Does locally compiled Julia emit faster code?

I’m working on a reasonably high-end platform (Intel Xeon E5). Will a generic binary of julia be able to leverage all the various optimizations of this CPU when compiling Julia code? Or do I need to compile from source locally?

I read a thread somewhere on here that generic Julia binaries don’t use AVX CPU instructions and locally compiled ones do. Is this still true and does the same apply to other CPU abilities?

Thanks!

3 Likes

On the early days of Julia, that was true, but currently most CPU names and fearures are detected and supported automatically. So, in most cases there is no actual benefit from building a local binary on our machines. See for example this file: https://github.com/JuliaLang/julia/blob/master/src/processor_x86.cpp and this: https://github.com/JuliaLang/julia/blob/master/src/features_x86.h.

3 Likes