[Gasoonjia]

Stack from ghstack (oldest at bottom):

• [executorch][cuda] Run transposed convolutions on the implicit-GEMM template #25091
• [executorch][cuda] Test 3-D convolutions on the implicit-GEMM template #25090
• [executorch][cuda] Run 1-D convolutions on the implicit-GEMM template #25089
• [executorch][cuda] Converted convolutions follow the output layout the graph recorded #25088
• [executorch][cuda] Route 3-input-channel window and kernel-equals-stride convolutions to the implicit-GEMM template #25087
• [executorch][cuda] Add an implicit-GEMM template for dense triton::convolution #25086
• [executorch][cuda] Route 2-D depthwise and few-channel grouped convolutions to a Triton direct-stencil kernel #25085
• -> [executorch][cuda] Add triton::convolution and the conv routing framework #25084
• [executorch][cuda] Add planned replacement to the Triton replacement pass #25083
• [executorch][cuda] Deprecate triton_kernel_mode=OFF #25082

The backend's Triton kernels have no convolution entry point. ATen spreads convolutions over many overloads whose arguments cannot express asymmetric or "same" padding once lowered, and Inductor's templates cannot compile some convolutions at all (dilated, most transposed and 1-D ones).

This diff adds triton::convolution, one canonical op with explicit padding_before/padding_after and separately defined forward and transposed semantics. bind_conv maps every ATen convolution overload of both lowering APIs to it, filling defaults and validating the arguments. out_shape_and_stride computes the output layout ATen's CUDA convolution returns; the op body and its fake share it, so the output is allocated once and templates only write into it. Templates declare supports(), worth_routing() and schedules, and force_conv_candidate lets tests run one schedule. conv.routing.plan is the planner for every convolution overload: it refuses symbolic shapes, non-CUDA inputs, ROCm and calls whose output layout differs from the node's, and marks calls Inductor cannot compile as REQUIRED. No template is registered yet, so lowering is unchanged.

test_conv_binding.py checks the binding, canonical arguments and output layouts against ATen; test_conv_routing.py checks the routing gates and failure contract on graphs from both lowering APIs, without a GPU.

This diff was authored with Claude Code.

Differential Revision: D129595183

NOTE FOR REVIEWERS: This PR has internal Meta-specific changes or comments, please review them on Phabricator!

[pytorch-bot]

🔗 Helpful Links

🧪 See artifacts and rendered test results at hud.pytorch.org/pr/pytorch/executorch/25084

• 📄 Preview Python docs built from this PR

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

❌ 1 New Failure, 1 Pending, 1 Unrelated Failure

As of commit d941f69 with merge base dea39f0 ():

NEW FAILURE - The following job has failed:

• pull / android / run-emulator (gh)
  adb: failed to install android-test-debug-androidTest.apk: cmd: Can't find service: package

BROKEN TRUNK - The following job failed but were present on the merge base:

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

• Test ROCm AOTI / test-voxtral-realtime-rocm-gfx950-rocm7.2 / linux-job (gh) (trunk failure)
  ##[error]The operation was canceled.

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.