avx2_pshufb"vpshufb like it was meant to be" on AVX2
swizzle_dyn_splitImplements an arbitrary shuffle over double the native vector width
using 4 native-width shuffles
transizeThis sets up a call to an architecture-specific function, and in doing so
it persuades rustc that everything is the correct size. Which it is.
This would not be needed if one could convince Rust that, by matching on N,
N is that value, and thus it would be valid to substitute e.g. 16.