I am trying to reproduce the results of the MKL FFTW interface in this report:
Does anybody know where I can get the source code used in that report?
So far, I have been running the source code in this webpage:
but the MKL FFTW interface shows really poor performance. I wonder if there is something I am missing.
The MKL performance charts show the performance of other FFT problems than that measured by the benchmark code from Heiko's page.
In particular, the MKL charts show performance of a single transform, while the benchmark runs FFTs of 2 or 4 interwoven data sets. Genuine FFTW is good at such layouts.
A good starting place for benchmarking FFTs is http://www.fftw.org/benchfft/.