Data structures for threaded computing

General Usage Performance
20

If you have a parallel mapreduce(f, op, x) (e.g., reduce(op, Map(f), x) from Transducers.jl), a neat way to minimize allocation and call mul! would be to use LazyArrays. Something like this (untested):

using LazyArrays: @~
using Transducers: Map

z = reduce(Map(x -> @~ x'x), x; init=nothing) do a, b
    a === nothing ? copy(b) : a .+= b
end

See also: Parallel reductions

2 Likes