Video playback power consumption regression |
||||
Issue descriptionhttps://chromeperf.appspot.com/report?sid=a613e57a93bfefa1c695907ff156ab7b3432f044d038ee5b71e90464bd9d63bb&start_rev=35190031097200000&end_rev=35440001105100000 It's a regression across the board, i.e. the change can be seen with and without overlays, and when using VaAPI or Sw decode. https://screenshot.googleplex.com/jbwvabr4Py5 CrOs blame range: https://crosland.corp.google.com/log/11022.0.0..11026.0.0 Chromium blame range: https://chromium.googlesource.com/chromium/src/+log/70.0.3532.0..70.0.3538.0?pretty=fuller&n=10000 (Initially I thought it was my decode-on-client-pixmaps but that landed in 70.0.3529.0 crrev.com/c/1181464)
,
Sep 11
I've created Issue 882879 from the Perf console.
,
Sep 13
Perhaps, this is caused by crbug.com/879962? A CL for it was landed on Chrome 71.0.3551.0. crrev.com/c/1217584 The CL will enable blur on shelf if blur on launcher is enabled. At this moment, it is enabled only on eve for performance reason. Miguel, may I ask you to investigate this regression with Chrome ToT? Thanks!
,
Sep 19
We're back to normal numbers after : https://crosland.corp.google.com/log/11066.0.0..11067.0.0 https://chromium.googlesource.com/chromium/src/+log/71.0.3544.0..71.0.3552.0 If the original regression was caused by https://chromium-review.googlesource.com/c/chromium/src/+/1217584 then the fix could be https://chromium-review.googlesource.com/c/chromium/src/+/1224691 manucornet@ do you think your CLs can be causing these regression (and fix)? I don't know if the tests show the shelf or not. Maybe the changes introduced re the spacing rubs the play back the wrong way...?
,
Sep 19
Issue 882879 has been merged into this issue.
,
Sep 20
Hmm, it seems unlikely that the regression was caused by CL 1217584, I would say this would be more likely to be a fix because it disables blur on the shelf for a bunch of devices. Before that CL, blur was enabled everywhere, and this CL only enables it for devices where the app list background is also blurred. It also seems very unlikely that CL 1224691 would have any impact on performance as it's just tweaking slightly the padding around a few elements.
,
Sep 20
So what is the status on this right now? Comment #4 ("back to normal numbers") makes it sound like this is fixed. Is that correct? If not, what needs to be done here? Thanks!
,
Sep 20
#6,7: thanks for taking a look. Numbers are back to normal indeed, so no action is needed, beyond learning from what happened to avoid repeated regressions -- sometimes power consumption regressions are nonintuitive, because they might affect e.g. the number of sleep states engaged, or the transitions among them or what not... |
||||
►
Sign in to add a comment |
||||
Comment 1 by mcasas@chromium.org
, Sep 11