New issue
Advanced search Search tips

Issue 620687 link

Starred by 1 user

Issue metadata

Status: Fixed
Owner:
Closed: Jun 2016
Cc:
EstimatedDays: ----
NextAction: ----
OS: Windows
Pri: 1
Type: Bug-Regression



Sign in to add a comment

Win 7 Nvidia GPU Perf (2) failing on angle_perftests

Project Member Reported by picksi@chromium.org, Jun 16 2016

Issue description

Link to bot : https://build.chromium.org/p/chromium.perf/builders/Win%207%20Nvidia%20GPU%20Perf%20%282%29

angle_perftests have been failing since #5921. top_10_mobile_memory is flaky and occasionally is reported as a reason for failure.

[40/40] DrawCallPerfBenchmark.Run/gl_null (TIMED OUT)
1 test timed out:
    DrawCallPerfBenchmark.Run/gl_null (c:\b\build\slave\win_x64_builder\build\src\third_party\angle\src\tests\perf_tests\drawcallperf.cpp:231)

Will look into disabling the test.
 

Comment 1 by picksi@chromium.org, Jun 16 2016

I don't see the named test in drawcallperf.cpp. Line 231 is just calls a virtual run() function.

I tried kicking off a bisect (via this graph: https://chromeperf.appspot.com/report?sid=e94a162abb649d5a83aa05bd325625e0367123b39cdd90b4abb931c83d9ad10a&rev=339862) but it looks like it failed. :(


Comment 2 by picksi@chromium.org, Jun 16 2016

Cc: jmad...@chromium.org
Labels: OS-Windows
@jmadill, I'm looking for a possible cause for the failure in #1 (failure in angle_perftests). It looks like you touched this file recently ... although a couple of days before the bots turned red :$

Do you have any idea what might be causing the time out?
Project Member

Comment 3 by 42576172...@developer.gserviceaccount.com, Jun 16 2016


===== BISECT JOB RESULTS =====
Status: completed


===== TESTED REVISIONS =====
Revision         Exit Code  Std Dev  N   Good?
chromium@399929  1          N/A      20  good
chromium@399973  1          N/A      20  bad

Bisect job ran on: winx64nvidia_perf_bisect
Bug ID: 620687

Test Command: .\src\out\Release_x64\angle_perftests.exe --test-launcher-print-test-stdio=always --test-launcher-jobs=1
Test Metric: DrawCallPerf_gl_null/score
Relative Change: 0.00%
Score: 0

Buildbot stdio: http://build.chromium.org/p/tryserver.chromium.perf/builders/winx64nvidia_perf_bisect/builds/1656
Job details: https://chromeperf.appspot.com/buildbucket_job_status/9009701586548437664


Not what you expected? We'll investigate and get back to you!
  https://chromeperf.appspot.com/bad_bisect?try_job_id=5307840030310400

| 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 Tests>AutoBisect.  Thank you!
Owner: jmad...@chromium.org
Status: Assigned (was: Untriaged)
I forget, how do I view the stdio? I'll see if I can figure this out today.

Comment 5 by picksi@chromium.org, Jun 16 2016

Here's the top level builder showing all the builds:
https://build.chromium.org/p/chromium.perf/builders/Win%207%20Nvidia%20GPU%20Perf%20%282%29

If you click the build ID it takes you to a specific build. e.g. clicking on #5929 takes you here:

https://build.chromium.org/p/chromium.perf/builders/Win%207%20Nvidia%20GPU%20Perf%20%282%29/builds/5929

The stdio is then available as a link under each build step. So the stdio for angle_perftests (number 14) in the list, takes you here:

https://build.chromium.org/p/chromium.perf/builders/Win%207%20Nvidia%20GPU%20Perf%20%282%29/builds/5929/steps/angle_perftests/logs/stdio



Owner: geoffl...@chromium.org
Geoff is going to look into disabling the tests. I have no idea how to debug this, cannot repro locally.
Cc: geoffl...@chromium.org
Owner: jmad...@chromium.org
We've figure this out, working on a fix.
Project Member

Comment 8 by bugdroid1@chromium.org, Jun 16 2016

The following revision refers to this bug:
  https://chromium.googlesource.com/angle/angle/+/cd1b12260360cbf38e9dbcb6f9a8ffff4ed2c393

commit cd1b12260360cbf38e9dbcb6f9a8ffff4ed2c393
Author: Jamie Madill <jmadill@chromium.org>
Date: Thu Jun 16 17:58:54 2016

perf_tests: Work around stalls in gl_null test teardown.

The Clear command in the draw call benchmark was not being properly
no-op'ed. This would cause the driver to get a huge queue of GL
command, which it would only flush when the test was torn down. This
manifested as a 5+ second stall on test teardown.

Looking at the clear command, it does some work with binding
and then calling GL functions, so properly filtering out the driver
commands would be a bit more complicated. For now, we can hack in a
workaround for the draw call test (the only one that uses the GL
NULL renderer at the moment) to not call clear when it would give
us problems.

BUG= 620687 

Change-Id: I64a2ed1883e8d3bf0d4c423c8bce0902ae9bf8df
Reviewed-on: https://chromium-review.googlesource.com/352828
Reviewed-by: Geoff Lang <geofflang@chromium.org>

[modify] https://crrev.com/cd1b12260360cbf38e9dbcb6f9a8ffff4ed2c393/src/tests/perf_tests/DrawCallPerf.cpp

Project Member

Comment 9 by bugdroid1@chromium.org, Jun 16 2016

The following revision refers to this bug:
  https://chromium.googlesource.com/chromium/src.git/+/e973b43fd81bb5066ec3c95034e8290cde5068eb

commit e973b43fd81bb5066ec3c95034e8290cde5068eb
Author: jmadill <jmadill@chromium.org>
Date: Thu Jun 16 21:19:09 2016

Roll ANGLE 9fc7b4c..cd1b122

https://chromium.googlesource.com/angle/angle.git/+log/9fc7b4c..cd1b122

BUG= chromium:616176 , 620687 

TEST=bots
TBR=geofflang@chromium.org
CQ_INCLUDE_TRYBOTS=tryserver.chromium.win:win_optional_gpu_tests_rel;tryserver.chromium.mac:mac_optional_gpu_tests_rel;tryserver.chromium.linux:linux_optional_gpu_tests_rel

Review-Url: https://codereview.chromium.org/2076523003
Cr-Commit-Position: refs/heads/master@{#400258}

[modify] https://crrev.com/e973b43fd81bb5066ec3c95034e8290cde5068eb/DEPS

Status: Fixed (was: Assigned)
Looks good now.

Sign in to add a comment