This module drops to C++ to launch threaded FFTs over a block of numpy ndarrays.
In particular, the transforms are defined over an ND grid ranging in each
dimension from [-N,N-1]. Of course this support can be toggled to regular
FFT support.
From the notes..
This module provides N-D FFTs for functions taken on the interval
n = [-N/2, ..., N/2-1] in all transformed directions. This is accomplished
quickly by making a change of variables in the DFT expression, leading to
multiplication of exp(+/-jPIk) * DFT{exp(+/-jPIn) * [n]}. Take notice that
BOTH your input and output arrays will be arranged on the negative-to-positive
interval. To take regular FFTs, shifting can be turned off.