godotengine/godot-angle-static
1
0
Fork 0
mirror of https://github.com/godotengine/godot-angle-static.git synced 2026-01-03 14:09:33 +03:00
Code Issues Packages Projects Releases Wiki Activity

Metal: Always prefer staged GPU uploads for textures.

Browse Source 
Uploading Metal texture data via MTLTexture::replaceRegion is a source
of lots of CPU hangs and jank. There may be better heuristics to
determine if we should do a CPU vs GPU upload but for now preferring
GPU uploads has better overall performance.

This also improves the MotionMark images test:
 - 40 -> 100 when using Metal.
 - 217 -> 235 when using Metal +
    https://chromium-review.googlesource.com/c/chromium/src/+/4091749

Bug: angleproject:8024, angleproject:8092, angleproject:8109
Change-Id: I36b5f585884391b4cc416365ae65f8542745beee
Reviewed-on: https://chromium-review.googlesource.com/c/angle/angle/+/4264963
Reviewed-by: Kenneth Russell <kbr@chromium.org>
Commit-Queue: Geoff Lang <geofflang@chromium.org>
Reviewed-by: Jonah Ryan-Davis <jonahr@google.com>
This commit is contained in:
Geoff Lang
2023-03-22 12:54:08 -04:00
committed by Angle LUCI CQ
parent ec307af4ad
commit f1f80e1e3f
10 changed files with 98 additions and 17 deletions
Download Patch File Download Diff File Expand all files Collapse all files

11
include/platform/FeaturesMtl_autogen.h
View File

@@ -265,6 +265,17 @@ struct FeaturesMtl : FeatureSetBase
FeatureInfo enableParallelMtlLibraryCompilation = {
"enableParallelMtlLibraryCompilation", FeatureCategory::MetalFeatures,
"Compile MTLLibrary in multiple threads.", &members, "http://crbug.com/1385510"};
FeatureInfo alwaysPreferStagedTextureUploads = {
"alwaysPreferStagedTextureUploads", FeatureCategory::MetalFeatures,
"Always prefer to upload texture data via a staging buffer and avoid "
"MTLTexture::replaceRegion.",
&members, "http://crbug.com/1380790"};
FeatureInfo disableStagedInitializationOfPackedTextureFormats = {
"disableStagedInitializationOfPackedTextureFormats", FeatureCategory::MetalFeatures,
"Staged GPU upload of some packed texture formats such as RGB9_E5 fail on Intel GPUs.",
&members, "http://anglebug.com/8092"};
};
inline FeaturesMtl::FeaturesMtl() = default;

16
include/platform/mtl_features.json
View File

@@ -289,6 +289,22 @@
"Compile MTLLibrary in multiple threads."
],
"issue": "http://crbug.com/1385510"
},
{
"name": "always_prefer_staged_texture_uploads",
"category": "Features",
"description": [
"Always prefer to upload texture data via a staging buffer and avoid MTLTexture::replaceRegion."
],
"issue": "http://crbug.com/1380790"
},
{
"name": "disable_staged_initialization_of_packed_texture_formats",
"category": "Features",
"description": [
"Staged GPU upload of some packed texture formats such as RGB9_E5 fail on Intel GPUs."
],
"issue": "http://anglebug.com/8092"
}
]
}

8
scripts/code_generation_hashes/ANGLE_features.json
View File

@@ -4,7 +4,7 @@
"include/platform/FeaturesGL_autogen.h":
"331b1660b3982f6540e362764df26021",
"include/platform/FeaturesMtl_autogen.h":
"d67e2035dabd6e495737da385d5dbe94",
"fc2e0328446f01a76c4b301e3b3eae3c",
"include/platform/FeaturesVk_autogen.h":
"5ab2618c7c158a4af52342eed3114e5f",
"include/platform/FrontendFeatures_autogen.h":
@@ -18,11 +18,11 @@
"include/platform/gl_features.json":
"cef92e7ede4c824b8bae37123401d354",
"include/platform/mtl_features.json":
"cd4de616ed3d18620f8c5b16cb076e66",
"17c058de70379f2f5695e1791d45d63c",
"include/platform/vk_features.json":
"0bc86b174a5f2ec7d1efcf396453693a",
"util/angle_features_autogen.cpp":
"8cb93987ff50d8fb12e117421b904001",
"7ef32f9454ec37566a93741815f65a3f",
"util/angle_features_autogen.h":
"0f74a51d5cfe16fb2a2087467f1e9362"
"1b7b43d4c28184cb6b74f8982688e6ce"
}

7
src/libANGLE/renderer/metal/DisplayMtl.mm
View File

@@ -1274,6 +1274,13 @@ void DisplayMtl::initializeFeatures()
ANGLE_FEATURE_CONDITION((&mFeatures), enableInMemoryMtlLibraryCache, true);
ANGLE_FEATURE_CONDITION((&mFeatures), enableParallelMtlLibraryCompilation, true);
// Uploading texture data via staging buffers improves performance on all tested systems.
// http://anglebug.com/8092: Disabled on intel due to some texture formats uploading incorrectly
// with staging buffers
ANGLE_FEATURE_CONDITION(&mFeatures, alwaysPreferStagedTextureUploads, true);
ANGLE_FEATURE_CONDITION(&mFeatures, disableStagedInitializationOfPackedTextureFormats,
isIntel());
ApplyFeatureOverrides(&mFeatures, getState());
}

11
src/libANGLE/renderer/metal/TextureMtl.mm
View File

@@ -525,13 +525,11 @@ angle::Result UploadTextureContents(const gl::Context *context,
{
ASSERT(texture && texture->valid());
ContextMtl *contextMtl = mtl::GetImpl(context);
#if !TARGET_OS_SIMULATOR
const angle::FeaturesMtl &features = contextMtl->getDisplay()->getFeatures();
ContextMtl *contextMtl = mtl::GetImpl(context);
const mtl::Format &mtlFormat = contextMtl->getPixelFormat(textureAngleFormat.id);
bool forceStagedUpload =
texture->hasIOSurface() && features.uploadDataToIosurfacesWithStagingBuffers.enabled;
if (texture->isCPUAccessible() && !forceStagedUpload)
bool preferGPUInitialization = PreferStagedTextureUploads(context, texture, mtlFormat);
if (texture->isCPUAccessible() && !preferGPUInitialization)
{
// If texture is CPU accessible, just call replaceRegion() directly.
texture->replaceRegion(contextMtl, region, mipmapLevel, slice, data, bytesPerRow,
@@ -539,7 +537,6 @@ angle::Result UploadTextureContents(const gl::Context *context,
return angle::Result::Continue;
}
#endif
// Texture is not CPU accessible or staging is forced due to a workaround
if (!textureAngleFormat.depthBits && !textureAngleFormat.stencilBits)

6
src/libANGLE/renderer/metal/mtl_utils.h
View File

@@ -33,12 +33,16 @@ void StopFrameCapture();
namespace mtl
{
bool PreferStagedTextureUploads(const gl::Context *context,
const TextureRef &texture,
const Format &textureObjFormat);
// Initialize texture content to black.
angle::Result InitializeTextureContents(const gl::Context *context,
const TextureRef &texture,
const Format &textureObjFormat,
const ImageNativeIndex &index);
// Same as above but using GPU clear operation instead of CPU.
// Same as above but using GPU clear operation instead of CPU.forma
// - channelsToInit parameter controls which channels will get their content initialized.
angle::Result InitializeTextureContentsGPU(const gl::Context *context,
const TextureRef &texture,

48
src/libANGLE/renderer/metal/mtl_utils.mm
View File

@@ -346,6 +346,47 @@ static angle::Result InitializeCompressedTextureContents(const gl::Context *cont
} // namespace
bool PreferStagedTextureUploads(const gl::Context *context,
const TextureRef &texture,
const Format &textureObjFormat)
{
// The simulator MUST upload all textures as staged.
if (TARGET_OS_SIMULATOR)
{
return true;
}
ContextMtl *contextMtl = mtl::GetImpl(context);
const angle::FeaturesMtl &features = contextMtl->getDisplay()->getFeatures();
const gl::InternalFormat &intendedInternalFormat = textureObjFormat.intendedInternalFormat();
if (intendedInternalFormat.compressed || textureObjFormat.actualAngleFormat().isBlock)
{
return false;
}
if (intendedInternalFormat.isLUMA())
{
return false;
}
if (features.disableStagedInitializationOfPackedTextureFormats.enabled)
{
switch (intendedInternalFormat.sizedInternalFormat)
{
case GL_RGB9_E5:
case GL_R11F_G11F_B10F:
return false;
default:
break;
}
}
return (texture->hasIOSurface() && features.uploadDataToIosurfacesWithStagingBuffers.enabled) ||
features.alwaysPreferStagedTextureUploads.enabled;
}
angle::Result InitializeTextureContents(const gl::Context *context,
const TextureRef &texture,
const Format &textureObjFormat,
@@ -358,10 +399,7 @@ angle::Result InitializeTextureContents(const gl::Context *context,
const gl::InternalFormat &intendedInternalFormat = textureObjFormat.intendedInternalFormat();
bool forceGPUInitialization = false;
#if TARGET_OS_SIMULATOR
forceGPUInitialization = true;
#endif // TARGET_OS_SIMULATOR
bool preferGPUInitialization = PreferStagedTextureUploads(context, texture, textureObjFormat);
// This function is called in many places to initialize the content of a texture.
// So it's better we do the initial check here instead of let the callers do it themselves:
@@ -385,7 +423,7 @@ angle::Result InitializeTextureContents(const gl::Context *context,
startDepth);
}
else if (texture->isCPUAccessible() && index.getType() != gl::TextureType::_2DMultisample &&
index.getType() != gl::TextureType::_2DMultisampleArray && !forceGPUInitialization)
index.getType() != gl::TextureType::_2DMultisampleArray && !preferGPUInitialization)
{
const angle::Format &dstFormat = angle::Format::Get(textureObjFormat.actualFormatId);
const size_t dstRowPitch = dstFormat.pixelBytes * size.width;

3
src/tests/angle_end2end_tests_expectations.txt
View File

@@ -987,6 +987,9 @@
6824 METAL INTEL : CubeMapTextureTest.SampleCoordinateTransformGrad/* = SKIP
6824 METAL INTEL : CubeMapTextureTest.SampleCoordinateTransformGrad_ES3/* = SKIP
// Staged texture uploads causes some incorrect texture data on Intel
8109 MAC METAL INTEL : Texture2DTest.ManySupersedingTextureUpdates/ES2_Metal = SKIP
2799 NVIDIA OPENGL WIN : RobustBufferAccessBehaviorTest.DrawElementsIndexOutOfRangeWithStaticDraw/* = SKIP
2799 NVIDIA OPENGL WIN : RobustBufferAccessBehaviorTest.DrawElementsIndexOutOfRangeWithDynamicDraw/* = SKIP
2861 ANDROID GLES : RobustBufferAccessBehaviorTest.NoBufferData/* = SKIP

3
util/angle_features_autogen.cpp
View File

@@ -38,6 +38,7 @@ constexpr PackedEnumMap<Feature, const char *> kFeatureNames = {{
{Feature::AllowTranslateUniformBlockToStructuredBuffer,
"allowTranslateUniformBlockToStructuredBuffer"},
{Feature::AlwaysCallUseProgramAfterLink, "alwaysCallUseProgramAfterLink"},
{Feature::AlwaysPreferStagedTextureUploads, "alwaysPreferStagedTextureUploads"},
{Feature::AlwaysUnbindFramebufferTexture2D, "alwaysUnbindFramebufferTexture2D"},
{Feature::AppendAliasedMemoryDecorationsToSsbo, "appendAliasedMemoryDecorationsToSsbo"},
{Feature::AsyncCommandBufferReset, "asyncCommandBufferReset"},
@@ -82,6 +83,8 @@ constexpr PackedEnumMap<Feature, const char *> kFeatureNames = {{
{Feature::DisableRasterOrderGroups, "disableRasterOrderGroups"},
{Feature::DisableRWTextureTier2Support, "disableRWTextureTier2Support"},
{Feature::DisableSemaphoreFd, "disableSemaphoreFd"},
{Feature::DisableStagedInitializationOfPackedTextureFormats,
"disableStagedInitializationOfPackedTextureFormats"},
{Feature::DisableSyncControlSupport, "disableSyncControlSupport"},
{Feature::DisableTextureClampToBorder, "disableTextureClampToBorder"},
{Feature::DisableTimestampQueries, "disableTimestampQueries"},

2
util/angle_features_autogen.h
View File

@@ -37,6 +37,7 @@ enum class Feature
AllowSeparateDepthStencilBuffers,
AllowTranslateUniformBlockToStructuredBuffer,
AlwaysCallUseProgramAfterLink,
AlwaysPreferStagedTextureUploads,
AlwaysUnbindFramebufferTexture2D,
AppendAliasedMemoryDecorationsToSsbo,
AsyncCommandBufferReset,
@@ -78,6 +79,7 @@ enum class Feature
DisableRasterOrderGroups,
DisableRWTextureTier2Support,
DisableSemaphoreFd,
DisableStagedInitializationOfPackedTextureFormats,
DisableSyncControlSupport,
DisableTextureClampToBorder,
DisableTimestampQueries,