for loop and data in GPU memory

If the for loop is parallelizable, does the Parallel Computing Toolbox execute for loop in a parallel fashion when it is acting on data stored in GPU memory?

[Jill Reese]

It would be a bit easier to answer your question if you provided some sample code.

Thanks, Jill

[Milos]

Suppose that f is a time-consuming function to compute, and that you want to compute its value on each element of array A and place the corresponding results in array B

A = gpuArray(X);

for i = 1:length(A)
    B(i) = f(A(i));
end

Because the loop iteration can occur in parallel, this evaluation could complete much faster if for loop run in parallel on the GPU - in similar fashion like parfor loop is executed. My question is does this happen automatically or for loop is executed in the ordinary non-parallel fashion?

Answer 1

[Jill Reese]

The for loop is not performed in parallel when you write code like this:

A = gpuArray(X);
for i = 1:length(A)
    B(i) = f(A(i));
end

However, if your function f contains only elementwise operations, then you could achieve more parallelism by using arrayfun like so:

B = arrayfun(@f, A);