mpz_gcd

[Kevin Ryde]

Paul.Zimmermann at loria.fr (Paul Zimmermann) writes:
>
> 1)   find two 1-limb numbers u, v such that B^2 divides u*a + v*b

Bear in mind that this takes a fair bit of time compared to smallish
sized multiplies.  About 1100 cycles on athlon for instance.  You can
measure with

	./speed -s1 -c find_a

The -s size is not used, but you have to give it to make the general
framework happy.

I guess this has the effect of being a biggish "linear" term in the
time formula.