Replacement mpn/generic/mul.c

[Torbjorn Granlund]

In the end, I decided for a slightly more complex mul.c, adding a few
more threshold parameters for unbalanced toom primitives.  The result is
a bit more complex that desirable.  The speedup of the new code is
however considerable.

I committed the code to the public GMP repository.

With this checking, more of the new multiplication code developed by
Marco Bodrato, Niels Möller and myself is used by mainstream GMP
multiplication.

-- 
Torbjörn