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

Issue 694615 link

Starred by 1 user

Issue metadata

Status: WontFix
Owner:
Closed: Mar 2017
Cc:
Components:
EstimatedDays: ----
NextAction: ----
OS: ----
Pri: 2
Type: Bug-Regression



Sign in to add a comment

2.5%-2.7% regression in smoothness.tough_path_rendering_cases at 450321:450463

Project Member Reported by rsch...@chromium.org, Feb 21 2017

Issue description

See the link to graphs below.
 
All graphs for this bug:
  https://chromeperf.appspot.com/group_report?bug_id=694615

Original alerts at time of bug-filing:
  https://chromeperf.appspot.com/group_report?keys=agxzfmNocm9tZXBlcmZyFAsSB0Fub21hbHkYgIDg-JacpgkM,agxzfmNocm9tZXBlcmZyFAsSB0Fub21hbHkYgIDghOSnvgoM


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

chromium-rel-mac-retina
chromium-rel-mac11-pro
Project Member

Comment 3 by 42576172...@developer.gserviceaccount.com, Feb 21 2017

Cc: yosin@chromium.org
Owner: yosin@chromium.org

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

Hi yosin@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 : yosin
  Commit : d892f9592860691ae9a782c12260c94ed6bd1a63
  Date   : Tue Feb 14 15:56:00 2017
  Subject: Make FrameSelection to hold non-canonicalized DOM positions

Bisect Details
  Configuration: mac_retina_perf_bisect
  Benchmark    : smoothness.tough_path_rendering_cases
  Metric       : frame_times/frame_times
  Change       : 2.01% | 24.1652206804 -> 24.6519266086

Revision             Result                   N
chromium@450369      24.1652 +- 0.151109      6      good
chromium@450370      24.7861 +- 0.306275      6      bad       <--
chromium@450371      24.7307 +- 0.314754      6      bad
chromium@450372      24.7445 +- 0.3803        6      bad
chromium@450375      24.7557 +- 0.14461       6      bad
chromium@450381      24.8083 +- 0.266947      6      bad
chromium@450393      24.6715 +- 0.387396      6      bad
chromium@450416      24.8992 +- 0.25965       6      bad
chromium@450463      24.6519 +- 0.210773      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 smoothness.tough_path_rendering_cases

Debug Info
  https://chromeperf.appspot.com/buildbucket_job_status/8987029402872497200

Is this bisect wrong?
  https://chromeperf.appspot.com/bad_bisect?try_job_id=6146752631013376


| 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!

Comment 4 by ericrk@chromium.org, Feb 22 2017

Components: -Internals>GPU Blink>Editing>Selection

Comment 5 by yosin@chromium.org, Feb 23 2017

Status: WontFix (was: Untriaged)
Not sue how r450370 increase time 2.7% = 0.656 ms.
Cc: senorblanco@chromium.org
Status: Assigned (was: WontFix)
Could you revert your change to be sure? You cannot just say "I don't know" and close as wont fix. Bisect pointed to your change

Comment 7 by yosin@chromium.org, Mar 10 2017

Since r450370 is fundamental change, it is hard to revert.
I'll investigate more.
Note that I kicked off another bisect on the Nexus7v2 regression (since it's much larger), but then noticed that the reference build was affected too. Also, that regression has a much wider range, and it's likely due to the fact that new tests were added to the suite while the bot was down. So you can ignore that regression.

Comment 10 by yosin@chromium.org, Mar 10 2017

Owner: ----
Status: WontFix (was: Assigned)
Mark WontFix regarding to #c9.

Thanks senorblanco@ for providing insight.
Thanks for pointing out the change in test cases. I changed the monitoring so we'll see alerts for individual pages, which will stop this from happening again and also clarify to the bisect what case it's trying to repro. 
Actually, by "that regression", I meant the large one on N7v2. The smaller one here (2.5%-2.7% on mac-retina and mac11-pro) is still unsolved, and doesn't include the change in test cases. Note that the intersection of the two bots has a much narrower range (450370-450375), and the bisect found here is in that range.

I have been eking out small gains (1-2%) with subsequent changes in Skia (which is why the graph is trending down). I have a larger optimization which if it pans out, should yield a 10-20% boost. It would be nice to figure out this regression and get it back but it's almost in the noise, so I wouldn't lose sleep over it.

sullivan@: Thanks! Having alerts fire for individual tests will be really helpful.
Labels: Performance-Responsiveness
Project Member

Comment 15 by 42576172...@developer.gserviceaccount.com, Mar 24 2017


=== BISECT JOB RESULTS ===
Bisect was unable to run to completion

Error: TIMEOUT

The bisect was able to narrow the range, you can try running with:
  good_revision: 255e749e57448c43f5f9f49ba0bf842cd92a6b69
  bad_revision : 3301fc1e1974b51ccb59f24519de0dcde34b9333

If failures persist contact the team (see below) and report the error.


Bisect Details
  Configuration: android_nexus7_perf_bisect
  Benchmark    : smoothness.tough_path_rendering_cases
  Metric       : frame_times/frame_times
  Change       : 42.17% | 56.8431073193 -> 80.8136808264

Revision             Result                  N
chromium@448737      56.8431 +- 1.01965      6      good
chromium@449130      79.9008 +- 1.54395      6      bad
chromium@449522      80.7093 +- 2.05891      6      bad
chromium@450307      80.9856 +- 2.36922      6      bad
chromium@451858      80.8137 +- 2.95976      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.tough_path_rendering_cases

Debug Info
  https://chromeperf.appspot.com/buildbucket_job_status/8984378573961521744

Is this bisect wrong?
  https://chromeperf.appspot.com/bad_bisect?try_job_id=4933524689256448


| 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!
Project Member

Comment 17 by 42576172...@developer.gserviceaccount.com, Mar 30 2017

Owner: yosin@chromium.org

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

Hi yosin@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 : yosin
  Commit : d892f9592860691ae9a782c12260c94ed6bd1a63
  Date   : Tue Feb 14 15:56:00 2017
  Subject: Make FrameSelection to hold non-canonicalized DOM positions

Bisect Details
  Configuration: mac_pro_perf_bisect
  Benchmark    : smoothness.tough_path_rendering_cases
  Metric       : frame_times/frame_times
  Change       : 2.53% | 27.1247354913 -> 27.8109416294

Revision             Result                   N
chromium@450320      27.1247 +- 0.134372      6      good
chromium@450348      27.036 +- 0.0799685      6      good
chromium@450362      27.1154 +- 0.111862      6      good
chromium@450369      27.2018 +- 0.292346      6      good
chromium@450370      27.7719 +- 0.328647      6      bad       <--
chromium@450371      27.7383 +- 0.3804        6      bad
chromium@450372      27.7358 +- 0.203606      6      bad
chromium@450375      27.8109 +- 0.263225      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 smoothness.tough_path_rendering_cases

Debug Info
  https://chromeperf.appspot.com/buildbucket_job_status/8983687242964032912

Is this bisect wrong?
  https://chromeperf.appspot.com/bad_bisect?try_job_id=4933524689256448


| 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!
Hmm. seems like there's a separate regression in the GUIMark Vector Chart test. Seems to have mostly recovered, but thought I'd isolate it here:

https://chromeperf.appspot.com/report?sid=017f2fcb7aa9b03430595379c526985a1a69adedc11dc403853aa7583fda42a4&start_rev=445000&end_rev=470000

Sign in to add a comment