System Environment: -------------------------- Platform: IVB Libdrm: (master)libdrm-2.4.59-32-g080b4929b7452dc1fea32ac1d32e7e571e7fb38b Mesa: (master)b77576edc1a8010e5457f82b41c335ae27cb066b Xserver: (master)xorg-server-1.17.0-16-g3a06faf3fcdb7451125a46181f9152e8e59e9770 Xf86_video_intel: (master)2.99.917-174-g3e390ec4110519746f2edbb3c38a40f0fc415430 Libva: (master)f9741725839ea144e9a6a1827f74503ee39946c3 Libva_intel_driver:(master)e8fde1cdaafb93c2b54d6092a728d099ad7cdd11 Kernel: (drm-intel-nightly)a5217f77503a1089aeb8a9f4e3731e29c1ac2d41 Bug detailed description: --------------------------- It sporadically fails on IVB with mesa master branch, fail rate:1/10 output: libGL: OpenDriver: trying /opt/X11R7/lib/dri/tls/i965_dri.so libGL: OpenDriver: trying /opt/X11R7/lib/dri/i965_dri.so Wallclock time between MSCs 16716.000000us does not match glXGetMscRateOML 16665.600000us PIGLIT: {"result": "fail" } ==Reproduce steps== ---------------------------- 1. xinit 2. ./bin/glx-oml-sync-control-timing -fullscreen -divisor 2 -auto
It also happens on 10.5 branch.
Test on the latest mesa master branch 10 cycles, it always fails. libGL: OpenDriver: trying /opt/X11R7/lib/dri/tls/i965_dri.so libGL: OpenDriver: trying /opt/X11R7/lib/dri/i965_dri.so Wallclock time between MSCs 16766.666667us does not match glXGetMscRateOML 16699.159664us PIGLIT: {"result": "fail" }
This seems to has improved. Using Mesa 13.1.0-devel (git-9f150ff) in IVB, almost all times the test pass without problems. From time to time I get a warn instead, "iteration 0 woke up 2 MSCs later than expected"
I made an investigation and came to the conclusion that the behavior of this test is based on such characteristics of the monitor as resolution and frequency, and on the window manager(different behavior on gnome and unity ).With resolution 1600x900 this test pass 88/100 times and with resolution 1920x1080 this test pass 11/100 times. Analyzing how the GLX_OML_sync_control extension implements in mesa, you can see that there are no calculation based on the influencing characteristics. That means this is not a Mesa bug. This bug doesn't depend on gpu intel gpu architecture, because the test behavior the same on the KBL, IVB and HSW. You can have different behavior, if you change the characteristics of the display, or install another window manager. I think it's necessary to make at least one change in this test. I can’t quite understand where this constant(50) was taken. timing.c:311 if (fabs (expected_msc_wallclock_duration - msc_wallclock_duration_stats.mean)> 50) I think we can increase it to 200, after that test will be working more stable.
-- GitLab Migration Automatic Message -- This bug has been migrated to freedesktop.org's GitLab instance and has been closed from further activity. You can subscribe and participate further through the new bug through this link to our GitLab instance: https://gitlab.freedesktop.org/mesa/mesa/issues/1475.
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.