
- rotated fft, prime factor (gpfa) fft

- meta fxt generator 
  (prog that writes generator code from existing fxt code, cf. metagen)

- radix 32,16,8,4,2 - fft

- weighted (irrational base) ffts a la crandall

- localised ffts (depth-first, cf. fftdif2l.cc)

- fermat number NTTs

- special NTT for P64 = 2^64-2^32+1

- galois field NTT

- selfsorting fft

- multithreading FFTs

- multithreading disk I/O (partly finished)


// ==== postponed:

- faster NTTs
- exact convolution with several NTTs and CRT
- montgomery multiplication for NTTs
  (probably thats _not_ faster than the present code)

- successive (zero padded) FFTs of increasing length
- zero padded fft thats asymptotcally cheaper than full-fft
  (that may be impossible)

- length-9 real-fft, fht

- efficient inplace transpose routine, probably
  interlaced with trig-multiplications

- rectangular transform

- winograd fft
