[ExecuTorch][WebGPU] Switch native backend from wgpu-native to Dawn (Tint) + SwiftShader#20192
Merged
Conversation
…Tint) + SwiftShader Pull Request resolved: #20079 Make Dawn (Chrome's WebGPU implementation, whose WGSL compiler Tint is the spec reference) running on SwiftShader the sole native WebGPU backend, replacing wgpu-native (naga), so the op tests run on a spec-faithful, headless, deterministic CLI backend. The `WEBGPU_IMPL` cache variable, the wgpu-native CMake branch, and the `WEBGPU_IMPL_DAWN` compile define are removed -- CMake now unconditionally `find_package(Dawn REQUIRED)` and links `dawn::webgpu_dawn`. `WebGPUCompat.h` drives pending callbacks via Dawn's `wgpuInstanceProcessEvents` on native and yields to the JS event loop under Emscripten. Dawn is vendored with NO new S3 artifact: `oss/.ci/scripts/setup-webgpu-linux-deps.sh` downloads Google's official `ubuntu-latest-Release` prebuilt directly from github.com/google/dawn/releases (pinned tag + sha256, the same pattern as `setup-wgpu-native.sh`), and reuses the SwiftShader prebuilt already on the ossci bucket. The release exports `dawn::webgpu_dawn` (a static lib) which drops into the existing `find_package(Dawn)`. It has no bundled SwiftShader, so `WebGPUDevice.cpp` requests a normal Vulkan adapter (`forceFallbackAdapter=false`) and `VK_ICD_FILENAMES` makes SwiftShader the only device. The release is built with a recent GCC, so the deps script also pulls a current libstdc++ from the `ubuntu-toolchain-r` PPA (its lib references `_M_replace_cold`, a GCC 13+ symbol) plus `libvulkan1` (Dawn dlopens the Vulkan loader) -- all scoped to the WebGPU CI job, backward-compatible, no repo-wide impact. Authored with assistance from Claude. ghstack-source-id: 391968389 @exported-using-ghexport Differential Revision: [D107589774](https://our.internmc.facebook.com/intern/diff/D107589774/)
🔗 Helpful Links🧪 See artifacts and rendered test results at hud.pytorch.org/pr/pytorch/executorch/20192
Note: Links to docs will display an error until the docs builds have been completed. This comment was automatically generated by Dr. CI and updates every 15 minutes. |
meta-cla
Bot
added
the
CLA Signed
This PR needs a
|
![github-advanced-security[bot]](https://avatars.githubusercontent.com/in/57789?s=60&v=4){kind=small}
Comment on lines
+33
to
+54
| uses: pytorch/test-infra/.github/workflows/linux_job_v2.yml@main | ||
| with: | ||
| ref: ${{ github.event_name == 'pull_request' && github.event.pull_request.head.sha || github.sha }} | ||
| runner: linux.4xlarge.memory | ||
| docker-image: ci-image:executorch-ubuntu-22.04-clang12 | ||
| submodules: recursive | ||
| timeout: 120 | ||
| script: | | ||
| set -eux | ||
|
|
||
| # The generic Linux job uses the base conda env, not the image's; activate | ||
| # the image env (it has the pinned from-source torch). Mirrors | ||
| # test-vulkan-operators-linux in pull.yml. | ||
| CONDA_ENV=$(conda env list --json | jq -r ".envs | .[-1]") | ||
| conda activate "${CONDA_ENV}" | ||
|
|
||
| # Install the python package + runtime deps (the .pte exporters). | ||
| PYTHON_EXECUTABLE=python bash .ci/scripts/setup-linux.sh --build-tool cmake | ||
|
|
||
| # Vendor Dawn (Tint) + SwiftShader, then build + run the native executables. | ||
| source .ci/scripts/setup-webgpu-linux-deps.sh | ||
| bash backends/webgpu/scripts/test_webgpu_native_ci.sh |
…tests Pull Request resolved: #20080 Native unit tests for two runtime enablers: per-pass compute-dispatch ordering (D107543258) and graph-owned scratch buffers (D107543259). `test/native/test_dispatch_order.cpp` exercises multi-dispatch read-after-write ordering through a single `execute()` using dependency chains -- a single-input `add` self-chain and a heterogeneous `rms_norm` -> `add` chain, both lowered via `VulkanPartitioner` -- comparing GPU output to a torch-computed golden per element. `test/native/test_scratch_buffer.cpp` is a white-box test of `WebGPUGraph::create_scratch_buffer` (no black-box consumer exists below the SDPA op): allocation + zero-size guard, copy round-trip, a compute Storage round-trip (its actual use), and a create/destroy lifecycle stress. Authored with assistance from Claude. ghstack-source-id: 391979580 @exported-using-ghexport Differential Revision: [D107576199](https://our.internmc.facebook.com/intern/diff/D107576199/)
JulianCloudNTH
approved these changes
Jun 10, 2026
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
3 participants
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
This PR was created by the merge bot to help merge the original PR into the main branch.
ghstack PR number: #20079 by @JulianCloudNTH
^ Please use this as the source of truth for the PR details, comments, and reviews
ghstack PR base: https://github.com/pytorch/executorch/tree/gh/JulianCloudNTH/14/base
ghstack PR head: https://github.com/pytorch/executorch/tree/gh/JulianCloudNTH/14/head
Merge bot PR base: https://github.com/pytorch/executorch/tree/main
Merge bot PR head: https://github.com/pytorch/executorch/tree/gh/JulianCloudNTH/14/orig
@diff-train-skip-merge