fix execute/intrinsic_nearest.f90

[Paul Brook]

On Friday 30 July 2004 19:28, Richard Henderson wrote:
> This allows the test to pass on i686 and (if I force -mieee) on alpha.
>
> The wording from the F2003 standard for NEAREST matches the wording
> for C99 nextafter, so I assume from that denormals are fair game.
>
> Ok?

This breaks on systems that round denormals to zero (mips), hence the current 
use of TINY. Subnormal numbers are at best a grey area of the fortran 
standard.

Maybe make your change and move the test into the gfortran.dg testsuilte so it 
can be xfailed on targets where it doesn't work.

Paul