Skip to content

Refactor CI to produce 64-bit integer interface wheels for WoA #199

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
merged 1 commit into from
Jun 4, 2025

Conversation

Harishmcw
Copy link
Contributor

This PR refactors the CI workflow to use a matrix strategy for building both 32-bit and 64-bit integer interface wheels for Windows-on-ARM (WoA).

  • Simplifies the workflow and avoids code duplication.

  • Ensures both 32-bit and 64-bit integer interface wheels are produced for WoA.

  • I updated the version in pyproject.toml and made sure it matches git describe --tags --abbrev=8 in OpenBLAS at the OPENBLAS_COMMIT

@Harishmcw
Copy link
Contributor Author

Hi @mattip and @matthew-brett ,

This updated WoA CI script now also generates the 64-bit integer interface wheel, which we can use to build NumPy for Windows-on-ARM.

strategy:
matrix:
include:
- INTERFACE64: 1
Copy link
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Is anyone using 32-but builds of Python for Windows on ARM?

If so, I guess we do need the INTERFACE64=0; BUILD_BITS=32 option. But I guess we also need an INTERFACE64=0; BUILD_BITS=64 option, for people building Numpy or Scipy with 32-bit interfaces, but 64-bit builds.

Copy link
Collaborator

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

No, 32-bit Python for Windows on ARM is not a thing, there's only win_arm64.

We do need LP64 (scipy-openblas32 ) and ILP64 (scipy-openblas64) wheels though, that's orthogonal to interpreter bitness.

Copy link
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

If so, I guess we do need the INTERFACE64=0; BUILD_BITS=32 option. But I guess we also need an INTERFACE64=0; BUILD_BITS=64 option, for people building Numpy or Scipy with 32-bit interfaces, but 64-bit builds.

Maybe I was incorrect here - but my comment was meant to say that we do not want or need a INTERFACE64=0; BUILD_BITS=32 build - as we have here, but we do want and need an INTERFACE64=0; BUILD_BITS=64 option, that I think we do not have.

Copy link
Collaborator

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Maybe the matrix/environment variables needs improvement. Both wheels can be imported by 64-bit python win-arm64.

The "32-bit" build uses INTERFACE64: 0, BUILD_BITS: 32. It prints Building for 64-bit binary, 32-bit interface at the beginning of thebuild step, and later prints out Compiling a 64-bit binary

Copy link
Contributor Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Hi @matthew-brett,

Just to clarify:

  • In our workflow, BUILD_BITS does not refer to the binary architecture — it always builds 64-bit binaries for WoA. Instead, BUILD_BITS maps to the integer interface width used by OpenBLAS:

  • INTERFACE64=0 → 32-bit integer interface → BUILD_BITS=32 → scipy_openblas32

  • INTERFACE64=1 → 64-bit integer interface → BUILD_BITS=64 → scipy_openblas64

So we already have both:

  • INTERFACE64=0; BUILD_BITS=32 (LP64)
  • INTERFACE64=1; BUILD_BITS=64 (ILP64)

We use the BUILD_BITS variable purely as a naming and testing label — as seen here in the CI script:
windows-arm.yml#L76-L80

Copy link
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Shouldn't we then remove BUILD_BITS on the basis that it's confusing, and have some variable like INTERFACE_BITS set internally to 32 for INTERFACE64=0 etc?

@mattip mattip merged commit a9daa2a into MacPython:main Jun 4, 2025
19 checks passed
@mattip
Copy link
Collaborator

mattip commented Jun 4, 2025

Thanks @Harishmcw. Once the wheels are uploaded to PyPI we can consider using scipy-openblas64 in NumPy.

@mattip
Copy link
Collaborator

mattip commented Jun 5, 2025

I uploaded the 64-bit wheel. The next step is a PR to numpy/numpy to redo the win-arm64 wheel build to use it (simplifying the logic in tools/wheels/cibw_before_build.sh closer to what it was before numpy/numpy#29039) and releasing the condition on requirements/ci_requirements.txt.

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Labels
None yet
Projects
None yet
Development

Successfully merging this pull request may close these issues.

4 participants