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

Issue 771862 link

Starred by 1 user

Issue metadata

Status: Fixed
Owner:
Closed: Oct 2017
Cc:
EstimatedDays: ----
NextAction: ----
OS: ----
Pri: 1
Type: Bug-Regression



Sign in to add a comment

video_VDAPerf: "delivery_time" regression between 9901.32.0 and 9968.0.0 on eve

Project Member Reported by hiroh@chromium.org, Oct 5 2017

Issue description

"delivery_time" metrics got worse between 62.9901.32.0 and 62.9968.0.0 on eve. The same regression happened on other x86 boards (e.g. poppy).
The regression between 9878.0.0 and 9879.0.0 is already filed.  crbug.com/771860 
https://chromeperf.appspot.com/report?sid=77971c40a22811935dedea6d36eaef568f33cc8a1ad9601d020752eec7c5f5e8

This regression was perhaps happened between 63.9957.0.0 and 63.9968.0.0 according to the graph for poppy?
https://chromeperf.appspot.com/report?sid=f55388bae5d3adc19b46258c429494921c2684c0b296a6863ab8ce2db56baa0a

 

Comment 1 by hiroh@chromium.org, Oct 5 2017

Cc: owenlin@chromium.org
crrev.com/c/654459 is the root cause of this regression, similar to  crbug.com/770596 .
Please see https://bugs.chromium.org/p/chromium/issues/detail?id=770596#c17 too.
hiroh, I think you can probably close this bug as per https://bugs.chromium.org/p/chromium/issues/detail?id=770596#c18.

Comment 3 by hiroh@chromium.org, Oct 5 2017

https://bugs.chromium.org/p/chromium/issues/detail?id=770596#c18 seems to not make sense to this regression.
I am concerned what Owen said in crbug.com/771919#c4.
So I still think render helper CL (crrev.com/c/654459) maybe contains some bug.

Comment 4 Deleted

Comment 5 by hiroh@chromium.org, Oct 6 2017

I attach the data before and after render helper CL (precisely, before = Chrome 62.0.3222.0 reverting render CL, and after = Chrome 62.0.3222.0)

9964_VDAPerf_h264-results-chart.json
19.7 KB View Download
9964_VDAPerf_h264_crowd1080_h264_no_rendering_test_output.log
9.8 KB View Download
9964_VDAPerf_h264_without_renderCL-results-char.json
19.7 KB View Download
9964_VDAPerf_h264_without_renderCL_crowd1080_h264_no_rendering_test_output.log
9.8 KB View Download

Comment 6 by hiroh@chromium.org, Oct 8 2017

According to the CL commit message, the CL only disables rendering.
So the CL shouldn't affect "delivery_time" at all, because it is computed from "delivery_time" with --disable-rendering.
Seeing the "delivery_time" outputs, the half of them got worse, although the half of faster ones weren't change so much.
I think something goes wrong due to the CL.
owenlin@, can you investigate this issue? It would take less time than I will do.

Comment 7 by hiroh@chromium.org, Oct 11 2017

I reverted the bad commit in libva-intel-driver. 
https://chromium-review.googlesource.com/c/chromiumos/overlays/chromiumos-overlay/+/710734
As far as I tested, looks like it resolves this regression.
Let us wait and see the metrics measured after the CL will be merged.

Comment 8 by hiroh@chromium.org, Oct 16 2017

Looks like the regressions has been fixed since the above CL was merged.
Enough data are not shown yet though.
Let us close this issue, and reopen if it is not really fixed. 
https://chromeperf.appspot.com/report?sid=08378ca2ee7f799071ffd123aefe13f3c5910d1bc2133632acf470a19440a2f3

Comment 9 by hiroh@chromium.org, Oct 16 2017

Status: Fixed (was: Untriaged)

Sign in to add a comment