Bad performance on multivariable functions.

[adamczykm]

Have a look:
a=randn(5000000);

@time(reduce((x,y) -> x,a))
elapsed time: 0.250185444 seconds (60000812 bytes allocated, 51.42% gc time)
0.5564212271634134

@time(reduce(@anon((x,y) -> x),a))
elapsed time: 0.815615445 seconds (120012708 bytes allocated, 35.23% gc time)
0.5564212271634134

[timholy]

I bet it's just that there needs to be a specialized version of reduce. Only map and map! have been implemented. Care to contribute reduce?

But there also seems to be a dispatch issue, I'm not seeing it call the specialized versions in FastAnonymous:

julia> @which map((@anon x->x), a)
map(f::Union(DataType,Function),A::AbstractArray{T,N}) at abstractarray.jl:1303

when instead it should be calling this method.

So if you contribute reduce, you could call it reduce_fa or something. Meanwhile I'll try to figure out the dispatch problem.

[adamczykm]

I'll be happy to contribute as soon as I get some spare time. Lots of things are happening to me right now. I'll check if it's still an issue on January.

[timholy]

This is fixed in julia 0.4 because of JuliaLang/julia#10314. I'll close the issue, but note that on the 0.3 version of this package it would still require a pull request to implement efficient reduce. If you decide to do this, I can create a 0.3 branch; as it is, further development of this package on 0.3 is frozen.