From: Sukanta on
How do I take advantage of parfor for fft2 and ifft2 computations? On my Cray CX1-iws (each node has 8 cores, 24 GB RAM), the parfor-based calculation is always ~8-10 times slower (I tested for N = 128, 256, and 512). Is there any way I can speed up the computation?

N = 512; u = randn(N,N,N); uf = zeros(N,N,N);

tic;
parfor k = 1:N
uf(:,:,k) = fftn(u(:,:,k));
end
toc;

tic;
uf = fft2(u);
toc;