This helps performance with SSE (somewhat) and AVX-512 (quite a bit), but not AVX2 for some reason.
@(optimization_mode)