Wraparound multiplicaton vs mullo

[Torbjorn Granlund]

I am following the discussion with great interest!

If we indeed can get a Schönhage-Strassen like algorithm that needs
dyadic products mod 2^t-1, we should get a nice speedup for all FFT
operand sizes except those that use FFT for the dyadic products.

(Today FFT is used for cofficient sizes of 300 to 700 limbs.  This would
probably increase sharply to around the plain FFT threshold, i.e. 1500
to 9000.)

-- 
Torbjörn