vkCreateImage when bound to imported memory, must also have
a VkExternalImageCreateInfo struct in its pNext chain that
has VK_EXTERNAL_MEMORY_HANDLE_TYPE_OPAQUE_FD_BIT as a
handleType.
This was discovered when tests started failing after a roll
of the Vulkan validation layers.
Tbr=cwallez@chromium.org, senorblanco@chromium.com
Bug: none
Change-Id: I4692d650af27b9325342baeda34a0d982fc49041
Reviewed-on: https://dawn-review.googlesource.com/c/dawn/+/24200
Reviewed-by: Austin Eng <enga@chromium.org>
Commit-Queue: Austin Eng <enga@chromium.org>
This unblocks a roll into Skia.
Allow comparison against type limits that might be tautological on
32bit or 64bit systems. Without this the following produces an error
on 64bit:
if (myUint64 > std::numeric_limits<size_t>::max()) {...}
Bug: None
Change-Id: I31fc579738b6fb85b40cfeac8f30936161a15da1
Reviewed-on: https://dawn-review.googlesource.com/c/dawn/+/24162
Reviewed-by: Corentin Wallez <cwallez@chromium.org>
Commit-Queue: Corentin Wallez <cwallez@chromium.org>
Since the build/ directory is checked out behind 'dawn_standalone', the
'lastchange' hook depending on that should be guarded by the same
condition.
Bug: dawn:437
Change-Id: If6d1f5ee505eab5bc5519ec6d29b2efe5d69d9aa
Reviewed-on: https://dawn-review.googlesource.com/c/dawn/+/24123
Reviewed-by: Corentin Wallez <cwallez@chromium.org>
Reviewed-by: Austin Eng <enga@chromium.org>
Commit-Queue: Jiajie Hu <jiajie.hu@intel.com>
Added a constraint in ValidateBufferBinding function checking for too
large uniform buffer bindings. Added MaxUniformBufferBindingSize
test in BindGroupValidationTest.
Bug: dawn:436
Change-Id: I31c6e2236ce928d5e81c43455eb18cf4eacdc0f1
Reviewed-on: https://dawn-review.googlesource.com/c/dawn/+/24081
Commit-Queue: Tomek Ponitka <tommek@google.com>
Reviewed-by: Austin Eng <enga@chromium.org>
The stress test is causing timeouts on slower GPUs
due to the TDR threshold being hit. This halves the
number of writes required to complete the test.
BUG=dawn:228
Change-Id: I18dfa68b39a2c079b89eeadc78a69675e5c3269f
Reviewed-on: https://dawn-review.googlesource.com/c/dawn/+/24100
Reviewed-by: Corentin Wallez <cwallez@chromium.org>
Reviewed-by: Bryan Bernhart <bryan.bernhart@intel.com>
Reviewed-by: Austin Eng <enga@chromium.org>
Commit-Queue: Bryan Bernhart <bryan.bernhart@intel.com>
SetSubData's count is a uint64_t while Queue::WriteBuffer's is a size_t
so we add a small validation check that no narrowing will occur.
No test is added because SetSubData will be removed soon, and it will be
hard to test because of the Null backend's artificial OOM.
This was found while debugging an unrelated fuzzer issue.
Bug: chromium:1099621
Change-Id: I27a9da2b94f51e889c5573f88d4a0a73fea5985c
Reviewed-on: https://dawn-review.googlesource.com/c/dawn/+/23961
Reviewed-by: Stephen White <senorblanco@chromium.org>
Reviewed-by: Austin Eng <enga@chromium.org>
Commit-Queue: Corentin Wallez <cwallez@chromium.org>
In the client code, we often need to translate between WGPUObject (the
API type) to Object* (the internal client type). This added a bunch of
reinterpret_casts that make the code less readable and more fragile.
This CL adds FromAPI and ToAPI helpers in the autogenerated
ApiObjects_autogen.h header, that convert between API and internal types
in a type-safe way.
Bug: dawn:445
Change-Id: Ia1bf624f0315ced496b95cb660adf88abd916d71
Reviewed-on: https://dawn-review.googlesource.com/c/dawn/+/24063
Commit-Queue: Corentin Wallez <cwallez@chromium.org>
Reviewed-by: Stephen White <senorblanco@chromium.org>
Reviewed-by: Austin Eng <enga@chromium.org>
This header was only used to call DeviceCreateErrorBuffer and
DeviceInjectError that are used in some handwritten client commands.
We remove the need for the header by making these two commands
handwritten.
This also improves readability, previously injecting errors read:
ClientDeviceInjectError(reinterpret_cast<WGPUDevice>(device),
WGPUErrorType_Validation,
"Some validation message");
And now reads:
device->InjectError(WGPUErrorType_Validation, "Some validation message");
Bug: dawn:445
Change-Id: Ie11570aacf3036e13abd174d91670ecb84661226
Reviewed-on: https://dawn-review.googlesource.com/c/dawn/+/24080
Commit-Queue: Corentin Wallez <cwallez@chromium.org>
Reviewed-by: Stephen White <senorblanco@chromium.org>
Reviewed-by: Austin Eng <enga@chromium.org>
This CL modifies code generation so that the generated client procs call
the handwritten methods on client objects directly.
Previously the flow was:
- wgpuBarDoStuff
- ClientBarDoStuff
- ClientHandwrittenBarDoStuff
- client::Bar::DoStuff
With this CL the flow is:
- wgpuBarDoStuff
- ClientBarDoStuff
- client::Bar::DoStuff
This required adding Buffer creation methods on client::Device instead
of calling client::Buffer static methods directly.
Bug: dawn:445
Change-Id: I1b332b71ac7a03685afcf8fd0617d3d27da468cf
Reviewed-on: https://dawn-review.googlesource.com/c/dawn/+/24062
Commit-Queue: Corentin Wallez <cwallez@chromium.org>
Reviewed-by: Stephen White <senorblanco@chromium.org>
Reviewed-by: Austin Eng <enga@chromium.org>
This CL only moves code, renames client::Fence members, and introduces
client::Queue. Additional fence methods are added for the interaction
with the queue. There are no functional changes.
With this ApiProcs.cpp is almost passthrough and will be removed in a
follow-up CL.
Bug: dawn:445
Change-Id: I65544ef76b54614452cf7c74a948a96cb35a4cfe
Reviewed-on: https://dawn-review.googlesource.com/c/dawn/+/24061
Reviewed-by: Austin Eng <enga@chromium.org>
Reviewed-by: Stephen White <senorblanco@chromium.org>
Commit-Queue: Corentin Wallez <cwallez@chromium.org>
This also removes the reliance on BufferBase::IsMapped to know whether
to unmap on destroy. This call was confusing because it was used by the
D3D12 backend to know if its own storage was mapped, but semantically
seemed to check for Buffer::State::Mapped (and not MappedAtCreation).
Bug: dawn:445
Change-Id: I3d6fde1d2996798d53264d5643545f0efb90551a
Reviewed-on: https://dawn-review.googlesource.com/c/dawn/+/24060
Commit-Queue: Corentin Wallez <cwallez@chromium.org>
Reviewed-by: Brandon Jones <brandon1.jones@intel.com>
This CL adds GetMappedRange reusing the existing GetMappedPointerImpl
call in dawn_native. In dawn_wire tracking is added to keep a
Buffer::mMappedData around.
Tests are added to test the result of Get[|Const]MappedRange in all
buffer states.
Bug: dawn:445
Change-Id: I3737dc4d36f31d392839952da0b5c0d10c7c8a88
Reviewed-on: https://dawn-review.googlesource.com/c/dawn/+/23861
Commit-Queue: Corentin Wallez <cwallez@chromium.org>
Reviewed-by: Kai Ninomiya <kainino@chromium.org>
In a follow-up CL we need to know if the buffer is in the
MappedAtCreation state for validation of GetMappedRange. Having the
state not 100% reflect the state in the spec meant that validation if
GetMappedRange would fail for buffers mapped at creation for which
IsMapWritable is true.
Bug: dawn:445
Change-Id: I4a64335a708b838526da8d65f907b21c782816e0
Reviewed-on: https://dawn-review.googlesource.com/c/dawn/+/23981
Commit-Queue: Corentin Wallez <cwallez@chromium.org>
Reviewed-by: Kai Ninomiya <kainino@chromium.org>
This CL only moves code, renames client::Buffer members and does
additional casts where needed. No functional changes.
Bug: dawn:445
Change-Id: I2bf83ecc1c9b36d5965d0365360dd981fcd41aac
Reviewed-on: https://dawn-review.googlesource.com/c/dawn/+/23860
Reviewed-by: Kai Ninomiya <kainino@chromium.org>
Reviewed-by: Austin Eng <enga@chromium.org>
Commit-Queue: Corentin Wallez <cwallez@chromium.org>
Otherwise on 32-bit platforms, creation of a buffer bigger than the max
size_t would overflow the check and no OOM would be produced.
Bug: chromium:1099621
Change-Id: I987a338b150d10c0eabc3eb5fb3b815a5c2a5ca2
Reviewed-on: https://dawn-review.googlesource.com/c/dawn/+/23960
Reviewed-by: Austin Eng <enga@chromium.org>
Reviewed-by: Kai Ninomiya <kainino@chromium.org>
Commit-Queue: Kai Ninomiya <kainino@chromium.org>
In a render/compute pass, if a texture is used as the same usages
across its subresources, and it is also used as different usages
across its subresources. We should mark that it is not used as the
same usages across its subresources for this pass, and we need to
visit every single subresource to track its usages transition.
Bug: dawn:462
Change-Id: Ide0770db62a3b7b85664757b21eb3280a0bb646c
Reviewed-on: https://dawn-review.googlesource.com/c/dawn/+/23300
Commit-Queue: Yunchao He <yunchao.he@intel.com>
Reviewed-by: Austin Eng <enga@chromium.org>
Added Device::GetOrCreateEmptyBindGroupLayout which caches the result
after being run and modified PipelineBase::getBindGroupLayout to use
that instead of creating a new empty bind group layout each time.
Bug: dawn:466
Change-Id: I70a5719265784eb8f60c2eedf6db7596462b21bd
Reviewed-on: https://dawn-review.googlesource.com/c/dawn/+/23980
Reviewed-by: Corentin Wallez <cwallez@chromium.org>
Reviewed-by: Austin Eng <enga@chromium.org>
Commit-Queue: Tomek Ponitka <tommek@google.com>
Also fix a misuse of VkBufferImageCopy that caused some Swiftshader
failures.
Bug: dawn:22
Change-Id: Ie812a590d70c7561dfcf2f78ce6c530187610e65
Reviewed-on: https://dawn-review.googlesource.com/c/dawn/+/23200
Commit-Queue: Corentin Wallez <cwallez@chromium.org>
Reviewed-by: Jiawei Shao <jiawei.shao@intel.com>
This patch moves the computation of TextureBufferCopySplit outside of
the loop over each copy texture array layer so that we only need to do
the computation once for each B2T and T2B copy command.
BUG=dawn:453
TEST=dawn_end2end_tests
Change-Id: I1fcfc972504f845467af92a77bc37870ed7b52a7
Reviewed-on: https://dawn-review.googlesource.com/c/dawn/+/23720
Commit-Queue: Jiawei Shao <jiawei.shao@intel.com>
Reviewed-by: Austin Eng <enga@chromium.org>
Reviewed-by: Corentin Wallez <cwallez@chromium.org>
Roll third_party/glslang/ 00965b9fa..839704450 (14 commits)
00965b9fa9..8397044502
$ git log 00965b9fa..839704450 --date=short --no-merges --format='%ad %ae %s'
2020-06-22 gleese Update test expected files with new magic number
2020-06-22 gleese Update SPIR-V generator version
2020-06-05 gleese Update test results to expect OpFUnordNotEqual
2020-06-05 gleese Use OpFUnordNotEqual for floating-point !=
2020-06-22 johnkslang Update README.md
2020-06-19 bclayton Add kokoro configs for android-ndk and cmake
2020-06-19 bclayton Switch ndk_test from gnustl_static to c++_static
2020-06-17 ShabbyX Add -g0 command line argument
2020-06-16 cepheus Build: use better MSVC subfolder names for the previous build changes.
2020-06-16 cepheus Bump version numbers.
2020-06-16 bclayton Move hlsl/ source to glslang/HLSL/
2020-06-16 cepheus Bump version.
2020-06-15 bclayton CMake: Fold HLSL source into glslang
2020-06-10 bclayton C Interface: Split SPIR-V C interface to own file
Roll third_party/shaderc/ c2ac70797..caa519ca5 (2 commits)
c2ac70797a..caa519ca53
$ git log c2ac70797..caa519ca5 --date=short --no-merges --format='%ad %ae %s'
2020-06-17 rharrison Add support for different default/trunks in roll-deps (#1086)
2020-06-15 rharrison Rolling 5 dependencies (#1085)
Roll third_party/spirv-cross/ 9e3df69d4..f9ae06512 (12 commits)
9e3df69d4e..f9ae06512e
$ git log 9e3df69d4..f9ae06512 --date=short --no-merges --format='%ad %ae %s'
2020-06-22 dsinclair Roll deps and update tests.
2020-06-22 post MSL: Remove the old VertexAttr API.
2020-06-19 cwallez Fix placement of SPIRV_CROSS_DEPRECATED.
2020-06-19 post Fix duplicated initialization for loop variables with initializers.
2020-06-18 post MSL: Add test case for constructing struct with non-value-type array.
2020-06-18 post MSL: Deal with loading non-value-type arrays.
2020-06-18 post MSL: Add tests for array copies in and out of buffers.
2020-06-18 post MSL: Improve handling of array types in buffer objects.
2020-06-18 post Clean up some deprecation warnings when building with Makefile.
2020-06-18 post Remove unused member in MSLShaderInput.
2020-06-13 cdavis MSL: Fix up input variables' vector lengths in all stages.
2020-06-16 post HLSL: Fix texProj in legacy HLSL.
Roll third_party/tint/ b8f493005..0c647a889 (15 commits)
https://dawn.googlesource.com/tint/+log/b8f4930055df..0c647a8896d1
$ git log b8f493005..0c647a889 --date=short --no-merges --format='%ad %ae %s'
2020-06-22 dneto sample: handle SPIR-V assembly input
2020-06-19 dj2 [spirv-writer] Emit logical and and logical or
2020-06-18 dneto [spirv-reader] Remove orphaned TODOs
2020-06-18 dsinclair Add support for sclamp and uclamp GLSL methods.
2020-06-18 dsinclair [spirv-writer] Add as casts.
2020-06-18 dneto [spirv-reader] Add ConvertFToU, ConvertFToS
2020-06-18 dsinclair [wgsl-reader] Fixup off by 1 bug.
2020-06-17 dneto [spirv-reader] Add mixed scalar/vector/matrix multiply
2020-06-17 dneto [spirv-reader] Add ConvertSToF, ConvertUToF
2020-06-17 dj2 Add GLSL MatrixInverse type determination.
2020-06-17 dneto [spirv-reader] Add vector shuffle
2020-06-17 dj2 Add type determination for GLSL determinant.
2020-06-17 dneto [spirv-reader] Add OpCopyObject
2020-06-17 dj2 [spirv-writer] Only extract composites for non-const constructors.
2020-06-16 dneto [spirv-reader] Add ArrayStride
Change-Id: I74d332c0972b2e6e0f4b6cc349d54ea5887b051c
Reviewed-on: https://dawn-review.googlesource.com/c/dawn/+/23661
Reviewed-by: Ryan Harrison <rharrison@chromium.org>
Reviewed-by: Austin Eng <enga@chromium.org>
Reviewed-by: Corentin Wallez <cwallez@chromium.org>
Commit-Queue: Ryan Harrison <rharrison@chromium.org>
This patch adds the support of copying with multiple texture array
layers in one buffer-to-texture and texture-to-buffer copy command.
BUG=dawn:453
TEST=dawn_end2end_tests
Change-Id: If009dbb29f2b0ef0667715eed0d66053b1491fd4
Reviewed-on: https://dawn-review.googlesource.com/c/dawn/+/23248
Reviewed-by: Corentin Wallez <cwallez@chromium.org>
Commit-Queue: Jiawei Shao <jiawei.shao@intel.com>
Bug: dawn:442
Change-Id: I889a943cbaf2d349c31a15fdf126d66964bdd0a7
Reviewed-on: https://dawn-review.googlesource.com/c/dawn/+/23247
Commit-Queue: Austin Eng <enga@chromium.org>
Reviewed-by: Stephen White <senorblanco@chromium.org>
This patch fixes a bug in buffer-to-texture and texture-to-buffer copies
when copySize.height == 0 or copySize.depth == 0. Previously we miss the
checks if either copySize.height or copySize.depth is 0 before doing
(copySize.height / blockHeight - 1) and (copySize.depth - 1) thus we
will get incorrect results because of arithmetic overflows.
This patch fixes this bug by adding the missed check and adds the
related regression tests in dawn_unittests
BUG=dawn:453
TEST=dawn_unittests
Change-Id: I970e420c0fa7f0b61c656365bef079c123a59e6a
Reviewed-on: https://dawn-review.googlesource.com/c/dawn/+/23520
Commit-Queue: Kai Ninomiya <kainino@chromium.org>
Reviewed-by: Austin Eng <enga@chromium.org>
Reviewed-by: Kai Ninomiya <kainino@chromium.org>
|MTLDevice maxBufferLength| is not available until 10.14, so on
lower versions of macOS, try using |recommendedMaxWorkingSetSize|
Bug: dawn:450
Change-Id: I52dc4211cf3d7014771d580a9af9c72abe92a375
Reviewed-on: https://dawn-review.googlesource.com/c/dawn/+/23263
Commit-Queue: Austin Eng <enga@chromium.org>
Reviewed-by: Stephen White <senorblanco@chromium.org>
ASAN throws when hitting OOM conditions which these tests intentionally
try to create.
Bug: dawn:450
Change-Id: I635480ae329cd5ace55287d77e2b79e663174f49
Reviewed-on: https://dawn-review.googlesource.com/c/dawn/+/23246
Commit-Queue: Austin Eng <enga@chromium.org>
Reviewed-by: Corentin Wallez <cwallez@chromium.org>
This CL adds support for multisampled sampling on the D3D12 backend.
This was already working on other backends. It also adds tests that all
of the sample locations are correct.
Bug: dawn:431
Change-Id: I6849e5e2d708ad4824e6db2665d668d43a4ef5ea
Reviewed-on: https://dawn-review.googlesource.com/c/dawn/+/23245
Commit-Queue: Austin Eng <enga@chromium.org>
Reviewed-by: Corentin Wallez <cwallez@chromium.org>
Change mTailPtr to mCurrent that's a VkBaseOutStructure*.
Bug: dawn:464
Change-Id: Ic10a0abc95e279e1537786601d006cf18305687a
Reviewed-on: https://dawn-review.googlesource.com/c/dawn/+/23540
Reviewed-by: Stephen White <senorblanco@chromium.org>
Reviewed-by: Austin Eng <enga@chromium.org>
Commit-Queue: Corentin Wallez <cwallez@chromium.org>
All the querying of extensions was correctly handled for the WebGPU
ShaderFloat16 extensions, except that we forgot to request it when
creating the VkDevice.
In practice this doesn't crash as most drivers won't check if the
extension is enabled, and just compile the shader code correctly.
However the Vulkan Validation Layers complain. This CL fixes it by
requesting the features on device creation.
Bug: dawn:426
Change-Id: Id7307bb4f1d08a10bec9fc7bf0500c5761f96ff7
Reviewed-on: https://dawn-review.googlesource.com/c/dawn/+/23203
Commit-Queue: Corentin Wallez <cwallez@chromium.org>
Reviewed-by: Austin Eng <enga@chromium.org>
Reviewed-by: Stephen White <senorblanco@chromium.org>
TypedInteger.h didn't include <functional> which made the reference to
std::hash cause a compilation failure. Fix this by moving TypedInteger
hashing to HashUtils.h so that TypedInteger doesn't include <functional>
in all of its users.
Bug: dawn:442
Change-Id: I5a0271bb187682616eb5ef3a13bb7f6271203453
Reviewed-on: https://dawn-review.googlesource.com/c/dawn/+/23440
Reviewed-by: Corentin Wallez <cwallez@chromium.org>
Commit-Queue: Corentin Wallez <cwallez@chromium.org>
tint has moved to using origin/main, and shaderc is soon to migrate
Change-Id: I578e4df91330351e10513874b2ca2728e4227803
Reviewed-on: https://dawn-review.googlesource.com/c/dawn/+/23320
Reviewed-by: Corentin Wallez <cwallez@chromium.org>
Reviewed-by: Austin Eng <enga@chromium.org>
Commit-Queue: Ryan Harrison <rharrison@chromium.org>
This CL adds a TypedInteger helper which provides additional type
safety in Dawn. It is a compile-time restriction that prevents integers
of different types from being used interchangably in Debug builds.
It also adds ityp::{array,bitset,span} as helper classes to wrap std::
versions (not span). These accept a template paramter as the Index type
so that typed integers, or enum classes, may be used as a type-safe
index.
For now, bind group layout binding indices use TypedInteger. Future
CLs will convert other indices to be type-safe as well.
Bug: dawn:442
Change-Id: I5b63b1e4f6154322db0227a7788a4e9b8303410e
Reviewed-on: https://dawn-review.googlesource.com/c/dawn/+/19902
Reviewed-by: Austin Eng <enga@chromium.org>
Commit-Queue: Austin Eng <enga@chromium.org>
Device shouldn't be used until after DeviceBase::Initialize so we need
to move the NextSerial call to after Device Initialization is completed.
Bug: chromium:1094477
Change-Id: I2ab317c1d75f853c022faac3fc0bbd66fbd7c14e
Reviewed-on: https://dawn-review.googlesource.com/c/dawn/+/23261
Reviewed-by: Austin Eng <enga@chromium.org>
Commit-Queue: Natasha Lee <natlee@microsoft.com>
Retrieval filter was receiving errors that shouldn't be
stored or not storing new errors. Since these errors are
never reaching the storage filter or being stored without
being filtered, this workaround removes the storage
filter entirely and filters them upon being retrieved.
After the change, two new E2E tests fail
backend validation and require further investigation.
Bug: dawn:460
Change-Id: I92d8c55c71832064b94e8ff0307e7af57ea81fda
Reviewed-on: https://dawn-review.googlesource.com/c/dawn/+/23144
Reviewed-by: Rafael Cintron <rafael.cintron@microsoft.com>
Reviewed-by: Austin Eng <enga@chromium.org>
Commit-Queue: Bryan Bernhart <bryan.bernhart@intel.com>
Fuzzer recently started sometimes injecting errors with
CheckHRESULT returning E_FAKE_ERROR_FOR_TESTING, which needs to also be ignored
in AssertAndIgnoreDeviceLossError so that we can continue with destruction.
Rename AssertAndIgnoreDeviceLossError to IgnoreErrors
This CL also adds Austin's pretty printing of HRESULT results as hex values
and adds printing of "E_FAKE_ERROR_FOR_TESTING" error.
Bug: chromium:1094448
Change-Id: I6715a5eec02919a3e3b4c86d493b3f7e84c5e206
Reviewed-on: https://dawn-review.googlesource.com/c/dawn/+/23241
Commit-Queue: Natasha Lee <natlee@microsoft.com>
Reviewed-by: Rafael Cintron <rafael.cintron@microsoft.com>
Reviewed-by: Corentin Wallez <cwallez@chromium.org>
Reviewed-by: Kai Ninomiya <kainino@chromium.org>
When the fuzzer passes the command DeviceLoseForTesting, the
DawnWireServerFuzzer will hang forever when waiting for a fence signal
because the device was lost. This commit sets a callback on device loss
which gives an extra condition to break the hang.
Bug: dawn:444
Change-Id: I0157358ed6da39cc6fcab7e1be8a3d5fde74266f
Reviewed-on: https://dawn-review.googlesource.com/c/dawn/+/23141
Reviewed-by: Corentin Wallez <cwallez@chromium.org>
Reviewed-by: Natasha Lee <natlee@microsoft.com>
Commit-Queue: Brandon Jones <brandon1.jones@intel.com>
This patch does several clean-ups in CopyTests as a preparation of
supporting buffer-to-texture and texture-to-buffer copies with multiple
texture array layers.
1. Remove the zero-initialization of textures in CopyTests_B2T
2. Use a big buffer that contains the pixel data in all the texture
array layers instead of several small ones that only contains one
texture array layer.
3. Make CopyTests_T2T::DoTest() also use CopyTests::TextureSpec as its
input parameter.
4. Add tests for the copies with multiple texture array layers. In this
patch we just copy the data once per texture array layer, while in
the next patch we will be able to copy the data for all the related
texture array layers in one copy command.
5. Move some helper functions for copy tests to WGPUHelpers.cpp
BUG=dawn:453
TEST=dawn_end2end_tests
Change-Id: Ia0150a6c55fffce3e24dda2cd45fd920b0873dc9
Reviewed-on: https://dawn-review.googlesource.com/c/dawn/+/23120
Commit-Queue: Corentin Wallez <cwallez@chromium.org>
Reviewed-by: Austin Eng <enga@chromium.org>
Reland after a refactor of how the extension handling work in the Vulkan
backend.
The original author is David Turner <david.turner.dev@gmail.com>.
This CL ensures that, on architectures with a varying subgroup size,
compute shaders are always compiled with a fixed subgroup size to
avoid consistency issues when one shader writes data in a subgroup-size
dependent layout to GPU memory, to be read by another shader in a
future dispatch.
At the moment, only Intel ICDs are known to implement this [1],
and the code uses a heuristics to chose the size of 16, which seems to
be the sweet spot according to Intel engineers.
+ Update the PNextChainBuilder class to deal with the fact that
VkComputePipelineCreateInfo::pNext is defined as a const void*,
which created compiler errors in the previous implementation.
[1] https://bugs.freedesktop.org/show_bug.cgi?id=108875
Bug: dawn:464
Change-Id: I035ee06084fcc964742f0bff4c54cff257c742ae
Reviewed-on: https://dawn-review.googlesource.com/c/dawn/+/23202
Commit-Queue: Corentin Wallez <cwallez@chromium.org>
Reviewed-by: Austin Eng <enga@chromium.org>
Reviewed-by: Stephen White <senorblanco@chromium.org>
Reland after a refactor of how the extension handling work in the Vulkan
backend.
The original author is David Turner <david.turner.dev@gmail.com>.
Certain Vulkan ICDs (Intel ones notably) will compile SPIR-V
shaders with an liberal, compiler-selected, subgroup size (i.e.
either 8, 16 or 32). For more context, see [1].
This can be a problem for compute, when one shader stores data
in device memory using a subgroup-size dependent layout, to be
consumed by a another shader. Problems arise when the compiler
decides to compile both shaders with different subgroup sizes.
To work-around this, the VK_EXT_subgroup_size_control device
extension was introduced recently: it allows the device to
report the min/max subgroup sizes it provides, and allows
the Vulkan program to control the subgroup size precisely
if it wants to.
This patch adds support to the Vulkan backend to report and
enable the extension if it is available. Note that:
- This changes the definition of VulkanDeviceKnobs to
make room for the required pNext-linked chains of
extensions.
- A helper class, PNextChainBuilder is also provided in
UtilsVulkan.h to make it easy to build pNext-linked
extension struct chains at runtime, as required when
probing device propertires/features, or when
creating a new VkDevice handle.
- This modifies VulkanDeviceInfo::GatherDeviceInfo() to
use PNextChainBuilder to query extension features and properties
in a single call to vkGetPhysicalDevice{Properties,Features}2.
Apart from that, there is no change in behaviour in this CL.
I.e. a later CL might force a specific subgroup size for
consistency, or introduce a new API to let Dawn clients
select a fixed subgroup size.
[1] https://bugs.freedesktop.org/show_bug.cgi?id=108875
Bug: dawn:464
Change-Id: I01e5c28e7dac66f0a57bf35532eb192912b254fa
Reviewed-on: https://dawn-review.googlesource.com/c/dawn/+/23201
Reviewed-by: Corentin Wallez <cwallez@chromium.org>
Commit-Queue: Corentin Wallez <cwallez@chromium.org>
Hao Li
Idan Raiter
Austin Eng
Corentin Wallez
Tomek Ponitka
Jiajie Hu
Bryan Bernhart
Yunchao He
Jiawei Shao
Brandon Jones
Ryan Harrison
Natasha Lee