Re: [eigen] A complex FFT for Eigen

[Tim Molteno <tim@xxxxxxxxxxxxxxxxxxx>]

Matthias Pospiech wrote:
> Tim Molteno schrieb:
> > Hi,
> >
> > Eigen is great. Thanks to everyone who has done so much work on it.
> > I've written a complex fft algorithm for eigen. It uses template 
> > metaprogramming, and is fairly snappy.
> Are you sure that it matches in performance with the FFTW Libary? I 
> just say that because I personally would use eigen for matrixes and 
> FFTW for FFT.
> But maybe you can convince of the improvements of you solution 
> compared to mine. The first that would come to my mind is, that FFTW 
> requires a plain 1D C-Array for a Matrix (of 2 or 3 dimensions).
Thanks for the comment. It will not be as fast as fftw, but might get 
close.

But it has other advantages

1) It is templated (fftw does not play nicely with fixed point for example)
2) It is very simple to use
3) If you're only doing one fft, it might even be faster than fftw since 
fftw requires a two-pass approach
4) It is GPL 3.
5) It is only a few lines of code (130)

Other Disadvantages

1) Requires a power of 2 length
2) Only one dimension at the moment. I'll get 2D transforms incorporated 
into Eigen when I next require them.

Cheers

Tim

---