cc: Avoid extra flush calls to Skia GrContext

cc/raster/gpu_rasterizer.cc

Index: cc/raster/gpu_rasterizer.cc
diff --git a/cc/raster/gpu_rasterizer.cc b/cc/raster/gpu_rasterizer.cc
index 9525d8f9f2220e8e5085b17f9ddb7d30bb848f29..4b385a55450595492213fa479f5208ba8c41c916 100644
--- a/cc/raster/gpu_rasterizer.cc
+++ b/cc/raster/gpu_rasterizer.cc
@@ -79,7 +79,7 @@ void GpuRasterizer::RasterizeSource(
 
     SkMultiPictureDraw multi_picture_draw;
     multi_picture_draw.add(sk_surface->getCanvas(), picture.get());
-    multi_picture_draw.draw(msaa_sample_count_ > 0);
+    multi_picture_draw.draw(true);

[Stephen White, 2016/02/02 17:00:01]

This means we'll be calling SkCanvas::flush() on each multi_picture_draw tile,
even in the non-MSAA case. Brian, does that seem acceptable to you?

[ericrk, 2016/02/08 18:08:19]

We were already flushing due to the ScopedGpuRaster - we'd call flush when this
scope exited.

On the other hand, do we need to flush here before releasing the surface via the
write_lock? or can we just pass false here, and handle the flush via the
scoped_gpu_raster?

It seems like we may be able to achieve the same effect by just changing this to
false and not modifying gl_rasterizer.cc or scoped_gpu_raster.cc - I might
prefer that.

     write_lock->ReleaseSkSurface();
   }
 }