cc: early return if unoccluded_region_in_target_surface is empty

cc/trees/occlusion_tracker.cc

 // Copyright 2012 The Chromium Authors. All rights reserved.
 // Use of this source code is governed by a BSD-style license that can be
 // found in the LICENSE file.
 
 #include "cc/trees/occlusion_tracker.h"
 
 #include <algorithm>
 
 #include "cc/base/math_util.h"
 #include "cc/layers/layer.h"
@@ skipping 549 matching lines @@
 
   // Take the ToEnclosingRect at each step, as we want to contain any unoccluded
   // partial pixels in the resulting Rect.
   Region unoccluded_region_in_target_surface =
       MathUtil::MapEnclosingClippedRect(draw_transform, content_rect);
   unoccluded_region_in_target_surface.Subtract(
       stack_.back().occlusion_from_inside_target);
   unoccluded_region_in_target_surface.Subtract(
       stack_.back().occlusion_from_outside_target);
 
+  if (unoccluded_region_in_target_surface.IsEmpty())

[danakj, 2014/07/14 15:14:53]

ProjectEnclosingClippedRect and Intersect both have early outs that would be hit
in this case. Is adding another early out here a win? Did you measure this in
some way?

[hyunki, 2014/07/14 15:29:34]

I think Intersect have early out.
However, when I checked ProjectEnclosingClippedRect, it checks transform only 
and doesn't check source rect (unoccluded_rect_in_target_surface), thus I think
it doesn't out early.   

[danakj, 2014/07/14 16:00:08]

Oh, right sorry. Ok sounds good, does this have any impact on the
occlusion_tracker_perftests?

[hyunki, 2014/07/14 16:25:09]

When I added logs, I've checked it falls early outs for many cases. (perftest
and unittest both)
perftest doesn't give big improvements on my machine as belows.

--> Before
[1/2] OcclusionTrackerPerfTest.UnoccludedContentRect_FullyOccluded (2005 ms)
[2/2] OcclusionTrackerPerfTest.UnoccludedContentRect_10OpaqueLayers (2005 ms)
--> After
[1/2] OcclusionTrackerPerfTest.UnoccludedContentRect_FullyOccluded (1999 ms)
[2/2] OcclusionTrackerPerfTest.UnoccludedContentRect_10OpaqueLayers (2000 ms)

[danakj, 2014/07/14 16:26:02]

It's not testing this case, so I would expect no change or a regression. No
change is good.

+    return gfx::Rect();
+
   gfx::Rect unoccluded_rect_in_target_surface =
       unoccluded_region_in_target_surface.bounds();
   gfx::Rect unoccluded_rect = MathUtil::ProjectEnclosingClippedRect(
       inverse_draw_transform, unoccluded_rect_in_target_surface);
   unoccluded_rect.Intersect(content_rect);
 
   return unoccluded_rect;
 }
 
 template <typename LayerType>
@@ skipping 21 matching lines @@
     return content_rect;
 
   // Take the ToEnclosingRect at each step, as we want to contain any unoccluded
   // partial pixels in the resulting Rect.
   Region unoccluded_region_in_target_surface =
       MathUtil::MapEnclosingClippedRect(draw_transform, content_rect);
   unoccluded_region_in_target_surface.Subtract(
       second_last.occlusion_from_inside_target);
   unoccluded_region_in_target_surface.Subtract(
       second_last.occlusion_from_outside_target);
 

[danakj, 2014/07/14 16:00:08]

How about here?

[hyunki, 2014/07/14 16:25:09]

Done.

   gfx::Rect unoccluded_rect_in_target_surface =
       unoccluded_region_in_target_surface.bounds();
   gfx::Rect unoccluded_rect = MathUtil::ProjectEnclosingClippedRect(
       inverse_draw_transform, unoccluded_rect_in_target_surface);
   unoccluded_rect.Intersect(content_rect);
 
   return unoccluded_rect;
 }
 
 // Instantiate (and export) templates here for the linker.
 template class OcclusionTracker<Layer>;
 template class OcclusionTracker<LayerImpl>;
 
 }  // namespace cc