GH-48960: [Python] Adds DEFAULT option in simd_level of cli.py

[chiranmoyf]

Rationale for this change

For aarch64 machines, except NONE, none of the existing options in simd_level of /dev/archery/archery/cli.py are valid. The error also occurs on x86_64, unless one of SSE4_2, AVX2, or AVX512 is specified.

With the DEFAULT option in simd_level, Neon paths will be activated for aarch64 machines, and SSE4_2 will be activated for x86_64.

What changes are included in this PR?

Adding DEFAULT option in simd_level of /dev/archery/archery/cli.py.

Are these changes tested?

Yes.

Are there any user-facing changes?

No.

• GitHub Issue: [Python] DEFAULT option in simd_level of cli.py missing #48960

[github-actions]

Thanks for opening a pull request!

If this is not a minor PR. Could you open an issue for this pull request on GitHub? https://github.com/apache/arrow/issues/new/choose

Opening GitHub issues ahead of time contributes to the Openness of the Apache Arrow project.

Then could you also rename the pull request title in the following format?

GH-${GITHUB_ISSUE_ID}: [${COMPONENT}] ${SUMMARY}

or

MINOR: [${COMPONENT}] ${SUMMARY}

See also:

• Other pull requests
• Contribution Guidelines - Contributing Overview

[raulcd]

Hi thanks for the PR.
This does not fit our definition of MINOR as we only allow minor PRs with small grammar or documentation changes. Could you open an issue?
Can you expand on how were you using, I am curious on the archery build usage.
DEFAULT is also the default value if you don't pass any simd_level.

[chiranmoyf]

Hi thanks for the PR. This does not fit our definition of MINOR as we only allow minor PRs with small grammar or documentation changes. Could you open an issue? Can you expand on how were you using, I am curious on the archery build usage. DEFAULT is also the default value if you don't pass any simd_level.

Sure, @raulcd we'll raise an issue.
Yes, DEFAULT is the default value if nothing is passed, but since it is missing in the simd_level list, DEFAULT is treated as an invalid argument.

Here's the exact error message.
Error: Invalid value for '--simd-level': 'DEFAULT' is not one of 'NONE', 'SSE4_2', 'AVX2', 'AVX512'.

[raulcd]

Thanks for the answer. My question about the usage is because we do not use the archery build command for Arrow's CI or on the codebase. It is not on documentation either. It seems the last reference to using it was around 6-7 years ago. I was curious on how are you using it as I don't think it's currently actively maintained.

[chiranmoyf]

We followed this documentation to build and benchmark Arrow using Archery.

[chiranmoyf]

Hi thanks for the PR. This does not fit our definition of MINOR as we only allow minor PRs with small grammar or documentation changes. Could you open an issue?
@raulcd We have raised a GitHub issue (#48960) for the build error.

[chiranmoyf]

@raulcd @kou Please let us know if any changes are required.
As per the documentation, we used Archery to benchmark and compare Arrow's performance across different architecture-specific code paths. Our main goal is to add SVE kernels to maximize performance on aarch64 machines. If there is another actively maintained tool that can be used to microbenchmark Arrow, please let us know.

[github-actions]

⚠️ GitHub issue #48960 has been automatically assigned in GitHub to PR creator.

[kou]

As per the documentation

It seems that the documentation uses archery benchmark run not archery build.

Does archery benchmark run (not archery build) accept the --simd-level option?

[raulcd]

I think we should deprecate and remove archery build as it's not maintained and is not expected to be up to date. In all documentations on how to build Arrow/PyArrow we never cover doing it via archery build and this seem to be a remnant from ancient times.

I am happy to merge this if this unblocks your current workflow but I still think we should remove the archery build command and you should possibly update your workflow.