Created attachment 76367 [details] transform_feedback_misc.test.log System Environment: -------------------------- Platform: IVB/HSW Libdrm: (master)libdrm-2.4.42-6-gbbf6e3dea3c79ea8e0c3e1dd8f80014201e003fc Mesa: (master)b21a9d46e4be0c666327569f07b9cddeb4d42d38 Xserver: (master)xorg-server-1.14.0-1-g5047810a4c20fab444b8c6eb146c55dcdb0d4219 Xf86_video_intel:(master)2.21.3-90-g93ecd5606eaf2e1db2 Cairo: (master)5ee136b2c067c92e85f04ba0c84966f7760f4556 Kernel: (drm-intel-nightly) 799c9b8dff5dc437f9a137c007ee2e1ecbb5e205 Bug detailed description: ------------------------- Ogles30/transform_feedback_misc.test fail with gnome-session enable compositing. The case is unsteadily and could repoduce by 5 times. It works well with gome-session with “fallback” mode(System tools/system settings/system info/Graphic), also well without window manager. Please see failure log and Xorg.0.log attached. Other failure cases: GL3Tests/sync/sync_functionality_clientwaitsync_flush.test GL3Tests/transform_feedback/transform_feedback_misc.test GL3Tests/sync/sync_coverage_getsynciv_defaults.test Reproduce steps: ------------------------- 1.xinit& 2.gnome-session enable compositing 3. ./GTF -width=64 -height=64 -run=GL3Tests/transform_feedback/transform_feedback_misc.test
Created attachment 76369 [details] Xorg.0.log
Ian, can you reproduce?
I would guess that the culprit is "i965: Actually use the user timeout in glClientWaitSync." but that's more recent than the sha meng tested. So I guess it must not be that (though I haven't tried to reproduce it yet).
These are the only ogles3conform failures on HSW. Can we take this as high priority?
I don't know that we've been able to reproduce this. Perhaps it only occurs on some HSW variants?
It also happens with raw X now. GL3Tests/sync/sync_functionality_clientwaitsync_flush.test unstable is easier to reproduce. ./GTF -width=64 -height=64 -run=GL3Tests/sync/sync_functionality_clientwaitsync_flush.test
Created attachment 83759 [details] sync_functionality_clientwaitsync_flush.test output
The failures with sync_functionality_clientwaitsync_flush.test are a known issue with the test itself. The problem is the test schedules too little work for the GPU to do before checking that the sync has not been signaled. At 64x64, a fast GPU always gets done before it can check. We don't care about those failures. Let's keep this bug about the transform_feedback_misc.test issue.
Reproduced the failure on my Ivybridge laptop. Test passes most of the times when run with -minfmt option. Even though symptoms of this bug and bug#67934 look similar, bisected commit(a3f48d9) in 67934 doesn't effect the output of this test in any way.
I noticed one transform feedback related workaround we weren't explicitly implementing, but it didn't help. It's on misc-workarounds of my tree.
In the updated gles3 conformance tests, the issue can not be reproduced.
The issue is also fixed in old gles3 version, so 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.