System Environment: ---------------------------------------------------------------------- Platform: IVB/HSW desktop (Ubuntu13.10 v.s Fedora19) Libdrm:(master)libdrm-2.4.49-2-gc3d96897de647bd5f6d4802c1 Mesa:(master)6b2b4cc8857a9163055c4e9c8007d53a9e668e75 Xf86_video_intel:(master)2.99.906-51-g80d1dee70ba3bdf38 Cairo:(master)31eff5c6eb57ad379689748fd8c60a5ffe0ba481 Libva:(staging)9aa8530ce272dc844428b7273afe7d4f0aaf391d Libva_intel_driver:(staging)34627c96f331f7a344270c Kernel:(drm-intel-nightly) 164a4cb4c1431a0689f855 Bug detailed description: ------------------------------------------------------------------------ Ubuntu 13.10 v.s. Fedora19, cairo-trace/ocitysmap.trace reduced by 60% on IVB/HSW with the same driver. But cairo-traces/xfce4-terminal-a1.trace improved by ~90% on IVB/HSW. Ubuntu 13.10 v.s. Fedora 19, there are some subcases also reduced by ~20% only on HSW. gnome-system-monitor.trace -26.39% gnome-terminal-vim.trace -24.51% Reproduce steps: -------------------------------------------------------------------- 1. xinit& 2.vblank_mode=0 CAIRO_TEST_TARGET=xlib perf/cairo-perf-trace -i 12 /home/cairo-traces/benchmark/ocitysmap.trace
Just to be clear here, what is the same and what is different between the two systems?
Ville can you please point at a branch with your fbc patches, presuming your tracking improvements should help here?
(In reply to comment #1) > Just to be clear here, what is the same and what is different between the > two systems? Test on HSW-D/IVB-D with same things: - the same driver: Kernel, X11R7(Mesa,cairo,xf86) - enabling SNA - Raw X(xinit) Only different things is OS: Ubuntu 13.10 and Fedora 19
Can you please attach Xorg.0.log from both to see if that shows any defaults that different between the two platforms?
Created attachment 90077 [details] HSW_Fedora19_Xorg.0.log
Created attachment 90078 [details] HSW_Ubuntu13.10_Xorg.0.log
Created attachment 90079 [details] IVB_Fedora19_Xorg.0.log
Created attachment 90080 [details] \IVB_Ubuntu13.10_Xorg.0.log
The IVB machine is broken. The fedora install has a broken libglx.so in the server that causes X to die in a fireball, and the ubuntu install is using the wrong kernel and has an old intel_drv.so which prevents X from starting. That needs to be resolved. The two HSW logs are from a Ubuntu system using the wrong kernel, and again have the invalid intel_drv.so preventing X from starting.
(In reply to comment #9) sorry,last time to the wrong log,update it now.
Created attachment 90130 [details] HSW_Fedora19_Xorg.0_new.log
Created attachment 90131 [details] HSW_Ubuntu13.10_Xorg.0_new.log
Created attachment 90132 [details] IVB_Fedora19_Xorg.0_new.log
Created attachment 90133 [details] IVB_Ubuntu13.10_Xorg.0_new .log
There's one last oddity present on all machines: GLX is being routed through swrast. Now that intel appears to be loaded on all setups, can you please confirm the original bug.
(In reply to comment #15) Retest with same driver on HSW,the results are same as last time: Ubuntu:0.638s Fedora:0.225s
What's the variance?
(In reply to comment #17) The results did not variance.
I see no reason to explain the discrepancy so keep digging. Perhaps a perf profile of the two might indicate the issue if it is CPU bound.
After updated system libs(by apt-get) on our mahcine (ubuntu 13.10), ocitysmap.trace improved to Fedora's result. We are trying to find out which system lib that cause the improvement.
If we haven't found out the relevant package by now, likely we won't until next time we succumb.
Closed it.
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.