[Documenting] Failed attempt at optimizing SipHash with SSE2 #17774

issue elichai opened this issue on December 19, 2019

elichai commented at 2:13 PM on December 19, 2019: contributor
This is mostly for future reference. I tried implementing SipHashUint256Extra with SSE2 instructions but when seeing it's half of the speed of the naive C impl I gave up on trying to optimize it more. (afterwards I tried benchmarking the implementations in supercop and the SSE ones were also slower than the naive)

diff: https://gist.github.com/elichai/134b95fee25f8170fcdc69535f8f8bd4

compiled with: CXXFLAGS='-g -O3 -march=native' and got:
```
# Benchmark, evals, iterations, total, min, max, median
SaltedSipHash, 5, 40000000, 4.27147, 2.09707e-08, 2.22874e-08, 2.10788e-08
SaltedSipHashX86, 5, 40000000, 8.53348, 4.12294e-08, 4.45225e-08, 4.23751e-08
```
fanquake added the label Brainstorming on Dec 19, 2019
MarcoFalke closed this on Apr 26, 2020
DrahtBot locked this on Feb 15, 2022

Contributors

Labels