This is the mail archive of the fortran@gcc.gnu.org mailing list for the GNU Fortran project.

Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]
Other format: [Raw text]

Re: Polyhedron 2005 Benchmarks

On Fri, Nov 18, 2005 at 07:09:51PM +0100, Paul Thomas wrote:
> Steve Kargl wrote:
> 
> >On Fri, Nov 18, 2005 at 06:44:05PM +0100, Paul Thomas wrote:
> >
> >>A word of caution,...
> >>
> >>>52.0s without and 36.7 seconds with -ftree-loop-linear.
> >>>
> >>... it also gives incorrect results.  I will figure out how to reduce it 
> >>and submit a PR.
> >>
> >
> >Is incorrect in that the two different gfortran runs give different
> >answers?  Are you using -O3 and --fast-math? 
> >
> Without it, ifort/gfortran/DF6 all give the same results. With 
> -ftree-loop-linear, it throws a NaN towards the end of the run and stops.
> 

OK. Smells like an off-by-one indexing problem caused by fortran
arrays starting by default at an index of 1 and C arrays starting
at 0.

-- 
Steve
Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]