Re: Scrolling performance

[Roo <oscosc gmail com>]

On Thu, 06 Jul 2006 22:59:35 +0200, Clemens Eisserer wrote:

> Just a side note: GTK-2.9.4 built with "-O2 -g -fPIC" is a lot slower
> than the stuff shipped with OpenSuSE (I guess 2.8.6+). maybe the
> debugging symbols cause cache misses ... I don't know but its about
> 25-75% slower.

What options do you suggest building the gtk2 srpm with. I think I know
how to alter those and rebuild the rpm... but I'm no expert with it.

>> Even with gtk-perf I'm still stuck being unable to make a comparison,
>> since they took away that option.
> Sorry I don't understand what you mean.

I could have done a run with gtk-perf with GTK using Cairo, and then
without it using Cairo. I could have had a direct and simple comparison if
the GTK developers hadn't simply ripped out the other options. It would
also have allowed me to used a GTK system that wasn't cripplingly slow.

> GtkPerf isn't really suited for a number/number comparison, however its
> not that bad to get hints whats going on. Keep in mind that some open
> windows larger than the root-window -> infalidate other windows. I
> recommend maximizing gtkperfs window before testing.
> 
>> I don't think that's acceptable at all. Cairo pretty obviously has
>> major performance problems (even if you accept that it is only on some
>> cards), and the GTK developers forced everyone to use it without even a
>> fallback.
> Where do you know that cairo is the problem?

That's the big change between the version of GTK used in FC4 and the one
in FC5. It's the logical place to start -- especially since there seem to
be plenty of complaints about its performance already.

> Maybe you could fire up your profiler and help a bit, the sooner more
> people work on performance the sooner we'll see better results.
> Especially the widget tuning stuff is something I don't know much about.

I'd be glad to... but I'm not an expert. If the instructions are
straightforward and don't involve major surgery on my system, I'll be glad
to collect information.