[gfortran] Fix NEAREST constant folder

[Paul Brook]

On Tuesday 12 April 2005 23:00, Steve Kargl wrote:
> On Tue, Apr 12, 2005 at 02:39:19PM -0700, Richard Henderson wrote:
> > On Tue, Apr 12, 2005 at 01:56:51PM -0700, Steve Kargl wrote:
> > > I suspect we will need to rip MPFR out of gfortran, and
> > > use gcc/real.c.
> >
> > I'm not thrilled about this, since there had been a vague push
> > to move real.c to use MPFR as well.  What about creating some
> > wrapper functions that allow us to fake subnormal numbers?
> >
> > We'd do this by allowing the exponent to go down to emin - p,
> > and force the least sigificant e - emin bits of the fraction
> > to be zero.  I.e. a sort of "normalized sub-normal".o
>
> I sent a message off to the MPFR list to ask about the
> future of subnormal numbers.  Still waiting for a reply.
> MPFR ensures a correctly rounded significand.  The
> exponent in principle has an unlimited range, so wrapping
> MPFR should be possible.

Wouldn't that cause double rounding?
MPFR would round to the full precision, then we'd round again to simulate the 
reduced precision from the denormal.

Paul