Reject non-default dim order in unfold_copy and select_scatter

[SuryanshSS1011]

Summary

Follow-up to #21865. Two more portable kernels index with getLeadingDims and getTrailingDims, which describe a contiguous run only in the default dim order.

unfold_copy carries no dim order check at all. Checked against the same logical data in two memory formats:

kernel	contiguous input	channels-last input
aten.unfold_copy	matches	wrong, max abs diff 14.0

It reads self as a contiguous run and writes out as a flat run, so neither side survives a non-default layout. unfold appears in shipped example models including the Qualcomm llama vision encoder and the gemma4 speech transform, where a channels-last input is ordinary.

select_scatter has the same arithmetic and the same defect, but it is not reachable from export today. It carries tensors_have_same_dim_order(in, src, out), which is not sufficient on its own, because is_channels_last_dim_order accepts 4 and 5 dims, so a rank-5 input and its rank-4 src can both be channels-last and pass together. At rank 4 the src is rank 3 and the check does reject. Reaching the rank-5 case needs a channels-last dim order that exir/dim_order_utils.py refuses to generate, since it raises for any rank other than 4. So the kernel is protected by a gap in the export layer rather than by its own checks, and the guard closes that before 5D channels-last support makes it reachable.

Both get the guard op_addmm, op_bmm and the kernels in #21865 use:

ET_KERNEL_CHECK(ctx, tensor_is_default_dim_order(in), InvalidArgument, out);

unfold_copy needs it on out as well, since out is one rank above self and is written as a flat run. Checking the pair with tensors_have_same_dim_order would not work there, for the same rank reason as above.

Test plan

A NonDefaultDimOrderDies per kernel, plus a second unfold_copy case covering the out guard. Reverting the two kernel sources while keeping the tests fails all three.

The select_scatter test has to be rank 5. A rank-4 case is rejected by the pre-existing same dim order check on current main, so it would pass without reaching the new guard. It builds its tensors with make_with_dimorder, because channels_last_like asserts sizes.size() == 4 and cannot express the failing case.

All three tests skip under is_aten, since the ATen kernels handle a non-default dim order and do not throw.

This also registers op_unfold_copy_test.cpp in kernels/test/CMakeLists.txt, without which neither the existing tests in that file nor the new ones run in a CMake build.

[pytorch-bot]

🔗 Helpful Links

🧪 See artifacts and rendered test results at hud.pytorch.org/pr/pytorch/executorch/21915

• 📄 Preview Python docs built from this PR

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

✅ No Failures

As of commit f7abaf0 with merge base 02b38d6 ():
💚 Looks good so far! There are no failures yet. 💚

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

[Copilot]

Copilot was unable to review this pull request because the user who requested the review has reached their quota limit.

[SuryanshSS1011]

@pytorchbot label "release notes: ops & kernels"

[Copilot]

Copilot was unable to review this pull request because the user who requested the review has reached their quota limit.