replace use_event with use_cuda, because use_event means the program running...

replace use_event with use_cuda, because use_event means the program running with CUDA, so use_cuda maybe more intuitive.

paddle/fluid/framework/details/execution_strategy.h

@@ -20,7 +20,7 @@ namespace details {
 
 struct ExecutionStrategy {
   size_t num_threads_{0};
-  bool use_event_{true};
+  bool use_cuda_{true};
   bool allow_op_delay_{false};
   size_t num_iteration_per_drop_scope_{100};
 };

paddle/fluid/framework/details/op_handle_base.cc

@@ -39,9 +39,9 @@ OpHandleBase::~OpHandleBase() {
 #endif
 }
 
-void OpHandleBase::Run(bool use_event) {
+void OpHandleBase::Run(bool use_cuda) {
 #ifdef PADDLE_WITH_CUDA
-  if (events_.empty() && use_event) {
+  if (events_.empty() && use_cuda) {
     for (auto &p : dev_ctxes_) {
       int dev_id = boost::get<platform::CUDAPlace>(p.first).device;
       PADDLE_ENFORCE(cudaSetDevice(dev_id));
@@ -50,7 +50,7 @@ void OpHandleBase::Run(bool use_event) {
     }
   }
 #else
-  PADDLE_ENFORCE(!use_event);
+  PADDLE_ENFORCE(!use_cuda);
 #endif
 
   RunImpl();

paddle/fluid/framework/details/op_handle_base.h

@@ -36,7 +36,7 @@ class OpHandleBase {
 
   virtual std::string Name() const = 0;
 
-  void Run(bool use_event);
+  void Run(bool use_cuda);
 
   virtual void RecordWaitEventOnCtx(platform::DeviceContext *waited_ctx);
 

paddle/fluid/framework/details/threaded_ssa_graph_executor.cc

@@ -192,7 +192,7 @@ void ThreadedSSAGraphExecutor::RunOp(
       if (VLOG_IS_ON(10)) {
         VLOG(10) << op << " " << op->Name() << " : " << op->DebugString();
       }
-      op->Run(strategy_.use_event_);
+      op->Run(strategy_.use_cuda_);
       VLOG(10) << op << " " << op->Name() << " Done ";
       running_ops_--;
       ready_var_q->Extend(op->Outputs());

paddle/fluid/framework/parallel_executor.cc

@@ -61,7 +61,7 @@ ParallelExecutor::ParallelExecutor(
     size_t num_trainers, size_t trainer_id)
     : member_(new ParallelExecutorPrivate(places)) {
   member_->global_scope_ = scope;
-  member_->use_cuda_ = exec_strategy.use_event_;
+  member_->use_cuda_ = exec_strategy.use_cuda_;
 
   // Step 1. Bcast the params to devs.
   // Create local scopes

paddle/fluid/pybind/pybind.cc

@@ -509,10 +509,10 @@ All parameter, weight, gradient are variables in Paddle.
             self.num_threads_ = num_threads;
           })
       .def_property(
-          "use_event",
-          [](const ExecutionStrategy &self) { return self.use_event_; },
-          [](ExecutionStrategy &self, bool use_event) {
-            self.use_event_ = use_event;
+          "use_cuda",
+          [](const ExecutionStrategy &self) { return self.use_cuda_; },
+          [](ExecutionStrategy &self, bool use_cuda) {
+            self.use_cuda_ = use_cuda;
           })
       .def_property(
           "allow_op_delay",

python/paddle/fluid/parallel_executor.py

@@ -113,7 +113,7 @@ class ParallelExecutor(object):
 
         if exec_strategy is None:
             exec_strategy = ExecutionStrategy()
-        exec_strategy.use_event = use_cuda
+        exec_strategy.use_cuda = use_cuda
 
         if exec_strategy.num_threads == 0:
             if use_cuda: