unbalanced mul/div

[Paul Zimmermann]

       Hi,

I noticed a nice speedup for unbalanced mul/div from GMP 4.3.2 to GMP 5.0.0.
See http://www.loria.fr/~zimmerma/tmuldiv10k.ps. This was tested on a 2.4Ghz
Opteron. The "mul" test multiplies x limbs (where x is in abscissa) by n-x
limbs, where n=10000. The "div" test divides n limbs by x limbs with remainder
(mpn_tdiv_qr).

There are still some irregularities and spikes, but the overall speedup is
in the 25%-33% range for both mul and div.

Congratulations!
Paul