Closed beomki-yeo closed 5 months ago
Ok then i will take the change to the detray benchmark side
Notice also how the ATLAS code skips on setting anything itself if an -march
flag is already set. Let's say through the $CXXFLAGS
environment variable. (Which is something that we do in some special nightlies of ours.) We may want to look out for such a thing even in the R&D projects. :thinking: (Since one could argue that CMake should never set such things at all. And it would be up to a layer above it, let's say in Spack, where we would request such optimizations. But since I'm not a fan of Spack myself, I'm still okay with CMake making some of these decisions itself...)
This PR turns on the CPU vectorization for
array_cmath
which will be followed by dedicated optimization in #116Not sure if we need to do this commonly for any other plugins. maybe I also need to add cmake condition that checks if the CPU supoorts
avx
. And is there any potential risk or downside in enabling vectorization?At the moment, it seems to be useful as detray CPU benchmark gets faster about 10% by enabling the vectorization.
detray main
detray with this PR