cc: Avoid extra flush calls to Skia GrContext

cc/raster/gpu_rasterizer.cc

 // Copyright 2015 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/raster/gpu_rasterizer.h"
 
 #include <algorithm>
 
 #include "base/bind.h"
 #include "base/metrics/histogram.h"
@@ skipping 61 matching lines @@
 
     SkSurface* sk_surface = write_lock->sk_surface();
 
     // Allocating an SkSurface will fail after a lost context.  Pretend we
     // rasterized, as the contents of the resource don't matter anymore.
     if (!sk_surface)
       return;
 
     SkMultiPictureDraw multi_picture_draw;
     multi_picture_draw.add(sk_surface->getCanvas(), picture.get());
-    multi_picture_draw.draw(msaa_sample_count_ > 0);
+    // This is undocumented:

[bsalomon, 2016/02/01 14:55:59]

What is undocumented? Are we missing something on the Skia side?

+    // SkCanvas::flush must be explicitly called in order to flush possibly
+    // deferred rendering in SkGpuDevice.
+    // This also prepares the GrRenderTarget of the sk_surface for external
+    // read.
+    // Correct point of preparing would be in
+    // ScopedWriteLockGr::ReleaseSkSurface, but this would be redundant, as
+    // SkGpuDevice::flush forces it.
+    // Passing true SkMultiPictureDraw::draw(bool flush) makes this happen.
+    multi_picture_draw.draw(true);
     write_lock->ReleaseSkSurface();
   }
 }
 
 }  // namespace cc