This is the mail archive of the
gcc-patches@gcc.gnu.org
mailing list for the GCC project.
Re: PATCH: PR target/23485: [ia64]: Integer dvide by zero doesn't raise a signal
On Sun, Aug 21, 2005 at 03:16:55PM -0700, H. J. Lu wrote:
> On Sun, Aug 21, 2005 at 11:04:34PM +0200, Gerald Pfeifer wrote:
> > On Sat, 20 Aug 2005, H. J. Lu wrote:
> > >> I am running SPEC CPU 2K with and without my patch. I will report the
> > >> results when I check in my patch.
> > > I got 1% slowdown in SPEC CPU2K INT and no significant change in SPEC
> > > CPU2K FP. I am checking my patch now.
> >
> > Checking the patch, or checking in the patch?
> >
> > I, for one, do quite disagree to accept a 1% slowdown in SPEC CPU2K INT
> > to change a behavior which was fully standards compliants to begin with.
> > This is not a bugfix, but a feature (which, I agree, may come handy when
> > compiling software which does not strictly honor the C language standard.)
> >
> > If it really slows down things, please make this an option, off by
> > default.
> >
>
> I got SPEC CPU 2K INT/FP data backwards. It should be
>
> 1. With devide-by-zero check, -O2 gives %1 slowdown in SPEC CPU 2K FP
> and no significant change in SPEC CPU 2K INT.
> 2. With devide-by-zero check, -O2 -minline-int-divide-max-throughput
> gives 0.2% slowdown and no significant change in SPEC CPU 2K INT.
> 3. -minline-int-divide-max-throughput may be slightly faster for SPEC
> CPU 2K FP and INT. I am running -minline-int-divide-min-latency now.
FWIW, -minline-int-divide-min-latency and
-minline-int-divide-max-throughput give similar numbers for SPEC CPU2K
INT and FP.
H.J.