Re: [Starpu-devel] Performances regression

[Xavier Lacoste <xavier.lacoste@inria.fr>]

Hello,

As we can see on the gantt diagrams, both result are not good with a lot 
of "blocked" time...
So maybe the performance regression i noticed is not so relevant...
I must be misusing StarPU...

Regards,

XL.

Le 01/02/2012 13:48, Xavier Lacoste a écrit :
> I was using heft
>
> Le 01/02/2012 13:30, Nicolas Collin a écrit :
> > Hello,
> >
> > Thank you for your feedback.
> > Which scheduler are you using during these tests ?
> >
> > Greetings,
> >   -- Nicolas
> >
> > On 02/01/2012 01:14 PM, Xavier Lacoste wrote:
> > > Hello,
> > >
> > > I noticed a performance regression with StarPU between revision 5000 
> > > and 5415 :
> > >
> > > I factorize my Matrix using starpu :
> > > First       factorization : StarPU r5000 45 seconds r5415 82 seconds.
> > > Second  factorization : StarPU r5000 25 seconds r5415 58 seconds.
> > >
> > > I'm using FXT and also StarPU_PROFILING_ENABLE which seems to slow 
> > > down a bit StarPU execution.
> > >
> > > Paje traces are available on Plafrim 
> > > ~lacoste/ricar/trunk/pasje_5000.trace 
> > > ~lacoste/ricar/trunk/pasje_5415.trace
> > >
> > > Have a nice day,
> > >
> > > XL.
> > >
> > >
> > >
> > >
> > > _______________________________________________
> > > Starpu-devel mailing list
> > > Starpu-devel@lists.gforge.inria.fr
> > > http://lists.gforge.inria.fr/cgi-bin/mailman/listinfo/starpu-devel
> >
> > _______________________________________________
> > Starpu-devel mailing list
> > Starpu-devel@lists.gforge.inria.fr
> > http://lists.gforge.inria.fr/cgi-bin/mailman/listinfo/starpu-devel
>
>
> _______________________________________________
> Starpu-devel mailing list
> Starpu-devel@lists.gforge.inria.fr
> http://lists.gforge.inria.fr/cgi-bin/mailman/listinfo/starpu-devel