Bug in map?

[alfaromartino]

I think there's a bug in ThreadsX.map. I initialized Julia with 16 threads.

MWE

using ThreadsX, BenchmarkTools

list1 = rand(50_000_000) .* 100
list2 = rand(50_000_000) .* 100

#OPERATION
foo(x,y) = 100 + x * y + log(x) * 2 / y

with results

@btime ThreadsX.map(foo, list1, list2); #302.361 ms (2771 allocations: 2.39 GiB)

@btime map(foo, list1, list2); #  213.384 ms (4 allocations: 381.47 MiB)

If I run @code_warntype ThreadsX.map(foo, list1, list2);, it shows that the function is type unstable.

REMARK: This is NOT a problem that occurs with ThreadsX.map! or with vmapntt! from LoopVectorization

using LoopVectorization

output = similar(list1)
@btime ThreadsX.map!(foo, output, list1, list2) # 32.265 ms (916 allocations: 88.42 KiB)
@btime vmapntt!(foo, output, list1, list2) # 26.963 ms (0 allocations: 0 bytes)

EDIT: I realized that this seems to be similar to #196 by @bkamins

[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 your foo.