[Patch] libstdc++/21193 (float, double, long double)

[Paolo Carlini]

Gabriel Dos Reis wrote:

>It would be nice if you had access to Sparc boxes or similar beasts,
>where a long double is 96 bits (no padding bits), I think (please have
>someone double check).   I'm deeply in something else; sorry for being
>quite since last week-end and not helping more. 
>  
>
No problem. Really I'm considering every possible detail (also studied
both Sparc and IBM long double formats, which, interestingly, still use
15 bits for the exponent). Thanks,

Paolo.