[Impeller] fix Adreno precision issue in gaussian downsample. (#57212)
Fixes observed sampling problems on the adreno series where slightly different mediump/lowp behavior, resulting in the downsample shader showing "chunky" rendeering.
diff --git a/impeller/entity/shaders/downsample.glsl b/impeller/entity/shaders/downsample.glsl
index 58973b3..9a1bcf7 100644
--- a/impeller/entity/shaders/downsample.glsl
+++ b/impeller/entity/shaders/downsample.glsl
@@ -2,7 +2,7 @@
// Use of this source code is governed by a BSD-style license that can be
// found in the LICENSE file.
-precision mediump float;
+precision highp float;
#include <impeller/constants.glsl>
#include <impeller/types.glsl>
diff --git a/impeller/tools/malioc.json b/impeller/tools/malioc.json
index 96e0def..33951a5 100644
--- a/impeller/tools/malioc.json
+++ b/impeller/tools/malioc.json
@@ -5274,7 +5274,7 @@
"uses_late_zs_update": false,
"variants": {
"Main": {
- "fp16_arithmetic": 80,
+ "fp16_arithmetic": 0,
"has_stack_spilling": false,
"performance": {
"longest_path_bound_pipelines": [
@@ -5303,9 +5303,9 @@
"arith_cvt"
],
"shortest_path_cycles": [
- 0.0625,
+ 0.09375,
0.0,
- 0.0625,
+ 0.09375,
0.0,
0.0,
0.0,
@@ -5313,14 +5313,12 @@
],
"total_bound_pipelines": [
"arith_total",
- "arith_cvt",
- "varying",
- "texture"
+ "arith_cvt"
],
"total_cycles": [
- 0.25,
- 0.125,
- 0.25,
+ 0.3125,
+ 0.140625,
+ 0.3125,
0.0,
0.0,
0.25,
@@ -5329,7 +5327,7 @@
},
"stack_spill_bytes": 0,
"thread_occupancy": 100,
- "uniform_registers_used": 6,
+ "uniform_registers_used": 8,
"work_registers_used": 19
}
}
@@ -5369,7 +5367,7 @@
"arithmetic"
],
"total_cycles": [
- 3.3333332538604736,
+ 2.6666667461395264,
1.0,
1.0
]
@@ -5394,7 +5392,7 @@
"uses_late_zs_update": false,
"variants": {
"Main": {
- "fp16_arithmetic": 57,
+ "fp16_arithmetic": 0,
"has_stack_spilling": false,
"performance": {
"longest_path_bound_pipelines": [
@@ -5423,9 +5421,9 @@
"arith_cvt"
],
"shortest_path_cycles": [
- 0.0625,
+ 0.09375,
0.0,
- 0.0625,
+ 0.09375,
0.0,
0.0,
0.0,
@@ -5436,9 +5434,9 @@
"arith_cvt"
],
"total_cycles": [
- 0.359375,
- 0.125,
- 0.359375,
+ 0.421875,
+ 0.140625,
+ 0.421875,
0.0,
0.0,
0.25,
@@ -5447,7 +5445,7 @@
},
"stack_spill_bytes": 0,
"thread_occupancy": 100,
- "uniform_registers_used": 6,
+ "uniform_registers_used": 8,
"work_registers_used": 19
}
}
@@ -5487,14 +5485,14 @@
"arithmetic"
],
"total_cycles": [
- 5.333333492279053,
+ 5.0,
1.0,
1.0
]
},
- "thread_occupancy": 100,
+ "thread_occupancy": 50,
"uniform_registers_used": 1,
- "work_registers_used": 4
+ "work_registers_used": 6
}
}
}
@@ -8316,7 +8314,7 @@
"uses_late_zs_update": false,
"variants": {
"Main": {
- "fp16_arithmetic": 40,
+ "fp16_arithmetic": 0,
"has_stack_spilling": false,
"performance": {
"longest_path_bound_pipelines": [
@@ -8345,9 +8343,9 @@
"arith_cvt"
],
"shortest_path_cycles": [
- 0.109375,
+ 0.140625,
0.0,
- 0.109375,
+ 0.140625,
0.0,
0.0,
0.0,
@@ -8358,9 +8356,9 @@
"arith_cvt"
],
"total_cycles": [
- 0.328125,
- 0.09375,
- 0.328125,
+ 0.34375,
+ 0.125,
+ 0.34375,
0.0,
0.0,
0.25,
@@ -8388,7 +8386,7 @@
"uses_late_zs_update": false,
"variants": {
"Main": {
- "fp16_arithmetic": 28,
+ "fp16_arithmetic": 0,
"has_stack_spilling": false,
"performance": {
"longest_path_bound_pipelines": [
@@ -8417,9 +8415,9 @@
"arith_cvt"
],
"shortest_path_cycles": [
- 0.109375,
+ 0.140625,
0.0,
- 0.109375,
+ 0.140625,
0.0,
0.0,
0.0,
@@ -8430,9 +8428,9 @@
"arith_cvt"
],
"total_cycles": [
- 0.4375,
- 0.09375,
- 0.4375,
+ 0.453125,
+ 0.125,
+ 0.453125,
0.0,
0.0,
0.25,
