Arm backend: Support for uint8 I/O

[per]

Summary

Add support for quantization to uint8 for input and output tensors.

Test plan

Tested through CI

cc @digantdesai @freddan80 @zingo @oscarandersson8218 @mansnils @Sebastian-Larsson @robell

[pytorch-bot]

🔗 Helpful Links

🧪 See artifacts and rendered test results at hud.pytorch.org/pr/pytorch/executorch/18869

• 📄 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:

• Rolling out OSDC (ARC) runners on pull workflow for PyTorch trunk commits

❌ 9 New Failures, 2 Cancelled Jobs, 4 Pending, 3 Unrelated Failures

As of commit 796d6ff with merge base 2c545f8 ():

NEW FAILURES - The following jobs have failed:

• Apple / build-benchmark-app / macos-job (gh)
  RuntimeError: Command bash /Users/runner/work/_temp/exec_script failed with exit code 1
• pull / android / run-emulator (gh)
  The process '/usr/local/lib/android/sdk/platform-tools/adb' failed with exit code 224
• pull / test-mcu-cortex-m-backend / linux-job (gh)
  RuntimeError: Command docker exec -t db0c1dce6351c9046e8109fcfad1c1c4f7ee266f8d2408aa3d3136c8bcd259e4 /exec failed with exit code 1
• pull / unittest / macos / macos-job (gh)
  export/tests/test_target_recipes.py::TestTargetRecipes::test_mv3_model
• trunk / test-arm-backend-ethos-u (test_pytest_ops_ethos_u85) / linux-job (gh)
  RuntimeError: Command docker exec -t fed03eb1b44f8fb5a87c1d52f69f343bf36b04c118c922add77d62dac3da2849 /exec failed with exit code 1
• trunk / test-cortex-m-e2e (mv2) / linux-job (gh)
  RuntimeError: Command docker exec -t b965df96bc3c50be6f98bef196bb1d88a5d9afc71ae9c29d64f0fff49d4a0cdc /exec failed with exit code 92
• trunk / test-torchao-huggingface-checkpoints (lfm2_5_1_2b, linux.arm64.2xlarge, executorch-ubuntu-22.04-g... / linux-job (gh)
  RuntimeError: Command docker exec -t 3df333c42689aeec8cc77a31bc07a40dfddbe93e44dfd7a8d9540b7ad0f28ec0 /exec failed with exit code 1
• trunk / test-torchao-huggingface-checkpoints (phi_4_mini, linux.arm64.2xlarge, executorch-ubuntu-22.04-gc... / linux-job (gh)
  RuntimeError: Command docker exec -t fdb58caec7be29c8b82c07df3ae6d11046fc848ec2d319dfe2c1d0366d81823f /exec failed with exit code 1
• trunk / test-torchao-huggingface-checkpoints (qwen3_4b, linux.arm64.2xlarge, executorch-ubuntu-22.04-gcc1... / linux-job (gh)
  RuntimeError: Command docker exec -t d8e79365a271b0e0165d31a15e27e346b73e139fd697d104f956ef1e4d5d6ce1 /exec failed with exit code 1

CANCELLED JOBS - The following jobs were cancelled. Please retry:

• pull / unittest-editable / windows / windows-job (gh)
• Test CoreML Backend / test-coreml / test-backend-macos (coreml, models) / macos-job (gh)
  ##[error]The operation was canceled.

FLAKY - The following job failed but was likely due to flakiness present on trunk:

• trunk / unittest-release / macos / macos-job (gh) (detected as infra flaky with no log or failing log classifier)

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

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

• pull / unittest / windows / windows-job (gh) (trunk failure)
  ##[error]The operation was canceled.
• trunk / unittest-release / windows / windows-job (gh) (trunk failure)
  ##[error]The operation was canceled.

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

[Copilot]

Pull request overview

Adds uint8 input/output quantization support for Arm TOSA-based backends by representing unsigned semantics via TOSA RESCALE flags (while keeping internal tensors signless int8), plus plumbing to preserve IO Q/DQ nodes when requested.

Changes:

• Add uint8 IO quantization config and tests validating IO-only uint8 and numeric correctness across TOSA/VGF/Ethos-U55 flows.
• Extend TOSA RESCALE op (fake + lowering) with input_unsigned / output_unsigned flags and map torch.uint8 to TOSA INT8 dtype.
• Add preserve_io_quantization compile spec option and propagate it through pipelines; extend Ethos-U runtime input type support to uint8.

Reviewed changes

Copilot reviewed 16 out of 16 changed files in this pull request and generated 3 comments.

Show a summary per file

File	Description
backends/arm/tosa/partitioner.py	Adds compile-spec-driven IO Q/DQ detagging and rejects partitions with invalid internal uint8 tensors.
backends/arm/tosa/mapping.py	Maps torch.uint8 to TOSA INT8 (unsigned semantics via RESCALE flags).
backends/arm/tosa/dialect/ops/rescale.py	Extends RESCALE schema and validation for unsigned IO semantics.
backends/arm/_passes/insert_rescales_pass.py	Folds dq→q into RESCALE and threads unsigned flags; enforces “uint8 at IO only”.
backends/arm/operators/op_tosa_rescale.py	Plumbs unsigned flags into flatbuffer RESCALE attributes and visitor parsing.
backends/arm/common/arm_compile_spec.py	Adds preserve_io_quantization to compile specs with parsing/roundtrip + warning behavior.
backends/arm/runtime/EthosUBackend.cpp	Allows uint8 (ScalarType::Byte) inputs in runtime IO validation.
backends/arm/quantizer/arm_quantizer.py	Adds get_uint8_io_quantization_config() and exports it.
backends/arm/quantizer/arm_quantizer_utils.py	Adjusts shared-qspec clique logic to avoid pulling uint8 IO qspecs into shared propagation.
backends/arm/test/common.py	Adds preserve_io_quantization option to VGF compile spec helper.
backends/arm/test/tester/test_pipeline.py	Plumbs preserve_io_quantization through test pipeline configuration.
backends/arm/test/quantizer/test_uint8_io_quantization.py	New test ensuring uint8 IO quantization config applies at IO.
backends/arm/test/passes/test_ioquantization_pass.py	Adds extensive end-to-end and pass-level tests for uint8 IO behavior.
backends/arm/test/misc/test_rescale_range.py	Adds unsigned zero-point range validation tests for RESCALE.
backends/arm/test/misc/test_compile_spec.py	Adds tests for preserve_io_quantization roundtrip and warning behavior.

Comments suppressed due to low confidence (1)

backends/arm/tosa/dialect/ops/rescale.py:38

• The function docstring is no longer a real docstring because tosa_spec = get_context_spec() appears before the triple-quoted string. This makes the string a no-op expression and drops the doc from introspection/tools. Move the docstring to be the first statement in RESCALE (and assign tosa_spec after it).

    tosa_spec = get_context_spec()
    """Casts the input tensor to dtype `dtype` to produce the correct tensor
    meta for a _rescale op.

    Additionally validates TOSA constraints of a RESCALE op.

    """

---

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

[zingo]

Hi @digantdesai this adds files and you might want to tests is from a buck2 perspective