Bug 582467. With accelerated layer backends, we should always treat video layers as active to avoid having to read back; no point in saving composition costs on those backends anyway. r=vlad

layout/generic/nsVideoFrame.cpp

@@ -64,6 +64,7 @@
 #include "nsIAccessibilityService.h"
 #endif
 
+using namespace mozilla;
 using namespace mozilla::layers;
 
 nsIFrame*
@@ -379,10 +380,19 @@ public:
   virtual LayerState GetLayerState(nsDisplayListBuilder* aBuilder,
                                    LayerManager* aManager)
   {
+    if (aManager->GetBackendType() != LayerManager::LAYERS_BASIC) {
+      // For non-basic layer managers we can assume that compositing
+      // layers is very cheap, and since ImageLayers don't require
+      // additional memory of the video frames we have to have anyway,
+      // we can't save much by making layers inactive. Also, for many
+      // accelerated layer managers calling
+      // imageContainer->GetCurrentAsSurface can be very expensive. So
+      // just always be active for these managers.
+      return LAYER_ACTIVE;
+    }
     nsHTMLMediaElement* elem =
       static_cast<nsHTMLMediaElement*>(mFrame->GetContent());
-    return elem->IsPotentiallyPlaying() ? mozilla::LAYER_ACTIVE :
-      mozilla::LAYER_INACTIVE;
+    return elem->IsPotentiallyPlaying() ? LAYER_ACTIVE : LAYER_INACTIVE;
   }
 };