[gfortran,ping] support for large kinds in front-end and library

[FX Coudert]

> __float80 and (__float128 or long double).
> 
>>And what math functions are available for each type?
> 
> I'm not sure.

Well, as long as there are not math functions for each type, I don't see 
the point in getting it right. In that precise case, real(16) will work, 
and reak(10) will not. We can then do better in another patch.

> There are a couple of folk interested in ia64-hpux -- Steve Ellcey
> <sje@cup.hp.com> for one, as well as various members of CodeSourcery.
> I imagine that you could convince some of them to do testing.

OK, noted. Although it's pretty annoying to develop a patch without 
being able to test it directly.

FX