diff --git a/paddle/operators/reduce_op.cc b/paddle/operators/reduce_op.cc
index 09b7091358e65221374a604122b742d763cfbafc..4a06babeda00f2420df80f81f876a0047a3285ef 100644
--- a/paddle/operators/reduce_op.cc
+++ b/paddle/operators/reduce_op.cc
@@ -190,10 +190,22 @@ REGISTER_OP(reduce_min, ops::ReduceOp, ops::ReduceMinOpMaker, reduce_min_grad,
 #define REGISTER_REDUCE_CPU_KERNEL(reduce_type, functor, grad_functor)         \
   REGISTER_OP_CPU_KERNEL(reduce_type,                                          \
                          ops::ReduceKernel<paddle::platform::CPUDeviceContext, \
-                                           float, ops::functor>);              \
+                                           float, ops::functor>,               \
+                         ops::ReduceKernel<paddle::platform::CPUDeviceContext, \
+                                           double, ops::functor>,              \
+                         ops::ReduceKernel<paddle::platform::CPUDeviceContext, \
+                                           int, ops::functor>,                 \
+                         ops::ReduceKernel<paddle::platform::CPUDeviceContext, \
+                                           int64_t, ops::functor>);            \
   REGISTER_OP_CPU_KERNEL(                                                      \
       reduce_type##_grad,                                                      \
       ops::ReduceGradKernel<paddle::platform::CPUDeviceContext, float,         \
+                            ops::grad_functor>,                                \
+      ops::ReduceGradKernel<paddle::platform::CPUDeviceContext, double,        \
+                            ops::grad_functor>,                                \
+      ops::ReduceGradKernel<paddle::platform::CPUDeviceContext, int,           \
+                            ops::grad_functor>,                                \
+      ops::ReduceGradKernel<paddle::platform::CPUDeviceContext, int64_t,       \
                             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 1dd948ed8a79cce8468f2fe210b5636e7dd1f99e..4ed1e051db4df579afe1c1ca24a06fa1baf3e13a 100644
--- a/paddle/operators/reduce_op.cu
+++ b/paddle/operators/reduce_op.cu
@@ -20,10 +20,22 @@ namespace ops = paddle::operators;
 #define REGISTER_REDUCE_GPU_KERNEL(reduce_type, functor, grad_functor)    \
   REGISTER_OP_CUDA_KERNEL(                                                \
       reduce_type, ops::ReduceKernel<paddle::platform::CUDADeviceContext, \
-                                     float, ops::functor>);               \
+                                     float, ops::functor>,                \
+      ops::ReduceKernel<paddle::platform::CUDADeviceContext, double,      \
+                        ops::functor>,                                    \
+      ops::ReduceKernel<paddle::platform::CUDADeviceContext, int,         \
+                        ops::functor>,                                    \
+      ops::ReduceKernel<paddle::platform::CUDADeviceContext, int64_t,     \
+                        ops::functor>);                                   \
   REGISTER_OP_CUDA_KERNEL(                                                \
       reduce_type##_grad,                                                 \
       ops::ReduceGradKernel<paddle::platform::CUDADeviceContext, float,   \
+                            ops::grad_functor>,                           \
+      ops::ReduceGradKernel<paddle::platform::CUDADeviceContext, double,  \
+                            ops::grad_functor>,                           \
+      ops::ReduceGradKernel<paddle::platform::CUDADeviceContext, int,     \
+                            ops::grad_functor>,                           \
+      ops::ReduceGradKernel<paddle::platform::CUDADeviceContext, int64_t, \
                             ops::grad_functor>);
 
 FOR_EACH_KERNEL_FUNCTOR(REGISTER_REDUCE_GPU_KERNEL);
diff --git a/python/paddle/v2/fluid/tests/test_reduce_op.py b/python/paddle/v2/fluid/tests/test_reduce_op.py
index 57ee307ba66b47bd15864e7be3943b4f5237eb1e..e2df0395ea9c99ed6d74d582dc9822780dd9ee1d 100644
--- a/python/paddle/v2/fluid/tests/test_reduce_op.py
+++ b/python/paddle/v2/fluid/tests/test_reduce_op.py
@@ -19,7 +19,7 @@ from op_test import OpTest
 class TestSumOp(OpTest):
     def setUp(self):
         self.op_type = "reduce_sum"
-        self.inputs = {'X': np.random.random((5, 6, 10)).astype("float32")}
+        self.inputs = {'X': np.random.random((5, 6, 10)).astype("float64")}
         self.outputs = {'Out': self.inputs['X'].sum(axis=0)}
 
     def test_check_output(self):
@@ -32,7 +32,7 @@ class TestSumOp(OpTest):
 class TestMeanOp(OpTest):
     def setUp(self):
         self.op_type = "reduce_mean"
-        self.inputs = {'X': np.random.random((5, 6, 2, 10)).astype("float32")}
+        self.inputs = {'X': np.random.random((5, 6, 2, 10)).astype("float64")}
         self.attrs = {'dim': 1}
         self.outputs = {'Out': self.inputs['X'].mean(axis=self.attrs['dim'])}
 
@@ -48,7 +48,7 @@ class TestMaxOp(OpTest):
 
     def setUp(self):
         self.op_type = "reduce_max"
-        self.inputs = {'X': np.random.random((5, 6, 10)).astype("float32")}
+        self.inputs = {'X': np.random.random((5, 6, 10)).astype("float64")}
         self.attrs = {'dim': -1}
         self.outputs = {'Out': self.inputs['X'].max(axis=self.attrs['dim'])}
 
@@ -61,7 +61,7 @@ class TestMinOp(OpTest):
 
     def setUp(self):
         self.op_type = "reduce_min"
-        self.inputs = {'X': np.random.random((5, 6, 10)).astype("float32")}
+        self.inputs = {'X': np.random.random((5, 6, 10)).astype("float64")}
         self.attrs = {'dim': 2}
         self.outputs = {'Out': self.inputs['X'].min(axis=self.attrs['dim'])}
 
@@ -72,7 +72,7 @@ class TestMinOp(OpTest):
 class TestKeepDimReduce(OpTest):
     def setUp(self):
         self.op_type = "reduce_sum"
-        self.inputs = {'X': np.random.random((5, 6, 10)).astype("float32")}
+        self.inputs = {'X': np.random.random((5, 6, 10)).astype("float64")}
         self.attrs = {'dim': -2, 'keep_dim': True}
         self.outputs = {
             'Out': self.inputs['X'].sum(axis=self.attrs['dim'], keepdims=True)
@@ -88,7 +88,7 @@ class TestKeepDimReduce(OpTest):
 class Test1DReduce(OpTest):
     def setUp(self):
         self.op_type = "reduce_sum"
-        self.inputs = {'X': np.random.random(20).astype("float32")}
+        self.inputs = {'X': np.random.random(20).astype("float64")}
         self.outputs = {'Out': self.inputs['X'].sum(axis=0)}
 
     def test_check_output(self):
@@ -101,7 +101,7 @@ class Test1DReduce(OpTest):
 class TestReduceAll(OpTest):
     def setUp(self):
         self.op_type = "reduce_sum"
-        self.inputs = {'X': np.random.random((5, 6, 2, 10)).astype("float32")}
+        self.inputs = {'X': np.random.random((5, 6, 2, 10)).astype("float64")}
         self.attrs = {'reduce_all': True}
         self.outputs = {'Out': self.inputs['X'].sum()}