Bug#583653: xserver-xorg-video-radeon: 15x-slower performance regression in KMS mode for 2D operations
Hi,
On Sat, May 29, 2010 at 12:55:45PM +0200, Michel Dänzer wrote:
> On Sam, 2010-05-29 at 10:21 +0200, Sylvain Beucler wrote:
> >
> > When using the R300 driver in KMS mode (current default), performances
> > for 2D operations are 15x slower than without KMS mode (previous
> > default).
> >
> > The attached simple test runs at 17FPS in KMS mode on my computer,
> > against 300FPS in non-KMS mode.
>
> On a similar setup (RV350 in a PowerBook) here, it runs at about 60-70
> fps with compiz and just above 100 fps without compositing.
For reference I get 130 FPS on a Radeon Mobility M6 and 80 FPS on a S3
ProSavage (Lenny w/o compositing).
> BTW, are your UMS numbers from driver version 6.13?
Yes, I get 300FPS with:
$ cat /etc/modprobe.d/radeon-kms.conf
options radeon modeset=0
> With EXA or XAA?
Xorg.0.log says EXA is enabled.
With KMS: I tried without both, then with EXA and then with XAA.
AFAICT this didn't changed the performances. I mean I used:
Option "AccelMethod" "EXA"
or
Option "AccelMethod" "XAA"
in the configuration, possibly one of them is enabled by default.
> > This makes most 2D games unplayable.
>
> I'm not sure the test app is really representative for 'most 2D games'.
Sure.
'freedink', for instance, runs at < 10 FPS at 100% CPU with KMS
(vs. 60 FPS with at 75% CPU usage without KMS), which results in
choppy animation and sound.
The scrolling title screen 2D intro of Cytadela (which is otherwise a
3D game) is jumpy too.
I realized there was a problem when working on a small test game and
checking the FPS. After blaming Python and writing the C test, I had
to witness Python was not faulty this time ;)
> Forgot to mention: sysprof or oprofile profiles of slow and fast runs
> might be interesting, at least if the CPU is pegged during the runs.
Indeed CPU is at 100%.
Do you have instructions on how to best run sysprof or oprofile?
--
Sylvain
Reply to: