Arm backend: Add CanonicalizeViewCopyPermutePass

[AdrianLundell]

This pass replaces PostponePermuteOpBelowSqueezeOrUnsqueezeLikeView
using a more general condition for which permute-view combinations may be
fused, see docstrings.

• remove stale comments in mha test

cc @digantdesai @freddan80 @per @zingo @oscarandersson8218 @mansnils @Sebastian-Larsson @robell @rascani

[pytorch-bot]

🔗 Helpful Links

🧪 See artifacts and rendered test results at hud.pytorch.org/pr/pytorch/executorch/20276

• 📄 Preview Python docs built from this PR

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

❌ 5 New Failures, 1 Cancelled Job, 1 Unclassified Failure

As of commit 9913688 with merge base e88fd04 ():

NEW FAILURES - The following jobs have failed:

• Cadence Build & Test / hifi-build / hifi4 (gh)
  Input required and not supplied: aws-region
• Cadence Build & Test / vision-build / vision (gh)
  Input required and not supplied: aws-region
• pull / test-lora-multimethod-linux / linux-job (gh)
  RuntimeError: Command docker exec -t eafba0428e6c8343c58a76589e689a771bd1d361303cfdcbb147a2e27fbf6e05 /exec failed with exit code 1
• pull / test-phi-3-mini-runner-linux / linux-job (gh)
  RuntimeError: Command docker exec -t 4ae6ac11562fdc357ad1f54a8b895f4744fbb030f52b4e845cc0dee63ef3f573 /exec failed with exit code 1
• trunk / test-qnn-optimum-model (fp32, focalnet) / linux-job (gh)
  RuntimeError: Command docker exec -t 38c5bd993280da78bf3058bb3c8d35c3970e6288a36b144996ae4c9f085978e6 /exec failed with exit code 92

UNCLASSIFIED FAILURE - DrCI could not classify the following job because the workflow did not run on the merge base. The failure may be pre-existing on trunk or introduced by this PR:

• trunk / test-selective-build-macos (cmake) / macos-job (gh) (this job did not run on the merge base, so DrCI cannot tell whether the failure is pre-existing)
  RuntimeError: Command bash /Users/ec2-user/runner/_work/_temp/exec_script failed with exit code 1

CANCELLED JOB - The following job was cancelled. Please retry:

• pull / unittest-editable / linux / linux-job (gh)

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

[zingo]

OK to merged but make sure no buck2 changes are needed.

[Copilot]

Pull request overview

This PR introduces a new Arm backend graph pass, CanonicalizeViewCopyPermutePass, to more generally canonicalize/fuse view_copy/permute_copy chains (including swapping adjacent view/permute when it enables additional fusions), replacing the previous squeeze/unsqueeze-focused permute postponing transform in the Arm pipeline.

Changes:

• Add CanonicalizeViewCopyPermutePass to canonicalize/fuse/swap view_copy/permute_copy chains and handle singleton-only permutes (static shapes) as views.
• Wire the new pass into the Arm TOSA pipeline and remove the old PostponePermuteOpBelowSqueezeOrUnsqueezeLikeView/ConvertPermuteSingletonToViewPass steps from that pipeline.
• Add targeted unit tests for swap/canonicalization behavior and update transpose-count expectations impacted by the new canonicalization.

Reviewed changes

Copilot reviewed 9 out of 9 changed files in this pull request and generated 2 comments.

Show a summary per file

File	Description
backends/arm/_passes/canonicalize_view_copy_permute_pass.py	New canonicalization pass that fuses adjacent ops and swaps compatible view/permute pairs while preserving shape meta (incl. SymInt).
backends/arm/_passes/arm_pass_manager.py	Integrates the new pass into the Arm pipeline and removes older permute-view reordering steps from that pipeline stage.
backends/arm/_passes/init.py	Exposes CanonicalizeViewCopyPermutePass from the Arm passes package.
backends/arm/test/passes/test_canonicalize_view_copy_permute_pass.py	New end-to-end tests validating graph rewrites and numerical equivalence for canonicalization behavior (incl. symbolic shapes).
backends/arm/test/passes/test_view_permute_swap.py	New unit tests validating view -> permute swap logic against tensor semantics (static + symbolic cases).
backends/arm/test/passes/test_permute_view_swap.py	New unit tests validating permute -> view swap logic against tensor semantics (static + symbolic cases).
backends/arm/test/misc/test_transpose_counts.py	Updates expected transpose counts for affected models/cases after canonicalization changes.
backends/arm/test/misc/test_high_rank_permute_view_invariants.py	Updates expected transpose counts for generated high-rank permute/view invariant cases.
backends/arm/test/ops/test_multihead_attention.py	Removes now-stale MHA test configuration/comments related to quantization settings.

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

[AdrianLundell]

• No required Buck2 changes for the latest commit.

The relevant targets already cover the new files:

• New pass source is picked up by glob(["*.py"]) in backends/arm/_passes/TARGETS:44.
• New pass tests are picked up by native.glob(["passes/test_*.py"]) in backends/arm/test/targets.bzl:15.

One optional cleanup: backends/arm/_passes/TARGETS:29 still has //executorch/backends/transforms:postpone_permute_below_squeeze_view, but arm_pass_manager.py no longer imports that pass.
It is stale, not necessary to fix target resolution.

So nothing should break but I could do the clean up.