All graphs for this bug:
  https://chromeperf.appspot.com/group_report?bug_id=792458

(For debugging:) Original alerts at time of bug-filing:
  https://chromeperf.appspot.com/group_report?sid=97bdc0ccf668125204a7d2aca567047e893cbae9e7bcbcc2ae9c6284416d7300


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

chromium-rel-mac11

📍 Pinpoint job started.
https://pinpoint-dot-chromeperf.appspot.com/job/14ed3114040000

📍 Found significant differences after each of 2 commits.
https://pinpoint-dot-chromeperf.appspot.com/job/14ed3114040000

[Extension] Implement ExtensionSystem::InstallUpdate.
By mxnguyen@chromium.org · Mon Dec 04 22:02:41 2017
chromium @ 30975340db017def1217499e2ad5ff90c55d286e

[PE] Turn on half-width tiles for all platforms and devices.
By chrishtr@chromium.org · Tue Dec 05 01:51:58 2017
chromium @ a3e4477c29d087d3708dbcb4dc5a90985c2b45e3

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

Looking at the job output, I think the identification of Minh's change is a false positive. It looks like the half-width tiles is the issue.

The following revision refers to this bug:
  https://chromium.googlesource.com/chromium/src.git/+/2698e4b83adaa801bec420545d483e8ce76b9fce

commit 2698e4b83adaa801bec420545d483e8ce76b9fce
Author: Eric Karl <ericrk@chromium.org>
Date: Thu Dec 21 03:30:02 2017

Round half-width tiles and correct order

The GPU half-width tile optimization didn't round up, so we could end
up needing to draw three half-width tiles in place of one full-width
one, leading to a regression.

This logic should also be above the border-texel padding logic, moved.

Bug:  792458 
Cq-Include-Trybots: master.tryserver.blink:linux_trusty_blink_rel;master.tryserver.chromium.android:android_optional_gpu_tests_rel
Change-Id: Idc4f2e101d52b2ff1c0c65571506704a4238d7b3
Reviewed-on: https://chromium-review.googlesource.com/837621
Commit-Queue: Eric Karl <ericrk@chromium.org>
Reviewed-by: vmpstr <vmpstr@chromium.org>
Cr-Commit-Position: refs/heads/master@{#525580}
[modify] https://crrev.com/2698e4b83adaa801bec420545d483e8ce76b9fce/cc/layers/picture_layer_impl.cc
[modify] https://crrev.com/2698e4b83adaa801bec420545d483e8ce76b9fce/cc/layers/picture_layer_impl_unittest.cc

Due to other changes, this is no longer reproducible on the original mac-11 bot, the only bot on which this was seen. I believe my change should have fixed things, but can't verify. Duping this in to the larger perf regression bug.