Add extended support from new Neutron C flow for Clamp operator

[StrycekSimon]

Summary

• Moves flag indicating use of the new Neutron C flow from CustomCompileConfig to NeutronTargetSpec
• Adds new Neutron C flow support for Clamp operator

Test plan

New test cases for Clamp are introduced. The relocation of new flag is covered by already existing unit tests.

cc @robert-kalmar @JakeStevens @digantdesai @rascani

[pytorch-bot]

🔗 Helpful Links

🧪 See artifacts and rendered test results at hud.pytorch.org/pr/pytorch/executorch/19510

• 📄 Preview Python docs built from this PR

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

✅ No Failures

As of commit 54e327d with merge base b04cc65 ():
💚 Looks good so far! There are no failures yet. 💚

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

[Copilot]

Pull request overview

This PR moves the use_new_flow_neutron_c flag from CustomDelegationOptions onto NeutronTargetSpec so it can be threaded through the quantizer and converters as part of target configuration, and extends the new Neutron‑C flow's clamp support beyond Relu‑shaped bounds by emitting Maximum/Minimum ops with quantized constants.

Changes:

• Move use_new_flow_neutron_c from CustomDelegationOptions to NeutronTargetSpec, updating all call sites and converters that read the flag.
• Extend ClampConverter to lower arbitrary clamp bounds to a Max+Min sequence under the new flow, factoring _is_convertible_to_relu out as a helper used by both the converter and the new ClampPattern.get_anchors.
• Refactor QuantizationPattern to require neutron_quantizer in the base constructor, propagate this through all subclasses, and rework ClampPattern to use a shared‑spec anchor under the new flow.

Reviewed changes

Copilot reviewed 13 out of 13 changed files in this pull request and generated 2 comments.

Show a summary per file

File	Description
backends/nxp/backend/custom_delegation_options.py	Removes use_new_flow_neutron_c from CustomDelegationOptions.
backends/nxp/backend/neutron_target_spec.py	Adds use_new_flow_neutron_c constructor argument and attribute.
backends/nxp/nxp_backend.py	Threads flag into NeutronTargetSpec constructions; drops it from CustomDelegationOptions.
backends/nxp/backend/ir/converter/node_converters/ops_converters/abs_converter.py	Reads new flag from target spec instead of delegation options.
backends/nxp/backend/ir/converter/node_converters/ops_converters/avg_pool_2d_converter.py	Same flag-source switch.
backends/nxp/backend/ir/converter/node_converters/ops_converters/max_pool2d_with_indices_converter.py	Same flag-source switch.
backends/nxp/backend/ir/converter/node_converters/ops_converters/mul_tensor_converter.py	Same flag-source switch.
backends/nxp/backend/ir/converter/node_converters/ops_converters/clamp_converter.py	Adds Max/Min lowering for non‑Relu clamp bounds under new flow; introduces _is_convertible_to_relu.
backends/nxp/quantizer/patterns.py	QuantizationPattern now takes neutron_quantizer; ClampPattern adds shared‑spec anchors when new flow is on.
backends/nxp/quantizer/neutron_quantizer.py	Passes self into all pattern constructors.
backends/nxp/tests/executorch_pipeline.py	Construct NeutronTargetSpec with the new flag; minor formatting.
backends/nxp/tests/ir/converter/node_converter/test_clamp_converter.py	Adds parameterized tests for clamp under the new Neutron‑C flow.
examples/nxp/aot_neutron_compile.py	Forwards CLI flag into NeutronTargetSpec.

Comments suppressed due to low confidence (1)

backends/nxp/backend/ir/converter/node_converters/ops_converters/clamp_converter.py:169

• The min/max constant tensors are hard-coded to np.int8, but _is_supported_on_target returns True unconditionally when use_new_flow_neutron_c is set, with no dtype filter. If the input is uint8 (which is listed as a supported type by other converters in this flow, e.g. abs_converter.py, avg_pool_2d_converter.py), values >127 will silently overflow when stored as int8, producing incorrect outputs. The dtype passed to np.array should be derived from the dequant node's dtype (or from quant_min/quant_max).

        min_tensor = self.builder.create_tensor_for_data(
            np.array([min_value], np.int8), "min"
        )
        self.propagate_quantization(x, min_tensor)
        max_tensor = self.builder.create_tensor_for_data(
            np.array([max_value], np.int8), "max"
        )

---

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

[MartinPavella]

Please add the new tests inside a class. It's a new convention we have started with the new flow:
class TestClampNewNeutronFlow:
We can then use shorter test names with improved clarity.

Please draw inspiration from other open PRs which enable new ops (e.g. here). The TestClampNewNeutronFlow class should test all aspects of the operator with the new flow, because eventually we will remove the old tests with the old flow. So please add more tests.

[StrycekSimon]

The tests are now inside a class. Also, added a broader set of test cases + more robust testing including verification of producing ReLU vs. Min/Max. I had to resort to spying on TFlite builder itself. Can you have a look?

[MartinPavella]

Would be nice to see some assertions that the node is indeed converted to a Relu variant when possible.

[StrycekSimon]

Added checking in TFLite. (See my response in your other comment.)

[roman-janik-nxp]

Needs a rework.

[linux-foundation-easycla]

• ❌ - login: @StrycekSimon / name: Simon Strycek. The commit (54e327d, 683638e, cd18329, fb27dd9) is not authorized under a signed CLA. Please click here to be authorized. For further assistance with EasyCLA, please visit our EasyCLA portal and chat with our support bot.

[StrycekSimon]

Here we go again with Copilot... 😄

[Copilot]

Pull request overview

Copilot reviewed 14 out of 14 changed files in this pull request and generated 4 comments.

Comments suppressed due to low confidence (2)

backends/nxp/tests/ir/converter/node_converter/test_clamp_converter.py:37

• exir_ops is imported but not used anywhere in this test module, which will fail unused-import linting (F401). Remove this import to keep the test file lint-clean.

)
from executorch.exir.dialects._ops import ops as exir_ops
from executorch.backends.nxp.tests.use_qat import *

backends/nxp/backend/ir/converter/node_converters/ops_converters/clamp_converter.py:210

• max_tensor is created with a hard-coded np.int8 dtype. If the clamp IO quantization dtype is uint8, this will produce an incorrect constant tensor type/value. Create the constant using the same dtype as the clamp input/output quantization.

        if max_value is not None:
            max_value = self._quantize_value(max_value, zp, scale, quant_min, quant_max)
            max_tensor = self.builder.create_tensor_for_data(
                np.array([max_value], np.int8), "max"
            )

[MartinPavella]

Well done 👍🏻

[Copilot]

Pull request overview

Copilot reviewed 14 out of 14 changed files in this pull request and generated 4 comments.