Julia High Performance - Second Edition

Parallel for loops

Julia includes an inbuilt parallel for loop that can automatically distribute the computation within the loop across all nodes in a cluster. This can sometimes allow code to speed up across machines with little programmer intervention.

In the following code, we generate a million random numbers and add them. The first function computes each step serially:

function serial_add()     s=0.0     for i = 1:1000000          s=s+randn()     end     return s end 

Each step in this loop can be computed independently, and thus should be easy to parallelize. The second function here attempts to distribute the steps across the cluster: 

function parallel_add()     return @distributed (+) for i=1:1000000        randn()     end end

We can see that the parallel function ...