All graphs for this bug:
  https://chromeperf.appspot.com/group_report?bug_id=767100

(For debugging:) Original alerts at time of bug-filing:
  https://chromeperf.appspot.com/group_report?sid=e43b0d75999c315ae97d27fb64c42704a95d21f2145b15361a269ba46eb1f11a


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

chromium-rel-win10

Started bisect job https://chromeperf.appspot.com/buildbucket_job_status/8967912656147143168

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

Hi fdoray@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 : Francois Doray
  Commit : a038ca797f70caabc2342e5da6b6e687ab0bfd96
  Date   : Sun Sep 17 03:26:06 2017
  Subject: Delete skipped tasks in the environment where the tasks would have run.

Bisect Details
  Configuration: winx64_10_perf_bisect
  Benchmark    : media_perftests
  Metric       : audio_bus_from_interleaved/int8_t
  Change       : 3.69% | 11.1137833333 -> 11.52415

Revision             Result                    N
chromium@502519      11.1138 +- 0.0411053      6      good
chromium@502521      11.1246 +- 0.0376883      6      good
chromium@502522      11.5242 +- 0.0467379      6      bad       <--

To Run This Test
  .\src\out\Release_x64\media_perftests.exe --single-process-tests

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

Debug information about this bisect:
  https://chromeperf.appspot.com/buildbucket_job_status/8967912656147143168


For feedback, file a bug with component Speed>Bisection

Started bisect job https://chromeperf.appspot.com/buildbucket_job_status/8967276174759678320

Re-running bisect. I verified with a CHECK(false) that the code modified by my CL doesn't run as part of this test...

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

Suspected Commit
  Author : Francois Doray
  Commit : a038ca797f70caabc2342e5da6b6e687ab0bfd96
  Date   : Sun Sep 17 03:26:06 2017
  Subject: Delete skipped tasks in the environment where the tasks would have run.

Bisect Details
  Configuration: winx64_10_perf_bisect
  Benchmark    : media_perftests
  Metric       : audio_bus_from_interleaved/int8_t
  Change       : 3.51% | 11.1318166667 -> 11.5225333333

Revision             Result                    N
chromium@502515      11.1318 +- 0.0152029      6      good
chromium@502520      11.127 +- 0.0199983       6      good
chromium@502521      11.1387 +- 0.0791093      6      good
chromium@502522      11.5232 +- 0.0528875      6      bad       <--
chromium@502523      11.5148 +- 0.0506165      6      bad
chromium@502525      11.5225 +- 0.0403535      6      bad

To Run This Test
  .\src\out\Release_x64\media_perftests.exe --single-process-tests

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

Debug information about this bisect:
  https://chromeperf.appspot.com/buildbucket_job_status/8967276174759678320


For feedback, file a bug with component Speed>Bisection

It's probably some optimization in the compiler that didn't work the same way because of your code. +brucedawson@, do you know why this could happen, and is there a way for us to configure the perf tests to not be affected by it?

It looks like that test is bi-modal for performance. Either the measurement technique is unstable or else something keeps tickling the optimizer to make it produce different code. I'm not sure what would do that.

Sorry, no ideas on how to avoid this.