All graphs for this bug:
  https://chromeperf.appspot.com/group_report?bug_id=844549

(For debugging:) Original alerts at time of bug-filing:
  https://chromeperf.appspot.com/group_report?sid=7619028846e8ad5eb9b1bb577f48ce330899c5b216cb1a6ee82d2409aa1bc5ac


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

android-nexus5

📍 Pinpoint job started.
https://pinpoint-dot-chromeperf.appspot.com/job/14d75914240000

📍 Found significant differences after each of 18 commits.
https://pinpoint-dot-chromeperf.appspot.com/job/14d75914240000

[vr] Add runtime flag to enable tabs view by tiborg@chromium.org
https://chromium.googlesource.com/chromium/src/+/4cdea8db5c2f1067b3b26154c3e2abd6780e4466

Trace scheduler/base independently. by kraynov@chromium.org
https://chromium.googlesource.com/chromium/src/+/46290e3f2b9f12cdc16e1d5856be3a35836492bb

Audio-service-based media::AudioInputIPC, to make AudioInputDevice work with the audio service. by jonasolsson@chromium.org
https://chromium.googlesource.com/chromium/src/+/8c97dbc39ecc653c073cdeff47c77cffd1706157

[omnibox] Leverage the existing file URL parsing routine by krb@chromium.org
https://chromium.googlesource.com/chromium/src/+/13622d510f53ddbd7633727c6834a905d6105eef

update frame sink ids on BeginFrameProvider when GPU crashes by fserb@chromium.org
https://chromium.googlesource.com/chromium/src/+/290fa6c74fa7e70c9d3aabdf57cb8dedf8e09b88

Cleanup arraysize usage in //net/base by rsleevi@chromium.org
https://chromium.googlesource.com/chromium/src/+/4625214fae133516f909e6305c858feb02944f4a

Revert "Revert "Don't use GrRRectBlurEffect for large rects when highp is not full float"" by bsalomon@google.com
https://skia.googlesource.com/skia/+/ff6a7cd4d2da9a8c992a20da01cd439bfb008b5c

Add driver bug workarounds to GrCaps by enne@chromium.org
https://skia.googlesource.com/skia/+/328490c6a1625ce51d0e81688e0c85c79c400d86

Roll src/third_party/skia/ d31b2f674..2d46a0092 (9 commits) by skia-chromium-autoroll@skia-buildbots.google.com.iam.gserviceaccount.com
https://chromium.googlesource.com/chromium/src/+/035c0c8f116fe792a3052e44cefdd99409ee8283

PresentationCallback: Fix a TODO in viz::Display by penghuang@chromium.org
https://chromium.googlesource.com/chromium/src/+/7ef949f51ec608a970a51f86c568d52f3e5ec959

Add device monitoring to the audio service. by guidou@chromium.org
https://chromium.googlesource.com/chromium/src/+/d4af62ded717ec9fe2bb54d9d05439d4dfeff0b5

Revert "Roll src/third_party/skia/ d31b2f674..2d46a0092 (9 commits)" by robertphillips@google.com
https://chromium.googlesource.com/chromium/src/+/4b2722307920997a657fbd5472c10f3d27409b07

Remove Document#selectedStylesheetSet/preferredStylesheetSet by cnardi@chromium.org
https://chromium.googlesource.com/chromium/src/+/62e388ac4d084b73649d99df72e651559be44077

[oilpan] Enable incremental marking buildflag by mlippautz@chromium.org
https://chromium.googlesource.com/chromium/src/+/838fb3bd2db6a2869a0efc6e7e3154f954f74b70

[SPv175+] Combine clips in more cases by wangxianzhu@chromium.org
https://chromium.googlesource.com/chromium/src/+/7f6fc0f6b0d391350bf6c6745514ec452873f081

Revert "Reland "Composite ScrollTimeline scrollSource when appropriate."" by dmurph@chromium.org
https://chromium.googlesource.com/chromium/src/+/35bab68abc419e3242d5388c7292c3fc1a338099

Display Cutout: Add viewport-fit descriptor by beccahughes@chromium.org
https://chromium.googlesource.com/chromium/src/+/fda642f0f944120b331a5df95ae08ea2751776d7

Radically simplify Blink scrollbar painting. by jbroman@chromium.org
https://chromium.googlesource.com/chromium/src/+/fd620b69bd94904966229f927f9f408f28edf8f5

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

This hopefully shouldn't be due to my change -- all I did was remove code, which should've made less work be performed, if anything.

I'm pretty sure it's not my DedicatedWorker-only change.

I'm pretty sure it is not my change.

Based on the pinpoint graph, https://chromium.googlesource.com/chromium/src/+/838fb3bd2db6a2869a0efc6e7e3154f954f74b70 looks like the culprit.

Agree this is mine. It also fits the land/revert/reland pattern. Thanks for triaging.

We decided to go with the regressions on microbenchmarks right now as a Finch study is showing large improvements for real world. We still keep them open though as they are worth looking at for optimizing the write barrier.