[Nyan]

Very useful. In fact, it speeds up a single instance (i.e. not taking advantage of SIMD) of MD5 by 20%: https://github.com/animetosho/md5-optimisation#x86-avx512-vl...