cc: Fix the format of GpuMemoryBuffer for SurfaceTexture

cc/trees/layer_tree_settings.h

Index: cc/trees/layer_tree_settings.h
diff --git a/cc/trees/layer_tree_settings.h b/cc/trees/layer_tree_settings.h
index 0653de5210660be25d5f055393db3c77bd913685..837ac8e3b8621c75ff07e588b96299044dff0dc0 100644
--- a/cc/trees/layer_tree_settings.h
+++ b/cc/trees/layer_tree_settings.h
@@ -5,6 +5,8 @@
 #ifndef CC_TREES_LAYER_TREE_SETTINGS_H_
 #define CC_TREES_LAYER_TREE_SETTINGS_H_
 
+#include <vector>
+
 #include "base/basictypes.h"
 #include "cc/base/cc_export.h"
 #include "cc/debug/layer_tree_debug_state.h"
@@ -77,7 +79,8 @@ class CC_EXPORT LayerTreeSettings {
   bool use_zero_copy;
   bool use_persistent_map_for_gpu_memory_buffers;
   bool enable_elastic_overscroll;
-  unsigned use_image_texture_target;
+  // an array of image texture targets for each GpuMemoryBuffer format

[reveman, 2015/07/28 08:58:26]

nit: s/an array../An array/ and s/format/format./

+  std::vector<unsigned> use_image_texture_targets;

[piman, 2015/07/28 16:33:22]

Why not an array? The size is known statically.

[jchen10, 2015/07/29 04:33:47]

I had wished to use an array from the start. But gave up it in order to keep the
patch more concise, since vector is handier than raw array. The initial thoughts
at that moment were:
1. The initializer function vector(count, value) could be used frequently in the
patch, so the code to re-invent it for array could be saved.
2. Vector can be passed as an parameter either by value or reference. Whereas
you have to manually copy all elements for an array in case of pass-by-value.
3. No need to include "ui/gfx/gpu_memory_buffer.h" everywhere merely for the
array size defined in it.
4. Performance penalty should be insignificant because there are a very small
number of elements.

   bool ignore_root_layer_flings;
   size_t scheduled_raster_task_limit;
   bool use_occlusion_for_tile_prioritization;