4),%edx
-	xorl	%r8d,%r12d
-	movl	%eax,(%rdi,%rcx,4)
-	xorl	%r11d,%r12d
-	addl	28(%r15),%r10d
-	addb	%dl,%al
-	movl	28(%rsi),%ebx
-	addl	$4139469664,%r10d
-	movzbl	%al,%eax
-	addl	%r12d,%r10d
-	movl	%edx,24(%rsi)
-	addb	%bl,%cl
-	roll	$16,%r10d
-	movl	%r8d,%r12d
-	pinsrw	$3,(%rdi,%rax,4),%xmm0
-
-	addl	%r11d,%r10d
-	movl	(%rdi,%rcx,4),%edx
-	xorl	%r11d,%r12d
-	movl	%ebx,(%rdi,%rcx,4)
-	xorl	%r10d,%r12d
-	addl	40(%r15),%r9d
-	addb	%dl,%bl
-	movl	32(%rsi),%eax
-	addl	$3200236656,%r9d
-	movzbl	%bl,%ebx
-	addl	%r12d,%r9d
-	movl	%edx,28(%rsi)
-	addb	%al,%cl
-	roll	$23,%r9d
-	movl	%r11d,%r12d
-	pinsrw	$3,(%rdi,%rbx,4),%xmm1
-
-	addl	%r10d,%r9d
-	movl	(%rdi,%rcx,4),%edx
-	xorl	%r10d,%r12d
-	movl	%eax,(%rdi,%rcx,4)
-	xorl	%r9d,%r12d
-	addl	52(%r15),%r8d
-	addb	%dl,%al
-	movl	36(%rsi),%ebx
-	addl	$681279174,%r8d
-	movzbl	%al,%eax
-	addl	%r12d,%r8d
-	movl	%edx,32(%rsi)
-	addb	%bl,%cl
-	roll	$4,%r8d
-	movl	%r10d,%r12d
-	pinsrw	$4,(%rdi,%rax,4),%xmm0
-
-	addl	%r9d,%r8d
-	movl	(%rdi,%rcx,4),%edx
-	xorl	%r9d,%r12d
-	movl	%ebx,(%rdi,%rcx,4)
-	xorl	%r8d,%r12d
-	addl	0(%r15),%r11d
-	addb	%dl,%bl
-	movl	40(%rsi),%eax
-	addl	$3936430074,%r11d
-	movzbl	%bl,%ebx
-	addl	%r12d,%r11d
-	movl	%edx,36(%rsi)
-	addb	%al,%cl
-	roll	$11,%r11d
-	movl	%r9d,%r12d
-	pinsrw	$4,(%rdi,%rbx,4),%xmm1
-
-	addl	%r8d,%r11d
-	movl	(%rdi,%rcx,4),%edx
-	xorl	%r8d,%r12d
-	movl	%eax,(%rdi,%rcx,4)
-	xorl	%r11d,%r12d
-	addl	12(%r15),%r10d
-	addb	%dl,%al
-	movl	44(%rsi),%ebx
-	addl	$3572445317,%r10d
-	movzbl	%al,%eax
-	addl	%r12d,%r10d
-	movl	%edx,40(%rsi)
-	addb	%bl,%cl
-	roll	$16,%r10d
-	movl	%r8d,%r12d
-	pinsrw	$5,(%rdi,%rax,4),%xmm0
-
-	addl	%r11d,%r10d
-	movl	(%rdi,%rcx,4),%edx
-	xorl	%r11d,%r12d
-	movl	%ebx,(%rdi,%rcx,4)
-	xorl	%r10d,%r12d
-	addl	24(%r15),%r9d
-	addb	%dl,%bl
-	movl	48(%rsi),%eax
-	addl	$76029189,%r9d
-	movzbl	%bl,%ebx
-	addl	%r12d,%r9d
-	movl	%edx,44(%rsi)
-	addb	%al,%cl
-	roll	$23,%r9d
-	movl	%r11d,%r12d
-	pinsrw	$5,(%rdi,%rbx,4),%xmm1
-
-	addl	%r10d,%r9d
-	movl	(%rdi,%rcx,4),%edx
-	xorl	%r10d,%r12d
-	movl	%eax,(%rdi,%rcx,4)
-	xorl	%r9d,%r12d
-	addl	36(%r15),%r8d
-	addb	%dl,%al
-	movl	52(%rsi),%ebx
-	addl	$3654602809,%r8d
-	movzbl	%al,%eax
-	addl	%r12d,%r8d
-	movl	%edx,48(%rsi)
-	addb	%bl,%cl
-	roll	$4,%r8d
-	movl	%r10d,%r12d
-	pinsrw	$6,(%rdi,%rax,4),%xmm0
-
-	addl	%r9d,%r8d
-	movl	(%rdi,%rcx,4),%edx
-	xorl	%r9d,%r12d
-	movl	%ebx,(%rdi,%rcx,4)
-	xorl	%r8d,%r12d
-	addl	48(%r15),%r11d
-	addb	%dl,%bl
-	movl	56(%rsi),%eax
-	addl	$3873151461,%r11d
-	movzbl	%bl,%ebx
-	addl	%r12d,%r11d
-	movl	%edx,52(%rsi)
-	addb	%al,%cl
-	roll	$11,%r11d
-	movl	%r9d,%r12d
-	pinsrw	$6,(%rdi,%rbx,4),%xmm1
-
-	addl	%r8d,%r11d
-	movl	(%rdi,%rcx,4),%edx
-	xorl	%r8d,%r12d
-	movl	%eax,(%rdi,%rcx,4)
-	xorl	%r11d,%r12d
-	addl	60(%r15),%r10d
-	addb	%dl,%al
-	movl	60(%rsi),%ebx
-	addl	$530742520,%r10d
-	movzbl	%al,%eax
-	addl	%r12d,%r10d
-	movl	%edx,56(%rsi)
-	addb	%bl,%cl
-	roll	$16,%r10d
-	movl	%r8d,%r12d
-	pinsrw	$7,(%rdi,%rax,4),%xmm0
-
-	addl	%r11d,%r10d
-	movdqu	32(%r13),%xmm4
-	movl	(%rdi,%rcx,4),%edx
-	xorl	%r11d,%r12d
-	movl	%ebx,(%rdi,%rcx,4)
-	xorl	%r10d,%r12d
-	addl	8(%r15),%r9d
-	addb	%dl,%bl
-	movl	64(%rsi),%eax
-	addl	$3299628645,%r9d
-	movzbl	%bl,%ebx
-	addl	%r12d,%r9d
-	movl	%edx,60(%rsi)
-	addb	%al,%cl
-	roll	$23,%r9d
-	movl	$-1,%r12d
-	pinsrw	$7,(%rdi,%rbx,4),%xmm1
-
-	addl	%r10d,%r9d
-	psllq	$8,%xmm1
-	pxor	%xmm0,%xmm4
-	pxor	%xmm1,%xmm4
-	pxor	%xmm0,%xmm0
-	movl	(%rdi,%rcx,4),%edx
-	xorl	%r11d,%r12d
-	movl	%eax,(%rdi,%rcx,4)
-	orl	%r9d,%r12d
-	addl	0(%r15),%r8d
-	addb	%dl,%al
-	movl	68(%rsi),%ebx
-	addl	$4096336452,%r8d
-	movzbl	%al,%eax
-	xorl	%r10d,%r12d
-	movl	%edx,64(%rsi)
-	addl	%r12d,%r8d
-	addb	%bl,%cl
-	roll	$6,%r8d
-	movl	$-1,%r12d
-	movd	(%rdi,%rax,4),%xmm0
-
-	addl	%r9d,%r8d
-	pxor	%xmm1,%xmm1
-	movl	(%rdi,%rcx,4),%edx
-	xorl	%r10d,%r12d
-	movl	%ebx,(%rdi,%rcx,4)
-	orl	%r8d,%r12d
-	addl	28(%r15),%r11d
-	addb	%dl,%bl
-	movl	72(%rsi),%eax
-	addl	$1126891415,%r11d
-	movzbl	%bl,%ebx
-	xorl	%r9d,%r12d
-	movl	%edx,68(%rsi)
-	addl	%r12d,%r11d
-	addb	%al,%cl
-	roll	$10,%r11d
-	movl	$-1,%r12d
-	movd	(%rdi,%rbx,4),%xmm1
-
-	addl	%r8d,%r11d
-	movl	(%rdi,%rcx,4),%edx
-	xorl	%r9d,%r12d
-	movl	%eax,(%rdi,%rcx,4)
-	orl	%r11d,%r12d
-	addl	56(%r15),%r10d
-	addb	%dl,%al
-	movl	76(%rsi),%ebx
-	addl	$2878612391,%r10d
-	movzbl	%al,%eax
-	xorl	%r8d,%r12d
-	movl	%edx,72(%rsi)
-	addl	%r12d,%r10d
-	addb	%bl,%cl
-	roll	$15,%r10d
-	movl	$-1,%r12d
-	pinsrw	$1,(%rdi,%rax,4),%xmm0
-
-	addl	%r11d,%r10d
-	movl	(%rdi,%rcx,4),%edx
-	xorl	%r8d,%r12d
-	movl	%ebx,(%rdi,%rcx,4)
-	orl	%r10d,%r12d
-	addl	20(%r15),%r9d
-	addb	%dl,%bl
-	movl	80(%rsi),%eax
-	addl	$4237533241,%r9d
-	movzbl	%bl,%ebx
-	xorl	%r11d,%r12d
-	movl	%edx,76(%rsi)
-	addl	%r12d,%r9d
-	addb	%al,%cl
-	roll	$21,%r9d
-	movl	$-1,%r12d
-	pinsrw	$1,(%rdi,%rbx,4),%xmm1
-
-	addl	%r10d,%r9d
-	movl	(%rdi,%rcx,4),%edx
-	xorl	%r11d,%r12d
-	movl	%eax,(%rdi,%rcx,4)
-	orl	%r9d,%r12d
-	addl	48(%r15),%r8d
-	addb	%dl,%al
-	movl	84(%rsi),%ebx
-	addl	$1700485571,%r8d
-	movzbl	%al,%eax
-	xorl	%r10d,%r12d
-	movl	%edx,80(%rsi)
-	addl	%r12d,%r8d
-	addb	%bl,%cl
-	roll	$6,%r8d
-	movl	$-1,%r12d
-	pinsrw	$2,(%rdi,%rax,4),%xmm0
-
-	addl	%r9d,%r8d
-	movl	(%rdi,%rcx,4),%edx
-	xorl	%r10d,%r12d
-	movl	%ebx,(%rdi,%rcx,4)
-	orl	%r8d,%r12d
-	addl	12(%r15),%r11d
-	addb	%dl,%bl
-	movl	88(%rsi),%eax
-	addl	$2399980690,%r11d
-	movzbl	%bl,%ebx
-	xorl	%r9d,%r12d
-	movl	%edx,84(%rsi)
-	addl	%r12d,%r11d
-	addb	%al,%cl
-	roll	$10,%r11d
-	movl	$-1,%r12d
-	pinsrw	$2,(%rdi,%rbx,4),%xmm1
-
-	addl	%r8d,%r11d
-	movl	(%rdi,%rcx,4),%edx
-	xorl	%r9d,%r12d
-	movl	%eax,(%rdi,%rcx,4)
-	orl	%r11d,%r12d
-	addl	40(%r15),%r10d
-	addb	%dl,%al
-	movl	92(%rsi),%ebx
-	addl	$4293915773,%r10d
-	movzbl	%al,%eax
-	xorl	%r8d,%r12d
-	movl	%edx,88(%rsi)
-	addl	%r12d,%r10d
-	addb	%bl,%cl
-	roll	$15,%r10d
-	movl	$-1,%r12d
-	pinsrw	$3,(%rdi,%rax,4),%xmm0
-
-	addl	%r11d,%r10d
-	movl	(%rdi,%rcx,4),%edx
-	xorl	%r8d,%r12d
-	movl	%ebx,(%rdi,%rcx,4)
-	orl	%r10d,%r12d
-	addl	4(%r15),%r9d
-	addb	%dl,%bl
-	movl	96(%rsi),%eax
-	addl	$2240044497,%r9d
-	movzbl	%bl,%ebx
-	xorl	%r11d,%r12d
-	movl	%edx,92(%rsi)
-	addl	%r12d,%r9d
-	addb	%al,%cl
-	roll	$21,%r9d
-	movl	$-1,%r12d
-	pinsrw	$3,(%rdi,%rbx,4),%xmm1
-
-	addl	%r10d,%r9d
-	movl	(%rdi,%rcx,4),%edx
-	xorl	%r11d,%r12d
-	movl	%eax,(%rdi,%rcx,4)
-	orl	%r9d,%r12d
-	addl	32(%r15),%r8d
-	addb	%dl,%al
-	movl	100(%rsi),%ebx
-	addl	$1873313359,%r8d
-	movzbl	%al,%eax
-	xorl	%r10d,%r12d
-	movl	%edx,96(%rsi)
-	addl	%r12d,%r8d
-	addb	%bl,%cl
-	roll	$6,%r8d
-	movl	$-1,%r12d
-	pinsrw	$4,(%rdi,%rax,4),%xmm0
-
-	addl	%r9d,%r8d
-	movl	(%rdi,%rcx,4),%edx
-	xorl	%r10d,%r12d
-	movl	%ebx,(%rdi,%rcx,4)
-	orl	%r8d,%r12d
-	addl	60(%r15),%r11d
-	addb	%dl,%bl
-	movl	104(%rsi),%eax
-	addl	$4264355552,%r11d
-	movzbl	%bl,%ebx
-	xorl	%r9d,%r12d
-	movl	%edx,100(%rsi)
-	addl	%r12d,%r11d
-	addb	%al,%cl
-	roll	$10,%r11d
-	movl	$-1,%r12d
-	pinsrw	$4,(%rdi,%rbx,4),%xmm1
-
-	addl	%r8d,%r11d
-	movl	(%rdi,%rcx,4),%edx
-	xorl	%r9d,%r12d
-	movl	%eax,(%rdi,%rcx,4)
-	orl	%r11d,%r12d
-	addl	24(%r15),%r10d
-	addb	%dl,%al
-	movl	108(%rsi),%ebx
-	addl	$2734768916,%r10d
-	movzbl	%al,%eax
-	xorl	%r8d,%r12d
-	movl	%edx,104(%rsi)
-	addl	%r12d,%r10d
-	addb	%bl,%cl
-	roll	$15,%r10d
-	movl	$-1,%r12d
-	pinsrw	$5,(%rdi,%rax,4),%xmm0
-
-	addl	%r11d,%r10d
-	movl	(%rdi,%rcx,4),%edx
-	xorl	%r8d,%r12d
-	movl	%ebx,(%rdi,%rcx,4)
-	orl	%r10d,%r12d
-	addl	52(%r15),%r9d
-	addb	%dl,%bl
-	movl	112(%rsi),%eax
-	addl	$1309151649,%r9d
-	movzbl	%bl,%ebx
-	xorl	%r11d,%r12d
-	movl	%edx,108(%rsi)
-	addl	%r12d,%r9d
-	addb	%al,%cl
-	roll	$21,%r9d
-	movl	$-1,%r12d
-	pinsrw	$5,(%rdi,%rbx,4),%xmm1
-
-	addl	%r10d,%r9d
-	movl	(%rdi,%rcx,4),%edx
-	xorl	%r11d,%r12d
-	movl	%eax,(%rdi,%rcx,4)
-	orl	%r9d,%r12d
-	addl	16(%r15),%r8d
-	addb	%dl,%al
-	movl	116(%rsi),%ebx
-	addl	$4149444226,%r8d
-	movzbl	%al,%eax
-	xorl	%r10d,%r12d
-	movl	%edx,112(%rsi)
-	addl	%r12d,%r8d
-	addb	%bl,%cl
-	roll	$6,%r8d
-	movl	$-1,%r12d
-	pinsrw	$6,(%rdi,%rax,4),%xmm0
-
-	addl	%r9d,%r8d
-	movl	(%rdi,%rcx,4),%edx
-	xorl	%r10d,%r12d
-	movl	%ebx,(%rdi,%rcx,4)
-	orl	%r8d,%r12d
-	addl	44(%r15),%r11d
-	addb	%dl,%bl
-	movl	120(%rsi),%eax
-	addl	$3174756917,%r11d
-	movzbl	%bl,%ebx
-	xorl	%r9d,%r12d
-	movl	%edx,116(%rsi)
-	addl	%r12d,%r11d
-	addb	%al,%cl
-	roll	$10,%r11d
-	movl	$-1,%r12d
-	pinsrw	$6,(%rdi,%rbx,4),%xmm1
-
-	addl	%r8d,%r11d
-	movl	(%rdi,%rcx,4),%edx
-	xorl	%r9d,%r12d
-	movl	%eax,(%rdi,%rcx,4)
-	orl	%r11d,%r12d
-	addl	8(%r15),%r10d
-	addb	%dl,%al
-	movl	124(%rsi),%ebx
-	addl	$718787259,%r10d
-	movzbl	%al,%eax
-	xorl	%r8d,%r12d
-	movl	%edx,120(%rsi)
-	addl	%r12d,%r10d
-	addb	%bl,%cl
-	roll	$15,%r10d
-	movl	$-1,%r12d
-	pinsrw	$7,(%rdi,%rax,4),%xmm0
-
-	addl	%r11d,%r10d
-	movdqu	48(%r13),%xmm5
-	addb	$32,%bpl
-	movl	(%rdi,%rcx,4),%edx
-	xorl	%r8d,%r12d
-	movl	%ebx,(%rdi,%rcx,4)
-	orl	%r10d,%r12d
-	addl	36(%r15),%r9d
-	addb	%dl,%bl
-	movl	0(%rdi,%rbp,4),%eax
-	addl	$3951481745,%r9d
-	movzbl	%bl,%ebx
-	xorl	%r11d,%r12d
-	movl	%edx,124(%rsi)
-	addl	%r12d,%r9d
-	addb	%al,%cl
-	roll	$21,%r9d
-	movl	$-1,%r12d
-	pinsrw	$7,(%rdi,%rbx,4),%xmm1
-
-	addl	%r10d,%r9d
-	movq	%rbp,%rsi
-	xorq	%rbp,%rbp
-	movb	%sil,%bpl
-	movq	%rcx,%rsi
-	xorq	%rcx,%rcx
-	movb	%sil,%cl
-	leaq	(%rdi,%rbp,4),%rsi
-	psllq	$8,%xmm1
-	pxor	%xmm0,%xmm5
-	pxor	%xmm1,%xmm5
-	addl	0(%rsp),%r8d
-	addl	4(%rsp),%r9d
-	addl	8(%rsp),%r10d
-	addl	12(%rsp),%r11d
-
-	movdqu	%xmm2,(%r14,%r13,1)
-	movdqu	%xmm3,16(%r14,%r13,1)
-	movdqu	%xmm4,32(%r14,%r13,1)
-	movdqu	%xmm5,48(%r14,%r13,1)
-	leaq	64(%r15),%r15
-	leaq	64(%r13),%r13
-	cmpq	16(%rsp),%r15
-	jb	L$oop
-
-	movq	24(%rsp),%r12
-	subb	%al,%cl
-	movl	%r8d,0(%r12)
-	movl	%r9d,4(%r12)
-	movl	%r10d,8(%r12)
-	movl	%r11d,12(%r12)
-	subb	$1,%bpl
-	movl	%ebp,-8(%rdi)
-	movl	%ecx,-4(%rdi)
-
-	movq	40(%rsp),%r15
-	movq	48(%rsp),%r14
-	movq	56(%rsp),%r13
-	movq	64(%rsp),%r12
-	movq	72(%rsp),%rbp
-	movq	80(%rsp),%rbx
-	leaq	88(%rsp),%rsp
-L$epilogue:
-L$abort:
-	.byte	0xf3,0xc3
-
-#endif

third_party/boringssl/mac-x86_64/crypto/sha/sha1-x86_64.S

diff --git a/third_party/boringssl/mac-x86_64/crypto/sha/sha1-x86_64.S b/third_party/boringssl/mac-x86_64/crypto/sha/sha1-x86_64.S
index 044dc5b..0509d45 100644
--- a/third_party/boringssl/mac-x86_64/crypto/sha/sha1-x86_64.S
+++ b/third_party/boringssl/mac-x86_64/crypto/sha/sha1-x86_64.S
@@ -12,6 +12,11 @@
 	movl	_OPENSSL_ia32cap_P+8(%rip),%r10d
 	testl	$512,%r8d
 	jz	L$ialu
+	andl	$268435456,%r8d
+	andl	$1073741824,%r9d
+	orl	%r9d,%r8d
+	cmpl	$1342177280,%r8d
+	je	_avx_shortcut
 	jmp	_ssse3_shortcut
 
 .p2align	4
@@ -2407,6 +2412,1122 @@
 L$epilogue_ssse3:
 	.byte	0xf3,0xc3
 
+
+.p2align	4
+sha1_block_data_order_avx:
+_avx_shortcut:
+	movq	%rsp,%rax
+	pushq	%rbx
+	pushq	%rbp
+	pushq	%r12
+	pushq	%r13
+	pushq	%r14
+	leaq	-64(%rsp),%rsp
+	vzeroupper
+	movq	%rax,%r14
+	andq	$-64,%rsp
+	movq	%rdi,%r8
+	movq	%rsi,%r9
+	movq	%rdx,%r10
+
+	shlq	$6,%r10
+	addq	%r9,%r10
+	leaq	K_XX_XX+64(%rip),%r11
+
+	movl	0(%r8),%eax
+	movl	4(%r8),%ebx
+	movl	8(%r8),%ecx
+	movl	12(%r8),%edx
+	movl	%ebx,%esi
+	movl	16(%r8),%ebp
+	movl	%ecx,%edi
+	xorl	%edx,%edi
+	andl	%edi,%esi
+
+	vmovdqa	64(%r11),%xmm6
+	vmovdqa	-64(%r11),%xmm11
+	vmovdqu	0(%r9),%xmm0
+	vmovdqu	16(%r9),%xmm1
+	vmovdqu	32(%r9),%xmm2
+	vmovdqu	48(%r9),%xmm3
+	vpshufb	%xmm6,%xmm0,%xmm0
+	addq	$64,%r9
+	vpshufb	%xmm6,%xmm1,%xmm1
+	vpshufb	%xmm6,%xmm2,%xmm2
+	vpshufb	%xmm6,%xmm3,%xmm3
+	vpaddd	%xmm11,%xmm0,%xmm4
+	vpaddd	%xmm11,%xmm1,%xmm5
+	vpaddd	%xmm11,%xmm2,%xmm6
+	vmovdqa	%xmm4,0(%rsp)
+	vmovdqa	%xmm5,16(%rsp)
+	vmovdqa	%xmm6,32(%rsp)
+	jmp	L$oop_avx
+.p2align	4
+L$oop_avx:
+	shrdl	$2,%ebx,%ebx
+	xorl	%edx,%esi
+	vpalignr	$8,%xmm0,%xmm1,%xmm4
+	movl	%eax,%edi
+	addl	0(%rsp),%ebp
+	vpaddd	%xmm3,%xmm11,%xmm9
+	xorl	%ecx,%ebx
+	shldl	$5,%eax,%eax
+	vpsrldq	$4,%xmm3,%xmm8
+	addl	%esi,%ebp
+	andl	%ebx,%edi
+	vpxor	%xmm0,%xmm4,%xmm4
+	xorl	%ecx,%ebx
+	addl	%eax,%ebp
+	vpxor	%xmm2,%xmm8,%xmm8
+	shrdl	$7,%eax,%eax
+	xorl	%ecx,%edi
+	movl	%ebp,%esi
+	addl	4(%rsp),%edx
+	vpxor	%xmm8,%xmm4,%xmm4
+	xorl	%ebx,%eax
+	shldl	$5,%ebp,%ebp
+	vmovdqa	%xmm9,48(%rsp)
+	addl	%edi,%edx
+	andl	%eax,%esi
+	vpsrld	$31,%xmm4,%xmm8
+	xorl	%ebx,%eax
+	addl	%ebp,%edx
+	shrdl	$7,%ebp,%ebp
+	xorl	%ebx,%esi
+	vpslldq	$12,%xmm4,%xmm10
+	vpaddd	%xmm4,%xmm4,%xmm4
+	movl	%edx,%edi
+	addl	8(%rsp),%ecx
+	xorl	%eax,%ebp
+	shldl	$5,%edx,%edx
+	vpsrld	$30,%xmm10,%xmm9
+	vpor	%xmm8,%xmm4,%xmm4
+	addl	%esi,%ecx
+	andl	%ebp,%edi
+	xorl	%eax,%ebp
+	addl	%edx,%ecx
+	vpslld	$2,%xmm10,%xmm10
+	vpxor	%xmm9,%xmm4,%xmm4
+	shrdl	$7,%edx,%edx
+	xorl	%eax,%edi
+	movl	%ecx,%esi
+	addl	12(%rsp),%ebx
+	vpxor	%xmm10,%xmm4,%xmm4
+	xorl	%ebp,%edx
+	shldl	$5,%ecx,%ecx
+	addl	%edi,%ebx
+	andl	%edx,%esi
+	xorl	%ebp,%edx
+	addl	%ecx,%ebx
+	shrdl	$7,%ecx,%ecx
+	xorl	%ebp,%esi
+	vpalignr	$8,%xmm1,%xmm2,%xmm5
+	movl	%ebx,%edi
+	addl	16(%rsp),%eax
+	vpaddd	%xmm4,%xmm11,%xmm9
+	xorl	%edx,%ecx
+	shldl	$5,%ebx,%ebx
+	vpsrldq	$4,%xmm4,%xmm8
+	addl	%esi,%eax
+	andl	%ecx,%edi
+	vpxor	%xmm1,%xmm5,%xmm5
+	xorl	%edx,%ecx
+	addl	%ebx,%eax
+	vpxor	%xmm3,%xmm8,%xmm8
+	shrdl	$7,%ebx,%ebx
+	xorl	%edx,%edi
+	movl	%eax,%esi
+	addl	20(%rsp),%ebp
+	vpxor	%xmm8,%xmm5,%xmm5
+	xorl	%ecx,%ebx
+	shldl	$5,%eax,%eax
+	vmovdqa	%xmm9,0(%rsp)
+	addl	%edi,%ebp
+	andl	%ebx,%esi
+	vpsrld	$31,%xmm5,%xmm8
+	xorl	%ecx,%ebx
+	addl	%eax,%ebp
+	shrdl	$7,%eax,%eax
+	xorl	%ecx,%esi
+	vpslldq	$12,%xmm5,%xmm10
+	vpaddd	%xmm5,%xmm5,%xmm5
+	movl	%ebp,%edi
+	addl	24(%rsp),%edx
+	xorl	%ebx,%eax
+	shldl	$5,%ebp,%ebp
+	vpsrld	$30,%xmm10,%xmm9
+	vpor	%xmm8,%xmm5,%xmm5
+	addl	%esi,%edx
+	andl	%eax,%edi
+	xorl	%ebx,%eax
+	addl	%ebp,%edx
+	vpslld	$2,%xmm10,%xmm10
+	vpxor	%xmm9,%xmm5,%xmm5
+	shrdl	$7,%ebp,%ebp
+	xorl	%ebx,%edi
+	movl	%edx,%esi
+	addl	28(%rsp),%ecx
+	vpxor	%xmm10,%xmm5,%xmm5
+	xorl	%eax,%ebp
+	shldl	$5,%edx,%edx
+	vmovdqa	-32(%r11),%xmm11
+	addl	%edi,%ecx
+	andl	%ebp,%esi
+	xorl	%eax,%ebp
+	addl	%edx,%ecx
+	shrdl	$7,%edx,%edx
+	xorl	%eax,%esi
+	vpalignr	$8,%xmm2,%xmm3,%xmm6
+	movl	%ecx,%edi
+	addl	32(%rsp),%ebx
+	vpaddd	%xmm5,%xmm11,%xmm9
+	xorl	%ebp,%edx
+	shldl	$5,%ecx,%ecx
+	vpsrldq	$4,%xmm5,%xmm8
+	addl	%esi,%ebx
+	andl	%edx,%edi
+	vpxor	%xmm2,%xmm6,%xmm6
+	xorl	%ebp,%edx
+	addl	%ecx,%ebx
+	vpxor	%xmm4,%xmm8,%xmm8
+	shrdl	$7,%ecx,%ecx
+	xorl	%ebp,%edi
+	movl	%ebx,%esi
+	addl	36(%rsp),%eax
+	vpxor	%xmm8,%xmm6,%xmm6
+	xorl	%edx,%ecx
+	shldl	$5,%ebx,%ebx
+	vmovdqa	%xmm9,16(%rsp)
+	addl	%edi,%eax
+	andl	%ecx,%esi
+	vpsrld	$31,%xmm6,%xmm8
+	xorl	%edx,%ecx
+	addl	%ebx,%eax
+	shrdl	$7,%ebx,%ebx
+	xorl	%edx,%esi
+	vpslldq	$12,%xmm6,%xmm10
+	vpaddd	%xmm6,%xmm6,%xmm6
+	movl	%eax,%edi
+	addl	40(%rsp),%ebp
+	xorl	%ecx,%ebx
+	shldl	$5,%eax,%eax
+	vpsrld	$30,%xmm10,%xmm9
+	vpor	%xmm8,%xmm6,%xmm6
+	addl	%esi,%ebp
+	andl	%ebx,%edi
+	xorl	%ecx,%ebx
+	addl	%eax,%ebp
+	vpslld	$2,%xmm10,%xmm10
+	vpxor	%xmm9,%xmm6,%xmm6
+	shrdl	$7,%eax,%eax
+	xorl	%ecx,%edi
+	movl	%ebp,%esi
+	addl	44(%rsp),%edx
+	vpxor	%xmm10,%xmm6,%xmm6
+	xorl	%ebx,%eax
+	shldl	$5,%ebp,%ebp
+	addl	%edi,%edx
+	andl	%eax,%esi
+	xorl	%ebx,%eax
+	addl	%ebp,%edx
+	shrdl	$7,%ebp,%ebp
+	xorl	%ebx,%esi
+	vpalignr	$8,%xmm3,%xmm4,%xmm7
+	movl	%edx,%edi
+	addl	48(%rsp),%ecx
+	vpaddd	%xmm6,%xmm11,%xmm9
+	xorl	%eax,%ebp
+	shldl	$5,%edx,%edx
+	vpsrldq	$4,%xmm6,%xmm8
+	addl	%esi,%ecx
+	andl	%ebp,%edi
+	vpxor	%xmm3,%xmm7,%xmm7
+	xorl	%eax,%ebp
+	addl	%edx,%ecx
+	vpxor	%xmm5,%xmm8,%xmm8
+	shrdl	$7,%edx,%edx
+	xorl	%eax,%edi
+	movl	%ecx,%esi
+	addl	52(%rsp),%ebx
+	vpxor	%xmm8,%xmm7,%xmm7
+	xorl	%ebp,%edx
+	shldl	$5,%ecx,%ecx
+	vmovdqa	%xmm9,32(%rsp)
+	addl	%edi,%ebx
+	andl	%edx,%esi
+	vpsrld	$31,%xmm7,%xmm8
+	xorl	%ebp,%edx
+	addl	%ecx,%ebx
+	shrdl	$7,%ecx,%ecx
+	xorl	%ebp,%esi
+	vpslldq	$12,%xmm7,%xmm10
+	vpaddd	%xmm7,%xmm7,%xmm7
+	movl	%ebx,%edi
+	addl	56(%rsp),%eax
+	xorl	%edx,%ecx
+	shldl	$5,%ebx,%ebx
+	vpsrld	$30,%xmm10,%xmm9
+	vpor	%xmm8,%xmm7,%xmm7
+	addl	%esi,%eax
+	andl	%ecx,%edi
+	xorl	%edx,%ecx
+	addl	%ebx,%eax
+	vpslld	$2,%xmm10,%xmm10
+	vpxor	%xmm9,%xmm7,%xmm7
+	shrdl	$7,%ebx,%ebx
+	xorl	%edx,%edi
+	movl	%eax,%esi
+	addl	60(%rsp),%ebp
+	vpxor	%xmm10,%xmm7,%xmm7
+	xorl	%ecx,%ebx
+	shldl	$5,%eax,%eax
+	addl	%edi,%ebp
+	andl	%ebx,%esi
+	xorl	%ecx,%ebx
+	addl	%eax,%ebp
+	vpalignr	$8,%xmm6,%xmm7,%xmm8
+	vpxor	%xmm4,%xmm0,%xmm0
+	shrdl	$7,%eax,%eax
+	xorl	%ecx,%esi
+	movl	%ebp,%edi
+	addl	0(%rsp),%edx
+	vpxor	%xmm1,%xmm0,%xmm0
+	xorl	%ebx,%eax
+	shldl	$5,%ebp,%ebp
+	vpaddd	%xmm7,%xmm11,%xmm9
+	addl	%esi,%edx
+	andl	%eax,%edi
+	vpxor	%xmm8,%xmm0,%xmm0
+	xorl	%ebx,%eax
+	addl	%ebp,%edx
+	shrdl	$7,%ebp,%ebp
+	xorl	%ebx,%edi
+	vpsrld	$30,%xmm0,%xmm8
+	vmovdqa	%xmm9,48(%rsp)
+	movl	%edx,%esi
+	addl	4(%rsp),%ecx
+	xorl	%eax,%ebp
+	shldl	$5,%edx,%edx
+	vpslld	$2,%xmm0,%xmm0
+	addl	%edi,%ecx
+	andl	%ebp,%esi
+	xorl	%eax,%ebp
+	addl	%edx,%ecx
+	shrdl	$7,%edx,%edx
+	xorl	%eax,%esi
+	movl	%ecx,%edi
+	addl	8(%rsp),%ebx
+	vpor	%xmm8,%xmm0,%xmm0
+	xorl	%ebp,%edx
+	shldl	$5,%ecx,%ecx
+	addl	%esi,%ebx
+	andl	%edx,%edi
+	xorl	%ebp,%edx
+	addl	%ecx,%ebx
+	addl	12(%rsp),%eax
+	xorl	%ebp,%edi
+	movl	%ebx,%esi
+	shldl	$5,%ebx,%ebx
+	addl	%edi,%eax
+	xorl	%edx,%esi
+	shrdl	$7,%ecx,%ecx
+	addl	%ebx,%eax
+	vpalignr	$8,%xmm7,%xmm0,%xmm8
+	vpxor	%xmm5,%xmm1,%xmm1
+	addl	16(%rsp),%ebp
+	xorl	%ecx,%esi
+	movl	%eax,%edi
+	shldl	$5,%eax,%eax
+	vpxor	%xmm2,%xmm1,%xmm1
+	addl	%esi,%ebp
+	xorl	%ecx,%edi
+	vpaddd	%xmm0,%xmm11,%xmm9
+	shrdl	$7,%ebx,%ebx
+	addl	%eax,%ebp
+	vpxor	%xmm8,%xmm1,%xmm1
+	addl	20(%rsp),%edx
+	xorl	%ebx,%edi
+	movl	%ebp,%esi
+	shldl	$5,%ebp,%ebp
+	vpsrld	$30,%xmm1,%xmm8
+	vmovdqa	%xmm9,0(%rsp)
+	addl	%edi,%edx
+	xorl	%ebx,%esi
+	shrdl	$7,%eax,%eax
+	addl	%ebp,%edx
+	vpslld	$2,%xmm1,%xmm1
+	addl	24(%rsp),%ecx
+	xorl	%eax,%esi
+	movl	%edx,%edi
+	shldl	$5,%edx,%edx
+	addl	%esi,%ecx
+	xorl	%eax,%edi
+	shrdl	$7,%ebp,%ebp
+	addl	%edx,%ecx
+	vpor	%xmm8,%xmm1,%xmm1
+	addl	28(%rsp),%ebx
+	xorl	%ebp,%edi
+	movl	%ecx,%esi
+	shldl	$5,%ecx,%ecx
+	addl	%edi,%ebx
+	xorl	%ebp,%esi
+	shrdl	$7,%edx,%edx
+	addl	%ecx,%ebx
+	vpalignr	$8,%xmm0,%xmm1,%xmm8
+	vpxor	%xmm6,%xmm2,%xmm2
+	addl	32(%rsp),%eax
+	xorl	%edx,%esi
+	movl	%ebx,%edi
+	shldl	$5,%ebx,%ebx
+	vpxor	%xmm3,%xmm2,%xmm2
+	addl	%esi,%eax
+	xorl	%edx,%edi
+	vpaddd	%xmm1,%xmm11,%xmm9
+	vmovdqa	0(%r11),%xmm11
+	shrdl	$7,%ecx,%ecx
+	addl	%ebx,%eax
+	vpxor	%xmm8,%xmm2,%xmm2
+	addl	36(%rsp),%ebp
+	xorl	%ecx,%edi
+	movl	%eax,%esi
+	shldl	$5,%eax,%eax
+	vpsrld	$30,%xmm2,%xmm8
+	vmovdqa	%xmm9,16(%rsp)
+	addl	%edi,%ebp
+	xorl	%ecx,%esi
+	shrdl	$7,%ebx,%ebx
+	addl	%eax,%ebp
+	vpslld	$2,%xmm2,%xmm2
+	addl	40(%rsp),%edx
+	xorl	%ebx,%esi
+	movl	%ebp,%edi
+	shldl	$5,%ebp,%ebp
+	addl	%esi,%edx
+	xorl	%ebx,%edi
+	shrdl	$7,%eax,%eax
+	addl	%ebp,%edx
+	vpor	%xmm8,%xmm2,%xmm2
+	addl	44(%rsp),%ecx
+	xorl	%eax,%edi
+	movl	%edx,%esi
+	shldl	$5,%edx,%edx
+	addl	%edi,%ecx
+	xorl	%eax,%esi
+	shrdl	$7,%ebp,%ebp
+	addl	%edx,%ecx
+	vpalignr	$8,%xmm1,%xmm2,%xmm8
+	vpxor	%xmm7,%xmm3,%xmm3
+	addl	48(%rsp),%ebx
+	xorl	%ebp,%esi
+	movl	%ecx,%edi
+	shldl	$5,%ecx,%ecx
+	vpxor	%xmm4,%xmm3,%xmm3
+	addl	%esi,%ebx
+	xorl	%ebp,%edi
+	vpaddd	%xmm2,%xmm11,%xmm9
+	shrdl	$7,%edx,%edx
+	addl	%ecx,%ebx
+	vpxor	%xmm8,%xmm3,%xmm3
+	addl	52(%rsp),%eax
+	xorl	%edx,%edi
+	movl	%ebx,%esi
+	shldl	$5,%ebx,%ebx
+	vpsrld	$30,%xmm3,%xmm8
+	vmovdqa	%xmm9,32(%rsp)
+	addl	%edi,%eax
+	xorl	%edx,%esi
+	shrdl	$7,%ecx,%ecx
+	addl	%ebx,%eax
+	vpslld	$2,%xmm3,%xmm3
+	addl	56(%rsp),%ebp
+	xorl	%ecx,%esi
+	movl	%eax,%edi
+	shldl	$5,%eax,%eax
+	addl	%esi,%ebp
+	xorl	%ecx,%edi
+	shrdl	$7,%ebx,%ebx
+	addl	%eax,%ebp
+	vpor	%xmm8,%xmm3,%xmm3
+	addl	60(%rsp),%edx
+	xorl	%ebx,%edi
+	movl	%ebp,%esi
+	shldl	$5,%ebp,%ebp
+	addl	%edi,%edx
+	xorl	%ebx,%esi
+	shrdl	$7,%eax,%eax
+	addl	%ebp,%edx
+	vpalignr	$8,%xmm2,%xmm3,%xmm8
+	vpxor	%xmm0,%xmm4,%xmm4
+	addl	0(%rsp),%ecx
+	xorl	%eax,%esi
+	movl	%edx,%edi
+	shldl	$5,%edx,%edx
+	vpxor	%xmm5,%xmm4,%xmm4
+	addl	%esi,%ecx
+	xorl	%eax,%edi
+	vpaddd	%xmm3,%xmm11,%xmm9
+	shrdl	$7,%ebp,%ebp
+	addl	%edx,%ecx
+	vpxor	%xmm8,%xmm4,%xmm4
+	addl	4(%rsp),%ebx
+	xorl	%ebp,%edi
+	movl	%ecx,%esi
+	shldl	$5,%ecx,%ecx
+	vpsrld	$30,%xmm4,%xmm8
+	vmovdqa	%xmm9,48(%rsp)
+	addl	%edi,%ebx
+	xorl	%ebp,%esi
+	shrdl	$7,%edx,%edx
+	addl	%ecx,%ebx
+	vpslld	$2,%xmm4,%xmm4
+	addl	8(%rsp),%eax
+	xorl	%edx,%esi
+	movl	%ebx,%edi
+	shldl	$5,%ebx,%ebx
+	addl	%esi,%eax
+	xorl	%edx,%edi
+	shrdl	$7,%ecx,%ecx
+	addl	%ebx,%eax
+	vpor	%xmm8,%xmm4,%xmm4
+	addl	12(%rsp),%ebp
+	xorl	%ecx,%edi
+	movl	%eax,%esi
+	shldl	$5,%eax,%eax
+	addl	%edi,%ebp
+	xorl	%ecx,%esi
+	shrdl	$7,%ebx,%ebx
+	addl	%eax,%ebp
+	vpalignr	$8,%xmm3,%xmm4,%xmm8
+	vpxor	%xmm1,%xmm5,%xmm5
+	addl	16(%rsp),%edx
+	xorl	%ebx,%esi
+	movl	%ebp,%edi
+	shldl	$5,%ebp,%ebp
+	vpxor	%xmm6,%xmm5,%xmm5
+	addl	%esi,%edx
+	xorl	%ebx,%edi
+	vpaddd	%xmm4,%xmm11,%xmm9
+	shrdl	$7,%eax,%eax
+	addl	%ebp,%edx
+	vpxor	%xmm8,%xmm5,%xmm5
+	addl	20(%rsp),%ecx
+	xorl	%eax,%edi
+	movl	%edx,%esi
+	shldl	$5,%edx,%edx
+	vpsrld	$30,%xmm5,%xmm8
+	vmovdqa	%xmm9,0(%rsp)
+	addl	%edi,%ecx
+	xorl	%eax,%esi
+	shrdl	$7,%ebp,%ebp
+	addl	%edx,%ecx
+	vpslld	$2,%xmm5,%xmm5
+	addl	24(%rsp),%ebx
+	xorl	%ebp,%esi
+	movl	%ecx,%edi
+	shldl	$5,%ecx,%ecx
+	addl	%esi,%ebx
+	xorl	%ebp,%edi
+	shrdl	$7,%edx,%edx
+	addl	%ecx,%ebx
+	vpor	%xmm8,%xmm5,%xmm5
+	addl	28(%rsp),%eax
+	shrdl	$7,%ecx,%ecx
+	movl	%ebx,%esi
+	xorl	%edx,%edi
+	shldl	$5,%ebx,%ebx
+	addl	%edi,%eax
+	xorl	%ecx,%esi
+	xorl	%edx,%ecx
+	addl	%ebx,%eax
+	vpalignr	$8,%xmm4,%xmm5,%xmm8
+	vpxor	%xmm2,%xmm6,%xmm6
+	addl	32(%rsp),%ebp
+	andl	%ecx,%esi
+	xorl	%edx,%ecx
+	shrdl	$7,%ebx,%ebx
+	vpxor	%xmm7,%xmm6,%xmm6
+	movl	%eax,%edi
+	xorl	%ecx,%esi
+	vpaddd	%xmm5,%xmm11,%xmm9
+	shldl	$5,%eax,%eax
+	addl	%esi,%ebp
+	vpxor	%xmm8,%xmm6,%xmm6
+	xorl	%ebx,%edi
+	xorl	%ecx,%ebx
+	addl	%eax,%ebp
+	addl	36(%rsp),%edx
+	vpsrld	$30,%xmm6,%xmm8
+	vmovdqa	%xmm9,16(%rsp)
+	andl	%ebx,%edi
+	xorl	%ecx,%ebx
+	shrdl	$7,%eax,%eax
+	movl	%ebp,%esi
+	vpslld	$2,%xmm6,%xmm6
+	xorl	%ebx,%edi
+	shldl	$5,%ebp,%ebp
+	addl	%edi,%edx
+	xorl	%eax,%esi
+	xorl	%ebx,%eax
+	addl	%ebp,%edx
+	addl	40(%rsp),%ecx
+	andl	%eax,%esi
+	vpor	%xmm8,%xmm6,%xmm6
+	xorl	%ebx,%eax
+	shrdl	$7,%ebp,%ebp
+	movl	%edx,%edi
+	xorl	%eax,%esi
+	shldl	$5,%edx,%edx
+	addl	%esi,%ecx
+	xorl	%ebp,%edi
+	xorl	%eax,%ebp
+	addl	%edx,%ecx
+	addl	44(%rsp),%ebx
+	andl	%ebp,%edi
+	xorl	%eax,%ebp
+	shrdl	$7,%edx,%edx
+	movl	%ecx,%esi
+	xorl	%ebp,%edi
+	shldl	$5,%ecx,%ecx
+	addl	%edi,%ebx
+	xorl	%edx,%esi
+	xorl	%ebp,%edx
+	addl	%ecx,%ebx
+	vpalignr	$8,%xmm5,%xmm6,%xmm8
+	vpxor	%xmm3,%xmm7,%xmm7
+	addl	48(%rsp),%eax
+	andl	%edx,%esi
+	xorl	%ebp,%edx
+	shrdl	$7,%ecx,%ecx
+	vpxor	%xmm0,%xmm7,%xmm7
+	movl	%ebx,%edi
+	xorl	%edx,%esi
+	vpaddd	%xmm6,%xmm11,%xmm9
+	vmovdqa	32(%r11),%xmm11
+	shldl	$5,%ebx,%ebx
+	addl	%esi,%eax
+	vpxor	%xmm8,%xmm7,%xmm7
+	xorl	%ecx,%edi
+	xorl	%edx,%ecx
+	addl	%ebx,%eax
+	addl	52(%rsp),%ebp
+	vpsrld	$30,%xmm7,%xmm8
+	vmovdqa	%xmm9,32(%rsp)
+	andl	%ecx,%edi
+	xorl	%edx,%ecx
+	shrdl	$7,%ebx,%ebx
+	movl	%eax,%esi
+	vpslld	$2,%xmm7,%xmm7
+	xorl	%ecx,%edi
+	shldl	$5,%eax,%eax
+	addl	%edi,%ebp
+	xorl	%ebx,%esi
+	xorl	%ecx,%ebx
+	addl	%eax,%ebp
+	addl	56(%rsp),%edx
+	andl	%ebx,%esi
+	vpor	%xmm8,%xmm7,%xmm7
+	xorl	%ecx,%ebx
+	shrdl	$7,%eax,%eax
+	movl	%ebp,%edi
+	xorl	%ebx,%esi
+	shldl	$5,%ebp,%ebp
+	addl	%esi,%edx
+	xorl	%eax,%edi
+	xorl	%ebx,%eax
+	addl	%ebp,%edx
+	addl	60(%rsp),%ecx
+	andl	%eax,%edi
+	xorl	%ebx,%eax
+	shrdl	$7,%ebp,%ebp
+	movl	%edx,%esi
+	xorl	%eax,%edi
+	shldl	$5,%edx,%edx
+	addl	%edi,%ecx
+	xorl	%ebp,%esi
+	xorl	%eax,%ebp
+	addl	%edx,%ecx
+	vpalignr	$8,%xmm6,%xmm7,%xmm8
+	vpxor	%xmm4,%xmm0,%xmm0
+	addl	0(%rsp),%ebx
+	andl	%ebp,%esi
+	xorl	%eax,%ebp
+	shrdl	$7,%edx,%edx
+	vpxor	%xmm1,%xmm0,%xmm0
+	movl	%ecx,%edi
+	xorl	%ebp,%esi
+	vpaddd	%xmm7,%xmm11,%xmm9
+	shldl	$5,%ecx,%ecx
+	addl	%esi,%ebx
+	vpxor	%xmm8,%xmm0,%xmm0
+	xorl	%edx,%edi
+	xorl	%ebp,%edx
+	addl	%ecx,%ebx
+	addl	4(%rsp),%eax
+	vpsrld	$30,%xmm0,%xmm8
+	vmovdqa	%xmm9,48(%rsp)
+	andl	%edx,%edi
+	xorl	%ebp,%edx
+	shrdl	$7,%ecx,%ecx
+	movl	%ebx,%esi
+	vpslld	$2,%xmm0,%xmm0
+	xorl	%edx,%edi
+	shldl	$5,%ebx,%ebx
+	addl	%edi,%eax
+	xorl	%ecx,%esi
+	xorl	%edx,%ecx
+	addl	%ebx,%eax
+	addl	8(%rsp),%ebp
+	andl	%ecx,%esi
+	vpor	%xmm8,%xmm0,%xmm0
+	xorl	%edx,%ecx
+	shrdl	$7,%ebx,%ebx
+	movl	%eax,%edi
+	xorl	%ecx,%esi
+	shldl	$5,%eax,%eax
+	addl	%esi,%ebp
+	xorl	%ebx,%edi
+	xorl	%ecx,%ebx
+	addl	%eax,%ebp
+	addl	12(%rsp),%edx
+	andl	%ebx,%edi
+	xorl	%ecx,%ebx
+	shrdl	$7,%eax,%eax
+	movl	%ebp,%esi
+	xorl	%ebx,%edi
+	shldl	$5,%ebp,%ebp
+	addl	%edi,%edx
+	xorl	%eax,%esi
+	xorl	%ebx,%eax
+	addl	%ebp,%edx
+	vpalignr	$8,%xmm7,%xmm0,%xmm8
+	vpxor	%xmm5,%xmm1,%xmm1
+	addl	16(%rsp),%ecx
+	andl	%eax,%esi
+	xorl	%ebx,%eax
+	shrdl	$7,%ebp,%ebp
+	vpxor	%xmm2,%xmm1,%xmm1
+	movl	%edx,%edi
+	xorl	%eax,%esi
+	vpaddd	%xmm0,%xmm11,%xmm9
+	shldl	$5,%edx,%edx
+	addl	%esi,%ecx
+	vpxor	%xmm8,%xmm1,%xmm1
+	xorl	%ebp,%edi
+	xorl	%eax,%ebp
+	addl	%edx,%ecx
+	addl	20(%rsp),%ebx
+	vpsrld	$30,%xmm1,%xmm8
+	vmovdqa	%xmm9,0(%rsp)
+	andl	%ebp,%edi
+	xorl	%eax,%ebp
+	shrdl	$7,%edx,%edx
+	movl	%ecx,%esi
+	vpslld	$2,%xmm1,%xmm1
+	xorl	%ebp,%edi
+	shldl	$5,%ecx,%ecx
+	addl	%edi,%ebx
+	xorl	%edx,%esi
+	xorl	%ebp,%edx
+	addl	%ecx,%ebx
+	addl	24(%rsp),%eax
+	andl	%edx,%esi
+	vpor	%xmm8,%xmm1,%xmm1
+	xorl	%ebp,%edx
+	shrdl	$7,%ecx,%ecx
+	movl	%ebx,%edi
+	xorl	%edx,%esi
+	shldl	$5,%ebx,%ebx
+	addl	%esi,%eax
+	xorl	%ecx,%edi
+	xorl	%edx,%ecx
+	addl	%ebx,%eax
+	addl	28(%rsp),%ebp
+	andl	%ecx,%edi
+	xorl	%edx,%ecx
+	shrdl	$7,%ebx,%ebx
+	movl	%eax,%esi
+	xorl	%ecx,%edi
+	shldl	$5,%eax,%eax
+	addl	%edi,%ebp
+	xorl	%ebx,%esi
+	xorl	%ecx,%ebx
+	addl	%eax,%ebp
+	vpalignr	$8,%xmm0,%xmm1,%xmm8
+	vpxor	%xmm6,%xmm2,%xmm2
+	addl	32(%rsp),%edx
+	andl	%ebx,%esi
+	xorl	%ecx,%ebx
+	shrdl	$7,%eax,%eax
+	vpxor	%xmm3,%xmm2,%xmm2
+	movl	%ebp,%edi
+	xorl	%ebx,%esi
+	vpaddd	%xmm1,%xmm11,%xmm9
+	shldl	$5,%ebp,%ebp
+	addl	%esi,%edx
+	vpxor	%xmm8,%xmm2,%xmm2
+	xorl	%eax,%edi
+	xorl	%ebx,%eax
+	addl	%ebp,%edx
+	addl	36(%rsp),%ecx
+	vpsrld	$30,%xmm2,%xmm8
+	vmovdqa	%xmm9,16(%rsp)
+	andl	%eax,%edi
+	xorl	%ebx,%eax
+	shrdl	$7,%ebp,%ebp
+	movl	%edx,%esi
+	vpslld	$2,%xmm2,%xmm2
+	xorl	%eax,%edi
+	shldl	$5,%edx,%edx
+	addl	%edi,%ecx
+	xorl	%ebp,%esi
+	xorl	%eax,%ebp
+	addl	%edx,%ecx
+	addl	40(%rsp),%ebx
+	andl	%ebp,%esi
+	vpor	%xmm8,%xmm2,%xmm2
+	xorl	%eax,%ebp
+	shrdl	$7,%edx,%edx
+	movl	%ecx,%edi
+	xorl	%ebp,%esi
+	shldl	$5,%ecx,%ecx
+	addl	%esi,%ebx
+	xorl	%edx,%edi
+	xorl	%ebp,%edx
+	addl	%ecx,%ebx
+	addl	44(%rsp),%eax
+	andl	%edx,%edi
+	xorl	%ebp,%edx
+	shrdl	$7,%ecx,%ecx
+	movl	%ebx,%esi
+	xorl	%edx,%edi
+	shldl	$5,%ebx,%ebx
+	addl	%edi,%eax
+	xorl	%edx,%esi
+	addl	%ebx,%eax
+	vpalignr	$8,%xmm1,%xmm2,%xmm8
+	vpxor	%xmm7,%xmm3,%xmm3
+	addl	48(%rsp),%ebp
+	xorl	%ecx,%esi
+	movl	%eax,%edi
+	shldl	$5,%eax,%eax
+	vpxor	%xmm4,%xmm3,%xmm3
+	addl	%esi,%ebp
+	xorl	%ecx,%edi
+	vpaddd	%xmm2,%xmm11,%xmm9
+	shrdl	$7,%ebx,%ebx
+	addl	%eax,%ebp
+	vpxor	%xmm8,%xmm3,%xmm3
+	addl	52(%rsp),%edx
+	xorl	%ebx,%edi
+	movl	%ebp,%esi
+	shldl	$5,%ebp,%ebp
+	vpsrld	$30,%xmm3,%xmm8
+	vmovdqa	%xmm9,32(%rsp)
+	addl	%edi,%edx
+	xorl	%ebx,%esi
+	shrdl	$7,%eax,%eax
+	addl	%ebp,%edx
+	vpslld	$2,%xmm3,%xmm3
+	addl	56(%rsp),%ecx
+	xorl	%eax,%esi
+	movl	%edx,%edi
+	shldl	$5,%edx,%edx
+	addl	%esi,%ecx
+	xorl	%eax,%edi
+	shrdl	$7,%ebp,%ebp
+	addl	%edx,%ecx
+	vpor	%xmm8,%xmm3,%xmm3
+	addl	60(%rsp),%ebx
+	xorl	%ebp,%edi
+	movl	%ecx,%esi
+	shldl	$5,%ecx,%ecx
+	addl	%edi,%ebx
+	xorl	%ebp,%esi
+	shrdl	$7,%edx,%edx
+	addl	%ecx,%ebx
+	addl	0(%rsp),%eax
+	vpaddd	%xmm3,%xmm11,%xmm9
+	xorl	%edx,%esi
+	movl	%ebx,%edi
+	shldl	$5,%ebx,%ebx
+	addl	%esi,%eax
+	vmovdqa	%xmm9,48(%rsp)
+	xorl	%edx,%edi
+	shrdl	$7,%ecx,%ecx
+	addl	%ebx,%eax
+	addl	4(%rsp),%ebp
+	xorl	%ecx,%edi
+	movl	%eax,%esi
+	shldl	$5,%eax,%eax
+	addl	%edi,%ebp
+	xorl	%ecx,%esi
+	shrdl	$7,%ebx,%ebx
+	addl	%eax,%ebp
+	addl	8(%rsp),%edx
+	xorl	%ebx,%esi
+	movl	%ebp,%edi
+	shldl	$5,%ebp,%ebp
+	addl	%esi,%edx
+	xorl	%ebx,%edi
+	shrdl	$7,%eax,%eax
+	addl	%ebp,%edx
+	addl	12(%rsp),%ecx
+	xorl	%eax,%edi
+	movl	%edx,%esi
+	shldl	$5,%edx,%edx
+	addl	%edi,%ecx
+	xorl	%eax,%esi
+	shrdl	$7,%ebp,%ebp
+	addl	%edx,%ecx
+	cmpq	%r10,%r9
+	je	L$done_avx
+	vmovdqa	64(%r11),%xmm6
+	vmovdqa	-64(%r11),%xmm11
+	vmovdqu	0(%r9),%xmm0
+	vmovdqu	16(%r9),%xmm1
+	vmovdqu	32(%r9),%xmm2
+	vmovdqu	48(%r9),%xmm3
+	vpshufb	%xmm6,%xmm0,%xmm0
+	addq	$64,%r9
+	addl	16(%rsp),%ebx
+	xorl	%ebp,%esi
+	vpshufb	%xmm6,%xmm1,%xmm1
+	movl	%ecx,%edi
+	shldl	$5,%ecx,%ecx
+	vpaddd	%xmm11,%xmm0,%xmm4
+	addl	%esi,%ebx
+	xorl	%ebp,%edi
+	shrdl	$7,%edx,%edx
+	addl	%ecx,%ebx
+	vmovdqa	%xmm4,0(%rsp)
+	addl	20(%rsp),%eax
+	xorl	%edx,%edi
+	movl	%ebx,%esi
+	shldl	$5,%ebx,%ebx
+	addl	%edi,%eax
+	xorl	%edx,%esi
+	shrdl	$7,%ecx,%ecx
+	addl	%ebx,%eax
+	addl	24(%rsp),%ebp
+	xorl	%ecx,%esi
+	movl	%eax,%edi
+	shldl	$5,%eax,%eax
+	addl	%esi,%ebp
+	xorl	%ecx,%edi
+	shrdl	$7,%ebx,%ebx
+	addl	%eax,%ebp
+	addl	28(%rsp),%edx
+	xorl	%ebx,%edi
+	movl	%ebp,%esi
+	shldl	$5,%ebp,%ebp
+	addl	%edi,%edx
+	xorl	%ebx,%esi
+	shrdl	$7,%eax,%eax
+	addl	%ebp,%edx
+	addl	32(%rsp),%ecx
+	xorl	%eax,%esi
+	vpshufb	%xmm6,%xmm2,%xmm2
+	movl	%edx,%edi
+	shldl	$5,%edx,%edx
+	vpaddd	%xmm11,%xmm1,%xmm5
+	addl	%esi,%ecx
+	xorl	%eax,%edi
+	shrdl	$7,%ebp,%ebp
+	addl	%edx,%ecx
+	vmovdqa	%xmm5,16(%rsp)
+	addl	36(%rsp),%ebx
+	xorl	%ebp,%edi
+	movl	%ecx,%esi
+	shldl	$5,%ecx,%ecx
+	addl	%edi,%ebx
+	xorl	%ebp,%esi
+	shrdl	$7,%edx,%edx
+	addl	%ecx,%ebx
+	addl	40(%rsp),%eax
+	xorl	%edx,%esi
+	movl	%ebx,%edi
+	shldl	$5,%ebx,%ebx
+	addl	%esi,%eax
+	xorl	%edx,%edi
+	shrdl	$7,%ecx,%ecx
+	addl	%ebx,%eax
+	addl	44(%rsp),%ebp
+	xorl	%ecx,%edi
+	movl	%eax,%esi
+	shldl	$5,%eax,%eax
+	addl	%edi,%ebp
+	xorl	%ecx,%esi
+	shrdl	$7,%ebx,%ebx
+	addl	%eax,%ebp
+	addl	48(%rsp),%edx
+	xorl	%ebx,%esi
+	vpshufb	%xmm6,%xmm3,%xmm3
+	movl	%ebp,%edi
+	shldl	$5,%ebp,%ebp
+	vpaddd	%xmm11,%xmm2,%xmm6
+	addl	%esi,%edx
+	xorl	%ebx,%edi
+	shrdl	$7,%eax,%eax
+	addl	%ebp,%edx
+	vmovdqa	%xmm6,32(%rsp)
+	addl	52(%rsp),%ecx
+	xorl	%eax,%edi
+	movl	%edx,%esi
+	shldl	$5,%edx,%edx
+	addl	%edi,%ecx
+	xorl	%eax,%esi
+	shrdl	$7,%ebp,%ebp
+	addl	%edx,%ecx
+	addl	56(%rsp),%ebx
+	xorl	%ebp,%esi
+	movl	%ecx,%edi
+	shldl	$5,%ecx,%ecx
+	addl	%esi,%ebx
+	xorl	%ebp,%edi
+	shrdl	$7,%edx,%edx
+	addl	%ecx,%ebx
+	addl	60(%rsp),%eax
+	xorl	%edx,%edi
+	movl	%ebx,%esi
+	shldl	$5,%ebx,%ebx
+	addl	%edi,%eax
+	shrdl	$7,%ecx,%ecx
+	addl	%ebx,%eax
+	addl	0(%r8),%eax
+	addl	4(%r8),%esi
+	addl	8(%r8),%ecx
+	addl	12(%r8),%edx
+	movl	%eax,0(%r8)
+	addl	16(%r8),%ebp
+	movl	%esi,4(%r8)
+	movl	%esi,%ebx
+	movl	%ecx,8(%r8)
+	movl	%ecx,%edi
+	movl	%edx,12(%r8)
+	xorl	%edx,%edi
+	movl	%ebp,16(%r8)
+	andl	%edi,%esi
+	jmp	L$oop_avx
+
+.p2align	4
+L$done_avx:
+	addl	16(%rsp),%ebx
+	xorl	%ebp,%esi
+	movl	%ecx,%edi
+	shldl	$5,%ecx,%ecx
+	addl	%esi,%ebx
+	xorl	%ebp,%edi
+	shrdl	$7,%edx,%edx
+	addl	%ecx,%ebx
+	addl	20(%rsp),%eax
+	xorl	%edx,%edi
+	movl	%ebx,%esi
+	shldl	$5,%ebx,%ebx
+	addl	%edi,%eax
+	xorl	%edx,%esi
+	shrdl	$7,%ecx,%ecx
+	addl	%ebx,%eax
+	addl	24(%rsp),%ebp
+	xorl	%ecx,%esi
+	movl	%eax,%edi
+	shldl	$5,%eax,%eax
+	addl	%esi,%ebp
+	xorl	%ecx,%edi
+	shrdl	$7,%ebx,%ebx
+	addl	%eax,%ebp
+	addl	28(%rsp),%edx
+	xorl	%ebx,%edi
+	movl	%ebp,%esi
+	shldl	$5,%ebp,%ebp
+	addl	%edi,%edx
+	xorl	%ebx,%esi
+	shrdl	$7,%eax,%eax
+	addl	%ebp,%edx
+	addl	32(%rsp),%ecx
+	xorl	%eax,%esi
+	movl	%edx,%edi
+	shldl	$5,%edx,%edx
+	addl	%esi,%ecx
+	xorl	%eax,%edi
+	shrdl	$7,%ebp,%ebp
+	addl	%edx,%ecx
+	addl	36(%rsp),%ebx
+	xorl	%ebp,%edi
+	movl	%ecx,%esi
+	shldl	$5,%ecx,%ecx
+	addl	%edi,%ebx
+	xorl	%ebp,%esi
+	shrdl	$7,%edx,%edx
+	addl	%ecx,%ebx
+	addl	40(%rsp),%eax
+	xorl	%edx,%esi
+	movl	%ebx,%edi
+	shldl	$5,%ebx,%ebx
+	addl	%esi,%eax
+	xorl	%edx,%edi
+	shrdl	$7,%ecx,%ecx
+	addl	%ebx,%eax
+	addl	44(%rsp),%ebp
+	xorl	%ecx,%edi
+	movl	%eax,%esi
+	shldl	$5,%eax,%eax
+	addl	%edi,%ebp
+	xorl	%ecx,%esi
+	shrdl	$7,%ebx,%ebx
+	addl	%eax,%ebp
+	addl	48(%rsp),%edx
+	xorl	%ebx,%esi
+	movl	%ebp,%edi
+	shldl	$5,%ebp,%ebp
+	addl	%esi,%edx
+	xorl	%ebx,%edi
+	shrdl	$7,%eax,%eax
+	addl	%ebp,%edx
+	addl	52(%rsp),%ecx
+	xorl	%eax,%edi
+	movl	%edx,%esi
+	shldl	$5,%edx,%edx
+	addl	%edi,%ecx
+	xorl	%eax,%esi
+	shrdl	$7,%ebp,%ebp
+	addl	%edx,%ecx
+	addl	56(%rsp),%ebx
+	xorl	%ebp,%esi
+	movl	%ecx,%edi
+	shldl	$5,%ecx,%ecx
+	addl	%esi,%ebx
+	xorl	%ebp,%edi
+	shrdl	$7,%edx,%edx
+	addl	%ecx,%ebx
+	addl	60(%rsp),%eax
+	xorl	%edx,%edi
+	movl	%ebx,%esi
+	shldl	$5,%ebx,%ebx
+	addl	%edi,%eax
+	shrdl	$7,%ecx,%ecx
+	addl	%ebx,%eax
+	vzeroupper
+
+	addl	0(%r8),%eax
+	addl	4(%r8),%esi
+	addl	8(%r8),%ecx
+	movl	%eax,0(%r8)
+	addl	12(%r8),%edx
+	movl	%esi,4(%r8)
+	addl	16(%r8),%ebp
+	movl	%ecx,8(%r8)
+	movl	%edx,12(%r8)
+	movl	%ebp,16(%r8)
+	leaq	(%r14),%rsi
+	movq	-40(%rsi),%r14
+	movq	-32(%rsi),%r13
+	movq	-24(%rsi),%r12
+	movq	-16(%rsi),%rbp
+	movq	-8(%rsi),%rbx
+	leaq	(%rsi),%rsp
+L$epilogue_avx:
+	.byte	0xf3,0xc3
+
 .p2align	6
 K_XX_XX:
 .long	0x5a827999,0x5a827999,0x5a827999,0x5a827999

third_party/boringssl/mac-x86_64/crypto/sha/sha256-x86_64.S

diff --git a/third_party/boringssl/mac-x86_64/crypto/sha/sha256-x86_64.S b/third_party/boringssl/mac-x86_64/crypto/sha/sha256-x86_64.S
index da02d4c..0146ff5 100644
--- a/third_party/boringssl/mac-x86_64/crypto/sha/sha256-x86_64.S
+++ b/third_party/boringssl/mac-x86_64/crypto/sha/sha256-x86_64.S
@@ -11,6 +11,11 @@
 	movl	0(%r11),%r9d
 	movl	4(%r11),%r10d
 	movl	8(%r11),%r11d
+	andl	$1073741824,%r9d
+	andl	$268435968,%r10d
+	orl	%r9d,%r10d
+	cmpl	$1342177792,%r10d
+	je	L$avx_shortcut
 	testl	$512,%r10d
 	jnz	L$ssse3_shortcut
 	pushq	%rbx
@@ -2840,4 +2845,1061 @@
 L$epilogue_ssse3:
 	.byte	0xf3,0xc3
 
+
+.p2align	6
+sha256_block_data_order_avx:
+L$avx_shortcut:
+	pushq	%rbx
+	pushq	%rbp
+	pushq	%r12
+	pushq	%r13
+	pushq	%r14
+	pushq	%r15
+	movq	%rsp,%r11
+	shlq	$4,%rdx
+	subq	$96,%rsp
+	leaq	(%rsi,%rdx,4),%rdx
+	andq	$-64,%rsp
+	movq	%rdi,64+0(%rsp)
+	movq	%rsi,64+8(%rsp)
+	movq	%rdx,64+16(%rsp)
+	movq	%r11,64+24(%rsp)
+L$prologue_avx:
+
+	vzeroupper
+	movl	0(%rdi),%eax
+	movl	4(%rdi),%ebx
+	movl	8(%rdi),%ecx
+	movl	12(%rdi),%edx
+	movl	16(%rdi),%r8d
+	movl	20(%rdi),%r9d
+	movl	24(%rdi),%r10d
+	movl	28(%rdi),%r11d
+	vmovdqa	K256+512+32(%rip),%xmm8
+	vmovdqa	K256+512+64(%rip),%xmm9
+	jmp	L$loop_avx
+.p2align	4
+L$loop_avx:
+	vmovdqa	K256+512(%rip),%xmm7
+	vmovdqu	0(%rsi),%xmm0
+	vmovdqu	16(%rsi),%xmm1
+	vmovdqu	32(%rsi),%xmm2
+	vmovdqu	48(%rsi),%xmm3
+	vpshufb	%xmm7,%xmm0,%xmm0
+	leaq	K256(%rip),%rbp
+	vpshufb	%xmm7,%xmm1,%xmm1
+	vpshufb	%xmm7,%xmm2,%xmm2
+	vpaddd	0(%rbp),%xmm0,%xmm4
+	vpshufb	%xmm7,%xmm3,%xmm3
+	vpaddd	32(%rbp),%xmm1,%xmm5
+	vpaddd	64(%rbp),%xmm2,%xmm6
+	vpaddd	96(%rbp),%xmm3,%xmm7
+	vmovdqa	%xmm4,0(%rsp)
+	movl	%eax,%r14d
+	vmovdqa	%xmm5,16(%rsp)
+	movl	%ebx,%edi
+	vmovdqa	%xmm6,32(%rsp)
+	xorl	%ecx,%edi
+	vmovdqa	%xmm7,48(%rsp)
+	movl	%r8d,%r13d
+	jmp	L$avx_00_47
+
+.p2align	4
+L$avx_00_47:
+	subq	$-128,%rbp
+	vpalignr	$4,%xmm0,%xmm1,%xmm4
+	shrdl	$14,%r13d,%r13d
+	movl	%r14d,%eax
+	movl	%r9d,%r12d
+	vpalignr	$4,%xmm2,%xmm3,%xmm7
+	shrdl	$9,%r14d,%r14d
+	xorl	%r8d,%r13d
+	xorl	%r10d,%r12d
+	vpsrld	$7,%xmm4,%xmm6
+	shrdl	$5,%r13d,%r13d
+	xorl	%eax,%r14d
+	andl	%r8d,%r12d
+	vpaddd	%xmm7,%xmm0,%xmm0
+	xorl	%r8d,%r13d
+	addl	0(%rsp),%r11d
+	movl	%eax,%r15d
+	vpsrld	$3,%xmm4,%xmm7
+	xorl	%r10d,%r12d
+	shrdl	$11,%r14d,%r14d
+	xorl	%ebx,%r15d
+	vpslld	$14,%xmm4,%xmm5
+	addl	%r12d,%r11d
+	shrdl	$6,%r13d,%r13d
+	andl	%r15d,%edi
+	vpxor	%xmm6,%xmm7,%xmm4
+	xorl	%eax,%r14d
+	addl	%r13d,%r11d
+	xorl	%ebx,%edi
+	vpshufd	$250,%xmm3,%xmm7
+	shrdl	$2,%r14d,%r14d
+	addl	%r11d,%edx
+	addl	%edi,%r11d
+	vpsrld	$11,%xmm6,%xmm6
+	movl	%edx,%r13d
+	addl	%r11d,%r14d
+	shrdl	$14,%r13d,%r13d
+	vpxor	%xmm5,%xmm4,%xmm4
+	movl	%r14d,%r11d
+	movl	%r8d,%r12d
+	shrdl	$9,%r14d,%r14d
+	vpslld	$11,%xmm5,%xmm5
+	xorl	%edx,%r13d
+	xorl	%r9d,%r12d
+	shrdl	$5,%r13d,%r13d
+	vpxor	%xmm6,%xmm4,%xmm4
+	xorl	%r11d,%r14d
+	andl	%edx,%r12d
+	xorl	%edx,%r13d
+	vpsrld	$10,%xmm7,%xmm6
+	addl	4(%rsp),%r10d
+	movl	%r11d,%edi
+	xorl	%r9d,%r12d
+	vpxor	%xmm5,%xmm4,%xmm4
+	shrdl	$11,%r14d,%r14d
+	xorl	%eax,%edi
+	addl	%r12d,%r10d
+	vpsrlq	$17,%xmm7,%xmm7
+	shrdl	$6,%r13d,%r13d
+	andl	%edi,%r15d
+	xorl	%r11d,%r14d
+	vpaddd	%xmm4,%xmm0,%xmm0
+	addl	%r13d,%r10d
+	xorl	%eax,%r15d
+	shrdl	$2,%r14d,%r14d
+	vpxor	%xmm7,%xmm6,%xmm6
+	addl	%r10d,%ecx
+	addl	%r15d,%r10d
+	movl	%ecx,%r13d
+	vpsrlq	$2,%xmm7,%xmm7
+	addl	%r10d,%r14d
+	shrdl	$14,%r13d,%r13d
+	movl	%r14d,%r10d
+	vpxor	%xmm7,%xmm6,%xmm6
+	movl	%edx,%r12d
+	shrdl	$9,%r14d,%r14d
+	xorl	%ecx,%r13d
+	vpshufb	%xmm8,%xmm6,%xmm6
+	xorl	%r8d,%r12d
+	shrdl	$5,%r13d,%r13d
+	xorl	%r10d,%r14d
+	vpaddd	%xmm6,%xmm0,%xmm0
+	andl	%ecx,%r12d
+	xorl	%ecx,%r13d
+	addl	8(%rsp),%r9d
+	vpshufd	$80,%xmm0,%xmm7
+	movl	%r10d,%r15d
+	xorl	%r8d,%r12d
+	shrdl	$11,%r14d,%r14d
+	vpsrld	$10,%xmm7,%xmm6
+	xorl	%r11d,%r15d
+	addl	%r12d,%r9d
+	shrdl	$6,%r13d,%r13d
+	vpsrlq	$17,%xmm7,%xmm7
+	andl	%r15d,%edi
+	xorl	%r10d,%r14d
+	addl	%r13d,%r9d
+	vpxor	%xmm7,%xmm6,%xmm6
+	xorl	%r11d,%edi
+	shrdl	$2,%r14d,%r14d
+	addl	%r9d,%ebx
+	vpsrlq	$2,%xmm7,%xmm7
+	addl	%edi,%r9d
+	movl	%ebx,%r13d
+	addl	%r9d,%r14d
+	vpxor	%xmm7,%xmm6,%xmm6
+	shrdl	$14,%r13d,%r13d
+	movl	%r14d,%r9d
+	movl	%ecx,%r12d
+	vpshufb	%xmm9,%xmm6,%xmm6
+	shrdl	$9,%r14d,%r14d
+	xorl	%ebx,%r13d
+	xorl	%edx,%r12d
+	vpaddd	%xmm6,%xmm0,%xmm0
+	shrdl	$5,%r13d,%r13d
+	xorl	%r9d,%r14d
+	andl	%ebx,%r12d
+	vpaddd	0(%rbp),%xmm0,%xmm6
+	xorl	%ebx,%r13d
+	addl	12(%rsp),%r8d
+	movl	%r9d,%edi
+	xorl	%edx,%r12d
+	shrdl	$11,%r14d,%r14d
+	xorl	%r10d,%edi
+	addl	%r12d,%r8d
+	shrdl	$6,%r13d,%r13d
+	andl	%edi,%r15d
+	xorl	%r9d,%r14d
+	addl	%r13d,%r8d
+	xorl	%r10d,%r15d
+	shrdl	$2,%r14d,%r14d
+	addl	%r8d,%eax
+	addl	%r15d,%r8d
+	movl	%eax,%r13d
+	addl	%r8d,%r14d
+	vmovdqa	%xmm6,0(%rsp)
+	vpalignr	$4,%xmm1,%xmm2,%xmm4
+	shrdl	$14,%r13d,%r13d
+	movl	%r14d,%r8d
+	movl	%ebx,%r12d
+	vpalignr	$4,%xmm3,%xmm0,%xmm7
+	shrdl	$9,%r14d,%r14d
+	xorl	%eax,%r13d
+	xorl	%ecx,%r12d
+	vpsrld	$7,%xmm4,%xmm6
+	shrdl	$5,%r13d,%r13d
+	xorl	%r8d,%r14d
+	andl	%eax,%r12d
+	vpaddd	%xmm7,%xmm1,%xmm1
+	xorl	%eax,%r13d
+	addl	16(%rsp),%edx
+	movl	%r8d,%r15d
+	vpsrld	$3,%xmm4,%xmm7
+	xorl	%ecx,%r12d
+	shrdl	$11,%r14d,%r14d
+	xorl	%r9d,%r15d
+	vpslld	$14,%xmm4,%xmm5
+	addl	%r12d,%edx
+	shrdl	$6,%r13d,%r13d
+	andl	%r15d,%edi
+	vpxor	%xmm6,%xmm7,%xmm4
+	xorl	%r8d,%r14d
+	addl	%r13d,%edx
+	xorl	%r9d,%edi
+	vpshufd	$250,%xmm0,%xmm7
+	shrdl	$2,%r14d,%r14d
+	addl	%edx,%r11d
+	addl	%edi,%edx
+	vpsrld	$11,%xmm6,%xmm6
+	movl	%r11d,%r13d
+	addl	%edx,%r14d
+	shrdl	$14,%r13d,%r13d
+	vpxor	%xmm5,%xmm4,%xmm4
+	movl	%r14d,%edx
+	movl	%eax,%r12d
+	shrdl	$9,%r14d,%r14d
+	vpslld	$11,%xmm5,%xmm5
+	xorl	%r11d,%r13d
+	xorl	%ebx,%r12d
+	shrdl	$5,%r13d,%r13d
+	vpxor	%xmm6,%xmm4,%xmm4
+	xorl	%edx,%r14d
+	andl	%r11d,%r12d
+	xorl	%r11d,%r13d
+	vpsrld	$10,%xmm7,%xmm6
+	addl	20(%rsp),%ecx
+	movl	%edx,%edi
+	xorl	%ebx,%r12d
+	vpxor	%xmm5,%xmm4,%xmm4
+	shrdl	$11,%r14d,%r14d
+	xorl	%r8d,%edi
+	addl	%r12d,%ecx
+	vpsrlq	$17,%xmm7,%xmm7
+	shrdl	$6,%r13d,%r13d
+	andl	%edi,%r15d
+	xorl	%edx,%r14d
+	vpaddd	%xmm4,%xmm1,%xmm1
+	addl	%r13d,%ecx
+	xorl	%r8d,%r15d
+	shrdl	$2,%r14d,%r14d
+	vpxor	%xmm7,%xmm6,%xmm6
+	addl	%ecx,%r10d
+	addl	%r15d,%ecx
+	movl	%r10d,%r13d
+	vpsrlq	$2,%xmm7,%xmm7
+	addl	%ecx,%r14d
+	shrdl	$14,%r13d,%r13d
+	movl	%r14d,%ecx
+	vpxor	%xmm7,%xmm6,%xmm6
+	movl	%r11d,%r12d
+	shrdl	$9,%r14d,%r14d
+	xorl	%r10d,%r13d
+	vpshufb	%xmm8,%xmm6,%xmm6
+	xorl	%eax,%r12d
+	shrdl	$5,%r13d,%r13d
+	xorl	%ecx,%r14d
+	vpaddd	%xmm6,%xmm1,%xmm1
+	andl	%r10d,%r12d
+	xorl	%r10d,%r13d
+	addl	24(%rsp),%ebx
+	vpshufd	$80,%xmm1,%xmm7
+	movl	%ecx,%r15d
+	xorl	%eax,%r12d
+	shrdl	$11,%r14d,%r14d
+	vpsrld	$10,%xmm7,%xmm6
+	xorl	%edx,%r15d
+	addl	%r12d,%ebx
+	shrdl	$6,%r13d,%r13d
+	vpsrlq	$17,%xmm7,%xmm7
+	andl	%r15d,%edi
+	xorl	%ecx,%r14d
+	addl	%r13d,%ebx
+	vpxor	%xmm7,%xmm6,%xmm6
+	xorl	%edx,%edi
+	shrdl	$2,%r14d,%r14d
+	addl	%ebx,%r9d
+	vpsrlq	$2,%xmm7,%xmm7
+	addl	%edi,%ebx
+	movl	%r9d,%r13d
+	addl	%ebx,%r14d
+	vpxor	%xmm7,%xmm6,%xmm6
+	shrdl	$14,%r13d,%r13d
+	movl	%r14d,%ebx
+	movl	%r10d,%r12d
+	vpshufb	%xmm9,%xmm6,%xmm6
+	shrdl	$9,%r14d,%r14d
+	xorl	%r9d,%r13d
+	xorl	%r11d,%r12d
+	vpaddd	%xmm6,%xmm1,%xmm1
+	shrdl	$5,%r13d,%r13d
+	xorl	%ebx,%r14d
+	andl	%r9d,%r12d
+	vpaddd	32(%rbp),%xmm1,%xmm6
+	xorl	%r9d,%r13d
+	addl	28(%rsp),%eax
+	movl	%ebx,%edi
+	xorl	%r11d,%r12d
+	shrdl	$11,%r14d,%r14d
+	xorl	%ecx,%edi
+	addl	%r12d,%eax
+	shrdl	$6,%r13d,%r13d
+	andl	%edi,%r15d
+	xorl	%ebx,%r14d
+	addl	%r13d,%eax
+	xorl	%ecx,%r15d
+	shrdl	$2,%r14d,%r14d
+	addl	%eax,%r8d
+	addl	%r15d,%eax
+	movl	%r8d,%r13d
+	addl	%eax,%r14d
+	vmovdqa	%xmm6,16(%rsp)
+	vpalignr	$4,%xmm2,%xmm3,%xmm4
+	shrdl	$14,%r13d,%r13d
+	movl	%r14d,%eax
+	movl	%r9d,%r12d
+	vpalignr	$4,%xmm0,%xmm1,%xmm7
+	shrdl	$9,%r14d,%r14d
+	xorl	%r8d,%r13d
+	xorl	%r10d,%r12d
+	vpsrld	$7,%xmm4,%xmm6
+	shrdl	$5,%r13d,%r13d
+	xorl	%eax,%r14d
+	andl	%r8d,%r12d
+	vpaddd	%xmm7,%xmm2,%xmm2
+	xorl	%r8d,%r13d
+	addl	32(%rsp),%r11d
+	movl	%eax,%r15d
+	vpsrld	$3,%xmm4,%xmm7
+	xorl	%r10d,%r12d
+	shrdl	$11,%r14d,%r14d
+	xorl	%ebx,%r15d
+	vpslld	$14,%xmm4,%xmm5
+	addl	%r12d,%r11d
+	shrdl	$6,%r13d,%r13d
+	andl	%r15d,%edi
+	vpxor	%xmm6,%xmm7,%xmm4
+	xorl	%eax,%r14d
+	addl	%r13d,%r11d
+	xorl	%ebx,%edi
+	vpshufd	$250,%xmm1,%xmm7
+	shrdl	$2,%r14d,%r14d
+	addl	%r11d,%edx
+	addl	%edi,%r11d
+	vpsrld	$11,%xmm6,%xmm6
+	movl	%edx,%r13d
+	addl	%r11d,%r14d
+	shrdl	$14,%r13d,%r13d
+	vpxor	%xmm5,%xmm4,%xmm4
+	movl	%r14d,%r11d
+	movl	%r8d,%r12d
+	shrdl	$9,%r14d,%r14d
+	vpslld	$11,%xmm5,%xmm5
+	xorl	%edx,%r13d
+	xorl	%r9d,%r12d
+	shrdl	$5,%r13d,%r13d
+	vpxor	%xmm6,%xmm4,%xmm4
+	xorl	%r11d,%r14d
+	andl	%edx,%r12d
+	xorl	%edx,%r13d
+	vpsrld	$10,%xmm7,%xmm6
+	addl	36(%rsp),%r10d
+	movl	%r11d,%edi
+	xorl	%r9d,%r12d
+	vpxor	%xmm5,%xmm4,%xmm4
+	shrdl	$11,%r14d,%r14d
+	xorl	%eax,%edi
+	addl	%r12d,%r10d
+	vpsrlq	$17,%xmm7,%xmm7
+	shrdl	$6,%r13d,%r13d
+	andl	%edi,%r15d
+	xorl	%r11d,%r14d
+	vpaddd	%xmm4,%xmm2,%xmm2
+	addl	%r13d,%r10d
+	xorl	%eax,%r15d
+	shrdl	$2,%r14d,%r14d
+	vpxor	%xmm7,%xmm6,%xmm6
+	addl	%r10d,%ecx
+	addl	%r15d,%r10d
+	movl	%ecx,%r13d
+	vpsrlq	$2,%xmm7,%xmm7
+	addl	%r10d,%r14d
+	shrdl	$14,%r13d,%r13d
+	movl	%r14d,%r10d
+	vpxor	%xmm7,%xmm6,%xmm6
+	movl	%edx,%r12d
+	shrdl	$9,%r14d,%r14d
+	xorl	%ecx,%r13d
+	vpshufb	%xmm8,%xmm6,%xmm6
+	xorl	%r8d,%r12d
+	shrdl	$5,%r13d,%r13d
+	xorl	%r10d,%r14d
+	vpaddd	%xmm6,%xmm2,%xmm2
+	andl	%ecx,%r12d
+	xorl	%ecx,%r13d
+	addl	40(%rsp),%r9d
+	vpshufd	$80,%xmm2,%xmm7
+	movl	%r10d,%r15d
+	xorl	%r8d,%r12d
+	shrdl	$11,%r14d,%r14d
+	vpsrld	$10,%xmm7,%xmm6
+	xorl	%r11d,%r15d
+	addl	%r12d,%r9d
+	shrdl	$6,%r13d,%r13d
+	vpsrlq	$17,%xmm7,%xmm7
+	andl	%r15d,%edi
+	xorl	%r10d,%r14d
+	addl	%r13d,%r9d
+	vpxor	%xmm7,%xmm6,%xmm6
+	xorl	%r11d,%edi
+	shrdl	$2,%r14d,%r14d
+	addl	%r9d,%ebx
+	vpsrlq	$2,%xmm7,%xmm7
+	addl	%edi,%r9d
+	movl	%ebx,%r13d
+	addl	%r9d,%r14d
+	vpxor	%xmm7,%xmm6,%xmm6
+	shrdl	$14,%r13d,%r13d
+	movl	%r14d,%r9d
+	movl	%ecx,%r12d
+	vpshufb	%xmm9,%xmm6,%xmm6
+	shrdl	$9,%r14d,%r14d
+	xorl	%ebx,%r13d
+	xorl	%edx,%r12d
+	vpaddd	%xmm6,%xmm2,%xmm2
+	shrdl	$5,%r13d,%r13d
+	xorl	%r9d,%r14d
+	andl	%ebx,%r12d
+	vpaddd	64(%rbp),%xmm2,%xmm6
+	xorl	%ebx,%r13d
+	addl	44(%rsp),%r8d
+	movl	%r9d,%edi
+	xorl	%edx,%r12d
+	shrdl	$11,%r14d,%r14d
+	xorl	%r10d,%edi
+	addl	%r12d,%r8d
+	shrdl	$6,%r13d,%r13d
+	andl	%edi,%r15d
+	xorl	%r9d,%r14d
+	addl	%r13d,%r8d
+	xorl	%r10d,%r15d
+	shrdl	$2,%r14d,%r14d
+	addl	%r8d,%eax
+	addl	%r15d,%r8d
+	movl	%eax,%r13d
+	addl	%r8d,%r14d
+	vmovdqa	%xmm6,32(%rsp)
+	vpalignr	$4,%xmm3,%xmm0,%xmm4
+	shrdl	$14,%r13d,%r13d
+	movl	%r14d,%r8d
+	movl	%ebx,%r12d
+	vpalignr	$4,%xmm1,%xmm2,%xmm7
+	shrdl	$9,%r14d,%r14d
+	xorl	%eax,%r13d
+	xorl	%ecx,%r12d
+	vpsrld	$7,%xmm4,%xmm6
+	shrdl	$5,%r13d,%r13d
+	xorl	%r8d,%r14d
+	andl	%eax,%r12d
+	vpaddd	%xmm7,%xmm3,%xmm3
+	xorl	%eax,%r13d
+	addl	48(%rsp),%edx
+	movl	%r8d,%r15d
+	vpsrld	$3,%xmm4,%xmm7
+	xorl	%ecx,%r12d
+	shrdl	$11,%r14d,%r14d
+	xorl	%r9d,%r15d
+	vpslld	$14,%xmm4,%xmm5
+	addl	%r12d,%edx
+	shrdl	$6,%r13d,%r13d
+	andl	%r15d,%edi
+	vpxor	%xmm6,%xmm7,%xmm4
+	xorl	%r8d,%r14d
+	addl	%r13d,%edx
+	xorl	%r9d,%edi
+	vpshufd	$250,%xmm2,%xmm7
+	shrdl	$2,%r14d,%r14d
+	addl	%edx,%r11d
+	addl	%edi,%edx
+	vpsrld	$11,%xmm6,%xmm6
+	movl	%r11d,%r13d
+	addl	%edx,%r14d
+	shrdl	$14,%r13d,%r13d
+	vpxor	%xmm5,%xmm4,%xmm4
+	movl	%r14d,%edx
+	movl	%eax,%r12d
+	shrdl	$9,%r14d,%r14d
+	vpslld	$11,%xmm5,%xmm5
+	xorl	%r11d,%r13d
+	xorl	%ebx,%r12d
+	shrdl	$5,%r13d,%r13d
+	vpxor	%xmm6,%xmm4,%xmm4
+	xorl	%edx,%r14d
+	andl	%r11d,%r12d
+	xorl	%r11d,%r13d
+	vpsrld	$10,%xmm7,%xmm6
+	addl	52(%rsp),%ecx
+	movl	%edx,%edi
+	xorl	%ebx,%r12d
+	vpxor	%xmm5,%xmm4,%xmm4
+	shrdl	$11,%r14d,%r14d
+	xorl	%r8d,%edi
+	addl	%r12d,%ecx
+	vpsrlq	$17,%xmm7,%xmm7
+	shrdl	$6,%r13d,%r13d
+	andl	%edi,%r15d
+	xorl	%edx,%r14d
+	vpaddd	%xmm4,%xmm3,%xmm3
+	addl	%r13d,%ecx
+	xorl	%r8d,%r15d
+	shrdl	$2,%r14d,%r14d
+	vpxor	%xmm7,%xmm6,%xmm6
+	addl	%ecx,%r10d
+	addl	%r15d,%ecx
+	movl	%r10d,%r13d
+	vpsrlq	$2,%xmm7,%xmm7
+	addl	%ecx,%r14d
+	shrdl	$14,%r13d,%r13d
+	movl	%r14d,%ecx
+	vpxor	%xmm7,%xmm6,%xmm6
+	movl	%r11d,%r12d
+	shrdl	$9,%r14d,%r14d
+	xorl	%r10d,%r13d
+	vpshufb	%xmm8,%xmm6,%xmm6
+	xorl	%eax,%r12d
+	shrdl	$5,%r13d,%r13d
+	xorl	%ecx,%r14d
+	vpaddd	%xmm6,%xmm3,%xmm3
+	andl	%r10d,%r12d
+	xorl	%r10d,%r13d
+	addl	56(%rsp),%ebx
+	vpshufd	$80,%xmm3,%xmm7
+	movl	%ecx,%r15d
+	xorl	%eax,%r12d
+	shrdl	$11,%r14d,%r14d
+	vpsrld	$10,%xmm7,%xmm6
+	xorl	%edx,%r15d
+	addl	%r12d,%ebx
+	shrdl	$6,%r13d,%r13d
+	vpsrlq	$17,%xmm7,%xmm7
+	andl	%r15d,%edi
+	xorl	%ecx,%r14d
+	addl	%r13d,%ebx
+	vpxor	%xmm7,%xmm6,%xmm6
+	xorl	%edx,%edi
+	shrdl	$2,%r14d,%r14d
+	addl	%ebx,%r9d
+	vpsrlq	$2,%xmm7,%xmm7
+	addl	%edi,%ebx
+	movl	%r9d,%r13d
+	addl	%ebx,%r14d
+	vpxor	%xmm7,%xmm6,%xmm6
+	shrdl	$14,%r13d,%r13d
+	movl	%r14d,%ebx
+	movl	%r10d,%r12d
+	vpshufb	%xmm9,%xmm6,%xmm6
+	shrdl	$9,%r14d,%r14d
+	xorl	%r9d,%r13d
+	xorl	%r11d,%r12d
+	vpaddd	%xmm6,%xmm3,%xmm3
+	shrdl	$5,%r13d,%r13d
+	xorl	%ebx,%r14d
+	andl	%r9d,%r12d
+	vpaddd	96(%rbp),%xmm3,%xmm6
+	xorl	%r9d,%r13d
+	addl	60(%rsp),%eax
+	movl	%ebx,%edi
+	xorl	%r11d,%r12d
+	shrdl	$11,%r14d,%r14d
+	xorl	%ecx,%edi
+	addl	%r12d,%eax
+	shrdl	$6,%r13d,%r13d
+	andl	%edi,%r15d
+	xorl	%ebx,%r14d
+	addl	%r13d,%eax
+	xorl	%ecx,%r15d
+	shrdl	$2,%r14d,%r14d
+	addl	%eax,%r8d
+	addl	%r15d,%eax
+	movl	%r8d,%r13d
+	addl	%eax,%r14d
+	vmovdqa	%xmm6,48(%rsp)
+	cmpb	$0,131(%rbp)
+	jne	L$avx_00_47
+	shrdl	$14,%r13d,%r13d
+	movl	%r14d,%eax
+	movl	%r9d,%r12d
+	shrdl	$9,%r14d,%r14d
+	xorl	%r8d,%r13d
+	xorl	%r10d,%r12d
+	shrdl	$5,%r13d,%r13d
+	xorl	%eax,%r14d
+	andl	%r8d,%r12d
+	xorl	%r8d,%r13d
+	addl	0(%rsp),%r11d
+	movl	%eax,%r15d
+	xorl	%r10d,%r12d
+	shrdl	$11,%r14d,%r14d
+	xorl	%ebx,%r15d
+	addl	%r12d,%r11d
+	shrdl	$6,%r13d,%r13d
+	andl	%r15d,%edi
+	xorl	%eax,%r14d
+	addl	%r13d,%r11d
+	xorl	%ebx,%edi
+	shrdl	$2,%r14d,%r14d
+	addl	%r11d,%edx
+	addl	%edi,%r11d
+	movl	%edx,%r13d
+	addl	%r11d,%r14d
+	shrdl	$14,%r13d,%r13d
+	movl	%r14d,%r11d
+	movl	%r8d,%r12d
+	shrdl	$9,%r14d,%r14d
+	xorl	%edx,%r13d
+	xorl	%r9d,%r12d
+	shrdl	$5,%r13d,%r13d
+	xorl	%r11d,%r14d
+	andl	%edx,%r12d
+	xorl	%edx,%r13d
+	addl	4(%rsp),%r10d
+	movl	%r11d,%edi
+	xorl	%r9d,%r12d
+	shrdl	$11,%r14d,%r14d
+	xorl	%eax,%edi
+	addl	%r12d,%r10d
+	shrdl	$6,%r13d,%r13d
+	andl	%edi,%r15d
+	xorl	%r11d,%r14d
+	addl	%r13d,%r10d
+	xorl	%eax,%r15d
+	shrdl	$2,%r14d,%r14d
+	addl	%r10d,%ecx
+	addl	%r15d,%r10d
+	movl	%ecx,%r13d
+	addl	%r10d,%r14d
+	shrdl	$14,%r13d,%r13d
+	movl	%r14d,%r10d
+	movl	%edx,%r12d
+	shrdl	$9,%r14d,%r14d
+	xorl	%ecx,%r13d
+	xorl	%r8d,%r12d
+	shrdl	$5,%r13d,%r13d
+	xorl	%r10d,%r14d
+	andl	%ecx,%r12d
+	xorl	%ecx,%r13d
+	addl	8(%rsp),%r9d
+	movl	%r10d,%r15d
+	xorl	%r8d,%r12d
+	shrdl	$11,%r14d,%r14d
+	xorl	%r11d,%r15d
+	addl	%r12d,%r9d
+	shrdl	$6,%r13d,%r13d
+	andl	%r15d,%edi
+	xorl	%r10d,%r14d
+	addl	%r13d,%r9d
+	xorl	%r11d,%edi
+	shrdl	$2,%r14d,%r14d
+	addl	%r9d,%ebx
+	addl	%edi,%r9d
+	movl	%ebx,%r13d
+	addl	%r9d,%r14d
+	shrdl	$14,%r13d,%r13d
+	movl	%r14d,%r9d
+	movl	%ecx,%r12d
+	shrdl	$9,%r14d,%r14d
+	xorl	%ebx,%r13d
+	xorl	%edx,%r12d
+	shrdl	$5,%r13d,%r13d
+	xorl	%r9d,%r14d
+	andl	%ebx,%r12d
+	xorl	%ebx,%r13d
+	addl	12(%rsp),%r8d
+	movl	%r9d,%edi
+	xorl	%edx,%r12d
+	shrdl	$11,%r14d,%r14d
+	xorl	%r10d,%edi
+	addl	%r12d,%r8d
+	shrdl	$6,%r13d,%r13d
+	andl	%edi,%r15d
+	xorl	%r9d,%r14d
+	addl	%r13d,%r8d
+	xorl	%r10d,%r15d
+	shrdl	$2,%r14d,%r14d
+	addl	%r8d,%eax
+	addl	%r15d,%r8d
+	movl	%eax,%r13d
+	addl	%r8d,%r14d
+	shrdl	$14,%r13d,%r13d
+	movl	%r14d,%r8d
+	movl	%ebx,%r12d
+	shrdl	$9,%r14d,%r14d
+	xorl	%eax,%r13d
+	xorl	%ecx,%r12d
+	shrdl	$5,%r13d,%r13d
+	xorl	%r8d,%r14d
+	andl	%eax,%r12d
+	xorl	%eax,%r13d
+	addl	16(%rsp),%edx
+	movl	%r8d,%r15d
+	xorl	%ecx,%r12d
+	shrdl	$11,%r14d,%r14d
+	xorl	%r9d,%r15d
+	addl	%r12d,%edx
+	shrdl	$6,%r13d,%r13d
+	andl	%r15d,%edi
+	xorl	%r8d,%r14d
+	addl	%r13d,%edx
+	xorl	%r9d,%edi
+	shrdl	$2,%r14d,%r14d
+	addl	%edx,%r11d
+	addl	%edi,%edx
+	movl	%r11d,%r13d
+	addl	%edx,%r14d
+	shrdl	$14,%r13d,%r13d
+	movl	%r14d,%edx
+	movl	%eax,%r12d
+	shrdl	$9,%r14d,%r14d
+	xorl	%r11d,%r13d
+	xorl	%ebx,%r12d
+	shrdl	$5,%r13d,%r13d
+	xorl	%edx,%r14d
+	andl	%r11d,%r12d
+	xorl	%r11d,%r13d
+	addl	20(%rsp),%ecx
+	movl	%edx,%edi
+	xorl	%ebx,%r12d
+	shrdl	$11,%r14d,%r14d
+	xorl	%r8d,%edi
+	addl	%r12d,%ecx
+	shrdl	$6,%r13d,%r13d
+	andl	%edi,%r15d
+	xorl	%edx,%r14d
+	addl	%r13d,%ecx
+	xorl	%r8d,%r15d
+	shrdl	$2,%r14d,%r14d
+	addl	%ecx,%r10d
+	addl	%r15d,%ecx
+	movl	%r10d,%r13d
+	addl	%ecx,%r14d
+	shrdl	$14,%r13d,%r13d
+	movl	%r14d,%ecx
+	movl	%r11d,%r12d
+	shrdl	$9,%r14d,%r14d
+	xorl	%r10d,%r13d
+	xorl	%eax,%r12d
+	shrdl	$5,%r13d,%r13d
+	xorl	%ecx,%r14d
+	andl	%r10d,%r12d
+	xorl	%r10d,%r13d
+	addl	24(%rsp),%ebx
+	movl	%ecx,%r15d
+	xorl	%eax,%r12d
+	shrdl	$11,%r14d,%r14d
+	xorl	%edx,%r15d
+	addl	%r12d,%ebx
+	shrdl	$6,%r13d,%r13d
+	andl	%r15d,%edi
+	xorl	%ecx,%r14d
+	addl	%r13d,%ebx
+	xorl	%edx,%edi
+	shrdl	$2,%r14d,%r14d
+	addl	%ebx,%r9d
+	addl	%edi,%ebx
+	movl	%r9d,%r13d
+	addl	%ebx,%r14d
+	shrdl	$14,%r13d,%r13d
+	movl	%r14d,%ebx
+	movl	%r10d,%r12d
+	shrdl	$9,%r14d,%r14d
+	xorl	%r9d,%r13d
+	xorl	%r11d,%r12d
+	shrdl	$5,%r13d,%r13d
+	xorl	%ebx,%r14d
+	andl	%r9d,%r12d
+	xorl	%r9d,%r13d
+	addl	28(%rsp),%eax
+	movl	%ebx,%edi
+	xorl	%r11d,%r12d
+	shrdl	$11,%r14d,%r14d
+	xorl	%ecx,%edi
+	addl	%r12d,%eax
+	shrdl	$6,%r13d,%r13d
+	andl	%edi,%r15d
+	xorl	%ebx,%r14d
+	addl	%r13d,%eax
+	xorl	%ecx,%r15d
+	shrdl	$2,%r14d,%r14d
+	addl	%eax,%r8d
+	addl	%r15d,%eax
+	movl	%r8d,%r13d
+	addl	%eax,%r14d
+	shrdl	$14,%r13d,%r13d
+	movl	%r14d,%eax
+	movl	%r9d,%r12d
+	shrdl	$9,%r14d,%r14d
+	xorl	%r8d,%r13d
+	xorl	%r10d,%r12d
+	shrdl	$5,%r13d,%r13d
+	xorl	%eax,%r14d
+	andl	%r8d,%r12d
+	xorl	%r8d,%r13d
+	addl	32(%rsp),%r11d
+	movl	%eax,%r15d
+	xorl	%r10d,%r12d
+	shrdl	$11,%r14d,%r14d
+	xorl	%ebx,%r15d
+	addl	%r12d,%r11d
+	shrdl	$6,%r13d,%r13d
+	andl	%r15d,%edi
+	xorl	%eax,%r14d
+	addl	%r13d,%r11d
+	xorl	%ebx,%edi
+	shrdl	$2,%r14d,%r14d
+	addl	%r11d,%edx
+	addl	%edi,%r11d
+	movl	%edx,%r13d
+	addl	%r11d,%r14d
+	shrdl	$14,%r13d,%r13d
+	movl	%r14d,%r11d
+	movl	%r8d,%r12d
+	shrdl	$9,%r14d,%r14d
+	xorl	%edx,%r13d
+	xorl	%r9d,%r12d
+	shrdl	$5,%r13d,%r13d
+	xorl	%r11d,%r14d
+	andl	%edx,%r12d
+	xorl	%edx,%r13d
+	addl	36(%rsp),%r10d
+	movl	%r11d,%edi
+	xorl	%r9d,%r12d
+	shrdl	$11,%r14d,%r14d
+	xorl	%eax,%edi
+	addl	%r12d,%r10d
+	shrdl	$6,%r13d,%r13d
+	andl	%edi,%r15d
+	xorl	%r11d,%r14d
+	addl	%r13d,%r10d
+	xorl	%eax,%r15d
+	shrdl	$2,%r14d,%r14d
+	addl	%r10d,%ecx
+	addl	%r15d,%r10d
+	movl	%ecx,%r13d
+	addl	%r10d,%r14d
+	shrdl	$14,%r13d,%r13d
+	movl	%r14d,%r10d
+	movl	%edx,%r12d
+	shrdl	$9,%r14d,%r14d
+	xorl	%ecx,%r13d
+	xorl	%r8d,%r12d
+	shrdl	$5,%r13d,%r13d
+	xorl	%r10d,%r14d
+	andl	%ecx,%r12d
+	xorl	%ecx,%r13d
+	addl	40(%rsp),%r9d
+	movl	%r10d,%r15d
+	xorl	%r8d,%r12d
+	shrdl	$11,%r14d,%r14d
+	xorl	%r11d,%r15d
+	addl	%r12d,%r9d
+	shrdl	$6,%r13d,%r13d
+	andl	%r15d,%edi
+	xorl	%r10d,%r14d
+	addl	%r13d,%r9d
+	xorl	%r11d,%edi
+	shrdl	$2,%r14d,%r14d
+	addl	%r9d,%ebx
+	addl	%edi,%r9d
+	movl	%ebx,%r13d
+	addl	%r9d,%r14d
+	shrdl	$14,%r13d,%r13d
+	movl	%r14d,%r9d
+	movl	%ecx,%r12d
+	shrdl	$9,%r14d,%r14d
+	xorl	%ebx,%r13d
+	xorl	%edx,%r12d
+	shrdl	$5,%r13d,%r13d
+	xorl	%r9d,%r14d
+	andl	%ebx,%r12d
+	xorl	%ebx,%r13d
+	addl	44(%rsp),%r8d
+	movl	%r9d,%edi
+	xorl	%edx,%r12d
+	shrdl	$11,%r14d,%r14d
+	xorl	%r10d,%edi
+	addl	%r12d,%r8d
+	shrdl	$6,%r13d,%r13d
+	andl	%edi,%r15d
+	xorl	%r9d,%r14d
+	addl	%r13d,%r8d
+	xorl	%r10d,%r15d
+	shrdl	$2,%r14d,%r14d
+	addl	%r8d,%eax
+	addl	%r15d,%r8d
+	movl	%eax,%r13d
+	addl	%r8d,%r14d
+	shrdl	$14,%r13d,%r13d
+	movl	%r14d,%r8d
+	movl	%ebx,%r12d
+	shrdl	$9,%r14d,%r14d
+	xorl	%eax,%r13d
+	xorl	%ecx,%r12d
+	shrdl	$5,%r13d,%r13d
+	xorl	%r8d,%r14d
+	andl	%eax,%r12d
+	xorl	%eax,%r13d
+	addl	48(%rsp),%edx
+	movl	%r8d,%r15d
+	xorl	%ecx,%r12d
+	shrdl	$11,%r14d,%r14d
+	xorl	%r9d,%r15d
+	addl	%r12d,%edx
+	shrdl	$6,%r13d,%r13d
+	andl	%r15d,%edi
+	xorl	%r8d,%r14d
+	addl	%r13d,%edx
+	xorl	%r9d,%edi
+	shrdl	$2,%r14d,%r14d
+	addl	%edx,%r11d
+	addl	%edi,%edx
+	movl	%r11d,%r13d
+	addl	%edx,%r14d
+	shrdl	$14,%r13d,%r13d
+	movl	%r14d,%edx
+	movl	%eax,%r12d
+	shrdl	$9,%r14d,%r14d
+	xorl	%r11d,%r13d
+	xorl	%ebx,%r12d
+	shrdl	$5,%r13d,%r13d
+	xorl	%edx,%r14d
+	andl	%r11d,%r12d
+	xorl	%r11d,%r13d
+	addl	52(%rsp),%ecx
+	movl	%edx,%edi
+	xorl	%ebx,%r12d
+	shrdl	$11,%r14d,%r14d
+	xorl	%r8d,%edi
+	addl	%r12d,%ecx
+	shrdl	$6,%r13d,%r13d
+	andl	%edi,%r15d
+	xorl	%edx,%r14d
+	addl	%r13d,%ecx
+	xorl	%r8d,%r15d
+	shrdl	$2,%r14d,%r14d
+	addl	%ecx,%r10d
+	addl	%r15d,%ecx
+	movl	%r10d,%r13d
+	addl	%ecx,%r14d
+	shrdl	$14,%r13d,%r13d
+	movl	%r14d,%ecx
+	movl	%r11d,%r12d
+	shrdl	$9,%r14d,%r14d
+	xorl	%r10d,%r13d
+	xorl	%eax,%r12d
+	shrdl	$5,%r13d,%r13d
+	xorl	%ecx,%r14d
+	andl	%r10d,%r12d
+	xorl	%r10d,%r13d
+	addl	56(%rsp),%ebx
+	movl	%ecx,%r15d
+	xorl	%eax,%r12d
+	shrdl	$11,%r14d,%r14d
+	xorl	%edx,%r15d
+	addl	%r12d,%ebx
+	shrdl	$6,%r13d,%r13d
+	andl	%r15d,%edi
+	xorl	%ecx,%r14d
+	addl	%r13d,%ebx
+	xorl	%edx,%edi
+	shrdl	$2,%r14d,%r14d
+	addl	%ebx,%r9d
+	addl	%edi,%ebx
+	movl	%r9d,%r13d
+	addl	%ebx,%r14d
+	shrdl	$14,%r13d,%r13d
+	movl	%r14d,%ebx
+	movl	%r10d,%r12d
+	shrdl	$9,%r14d,%r14d
+	xorl	%r9d,%r13d
+	xorl	%r11d,%r12d
+	shrdl	$5,%r13d,%r13d
+	xorl	%ebx,%r14d
+	andl	%r9d,%r12d
+	xorl	%r9d,%r13d
+	addl	60(%rsp),%eax
+	movl	%ebx,%edi
+	xorl	%r11d,%r12d
+	shrdl	$11,%r14d,%r14d
+	xorl	%ecx,%edi
+	addl	%r12d,%eax
+	shrdl	$6,%r13d,%r13d
+	andl	%edi,%r15d
+	xorl	%ebx,%r14d
+	addl	%r13d,%eax
+	xorl	%ecx,%r15d
+	shrdl	$2,%r14d,%r14d
+	addl	%eax,%r8d
+	addl	%r15d,%eax
+	movl	%r8d,%r13d
+	addl	%eax,%r14d
+	movq	64+0(%rsp),%rdi
+	movl	%r14d,%eax
+
+	addl	0(%rdi),%eax
+	leaq	64(%rsi),%rsi
+	addl	4(%rdi),%ebx
+	addl	8(%rdi),%ecx
+	addl	12(%rdi),%edx
+	addl	16(%rdi),%r8d
+	addl	20(%rdi),%r9d
+	addl	24(%rdi),%r10d
+	addl	28(%rdi),%r11d
+
+	cmpq	64+16(%rsp),%rsi
+
+	movl	%eax,0(%rdi)
+	movl	%ebx,4(%rdi)
+	movl	%ecx,8(%rdi)
+	movl	%edx,12(%rdi)
+	movl	%r8d,16(%rdi)
+	movl	%r9d,20(%rdi)
+	movl	%r10d,24(%rdi)
+	movl	%r11d,28(%rdi)
+	jb	L$loop_avx
+
+	movq	64+24(%rsp),%rsi
+	vzeroupper
+	movq	(%rsi),%r15
+	movq	8(%rsi),%r14
+	movq	16(%rsi),%r13
+	movq	24(%rsi),%r12
+	movq	32(%rsi),%rbp
+	movq	40(%rsi),%rbx
+	leaq	48(%rsi),%rsp
+L$epilogue_avx:
+	.byte	0xf3,0xc3
+
 #endif

third_party/boringssl/mac-x86_64/crypto/sha/sha512-x86_64.S

diff --git a/third_party/boringssl/mac-x86_64/crypto/sha/sha512-x86_64.S b/third_party/boringssl/mac-x86_64/crypto/sha/sha512-x86_64.S
index 2f5d912..aeabd3f 100644
--- a/third_party/boringssl/mac-x86_64/crypto/sha/sha512-x86_64.S
+++ b/third_party/boringssl/mac-x86_64/crypto/sha/sha512-x86_64.S
@@ -7,6 +7,17 @@
 
 .p2align	4
 _sha512_block_data_order:
+	leaq	_OPENSSL_ia32cap_P(%rip),%r11
+	movl	0(%r11),%r9d
+	movl	4(%r11),%r10d
+	movl	8(%r11),%r11d
+	testl	$2048,%r10d
+	jnz	L$xop_shortcut
+	andl	$1073741824,%r9d
+	andl	$268435968,%r10d
+	orl	%r9d,%r10d
+	cmpl	$1342177792,%r10d
+	je	L$avx_shortcut
 	pushq	%rbx
 	pushq	%rbp
 	pushq	%r12
@@ -1783,4 +1794,2234 @@
 .quad	0x0001020304050607,0x08090a0b0c0d0e0f
 .quad	0x0001020304050607,0x08090a0b0c0d0e0f
 .byte	83,72,65,53,49,50,32,98,108,111,99,107,32,116,114,97,110,115,102,111,114,109,32,102,111,114,32,120,56,54,95,54,52,44,32,67,82,89,80,84,79,71,65,77,83,32,98,121,32,60,97,112,112,114,111,64,111,112,101,110,115,115,108,46,111,114,103,62,0
+
+.p2align	6
+sha512_block_data_order_xop:
+L$xop_shortcut:
+	pushq	%rbx
+	pushq	%rbp
+	pushq	%r12
+	pushq	%r13
+	pushq	%r14
+	pushq	%r15
+	movq	%rsp,%r11
+	shlq	$4,%rdx
+	subq	$160,%rsp
+	leaq	(%rsi,%rdx,8),%rdx
+	andq	$-64,%rsp
+	movq	%rdi,128+0(%rsp)
+	movq	%rsi,128+8(%rsp)
+	movq	%rdx,128+16(%rsp)
+	movq	%r11,128+24(%rsp)
+L$prologue_xop:
+
+	vzeroupper
+	movq	0(%rdi),%rax
+	movq	8(%rdi),%rbx
+	movq	16(%rdi),%rcx
+	movq	24(%rdi),%rdx
+	movq	32(%rdi),%r8
+	movq	40(%rdi),%r9
+	movq	48(%rdi),%r10
+	movq	56(%rdi),%r11
+	jmp	L$loop_xop
+.p2align	4
+L$loop_xop:
+	vmovdqa	K512+1280(%rip),%xmm11
+	vmovdqu	0(%rsi),%xmm0
+	leaq	K512+128(%rip),%rbp
+	vmovdqu	16(%rsi),%xmm1
+	vmovdqu	32(%rsi),%xmm2
+	vpshufb	%xmm11,%xmm0,%xmm0
+	vmovdqu	48(%rsi),%xmm3
+	vpshufb	%xmm11,%xmm1,%xmm1
+	vmovdqu	64(%rsi),%xmm4
+	vpshufb	%xmm11,%xmm2,%xmm2
+	vmovdqu	80(%rsi),%xmm5
+	vpshufb	%xmm11,%xmm3,%xmm3
+	vmovdqu	96(%rsi),%xmm6
+	vpshufb	%xmm11,%xmm4,%xmm4
+	vmovdqu	112(%rsi),%xmm7
+	vpshufb	%xmm11,%xmm5,%xmm5
+	vpaddq	-128(%rbp),%xmm0,%xmm8
+	vpshufb	%xmm11,%xmm6,%xmm6
+	vpaddq	-96(%rbp),%xmm1,%xmm9
+	vpshufb	%xmm11,%xmm7,%xmm7
+	vpaddq	-64(%rbp),%xmm2,%xmm10
+	vpaddq	-32(%rbp),%xmm3,%xmm11
+	vmovdqa	%xmm8,0(%rsp)
+	vpaddq	0(%rbp),%xmm4,%xmm8
+	vmovdqa	%xmm9,16(%rsp)
+	vpaddq	32(%rbp),%xmm5,%xmm9
+	vmovdqa	%xmm10,32(%rsp)
+	vpaddq	64(%rbp),%xmm6,%xmm10
+	vmovdqa	%xmm11,48(%rsp)
+	vpaddq	96(%rbp),%xmm7,%xmm11
+	vmovdqa	%xmm8,64(%rsp)
+	movq	%rax,%r14
+	vmovdqa	%xmm9,80(%rsp)
+	movq	%rbx,%rdi
+	vmovdqa	%xmm10,96(%rsp)
+	xorq	%rcx,%rdi
+	vmovdqa	%xmm11,112(%rsp)
+	movq	%r8,%r13
+	jmp	L$xop_00_47
+
+.p2align	4
+L$xop_00_47:
+	addq	$256,%rbp
+	vpalignr	$8,%xmm0,%xmm1,%xmm8
+	rorq	$23,%r13
+	movq	%r14,%rax
+	vpalignr	$8,%xmm4,%xmm5,%xmm11
+	movq	%r9,%r12
+	rorq	$5,%r14
+.byte	143,72,120,195,200,56
+	xorq	%r8,%r13
+	xorq	%r10,%r12
+	vpsrlq	$7,%xmm8,%xmm8
+	rorq	$4,%r13
+	xorq	%rax,%r14
+	vpaddq	%xmm11,%xmm0,%xmm0
+	andq	%r8,%r12
+	xorq	%r8,%r13
+	addq	0(%rsp),%r11
+	movq	%rax,%r15
+.byte	143,72,120,195,209,7
+	xorq	%r10,%r12
+	rorq	$6,%r14
+	vpxor	%xmm9,%xmm8,%xmm8
+	xorq	%rbx,%r15
+	addq	%r12,%r11
+	rorq	$14,%r13
+	andq	%r15,%rdi
+.byte	143,104,120,195,223,3
+	xorq	%rax,%r14
+	addq	%r13,%r11
+	vpxor	%xmm10,%xmm8,%xmm8
+	xorq	%rbx,%rdi
+	rorq	$28,%r14
+	vpsrlq	$6,%xmm7,%xmm10
+	addq	%r11,%rdx
+	addq	%rdi,%r11
+	vpaddq	%xmm8,%xmm0,%xmm0
+	movq	%rdx,%r13
+	addq	%r11,%r14
+.byte	143,72,120,195,203,42
+	rorq	$23,%r13
+	movq	%r14,%r11
+	vpxor	%xmm10,%xmm11,%xmm11
+	movq	%r8,%r12
+	rorq	$5,%r14
+	xorq	%rdx,%r13
+	xorq	%r9,%r12
+	vpxor	%xmm9,%xmm11,%xmm11
+	rorq	$4,%r13
+	xorq	%r11,%r14
+	andq	%rdx,%r12
+	xorq	%rdx,%r13
+	vpaddq	%xmm11,%xmm0,%xmm0
+	addq	8(%rsp),%r10
+	movq	%r11,%rdi
+	xorq	%r9,%r12
+	rorq	$6,%r14
+	vpaddq	-128(%rbp),%xmm0,%xmm10
+	xorq	%rax,%rdi
+	addq	%r12,%r10
+	rorq	$14,%r13
+	andq	%rdi,%r15
+	xorq	%r11,%r14
+	addq	%r13,%r10
+	xorq	%rax,%r15
+	rorq	$28,%r14
+	addq	%r10,%rcx
+	addq	%r15,%r10
+	movq	%rcx,%r13
+	addq	%r10,%r14
+	vmovdqa	%xmm10,0(%rsp)
+	vpalignr	$8,%xmm1,%xmm2,%xmm8
+	rorq	$23,%r13
+	movq	%r14,%r10
+	vpalignr	$8,%xmm5,%xmm6,%xmm11
+	movq	%rdx,%r12
+	rorq	$5,%r14
+.byte	143,72,120,195,200,56
+	xorq	%rcx,%r13
+	xorq	%r8,%r12
+	vpsrlq	$7,%xmm8,%xmm8
+	rorq	$4,%r13
+	xorq	%r10,%r14
+	vpaddq	%xmm11,%xmm1,%xmm1
+	andq	%rcx,%r12
+	xorq	%rcx,%r13
+	addq	16(%rsp),%r9
+	movq	%r10,%r15
+.byte	143,72,120,195,209,7
+	xorq	%r8,%r12
+	rorq	$6,%r14
+	vpxor	%xmm9,%xmm8,%xmm8
+	xorq	%r11,%r15
+	addq	%r12,%r9
+	rorq	$14,%r13
+	andq	%r15,%rdi
+.byte	143,104,120,195,216,3
+	xorq	%r10,%r14
+	addq	%r13,%r9
+	vpxor	%xmm10,%xmm8,%xmm8
+	xorq	%r11,%rdi
+	rorq	$28,%r14
+	vpsrlq	$6,%xmm0,%xmm10
+	addq	%r9,%rbx
+	addq	%rdi,%r9
+	vpaddq	%xmm8,%xmm1,%xmm1
+	movq	%rbx,%r13
+	addq	%r9,%r14
+.byte	143,72,120,195,203,42
+	rorq	$23,%r13
+	movq	%r14,%r9
+	vpxor	%xmm10,%xmm11,%xmm11
+	movq	%rcx,%r12
+	rorq	$5,%r14
+	xorq	%rbx,%r13
+	xorq	%rdx,%r12
+	vpxor	%xmm9,%xmm11,%xmm11
+	rorq	$4,%r13
+	xorq	%r9,%r14
+	andq	%rbx,%r12
+	xorq	%rbx,%r13
+	vpaddq	%xmm11,%xmm1,%xmm1
+	addq	24(%rsp),%r8
+	movq	%r9,%rdi
+	xorq	%rdx,%r12
+	rorq	$6,%r14
+	vpaddq	-96(%rbp),%xmm1,%xmm10
+	xorq	%r10,%rdi
+	addq	%r12,%r8
+	rorq	$14,%r13
+	andq	%rdi,%r15
+	xorq	%r9,%r14
+	addq	%r13,%r8
+	xorq	%r10,%r15
+	rorq	$28,%r14
+	addq	%r8,%rax
+	addq	%r15,%r8
+	movq	%rax,%r13
+	addq	%r8,%r14
+	vmovdqa	%xmm10,16(%rsp)
+	vpalignr	$8,%xmm2,%xmm3,%xmm8
+	rorq	$23,%r13
+	movq	%r14,%r8
+	vpalignr	$8,%xmm6,%xmm7,%xmm11
+	movq	%rbx,%r12
+	rorq	$5,%r14
+.byte	143,72,120,195,200,56
+	xorq	%rax,%r13
+	xorq	%rcx,%r12
+	vpsrlq	$7,%xmm8,%xmm8
+	rorq	$4,%r13
+	xorq	%r8,%r14
+	vpaddq	%xmm11,%xmm2,%xmm2
+	andq	%rax,%r12
+	xorq	%rax,%r13
+	addq	32(%rsp),%rdx
+	movq	%r8,%r15
+.byte	143,72,120,195,209,7
+	xorq	%rcx,%r12
+	rorq	$6,%r14
+	vpxor	%xmm9,%xmm8,%xmm8
+	xorq	%r9,%r15
+	addq	%r12,%rdx
+	rorq	$14,%r13
+	andq	%r15,%rdi
+.byte	143,104,120,195,217,3
+	xorq	%r8,%r14
+	addq	%r13,%rdx
+	vpxor	%xmm10,%xmm8,%xmm8
+	xorq	%r9,%rdi
+	rorq	$28,%r14
+	vpsrlq	$6,%xmm1,%xmm10
+	addq	%rdx,%r11
+	addq	%rdi,%rdx
+	vpaddq	%xmm8,%xmm2,%xmm2
+	movq	%r11,%r13
+	addq	%rdx,%r14
+.byte	143,72,120,195,203,42
+	rorq	$23,%r13
+	movq	%r14,%rdx
+	vpxor	%xmm10,%xmm11,%xmm11
+	movq	%rax,%r12
+	rorq	$5,%r14
+	xorq	%r11,%r13
+	xorq	%rbx,%r12
+	vpxor	%xmm9,%xmm11,%xmm11
+	rorq	$4,%r13
+	xorq	%rdx,%r14
+	andq	%r11,%r12
+	xorq	%r11,%r13
+	vpaddq	%xmm11,%xmm2,%xmm2
+	addq	40(%rsp),%rcx
+	movq	%rdx,%rdi
+	xorq	%rbx,%r12
+	rorq	$6,%r14
+	vpaddq	-64(%rbp),%xmm2,%xmm10
+	xorq	%r8,%rdi
+	addq	%r12,%rcx
+	rorq	$14,%r13
+	andq	%rdi,%r15
+	xorq	%rdx,%r14
+	addq	%r13,%rcx
+	xorq	%r8,%r15
+	rorq	$28,%r14
+	addq	%rcx,%r10
+	addq	%r15,%rcx
+	movq	%r10,%r13
+	addq	%rcx,%r14
+	vmovdqa	%xmm10,32(%rsp)
+	vpalignr	$8,%xmm3,%xmm4,%xmm8
+	rorq	$23,%r13
+	movq	%r14,%rcx
+	vpalignr	$8,%xmm7,%xmm0,%xmm11
+	movq	%r11,%r12
+	rorq	$5,%r14
+.byte	143,72,120,195,200,56
+	xorq	%r10,%r13
+	xorq	%rax,%r12
+	vpsrlq	$7,%xmm8,%xmm8
+	rorq	$4,%r13
+	xorq	%rcx,%r14
+	vpaddq	%xmm11,%xmm3,%xmm3
+	andq	%r10,%r12
+	xorq	%r10,%r13
+	addq	48(%rsp),%rbx
+	movq	%rcx,%r15
+.byte	143,72,120,195,209,7
+	xorq	%rax,%r12
+	rorq	$6,%r14
+	vpxor	%xmm9,%xmm8,%xmm8
+	xorq	%rdx,%r15
+	addq	%r12,%rbx
+	rorq	$14,%r13
+	andq	%r15,%rdi
+.byte	143,104,120,195,218,3
+	xorq	%rcx,%r14
+	addq	%r13,%rbx
+	vpxor	%xmm10,%xmm8,%xmm8
+	xorq	%rdx,%rdi
+	rorq	$28,%r14
+	vpsrlq	$6,%xmm2,%xmm10
+	addq	%rbx,%r9
+	addq	%rdi,%rbx
+	vpaddq	%xmm8,%xmm3,%xmm3
+	movq	%r9,%r13
+	addq	%rbx,%r14
+.byte	143,72,120,195,203,42
+	rorq	$23,%r13
+	movq	%r14,%rbx
+	vpxor	%xmm10,%xmm11,%xmm11
+	movq	%r10,%r12
+	rorq	$5,%r14
+	xorq	%r9,%r13
+	xorq	%r11,%r12
+	vpxor	%xmm9,%xmm11,%xmm11
+	rorq	$4,%r13
+	xorq	%rbx,%r14
+	andq	%r9,%r12
+	xorq	%r9,%r13
+	vpaddq	%xmm11,%xmm3,%xmm3
+	addq	56(%rsp),%rax
+	movq	%rbx,%rdi
+	xorq	%r11,%r12
+	rorq	$6,%r14
+	vpaddq	-32(%rbp),%xmm3,%xmm10
+	xorq	%rcx,%rdi
+	addq	%r12,%rax
+	rorq	$14,%r13
+	andq	%rdi,%r15
+	xorq	%rbx,%r14
+	addq	%r13,%rax
+	xorq	%rcx,%r15
+	rorq	$28,%r14
+	addq	%rax,%r8
+	addq	%r15,%rax
+	movq	%r8,%r13
+	addq	%rax,%r14
+	vmovdqa	%xmm10,48(%rsp)
+	vpalignr	$8,%xmm4,%xmm5,%xmm8
+	rorq	$23,%r13
+	movq	%r14,%rax
+	vpalignr	$8,%xmm0,%xmm1,%xmm11
+	movq	%r9,%r12
+	rorq	$5,%r14
+.byte	143,72,120,195,200,56
+	xorq	%r8,%r13
+	xorq	%r10,%r12
+	vpsrlq	$7,%xmm8,%xmm8
+	rorq	$4,%r13
+	xorq	%rax,%r14
+	vpaddq	%xmm11,%xmm4,%xmm4
+	andq	%r8,%r12
+	xorq	%r8,%r13
+	addq	64(%rsp),%r11
+	movq	%rax,%r15
+.byte	143,72,120,195,209,7
+	xorq	%r10,%r12
+	rorq	$6,%r14
+	vpxor	%xmm9,%xmm8,%xmm8
+	xorq	%rbx,%r15
+	addq	%r12,%r11
+	rorq	$14,%r13
+	andq	%r15,%rdi
+.byte	143,104,120,195,219,3
+	xorq	%rax,%r14
+	addq	%r13,%r11
+	vpxor	%xmm10,%xmm8,%xmm8
+	xorq	%rbx,%rdi
+	rorq	$28,%r14
+	vpsrlq	$6,%xmm3,%xmm10
+	addq	%r11,%rdx
+	addq	%rdi,%r11
+	vpaddq	%xmm8,%xmm4,%xmm4
+	movq	%rdx,%r13
+	addq	%r11,%r14
+.byte	143,72,120,195,203,42
+	rorq	$23,%r13
+	movq	%r14,%r11
+	vpxor	%xmm10,%xmm11,%xmm11
+	movq	%r8,%r12
+	rorq	$5,%r14
+	xorq	%rdx,%r13
+	xorq	%r9,%r12
+	vpxor	%xmm9,%xmm11,%xmm11
+	rorq	$4,%r13
+	xorq	%r11,%r14
+	andq	%rdx,%r12
+	xorq	%rdx,%r13
+	vpaddq	%xmm11,%xmm4,%xmm4
+	addq	72(%rsp),%r10
+	movq	%r11,%rdi
+	xorq	%r9,%r12
+	rorq	$6,%r14
+	vpaddq	0(%rbp),%xmm4,%xmm10
+	xorq	%rax,%rdi
+	addq	%r12,%r10
+	rorq	$14,%r13
+	andq	%rdi,%r15
+	xorq	%r11,%r14
+	addq	%r13,%r10
+	xorq	%rax,%r15
+	rorq	$28,%r14
+	addq	%r10,%rcx
+	addq	%r15,%r10
+	movq	%rcx,%r13
+	addq	%r10,%r14
+	vmovdqa	%xmm10,64(%rsp)
+	vpalignr	$8,%xmm5,%xmm6,%xmm8
+	rorq	$23,%r13
+	movq	%r14,%r10
+	vpalignr	$8,%xmm1,%xmm2,%xmm11
+	movq	%rdx,%r12
+	rorq	$5,%r14
+.byte	143,72,120,195,200,56
+	xorq	%rcx,%r13
+	xorq	%r8,%r12
+	vpsrlq	$7,%xmm8,%xmm8
+	rorq	$4,%r13
+	xorq	%r10,%r14
+	vpaddq	%xmm11,%xmm5,%xmm5
+	andq	%rcx,%r12
+	xorq	%rcx,%r13
+	addq	80(%rsp),%r9
+	movq	%r10,%r15
+.byte	143,72,120,195,209,7
+	xorq	%r8,%r12
+	rorq	$6,%r14
+	vpxor	%xmm9,%xmm8,%xmm8
+	xorq	%r11,%r15
+	addq	%r12,%r9
+	rorq	$14,%r13
+	andq	%r15,%rdi
+.byte	143,104,120,195,220,3
+	xorq	%r10,%r14
+	addq	%r13,%r9
+	vpxor	%xmm10,%xmm8,%xmm8
+	xorq	%r11,%rdi
+	rorq	$28,%r14
+	vpsrlq	$6,%xmm4,%xmm10
+	addq	%r9,%rbx
+	addq	%rdi,%r9
+	vpaddq	%xmm8,%xmm5,%xmm5
+	movq	%rbx,%r13
+	addq	%r9,%r14
+.byte	143,72,120,195,203,42
+	rorq	$23,%r13
+	movq	%r14,%r9
+	vpxor	%xmm10,%xmm11,%xmm11
+	movq	%rcx,%r12
+	rorq	$5,%r14
+	xorq	%rbx,%r13
+	xorq	%rdx,%r12
+	vpxor	%xmm9,%xmm11,%xmm11
+	rorq	$4,%r13
+	xorq	%r9,%r14
+	andq	%rbx,%r12
+	xorq	%rbx,%r13
+	vpaddq	%xmm11,%xmm5,%xmm5
+	addq	88(%rsp),%r8
+	movq	%r9,%rdi
+	xorq	%rdx,%r12
+	rorq	$6,%r14
+	vpaddq	32(%rbp),%xmm5,%xmm10
+	xorq	%r10,%rdi
+	addq	%r12,%r8
+	rorq	$14,%r13
+	andq	%rdi,%r15
+	xorq	%r9,%r14
+	addq	%r13,%r8
+	xorq	%r10,%r15
+	rorq	$28,%r14
+	addq	%r8,%rax
+	addq	%r15,%r8
+	movq	%rax,%r13
+	addq	%r8,%r14
+	vmovdqa	%xmm10,80(%rsp)
+	vpalignr	$8,%xmm6,%xmm7,%xmm8
+	rorq	$23,%r13
+	movq	%r14,%r8
+	vpalignr	$8,%xmm2,%xmm3,%xmm11
+	movq	%rbx,%r12
+	rorq	$5,%r14
+.byte	143,72,120,195,200,56
+	xorq	%rax,%r13
+	xorq	%rcx,%r12
+	vpsrlq	$7,%xmm8,%xmm8
+	rorq	$4,%r13
+	xorq	%r8,%r14
+	vpaddq	%xmm11,%xmm6,%xmm6
+	andq	%rax,%r12
+	xorq	%rax,%r13
+	addq	96(%rsp),%rdx
+	movq	%r8,%r15
+.byte	143,72,120,195,209,7
+	xorq	%rcx,%r12
+	rorq	$6,%r14
+	vpxor	%xmm9,%xmm8,%xmm8
+	xorq	%r9,%r15
+	addq	%r12,%rdx
+	rorq	$14,%r13
+	andq	%r15,%rdi
+.byte	143,104,120,195,221,3
+	xorq	%r8,%r14
+	addq	%r13,%rdx
+	vpxor	%xmm10,%xmm8,%xmm8
+	xorq	%r9,%rdi
+	rorq	$28,%r14
+	vpsrlq	$6,%xmm5,%xmm10
+	addq	%rdx,%r11
+	addq	%rdi,%rdx
+	vpaddq	%xmm8,%xmm6,%xmm6
+	movq	%r11,%r13
+	addq	%rdx,%r14
+.byte	143,72,120,195,203,42
+	rorq	$23,%r13
+	movq	%r14,%rdx
+	vpxor	%xmm10,%xmm11,%xmm11
+	movq	%rax,%r12
+	rorq	$5,%r14
+	xorq	%r11,%r13
+	xorq	%rbx,%r12
+	vpxor	%xmm9,%xmm11,%xmm11
+	rorq	$4,%r13
+	xorq	%rdx,%r14
+	andq	%r11,%r12
+	xorq	%r11,%r13
+	vpaddq	%xmm11,%xmm6,%xmm6
+	addq	104(%rsp),%rcx
+	movq	%rdx,%rdi
+	xorq	%rbx,%r12
+	rorq	$6,%r14
+	vpaddq	64(%rbp),%xmm6,%xmm10
+	xorq	%r8,%rdi
+	addq	%r12,%rcx
+	rorq	$14,%r13
+	andq	%rdi,%r15
+	xorq	%rdx,%r14
+	addq	%r13,%rcx
+	xorq	%r8,%r15
+	rorq	$28,%r14
+	addq	%rcx,%r10
+	addq	%r15,%rcx
+	movq	%r10,%r13
+	addq	%rcx,%r14
+	vmovdqa	%xmm10,96(%rsp)
+	vpalignr	$8,%xmm7,%xmm0,%xmm8
+	rorq	$23,%r13
+	movq	%r14,%rcx
+	vpalignr	$8,%xmm3,%xmm4,%xmm11
+	movq	%r11,%r12
+	rorq	$5,%r14
+.byte	143,72,120,195,200,56
+	xorq	%r10,%r13
+	xorq	%rax,%r12
+	vpsrlq	$7,%xmm8,%xmm8
+	rorq	$4,%r13
+	xorq	%rcx,%r14
+	vpaddq	%xmm11,%xmm7,%xmm7
+	andq	%r10,%r12
+	xorq	%r10,%r13
+	addq	112(%rsp),%rbx
+	movq	%rcx,%r15
+.byte	143,72,120,195,209,7
+	xorq	%rax,%r12
+	rorq	$6,%r14
+	vpxor	%xmm9,%xmm8,%xmm8
+	xorq	%rdx,%r15
+	addq	%r12,%rbx
+	rorq	$14,%r13
+	andq	%r15,%rdi
+.byte	143,104,120,195,222,3
+	xorq	%rcx,%r14
+	addq	%r13,%rbx
+	vpxor	%xmm10,%xmm8,%xmm8
+	xorq	%rdx,%rdi
+	rorq	$28,%r14
+	vpsrlq	$6,%xmm6,%xmm10
+	addq	%rbx,%r9
+	addq	%rdi,%rbx
+	vpaddq	%xmm8,%xmm7,%xmm7
+	movq	%r9,%r13
+	addq	%rbx,%r14
+.byte	143,72,120,195,203,42
+	rorq	$23,%r13
+	movq	%r14,%rbx
+	vpxor	%xmm10,%xmm11,%xmm11
+	movq	%r10,%r12
+	rorq	$5,%r14
+	xorq	%r9,%r13
+	xorq	%r11,%r12
+	vpxor	%xmm9,%xmm11,%xmm11
+	rorq	$4,%r13
+	xorq	%rbx,%r14
+	andq	%r9,%r12
+	xorq	%r9,%r13
+	vpaddq	%xmm11,%xmm7,%xmm7
+	addq	120(%rsp),%rax
+	movq	%rbx,%rdi
+	xorq	%r11,%r12
+	rorq	$6,%r14
+	vpaddq	96(%rbp),%xmm7,%xmm10
+	xorq	%rcx,%rdi
+	addq	%r12,%rax
+	rorq	$14,%r13
+	andq	%rdi,%r15
+	xorq	%rbx,%r14
+	addq	%r13,%rax
+	xorq	%rcx,%r15
+	rorq	$28,%r14
+	addq	%rax,%r8
+	addq	%r15,%rax
+	movq	%r8,%r13
+	addq	%rax,%r14
+	vmovdqa	%xmm10,112(%rsp)
+	cmpb	$0,135(%rbp)
+	jne	L$xop_00_47
+	rorq	$23,%r13
+	movq	%r14,%rax
+	movq	%r9,%r12
+	rorq	$5,%r14
+	xorq	%r8,%r13
+	xorq	%r10,%r12
+	rorq	$4,%r13
+	xorq	%rax,%r14
+	andq	%r8,%r12
+	xorq	%r8,%r13
+	addq	0(%rsp),%r11
+	movq	%rax,%r15
+	xorq	%r10,%r12
+	rorq	$6,%r14
+	xorq	%rbx,%r15
+	addq	%r12,%r11
+	rorq	$14,%r13
+	andq	%r15,%rdi
+	xorq	%rax,%r14
+	addq	%r13,%r11
+	xorq	%rbx,%rdi
+	rorq	$28,%r14
+	addq	%r11,%rdx
+	addq	%rdi,%r11
+	movq	%rdx,%r13
+	addq	%r11,%r14
+	rorq	$23,%r13
+	movq	%r14,%r11
+	movq	%r8,%r12
+	rorq	$5,%r14
+	xorq	%rdx,%r13
+	xorq	%r9,%r12
+	rorq	$4,%r13
+	xorq	%r11,%r14
+	andq	%rdx,%r12
+	xorq	%rdx,%r13
+	addq	8(%rsp),%r10
+	movq	%r11,%rdi
+	xorq	%r9,%r12
+	rorq	$6,%r14
+	xorq	%rax,%rdi
+	addq	%r12,%r10
+	rorq	$14,%r13
+	andq	%rdi,%r15
+	xorq	%r11,%r14
+	addq	%r13,%r10
+	xorq	%rax,%r15
+	rorq	$28,%r14
+	addq	%r10,%rcx
+	addq	%r15,%r10
+	movq	%rcx,%r13
+	addq	%r10,%r14
+	rorq	$23,%r13
+	movq	%r14,%r10
+	movq	%rdx,%r12
+	rorq	$5,%r14
+	xorq	%rcx,%r13
+	xorq	%r8,%r12
+	rorq	$4,%r13
+	xorq	%r10,%r14
+	andq	%rcx,%r12
+	xorq	%rcx,%r13
+	addq	16(%rsp),%r9
+	movq	%r10,%r15
+	xorq	%r8,%r12
+	rorq	$6,%r14
+	xorq	%r11,%r15
+	addq	%r12,%r9
+	rorq	$14,%r13
+	andq	%r15,%rdi
+	xorq	%r10,%r14
+	addq	%r13,%r9
+	xorq	%r11,%rdi
+	rorq	$28,%r14
+	addq	%r9,%rbx
+	addq	%rdi,%r9
+	movq	%rbx,%r13
+	addq	%r9,%r14
+	rorq	$23,%r13
+	movq	%r14,%r9
+	movq	%rcx,%r12
+	rorq	$5,%r14
+	xorq	%rbx,%r13
+	xorq	%rdx,%r12
+	rorq	$4,%r13
+	xorq	%r9,%r14
+	andq	%rbx,%r12
+	xorq	%rbx,%r13
+	addq	24(%rsp),%r8
+	movq	%r9,%rdi
+	xorq	%rdx,%r12
+	rorq	$6,%r14
+	xorq	%r10,%rdi
+	addq	%r12,%r8
+	rorq	$14,%r13
+	andq	%rdi,%r15
+	xorq	%r9,%r14
+	addq	%r13,%r8
+	xorq	%r10,%r15
+	rorq	$28,%r14
+	addq	%r8,%rax
+	addq	%r15,%r8
+	movq	%rax,%r13
+	addq	%r8,%r14
+	rorq	$23,%r13
+	movq	%r14,%r8
+	movq	%rbx,%r12
+	rorq	$5,%r14
+	xorq	%rax,%r13
+	xorq	%rcx,%r12
+	rorq	$4,%r13
+	xorq	%r8,%r14
+	andq	%rax,%r12
+	xorq	%rax,%r13
+	addq	32(%rsp),%rdx
+	movq	%r8,%r15
+	xorq	%rcx,%r12
+	rorq	$6,%r14
+	xorq	%r9,%r15
+	addq	%r12,%rdx
+	rorq	$14,%r13
+	andq	%r15,%rdi
+	xorq	%r8,%r14
+	addq	%r13,%rdx
+	xorq	%r9,%rdi
+	rorq	$28,%r14
+	addq	%rdx,%r11
+	addq	%rdi,%rdx
+	movq	%r11,%r13
+	addq	%rdx,%r14
+	rorq	$23,%r13
+	movq	%r14,%rdx
+	movq	%rax,%r12
+	rorq	$5,%r14
+	xorq	%r11,%r13
+	xorq	%rbx,%r12
+	rorq	$4,%r13
+	xorq	%rdx,%r14
+	andq	%r11,%r12
+	xorq	%r11,%r13
+	addq	40(%rsp),%rcx
+	movq	%rdx,%rdi
+	xorq	%rbx,%r12
+	rorq	$6,%r14
+	xorq	%r8,%rdi
+	addq	%r12,%rcx
+	rorq	$14,%r13
+	andq	%rdi,%r15
+	xorq	%rdx,%r14
+	addq	%r13,%rcx
+	xorq	%r8,%r15
+	rorq	$28,%r14
+	addq	%rcx,%r10
+	addq	%r15,%rcx
+	movq	%r10,%r13
+	addq	%rcx,%r14
+	rorq	$23,%r13
+	movq	%r14,%rcx
+	movq	%r11,%r12
+	rorq	$5,%r14
+	xorq	%r10,%r13
+	xorq	%rax,%r12
+	rorq	$4,%r13
+	xorq	%rcx,%r14
+	andq	%r10,%r12
+	xorq	%r10,%r13
+	addq	48(%rsp),%rbx
+	movq	%rcx,%r15
+	xorq	%rax,%r12
+	rorq	$6,%r14
+	xorq	%rdx,%r15
+	addq	%r12,%rbx
+	rorq	$14,%r13
+	andq	%r15,%rdi
+	xorq	%rcx,%r14
+	addq	%r13,%rbx
+	xorq	%rdx,%rdi
+	rorq	$28,%r14
+	addq	%rbx,%r9
+	addq	%rdi,%rbx
+	movq	%r9,%r13
+	addq	%rbx,%r14
+	rorq	$23,%r13
+	movq	%r14,%rbx
+	movq	%r10,%r12
+	rorq	$5,%r14
+	xorq	%r9,%r13
+	xorq	%r11,%r12
+	rorq	$4,%r13
+	xorq	%rbx,%r14
+	andq	%r9,%r12
+	xorq	%r9,%r13
+	addq	56(%rsp),%rax
+	movq	%rbx,%rdi
+	xorq	%r11,%r12
+	rorq	$6,%r14
+	xorq	%rcx,%rdi
+	addq	%r12,%rax
+	rorq	$14,%r13
+	andq	%rdi,%r15
+	xorq	%rbx,%r14
+	addq	%r13,%rax
+	xorq	%rcx,%r15
+	rorq	$28,%r14
+	addq	%rax,%r8
+	addq	%r15,%rax
+	movq	%r8,%r13
+	addq	%rax,%r14
+	rorq	$23,%r13
+	movq	%r14,%rax
+	movq	%r9,%r12
+	rorq	$5,%r14
+	xorq	%r8,%r13
+	xorq	%r10,%r12
+	rorq	$4,%r13
+	xorq	%rax,%r14
+	andq	%r8,%r12
+	xorq	%r8,%r13
+	addq	64(%rsp),%r11
+	movq	%rax,%r15
+	xorq	%r10,%r12
+	rorq	$6,%r14
+	xorq	%rbx,%r15
+	addq	%r12,%r11
+	rorq	$14,%r13
+	andq	%r15,%rdi
+	xorq	%rax,%r14
+	addq	%r13,%r11
+	xorq	%rbx,%rdi
+	rorq	$28,%r14
+	addq	%r11,%rdx
+	addq	%rdi,%r11
+	movq	%rdx,%r13
+	addq	%r11,%r14
+	rorq	$23,%r13
+	movq	%r14,%r11
+	movq	%r8,%r12
+	rorq	$5,%r14
+	xorq	%rdx,%r13
+	xorq	%r9,%r12
+	rorq	$4,%r13
+	xorq	%r11,%r14
+	andq	%rdx,%r12
+	xorq	%rdx,%r13
+	addq	72(%rsp),%r10
+	movq	%r11,%rdi
+	xorq	%r9,%r12
+	rorq	$6,%r14
+	xorq	%rax,%rdi
+	addq	%r12,%r10
+	rorq	$14,%r13
+	andq	%rdi,%r15
+	xorq	%r11,%r14
+	addq	%r13,%r10
+	xorq	%rax,%r15
+	rorq	$28,%r14
+	addq	%r10,%rcx
+	addq	%r15,%r10
+	movq	%rcx,%r13
+	addq	%r10,%r14
+	rorq	$23,%r13
+	movq	%r14,%r10
+	movq	%rdx,%r12
+	rorq	$5,%r14
+	xorq	%rcx,%r13
+	xorq	%r8,%r12
+	rorq	$4,%r13
+	xorq	%r10,%r14
+	andq	%rcx,%r12
+	xorq	%rcx,%r13
+	addq	80(%rsp),%r9
+	movq	%r10,%r15
+	xorq	%r8,%r12
+	rorq	$6,%r14
+	xorq	%r11,%r15
+	addq	%r12,%r9
+	rorq	$14,%r13
+	andq	%r15,%rdi
+	xorq	%r10,%r14
+	addq	%r13,%r9
+	xorq	%r11,%rdi
+	rorq	$28,%r14
+	addq	%r9,%rbx
+	addq	%rdi,%r9
+	movq	%rbx,%r13
+	addq	%r9,%r14
+	rorq	$23,%r13
+	movq	%r14,%r9
+	movq	%rcx,%r12
+	rorq	$5,%r14
+	xorq	%rbx,%r13
+	xorq	%rdx,%r12
+	rorq	$4,%r13
+	xorq	%r9,%r14
+	andq	%rbx,%r12
+	xorq	%rbx,%r13
+	addq	88(%rsp),%r8
+	movq	%r9,%rdi
+	xorq	%rdx,%r12
+	rorq	$6,%r14
+	xorq	%r10,%rdi
+	addq	%r12,%r8
+	rorq	$14,%r13
+	andq	%rdi,%r15
+	xorq	%r9,%r14
+	addq	%r13,%r8
+	xorq	%r10,%r15
+	rorq	$28,%r14
+	addq	%r8,%rax
+	addq	%r15,%r8
+	movq	%rax,%r13
+	addq	%r8,%r14
+	rorq	$23,%r13
+	movq	%r14,%r8
+	movq	%rbx,%r12
+	rorq	$5,%r14
+	xorq	%rax,%r13
+	xorq	%rcx,%r12
+	rorq	$4,%r13
+	xorq	%r8,%r14
+	andq	%rax,%r12
+	xorq	%rax,%r13
+	addq	96(%rsp),%rdx
+	movq	%r8,%r15
+	xorq	%rcx,%r12
+	rorq	$6,%r14
+	xorq	%r9,%r15
+	addq	%r12,%rdx
+	rorq	$14,%r13
+	andq	%r15,%rdi
+	xorq	%r8,%r14
+	addq	%r13,%rdx
+	xorq	%r9,%rdi
+	rorq	$28,%r14
+	addq	%rdx,%r11
+	addq	%rdi,%rdx
+	movq	%r11,%r13
+	addq	%rdx,%r14
+	rorq	$23,%r13
+	movq	%r14,%rdx
+	movq	%rax,%r12
+	rorq	$5,%r14
+	xorq	%r11,%r13
+	xorq	%rbx,%r12
+	rorq	$4,%r13
+	xorq	%rdx,%r14
+	andq	%r11,%r12
+	xorq	%r11,%r13
+	addq	104(%rsp),%rcx
+	movq	%rdx,%rdi
+	xorq	%rbx,%r12
+	rorq	$6,%r14
+	xorq	%r8,%rdi
+	addq	%r12,%rcx
+	rorq	$14,%r13
+	andq	%rdi,%r15
+	xorq	%rdx,%r14
+	addq	%r13,%rcx
+	xorq	%r8,%r15
+	rorq	$28,%r14
+	addq	%rcx,%r10
+	addq	%r15,%rcx
+	movq	%r10,%r13
+	addq	%rcx,%r14
+	rorq	$23,%r13
+	movq	%r14,%rcx
+	movq	%r11,%r12
+	rorq	$5,%r14
+	xorq	%r10,%r13
+	xorq	%rax,%r12
+	rorq	$4,%r13
+	xorq	%rcx,%r14
+	andq	%r10,%r12
+	xorq	%r10,%r13
+	addq	112(%rsp),%rbx
+	movq	%rcx,%r15
+	xorq	%rax,%r12
+	rorq	$6,%r14
+	xorq	%rdx,%r15
+	addq	%r12,%rbx
+	rorq	$14,%r13
+	andq	%r15,%rdi
+	xorq	%rcx,%r14
+	addq	%r13,%rbx
+	xorq	%rdx,%rdi
+	rorq	$28,%r14
+	addq	%rbx,%r9
+	addq	%rdi,%rbx
+	movq	%r9,%r13
+	addq	%rbx,%r14
+	rorq	$23,%r13
+	movq	%r14,%rbx
+	movq	%r10,%r12
+	rorq	$5,%r14
+	xorq	%r9,%r13
+	xorq	%r11,%r12
+	rorq	$4,%r13
+	xorq	%rbx,%r14
+	andq	%r9,%r12
+	xorq	%r9,%r13
+	addq	120(%rsp),%rax
+	movq	%rbx,%rdi
+	xorq	%r11,%r12
+	rorq	$6,%r14
+	xorq	%rcx,%rdi
+	addq	%r12,%rax
+	rorq	$14,%r13
+	andq	%rdi,%r15
+	xorq	%rbx,%r14
+	addq	%r13,%rax
+	xorq	%rcx,%r15
+	rorq	$28,%r14
+	addq	%rax,%r8
+	addq	%r15,%rax
+	movq	%r8,%r13
+	addq	%rax,%r14
+	movq	128+0(%rsp),%rdi
+	movq	%r14,%rax
+
+	addq	0(%rdi),%rax
+	leaq	128(%rsi),%rsi
+	addq	8(%rdi),%rbx
+	addq	16(%rdi),%rcx
+	addq	24(%rdi),%rdx
+	addq	32(%rdi),%r8
+	addq	40(%rdi),%r9
+	addq	48(%rdi),%r10
+	addq	56(%rdi),%r11
+
+	cmpq	128+16(%rsp),%rsi
+
+	movq	%rax,0(%rdi)
+	movq	%rbx,8(%rdi)
+	movq	%rcx,16(%rdi)
+	movq	%rdx,24(%rdi)
+	movq	%r8,32(%rdi)
+	movq	%r9,40(%rdi)
+	movq	%r10,48(%rdi)
+	movq	%r11,56(%rdi)
+	jb	L$loop_xop
+
+	movq	128+24(%rsp),%rsi
+	vzeroupper
+	movq	(%rsi),%r15
+	movq	8(%rsi),%r14
+	movq	16(%rsi),%r13
+	movq	24(%rsi),%r12
+	movq	32(%rsi),%rbp
+	movq	40(%rsi),%rbx
+	leaq	48(%rsi),%rsp
+L$epilogue_xop:
+	.byte	0xf3,0xc3
+
+
+.p2align	6
+sha512_block_data_order_avx:
+L$avx_shortcut:
+	pushq	%rbx
+	pushq	%rbp
+	pushq	%r12
+	pushq	%r13
+	pushq	%r14
+	pushq	%r15
+	movq	%rsp,%r11
+	shlq	$4,%rdx
+	subq	$160,%rsp
+	leaq	(%rsi,%rdx,8),%rdx
+	andq	$-64,%rsp
+	movq	%rdi,128+0(%rsp)
+	movq	%rsi,128+8(%rsp)
+	movq	%rdx,128+16(%rsp)
+	movq	%r11,128+24(%rsp)
+L$prologue_avx:
+
+	vzeroupper
+	movq	0(%rdi),%rax
+	movq	8(%rdi),%rbx
+	movq	16(%rdi),%rcx
+	movq	24(%rdi),%rdx
+	movq	32(%rdi),%r8
+	movq	40(%rdi),%r9
+	movq	48(%rdi),%r10
+	movq	56(%rdi),%r11
+	jmp	L$loop_avx
+.p2align	4
+L$loop_avx:
+	vmovdqa	K512+1280(%rip),%xmm11
+	vmovdqu	0(%rsi),%xmm0
+	leaq	K512+128(%rip),%rbp
+	vmovdqu	16(%rsi),%xmm1
+	vmovdqu	32(%rsi),%xmm2
+	vpshufb	%xmm11,%xmm0,%xmm0
+	vmovdqu	48(%rsi),%xmm3
+	vpshufb	%xmm11,%xmm1,%xmm1
+	vmovdqu	64(%rsi),%xmm4
+	vpshufb	%xmm11,%xmm2,%xmm2
+	vmovdqu	80(%rsi),%xmm5
+	vpshufb	%xmm11,%xmm3,%xmm3
+	vmovdqu	96(%rsi),%xmm6
+	vpshufb	%xmm11,%xmm4,%xmm4
+	vmovdqu	112(%rsi),%xmm7
+	vpshufb	%xmm11,%xmm5,%xmm5
+	vpaddq	-128(%rbp),%xmm0,%xmm8
+	vpshufb	%xmm11,%xmm6,%xmm6
+	vpaddq	-96(%rbp),%xmm1,%xmm9
+	vpshufb	%xmm11,%xmm7,%xmm7
+	vpaddq	-64(%rbp),%xmm2,%xmm10
+	vpaddq	-32(%rbp),%xmm3,%xmm11
+	vmovdqa	%xmm8,0(%rsp)
+	vpaddq	0(%rbp),%xmm4,%xmm8
+	vmovdqa	%xmm9,16(%rsp)
+	vpaddq	32(%rbp),%xmm5,%xmm9
+	vmovdqa	%xmm10,32(%rsp)
+	vpaddq	64(%rbp),%xmm6,%xmm10
+	vmovdqa	%xmm11,48(%rsp)
+	vpaddq	96(%rbp),%xmm7,%xmm11
+	vmovdqa	%xmm8,64(%rsp)
+	movq	%rax,%r14
+	vmovdqa	%xmm9,80(%rsp)
+	movq	%rbx,%rdi
+	vmovdqa	%xmm10,96(%rsp)
+	xorq	%rcx,%rdi
+	vmovdqa	%xmm11,112(%rsp)
+	movq	%r8,%r13
+	jmp	L$avx_00_47
+
+.p2align	4
+L$avx_00_47:
+	addq	$256,%rbp
+	vpalignr	$8,%xmm0,%xmm1,%xmm8
+	shrdq	$23,%r13,%r13
+	movq	%r14,%rax
+	vpalignr	$8,%xmm4,%xmm5,%xmm11
+	movq	%r9,%r12
+	shrdq	$5,%r14,%r14
+	vpsrlq	$1,%xmm8,%xmm10
+	xorq	%r8,%r13
+	xorq	%r10,%r12
+	vpaddq	%xmm11,%xmm0,%xmm0
+	shrdq	$4,%r13,%r13
+	xorq	%rax,%r14
+	vpsrlq	$7,%xmm8,%xmm11
+	andq	%r8,%r12
+	xorq	%r8,%r13
+	vpsllq	$56,%xmm8,%xmm9
+	addq	0(%rsp),%r11
+	movq	%rax,%r15
+	vpxor	%xmm10,%xmm11,%xmm8
+	xorq	%r10,%r12
+	shrdq	$6,%r14,%r14
+	vpsrlq	$7,%xmm10,%xmm10
+	xorq	%rbx,%r15
+	addq	%r12,%r11
+	vpxor	%xmm9,%xmm8,%xmm8
+	shrdq	$14,%r13,%r13
+	andq	%r15,%rdi
+	vpsllq	$7,%xmm9,%xmm9
+	xorq	%rax,%r14
+	addq	%r13,%r11
+	vpxor	%xmm10,%xmm8,%xmm8
+	xorq	%rbx,%rdi
+	shrdq	$28,%r14,%r14
+	vpsrlq	$6,%xmm7,%xmm11
+	addq	%r11,%rdx
+	addq	%rdi,%r11
+	vpxor	%xmm9,%xmm8,%xmm8
+	movq	%rdx,%r13
+	addq	%r11,%r14
+	vpsllq	$3,%xmm7,%xmm10
+	shrdq	$23,%r13,%r13
+	movq	%r14,%r11
+	vpaddq	%xmm8,%xmm0,%xmm0
+	movq	%r8,%r12
+	shrdq	$5,%r14,%r14
+	vpsrlq	$19,%xmm7,%xmm9
+	xorq	%rdx,%r13
+	xorq	%r9,%r12
+	vpxor	%xmm10,%xmm11,%xmm11
+	shrdq	$4,%r13,%r13
+	xorq	%r11,%r14
+	vpsllq	$42,%xmm10,%xmm10
+	andq	%rdx,%r12
+	xorq	%rdx,%r13
+	vpxor	%xmm9,%xmm11,%xmm11
+	addq	8(%rsp),%r10
+	movq	%r11,%rdi
+	vpsrlq	$42,%xmm9,%xmm9
+	xorq	%r9,%r12
+	shrdq	$6,%r14,%r14
+	vpxor	%xmm10,%xmm11,%xmm11
+	xorq	%rax,%rdi
+	addq	%r12,%r10
+	vpxor	%xmm9,%xmm11,%xmm11
+	shrdq	$14,%r13,%r13
+	andq	%rdi,%r15
+	vpaddq	%xmm11,%xmm0,%xmm0
+	xorq	%r11,%r14
+	addq	%r13,%r10
+	vpaddq	-128(%rbp),%xmm0,%xmm10
+	xorq	%rax,%r15
+	shrdq	$28,%r14,%r14
+	addq	%r10,%rcx
+	addq	%r15,%r10
+	movq	%rcx,%r13
+	addq	%r10,%r14
+	vmovdqa	%xmm10,0(%rsp)
+	vpalignr	$8,%xmm1,%xmm2,%xmm8
+	shrdq	$23,%r13,%r13
+	movq	%r14,%r10
+	vpalignr	$8,%xmm5,%xmm6,%xmm11
+	movq	%rdx,%r12
+	shrdq	$5,%r14,%r14
+	vpsrlq	$1,%xmm8,%xmm10
+	xorq	%rcx,%r13
+	xorq	%r8,%r12
+	vpaddq	%xmm11,%xmm1,%xmm1
+	shrdq	$4,%r13,%r13
+	xorq	%r10,%r14
+	vpsrlq	$7,%xmm8,%xmm11
+	andq	%rcx,%r12
+	xorq	%rcx,%r13
+	vpsllq	$56,%xmm8,%xmm9
+	addq	16(%rsp),%r9
+	movq	%r10,%r15
+	vpxor	%xmm10,%xmm11,%xmm8
+	xorq	%r8,%r12
+	shrdq	$6,%r14,%r14
+	vpsrlq	$7,%xmm10,%xmm10
+	xorq	%r11,%r15
+	addq	%r12,%r9
+	vpxor	%xmm9,%xmm8,%xmm8
+	shrdq	$14,%r13,%r13
+	andq	%r15,%rdi
+	vpsllq	$7,%xmm9,%xmm9
+	xorq	%r10,%r14
+	addq	%r13,%r9
+	vpxor	%xmm10,%xmm8,%xmm8
+	xorq	%r11,%rdi
+	shrdq	$28,%r14,%r14
+	vpsrlq	$6,%xmm0,%xmm11
+	addq	%r9,%rbx
+	addq	%rdi,%r9
+	vpxor	%xmm9,%xmm8,%xmm8
+	movq	%rbx,%r13
+	addq	%r9,%r14
+	vpsllq	$3,%xmm0,%xmm10
+	shrdq	$23,%r13,%r13
+	movq	%r14,%r9
+	vpaddq	%xmm8,%xmm1,%xmm1
+	movq	%rcx,%r12
+	shrdq	$5,%r14,%r14
+	vpsrlq	$19,%xmm0,%xmm9
+	xorq	%rbx,%r13
+	xorq	%rdx,%r12
+	vpxor	%xmm10,%xmm11,%xmm11
+	shrdq	$4,%r13,%r13
+	xorq	%r9,%r14
+	vpsllq	$42,%xmm10,%xmm10
+	andq	%rbx,%r12
+	xorq	%rbx,%r13
+	vpxor	%xmm9,%xmm11,%xmm11
+	addq	24(%rsp),%r8
+	movq	%r9,%rdi
+	vpsrlq	$42,%xmm9,%xmm9
+	xorq	%rdx,%r12
+	shrdq	$6,%r14,%r14
+	vpxor	%xmm10,%xmm11,%xmm11
+	xorq	%r10,%rdi
+	addq	%r12,%r8
+	vpxor	%xmm9,%xmm11,%xmm11
+	shrdq	$14,%r13,%r13
+	andq	%rdi,%r15
+	vpaddq	%xmm11,%xmm1,%xmm1
+	xorq	%r9,%r14
+	addq	%r13,%r8
+	vpaddq	-96(%rbp),%xmm1,%xmm10
+	xorq	%r10,%r15
+	shrdq	$28,%r14,%r14
+	addq	%r8,%rax
+	addq	%r15,%r8
+	movq	%rax,%r13
+	addq	%r8,%r14
+	vmovdqa	%xmm10,16(%rsp)
+	vpalignr	$8,%xmm2,%xmm3,%xmm8
+	shrdq	$23,%r13,%r13
+	movq	%r14,%r8
+	vpalignr	$8,%xmm6,%xmm7,%xmm11
+	movq	%rbx,%r12
+	shrdq	$5,%r14,%r14
+	vpsrlq	$1,%xmm8,%xmm10
+	xorq	%rax,%r13
+	xorq	%rcx,%r12
+	vpaddq	%xmm11,%xmm2,%xmm2
+	shrdq	$4,%r13,%r13
+	xorq	%r8,%r14
+	vpsrlq	$7,%xmm8,%xmm11
+	andq	%rax,%r12
+	xorq	%rax,%r13
+	vpsllq	$56,%xmm8,%xmm9
+	addq	32(%rsp),%rdx
+	movq	%r8,%r15
+	vpxor	%xmm10,%xmm11,%xmm8
+	xorq	%rcx,%r12
+	shrdq	$6,%r14,%r14
+	vpsrlq	$7,%xmm10,%xmm10
+	xorq	%r9,%r15
+	addq	%r12,%rdx
+	vpxor	%xmm9,%xmm8,%xmm8
+	shrdq	$14,%r13,%r13
+	andq	%r15,%rdi
+	vpsllq	$7,%xmm9,%xmm9
+	xorq	%r8,%r14
+	addq	%r13,%rdx
+	vpxor	%xmm10,%xmm8,%xmm8
+	xorq	%r9,%rdi
+	shrdq	$28,%r14,%r14
+	vpsrlq	$6,%xmm1,%xmm11
+	addq	%rdx,%r11
+	addq	%rdi,%rdx
+	vpxor	%xmm9,%xmm8,%xmm8
+	movq	%r11,%r13
+	addq	%rdx,%r14
+	vpsllq	$3,%xmm1,%xmm10
+	shrdq	$23,%r13,%r13
+	movq	%r14,%rdx
+	vpaddq	%xmm8,%xmm2,%xmm2
+	movq	%rax,%r12
+	shrdq	$5,%r14,%r14
+	vpsrlq	$19,%xmm1,%xmm9
+	xorq	%r11,%r13
+	xorq	%rbx,%r12
+	vpxor	%xmm10,%xmm11,%xmm11
+	shrdq	$4,%r13,%r13
+	xorq	%rdx,%r14
+	vpsllq	$42,%xmm10,%xmm10
+	andq	%r11,%r12
+	xorq	%r11,%r13
+	vpxor	%xmm9,%xmm11,%xmm11
+	addq	40(%rsp),%rcx
+	movq	%rdx,%rdi
+	vpsrlq	$42,%xmm9,%xmm9
+	xorq	%rbx,%r12
+	shrdq	$6,%r14,%r14
+	vpxor	%xmm10,%xmm11,%xmm11
+	xorq	%r8,%rdi
+	addq	%r12,%rcx
+	vpxor	%xmm9,%xmm11,%xmm11
+	shrdq	$14,%r13,%r13
+	andq	%rdi,%r15
+	vpaddq	%xmm11,%xmm2,%xmm2
+	xorq	%rdx,%r14
+	addq	%r13,%rcx
+	vpaddq	-64(%rbp),%xmm2,%xmm10
+	xorq	%r8,%r15
+	shrdq	$28,%r14,%r14
+	addq	%rcx,%r10
+	addq	%r15,%rcx
+	movq	%r10,%r13
+	addq	%rcx,%r14
+	vmovdqa	%xmm10,32(%rsp)
+	vpalignr	$8,%xmm3,%xmm4,%xmm8
+	shrdq	$23,%r13,%r13
+	movq	%r14,%rcx
+	vpalignr	$8,%xmm7,%xmm0,%xmm11
+	movq	%r11,%r12
+	shrdq	$5,%r14,%r14
+	vpsrlq	$1,%xmm8,%xmm10
+	xorq	%r10,%r13
+	xorq	%rax,%r12
+	vpaddq	%xmm11,%xmm3,%xmm3
+	shrdq	$4,%r13,%r13
+	xorq	%rcx,%r14
+	vpsrlq	$7,%xmm8,%xmm11
+	andq	%r10,%r12
+	xorq	%r10,%r13
+	vpsllq	$56,%xmm8,%xmm9
+	addq	48(%rsp),%rbx
+	movq	%rcx,%r15
+	vpxor	%xmm10,%xmm11,%xmm8
+	xorq	%rax,%r12
+	shrdq	$6,%r14,%r14
+	vpsrlq	$7,%xmm10,%xmm10
+	xorq	%rdx,%r15
+	addq	%r12,%rbx
+	vpxor	%xmm9,%xmm8,%xmm8
+	shrdq	$14,%r13,%r13
+	andq	%r15,%rdi
+	vpsllq	$7,%xmm9,%xmm9
+	xorq	%rcx,%r14
+	addq	%r13,%rbx
+	vpxor	%xmm10,%xmm8,%xmm8
+	xorq	%rdx,%rdi
+	shrdq	$28,%r14,%r14
+	vpsrlq	$6,%xmm2,%xmm11
+	addq	%rbx,%r9
+	addq	%rdi,%rbx
+	vpxor	%xmm9,%xmm8,%xmm8
+	movq	%r9,%r13
+	addq	%rbx,%r14
+	vpsllq	$3,%xmm2,%xmm10
+	shrdq	$23,%r13,%r13
+	movq	%r14,%rbx
+	vpaddq	%xmm8,%xmm3,%xmm3
+	movq	%r10,%r12
+	shrdq	$5,%r14,%r14
+	vpsrlq	$19,%xmm2,%xmm9
+	xorq	%r9,%r13
+	xorq	%r11,%r12
+	vpxor	%xmm10,%xmm11,%xmm11
+	shrdq	$4,%r13,%r13
+	xorq	%rbx,%r14
+	vpsllq	$42,%xmm10,%xmm10
+	andq	%r9,%r12
+	xorq	%r9,%r13
+	vpxor	%xmm9,%xmm11,%xmm11
+	addq	56(%rsp),%rax
+	movq	%rbx,%rdi
+	vpsrlq	$42,%xmm9,%xmm9
+	xorq	%r11,%r12
+	shrdq	$6,%r14,%r14
+	vpxor	%xmm10,%xmm11,%xmm11
+	xorq	%rcx,%rdi
+	addq	%r12,%rax
+	vpxor	%xmm9,%xmm11,%xmm11
+	shrdq	$14,%r13,%r13
+	andq	%rdi,%r15
+	vpaddq	%xmm11,%xmm3,%xmm3
+	xorq	%rbx,%r14
+	addq	%r13,%rax
+	vpaddq	-32(%rbp),%xmm3,%xmm10
+	xorq	%rcx,%r15
+	shrdq	$28,%r14,%r14
+	addq	%rax,%r8
+	addq	%r15,%rax
+	movq	%r8,%r13
+	addq	%rax,%r14
+	vmovdqa	%xmm10,48(%rsp)
+	vpalignr	$8,%xmm4,%xmm5,%xmm8
+	shrdq	$23,%r13,%r13
+	movq	%r14,%rax
+	vpalignr	$8,%xmm0,%xmm1,%xmm11
+	movq	%r9,%r12
+	shrdq	$5,%r14,%r14
+	vpsrlq	$1,%xmm8,%xmm10
+	xorq	%r8,%r13
+	xorq	%r10,%r12
+	vpaddq	%xmm11,%xmm4,%xmm4
+	shrdq	$4,%r13,%r13
+	xorq	%rax,%r14
+	vpsrlq	$7,%xmm8,%xmm11
+	andq	%r8,%r12
+	xorq	%r8,%r13
+	vpsllq	$56,%xmm8,%xmm9
+	addq	64(%rsp),%r11
+	movq	%rax,%r15
+	vpxor	%xmm10,%xmm11,%xmm8
+	xorq	%r10,%r12
+	shrdq	$6,%r14,%r14
+	vpsrlq	$7,%xmm10,%xmm10
+	xorq	%rbx,%r15
+	addq	%r12,%r11
+	vpxor	%xmm9,%xmm8,%xmm8
+	shrdq	$14,%r13,%r13
+	andq	%r15,%rdi
+	vpsllq	$7,%xmm9,%xmm9
+	xorq	%rax,%r14
+	addq	%r13,%r11
+	vpxor	%xmm10,%xmm8,%xmm8
+	xorq	%rbx,%rdi
+	shrdq	$28,%r14,%r14
+	vpsrlq	$6,%xmm3,%xmm11
+	addq	%r11,%rdx
+	addq	%rdi,%r11
+	vpxor	%xmm9,%xmm8,%xmm8
+	movq	%rdx,%r13
+	addq	%r11,%r14
+	vpsllq	$3,%xmm3,%xmm10
+	shrdq	$23,%r13,%r13
+	movq	%r14,%r11
+	vpaddq	%xmm8,%xmm4,%xmm4
+	movq	%r8,%r12
+	shrdq	$5,%r14,%r14
+	vpsrlq	$19,%xmm3,%xmm9
+	xorq	%rdx,%r13
+	xorq	%r9,%r12
+	vpxor	%xmm10,%xmm11,%xmm11
+	shrdq	$4,%r13,%r13
+	xorq	%r11,%r14
+	vpsllq	$42,%xmm10,%xmm10
+	andq	%rdx,%r12
+	xorq	%rdx,%r13
+	vpxor	%xmm9,%xmm11,%xmm11
+	addq	72(%rsp),%r10
+	movq	%r11,%rdi
+	vpsrlq	$42,%xmm9,%xmm9
+	xorq	%r9,%r12
+	shrdq	$6,%r14,%r14
+	vpxor	%xmm10,%xmm11,%xmm11
+	xorq	%rax,%rdi
+	addq	%r12,%r10
+	vpxor	%xmm9,%xmm11,%xmm11
+	shrdq	$14,%r13,%r13
+	andq	%rdi,%r15
+	vpaddq	%xmm11,%xmm4,%xmm4
+	xorq	%r11,%r14
+	addq	%r13,%r10
+	vpaddq	0(%rbp),%xmm4,%xmm10
+	xorq	%rax,%r15
+	shrdq	$28,%r14,%r14
+	addq	%r10,%rcx
+	addq	%r15,%r10
+	movq	%rcx,%r13
+	addq	%r10,%r14
+	vmovdqa	%xmm10,64(%rsp)
+	vpalignr	$8,%xmm5,%xmm6,%xmm8
+	shrdq	$23,%r13,%r13
+	movq	%r14,%r10
+	vpalignr	$8,%xmm1,%xmm2,%xmm11
+	movq	%rdx,%r12
+	shrdq	$5,%r14,%r14
+	vpsrlq	$1,%xmm8,%xmm10
+	xorq	%rcx,%r13
+	xorq	%r8,%r12
+	vpaddq	%xmm11,%xmm5,%xmm5
+	shrdq	$4,%r13,%r13
+	xorq	%r10,%r14
+	vpsrlq	$7,%xmm8,%xmm11
+	andq	%rcx,%r12
+	xorq	%rcx,%r13
+	vpsllq	$56,%xmm8,%xmm9
+	addq	80(%rsp),%r9
+	movq	%r10,%r15
+	vpxor	%xmm10,%xmm11,%xmm8
+	xorq	%r8,%r12
+	shrdq	$6,%r14,%r14
+	vpsrlq	$7,%xmm10,%xmm10
+	xorq	%r11,%r15
+	addq	%r12,%r9
+	vpxor	%xmm9,%xmm8,%xmm8
+	shrdq	$14,%r13,%r13
+	andq	%r15,%rdi
+	vpsllq	$7,%xmm9,%xmm9
+	xorq	%r10,%r14
+	addq	%r13,%r9
+	vpxor	%xmm10,%xmm8,%xmm8
+	xorq	%r11,%rdi
+	shrdq	$28,%r14,%r14
+	vpsrlq	$6,%xmm4,%xmm11
+	addq	%r9,%rbx
+	addq	%rdi,%r9
+	vpxor	%xmm9,%xmm8,%xmm8
+	movq	%rbx,%r13
+	addq	%r9,%r14
+	vpsllq	$3,%xmm4,%xmm10
+	shrdq	$23,%r13,%r13
+	movq	%r14,%r9
+	vpaddq	%xmm8,%xmm5,%xmm5
+	movq	%rcx,%r12
+	shrdq	$5,%r14,%r14
+	vpsrlq	$19,%xmm4,%xmm9
+	xorq	%rbx,%r13
+	xorq	%rdx,%r12
+	vpxor	%xmm10,%xmm11,%xmm11
+	shrdq	$4,%r13,%r13
+	xorq	%r9,%r14
+	vpsllq	$42,%xmm10,%xmm10
+	andq	%rbx,%r12
+	xorq	%rbx,%r13
+	vpxor	%xmm9,%xmm11,%xmm11
+	addq	88(%rsp),%r8
+	movq	%r9,%rdi
+	vpsrlq	$42,%xmm9,%xmm9
+	xorq	%rdx,%r12
+	shrdq	$6,%r14,%r14
+	vpxor	%xmm10,%xmm11,%xmm11
+	xorq	%r10,%rdi
+	addq	%r12,%r8
+	vpxor	%xmm9,%xmm11,%xmm11
+	shrdq	$14,%r13,%r13
+	andq	%rdi,%r15
+	vpaddq	%xmm11,%xmm5,%xmm5
+	xorq	%r9,%r14
+	addq	%r13,%r8
+	vpaddq	32(%rbp),%xmm5,%xmm10
+	xorq	%r10,%r15
+	shrdq	$28,%r14,%r14
+	addq	%r8,%rax
+	addq	%r15,%r8
+	movq	%rax,%r13
+	addq	%r8,%r14
+	vmovdqa	%xmm10,80(%rsp)
+	vpalignr	$8,%xmm6,%xmm7,%xmm8
+	shrdq	$23,%r13,%r13
+	movq	%r14,%r8
+	vpalignr	$8,%xmm2,%xmm3,%xmm11
+	movq	%rbx,%r12
+	shrdq	$5,%r14,%r14
+	vpsrlq	$1,%xmm8,%xmm10
+	xorq	%rax,%r13
+	xorq	%rcx,%r12
+	vpaddq	%xmm11,%xmm6,%xmm6
+	shrdq	$4,%r13,%r13
+	xorq	%r8,%r14
+	vpsrlq	$7,%xmm8,%xmm11
+	andq	%rax,%r12
+	xorq	%rax,%r13
+	vpsllq	$56,%xmm8,%xmm9
+	addq	96(%rsp),%rdx
+	movq	%r8,%r15
+	vpxor	%xmm10,%xmm11,%xmm8
+	xorq	%rcx,%r12
+	shrdq	$6,%r14,%r14
+	vpsrlq	$7,%xmm10,%xmm10
+	xorq	%r9,%r15
+	addq	%r12,%rdx
+	vpxor	%xmm9,%xmm8,%xmm8
+	shrdq	$14,%r13,%r13
+	andq	%r15,%rdi
+	vpsllq	$7,%xmm9,%xmm9
+	xorq	%r8,%r14
+	addq	%r13,%rdx
+	vpxor	%xmm10,%xmm8,%xmm8
+	xorq	%r9,%rdi
+	shrdq	$28,%r14,%r14
+	vpsrlq	$6,%xmm5,%xmm11
+	addq	%rdx,%r11
+	addq	%rdi,%rdx
+	vpxor	%xmm9,%xmm8,%xmm8
+	movq	%r11,%r13
+	addq	%rdx,%r14
+	vpsllq	$3,%xmm5,%xmm10
+	shrdq	$23,%r13,%r13
+	movq	%r14,%rdx
+	vpaddq	%xmm8,%xmm6,%xmm6
+	movq	%rax,%r12
+	shrdq	$5,%r14,%r14
+	vpsrlq	$19,%xmm5,%xmm9
+	xorq	%r11,%r13
+	xorq	%rbx,%r12
+	vpxor	%xmm10,%xmm11,%xmm11
+	shrdq	$4,%r13,%r13
+	xorq	%rdx,%r14
+	vpsllq	$42,%xmm10,%xmm10
+	andq	%r11,%r12
+	xorq	%r11,%r13
+	vpxor	%xmm9,%xmm11,%xmm11
+	addq	104(%rsp),%rcx
+	movq	%rdx,%rdi
+	vpsrlq	$42,%xmm9,%xmm9
+	xorq	%rbx,%r12
+	shrdq	$6,%r14,%r14
+	vpxor	%xmm10,%xmm11,%xmm11
+	xorq	%r8,%rdi
+	addq	%r12,%rcx
+	vpxor	%xmm9,%xmm11,%xmm11
+	shrdq	$14,%r13,%r13
+	andq	%rdi,%r15
+	vpaddq	%xmm11,%xmm6,%xmm6
+	xorq	%rdx,%r14
+	addq	%r13,%rcx
+	vpaddq	64(%rbp),%xmm6,%xmm10
+	xorq	%r8,%r15
+	shrdq	$28,%r14,%r14
+	addq	%rcx,%r10
+	addq	%r15,%rcx
+	movq	%r10,%r13
+	addq	%rcx,%r14
+	vmovdqa	%xmm10,96(%rsp)
+	vpalignr	$8,%xmm7,%xmm0,%xmm8
+	shrdq	$23,%r13,%r13
+	movq	%r14,%rcx
+	vpalignr	$8,%xmm3,%xmm4,%xmm11
+	movq	%r11,%r12
+	shrdq	$5,%r14,%r14
+	vpsrlq	$1,%xmm8,%xmm10
+	xorq	%r10,%r13
+	xorq	%rax,%r12
+	vpaddq	%xmm11,%xmm7,%xmm7
+	shrdq	$4,%r13,%r13
+	xorq	%rcx,%r14
+	vpsrlq	$7,%xmm8,%xmm11
+	andq	%r10,%r12
+	xorq	%r10,%r13
+	vpsllq	$56,%xmm8,%xmm9
+	addq	112(%rsp),%rbx
+	movq	%rcx,%r15
+	vpxor	%xmm10,%xmm11,%xmm8
+	xorq	%rax,%r12
+	shrdq	$6,%r14,%r14
+	vpsrlq	$7,%xmm10,%xmm10
+	xorq	%rdx,%r15
+	addq	%r12,%rbx
+	vpxor	%xmm9,%xmm8,%xmm8
+	shrdq	$14,%r13,%r13
+	andq	%r15,%rdi
+	vpsllq	$7,%xmm9,%xmm9
+	xorq	%rcx,%r14
+	addq	%r13,%rbx
+	vpxor	%xmm10,%xmm8,%xmm8
+	xorq	%rdx,%rdi
+	shrdq	$28,%r14,%r14
+	vpsrlq	$6,%xmm6,%xmm11
+	addq	%rbx,%r9
+	addq	%rdi,%rbx
+	vpxor	%xmm9,%xmm8,%xmm8
+	movq	%r9,%r13
+	addq	%rbx,%r14
+	vpsllq	$3,%xmm6,%xmm10
+	shrdq	$23,%r13,%r13
+	movq	%r14,%rbx
+	vpaddq	%xmm8,%xmm7,%xmm7
+	movq	%r10,%r12
+	shrdq	$5,%r14,%r14
+	vpsrlq	$19,%xmm6,%xmm9
+	xorq	%r9,%r13
+	xorq	%r11,%r12
+	vpxor	%xmm10,%xmm11,%xmm11
+	shrdq	$4,%r13,%r13
+	xorq	%rbx,%r14
+	vpsllq	$42,%xmm10,%xmm10
+	andq	%r9,%r12
+	xorq	%r9,%r13
+	vpxor	%xmm9,%xmm11,%xmm11
+	addq	120(%rsp),%rax
+	movq	%rbx,%rdi
+	vpsrlq	$42,%xmm9,%xmm9
+	xorq	%r11,%r12
+	shrdq	$6,%r14,%r14
+	vpxor	%xmm10,%xmm11,%xmm11
+	xorq	%rcx,%rdi
+	addq	%r12,%rax
+	vpxor	%xmm9,%xmm11,%xmm11
+	shrdq	$14,%r13,%r13
+	andq	%rdi,%r15
+	vpaddq	%xmm11,%xmm7,%xmm7
+	xorq	%rbx,%r14
+	addq	%r13,%rax
+	vpaddq	96(%rbp),%xmm7,%xmm10
+	xorq	%rcx,%r15
+	shrdq	$28,%r14,%r14
+	addq	%rax,%r8
+	addq	%r15,%rax
+	movq	%r8,%r13
+	addq	%rax,%r14
+	vmovdqa	%xmm10,112(%rsp)
+	cmpb	$0,135(%rbp)
+	jne	L$avx_00_47
+	shrdq	$23,%r13,%r13
+	movq	%r14,%rax
+	movq	%r9,%r12
+	shrdq	$5,%r14,%r14
+	xorq	%r8,%r13
+	xorq	%r10,%r12
+	shrdq	$4,%r13,%r13
+	xorq	%rax,%r14
+	andq	%r8,%r12
+	xorq	%r8,%r13
+	addq	0(%rsp),%r11
+	movq	%rax,%r15
+	xorq	%r10,%r12
+	shrdq	$6,%r14,%r14
+	xorq	%rbx,%r15
+	addq	%r12,%r11
+	shrdq	$14,%r13,%r13
+	andq	%r15,%rdi
+	xorq	%rax,%r14
+	addq	%r13,%r11
+	xorq	%rbx,%rdi
+	shrdq	$28,%r14,%r14
+	addq	%r11,%rdx
+	addq	%rdi,%r11
+	movq	%rdx,%r13
+	addq	%r11,%r14
+	shrdq	$23,%r13,%r13
+	movq	%r14,%r11
+	movq	%r8,%r12
+	shrdq	$5,%r14,%r14
+	xorq	%rdx,%r13
+	xorq	%r9,%r12
+	shrdq	$4,%r13,%r13
+	xorq	%r11,%r14
+	andq	%rdx,%r12
+	xorq	%rdx,%r13
+	addq	8(%rsp),%r10
+	movq	%r11,%rdi
+	xorq	%r9,%r12
+	shrdq	$6,%r14,%r14
+	xorq	%rax,%rdi
+	addq	%r12,%r10
+	shrdq	$14,%r13,%r13
+	andq	%rdi,%r15
+	xorq	%r11,%r14
+	addq	%r13,%r10
+	xorq	%rax,%r15
+	shrdq	$28,%r14,%r14
+	addq	%r10,%rcx
+	addq	%r15,%r10
+	movq	%rcx,%r13
+	addq	%r10,%r14
+	shrdq	$23,%r13,%r13
+	movq	%r14,%r10
+	movq	%rdx,%r12
+	shrdq	$5,%r14,%r14
+	xorq	%rcx,%r13
+	xorq	%r8,%r12
+	shrdq	$4,%r13,%r13
+	xorq	%r10,%r14
+	andq	%rcx,%r12
+	xorq	%rcx,%r13
+	addq	16(%rsp),%r9
+	movq	%r10,%r15
+	xorq	%r8,%r12
+	shrdq	$6,%r14,%r14
+	xorq	%r11,%r15
+	addq	%r12,%r9
+	shrdq	$14,%r13,%r13
+	andq	%r15,%rdi
+	xorq	%r10,%r14
+	addq	%r13,%r9
+	xorq	%r11,%rdi
+	shrdq	$28,%r14,%r14
+	addq	%r9,%rbx
+	addq	%rdi,%r9
+	movq	%rbx,%r13
+	addq	%r9,%r14
+	shrdq	$23,%r13,%r13
+	movq	%r14,%r9
+	movq	%rcx,%r12
+	shrdq	$5,%r14,%r14
+	xorq	%rbx,%r13
+	xorq	%rdx,%r12
+	shrdq	$4,%r13,%r13
+	xorq	%r9,%r14
+	andq	%rbx,%r12
+	xorq	%rbx,%r13
+	addq	24(%rsp),%r8
+	movq	%r9,%rdi
+	xorq	%rdx,%r12
+	shrdq	$6,%r14,%r14
+	xorq	%r10,%rdi
+	addq	%r12,%r8
+	shrdq	$14,%r13,%r13
+	andq	%rdi,%r15
+	xorq	%r9,%r14
+	addq	%r13,%r8
+	xorq	%r10,%r15
+	shrdq	$28,%r14,%r14
+	addq	%r8,%rax
+	addq	%r15,%r8
+	movq	%rax,%r13
+	addq	%r8,%r14
+	shrdq	$23,%r13,%r13
+	movq	%r14,%r8
+	movq	%rbx,%r12
+	shrdq	$5,%r14,%r14
+	xorq	%rax,%r13
+	xorq	%rcx,%r12
+	shrdq	$4,%r13,%r13
+	xorq	%r8,%r14
+	andq	%rax,%r12
+	xorq	%rax,%r13
+	addq	32(%rsp),%rdx
+	movq	%r8,%r15
+	xorq	%rcx,%r12
+	shrdq	$6,%r14,%r14
+	xorq	%r9,%r15
+	addq	%r12,%rdx
+	shrdq	$14,%r13,%r13
+	andq	%r15,%rdi
+	xorq	%r8,%r14
+	addq	%r13,%rdx
+	xorq	%r9,%rdi
+	shrdq	$28,%r14,%r14
+	addq	%rdx,%r11
+	addq	%rdi,%rdx
+	movq	%r11,%r13
+	addq	%rdx,%r14
+	shrdq	$23,%r13,%r13
+	movq	%r14,%rdx
+	movq	%rax,%r12
+	shrdq	$5,%r14,%r14
+	xorq	%r11,%r13
+	xorq	%rbx,%r12
+	shrdq	$4,%r13,%r13
+	xorq	%rdx,%r14
+	andq	%r11,%r12
+	xorq	%r11,%r13
+	addq	40(%rsp),%rcx
+	movq	%rdx,%rdi
+	xorq	%rbx,%r12
+	shrdq	$6,%r14,%r14
+	xorq	%r8,%rdi
+	addq	%r12,%rcx
+	shrdq	$14,%r13,%r13
+	andq	%rdi,%r15
+	xorq	%rdx,%r14
+	addq	%r13,%rcx
+	xorq	%r8,%r15
+	shrdq	$28,%r14,%r14
+	addq	%rcx,%r10
+	addq	%r15,%rcx
+	movq	%r10,%r13
+	addq	%rcx,%r14
+	shrdq	$23,%r13,%r13
+	movq	%r14,%rcx
+	movq	%r11,%r12
+	shrdq	$5,%r14,%r14
+	xorq	%r10,%r13
+	xorq	%rax,%r12
+	shrdq	$4,%r13,%r13
+	xorq	%rcx,%r14
+	andq	%r10,%r12
+	xorq	%r10,%r13
+	addq	48(%rsp),%rbx
+	movq	%rcx,%r15
+	xorq	%rax,%r12
+	shrdq	$6,%r14,%r14
+	xorq	%rdx,%r15
+	addq	%r12,%rbx
+	shrdq	$14,%r13,%r13
+	andq	%r15,%rdi
+	xorq	%rcx,%r14
+	addq	%r13,%rbx
+	xorq	%rdx,%rdi
+	shrdq	$28,%r14,%r14
+	addq	%rbx,%r9
+	addq	%rdi,%rbx
+	movq	%r9,%r13
+	addq	%rbx,%r14
+	shrdq	$23,%r13,%r13
+	movq	%r14,%rbx
+	movq	%r10,%r12
+	shrdq	$5,%r14,%r14
+	xorq	%r9,%r13
+	xorq	%r11,%r12
+	shrdq	$4,%r13,%r13
+	xorq	%rbx,%r14
+	andq	%r9,%r12
+	xorq	%r9,%r13
+	addq	56(%rsp),%rax
+	movq	%rbx,%rdi
+	xorq	%r11,%r12
+	shrdq	$6,%r14,%r14
+	xorq	%rcx,%rdi
+	addq	%r12,%rax
+	shrdq	$14,%r13,%r13
+	andq	%rdi,%r15
+	xorq	%rbx,%r14
+	addq	%r13,%rax
+	xorq	%rcx,%r15
+	shrdq	$28,%r14,%r14
+	addq	%rax,%r8
+	addq	%r15,%rax
+	movq	%r8,%r13
+	addq	%rax,%r14
+	shrdq	$23,%r13,%r13
+	movq	%r14,%rax
+	movq	%r9,%r12
+	shrdq	$5,%r14,%r14
+	xorq	%r8,%r13
+	xorq	%r10,%r12
+	shrdq	$4,%r13,%r13
+	xorq	%rax,%r14
+	andq	%r8,%r12
+	xorq	%r8,%r13
+	addq	64(%rsp),%r11
+	movq	%rax,%r15
+	xorq	%r10,%r12
+	shrdq	$6,%r14,%r14
+	xorq	%rbx,%r15
+	addq	%r12,%r11
+	shrdq	$14,%r13,%r13
+	andq	%r15,%rdi
+	xorq	%rax,%r14
+	addq	%r13,%r11
+	xorq	%rbx,%rdi
+	shrdq	$28,%r14,%r14
+	addq	%r11,%rdx
+	addq	%rdi,%r11
+	movq	%rdx,%r13
+	addq	%r11,%r14
+	shrdq	$23,%r13,%r13
+	movq	%r14,%r11
+	movq	%r8,%r12
+	shrdq	$5,%r14,%r14
+	xorq	%rdx,%r13
+	xorq	%r9,%r12
+	shrdq	$4,%r13,%r13
+	xorq	%r11,%r14
+	andq	%rdx,%r12
+	xorq	%rdx,%r13
+	addq	72(%rsp),%r10
+	movq	%r11,%rdi
+	xorq	%r9,%r12
+	shrdq	$6,%r14,%r14
+	xorq	%rax,%rdi
+	addq	%r12,%r10
+	shrdq	$14,%r13,%r13
+	andq	%rdi,%r15
+	xorq	%r11,%r14
+	addq	%r13,%r10
+	xorq	%rax,%r15
+	shrdq	$28,%r14,%r14
+	addq	%r10,%rcx
+	addq	%r15,%r10
+	movq	%rcx,%r13
+	addq	%r10,%r14
+	shrdq	$23,%r13,%r13
+	movq	%r14,%r10
+	movq	%rdx,%r12
+	shrdq	$5,%r14,%r14
+	xorq	%rcx,%r13
+	xorq	%r8,%r12
+	shrdq	$4,%r13,%r13
+	xorq	%r10,%r14
+	andq	%rcx,%r12
+	xorq	%rcx,%r13
+	addq	80(%rsp),%r9
+	movq	%r10,%r15
+	xorq	%r8,%r12
+	shrdq	$6,%r14,%r14
+	xorq	%r11,%r15
+	addq	%r12,%r9
+	shrdq	$14,%r13,%r13
+	andq	%r15,%rdi
+	xorq	%r10,%r14
+	addq	%r13,%r9
+	xorq	%r11,%rdi
+	shrdq	$28,%r14,%r14
+	addq	%r9,%rbx
+	addq	%rdi,%r9
+	movq	%rbx,%r13
+	addq	%r9,%r14
+	shrdq	$23,%r13,%r13
+	movq	%r14,%r9
+	movq	%rcx,%r12
+	shrdq	$5,%r14,%r14
+	xorq	%rbx,%r13
+	xorq	%rdx,%r12
+	shrdq	$4,%r13,%r13
+	xorq	%r9,%r14
+	andq	%rbx,%r12
+	xorq	%rbx,%r13
+	addq	88(%rsp),%r8
+	movq	%r9,%rdi
+	xorq	%rdx,%r12
+	shrdq	$6,%r14,%r14
+	xorq	%r10,%rdi
+	addq	%r12,%r8
+	shrdq	$14,%r13,%r13
+	andq	%rdi,%r15
+	xorq	%r9,%r14
+	addq	%r13,%r8
+	xorq	%r10,%r15
+	shrdq	$28,%r14,%r14
+	addq	%r8,%rax
+	addq	%r15,%r8
+	movq	%rax,%r13
+	addq	%r8,%r14
+	shrdq	$23,%r13,%r13
+	movq	%r14,%r8
+	movq	%rbx,%r12
+	shrdq	$5,%r14,%r14
+	xorq	%rax,%r13
+	xorq	%rcx,%r12
+	shrdq	$4,%r13,%r13
+	xorq	%r8,%r14
+	andq	%rax,%r12
+	xorq	%rax,%r13
+	addq	96(%rsp),%rdx
+	movq	%r8,%r15
+	xorq	%rcx,%r12
+	shrdq	$6,%r14,%r14
+	xorq	%r9,%r15
+	addq	%r12,%rdx
+	shrdq	$14,%r13,%r13
+	andq	%r15,%rdi
+	xorq	%r8,%r14
+	addq	%r13,%rdx
+	xorq	%r9,%rdi
+	shrdq	$28,%r14,%r14
+	addq	%rdx,%r11
+	addq	%rdi,%rdx
+	movq	%r11,%r13
+	addq	%rdx,%r14
+	shrdq	$23,%r13,%r13
+	movq	%r14,%rdx
+	movq	%rax,%r12
+	shrdq	$5,%r14,%r14
+	xorq	%r11,%r13
+	xorq	%rbx,%r12
+	shrdq	$4,%r13,%r13
+	xorq	%rdx,%r14
+	andq	%r11,%r12
+	xorq	%r11,%r13
+	addq	104(%rsp),%rcx
+	movq	%rdx,%rdi
+	xorq	%rbx,%r12
+	shrdq	$6,%r14,%r14
+	xorq	%r8,%rdi
+	addq	%r12,%rcx
+	shrdq	$14,%r13,%r13
+	andq	%rdi,%r15
+	xorq	%rdx,%r14
+	addq	%r13,%rcx
+	xorq	%r8,%r15
+	shrdq	$28,%r14,%r14
+	addq	%rcx,%r10
+	addq	%r15,%rcx
+	movq	%r10,%r13
+	addq	%rcx,%r14
+	shrdq	$23,%r13,%r13
+	movq	%r14,%rcx
+	movq	%r11,%r12
+	shrdq	$5,%r14,%r14
+	xorq	%r10,%r13
+	xorq	%rax,%r12
+	shrdq	$4,%r13,%r13
+	xorq	%rcx,%r14
+	andq	%r10,%r12
+	xorq	%r10,%r13
+	addq	112(%rsp),%rbx
+	movq	%rcx,%r15
+	xorq	%rax,%r12
+	shrdq	$6,%r14,%r14
+	xorq	%rdx,%r15
+	addq	%r12,%rbx
+	shrdq	$14,%r13,%r13
+	andq	%r15,%rdi
+	xorq	%rcx,%r14
+	addq	%r13,%rbx
+	xorq	%rdx,%rdi
+	shrdq	$28,%r14,%r14
+	addq	%rbx,%r9
+	addq	%rdi,%rbx
+	movq	%r9,%r13
+	addq	%rbx,%r14
+	shrdq	$23,%r13,%r13
+	movq	%r14,%rbx
+	movq	%r10,%r12
+	shrdq	$5,%r14,%r14
+	xorq	%r9,%r13
+	xorq	%r11,%r12
+	shrdq	$4,%r13,%r13
+	xorq	%rbx,%r14
+	andq	%r9,%r12
+	xorq	%r9,%r13
+	addq	120(%rsp),%rax
+	movq	%rbx,%rdi
+	xorq	%r11,%r12
+	shrdq	$6,%r14,%r14
+	xorq	%rcx,%rdi
+	addq	%r12,%rax
+	shrdq	$14,%r13,%r13
+	andq	%rdi,%r15
+	xorq	%rbx,%r14
+	addq	%r13,%rax
+	xorq	%rcx,%r15
+	shrdq	$28,%r14,%r14
+	addq	%rax,%r8
+	addq	%r15,%rax
+	movq	%r8,%r13
+	addq	%rax,%r14
+	movq	128+0(%rsp),%rdi
+	movq	%r14,%rax
+
+	addq	0(%rdi),%rax
+	leaq	128(%rsi),%rsi
+	addq	8(%rdi),%rbx
+	addq	16(%rdi),%rcx
+	addq	24(%rdi),%rdx
+	addq	32(%rdi),%r8
+	addq	40(%rdi),%r9
+	addq	48(%rdi),%r10
+	addq	56(%rdi),%r11
+
+	cmpq	128+16(%rsp),%rsi
+
+	movq	%rax,0(%rdi)
+	movq	%rbx,8(%rdi)
+	movq	%rcx,16(%rdi)
+	movq	%rdx,24(%rdi)
+	movq	%r8,32(%rdi)
+	movq	%r9,40(%rdi)
+	movq	%r10,48(%rdi)
+	movq	%r11,56(%rdi)
+	jb	L$loop_avx
+
+	movq	128+24(%rsp),%rsi
+	vzeroupper
+	movq	(%rsi),%r15
+	movq	8(%rsi),%r14
+	movq	16(%rsi),%r13
+	movq	24(%rsi),%r12
+	movq	32(%rsi),%rbp
+	movq	40(%rsi),%rbx
+	leaq	48(%rsi),%rsp
+L$epilogue_avx:
+	.byte	0xf3,0xc3
+
 #endif

third_party/boringssl/win-x86/crypto/chacha/chacha-x86.asm

diff --git a/third_party/boringssl/win-x86/crypto/chacha/chacha-x86.asm b/third_party/boringssl/win-x86/crypto/chacha/chacha-x86.asm
new file mode 100644
index 0000000..3ba31a2
--- /dev/null
+++ b/third_party/boringssl/win-x86/crypto/chacha/chacha-x86.asm
@@ -0,0 +1,977 @@
+%ifidn __OUTPUT_FORMAT__,obj
+section	code	use32 class=code align=64
+%elifidn __OUTPUT_FORMAT__,win32
+%ifdef __YASM_VERSION_ID__
+%if __YASM_VERSION_ID__ < 01010000h
+%error yasm version 1.1.0 or later needed.
+%endif
+; Yasm automatically includes .00 and complains about redefining it.
+; https://www.tortall.net/projects/yasm/manual/html/objfmt-win32-safeseh.html
+%else
+$@feat.00 equ 1
+%endif
+section	.text	code align=64
+%else
+section	.text	code
+%endif
+global	_ChaCha20_ctr32
+align	16
+_ChaCha20_ctr32:
+L$_ChaCha20_ctr32_begin:
+	push	ebp
+	push	ebx
+	push	esi
+	push	edi
+	xor	eax,eax
+	cmp	eax,DWORD [28+esp]
+	je	NEAR L$000no_data
+	call	L$pic_point
+L$pic_point:
+	pop	eax
+	lea	ebp,[_OPENSSL_ia32cap_P]
+	test	DWORD [ebp],16777216
+	jz	NEAR L$001x86
+	test	DWORD [4+ebp],512
+	jz	NEAR L$001x86
+	jmp	NEAR L$ssse3_shortcut
+L$001x86:
+	mov	esi,DWORD [32+esp]
+	mov	edi,DWORD [36+esp]
+	sub	esp,132
+	mov	eax,DWORD [esi]
+	mov	ebx,DWORD [4+esi]
+	mov	ecx,DWORD [8+esi]
+	mov	edx,DWORD [12+esi]
+	mov	DWORD [80+esp],eax
+	mov	DWORD [84+esp],ebx
+	mov	DWORD [88+esp],ecx
+	mov	DWORD [92+esp],edx
+	mov	eax,DWORD [16+esi]
+	mov	ebx,DWORD [20+esi]
+	mov	ecx,DWORD [24+esi]
+	mov	edx,DWORD [28+esi]
+	mov	DWORD [96+esp],eax
+	mov	DWORD [100+esp],ebx
+	mov	DWORD [104+esp],ecx
+	mov	DWORD [108+esp],edx
+	mov	eax,DWORD [edi]
+	mov	ebx,DWORD [4+edi]
+	mov	ecx,DWORD [8+edi]
+	mov	edx,DWORD [12+edi]
+	sub	eax,1
+	mov	DWORD [112+esp],eax
+	mov	DWORD [116+esp],ebx
+	mov	DWORD [120+esp],ecx
+	mov	DWORD [124+esp],edx
+	jmp	NEAR L$002entry
+align	16
+L$003outer_loop:
+	mov	DWORD [156+esp],ebx
+	mov	DWORD [152+esp],eax
+	mov	DWORD [160+esp],ecx
+L$002entry:
+	mov	eax,1634760805
+	mov	DWORD [4+esp],857760878
+	mov	DWORD [8+esp],2036477234
+	mov	DWORD [12+esp],1797285236
+	mov	ebx,DWORD [84+esp]
+	mov	ebp,DWORD [88+esp]
+	mov	ecx,DWORD [104+esp]
+	mov	esi,DWORD [108+esp]
+	mov	edx,DWORD [116+esp]
+	mov	edi,DWORD [120+esp]
+	mov	DWORD [20+esp],ebx
+	mov	DWORD [24+esp],ebp
+	mov	DWORD [40+esp],ecx
+	mov	DWORD [44+esp],esi
+	mov	DWORD [52+esp],edx
+	mov	DWORD [56+esp],edi
+	mov	ebx,DWORD [92+esp]
+	mov	edi,DWORD [124+esp]
+	mov	edx,DWORD [112+esp]
+	mov	ebp,DWORD [80+esp]
+	mov	ecx,DWORD [96+esp]
+	mov	esi,DWORD [100+esp]
+	add	edx,1
+	mov	DWORD [28+esp],ebx
+	mov	DWORD [60+esp],edi
+	mov	DWORD [112+esp],edx
+	mov	ebx,10
+	jmp	NEAR L$004loop
+align	16
+L$004loop:
+	add	eax,ebp
+	mov	DWORD [128+esp],ebx
+	mov	ebx,ebp
+	xor	edx,eax
+	rol	edx,16
+	add	ecx,edx
+	xor	ebx,ecx
+	mov	edi,DWORD [52+esp]
+	rol	ebx,12
+	mov	ebp,DWORD [20+esp]
+	add	eax,ebx
+	xor	edx,eax
+	mov	DWORD [esp],eax
+	rol	edx,8
+	mov	eax,DWORD [4+esp]
+	add	ecx,edx
+	mov	DWORD [48+esp],edx
+	xor	ebx,ecx
+	add	eax,ebp
+	rol	ebx,7
+	xor	edi,eax
+	mov	DWORD [32+esp],ecx
+	rol	edi,16
+	mov	DWORD [16+esp],ebx
+	add	esi,edi
+	mov	ecx,DWORD [40+esp]
+	xor	ebp,esi
+	mov	edx,DWORD [56+esp]
+	rol	ebp,12
+	mov	ebx,DWORD [24+esp]
+	add	eax,ebp
+	xor	edi,eax
+	mov	DWORD [4+esp],eax
+	rol	edi,8
+	mov	eax,DWORD [8+esp]
+	add	esi,edi
+	mov	DWORD [52+esp],edi
+	xor	ebp,esi
+	add	eax,ebx
+	rol	ebp,7
+	xor	edx,eax
+	mov	DWORD [36+esp],esi
+	rol	edx,16
+	mov	DWORD [20+esp],ebp
+	add	ecx,edx
+	mov	esi,DWORD [44+esp]
+	xor	ebx,ecx
+	mov	edi,DWORD [60+esp]
+	rol	ebx,12
+	mov	ebp,DWORD [28+esp]
+	add	eax,ebx
+	xor	edx,eax
+	mov	DWORD [8+esp],eax
+	rol	edx,8
+	mov	eax,DWORD [12+esp]
+	add	ecx,edx
+	mov	DWORD [56+esp],edx
+	xor	ebx,ecx
+	add	eax,ebp
+	rol	ebx,7
+	xor	edi,eax
+	rol	edi,16
+	mov	DWORD [24+esp],ebx
+	add	esi,edi
+	xor	ebp,esi
+	rol	ebp,12
+	mov	ebx,DWORD [20+esp]
+	add	eax,ebp
+	xor	edi,eax
+	mov	DWORD [12+esp],eax
+	rol	edi,8
+	mov	eax,DWORD [esp]
+	add	esi,edi
+	mov	edx,edi
+	xor	ebp,esi
+	add	eax,ebx
+	rol	ebp,7
+	xor	edx,eax
+	rol	edx,16
+	mov	DWORD [28+esp],ebp
+	add	ecx,edx
+	xor	ebx,ecx
+	mov	edi,DWORD [48+esp]
+	rol	ebx,12
+	mov	ebp,DWORD [24+esp]
+	add	eax,ebx
+	xor	edx,eax
+	mov	DWORD [esp],eax
+	rol	edx,8
+	mov	eax,DWORD [4+esp]
+	add	ecx,edx
+	mov	DWORD [60+esp],edx
+	xor	ebx,ecx
+	add	eax,ebp
+	rol	ebx,7
+	xor	edi,eax
+	mov	DWORD [40+esp],ecx
+	rol	edi,16
+	mov	DWORD [20+esp],ebx
+	add	esi,edi
+	mov	ecx,DWORD [32+esp]
+	xor	ebp,esi
+	mov	edx,DWORD [52+esp]
+	rol	ebp,12
+	mov	ebx,DWORD [28+esp]
+	add	eax,ebp
+	xor	edi,eax
+	mov	DWORD [4+esp],eax
+	rol	edi,8
+	mov	eax,DWORD [8+esp]
+	add	esi,edi
+	mov	DWORD [48+esp],edi
+	xor	ebp,esi
+	add	eax,ebx
+	rol	ebp,7
+	xor	edx,eax
+	mov	DWORD [44+esp],esi
+	rol	edx,16
+	mov	DWORD [24+esp],ebp
+	add	ecx,edx
+	mov	esi,DWORD [36+esp]
+	xor	ebx,ecx
+	mov	edi,DWORD [56+esp]
+	rol	ebx,12
+	mov	ebp,DWORD [16+esp]
+	add	eax,ebx
+	xor	edx,eax
+	mov	DWORD [8+esp],eax
+	rol	edx,8
+	mov	eax,DWORD [12+esp]
+	add	ecx,edx
+	mov	DWORD [52+esp],edx
+	xor	ebx,ecx
+	add	eax,ebp
+	rol	ebx,7
+	xor	edi,eax
+	rol	edi,16
+	mov	DWORD [28+esp],ebx
+	add	esi,edi
+	xor	ebp,esi
+	mov	edx,DWORD [48+esp]
+	rol	ebp,12
+	mov	ebx,DWORD [128+esp]
+	add	eax,ebp
+	xor	edi,eax
+	mov	DWORD [12+esp],eax
+	rol	edi,8
+	mov	eax,DWORD [esp]
+	add	esi,edi
+	mov	DWORD [56+esp],edi
+	xor	ebp,esi
+	rol	ebp,7
+	dec	ebx
+	jnz	NEAR L$004loop
+	mov	ebx,DWORD [160+esp]
+	add	eax,1634760805
+	add	ebp,DWORD [80+esp]
+	add	ecx,DWORD [96+esp]
+	add	esi,DWORD [100+esp]
+	cmp	ebx,64
+	jb	NEAR L$005tail
+	mov	ebx,DWORD [156+esp]
+	add	edx,DWORD [112+esp]
+	add	edi,DWORD [120+esp]
+	xor	eax,DWORD [ebx]
+	xor	ebp,DWORD [16+ebx]
+	mov	DWORD [esp],eax
+	mov	eax,DWORD [152+esp]
+	xor	ecx,DWORD [32+ebx]
+	xor	esi,DWORD [36+ebx]
+	xor	edx,DWORD [48+ebx]
+	xor	edi,DWORD [56+ebx]
+	mov	DWORD [16+eax],ebp
+	mov	DWORD [32+eax],ecx
+	mov	DWORD [36+eax],esi
+	mov	DWORD [48+eax],edx
+	mov	DWORD [56+eax],edi
+	mov	ebp,DWORD [4+esp]
+	mov	ecx,DWORD [8+esp]
+	mov	esi,DWORD [12+esp]
+	mov	edx,DWORD [20+esp]
+	mov	edi,DWORD [24+esp]
+	add	ebp,857760878
+	add	ecx,2036477234
+	add	esi,1797285236
+	add	edx,DWORD [84+esp]
+	add	edi,DWORD [88+esp]
+	xor	ebp,DWORD [4+ebx]
+	xor	ecx,DWORD [8+ebx]
+	xor	esi,DWORD [12+ebx]
+	xor	edx,DWORD [20+ebx]
+	xor	edi,DWORD [24+ebx]
+	mov	DWORD [4+eax],ebp
+	mov	DWORD [8+eax],ecx
+	mov	DWORD [12+eax],esi
+	mov	DWORD [20+eax],edx
+	mov	DWORD [24+eax],edi
+	mov	ebp,DWORD [28+esp]
+	mov	ecx,DWORD [40+esp]
+	mov	esi,DWORD [44+esp]
+	mov	edx,DWORD [52+esp]
+	mov	edi,DWORD [60+esp]
+	add	ebp,DWORD [92+esp]
+	add	ecx,DWORD [104+esp]
+	add	esi,DWORD [108+esp]
+	add	edx,DWORD [116+esp]
+	add	edi,DWORD [124+esp]
+	xor	ebp,DWORD [28+ebx]
+	xor	ecx,DWORD [40+ebx]
+	xor	esi,DWORD [44+ebx]
+	xor	edx,DWORD [52+ebx]
+	xor	edi,DWORD [60+ebx]
+	lea	ebx,[64+ebx]
+	mov	DWORD [28+eax],ebp
+	mov	ebp,DWORD [esp]
+	mov	DWORD [40+eax],ecx
+	mov	ecx,DWORD [160+esp]
+	mov	DWORD [44+eax],esi
+	mov	DWORD [52+eax],edx
+	mov	DWORD [60+eax],edi
+	mov	DWORD [eax],ebp
+	lea	eax,[64+eax]
+	sub	ecx,64
+	jnz	NEAR L$003outer_loop
+	jmp	NEAR L$006done
+L$005tail:
+	add	edx,DWORD [112+esp]
+	add	edi,DWORD [120+esp]
+	mov	DWORD [esp],eax
+	mov	DWORD [16+esp],ebp
+	mov	DWORD [32+esp],ecx
+	mov	DWORD [36+esp],esi
+	mov	DWORD [48+esp],edx
+	mov	DWORD [56+esp],edi
+	mov	ebp,DWORD [4+esp]
+	mov	ecx,DWORD [8+esp]
+	mov	esi,DWORD [12+esp]
+	mov	edx,DWORD [20+esp]
+	mov	edi,DWORD [24+esp]
+	add	ebp,857760878
+	add	ecx,2036477234
+	add	esi,1797285236
+	add	edx,DWORD [84+esp]
+	add	edi,DWORD [88+esp]
+	mov	DWORD [4+esp],ebp
+	mov	DWORD [8+esp],ecx
+	mov	DWORD [12+esp],esi
+	mov	DWORD [20+esp],edx
+	mov	DWORD [24+esp],edi
+	mov	ebp,DWORD [28+esp]
+	mov	ecx,DWORD [40+esp]
+	mov	esi,DWORD [44+esp]
+	mov	edx,DWORD [52+esp]
+	mov	edi,DWORD [60+esp]
+	add	ebp,DWORD [92+esp]
+	add	ecx,DWORD [104+esp]
+	add	esi,DWORD [108+esp]
+	add	edx,DWORD [116+esp]
+	add	edi,DWORD [124+esp]
+	mov	DWORD [28+esp],ebp
+	mov	ebp,DWORD [156+esp]
+	mov	DWORD [40+esp],ecx
+	mov	ecx,DWORD [152+esp]
+	mov	DWORD [44+esp],esi
+	xor	esi,esi
+	mov	DWORD [52+esp],edx
+	mov	DWORD [60+esp],edi
+	xor	eax,eax
+	xor	edx,edx
+L$007tail_loop:
+	mov	al,BYTE [ebp*1+esi]
+	mov	dl,BYTE [esi*1+esp]
+	lea	esi,[1+esi]
+	xor	al,dl
+	mov	BYTE [esi*1+ecx-1],al
+	dec	ebx
+	jnz	NEAR L$007tail_loop
+L$006done:
+	add	esp,132
+L$000no_data:
+	pop	edi
+	pop	esi
+	pop	ebx
+	pop	ebp
+	ret
+global	_ChaCha20_ssse3
+align	16
+_ChaCha20_ssse3:
+L$_ChaCha20_ssse3_begin:
+	push	ebp
+	push	ebx
+	push	esi
+	push	edi
+L$ssse3_shortcut:
+	mov	edi,DWORD [20+esp]
+	mov	esi,DWORD [24+esp]
+	mov	ecx,DWORD [28+esp]
+	mov	edx,DWORD [32+esp]
+	mov	ebx,DWORD [36+esp]
+	mov	ebp,esp
+	sub	esp,524
+	and	esp,-64
+	mov	DWORD [512+esp],ebp
+	lea	eax,[(L$ssse3_data-L$pic_point)+eax]
+	movdqu	xmm3,[ebx]
+	cmp	ecx,256
+	jb	NEAR L$0081x
+	mov	DWORD [516+esp],edx
+	mov	DWORD [520+esp],ebx
+	sub	ecx,256
+	lea	ebp,[384+esp]
+	movdqu	xmm7,[edx]
+	pshufd	xmm0,xmm3,0
+	pshufd	xmm1,xmm3,85
+	pshufd	xmm2,xmm3,170
+	pshufd	xmm3,xmm3,255
+	paddd	xmm0,[48+eax]
+	pshufd	xmm4,xmm7,0
+	pshufd	xmm5,xmm7,85
+	psubd	xmm0,[64+eax]
+	pshufd	xmm6,xmm7,170
+	pshufd	xmm7,xmm7,255
+	movdqa	[64+ebp],xmm0
+	movdqa	[80+ebp],xmm1
+	movdqa	[96+ebp],xmm2
+	movdqa	[112+ebp],xmm3
+	movdqu	xmm3,[16+edx]
+	movdqa	[ebp-64],xmm4
+	movdqa	[ebp-48],xmm5
+	movdqa	[ebp-32],xmm6
+	movdqa	[ebp-16],xmm7
+	movdqa	xmm7,[32+eax]
+	lea	ebx,[128+esp]
+	pshufd	xmm0,xmm3,0
+	pshufd	xmm1,xmm3,85
+	pshufd	xmm2,xmm3,170
+	pshufd	xmm3,xmm3,255
+	pshufd	xmm4,xmm7,0
+	pshufd	xmm5,xmm7,85
+	pshufd	xmm6,xmm7,170
+	pshufd	xmm7,xmm7,255
+	movdqa	[ebp],xmm0
+	movdqa	[16+ebp],xmm1
+	movdqa	[32+ebp],xmm2
+	movdqa	[48+ebp],xmm3
+	movdqa	[ebp-128],xmm4
+	movdqa	[ebp-112],xmm5
+	movdqa	[ebp-96],xmm6
+	movdqa	[ebp-80],xmm7
+	lea	esi,[128+esi]
+	lea	edi,[128+edi]
+	jmp	NEAR L$009outer_loop
+align	16
+L$009outer_loop:
+	movdqa	xmm1,[ebp-112]
+	movdqa	xmm2,[ebp-96]
+	movdqa	xmm3,[ebp-80]
+	movdqa	xmm5,[ebp-48]
+	movdqa	xmm6,[ebp-32]
+	movdqa	xmm7,[ebp-16]
+	movdqa	[ebx-112],xmm1
+	movdqa	[ebx-96],xmm2
+	movdqa	[ebx-80],xmm3
+	movdqa	[ebx-48],xmm5
+	movdqa	[ebx-32],xmm6
+	movdqa	[ebx-16],xmm7
+	movdqa	xmm2,[32+ebp]
+	movdqa	xmm3,[48+ebp]
+	movdqa	xmm4,[64+ebp]
+	movdqa	xmm5,[80+ebp]
+	movdqa	xmm6,[96+ebp]
+	movdqa	xmm7,[112+ebp]
+	paddd	xmm4,[64+eax]
+	movdqa	[32+ebx],xmm2
+	movdqa	[48+ebx],xmm3
+	movdqa	[64+ebx],xmm4
+	movdqa	[80+ebx],xmm5
+	movdqa	[96+ebx],xmm6
+	movdqa	[112+ebx],xmm7
+	movdqa	[64+ebp],xmm4
+	movdqa	xmm0,[ebp-128]
+	movdqa	xmm6,xmm4
+	movdqa	xmm3,[ebp-64]
+	movdqa	xmm4,[ebp]
+	movdqa	xmm5,[16+ebp]
+	mov	edx,10
+	nop
+align	16
+L$010loop:
+	paddd	xmm0,xmm3
+	movdqa	xmm2,xmm3
+	pxor	xmm6,xmm0
+	pshufb	xmm6,[eax]
+	paddd	xmm4,xmm6
+	pxor	xmm2,xmm4
+	movdqa	xmm3,[ebx-48]
+	movdqa	xmm1,xmm2
+	pslld	xmm2,12
+	psrld	xmm1,20
+	por	xmm2,xmm1
+	movdqa	xmm1,[ebx-112]
+	paddd	xmm0,xmm2
+	movdqa	xmm7,[80+ebx]
+	pxor	xmm6,xmm0
+	movdqa	[ebx-128],xmm0
+	pshufb	xmm6,[16+eax]
+	paddd	xmm4,xmm6
+	movdqa	[64+ebx],xmm6
+	pxor	xmm2,xmm4
+	paddd	xmm1,xmm3
+	movdqa	xmm0,xmm2
+	pslld	xmm2,7
+	psrld	xmm0,25
+	pxor	xmm7,xmm1
+	por	xmm2,xmm0
+	movdqa	[ebx],xmm4
+	pshufb	xmm7,[eax]
+	movdqa	[ebx-64],xmm2
+	paddd	xmm5,xmm7
+	movdqa	xmm4,[32+ebx]
+	pxor	xmm3,xmm5
+	movdqa	xmm2,[ebx-32]
+	movdqa	xmm0,xmm3
+	pslld	xmm3,12
+	psrld	xmm0,20
+	por	xmm3,xmm0
+	movdqa	xmm0,[ebx-96]
+	paddd	xmm1,xmm3
+	movdqa	xmm6,[96+ebx]
+	pxor	xmm7,xmm1
+	movdqa	[ebx-112],xmm1
+	pshufb	xmm7,[16+eax]
+	paddd	xmm5,xmm7
+	movdqa	[80+ebx],xmm7
+	pxor	xmm3,xmm5
+	paddd	xmm0,xmm2
+	movdqa	xmm1,xmm3
+	pslld	xmm3,7
+	psrld	xmm1,25
+	pxor	xmm6,xmm0
+	por	xmm3,xmm1
+	movdqa	[16+ebx],xmm5
+	pshufb	xmm6,[eax]
+	movdqa	[ebx-48],xmm3
+	paddd	xmm4,xmm6
+	movdqa	xmm5,[48+ebx]
+	pxor	xmm2,xmm4
+	movdqa	xmm3,[ebx-16]
+	movdqa	xmm1,xmm2
+	pslld	xmm2,12
+	psrld	xmm1,20
+	por	xmm2,xmm1
+	movdqa	xmm1,[ebx-80]
+	paddd	xmm0,xmm2
+	movdqa	xmm7,[112+ebx]
+	pxor	xmm6,xmm0
+	movdqa	[ebx-96],xmm0
+	pshufb	xmm6,[16+eax]
+	paddd	xmm4,xmm6
+	movdqa	[96+ebx],xmm6
+	pxor	xmm2,xmm4
+	paddd	xmm1,xmm3
+	movdqa	xmm0,xmm2
+	pslld	xmm2,7
+	psrld	xmm0,25
+	pxor	xmm7,xmm1
+	por	xmm2,xmm0
+	pshufb	xmm7,[eax]
+	movdqa	[ebx-32],xmm2
+	paddd	xmm5,xmm7
+	pxor	xmm3,xmm5
+	movdqa	xmm2,[ebx-48]
+	movdqa	xmm0,xmm3
+	pslld	xmm3,12
+	psrld	xmm0,20
+	por	xmm3,xmm0
+	movdqa	xmm0,[ebx-128]
+	paddd	xmm1,xmm3
+	pxor	xmm7,xmm1
+	movdqa	[ebx-80],xmm1
+	pshufb	xmm7,[16+eax]
+	paddd	xmm5,xmm7
+	movdqa	xmm6,xmm7
+	pxor	xmm3,xmm5
+	paddd	xmm0,xmm2
+	movdqa	xmm1,xmm3
+	pslld	xmm3,7
+	psrld	xmm1,25
+	pxor	xmm6,xmm0
+	por	xmm3,xmm1
+	pshufb	xmm6,[eax]
+	movdqa	[ebx-16],xmm3
+	paddd	xmm4,xmm6
+	pxor	xmm2,xmm4
+	movdqa	xmm3,[ebx-32]
+	movdqa	xmm1,xmm2
+	pslld	xmm2,12
+	psrld	xmm1,20
+	por	xmm2,xmm1
+	movdqa	xmm1,[ebx-112]
+	paddd	xmm0,xmm2
+	movdqa	xmm7,[64+ebx]
+	pxor	xmm6,xmm0
+	movdqa	[ebx-128],xmm0
+	pshufb	xmm6,[16+eax]
+	paddd	xmm4,xmm6
+	movdqa	[112+ebx],xmm6
+	pxor	xmm2,xmm4
+	paddd	xmm1,xmm3
+	movdqa	xmm0,xmm2
+	pslld	xmm2,7
+	psrld	xmm0,25
+	pxor	xmm7,xmm1
+	por	xmm2,xmm0
+	movdqa	[32+ebx],xmm4
+	pshufb	xmm7,[eax]
+	movdqa	[ebx-48],xmm2
+	paddd	xmm5,xmm7
+	movdqa	xmm4,[ebx]
+	pxor	xmm3,xmm5
+	movdqa	xmm2,[ebx-16]
+	movdqa	xmm0,xmm3
+	pslld	xmm3,12
+	psrld	xmm0,20
+	por	xmm3,xmm0
+	movdqa	xmm0,[ebx-96]
+	paddd	xmm1,xmm3
+	movdqa	xmm6,[80+ebx]
+	pxor	xmm7,xmm1
+	movdqa	[ebx-112],xmm1
+	pshufb	xmm7,[16+eax]
+	paddd	xmm5,xmm7
+	movdqa	[64+ebx],xmm7
+	pxor	xmm3,xmm5
+	paddd	xmm0,xmm2
+	movdqa	xmm1,xmm3
+	pslld	xmm3,7
+	psrld	xmm1,25
+	pxor	xmm6,xmm0
+	por	xmm3,xmm1
+	movdqa	[48+ebx],xmm5
+	pshufb	xmm6,[eax]
+	movdqa	[ebx-32],xmm3
+	paddd	xmm4,xmm6
+	movdqa	xmm5,[16+ebx]
+	pxor	xmm2,xmm4
+	movdqa	xmm3,[ebx-64]
+	movdqa	xmm1,xmm2
+	pslld	xmm2,12
+	psrld	xmm1,20
+	por	xmm2,xmm1
+	movdqa	xmm1,[ebx-80]
+	paddd	xmm0,xmm2
+	movdqa	xmm7,[96+ebx]
+	pxor	xmm6,xmm0
+	movdqa	[ebx-96],xmm0
+	pshufb	xmm6,[16+eax]
+	paddd	xmm4,xmm6
+	movdqa	[80+ebx],xmm6
+	pxor	xmm2,xmm4
+	paddd	xmm1,xmm3
+	movdqa	xmm0,xmm2
+	pslld	xmm2,7
+	psrld	xmm0,25
+	pxor	xmm7,xmm1
+	por	xmm2,xmm0
+	pshufb	xmm7,[eax]
+	movdqa	[ebx-16],xmm2
+	paddd	xmm5,xmm7
+	pxor	xmm3,xmm5
+	movdqa	xmm0,xmm3
+	pslld	xmm3,12
+	psrld	xmm0,20
+	por	xmm3,xmm0
+	movdqa	xmm0,[ebx-128]
+	paddd	xmm1,xmm3
+	movdqa	xmm6,[64+ebx]
+	pxor	xmm7,xmm1
+	movdqa	[ebx-80],xmm1
+	pshufb	xmm7,[16+eax]
+	paddd	xmm5,xmm7
+	movdqa	[96+ebx],xmm7
+	pxor	xmm3,xmm5
+	movdqa	xmm1,xmm3
+	pslld	xmm3,7
+	psrld	xmm1,25
+	por	xmm3,xmm1
+	dec	edx
+	jnz	NEAR L$010loop
+	movdqa	[ebx-64],xmm3
+	movdqa	[ebx],xmm4
+	movdqa	[16+ebx],xmm5
+	movdqa	[64+ebx],xmm6
+	movdqa	[96+ebx],xmm7
+	movdqa	xmm1,[ebx-112]
+	movdqa	xmm2,[ebx-96]
+	movdqa	xmm3,[ebx-80]
+	paddd	xmm0,[ebp-128]
+	paddd	xmm1,[ebp-112]
+	paddd	xmm2,[ebp-96]
+	paddd	xmm3,[ebp-80]
+	movdqa	xmm6,xmm0
+	punpckldq	xmm0,xmm1
+	movdqa	xmm7,xmm2
+	punpckldq	xmm2,xmm3
+	punpckhdq	xmm6,xmm1
+	punpckhdq	xmm7,xmm3
+	movdqa	xmm1,xmm0
+	punpcklqdq	xmm0,xmm2
+	movdqa	xmm3,xmm6
+	punpcklqdq	xmm6,xmm7
+	punpckhqdq	xmm1,xmm2
+	punpckhqdq	xmm3,xmm7
+	movdqu	xmm4,[esi-128]
+	movdqu	xmm5,[esi-64]
+	movdqu	xmm2,[esi]
+	movdqu	xmm7,[64+esi]
+	lea	esi,[16+esi]
+	pxor	xmm4,xmm0
+	movdqa	xmm0,[ebx-64]
+	pxor	xmm5,xmm1
+	movdqa	xmm1,[ebx-48]
+	pxor	xmm6,xmm2
+	movdqa	xmm2,[ebx-32]
+	pxor	xmm7,xmm3
+	movdqa	xmm3,[ebx-16]
+	movdqu	[edi-128],xmm4
+	movdqu	[edi-64],xmm5
+	movdqu	[edi],xmm6
+	movdqu	[64+edi],xmm7
+	lea	edi,[16+edi]
+	paddd	xmm0,[ebp-64]
+	paddd	xmm1,[ebp-48]
+	paddd	xmm2,[ebp-32]
+	paddd	xmm3,[ebp-16]
+	movdqa	xmm6,xmm0
+	punpckldq	xmm0,xmm1
+	movdqa	xmm7,xmm2
+	punpckldq	xmm2,xmm3
+	punpckhdq	xmm6,xmm1
+	punpckhdq	xmm7,xmm3
+	movdqa	xmm1,xmm0
+	punpcklqdq	xmm0,xmm2
+	movdqa	xmm3,xmm6
+	punpcklqdq	xmm6,xmm7
+	punpckhqdq	xmm1,xmm2
+	punpckhqdq	xmm3,xmm7
+	movdqu	xmm4,[esi-128]
+	movdqu	xmm5,[esi-64]
+	movdqu	xmm2,[esi]
+	movdqu	xmm7,[64+esi]
+	lea	esi,[16+esi]
+	pxor	xmm4,xmm0
+	movdqa	xmm0,[ebx]
+	pxor	xmm5,xmm1
+	movdqa	xmm1,[16+ebx]
+	pxor	xmm6,xmm2
+	movdqa	xmm2,[32+ebx]
+	pxor	xmm7,xmm3
+	movdqa	xmm3,[48+ebx]
+	movdqu	[edi-128],xmm4
+	movdqu	[edi-64],xmm5
+	movdqu	[edi],xmm6
+	movdqu	[64+edi],xmm7
+	lea	edi,[16+edi]
+	paddd	xmm0,[ebp]
+	paddd	xmm1,[16+ebp]
+	paddd	xmm2,[32+ebp]
+	paddd	xmm3,[48+ebp]
+	movdqa	xmm6,xmm0
+	punpckldq	xmm0,xmm1
+	movdqa	xmm7,xmm2
+	punpckldq	xmm2,xmm3
+	punpckhdq	xmm6,xmm1
+	punpckhdq	xmm7,xmm3
+	movdqa	xmm1,xmm0
+	punpcklqdq	xmm0,xmm2
+	movdqa	xmm3,xmm6
+	punpcklqdq	xmm6,xmm7
+	punpckhqdq	xmm1,xmm2
+	punpckhqdq	xmm3,xmm7
+	movdqu	xmm4,[esi-128]
+	movdqu	xmm5,[esi-64]
+	movdqu	xmm2,[esi]
+	movdqu	xmm7,[64+esi]
+	lea	esi,[16+esi]
+	pxor	xmm4,xmm0
+	movdqa	xmm0,[64+ebx]
+	pxor	xmm5,xmm1
+	movdqa	xmm1,[80+ebx]
+	pxor	xmm6,xmm2
+	movdqa	xmm2,[96+ebx]
+	pxor	xmm7,xmm3
+	movdqa	xmm3,[112+ebx]
+	movdqu	[edi-128],xmm4
+	movdqu	[edi-64],xmm5
+	movdqu	[edi],xmm6
+	movdqu	[64+edi],xmm7
+	lea	edi,[16+edi]
+	paddd	xmm0,[64+ebp]
+	paddd	xmm1,[80+ebp]
+	paddd	xmm2,[96+ebp]
+	paddd	xmm3,[112+ebp]
+	movdqa	xmm6,xmm0
+	punpckldq	xmm0,xmm1
+	movdqa	xmm7,xmm2
+	punpckldq	xmm2,xmm3
+	punpckhdq	xmm6,xmm1
+	punpckhdq	xmm7,xmm3
+	movdqa	xmm1,xmm0
+	punpcklqdq	xmm0,xmm2
+	movdqa	xmm3,xmm6
+	punpcklqdq	xmm6,xmm7
+	punpckhqdq	xmm1,xmm2
+	punpckhqdq	xmm3,xmm7
+	movdqu	xmm4,[esi-128]
+	movdqu	xmm5,[esi-64]
+	movdqu	xmm2,[esi]
+	movdqu	xmm7,[64+esi]
+	lea	esi,[208+esi]
+	pxor	xmm4,xmm0
+	pxor	xmm5,xmm1
+	pxor	xmm6,xmm2
+	pxor	xmm7,xmm3
+	movdqu	[edi-128],xmm4
+	movdqu	[edi-64],xmm5
+	movdqu	[edi],xmm6
+	movdqu	[64+edi],xmm7
+	lea	edi,[208+edi]
+	sub	ecx,256
+	jnc	NEAR L$009outer_loop
+	add	ecx,256
+	jz	NEAR L$011done
+	mov	ebx,DWORD [520+esp]
+	lea	esi,[esi-128]
+	mov	edx,DWORD [516+esp]
+	lea	edi,[edi-128]
+	movd	xmm2,DWORD [64+ebp]
+	movdqu	xmm3,[ebx]
+	paddd	xmm2,[96+eax]
+	pand	xmm3,[112+eax]
+	por	xmm3,xmm2
+L$0081x:
+	movdqa	xmm0,[32+eax]
+	movdqu	xmm1,[edx]
+	movdqu	xmm2,[16+edx]
+	movdqa	xmm6,[eax]
+	movdqa	xmm7,[16+eax]
+	mov	DWORD [48+esp],ebp
+	movdqa	[esp],xmm0
+	movdqa	[16+esp],xmm1
+	movdqa	[32+esp],xmm2
+	movdqa	[48+esp],xmm3
+	mov	edx,10
+	jmp	NEAR L$012loop1x
+align	16
+L$013outer1x:
+	movdqa	xmm3,[80+eax]
+	movdqa	xmm0,[esp]
+	movdqa	xmm1,[16+esp]
+	movdqa	xmm2,[32+esp]
+	paddd	xmm3,[48+esp]
+	mov	edx,10
+	movdqa	[48+esp],xmm3
+	jmp	NEAR L$012loop1x
+align	16
+L$012loop1x:
+	paddd	xmm0,xmm1
+	pxor	xmm3,xmm0
+db	102,15,56,0,222
+	paddd	xmm2,xmm3
+	pxor	xmm1,xmm2
+	movdqa	xmm4,xmm1
+	psrld	xmm1,20
+	pslld	xmm4,12
+	por	xmm1,xmm4
+	paddd	xmm0,xmm1
+	pxor	xmm3,xmm0
+db	102,15,56,0,223
+	paddd	xmm2,xmm3
+	pxor	xmm1,xmm2
+	movdqa	xmm4,xmm1
+	psrld	xmm1,25
+	pslld	xmm4,7
+	por	xmm1,xmm4
+	pshufd	xmm2,xmm2,78
+	pshufd	xmm1,xmm1,57
+	pshufd	xmm3,xmm3,147
+	nop
+	paddd	xmm0,xmm1
+	pxor	xmm3,xmm0
+db	102,15,56,0,222
+	paddd	xmm2,xmm3
+	pxor	xmm1,xmm2
+	movdqa	xmm4,xmm1
+	psrld	xmm1,20
+	pslld	xmm4,12
+	por	xmm1,xmm4
+	paddd	xmm0,xmm1
+	pxor	xmm3,xmm0
+db	102,15,56,0,223
+	paddd	xmm2,xmm3
+	pxor	xmm1,xmm2
+	movdqa	xmm4,xmm1
+	psrld	xmm1,25
+	pslld	xmm4,7
+	por	xmm1,xmm4
+	pshufd	xmm2,xmm2,78
+	pshufd	xmm1,xmm1,147
+	pshufd	xmm3,xmm3,57
+	dec	edx
+	jnz	NEAR L$012loop1x
+	paddd	xmm0,[esp]
+	paddd	xmm1,[16+esp]
+	paddd	xmm2,[32+esp]
+	paddd	xmm3,[48+esp]
+	cmp	ecx,64
+	jb	NEAR L$014tail
+	movdqu	xmm4,[esi]
+	movdqu	xmm5,[16+esi]
+	pxor	xmm0,xmm4
+	movdqu	xmm4,[32+esi]
+	pxor	xmm1,xmm5
+	movdqu	xmm5,[48+esi]
+	pxor	xmm2,xmm4
+	pxor	xmm3,xmm5
+	lea	esi,[64+esi]
+	movdqu	[edi],xmm0
+	movdqu	[16+edi],xmm1
+	movdqu	[32+edi],xmm2
+	movdqu	[48+edi],xmm3
+	lea	edi,[64+edi]
+	sub	ecx,64
+	jnz	NEAR L$013outer1x
+	jmp	NEAR L$011done
+L$014tail:
+	movdqa	[esp],xmm0
+	movdqa	[16+esp],xmm1
+	movdqa	[32+esp],xmm2
+	movdqa	[48+esp],xmm3
+	xor	eax,eax
+	xor	edx,edx
+	xor	ebp,ebp
+L$015tail_loop:
+	mov	al,BYTE [ebp*1+esp]
+	mov	dl,BYTE [ebp*1+esi]
+	lea	ebp,[1+ebp]
+	xor	al,dl
+	mov	BYTE [ebp*1+edi-1],al
+	dec	ecx
+	jnz	NEAR L$015tail_loop
+L$011done:
+	mov	esp,DWORD [512+esp]
+	pop	edi
+	pop	esi
+	pop	ebx
+	pop	ebp
+	ret
+align	64
+L$ssse3_data:
+db	2,3,0,1,6,7,4,5,10,11,8,9,14,15,12,13
+db	3,0,1,2,7,4,5,6,11,8,9,10,15,12,13,14
+dd	1634760805,857760878,2036477234,1797285236
+dd	0,1,2,3
+dd	4,4,4,4
+dd	1,0,0,0
+dd	4,0,0,0
+dd	0,-1,-1,-1
+align	64
+db	67,104,97,67,104,97,50,48,32,102,111,114,32,120,56,54
+db	44,32,67,82,89,80,84,79,71,65,77,83,32,98,121,32
+db	60,97,112,112,114,111,64,111,112,101,110,115,115,108,46,111
+db	114,103,62,0
+segment	.bss
+common	_OPENSSL_ia32cap_P 16

third_party/boringssl/win-x86/crypto/cpu-x86-asm.asm

diff --git a/third_party/boringssl/win-x86/crypto/cpu-x86-asm.asm b/third_party/boringssl/win-x86/crypto/cpu-x86-asm.asm
deleted file mode 100644
index 4317a73..0000000
--- a/third_party/boringssl/win-x86/crypto/cpu-x86-asm.asm
+++ /dev/null
@@ -1,303 +0,0 @@
-%ifidn __OUTPUT_FORMAT__,obj
-section	code	use32 class=code align=64
-%elifidn __OUTPUT_FORMAT__,win32
-%ifdef __YASM_VERSION_ID__
-%if __YASM_VERSION_ID__ < 01010000h
-%error yasm version 1.1.0 or later needed.
-%endif
-; Yasm automatically includes .00 and complains about redefining it.
-; https://www.tortall.net/projects/yasm/manual/html/objfmt-win32-safeseh.html
-%else
-$@feat.00 equ 1
-%endif
-section	.text	code align=64
-%else
-section	.text	code
-%endif
-global	_OPENSSL_ia32_cpuid
-align	16
-_OPENSSL_ia32_cpuid:
-L$_OPENSSL_ia32_cpuid_begin:
-	push	ebp
-	push	ebx
-	push	esi
-	push	edi
-	xor	edx,edx
-	pushfd
-	pop	eax
-	mov	ecx,eax
-	xor	eax,2097152
-	push	eax
-	popfd
-	pushfd
-	pop	eax
-	xor	ecx,eax
-	xor	eax,eax
-	bt	ecx,21
-	jnc	NEAR L$000nocpuid
-	mov	esi,DWORD [20+esp]
-	mov	DWORD [8+esi],eax
-	cpuid
-	mov	edi,eax
-	xor	eax,eax
-	cmp	ebx,1970169159
-	setne	al
-	mov	ebp,eax
-	cmp	edx,1231384169
-	setne	al
-	or	ebp,eax
-	cmp	ecx,1818588270
-	setne	al
-	or	ebp,eax
-	jz	NEAR L$001intel
-	cmp	ebx,1752462657
-	setne	al
-	mov	esi,eax
-	cmp	edx,1769238117
-	setne	al
-	or	esi,eax
-	cmp	ecx,1145913699
-	setne	al
-	or	esi,eax
-	jnz	NEAR L$001intel
-	mov	eax,2147483648
-	cpuid
-	cmp	eax,2147483649
-	jb	NEAR L$001intel
-	mov	esi,eax
-	mov	eax,2147483649
-	cpuid
-	or	ebp,ecx
-	and	ebp,2049
-	cmp	esi,2147483656
-	jb	NEAR L$001intel
-	mov	eax,2147483656
-	cpuid
-	movzx	esi,cl
-	inc	esi
-	mov	eax,1
-	xor	ecx,ecx
-	cpuid
-	bt	edx,28
-	jnc	NEAR L$002generic
-	shr	ebx,16
-	and	ebx,255
-	cmp	ebx,esi
-	ja	NEAR L$002generic
-	and	edx,4026531839
-	jmp	NEAR L$002generic
-L$001intel:
-	cmp	edi,7
-	jb	NEAR L$003cacheinfo
-	mov	esi,DWORD [20+esp]
-	mov	eax,7
-	xor	ecx,ecx
-	cpuid
-	mov	DWORD [8+esi],ebx
-L$003cacheinfo:
-	cmp	edi,4
-	mov	edi,-1
-	jb	NEAR L$004nocacheinfo
-	mov	eax,4
-	mov	ecx,0
-	cpuid
-	mov	edi,eax
-	shr	edi,14
-	and	edi,4095
-L$004nocacheinfo:
-	mov	eax,1
-	xor	ecx,ecx
-	cpuid
-	and	edx,3220176895
-	cmp	ebp,0
-	jne	NEAR L$005notintel
-	or	edx,1073741824
-L$005notintel:
-	bt	edx,28
-	jnc	NEAR L$002generic
-	and	edx,4026531839
-	cmp	edi,0
-	je	NEAR L$002generic
-	or	edx,268435456
-	shr	ebx,16
-	cmp	bl,1
-	ja	NEAR L$002generic
-	and	edx,4026531839
-L$002generic:
-	and	ebp,2048
-	and	ecx,4294965247
-	mov	esi,edx
-	or	ebp,ecx
-	bt	ecx,27
-	jnc	NEAR L$006clear_avx
-	xor	ecx,ecx
-db	15,1,208
-	and	eax,6
-	cmp	eax,6
-	je	NEAR L$007done
-	cmp	eax,2
-	je	NEAR L$006clear_avx
-L$008clear_xmm:
-	and	ebp,4261412861
-	and	esi,4278190079
-L$006clear_avx:
-	and	ebp,4026525695
-	mov	edi,DWORD [20+esp]
-	and	DWORD [8+edi],4294967263
-L$007done:
-	mov	eax,esi
-	mov	edx,ebp
-L$000nocpuid:
-	pop	edi
-	pop	esi
-	pop	ebx
-	pop	ebp
-	ret
-;extern	_OPENSSL_ia32cap_P
-global	_OPENSSL_rdtsc
-align	16
-_OPENSSL_rdtsc:
-L$_OPENSSL_rdtsc_begin:
-	xor	eax,eax
-	xor	edx,edx
-	lea	ecx,[_OPENSSL_ia32cap_P]
-	bt	DWORD [ecx],4
-	jnc	NEAR L$009notsc
-	rdtsc
-L$009notsc:
-	ret
-global	_OPENSSL_instrument_halt
-align	16
-_OPENSSL_instrument_halt:
-L$_OPENSSL_instrument_halt_begin:
-	lea	ecx,[_OPENSSL_ia32cap_P]
-	bt	DWORD [ecx],4
-	jnc	NEAR L$010nohalt
-dd	2421723150
-	and	eax,3
-	jnz	NEAR L$010nohalt
-	pushfd
-	pop	eax
-	bt	eax,9
-	jnc	NEAR L$010nohalt
-	rdtsc
-	push	edx
-	push	eax
-	hlt
-	rdtsc
-	sub	eax,DWORD [esp]
-	sbb	edx,DWORD [4+esp]
-	add	esp,8
-	ret
-L$010nohalt:
-	xor	eax,eax
-	xor	edx,edx
-	ret
-global	_OPENSSL_far_spin
-align	16
-_OPENSSL_far_spin:
-L$_OPENSSL_far_spin_begin:
-	pushfd
-	pop	eax
-	bt	eax,9
-	jnc	NEAR L$011nospin
-	mov	eax,DWORD [4+esp]
-	mov	ecx,DWORD [8+esp]
-dd	2430111262
-	xor	eax,eax
-	mov	edx,DWORD [ecx]
-	jmp	NEAR L$012spin
-align	16
-L$012spin:
-	inc	eax
-	cmp	edx,DWORD [ecx]
-	je	NEAR L$012spin
-dd	529567888
-	ret
-L$011nospin:
-	xor	eax,eax
-	xor	edx,edx
-	ret
-global	_OPENSSL_wipe_cpu
-align	16
-_OPENSSL_wipe_cpu:
-L$_OPENSSL_wipe_cpu_begin:
-	xor	eax,eax
-	xor	edx,edx
-	lea	ecx,[_OPENSSL_ia32cap_P]
-	mov	ecx,DWORD [ecx]
-	bt	DWORD [ecx],1
-	jnc	NEAR L$013no_x87
-	and	ecx,83886080
-	cmp	ecx,83886080
-	jne	NEAR L$014no_sse2
-	pxor	xmm0,xmm0
-	pxor	xmm1,xmm1
-	pxor	xmm2,xmm2
-	pxor	xmm3,xmm3
-	pxor	xmm4,xmm4
-	pxor	xmm5,xmm5
-	pxor	xmm6,xmm6
-	pxor	xmm7,xmm7
-L$014no_sse2:
-dd	4007259865,4007259865,4007259865,4007259865,2430851995
-L$013no_x87:
-	lea	eax,[4+esp]
-	ret
-global	_OPENSSL_atomic_add
-align	16
-_OPENSSL_atomic_add:
-L$_OPENSSL_atomic_add_begin:
-	mov	edx,DWORD [4+esp]
-	mov	ecx,DWORD [8+esp]
-	push	ebx
-	nop
-	mov	eax,DWORD [edx]
-L$015spin:
-	lea	ebx,[ecx*1+eax]
-	nop
-dd	447811568
-	jne	NEAR L$015spin
-	mov	eax,ebx
-	pop	ebx
-	ret
-global	_OPENSSL_indirect_call
-align	16
-_OPENSSL_indirect_call:
-L$_OPENSSL_indirect_call_begin:
-	push	ebp
-	mov	ebp,esp
-	sub	esp,28
-	mov	ecx,DWORD [12+ebp]
-	mov	DWORD [esp],ecx
-	mov	edx,DWORD [16+ebp]
-	mov	DWORD [4+esp],edx
-	mov	eax,DWORD [20+ebp]
-	mov	DWORD [8+esp],eax
-	mov	eax,DWORD [24+ebp]
-	mov	DWORD [12+esp],eax
-	mov	eax,DWORD [28+ebp]
-	mov	DWORD [16+esp],eax
-	mov	eax,DWORD [32+ebp]
-	mov	DWORD [20+esp],eax
-	mov	eax,DWORD [36+ebp]
-	mov	DWORD [24+esp],eax
-	call	DWORD [8+ebp]
-	mov	esp,ebp
-	pop	ebp
-	ret
-global	_OPENSSL_ia32_rdrand
-align	16
-_OPENSSL_ia32_rdrand:
-L$_OPENSSL_ia32_rdrand_begin:
-	mov	ecx,8
-L$016loop:
-db	15,199,240
-	jc	NEAR L$017break
-	loop	L$016loop
-L$017break:
-	cmp	eax,0
-	cmove	eax,ecx
-	ret
-segment	.bss
-common	_OPENSSL_ia32cap_P 16

third_party/boringssl/win-x86/crypto/rc4/rc4-586.asm

diff --git a/third_party/boringssl/win-x86/crypto/rc4/rc4-586.asm b/third_party/boringssl/win-x86/crypto/rc4/rc4-586.asm
index 08cd9f6..0bab2be 100644
--- a/third_party/boringssl/win-x86/crypto/rc4/rc4-586.asm
+++ b/third_party/boringssl/win-x86/crypto/rc4/rc4-586.asm
@@ -349,34 +349,5 @@
 	pop	ebx
 	pop	ebp
 	ret
-global	_RC4_options
-align	16
-_RC4_options:
-L$_RC4_options_begin:
-	call	L$016pic_point
-L$016pic_point:
-	pop	eax
-	lea	eax,[(L$017opts-L$016pic_point)+eax]
-	lea	edx,[_OPENSSL_ia32cap_P]
-	mov	edx,DWORD [edx]
-	bt	edx,20
-	jc	NEAR L$0181xchar
-	bt	edx,26
-	jnc	NEAR L$019ret
-	add	eax,25
-	ret
-L$0181xchar:
-	add	eax,12
-L$019ret:
-	ret
-align	64
-L$017opts:
-db	114,99,52,40,52,120,44,105,110,116,41,0
-db	114,99,52,40,49,120,44,99,104,97,114,41,0
-db	114,99,52,40,56,120,44,109,109,120,41,0
-db	82,67,52,32,102,111,114,32,120,56,54,44,32,67,82,89
-db	80,84,79,71,65,77,83,32,98,121,32,60,97,112,112,114
-db	111,64,111,112,101,110,115,115,108,46,111,114,103,62,0
-align	64
 segment	.bss
 common	_OPENSSL_ia32cap_P 16

third_party/boringssl/win-x86/crypto/sha/sha1-586.asm

diff --git a/third_party/boringssl/win-x86/crypto/sha/sha1-586.asm b/third_party/boringssl/win-x86/crypto/sha/sha1-586.asm
index e24449d..cee8c6262 100644
--- a/third_party/boringssl/win-x86/crypto/sha/sha1-586.asm
+++ b/third_party/boringssl/win-x86/crypto/sha/sha1-586.asm
@@ -35,8 +35,11 @@
 	mov	ecx,DWORD [8+esi]
 	test	eax,16777216
 	jz	NEAR L$001x86
-	test	ecx,536870912
-	jnz	NEAR L$shaext_shortcut
+	and	edx,268435456
+	and	eax,1073741824
+	or	eax,edx
+	cmp	eax,1342177280
+	je	NEAR L$avx_shortcut
 	jmp	NEAR L$ssse3_shortcut
 align	16
 L$001x86:
@@ -1405,7 +1408,7 @@
 	pop	ebp
 	ret
 align	16
-__sha1_block_data_order_shaext:
+__sha1_block_data_order_ssse3:
 	push	ebp
 	push	ebx
 	push	esi
@@ -1414,174 +1417,6 @@
 L$003pic_point:
 	pop	ebp
 	lea	ebp,[(L$K_XX_XX-L$003pic_point)+ebp]
-L$shaext_shortcut:
-	mov	edi,DWORD [20+esp]
-	mov	ebx,esp
-	mov	esi,DWORD [24+esp]
-	mov	ecx,DWORD [28+esp]
-	sub	esp,32
-	movdqu	xmm0,[edi]
-	movd	xmm1,DWORD [16+edi]
-	and	esp,-32
-	movdqa	xmm3,[80+ebp]
-	movdqu	xmm4,[esi]
-	pshufd	xmm0,xmm0,27
-	movdqu	xmm5,[16+esi]
-	pshufd	xmm1,xmm1,27
-	movdqu	xmm6,[32+esi]
-db	102,15,56,0,227
-	movdqu	xmm7,[48+esi]
-db	102,15,56,0,235
-db	102,15,56,0,243
-db	102,15,56,0,251
-	jmp	NEAR L$004loop_shaext
-align	16
-L$004loop_shaext:
-	dec	ecx
-	lea	eax,[64+esi]
-	movdqa	[esp],xmm1
-	paddd	xmm1,xmm4
-	cmovne	esi,eax
-	movdqa	[16+esp],xmm0
-db	15,56,201,229
-	movdqa	xmm2,xmm0
-db	15,58,204,193,0
-db	15,56,200,213
-	pxor	xmm4,xmm6
-db	15,56,201,238
-db	15,56,202,231
-	movdqa	xmm1,xmm0
-db	15,58,204,194,0
-db	15,56,200,206
-	pxor	xmm5,xmm7
-db	15,56,202,236
-db	15,56,201,247
-	movdqa	xmm2,xmm0
-db	15,58,204,193,0
-db	15,56,200,215
-	pxor	xmm6,xmm4
-db	15,56,201,252
-db	15,56,202,245
-	movdqa	xmm1,xmm0
-db	15,58,204,194,0
-db	15,56,200,204
-	pxor	xmm7,xmm5
-db	15,56,202,254
-db	15,56,201,229
-	movdqa	xmm2,xmm0
-db	15,58,204,193,0
-db	15,56,200,213
-	pxor	xmm4,xmm6
-db	15,56,201,238
-db	15,56,202,231
-	movdqa	xmm1,xmm0
-db	15,58,204,194,1
-db	15,56,200,206
-	pxor	xmm5,xmm7
-db	15,56,202,236
-db	15,56,201,247
-	movdqa	xmm2,xmm0
-db	15,58,204,193,1
-db	15,56,200,215
-	pxor	xmm6,xmm4
-db	15,56,201,252
-db	15,56,202,245
-	movdqa	xmm1,xmm0
-db	15,58,204,194,1
-db	15,56,200,204
-	pxor	xmm7,xmm5
-db	15,56,202,254
-db	15,56,201,229
-	movdqa	xmm2,xmm0
-db	15,58,204,193,1
-db	15,56,200,213
-	pxor	xmm4,xmm6
-db	15,56,201,238
-db	15,56,202,231
-	movdqa	xmm1,xmm0
-db	15,58,204,194,1
-db	15,56,200,206
-	pxor	xmm5,xmm7
-db	15,56,202,236
-db	15,56,201,247
-	movdqa	xmm2,xmm0
-db	15,58,204,193,2
-db	15,56,200,215
-	pxor	xmm6,xmm4
-db	15,56,201,252
-db	15,56,202,245
-	movdqa	xmm1,xmm0
-db	15,58,204,194,2
-db	15,56,200,204
-	pxor	xmm7,xmm5
-db	15,56,202,254
-db	15,56,201,229
-	movdqa	xmm2,xmm0
-db	15,58,204,193,2
-db	15,56,200,213
-	pxor	xmm4,xmm6
-db	15,56,201,238
-db	15,56,202,231
-	movdqa	xmm1,xmm0
-db	15,58,204,194,2
-db	15,56,200,206
-	pxor	xmm5,xmm7
-db	15,56,202,236
-db	15,56,201,247
-	movdqa	xmm2,xmm0
-db	15,58,204,193,2
-db	15,56,200,215
-	pxor	xmm6,xmm4
-db	15,56,201,252
-db	15,56,202,245
-	movdqa	xmm1,xmm0
-db	15,58,204,194,3
-db	15,56,200,204
-	pxor	xmm7,xmm5
-db	15,56,202,254
-	movdqu	xmm4,[esi]
-	movdqa	xmm2,xmm0
-db	15,58,204,193,3
-db	15,56,200,213
-	movdqu	xmm5,[16+esi]
-db	102,15,56,0,227
-	movdqa	xmm1,xmm0
-db	15,58,204,194,3
-db	15,56,200,206
-	movdqu	xmm6,[32+esi]
-db	102,15,56,0,235
-	movdqa	xmm2,xmm0
-db	15,58,204,193,3
-db	15,56,200,215
-	movdqu	xmm7,[48+esi]
-db	102,15,56,0,243
-	movdqa	xmm1,xmm0
-db	15,58,204,194,3
-	movdqa	xmm2,[esp]
-db	102,15,56,0,251
-db	15,56,200,202
-	paddd	xmm0,[16+esp]
-	jnz	NEAR L$004loop_shaext
-	pshufd	xmm0,xmm0,27
-	pshufd	xmm1,xmm1,27
-	movdqu	[edi],xmm0
-	movd	DWORD [16+edi],xmm1
-	mov	esp,ebx
-	pop	edi
-	pop	esi
-	pop	ebx
-	pop	ebp
-	ret
-align	16
-__sha1_block_data_order_ssse3:
-	push	ebp
-	push	ebx
-	push	esi
-	push	edi
-	call	L$005pic_point
-L$005pic_point:
-	pop	ebp
-	lea	ebp,[(L$K_XX_XX-L$005pic_point)+ebp]
 L$ssse3_shortcut:
 	movdqa	xmm7,[ebp]
 	movdqa	xmm0,[16+ebp]
@@ -1634,9 +1469,9 @@
 	xor	ebp,edx
 	pshufd	xmm4,xmm0,238
 	and	esi,ebp
-	jmp	NEAR L$006loop
+	jmp	NEAR L$004loop
 align	16
-L$006loop:
+L$004loop:
 	ror	ebx,2
 	xor	esi,edx
 	mov	ebp,eax
@@ -2539,7 +2374,7 @@
 	add	ecx,edx
 	mov	ebp,DWORD [196+esp]
 	cmp	ebp,DWORD [200+esp]
-	je	NEAR L$007done
+	je	NEAR L$005done
 	movdqa	xmm7,[160+esp]
 	movdqa	xmm6,[176+esp]
 	movdqu	xmm0,[ebp]
@@ -2674,9 +2509,9 @@
 	pshufd	xmm4,xmm0,238
 	and	esi,ebx
 	mov	ebx,ebp
-	jmp	NEAR L$006loop
+	jmp	NEAR L$004loop
 align	16
-L$007done:
+L$005done:
 	add	ebx,DWORD [16+esp]
 	xor	esi,edi
 	mov	ebp,ecx
@@ -2789,6 +2624,1174 @@
 	pop	ebx
 	pop	ebp
 	ret
+align	16
+__sha1_block_data_order_avx:
+	push	ebp
+	push	ebx
+	push	esi
+	push	edi
+	call	L$006pic_point
+L$006pic_point:
+	pop	ebp
+	lea	ebp,[(L$K_XX_XX-L$006pic_point)+ebp]
+L$avx_shortcut:
+	vzeroall
+	vmovdqa	xmm7,[ebp]
+	vmovdqa	xmm0,[16+ebp]
+	vmovdqa	xmm1,[32+ebp]
+	vmovdqa	xmm2,[48+ebp]
+	vmovdqa	xmm6,[64+ebp]
+	mov	edi,DWORD [20+esp]
+	mov	ebp,DWORD [24+esp]
+	mov	edx,DWORD [28+esp]
+	mov	esi,esp
+	sub	esp,208
+	and	esp,-64
+	vmovdqa	[112+esp],xmm0
+	vmovdqa	[128+esp],xmm1
+	vmovdqa	[144+esp],xmm2
+	shl	edx,6
+	vmovdqa	[160+esp],xmm7
+	add	edx,ebp
+	vmovdqa	[176+esp],xmm6
+	add	ebp,64
+	mov	DWORD [192+esp],edi
+	mov	DWORD [196+esp],ebp
+	mov	DWORD [200+esp],edx
+	mov	DWORD [204+esp],esi
+	mov	eax,DWORD [edi]
+	mov	ebx,DWORD [4+edi]
+	mov	ecx,DWORD [8+edi]
+	mov	edx,DWORD [12+edi]
+	mov	edi,DWORD [16+edi]
+	mov	esi,ebx
+	vmovdqu	xmm0,[ebp-64]
+	vmovdqu	xmm1,[ebp-48]
+	vmovdqu	xmm2,[ebp-32]
+	vmovdqu	xmm3,[ebp-16]
+	vpshufb	xmm0,xmm0,xmm6
+	vpshufb	xmm1,xmm1,xmm6
+	vpshufb	xmm2,xmm2,xmm6
+	vmovdqa	[96+esp],xmm7
+	vpshufb	xmm3,xmm3,xmm6
+	vpaddd	xmm4,xmm0,xmm7
+	vpaddd	xmm5,xmm1,xmm7
+	vpaddd	xmm6,xmm2,xmm7
+	vmovdqa	[esp],xmm4
+	mov	ebp,ecx
+	vmovdqa	[16+esp],xmm5
+	xor	ebp,edx
+	vmovdqa	[32+esp],xmm6
+	and	esi,ebp
+	jmp	NEAR L$007loop
+align	16
+L$007loop:
+	shrd	ebx,ebx,2
+	xor	esi,edx
+	vpalignr	xmm4,xmm1,xmm0,8
+	mov	ebp,eax
+	add	edi,DWORD [esp]
+	vpaddd	xmm7,xmm7,xmm3
+	vmovdqa	[64+esp],xmm0
+	xor	ebx,ecx
+	shld	eax,eax,5
+	vpsrldq	xmm6,xmm3,4
+	add	edi,esi
+	and	ebp,ebx
+	vpxor	xmm4,xmm4,xmm0
+	xor	ebx,ecx
+	add	edi,eax
+	vpxor	xmm6,xmm6,xmm2
+	shrd	eax,eax,7
+	xor	ebp,ecx
+	vmovdqa	[48+esp],xmm7
+	mov	esi,edi
+	add	edx,DWORD [4+esp]
+	vpxor	xmm4,xmm4,xmm6
+	xor	eax,ebx
+	shld	edi,edi,5
+	add	edx,ebp
+	and	esi,eax
+	vpsrld	xmm6,xmm4,31
+	xor	eax,ebx
+	add	edx,edi
+	shrd	edi,edi,7
+	xor	esi,ebx
+	vpslldq	xmm0,xmm4,12
+	vpaddd	xmm4,xmm4,xmm4
+	mov	ebp,edx
+	add	ecx,DWORD [8+esp]
+	xor	edi,eax
+	shld	edx,edx,5
+	vpsrld	xmm7,xmm0,30
+	vpor	xmm4,xmm4,xmm6
+	add	ecx,esi
+	and	ebp,edi
+	xor	edi,eax
+	add	ecx,edx
+	vpslld	xmm0,xmm0,2
+	shrd	edx,edx,7
+	xor	ebp,eax
+	vpxor	xmm4,xmm4,xmm7
+	mov	esi,ecx
+	add	ebx,DWORD [12+esp]
+	xor	edx,edi
+	shld	ecx,ecx,5
+	vpxor	xmm4,xmm4,xmm0
+	add	ebx,ebp
+	and	esi,edx
+	vmovdqa	xmm0,[96+esp]
+	xor	edx,edi
+	add	ebx,ecx
+	shrd	ecx,ecx,7
+	xor	esi,edi
+	vpalignr	xmm5,xmm2,xmm1,8
+	mov	ebp,ebx
+	add	eax,DWORD [16+esp]
+	vpaddd	xmm0,xmm0,xmm4
+	vmovdqa	[80+esp],xmm1
+	xor	ecx,edx
+	shld	ebx,ebx,5
+	vpsrldq	xmm7,xmm4,4
+	add	eax,esi
+	and	ebp,ecx
+	vpxor	xmm5,xmm5,xmm1
+	xor	ecx,edx
+	add	eax,ebx
+	vpxor	xmm7,xmm7,xmm3
+	shrd	ebx,ebx,7
+	xor	ebp,edx
+	vmovdqa	[esp],xmm0
+	mov	esi,eax
+	add	edi,DWORD [20+esp]
+	vpxor	xmm5,xmm5,xmm7
+	xor	ebx,ecx
+	shld	eax,eax,5
+	add	edi,ebp
+	and	esi,ebx
+	vpsrld	xmm7,xmm5,31
+	xor	ebx,ecx
+	add	edi,eax
+	shrd	eax,eax,7
+	xor	esi,ecx
+	vpslldq	xmm1,xmm5,12
+	vpaddd	xmm5,xmm5,xmm5
+	mov	ebp,edi
+	add	edx,DWORD [24+esp]
+	xor	eax,ebx
+	shld	edi,edi,5
+	vpsrld	xmm0,xmm1,30
+	vpor	xmm5,xmm5,xmm7
+	add	edx,esi
+	and	ebp,eax
+	xor	eax,ebx
+	add	edx,edi
+	vpslld	xmm1,xmm1,2
+	shrd	edi,edi,7
+	xor	ebp,ebx
+	vpxor	xmm5,xmm5,xmm0
+	mov	esi,edx
+	add	ecx,DWORD [28+esp]
+	xor	edi,eax
+	shld	edx,edx,5
+	vpxor	xmm5,xmm5,xmm1
+	add	ecx,ebp
+	and	esi,edi
+	vmovdqa	xmm1,[112+esp]
+	xor	edi,eax
+	add	ecx,edx
+	shrd	edx,edx,7
+	xor	esi,eax
+	vpalignr	xmm6,xmm3,xmm2,8
+	mov	ebp,ecx
+	add	ebx,DWORD [32+esp]
+	vpaddd	xmm1,xmm1,xmm5
+	vmovdqa	[96+esp],xmm2
+	xor	edx,edi
+	shld	ecx,ecx,5
+	vpsrldq	xmm0,xmm5,4
+	add	ebx,esi
+	and	ebp,edx
+	vpxor	xmm6,xmm6,xmm2
+	xor	edx,edi
+	add	ebx,ecx
+	vpxor	xmm0,xmm0,xmm4
+	shrd	ecx,ecx,7
+	xor	ebp,edi
+	vmovdqa	[16+esp],xmm1
+	mov	esi,ebx
+	add	eax,DWORD [36+esp]
+	vpxor	xmm6,xmm6,xmm0
+	xor	ecx,edx
+	shld	ebx,ebx,5
+	add	eax,ebp
+	and	esi,ecx
+	vpsrld	xmm0,xmm6,31
+	xor	ecx,edx
+	add	eax,ebx
+	shrd	ebx,ebx,7
+	xor	esi,edx
+	vpslldq	xmm2,xmm6,12
+	vpaddd	xmm6,xmm6,xmm6
+	mov	ebp,eax
+	add	edi,DWORD [40+esp]
+	xor	ebx,ecx
+	shld	eax,eax,5
+	vpsrld	xmm1,xmm2,30
+	vpor	xmm6,xmm6,xmm0
+	add	edi,esi
+	and	ebp,ebx
+	xor	ebx,ecx
+	add	edi,eax
+	vpslld	xmm2,xmm2,2
+	vmovdqa	xmm0,[64+esp]
+	shrd	eax,eax,7
+	xor	ebp,ecx
+	vpxor	xmm6,xmm6,xmm1
+	mov	esi,edi
+	add	edx,DWORD [44+esp]
+	xor	eax,ebx
+	shld	edi,edi,5
+	vpxor	xmm6,xmm6,xmm2
+	add	edx,ebp
+	and	esi,eax
+	vmovdqa	xmm2,[112+esp]
+	xor	eax,ebx
+	add	edx,edi
+	shrd	edi,edi,7
+	xor	esi,ebx
+	vpalignr	xmm7,xmm4,xmm3,8
+	mov	ebp,edx
+	add	ecx,DWORD [48+esp]
+	vpaddd	xmm2,xmm2,xmm6
+	vmovdqa	[64+esp],xmm3
+	xor	edi,eax
+	shld	edx,edx,5
+	vpsrldq	xmm1,xmm6,4
+	add	ecx,esi
+	and	ebp,edi
+	vpxor	xmm7,xmm7,xmm3
+	xor	edi,eax
+	add	ecx,edx
+	vpxor	xmm1,xmm1,xmm5
+	shrd	edx,edx,7
+	xor	ebp,eax
+	vmovdqa	[32+esp],xmm2
+	mov	esi,ecx
+	add	ebx,DWORD [52+esp]
+	vpxor	xmm7,xmm7,xmm1
+	xor	edx,edi
+	shld	ecx,ecx,5
+	add	ebx,ebp
+	and	esi,edx
+	vpsrld	xmm1,xmm7,31
+	xor	edx,edi
+	add	ebx,ecx
+	shrd	ecx,ecx,7
+	xor	esi,edi
+	vpslldq	xmm3,xmm7,12
+	vpaddd	xmm7,xmm7,xmm7
+	mov	ebp,ebx
+	add	eax,DWORD [56+esp]
+	xor	ecx,edx
+	shld	ebx,ebx,5
+	vpsrld	xmm2,xmm3,30
+	vpor	xmm7,xmm7,xmm1
+	add	eax,esi
+	and	ebp,ecx
+	xor	ecx,edx
+	add	eax,ebx
+	vpslld	xmm3,xmm3,2
+	vmovdqa	xmm1,[80+esp]
+	shrd	ebx,ebx,7
+	xor	ebp,edx
+	vpxor	xmm7,xmm7,xmm2
+	mov	esi,eax
+	add	edi,DWORD [60+esp]
+	xor	ebx,ecx
+	shld	eax,eax,5
+	vpxor	xmm7,xmm7,xmm3
+	add	edi,ebp
+	and	esi,ebx
+	vmovdqa	xmm3,[112+esp]
+	xor	ebx,ecx
+	add	edi,eax
+	vpalignr	xmm2,xmm7,xmm6,8
+	vpxor	xmm0,xmm0,xmm4
+	shrd	eax,eax,7
+	xor	esi,ecx
+	mov	ebp,edi
+	add	edx,DWORD [esp]
+	vpxor	xmm0,xmm0,xmm1
+	vmovdqa	[80+esp],xmm4
+	xor	eax,ebx
+	shld	edi,edi,5
+	vmovdqa	xmm4,xmm3
+	vpaddd	xmm3,xmm3,xmm7
+	add	edx,esi
+	and	ebp,eax
+	vpxor	xmm0,xmm0,xmm2
+	xor	eax,ebx
+	add	edx,edi
+	shrd	edi,edi,7
+	xor	ebp,ebx
+	vpsrld	xmm2,xmm0,30
+	vmovdqa	[48+esp],xmm3
+	mov	esi,edx
+	add	ecx,DWORD [4+esp]
+	xor	edi,eax
+	shld	edx,edx,5
+	vpslld	xmm0,xmm0,2
+	add	ecx,ebp
+	and	esi,edi
+	xor	edi,eax
+	add	ecx,edx
+	shrd	edx,edx,7
+	xor	esi,eax
+	mov	ebp,ecx
+	add	ebx,DWORD [8+esp]
+	vpor	xmm0,xmm0,xmm2
+	xor	edx,edi
+	shld	ecx,ecx,5
+	vmovdqa	xmm2,[96+esp]
+	add	ebx,esi
+	and	ebp,edx
+	xor	edx,edi
+	add	ebx,ecx
+	add	eax,DWORD [12+esp]
+	xor	ebp,edi
+	mov	esi,ebx
+	shld	ebx,ebx,5
+	add	eax,ebp
+	xor	esi,edx
+	shrd	ecx,ecx,7
+	add	eax,ebx
+	vpalignr	xmm3,xmm0,xmm7,8
+	vpxor	xmm1,xmm1,xmm5
+	add	edi,DWORD [16+esp]
+	xor	esi,ecx
+	mov	ebp,eax
+	shld	eax,eax,5
+	vpxor	xmm1,xmm1,xmm2
+	vmovdqa	[96+esp],xmm5
+	add	edi,esi
+	xor	ebp,ecx
+	vmovdqa	xmm5,xmm4
+	vpaddd	xmm4,xmm4,xmm0
+	shrd	ebx,ebx,7
+	add	edi,eax
+	vpxor	xmm1,xmm1,xmm3
+	add	edx,DWORD [20+esp]
+	xor	ebp,ebx
+	mov	esi,edi
+	shld	edi,edi,5
+	vpsrld	xmm3,xmm1,30
+	vmovdqa	[esp],xmm4
+	add	edx,ebp
+	xor	esi,ebx
+	shrd	eax,eax,7
+	add	edx,edi
+	vpslld	xmm1,xmm1,2
+	add	ecx,DWORD [24+esp]
+	xor	esi,eax
+	mov	ebp,edx
+	shld	edx,edx,5
+	add	ecx,esi
+	xor	ebp,eax
+	shrd	edi,edi,7
+	add	ecx,edx
+	vpor	xmm1,xmm1,xmm3
+	add	ebx,DWORD [28+esp]
+	xor	ebp,edi
+	vmovdqa	xmm3,[64+esp]
+	mov	esi,ecx
+	shld	ecx,ecx,5
+	add	ebx,ebp
+	xor	esi,edi
+	shrd	edx,edx,7
+	add	ebx,ecx
+	vpalignr	xmm4,xmm1,xmm0,8
+	vpxor	xmm2,xmm2,xmm6
+	add	eax,DWORD [32+esp]
+	xor	esi,edx
+	mov	ebp,ebx
+	shld	ebx,ebx,5
+	vpxor	xmm2,xmm2,xmm3
+	vmovdqa	[64+esp],xmm6
+	add	eax,esi
+	xor	ebp,edx
+	vmovdqa	xmm6,[128+esp]
+	vpaddd	xmm5,xmm5,xmm1
+	shrd	ecx,ecx,7
+	add	eax,ebx
+	vpxor	xmm2,xmm2,xmm4
+	add	edi,DWORD [36+esp]
+	xor	ebp,ecx
+	mov	esi,eax
+	shld	eax,eax,5
+	vpsrld	xmm4,xmm2,30
+	vmovdqa	[16+esp],xmm5
+	add	edi,ebp
+	xor	esi,ecx
+	shrd	ebx,ebx,7
+	add	edi,eax
+	vpslld	xmm2,xmm2,2
+	add	edx,DWORD [40+esp]
+	xor	esi,ebx
+	mov	ebp,edi
+	shld	edi,edi,5
+	add	edx,esi
+	xor	ebp,ebx
+	shrd	eax,eax,7
+	add	edx,edi
+	vpor	xmm2,xmm2,xmm4
+	add	ecx,DWORD [44+esp]
+	xor	ebp,eax
+	vmovdqa	xmm4,[80+esp]
+	mov	esi,edx
+	shld	edx,edx,5
+	add	ecx,ebp
+	xor	esi,eax
+	shrd	edi,edi,7
+	add	ecx,edx
+	vpalignr	xmm5,xmm2,xmm1,8
+	vpxor	xmm3,xmm3,xmm7
+	add	ebx,DWORD [48+esp]
+	xor	esi,edi
+	mov	ebp,ecx
+	shld	ecx,ecx,5
+	vpxor	xmm3,xmm3,xmm4
+	vmovdqa	[80+esp],xmm7
+	add	ebx,esi
+	xor	ebp,edi
+	vmovdqa	xmm7,xmm6
+	vpaddd	xmm6,xmm6,xmm2
+	shrd	edx,edx,7
+	add	ebx,ecx
+	vpxor	xmm3,xmm3,xmm5
+	add	eax,DWORD [52+esp]
+	xor	ebp,edx
+	mov	esi,ebx
+	shld	ebx,ebx,5
+	vpsrld	xmm5,xmm3,30
+	vmovdqa	[32+esp],xmm6
+	add	eax,ebp
+	xor	esi,edx
+	shrd	ecx,ecx,7
+	add	eax,ebx
+	vpslld	xmm3,xmm3,2
+	add	edi,DWORD [56+esp]
+	xor	esi,ecx
+	mov	ebp,eax
+	shld	eax,eax,5
+	add	edi,esi
+	xor	ebp,ecx
+	shrd	ebx,ebx,7
+	add	edi,eax
+	vpor	xmm3,xmm3,xmm5
+	add	edx,DWORD [60+esp]
+	xor	ebp,ebx
+	vmovdqa	xmm5,[96+esp]
+	mov	esi,edi
+	shld	edi,edi,5
+	add	edx,ebp
+	xor	esi,ebx
+	shrd	eax,eax,7
+	add	edx,edi
+	vpalignr	xmm6,xmm3,xmm2,8
+	vpxor	xmm4,xmm4,xmm0
+	add	ecx,DWORD [esp]
+	xor	esi,eax
+	mov	ebp,edx
+	shld	edx,edx,5
+	vpxor	xmm4,xmm4,xmm5
+	vmovdqa	[96+esp],xmm0
+	add	ecx,esi
+	xor	ebp,eax
+	vmovdqa	xmm0,xmm7
+	vpaddd	xmm7,xmm7,xmm3
+	shrd	edi,edi,7
+	add	ecx,edx
+	vpxor	xmm4,xmm4,xmm6
+	add	ebx,DWORD [4+esp]
+	xor	ebp,edi
+	mov	esi,ecx
+	shld	ecx,ecx,5
+	vpsrld	xmm6,xmm4,30
+	vmovdqa	[48+esp],xmm7
+	add	ebx,ebp
+	xor	esi,edi
+	shrd	edx,edx,7
+	add	ebx,ecx
+	vpslld	xmm4,xmm4,2
+	add	eax,DWORD [8+esp]
+	xor	esi,edx
+	mov	ebp,ebx
+	shld	ebx,ebx,5
+	add	eax,esi
+	xor	ebp,edx
+	shrd	ecx,ecx,7
+	add	eax,ebx
+	vpor	xmm4,xmm4,xmm6
+	add	edi,DWORD [12+esp]
+	xor	ebp,ecx
+	vmovdqa	xmm6,[64+esp]
+	mov	esi,eax
+	shld	eax,eax,5
+	add	edi,ebp
+	xor	esi,ecx
+	shrd	ebx,ebx,7
+	add	edi,eax
+	vpalignr	xmm7,xmm4,xmm3,8
+	vpxor	xmm5,xmm5,xmm1
+	add	edx,DWORD [16+esp]
+	xor	esi,ebx
+	mov	ebp,edi
+	shld	edi,edi,5
+	vpxor	xmm5,xmm5,xmm6
+	vmovdqa	[64+esp],xmm1
+	add	edx,esi
+	xor	ebp,ebx
+	vmovdqa	xmm1,xmm0
+	vpaddd	xmm0,xmm0,xmm4
+	shrd	eax,eax,7
+	add	edx,edi
+	vpxor	xmm5,xmm5,xmm7
+	add	ecx,DWORD [20+esp]
+	xor	ebp,eax
+	mov	esi,edx
+	shld	edx,edx,5
+	vpsrld	xmm7,xmm5,30
+	vmovdqa	[esp],xmm0
+	add	ecx,ebp
+	xor	esi,eax
+	shrd	edi,edi,7
+	add	ecx,edx
+	vpslld	xmm5,xmm5,2
+	add	ebx,DWORD [24+esp]
+	xor	esi,edi
+	mov	ebp,ecx
+	shld	ecx,ecx,5
+	add	ebx,esi
+	xor	ebp,edi
+	shrd	edx,edx,7
+	add	ebx,ecx
+	vpor	xmm5,xmm5,xmm7
+	add	eax,DWORD [28+esp]
+	vmovdqa	xmm7,[80+esp]
+	shrd	ecx,ecx,7
+	mov	esi,ebx
+	xor	ebp,edx
+	shld	ebx,ebx,5
+	add	eax,ebp
+	xor	esi,ecx
+	xor	ecx,edx
+	add	eax,ebx
+	vpalignr	xmm0,xmm5,xmm4,8
+	vpxor	xmm6,xmm6,xmm2
+	add	edi,DWORD [32+esp]
+	and	esi,ecx
+	xor	ecx,edx
+	shrd	ebx,ebx,7
+	vpxor	xmm6,xmm6,xmm7
+	vmovdqa	[80+esp],xmm2
+	mov	ebp,eax
+	xor	esi,ecx
+	vmovdqa	xmm2,xmm1
+	vpaddd	xmm1,xmm1,xmm5
+	shld	eax,eax,5
+	add	edi,esi
+	vpxor	xmm6,xmm6,xmm0
+	xor	ebp,ebx
+	xor	ebx,ecx
+	add	edi,eax
+	add	edx,DWORD [36+esp]
+	vpsrld	xmm0,xmm6,30
+	vmovdqa	[16+esp],xmm1
+	and	ebp,ebx
+	xor	ebx,ecx
+	shrd	eax,eax,7
+	mov	esi,edi
+	vpslld	xmm6,xmm6,2
+	xor	ebp,ebx
+	shld	edi,edi,5
+	add	edx,ebp
+	xor	esi,eax
+	xor	eax,ebx
+	add	edx,edi
+	add	ecx,DWORD [40+esp]
+	and	esi,eax
+	vpor	xmm6,xmm6,xmm0
+	xor	eax,ebx
+	shrd	edi,edi,7
+	vmovdqa	xmm0,[96+esp]
+	mov	ebp,edx
+	xor	esi,eax
+	shld	edx,edx,5
+	add	ecx,esi
+	xor	ebp,edi
+	xor	edi,eax
+	add	ecx,edx
+	add	ebx,DWORD [44+esp]
+	and	ebp,edi
+	xor	edi,eax
+	shrd	edx,edx,7
+	mov	esi,ecx
+	xor	ebp,edi
+	shld	ecx,ecx,5
+	add	ebx,ebp
+	xor	esi,edx
+	xor	edx,edi
+	add	ebx,ecx
+	vpalignr	xmm1,xmm6,xmm5,8
+	vpxor	xmm7,xmm7,xmm3
+	add	eax,DWORD [48+esp]
+	and	esi,edx
+	xor	edx,edi
+	shrd	ecx,ecx,7
+	vpxor	xmm7,xmm7,xmm0
+	vmovdqa	[96+esp],xmm3
+	mov	ebp,ebx
+	xor	esi,edx
+	vmovdqa	xmm3,[144+esp]
+	vpaddd	xmm2,xmm2,xmm6
+	shld	ebx,ebx,5
+	add	eax,esi
+	vpxor	xmm7,xmm7,xmm1
+	xor	ebp,ecx
+	xor	ecx,edx
+	add	eax,ebx
+	add	edi,DWORD [52+esp]
+	vpsrld	xmm1,xmm7,30
+	vmovdqa	[32+esp],xmm2
+	and	ebp,ecx
+	xor	ecx,edx
+	shrd	ebx,ebx,7
+	mov	esi,eax
+	vpslld	xmm7,xmm7,2
+	xor	ebp,ecx
+	shld	eax,eax,5
+	add	edi,ebp
+	xor	esi,ebx
+	xor	ebx,ecx
+	add	edi,eax
+	add	edx,DWORD [56+esp]
+	and	esi,ebx
+	vpor	xmm7,xmm7,xmm1
+	xor	ebx,ecx
+	shrd	eax,eax,7
+	vmovdqa	xmm1,[64+esp]
+	mov	ebp,edi
+	xor	esi,ebx
+	shld	edi,edi,5
+	add	edx,esi
+	xor	ebp,eax
+	xor	eax,ebx
+	add	edx,edi
+	add	ecx,DWORD [60+esp]
+	and	ebp,eax
+	xor	eax,ebx
+	shrd	edi,edi,7
+	mov	esi,edx
+	xor	ebp,eax
+	shld	edx,edx,5
+	add	ecx,ebp
+	xor	esi,edi
+	xor	edi,eax
+	add	ecx,edx
+	vpalignr	xmm2,xmm7,xmm6,8
+	vpxor	xmm0,xmm0,xmm4
+	add	ebx,DWORD [esp]
+	and	esi,edi
+	xor	edi,eax
+	shrd	edx,edx,7
+	vpxor	xmm0,xmm0,xmm1
+	vmovdqa	[64+esp],xmm4
+	mov	ebp,ecx
+	xor	esi,edi
+	vmovdqa	xmm4,xmm3
+	vpaddd	xmm3,xmm3,xmm7
+	shld	ecx,ecx,5
+	add	ebx,esi
+	vpxor	xmm0,xmm0,xmm2
+	xor	ebp,edx
+	xor	edx,edi
+	add	ebx,ecx
+	add	eax,DWORD [4+esp]
+	vpsrld	xmm2,xmm0,30
+	vmovdqa	[48+esp],xmm3
+	and	ebp,edx
+	xor	edx,edi
+	shrd	ecx,ecx,7
+	mov	esi,ebx
+	vpslld	xmm0,xmm0,2
+	xor	ebp,edx
+	shld	ebx,ebx,5
+	add	eax,ebp
+	xor	esi,ecx
+	xor	ecx,edx
+	add	eax,ebx
+	add	edi,DWORD [8+esp]
+	and	esi,ecx
+	vpor	xmm0,xmm0,xmm2
+	xor	ecx,edx
+	shrd	ebx,ebx,7
+	vmovdqa	xmm2,[80+esp]
+	mov	ebp,eax
+	xor	esi,ecx
+	shld	eax,eax,5
+	add	edi,esi
+	xor	ebp,ebx
+	xor	ebx,ecx
+	add	edi,eax
+	add	edx,DWORD [12+esp]
+	and	ebp,ebx
+	xor	ebx,ecx
+	shrd	eax,eax,7
+	mov	esi,edi
+	xor	ebp,ebx
+	shld	edi,edi,5
+	add	edx,ebp
+	xor	esi,eax
+	xor	eax,ebx
+	add	edx,edi
+	vpalignr	xmm3,xmm0,xmm7,8
+	vpxor	xmm1,xmm1,xmm5
+	add	ecx,DWORD [16+esp]
+	and	esi,eax
+	xor	eax,ebx
+	shrd	edi,edi,7
+	vpxor	xmm1,xmm1,xmm2
+	vmovdqa	[80+esp],xmm5
+	mov	ebp,edx
+	xor	esi,eax
+	vmovdqa	xmm5,xmm4
+	vpaddd	xmm4,xmm4,xmm0
+	shld	edx,edx,5
+	add	ecx,esi
+	vpxor	xmm1,xmm1,xmm3
+	xor	ebp,edi
+	xor	edi,eax
+	add	ecx,edx
+	add	ebx,DWORD [20+esp]
+	vpsrld	xmm3,xmm1,30
+	vmovdqa	[esp],xmm4
+	and	ebp,edi
+	xor	edi,eax
+	shrd	edx,edx,7
+	mov	esi,ecx
+	vpslld	xmm1,xmm1,2
+	xor	ebp,edi
+	shld	ecx,ecx,5
+	add	ebx,ebp
+	xor	esi,edx
+	xor	edx,edi
+	add	ebx,ecx
+	add	eax,DWORD [24+esp]
+	and	esi,edx
+	vpor	xmm1,xmm1,xmm3
+	xor	edx,edi
+	shrd	ecx,ecx,7
+	vmovdqa	xmm3,[96+esp]
+	mov	ebp,ebx
+	xor	esi,edx
+	shld	ebx,ebx,5
+	add	eax,esi
+	xor	ebp,ecx
+	xor	ecx,edx
+	add	eax,ebx
+	add	edi,DWORD [28+esp]
+	and	ebp,ecx
+	xor	ecx,edx
+	shrd	ebx,ebx,7
+	mov	esi,eax
+	xor	ebp,ecx
+	shld	eax,eax,5
+	add	edi,ebp
+	xor	esi,ebx
+	xor	ebx,ecx
+	add	edi,eax
+	vpalignr	xmm4,xmm1,xmm0,8
+	vpxor	xmm2,xmm2,xmm6
+	add	edx,DWORD [32+esp]
+	and	esi,ebx
+	xor	ebx,ecx
+	shrd	eax,eax,7
+	vpxor	xmm2,xmm2,xmm3
+	vmovdqa	[96+esp],xmm6
+	mov	ebp,edi
+	xor	esi,ebx
+	vmovdqa	xmm6,xmm5
+	vpaddd	xmm5,xmm5,xmm1
+	shld	edi,edi,5
+	add	edx,esi
+	vpxor	xmm2,xmm2,xmm4
+	xor	ebp,eax
+	xor	eax,ebx
+	add	edx,edi
+	add	ecx,DWORD [36+esp]
+	vpsrld	xmm4,xmm2,30
+	vmovdqa	[16+esp],xmm5
+	and	ebp,eax
+	xor	eax,ebx
+	shrd	edi,edi,7
+	mov	esi,edx
+	vpslld	xmm2,xmm2,2
+	xor	ebp,eax
+	shld	edx,edx,5
+	add	ecx,ebp
+	xor	esi,edi
+	xor	edi,eax
+	add	ecx,edx
+	add	ebx,DWORD [40+esp]
+	and	esi,edi
+	vpor	xmm2,xmm2,xmm4
+	xor	edi,eax
+	shrd	edx,edx,7
+	vmovdqa	xmm4,[64+esp]
+	mov	ebp,ecx
+	xor	esi,edi
+	shld	ecx,ecx,5
+	add	ebx,esi
+	xor	ebp,edx
+	xor	edx,edi
+	add	ebx,ecx
+	add	eax,DWORD [44+esp]
+	and	ebp,edx
+	xor	edx,edi
+	shrd	ecx,ecx,7
+	mov	esi,ebx
+	xor	ebp,edx
+	shld	ebx,ebx,5
+	add	eax,ebp
+	xor	esi,edx
+	add	eax,ebx
+	vpalignr	xmm5,xmm2,xmm1,8
+	vpxor	xmm3,xmm3,xmm7
+	add	edi,DWORD [48+esp]
+	xor	esi,ecx
+	mov	ebp,eax
+	shld	eax,eax,5
+	vpxor	xmm3,xmm3,xmm4
+	vmovdqa	[64+esp],xmm7
+	add	edi,esi
+	xor	ebp,ecx
+	vmovdqa	xmm7,xmm6
+	vpaddd	xmm6,xmm6,xmm2
+	shrd	ebx,ebx,7
+	add	edi,eax
+	vpxor	xmm3,xmm3,xmm5
+	add	edx,DWORD [52+esp]
+	xor	ebp,ebx
+	mov	esi,edi
+	shld	edi,edi,5
+	vpsrld	xmm5,xmm3,30
+	vmovdqa	[32+esp],xmm6
+	add	edx,ebp
+	xor	esi,ebx
+	shrd	eax,eax,7
+	add	edx,edi
+	vpslld	xmm3,xmm3,2
+	add	ecx,DWORD [56+esp]
+	xor	esi,eax
+	mov	ebp,edx
+	shld	edx,edx,5
+	add	ecx,esi
+	xor	ebp,eax
+	shrd	edi,edi,7
+	add	ecx,edx
+	vpor	xmm3,xmm3,xmm5
+	add	ebx,DWORD [60+esp]
+	xor	ebp,edi
+	mov	esi,ecx
+	shld	ecx,ecx,5
+	add	ebx,ebp
+	xor	esi,edi
+	shrd	edx,edx,7
+	add	ebx,ecx
+	add	eax,DWORD [esp]
+	vpaddd	xmm7,xmm7,xmm3
+	xor	esi,edx
+	mov	ebp,ebx
+	shld	ebx,ebx,5
+	add	eax,esi
+	vmovdqa	[48+esp],xmm7
+	xor	ebp,edx
+	shrd	ecx,ecx,7
+	add	eax,ebx
+	add	edi,DWORD [4+esp]
+	xor	ebp,ecx
+	mov	esi,eax
+	shld	eax,eax,5
+	add	edi,ebp
+	xor	esi,ecx
+	shrd	ebx,ebx,7
+	add	edi,eax
+	add	edx,DWORD [8+esp]
+	xor	esi,ebx
+	mov	ebp,edi
+	shld	edi,edi,5
+	add	edx,esi
+	xor	ebp,ebx
+	shrd	eax,eax,7
+	add	edx,edi
+	add	ecx,DWORD [12+esp]
+	xor	ebp,eax
+	mov	esi,edx
+	shld	edx,edx,5
+	add	ecx,ebp
+	xor	esi,eax
+	shrd	edi,edi,7
+	add	ecx,edx
+	mov	ebp,DWORD [196+esp]
+	cmp	ebp,DWORD [200+esp]
+	je	NEAR L$008done
+	vmovdqa	xmm7,[160+esp]
+	vmovdqa	xmm6,[176+esp]
+	vmovdqu	xmm0,[ebp]
+	vmovdqu	xmm1,[16+ebp]
+	vmovdqu	xmm2,[32+ebp]
+	vmovdqu	xmm3,[48+ebp]
+	add	ebp,64
+	vpshufb	xmm0,xmm0,xmm6
+	mov	DWORD [196+esp],ebp
+	vmovdqa	[96+esp],xmm7
+	add	ebx,DWORD [16+esp]
+	xor	esi,edi
+	vpshufb	xmm1,xmm1,xmm6
+	mov	ebp,ecx
+	shld	ecx,ecx,5
+	vpaddd	xmm4,xmm0,xmm7
+	add	ebx,esi
+	xor	ebp,edi
+	shrd	edx,edx,7
+	add	ebx,ecx
+	vmovdqa	[esp],xmm4
+	add	eax,DWORD [20+esp]
+	xor	ebp,edx
+	mov	esi,ebx
+	shld	ebx,ebx,5
+	add	eax,ebp
+	xor	esi,edx
+	shrd	ecx,ecx,7
+	add	eax,ebx
+	add	edi,DWORD [24+esp]
+	xor	esi,ecx
+	mov	ebp,eax
+	shld	eax,eax,5
+	add	edi,esi
+	xor	ebp,ecx
+	shrd	ebx,ebx,7
+	add	edi,eax
+	add	edx,DWORD [28+esp]
+	xor	ebp,ebx
+	mov	esi,edi
+	shld	edi,edi,5
+	add	edx,ebp
+	xor	esi,ebx
+	shrd	eax,eax,7
+	add	edx,edi
+	add	ecx,DWORD [32+esp]
+	xor	esi,eax
+	vpshufb	xmm2,xmm2,xmm6
+	mov	ebp,edx
+	shld	edx,edx,5
+	vpaddd	xmm5,xmm1,xmm7
+	add	ecx,esi
+	xor	ebp,eax
+	shrd	edi,edi,7
+	add	ecx,edx
+	vmovdqa	[16+esp],xmm5
+	add	ebx,DWORD [36+esp]
+	xor	ebp,edi
+	mov	esi,ecx
+	shld	ecx,ecx,5
+	add	ebx,ebp
+	xor	esi,edi
+	shrd	edx,edx,7
+	add	ebx,ecx
+	add	eax,DWORD [40+esp]
+	xor	esi,edx
+	mov	ebp,ebx
+	shld	ebx,ebx,5
+	add	eax,esi
+	xor	ebp,edx
+	shrd	ecx,ecx,7
+	add	eax,ebx
+	add	edi,DWORD [44+esp]
+	xor	ebp,ecx
+	mov	esi,eax
+	shld	eax,eax,5
+	add	edi,ebp
+	xor	esi,ecx
+	shrd	ebx,ebx,7
+	add	edi,eax
+	add	edx,DWORD [48+esp]
+	xor	esi,ebx
+	vpshufb	xmm3,xmm3,xmm6
+	mov	ebp,edi
+	shld	edi,edi,5
+	vpaddd	xmm6,xmm2,xmm7
+	add	edx,esi
+	xor	ebp,ebx
+	shrd	eax,eax,7
+	add	edx,edi
+	vmovdqa	[32+esp],xmm6
+	add	ecx,DWORD [52+esp]
+	xor	ebp,eax
+	mov	esi,edx
+	shld	edx,edx,5
+	add	ecx,ebp
+	xor	esi,eax
+	shrd	edi,edi,7
+	add	ecx,edx
+	add	ebx,DWORD [56+esp]
+	xor	esi,edi
+	mov	ebp,ecx
+	shld	ecx,ecx,5
+	add	ebx,esi
+	xor	ebp,edi
+	shrd	edx,edx,7
+	add	ebx,ecx
+	add	eax,DWORD [60+esp]
+	xor	ebp,edx
+	mov	esi,ebx
+	shld	ebx,ebx,5
+	add	eax,ebp
+	shrd	ecx,ecx,7
+	add	eax,ebx
+	mov	ebp,DWORD [192+esp]
+	add	eax,DWORD [ebp]
+	add	esi,DWORD [4+ebp]
+	add	ecx,DWORD [8+ebp]
+	mov	DWORD [ebp],eax
+	add	edx,DWORD [12+ebp]
+	mov	DWORD [4+ebp],esi
+	add	edi,DWORD [16+ebp]
+	mov	ebx,ecx
+	mov	DWORD [8+ebp],ecx
+	xor	ebx,edx
+	mov	DWORD [12+ebp],edx
+	mov	DWORD [16+ebp],edi
+	mov	ebp,esi
+	and	esi,ebx
+	mov	ebx,ebp
+	jmp	NEAR L$007loop
+align	16
+L$008done:
+	add	ebx,DWORD [16+esp]
+	xor	esi,edi
+	mov	ebp,ecx
+	shld	ecx,ecx,5
+	add	ebx,esi
+	xor	ebp,edi
+	shrd	edx,edx,7
+	add	ebx,ecx
+	add	eax,DWORD [20+esp]
+	xor	ebp,edx
+	mov	esi,ebx
+	shld	ebx,ebx,5
+	add	eax,ebp
+	xor	esi,edx
+	shrd	ecx,ecx,7
+	add	eax,ebx
+	add	edi,DWORD [24+esp]
+	xor	esi,ecx
+	mov	ebp,eax
+	shld	eax,eax,5
+	add	edi,esi
+	xor	ebp,ecx
+	shrd	ebx,ebx,7
+	add	edi,eax
+	add	edx,DWORD [28+esp]
+	xor	ebp,ebx
+	mov	esi,edi
+	shld	edi,edi,5
+	add	edx,ebp
+	xor	esi,ebx
+	shrd	eax,eax,7
+	add	edx,edi
+	add	ecx,DWORD [32+esp]
+	xor	esi,eax
+	mov	ebp,edx
+	shld	edx,edx,5
+	add	ecx,esi
+	xor	ebp,eax
+	shrd	edi,edi,7
+	add	ecx,edx
+	add	ebx,DWORD [36+esp]
+	xor	ebp,edi
+	mov	esi,ecx
+	shld	ecx,ecx,5
+	add	ebx,ebp
+	xor	esi,edi
+	shrd	edx,edx,7
+	add	ebx,ecx
+	add	eax,DWORD [40+esp]
+	xor	esi,edx
+	mov	ebp,ebx
+	shld	ebx,ebx,5
+	add	eax,esi
+	xor	ebp,edx
+	shrd	ecx,ecx,7
+	add	eax,ebx
+	add	edi,DWORD [44+esp]
+	xor	ebp,ecx
+	mov	esi,eax
+	shld	eax,eax,5
+	add	edi,ebp
+	xor	esi,ecx
+	shrd	ebx,ebx,7
+	add	edi,eax
+	add	edx,DWORD [48+esp]
+	xor	esi,ebx
+	mov	ebp,edi
+	shld	edi,edi,5
+	add	edx,esi
+	xor	ebp,ebx
+	shrd	eax,eax,7
+	add	edx,edi
+	add	ecx,DWORD [52+esp]
+	xor	ebp,eax
+	mov	esi,edx
+	shld	edx,edx,5
+	add	ecx,ebp
+	xor	esi,eax
+	shrd	edi,edi,7
+	add	ecx,edx
+	add	ebx,DWORD [56+esp]
+	xor	esi,edi
+	mov	ebp,ecx
+	shld	ecx,ecx,5
+	add	ebx,esi
+	xor	ebp,edi
+	shrd	edx,edx,7
+	add	ebx,ecx
+	add	eax,DWORD [60+esp]
+	xor	ebp,edx
+	mov	esi,ebx
+	shld	ebx,ebx,5
+	add	eax,ebp
+	shrd	ecx,ecx,7
+	add	eax,ebx
+	vzeroall
+	mov	ebp,DWORD [192+esp]
+	add	eax,DWORD [ebp]
+	mov	esp,DWORD [204+esp]
+	add	esi,DWORD [4+ebp]
+	add	ecx,DWORD [8+ebp]
+	mov	DWORD [ebp],eax
+	add	edx,DWORD [12+ebp]
+	mov	DWORD [4+ebp],esi
+	add	edi,DWORD [16+ebp]
+	mov	DWORD [8+ebp],ecx
+	mov	DWORD [12+ebp],edx
+	mov	DWORD [16+ebp],edi
+	pop	edi
+	pop	esi
+	pop	ebx
+	pop	ebp
+	ret
 align	64
 L$K_XX_XX:
 dd	1518500249,1518500249,1518500249,1518500249

third_party/boringssl/win-x86/crypto/sha/sha256-586.asm

diff --git a/third_party/boringssl/win-x86/crypto/sha/sha256-586.asm b/third_party/boringssl/win-x86/crypto/sha/sha256-586.asm
index fe36bc5..3e7cfcc 100644
--- a/third_party/boringssl/win-x86/crypto/sha/sha256-586.asm
+++ b/third_party/boringssl/win-x86/crypto/sha/sha256-586.asm
@@ -49,11 +49,10 @@
 	jz	NEAR L$003no_xmm
 	and	ecx,1073741824
 	and	ebx,268435968
-	test	edx,536870912
-	jnz	NEAR L$004shaext
 	or	ecx,ebx
 	and	ecx,1342177280
 	cmp	ecx,1342177280
+	je	NEAR L$004AVX
 	test	ebx,512
 	jnz	NEAR L$005SSSE3
 L$003no_xmm:
@@ -3179,204 +3178,6 @@
 	pop	ebp
 	ret
 align	32
-L$004shaext:
-	sub	esp,32
-	movdqu	xmm1,[esi]
-	lea	ebp,[128+ebp]
-	movdqu	xmm2,[16+esi]
-	movdqa	xmm7,[128+ebp]
-	pshufd	xmm0,xmm1,27
-	pshufd	xmm1,xmm1,177
-	pshufd	xmm2,xmm2,27
-db	102,15,58,15,202,8
-	punpcklqdq	xmm2,xmm0
-	jmp	NEAR L$010loop_shaext
-align	16
-L$010loop_shaext:
-	movdqu	xmm3,[edi]
-	movdqu	xmm4,[16+edi]
-	movdqu	xmm5,[32+edi]
-db	102,15,56,0,223
-	movdqu	xmm6,[48+edi]
-	movdqa	[16+esp],xmm2
-	movdqa	xmm0,[ebp-128]
-	paddd	xmm0,xmm3
-db	102,15,56,0,231
-db	15,56,203,209
-	pshufd	xmm0,xmm0,14
-	nop
-	movdqa	[esp],xmm1
-db	15,56,203,202
-	movdqa	xmm0,[ebp-112]
-	paddd	xmm0,xmm4
-db	102,15,56,0,239
-db	15,56,203,209
-	pshufd	xmm0,xmm0,14
-	lea	edi,[64+edi]
-db	15,56,204,220
-db	15,56,203,202
-	movdqa	xmm0,[ebp-96]
-	paddd	xmm0,xmm5
-db	102,15,56,0,247
-db	15,56,203,209
-	pshufd	xmm0,xmm0,14
-	movdqa	xmm7,xmm6
-db	102,15,58,15,253,4
-	nop
-	paddd	xmm3,xmm7
-db	15,56,204,229
-db	15,56,203,202
-	movdqa	xmm0,[ebp-80]
-	paddd	xmm0,xmm6
-db	15,56,205,222
-db	15,56,203,209
-	pshufd	xmm0,xmm0,14
-	movdqa	xmm7,xmm3
-db	102,15,58,15,254,4
-	nop
-	paddd	xmm4,xmm7
-db	15,56,204,238
-db	15,56,203,202
-	movdqa	xmm0,[ebp-64]
-	paddd	xmm0,xmm3
-db	15,56,205,227
-db	15,56,203,209
-	pshufd	xmm0,xmm0,14
-	movdqa	xmm7,xmm4
-db	102,15,58,15,251,4
-	nop
-	paddd	xmm5,xmm7
-db	15,56,204,243
-db	15,56,203,202
-	movdqa	xmm0,[ebp-48]
-	paddd	xmm0,xmm4
-db	15,56,205,236
-db	15,56,203,209
-	pshufd	xmm0,xmm0,14
-	movdqa	xmm7,xmm5
-db	102,15,58,15,252,4
-	nop
-	paddd	xmm6,xmm7
-db	15,56,204,220
-db	15,56,203,202
-	movdqa	xmm0,[ebp-32]
-	paddd	xmm0,xmm5
-db	15,56,205,245
-db	15,56,203,209
-	pshufd	xmm0,xmm0,14
-	movdqa	xmm7,xmm6
-db	102,15,58,15,253,4
-	nop
-	paddd	xmm3,xmm7
-db	15,56,204,229
-db	15,56,203,202
-	movdqa	xmm0,[ebp-16]
-	paddd	xmm0,xmm6
-db	15,56,205,222
-db	15,56,203,209
-	pshufd	xmm0,xmm0,14
-	movdqa	xmm7,xmm3
-db	102,15,58,15,254,4
-	nop
-	paddd	xmm4,xmm7
-db	15,56,204,238
-db	15,56,203,202
-	movdqa	xmm0,[ebp]
-	paddd	xmm0,xmm3
-db	15,56,205,227
-db	15,56,203,209
-	pshufd	xmm0,xmm0,14
-	movdqa	xmm7,xmm4
-db	102,15,58,15,251,4
-	nop
-	paddd	xmm5,xmm7
-db	15,56,204,243
-db	15,56,203,202
-	movdqa	xmm0,[16+ebp]
-	paddd	xmm0,xmm4
-db	15,56,205,236
-db	15,56,203,209
-	pshufd	xmm0,xmm0,14
-	movdqa	xmm7,xmm5
-db	102,15,58,15,252,4
-	nop
-	paddd	xmm6,xmm7
-db	15,56,204,220
-db	15,56,203,202
-	movdqa	xmm0,[32+ebp]
-	paddd	xmm0,xmm5
-db	15,56,205,245
-db	15,56,203,209
-	pshufd	xmm0,xmm0,14
-	movdqa	xmm7,xmm6
-db	102,15,58,15,253,4
-	nop
-	paddd	xmm3,xmm7
-db	15,56,204,229
-db	15,56,203,202
-	movdqa	xmm0,[48+ebp]
-	paddd	xmm0,xmm6
-db	15,56,205,222
-db	15,56,203,209
-	pshufd	xmm0,xmm0,14
-	movdqa	xmm7,xmm3
-db	102,15,58,15,254,4
-	nop
-	paddd	xmm4,xmm7
-db	15,56,204,238
-db	15,56,203,202
-	movdqa	xmm0,[64+ebp]
-	paddd	xmm0,xmm3
-db	15,56,205,227
-db	15,56,203,209
-	pshufd	xmm0,xmm0,14
-	movdqa	xmm7,xmm4
-db	102,15,58,15,251,4
-	nop
-	paddd	xmm5,xmm7
-db	15,56,204,243
-db	15,56,203,202
-	movdqa	xmm0,[80+ebp]
-	paddd	xmm0,xmm4
-db	15,56,205,236
-db	15,56,203,209
-	pshufd	xmm0,xmm0,14
-	movdqa	xmm7,xmm5
-db	102,15,58,15,252,4
-db	15,56,203,202
-	paddd	xmm6,xmm7
-	movdqa	xmm0,[96+ebp]
-	paddd	xmm0,xmm5
-db	15,56,203,209
-	pshufd	xmm0,xmm0,14
-db	15,56,205,245
-	movdqa	xmm7,[128+ebp]
-db	15,56,203,202
-	movdqa	xmm0,[112+ebp]
-	paddd	xmm0,xmm6
-	nop
-db	15,56,203,209
-	pshufd	xmm0,xmm0,14
-	cmp	eax,edi
-	nop
-db	15,56,203,202
-	paddd	xmm2,[16+esp]
-	paddd	xmm1,[esp]
-	jnz	NEAR L$010loop_shaext
-	pshufd	xmm2,xmm2,177
-	pshufd	xmm7,xmm1,27
-	pshufd	xmm1,xmm1,177
-	punpckhqdq	xmm1,xmm2
-db	102,15,58,15,215,8
-	mov	esp,DWORD [44+esp]
-	movdqu	[esi],xmm1
-	movdqu	[16+esi],xmm2
-	pop	edi
-	pop	esi
-	pop	ebx
-	pop	ebp
-	ret
-align	32
 L$005SSSE3:
 	lea	esp,[esp-96]
 	mov	eax,DWORD [esi]
@@ -3396,9 +3197,9 @@
 	mov	DWORD [24+esp],ecx
 	mov	DWORD [28+esp],esi
 	movdqa	xmm7,[256+ebp]
-	jmp	NEAR L$011grand_ssse3
+	jmp	NEAR L$010grand_ssse3
 align	16
-L$011grand_ssse3:
+L$010grand_ssse3:
 	movdqu	xmm0,[edi]
 	movdqu	xmm1,[16+edi]
 	movdqu	xmm2,[32+edi]
@@ -3421,9 +3222,9 @@
 	paddd	xmm7,xmm3
 	movdqa	[64+esp],xmm6
 	movdqa	[80+esp],xmm7
-	jmp	NEAR L$012ssse3_00_47
+	jmp	NEAR L$011ssse3_00_47
 align	16
-L$012ssse3_00_47:
+L$011ssse3_00_47:
 	add	ebp,64
 	mov	ecx,edx
 	movdqa	xmm4,xmm1
@@ -4066,7 +3867,7 @@
 	add	eax,ecx
 	movdqa	[80+esp],xmm6
 	cmp	DWORD [64+ebp],66051
-	jne	NEAR L$012ssse3_00_47
+	jne	NEAR L$011ssse3_00_47
 	mov	ecx,edx
 	ror	edx,14
 	mov	esi,DWORD [20+esp]
@@ -4580,12 +4381,1193 @@
 	movdqa	xmm7,[64+ebp]
 	sub	ebp,192
 	cmp	edi,DWORD [104+esp]
-	jb	NEAR L$011grand_ssse3
+	jb	NEAR L$010grand_ssse3
 	mov	esp,DWORD [108+esp]
 	pop	edi
 	pop	esi
 	pop	ebx
 	pop	ebp
 	ret
+align	32
+L$004AVX:
+	lea	esp,[esp-96]
+	vzeroall
+	mov	eax,DWORD [esi]
+	mov	ebx,DWORD [4+esi]
+	mov	ecx,DWORD [8+esi]
+	mov	edi,DWORD [12+esi]
+	mov	DWORD [4+esp],ebx
+	xor	ebx,ecx
+	mov	DWORD [8+esp],ecx
+	mov	DWORD [12+esp],edi
+	mov	edx,DWORD [16+esi]
+	mov	edi,DWORD [20+esi]
+	mov	ecx,DWORD [24+esi]
+	mov	esi,DWORD [28+esi]
+	mov	DWORD [20+esp],edi
+	mov	edi,DWORD [100+esp]
+	mov	DWORD [24+esp],ecx
+	mov	DWORD [28+esp],esi
+	vmovdqa	xmm7,[256+ebp]
+	jmp	NEAR L$012grand_avx
+align	32
+L$012grand_avx:
+	vmovdqu	xmm0,[edi]
+	vmovdqu	xmm1,[16+edi]
+	vmovdqu	xmm2,[32+edi]
+	vmovdqu	xmm3,[48+edi]
+	add	edi,64
+	vpshufb	xmm0,xmm0,xmm7
+	mov	DWORD [100+esp],edi
+	vpshufb	xmm1,xmm1,xmm7
+	vpshufb	xmm2,xmm2,xmm7
+	vpaddd	xmm4,xmm0,[ebp]
+	vpshufb	xmm3,xmm3,xmm7
+	vpaddd	xmm5,xmm1,[16+ebp]
+	vpaddd	xmm6,xmm2,[32+ebp]
+	vpaddd	xmm7,xmm3,[48+ebp]
+	vmovdqa	[32+esp],xmm4
+	vmovdqa	[48+esp],xmm5
+	vmovdqa	[64+esp],xmm6
+	vmovdqa	[80+esp],xmm7
+	jmp	NEAR L$013avx_00_47
+align	16
+L$013avx_00_47:
+	add	ebp,64
+	vpalignr	xmm4,xmm1,xmm0,4
+	mov	ecx,edx
+	shrd	edx,edx,14
+	mov	esi,DWORD [20+esp]
+	vpalignr	xmm7,xmm3,xmm2,4
+	xor	edx,ecx
+	mov	edi,DWORD [24+esp]
+	xor	esi,edi
+	vpsrld	xmm6,xmm4,7
+	shrd	edx,edx,5
+	and	esi,ecx
+	mov	DWORD [16+esp],ecx
+	vpaddd	xmm0,xmm0,xmm7
+	xor	edx,ecx
+	xor	edi,esi
+	shrd	edx,edx,6
+	vpsrld	xmm7,xmm4,3
+	mov	ecx,eax
+	add	edx,edi
+	mov	edi,DWORD [4+esp]
+	vpslld	xmm5,xmm4,14
+	mov	esi,eax
+	shrd	ecx,ecx,9
+	mov	DWORD [esp],eax
+	vpxor	xmm4,xmm7,xmm6
+	xor	ecx,eax
+	xor	eax,edi
+	add	edx,DWORD [28+esp]
+	vpshufd	xmm7,xmm3,250
+	shrd	ecx,ecx,11
+	and	ebx,eax
+	xor	ecx,esi
+	vpsrld	xmm6,xmm6,11
+	add	edx,DWORD [32+esp]
+	xor	ebx,edi
+	shrd	ecx,ecx,2
+	vpxor	xmm4,xmm4,xmm5
+	add	ebx,edx
+	add	edx,DWORD [12+esp]
+	add	ebx,ecx
+	vpslld	xmm5,xmm5,11
+	mov	ecx,edx
+	shrd	edx,edx,14
+	mov	esi,DWORD [16+esp]
+	vpxor	xmm4,xmm4,xmm6
+	xor	edx,ecx
+	mov	edi,DWORD [20+esp]
+	xor	esi,edi
+	vpsrld	xmm6,xmm7,10
+	shrd	edx,edx,5
+	and	esi,ecx
+	mov	DWORD [12+esp],ecx
+	vpxor	xmm4,xmm4,xmm5
+	xor	edx,ecx
+	xor	edi,esi
+	shrd	edx,edx,6
+	vpsrlq	xmm5,xmm7,17
+	mov	ecx,ebx
+	add	edx,edi
+	mov	edi,DWORD [esp]
+	vpaddd	xmm0,xmm0,xmm4
+	mov	esi,ebx
+	shrd	ecx,ecx,9
+	mov	DWORD [28+esp],ebx
+	vpxor	xmm6,xmm6,xmm5
+	xor	ecx,ebx
+	xor	ebx,edi
+	add	edx,DWORD [24+esp]
+	vpsrlq	xmm7,xmm7,19
+	shrd	ecx,ecx,11
+	and	eax,ebx
+	xor	ecx,esi
+	vpxor	xmm6,xmm6,xmm7
+	add	edx,DWORD [36+esp]
+	xor	eax,edi
+	shrd	ecx,ecx,2
+	vpshufd	xmm7,xmm6,132
+	add	eax,edx
+	add	edx,DWORD [8+esp]
+	add	eax,ecx
+	vpsrldq	xmm7,xmm7,8
+	mov	ecx,edx
+	shrd	edx,edx,14
+	mov	esi,DWORD [12+esp]
+	vpaddd	xmm0,xmm0,xmm7
+	xor	edx,ecx
+	mov	edi,DWORD [16+esp]
+	xor	esi,edi
+	vpshufd	xmm7,xmm0,80
+	shrd	edx,edx,5
+	and	esi,ecx
+	mov	DWORD [8+esp],ecx
+	vpsrld	xmm6,xmm7,10
+	xor	edx,ecx
+	xor	edi,esi
+	shrd	edx,edx,6
+	vpsrlq	xmm5,xmm7,17
+	mov	ecx,eax
+	add	edx,edi
+	mov	edi,DWORD [28+esp]
+	vpxor	xmm6,xmm6,xmm5
+	mov	esi,eax
+	shrd	ecx,ecx,9
+	mov	DWORD [24+esp],eax
+	vpsrlq	xmm7,xmm7,19
+	xor	ecx,eax
+	xor	eax,edi
+	add	edx,DWORD [20+esp]
+	vpxor	xmm6,xmm6,xmm7
+	shrd	ecx,ecx,11
+	and	ebx,eax
+	xor	ecx,esi
+	vpshufd	xmm7,xmm6,232
+	add	edx,DWORD [40+esp]
+	xor	ebx,edi
+	shrd	ecx,ecx,2
+	vpslldq	xmm7,xmm7,8
+	add	ebx,edx
+	add	edx,DWORD [4+esp]
+	add	ebx,ecx
+	vpaddd	xmm0,xmm0,xmm7
+	mov	ecx,edx
+	shrd	edx,edx,14
+	mov	esi,DWORD [8+esp]
+	vpaddd	xmm6,xmm0,[ebp]
+	xor	edx,ecx
+	mov	edi,DWORD [12+esp]
+	xor	esi,edi
+	shrd	edx,edx,5
+	and	esi,ecx
+	mov	DWORD [4+esp],ecx
+	xor	edx,ecx
+	xor	edi,esi
+	shrd	edx,edx,6
+	mov	ecx,ebx
+	add	edx,edi
+	mov	edi,DWORD [24+esp]
+	mov	esi,ebx
+	shrd	ecx,ecx,9
+	mov	DWORD [20+esp],ebx
+	xor	ecx,ebx
+	xor	ebx,edi
+	add	edx,DWORD [16+esp]
+	shrd	ecx,ecx,11
+	and	eax,ebx
+	xor	ecx,esi
+	add	edx,DWORD [44+esp]
+	xor	eax,edi
+	shrd	ecx,ecx,2
+	add	eax,edx
+	add	edx,DWORD [esp]
+	add	eax,ecx
+	vmovdqa	[32+esp],xmm6
+	vpalignr	xmm4,xmm2,xmm1,4
+	mov	ecx,edx
+	shrd	edx,edx,14
+	mov	esi,DWORD [4+esp]
+	vpalignr	xmm7,xmm0,xmm3,4
+	xor	edx,ecx
+	mov	edi,DWORD [8+esp]
+	xor	esi,edi
+	vpsrld	xmm6,xmm4,7
+	shrd	edx,edx,5
+	and	esi,ecx
+	mov	DWORD [esp],ecx
+	vpaddd	xmm1,xmm1,xmm7
+	xor	edx,ecx
+	xor	edi,esi
+	shrd	edx,edx,6
+	vpsrld	xmm7,xmm4,3
+	mov	ecx,eax
+	add	edx,edi
+	mov	edi,DWORD [20+esp]
+	vpslld	xmm5,xmm4,14
+	mov	esi,eax
+	shrd	ecx,ecx,9
+	mov	DWORD [16+esp],eax
+	vpxor	xmm4,xmm7,xmm6
+	xor	ecx,eax
+	xor	eax,edi
+	add	edx,DWORD [12+esp]
+	vpshufd	xmm7,xmm0,250
+	shrd	ecx,ecx,11
+	and	ebx,eax
+	xor	ecx,esi
+	vpsrld	xmm6,xmm6,11
+	add	edx,DWORD [48+esp]
+	xor	ebx,edi
+	shrd	ecx,ecx,2
+	vpxor	xmm4,xmm4,xmm5
+	add	ebx,edx
+	add	edx,DWORD [28+esp]
+	add	ebx,ecx
+	vpslld	xmm5,xmm5,11
+	mov	ecx,edx
+	shrd	edx,edx,14
+	mov	esi,DWORD [esp]
+	vpxor	xmm4,xmm4,xmm6
+	xor	edx,ecx
+	mov	edi,DWORD [4+esp]
+	xor	esi,edi
+	vpsrld	xmm6,xmm7,10
+	shrd	edx,edx,5
+	and	esi,ecx
+	mov	DWORD [28+esp],ecx
+	vpxor	xmm4,xmm4,xmm5
+	xor	edx,ecx
+	xor	edi,esi
+	shrd	edx,edx,6
+	vpsrlq	xmm5,xmm7,17
+	mov	ecx,ebx
+	add	edx,edi
+	mov	edi,DWORD [16+esp]
+	vpaddd	xmm1,xmm1,xmm4
+	mov	esi,ebx
+	shrd	ecx,ecx,9
+	mov	DWORD [12+esp],ebx
+	vpxor	xmm6,xmm6,xmm5
+	xor	ecx,ebx
+	xor	ebx,edi
+	add	edx,DWORD [8+esp]
+	vpsrlq	xmm7,xmm7,19
+	shrd	ecx,ecx,11
+	and	eax,ebx
+	xor	ecx,esi
+	vpxor	xmm6,xmm6,xmm7
+	add	edx,DWORD [52+esp]
+	xor	eax,edi
+	shrd	ecx,ecx,2
+	vpshufd	xmm7,xmm6,132
+	add	eax,edx
+	add	edx,DWORD [24+esp]
+	add	eax,ecx
+	vpsrldq	xmm7,xmm7,8
+	mov	ecx,edx
+	shrd	edx,edx,14
+	mov	esi,DWORD [28+esp]
+	vpaddd	xmm1,xmm1,xmm7
+	xor	edx,ecx
+	mov	edi,DWORD [esp]
+	xor	esi,edi
+	vpshufd	xmm7,xmm1,80
+	shrd	edx,edx,5
+	and	esi,ecx
+	mov	DWORD [24+esp],ecx
+	vpsrld	xmm6,xmm7,10
+	xor	edx,ecx
+	xor	edi,esi
+	shrd	edx,edx,6
+	vpsrlq	xmm5,xmm7,17
+	mov	ecx,eax
+	add	edx,edi
+	mov	edi,DWORD [12+esp]
+	vpxor	xmm6,xmm6,xmm5
+	mov	esi,eax
+	shrd	ecx,ecx,9
+	mov	DWORD [8+esp],eax
+	vpsrlq	xmm7,xmm7,19
+	xor	ecx,eax
+	xor	eax,edi
+	add	edx,DWORD [4+esp]
+	vpxor	xmm6,xmm6,xmm7
+	shrd	ecx,ecx,11
+	and	ebx,eax
+	xor	ecx,esi
+	vpshufd	xmm7,xmm6,232
+	add	edx,DWORD [56+esp]
+	xor	ebx,edi
+	shrd	ecx,ecx,2
+	vpslldq	xmm7,xmm7,8
+	add	ebx,edx
+	add	edx,DWORD [20+esp]
+	add	ebx,ecx
+	vpaddd	xmm1,xmm1,xmm7
+	mov	ecx,edx
+	shrd	edx,edx,14
+	mov	esi,DWORD [24+esp]
+	vpaddd	xmm6,xmm1,[16+ebp]
+	xor	edx,ecx
+	mov	edi,DWORD [28+esp]
+	xor	esi,edi
+	shrd	edx,edx,5
+	and	esi,ecx
+	mov	DWORD [20+esp],ecx
+	xor	edx,ecx
+	xor	edi,esi
+	shrd	edx,edx,6
+	mov	ecx,ebx
+	add	edx,edi
+	mov	edi,DWORD [8+esp]
+	mov	esi,ebx
+	shrd	ecx,ecx,9
+	mov	DWORD [4+esp],ebx
+	xor	ecx,ebx
+	xor	ebx,edi
+	add	edx,DWORD [esp]
+	shrd	ecx,ecx,11
+	and	eax,ebx
+	xor	ecx,esi
+	add	edx,DWORD [60+esp]
+	xor	eax,edi
+	shrd	ecx,ecx,2
+	add	eax,edx
+	add	edx,DWORD [16+esp]
+	add	eax,ecx
+	vmovdqa	[48+esp],xmm6
+	vpalignr	xmm4,xmm3,xmm2,4
+	mov	ecx,edx
+	shrd	edx,edx,14
+	mov	esi,DWORD [20+esp]
+	vpalignr	xmm7,xmm1,xmm0,4
+	xor	edx,ecx
+	mov	edi,DWORD [24+esp]
+	xor	esi,edi
+	vpsrld	xmm6,xmm4,7
+	shrd	edx,edx,5
+	and	esi,ecx
+	mov	DWORD [16+esp],ecx
+	vpaddd	xmm2,xmm2,xmm7
+	xor	edx,ecx
+	xor	edi,esi
+	shrd	edx,edx,6
+	vpsrld	xmm7,xmm4,3
+	mov	ecx,eax
+	add	edx,edi
+	mov	edi,DWORD [4+esp]
+	vpslld	xmm5,xmm4,14
+	mov	esi,eax
+	shrd	ecx,ecx,9
+	mov	DWORD [esp],eax
+	vpxor	xmm4,xmm7,xmm6
+	xor	ecx,eax
+	xor	eax,edi
+	add	edx,DWORD [28+esp]
+	vpshufd	xmm7,xmm1,250
+	shrd	ecx,ecx,11
+	and	ebx,eax
+	xor	ecx,esi
+	vpsrld	xmm6,xmm6,11
+	add	edx,DWORD [64+esp]
+	xor	ebx,edi
+	shrd	ecx,ecx,2
+	vpxor	xmm4,xmm4,xmm5
+	add	ebx,edx
+	add	edx,DWORD [12+esp]
+	add	ebx,ecx
+	vpslld	xmm5,xmm5,11
+	mov	ecx,edx
+	shrd	edx,edx,14
+	mov	esi,DWORD [16+esp]
+	vpxor	xmm4,xmm4,xmm6
+	xor	edx,ecx
+	mov	edi,DWORD [20+esp]
+	xor	esi,edi
+	vpsrld	xmm6,xmm7,10
+	shrd	edx,edx,5
+	and	esi,ecx
+	mov	DWORD [12+esp],ecx
+	vpxor	xmm4,xmm4,xmm5
+	xor	edx,ecx
+	xor	edi,esi
+	shrd	edx,edx,6
+	vpsrlq	xmm5,xmm7,17
+	mov	ecx,ebx
+	add	edx,edi
+	mov	edi,DWORD [esp]
+	vpaddd	xmm2,xmm2,xmm4
+	mov	esi,ebx
+	shrd	ecx,ecx,9
+	mov	DWORD [28+esp],ebx
+	vpxor	xmm6,xmm6,xmm5
+	xor	ecx,ebx
+	xor	ebx,edi
+	add	edx,DWORD [24+esp]
+	vpsrlq	xmm7,xmm7,19
+	shrd	ecx,ecx,11
+	and	eax,ebx
+	xor	ecx,esi
+	vpxor	xmm6,xmm6,xmm7
+	add	edx,DWORD [68+esp]
+	xor	eax,edi
+	shrd	ecx,ecx,2
+	vpshufd	xmm7,xmm6,132
+	add	eax,edx
+	add	edx,DWORD [8+esp]
+	add	eax,ecx
+	vpsrldq	xmm7,xmm7,8
+	mov	ecx,edx
+	shrd	edx,edx,14
+	mov	esi,DWORD [12+esp]
+	vpaddd	xmm2,xmm2,xmm7
+	xor	edx,ecx
+	mov	edi,DWORD [16+esp]
+	xor	esi,edi
+	vpshufd	xmm7,xmm2,80
+	shrd	edx,edx,5
+	and	esi,ecx
+	mov	DWORD [8+esp],ecx
+	vpsrld	xmm6,xmm7,10
+	xor	edx,ecx
+	xor	edi,esi
+	shrd	edx,edx,6
+	vpsrlq	xmm5,xmm7,17
+	mov	ecx,eax
+	add	edx,edi
+	mov	edi,DWORD [28+esp]
+	vpxor	xmm6,xmm6,xmm5
+	mov	esi,eax
+	shrd	ecx,ecx,9
+	mov	DWORD [24+esp],eax
+	vpsrlq	xmm7,xmm7,19
+	xor	ecx,eax
+	xor	eax,edi
+	add	edx,DWORD [20+esp]
+	vpxor	xmm6,xmm6,xmm7
+	shrd	ecx,ecx,11
+	and	ebx,eax
+	xor	ecx,esi
+	vpshufd	xmm7,xmm6,232
+	add	edx,DWORD [72+esp]
+	xor	ebx,edi
+	shrd	ecx,ecx,2
+	vpslldq	xmm7,xmm7,8
+	add	ebx,edx
+	add	edx,DWORD [4+esp]
+	add	ebx,ecx
+	vpaddd	xmm2,xmm2,xmm7
+	mov	ecx,edx
+	shrd	edx,edx,14
+	mov	esi,DWORD [8+esp]
+	vpaddd	xmm6,xmm2,[32+ebp]
+	xor	edx,ecx
+	mov	edi,DWORD [12+esp]
+	xor	esi,edi
+	shrd	edx,edx,5
+	and	esi,ecx
+	mov	DWORD [4+esp],ecx
+	xor	edx,ecx
+	xor	edi,esi
+	shrd	edx,edx,6
+	mov	ecx,ebx
+	add	edx,edi
+	mov	edi,DWORD [24+esp]
+	mov	esi,ebx
+	shrd	ecx,ecx,9
+	mov	DWORD [20+esp],ebx
+	xor	ecx,ebx
+	xor	ebx,edi
+	add	edx,DWORD [16+esp]
+	shrd	ecx,ecx,11
+	and	eax,ebx
+	xor	ecx,esi
+	add	edx,DWORD [76+esp]
+	xor	eax,edi
+	shrd	ecx,ecx,2
+	add	eax,edx
+	add	edx,DWORD [esp]
+	add	eax,ecx
+	vmovdqa	[64+esp],xmm6
+	vpalignr	xmm4,xmm0,xmm3,4
+	mov	ecx,edx
+	shrd	edx,edx,14
+	mov	esi,DWORD [4+esp]
+	vpalignr	xmm7,xmm2,xmm1,4
+	xor	edx,ecx
+	mov	edi,DWORD [8+esp]
+	xor	esi,edi
+	vpsrld	xmm6,xmm4,7
+	shrd	edx,edx,5
+	and	esi,ecx
+	mov	DWORD [esp],ecx
+	vpaddd	xmm3,xmm3,xmm7
+	xor	edx,ecx
+	xor	edi,esi
+	shrd	edx,edx,6
+	vpsrld	xmm7,xmm4,3
+	mov	ecx,eax
+	add	edx,edi
+	mov	edi,DWORD [20+esp]
+	vpslld	xmm5,xmm4,14
+	mov	esi,eax
+	shrd	ecx,ecx,9
+	mov	DWORD [16+esp],eax
+	vpxor	xmm4,xmm7,xmm6
+	xor	ecx,eax
+	xor	eax,edi
+	add	edx,DWORD [12+esp]
+	vpshufd	xmm7,xmm2,250
+	shrd	ecx,ecx,11
+	and	ebx,eax
+	xor	ecx,esi
+	vpsrld	xmm6,xmm6,11
+	add	edx,DWORD [80+esp]
+	xor	ebx,edi
+	shrd	ecx,ecx,2
+	vpxor	xmm4,xmm4,xmm5
+	add	ebx,edx
+	add	edx,DWORD [28+esp]
+	add	ebx,ecx
+	vpslld	xmm5,xmm5,11
+	mov	ecx,edx
+	shrd	edx,edx,14
+	mov	esi,DWORD [esp]
+	vpxor	xmm4,xmm4,xmm6
+	xor	edx,ecx
+	mov	edi,DWORD [4+esp]
+	xor	esi,edi
+	vpsrld	xmm6,xmm7,10
+	shrd	edx,edx,5
+	and	esi,ecx
+	mov	DWORD [28+esp],ecx
+	vpxor	xmm4,xmm4,xmm5
+	xor	edx,ecx
+	xor	edi,esi
+	shrd	edx,edx,6
+	vpsrlq	xmm5,xmm7,17
+	mov	ecx,ebx
+	add	edx,edi
+	mov	edi,DWORD [16+esp]
+	vpaddd	xmm3,xmm3,xmm4
+	mov	esi,ebx
+	shrd	ecx,ecx,9
+	mov	DWORD [12+esp],ebx
+	vpxor	xmm6,xmm6,xmm5
+	xor	ecx,ebx
+	xor	ebx,edi
+	add	edx,DWORD [8+esp]
+	vpsrlq	xmm7,xmm7,19
+	shrd	ecx,ecx,11
+	and	eax,ebx
+	xor	ecx,esi
+	vpxor	xmm6,xmm6,xmm7
+	add	edx,DWORD [84+esp]
+	xor	eax,edi
+	shrd	ecx,ecx,2
+	vpshufd	xmm7,xmm6,132
+	add	eax,edx
+	add	edx,DWORD [24+esp]
+	add	eax,ecx
+	vpsrldq	xmm7,xmm7,8
+	mov	ecx,edx
+	shrd	edx,edx,14
+	mov	esi,DWORD [28+esp]
+	vpaddd	xmm3,xmm3,xmm7
+	xor	edx,ecx
+	mov	edi,DWORD [esp]
+	xor	esi,edi
+	vpshufd	xmm7,xmm3,80
+	shrd	edx,edx,5
+	and	esi,ecx
+	mov	DWORD [24+esp],ecx
+	vpsrld	xmm6,xmm7,10
+	xor	edx,ecx
+	xor	edi,esi
+	shrd	edx,edx,6
+	vpsrlq	xmm5,xmm7,17
+	mov	ecx,eax
+	add	edx,edi
+	mov	edi,DWORD [12+esp]
+	vpxor	xmm6,xmm6,xmm5
+	mov	esi,eax
+	shrd	ecx,ecx,9
+	mov	DWORD [8+esp],eax
+	vpsrlq	xmm7,xmm7,19
+	xor	ecx,eax
+	xor	eax,edi
+	add	edx,DWORD [4+esp]
+	vpxor	xmm6,xmm6,xmm7
+	shrd	ecx,ecx,11
+	and	ebx,eax
+	xor	ecx,esi
+	vpshufd	xmm7,xmm6,232
+	add	edx,DWORD [88+esp]
+	xor	ebx,edi
+	shrd	ecx,ecx,2
+	vpslldq	xmm7,xmm7,8
+	add	ebx,edx
+	add	edx,DWORD [20+esp]
+	add	ebx,ecx
+	vpaddd	xmm3,xmm3,xmm7
+	mov	ecx,edx
+	shrd	edx,edx,14
+	mov	esi,DWORD [24+esp]
+	vpaddd	xmm6,xmm3,[48+ebp]
+	xor	edx,ecx
+	mov	edi,DWORD [28+esp]
+	xor	esi,edi
+	shrd	edx,edx,5
+	and	esi,ecx
+	mov	DWORD [20+esp],ecx
+	xor	edx,ecx
+	xor	edi,esi
+	shrd	edx,edx,6
+	mov	ecx,ebx
+	add	edx,edi
+	mov	edi,DWORD [8+esp]
+	mov	esi,ebx
+	shrd	ecx,ecx,9
+	mov	DWORD [4+esp],ebx
+	xor	ecx,ebx
+	xor	ebx,edi
+	add	edx,DWORD [esp]
+	shrd	ecx,ecx,11
+	and	eax,ebx
+	xor	ecx,esi
+	add	edx,DWORD [92+esp]
+	xor	eax,edi
+	shrd	ecx,ecx,2
+	add	eax,edx
+	add	edx,DWORD [16+esp]
+	add	eax,ecx
+	vmovdqa	[80+esp],xmm6
+	cmp	DWORD [64+ebp],66051
+	jne	NEAR L$013avx_00_47
+	mov	ecx,edx
+	shrd	edx,edx,14
+	mov	esi,DWORD [20+esp]
+	xor	edx,ecx
+	mov	edi,DWORD [24+esp]
+	xor	esi,edi
+	shrd	edx,edx,5
+	and	esi,ecx
+	mov	DWORD [16+esp],ecx
+	xor	edx,ecx
+	xor	edi,esi
+	shrd	edx,edx,6
+	mov	ecx,eax
+	add	edx,edi
+	mov	edi,DWORD [4+esp]
+	mov	esi,eax
+	shrd	ecx,ecx,9
+	mov	DWORD [esp],eax
+	xor	ecx,eax
+	xor	eax,edi
+	add	edx,DWORD [28+esp]
+	shrd	ecx,ecx,11
+	and	ebx,eax
+	xor	ecx,esi
+	add	edx,DWORD [32+esp]
+	xor	ebx,edi
+	shrd	ecx,ecx,2
+	add	ebx,edx
+	add	edx,DWORD [12+esp]
+	add	ebx,ecx
+	mov	ecx,edx
+	shrd	edx,edx,14
+	mov	esi,DWORD [16+esp]
+	xor	edx,ecx
+	mov	edi,DWORD [20+esp]
+	xor	esi,edi
+	shrd	edx,edx,5
+	and	esi,ecx
+	mov	DWORD [12+esp],ecx
+	xor	edx,ecx
+	xor	edi,esi
+	shrd	edx,edx,6
+	mov	ecx,ebx
+	add	edx,edi
+	mov	edi,DWORD [esp]
+	mov	esi,ebx
+	shrd	ecx,ecx,9
+	mov	DWORD [28+esp],ebx
+	xor	ecx,ebx
+	xor	ebx,edi
+	add	edx,DWORD [24+esp]
+	shrd	ecx,ecx,11
+	and	eax,ebx
+	xor	ecx,esi
+	add	edx,DWORD [36+esp]
+	xor	eax,edi
+	shrd	ecx,ecx,2
+	add	eax,edx
+	add	edx,DWORD [8+esp]
+	add	eax,ecx
+	mov	ecx,edx
+	shrd	edx,edx,14
+	mov	esi,DWORD [12+esp]
+	xor	edx,ecx
+	mov	edi,DWORD [16+esp]
+	xor	esi,edi
+	shrd	edx,edx,5
+	and	esi,ecx
+	mov	DWORD [8+esp],ecx
+	xor	edx,ecx
+	xor	edi,esi
+	shrd	edx,edx,6
+	mov	ecx,eax
+	add	edx,edi
+	mov	edi,DWORD [28+esp]
+	mov	esi,eax
+	shrd	ecx,ecx,9
+	mov	DWORD [24+esp],eax
+	xor	ecx,eax
+	xor	eax,edi
+	add	edx,DWORD [20+esp]
+	shrd	ecx,ecx,11
+	and	ebx,eax
+	xor	ecx,esi
+	add	edx,DWORD [40+esp]
+	xor	ebx,edi
+	shrd	ecx,ecx,2
+	add	ebx,edx
+	add	edx,DWORD [4+esp]
+	add	ebx,ecx
+	mov	ecx,edx
+	shrd	edx,edx,14
+	mov	esi,DWORD [8+esp]
+	xor	edx,ecx
+	mov	edi,DWORD [12+esp]
+	xor	esi,edi
+	shrd	edx,edx,5
+	and	esi,ecx
+	mov	DWORD [4+esp],ecx
+	xor	edx,ecx
+	xor	edi,esi
+	shrd	edx,edx,6
+	mov	ecx,ebx
+	add	edx,edi
+	mov	edi,DWORD [24+esp]
+	mov	esi,ebx
+	shrd	ecx,ecx,9
+	mov	DWORD [20+esp],ebx
+	xor	ecx,ebx
+	xor	ebx,edi
+	add	edx,DWORD [16+esp]
+	shrd	ecx,ecx,11
+	and	eax,ebx
+	xor	ecx,esi
+	add	edx,DWORD [44+esp]
+	xor	eax,edi
+	shrd	ecx,ecx,2
+	add	eax,edx
+	add	edx,DWORD [esp]
+	add	eax,ecx
+	mov	ecx,edx
+	shrd	edx,edx,14
+	mov	esi,DWORD [4+esp]
+	xor	edx,ecx
+	mov	edi,DWORD [8+esp]
+	xor	esi,edi
+	shrd	edx,edx,5
+	and	esi,ecx
+	mov	DWORD [esp],ecx
+	xor	edx,ecx
+	xor	edi,esi
+	shrd	edx,edx,6
+	mov	ecx,eax
+	add	edx,edi
+	mov	edi,DWORD [20+esp]
+	mov	esi,eax
+	shrd	ecx,ecx,9
+	mov	DWORD [16+esp],eax
+	xor	ecx,eax
+	xor	eax,edi
+	add	edx,DWORD [12+esp]
+	shrd	ecx,ecx,11
+	and	ebx,eax
+	xor	ecx,esi
+	add	edx,DWORD [48+esp]
+	xor	ebx,edi
+	shrd	ecx,ecx,2
+	add	ebx,edx
+	add	edx,DWORD [28+esp]
+	add	ebx,ecx
+	mov	ecx,edx
+	shrd	edx,edx,14
+	mov	esi,DWORD [esp]
+	xor	edx,ecx
+	mov	edi,DWORD [4+esp]
+	xor	esi,edi
+	shrd	edx,edx,5
+	and	esi,ecx
+	mov	DWORD [28+esp],ecx
+	xor	edx,ecx
+	xor	edi,esi
+	shrd	edx,edx,6
+	mov	ecx,ebx
+	add	edx,edi
+	mov	edi,DWORD [16+esp]
+	mov	esi,ebx
+	shrd	ecx,ecx,9
+	mov	DWORD [12+esp],ebx
+	xor	ecx,ebx
+	xor	ebx,edi
+	add	edx,DWORD [8+esp]
+	shrd	ecx,ecx,11
+	and	eax,ebx
+	xor	ecx,esi
+	add	edx,DWORD [52+esp]
+	xor	eax,edi
+	shrd	ecx,ecx,2
+	add	eax,edx
+	add	edx,DWORD [24+esp]
+	add	eax,ecx
+	mov	ecx,edx
+	shrd	edx,edx,14
+	mov	esi,DWORD [28+esp]
+	xor	edx,ecx
+	mov	edi,DWORD [esp]
+	xor	esi,edi
+	shrd	edx,edx,5
+	and	esi,ecx
+	mov	DWORD [24+esp],ecx
+	xor	edx,ecx
+	xor	edi,esi
+	shrd	edx,edx,6
+	mov	ecx,eax
+	add	edx,edi
+	mov	edi,DWORD [12+esp]
+	mov	esi,eax
+	shrd	ecx,ecx,9
+	mov	DWORD [8+esp],eax
+	xor	ecx,eax
+	xor	eax,edi
+	add	edx,DWORD [4+esp]
+	shrd	ecx,ecx,11
+	and	ebx,eax
+	xor	ecx,esi
+	add	edx,DWORD [56+esp]
+	xor	ebx,edi
+	shrd	ecx,ecx,2
+	add	ebx,edx
+	add	edx,DWORD [20+esp]
+	add	ebx,ecx
+	mov	ecx,edx
+	shrd	edx,edx,14
+	mov	esi,DWORD [24+esp]
+	xor	edx,ecx
+	mov	edi,DWORD [28+esp]
+	xor	esi,edi
+	shrd	edx,edx,5
+	and	esi,ecx
+	mov	DWORD [20+esp],ecx
+	xor	edx,ecx
+	xor	edi,esi
+	shrd	edx,edx,6
+	mov	ecx,ebx
+	add	edx,edi
+	mov	edi,DWORD [8+esp]
+	mov	esi,ebx
+	shrd	ecx,ecx,9
+	mov	DWORD [4+esp],ebx
+	xor	ecx,ebx
+	xor	ebx,edi
+	add	edx,DWORD [esp]
+	shrd	ecx,ecx,11
+	and	eax,ebx
+	xor	ecx,esi
+	add	edx,DWORD [60+esp]
+	xor	eax,edi
+	shrd	ecx,ecx,2
+	add	eax,edx
+	add	edx,DWORD [16+esp]
+	add	eax,ecx
+	mov	ecx,edx
+	shrd	edx,edx,14
+	mov	esi,DWORD [20+esp]
+	xor	edx,ecx
+	mov	edi,DWORD [24+esp]
+	xor	esi,edi
+	shrd	edx,edx,5
+	and	esi,ecx
+	mov	DWORD [16+esp],ecx
+	xor	edx,ecx
+	xor	edi,esi
+	shrd	edx,edx,6
+	mov	ecx,eax
+	add	edx,edi
+	mov	edi,DWORD [4+esp]
+	mov	esi,eax
+	shrd	ecx,ecx,9
+	mov	DWORD [esp],eax
+	xor	ecx,eax
+	xor	eax,edi
+	add	edx,DWORD [28+esp]
+	shrd	ecx,ecx,11
+	and	ebx,eax
+	xor	ecx,esi
+	add	edx,DWORD [64+esp]
+	xor	ebx,edi
+	shrd	ecx,ecx,2
+	add	ebx,edx
+	add	edx,DWORD [12+esp]
+	add	ebx,ecx
+	mov	ecx,edx
+	shrd	edx,edx,14
+	mov	esi,DWORD [16+esp]
+	xor	edx,ecx
+	mov	edi,DWORD [20+esp]
+	xor	esi,edi
+	shrd	edx,edx,5
+	and	esi,ecx
+	mov	DWORD [12+esp],ecx
+	xor	edx,ecx
+	xor	edi,esi
+	shrd	edx,edx,6
+	mov	ecx,ebx
+	add	edx,edi
+	mov	edi,DWORD [esp]
+	mov	esi,ebx
+	shrd	ecx,ecx,9
+	mov	DWORD [28+esp],ebx
+	xor	ecx,ebx
+	xor	ebx,edi
+	add	edx,DWORD [24+esp]
+	shrd	ecx,ecx,11
+	and	eax,ebx
+	xor	ecx,esi
+	add	edx,DWORD [68+esp]
+	xor	eax,edi
+	shrd	ecx,ecx,2
+	add	eax,edx
+	add	edx,DWORD [8+esp]
+	add	eax,ecx
+	mov	ecx,edx
+	shrd	edx,edx,14
+	mov	esi,DWORD [12+esp]
+	xor	edx,ecx
+	mov	edi,DWORD [16+esp]
+	xor	esi,edi
+	shrd	edx,edx,5
+	and	esi,ecx
+	mov	DWORD [8+esp],ecx
+	xor	edx,ecx
+	xor	edi,esi
+	shrd	edx,edx,6
+	mov	ecx,eax
+	add	edx,edi
+	mov	edi,DWORD [28+esp]
+	mov	esi,eax
+	shrd	ecx,ecx,9
+	mov	DWORD [24+esp],eax
+	xor	ecx,eax
+	xor	eax,edi
+	add	edx,DWORD [20+esp]
+	shrd	ecx,ecx,11
+	and	ebx,eax
+	xor	ecx,esi
+	add	edx,DWORD [72+esp]
+	xor	ebx,edi
+	shrd	ecx,ecx,2
+	add	ebx,edx
+	add	edx,DWORD [4+esp]
+	add	ebx,ecx
+	mov	ecx,edx
+	shrd	edx,edx,14
+	mov	esi,DWORD [8+esp]
+	xor	edx,ecx
+	mov	edi,DWORD [12+esp]
+	xor	esi,edi
+	shrd	edx,edx,5
+	and	esi,ecx
+	mov	DWORD [4+esp],ecx
+	xor	edx,ecx
+	xor	edi,esi
+	shrd	edx,edx,6
+	mov	ecx,ebx
+	add	edx,edi
+	mov	edi,DWORD [24+esp]
+	mov	esi,ebx
+	shrd	ecx,ecx,9
+	mov	DWORD [20+esp],ebx
+	xor	ecx,ebx
+	xor	ebx,edi
+	add	edx,DWORD [16+esp]
+	shrd	ecx,ecx,11
+	and	eax,ebx
+	xor	ecx,esi
+	add	edx,DWORD [76+esp]
+	xor	eax,edi
+	shrd	ecx,ecx,2
+	add	eax,edx
+	add	edx,DWORD [esp]
+	add	eax,ecx
+	mov	ecx,edx
+	shrd	edx,edx,14
+	mov	esi,DWORD [4+esp]
+	xor	edx,ecx
+	mov	edi,DWORD [8+esp]
+	xor	esi,edi
+	shrd	edx,edx,5
+	and	esi,ecx
+	mov	DWORD [esp],ecx
+	xor	edx,ecx
+	xor	edi,esi
+	shrd	edx,edx,6
+	mov	ecx,eax
+	add	edx,edi
+	mov	edi,DWORD [20+esp]
+	mov	esi,eax
+	shrd	ecx,ecx,9
+	mov	DWORD [16+esp],eax
+	xor	ecx,eax
+	xor	eax,edi
+	add	edx,DWORD [12+esp]
+	shrd	ecx,ecx,11
+	and	ebx,eax
+	xor	ecx,esi
+	add	edx,DWORD [80+esp]
+	xor	ebx,edi
+	shrd	ecx,ecx,2
+	add	ebx,edx
+	add	edx,DWORD [28+esp]
+	add	ebx,ecx
+	mov	ecx,edx
+	shrd	edx,edx,14
+	mov	esi,DWORD [esp]
+	xor	edx,ecx
+	mov	edi,DWORD [4+esp]
+	xor	esi,edi
+	shrd	edx,edx,5
+	and	esi,ecx
+	mov	DWORD [28+esp],ecx
+	xor	edx,ecx
+	xor	edi,esi
+	shrd	edx,edx,6
+	mov	ecx,ebx
+	add	edx,edi
+	mov	edi,DWORD [16+esp]
+	mov	esi,ebx
+	shrd	ecx,ecx,9
+	mov	DWORD [12+esp],ebx
+	xor	ecx,ebx
+	xor	ebx,edi
+	add	edx,DWORD [8+esp]
+	shrd	ecx,ecx,11
+	and	eax,ebx
+	xor	ecx,esi
+	add	edx,DWORD [84+esp]
+	xor	eax,edi
+	shrd	ecx,ecx,2
+	add	eax,edx
+	add	edx,DWORD [24+esp]
+	add	eax,ecx
+	mov	ecx,edx
+	shrd	edx,edx,14
+	mov	esi,DWORD [28+esp]
+	xor	edx,ecx
+	mov	edi,DWORD [esp]
+	xor	esi,edi
+	shrd	edx,edx,5
+	and	esi,ecx
+	mov	DWORD [24+esp],ecx
+	xor	edx,ecx
+	xor	edi,esi
+	shrd	edx,edx,6
+	mov	ecx,eax
+	add	edx,edi
+	mov	edi,DWORD [12+esp]
+	mov	esi,eax
+	shrd	ecx,ecx,9
+	mov	DWORD [8+esp],eax
+	xor	ecx,eax
+	xor	eax,edi
+	add	edx,DWORD [4+esp]
+	shrd	ecx,ecx,11
+	and	ebx,eax
+	xor	ecx,esi
+	add	edx,DWORD [88+esp]
+	xor	ebx,edi
+	shrd	ecx,ecx,2
+	add	ebx,edx
+	add	edx,DWORD [20+esp]
+	add	ebx,ecx
+	mov	ecx,edx
+	shrd	edx,edx,14
+	mov	esi,DWORD [24+esp]
+	xor	edx,ecx
+	mov	edi,DWORD [28+esp]
+	xor	esi,edi
+	shrd	edx,edx,5
+	and	esi,ecx
+	mov	DWORD [20+esp],ecx
+	xor	edx,ecx
+	xor	edi,esi
+	shrd	edx,edx,6
+	mov	ecx,ebx
+	add	edx,edi
+	mov	edi,DWORD [8+esp]
+	mov	esi,ebx
+	shrd	ecx,ecx,9
+	mov	DWORD [4+esp],ebx
+	xor	ecx,ebx
+	xor	ebx,edi
+	add	edx,DWORD [esp]
+	shrd	ecx,ecx,11
+	and	eax,ebx
+	xor	ecx,esi
+	add	edx,DWORD [92+esp]
+	xor	eax,edi
+	shrd	ecx,ecx,2
+	add	eax,edx
+	add	edx,DWORD [16+esp]
+	add	eax,ecx
+	mov	esi,DWORD [96+esp]
+	xor	ebx,edi
+	mov	ecx,DWORD [12+esp]
+	add	eax,DWORD [esi]
+	add	ebx,DWORD [4+esi]
+	add	edi,DWORD [8+esi]
+	add	ecx,DWORD [12+esi]
+	mov	DWORD [esi],eax
+	mov	DWORD [4+esi],ebx
+	mov	DWORD [8+esi],edi
+	mov	DWORD [12+esi],ecx
+	mov	DWORD [4+esp],ebx
+	xor	ebx,edi
+	mov	DWORD [8+esp],edi
+	mov	DWORD [12+esp],ecx
+	mov	edi,DWORD [20+esp]
+	mov	ecx,DWORD [24+esp]
+	add	edx,DWORD [16+esi]
+	add	edi,DWORD [20+esi]
+	add	ecx,DWORD [24+esi]
+	mov	DWORD [16+esi],edx
+	mov	DWORD [20+esi],edi
+	mov	DWORD [20+esp],edi
+	mov	edi,DWORD [28+esp]
+	mov	DWORD [24+esi],ecx
+	add	edi,DWORD [28+esi]
+	mov	DWORD [24+esp],ecx
+	mov	DWORD [28+esi],edi
+	mov	DWORD [28+esp],edi
+	mov	edi,DWORD [100+esp]
+	vmovdqa	xmm7,[64+ebp]
+	sub	ebp,192
+	cmp	edi,DWORD [104+esp]
+	jb	NEAR L$012grand_avx
+	mov	esp,DWORD [108+esp]
+	vzeroall
+	pop	edi
+	pop	esi
+	pop	ebx
+	pop	ebp
+	ret
 segment	.bss
 common	_OPENSSL_ia32cap_P 16

third_party/boringssl/win-x86_64/crypto/bn/rsaz-x86_64.asm

diff --git a/third_party/boringssl/win-x86_64/crypto/bn/rsaz-x86_64.asm b/third_party/boringssl/win-x86_64/crypto/bn/rsaz-x86_64.asm
index 04d5e39..72ec505 100644
--- a/third_party/boringssl/win-x86_64/crypto/bn/rsaz-x86_64.asm
+++ b/third_party/boringssl/win-x86_64/crypto/bn/rsaz-x86_64.asm
@@ -504,48 +504,104 @@
 	push	r14
 	push	r15
 
-	mov	r9d,r9d
-	sub	rsp,128+24
+	sub	rsp,328
+	movaps	XMMWORD[160+rsp],xmm6
+	movaps	XMMWORD[176+rsp],xmm7
+	movaps	XMMWORD[192+rsp],xmm8
+	movaps	XMMWORD[208+rsp],xmm9
+	movaps	XMMWORD[224+rsp],xmm10
+	movaps	XMMWORD[240+rsp],xmm11
+	movaps	XMMWORD[256+rsp],xmm12
+	movaps	XMMWORD[272+rsp],xmm13
+	movaps	XMMWORD[288+rsp],xmm14
+	movaps	XMMWORD[304+rsp],xmm15
 $L$mul_gather4_body:
-	mov	eax,DWORD[64+r9*4+rdx]
-DB	102,72,15,110,199
-	mov	ebx,DWORD[r9*4+rdx]
-DB	102,72,15,110,201
-	mov	QWORD[128+rsp],r8
+	movd	xmm8,r9d
+	movdqa	xmm1,XMMWORD[(($L$inc+16))]
+	movdqa	xmm0,XMMWORD[$L$inc]
 
-	shl	rax,32
-	or	rbx,rax
+	pshufd	xmm8,xmm8,0
+	movdqa	xmm7,xmm1
+	movdqa	xmm2,xmm1
+	paddd	xmm1,xmm0
+	pcmpeqd	xmm0,xmm8
+	movdqa	xmm3,xmm7
+	paddd	xmm2,xmm1
+	pcmpeqd	xmm1,xmm8
+	movdqa	xmm4,xmm7
+	paddd	xmm3,xmm2
+	pcmpeqd	xmm2,xmm8
+	movdqa	xmm5,xmm7
+	paddd	xmm4,xmm3
+	pcmpeqd	xmm3,xmm8
+	movdqa	xmm6,xmm7
+	paddd	xmm5,xmm4
+	pcmpeqd	xmm4,xmm8
+	paddd	xmm6,xmm5
+	pcmpeqd	xmm5,xmm8
+	paddd	xmm7,xmm6
+	pcmpeqd	xmm6,xmm8
+	pcmpeqd	xmm7,xmm8
+
+	movdqa	xmm8,XMMWORD[rdx]
+	movdqa	xmm9,XMMWORD[16+rdx]
+	movdqa	xmm10,XMMWORD[32+rdx]
+	movdqa	xmm11,XMMWORD[48+rdx]
+	pand	xmm8,xmm0
+	movdqa	xmm12,XMMWORD[64+rdx]
+	pand	xmm9,xmm1
+	movdqa	xmm13,XMMWORD[80+rdx]
+	pand	xmm10,xmm2
+	movdqa	xmm14,XMMWORD[96+rdx]
+	pand	xmm11,xmm3
+	movdqa	xmm15,XMMWORD[112+rdx]
+	lea	rbp,[128+rdx]
+	pand	xmm12,xmm4
+	pand	xmm13,xmm5
+	pand	xmm14,xmm6
+	pand	xmm15,xmm7
+	por	xmm8,xmm10
+	por	xmm9,xmm11
+	por	xmm8,xmm12
+	por	xmm9,xmm13
+	por	xmm8,xmm14
+	por	xmm9,xmm15
+
+	por	xmm8,xmm9
+	pshufd	xmm9,xmm8,0x4e
+	por	xmm8,xmm9
+DB	102,76,15,126,195
+
+	mov	QWORD[128+rsp],r8
+	mov	QWORD[((128+8))+rsp],rdi
+	mov	QWORD[((128+16))+rsp],rcx
+
 	mov	rax,QWORD[rsi]
 	mov	rcx,QWORD[8+rsi]
-	lea	rbp,[128+r9*4+rdx]
 	mul	rbx
 	mov	QWORD[rsp],rax
 	mov	rax,rcx
 	mov	r8,rdx
 
 	mul	rbx
-	movd	xmm4,DWORD[rbp]
 	add	r8,rax
 	mov	rax,QWORD[16+rsi]
 	mov	r9,rdx
 	adc	r9,0
 
 	mul	rbx
-	movd	xmm5,DWORD[64+rbp]
 	add	r9,rax
 	mov	rax,QWORD[24+rsi]
 	mov	r10,rdx
 	adc	r10,0
 
 	mul	rbx
-	pslldq	xmm5,4
 	add	r10,rax
 	mov	rax,QWORD[32+rsi]
 	mov	r11,rdx
 	adc	r11,0
 
 	mul	rbx
-	por	xmm4,xmm5
 	add	r11,rax
 	mov	rax,QWORD[40+rsi]
 	mov	r12,rdx
@@ -558,14 +614,12 @@
 	adc	r13,0
 
 	mul	rbx
-	lea	rbp,[128+rbp]
 	add	r13,rax
 	mov	rax,QWORD[56+rsi]
 	mov	r14,rdx
 	adc	r14,0
 
 	mul	rbx
-DB	102,72,15,126,227
 	add	r14,rax
 	mov	rax,QWORD[rsi]
 	mov	r15,rdx
@@ -577,6 +631,35 @@
 
 ALIGN	32
 $L$oop_mul_gather:
+	movdqa	xmm8,XMMWORD[rbp]
+	movdqa	xmm9,XMMWORD[16+rbp]
+	movdqa	xmm10,XMMWORD[32+rbp]
+	movdqa	xmm11,XMMWORD[48+rbp]
+	pand	xmm8,xmm0
+	movdqa	xmm12,XMMWORD[64+rbp]
+	pand	xmm9,xmm1
+	movdqa	xmm13,XMMWORD[80+rbp]
+	pand	xmm10,xmm2
+	movdqa	xmm14,XMMWORD[96+rbp]
+	pand	xmm11,xmm3
+	movdqa	xmm15,XMMWORD[112+rbp]
+	lea	rbp,[128+rbp]
+	pand	xmm12,xmm4
+	pand	xmm13,xmm5
+	pand	xmm14,xmm6
+	pand	xmm15,xmm7
+	por	xmm8,xmm10
+	por	xmm9,xmm11
+	por	xmm8,xmm12
+	por	xmm9,xmm13
+	por	xmm8,xmm14
+	por	xmm9,xmm15
+
+	por	xmm8,xmm9
+	pshufd	xmm9,xmm8,0x4e
+	por	xmm8,xmm9
+DB	102,76,15,126,195
+
 	mul	rbx
 	add	r8,rax
 	mov	rax,QWORD[8+rsi]
@@ -585,7 +668,6 @@
 	adc	r8,0
 
 	mul	rbx
-	movd	xmm4,DWORD[rbp]
 	add	r9,rax
 	mov	rax,QWORD[16+rsi]
 	adc	rdx,0
@@ -594,7 +676,6 @@
 	adc	r9,0
 
 	mul	rbx
-	movd	xmm5,DWORD[64+rbp]
 	add	r10,rax
 	mov	rax,QWORD[24+rsi]
 	adc	rdx,0
@@ -603,7 +684,6 @@
 	adc	r10,0
 
 	mul	rbx
-	pslldq	xmm5,4
 	add	r11,rax
 	mov	rax,QWORD[32+rsi]
 	adc	rdx,0
@@ -612,7 +692,6 @@
 	adc	r11,0
 
 	mul	rbx
-	por	xmm4,xmm5
 	add	r12,rax
 	mov	rax,QWORD[40+rsi]
 	adc	rdx,0
@@ -637,7 +716,6 @@
 	adc	r14,0
 
 	mul	rbx
-DB	102,72,15,126,227
 	add	r15,rax
 	mov	rax,QWORD[rsi]
 	adc	rdx,0
@@ -645,7 +723,6 @@
 	mov	r15,rdx
 	adc	r15,0
 
-	lea	rbp,[128+rbp]
 	lea	rdi,[8+rdi]
 
 	dec	ecx
@@ -660,8 +737,8 @@
 	mov	QWORD[48+rdi],r14
 	mov	QWORD[56+rdi],r15
 
-DB	102,72,15,126,199
-DB	102,72,15,126,205
+	mov	rdi,QWORD[((128+8))+rsp]
+	mov	rbp,QWORD[((128+16))+rsp]
 
 	mov	r8,QWORD[rsp]
 	mov	r9,QWORD[8+rsp]
@@ -686,6 +763,17 @@
 	call	__rsaz_512_subtract
 
 	lea	rax,[((128+24+48))+rsp]
+	movaps	xmm6,XMMWORD[((160-200))+rax]
+	movaps	xmm7,XMMWORD[((176-200))+rax]
+	movaps	xmm8,XMMWORD[((192-200))+rax]
+	movaps	xmm9,XMMWORD[((208-200))+rax]
+	movaps	xmm10,XMMWORD[((224-200))+rax]
+	movaps	xmm11,XMMWORD[((240-200))+rax]
+	movaps	xmm12,XMMWORD[((256-200))+rax]
+	movaps	xmm13,XMMWORD[((272-200))+rax]
+	movaps	xmm14,XMMWORD[((288-200))+rax]
+	movaps	xmm15,XMMWORD[((304-200))+rax]
+	lea	rax,[176+rax]
 	mov	r15,QWORD[((-48))+rax]
 	mov	r14,QWORD[((-40))+rax]
 	mov	r13,QWORD[((-32))+rax]
@@ -724,7 +812,7 @@
 	mov	r9d,r9d
 	sub	rsp,128+24
 $L$mul_scatter4_body:
-	lea	r8,[r9*4+r8]
+	lea	r8,[r9*8+r8]
 DB	102,72,15,110,199
 DB	102,72,15,110,202
 DB	102,73,15,110,208
@@ -760,30 +848,14 @@
 
 	call	__rsaz_512_subtract
 
-	mov	DWORD[rsi],r8d
-	shr	r8,32
-	mov	DWORD[128+rsi],r9d
-	shr	r9,32
-	mov	DWORD[256+rsi],r10d
-	shr	r10,32
-	mov	DWORD[384+rsi],r11d
-	shr	r11,32
-	mov	DWORD[512+rsi],r12d
-	shr	r12,32
-	mov	DWORD[640+rsi],r13d
-	shr	r13,32
-	mov	DWORD[768+rsi],r14d
-	shr	r14,32
-	mov	DWORD[896+rsi],r15d
-	shr	r15,32
-	mov	DWORD[64+rsi],r8d
-	mov	DWORD[192+rsi],r9d
-	mov	DWORD[320+rsi],r10d
-	mov	DWORD[448+rsi],r11d
-	mov	DWORD[576+rsi],r12d
-	mov	DWORD[704+rsi],r13d
-	mov	DWORD[832+rsi],r14d
-	mov	DWORD[960+rsi],r15d
+	mov	QWORD[rsi],r8
+	mov	QWORD[128+rsi],r9
+	mov	QWORD[256+rsi],r10
+	mov	QWORD[384+rsi],r11
+	mov	QWORD[512+rsi],r12
+	mov	QWORD[640+rsi],r13
+	mov	QWORD[768+rsi],r14
+	mov	QWORD[896+rsi],r15
 
 	lea	rax,[((128+24+48))+rsp]
 	mov	r15,QWORD[((-48))+rax]
@@ -1150,16 +1222,14 @@
 
 ALIGN	16
 rsaz_512_scatter4:
-	lea	rcx,[r8*4+rcx]
+	lea	rcx,[r8*8+rcx]
 	mov	r9d,8
 	jmp	NEAR $L$oop_scatter
 ALIGN	16
 $L$oop_scatter:
 	mov	rax,QWORD[rdx]
 	lea	rdx,[8+rdx]
-	mov	DWORD[rcx],eax
-	shr	rax,32
-	mov	DWORD[64+rcx],eax
+	mov	QWORD[rcx],rax
 	lea	rcx,[128+rcx]
 	dec	r9d
 	jnz	NEAR $L$oop_scatter
@@ -1170,22 +1240,98 @@
 
 ALIGN	16
 rsaz_512_gather4:
-	lea	rdx,[r8*4+rdx]
+$L$SEH_begin_rsaz_512_gather4:
+DB	0x48,0x81,0xec,0xa8,0x00,0x00,0x00
+DB	0x0f,0x29,0x34,0x24
+DB	0x0f,0x29,0x7c,0x24,0x10
+DB	0x44,0x0f,0x29,0x44,0x24,0x20
+DB	0x44,0x0f,0x29,0x4c,0x24,0x30
+DB	0x44,0x0f,0x29,0x54,0x24,0x40
+DB	0x44,0x0f,0x29,0x5c,0x24,0x50
+DB	0x44,0x0f,0x29,0x64,0x24,0x60
+DB	0x44,0x0f,0x29,0x6c,0x24,0x70
+DB	0x44,0x0f,0x29,0xb4,0x24,0x80,0,0,0
+DB	0x44,0x0f,0x29,0xbc,0x24,0x90,0,0,0
+	movd	xmm8,r8d
+	movdqa	xmm1,XMMWORD[(($L$inc+16))]
+	movdqa	xmm0,XMMWORD[$L$inc]
+
+	pshufd	xmm8,xmm8,0
+	movdqa	xmm7,xmm1
+	movdqa	xmm2,xmm1
+	paddd	xmm1,xmm0
+	pcmpeqd	xmm0,xmm8
+	movdqa	xmm3,xmm7
+	paddd	xmm2,xmm1
+	pcmpeqd	xmm1,xmm8
+	movdqa	xmm4,xmm7
+	paddd	xmm3,xmm2
+	pcmpeqd	xmm2,xmm8
+	movdqa	xmm5,xmm7
+	paddd	xmm4,xmm3
+	pcmpeqd	xmm3,xmm8
+	movdqa	xmm6,xmm7
+	paddd	xmm5,xmm4
+	pcmpeqd	xmm4,xmm8
+	paddd	xmm6,xmm5
+	pcmpeqd	xmm5,xmm8
+	paddd	xmm7,xmm6
+	pcmpeqd	xmm6,xmm8
+	pcmpeqd	xmm7,xmm8
 	mov	r9d,8
 	jmp	NEAR $L$oop_gather
 ALIGN	16
 $L$oop_gather:
-	mov	eax,DWORD[rdx]
-	mov	r8d,DWORD[64+rdx]
+	movdqa	xmm8,XMMWORD[rdx]
+	movdqa	xmm9,XMMWORD[16+rdx]
+	movdqa	xmm10,XMMWORD[32+rdx]
+	movdqa	xmm11,XMMWORD[48+rdx]
+	pand	xmm8,xmm0
+	movdqa	xmm12,XMMWORD[64+rdx]
+	pand	xmm9,xmm1
+	movdqa	xmm13,XMMWORD[80+rdx]
+	pand	xmm10,xmm2
+	movdqa	xmm14,XMMWORD[96+rdx]
+	pand	xmm11,xmm3
+	movdqa	xmm15,XMMWORD[112+rdx]
 	lea	rdx,[128+rdx]
-	shl	r8,32
-	or	rax,r8
-	mov	QWORD[rcx],rax
+	pand	xmm12,xmm4
+	pand	xmm13,xmm5
+	pand	xmm14,xmm6
+	pand	xmm15,xmm7
+	por	xmm8,xmm10
+	por	xmm9,xmm11
+	por	xmm8,xmm12
+	por	xmm9,xmm13
+	por	xmm8,xmm14
+	por	xmm9,xmm15
+
+	por	xmm8,xmm9
+	pshufd	xmm9,xmm8,0x4e
+	por	xmm8,xmm9
+	movq	QWORD[rcx],xmm8
 	lea	rcx,[8+rcx]
 	dec	r9d
 	jnz	NEAR $L$oop_gather
+	movaps	xmm6,XMMWORD[rsp]
+	movaps	xmm7,XMMWORD[16+rsp]
+	movaps	xmm8,XMMWORD[32+rsp]
+	movaps	xmm9,XMMWORD[48+rsp]
+	movaps	xmm10,XMMWORD[64+rsp]
+	movaps	xmm11,XMMWORD[80+rsp]
+	movaps	xmm12,XMMWORD[96+rsp]
+	movaps	xmm13,XMMWORD[112+rsp]
+	movaps	xmm14,XMMWORD[128+rsp]
+	movaps	xmm15,XMMWORD[144+rsp]
+	add	rsp,0xa8
 	DB	0F3h,0C3h		;repret
+$L$SEH_end_rsaz_512_gather4:
 
+
+ALIGN	64
+$L$inc:
+	DD	0,0,1,1
+	DD	2,2,2,2
 EXTERN	__imp_RtlVirtualUnwind
 
 ALIGN	16
@@ -1221,6 +1367,18 @@
 
 	lea	rax,[((128+24+48))+rax]
 
+	lea	rbx,[$L$mul_gather4_epilogue]
+	cmp	rbx,r10
+	jne	NEAR $L$se_not_in_mul_gather4
+
+	lea	rax,[176+rax]
+
+	lea	rsi,[((-48-168))+rax]
+	lea	rdi,[512+r8]
+	mov	ecx,20
+	DD	0xa548f3fc
+
+$L$se_not_in_mul_gather4:
 	mov	rbx,QWORD[((-8))+rax]
 	mov	rbp,QWORD[((-16))+rax]
 	mov	r12,QWORD[((-24))+rax]
@@ -1296,6 +1454,10 @@
 	DD	$L$SEH_end_rsaz_512_mul_by_one wrt ..imagebase
 	DD	$L$SEH_info_rsaz_512_mul_by_one wrt ..imagebase
 
+	DD	$L$SEH_begin_rsaz_512_gather4 wrt ..imagebase
+	DD	$L$SEH_end_rsaz_512_gather4 wrt ..imagebase
+	DD	$L$SEH_info_rsaz_512_gather4 wrt ..imagebase
+
 section	.xdata rdata align=8
 ALIGN	8
 $L$SEH_info_rsaz_512_sqr:
@@ -1318,3 +1480,16 @@
 DB	9,0,0,0
 	DD	se_handler wrt ..imagebase
 	DD	$L$mul_by_one_body wrt ..imagebase,$L$mul_by_one_epilogue wrt ..imagebase
+$L$SEH_info_rsaz_512_gather4:
+DB	0x01,0x46,0x16,0x00
+DB	0x46,0xf8,0x09,0x00
+DB	0x3d,0xe8,0x08,0x00
+DB	0x34,0xd8,0x07,0x00
+DB	0x2e,0xc8,0x06,0x00
+DB	0x28,0xb8,0x05,0x00
+DB	0x22,0xa8,0x04,0x00
+DB	0x1c,0x98,0x03,0x00
+DB	0x16,0x88,0x02,0x00
+DB	0x10,0x78,0x01,0x00
+DB	0x0b,0x68,0x00,0x00
+DB	0x07,0x01,0x15,0x00

third_party/boringssl/win-x86_64/crypto/bn/x86_64-mont.asm

diff --git a/third_party/boringssl/win-x86_64/crypto/bn/x86_64-mont.asm b/third_party/boringssl/win-x86_64/crypto/bn/x86_64-mont.asm
index db0d1b9..4d8e1cb 100644
--- a/third_party/boringssl/win-x86_64/crypto/bn/x86_64-mont.asm
+++ b/third_party/boringssl/win-x86_64/crypto/bn/x86_64-mont.asm
@@ -677,20 +677,20 @@
 
 
 
-	lea	r11,[((-64))+r9*4+rsp]
+	lea	r11,[((-64))+r9*2+rsp]
 	mov	r8,QWORD[r8]
 	sub	r11,rsi
 	and	r11,4095
 	cmp	r10,r11
 	jb	NEAR $L$sqr8x_sp_alt
 	sub	rsp,r11
-	lea	rsp,[((-64))+r9*4+rsp]
+	lea	rsp,[((-64))+r9*2+rsp]
 	jmp	NEAR $L$sqr8x_sp_done
 
 ALIGN	32
 $L$sqr8x_sp_alt:
-	lea	r10,[((4096-64))+r9*4]
-	lea	rsp,[((-64))+r9*4+rsp]
+	lea	r10,[((4096-64))+r9*2]
+	lea	rsp,[((-64))+r9*2+rsp]
 	sub	r11,r10
 	mov	r10,0
 	cmovc	r11,r10
@@ -700,58 +700,80 @@
 	mov	r10,r9
 	neg	r9
 
-	lea	r11,[64+r9*2+rsp]
 	mov	QWORD[32+rsp],r8
 	mov	QWORD[40+rsp],rax
 $L$sqr8x_body:
 
-	mov	rbp,r9
-DB	102,73,15,110,211
-	shr	rbp,3+2
-	mov	eax,DWORD[((OPENSSL_ia32cap_P+8))]
-	jmp	NEAR $L$sqr8x_copy_n
-
-ALIGN	32
-$L$sqr8x_copy_n:
-	movq	xmm0,QWORD[rcx]
-	movq	xmm1,QWORD[8+rcx]
-	movq	xmm3,QWORD[16+rcx]
-	movq	xmm4,QWORD[24+rcx]
-	lea	rcx,[32+rcx]
-	movdqa	XMMWORD[r11],xmm0
-	movdqa	XMMWORD[16+r11],xmm1
-	movdqa	XMMWORD[32+r11],xmm3
-	movdqa	XMMWORD[48+r11],xmm4
-	lea	r11,[64+r11]
-	dec	rbp
-	jnz	NEAR $L$sqr8x_copy_n
-
+DB	102,72,15,110,209
 	pxor	xmm0,xmm0
 DB	102,72,15,110,207
 DB	102,73,15,110,218
 	call	bn_sqr8x_internal
 
-	pxor	xmm0,xmm0
-	lea	rax,[48+rsp]
-	lea	rdx,[64+r9*2+rsp]
-	shr	r9,3+2
-	mov	rsi,QWORD[40+rsp]
-	jmp	NEAR $L$sqr8x_zero
+
+
+
+	lea	rbx,[r9*1+rdi]
+	mov	rcx,r9
+	mov	rdx,r9
+DB	102,72,15,126,207
+	sar	rcx,3+2
+	jmp	NEAR $L$sqr8x_sub
 
 ALIGN	32
-$L$sqr8x_zero:
-	movdqa	XMMWORD[rax],xmm0
-	movdqa	XMMWORD[16+rax],xmm0
-	movdqa	XMMWORD[32+rax],xmm0
-	movdqa	XMMWORD[48+rax],xmm0
-	lea	rax,[64+rax]
-	movdqa	XMMWORD[rdx],xmm0
-	movdqa	XMMWORD[16+rdx],xmm0
-	movdqa	XMMWORD[32+rdx],xmm0
-	movdqa	XMMWORD[48+rdx],xmm0
-	lea	rdx,[64+rdx]
-	dec	r9
-	jnz	NEAR $L$sqr8x_zero
+$L$sqr8x_sub:
+	mov	r12,QWORD[rbx]
+	mov	r13,QWORD[8+rbx]
+	mov	r14,QWORD[16+rbx]
+	mov	r15,QWORD[24+rbx]
+	lea	rbx,[32+rbx]
+	sbb	r12,QWORD[rbp]
+	sbb	r13,QWORD[8+rbp]
+	sbb	r14,QWORD[16+rbp]
+	sbb	r15,QWORD[24+rbp]
+	lea	rbp,[32+rbp]
+	mov	QWORD[rdi],r12
+	mov	QWORD[8+rdi],r13
+	mov	QWORD[16+rdi],r14
+	mov	QWORD[24+rdi],r15
+	lea	rdi,[32+rdi]
+	inc	rcx
+	jnz	NEAR $L$sqr8x_sub
+
+	sbb	rax,0
+	lea	rbx,[r9*1+rbx]
+	lea	rdi,[r9*1+rdi]
+
+DB	102,72,15,110,200
+	pxor	xmm0,xmm0
+	pshufd	xmm1,xmm1,0
+	mov	rsi,QWORD[40+rsp]
+	jmp	NEAR $L$sqr8x_cond_copy
+
+ALIGN	32
+$L$sqr8x_cond_copy:
+	movdqa	xmm2,XMMWORD[rbx]
+	movdqa	xmm3,XMMWORD[16+rbx]
+	lea	rbx,[32+rbx]
+	movdqu	xmm4,XMMWORD[rdi]
+	movdqu	xmm5,XMMWORD[16+rdi]
+	lea	rdi,[32+rdi]
+	movdqa	XMMWORD[(-32)+rbx],xmm0
+	movdqa	XMMWORD[(-16)+rbx],xmm0
+	movdqa	XMMWORD[(-32)+rdx*1+rbx],xmm0
+	movdqa	XMMWORD[(-16)+rdx*1+rbx],xmm0
+	pcmpeqd	xmm0,xmm1
+	pand	xmm2,xmm1
+	pand	xmm3,xmm1
+	pand	xmm4,xmm0
+	pand	xmm5,xmm0
+	pxor	xmm0,xmm0
+	por	xmm4,xmm2
+	por	xmm5,xmm3
+	movdqu	XMMWORD[(-32)+rdi],xmm4
+	movdqu	XMMWORD[(-16)+rdi],xmm5
+	add	r9,32
+	jnz	NEAR $L$sqr8x_cond_copy
 
 	mov	rax,1
 	mov	r15,QWORD[((-48))+rsi]

third_party/boringssl/win-x86_64/crypto/bn/x86_64-mont5.asm

diff --git a/third_party/boringssl/win-x86_64/crypto/bn/x86_64-mont5.asm b/third_party/boringssl/win-x86_64/crypto/bn/x86_64-mont5.asm
index 284318a..cd9a6e5 100644
--- a/third_party/boringssl/win-x86_64/crypto/bn/x86_64-mont5.asm
+++ b/third_party/boringssl/win-x86_64/crypto/bn/x86_64-mont5.asm
@@ -31,49 +31,151 @@
 $L$mul_enter:
 	mov	r9d,r9d
 	mov	rax,rsp
-	mov	r10d,DWORD[56+rsp]
+	movd	xmm5,DWORD[56+rsp]
+	lea	r10,[$L$inc]
 	push	rbx
 	push	rbp
 	push	r12
 	push	r13
 	push	r14
 	push	r15
-	lea	rsp,[((-40))+rsp]
-	movaps	XMMWORD[rsp],xmm6
-	movaps	XMMWORD[16+rsp],xmm7
+
 	lea	r11,[2+r9]
 	neg	r11
-	lea	rsp,[r11*8+rsp]
+	lea	rsp,[((-264))+r11*8+rsp]
 	and	rsp,-1024
 
 	mov	QWORD[8+r9*8+rsp],rax
 $L$mul_body:
-	mov	r12,rdx
-	mov	r11,r10
-	shr	r10,3
-	and	r11,7
-	not	r10
-	lea	rax,[$L$magic_masks]
-	and	r10,3
-	lea	r12,[96+r11*8+r12]
-	movq	xmm4,QWORD[r10*8+rax]
-	movq	xmm5,QWORD[8+r10*8+rax]
-	movq	xmm6,QWORD[16+r10*8+rax]
-	movq	xmm7,QWORD[24+r10*8+rax]
+	lea	r12,[128+rdx]
+	movdqa	xmm0,XMMWORD[r10]
+	movdqa	xmm1,XMMWORD[16+r10]
+	lea	r10,[((24-112))+r9*8+rsp]
+	and	r10,-16
 
-	movq	xmm0,QWORD[(((-96)))+r12]
-	movq	xmm1,QWORD[((-32))+r12]
-	pand	xmm0,xmm4
-	movq	xmm2,QWORD[32+r12]
-	pand	xmm1,xmm5
-	movq	xmm3,QWORD[96+r12]
-	pand	xmm2,xmm6
-	por	xmm0,xmm1
-	pand	xmm3,xmm7
+	pshufd	xmm5,xmm5,0
+	movdqa	xmm4,xmm1
+	movdqa	xmm2,xmm1
+	paddd	xmm1,xmm0
+	pcmpeqd	xmm0,xmm5
+DB	0x67
+	movdqa	xmm3,xmm4
+	paddd	xmm2,xmm1
+	pcmpeqd	xmm1,xmm5
+	movdqa	XMMWORD[112+r10],xmm0
+	movdqa	xmm0,xmm4
+
+	paddd	xmm3,xmm2
+	pcmpeqd	xmm2,xmm5
+	movdqa	XMMWORD[128+r10],xmm1
+	movdqa	xmm1,xmm4
+
+	paddd	xmm0,xmm3
+	pcmpeqd	xmm3,xmm5
+	movdqa	XMMWORD[144+r10],xmm2
+	movdqa	xmm2,xmm4
+
+	paddd	xmm1,xmm0
+	pcmpeqd	xmm0,xmm5
+	movdqa	XMMWORD[160+r10],xmm3
+	movdqa	xmm3,xmm4
+	paddd	xmm2,xmm1
+	pcmpeqd	xmm1,xmm5
+	movdqa	XMMWORD[176+r10],xmm0
+	movdqa	xmm0,xmm4
+
+	paddd	xmm3,xmm2
+	pcmpeqd	xmm2,xmm5
+	movdqa	XMMWORD[192+r10],xmm1
+	movdqa	xmm1,xmm4
+
+	paddd	xmm0,xmm3
+	pcmpeqd	xmm3,xmm5
+	movdqa	XMMWORD[208+r10],xmm2
+	movdqa	xmm2,xmm4
+
+	paddd	xmm1,xmm0
+	pcmpeqd	xmm0,xmm5
+	movdqa	XMMWORD[224+r10],xmm3
+	movdqa	xmm3,xmm4
+	paddd	xmm2,xmm1
+	pcmpeqd	xmm1,xmm5
+	movdqa	XMMWORD[240+r10],xmm0
+	movdqa	xmm0,xmm4
+
+	paddd	xmm3,xmm2
+	pcmpeqd	xmm2,xmm5
+	movdqa	XMMWORD[256+r10],xmm1
+	movdqa	xmm1,xmm4
+
+	paddd	xmm0,xmm3
+	pcmpeqd	xmm3,xmm5
+	movdqa	XMMWORD[272+r10],xmm2
+	movdqa	xmm2,xmm4
+
+	paddd	xmm1,xmm0
+	pcmpeqd	xmm0,xmm5
+	movdqa	XMMWORD[288+r10],xmm3
+	movdqa	xmm3,xmm4
+	paddd	xmm2,xmm1
+	pcmpeqd	xmm1,xmm5
+	movdqa	XMMWORD[304+r10],xmm0
+
+	paddd	xmm3,xmm2
+DB	0x67
+	pcmpeqd	xmm2,xmm5
+	movdqa	XMMWORD[320+r10],xmm1
+
+	pcmpeqd	xmm3,xmm5
+	movdqa	XMMWORD[336+r10],xmm2
+	pand	xmm0,XMMWORD[64+r12]
+
+	pand	xmm1,XMMWORD[80+r12]
+	pand	xmm2,XMMWORD[96+r12]
+	movdqa	XMMWORD[352+r10],xmm3
+	pand	xmm3,XMMWORD[112+r12]
 	por	xmm0,xmm2
+	por	xmm1,xmm3
+	movdqa	xmm4,XMMWORD[((-128))+r12]
+	movdqa	xmm5,XMMWORD[((-112))+r12]
+	movdqa	xmm2,XMMWORD[((-96))+r12]
+	pand	xmm4,XMMWORD[112+r10]
+	movdqa	xmm3,XMMWORD[((-80))+r12]
+	pand	xmm5,XMMWORD[128+r10]
+	por	xmm0,xmm4
+	pand	xmm2,XMMWORD[144+r10]
+	por	xmm1,xmm5
+	pand	xmm3,XMMWORD[160+r10]
+	por	xmm0,xmm2
+	por	xmm1,xmm3
+	movdqa	xmm4,XMMWORD[((-64))+r12]
+	movdqa	xmm5,XMMWORD[((-48))+r12]
+	movdqa	xmm2,XMMWORD[((-32))+r12]
+	pand	xmm4,XMMWORD[176+r10]
+	movdqa	xmm3,XMMWORD[((-16))+r12]
+	pand	xmm5,XMMWORD[192+r10]
+	por	xmm0,xmm4
+	pand	xmm2,XMMWORD[208+r10]
+	por	xmm1,xmm5
+	pand	xmm3,XMMWORD[224+r10]
+	por	xmm0,xmm2
+	por	xmm1,xmm3
+	movdqa	xmm4,XMMWORD[r12]
+	movdqa	xmm5,XMMWORD[16+r12]
+	movdqa	xmm2,XMMWORD[32+r12]
+	pand	xmm4,XMMWORD[240+r10]
+	movdqa	xmm3,XMMWORD[48+r12]
+	pand	xmm5,XMMWORD[256+r10]
+	por	xmm0,xmm4
+	pand	xmm2,XMMWORD[272+r10]
+	por	xmm1,xmm5
+	pand	xmm3,XMMWORD[288+r10]
+	por	xmm0,xmm2
+	por	xmm1,xmm3
+	por	xmm0,xmm1
+	pshufd	xmm1,xmm0,0x4e
+	por	xmm0,xmm1
 	lea	r12,[256+r12]
-	por	xmm0,xmm3
-
 DB	102,72,15,126,195
 
 	mov	r8,QWORD[r8]
@@ -82,29 +184,14 @@
 	xor	r14,r14
 	xor	r15,r15
 
-	movq	xmm0,QWORD[(((-96)))+r12]
-	movq	xmm1,QWORD[((-32))+r12]
-	pand	xmm0,xmm4
-	movq	xmm2,QWORD[32+r12]
-	pand	xmm1,xmm5
-
 	mov	rbp,r8
 	mul	rbx
 	mov	r10,rax
 	mov	rax,QWORD[rcx]
 
-	movq	xmm3,QWORD[96+r12]
-	pand	xmm2,xmm6
-	por	xmm0,xmm1
-	pand	xmm3,xmm7
-
 	imul	rbp,r10
 	mov	r11,rdx
 
-	por	xmm0,xmm2
-	lea	r12,[256+r12]
-	por	xmm0,xmm3
-
 	mul	rbp
 	add	r10,rax
 	mov	rax,QWORD[8+rsi]
@@ -137,14 +224,12 @@
 	cmp	r15,r9
 	jne	NEAR $L$1st
 
-DB	102,72,15,126,195
 
 	add	r13,rax
-	mov	rax,QWORD[rsi]
 	adc	rdx,0
 	add	r13,r11
 	adc	rdx,0
-	mov	QWORD[((-16))+r15*8+rsp],r13
+	mov	QWORD[((-16))+r9*8+rsp],r13
 	mov	r13,rdx
 	mov	r11,r10
 
@@ -158,33 +243,78 @@
 	jmp	NEAR $L$outer
 ALIGN	16
 $L$outer:
+	lea	rdx,[((24+128))+r9*8+rsp]
+	and	rdx,-16
+	pxor	xmm4,xmm4
+	pxor	xmm5,xmm5
+	movdqa	xmm0,XMMWORD[((-128))+r12]
+	movdqa	xmm1,XMMWORD[((-112))+r12]
+	movdqa	xmm2,XMMWORD[((-96))+r12]
+	movdqa	xmm3,XMMWORD[((-80))+r12]
+	pand	xmm0,XMMWORD[((-128))+rdx]
+	pand	xmm1,XMMWORD[((-112))+rdx]
+	por	xmm4,xmm0
+	pand	xmm2,XMMWORD[((-96))+rdx]
+	por	xmm5,xmm1
+	pand	xmm3,XMMWORD[((-80))+rdx]
+	por	xmm4,xmm2
+	por	xmm5,xmm3
+	movdqa	xmm0,XMMWORD[((-64))+r12]
+	movdqa	xmm1,XMMWORD[((-48))+r12]
+	movdqa	xmm2,XMMWORD[((-32))+r12]
+	movdqa	xmm3,XMMWORD[((-16))+r12]
+	pand	xmm0,XMMWORD[((-64))+rdx]
+	pand	xmm1,XMMWORD[((-48))+rdx]
+	por	xmm4,xmm0
+	pand	xmm2,XMMWORD[((-32))+rdx]
+	por	xmm5,xmm1
+	pand	xmm3,XMMWORD[((-16))+rdx]
+	por	xmm4,xmm2
+	por	xmm5,xmm3
+	movdqa	xmm0,XMMWORD[r12]
+	movdqa	xmm1,XMMWORD[16+r12]
+	movdqa	xmm2,XMMWORD[32+r12]
+	movdqa	xmm3,XMMWORD[48+r12]
+	pand	xmm0,XMMWORD[rdx]
+	pand	xmm1,XMMWORD[16+rdx]
+	por	xmm4,xmm0
+	pand	xmm2,XMMWORD[32+rdx]
+	por	xmm5,xmm1
+	pand	xmm3,XMMWORD[48+rdx]
+	por	xmm4,xmm2
+	por	xmm5,xmm3
+	movdqa	xmm0,XMMWORD[64+r12]
+	movdqa	xmm1,XMMWORD[80+r12]
+	movdqa	xmm2,XMMWORD[96+r12]
+	movdqa	xmm3,XMMWORD[112+r12]
+	pand	xmm0,XMMWORD[64+rdx]
+	pand	xmm1,XMMWORD[80+rdx]
+	por	xmm4,xmm0
+	pand	xmm2,XMMWORD[96+rdx]
+	por	xmm5,xmm1
+	pand	xmm3,XMMWORD[112+rdx]
+	por	xmm4,xmm2
+	por	xmm5,xmm3
+	por	xmm4,xmm5
+	pshufd	xmm0,xmm4,0x4e
+	por	xmm0,xmm4
+	lea	r12,[256+r12]
+
+	mov	rax,QWORD[rsi]
+DB	102,72,15,126,195
+
 	xor	r15,r15
 	mov	rbp,r8
 	mov	r10,QWORD[rsp]
 
-	movq	xmm0,QWORD[(((-96)))+r12]
-	movq	xmm1,QWORD[((-32))+r12]
-	pand	xmm0,xmm4
-	movq	xmm2,QWORD[32+r12]
-	pand	xmm1,xmm5
-
 	mul	rbx
 	add	r10,rax
 	mov	rax,QWORD[rcx]
 	adc	rdx,0
 
-	movq	xmm3,QWORD[96+r12]
-	pand	xmm2,xmm6
-	por	xmm0,xmm1
-	pand	xmm3,xmm7
-
 	imul	rbp,r10
 	mov	r11,rdx
 
-	por	xmm0,xmm2
-	lea	r12,[256+r12]
-	por	xmm0,xmm3
-
 	mul	rbp
 	add	r10,rax
 	mov	rax,QWORD[8+rsi]
@@ -220,15 +350,12 @@
 	cmp	r15,r9
 	jne	NEAR $L$inner
 
-DB	102,72,15,126,195
-
 	add	r13,rax
-	mov	rax,QWORD[rsi]
 	adc	rdx,0
 	add	r13,r10
-	mov	r10,QWORD[r15*8+rsp]
+	mov	r10,QWORD[r9*8+rsp]
 	adc	rdx,0
-	mov	QWORD[((-16))+r15*8+rsp],r13
+	mov	QWORD[((-16))+r9*8+rsp],r13
 	mov	r13,rdx
 
 	xor	rdx,rdx
@@ -274,8 +401,7 @@
 
 	mov	rsi,QWORD[8+r9*8+rsp]
 	mov	rax,1
-	movaps	xmm6,XMMWORD[((-88))+rsi]
-	movaps	xmm7,XMMWORD[((-72))+rsi]
+
 	mov	r15,QWORD[((-48))+rsi]
 	mov	r14,QWORD[((-40))+rsi]
 	mov	r13,QWORD[((-32))+rsi]
@@ -312,13 +438,10 @@
 	push	r13
 	push	r14
 	push	r15
-	lea	rsp,[((-40))+rsp]
-	movaps	XMMWORD[rsp],xmm6
-	movaps	XMMWORD[16+rsp],xmm7
+
 DB	0x67
-	mov	r10d,r9d
 	shl	r9d,3
-	shl	r10d,3+2
+	lea	r10,[r9*2+r9]
 	neg	r9
 
 
@@ -328,19 +451,21 @@
 
 
 
-	lea	r11,[((-64))+r9*2+rsp]
-	sub	r11,rsi
+
+
+	lea	r11,[((-320))+r9*2+rsp]
+	sub	r11,rdi
 	and	r11,4095
 	cmp	r10,r11
 	jb	NEAR $L$mul4xsp_alt
 	sub	rsp,r11
-	lea	rsp,[((-64))+r9*2+rsp]
+	lea	rsp,[((-320))+r9*2+rsp]
 	jmp	NEAR $L$mul4xsp_done
 
 ALIGN	32
 $L$mul4xsp_alt:
-	lea	r10,[((4096-64))+r9*2]
-	lea	rsp,[((-64))+r9*2+rsp]
+	lea	r10,[((4096-320))+r9*2]
+	lea	rsp,[((-320))+r9*2+rsp]
 	sub	r11,r10
 	mov	r10,0
 	cmovc	r11,r10
@@ -356,8 +481,7 @@
 
 	mov	rsi,QWORD[40+rsp]
 	mov	rax,1
-	movaps	xmm6,XMMWORD[((-88))+rsi]
-	movaps	xmm7,XMMWORD[((-72))+rsi]
+
 	mov	r15,QWORD[((-48))+rsi]
 	mov	r14,QWORD[((-40))+rsi]
 	mov	r13,QWORD[((-32))+rsi]
@@ -375,47 +499,141 @@
 ALIGN	32
 mul4x_internal:
 	shl	r9,5
-	mov	r10d,DWORD[56+rax]
-	lea	r13,[256+r9*1+rdx]
+	movd	xmm5,DWORD[56+rax]
+	lea	rax,[$L$inc]
+	lea	r13,[128+r9*1+rdx]
 	shr	r9,5
-	mov	r11,r10
-	shr	r10,3
-	and	r11,7
-	not	r10
-	lea	rax,[$L$magic_masks]
-	and	r10,3
-	lea	r12,[96+r11*8+rdx]
-	movq	xmm4,QWORD[r10*8+rax]
-	movq	xmm5,QWORD[8+r10*8+rax]
-	add	r11,7
-	movq	xmm6,QWORD[16+r10*8+rax]
-	movq	xmm7,QWORD[24+r10*8+rax]
-	and	r11,7
+	movdqa	xmm0,XMMWORD[rax]
+	movdqa	xmm1,XMMWORD[16+rax]
+	lea	r10,[((88-112))+r9*1+rsp]
+	lea	r12,[128+rdx]
 
-	movq	xmm0,QWORD[(((-96)))+r12]
-	lea	r14,[256+r12]
-	movq	xmm1,QWORD[((-32))+r12]
-	pand	xmm0,xmm4
-	movq	xmm2,QWORD[32+r12]
-	pand	xmm1,xmm5
-	movq	xmm3,QWORD[96+r12]
-	pand	xmm2,xmm6
+	pshufd	xmm5,xmm5,0
+	movdqa	xmm4,xmm1
+DB	0x67,0x67
+	movdqa	xmm2,xmm1
+	paddd	xmm1,xmm0
+	pcmpeqd	xmm0,xmm5
 DB	0x67
-	por	xmm0,xmm1
-	movq	xmm1,QWORD[((-96))+r14]
+	movdqa	xmm3,xmm4
+	paddd	xmm2,xmm1
+	pcmpeqd	xmm1,xmm5
+	movdqa	XMMWORD[112+r10],xmm0
+	movdqa	xmm0,xmm4
+
+	paddd	xmm3,xmm2
+	pcmpeqd	xmm2,xmm5
+	movdqa	XMMWORD[128+r10],xmm1
+	movdqa	xmm1,xmm4
+
+	paddd	xmm0,xmm3
+	pcmpeqd	xmm3,xmm5
+	movdqa	XMMWORD[144+r10],xmm2
+	movdqa	xmm2,xmm4
+
+	paddd	xmm1,xmm0
+	pcmpeqd	xmm0,xmm5
+	movdqa	XMMWORD[160+r10],xmm3
+	movdqa	xmm3,xmm4
+	paddd	xmm2,xmm1
+	pcmpeqd	xmm1,xmm5
+	movdqa	XMMWORD[176+r10],xmm0
+	movdqa	xmm0,xmm4
+
+	paddd	xmm3,xmm2
+	pcmpeqd	xmm2,xmm5
+	movdqa	XMMWORD[192+r10],xmm1
+	movdqa	xmm1,xmm4
+
+	paddd	xmm0,xmm3
+	pcmpeqd	xmm3,xmm5
+	movdqa	XMMWORD[208+r10],xmm2
+	movdqa	xmm2,xmm4
+
+	paddd	xmm1,xmm0
+	pcmpeqd	xmm0,xmm5
+	movdqa	XMMWORD[224+r10],xmm3
+	movdqa	xmm3,xmm4
+	paddd	xmm2,xmm1
+	pcmpeqd	xmm1,xmm5
+	movdqa	XMMWORD[240+r10],xmm0
+	movdqa	xmm0,xmm4
+
+	paddd	xmm3,xmm2
+	pcmpeqd	xmm2,xmm5
+	movdqa	XMMWORD[256+r10],xmm1
+	movdqa	xmm1,xmm4
+
+	paddd	xmm0,xmm3
+	pcmpeqd	xmm3,xmm5
+	movdqa	XMMWORD[272+r10],xmm2
+	movdqa	xmm2,xmm4
+
+	paddd	xmm1,xmm0
+	pcmpeqd	xmm0,xmm5
+	movdqa	XMMWORD[288+r10],xmm3
+	movdqa	xmm3,xmm4
+	paddd	xmm2,xmm1
+	pcmpeqd	xmm1,xmm5
+	movdqa	XMMWORD[304+r10],xmm0
+
+	paddd	xmm3,xmm2
 DB	0x67
-	pand	xmm3,xmm7
-DB	0x67
+	pcmpeqd	xmm2,xmm5
+	movdqa	XMMWORD[320+r10],xmm1
+
+	pcmpeqd	xmm3,xmm5
+	movdqa	XMMWORD[336+r10],xmm2
+	pand	xmm0,XMMWORD[64+r12]
+
+	pand	xmm1,XMMWORD[80+r12]
+	pand	xmm2,XMMWORD[96+r12]
+	movdqa	XMMWORD[352+r10],xmm3
+	pand	xmm3,XMMWORD[112+r12]
 	por	xmm0,xmm2
-	movq	xmm2,QWORD[((-32))+r14]
-DB	0x67
-	pand	xmm1,xmm4
-DB	0x67
-	por	xmm0,xmm3
-	movq	xmm3,QWORD[32+r14]
-
+	por	xmm1,xmm3
+	movdqa	xmm4,XMMWORD[((-128))+r12]
+	movdqa	xmm5,XMMWORD[((-112))+r12]
+	movdqa	xmm2,XMMWORD[((-96))+r12]
+	pand	xmm4,XMMWORD[112+r10]
+	movdqa	xmm3,XMMWORD[((-80))+r12]
+	pand	xmm5,XMMWORD[128+r10]
+	por	xmm0,xmm4
+	pand	xmm2,XMMWORD[144+r10]
+	por	xmm1,xmm5
+	pand	xmm3,XMMWORD[160+r10]
+	por	xmm0,xmm2
+	por	xmm1,xmm3
+	movdqa	xmm4,XMMWORD[((-64))+r12]
+	movdqa	xmm5,XMMWORD[((-48))+r12]
+	movdqa	xmm2,XMMWORD[((-32))+r12]
+	pand	xmm4,XMMWORD[176+r10]
+	movdqa	xmm3,XMMWORD[((-16))+r12]
+	pand	xmm5,XMMWORD[192+r10]
+	por	xmm0,xmm4
+	pand	xmm2,XMMWORD[208+r10]
+	por	xmm1,xmm5
+	pand	xmm3,XMMWORD[224+r10]
+	por	xmm0,xmm2
+	por	xmm1,xmm3
+	movdqa	xmm4,XMMWORD[r12]
+	movdqa	xmm5,XMMWORD[16+r12]
+	movdqa	xmm2,XMMWORD[32+r12]
+	pand	xmm4,XMMWORD[240+r10]
+	movdqa	xmm3,XMMWORD[48+r12]
+	pand	xmm5,XMMWORD[256+r10]
+	por	xmm0,xmm4
+	pand	xmm2,XMMWORD[272+r10]
+	por	xmm1,xmm5
+	pand	xmm3,XMMWORD[288+r10]
+	por	xmm0,xmm2
+	por	xmm1,xmm3
+	por	xmm0,xmm1
+	pshufd	xmm1,xmm0,0x4e
+	por	xmm0,xmm1
+	lea	r12,[256+r12]
 DB	102,72,15,126,195
-	movq	xmm0,QWORD[96+r14]
+
 	mov	QWORD[((16+8))+rsp],r13
 	mov	QWORD[((56+8))+rsp],rdi
 
@@ -429,26 +647,10 @@
 	mov	r10,rax
 	mov	rax,QWORD[rcx]
 
-	pand	xmm2,xmm5
-	pand	xmm3,xmm6
-	por	xmm1,xmm2
-
 	imul	rbp,r10
-
-
-
-
-
-
-
-	lea	r14,[((64+8))+r11*8+rsp]
+	lea	r14,[((64+8))+rsp]
 	mov	r11,rdx
 
-	pand	xmm0,xmm7
-	por	xmm1,xmm3
-	lea	r12,[512+r12]
-	por	xmm0,xmm1
-
 	mul	rbp
 	add	r10,rax
 	mov	rax,QWORD[8+r9*1+rsi]
@@ -457,7 +659,7 @@
 
 	mul	rbx
 	add	r11,rax
-	mov	rax,QWORD[16+rcx]
+	mov	rax,QWORD[8+rcx]
 	adc	rdx,0
 	mov	r10,rdx
 
@@ -467,7 +669,7 @@
 	adc	rdx,0
 	add	rdi,r11
 	lea	r15,[32+r9]
-	lea	rcx,[64+rcx]
+	lea	rcx,[32+rcx]
 	adc	rdx,0
 	mov	QWORD[r14],rdi
 	mov	r13,rdx
@@ -477,7 +679,7 @@
 $L$1st4x:
 	mul	rbx
 	add	r10,rax
-	mov	rax,QWORD[((-32))+rcx]
+	mov	rax,QWORD[((-16))+rcx]
 	lea	r14,[32+r14]
 	adc	rdx,0
 	mov	r11,rdx
@@ -493,7 +695,7 @@
 
 	mul	rbx
 	add	r11,rax
-	mov	rax,QWORD[((-16))+rcx]
+	mov	rax,QWORD[((-8))+rcx]
 	adc	rdx,0
 	mov	r10,rdx
 
@@ -523,7 +725,7 @@
 
 	mul	rbx
 	add	r11,rax
-	mov	rax,QWORD[16+rcx]
+	mov	rax,QWORD[8+rcx]
 	adc	rdx,0
 	mov	r10,rdx
 
@@ -532,7 +734,7 @@
 	mov	rax,QWORD[16+r15*1+rsi]
 	adc	rdx,0
 	add	rdi,r11
-	lea	rcx,[64+rcx]
+	lea	rcx,[32+rcx]
 	adc	rdx,0
 	mov	QWORD[r14],rdi
 	mov	r13,rdx
@@ -542,7 +744,7 @@
 
 	mul	rbx
 	add	r10,rax
-	mov	rax,QWORD[((-32))+rcx]
+	mov	rax,QWORD[((-16))+rcx]
 	lea	r14,[32+r14]
 	adc	rdx,0
 	mov	r11,rdx
@@ -558,7 +760,7 @@
 
 	mul	rbx
 	add	r11,rax
-	mov	rax,QWORD[((-16))+rcx]
+	mov	rax,QWORD[((-8))+rcx]
 	adc	rdx,0
 	mov	r10,rdx
 
@@ -571,8 +773,7 @@
 	mov	QWORD[((-16))+r14],rdi
 	mov	r13,rdx
 
-DB	102,72,15,126,195
-	lea	rcx,[r9*2+rcx]
+	lea	rcx,[r9*1+rcx]
 
 	xor	rdi,rdi
 	add	r13,r10
@@ -583,6 +784,63 @@
 
 ALIGN	32
 $L$outer4x:
+	lea	rdx,[((16+128))+r14]
+	pxor	xmm4,xmm4
+	pxor	xmm5,xmm5
+	movdqa	xmm0,XMMWORD[((-128))+r12]
+	movdqa	xmm1,XMMWORD[((-112))+r12]
+	movdqa	xmm2,XMMWORD[((-96))+r12]
+	movdqa	xmm3,XMMWORD[((-80))+r12]
+	pand	xmm0,XMMWORD[((-128))+rdx]
+	pand	xmm1,XMMWORD[((-112))+rdx]
+	por	xmm4,xmm0
+	pand	xmm2,XMMWORD[((-96))+rdx]
+	por	xmm5,xmm1
+	pand	xmm3,XMMWORD[((-80))+rdx]
+	por	xmm4,xmm2
+	por	xmm5,xmm3
+	movdqa	xmm0,XMMWORD[((-64))+r12]
+	movdqa	xmm1,XMMWORD[((-48))+r12]
+	movdqa	xmm2,XMMWORD[((-32))+r12]
+	movdqa	xmm3,XMMWORD[((-16))+r12]
+	pand	xmm0,XMMWORD[((-64))+rdx]
+	pand	xmm1,XMMWORD[((-48))+rdx]
+	por	xmm4,xmm0
+	pand	xmm2,XMMWORD[((-32))+rdx]
+	por	xmm5,xmm1
+	pand	xmm3,XMMWORD[((-16))+rdx]
+	por	xmm4,xmm2
+	por	xmm5,xmm3
+	movdqa	xmm0,XMMWORD[r12]
+	movdqa	xmm1,XMMWORD[16+r12]
+	movdqa	xmm2,XMMWORD[32+r12]
+	movdqa	xmm3,XMMWORD[48+r12]
+	pand	xmm0,XMMWORD[rdx]
+	pand	xmm1,XMMWORD[16+rdx]
+	por	xmm4,xmm0
+	pand	xmm2,XMMWORD[32+rdx]
+	por	xmm5,xmm1
+	pand	xmm3,XMMWORD[48+rdx]
+	por	xmm4,xmm2
+	por	xmm5,xmm3
+	movdqa	xmm0,XMMWORD[64+r12]
+	movdqa	xmm1,XMMWORD[80+r12]
+	movdqa	xmm2,XMMWORD[96+r12]
+	movdqa	xmm3,XMMWORD[112+r12]
+	pand	xmm0,XMMWORD[64+rdx]
+	pand	xmm1,XMMWORD[80+rdx]
+	por	xmm4,xmm0
+	pand	xmm2,XMMWORD[96+rdx]
+	por	xmm5,xmm1
+	pand	xmm3,XMMWORD[112+rdx]
+	por	xmm4,xmm2
+	por	xmm5,xmm3
+	por	xmm4,xmm5
+	pshufd	xmm0,xmm4,0x4e
+	por	xmm0,xmm4
+	lea	r12,[256+r12]
+DB	102,72,15,126,195
+
 	mov	r10,QWORD[r9*1+r14]
 	mov	rbp,r8
 	mul	rbx
@@ -590,25 +848,11 @@
 	mov	rax,QWORD[rcx]
 	adc	rdx,0
 
-	movq	xmm0,QWORD[(((-96)))+r12]
-	movq	xmm1,QWORD[((-32))+r12]
-	pand	xmm0,xmm4
-	movq	xmm2,QWORD[32+r12]
-	pand	xmm1,xmm5
-	movq	xmm3,QWORD[96+r12]
-
 	imul	rbp,r10
-DB	0x67
 	mov	r11,rdx
 	mov	QWORD[r14],rdi
 
-	pand	xmm2,xmm6
-	por	xmm0,xmm1
-	pand	xmm3,xmm7
-	por	xmm0,xmm2
 	lea	r14,[r9*1+r14]
-	lea	r12,[256+r12]
-	por	xmm0,xmm3
 
 	mul	rbp
 	add	r10,rax
@@ -618,7 +862,7 @@
 
 	mul	rbx
 	add	r11,rax
-	mov	rax,QWORD[16+rcx]
+	mov	rax,QWORD[8+rcx]
 	adc	rdx,0
 	add	r11,QWORD[8+r14]
 	adc	rdx,0
@@ -630,7 +874,7 @@
 	adc	rdx,0
 	add	rdi,r11
 	lea	r15,[32+r9]
-	lea	rcx,[64+rcx]
+	lea	rcx,[32+rcx]
 	adc	rdx,0
 	mov	r13,rdx
 	jmp	NEAR $L$inner4x
@@ -639,7 +883,7 @@
 $L$inner4x:
 	mul	rbx
 	add	r10,rax
-	mov	rax,QWORD[((-32))+rcx]
+	mov	rax,QWORD[((-16))+rcx]
 	adc	rdx,0
 	add	r10,QWORD[16+r14]
 	lea	r14,[32+r14]
@@ -657,7 +901,7 @@
 
 	mul	rbx
 	add	r11,rax
-	mov	rax,QWORD[((-16))+rcx]
+	mov	rax,QWORD[((-8))+rcx]
 	adc	rdx,0
 	add	r11,QWORD[((-8))+r14]
 	adc	rdx,0
@@ -691,7 +935,7 @@
 
 	mul	rbx
 	add	r11,rax
-	mov	rax,QWORD[16+rcx]
+	mov	rax,QWORD[8+rcx]
 	adc	rdx,0
 	add	r11,QWORD[8+r14]
 	adc	rdx,0
@@ -702,7 +946,7 @@
 	mov	rax,QWORD[16+r15*1+rsi]
 	adc	rdx,0
 	add	rdi,r11
-	lea	rcx,[64+rcx]
+	lea	rcx,[32+rcx]
 	adc	rdx,0
 	mov	QWORD[((-8))+r14],r13
 	mov	r13,rdx
@@ -712,7 +956,7 @@
 
 	mul	rbx
 	add	r10,rax
-	mov	rax,QWORD[((-32))+rcx]
+	mov	rax,QWORD[((-16))+rcx]
 	adc	rdx,0
 	add	r10,QWORD[16+r14]
 	lea	r14,[32+r14]
@@ -731,7 +975,7 @@
 	mul	rbx
 	add	r11,rax
 	mov	rax,rbp
-	mov	rbp,QWORD[((-16))+rcx]
+	mov	rbp,QWORD[((-8))+rcx]
 	adc	rdx,0
 	add	r11,QWORD[((-8))+r14]
 	adc	rdx,0
@@ -746,9 +990,8 @@
 	mov	QWORD[((-24))+r14],r13
 	mov	r13,rdx
 
-DB	102,72,15,126,195
 	mov	QWORD[((-16))+r14],rdi
-	lea	rcx,[r9*2+rcx]
+	lea	rcx,[r9*1+rcx]
 
 	xor	rdi,rdi
 	add	r13,r10
@@ -759,16 +1002,23 @@
 
 	cmp	r12,QWORD[((16+8))+rsp]
 	jb	NEAR $L$outer4x
+	xor	rax,rax
 	sub	rbp,r13
 	adc	r15,r15
 	or	rdi,r15
-	xor	rdi,1
+	sub	rax,rdi
 	lea	rbx,[r9*1+r14]
-	lea	rbp,[rdi*8+rcx]
+	mov	r12,QWORD[rcx]
+	lea	rbp,[rcx]
 	mov	rcx,r9
 	sar	rcx,3+2
 	mov	rdi,QWORD[((56+8))+rsp]
-	jmp	NEAR $L$sqr4x_sub
+	dec	r12
+	xor	r10,r10
+	mov	r13,QWORD[8+rbp]
+	mov	r14,QWORD[16+rbp]
+	mov	r15,QWORD[24+rbp]
+	jmp	NEAR $L$sqr4x_sub_entry
 
 global	bn_power5
 
@@ -793,12 +1043,9 @@
 	push	r13
 	push	r14
 	push	r15
-	lea	rsp,[((-40))+rsp]
-	movaps	XMMWORD[rsp],xmm6
-	movaps	XMMWORD[16+rsp],xmm7
-	mov	r10d,r9d
+
 	shl	r9d,3
-	shl	r10d,3+2
+	lea	r10d,[r9*2+r9]
 	neg	r9
 	mov	r8,QWORD[r8]
 
@@ -808,19 +1055,20 @@
 
 
 
-	lea	r11,[((-64))+r9*2+rsp]
-	sub	r11,rsi
+
+	lea	r11,[((-320))+r9*2+rsp]
+	sub	r11,rdi
 	and	r11,4095
 	cmp	r10,r11
 	jb	NEAR $L$pwr_sp_alt
 	sub	rsp,r11
-	lea	rsp,[((-64))+r9*2+rsp]
+	lea	rsp,[((-320))+r9*2+rsp]
 	jmp	NEAR $L$pwr_sp_done
 
 ALIGN	32
 $L$pwr_sp_alt:
-	lea	r10,[((4096-64))+r9*2]
-	lea	rsp,[((-64))+r9*2+rsp]
+	lea	r10,[((4096-320))+r9*2]
+	lea	rsp,[((-320))+r9*2+rsp]
 	sub	r11,r10
 	mov	r10,0
 	cmovc	r11,r10
@@ -848,10 +1096,15 @@
 DB	102,72,15,110,226
 
 	call	__bn_sqr8x_internal
+	call	__bn_post4x_internal
 	call	__bn_sqr8x_internal
+	call	__bn_post4x_internal
 	call	__bn_sqr8x_internal
+	call	__bn_post4x_internal
 	call	__bn_sqr8x_internal
+	call	__bn_post4x_internal
 	call	__bn_sqr8x_internal
+	call	__bn_post4x_internal
 
 DB	102,72,15,126,209
 DB	102,72,15,126,226
@@ -1397,9 +1650,9 @@
 	mov	QWORD[((-16))+rdi],rbx
 	mov	QWORD[((-8))+rdi],r8
 DB	102,72,15,126,213
-sqr8x_reduction:
+__bn_sqr8x_reduction:
 	xor	rax,rax
-	lea	rcx,[r9*2+rbp]
+	lea	rcx,[rbp*1+r9]
 	lea	rdx,[((48+8))+r9*2+rsp]
 	mov	QWORD[((0+8))+rsp],rcx
 	lea	rdi,[((48+8))+r9*1+rsp]
@@ -1432,14 +1685,14 @@
 ALIGN	32
 $L$8x_reduce:
 	mul	rbx
-	mov	rax,QWORD[16+rbp]
+	mov	rax,QWORD[8+rbp]
 	neg	r8
 	mov	r8,rdx
 	adc	r8,0
 
 	mul	rbx
 	add	r9,rax
-	mov	rax,QWORD[32+rbp]
+	mov	rax,QWORD[16+rbp]
 	adc	rdx,0
 	add	r8,r9
 	mov	QWORD[((48-8+8))+rcx*8+rsp],rbx
@@ -1448,7 +1701,7 @@
 
 	mul	rbx
 	add	r10,rax
-	mov	rax,QWORD[48+rbp]
+	mov	rax,QWORD[24+rbp]
 	adc	rdx,0
 	add	r9,r10
 	mov	rsi,QWORD[((32+8))+rsp]
@@ -1457,7 +1710,7 @@
 
 	mul	rbx
 	add	r11,rax
-	mov	rax,QWORD[64+rbp]
+	mov	rax,QWORD[32+rbp]
 	adc	rdx,0
 	imul	rsi,r8
 	add	r10,r11
@@ -1466,7 +1719,7 @@
 
 	mul	rbx
 	add	r12,rax
-	mov	rax,QWORD[80+rbp]
+	mov	rax,QWORD[40+rbp]
 	adc	rdx,0
 	add	r11,r12
 	mov	r12,rdx
@@ -1474,7 +1727,7 @@
 
 	mul	rbx
 	add	r13,rax
-	mov	rax,QWORD[96+rbp]
+	mov	rax,QWORD[48+rbp]
 	adc	rdx,0
 	add	r12,r13
 	mov	r13,rdx
@@ -1482,7 +1735,7 @@
 
 	mul	rbx
 	add	r14,rax
-	mov	rax,QWORD[112+rbp]
+	mov	rax,QWORD[56+rbp]
 	adc	rdx,0
 	add	r13,r14
 	mov	r14,rdx
@@ -1500,7 +1753,7 @@
 	dec	ecx
 	jnz	NEAR $L$8x_reduce
 
-	lea	rbp,[128+rbp]
+	lea	rbp,[64+rbp]
 	xor	rax,rax
 	mov	rdx,QWORD[((8+8))+rsp]
 	cmp	rbp,QWORD[((0+8))+rsp]
@@ -1526,14 +1779,14 @@
 $L$8x_tail:
 	mul	rbx
 	add	r8,rax
-	mov	rax,QWORD[16+rbp]
+	mov	rax,QWORD[8+rbp]
 	mov	QWORD[rdi],r8
 	mov	r8,rdx
 	adc	r8,0
 
 	mul	rbx
 	add	r9,rax
-	mov	rax,QWORD[32+rbp]
+	mov	rax,QWORD[16+rbp]
 	adc	rdx,0
 	add	r8,r9
 	lea	rdi,[8+rdi]
@@ -1542,7 +1795,7 @@
 
 	mul	rbx
 	add	r10,rax
-	mov	rax,QWORD[48+rbp]
+	mov	rax,QWORD[24+rbp]
 	adc	rdx,0
 	add	r9,r10
 	mov	r10,rdx
@@ -1550,7 +1803,7 @@
 
 	mul	rbx
 	add	r11,rax
-	mov	rax,QWORD[64+rbp]
+	mov	rax,QWORD[32+rbp]
 	adc	rdx,0
 	add	r10,r11
 	mov	r11,rdx
@@ -1558,7 +1811,7 @@
 
 	mul	rbx
 	add	r12,rax
-	mov	rax,QWORD[80+rbp]
+	mov	rax,QWORD[40+rbp]
 	adc	rdx,0
 	add	r11,r12
 	mov	r12,rdx
@@ -1566,7 +1819,7 @@
 
 	mul	rbx
 	add	r13,rax
-	mov	rax,QWORD[96+rbp]
+	mov	rax,QWORD[48+rbp]
 	adc	rdx,0
 	add	r12,r13
 	mov	r13,rdx
@@ -1574,7 +1827,7 @@
 
 	mul	rbx
 	add	r14,rax
-	mov	rax,QWORD[112+rbp]
+	mov	rax,QWORD[56+rbp]
 	adc	rdx,0
 	add	r13,r14
 	mov	r14,rdx
@@ -1592,7 +1845,7 @@
 	dec	ecx
 	jnz	NEAR $L$8x_tail
 
-	lea	rbp,[128+rbp]
+	lea	rbp,[64+rbp]
 	mov	rdx,QWORD[((8+8))+rsp]
 	cmp	rbp,QWORD[((0+8))+rsp]
 	jae	NEAR $L$8x_tail_done
@@ -1616,6 +1869,15 @@
 ALIGN	32
 $L$8x_tail_done:
 	add	r8,QWORD[rdx]
+	adc	r9,0
+	adc	r10,0
+	adc	r11,0
+	adc	r12,0
+	adc	r13,0
+	adc	r14,0
+	adc	r15,0
+
+
 	xor	rax,rax
 
 	neg	rsi
@@ -1629,7 +1891,7 @@
 	adc	r14,QWORD[48+rdi]
 	adc	r15,QWORD[56+rdi]
 	adc	rax,0
-	mov	rcx,QWORD[((-16))+rbp]
+	mov	rcx,QWORD[((-8))+rbp]
 	xor	rsi,rsi
 
 DB	102,72,15,126,213
@@ -1647,40 +1909,58 @@
 
 	cmp	rdi,rdx
 	jb	NEAR $L$8x_reduction_loop
+	DB	0F3h,0C3h		;repret
 
-	sub	rcx,r15
-	lea	rbx,[r9*1+rdi]
-	adc	rsi,rsi
-	mov	rcx,r9
-	or	rax,rsi
-DB	102,72,15,126,207
-	xor	rax,1
-DB	102,72,15,126,206
-	lea	rbp,[rax*8+rbp]
-	sar	rcx,3+2
-	jmp	NEAR $L$sqr4x_sub
 
 ALIGN	32
+__bn_post4x_internal:
+	mov	r12,QWORD[rbp]
+	lea	rbx,[r9*1+rdi]
+	mov	rcx,r9
+DB	102,72,15,126,207
+	neg	rax
+DB	102,72,15,126,206
+	sar	rcx,3+2
+	dec	r12
+	xor	r10,r10
+	mov	r13,QWORD[8+rbp]
+	mov	r14,QWORD[16+rbp]
+	mov	r15,QWORD[24+rbp]
+	jmp	NEAR $L$sqr4x_sub_entry
+
+ALIGN	16
 $L$sqr4x_sub:
-DB	0x66
-	mov	r12,QWORD[rbx]
-	mov	r13,QWORD[8+rbx]
-	sbb	r12,QWORD[rbp]
-	mov	r14,QWORD[16+rbx]
-	sbb	r13,QWORD[16+rbp]
-	mov	r15,QWORD[24+rbx]
-	lea	rbx,[32+rbx]
-	sbb	r14,QWORD[32+rbp]
+	mov	r12,QWORD[rbp]
+	mov	r13,QWORD[8+rbp]
+	mov	r14,QWORD[16+rbp]
+	mov	r15,QWORD[24+rbp]
+$L$sqr4x_sub_entry:
+	lea	rbp,[32+rbp]
+	not	r12
+	not	r13
+	not	r14
+	not	r15
+	and	r12,rax
+	and	r13,rax
+	and	r14,rax
+	and	r15,rax
+
+	neg	r10
+	adc	r12,QWORD[rbx]
+	adc	r13,QWORD[8+rbx]
+	adc	r14,QWORD[16+rbx]
+	adc	r15,QWORD[24+rbx]
 	mov	QWORD[rdi],r12
-	sbb	r15,QWORD[48+rbp]
-	lea	rbp,[64+rbp]
+	lea	rbx,[32+rbx]
 	mov	QWORD[8+rdi],r13
+	sbb	r10,r10
 	mov	QWORD[16+rdi],r14
 	mov	QWORD[24+rdi],r15
 	lea	rdi,[32+rdi]
 
 	inc	rcx
 	jnz	NEAR $L$sqr4x_sub
+
 	mov	r10,r9
 	neg	r9
 	DB	0F3h,0C3h		;repret
@@ -1718,13 +1998,9 @@
 	push	r13
 	push	r14
 	push	r15
-	lea	rsp,[((-40))+rsp]
-	movaps	XMMWORD[rsp],xmm6
-	movaps	XMMWORD[16+rsp],xmm7
-DB	0x67
-	mov	r10d,r9d
+
 	shl	r9d,3
-	shl	r10d,3+2
+	lea	r10,[r9*2+r9]
 	neg	r9
 	mov	r8,QWORD[r8]
 
@@ -1734,19 +2010,20 @@
 
 
 
-	lea	r11,[((-64))+r9*2+rsp]
-	sub	r11,rsi
+
+	lea	r11,[((-320))+r9*2+rsp]
+	sub	r11,rdi
 	and	r11,4095
 	cmp	r10,r11
 	jb	NEAR $L$from_sp_alt
 	sub	rsp,r11
-	lea	rsp,[((-64))+r9*2+rsp]
+	lea	rsp,[((-320))+r9*2+rsp]
 	jmp	NEAR $L$from_sp_done
 
 ALIGN	32
 $L$from_sp_alt:
-	lea	r10,[((4096-64))+r9*2]
-	lea	rsp,[((-64))+r9*2+rsp]
+	lea	r10,[((4096-320))+r9*2]
+	lea	rsp,[((-320))+r9*2+rsp]
 	sub	r11,r10
 	mov	r10,0
 	cmovc	r11,r10
@@ -1797,7 +2074,8 @@
 DB	0x67
 	mov	rbp,rcx
 DB	102,73,15,110,218
-	call	sqr8x_reduction
+	call	__bn_sqr8x_reduction
+	call	__bn_post4x_internal
 
 	pxor	xmm0,xmm0
 	lea	rax,[48+rsp]
@@ -1847,55 +2125,171 @@
 
 global	bn_gather5
 
-ALIGN	16
+ALIGN	32
 bn_gather5:
 $L$SEH_begin_bn_gather5:
 
-DB	0x48,0x83,0xec,0x28
-DB	0x0f,0x29,0x34,0x24
-DB	0x0f,0x29,0x7c,0x24,0x10
-	mov	r11d,r9d
-	shr	r9d,3
-	and	r11,7
-	not	r9d
-	lea	rax,[$L$magic_masks]
-	and	r9d,3
-	lea	r8,[128+r11*8+r8]
-	movq	xmm4,QWORD[r9*8+rax]
-	movq	xmm5,QWORD[8+r9*8+rax]
-	movq	xmm6,QWORD[16+r9*8+rax]
-	movq	xmm7,QWORD[24+r9*8+rax]
-	jmp	NEAR $L$gather
-ALIGN	16
-$L$gather:
-	movq	xmm0,QWORD[(((-128)))+r8]
-	movq	xmm1,QWORD[((-64))+r8]
-	pand	xmm0,xmm4
-	movq	xmm2,QWORD[r8]
-	pand	xmm1,xmm5
-	movq	xmm3,QWORD[64+r8]
-	pand	xmm2,xmm6
-	por	xmm0,xmm1
-	pand	xmm3,xmm7
-DB	0x67,0x67
-	por	xmm0,xmm2
-	lea	r8,[256+r8]
-	por	xmm0,xmm3
+DB	0x4c,0x8d,0x14,0x24
+DB	0x48,0x81,0xec,0x08,0x01,0x00,0x00
+	lea	rax,[$L$inc]
+	and	rsp,-16
 
+	movd	xmm5,r9d
+	movdqa	xmm0,XMMWORD[rax]
+	movdqa	xmm1,XMMWORD[16+rax]
+	lea	r11,[128+r8]
+	lea	rax,[128+rsp]
+
+	pshufd	xmm5,xmm5,0
+	movdqa	xmm4,xmm1
+	movdqa	xmm2,xmm1
+	paddd	xmm1,xmm0
+	pcmpeqd	xmm0,xmm5
+	movdqa	xmm3,xmm4
+
+	paddd	xmm2,xmm1
+	pcmpeqd	xmm1,xmm5
+	movdqa	XMMWORD[(-128)+rax],xmm0
+	movdqa	xmm0,xmm4
+
+	paddd	xmm3,xmm2
+	pcmpeqd	xmm2,xmm5
+	movdqa	XMMWORD[(-112)+rax],xmm1
+	movdqa	xmm1,xmm4
+
+	paddd	xmm0,xmm3
+	pcmpeqd	xmm3,xmm5
+	movdqa	XMMWORD[(-96)+rax],xmm2
+	movdqa	xmm2,xmm4
+	paddd	xmm1,xmm0
+	pcmpeqd	xmm0,xmm5
+	movdqa	XMMWORD[(-80)+rax],xmm3
+	movdqa	xmm3,xmm4
+
+	paddd	xmm2,xmm1
+	pcmpeqd	xmm1,xmm5
+	movdqa	XMMWORD[(-64)+rax],xmm0
+	movdqa	xmm0,xmm4
+
+	paddd	xmm3,xmm2
+	pcmpeqd	xmm2,xmm5
+	movdqa	XMMWORD[(-48)+rax],xmm1
+	movdqa	xmm1,xmm4
+
+	paddd	xmm0,xmm3
+	pcmpeqd	xmm3,xmm5
+	movdqa	XMMWORD[(-32)+rax],xmm2
+	movdqa	xmm2,xmm4
+	paddd	xmm1,xmm0
+	pcmpeqd	xmm0,xmm5
+	movdqa	XMMWORD[(-16)+rax],xmm3
+	movdqa	xmm3,xmm4
+
+	paddd	xmm2,xmm1
+	pcmpeqd	xmm1,xmm5
+	movdqa	XMMWORD[rax],xmm0
+	movdqa	xmm0,xmm4
+
+	paddd	xmm3,xmm2
+	pcmpeqd	xmm2,xmm5
+	movdqa	XMMWORD[16+rax],xmm1
+	movdqa	xmm1,xmm4
+
+	paddd	xmm0,xmm3
+	pcmpeqd	xmm3,xmm5
+	movdqa	XMMWORD[32+rax],xmm2
+	movdqa	xmm2,xmm4
+	paddd	xmm1,xmm0
+	pcmpeqd	xmm0,xmm5
+	movdqa	XMMWORD[48+rax],xmm3
+	movdqa	xmm3,xmm4
+
+	paddd	xmm2,xmm1
+	pcmpeqd	xmm1,xmm5
+	movdqa	XMMWORD[64+rax],xmm0
+	movdqa	xmm0,xmm4
+
+	paddd	xmm3,xmm2
+	pcmpeqd	xmm2,xmm5
+	movdqa	XMMWORD[80+rax],xmm1
+	movdqa	xmm1,xmm4
+
+	paddd	xmm0,xmm3
+	pcmpeqd	xmm3,xmm5
+	movdqa	XMMWORD[96+rax],xmm2
+	movdqa	xmm2,xmm4
+	movdqa	XMMWORD[112+rax],xmm3
+	jmp	NEAR $L$gather
+
+ALIGN	32
+$L$gather:
+	pxor	xmm4,xmm4
+	pxor	xmm5,xmm5
+	movdqa	xmm0,XMMWORD[((-128))+r11]
+	movdqa	xmm1,XMMWORD[((-112))+r11]
+	movdqa	xmm2,XMMWORD[((-96))+r11]
+	pand	xmm0,XMMWORD[((-128))+rax]
+	movdqa	xmm3,XMMWORD[((-80))+r11]
+	pand	xmm1,XMMWORD[((-112))+rax]
+	por	xmm4,xmm0
+	pand	xmm2,XMMWORD[((-96))+rax]
+	por	xmm5,xmm1
+	pand	xmm3,XMMWORD[((-80))+rax]
+	por	xmm4,xmm2
+	por	xmm5,xmm3
+	movdqa	xmm0,XMMWORD[((-64))+r11]
+	movdqa	xmm1,XMMWORD[((-48))+r11]
+	movdqa	xmm2,XMMWORD[((-32))+r11]
+	pand	xmm0,XMMWORD[((-64))+rax]
+	movdqa	xmm3,XMMWORD[((-16))+r11]
+	pand	xmm1,XMMWORD[((-48))+rax]
+	por	xmm4,xmm0
+	pand	xmm2,XMMWORD[((-32))+rax]
+	por	xmm5,xmm1
+	pand	xmm3,XMMWORD[((-16))+rax]
+	por	xmm4,xmm2
+	por	xmm5,xmm3
+	movdqa	xmm0,XMMWORD[r11]
+	movdqa	xmm1,XMMWORD[16+r11]
+	movdqa	xmm2,XMMWORD[32+r11]
+	pand	xmm0,XMMWORD[rax]
+	movdqa	xmm3,XMMWORD[48+r11]
+	pand	xmm1,XMMWORD[16+rax]
+	por	xmm4,xmm0
+	pand	xmm2,XMMWORD[32+rax]
+	por	xmm5,xmm1
+	pand	xmm3,XMMWORD[48+rax]
+	por	xmm4,xmm2
+	por	xmm5,xmm3
+	movdqa	xmm0,XMMWORD[64+r11]
+	movdqa	xmm1,XMMWORD[80+r11]
+	movdqa	xmm2,XMMWORD[96+r11]
+	pand	xmm0,XMMWORD[64+rax]
+	movdqa	xmm3,XMMWORD[112+r11]
+	pand	xmm1,XMMWORD[80+rax]
+	por	xmm4,xmm0
+	pand	xmm2,XMMWORD[96+rax]
+	por	xmm5,xmm1
+	pand	xmm3,XMMWORD[112+rax]
+	por	xmm4,xmm2
+	por	xmm5,xmm3
+	por	xmm4,xmm5
+	lea	r11,[256+r11]
+	pshufd	xmm0,xmm4,0x4e
+	por	xmm0,xmm4
 	movq	QWORD[rcx],xmm0
 	lea	rcx,[8+rcx]
 	sub	edx,1
 	jnz	NEAR $L$gather
-	movaps	xmm6,XMMWORD[rsp]
-	movaps	xmm7,XMMWORD[16+rsp]
-	lea	rsp,[40+rsp]
+
+	lea	rsp,[r10]
 	DB	0F3h,0C3h		;repret
 $L$SEH_end_bn_gather5:
 
 ALIGN	64
-$L$magic_masks:
-	DD	0,0,0,0,0,0,-1,-1
-	DD	0,0,0,0,0,0,0,0
+$L$inc:
+	DD	0,0,1,1
+	DD	2,2,2,2
 DB	77,111,110,116,103,111,109,101,114,121,32,77,117,108,116,105
 DB	112,108,105,99,97,116,105,111,110,32,119,105,116,104,32,115
 DB	99,97,116,116,101,114,47,103,97,116,104,101,114,32,102,111
@@ -1937,19 +2331,16 @@
 
 	lea	r10,[$L$mul_epilogue]
 	cmp	rbx,r10
-	jb	NEAR $L$body_40
+	ja	NEAR $L$body_40
 
 	mov	r10,QWORD[192+r8]
 	mov	rax,QWORD[8+r10*8+rax]
+
 	jmp	NEAR $L$body_proceed
 
 $L$body_40:
 	mov	rax,QWORD[40+rax]
 $L$body_proceed:
-
-	movaps	xmm0,XMMWORD[((-88))+rax]
-	movaps	xmm1,XMMWORD[((-72))+rax]
-
 	mov	rbx,QWORD[((-8))+rax]
 	mov	rbp,QWORD[((-16))+rax]
 	mov	r12,QWORD[((-24))+rax]
@@ -1962,8 +2353,6 @@
 	mov	QWORD[224+r8],r13
 	mov	QWORD[232+r8],r14
 	mov	QWORD[240+r8],r15
-	movups	XMMWORD[512+r8],xmm0
-	movups	XMMWORD[528+r8],xmm1
 
 $L$common_seh_tail:
 	mov	rdi,QWORD[8+rax]
@@ -2049,8 +2438,7 @@
 	DD	$L$from_body wrt ..imagebase,$L$from_epilogue wrt ..imagebase
 ALIGN	8
 $L$SEH_info_bn_gather5:
-DB	0x01,0x0d,0x05,0x00
-DB	0x0d,0x78,0x01,0x00
-DB	0x08,0x68,0x00,0x00
-DB	0x04,0x42,0x00,0x00
+DB	0x01,0x0b,0x03,0x0a
+DB	0x0b,0x01,0x21,0x00
+DB	0x04,0xa3,0x00,0x00
 ALIGN	8

third_party/boringssl/win-x86_64/crypto/chacha/chacha-x86_64.asm

diff --git a/third_party/boringssl/win-x86_64/crypto/chacha/chacha-x86_64.asm b/third_party/boringssl/win-x86_64/crypto/chacha/chacha-x86_64.asm
new file mode 100644
index 0000000..afebd2e
--- /dev/null
+++ b/third_party/boringssl/win-x86_64/crypto/chacha/chacha-x86_64.asm
@@ -0,0 +1,1689 @@
+default	rel
+%define XMMWORD
+%define YMMWORD
+%define ZMMWORD
+section	.text code align=64
+
+
+EXTERN	OPENSSL_ia32cap_P
+
+ALIGN	64
+$L$zero:
+	DD	0,0,0,0
+$L$one:
+	DD	1,0,0,0
+$L$inc:
+	DD	0,1,2,3
+$L$four:
+	DD	4,4,4,4
+$L$incy:
+	DD	0,2,4,6,1,3,5,7
+$L$eight:
+	DD	8,8,8,8,8,8,8,8
+$L$rot16:
+DB	0x2,0x3,0x0,0x1,0x6,0x7,0x4,0x5,0xa,0xb,0x8,0x9,0xe,0xf,0xc,0xd
+$L$rot24:
+DB	0x3,0x0,0x1,0x2,0x7,0x4,0x5,0x6,0xb,0x8,0x9,0xa,0xf,0xc,0xd,0xe
+$L$sigma:
+DB	101,120,112,97,110,100,32,51,50,45,98,121,116,101,32,107
+DB	0
+DB	67,104,97,67,104,97,50,48,32,102,111,114,32,120,56,54
+DB	95,54,52,44,32,67,82,89,80,84,79,71,65,77,83,32
+DB	98,121,32,60,97,112,112,114,111,64,111,112,101,110,115,115
+DB	108,46,111,114,103,62,0
+global	ChaCha20_ctr32
+
+ALIGN	64
+ChaCha20_ctr32:
+	mov	QWORD[8+rsp],rdi	;WIN64 prologue
+	mov	QWORD[16+rsp],rsi
+	mov	rax,rsp
+$L$SEH_begin_ChaCha20_ctr32:
+	mov	rdi,rcx
+	mov	rsi,rdx
+	mov	rdx,r8
+	mov	rcx,r9
+	mov	r8,QWORD[40+rsp]
+
+
+	cmp	rdx,0
+	je	NEAR $L$no_data
+	mov	r10,QWORD[((OPENSSL_ia32cap_P+4))]
+	test	r10d,512
+	jnz	NEAR $L$ChaCha20_ssse3
+
+	push	rbx
+	push	rbp
+	push	r12
+	push	r13
+	push	r14
+	push	r15
+	sub	rsp,64+24
+
+
+	movdqu	xmm1,XMMWORD[rcx]
+	movdqu	xmm2,XMMWORD[16+rcx]
+	movdqu	xmm3,XMMWORD[r8]
+	movdqa	xmm4,XMMWORD[$L$one]
+
+
+	movdqa	XMMWORD[16+rsp],xmm1
+	movdqa	XMMWORD[32+rsp],xmm2
+	movdqa	XMMWORD[48+rsp],xmm3
+	mov	rbp,rdx
+	jmp	NEAR $L$oop_outer
+
+ALIGN	32
+$L$oop_outer:
+	mov	eax,0x61707865
+	mov	ebx,0x3320646e
+	mov	ecx,0x79622d32
+	mov	edx,0x6b206574
+	mov	r8d,DWORD[16+rsp]
+	mov	r9d,DWORD[20+rsp]
+	mov	r10d,DWORD[24+rsp]
+	mov	r11d,DWORD[28+rsp]
+	movd	r12d,xmm3
+	mov	r13d,DWORD[52+rsp]
+	mov	r14d,DWORD[56+rsp]
+	mov	r15d,DWORD[60+rsp]
+
+	mov	QWORD[((64+0))+rsp],rbp
+	mov	ebp,10
+	mov	QWORD[((64+8))+rsp],rsi
+DB	102,72,15,126,214
+	mov	QWORD[((64+16))+rsp],rdi
+	mov	rdi,rsi
+	shr	rdi,32
+	jmp	NEAR $L$oop
+
+ALIGN	32
+$L$oop:
+	add	eax,r8d
+	xor	r12d,eax
+	rol	r12d,16
+	add	ebx,r9d
+	xor	r13d,ebx
+	rol	r13d,16
+	add	esi,r12d
+	xor	r8d,esi
+	rol	r8d,12
+	add	edi,r13d
+	xor	r9d,edi
+	rol	r9d,12
+	add	eax,r8d
+	xor	r12d,eax
+	rol	r12d,8
+	add	ebx,r9d
+	xor	r13d,ebx
+	rol	r13d,8
+	add	esi,r12d
+	xor	r8d,esi
+	rol	r8d,7
+	add	edi,r13d
+	xor	r9d,edi
+	rol	r9d,7
+	mov	DWORD[32+rsp],esi
+	mov	DWORD[36+rsp],edi
+	mov	esi,DWORD[40+rsp]
+	mov	edi,DWORD[44+rsp]
+	add	ecx,r10d
+	xor	r14d,ecx
+	rol	r14d,16
+	add	edx,r11d
+	xor	r15d,edx
+	rol	r15d,16
+	add	esi,r14d
+	xor	r10d,esi
+	rol	r10d,12
+	add	edi,r15d
+	xor	r11d,edi
+	rol	r11d,12
+	add	ecx,r10d
+	xor	r14d,ecx
+	rol	r14d,8
+	add	edx,r11d
+	xor	r15d,edx
+	rol	r15d,8
+	add	esi,r14d
+	xor	r10d,esi
+	rol	r10d,7
+	add	edi,r15d
+	xor	r11d,edi
+	rol	r11d,7
+	add	eax,r9d
+	xor	r15d,eax
+	rol	r15d,16
+	add	ebx,r10d
+	xor	r12d,ebx
+	rol	r12d,16
+	add	esi,r15d
+	xor	r9d,esi
+	rol	r9d,12
+	add	edi,r12d
+	xor	r10d,edi
+	rol	r10d,12
+	add	eax,r9d
+	xor	r15d,eax
+	rol	r15d,8
+	add	ebx,r10d
+	xor	r12d,ebx
+	rol	r12d,8
+	add	esi,r15d
+	xor	r9d,esi
+	rol	r9d,7
+	add	edi,r12d
+	xor	r10d,edi
+	rol	r10d,7
+	mov	DWORD[40+rsp],esi
+	mov	DWORD[44+rsp],edi
+	mov	esi,DWORD[32+rsp]
+	mov	edi,DWORD[36+rsp]
+	add	ecx,r11d
+	xor	r13d,ecx
+	rol	r13d,16
+	add	edx,r8d
+	xor	r14d,edx
+	rol	r14d,16
+	add	esi,r13d
+	xor	r11d,esi
+	rol	r11d,12
+	add	edi,r14d
+	xor	r8d,edi
+	rol	r8d,12
+	add	ecx,r11d
+	xor	r13d,ecx
+	rol	r13d,8
+	add	edx,r8d
+	xor	r14d,edx
+	rol	r14d,8
+	add	esi,r13d
+	xor	r11d,esi
+	rol	r11d,7
+	add	edi,r14d
+	xor	r8d,edi
+	rol	r8d,7
+	dec	ebp
+	jnz	NEAR $L$oop
+	mov	DWORD[36+rsp],edi
+	mov	DWORD[32+rsp],esi
+	mov	rbp,QWORD[64+rsp]
+	movdqa	xmm1,xmm2
+	mov	rsi,QWORD[((64+8))+rsp]
+	paddd	xmm3,xmm4
+	mov	rdi,QWORD[((64+16))+rsp]
+
+	add	eax,0x61707865
+	add	ebx,0x3320646e
+	add	ecx,0x79622d32
+	add	edx,0x6b206574
+	add	r8d,DWORD[16+rsp]
+	add	r9d,DWORD[20+rsp]
+	add	r10d,DWORD[24+rsp]
+	add	r11d,DWORD[28+rsp]
+	add	r12d,DWORD[48+rsp]
+	add	r13d,DWORD[52+rsp]
+	add	r14d,DWORD[56+rsp]
+	add	r15d,DWORD[60+rsp]
+	paddd	xmm1,XMMWORD[32+rsp]
+
+	cmp	rbp,64
+	jb	NEAR $L$tail
+
+	xor	eax,DWORD[rsi]
+	xor	ebx,DWORD[4+rsi]
+	xor	ecx,DWORD[8+rsi]
+	xor	edx,DWORD[12+rsi]
+	xor	r8d,DWORD[16+rsi]
+	xor	r9d,DWORD[20+rsi]
+	xor	r10d,DWORD[24+rsi]
+	xor	r11d,DWORD[28+rsi]
+	movdqu	xmm0,XMMWORD[32+rsi]
+	xor	r12d,DWORD[48+rsi]
+	xor	r13d,DWORD[52+rsi]
+	xor	r14d,DWORD[56+rsi]
+	xor	r15d,DWORD[60+rsi]
+	lea	rsi,[64+rsi]
+	pxor	xmm0,xmm1
+
+	movdqa	XMMWORD[32+rsp],xmm2
+	movd	DWORD[48+rsp],xmm3
+
+	mov	DWORD[rdi],eax
+	mov	DWORD[4+rdi],ebx
+	mov	DWORD[8+rdi],ecx
+	mov	DWORD[12+rdi],edx
+	mov	DWORD[16+rdi],r8d
+	mov	DWORD[20+rdi],r9d
+	mov	DWORD[24+rdi],r10d
+	mov	DWORD[28+rdi],r11d
+	movdqu	XMMWORD[32+rdi],xmm0
+	mov	DWORD[48+rdi],r12d
+	mov	DWORD[52+rdi],r13d
+	mov	DWORD[56+rdi],r14d
+	mov	DWORD[60+rdi],r15d
+	lea	rdi,[64+rdi]
+
+	sub	rbp,64
+	jnz	NEAR $L$oop_outer
+
+	jmp	NEAR $L$done
+
+ALIGN	16
+$L$tail:
+	mov	DWORD[rsp],eax
+	mov	DWORD[4+rsp],ebx
+	xor	rbx,rbx
+	mov	DWORD[8+rsp],ecx
+	mov	DWORD[12+rsp],edx
+	mov	DWORD[16+rsp],r8d
+	mov	DWORD[20+rsp],r9d
+	mov	DWORD[24+rsp],r10d
+	mov	DWORD[28+rsp],r11d
+	movdqa	XMMWORD[32+rsp],xmm1
+	mov	DWORD[48+rsp],r12d
+	mov	DWORD[52+rsp],r13d
+	mov	DWORD[56+rsp],r14d
+	mov	DWORD[60+rsp],r15d
+
+$L$oop_tail:
+	movzx	eax,BYTE[rbx*1+rsi]
+	movzx	edx,BYTE[rbx*1+rsp]
+	lea	rbx,[1+rbx]
+	xor	eax,edx
+	mov	BYTE[((-1))+rbx*1+rdi],al
+	dec	rbp
+	jnz	NEAR $L$oop_tail
+
+$L$done:
+	add	rsp,64+24
+	pop	r15
+	pop	r14
+	pop	r13
+	pop	r12
+	pop	rbp
+	pop	rbx
+$L$no_data:
+	mov	rdi,QWORD[8+rsp]	;WIN64 epilogue
+	mov	rsi,QWORD[16+rsp]
+	DB	0F3h,0C3h		;repret
+$L$SEH_end_ChaCha20_ctr32:
+
+ALIGN	32
+ChaCha20_ssse3:
+	mov	QWORD[8+rsp],rdi	;WIN64 prologue
+	mov	QWORD[16+rsp],rsi
+	mov	rax,rsp
+$L$SEH_begin_ChaCha20_ssse3:
+	mov	rdi,rcx
+	mov	rsi,rdx
+	mov	rdx,r8
+	mov	rcx,r9
+	mov	r8,QWORD[40+rsp]
+
+
+$L$ChaCha20_ssse3:
+	cmp	rdx,128
+	ja	NEAR $L$ChaCha20_4x
+
+$L$do_sse3_after_all:
+	push	rbx
+	push	rbp
+	push	r12
+	push	r13
+	push	r14
+	push	r15
+
+	sub	rsp,64+72
+	movaps	XMMWORD[(64+32)+rsp],xmm6
+	movaps	XMMWORD[(64+48)+rsp],xmm7
+	movdqa	xmm0,XMMWORD[$L$sigma]
+	movdqu	xmm1,XMMWORD[rcx]
+	movdqu	xmm2,XMMWORD[16+rcx]
+	movdqu	xmm3,XMMWORD[r8]
+	movdqa	xmm6,XMMWORD[$L$rot16]
+	movdqa	xmm7,XMMWORD[$L$rot24]
+
+	movdqa	XMMWORD[rsp],xmm0
+	movdqa	XMMWORD[16+rsp],xmm1
+	movdqa	XMMWORD[32+rsp],xmm2
+	movdqa	XMMWORD[48+rsp],xmm3
+	mov	ebp,10
+	jmp	NEAR $L$oop_ssse3
+
+ALIGN	32
+$L$oop_outer_ssse3:
+	movdqa	xmm3,XMMWORD[$L$one]
+	movdqa	xmm0,XMMWORD[rsp]
+	movdqa	xmm1,XMMWORD[16+rsp]
+	movdqa	xmm2,XMMWORD[32+rsp]
+	paddd	xmm3,XMMWORD[48+rsp]
+	mov	ebp,10
+	movdqa	XMMWORD[48+rsp],xmm3
+	jmp	NEAR $L$oop_ssse3
+
+ALIGN	32
+$L$oop_ssse3:
+	paddd	xmm0,xmm1
+	pxor	xmm3,xmm0
+DB	102,15,56,0,222
+	paddd	xmm2,xmm3
+	pxor	xmm1,xmm2
+	movdqa	xmm4,xmm1
+	psrld	xmm1,20
+	pslld	xmm4,12
+	por	xmm1,xmm4
+	paddd	xmm0,xmm1
+	pxor	xmm3,xmm0
+DB	102,15,56,0,223
+	paddd	xmm2,xmm3
+	pxor	xmm1,xmm2
+	movdqa	xmm4,xmm1
+	psrld	xmm1,25
+	pslld	xmm4,7
+	por	xmm1,xmm4
+	pshufd	xmm2,xmm2,78
+	pshufd	xmm1,xmm1,57
+	pshufd	xmm3,xmm3,147
+	nop
+	paddd	xmm0,xmm1
+	pxor	xmm3,xmm0
+DB	102,15,56,0,222
+	paddd	xmm2,xmm3
+	pxor	xmm1,xmm2
+	movdqa	xmm4,xmm1
+	psrld	xmm1,20
+	pslld	xmm4,12
+	por	xmm1,xmm4
+	paddd	xmm0,xmm1
+	pxor	xmm3,xmm0
+DB	102,15,56,0,223
+	paddd	xmm2,xmm3
+	pxor	xmm1,xmm2
+	movdqa	xmm4,xmm1
+	psrld	xmm1,25
+	pslld	xmm4,7
+	por	xmm1,xmm4
+	pshufd	xmm2,xmm2,78
+	pshufd	xmm1,xmm1,147
+	pshufd	xmm3,xmm3,57
+	dec	ebp
+	jnz	NEAR $L$oop_ssse3
+	paddd	xmm0,XMMWORD[rsp]
+	paddd	xmm1,XMMWORD[16+rsp]
+	paddd	xmm2,XMMWORD[32+rsp]
+	paddd	xmm3,XMMWORD[48+rsp]
+
+	cmp	rdx,64
+	jb	NEAR $L$tail_ssse3
+
+	movdqu	xmm4,XMMWORD[rsi]
+	movdqu	xmm5,XMMWORD[16+rsi]
+	pxor	xmm0,xmm4
+	movdqu	xmm4,XMMWORD[32+rsi]
+	pxor	xmm1,xmm5
+	movdqu	xmm5,XMMWORD[48+rsi]
+	lea	rsi,[64+rsi]
+	pxor	xmm2,xmm4
+	pxor	xmm3,xmm5
+
+	movdqu	XMMWORD[rdi],xmm0
+	movdqu	XMMWORD[16+rdi],xmm1
+	movdqu	XMMWORD[32+rdi],xmm2
+	movdqu	XMMWORD[48+rdi],xmm3
+	lea	rdi,[64+rdi]
+
+	sub	rdx,64
+	jnz	NEAR $L$oop_outer_ssse3
+
+	jmp	NEAR $L$done_ssse3
+
+ALIGN	16
+$L$tail_ssse3:
+	movdqa	XMMWORD[rsp],xmm0
+	movdqa	XMMWORD[16+rsp],xmm1
+	movdqa	XMMWORD[32+rsp],xmm2
+	movdqa	XMMWORD[48+rsp],xmm3
+	xor	rbx,rbx
+
+$L$oop_tail_ssse3:
+	movzx	eax,BYTE[rbx*1+rsi]
+	movzx	ecx,BYTE[rbx*1+rsp]
+	lea	rbx,[1+rbx]
+	xor	eax,ecx
+	mov	BYTE[((-1))+rbx*1+rdi],al
+	dec	rdx
+	jnz	NEAR $L$oop_tail_ssse3
+
+$L$done_ssse3:
+	movaps	xmm6,XMMWORD[((64+32))+rsp]
+	movaps	xmm7,XMMWORD[((64+48))+rsp]
+	add	rsp,64+72
+	pop	r15
+	pop	r14
+	pop	r13
+	pop	r12
+	pop	rbp
+	pop	rbx
+	mov	rdi,QWORD[8+rsp]	;WIN64 epilogue
+	mov	rsi,QWORD[16+rsp]
+	DB	0F3h,0C3h		;repret
+$L$SEH_end_ChaCha20_ssse3:
+
+ALIGN	32
+ChaCha20_4x:
+	mov	QWORD[8+rsp],rdi	;WIN64 prologue
+	mov	QWORD[16+rsp],rsi
+	mov	rax,rsp
+$L$SEH_begin_ChaCha20_4x:
+	mov	rdi,rcx
+	mov	rsi,rdx
+	mov	rdx,r8
+	mov	rcx,r9
+	mov	r8,QWORD[40+rsp]
+
+
+$L$ChaCha20_4x:
+	mov	r11,r10
+	shr	r10,32
+	test	r10,32
+	jnz	NEAR $L$ChaCha20_8x
+	cmp	rdx,192
+	ja	NEAR $L$proceed4x
+
+	and	r11,71303168
+	cmp	r11,4194304
+	je	NEAR $L$do_sse3_after_all
+
+$L$proceed4x:
+	lea	r11,[((-120))+rsp]
+	sub	rsp,0x148+160
+	movaps	XMMWORD[(-48)+r11],xmm6
+	movaps	XMMWORD[(-32)+r11],xmm7
+	movaps	XMMWORD[(-16)+r11],xmm8
+	movaps	XMMWORD[r11],xmm9
+	movaps	XMMWORD[16+r11],xmm10
+	movaps	XMMWORD[32+r11],xmm11
+	movaps	XMMWORD[48+r11],xmm12
+	movaps	XMMWORD[64+r11],xmm13
+	movaps	XMMWORD[80+r11],xmm14
+	movaps	XMMWORD[96+r11],xmm15
+	movdqa	xmm11,XMMWORD[$L$sigma]
+	movdqu	xmm15,XMMWORD[rcx]
+	movdqu	xmm7,XMMWORD[16+rcx]
+	movdqu	xmm3,XMMWORD[r8]
+	lea	rcx,[256+rsp]
+	lea	r10,[$L$rot16]
+	lea	r11,[$L$rot24]
+
+	pshufd	xmm8,xmm11,0x00
+	pshufd	xmm9,xmm11,0x55
+	movdqa	XMMWORD[64+rsp],xmm8
+	pshufd	xmm10,xmm11,0xaa
+	movdqa	XMMWORD[80+rsp],xmm9
+	pshufd	xmm11,xmm11,0xff
+	movdqa	XMMWORD[96+rsp],xmm10
+	movdqa	XMMWORD[112+rsp],xmm11
+
+	pshufd	xmm12,xmm15,0x00
+	pshufd	xmm13,xmm15,0x55
+	movdqa	XMMWORD[(128-256)+rcx],xmm12
+	pshufd	xmm14,xmm15,0xaa
+	movdqa	XMMWORD[(144-256)+rcx],xmm13
+	pshufd	xmm15,xmm15,0xff
+	movdqa	XMMWORD[(160-256)+rcx],xmm14
+	movdqa	XMMWORD[(176-256)+rcx],xmm15
+
+	pshufd	xmm4,xmm7,0x00
+	pshufd	xmm5,xmm7,0x55
+	movdqa	XMMWORD[(192-256)+rcx],xmm4
+	pshufd	xmm6,xmm7,0xaa
+	movdqa	XMMWORD[(208-256)+rcx],xmm5
+	pshufd	xmm7,xmm7,0xff
+	movdqa	XMMWORD[(224-256)+rcx],xmm6
+	movdqa	XMMWORD[(240-256)+rcx],xmm7
+
+	pshufd	xmm0,xmm3,0x00
+	pshufd	xmm1,xmm3,0x55
+	paddd	xmm0,XMMWORD[$L$inc]
+	pshufd	xmm2,xmm3,0xaa
+	movdqa	XMMWORD[(272-256)+rcx],xmm1
+	pshufd	xmm3,xmm3,0xff
+	movdqa	XMMWORD[(288-256)+rcx],xmm2
+	movdqa	XMMWORD[(304-256)+rcx],xmm3
+
+	jmp	NEAR $L$oop_enter4x
+
+ALIGN	32
+$L$oop_outer4x:
+	movdqa	xmm8,XMMWORD[64+rsp]
+	movdqa	xmm9,XMMWORD[80+rsp]
+	movdqa	xmm10,XMMWORD[96+rsp]
+	movdqa	xmm11,XMMWORD[112+rsp]
+	movdqa	xmm12,XMMWORD[((128-256))+rcx]
+	movdqa	xmm13,XMMWORD[((144-256))+rcx]
+	movdqa	xmm14,XMMWORD[((160-256))+rcx]
+	movdqa	xmm15,XMMWORD[((176-256))+rcx]
+	movdqa	xmm4,XMMWORD[((192-256))+rcx]
+	movdqa	xmm5,XMMWORD[((208-256))+rcx]
+	movdqa	xmm6,XMMWORD[((224-256))+rcx]
+	movdqa	xmm7,XMMWORD[((240-256))+rcx]
+	movdqa	xmm0,XMMWORD[((256-256))+rcx]
+	movdqa	xmm1,XMMWORD[((272-256))+rcx]
+	movdqa	xmm2,XMMWORD[((288-256))+rcx]
+	movdqa	xmm3,XMMWORD[((304-256))+rcx]
+	paddd	xmm0,XMMWORD[$L$four]
+
+$L$oop_enter4x:
+	movdqa	XMMWORD[32+rsp],xmm6
+	movdqa	XMMWORD[48+rsp],xmm7
+	movdqa	xmm7,XMMWORD[r10]
+	mov	eax,10
+	movdqa	XMMWORD[(256-256)+rcx],xmm0
+	jmp	NEAR $L$oop4x
+
+ALIGN	32
+$L$oop4x:
+	paddd	xmm8,xmm12
+	paddd	xmm9,xmm13
+	pxor	xmm0,xmm8
+	pxor	xmm1,xmm9
+DB	102,15,56,0,199
+DB	102,15,56,0,207
+	paddd	xmm4,xmm0
+	paddd	xmm5,xmm1
+	pxor	xmm12,xmm4
+	pxor	xmm13,xmm5
+	movdqa	xmm6,xmm12
+	pslld	xmm12,12
+	psrld	xmm6,20
+	movdqa	xmm7,xmm13
+	pslld	xmm13,12
+	por	xmm12,xmm6
+	psrld	xmm7,20
+	movdqa	xmm6,XMMWORD[r11]
+	por	xmm13,xmm7
+	paddd	xmm8,xmm12
+	paddd	xmm9,xmm13
+	pxor	xmm0,xmm8
+	pxor	xmm1,xmm9
+DB	102,15,56,0,198
+DB	102,15,56,0,206
+	paddd	xmm4,xmm0
+	paddd	xmm5,xmm1
+	pxor	xmm12,xmm4
+	pxor	xmm13,xmm5
+	movdqa	xmm7,xmm12
+	pslld	xmm12,7
+	psrld	xmm7,25
+	movdqa	xmm6,xmm13
+	pslld	xmm13,7
+	por	xmm12,xmm7
+	psrld	xmm6,25
+	movdqa	xmm7,XMMWORD[r10]
+	por	xmm13,xmm6
+	movdqa	XMMWORD[rsp],xmm4
+	movdqa	XMMWORD[16+rsp],xmm5
+	movdqa	xmm4,XMMWORD[32+rsp]
+	movdqa	xmm5,XMMWORD[48+rsp]
+	paddd	xmm10,xmm14
+	paddd	xmm11,xmm15
+	pxor	xmm2,xmm10
+	pxor	xmm3,xmm11
+DB	102,15,56,0,215
+DB	102,15,56,0,223
+	paddd	xmm4,xmm2
+	paddd	xmm5,xmm3
+	pxor	xmm14,xmm4
+	pxor	xmm15,xmm5
+	movdqa	xmm6,xmm14
+	pslld	xmm14,12
+	psrld	xmm6,20
+	movdqa	xmm7,xmm15
+	pslld	xmm15,12
+	por	xmm14,xmm6
+	psrld	xmm7,20
+	movdqa	xmm6,XMMWORD[r11]
+	por	xmm15,xmm7
+	paddd	xmm10,xmm14
+	paddd	xmm11,xmm15
+	pxor	xmm2,xmm10
+	pxor	xmm3,xmm11
+DB	102,15,56,0,214
+DB	102,15,56,0,222
+	paddd	xmm4,xmm2
+	paddd	xmm5,xmm3
+	pxor	xmm14,xmm4
+	pxor	xmm15,xmm5
+	movdqa	xmm7,xmm14
+	pslld	xmm14,7
+	psrld	xmm7,25
+	movdqa	xmm6,xmm15
+	pslld	xmm15,7
+	por	xmm14,xmm7
+	psrld	xmm6,25
+	movdqa	xmm7,XMMWORD[r10]
+	por	xmm15,xmm6
+	paddd	xmm8,xmm13
+	paddd	xmm9,xmm14
+	pxor	xmm3,xmm8
+	pxor	xmm0,xmm9
+DB	102,15,56,0,223
+DB	102,15,56,0,199
+	paddd	xmm4,xmm3
+	paddd	xmm5,xmm0
+	pxor	xmm13,xmm4
+	pxor	xmm14,xmm5
+	movdqa	xmm6,xmm13
+	pslld	xmm13,12
+	psrld	xmm6,20
+	movdqa	xmm7,xmm14
+	pslld	xmm14,12
+	por	xmm13,xmm6
+	psrld	xmm7,20
+	movdqa	xmm6,XMMWORD[r11]
+	por	xmm14,xmm7
+	paddd	xmm8,xmm13
+	paddd	xmm9,xmm14
+	pxor	xmm3,xmm8
+	pxor	xmm0,xmm9
+DB	102,15,56,0,222
+DB	102,15,56,0,198
+	paddd	xmm4,xmm3
+	paddd	xmm5,xmm0
+	pxor	xmm13,xmm4
+	pxor	xmm14,xmm5
+	movdqa	xmm7,xmm13
+	pslld	xmm13,7
+	psrld	xmm7,25
+	movdqa	xmm6,xmm14
+	pslld	xmm14,7
+	por	xmm13,xmm7
+	psrld	xmm6,25
+	movdqa	xmm7,XMMWORD[r10]
+	por	xmm14,xmm6
+	movdqa	XMMWORD[32+rsp],xmm4
+	movdqa	XMMWORD[48+rsp],xmm5
+	movdqa	xmm4,XMMWORD[rsp]
+	movdqa	xmm5,XMMWORD[16+rsp]
+	paddd	xmm10,xmm15
+	paddd	xmm11,xmm12
+	pxor	xmm1,xmm10
+	pxor	xmm2,xmm11
+DB	102,15,56,0,207
+DB	102,15,56,0,215
+	paddd	xmm4,xmm1
+	paddd	xmm5,xmm2
+	pxor	xmm15,xmm4
+	pxor	xmm12,xmm5
+	movdqa	xmm6,xmm15
+	pslld	xmm15,12
+	psrld	xmm6,20
+	movdqa	xmm7,xmm12
+	pslld	xmm12,12
+	por	xmm15,xmm6
+	psrld	xmm7,20
+	movdqa	xmm6,XMMWORD[r11]
+	por	xmm12,xmm7
+	paddd	xmm10,xmm15
+	paddd	xmm11,xmm12
+	pxor	xmm1,xmm10
+	pxor	xmm2,xmm11
+DB	102,15,56,0,206
+DB	102,15,56,0,214
+	paddd	xmm4,xmm1
+	paddd	xmm5,xmm2
+	pxor	xmm15,xmm4
+	pxor	xmm12,xmm5
+	movdqa	xmm7,xmm15
+	pslld	xmm15,7
+	psrld	xmm7,25
+	movdqa	xmm6,xmm12
+	pslld	xmm12,7
+	por	xmm15,xmm7
+	psrld	xmm6,25
+	movdqa	xmm7,XMMWORD[r10]
+	por	xmm12,xmm6
+	dec	eax
+	jnz	NEAR $L$oop4x
+
+	paddd	xmm8,XMMWORD[64+rsp]
+	paddd	xmm9,XMMWORD[80+rsp]
+	paddd	xmm10,XMMWORD[96+rsp]
+	paddd	xmm11,XMMWORD[112+rsp]
+
+	movdqa	xmm6,xmm8
+	punpckldq	xmm8,xmm9
+	movdqa	xmm7,xmm10
+	punpckldq	xmm10,xmm11
+	punpckhdq	xmm6,xmm9
+	punpckhdq	xmm7,xmm11
+	movdqa	xmm9,xmm8
+	punpcklqdq	xmm8,xmm10
+	movdqa	xmm11,xmm6
+	punpcklqdq	xmm6,xmm7
+	punpckhqdq	xmm9,xmm10
+	punpckhqdq	xmm11,xmm7
+	paddd	xmm12,XMMWORD[((128-256))+rcx]
+	paddd	xmm13,XMMWORD[((144-256))+rcx]
+	paddd	xmm14,XMMWORD[((160-256))+rcx]
+	paddd	xmm15,XMMWORD[((176-256))+rcx]
+
+	movdqa	XMMWORD[rsp],xmm8
+	movdqa	XMMWORD[16+rsp],xmm9
+	movdqa	xmm8,XMMWORD[32+rsp]
+	movdqa	xmm9,XMMWORD[48+rsp]
+
+	movdqa	xmm10,xmm12
+	punpckldq	xmm12,xmm13
+	movdqa	xmm7,xmm14
+	punpckldq	xmm14,xmm15
+	punpckhdq	xmm10,xmm13
+	punpckhdq	xmm7,xmm15
+	movdqa	xmm13,xmm12
+	punpcklqdq	xmm12,xmm14
+	movdqa	xmm15,xmm10
+	punpcklqdq	xmm10,xmm7
+	punpckhqdq	xmm13,xmm14
+	punpckhqdq	xmm15,xmm7
+	paddd	xmm4,XMMWORD[((192-256))+rcx]
+	paddd	xmm5,XMMWORD[((208-256))+rcx]
+	paddd	xmm8,XMMWORD[((224-256))+rcx]
+	paddd	xmm9,XMMWORD[((240-256))+rcx]
+
+	movdqa	XMMWORD[32+rsp],xmm6
+	movdqa	XMMWORD[48+rsp],xmm11
+
+	movdqa	xmm14,xmm4
+	punpckldq	xmm4,xmm5
+	movdqa	xmm7,xmm8
+	punpckldq	xmm8,xmm9
+	punpckhdq	xmm14,xmm5
+	punpckhdq	xmm7,xmm9
+	movdqa	xmm5,xmm4
+	punpcklqdq	xmm4,xmm8
+	movdqa	xmm9,xmm14
+	punpcklqdq	xmm14,xmm7
+	punpckhqdq	xmm5,xmm8
+	punpckhqdq	xmm9,xmm7
+	paddd	xmm0,XMMWORD[((256-256))+rcx]
+	paddd	xmm1,XMMWORD[((272-256))+rcx]
+	paddd	xmm2,XMMWORD[((288-256))+rcx]
+	paddd	xmm3,XMMWORD[((304-256))+rcx]
+
+	movdqa	xmm8,xmm0
+	punpckldq	xmm0,xmm1
+	movdqa	xmm7,xmm2
+	punpckldq	xmm2,xmm3
+	punpckhdq	xmm8,xmm1
+	punpckhdq	xmm7,xmm3
+	movdqa	xmm1,xmm0
+	punpcklqdq	xmm0,xmm2
+	movdqa	xmm3,xmm8
+	punpcklqdq	xmm8,xmm7
+	punpckhqdq	xmm1,xmm2
+	punpckhqdq	xmm3,xmm7
+	cmp	rdx,64*4
+	jb	NEAR $L$tail4x
+
+	movdqu	xmm6,XMMWORD[rsi]
+	movdqu	xmm11,XMMWORD[16+rsi]
+	movdqu	xmm2,XMMWORD[32+rsi]
+	movdqu	xmm7,XMMWORD[48+rsi]
+	pxor	xmm6,XMMWORD[rsp]
+	pxor	xmm11,xmm12
+	pxor	xmm2,xmm4
+	pxor	xmm7,xmm0
+
+	movdqu	XMMWORD[rdi],xmm6
+	movdqu	xmm6,XMMWORD[64+rsi]
+	movdqu	XMMWORD[16+rdi],xmm11
+	movdqu	xmm11,XMMWORD[80+rsi]
+	movdqu	XMMWORD[32+rdi],xmm2
+	movdqu	xmm2,XMMWORD[96+rsi]
+	movdqu	XMMWORD[48+rdi],xmm7
+	movdqu	xmm7,XMMWORD[112+rsi]
+	lea	rsi,[128+rsi]
+	pxor	xmm6,XMMWORD[16+rsp]
+	pxor	xmm11,xmm13
+	pxor	xmm2,xmm5
+	pxor	xmm7,xmm1
+
+	movdqu	XMMWORD[64+rdi],xmm6
+	movdqu	xmm6,XMMWORD[rsi]
+	movdqu	XMMWORD[80+rdi],xmm11
+	movdqu	xmm11,XMMWORD[16+rsi]
+	movdqu	XMMWORD[96+rdi],xmm2
+	movdqu	xmm2,XMMWORD[32+rsi]
+	movdqu	XMMWORD[112+rdi],xmm7
+	lea	rdi,[128+rdi]
+	movdqu	xmm7,XMMWORD[48+rsi]
+	pxor	xmm6,XMMWORD[32+rsp]
+	pxor	xmm11,xmm10
+	pxor	xmm2,xmm14
+	pxor	xmm7,xmm8
+
+	movdqu	XMMWORD[rdi],xmm6
+	movdqu	xmm6,XMMWORD[64+rsi]
+	movdqu	XMMWORD[16+rdi],xmm11
+	movdqu	xmm11,XMMWORD[80+rsi]
+	movdqu	XMMWORD[32+rdi],xmm2
+	movdqu	xmm2,XMMWORD[96+rsi]
+	movdqu	XMMWORD[48+rdi],xmm7
+	movdqu	xmm7,XMMWORD[112+rsi]
+	lea	rsi,[128+rsi]
+	pxor	xmm6,XMMWORD[48+rsp]
+	pxor	xmm11,xmm15
+	pxor	xmm2,xmm9
+	pxor	xmm7,xmm3
+	movdqu	XMMWORD[64+rdi],xmm6
+	movdqu	XMMWORD[80+rdi],xmm11
+	movdqu	XMMWORD[96+rdi],xmm2
+	movdqu	XMMWORD[112+rdi],xmm7
+	lea	rdi,[128+rdi]
+
+	sub	rdx,64*4
+	jnz	NEAR $L$oop_outer4x
+
+	jmp	NEAR $L$done4x
+
+$L$tail4x:
+	cmp	rdx,192
+	jae	NEAR $L$192_or_more4x
+	cmp	rdx,128
+	jae	NEAR $L$128_or_more4x
+	cmp	rdx,64
+	jae	NEAR $L$64_or_more4x
+
+
+	xor	r10,r10
+
+	movdqa	XMMWORD[16+rsp],xmm12
+	movdqa	XMMWORD[32+rsp],xmm4
+	movdqa	XMMWORD[48+rsp],xmm0
+	jmp	NEAR $L$oop_tail4x
+
+ALIGN	32
+$L$64_or_more4x:
+	movdqu	xmm6,XMMWORD[rsi]
+	movdqu	xmm11,XMMWORD[16+rsi]
+	movdqu	xmm2,XMMWORD[32+rsi]
+	movdqu	xmm7,XMMWORD[48+rsi]
+	pxor	xmm6,XMMWORD[rsp]
+	pxor	xmm11,xmm12
+	pxor	xmm2,xmm4
+	pxor	xmm7,xmm0
+	movdqu	XMMWORD[rdi],xmm6
+	movdqu	XMMWORD[16+rdi],xmm11
+	movdqu	XMMWORD[32+rdi],xmm2
+	movdqu	XMMWORD[48+rdi],xmm7
+	je	NEAR $L$done4x
+
+	movdqa	xmm6,XMMWORD[16+rsp]
+	lea	rsi,[64+rsi]
+	xor	r10,r10
+	movdqa	XMMWORD[rsp],xmm6
+	movdqa	XMMWORD[16+rsp],xmm13
+	lea	rdi,[64+rdi]
+	movdqa	XMMWORD[32+rsp],xmm5
+	sub	rdx,64
+	movdqa	XMMWORD[48+rsp],xmm1
+	jmp	NEAR $L$oop_tail4x
+
+ALIGN	32
+$L$128_or_more4x:
+	movdqu	xmm6,XMMWORD[rsi]
+	movdqu	xmm11,XMMWORD[16+rsi]
+	movdqu	xmm2,XMMWORD[32+rsi]
+	movdqu	xmm7,XMMWORD[48+rsi]
+	pxor	xmm6,XMMWORD[rsp]
+	pxor	xmm11,xmm12
+	pxor	xmm2,xmm4
+	pxor	xmm7,xmm0
+
+	movdqu	XMMWORD[rdi],xmm6
+	movdqu	xmm6,XMMWORD[64+rsi]
+	movdqu	XMMWORD[16+rdi],xmm11
+	movdqu	xmm11,XMMWORD[80+rsi]
+	movdqu	XMMWORD[32+rdi],xmm2
+	movdqu	xmm2,XMMWORD[96+rsi]
+	movdqu	XMMWORD[48+rdi],xmm7
+	movdqu	xmm7,XMMWORD[112+rsi]
+	pxor	xmm6,XMMWORD[16+rsp]
+	pxor	xmm11,xmm13
+	pxor	xmm2,xmm5
+	pxor	xmm7,xmm1
+	movdqu	XMMWORD[64+rdi],xmm6
+	movdqu	XMMWORD[80+rdi],xmm11
+	movdqu	XMMWORD[96+rdi],xmm2
+	movdqu	XMMWORD[112+rdi],xmm7
+	je	NEAR $L$done4x
+
+	movdqa	xmm6,XMMWORD[32+rsp]
+	lea	rsi,[128+rsi]
+	xor	r10,r10
+	movdqa	XMMWORD[rsp],xmm6
+	movdqa	XMMWORD[16+rsp],xmm10
+	lea	rdi,[128+rdi]
+	movdqa	XMMWORD[32+rsp],xmm14
+	sub	rdx,128
+	movdqa	XMMWORD[48+rsp],xmm8
+	jmp	NEAR $L$oop_tail4x
+
+ALIGN	32
+$L$192_or_more4x:
+	movdqu	xmm6,XMMWORD[rsi]
+	movdqu	xmm11,XMMWORD[16+rsi]
+	movdqu	xmm2,XMMWORD[32+rsi]
+	movdqu	xmm7,XMMWORD[48+rsi]
+	pxor	xmm6,XMMWORD[rsp]
+	pxor	xmm11,xmm12
+	pxor	xmm2,xmm4
+	pxor	xmm7,xmm0
+
+	movdqu	XMMWORD[rdi],xmm6
+	movdqu	xmm6,XMMWORD[64+rsi]
+	movdqu	XMMWORD[16+rdi],xmm11
+	movdqu	xmm11,XMMWORD[80+rsi]
+	movdqu	XMMWORD[32+rdi],xmm2
+	movdqu	xmm2,XMMWORD[96+rsi]
+	movdqu	XMMWORD[48+rdi],xmm7
+	movdqu	xmm7,XMMWORD[112+rsi]
+	lea	rsi,[128+rsi]
+	pxor	xmm6,XMMWORD[16+rsp]
+	pxor	xmm11,xmm13
+	pxor	xmm2,xmm5
+	pxor	xmm7,xmm1
+
+	movdqu	XMMWORD[64+rdi],xmm6
+	movdqu	xmm6,XMMWORD[rsi]
+	movdqu	XMMWORD[80+rdi],xmm11
+	movdqu	xmm11,XMMWORD[16+rsi]
+	movdqu	XMMWORD[96+rdi],xmm2
+	movdqu	xmm2,XMMWORD[32+rsi]
+	movdqu	XMMWORD[112+rdi],xmm7
+	lea	rdi,[128+rdi]
+	movdqu	xmm7,XMMWORD[48+rsi]
+	pxor	xmm6,XMMWORD[32+rsp]
+	pxor	xmm11,xmm10
+	pxor	xmm2,xmm14
+	pxor	xmm7,xmm8
+	movdqu	XMMWORD[rdi],xmm6
+	movdqu	XMMWORD[16+rdi],xmm11
+	movdqu	XMMWORD[32+rdi],xmm2
+	movdqu	XMMWORD[48+rdi],xmm7
+	je	NEAR $L$done4x
+
+	movdqa	xmm6,XMMWORD[48+rsp]
+	lea	rsi,[64+rsi]
+	xor	r10,r10
+	movdqa	XMMWORD[rsp],xmm6
+	movdqa	XMMWORD[16+rsp],xmm15
+	lea	rdi,[64+rdi]
+	movdqa	XMMWORD[32+rsp],xmm9
+	sub	rdx,192
+	movdqa	XMMWORD[48+rsp],xmm3
+
+$L$oop_tail4x:
+	movzx	eax,BYTE[r10*1+rsi]
+	movzx	ecx,BYTE[r10*1+rsp]
+	lea	r10,[1+r10]
+	xor	eax,ecx
+	mov	BYTE[((-1))+r10*1+rdi],al
+	dec	rdx
+	jnz	NEAR $L$oop_tail4x
+
+$L$done4x:
+	lea	r11,[((320+48))+rsp]
+	movaps	xmm6,XMMWORD[((-48))+r11]
+	movaps	xmm7,XMMWORD[((-32))+r11]
+	movaps	xmm8,XMMWORD[((-16))+r11]
+	movaps	xmm9,XMMWORD[r11]
+	movaps	xmm10,XMMWORD[16+r11]
+	movaps	xmm11,XMMWORD[32+r11]
+	movaps	xmm12,XMMWORD[48+r11]
+	movaps	xmm13,XMMWORD[64+r11]
+	movaps	xmm14,XMMWORD[80+r11]
+	movaps	xmm15,XMMWORD[96+r11]
+	add	rsp,0x148+160
+	mov	rdi,QWORD[8+rsp]	;WIN64 epilogue
+	mov	rsi,QWORD[16+rsp]
+	DB	0F3h,0C3h		;repret
+$L$SEH_end_ChaCha20_4x:
+
+ALIGN	32
+ChaCha20_8x:
+	mov	QWORD[8+rsp],rdi	;WIN64 prologue
+	mov	QWORD[16+rsp],rsi
+	mov	rax,rsp
+$L$SEH_begin_ChaCha20_8x:
+	mov	rdi,rcx
+	mov	rsi,rdx
+	mov	rdx,r8
+	mov	rcx,r9
+	mov	r8,QWORD[40+rsp]
+
+
+$L$ChaCha20_8x:
+	mov	r10,rsp
+	sub	rsp,0x280+176
+	and	rsp,-32
+	lea	r11,[((656+48))+rsp]
+	movaps	XMMWORD[(-48)+r11],xmm6
+	movaps	XMMWORD[(-32)+r11],xmm7
+	movaps	XMMWORD[(-16)+r11],xmm8
+	movaps	XMMWORD[r11],xmm9
+	movaps	XMMWORD[16+r11],xmm10
+	movaps	XMMWORD[32+r11],xmm11
+	movaps	XMMWORD[48+r11],xmm12
+	movaps	XMMWORD[64+r11],xmm13
+	movaps	XMMWORD[80+r11],xmm14
+	movaps	XMMWORD[96+r11],xmm15
+	vzeroupper
+	mov	QWORD[640+rsp],r10
+
+
+
+
+
+
+
+
+
+
+	vbroadcasti128	ymm11,XMMWORD[$L$sigma]
+	vbroadcasti128	ymm3,XMMWORD[rcx]
+	vbroadcasti128	ymm15,XMMWORD[16+rcx]
+	vbroadcasti128	ymm7,XMMWORD[r8]
+	lea	rcx,[256+rsp]
+	lea	rax,[512+rsp]
+	lea	r10,[$L$rot16]
+	lea	r11,[$L$rot24]
+
+	vpshufd	ymm8,ymm11,0x00
+	vpshufd	ymm9,ymm11,0x55
+	vmovdqa	YMMWORD[(128-256)+rcx],ymm8
+	vpshufd	ymm10,ymm11,0xaa
+	vmovdqa	YMMWORD[(160-256)+rcx],ymm9
+	vpshufd	ymm11,ymm11,0xff
+	vmovdqa	YMMWORD[(192-256)+rcx],ymm10
+	vmovdqa	YMMWORD[(224-256)+rcx],ymm11
+
+	vpshufd	ymm0,ymm3,0x00
+	vpshufd	ymm1,ymm3,0x55
+	vmovdqa	YMMWORD[(256-256)+rcx],ymm0
+	vpshufd	ymm2,ymm3,0xaa
+	vmovdqa	YMMWORD[(288-256)+rcx],ymm1
+	vpshufd	ymm3,ymm3,0xff
+	vmovdqa	YMMWORD[(320-256)+rcx],ymm2
+	vmovdqa	YMMWORD[(352-256)+rcx],ymm3
+
+	vpshufd	ymm12,ymm15,0x00
+	vpshufd	ymm13,ymm15,0x55
+	vmovdqa	YMMWORD[(384-512)+rax],ymm12
+	vpshufd	ymm14,ymm15,0xaa
+	vmovdqa	YMMWORD[(416-512)+rax],ymm13
+	vpshufd	ymm15,ymm15,0xff
+	vmovdqa	YMMWORD[(448-512)+rax],ymm14
+	vmovdqa	YMMWORD[(480-512)+rax],ymm15
+
+	vpshufd	ymm4,ymm7,0x00
+	vpshufd	ymm5,ymm7,0x55
+	vpaddd	ymm4,ymm4,YMMWORD[$L$incy]
+	vpshufd	ymm6,ymm7,0xaa
+	vmovdqa	YMMWORD[(544-512)+rax],ymm5
+	vpshufd	ymm7,ymm7,0xff
+	vmovdqa	YMMWORD[(576-512)+rax],ymm6
+	vmovdqa	YMMWORD[(608-512)+rax],ymm7
+
+	jmp	NEAR $L$oop_enter8x
+
+ALIGN	32
+$L$oop_outer8x:
+	vmovdqa	ymm8,YMMWORD[((128-256))+rcx]
+	vmovdqa	ymm9,YMMWORD[((160-256))+rcx]
+	vmovdqa	ymm10,YMMWORD[((192-256))+rcx]
+	vmovdqa	ymm11,YMMWORD[((224-256))+rcx]
+	vmovdqa	ymm0,YMMWORD[((256-256))+rcx]
+	vmovdqa	ymm1,YMMWORD[((288-256))+rcx]
+	vmovdqa	ymm2,YMMWORD[((320-256))+rcx]
+	vmovdqa	ymm3,YMMWORD[((352-256))+rcx]
+	vmovdqa	ymm12,YMMWORD[((384-512))+rax]
+	vmovdqa	ymm13,YMMWORD[((416-512))+rax]
+	vmovdqa	ymm14,YMMWORD[((448-512))+rax]
+	vmovdqa	ymm15,YMMWORD[((480-512))+rax]
+	vmovdqa	ymm4,YMMWORD[((512-512))+rax]
+	vmovdqa	ymm5,YMMWORD[((544-512))+rax]
+	vmovdqa	ymm6,YMMWORD[((576-512))+rax]
+	vmovdqa	ymm7,YMMWORD[((608-512))+rax]
+	vpaddd	ymm4,ymm4,YMMWORD[$L$eight]
+
+$L$oop_enter8x:
+	vmovdqa	YMMWORD[64+rsp],ymm14
+	vmovdqa	YMMWORD[96+rsp],ymm15
+	vbroadcasti128	ymm15,XMMWORD[r10]
+	vmovdqa	YMMWORD[(512-512)+rax],ymm4
+	mov	eax,10
+	jmp	NEAR $L$oop8x
+
+ALIGN	32
+$L$oop8x:
+	vpaddd	ymm8,ymm8,ymm0
+	vpxor	ymm4,ymm8,ymm4
+	vpshufb	ymm4,ymm4,ymm15
+	vpaddd	ymm9,ymm9,ymm1
+	vpxor	ymm5,ymm9,ymm5
+	vpshufb	ymm5,ymm5,ymm15
+	vpaddd	ymm12,ymm12,ymm4
+	vpxor	ymm0,ymm12,ymm0
+	vpslld	ymm14,ymm0,12
+	vpsrld	ymm0,ymm0,20
+	vpor	ymm0,ymm14,ymm0
+	vbroadcasti128	ymm14,XMMWORD[r11]
+	vpaddd	ymm13,ymm13,ymm5
+	vpxor	ymm1,ymm13,ymm1
+	vpslld	ymm15,ymm1,12
+	vpsrld	ymm1,ymm1,20
+	vpor	ymm1,ymm15,ymm1
+	vpaddd	ymm8,ymm8,ymm0
+	vpxor	ymm4,ymm8,ymm4
+	vpshufb	ymm4,ymm4,ymm14
+	vpaddd	ymm9,ymm9,ymm1
+	vpxor	ymm5,ymm9,ymm5
+	vpshufb	ymm5,ymm5,ymm14
+	vpaddd	ymm12,ymm12,ymm4
+	vpxor	ymm0,ymm12,ymm0
+	vpslld	ymm15,ymm0,7
+	vpsrld	ymm0,ymm0,25
+	vpor	ymm0,ymm15,ymm0
+	vbroadcasti128	ymm15,XMMWORD[r10]
+	vpaddd	ymm13,ymm13,ymm5
+	vpxor	ymm1,ymm13,ymm1
+	vpslld	ymm14,ymm1,7
+	vpsrld	ymm1,ymm1,25
+	vpor	ymm1,ymm14,ymm1
+	vmovdqa	YMMWORD[rsp],ymm12
+	vmovdqa	YMMWORD[32+rsp],ymm13
+	vmovdqa	ymm12,YMMWORD[64+rsp]
+	vmovdqa	ymm13,YMMWORD[96+rsp]
+	vpaddd	ymm10,ymm10,ymm2
+	vpxor	ymm6,ymm10,ymm6
+	vpshufb	ymm6,ymm6,ymm15
+	vpaddd	ymm11,ymm11,ymm3
+	vpxor	ymm7,ymm11,ymm7
+	vpshufb	ymm7,ymm7,ymm15
+	vpaddd	ymm12,ymm12,ymm6
+	vpxor	ymm2,ymm12,ymm2
+	vpslld	ymm14,ymm2,12
+	vpsrld	ymm2,ymm2,20
+	vpor	ymm2,ymm14,ymm2
+	vbroadcasti128	ymm14,XMMWORD[r11]
+	vpaddd	ymm13,ymm13,ymm7
+	vpxor	ymm3,ymm13,ymm3
+	vpslld	ymm15,ymm3,12
+	vpsrld	ymm3,ymm3,20
+	vpor	ymm3,ymm15,ymm3
+	vpaddd	ymm10,ymm10,ymm2
+	vpxor	ymm6,ymm10,ymm6
+	vpshufb	ymm6,ymm6,ymm14
+	vpaddd	ymm11,ymm11,ymm3
+	vpxor	ymm7,ymm11,ymm7
+	vpshufb	ymm7,ymm7,ymm14
+	vpaddd	ymm12,ymm12,ymm6
+	vpxor	ymm2,ymm12,ymm2
+	vpslld	ymm15,ymm2,7
+	vpsrld	ymm2,ymm2,25
+	vpor	ymm2,ymm15,ymm2
+	vbroadcasti128	ymm15,XMMWORD[r10]
+	vpaddd	ymm13,ymm13,ymm7
+	vpxor	ymm3,ymm13,ymm3
+	vpslld	ymm14,ymm3,7
+	vpsrld	ymm3,ymm3,25
+	vpor	ymm3,ymm14,ymm3
+	vpaddd	ymm8,ymm8,ymm1
+	vpxor	ymm7,ymm8,ymm7
+	vpshufb	ymm7,ymm7,ymm15
+	vpaddd	ymm9,ymm9,ymm2
+	vpxor	ymm4,ymm9,ymm4
+	vpshufb	ymm4,ymm4,ymm15
+	vpaddd	ymm12,ymm12,ymm7
+	vpxor	ymm1,ymm12,ymm1
+	vpslld	ymm14,ymm1,12
+	vpsrld	ymm1,ymm1,20
+	vpor	ymm1,ymm14,ymm1
+	vbroadcasti128	ymm14,XMMWORD[r11]
+	vpaddd	ymm13,ymm13,ymm4
+	vpxor	ymm2,ymm13,ymm2
+	vpslld	ymm15,ymm2,12
+	vpsrld	ymm2,ymm2,20
+	vpor	ymm2,ymm15,ymm2
+	vpaddd	ymm8,ymm8,ymm1
+	vpxor	ymm7,ymm8,ymm7
+	vpshufb	ymm7,ymm7,ymm14
+	vpaddd	ymm9,ymm9,ymm2
+	vpxor	ymm4,ymm9,ymm4
+	vpshufb	ymm4,ymm4,ymm14
+	vpaddd	ymm12,ymm12,ymm7
+	vpxor	ymm1,ymm12,ymm1
+	vpslld	ymm15,ymm1,7
+	vpsrld	ymm1,ymm1,25
+	vpor	ymm1,ymm15,ymm1
+	vbroadcasti128	ymm15,XMMWORD[r10]
+	vpaddd	ymm13,ymm13,ymm4
+	vpxor	ymm2,ymm13,ymm2
+	vpslld	ymm14,ymm2,7
+	vpsrld	ymm2,ymm2,25
+	vpor	ymm2,ymm14,ymm2
+	vmovdqa	YMMWORD[64+rsp],ymm12
+	vmovdqa	YMMWORD[96+rsp],ymm13
+	vmovdqa	ymm12,YMMWORD[rsp]
+	vmovdqa	ymm13,YMMWORD[32+rsp]
+	vpaddd	ymm10,ymm10,ymm3
+	vpxor	ymm5,ymm10,ymm5
+	vpshufb	ymm5,ymm5,ymm15
+	vpaddd	ymm11,ymm11,ymm0
+	vpxor	ymm6,ymm11,ymm6
+	vpshufb	ymm6,ymm6,ymm15
+	vpaddd	ymm12,ymm12,ymm5
+	vpxor	ymm3,ymm12,ymm3
+	vpslld	ymm14,ymm3,12
+	vpsrld	ymm3,ymm3,20
+	vpor	ymm3,ymm14,ymm3
+	vbroadcasti128	ymm14,XMMWORD[r11]
+	vpaddd	ymm13,ymm13,ymm6
+	vpxor	ymm0,ymm13,ymm0
+	vpslld	ymm15,ymm0,12
+	vpsrld	ymm0,ymm0,20
+	vpor	ymm0,ymm15,ymm0
+	vpaddd	ymm10,ymm10,ymm3
+	vpxor	ymm5,ymm10,ymm5
+	vpshufb	ymm5,ymm5,ymm14
+	vpaddd	ymm11,ymm11,ymm0
+	vpxor	ymm6,ymm11,ymm6
+	vpshufb	ymm6,ymm6,ymm14
+	vpaddd	ymm12,ymm12,ymm5
+	vpxor	ymm3,ymm12,ymm3
+	vpslld	ymm15,ymm3,7
+	vpsrld	ymm3,ymm3,25
+	vpor	ymm3,ymm15,ymm3
+	vbroadcasti128	ymm15,XMMWORD[r10]
+	vpaddd	ymm13,ymm13,ymm6
+	vpxor	ymm0,ymm13,ymm0
+	vpslld	ymm14,ymm0,7
+	vpsrld	ymm0,ymm0,25
+	vpor	ymm0,ymm14,ymm0
+	dec	eax
+	jnz	NEAR $L$oop8x
+
+	lea	rax,[512+rsp]
+	vpaddd	ymm8,ymm8,YMMWORD[((128-256))+rcx]
+	vpaddd	ymm9,ymm9,YMMWORD[((160-256))+rcx]
+	vpaddd	ymm10,ymm10,YMMWORD[((192-256))+rcx]
+	vpaddd	ymm11,ymm11,YMMWORD[((224-256))+rcx]
+
+	vpunpckldq	ymm14,ymm8,ymm9
+	vpunpckldq	ymm15,ymm10,ymm11
+	vpunpckhdq	ymm8,ymm8,ymm9
+	vpunpckhdq	ymm10,ymm10,ymm11
+	vpunpcklqdq	ymm9,ymm14,ymm15
+	vpunpckhqdq	ymm14,ymm14,ymm15
+	vpunpcklqdq	ymm11,ymm8,ymm10
+	vpunpckhqdq	ymm8,ymm8,ymm10
+	vpaddd	ymm0,ymm0,YMMWORD[((256-256))+rcx]
+	vpaddd	ymm1,ymm1,YMMWORD[((288-256))+rcx]
+	vpaddd	ymm2,ymm2,YMMWORD[((320-256))+rcx]
+	vpaddd	ymm3,ymm3,YMMWORD[((352-256))+rcx]
+
+	vpunpckldq	ymm10,ymm0,ymm1
+	vpunpckldq	ymm15,ymm2,ymm3
+	vpunpckhdq	ymm0,ymm0,ymm1
+	vpunpckhdq	ymm2,ymm2,ymm3
+	vpunpcklqdq	ymm1,ymm10,ymm15
+	vpunpckhqdq	ymm10,ymm10,ymm15
+	vpunpcklqdq	ymm3,ymm0,ymm2
+	vpunpckhqdq	ymm0,ymm0,ymm2
+	vperm2i128	ymm15,ymm9,ymm1,0x20
+	vperm2i128	ymm1,ymm9,ymm1,0x31
+	vperm2i128	ymm9,ymm14,ymm10,0x20
+	vperm2i128	ymm10,ymm14,ymm10,0x31
+	vperm2i128	ymm14,ymm11,ymm3,0x20
+	vperm2i128	ymm3,ymm11,ymm3,0x31
+	vperm2i128	ymm11,ymm8,ymm0,0x20
+	vperm2i128	ymm0,ymm8,ymm0,0x31
+	vmovdqa	YMMWORD[rsp],ymm15
+	vmovdqa	YMMWORD[32+rsp],ymm9
+	vmovdqa	ymm15,YMMWORD[64+rsp]
+	vmovdqa	ymm9,YMMWORD[96+rsp]
+
+	vpaddd	ymm12,ymm12,YMMWORD[((384-512))+rax]
+	vpaddd	ymm13,ymm13,YMMWORD[((416-512))+rax]
+	vpaddd	ymm15,ymm15,YMMWORD[((448-512))+rax]
+	vpaddd	ymm9,ymm9,YMMWORD[((480-512))+rax]
+
+	vpunpckldq	ymm2,ymm12,ymm13
+	vpunpckldq	ymm8,ymm15,ymm9
+	vpunpckhdq	ymm12,ymm12,ymm13
+	vpunpckhdq	ymm15,ymm15,ymm9
+	vpunpcklqdq	ymm13,ymm2,ymm8
+	vpunpckhqdq	ymm2,ymm2,ymm8
+	vpunpcklqdq	ymm9,ymm12,ymm15
+	vpunpckhqdq	ymm12,ymm12,ymm15
+	vpaddd	ymm4,ymm4,YMMWORD[((512-512))+rax]
+	vpaddd	ymm5,ymm5,YMMWORD[((544-512))+rax]
+	vpaddd	ymm6,ymm6,YMMWORD[((576-512))+rax]
+	vpaddd	ymm7,ymm7,YMMWORD[((608-512))+rax]
+
+	vpunpckldq	ymm15,ymm4,ymm5
+	vpunpckldq	ymm8,ymm6,ymm7
+	vpunpckhdq	ymm4,ymm4,ymm5
+	vpunpckhdq	ymm6,ymm6,ymm7
+	vpunpcklqdq	ymm5,ymm15,ymm8
+	vpunpckhqdq	ymm15,ymm15,ymm8
+	vpunpcklqdq	ymm7,ymm4,ymm6
+	vpunpckhqdq	ymm4,ymm4,ymm6
+	vperm2i128	ymm8,ymm13,ymm5,0x20
+	vperm2i128	ymm5,ymm13,ymm5,0x31
+	vperm2i128	ymm13,ymm2,ymm15,0x20
+	vperm2i128	ymm15,ymm2,ymm15,0x31
+	vperm2i128	ymm2,ymm9,ymm7,0x20
+	vperm2i128	ymm7,ymm9,ymm7,0x31
+	vperm2i128	ymm9,ymm12,ymm4,0x20
+	vperm2i128	ymm4,ymm12,ymm4,0x31
+	vmovdqa	ymm6,YMMWORD[rsp]
+	vmovdqa	ymm12,YMMWORD[32+rsp]
+
+	cmp	rdx,64*8
+	jb	NEAR $L$tail8x
+
+	vpxor	ymm6,ymm6,YMMWORD[rsi]
+	vpxor	ymm8,ymm8,YMMWORD[32+rsi]
+	vpxor	ymm1,ymm1,YMMWORD[64+rsi]
+	vpxor	ymm5,ymm5,YMMWORD[96+rsi]
+	lea	rsi,[128+rsi]
+	vmovdqu	YMMWORD[rdi],ymm6
+	vmovdqu	YMMWORD[32+rdi],ymm8
+	vmovdqu	YMMWORD[64+rdi],ymm1
+	vmovdqu	YMMWORD[96+rdi],ymm5
+	lea	rdi,[128+rdi]
+
+	vpxor	ymm12,ymm12,YMMWORD[rsi]
+	vpxor	ymm13,ymm13,YMMWORD[32+rsi]
+	vpxor	ymm10,ymm10,YMMWORD[64+rsi]
+	vpxor	ymm15,ymm15,YMMWORD[96+rsi]
+	lea	rsi,[128+rsi]
+	vmovdqu	YMMWORD[rdi],ymm12
+	vmovdqu	YMMWORD[32+rdi],ymm13
+	vmovdqu	YMMWORD[64+rdi],ymm10
+	vmovdqu	YMMWORD[96+rdi],ymm15
+	lea	rdi,[128+rdi]
+
+	vpxor	ymm14,ymm14,YMMWORD[rsi]
+	vpxor	ymm2,ymm2,YMMWORD[32+rsi]
+	vpxor	ymm3,ymm3,YMMWORD[64+rsi]
+	vpxor	ymm7,ymm7,YMMWORD[96+rsi]
+	lea	rsi,[128+rsi]
+	vmovdqu	YMMWORD[rdi],ymm14
+	vmovdqu	YMMWORD[32+rdi],ymm2
+	vmovdqu	YMMWORD[64+rdi],ymm3
+	vmovdqu	YMMWORD[96+rdi],ymm7
+	lea	rdi,[128+rdi]
+
+	vpxor	ymm11,ymm11,YMMWORD[rsi]
+	vpxor	ymm9,ymm9,YMMWORD[32+rsi]
+	vpxor	ymm0,ymm0,YMMWORD[64+rsi]
+	vpxor	ymm4,ymm4,YMMWORD[96+rsi]
+	lea	rsi,[128+rsi]
+	vmovdqu	YMMWORD[rdi],ymm11
+	vmovdqu	YMMWORD[32+rdi],ymm9
+	vmovdqu	YMMWORD[64+rdi],ymm0
+	vmovdqu	YMMWORD[96+rdi],ymm4
+	lea	rdi,[128+rdi]
+
+	sub	rdx,64*8
+	jnz	NEAR $L$oop_outer8x
+
+	jmp	NEAR $L$done8x
+
+$L$tail8x:
+	cmp	rdx,448
+	jae	NEAR $L$448_or_more8x
+	cmp	rdx,384
+	jae	NEAR $L$384_or_more8x
+	cmp	rdx,320
+	jae	NEAR $L$320_or_more8x
+	cmp	rdx,256
+	jae	NEAR $L$256_or_more8x
+	cmp	rdx,192
+	jae	NEAR $L$192_or_more8x
+	cmp	rdx,128
+	jae	NEAR $L$128_or_more8x
+	cmp	rdx,64
+	jae	NEAR $L$64_or_more8x
+
+	xor	r10,r10
+	vmovdqa	YMMWORD[rsp],ymm6
+	vmovdqa	YMMWORD[32+rsp],ymm8
+	jmp	NEAR $L$oop_tail8x
+
+ALIGN	32
+$L$64_or_more8x:
+	vpxor	ymm6,ymm6,YMMWORD[rsi]
+	vpxor	ymm8,ymm8,YMMWORD[32+rsi]
+	vmovdqu	YMMWORD[rdi],ymm6
+	vmovdqu	YMMWORD[32+rdi],ymm8
+	je	NEAR $L$done8x
+
+	lea	rsi,[64+rsi]
+	xor	r10,r10
+	vmovdqa	YMMWORD[rsp],ymm1
+	lea	rdi,[64+rdi]
+	sub	rdx,64
+	vmovdqa	YMMWORD[32+rsp],ymm5
+	jmp	NEAR $L$oop_tail8x
+
+ALIGN	32
+$L$128_or_more8x:
+	vpxor	ymm6,ymm6,YMMWORD[rsi]
+	vpxor	ymm8,ymm8,YMMWORD[32+rsi]
+	vpxor	ymm1,ymm1,YMMWORD[64+rsi]
+	vpxor	ymm5,ymm5,YMMWORD[96+rsi]
+	vmovdqu	YMMWORD[rdi],ymm6
+	vmovdqu	YMMWORD[32+rdi],ymm8
+	vmovdqu	YMMWORD[64+rdi],ymm1
+	vmovdqu	YMMWORD[96+rdi],ymm5
+	je	NEAR $L$done8x
+
+	lea	rsi,[128+rsi]
+	xor	r10,r10
+	vmovdqa	YMMWORD[rsp],ymm12
+	lea	rdi,[128+rdi]
+	sub	rdx,128
+	vmovdqa	YMMWORD[32+rsp],ymm13
+	jmp	NEAR $L$oop_tail8x
+
+ALIGN	32
+$L$192_or_more8x:
+	vpxor	ymm6,ymm6,YMMWORD[rsi]
+	vpxor	ymm8,ymm8,YMMWORD[32+rsi]
+	vpxor	ymm1,ymm1,YMMWORD[64+rsi]
+	vpxor	ymm5,ymm5,YMMWORD[96+rsi]
+	vpxor	ymm12,ymm12,YMMWORD[128+rsi]
+	vpxor	ymm13,ymm13,YMMWORD[160+rsi]
+	vmovdqu	YMMWORD[rdi],ymm6
+	vmovdqu	YMMWORD[32+rdi],ymm8
+	vmovdqu	YMMWORD[64+rdi],ymm1
+	vmovdqu	YMMWORD[96+rdi],ymm5
+	vmovdqu	YMMWORD[128+rdi],ymm12
+	vmovdqu	YMMWORD[160+rdi],ymm13
+	je	NEAR $L$done8x
+
+	lea	rsi,[192+rsi]
+	xor	r10,r10
+	vmovdqa	YMMWORD[rsp],ymm10
+	lea	rdi,[192+rdi]
+	sub	rdx,192
+	vmovdqa	YMMWORD[32+rsp],ymm15
+	jmp	NEAR $L$oop_tail8x
+
+ALIGN	32
+$L$256_or_more8x:
+	vpxor	ymm6,ymm6,YMMWORD[rsi]
+	vpxor	ymm8,ymm8,YMMWORD[32+rsi]
+	vpxor	ymm1,ymm1,YMMWORD[64+rsi]
+	vpxor	ymm5,ymm5,YMMWORD[96+rsi]
+	vpxor	ymm12,ymm12,YMMWORD[128+rsi]
+	vpxor	ymm13,ymm13,YMMWORD[160+rsi]
+	vpxor	ymm10,ymm10,YMMWORD[192+rsi]
+	vpxor	ymm15,ymm15,YMMWORD[224+rsi]
+	vmovdqu	YMMWORD[rdi],ymm6
+	vmovdqu	YMMWORD[32+rdi],ymm8
+	vmovdqu	YMMWORD[64+rdi],ymm1
+	vmovdqu	YMMWORD[96+rdi],ymm5
+	vmovdqu	YMMWORD[128+rdi],ymm12
+	vmovdqu	YMMWORD[160+rdi],ymm13
+	vmovdqu	YMMWORD[192+rdi],ymm10
+	vmovdqu	YMMWORD[224+rdi],ymm15
+	je	NEAR $L$done8x
+
+	lea	rsi,[256+rsi]
+	xor	r10,r10
+	vmovdqa	YMMWORD[rsp],ymm14
+	lea	rdi,[256+rdi]
+	sub	rdx,256
+	vmovdqa	YMMWORD[32+rsp],ymm2
+	jmp	NEAR $L$oop_tail8x
+
+ALIGN	32
+$L$320_or_more8x:
+	vpxor	ymm6,ymm6,YMMWORD[rsi]
+	vpxor	ymm8,ymm8,YMMWORD[32+rsi]
+	vpxor	ymm1,ymm1,YMMWORD[64+rsi]
+	vpxor	ymm5,ymm5,YMMWORD[96+rsi]
+	vpxor	ymm12,ymm12,YMMWORD[128+rsi]
+	vpxor	ymm13,ymm13,YMMWORD[160+rsi]
+	vpxor	ymm10,ymm10,YMMWORD[192+rsi]
+	vpxor	ymm15,ymm15,YMMWORD[224+rsi]
+	vpxor	ymm14,ymm14,YMMWORD[256+rsi]
+	vpxor	ymm2,ymm2,YMMWORD[288+rsi]
+	vmovdqu	YMMWORD[rdi],ymm6
+	vmovdqu	YMMWORD[32+rdi],ymm8
+	vmovdqu	YMMWORD[64+rdi],ymm1
+	vmovdqu	YMMWORD[96+rdi],ymm5
+	vmovdqu	YMMWORD[128+rdi],ymm12
+	vmovdqu	YMMWORD[160+rdi],ymm13
+	vmovdqu	YMMWORD[192+rdi],ymm10
+	vmovdqu	YMMWORD[224+rdi],ymm15
+	vmovdqu	YMMWORD[256+rdi],ymm14
+	vmovdqu	YMMWORD[288+rdi],ymm2
+	je	NEAR $L$done8x
+
+	lea	rsi,[320+rsi]
+	xor	r10,r10
+	vmovdqa	YMMWORD[rsp],ymm3
+	lea	rdi,[320+rdi]
+	sub	rdx,320
+	vmovdqa	YMMWORD[32+rsp],ymm7
+	jmp	NEAR $L$oop_tail8x
+
+ALIGN	32
+$L$384_or_more8x:
+	vpxor	ymm6,ymm6,YMMWORD[rsi]
+	vpxor	ymm8,ymm8,YMMWORD[32+rsi]
+	vpxor	ymm1,ymm1,YMMWORD[64+rsi]
+	vpxor	ymm5,ymm5,YMMWORD[96+rsi]
+	vpxor	ymm12,ymm12,YMMWORD[128+rsi]
+	vpxor	ymm13,ymm13,YMMWORD[160+rsi]
+	vpxor	ymm10,ymm10,YMMWORD[192+rsi]
+	vpxor	ymm15,ymm15,YMMWORD[224+rsi]
+	vpxor	ymm14,ymm14,YMMWORD[256+rsi]
+	vpxor	ymm2,ymm2,YMMWORD[288+rsi]
+	vpxor	ymm3,ymm3,YMMWORD[320+rsi]
+	vpxor	ymm7,ymm7,YMMWORD[352+rsi]
+	vmovdqu	YMMWORD[rdi],ymm6
+	vmovdqu	YMMWORD[32+rdi],ymm8
+	vmovdqu	YMMWORD[64+rdi],ymm1
+	vmovdqu	YMMWORD[96+rdi],ymm5
+	vmovdqu	YMMWORD[128+rdi],ymm12
+	vmovdqu	YMMWORD[160+rdi],ymm13
+	vmovdqu	YMMWORD[192+rdi],ymm10
+	vmovdqu	YMMWORD[224+rdi],ymm15
+	vmovdqu	YMMWORD[256+rdi],ymm14
+	vmovdqu	YMMWORD[288+rdi],ymm2
+	vmovdqu	YMMWORD[320+rdi],ymm3
+	vmovdqu	YMMWORD[352+rdi],ymm7
+	je	NEAR $L$done8x
+
+	lea	rsi,[384+rsi]
+	xor	r10,r10
+	vmovdqa	YMMWORD[rsp],ymm11
+	lea	rdi,[384+rdi]
+	sub	rdx,384
+	vmovdqa	YMMWORD[32+rsp],ymm9
+	jmp	NEAR $L$oop_tail8x
+
+ALIGN	32
+$L$448_or_more8x:
+	vpxor	ymm6,ymm6,YMMWORD[rsi]
+	vpxor	ymm8,ymm8,YMMWORD[32+rsi]
+	vpxor	ymm1,ymm1,YMMWORD[64+rsi]
+	vpxor	ymm5,ymm5,YMMWORD[96+rsi]
+	vpxor	ymm12,ymm12,YMMWORD[128+rsi]
+	vpxor	ymm13,ymm13,YMMWORD[160+rsi]
+	vpxor	ymm10,ymm10,YMMWORD[192+rsi]
+	vpxor	ymm15,ymm15,YMMWORD[224+rsi]
+	vpxor	ymm14,ymm14,YMMWORD[256+rsi]
+	vpxor	ymm2,ymm2,YMMWORD[288+rsi]
+	vpxor	ymm3,ymm3,YMMWORD[320+rsi]
+	vpxor	ymm7,ymm7,YMMWORD[352+rsi]
+	vpxor	ymm11,ymm11,YMMWORD[384+rsi]
+	vpxor	ymm9,ymm9,YMMWORD[416+rsi]
+	vmovdqu	YMMWORD[rdi],ymm6
+	vmovdqu	YMMWORD[32+rdi],ymm8
+	vmovdqu	YMMWORD[64+rdi],ymm1
+	vmovdqu	YMMWORD[96+rdi],ymm5
+	vmovdqu	YMMWORD[128+rdi],ymm12
+	vmovdqu	YMMWORD[160+rdi],ymm13
+	vmovdqu	YMMWORD[192+rdi],ymm10
+	vmovdqu	YMMWORD[224+rdi],ymm15
+	vmovdqu	YMMWORD[256+rdi],ymm14
+	vmovdqu	YMMWORD[288+rdi],ymm2
+	vmovdqu	YMMWORD[320+rdi],ymm3
+	vmovdqu	YMMWORD[352+rdi],ymm7
+	vmovdqu	YMMWORD[384+rdi],ymm11
+	vmovdqu	YMMWORD[416+rdi],ymm9
+	je	NEAR $L$done8x
+
+	lea	rsi,[448+rsi]
+	xor	r10,r10
+	vmovdqa	YMMWORD[rsp],ymm0
+	lea	rdi,[448+rdi]
+	sub	rdx,448
+	vmovdqa	YMMWORD[32+rsp],ymm4
+
+$L$oop_tail8x:
+	movzx	eax,BYTE[r10*1+rsi]
+	movzx	ecx,BYTE[r10*1+rsp]
+	lea	r10,[1+r10]
+	xor	eax,ecx
+	mov	BYTE[((-1))+r10*1+rdi],al
+	dec	rdx
+	jnz	NEAR $L$oop_tail8x
+
+$L$done8x:
+	vzeroall
+	lea	r11,[((656+48))+rsp]
+	movaps	xmm6,XMMWORD[((-48))+r11]
+	movaps	xmm7,XMMWORD[((-32))+r11]
+	movaps	xmm8,XMMWORD[((-16))+r11]
+	movaps	xmm9,XMMWORD[r11]
+	movaps	xmm10,XMMWORD[16+r11]
+	movaps	xmm11,XMMWORD[32+r11]
+	movaps	xmm12,XMMWORD[48+r11]
+	movaps	xmm13,XMMWORD[64+r11]
+	movaps	xmm14,XMMWORD[80+r11]
+	movaps	xmm15,XMMWORD[96+r11]
+	mov	rsp,QWORD[640+rsp]
+	mov	rdi,QWORD[8+rsp]	;WIN64 epilogue
+	mov	rsi,QWORD[16+rsp]
+	DB	0F3h,0C3h		;repret
+$L$SEH_end_ChaCha20_8x:

third_party/boringssl/win-x86_64/crypto/cpu-x86_64-asm.asm

diff --git a/third_party/boringssl/win-x86_64/crypto/cpu-x86_64-asm.asm b/third_party/boringssl/win-x86_64/crypto/cpu-x86_64-asm.asm
deleted file mode 100644
index c92d7bb..0000000
--- a/third_party/boringssl/win-x86_64/crypto/cpu-x86_64-asm.asm
+++ /dev/null
@@ -1,154 +0,0 @@
-default	rel
-%define XMMWORD
-%define YMMWORD
-%define ZMMWORD
-section	.text code align=64
-
-
-global	OPENSSL_ia32_cpuid
-
-ALIGN	16
-OPENSSL_ia32_cpuid:
-	mov	QWORD[8+rsp],rdi	;WIN64 prologue
-	mov	QWORD[16+rsp],rsi
-	mov	rax,rsp
-$L$SEH_begin_OPENSSL_ia32_cpuid:
-	mov	rdi,rcx
-
-
-
-
-	mov	rdi,rcx
-	mov	r8,rbx
-
-	xor	eax,eax
-	mov	DWORD[8+rdi],eax
-	cpuid
-	mov	r11d,eax
-
-	xor	eax,eax
-	cmp	ebx,0x756e6547
-	setne	al
-	mov	r9d,eax
-	cmp	edx,0x49656e69
-	setne	al
-	or	r9d,eax
-	cmp	ecx,0x6c65746e
-	setne	al
-	or	r9d,eax
-	jz	NEAR $L$intel
-
-	cmp	ebx,0x68747541
-	setne	al
-	mov	r10d,eax
-	cmp	edx,0x69746E65
-	setne	al
-	or	r10d,eax
-	cmp	ecx,0x444D4163
-	setne	al
-	or	r10d,eax
-	jnz	NEAR $L$intel
-
-
-
-
-	mov	eax,0x80000000
-	cpuid
-
-
-	cmp	eax,0x80000001
-	jb	NEAR $L$intel
-	mov	r10d,eax
-	mov	eax,0x80000001
-	cpuid
-
-
-	or	r9d,ecx
-	and	r9d,0x00000801
-
-	cmp	r10d,0x80000008
-	jb	NEAR $L$intel
-
-	mov	eax,0x80000008
-	cpuid
-
-	movzx	r10,cl
-	inc	r10
-
-	mov	eax,1
-	cpuid
-
-	bt	edx,28
-	jnc	NEAR $L$generic
-	shr	ebx,16
-	cmp	bl,r10b
-	ja	NEAR $L$generic
-	and	edx,0xefffffff
-	jmp	NEAR $L$generic
-
-$L$intel:
-	cmp	r11d,4
-	mov	r10d,-1
-	jb	NEAR $L$nocacheinfo
-
-	mov	eax,4
-	mov	ecx,0
-	cpuid
-	mov	r10d,eax
-	shr	r10d,14
-	and	r10d,0xfff
-
-	cmp	r11d,7
-	jb	NEAR $L$nocacheinfo
-
-	mov	eax,7
-	xor	ecx,ecx
-	cpuid
-	mov	DWORD[8+rdi],ebx
-
-$L$nocacheinfo:
-	mov	eax,1
-	cpuid
-
-	and	edx,0xbfefffff
-	cmp	r9d,0
-	jne	NEAR $L$notintel
-	or	edx,0x40000000
-$L$notintel:
-	bt	edx,28
-	jnc	NEAR $L$generic
-	and	edx,0xefffffff
-	cmp	r10d,0
-	je	NEAR $L$generic
-
-	or	edx,0x10000000
-	shr	ebx,16
-	cmp	bl,1
-	ja	NEAR $L$generic
-	and	edx,0xefffffff
-$L$generic:
-	and	r9d,0x00000800
-	and	ecx,0xfffff7ff
-	or	r9d,ecx
-
-	mov	r10d,edx
-	bt	r9d,27
-	jnc	NEAR $L$clear_avx
-	xor	ecx,ecx
-DB	0x0f,0x01,0xd0
-	and	eax,6
-	cmp	eax,6
-	je	NEAR $L$done
-$L$clear_avx:
-	mov	eax,0xefffe7ff
-	and	r9d,eax
-	and	DWORD[8+rdi],0xffffffdf
-$L$done:
-	mov	DWORD[4+rdi],r9d
-	mov	DWORD[rdi],r10d
-	mov	rbx,r8
-	mov	rdi,QWORD[8+rsp]	;WIN64 epilogue
-	mov	rsi,QWORD[16+rsp]
-	DB	0F3h,0C3h		;repret
-$L$SEH_end_OPENSSL_ia32_cpuid:
-

third_party/boringssl/win-x86_64/crypto/ec/p256-x86_64-asm.asm

diff --git a/third_party/boringssl/win-x86_64/crypto/ec/p256-x86_64-asm.asm b/third_party/boringssl/win-x86_64/crypto/ec/p256-x86_64-asm.asm
new file mode 100644
index 0000000..a2e4075
--- /dev/null
+++ b/third_party/boringssl/win-x86_64/crypto/ec/p256-x86_64-asm.asm
@@ -0,0 +1,1925 @@
+default	rel
+%define XMMWORD
+%define YMMWORD
+%define ZMMWORD
+section	.text code align=64
+
+EXTERN	OPENSSL_ia32cap_P
+
+
+ALIGN	64
+$L$poly:
+	DQ	0xffffffffffffffff,0x00000000ffffffff,0x0000000000000000,0xffffffff00000001
+
+$L$One:
+	DD	1,1,1,1,1,1,1,1
+$L$Two:
+	DD	2,2,2,2,2,2,2,2
+$L$Three:
+	DD	3,3,3,3,3,3,3,3
+$L$ONE_mont:
+	DQ	0x0000000000000001,0xffffffff00000000,0xffffffffffffffff,0x00000000fffffffe
+
+
+ALIGN	64
+ecp_nistz256_mul_by_2:
+	mov	QWORD[8+rsp],rdi	;WIN64 prologue
+	mov	QWORD[16+rsp],rsi
+	mov	rax,rsp
+$L$SEH_begin_ecp_nistz256_mul_by_2:
+	mov	rdi,rcx
+	mov	rsi,rdx
+
+
+	push	r12
+	push	r13
+
+	mov	r8,QWORD[rsi]
+	mov	r9,QWORD[8+rsi]
+	add	r8,r8
+	mov	r10,QWORD[16+rsi]
+	adc	r9,r9
+	mov	r11,QWORD[24+rsi]
+	lea	rsi,[$L$poly]
+	mov	rax,r8
+	adc	r10,r10
+	adc	r11,r11
+	mov	rdx,r9
+	sbb	r13,r13
+
+	sub	r8,QWORD[rsi]
+	mov	rcx,r10
+	sbb	r9,QWORD[8+rsi]
+	sbb	r10,QWORD[16+rsi]
+	mov	r12,r11
+	sbb	r11,QWORD[24+rsi]
+	test	r13,r13
+
+	cmovz	r8,rax
+	cmovz	r9,rdx
+	mov	QWORD[rdi],r8
+	cmovz	r10,rcx
+	mov	QWORD[8+rdi],r9
+	cmovz	r11,r12
+	mov	QWORD[16+rdi],r10
+	mov	QWORD[24+rdi],r11
+
+	pop	r13
+	pop	r12
+	mov	rdi,QWORD[8+rsp]	;WIN64 epilogue
+	mov	rsi,QWORD[16+rsp]
+	DB	0F3h,0C3h		;repret
+$L$SEH_end_ecp_nistz256_mul_by_2:
+
+
+
+global	ecp_nistz256_neg
+
+ALIGN	32
+ecp_nistz256_neg:
+	mov	QWORD[8+rsp],rdi	;WIN64 prologue
+	mov	QWORD[16+rsp],rsi
+	mov	rax,rsp
+$L$SEH_begin_ecp_nistz256_neg:
+	mov	rdi,rcx
+	mov	rsi,rdx
+
+
+	push	r12
+	push	r13
+
+	xor	r8,r8
+	xor	r9,r9
+	xor	r10,r10
+	xor	r11,r11
+	xor	r13,r13
+
+	sub	r8,QWORD[rsi]
+	sbb	r9,QWORD[8+rsi]
+	sbb	r10,QWORD[16+rsi]
+	mov	rax,r8
+	sbb	r11,QWORD[24+rsi]
+	lea	rsi,[$L$poly]
+	mov	rdx,r9
+	sbb	r13,0
+
+	add	r8,QWORD[rsi]
+	mov	rcx,r10
+	adc	r9,QWORD[8+rsi]
+	adc	r10,QWORD[16+rsi]
+	mov	r12,r11
+	adc	r11,QWORD[24+rsi]
+	test	r13,r13
+
+	cmovz	r8,rax
+	cmovz	r9,rdx
+	mov	QWORD[rdi],r8
+	cmovz	r10,rcx
+	mov	QWORD[8+rdi],r9
+	cmovz	r11,r12
+	mov	QWORD[16+rdi],r10
+	mov	QWORD[24+rdi],r11
+
+	pop	r13
+	pop	r12
+	mov	rdi,QWORD[8+rsp]	;WIN64 epilogue
+	mov	rsi,QWORD[16+rsp]
+	DB	0F3h,0C3h		;repret
+$L$SEH_end_ecp_nistz256_neg:
+
+
+
+
+
+
+global	ecp_nistz256_mul_mont
+
+ALIGN	32
+ecp_nistz256_mul_mont:
+	mov	QWORD[8+rsp],rdi	;WIN64 prologue
+	mov	QWORD[16+rsp],rsi
+	mov	rax,rsp
+$L$SEH_begin_ecp_nistz256_mul_mont:
+	mov	rdi,rcx
+	mov	rsi,rdx
+	mov	rdx,r8
+
+
+$L$mul_mont:
+	push	rbp
+	push	rbx
+	push	r12
+	push	r13
+	push	r14
+	push	r15
+	mov	rbx,rdx
+	mov	rax,QWORD[rdx]
+	mov	r9,QWORD[rsi]
+	mov	r10,QWORD[8+rsi]
+	mov	r11,QWORD[16+rsi]
+	mov	r12,QWORD[24+rsi]
+
+	call	__ecp_nistz256_mul_montq
+$L$mul_mont_done:
+	pop	r15
+	pop	r14
+	pop	r13
+	pop	r12
+	pop	rbx
+	pop	rbp
+	mov	rdi,QWORD[8+rsp]	;WIN64 epilogue
+	mov	rsi,QWORD[16+rsp]
+	DB	0F3h,0C3h		;repret
+$L$SEH_end_ecp_nistz256_mul_mont:
+
+
+ALIGN	32
+__ecp_nistz256_mul_montq:
+
+
+	mov	rbp,rax
+	mul	r9
+	mov	r14,QWORD[(($L$poly+8))]
+	mov	r8,rax
+	mov	rax,rbp
+	mov	r9,rdx
+
+	mul	r10
+	mov	r15,QWORD[(($L$poly+24))]
+	add	r9,rax
+	mov	rax,rbp
+	adc	rdx,0
+	mov	r10,rdx
+
+	mul	r11
+	add	r10,rax
+	mov	rax,rbp
+	adc	rdx,0
+	mov	r11,rdx
+
+	mul	r12
+	add	r11,rax
+	mov	rax,r8
+	adc	rdx,0
+	xor	r13,r13
+	mov	r12,rdx
+
+
+
+
+
+
+
+
+
+
+	mov	rbp,r8
+	shl	r8,32
+	mul	r15
+	shr	rbp,32
+	add	r9,r8
+	adc	r10,rbp
+	adc	r11,rax
+	mov	rax,QWORD[8+rbx]
+	adc	r12,rdx
+	adc	r13,0
+	xor	r8,r8
+
+
+
+	mov	rbp,rax
+	mul	QWORD[rsi]
+	add	r9,rax
+	mov	rax,rbp
+	adc	rdx,0
+	mov	rcx,rdx
+
+	mul	QWORD[8+rsi]
+	add	r10,rcx
+	adc	rdx,0
+	add	r10,rax
+	mov	rax,rbp
+	adc	rdx,0
+	mov	rcx,rdx
+
+	mul	QWORD[16+rsi]
+	add	r11,rcx
+	adc	rdx,0
+	add	r11,rax
+	mov	rax,rbp
+	adc	rdx,0
+	mov	rcx,rdx
+
+	mul	QWORD[24+rsi]
+	add	r12,rcx
+	adc	rdx,0
+	add	r12,rax
+	mov	rax,r9
+	adc	r13,rdx
+	adc	r8,0
+
+
+
+	mov	rbp,r9
+	shl	r9,32
+	mul	r15
+	shr	rbp,32
+	add	r10,r9
+	adc	r11,rbp
+	adc	r12,rax
+	mov	rax,QWORD[16+rbx]
+	adc	r13,rdx
+	adc	r8,0
+	xor	r9,r9
+
+
+
+	mov	rbp,rax
+	mul	QWORD[rsi]
+	add	r10,rax
+	mov	rax,rbp
+	adc	rdx,0
+	mov	rcx,rdx
+
+	mul	QWORD[8+rsi]
+	add	r11,rcx
+	adc	rdx,0
+	add	r11,rax
+	mov	rax,rbp
+	adc	rdx,0
+	mov	rcx,rdx
+
+	mul	QWORD[16+rsi]
+	add	r12,rcx
+	adc	rdx,0
+	add	r12,rax
+	mov	rax,rbp
+	adc	rdx,0
+	mov	rcx,rdx
+
+	mul	QWORD[24+rsi]
+	add	r13,rcx
+	adc	rdx,0
+	add	r13,rax
+	mov	rax,r10
+	adc	r8,rdx
+	adc	r9,0
+
+
+
+	mov	rbp,r10
+	shl	r10,32
+	mul	r15
+	shr	rbp,32
+	add	r11,r10
+	adc	r12,rbp
+	adc	r13,rax
+	mov	rax,QWORD[24+rbx]
+	adc	r8,rdx
+	adc	r9,0
+	xor	r10,r10
+
+
+
+	mov	rbp,rax
+	mul	QWORD[rsi]
+	add	r11,rax
+	mov	rax,rbp
+	adc	rdx,0
+	mov	rcx,rdx
+
+	mul	QWORD[8+rsi]
+	add	r12,rcx
+	adc	rdx,0
+	add	r12,rax
+	mov	rax,rbp
+	adc	rdx,0
+	mov	rcx,rdx
+
+	mul	QWORD[16+rsi]
+	add	r13,rcx
+	adc	rdx,0
+	add	r13,rax
+	mov	rax,rbp
+	adc	rdx,0
+	mov	rcx,rdx
+
+	mul	QWORD[24+rsi]
+	add	r8,rcx
+	adc	rdx,0
+	add	r8,rax
+	mov	rax,r11
+	adc	r9,rdx
+	adc	r10,0
+
+
+
+	mov	rbp,r11
+	shl	r11,32
+	mul	r15
+	shr	rbp,32
+	add	r12,r11
+	adc	r13,rbp
+	mov	rcx,r12
+	adc	r8,rax
+	adc	r9,rdx
+	mov	rbp,r13
+	adc	r10,0
+
+
+
+	sub	r12,-1
+	mov	rbx,r8
+	sbb	r13,r14
+	sbb	r8,0
+	mov	rdx,r9
+	sbb	r9,r15
+	sbb	r10,0
+
+	cmovc	r12,rcx
+	cmovc	r13,rbp
+	mov	QWORD[rdi],r12
+	cmovc	r8,rbx
+	mov	QWORD[8+rdi],r13
+	cmovc	r9,rdx
+	mov	QWORD[16+rdi],r8
+	mov	QWORD[24+rdi],r9
+
+	DB	0F3h,0C3h		;repret
+
+
+
+
+
+
+
+
+
+global	ecp_nistz256_sqr_mont
+
+ALIGN	32
+ecp_nistz256_sqr_mont:
+	mov	QWORD[8+rsp],rdi	;WIN64 prologue
+	mov	QWORD[16+rsp],rsi
+	mov	rax,rsp
+$L$SEH_begin_ecp_nistz256_sqr_mont:
+	mov	rdi,rcx
+	mov	rsi,rdx
+
+
+	push	rbp
+	push	rbx
+	push	r12
+	push	r13
+	push	r14
+	push	r15
+	mov	rax,QWORD[rsi]
+	mov	r14,QWORD[8+rsi]
+	mov	r15,QWORD[16+rsi]
+	mov	r8,QWORD[24+rsi]
+
+	call	__ecp_nistz256_sqr_montq
+$L$sqr_mont_done:
+	pop	r15
+	pop	r14
+	pop	r13
+	pop	r12
+	pop	rbx
+	pop	rbp
+	mov	rdi,QWORD[8+rsp]	;WIN64 epilogue
+	mov	rsi,QWORD[16+rsp]
+	DB	0F3h,0C3h		;repret
+$L$SEH_end_ecp_nistz256_sqr_mont:
+
+
+ALIGN	32
+__ecp_nistz256_sqr_montq:
+	mov	r13,rax
+	mul	r14
+	mov	r9,rax
+	mov	rax,r15
+	mov	r10,rdx
+
+	mul	r13
+	add	r10,rax
+	mov	rax,r8
+	adc	rdx,0
+	mov	r11,rdx
+
+	mul	r13
+	add	r11,rax
+	mov	rax,r15
+	adc	rdx,0
+	mov	r12,rdx
+
+
+	mul	r14
+	add	r11,rax
+	mov	rax,r8
+	adc	rdx,0
+	mov	rbp,rdx
+
+	mul	r14
+	add	r12,rax
+	mov	rax,r8
+	adc	rdx,0
+	add	r12,rbp
+	mov	r13,rdx
+	adc	r13,0
+
+
+	mul	r15
+	xor	r15,r15
+	add	r13,rax
+	mov	rax,QWORD[rsi]
+	mov	r14,rdx
+	adc	r14,0
+
+	add	r9,r9
+	adc	r10,r10
+	adc	r11,r11
+	adc	r12,r12
+	adc	r13,r13
+	adc	r14,r14
+	adc	r15,0
+
+	mul	rax
+	mov	r8,rax
+	mov	rax,QWORD[8+rsi]
+	mov	rcx,rdx
+
+	mul	rax
+	add	r9,rcx
+	adc	r10,rax
+	mov	rax,QWORD[16+rsi]
+	adc	rdx,0
+	mov	rcx,rdx
+
+	mul	rax
+	add	r11,rcx
+	adc	r12,rax
+	mov	rax,QWORD[24+rsi]
+	adc	rdx,0
+	mov	rcx,rdx
+
+	mul	rax
+	add	r13,rcx
+	adc	r14,rax
+	mov	rax,r8
+	adc	r15,rdx
+
+	mov	rsi,QWORD[(($L$poly+8))]
+	mov	rbp,QWORD[(($L$poly+24))]
+
+
+
+
+	mov	rcx,r8
+	shl	r8,32
+	mul	rbp
+	shr	rcx,32
+	add	r9,r8
+	adc	r10,rcx
+	adc	r11,rax
+	mov	rax,r9
+	adc	rdx,0
+
+
+
+	mov	rcx,r9
+	shl	r9,32
+	mov	r8,rdx
+	mul	rbp
+	shr	rcx,32
+	add	r10,r9
+	adc	r11,rcx
+	adc	r8,rax
+	mov	rax,r10
+	adc	rdx,0
+
+
+
+	mov	rcx,r10
+	shl	r10,32
+	mov	r9,rdx
+	mul	rbp
+	shr	rcx,32
+	add	r11,r10
+	adc	r8,rcx
+	adc	r9,rax
+	mov	rax,r11
+	adc	rdx,0
+
+
+
+	mov	rcx,r11
+	shl	r11,32
+	mov	r10,rdx
+	mul	rbp
+	shr	rcx,32
+	add	r8,r11
+	adc	r9,rcx
+	adc	r10,rax
+	adc	rdx,0
+	xor	r11,r11
+
+
+
+	add	r12,r8
+	adc	r13,r9
+	mov	r8,r12
+	adc	r14,r10
+	adc	r15,rdx
+	mov	r9,r13
+	adc	r11,0
+
+	sub	r12,-1
+	mov	r10,r14
+	sbb	r13,rsi
+	sbb	r14,0
+	mov	rcx,r15
+	sbb	r15,rbp
+	sbb	r11,0
+
+	cmovc	r12,r8
+	cmovc	r13,r9
+	mov	QWORD[rdi],r12
+	cmovc	r14,r10
+	mov	QWORD[8+rdi],r13
+	cmovc	r15,rcx
+	mov	QWORD[16+rdi],r14
+	mov	QWORD[24+rdi],r15
+
+	DB	0F3h,0C3h		;repret
+
+
+
+
+
+
+
+global	ecp_nistz256_from_mont
+
+ALIGN	32
+ecp_nistz256_from_mont:
+	mov	QWORD[8+rsp],rdi	;WIN64 prologue
+	mov	QWORD[16+rsp],rsi
+	mov	rax,rsp
+$L$SEH_begin_ecp_nistz256_from_mont:
+	mov	rdi,rcx
+	mov	rsi,rdx
+
+
+	push	r12
+	push	r13
+
+	mov	rax,QWORD[rsi]
+	mov	r13,QWORD[(($L$poly+24))]
+	mov	r9,QWORD[8+rsi]
+	mov	r10,QWORD[16+rsi]
+	mov	r11,QWORD[24+rsi]
+	mov	r8,rax
+	mov	r12,QWORD[(($L$poly+8))]
+
+
+
+	mov	rcx,rax
+	shl	r8,32
+	mul	r13
+	shr	rcx,32
+	add	r9,r8
+	adc	r10,rcx
+	adc	r11,rax
+	mov	rax,r9
+	adc	rdx,0
+
+
+
+	mov	rcx,r9
+	shl	r9,32
+	mov	r8,rdx
+	mul	r13
+	shr	rcx,32
+	add	r10,r9
+	adc	r11,rcx
+	adc	r8,rax
+	mov	rax,r10
+	adc	rdx,0
+
+
+
+	mov	rcx,r10
+	shl	r10,32
+	mov	r9,rdx
+	mul	r13
+	shr	rcx,32
+	add	r11,r10
+	adc	r8,rcx
+	adc	r9,rax
+	mov	rax,r11
+	adc	rdx,0
+
+
+
+	mov	rcx,r11
+	shl	r11,32
+	mov	r10,rdx
+	mul	r13
+	shr	rcx,32
+	add	r8,r11
+	adc	r9,rcx
+	mov	rcx,r8
+	adc	r10,rax
+	mov	rsi,r9
+	adc	rdx,0
+
+	sub	r8,-1
+	mov	rax,r10
+	sbb	r9,r12
+	sbb	r10,0
+	mov	r11,rdx
+	sbb	rdx,r13
+	sbb	r13,r13
+
+	cmovnz	r8,rcx
+	cmovnz	r9,rsi
+	mov	QWORD[rdi],r8
+	cmovnz	r10,rax
+	mov	QWORD[8+rdi],r9
+	cmovz	r11,rdx
+	mov	QWORD[16+rdi],r10
+	mov	QWORD[24+rdi],r11
+
+	pop	r13
+	pop	r12
+	mov	rdi,QWORD[8+rsp]	;WIN64 epilogue
+	mov	rsi,QWORD[16+rsp]
+	DB	0F3h,0C3h		;repret
+$L$SEH_end_ecp_nistz256_from_mont:
+
+
+global	ecp_nistz256_select_w5
+
+ALIGN	32
+ecp_nistz256_select_w5:
+	lea	rax,[((-136))+rsp]
+$L$SEH_begin_ecp_nistz256_select_w5:
+DB	0x48,0x8d,0x60,0xe0
+DB	0x0f,0x29,0x70,0xe0
+DB	0x0f,0x29,0x78,0xf0
+DB	0x44,0x0f,0x29,0x00
+DB	0x44,0x0f,0x29,0x48,0x10
+DB	0x44,0x0f,0x29,0x50,0x20
+DB	0x44,0x0f,0x29,0x58,0x30
+DB	0x44,0x0f,0x29,0x60,0x40
+DB	0x44,0x0f,0x29,0x68,0x50
+DB	0x44,0x0f,0x29,0x70,0x60
+DB	0x44,0x0f,0x29,0x78,0x70
+	movdqa	xmm0,XMMWORD[$L$One]
+	movd	xmm1,r8d
+
+	pxor	xmm2,xmm2
+	pxor	xmm3,xmm3
+	pxor	xmm4,xmm4
+	pxor	xmm5,xmm5
+	pxor	xmm6,xmm6
+	pxor	xmm7,xmm7
+
+	movdqa	xmm8,xmm0
+	pshufd	xmm1,xmm1,0
+
+	mov	rax,16
+$L$select_loop_sse_w5:
+
+	movdqa	xmm15,xmm8
+	paddd	xmm8,xmm0
+	pcmpeqd	xmm15,xmm1
+
+	movdqa	xmm9,XMMWORD[rdx]
+	movdqa	xmm10,XMMWORD[16+rdx]
+	movdqa	xmm11,XMMWORD[32+rdx]
+	movdqa	xmm12,XMMWORD[48+rdx]
+	movdqa	xmm13,XMMWORD[64+rdx]
+	movdqa	xmm14,XMMWORD[80+rdx]
+	lea	rdx,[96+rdx]
+
+	pand	xmm9,xmm15
+	pand	xmm10,xmm15
+	por	xmm2,xmm9
+	pand	xmm11,xmm15
+	por	xmm3,xmm10
+	pand	xmm12,xmm15
+	por	xmm4,xmm11
+	pand	xmm13,xmm15
+	por	xmm5,xmm12
+	pand	xmm14,xmm15
+	por	xmm6,xmm13
+	por	xmm7,xmm14
+
+	dec	rax
+	jnz	NEAR $L$select_loop_sse_w5
+
+	movdqu	XMMWORD[rcx],xmm2
+	movdqu	XMMWORD[16+rcx],xmm3
+	movdqu	XMMWORD[32+rcx],xmm4
+	movdqu	XMMWORD[48+rcx],xmm5
+	movdqu	XMMWORD[64+rcx],xmm6
+	movdqu	XMMWORD[80+rcx],xmm7
+	movaps	xmm6,XMMWORD[rsp]
+	movaps	xmm7,XMMWORD[16+rsp]
+	movaps	xmm8,XMMWORD[32+rsp]
+	movaps	xmm9,XMMWORD[48+rsp]
+	movaps	xmm10,XMMWORD[64+rsp]
+	movaps	xmm11,XMMWORD[80+rsp]
+	movaps	xmm12,XMMWORD[96+rsp]
+	movaps	xmm13,XMMWORD[112+rsp]
+	movaps	xmm14,XMMWORD[128+rsp]
+	movaps	xmm15,XMMWORD[144+rsp]
+	lea	rsp,[168+rsp]
+$L$SEH_end_ecp_nistz256_select_w5:
+	DB	0F3h,0C3h		;repret
+
+
+
+
+global	ecp_nistz256_select_w7
+
+ALIGN	32
+ecp_nistz256_select_w7:
+	lea	rax,[((-136))+rsp]
+$L$SEH_begin_ecp_nistz256_select_w7:
+DB	0x48,0x8d,0x60,0xe0
+DB	0x0f,0x29,0x70,0xe0
+DB	0x0f,0x29,0x78,0xf0
+DB	0x44,0x0f,0x29,0x00
+DB	0x44,0x0f,0x29,0x48,0x10
+DB	0x44,0x0f,0x29,0x50,0x20
+DB	0x44,0x0f,0x29,0x58,0x30
+DB	0x44,0x0f,0x29,0x60,0x40
+DB	0x44,0x0f,0x29,0x68,0x50
+DB	0x44,0x0f,0x29,0x70,0x60
+DB	0x44,0x0f,0x29,0x78,0x70
+	movdqa	xmm8,XMMWORD[$L$One]
+	movd	xmm1,r8d
+
+	pxor	xmm2,xmm2
+	pxor	xmm3,xmm3
+	pxor	xmm4,xmm4
+	pxor	xmm5,xmm5
+
+	movdqa	xmm0,xmm8
+	pshufd	xmm1,xmm1,0
+	mov	rax,64
+
+$L$select_loop_sse_w7:
+	movdqa	xmm15,xmm8
+	paddd	xmm8,xmm0
+	movdqa	xmm9,XMMWORD[rdx]
+	movdqa	xmm10,XMMWORD[16+rdx]
+	pcmpeqd	xmm15,xmm1
+	movdqa	xmm11,XMMWORD[32+rdx]
+	movdqa	xmm12,XMMWORD[48+rdx]
+	lea	rdx,[64+rdx]
+
+	pand	xmm9,xmm15
+	pand	xmm10,xmm15
+	por	xmm2,xmm9
+	pand	xmm11,xmm15
+	por	xmm3,xmm10
+	pand	xmm12,xmm15
+	por	xmm4,xmm11
+	prefetcht0	[255+rdx]
+	por	xmm5,xmm12
+
+	dec	rax
+	jnz	NEAR $L$select_loop_sse_w7
+
+	movdqu	XMMWORD[rcx],xmm2
+	movdqu	XMMWORD[16+rcx],xmm3
+	movdqu	XMMWORD[32+rcx],xmm4
+	movdqu	XMMWORD[48+rcx],xmm5
+	movaps	xmm6,XMMWORD[rsp]
+	movaps	xmm7,XMMWORD[16+rsp]
+	movaps	xmm8,XMMWORD[32+rsp]
+	movaps	xmm9,XMMWORD[48+rsp]
+	movaps	xmm10,XMMWORD[64+rsp]
+	movaps	xmm11,XMMWORD[80+rsp]
+	movaps	xmm12,XMMWORD[96+rsp]
+	movaps	xmm13,XMMWORD[112+rsp]
+	movaps	xmm14,XMMWORD[128+rsp]
+	movaps	xmm15,XMMWORD[144+rsp]
+	lea	rsp,[168+rsp]
+$L$SEH_end_ecp_nistz256_select_w7:
+	DB	0F3h,0C3h		;repret
+
+global	ecp_nistz256_avx2_select_w7
+
+ALIGN	32
+ecp_nistz256_avx2_select_w7:
+	mov	QWORD[8+rsp],rdi	;WIN64 prologue
+	mov	QWORD[16+rsp],rsi
+	mov	rax,rsp
+$L$SEH_begin_ecp_nistz256_avx2_select_w7:
+	mov	rdi,rcx
+	mov	rsi,rdx
+	mov	rdx,r8
+
+
+DB	0x0f,0x0b
+	mov	rdi,QWORD[8+rsp]	;WIN64 epilogue
+	mov	rsi,QWORD[16+rsp]
+	DB	0F3h,0C3h		;repret
+$L$SEH_end_ecp_nistz256_avx2_select_w7:
+
+ALIGN	32
+__ecp_nistz256_add_toq:
+	add	r12,QWORD[rbx]
+	adc	r13,QWORD[8+rbx]
+	mov	rax,r12
+	adc	r8,QWORD[16+rbx]
+	adc	r9,QWORD[24+rbx]
+	mov	rbp,r13
+	sbb	r11,r11
+
+	sub	r12,-1
+	mov	rcx,r8
+	sbb	r13,r14
+	sbb	r8,0
+	mov	r10,r9
+	sbb	r9,r15
+	test	r11,r11
+
+	cmovz	r12,rax
+	cmovz	r13,rbp
+	mov	QWORD[rdi],r12
+	cmovz	r8,rcx
+	mov	QWORD[8+rdi],r13
+	cmovz	r9,r10
+	mov	QWORD[16+rdi],r8
+	mov	QWORD[24+rdi],r9
+
+	DB	0F3h,0C3h		;repret
+
+
+
+ALIGN	32
+__ecp_nistz256_sub_fromq:
+	sub	r12,QWORD[rbx]
+	sbb	r13,QWORD[8+rbx]
+	mov	rax,r12
+	sbb	r8,QWORD[16+rbx]
+	sbb	r9,QWORD[24+rbx]
+	mov	rbp,r13
+	sbb	r11,r11
+
+	add	r12,-1
+	mov	rcx,r8
+	adc	r13,r14
+	adc	r8,0
+	mov	r10,r9
+	adc	r9,r15
+	test	r11,r11
+
+	cmovz	r12,rax
+	cmovz	r13,rbp
+	mov	QWORD[rdi],r12
+	cmovz	r8,rcx
+	mov	QWORD[8+rdi],r13
+	cmovz	r9,r10
+	mov	QWORD[16+rdi],r8
+	mov	QWORD[24+rdi],r9
+
+	DB	0F3h,0C3h		;repret
+
+
+
+ALIGN	32
+__ecp_nistz256_subq:
+	sub	rax,r12
+	sbb	rbp,r13
+	mov	r12,rax
+	sbb	rcx,r8
+	sbb	r10,r9
+	mov	r13,rbp
+	sbb	r11,r11
+
+	add	rax,-1
+	mov	r8,rcx
+	adc	rbp,r14
+	adc	rcx,0
+	mov	r9,r10
+	adc	r10,r15
+	test	r11,r11
+
+	cmovnz	r12,rax
+	cmovnz	r13,rbp
+	cmovnz	r8,rcx
+	cmovnz	r9,r10
+
+	DB	0F3h,0C3h		;repret
+
+
+
+ALIGN	32
+__ecp_nistz256_mul_by_2q:
+	add	r12,r12
+	adc	r13,r13
+	mov	rax,r12
+	adc	r8,r8
+	adc	r9,r9
+	mov	rbp,r13
+	sbb	r11,r11
+
+	sub	r12,-1
+	mov	rcx,r8
+	sbb	r13,r14
+	sbb	r8,0
+	mov	r10,r9
+	sbb	r9,r15
+	test	r11,r11
+
+	cmovz	r12,rax
+	cmovz	r13,rbp
+	mov	QWORD[rdi],r12
+	cmovz	r8,rcx
+	mov	QWORD[8+rdi],r13
+	cmovz	r9,r10
+	mov	QWORD[16+rdi],r8
+	mov	QWORD[24+rdi],r9
+
+	DB	0F3h,0C3h		;repret
+
+global	ecp_nistz256_point_double
+
+ALIGN	32
+ecp_nistz256_point_double:
+	mov	QWORD[8+rsp],rdi	;WIN64 prologue
+	mov	QWORD[16+rsp],rsi
+	mov	rax,rsp
+$L$SEH_begin_ecp_nistz256_point_double:
+	mov	rdi,rcx
+	mov	rsi,rdx
+
+
+	push	rbp
+	push	rbx
+	push	r12
+	push	r13
+	push	r14
+	push	r15
+	sub	rsp,32*5+8
+
+$L$point_double_shortcutq:
+	movdqu	xmm0,XMMWORD[rsi]
+	mov	rbx,rsi
+	movdqu	xmm1,XMMWORD[16+rsi]
+	mov	r12,QWORD[((32+0))+rsi]
+	mov	r13,QWORD[((32+8))+rsi]
+	mov	r8,QWORD[((32+16))+rsi]
+	mov	r9,QWORD[((32+24))+rsi]
+	mov	r14,QWORD[(($L$poly+8))]
+	mov	r15,QWORD[(($L$poly+24))]
+	movdqa	XMMWORD[96+rsp],xmm0
+	movdqa	XMMWORD[(96+16)+rsp],xmm1
+	lea	r10,[32+rdi]
+	lea	r11,[64+rdi]
+DB	102,72,15,110,199
+DB	102,73,15,110,202
+DB	102,73,15,110,211
+
+	lea	rdi,[rsp]
+	call	__ecp_nistz256_mul_by_2q
+
+	mov	rax,QWORD[((64+0))+rsi]
+	mov	r14,QWORD[((64+8))+rsi]
+	mov	r15,QWORD[((64+16))+rsi]
+	mov	r8,QWORD[((64+24))+rsi]
+	lea	rsi,[((64-0))+rsi]
+	lea	rdi,[64+rsp]
+	call	__ecp_nistz256_sqr_montq
+
+	mov	rax,QWORD[((0+0))+rsp]
+	mov	r14,QWORD[((8+0))+rsp]
+	lea	rsi,[((0+0))+rsp]
+	mov	r15,QWORD[((16+0))+rsp]
+	mov	r8,QWORD[((24+0))+rsp]
+	lea	rdi,[rsp]
+	call	__ecp_nistz256_sqr_montq
+
+	mov	rax,QWORD[32+rbx]
+	mov	r9,QWORD[((64+0))+rbx]
+	mov	r10,QWORD[((64+8))+rbx]
+	mov	r11,QWORD[((64+16))+rbx]
+	mov	r12,QWORD[((64+24))+rbx]
+	lea	rsi,[((64-0))+rbx]
+	lea	rbx,[32+rbx]
+DB	102,72,15,126,215
+	call	__ecp_nistz256_mul_montq
+	call	__ecp_nistz256_mul_by_2q
+
+	mov	r12,QWORD[((96+0))+rsp]
+	mov	r13,QWORD[((96+8))+rsp]
+	lea	rbx,[64+rsp]
+	mov	r8,QWORD[((96+16))+rsp]
+	mov	r9,QWORD[((96+24))+rsp]
+	lea	rdi,[32+rsp]
+	call	__ecp_nistz256_add_toq
+
+	mov	r12,QWORD[((96+0))+rsp]
+	mov	r13,QWORD[((96+8))+rsp]
+	lea	rbx,[64+rsp]
+	mov	r8,QWORD[((96+16))+rsp]
+	mov	r9,QWORD[((96+24))+rsp]
+	lea	rdi,[64+rsp]
+	call	__ecp_nistz256_sub_fromq
+
+	mov	rax,QWORD[((0+0))+rsp]
+	mov	r14,QWORD[((8+0))+rsp]
+	lea	rsi,[((0+0))+rsp]
+	mov	r15,QWORD[((16+0))+rsp]
+	mov	r8,QWORD[((24+0))+rsp]
+DB	102,72,15,126,207
+	call	__ecp_nistz256_sqr_montq
+	xor	r9,r9
+	mov	rax,r12
+	add	r12,-1
+	mov	r10,r13
+	adc	r13,rsi
+	mov	rcx,r14
+	adc	r14,0
+	mov	r8,r15
+	adc	r15,rbp
+	adc	r9,0
+	xor	rsi,rsi
+	test	rax,1
+
+	cmovz	r12,rax
+	cmovz	r13,r10
+	cmovz	r14,rcx
+	cmovz	r15,r8
+	cmovz	r9,rsi
+
+	mov	rax,r13
+	shr	r12,1
+	shl	rax,63
+	mov	r10,r14
+	shr	r13,1
+	or	r12,rax
+	shl	r10,63
+	mov	rcx,r15
+	shr	r14,1
+	or	r13,r10
+	shl	rcx,63
+	mov	QWORD[rdi],r12
+	shr	r15,1
+	mov	QWORD[8+rdi],r13
+	shl	r9,63
+	or	r14,rcx
+	or	r15,r9
+	mov	QWORD[16+rdi],r14
+	mov	QWORD[24+rdi],r15
+	mov	rax,QWORD[64+rsp]
+	lea	rbx,[64+rsp]
+	mov	r9,QWORD[((0+32))+rsp]
+	mov	r10,QWORD[((8+32))+rsp]
+	lea	rsi,[((0+32))+rsp]
+	mov	r11,QWORD[((16+32))+rsp]
+	mov	r12,QWORD[((24+32))+rsp]
+	lea	rdi,[32+rsp]
+	call	__ecp_nistz256_mul_montq
+
+	lea	rdi,[128+rsp]
+	call	__ecp_nistz256_mul_by_2q
+
+	lea	rbx,[32+rsp]
+	lea	rdi,[32+rsp]
+	call	__ecp_nistz256_add_toq
+
+	mov	rax,QWORD[96+rsp]
+	lea	rbx,[96+rsp]
+	mov	r9,QWORD[((0+0))+rsp]
+	mov	r10,QWORD[((8+0))+rsp]
+	lea	rsi,[((0+0))+rsp]
+	mov	r11,QWORD[((16+0))+rsp]
+	mov	r12,QWORD[((24+0))+rsp]
+	lea	rdi,[rsp]
+	call	__ecp_nistz256_mul_montq
+
+	lea	rdi,[128+rsp]
+	call	__ecp_nistz256_mul_by_2q
+
+	mov	rax,QWORD[((0+32))+rsp]
+	mov	r14,QWORD[((8+32))+rsp]
+	lea	rsi,[((0+32))+rsp]
+	mov	r15,QWORD[((16+32))+rsp]
+	mov	r8,QWORD[((24+32))+rsp]
+DB	102,72,15,126,199
+	call	__ecp_nistz256_sqr_montq
+
+	lea	rbx,[128+rsp]
+	mov	r8,r14
+	mov	r9,r15
+	mov	r14,rsi
+	mov	r15,rbp
+	call	__ecp_nistz256_sub_fromq
+
+	mov	rax,QWORD[((0+0))+rsp]
+	mov	rbp,QWORD[((0+8))+rsp]
+	mov	rcx,QWORD[((0+16))+rsp]
+	mov	r10,QWORD[((0+24))+rsp]
+	lea	rdi,[rsp]
+	call	__ecp_nistz256_subq
+
+	mov	rax,QWORD[32+rsp]
+	lea	rbx,[32+rsp]
+	mov	r14,r12
+	xor	ecx,ecx
+	mov	QWORD[((0+0))+rsp],r12
+	mov	r10,r13
+	mov	QWORD[((0+8))+rsp],r13
+	cmovz	r11,r8
+	mov	QWORD[((0+16))+rsp],r8
+	lea	rsi,[((0-0))+rsp]
+	cmovz	r12,r9
+	mov	QWORD[((0+24))+rsp],r9
+	mov	r9,r14
+	lea	rdi,[rsp]
+	call	__ecp_nistz256_mul_montq
+
+DB	102,72,15,126,203
+DB	102,72,15,126,207
+	call	__ecp_nistz256_sub_fromq
+
+	add	rsp,32*5+8
+	pop	r15
+	pop	r14
+	pop	r13
+	pop	r12
+	pop	rbx
+	pop	rbp
+	mov	rdi,QWORD[8+rsp]	;WIN64 epilogue
+	mov	rsi,QWORD[16+rsp]
+	DB	0F3h,0C3h		;repret
+$L$SEH_end_ecp_nistz256_point_double:
+global	ecp_nistz256_point_add
+
+ALIGN	32
+ecp_nistz256_point_add:
+	mov	QWORD[8+rsp],rdi	;WIN64 prologue
+	mov	QWORD[16+rsp],rsi
+	mov	rax,rsp
+$L$SEH_begin_ecp_nistz256_point_add:
+	mov	rdi,rcx
+	mov	rsi,rdx
+	mov	rdx,r8
+
+
+	push	rbp
+	push	rbx
+	push	r12
+	push	r13
+	push	r14
+	push	r15
+	sub	rsp,32*18+8
+
+	movdqu	xmm0,XMMWORD[rsi]
+	movdqu	xmm1,XMMWORD[16+rsi]
+	movdqu	xmm2,XMMWORD[32+rsi]
+	movdqu	xmm3,XMMWORD[48+rsi]
+	movdqu	xmm4,XMMWORD[64+rsi]
+	movdqu	xmm5,XMMWORD[80+rsi]
+	mov	rbx,rsi
+	mov	rsi,rdx
+	movdqa	XMMWORD[384+rsp],xmm0
+	movdqa	XMMWORD[(384+16)+rsp],xmm1
+	por	xmm1,xmm0
+	movdqa	XMMWORD[416+rsp],xmm2
+	movdqa	XMMWORD[(416+16)+rsp],xmm3
+	por	xmm3,xmm2
+	movdqa	XMMWORD[448+rsp],xmm4
+	movdqa	XMMWORD[(448+16)+rsp],xmm5
+	por	xmm3,xmm1
+
+	movdqu	xmm0,XMMWORD[rsi]
+	pshufd	xmm5,xmm3,0xb1
+	movdqu	xmm1,XMMWORD[16+rsi]
+	movdqu	xmm2,XMMWORD[32+rsi]
+	por	xmm5,xmm3
+	movdqu	xmm3,XMMWORD[48+rsi]
+	mov	rax,QWORD[((64+0))+rsi]
+	mov	r14,QWORD[((64+8))+rsi]
+	mov	r15,QWORD[((64+16))+rsi]
+	mov	r8,QWORD[((64+24))+rsi]
+	movdqa	XMMWORD[480+rsp],xmm0
+	pshufd	xmm4,xmm5,0x1e
+	movdqa	XMMWORD[(480+16)+rsp],xmm1
+	por	xmm1,xmm0
+DB	102,72,15,110,199
+	movdqa	XMMWORD[512+rsp],xmm2
+	movdqa	XMMWORD[(512+16)+rsp],xmm3
+	por	xmm3,xmm2
+	por	xmm5,xmm4
+	pxor	xmm4,xmm4
+	por	xmm3,xmm1
+
+	lea	rsi,[((64-0))+rsi]
+	mov	QWORD[((544+0))+rsp],rax
+	mov	QWORD[((544+8))+rsp],r14
+	mov	QWORD[((544+16))+rsp],r15
+	mov	QWORD[((544+24))+rsp],r8
+	lea	rdi,[96+rsp]
+	call	__ecp_nistz256_sqr_montq
+
+	pcmpeqd	xmm5,xmm4
+	pshufd	xmm4,xmm3,0xb1
+	por	xmm4,xmm3
+	pshufd	xmm5,xmm5,0
+	pshufd	xmm3,xmm4,0x1e
+	por	xmm4,xmm3
+	pxor	xmm3,xmm3
+	pcmpeqd	xmm4,xmm3
+	pshufd	xmm4,xmm4,0
+	mov	rax,QWORD[((64+0))+rbx]
+	mov	r14,QWORD[((64+8))+rbx]
+	mov	r15,QWORD[((64+16))+rbx]
+	mov	r8,QWORD[((64+24))+rbx]
+DB	102,72,15,110,203
+
+	lea	rsi,[((64-0))+rbx]
+	lea	rdi,[32+rsp]
+	call	__ecp_nistz256_sqr_montq
+
+	mov	rax,QWORD[544+rsp]
+	lea	rbx,[544+rsp]
+	mov	r9,QWORD[((0+96))+rsp]
+	mov	r10,QWORD[((8+96))+rsp]
+	lea	rsi,[((0+96))+rsp]
+	mov	r11,QWORD[((16+96))+rsp]
+	mov	r12,QWORD[((24+96))+rsp]
+	lea	rdi,[224+rsp]
+	call	__ecp_nistz256_mul_montq
+
+	mov	rax,QWORD[448+rsp]
+	lea	rbx,[448+rsp]
+	mov	r9,QWORD[((0+32))+rsp]
+	mov	r10,QWORD[((8+32))+rsp]
+	lea	rsi,[((0+32))+rsp]
+	mov	r11,QWORD[((16+32))+rsp]
+	mov	r12,QWORD[((24+32))+rsp]
+	lea	rdi,[256+rsp]
+	call	__ecp_nistz256_mul_montq
+
+	mov	rax,QWORD[416+rsp]
+	lea	rbx,[416+rsp]
+	mov	r9,QWORD[((0+224))+rsp]
+	mov	r10,QWORD[((8+224))+rsp]
+	lea	rsi,[((0+224))+rsp]
+	mov	r11,QWORD[((16+224))+rsp]
+	mov	r12,QWORD[((24+224))+rsp]
+	lea	rdi,[224+rsp]
+	call	__ecp_nistz256_mul_montq
+
+	mov	rax,QWORD[512+rsp]
+	lea	rbx,[512+rsp]
+	mov	r9,QWORD[((0+256))+rsp]
+	mov	r10,QWORD[((8+256))+rsp]
+	lea	rsi,[((0+256))+rsp]
+	mov	r11,QWORD[((16+256))+rsp]
+	mov	r12,QWORD[((24+256))+rsp]
+	lea	rdi,[256+rsp]
+	call	__ecp_nistz256_mul_montq
+
+	lea	rbx,[224+rsp]
+	lea	rdi,[64+rsp]
+	call	__ecp_nistz256_sub_fromq
+
+	or	r12,r13
+	movdqa	xmm2,xmm4
+	or	r12,r8
+	or	r12,r9
+	por	xmm2,xmm5
+DB	102,73,15,110,220
+
+	mov	rax,QWORD[384+rsp]
+	lea	rbx,[384+rsp]
+	mov	r9,QWORD[((0+96))+rsp]
+	mov	r10,QWORD[((8+96))+rsp]
+	lea	rsi,[((0+96))+rsp]
+	mov	r11,QWORD[((16+96))+rsp]
+	mov	r12,QWORD[((24+96))+rsp]
+	lea	rdi,[160+rsp]
+	call	__ecp_nistz256_mul_montq
+
+	mov	rax,QWORD[480+rsp]
+	lea	rbx,[480+rsp]
+	mov	r9,QWORD[((0+32))+rsp]
+	mov	r10,QWORD[((8+32))+rsp]
+	lea	rsi,[((0+32))+rsp]
+	mov	r11,QWORD[((16+32))+rsp]
+	mov	r12,QWORD[((24+32))+rsp]
+	lea	rdi,[192+rsp]
+	call	__ecp_nistz256_mul_montq
+
+	lea	rbx,[160+rsp]
+	lea	rdi,[rsp]
+	call	__ecp_nistz256_sub_fromq
+
+	or	r12,r13
+	or	r12,r8
+	or	r12,r9
+
+DB	0x3e
+	jnz	NEAR $L$add_proceedq
+DB	102,73,15,126,208
+DB	102,73,15,126,217
+	test	r8,r8
+	jnz	NEAR $L$add_proceedq
+	test	r9,r9
+	jz	NEAR $L$add_doubleq
+
+DB	102,72,15,126,199
+	pxor	xmm0,xmm0
+	movdqu	XMMWORD[rdi],xmm0
+	movdqu	XMMWORD[16+rdi],xmm0
+	movdqu	XMMWORD[32+rdi],xmm0
+	movdqu	XMMWORD[48+rdi],xmm0
+	movdqu	XMMWORD[64+rdi],xmm0
+	movdqu	XMMWORD[80+rdi],xmm0
+	jmp	NEAR $L$add_doneq
+
+ALIGN	32
+$L$add_doubleq:
+DB	102,72,15,126,206
+DB	102,72,15,126,199
+	add	rsp,416
+	jmp	NEAR $L$point_double_shortcutq
+
+ALIGN	32
+$L$add_proceedq:
+	mov	rax,QWORD[((0+64))+rsp]
+	mov	r14,QWORD[((8+64))+rsp]
+	lea	rsi,[((0+64))+rsp]
+	mov	r15,QWORD[((16+64))+rsp]
+	mov	r8,QWORD[((24+64))+rsp]
+	lea	rdi,[96+rsp]
+	call	__ecp_nistz256_sqr_montq
+
+	mov	rax,QWORD[448+rsp]
+	lea	rbx,[448+rsp]
+	mov	r9,QWORD[((0+0))+rsp]
+	mov	r10,QWORD[((8+0))+rsp]
+	lea	rsi,[((0+0))+rsp]
+	mov	r11,QWORD[((16+0))+rsp]
+	mov	r12,QWORD[((24+0))+rsp]
+	lea	rdi,[352+rsp]
+	call	__ecp_nistz256_mul_montq
+
+	mov	rax,QWORD[((0+0))+rsp]
+	mov	r14,QWORD[((8+0))+rsp]
+	lea	rsi,[((0+0))+rsp]
+	mov	r15,QWORD[((16+0))+rsp]
+	mov	r8,QWORD[((24+0))+rsp]
+	lea	rdi,[32+rsp]
+	call	__ecp_nistz256_sqr_montq
+
+	mov	rax,QWORD[544+rsp]
+	lea	rbx,[544+rsp]
+	mov	r9,QWORD[((0+352))+rsp]
+	mov	r10,QWORD[((8+352))+rsp]
+	lea	rsi,[((0+352))+rsp]
+	mov	r11,QWORD[((16+352))+rsp]
+	mov	r12,QWORD[((24+352))+rsp]
+	lea	rdi,[352+rsp]
+	call	__ecp_nistz256_mul_montq
+
+	mov	rax,QWORD[rsp]
+	lea	rbx,[rsp]
+	mov	r9,QWORD[((0+32))+rsp]
+	mov	r10,QWORD[((8+32))+rsp]
+	lea	rsi,[((0+32))+rsp]
+	mov	r11,QWORD[((16+32))+rsp]
+	mov	r12,QWORD[((24+32))+rsp]
+	lea	rdi,[128+rsp]
+	call	__ecp_nistz256_mul_montq
+
+	mov	rax,QWORD[160+rsp]
+	lea	rbx,[160+rsp]
+	mov	r9,QWORD[((0+32))+rsp]
+	mov	r10,QWORD[((8+32))+rsp]
+	lea	rsi,[((0+32))+rsp]
+	mov	r11,QWORD[((16+32))+rsp]
+	mov	r12,QWORD[((24+32))+rsp]
+	lea	rdi,[192+rsp]
+	call	__ecp_nistz256_mul_montq
+
+
+
+
+	add	r12,r12
+	lea	rsi,[96+rsp]
+	adc	r13,r13
+	mov	rax,r12
+	adc	r8,r8
+	adc	r9,r9
+	mov	rbp,r13
+	sbb	r11,r11
+
+	sub	r12,-1
+	mov	rcx,r8
+	sbb	r13,r14
+	sbb	r8,0
+	mov	r10,r9
+	sbb	r9,r15
+	test	r11,r11
+
+	cmovz	r12,rax
+	mov	rax,QWORD[rsi]
+	cmovz	r13,rbp
+	mov	rbp,QWORD[8+rsi]
+	cmovz	r8,rcx
+	mov	rcx,QWORD[16+rsi]
+	cmovz	r9,r10
+	mov	r10,QWORD[24+rsi]
+
+	call	__ecp_nistz256_subq
+
+	lea	rbx,[128+rsp]
+	lea	rdi,[288+rsp]
+	call	__ecp_nistz256_sub_fromq
+
+	mov	rax,QWORD[((192+0))+rsp]
+	mov	rbp,QWORD[((192+8))+rsp]
+	mov	rcx,QWORD[((192+16))+rsp]
+	mov	r10,QWORD[((192+24))+rsp]
+	lea	rdi,[320+rsp]
+
+	call	__ecp_nistz256_subq
+
+	mov	QWORD[rdi],r12
+	mov	QWORD[8+rdi],r13
+	mov	QWORD[16+rdi],r8
+	mov	QWORD[24+rdi],r9
+	mov	rax,QWORD[128+rsp]
+	lea	rbx,[128+rsp]
+	mov	r9,QWORD[((0+224))+rsp]
+	mov	r10,QWORD[((8+224))+rsp]
+	lea	rsi,[((0+224))+rsp]
+	mov	r11,QWORD[((16+224))+rsp]
+	mov	r12,QWORD[((24+224))+rsp]
+	lea	rdi,[256+rsp]
+	call	__ecp_nistz256_mul_montq
+
+	mov	rax,QWORD[320+rsp]
+	lea	rbx,[320+rsp]
+	mov	r9,QWORD[((0+64))+rsp]
+	mov	r10,QWORD[((8+64))+rsp]
+	lea	rsi,[((0+64))+rsp]
+	mov	r11,QWORD[((16+64))+rsp]
+	mov	r12,QWORD[((24+64))+rsp]
+	lea	rdi,[320+rsp]
+	call	__ecp_nistz256_mul_montq
+
+	lea	rbx,[256+rsp]
+	lea	rdi,[320+rsp]
+	call	__ecp_nistz256_sub_fromq
+
+DB	102,72,15,126,199
+
+	movdqa	xmm0,xmm5
+	movdqa	xmm1,xmm5
+	pandn	xmm0,XMMWORD[352+rsp]
+	movdqa	xmm2,xmm5
+	pandn	xmm1,XMMWORD[((352+16))+rsp]
+	movdqa	xmm3,xmm5
+	pand	xmm2,XMMWORD[544+rsp]
+	pand	xmm3,XMMWORD[((544+16))+rsp]
+	por	xmm2,xmm0
+	por	xmm3,xmm1
+
+	movdqa	xmm0,xmm4
+	movdqa	xmm1,xmm4
+	pandn	xmm0,xmm2
+	movdqa	xmm2,xmm4
+	pandn	xmm1,xmm3
+	movdqa	xmm3,xmm4
+	pand	xmm2,XMMWORD[448+rsp]
+	pand	xmm3,XMMWORD[((448+16))+rsp]
+	por	xmm2,xmm0
+	por	xmm3,xmm1
+	movdqu	XMMWORD[64+rdi],xmm2
+	movdqu	XMMWORD[80+rdi],xmm3
+
+	movdqa	xmm0,xmm5
+	movdqa	xmm1,xmm5
+	pandn	xmm0,XMMWORD[288+rsp]
+	movdqa	xmm2,xmm5
+	pandn	xmm1,XMMWORD[((288+16))+rsp]
+	movdqa	xmm3,xmm5
+	pand	xmm2,XMMWORD[480+rsp]
+	pand	xmm3,XMMWORD[((480+16))+rsp]
+	por	xmm2,xmm0
+	por	xmm3,xmm1
+
+	movdqa	xmm0,xmm4
+	movdqa	xmm1,xmm4
+	pandn	xmm0,xmm2
+	movdqa	xmm2,xmm4
+	pandn	xmm1,xmm3
+	movdqa	xmm3,xmm4
+	pand	xmm2,XMMWORD[384+rsp]
+	pand	xmm3,XMMWORD[((384+16))+rsp]
+	por	xmm2,xmm0
+	por	xmm3,xmm1
+	movdqu	XMMWORD[rdi],xmm2
+	movdqu	XMMWORD[16+rdi],xmm3
+
+	movdqa	xmm0,xmm5
+	movdqa	xmm1,xmm5
+	pandn	xmm0,XMMWORD[320+rsp]
+	movdqa	xmm2,xmm5
+	pandn	xmm1,XMMWORD[((320+16))+rsp]
+	movdqa	xmm3,xmm5
+	pand	xmm2,XMMWORD[512+rsp]
+	pand	xmm3,XMMWORD[((512+16))+rsp]
+	por	xmm2,xmm0
+	por	xmm3,xmm1
+
+	movdqa	xmm0,xmm4
+	movdqa	xmm1,xmm4
+	pandn	xmm0,xmm2
+	movdqa	xmm2,xmm4
+	pandn	xmm1,xmm3
+	movdqa	xmm3,xmm4
+	pand	xmm2,XMMWORD[416+rsp]
+	pand	xmm3,XMMWORD[((416+16))+rsp]
+	por	xmm2,xmm0
+	por	xmm3,xmm1
+	movdqu	XMMWORD[32+rdi],xmm2
+	movdqu	XMMWORD[48+rdi],xmm3
+
+$L$add_doneq:
+	add	rsp,32*18+8
+	pop	r15
+	pop	r14
+	pop	r13
+	pop	r12
+	pop	rbx
+	pop	rbp
+	mov	rdi,QWORD[8+rsp]	;WIN64 epilogue
+	mov	rsi,QWORD[16+rsp]
+	DB	0F3h,0C3h		;repret
+$L$SEH_end_ecp_nistz256_point_add:
+global	ecp_nistz256_point_add_affine
+
+ALIGN	32
+ecp_nistz256_point_add_affine:
+	mov	QWORD[8+rsp],rdi	;WIN64 prologue
+	mov	QWORD[16+rsp],rsi
+	mov	rax,rsp
+$L$SEH_begin_ecp_nistz256_point_add_affine:
+	mov	rdi,rcx
+	mov	rsi,rdx
+	mov	rdx,r8
+
+
+	push	rbp
+	push	rbx
+	push	r12
+	push	r13
+	push	r14
+	push	r15
+	sub	rsp,32*15+8
+
+	movdqu	xmm0,XMMWORD[rsi]
+	mov	rbx,rdx
+	movdqu	xmm1,XMMWORD[16+rsi]
+	movdqu	xmm2,XMMWORD[32+rsi]
+	movdqu	xmm3,XMMWORD[48+rsi]
+	movdqu	xmm4,XMMWORD[64+rsi]
+	movdqu	xmm5,XMMWORD[80+rsi]
+	mov	rax,QWORD[((64+0))+rsi]
+	mov	r14,QWORD[((64+8))+rsi]
+	mov	r15,QWORD[((64+16))+rsi]
+	mov	r8,QWORD[((64+24))+rsi]
+	movdqa	XMMWORD[320+rsp],xmm0
+	movdqa	XMMWORD[(320+16)+rsp],xmm1
+	por	xmm1,xmm0
+	movdqa	XMMWORD[352+rsp],xmm2
+	movdqa	XMMWORD[(352+16)+rsp],xmm3
+	por	xmm3,xmm2
+	movdqa	XMMWORD[384+rsp],xmm4
+	movdqa	XMMWORD[(384+16)+rsp],xmm5
+	por	xmm3,xmm1
+
+	movdqu	xmm0,XMMWORD[rbx]
+	pshufd	xmm5,xmm3,0xb1
+	movdqu	xmm1,XMMWORD[16+rbx]
+	movdqu	xmm2,XMMWORD[32+rbx]
+	por	xmm5,xmm3
+	movdqu	xmm3,XMMWORD[48+rbx]
+	movdqa	XMMWORD[416+rsp],xmm0
+	pshufd	xmm4,xmm5,0x1e
+	movdqa	XMMWORD[(416+16)+rsp],xmm1
+	por	xmm1,xmm0
+DB	102,72,15,110,199
+	movdqa	XMMWORD[448+rsp],xmm2
+	movdqa	XMMWORD[(448+16)+rsp],xmm3
+	por	xmm3,xmm2
+	por	xmm5,xmm4
+	pxor	xmm4,xmm4
+	por	xmm3,xmm1
+
+	lea	rsi,[((64-0))+rsi]
+	lea	rdi,[32+rsp]
+	call	__ecp_nistz256_sqr_montq
+
+	pcmpeqd	xmm5,xmm4
+	pshufd	xmm4,xmm3,0xb1
+	mov	rax,QWORD[rbx]
+
+	mov	r9,r12
+	por	xmm4,xmm3
+	pshufd	xmm5,xmm5,0
+	pshufd	xmm3,xmm4,0x1e
+	mov	r10,r13
+	por	xmm4,xmm3
+	pxor	xmm3,xmm3
+	mov	r11,r14
+	pcmpeqd	xmm4,xmm3
+	pshufd	xmm4,xmm4,0
+
+	lea	rsi,[((32-0))+rsp]
+	mov	r12,r15
+	lea	rdi,[rsp]
+	call	__ecp_nistz256_mul_montq
+
+	lea	rbx,[320+rsp]
+	lea	rdi,[64+rsp]
+	call	__ecp_nistz256_sub_fromq
+
+	mov	rax,QWORD[384+rsp]
+	lea	rbx,[384+rsp]
+	mov	r9,QWORD[((0+32))+rsp]
+	mov	r10,QWORD[((8+32))+rsp]
+	lea	rsi,[((0+32))+rsp]
+	mov	r11,QWORD[((16+32))+rsp]
+	mov	r12,QWORD[((24+32))+rsp]
+	lea	rdi,[32+rsp]
+	call	__ecp_nistz256_mul_montq
+
+	mov	rax,QWORD[384+rsp]
+	lea	rbx,[384+rsp]
+	mov	r9,QWORD[((0+64))+rsp]
+	mov	r10,QWORD[((8+64))+rsp]
+	lea	rsi,[((0+64))+rsp]
+	mov	r11,QWORD[((16+64))+rsp]
+	mov	r12,QWORD[((24+64))+rsp]
+	lea	rdi,[288+rsp]
+	call	__ecp_nistz256_mul_montq
+
+	mov	rax,QWORD[448+rsp]
+	lea	rbx,[448+rsp]
+	mov	r9,QWORD[((0+32))+rsp]
+	mov	r10,QWORD[((8+32))+rsp]
+	lea	rsi,[((0+32))+rsp]
+	mov	r11,QWORD[((16+32))+rsp]
+	mov	r12,QWORD[((24+32))+rsp]
+	lea	rdi,[32+rsp]
+	call	__ecp_nistz256_mul_montq
+
+	lea	rbx,[352+rsp]
+	lea	rdi,[96+rsp]
+	call	__ecp_nistz256_sub_fromq
+
+	mov	rax,QWORD[((0+64))+rsp]
+	mov	r14,QWORD[((8+64))+rsp]
+	lea	rsi,[((0+64))+rsp]
+	mov	r15,QWORD[((16+64))+rsp]
+	mov	r8,QWORD[((24+64))+rsp]
+	lea	rdi,[128+rsp]
+	call	__ecp_nistz256_sqr_montq
+
+	mov	rax,QWORD[((0+96))+rsp]
+	mov	r14,QWORD[((8+96))+rsp]
+	lea	rsi,[((0+96))+rsp]
+	mov	r15,QWORD[((16+96))+rsp]
+	mov	r8,QWORD[((24+96))+rsp]
+	lea	rdi,[192+rsp]
+	call	__ecp_nistz256_sqr_montq
+
+	mov	rax,QWORD[128+rsp]
+	lea	rbx,[128+rsp]
+	mov	r9,QWORD[((0+64))+rsp]
+	mov	r10,QWORD[((8+64))+rsp]
+	lea	rsi,[((0+64))+rsp]
+	mov	r11,QWORD[((16+64))+rsp]
+	mov	r12,QWORD[((24+64))+rsp]
+	lea	rdi,[160+rsp]
+	call	__ecp_nistz256_mul_montq
+
+	mov	rax,QWORD[320+rsp]
+	lea	rbx,[320+rsp]
+	mov	r9,QWORD[((0+128))+rsp]
+	mov	r10,QWORD[((8+128))+rsp]
+	lea	rsi,[((0+128))+rsp]
+	mov	r11,QWORD[((16+128))+rsp]
+	mov	r12,QWORD[((24+128))+rsp]
+	lea	rdi,[rsp]
+	call	__ecp_nistz256_mul_montq
+
+
+
+
+	add	r12,r12
+	lea	rsi,[192+rsp]
+	adc	r13,r13
+	mov	rax,r12
+	adc	r8,r8
+	adc	r9,r9
+	mov	rbp,r13
+	sbb	r11,r11
+
+	sub	r12,-1
+	mov	rcx,r8
+	sbb	r13,r14
+	sbb	r8,0
+	mov	r10,r9
+	sbb	r9,r15
+	test	r11,r11
+
+	cmovz	r12,rax
+	mov	rax,QWORD[rsi]
+	cmovz	r13,rbp
+	mov	rbp,QWORD[8+rsi]
+	cmovz	r8,rcx
+	mov	rcx,QWORD[16+rsi]
+	cmovz	r9,r10
+	mov	r10,QWORD[24+rsi]
+
+	call	__ecp_nistz256_subq
+
+	lea	rbx,[160+rsp]
+	lea	rdi,[224+rsp]
+	call	__ecp_nistz256_sub_fromq
+
+	mov	rax,QWORD[((0+0))+rsp]
+	mov	rbp,QWORD[((0+8))+rsp]
+	mov	rcx,QWORD[((0+16))+rsp]
+	mov	r10,QWORD[((0+24))+rsp]
+	lea	rdi,[64+rsp]
+
+	call	__ecp_nistz256_subq
+
+	mov	QWORD[rdi],r12
+	mov	QWORD[8+rdi],r13
+	mov	QWORD[16+rdi],r8
+	mov	QWORD[24+rdi],r9
+	mov	rax,QWORD[352+rsp]
+	lea	rbx,[352+rsp]
+	mov	r9,QWORD[((0+160))+rsp]
+	mov	r10,QWORD[((8+160))+rsp]
+	lea	rsi,[((0+160))+rsp]
+	mov	r11,QWORD[((16+160))+rsp]
+	mov	r12,QWORD[((24+160))+rsp]
+	lea	rdi,[32+rsp]
+	call	__ecp_nistz256_mul_montq
+
+	mov	rax,QWORD[96+rsp]
+	lea	rbx,[96+rsp]
+	mov	r9,QWORD[((0+64))+rsp]
+	mov	r10,QWORD[((8+64))+rsp]
+	lea	rsi,[((0+64))+rsp]
+	mov	r11,QWORD[((16+64))+rsp]
+	mov	r12,QWORD[((24+64))+rsp]
+	lea	rdi,[64+rsp]
+	call	__ecp_nistz256_mul_montq
+
+	lea	rbx,[32+rsp]
+	lea	rdi,[256+rsp]
+	call	__ecp_nistz256_sub_fromq
+
+DB	102,72,15,126,199
+
+	movdqa	xmm0,xmm5
+	movdqa	xmm1,xmm5
+	pandn	xmm0,XMMWORD[288+rsp]
+	movdqa	xmm2,xmm5
+	pandn	xmm1,XMMWORD[((288+16))+rsp]
+	movdqa	xmm3,xmm5
+	pand	xmm2,XMMWORD[$L$ONE_mont]
+	pand	xmm3,XMMWORD[(($L$ONE_mont+16))]
+	por	xmm2,xmm0
+	por	xmm3,xmm1
+
+	movdqa	xmm0,xmm4
+	movdqa	xmm1,xmm4
+	pandn	xmm0,xmm2
+	movdqa	xmm2,xmm4
+	pandn	xmm1,xmm3
+	movdqa	xmm3,xmm4
+	pand	xmm2,XMMWORD[384+rsp]
+	pand	xmm3,XMMWORD[((384+16))+rsp]
+	por	xmm2,xmm0
+	por	xmm3,xmm1
+	movdqu	XMMWORD[64+rdi],xmm2
+	movdqu	XMMWORD[80+rdi],xmm3
+
+	movdqa	xmm0,xmm5
+	movdqa	xmm1,xmm5
+	pandn	xmm0,XMMWORD[224+rsp]
+	movdqa	xmm2,xmm5
+	pandn	xmm1,XMMWORD[((224+16))+rsp]
+	movdqa	xmm3,xmm5
+	pand	xmm2,XMMWORD[416+rsp]
+	pand	xmm3,XMMWORD[((416+16))+rsp]
+	por	xmm2,xmm0
+	por	xmm3,xmm1
+
+	movdqa	xmm0,xmm4
+	movdqa	xmm1,xmm4
+	pandn	xmm0,xmm2
+	movdqa	xmm2,xmm4
+	pandn	xmm1,xmm3
+	movdqa	xmm3,xmm4
+	pand	xmm2,XMMWORD[320+rsp]
+	pand	xmm3,XMMWORD[((320+16))+rsp]
+	por	xmm2,xmm0
+	por	xmm3,xmm1
+	movdqu	XMMWORD[rdi],xmm2
+	movdqu	XMMWORD[16+rdi],xmm3
+
+	movdqa	xmm0,xmm5
+	movdqa	xmm1,xmm5
+	pandn	xmm0,XMMWORD[256+rsp]
+	movdqa	xmm2,xmm5
+	pandn	xmm1,XMMWORD[((256+16))+rsp]
+	movdqa	xmm3,xmm5
+	pand	xmm2,XMMWORD[448+rsp]
+	pand	xmm3,XMMWORD[((448+16))+rsp]
+	por	xmm2,xmm0
+	por	xmm3,xmm1
+
+	movdqa	xmm0,xmm4
+	movdqa	xmm1,xmm4
+	pandn	xmm0,xmm2
+	movdqa	xmm2,xmm4
+	pandn	xmm1,xmm3
+	movdqa	xmm3,xmm4
+	pand	xmm2,XMMWORD[352+rsp]
+	pand	xmm3,XMMWORD[((352+16))+rsp]
+	por	xmm2,xmm0
+	por	xmm3,xmm1
+	movdqu	XMMWORD[32+rdi],xmm2
+	movdqu	XMMWORD[48+rdi],xmm3
+
+	add	rsp,32*15+8
+	pop	r15
+	pop	r14
+	pop	r13
+	pop	r12
+	pop	rbx
+	pop	rbp
+	mov	rdi,QWORD[8+rsp]	;WIN64 epilogue
+	mov	rsi,QWORD[16+rsp]
+	DB	0F3h,0C3h		;repret
+$L$SEH_end_ecp_nistz256_point_add_affine:

third_party/boringssl/win-x86_64/crypto/rc4/rc4-md5-x86_64.asm

diff --git a/third_party/boringssl/win-x86_64/crypto/rc4/rc4-md5-x86_64.asm b/third_party/boringssl/win-x86_64/crypto/rc4/rc4-md5-x86_64.asm
deleted file mode 100644
index f1ea965..0000000
--- a/third_party/boringssl/win-x86_64/crypto/rc4/rc4-md5-x86_64.asm
+++ /dev/null
@@ -1,1372 +0,0 @@
-default	rel
-%define XMMWORD
-%define YMMWORD
-%define ZMMWORD
-section	.text code align=64
-
-ALIGN	16
-
-global	rc4_md5_enc
-
-rc4_md5_enc:
-	mov	QWORD[8+rsp],rdi	;WIN64 prologue
-	mov	QWORD[16+rsp],rsi
-	mov	rax,rsp
-$L$SEH_begin_rc4_md5_enc:
-	mov	rdi,rcx
-	mov	rsi,rdx
-	mov	rdx,r8
-	mov	rcx,r9
-	mov	r8,QWORD[40+rsp]
-	mov	r9,QWORD[48+rsp]
-
-
-	cmp	r9,0
-	je	NEAR $L$abort
-	push	rbx
-	push	rbp
-	push	r12
-	push	r13
-	push	r14
-	push	r15
-	sub	rsp,40
-$L$body:
-	mov	r11,rcx
-	mov	r12,r9
-	mov	r13,rsi
-	mov	r14,rdx
-	mov	r15,r8
-	xor	rbp,rbp
-	xor	rcx,rcx
-
-	lea	rdi,[8+rdi]
-	mov	bpl,BYTE[((-8))+rdi]
-	mov	cl,BYTE[((-4))+rdi]
-
-	inc	bpl
-	sub	r14,r13
-	mov	eax,DWORD[rbp*4+rdi]
-	add	cl,al
-	lea	rsi,[rbp*4+rdi]
-	shl	r12,6
-	add	r12,r15
-	mov	QWORD[16+rsp],r12
-
-	mov	QWORD[24+rsp],r11
-	mov	r8d,DWORD[r11]
-	mov	r9d,DWORD[4+r11]
-	mov	r10d,DWORD[8+r11]
-	mov	r11d,DWORD[12+r11]
-	jmp	NEAR $L$oop
-
-ALIGN	16
-$L$oop:
-	mov	DWORD[rsp],r8d
-	mov	DWORD[4+rsp],r9d
-	mov	DWORD[8+rsp],r10d
-	mov	r12d,r11d
-	mov	DWORD[12+rsp],r11d
-	pxor	xmm0,xmm0
-	mov	edx,DWORD[rcx*4+rdi]
-	xor	r12d,r10d
-	mov	DWORD[rcx*4+rdi],eax
-	and	r12d,r9d
-	add	r8d,DWORD[r15]
-	add	al,dl
-	mov	ebx,DWORD[4+rsi]
-	add	r8d,3614090360
-	xor	r12d,r11d
-	movzx	eax,al
-	mov	DWORD[rsi],edx
-	add	r8d,r12d
-	add	cl,bl
-	rol	r8d,7
-	mov	r12d,r10d
-	movd	xmm0,DWORD[rax*4+rdi]
-
-	add	r8d,r9d
-	pxor	xmm1,xmm1
-	mov	edx,DWORD[rcx*4+rdi]
-	xor	r12d,r9d
-	mov	DWORD[rcx*4+rdi],ebx
-	and	r12d,r8d
-	add	r11d,DWORD[4+r15]
-	add	bl,dl
-	mov	eax,DWORD[8+rsi]
-	add	r11d,3905402710
-	xor	r12d,r10d
-	movzx	ebx,bl
-	mov	DWORD[4+rsi],edx
-	add	r11d,r12d
-	add	cl,al
-	rol	r11d,12
-	mov	r12d,r9d
-	movd	xmm1,DWORD[rbx*4+rdi]
-
-	add	r11d,r8d
-	mov	edx,DWORD[rcx*4+rdi]
-	xor	r12d,r8d
-	mov	DWORD[rcx*4+rdi],eax
-	and	r12d,r11d
-	add	r10d,DWORD[8+r15]
-	add	al,dl
-	mov	ebx,DWORD[12+rsi]
-	add	r10d,606105819
-	xor	r12d,r9d
-	movzx	eax,al
-	mov	DWORD[8+rsi],edx
-	add	r10d,r12d
-	add	cl,bl
-	rol	r10d,17
-	mov	r12d,r8d
-	pinsrw	xmm0,WORD[rax*4+rdi],1
-
-	add	r10d,r11d
-	mov	edx,DWORD[rcx*4+rdi]
-	xor	r12d,r11d
-	mov	DWORD[rcx*4+rdi],ebx
-	and	r12d,r10d
-	add	r9d,DWORD[12+r15]
-	add	bl,dl
-	mov	eax,DWORD[16+rsi]
-	add	r9d,3250441966
-	xor	r12d,r8d
-	movzx	ebx,bl
-	mov	DWORD[12+rsi],edx
-	add	r9d,r12d
-	add	cl,al
-	rol	r9d,22
-	mov	r12d,r11d
-	pinsrw	xmm1,WORD[rbx*4+rdi],1
-
-	add	r9d,r10d
-	mov	edx,DWORD[rcx*4+rdi]
-	xor	r12d,r10d
-	mov	DWORD[rcx*4+rdi],eax
-	and	r12d,r9d
-	add	r8d,DWORD[16+r15]
-	add	al,dl
-	mov	ebx,DWORD[20+rsi]
-	add	r8d,4118548399
-	xor	r12d,r11d
-	movzx	eax,al
-	mov	DWORD[16+rsi],edx
-	add	r8d,r12d
-	add	cl,bl
-	rol	r8d,7
-	mov	r12d,r10d
-	pinsrw	xmm0,WORD[rax*4+rdi],2
-
-	add	r8d,r9d
-	mov	edx,DWORD[rcx*4+rdi]
-	xor	r12d,r9d
-	mov	DWORD[rcx*4+rdi],ebx
-	and	r12d,r8d
-	add	r11d,DWORD[20+r15]
-	add	bl,dl
-	mov	eax,DWORD[24+rsi]
-	add	r11d,1200080426
-	xor	r12d,r10d
-	movzx	ebx,bl
-	mov	DWORD[20+rsi],edx
-	add	r11d,r12d
-	add	cl,al
-	rol	r11d,12
-	mov	r12d,r9d
-	pinsrw	xmm1,WORD[rbx*4+rdi],2
-
-	add	r11d,r8d
-	mov	edx,DWORD[rcx*4+rdi]
-	xor	r12d,r8d
-	mov	DWORD[rcx*4+rdi],eax
-	and	r12d,r11d
-	add	r10d,DWORD[24+r15]
-	add	al,dl
-	mov	ebx,DWORD[28+rsi]
-	add	r10d,2821735955
-	xor	r12d,r9d
-	movzx	eax,al
-	mov	DWORD[24+rsi],edx
-	add	r10d,r12d
-	add	cl,bl
-	rol	r10d,17
-	mov	r12d,r8d
-	pinsrw	xmm0,WORD[rax*4+rdi],3
-
-	add	r10d,r11d
-	mov	edx,DWORD[rcx*4+rdi]
-	xor	r12d,r11d
-	mov	DWORD[rcx*4+rdi],ebx
-	and	r12d,r10d
-	add	r9d,DWORD[28+r15]
-	add	bl,dl
-	mov	eax,DWORD[32+rsi]
-	add	r9d,4249261313
-	xor	r12d,r8d
-	movzx	ebx,bl
-	mov	DWORD[28+rsi],edx
-	add	r9d,r12d
-	add	cl,al
-	rol	r9d,22
-	mov	r12d,r11d
-	pinsrw	xmm1,WORD[rbx*4+rdi],3
-
-	add	r9d,r10d
-	mov	edx,DWORD[rcx*4+rdi]
-	xor	r12d,r10d
-	mov	DWORD[rcx*4+rdi],eax
-	and	r12d,r9d
-	add	r8d,DWORD[32+r15]
-	add	al,dl
-	mov	ebx,DWORD[36+rsi]
-	add	r8d,1770035416
-	xor	r12d,r11d
-	movzx	eax,al
-	mov	DWORD[32+rsi],edx
-	add	r8d,r12d
-	add	cl,bl
-	rol	r8d,7
-	mov	r12d,r10d
-	pinsrw	xmm0,WORD[rax*4+rdi],4
-
-	add	r8d,r9d
-	mov	edx,DWORD[rcx*4+rdi]
-	xor	r12d,r9d
-	mov	DWORD[rcx*4+rdi],ebx
-	and	r12d,r8d
-	add	r11d,DWORD[36+r15]
-	add	bl,dl
-	mov	eax,DWORD[40+rsi]
-	add	r11d,2336552879
-	xor	r12d,r10d
-	movzx	ebx,bl
-	mov	DWORD[36+rsi],edx
-	add	r11d,r12d
-	add	cl,al
-	rol	r11d,12
-	mov	r12d,r9d
-	pinsrw	xmm1,WORD[rbx*4+rdi],4
-
-	add	r11d,r8d
-	mov	edx,DWORD[rcx*4+rdi]
-	xor	r12d,r8d
-	mov	DWORD[rcx*4+rdi],eax
-	and	r12d,r11d
-	add	r10d,DWORD[40+r15]
-	add	al,dl
-	mov	ebx,DWORD[44+rsi]
-	add	r10d,4294925233
-	xor	r12d,r9d
-	movzx	eax,al
-	mov	DWORD[40+rsi],edx
-	add	r10d,r12d
-	add	cl,bl
-	rol	r10d,17
-	mov	r12d,r8d
-	pinsrw	xmm0,WORD[rax*4+rdi],5
-
-	add	r10d,r11d
-	mov	edx,DWORD[rcx*4+rdi]
-	xor	r12d,r11d
-	mov	DWORD[rcx*4+rdi],ebx
-	and	r12d,r10d
-	add	r9d,DWORD[44+r15]
-	add	bl,dl
-	mov	eax,DWORD[48+rsi]
-	add	r9d,2304563134
-	xor	r12d,r8d
-	movzx	ebx,bl
-	mov	DWORD[44+rsi],edx
-	add	r9d,r12d
-	add	cl,al
-	rol	r9d,22
-	mov	r12d,r11d
-	pinsrw	xmm1,WORD[rbx*4+rdi],5
-
-	add	r9d,r10d
-	mov	edx,DWORD[rcx*4+rdi]
-	xor	r12d,r10d
-	mov	DWORD[rcx*4+rdi],eax
-	and	r12d,r9d
-	add	r8d,DWORD[48+r15]
-	add	al,dl
-	mov	ebx,DWORD[52+rsi]
-	add	r8d,1804603682
-	xor	r12d,r11d
-	movzx	eax,al
-	mov	DWORD[48+rsi],edx
-	add	r8d,r12d
-	add	cl,bl
-	rol	r8d,7
-	mov	r12d,r10d
-	pinsrw	xmm0,WORD[rax*4+rdi],6
-
-	add	r8d,r9d
-	mov	edx,DWORD[rcx*4+rdi]
-	xor	r12d,r9d
-	mov	DWORD[rcx*4+rdi],ebx
-	and	r12d,r8d
-	add	r11d,DWORD[52+r15]
-	add	bl,dl
-	mov	eax,DWORD[56+rsi]
-	add	r11d,4254626195
-	xor	r12d,r10d
-	movzx	ebx,bl
-	mov	DWORD[52+rsi],edx
-	add	r11d,r12d
-	add	cl,al
-	rol	r11d,12
-	mov	r12d,r9d
-	pinsrw	xmm1,WORD[rbx*4+rdi],6
-
-	add	r11d,r8d
-	mov	edx,DWORD[rcx*4+rdi]
-	xor	r12d,r8d
-	mov	DWORD[rcx*4+rdi],eax
-	and	r12d,r11d
-	add	r10d,DWORD[56+r15]
-	add	al,dl
-	mov	ebx,DWORD[60+rsi]
-	add	r10d,2792965006
-	xor	r12d,r9d
-	movzx	eax,al
-	mov	DWORD[56+rsi],edx
-	add	r10d,r12d
-	add	cl,bl
-	rol	r10d,17
-	mov	r12d,r8d
-	pinsrw	xmm0,WORD[rax*4+rdi],7
-
-	add	r10d,r11d
-	movdqu	xmm2,XMMWORD[r13]
-	mov	edx,DWORD[rcx*4+rdi]
-	xor	r12d,r11d
-	mov	DWORD[rcx*4+rdi],ebx
-	and	r12d,r10d
-	add	r9d,DWORD[60+r15]
-	add	bl,dl
-	mov	eax,DWORD[64+rsi]
-	add	r9d,1236535329
-	xor	r12d,r8d
-	movzx	ebx,bl
-	mov	DWORD[60+rsi],edx
-	add	r9d,r12d
-	add	cl,al
-	rol	r9d,22
-	mov	r12d,r10d
-	pinsrw	xmm1,WORD[rbx*4+rdi],7
-
-	add	r9d,r10d
-	psllq	xmm1,8
-	pxor	xmm2,xmm0
-	pxor	xmm2,xmm1
-	pxor	xmm0,xmm0
-	mov	edx,DWORD[rcx*4+rdi]
-	xor	r12d,r9d
-	mov	DWORD[rcx*4+rdi],eax
-	and	r12d,r11d
-	add	r8d,DWORD[4+r15]
-	add	al,dl
-	mov	ebx,DWORD[68+rsi]
-	add	r8d,4129170786
-	xor	r12d,r10d
-	movzx	eax,al
-	mov	DWORD[64+rsi],edx
-	add	r8d,r12d
-	add	cl,bl
-	rol	r8d,5
-	mov	r12d,r9d
-	movd	xmm0,DWORD[rax*4+rdi]
-
-	add	r8d,r9d
-	pxor	xmm1,xmm1
-	mov	edx,DWORD[rcx*4+rdi]
-	xor	r12d,r8d
-	mov	DWORD[rcx*4+rdi],ebx
-	and	r12d,r10d
-	add	r11d,DWORD[24+r15]
-	add	bl,dl
-	mov	eax,DWORD[72+rsi]
-	add	r11d,3225465664
-	xor	r12d,r9d
-	movzx	ebx,bl
-	mov	DWORD[68+rsi],edx
-	add	r11d,r12d
-	add	cl,al
-	rol	r11d,9
-	mov	r12d,r8d
-	movd	xmm1,DWORD[rbx*4+rdi]
-
-	add	r11d,r8d
-	mov	edx,DWORD[rcx*4+rdi]
-	xor	r12d,r11d
-	mov	DWORD[rcx*4+rdi],eax
-	and	r12d,r9d
-	add	r10d,DWORD[44+r15]
-	add	al,dl
-	mov	ebx,DWORD[76+rsi]
-	add	r10d,643717713
-	xor	r12d,r8d
-	movzx	eax,al
-	mov	DWORD[72+rsi],edx
-	add	r10d,r12d
-	add	cl,bl
-	rol	r10d,14
-	mov	r12d,r11d
-	pinsrw	xmm0,WORD[rax*4+rdi],1
-
-	add	r10d,r11d
-	mov	edx,DWORD[rcx*4+rdi]
-	xor	r12d,r10d
-	mov	DWORD[rcx*4+rdi],ebx
-	and	r12d,r8d
-	add	r9d,DWORD[r15]
-	add	bl,dl
-	mov	eax,DWORD[80+rsi]
-	add	r9d,3921069994
-	xor	r12d,r11d
-	movzx	ebx,bl
-	mov	DWORD[76+rsi],edx
-	add	r9d,r12d
-	add	cl,al
-	rol	r9d,20
-	mov	r12d,r10d
-	pinsrw	xmm1,WORD[rbx*4+rdi],1
-
-	add	r9d,r10d
-	mov	edx,DWORD[rcx*4+rdi]
-	xor	r12d,r9d
-	mov	DWORD[rcx*4+rdi],eax
-	and	r12d,r11d
-	add	r8d,DWORD[20+r15]
-	add	al,dl
-	mov	ebx,DWORD[84+rsi]
-	add	r8d,3593408605
-	xor	r12d,r10d
-	movzx	eax,al
-	mov	DWORD[80+rsi],edx
-	add	r8d,r12d
-	add	cl,bl
-	rol	r8d,5
-	mov	r12d,r9d
-	pinsrw	xmm0,WORD[rax*4+rdi],2
-
-	add	r8d,r9d
-	mov	edx,DWORD[rcx*4+rdi]
-	xor	r12d,r8d
-	mov	DWORD[rcx*4+rdi],ebx
-	and	r12d,r10d
-	add	r11d,DWORD[40+r15]
-	add	bl,dl
-	mov	eax,DWORD[88+rsi]
-	add	r11d,38016083
-	xor	r12d,r9d
-	movzx	ebx,bl
-	mov	DWORD[84+rsi],edx
-	add	r11d,r12d
-	add	cl,al
-	rol	r11d,9
-	mov	r12d,r8d
-	pinsrw	xmm1,WORD[rbx*4+rdi],2
-
-	add	r11d,r8d
-	mov	edx,DWORD[rcx*4+rdi]
-	xor	r12d,r11d
-	mov	DWORD[rcx*4+rdi],eax
-	and	r12d,r9d
-	add	r10d,DWORD[60+r15]
-	add	al,dl
-	mov	ebx,DWORD[92+rsi]
-	add	r10d,3634488961
-	xor	r12d,r8d
-	movzx	eax,al
-	mov	DWORD[88+rsi],edx
-	add	r10d,r12d
-	add	cl,bl
-	rol	r10d,14
-	mov	r12d,r11d
-	pinsrw	xmm0,WORD[rax*4+rdi],3
-
-	add	r10d,r11d
-	mov	edx,DWORD[rcx*4+rdi]
-	xor	r12d,r10d
-	mov	DWORD[rcx*4+rdi],ebx
-	and	r12d,r8d
-	add	r9d,DWORD[16+r15]
-	add	bl,dl
-	mov	eax,DWORD[96+rsi]
-	add	r9d,3889429448
-	xor	r12d,r11d
-	movzx	ebx,bl
-	mov	DWORD[92+rsi],edx
-	add	r9d,r12d
-	add	cl,al
-	rol	r9d,20
-	mov	r12d,r10d
-	pinsrw	xmm1,WORD[rbx*4+rdi],3
-
-	add	r9d,r10d
-	mov	edx,DWORD[rcx*4+rdi]
-	xor	r12d,r9d
-	mov	DWORD[rcx*4+rdi],eax
-	and	r12d,r11d
-	add	r8d,DWORD[36+r15]
-	add	al,dl
-	mov	ebx,DWORD[100+rsi]
-	add	r8d,568446438
-	xor	r12d,r10d
-	movzx	eax,al
-	mov	DWORD[96+rsi],edx
-	add	r8d,r12d
-	add	cl,bl
-	rol	r8d,5
-	mov	r12d,r9d
-	pinsrw	xmm0,WORD[rax*4+rdi],4
-
-	add	r8d,r9d
-	mov	edx,DWORD[rcx*4+rdi]
-	xor	r12d,r8d
-	mov	DWORD[rcx*4+rdi],ebx
-	and	r12d,r10d
-	add	r11d,DWORD[56+r15]
-	add	bl,dl
-	mov	eax,DWORD[104+rsi]
-	add	r11d,3275163606
-	xor	r12d,r9d
-	movzx	ebx,bl
-	mov	DWORD[100+rsi],edx
-	add	r11d,r12d
-	add	cl,al
-	rol	r11d,9
-	mov	r12d,r8d
-	pinsrw	xmm1,WORD[rbx*4+rdi],4
-
-	add	r11d,r8d
-	mov	edx,DWORD[rcx*4+rdi]
-	xor	r12d,r11d
-	mov	DWORD[rcx*4+rdi],eax
-	and	r12d,r9d
-	add	r10d,DWORD[12+r15]
-	add	al,dl
-	mov	ebx,DWORD[108+rsi]
-	add	r10d,4107603335
-	xor	r12d,r8d
-	movzx	eax,al
-	mov	DWORD[104+rsi],edx
-	add	r10d,r12d
-	add	cl,bl
-	rol	r10d,14
-	mov	r12d,r11d
-	pinsrw	xmm0,WORD[rax*4+rdi],5
-
-	add	r10d,r11d
-	mov	edx,DWORD[rcx*4+rdi]
-	xor	r12d,r10d
-	mov	DWORD[rcx*4+rdi],ebx
-	and	r12d,r8d
-	add	r9d,DWORD[32+r15]
-	add	bl,dl
-	mov	eax,DWORD[112+rsi]
-	add	r9d,1163531501
-	xor	r12d,r11d
-	movzx	ebx,bl
-	mov	DWORD[108+rsi],edx
-	add	r9d,r12d
-	add	cl,al
-	rol	r9d,20
-	mov	r12d,r10d
-	pinsrw	xmm1,WORD[rbx*4+rdi],5
-
-	add	r9d,r10d
-	mov	edx,DWORD[rcx*4+rdi]
-	xor	r12d,r9d
-	mov	DWORD[rcx*4+rdi],eax
-	and	r12d,r11d
-	add	r8d,DWORD[52+r15]
-	add	al,dl
-	mov	ebx,DWORD[116+rsi]
-	add	r8d,2850285829
-	xor	r12d,r10d
-	movzx	eax,al
-	mov	DWORD[112+rsi],edx
-	add	r8d,r12d
-	add	cl,bl
-	rol	r8d,5
-	mov	r12d,r9d
-	pinsrw	xmm0,WORD[rax*4+rdi],6
-
-	add	r8d,r9d
-	mov	edx,DWORD[rcx*4+rdi]
-	xor	r12d,r8d
-	mov	DWORD[rcx*4+rdi],ebx
-	and	r12d,r10d
-	add	r11d,DWORD[8+r15]
-	add	bl,dl
-	mov	eax,DWORD[120+rsi]
-	add	r11d,4243563512
-	xor	r12d,r9d
-	movzx	ebx,bl
-	mov	DWORD[116+rsi],edx
-	add	r11d,r12d
-	add	cl,al
-	rol	r11d,9
-	mov	r12d,r8d
-	pinsrw	xmm1,WORD[rbx*4+rdi],6
-
-	add	r11d,r8d
-	mov	edx,DWORD[rcx*4+rdi]
-	xor	r12d,r11d
-	mov	DWORD[rcx*4+rdi],eax
-	and	r12d,r9d
-	add	r10d,DWORD[28+r15]
-	add	al,dl
-	mov	ebx,DWORD[124+rsi]
-	add	r10d,1735328473
-	xor	r12d,r8d
-	movzx	eax,al
-	mov	DWORD[120+rsi],edx
-	add	r10d,r12d
-	add	cl,bl
-	rol	r10d,14
-	mov	r12d,r11d
-	pinsrw	xmm0,WORD[rax*4+rdi],7
-
-	add	r10d,r11d
-	movdqu	xmm3,XMMWORD[16+r13]
-	add	bpl,32
-	mov	edx,DWORD[rcx*4+rdi]
-	xor	r12d,r10d
-	mov	DWORD[rcx*4+rdi],ebx
-	and	r12d,r8d
-	add	r9d,DWORD[48+r15]
-	add	bl,dl
-	mov	eax,DWORD[rbp*4+rdi]
-	add	r9d,2368359562
-	xor	r12d,r11d
-	movzx	ebx,bl
-	mov	DWORD[124+rsi],edx
-	add	r9d,r12d
-	add	cl,al
-	rol	r9d,20
-	mov	r12d,r11d
-	pinsrw	xmm1,WORD[rbx*4+rdi],7
-
-	add	r9d,r10d
-	mov	rsi,rcx
-	xor	rcx,rcx
-	mov	cl,sil
-	lea	rsi,[rbp*4+rdi]
-	psllq	xmm1,8
-	pxor	xmm3,xmm0
-	pxor	xmm3,xmm1
-	pxor	xmm0,xmm0
-	mov	edx,DWORD[rcx*4+rdi]
-	xor	r12d,r10d
-	mov	DWORD[rcx*4+rdi],eax
-	xor	r12d,r9d
-	add	r8d,DWORD[20+r15]
-	add	al,dl
-	mov	ebx,DWORD[4+rsi]
-	add	r8d,4294588738
-	movzx	eax,al
-	add	r8d,r12d
-	mov	DWORD[rsi],edx
-	add	cl,bl
-	rol	r8d,4
-	mov	r12d,r10d
-	movd	xmm0,DWORD[rax*4+rdi]
-
-	add	r8d,r9d
-	pxor	xmm1,xmm1
-	mov	edx,DWORD[rcx*4+rdi]
-	xor	r12d,r9d
-	mov	DWORD[rcx*4+rdi],ebx
-	xor	r12d,r8d
-	add	r11d,DWORD[32+r15]
-	add	bl,dl
-	mov	eax,DWORD[8+rsi]
-	add	r11d,2272392833
-	movzx	ebx,bl
-	add	r11d,r12d
-	mov	DWORD[4+rsi],edx
-	add	cl,al
-	rol	r11d,11
-	mov	r12d,r9d
-	movd	xmm1,DWORD[rbx*4+rdi]
-
-	add	r11d,r8d
-	mov	edx,DWORD[rcx*4+rdi]
-	xor	r12d,r8d
-	mov	DWORD[rcx*4+rdi],eax
-	xor	r12d,r11d
-	add	r10d,DWORD[44+r15]
-	add	al,dl
-	mov	ebx,DWORD[12+rsi]
-	add	r10d,1839030562
-	movzx	eax,al
-	add	r10d,r12d
-	mov	DWORD[8+rsi],edx
-	add	cl,bl
-	rol	r10d,16
-	mov	r12d,r8d
-	pinsrw	xmm0,WORD[rax*4+rdi],1
-
-	add	r10d,r11d
-	mov	edx,DWORD[rcx*4+rdi]
-	xor	r12d,r11d
-	mov	DWORD[rcx*4+rdi],ebx
-	xor	r12d,r10d
-	add	r9d,DWORD[56+r15]
-	add	bl,dl
-	mov	eax,DWORD[16+rsi]
-	add	r9d,4259657740
-	movzx	ebx,bl
-	add	r9d,r12d
-	mov	DWORD[12+rsi],edx
-	add	cl,al
-	rol	r9d,23
-	mov	r12d,r11d
-	pinsrw	xmm1,WORD[rbx*4+rdi],1
-
-	add	r9d,r10d
-	mov	edx,DWORD[rcx*4+rdi]
-	xor	r12d,r10d
-	mov	DWORD[rcx*4+rdi],eax
-	xor	r12d,r9d
-	add	r8d,DWORD[4+r15]
-	add	al,dl
-	mov	ebx,DWORD[20+rsi]
-	add	r8d,2763975236
-	movzx	eax,al
-	add	r8d,r12d
-	mov	DWORD[16+rsi],edx
-	add	cl,bl
-	rol	r8d,4
-	mov	r12d,r10d
-	pinsrw	xmm0,WORD[rax*4+rdi],2
-
-	add	r8d,r9d
-	mov	edx,DWORD[rcx*4+rdi]
-	xor	r12d,r9d
-	mov	DWORD[rcx*4+rdi],ebx
-	xor	r12d,r8d
-	add	r11d,DWORD[16+r15]
-	add	bl,dl
-	mov	eax,DWORD[24+rsi]
-	add	r11d,1272893353
-	movzx	ebx,bl
-	add	r11d,r12d
-	mov	DWORD[20+rsi],edx
-	add	cl,al
-	rol	r11d,11
-	mov	r12d,r9d
-	pinsrw	xmm1,WORD[rbx*4+rdi],2
-
-	add	r11d,r8d
-	mov	edx,DWORD[rcx*4+rdi]
-	xor	r12d,r8d
-	mov	DWORD[rcx*4+rdi],eax
-	xor	r12d,r11d
-	add	r10d,DWORD[28+r15]
-	add	al,dl
-	mov	ebx,DWORD[28+rsi]
-	add	r10d,4139469664
-	movzx	eax,al
-	add	r10d,r12d
-	mov	DWORD[24+rsi],edx
-	add	cl,bl
-	rol	r10d,16
-	mov	r12d,r8d
-	pinsrw	xmm0,WORD[rax*4+rdi],3
-
-	add	r10d,r11d
-	mov	edx,DWORD[rcx*4+rdi]
-	xor	r12d,r11d
-	mov	DWORD[rcx*4+rdi],ebx
-	xor	r12d,r10d
-	add	r9d,DWORD[40+r15]
-	add	bl,dl
-	mov	eax,DWORD[32+rsi]
-	add	r9d,3200236656
-	movzx	ebx,bl
-	add	r9d,r12d
-	mov	DWORD[28+rsi],edx
-	add	cl,al
-	rol	r9d,23
-	mov	r12d,r11d
-	pinsrw	xmm1,WORD[rbx*4+rdi],3
-
-	add	r9d,r10d
-	mov	edx,DWORD[rcx*4+rdi]
-	xor	r12d,r10d
-	mov	DWORD[rcx*4+rdi],eax
-	xor	r12d,r9d
-	add	r8d,DWORD[52+r15]
-	add	al,dl
-	mov	ebx,DWORD[36+rsi]
-	add	r8d,681279174
-	movzx	eax,al
-	add	r8d,r12d
-	mov	DWORD[32+rsi],edx
-	add	cl,bl
-	rol	r8d,4
-	mov	r12d,r10d
-	pinsrw	xmm0,WORD[rax*4+rdi],4
-
-	add	r8d,r9d
-	mov	edx,DWORD[rcx*4+rdi]
-	xor	r12d,r9d
-	mov	DWORD[rcx*4+rdi],ebx
-	xor	r12d,r8d
-	add	r11d,DWORD[r15]
-	add	bl,dl
-	mov	eax,DWORD[40+rsi]
-	add	r11d,3936430074
-	movzx	ebx,bl
-	add	r11d,r12d
-	mov	DWORD[36+rsi],edx
-	add	cl,al
-	rol	r11d,11
-	mov	r12d,r9d
-	pinsrw	xmm1,WORD[rbx*4+rdi],4
-
-	add	r11d,r8d
-	mov	edx,DWORD[rcx*4+rdi]
-	xor	r12d,r8d
-	mov	DWORD[rcx*4+rdi],eax
-	xor	r12d,r11d
-	add	r10d,DWORD[12+r15]
-	add	al,dl
-	mov	ebx,DWORD[44+rsi]
-	add	r10d,3572445317
-	movzx	eax,al
-	add	r10d,r12d
-	mov	DWORD[40+rsi],edx
-	add	cl,bl
-	rol	r10d,16
-	mov	r12d,r8d
-	pinsrw	xmm0,WORD[rax*4+rdi],5
-
-	add	r10d,r11d
-	mov	edx,DWORD[rcx*4+rdi]
-	xor	r12d,r11d
-	mov	DWORD[rcx*4+rdi],ebx
-	xor	r12d,r10d
-	add	r9d,DWORD[24+r15]
-	add	bl,dl
-	mov	eax,DWORD[48+rsi]
-	add	r9d,76029189
-	movzx	ebx,bl
-	add	r9d,r12d
-	mov	DWORD[44+rsi],edx
-	add	cl,al
-	rol	r9d,23
-	mov	r12d,r11d
-	pinsrw	xmm1,WORD[rbx*4+rdi],5
-
-	add	r9d,r10d
-	mov	edx,DWORD[rcx*4+rdi]
-	xor	r12d,r10d
-	mov	DWORD[rcx*4+rdi],eax
-	xor	r12d,r9d
-	add	r8d,DWORD[36+r15]
-	add	al,dl
-	mov	ebx,DWORD[52+rsi]
-	add	r8d,3654602809
-	movzx	eax,al
-	add	r8d,r12d
-	mov	DWORD[48+rsi],edx
-	add	cl,bl
-	rol	r8d,4
-	mov	r12d,r10d
-	pinsrw	xmm0,WORD[rax*4+rdi],6
-
-	add	r8d,r9d
-	mov	edx,DWORD[rcx*4+rdi]
-	xor	r12d,r9d
-	mov	DWORD[rcx*4+rdi],ebx
-	xor	r12d,r8d
-	add	r11d,DWORD[48+r15]
-	add	bl,dl
-	mov	eax,DWORD[56+rsi]
-	add	r11d,3873151461
-	movzx	ebx,bl
-	add	r11d,r12d
-	mov	DWORD[52+rsi],edx
-	add	cl,al
-	rol	r11d,11
-	mov	r12d,r9d
-	pinsrw	xmm1,WORD[rbx*4+rdi],6
-
-	add	r11d,r8d
-	mov	edx,DWORD[rcx*4+rdi]
-	xor	r12d,r8d
-	mov	DWORD[rcx*4+rdi],eax
-	xor	r12d,r11d
-	add	r10d,DWORD[60+r15]
-	add	al,dl
-	mov	ebx,DWORD[60+rsi]
-	add	r10d,530742520
-	movzx	eax,al
-	add	r10d,r12d
-	mov	DWORD[56+rsi],edx
-	add	cl,bl
-	rol	r10d,16
-	mov	r12d,r8d
-	pinsrw	xmm0,WORD[rax*4+rdi],7
-
-	add	r10d,r11d
-	movdqu	xmm4,XMMWORD[32+r13]
-	mov	edx,DWORD[rcx*4+rdi]
-	xor	r12d,r11d
-	mov	DWORD[rcx*4+rdi],ebx
-	xor	r12d,r10d
-	add	r9d,DWORD[8+r15]
-	add	bl,dl
-	mov	eax,DWORD[64+rsi]
-	add	r9d,3299628645
-	movzx	ebx,bl
-	add	r9d,r12d
-	mov	DWORD[60+rsi],edx
-	add	cl,al
-	rol	r9d,23
-	mov	r12d,-1
-	pinsrw	xmm1,WORD[rbx*4+rdi],7
-
-	add	r9d,r10d
-	psllq	xmm1,8
-	pxor	xmm4,xmm0
-	pxor	xmm4,xmm1
-	pxor	xmm0,xmm0
-	mov	edx,DWORD[rcx*4+rdi]
-	xor	r12d,r11d
-	mov	DWORD[rcx*4+rdi],eax
-	or	r12d,r9d
-	add	r8d,DWORD[r15]
-	add	al,dl
-	mov	ebx,DWORD[68+rsi]
-	add	r8d,4096336452
-	movzx	eax,al
-	xor	r12d,r10d
-	mov	DWORD[64+rsi],edx
-	add	r8d,r12d
-	add	cl,bl
-	rol	r8d,6
-	mov	r12d,-1
-	movd	xmm0,DWORD[rax*4+rdi]
-
-	add	r8d,r9d
-	pxor	xmm1,xmm1
-	mov	edx,DWORD[rcx*4+rdi]
-	xor	r12d,r10d
-	mov	DWORD[rcx*4+rdi],ebx
-	or	r12d,r8d
-	add	r11d,DWORD[28+r15]
-	add	bl,dl
-	mov	eax,DWORD[72+rsi]
-	add	r11d,1126891415
-	movzx	ebx,bl
-	xor	r12d,r9d
-	mov	DWORD[68+rsi],edx
-	add	r11d,r12d
-	add	cl,al
-	rol	r11d,10
-	mov	r12d,-1
-	movd	xmm1,DWORD[rbx*4+rdi]
-
-	add	r11d,r8d
-	mov	edx,DWORD[rcx*4+rdi]
-	xor	r12d,r9d
-	mov	DWORD[rcx*4+rdi],eax
-	or	r12d,r11d
-	add	r10d,DWORD[56+r15]
-	add	al,dl
-	mov	ebx,DWORD[76+rsi]
-	add	r10d,2878612391
-	movzx	eax,al
-	xor	r12d,r8d
-	mov	DWORD[72+rsi],edx
-	add	r10d,r12d
-	add	cl,bl
-	rol	r10d,15
-	mov	r12d,-1
-	pinsrw	xmm0,WORD[rax*4+rdi],1
-
-	add	r10d,r11d
-	mov	edx,DWORD[rcx*4+rdi]
-	xor	r12d,r8d
-	mov	DWORD[rcx*4+rdi],ebx
-	or	r12d,r10d
-	add	r9d,DWORD[20+r15]
-	add	bl,dl
-	mov	eax,DWORD[80+rsi]
-	add	r9d,4237533241
-	movzx	ebx,bl
-	xor	r12d,r11d
-	mov	DWORD[76+rsi],edx
-	add	r9d,r12d
-	add	cl,al
-	rol	r9d,21
-	mov	r12d,-1
-	pinsrw	xmm1,WORD[rbx*4+rdi],1
-
-	add	r9d,r10d
-	mov	edx,DWORD[rcx*4+rdi]
-	xor	r12d,r11d
-	mov	DWORD[rcx*4+rdi],eax
-	or	r12d,r9d
-	add	r8d,DWORD[48+r15]
-	add	al,dl
-	mov	ebx,DWORD[84+rsi]
-	add	r8d,1700485571
-	movzx	eax,al
-	xor	r12d,r10d
-	mov	DWORD[80+rsi],edx
-	add	r8d,r12d
-	add	cl,bl
-	rol	r8d,6
-	mov	r12d,-1
-	pinsrw	xmm0,WORD[rax*4+rdi],2
-
-	add	r8d,r9d
-	mov	edx,DWORD[rcx*4+rdi]
-	xor	r12d,r10d
-	mov	DWORD[rcx*4+rdi],ebx
-	or	r12d,r8d
-	add	r11d,DWORD[12+r15]
-	add	bl,dl
-	mov	eax,DWORD[88+rsi]
-	add	r11d,2399980690
-	movzx	ebx,bl
-	xor	r12d,r9d
-	mov	DWORD[84+rsi],edx
-	add	r11d,r12d
-	add	cl,al
-	rol	r11d,10
-	mov	r12d,-1
-	pinsrw	xmm1,WORD[rbx*4+rdi],2
-
-	add	r11d,r8d
-	mov	edx,DWORD[rcx*4+rdi]
-	xor	r12d,r9d
-	mov	DWORD[rcx*4+rdi],eax
-	or	r12d,r11d
-	add	r10d,DWORD[40+r15]
-	add	al,dl
-	mov	ebx,DWORD[92+rsi]
-	add	r10d,4293915773
-	movzx	eax,al
-	xor	r12d,r8d
-	mov	DWORD[88+rsi],edx
-	add	r10d,r12d
-	add	cl,bl
-	rol	r10d,15
-	mov	r12d,-1
-	pinsrw	xmm0,WORD[rax*4+rdi],3
-
-	add	r10d,r11d
-	mov	edx,DWORD[rcx*4+rdi]
-	xor	r12d,r8d
-	mov	DWORD[rcx*4+rdi],ebx
-	or	r12d,r10d
-	add	r9d,DWORD[4+r15]
-	add	bl,dl
-	mov	eax,DWORD[96+rsi]
-	add	r9d,2240044497
-	movzx	ebx,bl
-	xor	r12d,r11d
-	mov	DWORD[92+rsi],edx
-	add	r9d,r12d
-	add	cl,al
-	rol	r9d,21
-	mov	r12d,-1
-	pinsrw	xmm1,WORD[rbx*4+rdi],3
-
-	add	r9d,r10d
-	mov	edx,DWORD[rcx*4+rdi]
-	xor	r12d,r11d
-	mov	DWORD[rcx*4+rdi],eax
-	or	r12d,r9d
-	add	r8d,DWORD[32+r15]
-	add	al,dl
-	mov	ebx,DWORD[100+rsi]
-	add	r8d,1873313359
-	movzx	eax,al
-	xor	r12d,r10d
-	mov	DWORD[96+rsi],edx
-	add	r8d,r12d
-	add	cl,bl
-	rol	r8d,6
-	mov	r12d,-1
-	pinsrw	xmm0,WORD[rax*4+rdi],4
-
-	add	r8d,r9d
-	mov	edx,DWORD[rcx*4+rdi]
-	xor	r12d,r10d
-	mov	DWORD[rcx*4+rdi],ebx
-	or	r12d,r8d
-	add	r11d,DWORD[60+r15]
-	add	bl,dl
-	mov	eax,DWORD[104+rsi]
-	add	r11d,4264355552
-	movzx	ebx,bl
-	xor	r12d,r9d
-	mov	DWORD[100+rsi],edx
-	add	r11d,r12d
-	add	cl,al
-	rol	r11d,10
-	mov	r12d,-1
-	pinsrw	xmm1,WORD[rbx*4+rdi],4
-
-	add	r11d,r8d
-	mov	edx,DWORD[rcx*4+rdi]
-	xor	r12d,r9d
-	mov	DWORD[rcx*4+rdi],eax
-	or	r12d,r11d
-	add	r10d,DWORD[24+r15]
-	add	al,dl
-	mov	ebx,DWORD[108+rsi]
-	add	r10d,2734768916
-	movzx	eax,al
-	xor	r12d,r8d
-	mov	DWORD[104+rsi],edx
-	add	r10d,r12d
-	add	cl,bl
-	rol	r10d,15
-	mov	r12d,-1
-	pinsrw	xmm0,WORD[rax*4+rdi],5
-
-	add	r10d,r11d
-	mov	edx,DWORD[rcx*4+rdi]
-	xor	r12d,r8d
-	mov	DWORD[rcx*4+rdi],ebx
-	or	r12d,r10d
-	add	r9d,DWORD[52+r15]
-	add	bl,dl
-	mov	eax,DWORD[112+rsi]
-	add	r9d,1309151649
-	movzx	ebx,bl
-	xor	r12d,r11d
-	mov	DWORD[108+rsi],edx
-	add	r9d,r12d
-	add	cl,al
-	rol	r9d,21
-	mov	r12d,-1
-	pinsrw	xmm1,WORD[rbx*4+rdi],5
-
-	add	r9d,r10d
-	mov	edx,DWORD[rcx*4+rdi]
-	xor	r12d,r11d
-	mov	DWORD[rcx*4+rdi],eax
-	or	r12d,r9d
-	add	r8d,DWORD[16+r15]
-	add	al,dl
-	mov	ebx,DWORD[116+rsi]
-	add	r8d,4149444226
-	movzx	eax,al
-	xor	r12d,r10d
-	mov	DWORD[112+rsi],edx
-	add	r8d,r12d
-	add	cl,bl
-	rol	r8d,6
-	mov	r12d,-1
-	pinsrw	xmm0,WORD[rax*4+rdi],6
-
-	add	r8d,r9d
-	mov	edx,DWORD[rcx*4+rdi]
-	xor	r12d,r10d
-	mov	DWORD[rcx*4+rdi],ebx
-	or	r12d,r8d
-	add	r11d,DWORD[44+r15]
-	add	bl,dl
-	mov	eax,DWORD[120+rsi]
-	add	r11d,3174756917
-	movzx	ebx,bl
-	xor	r12d,r9d
-	mov	DWORD[116+rsi],edx
-	add	r11d,r12d
-	add	cl,al
-	rol	r11d,10
-	mov	r12d,-1
-	pinsrw	xmm1,WORD[rbx*4+rdi],6
-
-	add	r11d,r8d
-	mov	edx,DWORD[rcx*4+rdi]
-	xor	r12d,r9d
-	mov	DWORD[rcx*4+rdi],eax
-	or	r12d,r11d
-	add	r10d,DWORD[8+r15]
-	add	al,dl
-	mov	ebx,DWORD[124+rsi]
-	add	r10d,718787259
-	movzx	eax,al
-	xor	r12d,r8d
-	mov	DWORD[120+rsi],edx
-	add	r10d,r12d
-	add	cl,bl
-	rol	r10d,15
-	mov	r12d,-1
-	pinsrw	xmm0,WORD[rax*4+rdi],7
-
-	add	r10d,r11d
-	movdqu	xmm5,XMMWORD[48+r13]
-	add	bpl,32
-	mov	edx,DWORD[rcx*4+rdi]
-	xor	r12d,r8d
-	mov	DWORD[rcx*4+rdi],ebx
-	or	r12d,r10d
-	add	r9d,DWORD[36+r15]
-	add	bl,dl
-	mov	eax,DWORD[rbp*4+rdi]
-	add	r9d,3951481745
-	movzx	ebx,bl
-	xor	r12d,r11d
-	mov	DWORD[124+rsi],edx
-	add	r9d,r12d
-	add	cl,al
-	rol	r9d,21
-	mov	r12d,-1
-	pinsrw	xmm1,WORD[rbx*4+rdi],7
-
-	add	r9d,r10d
-	mov	rsi,rbp
-	xor	rbp,rbp
-	mov	bpl,sil
-	mov	rsi,rcx
-	xor	rcx,rcx
-	mov	cl,sil
-	lea	rsi,[rbp*4+rdi]
-	psllq	xmm1,8
-	pxor	xmm5,xmm0
-	pxor	xmm5,xmm1
-	add	r8d,DWORD[rsp]
-	add	r9d,DWORD[4+rsp]
-	add	r10d,DWORD[8+rsp]
-	add	r11d,DWORD[12+rsp]
-
-	movdqu	XMMWORD[r13*1+r14],xmm2
-	movdqu	XMMWORD[16+r13*1+r14],xmm3
-	movdqu	XMMWORD[32+r13*1+r14],xmm4
-	movdqu	XMMWORD[48+r13*1+r14],xmm5
-	lea	r15,[64+r15]
-	lea	r13,[64+r13]
-	cmp	r15,QWORD[16+rsp]
-	jb	NEAR $L$oop
-
-	mov	r12,QWORD[24+rsp]
-	sub	cl,al
-	mov	DWORD[r12],r8d
-	mov	DWORD[4+r12],r9d
-	mov	DWORD[8+r12],r10d
-	mov	DWORD[12+r12],r11d
-	sub	bpl,1
-	mov	DWORD[((-8))+rdi],ebp
-	mov	DWORD[((-4))+rdi],ecx
-
-	mov	r15,QWORD[40+rsp]
-	mov	r14,QWORD[48+rsp]
-	mov	r13,QWORD[56+rsp]
-	mov	r12,QWORD[64+rsp]
-	mov	rbp,QWORD[72+rsp]
-	mov	rbx,QWORD[80+rsp]
-	lea	rsp,[88+rsp]
-$L$epilogue:
-$L$abort:
-	mov	rdi,QWORD[8+rsp]	;WIN64 epilogue
-	mov	rsi,QWORD[16+rsp]
-	DB	0F3h,0C3h		;repret
-$L$SEH_end_rc4_md5_enc:
-EXTERN	__imp_RtlVirtualUnwind
-
-ALIGN	16
-se_handler:
-	push	rsi
-	push	rdi
-	push	rbx
-	push	rbp
-	push	r12
-	push	r13
-	push	r14
-	push	r15
-	pushfq
-	sub	rsp,64
-
-	mov	rax,QWORD[120+r8]
-	mov	rbx,QWORD[248+r8]
-
-	lea	r10,[$L$body]
-	cmp	rbx,r10
-	jb	NEAR $L$in_prologue
-
-	mov	rax,QWORD[152+r8]
-
-	lea	r10,[$L$epilogue]
-	cmp	rbx,r10
-	jae	NEAR $L$in_prologue
-
-	mov	r15,QWORD[40+rax]
-	mov	r14,QWORD[48+rax]
-	mov	r13,QWORD[56+rax]
-	mov	r12,QWORD[64+rax]
-	mov	rbp,QWORD[72+rax]
-	mov	rbx,QWORD[80+rax]
-	lea	rax,[88+rax]
-
-	mov	QWORD[144+r8],rbx
-	mov	QWORD[160+r8],rbp
-	mov	QWORD[216+r8],r12
-	mov	QWORD[224+r8],r13
-	mov	QWORD[232+r8],r14
-	mov	QWORD[240+r8],r15
-
-$L$in_prologue:
-	mov	rdi,QWORD[8+rax]
-	mov	rsi,QWORD[16+rax]
-	mov	QWORD[152+r8],rax
-	mov	QWORD[168+r8],rsi
-	mov	QWORD[176+r8],rdi
-
-	mov	rdi,QWORD[40+r9]
-	mov	rsi,r8
-	mov	ecx,154
-	DD	0xa548f3fc
-
-	mov	rsi,r9
-	xor	rcx,rcx
-	mov	rdx,QWORD[8+rsi]
-	mov	r8,QWORD[rsi]
-	mov	r9,QWORD[16+rsi]
-	mov	r10,QWORD[40+rsi]
-	lea	r11,[56+rsi]
-	lea	r12,[24+rsi]
-	mov	QWORD[32+rsp],r10
-	mov	QWORD[40+rsp],r11
-	mov	QWORD[48+rsp],r12
-	mov	QWORD[56+rsp],rcx
-	call	QWORD[__imp_RtlVirtualUnwind]
-
-	mov	eax,1
-	add	rsp,64
-	popfq
-	pop	r15
-	pop	r14
-	pop	r13
-	pop	r12
-	pop	rbp
-	pop	rbx
-	pop	rdi
-	pop	rsi
-	DB	0F3h,0C3h		;repret
-
-
-section	.pdata rdata align=4
-ALIGN	4
-	DD	$L$SEH_begin_rc4_md5_enc wrt ..imagebase
-	DD	$L$SEH_end_rc4_md5_enc wrt ..imagebase
-	DD	$L$SEH_info_rc4_md5_enc wrt ..imagebase
-
-section	.xdata rdata align=8
-ALIGN	8
-$L$SEH_info_rc4_md5_enc:
-DB	9,0,0,0
-	DD	se_handler wrt ..imagebase

third_party/boringssl/win-x86_64/crypto/sha/sha1-x86_64.asm

diff --git a/third_party/boringssl/win-x86_64/crypto/sha/sha1-x86_64.asm b/third_party/boringssl/win-x86_64/crypto/sha/sha1-x86_64.asm
index 0f5361a..168f78d 100644
--- a/third_party/boringssl/win-x86_64/crypto/sha/sha1-x86_64.asm
+++ b/third_party/boringssl/win-x86_64/crypto/sha/sha1-x86_64.asm
@@ -24,6 +24,11 @@
 	mov	r10d,DWORD[((OPENSSL_ia32cap_P+8))]
 	test	r8d,512
 	jz	NEAR $L$ialu
+	and	r8d,268435456
+	and	r9d,1073741824
+	or	r8d,r9d
+	cmp	r8d,1342177280
+	je	NEAR _avx_shortcut
 	jmp	NEAR _ssse3_shortcut
 
 ALIGN	16
@@ -2445,6 +2450,1146 @@
 	mov	rsi,QWORD[16+rsp]
 	DB	0F3h,0C3h		;repret
 $L$SEH_end_sha1_block_data_order_ssse3:
+
+ALIGN	16
+sha1_block_data_order_avx:
+	mov	QWORD[8+rsp],rdi	;WIN64 prologue
+	mov	QWORD[16+rsp],rsi
+	mov	rax,rsp
+$L$SEH_begin_sha1_block_data_order_avx:
+	mov	rdi,rcx
+	mov	rsi,rdx
+	mov	rdx,r8
+
+
+_avx_shortcut:
+	mov	rax,rsp
+	push	rbx
+	push	rbp
+	push	r12
+	push	r13
+	push	r14
+	lea	rsp,[((-160))+rsp]
+	vzeroupper
+	vmovaps	XMMWORD[(-40-96)+rax],xmm6
+	vmovaps	XMMWORD[(-40-80)+rax],xmm7
+	vmovaps	XMMWORD[(-40-64)+rax],xmm8
+	vmovaps	XMMWORD[(-40-48)+rax],xmm9
+	vmovaps	XMMWORD[(-40-32)+rax],xmm10
+	vmovaps	XMMWORD[(-40-16)+rax],xmm11
+$L$prologue_avx:
+	mov	r14,rax
+	and	rsp,-64
+	mov	r8,rdi
+	mov	r9,rsi
+	mov	r10,rdx
+
+	shl	r10,6
+	add	r10,r9
+	lea	r11,[((K_XX_XX+64))]
+
+	mov	eax,DWORD[r8]
+	mov	ebx,DWORD[4+r8]
+	mov	ecx,DWORD[8+r8]
+	mov	edx,DWORD[12+r8]
+	mov	esi,ebx
+	mov	ebp,DWORD[16+r8]
+	mov	edi,ecx
+	xor	edi,edx
+	and	esi,edi
+
+	vmovdqa	xmm6,XMMWORD[64+r11]
+	vmovdqa	xmm11,XMMWORD[((-64))+r11]
+	vmovdqu	xmm0,XMMWORD[r9]
+	vmovdqu	xmm1,XMMWORD[16+r9]
+	vmovdqu	xmm2,XMMWORD[32+r9]
+	vmovdqu	xmm3,XMMWORD[48+r9]
+	vpshufb	xmm0,xmm0,xmm6
+	add	r9,64
+	vpshufb	xmm1,xmm1,xmm6
+	vpshufb	xmm2,xmm2,xmm6
+	vpshufb	xmm3,xmm3,xmm6
+	vpaddd	xmm4,xmm0,xmm11
+	vpaddd	xmm5,xmm1,xmm11
+	vpaddd	xmm6,xmm2,xmm11
+	vmovdqa	XMMWORD[rsp],xmm4
+	vmovdqa	XMMWORD[16+rsp],xmm5
+	vmovdqa	XMMWORD[32+rsp],xmm6
+	jmp	NEAR $L$oop_avx
+ALIGN	16
+$L$oop_avx:
+	shrd	ebx,ebx,2
+	xor	esi,edx
+	vpalignr	xmm4,xmm1,xmm0,8
+	mov	edi,eax
+	add	ebp,DWORD[rsp]
+	vpaddd	xmm9,xmm11,xmm3
+	xor	ebx,ecx
+	shld	eax,eax,5
+	vpsrldq	xmm8,xmm3,4
+	add	ebp,esi
+	and	edi,ebx
+	vpxor	xmm4,xmm4,xmm0
+	xor	ebx,ecx
+	add	ebp,eax
+	vpxor	xmm8,xmm8,xmm2
+	shrd	eax,eax,7
+	xor	edi,ecx
+	mov	esi,ebp
+	add	edx,DWORD[4+rsp]
+	vpxor	xmm4,xmm4,xmm8
+	xor	eax,ebx
+	shld	ebp,ebp,5
+	vmovdqa	XMMWORD[48+rsp],xmm9
+	add	edx,edi
+	and	esi,eax
+	vpsrld	xmm8,xmm4,31
+	xor	eax,ebx
+	add	edx,ebp
+	shrd	ebp,ebp,7
+	xor	esi,ebx
+	vpslldq	xmm10,xmm4,12
+	vpaddd	xmm4,xmm4,xmm4
+	mov	edi,edx
+	add	ecx,DWORD[8+rsp]
+	xor	ebp,eax
+	shld	edx,edx,5
+	vpsrld	xmm9,xmm10,30
+	vpor	xmm4,xmm4,xmm8
+	add	ecx,esi
+	and	edi,ebp
+	xor	ebp,eax
+	add	ecx,edx
+	vpslld	xmm10,xmm10,2
+	vpxor	xmm4,xmm4,xmm9
+	shrd	edx,edx,7
+	xor	edi,eax
+	mov	esi,ecx
+	add	ebx,DWORD[12+rsp]
+	vpxor	xmm4,xmm4,xmm10
+	xor	edx,ebp
+	shld	ecx,ecx,5
+	add	ebx,edi
+	and	esi,edx
+	xor	edx,ebp
+	add	ebx,ecx
+	shrd	ecx,ecx,7
+	xor	esi,ebp
+	vpalignr	xmm5,xmm2,xmm1,8
+	mov	edi,ebx
+	add	eax,DWORD[16+rsp]
+	vpaddd	xmm9,xmm11,xmm4
+	xor	ecx,edx
+	shld	ebx,ebx,5
+	vpsrldq	xmm8,xmm4,4
+	add	eax,esi
+	and	edi,ecx
+	vpxor	xmm5,xmm5,xmm1
+	xor	ecx,edx
+	add	eax,ebx
+	vpxor	xmm8,xmm8,xmm3
+	shrd	ebx,ebx,7
+	xor	edi,edx
+	mov	esi,eax
+	add	ebp,DWORD[20+rsp]
+	vpxor	xmm5,xmm5,xmm8
+	xor	ebx,ecx
+	shld	eax,eax,5
+	vmovdqa	XMMWORD[rsp],xmm9
+	add	ebp,edi
+	and	esi,ebx
+	vpsrld	xmm8,xmm5,31
+	xor	ebx,ecx
+	add	ebp,eax
+	shrd	eax,eax,7
+	xor	esi,ecx
+	vpslldq	xmm10,xmm5,12
+	vpaddd	xmm5,xmm5,xmm5
+	mov	edi,ebp
+	add	edx,DWORD[24+rsp]
+	xor	eax,ebx
+	shld	ebp,ebp,5
+	vpsrld	xmm9,xmm10,30
+	vpor	xmm5,xmm5,xmm8
+	add	edx,esi
+	and	edi,eax
+	xor	eax,ebx
+	add	edx,ebp
+	vpslld	xmm10,xmm10,2
+	vpxor	xmm5,xmm5,xmm9
+	shrd	ebp,ebp,7
+	xor	edi,ebx
+	mov	esi,edx
+	add	ecx,DWORD[28+rsp]
+	vpxor	xmm5,xmm5,xmm10
+	xor	ebp,eax
+	shld	edx,edx,5
+	vmovdqa	xmm11,XMMWORD[((-32))+r11]
+	add	ecx,edi
+	and	esi,ebp
+	xor	ebp,eax
+	add	ecx,edx
+	shrd	edx,edx,7
+	xor	esi,eax
+	vpalignr	xmm6,xmm3,xmm2,8
+	mov	edi,ecx
+	add	ebx,DWORD[32+rsp]
+	vpaddd	xmm9,xmm11,xmm5
+	xor	edx,ebp
+	shld	ecx,ecx,5
+	vpsrldq	xmm8,xmm5,4
+	add	ebx,esi
+	and	edi,edx
+	vpxor	xmm6,xmm6,xmm2
+	xor	edx,ebp
+	add	ebx,ecx
+	vpxor	xmm8,xmm8,xmm4
+	shrd	ecx,ecx,7
+	xor	edi,ebp
+	mov	esi,ebx
+	add	eax,DWORD[36+rsp]
+	vpxor	xmm6,xmm6,xmm8
+	xor	ecx,edx
+	shld	ebx,ebx,5
+	vmovdqa	XMMWORD[16+rsp],xmm9
+	add	eax,edi
+	and	esi,ecx
+	vpsrld	xmm8,xmm6,31
+	xor	ecx,edx
+	add	eax,ebx
+	shrd	ebx,ebx,7
+	xor	esi,edx
+	vpslldq	xmm10,xmm6,12
+	vpaddd	xmm6,xmm6,xmm6
+	mov	edi,eax
+	add	ebp,DWORD[40+rsp]
+	xor	ebx,ecx
+	shld	eax,eax,5
+	vpsrld	xmm9,xmm10,30
+	vpor	xmm6,xmm6,xmm8
+	add	ebp,esi
+	and	edi,ebx
+	xor	ebx,ecx
+	add	ebp,eax
+	vpslld	xmm10,xmm10,2
+	vpxor	xmm6,xmm6,xmm9
+	shrd	eax,eax,7
+	xor	edi,ecx
+	mov	esi,ebp
+	add	edx,DWORD[44+rsp]
+	vpxor	xmm6,xmm6,xmm10
+	xor	eax,ebx
+	shld	ebp,ebp,5
+	add	edx,edi
+	and	esi,eax
+	xor	eax,ebx
+	add	edx,ebp
+	shrd	ebp,ebp,7
+	xor	esi,ebx
+	vpalignr	xmm7,xmm4,xmm3,8
+	mov	edi,edx
+	add	ecx,DWORD[48+rsp]
+	vpaddd	xmm9,xmm11,xmm6
+	xor	ebp,eax
+	shld	edx,edx,5
+	vpsrldq	xmm8,xmm6,4
+	add	ecx,esi
+	and	edi,ebp
+	vpxor	xmm7,xmm7,xmm3
+	xor	ebp,eax
+	add	ecx,edx
+	vpxor	xmm8,xmm8,xmm5
+	shrd	edx,edx,7
+	xor	edi,eax
+	mov	esi,ecx
+	add	ebx,DWORD[52+rsp]
+	vpxor	xmm7,xmm7,xmm8
+	xor	edx,ebp
+	shld	ecx,ecx,5
+	vmovdqa	XMMWORD[32+rsp],xmm9
+	add	ebx,edi
+	and	esi,edx
+	vpsrld	xmm8,xmm7,31
+	xor	edx,ebp
+	add	ebx,ecx
+	shrd	ecx,ecx,7
+	xor	esi,ebp
+	vpslldq	xmm10,xmm7,12
+	vpaddd	xmm7,xmm7,xmm7
+	mov	edi,ebx
+	add	eax,DWORD[56+rsp]
+	xor	ecx,edx
+	shld	ebx,ebx,5
+	vpsrld	xmm9,xmm10,30
+	vpor	xmm7,xmm7,xmm8
+	add	eax,esi
+	and	edi,ecx
+	xor	ecx,edx
+	add	eax,ebx
+	vpslld	xmm10,xmm10,2
+	vpxor	xmm7,xmm7,xmm9
+	shrd	ebx,ebx,7
+	xor	edi,edx
+	mov	esi,eax
+	add	ebp,DWORD[60+rsp]
+	vpxor	xmm7,xmm7,xmm10
+	xor	ebx,ecx
+	shld	eax,eax,5
+	add	ebp,edi
+	and	esi,ebx
+	xor	ebx,ecx
+	add	ebp,eax
+	vpalignr	xmm8,xmm7,xmm6,8
+	vpxor	xmm0,xmm0,xmm4
+	shrd	eax,eax,7
+	xor	esi,ecx
+	mov	edi,ebp
+	add	edx,DWORD[rsp]
+	vpxor	xmm0,xmm0,xmm1
+	xor	eax,ebx
+	shld	ebp,ebp,5
+	vpaddd	xmm9,xmm11,xmm7
+	add	edx,esi
+	and	edi,eax
+	vpxor	xmm0,xmm0,xmm8
+	xor	eax,ebx
+	add	edx,ebp
+	shrd	ebp,ebp,7
+	xor	edi,ebx
+	vpsrld	xmm8,xmm0,30
+	vmovdqa	XMMWORD[48+rsp],xmm9
+	mov	esi,edx
+	add	ecx,DWORD[4+rsp]
+	xor	ebp,eax
+	shld	edx,edx,5
+	vpslld	xmm0,xmm0,2
+	add	ecx,edi
+	and	esi,ebp
+	xor	ebp,eax
+	add	ecx,edx
+	shrd	edx,edx,7
+	xor	esi,eax
+	mov	edi,ecx
+	add	ebx,DWORD[8+rsp]
+	vpor	xmm0,xmm0,xmm8
+	xor	edx,ebp
+	shld	ecx,ecx,5
+	add	ebx,esi
+	and	edi,edx
+	xor	edx,ebp
+	add	ebx,ecx
+	add	eax,DWORD[12+rsp]
+	xor	edi,ebp
+	mov	esi,ebx
+	shld	ebx,ebx,5
+	add	eax,edi
+	xor	esi,edx
+	shrd	ecx,ecx,7
+	add	eax,ebx
+	vpalignr	xmm8,xmm0,xmm7,8
+	vpxor	xmm1,xmm1,xmm5
+	add	ebp,DWORD[16+rsp]
+	xor	esi,ecx
+	mov	edi,eax
+	shld	eax,eax,5
+	vpxor	xmm1,xmm1,xmm2
+	add	ebp,esi
+	xor	edi,ecx
+	vpaddd	xmm9,xmm11,xmm0
+	shrd	ebx,ebx,7
+	add	ebp,eax
+	vpxor	xmm1,xmm1,xmm8
+	add	edx,DWORD[20+rsp]
+	xor	edi,ebx
+	mov	esi,ebp
+	shld	ebp,ebp,5
+	vpsrld	xmm8,xmm1,30
+	vmovdqa	XMMWORD[rsp],xmm9
+	add	edx,edi
+	xor	esi,ebx
+	shrd	eax,eax,7
+	add	edx,ebp
+	vpslld	xmm1,xmm1,2
+	add	ecx,DWORD[24+rsp]
+	xor	esi,eax
+	mov	edi,edx
+	shld	edx,edx,5
+	add	ecx,esi
+	xor	edi,eax
+	shrd	ebp,ebp,7
+	add	ecx,edx
+	vpor	xmm1,xmm1,xmm8
+	add	ebx,DWORD[28+rsp]
+	xor	edi,ebp
+	mov	esi,ecx
+	shld	ecx,ecx,5
+	add	ebx,edi
+	xor	esi,ebp
+	shrd	edx,edx,7
+	add	ebx,ecx
+	vpalignr	xmm8,xmm1,xmm0,8
+	vpxor	xmm2,xmm2,xmm6
+	add	eax,DWORD[32+rsp]
+	xor	esi,edx
+	mov	edi,ebx
+	shld	ebx,ebx,5
+	vpxor	xmm2,xmm2,xmm3
+	add	eax,esi
+	xor	edi,edx
+	vpaddd	xmm9,xmm11,xmm1
+	vmovdqa	xmm11,XMMWORD[r11]
+	shrd	ecx,ecx,7
+	add	eax,ebx
+	vpxor	xmm2,xmm2,xmm8
+	add	ebp,DWORD[36+rsp]
+	xor	edi,ecx
+	mov	esi,eax
+	shld	eax,eax,5
+	vpsrld	xmm8,xmm2,30
+	vmovdqa	XMMWORD[16+rsp],xmm9
+	add	ebp,edi
+	xor	esi,ecx
+	shrd	ebx,ebx,7
+	add	ebp,eax
+	vpslld	xmm2,xmm2,2
+	add	edx,DWORD[40+rsp]
+	xor	esi,ebx
+	mov	edi,ebp
+	shld	ebp,ebp,5
+	add	edx,esi
+	xor	edi,ebx
+	shrd	eax,eax,7
+	add	edx,ebp
+	vpor	xmm2,xmm2,xmm8
+	add	ecx,DWORD[44+rsp]
+	xor	edi,eax
+	mov	esi,edx
+	shld	edx,edx,5
+	add	ecx,edi
+	xor	esi,eax
+	shrd	ebp,ebp,7
+	add	ecx,edx
+	vpalignr	xmm8,xmm2,xmm1,8
+	vpxor	xmm3,xmm3,xmm7
+	add	ebx,DWORD[48+rsp]
+	xor	esi,ebp
+	mov	edi,ecx
+	shld	ecx,ecx,5
+	vpxor	xmm3,xmm3,xmm4
+	add	ebx,esi
+	xor	edi,ebp
+	vpaddd	xmm9,xmm11,xmm2
+	shrd	edx,edx,7
+	add	ebx,ecx
+	vpxor	xmm3,xmm3,xmm8
+	add	eax,DWORD[52+rsp]
+	xor	edi,edx
+	mov	esi,ebx
+	shld	ebx,ebx,5
+	vpsrld	xmm8,xmm3,30
+	vmovdqa	XMMWORD[32+rsp],xmm9
+	add	eax,edi
+	xor	esi,edx
+	shrd	ecx,ecx,7
+	add	eax,ebx
+	vpslld	xmm3,xmm3,2
+	add	ebp,DWORD[56+rsp]
+	xor	esi,ecx
+	mov	edi,eax
+	shld	eax,eax,5
+	add	ebp,esi
+	xor	edi,ecx
+	shrd	ebx,ebx,7
+	add	ebp,eax
+	vpor	xmm3,xmm3,xmm8
+	add	edx,DWORD[60+rsp]
+	xor	edi,ebx
+	mov	esi,ebp
+	shld	ebp,ebp,5
+	add	edx,edi
+	xor	esi,ebx
+	shrd	eax,eax,7
+	add	edx,ebp
+	vpalignr	xmm8,xmm3,xmm2,8
+	vpxor	xmm4,xmm4,xmm0
+	add	ecx,DWORD[rsp]
+	xor	esi,eax
+	mov	edi,edx
+	shld	edx,edx,5
+	vpxor	xmm4,xmm4,xmm5
+	add	ecx,esi
+	xor	edi,eax
+	vpaddd	xmm9,xmm11,xmm3
+	shrd	ebp,ebp,7
+	add	ecx,edx
+	vpxor	xmm4,xmm4,xmm8
+	add	ebx,DWORD[4+rsp]
+	xor	edi,ebp
+	mov	esi,ecx
+	shld	ecx,ecx,5
+	vpsrld	xmm8,xmm4,30
+	vmovdqa	XMMWORD[48+rsp],xmm9
+	add	ebx,edi
+	xor	esi,ebp
+	shrd	edx,edx,7
+	add	ebx,ecx
+	vpslld	xmm4,xmm4,2
+	add	eax,DWORD[8+rsp]
+	xor	esi,edx
+	mov	edi,ebx
+	shld	ebx,ebx,5
+	add	eax,esi
+	xor	edi,edx
+	shrd	ecx,ecx,7
+	add	eax,ebx
+	vpor	xmm4,xmm4,xmm8
+	add	ebp,DWORD[12+rsp]
+	xor	edi,ecx
+	mov	esi,eax
+	shld	eax,eax,5
+	add	ebp,edi
+	xor	esi,ecx
+	shrd	ebx,ebx,7
+	add	ebp,eax
+	vpalignr	xmm8,xmm4,xmm3,8
+	vpxor	xmm5,xmm5,xmm1
+	add	edx,DWORD[16+rsp]
+	xor	esi,ebx
+	mov	edi,ebp
+	shld	ebp,ebp,5
+	vpxor	xmm5,xmm5,xmm6
+	add	edx,esi
+	xor	edi,ebx
+	vpaddd	xmm9,xmm11,xmm4
+	shrd	eax,eax,7
+	add	edx,ebp
+	vpxor	xmm5,xmm5,xmm8
+	add	ecx,DWORD[20+rsp]
+	xor	edi,eax
+	mov	esi,edx
+	shld	edx,edx,5
+	vpsrld	xmm8,xmm5,30
+	vmovdqa	XMMWORD[rsp],xmm9
+	add	ecx,edi
+	xor	esi,eax
+	shrd	ebp,ebp,7
+	add	ecx,edx
+	vpslld	xmm5,xmm5,2
+	add	ebx,DWORD[24+rsp]
+	xor	esi,ebp
+	mov	edi,ecx
+	shld	ecx,ecx,5
+	add	ebx,esi
+	xor	edi,ebp
+	shrd	edx,edx,7
+	add	ebx,ecx
+	vpor	xmm5,xmm5,xmm8
+	add	eax,DWORD[28+rsp]
+	shrd	ecx,ecx,7
+	mov	esi,ebx
+	xor	edi,edx
+	shld	ebx,ebx,5
+	add	eax,edi
+	xor	esi,ecx
+	xor	ecx,edx
+	add	eax,ebx
+	vpalignr	xmm8,xmm5,xmm4,8
+	vpxor	xmm6,xmm6,xmm2
+	add	ebp,DWORD[32+rsp]
+	and	esi,ecx
+	xor	ecx,edx
+	shrd	ebx,ebx,7
+	vpxor	xmm6,xmm6,xmm7
+	mov	edi,eax
+	xor	esi,ecx
+	vpaddd	xmm9,xmm11,xmm5
+	shld	eax,eax,5
+	add	ebp,esi
+	vpxor	xmm6,xmm6,xmm8
+	xor	edi,ebx
+	xor	ebx,ecx
+	add	ebp,eax
+	add	edx,DWORD[36+rsp]
+	vpsrld	xmm8,xmm6,30
+	vmovdqa	XMMWORD[16+rsp],xmm9
+	and	edi,ebx
+	xor	ebx,ecx
+	shrd	eax,eax,7
+	mov	esi,ebp
+	vpslld	xmm6,xmm6,2
+	xor	edi,ebx
+	shld	ebp,ebp,5
+	add	edx,edi
+	xor	esi,eax
+	xor	eax,ebx
+	add	edx,ebp
+	add	ecx,DWORD[40+rsp]
+	and	esi,eax
+	vpor	xmm6,xmm6,xmm8
+	xor	eax,ebx
+	shrd	ebp,ebp,7
+	mov	edi,edx
+	xor	esi,eax
+	shld	edx,edx,5
+	add	ecx,esi
+	xor	edi,ebp
+	xor	ebp,eax
+	add	ecx,edx
+	add	ebx,DWORD[44+rsp]
+	and	edi,ebp
+	xor	ebp,eax
+	shrd	edx,edx,7
+	mov	esi,ecx
+	xor	edi,ebp
+	shld	ecx,ecx,5
+	add	ebx,edi
+	xor	esi,edx
+	xor	edx,ebp
+	add	ebx,ecx
+	vpalignr	xmm8,xmm6,xmm5,8
+	vpxor	xmm7,xmm7,xmm3
+	add	eax,DWORD[48+rsp]
+	and	esi,edx
+	xor	edx,ebp
+	shrd	ecx,ecx,7
+	vpxor	xmm7,xmm7,xmm0
+	mov	edi,ebx
+	xor	esi,edx
+	vpaddd	xmm9,xmm11,xmm6
+	vmovdqa	xmm11,XMMWORD[32+r11]
+	shld	ebx,ebx,5
+	add	eax,esi
+	vpxor	xmm7,xmm7,xmm8
+	xor	edi,ecx
+	xor	ecx,edx
+	add	eax,ebx
+	add	ebp,DWORD[52+rsp]
+	vpsrld	xmm8,xmm7,30
+	vmovdqa	XMMWORD[32+rsp],xmm9
+	and	edi,ecx
+	xor	ecx,edx
+	shrd	ebx,ebx,7
+	mov	esi,eax
+	vpslld	xmm7,xmm7,2
+	xor	edi,ecx
+	shld	eax,eax,5
+	add	ebp,edi
+	xor	esi,ebx
+	xor	ebx,ecx
+	add	ebp,eax
+	add	edx,DWORD[56+rsp]
+	and	esi,ebx
+	vpor	xmm7,xmm7,xmm8
+	xor	ebx,ecx
+	shrd	eax,eax,7
+	mov	edi,ebp
+	xor	esi,ebx
+	shld	ebp,ebp,5
+	add	edx,esi
+	xor	edi,eax
+	xor	eax,ebx
+	add	edx,ebp
+	add	ecx,DWORD[60+rsp]
+	and	edi,eax
+	xor	eax,ebx
+	shrd	ebp,ebp,7
+	mov	esi,edx
+	xor	edi,eax
+	shld	edx,edx,5
+	add	ecx,edi
+	xor	esi,ebp
+	xor	ebp,eax
+	add	ecx,edx
+	vpalignr	xmm8,xmm7,xmm6,8
+	vpxor	xmm0,xmm0,xmm4
+	add	ebx,DWORD[rsp]
+	and	esi,ebp
+	xor	ebp,eax
+	shrd	edx,edx,7
+	vpxor	xmm0,xmm0,xmm1
+	mov	edi,ecx
+	xor	esi,ebp
+	vpaddd	xmm9,xmm11,xmm7
+	shld	ecx,ecx,5
+	add	ebx,esi
+	vpxor	xmm0,xmm0,xmm8
+	xor	edi,edx
+	xor	edx,ebp
+	add	ebx,ecx
+	add	eax,DWORD[4+rsp]
+	vpsrld	xmm8,xmm0,30
+	vmovdqa	XMMWORD[48+rsp],xmm9
+	and	edi,edx
+	xor	edx,ebp
+	shrd	ecx,ecx,7
+	mov	esi,ebx
+	vpslld	xmm0,xmm0,2
+	xor	edi,edx
+	shld	ebx,ebx,5
+	add	eax,edi
+	xor	esi,ecx
+	xor	ecx,edx
+	add	eax,ebx
+	add	ebp,DWORD[8+rsp]
+	and	esi,ecx
+	vpor	xmm0,xmm0,xmm8
+	xor	ecx,edx
+	shrd	ebx,ebx,7
+	mov	edi,eax
+	xor	esi,ecx
+	shld	eax,eax,5
+	add	ebp,esi
+	xor	edi,ebx
+	xor	ebx,ecx
+	add	ebp,eax
+	add	edx,DWORD[12+rsp]
+	and	edi,ebx
+	xor	ebx,ecx
+	shrd	eax,eax,7
+	mov	esi,ebp
+	xor	edi,ebx
+	shld	ebp,ebp,5
+	add	edx,edi
+	xor	esi,eax
+	xor	eax,ebx
+	add	edx,ebp
+	vpalignr	xmm8,xmm0,xmm7,8
+	vpxor	xmm1,xmm1,xmm5
+	add	ecx,DWORD[16+rsp]
+	and	esi,eax
+	xor	eax,ebx
+	shrd	ebp,ebp,7
+	vpxor	xmm1,xmm1,xmm2
+	mov	edi,edx
+	xor	esi,eax
+	vpaddd	xmm9,xmm11,xmm0
+	shld	edx,edx,5
+	add	ecx,esi
+	vpxor	xmm1,xmm1,xmm8
+	xor	edi,ebp
+	xor	ebp,eax
+	add	ecx,edx
+	add	ebx,DWORD[20+rsp]
+	vpsrld	xmm8,xmm1,30
+	vmovdqa	XMMWORD[rsp],xmm9
+	and	edi,ebp
+	xor	ebp,eax
+	shrd	edx,edx,7
+	mov	esi,ecx
+	vpslld	xmm1,xmm1,2
+	xor	edi,ebp
+	shld	ecx,ecx,5
+	add	ebx,edi
+	xor	esi,edx
+	xor	edx,ebp
+	add	ebx,ecx
+	add	eax,DWORD[24+rsp]
+	and	esi,edx
+	vpor	xmm1,xmm1,xmm8
+	xor	edx,ebp
+	shrd	ecx,ecx,7
+	mov	edi,ebx
+	xor	esi,edx
+	shld	ebx,ebx,5
+	add	eax,esi
+	xor	edi,ecx
+	xor	ecx,edx
+	add	eax,ebx
+	add	ebp,DWORD[28+rsp]
+	and	edi,ecx
+	xor	ecx,edx
+	shrd	ebx,ebx,7
+	mov	esi,eax
+	xor	edi,ecx
+	shld	eax,eax,5
+	add	ebp,edi
+	xor	esi,ebx
+	xor	ebx,ecx
+	add	ebp,eax
+	vpalignr	xmm8,xmm1,xmm0,8
+	vpxor	xmm2,xmm2,xmm6
+	add	edx,DWORD[32+rsp]
+	and	esi,ebx
+	xor	ebx,ecx
+	shrd	eax,eax,7
+	vpxor	xmm2,xmm2,xmm3
+	mov	edi,ebp
+	xor	esi,ebx
+	vpaddd	xmm9,xmm11,xmm1
+	shld	ebp,ebp,5
+	add	edx,esi
+	vpxor	xmm2,xmm2,xmm8
+	xor	edi,eax
+	xor	eax,ebx
+	add	edx,ebp
+	add	ecx,DWORD[36+rsp]
+	vpsrld	xmm8,xmm2,30
+	vmovdqa	XMMWORD[16+rsp],xmm9
+	and	edi,eax
+	xor	eax,ebx
+	shrd	ebp,ebp,7
+	mov	esi,edx
+	vpslld	xmm2,xmm2,2
+	xor	edi,eax
+	shld	edx,edx,5
+	add	ecx,edi
+	xor	esi,ebp
+	xor	ebp,eax
+	add	ecx,edx
+	add	ebx,DWORD[40+rsp]
+	and	esi,ebp
+	vpor	xmm2,xmm2,xmm8
+	xor	ebp,eax
+	shrd	edx,edx,7
+	mov	edi,ecx
+	xor	esi,ebp
+	shld	ecx,ecx,5
+	add	ebx,esi
+	xor	edi,edx
+	xor	edx,ebp
+	add	ebx,ecx
+	add	eax,DWORD[44+rsp]
+	and	edi,edx
+	xor	edx,ebp
+	shrd	ecx,ecx,7
+	mov	esi,ebx
+	xor	edi,edx
+	shld	ebx,ebx,5
+	add	eax,edi
+	xor	esi,edx
+	add	eax,ebx
+	vpalignr	xmm8,xmm2,xmm1,8
+	vpxor	xmm3,xmm3,xmm7
+	add	ebp,DWORD[48+rsp]
+	xor	esi,ecx
+	mov	edi,eax
+	shld	eax,eax,5
+	vpxor	xmm3,xmm3,xmm4
+	add	ebp,esi
+	xor	edi,ecx
+	vpaddd	xmm9,xmm11,xmm2
+	shrd	ebx,ebx,7
+	add	ebp,eax
+	vpxor	xmm3,xmm3,xmm8
+	add	edx,DWORD[52+rsp]
+	xor	edi,ebx
+	mov	esi,ebp
+	shld	ebp,ebp,5
+	vpsrld	xmm8,xmm3,30
+	vmovdqa	XMMWORD[32+rsp],xmm9
+	add	edx,edi
+	xor	esi,ebx
+	shrd	eax,eax,7
+	add	edx,ebp
+	vpslld	xmm3,xmm3,2
+	add	ecx,DWORD[56+rsp]
+	xor	esi,eax
+	mov	edi,edx
+	shld	edx,edx,5
+	add	ecx,esi
+	xor	edi,eax
+	shrd	ebp,ebp,7
+	add	ecx,edx
+	vpor	xmm3,xmm3,xmm8
+	add	ebx,DWORD[60+rsp]
+	xor	edi,ebp
+	mov	esi,ecx
+	shld	ecx,ecx,5
+	add	ebx,edi
+	xor	esi,ebp
+	shrd	edx,edx,7
+	add	ebx,ecx
+	add	eax,DWORD[rsp]
+	vpaddd	xmm9,xmm11,xmm3
+	xor	esi,edx
+	mov	edi,ebx
+	shld	ebx,ebx,5
+	add	eax,esi
+	vmovdqa	XMMWORD[48+rsp],xmm9
+	xor	edi,edx
+	shrd	ecx,ecx,7
+	add	eax,ebx
+	add	ebp,DWORD[4+rsp]
+	xor	edi,ecx
+	mov	esi,eax
+	shld	eax,eax,5
+	add	ebp,edi
+	xor	esi,ecx
+	shrd	ebx,ebx,7
+	add	ebp,eax
+	add	edx,DWORD[8+rsp]
+	xor	esi,ebx
+	mov	edi,ebp
+	shld	ebp,ebp,5
+	add	edx,esi
+	xor	edi,ebx
+	shrd	eax,eax,7
+	add	edx,ebp
+	add	ecx,DWORD[12+rsp]
+	xor	edi,eax
+	mov	esi,edx
+	shld	edx,edx,5
+	add	ecx,edi
+	xor	esi,eax
+	shrd	ebp,ebp,7
+	add	ecx,edx
+	cmp	r9,r10
+	je	NEAR $L$done_avx
+	vmovdqa	xmm6,XMMWORD[64+r11]
+	vmovdqa	xmm11,XMMWORD[((-64))+r11]
+	vmovdqu	xmm0,XMMWORD[r9]
+	vmovdqu	xmm1,XMMWORD[16+r9]
+	vmovdqu	xmm2,XMMWORD[32+r9]
+	vmovdqu	xmm3,XMMWORD[48+r9]
+	vpshufb	xmm0,xmm0,xmm6
+	add	r9,64
+	add	ebx,DWORD[16+rsp]
+	xor	esi,ebp
+	vpshufb	xmm1,xmm1,xmm6
+	mov	edi,ecx
+	shld	ecx,ecx,5
+	vpaddd	xmm4,xmm0,xmm11
+	add	ebx,esi
+	xor	edi,ebp
+	shrd	edx,edx,7
+	add	ebx,ecx
+	vmovdqa	XMMWORD[rsp],xmm4
+	add	eax,DWORD[20+rsp]
+	xor	edi,edx
+	mov	esi,ebx
+	shld	ebx,ebx,5
+	add	eax,edi
+	xor	esi,edx
+	shrd	ecx,ecx,7
+	add	eax,ebx
+	add	ebp,DWORD[24+rsp]
+	xor	esi,ecx
+	mov	edi,eax
+	shld	eax,eax,5
+	add	ebp,esi
+	xor	edi,ecx
+	shrd	ebx,ebx,7
+	add	ebp,eax
+	add	edx,DWORD[28+rsp]
+	xor	edi,ebx
+	mov	esi,ebp
+	shld	ebp,ebp,5
+	add	edx,edi
+	xor	esi,ebx
+	shrd	eax,eax,7
+	add	edx,ebp
+	add	ecx,DWORD[32+rsp]
+	xor	esi,eax
+	vpshufb	xmm2,xmm2,xmm6
+	mov	edi,edx
+	shld	edx,edx,5
+	vpaddd	xmm5,xmm1,xmm11
+	add	ecx,esi
+	xor	edi,eax
+	shrd	ebp,ebp,7
+	add	ecx,edx
+	vmovdqa	XMMWORD[16+rsp],xmm5
+	add	ebx,DWORD[36+rsp]
+	xor	edi,ebp
+	mov	esi,ecx
+	shld	ecx,ecx,5
+	add	ebx,edi
+	xor	esi,ebp
+	shrd	edx,edx,7
+	add	ebx,ecx
+	add	eax,DWORD[40+rsp]
+	xor	esi,edx
+	mov	edi,ebx
+	shld	ebx,ebx,5
+	add	eax,esi
+	xor	edi,edx
+	shrd	ecx,ecx,7
+	add	eax,ebx
+	add	ebp,DWORD[44+rsp]
+	xor	edi,ecx
+	mov	esi,eax
+	shld	eax,eax,5
+	add	ebp,edi
+	xor	esi,ecx
+	shrd	ebx,ebx,7
+	add	ebp,eax
+	add	edx,DWORD[48+rsp]
+	xor	esi,ebx
+	vpshufb	xmm3,xmm3,xmm6
+	mov	edi,ebp
+	shld	ebp,ebp,5
+	vpaddd	xmm6,xmm2,xmm11
+	add	edx,esi
+	xor	edi,ebx
+	shrd	eax,eax,7
+	add	edx,ebp
+	vmovdqa	XMMWORD[32+rsp],xmm6
+	add	ecx,DWORD[52+rsp]
+	xor	edi,eax
+	mov	esi,edx
+	shld	edx,edx,5
+	add	ecx,edi
+	xor	esi,eax
+	shrd	ebp,ebp,7
+	add	ecx,edx
+	add	ebx,DWORD[56+rsp]
+	xor	esi,ebp
+	mov	edi,ecx
+	shld	ecx,ecx,5
+	add	ebx,esi
+	xor	edi,ebp
+	shrd	edx,edx,7
+	add	ebx,ecx
+	add	eax,DWORD[60+rsp]
+	xor	edi,edx
+	mov	esi,ebx
+	shld	ebx,ebx,5
+	add	eax,edi
+	shrd	ecx,ecx,7
+	add	eax,ebx
+	add	eax,DWORD[r8]
+	add	esi,DWORD[4+r8]
+	add	ecx,DWORD[8+r8]
+	add	edx,DWORD[12+r8]
+	mov	DWORD[r8],eax
+	add	ebp,DWORD[16+r8]
+	mov	DWORD[4+r8],esi
+	mov	ebx,esi
+	mov	DWORD[8+r8],ecx
+	mov	edi,ecx
+	mov	DWORD[12+r8],edx
+	xor	edi,edx
+	mov	DWORD[16+r8],ebp
+	and	esi,edi
+	jmp	NEAR $L$oop_avx
+
+ALIGN	16
+$L$done_avx:
+	add	ebx,DWORD[16+rsp]
+	xor	esi,ebp
+	mov	edi,ecx
+	shld	ecx,ecx,5
+	add	ebx,esi
+	xor	edi,ebp
+	shrd	edx,edx,7
+	add	ebx,ecx
+	add	eax,DWORD[20+rsp]
+	xor	edi,edx
+	mov	esi,ebx
+	shld	ebx,ebx,5
+	add	eax,edi
+	xor	esi,edx
+	shrd	ecx,ecx,7
+	add	eax,ebx
+	add	ebp,DWORD[24+rsp]
+	xor	esi,ecx
+	mov	edi,eax
+	shld	eax,eax,5
+	add	ebp,esi
+	xor	edi,ecx
+	shrd	ebx,ebx,7
+	add	ebp,eax
+	add	edx,DWORD[28+rsp]
+	xor	edi,ebx
+	mov	esi,ebp
+	shld	ebp,ebp,5
+	add	edx,edi
+	xor	esi,ebx
+	shrd	eax,eax,7
+	add	edx,ebp
+	add	ecx,DWORD[32+rsp]
+	xor	esi,eax
+	mov	edi,edx
+	shld	edx,edx,5
+	add	ecx,esi
+	xor	edi,eax
+	shrd	ebp,ebp,7
+	add	ecx,edx
+	add	ebx,DWORD[36+rsp]
+	xor	edi,ebp
+	mov	esi,ecx
+	shld	ecx,ecx,5
+	add	ebx,edi
+	xor	esi,ebp
+	shrd	edx,edx,7
+	add	ebx,ecx
+	add	eax,DWORD[40+rsp]
+	xor	esi,edx
+	mov	edi,ebx
+	shld	ebx,ebx,5
+	add	eax,esi
+	xor	edi,edx
+	shrd	ecx,ecx,7
+	add	eax,ebx
+	add	ebp,DWORD[44+rsp]
+	xor	edi,ecx
+	mov	esi,eax
+	shld	eax,eax,5
+	add	ebp,edi
+	xor	esi,ecx
+	shrd	ebx,ebx,7
+	add	ebp,eax
+	add	edx,DWORD[48+rsp]
+	xor	esi,ebx
+	mov	edi,ebp
+	shld	ebp,ebp,5
+	add	edx,esi
+	xor	edi,ebx
+	shrd	eax,eax,7
+	add	edx,ebp
+	add	ecx,DWORD[52+rsp]
+	xor	edi,eax
+	mov	esi,edx
+	shld	edx,edx,5
+	add	ecx,edi
+	xor	esi,eax
+	shrd	ebp,ebp,7
+	add	ecx,edx
+	add	ebx,DWORD[56+rsp]
+	xor	esi,ebp
+	mov	edi,ecx
+	shld	ecx,ecx,5
+	add	ebx,esi
+	xor	edi,ebp
+	shrd	edx,edx,7
+	add	ebx,ecx
+	add	eax,DWORD[60+rsp]
+	xor	edi,edx
+	mov	esi,ebx
+	shld	ebx,ebx,5
+	add	eax,edi
+	shrd	ecx,ecx,7
+	add	eax,ebx
+	vzeroupper
+
+	add	eax,DWORD[r8]
+	add	esi,DWORD[4+r8]
+	add	ecx,DWORD[8+r8]
+	mov	DWORD[r8],eax
+	add	edx,DWORD[12+r8]
+	mov	DWORD[4+r8],esi
+	add	ebp,DWORD[16+r8]
+	mov	DWORD[8+r8],ecx
+	mov	DWORD[12+r8],edx
+	mov	DWORD[16+r8],ebp
+	movaps	xmm6,XMMWORD[((-40-96))+r14]
+	movaps	xmm7,XMMWORD[((-40-80))+r14]
+	movaps	xmm8,XMMWORD[((-40-64))+r14]
+	movaps	xmm9,XMMWORD[((-40-48))+r14]
+	movaps	xmm10,XMMWORD[((-40-32))+r14]
+	movaps	xmm11,XMMWORD[((-40-16))+r14]
+	lea	rsi,[r14]
+	mov	r14,QWORD[((-40))+rsi]
+	mov	r13,QWORD[((-32))+rsi]
+	mov	r12,QWORD[((-24))+rsi]
+	mov	rbp,QWORD[((-16))+rsi]
+	mov	rbx,QWORD[((-8))+rsi]
+	lea	rsp,[rsi]
+$L$epilogue_avx:
+	mov	rdi,QWORD[8+rsp]	;WIN64 epilogue
+	mov	rsi,QWORD[16+rsp]
+	DB	0F3h,0C3h		;repret
+$L$SEH_end_sha1_block_data_order_avx:
 ALIGN	64
 K_XX_XX:
 	DD	0x5a827999,0x5a827999,0x5a827999,0x5a827999
@@ -2605,6 +3750,9 @@
 	DD	$L$SEH_begin_sha1_block_data_order_ssse3 wrt ..imagebase
 	DD	$L$SEH_end_sha1_block_data_order_ssse3 wrt ..imagebase
 	DD	$L$SEH_info_sha1_block_data_order_ssse3 wrt ..imagebase
+	DD	$L$SEH_begin_sha1_block_data_order_avx wrt ..imagebase
+	DD	$L$SEH_end_sha1_block_data_order_avx wrt ..imagebase
+	DD	$L$SEH_info_sha1_block_data_order_avx wrt ..imagebase
 section	.xdata rdata align=8
 ALIGN	8
 $L$SEH_info_sha1_block_data_order:
@@ -2614,3 +3762,7 @@
 DB	9,0,0,0
 	DD	ssse3_handler wrt ..imagebase
 	DD	$L$prologue_ssse3 wrt ..imagebase,$L$epilogue_ssse3 wrt ..imagebase
+$L$SEH_info_sha1_block_data_order_avx:
+DB	9,0,0,0
+	DD	ssse3_handler wrt ..imagebase
+	DD	$L$prologue_avx wrt ..imagebase,$L$epilogue_avx wrt ..imagebase

third_party/boringssl/win-x86_64/crypto/sha/sha256-x86_64.asm

diff --git a/third_party/boringssl/win-x86_64/crypto/sha/sha256-x86_64.asm b/third_party/boringssl/win-x86_64/crypto/sha/sha256-x86_64.asm
index e6193c5..efaf9b5 100644
--- a/third_party/boringssl/win-x86_64/crypto/sha/sha256-x86_64.asm
+++ b/third_party/boringssl/win-x86_64/crypto/sha/sha256-x86_64.asm
@@ -23,6 +23,11 @@
 	mov	r9d,DWORD[r11]
 	mov	r10d,DWORD[4+r11]
 	mov	r11d,DWORD[8+r11]
+	and	r9d,1073741824
+	and	r10d,268435968
+	or	r10d,r9d
+	cmp	r10d,1342177792
+	je	NEAR $L$avx_shortcut
 	test	r10d,512
 	jnz	NEAR $L$ssse3_shortcut
 	push	rbx
@@ -2877,6 +2882,1082 @@
 	mov	rsi,QWORD[16+rsp]
 	DB	0F3h,0C3h		;repret
 $L$SEH_end_sha256_block_data_order_ssse3:
+
+ALIGN	64
+sha256_block_data_order_avx:
+	mov	QWORD[8+rsp],rdi	;WIN64 prologue
+	mov	QWORD[16+rsp],rsi
+	mov	rax,rsp
+$L$SEH_begin_sha256_block_data_order_avx:
+	mov	rdi,rcx
+	mov	rsi,rdx
+	mov	rdx,r8
+
+
+$L$avx_shortcut:
+	push	rbx
+	push	rbp
+	push	r12
+	push	r13
+	push	r14
+	push	r15
+	mov	r11,rsp
+	shl	rdx,4
+	sub	rsp,160
+	lea	rdx,[rdx*4+rsi]
+	and	rsp,-64
+	mov	QWORD[((64+0))+rsp],rdi
+	mov	QWORD[((64+8))+rsp],rsi
+	mov	QWORD[((64+16))+rsp],rdx
+	mov	QWORD[((64+24))+rsp],r11
+	movaps	XMMWORD[(64+32)+rsp],xmm6
+	movaps	XMMWORD[(64+48)+rsp],xmm7
+	movaps	XMMWORD[(64+64)+rsp],xmm8
+	movaps	XMMWORD[(64+80)+rsp],xmm9
+$L$prologue_avx:
+
+	vzeroupper
+	mov	eax,DWORD[rdi]
+	mov	ebx,DWORD[4+rdi]
+	mov	ecx,DWORD[8+rdi]
+	mov	edx,DWORD[12+rdi]
+	mov	r8d,DWORD[16+rdi]
+	mov	r9d,DWORD[20+rdi]
+	mov	r10d,DWORD[24+rdi]
+	mov	r11d,DWORD[28+rdi]
+	vmovdqa	xmm8,XMMWORD[((K256+512+32))]
+	vmovdqa	xmm9,XMMWORD[((K256+512+64))]
+	jmp	NEAR $L$loop_avx
+ALIGN	16
+$L$loop_avx:
+	vmovdqa	xmm7,XMMWORD[((K256+512))]
+	vmovdqu	xmm0,XMMWORD[rsi]
+	vmovdqu	xmm1,XMMWORD[16+rsi]
+	vmovdqu	xmm2,XMMWORD[32+rsi]
+	vmovdqu	xmm3,XMMWORD[48+rsi]
+	vpshufb	xmm0,xmm0,xmm7
+	lea	rbp,[K256]
+	vpshufb	xmm1,xmm1,xmm7
+	vpshufb	xmm2,xmm2,xmm7
+	vpaddd	xmm4,xmm0,XMMWORD[rbp]
+	vpshufb	xmm3,xmm3,xmm7
+	vpaddd	xmm5,xmm1,XMMWORD[32+rbp]
+	vpaddd	xmm6,xmm2,XMMWORD[64+rbp]
+	vpaddd	xmm7,xmm3,XMMWORD[96+rbp]
+	vmovdqa	XMMWORD[rsp],xmm4
+	mov	r14d,eax
+	vmovdqa	XMMWORD[16+rsp],xmm5
+	mov	edi,ebx
+	vmovdqa	XMMWORD[32+rsp],xmm6
+	xor	edi,ecx
+	vmovdqa	XMMWORD[48+rsp],xmm7
+	mov	r13d,r8d
+	jmp	NEAR $L$avx_00_47
+
+ALIGN	16
+$L$avx_00_47:
+	sub	rbp,-128
+	vpalignr	xmm4,xmm1,xmm0,4
+	shrd	r13d,r13d,14
+	mov	eax,r14d
+	mov	r12d,r9d
+	vpalignr	xmm7,xmm3,xmm2,4
+	shrd	r14d,r14d,9
+	xor	r13d,r8d
+	xor	r12d,r10d
+	vpsrld	xmm6,xmm4,7
+	shrd	r13d,r13d,5
+	xor	r14d,eax
+	and	r12d,r8d
+	vpaddd	xmm0,xmm0,xmm7
+	xor	r13d,r8d
+	add	r11d,DWORD[rsp]
+	mov	r15d,eax
+	vpsrld	xmm7,xmm4,3
+	xor	r12d,r10d
+	shrd	r14d,r14d,11
+	xor	r15d,ebx
+	vpslld	xmm5,xmm4,14
+	add	r11d,r12d
+	shrd	r13d,r13d,6
+	and	edi,r15d
+	vpxor	xmm4,xmm7,xmm6
+	xor	r14d,eax
+	add	r11d,r13d
+	xor	edi,ebx
+	vpshufd	xmm7,xmm3,250
+	shrd	r14d,r14d,2
+	add	edx,r11d
+	add	r11d,edi
+	vpsrld	xmm6,xmm6,11
+	mov	r13d,edx
+	add	r14d,r11d
+	shrd	r13d,r13d,14
+	vpxor	xmm4,xmm4,xmm5
+	mov	r11d,r14d
+	mov	r12d,r8d
+	shrd	r14d,r14d,9
+	vpslld	xmm5,xmm5,11
+	xor	r13d,edx
+	xor	r12d,r9d
+	shrd	r13d,r13d,5
+	vpxor	xmm4,xmm4,xmm6
+	xor	r14d,r11d
+	and	r12d,edx
+	xor	r13d,edx
+	vpsrld	xmm6,xmm7,10
+	add	r10d,DWORD[4+rsp]
+	mov	edi,r11d
+	xor	r12d,r9d
+	vpxor	xmm4,xmm4,xmm5
+	shrd	r14d,r14d,11
+	xor	edi,eax
+	add	r10d,r12d
+	vpsrlq	xmm7,xmm7,17
+	shrd	r13d,r13d,6
+	and	r15d,edi
+	xor	r14d,r11d
+	vpaddd	xmm0,xmm0,xmm4
+	add	r10d,r13d
+	xor	r15d,eax
+	shrd	r14d,r14d,2
+	vpxor	xmm6,xmm6,xmm7
+	add	ecx,r10d
+	add	r10d,r15d
+	mov	r13d,ecx
+	vpsrlq	xmm7,xmm7,2
+	add	r14d,r10d
+	shrd	r13d,r13d,14
+	mov	r10d,r14d
+	vpxor	xmm6,xmm6,xmm7
+	mov	r12d,edx
+	shrd	r14d,r14d,9
+	xor	r13d,ecx
+	vpshufb	xmm6,xmm6,xmm8
+	xor	r12d,r8d
+	shrd	r13d,r13d,5
+	xor	r14d,r10d
+	vpaddd	xmm0,xmm0,xmm6
+	and	r12d,ecx
+	xor	r13d,ecx
+	add	r9d,DWORD[8+rsp]
+	vpshufd	xmm7,xmm0,80
+	mov	r15d,r10d
+	xor	r12d,r8d
+	shrd	r14d,r14d,11
+	vpsrld	xmm6,xmm7,10
+	xor	r15d,r11d
+	add	r9d,r12d
+	shrd	r13d,r13d,6
+	vpsrlq	xmm7,xmm7,17
+	and	edi,r15d
+	xor	r14d,r10d
+	add	r9d,r13d
+	vpxor	xmm6,xmm6,xmm7
+	xor	edi,r11d
+	shrd	r14d,r14d,2
+	add	ebx,r9d
+	vpsrlq	xmm7,xmm7,2
+	add	r9d,edi
+	mov	r13d,ebx
+	add	r14d,r9d
+	vpxor	xmm6,xmm6,xmm7
+	shrd	r13d,r13d,14
+	mov	r9d,r14d
+	mov	r12d,ecx
+	vpshufb	xmm6,xmm6,xmm9
+	shrd	r14d,r14d,9
+	xor	r13d,ebx
+	xor	r12d,edx
+	vpaddd	xmm0,xmm0,xmm6
+	shrd	r13d,r13d,5
+	xor	r14d,r9d
+	and	r12d,ebx
+	vpaddd	xmm6,xmm0,XMMWORD[rbp]
+	xor	r13d,ebx
+	add	r8d,DWORD[12+rsp]
+	mov	edi,r9d
+	xor	r12d,edx
+	shrd	r14d,r14d,11
+	xor	edi,r10d
+	add	r8d,r12d
+	shrd	r13d,r13d,6
+	and	r15d,edi
+	xor	r14d,r9d
+	add	r8d,r13d
+	xor	r15d,r10d
+	shrd	r14d,r14d,2
+	add	eax,r8d
+	add	r8d,r15d
+	mov	r13d,eax
+	add	r14d,r8d
+	vmovdqa	XMMWORD[rsp],xmm6
+	vpalignr	xmm4,xmm2,xmm1,4
+	shrd	r13d,r13d,14
+	mov	r8d,r14d
+	mov	r12d,ebx
+	vpalignr	xmm7,xmm0,xmm3,4
+	shrd	r14d,r14d,9
+	xor	r13d,eax
+	xor	r12d,ecx
+	vpsrld	xmm6,xmm4,7
+	shrd	r13d,r13d,5
+	xor	r14d,r8d
+	and	r12d,eax
+	vpaddd	xmm1,xmm1,xmm7
+	xor	r13d,eax
+	add	edx,DWORD[16+rsp]
+	mov	r15d,r8d
+	vpsrld	xmm7,xmm4,3
+	xor	r12d,ecx
+	shrd	r14d,r14d,11
+	xor	r15d,r9d
+	vpslld	xmm5,xmm4,14
+	add	edx,r12d
+	shrd	r13d,r13d,6
+	and	edi,r15d
+	vpxor	xmm4,xmm7,xmm6
+	xor	r14d,r8d
+	add	edx,r13d
+	xor	edi,r9d
+	vpshufd	xmm7,xmm0,250
+	shrd	r14d,r14d,2
+	add	r11d,edx
+	add	edx,edi
+	vpsrld	xmm6,xmm6,11
+	mov	r13d,r11d
+	add	r14d,edx
+	shrd	r13d,r13d,14
+	vpxor	xmm4,xmm4,xmm5
+	mov	edx,r14d
+	mov	r12d,eax
+	shrd	r14d,r14d,9
+	vpslld	xmm5,xmm5,11
+	xor	r13d,r11d
+	xor	r12d,ebx
+	shrd	r13d,r13d,5
+	vpxor	xmm4,xmm4,xmm6
+	xor	r14d,edx
+	and	r12d,r11d
+	xor	r13d,r11d
+	vpsrld	xmm6,xmm7,10
+	add	ecx,DWORD[20+rsp]
+	mov	edi,edx
+	xor	r12d,ebx
+	vpxor	xmm4,xmm4,xmm5
+	shrd	r14d,r14d,11
+	xor	edi,r8d
+	add	ecx,r12d
+	vpsrlq	xmm7,xmm7,17
+	shrd	r13d,r13d,6
+	and	r15d,edi
+	xor	r14d,edx
+	vpaddd	xmm1,xmm1,xmm4
+	add	ecx,r13d
+	xor	r15d,r8d
+	shrd	r14d,r14d,2
+	vpxor	xmm6,xmm6,xmm7
+	add	r10d,ecx
+	add	ecx,r15d
+	mov	r13d,r10d
+	vpsrlq	xmm7,xmm7,2
+	add	r14d,ecx
+	shrd	r13d,r13d,14
+	mov	ecx,r14d
+	vpxor	xmm6,xmm6,xmm7
+	mov	r12d,r11d
+	shrd	r14d,r14d,9
+	xor	r13d,r10d
+	vpshufb	xmm6,xmm6,xmm8
+	xor	r12d,eax
+	shrd	r13d,r13d,5
+	xor	r14d,ecx
+	vpaddd	xmm1,xmm1,xmm6
+	and	r12d,r10d
+	xor	r13d,r10d
+	add	ebx,DWORD[24+rsp]
+	vpshufd	xmm7,xmm1,80
+	mov	r15d,ecx
+	xor	r12d,eax
+	shrd	r14d,r14d,11
+	vpsrld	xmm6,xmm7,10
+	xor	r15d,edx
+	add	ebx,r12d
+	shrd	r13d,r13d,6
+	vpsrlq	xmm7,xmm7,17
+	and	edi,r15d
+	xor	r14d,ecx
+	add	ebx,r13d
+	vpxor	xmm6,xmm6,xmm7
+	xor	edi,edx
+	shrd	r14d,r14d,2
+	add	r9d,ebx
+	vpsrlq	xmm7,xmm7,2
+	add	ebx,edi
+	mov	r13d,r9d
+	add	r14d,ebx
+	vpxor	xmm6,xmm6,xmm7
+	shrd	r13d,r13d,14
+	mov	ebx,r14d
+	mov	r12d,r10d
+	vpshufb	xmm6,xmm6,xmm9
+	shrd	r14d,r14d,9
+	xor	r13d,r9d
+	xor	r12d,r11d
+	vpaddd	xmm1,xmm1,xmm6
+	shrd	r13d,r13d,5
+	xor	r14d,ebx
+	and	r12d,r9d
+	vpaddd	xmm6,xmm1,XMMWORD[32+rbp]
+	xor	r13d,r9d
+	add	eax,DWORD[28+rsp]
+	mov	edi,ebx
+	xor	r12d,r11d
+	shrd	r14d,r14d,11
+	xor	edi,ecx
+	add	eax,r12d
+	shrd	r13d,r13d,6
+	and	r15d,edi
+	xor	r14d,ebx
+	add	eax,r13d
+	xor	r15d,ecx
+	shrd	r14d,r14d,2
+	add	r8d,eax
+	add	eax,r15d
+	mov	r13d,r8d
+	add	r14d,eax
+	vmovdqa	XMMWORD[16+rsp],xmm6
+	vpalignr	xmm4,xmm3,xmm2,4
+	shrd	r13d,r13d,14
+	mov	eax,r14d
+	mov	r12d,r9d
+	vpalignr	xmm7,xmm1,xmm0,4
+	shrd	r14d,r14d,9
+	xor	r13d,r8d
+	xor	r12d,r10d
+	vpsrld	xmm6,xmm4,7
+	shrd	r13d,r13d,5
+	xor	r14d,eax
+	and	r12d,r8d
+	vpaddd	xmm2,xmm2,xmm7
+	xor	r13d,r8d
+	add	r11d,DWORD[32+rsp]
+	mov	r15d,eax
+	vpsrld	xmm7,xmm4,3
+	xor	r12d,r10d
+	shrd	r14d,r14d,11
+	xor	r15d,ebx
+	vpslld	xmm5,xmm4,14
+	add	r11d,r12d
+	shrd	r13d,r13d,6
+	and	edi,r15d
+	vpxor	xmm4,xmm7,xmm6
+	xor	r14d,eax
+	add	r11d,r13d
+	xor	edi,ebx
+	vpshufd	xmm7,xmm1,250
+	shrd	r14d,r14d,2
+	add	edx,r11d
+	add	r11d,edi
+	vpsrld	xmm6,xmm6,11
+	mov	r13d,edx
+	add	r14d,r11d
+	shrd	r13d,r13d,14
+	vpxor	xmm4,xmm4,xmm5
+	mov	r11d,r14d
+	mov	r12d,r8d
+	shrd	r14d,r14d,9
+	vpslld	xmm5,xmm5,11
+	xor	r13d,edx
+	xor	r12d,r9d
+	shrd	r13d,r13d,5
+	vpxor	xmm4,xmm4,xmm6
+	xor	r14d,r11d
+	and	r12d,edx
+	xor	r13d,edx
+	vpsrld	xmm6,xmm7,10
+	add	r10d,DWORD[36+rsp]
+	mov	edi,r11d
+	xor	r12d,r9d
+	vpxor	xmm4,xmm4,xmm5
+	shrd	r14d,r14d,11
+	xor	edi,eax
+	add	r10d,r12d
+	vpsrlq	xmm7,xmm7,17
+	shrd	r13d,r13d,6
+	and	r15d,edi
+	xor	r14d,r11d
+	vpaddd	xmm2,xmm2,xmm4
+	add	r10d,r13d
+	xor	r15d,eax
+	shrd	r14d,r14d,2
+	vpxor	xmm6,xmm6,xmm7
+	add	ecx,r10d
+	add	r10d,r15d
+	mov	r13d,ecx
+	vpsrlq	xmm7,xmm7,2
+	add	r14d,r10d
+	shrd	r13d,r13d,14
+	mov	r10d,r14d
+	vpxor	xmm6,xmm6,xmm7
+	mov	r12d,edx
+	shrd	r14d,r14d,9
+	xor	r13d,ecx
+	vpshufb	xmm6,xmm6,xmm8
+	xor	r12d,r8d
+	shrd	r13d,r13d,5
+	xor	r14d,r10d
+	vpaddd	xmm2,xmm2,xmm6
+	and	r12d,ecx
+	xor	r13d,ecx
+	add	r9d,DWORD[40+rsp]
+	vpshufd	xmm7,xmm2,80
+	mov	r15d,r10d
+	xor	r12d,r8d
+	shrd	r14d,r14d,11
+	vpsrld	xmm6,xmm7,10
+	xor	r15d,r11d
+	add	r9d,r12d
+	shrd	r13d,r13d,6
+	vpsrlq	xmm7,xmm7,17
+	and	edi,r15d
+	xor	r14d,r10d
+	add	r9d,r13d
+	vpxor	xmm6,xmm6,xmm7
+	xor	edi,r11d
+	shrd	r14d,r14d,2
+	add	ebx,r9d
+	vpsrlq	xmm7,xmm7,2
+	add	r9d,edi
+	mov	r13d,ebx
+	add	r14d,r9d
+	vpxor	xmm6,xmm6,xmm7
+	shrd	r13d,r13d,14
+	mov	r9d,r14d
+	mov	r12d,ecx
+	vpshufb	xmm6,xmm6,xmm9
+	shrd	r14d,r14d,9
+	xor	r13d,ebx
+	xor	r12d,edx
+	vpaddd	xmm2,xmm2,xmm6
+	shrd	r13d,r13d,5
+	xor	r14d,r9d
+	and	r12d,ebx
+	vpaddd	xmm6,xmm2,XMMWORD[64+rbp]
+	xor	r13d,ebx
+	add	r8d,DWORD[44+rsp]
+	mov	edi,r9d
+	xor	r12d,edx
+	shrd	r14d,r14d,11
+	xor	edi,r10d
+	add	r8d,r12d
+	shrd	r13d,r13d,6
+	and	r15d,edi
+	xor	r14d,r9d
+	add	r8d,r13d
+	xor	r15d,r10d
+	shrd	r14d,r14d,2
+	add	eax,r8d
+	add	r8d,r15d
+	mov	r13d,eax
+	add	r14d,r8d
+	vmovdqa	XMMWORD[32+rsp],xmm6
+	vpalignr	xmm4,xmm0,xmm3,4
+	shrd	r13d,r13d,14
+	mov	r8d,r14d
+	mov	r12d,ebx
+	vpalignr	xmm7,xmm2,xmm1,4
+	shrd	r14d,r14d,9
+	xor	r13d,eax
+	xor	r12d,ecx
+	vpsrld	xmm6,xmm4,7
+	shrd	r13d,r13d,5
+	xor	r14d,r8d
+	and	r12d,eax
+	vpaddd	xmm3,xmm3,xmm7
+	xor	r13d,eax
+	add	edx,DWORD[48+rsp]
+	mov	r15d,r8d
+	vpsrld	xmm7,xmm4,3
+	xor	r12d,ecx
+	shrd	r14d,r14d,11
+	xor	r15d,r9d
+	vpslld	xmm5,xmm4,14
+	add	edx,r12d
+	shrd	r13d,r13d,6
+	and	edi,r15d
+	vpxor	xmm4,xmm7,xmm6
+	xor	r14d,r8d
+	add	edx,r13d
+	xor	edi,r9d
+	vpshufd	xmm7,xmm2,250
+	shrd	r14d,r14d,2
+	add	r11d,edx
+	add	edx,edi
+	vpsrld	xmm6,xmm6,11
+	mov	r13d,r11d
+	add	r14d,edx
+	shrd	r13d,r13d,14
+	vpxor	xmm4,xmm4,xmm5
+	mov	edx,r14d
+	mov	r12d,eax
+	shrd	r14d,r14d,9
+	vpslld	xmm5,xmm5,11
+	xor	r13d,r11d
+	xor	r12d,ebx
+	shrd	r13d,r13d,5
+	vpxor	xmm4,xmm4,xmm6
+	xor	r14d,edx
+	and	r12d,r11d
+	xor	r13d,r11d
+	vpsrld	xmm6,xmm7,10
+	add	ecx,DWORD[52+rsp]
+	mov	edi,edx
+	xor	r12d,ebx
+	vpxor	xmm4,xmm4,xmm5
+	shrd	r14d,r14d,11
+	xor	edi,r8d
+	add	ecx,r12d
+	vpsrlq	xmm7,xmm7,17
+	shrd	r13d,r13d,6
+	and	r15d,edi
+	xor	r14d,edx
+	vpaddd	xmm3,xmm3,xmm4
+	add	ecx,r13d
+	xor	r15d,r8d
+	shrd	r14d,r14d,2
+	vpxor	xmm6,xmm6,xmm7
+	add	r10d,ecx
+	add	ecx,r15d
+	mov	r13d,r10d
+	vpsrlq	xmm7,xmm7,2
+	add	r14d,ecx
+	shrd	r13d,r13d,14
+	mov	ecx,r14d
+	vpxor	xmm6,xmm6,xmm7
+	mov	r12d,r11d
+	shrd	r14d,r14d,9
+	xor	r13d,r10d
+	vpshufb	xmm6,xmm6,xmm8
+	xor	r12d,eax
+	shrd	r13d,r13d,5
+	xor	r14d,ecx
+	vpaddd	xmm3,xmm3,xmm6
+	and	r12d,r10d
+	xor	r13d,r10d
+	add	ebx,DWORD[56+rsp]
+	vpshufd	xmm7,xmm3,80
+	mov	r15d,ecx
+	xor	r12d,eax
+	shrd	r14d,r14d,11
+	vpsrld	xmm6,xmm7,10
+	xor	r15d,edx
+	add	ebx,r12d
+	shrd	r13d,r13d,6
+	vpsrlq	xmm7,xmm7,17
+	and	edi,r15d
+	xor	r14d,ecx
+	add	ebx,r13d
+	vpxor	xmm6,xmm6,xmm7
+	xor	edi,edx
+	shrd	r14d,r14d,2
+	add	r9d,ebx
+	vpsrlq	xmm7,xmm7,2
+	add	ebx,edi
+	mov	r13d,r9d
+	add	r14d,ebx
+	vpxor	xmm6,xmm6,xmm7
+	shrd	r13d,r13d,14
+	mov	ebx,r14d
+	mov	r12d,r10d
+	vpshufb	xmm6,xmm6,xmm9
+	shrd	r14d,r14d,9
+	xor	r13d,r9d
+	xor	r12d,r11d
+	vpaddd	xmm3,xmm3,xmm6
+	shrd	r13d,r13d,5
+	xor	r14d,ebx
+	and	r12d,r9d
+	vpaddd	xmm6,xmm3,XMMWORD[96+rbp]
+	xor	r13d,r9d
+	add	eax,DWORD[60+rsp]
+	mov	edi,ebx
+	xor	r12d,r11d
+	shrd	r14d,r14d,11
+	xor	edi,ecx
+	add	eax,r12d
+	shrd	r13d,r13d,6
+	and	r15d,edi
+	xor	r14d,ebx
+	add	eax,r13d
+	xor	r15d,ecx
+	shrd	r14d,r14d,2
+	add	r8d,eax
+	add	eax,r15d
+	mov	r13d,r8d
+	add	r14d,eax
+	vmovdqa	XMMWORD[48+rsp],xmm6
+	cmp	BYTE[131+rbp],0
+	jne	NEAR $L$avx_00_47
+	shrd	r13d,r13d,14
+	mov	eax,r14d
+	mov	r12d,r9d
+	shrd	r14d,r14d,9
+	xor	r13d,r8d
+	xor	r12d,r10d
+	shrd	r13d,r13d,5
+	xor	r14d,eax
+	and	r12d,r8d
+	xor	r13d,r8d
+	add	r11d,DWORD[rsp]
+	mov	r15d,eax
+	xor	r12d,r10d
+	shrd	r14d,r14d,11
+	xor	r15d,ebx
+	add	r11d,r12d
+	shrd	r13d,r13d,6
+	and	edi,r15d
+	xor	r14d,eax
+	add	r11d,r13d
+	xor	edi,ebx
+	shrd	r14d,r14d,2
+	add	edx,r11d
+	add	r11d,edi
+	mov	r13d,edx
+	add	r14d,r11d
+	shrd	r13d,r13d,14
+	mov	r11d,r14d
+	mov	r12d,r8d
+	shrd	r14d,r14d,9
+	xor	r13d,edx
+	xor	r12d,r9d
+	shrd	r13d,r13d,5
+	xor	r14d,r11d
+	and	r12d,edx
+	xor	r13d,edx
+	add	r10d,DWORD[4+rsp]
+	mov	edi,r11d
+	xor	r12d,r9d
+	shrd	r14d,r14d,11
+	xor	edi,eax
+	add	r10d,r12d
+	shrd	r13d,r13d,6
+	and	r15d,edi
+	xor	r14d,r11d
+	add	r10d,r13d
+	xor	r15d,eax
+	shrd	r14d,r14d,2
+	add	ecx,r10d
+	add	r10d,r15d
+	mov	r13d,ecx
+	add	r14d,r10d
+	shrd	r13d,r13d,14
+	mov	r10d,r14d
+	mov	r12d,edx
+	shrd	r14d,r14d,9
+	xor	r13d,ecx
+	xor	r12d,r8d
+	shrd	r13d,r13d,5
+	xor	r14d,r10d
+	and	r12d,ecx
+	xor	r13d,ecx
+	add	r9d,DWORD[8+rsp]
+	mov	r15d,r10d
+	xor	r12d,r8d
+	shrd	r14d,r14d,11
+	xor	r15d,r11d
+	add	r9d,r12d
+	shrd	r13d,r13d,6
+	and	edi,r15d
+	xor	r14d,r10d
+	add	r9d,r13d
+	xor	edi,r11d
+	shrd	r14d,r14d,2
+	add	ebx,r9d
+	add	r9d,edi
+	mov	r13d,ebx
+	add	r14d,r9d
+	shrd	r13d,r13d,14
+	mov	r9d,r14d
+	mov	r12d,ecx
+	shrd	r14d,r14d,9
+	xor	r13d,ebx
+	xor	r12d,edx
+	shrd	r13d,r13d,5
+	xor	r14d,r9d
+	and	r12d,ebx
+	xor	r13d,ebx
+	add	r8d,DWORD[12+rsp]
+	mov	edi,r9d
+	xor	r12d,edx
+	shrd	r14d,r14d,11
+	xor	edi,r10d
+	add	r8d,r12d
+	shrd	r13d,r13d,6
+	and	r15d,edi
+	xor	r14d,r9d
+	add	r8d,r13d
+	xor	r15d,r10d
+	shrd	r14d,r14d,2
+	add	eax,r8d
+	add	r8d,r15d
+	mov	r13d,eax
+	add	r14d,r8d
+	shrd	r13d,r13d,14
+	mov	r8d,r14d
+	mov	r12d,ebx
+	shrd	r14d,r14d,9
+	xor	r13d,eax
+	xor	r12d,ecx
+	shrd	r13d,r13d,5
+	xor	r14d,r8d
+	and	r12d,eax
+	xor	r13d,eax
+	add	edx,DWORD[16+rsp]
+	mov	r15d,r8d
+	xor	r12d,ecx
+	shrd	r14d,r14d,11
+	xor	r15d,r9d
+	add	edx,r12d
+	shrd	r13d,r13d,6
+	and	edi,r15d
+	xor	r14d,r8d
+	add	edx,r13d
+	xor	edi,r9d
+	shrd	r14d,r14d,2
+	add	r11d,edx
+	add	edx,edi
+	mov	r13d,r11d
+	add	r14d,edx
+	shrd	r13d,r13d,14
+	mov	edx,r14d
+	mov	r12d,eax
+	shrd	r14d,r14d,9
+	xor	r13d,r11d
+	xor	r12d,ebx
+	shrd	r13d,r13d,5
+	xor	r14d,edx
+	and	r12d,r11d
+	xor	r13d,r11d
+	add	ecx,DWORD[20+rsp]
+	mov	edi,edx
+	xor	r12d,ebx
+	shrd	r14d,r14d,11
+	xor	edi,r8d
+	add	ecx,r12d
+	shrd	r13d,r13d,6
+	and	r15d,edi
+	xor	r14d,edx
+	add	ecx,r13d
+	xor	r15d,r8d
+	shrd	r14d,r14d,2
+	add	r10d,ecx
+	add	ecx,r15d
+	mov	r13d,r10d
+	add	r14d,ecx
+	shrd	r13d,r13d,14
+	mov	ecx,r14d
+	mov	r12d,r11d
+	shrd	r14d,r14d,9
+	xor	r13d,r10d
+	xor	r12d,eax
+	shrd	r13d,r13d,5
+	xor	r14d,ecx
+	and	r12d,r10d
+	xor	r13d,r10d
+	add	ebx,DWORD[24+rsp]
+	mov	r15d,ecx
+	xor	r12d,eax
+	shrd	r14d,r14d,11
+	xor	r15d,edx
+	add	ebx,r12d
+	shrd	r13d,r13d,6
+	and	edi,r15d
+	xor	r14d,ecx
+	add	ebx,r13d
+	xor	edi,edx
+	shrd	r14d,r14d,2
+	add	r9d,ebx
+	add	ebx,edi
+	mov	r13d,r9d
+	add	r14d,ebx
+	shrd	r13d,r13d,14
+	mov	ebx,r14d
+	mov	r12d,r10d
+	shrd	r14d,r14d,9
+	xor	r13d,r9d
+	xor	r12d,r11d
+	shrd	r13d,r13d,5
+	xor	r14d,ebx
+	and	r12d,r9d
+	xor	r13d,r9d
+	add	eax,DWORD[28+rsp]
+	mov	edi,ebx
+	xor	r12d,r11d
+	shrd	r14d,r14d,11
+	xor	edi,ecx
+	add	eax,r12d
+	shrd	r13d,r13d,6
+	and	r15d,edi
+	xor	r14d,ebx
+	add	eax,r13d
+	xor	r15d,ecx
+	shrd	r14d,r14d,2
+	add	r8d,eax
+	add	eax,r15d
+	mov	r13d,r8d
+	add	r14d,eax
+	shrd	r13d,r13d,14
+	mov	eax,r14d
+	mov	r12d,r9d
+	shrd	r14d,r14d,9
+	xor	r13d,r8d
+	xor	r12d,r10d
+	shrd	r13d,r13d,5
+	xor	r14d,eax
+	and	r12d,r8d
+	xor	r13d,r8d
+	add	r11d,DWORD[32+rsp]
+	mov	r15d,eax
+	xor	r12d,r10d
+	shrd	r14d,r14d,11
+	xor	r15d,ebx
+	add	r11d,r12d
+	shrd	r13d,r13d,6
+	and	edi,r15d
+	xor	r14d,eax
+	add	r11d,r13d
+	xor	edi,ebx
+	shrd	r14d,r14d,2
+	add	edx,r11d
+	add	r11d,edi
+	mov	r13d,edx
+	add	r14d,r11d
+	shrd	r13d,r13d,14
+	mov	r11d,r14d
+	mov	r12d,r8d
+	shrd	r14d,r14d,9
+	xor	r13d,edx
+	xor	r12d,r9d
+	shrd	r13d,r13d,5
+	xor	r14d,r11d
+	and	r12d,edx
+	xor	r13d,edx
+	add	r10d,DWORD[36+rsp]
+	mov	edi,r11d
+	xor	r12d,r9d
+	shrd	r14d,r14d,11
+	xor	edi,eax
+	add	r10d,r12d
+	shrd	r13d,r13d,6
+	and	r15d,edi
+	xor	r14d,r11d
+	add	r10d,r13d
+	xor	r15d,eax
+	shrd	r14d,r14d,2
+	add	ecx,r10d
+	add	r10d,r15d
+	mov	r13d,ecx
+	add	r14d,r10d
+	shrd	r13d,r13d,14
+	mov	r10d,r14d
+	mov	r12d,edx
+	shrd	r14d,r14d,9
+	xor	r13d,ecx
+	xor	r12d,r8d
+	shrd	r13d,r13d,5
+	xor	r14d,r10d
+	and	r12d,ecx
+	xor	r13d,ecx
+	add	r9d,DWORD[40+rsp]
+	mov	r15d,r10d
+	xor	r12d,r8d
+	shrd	r14d,r14d,11
+	xor	r15d,r11d
+	add	r9d,r12d
+	shrd	r13d,r13d,6
+	and	edi,r15d
+	xor	r14d,r10d
+	add	r9d,r13d
+	xor	edi,r11d
+	shrd	r14d,r14d,2
+	add	ebx,r9d
+	add	r9d,edi
+	mov	r13d,ebx
+	add	r14d,r9d
+	shrd	r13d,r13d,14
+	mov	r9d,r14d
+	mov	r12d,ecx
+	shrd	r14d,r14d,9
+	xor	r13d,ebx
+	xor	r12d,edx
+	shrd	r13d,r13d,5
+	xor	r14d,r9d
+	and	r12d,ebx
+	xor	r13d,ebx
+	add	r8d,DWORD[44+rsp]
+	mov	edi,r9d
+	xor	r12d,edx
+	shrd	r14d,r14d,11
+	xor	edi,r10d
+	add	r8d,r12d
+	shrd	r13d,r13d,6
+	and	r15d,edi
+	xor	r14d,r9d
+	add	r8d,r13d
+	xor	r15d,r10d
+	shrd	r14d,r14d,2
+	add	eax,r8d
+	add	r8d,r15d
+	mov	r13d,eax
+	add	r14d,r8d
+	shrd	r13d,r13d,14
+	mov	r8d,r14d
+	mov	r12d,ebx
+	shrd	r14d,r14d,9
+	xor	r13d,eax
+	xor	r12d,ecx
+	shrd	r13d,r13d,5
+	xor	r14d,r8d
+	and	r12d,eax
+	xor	r13d,eax
+	add	edx,DWORD[48+rsp]
+	mov	r15d,r8d
+	xor	r12d,ecx
+	shrd	r14d,r14d,11
+	xor	r15d,r9d
+	add	edx,r12d
+	shrd	r13d,r13d,6
+	and	edi,r15d
+	xor	r14d,r8d
+	add	edx,r13d
+	xor	edi,r9d
+	shrd	r14d,r14d,2
+	add	r11d,edx
+	add	edx,edi
+	mov	r13d,r11d
+	add	r14d,edx
+	shrd	r13d,r13d,14
+	mov	edx,r14d
+	mov	r12d,eax
+	shrd	r14d,r14d,9
+	xor	r13d,r11d
+	xor	r12d,ebx
+	shrd	r13d,r13d,5
+	xor	r14d,edx
+	and	r12d,r11d
+	xor	r13d,r11d
+	add	ecx,DWORD[52+rsp]
+	mov	edi,edx
+	xor	r12d,ebx
+	shrd	r14d,r14d,11
+	xor	edi,r8d
+	add	ecx,r12d
+	shrd	r13d,r13d,6
+	and	r15d,edi
+	xor	r14d,edx
+	add	ecx,r13d
+	xor	r15d,r8d
+	shrd	r14d,r14d,2
+	add	r10d,ecx
+	add	ecx,r15d
+	mov	r13d,r10d
+	add	r14d,ecx
+	shrd	r13d,r13d,14
+	mov	ecx,r14d
+	mov	r12d,r11d
+	shrd	r14d,r14d,9
+	xor	r13d,r10d
+	xor	r12d,eax
+	shrd	r13d,r13d,5
+	xor	r14d,ecx
+	and	r12d,r10d
+	xor	r13d,r10d
+	add	ebx,DWORD[56+rsp]
+	mov	r15d,ecx
+	xor	r12d,eax
+	shrd	r14d,r14d,11
+	xor	r15d,edx
+	add	ebx,r12d
+	shrd	r13d,r13d,6
+	and	edi,r15d
+	xor	r14d,ecx
+	add	ebx,r13d
+	xor	edi,edx
+	shrd	r14d,r14d,2
+	add	r9d,ebx
+	add	ebx,edi
+	mov	r13d,r9d
+	add	r14d,ebx
+	shrd	r13d,r13d,14
+	mov	ebx,r14d
+	mov	r12d,r10d
+	shrd	r14d,r14d,9
+	xor	r13d,r9d
+	xor	r12d,r11d
+	shrd	r13d,r13d,5
+	xor	r14d,ebx
+	and	r12d,r9d
+	xor	r13d,r9d
+	add	eax,DWORD[60+rsp]
+	mov	edi,ebx
+	xor	r12d,r11d
+	shrd	r14d,r14d,11
+	xor	edi,ecx
+	add	eax,r12d
+	shrd	r13d,r13d,6
+	and	r15d,edi
+	xor	r14d,ebx
+	add	eax,r13d
+	xor	r15d,ecx
+	shrd	r14d,r14d,2
+	add	r8d,eax
+	add	eax,r15d
+	mov	r13d,r8d
+	add	r14d,eax
+	mov	rdi,QWORD[((64+0))+rsp]
+	mov	eax,r14d
+
+	add	eax,DWORD[rdi]
+	lea	rsi,[64+rsi]
+	add	ebx,DWORD[4+rdi]
+	add	ecx,DWORD[8+rdi]
+	add	edx,DWORD[12+rdi]
+	add	r8d,DWORD[16+rdi]
+	add	r9d,DWORD[20+rdi]
+	add	r10d,DWORD[24+rdi]
+	add	r11d,DWORD[28+rdi]
+
+	cmp	rsi,QWORD[((64+16))+rsp]
+
+	mov	DWORD[rdi],eax
+	mov	DWORD[4+rdi],ebx
+	mov	DWORD[8+rdi],ecx
+	mov	DWORD[12+rdi],edx
+	mov	DWORD[16+rdi],r8d
+	mov	DWORD[20+rdi],r9d
+	mov	DWORD[24+rdi],r10d
+	mov	DWORD[28+rdi],r11d
+	jb	NEAR $L$loop_avx
+
+	mov	rsi,QWORD[((64+24))+rsp]
+	vzeroupper
+	movaps	xmm6,XMMWORD[((64+32))+rsp]
+	movaps	xmm7,XMMWORD[((64+48))+rsp]
+	movaps	xmm8,XMMWORD[((64+64))+rsp]
+	movaps	xmm9,XMMWORD[((64+80))+rsp]
+	mov	r15,QWORD[rsi]
+	mov	r14,QWORD[8+rsi]
+	mov	r13,QWORD[16+rsi]
+	mov	r12,QWORD[24+rsi]
+	mov	rbp,QWORD[32+rsi]
+	mov	rbx,QWORD[40+rsi]
+	lea	rsp,[48+rsi]
+$L$epilogue_avx:
+	mov	rdi,QWORD[8+rsp]	;WIN64 epilogue
+	mov	rsi,QWORD[16+rsp]
+	DB	0F3h,0C3h		;repret
+$L$SEH_end_sha256_block_data_order_avx:
 EXTERN	__imp_RtlVirtualUnwind
 
 ALIGN	16
@@ -2982,6 +4063,9 @@
 	DD	$L$SEH_begin_sha256_block_data_order_ssse3 wrt ..imagebase
 	DD	$L$SEH_end_sha256_block_data_order_ssse3 wrt ..imagebase
 	DD	$L$SEH_info_sha256_block_data_order_ssse3 wrt ..imagebase
+	DD	$L$SEH_begin_sha256_block_data_order_avx wrt ..imagebase
+	DD	$L$SEH_end_sha256_block_data_order_avx wrt ..imagebase
+	DD	$L$SEH_info_sha256_block_data_order_avx wrt ..imagebase
 section	.xdata rdata align=8
 ALIGN	8
 $L$SEH_info_sha256_block_data_order:
@@ -2992,3 +4076,7 @@
 DB	9,0,0,0
 	DD	se_handler wrt ..imagebase
 	DD	$L$prologue_ssse3 wrt ..imagebase,$L$epilogue_ssse3 wrt ..imagebase
+$L$SEH_info_sha256_block_data_order_avx:
+DB	9,0,0,0
+	DD	se_handler wrt ..imagebase
+	DD	$L$prologue_avx wrt ..imagebase,$L$epilogue_avx wrt ..imagebase

third_party/boringssl/win-x86_64/crypto/sha/sha512-x86_64.asm

diff --git a/third_party/boringssl/win-x86_64/crypto/sha/sha512-x86_64.asm b/third_party/boringssl/win-x86_64/crypto/sha/sha512-x86_64.asm
index b76cc0e..71449cd 100644
--- a/third_party/boringssl/win-x86_64/crypto/sha/sha512-x86_64.asm
+++ b/third_party/boringssl/win-x86_64/crypto/sha/sha512-x86_64.asm
@@ -19,6 +19,17 @@
 	mov	rdx,r8
 
 
+	lea	r11,[OPENSSL_ia32cap_P]
+	mov	r9d,DWORD[r11]
+	mov	r10d,DWORD[4+r11]
+	mov	r11d,DWORD[8+r11]
+	test	r10d,2048
+	jnz	NEAR $L$xop_shortcut
+	and	r9d,1073741824
+	and	r10d,268435968
+	or	r10d,r9d
+	cmp	r10d,1342177792
+	je	NEAR $L$avx_shortcut
 	push	rbx
 	push	rbp
 	push	r12
@@ -1801,6 +1812,2282 @@
 DB	52,44,32,67,82,89,80,84,79,71,65,77,83,32,98,121
 DB	32,60,97,112,112,114,111,64,111,112,101,110,115,115,108,46
 DB	111,114,103,62,0
+
+ALIGN	64
+sha512_block_data_order_xop:
+	mov	QWORD[8+rsp],rdi	;WIN64 prologue
+	mov	QWORD[16+rsp],rsi
+	mov	rax,rsp
+$L$SEH_begin_sha512_block_data_order_xop:
+	mov	rdi,rcx
+	mov	rsi,rdx
+	mov	rdx,r8
+
+
+$L$xop_shortcut:
+	push	rbx
+	push	rbp
+	push	r12
+	push	r13
+	push	r14
+	push	r15
+	mov	r11,rsp
+	shl	rdx,4
+	sub	rsp,256
+	lea	rdx,[rdx*8+rsi]
+	and	rsp,-64
+	mov	QWORD[((128+0))+rsp],rdi
+	mov	QWORD[((128+8))+rsp],rsi
+	mov	QWORD[((128+16))+rsp],rdx
+	mov	QWORD[((128+24))+rsp],r11
+	movaps	XMMWORD[(128+32)+rsp],xmm6
+	movaps	XMMWORD[(128+48)+rsp],xmm7
+	movaps	XMMWORD[(128+64)+rsp],xmm8
+	movaps	XMMWORD[(128+80)+rsp],xmm9
+	movaps	XMMWORD[(128+96)+rsp],xmm10
+	movaps	XMMWORD[(128+112)+rsp],xmm11
+$L$prologue_xop:
+
+	vzeroupper
+	mov	rax,QWORD[rdi]
+	mov	rbx,QWORD[8+rdi]
+	mov	rcx,QWORD[16+rdi]
+	mov	rdx,QWORD[24+rdi]
+	mov	r8,QWORD[32+rdi]
+	mov	r9,QWORD[40+rdi]
+	mov	r10,QWORD[48+rdi]
+	mov	r11,QWORD[56+rdi]
+	jmp	NEAR $L$loop_xop
+ALIGN	16
+$L$loop_xop:
+	vmovdqa	xmm11,XMMWORD[((K512+1280))]
+	vmovdqu	xmm0,XMMWORD[rsi]
+	lea	rbp,[((K512+128))]
+	vmovdqu	xmm1,XMMWORD[16+rsi]
+	vmovdqu	xmm2,XMMWORD[32+rsi]
+	vpshufb	xmm0,xmm0,xmm11
+	vmovdqu	xmm3,XMMWORD[48+rsi]
+	vpshufb	xmm1,xmm1,xmm11
+	vmovdqu	xmm4,XMMWORD[64+rsi]
+	vpshufb	xmm2,xmm2,xmm11
+	vmovdqu	xmm5,XMMWORD[80+rsi]
+	vpshufb	xmm3,xmm3,xmm11
+	vmovdqu	xmm6,XMMWORD[96+rsi]
+	vpshufb	xmm4,xmm4,xmm11
+	vmovdqu	xmm7,XMMWORD[112+rsi]
+	vpshufb	xmm5,xmm5,xmm11
+	vpaddq	xmm8,xmm0,XMMWORD[((-128))+rbp]
+	vpshufb	xmm6,xmm6,xmm11
+	vpaddq	xmm9,xmm1,XMMWORD[((-96))+rbp]
+	vpshufb	xmm7,xmm7,xmm11
+	vpaddq	xmm10,xmm2,XMMWORD[((-64))+rbp]
+	vpaddq	xmm11,xmm3,XMMWORD[((-32))+rbp]
+	vmovdqa	XMMWORD[rsp],xmm8
+	vpaddq	xmm8,xmm4,XMMWORD[rbp]
+	vmovdqa	XMMWORD[16+rsp],xmm9
+	vpaddq	xmm9,xmm5,XMMWORD[32+rbp]
+	vmovdqa	XMMWORD[32+rsp],xmm10
+	vpaddq	xmm10,xmm6,XMMWORD[64+rbp]
+	vmovdqa	XMMWORD[48+rsp],xmm11
+	vpaddq	xmm11,xmm7,XMMWORD[96+rbp]
+	vmovdqa	XMMWORD[64+rsp],xmm8
+	mov	r14,rax
+	vmovdqa	XMMWORD[80+rsp],xmm9
+	mov	rdi,rbx
+	vmovdqa	XMMWORD[96+rsp],xmm10
+	xor	rdi,rcx
+	vmovdqa	XMMWORD[112+rsp],xmm11
+	mov	r13,r8
+	jmp	NEAR $L$xop_00_47
+
+ALIGN	16
+$L$xop_00_47:
+	add	rbp,256
+	vpalignr	xmm8,xmm1,xmm0,8
+	ror	r13,23
+	mov	rax,r14
+	vpalignr	xmm11,xmm5,xmm4,8
+	mov	r12,r9
+	ror	r14,5
+DB	143,72,120,195,200,56
+	xor	r13,r8
+	xor	r12,r10
+	vpsrlq	xmm8,xmm8,7
+	ror	r13,4
+	xor	r14,rax
+	vpaddq	xmm0,xmm0,xmm11
+	and	r12,r8
+	xor	r13,r8
+	add	r11,QWORD[rsp]
+	mov	r15,rax
+DB	143,72,120,195,209,7
+	xor	r12,r10
+	ror	r14,6
+	vpxor	xmm8,xmm8,xmm9
+	xor	r15,rbx
+	add	r11,r12
+	ror	r13,14
+	and	rdi,r15
+DB	143,104,120,195,223,3
+	xor	r14,rax
+	add	r11,r13
+	vpxor	xmm8,xmm8,xmm10
+	xor	rdi,rbx
+	ror	r14,28
+	vpsrlq	xmm10,xmm7,6
+	add	rdx,r11
+	add	r11,rdi
+	vpaddq	xmm0,xmm0,xmm8
+	mov	r13,rdx
+	add	r14,r11
+DB	143,72,120,195,203,42
+	ror	r13,23
+	mov	r11,r14
+	vpxor	xmm11,xmm11,xmm10
+	mov	r12,r8
+	ror	r14,5
+	xor	r13,rdx
+	xor	r12,r9
+	vpxor	xmm11,xmm11,xmm9
+	ror	r13,4
+	xor	r14,r11
+	and	r12,rdx
+	xor	r13,rdx
+	vpaddq	xmm0,xmm0,xmm11
+	add	r10,QWORD[8+rsp]
+	mov	rdi,r11
+	xor	r12,r9
+	ror	r14,6
+	vpaddq	xmm10,xmm0,XMMWORD[((-128))+rbp]
+	xor	rdi,rax
+	add	r10,r12
+	ror	r13,14
+	and	r15,rdi
+	xor	r14,r11
+	add	r10,r13
+	xor	r15,rax
+	ror	r14,28
+	add	rcx,r10
+	add	r10,r15
+	mov	r13,rcx
+	add	r14,r10
+	vmovdqa	XMMWORD[rsp],xmm10
+	vpalignr	xmm8,xmm2,xmm1,8
+	ror	r13,23
+	mov	r10,r14
+	vpalignr	xmm11,xmm6,xmm5,8
+	mov	r12,rdx
+	ror	r14,5
+DB	143,72,120,195,200,56
+	xor	r13,rcx
+	xor	r12,r8
+	vpsrlq	xmm8,xmm8,7
+	ror	r13,4
+	xor	r14,r10
+	vpaddq	xmm1,xmm1,xmm11
+	and	r12,rcx
+	xor	r13,rcx
+	add	r9,QWORD[16+rsp]
+	mov	r15,r10
+DB	143,72,120,195,209,7
+	xor	r12,r8
+	ror	r14,6
+	vpxor	xmm8,xmm8,xmm9
+	xor	r15,r11
+	add	r9,r12
+	ror	r13,14
+	and	rdi,r15
+DB	143,104,120,195,216,3
+	xor	r14,r10
+	add	r9,r13
+	vpxor	xmm8,xmm8,xmm10
+	xor	rdi,r11
+	ror	r14,28
+	vpsrlq	xmm10,xmm0,6
+	add	rbx,r9
+	add	r9,rdi
+	vpaddq	xmm1,xmm1,xmm8
+	mov	r13,rbx
+	add	r14,r9
+DB	143,72,120,195,203,42
+	ror	r13,23
+	mov	r9,r14
+	vpxor	xmm11,xmm11,xmm10
+	mov	r12,rcx
+	ror	r14,5
+	xor	r13,rbx
+	xor	r12,rdx
+	vpxor	xmm11,xmm11,xmm9
+	ror	r13,4
+	xor	r14,r9
+	and	r12,rbx
+	xor	r13,rbx
+	vpaddq	xmm1,xmm1,xmm11
+	add	r8,QWORD[24+rsp]
+	mov	rdi,r9
+	xor	r12,rdx
+	ror	r14,6
+	vpaddq	xmm10,xmm1,XMMWORD[((-96))+rbp]
+	xor	rdi,r10
+	add	r8,r12
+	ror	r13,14
+	and	r15,rdi
+	xor	r14,r9
+	add	r8,r13
+	xor	r15,r10
+	ror	r14,28
+	add	rax,r8
+	add	r8,r15
+	mov	r13,rax
+	add	r14,r8
+	vmovdqa	XMMWORD[16+rsp],xmm10
+	vpalignr	xmm8,xmm3,xmm2,8
+	ror	r13,23
+	mov	r8,r14
+	vpalignr	xmm11,xmm7,xmm6,8
+	mov	r12,rbx
+	ror	r14,5
+DB	143,72,120,195,200,56
+	xor	r13,rax
+	xor	r12,rcx
+	vpsrlq	xmm8,xmm8,7
+	ror	r13,4
+	xor	r14,r8
+	vpaddq	xmm2,xmm2,xmm11
+	and	r12,rax
+	xor	r13,rax
+	add	rdx,QWORD[32+rsp]
+	mov	r15,r8
+DB	143,72,120,195,209,7
+	xor	r12,rcx
+	ror	r14,6
+	vpxor	xmm8,xmm8,xmm9
+	xor	r15,r9
+	add	rdx,r12
+	ror	r13,14
+	and	rdi,r15
+DB	143,104,120,195,217,3
+	xor	r14,r8
+	add	rdx,r13
+	vpxor	xmm8,xmm8,xmm10
+	xor	rdi,r9
+	ror	r14,28
+	vpsrlq	xmm10,xmm1,6
+	add	r11,rdx
+	add	rdx,rdi
+	vpaddq	xmm2,xmm2,xmm8
+	mov	r13,r11
+	add	r14,rdx
+DB	143,72,120,195,203,42
+	ror	r13,23
+	mov	rdx,r14
+	vpxor	xmm11,xmm11,xmm10
+	mov	r12,rax
+	ror	r14,5
+	xor	r13,r11
+	xor	r12,rbx
+	vpxor	xmm11,xmm11,xmm9
+	ror	r13,4
+	xor	r14,rdx
+	and	r12,r11
+	xor	r13,r11
+	vpaddq	xmm2,xmm2,xmm11
+	add	rcx,QWORD[40+rsp]
+	mov	rdi,rdx
+	xor	r12,rbx
+	ror	r14,6
+	vpaddq	xmm10,xmm2,XMMWORD[((-64))+rbp]
+	xor	rdi,r8
+	add	rcx,r12
+	ror	r13,14
+	and	r15,rdi
+	xor	r14,rdx
+	add	rcx,r13
+	xor	r15,r8
+	ror	r14,28
+	add	r10,rcx
+	add	rcx,r15
+	mov	r13,r10
+	add	r14,rcx
+	vmovdqa	XMMWORD[32+rsp],xmm10
+	vpalignr	xmm8,xmm4,xmm3,8
+	ror	r13,23
+	mov	rcx,r14
+	vpalignr	xmm11,xmm0,xmm7,8
+	mov	r12,r11
+	ror	r14,5
+DB	143,72,120,195,200,56
+	xor	r13,r10
+	xor	r12,rax
+	vpsrlq	xmm8,xmm8,7
+	ror	r13,4
+	xor	r14,rcx
+	vpaddq	xmm3,xmm3,xmm11
+	and	r12,r10
+	xor	r13,r10
+	add	rbx,QWORD[48+rsp]
+	mov	r15,rcx
+DB	143,72,120,195,209,7
+	xor	r12,rax
+	ror	r14,6
+	vpxor	xmm8,xmm8,xmm9
+	xor	r15,rdx
+	add	rbx,r12
+	ror	r13,14
+	and	rdi,r15
+DB	143,104,120,195,218,3
+	xor	r14,rcx
+	add	rbx,r13
+	vpxor	xmm8,xmm8,xmm10
+	xor	rdi,rdx
+	ror	r14,28
+	vpsrlq	xmm10,xmm2,6
+	add	r9,rbx
+	add	rbx,rdi
+	vpaddq	xmm3,xmm3,xmm8
+	mov	r13,r9
+	add	r14,rbx
+DB	143,72,120,195,203,42
+	ror	r13,23
+	mov	rbx,r14
+	vpxor	xmm11,xmm11,xmm10
+	mov	r12,r10
+	ror	r14,5
+	xor	r13,r9
+	xor	r12,r11
+	vpxor	xmm11,xmm11,xmm9
+	ror	r13,4
+	xor	r14,rbx
+	and	r12,r9
+	xor	r13,r9
+	vpaddq	xmm3,xmm3,xmm11
+	add	rax,QWORD[56+rsp]
+	mov	rdi,rbx
+	xor	r12,r11
+	ror	r14,6
+	vpaddq	xmm10,xmm3,XMMWORD[((-32))+rbp]
+	xor	rdi,rcx
+	add	rax,r12
+	ror	r13,14
+	and	r15,rdi
+	xor	r14,rbx
+	add	rax,r13
+	xor	r15,rcx
+	ror	r14,28
+	add	r8,rax
+	add	rax,r15
+	mov	r13,r8
+	add	r14,rax
+	vmovdqa	XMMWORD[48+rsp],xmm10
+	vpalignr	xmm8,xmm5,xmm4,8
+	ror	r13,23
+	mov	rax,r14
+	vpalignr	xmm11,xmm1,xmm0,8
+	mov	r12,r9
+	ror	r14,5
+DB	143,72,120,195,200,56
+	xor	r13,r8
+	xor	r12,r10
+	vpsrlq	xmm8,xmm8,7
+	ror	r13,4
+	xor	r14,rax
+	vpaddq	xmm4,xmm4,xmm11
+	and	r12,r8
+	xor	r13,r8
+	add	r11,QWORD[64+rsp]
+	mov	r15,rax
+DB	143,72,120,195,209,7
+	xor	r12,r10
+	ror	r14,6
+	vpxor	xmm8,xmm8,xmm9
+	xor	r15,rbx
+	add	r11,r12
+	ror	r13,14
+	and	rdi,r15
+DB	143,104,120,195,219,3
+	xor	r14,rax
+	add	r11,r13
+	vpxor	xmm8,xmm8,xmm10
+	xor	rdi,rbx
+	ror	r14,28
+	vpsrlq	xmm10,xmm3,6
+	add	rdx,r11
+	add	r11,rdi
+	vpaddq	xmm4,xmm4,xmm8
+	mov	r13,rdx
+	add	r14,r11
+DB	143,72,120,195,203,42
+	ror	r13,23
+	mov	r11,r14
+	vpxor	xmm11,xmm11,xmm10
+	mov	r12,r8
+	ror	r14,5
+	xor	r13,rdx
+	xor	r12,r9
+	vpxor	xmm11,xmm11,xmm9
+	ror	r13,4
+	xor	r14,r11
+	and	r12,rdx
+	xor	r13,rdx
+	vpaddq	xmm4,xmm4,xmm11
+	add	r10,QWORD[72+rsp]
+	mov	rdi,r11
+	xor	r12,r9
+	ror	r14,6
+	vpaddq	xmm10,xmm4,XMMWORD[rbp]
+	xor	rdi,rax
+	add	r10,r12
+	ror	r13,14
+	and	r15,rdi
+	xor	r14,r11
+	add	r10,r13
+	xor	r15,rax
+	ror	r14,28
+	add	rcx,r10
+	add	r10,r15
+	mov	r13,rcx
+	add	r14,r10
+	vmovdqa	XMMWORD[64+rsp],xmm10
+	vpalignr	xmm8,xmm6,xmm5,8
+	ror	r13,23
+	mov	r10,r14
+	vpalignr	xmm11,xmm2,xmm1,8
+	mov	r12,rdx
+	ror	r14,5
+DB	143,72,120,195,200,56
+	xor	r13,rcx
+	xor	r12,r8
+	vpsrlq	xmm8,xmm8,7
+	ror	r13,4
+	xor	r14,r10
+	vpaddq	xmm5,xmm5,xmm11
+	and	r12,rcx
+	xor	r13,rcx
+	add	r9,QWORD[80+rsp]
+	mov	r15,r10
+DB	143,72,120,195,209,7
+	xor	r12,r8
+	ror	r14,6
+	vpxor	xmm8,xmm8,xmm9
+	xor	r15,r11
+	add	r9,r12
+	ror	r13,14
+	and	rdi,r15
+DB	143,104,120,195,220,3
+	xor	r14,r10
+	add	r9,r13
+	vpxor	xmm8,xmm8,xmm10
+	xor	rdi,r11
+	ror	r14,28
+	vpsrlq	xmm10,xmm4,6
+	add	rbx,r9
+	add	r9,rdi
+	vpaddq	xmm5,xmm5,xmm8
+	mov	r13,rbx
+	add	r14,r9
+DB	143,72,120,195,203,42
+	ror	r13,23
+	mov	r9,r14
+	vpxor	xmm11,xmm11,xmm10
+	mov	r12,rcx
+	ror	r14,5
+	xor	r13,rbx
+	xor	r12,rdx
+	vpxor	xmm11,xmm11,xmm9
+	ror	r13,4
+	xor	r14,r9
+	and	r12,rbx
+	xor	r13,rbx
+	vpaddq	xmm5,xmm5,xmm11
+	add	r8,QWORD[88+rsp]
+	mov	rdi,r9
+	xor	r12,rdx
+	ror	r14,6
+	vpaddq	xmm10,xmm5,XMMWORD[32+rbp]
+	xor	rdi,r10
+	add	r8,r12
+	ror	r13,14
+	and	r15,rdi
+	xor	r14,r9
+	add	r8,r13
+	xor	r15,r10
+	ror	r14,28
+	add	rax,r8
+	add	r8,r15
+	mov	r13,rax
+	add	r14,r8
+	vmovdqa	XMMWORD[80+rsp],xmm10
+	vpalignr	xmm8,xmm7,xmm6,8
+	ror	r13,23
+	mov	r8,r14
+	vpalignr	xmm11,xmm3,xmm2,8
+	mov	r12,rbx
+	ror	r14,5
+DB	143,72,120,195,200,56
+	xor	r13,rax
+	xor	r12,rcx
+	vpsrlq	xmm8,xmm8,7
+	ror	r13,4
+	xor	r14,r8
+	vpaddq	xmm6,xmm6,xmm11
+	and	r12,rax
+	xor	r13,rax
+	add	rdx,QWORD[96+rsp]
+	mov	r15,r8
+DB	143,72,120,195,209,7
+	xor	r12,rcx
+	ror	r14,6
+	vpxor	xmm8,xmm8,xmm9
+	xor	r15,r9
+	add	rdx,r12
+	ror	r13,14
+	and	rdi,r15
+DB	143,104,120,195,221,3
+	xor	r14,r8
+	add	rdx,r13
+	vpxor	xmm8,xmm8,xmm10
+	xor	rdi,r9
+	ror	r14,28
+	vpsrlq	xmm10,xmm5,6
+	add	r11,rdx
+	add	rdx,rdi
+	vpaddq	xmm6,xmm6,xmm8
+	mov	r13,r11
+	add	r14,rdx
+DB	143,72,120,195,203,42
+	ror	r13,23
+	mov	rdx,r14
+	vpxor	xmm11,xmm11,xmm10
+	mov	r12,rax
+	ror	r14,5
+	xor	r13,r11
+	xor	r12,rbx
+	vpxor	xmm11,xmm11,xmm9
+	ror	r13,4
+	xor	r14,rdx
+	and	r12,r11
+	xor	r13,r11
+	vpaddq	xmm6,xmm6,xmm11
+	add	rcx,QWORD[104+rsp]
+	mov	rdi,rdx
+	xor	r12,rbx
+	ror	r14,6
+	vpaddq	xmm10,xmm6,XMMWORD[64+rbp]
+	xor	rdi,r8
+	add	rcx,r12
+	ror	r13,14
+	and	r15,rdi
+	xor	r14,rdx
+	add	rcx,r13
+	xor	r15,r8
+	ror	r14,28
+	add	r10,rcx
+	add	rcx,r15
+	mov	r13,r10
+	add	r14,rcx
+	vmovdqa	XMMWORD[96+rsp],xmm10
+	vpalignr	xmm8,xmm0,xmm7,8
+	ror	r13,23
+	mov	rcx,r14
+	vpalignr	xmm11,xmm4,xmm3,8
+	mov	r12,r11
+	ror	r14,5
+DB	143,72,120,195,200,56
+	xor	r13,r10
+	xor	r12,rax
+	vpsrlq	xmm8,xmm8,7
+	ror	r13,4
+	xor	r14,rcx
+	vpaddq	xmm7,xmm7,xmm11
+	and	r12,r10
+	xor	r13,r10
+	add	rbx,QWORD[112+rsp]
+	mov	r15,rcx
+DB	143,72,120,195,209,7
+	xor	r12,rax
+	ror	r14,6
+	vpxor	xmm8,xmm8,xmm9
+	xor	r15,rdx
+	add	rbx,r12
+	ror	r13,14
+	and	rdi,r15
+DB	143,104,120,195,222,3
+	xor	r14,rcx
+	add	rbx,r13
+	vpxor	xmm8,xmm8,xmm10
+	xor	rdi,rdx
+	ror	r14,28
+	vpsrlq	xmm10,xmm6,6
+	add	r9,rbx
+	add	rbx,rdi
+	vpaddq	xmm7,xmm7,xmm8
+	mov	r13,r9
+	add	r14,rbx
+DB	143,72,120,195,203,42
+	ror	r13,23
+	mov	rbx,r14
+	vpxor	xmm11,xmm11,xmm10
+	mov	r12,r10
+	ror	r14,5
+	xor	r13,r9
+	xor	r12,r11
+	vpxor	xmm11,xmm11,xmm9
+	ror	r13,4
+	xor	r14,rbx
+	and	r12,r9
+	xor	r13,r9
+	vpaddq	xmm7,xmm7,xmm11
+	add	rax,QWORD[120+rsp]
+	mov	rdi,rbx
+	xor	r12,r11
+	ror	r14,6
+	vpaddq	xmm10,xmm7,XMMWORD[96+rbp]
+	xor	rdi,rcx
+	add	rax,r12
+	ror	r13,14
+	and	r15,rdi
+	xor	r14,rbx
+	add	rax,r13
+	xor	r15,rcx
+	ror	r14,28
+	add	r8,rax
+	add	rax,r15
+	mov	r13,r8
+	add	r14,rax
+	vmovdqa	XMMWORD[112+rsp],xmm10
+	cmp	BYTE[135+rbp],0
+	jne	NEAR $L$xop_00_47
+	ror	r13,23
+	mov	rax,r14
+	mov	r12,r9
+	ror	r14,5
+	xor	r13,r8
+	xor	r12,r10
+	ror	r13,4
+	xor	r14,rax
+	and	r12,r8
+	xor	r13,r8
+	add	r11,QWORD[rsp]
+	mov	r15,rax
+	xor	r12,r10
+	ror	r14,6
+	xor	r15,rbx
+	add	r11,r12
+	ror	r13,14
+	and	rdi,r15
+	xor	r14,rax
+	add	r11,r13
+	xor	rdi,rbx
+	ror	r14,28
+	add	rdx,r11
+	add	r11,rdi
+	mov	r13,rdx
+	add	r14,r11
+	ror	r13,23
+	mov	r11,r14
+	mov	r12,r8
+	ror	r14,5
+	xor	r13,rdx
+	xor	r12,r9
+	ror	r13,4
+	xor	r14,r11
+	and	r12,rdx
+	xor	r13,rdx
+	add	r10,QWORD[8+rsp]
+	mov	rdi,r11
+	xor	r12,r9
+	ror	r14,6
+	xor	rdi,rax
+	add	r10,r12
+	ror	r13,14
+	and	r15,rdi
+	xor	r14,r11
+	add	r10,r13
+	xor	r15,rax
+	ror	r14,28
+	add	rcx,r10
+	add	r10,r15
+	mov	r13,rcx
+	add	r14,r10
+	ror	r13,23
+	mov	r10,r14
+	mov	r12,rdx
+	ror	r14,5
+	xor	r13,rcx
+	xor	r12,r8
+	ror	r13,4
+	xor	r14,r10
+	and	r12,rcx
+	xor	r13,rcx
+	add	r9,QWORD[16+rsp]
+	mov	r15,r10
+	xor	r12,r8
+	ror	r14,6
+	xor	r15,r11
+	add	r9,r12
+	ror	r13,14
+	and	rdi,r15
+	xor	r14,r10
+	add	r9,r13
+	xor	rdi,r11
+	ror	r14,28
+	add	rbx,r9
+	add	r9,rdi
+	mov	r13,rbx
+	add	r14,r9
+	ror	r13,23
+	mov	r9,r14
+	mov	r12,rcx
+	ror	r14,5
+	xor	r13,rbx
+	xor	r12,rdx
+	ror	r13,4
+	xor	r14,r9
+	and	r12,rbx
+	xor	r13,rbx
+	add	r8,QWORD[24+rsp]
+	mov	rdi,r9
+	xor	r12,rdx
+	ror	r14,6
+	xor	rdi,r10
+	add	r8,r12
+	ror	r13,14
+	and	r15,rdi
+	xor	r14,r9
+	add	r8,r13
+	xor	r15,r10
+	ror	r14,28
+	add	rax,r8
+	add	r8,r15
+	mov	r13,rax
+	add	r14,r8
+	ror	r13,23
+	mov	r8,r14
+	mov	r12,rbx
+	ror	r14,5
+	xor	r13,rax
+	xor	r12,rcx
+	ror	r13,4
+	xor	r14,r8
+	and	r12,rax
+	xor	r13,rax
+	add	rdx,QWORD[32+rsp]
+	mov	r15,r8
+	xor	r12,rcx
+	ror	r14,6
+	xor	r15,r9
+	add	rdx,r12
+	ror	r13,14
+	and	rdi,r15
+	xor	r14,r8
+	add	rdx,r13
+	xor	rdi,r9
+	ror	r14,28
+	add	r11,rdx
+	add	rdx,rdi
+	mov	r13,r11
+	add	r14,rdx
+	ror	r13,23
+	mov	rdx,r14
+	mov	r12,rax
+	ror	r14,5
+	xor	r13,r11
+	xor	r12,rbx
+	ror	r13,4
+	xor	r14,rdx
+	and	r12,r11
+	xor	r13,r11
+	add	rcx,QWORD[40+rsp]
+	mov	rdi,rdx
+	xor	r12,rbx
+	ror	r14,6
+	xor	rdi,r8
+	add	rcx,r12
+	ror	r13,14
+	and	r15,rdi
+	xor	r14,rdx
+	add	rcx,r13
+	xor	r15,r8
+	ror	r14,28
+	add	r10,rcx
+	add	rcx,r15
+	mov	r13,r10
+	add	r14,rcx
+	ror	r13,23
+	mov	rcx,r14
+	mov	r12,r11
+	ror	r14,5
+	xor	r13,r10
+	xor	r12,rax
+	ror	r13,4
+	xor	r14,rcx
+	and	r12,r10
+	xor	r13,r10
+	add	rbx,QWORD[48+rsp]
+	mov	r15,rcx
+	xor	r12,rax
+	ror	r14,6
+	xor	r15,rdx
+	add	rbx,r12
+	ror	r13,14
+	and	rdi,r15
+	xor	r14,rcx
+	add	rbx,r13
+	xor	rdi,rdx
+	ror	r14,28
+	add	r9,rbx
+	add	rbx,rdi
+	mov	r13,r9
+	add	r14,rbx
+	ror	r13,23
+	mov	rbx,r14
+	mov	r12,r10
+	ror	r14,5
+	xor	r13,r9
+	xor	r12,r11
+	ror	r13,4
+	xor	r14,rbx
+	and	r12,r9
+	xor	r13,r9
+	add	rax,QWORD[56+rsp]
+	mov	rdi,rbx
+	xor	r12,r11
+	ror	r14,6
+	xor	rdi,rcx
+	add	rax,r12
+	ror	r13,14
+	and	r15,rdi
+	xor	r14,rbx
+	add	rax,r13
+	xor	r15,rcx
+	ror	r14,28
+	add	r8,rax
+	add	rax,r15
+	mov	r13,r8
+	add	r14,rax
+	ror	r13,23
+	mov	rax,r14
+	mov	r12,r9
+	ror	r14,5
+	xor	r13,r8
+	xor	r12,r10
+	ror	r13,4
+	xor	r14,rax
+	and	r12,r8
+	xor	r13,r8
+	add	r11,QWORD[64+rsp]
+	mov	r15,rax
+	xor	r12,r10
+	ror	r14,6
+	xor	r15,rbx
+	add	r11,r12
+	ror	r13,14
+	and	rdi,r15
+	xor	r14,rax
+	add	r11,r13
+	xor	rdi,rbx
+	ror	r14,28
+	add	rdx,r11
+	add	r11,rdi
+	mov	r13,rdx
+	add	r14,r11
+	ror	r13,23
+	mov	r11,r14
+	mov	r12,r8
+	ror	r14,5
+	xor	r13,rdx
+	xor	r12,r9
+	ror	r13,4
+	xor	r14,r11
+	and	r12,rdx
+	xor	r13,rdx
+	add	r10,QWORD[72+rsp]
+	mov	rdi,r11
+	xor	r12,r9
+	ror	r14,6
+	xor	rdi,rax
+	add	r10,r12
+	ror	r13,14
+	and	r15,rdi
+	xor	r14,r11
+	add	r10,r13
+	xor	r15,rax
+	ror	r14,28
+	add	rcx,r10
+	add	r10,r15
+	mov	r13,rcx
+	add	r14,r10
+	ror	r13,23
+	mov	r10,r14
+	mov	r12,rdx
+	ror	r14,5
+	xor	r13,rcx
+	xor	r12,r8
+	ror	r13,4
+	xor	r14,r10
+	and	r12,rcx
+	xor	r13,rcx
+	add	r9,QWORD[80+rsp]
+	mov	r15,r10
+	xor	r12,r8
+	ror	r14,6
+	xor	r15,r11
+	add	r9,r12
+	ror	r13,14
+	and	rdi,r15
+	xor	r14,r10
+	add	r9,r13
+	xor	rdi,r11
+	ror	r14,28
+	add	rbx,r9
+	add	r9,rdi
+	mov	r13,rbx
+	add	r14,r9
+	ror	r13,23
+	mov	r9,r14
+	mov	r12,rcx
+	ror	r14,5
+	xor	r13,rbx
+	xor	r12,rdx
+	ror	r13,4
+	xor	r14,r9
+	and	r12,rbx
+	xor	r13,rbx
+	add	r8,QWORD[88+rsp]
+	mov	rdi,r9
+	xor	r12,rdx
+	ror	r14,6
+	xor	rdi,r10
+	add	r8,r12
+	ror	r13,14
+	and	r15,rdi
+	xor	r14,r9
+	add	r8,r13
+	xor	r15,r10
+	ror	r14,28
+	add	rax,r8
+	add	r8,r15
+	mov	r13,rax
+	add	r14,r8
+	ror	r13,23
+	mov	r8,r14
+	mov	r12,rbx
+	ror	r14,5
+	xor	r13,rax
+	xor	r12,rcx
+	ror	r13,4
+	xor	r14,r8
+	and	r12,rax
+	xor	r13,rax
+	add	rdx,QWORD[96+rsp]
+	mov	r15,r8
+	xor	r12,rcx
+	ror	r14,6
+	xor	r15,r9
+	add	rdx,r12
+	ror	r13,14
+	and	rdi,r15
+	xor	r14,r8
+	add	rdx,r13
+	xor	rdi,r9
+	ror	r14,28
+	add	r11,rdx
+	add	rdx,rdi
+	mov	r13,r11
+	add	r14,rdx
+	ror	r13,23
+	mov	rdx,r14
+	mov	r12,rax
+	ror	r14,5
+	xor	r13,r11
+	xor	r12,rbx
+	ror	r13,4
+	xor	r14,rdx
+	and	r12,r11
+	xor	r13,r11
+	add	rcx,QWORD[104+rsp]
+	mov	rdi,rdx
+	xor	r12,rbx
+	ror	r14,6
+	xor	rdi,r8
+	add	rcx,r12
+	ror	r13,14
+	and	r15,rdi
+	xor	r14,rdx
+	add	rcx,r13
+	xor	r15,r8
+	ror	r14,28
+	add	r10,rcx
+	add	rcx,r15
+	mov	r13,r10
+	add	r14,rcx
+	ror	r13,23
+	mov	rcx,r14
+	mov	r12,r11
+	ror	r14,5
+	xor	r13,r10
+	xor	r12,rax
+	ror	r13,4
+	xor	r14,rcx
+	and	r12,r10
+	xor	r13,r10
+	add	rbx,QWORD[112+rsp]
+	mov	r15,rcx
+	xor	r12,rax
+	ror	r14,6
+	xor	r15,rdx
+	add	rbx,r12
+	ror	r13,14
+	and	rdi,r15
+	xor	r14,rcx
+	add	rbx,r13
+	xor	rdi,rdx
+	ror	r14,28
+	add	r9,rbx
+	add	rbx,rdi
+	mov	r13,r9
+	add	r14,rbx
+	ror	r13,23
+	mov	rbx,r14
+	mov	r12,r10
+	ror	r14,5
+	xor	r13,r9
+	xor	r12,r11
+	ror	r13,4
+	xor	r14,rbx
+	and	r12,r9
+	xor	r13,r9
+	add	rax,QWORD[120+rsp]
+	mov	rdi,rbx
+	xor	r12,r11
+	ror	r14,6
+	xor	rdi,rcx
+	add	rax,r12
+	ror	r13,14
+	and	r15,rdi
+	xor	r14,rbx
+	add	rax,r13
+	xor	r15,rcx
+	ror	r14,28
+	add	r8,rax
+	add	rax,r15
+	mov	r13,r8
+	add	r14,rax
+	mov	rdi,QWORD[((128+0))+rsp]
+	mov	rax,r14
+
+	add	rax,QWORD[rdi]
+	lea	rsi,[128+rsi]
+	add	rbx,QWORD[8+rdi]
+	add	rcx,QWORD[16+rdi]
+	add	rdx,QWORD[24+rdi]
+	add	r8,QWORD[32+rdi]
+	add	r9,QWORD[40+rdi]
+	add	r10,QWORD[48+rdi]
+	add	r11,QWORD[56+rdi]
+
+	cmp	rsi,QWORD[((128+16))+rsp]
+
+	mov	QWORD[rdi],rax
+	mov	QWORD[8+rdi],rbx
+	mov	QWORD[16+rdi],rcx
+	mov	QWORD[24+rdi],rdx
+	mov	QWORD[32+rdi],r8
+	mov	QWORD[40+rdi],r9
+	mov	QWORD[48+rdi],r10
+	mov	QWORD[56+rdi],r11
+	jb	NEAR $L$loop_xop
+
+	mov	rsi,QWORD[((128+24))+rsp]
+	vzeroupper
+	movaps	xmm6,XMMWORD[((128+32))+rsp]
+	movaps	xmm7,XMMWORD[((128+48))+rsp]
+	movaps	xmm8,XMMWORD[((128+64))+rsp]
+	movaps	xmm9,XMMWORD[((128+80))+rsp]
+	movaps	xmm10,XMMWORD[((128+96))+rsp]
+	movaps	xmm11,XMMWORD[((128+112))+rsp]
+	mov	r15,QWORD[rsi]
+	mov	r14,QWORD[8+rsi]
+	mov	r13,QWORD[16+rsi]
+	mov	r12,QWORD[24+rsi]
+	mov	rbp,QWORD[32+rsi]
+	mov	rbx,QWORD[40+rsi]
+	lea	rsp,[48+rsi]
+$L$epilogue_xop:
+	mov	rdi,QWORD[8+rsp]	;WIN64 epilogue
+	mov	rsi,QWORD[16+rsp]
+	DB	0F3h,0C3h		;repret
+$L$SEH_end_sha512_block_data_order_xop:
+
+ALIGN	64
+sha512_block_data_order_avx:
+	mov	QWORD[8+rsp],rdi	;WIN64 prologue
+	mov	QWORD[16+rsp],rsi
+	mov	rax,rsp
+$L$SEH_begin_sha512_block_data_order_avx:
+	mov	rdi,rcx
+	mov	rsi,rdx
+	mov	rdx,r8
+
+
+$L$avx_shortcut:
+	push	rbx
+	push	rbp
+	push	r12
+	push	r13
+	push	r14
+	push	r15
+	mov	r11,rsp
+	shl	rdx,4
+	sub	rsp,256
+	lea	rdx,[rdx*8+rsi]
+	and	rsp,-64
+	mov	QWORD[((128+0))+rsp],rdi
+	mov	QWORD[((128+8))+rsp],rsi
+	mov	QWORD[((128+16))+rsp],rdx
+	mov	QWORD[((128+24))+rsp],r11
+	movaps	XMMWORD[(128+32)+rsp],xmm6
+	movaps	XMMWORD[(128+48)+rsp],xmm7
+	movaps	XMMWORD[(128+64)+rsp],xmm8
+	movaps	XMMWORD[(128+80)+rsp],xmm9
+	movaps	XMMWORD[(128+96)+rsp],xmm10
+	movaps	XMMWORD[(128+112)+rsp],xmm11
+$L$prologue_avx:
+
+	vzeroupper
+	mov	rax,QWORD[rdi]
+	mov	rbx,QWORD[8+rdi]
+	mov	rcx,QWORD[16+rdi]
+	mov	rdx,QWORD[24+rdi]
+	mov	r8,QWORD[32+rdi]
+	mov	r9,QWORD[40+rdi]
+	mov	r10,QWORD[48+rdi]
+	mov	r11,QWORD[56+rdi]
+	jmp	NEAR $L$loop_avx
+ALIGN	16
+$L$loop_avx:
+	vmovdqa	xmm11,XMMWORD[((K512+1280))]
+	vmovdqu	xmm0,XMMWORD[rsi]
+	lea	rbp,[((K512+128))]
+	vmovdqu	xmm1,XMMWORD[16+rsi]
+	vmovdqu	xmm2,XMMWORD[32+rsi]
+	vpshufb	xmm0,xmm0,xmm11
+	vmovdqu	xmm3,XMMWORD[48+rsi]
+	vpshufb	xmm1,xmm1,xmm11
+	vmovdqu	xmm4,XMMWORD[64+rsi]
+	vpshufb	xmm2,xmm2,xmm11
+	vmovdqu	xmm5,XMMWORD[80+rsi]
+	vpshufb	xmm3,xmm3,xmm11
+	vmovdqu	xmm6,XMMWORD[96+rsi]
+	vpshufb	xmm4,xmm4,xmm11
+	vmovdqu	xmm7,XMMWORD[112+rsi]
+	vpshufb	xmm5,xmm5,xmm11
+	vpaddq	xmm8,xmm0,XMMWORD[((-128))+rbp]
+	vpshufb	xmm6,xmm6,xmm11
+	vpaddq	xmm9,xmm1,XMMWORD[((-96))+rbp]
+	vpshufb	xmm7,xmm7,xmm11
+	vpaddq	xmm10,xmm2,XMMWORD[((-64))+rbp]
+	vpaddq	xmm11,xmm3,XMMWORD[((-32))+rbp]
+	vmovdqa	XMMWORD[rsp],xmm8
+	vpaddq	xmm8,xmm4,XMMWORD[rbp]
+	vmovdqa	XMMWORD[16+rsp],xmm9
+	vpaddq	xmm9,xmm5,XMMWORD[32+rbp]
+	vmovdqa	XMMWORD[32+rsp],xmm10
+	vpaddq	xmm10,xmm6,XMMWORD[64+rbp]
+	vmovdqa	XMMWORD[48+rsp],xmm11
+	vpaddq	xmm11,xmm7,XMMWORD[96+rbp]
+	vmovdqa	XMMWORD[64+rsp],xmm8
+	mov	r14,rax
+	vmovdqa	XMMWORD[80+rsp],xmm9
+	mov	rdi,rbx
+	vmovdqa	XMMWORD[96+rsp],xmm10
+	xor	rdi,rcx
+	vmovdqa	XMMWORD[112+rsp],xmm11
+	mov	r13,r8
+	jmp	NEAR $L$avx_00_47
+
+ALIGN	16
+$L$avx_00_47:
+	add	rbp,256
+	vpalignr	xmm8,xmm1,xmm0,8
+	shrd	r13,r13,23
+	mov	rax,r14
+	vpalignr	xmm11,xmm5,xmm4,8
+	mov	r12,r9
+	shrd	r14,r14,5
+	vpsrlq	xmm10,xmm8,1
+	xor	r13,r8
+	xor	r12,r10
+	vpaddq	xmm0,xmm0,xmm11
+	shrd	r13,r13,4
+	xor	r14,rax
+	vpsrlq	xmm11,xmm8,7
+	and	r12,r8
+	xor	r13,r8
+	vpsllq	xmm9,xmm8,56
+	add	r11,QWORD[rsp]
+	mov	r15,rax
+	vpxor	xmm8,xmm11,xmm10
+	xor	r12,r10
+	shrd	r14,r14,6
+	vpsrlq	xmm10,xmm10,7
+	xor	r15,rbx
+	add	r11,r12
+	vpxor	xmm8,xmm8,xmm9
+	shrd	r13,r13,14
+	and	rdi,r15
+	vpsllq	xmm9,xmm9,7
+	xor	r14,rax
+	add	r11,r13
+	vpxor	xmm8,xmm8,xmm10
+	xor	rdi,rbx
+	shrd	r14,r14,28
+	vpsrlq	xmm11,xmm7,6
+	add	rdx,r11
+	add	r11,rdi
+	vpxor	xmm8,xmm8,xmm9
+	mov	r13,rdx
+	add	r14,r11
+	vpsllq	xmm10,xmm7,3
+	shrd	r13,r13,23
+	mov	r11,r14
+	vpaddq	xmm0,xmm0,xmm8
+	mov	r12,r8
+	shrd	r14,r14,5
+	vpsrlq	xmm9,xmm7,19
+	xor	r13,rdx
+	xor	r12,r9
+	vpxor	xmm11,xmm11,xmm10
+	shrd	r13,r13,4
+	xor	r14,r11
+	vpsllq	xmm10,xmm10,42
+	and	r12,rdx
+	xor	r13,rdx
+	vpxor	xmm11,xmm11,xmm9
+	add	r10,QWORD[8+rsp]
+	mov	rdi,r11
+	vpsrlq	xmm9,xmm9,42
+	xor	r12,r9
+	shrd	r14,r14,6
+	vpxor	xmm11,xmm11,xmm10
+	xor	rdi,rax
+	add	r10,r12
+	vpxor	xmm11,xmm11,xmm9
+	shrd	r13,r13,14
+	and	r15,rdi
+	vpaddq	xmm0,xmm0,xmm11
+	xor	r14,r11
+	add	r10,r13
+	vpaddq	xmm10,xmm0,XMMWORD[((-128))+rbp]
+	xor	r15,rax
+	shrd	r14,r14,28
+	add	rcx,r10
+	add	r10,r15
+	mov	r13,rcx
+	add	r14,r10
+	vmovdqa	XMMWORD[rsp],xmm10
+	vpalignr	xmm8,xmm2,xmm1,8
+	shrd	r13,r13,23
+	mov	r10,r14
+	vpalignr	xmm11,xmm6,xmm5,8
+	mov	r12,rdx
+	shrd	r14,r14,5
+	vpsrlq	xmm10,xmm8,1
+	xor	r13,rcx
+	xor	r12,r8
+	vpaddq	xmm1,xmm1,xmm11
+	shrd	r13,r13,4
+	xor	r14,r10
+	vpsrlq	xmm11,xmm8,7
+	and	r12,rcx
+	xor	r13,rcx
+	vpsllq	xmm9,xmm8,56
+	add	r9,QWORD[16+rsp]
+	mov	r15,r10
+	vpxor	xmm8,xmm11,xmm10
+	xor	r12,r8
+	shrd	r14,r14,6
+	vpsrlq	xmm10,xmm10,7
+	xor	r15,r11
+	add	r9,r12
+	vpxor	xmm8,xmm8,xmm9
+	shrd	r13,r13,14
+	and	rdi,r15
+	vpsllq	xmm9,xmm9,7
+	xor	r14,r10
+	add	r9,r13
+	vpxor	xmm8,xmm8,xmm10
+	xor	rdi,r11
+	shrd	r14,r14,28
+	vpsrlq	xmm11,xmm0,6
+	add	rbx,r9
+	add	r9,rdi
+	vpxor	xmm8,xmm8,xmm9
+	mov	r13,rbx
+	add	r14,r9
+	vpsllq	xmm10,xmm0,3
+	shrd	r13,r13,23
+	mov	r9,r14
+	vpaddq	xmm1,xmm1,xmm8
+	mov	r12,rcx
+	shrd	r14,r14,5
+	vpsrlq	xmm9,xmm0,19
+	xor	r13,rbx
+	xor	r12,rdx
+	vpxor	xmm11,xmm11,xmm10
+	shrd	r13,r13,4
+	xor	r14,r9
+	vpsllq	xmm10,xmm10,42
+	and	r12,rbx
+	xor	r13,rbx
+	vpxor	xmm11,xmm11,xmm9
+	add	r8,QWORD[24+rsp]
+	mov	rdi,r9
+	vpsrlq	xmm9,xmm9,42
+	xor	r12,rdx
+	shrd	r14,r14,6
+	vpxor	xmm11,xmm11,xmm10
+	xor	rdi,r10
+	add	r8,r12
+	vpxor	xmm11,xmm11,xmm9
+	shrd	r13,r13,14
+	and	r15,rdi
+	vpaddq	xmm1,xmm1,xmm11
+	xor	r14,r9
+	add	r8,r13
+	vpaddq	xmm10,xmm1,XMMWORD[((-96))+rbp]
+	xor	r15,r10
+	shrd	r14,r14,28
+	add	rax,r8
+	add	r8,r15
+	mov	r13,rax
+	add	r14,r8
+	vmovdqa	XMMWORD[16+rsp],xmm10
+	vpalignr	xmm8,xmm3,xmm2,8
+	shrd	r13,r13,23
+	mov	r8,r14
+	vpalignr	xmm11,xmm7,xmm6,8
+	mov	r12,rbx
+	shrd	r14,r14,5
+	vpsrlq	xmm10,xmm8,1
+	xor	r13,rax
+	xor	r12,rcx
+	vpaddq	xmm2,xmm2,xmm11
+	shrd	r13,r13,4
+	xor	r14,r8
+	vpsrlq	xmm11,xmm8,7
+	and	r12,rax
+	xor	r13,rax
+	vpsllq	xmm9,xmm8,56
+	add	rdx,QWORD[32+rsp]
+	mov	r15,r8
+	vpxor	xmm8,xmm11,xmm10
+	xor	r12,rcx
+	shrd	r14,r14,6
+	vpsrlq	xmm10,xmm10,7
+	xor	r15,r9
+	add	rdx,r12
+	vpxor	xmm8,xmm8,xmm9
+	shrd	r13,r13,14
+	and	rdi,r15
+	vpsllq	xmm9,xmm9,7
+	xor	r14,r8
+	add	rdx,r13
+	vpxor	xmm8,xmm8,xmm10
+	xor	rdi,r9
+	shrd	r14,r14,28
+	vpsrlq	xmm11,xmm1,6
+	add	r11,rdx
+	add	rdx,rdi
+	vpxor	xmm8,xmm8,xmm9
+	mov	r13,r11
+	add	r14,rdx
+	vpsllq	xmm10,xmm1,3
+	shrd	r13,r13,23
+	mov	rdx,r14
+	vpaddq	xmm2,xmm2,xmm8
+	mov	r12,rax
+	shrd	r14,r14,5
+	vpsrlq	xmm9,xmm1,19
+	xor	r13,r11
+	xor	r12,rbx
+	vpxor	xmm11,xmm11,xmm10
+	shrd	r13,r13,4
+	xor	r14,rdx
+	vpsllq	xmm10,xmm10,42
+	and	r12,r11
+	xor	r13,r11
+	vpxor	xmm11,xmm11,xmm9
+	add	rcx,QWORD[40+rsp]
+	mov	rdi,rdx
+	vpsrlq	xmm9,xmm9,42
+	xor	r12,rbx
+	shrd	r14,r14,6
+	vpxor	xmm11,xmm11,xmm10
+	xor	rdi,r8
+	add	rcx,r12
+	vpxor	xmm11,xmm11,xmm9
+	shrd	r13,r13,14
+	and	r15,rdi
+	vpaddq	xmm2,xmm2,xmm11
+	xor	r14,rdx
+	add	rcx,r13
+	vpaddq	xmm10,xmm2,XMMWORD[((-64))+rbp]
+	xor	r15,r8
+	shrd	r14,r14,28
+	add	r10,rcx
+	add	rcx,r15
+	mov	r13,r10
+	add	r14,rcx
+	vmovdqa	XMMWORD[32+rsp],xmm10
+	vpalignr	xmm8,xmm4,xmm3,8
+	shrd	r13,r13,23
+	mov	rcx,r14
+	vpalignr	xmm11,xmm0,xmm7,8
+	mov	r12,r11
+	shrd	r14,r14,5
+	vpsrlq	xmm10,xmm8,1
+	xor	r13,r10
+	xor	r12,rax
+	vpaddq	xmm3,xmm3,xmm11
+	shrd	r13,r13,4
+	xor	r14,rcx
+	vpsrlq	xmm11,xmm8,7
+	and	r12,r10
+	xor	r13,r10
+	vpsllq	xmm9,xmm8,56
+	add	rbx,QWORD[48+rsp]
+	mov	r15,rcx
+	vpxor	xmm8,xmm11,xmm10
+	xor	r12,rax
+	shrd	r14,r14,6
+	vpsrlq	xmm10,xmm10,7
+	xor	r15,rdx
+	add	rbx,r12
+	vpxor	xmm8,xmm8,xmm9
+	shrd	r13,r13,14
+	and	rdi,r15
+	vpsllq	xmm9,xmm9,7
+	xor	r14,rcx
+	add	rbx,r13
+	vpxor	xmm8,xmm8,xmm10
+	xor	rdi,rdx
+	shrd	r14,r14,28
+	vpsrlq	xmm11,xmm2,6
+	add	r9,rbx
+	add	rbx,rdi
+	vpxor	xmm8,xmm8,xmm9
+	mov	r13,r9
+	add	r14,rbx
+	vpsllq	xmm10,xmm2,3
+	shrd	r13,r13,23
+	mov	rbx,r14
+	vpaddq	xmm3,xmm3,xmm8
+	mov	r12,r10
+	shrd	r14,r14,5
+	vpsrlq	xmm9,xmm2,19
+	xor	r13,r9
+	xor	r12,r11
+	vpxor	xmm11,xmm11,xmm10
+	shrd	r13,r13,4
+	xor	r14,rbx
+	vpsllq	xmm10,xmm10,42
+	and	r12,r9
+	xor	r13,r9
+	vpxor	xmm11,xmm11,xmm9
+	add	rax,QWORD[56+rsp]
+	mov	rdi,rbx
+	vpsrlq	xmm9,xmm9,42
+	xor	r12,r11
+	shrd	r14,r14,6
+	vpxor	xmm11,xmm11,xmm10
+	xor	rdi,rcx
+	add	rax,r12
+	vpxor	xmm11,xmm11,xmm9
+	shrd	r13,r13,14
+	and	r15,rdi
+	vpaddq	xmm3,xmm3,xmm11
+	xor	r14,rbx
+	add	rax,r13
+	vpaddq	xmm10,xmm3,XMMWORD[((-32))+rbp]
+	xor	r15,rcx
+	shrd	r14,r14,28
+	add	r8,rax
+	add	rax,r15
+	mov	r13,r8
+	add	r14,rax
+	vmovdqa	XMMWORD[48+rsp],xmm10
+	vpalignr	xmm8,xmm5,xmm4,8
+	shrd	r13,r13,23
+	mov	rax,r14
+	vpalignr	xmm11,xmm1,xmm0,8
+	mov	r12,r9
+	shrd	r14,r14,5
+	vpsrlq	xmm10,xmm8,1
+	xor	r13,r8
+	xor	r12,r10
+	vpaddq	xmm4,xmm4,xmm11
+	shrd	r13,r13,4
+	xor	r14,rax
+	vpsrlq	xmm11,xmm8,7
+	and	r12,r8
+	xor	r13,r8
+	vpsllq	xmm9,xmm8,56
+	add	r11,QWORD[64+rsp]
+	mov	r15,rax
+	vpxor	xmm8,xmm11,xmm10
+	xor	r12,r10
+	shrd	r14,r14,6
+	vpsrlq	xmm10,xmm10,7
+	xor	r15,rbx
+	add	r11,r12
+	vpxor	xmm8,xmm8,xmm9
+	shrd	r13,r13,14
+	and	rdi,r15
+	vpsllq	xmm9,xmm9,7
+	xor	r14,rax
+	add	r11,r13
+	vpxor	xmm8,xmm8,xmm10
+	xor	rdi,rbx
+	shrd	r14,r14,28
+	vpsrlq	xmm11,xmm3,6
+	add	rdx,r11
+	add	r11,rdi
+	vpxor	xmm8,xmm8,xmm9
+	mov	r13,rdx
+	add	r14,r11
+	vpsllq	xmm10,xmm3,3
+	shrd	r13,r13,23
+	mov	r11,r14
+	vpaddq	xmm4,xmm4,xmm8
+	mov	r12,r8
+	shrd	r14,r14,5
+	vpsrlq	xmm9,xmm3,19
+	xor	r13,rdx
+	xor	r12,r9
+	vpxor	xmm11,xmm11,xmm10
+	shrd	r13,r13,4
+	xor	r14,r11
+	vpsllq	xmm10,xmm10,42
+	and	r12,rdx
+	xor	r13,rdx
+	vpxor	xmm11,xmm11,xmm9
+	add	r10,QWORD[72+rsp]
+	mov	rdi,r11
+	vpsrlq	xmm9,xmm9,42
+	xor	r12,r9
+	shrd	r14,r14,6
+	vpxor	xmm11,xmm11,xmm10
+	xor	rdi,rax
+	add	r10,r12
+	vpxor	xmm11,xmm11,xmm9
+	shrd	r13,r13,14
+	and	r15,rdi
+	vpaddq	xmm4,xmm4,xmm11
+	xor	r14,r11
+	add	r10,r13
+	vpaddq	xmm10,xmm4,XMMWORD[rbp]
+	xor	r15,rax
+	shrd	r14,r14,28
+	add	rcx,r10
+	add	r10,r15
+	mov	r13,rcx
+	add	r14,r10
+	vmovdqa	XMMWORD[64+rsp],xmm10
+	vpalignr	xmm8,xmm6,xmm5,8
+	shrd	r13,r13,23
+	mov	r10,r14
+	vpalignr	xmm11,xmm2,xmm1,8
+	mov	r12,rdx
+	shrd	r14,r14,5
+	vpsrlq	xmm10,xmm8,1
+	xor	r13,rcx
+	xor	r12,r8
+	vpaddq	xmm5,xmm5,xmm11
+	shrd	r13,r13,4
+	xor	r14,r10
+	vpsrlq	xmm11,xmm8,7
+	and	r12,rcx
+	xor	r13,rcx
+	vpsllq	xmm9,xmm8,56
+	add	r9,QWORD[80+rsp]
+	mov	r15,r10
+	vpxor	xmm8,xmm11,xmm10
+	xor	r12,r8
+	shrd	r14,r14,6
+	vpsrlq	xmm10,xmm10,7
+	xor	r15,r11
+	add	r9,r12
+	vpxor	xmm8,xmm8,xmm9
+	shrd	r13,r13,14
+	and	rdi,r15
+	vpsllq	xmm9,xmm9,7
+	xor	r14,r10
+	add	r9,r13
+	vpxor	xmm8,xmm8,xmm10
+	xor	rdi,r11
+	shrd	r14,r14,28
+	vpsrlq	xmm11,xmm4,6
+	add	rbx,r9
+	add	r9,rdi
+	vpxor	xmm8,xmm8,xmm9
+	mov	r13,rbx
+	add	r14,r9
+	vpsllq	xmm10,xmm4,3
+	shrd	r13,r13,23
+	mov	r9,r14
+	vpaddq	xmm5,xmm5,xmm8
+	mov	r12,rcx
+	shrd	r14,r14,5
+	vpsrlq	xmm9,xmm4,19
+	xor	r13,rbx
+	xor	r12,rdx
+	vpxor	xmm11,xmm11,xmm10
+	shrd	r13,r13,4
+	xor	r14,r9
+	vpsllq	xmm10,xmm10,42
+	and	r12,rbx
+	xor	r13,rbx
+	vpxor	xmm11,xmm11,xmm9
+	add	r8,QWORD[88+rsp]
+	mov	rdi,r9
+	vpsrlq	xmm9,xmm9,42
+	xor	r12,rdx
+	shrd	r14,r14,6
+	vpxor	xmm11,xmm11,xmm10
+	xor	rdi,r10
+	add	r8,r12
+	vpxor	xmm11,xmm11,xmm9
+	shrd	r13,r13,14
+	and	r15,rdi
+	vpaddq	xmm5,xmm5,xmm11
+	xor	r14,r9
+	add	r8,r13
+	vpaddq	xmm10,xmm5,XMMWORD[32+rbp]
+	xor	r15,r10
+	shrd	r14,r14,28
+	add	rax,r8
+	add	r8,r15
+	mov	r13,rax
+	add	r14,r8
+	vmovdqa	XMMWORD[80+rsp],xmm10
+	vpalignr	xmm8,xmm7,xmm6,8
+	shrd	r13,r13,23
+	mov	r8,r14
+	vpalignr	xmm11,xmm3,xmm2,8
+	mov	r12,rbx
+	shrd	r14,r14,5
+	vpsrlq	xmm10,xmm8,1
+	xor	r13,rax
+	xor	r12,rcx
+	vpaddq	xmm6,xmm6,xmm11
+	shrd	r13,r13,4
+	xor	r14,r8
+	vpsrlq	xmm11,xmm8,7
+	and	r12,rax
+	xor	r13,rax
+	vpsllq	xmm9,xmm8,56
+	add	rdx,QWORD[96+rsp]
+	mov	r15,r8
+	vpxor	xmm8,xmm11,xmm10
+	xor	r12,rcx
+	shrd	r14,r14,6
+	vpsrlq	xmm10,xmm10,7
+	xor	r15,r9
+	add	rdx,r12
+	vpxor	xmm8,xmm8,xmm9
+	shrd	r13,r13,14
+	and	rdi,r15
+	vpsllq	xmm9,xmm9,7
+	xor	r14,r8
+	add	rdx,r13
+	vpxor	xmm8,xmm8,xmm10
+	xor	rdi,r9
+	shrd	r14,r14,28
+	vpsrlq	xmm11,xmm5,6
+	add	r11,rdx
+	add	rdx,rdi
+	vpxor	xmm8,xmm8,xmm9
+	mov	r13,r11
+	add	r14,rdx
+	vpsllq	xmm10,xmm5,3
+	shrd	r13,r13,23
+	mov	rdx,r14
+	vpaddq	xmm6,xmm6,xmm8
+	mov	r12,rax
+	shrd	r14,r14,5
+	vpsrlq	xmm9,xmm5,19
+	xor	r13,r11
+	xor	r12,rbx
+	vpxor	xmm11,xmm11,xmm10
+	shrd	r13,r13,4
+	xor	r14,rdx
+	vpsllq	xmm10,xmm10,42
+	and	r12,r11
+	xor	r13,r11
+	vpxor	xmm11,xmm11,xmm9
+	add	rcx,QWORD[104+rsp]
+	mov	rdi,rdx
+	vpsrlq	xmm9,xmm9,42
+	xor	r12,rbx
+	shrd	r14,r14,6
+	vpxor	xmm11,xmm11,xmm10
+	xor	rdi,r8
+	add	rcx,r12
+	vpxor	xmm11,xmm11,xmm9
+	shrd	r13,r13,14
+	and	r15,rdi
+	vpaddq	xmm6,xmm6,xmm11
+	xor	r14,rdx
+	add	rcx,r13
+	vpaddq	xmm10,xmm6,XMMWORD[64+rbp]
+	xor	r15,r8
+	shrd	r14,r14,28
+	add	r10,rcx
+	add	rcx,r15
+	mov	r13,r10
+	add	r14,rcx
+	vmovdqa	XMMWORD[96+rsp],xmm10
+	vpalignr	xmm8,xmm0,xmm7,8
+	shrd	r13,r13,23
+	mov	rcx,r14
+	vpalignr	xmm11,xmm4,xmm3,8
+	mov	r12,r11
+	shrd	r14,r14,5
+	vpsrlq	xmm10,xmm8,1
+	xor	r13,r10
+	xor	r12,rax
+	vpaddq	xmm7,xmm7,xmm11
+	shrd	r13,r13,4
+	xor	r14,rcx
+	vpsrlq	xmm11,xmm8,7
+	and	r12,r10
+	xor	r13,r10
+	vpsllq	xmm9,xmm8,56
+	add	rbx,QWORD[112+rsp]
+	mov	r15,rcx
+	vpxor	xmm8,xmm11,xmm10
+	xor	r12,rax
+	shrd	r14,r14,6
+	vpsrlq	xmm10,xmm10,7
+	xor	r15,rdx
+	add	rbx,r12
+	vpxor	xmm8,xmm8,xmm9
+	shrd	r13,r13,14
+	and	rdi,r15
+	vpsllq	xmm9,xmm9,7
+	xor	r14,rcx
+	add	rbx,r13
+	vpxor	xmm8,xmm8,xmm10
+	xor	rdi,rdx
+	shrd	r14,r14,28
+	vpsrlq	xmm11,xmm6,6
+	add	r9,rbx
+	add	rbx,rdi
+	vpxor	xmm8,xmm8,xmm9
+	mov	r13,r9
+	add	r14,rbx
+	vpsllq	xmm10,xmm6,3
+	shrd	r13,r13,23
+	mov	rbx,r14
+	vpaddq	xmm7,xmm7,xmm8
+	mov	r12,r10
+	shrd	r14,r14,5
+	vpsrlq	xmm9,xmm6,19
+	xor	r13,r9
+	xor	r12,r11
+	vpxor	xmm11,xmm11,xmm10
+	shrd	r13,r13,4
+	xor	r14,rbx
+	vpsllq	xmm10,xmm10,42
+	and	r12,r9
+	xor	r13,r9
+	vpxor	xmm11,xmm11,xmm9
+	add	rax,QWORD[120+rsp]
+	mov	rdi,rbx
+	vpsrlq	xmm9,xmm9,42
+	xor	r12,r11
+	shrd	r14,r14,6
+	vpxor	xmm11,xmm11,xmm10
+	xor	rdi,rcx
+	add	rax,r12
+	vpxor	xmm11,xmm11,xmm9
+	shrd	r13,r13,14
+	and	r15,rdi
+	vpaddq	xmm7,xmm7,xmm11
+	xor	r14,rbx
+	add	rax,r13
+	vpaddq	xmm10,xmm7,XMMWORD[96+rbp]
+	xor	r15,rcx
+	shrd	r14,r14,28
+	add	r8,rax
+	add	rax,r15
+	mov	r13,r8
+	add	r14,rax
+	vmovdqa	XMMWORD[112+rsp],xmm10
+	cmp	BYTE[135+rbp],0
+	jne	NEAR $L$avx_00_47
+	shrd	r13,r13,23
+	mov	rax,r14
+	mov	r12,r9
+	shrd	r14,r14,5
+	xor	r13,r8
+	xor	r12,r10
+	shrd	r13,r13,4
+	xor	r14,rax
+	and	r12,r8
+	xor	r13,r8
+	add	r11,QWORD[rsp]
+	mov	r15,rax
+	xor	r12,r10
+	shrd	r14,r14,6
+	xor	r15,rbx
+	add	r11,r12
+	shrd	r13,r13,14
+	and	rdi,r15
+	xor	r14,rax
+	add	r11,r13
+	xor	rdi,rbx
+	shrd	r14,r14,28
+	add	rdx,r11
+	add	r11,rdi
+	mov	r13,rdx
+	add	r14,r11
+	shrd	r13,r13,23
+	mov	r11,r14
+	mov	r12,r8
+	shrd	r14,r14,5
+	xor	r13,rdx
+	xor	r12,r9
+	shrd	r13,r13,4
+	xor	r14,r11
+	and	r12,rdx
+	xor	r13,rdx
+	add	r10,QWORD[8+rsp]
+	mov	rdi,r11
+	xor	r12,r9
+	shrd	r14,r14,6
+	xor	rdi,rax
+	add	r10,r12
+	shrd	r13,r13,14
+	and	r15,rdi
+	xor	r14,r11
+	add	r10,r13
+	xor	r15,rax
+	shrd	r14,r14,28
+	add	rcx,r10
+	add	r10,r15
+	mov	r13,rcx
+	add	r14,r10
+	shrd	r13,r13,23
+	mov	r10,r14
+	mov	r12,rdx
+	shrd	r14,r14,5
+	xor	r13,rcx
+	xor	r12,r8
+	shrd	r13,r13,4
+	xor	r14,r10
+	and	r12,rcx
+	xor	r13,rcx
+	add	r9,QWORD[16+rsp]
+	mov	r15,r10
+	xor	r12,r8
+	shrd	r14,r14,6
+	xor	r15,r11
+	add	r9,r12
+	shrd	r13,r13,14
+	and	rdi,r15
+	xor	r14,r10
+	add	r9,r13
+	xor	rdi,r11
+	shrd	r14,r14,28
+	add	rbx,r9
+	add	r9,rdi
+	mov	r13,rbx
+	add	r14,r9
+	shrd	r13,r13,23
+	mov	r9,r14
+	mov	r12,rcx
+	shrd	r14,r14,5
+	xor	r13,rbx
+	xor	r12,rdx
+	shrd	r13,r13,4
+	xor	r14,r9
+	and	r12,rbx
+	xor	r13,rbx
+	add	r8,QWORD[24+rsp]
+	mov	rdi,r9
+	xor	r12,rdx
+	shrd	r14,r14,6
+	xor	rdi,r10
+	add	r8,r12
+	shrd	r13,r13,14
+	and	r15,rdi
+	xor	r14,r9
+	add	r8,r13
+	xor	r15,r10
+	shrd	r14,r14,28
+	add	rax,r8
+	add	r8,r15
+	mov	r13,rax
+	add	r14,r8
+	shrd	r13,r13,23
+	mov	r8,r14
+	mov	r12,rbx
+	shrd	r14,r14,5
+	xor	r13,rax
+	xor	r12,rcx
+	shrd	r13,r13,4
+	xor	r14,r8
+	and	r12,rax
+	xor	r13,rax
+	add	rdx,QWORD[32+rsp]
+	mov	r15,r8
+	xor	r12,rcx
+	shrd	r14,r14,6
+	xor	r15,r9
+	add	rdx,r12
+	shrd	r13,r13,14
+	and	rdi,r15
+	xor	r14,r8
+	add	rdx,r13
+	xor	rdi,r9
+	shrd	r14,r14,28
+	add	r11,rdx
+	add	rdx,rdi
+	mov	r13,r11
+	add	r14,rdx
+	shrd	r13,r13,23
+	mov	rdx,r14
+	mov	r12,rax
+	shrd	r14,r14,5
+	xor	r13,r11
+	xor	r12,rbx
+	shrd	r13,r13,4
+	xor	r14,rdx
+	and	r12,r11
+	xor	r13,r11
+	add	rcx,QWORD[40+rsp]
+	mov	rdi,rdx
+	xor	r12,rbx
+	shrd	r14,r14,6
+	xor	rdi,r8
+	add	rcx,r12
+	shrd	r13,r13,14
+	and	r15,rdi
+	xor	r14,rdx
+	add	rcx,r13
+	xor	r15,r8
+	shrd	r14,r14,28
+	add	r10,rcx
+	add	rcx,r15
+	mov	r13,r10
+	add	r14,rcx
+	shrd	r13,r13,23
+	mov	rcx,r14
+	mov	r12,r11
+	shrd	r14,r14,5
+	xor	r13,r10
+	xor	r12,rax
+	shrd	r13,r13,4
+	xor	r14,rcx
+	and	r12,r10
+	xor	r13,r10
+	add	rbx,QWORD[48+rsp]
+	mov	r15,rcx
+	xor	r12,rax
+	shrd	r14,r14,6
+	xor	r15,rdx
+	add	rbx,r12
+	shrd	r13,r13,14
+	and	rdi,r15
+	xor	r14,rcx
+	add	rbx,r13
+	xor	rdi,rdx
+	shrd	r14,r14,28
+	add	r9,rbx
+	add	rbx,rdi
+	mov	r13,r9
+	add	r14,rbx
+	shrd	r13,r13,23
+	mov	rbx,r14
+	mov	r12,r10
+	shrd	r14,r14,5
+	xor	r13,r9
+	xor	r12,r11
+	shrd	r13,r13,4
+	xor	r14,rbx
+	and	r12,r9
+	xor	r13,r9
+	add	rax,QWORD[56+rsp]
+	mov	rdi,rbx
+	xor	r12,r11
+	shrd	r14,r14,6
+	xor	rdi,rcx
+	add	rax,r12
+	shrd	r13,r13,14
+	and	r15,rdi
+	xor	r14,rbx
+	add	rax,r13
+	xor	r15,rcx
+	shrd	r14,r14,28
+	add	r8,rax
+	add	rax,r15
+	mov	r13,r8
+	add	r14,rax
+	shrd	r13,r13,23
+	mov	rax,r14
+	mov	r12,r9
+	shrd	r14,r14,5
+	xor	r13,r8
+	xor	r12,r10
+	shrd	r13,r13,4
+	xor	r14,rax
+	and	r12,r8
+	xor	r13,r8
+	add	r11,QWORD[64+rsp]
+	mov	r15,rax
+	xor	r12,r10
+	shrd	r14,r14,6
+	xor	r15,rbx
+	add	r11,r12
+	shrd	r13,r13,14
+	and	rdi,r15
+	xor	r14,rax
+	add	r11,r13
+	xor	rdi,rbx
+	shrd	r14,r14,28
+	add	rdx,r11
+	add	r11,rdi
+	mov	r13,rdx
+	add	r14,r11
+	shrd	r13,r13,23
+	mov	r11,r14
+	mov	r12,r8
+	shrd	r14,r14,5
+	xor	r13,rdx
+	xor	r12,r9
+	shrd	r13,r13,4
+	xor	r14,r11
+	and	r12,rdx
+	xor	r13,rdx
+	add	r10,QWORD[72+rsp]
+	mov	rdi,r11
+	xor	r12,r9
+	shrd	r14,r14,6
+	xor	rdi,rax
+	add	r10,r12
+	shrd	r13,r13,14
+	and	r15,rdi
+	xor	r14,r11
+	add	r10,r13
+	xor	r15,rax
+	shrd	r14,r14,28
+	add	rcx,r10
+	add	r10,r15
+	mov	r13,rcx
+	add	r14,r10
+	shrd	r13,r13,23
+	mov	r10,r14
+	mov	r12,rdx
+	shrd	r14,r14,5
+	xor	r13,rcx
+	xor	r12,r8
+	shrd	r13,r13,4
+	xor	r14,r10
+	and	r12,rcx
+	xor	r13,rcx
+	add	r9,QWORD[80+rsp]
+	mov	r15,r10
+	xor	r12,r8
+	shrd	r14,r14,6
+	xor	r15,r11
+	add	r9,r12
+	shrd	r13,r13,14
+	and	rdi,r15
+	xor	r14,r10
+	add	r9,r13
+	xor	rdi,r11
+	shrd	r14,r14,28
+	add	rbx,r9
+	add	r9,rdi
+	mov	r13,rbx
+	add	r14,r9
+	shrd	r13,r13,23
+	mov	r9,r14
+	mov	r12,rcx
+	shrd	r14,r14,5
+	xor	r13,rbx
+	xor	r12,rdx
+	shrd	r13,r13,4
+	xor	r14,r9
+	and	r12,rbx
+	xor	r13,rbx
+	add	r8,QWORD[88+rsp]
+	mov	rdi,r9
+	xor	r12,rdx
+	shrd	r14,r14,6
+	xor	rdi,r10
+	add	r8,r12
+	shrd	r13,r13,14
+	and	r15,rdi
+	xor	r14,r9
+	add	r8,r13
+	xor	r15,r10
+	shrd	r14,r14,28
+	add	rax,r8
+	add	r8,r15
+	mov	r13,rax
+	add	r14,r8
+	shrd	r13,r13,23
+	mov	r8,r14
+	mov	r12,rbx
+	shrd	r14,r14,5
+	xor	r13,rax
+	xor	r12,rcx
+	shrd	r13,r13,4
+	xor	r14,r8
+	and	r12,rax
+	xor	r13,rax
+	add	rdx,QWORD[96+rsp]
+	mov	r15,r8
+	xor	r12,rcx
+	shrd	r14,r14,6
+	xor	r15,r9
+	add	rdx,r12
+	shrd	r13,r13,14
+	and	rdi,r15
+	xor	r14,r8
+	add	rdx,r13
+	xor	rdi,r9
+	shrd	r14,r14,28
+	add	r11,rdx
+	add	rdx,rdi
+	mov	r13,r11
+	add	r14,rdx
+	shrd	r13,r13,23
+	mov	rdx,r14
+	mov	r12,rax
+	shrd	r14,r14,5
+	xor	r13,r11
+	xor	r12,rbx
+	shrd	r13,r13,4
+	xor	r14,rdx
+	and	r12,r11
+	xor	r13,r11
+	add	rcx,QWORD[104+rsp]
+	mov	rdi,rdx
+	xor	r12,rbx
+	shrd	r14,r14,6
+	xor	rdi,r8
+	add	rcx,r12
+	shrd	r13,r13,14
+	and	r15,rdi
+	xor	r14,rdx
+	add	rcx,r13
+	xor	r15,r8
+	shrd	r14,r14,28
+	add	r10,rcx
+	add	rcx,r15
+	mov	r13,r10
+	add	r14,rcx
+	shrd	r13,r13,23
+	mov	rcx,r14
+	mov	r12,r11
+	shrd	r14,r14,5
+	xor	r13,r10
+	xor	r12,rax
+	shrd	r13,r13,4
+	xor	r14,rcx
+	and	r12,r10
+	xor	r13,r10
+	add	rbx,QWORD[112+rsp]
+	mov	r15,rcx
+	xor	r12,rax
+	shrd	r14,r14,6
+	xor	r15,rdx
+	add	rbx,r12
+	shrd	r13,r13,14
+	and	rdi,r15
+	xor	r14,rcx
+	add	rbx,r13
+	xor	rdi,rdx
+	shrd	r14,r14,28
+	add	r9,rbx
+	add	rbx,rdi
+	mov	r13,r9
+	add	r14,rbx
+	shrd	r13,r13,23
+	mov	rbx,r14
+	mov	r12,r10
+	shrd	r14,r14,5
+	xor	r13,r9
+	xor	r12,r11
+	shrd	r13,r13,4
+	xor	r14,rbx
+	and	r12,r9
+	xor	r13,r9
+	add	rax,QWORD[120+rsp]
+	mov	rdi,rbx
+	xor	r12,r11
+	shrd	r14,r14,6
+	xor	rdi,rcx
+	add	rax,r12
+	shrd	r13,r13,14
+	and	r15,rdi
+	xor	r14,rbx
+	add	rax,r13
+	xor	r15,rcx
+	shrd	r14,r14,28
+	add	r8,rax
+	add	rax,r15
+	mov	r13,r8
+	add	r14,rax
+	mov	rdi,QWORD[((128+0))+rsp]
+	mov	rax,r14
+
+	add	rax,QWORD[rdi]
+	lea	rsi,[128+rsi]
+	add	rbx,QWORD[8+rdi]
+	add	rcx,QWORD[16+rdi]
+	add	rdx,QWORD[24+rdi]
+	add	r8,QWORD[32+rdi]
+	add	r9,QWORD[40+rdi]
+	add	r10,QWORD[48+rdi]
+	add	r11,QWORD[56+rdi]
+
+	cmp	rsi,QWORD[((128+16))+rsp]
+
+	mov	QWORD[rdi],rax
+	mov	QWORD[8+rdi],rbx
+	mov	QWORD[16+rdi],rcx
+	mov	QWORD[24+rdi],rdx
+	mov	QWORD[32+rdi],r8
+	mov	QWORD[40+rdi],r9
+	mov	QWORD[48+rdi],r10
+	mov	QWORD[56+rdi],r11
+	jb	NEAR $L$loop_avx
+
+	mov	rsi,QWORD[((128+24))+rsp]
+	vzeroupper
+	movaps	xmm6,XMMWORD[((128+32))+rsp]
+	movaps	xmm7,XMMWORD[((128+48))+rsp]
+	movaps	xmm8,XMMWORD[((128+64))+rsp]
+	movaps	xmm9,XMMWORD[((128+80))+rsp]
+	movaps	xmm10,XMMWORD[((128+96))+rsp]
+	movaps	xmm11,XMMWORD[((128+112))+rsp]
+	mov	r15,QWORD[rsi]
+	mov	r14,QWORD[8+rsi]
+	mov	r13,QWORD[16+rsi]
+	mov	r12,QWORD[24+rsi]
+	mov	rbp,QWORD[32+rsi]
+	mov	rbx,QWORD[40+rsi]
+	lea	rsp,[48+rsi]
+$L$epilogue_avx:
+	mov	rdi,QWORD[8+rsp]	;WIN64 epilogue
+	mov	rsi,QWORD[16+rsp]
+	DB	0F3h,0C3h		;repret
+$L$SEH_end_sha512_block_data_order_avx:
 EXTERN	__imp_RtlVirtualUnwind
 
 ALIGN	16
@@ -1903,9 +4190,23 @@
 	DD	$L$SEH_begin_sha512_block_data_order wrt ..imagebase
 	DD	$L$SEH_end_sha512_block_data_order wrt ..imagebase
 	DD	$L$SEH_info_sha512_block_data_order wrt ..imagebase
+	DD	$L$SEH_begin_sha512_block_data_order_xop wrt ..imagebase
+	DD	$L$SEH_end_sha512_block_data_order_xop wrt ..imagebase
+	DD	$L$SEH_info_sha512_block_data_order_xop wrt ..imagebase
+	DD	$L$SEH_begin_sha512_block_data_order_avx wrt ..imagebase
+	DD	$L$SEH_end_sha512_block_data_order_avx wrt ..imagebase
+	DD	$L$SEH_info_sha512_block_data_order_avx wrt ..imagebase
 section	.xdata rdata align=8
 ALIGN	8
 $L$SEH_info_sha512_block_data_order:
 DB	9,0,0,0
 	DD	se_handler wrt ..imagebase
 	DD	$L$prologue wrt ..imagebase,$L$epilogue wrt ..imagebase
+$L$SEH_info_sha512_block_data_order_xop:
+DB	9,0,0,0
+	DD	se_handler wrt ..imagebase
+	DD	$L$prologue_xop wrt ..imagebase,$L$epilogue_xop wrt ..imagebase
+$L$SEH_info_sha512_block_data_order_avx:
+DB	9,0,0,0
+	DD	se_handler wrt ..imagebase
+	DD	$L$prologue_avx wrt ..imagebase,$L$epilogue_avx wrt ..imagebase

tools/VERSION

diff --git a/tools/VERSION b/tools/VERSION
index 3eb18d8..8bec38a 100644
--- a/tools/VERSION
+++ b/tools/VERSION
@@ -27,5 +27,5 @@
 MAJOR 1
 MINOR 19
 PATCH 0
-PRERELEASE 3
+PRERELEASE 4
 PRERELEASE_PATCH 0

tools/bots/dart_sdk.py

diff --git a/tools/bots/dart_sdk.py b/tools/bots/dart_sdk.py
index 00e7d35..92f89f8 100644
--- a/tools/bots/dart_sdk.py
+++ b/tools/bots/dart_sdk.py
@@ -81,8 +81,9 @@
 
 def DartArchiveUploadSDKs(system, sdk32_zip, sdk64_zip):
   namer = bot_utils.GCSNamer(CHANNEL, bot_utils.ReleaseType.RAW)
-  revision = utils.GetArchiveVersion()
-  for revision in [revision, 'latest']:
+  git_number = utils.GetArchiveVersion()
+  git_hash = 'hash/%s' % utils.GetGitRevision()
+  for revision in [git_number, git_hash, 'latest']:
     path32 = namer.sdk_zipfilepath(revision, system, 'ia32', 'release')
     path64 = namer.sdk_zipfilepath(revision, system, 'x64', 'release')
     DartArchiveFile(sdk32_zip, path32, checksum_files=True)

tools/create_sdk.py

diff --git a/tools/create_sdk.py b/tools/create_sdk.py
index dad929b..85ba811 100755
--- a/tools/create_sdk.py
+++ b/tools/create_sdk.py
@@ -44,7 +44,6 @@
 # ......dart_client.platform
 # ......dart_server.platform
 # ......dart_shared.platform
-# ......dart2dart.platform
 # ......_internal/
 #.........spec.sum
 #.........strong.sum
@@ -267,8 +266,7 @@
   # Copy the platform descriptors.
   for file_name in ["dart_client.platform",
                     "dart_server.platform",
-                    "dart_shared.platform",
-                    "dart2dart.platform"]:
+                    "dart_shared.platform"]:
     copyfile(join(HOME, 'sdk', 'lib', file_name), join(LIB, file_name));
 
   # Copy libraries.dart to lib/_internal/libraries.dart for backwards

tools/deps/dartium.deps/DEPS

diff --git a/tools/deps/dartium.deps/DEPS b/tools/deps/dartium.deps/DEPS
index b73f612..9457e56 100644
--- a/tools/deps/dartium.deps/DEPS
+++ b/tools/deps/dartium.deps/DEPS
@@ -9,7 +9,7 @@
 
 vars.update({
   "dartium_chromium_commit": "67a7ba9669f7bb0300ef35085d4e6bb98b1966cc",
-  "dartium_webkit_commit": "bfb8649f69c7b02100aa6fd5a9c77d7858106d1b",
+  "dartium_webkit_commit": "006cf3ee53f37973d2db812a62fc8f5f83b52948",
   "chromium_base_revision": "338390",
 
   # We use mirrors of all github repos to guarantee reproducibility and

tools/utils.py

diff --git a/tools/utils.py b/tools/utils.py
index 0611bdd..51a3a03 100644
--- a/tools/utils.py
+++ b/tools/utils.py
@@ -103,7 +103,7 @@
 
 # Try to guess Visual Studio location when buiding on Windows.
 def GuessVisualStudioPath():
-  defaultPath = r"C:\Program Files (x86)\Microsoft Visual Studio 12.0\Common7" \
+  defaultPath = r"C:\Program Files (x86)\Microsoft Visual Studio 14.0\Common7" \
                 r"\IDE"
   defaultExecutable = "devenv.com"
 
@@ -148,8 +148,8 @@
               # Can't find value under the key - continue to the next key.
               continue
             isExpress = executable != 'devenv.com'
-            if not isExpress and subkeyName == '12.0':
-              # Stop search since if we found non-Express VS2013 version
+            if not isExpress and subkeyName == '14.0':
+              # Stop search since if we found non-Express VS2015 version
               # installed, which is preferred version.
               return installDir, executable
             else: