Roland Schilling wrote:
> and assume it to be compiled for double-precision reals.
This is not standard-conforming. That is, -r8 switch, which would
promote both REAL and DOUBLE PRECISION to 8 bytes (on your platform),
is *not* standard conforming.
That said,
r4 = huge(1.)
if default real is 64-bits and r4 is 32-bits, is legal, but I suspect it
is an overflow, and then IEEE folks can say what it ought to produce
and what comparison operators produce under IEEE conformance. I don't
think the Fortran standard says anything specific about how conversions
between real kinds are done.
Best,
Aleks
|