Bug 44014 – [Qt] Path: Fast approximation of stroke bounding rects

• Component: WebCore Misc.
• Status: RESOLVED FIXED

[Andreas Kling]

Path::strokeBoundingRect() is used by CanvasRenderingContext2D to determine the dirty region following a stroke() call.

We currently implement this using QPainterPathStroker which is very expensive for complex paths.

Instead of this we can use QPainterPath::controlPointRect() to approximate the bounding rect. The result will be slightly larger than what the current implementation would return but will run in a fraction of the time.

[Andreas Kling]

Example page that spends a lot of time in Qt's strokeBoundingRect():
http://sebleedelisle.com/demos/html5landscape.html

[Andreas Kling]

Created attachment 64425 [details]
Proposed patch

[Andreas Kling]

(In reply to comment #1)
> Example page that spends a lot of time in Qt's strokeBoundingRect():
> http://sebleedelisle.com/demos/html5landscape.html

Speed improvement for this page with the proposed patch: 18.6%

About the patch, if we want to narrow down the returned rect further, I suppose the lineWidth could be divided by 2.

[Ariya Hidayat]

Comment on attachment 64425 [details]
Proposed patch

Very nice!

[Andreas Kling]

Comment on attachment 64425 [details]
Proposed patch

Clearing flags on attachment: 64425

Committed r65374: <http://trac.webkit.org/changeset/65374>

[Andreas Kling]

All reviewed patches have been landed.  Closing bug.

[WebKit Review Bot]

http://trac.webkit.org/changeset/65374 might have broken Qt Linux Release

[Andreas Kling]

(In reply to comment #7)
> http://trac.webkit.org/changeset/65374 might have broken Qt Linux Release

Gaaahhh, that's what I get for compiling without SVG support..

[Andreas Kling]

http://trac.webkit.org/changeset/65374 broke a bunch of SVG tests. Reopening.

[Andreas Kling]

Created attachment 64609 [details]
Proposed patch v2

Move the (Qt-specific) logic in CRC2D::stroke() this time so as not to affect SVG.

[Ariya Hidayat]

Comment on attachment 64609 [details]
Proposed patch v2

WebCore/html/canvas/CanvasRenderingContext2D.cpp:833
 +          boundingRect.inflate(state().m_miterLimit + state().m_lineWidth);
This means we still ignore the CanvasStrokeStyleApplier, not sure how severe it would be.
A workaround is to expand the boundingRect by the stroke width, sadly this does not take into account miter join.

Clearing the review flag while this is being thought off.

[Ariya Hidayat]

Comment on attachment 64609 [details]
Proposed patch v2

LGTM. re=me

[Andreas Kling]

Comment on attachment 64609 [details]
Proposed patch v2

Clearing flags on attachment: 64609

Committed r65525: <http://trac.webkit.org/changeset/65525>

[Andreas Kling]

All reviewed patches have been landed.  Closing bug.