FetchOp Force sync

paddle/fluid/framework/details/fetch_op_handle.cc

@@ -47,9 +47,11 @@ void FetchOpHandle::WaitAndMergeCPUTensors() const {
 }
 
 void FetchOpHandle::RunImpl() {
+  auto cpu_ctx =
+      platform::DeviceContextPool::Instance().Get(platform::CPUPlace());
   for (auto *input : inputs_) {
     auto *var = static_cast<VarHandle *>(input);
-    var->generated_op_->Wait(this->dev_ctx_[var->place_]);
+    var->generated_op_->Wait(cpu_ctx);
   }
 
   tensors_.resize(inputs_.size());

paddle/fluid/framework/details/threaded_ssa_graph_executor.cc

@@ -209,10 +209,6 @@ void ThreadedSSAGraphExecutor::RunOp(
       VLOG(10) << op->DebugString();
       op->Run(use_event_);
 
-      for (auto &dev_ctx : op->dev_ctx_) {
-        dev_ctx.second->Wait();  // Sync error
-      }
-
       for (auto *ready : *ready_buffer) {
         ready->store(true, std::memory_order_release);
       }