afft?

[OverLordGoldDragon]

Can modulus be reasonably integrated into FFTs?

It'd notably accelerate and cut memory use for CWT, STFT, and others where we take abs after fft (many uses), by sparing allocations and assignments. Idea being to take modulus before the final assignment to each FFT bin. Less performant but still improving would be a simple abs on top of regular output, but done internally so sparing allocation.

Would FFTW be interested in implementing this? Is it already implemented somewhere? I'd appreciate pointers on this topic.