How can I make fft2 parallel?

Hi guys

I got a code consisting of 3 loops and packed with fft2 command. Since it is not possible to make the code itself parallel, I wonder if it is possible to make the fft2 parallel instead and see how much it can speed up the code.

Any help in this regard is appreciated

Thanks

Mohammad

Answer 1

[michio]

Are you considering using gpu? fft2 is one of the built-in functions that support gpuArray. It requires Parallel Computing Toolbox though.

http://www.mathworks.com/help/distcomp/run-built-in-functions-on-a-gpu.html http://www.mathworks.com/matlabcentral/newsreader/view_thread/309375

[[e-mail]]

Thanks Michio

I wonder whether it may take more time than the traditional fft2 to send a huge matrix regularly to gpu and apply fft2 on it. Is that a right thing to consider or using gpu is always faster?

[Walter Roberson]

gpu is not always faster. You need to do tests. What would a sample array size be for you?

[Walter Roberson]

One thing to watch out for is that GPUs often have limited memory. It is common for the largest practical input array size to be about 1/8th of the amount of memory on the GPU.

On the test I just did on my Geforce GTX 650M (not a high end device), gputimeit() of fft2 of a random 1000 x 1000 array took aproximately twice as long as on the CPU. I could not go much faster than 1000 x 1000 before running out of memory on the gpu.