Re: Question about the performance of gtk+ in neo1973 mobile phone

["Daniel Amelang" <daniel amelang gmail com>]

On 5/23/07, Koen Kooi <koen dominion kabel utwente nl> wrote:
> -----BEGIN PGP SIGNED MESSAGE-----
> Hash: SHA1
>
> Daniel Amelang schreef:
>
> > If slow pixbuf rendering were a problem, you may be interested in a
> > patch I submitted recently to xorg that speeded up pixbuf rendering on
> > R5G6B5 framebuffers by 3.4x :
> >
> > http://lists.freedesktop.org/archives/xorg/2007-April/024286.html
> >
> > That speedup was measured on a N800, but the neo1973 should see
> > similar improvement.
>
> I gave the testapp and patch a spin:
>
> ipaq hx4700:
>
> unpatched Xw100:  6.04 FPS
> patched Xw100:    6.04 FPS
> unpatched Xfbdev: 6.04 FPS
> patched Xfbdev:   6.04 FPS
>
> ipaq h2200:
>
> unpatched Xfbdev: 4.18 FPS
> patched Xfbdev:   4.17 FPS
>
> Any clues why the patch doesn't seem to have an effect?

Assuming you didn't make any errors in applying the patch and getting
the patched server compiled and up and running (I do have faith in you
:), it's likely that for your combination of Gdk+Xserver+hardware, the
test program doesn't hit the exact same composite case
(fbCompositeSrc_8888RevNPx0565) as it does on my N800.

Easiest way to tell that would to add a couple ErrorF calls in
fbpict.c to see what functions are called. ErrorF messages get spat
out to stderr, if I remember correctly.

You didn't happen to apply that patch by hand, did you? I can think of
a couple tricky places that might have tripped you up if you did.
Especially the part about adding the new function to the fbComposite
"switch case of doom". Care to send me your patched fbpict.c?

Dan Amelang