search

tahoe-lafs / zfec

zfec -- an efficient, portable erasure coding tool
Other
373 stars 44 forks source link

Try switching to -O3, see if it speeds things up #102

Open itamarst opened 8 months ago

itamarst commented 8 months ago

-O3 enables auto-vectorization, among other compiler optimizations, so might make things run faster. Blocked on #81

itamarst commented 8 months ago

Some numbers from experiments:

Compiler Options Speed (MB/sec)
gcc 11 Default 360
gcc 11 -O3 230
clang 14 Default 375
clang 14 -O3 377
gcc 11 -march=x86-64-v2 478
gcc 11 -O3 -march=x86-64-v2 446
gcc 11 -O3 -march=x86-64-v3 446
itamarst commented 8 months ago

-O3 was not helpful!

Main speed up for compiler options, as seen above, is telling the compiler you're not on an ancient CPU. Here was RHEL 9's explanation of why they switched to this: https://developers.redhat.com/blog/2021/01/05/building-red-hat-enterprise-linux-9-for-the-x86-64-v2-microarchitecture-level

It's hard to get survey data, best I know of is Steam's. It suggests maybe... 0.6% of computers in their survey won't work with x86-64-v2 due to e.g. lack of SSE4.2. https://store.steampowered.com/hwsurvey/Steam-Hardware-Software-Survey-Welcome-to-Steam