Add support for AVX2?

[danielealbano]

It looks like more or less all the SIMD instructions in use have an AVX2 variant and this would provide about a 2x speed improvement, both on Intel and AMD cpus.

Would be cool to get that 😎

[Daniel-Liu-c0deb0t]

there's going to be one difficulty with shifting the simd vector by an amount only known at runtime. avx2 is really annoying to deal with due to its 128-bit lanes