Benchmarks?

[kozross]

I'd like to see some. I'd be happy to run them myself, but as I have no idea how FFT performance is evaluated, I'd love some help finding a suitable test.

[adis300]

I would just create some pseudo signal that of size 2^N and a pseudo signal of size that is not a power of 2. And run some comparison test on local machine thousands of times and compare the results. It would be great if you could help. That's a very generous offer. :)

[kozross]

@adis300 I can do that, but I know pretty much nothing about DSP, which means I don't even know how to generate a pseudo-signal! Additionally, if there's some existing benchmark or test for this kinda thing, it'd probably be better and more convincing. Can you suggest anything?

[adis300]

In this project, the example shows how to generate a pseudo sine wave and applied FFT. I think this gentlemen provided some benchmark in his README file (https://bitbucket.org/jpommier/pffft). You can take a look at the FFTPACK vs FFTW performance comparison. The work I did is just repackaging Dr. Monty's work from 1997 that translates 1d transform from fortran to C, making it very easy to use for all developers.