All graphs for this bug:
  https://chromeperf.appspot.com/group_report?bug_id=645429

Original alerts at time of bug-filing:
  https://chromeperf.appspot.com/group_report?keys=agxzfmNocm9tZXBlcmZyFAsSB0Fub21hbHkYgICgycytrwoM,agxzfmNocm9tZXBlcmZyFAsSB0Fub21hbHkYgICgyemXvQkM,agxzfmNocm9tZXBlcmZyFAsSB0Fub21hbHkYgICgicWRjQgM,agxzfmNocm9tZXBlcmZyFAsSB0Fub21hbHkYgICgyYCGowkM,agxzfmNocm9tZXBlcmZyFAsSB0Fub21hbHkYgICgicuMoAoM,agxzfmNocm9tZXBlcmZyFAsSB0Fub21hbHkYgICgifCG_goM,agxzfmNocm9tZXBlcmZyFAsSB0Fub21hbHkYgICg8YWk9AkM,agxzfmNocm9tZXBlcmZyFAsSB0Fub21hbHkYgICgyerCsAoM,agxzfmNocm9tZXBlcmZyFAsSB0Fub21hbHkYgICgiczc5QoM


Bot(s) for this bug's original alert(s):

webrtc-android-tests-nexus5
webrtc-android-tests-nexus72
webrtc-android-tests-nexus9

Hey perkj@,

Your change in https://codereview.webrtc.org/2255463002 seems to have led to some quite large improvements in the encode time on the Nexus perf bots.

I just created this bug to let you verify that these large improvements are expected? :)

It looks like this cl made quite an impact when the CPU is heavily loaded. 
As far as I can understand, when the cpu is less loaded (about 50%) encode time on the CIF VP encode, it has no effect and on Linux I can not see any change in numbers in any of the test cases. 
However, on Android, where the encode_usage is more than 100% it can have a huge impact.  

Åsa, would you mind taking a look and see if you come to the same conclusion?