* Use while loops instead of do while, since loop_next checks for zero before
decrementing, not after.
Regular FFT seems to work now, though there seem to be quite some rounding errors.
Testing a larger FFT is hard, since decoding the results is still manual.