Skip to content

Fix PyTorch matmul device placement#3638

Closed
FlorianPfaff wants to merge 4 commits into
mainfrom
matmul-device
Closed

Fix PyTorch matmul device placement#3638
FlorianPfaff wants to merge 4 commits into
mainfrom
matmul-device

Conversation

@FlorianPfaff

@FlorianPfaff FlorianPfaff commented Jul 1, 2026

Copy link
Copy Markdown
Owner

Summary

  • Preserve an existing non-CPU PyTorch tensor device when matmul receives mixed CPU/accelerator operands.
  • Keep raw pyrecest._backend.pytorch.matmul and the active public PyTorch backend facade synchronized.
  • Preserve existing dtype promotion before delegating to torch.matmul.

Bug fixed

matmul converted operands independently and then delegated to torch.matmul. A CPU tensor or array-like left operand paired with a CUDA/meta right operand could therefore run on or return a CPU tensor instead of preserving the existing accelerator operand.

Validation

  • Syntax-checked the updated backend-support module locally with py_compile.
  • Smoke-checked the wrapper logic locally with CPU-left/meta-right tensors.
  • Full repository test suite not run in this connector session.

@github-actions

github-actions Bot commented Jul 1, 2026

Copy link
Copy Markdown
Contributor

✅MegaLinter analysis: Success

Descriptor Linter Files Fixed Errors Warnings Elapsed time
✅ COPYPASTE jscpd yes no no 70.99s
✅ JSON prettier 7 0 0 0 1.0s
✅ JSON v8r 7 0 0 5.04s
✅ MARKDOWN markdownlint 68 0 0 0 1.72s
✅ MARKDOWN markdown-table-formatter 68 0 0 0 0.82s
✅ PYTHON black 1184 29 0 0 57.54s
✅ PYTHON isort 1184 52 0 0 3.29s
✅ REPOSITORY checkov yes no no 45.95s
✅ REPOSITORY gitleaks yes no no 11.78s
✅ REPOSITORY git_diff yes no no 0.17s
✅ REPOSITORY secretlint yes no no 44.04s
✅ REPOSITORY syft yes no no 4.6s
✅ REPOSITORY trivy-sbom yes no no 5.12s
✅ REPOSITORY trufflehog yes no no 17.34s
✅ YAML prettier 11 0 0 0 0.74s
✅ YAML v8r 11 0 0 10.85s
✅ YAML yamllint 11 0 0 0.47s

Notices

📣 MegaLinter 9.5.0 is out! Discover the new features and security recommendations in the release announcement. (Skip this info by defining SECURITY_SUGGESTIONS: false)

See detailed reports in MegaLinter artifacts

Your project could benefit from a custom flavor, which would allow you to run only the linters you need, and thus improve runtime performances. (Skip this info by defining FLAVOR_SUGGESTIONS: false)

  • Documentation: Custom Flavors
  • Command: npx mega-linter-runner@9.5.0 --custom-flavor-setup --custom-flavor-linters PYTHON_BLACK,PYTHON_ISORT,COPYPASTE_JSCPD,JSON_V8R,JSON_PRETTIER,MARKDOWN_MARKDOWNLINT,MARKDOWN_MARKDOWN_TABLE_FORMATTER,REPOSITORY_CHECKOV,REPOSITORY_GIT_DIFF,REPOSITORY_GITLEAKS,REPOSITORY_SECRETLINT,REPOSITORY_SYFT,REPOSITORY_TRIVY_SBOM,REPOSITORY_TRUFFLEHOG,YAML_PRETTIER,YAML_YAMLLINT,YAML_V8R

MegaLinter is graciously provided by OX Security
Show us your support by starring ⭐ the repository

@FlorianPfaff FlorianPfaff enabled auto-merge (squash) July 1, 2026 23:19
@FlorianPfaff ChatGPT Codex Connector

FlorianPfaff commented Jul 4, 2026

Copy link
Copy Markdown
Owner Author

Closing this as superseded. Current main already has the dedicated backend_support/_pytorch_matmul_device_contract.py hook plus real regression coverage in tests/backend_support/test_pytorch_matmul_device_contract.py. This older branch conflicts with that test file and carries a less complete inline wrapper/placeholder-test history.

auto-merge was automatically disabled July 4, 2026 17:32

Pull request was closed

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Reviewers

No reviews

Assignees

No one assigned

Labels

None yet

Projects

None yet

Milestone

No milestone

Development

Successfully merging this pull request may close these issues.

1 participant

@FlorianPfaff