[gfortran] Fix NEAREST constant folder

[Steve Kargl]

On Tue, Apr 12, 2005 at 01:16:19PM -0700, Richard E Maine wrote:
> On Apr 12, 2005, at 1:06 PM, Steve Kargl wrote:
> 
> >AFAIK, the standard does not define
> >the meaning of "machine representable number".
> 
> I think the normal English definition applies (as it generally does 
> when there is a normal English meaning and there isn't an overriding 
> technical definition in the standard).  Namely...
> 
> I would say that "machine representable number" means a number that the 
> machine can represent.
> 
> Note that this is definitely not the same thing as a "model number". 
> The standard does define what a model number is, and it distinguishes 
> model numbers from machine representable numbers.  (Which isn't to say 
> that there might not be places where it gets it wrong, but the 
> distinction is made).
> 

Yes, that's the interpretation that Tobi and I want to make.
nearest(0.,1.) should return a subnormal number if that is the 
nearest "machine representable number".  I'm somewhat surprised 
that standard did not state "processor-dependent representable
number" in the sense of a Fortran processor.  Afterall, a
Fortran processor does not need to be tied to a "machine".

-- 
Steve