NXP backend: added support for aten.conv_transpose1 and refactored convolution_converter

[novak-vaclav]

Summary

Added support for aten.conv_transpose1d by moving functionality from convolution_converter to brand new convert_1d_conv_to2d aten pass, and extending it.

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/19004

• 📄 Preview Python docs built from this PR

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

✅ You can merge normally! (1 Unrelated Failure)

As of commit 2b79e1a with merge base a49171d ():

FLAKY - The following job failed but was likely due to flakiness present on trunk:

• pull / test-moshi-linux / linux-job (gh) (matched linux rule in flaky-rules.json)
  E: Failed to fetch http://security.ubuntu.com/ubuntu/pool/main/o/openjpeg2/libopenjp2-7_2.4.0-6ubuntu0.4_amd64.deb 404 Not Found [IP: 104.20.28.246 80]

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"

[Copilot]

Pull request overview

Adds NXP backend support for aten.conv_transpose1d by moving 1D-to-2D convolution lowering out of the IR converter and into a dedicated ATen graph rewrite pass, plus adjusts quantization handling for grouped transposed convolutions.

Changes:

• Introduce ConvertConv1dToConv2dPass to rewrite aten.conv1d / aten.conv_transpose1d into 2D equivalents via unsqueeze/conv2d(or conv_transpose2d)/squeeze.
• Remove 1D-convolution handling from the TFLite convolution_converter and enable the new pass in the default Neutron ATen pass pipeline.
• Update quantizer patterns/utilities to correctly derive bias qparams for grouped conv_transpose2d and fix per-channel axis handling; add comprehensive tests for the new pass.

Reviewed changes

Copilot reviewed 9 out of 9 changed files in this pull request and generated 4 comments.

Show a summary per file

File	Description
backends/nxp/aten_passes/convert_1d_conv_to_2d.py	New ATen pass converting 1D conv/transposed conv to 2D form with shape/meta propagation.
backends/nxp/aten_passes/neutron_aten_pass_manager.py	Registers the new pass in the default Neutron ATen pass sequence.
backends/nxp/backend/ir/converter/node_converters/ops_converters/convolution_converter.py	Removes 1D convolution conversion logic; now expects only 2D weights rank.
backends/nxp/quantizer/utils.py	Adds helper to “pad”/repeat weight scales when deriving bias qparams for grouped transposed conv.
backends/nxp/quantizer/patterns.py	Drops 1D conv patterns; updates ConvTranspose2d quantization (bias qparams + correct per-channel axis).
backends/nxp/quantizer/neutron_quantizer.py	Removes the Conv1dPattern registration (since 1D conv is rewritten earlier).
backends/nxp/tests/test_convert_1d_conv_to_2d.py	New test suite covering conv1d + conv_transpose1d rewrite and full pipeline delegation.
backends/nxp/tests/models.py	Updates Conv1d test module API and adds ConvTranspose1d + runtime-weight conv1d models for testing.
backends/nxp/tests/ir/converter/node_converter/test_conv_converter.py	Removes prior conv1d conversion tests (superseded by new pass tests).

---

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

[novak-vaclav]

Refactored the whole solution. Solved cases where there was batch_norm after conv1d, since insertion of squeeze prevented batch_norm fusion with the conv.
I also thought that converting conv1d to conv2d with bmm behind it would break the bmm, but it should not happen.

[novak-vaclav]

Please review @MartinPavella and @StrycekSimon. Thank you 😊

[Copilot]

Pull request overview

Copilot reviewed 13 out of 13 changed files in this pull request and generated 5 comments.

---

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

[MartinPavella]

Great work!
I only have cosmetic suggestions and questions. Perhaps for future PRs you could split the implementation into multiple commits. Ultimately it would still get squashed to 1, but it would make the review easier as the commit messages would explain the motivation for some changes. Here I'm really not sure why some changes were made.

Also Copilot had some good comments, so please address them too if you see fit.

[StrycekSimon]

Please, handle the 2D inputs or at least mark them as unsupported and create an issue.

[novak-vaclav]

Enabled conv_transpose1d test case in test_quantizer

Fixed some of the issues and provided comments on some. Please review it @StrycekSimon @MartinPavella. Thank you! 😊

[MartinPavella]

LGTM. Please resolve the conflicts so the tests can run.

[novak-vaclav]

Rebased and resolved all comments. Please run the tests @MartinPavella
Thank you for reviews

[Copilot]

Pull request overview

Copilot reviewed 13 out of 13 changed files in this pull request and generated 4 comments.

Comments suppressed due to low confidence (1)

backends/nxp/backend/ir/converter/node_converters/ops_converters/convolution_converter.py:437

• After removing in-converter 1D convolution lowering, the NotImplementedError message will be hit if a model reaches IR conversion with rank-3 weights (i.e., the ConvertConv1dToConv2dPass wasn’t run). Consider updating the error message to explicitly point users to NeutronAtenPassManager / ConvertConv1dToConv2dPass so the failure is actionable.

        rank = t_op.tmp_inputs[1].shape.len()
        if rank == 4:  # Conv2D
            ops_to_add = self._convert_2d_conv(t_op, conv_params)
        else:
            raise NotImplementedError(
                f"{rank - 2}D convolution is not supported."
            )  # Should never get here.

---

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

[novak-vaclav]

Fixed some issues mentioned by Copilot.

[Copilot]

Pull request overview

Copilot reviewed 13 out of 13 changed files in this pull request and generated 7 comments.

[novak-vaclav]

@MartinPavella
Had to fix one failing test suite - split_group_convolution as it assumed direct support of conv1d without its conversion to 2d.
Sorry for pushing the first commit together with the rebase. To ease the pain in the many changes, I found out that ctrl+f with string "nxp" can be used to search the relevant changes, as I did not find an option to collapse all the differences :(

Hopefully everything will pass now, please run the tests. Thank you 😊

[MartinPavella]

LGTM. Once tests pass, we can merge.

[novak-vaclav]

@MartinPavella

Internal tests are passing.

The Github failing tests don't seem to be related to NXP code.

[MartinPavella]

The failing checks seem unrelated and the internal CI is passing. Merging.