perf: vectorize small u8 table take with AVX2

[joseph-isaacs]

Adds AVX2 VPSHUFB small-table lookup on top of the NEON implementation merged in #9571. It handles u8 indices, up to 16 one-byte values, and at least 64 rows, with runtime AVX2 detection and the existing gather/scalar fallbacks.

The byte-value trait implementations live in the architecture-gated module. Table setup uses copy_from_slice, scalar tails use safe writes, and output buffers preserve the caller's allocator. Bounds tests cover invalid indices in both vector and scalar-tail positions on all targets.

For the 1M-row small-u8 dictionary benchmark, CodSpeed reports 55.4 µs versus 422.1 µs, a 7.6× speedup, on the AVX2 build. The AVX-512-enabled build runs the same AVX2 kernel and reports 55.2 µs. These measurements are from commit d138ae05b5, before the subsequent merge of develop.

Benchmark comparison

Validation: all reported CI checks on d138ae05b5 passed apart from pending review policy. After merging develop, the final diff was reviewed and new CI will validate the updated head; local tests were not rerun.

[codspeed]

Merging this PR will regress 1 benchmark

⚠️ 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

⚡ 8 improved benchmarks
❌ 1 regressed benchmark
✅ 2092 untouched benchmarks
⏩ 518 skipped benchmarks¹

Warning

Please fix the performance issues or acknowledge them on CodSpeed.

Performance Changes

	Mode	Benchmark	BASE	HEAD	Efficiency
❌	Simulation	take[small_m/shuffled/primitive/nonnull/chunks=2048/indices=64]	280.1 µs	481 µs	-41.76%
⚡	WallTime	dict_canonicalize_gt_u8_avx512[1000000]	422.3 µs	55.4 µs	×7.6
⚡	WallTime	dict_canonicalize_gt_u8_avx2[1000000]	423.4 µs	55.9 µs	×7.6
⚡	Simulation	decode_primitives[u8, (2000, 8)]	44.5 µs	33.5 µs	+32.83%
⚡	Simulation	decode_primitives[u8, (2000, 4)]	44.5 µs	33.5 µs	+32.83%
⚡	Simulation	decode_primitives[u8, (2000, 2)]	44.5 µs	33.5 µs	+32.83%
⚡	Simulation	decode_primitives[u8, (1000, 8)]	37 µs	31.5 µs	+17.45%
⚡	Simulation	decode_primitives[u8, (1000, 4)]	37 µs	31.5 µs	+17.45%
⚡	Simulation	decode_primitives[u8, (1000, 2)]	37.4 µs	31.8 µs	+17.43%

Tip

Investigate this regression by commenting @codspeedbot fix this regression on this PR, or directly use the CodSpeed MCP with your agent.

---

_{Comparing ji/small-u8-table-take-avx2 (25dce5c) with develop (e8e48c8)}

Footnotes

1. 518 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. ↩

[github-actions]

This PR has been marked as stale because it has been open for 14 days with no activity. Please comment or remove the stale label if you wish to keep it active, otherwise it will be closed in 7 days

[github-actions]

This PR has been marked as stale because it has been open for 14 days with no activity. Please comment or remove the stale label if you wish to keep it active, otherwise it will be closed in 7 days