Though worth noting only for specific AVX-512 CPUs.

Though you can still have vectorised support for regular precision for AVX2.