Bug#523844: Upgrade to xserver-xorg-video-intel 2:2.7.0-1 doesn't help.
sorry,post from google groups seems not working,So I post again. Hi Gunter, thanks for going through this pain of kernel testing :) On Sat, 2009-04-18 at 11:24 +0200, Gunter Ohrner wrote: Unfortunately PAT does not seem to be the (only) culprit, I still get: $ egrep 'tiling' /var/log/Xorg.0.log (EE) intel(0): Failed to set tiling on front buffer: rejected by kernel (EE) intel(0): Failed to set tiling on back buffer: rejected by kernel (EE) intel(0): Failed to set tiling on depth buffer: rejected by kernel as with the original 2.6.29 image and Xorg compositing performance is still unusuably slow. with 2.6.29 If use EXA ,I also get these three error lines but if use UXA, only the front buffer error occurs AIUI there are tiling fixes queued for upcoming kernels. Hopefully 2.6.30 will behave better... Yes,I followed this ,and built a custom 2.6.30-rc3 kernel from git the errors disappear.but still UXA is much faster than EXA here's my simple benchmark result: x11perf -aa10text glxgears EXA: 459000.0/sec 45FPS UXA :654000.0/sec430FPS -- To UNSUBSCRIBE, email to debian-bugs-dist-requ...@lists.debian.org with a subject of unsubscribe. Trouble? Contact listmas...@lists.debian.org
Bug#523844: Upgrade to xserver-xorg-video-intel 2:2.7.0-1 doesn't help.
darren wrote: x11perf -aa10text glxgears EXA: 459000.0/sec 45FPS UXA :654000.0/sec430FPS The glxgears column is meaningless. Brice -- To UNSUBSCRIBE, email to debian-bugs-dist-requ...@lists.debian.org with a subject of unsubscribe. Trouble? Contact listmas...@lists.debian.org
Bug#523844: Upgrade to xserver-xorg-video-intel 2:2.7.0-1 doesn't help.
On Mon, Apr 27, 2009 at 8:17 PM, Brice Goglin brice.gog...@ens-lyon.org wrote: darren wrote: x11perf -aa10text glxgears EXA: 459000.0/sec 45FPS UXA : 654000.0/sec 430FPS The glxgears column is meaningless. May I quote Gunter Ohrne: glxgear which is no meaningful benchmark, I know, but it's a tendency indicator nontheless I benchmarked the 2D performance using gtkperf and render_bench[1] again (KMS is disabled in all four cases,since I can't get the X running,it just hangs and doesn't respond to any keyboard input,I can not use ssh to connect to it from another host) which gives me the following result: -- gtkperf render_bench Kernel-2.6.29/exa 14.03 36.52 kernel-2.6.29/uxa 14.9 36.04 kernel-2.6.30-rc3/exa16.136.18 kernel-2.6.30-rc3/uxa13.32 35.39 --- gtkperf runs for five times and the result is total average render time. render_bench just runs once in each case,I simply add the total time . No wonder kernel-2.6.30-rc3/uxa is the fastest combination both with gtkperf and render_bench. gtkperf shows that performace in kernel-2.6.30-rc3/exa is even worse than in Kernel-2.6.29/exa Maybe the benchmark I did has flaws and meaningless,but the performace really improve with kernel-2.6.30/uxa from my feeling with running some applications. with kernel-2.6.29 suspend and wake will crash the X,while with kernel-2.6.30rc3 it works again.So I am anxious for the release of kernel-2.6.30 [1] http://www.rasterman.com/files/render_bench.tar.gz Brice first second third forth fifth AVERAGE TIME kernel 2.6.30rc3/EXA GtkEntry0.090.110.160.090.1 0.11 GtkComboBox 3.363.563.513.293.343.41 GtkComboBoxEntry3.513.643.242.823.543.35 GtkSpinButton 0.430.560.430.510.550.5 GtkProgressBar 0.290.390.430.360.480.39 GtkToggleButton 0.270.390.470.3 0.560.4 GtkCheckButton 0.080.130.120.080.310.14 GtkRadioButton 0.130.250.330.190.240.23 GtkTextView - Add text 0.560.690.670.720.9 0.71 GtkTextView – Scroll0.430.590.540.5 0.020.42 GtkDrawingArea – Lines 1.5 1.791.831.742.4 1.85 GtkDrawingArea – Circles1.952.332.182.252.73 2.29 GtkDrawingArea - Text 1.881.841.781.821.541.77 GtkDrawingArea – Pixbufs0.550.590.550.5 0.5 0.54 --- Total time 15.03 16.86 16.24 15.17 17.21 16.1 kernel 2.6.30rc3/UXA GtkEntry - time 0.090.070.090.110.1 0.09 GtkComboBox - time 2.4 2.582.822.893.1 2.76 GtkComboBoxEntry - time 2.782.792.872.6 2.462.7 GtkSpinButton - time0.420.520.410.360.520.45 GtkProgressBar - time 0.250.230.240.240.250.24 GtkToggleButton - time 0.320.260.250.3 0.240.27 GtkCheckButton - time 0.110.140.140.120.080.12 GtkRadioButton - time 0.160.130.180.120.130.14 GtkTextView - Add text - time 0.550.630.630.640.63 0.62 GtkTextView - Scroll - time 0.410.540.430.440.56 0.48 GtkDrawingArea - Lines - time 1.5 1.361.361.351.51 1.42 GtkDrawingArea - Circles - time 2.852.812.832.812.82 2.82 GtkDrawingArea - Text - time1.030.921.021.150.84 0.99 GtkDrawingArea - Pixbufs - time 0.2 0.240.220.210.22 0.22 --- Total time 13.07 13.24 13.49 13.34 13.46 13.32 kernel-2.6.29/EXA GtkEntry - time 0.1 0.090.080.070.06
Bug#523844: Upgrade to xserver-xorg-video-intel 2:2.7.0-1 doesn't help.
Am Freitag, 17. April 2009 schrieb Brice Goglin: $ uname -a Linux Zweiblum 2.6.26-2-686 #1 SMP Thu Mar 26 01:08:11 UTC 2009 i686 GNU/Linux $ egrep '_PAT(_| )' /boot/config-2.6.26-2-686 # CONFIG_X86_PAT is not set PAT wasn't supported at all in 2.6.26. Mh, at least the configuration option is mentioned in the kernel config, even though it's disabled. But it doesn't mean that it's the same than having PAT disabled in 2.6.29 since the whole ioremap implementation has been intrusively modified. I'm reporting back with some delay as I struggled a bit to produce a custom compiled 2.6.29 kernel - I was not prepared for the initrd and kernel-patch changes introduced by kernel-package 12 and documentation of the new behaviour is still rather sparse... I now rebuild the Debian 2.6.29 kernel source package using the same configuration as the official image, but I enabled PAT. Unfortunately PAT does not seem to be the (only) culprit, I still get: $ egrep 'tiling' /var/log/Xorg.0.log (EE) intel(0): Failed to set tiling on front buffer: rejected by kernel (EE) intel(0): Failed to set tiling on back buffer: rejected by kernel (EE) intel(0): Failed to set tiling on depth buffer: rejected by kernel as with the original 2.6.29 image and Xorg compositing performance is still unusuably slow. $ uname -a Linux Zweiblum 2.6.29.debian.pat.20090418.1 #1 SMP Sat Apr 18 10:26:43 CEST 2009 i686 GNU/Linux $ egrep '_PAT(_|=| )' /boot/config-2.6.29.debian.pat.20090418.1 CONFIG_X86_PAT=y $ egrep PAT /var/log/dmesg [0.00] x86 PAT enabled: cpu 0, old 0x7040600070406, new 0x7010600070106 [0.004000] x86 PAT enabled: cpu 1, old 0x7040600070406, new 0x7010600070106 Compared to the official 2.6.29 image, x11perf shows doubled performance: 120 trep @ 0.0218 msec ( 46000.0/sec): Char in 80-char aa line (Charter 10) However this is still several times slower than with the Debian kernel 2.6.26 image and does not improve usability. Greetings, Gunter signature.asc Description: This is a digitally signed message part.
Bug#523844: Upgrade to xserver-xorg-video-intel 2:2.7.0-1 doesn't help.
Hi Gunter, thanks for going through this pain of kernel testing :) On Sat, 2009-04-18 at 11:24 +0200, Gunter Ohrner wrote: Unfortunately PAT does not seem to be the (only) culprit, I still get: $ egrep 'tiling' /var/log/Xorg.0.log (EE) intel(0): Failed to set tiling on front buffer: rejected by kernel (EE) intel(0): Failed to set tiling on back buffer: rejected by kernel (EE) intel(0): Failed to set tiling on depth buffer: rejected by kernel as with the original 2.6.29 image and Xorg compositing performance is still unusuably slow. AIUI there are tiling fixes queued for upcoming kernels. Hopefully 2.6.30 will behave better... Cheers, Julien -- To UNSUBSCRIBE, email to debian-bugs-dist-requ...@lists.debian.org with a subject of unsubscribe. Trouble? Contact listmas...@lists.debian.org
Bug#523844: Upgrade to xserver-xorg-video-intel 2:2.7.0-1 doesn't help.
Am Donnerstag, 16. April 2009 schrieb Brice Goglin: Gunter Ohrner wrote: So, what's wrong - or at least that different - with Debian's 2.6.29 kernel? PAT is disabled in 2.6.29, we feel like it may explain some performance problems if the video memory isn't in write combining anymore. Using EXA, the measured x11perf aa font rendering performance is 5 to 6 times better with the PAT kernel. The subjective impact to compositing performance is even larger - significantly larger. While using the KDE 4 desktop is virtually impossible with EXA and the Debian Linux 2.6.29 image - the window gallery task switcher resembles a slide show, no animation is visible at all - it feels responsive and fluent using the 2.6.29 linux image. Greetings, Gunter Ohrner signature.asc Description: This is a digitally signed message part.
Bug#523844: Upgrade to xserver-xorg-video-intel 2:2.7.0-1 doesn't help.
Am Donnerstag, 16. April 2009 schrieb Brice Goglin: PAT is disabled in 2.6.29, we feel like it may explain some performance problems if the video memory isn't in write combining anymore. Mh, I just checked my 2.6.26's image configuration, and it also seems to have PAT disabled: $ uname -a Linux Zweiblum 2.6.26-2-686 #1 SMP Thu Mar 26 01:08:11 UTC 2009 i686 GNU/Linux $ egrep '_PAT(_| )' /boot/config-2.6.26-2-686 # CONFIG_X86_PAT is not set $ egrep '_PAT(_| )' /boot/config-2.6.29-1-686 # CONFIG_X86_PAT is not set I diffed the Xorg.0.log with kernel 2.6.29 (extremely slow) and 2.6.26 (well usuable) and attached the results. Maybe the following Failed to set tiling are an indication of the problem - lines starting with - are from the 2.6.29 log, lines starting with + are from xorg running on 2.6.26: @@ -440,11 +440,14 @@ (**) intel(0): Tiling enabled (==) intel(0): VideoRam: 262144 KB (II) intel(0): Attempting memory allocation with tiled buffers. -(EE) intel(0): Failed to set tiling on front buffer: rejected by kernel -(EE) intel(0): Failed to set tiling on back buffer: rejected by kernel -(EE) intel(0): Failed to set tiling on depth buffer: rejected by kernel (II) intel(0): Tiled allocation successful. (II) intel(0): [drm] Registers = 0xdc10 +(II) intel(0): [drm] ring buffer = 0xc000 +(II) intel(0): [drm] mapped front buffer at 0xc100, handle = 0xc100 +(II) intel(0): [drm] mapped back buffer at 0xc500, handle = 0xc500 +(II) intel(0): [drm] mapped depth buffer at 0xc600, handle = 0xc600 +(II) intel(0): [drm] mapped classic textures at 0xc700, handle = 0xc700 +(II) intel(0): [drm] Initialized kernel agp heap manager, 33554432 (II) intel(0): [dri] visual configs initialized (II) intel(0): vgaHWGetIOBase: hwp-IOBase is 0x03d0, hwp-PIOOffset is 0x (**) intel(0): Option MigrationHeuristic greedy @@ -458,32 +461,35 @@ Greetings, Gunter xorg_log_linux_2.6.29_2.6.26.diff.gz Description: GNU Zip compressed data signature.asc Description: This is a digitally signed message part.
Bug#523844: Upgrade to xserver-xorg-video-intel 2:2.7.0-1 doesn't help.
Gunter Ohrner wrote: Mh, I just checked my 2.6.26's image configuration, and it also seems to have PAT disabled: $ uname -a Linux Zweiblum 2.6.26-2-686 #1 SMP Thu Mar 26 01:08:11 UTC 2009 i686 GNU/Linux $ egrep '_PAT(_| )' /boot/config-2.6.26-2-686 # CONFIG_X86_PAT is not set $ egrep '_PAT(_| )' /boot/config-2.6.29-1-686 # CONFIG_X86_PAT is not set PAT wasn't supported at all in 2.6.26. But it doesn't mean that it's the same than having PAT disabled in 2.6.29 since the whole ioremap implementation has been intrusively modified. Brice -- To UNSUBSCRIBE, email to debian-bugs-dist-requ...@lists.debian.org with a subject of unsubscribe. Trouble? Contact listmas...@lists.debian.org
Bug#523844: Upgrade to xserver-xorg-video-intel 2:2.7.0-1 doesn't help.
Gunter Ohrner wrote: Hi! I just upgraded to package version 2:2.7.0-1 with no improvement. Composite mode with EXA still is unusuably slow. Can you try without another kernel? Either Debian's old 2.6.26 or a custom 2.6.29 with PAT enabled. I still have to check whether XAA works again with this driver release, however. XAA will be dropped in the near anyway... Brice -- To UNSUBSCRIBE, email to debian-bugs-dist-requ...@lists.debian.org with a subject of unsubscribe. Trouble? Contact listmas...@lists.debian.org
Bug#523844: Upgrade to xserver-xorg-video-intel 2:2.7.0-1 doesn't help.
Am Donnerstag, 16. April 2009 schrieb Brice Goglin: I just upgraded to package version 2:2.7.0-1 with no improvement. Composite mode with EXA still is unusuably slow. Can you try without another kernel? Either Debian's old 2.6.26 or a custom 2.6.29 with PAT enabled. # uname -a Linux Zweiblum 2.6.26-2-686 #1 SMP Thu Mar 26 01:08:11 UTC 2009 i686 GNU/Linux Ok, guess what - with Debian's 2.6.26 package it feels quick and snappy. # x11perf -aa10text (...) 360 trep @ 0.0073 msec (137000.0/sec): Char in 80-char aa line (Charter 10) # glxgears -info (...) 1912 frames in 5.0 seconds = 382.250 FPS So, what's wrong - or at least that different - with Debian's 2.6.29 kernel? Thanks a lot in advance, now I'd just like to know what I've just observed... Greetings, Gunter -- +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ kyourek There was a 23% drop in temperature. nappyjallapy That's almost 25%! kyourek ... That was one of the most worthless comments I've ever heard.-- http://www.bash.org/?2999 +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ + PGP-verschlüsselte Mails bevorzugt! + +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ signature.asc Description: This is a digitally signed message part.
Bug#523844: Upgrade to xserver-xorg-video-intel 2:2.7.0-1 doesn't help.
Gunter Ohrner wrote: # uname -a Linux Zweiblum 2.6.26-2-686 #1 SMP Thu Mar 26 01:08:11 UTC 2009 i686 GNU/Linux Ok, guess what - with Debian's 2.6.26 package it feels quick and snappy. # x11perf -aa10text (...) 360 trep @ 0.0073 msec (137000.0/sec): Char in 80-char aa line (Charter 10) # glxgears -info (...) 1912 frames in 5.0 seconds = 382.250 FPS So, what's wrong - or at least that different - with Debian's 2.6.29 kernel? PAT is disabled in 2.6.29, we feel like it may explain some performance problems if the video memory isn't in write combining anymore. Brice -- To UNSUBSCRIBE, email to debian-bugs-dist-requ...@lists.debian.org with a subject of unsubscribe. Trouble? Contact listmas...@lists.debian.org
