Hello,
I am often experiencing perfomance degradations with KDE 4.10.1.
After login the system seems to work rather smooth, but during day by day usage it slowly degrades to the point of some huge perfomance loss.
I noticed that sometimes, after opening some new Chromium-browser tab (the tab content doesn't matter) or after restoring a previous chromium session, compositing perfomance is brought to its knees and Xorg and kwin processes reach ~20% cpu usage respectively.
Closing chromium-browser and switching OpenGL effects off and back on reinstate cpu usage to normal values and compositing perfomance to an acceptable value, but not to the original (smooth) one.
I don't know who to blame (chromium, catalyst drivers, kwin) but AFAIR I didn't have these perfomance degradations with previous KDE versions, at least not to this extent.

Anyone having same issues? How may I debug further?
My setup: triple monitor setup, latest Catalyst driver, kubuntu KDE 4.10.1. (no OpenGL 2 shader enabled, no vsync)

Last thing to note: On the same system with a single monitor setup, compositing perfomance doesn't suffer at all.

EDIT: apart the avarage 10-20FPS loss with window operations, KDE Cube effect and window resizing turned out to be the two with worse behavior, barely reaching 8FPS, whilst OpenGL games still play _smooth_ at full speed, without any glitches.

I'd say: don't use Catalyst

mgraesslin wrote:I'd say: don't use Catalyst

I need Catalyst driver and it has worked for years.
I would rather use radeon driver if it had support for dual PCIe radeon cards (another Xorg is attached to a fourth monitor, as it's a multiseat environment), a three monitor setup (eyefinity) with OpenGL and some sort of video deconding acceleration. But last time I tried it hadn't.

Furthermore that problem appeared after upgrading to KDE 4.10.1 whereas I didn't upgrade catalyst driver in a while.

frasty wrote:Furthermore that problem appeared after upgrading to KDE 4.10.1 whereas I didn't upgrade catalyst driver in a while.

I have just looked through all changes in KWin between 4.10.0 and 4.10.1 - there are none which could affect Catalyst. The compositor has only seen one change and that's in the EGL backend which is obviously unsupported by Catalyst anyway.

qdbus org.kde.kwin /KWin supportInformation

Did you use 4.10.0 or 4.9.x before?
Any downstream patches ("backports") applied?

mgraesslin wrote:

frasty wrote:Furthermore that problem appeared after upgrading to KDE 4.10.1 whereas I didn't upgrade catalyst driver in a while.

I have just looked through all changes in KWin between 4.10.0 and 4.10.1 - there are none which could affect Catalyst. The compositor has only seen one change and that's in the EGL backend which is obviously unsupported by Catalyst anyway.

Thanks mgraesslin, I'll try with a blank KDE session and see if problem persists.

luebking wrote:qdbus org.kde.kwin /KWin supportInformation

http://pastebin.ca/2333090

luebking wrote:Did you use 4.10.0 or 4.9.x before?

4.10.0

luebking wrote:Any downstream patches ("backports") applied?

I just install packages from the backports repository:

plasma-desktop:
Installed: 4:4.10.1b-0ubuntu1~ubuntu12.10~ppa1
Candidate: 4:4.10.1b-0ubuntu1~ubuntu12.10~ppa1
Version table:
*** 4:4.10.1b-0ubuntu1~ubuntu12.10~ppa1 0
500 http://ppa.launchpad.net/kubuntu-ppa/backports/ubuntu/ quantal/main amd64 Packages

erm better do not active ShowFPS, it's making the system slower.

Did you give a try to OpenGL 2 shaders? With your GPU/driver it should perform better.

wild guess:
> cacheTexture: true
uncheck this (only checkable blur option) and maybe try w/o blur to track down the issue.

Finally, I found that logging out and back in also with a different user didn't improve compositing perfomance.
So, after killing X servers and done reloading of fglrx module, compositing perfomance is back to its initial smoothness. Probably some catalyst issue triggered by [ ]?? Now I am waiting how much time it'll take to get things slow down again.

Sounds like either the kernel module leaks or you might drop into some performance saving mode (can one inspect that with fglrx)?
At least the nvidia blob keeps one in performance mode when more than one screen is attached/enabled - reducing clock rates is usually no good idea (that's not a resolution matter but because it has to feed two different destinations simultaneously, ie. at doubled frequency)

luebking wrote:Sounds like either the kernel module leaks or you might drop into some performance saving mode (can one inspect that with fglrx)?
At least the nvidia blob keeps one in performance mode when more than one screen is attached/enabled - reducing clock rates is usually no good idea (that's not a resolution matter but because it has to feed two different destinations simultaneously, ie. at doubled frequency)

As you suggested, disabling the blur effect actually seems to have brought the system back to its normal smoothness! No more lagginess so far (two days uptime and still logged in). Previously (at least prior to upgrade to 4.10.1) blur effect didn't exposed any problems, though.

Anyway, last time I experienced lagginess I found that GPU load started to linger at ~60% even with no window movements. And yes I can confirm that also fglrx keeps clock rates fixed at their maximum when behaving with multiple screens attached.

Remains the question on whether it lacks on FBOs or cache, please deactivate
"cacheTexture: true" in "kcmshell4 kwincompositing", 2nd tabm blur config, the only checkbox.

Blurring and Lanczos filter were suppressed before 4.10 for fglrx (because w/o dri there was no glsl and the driver additionally crashed in the ARB shader)

luebking wrote:Remains the question on whether it lacks on FBOs or cache, please deactivate
"cacheTexture: true" in "kcmshell4 kwincompositing", 2nd tabm blur config, the only checkbox.

Blurring and Lanczos filter were suppressed before 4.10 for fglrx (because w/o dri there was no glsl and the driver additionally crashed in the ARB shader)

I Spoked too soon. After some browsing with chromium-browser, compositing lagginess kicked in. This time, killing chromium-browser restored normal behavior, although after some minutes I got two Xorg segfault:

Code: Select all

[615511.611] (EE) Backtrace:
[615511.633] (EE) 0: /usr/bin/X (xorg_backtrace+0x36) [0x7f2e92440b76]
[615511.633] (EE) 1: /usr/bin/X (0x7f2e92298000+0x1ac9a9) [0x7f2e924449a9]
[615511.633] (EE) 2: /lib/x86_64-linux-gnu/libpthread.so.0 (0x7f2e915be000+0xfcb0) [0x7f2e915cdcb0]
[615511.633] (EE) 3: /usr/lib/dri/fglrx_dri.so (0x7f2e882d3000+0x10e5421) [0x7f2e893b8421]
[615511.634] (EE) 4: /usr/lib/dri/fglrx_dri.so (0x7f2e882d3000+0x10e555d) [0x7f2e893b855d]
[615511.634] (EE) 5: /usr/lib/dri/fglrx_dri.so (0x7f2e882d3000+0x10f1708) [0x7f2e893c4708]
[615511.634] (EE) 6: /usr/lib/dri/fglrx_dri.so (0x7f2e882d3000+0x10f1884) [0x7f2e893c4884]
[615511.634] (EE) 7: /usr/lib/dri/fglrx_dri.so (0x7f2e882d3000+0x10e5647) [0x7f2e893b8647]
[615511.634] (EE) 8: /usr/lib/dri/fglrx_dri.so (0x7f2e882d3000+0xcde14a) [0x7f2e88fb114a]
[615511.634] (EE) 9: /usr/lib/dri/fglrx_dri.so (0x7f2e882d3000+0xca99d6) [0x7f2e88f7c9d6]
[615511.634] (EE) 10: /usr/lib/dri/fglrx_dri.so (0x7f2e882d3000+0xaf9891) [0x7f2e88dcc891]
[615511.634] (EE) 11: /usr/lib/dri/fglrx_dri.so (0x7f2e882d3000+0xed93d8) [0x7f2e891ac3d8]
[615511.634] (EE) 12: /usr/lib/dri/fglrx_dri.so (0x7f2e882d3000+0xef3489) [0x7f2e891c6489]
[615511.634] (EE) 13: /usr/lib/dri/fglrx_dri.so (0x7f2e882d3000+0xef3885) [0x7f2e891c6885]
[615511.634] (EE) 14: /usr/lib/dri/fglrx_dri.so (0x7f2e882d3000+0xed7054) [0x7f2e891aa054]
[615511.634] (EE) 15: /usr/lib/dri/fglrx_dri.so (0x7f2e882d3000+0xf0be3c) [0x7f2e891dee3c]
[615511.634] (EE) 16: /usr/lib/dri/fglrx_dri.so (0x7f2e882d3000+0x270d55) [0x7f2e88543d55]
[615511.634] (EE) 17: /usr/lib/dri/fglrx_dri.so (0x7f2e882d3000+0x276c78) [0x7f2e88549c78]
[615511.634] (EE) 18: /usr/lib/dri/fglrx_dri.so (0x7f2e882d3000+0x2cd5d5) [0x7f2e885a05d5]
[615511.634] (EE) 19: /usr/lib/dri/fglrx_dri.so (0x7f2e882d3000+0x2cda1d) [0x7f2e885a0a1d]
[615511.634] (EE) 20: /usr/lib/dri/fglrx_dri.so (0x7f2e882d3000+0x2cdb5a) [0x7f2e885a0b5a]
[615511.634] (EE) 21: /usr/lib/dri/fglrx_dri.so (0x7f2e882d3000+0x3e38b5) [0x7f2e886b68b5]
[615511.634] (EE) 22: /usr/lib/dri/fglrx_dri.so (0x7f2e882d3000+0x3e3b78) [0x7f2e886b6b78]
[615511.634] (EE) 23: /usr/lib/dri/fglrx_dri.so (0x7f2e882d3000+0xf0a2b4) [0x7f2e891dd2b4]
[615511.634] (EE) 24: /usr/lib/dri/fglrx_dri.so (0x7f2e882d3000+0x186dc68) [0x7f2e89b40c68]
[615511.634] (EE) 25: /usr/lib/dri/fglrx_dri.so (0x7f2e882d3000+0x186b15c) [0x7f2e89b3e15c]
[615511.634] (EE) 26: /usr/lib/x86_64-linux-gnu/xorg/extra-modules/extra-modules.dpkg-tmp/modules/extensions/libglx.so (0x7f2e8f33d000+0x1ac07) [0x7f2e8f357c07]
[615511.634] (EE) 27: /usr/lib/x86_64-linux-gnu/xorg/extra-modules/extra-modules.dpkg-tmp/modules/extensions/libglx.so (0x7f2e8f33d000+0x18edf) [0x7f2e8f355edf]
[615511.634] (EE) 28: /usr/lib/x86_64-linux-gnu/xorg/extra-modules/extra-modules.dpkg-tmp/modules/extensions/libglx.so (0x7f2e8f33d000+0x1f147) [0x7f2e8f35c147]
[615511.634] (EE) 29: /usr/bin/X (0x7f2e92298000+0x55a91) [0x7f2e922eda91]
[615511.634] (EE) 30: /usr/bin/X (0x7f2e92298000+0x4456a) [0x7f2e922dc56a]
[615511.634] (EE) 31: /lib/x86_64-linux-gnu/libc.so.6 (__libc_start_main+0xed) [0x7f2e9024476d]
[615511.634] (EE) 32: /usr/bin/X (0x7f2e92298000+0x448ad) [0x7f2e922dc8ad]
[615511.634] (EE)
[615511.634] (EE) Segmentation fault at address 0xf0

also chromium-browser crashed:

Code: Select all

[614819.544741] chromium-browse[16744]: segfault at f0 ip 00007f4dd9e262d3 sp 00007fff77f64ac0 error 4 in fglrx_dri.so[7f4dd8d44000+1c9b000]
[614868.345072] chromium-browse[14040]: segfault at f0 ip 00007f5fc208a421 sp 00007fffa1d55230 error 4 in fglrx_dri.so[7f5fc0fa5000+1c9b000]
[614960.489910] chromium-browse[14202]: segfault at f0 ip 00007f6008d962d3 sp 00007fffa0825d20 error 4 in fglrx_dri.so[7f6007cb4000+1c9b000]
[615225.760988] chromium-browse[15607]: segfault at 30 ip 00007f2e9f39cdbd sp 00007fff3b43c838 error 4 in libflashplayer.so[7f2e9ecc1000+117d000]
[615273.077503] chromium-browse[15258]: segfault at f0 ip 00007f422855f421 sp 00007fff1537e490 error 4 in fglrx_dri.so[7f422747a000+1c9b000]
[615486.955801] Assertion failed in ../../../../../../../../drivers/2d/lnx/fgl/drm/kernel/hal_evergreen.c at line: 69
[615834.908906] chromium-browse[19487]: segfault at f0 ip 00007f2622389421 sp 00007ffff7966870 error 4 in fglrx_dri.so[7f26212a4000+1c9b000]
[615836.538959] chromium-browse[21200]: segfault at f0 ip 00007f78220dc421 sp 00007fff50f7cfc0 error 4 in fglrx_dri.so[7f7820ff7000+1c9b000]
[615887.634483] chromium-browse[21221]: segfault at f0 ip 00007fd0a492d421 sp 00007fff1a7c52a0 error 4 in fglrx_dri.so[7fd0a3848000+1c9b000]

After that, I was no more able to enable OpenGL compositing (only Xrender) and I started suffering from ghost windows and redrawing issues all over the desktop.

Now I'm going to reload the fglrx module and I'll try with "cacheTexture: true" disabled.

Today's behavior (with "cacheTexture: true" deactivated):
Window effects and general window movements are working flawlessly on each screen for Activities 1 & 2.
Suddenly on third Activity, on second and third screens, desktop effects and also simple window movements started to tremendously lag, no matter which application is running on, whilst on the first screen everything still runs smooth as do on all the three screens of activities 1 & 2.
Immediately after, plasma-desktop crashed and chromium-browser tabs occasionally started to flicker.

I've created another activity and it resambles the behavior of the third activity, with screen0 ok and screen1 and screen2 simply unusable.

This is going to be frustrating. Now I'll try with only two activities.........