ChipSort is a sorting module containing SIMD and cache-aware techniques. It's based on a couple of academic papers from 2008. More details can be found in our documentation.
Like any experimental Julia package on GitHub you can install ChipSort from the Julia REPL by first typing ] to enter the package management prompt, and then
pkg> add https://github.com/nlw0/ChipSort.jl
You can now try out the basic functions offered by the package such as sort_net to use a sorting network, or try the full array sort function prototype chipsort.
julia> using ChipSort
julia> using SIMD
julia> data = [Vec(tuple(rand(Int8, 4)...)) for _ in 1:4]
4-element Array{Vec{4,Int8},1}:
<4 x Int8>[-15, 98, 5, -28]
<4 x Int8>[47, -112, 98, -14]
<4 x Int8>[-18, -3, -111, 85]
<4 x Int8>[79, -12, -44, -85]
julia> x = sort_net(data...)
(<4 x Int8>[-18, -112, -111, -85], <4 x Int8>[-15, -12, -44, -28], <4 x Int8>[47, -3, 5, -14], <4 x Int8>[79, 98, 98, 85])
julia> y = transpose_vecs(x...)
(<4 x Int8>[-18, -15, 47, 79], <4 x Int8>[-112, -12, -3, 98], <4 x Int8>[-111, -44, 5, 98], <4 x Int8>[-85, -28, -14, 85])
julia> z = merge_vecs(y...)
<16 x Int8>[-112, -111, -85, -44, -28, -18, -15, -14, -12, -3, 5, 47, 79, 85, 98, 98]
julia> bigdata = rand(Int32, 2^20);
julia> chipsort_large(bigdata, Val(8), Val(32)) == sort(bigdata)
true
Make sure you check our documentation for more information.
Latest benchmark results are: 81% speedup on a 1M Int32 array, 2x speedup on 8k Int32 and 4x on 64 values.