perf: add AVX512-optimized is_ascii to fix native CPU regression

[bonega]

Summary

• Fix severe performance regression when compiling with -C target-cpu=native on AVX512 CPUs
• Add custom is_ascii implementation using AVX512BW intrinsics
• Runtime feature detection ensures compatibility with all systems

Problem

When compiling with -C target-cpu=native on AVX512-capable CPUs, LLVM generates poor code for stdlib's is_ascii(), causing 30x slowdowns in ASCII-heavy workloads.

Solution

Add src/simd.rs with AVX512-optimized is_ascii using:

• _mm512_movepi8_mask to extract MSB of 64 bytes at once
• _mm512_maskz_loadu_epi8 for safe tail handling
• Runtime detection via is_x86_feature_detected!("avx512bw")
• Fallback to stdlib on non-AVX512 systems

Benchmarks

Workload	Default compilation	With -C target-cpu=native
ASCII decode/encode	1.3-1.8x faster	30-35x faster

Test plan

• All existing tests pass (cargo test)
• Benchmarks show expected improvements
• Works on non-AVX512 systems (falls back to stdlib)

[github-actions]

Yore Benchmarks

Details

Benchmark suite	Current: a6dcb41	Previous: dd07916	Ratio
decode_checked/mostly_ascii/8	33 ns/iter (± 12)	34 ns/iter (± 13)	0.97
decode_checked/mostly_ascii/64	66 ns/iter (± 7)	70 ns/iter (± 14)	0.94
decode_checked/mostly_ascii/256	139 ns/iter (± 13)	143 ns/iter (± 12)	0.97
decode_checked/mostly_ascii/512	232 ns/iter (± 19)	228 ns/iter (± 21)	1.02
decode_checked/mostly_ascii/1024	415 ns/iter (± 24)	423 ns/iter (± 28)	0.98
decode_checked/mostly_ascii/2048	879 ns/iter (± 53)	888 ns/iter (± 47)	0.99
decode_checked/mostly_ascii/4096	1964 ns/iter (± 74)	2004 ns/iter (± 75)	0.98
decode_checked/ascii/8	9 ns/iter (± 0)	8 ns/iter (± 0)	1.13
decode_checked/ascii/64	8 ns/iter (± 0)	8 ns/iter (± 0)	1
decode_checked/ascii/256	10 ns/iter (± 0)	10 ns/iter (± 0)	1
decode_checked/ascii/512	16 ns/iter (± 0)	15 ns/iter (± 0)	1.07
decode_checked/ascii/1024	26 ns/iter (± 3)	25 ns/iter (± 0)	1.04
decode_checked/ascii/2048	51 ns/iter (± 4)	50 ns/iter (± 11)	1.02
decode_checked/ascii/4096	91 ns/iter (± 0)	91 ns/iter (± 3)	1
decode_checked/extended/8	35 ns/iter (± 1)	36 ns/iter (± 1)	0.97
decode_checked/extended/64	84 ns/iter (± 3)	85 ns/iter (± 2)	0.99
decode_checked/extended/256	259 ns/iter (± 3)	260 ns/iter (± 6)	1.00
decode_checked/extended/512	484 ns/iter (± 19)	485 ns/iter (± 89)	1.00
decode_checked/extended/1024	904 ns/iter (± 5)	902 ns/iter (± 11)	1.00
decode_checked/extended/2048	1758 ns/iter (± 48)	1760 ns/iter (± 100)	1.00
decode_checked/extended/4096	3464 ns/iter (± 392)	3478 ns/iter (± 32)	1.00
decode_lossy/all_bad/8	48 ns/iter (± 1)	39 ns/iter (± 0)	1.23
decode_lossy/all_bad/64	63 ns/iter (± 0)	63 ns/iter (± 1)	1
decode_lossy/all_bad/256	178 ns/iter (± 10)	173 ns/iter (± 17)	1.03
decode_lossy/all_bad/512	336 ns/iter (± 42)	323 ns/iter (± 7)	1.04
decode_lossy/all_bad/1024	647 ns/iter (± 21)	622 ns/iter (± 78)	1.04
decode_lossy/all_bad/2048	1256 ns/iter (± 7)	1220 ns/iter (± 36)	1.03
decode_lossy/all_bad/4096	2482 ns/iter (± 22)	2394 ns/iter (± 35)	1.04
decode_lossy/mostly_ascii/8	47 ns/iter (± 21)	39 ns/iter (± 15)	1.21
decode_lossy/mostly_ascii/64	65 ns/iter (± 6)	65 ns/iter (± 15)	1
decode_lossy/mostly_ascii/256	155 ns/iter (± 16)	148 ns/iter (± 17)	1.05
decode_lossy/mostly_ascii/512	245 ns/iter (± 21)	236 ns/iter (± 23)	1.04
decode_lossy/mostly_ascii/1024	429 ns/iter (± 24)	414 ns/iter (± 31)	1.04
decode_lossy/mostly_ascii/2048	806 ns/iter (± 40)	789 ns/iter (± 39)	1.02
decode_lossy/mostly_ascii/4096	1583 ns/iter (± 130)	1545 ns/iter (± 55)	1.02
encode_checked/mostly_ascii/8	42 ns/iter (± 17)	42 ns/iter (± 17)	1
encode_checked/mostly_ascii/64	126 ns/iter (± 3)	128 ns/iter (± 20)	0.98
encode_checked/mostly_ascii/256	456 ns/iter (± 6)	457 ns/iter (± 17)	1.00
encode_checked/mostly_ascii/512	905 ns/iter (± 70)	911 ns/iter (± 180)	0.99
encode_checked/mostly_ascii/1024	1921 ns/iter (± 46)	1801 ns/iter (± 103)	1.07
encode_checked/mostly_ascii/2048	3693 ns/iter (± 192)	3541 ns/iter (± 331)	1.04
encode_checked/mostly_ascii/4096	7053 ns/iter (± 294)	7027 ns/iter (± 206)	1.00
encode_checked/ascii/8	9 ns/iter (± 0)	8 ns/iter (± 0)	1.13
encode_checked/ascii/64	8 ns/iter (± 0)	8 ns/iter (± 0)	1
encode_checked/ascii/256	10 ns/iter (± 0)	11 ns/iter (± 0)	0.91
encode_checked/ascii/512	14 ns/iter (± 0)	14 ns/iter (± 0)	1
encode_checked/ascii/1024	25 ns/iter (± 0)	24 ns/iter (± 0)	1.04
encode_checked/ascii/2048	50 ns/iter (± 1)	50 ns/iter (± 6)	1
encode_checked/ascii/4096	91 ns/iter (± 2)	89 ns/iter (± 1)	1.02
encode_checked/extended/8	54 ns/iter (± 1)	52 ns/iter (± 1)	1.04
encode_checked/extended/64	206 ns/iter (± 3)	208 ns/iter (± 8)	0.99
encode_checked/extended/256	716 ns/iter (± 6)	705 ns/iter (± 83)	1.02
encode_checked/extended/512	1400 ns/iter (± 15)	1382 ns/iter (± 283)	1.01
encode_checked/extended/1024	2749 ns/iter (± 26)	2747 ns/iter (± 162)	1.00
encode_checked/extended/2048	5513 ns/iter (± 45)	5425 ns/iter (± 338)	1.02
encode_checked/extended/4096	10854 ns/iter (± 165)	10785 ns/iter (± 218)	1.01
encode_lossy/all_bad/8	55 ns/iter (± 1)	52 ns/iter (± 0)	1.06
encode_lossy/all_bad/64	251 ns/iter (± 4)	227 ns/iter (± 2)	1.11
encode_lossy/all_bad/256	890 ns/iter (± 11)	797 ns/iter (± 5)	1.12
encode_lossy/all_bad/512	1747 ns/iter (± 72)	1559 ns/iter (± 10)	1.12
encode_lossy/all_bad/1024	3484 ns/iter (± 54)	3094 ns/iter (± 20)	1.13
encode_lossy/all_bad/2048	6797 ns/iter (± 1146)	6138 ns/iter (± 129)	1.11
encode_lossy/all_bad/4096	13522 ns/iter (± 602)	12228 ns/iter (± 90)	1.11

This comment was automatically generated by workflow using github-action-benchmark.

[github-actions]

⚠️ Performance Alert ⚠️

Possible performance regression was detected for benchmark 'Yore Benchmarks'.
Benchmark result of this commit is worse than the previous benchmark result exceeding threshold 1.10.

Benchmark suite	Current: a6dcb41	Previous: dd07916	Ratio
decode_checked/ascii/8	9 ns/iter (± 0)	8 ns/iter (± 0)	1.13
decode_lossy/all_bad/8	48 ns/iter (± 1)	39 ns/iter (± 0)	1.23
decode_lossy/mostly_ascii/8	47 ns/iter (± 21)	39 ns/iter (± 15)	1.21
encode_checked/ascii/8	9 ns/iter (± 0)	8 ns/iter (± 0)	1.13
encode_lossy/all_bad/64	251 ns/iter (± 4)	227 ns/iter (± 2)	1.11
encode_lossy/all_bad/256	890 ns/iter (± 11)	797 ns/iter (± 5)	1.12
encode_lossy/all_bad/512	1747 ns/iter (± 72)	1559 ns/iter (± 10)	1.12
encode_lossy/all_bad/1024	3484 ns/iter (± 54)	3094 ns/iter (± 20)	1.13
encode_lossy/all_bad/2048	6797 ns/iter (± 1146)	6138 ns/iter (± 129)	1.11
encode_lossy/all_bad/4096	13522 ns/iter (± 602)	12228 ns/iter (± 90)	1.11

This comment was automatically generated by workflow using github-action-benchmark.