GMP 4.4 remaining tasks

[bodrato at mail.dm.unipi.it]

> For GMP 4.4 we need mpn_invert code that uses FFT wraparound.  If I
> understand it correctly, your mulmid code will help in a (large) operand
> size interval before FFT kicks in.

I'd like to try the mod_bnm1 wraparound... It should be slightly faster
than (B^n+1) for big sizes, and it should be a gain wrt plain mul far
earlier than FFT.
Unfortunately this wraparound is not described (yet) in the book... I'll
need much more attention :-)

-- 
http://bodrato.it/papers/