All graphs for this bug:
  https://chromeperf.appspot.com/group_report?bug_id=821796

(For debugging:) Original alerts at time of bug-filing:
  https://chromeperf.appspot.com/group_report?sid=9025bc991e66741dc4ef8e9192659fac92c606907671c712c74da24a426a5345


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

chromium-rel-win7-gpu-nvidia

📍 Pinpoint job started.
https://pinpoint-dot-chromeperf.appspot.com/job/13916916440000

📍 Couldn't reproduce a difference.
https://pinpoint-dot-chromeperf.appspot.com/job/13916916440000

📍 Pinpoint job started.
https://pinpoint-dot-chromeperf.appspot.com/job/13d1f5a3440000

📍 Found significant differences after each of 4 commits.
https://pinpoint-dot-chromeperf.appspot.com/job/13d1f5a3440000

Adds user setting to allow speech to continue after text goes away. by katie@chromium.org
https://chromium.googlesource.com/chromium/src/+/80d1057bdfc7b78cc1344c71b26b27835f21aca6

make SkJumper stages normal Skia code by mtklein@chromium.org
https://skia.googlesource.com/skia/+/22e536e3a1a09405d1c0e6f071717a726d86e8d4

QUIC connections will always have at least one retransmittable packet on the wire if by wangyix@chromium.org
https://chromium.googlesource.com/chromium/src/+/32e4b32b7afcac36791ae29af2d5dc7416fba19c

Watch chromeos lkgm changes. by achuith@chromium.org
https://chromium.googlesource.com/chromium/src/+/434e2459561fc31840ebb4517e3b5479c4ae3557

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

mtklein: There are a lot of small performance changes seen in the bisect above, but it looks like there is a 1ms jump at "make SkJumper stages normal Skia code". Can you take a look?

Yes, that's very likely that change.  It's somewhat expected, and we thought we'd be able to correct for it after the fact, but I think we're blocked on a bug in LLVM.

https://bugs.llvm.org/show_bug.cgi?id=36806