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

Issue 639065 link

Starred by 1 user

Issue metadata

Status: WontFix
Owner:
Last visit > 30 days ago
Closed: May 2017
Cc:
EstimatedDays: ----
NextAction: ----
OS: ----
Pri: 2
Type: Bug-Regression



Sign in to add a comment

39.3%-61.8% regression in cpu_utilization for media.android.tough_video_cases at 412621:412670

Project Member Reported by chcunningham@google.com, Aug 18 2016

Issue description

Significant spike in cpu utilization across several android devices. 
 
Project Member

Comment 3 by 42576172...@developer.gserviceaccount.com, Aug 19 2016


===== BISECT JOB RESULTS =====
Status: failed


===== TESTED REVISIONS =====
Revision         Mean     Std Dev  N   Good?
chromium@412620  27.052   1.95714  12  good
chromium@412645  22.3822  1.7966   5   bad
chromium@412670  24.0651  2.89448  8   bad

Bisect job ran on: android_nexus5X_perf_bisect
Bug ID: 639065

Test Command: src/tools/perf/run_benchmark -v --browser=android-chromium --output-format=chartjson --upload-results --also-run-disabled-tests media.android.tough_video_cases
Test Metric: cpu_cstate_WFI_residency_percent/video.html?src_smpte_3840x2160_60fps_vp9.webm
Relative Change: 13.37%
Score: 0

Buildbot stdio: http://build.chromium.org/p/tryserver.chromium.perf/builders/android_nexus5X_perf_bisect/builds/532
Job details: https://chromeperf.appspot.com/buildbucket_job_status/9003959466683254768


Not what you expected? We'll investigate and get back to you!
  https://chromeperf.appspot.com/bad_bisect?try_job_id=5801990349651968

| 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!
Project Member

Comment 5 by 42576172...@developer.gserviceaccount.com, Aug 19 2016


===== BISECT JOB RESULTS =====
Status: failed


===== TESTED REVISIONS =====
Revision         Mean     Std Dev  N   Good?
chromium@412620  27.052   1.95714  12  good
chromium@412645  22.3822  1.7966   5   bad
chromium@412670  24.0651  2.89448  8   bad

Bisect job ran on: android_nexus5X_perf_bisect
Bug ID: 639065

Test Command: src/tools/perf/run_benchmark -v --browser=android-chromium --output-format=chartjson --upload-results --also-run-disabled-tests media.android.tough_video_cases
Test Metric: cpu_cstate_WFI_residency_percent/video.html?src_smpte_3840x2160_60fps_vp9.webm
Relative Change: 13.37%
Score: 0

Buildbot stdio: http://build.chromium.org/p/tryserver.chromium.perf/builders/android_nexus5X_perf_bisect/builds/532
Job details: https://chromeperf.appspot.com/buildbucket_job_status/9003959466683254768


Not what you expected? We'll investigate and get back to you!
  https://chromeperf.appspot.com/bad_bisect?try_job_id=5801990349651968

| 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!
Project Member

Comment 6 by sheriffbot@chromium.org, Aug 19 2016

Labels: Hotlist-Google
Cc: pras...@chromium.org
Retrying bisects.

prasadv@ - sorry to randomly include, pls re-route as needed. Can you help me understand why the firsts bisect failed? They don't really give an error msg AFAIK.
Cc: robert...@chromium.org
Hi chcunningham@chromium.org, Looks like while retrying the tests, the step command, after 60 mins without any output it timed out.
Previous test runs took around 8~9 minutes.

https://build.chromium.org/p/tryserver.chromium.perf/builders/android_nexus5X_perf_bisect/builds/532/steps/Working%20on%20revision%20chromium%40412633.Waiting%20for%20chromium%40412633.Waiting%20for%20chromium%40412633.Waiting%20for%20chromium%40412633.Performance%20Test%206%20of%206/logs/stdio


Hi Roberto, Is this has to do anything with multiple nesting?
Project Member

Comment 12 by 42576172...@developer.gserviceaccount.com, Aug 20 2016


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


=== Bisection aborted ===
The bisect was aborted because The metric values for the initial "good" and "bad" revisions do not represent a clear regression.
Please contact the the team (see below) if you believe this is in error.

=== Warnings ===
The following warnings were raised by the bisect job:

 * Bisect failed to reproduce the regression with enough confidence.

===== TESTED REVISIONS =====
Revision         Mean     Std Dev   N   Good?
chromium@412620  7.56403  0.104825  12  good
chromium@412670  7.52792  0.202036  12  bad

Bisect job ran on: android_nexus5X_perf_bisect
Bug ID: 639065

Test Command: src/tools/perf/run_benchmark -v --browser=android-chromium --output-format=chartjson --upload-results --also-run-disabled-tests media.android.tough_video_cases
Test Metric: cpu_utilization/smpte_3840x2160_60fps_vp9.webm_browser
Relative Change: 1.34%
Score: 0

Buildbot stdio: http://build.chromium.org/p/tryserver.chromium.perf/builders/android_nexus5X_perf_bisect/builds/544
Job details: https://chromeperf.appspot.com/buildbucket_job_status/9003866371072079504


Not what you expected? We'll investigate and get back to you!
  https://chromeperf.appspot.com/bad_bisect?try_job_id=4997142469410816

| 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!
Project Member

Comment 13 by 42576172...@developer.gserviceaccount.com, Aug 20 2016


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


=== Bisection aborted ===
The bisect was aborted because The metric values for the initial "good" and "bad" revisions do not represent a clear regression.
Please contact the the team (see below) if you believe this is in error.

=== Warnings ===
The following warnings were raised by the bisect job:

 * Bisect failed to reproduce the regression with enough confidence.

===== TESTED REVISIONS =====
Revision         Mean     Std Dev   N   Good?
chromium@412620  19.0452  0.413545  12  good
chromium@412670  19.0991  0.235241  12  bad

Bisect job ran on: android_nexus5X_perf_bisect
Bug ID: 639065

Test Command: src/tools/perf/run_benchmark -v --browser=android-chromium --output-format=chartjson --upload-results --also-run-disabled-tests media.android.tough_video_cases
Test Metric: cpu_utilization/smpte_3840x2160_60fps_vp9.webm_gpu
Relative Change: 0.47%
Score: 0

Buildbot stdio: http://build.chromium.org/p/tryserver.chromium.perf/builders/android_nexus5X_perf_bisect/builds/545
Job details: https://chromeperf.appspot.com/buildbucket_job_status/9003866374692972672


Not what you expected? We'll investigate and get back to you!
  https://chromeperf.appspot.com/bad_bisect?try_job_id=5334297133711360

| 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!
Project Member

Comment 14 by 42576172...@developer.gserviceaccount.com, Aug 20 2016


===== BISECT JOB RESULTS =====
Status: failed


===== TESTED REVISIONS =====
Revision         Mean     Std Dev    N  Good?
chromium@412620  19.5373  0.15475    5  good
chromium@412670  19.2636  0.0939957  5  bad

Bisect job ran on: android_nexus5X_perf_bisect
Bug ID: 639065

Test Command: src/tools/perf/run_benchmark -v --browser=android-chromium --output-format=chartjson --upload-results --also-run-disabled-tests media.android.tough_video_cases
Test Metric: cpu_utilization/video.html?src_smpte_3840x2160_60fps_vp9.webm
Relative Change: 1.40%
Score: 0

Buildbot stdio: http://build.chromium.org/p/tryserver.chromium.perf/builders/android_nexus5X_perf_bisect/builds/543
Job details: https://chromeperf.appspot.com/buildbucket_job_status/9003866387306146688


Not what you expected? We'll investigate and get back to you!
  https://chromeperf.appspot.com/bad_bisect?try_job_id=6159970236628992

| 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!
Project Member

Comment 16 by 42576172...@developer.gserviceaccount.com, Aug 20 2016


===== BISECT JOB RESULTS =====
Status: failed


===== TESTED REVISIONS =====
Revision         Mean     Std Dev    N  Good?
chromium@412620  19.5373  0.15475    5  good
chromium@412670  19.2636  0.0939957  5  bad

Bisect job ran on: android_nexus5X_perf_bisect
Bug ID: 639065

Test Command: src/tools/perf/run_benchmark -v --browser=android-chromium --output-format=chartjson --upload-results --also-run-disabled-tests media.android.tough_video_cases
Test Metric: cpu_utilization/video.html?src_smpte_3840x2160_60fps_vp9.webm
Relative Change: 1.40%
Score: 0

Buildbot stdio: http://build.chromium.org/p/tryserver.chromium.perf/builders/android_nexus5X_perf_bisect/builds/543
Job details: https://chromeperf.appspot.com/buildbucket_job_status/9003866387306146688


Not what you expected? We'll investigate and get back to you!
  https://chromeperf.appspot.com/bad_bisect?try_job_id=6159970236628992

| 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!
Project Member

Comment 18 by 42576172...@developer.gserviceaccount.com, Aug 20 2016


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


=== Bisection aborted ===
The bisect was aborted because The metric values for the initial "good" and "bad" revisions do not represent a clear regression.
Please contact the the team (see below) if you believe this is in error.

=== Warnings ===
The following warnings were raised by the bisect job:

 * Bisect failed to reproduce the regression with enough confidence.

===== TESTED REVISIONS =====
Revision         Mean     Std Dev  N   Good?
chromium@412620  24.7331  1.56031  12  good
chromium@412670  25.4304  1.96894  12  bad

Bisect job ran on: android_nexus5X_perf_bisect
Bug ID: 639065

Test Command: src/tools/perf/run_benchmark -v --browser=android-chromium --output-format=chartjson --upload-results --also-run-disabled-tests media.android.tough_video_cases
Test Metric: cpu_cstate_WFI_residency_percent/video.html?src_smpte_3840x2160_60fps_vp9.webm
Relative Change: 0.68%
Score: 0

Buildbot stdio: http://build.chromium.org/p/tryserver.chromium.perf/builders/android_nexus5X_perf_bisect/builds/533
Job details: https://chromeperf.appspot.com/buildbucket_job_status/9003959466683254768


Not what you expected? We'll investigate and get back to you!
  https://chromeperf.appspot.com/bad_bisect?try_job_id=5801990349651968

| 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!
Re comment 11: thanks for digging into it. Seems to be somewhat transient... other bisects have since gotten further.

Since filing this bug, many of the graphs have recovered, but some have not. I've just kicked off 3 new bisects for the 3 largest unrecovered regressions - 3 graphs for "ChromiumPerf/android-nexus5X/media.android.tough_video_cases / cpu_utilization /" (not sure whats different about these ... why aren't they just one graph?). For these, bisect should really not fail to find a significant difference between good and bad. Fingers crossed.
Project Member

Comment 23 by 42576172...@developer.gserviceaccount.com, Aug 22 2016


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


=== Bisection aborted ===
The bisect was aborted because The metric values for the initial "good" and "bad" revisions do not represent a clear regression.
Please contact the the team (see below) if you believe this is in error.

=== Warnings ===
The following warnings were raised by the bisect job:

 * Bisect failed to reproduce the regression with enough confidence.

===== TESTED REVISIONS =====
Revision         Mean     Std Dev   N   Good?
chromium@412620  19.1425  0.357672  12  good
chromium@412670  19.6262  0.469599  12  bad

Bisect job ran on: android_nexus5X_perf_bisect
Bug ID: 639065

Test Command: src/tools/perf/run_benchmark -v --browser=android-chromium --output-format=chartjson --upload-results --also-run-disabled-tests media.android.tough_video_cases
Test Metric: cpu_utilization/smpte_3840x2160_60fps_vp9.webm_gpu
Relative Change: 0.53%
Score: 0

Buildbot stdio: http://build.chromium.org/p/tryserver.chromium.perf/builders/android_nexus5X_perf_bisect/builds/555
Job details: https://chromeperf.appspot.com/buildbucket_job_status/9003611500369669344


Not what you expected? We'll investigate and get back to you!
  https://chromeperf.appspot.com/bad_bisect?try_job_id=5322328100044800

| 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!
Project Member

Comment 24 by 42576172...@developer.gserviceaccount.com, Aug 22 2016


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


=== Bisection aborted ===
The bisect was aborted because The metric values for the initial "good" and "bad" revisions do not represent a clear regression.
Please contact the the team (see below) if you believe this is in error.

=== Warnings ===
The following warnings were raised by the bisect job:

 * Bisect failed to reproduce the regression with enough confidence.

===== TESTED REVISIONS =====
Revision         Mean     Std Dev   N   Good?
chromium@412620  7.562    0.110904  12  good
chromium@412670  7.56092  0.126664  12  bad

Bisect job ran on: android_nexus5X_perf_bisect
Bug ID: 639065

Test Command: src/tools/perf/run_benchmark -v --browser=android-chromium --output-format=chartjson --upload-results --also-run-disabled-tests media.android.tough_video_cases
Test Metric: cpu_utilization/smpte_3840x2160_60fps_vp9.webm_browser
Relative Change: 0.34%
Score: 0

Buildbot stdio: http://build.chromium.org/p/tryserver.chromium.perf/builders/android_nexus5X_perf_bisect/builds/556
Job details: https://chromeperf.appspot.com/buildbucket_job_status/9003611476920974496


Not what you expected? We'll investigate and get back to you!
  https://chromeperf.appspot.com/bad_bisect?try_job_id=5794633305554944

| 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!
Project Member

Comment 25 by 42576172...@developer.gserviceaccount.com, Aug 22 2016


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


=== Bisection aborted ===
The bisect was aborted because The metric values for the initial "good" and "bad" revisions do not represent a clear regression.
Please contact the the team (see below) if you believe this is in error.

=== Warnings ===
The following warnings were raised by the bisect job:

 * Bisect failed to reproduce the regression with enough confidence.

===== TESTED REVISIONS =====
Revision         Mean     Std Dev   N   Good?
chromium@412620  19.2328  0.421451  12  good
chromium@412670  19.4033  0.361467  12  bad

Bisect job ran on: android_nexus5X_perf_bisect
Bug ID: 639065

Test Command: src/tools/perf/run_benchmark -v --browser=android-chromium --output-format=chartjson --upload-results --also-run-disabled-tests media.android.tough_video_cases
Test Metric: cpu_utilization/video.html?src_smpte_3840x2160_60fps_vp9.webm
Relative Change: 1.64%
Score: 0

Buildbot stdio: http://build.chromium.org/p/tryserver.chromium.perf/builders/android_nexus5X_perf_bisect/builds/554
Job details: https://chromeperf.appspot.com/buildbucket_job_status/9003611521356408768


Not what you expected? We'll investigate and get back to you!
  https://chromeperf.appspot.com/bad_bisect?try_job_id=5866898277793792

| 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!
New bisect pulls back the starting revision... all of the good/bad numbers from previous attempts are really "bad". Standby for results.
Project Member

Comment 28 by 42576172...@developer.gserviceaccount.com, Aug 23 2016

Bisect failed: http://build.chromium.org/p/tryserver.chromium.perf/builders/android_nexus5X_perf_bisect/builds/558
Failure reason: the build has failed due to infrastructure failure.

prasadv/robertocn: new bisect failure. Can you help me get this working?
Project Member

Comment 31 by 42576172...@developer.gserviceaccount.com, Aug 23 2016


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


=== Bisection aborted ===
The bisect was aborted because The metric values for the initial "good" and "bad" revisions do not represent a clear regression.
Please contact the the team (see below) if you believe this is in error.

=== Warnings ===
The following warnings were raised by the bisect job:

 * Bisect failed to reproduce the regression with enough confidence.

===== TESTED REVISIONS =====
Revision         Mean     Std Dev   N   Good?
chromium@412392  19.2152  0.33522   12  good
chromium@412724  19.2922  0.364542  12  bad

Bisect job ran on: android_nexus5X_perf_bisect
Bug ID: 639065

Test Command: src/tools/perf/run_benchmark -v --browser=android-chromium --output-format=chartjson --upload-results --also-run-disabled-tests media.android.tough_video_cases
Test Metric: cpu_utilization/video.html?src_smpte_3840x2160_60fps_vp9.webm
Relative Change: 1.38%
Score: 0

Buildbot stdio: http://build.chromium.org/p/tryserver.chromium.perf/builders/android_nexus5X_perf_bisect/builds/561
Job details: https://chromeperf.appspot.com/buildbucket_job_status/9003578112784539888


Not what you expected? We'll investigate and get back to you!
  https://chromeperf.appspot.com/bad_bisect?try_job_id=6401452591808512

| 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!
Pulling start range back even further. Not hopeful. 

Again, many of the graphs associated with this bug have resolved, but a few for cpu utilization are still significantly regressed. My bisects focus on these remaining cases.

prasadv/robertocn: Has the bot/infra received patches recently? The reference graph for this metric hasn't changed, but it may be that the reference build and current build use totally different OS apis - not sure if this is a reasonable explanation. 
Project Member

Comment 34 by 42576172...@developer.gserviceaccount.com, Aug 23 2016


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


=== Bisection aborted ===
The bisect was aborted because The metric values for the initial "good" and "bad" revisions do not represent a clear regression.
Please contact the the team (see below) if you believe this is in error.

=== Warnings ===
The following warnings were raised by the bisect job:

 * Bisect failed to reproduce the regression with enough confidence.

===== TESTED REVISIONS =====
Revision         Mean     Std Dev   N   Good?
chromium@412148  18.9281  0.388502  12  good
chromium@412670  19.3648  0.396707  12  bad

Bisect job ran on: android_nexus5X_perf_bisect
Bug ID: 639065

Test Command: src/tools/perf/run_benchmark -v --browser=android-chromium --output-format=chartjson --upload-results --also-run-disabled-tests media.android.tough_video_cases
Test Metric: cpu_utilization/video.html?src_smpte_3840x2160_60fps_vp9.webm
Relative Change: 2.96%
Score: 0

Buildbot stdio: http://build.chromium.org/p/tryserver.chromium.perf/builders/android_nexus5X_perf_bisect/builds/586
Job details: https://chromeperf.appspot.com/buildbucket_job_status/9003515413635006256


Not what you expected? We'll investigate and get back to you!
  https://chromeperf.appspot.com/bad_bisect?try_job_id=5857205207695360

| 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!
Cc: -pras...@chromium.org
Owner: pras...@chromium.org
prasadv@ - bisect has not been helpful for this regression. Please take a look. Note my details about the particular regression (see Comment 33) - some graphs associated with this bug have recovered. I'm focused on those that remain regressed.

please re-assign to appropriate tooling owner if not you :)
Cc: chcunningham@chromium.org pras...@chromium.org
Labels: BadBisect
Owner: ----
Status: Available (was: Assigned)
Chris, #5 and #7 failed due to this bug 639813, which has been fixed since. I will attempt relaunching those bisects that seemed to be more close to reproducing a regression than those launched after.
Project Member

Comment 38 by 42576172...@developer.gserviceaccount.com, Sep 10 2016

Cc: marpan@chromium.org
Owner: marpan@chromium.org

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

Hi marpan@chromium.org, the bisect results pointed to your CL below as possibly
causing a regression. Please have a look at this info and see whether
your CL be related.


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


===== SUSPECTED CL(s) =====
Subject : Roll src/third_party/libvpx/source/libvpx/ 2d1e63d0c..fe4dd4f43 (13 commits).
Author  : marpan
Commit description:
  
https://chromium.googlesource.com/webm/libvpx.git/+log/2d1e63d0c50c..fe4dd4f43ff6

$ git log 2d1e63d0c..fe4dd4f43 --date=short --no-merges --format='%ad %ae %s'
2016-08-11 jzern variance_impl_avx2: restore table layout
2016-08-03 linfengz NEON intrinsics for 4 loopfilter functions
2016-08-11 yunqingwang Fix another motion vector out of range bug
2016-08-10 marpan vp8: Fix denoiser setting in multi-res sample encoder.
2016-08-09 paulwilkins Modified ARF group allocation.
2016-07-29 margtu-fivt Align thread entry point stack
2016-08-08 jzern tests: use scoped_ptr for local video source vars
2016-08-08 jzern y4m_test: init members in the constructor
2016-07-15 yuryg Add cpi parameter for forcing segmentation update
2016-08-08 aconverse Refactor mv limits.
2016-08-05 jzern *_perf_test.cc: correct DoDecode signature
2016-08-04 noreply test: apply clang-tidy google-readability-braces-around-statements
2016-08-05 yunqingwang Fix a motion vector out of range bug

R=johannkoenig@google.com

Review-Url: https://codereview.chromium.org/2259483002
Cr-Commit-Position: refs/heads/master@{#412644}
Commit  : db041e863ac318d054f13a16885a557c7cc93b5e
Date    : Wed Aug 17 21:33:19 2016


===== TESTED REVISIONS =====
Revision         Mean     Std Dev   N  Good?
chromium@412620  12.8387  1.21217   5  good
chromium@412633  13.4868  1.52297   5  good
chromium@412639  13.503   0.809054  5  good
chromium@412642  12.7857  1.01754   5  good
chromium@412643  13.7466  1.30273   5  good
chromium@412644  19.2652  1.41711   5  bad    <--
chromium@412645  20.538   2.4618    5  bad
chromium@412670  21.8589  2.82197   5  bad

Bisect job ran on: android_nexus5X_perf_bisect
Bug ID: 639065

Test Command: src/tools/perf/run_benchmark -v --browser=android-chromium --output-format=chartjson --upload-results --also-run-disabled-tests media.android.tough_video_cases
Test Metric: cpu_cstate_WFI_residency_percent/video.html?src_smpte_3840x2160_60fps_vp9.webm
Relative Change: 70.26%
Score: 99.9

Buildbot stdio: http://build.chromium.org/p/tryserver.chromium.perf/builders/android_nexus5X_perf_bisect/builds/661
Job details: https://chromeperf.appspot.com/buildbucket_job_status/9001972706948736672


Not what you expected? We'll investigate and get back to you!
  https://chromeperf.appspot.com/bad_bisect?try_job_id=6248374276194304

| 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!
Cc: johannkoenig@chromium.org
Johann: do you see anything in that roll (mentioned in #38) that would cause
the cpu utilization regression on android.
Is a higher number better or worse? In the previous bisects near the top it looked like a high number going to a low number, maybe fps? and the drop is clearly bad.

This change brings in some arm improvements so I find it strange that it is implicated.

Looking at third_party/libvpx I would expect the following (unfortunately I can't find the chromium@N versions in the commit log - there is Cr-Commit-Position, but it is in the 415xxx range, not 412xxx)

I would expect a progression like this:

Speed regression hits with:
https://codereview.chromium.org/2221193003
as I remove the armv6 fallback for halfpix variance in "2016-07-29 johannkoenig Remove armv6 target"

Maybe a tiny speed change with:
https://codereview.chromium.org/2259483002
for "2016-08-03 linfengz NEON intrinsics for 4 loopfilter functions"
and
https://codereview.chromium.org/2266453002
as it picks up loopfilter improvements in "2016-08-12 linfengz NEON asm of vpx_lpf_{horizontal,vertical}_8_dual_neon()"

but those are just replacing existing neon assembly with intrinsics, which should be ~approximately the same speed.

Things should mostly be recovered with:
https://codereview.chromium.org/2295893002
That brings in "2016-08-23 johannkoenig Remove halfpix specialization" which fixes the largest hole in the missing neon optimizations. This is something there *were* armv6/media optimizations for, but they were removed.

There are still two functions missing which I have changes in flight for:
https://chromium-review.googlesource.com/#/c/380311/
https://chromium-review.googlesource.com/#/c/380505/

However, these are 4x4 transforms and should not have a large impact in coding speed in most cases..

marpan/johan, thanks for looking into this. 

> Is a higher number better or worse? In the previous bisects near the top it looked like a high number going to a low number

A higher number is worse. Those early bisects weren't good. The issue here is not that the speed or fps of video playback is regressed, but rather the CPU usage is higher during the playback.


Labels: Performance
Status: Assigned (was: Available)
Status: WontFix (was: Assigned)
This regression has been included in a stable release and that stable channel is now deprecated. I'm closing this so that we won't have unresponsive performance regressions in the future. 

Sign in to add a comment