Started bisect job https://chromeperf.appspot.com/buildbucket_job_status/8970870024613860416

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

Suspected Commit
  Author : Christopher Cameron
  Commit : 1894d423068be735560e0171922c833c4091b5d1
  Date   : Wed Jun 28 06:31:03 2017
  Subject: color: Enable color correct rendering by default

Bisect Details
  Configuration: android_nexus5_perf_bisect
  Benchmark    : thread_times.tough_compositor_cases
  Metric       : thread_raster_cpu_time_per_frame/http___jsbin.com_covoqi_1_quiet?NEW_TILINGS
  Change       : 6.89% | 0.514406848365 -> 0.549860826317

Revision             Result                      N
chromium@482861      0.514407 +- 0.014739        6      good
chromium@482891      0.525458 +- 0.0106535       6      good
chromium@482899      0.524574 +- 0.0201665       6      good
chromium@482903      0.527782 +- 0.0181151       6      good
chromium@482904      0.556453 +- 0.00786616      6      bad       <--
chromium@482905      0.544825 +- 0.00796991      6      bad
chromium@482906      0.556271 +- 0.00624016      6      bad
chromium@482921      0.549861 +- 0.00834412      6      bad

To Run This Test
  src/tools/perf/run_benchmark -v --browser=android-chromium --output-format=chartjson --upload-results --pageset-repeat=1 --also-run-disabled-tests --story-filter=http...jsbin.com.covoqi.1.quiet.NEW.TILINGS thread_times.tough_compositor_cases

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

Debug information about this bisect:
  https://chromeperf.appspot.com/buildbucket_job_status/8970870024613860416


For feedback, file a bug with component Speed>Bisection

Hey Chris, I know this went in a while ago and we closed other related alerts.  I've been making a pass through all GPU thread time regressions that seem to have not recovered.

Here are two cases, the first (http://output.jsbin.com/beqojupo/1/quiet?JS_FULL_SCREEN_INVALIDATION) has less noise.  The raster task time went up and didn't recover.

https://chromeperf.appspot.com/report?sid=bb6d70462455e3e7b5571fa9d154cc79a54dbfd6aa3362bfb843cfc9edba3711&start_rev=476959&end_rev=495640

There are no images, this is basically drawing colored rects.  Do you know why raster CPU time may have increased?

btw, this benchmark is using Software Raster.

For an Android device which will always be sRGB, an increase in raster time for solid colors doesn't make much sense. I can take a look to see if anything jumps out at me.

While we're in the neighborhood, notice on the graph that there are some much larger regressions over the last year. Do we know what happened there?

Deleted: fullcontext.png 322 KB

	fullcontext.png 322 KB View Download

Re #6: I openend a few issues on Friday.  Issue 757134 is the GPU scheduler getting turned on.

There seem to be upticks on 07-22 and 08-16 as well which I plan to bisect.