This is a reland of f84daa070f
Eliminate static initializer(disallowed in Chromium) by replacing
shader string living in anonymous namespace from type "std::string"
to char array.
Original change's description:
> CopyTextureForBrowser: Support color format conversion
>
> This CL enables blit from RGBA8Unorm soruce texture to dst texture that
> |CopyImageBitmapToTexture| supported dst format.
>
> BUG=dawn:465
>
> Change-Id: I99846cf8dc37bc89e0c168a3d86193bb3a0c0ebb
> Reviewed-on: https://dawn-review.googlesource.com/c/dawn/+/44020
> Commit-Queue: Shaobo Yan <shaobo.yan@intel.com>
> Reviewed-by: Corentin Wallez <cwallez@chromium.org>
> Reviewed-by: Austin Eng <enga@chromium.org>
Bug: dawn:465
Change-Id: Ic3a156f5a20b217fd2aa5f86b01bad8ce77dc41c
Reviewed-on: https://dawn-review.googlesource.com/c/dawn/+/45443
Commit-Queue: Shaobo Yan <shaobo.yan@intel.com>
Reviewed-by: Corentin Wallez <cwallez@chromium.org>
Reviewed-by: Austin Eng <enga@chromium.org>
Still leave deprecated `depth` functional as there are some references in
other clients. Using `depth` and `depthOrArrayLayers` at the same time is
invalid. Add DeprecatedAPITests.
Bug: chromium:1176969
Change-Id: Ia06645e4f3c17588323dd36b11f9f3988b2e3aba
Reviewed-on: https://dawn-review.googlesource.com/c/dawn/+/44640
Commit-Queue: Shrek Shao <shrekshao@google.com>
Reviewed-by: Austin Eng <enga@chromium.org>
The CL that fixed deprecation warning expectations landed at the same
time as the CL which added this expectation and broke CQ.
We shouldn't check deprecation warnings here now because the warnings
for CreateRenderPipeline are temporarily disabled.
Bug: none
Change-Id: I99a8b6643e64f87b2354058dbac9a236c51a2afd
Reviewed-on: https://dawn-review.googlesource.com/c/dawn/+/45180
Reviewed-by: Austin Eng <enga@chromium.org>
Reviewed-by: Brandon Jones <bajones@chromium.org>
Commit-Queue: Austin Eng <enga@chromium.org>
The WGSL spec has been updated with 'Default Struct Layouts':
https://github.com/gpuweb/gpuweb/pull/1447
This removes the `[[offset(n)]]` decoration, and replaces it with two optional decorations: `[[size(n)]]` and `[[align(n)]]`, and a sensible set of sizes and alignments for each type.
Most `[[stride(n)]]` decorations have also been removed from arrays.
Bug: tint:626
Bug: tint:629
Change-Id: Ib0d2741f61ea943e6fb99d00cbb5cab2f97ae7be
Reviewed-on: https://dawn-review.googlesource.com/c/dawn/+/44280
Commit-Queue: Ben Clayton <bclayton@google.com>
Reviewed-by: Corentin Wallez <cwallez@chromium.org>
This patch fixes a bug in the computation of temporary buffer size in
the implementation of the workaround for the T2T copy issue on Intel
GPUs.
With this tests all the T2T operation tests in WebGPU CTS will be able
to pass on Intel Gen9 and Gen9.5 GPUs.
BUG=chromium:1161355
TEST=dawn_end2end_tests
Change-Id: Ic4fb0c0056cf7b01720e47cc91c37a9f293cd194
Reviewed-on: https://dawn-review.googlesource.com/c/dawn/+/44840
Reviewed-by: Corentin Wallez <cwallez@chromium.org>
Commit-Queue: Jiawei Shao <jiawei.shao@intel.com>
Remove usage of the depth offset in TextureCopySplitter. Because there
is no slice pitch alignment requirement, it is simpler to express the
subresource footprint with just width and height. Fixes incorrect copies
when using WriteTexture on a texture with 64 or less width.
Bug: dawn:573
Change-Id: I51cb73b522a443b8b42d9ba290c69541211be4bb
Reviewed-on: https://dawn-review.googlesource.com/c/dawn/+/39200
Reviewed-by: Rafael Cintron <rafael.cintron@microsoft.com>
Reviewed-by: Corentin Wallez <cwallez@chromium.org>
Commit-Queue: Brandon Jones <brandon1.jones@intel.com>
Allow external images to specify all allowed texture usages which will
be checked by ProduceTexture. This avoids needing to assume the usage
between multiple calls to ProduceTexture is always the same.
Bug=dawn:625
Change-Id: I9fdb59af23e6c160e939ab1c51de03542248c0c8
Reviewed-on: https://dawn-review.googlesource.com/c/dawn/+/44260
Commit-Queue: Bryan Bernhart <bryan.bernhart@intel.com>
Reviewed-by: Corentin Wallez <cwallez@chromium.org>
Currently normalizes anything using the new layout to the old one for
the sake of getting things working as quickly as possible. Follow up
changes will gradually push the new layout through more of the stack.
Bug: dawn:642
Change-Id: Ie92fa9dde21174f62ceba1a1f4866cbc24c5fc6f
Reviewed-on: https://dawn-review.googlesource.com/c/dawn/+/38600
Commit-Queue: Brandon Jones <bajones@chromium.org>
Reviewed-by: Corentin Wallez <cwallez@chromium.org>
On Intel Gen9 (SKL) and Gen9.5 (KBL, CFL, CML) GPUs with latest
Intel D3D12 driver (27.20.100.9316), there is a bug in the command
CopyTextureRegion() when we want to do the texture-to-texture copy
with the formats whose texel block size < 4 bytes and source mipmap
level > destination mipmap level.
This patch adds a workaround for this driver bug by implementing
the functionality of the T2T copy with one T2B copy and one B2T
copy.
BUG=chromium:1161355
TEST=dawn_end2end_tests
Change-Id: I688bb8bae277832aaba1be2680012040ee8e1160
Reviewed-on: https://dawn-review.googlesource.com/c/dawn/+/43860
Reviewed-by: Corentin Wallez <cwallez@chromium.org>
Commit-Queue: Austin Eng <enga@chromium.org>
CopyTextureForBrowserTests will cover incoming color format conversion
cases. While it is OK for |unorm| formats to compare their pixel values
on the CPU side, we cannot do such comparisons for the |float| formats
because we may meet the precision issues when comparing a value
generated at the CPU side to the one from the GPU side.
Refactor this test suites by using compute shader and do bit-by-bit
comparison from source texture and destination texture.
BUG=dawn:465
Change-Id: I979fcf1a1d96bbe9f8a4cf2f1a305d488e88b257
Reviewed-on: https://dawn-review.googlesource.com/c/dawn/+/36140
Commit-Queue: Shaobo Yan <shaobo.yan@intel.com>
Reviewed-by: Austin Eng <enga@chromium.org>
This patch adds a regression test to dawn_end2end_tests for an
Intel D3D12 driver bug about the T2T copies with specific texture
formats when the source mipmap level is greater than 0.
This patch also does several clean-ups in CopyTests.cpp to support
testing arbitrary texture formats:
1. Initialize the source textures with device.WriteTexture().
2. Set default values for TextureSpec.copyLevel (0),
TextureSpec.levelCount (1) and TextureSpec.copyOrigin ({0, 0, 0}).
3. Add TextureSpec.format and set the default value for it (RGBA8Unorm).
4. Support initializing and comparing the texture data with arbitrary
texture format instead of RGBA8Unorm only in CopyTests_T2T.
5. Check all the texels in the related destination texture subresources
instead of only the copied texels.
BUG=chromium:1161355
TEST=dawn_end2end_tests
Change-Id: Ib32aac57dc7bdc9a4c2edc96093afa945aa08a5c
Reviewed-on: https://dawn-review.googlesource.com/c/dawn/+/42240
Commit-Queue: Jiawei Shao <jiawei.shao@intel.com>
Reviewed-by: Austin Eng <enga@chromium.org>
This change allows multiple Dawn textures to be created from the same
D3D11 resource. This avoids re-opening the shared handle by caching the
D3D12 resource outside of the Dawn texture.
Re-opening the handle costs 5-10% of CPU cycles per frame, which far
exceeded syncronization costs.
In a future change, WrapSharedHandle will be depreciated.
BUG=dawn:625
Change-Id: If0d2dc9b7445ec3ae718bc5305164db88057c4ea
Reviewed-on: https://dawn-review.googlesource.com/c/dawn/+/42140
Commit-Queue: Bryan Bernhart <bryan.bernhart@intel.com>
Reviewed-by: Austin Eng <enga@chromium.org>
Reviewed-by: Corentin Wallez <cwallez@chromium.org>
Async functions should fail by rejecting the callback instead of
generating a device error. This fixes a leak in the wire where
the allocation for the callback was never delete since it was
never called.
Fixed: chromium:1181627
Change-Id: I840073c1d1b5f1401aa8ed29d3c8f0e1e4fefd35
Reviewed-on: https://dawn-review.googlesource.com/c/dawn/+/42540
Reviewed-by: Austin Eng <enga@chromium.org>
Reviewed-by: Corentin Wallez <cwallez@chromium.org>
Commit-Queue: Austin Eng <enga@chromium.org>
Currently on D3D12 backend a device lost will occur when we do a T2T
copy under the following situations:
1. with Depth32Float
2. only copy one row
3. bufferCopyOffset == 256
This is because in current implementation it is possible for us to do
a copy with D3D12_SUBRESOURCE_FOOTPRINT.Depth > 1, which is not allowed
with DXGI_FORMAT_D32_FLOAT because this format is not supported to be
used as 3D textures.
This patch adds a regression test for this bug and we will fix it
later.
BUG=dawn:693
TEST=dawn_end2end_tests
Change-Id: Ib6fe70988b5b217d5f14d3f32999b3841e5d23b9
Reviewed-on: https://dawn-review.googlesource.com/c/dawn/+/42600
Commit-Queue: Austin Eng <enga@chromium.org>
Reviewed-by: Corentin Wallez <cwallez@chromium.org>
Reviewed-by: Brandon Jones <brandon1.jones@intel.com>
Reviewed-by: Austin Eng <enga@chromium.org>
The are replaced by Queue::OnSubmittedWorkDone. Only Queue::CreateFence
is deprecated since all other fence-related calls require a fence to
work.
Also ports a number of uses of fences in test harness to use
Queue::OnSubmittedWorkDone instead.
Bug: chromium:1177476
Change-Id: I479415f72b08158a3484013e00db8facd11e6f33
Reviewed-on: https://dawn-review.googlesource.com/c/dawn/+/42660
Auto-Submit: Corentin Wallez <cwallez@chromium.org>
Commit-Queue: Brandon Jones <bajones@chromium.org>
Reviewed-by: Austin Eng <enga@chromium.org>
Reviewed-by: Brandon Jones <bajones@chromium.org>
