We had previously shipped both MSL 2.0 and 2.1 shaders to support
MacOS 10.13 (MSL 2.0 only) while also supporting the stencil blit
fast path that MSL 2.1 provides with the has_shader_stencil_output
feature.
Each configuration of precompiled shaders is ~300kb so we will drop
support for MacOS 10.13 and only ship the MSL 2.1 shaders.
Rework of the disablement for NVIDIA and GPU family 1 to be more
readable. The features themselves are always true because "disable on
NVIDIA" is always true and the check for "is NVIDIA" is moved to the
code which fails the initialization.
Bug: angleproject:8258
Change-Id: Icc8c69540e43fd2b0b237fffbfe170bb3422903f
Reviewed-on: https://chromium-review.googlesource.com/c/angle/angle/+/4681130
Commit-Queue: Geoff Lang <geofflang@chromium.org>
Reviewed-by: Kenneth Russell <kbr@chromium.org>
Reviewed-by: Shahbaz Youssefi <syoussefi@chromium.org>
Apple OpenGL drivers sometimes crash when given an instanced draw with
a buffer that has never been given data.
It's not efficient to check if the attribute is both zero-sized and
instanced so just ensure that every time a zero-sized buffer is bound
to an attribute, it gets initialized with some data.
Bug: chromium:1456243
Change-Id: I66b7c7017843153db2df3bc50010cba765d03c5f
Reviewed-on: https://chromium-review.googlesource.com/c/angle/angle/+/4642048
Commit-Queue: Geoff Lang <geofflang@chromium.org>
Reviewed-by: Shahbaz Youssefi <syoussefi@chromium.org>
This transformation was buggy and was disabled. Originally, it was
intended to be used everywhere. It is now needed for a GL driver
workaround.
This change reimplements this transformation and uses it as a GL
workaround.
Bug: chromium:1420130
Change-Id: I42d63fa5844bcf683ac41e61925aa637e033ca2e
Reviewed-on: https://chromium-review.googlesource.com/c/angle/angle/+/4676634
Commit-Queue: Shahbaz Youssefi <syoussefi@chromium.org>
Reviewed-by: Geoff Lang <geofflang@chromium.org>
The Vulkan loader somewhat recently introduced a requirement that
clients must opt-in to using portability implementations of Vulkan (such
as MoltenVK). Since there is no native Vulkan driver for macOS (and
therefore no alternative), unconditionally enable the portability
enumeration extension there.
Bug: angleproject:8229
Change-Id: I24f0f24e25abd277855ed9ac4de370cfb47d3266
Reviewed-on: https://chromium-review.googlesource.com/c/angle/angle/+/4639495
Commit-Queue: Shahbaz Youssefi <syoussefi@chromium.org>
Reviewed-by: Shahbaz Youssefi <syoussefi@chromium.org>
Reviewed-by: Steven Noonan <steven@uplinklabs.net>
Reviewed-by: Charlie Lao <cclao@google.com>
Auto-Submit: Steven Noonan <steven@uplinklabs.net>
Overwritten features were never reset even if a display was
terminated. On platforms that reuse displays for all tests
in the end2end suite, overridden feature would leak into
subsequent tests causing unexpected failure.
Bug: angleproject:8235
Change-Id: I1b359bc762a2bca8db4e4dbc7a587604e5bd6a5b
Reviewed-on: https://chromium-review.googlesource.com/c/angle/angle/+/4643453
Reviewed-by: Shahbaz Youssefi <syoussefi@chromium.org>
Commit-Queue: mohan maiya <m.maiya@samsung.com>
On outdated (but recent) AMD drivers, the Windows-only Vulkan extension
VK_EXT_full_screen_exclusive appeared to be implicitly enabled and set
to VK_FULL_SCREEN_EXCLUSIVE_APPLICATION_CONTROLLED_EXT mode. Even though
ANGLE did not enable or interact with this extension at all, the driver
was incorrectly returning VK_ERROR_FULL_SCREEN_EXCLUSIVE_MODE_LOST_EXT
error codes on various swapchain operations when the full screen window
focus was lost (i.e. alt-tab out and back in). Naturally, ANGLE was not
expecting these error codes and did not know how to handle them.
Depending on where the errors occurred, ANGLE might crash or retry
creating the swapchain repeatedly.
Treating the unexpected VK_ERROR_FULL_SCREEN_EXCLUSIVE_MODE_LOST_EXT
error code as VK_ERROR_OUT_OF_DATE_KHR/VK_SUBOPTIMAL_KHR was not
sufficient, because the driver would repeat the error on every swapchain
operation, apparently expecting the error to be handled by
a vkAcquireFullScreenExclusiveModeEXT call (even though that would make
no sense, since the extension was not enabled).
The incorrect driver behavior was reported to AMD and was fixed in
recent driver releases. The earliest driver I've tested and know to be
working is AMD's Adrenaline driver version 23.5.2
(VkPhysicalDeviceProperties calls this driverVersion 2.0.262/0x800106).
The last known bad version was 0x8000e9.
The simplest workaround on these older AMD graphics drivers is to
explicitly enable the extension, but set it to
VK_FULL_SCREEN_EXCLUSIVE_DISALLOWED_EXT mode. On newer drivers we do not
need to do anything with the extension and can ignore it.
Bug: angleproject:8215
Change-Id: I7c58d47a0350f4b0bc1a77f200c1e2f72fcde8d8
Reviewed-on: https://chromium-review.googlesource.com/c/angle/angle/+/4627279
Commit-Queue: Shahbaz Youssefi <syoussefi@chromium.org>
Reviewed-by: Shahbaz Youssefi <syoussefi@chromium.org>
Reviewed-by: Charlie Lao <cclao@google.com>
Updates the script to produce reasonably formatted code without
clang-format.
Autogen files moved to autogen/ sub-directories because clang-format
does not support per-file settings ;(
This allows to run this codegen very quickly
(~50ms on my machine)
Bug: angleproject:8193
Change-Id: Ie84282090d574ebb4debe3edcfd82f983f27a5ff
Reviewed-on: https://chromium-review.googlesource.com/c/angle/angle/+/4604578
Commit-Queue: Roman Lavrov <romanl@google.com>
Reviewed-by: Shahbaz Youssefi <syoussefi@chromium.org>
When a driver bug with dynamic state is encountered, it is hard to debug
which dynamic state exactly is causing an issue, due to the current
granularity of disabling all entire state from an extension. With this
change, every dynamic state gets its own ANGLE feature, and can be
toggled as necessary.
Disabling the supportsExtendedDynamicState* features implicitly
disables all dependent features.
Bug: b/285124778
Bug: b/275210062
Bug: fuchsia:107106
Bug: angleproject:5906
Change-Id: Ic291279872df2d0eb58618ff364ab118bdcc4a9f
Reviewed-on: https://chromium-review.googlesource.com/c/angle/angle/+/4577553
Reviewed-by: Cody Northrop <cnorthrop@google.com>
Auto-Submit: Shahbaz Youssefi <syoussefi@chromium.org>
Commit-Queue: Shahbaz Youssefi <syoussefi@chromium.org>
Reviewed-by: Yuxin Hu <yuxinhu@google.com>
This reverts commit ee64836f70 ,
relanding the patch stack described there.
Between patchsets 1 and 5:
- The shadow buffer allocation has been replaced with a multimap of
precisely-sized buffers, rather than rounding up buffer sizes.
- Garbage collection of shadow buffers is triggered in three situations:
- A certain number of context switches have occurred; this number
was hand-tuned to avoid GC every frame.
- A certain number of command buffer submissions has occurred; this
number was hand-tuned to GC no more often than every few seconds
on representative workloads.
- The total size of the allocated shadow buffers is more than 1 MB,
and either more than twice the size at the last garbage
collection, or 64 MB more than at the last garbage collection. In
this case, aggressive GC is performed in order to reclaim shadow
buffers more quickly.
Performance before and after these changes appears identical on
microbenchmarks. On one Figma test case, comparing GPU memory
allocated inside the BufferManager, peak consumption is decreased by
over 75%, and steady-state consumption decreases by over 88%.
Patchset 6 adds a needed workaround for a bug in the
AMDMTLBronzeDriver affecting uploads of client-side data, and
therefore some dEQP tests. It also streamlines the aggressive GC.
Bug: angleproject:7544
Change-Id: I81b061f0b33c27fa403527fa12d626f4e9c88ebe
Reviewed-on: https://chromium-review.googlesource.com/c/angle/angle/+/4497413
Reviewed-by: Shahbaz Youssefi <syoussefi@chromium.org>
Reviewed-by: Geoff Lang <geofflang@chromium.org>
Commit-Queue: Geoff Lang <geofflang@chromium.org>
Commit-Queue: Shahbaz Youssefi <syoussefi@chromium.org>
NVIDIA GPUs are in a very small number of Macs and we don't have have
adequate test coverage to properly support them.
Add the disable_metal_on_nvidia feature to control support for NVIDIA
at runtime. Also factor the support of GPU family 1 into a feature.
Bug: angleproject:8170
Change-Id: I160036d710cc9e278eb77e351a483b9f5b69e8ce
Reviewed-on: https://chromium-review.googlesource.com/c/angle/angle/+/4544659
Commit-Queue: Geoff Lang <geofflang@chromium.org>
Reviewed-by: Kenneth Russell <kbr@chromium.org>
On newer Mali GPUs (Valhall), the glDraw*BaseVertex* implementations
seem to be buggy and cause noticable artifacts. Chrome uses the
presence of baseVertexBaseInstanceANGLE and baseInstanceEXT
extensions to determine whether it can draw with a base vertex or to
fallback to manually offsetting the vbo.
This CL disables these extensions on Mali Valhall so that Chrome
takes the fallback path instead to fix the visual artifacts.
Bug: b/280931795
Change-Id: Ic27e37465283ca3338747b53aa7888c7eba4b7b8
Reviewed-on: https://chromium-review.googlesource.com/c/angle/angle/+/4549682
Reviewed-by: Geoff Lang <geofflang@chromium.org>
Commit-Queue: Geoff Lang <geofflang@chromium.org>
Querying the clip control state from the driver when sychronizing
external contexts generates driver errors.
Simply disable this extension in this configuration for now.
Chrome is currently not using it for rasterization.
This issue was noticed on specific Huawei devices in the wild. It
hasn't been observed on other OEMs.
Bug: chromium:1434317
Change-Id: I88d9bff1933274f61020b75e68aee9c94b0b684a
Reviewed-on: https://chromium-review.googlesource.com/c/angle/angle/+/4518062
Reviewed-by: Peng Huang <penghuang@chromium.org>
Commit-Queue: Geoff Lang <geofflang@chromium.org>
Reviewed-by: Colin Blundell <blundell@chromium.org>
This adds a feature (enabled by default) that results in
generating the same code regardless of whether emulateAlphaToCoverage
is enabled or addExplicitBoolCasts. This is done to maximize
cache sharing.
Bug: chromium:1423136
Change-Id: Ia491f13469c750fc2c45aecf1f93fed53b782dc5
Reviewed-on: https://chromium-review.googlesource.com/c/angle/angle/+/4496254
Reviewed-by: Alexey Knyazev <lexa.knyazev@gmail.com>
Reviewed-by: Kenneth Russell <kbr@chromium.org>
Commit-Queue: Scott Violet <sky@chromium.org>
Makes EAGL and CGL compilation exclusive to each other. Catalyst would
compile with both on ARM64, but only use EAGL.
Remove use of GLSL generation in MSL. Generating GLSL for MSL
compute and geometry shaders do not make sense.
Remove compilation of RewriteRowMajorMatrices for Metal.
It was removed from Metal in:
b23bf47c9e
Reland "Metal: rewrite default uniforms and uniform blocks"
Remove use of ANGLE_ENABLE_APPLE_WORKAROUNDS and gn variable
angle_enable_apple_translator_workarounds. The workarounds are just
unconditional business-as-usual implementation of ANGLE when ANGLE is
compile for PLATFORM_APPLE.
Fixed: angleproject:8129
Change-Id: I3af249708803fba816e2a7c92c35dddb84ce6946
Reviewed-on: https://chromium-review.googlesource.com/c/angle/angle/+/4428875
Auto-Submit: Kimmo Kinnunen <kkinnunen@apple.com>
Reviewed-by: Kyle Piddington <kpiddington@apple.com>
Commit-Queue: Kimmo Kinnunen <kkinnunen@apple.com>
Reviewed-by: Dan Glastonbury <djg@apple.com>
Reviewed-by: Kenneth Russell <kbr@chromium.org>
If the GLSL shader code expects the comparison between
two mediump float values returning true, but the SpirV
compiler treats one of them as 32-bit, and treats the
other one as 16-bit, the comparison would return false
instead. The SpirV compiler may not automatically cast
the mediump float values to 16 bits, because it may
utilize the RelaxedPrecision decoration to keep a mediump
float as 32-bit, so that the compiler can avoid the
type cast from 32-bit highp to 16-bit mediump.
This change adds an additional OpQuantizeToF16 SpirV
instruction to explicitly cast mediump float scalar
or mediump float vector to 16 bit, if they are assigned
with a highp float value. This ensures that if the
GLSL shader code ever compares two meiump float values,
the SpirV shader compiler is not accidentally comparing
a 16 bit with a 32 bit float value.
This fixe the deqp test failure on Pixel 6 and Pixel 7:
dEQP-GLES2.functional.shaders.algorithm.rgb_to_hsl_vertex.
Bug: b/274859104
Bug: b/274408172
Change-Id: Ifd996cea14c0f77f45ae90f38c8e53cf5035139f
Reviewed-on: https://chromium-review.googlesource.com/c/angle/angle/+/4400404
Commit-Queue: Yuxin Hu <yuxinhu@google.com>
Reviewed-by: Shahbaz Youssefi <syoussefi@chromium.org>
Reviewed-by: Cody Northrop <cnorthrop@google.com>
CompileMetalShaders results in the appropriate command line tool
being run to generate metallib blobs from shader source.
DisableProgramCaching results in not saving programs to BlobCache
LoadMetalShadersFromBlobCache results in trying to load metallibs
from BlobCache.
Bug: chromium:1423136
Change-Id: I01a4d7a5d60ed5ac978fb99db01b741e0f19e76b
Reviewed-on: https://chromium-review.googlesource.com/c/angle/angle/+/4434293
Reviewed-by: Geoff Lang <geofflang@chromium.org>
Commit-Queue: Scott Violet <sky@chromium.org>
New ESSL built-ins are mapped to their Metal
counterparts and tweaked to follow OpenGL ES
semantics when needed.
Fixed A2C interaction with sample coverage
by emulating the former on non-Apple GPUs.
Bug: angleproject:8097
Fixed: angleproject:5087
Change-Id: I5d28a941af5cbc14743a3930731529f11f55febd
Reviewed-on: https://chromium-review.googlesource.com/c/angle/angle/+/4404896
Reviewed-by: Kenneth Russell <kbr@chromium.org>
Reviewed-by: Kyle Piddington <kpiddington@apple.com>
Commit-Queue: Alexey Knyazev <lexa.knyazev@gmail.com>
This workaround turned out to affect pretty much every driver, and
numerous GL implementations were found to work around it similarly to
ANGLE. It seems like this workaround may only be necessary for
colorburn and colordodge, but for now ANGLE applies it to all modes.
Bug: b/274528004
Bug: b/277777623
Change-Id: Id555c981a9775f949a3022b7e92c755accea7cea
Reviewed-on: https://chromium-review.googlesource.com/c/angle/angle/+/4416158
Reviewed-by: Yuxin Hu <yuxinhu@google.com>
Reviewed-by: Yiwei Zhang <zzyiwei@chromium.org>
Commit-Queue: Shahbaz Youssefi <syoussefi@chromium.org>
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>
glBeginQuery(GL_TIME_ELAPSED/GL_TIMESTAMP) on Mali implementation
assumes a complete FrameBuffer. Without it glGetQueryObject will return
a meaningless value, causing some applications to misbehave.
This workaround caches and binds a default FBO in this case.
Bug: chromium:1356053,b/269068358
Change-Id: I756ded948c2c5aada744f9dd428ad77c37a009c6
Reviewed-on: https://chromium-review.googlesource.com/c/angle/angle/+/4359032
Reviewed-by: Geoff Lang <geofflang@chromium.org>
Commit-Queue: Kramer Ge <fangzhoug@chromium.org>
Used texture views to sample stencil values
from combined depth/stencil textures.
Texture2DDepthStencilTestES3.TexSampleModes* tests
were split into swizzled and non-swizzled variants
to suppress only swizzled cases on some platforms.
Added a new avoidStencilTextureSwizzle workaround to
skip creating swizzled texture views of stencil-only
textures on platforms that fail to sample from them.
Fixed: angleproject:8051
Change-Id: I0b1148f8d30fc6459239efcdaeee6c0364633cc8
Reviewed-on: https://chromium-review.googlesource.com/c/angle/angle/+/4304058
Commit-Queue: Alexey Knyazev <lexa.knyazev@gmail.com>
Reviewed-by: Kenneth Russell <kbr@chromium.org>
Reviewed-by: Geoff Lang <geofflang@chromium.org>
Currently ANGLE does not use "vkResetCommandBuffer()" on Vulkan
Secondary Command Buffers. Instead it uses "vkFreeCommandBuffers()" and
"vkAllocateCommandBuffers".
According to spec, "VK_COMMAND_POOL_CREATE_RESET_COMMAND_BUFFER_BIT" is
required only for command buffer reset operations.
However, some ARM drivers may not free memory in
"vkFreeCommandBuffers()" without this flag.
Bug: angleproject:8059
Change-Id: Ibfe45bca345dc48484b625c450369d30805cec77
Reviewed-on: https://chromium-review.googlesource.com/c/angle/angle/+/4306722
Reviewed-by: Charlie Lao <cclao@google.com>
Reviewed-by: Shahbaz Youssefi <syoussefi@chromium.org>
Commit-Queue: Igor Nazarov <i.nazarov@samsung.com>
Support GL_MIRROR_CLAMP_TO_EDGE_EXT address
mode when native support is available.
Adjusted validation, sorted switch cases by
target enums order.
Added new TextureMirrorClampToEdge* end2end
tests.
Drive-by: fixed texture address computation
for integer formats with GL_MIRRORED_REPEAT
on D3D11.
Fixed: angleproject:7968
Change-Id: Iaf29c8b4b32a7630c2a871f832d171f4bc4e2672
Reviewed-on: https://chromium-review.googlesource.com/c/angle/angle/+/4289137
Commit-Queue: Alexey Knyazev <lexa.knyazev@gmail.com>
Reviewed-by: Geoff Lang <geofflang@chromium.org>
Reviewed-by: Yuxin Hu <yuxinhu@google.com>
This CL separate out the logic of mLastCompletedQueueSerial update and
retireFinishedCommands and garbage clean up into different functions. At
submission, previously we are always check fence and update
mLastCompletedQueueSerials and calling reset on finished commands and do
garbage clean up. With this CL, we only do the fence check and update
mLastCompletedQueueSerials. Then it request worker thread to do the
command buffer reset and garbage cleanup. We uses the CommandProcessor's
thread for the reset and cleanup, since async submission path needs to
handle this clean up anyway.
This CL also added a new feature flag asyncCommandBufferReset and it is
disabled right now. This will be enabled in the follow up CL.
Bug: b/255411748
Change-Id: I6da558f8d4c962eb038e2378ccc76c464101cde2
Reviewed-on: https://chromium-review.googlesource.com/c/angle/angle/+/4244823
Reviewed-by: Shahbaz Youssefi <syoussefi@chromium.org>
Auto-Submit: Charlie Lao <cclao@google.com>
Commit-Queue: Charlie Lao <cclao@google.com>
Reviewed-by: Amirali Abdolrashidi <abdolrashidi@google.com>
D3D11:
* Exposed EXT_texture_border_clamp extension string,
to ensure that the relevant CTS tests are running.
* Updated StateManager11::setSamplerState to adjust
the border color based on the texture format.
* Refactored ShaderConstants11::updateSamplerMetadata to
correctly adjust the border color for integer formats.
* Removed unused SamplerMetadata.internalFormatBits
D3D9:
* Updated Renderer9::setSamplerState to adjust the border
color value based on the current texture format.
* Added borderColorSrgb feature required for some drivers.
GL:
* Copy alpha value to green for A and LA legacy formats to
workaround driver bugs when lumaWorkaround is not used.
Tests:
* Added ES 2.0 tests for texture formats
that require border color adjustments.
Fixed: angleproject:7969
Change-Id: I3d36cce43e76e6d5069a51865152c2250ecbb017
Reviewed-on: https://chromium-review.googlesource.com/c/angle/angle/+/4291000
Commit-Queue: Alexey Knyazev <lexa.knyazev@gmail.com>
Reviewed-by: Geoff Lang <geofflang@chromium.org>
There is a maximum limit for device memory object allocation. On some
platforms, there can be an error regarding too many object allocations
when 4096 device memory handles have been allocated. Suballocation can
help mitigate this issue. In this CL, some images will be allocated
using VMA API calls, which use suballocation.
* Added a new feature (useVmaForImageSuballocation).
* Added VMA allocation for ImageHelper, which is used in initMemory().
* Suballocation is used for VMA image allocation.
* If enabled, mVmaAllocation will be initialized in the ImageHelper
object (instead of mDeviceMemory).
* It is currently used for all platforms.
* Minor change to the name of an arg in CreateBuffer() declaration.
* Added test to make sure we can allocate at least 4096 images on
supported platforms (8000 in the test).
* Skipped the test "NonZeroBaseEmulatedClear" when run on Linux/Intel
if this feature is enabled (due to output color mismatch).
* Skipped several tests for capture/replay on Windows.
Bug: b/218891184
Change-Id: Ibf80c9c8c485b301da7d23b5ba4bcbb1a8e3194f
Reviewed-on: https://chromium-review.googlesource.com/c/angle/angle/+/4191202
Reviewed-by: Shahbaz Youssefi <syoussefi@chromium.org>
Reviewed-by: Charlie Lao <cclao@google.com>
Commit-Queue: Amirali Abdolrashidi <abdolrashidi@google.com>
Update the Metal library cache to be thread safe. Change
absl::flat_hash_map back to std::unordered_map because the value types
now contain a mutex which must not move.
Only generate async compilation tasks for shaders that were not already
compiled and in the cache.
Collapse some of the link methods in ProgramMtl that only had one call
site. All linking is now done in ProgramMtl::link and
ProgramMtl::load.
Support disabling parallel linking using the new
enableParallelMtlLibraryCompilation feature.
Bug: chromium:1385510
Change-Id: I71ba71a34d994066729df7e4170911f88c89de4a
Reviewed-on: https://chromium-review.googlesource.com/c/angle/angle/+/4234153
Reviewed-by: Quyen Le <lehoangquyen@chromium.org>
Reviewed-by: Jonah Ryan-Davis <jonahr@google.com>
Commit-Queue: Geoff Lang <geofflang@chromium.org>
Extensions are now grouped based on promotion-to-core status, and
initialization code is generally cleaned up. ANGLE currently doesn't
attempt to use a core version above Vulkan 1.1, as extensions are always
assumed to be exposed even if promoted to core in later versions.
With this change, it would be simpler to use a newer core version if
ever necessary.
Bug: angleproject:7959
Change-Id: Ia5fd3e06c18f7289e9e5a63af0f039a6dc89c687
Reviewed-on: https://chromium-review.googlesource.com/c/angle/angle/+/4224582
Auto-Submit: Shahbaz Youssefi <syoussefi@chromium.org>
Reviewed-by: Ian Elliott <ianelliott@google.com>
Reviewed-by: Charlie Lao <cclao@google.com>
Commit-Queue: Shahbaz Youssefi <syoussefi@chromium.org>
Add a small cache for (msl + compile parameters) -> MTLLibrary at the
egl::Display level. In regular executions of Chrome, the same shaders
(particularly vertex) are compiled multiple times in different programs.
Tested for a regular Chrome startup + open wikipedia + motionmark 1.2:
112/282 (40%) cache hits.
Several different caching methods were profiled (LinkProgram perf test)
- struct key with std::map : 303309
- struct key with std::unordered_map : 308090
- binary blob key with std::map : 263595
- binary blob key with std::unordered_map : 286051
- struct key + is_transparent with std::unordered_map : 304877
- struct key + is_transparent with absl::flat_hash_map : 335686
Using is_transparent allows us to search the hash map without
copying the shader source string to construct the key structure.
Bug: chromium:1385510
Change-Id: Ieec4ba526fe286276a4af7114d89cde32a8f9e1d
Reviewed-on: https://chromium-review.googlesource.com/c/angle/angle/+/4214012
Reviewed-by: Shahbaz Youssefi <syoussefi@chromium.org>
Commit-Queue: Geoff Lang <geofflang@chromium.org>
Reviewed-by: Kenneth Russell <kbr@chromium.org>
