I am writing one program which applies 13x13 filter on image in CPU. I want to optimize this code with using GPU. For that I have one way to do this, use CUFFT libraries. In this library there is some functions for fourier transform, like cufftExecR2C, cufftExecC2C and cufftExecC2R. But my image data and filter kernel is in real format.

So how can I apply real to real operation in FFT calculation using CUFFT library?


Now I have got the solution. If any one has this type of problem feel free to ask me.


I’m new in CUDA and I have the same problem.

Do you use complex-to-real with imaginary part set to 0?

This is the only I can think, I’m a newbie in this. :(

I would like to know the solution that you found, please.

Thank you!! :thumbup: