Recognise the repeat_interleave idiom in RemovePermutesAroundElementwiseOps (#21790)

[mcremon-meta]

Summary:

repeat_interleave(scale, dim) lowers to a three-node idiom::

unsqueeze(dim + 1) -> expand_copy(scale at dim + 1)
                   -> view_copy(merge dim, dim + 1)

which the region walker could not cross, so any permuted region containing one
was rejected. torchaudio's Stretch2d is built from this, so it blocks every
upsampling network -- on wavernn it strands the permutes ConvToChannelsLast
put around each conv in the upsample path.

Adding expand_copy to _permutable_ops on its own does not help: traversal
then dies one node later at the merging view_copy. That merge is not
unconditionally layout-invariant the way cat/slice are -- flattening two
dims only commutes with a permutation that leaves them adjacent and in order,
otherwise the flatten reorders elements.

Worse, handling the three nodes separately would force a choice of un-permuted
position for the intermediate unit dim (inserting a size-1 dim is ambiguous:
any position gives a consistent permutation). That arbitrary choice then
decides whether the later merge is still legal, coupling two rewrites that
look independent.

So the triple is matched as a unit (_interleave_triple). It is rank-preserving
overall, so the permutation flows through unchanged and the only rewrite needed
is remapping the interleaved dim, new_dim = P[old_dim], plus rebuilding the
three shape args in un-permuted space (update_interleave). No intermediate
layout is ever materialised, so the ambiguity does not arise.

Guards: head and expand must each have exactly one user (otherwise the in-place
rewrite would corrupt another consumer), every non-inserted dim of the expand
must pass through untouched, and the view's output shape must be exactly the
merge. Symbolic shapes are declined. The head may be either an explicit
unsqueeze_copy or a view_copy that inserts one unit dim, matching how the
pass already treats the two spellings interchangeably.

Differential Revision: D114508262

[pytorch-bot]

🔗 Helpful Links

🧪 See artifacts and rendered test results at hud.pytorch.org/pr/pytorch/executorch/21790

• 📄 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 2ec948f with merge base c6213ae ():
💚 Looks good so far! There are no failures yet. 💚

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

[meta-codesync]

@mcremon-meta has exported this pull request. If you are a Meta employee, you can view the originating Diff in D114508262.

[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.