Doing an FFT with integer math means doing it in a ring -- and thus dealing with modular reduction. Using complex floating-point math the "modular reduction" c

Doing an FFT with integer math means doing it in a ring -- and thus dealing with modular reduction. Using complex floating-point math the "modular reduction" comes for free thanks to w^N=1. I would have to write some code to confirm which is faster today -- especially if you choose rings which make the modular reduction easier -- but CPUs have typically optimized FP math more for throughout than integer math, since mostly-integer code has more unpredictable branches and less ILP than FP-heavy code.