git clone http://leto.net/code/Math-GMP-SWIG.git
The first thing I have implemented is nextprime, which returns the next prime larger than a given $x. This breaks the dependence of Crypt::Primes on Math::Pari's nextprime function. The reason that many people have voiced concerns over Perl continuing to use Math::Pari is that Math::Pari is extremely sensitive to new versions of Perl. Basically, a new version of Math::Pari must be made for almost every new version of Perl because it depends on XS internals that occasionally change. Math::Pari is written in XS, so it is a challenge to maintain.
A SWIG interface to Pari could be possible but I am more familiar with GMP and so I decided in that direction. I wrote some code in 2001 that calculated Pi with Ramunjan's famous formulas using GMP. One algorithm gets 8 digits of precision per iteration and the other 14 digits per iteration.
I am preparing a TPF Grant proposal for this, so if any one has comments or suggestions, please comment or email me personally.