Add quantize/dequantize_per_channel_group support to QNN backend

[Hyungkeun-Park-Nota]

Motivation

quantized_decomposed.quantize_per_channel_group and dequantize_per_channel_group are used for LLM weight-only quantization (e.g. int4 group-wise / per-channel-group). The QNN backend did not recognize these ops, causing two distinct failures:

1. The ops were decomposed away during torch.export instead of being preserved for the backend.
2. InsertIOQDQ raised a KeyError when a pre-quantized weight node annotated with dequantize_per_channel_group encoding fed the graph output, because the op was absent from q_dq_map.

This change adds the same treatment that torchao.quantize_affine / dequantize_affine already receive, applying it to the quantized_decomposed per-channel-group variants.

Changes

backends/qualcomm/builders/node_visitor.py

• Add both ops to q_ops / dq_ops sets.
• Add both ops to per_block_encoding so get_quant_encoding_conf routes them through make_qnn_per_block_config.

backends/qualcomm/partition/utils.py

• Add both ops to get_skip_decomp_table so they are preserved during export and reach the backend unchanged.

backends/qualcomm/_passes/utils.py

• Remap "scales" → QCOM_SCALE in get_quant_attrs, parallel to the existing "input_dtype" → QCOM_DTYPE remap for torchao ops, so AnnotateQuantAttrs correctly propagates the scale tensor for per-channel-group nodes.

backends/qualcomm/_passes/qnn_pass_manager.py

• Dynamically register both ops into node_visitor.q_ops / dq_ops in get_to_edge_transform_passes, following the same pattern used for torchao ops.

backends/qualcomm/_passes/insert_io_qdq.py

• Add per-channel-group entries to q_dq_map to fix the KeyError.

Testing

Unit test added to backends/qualcomm/tests/test_passes.py:

python -m pytest backends/qualcomm/tests/test_passes.py::TestPasses::test_insert_io_qdq_per_channel_group_no_key_error -v

The existing InsertIOQDQ tests also continue to pass:

python -m pytest backends/qualcomm/tests/test_passes.py::TestPasses::test_insert_io_qdq_handles_dequant_encoding backends/qualcomm/tests/test_passes.py::TestPasses::test_insert_io_qdq_no_revisit -v

The full QNN export pipeline (annotation → fold QDQ → insert IO QDQ → layout transform → partition) completes without errors for a quantized linear model. End-to-end execution on device was not verified in this environment.

[pytorch-bot]

🔗 Helpful Links

🧪 See artifacts and rendered test results at hud.pytorch.org/pr/pytorch/executorch/19629

• 📄 Preview Python docs built from this PR

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

❗ 2 Active SEVs

There are 2 currently active SEVs. If your PR is affected, please view them below:

• Run pull request jobs on OSDC runners in shadow mode
• Network migration for OSDC completed

⚠️ 11 Awaiting Approval

As of commit 79a29da with merge base 824cbff ():

AWAITING APPROVAL - The following workflows need approval before CI can run:

• Build documentation (gh)
• Build Presets (gh)
• Cadence Build & Test (gh)
• Lint (gh)
• periodic (gh)
• pull (gh)
• Test ARM Backend (gh)
• Test CoreML Backend (gh)
• Test QNN Backend (gh)
• Test Vulkan Backend (gh)
• Test XNNPACK Backend (gh)

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

[Hyungkeun-Park-Nota]

@pytorchbot label "release notes: qualcomm"