GPU benchmarks should track the frequency of impl-thread vs. main-thread scrolls

cc/CCRenderingStats.h

Index: cc/CCRenderingStats.h
diff --git a/cc/CCRenderingStats.h b/cc/CCRenderingStats.h
index 0203df3872a0fce38692587bdb9cbfbed120c9e0..fb178a89722fe38fb7fd69a483c6023b305e44c1 100644
--- a/cc/CCRenderingStats.h
+++ b/cc/CCRenderingStats.h
@@ -16,6 +16,8 @@ struct CCRenderingStats {
     double totalRasterizeTimeInSeconds;
     double totalCommitTimeInSeconds;
     size_t totalCommitCount;
+    size_t numImplThreadScrolls;
+    size_t numMainThreadScrolls;
 
     CCRenderingStats()
         : numAnimationFrames(0)
@@ -25,6 +27,8 @@ struct CCRenderingStats {
         , totalRasterizeTimeInSeconds(0)
         , totalCommitTimeInSeconds(0)
         , totalCommitCount(0)
+        , numImplThreadScrolls(0)
+        , numMainThreadScrolls(0)
     {
     }
 };