// Copyright (c) 2019, 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.

#include "vm/globals.h"

// For `AllocateObjectInstr::WillAllocateNewOrRemembered`
// For `GenericCheckBoundInstr::UseUnboxedRepresentation`
#include "vm/compiler/backend/il.h"

#define SHOULD_NOT_INCLUDE_RUNTIME

#include "vm/compiler/stub_code_compiler.h"

#if defined(TARGET_ARCH_ARM64)

#include "vm/class_id.h"
#include "vm/code_entry_kind.h"
#include "vm/compiler/api/type_check_mode.h"
#include "vm/compiler/assembler/assembler.h"
#include "vm/compiler/backend/locations.h"
#include "vm/constants.h"
#include "vm/ffi_callback_metadata.h"
#include "vm/instructions.h"
#include "vm/static_type_exactness_state.h"
#include "vm/tags.h"

#define __ assembler->

namespace dart {
namespace compiler {

// Ensures that [R0] is a new object, if not it will be added to the remembered
// set via a leaf runtime call.
//
// WARNING: This might clobber all registers except for [R0], [THR] and [FP].
// The caller should simply call LeaveStubFrame() and return.
void StubCodeCompiler::EnsureIsNewOrRemembered() {
  // If the object is not in an active TLAB, we call a leaf-runtime to add it to
  // the remembered set and/or deferred marking worklist. This test assumes a
  // Page's TLAB use is always ascending.
  Label done;
  __ AndImmediate(TMP, R0, target::kPageMask);
  __ LoadFromOffset(TMP, TMP, target::Page::original_top_offset());
  __ CompareRegisters(R0, TMP);
  __ BranchIf(UNSIGNED_GREATER_EQUAL, &done);

  {
    LeafRuntimeScope rt(assembler, /*frame_size=*/0,
                        /*preserve_registers=*/false);
    // R0 already loaded.
    __ mov(R1, THR);
    rt.Call(kEnsureRememberedAndMarkingDeferredRuntimeEntry,
            /*argument_count=*/2);
  }

  __ Bind(&done);
}

// In TSAN mode the runtime will throw an exception using an intermediary
// longjmp() call to unwind the C frames in a way that TSAN can understand.
//
// This wrapper will setup a [jmp_buf] on the stack and initialize it to be a
// target for a possible longjmp(). In the exceptional case we'll forward
// control of execution to the usual JumpToFrame stub.
//
// In non-TSAN mode this will do nothing and the runtime will call the
// JumpToFrame stub directly.
//
// The callback [fun] may be invoked with a modified [RSP] due to allocating
// a [jmp_buf] allocating structure on the stack (as well as the saved old
// [Thread::tsan_utils_->setjmp_buffer_]).
static void WithExceptionCatchingTrampoline(Assembler* assembler,
                                            std::function<void()> fun) {
#if !defined(USING_SIMULATOR)
  const Register kTsanUtilsReg = R3;

  // Reserve space for arguments and align frame before entering C++ world.
  const intptr_t kJumpBufferSize = sizeof(jmp_buf);
  // Save & Restore the volatile CPU registers across the setjmp() call.
  const RegisterSet volatile_registers(
      kAbiVolatileCpuRegs & ~(1 << R0) & ~(1 << SP),
      /*fpu_registers=*/0);

  const Register kSavedRspReg = R20;
  COMPILE_ASSERT(IsCalleeSavedRegister(kSavedRspReg));
  // We rely on THR being preserved across the setjmp() call.
  COMPILE_ASSERT(IsCalleeSavedRegister(THR));

  if (FLAG_target_thread_sanitizer) {
    Label do_native_call;

    // Save old jmp_buf.
    __ ldr(kTsanUtilsReg, Address(THR, target::Thread::tsan_utils_offset()));
    __ ldr(TMP,
           Address(kTsanUtilsReg, target::TsanUtils::setjmp_buffer_offset()));
    __ Push(TMP);

    // Allocate jmp_buf struct on stack & remember pointer to it on the
    // [Thread::tsan_utils_->setjmp_buffer] (which exceptions.cc will longjmp()
    // to)
    __ AddImmediate(SP, -kJumpBufferSize);
    __ str(SP,
           Address(kTsanUtilsReg, target::TsanUtils::setjmp_buffer_offset()));

    // Call setjmp() with a pointer to the allocated jmp_buf struct.
    __ MoveRegister(R0, SP);
    __ PushRegisters(volatile_registers);
    __ EnterCFrame(0);
    __ mov(R25, CSP);
    __ mov(CSP, SP);
    __ ldr(kTsanUtilsReg, Address(THR, target::Thread::tsan_utils_offset()));
    __ CallCFunction(
        Address(kTsanUtilsReg, target::TsanUtils::setjmp_function_offset()));
    __ mov(SP, CSP);
    __ mov(CSP, R25);
    __ LeaveCFrame();
    __ PopRegisters(volatile_registers);

    // We are the target of a longjmp() iff setjmp() returns non-0.
    __ cbz(&do_native_call, R0);

    // We are the target of a longjmp: Cleanup the stack and tail-call the
    // JumpToFrame stub which will take care of unwinding the stack and hand
    // execution to the catch entry.
    __ AddImmediate(SP, kJumpBufferSize);
    __ ldr(kTsanUtilsReg, Address(THR, target::Thread::tsan_utils_offset()));
    __ Pop(TMP);
    __ str(TMP,
           Address(kTsanUtilsReg, target::TsanUtils::setjmp_buffer_offset()));

    __ ldr(R0,
           Address(kTsanUtilsReg, target::TsanUtils::exception_pc_offset()));
    __ ldr(R1,
           Address(kTsanUtilsReg, target::TsanUtils::exception_sp_offset()));
    __ ldr(R2,
           Address(kTsanUtilsReg, target::TsanUtils::exception_fp_offset()));
    __ MoveRegister(R3, THR);
    __ Jump(Address(THR, target::Thread::jump_to_frame_entry_point_offset()));

    // We leave the created [jump_buf] structure on the stack as well as the
    // pushed old [Thread::tsan_utils_->setjmp_buffer_].
    __ Bind(&do_native_call);
    __ MoveRegister(kSavedRspReg, SP);
  }
#endif  // !defined(USING_SIMULATOR)

  fun();

#if !defined(USING_SIMULATOR)
  if (FLAG_target_thread_sanitizer) {
    __ MoveRegister(SP, kSavedRspReg);
    __ AddImmediate(SP, kJumpBufferSize);
    const Register kTsanUtilsReg2 = kSavedRspReg;
    __ ldr(kTsanUtilsReg2, Address(THR, target::Thread::tsan_utils_offset()));
    __ Pop(TMP);
    __ str(TMP,
           Address(kTsanUtilsReg2, target::TsanUtils::setjmp_buffer_offset()));
  }
#endif  // !defined(USING_SIMULATOR)
}

// Input parameters:
//   LR : return address.
//   SP : address of last argument in argument array.
//   SP + 8*R4 - 8 : address of first argument in argument array.
//   SP + 8*R4 : address of return value.
//   R5 : address of the runtime function to call.
//   R4 : number of arguments to the call.
void StubCodeCompiler::GenerateCallToRuntimeStub() {
  const intptr_t thread_offset = target::NativeArguments::thread_offset();
  const intptr_t argc_tag_offset = target::NativeArguments::argc_tag_offset();
  const intptr_t argv_offset = target::NativeArguments::argv_offset();
  const intptr_t retval_offset = target::NativeArguments::retval_offset();

  __ Comment("CallToRuntimeStub");
  __ ldr(CODE_REG, Address(THR, target::Thread::call_to_runtime_stub_offset()));
  __ SetPrologueOffset();
  __ EnterStubFrame();

  // Save exit frame information to enable stack walking as we are about
  // to transition to Dart VM C++ code.
  __ StoreToOffset(FP, THR, target::Thread::top_exit_frame_info_offset());

  // Mark that the thread exited generated code through a runtime call.
  __ LoadImmediate(R8, target::Thread::exit_through_runtime_call());
  __ StoreToOffset(R8, THR, target::Thread::exit_through_ffi_offset());

#if defined(DEBUG)
  {
    Label ok;
    // Check that we are always entering from Dart code.
    __ LoadFromOffset(R8, THR, target::Thread::vm_tag_offset());
    __ CompareImmediate(R8, VMTag::kDartTagId);
    __ b(&ok, EQ);
    __ Stop("Not coming from Dart code.");
    __ Bind(&ok);
  }
#endif

  // Mark that the thread is executing VM code.
  __ StoreToOffset(R5, THR, target::Thread::vm_tag_offset());

  WithExceptionCatchingTrampoline(assembler, [&]() {
    // Reserve space for arguments and align frame before entering C++ world.
    // target::NativeArguments are passed in registers.
    __ Comment("align stack");
    // Reserve space for arguments.
    ASSERT(target::NativeArguments::StructSize() == 4 * target::kWordSize);
    __ ReserveAlignedFrameSpace(target::NativeArguments::StructSize());

    // Pass target::NativeArguments structure by value and call runtime.
    // Registers R0, R1, R2, and R3 are used.

    ASSERT(thread_offset == 0 * target::kWordSize);
    // Set thread in NativeArgs.
    __ mov(R0, THR);

    ASSERT(argc_tag_offset == 1 * target::kWordSize);
    __ mov(R1, R4);  // Set argc in target::NativeArguments.

    ASSERT(argv_offset == 2 * target::kWordSize);
    __ add(R2, ZR, Operand(R4, LSL, 3));
    __ add(R2, FP, Operand(R2));  // Compute argv.
    // Set argv in target::NativeArguments.
    __ AddImmediate(R2,
                    target::frame_layout.param_end_from_fp * target::kWordSize);

    ASSERT(retval_offset == 3 * target::kWordSize);
    __ AddImmediate(R3, R2, target::kWordSize);

    __ StoreToOffset(R0, SP, thread_offset);
    __ StoreToOffset(R1, SP, argc_tag_offset);
    __ StoreToOffset(R2, SP, argv_offset);
    __ StoreToOffset(R3, SP, retval_offset);
    __ mov(R0, SP);  // Pass the pointer to the target::NativeArguments.

    // We are entering runtime code, so the C stack pointer must be restored
    // from the stack limit to the top of the stack. We cache the stack limit
    // address in a callee-saved register.
    __ mov(R25, CSP);
    __ mov(CSP, SP);

    __ blr(R5);
    __ Comment("CallToRuntimeStub return");

    // Restore SP and CSP.
    __ mov(SP, CSP);
    __ mov(CSP, R25);

    // Refresh pinned registers (write barrier mask, null, dispatch table, etc).
    __ RestorePinnedRegisters();

    // Retval is next to 1st argument.
    // Mark that the thread is executing Dart code.
    __ LoadImmediate(R2, VMTag::kDartTagId);
    __ StoreToOffset(R2, THR, target::Thread::vm_tag_offset());

    // Mark that the thread has not exited generated Dart code.
    __ StoreToOffset(ZR, THR, target::Thread::exit_through_ffi_offset());

    // Reset exit frame information in Isolate's mutator thread structure.
    __ StoreToOffset(ZR, THR, target::Thread::top_exit_frame_info_offset());

    // Restore the global object pool after returning from runtime (old space is
    // moving, so the GOP could have been relocated).
    if (FLAG_precompiled_mode) {
      __ SetupGlobalPoolAndDispatchTable();
    }
  });

  __ LeaveStubFrame();

  // The following return can jump to a lazy-deopt stub, which assumes R0
  // contains a return value and will save it in a GC-visible way.  We therefore
  // have to ensure R0 does not contain any garbage value left from the C
  // function we called (which has return type "void").
  // (See GenerateDeoptimizationSequence::saved_result_slot_from_fp.)
  __ LoadImmediate(R0, 0);
  __ ret();
}

void StubCodeCompiler::GenerateSharedStubGeneric(
    bool save_fpu_registers,
    intptr_t self_code_stub_offset_from_thread,
    bool allow_return,
    std::function<void()> perform_runtime_call) {
  // We want the saved registers to appear like part of the caller's frame, so
  // we push them before calling EnterStubFrame.
  RegisterSet all_registers;
  all_registers.AddAllNonReservedRegisters(save_fpu_registers);

  // To make the stack map calculation architecture independent we do the same
  // as on intel.
  READS_RETURN_ADDRESS_FROM_LR(__ Push(LR));
  __ PushRegisters(all_registers);
  __ ldr(CODE_REG, Address(THR, self_code_stub_offset_from_thread));
  __ EnterStubFrame();
  perform_runtime_call();
  if (!allow_return) {
    __ Breakpoint();
    return;
  }
  __ LeaveStubFrame();
  __ PopRegisters(all_registers);
  __ Drop(1);  // We use the LR restored via LeaveStubFrame.
  READS_RETURN_ADDRESS_FROM_LR(__ ret(LR));
}

void StubCodeCompiler::GenerateSharedStub(
    bool save_fpu_registers,
    const RuntimeEntry* target,
    intptr_t self_code_stub_offset_from_thread,
    bool allow_return,
    bool store_runtime_result_in_result_register) {
  ASSERT(!store_runtime_result_in_result_register || allow_return);
  auto perform_runtime_call = [&]() {
    if (store_runtime_result_in_result_register) {
      __ PushRegister(NULL_REG);
    }
    __ CallRuntime(*target, /*argument_count=*/0);
    if (store_runtime_result_in_result_register) {
      __ PopRegister(R0);
      __ str(R0,
             Address(FP, target::kWordSize *
                             StubCodeCompiler::WordOffsetFromFpToCpuRegister(
                                 SharedSlowPathStubABI::kResultReg)));
    }
  };
  GenerateSharedStubGeneric(save_fpu_registers,
                            self_code_stub_offset_from_thread, allow_return,
                            perform_runtime_call);
}

void StubCodeCompiler::GenerateEnterSafepointStub() {
  RegisterSet all_registers;
  all_registers.AddAllGeneralRegisters();

  __ EnterFrame(0);
  __ PushRegisters(all_registers);

  __ mov(CALLEE_SAVED_TEMP, CSP);
  __ mov(CALLEE_SAVED_TEMP2, SP);
  __ ReserveAlignedFrameSpace(0);
  __ mov(CSP, SP);

  __ ldr(R0, Address(THR, kEnterSafepointRuntimeEntry.OffsetFromThread()));
  __ blr(R0);

  __ mov(SP, CALLEE_SAVED_TEMP2);
  __ mov(CSP, CALLEE_SAVED_TEMP);

  __ PopRegisters(all_registers);
  __ LeaveFrame();

  __ Ret();
}

static void GenerateExitSafepointStubCommon(Assembler* assembler,
                                            uword runtime_entry_offset) {
  RegisterSet all_registers;
  all_registers.AddAllGeneralRegisters();

  __ EnterFrame(0);
  __ PushRegisters(all_registers);

  __ mov(CALLEE_SAVED_TEMP, CSP);
  __ mov(CALLEE_SAVED_TEMP2, SP);
  __ ReserveAlignedFrameSpace(0);
  __ mov(CSP, SP);

  // Set the execution state to VM while waiting for the safepoint to end.
  // This isn't strictly necessary but enables tests to check that we're not
  // in native code anymore. See tests/ffi/function_gc_test.dart for example.
  __ LoadImmediate(R0, target::Thread::vm_execution_state());
  __ str(R0, Address(THR, target::Thread::execution_state_offset()));

  __ ldr(R0, Address(THR, runtime_entry_offset));
  __ blr(R0);

  __ mov(SP, CALLEE_SAVED_TEMP2);
  __ mov(CSP, CALLEE_SAVED_TEMP);

  __ PopRegisters(all_registers);
  __ LeaveFrame();

  __ Ret();
}

void StubCodeCompiler::GenerateExitSafepointStub() {
  GenerateExitSafepointStubCommon(
      assembler, kExitSafepointRuntimeEntry.OffsetFromThread());
}

void StubCodeCompiler::GenerateExitSafepointIgnoreUnwindInProgressStub() {
  GenerateExitSafepointStubCommon(
      assembler,
      kExitSafepointIgnoreUnwindInProgressRuntimeEntry.OffsetFromThread());
}

// Calls native code within a safepoint.
//
// On entry:
//   R9: target to call
//   Stack: set up for native call (SP), aligned, CSP < SP
//
// On exit:
//   R19: clobbered, although normally callee-saved
//   Stack: preserved, CSP == SP
void StubCodeCompiler::GenerateCallNativeThroughSafepointStub() {
  COMPILE_ASSERT(IsAbiPreservedRegister(R19));

  SPILLS_RETURN_ADDRESS_FROM_LR_TO_REGISTER(__ mov(R19, LR));
  __ LoadImmediate(R10, target::Thread::exit_through_ffi());
  __ TransitionGeneratedToNative(R9, FPREG, R10 /*volatile*/,
                                 /*enter_safepoint=*/true);
  __ mov(R25, CSP);
  __ mov(CSP, SP);

#if defined(DEBUG)
  // Check CSP alignment.
  __ andi(R11 /*volatile*/, SP,
          Immediate(~(OS::ActivationFrameAlignment() - 1)));
  __ cmp(R11, Operand(SP));
  Label done;
  __ b(&done, EQ);
  __ Breakpoint();
  __ Bind(&done);
#endif

  __ blr(R9);

  __ mov(SP, CSP);
  __ mov(CSP, R25);

  __ TransitionNativeToGenerated(R10, /*leave_safepoint=*/true);
  __ ret(R19);
}

void StubCodeCompiler::GenerateLoadBSSEntry(BSS::Relocation relocation,
                                            Register dst,
                                            Register tmp) {
  compiler::Label skip_reloc;
  __ b(&skip_reloc);
  InsertBSSRelocation(relocation);
  __ Bind(&skip_reloc);

  __ adr(tmp, compiler::Immediate(-compiler::target::kWordSize));

  // tmp holds the address of the relocation.
  __ ldr(dst, compiler::Address(tmp));

  // dst holds the relocation itself: tmp - bss_start.
  // tmp = tmp + (bss_start - tmp) = bss_start
  __ add(tmp, tmp, compiler::Operand(dst));

  // tmp holds the start of the BSS section.
  // Load the "get-thread" routine: *bss_start.
  __ ldr(dst, compiler::Address(tmp));
}

void StubCodeCompiler::GenerateLoadFfiCallbackMetadataRuntimeFunction(
    uword function_index,
    Register dst) {
  // Keep in sync with FfiCallbackMetadata::EnsureFirstTrampolinePageLocked.
  // Note: If the stub was aligned, this could be a single PC relative load.

  // Load a pointer to the beginning of the stub into dst.
  const intptr_t code_size = __ CodeSize();
  __ adr(dst, Immediate(-code_size));

  // Round dst down to the page size.
  __ andi(dst, dst, Immediate(FfiCallbackMetadata::kPageMask));

  // Load the function from the function table.
  __ LoadFromOffset(dst, dst,
                    FfiCallbackMetadata::RuntimeFunctionOffset(function_index));
}

void StubCodeCompiler::GenerateFfiCallbackTrampolineStub() {
#if defined(USING_SIMULATOR) && !defined(DART_PRECOMPILER)
  // TODO(37299): FFI is not supported in SIMARM64.
  __ Breakpoint();
#else
  Label body;

  // R9 is volatile and not used for passing any arguments.
  COMPILE_ASSERT(!IsCalleeSavedRegister(R9) && !IsArgumentRegister(R9));
  for (intptr_t i = 0; i < FfiCallbackMetadata::NumCallbackTrampolinesPerPage();
       ++i) {
    // The FfiCallbackMetadata table is keyed by the trampoline entry point. So
    // look up the current PC, then jump to the shared section.
    __ adr(R9, Immediate(0));
    __ b(&body);
  }

  ASSERT_EQUAL(__ CodeSize(),
               FfiCallbackMetadata::kNativeCallbackTrampolineSize *
                   FfiCallbackMetadata::NumCallbackTrampolinesPerPage());

  __ Bind(&body);

  const intptr_t shared_stub_start = __ CodeSize();

  // Save THR (callee-saved) and LR on the real C stack (CSP). Keeps it
  // aligned.
  COMPILE_ASSERT(FfiCallbackMetadata::kNativeCallbackTrampolineStackDelta == 2);
  SPILLS_LR_TO_FRAME(__ stp(
      THR, LR, Address(CSP, -2 * target::kWordSize, Address::PairPreIndex)));

  COMPILE_ASSERT(!IsArgumentRegister(THR));

  RegisterSet all_registers;
  all_registers.AddAllArgumentRegisters();
  all_registers.Add(Location::RegisterLocation(
      CallingConventions::kPointerToReturnStructRegisterCall));

  // Load the thread, verify the callback ID and exit the safepoint.
  //
  // We exit the safepoint inside DLRT_GetFfiCallbackMetadata in order to save
  // code size on this shared stub.
  {
    __ SetupDartSP();

    __ EnterFrame(0);
    __ PushRegisters(all_registers);

    __ mov(R0, R9);

    // We also need to look up the entry point for the trampoline. This is
    // returned using a pointer passed to the second arg of the C function
    // below. We aim that pointer at a reserved stack slot.
    __ AddImmediate(SP, SP, -compiler::target::kWordSize);
    __ mov(R1, SP);

    // We also need to know if this is a sync or async callback. This is also
    // returned by pointer.
    __ AddImmediate(SP, SP, -compiler::target::kWordSize);
    __ mov(R2, SP);

    __ EnterFrame(0);
    __ ReserveAlignedFrameSpace(0);

#if defined(DART_TARGET_OS_FUCHSIA)
    // TODO(https://dartbug.com/52579): Remove.
    if (FLAG_precompiled_mode) {
      GenerateLoadBSSEntry(BSS::Relocation::DRT_GetFfiCallbackMetadata, R4, R9);
    } else {
      Label call;
      __ ldr(R4, compiler::Address::PC(2 * Instr::kInstrSize));
      __ b(&call);
      __ Emit64(reinterpret_cast<int64_t>(&DLRT_GetFfiCallbackMetadata));
      __ Bind(&call);
    }
#else
    GenerateLoadFfiCallbackMetadataRuntimeFunction(
        FfiCallbackMetadata::kGetFfiCallbackMetadata, R4);
#endif  // defined(DART_TARGET_OS_FUCHSIA)

    __ mov(CSP, SP);
    __ blr(R4);
    __ mov(SP, CSP);
    __ mov(THR, R0);

    __ LeaveFrame();

    // The trampoline type is at the top of the stack. Pop it into R9.
    __ Pop(R9);

    // Entry point is now at the top of the stack. Pop it into R10.
    COMPILE_ASSERT(!IsCalleeSavedRegister(R10) && !IsArgumentRegister(R10));
    __ Pop(R10);

    __ PopRegisters(all_registers);
    __ LeaveFrame();

    __ RestoreCSP();
  }

  Label async_callback;
  Label done;

  // If GetFfiCallbackMetadata returned a null thread, it means that the async
  // callback was invoked after it was deleted. In this case, do nothing.
  __ cmp(THR, Operand(0));
  __ b(&done, EQ);

  // Check the trampoline type to see how the callback should be invoked.
  __ cmp(
      R9,
      Operand(static_cast<uword>(FfiCallbackMetadata::TrampolineType::kAsync)));
  __ b(&async_callback, EQ);

  // Sync callback. The entry point contains the target function, so just call
  // it. DLRT_GetThreadForNativeCallbackTrampoline exited the safepoint, so
  // re-enter it afterwards.

  // Clobbers all volatile registers, including the callback ID in R9.
  // Resets CSP and SP, important for EnterSafepoint below.
  __ blr(R10);

  // Clobbers TMP, TMP2 and R9 -- all volatile and not holding return values.
  __ EnterFullSafepoint(/*scratch=*/R9);

  __ b(&done);
  __ Bind(&async_callback);

  // Async callback. The entrypoint marshals the arguments into a message and
  // sends it over the send port. DLRT_GetThreadForNativeCallbackTrampoline
  // entered a temporary isolate, so exit it afterwards.

  // Clobbers all volatile registers, including the callback ID in R9.
  // Resets CSP and SP, important for EnterSafepoint below.
  __ blr(R10);

  // Exit the temporary isolate.
  {
    __ SetupDartSP();
    __ EnterFrame(0);
    __ ReserveAlignedFrameSpace(0);

#if defined(DART_TARGET_OS_FUCHSIA)
    // TODO(https://dartbug.com/52579): Remove.
    if (FLAG_precompiled_mode) {
      GenerateLoadBSSEntry(BSS::Relocation::DRT_ExitTemporaryIsolate, R4, R9);
    } else {
      Label call;
      __ ldr(R4, compiler::Address::PC(2 * Instr::kInstrSize));
      __ b(&call);
      __ Emit64(reinterpret_cast<int64_t>(&DLRT_ExitTemporaryIsolate));
      __ Bind(&call);
    }
#else
    GenerateLoadFfiCallbackMetadataRuntimeFunction(
        FfiCallbackMetadata::kExitTemporaryIsolate, R4);
#endif

    __ mov(CSP, SP);
    __ blr(R4);
    __ mov(SP, CSP);
    __ mov(THR, R0);

    __ LeaveFrame();
    __ RestoreCSP();
  }

  __ Bind(&done);

  // Pop LR and THR from the real stack (CSP).
  RESTORES_LR_FROM_FRAME(__ ldp(
      THR, LR, Address(CSP, 2 * target::kWordSize, Address::PairPostIndex)));

  __ ret();

  ASSERT_LESS_OR_EQUAL(__ CodeSize() - shared_stub_start,
                       FfiCallbackMetadata::kNativeCallbackSharedStubSize);
  ASSERT_LESS_OR_EQUAL(__ CodeSize(), FfiCallbackMetadata::kPageSize);

#if defined(DEBUG)
  while (__ CodeSize() < FfiCallbackMetadata::kPageSize) {
    __ Breakpoint();
  }
#endif
#endif  // !defined(HOST_ARCH_ARM64)
}

void StubCodeCompiler::GenerateDispatchTableNullErrorStub() {
  __ EnterStubFrame();
  __ SmiTag(DispatchTableNullErrorABI::kClassIdReg);
  __ PushRegister(DispatchTableNullErrorABI::kClassIdReg);
  __ CallRuntime(kDispatchTableNullErrorRuntimeEntry, /*argument_count=*/1);
  // The NullError runtime entry does not return.
  __ Breakpoint();
}

void StubCodeCompiler::GenerateRangeError(bool with_fpu_regs) {
  auto perform_runtime_call = [&]() {
    // If the generated code has unboxed index/length we need to box them before
    // calling the runtime entry.
    if (GenericCheckBoundInstr::UseUnboxedRepresentation()) {
      Label length, smi_case;

      // The user-controlled index might not fit into a Smi.
#if !defined(DART_COMPRESSED_POINTERS)
      __ adds(RangeErrorABI::kIndexReg, RangeErrorABI::kIndexReg,
              compiler::Operand(RangeErrorABI::kIndexReg));
      __ BranchIf(NO_OVERFLOW, &length);
#else
      __ mov(TMP, RangeErrorABI::kIndexReg);
      __ SmiTag(RangeErrorABI::kIndexReg);
      __ sxtw(RangeErrorABI::kIndexReg, RangeErrorABI::kIndexReg);
      __ cmp(TMP,
             compiler::Operand(RangeErrorABI::kIndexReg, ASR, kSmiTagSize));
      __ BranchIf(EQ, &length);
#endif
      {
        // Allocate a mint, reload the two registers and populate the mint.
        __ PushRegister(NULL_REG);
        __ CallRuntime(kAllocateMintRuntimeEntry, /*argument_count=*/0);
        __ PopRegister(RangeErrorABI::kIndexReg);
        __ ldr(TMP,
               Address(FP, target::kWordSize *
                               StubCodeCompiler::WordOffsetFromFpToCpuRegister(
                                   RangeErrorABI::kIndexReg)));
        __ str(TMP, FieldAddress(RangeErrorABI::kIndexReg,
                                 target::Mint::value_offset()));
        __ ldr(RangeErrorABI::kLengthReg,
               Address(FP, target::kWordSize *
                               StubCodeCompiler::WordOffsetFromFpToCpuRegister(
                                   RangeErrorABI::kLengthReg)));
      }

      // Length is guaranteed to be in positive Smi range (it comes from a load
      // of a vm recognized array).
      __ Bind(&length);
      __ SmiTag(RangeErrorABI::kLengthReg);
    }
    __ PushRegistersInOrder(
        {RangeErrorABI::kLengthReg, RangeErrorABI::kIndexReg});
    __ CallRuntime(kRangeErrorRuntimeEntry, /*argument_count=*/2);
    __ Breakpoint();
  };

  GenerateSharedStubGeneric(
      /*save_fpu_registers=*/with_fpu_regs,
      with_fpu_regs
          ? target::Thread::range_error_shared_with_fpu_regs_stub_offset()
          : target::Thread::range_error_shared_without_fpu_regs_stub_offset(),
      /*allow_return=*/false, perform_runtime_call);
}

void StubCodeCompiler::GenerateWriteError(bool with_fpu_regs) {
  auto perform_runtime_call = [&]() {
    __ CallRuntime(kWriteErrorRuntimeEntry, /*argument_count=*/2);
    __ Breakpoint();
  };

  GenerateSharedStubGeneric(
      /*save_fpu_registers=*/with_fpu_regs,
      with_fpu_regs
          ? target::Thread::write_error_shared_with_fpu_regs_stub_offset()
          : target::Thread::write_error_shared_without_fpu_regs_stub_offset(),
      /*allow_return=*/false, perform_runtime_call);
}

// Input parameters:
//   LR : return address.
//   SP : address of return value.
//   R5 : address of the native function to call.
//   R2 : address of first argument in argument array.
//   R1 : argc_tag including number of arguments and function kind.
static void GenerateCallNativeWithWrapperStub(Assembler* assembler,
                                              Address wrapper) {
  const intptr_t thread_offset = target::NativeArguments::thread_offset();
  const intptr_t argc_tag_offset = target::NativeArguments::argc_tag_offset();
  const intptr_t argv_offset = target::NativeArguments::argv_offset();
  const intptr_t retval_offset = target::NativeArguments::retval_offset();

  __ EnterStubFrame();

  // Save exit frame information to enable stack walking as we are about
  // to transition to native code.
  __ StoreToOffset(FP, THR, target::Thread::top_exit_frame_info_offset());

  // Mark that the thread exited generated code through a runtime call.
  __ LoadImmediate(R6, target::Thread::exit_through_runtime_call());
  __ StoreToOffset(R6, THR, target::Thread::exit_through_ffi_offset());

#if defined(DEBUG)
  {
    Label ok;
    // Check that we are always entering from Dart code.
    __ LoadFromOffset(R6, THR, target::Thread::vm_tag_offset());
    __ CompareImmediate(R6, VMTag::kDartTagId);
    __ b(&ok, EQ);
    __ Stop("Not coming from Dart code.");
    __ Bind(&ok);
  }
#endif

  // Mark that the thread is executing native code.
  __ StoreToOffset(R5, THR, target::Thread::vm_tag_offset());

  WithExceptionCatchingTrampoline(assembler, [&]() {
    // Reserve space for the native arguments structure passed on the stack (the
    // outgoing pointer parameter to the native arguments structure is passed in
    // R0) and align frame before entering the C++ world.
    __ ReserveAlignedFrameSpace(target::NativeArguments::StructSize());

    // Initialize target::NativeArguments structure and call native function.
    // Registers R0, R1, R2, and R3 are used.

    ASSERT(thread_offset == 0 * target::kWordSize);
    // Set thread in NativeArgs.
    __ mov(R0, THR);

    ASSERT(argc_tag_offset == 1 * target::kWordSize);
    // Set argc in target::NativeArguments: R1 already contains argc.

    ASSERT(argv_offset == 2 * target::kWordSize);
    // Set argv in target::NativeArguments: R2 already contains argv.

    // Set retval in NativeArgs.
    ASSERT(retval_offset == 3 * target::kWordSize);
    __ AddImmediate(
        R3, FP,
        (target::frame_layout.param_end_from_fp + 1) * target::kWordSize);

    // Passing the structure by value as in runtime calls would require changing
    // Dart API for native functions.
    // For now, space is reserved on the stack and we pass a pointer to it.
    __ StoreToOffset(R0, SP, thread_offset);
    __ StoreToOffset(R1, SP, argc_tag_offset);
    __ StoreToOffset(R2, SP, argv_offset);
    __ StoreToOffset(R3, SP, retval_offset);
    __ mov(R0, SP);  // Pass the pointer to the target::NativeArguments.

    // We are entering runtime code, so the C stack pointer must be restored
    // from the stack limit to the top of the stack. We cache the stack limit
    // address in the Dart SP register, which is callee-saved in the C ABI.
    __ mov(R25, CSP);
    __ mov(CSP, SP);

    __ mov(R1, R5);  // Pass the function entrypoint to call.

    // Call native function invocation wrapper or redirection via simulator.
    __ Call(wrapper);

    // Restore SP and CSP.
    __ mov(SP, CSP);
    __ mov(CSP, R25);

    // Refresh pinned registers (write barrier mask, null, dispatch table, etc).
    __ RestorePinnedRegisters();

    // Mark that the thread is executing Dart code.
    __ LoadImmediate(R2, VMTag::kDartTagId);
    __ StoreToOffset(R2, THR, target::Thread::vm_tag_offset());

    // Mark that the thread has not exited generated Dart code.
    __ StoreToOffset(ZR, THR, target::Thread::exit_through_ffi_offset());

    // Reset exit frame information in Isolate's mutator thread structure.
    __ StoreToOffset(ZR, THR, target::Thread::top_exit_frame_info_offset());

    // Restore the global object pool after returning from runtime (old space is
    // moving, so the GOP could have been relocated).
    if (FLAG_precompiled_mode) {
      __ SetupGlobalPoolAndDispatchTable();
    }
  });

  __ LeaveStubFrame();
  __ ret();
}

void StubCodeCompiler::GenerateCallNoScopeNativeStub() {
  GenerateCallNativeWithWrapperStub(
      assembler,
      Address(THR,
              target::Thread::no_scope_native_wrapper_entry_point_offset()));
}

void StubCodeCompiler::GenerateCallAutoScopeNativeStub() {
  GenerateCallNativeWithWrapperStub(
      assembler,
      Address(THR,
              target::Thread::auto_scope_native_wrapper_entry_point_offset()));
}

// Input parameters:
//   LR : return address.
//   SP : address of return value.
//   R5 : address of the native function to call.
//   R2 : address of first argument in argument array.
//   R1 : argc_tag including number of arguments and function kind.
void StubCodeCompiler::GenerateCallBootstrapNativeStub() {
  GenerateCallNativeWithWrapperStub(
      assembler,
      Address(THR,
              target::Thread::bootstrap_native_wrapper_entry_point_offset()));
}

// Input parameters:
//   ARGS_DESC_REG: arguments descriptor array.
void StubCodeCompiler::GenerateCallStaticFunctionStub() {
  // Create a stub frame as we are pushing some objects on the stack before
  // calling into the runtime.
  __ EnterStubFrame();
  // Setup space on stack for return value and preserve arguments descriptor.
  __ Push(ARGS_DESC_REG);
  __ Push(ZR);
  __ CallRuntime(kPatchStaticCallRuntimeEntry, 0);
  // Get Code object result and restore arguments descriptor array.
  __ Pop(CODE_REG);
  __ Pop(ARGS_DESC_REG);
  // Remove the stub frame.
  __ LeaveStubFrame();
  // Jump to the dart function.
  __ LoadFieldFromOffset(R0, CODE_REG, target::Code::entry_point_offset());
  __ br(R0);
}

// Called from a static call only when an invalid code has been entered
// (invalid because its function was optimized or deoptimized).
// ARGS_DESC_REG: arguments descriptor array.
void StubCodeCompiler::GenerateFixCallersTargetStub() {
  Label monomorphic;
  __ BranchOnMonomorphicCheckedEntryJIT(&monomorphic);

  // Load code pointer to this stub from the thread:
  // The one that is passed in, is not correct - it points to the code object
  // that needs to be replaced.
  __ ldr(CODE_REG,
         Address(THR, target::Thread::fix_callers_target_code_offset()));
  // Create a stub frame as we are pushing some objects on the stack before
  // calling into the runtime.
  __ EnterStubFrame();
  // Setup space on stack for return value and preserve arguments descriptor.
  __ Push(ARGS_DESC_REG);
  __ Push(ZR);
  __ CallRuntime(kFixCallersTargetRuntimeEntry, 0);
  // Get Code object result and restore arguments descriptor array.
  __ Pop(CODE_REG);
  __ Pop(ARGS_DESC_REG);
  // Remove the stub frame.
  __ LeaveStubFrame();
  // Jump to the dart function.
  __ LoadFieldFromOffset(R0, CODE_REG, target::Code::entry_point_offset());
  __ br(R0);

  __ Bind(&monomorphic);
  // Load code pointer to this stub from the thread:
  // The one that is passed in, is not correct - it points to the code object
  // that needs to be replaced.
  __ ldr(CODE_REG,
         Address(THR, target::Thread::fix_callers_target_code_offset()));
  // Create a stub frame as we are pushing some objects on the stack before
  // calling into the runtime.
  __ EnterStubFrame();
  __ Push(ZR);  // Result slot.
  __ Push(R0);  // Preserve receiver.
  __ Push(R5);  // Old cache value (also 2nd return value).
  __ CallRuntime(kFixCallersTargetMonomorphicRuntimeEntry, 2);
  __ Pop(R5);        // Get target cache object.
  __ Pop(R0);        // Restore receiver.
  __ Pop(CODE_REG);  // Get target Code object.
  // Remove the stub frame.
  __ LeaveStubFrame();
  // Jump to the dart function.
  __ LoadFieldFromOffset(
      R1, CODE_REG,
      target::Code::entry_point_offset(CodeEntryKind::kMonomorphic));
  __ br(R1);
}

// Called from object allocate instruction when the allocation stub has been
// disabled.
void StubCodeCompiler::GenerateFixAllocationStubTargetStub() {
  // Load code pointer to this stub from the thread:
  // The one that is passed in, is not correct - it points to the code object
  // that needs to be replaced.
  __ ldr(CODE_REG,
         Address(THR, target::Thread::fix_allocation_stub_code_offset()));
  __ EnterStubFrame();
  // Setup space on stack for return value.
  __ Push(ZR);
  __ CallRuntime(kFixAllocationStubTargetRuntimeEntry, 0);
  // Get Code object result.
  __ Pop(CODE_REG);
  // Remove the stub frame.
  __ LeaveStubFrame();
  // Jump to the dart function.
  __ LoadFieldFromOffset(R0, CODE_REG, target::Code::entry_point_offset());
  __ br(R0);
}

// Called from object allocate instruction when the allocation stub for a
// generic class has been disabled.
void StubCodeCompiler::GenerateFixParameterizedAllocationStubTargetStub() {
  // Load code pointer to this stub from the thread:
  // The one that is passed in, is not correct - it points to the code object
  // that needs to be replaced.
  __ ldr(CODE_REG,
         Address(THR, target::Thread::fix_allocation_stub_code_offset()));
  __ EnterStubFrame();
  // Preserve type arguments register.
  __ Push(AllocateObjectABI::kTypeArgumentsReg);
  // Setup space on stack for return value.
  __ Push(ZR);
  __ CallRuntime(kFixAllocationStubTargetRuntimeEntry, 0);
  // Get Code object result.
  __ Pop(CODE_REG);
  // Restore type arguments register.
  __ Pop(AllocateObjectABI::kTypeArgumentsReg);
  // Remove the stub frame.
  __ LeaveStubFrame();
  // Jump to the dart function.
  __ LoadFieldFromOffset(R0, CODE_REG, target::Code::entry_point_offset());
  __ br(R0);
}

// Input parameters:
//   R2: smi-tagged argument count, may be zero.
//   FP[target::frame_layout.param_end_from_fp + 1]: last argument.
static void PushArrayOfArguments(Assembler* assembler) {
  // Allocate array to store arguments of caller.
  __ LoadObject(R1, NullObject());
  // R1: null element type for raw Array.
  // R2: smi-tagged argument count, may be zero.
  __ BranchLink(StubCodeAllocateArray());
  // R0: newly allocated array.
  // R2: smi-tagged argument count, may be zero (was preserved by the stub).
  __ Push(R0);  // Array is in R0 and on top of stack.
  __ SmiUntag(R2);
  __ add(R1, FP, Operand(R2, LSL, target::kWordSizeLog2));
  __ AddImmediate(R1,
                  target::frame_layout.param_end_from_fp * target::kWordSize);
  __ AddImmediate(R3, R0, target::Array::data_offset() - kHeapObjectTag);
  // R1: address of first argument on stack.
  // R3: address of first argument in array.

  Label loop, loop_exit;
  __ Bind(&loop);
  __ CompareRegisters(R2, ZR);
  __ b(&loop_exit, LE);
  __ ldr(R7, Address(R1));
  __ AddImmediate(R1, -target::kWordSize);
  __ AddImmediate(R3, target::kCompressedWordSize);
  __ AddImmediate(R2, R2, -1);
  __ StoreCompressedIntoObject(R0, Address(R3, -target::kCompressedWordSize),
                               R7);
  __ b(&loop);
  __ Bind(&loop_exit);
}

// Used by eager and lazy deoptimization. Preserve result in RAX if necessary.
// This stub translates optimized frame into unoptimized frame. The optimized
// frame can contain values in registers and on stack, the unoptimized
// frame contains all values on stack.
// Deoptimization occurs in following steps:
// - Push all registers that can contain values.
// - Call C routine to copy the stack and saved registers into temporary buffer.
// - Adjust caller's frame to correct unoptimized frame size.
// - Fill the unoptimized frame.
// - Materialize objects that require allocation (e.g. Double instances).
// GC can occur only after frame is fully rewritten.
// Stack after TagAndPushPP() below:
//   +------------------+
//   | Saved PP         | <- PP
//   +------------------+
//   | PC marker        | <- TOS
//   +------------------+
//   | Saved FP         | <- FP of stub
//   +------------------+
//   | return-address   |  (deoptimization point)
//   +------------------+
//   | Saved CODE_REG   |
//   +------------------+
//   | ...              | <- SP of optimized frame
//
// Parts of the code cannot GC, part of the code can GC.
static void GenerateDeoptimizationSequence(Assembler* assembler,
                                           DeoptStubKind kind) {
  // DeoptimizeCopyFrame expects a Dart frame, i.e. EnterDartFrame(0), but there
  // is no need to set the correct PC marker or load PP, since they get patched.
  __ EnterStubFrame();

  // The code in this frame may not cause GC. kDeoptimizeCopyFrameRuntimeEntry
  // and kDeoptimizeFillFrameRuntimeEntry are leaf runtime calls.
  const intptr_t saved_result_slot_from_fp =
      target::frame_layout.first_local_from_fp + 1 -
      (kNumberOfCpuRegisters - R0);
  const intptr_t saved_exception_slot_from_fp =
      target::frame_layout.first_local_from_fp + 1 -
      (kNumberOfCpuRegisters - R0);
  const intptr_t saved_stacktrace_slot_from_fp =
      target::frame_layout.first_local_from_fp + 1 -
      (kNumberOfCpuRegisters - R1);
  // Result in R0 is preserved as part of pushing all registers below.

  // Push registers in their enumeration order: lowest register number at
  // lowest address.
  for (intptr_t i = kNumberOfCpuRegisters - 1; i >= 0; i--) {
    const Register r = static_cast<Register>(i);
    if (r == CODE_REG) {
      // Save the original value of CODE_REG pushed before invoking this stub
      // instead of the value used to call this stub.
      COMPILE_ASSERT(R25 > CODE_REG);
      __ ldr(R25, Address(FP, 2 * target::kWordSize));
      __ str(R25, Address(SP, -1 * target::kWordSize, Address::PreIndex));
    } else if (r == R15) {
      // Because we save registers in decreasing order, IP0 will already be
      // saved.
      COMPILE_ASSERT(IP0 == R16);
      __ mov(IP0, R15);
      __ str(IP0, Address(SP, -1 * target::kWordSize, Address::PreIndex));
    } else if (r == R31) {
      __ str(ZR, Address(SP, -1 * target::kWordSize, Address::PreIndex));
    } else {
      __ str(r, Address(SP, -1 * target::kWordSize, Address::PreIndex));
    }
  }

  for (intptr_t reg_idx = kNumberOfVRegisters - 1; reg_idx >= 0; reg_idx--) {
    VRegister vreg = static_cast<VRegister>(reg_idx);
    __ PushQuad(vreg);
  }

  {
    __ mov(R0, SP);  // Pass address of saved registers block.
    LeafRuntimeScope rt(assembler,
                        /*frame_size=*/0,
                        /*preserve_registers=*/false);
    bool is_lazy =
        (kind == kLazyDeoptFromReturn) || (kind == kLazyDeoptFromThrow);
    __ LoadImmediate(R1, is_lazy ? 1 : 0);
    rt.Call(kDeoptimizeCopyFrameRuntimeEntry, 2);
    // Result (R0) is stack-size (FP - SP) in bytes.
  }

  if (kind == kLazyDeoptFromReturn) {
    // Restore result into R1 temporarily.
    __ LoadFromOffset(R1, FP, saved_result_slot_from_fp * target::kWordSize);
  } else if (kind == kLazyDeoptFromThrow) {
    // Restore result into R1 temporarily.
    __ LoadFromOffset(R1, FP, saved_exception_slot_from_fp * target::kWordSize);
    __ LoadFromOffset(R2, FP,
                      saved_stacktrace_slot_from_fp * target::kWordSize);
  }

  // There is a Dart Frame on the stack. We must restore PP and leave frame.
  __ RestoreCodePointer();
  __ LeaveStubFrame();
  __ sub(SP, FP, Operand(R0));

  // DeoptimizeFillFrame expects a Dart frame, i.e. EnterDartFrame(0), but there
  // is no need to set the correct PC marker or load PP, since they get patched.
  __ EnterStubFrame();

  if (kind == kLazyDeoptFromReturn) {
    __ Push(R1);  // Preserve result as first local.
  } else if (kind == kLazyDeoptFromThrow) {
    __ Push(R1);  // Preserve exception as first local.
    __ Push(R2);  // Preserve stacktrace as second local.
  }
  {
    __ mov(R0, FP);  // Pass last FP as parameter in R0.
    LeafRuntimeScope rt(assembler,
                        /*frame_size=*/0,
                        /*preserve_registers=*/false);
    rt.Call(kDeoptimizeFillFrameRuntimeEntry, 1);
  }
  if (kind == kLazyDeoptFromReturn) {
    // Restore result into R1.
    __ LoadFromOffset(
        R1, FP, target::frame_layout.first_local_from_fp * target::kWordSize);
  } else if (kind == kLazyDeoptFromThrow) {
    // Restore result into R1.
    __ LoadFromOffset(
        R1, FP, target::frame_layout.first_local_from_fp * target::kWordSize);
    __ LoadFromOffset(
        R2, FP,
        (target::frame_layout.first_local_from_fp - 1) * target::kWordSize);
  }
  // Code above cannot cause GC.
  // There is a Dart Frame on the stack. We must restore PP and leave frame.
  __ RestoreCodePointer();
  __ LeaveStubFrame();

  // Frame is fully rewritten at this point and it is safe to perform a GC.
  // Materialize any objects that were deferred by FillFrame because they
  // require allocation.
  // Enter stub frame with loading PP. The caller's PP is not materialized yet.
  __ EnterStubFrame();
  if (kind == kLazyDeoptFromReturn) {
    __ Push(R1);  // Preserve result, it will be GC-d here.
  } else if (kind == kLazyDeoptFromThrow) {
    // Preserve CODE_REG for one more runtime call.
    __ Push(CODE_REG);
    __ Push(R1);  // Preserve exception, it will be GC-d here.
    __ Push(R2);  // Preserve stacktrace, it will be GC-d here.
  }

  __ Push(ZR);  // Space for the result.
  __ CallRuntime(kDeoptimizeMaterializeRuntimeEntry, 0);
  // Result tells stub how many bytes to remove from the expression stack
  // of the bottom-most frame. They were used as materialization arguments.
  __ Pop(R2);
  __ SmiUntag(R2);
  if (kind == kLazyDeoptFromReturn) {
    __ Pop(R0);  // Restore result.
  } else if (kind == kLazyDeoptFromThrow) {
    __ Pop(R1);  // Restore stacktrace.
    __ Pop(R0);  // Restore exception.
    __ Pop(CODE_REG);
  }
  __ LeaveStubFrame();
  // Remove materialization arguments.
  __ add(SP, SP, Operand(R2));
  // The caller is responsible for emitting the return instruction.

  if (kind == kLazyDeoptFromThrow) {
    // Unoptimized frame is now ready to accept the exception. Rethrow it to
    // find the right handler. Ask rethrow machinery to bypass debugger it
    // was already notified about this exception.
    __ EnterStubFrame();
    __ Push(ZR);  // Space for the return value (unused).
    __ Push(R0);  // Exception
    __ Push(R1);  // Stacktrace
    __ PushImmediate(target::ToRawSmi(1));  // Bypass debugger
    __ CallRuntime(kReThrowRuntimeEntry, 3);
    __ LeaveStubFrame();
  }
}

// R0: result, must be preserved
void StubCodeCompiler::GenerateDeoptimizeLazyFromReturnStub() {
  // Push zap value instead of CODE_REG for lazy deopt.
  __ LoadImmediate(TMP, kZapCodeReg);
  __ Push(TMP);
  // Return address for "call" to deopt stub.
  WRITES_RETURN_ADDRESS_TO_LR(__ LoadImmediate(LR, kZapReturnAddress));
  __ ldr(CODE_REG,
         Address(THR, target::Thread::lazy_deopt_from_return_stub_offset()));
  GenerateDeoptimizationSequence(assembler, kLazyDeoptFromReturn);
  __ ret();
}

// R0: exception, must be preserved
// R1: stacktrace, must be preserved
void StubCodeCompiler::GenerateDeoptimizeLazyFromThrowStub() {
  // Push zap value instead of CODE_REG for lazy deopt.
  __ LoadImmediate(TMP, kZapCodeReg);
  __ Push(TMP);
  // Return address for "call" to deopt stub.
  WRITES_RETURN_ADDRESS_TO_LR(__ LoadImmediate(LR, kZapReturnAddress));
  __ ldr(CODE_REG,
         Address(THR, target::Thread::lazy_deopt_from_throw_stub_offset()));
  GenerateDeoptimizationSequence(assembler, kLazyDeoptFromThrow);
  __ ret();
}

void StubCodeCompiler::GenerateDeoptimizeStub() {
  __ Push(CODE_REG);
  __ ldr(CODE_REG, Address(THR, target::Thread::deoptimize_stub_offset()));
  GenerateDeoptimizationSequence(assembler, kEagerDeopt);
  __ ret();
}

// IC_DATA_REG: ICData/MegamorphicCache
static void GenerateNoSuchMethodDispatcherBody(Assembler* assembler) {
  __ EnterStubFrame();

  __ ldr(ARGS_DESC_REG,
         FieldAddress(IC_DATA_REG,
                      target::CallSiteData::arguments_descriptor_offset()));

  // Load the receiver.
  __ LoadCompressedSmiFieldFromOffset(
      R2, ARGS_DESC_REG, target::ArgumentsDescriptor::size_offset());
  __ add(TMP, FP, Operand(R2, LSL, target::kWordSizeLog2 - 1));  // R2 is Smi.
  __ LoadFromOffset(R6, TMP,
                    target::frame_layout.param_end_from_fp * target::kWordSize);
  __ Push(ZR);             // Result slot.
  __ Push(R6);             // Receiver.
  __ Push(IC_DATA_REG);    // ICData/MegamorphicCache.
  __ Push(ARGS_DESC_REG);  // Arguments descriptor.

  // Adjust arguments count.
  __ LoadCompressedSmiFieldFromOffset(
      R3, ARGS_DESC_REG, target::ArgumentsDescriptor::type_args_len_offset());
  __ AddImmediate(TMP, R2, 1, kObjectBytes);  // Include the type arguments.
  __ cmp(R3, Operand(0), kObjectBytes);
  // R2 <- (R3 == 0) ? R2 : TMP + 1 (R2 : R2 + 2).
  __ csinc(R2, R2, TMP, EQ, kObjectBytes);

  // R2: Smi-tagged arguments array length.
  PushArrayOfArguments(assembler);
  const intptr_t kNumArgs = 4;
  __ CallRuntime(kNoSuchMethodFromCallStubRuntimeEntry, kNumArgs);
  __ Drop(4);
  __ Pop(R0);  // Return value.
  __ LeaveStubFrame();
  __ ret();
}

static void GenerateDispatcherCode(Assembler* assembler,
                                   Label* call_target_function) {
  __ Comment("NoSuchMethodDispatch");
  // When lazily generated invocation dispatchers are disabled, the
  // miss-handler may return null.
  __ CompareObject(R0, NullObject());
  __ b(call_target_function, NE);

  GenerateNoSuchMethodDispatcherBody(assembler);
}

// Input:
//   ARGS_DESC_REG - arguments descriptor
//   IC_DATA_REG - icdata/megamorphic_cache
void StubCodeCompiler::GenerateNoSuchMethodDispatcherStub() {
  GenerateNoSuchMethodDispatcherBody(assembler);
}

// Called for inline allocation of arrays.
// Input registers (preserved):
//   LR: return address.
//   AllocateArrayABI::kLengthReg: array length as Smi.
//   AllocateArrayABI::kTypeArgumentsReg: type arguments of array.
// Output registers:
//   AllocateArrayABI::kResultReg: newly allocated array.
// Clobbered:
//   R3, R7
void StubCodeCompiler::GenerateAllocateArrayStub() {
  if (!FLAG_use_slow_path && FLAG_inline_alloc) {
    Label slow_case;
    // Compute the size to be allocated, it is based on the array length
    // and is computed as:
    // RoundedAllocationSize(
    //     (array_length * kCompressedWordSize) + target::Array::header_size()).
    // Check that length is a Smi.
    __ BranchIfNotSmi(AllocateArrayABI::kLengthReg, &slow_case);

    // Check length >= 0 && length <= kMaxNewSpaceElements
    const intptr_t max_len =
        target::ToRawSmi(target::Array::kMaxNewSpaceElements);
    __ CompareImmediate(AllocateArrayABI::kLengthReg, max_len, kObjectBytes);
    __ b(&slow_case, HI);

    const intptr_t cid = kArrayCid;
    NOT_IN_PRODUCT(__ MaybeTraceAllocation(kArrayCid, &slow_case, R4));

    // Calculate and align allocation size.
    // Load new object start and calculate next object start.
    // AllocateArrayABI::kTypeArgumentsReg: type arguments of array.
    // AllocateArrayABI::kLengthReg: array length as Smi.
    __ ldr(AllocateArrayABI::kResultReg,
           Address(THR, target::Thread::top_offset()));
    intptr_t fixed_size_plus_alignment_padding =
        target::Array::header_size() +
        target::ObjectAlignment::kObjectAlignment - 1;
    __ LoadImmediate(R3, fixed_size_plus_alignment_padding);
// AllocateArrayABI::kLengthReg is Smi.
#if defined(DART_COMPRESSED_POINTERS)
    __ add(R3, R3, Operand(AllocateArrayABI::kLengthReg, LSL, 1), kObjectBytes);
#else
    __ add(R3, R3, Operand(AllocateArrayABI::kLengthReg, LSL, 2), kObjectBytes);
#endif
    ASSERT(kSmiTagShift == 1);
    __ andi(R3, R3,
            Immediate(~(target::ObjectAlignment::kObjectAlignment - 1)));
    // AllocateArrayABI::kResultReg: potential new object start.
    // R3: object size in bytes.
    __ adds(R7, R3, Operand(AllocateArrayABI::kResultReg));
    __ b(&slow_case, CS);  // Branch if unsigned overflow.

    // Check if the allocation fits into the remaining space.
    // AllocateArrayABI::kResultReg: potential new object start.
    // AllocateArrayABI::kTypeArgumentsReg: type arguments of array.
    // AllocateArrayABI::kLengthReg: array length as Smi.
    // R3: array size.
    // R7: potential next object start.
    __ LoadFromOffset(TMP, THR, target::Thread::end_offset());
    __ CompareRegisters(R7, TMP);
    __ b(&slow_case, CS);  // Branch if unsigned higher or equal.
    __ CheckAllocationCanary(AllocateArrayABI::kResultReg);

    // Successfully allocated the object(s), now update top to point to
    // next object start and initialize the object.
    // AllocateArrayABI::kResultReg: potential new object start.
    // R3: array size.
    // R7: potential next object start.
    __ str(R7, Address(THR, target::Thread::top_offset()));
    __ add(AllocateArrayABI::kResultReg, AllocateArrayABI::kResultReg,
           Operand(kHeapObjectTag));

    // AllocateArrayABI::kResultReg: new object start as a tagged pointer.
    // AllocateArrayABI::kTypeArgumentsReg: type arguments of array.
    // AllocateArrayABI::kLengthReg: array length as Smi.
    // R3: array size.
    // R7: new object end address.

    // Calculate the size tag.
    // AllocateArrayABI::kResultReg: new object start as a tagged pointer.
    // AllocateArrayABI::kLengthReg: array length as Smi.
    // R3: array size.
    // R7: new object end address.
    const intptr_t shift = target::UntaggedObject::kTagBitsSizeTagPos -
                           target::ObjectAlignment::kObjectAlignmentLog2;
    __ CompareImmediate(R3, target::UntaggedObject::kSizeTagMaxSizeTag);
    // If no size tag overflow, shift R3 left, else set R3 to zero.
    __ LslImmediate(TMP, R3, shift);
    __ csel(R3, TMP, R3, LS);
    __ csel(R3, ZR, R3, HI);

    // Get the class index and insert it into the tags.
    const uword tags =
        target::MakeTagWordForNewSpaceObject(cid, /*instance_size=*/0);

    __ LoadImmediate(TMP, tags);
    __ orr(R3, R3, Operand(TMP));
    __ StoreFieldToOffset(R3, AllocateArrayABI::kResultReg,
                          target::Array::tags_offset());

    // Store the type argument field.
    __ StoreCompressedIntoObjectOffsetNoBarrier(
        AllocateArrayABI::kResultReg, target::Array::type_arguments_offset(),
        AllocateArrayABI::kTypeArgumentsReg);

    // Set the length field.
    __ StoreCompressedIntoObjectOffsetNoBarrier(AllocateArrayABI::kResultReg,
                                                target::Array::length_offset(),
                                                AllocateArrayABI::kLengthReg);

    // Initialize all array elements to raw_null.
    // AllocateArrayABI::kResultReg: new object start as a tagged pointer.
    // R7: new object end address.
    // AllocateArrayABI::kLengthReg: array length as Smi.
    __ AddImmediate(R3, AllocateArrayABI::kResultReg,
                    target::Array::data_offset() - kHeapObjectTag);
    // R3: iterator which initially points to the start of the variable
    // data area to be initialized.
#if defined(DART_COMPRESSED_POINTERS)
    const Register kWordOfNulls = TMP;
    __ andi(kWordOfNulls, NULL_REG, Immediate(0xFFFFFFFF));
    __ orr(kWordOfNulls, kWordOfNulls, Operand(kWordOfNulls, LSL, 32));
#else
    const Register kWordOfNulls = NULL_REG;
#endif
    Label loop;
    __ Bind(&loop);
    ASSERT(target::kObjectAlignment == 2 * target::kWordSize);
    __ stp(kWordOfNulls, kWordOfNulls,
           Address(R3, 2 * target::kWordSize, Address::PairPostIndex));
    // Safe to only check every kObjectAlignment bytes instead of each word.
    ASSERT(kAllocationRedZoneSize >= target::kObjectAlignment);
    __ CompareRegisters(R3, R7);
    __ b(&loop, UNSIGNED_LESS);
    __ WriteAllocationCanary(R7);  // Fix overshoot.

    // Done allocating and initializing the array.
    // AllocateArrayABI::kResultReg: new object.
    // AllocateArrayABI::kLengthReg: array length as Smi (preserved).
    __ ret();

    // Unable to allocate the array using the fast inline code, just call
    // into the runtime.
    __ Bind(&slow_case);
  }
  // Create a stub frame as we are pushing some objects on the stack before
  // calling into the runtime.
  __ EnterStubFrame();
  // Setup space on stack for return value.
  // Push array length as Smi and element type.
  __ Push(ZR);
  __ Push(AllocateArrayABI::kLengthReg);
  __ Push(AllocateArrayABI::kTypeArgumentsReg);
  __ CallRuntime(kAllocateArrayRuntimeEntry, 2);

  // Write-barrier elimination might be enabled for this array (depending on the
  // array length). To be sure we will check if the allocated object is in old
  // space and if so call a leaf runtime to add it to the remembered set.
  __ ldr(AllocateArrayABI::kResultReg, Address(SP, 2 * target::kWordSize));
  EnsureIsNewOrRemembered();

  // Pop arguments; result is popped in IP.
  __ Pop(AllocateArrayABI::kTypeArgumentsReg);
  __ Pop(AllocateArrayABI::kLengthReg);
  __ Pop(AllocateArrayABI::kResultReg);
  __ LeaveStubFrame();
  __ ret();
}

void StubCodeCompiler::GenerateAllocateMintSharedWithFPURegsStub() {
  // For test purpose call allocation stub without inline allocation attempt.
  if (!FLAG_use_slow_path && FLAG_inline_alloc) {
    Label slow_case;
    __ TryAllocate(compiler::MintClass(), &slow_case, Assembler::kNearJump,
                   AllocateMintABI::kResultReg, AllocateMintABI::kTempReg);
    __ Ret();

    __ Bind(&slow_case);
  }
  COMPILE_ASSERT(AllocateMintABI::kResultReg ==
                 SharedSlowPathStubABI::kResultReg);
  GenerateSharedStub(/*save_fpu_registers=*/true, &kAllocateMintRuntimeEntry,
                     target::Thread::allocate_mint_with_fpu_regs_stub_offset(),
                     /*allow_return=*/true,
                     /*store_runtime_result_in_result_register=*/true);
}

void StubCodeCompiler::GenerateAllocateMintSharedWithoutFPURegsStub() {
  // For test purpose call allocation stub without inline allocation attempt.
  if (!FLAG_use_slow_path && FLAG_inline_alloc) {
    Label slow_case;
    __ TryAllocate(compiler::MintClass(), &slow_case, Assembler::kNearJump,
                   AllocateMintABI::kResultReg, AllocateMintABI::kTempReg);
    __ Ret();

    __ Bind(&slow_case);
  }
  COMPILE_ASSERT(AllocateMintABI::kResultReg ==
                 SharedSlowPathStubABI::kResultReg);
  GenerateSharedStub(
      /*save_fpu_registers=*/false, &kAllocateMintRuntimeEntry,
      target::Thread::allocate_mint_without_fpu_regs_stub_offset(),
      /*allow_return=*/true,
      /*store_runtime_result_in_result_register=*/true);
}

// Called when invoking Dart code from C++ (VM code).
// Input parameters:
//   LR : points to return address.
//   R0 : target code or entry point (in bare instructions mode).
//   R1 : arguments descriptor array.
//   R2 : arguments array.
//   R3 : current thread.
void StubCodeCompiler::GenerateInvokeDartCodeStub() {
  __ Comment("InvokeDartCodeStub");

  // Copy the C stack pointer (CSP/R31) into the stack pointer we'll actually
  // use to access the stack (SP/R15) and set the C stack pointer to near the
  // stack limit, loaded from the Thread held in R3, to prevent signal handlers
  // from over-writing Dart frames.
  __ mov(SP, CSP);
  __ SetupCSPFromThread(R3);
  __ EnterFrame(0);

  // Push code object to PC marker slot.
  __ ldr(TMP, Address(R3, target::Thread::invoke_dart_code_stub_offset()));
  __ Push(TMP);

#if defined(DART_TARGET_OS_FUCHSIA)
  __ str(R18, Address(R3, target::Thread::saved_shadow_call_stack_offset()));
#elif defined(USING_SHADOW_CALL_STACK)
#error Unimplemented
#endif

  __ PushNativeCalleeSavedRegisters();

  // Set up THR, which caches the current thread in Dart code.
  if (THR != R3) {
    __ mov(THR, R3);
  }

  // Refresh pinned registers (write barrier mask, null, dispatch table, etc).
  __ RestorePinnedRegisters();

  // Save the current VMTag on the stack.
  __ LoadFromOffset(R4, THR, target::Thread::vm_tag_offset());
  __ Push(R4);

  // Save top resource and top exit frame info. Use R6 as a temporary register.
  // StackFrameIterator reads the top exit frame info saved in this frame.
  __ LoadFromOffset(R6, THR, target::Thread::top_resource_offset());
  __ StoreToOffset(ZR, THR, target::Thread::top_resource_offset());
  __ Push(R6);

  __ LoadFromOffset(R6, THR, target::Thread::exit_through_ffi_offset());
  __ Push(R6);
  __ StoreToOffset(ZR, THR, target::Thread::exit_through_ffi_offset());

  __ LoadFromOffset(R6, THR, target::Thread::top_exit_frame_info_offset());
  __ StoreToOffset(ZR, THR, target::Thread::top_exit_frame_info_offset());
  // target::frame_layout.exit_link_slot_from_entry_fp must be kept in sync
  // with the code below.
#if defined(DART_TARGET_OS_FUCHSIA)
  ASSERT(target::frame_layout.exit_link_slot_from_entry_fp == -24);
#else
  ASSERT(target::frame_layout.exit_link_slot_from_entry_fp == -23);
#endif
  __ Push(R6);
  // In debug mode, verify that we've pushed the top exit frame info at the
  // correct offset from FP.
  __ EmitEntryFrameVerification();

  // Mark that the thread is executing Dart code. Do this after initializing the
  // exit link for the profiler.
  __ LoadImmediate(R6, VMTag::kDartTagId);
  __ StoreToOffset(R6, THR, target::Thread::vm_tag_offset());

  // Load arguments descriptor array into R4, which is passed to Dart code.
  __ mov(R4, R1);

  // Load number of arguments into R5 and adjust count for type arguments.
  __ LoadCompressedSmiFieldFromOffset(
      R5, R4, target::ArgumentsDescriptor::count_offset());
  __ LoadCompressedSmiFieldFromOffset(
      R3, R4, target::ArgumentsDescriptor::type_args_len_offset());
  __ SmiUntag(R5);
  // Include the type arguments.
  __ cmp(R3, Operand(0), kObjectBytes);
  __ csinc(R5, R5, R5, EQ);  // R5 <- (R3 == 0) ? R5 : R5 + 1

  // Compute address of 'arguments array' data area into R2.
  __ AddImmediate(R2, R2, target::Array::data_offset() - kHeapObjectTag);

  // Set up arguments for the Dart call.
  Label push_arguments;
  Label done_push_arguments;
  __ cmp(R5, Operand(0));
  __ b(&done_push_arguments, EQ);  // check if there are arguments.
  __ LoadImmediate(R1, 0);
  __ Bind(&push_arguments);
  __ LoadCompressed(R3, Address(R2));
  __ Push(R3);
  __ add(R1, R1, Operand(1));
  __ add(R2, R2, Operand(target::kCompressedWordSize));
  __ cmp(R1, Operand(R5));
  __ b(&push_arguments, LT);
  __ Bind(&done_push_arguments);

  if (FLAG_precompiled_mode) {
    __ SetupGlobalPoolAndDispatchTable();
    __ mov(CODE_REG, ZR);  // GC-safe value into CODE_REG.
  } else {
    // We now load the pool pointer(PP) with a GC safe value as we are about to
    // invoke dart code. We don't need a real object pool here.
    // Smi zero does not work because ARM64 assumes PP to be untagged.
    __ LoadObject(PP, NullObject());
    __ mov(CODE_REG, R0);
    __ ldr(R0, FieldAddress(CODE_REG, target::Code::entry_point_offset()));
  }

  // Call the Dart code entrypoint.
  __ blr(R0);  // R4 is the arguments descriptor array.
  __ Comment("InvokeDartCodeStub return");

  // Get rid of arguments pushed on the stack.
  __ AddImmediate(
      SP, FP,
      target::frame_layout.exit_link_slot_from_entry_fp * target::kWordSize);

  // Restore the saved top exit frame info and top resource back into the
  // Isolate structure. Uses R6 as a temporary register for this.
  __ Pop(R6);
  __ StoreToOffset(R6, THR, target::Thread::top_exit_frame_info_offset());
  __ Pop(R6);
  __ StoreToOffset(R6, THR, target::Thread::exit_through_ffi_offset());
  __ Pop(R6);
  __ StoreToOffset(R6, THR, target::Thread::top_resource_offset());

  // Restore the current VMTag from the stack.
  __ Pop(R4);
  __ StoreToOffset(R4, THR, target::Thread::vm_tag_offset());

  __ PopNativeCalleeSavedRegisters();

  // Restore the frame pointer and C stack pointer and return.
  __ LeaveFrame();
  __ RestoreCSP();
  __ ret();
}

// Helper to generate space allocation of context stub.
// This does not initialise the fields of the context.
// Input:
//   R1: number of context variables.
// Output:
//   R0: new allocated Context object.
// Clobbered:
//   R2, R3, R4, TMP
static void GenerateAllocateContextSpaceStub(Assembler* assembler,
                                             Label* slow_case) {
  // First compute the rounded instance size.
  // R1: number of context variables.
  intptr_t fixed_size_plus_alignment_padding =
      target::Context::header_size() +
      target::ObjectAlignment::kObjectAlignment - 1;
  __ LoadImmediate(R2, fixed_size_plus_alignment_padding);
  __ add(R2, R2, Operand(R1, LSL, kCompressedWordSizeLog2));
  ASSERT(kSmiTagShift == 1);
  __ andi(R2, R2, Immediate(~(target::ObjectAlignment::kObjectAlignment - 1)));

  NOT_IN_PRODUCT(__ MaybeTraceAllocation(kContextCid, slow_case, R4));
  // Now allocate the object.
  // R1: number of context variables.
  // R2: object size.
  __ ldr(R0, Address(THR, target::Thread::top_offset()));
  __ add(R3, R2, Operand(R0));
  // Check if the allocation fits into the remaining space.
  // R0: potential new object.
  // R1: number of context variables.
  // R2: object size.
  // R3: potential next object start.
  __ ldr(TMP, Address(THR, target::Thread::end_offset()));
  __ CompareRegisters(R3, TMP);
  __ b(slow_case, CS);  // Branch if unsigned higher or equal.
  __ CheckAllocationCanary(R0);

  // Successfully allocated the object, now update top to point to
  // next object start and initialize the object.
  // R0: new object.
  // R1: number of context variables.
  // R2: object size.
  // R3: next object start.
  __ str(R3, Address(THR, target::Thread::top_offset()));
  __ add(R0, R0, Operand(kHeapObjectTag));

  // Calculate the size tag.
  // R0: new object.
  // R1: number of context variables.
  // R2: object size.
  const intptr_t shift = target::UntaggedObject::kTagBitsSizeTagPos -
                         target::ObjectAlignment::kObjectAlignmentLog2;
  __ CompareImmediate(R2, target::UntaggedObject::kSizeTagMaxSizeTag);
  // If no size tag overflow, shift R2 left, else set R2 to zero.
  __ LslImmediate(TMP, R2, shift);
  __ csel(R2, TMP, R2, LS);
  __ csel(R2, ZR, R2, HI);

  // Get the class index and insert it into the tags.
  // R2: size and bit tags.
  const uword tags =
      target::MakeTagWordForNewSpaceObject(kContextCid, /*instance_size=*/0);

  __ LoadImmediate(TMP, tags);
  __ orr(R2, R2, Operand(TMP));
  __ StoreFieldToOffset(R2, R0, target::Object::tags_offset());

  // Setup up number of context variables field.
  // R0: new object.
  // R1: number of context variables as integer value (not object).
  __ StoreFieldToOffset(R1, R0, target::Context::num_variables_offset(),
                        kFourBytes);
}

// Called for inline allocation of contexts.
// Input:
//   R1: number of context variables.
// Output:
//   R0: new allocated Context object.
// Clobbered:
//   R2, R3, R4, TMP
void StubCodeCompiler::GenerateAllocateContextStub() {
  if (!FLAG_use_slow_path && FLAG_inline_alloc) {
    Label slow_case;

    GenerateAllocateContextSpaceStub(assembler, &slow_case);

    // Setup the parent field.
    // R0: new object.
    // R1: number of context variables.
    __ StoreCompressedIntoObjectOffset(R0, target::Context::parent_offset(),
                                       NULL_REG);

    // Initialize the context variables.
    // R0: new object.
    // R1: number of context variables.
    __ AddImmediate(R3, R0,
                    target::Context::variable_offset(0) - kHeapObjectTag);
#if defined(DART_COMPRESSED_POINTERS)
    const Register kWordOfNulls = TMP;
    __ andi(kWordOfNulls, NULL_REG, Immediate(0xFFFFFFFF));
    __ orr(kWordOfNulls, kWordOfNulls, Operand(kWordOfNulls, LSL, 32));
#else
    const Register kWordOfNulls = NULL_REG;
#endif
    Label loop;
    __ Bind(&loop);
    ASSERT(target::kObjectAlignment == 2 * target::kWordSize);
    __ stp(kWordOfNulls, kWordOfNulls,
           Address(R3, 2 * target::kWordSize, Address::PairPostIndex));
    // Safe to only check every kObjectAlignment bytes instead of each word.
    ASSERT(kAllocationRedZoneSize >= target::kObjectAlignment);
    __ subs(R1, R1,
            Operand(target::kObjectAlignment / target::kCompressedWordSize));
    __ b(&loop, HI);
#if defined(DEBUG)
    __ ldr(TMP2, Address(THR, target::Thread::top_offset()));
    __ WriteAllocationCanary(TMP2);  // Fix overshoot.
#endif

    // Done allocating and initializing the context.
    // R0: new object.
    __ ret();

    __ Bind(&slow_case);
  }
  // Create a stub frame as we are pushing some objects on the stack before
  // calling into the runtime.
  __ EnterStubFrame();
  // Setup space on stack for return value.
  __ SmiTag(R1);
  __ PushObject(NullObject());
  __ Push(R1);
  __ CallRuntime(kAllocateContextRuntimeEntry, 1);  // Allocate context.
  __ Drop(1);  // Pop number of context variables argument.
  __ Pop(R0);  // Pop the new context object.

  // Write-barrier elimination might be enabled for this context (depending on
  // the size). To be sure we will check if the allocated object is in old
  // space and if so call a leaf runtime to add it to the remembered set.
  EnsureIsNewOrRemembered();

  // R0: new object
  // Restore the frame pointer.
  __ LeaveStubFrame();

  __ ret();
}

// Called for clone of contexts.
// Input:
//   R5: context variable to clone.
// Output:
//   R0: new allocated Context object.
// Clobbered:
//   R1, (R2), R3, R4, (TMP)
void StubCodeCompiler::GenerateCloneContextStub() {
  if (!FLAG_use_slow_path && FLAG_inline_alloc) {
    Label slow_case;

    // Load num. variable (int32) in the existing context.
    __ ldr(R1, FieldAddress(R5, target::Context::num_variables_offset()),
           kFourBytes);

    GenerateAllocateContextSpaceStub(assembler, &slow_case);

    // Load parent in the existing context.
    __ LoadCompressed(R3, FieldAddress(R5, target::Context::parent_offset()));
    // Setup the parent field.
    // R0: new context.
    __ StoreCompressedIntoObjectNoBarrier(
        R0, FieldAddress(R0, target::Context::parent_offset()), R3);

    // Clone the context variables.
    // R0: new context.
    // R1: number of context variables.
    {
      Label loop, done;
      // R3: Variable array address, new context.
      __ AddImmediate(R3, R0,
                      target::Context::variable_offset(0) - kHeapObjectTag);
      // R4: Variable array address, old context.
      __ AddImmediate(R4, R5,
                      target::Context::variable_offset(0) - kHeapObjectTag);

      __ Bind(&loop);
      __ subs(R1, R1, Operand(1));
      __ b(&done, MI);

      __ ldr(R5, Address(R4, R1, UXTX, Address::Scaled), kObjectBytes);
      __ str(R5, Address(R3, R1, UXTX, Address::Scaled), kObjectBytes);
      __ b(&loop, NE);  // Loop if R1 not zero.

      __ Bind(&done);
    }

    // Done allocating and initializing the context.
    // R0: new object.
    __ ret();

    __ Bind(&slow_case);
  }

  // Create a stub frame as we are pushing some objects on the stack before
  // calling into the runtime.
  __ EnterStubFrame();
  // Setup space on stack for return value.
  __ PushPair(R5, NULL_REG);
  __ CallRuntime(kCloneContextRuntimeEntry, 1);  // Clone context.
  // Pop number of context variables argument.
  // Pop the new context object.
  __ PopPair(R1, R0);

  // Write-barrier elimination might be enabled for this context (depending on
  // the size). To be sure we will check if the allocated object is in old
  // space and if so call a leaf runtime to add it to the remembered set.
  EnsureIsNewOrRemembered();

  // R0: new object
  // Restore the frame pointer.
  __ LeaveStubFrame();
  __ ret();
}

void StubCodeCompiler::GenerateWriteBarrierWrappersStub() {
  for (intptr_t i = 0; i < kNumberOfCpuRegisters; ++i) {
    if ((kDartAvailableCpuRegs & (1 << i)) == 0) continue;

    Register reg = static_cast<Register>(i);
    intptr_t start = __ CodeSize();
    SPILLS_LR_TO_FRAME(__ Push(LR));
    __ Push(kWriteBarrierObjectReg);
    __ mov(kWriteBarrierObjectReg, reg);
    __ Call(Address(THR, target::Thread::write_barrier_entry_point_offset()));
    __ Pop(kWriteBarrierObjectReg);
    RESTORES_LR_FROM_FRAME(__ Pop(LR));
    READS_RETURN_ADDRESS_FROM_LR(__ ret(LR));
    intptr_t end = __ CodeSize();

    RELEASE_ASSERT(end - start == kStoreBufferWrapperSize);
  }
}

// Helper stub to implement Assembler::StoreIntoObject/Array.
// Input parameters:
//   R1: Object (old)
//   R0: Value (old or new)
//  R25: Slot
// If R0 is new, add R1 to the store buffer. Otherwise R0 is old, mark R0
// and add it to the mark list.
COMPILE_ASSERT(kWriteBarrierObjectReg == R1);
COMPILE_ASSERT(kWriteBarrierValueReg == R0);
COMPILE_ASSERT(kWriteBarrierSlotReg == R25);
static void GenerateWriteBarrierStubHelper(Assembler* assembler, bool cards) {
  RegisterSet spill_set((1 << R2) | (1 << R3) | (1 << R4), 0);

  Label skip_marking;
  __ ldr(TMP, FieldAddress(R0, target::Object::tags_offset()));
  __ ldr(TMP2, Address(THR, target::Thread::write_barrier_mask_offset()));
  __ and_(TMP, TMP, Operand(TMP2));
  __ tsti(TMP, Immediate(target::UntaggedObject::kIncrementalBarrierMask));
  __ b(&skip_marking, ZERO);

  {
    // Atomically clear kNotMarkedBit.
    Label retry, is_new, done;
    __ PushRegisters(spill_set);
    __ AddImmediate(R3, R0, target::Object::tags_offset() - kHeapObjectTag);
    // R3: Untagged address of header word (atomics do not support offsets).
    if (TargetCPUFeatures::atomic_memory_supported()) {
      __ LoadImmediate(TMP, 1 << target::UntaggedObject::kNotMarkedBit);
      __ ldclr(TMP, TMP, R3);
      __ tbz(&done, TMP, target::UntaggedObject::kNotMarkedBit);
    } else {
      __ Bind(&retry);
      __ ldxr(R2, R3, kEightBytes);
      __ tbz(&done, R2, target::UntaggedObject::kNotMarkedBit);
      __ AndImmediate(R2, R2, ~(1 << target::UntaggedObject::kNotMarkedBit));
      __ stxr(R4, R2, R3, kEightBytes);
      __ cbnz(&retry, R4);
    }

    __ tbnz(&is_new, R0, target::ObjectAlignment::kNewObjectBitPosition);

    auto mark_stack_push = [&](intptr_t offset, const RuntimeEntry& entry) {
      __ LoadFromOffset(R4, THR, offset);
      __ LoadFromOffset(R2, R4, target::MarkingStackBlock::top_offset(),
                        kUnsignedFourBytes);
      __ add(R3, R4, Operand(R2, LSL, target::kWordSizeLog2));
      __ StoreToOffset(R0, R3, target::MarkingStackBlock::pointers_offset());
      __ add(R2, R2, Operand(1));
      __ StoreToOffset(R2, R4, target::MarkingStackBlock::top_offset(),
                       kUnsignedFourBytes);
      __ CompareImmediate(R2, target::MarkingStackBlock::kSize);
      __ b(&done, NE);

      {
        LeafRuntimeScope rt(assembler,
                            /*frame_size=*/0,
                            /*preserve_registers=*/true);
        __ mov(R0, THR);
        rt.Call(entry, /*argument_count=*/1);
      }
    };

    mark_stack_push(target::Thread::old_marking_stack_block_offset(),
                    kOldMarkingStackBlockProcessRuntimeEntry);
    __ b(&done);

    __ Bind(&is_new);
    mark_stack_push(target::Thread::new_marking_stack_block_offset(),
                    kNewMarkingStackBlockProcessRuntimeEntry);

    __ Bind(&done);
    __ clrex();
    __ PopRegisters(spill_set);
  }

  Label add_to_remembered_set, remember_card;
  __ Bind(&skip_marking);
  __ ldr(TMP, FieldAddress(R1, target::Object::tags_offset()));
  __ ldr(TMP2, FieldAddress(R0, target::Object::tags_offset()));
  __ and_(TMP, TMP2,
          Operand(TMP, LSR, target::UntaggedObject::kBarrierOverlapShift));
  __ tsti(TMP, Immediate(target::UntaggedObject::kGenerationalBarrierMask));
  __ b(&add_to_remembered_set, NOT_ZERO);
  __ ret();

  __ Bind(&add_to_remembered_set);
  if (cards) {
    __ LoadFieldFromOffset(TMP, R1, target::Object::tags_offset(), kFourBytes);
    __ tbnz(&remember_card, TMP, target::UntaggedObject::kCardRememberedBit);
  } else {
#if defined(DEBUG)
    Label ok;
    __ LoadFieldFromOffset(TMP, R1, target::Object::tags_offset(), kFourBytes);
    __ tbz(&ok, TMP, target::UntaggedObject::kCardRememberedBit);
    __ Stop("Wrong barrier");
    __ Bind(&ok);
#endif
  }
  {
    // Atomically clear kOldAndNotRememberedBit.
    Label retry, done;
    __ PushRegisters(spill_set);
    __ AddImmediate(R3, R1, target::Object::tags_offset() - kHeapObjectTag);
    // R3: Untagged address of header word (atomics do not support offsets).
    if (TargetCPUFeatures::atomic_memory_supported()) {
      __ LoadImmediate(TMP,
                       1 << target::UntaggedObject::kOldAndNotRememberedBit);
      __ ldclr(TMP, TMP, R3);
      __ tbz(&done, TMP, target::UntaggedObject::kOldAndNotRememberedBit);
    } else {
      __ Bind(&retry);
      __ ldxr(R2, R3, kEightBytes);
      __ tbz(&done, R2, target::UntaggedObject::kOldAndNotRememberedBit);
      __ AndImmediate(R2, R2,
                      ~(1 << target::UntaggedObject::kOldAndNotRememberedBit));
      __ stxr(R4, R2, R3, kEightBytes);
      __ cbnz(&retry, R4);
    }

    // Load the StoreBuffer block out of the thread. Then load top_ out of the
    // StoreBufferBlock and add the address to the pointers_.
    __ LoadFromOffset(R4, THR, target::Thread::store_buffer_block_offset());
    __ LoadFromOffset(R2, R4, target::StoreBufferBlock::top_offset(),
                      kUnsignedFourBytes);
    __ add(R3, R4, Operand(R2, LSL, target::kWordSizeLog2));
    __ StoreToOffset(R1, R3, target::StoreBufferBlock::pointers_offset());

    // Increment top_ and check for overflow.
    // R2: top_.
    // R4: StoreBufferBlock.
    __ add(R2, R2, Operand(1));
    __ StoreToOffset(R2, R4, target::StoreBufferBlock::top_offset(),
                     kUnsignedFourBytes);
    __ CompareImmediate(R2, target::StoreBufferBlock::kSize);
    __ b(&done, NE);

    {
      LeafRuntimeScope rt(assembler,
                          /*frame_size=*/0,
                          /*preserve_registers=*/true);
      __ mov(R0, THR);
      rt.Call(kStoreBufferBlockProcessRuntimeEntry, /*argument_count=*/1);
    }

    __ Bind(&done);
    __ PopRegisters(spill_set);
    __ ret();
  }
  if (cards) {
    Label remember_card_slow, retry;

    // Get card table.
    __ Bind(&remember_card);
    __ AndImmediate(TMP, R1, target::kPageMask);  // Page.
    __ ldr(TMP2,
           Address(TMP, target::Page::card_table_offset()));  // Card table.
    __ cbz(&remember_card_slow, TMP2);

    // Atomically dirty the card.
    __ sub(R25, R25, Operand(TMP));  // Offset in page.
    __ LsrImmediate(R25, R25, target::Page::kBytesPerCardLog2);  // Index.
    __ LoadImmediate(TMP, 1);
    __ lslv(TMP, TMP, R25);  // Bit mask. (Shift amount is mod 64.)
    __ LsrImmediate(R25, R25, target::kBitsPerWordLog2);  // Word index.
    __ add(TMP2, TMP2, Operand(R25, LSL, target::kWordSizeLog2));  // Word addr.

    if (TargetCPUFeatures::atomic_memory_supported()) {
      __ ldset(TMP, ZR, TMP2);
    } else {
      __ PushRegister(R0);
      __ Bind(&retry);
      __ ldxr(R25, TMP2);
      __ orr(R25, R25, Operand(TMP));
      __ stxr(R0, R25, TMP2);
      __ cbnz(&retry, R0);
      __ PopRegister(R0);
    }
    __ ret();

    // Card table not yet allocated.
    __ Bind(&remember_card_slow);
    {
      LeafRuntimeScope rt(assembler,
                          /*frame_size=*/0,
                          /*preserve_registers=*/true);
      __ mov(R0, R1);   // Arg0 = Object
      __ mov(R1, R25);  // Arg1 = Slot
      rt.Call(kRememberCardRuntimeEntry, /*argument_count=*/2);
    }
    __ ret();
  }
}

void StubCodeCompiler::GenerateWriteBarrierStub() {
  GenerateWriteBarrierStubHelper(assembler, false);
}

void StubCodeCompiler::GenerateArrayWriteBarrierStub() {
  GenerateWriteBarrierStubHelper(assembler, true);
}

static void GenerateAllocateObjectHelper(Assembler* assembler,
                                         bool is_cls_parameterized) {
  const Register kTagsReg = AllocateObjectABI::kTagsReg;

  {
    Label slow_case;

#if !defined(PRODUCT)
    {
      const Register kTraceAllocationTempReg = R8;
      const Register kCidRegister = R9;
      __ ExtractClassIdFromTags(kCidRegister, AllocateObjectABI::kTagsReg);
      __ MaybeTraceAllocation(kCidRegister, &slow_case,
                              kTraceAllocationTempReg);
    }
#endif

    const Register kNewTopReg = R3;

    // Bump allocation.
    {
      const Register kInstanceSizeReg = R4;
      const Register kEndReg = R5;

      __ ExtractInstanceSizeFromTags(kInstanceSizeReg, kTagsReg);

      // Load two words from Thread::top: top and end.
      // AllocateObjectABI::kResultReg: potential next object start.
      __ ldp(AllocateObjectABI::kResultReg, kEndReg,
             Address(THR, target::Thread::top_offset(), Address::PairOffset));

      __ add(kNewTopReg, AllocateObjectABI::kResultReg,
             Operand(kInstanceSizeReg));

      __ CompareRegisters(kEndReg, kNewTopReg);
      __ b(&slow_case, UNSIGNED_LESS_EQUAL);

      // Successfully allocated the object, now update top to point to
      // next object start and store the class in the class field of object.
      __ str(kNewTopReg, Address(THR, target::Thread::top_offset()));
    }  // kInstanceSizeReg = R4, kEndReg = R5

    // Tags.
    __ str(kTagsReg, Address(AllocateObjectABI::kResultReg,
                             target::Object::tags_offset()));

    // Initialize the remaining words of the object.
    {
      const Register kFieldReg = R4;
      __ AddImmediate(kFieldReg, AllocateObjectABI::kResultReg,
                      target::Instance::first_field_offset());
#if defined(DART_COMPRESSED_POINTERS)
      const Register kWordOfNulls = TMP;
      __ andi(kWordOfNulls, NULL_REG, Immediate(0xFFFFFFFF));
      __ orr(kWordOfNulls, kWordOfNulls, Operand(kWordOfNulls, LSL, 32));
#else
      const Register kWordOfNulls = NULL_REG;
#endif
      Label loop;
      __ Bind(&loop);
      ASSERT(target::kObjectAlignment == 2 * target::kWordSize);
      __ stp(kWordOfNulls, kWordOfNulls,
             Address(kFieldReg, 2 * target::kWordSize, Address::PairPostIndex));
      // Safe to only check every kObjectAlignment bytes instead of each word.
      ASSERT(kAllocationRedZoneSize >= target::kObjectAlignment);
      __ CompareRegisters(kFieldReg, kNewTopReg);
      __ b(&loop, UNSIGNED_LESS);
      __ WriteAllocationCanary(kNewTopReg);  // Fix overshoot.
    }  // kFieldReg = R4

    __ AddImmediate(AllocateObjectABI::kResultReg,
                    AllocateObjectABI::kResultReg, kHeapObjectTag);

    if (is_cls_parameterized) {
      Label not_parameterized_case;

      const Register kClsIdReg = R4;
      const Register kTypeOffsetReg = R5;

      __ ExtractClassIdFromTags(kClsIdReg, kTagsReg);

      // Load class' type_arguments_field offset in words.
      __ LoadClassById(kTypeOffsetReg, kClsIdReg);
      __ ldr(
          kTypeOffsetReg,
          FieldAddress(kTypeOffsetReg,
                       target::Class::
                           host_type_arguments_field_offset_in_words_offset()),
          kFourBytes);

      // Set the type arguments in the new object.
      __ add(kTypeOffsetReg, AllocateObjectABI::kResultReg,
             Operand(kTypeOffsetReg, LSL, target::kCompressedWordSizeLog2));
      __ StoreCompressedIntoObjectNoBarrier(
          AllocateObjectABI::kResultReg, FieldAddress(kTypeOffsetReg, 0),
          AllocateObjectABI::kTypeArgumentsReg);

      __ Bind(&not_parameterized_case);
    }  // kClsIdReg = R4, kTypeOffsetReg = R5

    __ ret();

    __ Bind(&slow_case);
  }  // kNewTopReg = R3

  // Fall back on slow case:
  if (!is_cls_parameterized) {
    __ mov(AllocateObjectABI::kTypeArgumentsReg, NULL_REG);
  }
  // Tail call to generic allocation stub.
  __ ldr(
      R3,
      Address(THR, target::Thread::allocate_object_slow_entry_point_offset()));
  __ br(R3);
}

// Called for inline allocation of objects (any class).
void StubCodeCompiler::GenerateAllocateObjectStub() {
  GenerateAllocateObjectHelper(assembler, /*is_cls_parameterized=*/false);
}

void StubCodeCompiler::GenerateAllocateObjectParameterizedStub() {
  GenerateAllocateObjectHelper(assembler, /*is_cls_parameterized=*/true);
}

void StubCodeCompiler::GenerateAllocateObjectSlowStub() {
  if (!FLAG_precompiled_mode) {
    __ ldr(CODE_REG,
           Address(THR, target::Thread::call_to_runtime_stub_offset()));
  }

  __ ExtractClassIdFromTags(AllocateObjectABI::kTagsReg,
                            AllocateObjectABI::kTagsReg);

  // Create a stub frame as we are pushing some objects on the stack before
  // calling into the runtime.
  __ EnterStubFrame();

  __ LoadClassById(R0, AllocateObjectABI::kTagsReg);
  __ PushPair(R0, NULL_REG);  // Pushes result slot, then class object.

  // Should be Object::null() if class is non-parameterized.
  __ Push(AllocateObjectABI::kTypeArgumentsReg);

  __ CallRuntime(kAllocateObjectRuntimeEntry, 2);

  // Load result off the stack into result register.
  __ ldr(AllocateObjectABI::kResultReg, Address(SP, 2 * target::kWordSize));

  // Write-barrier elimination is enabled for [cls] and we therefore need to
  // ensure that the object is in new-space or has remembered bit set.
  EnsureIsNewOrRemembered();

  __ LeaveStubFrame();

  __ ret();
}

// Called for inline allocation of objects.
void StubCodeCompiler::GenerateAllocationStubForClass(
    UnresolvedPcRelativeCalls* unresolved_calls,
    const Class& cls,
    const Code& allocate_object,
    const Code& allocat_object_parametrized) {
  classid_t cls_id = target::Class::GetId(cls);
  ASSERT(cls_id != kIllegalCid);

  // The generated code is different if the class is parameterized.
  const bool is_cls_parameterized = target::Class::NumTypeArguments(cls) > 0;
  ASSERT(!is_cls_parameterized || target::Class::TypeArgumentsFieldOffset(
                                      cls) != target::Class::kNoTypeArguments);

  const intptr_t instance_size = target::Class::GetInstanceSize(cls);
  ASSERT(instance_size > 0);

  const uword tags =
      target::MakeTagWordForNewSpaceObject(cls_id, instance_size);

  // Note: Keep in sync with helper function.
  const Register kTagsReg = AllocateObjectABI::kTagsReg;

  __ LoadImmediate(kTagsReg, tags);

  if (!FLAG_use_slow_path && FLAG_inline_alloc &&
      !target::Class::TraceAllocation(cls) &&
      target::SizeFitsInSizeTag(instance_size)) {
    RELEASE_ASSERT(AllocateObjectInstr::WillAllocateNewOrRemembered(cls));
    RELEASE_ASSERT(target::Heap::IsAllocatableInNewSpace(instance_size));

    if (is_cls_parameterized) {
      if (!IsSameObject(NullObject(),
                        CastHandle<Object>(allocat_object_parametrized))) {
        __ GenerateUnRelocatedPcRelativeTailCall();
        unresolved_calls->Add(new UnresolvedPcRelativeCall(
            __ CodeSize(), allocat_object_parametrized, /*is_tail_call=*/true));
      } else {
        __ ldr(R4,
               Address(THR,
                       target::Thread::
                           allocate_object_parameterized_entry_point_offset()));
        __ br(R4);
      }
    } else {
      if (!IsSameObject(NullObject(), CastHandle<Object>(allocate_object))) {
        __ GenerateUnRelocatedPcRelativeTailCall();
        unresolved_calls->Add(new UnresolvedPcRelativeCall(
            __ CodeSize(), allocate_object, /*is_tail_call=*/true));
      } else {
        __ ldr(
            R4,
            Address(THR, target::Thread::allocate_object_entry_point_offset()));
        __ br(R4);
      }
    }
  } else {
    if (!is_cls_parameterized) {
      __ LoadObject(AllocateObjectABI::kTypeArgumentsReg, NullObject());
    }
    __ ldr(R4,
           Address(THR,
                   target::Thread::allocate_object_slow_entry_point_offset()));
    __ br(R4);
  }
}

// Called for invoking "dynamic noSuchMethod(Invocation invocation)" function
// from the entry code of a dart function after an error in passed argument
// name or number is detected.
// Input parameters:
//  LR : return address.
//  SP : address of last argument.
//  R4: arguments descriptor array.
void StubCodeCompiler::GenerateCallClosureNoSuchMethodStub() {
  __ EnterStubFrame();

  // Load the receiver.
  __ LoadCompressedSmiFieldFromOffset(
      R2, R4, target::ArgumentsDescriptor::size_offset());
  __ add(TMP, FP, Operand(R2, LSL, target::kWordSizeLog2 - 1));
  __ LoadFromOffset(R6, TMP,
                    target::frame_layout.param_end_from_fp * target::kWordSize);

  // Load the function.
  __ LoadCompressedFieldFromOffset(TMP, R6, target::Closure::function_offset());

  __ Push(ZR);   // Result slot.
  __ Push(R6);   // Receiver.
  __ Push(TMP);  // Function
  __ Push(R4);   // Arguments descriptor.

  // Adjust arguments count.
  __ LoadCompressedSmiFieldFromOffset(
      R3, R4, target::ArgumentsDescriptor::type_args_len_offset());
  __ AddImmediate(TMP, R2, 1, kObjectBytes);  // Include the type arguments.
  __ cmp(R3, Operand(0), kObjectBytes);
  // R2 <- (R3 == 0) ? R2 : TMP + 1 (R2 : R2 + 2).
  __ csinc(R2, R2, TMP, EQ, kObjectBytes);

  // R2: Smi-tagged arguments array length.
  PushArrayOfArguments(assembler);

  const intptr_t kNumArgs = 4;
  __ CallRuntime(kNoSuchMethodFromPrologueRuntimeEntry, kNumArgs);
  // noSuchMethod on closures always throws an error, so it will never return.
  __ brk(0);
}

//  R6: function object.
//  R5: inline cache data object.
// Cannot use function object from ICData as it may be the inlined
// function and not the top-scope function.
void StubCodeCompiler::GenerateOptimizedUsageCounterIncrement() {
  Register ic_reg = R5;
  Register func_reg = R6;
  if (FLAG_precompiled_mode) {
    __ Breakpoint();
    return;
  }
  if (FLAG_trace_optimized_ic_calls) {
    __ EnterStubFrame();
    __ Push(R6);        // Preserve.
    __ Push(R5);        // Preserve.
    __ Push(ic_reg);    // Argument.
    __ Push(func_reg);  // Argument.
    __ CallRuntime(kTraceICCallRuntimeEntry, 2);
    __ Drop(2);  // Discard argument;
    __ Pop(R5);  // Restore.
    __ Pop(R6);  // Restore.
    __ LeaveStubFrame();
  }
  __ LoadFieldFromOffset(R7, func_reg, target::Function::usage_counter_offset(),
                         kFourBytes);
  __ add(R7, R7, Operand(1));
  __ StoreFieldToOffset(R7, func_reg, target::Function::usage_counter_offset(),
                        kFourBytes);
}

// Loads function into 'temp_reg'.
void StubCodeCompiler::GenerateUsageCounterIncrement(Register temp_reg) {
  if (FLAG_precompiled_mode) {
    __ Breakpoint();
    return;
  }
  if (FLAG_optimization_counter_threshold >= 0) {
    Register func_reg = temp_reg;
    ASSERT(temp_reg == R6);
    __ Comment("Increment function counter");
    __ LoadFieldFromOffset(func_reg, IC_DATA_REG,
                           target::ICData::owner_offset());
    __ LoadFieldFromOffset(
        R7, func_reg, target::Function::usage_counter_offset(), kFourBytes);
    __ AddImmediate(R7, 1);
    __ StoreFieldToOffset(R7, func_reg,
                          target::Function::usage_counter_offset(), kFourBytes);
  }
}

// Note: R5 must be preserved.
// Attempt a quick Smi operation for known operations ('kind'). The ICData
// must have been primed with a Smi/Smi check that will be used for counting
// the invocations.
static void EmitFastSmiOp(Assembler* assembler,
                          Token::Kind kind,
                          intptr_t num_args,
                          Label* not_smi_or_overflow) {
  __ Comment("Fast Smi op");
  __ ldr(R0, Address(SP, +1 * target::kWordSize));  // Left.
  __ ldr(R1, Address(SP, +0 * target::kWordSize));  // Right.
  __ orr(TMP, R0, Operand(R1));
  __ BranchIfNotSmi(TMP, not_smi_or_overflow);
  switch (kind) {
    case Token::kADD: {
      __ adds(R0, R1, Operand(R0), kObjectBytes);  // Add.
      __ b(not_smi_or_overflow, VS);               // Branch if overflow.
      break;
    }
    case Token::kLT: {
      __ CompareObjectRegisters(R0, R1);
      __ LoadObject(R0, CastHandle<Object>(TrueObject()));
      __ LoadObject(R1, CastHandle<Object>(FalseObject()));
      __ csel(R0, R0, R1, LT);
      break;
    }
    case Token::kEQ: {
      __ CompareObjectRegisters(R0, R1);
      __ LoadObject(R0, CastHandle<Object>(TrueObject()));
      __ LoadObject(R1, CastHandle<Object>(FalseObject()));
      __ csel(R0, R0, R1, EQ);
      break;
    }
    default:
      UNIMPLEMENTED();
  }

  // R5: IC data object (preserved).
  __ LoadFieldFromOffset(R6, R5, target::ICData::entries_offset());
  // R6: ic_data_array with check entries: classes and target functions.
  __ AddImmediate(R6, target::Array::data_offset() - kHeapObjectTag);
// R6: points directly to the first ic data array element.
#if defined(DEBUG)
  // Check that first entry is for Smi/Smi.
  Label error, ok;
  const intptr_t imm_smi_cid = target::ToRawSmi(kSmiCid);
  __ LoadCompressedSmiFromOffset(R1, R6, 0);
  __ CompareImmediate(R1, imm_smi_cid, kObjectBytes);
  __ b(&error, NE);
  __ LoadCompressedSmiFromOffset(R1, R6, target::kCompressedWordSize);
  __ CompareImmediate(R1, imm_smi_cid, kObjectBytes);
  __ b(&ok, EQ);
  __ Bind(&error);
  __ Stop("Incorrect IC data");
  __ Bind(&ok);
#endif
  if (FLAG_optimization_counter_threshold >= 0) {
    const intptr_t count_offset =
        target::ICData::CountIndexFor(num_args) * target::kCompressedWordSize;
    // Update counter, ignore overflow.
    __ LoadCompressedSmiFromOffset(R1, R6, count_offset);
    __ adds(R1, R1, Operand(target::ToRawSmi(1)), kObjectBytes);
    __ StoreToOffset(R1, R6, count_offset, kObjectBytes);
  }

  __ ret();
}

// Saves the offset of the target entry-point (from the Function) into R8.
//
// Must be the first code generated, since any code before will be skipped in
// the unchecked entry-point.
static void GenerateRecordEntryPoint(Assembler* assembler) {
  Label done;
  __ LoadImmediate(R8, target::Function::entry_point_offset() - kHeapObjectTag);
  __ b(&done);
  __ BindUncheckedEntryPoint();
  __ LoadImmediate(
      R8, target::Function::entry_point_offset(CodeEntryKind::kUnchecked) -
              kHeapObjectTag);
  __ Bind(&done);
}

// Generate inline cache check for 'num_args'.
//  R0: receiver (if instance call)
//  R5: ICData
//  LR: return address
// Control flow:
// - If receiver is null -> jump to IC miss.
// - If receiver is Smi -> load Smi class.
// - If receiver is not-Smi -> load receiver's class.
// - Check if 'num_args' (including receiver) match any IC data group.
// - Match found -> jump to target.
// - Match not found -> jump to IC miss.
void StubCodeCompiler::GenerateNArgsCheckInlineCacheStub(
    intptr_t num_args,
    const RuntimeEntry& handle_ic_miss,
    Token::Kind kind,
    Optimized optimized,
    CallType type,
    Exactness exactness) {
  const bool save_entry_point = kind == Token::kILLEGAL;
  if (FLAG_precompiled_mode) {
    __ Breakpoint();
    return;
  }

  if (save_entry_point) {
    GenerateRecordEntryPoint(assembler);
  }

  if (optimized == kOptimized) {
    GenerateOptimizedUsageCounterIncrement();
  } else {
    GenerateUsageCounterIncrement(/*scratch=*/R6);
  }

  ASSERT(num_args == 1 || num_args == 2);
#if defined(DEBUG)
  {
    Label ok;
    // Check that the IC data array has NumArgsTested() == num_args.
    // 'NumArgsTested' is stored in the least significant bits of 'state_bits'.
    __ LoadFromOffset(R6, R5,
                      target::ICData::state_bits_offset() - kHeapObjectTag,
                      kUnsignedFourBytes);
    ASSERT(target::ICData::NumArgsTestedShift() == 0);  // No shift needed.
    __ andi(R6, R6, Immediate(target::ICData::NumArgsTestedMask()));
    __ CompareImmediate(R6, num_args);
    __ b(&ok, EQ);
    __ Stop("Incorrect stub for IC data");
    __ Bind(&ok);
  }
#endif  // DEBUG

#if !defined(PRODUCT)
  Label stepping, done_stepping;
  if (optimized == kUnoptimized) {
    __ Comment("Check single stepping");
    __ LoadIsolate(R6);
    __ LoadFromOffset(R6, R6, target::Isolate::single_step_offset(),
                      kUnsignedByte);
    __ CompareRegisters(R6, ZR);
    __ b(&stepping, NE);
    __ Bind(&done_stepping);
  }
#endif

  Label not_smi_or_overflow;
  if (kind != Token::kILLEGAL) {
    EmitFastSmiOp(assembler, kind, num_args, &not_smi_or_overflow);
  }
  __ Bind(&not_smi_or_overflow);

  __ Comment("Extract ICData initial values and receiver cid");
  // R5: IC data object (preserved).
  __ LoadFieldFromOffset(R6, R5, target::ICData::entries_offset());
  // R6: ic_data_array with check entries: classes and target functions.
  __ AddImmediate(R6, target::Array::data_offset() - kHeapObjectTag);
  // R6: points directly to the first ic data array element.

  if (type == kInstanceCall) {
    __ LoadTaggedClassIdMayBeSmi(R3, R0);
    __ LoadFieldFromOffset(ARGS_DESC_REG, R5,
                           target::CallSiteData::arguments_descriptor_offset());
    if (num_args == 2) {
      __ LoadCompressedSmiFieldFromOffset(
          R7, ARGS_DESC_REG, target::ArgumentsDescriptor::count_offset());
      __ SmiUntag(R7);  // Untag so we can use the LSL 3 addressing mode.
      __ sub(R7, R7, Operand(2));
      // R1 <- [SP + (R1 << 3)]
      __ ldr(R1, Address(SP, R7, UXTX, Address::Scaled));
      __ LoadTaggedClassIdMayBeSmi(R1, R1);
    }
  } else {
    __ LoadFieldFromOffset(ARGS_DESC_REG, R5,
                           target::CallSiteData::arguments_descriptor_offset());
    // Get the receiver's class ID (first read number of arguments from
    // arguments descriptor array and then access the receiver from the stack).
    __ LoadCompressedSmiFieldFromOffset(
        R7, ARGS_DESC_REG, target::ArgumentsDescriptor::count_offset());
    __ SmiUntag(R7);  // Untag so we can use the LSL 3 addressing mode.
    __ sub(R7, R7, Operand(1));
    // R0 <- [SP + (R7 << 3)]
    __ ldr(R0, Address(SP, R7, UXTX, Address::Scaled));
    __ LoadTaggedClassIdMayBeSmi(R3, R0);
    if (num_args == 2) {
      __ AddImmediate(R1, R7, -1);
      // R1 <- [SP + (R1 << 3)]
      __ ldr(R1, Address(SP, R1, UXTX, Address::Scaled));
      __ LoadTaggedClassIdMayBeSmi(R1, R1);
    }
  }
  // R3: first argument class ID as Smi.
  // R1: second argument class ID as Smi.
  // R4: args descriptor

  // We unroll the generic one that is generated once more than the others.
  const bool optimize = kind == Token::kILLEGAL;

  // Loop that checks if there is an IC data match.
  Label loop, found, miss;
  __ Comment("ICData loop");

  __ Bind(&loop);
  for (int unroll = optimize ? 4 : 2; unroll >= 0; unroll--) {
    Label update;

    __ LoadCompressedSmiFromOffset(R2, R6, 0);
    __ CompareObjectRegisters(R3, R2);  // Class id match?
    if (num_args == 2) {
      __ b(&update, NE);  // Continue.
      __ LoadCompressedSmiFromOffset(R2, R6, target::kCompressedWordSize);
      __ CompareObjectRegisters(R1, R2);  // Class id match?
    }
    __ b(&found, EQ);  // Break.

    __ Bind(&update);

    const intptr_t entry_size = target::ICData::TestEntryLengthFor(
                                    num_args, exactness == kCheckExactness) *
                                target::kCompressedWordSize;
    __ AddImmediate(R6, entry_size);  // Next entry.

    __ CompareImmediate(R2, target::ToRawSmi(kIllegalCid));  // Done?
    if (unroll == 0) {
      __ b(&loop, NE);
    } else {
      __ b(&miss, EQ);
    }
  }

  __ Bind(&miss);
  __ Comment("IC miss");

  // Compute address of arguments.
  __ LoadCompressedSmiFieldFromOffset(
      R7, ARGS_DESC_REG, target::ArgumentsDescriptor::count_offset());
  __ SmiUntag(R7);  // Untag so we can use the LSL 3 addressing mode.
  __ sub(R7, R7, Operand(1));
  // R7: argument_count - 1 (untagged).
  // R7 <- SP + (R7 << 3)
  __ add(R7, SP, Operand(R7, UXTX, 3));  // R7 is Untagged.
  // R7: address of receiver.
  // Create a stub frame as we are pushing some objects on the stack before
  // calling into the runtime.
  __ EnterStubFrame();
  // Preserve IC data object and arguments descriptor array and
  // setup space on stack for result (target code object).
  __ Push(ARGS_DESC_REG);  // Preserve arguments descriptor array.
  __ Push(R5);             // Preserve IC Data.
  if (save_entry_point) {
    __ SmiTag(R8);
    __ Push(R8);
  }
  // Setup space on stack for the result (target code object).
  __ Push(ZR);
  // Push call arguments.
  for (intptr_t i = 0; i < num_args; i++) {
    __ LoadFromOffset(TMP, R7, -i * target::kWordSize);
    __ Push(TMP);
  }
  // Pass IC data object.
  __ Push(R5);
  __ CallRuntime(handle_ic_miss, num_args + 1);
  // Remove the call arguments pushed earlier, including the IC data object.
  __ Drop(num_args + 1);
  // Pop returned function object into R0.
  // Restore arguments descriptor array and IC data array.
  __ Pop(FUNCTION_REG);  // Pop returned function object into R0.
  if (save_entry_point) {
    __ Pop(R8);
    __ SmiUntag(R8);
  }
  __ Pop(R5);             // Restore IC Data.
  __ Pop(ARGS_DESC_REG);  // Restore arguments descriptor array.
  __ RestoreCodePointer();
  __ LeaveStubFrame();
  Label call_target_function;
  if (FLAG_precompiled_mode) {
    GenerateDispatcherCode(assembler, &call_target_function);
  } else {
    __ b(&call_target_function);
  }

  __ Bind(&found);
  // R6: pointer to an IC data check group.
  const intptr_t target_offset =
      target::ICData::TargetIndexFor(num_args) * target::kCompressedWordSize;
  const intptr_t count_offset =
      target::ICData::CountIndexFor(num_args) * target::kCompressedWordSize;
  const intptr_t exactness_offset =
      target::ICData::ExactnessIndexFor(num_args) * target::kCompressedWordSize;

  Label call_target_function_through_unchecked_entry;
  if (exactness == kCheckExactness) {
    Label exactness_ok;
    ASSERT(num_args == 1);
    __ LoadCompressedSmi(R1, Address(R6, exactness_offset));
    __ CompareImmediate(
        R1,
        target::ToRawSmi(
            StaticTypeExactnessState::HasExactSuperType().Encode()),
        kObjectBytes);
    __ BranchIf(LESS, &exactness_ok);
    __ BranchIf(EQUAL, &call_target_function_through_unchecked_entry);

    // Check trivial exactness.
    // Note: UntaggedICData::receivers_static_type_ is guaranteed to be not null
    // because we only emit calls to this stub when it is not null.
    __ LoadCompressed(
        R2, FieldAddress(R5, target::ICData::receivers_static_type_offset()));
    __ LoadCompressed(R2, FieldAddress(R2, target::Type::arguments_offset()));
    // R1 contains an offset to type arguments in words as a smi,
    // hence TIMES_4. R0 is guaranteed to be non-smi because it is expected
    // to have type arguments.
#if defined(DART_COMPRESSED_POINTERS)
    __ sxtw(R1, R1);
#endif
    __ LoadIndexedPayload(R3, R0, 0, R1, TIMES_COMPRESSED_HALF_WORD_SIZE,
                          kObjectBytes);
    __ CompareObjectRegisters(R2, R3);
    __ BranchIf(EQUAL, &call_target_function_through_unchecked_entry);

    // Update exactness state (not-exact anymore).
    __ LoadImmediate(
        R1, target::ToRawSmi(StaticTypeExactnessState::NotExact().Encode()));
    __ StoreToOffset(R1, R6, exactness_offset, kObjectBytes);
    __ Bind(&exactness_ok);
  }
  __ LoadCompressedFromOffset(FUNCTION_REG, R6, target_offset);

  if (FLAG_optimization_counter_threshold >= 0) {
    __ Comment("Update caller's counter");
    __ LoadCompressedSmiFromOffset(R1, R6, count_offset);
    // Ignore overflow.
    __ add(R1, R1, Operand(target::ToRawSmi(1)), kObjectBytes);
    __ StoreToOffset(R1, R6, count_offset, kObjectBytes);
  }

  __ Comment("Call target");
  __ Bind(&call_target_function);
  // R0: target function.
  __ LoadCompressedFieldFromOffset(CODE_REG, FUNCTION_REG,
                                   target::Function::code_offset());
  if (save_entry_point) {
    __ add(R2, FUNCTION_REG, Operand(R8));
    __ ldr(R2, Address(R2, 0));
  } else {
    __ LoadFieldFromOffset(R2, FUNCTION_REG,
                           target::Function::entry_point_offset());
  }
  __ br(R2);

  if (exactness == kCheckExactness) {
    __ Bind(&call_target_function_through_unchecked_entry);
    if (FLAG_optimization_counter_threshold >= 0) {
      __ Comment("Update ICData counter");
      __ LoadCompressedSmiFromOffset(R1, R6, count_offset);
      // Ignore overflow.
      __ add(R1, R1, Operand(target::ToRawSmi(1)), kObjectBytes);
      __ StoreToOffset(R1, R6, count_offset, kObjectBytes);
    }
    __ Comment("Call target (via unchecked entry point)");
    __ LoadCompressedFromOffset(FUNCTION_REG, R6, target_offset);
    __ LoadCompressedFieldFromOffset(CODE_REG, FUNCTION_REG,
                                     target::Function::code_offset());
    __ LoadFieldFromOffset(
        R2, FUNCTION_REG,
        target::Function::entry_point_offset(CodeEntryKind::kUnchecked));
    __ br(R2);
  }

#if !defined(PRODUCT)
  if (optimized == kUnoptimized) {
    __ Bind(&stepping);
    __ EnterStubFrame();
    if (type == kInstanceCall) {
      __ Push(R0);  // Preserve receiver.
    }
    if (save_entry_point) {
      __ SmiTag(R8);
      __ Push(R8);
    }
    __ Push(R5);  // Preserve IC data.
    __ CallRuntime(kSingleStepHandlerRuntimeEntry, 0);
    __ Pop(R5);
    if (save_entry_point) {
      __ Pop(R8);
      __ SmiUntag(R8);
    }
    if (type == kInstanceCall) {
      __ Pop(R0);
    }
    __ RestoreCodePointer();
    __ LeaveStubFrame();
    __ b(&done_stepping);
  }
#endif
}

// R0: receiver
// R5: ICData
// LR: return address
void StubCodeCompiler::GenerateOneArgCheckInlineCacheStub() {
  GenerateNArgsCheckInlineCacheStub(
      1, kInlineCacheMissHandlerOneArgRuntimeEntry, Token::kILLEGAL,
      kUnoptimized, kInstanceCall, kIgnoreExactness);
}

// R0: receiver
// R5: ICData
// LR: return address
void StubCodeCompiler::GenerateOneArgCheckInlineCacheWithExactnessCheckStub() {
  GenerateNArgsCheckInlineCacheStub(
      1, kInlineCacheMissHandlerOneArgRuntimeEntry, Token::kILLEGAL,
      kUnoptimized, kInstanceCall, kCheckExactness);
}

// R0: receiver
// R5: ICData
// LR: return address
void StubCodeCompiler::GenerateTwoArgsCheckInlineCacheStub() {
  GenerateNArgsCheckInlineCacheStub(
      2, kInlineCacheMissHandlerTwoArgsRuntimeEntry, Token::kILLEGAL,
      kUnoptimized, kInstanceCall, kIgnoreExactness);
}

// R0: receiver
// R5: ICData
// LR: return address
void StubCodeCompiler::GenerateSmiAddInlineCacheStub() {
  GenerateNArgsCheckInlineCacheStub(
      2, kInlineCacheMissHandlerTwoArgsRuntimeEntry, Token::kADD, kUnoptimized,
      kInstanceCall, kIgnoreExactness);
}

// R0: receiver
// R5: ICData
// LR: return address
void StubCodeCompiler::GenerateSmiLessInlineCacheStub() {
  GenerateNArgsCheckInlineCacheStub(
      2, kInlineCacheMissHandlerTwoArgsRuntimeEntry, Token::kLT, kUnoptimized,
      kInstanceCall, kIgnoreExactness);
}

// R0: receiver
// R5: ICData
// LR: return address
void StubCodeCompiler::GenerateSmiEqualInlineCacheStub() {
  GenerateNArgsCheckInlineCacheStub(
      2, kInlineCacheMissHandlerTwoArgsRuntimeEntry, Token::kEQ, kUnoptimized,
      kInstanceCall, kIgnoreExactness);
}

// R0: receiver
// R5: ICData
// R6: Function
// LR: return address
void StubCodeCompiler::GenerateOneArgOptimizedCheckInlineCacheStub() {
  GenerateNArgsCheckInlineCacheStub(
      1, kInlineCacheMissHandlerOneArgRuntimeEntry, Token::kILLEGAL, kOptimized,
      kInstanceCall, kIgnoreExactness);
}

// R0: receiver
// R5: ICData
// R6: Function
// LR: return address
void StubCodeCompiler::
    GenerateOneArgOptimizedCheckInlineCacheWithExactnessCheckStub() {
  GenerateNArgsCheckInlineCacheStub(
      1, kInlineCacheMissHandlerOneArgRuntimeEntry, Token::kILLEGAL, kOptimized,
      kInstanceCall, kCheckExactness);
}

// R0: receiver
// R5: ICData
// R6: Function
// LR: return address
void StubCodeCompiler::GenerateTwoArgsOptimizedCheckInlineCacheStub() {
  GenerateNArgsCheckInlineCacheStub(
      2, kInlineCacheMissHandlerTwoArgsRuntimeEntry, Token::kILLEGAL,
      kOptimized, kInstanceCall, kIgnoreExactness);
}

// R5: ICData
// LR: return address
void StubCodeCompiler::GenerateZeroArgsUnoptimizedStaticCallStub() {
  GenerateRecordEntryPoint(assembler);
  GenerateUsageCounterIncrement(/* scratch */ R6);
#if defined(DEBUG)
  {
    Label ok;
    // Check that the IC data array has NumArgsTested() == 0.
    // 'NumArgsTested' is stored in the least significant bits of 'state_bits'.
    __ LoadFromOffset(R6, R5,
                      target::ICData::state_bits_offset() - kHeapObjectTag,
                      kUnsignedFourBytes);
    ASSERT(target::ICData::NumArgsTestedShift() == 0);  // No shift needed.
    __ andi(R6, R6, Immediate(target::ICData::NumArgsTestedMask()));
    __ CompareImmediate(R6, 0);
    __ b(&ok, EQ);
    __ Stop("Incorrect IC data for unoptimized static call");
    __ Bind(&ok);
  }
#endif  // DEBUG

  // Check single stepping.
#if !defined(PRODUCT)
  Label stepping, done_stepping;
  __ LoadIsolate(R6);
  __ LoadFromOffset(R6, R6, target::Isolate::single_step_offset(),
                    kUnsignedByte);
  __ CompareImmediate(R6, 0);
  __ b(&stepping, NE);
  __ Bind(&done_stepping);
#endif

  // R5: IC data object (preserved).
  __ LoadFieldFromOffset(R6, R5, target::ICData::entries_offset());
  // R6: ic_data_array with entries: target functions and count.
  __ AddImmediate(R6, target::Array::data_offset() - kHeapObjectTag);
  // R6: points directly to the first ic data array element.
  const intptr_t target_offset =
      target::ICData::TargetIndexFor(0) * target::kCompressedWordSize;
  const intptr_t count_offset =
      target::ICData::CountIndexFor(0) * target::kCompressedWordSize;

  if (FLAG_optimization_counter_threshold >= 0) {
    // Increment count for this call, ignore overflow.
    __ LoadCompressedSmiFromOffset(R1, R6, count_offset);
    __ adds(R1, R1, Operand(target::ToRawSmi(1)), kObjectBytes);
    __ StoreToOffset(R1, R6, count_offset, kObjectBytes);
  }

  // Load arguments descriptor into R4.
  __ LoadFieldFromOffset(ARGS_DESC_REG, R5,
                         target::CallSiteData::arguments_descriptor_offset());

  // Get function and call it, if possible.
  __ LoadCompressedFromOffset(FUNCTION_REG, R6, target_offset);
  __ LoadCompressedFieldFromOffset(CODE_REG, FUNCTION_REG,
                                   target::Function::code_offset());
  __ add(R2, FUNCTION_REG, Operand(R8));
  __ ldr(R2, Address(R2, 0));
  __ br(R2);

#if !defined(PRODUCT)
  __ Bind(&stepping);
  __ EnterStubFrame();
  __ Push(R5);  // Preserve IC data.
  __ SmiTag(R8);
  __ Push(R8);
  __ CallRuntime(kSingleStepHandlerRuntimeEntry, 0);
  __ Pop(R8);
  __ SmiUntag(R8);
  __ Pop(R5);
  __ RestoreCodePointer();
  __ LeaveStubFrame();
  __ b(&done_stepping);
#endif
}

// R5: ICData
// LR: return address
void StubCodeCompiler::GenerateOneArgUnoptimizedStaticCallStub() {
  GenerateUsageCounterIncrement(/* scratch */ R6);
  GenerateNArgsCheckInlineCacheStub(1, kStaticCallMissHandlerOneArgRuntimeEntry,
                                    Token::kILLEGAL, kUnoptimized, kStaticCall,
                                    kIgnoreExactness);
}

// R5: ICData
// LR: return address
void StubCodeCompiler::GenerateTwoArgsUnoptimizedStaticCallStub() {
  GenerateUsageCounterIncrement(/* scratch */ R6);
  GenerateNArgsCheckInlineCacheStub(
      2, kStaticCallMissHandlerTwoArgsRuntimeEntry, Token::kILLEGAL,
      kUnoptimized, kStaticCall, kIgnoreExactness);
}

// Stub for compiling a function and jumping to the compiled code.
// ARGS_DESC_REG: Arguments descriptor.
// FUNCTION_REG: Function.
void StubCodeCompiler::GenerateLazyCompileStub() {
  // Preserve arg desc.
  __ EnterStubFrame();
  __ Push(ARGS_DESC_REG);  // Save arg. desc.
  __ Push(FUNCTION_REG);   // Pass function.
  __ CallRuntime(kCompileFunctionRuntimeEntry, 1);
  __ Pop(FUNCTION_REG);   // Restore function.
  __ Pop(ARGS_DESC_REG);  // Restore arg desc.
  __ LeaveStubFrame();

  __ LoadCompressedFieldFromOffset(CODE_REG, FUNCTION_REG,
                                   target::Function::code_offset());
  __ LoadFieldFromOffset(R2, FUNCTION_REG,
                         target::Function::entry_point_offset());
  __ br(R2);
}

// R5: Contains an ICData.
void StubCodeCompiler::GenerateICCallBreakpointStub() {
#if defined(PRODUCT)
  __ Stop("No debugging in PRODUCT mode");
#else
  __ EnterStubFrame();
  __ Push(R0);  // Preserve receiver.
  __ Push(R5);  // Preserve IC data.
  __ Push(ZR);  // Space for result.
  __ CallRuntime(kBreakpointRuntimeHandlerRuntimeEntry, 0);
  __ Pop(CODE_REG);  // Original stub.
  __ Pop(R5);        // Restore IC data.
  __ Pop(R0);        // Restore receiver.
  __ LeaveStubFrame();
  __ LoadFieldFromOffset(TMP, CODE_REG, target::Code::entry_point_offset());
  __ br(TMP);
#endif  // defined(PRODUCT)
}

void StubCodeCompiler::GenerateUnoptStaticCallBreakpointStub() {
#if defined(PRODUCT)
  __ Stop("No debugging in PRODUCT mode");
#else
  __ EnterStubFrame();
  __ Push(R5);  // Preserve IC data.
  __ Push(ZR);  // Space for result.
  __ CallRuntime(kBreakpointRuntimeHandlerRuntimeEntry, 0);
  __ Pop(CODE_REG);  // Original stub.
  __ Pop(R5);        // Restore IC data.
  __ LeaveStubFrame();
  __ LoadFieldFromOffset(TMP, CODE_REG, target::Code::entry_point_offset());
  __ br(TMP);
#endif  // defined(PRODUCT)
}

void StubCodeCompiler::GenerateRuntimeCallBreakpointStub() {
#if defined(PRODUCT)
  __ Stop("No debugging in PRODUCT mode");
#else
  __ EnterStubFrame();
  __ Push(ZR);  // Space for result.
  __ CallRuntime(kBreakpointRuntimeHandlerRuntimeEntry, 0);
  __ Pop(CODE_REG);
  __ LeaveStubFrame();
  __ LoadFieldFromOffset(R0, CODE_REG, target::Code::entry_point_offset());
  __ br(R0);
#endif  // defined(PRODUCT)
}

// Called only from unoptimized code. All relevant registers have been saved.
void StubCodeCompiler::GenerateDebugStepCheckStub() {
#if defined(PRODUCT)
  __ Stop("No debugging in PRODUCT mode");
#else
  // Check single stepping.
  Label stepping, done_stepping;
  __ LoadIsolate(R1);
  __ LoadFromOffset(R1, R1, target::Isolate::single_step_offset(),
                    kUnsignedByte);
  __ CompareImmediate(R1, 0);
  __ b(&stepping, NE);
  __ Bind(&done_stepping);
  __ ret();

  __ Bind(&stepping);
  __ EnterStubFrame();
  __ CallRuntime(kSingleStepHandlerRuntimeEntry, 0);
  __ LeaveStubFrame();
  __ b(&done_stepping);
#endif  // defined(PRODUCT)
}

// Used to check class and type arguments. Arguments passed in registers:
//
// Inputs (all preserved, mostly from TypeTestABI struct):
//   - kSubtypeTestCacheReg: UntaggedSubtypeTestCache
//   - kInstanceReg: instance to test against.
//   - kDstTypeReg: destination type (for n>=7).
//   - kInstantiatorTypeArgumentsReg: instantiator type arguments (for n>=3).
//   - kFunctionTypeArgumentsReg: function type arguments (for n>=4).
//   - LR: return address.
//
// Outputs (from TypeTestABI struct):
//   - kSubtypeTestCacheResultReg: the cached result, or null if not found.
void StubCodeCompiler::GenerateSubtypeNTestCacheStub(Assembler* assembler,
                                                     int n) {
  ASSERT(n >= 1);
  ASSERT(n <= SubtypeTestCache::kMaxInputs);
  // If we need the parent function type arguments for a closure, we also need
  // the delayed type arguments, so this case will never happen.
  ASSERT(n != 5);

  // We could initialize kSubtypeTestCacheResultReg with null and use that as
  // the null register up until exit, which means we'd just need to return
  // without setting it in the not_found case.
  //
  // However, that would mean the expense of keeping another register live
  // across the loop to hold the cache entry address, and the not_found case
  // means we're going to runtime, so optimize for the found case instead.
  //
  // Thus, we use it to store the current cache entry, since it's distinct from
  // all the preserved input registers and the scratch register, and the last
  // use of the current cache entry is to set kSubtypeTestCacheResultReg.
  const Register kCacheArrayReg = TypeTestABI::kSubtypeTestCacheResultReg;

  Label not_found;
  GenerateSubtypeTestCacheSearch(
      assembler, n, NULL_REG, kCacheArrayReg,
      STCInternalRegs::kInstanceCidOrSignatureReg,
      STCInternalRegs::kInstanceInstantiatorTypeArgumentsReg,
      STCInternalRegs::kInstanceParentFunctionTypeArgumentsReg,
      STCInternalRegs::kInstanceDelayedFunctionTypeArgumentsReg,
      STCInternalRegs::kCacheEntriesEndReg,
      STCInternalRegs::kCacheContentsSizeReg,
      STCInternalRegs::kProbeDistanceReg,
      [&](Assembler* assembler, int n) {
        __ LoadCompressed(
            TypeTestABI::kSubtypeTestCacheResultReg,
            Address(kCacheArrayReg, target::kCompressedWordSize *
                                        target::SubtypeTestCache::kTestResult));
        __ Ret();
      },
      [&](Assembler* assembler, int n) {
        __ MoveRegister(TypeTestABI::kSubtypeTestCacheResultReg, NULL_REG);
        __ Ret();
      });
}

void StubCodeCompiler::GenerateGetCStackPointerStub() {
  __ mov(R0, CSP);
  __ ret();
}

// Jump to a frame on the call stack.
// LR: return address.
// R0: program_counter.
// R1: stack_pointer.
// R2: frame_pointer.
// R3: thread.
// Does not return.
//
// Notice: We need to keep this in sync with `Simulator::JumpToFrame()`.
void StubCodeCompiler::GenerateJumpToFrameStub() {
  ASSERT(kExceptionObjectReg == R0);
  ASSERT(kStackTraceObjectReg == R1);
  __ set_lr_state(compiler::LRState::Clobbered());
  __ mov(CALLEE_SAVED_TEMP, R0);  // Program counter.
  __ mov(SP, R1);                 // Stack pointer.
  __ mov(FP, R2);                 // Frame_pointer.
  __ mov(THR, R3);
  __ SetupCSPFromThread(THR);
#if defined(DART_TARGET_OS_FUCHSIA)
  // We need to restore the shadow call stack pointer like longjmp would,
  // effectively popping all the return addresses between the Dart exit frame
  // and Exceptions::JumpToFrame, otherwise the shadow call stack might
  // eventually overflow.
  __ ldr(R18, Address(THR, target::Thread::saved_shadow_call_stack_offset()));
#elif defined(USING_SHADOW_CALL_STACK)
#error Unimplemented
#endif
  Label exit_through_non_ffi;
  Register tmp1 = R0, tmp2 = R1;
  // Check if we exited generated from FFI. If so do transition - this is needed
  // because normally runtime calls transition back to generated via destructor
  // of TransitionGeneratedToVM/Native that is part of runtime boilerplate
  // code (see DEFINE_RUNTIME_ENTRY_IMPL in runtime_entry.h). Ffi calls don't
  // have this boilerplate, don't have this stack resource, have to transition
  // explicitly.
  __ LoadFromOffset(tmp1, THR,
                    compiler::target::Thread::exit_through_ffi_offset());
  __ LoadImmediate(tmp2, target::Thread::exit_through_ffi());
  __ cmp(tmp1, Operand(tmp2));
  __ b(&exit_through_non_ffi, NE);
  __ TransitionNativeToGenerated(tmp1, /*leave_safepoint=*/true,
                                 /*ignore_unwind_in_progress=*/true);
  __ Bind(&exit_through_non_ffi);

  // Refresh pinned registers (write barrier mask, null, dispatch table, etc).
  __ RestorePinnedRegisters();
  // Set the tag.
  __ LoadImmediate(R2, VMTag::kDartTagId);
  __ StoreToOffset(R2, THR, target::Thread::vm_tag_offset());
  // Clear top exit frame.
  __ StoreToOffset(ZR, THR, target::Thread::top_exit_frame_info_offset());
  // Restore the pool pointer.
  __ RestoreCodePointer();
  if (FLAG_precompiled_mode) {
    __ SetupGlobalPoolAndDispatchTable();
  } else {
    __ LoadPoolPointer();
  }
  __ ret(CALLEE_SAVED_TEMP);  // Jump to continuation point.
}

// Run an exception handler.  Execution comes from JumpToFrame
// stub or from the simulator.
//
// The arguments are stored in the Thread object.
// Does not return.
void StubCodeCompiler::GenerateRunExceptionHandlerStub() {
  WRITES_RETURN_ADDRESS_TO_LR(
      __ LoadFromOffset(LR, THR, target::Thread::resume_pc_offset()));

  word offset_from_thread = 0;
  bool ok = target::CanLoadFromThread(NullObject(), &offset_from_thread);
  ASSERT(ok);
  __ LoadFromOffset(R2, THR, offset_from_thread);

  // Exception object.
  __ LoadFromOffset(R0, THR, target::Thread::active_exception_offset());
  __ StoreToOffset(R2, THR, target::Thread::active_exception_offset());

  // StackTrace object.
  __ LoadFromOffset(R1, THR, target::Thread::active_stacktrace_offset());
  __ StoreToOffset(R2, THR, target::Thread::active_stacktrace_offset());

  __ ret();  // Jump to the exception handler code.
}

// Deoptimize a frame on the call stack before rewinding.
// The arguments are stored in the Thread object.
// No result.
void StubCodeCompiler::GenerateDeoptForRewindStub() {
  // Push zap value instead of CODE_REG.
  __ LoadImmediate(TMP, kZapCodeReg);
  __ Push(TMP);

  // Load the deopt pc into LR.
  WRITES_RETURN_ADDRESS_TO_LR(
      __ LoadFromOffset(LR, THR, target::Thread::resume_pc_offset()));
  GenerateDeoptimizationSequence(assembler, kEagerDeopt);

  // After we have deoptimized, jump to the correct frame.
  __ EnterStubFrame();
  __ CallRuntime(kRewindPostDeoptRuntimeEntry, 0);
  __ LeaveStubFrame();
  __ brk(0);
}

// Calls to the runtime to optimize the given function.
// R6: function to be re-optimized.
// ARGS_DESC_REG: argument descriptor (preserved).
void StubCodeCompiler::GenerateOptimizeFunctionStub() {
  __ LoadFromOffset(CODE_REG, THR, target::Thread::optimize_stub_offset());
  __ EnterStubFrame();
  __ Push(ARGS_DESC_REG);
  // Setup space on stack for the return value.
  __ Push(ZR);
  __ Push(R6);
  __ CallRuntime(kOptimizeInvokedFunctionRuntimeEntry, 1);
  __ Pop(R0);             // Discard argument.
  __ Pop(FUNCTION_REG);   // Get Function object
  __ Pop(ARGS_DESC_REG);  // Restore argument descriptor.
  __ LoadCompressedFieldFromOffset(CODE_REG, FUNCTION_REG,
                                   target::Function::code_offset());
  __ LoadFieldFromOffset(R1, FUNCTION_REG,
                         target::Function::entry_point_offset());
  __ LeaveStubFrame();
  __ br(R1);
  __ brk(0);
}

// Does identical check (object references are equal or not equal) with special
// checks for boxed numbers and returns with ZF set iff left and right are
// identical.
static void GenerateIdenticalWithNumberCheckStub(Assembler* assembler,
                                                 const Register left,
                                                 const Register right) {
  Label reference_compare, check_mint;
  // If any of the arguments is Smi do reference compare.
  // Note: A Mint cannot contain a value that would fit in Smi.
  __ BranchIfSmi(left, &reference_compare);
  __ BranchIfSmi(right, &reference_compare);

  // Value compare for two doubles.
  __ CompareClassId(left, kDoubleCid);
  __ b(&check_mint, NE);
  __ CompareClassId(right, kDoubleCid);
  __ b(&reference_compare, NE);  // Do not branch directly to ret! See below.

  // Double values bitwise compare.
  __ LoadFieldFromOffset(left, left, target::Double::value_offset());
  __ LoadFieldFromOffset(right, right, target::Double::value_offset());
  __ CompareRegisters(left, right);
  __ ret();

  __ Bind(&check_mint);
  __ CompareClassId(left, kMintCid);
  __ b(&reference_compare, NE);
  __ CompareClassId(right, kMintCid);
  __ b(&reference_compare, NE);  // Do not branch directly to ret! See below.
  __ LoadFieldFromOffset(left, left, target::Mint::value_offset());
  __ LoadFieldFromOffset(right, right, target::Mint::value_offset());
  __ CompareRegisters(left, right);
  __ ret();

  __ Bind(&reference_compare);
  __ CompareObjectRegisters(left, right);
  // None of the branches above go directly here to avoid generating a
  // conditional branch to a ret instruction.
  // This is an attempt to workaround a possible CPU on Exynos 2100 SoC.
  // See https://github.com/flutter/flutter/issues/88261
  __ ret();
}

// Called only from unoptimized code. All relevant registers have been saved.
// LR: return address.
// SP + 4: left operand.
// SP + 0: right operand.
// Return Zero condition flag set if equal.
void StubCodeCompiler::GenerateUnoptimizedIdenticalWithNumberCheckStub() {
#if !defined(PRODUCT)
  // Check single stepping.
  Label stepping, done_stepping;
  __ LoadIsolate(R1);
  __ LoadFromOffset(R1, R1, target::Isolate::single_step_offset(),
                    kUnsignedByte);
  __ CompareImmediate(R1, 0);
  __ b(&stepping, NE);
  __ Bind(&done_stepping);
#endif

  const Register left = R1;
  const Register right = R0;
  __ LoadFromOffset(left, SP, 1 * target::kWordSize);
  __ LoadFromOffset(right, SP, 0 * target::kWordSize);
  GenerateIdenticalWithNumberCheckStub(assembler, left, right);

#if !defined(PRODUCT)
  __ Bind(&stepping);
  __ EnterStubFrame();
  __ CallRuntime(kSingleStepHandlerRuntimeEntry, 0);
  __ RestoreCodePointer();
  __ LeaveStubFrame();
  __ b(&done_stepping);
#endif
}

// Called from optimized code only.
// LR: return address.
// SP + 4: left operand.
// SP + 0: right operand.
// Return Zero condition flag set if equal.
void StubCodeCompiler::GenerateOptimizedIdenticalWithNumberCheckStub() {
  const Register left = R1;
  const Register right = R0;
  __ LoadFromOffset(left, SP, 1 * target::kWordSize);
  __ LoadFromOffset(right, SP, 0 * target::kWordSize);
  GenerateIdenticalWithNumberCheckStub(assembler, left, right);
}

// Called from megamorphic call sites.
//  R0: receiver (passed to target)
//  IC_DATA_REG: MegamorphicCache (preserved)
// Passed to target:
//  FUNCTION_REG: target function
//  CODE_REG: target Code
//  ARGS_DESC_REG: arguments descriptor
void StubCodeCompiler::GenerateMegamorphicCallStub() {
  // Jump if receiver is a smi.
  Label smi_case;
  __ BranchIfSmi(R0, &smi_case);

  // Loads the cid of the object.
  __ LoadClassId(R8, R0);

  Label cid_loaded;
  __ Bind(&cid_loaded);
  __ ldr(R2,
         FieldAddress(IC_DATA_REG, target::MegamorphicCache::buckets_offset()));
  __ ldr(R1,
         FieldAddress(IC_DATA_REG, target::MegamorphicCache::mask_offset()));
  // R2: cache buckets array.
  // R1: mask as a smi.

  // Make the cid into a smi.
  __ SmiTag(R8);
  // R8: class ID of the receiver (smi).

  // Compute the table index.
  ASSERT(target::MegamorphicCache::kSpreadFactor == 7);
  // Use lsl and sub to multiply with 7 == 8 - 1.
  __ LslImmediate(R3, R8, 3);
  __ sub(R3, R3, Operand(R8));
  // R3: probe.
  Label loop;
  __ Bind(&loop);
  __ and_(R3, R3, Operand(R1));

  const intptr_t base = target::Array::data_offset();
  // R3 is smi tagged, but table entries are 16 bytes, so LSL 3.
  __ add(TMP, R2, Operand(R3, LSL, kCompressedWordSizeLog2));
  __ LoadCompressedSmiFieldFromOffset(R6, TMP, base);
  Label probe_failed;
  __ CompareObjectRegisters(R6, R8);
  __ b(&probe_failed, NE);

  Label load_target;
  __ Bind(&load_target);
  // Call the target found in the cache.  For a class id match, this is a
  // proper target for the given name and arguments descriptor.  If the
  // illegal class id was found, the target is a cache miss handler that can
  // be invoked as a normal Dart function.
  __ LoadCompressed(FUNCTION_REG,
                    FieldAddress(TMP, base + target::kCompressedWordSize));
  __ ldr(R1,
         FieldAddress(FUNCTION_REG, target::Function::entry_point_offset()));
  __ ldr(ARGS_DESC_REG,
         FieldAddress(IC_DATA_REG,
                      target::CallSiteData::arguments_descriptor_offset()));
  if (!FLAG_precompiled_mode) {
    __ LoadCompressed(
        CODE_REG, FieldAddress(FUNCTION_REG, target::Function::code_offset()));
  }
  __ br(R1);

  // Probe failed, check if it is a miss.
  __ Bind(&probe_failed);
  ASSERT(kIllegalCid == 0);
  __ tst(R6, Operand(R6), kObjectBytes);
  Label miss;
  __ b(&miss, EQ);  // branch if miss.

  // Try next extry in the table.
  __ AddImmediate(R3, target::ToRawSmi(1));
  __ b(&loop);

  // Load cid for the Smi case.
  __ Bind(&smi_case);
  __ LoadImmediate(R8, kSmiCid);
  __ b(&cid_loaded);

  __ Bind(&miss);
  GenerateSwitchableCallMissStub();
}

// Input:
//   R0 - receiver
//   IC_DATA_REG - icdata
void StubCodeCompiler::GenerateICCallThroughCodeStub() {
  Label loop, found, miss;
  __ ldr(R8, FieldAddress(IC_DATA_REG, target::ICData::entries_offset()));
  __ ldr(ARGS_DESC_REG,
         FieldAddress(IC_DATA_REG,
                      target::CallSiteData::arguments_descriptor_offset()));
  __ AddImmediate(R8, target::Array::data_offset() - kHeapObjectTag);
  // R8: first IC entry
  __ LoadTaggedClassIdMayBeSmi(R1, R0);
  // R1: receiver cid as Smi

  __ Bind(&loop);
  __ LoadCompressedSmi(R2, Address(R8, 0));
  __ cmp(R1, Operand(R2), kObjectBytes);
  __ b(&found, EQ);
  __ CompareImmediate(R2, target::ToRawSmi(kIllegalCid), kObjectBytes);
  __ b(&miss, EQ);

  const intptr_t entry_length =
      target::ICData::TestEntryLengthFor(1, /*tracking_exactness=*/false) *
      target::kCompressedWordSize;
  __ AddImmediate(R8, entry_length);  // Next entry.
  __ b(&loop);

  __ Bind(&found);
  if (FLAG_precompiled_mode) {
    const intptr_t entry_offset =
        target::ICData::EntryPointIndexFor(1) * target::kCompressedWordSize;
    __ LoadCompressed(R1, Address(R8, entry_offset));
    __ ldr(R1, FieldAddress(R1, target::Function::entry_point_offset()));
  } else {
    const intptr_t code_offset =
        target::ICData::CodeIndexFor(1) * target::kCompressedWordSize;
    __ LoadCompressed(CODE_REG, Address(R8, code_offset));
    __ ldr(R1, FieldAddress(CODE_REG, target::Code::entry_point_offset()));
  }
  __ br(R1);

  __ Bind(&miss);
  __ ldr(R1, Address(THR, target::Thread::switchable_call_miss_entry_offset()));
  __ br(R1);
}

// Implement the monomorphic entry check for call-sites where the receiver
// might be a Smi.
//
//   R0: receiver
//   R5: MonomorphicSmiableCall object
//
//   R1: clobbered
void StubCodeCompiler::GenerateMonomorphicSmiableCheckStub() {
  Label miss;
  __ LoadClassIdMayBeSmi(IP0, R0);

  // Note: this stub is only used in AOT mode, hence the direct (bare) call.
  __ LoadField(
      IP1,
      FieldAddress(R5, target::MonomorphicSmiableCall::expected_cid_offset()));
  __ LoadField(
      R1,
      FieldAddress(R5, target::MonomorphicSmiableCall::entrypoint_offset()));
  __ cmp(IP0, Operand(IP1));
  __ b(&miss, NE);
  __ br(R1);

  __ Bind(&miss);
  __ ldr(IP0,
         Address(THR, target::Thread::switchable_call_miss_entry_offset()));
  __ br(IP0);
}

// Called from switchable IC calls.
//  R0: receiver
void StubCodeCompiler::GenerateSwitchableCallMissStub() {
  __ ldr(CODE_REG,
         Address(THR, target::Thread::switchable_call_miss_stub_offset()));
  __ EnterStubFrame();
  __ Push(R0);  // Preserve receiver.

  __ Push(ZR);  // Result slot.
  __ Push(ZR);  // Arg0: stub out.
  __ Push(R0);  // Arg1: Receiver
  __ CallRuntime(kSwitchableCallMissRuntimeEntry, 2);
  __ Drop(1);
  __ Pop(CODE_REG);  // result = stub
  __ Pop(R5);        // result = IC

  __ Pop(R0);  // Restore receiver.
  __ LeaveStubFrame();

  __ ldr(R1, FieldAddress(CODE_REG, target::Code::entry_point_offset(
                                        CodeEntryKind::kNormal)));
  __ br(R1);
}

// Called from switchable IC calls.
//  R0: receiver
//  R5: SingleTargetCache
// Passed to target:
//  CODE_REG: target Code object
void StubCodeCompiler::GenerateSingleTargetCallStub() {
  Label miss;
  __ LoadClassIdMayBeSmi(R1, R0);
  __ ldr(R2, FieldAddress(R5, target::SingleTargetCache::lower_limit_offset()),
         kUnsignedTwoBytes);
  __ ldr(R3, FieldAddress(R5, target::SingleTargetCache::upper_limit_offset()),
         kUnsignedTwoBytes);

  __ cmp(R1, Operand(R2));
  __ b(&miss, LT);
  __ cmp(R1, Operand(R3));
  __ b(&miss, GT);

  __ ldr(R1, FieldAddress(R5, target::SingleTargetCache::entry_point_offset()));
  __ ldr(CODE_REG,
         FieldAddress(R5, target::SingleTargetCache::target_offset()));
  __ br(R1);

  __ Bind(&miss);
  __ EnterStubFrame();
  __ Push(R0);  // Preserve receiver.

  __ Push(ZR);  // Result slot.
  __ Push(ZR);  // Arg0: Stub out.
  __ Push(R0);  // Arg1: Receiver
  __ CallRuntime(kSwitchableCallMissRuntimeEntry, 2);
  __ Drop(1);
  __ Pop(CODE_REG);  // result = stub
  __ Pop(R5);        // result = IC

  __ Pop(R0);  // Restore receiver.
  __ LeaveStubFrame();

  __ ldr(R1, FieldAddress(CODE_REG, target::Code::entry_point_offset(
                                        CodeEntryKind::kMonomorphic)));
  __ br(R1);
}

static int GetScaleFactor(intptr_t size) {
  switch (size) {
    case 1:
      return 0;
    case 2:
      return 1;
    case 4:
      return 2;
    case 8:
      return 3;
    case 16:
      return 4;
  }
  UNREACHABLE();
  return -1;
}

void StubCodeCompiler::GenerateAllocateTypedDataArrayStub(intptr_t cid) {
  const intptr_t element_size = TypedDataElementSizeInBytes(cid);
  const intptr_t max_len = TypedDataMaxNewSpaceElements(cid);
  const intptr_t scale_shift = GetScaleFactor(element_size);

  COMPILE_ASSERT(AllocateTypedDataArrayABI::kLengthReg == R4);
  COMPILE_ASSERT(AllocateTypedDataArrayABI::kResultReg == R0);

  if (!FLAG_use_slow_path && FLAG_inline_alloc) {
    Label call_runtime;
    NOT_IN_PRODUCT(__ MaybeTraceAllocation(cid, &call_runtime, R2));
    __ mov(R2, AllocateTypedDataArrayABI::kLengthReg);
    /* Check that length is a positive Smi. */
    /* R2: requested array length argument. */
    __ BranchIfNotSmi(R2, &call_runtime);
    __ SmiUntag(R2);
    /* Check for length >= 0 && length <= max_len. */
    /* R2: untagged array length. */
    __ CompareImmediate(R2, max_len, kObjectBytes);
    __ b(&call_runtime, HI);
    __ LslImmediate(R2, R2, scale_shift);
    const intptr_t fixed_size_plus_alignment_padding =
        target::TypedData::HeaderSize() +
        target::ObjectAlignment::kObjectAlignment - 1;
    __ AddImmediate(R2, fixed_size_plus_alignment_padding);
    __ andi(R2, R2,
            Immediate(~(target::ObjectAlignment::kObjectAlignment - 1)));
    __ ldr(R0, Address(THR, target::Thread::top_offset()));

    /* R2: allocation size. */
    __ adds(R1, R0, Operand(R2));
    __ b(&call_runtime, CS); /* Fail on unsigned overflow. */

    /* Check if the allocation fits into the remaining space. */
    /* R0: potential new object start. */
    /* R1: potential next object start. */
    /* R2: allocation size. */
    __ ldr(R6, Address(THR, target::Thread::end_offset()));
    __ cmp(R1, Operand(R6));
    __ b(&call_runtime, CS);
    __ CheckAllocationCanary(R0);

    /* Successfully allocated the object(s), now update top to point to */
    /* next object start and initialize the object. */
    __ str(R1, Address(THR, target::Thread::top_offset()));
    __ AddImmediate(R0, kHeapObjectTag);
    /* Initialize the tags. */
    /* R0: new object start as a tagged pointer. */
    /* R1: new object end address. */
    /* R2: allocation size. */
    {
      __ CompareImmediate(R2, target::UntaggedObject::kSizeTagMaxSizeTag);
      __ LslImmediate(R2, R2,
                      target::UntaggedObject::kTagBitsSizeTagPos -
                          target::ObjectAlignment::kObjectAlignmentLog2);
      __ csel(R2, ZR, R2, HI);

      /* Get the class index and insert it into the tags. */
      uword tags =
          target::MakeTagWordForNewSpaceObject(cid, /*instance_size=*/0);
      __ LoadImmediate(TMP, tags);
      __ orr(R2, R2, Operand(TMP));
      __ str(R2, FieldAddress(R0, target::Object::tags_offset())); /* Tags. */
    }
    /* Set the length field. */
    /* R0: new object start as a tagged pointer. */
    /* R1: new object end address. */
    __ mov(R2, AllocateTypedDataArrayABI::kLengthReg); /* Array length. */
    __ StoreCompressedIntoObjectNoBarrier(
        R0, FieldAddress(R0, target::TypedDataBase::length_offset()), R2);
    /* Initialize all array elements to 0. */
    /* R0: new object start as a tagged pointer. */
    /* R1: new object end address. */
    /* R2: iterator which initially points to the start of the variable */
    /* data area to be initialized. */
    __ AddImmediate(R2, R0, target::TypedData::HeaderSize() - 1);
    __ StoreInternalPointer(
        R0, FieldAddress(R0, target::PointerBase::data_offset()), R2);
    Label loop;
    __ Bind(&loop);
    ASSERT(target::kObjectAlignment == 2 * target::kWordSize);
    __ stp(ZR, ZR, Address(R2, 2 * target::kWordSize, Address::PairPostIndex));
    __ cmp(R2, Operand(R1));
    __ b(&loop, UNSIGNED_LESS);
    __ WriteAllocationCanary(R1);  // Fix overshoot.

    __ Ret();

    __ Bind(&call_runtime);
  }

  __ EnterStubFrame();
  __ Push(ZR);                                     // Result slot.
  __ PushImmediate(target::ToRawSmi(cid));         // Cid
  __ Push(AllocateTypedDataArrayABI::kLengthReg);  // Array length
  __ CallRuntime(kAllocateTypedDataRuntimeEntry, 2);
  __ Drop(2);  // Drop arguments.
  __ Pop(AllocateTypedDataArrayABI::kResultReg);
  __ LeaveStubFrame();
  __ Ret();
}

}  // namespace compiler

}  // namespace dart

#endif  // defined(TARGET_ARCH_ARM64)