2D (and 3D) Radeon acceleration partially solved.
 

I finally managed to get my Radeon 9k pro up with XFree 4.2.

That involved:

- Carefully recompiling the 2.4 kernel and installing the newest
ATI 3.7.0 driver as prescribed, compiling the fglrx module.

- Commenting out a wrong "BusID 1:0:1" that fglrxconfig inserted
into /etc/X11/XF86Config-4.

Now OpenGL (Tuxracer) works fine, but 2D acceleration does not!

By modifying a line in the config file thus:

Option "no_dri" "yes"

I get this message in /var/log/XFree86.0.log :

(WW) fglrx(0): ***********************************
(WW) fglrx(0): * DRI initialization disabled! *
(WW) fglrx(0): * 2D acceleraton available (MMIO) *
(WW) fglrx(0): * no 3D acceleration available *
(WW) fglrx(0): ***********************************

and now 2D-blits go 13 times faster, but OpenGL almost stops!

I measure my 2D performance with "x11perf -copypixpix500"
and get about 2000 pr sec with accel, 150 without.

It is unbearable if one has to choose between 2D and 3D
acceleration, really. I'm writing a smooth scrolling routine for
our open source tile-based game (Freeciv), so this worries me a
litte; I'm not much for 3D.

My question is: Do GeForce users have the same problem?

More to the point, should I expect that most linux users lack 2D
hardware acceleration? I sort of hope that there may be something
lacking with my setup, but this doesn't seem to be documented
anywhere.

Any help much appreciated.


-ali

My configuration shows exactly the same symptoms. Enabled 3D-Acceleration decreases 2D-Performance severely. The test "Copy 500x500 from pixmap to window" looses up to 92% performance.

Gnome's reactions when moving or resizing windows get very slow when 3D-Acceleration is enabled, especially when those cool window-decorations are enabled too.

I wonder why this thread got no replies, and I didn´t find a similar thread, so I´m bumping it up again with the following questions:

- Do other Radeon-Users experience this too?
- Is this a general Radeon-driver problem, or do you think my configuration is faulty?

Thanks for any hints.

Some System-Infos:

Duron 1.2GHZ, 384MB Ram, SuSE9.0 with actual online-updates, XFree86 Version 4.3.0.1, Gnome 2.2.2 with metacity-WM, Radeon9500 128MB with 3.7.0-25 Drivers from SuSE-FTP Server. I followed SuSE´s Radeon driver-installation guide.

I have made several test with all the radeon driver that i have (all downloaded at ati)and i think that it's a driver issue......i have the same problems as your's......up to 93% performance loss!!!
I have a athlon 2100+ radeon 9700pro via kt400 mobo 512 meg ram and mandrake 9.2/9.1.
I hope someone is able to solve this....if there's a solution....

Similar problem here
 

I have Suse 9.0 with the latest updates installed. I installed the ATI 3.7.0 driver from Suse's ftp site.

Using KDE and glxgears to test 3d performance, I get 2100 frames per second.

For kicks, I decided to install Ximian Desktop (gnome based from what I can tell). With NO other changes, glxgears performance dropped to 300 frames per second.

I verified that the driver was still installed and dri was enabled, yet I cannot explain the huge degradation in performance.

I don't want to incite any linux desktop war, but is it possible that this is a Gnome issue? Or at the very least an interoperability problem between the ATI drivers and Gnome?

I don't have tried gnome but i think that there are something wrong with the ati drivers....

This is not a desktop issue. It's XFree, your "windows", that communicates
with the hardware through the vendor-specific kernel-module driver.

Problem is, when 2D accelleration is not effective, all desktop performance
will suffer enormously, plus the CPU usage will increase a lot.

I've notified ATI about the problem, hoping they can fix it. I don't really
understand "DRI" or "MMIO", but since these messages from the ATI-
specific module appeared covertly in the logfile as described above, I'm
inclined to blame ATI. Still don't know if GeForce users have the same
problem..

I also have a huge problem with the fact that the "no_dri" trick was
not documented anywhere, and I had to find out the hard way in order
to make 2D work at all.