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

Issue 176612 link

Starred by 5 users

Issue metadata

Status: WontFix
Owner:
Closed: Feb 2013
Cc:
Components:
EstimatedDays: ----
NextAction: ----
OS: ----
Pri: 2
Type: Bug-Regression



Sign in to add a comment

Perf regression on Windows page cyclers

Project Member Reported by simonjam@chromium.org, Feb 15 2013

Issue description

I notice two small perf regressions on the page cyclers from this morning. Both are Windows only and the same small revision range.

http://chromium-perf.appspot.com/?tab=chromium-rel-win7-dual&graph=times&trace=t&rev=182970&history=150&master=ChromiumPerf&testSuite=page_cycler_intl2&details=true

...is just a typical page cycler going slower.

http://chromium-perf.appspot.com/?tab=chromium-rel-win7-gpu-ati&graph=megapixels_painted_per_second&trace=megapixels_painted_per_second&rev=182855&history=150&master=ChromiumPerf&testSuite=scrolling_benchmark&details=true

...says we're drawing fewer pixels than before. I'm not exactly sure how that's measured. It could be that a slower load means more time not painting anything.

Any thoughts?
 

Comment 1 by wtc@chromium.org, Feb 16 2013

simonjam: I guess you cc'ed me because my CL
https://codereview.chromium.org/12259038/ falls within the revision range?

My CL contains two changes:

1. A 64-bit Windows build system change. I think this one can be ruled out
because the page cycler should be running a 32-bit Windows build.

2. An SSL change that affects 32-bit x86 builds on all platforms. That
change is bigger on Windows. Does the page cycler do HTTPS a lot?
I believe the cyclers use HTTP. Looks like that one's out.

FTR, the revision range is:

http://build.chromium.org/f/chromium/perf/dashboard/ui/changelog.html?url=/trunk/src&mode=html&range=182721:182726

Comment 4 by kevers@chromium.org, Feb 19 2013

The scrollbar change at the end of the suspect rev list (https://codereview.chromium.org/12256011/) only affects layout on unthemed Windows (e.g. Windows Classic).  Prior to the change, unthemed scrollbars reported having 0 thickness.  I don't see how this could have affected perf.
Cc: -wtc@chromium.org
Wasn't the NSS change.
Cc: -kevers@chromium.org
Owner: kevers@chromium.org
Definitely the scrollbar change:

http://chromium-perf.appspot.com/?tab=chromium-rel-win7-dual&graph=times&trace=t&rev=-1&history=150&master=ChromiumPerf&testSuite=page_cycler_intl2&details=true

Please fix the perf regression and then reland.
Blocking: chromium:175335
Project Member

Comment 9 by bugdroid1@chromium.org, Feb 20 2013

The following revision refers to this bug:
    http://src.chromium.org/viewvc/chrome?view=rev&revision=183626

------------------------------------------------------------------------
r183626 | simonjam@chromium.org | 2013-02-20T21:48:37.845432Z

Changed paths:
   M http://src.chromium.org/viewvc/chrome/trunk/src/tools/perf_expectations/perf_expectations.json?r1=183626&r2=183625&pathrev=183626

Revert perf expectation change for win7 intl2.

The culprit has been identified and reverted. We should return to
expecting the old values.

BUG= 176612 
TBR=thakis

Review URL: https://codereview.chromium.org/12316014
------------------------------------------------------------------------
Cc: dharani@chromium.org sky@chromium.org girard@chromium.org abarth@chromium.org
Status: WontFix
The regression in perf is due to restoring proper layout when not using a themed scrollbar.  In WebKit r141652 a change was made to ask the theme engine for the scrollbar thickness.  The reason for the change is that the scrollbar thickness was previously being computed in pixels and not in DIP, and was therefore not compatible with high-DPI mode.  As a result of this change, scrollbars disappeared when not using a theme.  Perf "improved" for the simple reason that layout is faster when sizes don't change due to additional of the scrollbars.  Fixing to restore the scrollbar thickness, causes the page cycle time to return to expected values in line with numbers before the initial regression was introduced. 



Blocking: -chromium:175335
Project Member

Comment 12 by bugdroid1@chromium.org, Mar 9 2013

Labels: -Type-Regression -Area-Internals -Stability-Performance Type-Bug-Regression Cr-Internals Performance

Sign in to add a comment