This issue exists to document c++ compilation flags/annotations to try for speed.
Many are tried and benchmarked at #942; this documents others we may wish to include
[x] -funroll-loops
[x] -ffast-math
[x] -march=*
[ ] __restrict keyword
[ ] -ffp-contract=fast
[ ] other compilers: clang/lcc
[ ] std::max instead of masking
[ ] -flto and -fno-semantic-interposition
And note that zmm (AVX512) may hurt as it may cause CPUs to downclock.
This issue exists to document c++ compilation flags/annotations to try for speed. Many are tried and benchmarked at #942; this documents others we may wish to include
And note that zmm (AVX512) may hurt as it may cause CPUs to downclock.