NXP backend: added support for aten.bmm

[novak-vaclav]

Summary

adds support for aten.bmm operator.

The original PR is here, however I pushed to the branch without committing the work first and the PR closed itself auto-magically.

Test plan

tests can be manually run using pytest -c /dev/null backends/nxp/tests/

cc @robert-kalmar @JakeStevens @digantdesai @MartinPavella

[pytorch-bot]

🔗 Helpful Links

🧪 See artifacts and rendered test results at hud.pytorch.org/pr/pytorch/executorch/17818

• 📄 Preview Python docs built from this PR

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

⚠️ 1 Awaiting Approval

As of commit 8fc1396 with merge base abc0237 ():

AWAITING APPROVAL - The following workflow needs approval before CI can run:

• Claude Code (gh)

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

[novak-vaclav]

@pytorchbot label "release notes: nxp"

[novak-vaclav]

@pytorchbot label "module: nxp"

[novak-vaclav]

@MartinPavella
Internal tests passing here.

Please try to re-run the periodic / gather models Github test if possible, according to the log it was stopped because some other, higher priority task was scheduled.

After that, this can be merged ✔️

[Copilot]

Pull request overview

This PR adds support for the aten.bmm (batch matrix multiply) operator in the NXP Neutron backend for ExecuTorch. It includes a new BMMConverter node converter, a BMMPattern quantization pattern, test models, and test cases.

Changes:

• Adds BMMConverter class with IR support and target-level (num_macs) validation, converting aten.bmm to TFLite BatchMatMul with proper channels-first/last handling
• Registers BMMPattern quantizer and BMMConverter in the partitioner/converter registries
• Adds test models (BatchMatMulModel, BatchMatMulConvModel) and tests covering supported/unsupported shapes and conv+bmm fusion scenarios

Reviewed changes

Copilot reviewed 9 out of 9 changed files in this pull request and generated 8 comments.

Show a summary per file

File	Description
bmm_converter.py	Core converter implementation with IR and target support checks, channels-format handling
__init__.py	Exports BMMConverter
edge_program_converter.py	Registers BMMConverter for aten.bmm.default
neutron_partitioner.py	Registers BMMConverter in the supported ops map
patterns.py	Adds BMMPattern quantization pattern
neutron_quantizer.py	Registers BMMPattern in the quantizer
models.py	Adds BatchMatMulModel and BatchMatMulConvModel test models
test_bmm_converter.py	Tests for supported/unsupported configs and conv+bmm quantization
op-support.csv	Documents aten.bmm.default operator support

---

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

[Copilot]

Pull request overview

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

---

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

You can also share your feedback on Copilot code review. Take the survey.

[novak-vaclav]

Fixed the issues mentioned by @MartinPavella
When rebasing and splitting the feature changes and the rebased changes into separate commits, and then pushing them in the wrong order, it automatically closes the PR. However I found out that it can be reopened.

[MartinPavella]

x1_W not divisible by NUM_MACS

Unless W refers to the first dimension, this is not true,

[novak-vaclav]

During the implementation and review of the feedback, I forgot to also modify this test. The test cases were all a bit off, so I fixed them. Please have a look.

[novak-vaclav]

Fixed all the issues, please have a look @MartinPavella