Issue metadata
Sign in to add a comment
|
2.8%-32.9% regression in rasterize_and_record_micro.top_25 at 503836:503934 |
||||||||||||||||||||||
Issue descriptionSee the link to graphs below.
,
Sep 25 2017
Started bisect job https://chromeperf.appspot.com/buildbucket_job_status/8967446426408019232
,
Sep 25 2017
=== BISECT JOB RESULTS === Bisect failed for unknown reasons Please contact the team (see below) and report the error. Bisect Details Configuration: mac_10_11_perf_bisect Benchmark : rasterize_and_record_micro.top_25 Metric : rasterize_time/file___static_top_25_googleplus.html To Run This Test src/tools/perf/run_benchmark -v --browser=release --output-format=chartjson --upload-results --pageset-repeat=1 --also-run-disabled-tests --story-filter=file...static.top.25.googleplus.html rasterize_and_record_micro.top_25 More information on addressing performance regressions: http://g.co/ChromePerformanceRegressions Debug information about this bisect: https://chromeperf.appspot.com/buildbucket_job_status/8967446426408019232 For feedback, file a bug with component Speed>Bisection
,
Sep 25 2017
📍 Pinpoint job started. https://pinpoint-dot-chromeperf.appspot.com/job/104d53a8780000
,
Sep 25 2017
📍 Pinpoint job stopped with an error. https://pinpoint-dot-chromeperf.appspot.com/job/104d53a8780000 Couldn't reproduce a difference.
,
Sep 25 2017
khushalsagar: If you look at the bisect link from #5, it's pretty clear that it actually sees a regression at r503885: "Reland images: Move animation of images to cc" dtu: cc-ing since output in #5 doesn't match up with what I expect from examining the points on the chart.
,
Sep 25 2017
#6: https://codereview.chromium.org/3020483002/ will remove the incorrect output. Filed https://github.com/catapult-project/catapult/issues/3914 for the underlying error.
,
Sep 26 2017
📍 Pinpoint job started. https://pinpoint-dot-chromeperf.appspot.com/job/169f1a64780000
,
Sep 26 2017
📍 Pinpoint job started. https://pinpoint-dot-chromeperf.appspot.com/job/11e185b8780000
,
Sep 26 2017
📍 Pinpoint job completed. https://pinpoint-dot-chromeperf.appspot.com/job/169f1a64780000 Found significant differences after 1 commit: Reland images: Move animation of images to cc. By khushalsagar@chromium.org · Fri Sep 22 22:40:52 2017 chromium@fdacdc9f7bd8e4b0cf23b192e959c2350f693e7c
,
Sep 26 2017
📍 Pinpoint job completed. https://pinpoint-dot-chromeperf.appspot.com/job/11e185b8780000 Found significant differences after 1 commit: Reland images: Move animation of images to cc. By khushalsagar@chromium.org · Fri Sep 22 22:40:52 2017 chromium@fdacdc9f7bd8e4b0cf23b192e959c2350f693e7c
,
Sep 26 2017
Started bisect job https://chromeperf.appspot.com/buildbucket_job_status/8967364799143336256
,
Sep 26 2017
Started bisect job https://chromeperf.appspot.com/buildbucket_job_status/8967364739225121296
,
Sep 26 2017
=== BISECT JOB RESULTS === Perf regression found with culprit Suspected Commit Author : Herb Derby Commit : 0f68a2d3dfaeef91796c0e7e3f5e92a1d949ad2b Date : Thu Sep 14 22:38:06 2017 Subject: Use Skia's faster blur code. Bisect Details Configuration: mac_10_11_perf_bisect Benchmark : rasterize_and_record_micro.top_25 Metric : rasterize_time/file___static_top_25_google.html Change : 11.78% | 3.95466666667 -> 3.48866666667 Revision Result N chromium@502008 3.95467 +- 0.103225 6 good chromium@502031 3.98383 +- 0.0356768 6 good chromium@502043 3.98617 +- 0.0530362 6 good chromium@502045 3.971 +- 0.0252587 6 good chromium@502046 3.4615 +- 0.0221698 6 bad <-- chromium@502049 3.4545 +- 0.0747496 6 bad chromium@502054 3.47633 +- 0.0327923 6 bad chromium@502100 3.48867 +- 0.0264827 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 --story-filter=file...static.top.25.google.html rasterize_and_record_micro.top_25 More information on addressing performance regressions: http://g.co/ChromePerformanceRegressions Debug information about this bisect: https://chromeperf.appspot.com/buildbucket_job_status/8967364739225121296 For feedback, file a bug with component Speed>Bisection
,
Sep 26 2017
=== Auto-CCing suspected CL author herb@google.com === Hi herb@google.com, 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 : Herb Derby Commit : bf13c2bc8e5d9372c0ce50ef4637c7acaec3d401 Date : Fri Sep 08 23:26:09 2017 Subject: Use interpolation for small radii. Bisect Details Configuration: mac_retina_perf_bisect Benchmark : rasterize_and_record_micro.top_25 Metric : rasterize_time/file___static_top_25_google.html Change : 13.53% | 8.25133333333 -> 7.13533333333 Revision Result N chromium@500691 8.25133 +- 0.042111 6 good chromium@500714 8.2555 +- 0.103709 6 good chromium@500726 8.238 +- 0.0415933 6 good chromium@500732 8.25183 +- 0.0449314 6 good chromium@500735 8.2485 +- 0.0809166 6 good chromium@500736 7.10467 +- 0.0454899 6 bad <-- chromium@500737 7.1245 +- 0.080731 6 bad chromium@500783 7.13533 +- 0.140012 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 --story-filter=file...static.top.25.google.html rasterize_and_record_micro.top_25 More information on addressing performance regressions: http://g.co/ChromePerformanceRegressions Debug information about this bisect: https://chromeperf.appspot.com/buildbucket_job_status/8967364799143336256 For feedback, file a bug with component Speed>Bisection
,
Sep 26 2017
Assigning it back to me. The graph in a lot of cases looked like it had dropped a few days ago and jumped back to the same value with r503885 which looked a bit suspect. I was trying to see what improved it and whether that had any co-relation with the change that resulted in the regression.
,
Sep 28 2017
The regression is coming from the additional parameter added to PlaybackImageProvider. I tried a patch which only removes that map (which was empty for the benchmark anyway), and that fixes it. See https://chromium-review.googlesource.com/c/chromium/src/+/687872. Its very weird that a single extra param would regress the object construction time by such a significant amount. This is also the root cause for 768543.
,
Sep 29 2017
,
Sep 29 2017
The following revision refers to this bug: https://chromium.googlesource.com/chromium/src.git/+/5d1be840afed203dee9425853a0a64b9d61cdd3b commit 5d1be840afed203dee9425853a0a64b9d61cdd3b Author: Khushal <khushalsagar@chromium.org> Date: Fri Sep 29 17:51:17 2017 cc: Address raster regression from PlaybackImageProvider construction. Adding additional parameters to PlaybackImageProvider ctor causes a regression in total raster time on some mac configs. In order to address this, break out these settings to a separate struct which can optionally be given to the provider. The params are modified on the Settings object post construction. R=enne@chromium.org Bug: 768538 , 768543 Cq-Include-Trybots: master.tryserver.blink:linux_trusty_blink_rel Change-Id: I0e7fc0df225826e5a26f0f9d09c106a6b2d1f640 Reviewed-on: https://chromium-review.googlesource.com/691206 Reviewed-by: enne <enne@chromium.org> Reviewed-by: vmpstr <vmpstr@chromium.org> Commit-Queue: Khushal <khushalsagar@chromium.org> Cr-Commit-Position: refs/heads/master@{#505416} [modify] https://crrev.com/5d1be840afed203dee9425853a0a64b9d61cdd3b/cc/benchmarks/rasterize_and_record_benchmark_impl.cc [modify] https://crrev.com/5d1be840afed203dee9425853a0a64b9d61cdd3b/cc/raster/playback_image_provider.cc [modify] https://crrev.com/5d1be840afed203dee9425853a0a64b9d61cdd3b/cc/raster/playback_image_provider.h [modify] https://crrev.com/5d1be840afed203dee9425853a0a64b9d61cdd3b/cc/raster/playback_image_provider_unittest.cc [modify] https://crrev.com/5d1be840afed203dee9425853a0a64b9d61cdd3b/cc/tiles/tile_manager.cc
,
Oct 2 2017
Issue 768543 has been merged into this issue.
,
Oct 2 2017
All graphs have recovered. |
|||||||||||||||||||||||
►
Sign in to add a comment |
|||||||||||||||||||||||
Comment 1 by 42576172...@developer.gserviceaccount.com
, Sep 25 2017