x86 regression: optimizer bug in 1.1 and dev

[Dave Love]

>>>>> "UD" == Ulrich Drepper <drepper@cygnus.com> writes:

 UD> If you are doing numeric computations on ix86 (not a wise choice)
 UD> you have to be aware of the risks and the gcc features.

Is there something GCC specific involved?  (The x86 SunPro compiler
has a flag like -ffloat-store, for instance.)  I thought it was just a
function of the system-dependent startup code setting extended
precision (and I wonder if it's worth the grief it causes).

A question I've never seen answered is whether extended precision is
necessary for the GNU maths library, i.e. whether resetting it as
cautiously advocated in the g77 manual is asking for trouble.
Comments in the glibc fpu_control.h (?) on the subject are confusing.