Environment: ----------------------------------- Platform:BDW Libdrm: (master)libdrm-2.4.58-4-g00847fa48b83a85b0cb882594a12ed1511f780db Mesa: (master)cd745d46ce7ee9adc95c903670dd11cf3443e7a1 Xserver: (master)xorg-server-1.16.99.901-3-g63bb5c5ef16edf652179770294dcca4fc07dc992 Xf86_video_intel: (master)2.99.916-132-g0532a3313ad9c76a6e1d28e8a1c2ea495583fead Cairo: (master)adbeb3d53c6c6e8ddcc63988200da4c5c9627717 Libva: (master)ccd93de5a707e92a629cccd595757c8d436fa3cc Libva_intel_driver: (master)24cba20a119c96556ae4dc9a90043896ea70e567 Kernel: (drm-intel-nightly)0642a51748ff46a205cf2a3fc45ba18cc92c9bda Bug detailed description: --------------------------------------------- It's Xf86_video_intel regression GpuTest_v0.5_triangle_fullscreen show largest performance FPS reduce: -60% SynMark2_v6.0.0_OglBatch0: -33% SynMark2_v6.0.0_OglBatch1: -33% SynMark2_v6.0.0_OglBatch2: -32% SynMark2_v6.0.0_OglBatch3: -30% SynMark2_v6.0.0_OglBatch4: -34% From bisect result, below is first bad commit: commit 0532a3313ad9c76a6e1d28e8a1c2ea495583fead Author: Chris Wilson <chris@chris-wilson.co.uk (file://chris@chris-wilson.co.uk/)> AuthorDate: Wed Nov 5 20:11:54 2014 +0000 Commit: Chris Wilson <chris@chris-wilson.co.uk (file://chris@chris-wilson.co.uk/)> CommitDate: Wed Nov 5 21:06:57 2014 +0000 sna/gen8: Clear instancing enabled bit between batches gen8 sets the instancing bit relative to the vertex element, but we were clearing it for the vertex buffer. As the maximum number of vertex elements is fixed, just clear them all when emitting our header. Note that VF_SGVS is not sufficient by itself to disable all side-effects of instancing. Thanks to Kenneth Graunke for pointing out the change from vertex buffer to vertex element of the instancing enable bit. Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=84958 (https://bugs.freedesktop.org/show_bug.cgi?id=84958) Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk (file://chris@chris-wilson.co.uk/)> Below cases all will fail: Lightsmark v2008 Warsow v1.0 GpuTest_v0.5_triangle_fullscreen GpuTest_v0.5_triangle_Windowed_640*480 SynMark2_v6.0.0_OglBatch0 SynMark2_v6.0.0_OglBatch1 SynMark2_v6.0.0_OglBatch2 SynMark2_v6.0.0_OglBatch3 SynMark2_v6.0.0_OglBatch4 SynMark2_v6.0.0_OglBatch5 SynMark2_v6.0.0_OglGeomPoint SynMark2_v6.0.0_OglGeomTriList SynMark2_v6.0.0_OglGeomTriStrip SynMark2_v6.0.0_OglShMapVsm SynMark2_v6.0.0_OglTexFilterTri SynMark2_v6.0.0_OglTexFilterAniso SynMark2_v6.0.0_OglVSDiffuse1 SynMark2_v6.0.0_OglVSDiffuse8 SynMark2_v6.0.0_OglVSTangent Unigine-Valley v1.0 GLbenchmark v2.5.1 EgyptTestStandard Reproduce steps: --------------------------------------------- 1. xinit& 2. ./GpuTest_0.5 triangle_fullscreen log message will upload later.
Previous FPS values would have been skipping the copy operation and so these bandwidth tests would have been incorrect.
(In reply to Chris Wilson from comment #1) > Previous FPS values would have been skipping the copy operation and so these > bandwidth tests would have been incorrect. Chris, excuse me, I do not understand your comments, would you pls educate me more, why this is not bug?Thanks.
There was a bug in the ddx that was causing the GPU to not evaluate the blits for SwapBuffers(). This gave inflated results for swap/bandwidth bound benchmarks like Triangles. 0532a3313ad9c76a6e1d28e8a1c2ea495583fead fixes the copies so that they actually are performed by the GPU and corrects the benchmark results.
(In reply to Chris Wilson from comment #3) > There was a bug in the ddx that was causing the GPU to not evaluate the > blits for SwapBuffers(). This gave inflated results for swap/bandwidth bound > benchmarks like Triangles. 0532a3313ad9c76a6e1d28e8a1c2ea495583fead fixes > the copies so that they actually are performed by the GPU and corrects the > benchmark results. Thanks a lot Chris, so that means previous our performance values higher is wrong, right? We have many cases performance data decreased this time. Lightsmark v2008 Warsow v1.0 GpuTest_v0.5_triangle_fullscreen GpuTest_v0.5_triangle_Windowed_640*480 SynMark2_v6.0.0_OglBatch0 SynMark2_v6.0.0_OglBatch1 SynMark2_v6.0.0_OglBatch2 SynMark2_v6.0.0_OglBatch3 SynMark2_v6.0.0_OglBatch4 SynMark2_v6.0.0_OglBatch5 SynMark2_v6.0.0_OglGeomPoint SynMark2_v6.0.0_OglGeomTriList SynMark2_v6.0.0_OglGeomTriStrip SynMark2_v6.0.0_OglShMapVsm SynMark2_v6.0.0_OglTexFilterTri SynMark2_v6.0.0_OglTexFilterAniso SynMark2_v6.0.0_OglVSDiffuse1 SynMark2_v6.0.0_OglVSDiffuse8 SynMark2_v6.0.0_OglVSTangent Unigine-Valley v1.0 GLbenchmark v2.5.1 EgyptTestStandard
Right, it is going to affect almost everything that calls SwapBuffers or CopyRegion. It is more or less the same as employing a swap evasion technique for benchmarks. Hopefully, the game benchmarks are the least effected (or anything running at around refresh rate).
(In reply to Chris Wilson from comment #5) > Right, it is going to affect almost everything that calls SwapBuffers or > CopyRegion. It is more or less the same as employing a swap evasion > technique for benchmarks. > > Hopefully, the game benchmarks are the least effected (or anything running > at around refresh rate). Chris,Thanks for your kindly explanation!
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.