Arm backend: Rework reporting of qspecs and qconfigs

[martinlsm]

This PR contains four commits:

1. Move quantizer_reporter out from cortex_m.quantizer
   When importing backends.cortex_m.quantizer.quantizer_reporter from
   backends.arm.quantizer.arm_quantizer_utils, a cyclic dependency chain is
   is formed. The problem is that quantizer_reporter triggers
   backends/cortex_m/quantizer/__init__.py when imported, which in turn has
   imports leading back to the Arm backend. To fix this problem, move
   quantizer_reporter to backends/cortex_m so it can be imported without
   forming any cycle.

2. Arm backend: Remove _QuantizerReporterUserMixin
   _QuantizerReporterUserMixin was a duplicated class with the name
   QuantizerReporterUser. Remove the former to instead use the latter.

3. Arm backend: Add label attribute to QuantizationConfig
   The quantizer reporter logs the quantization config in a human-readable
   format. Prior to this patch, this was done with the help of a dict
   called SUPPORTED_QCONFIGS, which was defined in quantizer_reporter.py
   and populated by the user. This patch reworks this concept by instead
   adding a label attribute to QuantizationConfig that the reporter can
   use to print the config in a human-readable format.

4. Arm backend: Rework reporting of qspecs
   The quantization reporter prints quantization specs in human-readable
   format. Prior to this patch, this was implemented such that
   quantizer_reporter.py defined a dict SUPPORTED_QSPECS which was
   populated by the user. This dict would map qspec objects to string
   representations. This patch removes this dict and instead modifies
   the helper function _qspec_repr to return a compact string
   representation based on the attributes of the qspec.

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

[pytorch-bot]

🔗 Helpful Links

🧪 See artifacts and rendered test results at hud.pytorch.org/pr/pytorch/executorch/19016

• 📄 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 & trunk workflows in PyTorch main

❌ 3 New Failures, 3 Unrelated Failures

As of commit 1076083 with merge base 6be4fb5 ():

NEW FAILURES - The following jobs have failed:

• Cadence Build & Test / cpu-test / test-aot / test-aot (gh)
  backends/cadence/aot/tests/test_replace_ops_passes.py::TestReplaceOpsPasses::test_replace_transposed_conv_with_linear_0
• pull / test-multimodal-linux (gemma3-4b) / linux-job (gh)
  RuntimeError: Command docker exec -t e800aff921fdfda329dca66bfd3a664b1b0b392cec9d09dc1bb17dab25f359b4 /exec failed with exit code 139
• trunk / test-models-macos-cpu (resnet50, xnnpack-quantization-delegation) / macos-job (gh)
  RuntimeError: Command bash /Users/ec2-user/runner/_work/_temp/exec_script failed with exit code 1

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)
  ##[error]The operation was canceled.
• pull / unittest-editable / 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.

[martinlsm]

@pytorchbot label ciflow/trunk

[martinlsm]

@pytorchbot label "partner: arm"

[martinlsm]

@pytorchbot label "release notes: none"

[Copilot]

Pull request overview

This PR refactors quantization reporting for Arm/Cortex-M backends to avoid import cycles and to make qconfig/qspec reporting self-describing (via QuantizationConfig.label and attribute-based qspec formatting) rather than relying on external registration dicts.

Changes:

• Move quantizer_reporter to backends/cortex_m/ and update imports/build targets to eliminate a Cortex-M ↔ Arm cyclic dependency.
• Replace qconfig/qspec “registry dict” reporting with QuantizationConfig.label and a new qspec_repr() helper for compact, attribute-based qspec formatting.
• Update Arm/Cortex-M quantizers and tests to use the new reporter API (QuantizerInfo.qconfig_label) and new string representations.

Reviewed changes

Copilot reviewed 9 out of 9 changed files in this pull request and generated 1 comment.

Show a summary per file

File	Description
backends/cortex_m/test/misc/test_quantizer_reporter.py	Updates tests to use the relocated reporter module and asserts new qspec string formatting via qspec_repr().
backends/cortex_m/quantizer_reporter.py	Relocates and refactors reporter: removes qspec/qconfig registries, adds qspec_repr(), and changes QuantizerInfo to use qconfig_label.
backends/cortex_m/quantizer/quantizer.py	Updates reporter import path to the new module location.
backends/cortex_m/quantizer/quantization_configs.py	Stops registering qspec/qconfig dicts; instead sets human-readable labels directly on config instances.
backends/cortex_m/quantizer/TARGETS	Removes per-subdir reporter target; depends on the new top-level //executorch/backends/cortex_m:quantizer_reporter.
backends/cortex_m/TARGETS	Adds a new Buck target exporting quantizer_reporter.py from the Cortex-M root.
backends/arm/quantizer/quantization_config.py	Adds optional label field to QuantizationConfig for debug/reporting.
backends/arm/quantizer/arm_quantizer_utils.py	Removes duplicated reporter mixin and switches to QuantizerReporterUser; updates qconfig labeling logic.
backends/arm/quantizer/arm_quantizer.py	Constructs and propagates meaningful label strings when creating quantization configs (instead of external registration).

---

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

[Copilot]

Pull request overview

This PR restructures Cortex-M/Arm quantization reporting to eliminate import cycles and to make qconfig/qspec reporting more self-describing (via labels and attribute-based qspec formatting).

Changes:

• Move quantizer_reporter to backends/cortex_m/quantizer_reporter.py and adjust Buck targets/imports accordingly.
• Remove the Arm-side _QuantizerReporterUserMixin in favor of QuantizerReporterUser, and update QuantizerInfo to use a qconfig_label.
• Add label to QuantizationConfig and rework qspec reporting to use qspec_repr() instead of user-populated registries.

Reviewed changes

Copilot reviewed 9 out of 9 changed files in this pull request and generated 1 comment.

Show a summary per file

File	Description
backends/cortex_m/test/misc/test_quantizer_reporter.py	Updates tests for new reporter module path and new qspec_repr() output (adds unit tests for representations).
backends/cortex_m/quantizer_reporter.py	Introduces qspec_repr(), renames QuantizerInfo field, and changes QuantizerReport to store QuantizerInfo directly.
backends/cortex_m/quantizer/quantizer.py	Updates import to the relocated QuantizerReporter.
backends/cortex_m/quantizer/quantization_configs.py	Removes reporter registry updates and sets label on predefined Cortex-M qconfigs.
backends/cortex_m/quantizer/TARGETS	Removes local quantizer_reporter library and depends on the new top-level cortex_m reporter target.
backends/cortex_m/TARGETS	Adds new Buck python_library target for quantizer_reporter.py.
backends/arm/quantizer/quantization_config.py	Adds optional label field to QuantizationConfig for debugging/visualization/reporting.
backends/arm/quantizer/arm_quantizer_utils.py	Removes duplicated reporter-user mixin; uses QuantizerReporterUser and reports qconfig_label from QuantizationConfig.label.
backends/arm/quantizer/arm_quantizer.py	Builds human-readable labels for produced quantization configs; removes old reporter registration logic.

Comments suppressed due to low confidence (1)

backends/cortex_m/quantizer_reporter.py:53

• qspec_repr()'s QuantizationSpec formatting currently only includes dtype and (optional) quant_min/quant_max. This collapses distinct qspecs (e.g., per-tensor vs per-channel, different qscheme/ch_axis/is_dynamic) to the same string, which can make the reporter output misleading and less useful for debugging. Consider including qscheme (and ch_axis when per-channel) and possibly is_dynamic in the representation so different specs remain distinguishable.

---

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