Re: [eigen] A complex FFT for Eigen

[Matthias Pospiech <matthias.pospiech@xxxxxx>]

Benoit Jacob schrieb:
> Hi,
>
> Thank you for your interest and your code, it's at least interesting to read.
>
> Eigen is a linear algebra library, so FFT is beyond its basic scope,
> but we do have a couple of modules doing non-linear-algebra stuff
> (Eigen/Regression and Eigen/Geometry) so FFT is not ruled out at that
> level.
>
> I lack knowledge on FFT actually to make a decision whether or not we
> want it inside Eigen:
> - is it usual to restrict it to complex numbers and power-of-two fixed
> sizes ? Wouldn't it be desirable to allow reals, to allow any
> non-pow-of-2 size, to allow dynamic-size? If any is "yes" then we need
> it before inclusion in Eigen
>   
Standard FFT can only use powers of 2, others are usual normal DFT 
routine which are much less optimized.
Fourier Transform is by definition using complex numbers. Very often 
people are using FFTs with matrixes. However usual FFT routines require 
plain C Arrays. That would require to copy the data from Eigen to 
C-Array, do Fourier Transform, copy back to Eigen matrix and such. Thats 
my understanding why it could make sence to have an optimzed FFT routine 
which can directly use eigen matrices.

Matthias

---