diff --git a/paddle/operators/activation_op.cc b/paddle/operators/activation_op.cc
index 66e9d2c40138c26975f07cb544e54de6f00d6b09..2afa8a68b005f87dab952b59cf6d2e962049e0d9 100644
--- a/paddle/operators/activation_op.cc
+++ b/paddle/operators/activation_op.cc
@@ -285,11 +285,9 @@ REGISTER_OP(stanh, ops::ActivationOp, ops::STanhOpMaker<float>, stanh_grad,
 #define REGISTER_ACTIVATION_CPU_KERNEL(act_type, functor, grad_functor)        \
   REGISTER_OP_CPU_KERNEL(                                                      \
       act_type,                                                                \
-      paddle::operators::ActivationKernel<paddle::platform::CPUPlace,          \
-                                          paddle::operators::functor<float>>); \
+      ops::ActivationKernel<paddle::platform::CPUPlace, ops::functor<float>>); \
   REGISTER_OP_CPU_KERNEL(act_type##_grad,                                      \
-                         paddle::operators::ActivationGradKernel<              \
-                             paddle::platform::CPUPlace,                       \
-                             paddle::operators::grad_functor<float>>);
+                         ops::ActivationGradKernel<paddle::platform::CPUPlace, \
+                                                   ops::grad_functor<float>>);
 
 FOR_EACH_KERNEL_FUNCTOR(REGISTER_ACTIVATION_CPU_KERNEL);
diff --git a/paddle/operators/activation_op.cu b/paddle/operators/activation_op.cu
index 93e9f1c694bacba48c4f8c46f90fb5b512bead99..7b7644519d4e9cadcc4ca62ccb599262feffa660 100644
--- a/paddle/operators/activation_op.cu
+++ b/paddle/operators/activation_op.cu
@@ -15,14 +15,14 @@
 #define EIGEN_USE_GPU
 #include "paddle/operators/activation_op.h"
 
+namespace ops = paddle::operators;
+
 #define REGISTER_ACTIVATION_GPU_KERNEL(act_type, functor, grad_functor)        \
   REGISTER_OP_GPU_KERNEL(                                                      \
       act_type,                                                                \
-      paddle::operators::ActivationKernel<paddle::platform::GPUPlace,          \
-                                          paddle::operators::functor<float>>); \
+      ops::ActivationKernel<paddle::platform::GPUPlace, ops::functor<float>>); \
   REGISTER_OP_GPU_KERNEL(act_type##_grad,                                      \
-                         paddle::operators::ActivationGradKernel<              \
-                             paddle::platform::GPUPlace,                       \
-                             paddle::operators::grad_functor<float>>);
+                         ops::ActivationGradKernel<paddle::platform::GPUPlace, \
+                                                   ops::grad_functor<float>>);
 
 FOR_EACH_KERNEL_FUNCTOR(REGISTER_ACTIVATION_GPU_KERNEL);
diff --git a/paddle/operators/reduce_op.cc b/paddle/operators/reduce_op.cc
index 3ef443d1c7f475cbd578078db02fb5e0d500d060..55f294a9be6f441e2d0b3bd6f7854a9d07a4ee24 100644
--- a/paddle/operators/reduce_op.cc
+++ b/paddle/operators/reduce_op.cc
@@ -168,36 +168,22 @@ namespace ops = paddle::operators;
 
 REGISTER_OP(reduce_sum, ops::ReduceOp, ops::ReduceSumOpMaker, reduce_sum_grad,
             ops::ReduceGradOp);
-REGISTER_OP_CPU_KERNEL(
-    reduce_sum,
-    ops::ReduceKernel<paddle::platform::CPUPlace, float, ops::SumFunctor>);
-REGISTER_OP_CPU_KERNEL(reduce_sum_grad,
-                       ops::ReduceGradKernel<paddle::platform::CPUPlace, float,
-                                             ops::SumGradFunctor>);
 
 REGISTER_OP(reduce_mean, ops::ReduceOp, ops::ReduceMeanOpMaker,
             reduce_mean_grad, ops::ReduceGradOp);
-REGISTER_OP_CPU_KERNEL(
-    reduce_mean,
-    ops::ReduceKernel<paddle::platform::CPUPlace, float, ops::MeanFunctor>);
-REGISTER_OP_CPU_KERNEL(reduce_mean_grad,
-                       ops::ReduceGradKernel<paddle::platform::CPUPlace, float,
-                                             ops::MeanGradFunctor>);
 
 REGISTER_OP(reduce_max, ops::ReduceOp, ops::ReduceMaxOpMaker, reduce_max_grad,
             ops::ReduceGradOp);
-REGISTER_OP_CPU_KERNEL(
-    reduce_max,
-    ops::ReduceKernel<paddle::platform::CPUPlace, float, ops::MaxFunctor>);
-REGISTER_OP_CPU_KERNEL(reduce_max_grad,
-                       ops::ReduceGradKernel<paddle::platform::CPUPlace, float,
-                                             ops::MaxOrMinGradFunctor>);
-
-REGISTER_OP(reduce_min, ops::ReduceOp, ops::ReduceMaxOpMaker, reduce_min_grad,
+
+REGISTER_OP(reduce_min, ops::ReduceOp, ops::ReduceMinOpMaker, reduce_min_grad,
             ops::ReduceGradOp);
-REGISTER_OP_CPU_KERNEL(
-    reduce_min,
-    ops::ReduceKernel<paddle::platform::CPUPlace, float, ops::MinFunctor>);
-REGISTER_OP_CPU_KERNEL(reduce_min_grad,
-                       ops::ReduceGradKernel<paddle::platform::CPUPlace, float,
-                                             ops::MaxOrMinGradFunctor>);
+
+#define REGISTER_REDUCE_CPU_KERNEL(reduce_type, functor, grad_functor)     \
+  REGISTER_OP_CPU_KERNEL(                                                  \
+      reduce_type,                                                         \
+      ops::ReduceKernel<paddle::platform::CPUPlace, float, ops::functor>); \
+  REGISTER_OP_CPU_KERNEL(reduce_type##_grad,                               \
+                         ops::ReduceGradKernel<paddle::platform::CPUPlace, \
+                                               float, ops::grad_functor>);
+
+FOR_EACH_KERNEL_FUNCTOR(REGISTER_REDUCE_CPU_KERNEL);
diff --git a/paddle/operators/reduce_op.cu b/paddle/operators/reduce_op.cu
index 595127b858ea8eb41281f92e92c6467e4d90ff1a..d306e1a24096d737438d71d4d4abc35328d160cb 100644
--- a/paddle/operators/reduce_op.cu
+++ b/paddle/operators/reduce_op.cu
@@ -17,30 +17,12 @@
 
 namespace ops = paddle::operators;
 
-REGISTER_OP_GPU_KERNEL(
-    reduce_sum,
-    ops::ReduceKernel<paddle::platform::GPUPlace, float, ops::SumFunctor>);
-REGISTER_OP_GPU_KERNEL(reduce_sum_grad,
-                       ops::ReduceGradKernel<paddle::platform::GPUPlace, float,
-                                             ops::SumGradFunctor>);
-
-REGISTER_OP_GPU_KERNEL(
-    reduce_mean,
-    ops::ReduceKernel<paddle::platform::GPUPlace, float, ops::MeanFunctor>);
-REGISTER_OP_GPU_KERNEL(reduce_mean_grad,
-                       ops::ReduceGradKernel<paddle::platform::GPUPlace, float,
-                                             ops::MeanGradFunctor>);
-
-REGISTER_OP_GPU_KERNEL(
-    reduce_max,
-    ops::ReduceKernel<paddle::platform::GPUPlace, float, ops::MaxFunctor>);
-REGISTER_OP_GPU_KERNEL(reduce_max_grad,
-                       ops::ReduceGradKernel<paddle::platform::GPUPlace, float,
-                                             ops::MaxOrMinGradFunctor>);
-
-REGISTER_OP_GPU_KERNEL(
-    reduce_min,
-    ops::ReduceKernel<paddle::platform::GPUPlace, float, ops::MinFunctor>);
-REGISTER_OP_GPU_KERNEL(reduce_min_grad,
-                       ops::ReduceGradKernel<paddle::platform::GPUPlace, float,
-                                             ops::MaxOrMinGradFunctor>);
+#define REGISTER_REDUCE_GPU_KERNEL(reduce_type, functor, grad_functor)     \
+  REGISTER_OP_GPU_KERNEL(                                                  \
+      reduce_type,                                                         \
+      ops::ReduceKernel<paddle::platform::GPUPlace, float, ops::functor>); \
+  REGISTER_OP_GPU_KERNEL(reduce_type##_grad,                               \
+                         ops::ReduceGradKernel<paddle::platform::GPUPlace, \
+                                               float, ops::grad_functor>);
+
+FOR_EACH_KERNEL_FUNCTOR(REGISTER_REDUCE_GPU_KERNEL);
diff --git a/paddle/operators/reduce_op.h b/paddle/operators/reduce_op.h
index ba3f3db81dc6251a063d27e597fd7e486e7b6c14..45043c440bc8017e97f8be00d08f1cb60d201e20 100644
--- a/paddle/operators/reduce_op.h
+++ b/paddle/operators/reduce_op.h
@@ -198,3 +198,9 @@ class ReduceGradKernel : public framework::OpKernel<T> {
 
 }  // namespace operators
 }  // namespace paddle
+
+#define FOR_EACH_KERNEL_FUNCTOR(__macro)                \
+  __macro(reduce_sum, SumFunctor, SumGradFunctor);      \
+  __macro(reduce_mean, MeanFunctor, MeanGradFunctor);   \
+  __macro(reduce_max, MaxFunctor, MaxOrMinGradFunctor); \
+  __macro(reduce_min, MinFunctor, MaxOrMinGradFunctor);