For DIF, the phases are of course reversed in the way the step through the binary digits of f(m) and GN(k,f) |
One still has d phases labeled by p |
At first recursion p=0 one is manipulating f(m) with highest binary index md-1 and forming GN(k,f) with binary index k0 |
The first DIF computations are done at end of recursion and this is "natural" m=0 and kd-1 index for GN(k,f) |
DIF has an identical parallel performance to DIT although some of the steps differ in detail |