Code optimization with GPU

Hi everyone, I have the following line of codes which I would like to parallelize on the GPU :

U = zeros(q, M, q, 'gpuArray');
for j=1:q
    U(j,:,j) = my_line_vector;
end

Is there any way to suppress the loop?

Thes are the lines of code that run the slowest in my program.

Thank you in advance

Michel

Answer 1

[Joss Knight]

I = reshape(gpuArray.eye(q),q,1,q);
U = I.*my_line_vector;

[POHL Michel]

Hi, no it does not work, in your case U does not have any zeros,

whereas in my case U has a lot of zeros.

[Joss Knight]

Ah yes, sorry about that. I've edited the original, see if that works.

[POHL Michel]

Thank you, it works!

I did not know that the element wise product worked with matrices with different size.

[Joss Knight]

https://uk.mathworks.com/help/matlab/matlab_prog/compatible-array-sizes-for-basic-operations.html