Bug 23050 - Copying Fonts by value makes GraphicsContext::save() and restore() expensive
Summary: Copying Fonts by value makes GraphicsContext::save() and restore() expensive
Status: RESOLVED WONTFIX
Alias: None
Product: WebKit
Classification: Unclassified
Component: Layout and Rendering (show other bugs)
Version: 528+ (Nightly build)
Hardware: Mac OS X 10.5
: P2 Normal
Assignee: Nobody
URL: https://bugs.webkit.org/attachment.cg...
Keywords:
Depends on:
Blocks:
 
Reported: 2008-12-30 22:12 PST by Simon Fraser (smfr)
Modified: 2009-01-17 18:29 PST (History)
4 users (show)

See Also:

Attachments
Shark session (2.47 MB, application/octet-stream)
2009-01-01 19:08 PST, Simon Fraser (smfr) 		no flags Details
View All Add an attachment (proposed patch, testcase, etc.)

Note You need to log in before you can comment on or make changes to this bug.
Description Simon Fraser (smfr) 2008-12-30 22:12:19 PST 
I have a Shark sample (generated from the testcase in bug 19312) that shows that GraphicsContext::save() and restore() are pretty expensive; in this test. they take 11.7% and 23% of the entire sample (of which 87% is in painting).

Focusing on GraphicsContext::restore(), GraphicsContext::operator= is 65%, including 58% from Font::operator=. Vector::removeLast is 19%, and restorePlatformState() is 14%. Of the Font::operator= cost, almost all is in the GlyphPageTreeNode hash table copy.

Inside GraphicsContext::save(), 62% is in the Font constructor, and 22% is savePlatformState.
Comment 1 Simon Fraser (smfr) 2009-01-01 19:07:21 PST 
Testcase in the URL field.
Comment 2 Simon Fraser (smfr) 2009-01-01 19:08:45 PST 
Created attachment 26358 [details]
Shark session
Comment 3 Simon Fraser (smfr) 2009-01-01 19:10:36 PST 
My initial analysis was done on a debug build (stupid!). In a release build, things aren't nearly as bad. In the attached Shark profile of the testcase, GraphicsContext::restore() is 1.4%, and GraphicsContext::save() 0.6%.

However, Font::operator= is still 39% of the time spent in GraphicsContext::restore(), so I think there's still a win to be had here.
Comment 4 Simon Fraser (smfr) 2009-01-01 19:27:44 PST 
Another good testcase: http://www.bel.fi/~alankila/plasma.html
Comment 5 Simon Fraser (smfr) 2009-01-17 18:29:05 PST 
Obviated by http://trac.webkit.org/changeset/40007