Wraparound multiplicaton vs mullo

[Torbjorn Granlund]

nisse at lysator.liu.se (Niels Möller) writes:

  Nut it *is* mod 2^n - 1, for certain values of n.
  
Which values of n?

  > Do you represent 0 canonically as B^n-1?
  
  No, it's only "seminormalized", no canonical representation of zero.
  I'd expect that B^n - 1 is more likely than zero, since various terms
  are added together and then extra high limbs are wrapped around and
  added in at the low end. 
  
I don't see how zero can be generated ever in the ref code, except
when multiplying zero.  (The mul_2nm1_2 and mul_2nm1_4 functions
are more likely to generate zero, I think.)

The testing code does not seem to think that the two representations of
zero are equal.  :-)

-- 
Torbjörn