All graphs for this bug:
  https://chromeperf.appspot.com/group_report?bug_id=726916

Original alerts at time of bug-filing:
  https://chromeperf.appspot.com/group_report?keys=agxzfmNocm9tZXBlcmZyFAsSB0Fub21hbHkYgIDg-vG6tQkM,agxzfmNocm9tZXBlcmZyFAsSB0Fub21hbHkYgIDg-t3q9AoM,agxzfmNocm9tZXBlcmZyFAsSB0Fub21hbHkYgIDg-r_OrgoM,agxzfmNocm9tZXBlcmZyFAsSB0Fub21hbHkYgIDg-vaQqwsM,agxzfmNocm9tZXBlcmZyFAsSB0Fub21hbHkYgIDgut-64AgM,agxzfmNocm9tZXBlcmZyFAsSB0Fub21hbHkYgIDg-r3mpwkM,agxzfmNocm9tZXBlcmZyFAsSB0Fub21hbHkYgIDgupOqlwgM


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

android-nexus5
android-nexus5X
android-webview-nexus5X
android-webview-nexus6

Started bisect job https://chromeperf.appspot.com/buildbucket_job_status/8978489353045118064

=== Auto-CCing suspected CL author drott@chromium.org ===

Hi drott@chromium.org, the bisect results pointed to your CL, please take a look at the
results.


=== BISECT JOB RESULTS ===
Perf regression found with culprit

Suspected Commit
  Author : drott
  Commit : ba0e6cd12bf6f5e864e1f60993ceafef3f4a7cae
  Date   : Wed May 24 09:08:34 2017
  Subject: Do not build FreeType with HarfBuzz support on Android

Bisect Details
  Configuration: android_webview_arm64_aosp_perf_bisect
  Benchmark    : system_health.memory_mobile
  Metric       : memory:webview:all_processes:reported_by_chrome:gpu:effective_size_avg/load_media/load_media_imgur
  Change       : 238.72% | 3495253.33333 -> 11923911.1111

Revision             Result                    N
chromium@474204      3495253 +- 9386645        9       good
chromium@474220      3411383 +- 17458029       14      good
chromium@474221      7898908 +- 7571622        9       bad       <--
chromium@474222      12124160 +- 11801076      6       bad
chromium@474224      10824363 +- 8339075       6       bad
chromium@474228      9106773 +- 5289948        6       bad
chromium@474236      11923911 +- 18296194      9       bad

Please refer to the following doc on diagnosing memory regressions:
  https://chromium.googlesource.com/chromium/src/+/master/docs/memory-infra/memory_benchmarks.md

To Run This Test
  src/tools/perf/run_benchmark -v --browser=android-webview --output-format=chartjson --upload-results --pageset-repeat=1 --also-run-disabled-tests --story-filter=load.media.imgur system_health.memory_mobile

Debug Info
  https://chromeperf.appspot.com/buildbucket_job_status/8978489353045118064

Is this bisect wrong?
  https://chromeperf.appspot.com/bad_bisect?try_job_id=5050253926989824


| O O | Visit http://www.chromium.org/developers/speed-infra/perf-bug-faq
|  X  | for more information addressing perf regression bugs. For feedback,
| / \ | file a bug with component Speed>Bisection.  Thank you!

 Issue 726923  has been merged into this issue.

 Issue 726925  has been merged into this issue.

 Issue 726924  has been merged into this issue.

The blamed CL is a partial Android only revert of compiling FreeType with HarfBuzz support Android, see  issue 724095 . I have a hard time understanding why this would cause a 238.72% GPU memory increase, unless we saw an equivalent drop after the original change: https://codereview.chromium.org/2880223002

Additionally, the bisect seems to be based on a highly volatile metric. 

Also, Mac-Retina results idle power benchmark results have been merged into this bug which I doubt have anything to do with the Android change.

brianderson@, may I assign this back to you for untangling this?

 Issue 727355  has been merged into this issue.

 Issue 727356  has been merged into this issue.

 Issue 727357  has been merged into this issue.

 Issue 727358  has been merged into this issue.

 Issue 727497  has been merged into this issue.

Re #7:

I split off the improvements in PLT into  bug 727727 .

Four separate bisects seem to confirm that your CL causes an increase in GPU memory on imgur in webview:
Comment #3
 Bug 726923   #3 
 Bug 726925   #3 
 Bug 726924   #3 

I know the graph is noisy, but that still seems like an issue. The graph does show a drop at the oritinal change. 

I think the mac-retina power benchmark results already got merged out? I don't see them in the list.

There are also non-gpu 

I went through the graphs and split off anything that doesn't improve at the CL in #7. All these seem to be from the blamed CL now, reassigning to drott.

Thank you for untangling these, very helpful. I moved out one more. What's left covered in this bug now I will mark as WontFix, since it is a revert to the original behavior. This is for as as long as we can't enable the HarfBuzz support in FreeType on Android due to performance concerns, compare  issue 724095 .