@distributed (op) for with mutable types

[tkf]

@Marco_Lombardi

Yes, but I think mapreduce is not distributed, is it?

You are right. mapreduce in Base has no parallelism. Perhaps Distributed.jl should add it. (You can also use Transducers.dreduce(op, Map(f), itr))

@Marco_Lombardi

Also, from the Julia documentation I read that pmap (and therefore I guess also a parallel version of mapreduce ) should be used in situation where the function is slow to compute, while @distributed for is more appropriate for quick computation.

This is not generally true. Functional approach can have much better performance characteristics (see, e.g., [RFC/ANN] FLoops.jl: fast generic for loops (foldl for humans™)).

@Marco_Lombardi

I think that a simple change would make it return also the index i of the maximum, on top of the maximum value.

You need two accumulators to interact to get this. So, you can’t just independently annotate each reduction variable.