Howdy all,
I have some code that makes heavy use of the power of 2 FFT's in MKL v5.2. I'm going through trying to improve the code effeciency. Would appreciate any insight the community can give regarding the potential for speeding things up from:
1) updating to use the latest version with the DFT interface. I'm only doing 2^N sized fft's (usually 1k - 8k size). From the reading it looks like most of the improvements are related toothersizes of fft. Has anybody upgraded code from v5.2 to v8? can you comment on any speed improvements that you saw?
2) I see that the DFT interface supports using un-ordered FFT's. since a good chunk of my code is doing correlations, i don't need the ordering. about how much time can i expect to save on complex FFT's of 1k to 8k size? Qualitatively, is it closer to 2% or 20%?
Actually updating, testing, and timing the code will be a non-trivial effort, so i would like to have at least a vague feeling for the potential benefit. Any experience that you can share will be most helpful!
- Rob
