GPU process clients can perform a "shallow" finish.

cc/trees/thread_proxy.cc

Index: cc/trees/thread_proxy.cc
===================================================================
--- cc/trees/thread_proxy.cc	(revision 207109)
+++ cc/trees/thread_proxy.cc	(working copy)
@@ -1215,8 +1215,9 @@
 void ThreadProxy::FinishGLOnImplThread(CompletionEvent* completion) {
   TRACE_EVENT0("cc", "ThreadProxy::FinishGLOnImplThread");
   DCHECK(IsImplThread());
-  if (layer_tree_host_impl_->resource_provider())
-    layer_tree_host_impl_->resource_provider()->Finish();
+  // Can this be a flush?

[piman, 2013/06/21 02:31:10]

I think this can be a shallow finish. The idea here is to ensure the
asynchronous readpixels have a chance to complete. See caveat in gl_renderer
though.

+  //if (layer_tree_host_impl_->resource_provider())
+  //  layer_tree_host_impl_->resource_provider()->Finish();
   completion->Signal();
 }