follow comments

paddle/fluid/framework/details/fetch_op_handle.cc

@@ -49,7 +49,9 @@ void FetchOpHandle::RunImpl() {
       platform::DeviceContextPool::Instance().Get(platform::CPUPlace());
   for (auto *input : inputs_) {
     auto *var = static_cast<VarHandle *>(input);
-    if (var->generated_op_) var->generated_op_->Wait(cpu_ctx);
+    if (var->generated_op_) {
+      var->generated_op_->Wait(cpu_ctx);
+    }
   }
   tensors_.resize(inputs_.size());
   auto *var_handle = static_cast<VarHandle *>(inputs_[0]);
@@ -61,14 +63,9 @@ void FetchOpHandle::RunImpl() {
     auto &scope = scopes[i];
     auto *var =
         scope->FindVar(kLocalExecScopeName)->Get<Scope *>()->FindVar(var_name);
-    if (var == nullptr) {
-      scope->FindVar(var_name);
-    }
-
     PADDLE_ENFORCE_NOT_NULL(var, "Cannot find variable %s in execution scope",
                             var_name);
     auto &t = var->Get<framework::LoDTensor>();
-
     if (platform::is_gpu_place(t.place())) {
 #ifdef PADDLE_WITH_CUDA
       TensorCopySync(t, cpu, &tensors_[i]);

paddle/fluid/framework/details/nccl_all_reduce_op_handle.cc

@@ -36,7 +36,9 @@ void NCCLAllReduceOpHandle::RunImpl() {
     // Wait input done
     for (auto *in : inputs_) {
       auto &p = static_cast<VarHandle *>(in)->place_;
-      if (in->generated_op_) in->generated_op_->Wait(dev_ctxes_[p]);
+      if (in->generated_op_) {
+        in->generated_op_->Wait(dev_ctxes_[p]);
+      }
     }
 
     auto &var_name = static_cast<VarHandle *>(this->inputs_[0])->name_;

paddle/fluid/framework/details/send_op_handle.cc

@@ -32,7 +32,9 @@ void SendOpHandle::RunImpl() {
     if (in->DebugString() == "dummy") {  // HACK
       continue;
     }
-    if (in->generated_op_) in->generated_op_->Wait(dev_ctxes_[p]);
+    if (in->generated_op_) {
+      in->generated_op_->Wait(dev_ctxes_[p]);
+    }
   }
   auto &tmp_scope = local_scope_->FindVar(kLocalExecScopeName)->Get<Scope *>();
   // FIXME(wuyi): can not use RunAndRecordEvent here, for it will cause dead