Nested parallel loop

General Usage
1

Is there an easy way, to make the example below work?

using Distributed

addprocs(3)

@sync @distributed (+) for x in collect(1:10), y in collect(1:10)
    sum(rand(x,y))
end
2

What do you expect from the reducer (+) ?

Doing sum(rand(…)) for all combinations of x and y could be:

@sync @distributed for t in [ (x,y) for x in collect(1:10),y in collect(1:10)]
    sum(rand(t[1],t[2]))
end
1 Like
3

Thank you. this works.
Well, this was only a toy example. But I often have a reducer (maybe vcat or similar)

4

But beware, the comprehension [ (x,y) for x in collect(1:10),y in collect(1:10)] is not distributed, or in other words the code is like:

arrayOfTuples = [ (x,y) for x in collect(1:10),y in collect(1:10)];
@sync @distributed for t in arrayOfTuples
    sum(rand(t[1],t[2]))
end
5

that is perfectly fine. in my case ‘the workload’ in in place of sum(rand(…))