NXP backend: Enable Sub Tensor with new Neutron flow#19588
NXP backend: Enable Sub Tensor with new Neutron flow#19588roman-janik-nxp wants to merge 1 commit into
Conversation
roman-janik-nxp
added
module: nxp
🔗 Helpful Links🧪 See artifacts and rendered test results at hud.pytorch.org/pr/pytorch/executorch/19588
Note: Links to docs will display an error until the docs builds have been completed. ❗ 1 Active SEVsThere are 1 currently active SEVs. If your PR is affected, please view them below: ❌ 3 New FailuresAs of commit c91b4b8 with merge base 5b89d23 ( NEW FAILURES - The following jobs have failed:
This comment was automatically generated by Dr. CI and updates every 15 minutes. |
meta-cla
Bot
added
the
CLA Signed
| @@ -1,26 +1,40 @@ | |||
| # Copyright 2025 NXP | |||
|
|
||
|
|
||
| class TestSubTensorNewNeutronFlow: | ||
| @pytest.mark.skip("AIR-14602: incorrect results") |
There was a problem hiding this comment.
The AIR-14602 refers to the Add operator. Are you sure this is correct?
|
|
||
|
|
||
| class TestSubTensorNewNeutronFlow: | ||
| @pytest.mark.skip("AIR-14602: incorrect results") |
There was a problem hiding this comment.
I think using @pytest.mark.xfail(strict=True) would be better here, as once Neutron is updated and the bug is fixed. we will be alerted and will update the test.
With skip it's possible that we will forget about it.
This applies to other tests too, not just this one.
| [ | ||
| pytest.param((6, 8), id="2D."), | ||
| pytest.param((1, 4, 8), id="3D."), | ||
| pytest.param((1, 4, 8, 8), id="4D."), |
There was a problem hiding this comment.
Please don't use multiples of 8 in the shapes, to make sure there is no num_macs related restriction.
This applies to all tests, not just this one.
| pytest.param((1, 4, 8, 8), id="4D."), | ||
| ], | ||
| ) | ||
| def test__basic_nsys_inference(self, x_input_shape, mocker): |
There was a problem hiding this comment.
Please separate out the passing and failing cases (I assume at least some cases produce correct outputs. If not, please try to find some :D). Right now everything is skipped. It's desirable to have at least some passing tests.
| id="2 inputs 3D.", | ||
| ), | ||
| pytest.param( | ||
| [ModelInputSpec((4,)), ModelInputSpec((4, 4))], id="2 inputs 2D+3D." |
There was a problem hiding this comment.
The id seems incorrect. Looks like 1D - 2D to me.
| ) | ||
| assert graph_contains_any_of_ops(delegated_ep.graph, [SubTensor]) | ||
|
|
||
| @pytest.mark.skip("AIR-14602: incorrect results") |
There was a problem hiding this comment.
Would it be possible to find a working example so we can have a passing test?
| ), | ||
| ], | ||
| ) | ||
| def test__w_conv_unsupported(self, input_spec): |
There was a problem hiding this comment.
Please also add a test for the supported channels first broadcasting case.
2 participants
Summary
Add tests verifying correct support for sub.tensor by the Neutron backend using the new Neutron MLIR flow.
Test plan
Unit tests provided.
cc @robert-kalmar @JakeStevens @digantdesai @rascani