On Tue, 2006-01-03 at 08:30 -0600, David Schanen wrote:
> On 1/2/06, Stefan Teleman <[EMAIL PROTECTED]> wrote:
> > > Things have probably changed since then as there were major compiler
> > > releases since then. Maybe it would be a good time to rerun it with
> > > Sun Studio 11 (-fast) and equivalent on gcc 4.0 (-O3 -ffast-math
> > > -funroll-loops -mcpu=ultrasparc3 -m64 -mvis -ftree-vectorize).
> > > And after all, they are still just benchmarks. Results could be the
> > > other way around with different benchmarks as seen in your case.
> >
> > -fast is not the fastest option on either SPARC or X86. he who uses
> > -fast shall be disappointed. :-) it actually generates significantly
> > _slower_ code than other options (which require more work because they
> > need to be written out).
> I would be very careful using any optimization that sets the -fns on
> Sun Studio, as I've seen significant differences in results when
> enabling this.  For that matter, -ffast-math has a similar effect with
> gcc.
> I question how much any of this would affect KDE though, since
> enhancements to both Sun Studio and GCC optimization have had to do
> additions for the new incarnations of simd instructions and should
> really only affect floating point performance, so far as I am aware. 
> For something like KDE I would imagine that the memory footprint would
> be more important than anything else when it comes to performance.  Am
> I missing something?

No. You are absolutely right.

That was exactly my point when I was objecting the statement that "Sun
Studio" can imporve KDE performance. Overall KDE performance depends on
so many other things and compiler is a last thing to look at in that
list.

_______________________________________________
opensolaris-discuss mailing list
[email protected]

Reply via email to