Add a background timing thread to kilobench

include/gpu/gl/SkGLContext.h

Index: include/gpu/gl/SkGLContext.h
diff --git a/include/gpu/gl/SkGLContext.h b/include/gpu/gl/SkGLContext.h
index 63c3e9ca7ba515836ac2bd6871f6deb30754f184..dac4d778436b54fc2ccb002669f789eaee0bbe5a 100644
--- a/include/gpu/gl/SkGLContext.h
+++ b/include/gpu/gl/SkGLContext.h
@@ -65,7 +65,7 @@ public:
      * not perform some sort of synchronization, depending on whether the
      * drawing surface provided by the platform is double buffered.
      */
-    void swapBuffers();
+    void swapBuffers(bool usePlatformSwapBuffers = false);

[bsalomon, 2016/01/20 18:54:52]

Can we just undo the swap-as-fence-sync hack before landing this?

Maybe give tools that are relying on this method an alternative
(waitOnSyncOrSwap()?) that  has the old behavior?

 
     /**
      * This notifies the context that we are deliberately testing abandoning
@@ -83,6 +83,11 @@ public:
 
     class GLFenceSync;  // SkGpuFenceSync implementation that uses the OpenGL functionality.
 
+    /*
+     * returns the fencesync object owned by this SkGLContext
+     */
+    SkGpuFenceSync* fenceSync() { return fFenceSync.get(); }
+
 protected:
     SkGLContext();