[ExecuTorch][WebGPU] Add slice_copy op (aten.slice_copy.Tensor)

[JulianCloudNTH]

Stack from ghstack (oldest at bottom):

• [ExecuTorch][WebGPU] aten.index.Tensor test suite (export + native golden) #20465
• [ExecuTorch][WebGPU] Add aten.index.Tensor (1D-self gather) #20464
• [ExecuTorch][WebGPU] Add clone op (aten.clone.default) #20463
• [ExecuTorch][WebGPU] Flatten landed-op test dirs to test/ops/test_<op>.py #20435
• [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.slice_copy.Tensor to the WebGPU delegate as a gather: each output element is mapped back to its source input element along the sliced dim via start + coord * step.

Composition (single compute dispatch):

• runtime/ops/slice/Slice.cpp — reads args = [self, dim, start, end, step, out] via read_scalar (static Int/Null-sentinel default; throws on dynamic SymInt); normalizes negative dim/start, clamps start to [0, in_size]; builds two TensorMeta UBOs + a SliceParams{dim, start, step} uniform; guards fp32; dispatches over compute_1d_workgroup_count(out.numel) with override wg_size; releases all uniforms after the bind group.
• runtime/ops/slice/slice.wgsl — delinearizes the output index over the contiguous output strides, maps the sliced-dim coordinate back to the input (start + coord*step), relinearizes over the input strides.
  @exported-using-ghexport

Differential Revision: D108793168

Differential Revision: D108793168

[pytorch-bot]

🔗 Helpful Links

🧪 See artifacts and rendered test results at hud.pytorch.org/pr/pytorch/executorch/20394

• 📄 Preview Python docs built from this PR

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

❌ 3 New Failures, 2 Pending, 2 Unrelated Failures

As of commit fad5528 with merge base e03f777 ():

NEW FAILURES - The following jobs have failed:

• Propose to merge ghstack orig PRs to main / Try to create a PR with ghstack /orig branch (gh)
  Process completed with exit code 1.
• pull / test-qnn-models-linux (mv3) / linux-job (gh)
  RuntimeError: Command docker exec -t 31e2c1d82991b5e4eb9e96dd3ca4e65d4d335c38da2d5d21062292477877e9a4 /exec failed with exit code 92
• pull / unittest-buck / linux / linux-job (gh)
  RuntimeError: Command docker exec -t 9f8438410ffc2d982eabb4986e0a924b1d68be590149764f3a4535f63caaf3d2 /exec failed with exit code 3

BROKEN TRUNK - The following jobs failed but were 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-buck / macos / macos-job (gh) (trunk failure)
  RuntimeError: Command bash /Users/ec2-user/runner/_work/_temp/exec_script failed with exit code 3

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 2m 27s —— View job

---

Code Review: slice_copy WebGPU op

• Read changed files
• Check conventions vs sibling ops & TensorMeta
• Analyze correctness, edge cases, style
• Post review

Reviewed against the sibling select op (runtime/ops/select/), which this closely and correctly mirrors. The gather approach is sound: contiguous-output delinearization → map sliced-dim coord via start + coord*step → relinearize over input strides. The std140 TensorMeta UBO usage, fp32 guard, workgroup sizing, uniform lifetime handling, and bind-group layout all match the established pattern. Nice work — the negative-dim/start normalization and the Int/Null/SymInt sentinel handling in read_scalar are handled carefully.

A few points below; only the first is a real (if low-likelihood) gap.

Suggestions

1. step is not validated — a non-positive value silently misbehaves (Slice.cpp:73, 86-88)
step is read and cast straight to uint32_t. The other parameters get explicit range checks (dim out of range, start clamp), but step doesn't. A step <= 0 would wrap to a huge u32 and produce out-of-bounds reads in the shader (no bounds check on in_bufi). PyTorch/the partitioner guarantee step >= 1 for slice_copy.Tensor, so this is defensive, but it's cheap and matches the surrounding validation style:

const int64_t step = read_scalar(graph, args.at(4), 1, "step");
if (step < 1) {
  throw std::runtime_error("slice: step must be >= 1");
}

Fix this →

2. No bounds check on in_bufi in the shader (slice.wgsl:39)
The computed input index is used to read input[in_bufi] without a guard. This relies entirely on AOT-correct shapes + params (1 above). That's consistent with select.wgsl, so it's acceptable as-is — just noting the invariant the C++ side must uphold.

3. Test coverage and partitioner allowlist appear to be missing from this PR

• aten.slice_copy.Tensor is not added to WEBGPU_SUPPORTED_OPS in backends/webgpu/test/tester.py (cf. select_copy.int at line 26).
• No SliceModule/_SLICE_CONFIGS op-test was added — the _fn_config_suite builder in op_tests/cases.py:149 already lists slice in its docstring (view/select/slice), so wiring one up should be a near-copy of the select suite.

If those land in another PR of this ghstack, ignore this — but the runtime op isn't exercised by the op-test framework without them.

Nits

• The bind-group / pipeline / layout boilerplate (Slice.cpp:106-182) is byte-for-byte identical to Select.cpp. This is the existing per-op convention, so duplicating it is the right call here, but the WebGPU backend may eventually want a shared helper as the op count grows. No action needed.
• slice_wgsl.h carries a wgsl-sha256 — make sure test_wgsl_codegen.py passes (it should, since the header matches slice.wgsl).

Overall: clean, idiomatic, and consistent with the backend. The only thing I'd actually change before merge is the step guard (#1).
· branch gh/JulianCloudNTH/43/head