Fix quantized Conv1d same padding with even kernels in XNNPACK

[SakshamKapoor2911]

Fixes #20558.
Related to #20553.

Summary

Quantized nn.Conv1d(..., padding="same") with an even kernel exports as:

dequant -> constant_pad_nd -> convolution

XNNPACK's Conv1d lowering maps Conv1d through Conv2d by inserting an extra width dimension. For even-kernel same-padding, the explicit temporal pad has to be represented as asymmetric Conv2d input padding after that 1d-to-2d mapping. Otherwise the Conv1d case either remains partially undelegated or reaches XNNPACK with mismatched output metadata and fails/produces incorrect output.

This PR:

• pulls temporal-only zero constant_pad_nd into the quantized Conv1d partition,
• folds that temporal pad into XNNPACK asymmetric Conv2d input padding during Conv1d unsqueeze lowering,
• serializes folded asymmetric padding when present,
• repairs Conv1d folded-pad metadata after later XNNPACK passes retrace from the symmetric ATen convolution args,
• adds a regression for quantized Conv1d even-kernel padding="same".

Relationship to #20553

#20553 folds constant_pad_nd into convolution input padding for the Conv2d path. I tested its current head (a04093bba2766afe609ef325737effdd01c3c4ad) against the Conv1d repro from #20558. The odd-kernel case works, but the even-kernel Conv1d case still fails at runtime with xnn_status_invalid_parameter.

This PR handles the Conv1d-specific path where the temporal dimension is later mapped to Conv2d height and where metadata must be corrected after XNNPACK passes retrace from symmetric ATen conv args.

I noticed #20558 is assigned to @JakeStevens; happy to adapt or defer if there is already an internal Conv1d fix in flight.

Test plan

Local note: this checkout does not expose src/executorch/exir/_serialize/program.fbs as a package resource, so to_executorch() pytest runs used a temporary sitecustomize.py shim that redirects missing FlatBuffer resource reads to repo schema/. The manual repro uses the same schema-resource workaround. CI/dev install should package these resources normally.

Manual quantized Conv1d repro on patched main:

kernel_size=3: max abs diff = 0.000000
kernel_size=4: max abs diff = 0.000000

Additional manual quantized Conv1d same-padding matrix:

seed=0 length=16 k=2 dilation=1 bias=True:  max_abs_diff=0.000000 shape=(1,4,16)
seed=1 length=16 k=4 dilation=1 bias=True:  max_abs_diff=0.000000 shape=(1,4,16)
seed=2 length=15 k=4 dilation=1 bias=False: max_abs_diff=0.000000 shape=(1,4,15)
seed=3 length=17 k=6 dilation=1 bias=True:  max_abs_diff=0.000000 shape=(1,4,17)
seed=4 length=16 k=4 dilation=3 bias=True:  max_abs_diff=0.000000 shape=(1,4,16)

Commands run:

PYTHONPATH=/tmp/executorch20558_site:src python -m pytest backends/xnnpack/test/ops/test_conv1d.py::TestConv1d::test_qs8_conv1d_even_kernel_same_padding -q
# 1 passed

PYTHONPATH=/tmp/executorch20558_site:src python -m pytest backends/xnnpack/test/ops/test_conv1d.py -q
# 7 passed

PYTHONPATH=/tmp/executorch20558_site:src python -m pytest backends/xnnpack/test/ops/test_conv2d.py -q
# 28 passed

PYTHONPATH=/tmp/executorch20558_site:src python -m pytest backends/xnnpack/test/ops/test_static_constant_pad.py -q
# 8 passed

python -m py_compile \
  backends/xnnpack/_passes/conv1d_unsqueeze_pass.py \
  backends/xnnpack/operators/op_conv2d.py \
  backends/xnnpack/partition/config/gemm_configs.py \
  backends/xnnpack/test/ops/test_conv1d.py

python -m black --check \
  backends/xnnpack/_passes/__init__.py \
  backends/xnnpack/_passes/conv1d_unsqueeze_pass.py \
  backends/xnnpack/operators/op_conv2d.py \
  backends/xnnpack/partition/config/gemm_configs.py \
  backends/xnnpack/test/ops/test_conv1d.py

python -m flake8 \
  backends/xnnpack/_passes/__init__.py \
  backends/xnnpack/_passes/conv1d_unsqueeze_pass.py \
  backends/xnnpack/operators/op_conv2d.py \
  backends/xnnpack/partition/config/gemm_configs.py \
  backends/xnnpack/test/ops/test_conv1d.py

python -m ufmt check \
  backends/xnnpack/_passes/__init__.py \
  backends/xnnpack/_passes/conv1d_unsqueeze_pass.py \
  backends/xnnpack/operators/op_conv2d.py \
  backends/xnnpack/partition/config/gemm_configs.py \
  backends/xnnpack/test/ops/test_conv1d.py

git diff --check

[pytorch-bot]

🔗 Helpful Links

🧪 See artifacts and rendered test results at hud.pytorch.org/pr/pytorch/executorch/20734

• 📄 Preview Python docs built from this PR

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

⚠️ 11 Awaiting Approval

As of commit 529a792 with merge base 4af91c3 ():

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

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

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

[linux-foundation-easycla]

The committers listed above are authorized under a signed CLA.

• ✅ login: SakshamKapoor2911 / name: SakshamKapoor2911 (e7636f5)

[Copilot]

Pull request overview

This PR fixes an XNNPACK delegation/runtime correctness issue for statically-quantized nn.Conv1d(padding="same") with even kernel sizes by folding the explicit temporal constant_pad_nd into asymmetric Conv2d padding during Conv1d→Conv2d lowering, and ensuring the serialized graph’s metadata matches the folded asymmetric padding.

Changes:

• Extend XNNPACK partitioning and Conv1d unsqueeze lowering to absorb temporal-only zero constant_pad_nd into the quantized Conv1d partition and fold it into asymmetric xnnpack_input_padding.
• Teach Conv2d serialization to use folded asymmetric padding metadata when present, and add a late pass to repair Conv1d folded-pad tensor metadata after retracing passes.
• Add a regression test covering quantized Conv1d even-kernel padding="same".

Reviewed changes

Copilot reviewed 5 out of 5 changed files in this pull request and generated 1 comment.

Show a summary per file

File	Description
backends/xnnpack/test/ops/test_conv1d.py	Adds a regression test for quantized Conv1d even-kernel padding="same" and factors out calibration sample generation.
backends/xnnpack/partition/config/gemm_configs.py	Extends convolution activation-dependency discovery to include temporal-only zero constant_pad_nd for quantized Conv1d so it can be partitioned/delegated.
backends/xnnpack/operators/op_conv2d.py	Uses folded asymmetric xnnpack_input_padding metadata (top/right/bottom/left) when present instead of symmetric ATen padding args.
backends/xnnpack/_passes/conv1d_unsqueeze_pass.py	Folds eligible temporal-only constant_pad_nd into asymmetric input padding during Conv1d→Conv2d lowering and adds a pass to restore correct tensor metadata for folded-pad Conv1d.
backends/xnnpack/_passes/init.py	Registers the new folded-pad metadata repair pass in the default XNNPACK pass pipeline.

---

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

[pytorch-bot]

❌ 🤖 pytorchbot command failed:

Got EOF while in a quoted string```
Try `@pytorchbot --help` for more info.

[SakshamKapoor2911]

@pytorchbot label "release notes: xnnpack"