FLoops ThreadedEx scheduling

General Usage
3

The ... in this case includes a loop that performs simple operations corresponding to math factorizations and that can go up to i iterations, so for every iteration of the outer loop, the loop of the factorization has increased work.

In literal code its:

factor = 2
done = i
while (factor < done)
    if i % factor == 0
        sum += factor + div(i, factor)
        done = div(i, factor) 
        if done == factor
            sum -= factor
        end
    end
    factor += 1
end

I ran few more tests to check what I could relate to your answer and I noticed that thread 1 would always have the higher number of tasks. Based on what you said, could I say that the first iterations aren’t time consuming enough so the Julia scheduler re-uses the master thread? Does the Julia scheduler have higher priority in distributing the tasks than the implementation of the ThreadedEx?

These are some of the distribution of tasks that I got:

  • 50k iterations and 4 threads:

    • 18751
    • 12500
    • 12500
    • 6250

  • 350k iterations and 4 threads:

    • 131251
    • 87500
    • 87500
    • 43750