search

rurban / smhasher

Hash function quality and speed tests
https://rurban.github.io/smhasher/
Other
1.85k stars 178 forks source link

prvhash #275

Closed avaneev closed 1 year ago

avaneev commented 1 year ago

Please update the speeds of all presented prvhash hashes. The latest update includes "always_inline" attribute like in the recent "komihash" update. Maybe that can make a difference. There was also a change to function parameters (no InitVec is available now).

rurban commented 1 year ago

done

avaneev commented 1 year ago

Suspicious that previously it was 260.27 cycles and now its 361.73 cycles for the same code? But I won't be playing your game, your karma is your responsibility. Same applies for komihash speeds.

avaneev commented 1 year ago

The code of prvhash64_64m() was actually reverted to v4.3 when it had even lower timings per your test - 241.19 (cycl./map). Personal bias is definitely at play, that's without doubt now.

rurban commented 1 year ago

looks like the recent amd firmware patches slowed everything down. On Intel also. Would need a full test speed update for all. I'll try this over the weekend, but then I'd need to disable all cronjobs

avaneev commented 1 year ago

Hmm? Zenbleed affects Zen2 processors only, and komihash timings did not increase by 100... Just saying.

rurban commented 1 year ago

Updated all non-crypto speeds. Needed only 100 minutes

avaneev commented 1 year ago

Thanks.