Title: [149369] trunk/Source/WebCore
- Revision
- 149369
- Author
- [email protected]
- Date
- 2013-04-30 07:39:56 -0700 (Tue, 30 Apr 2013)
Log Message
[Texmap] Avoid using overlap/non-overlap region in cases where the overhead is bigger than the gain
https://bugs.webkit.org/show_bug.cgi?id=115226
Patch by Noam Rosenthal <[email protected]> on 2013-04-30
Reviewed by Allan Sandfeld Jensen.
When there are several fragmeneted overlap regions, or when the overlap region is bigger than the
non-overlap region, it's more efficient to have a single intermediate surface rather than split
the rendering to overlap/non-overlap.
Covered by tests in compositing/overlap-blending.
* platform/graphics/texmap/TextureMapperLayer.cpp:
(WebCore::TextureMapperLayer::shouldBlend):
Readability fix.
(WebCore::TextureMapperLayer::paintUsingOverlapRegions):
Consolidate to one overlap rect when one of the following occurs:
- The overlap region's area is larger than the non-overlap region area.
- There are many fragmented overlap regions and no non-overlap regions.
Modified Paths
Diff
Modified: trunk/Source/WebCore/ChangeLog (149368 => 149369)
--- trunk/Source/WebCore/ChangeLog 2013-04-30 14:37:25 UTC (rev 149368)
+++ trunk/Source/WebCore/ChangeLog 2013-04-30 14:39:56 UTC (rev 149369)
@@ -1,3 +1,25 @@
+2013-04-30 Noam Rosenthal <[email protected]>
+
+ [Texmap] Avoid using overlap/non-overlap region in cases where the overhead is bigger than the gain
+ https://bugs.webkit.org/show_bug.cgi?id=115226
+
+ Reviewed by Allan Sandfeld Jensen.
+
+ When there are several fragmeneted overlap regions, or when the overlap region is bigger than the
+ non-overlap region, it's more efficient to have a single intermediate surface rather than split
+ the rendering to overlap/non-overlap.
+
+ Covered by tests in compositing/overlap-blending.
+
+ * platform/graphics/texmap/TextureMapperLayer.cpp:
+ (WebCore::TextureMapperLayer::shouldBlend):
+ Readability fix.
+
+ (WebCore::TextureMapperLayer::paintUsingOverlapRegions):
+ Consolidate to one overlap rect when one of the following occurs:
+ - The overlap region's area is larger than the non-overlap region area.
+ - There are many fragmented overlap regions and no non-overlap regions.
+
2013-04-30 Christophe Dumez <[email protected]>
Stop using "in" keyword in IDL files
Modified: trunk/Source/WebCore/platform/graphics/texmap/TextureMapperLayer.cpp (149368 => 149369)
--- trunk/Source/WebCore/platform/graphics/texmap/TextureMapperLayer.cpp 2013-04-30 14:37:25 UTC (rev 149368)
+++ trunk/Source/WebCore/platform/graphics/texmap/TextureMapperLayer.cpp 2013-04-30 14:39:56 UTC (rev 149369)
@@ -197,11 +197,10 @@
if (m_state.preserves3D)
return false;
-#if ENABLE(CSS_FILTERS)
- if (m_currentFilters.size())
- return true;
-#endif
- return m_currentOpacity < 1 || m_state.maskLayer || (m_state.replicaLayer && m_state.replicaLayer->m_state.maskLayer);
+ return m_currentOpacity < 1
+ || hasFilters()
+ || m_state.maskLayer
+ || (m_state.replicaLayer && m_state.replicaLayer->m_state.maskLayer);
}
bool TextureMapperLayer::isVisible() const
@@ -334,8 +333,16 @@
return;
}
+ // Having both overlap and non-overlap regions carries some overhead. Avoid it if the overlap area
+ // is big anyway.
+ if (overlapRegion.bounds().size().area() > nonOverlapRegion.bounds().size().area()) {
+ overlapRegion.unite(nonOverlapRegion);
+ nonOverlapRegion = Region();
+ }
+
nonOverlapRegion.translate(options.offset);
Vector<IntRect> rects = nonOverlapRegion.rects();
+
for (size_t i = 0; i < rects.size(); ++i) {
IntRect rect = rects[i];
if (!rect.intersects(options.textureMapper->clipBounds()))
@@ -347,6 +354,12 @@
}
rects = overlapRegion.rects();
+ static const size_t OverlapRegionConsolidationThreshold = 4;
+ if (nonOverlapRegion.isEmpty() && rects.size() > OverlapRegionConsolidationThreshold) {
+ rects.clear();
+ rects.append(overlapRegion.bounds());
+ }
+
IntSize maxTextureSize = options.textureMapper->maxTextureSize();
IntRect adjustedClipBounds(options.textureMapper->clipBounds());
adjustedClipBounds.move(-options.offset);
_______________________________________________
webkit-changes mailing list
[email protected]
https://lists.webkit.org/mailman/listinfo/webkit-changes
