You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
{{ message }}
Repository navigation
Regression in average_cpu_usage mac_ios_post_backdrop_filter_perf_ios__timeline_summary #83113
It started around #83026, but I can't see how a plugins roll could have affected this benchmark. I'm going to start with giving it to the engine sheriff this week to bisect, but including the plugins team as well for triage.
Aha. I suspect it's not the plugin, but the engine; seeing similar regressions in average_cpu_usage mac_ios_new_gallery_ios__transition_perf around #83025.
Yeah, that sounds plausible. We should be okay with regression given that we are getting better timelines, right?
Until this PR we effectively could not tell how much actual CPU vs Wall time was spent on particular timeline event on iOS
@aam 15% CPU to query CPU time seems like a lot. Do you know if this is also what we see on other platforms?
Either way we should also verify locally that it is that change. If @aam doesn't have a setup for that we can identify someone at the issue triage meeting on Monday.
We may need to accept this regression in order for the timelines from iOS to include this information. Dropping to P2 for @aam to confirm and compare overhead with Android and/or Linux.
Yes, it looks like we don't track average CPU utilization on Android like we do on iOS. I think we'll have to chalk this up as the cost of the timeline in profile mode on iOS.
This thread has been automatically locked since there has not been any recent activity after it was closed. If you are still experiencing a similar issue, please open a new bug, including the output of flutter doctor -v and a minimal reproduction of the issue.
Activity
kf6gpe commentedon May 21, 2021
Aha. I suspect it's not the plugin, but the engine; seeing similar regressions in average_cpu_usage mac_ios_new_gallery_ios__transition_perf around #83025.
See
https://flutter-flutter-perf.skia.org/e/?begin=1621494841&end=1621555092&keys=X91281eb0cabcec0938c76e6e8a17e9e4&num_commits=50&request_type=1&xbaroffset=24192
zanderso commentedon May 21, 2021
Maybe https://dart.googlesource.com/sdk.git/+/626b48090a3b6fdfa429666dc0e7646d1bde3a3b%5E%21/#F0 ? /cc @aam
aam commentedon May 21, 2021
Yeah, that sounds plausible. We should be okay with regression given that we are getting better timelines, right?
Until this PR we effectively could not tell how much actual CPU vs Wall time was spent on particular timeline event on iOS
zanderso commentedon May 22, 2021
@aam 15% CPU to query CPU time seems like a lot. Do you know if this is also what we see on other platforms?
Either way we should also verify locally that it is that change. If @aam doesn't have a setup for that we can identify someone at the issue triage meeting on Monday.
jason-simmons commentedon May 22, 2021
Confirmed that reverting https://dart.googlesource.com/sdk.git/+/626b48090a3b6fdfa429666dc0e7646d1bde3a3b%5E%21/#F0 fixes the difference in
average_cpu_usageon themac_ios_simple_animation_perf_iosbenchmarkzanderso commentedon May 24, 2021
We may need to accept this regression in order for the timelines from iOS to include this information. Dropping to P2 for @aam to confirm and compare overhead with Android and/or Linux.
aam commentedon May 24, 2021
hmm, for some reason
average_cpu_usagemetric is not being extracted on android benchmarksaam commentedon May 25, 2021
It looks like cpu_usage metric is very ios-specific https://github.com/flutter/engine/blob/master/shell/platform/darwin/ios/framework/Source/profiler_metrics_ios.mm#L173. I think it's hard to make apples-to-apples comparison between ios and android for this kind of calculation.
zanderso commentedon May 25, 2021
Yes, it looks like we don't track average CPU utilization on Android like we do on iOS. I think we'll have to chalk this up as the cost of the timeline in profile mode on iOS.
github-actions commentedon Aug 1, 2021
This thread has been automatically locked since there has not been any recent activity after it was closed. If you are still experiencing a similar issue, please open a new bug, including the output of
flutter doctor -vand a minimal reproduction of the issue.