diff --git a/paddle/fluid/framework/details/threaded_ssa_graph_executor.cc b/paddle/fluid/framework/details/threaded_ssa_graph_executor.cc
index 0bf05c3c112c467b9cdae091e0d2d38d4d299655..fc8403155625fdf21c467aa189c65b14d3487b41 100644
--- a/paddle/fluid/framework/details/threaded_ssa_graph_executor.cc
+++ b/paddle/fluid/framework/details/threaded_ssa_graph_executor.cc
@@ -178,10 +178,7 @@ void ThreadedSSAGraphExecutor::RunOp(
     try {
       VLOG(10) << op->Name() << " : " << op->DebugString();
       op->Run(use_event_);
-
-      for (auto &each : op->outputs_) {
-        ready_var_q.Push(each);
-      }
+      ready_var_q.Extend(op->outputs_);
     } catch (platform::EnforceNotMet ex) {
       exception_.reset(new platform::EnforceNotMet(ex));
     } catch (...) {
diff --git a/paddle/fluid/framework/details/threaded_ssa_graph_executor.h b/paddle/fluid/framework/details/threaded_ssa_graph_executor.h
index 26ff14786397933382795c1371e6ab68185b0abe..839217031145a276b25992e334657608c3637758 100644
--- a/paddle/fluid/framework/details/threaded_ssa_graph_executor.h
+++ b/paddle/fluid/framework/details/threaded_ssa_graph_executor.h
@@ -35,6 +35,17 @@ class BlockingQueue {
     cv_.notify_one();
   }
 
+  template <typename U>
+  void Extend(const U &items) {
+    {
+      std::lock_guard<std::mutex> g(mutex_);
+      for (auto &item : items) {
+        q_.emplace_back(item);
+      }
+    }
+    cv_.notify_all();
+  }
+
   T Pop() {
     std::unique_lock<std::mutex> lock(mutex_);
     while (q_.empty()) {