Bug 102563 – Add an SVG repaint test

• Status: RESOLVED
• Resolution: WONTFIX
• Component: Tools / Tests

[Florin Malita]

Existing tests don't measure paint performance, adding one to track improvements in https://bugs.webkit.org/show_bug.cgi?id=102272.

[Florin Malita]

Created attachment 174765 [details]
Patch

[Ryosuke Niwa]

Comment on attachment 174765 [details]
Patch

View in context: https://bugs.webkit.org/attachment.cgi?id=174765&action=review

> PerformanceTests/SVG/SvgPainting.html:736
> +          testRunner.display();

display() does much more than just painting.

[Ryosuke Niwa]

Comment on attachment 174765 [details]
Patch

View in context: https://bugs.webkit.org/attachment.cgi?id=174765&action=review

> PerformanceTests/SVG/SvgPainting.html:736
> +          testRunner.display();

display() does much more than just painting.

[Florin Malita]

(In reply to comment #2)
> > PerformanceTests/SVG/SvgPainting.html:736
> > +          testRunner.display();
> 
> display() does much more than just painting.

That's OK as long as paint regressions/progressions are observable in the results (and they are for the parent patch).

Or is there a better way to isolate paint-only times?

[Florin Malita]

Bump.

[Ryosuke Niwa]

Comment on attachment 174765 [details]
Patch

I don't think we should be using testRunner.display() for this.

[Florin Malita]

So the previous question still stands: what is the better way for measuring paint performance?

[Ryosuke Niwa]

(In reply to comment #7)
> So the previous question still stands: what is the better way for measuring paint performance?

See other SVG tests.

[Florin Malita]

(In reply to comment #8)
> (In reply to comment #7)
> > So the previous question still stands: what is the better way for measuring paint performance?
> 
> See other SVG tests.

The existing tests don't paint. Try this:

--- a/Source/WebCore/rendering/svg/RenderSVGShape.cpp
+++ b/Source/WebCore/rendering/svg/RenderSVGShape.cpp
@@ -271,6 +271,7 @@ void RenderSVGShape::fillAndStrokeShape(GraphicsContext* context)
 
 void RenderSVGShape::paint(PaintInfo& paintInfo, const LayoutPoint&)
 {
+    fprintf(stderr, "*** paint\n");
     if (paintInfo.context->paintingDisabled() || style()->visibility() == HIDDEN || isEmpty
         return;
     FloatRect boundingBox = repaintRectInLocalCoordinates();


Nothing comes out unless display() is called.

[Ryosuke Niwa]

(In reply to comment #9)
> (In reply to comment #8)
> > (In reply to comment #7)
> > > So the previous question still stands: what is the better way for measuring paint performance?
> > 
> > See other SVG tests.
> 
> The existing tests don't paint. Try this:
> 
> --- a/Source/WebCore/rendering/svg/RenderSVGShape.cpp
> +++ b/Source/WebCore/rendering/svg/RenderSVGShape.cpp
> @@ -271,6 +271,7 @@ void RenderSVGShape::fillAndStrokeShape(GraphicsContext* context)
> 
>  void RenderSVGShape::paint(PaintInfo& paintInfo, const LayoutPoint&)
>  {
> +    fprintf(stderr, "*** paint\n");
>      if (paintInfo.context->paintingDisabled() || style()->visibility() == HIDDEN || isEmpty
>          return;
>      FloatRect boundingBox = repaintRectInLocalCoordinates();
> 
> 
> Nothing comes out unless display() is called.

Then we don't have a way to test paints. In general, using DRT is a terrible way of testing painting performance. It doesn't even use real graphics context.

[Florin Malita]

(In reply to comment #10)
> Then we don't have a way to test paints. In general, using DRT is a terrible way of testing painting performance. It doesn't even use real graphics context.

OK.

Not sure about other platforms, but CR/DRT seems to use a regular off-screen canvas, and thus paint performance variations are noticeable in display() timings.