Skip to content

perf: vectorize small u8 table take with AVX2 - #9572

Open
joseph-isaacs wants to merge 1 commit into
ji/small-u8-table-take-neonfrom
ji/small-u8-table-take-avx2
Open

perf: vectorize small u8 table take with AVX2#9572
joseph-isaacs wants to merge 1 commit into
ji/small-u8-table-take-neonfrom
ji/small-u8-table-take-avx2

perf: vectorize small u8 table take with AVX2

134d1c5
Select commit
Loading
Failed to load commit list.
CodSpeed HQ / CodSpeed Performance Analysis failed Sep 9, 2026 in 0s

1 benchmark regressed

⚠️ 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.

⚡ 11 improved benchmarks
❌ 1 regressed benchmark
✅ 2181 untouched benchmarks
⏩ 218 skipped benchmarks1

Warning

Please fix the performance issues or acknowledge them on CodSpeed.

Performance Changes

Mode Benchmark BASE HEAD Efficiency
WallTime filtered_owned_i64_avx2[OneNullInEight] 21.9 µs 26.8 µs -18.4%
WallTime dict_canonicalize_gt_u8_avx2[1000000] 419.8 µs 58.2 µs ×7.2
WallTime dict_canonicalize_gt_u8_avx512[1000000] 419.5 µs 58.8 µs ×7.1
WallTime dict_canonicalize_gt_u8_avx2[16000000] 6.8 ms 1.4 ms ×5
WallTime dict_canonicalize_gt_u8_avx512[16000000] 6.8 ms 1.4 ms ×5
Simulation decode_primitives[u8, (2000, 8)] 43.8 µs 33.4 µs +30.88%
Simulation decode_primitives[u8, (2000, 4)] 43.8 µs 33.4 µs +30.88%
Simulation decode_primitives[u8, (2000, 2)] 43.8 µs 33.4 µs +30.88%
Simulation decode_primitives[u8, (1000, 8)] 36.3 µs 31.5 µs +15.07%
Simulation decode_primitives[u8, (1000, 4)] 36.3 µs 31.5 µs +15.06%
Simulation decode_primitives[u8, (1000, 2)] 37.1 µs 32.3 µs +14.69%
WallTime words_gather_scalar_avx2[65536] 9.4 µs 8.3 µs +13.35%

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 (134d1c5) with ji/small-u8-table-take-neon (e97db31)

Open in CodSpeed

Footnotes

  1. 218 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.