Bug 10474

Summary: radeon r350: EXA makes gtk and qt apps unusable slow
Product: xorg Reporter: Hannes Janetzek <jeff>
Component: Server/Acceleration/EXAAssignee: Xorg Project Team <xorg-team>
Status: RESOLVED FIXED QA Contact: Xorg Project Team <xorg-team>
Severity: normal    
Priority: medium CC: jeff
Version: 7.2 (2007.02)Keywords: NEEDINFO
Hardware: Other   
OS: All   
Whiteboard:
i915 platform: i915 features:
Attachments:
Description Flags
Xorg log,etc
none
Xord logs, dmesg output none

Description Hannes Janetzek 2007-03-30 06:47:36 UTC
I have tested this with ubuntus ati driver 1:6.6.3.2 and the xf86-video-ati-6.6.191 snapshot. I have a dell inspiron 8600c laptop. 
The strange thing is that xterm and my e17wm(wich has its own widget tk) are drawn as fast as normally but gtk and qt apps became unusable slow when i use EXA.
For example sylpheed-claws needs one minute to become fully drawn and uses 100% cpu for so long.
Comment 1 Michel Dänzer 2007-03-30 07:14:55 UTC
Which version of the X server? EXA is only expected to run reasonably well as of xserver 1.3.
Comment 2 Hannes Janetzek 2007-03-30 07:35:09 UTC
Forgot to mention: The last few months EXA worked really good. 
Comment 3 Hannes Janetzek 2007-03-30 07:39:22 UTC
(In reply to comment #1)
> Which version of the X server? EXA is only expected to run reasonably well as
> of xserver 1.3.
> 
ah, ok. I will try that
Comment 4 Michel Dänzer 2007-03-30 07:47:46 UTC
(In reply to comment #2)
> Forgot to mention: The last few months EXA worked really good. 

What changed in the setup?

It might be useful if you attached (as opposed to paste) a full log file.
Comment 5 Hannes Janetzek 2007-03-30 08:57:53 UTC
Created attachment 9384 [details]
Xorg log,etc
Comment 6 Hannes Janetzek 2007-03-30 08:58:20 UTC
(In reply to comment #4)
> (In reply to comment #2)
> > Forgot to mention: The last few months EXA worked really good. 
> 
> What changed in the setup?
> 
Hard to tell, since I just updated with apt. I switched to ubuntu 7.0 recently, but then EXA still worked for some weeks.

> It might be useful if you attached (as opposed to paste) a full log file.
> 
I sent this bug report already to ubuntus bugzilla. I made the attached package for that. It includes Xorg.0.log.   i also added some profiling with oprofile. it shows the cpu-usage while resizing the frames inside sylpheed-claws. it seems that all cpu goes into memcpy operations. 

Comment 7 Michel Dänzer 2007-03-30 09:16:05 UTC
(In reply to comment #6)
> (In reply to comment #4)
> > (In reply to comment #2)
> > > Forgot to mention: The last few months EXA worked really good. 
> > 
> > What changed in the setup?
> > 
> Hard to tell, since I just updated with apt. 

/var/log/dpkg.log* might be useful.

> I switched to ubuntu 7.0 recently, but then EXA still worked for some weeks.

Could it be that xserver-xorg-core went from 1.1 to 1.2? If so, Option "MigrationHeuristic" "greedy" might give you back the previous performance.

Even better performance might be possible with xserver 1.3 RCs and xf86-video-ati 6.6.191, without the above option but with Option "AccelDFS" instead.
Comment 8 Hannes Janetzek 2007-03-30 09:49:54 UTC
> > (In reply to comment #4)
> > > (In reply to comment #2)
> > > > Forgot to mention: The last few months EXA worked really good. 
> > > 
> > > What changed in the setup?
> > > 
> > Hard to tell, since I just updated with apt. 
> 
> /var/log/dpkg.log* might be useful.
>  
hm no, mine has only entries from last 4 days


> Could it be that xserver-xorg-core went from 1.1 to 1.2? If so, Option
> "MigrationHeuristic" "greedy" 
what ?! :)
> might give you back the previous performance.
I now switched to xserver 1.3(git) and rebuild the drivers. Could some libraries
 also need to be updated? 
Drawing is already faster: about one update per second when drawing a big window. 
yes, indeed this Option makes it usable again, though xcompmgr doesn't work as g
ood as before

> Even better performance might be possible with xserver 1.3 RCs and
> xf86-video-ati 6.6.191, without the above option but with Option "AccelDFS"
> instead.

Nope this option doesn't increase drawing speed.  
>
i attached the Xorg logs with these option
I don't know if this has something to do with this, but I heard that MTRR has something to do with drm. dmesg says mtrr: no MTRR for d0000000,8000000 found 
and this region could be my graphics card memory, i think.
reg04: base=0xd0000000 (3328MB), size= 128MB: write-combining, count=1

Comment 9 Hannes Janetzek 2007-03-30 09:51:41 UTC
Created attachment 9391 [details]
Xord logs, dmesg output
Comment 10 Hannes Janetzek 2007-03-30 10:13:54 UTC
no, the mtrr message isn't the cause. after rebooting the message didn't appeared again. 
Comment 11 Michel Dänzer 2007-03-31 08:33:42 UTC
(In reply to comment #8)
> I now switched to xserver 1.3(git) and rebuild the drivers. Could some
> libraries also need to be updated? 

Shouldn't be necessary.

> i attached the Xorg logs with these option

The log files are all identical and show XAA being used instead of EXA, please clarify.
Comment 12 Hannes Janetzek 2007-04-01 08:00:58 UTC
sorry for the mess. i realized that the self compiled xserver doesn't produce a /var/lib/Xorg.0.log so that is why both show the same old log.

Comment 13 Michel Dänzer 2007-04-01 08:25:54 UTC
(In reply to comment #12)
> sorry for the mess. i realized that the self compiled xserver doesn't produce a
> /var/lib/Xorg.0.log so that is why both show the same old log.

Does it put it in /usr/local/var/log/Xorg.0.log? The X server stdout should tell.
Comment 14 Benjamin Close 2008-01-11 02:39:24 UTC
Bugzilla Upgrade Mass Bug Change

NEEDSINFO state was removed in Bugzilla 3.x, reopening any bugs previously listed as NEEDSINFO.

  - benjsc
    fd.o Wrangler
Comment 15 Michel Dänzer 2008-12-14 04:50:42 UTC
Is this still an issue with current versions of the X server and driver?
Comment 16 Jerome Glisse 2009-05-20 04:37:21 UTC
Closing this bug for inactivity, if you still have this issue with recent Xorg, ddx (xf86-video-ati) and mesa please reopen and update bugs informations.

Use of freedesktop.org services, including Bugzilla, is subject to our Code of Conduct. How we collect and use information is described in our Privacy Policy.