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

Issue 798351 link

Starred by 1 user

Issue metadata

Status: WontFix
Owner:
Closed: Jan 2018
Cc:
EstimatedDays: ----
NextAction: ----
OS: ----
Pri: 2
Type: Bug-Regression



Sign in to add a comment

15.3% regression in memory.desktop at 525717:525826

Project Member Reported by alexclarke@chromium.org, Jan 2 2018

Issue description

See the link to graphs below.
 
All graphs for this bug:
  https://chromeperf.appspot.com/group_report?bug_id=798351

(For debugging:) Original alerts at time of bug-filing:
  https://chromeperf.appspot.com/group_report?sid=bbeb291cb902d4981f2dfa927494baa61628ae1c722169f0d0adf58c7f10248d


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

chromium-rel-mac12
๐Ÿ“ Pinpoint job started.
https://pinpoint-dot-chromeperf.appspot.com/job/14d59603040000
Cc: herb@google.com mtkl...@google.com
Owner: herb@google.com
Status: Assigned (was: Untriaged)
๐Ÿ“ Found a significant difference after 1 commit.
https://pinpoint-dot-chromeperf.appspot.com/job/14d59603040000

Account for border with small sigma.
By herb@google.com ยท Wed Dec 20 14:17:24 2017 +0000
skia @ dca92e8a622cec7579dd2d44b8b7121603850b93

Understanding performance regressions:
  http://g.co/ChromePerformanceRegressions

Comment 4 by herb@google.com, Jan 2 2018

Status: WontFix (was: Assigned)
I think the graph shows one value getting smaller, and another getting larger. If you sum the two, you end up with overall less memory use.

I think this is working as intended.
Status: Assigned (was: WontFix)
The blue reference line is just the result from the stable build of Chrome (i.e., one without this patch) and is generally only useful for finding unanticipated changes that affect all builds (e.g., OS updates). Ignoring that, the red line does go up which could suggest a memory regression from this change.
Yes, the change in the skia repository could not have affected the ref build which is running against a pinned version of Chrome. I have to assume that drop is due to some other infrastructure issue which just happened to coincide.

On the other hand, the pinpoint results did reproduce pretty clearly the patterns observed on the dashboard for the regular (non-ref) build: namely the average values went up and the metric became noisier.

I believe the highlighted CL did cause a real regression of about 6MiB on average which should be addressed.
herb: can you please take a look at the memory regression (see #5 and 6)? 6MiB on Android is pretty significant.

Comment 8 by herb@google.com, Jan 25 2018

Status: WontFix (was: Assigned)
This is working as intended. Before we incorrectly returned the original with out a required border. But the way the image filters are defined, it is necessary to copy the image, and add a transparent black border. The extra copy is the cost.

Sign in to add a comment