This case baffles me: using StructArrays n, s = 10^5, 100; llp = [rand(1:n,s) for _=1:n]; llw = [rand(-1:0.001:1,s) for _=1:n] function sort_vec!(pp::Vector{tp}, ww::Vector{tw}, by::Function=first) ::Nothing where {tp<:Integer, tw} sort!(StructArray((pp,ww)), by=by, alg=QuickSort) return noth…

If it’s the specialization heuristic, then the official workaround is to introduce a type parameter, say by::F=first and add F or F<:Function to the where clause.

How is it possible that parallelized code causes fewer allocations?

abraemer December 30, 2024, 3:12pm 19

Have you seen Finch.jl?

2 Likes

How is it possible that parallelized code causes fewer allocations?