NXP backend: Separable convolution decomposition in executorch

[MartinPavella]

Summary

This PR introduces an aten dialect pre-processing pass which splits conv nodes with group>1 into multiple parallel conv nodes with group=1. This replaces the original implementation in Neutron IR.

Test plan

Unit tests provided in backends/nxp/tests/test_split_group_convolution.py

cc @robert-kalmar @roman-janik-nxp @StrycekSimon @jirioc

[pytorch-bot]

🔗 Helpful Links

🧪 See artifacts and rendered test results at hud.pytorch.org/pr/pytorch/executorch/13758

• 📄 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:

• ROCm MI2xx CI/CD workflows failing due to : download from https://api.github.com/repos/pytorch/pytorch timed out.

❌ 2 New Failures

As of commit fe37a89 with merge base 1520f9f ():

NEW FAILURES - The following jobs have failed:

• Build documentation / build (buck2) / Build doc (gh)
  At least one of the pre-conditions you specified did not hold
• pull / unittest / macos / macos-job (gh)
  RuntimeError: Command bash /Users/ec2-user/runner/_work/_temp/exec_script failed with exit code 5

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

[MartinPavella]

@pytorchbot label "module: nxp" "release notes: nxp"

[robert-kalmar]

Why do you remove the tests for the separable convolution and not update them? Now they can convert to Neutron, or we miss some Ops?

[MartinPavella]

8 tests were removed in total:

• 4 of them tested the group conv decomposition with a float32 model. These were used to test the IR decomposition implementation. They were not that useful as they did not test the intended use-case (quantized model).
• The last 4 used a quantized model, but as split is not yet supported, the group conv couldn't be delegated. So these tests only verified that the node was not delegated.

8 new tests were added in backends/nxp/tests/test_split_group_convolution.py, which test the decomposition and the delegation together.

Updating the 4 removed floating point tests would result in basically duplicate tests. And as the decomposition is no longer the responsibility of the ConvConverter, it made more sense to me to test the new functionality in a new module.

[robert-kalmar]

• Fix the tests. The unit tests are failing.

[shoumikhin]

@MartinPavella executorch/backends/nxp/tests:test_quantizer test failed for us internally on this PR, can you please check if this looks related?

ImportError while importing test module '/data/sandcastle/boxes/trunk-hg-full-fbsource/buck-out/v2/gen/fbcode/a9f8037888495e78/executorch/backends/nxp/tests/__test_quantizer__/test_quantizer#link-tree/executorch/backends/nxp/tests/test_quantizer.py'.
Hint: make sure your test modules/packages have valid Python names.
Traceback:
/usr/local/fbcode/platform010/lib/python3.12/importlib/__init__.py:135: in import_module
    return _bootstrap._gcd_import(name[level:], package, level)
../buck-out/v2/gen/fbcode/a9f8037888495e78/executorch/backends/nxp/tests/__test_quantizer__/test_quantizer#link-tree/executorch/backends/nxp/tests/test_quantizer.py:12: in <module>
    from executorch.backends.nxp.quantizer.neutron_quantizer import NeutronQuantizer
../buck-out/v2/gen/fbcode/a9f8037888495e78/executorch/backends/nxp/tests/__test_quantizer__/test_quantizer#link-tree/executorch/backends/nxp/quantizer/neutron_quantizer.py:11: in <module>
    from executorch.backends.nxp.aten_passes.neutron_aten_pass_manager import (
../buck-out/v2/gen/fbcode/a9f8037888495e78/executorch/backends/nxp/tests/__test_quantizer__/test_quantizer#link-tree/executorch/backends/nxp/aten_passes/neutron_aten_pass_manager.py:16: in <module>
    from executorch.backends.nxp.aten_passes.split_group_convolution import (
../buck-out/v2/gen/fbcode/a9f8037888495e78/executorch/backends/nxp/tests/__test_quantizer__/test_quantizer#link-tree/executorch/backends/nxp/aten_passes/split_group_convolution.py:10: in <module>
    from executorch.backends.nxp.backend.ir.converter.node_converters.shared.conv_utils import (
E   ModuleNotFoundError: No module named 'executorch.backends.nxp.backend'
=============================== warnings summary ===============================
../buck-out/v2/gen/fbcode/a9f8037888495e78/executorch/backends/nxp/tests/__test_quantizer__/test_quantizer#link-tree/_pytest/assertion/rewrite.py:934: 52 warnings
  /data/sandcastle/boxes/trunk-hg-full-fbsource/buck-out/v2/gen/fbcode/a9f8037888495e78/executorch/backends/nxp/tests/__test_quantizer__/test_quantizer#link-tree/_pytest/assertion/rewrite.py:934: DeprecationWarning: ast.NameConstant is deprecated and will be removed in Python 3.14; use ast.Constant instead
    clear = ast.Assign(variables, ast.NameConstant(None))