gpu: Reuse transfer buffers more aggresively

gpu/command_buffer/service/async_pixel_transfer_manager.h

Index: gpu/command_buffer/service/async_pixel_transfer_manager.h
diff --git a/gpu/command_buffer/service/async_pixel_transfer_manager.h b/gpu/command_buffer/service/async_pixel_transfer_manager.h
index d0ea5da153c4cf08cdd292d278702c70732949f2..a1fa64ab1aa430c72be65478eb010d20ccef9386 100644
--- a/gpu/command_buffer/service/async_pixel_transfer_manager.h
+++ b/gpu/command_buffer/service/async_pixel_transfer_manager.h
@@ -63,6 +63,9 @@ class GPU_EXPORT AsyncPixelTransferManager
   virtual void BindCompletedAsyncTransfers() = 0;
 
   // There's no guarantee that callback will run on the caller thread.
+  virtual void AsyncRunWhenCompleted(base::Callback<void()> callback) = 0;

[reveman, 2014/01/22 17:30:04]

Can you drop "When" from this somehow? "Async" prefix already indicate that this
will run asynchronously sometime in the future. Maybe
AsyncRunCompletionCallback()? Also s/base::Callback<void()>/base::Closure/

[epenner, 2014/01/23 02:26:12]

This is very similar to AsyncNotifyCompletion both in how it works and what it
does in this instance. Could merge this somehow? Some thoughts:
- Could we use some kind of token API rather than queries in the client and
ditch the queries?
- Could the queries just record the token in the client and ditch the service
side stuff?

Nit: use 'base::Closure'

+
+  // There's no guarantee that callback will run on the caller thread.
   virtual void AsyncNotifyCompletion(
       const AsyncMemoryParams& mem_params,
       AsyncPixelTransferCompletionObserver* observer) = 0;