[mapleFU]

Rationale for this change

See: https://godbolt.org/z/EPPqn7xco

What changes are included in this PR?

Extract the alp argument out, this brings optimization in clang ( at least clang-18 ).

According to the godbolt, gcc can optimize this well, so using gcc might not seeing this optimization.

Are these changes tested?

No

Are there any user-facing changes?

No

Was AI used for this PR?

AI help me to test the cases in different machines

PR code and description written by:

• AI

Reviewed before submission by:

• Human

• GitHub Issue: [C++][Parquet] Enhance ALP performance when decoding in clang #55247

[github-actions]

Thanks for opening a pull request!

This pull request has been automatically converted to a draft because its title doesn't match Arrow's required format.

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}

After updating the title, you can mark the pull request as ready for review.

See also:

• Other pull requests
• Contribution Guidelines - Contributing Overview

[github-actions]

⚠️ GitHub issue #55247 has been automatically assigned in GitHub to PR creator.

[Copilot]

🟢 Approval recommended

The localized change preserves decoding semantics and has no identified blocking issues; performance gains were not independently verified.

No open issues 🎉

What changed in this PR

This PR hoists ALP exponent/factor retrieval out of the C++ decompression loop to improve Clang vectorization without changing decoding behavior.

Changes:

• Copies exponent/factor metadata once before decoding.
• Documents how reading the byte-sized metadata inside the loop can inhibit vectorization.

File	Description
cpp/​src/​arrow/​util/​alp/​alp_compression.cc	Reuses cached exponent/factor metadata throughout the decode loop.

_{Review effort: Balanced}

[mapleFU]

Benchmark result on my EPYC EPYC 9654 and clang++18

benchmark | before | after | improvement
-- | -- | -- | --
BM_AlpDecodingFloat/1024 | 1015 ns | 354 ns | 2.9x
BM_AlpDecodingFloat/4096 | 3390 ns | 786 ns | 4.3x
BM_AlpDecodingFloat/32768 | 25.4 µs | 4.4–5.5 µs | 4.6–5.8x
BM_AlpDecodingFloat/65536 | 50.0 µs | 8.6 µs | 5.8x
BM_AlpDecodingDouble/65536 | 52.4 µs | 43.8 µs | 1.2x