All graphs for this bug:
  https://chromeperf.appspot.com/group_report?bug_id=648997

Original alerts at time of bug-filing:
  https://chromeperf.appspot.com/group_report?keys=agxzfmNocm9tZXBlcmZyFAsSB0Fub21hbHkYgICg-e3OrQkM,agxzfmNocm9tZXBlcmZyFAsSB0Fub21hbHkYgICg-diTsgsM,agxzfmNocm9tZXBlcmZyFAsSB0Fub21hbHkYgICg-cO74AsM,agxzfmNocm9tZXBlcmZyFAsSB0Fub21hbHkYgICghZS6uwsM


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

webrtc-android-tests-nexus72
webrtc-win-large-tests

The cl (https://codereview.webrtc.org/2347863002/) replaces some calls to ConvertToI420 to I420Copy, on the path reading YUV-frames from file. Which *might* be more efficient. But ConvertToI420 should boil down to I420Copy after a few levels of function calls, so 5% to 8% improvement is unexpectedly much.

Actually, the copying involved here is useless, I'm considering another cl to read the data with fread directly to the right location.

Closing, since it is an improvement.