Re: Q4 2009 report

[Steve Sakoman <sakoman gmail com>]

On Wed, Jan 13, 2010 at 1:08 AM, Dave Neary <dneary gnome org> wrote:

> Do you have any suggestions for useful baselining of performance?

Not really.  My main motivation is to look for performance regressions
over time.

> Running these figures over time will tell us if one component gets
> slower, but it won't tell us which components are slow already, compared
> to (say) relative performance on x86.

Quite true.

> So the GtkComboBox tests took ~29 times the GtkEntry tests on ARM - if
> they only take 6 times the tests on x86, we know that either there's a
> huge speed-up on GtkEntry or a slow-down on GtkComboBox. Would that be a
> useful thing to look for?

I think that would be a great starting point to help decide where to
focus for ARM improvements.  Of course we'd also need to factor in
whether a particular test measures something that matters :-)

> For reference, I found this gtkperf run on an xorg list:
> http://lists.x.org/archives/xorg-driver-ati/2009-June/010229.html
>
> Comparing your results and those, GtkSpinButton and GtkToggleButton
> appear between 25%  and 33% slower, GtkTextView is about 35% slower, and
> GtkDrawingArea is 30% slower for circles. Funny enough, GtkDrawingArea
> is faster relative for lines and pixbufs.

I'm going to focus first on selecting a set of tests and automating
them as part of my build/test process.  Sadly there is still a whole
bunch of work to do in fixing bugs, so performance work is a bit low
on the to do list!

Steve