[ExecuTorch][WebGPU] Remove the orphan q4gsw_linear_coop4 shader

[JCNTH]

Stack from ghstack (oldest at bottom):

• [ExecuTorch][WebGPU] Test: rms_norm workgroup size is runtime-configurable #20956
• -> [ExecuTorch][WebGPU] Remove the orphan q4gsw_linear_coop4 shader #20955
• [ExecuTorch][WebGPU] Make shader workgroup sizes runtime-configurable #20954

q4gsw_linear_coop4.wgsl is an unwired dead shader: no handler #includes its generated q4gsw_linear_coop4_wgsl.h, and it is superseded by q4gsw_linear_coop4_bicol.wgsl (the live GEMV decode kernel that QuantizedLinear.cpp uses). Remove the shader and its generated header.

Key changes:

• Deleted q4gsw_linear_coop4.wgsl and q4gsw_linear_coop4_wgsl.h.
• Dropped a stale reference to the removed kernel in test_quantized_linear.py and test_webgpu_native.cpp comments.

No build/CMake change: WEBGPU_SRCS lists only .cpp, and shaders wire via the gen_wgsl_headers.py glob (no static list), so removing both files leaves codegen drift-clean.

Co-authored-with: Claude Code.
@exported-using-ghexport

Differential Revision: D112060303

Differential Revision: D112060303

[pytorch-bot]

🔗 Helpful Links

🧪 See artifacts and rendered test results at hud.pytorch.org/pr/pytorch/executorch/20955

• 📄 Preview Python docs built from this PR

Note: Links to docs will display an error until the docs builds have been completed.

❌ 1 New Failure

As of commit 741c146 with merge base e500fee ():

NEW FAILURE - The following job has failed:

• pull / test-multimodal-linux (gemma3-4b) / linux-job (gh)
  RuntimeError: Command docker exec -t 63fe8e374fb4551159dc2f082d0113859466953bbf1967825a205d2e4f3dac10 /exec failed with exit code 139

This comment was automatically generated by Dr. CI and updates every 15 minutes.

[github-actions]

This PR needs a release notes: label

If your change should be included in the release notes (i.e. would users of this library care about this change?), please use a label starting with release notes:. This helps us keep track and include your important work in the next release notes.

To add a label, you can comment to pytorchbot, for example
@pytorchbot label "release notes: none"

For more information, see
https://github.com/pytorch/pytorch/wiki/PyTorch-AutoLabel-Bot#why-categorize-for-release-notes-and-how-does-it-work.

[JCNTH]

@claude Review this PR

[claude]

Claude finished @JCNTH's task in 59s —— View job

---

Review — Remove orphan q4gsw_linear_coop4 shader

• Gather context (PR diff, base branch)
• Verify the shader is truly orphaned (no #include of generated header)
• Verify the superseding kernel (_bicol) is the live one
• Check test comment / native test edits
• Confirm codegen wiring (glob, not static list)

Verdict: LGTM. This is a clean, correct dead-code removal. Every claim in the PR description is verifiable.

What I verified

1. The shader is genuinely orphaned. Grepping the whole tree for q4gsw_linear_coop4_wgsl.h (the generated header for the deleted shader) returns zero includes. Nothing consumes it.

2. _bicol is the live kernel. QuantizedLinear.cpp:13 includes q4gsw_linear_coop4_bicol_wgsl.h — a distinct file that is not touched by this PR. The naming is close enough to be a footgun, but they are two separate shaders and only the non-_bicol one is removed. Good.

3. Codegen wiring is glob-based, so no static list to update. gen_wgsl_headers.py:496 discovers shaders via (BACKEND_ROOT / "runtime/ops").glob("**/*.wgsl") — no hardcoded manifest. CMakeLists.txt:28 WEBGPU_SRCS lists only .cpp files, and the --check step (line 64) verifies committed _wgsl.h match their .wgsl. Deleting both the .wgsl and its _wgsl.h together keeps the drift check clean, as the description states.

4. Comment edits are accurate. Both test_quantized_linear.py and test_webgpu_native.cpp just drop the parenthetical "(mirrors coop4's general-N robustness)" from the bicol odd-N guard comment. The surrounding sentence still reads correctly and the guard behavior it describes is unchanged.

Notes

• The test_webgpu_native.cpp edit isn't mentioned in the PR description's "Key changes" list (which only cites the .py comment), but the change itself is the same benign comment cleanup, so no action needed — just worth flagging for the description.
• The reported CI failure (pull / android / run-emulator) is unrelated to this change — no Android or emulator surface is touched by deleting a WebGPU compute shader.

Nice tightly-scoped cleanup.
· branch gh/JCNTH/87/head