diff --git a/paddle/fluid/operators/reader/buffered_reader.h b/paddle/fluid/operators/reader/buffered_reader.h
index 7f5d3d9fff8f28c8533587f326796f846987412e..cbe2bc1b5fdd69d1a843b768e3289acd621369a6 100644
--- a/paddle/fluid/operators/reader/buffered_reader.h
+++ b/paddle/fluid/operators/reader/buffered_reader.h
@@ -53,8 +53,8 @@ class BufferedReader : public framework::DecoratedReader {
 
   // The buffer for reading data.
   // NOTE: the simplest way to implement buffered reader is do not use any
-  // buffer, just async read and create futures as buffer size. However, to
-  // malloc Tensor every time is extremely slow. Here we store all data in
+  // buffer, just read async and create futures as buffer size. However, to
+  // malloc tensors every time is extremely slow. Here we store all data in
   // buffers and prevent alloc every time.
   std::vector<TensorVec> cpu_buffer_;
   std::vector<TensorVec> gpu_buffer_;
diff --git a/paddle/fluid/operators/reader/open_files_op.cc b/paddle/fluid/operators/reader/open_files_op.cc
index aa55681ffca9e43113172cbff88d80f19962c74f..38223e069975a08791d58d6ae10e2112b79a61fe 100644
--- a/paddle/fluid/operators/reader/open_files_op.cc
+++ b/paddle/fluid/operators/reader/open_files_op.cc
@@ -134,7 +134,7 @@ class PreemptiveReaderContainer : public IReaderContainer {
       } else {
         *out = item.data_;
         // continue read async
-        AsyncRead(item.reader_it_, &future_it);
+        ReadAsync(item.reader_it_, &future_it);
       }
     } else {
       out->clear();
@@ -151,10 +151,10 @@ class PreemptiveReaderContainer : public IReaderContainer {
     auto future_it = futures_.end();
     --future_it;
 
-    AsyncRead(reader_it, &future_it);
+    ReadAsync(reader_it, &future_it);
   }
 
-  void AsyncRead(const ReaderList::iterator& reader_it,
+  void ReadAsync(const ReaderList::iterator& reader_it,
                  FutureList::iterator* future_it_ptr) {
     auto& future_it = *future_it_ptr;
     *future_it = pool_.enqueue([reader_it, future_it, this] {