diff --git a/paddle/fluid/eager/tests/performance_tests/benchmark_eager_cpu.cc b/paddle/fluid/eager/tests/performance_tests/benchmark_eager_cpu.cc
index b72b7cb87530e037f399d9925df0a5ad75ae867c..6c4bf9a4f17e6f88503f0a1d6ec2f3029000b6f0 100644
--- a/paddle/fluid/eager/tests/performance_tests/benchmark_eager_cpu.cc
+++ b/paddle/fluid/eager/tests/performance_tests/benchmark_eager_cpu.cc
@@ -176,4 +176,4 @@ TEST(Benchmark, EagerIntermediateMLPCPU) {
 USE_OP_ITSELF(scale);
 USE_OP_ITSELF(elementwise_add);
 USE_OP_ITSELF(matmul_v2);
-USE_OP(reduce_sum);
+USE_OP_ITSELF(reduce_sum);
diff --git a/paddle/fluid/eager/tests/performance_tests/benchmark_eager_cuda.cc b/paddle/fluid/eager/tests/performance_tests/benchmark_eager_cuda.cc
index 701a9b2cba195dd53323e88c1044eb995b9725c5..14e7ce8cfcfb4dea0907cd128873223c8e5859a2 100644
--- a/paddle/fluid/eager/tests/performance_tests/benchmark_eager_cuda.cc
+++ b/paddle/fluid/eager/tests/performance_tests/benchmark_eager_cuda.cc
@@ -185,7 +185,7 @@ TEST(Benchmark, EagerIntermediateMLPCUDA) {
 
 USE_OP_ITSELF(scale);
 USE_OP_ITSELF(matmul_v2);
-USE_OP(reduce_sum);
+USE_OP_ITSELF(reduce_sum);
 USE_OP(reduce_sum_grad);
 USE_OP_ITSELF(elementwise_add);
 
diff --git a/paddle/fluid/eager/tests/performance_tests/benchmark_fluid_cpu.cc b/paddle/fluid/eager/tests/performance_tests/benchmark_fluid_cpu.cc
index bea80809a3b17a81cf8d3e19dc535a1f733b1896..3292de9363696dae30d853980eca6fb1ba1055cc 100644
--- a/paddle/fluid/eager/tests/performance_tests/benchmark_fluid_cpu.cc
+++ b/paddle/fluid/eager/tests/performance_tests/benchmark_fluid_cpu.cc
@@ -214,4 +214,4 @@ TEST(Benchmark, FluidMLPCPU) {
 USE_OP_ITSELF(scale);
 USE_OP_ITSELF(elementwise_add);
 USE_OP_ITSELF(matmul_v2);
-USE_OP(reduce_sum);
+USE_OP_ITSELF(reduce_sum);
diff --git a/paddle/fluid/eager/tests/performance_tests/benchmark_fluid_cuda.cc b/paddle/fluid/eager/tests/performance_tests/benchmark_fluid_cuda.cc
index 86c1ad7e23a6a3d186d53c7ff14c1450883465cd..e9b7d10070dbf22f10e617d34f143992d19fb659 100644
--- a/paddle/fluid/eager/tests/performance_tests/benchmark_fluid_cuda.cc
+++ b/paddle/fluid/eager/tests/performance_tests/benchmark_fluid_cuda.cc
@@ -247,7 +247,7 @@ TEST(Benchmark, FluidMLPCUDA) {
 
 USE_OP_ITSELF(scale);
 USE_OP_ITSELF(matmul_v2);
-USE_OP(reduce_sum);
+USE_OP_ITSELF(reduce_sum);
 USE_OP(reduce_sum_grad);
 USE_OP_ITSELF(elementwise_add);
 
diff --git a/paddle/fluid/framework/new_executor/standalone_executor_test.cc b/paddle/fluid/framework/new_executor/standalone_executor_test.cc
index c1d449d30205e2e2495a921cadef7f7742a8cedc..2c3359ffa8e46f0d30a01d73fccb95d88771480a 100644
--- a/paddle/fluid/framework/new_executor/standalone_executor_test.cc
+++ b/paddle/fluid/framework/new_executor/standalone_executor_test.cc
@@ -35,8 +35,8 @@ USE_OP(sigmoid);
 USE_OP(tanh);
 USE_OP(elementwise_mul);
 USE_OP(softmax_with_cross_entropy);
-USE_OP(reduce_mean);
-USE_OP(reduce_sum);
+USE_OP_ITSELF(reduce_mean);
+USE_OP_ITSELF(reduce_sum);
 USE_OP(reduce_sum_grad);
 USE_OP(reduce_mean_grad);
 USE_OP_ITSELF(reshape2_grad);
diff --git a/paddle/fluid/imperative/tests/test_tracer.cc b/paddle/fluid/imperative/tests/test_tracer.cc
index afe1f92ca03b3bc0724c1d04c26cb8c57138c136..d05036f7a12ebdc3db5fbfda5eb50c295c0478e4 100644
--- a/paddle/fluid/imperative/tests/test_tracer.cc
+++ b/paddle/fluid/imperative/tests/test_tracer.cc
@@ -590,6 +590,6 @@ TEST(test_tracer, eager_tracer) {
 
 USE_OP(mul);
 USE_OP(mul_grad);
-USE_OP(reduce_sum);
+USE_OP_ITSELF(reduce_sum);
 USE_OP(reduce_sum_grad);
 USE_OP_ITSELF(elementwise_add);
diff --git a/paddle/fluid/operators/reduce_ops/reduce_mean_op.cc b/paddle/fluid/operators/reduce_ops/reduce_mean_op.cc
index c8d568c8c2cf73041549a138085b72b41c0c297a..e80df5f95bb4ab33a6c08cc646d0ef8311e38936 100644
--- a/paddle/fluid/operators/reduce_ops/reduce_mean_op.cc
+++ b/paddle/fluid/operators/reduce_ops/reduce_mean_op.cc
@@ -99,13 +99,6 @@ REGISTER_OPERATOR(reduce_mean_grad, ops::ReduceGradOp,
                   ops::ReduceMeanDoubleGradDescMaker,
                   ops::ReduceMeanDoubleGradOpBaseMaker,
                   ops::ReduceMeanGradNoNeedBufferVarInferer);
-REGISTER_OP_CPU_KERNEL(reduce_mean,
-                       ops::ReduceKernel<paddle::platform::CPUDeviceContext,
-                                         bool, ops::MeanFunctor>,
-                       ops::ReduceKernel<paddle::platform::CPUDeviceContext,
-                                         float, ops::MeanFunctor>,
-                       ops::ReduceKernel<paddle::platform::CPUDeviceContext,
-                                         double, ops::MeanFunctor>);
 
 template <typename T>
 using CPUReduceMeanGradKernel =
diff --git a/paddle/fluid/operators/reduce_ops/reduce_mean_op.cu b/paddle/fluid/operators/reduce_ops/reduce_mean_op.cu
deleted file mode 100644
index 30a699e979efc40190a5c83850340f1f15dd918a..0000000000000000000000000000000000000000
--- a/paddle/fluid/operators/reduce_ops/reduce_mean_op.cu
+++ /dev/null
@@ -1,27 +0,0 @@
-// Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserved.
-//
-// Licensed under the Apache License, Version 2.0 (the "License");
-// you may not use this file except in compliance with the License.
-// You may obtain a copy of the License at
-//
-//     http://www.apache.org/licenses/LICENSE-2.0
-//
-// Unless required by applicable law or agreed to in writing, software
-// distributed under the License is distributed on an "AS IS" BASIS,
-// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
-// See the License for the specific language governing permissions and
-// limitations under the License.
-
-#include <vector>
-#include "paddle/fluid/operators/reduce_ops/reduce_mean_op.h"
-#include "paddle/fluid/operators/reduce_ops/reduce_op.h"
-
-REGISTER_OP_CUDA_KERNEL(
-    reduce_mean,
-    ops::ReduceCudaKernel<bool, kps::AddFunctor, kps::DivideFunctor>,
-    ops::ReduceCudaKernel<paddle::platform::float16, kps::AddFunctor,
-                          kps::DivideFunctor>,
-    ops::ReduceCudaKernel<float, kps::AddFunctor, kps::DivideFunctor>,
-    ops::ReduceCudaKernel<double, kps::AddFunctor, kps::DivideFunctor>,
-    ops::ReduceCudaKernel<int, kps::AddFunctor, kps::DivideFunctor>,
-    ops::ReduceCudaKernel<int64_t, kps::AddFunctor, kps::DivideFunctor>);
diff --git a/paddle/fluid/operators/reduce_ops/reduce_sum_op.cc b/paddle/fluid/operators/reduce_ops/reduce_sum_op.cc
index cfafc11739948ba791654516147f49d6fda78bdf..bdab14a18a05ab3e0df1dbda57f3753033cfacb4 100644
--- a/paddle/fluid/operators/reduce_ops/reduce_sum_op.cc
+++ b/paddle/fluid/operators/reduce_ops/reduce_sum_op.cc
@@ -107,27 +107,6 @@ REGISTER_OPERATOR(reduce_sum_grad, ops::ReduceGradOp,
                   ops::ReduceSumDoubleOpGradMaker<paddle::imperative::OpBase>,
                   ops::ReduceSumGradNoNeedBufferVarInferer);
 
-REGISTER_OP_CPU_KERNEL(
-    reduce_sum, ops::ReduceKernel<paddle::platform::CPUDeviceContext, bool,
-                                  ops::SumFunctor>,
-    ops::ReduceKernel<paddle::platform::CPUDeviceContext, float,
-                      ops::SumFunctor>,
-    ops::ReduceKernel<paddle::platform::CPUDeviceContext, double,
-                      ops::SumFunctor>,
-    ops::ReduceKernel<paddle::platform::CPUDeviceContext,
-                      paddle::platform::float16, ops::SumFunctor>,
-    ops::ReduceKernel<paddle::platform::CPUDeviceContext, int16_t,
-                      ops::SumFunctor>,
-    ops::ReduceKernel<paddle::platform::CPUDeviceContext, int, ops::SumFunctor>,
-    ops::ReduceKernel<paddle::platform::CPUDeviceContext, int64_t,
-                      ops::SumFunctor>,
-    ops::ReduceKernel<paddle::platform::CPUDeviceContext,
-                      paddle::platform::complex<float>, ops::SumFunctor>,
-    ops::ReduceKernel<paddle::platform::CPUDeviceContext,
-                      paddle::platform::complex<double>,
-
-                      ops::SumFunctor>);
-
 template <typename T>
 using CPUReduceSumGradKernel =
     ops::ReduceSumGradKernel<paddle::platform::CPUDeviceContext, T,
diff --git a/paddle/fluid/operators/reduce_ops/reduce_sum_op.cu b/paddle/fluid/operators/reduce_ops/reduce_sum_op.cu
deleted file mode 100644
index 94ccb0965f06e9bd8b8264b7cec91220194487c4..0000000000000000000000000000000000000000
--- a/paddle/fluid/operators/reduce_ops/reduce_sum_op.cu
+++ /dev/null
@@ -1,29 +0,0 @@
-// Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserved.
-//
-// Licensed under the Apache License, Version 2.0 (the "License");
-// you may not use this file except in compliance with the License.
-// You may obtain a copy of the License at
-//
-//     http://www.apache.org/licenses/LICENSE-2.0
-//
-// Unless required by applicable law or agreed to in writing, software
-// distributed under the License is distributed on an "AS IS" BASIS,
-// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
-// See the License for the specific language governing permissions and
-// limitations under the License.
-#include "paddle/fluid/operators/reduce_ops/reduce_op.h"
-#include "paddle/fluid/operators/reduce_ops/reduce_sum_op.h"
-REGISTER_OP_CUDA_KERNEL(
-    reduce_sum,
-    ops::ReduceCudaKernel<bool, kps::AddFunctor, kps::IdentityFunctor>,
-    ops::ReduceCudaKernel<float, kps::AddFunctor, kps::IdentityFunctor>,
-    ops::ReduceCudaKernel<double, kps::AddFunctor, kps::IdentityFunctor>,
-    ops::ReduceCudaKernel<paddle::platform::float16, kps::AddFunctor,
-                          kps::IdentityFunctor>,
-    ops::ReduceCudaKernel<int16_t, kps::AddFunctor, kps::IdentityFunctor>,
-    ops::ReduceCudaKernel<int, kps::AddFunctor, kps::IdentityFunctor>,
-    ops::ReduceCudaKernel<int64_t, kps::AddFunctor, kps::IdentityFunctor>,
-    ops::ReduceCudaKernel<paddle::platform::complex<float>, kps::AddFunctor,
-                          kps::IdentityFunctor>,
-    ops::ReduceCudaKernel<paddle::platform::complex<double>, kps::AddFunctor,
-                          kps::IdentityFunctor>);
diff --git a/paddle/fluid/operators/reduce_ops/unity_build_rule.cmake b/paddle/fluid/operators/reduce_ops/unity_build_rule.cmake
index 74781ef6f0237ab34a6c6590f5d014f352acba39..c4f32a8d25764b4cd0a7ce0a8cb4691fd2c8ea05 100644
--- a/paddle/fluid/operators/reduce_ops/unity_build_rule.cmake
+++ b/paddle/fluid/operators/reduce_ops/unity_build_rule.cmake
@@ -21,5 +21,4 @@ register_unity_group(cu
 register_unity_group(cu frobenius_norm_op.cu)
 register_unity_group(cu logsumexp_op.cu)
 register_unity_group(cu reduce_max_op.cu)
-register_unity_group(cu reduce_mean_op.cu)
 register_unity_group(cu reduce_min_op.cu)
diff --git a/paddle/phi/kernels/gpu/math_kernel.cu b/paddle/phi/kernels/gpu/math_kernel.cu
index 2ae40bd4b1923bd59883bb474128624de85350f7..c3605ce655f2bde78c01a88db4dd70bd38c98431 100644
--- a/paddle/phi/kernels/gpu/math_kernel.cu
+++ b/paddle/phi/kernels/gpu/math_kernel.cu
@@ -165,4 +165,6 @@ PT_REGISTER_KERNEL(mean_raw,
                    float,
                    double,
                    bool,
-                   float16) {}
+                   float16,
+                   int,
+                   int64_t) {}
diff --git a/paddle/phi/kernels/math_kernel.cc b/paddle/phi/kernels/math_kernel.cc
index 7fb6cc0ba9cca504dfdf8abab201712b22c21e4e..e1e3679ea8be89ef438b59e70889e23adc360ea0 100644
--- a/paddle/phi/kernels/math_kernel.cc
+++ b/paddle/phi/kernels/math_kernel.cc
@@ -152,6 +152,8 @@ PT_REGISTER_KERNEL(mean,
                    float,
                    double,
                    bool,
+                   int,
+                   int64_t,
                    phi::dtype::float16) {}
 PT_REGISTER_KERNEL(sum,
                    GPU,