[ANN] ThreadsX.jl: Parallelized Base functions

[carstenbauer]

FYI: I’ve also wrapped a significant part of LIKWID in LIKWID.jl. It has a perf mode but also its own daemon for accessing low-level information like performance counters.

[pitsianis]

Does ThreadsX support thread schedulers like the WorkStealingEx from FoldsThreads? I would like to use it with the ThreadsX.maximum.

[tkf]

You can use Folds.maximum from GitHub - JuliaFolds/Folds.jl: A unified interface for sequential, threaded, and distributed fold for specifying the executor.

Folds.jl and ThreadsX.jl have some overlapping features but the differences are that Folds.jl can take executor while ThreadsX.jl implements non-fold algorithms like quicksort.

[Akatz]

@tkf

Also, it’s not so hard to put it in Strided.jl. For nested parallelism, I think making @threads for nestable with your PR (or the other one) is enough.

Having said that, I think transducer-based implementation is going to be appealing if we have some kind of filtering built into the broadcasting syntax. I keep thinking about this but it’s not obvious to me (in the sense it’s hard to come up with som

@rayegun This might be something to consider for your sparse codegen plans