i686 regression tests

[Donn Terry]

The recent patch I submitted for rbug.c plays in much the same
pool; you might wish to look at it.  I don't believe it solves
this problem directly, but may prove a leg up on it.

Donn


Sylvain Pion wrote:
> 
> On Thu, Jun 03, 1999 at 12:04:11AM -0600, Jeffrey A Law wrote:
> > 980619-1.c is failing with gcc-2.95, but not egcs-1.1.2 due to the extended
> > precision issues on the x86.  gcc-2.95 is holding a value in a reg rather
> > than flushing it to the stack.
> >
> > As a result it takes many more iterations before the program converges on a
> > final answer.
> >
> > I'm going to remove 980619-1.c from the failure list for x86 targets.
> 
> I would be happy if gcc had some kind of "-fstrict-ieee" option that takes
> care of working around the cpu "features" to conform to ieee 754 at a minimal
> cost.  I know we already have -ffloat-store, but it's not perfect
> (intermediate computations are still kept in registers, and sometimes it's not
> useful to flush to the stack).
> Perhaps deciding to flush to memory after each insn leading to a potential
> problem (+,-,*,/,sqrt) would be enough and simple ?
> 
> Maybe you should run 980619-1.c with -ffloat-store ?
> 
> --
> Sylvain

-- 

===================================================
Donn Terry                  mailto:donn@interix.com
Softway Systems, Inc.        http://www.interix.com
2850 McClelland Dr, Ste. 1800   Ft.Collins CO 80525
Tel: +1-970-204-9900           Fax: +1-970-204-9951
===================================================