The rules have been relaxed about function values always needing to be
consumed, so a bunch of these phony-assignments can be replaced with the
RHS, but this will be done as a follow-up.
Bug: tint:1213
Change-Id: Ie7c4280f87b4ad7e5a429994b0b88ac22c2f3a9f
Reviewed-on: https://dawn-review.googlesource.com/c/dawn/+/67300
Commit-Queue: Ben Clayton <bclayton@google.com>
Commit-Queue: Corentin Wallez <cwallez@chromium.org>
Auto-Submit: Ben Clayton <bclayton@google.com>
Reviewed-by: Corentin Wallez <cwallez@chromium.org>
Prevent reusing offsets from a previous direct draw.
Update test to verify that values are updated correctly
for each draw. Add tests for indirect draw offsets.
Bug: dawn:548
Change-Id: Ice8325a8a41b8a4375767156dbaba3ee3d714f3b
Reviewed-on: https://dawn-review.googlesource.com/c/dawn/+/67121
Reviewed-by: Austin Eng <enga@chromium.org>
Reviewed-by: Corentin Wallez <cwallez@chromium.org>
Commit-Queue: Austin Eng <enga@chromium.org>
This patch implements [[num_workgroups]] on the API side for
Dispatch() calls by setting num_workgroups.xyz as root constants.
This patch also adds a temporary validation that on D3D12 backend
using a compute pipeline with [[num_workgroups]] in a
DispatchIndirect call is not supported.
BUG=dawn:839
TEST=dawn_end2end_tests
Change-Id: Iaee2ffd162e9420e4e80944fbb222f10a4600c6a
Reviewed-on: https://dawn-review.googlesource.com/c/dawn/+/66580
Reviewed-by: Corentin Wallez <cwallez@chromium.org>
Reviewed-by: Austin Eng <enga@chromium.org>
Commit-Queue: Jiawei Shao <jiawei.shao@intel.com>
tested:
- maxComputeWorkgroupStorageSize
- maxUniformBufferBindingSize
- maxStorageBufferBindingSize
Two of these limits are exposed as configurable to the JS API
so it's important they are tested to work before we expose
them. maxUniformBufferBindingSize came along as well because
the test for storageBufferBindingSize was easy to parameterize.
Bug: dawn:685
Change-Id: I08de6df9d70a22aca0f48ac3fef0038f7aec727b
Reviewed-on: https://dawn-review.googlesource.com/c/dawn/+/66480
Reviewed-by: Corentin Wallez <cwallez@chromium.org>
Reviewed-by: Brandon Jones <bajones@chromium.org>
Commit-Queue: Austin Eng <enga@chromium.org>
For size parameter in mapAsync, use wgpu::kWholeMapSize rather than 0 to
indicate using the default size, i.e. remaining buffer size after
offset. Using size=0 is still available but will cause a deprecation
warning.
Bug: dawn:1159
Change-Id: I474d87ecae4a54ceb28d636f883a6233c91f16fa
Reviewed-on: https://dawn-review.googlesource.com/c/dawn/+/66284
Auto-Submit: Zhaoming Jiang <zhaoming.jiang@intel.com>
Commit-Queue: Zhaoming Jiang <zhaoming.jiang@intel.com>
Reviewed-by: Corentin Wallez <cwallez@chromium.org>
Reviewed-by: Austin Eng <enga@chromium.org>
`a && b` only evaluates b if a is true. `a & b` always evaluates
both a and b. If a and b are of type bool, `&&` is usually what you
want, so clang now warns on `&` where both arguments are of type bool.
From what I can tell, in Dawn it wasn't important if we evaluate both
branches or not in the places where this fired, so I went with `&&`
everywhere.
Bug: chromium:1255745
Change-Id: Ifa196a7150d5bdfb8527fe8b6f40d7e2e957d609
Reviewed-on: https://dawn-review.googlesource.com/c/dawn/+/66200
Auto-Submit: Nico Weber <thakis@chromium.org>
Reviewed-by: Corentin Wallez <cwallez@chromium.org>
Commit-Queue: Nico Weber <thakis@chromium.org>
This CL renames "extension" to "feature" to follow WebGPU. It still
supports both. A future Chromium CL will pick this change, then all
"extension" occurrences will be removed.
Change-Id: I070e32d7ae042f9b846df01f200b39f6741a0a14
Bug: dawn:1149
Reviewed-on: https://dawn-review.googlesource.com/c/dawn/+/65664
Reviewed-by: Corentin Wallez <cwallez@chromium.org>
Reviewed-by: Austin Eng <enga@chromium.org>
Commit-Queue: François Beaufort <beaufort.francois@gmail.com>
It wasn't implemented because Discard wasn't implemented for a long
time.
Bug: dawn:1151
Change-Id: Ie26559dd65a2c97a32df418521ca93c1021dd837
Reviewed-on: https://dawn-review.googlesource.com/c/dawn/+/65720
Commit-Queue: Corentin Wallez <cwallez@chromium.org>
Reviewed-by: Brandon Jones <bajones@chromium.org>
Add vulkan backend implementations and tests.
Disabled some tests that fail due to tint missing
features and issues.
Bug: dawn:1041, tint:1155
Change-Id: Iac161317450cff59627e08b1228bffde4cef71da
Reviewed-on: https://dawn-review.googlesource.com/c/dawn/+/64981
Commit-Queue: Shrek Shao <shrekshao@google.com>
Reviewed-by: Corentin Wallez <cwallez@chromium.org>
Breaking change, but it should only require small changes in any project
that relies on it, so just doing this instead of a two-stage deprecation.
Will require a manual roll into (at least) Chromium.
Bug: dawn:1080, chromium:1253721
Change-Id: I6699e0629c3b2fe63e7f9d5ba0a928f00316a588
Reviewed-on: https://dawn-review.googlesource.com/c/dawn/+/64520
Reviewed-by: Austin Eng <enga@chromium.org>
Reviewed-by: Corentin Wallez <cwallez@chromium.org>
For size parameter in setIndexBuffer and setVertexBuffer, use
wgpu::kWholeSize rather than 0 to indicate using the default
size, i.e. remaining buffer size after offset. Using size=0 is still
avaliable but will cause a deprecation warning.
Bug: dawn:1058
Change-Id: I4c4f6ca6fc7346c225815bdcd607fed110ce2b9a
Reviewed-on: https://dawn-review.googlesource.com/c/dawn/+/65020
Commit-Queue: Zhaoming Jiang <zhaoming.jiang@intel.com>
Reviewed-by: Corentin Wallez <cwallez@chromium.org>
Every render pass which invokes DrawIndexedIndirect, either directly or
through a RenderBundle execution, is now preceded immediately by at
least one validation pass.
All indirect buffer offests used with DII are validated, and their
validated values are copied into a separate scratch buffer (or zeroed
out there, in the case of validation failure). All encoded DII commands
are rewritten to use the validated parameters instead of the original
ones.
Bug: dawn:809
Change-Id: I5eead937f19536f84f89e2c8e6fed7f18f0aee9f
Reviewed-on: https://dawn-review.googlesource.com/c/dawn/+/63461
Commit-Queue: Ken Rockot <rockot@google.com>
Reviewed-by: Austin Eng <enga@chromium.org>
From macOS 11+, Metal requires the minimum buffer offset alignment to be
256 bytes. We have updated the destinationOffset alignment of
resolveQuerySet to 256 in SPEC. This PR is changing the implementation.
Bug: dawn:940
Change-Id: Ie3c69d6a90adb76038dbac95ae2158a38da583f3
Reviewed-on: https://dawn-review.googlesource.com/c/dawn/+/63762
Commit-Queue: Hao Li <hao.x.li@intel.com>
Reviewed-by: Austin Eng <enga@chromium.org>
This command copies data from host memory into a GPU buffer. It's
analogous to Queue::WriteBuffer, but executed in the context of a
command buffer, sequenced with other encoded commands. This is useful
for supporting a notion of a shared scratch buffer, with a single
allocation whose contents may need to be overwritten with new data
before each pass that uses it.
Bug: dawn:809
Change-Id: If58d49c52a41127e2980dd626fd687eb1c91fe28
Reviewed-on: https://dawn-review.googlesource.com/c/dawn/+/64001
Reviewed-by: Austin Eng <enga@chromium.org>
Reviewed-by: Corentin Wallez <cwallez@chromium.org>
Commit-Queue: Ken Rockot <rockot@google.com>
This patch implements the asynchronous path of CreateRenderPipelineAsync
on D3D12 backend.
1. Call the constructor of dawn_native::d3d12::RenderPipeline in main
thread.
2. Execute dawn_native::RenderPipelineBase::Initialize() (a virtual function)
asynchronously.
3. Ensure every operation in dawn_native::d3d12::RenderPipeline::Initialize()
is thread-safe.
4. Save all the return values (pipeline object or error message, userdata, etc)
in a CreateRenderPipelineAsyncWaitableCallbackTask object and insert this
callback task into CallbackTaskManager.
5. In Callback.Finish():
- Insert the pipeline object into the pipeline cache if necessary
- Call WGPUCreateRenderPipelineAsyncCallback
This patch also removes FlatRenderPipelineDescriptor as it is not needed
right now.
BUG=dawn:529
TEST=dawn_end2end_tests
Change-Id: I7fd30339ab7bea599c483dea4bd1100359982409
Reviewed-on: https://dawn-review.googlesource.com/c/dawn/+/64440
Commit-Queue: Jiawei Shao <jiawei.shao@intel.com>
Reviewed-by: Austin Eng <enga@chromium.org>
Tint passes threadgroup memory in MSL as entrypoint arguments since
threadgroup memory at the module scope cannot be default initialized.
MSL lacks default constructors for matrices in threadgroup memory.
Bug: dawn:1110
Change-Id: I7462fa448c6ebdb3cc4dc24bd5ff0a99287cdba0
Reviewed-on: https://dawn-review.googlesource.com/c/dawn/+/64240
Reviewed-by: James Price <jrprice@google.com>
Commit-Queue: Austin Eng <enga@chromium.org>
Support vertex-only render pipeline on D3D12, Vulkan, Metal, OpenGL
and OpenGL ES backends. Related validation tests and end to end tests
are also implemented.
Bug: dawn:136
Change-Id: If266fd441c1d39ccd940ea26b74b405f8abb351a
Reviewed-on: https://dawn-review.googlesource.com/c/dawn/+/63080
Commit-Queue: Zhaoming Jiang <zhaoming.jiang@intel.com>
Reviewed-by: Austin Eng <enga@chromium.org>
Suppresses LoadOpClearIntegerFormatsToLargeValues on Mac 11.5 with
Metal on Intel GPUs.
Bug: dawn:1109
Change-Id: Ia4316ee00ad790752fa3b5d1778d885c5eb3a188
Reviewed-on: https://dawn-review.googlesource.com/c/dawn/+/64160
Commit-Queue: Brian Sheedy <bsheedy@google.com>
Commit-Queue: Austin Eng <enga@chromium.org>
Auto-Submit: Brian Sheedy <bsheedy@google.com>
Reviewed-by: Austin Eng <enga@chromium.org>
Adds a supression to the failing test.
BUG=dawn:1106
Change-Id: Ib0cfd49b6d1af59c60d7bc4f2e3e1bec5ae80601
Reviewed-on: https://dawn-review.googlesource.com/c/dawn/+/64062
Reviewed-by: Austin Eng <enga@chromium.org>
Commit-Queue: Bryan Bernhart <bryan.bernhart@intel.com>
- We use Tint unconditionally so binding reflection is now correct.
- WGSL disallows textureSampleCompare in compute shaders and is not
likely to change.
- We now support reading back stencil in tests.
Bug: dawn:439, dawn:667, dawn:216, dawn:704
Change-Id: I7ba0026bb906ce985abb495581b56977da7be2da
Reviewed-on: https://dawn-review.googlesource.com/c/dawn/+/63680
Reviewed-by: Brandon Jones <bajones@google.com>
Commit-Queue: Corentin Wallez <cwallez@chromium.org>
We only remap bindings for the target entry point, so we need to strip
all other entry points to avoid generating invalid bindings for them.
Bug: tint:1170
Change-Id: Ia1a73601e8d620341fb4f7170dfa856632a04245
Reviewed-on: https://dawn-review.googlesource.com/c/dawn/+/63880
Commit-Queue: James Price <jrprice@google.com>
Auto-Submit: James Price <jrprice@google.com>
Reviewed-by: Austin Eng <enga@chromium.org>
This patch implements the struct FlatRenderPipelineDescriptor to
save all the pointer members of RenderPipelineDescriptor in
CreateRenderPipelineAsync so that the render pipeline descriptor
is always valid when the render pipeline is created asynchronously.
BUG=dawn:529
TEST=dawn_end2end_tests
Change-Id: I99e06581f84c52d484f877ba29e8cc1ca50d68b2
Reviewed-on: https://dawn-review.googlesource.com/c/dawn/+/63368
Reviewed-by: Austin Eng <enga@chromium.org>
Commit-Queue: Jiawei Shao <jiawei.shao@intel.com>
When creating a new render pass in D3D12, call the OMSetStencilRef
to set the stencil reference value to zero.
Bug: dawn:1097
Change-Id: I5dd94691b1b354b601c06a02d3d5fa619d8b58ee
Reviewed-on: https://dawn-review.googlesource.com/c/dawn/+/63360
Commit-Queue: Jiawei Shao <jiawei.shao@intel.com>
Reviewed-by: Jiawei Shao <jiawei.shao@intel.com>
Reviewed-by: Austin Eng <enga@chromium.org>
The end to end texture tests with the binary information for the textures are generated using the open source astc-encoder tool.
Also: Refactors end2end compressed texture tests to allow for non-4x4 texel compression formats (ASTC formats are not all 4x4).
Bug: dawn:955
Change-Id: I5311a5aa7bd017de72300bf7c87fba6c869b1db5
Reviewed-on: https://dawn-review.googlesource.com/c/dawn/+/63121
Commit-Queue: Loko Kung <lokokung@google.com>
Reviewed-by: Austin Eng <enga@chromium.org>
This patch fixes a use-after-free issue in FlatComputePipelineDescriptor
by always making its pointer members point to its own members.
BUG=dawn:529, chromium:1246158, chromium:1246748
TEST=dawn_end2end_tests
Change-Id: I49b98357444edcdb1b225e961d1e4d6e5b11e978
Reviewed-on: https://dawn-review.googlesource.com/c/dawn/+/63366
Commit-Queue: Jiawei Shao <jiawei.shao@intel.com>
Reviewed-by: Austin Eng <enga@chromium.org>
Fixes a crash found while fuzzing by returning early when SetDebugName
is called with an empty D3D12/Vulkan object or when the device is lost.
Adds tests.
Bug: chromium:1245720
Change-Id: Ie443f690cf5635d017295a13e21c33dd2e97dc46
Reviewed-on: https://dawn-review.googlesource.com/c/dawn/+/63363
Reviewed-by: Austin Eng <enga@chromium.org>
Commit-Queue: Brandon Jones (Intel) <brandon1.jones@intel.com>
Prevents bind group layouts created by a default pipeline layout from
being reusable with any other pipelines or layouts, as detailed in
https://github.com/gpuweb/gpuweb/pull/2068
Change-Id: Ic398eb6c6e089ac63ce6650f125dd20a9dfc8862
Bug: dawn:1094
Reviewed-on: https://dawn-review.googlesource.com/c/dawn/+/63220
Commit-Queue: Brandon Jones <bajones@chromium.org>
Reviewed-by: Corentin Wallez <cwallez@chromium.org>
The end to end texture tests with the binary information for the textures are generated with a combination of the open source etc2comp tool and manual crafting of the data based on the specification in the Kronos Data Format Specifications (https://www.khronos.org/registry/DataFormat/specs/1.1/dataformat.1.1.pdf) for the compressed formats.
Also: Refactors end2end compressed texture tests and some utilities so that we can use parameterized tests for easier new format additions.
Bug: dawn:955
Change-Id: I91686f26a693a85cb076ef0eca3029111a64c50f
Reviewed-on: https://dawn-review.googlesource.com/c/dawn/+/62740
Commit-Queue: Loko Kung <lokokung@google.com>
Reviewed-by: Corentin Wallez <cwallez@chromium.org>
Buffer allocations in Dawn may be padded. This padding
could be visible, depending on how the backend
implements robust vertex buffer access. This commit
updates buffer creation to clear all padding bytes
immediately after creation.
It is not counted as a lazy clear.
And, add a test which reads off the end of a padded
vertex buffer to check that the padding bytes are also
initialized.
Also: Update Metal buffers to always allocate enough
space for Tint's vertex pulling, and pass the full
allocated size to Tint. While writing the test in this
commit, a bug was found where Tint assumes there is at
least 4 bytes in the vertex pulling buffer. The WebGPU
API currently allows zero-sized bindings, so Dawn needs
to always allocate enough space for one element.
Also: Update Vulkan vertex/index buffers to allocate
at least one more byte than requested because
vkCmdSetVertexBuffers and vkCmdSetIndexBuffer disallow
the offset to be equal to the buffer size. We need
at least one more byte to make zero-sized bindings
as the end of the buffer valid.
Lastly: Update helper so that a T2B copy of a single
row that touches all bytes in a buffer updates the
buffer state to be fully initialized. This is a small
optimization that is necessary to write the test.
Bug: dawn:837
Change-Id: I829f4764509c4ec784c5aeaaf40b6dcbd4be6866
Reviewed-on: https://dawn-review.googlesource.com/c/dawn/+/62161
Commit-Queue: Austin Eng <enga@chromium.org>
Reviewed-by: Kai Ninomiya <kainino@chromium.org>
Reviewed-by: Corentin Wallez <cwallez@chromium.org>
Currently, the bind group layout tightly packs the shader registers for
each of the sampler and non-sampler descriptors sets. This reduces the
max shader register used and helps targeting shader model 5.0, which has
a relatively low max slot count per resource. It is safe in D3D, since a
shader register collision can be valid if the descriptor types differ.
To support Mesa's SPIR-V to DXIL compiler, we need to avoid possible
collisions between resources' shader registers because it uses SPIR-V as
an intermediate representation (which does not namespace bindings by
type). This change re-works BindGroupLayoutD3D12 to not assume the
resulting shader registers will be tightly packed and group per
descriptor type.
Change-Id: I0bb51106c4683bfe02ce15ecad71716734b7a91f
Reviewed-on: https://dawn-review.googlesource.com/c/dawn/+/60764
Reviewed-by: Corentin Wallez <cwallez@chromium.org>
Commit-Queue: Michael Tang <tangm@microsoft.com>
Maintains a 11on12 resource cache on the external image
to prevent re-creating the wrapped resource then
flushing it per produced texture. To prevent
unbounded growth, a basic cap is used which gets
cleared.
This change fixes signficant CPU time spent in the WebGPU command
decoder for video import workloads and excessive memory overhead
from swap chain buffers.
Fixed: dawn:625
Change-Id: I72c07b02f6ab6877a9f21758650962c895933bf9
Reviewed-on: https://dawn-review.googlesource.com/c/dawn/+/51421
Commit-Queue: Bryan Bernhart <bryan.bernhart@intel.com>
Reviewed-by: Corentin Wallez <cwallez@chromium.org>
This is a reland of 365d40b93a
Original change's description:
> Making D3D12 accept InternalUsageDescriptors
>
> This is a follow up from two previous CLs,
> https://dawn-review.googlesource.com/c/dawn/+/58440
> and
> https://dawn-review.googlesource.com/c/dawn/+/58140
>
> This CL adds in the D3D12Backend the possibility to use Dawn Texture
> Internal Usage Descriptor, and adds a series of tests to validate it.
>
> Change-Id: I94f6b39153b60ec8af61bda22d41a865898da613
> Bug: dawn:1027, dawn:1052
> Reviewed-on: https://dawn-review.googlesource.com/c/dawn/+/61520
> Commit-Queue: Juanmi Huertas <juanmihd@chromium.org>
> Reviewed-by: Kai Ninomiya <kainino@chromium.org>
Bug: dawn:1027, dawn:1052
Change-Id: I0d35cbf4dc1c6dc80e1a5dfb4770a5383ef543fa
Reviewed-on: https://dawn-review.googlesource.com/c/dawn/+/61840
Reviewed-by: Austin Eng <enga@chromium.org>
Commit-Queue: Juanmi Huertas <juanmihd@chromium.org>
Dawn only reapplies bind groups that have changed on the layout
since the last draw. However, the D3D12 backend was always resetting
the root signature state upon switching pipelines. This led to a
bug where the root signature could be changed and dirtied without
reapplying the dirty bind groups.
Moving application of the root signature state to the same loop
that applies bind groups helps ensure the state stays in sync.
Fixed: dawn:1055
Change-Id: Iae89088560e83f6104c921d42de27c03095d654f
Reviewed-on: https://dawn-review.googlesource.com/c/dawn/+/61582
Commit-Queue: Austin Eng <enga@chromium.org>
Reviewed-by: Bryan Bernhart <bryan.bernhart@intel.com>
Older names are kept around as an alias for a while. Unfortunately we
have no mechanism for producing deprecation error messages when they
are used.
Bug: dawn:1035
Change-Id: Ic6716fd526ecbedaa8e7925ab93e3ff32f642d97
Reviewed-on: https://dawn-review.googlesource.com/c/dawn/+/61382
Reviewed-by: Austin Eng <enga@chromium.org>
Reviewed-by: Kai Ninomiya <kainino@chromium.org>
Commit-Queue: Brandon Jones <bajones@chromium.org>
This CL adds an internal method to copy textures, that will disregard
the need of having CopySrc usage.
This CL adds an intermediate helped method to be used by both
APIContextTextureToTexture and APIContextTextureToTextureInternal.
Fixed: dawn:1052
Change-Id: Icd28e0ef58ecfaa412eefe8c95e41cd2298a9acc
Reviewed-on: https://dawn-review.googlesource.com/c/dawn/+/58440
Reviewed-by: Kai Ninomiya <kainino@chromium.org>
Commit-Queue: Juanmi Huertas <juanmihd@chromium.org>
Ben Clayton
Aleksi Sapon
Jiawei Shao
Austin Eng
Zhaoming Jiang
Corentin Wallez
Nico Weber
François Beaufort
shrekshao
Kai Ninomiya
Ken Rockot
Li Hao
Brandon Jones
Brian Sheedy
Bryan Bernhart
James Price
Loko Kung
Brandon Jones
Michael Tang
Juanmi