package a new USE_NO_KERNEL_OP for USE_OP_ITSELF

doc/howto/dev/new_op_cn.md

@@ -227,6 +227,12 @@ make mul_op
     USE_CPU_ONLY_OP(gather);
     ```
     
+    如果OP不带Kernel，则使用`USE_NO_KENREL_OP`:
+    
+    ```
+    USE_NO_KENREL_OP(recurrent);
+    ```
+    
     使用`USE_OP`告知编译器需要链接该Op的目标文件，具体解释参考[代码注释](https://github.com/PaddlePaddle/Paddle/blob/develop/paddle/framework/op_registry.h#L81)。
     
     

paddle/framework/backward.cc

@@ -182,7 +182,7 @@ static std::unique_ptr<OperatorBase> BackwardRecursive(
                    });
 
     // process recurrent gradient op as a special operator.
-    if (forwardOp.Type() == "recurrent_op") {
+    if (forwardOp.Type() == "recurrent") {
       // NOTE clean up cycle call somewhere (RNN's stepnet constains itself), or
       // this will result in infinite loop.
       const auto& rnnop =

paddle/framework/op_registry.h

@@ -194,6 +194,8 @@ class OpKernelRegistrar : public Registrar {
   USE_OP_DEVICE_KERNEL(op_type, GPU)
 #endif
 
+#define USE_NO_KERNEL_OP(op_type) USE_OP_ITSELF(op_type);
+
 #define USE_CPU_ONLY_OP(op_type) \
   USE_OP_ITSELF(op_type);        \
   USE_OP_DEVICE_KERNEL(op_type, CPU);

paddle/operators/minus_op.cc

@@ -79,7 +79,7 @@ class MinusGradOp : public NetOp {
 }  // namespace paddle
 
 USE_OP(scale);
-USE_OP_ITSELF(identity);
+USE_NO_KERNEL_OP(identity);
 namespace ops = paddle::operators;
 REGISTER_OP(minus, ops::MinusOp, ops::MinusOpMaker, ops::MinusGradOp<float>);
 REGISTER_OP_CPU_KERNEL(minus,

paddle/operators/recurrent_op.cc

@@ -235,5 +235,5 @@ RecurrentGradientOp::RecurrentGradientOp(
 }  // namespace paddle
 
 REGISTER_OP_WITHOUT_GRADIENT(
-    recurrent_op, paddle::operators::RecurrentOp,
+    recurrent, paddle::operators::RecurrentOp,
     paddle::operators::RecurrentAlgorithmProtoAndCheckerMaker);

paddle/pybind/pybind.cc

@@ -39,12 +39,12 @@ USE_OP(sigmoid);
 USE_OP(softmax);
 USE_OP(rowwise_add);
 USE_OP(fill_zeros_like);
-USE_OP_ITSELF(recurrent_op);
+USE_NO_KERNEL_OP(recurrent);
 USE_OP(gaussian_random);
 USE_OP(uniform_random);
 USE_OP(lookup_table);
 USE_OP(scale);
-USE_OP_ITSELF(identity);
+USE_NO_KERNEL_OP(identity);
 USE_OP(minus);
 USE_CPU_ONLY_OP(gather);
 USE_CPU_ONLY_OP(scatter);

python/paddle/v2/framework/op.py

@@ -179,7 +179,7 @@ class OperatorFactory(object):
 
 class __RecurrentOp__(object):
     __proto__ = None
-    type = 'recurrent_op'
+    type = 'recurrent'
 
     def __init__(self):
         # cache recurrent_op's proto