Parallelizing a for-loop with a matrix

General Usage Performance
1

I have a simple for-loop with a matrix, where I calculate a slice of the matrix via each step of the for-loop. e.g.

for ilat = 1 : nlat, ilon = 1 : nlon

    for ip = 1 : np; esat[ip+1] = t2esat(Ta[ilon,ilat,ip],p[ip+1]) end

    p[end] = ps[ilon,ilat]
    esat[end] = t2esat(Ts[ilon,ilat],p[end])
    swp[ilon,ilat,it] = calcswp(esat,p,ps[ilon,ilat])

end

I wish to parallelize the outer for-loop here. What packages should I use? Should I use DistributedArrays, or is Distributed enough? I’m not sure if I should use multi-threading, but I can use multiple cores. Do I need to specify the number of cores?

2

You might like

https://juliafolds.github.io/data-parallelism/tutorials/quick-introduction/

2 Likes
3

If you post runnable code (with inputs and maybe timings using BenchmarkTools.jl), you’ll likely get much more response.

But in any case, I think you may want to start with multithreading by starting julia with julia --threads 4 and adding Threads.@threads in front of your first for loop. Distributed is for distributed computing - when you have e.g. a cluster with multiple compute nodes.

1 Like