Re:XXH3 – a new speed-optimized hash algorithm
by Cyan4973 on Friday March 15, 2019 @04:14PM
(#19403793)
wyhash is a close relative to mumhashv2, and feature approximately the same performance strength and weaknesses.
Re:XXH3 – a new speed-optimized hash algorithm
by rurban on Friday March 15, 2019 @07:20PM
(#19405240)
Working on it tomorrow. This just got in today. I expect xxh3 to beat everybody on long keys, and because it uses optimized intrinsics it should be close to the optimized th1a variants. wyhash is purely portable code.
Re:XXH3 – a new speed-optimized hash algorithm
by rurban on Saturday March 16, 2019 @07:55AM
(#19407370)
Results are now at smhasher. I had to fix all the MSVC issues first.
wyhash still is the recommended hash function: portable and lowest cycles/hash.
However xxxh3 is spectacularly fast, for small and esp. large key sizes. This is now the recommended hash for file or db digests.
String hash table stats not yet done, there FNV1A is still the recommendation, with safe collision handling.
Re:XXH3 – a new speed-optimized hash algorithm
by JanecekPetr on Friday March 15, 2019 @05:34PM
(#19404530)
Here's the smhasher output for XXH3 on an i9: https://pastebin.com/ryLN24Qy, the wyhash has its output in its readme.
TLDR: XXH3 uses less cycles per key, by about 30%. The real throughput / latency need a more equal and thorough benchmark of both, though.
Anyone have a TL;DR regarding how these two compare?