I am trying to use Apple's AltiVec FFT routine fft2d_zrip in a scheme
for doing 2D sinc interpolation by zero-padding the 2D DFT of an image.
I am able to do forward and reverse 2D FFT's and recover the original
image, but when I zero pad I get clearly wrong results. The FFT routine
in question makes use of what Apple calls a "split complex vector"
format, which is a reordering of array elements to improve performance.
I know that such reordering is fairly common in high performance FFT's,
but Apple's scheme is not very well doented.