Am Donnerstag, 4. November 2004 01:18 schrieb Stefan Teleman:
> Feel free to time test KDE built with Forte vs KDE built with GCC, and
> post the results. Saying "30%, Come on" is not exactly computer
> science, or a counter argument.
>
> Or, get the Forte flags from my KDE patches, write a small program
> which calculates PI 10000 times, compile i with those flags (Forte 8
> or 9), and run it.

... which doesn't really give a good number ;-)
CPU architectures and compilers give very different time results, if it comes 
to integer vs. float calculation. KDE is a gui and thus you mainly have 
integer calculation - here calculating PI certainly isn't a good benchmark.
And don't ask me for a proposal an how to measure speed on KDE...

Greetings,
eva

> Do the same thing with GCC, with your favorite flags, and run it.
> Then please post the source code of the program, the compiler flags
> you used, and the results.
>
> Hint: it's been done before.
>
> Instruction prefetch _is_ supported on UltraSPARC II, not just III or
> better. And cache instruction prefetch is not the only way to achieve
> very aggressive optimizations with Forte on UltraSPARC. It's just one
> of the ways of optimizing, and there are specific ways of setting the
> optimization of prefetch-to-load and prefetch-to-store factors, and
> also specifiying cache properties. There's tons of other things one
> can do with registers, enabling the visual instruction set, pointer
> aliasing,
>
> I am not aware of anyone at GCC claiming they achieve the same
> optimization levels on UltraSPARC as the Sun compiler.
>
> --Stefan
>
> ------
>
> On Wednesday 03 November 2004 06:37, Christopher Layne wrote:
> > That's great. But how about we just back this up with actual timed
> > testing of KDE components and see for ourselves?
> >
> > -cl
> >
> > >   The 900 Mhz UltraSparc III-CU and later model processors has a
> > >   Branch prediction logic that enables the processor to prefetch
> > >   the right instructions after a branch and prevent Cache-miss
> > > main memory instruction fetches.
> > >   This circutry gets enabled with the "xtarget=ultra3"  or
> > > "xtarget=ultra3i" ( for the UltraSparc III-i processor )
> > > compiler directives. Then the performance differance is certainly
> > >  30%. Because you avoid a large number of the execution pipeline
> > > stall's that you have otherwise.
> > >
> > >   //Lars
> > >
> > >
> > >
> > >
> > >
> > >
> > >
> > > ___________________________________________________
> > > This message is from the kde-solaris mailing list.
> > > Account management:
> > > https://mail.kde.org/mailman/listinfo/kde-solaris. Archives:
> > > http://lists.kde.org/.
> > > More info: http://www.kde.org/faq.html.

-- 
-----------------------------------------
basysKom
Ingenieurbüro Brucherseifer und Ackermann GbR
IT System Kompetenz
-----------------------------------------
Dipl.-Ing. Eva Brucherseifer
Karlstr. 45
64283 Darmstadt
Fon: 06151-3969961   Fax: 06151-3969736
Mobil: 0170-5533642
eva.brucherseifer@basyskom.de
www.basyskom.de
-----------------------------------------
___________________________________________________
This message is from the kde-solaris mailing list.
Account management:  https://mail.kde.org/mailman/listinfo/kde-solaris.
Archives: http://lists.kde.org/.
More info: http://www.kde.org/faq.html.