Detect AVX-512 FMA units to choose best SIMD
Add a test program that times AVX-512 code to
detect single vs. dual AVX-512 FMA units.
Added CMake code to always compile this file
with AVX-512 flags, both at CMake configuration
and runtime.
Tested to work on both AVX2 and AVX-512 hardware
with dual FMAs, and by manually faking single
FMA units, but when we get access to hardware
with a single AVX-512 FMA unit we need to check
that we produce the correct result.
Change-Id: I6240e864bc77f95085c5cd3303a84ab581eb3662