[Insight-developers] VNL FFT

[Kent Williams]

I discovered the other day that I'd coded a bug into my regression test 
for the FFT classes -- it was getting 'false positive' results.  The 
corrected test code exposed that the VNL-based FFT works for the 
square/cubic test case -- images of dimension (4,4,4) but fails for 
images of dimension (3,5,4).

Further, I discovered that VNL's FFT routines for non square/cubic 
images return substantially different answers than FFTW.

Which leads me to believe that VNL's fft isn't probably a reliable 
default implementation to include as part of ITK.

So what I'm wondering is this: does anyone have knowledge of a free FFT 
implementation that is licensing compatible with ITK?
The second question is has anyone who understands FFT implementation 
better than I looked at what VNL is doing?