Add UMA metrics for V-Sync smoothness and latency |
||
Issue descriptionThese metrics would be useful when testing impact of GPU V-sync on Windows (see Issue 467617). V-sync latency is time from V-sync timestamp to when Scheduler receives or start handling it (as OnBeginFrame). For V-sync smoothness for now we could look at one of existing BeginFrame UMA metrics. Ideally we'd want it to be expressed as something like coefficient of variance.
,
Mar 15 2017
The following revision refers to this bug: https://chromium.googlesource.com/chromium/src.git/+/2b0182bca127df14fa7f3c8becb8ff682b3f1bd6 commit 2b0182bca127df14fa7f3c8becb8ff682b3f1bd6 Author: stanisc <stanisc@chromium.org> Date: Wed Mar 15 20:31:43 2017 Add BeginFrameLatency UMA The purpose of this UMA is to estimate v-sync latency i.e. the time from hardware v-sync which is represented by BeginFrame timestamp to the time BeginFrame is handled in Scheduler. This metric will be useful to understand impact of GPU V-sync which has been recently landed behind a feature switch. BUG= 701181 CQ_INCLUDE_TRYBOTS=master.tryserver.blink:linux_trusty_blink_rel Review-Url: https://codereview.chromium.org/2752573002 Cr-Commit-Position: refs/heads/master@{#457190} [modify] https://crrev.com/2b0182bca127df14fa7f3c8becb8ff682b3f1bd6/cc/scheduler/compositor_timing_history.cc [modify] https://crrev.com/2b0182bca127df14fa7f3c8becb8ff682b3f1bd6/cc/scheduler/compositor_timing_history.h [modify] https://crrev.com/2b0182bca127df14fa7f3c8becb8ff682b3f1bd6/cc/scheduler/compositor_timing_history_unittest.cc [modify] https://crrev.com/2b0182bca127df14fa7f3c8becb8ff682b3f1bd6/cc/scheduler/scheduler.cc [modify] https://crrev.com/2b0182bca127df14fa7f3c8becb8ff682b3f1bd6/tools/metrics/histograms/histograms.xml
,
Mar 17 2017
The following revision refers to this bug: https://chromium.googlesource.com/chromium/src.git/+/0828f5b83f4835d9772dab5bd0bea6d8d9704737 commit 0828f5b83f4835d9772dab5bd0bea6d8d9704737 Author: stanisc <stanisc@chromium.org> Date: Fri Mar 17 03:38:38 2017 Reimplement vsync latency UMA to be based on BeginImplFrame rather than BeginMainFrame The previous fix that I submitted a couple of days ago mistakenly hooked to a wrong Scheduler method which resulted in recording latency for throttled BeginMainFrame call. BeginImplFrame is a much better method to hook to. I verified this locally by looking at chrome://histograms. The browser side histogram now looks way cleaner than in the first implementation. BUG= 701181 CQ_INCLUDE_TRYBOTS=master.tryserver.blink:linux_trusty_blink_rel Review-Url: https://codereview.chromium.org/2754943002 Cr-Commit-Position: refs/heads/master@{#457676} [modify] https://crrev.com/0828f5b83f4835d9772dab5bd0bea6d8d9704737/cc/scheduler/compositor_timing_history.cc [modify] https://crrev.com/0828f5b83f4835d9772dab5bd0bea6d8d9704737/cc/scheduler/compositor_timing_history.h [modify] https://crrev.com/0828f5b83f4835d9772dab5bd0bea6d8d9704737/cc/scheduler/compositor_timing_history_unittest.cc [modify] https://crrev.com/0828f5b83f4835d9772dab5bd0bea6d8d9704737/cc/scheduler/scheduler.cc [modify] https://crrev.com/0828f5b83f4835d9772dab5bd0bea6d8d9704737/cc/scheduler/scheduler.h [modify] https://crrev.com/0828f5b83f4835d9772dab5bd0bea6d8d9704737/tools/metrics/histograms/histograms.xml
,
Mar 17 2017
|
||
►
Sign in to add a comment |
||
Comment 1 by stanisc@chromium.org
, Mar 14 2017