New issue
Advanced search Search tips
Note: Color blocks (like or ) mean that a user may not be available. Tooltip shows the reason.

Issue 698261 link

Starred by 1 user

Issue metadata

Status: WontFix
Owner:
Closed: Apr 2017
Cc:
Components:
EstimatedDays: ----
NextAction: ----
OS: Android
Pri: 1
Type: Bug



Sign in to add a comment

50% regression on smoothness.key_mobile_sites_smooth in 441343:441726 on Android

Project Member Reported by ajuma@chromium.org, Mar 3 2017

Issue description

mean_pixels_checkerboarded regressed by ~50% on Nexus 5X and Nexus 6:
https://chromeperf.appspot.com/report?sid=7bfb28f87cd369f7501ddce4c26d6b2946f7e8eddd1853e10162d8abe8f52a57&start_rev=440434&end_rev=445299

Kicking off bisects.

 

=== BISECT JOB RESULTS ===
Bisect failed unexpectedly

Bisect was aborted with the following:
  Infra Failure: Step('device_status (3)') returned 1


Bisect Details
  Configuration: android_nexus5X_perf_bisect
  Benchmark    : smoothness.key_mobile_sites_smooth
  Metric       : mean_pixels_checkerboarded/mean_pixels_checkerboarded

Revision             Result                      N
chromium@441343      0.0307326 +- 0.0612337      14      good
chromium@441726      0.0475256 +- 0.0721127      10      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 smoothness.key_mobile_sites_smooth

Debug Info
  https://chromeperf.appspot.com/buildbucket_job_status/8986128420550667952

Is this bisect wrong?
  https://chromeperf.appspot.com/bad_bisect?try_job_id=5853287774945280


| 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!

=== BISECT JOB RESULTS ===
Perf regression found but unable to narrow commit range

Build failures prevented the bisect from narrowing the range further.


Bisect Details
  Configuration: android_nexus6_perf_bisect
  Benchmark    : smoothness.key_mobile_sites_smooth
  Metric       : mean_pixels_checkerboarded/mean_pixels_checkerboarded

Suspected Commit Range
  48 commits in range
  https://chromium.googlesource.com/chromium/src/+log/97a6fdc4e9b2194e4d5f245dc6704db72bf52bde..566236e262e49531224e263885e6208259725ab0


Revision             Result                      N
chromium@441343      0.0220171 +- 0.0156808      9        good
chromium@441535      0.0269516 +- 0.0362414      9        good
chromium@441583      0.0284915 +- 0.012385       6        good
chromium@441584      ---                         ---      build failure
---                  ---                         ---      too many build failures to list
chromium@441630      ---                         ---      build failure
chromium@441631      0.0432381 +- 0.0368545      14       bad
chromium@441726      0.0441346 +- 0.0230693      8        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 smoothness.key_mobile_sites_smooth

Debug Info
  https://chromeperf.appspot.com/buildbucket_job_status/8986128448496073904

Is this bisect wrong?
  https://chromeperf.appspot.com/bad_bisect?try_job_id=5886850260008960


| 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!

=== BISECT JOB RESULTS ===
Perf regression found but unable to narrow commit range

Build failures prevented the bisect from narrowing the range further.


Bisect Details
  Configuration: android_nexus6_perf_bisect
  Benchmark    : smoothness.key_mobile_sites_smooth
  Metric       : mean_pixels_checkerboarded/mean_pixels_checkerboarded

Suspected Commit Range
  24 commits in range
  https://chromium.googlesource.com/chromium/src/+log/97a6fdc4e9b2194e4d5f245dc6704db72bf52bde..21385c25732a75a1e7172089afb08285566471f3


Revision             Result                      N
chromium@441583      0.0226374 +- 0.0248714      7        good
chromium@441584      ---                         ---      build failure
---                  ---                         ---      too many build failures to list
chromium@441606      ---                         ---      build failure
chromium@441607      0.0436239 +- 0.0282813      9        bad
chromium@441631      0.0429444 +- 0.0161666      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 smoothness.key_mobile_sites_smooth

Debug Info
  https://chromeperf.appspot.com/buildbucket_job_status/8986018998441351360

Is this bisect wrong?
  https://chromeperf.appspot.com/bad_bisect?try_job_id=6463146992074752


| 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!

=== BISECT JOB RESULTS ===
Perf regression found but unable to narrow commit range

Build failures prevented the bisect from narrowing the range further.


Bisect Details
  Configuration: android_nexus6_perf_bisect
  Benchmark    : smoothness.key_mobile_sites_smooth
  Metric       : mean_pixels_checkerboarded/mean_pixels_checkerboarded

Suspected Commit Range
  12 commits in range
  https://chromium.googlesource.com/chromium/src/+log/97a6fdc4e9b2194e4d5f245dc6704db72bf52bde..f6a0f4b7f8abe662390f61e1d5133a0894bb6080


Revision             Result                      N
chromium@441583      0.0196496 +- 0.0191583      6        good
chromium@441584      ---                         ---      build failure
---                  ---                         ---      too many build failures to list
chromium@441594      ---                         ---      build failure
chromium@441595      0.0407806 +- 0.0373329      9        bad
chromium@441607      0.039359 +- 0.00866614      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 smoothness.key_mobile_sites_smooth

Debug Info
  https://chromeperf.appspot.com/buildbucket_job_status/8985928859197403456

Is this bisect wrong?
  https://chromeperf.appspot.com/bad_bisect?try_job_id=5812761000411136


| 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!
Cc: schenney@chromium.org
Owner: schenney@chromium.org

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

Hi schenney@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 : schenney
  Commit : d3ba7664a7ebc6e74861602eb9696c7ca0a4ed2e
  Date   : Thu Jan 05 06:01:46 2017
  Subject: Fix border radius on composited children.

Bisect Details
  Configuration: android_nexus6_perf_bisect
  Benchmark    : smoothness.key_mobile_sites_smooth
  Metric       : mean_pixels_checkerboarded/mean_pixels_checkerboarded
  Change       : 94.65% | 0.0217264957265 -> 0.0422905982906

Revision             Result                      N
chromium@441583      0.0217265 +- 0.0170529      6      good
chromium@441589      0.0253952 +- 0.0219079      9      good
chromium@441591      0.0239801 +- 0.0210076      9      good
chromium@441592      0.0435869 +- 0.0309091      9      bad       <--
chromium@441595      0.0422906 +- 0.0132384      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 smoothness.key_mobile_sites_smooth

Debug Info
  https://chromeperf.appspot.com/buildbucket_job_status/8985862948855065840

Is this bisect wrong?
  https://chromeperf.appspot.com/bad_bisect?try_job_id=5837142456008704


| 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!
Cc: sunxd@chromium.org chrishtr@chromium.org wkorman@chromium.org
Components: Blink>Compositing
Labels: M-57
This might be case where the perf regression is the price we have to pay for correctness, though it would be worth double-checking that the additional border-radius clipping here really does clip something. I also wonder if sunxd@'s work will help with reducing the perf impact.
It does seem like it's my patch, because there is also a blip up when the original version was landed then reverted (Dec 12-13).

We already check, in a slightly fast and dirty version, that the mask is actually clipping something. We could try to make a more precise test but I very much doubt that it will make a difference. I'll need to look at the sites in the test to see which are causing problems.

There's another improvement I need to look into that might help, but that's more about memory than speed, I think.
Labels: PaintTeamTriaged-20170303 BugSource-Team
Labels: -M-57 M-60
Status: WontFix (was: Assigned)
The Nexus 5x has reverted back to very low numbers, so no action there.

The Nexus 6 has stayed where it was after this patch. We're not going to roll out the patch because it fixes a major problem. We're on the verge of landing a patch to improve performance of the masks introduced by my patch. So I think WontFix is the right strategy. 

Sign in to add a comment