[ExecuTorch][WebGPU] Add permute_copy + IntList graph support (aten.permute_copy.default)

[JulianCloudNTH]

Stack from ghstack (oldest at bottom):

• [ExecuTorch][WebGPU] cat op test suite (cases.py op-test framework) #20399
• [ExecuTorch][WebGPU] Add cat op + ValueList graph support (aten.cat.default) #20398
• [ExecuTorch][WebGPU] permute_copy op test suite (cases.py op-test framework) #20397
• -> [ExecuTorch][WebGPU] Add permute_copy + IntList graph support (aten.permute_copy.default) #20396
• [ExecuTorch][WebGPU] slice_copy op test suite (cases.py op-test framework) #20395
• [ExecuTorch][WebGPU] Add slice_copy op (aten.slice_copy.Tensor) #20394
• [ExecuTorch][WebGPU] squeeze_copy + unsqueeze_copy test suites (cases.py op-test framework) #20393
• [ExecuTorch][WebGPU] Add squeeze_copy + unsqueeze_copy (flat copies) #20392
• [ExecuTorch][WebGPU] sigmoid op test suite (cases.py op-test framework) #20391
• [ExecuTorch][WebGPU] Add sigmoid op (aten.sigmoid.default) #20390
• [ExecuTorch][WebGPU] select_copy op test suite (cases.py op-test framework) #20363
• [ExecuTorch][WebGPU] Add select_copy op (aten.select_copy.int) #20362
• [ExecuTorch][WebGPU] view_copy op test suite (cases.py op-test framework) #20361
• [ExecuTorch][WebGPU] Add view_copy op (aten.view_copy.default) #20360
• [ExecuTorch][WebGPU] mul op test suite (cases.py op-test framework) #20359

Adds aten.permute_copy.default (a coordinate-reorder gather) to the WebGPU delegate, and the IntList graph value type it needs to read its dims argument.

Composition:

• runtime/WebGPUGraph.{h,cpp} — adds ValueType::IntList backed by std::vector<std::vector<int64_t>> int_lists_ + get_int_list(int); build() deserializes vkgraph::GraphTypes::IntList via value_as_IntList()->items() (int64, matching the FlatBuffer [long]); mirrors the existing scalar value plumbing.
• runtime/ops/permute/Permute.cpp — reads the permutation via get_int_list, normalizes negative dims, validates it is a permutation of [0, ndim), builds two TensorMeta UBOs + a PermuteParams{perm: vec4<u32>} uniform, guards fp32 + rank≤4, dispatches over compute_1d_workgroup_count(out.numel) with override wg_size; releases all uniforms after the bind group.
• runtime/ops/permute/permute.wgsl — delinearizes the output index over the contiguous output strides, reads input at in.strides[perm[d]] per dim (mirrors Vulkan permute_buffer.glsl).
• Registers both aten.permute_copy.default and aten.permute.default to the same handler.
  @exported-using-ghexport

Differential Revision: D108793162

Differential Revision: D108793162

[pytorch-bot]

🔗 Helpful Links

🧪 See artifacts and rendered test results at hud.pytorch.org/pr/pytorch/executorch/20396

• 📄 Preview Python docs built from this PR

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

❌ 2 New Failures, 3 Unrelated Failures

As of commit dcc3a48 with merge base 0e65ba6 ():

NEW FAILURES - The following jobs have failed:

• pull / unittest / linux / linux-job (gh)
  RuntimeError: Command docker exec -t 924748f14259c6a7febf75f2aed6850ff6e48e1fc69ec8f8f2a8f732ead4c289 /exec failed with exit code 1
• pull / unittest-editable / linux / linux-job (gh)
  RuntimeError: Command docker exec -t 37db4da251f56493c8d09470335882c58556de9d396b55f25cf436e928add847 /exec failed with exit code 1

FLAKY - The following job failed but was likely due to flakiness present on trunk:

• pull / unittest / macos / macos-job (gh) (similar failure)
  RuntimeError: Command bash /Users/ec2-user/runner/_work/_temp/exec_script failed with exit code 1

BROKEN TRUNK - The following jobs failed but was present on the merge base:

👉 Rebase onto the `viable/strict` branch to avoid these failures

• pull / test-llama-runner-qnn-linux (fp32, qnn_16a16w, qnn) / linux-job (gh) (trunk failure)
• pull / unittest-editable / macos / macos-job (gh) (trunk failure)
  RuntimeError: Command bash /Users/ec2-user/runner/_work/_temp/exec_script failed with exit code 1

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.

[JulianCloudNTH]

@claude review

[claude]

Claude finished @JulianCloudNTH's task in 3m 11s —— View job

---

Code Review — PR #20396: permute_copy + IntList graph support

• Read changed files
• Review WebGPUGraph IntList plumbing
• Review Permute.cpp op
• Review permute.wgsl shader
• Post review feedback

Overall this is a clean, well-scoped addition that faithfully mirrors the existing per-op pattern (Slice.cpp) and the Vulkan permute_buffer.glsl reference. The IntList plumbing matches the FlatBuffer [long] schema (schema/program.fbs:193-195, Vulkan schema.fbs:88-89) and the existing scalar-value conventions. I verified the WGSL gather math, the bounds/validation logic, and the uniform/bind-group lifetime. A few notes below — nothing blocking.

Correctness ✅

• Gather math is correct. For output coord d, in_bufi += coord * in_strides[perm[d]] correctly maps output dim d to input dim dims[d] (PyTorch permute semantics: out.shape[d] == in.shape[dims[d]]). Output delinearization over contiguous out_meta.strides is sound since the AOT output is contiguous.
• No division-by-zero in the shader. out_meta.strides[d] for d < ndim is always ≥1 with contiguous strides; an empty tensor (numel == 0) dispatches 0 workgroups.
• No OOB on the fixed-size arrays. perm[]/seen[] are sized kTensorMetaMaxNdim (4) and the ndim > 4 guard (Permute.cpp:62) runs before the indexing loop. The permutation-validity check (negative-dim normalize + seen[] dup detection) is correct.
• fp32 + rank≤4 guards are present and consistent with Slice.cpp.

Suggestions (non-blocking)

1. Missing op test. Every other buffer op under backends/webgpu/test/ops/ (slice, squeeze, unsqueeze, select, view_copy, …) ships a test module, but there's no test/ops/permute/. Since permute has nontrivial index math, a correctness test across a few rank-2/3/4 permutations (including negative dims and identity) would be valuable and would also exercise the new IntList deserialization path. Add a permute op test →

2. Significant boilerplate duplication with Slice.cpp. Permute.cpp:111-187 (shader module → bind-group-layout → pipeline → bind-group → dispatch → release) is ~95% identical to Slice.cpp:106-182, differing only in the params struct and WGSL constant. This is a pre-existing pattern across ops, so following it here is reasonable — but the duplication is now large enough that a small shared helper (e.g. "build a 1-storage-in / 1-storage-out / N-uniform pipeline+dispatch") would pay off across the growing op set. Out of scope for this PR; flagging for the backend maintainers.

3. kPermuteWorkgroupSizeY / kPermuteWorkgroupSizeZ are unused (permute_wgsl.h:57-58). These appear to be emitted by the WGSL codegen for all shaders, so this matches the generated-header convention and is fine — just noting they're dead for this 1-D dispatch.

4. Redundant rank guard. fill_tensor_meta (TensorMeta.h:42-44) already throws on ndim > 4, so the explicit check at Permute.cpp:62-64 is belt-and-suspenders. Harmless, and arguably clearer since it guards the array writes directly.

Verified

• WGSL SHA d34f5973…00197 matches the embedded value in permute_wgsl.h:16 (generated header is in sync with the .wgsl source). ✅
• IntList enum addition and int_lists_ resize in build() correctly mirror the other value types. ✅
  · branch gh/JulianCloudNTH/45/head