Don't access DisplayItemClient::VisualRect() for cached display items.

third_party/WebKit/Source/platform/graphics/ContentLayerDelegate.cpp

 /*
  * Copyright (C) 2012 Google Inc. All rights reserved.
  *
  * Redistribution and use in source and binary forms, with or without
  * modification, are permitted provided that the following conditions
  * are met:
  * 1.  Redistributions of source code must retain the above copyright
  *     notice, this list of conditions and the following disclaimer.
  * 2.  Redistributions in binary form must reproduce the above copyright
  *     notice, this list of conditions and the following disclaimer in the
@@ skipping 72 matching lines @@
     disabled_mode = GraphicsContext::kFullyDisabled;
 
   // Anything other than PaintDefaultBehavior is for testing. In non-testing
   // scenarios, it is an error to call GraphicsLayer::paint. Actual painting
   // occurs in FrameView::paintTree(); this method merely copies the painted
   // output to the WebDisplayItemList.
   if (painting_control != kPaintDefaultBehavior)
     graphics_layer_->Paint(nullptr, disabled_mode);
 
   paint_controller.GetPaintArtifact().AppendToWebDisplayItemList(
+      graphics_layer_->OffsetFromLayoutObjectWithSubpixelAccumulation(),

[wkorman, 2017/05/19 00:55:08]

Does this add some small overhead vs. ToT, as (I think) we used to only do this
once per new item and save thereafter, whereas now we do it each time we append
regardless of old or new?

Seems ok but figured worth checkpoint as part of the review.

[Xianzhu, 2017/05/19 01:25:11]

Good catch.

The overhead happens when the GraphicsLayer has nothing invalidated, and we
reuse the whole cached DisplayItemList. I think we can optimize the case by also
reusing the original WebDisplayItemList instead of reconstructing it from the
cached DisplayItemList.

If anything is invalidated, regardless of this patch, we re-apply the offset for
each cached or new display item. Previously we called
DisplayItemClient::VisualRect() and now we call DisplayItem::VisualRect()
instead. The latter might be slightly faster because the call is inlined instead
of virtual.

Let's see how this affects performance tests.

       web_display_item_list);
 
   paint_controller.SetDisplayItemConstructionIsDisabled(false);
   paint_controller.SetSubsequenceCachingIsDisabled(false);
 }
 
 size_t ContentLayerDelegate::ApproximateUnsharedMemoryUsage() const {
   return graphics_layer_->GetPaintController().ApproximateUnsharedMemoryUsage();
 }
 
 }  // namespace blink