Performance of map

[bkamins]

@tkf what is the reason of the following performance comparison? (this is a fresh Julia session on Win11 with 8 threads):

julia> Threads.nthreads()
8

julia> x = rand(10^8) .- 0.5;

julia> @time map(abs, x);
  0.251657 seconds (99.43 k allocations: 768.457 MiB, 18.93% gc time, 14.49% compilation time)

julia> @time map(abs, x);
  0.239117 seconds (3 allocations: 762.940 MiB, 22.72% gc time)

julia> using ThreadsX

julia> @time ThreadsX.map(abs, x);
  1.842571 seconds (2.53 M allocations: 4.024 GiB, 10.90% gc time, 33.76% compilation time)

julia> @time ThreadsX.map(abs, x);
  1.176356 seconds (1.51 k allocations: 3.888 GiB, 14.73% gc time)

If I use 1, 2, or 4 threads the situation is similar.

Thank you!

[MasonProtter]

Hoping to mitigate this somewhat with JuliaFolds/Transducers.jl#553. However, Transducers.jl (and by extension, ThreadsX.jl), is at it's worst when dealing with very fast functions like abs.

The easiest fix for you to get better performance would be to use ThreadsX.map!(abs, similar(x), x).

[jamblejoe]

I have the same problem on Apple M1:

julia> using ThreadsX

julia> Threads.nthreads()
4

julia> x = rand(10^8) .- 0.5;

julia> @time map(abs, x);
  0.365691 seconds (45.60 k allocations: 765.970 MiB, 16.88% gc time, 6.83% compilation time)

julia> @time map(abs, x);
  0.471808 seconds (3 allocations: 762.940 MiB, 14.60% gc time)

julia> @time ThreadsX.map(abs, x);
  1.776120 seconds (3.21 M allocations: 1.576 GiB, 12.86% gc time, 47.61% compilation time)

julia> @time ThreadsX.map(abs, x);
  1.761694 seconds (120 allocations: 2.340 GiB)

julia> versioninfo()
Julia Version 1.10.3
Commit 0b4590a5507 (2024-04-30 10:59 UTC)
Build Info:
  Official https://julialang.org/ release
Platform Info:
  OS: macOS (arm64-apple-darwin22.4.0)
  CPU: 8 × Apple M1
  WORD_SIZE: 64
  LIBM: libopenlibm
  LLVM: libLLVM-15.0.7 (ORCJIT, apple-m1)
Threads: 4 default, 0 interactive, 2 GC (on 4 virtual cores)
Environment:
  JULIA_EDITOR = code
  JULIA_NUM_THREADS = 4