Kahan's Floating Point Test

[Geoff Keating]

> From: dewar@gnat.com
> Cc: gcc@gcc.gnu.org
> Date: Thu, 31 May 2001 20:25:40 -0400 (EDT)
> 
> <<Even those (old) PowerPC architectures where all computations are done
> in double precision should have single-precision IEEE arithmetic fully
> supported in GCC; GCC will add appropriate rounding instructions when
> necessary.
> >>
> 
> That's not quite good enough, due to double rounding phenomena

It is OK between double precision and single precision; double
rounding can't happen there because there are enough bits in a
'double'.  It's not OK between the Intel 80-bit extended precision and
double precision.

-- 
- Geoffrey Keating <geoffk@geoffk.org>