All graphs for this bug:
  https://chromeperf.appspot.com/group_report?bug_id=839147

(For debugging:) Original alerts at time of bug-filing:
  https://chromeperf.appspot.com/group_report?sid=8cd4413b1562d0ad41722ccc31bdfb30d1dcc81e95bee27638f9097e39f14355


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

android-nexus5
chromium-rel-win10

📍 Pinpoint job started.
https://pinpoint-dot-chromeperf.appspot.com/job/14c37fcbc40000

📍 Found significant differences after each of 2 commits.
https://pinpoint-dot-chromeperf.appspot.com/job/14c37fcbc40000

D3D: Don't test for device loss as frequenty in tight loops. by geofflang@chromium.org
https://chromium.googlesource.com/angle/angle/+/5528d11fe809ed95505995322ba3c4082d483fa7

Roll src/third_party/angle/ f1bcd017c..696434382 (2 commits) by angle-chromium-autoroll@skia-buildbots.google.com.iam.gserviceaccount.com
https://chromium.googlesource.com/chromium/src/+/d02a65d20a30dba57c18c297b3ffe1320e2f5eb6

Understanding performance regressions:
  http://g.co/ChromePerformanceRegressions

geofflang, can you take a look? Looking at the bisect, there's a pretty clear jump at "D3D: Don't test for device loss as frequenty in tight loops."

Also removing the improvement on Nexus 5 from this bug because I think it got associated incorrectly, this is a windows regression.

Fixed as of https://chromium-review.googlesource.com/c/chromium/src/+/1065232