Bug 32122 – [Qt] Preparation for tiling support

• Status: RESOLVED
• Resolution: DUPLICATE
• Component: WebKit Qt

[Kenneth Rohde Christiansen]

Tiling has the ability to improve the scrolling performance and experience for the two Qt web views.

We will use this bug to track and discuss this work.

[Kenneth Rohde Christiansen]

Created attachment 44249 [details]
Split out the renderPrivate method to two, one for absolute coords, and one relative to the viewport

[WebKit Review Bot]

style-queue ran check-webkit-style on attachment 44249 [details] without any errors.

[Kenneth Rohde Christiansen]

Created attachment 44251 [details]
Receive invalidate events for areas changed outside the viewport, if ScrollView::setPaintsEntireContents is true

[WebKit Review Bot]

style-queue ran check-webkit-style on attachment 44251 [details] without any errors.

[Kenneth Rohde Christiansen]

Created attachment 44253 [details]
Add tiling to QGraphicsWebView

[Kenneth Rohde Christiansen]

Comment on attachment 44251 [details]
Receive invalidate events for areas changed outside the viewport, if ScrollView::setPaintsEntireContents is true

We might want to do this at the frame level instead, removing r?

[Kenneth Rohde Christiansen]

Benjamin did a simple test (can you upload it Benjamin?) that automatically scrolls on offline webpages.

Here are the preliminary results:

QWebView (not tiles, xcopyregion):

RESULT : Test::benchWebView():"simplest painting":              
     2.1 msec per iteration (total: 35, iterations: 16)          
RESULT : Test::benchWebView():"slashdot page":                  
     16 msec per iteration (total: 33, iterations: 2)            
RESULT : Test::benchWebView():"amazon book page":                
     12 msec per iteration (total: 24, iterations: 2)            
RESULT : Test::benchWebView():"wikipedia Qt page":              
     12 msec per iteration (total: 24, iterations: 2)            

QGraphicsWebView (no tiling, should fall back to x copy region as it uses GraphicsView infrastructure for scrolling):

RESULT : Test::benchGraphicsWebView():"simplest painting":      
     8.8 msec per iteration (total: 71, iterations: 8)          
RESULT : Test::benchGraphicsWebView():"slashdot page":          
     100 msec per iteration (total: 100, iterations: 1)          
RESULT : Test::benchGraphicsWebView():"amazon book page":        
     88 msec per iteration (total: 88, iterations: 1)            
RESULT : Test::benchGraphicsWebView():"wikipedia Qt page":      
     137 msec per iteration (total: 137, iterations: 1)           

QGraphicsWebView (with my above tiling patch):

RESULT : Test::benchGraphicsWebView():"simplest painting":
     2.6 msec per iteration (total: 42, iterations: 16)
RESULT : Test::benchGraphicsWebView():"slashdot page":
     1.2 msec per iteration (total: 41, iterations: 32)
RESULT : Test::benchGraphicsWebView():"amazon book page":
     0.71 msec per iteration (total: 23, iterations: 32)
RESULT : Test::benchGraphicsWebView():"wikipedia Qt page":
     0.84 msec per iteration (total: 27, iterations: 32)

[Antti Koivisto]

Looks good, r=me

[WebKit Commit Bot]

Comment on attachment 44249 [details]
Split out the renderPrivate method to two, one for absolute coords, and one relative to the viewport

Clearing flags on attachment: 44249

Committed r51698: <http://trac.webkit.org/changeset/51698>

[WebKit Commit Bot]

All reviewed patches have been landed.  Closing bug.

[Kenneth Rohde Christiansen]

Reopening, as the bug report has uncommitted patches who are not yet up for review.

[Benjamin Poulain]

Created attachment 44395 [details]
Basic benchmark for the scrolling support

Basic benchmark used for testing the tiling support on real use cases.
A proper benchmark needs to be created for the integration in the source.

[Benjamin Poulain]

Created attachment 44396 [details]
Offscreen rendering of the mainframe at the QWebFrame level

This patch render the frame offscreen on a pixmap and always paint the pixmap in paint event.

The advantage of storing the pixmap at the frame level are:
 - improve the speed of graphics web view && qweb view
 - we can avoid completely GraphicsContext in some cases
 - we can have support for multi-frame

This is preliminary work, not meant to be integrated. It does not yet integrate the work of Kenneth

[Kenneth Rohde Christiansen]

Created attachment 44870 [details]
Slides from a presentation about the work

[Antti Koivisto]

The work for this is going on here: http://gitorious.org/~akoivisto/webkit/anttis-webkit/commits/tile-cache

[Simon Hausmann]

Work is now proceeding in 35146 as far as I can see.

*** This bug has been marked as a duplicate of bug 35146 ***