testing new AVX-512 support for vector indexing

[matthias314]

If you have a computer with AVX-512 (more precisely: AVX512BW), then you can help test the new code for vector indexing. The code supports AVX, AVX2 and AVX-512. However, I don't have access to a computer with AVX-512. On Linux, you can check if you have it by looking at the output of cat /proc/cpuinfo.

To test the AVX-512 code, download the shuffle branch

pkg> activate --temp
  Activating new project at `/tmp/jl_wVFDZB`

(jl_wVFDZB) pkg> add SmallCollections#shuffle

and then say

julia> using SmallCollections

julia> M = 32; v = -FixedVector{M,Int8}(1:M); p = ntuple(i -> Int8(i-1), M);

julia> SmallCollections.shuffle(Val(64), v, p)

The result should be equal to v. Once this works, we can compare it with AVX2 shuffles, for example via

julia> using Chairmarks

julia> @b SmallCollections.shuffle(Val(32), $v, $p), SmallCollections.shuffle(Val(64), $v, $p)

Your help is appreciated!