Add gfx950 ROCm CI for AOTI

[digantdesai]

Add an opt-in, path-filtered ROCm job on the gfx950 runner. Build the HIP runtime and native runner, exercise targeted Python and C++ coverage, verify generated AOTI code and HIP-only linkage, and provide a generic ROCm LLM preset.

[pytorch-bot]

🔗 Helpful Links

🧪 See artifacts and rendered test results at hud.pytorch.org/pr/pytorch/executorch/21957

• 📄 Preview Python docs built from this PR

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

❌ 2 New Failures

As of commit 2631173 with merge base 910f816 ():

NEW FAILURES - The following jobs have failed:

• Test CUDA Builds / export-model-cuda-artifact (SocialLocalMobile, Qwen3.5-35B-A3B-HQQ-INT4, quantized-int4-tile-packed) / linux-job (gh)
  RuntimeError: Command bash /__w/_temp/exec_script failed with exit code 1
• trunk / test-arm-backend-ethos-u (test_pytest_models_ethos_u85) / linux-job (gh)
  ##[error]fatal: unable to access 'https://github.com/pybind/pybind11.git/': Failed to connect to github.com port 443 after 132199 ms: Could not connect to server

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

[pytorch-bot]

Warning: Unknown label ciflow/rocm.
Currently recognized labels are

• ciflow/android
• ciflow/apple
• ciflow/cuda
• ciflow/cuda-perf
• ciflow/metal
• ciflow/mlx
• ciflow/nightly
• ciflow/trunk
• ciflow/binaries
• ciflow/binaries/all
• ciflow/periodic

Please add the new label to .github/pytorch-probot.yml

[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.

[Copilot]

Pull request overview

Adds an opt-in ROCm (HIP) CI workflow intended to validate ExecuTorch’s AOTI CUDA backend on AMD GPUs, along with build presets and documentation updates to support ROCm-focused LLM builds and targeted test execution.

Changes:

• Introduces a new GitHub Actions workflow (rocm.yml) and CI script to build and validate ROCm AOTI (native runner build, targeted C++ tests, targeted Python tests, linkage checks, and an example export).
• Adds ROCm CMake presets/workflow preset for LLM release builds and updates ROCm documentation to point to the preset workflow.
• Improves CUDA fused MoE test runner configurability by allowing the executor_runner path to be overridden via environment variable.

Reviewed changes

Copilot reviewed 6 out of 6 changed files in this pull request and generated 2 comments.

Show a summary per file

File	Description
CMakePresets.json	Adds ROCm LLM release configure/build/workflow presets (install-oriented).
backends/cuda/tests/test_fused_moe.py	Allows overriding the native runner path via EXECUTORCH_EXECUTOR_RUNNER.
backends/cuda/rocm.md	Updates ROCm build instructions to use the new CMake workflow preset and updates CI notes.
.github/workflows/rocm.yml	Adds a path-filtered ROCm AOTI CI workflow using the test-infra Linux job runner.
.github/pytorch-probot.yml	Registers ciflow/rocm push tag for triggering ROCm CI.
.ci/scripts/test-rocm-aoti.sh	Implements the end-to-end ROCm AOTI CI script (build, test, linkage verification, export).

---

💡 Add a code-review agent skill or configure MCP servers for context-aware, tailored reviews. Learn more in the docs.

[Copilot]

Pull request overview

Copilot reviewed 6 out of 6 changed files in this pull request and generated no new comments.

Suppressed comments (1)

.github/workflows/rocm.yml:84

• The changed-file gate only matches single-level paths like backends/cuda/*, so edits under nested directories (e.g. backends/cuda/tests/...) won’t set RUN_AOTI=true. That means the ROCm job can be silently skipped even though the workflow triggers on backends/cuda/** at the event level.

              backends/aoti/* | \
              backends/cuda/* | \
              extension/cuda/* | \
              examples/cuda/*)

[Copilot]

Pull request overview

Copilot reviewed 6 out of 6 changed files in this pull request and generated no new comments.

Suppressed comments (2)

.github/workflows/rocm.yml:77

• aoti-run-decision should also consider torch_pin.py as an input, since .ci/scripts/test-rocm-aoti.sh imports it to select the PyTorch version. Otherwise pushes that only change torch_pin.py may skip ROCm CI unexpectedly.

              CMakeLists.txt | \
              CMakePresets.json | \
              install_requirements.py | \
              backends/aoti/* | \
              backends/cuda/* | \

.github/workflows/rocm.yml:21

• The ROCm CI script reads TORCH_VERSION via from torch_pin import TORCH_VERSION (in .ci/scripts/test-rocm-aoti.sh), but torch_pin.py is not included in the pull_request.paths filter. Changes to torch pinning could affect this workflow without triggering ROCm CI on PRs.

This issue also appears on line 73 of the same file.

      - CMakeLists.txt
      - CMakePresets.json
      - install_requirements.py
      - backends/aoti/**

[Copilot]

Pull request overview

Copilot reviewed 6 out of 6 changed files in this pull request and generated no new comments.