Re: [eigen] A complex FFT for Eigen

[Tim Molteno <tim@xxxxxxxxxxxxxxxxxxx>]

Benjamin Schindler wrote:
> Benoit Jacob wrote:
> > <snip>
> >
> > Let's give us a bit more time to determine whether we want FFT inside
> > Eigen -- don't want to sound un-enthusiastic, it's rather that I'm
> > currently busy and I don't want to make the wrong decision.
> >
> > Anybody reading this, opinions?
> >   
> May be I'm completely wrong on this. But my understanding is that most 
> usually, fft/dft is done on dynamic size data, or at least on fixed 
> size data with N not too small. What I would look into whether it 
> would be easy to use fftw with eigen. From what I gather fftw is 
> really one of the fastest FFT library out there. Even if eigen could 
> get close to the performance of fftw the question still remains 
> whether it's worth it. May be an optional module that integrates Eigen 
> into fftw could be coded quickly and would reduce code duplication. 
> fftw is gpl, so there are no issues there
Hmm, depends on the application. Most FFT's are used for digital 
filtering or numerical problems where the data size is known at compile 
time.

I looked into fftw and it is a great piece of software, excruciatingly 
complicated and very good at what it does. It is 3.5 MB of source code! 
No chance of incorporating it as a plain header file though as it is 
written in C. The code I submitted was 130 lines of C++, templated and 
all in a header file. I think that FFTW and CFFT serve quite different 
purposes.

Cheers

Tim

---