I have cuda code in which I implemented several C2C 2D FFT. They all use the same plan, but for some reason, the times at 2D FFT are large and seem to be very different. The same FFT data size appears to be between 0.4 and 1.8 s.
This is for FFT 1920x1080. Do these times know the reasonable?
Anyway - I had a good experience with fast fast FFTs with CUDA 1-D. is it the same as taking 1D FFT row by row and then again through the columns of the matrix to get the same results as this 2D FFT? I have FFT experience that happens in a few hundreths of a second in large datasets for 1D FFTs, so I was hoping to possibly fix some of these results.
thank
Derek source
share