[Cortex-M] Support public edge transforms

[rascani]

Summary

Passing CortexMPassManager to edge.transform() currently fails because it inherits a graph-module pass manager while keeping its lowering pipeline in a separate .transform() method. Callers consequently have to replace edge._edge_programs["forward"] directly.

Make it an ExportedProgramPassManager with a program-aware adapter around the existing pipeline. Each pass is instantiated against the current exported program, preserving updates to constants and graph signatures. This enables edge.transform(CortexMPassManager(...)) and to_edge_transform_and_lower(..., transform_passes=CortexMPassManager(...)), while retaining the legacy bound-program constructor and .transform() entry point. Update the exporters, examples, documentation, and test harness to use the public APIs.

Authored with OpenAI Codex.

Test plan

The following tests passed: 307 passed, 1 skipped, and 24 expected failures. The new regression tests cover both layouts and all three entry points, serialization, multiple methods, target configuration, and constant lifting.

python -m pytest -q \
  backends/cortex_m/test/test_pass_manager.py \
  backends/cortex_m/test/test_explicit_layout_pipeline.py \
  backends/cortex_m/test/misc/test_target_config.py

OMP_NUM_THREADS=1 python -m pytest -q -n 4 \
  backends/cortex_m/test/ops exir/tests/test_pass_infra.py \
  -k 'test_dialect or TestExportedProgramPassManager'

Additional smoke checks passed for the Arm AOT exporter, CMSIS metadata, and ETRecord generation. Executed the export cells in examples/arm/ethos_u_cmsis_nn_fallback_example.ipynb with the pinned ethos-u-vela==5.1.0, confirming that the compiled delegate survives Cortex-M lowering and serialization.

Built a fresh Cortex-M55 semihosting runner from this checkout with CORTEX_M_ENABLE_RUNTIME_CHECKS=ON and EXECUTORCH_ENABLE_PROGRAM_VERIFICATION=ON. Ran four serialized programs on Corstone-300 FVP: mixed Ethos-U/Cortex-M, convolution with each layout, and linear. All four outputs matched quantized PyTorch exactly, with zero maximum absolute error; the mixed case executed one NPU delegation.

Lintrunner passed for all 18 changed files. Three independent Codex reviews found no actionable regressions. Buck execution has not been validated locally.

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

[pytorch-bot]

🔗 Helpful Links

🧪 See artifacts and rendered test results at hud.pytorch.org/pr/pytorch/executorch/22980

• 📄 Preview Python docs built from this PR

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

❗ 1 Active SEVs

There are 1 currently active SEVs. If your PR is affected, please view them below:

• linux.dgx.b200 runner queuing exceeds 4 hours and dgxb200-05 "CUDA not available"

❌ 1 New Failure, 2 Unrelated Failures

As of commit c472672 with merge base 9b91b43 ():

NEW FAILURE - The following job has failed:

• trunk / test-samsung-models-linux / linux-job (gh)
  test_mobilebert_finetuning_fp16

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

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

• Apple / build-benchmark-app / macos-job (gh) (trunk failure)
  RuntimeError: Command bash /Users/runner/work/_temp/exec_script failed with exit code 1
• Apple / build-frameworks-ios / macos-job (gh) (trunk failure)
  RuntimeError: Command bash /Users/runner/work/_temp/exec_script failed with exit code 1

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

[AdrianLundell]

Good stuff!

[rascani]

Failures unrelated.