GPU: Adding sync points for cross-channel synchronization

gpu/ipc/command_buffer_proxy.h

Index: gpu/ipc/command_buffer_proxy.h
diff --git a/gpu/ipc/command_buffer_proxy.h b/gpu/ipc/command_buffer_proxy.h
index 4f53fb0fbff1d422efd9da78eca42e1f39123843..3907a0ec83bdb327c13252d8893975a91caa601c 100644
--- a/gpu/ipc/command_buffer_proxy.h
+++ b/gpu/ipc/command_buffer_proxy.h
@@ -37,6 +37,17 @@ class GPU_EXPORT CommandBufferProxy : public gpu::CommandBuffer {
   virtual bool DiscardBackbuffer() = 0;
   virtual bool EnsureBackbuffer() = 0;
 
+  // Inserts a sync point, returning its ID. This is handled on the IO thread,

[apatrick_chromium, 2012/06/05 18:44:15]

nit: IO thread of the GPU process

[piman, 2012/06/05 20:09:50]

Done.

+  // and so should be relatively fast, but its effect is ordered wrt other
+  // messages (in particular, Flush). Sync point IDs are global and can be used
+  // for cross-channel synchronization.
+  virtual uint32 InsertSyncPoint() = 0;
+
+  // Makes this command buffer wait on a sync point. This command buffer will be
+  // unscheduled until the command buffer that inserted that sync point reaches
+  // it, or gets destroyed.
+  virtual void WaitSyncPoint(uint32) = 0;
+
   // Register a callback to invoke whenever we recieve a new memory allocation.
   virtual void SetMemoryAllocationChangedCallback(
       const base::Callback<void(const GpuMemoryAllocationForRenderer&)>&