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

[FX Coudert]

> You need to remember that some targets (e.g. ia64) have both real(10)
> and real(16), and only one of them corresponds to "long double".  There
> are quite a number of instances in the front-end patch that make this
> mistake.

Right now, these two real kinds are not working. So, I'd say we go for 
this first, and add refine this later (this patch is already large enough).

Plus, I'm not sure, since it appears that on ia64-unknown-linux-gnu, 
gfortran doesn't support real(16).

As I don't know anything about IA64, how are these two real types called 
in C (long double and ..., I guess)? And what math functions are 
available for each type? And, of course, where can one find such a 
computer to test and debug?

FX