Revert "Arm backend: Lower MXFP Linear to TOSA"

[zingo]

Reverts #19969

cc @digantdesai @freddan80 @per @oscarandersson8218 @mansnils @Sebastian-Larsson @robell @rascani

[pytorch-bot]

🔗 Helpful Links

🧪 See artifacts and rendered test results at hud.pytorch.org/pr/pytorch/executorch/20047

• 📄 Preview Python docs built from this PR

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

❌ 2 New Failures, 1 Cancelled Job

As of commit aa82cd0 with merge base 4c9c444 ():

NEW FAILURES - The following jobs have failed:

• Test WebGPU Backend / test-webgpu / test-backend-linux (webgpu, models) / linux-job (gh)
  RuntimeError: Command docker exec -t 65d5c8ad399fc19c3315b5b06803ef55da75abc91419c6e6f75fbe4d9dac946b /exec failed with exit code 1
• Test WebGPU Backend / test-webgpu / test-backend-linux (webgpu, operators) / linux-job (gh)
  RuntimeError: Command docker exec -t 909035fcc3097deaeded77396bf69bf1845809599d09f1355a991cde3012cb77 /exec failed with exit code 1

CANCELLED JOB - The following job was cancelled. Please retry:

• pull / unittest-nxp-neutron / linux-job (gh)

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

[github-actions]

This PR needs a release notes: label

If your change should be included in the release notes (i.e. would users of this library care about this change?), please use a label starting with release notes:. This helps us keep track and include your important work in the next release notes.

To add a label, you can comment to pytorchbot, for example
@pytorchbot label "release notes: none"

For more information, see
https://github.com/pytorch/pytorch/wiki/PyTorch-AutoLabel-Bot#why-categorize-for-release-notes-and-how-does-it-work.

[Copilot]

Pull request overview

This PR reverts the Arm backend change that lowered MXFP Linear into explicit TOSA block-scaled operators, restoring the previous behavior by removing the MXFP-specific dialect ops, visitors, rewrite pass, and associated tests.

Changes:

• Remove TOSA dialect ops and Arm operator visitors for CAST_TO_BLOCK_SCALED / MATMUL_T_BLOCK_SCALED, plus the MXFP linear rewrite pass.
• Tighten FP8 extension validation to require the dedicated fp8* extensions (rather than allowing mxfp as a substitute).
• Simplify/relocate MXFP linear tests to only validate eager CPU/reference behavior.

Reviewed changes

Copilot reviewed 19 out of 19 changed files in this pull request and generated 1 comment.

Show a summary per file

File	Description
backends/arm/tosa/mapping.py	Removes MXFP-as-fallback extension validation for FP8 dtypes.
backends/arm/tosa/dialect/ops/matmul_t_block_scaled.py	Deletes fake op definition for block-scaled matmul.
backends/arm/tosa/dialect/ops/cast_to_block_scaled.py	Deletes fake op definition for block-scaled cast.
backends/arm/tosa/dialect/init.py	Stops importing the removed dialect ops.
backends/arm/test/targets.bzl	Updates/removes test targets related to the reverted MXFP lowering pipeline.
backends/arm/test/passes/test_rewrite_mxfp_linear_pass.py	Deletes tests for the removed rewrite pass.
backends/arm/test/ops/test_mxfp_linear.py	Simplifies MXFP linear tests to eager CPU reference checks.
backends/arm/test/ops/mxfp/common.py	Deletes MXFP pipeline helper utilities.
backends/arm/test/ops/mxfp/init.py	Deletes empty package marker for removed MXFP test helpers.
backends/arm/test/misc/tosa_dialect/test_tosa_dialect_mxfp_linear.py	Deletes tests for removed TOSA dialect MXFP ops.
backends/arm/test/misc/tosa_dialect/test_tosa_dialect_cast_to_block_scaled.py	Deletes tests for removed TOSA dialect MXFP ops.
backends/arm/process_node.py	Removes float8_e8m0fnu handling from tensor serialization path.
backends/arm/operators/op_tosa_matmul_t_block_scaled.py	Deletes TOSA serializer visitor for block-scaled matmul.
backends/arm/operators/op_tosa_cast_to_block_scaled.py	Deletes TOSA serializer visitor for block-scaled cast.
backends/arm/operators/init.py	Stops importing the removed operator visitors.
backends/arm/operator_support/tosa_supported_operators.py	Removes MXFP custom-op support list and loosens dtype gating changes tied to MXFP.
backends/arm/_passes/rewrite_mxfp_linear.py	Deletes the MXFP linear rewrite pass implementation.
backends/arm/_passes/arm_pass_manager.py	Removes the rewrite pass from the Arm pass pipeline.
backends/arm/_passes/init.py	Stops exporting the removed rewrite pass.

Comments suppressed due to low confidence (2)

backends/arm/test/ops/test_mxfp_linear.py:185

• test_data is a callable (values in test_data_fp are lambda: (...)) produced by common.parametrize, but the helper annotates it as torch.Tensor. This is misleading for readers and type checkers.
  backends/arm/test/ops/test_mxfp_linear.py:214
• test_data here is a callable provided by common.parametrize (dict values are lambdas), not a torch.Tensor. The current annotation is incorrect/misleading.

---

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