NXP backend: Enable `MM` and `AddMM` with new Neutron flow. by MartinPavella · Pull Request #20278 · pytorch/executorch

MartinPavella · 2026-06-15T12:31:48Z

Summary

This PR enables the MM and AddMM operators in the NXP backend using the new Neutron C MLIR flow.

Test plan

Unit tests provided.

cc @robert-kalmar @JakeStevens @digantdesai @rascani

pytorch-bot · 2026-06-15T12:31:52Z

🔗 Helpful Links

🧪 See artifacts and rendered test results at hud.pytorch.org/pr/pytorch/executorch/20278

📄 Preview Python docs built from this PR

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

❌ 1 New Failure, 2 Unrelated Failures

As of commit b10f3f9 with merge base 6021a58 ():

NEW FAILURE - The following job has failed:

pull / unittest / macos / macos-job (gh)
export/tests/test_target_recipes.py::TestTargetRecipes::test_vit_model

BROKEN TRUNK - The following jobs failed but were present on the merge base:

👉 Rebase onto the `viable/strict` branch to avoid these failures

pull / unittest / windows / windows-job (gh) (trunk failure)
Process completed with exit code 2.
pull / unittest-editable / windows / windows-job (gh) (trunk failure)
Process completed with exit code 2.

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

linux-foundation-easycla · 2026-06-15T12:31:56Z

❌ - login: @MartinPavella / name: Martin Pavella. The commit (09dddfb, 6985351, a88805a, b10f3f9, cd4b36c, e82072c) 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.

MartinPavella · 2026-06-15T12:32:12Z

@novak-vaclav, @irtrukhina please feel free to have a look as well.

roman-janik-nxp

LGTM, need to wait on internal tests.

roman-janik-nxp · 2026-06-22T16:25:04Z

Internal tests: https://bamboo3.sw.nxp.com/browse/MLTECE-EXIGH188-8

roman-janik-nxp

Good job!

Copilot

Copilot was unable to review this pull request because the user who requested the review has reached their quota limit.

novak-vaclav

I have just a few comments and questions, otherwise very good 👌

novak-vaclav · 2026-06-25T12:34:12Z



+# The edge operator signature is: aten.addmm(bias, input, weight, *, beta=1, alpha=1)
+MAIN_INPUT_IDX = 1


nicely explained!

novak-vaclav · 2026-06-25T12:49:43Z

    def convert(self, node: Node):
+        """Convert the `aten.addmm` operator to NeutronIR `FullyConnected`.
+        The schema is:
+            addmm(


nit: this is the schema of high level torch API, however in aten, the addmm operator's signature looks different.
I find this a bit confusing, since the node_converter deals with only aten operators.

What do you mean? What lead you to believe the schema should be different?
When I debug the lowering and stop in the AddMMConverter, I can see the schema of the exact operator that is being converted, and it is exactly equal to what's in the doc-string (see attached).

What I had in mind was I'd rather see directly the "schema" of node.args, since that is what we directly work with in node_converter. Something like node.args = [input_node, mat1, ...].
I thought the schema was taken from the documentation online, which might be less reliable than directly identifing the nodes in node.args. However if it is taken from the node attribute directly, then it's fine by me.

novak-vaclav · 2026-06-25T12:54:16Z

I'm just thinking, we could make a precommit hook that would check that for each modified file in a commit, if there is an NXP license, the year must be up to date.
Something like: get modified files -> for each file, read the first line -> if the line starts with "# Copyright", it must also contain "2026"
What do you think? I don't know if we can just create precommit hooks in an opensource repository. An alternative might be a special CI test in .github/workflows?

That's a good idea!
Just a note, only "significant" changes to a file require the license to be updated. The word "significant" is vague and up to our interpretation (for example for the changes to test_integration.py here, I deemed the changes insignificant, as I just added a comment, updates some indices, and replaced operators with their aliases. No functional changes were made. But someone else may consider this significant anyway.).

Regarding the automatic check, please feel free to bring it up during the architecture meeting on Monday. I quite like it :)

…ly node in a partition.

MartinPavella requested a review from roman-janik-nxp June 15, 2026 12:31

MartinPavella self-assigned this Jun 15, 2026

MartinPavella requested a review from robert-kalmar as a code owner June 15, 2026 12:31

MartinPavella added module: nxp Issues related to NXP Neutron NPU delegation and code under backends/nxp/ release notes: nxp Changes to the NXP Neutron backend delegate labels Jun 15, 2026

MartinPavella had a problem deploying to cadence June 15, 2026 12:31 — with GitHub Actions Failure

meta-cla Bot added the CLA Signed This label is managed by the Facebook bot. Authors need to sign the CLA before a PR can be reviewed. label Jun 15, 2026

MartinPavella force-pushed the nxg01483/EIEX-862-add-mm-addmm-support-using-new-neutron-flow branch from 9c886de to 3495635 Compare June 17, 2026 06:36

MartinPavella had a problem deploying to cadence June 17, 2026 06:36 — with GitHub Actions Failure

MartinPavella force-pushed the nxg01483/EIEX-862-add-mm-addmm-support-using-new-neutron-flow branch from 3495635 to 605f986 Compare June 17, 2026 13:54

MartinPavella had a problem deploying to cadence June 17, 2026 13:54 — with GitHub Actions Failure

MartinPavella force-pushed the nxg01483/EIEX-862-add-mm-addmm-support-using-new-neutron-flow branch 2 times, most recently from 3044d1d to 896a8c2 Compare June 19, 2026 07:50

robert-kalmar reviewed Jun 19, 2026

View reviewed changes

Comment thread backends/nxp/backend/ir/converter/node_converter.py Outdated

roman-janik-nxp reviewed Jun 19, 2026

View reviewed changes

MartinPavella force-pushed the nxg01483/EIEX-862-add-mm-addmm-support-using-new-neutron-flow branch from 896a8c2 to e3aa36b Compare June 22, 2026 07:51

roman-janik-nxp approved these changes Jun 22, 2026

View reviewed changes

novak-vaclav requested a review from Copilot June 23, 2026 14:05

Copilot AI requested a review from novak-vaclav June 23, 2026 14:05

Copilot AI reviewed Jun 23, 2026

robert-kalmar approved these changes Jun 23, 2026

View reviewed changes

novak-vaclav reviewed Jun 25, 2026

View reviewed changes

MartinPavella added 6 commits June 26, 2026 09:58

NXP backend: Enable MM and AddMM with new Neutron flow.

e82072c

Do not delegate permute_copy if it has a static input and is the on…

6985351

…ly node in a partition.

Fix failing tests.

a88805a

Resolve conflicts.

09dddfb

Update based on review.

cd4b36c

Fixes based on review comments.

b10f3f9

MartinPavella force-pushed the nxg01483/EIEX-862-add-mm-addmm-support-using-new-neutron-flow branch from e3aa36b to b10f3f9 Compare June 26, 2026 08:07

MartinPavella requested a review from novak-vaclav June 26, 2026 09:10

novak-vaclav approved these changes Jun 26, 2026

View reviewed changes



		# The edge operator signature is: aten.addmm(bias, input, weight, *, beta=1, alpha=1)
		MAIN_INPUT_IDX = 1

Uh oh!

Conversation

MartinPavella commented Jun 15, 2026 • edited by pytorch-bot Bot Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Summary

Test plan

Uh oh!

pytorch-bot Bot commented Jun 15, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

🔗 Helpful Links

🧪 See artifacts and rendered test results at hud.pytorch.org/pr/pytorch/executorch/20278

❌ 1 New Failure, 2 Unrelated Failures

Uh oh!

linux-foundation-easycla Bot commented Jun 15, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

MartinPavella commented Jun 15, 2026

Uh oh!

Uh oh!

roman-janik-nxp left a comment

Choose a reason for hiding this comment

Uh oh!

roman-janik-nxp commented Jun 22, 2026

Uh oh!

roman-janik-nxp left a comment

Choose a reason for hiding this comment

Uh oh!

Copilot AI left a comment

Choose a reason for hiding this comment

Uh oh!

novak-vaclav left a comment

Choose a reason for hiding this comment

Uh oh!

novak-vaclav Jun 25, 2026

Choose a reason for hiding this comment

Uh oh!

novak-vaclav Jun 25, 2026

Choose a reason for hiding this comment

Uh oh!

MartinPavella Jun 26, 2026

Choose a reason for hiding this comment

Uh oh!

novak-vaclav Jun 26, 2026

Choose a reason for hiding this comment

Uh oh!

Uh oh!

novak-vaclav Jun 25, 2026

Choose a reason for hiding this comment

Uh oh!

novak-vaclav Jun 25, 2026

Choose a reason for hiding this comment

Uh oh!

MartinPavella Jun 26, 2026

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

5 participants

MartinPavella commented Jun 15, 2026 •

edited by pytorch-bot Bot

Loading

pytorch-bot Bot commented Jun 15, 2026 •

edited

Loading

linux-foundation-easycla Bot commented Jun 15, 2026 •

edited

Loading