All graphs for this bug:
  https://chromeperf.appspot.com/group_report?bug_id=700853

Original alerts at time of bug-filing:
  https://chromeperf.appspot.com/group_report?keys=agxzfmNocm9tZXBlcmZyFAsSB0Fub21hbHkYgIDgtMaKswoM


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

chromium-rel-mac-retina

Started bisect job https://chromeperf.appspot.com/buildbucket_job_status/8985238962234364864

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

Suspected Commit
  Author : Michael Hablich
  Commit : 7c3936c67a6361371d9a00d83e2c03ecc603ce76
  Date   : Thu Mar 09 08:05:42 2017
  Subject: Version 5.9.34.1 (Turn on I+TF)

Bisect Details
  Configuration: mac_retina_perf_bisect
  Benchmark    : blink_perf.shadow_dom
  Metric       : v1-slot-append/v1-slot-append
  Change       : 10.28% | 5231.29406211 -> 4693.47516699

Revision                           Result                  N
chromium@455719                    5231.29 +- 204.086      6      good
chromium@455719,v8@7c3936c67a      4484.07 +- 189.209      6      bad       <--
chromium@455719,v8@fbffc377e3      4545.83 +- 115.663      6      bad
chromium@455720                    4456.63 +- 268.699      6      bad
chromium@455721                    4526.42 +- 109.152      6      bad
chromium@455723                    4502.58 +- 199.341      6      bad
chromium@455726                    4519.57 +- 137.854      6      bad
chromium@455732                    4554.08 +- 179.615      6      bad
chromium@455744                    4481.26 +- 192.014      6      bad
chromium@455768                    4493.4 +- 161.221       6      bad
chromium@455816                    4693.48 +- 253.091      6      bad

To Run This Test
  src/tools/perf/run_benchmark -v --browser=release --output-format=chartjson --upload-results --pageset-repeat=1 --also-run-disabled-tests blink_perf.shadow_dom

Debug Info
  https://chromeperf.appspot.com/buildbucket_job_status/8985238962234364864

Is this bisect wrong?
  https://chromeperf.appspot.com/bad_bisect?try_job_id=6464502490464256


| 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!

Click the link for "Original alerts at time of bug-filing:" to get the original alert. Seems like "v1-slot-append" regresses over all platforms.

Is this the same regression that makes Polymer 2.0 slightly slower?

SmallDistributionWithLayout also got a higher variance around that time. The ref build is still ok, so I suspect that it is not a bot problem.

mvstanton@ is going to further triage this.

With --turbo on there is a function that gets optimized twice by TurboFan, once normally, and then again with TurboFan OSR:

[compiling method 0x1ddf214db5b1 <JSFunction GCController.collectAll (sfi = 0x1ddf214dac69)> using TurboFan]
[optimizing 0x1ddf214db5b1 <JSFunction GCController.collectAll (sfi = 0x1ddf214dac69)> - took 0.161, 0.627, 0.065 ms]
[completed optimizing 0x1ddf214db5b1 <JSFunction GCController.collectAll (sfi = 0x1ddf214dac69)>]
[compiling method 0x1ddf214db5b1 <JSFunction GCController.collectAll (sfi = 0x1ddf214dac69)> using TurboFan OSR]
[optimizing 0x1ddf214db5b1 <JSFunction GCController.collectAll (sfi = 0x1ddf214dac69)> - took 0.134, 0.448, 0.044 ms]

Aside from that, nothing about the run looks different. There are no deoptimizations taking place in either pipeline configuration.

So I don't see anything unusual to explain the degrade yet...

Running with profiler ticks, in the Crankshaft case we spend more time in optimized code (70 ticks in method run(), verses 44 with --turbo). Otherwise everything is the same. Only 1-2% of time is in JavaScript for this test, the majority of the time is spent in GC and api calls (~4600 ticks in Builtin_HandleApiCall).

I'd attribute this result to our known issue that we are running more in the interpreter than in optimized code across several scenarios (a google maps rendering test we are looking at, typescript). In a small way, the same thing is showing up here.


WITHOUT IGNITION+TURBOFAN
 [Summary]:
   ticks  total  nonlib   name
    149    1.7%    1.7%  JavaScript
   8711   97.7%   97.9%  C++
   1029   11.5%   11.6%  GC
     20    0.2%          Shared libraries
     34    0.4%          Unaccounted

WITH
 [Summary]:
   ticks  total  nonlib   name
    103    1.1%    1.1%  JavaScript
   8837   98.3%   98.5%  C++
    986   11.0%   11.0%  GC
     18    0.2%          Shared libraries
     33    0.4%          Unaccounted

Blocking to show the relationship but de-priotizing. This is not important for the M59 stable launch.

No movement in about 6 months, should we WontFix?

WontFix-ing after no update from #15, please reopen if you're still planning on fixing this.