Refactor bit-transpose benchmarks with macro and add walltime CI

[joseph-isaacs]

Summary

This PR refactors the bit-transpose benchmark suite to reduce code duplication and adds a dedicated CI job for measuring per-tier performance on x86 hardware with all CPU feature tiers available.

Key Changes

• Introduced bench_feature! macro: Consolidates the repetitive pattern of generating paired transpose/untranspose benchmarks for each CPU feature tier. The macro:

  • Generates <feature>_transpose and <feature>_untranspose benchmark functions
  • Supports optional runtime guards via guard = <expr> to skip benchmarks when CPU features are unavailable
  • Uses paste::paste! for identifier generation

• Refactored benchmark definitions: Replaced 16 individual benchmark functions with 5 macro invocations:

  • scalar and dispatch (always available)
  • bmi2 and vbmi (x86_64 only, with runtime guards)
  • neon (aarch64 only, always available)

• Added walltime CI job: New bench-bit-transpose-walltime workflow that:

  • Runs on dedicated x86 hardware (Ice Lake c6id.8xlarge) with all CPU feature tiers
  • Complements the existing CodSpeed simulation job which cannot execute AVX-512 VBMI
  • Measures actual wall-clock performance for each tier in CI
  • Uses -C target-feature=+avx2 to ensure baseline feature availability

Implementation Details

The macro approach enables "just works" local benchmarking—cargo bench exercises whatever the host CPU supports—while CI on feature-complete hardware measures all tiers. Runtime guards prevent benchmark failures on unsupported CPUs without requiring per-feature conditional compilation wiring.

https://claude.ai/code/session_018byoMez1xPQTscpobqk1ST

[CLAassistant]

Thank you for your submission! We really appreciate it. Like many open source projects, we ask that you sign our Contributor License Agreement before we can accept your contribution.
_{You have signed the CLA already but the status is still pending? Let us recheck it.}

[codspeed-hq]

Merging this PR will improve performance by 38.93%

⚠️ Unknown Walltime execution environment detected

Using the Walltime instrument on standard Hosted Runners will lead to inconsistent data.

For the most accurate results, we recommend using CodSpeed Macro Runners: bare-metal machines fine-tuned for performance measurement consistency.

⚠️ Different runtime environments detected

Some benchmarks with significant performance changes were compared across different runtime environments,
which may affect the accuracy of the results.

Open the report in CodSpeed to investigate

⚡ 4 improved benchmarks
✅ 139 untouched benchmarks
🆕 14 new benchmarks
⏩ 134 skipped benchmarks¹

Performance Changes

	Mode	Benchmark	BASE	HEAD	Efficiency
⚡	Simulation	for_pack_16_to_3_stack	1.8 µs	1.3 µs	+43.67%
⚡	Simulation	unpack_16_from_3_stack	2 µs	1.4 µs	+41.77%
⚡	Simulation	unchecked_unpack_16_from_3_stack	2.1 µs	1.5 µs	+37.65%
⚡	Simulation	pack_16_to_3_stack	2.4 µs	1.8 µs	+32.89%
🆕	WallTime	scalar_transpose	N/A	55.3 µs	N/A
🆕	WallTime	scalar_untranspose[u8]	N/A	45.3 µs	N/A
🆕	WallTime	dispatch_untranspose[u16]	N/A	51 µs	N/A
🆕	WallTime	dispatch_untranspose[u8]	N/A	45.8 µs	N/A
🆕	WallTime	dispatch_untranspose[u64]	N/A	5.6 µs	N/A
🆕	WallTime	dispatch_transpose	N/A	6 µs	N/A
🆕	WallTime	dispatch_untranspose[u32]	N/A	61.8 µs	N/A
🆕	WallTime	scalar_untranspose[u16]	N/A	51.3 µs	N/A
🆕	WallTime	scalar_untranspose[u64]	N/A	60.6 µs	N/A
🆕	WallTime	scalar_untranspose[u32]	N/A	62 µs	N/A
🆕	WallTime	vbmi_transpose	N/A	4.7 µs	N/A
🆕	WallTime	vbmi_untranspose	N/A	4.9 µs	N/A
🆕	WallTime	bmi2_untranspose	N/A	33.8 µs	N/A
🆕	WallTime	bmi2_transpose	N/A	33.7 µs	N/A

Tip

Curious why this is faster? Comment @codspeedbot explain why this is faster on this PR, or directly use the CodSpeed MCP with your agent.

---

_{Comparing claude/blissful-mccarthy-6wV2r (fd689c3) with develop (938100e)}

Footnotes

1. 134 benchmarks were skipped, so the baseline results were used instead. If they were deleted from the codebase, click here and archive them to remove them from the performance reports. ↩