FFT improvements

[Torbjorn Granlund]

"Vincent Diepeveen" <diep at xs4all.nl> writes:

  With Joel Veness i'm working on some FFT code. We happen to see that GMP is 
  using the same method nowadays in version 4.2.
  When we've got it to work low level optimized here we'll do a backport of 
  that to GMP, as that should speed it up quite some in 64 bits.
  
Once you have correctly working code, some speed comparisons would be
interesting.

  Especially: how much wanted is a parallel implementation of GMP?
  
We are not interested in that at this point.  We don't think such
parallellisation belong to a library like GMP.  Users need to
parallellize at a higher level.

-- 
Torbjörn