Issue metadata
Sign in to add a comment
|
31.7%-78.9% regression in smoothness.tough_path_rendering_cases at 485333:485452 |
||||||||||||||||||||
Issue descriptionSee the link to graphs below.
,
Jul 12 2017
Started bisect job https://chromeperf.appspot.com/buildbucket_job_status/8974256468033148240
,
Jul 12 2017
=== Auto-CCing suspected CL author junov@chromium.org === Hi junov@chromium.org, the bisect results pointed to your CL, please take a look at the results. === BISECT JOB RESULTS === Perf regression found with culprit Suspected Commit Author : Justin Novosad Commit : 3750a7881ce851e03ce6177dca30307eb4a7f5c6 Date : Mon Jul 10 20:50:59 2017 Subject: Disabling display list 2D canvas by default Bisect Details Configuration: winx64intel_perf_bisect Benchmark : smoothness.gpu_rasterization.tough_path_rendering_cases Metric : frame_times/MotionMark Canvas Stroke Shapes Change : 77.96% | 16.795869452 -> 29.8892924365 Revision Result N chromium@485332 16.7959 +- 0.30306 6 good chromium@485362 16.8577 +- 0.254555 6 good chromium@485370 16.9437 +- 0.462336 6 good chromium@485371 29.7545 +- 0.321784 6 bad <-- chromium@485372 29.9967 +- 1.01612 6 bad chromium@485374 30.0998 +- 0.259233 6 bad chromium@485377 30.3004 +- 0.745431 6 bad chromium@485392 29.9505 +- 0.585215 6 bad chromium@485452 29.8893 +- 0.26637 6 bad To Run This Test src/tools/perf/run_benchmark -v --browser=release_x64 --output-format=chartjson --upload-results --pageset-repeat=1 --also-run-disabled-tests --story-filter=MotionMark.Canvas.Stroke.Shapes smoothness.gpu_rasterization.tough_path_rendering_cases More information on addressing performance regressions: http://g.co/ChromePerformanceRegressions Debug information about this bisect: https://chromeperf.appspot.com/buildbucket_job_status/8974256468033148240 For feedback, file a bug with component Speed>Bisection
,
Jul 12 2017
Issue 741726 has been merged into this issue.
,
Jul 13 2017
,
Jul 17 2017
,
Jul 17 2017
This seems a bit odd. Display list 2D canvas should have only kicked in for canvases that were software rendered, but the "Stroke Shapes" uses a single large canvas which should always be GPU accelerated and never trigger display list 2D canvas in the first place. At any rate, if it was rasterized in software, it does mean we lose the parallelism we gained from rendering into multiple tiles on separate threads, so it's logical that it should run slower with display list 2D canvas disabled.
,
Jul 20 2017
,
Jul 22 2017
Issue 741705 has been merged into this issue. |
|||||||||||||||||||||
►
Sign in to add a comment |
|||||||||||||||||||||
Comment 1 by 42576172...@developer.gserviceaccount.com
, Jul 12 2017