9 releases
| new 0.2.24 | Mar 18, 2025 |
|---|---|
| 0.2.23 | Mar 17, 2025 |
| 0.2.22 | Feb 3, 2024 |
| 0.2.21 | Apr 20, 2022 |
| 0.1.1 | Apr 16, 2022 |
#4 in #avx2
160 downloads per month
Used in hamming-bitwise-fast
1.5MB
585 lines
Contains (WOFF font, 400KB) NanumBarunGothic-13b3dcba.ttf.woff2, (WOFF font, 135KB) FiraSans-Medium-e1aa3f0a.woff2, (WOFF font, 130KB) FiraSans-Regular-0fe48ade.woff2, (WOFF font, 82KB) SourceSerif4-Bold-6d4fd4c0.ttf.woff2, (WOFF font, 77KB) SourceSerif4-Regular-6b053e98.ttf.woff2, (WOFF font, 45KB) SourceCodePro-It-fc8b9304.ttf.woff2 and 3 more.
hamming_rs
Computes hamming distance and weight, possibly with avx/avx2 instructions for x86 processors
avx2 optimized version is used when inputs have same memory alignment and are large enough
otherwise, functions will fallback to slower versions but still faster than strsim or hamming crates.
references
Muła, Wojciech, Nathan Kurz, and Daniel Lemire. "Faster population counts using AVX2 instructions." The Computer Journal 61.1 (2018): 111-120.
https://arxiv.org/pdf/1611.07612.pdf
Thanks to @emschwartz (github), his fast implementation is now used when avx2 is not available. It allows auto-vectorization on non intel platforms.
performances
We ran benchmarks with cargo bench on a laptop with a AMD Ryzen 5 PRO 5675U @4.4Ghz with 32Gb of RAM.
Before that, we set export RUSTFLAGS="-C target-feature=+avx2,+fma".
We compare ourselves against hamming and strsim reference crates.
We notice a speedup of ~3 against hamming and of ~30 against strsim
