On Fri, 2009-02-27 at 20:25 -0500, Joshua C. wrote:
I was testing the latest xf86-video-radeon with f11 and the latest
xorg-x11-server. And to my surprise the fedora version performs much
worse than the vanilla code here:
http://lists.freedesktop.org/archives/xorg-announce/2009-February/000771.....
Installed are:
mesa-libGLU-7.3-5.fc11.x86_64
mesa-libGL-7.3-5.fc11.x86_64
mesa-dri-drivers-7.3-5.fc11.x86_64
libdrm-2.4.5-0.fc11.x86_64
xorg-x11-server-utils-7.4-5.fc11.x86_64
xorg-x11-server-common-1.6.0-2.fc11.x86_64
xorg-x11-server-Xorg-1.6.0-2.fc11.x86_64
xorg-x11-drv-ati-6.11.0-1.1.fc11.x86_64 (from vanilla source without
extra patches)
kernel-2.6.29-0.172.rc6.git4.fc11.x86_64
With the fedora version (with all the patches) I cannot do basic
things like watch flash files and video (almost no hardware
acceleration). This results in 100% cpu usage, screen distortions. At
the same time the vanilla code playes the video files smoothly and
everything is fine. Even the kms works! With the fedora code I need to
pass radeon.gartsize=128 at startup in order to make it start the
xserver. And in terms of speed it's far behind the vanilla code.
I saw that some patches date back to before 6.10.0. Are they still
needed? Has someone else tried this?
So if you run this vanilla driver please disable KMS, you are running
two drivers in direct conflict, and it might work sometimes, it might
also corrupt memory or oops randomly. So its not on the advised list.
However can you check if the driver you installed from upstream is
running XAA or EXA acceleration? We enable EXA acceleration by default
under KMS and non-kms, and it may be the case that XAA is faster for
some situations that we haven't identified yet. However EXA provides
support for more features going forward so we have to try and squash any
regressions at some point.
I think we will switch upstream to EXA quite soon, which sorta means you
just put off the inevitable at the moment.
Dave.