All graphs for this bug:
  https://chromeperf.appspot.com/group_report?bug_id=747873

(For debugging:) Original alerts at time of bug-filing:
  https://chromeperf.appspot.com/group_report?sid=47e02ab3c910ee3c58e2664f1bae34c533f0b2d333f732b2a31b2169226db636


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

android-webview-nexus6

Started bisect job https://chromeperf.appspot.com/buildbucket_job_status/8973192964674459136

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

Hi fmalita@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 : Florin Malita
  Commit : c785be2815a779a528834a06e2e4d00d9fe3dfce
  Date   : Thu Jul 20 22:06:51 2017
  Subject: Enable Skia rasterpipeline for tiling

Bisect Details
  Configuration: android_webview_nexus6_aosp_perf_bisect
  Benchmark    : thread_times.key_mobile_sites_smooth
  Metric       : thread_raster_cpu_time_per_frame/thread_raster_cpu_time_per_frame
  Change       : 62.31% | 2.33531714471 -> 3.79041112161

Revision             Result                    N
chromium@488276      2.33532 +- 0.0731864      6      good
chromium@488421      2.33681 +- 0.0723653      6      good
chromium@488430      2.35032 +- 0.0602549      6      good
chromium@488431      3.80275 +- 0.241872       6      bad       <--
chromium@488432      3.79793 +- 0.155502       6      bad
chromium@488433      3.86256 +- 0.157111       6      bad
chromium@488435      3.83577 +- 0.244423       6      bad
chromium@488440      3.90928 +- 0.298515       6      bad
chromium@488458      3.82323 +- 0.18966        6      bad
chromium@488494      3.83478 +- 0.281406       6      bad
chromium@488566      3.79041 +- 0.115792       6      bad

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 thread_times.key_mobile_sites_smooth

More information on addressing performance regressions:
  http://g.co/ChromePerformanceRegressions

Debug information about this bisect:
  https://chromeperf.appspot.com/buildbucket_job_status/8973192964674459136


For feedback, file a bug with component Speed>Bisection

We expected some regressions for sites which use asymmetric tiling (uncommon).  Punting to mtklein@ to see if there's anything we can do for Android.

Off the top of my head I don't have anything in particular in reserve to make image sampling on ARM devices run faster.  Did removing SK_SUPPORT_LEGACY_TILED_BITMAPS let us delete any code?  If not, maybe we should just put it all back?

Aren't these devices rasterizing with Ganesh?  Do we really care about regressing software rendering performance?

reed@ prolly knows how much code hangs off SK_SUPPORT_LEGACY_TILED_BITMAPS.

Good question about these Android devices hitting the software rasterizer though.  aelias@ may know: aren't these perf tests supposed to run with Ganesh on Android?

We still ship software raster on blacklisted drivers and on all desktop sites (desktop sites should be close to getting flipped to Ganesh once GPU scheduling is good enough), so we do care to some degree, but we're willing to let it modestly regress over time.  This seems modest enough and mostly site-specific ("cuteoverload.com" main culprit).

 Issue 747874  has been merged into this issue.

 Issue 751158  has been merged into this issue.

FWIW, we did eventually think of something to speed this back up a little, and many of the regressions went away (or softened) when we landed it as https://skia-review.googlesource.com/c/27701.

 Issue 752420  has been merged into this issue.