fast inversion

[paul zimmermann]

       Hi,

I did update my notes and code about fast inversion (which were used as basis
for Algorithm 3.5 ApproximateReciprocal in "Modern Computer Arithmetic"):

http://www.loria.fr/~zimmerma/papers/invert.{c,pdf}

The main change is that now the result is uniquely defined (instead of 2 ulps
margin in the previous version), thus it coincides with the mpn_invert
function from GMP.

Comments are welcome.
Paul