diff --git a/paddle/fluid/operators/elementwise/elementwise_floordiv_op.cc b/paddle/fluid/operators/elementwise/elementwise_floordiv_op.cc
index 67b9b665c6c85db39a35538f1fa0b7d8dc1c6d3d..b876438a1941f8626b713000575d6a3058624aae 100644
--- a/paddle/fluid/operators/elementwise/elementwise_floordiv_op.cc
+++ b/paddle/fluid/operators/elementwise/elementwise_floordiv_op.cc
@@ -12,6 +12,8 @@ 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/elementwise/elementwise_floordiv_op.h"
+
 #include <string>
 
 #include "paddle/fluid/operators/elementwise/elementwise_op.h"
@@ -61,6 +63,12 @@ namespace ops = paddle::operators;
 REGISTER_OP_WITHOUT_GRADIENT(elementwise_floordiv, ops::ElementwiseOp,
                              ops::ElementwiseFloorDivOpMaker);
 
+REGISTER_OP_CPU_KERNEL(
+    elementwise_floordiv,
+    ops::ElementwiseFloorDivKernel<paddle::platform::CPUDeviceContext, int>,
+    ops::ElementwiseFloorDivKernel<paddle::platform::CPUDeviceContext,
+                                   int64_t>);
+
 REGISTER_OP_VERSION(elementwise_floordiv)
     .AddCheckpoint(
         R"ROC(Register elementwise_floordiv for adding the attribute of Scale_y)ROC",
diff --git a/paddle/fluid/operators/elementwise/elementwise_floordiv_op.cu b/paddle/fluid/operators/elementwise/elementwise_floordiv_op.cu
new file mode 100644
index 0000000000000000000000000000000000000000..9b146fe7279deafb89a8d9526ddfd20f92da20a9
--- /dev/null
+++ b/paddle/fluid/operators/elementwise/elementwise_floordiv_op.cu
@@ -0,0 +1,46 @@
+/* Copyright (c) 2019 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/elementwise/elementwise_floordiv_op.h"
+
+namespace paddle {
+namespace operators {
+
+template <typename T>
+class ElementwiseFloorDivKernel<platform::CUDADeviceContext, T>
+    : public framework::OpKernel<T> {
+ public:
+  void Compute(const framework::ExecutionContext& ctx) const override {
+    std::vector<const framework::Tensor*> ins;
+    std::vector<framework::Tensor*> outs;
+    const auto& cuda_ctx =
+        ctx.template device_context<platform::CUDADeviceContext>();
+
+    int axis = PackTensorsIntoVector<T>(ctx, &ins, &outs);
+    paddle::operators::LaunchElementwiseCudaKernel<ElementwiseType::kBinary, T,
+                                                   T>(
+        cuda_ctx, ins, &outs, axis, FloorDivFunctor<T>());
+  }
+};
+
+}  // namespace operators
+}  // namespace paddle
+
+namespace ops = paddle::operators;
+namespace plat = paddle::platform;
+
+REGISTER_OP_CUDA_KERNEL(
+    elementwise_floordiv,
+    ops::ElementwiseFloorDivKernel<plat::CUDADeviceContext, int>,
+    ops::ElementwiseFloorDivKernel<plat::CUDADeviceContext, int64_t>);
diff --git a/paddle/fluid/operators/elementwise/elementwise_floordiv_op.h b/paddle/fluid/operators/elementwise/elementwise_floordiv_op.h
new file mode 100644
index 0000000000000000000000000000000000000000..fc8f18161990db34ff555d2f64c5d157ef5900b6
--- /dev/null
+++ b/paddle/fluid/operators/elementwise/elementwise_floordiv_op.h
@@ -0,0 +1,54 @@
+/* Copyright (c) 2019 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. */
+
+#pragma once
+
+#include "paddle/fluid/operators/elementwise/elementwise_op.h"
+
+namespace paddle {
+namespace operators {
+
+template <typename DeviceContext, typename T>
+void elementwise_floor_div(const framework::ExecutionContext &ctx,
+                           const framework::Tensor *x,
+                           const framework::Tensor *y, framework::Tensor *z) {
+  int axis = ctx.Attr<int>("axis");
+  auto x_dims = x->dims();
+  auto y_dims = y->dims();
+  if (x_dims.size() >= y_dims.size()) {
+    ElementwiseComputeEx<FloorDivFunctor<T>, DeviceContext, T>(
+        ctx, x, y, axis, FloorDivFunctor<T>(), z);
+  } else {
+    ElementwiseComputeEx<InverseFloorDivFunctor<T>, DeviceContext, T>(
+        ctx, x, y, axis, InverseFloorDivFunctor<T>(), z);
+  }
+}
+
+template <typename DeviceContext, typename T>
+class ElementwiseFloorDivKernel : public framework::OpKernel<T> {
+ public:
+  void Compute(const framework::ExecutionContext &ctx) const override {
+    auto *x = ctx.Input<framework::LoDTensor>("X");
+    auto *y = ctx.Input<framework::LoDTensor>("Y");
+    auto *z = ctx.Output<framework::LoDTensor>("Out");
+
+    z->mutable_data<T>(ctx.GetPlace());
+
+    // dtype of x and y is int64 or int32
+    elementwise_floor_div<DeviceContext, T>(ctx, x, y, z);
+  }
+};
+
+}  // namespace operators
+}  // namespace paddle
diff --git a/paddle/fluid/operators/elementwise/elementwise_functor.h b/paddle/fluid/operators/elementwise/elementwise_functor.h
index 844b0a195031f517bac647c885bc2cf279ab66f1..5dfb7eece9ec77b3f10196654e976e335af890e9 100644
--- a/paddle/fluid/operators/elementwise/elementwise_functor.h
+++ b/paddle/fluid/operators/elementwise/elementwise_functor.h
@@ -49,6 +49,23 @@ using DivFunctor = phi::funcs::DivideFunctor<T>;
 template <typename T>
 using InverseDivFunctor = phi::funcs::InverseDivideFunctor<T>;
 
+// Floor Divide
+template <typename T>
+struct FloorDivFunctor {
+  inline HOSTDEVICE T operator()(const T a, const T b) const {
+    PADDLE_ENFORCE(b != 0, DIV_ERROR_INFO);
+    return static_cast<T>(std::trunc(a / b));
+  }
+};
+
+template <typename T>
+struct InverseFloorDivFunctor {
+  inline HOSTDEVICE T operator()(const T a, const T b) const {
+    PADDLE_ENFORCE(a != 0, DIV_ERROR_INFO);
+    return static_cast<T>(std::trunc(b / a));
+  }
+};
+
 #undef DIV_ERROR_INFO
 
 // Maximum
diff --git a/paddle/fluid/operators/elementwise/elementwise_pow_op.cc b/paddle/fluid/operators/elementwise/elementwise_pow_op.cc
index c0dbb0df8c12f940014082accc017e01ac9c2bb5..eddbfd3b15ea46e55566a9dd030a437d8f25583b 100644
--- a/paddle/fluid/operators/elementwise/elementwise_pow_op.cc
+++ b/paddle/fluid/operators/elementwise/elementwise_pow_op.cc
@@ -9,6 +9,8 @@ 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/elementwise/elementwise_pow_op.h"
+
 #include <string>
 
 #include "paddle/fluid/operators/elementwise/elementwise_op.h"
@@ -68,6 +70,19 @@ REGISTER_OPERATOR(elementwise_pow, ops::ElementwiseOp,
                   ops::ElementwisePowOpGradMaker<paddle::imperative::OpBase>);
 REGISTER_OPERATOR(elementwise_pow_grad, ops::ElementwiseOpGrad);
 
+REGISTER_OP_CPU_KERNEL(
+    elementwise_pow,
+    ops::ElementwisePowKernel<paddle::platform::CPUDeviceContext, float>,
+    ops::ElementwisePowKernel<paddle::platform::CPUDeviceContext, double>,
+    ops::ElementwisePowKernel<paddle::platform::CPUDeviceContext, int>,
+    ops::ElementwisePowKernel<paddle::platform::CPUDeviceContext, int64_t>);
+REGISTER_OP_CPU_KERNEL(
+    elementwise_pow_grad,
+    ops::ElementwisePowGradKernel<paddle::platform::CPUDeviceContext, float>,
+    ops::ElementwisePowGradKernel<paddle::platform::CPUDeviceContext, double>,
+    ops::ElementwisePowGradKernel<paddle::platform::CPUDeviceContext, int>,
+    ops::ElementwisePowGradKernel<paddle::platform::CPUDeviceContext, int64_t>);
+
 REGISTER_OP_VERSION(elementwise_pow)
     .AddCheckpoint(
         R"ROC(Register elementwise_pow for adding the attribute of Scale_y)ROC",
diff --git a/paddle/fluid/operators/elementwise/elementwise_pow_op.cu b/paddle/fluid/operators/elementwise/elementwise_pow_op.cu
new file mode 100644
index 0000000000000000000000000000000000000000..1286064dacfda49d4e6e68cd1c18f60e15b14ea1
--- /dev/null
+++ b/paddle/fluid/operators/elementwise/elementwise_pow_op.cu
@@ -0,0 +1,54 @@
+/* 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/elementwise/elementwise_pow_op.h"
+
+namespace ops = paddle::operators;
+
+namespace paddle {
+namespace operators {
+
+template <typename T>
+class ElementwisePowKernel<platform::CUDADeviceContext, T>
+    : public framework::OpKernel<T> {
+ public:
+  void Compute(const framework::ExecutionContext& ctx) const override {
+    std::vector<const framework::Tensor*> ins;
+    std::vector<framework::Tensor*> outs;
+    const auto& cuda_ctx =
+        ctx.template device_context<platform::CUDADeviceContext>();
+
+    int axis = PackTensorsIntoVector<T>(ctx, &ins, &outs);
+    paddle::operators::LaunchElementwiseCudaKernel<ElementwiseType::kBinary, T,
+                                                   T>(cuda_ctx, ins, &outs,
+                                                      axis, PowFunctor<T>());
+  }
+};
+
+}  // namespace operators
+}  // namespace paddle
+
+REGISTER_OP_CUDA_KERNEL(
+    elementwise_pow,
+    ops::ElementwisePowKernel<paddle::platform::CUDADeviceContext, float>,
+    ops::ElementwisePowKernel<paddle::platform::CUDADeviceContext, double>,
+    ops::ElementwisePowKernel<paddle::platform::CUDADeviceContext, int>,
+    ops::ElementwisePowKernel<paddle::platform::CUDADeviceContext, int64_t>);
+REGISTER_OP_CUDA_KERNEL(
+    elementwise_pow_grad,
+    ops::ElementwisePowGradKernel<paddle::platform::CUDADeviceContext, float>,
+    ops::ElementwisePowGradKernel<paddle::platform::CUDADeviceContext, double>,
+    ops::ElementwisePowGradKernel<paddle::platform::CUDADeviceContext, int>,
+    ops::ElementwisePowGradKernel<paddle::platform::CUDADeviceContext,
+                                  int64_t>);
diff --git a/paddle/fluid/operators/elementwise/elementwise_pow_op.h b/paddle/fluid/operators/elementwise/elementwise_pow_op.h
new file mode 100644
index 0000000000000000000000000000000000000000..1dfe7ed2325f4429cd825e624b27c3f5ae092655
--- /dev/null
+++ b/paddle/fluid/operators/elementwise/elementwise_pow_op.h
@@ -0,0 +1,107 @@
+/* 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. */
+
+#pragma once
+
+#include <cmath>
+#include <type_traits>
+
+#include "paddle/fluid/operators/elementwise/elementwise_op.h"
+
+namespace paddle {
+namespace operators {
+
+template <typename T>
+struct PowFunctor {
+  inline HOSTDEVICE T operator()(const T a, const T b) const {
+// TODO(wujionghao): A potential speed improvement is supporting different
+// types in C++.
+#if defined(__CUDA_ARCH__) || defined(__HIPCC__)
+    // On CUDAPlace, std::pow(3, 1) calls pow(float, float), and
+    // it will return a float number like 2.99... , which floor to 2
+    // when cast to int by default and it is wrong.
+    // Use llrint to cast it to the nearest integer, which is 3.
+    if (std::is_integral<T>::value) {
+      return std::llrint(
+          std::pow(static_cast<double>(a), static_cast<double>(b)));
+    }
+#endif
+    return std::pow(a, b);
+  }
+};
+
+template <typename DeviceContext, typename T>
+class ElementwisePowKernel : public framework::OpKernel<T> {
+ public:
+  void Compute(const framework::ExecutionContext& ctx) const override {
+    using Tensor = framework::LoDTensor;
+    auto* x = ctx.Input<Tensor>("X");
+    PADDLE_ENFORCE_EQ(x != nullptr, true,
+                      platform::errors::NotFound(
+                          "Cannot get input Variable X, Variable name = %s",
+                          ctx.InputName("X")));
+    auto* y = ctx.Input<Tensor>("Y");
+    auto* z = ctx.Output<Tensor>("Out");
+    z->mutable_data<T>(ctx.GetPlace());
+    int axis = ctx.Attr<int>("axis");
+    ElementwiseComputeEx<PowFunctor<T>, DeviceContext, T>(ctx, x, y, axis,
+                                                          PowFunctor<T>(), z);
+  }
+};
+
+template <typename T>
+struct PowGradDX {
+  HOSTDEVICE T operator()(T x, T y, T out, T dout) const {
+#if defined(__CUDA_ARCH__) || defined(__HIPCC__)
+    if (std::is_integral<T>::value) {
+      return dout * y *
+             std::pow(static_cast<double>(x), static_cast<double>(y - 1));
+    }
+#endif
+    return dout * y * std::pow(x, y - 1);
+  }
+};
+
+template <typename T, typename Enable = void>
+struct PowGradDY {
+  HOSTDEVICE T operator()(T x, T y, T out, T dout) const {
+#if defined(__CUDA_ARCH__) || defined(__HIPCC__)
+    if (std::is_integral<T>::value) {
+      return dout * std::log(static_cast<double>(x)) *
+             std::pow(static_cast<double>(x), static_cast<double>(y));
+    }
+#endif
+    return dout * std::log(x) * std::pow(x, y);
+  }
+};
+
+template <typename DeviceContext, typename T>
+class ElementwisePowGradKernel : public ElemwiseGradKernel<T> {
+ public:
+  void Compute(const framework::ExecutionContext& ctx) const override {
+    ElemwiseGradKernel<T>::Compute(ctx);
+    using Tensor = framework::Tensor;
+    auto* x = ctx.Input<Tensor>("X");
+    auto* y = ctx.Input<Tensor>("Y");
+    auto* dout = ctx.Input<Tensor>(framework::GradVarName("Out"));
+    auto* out = dout;
+    auto* dx = ctx.Output<Tensor>(framework::GradVarName("X"));
+    auto* dy = ctx.Output<Tensor>(framework::GradVarName("Y"));
+    int axis = ctx.Attr<int>("axis");
+    ElemwiseGradCompute<DeviceContext, T, PowGradDX<T>, PowGradDY<T>>(
+        ctx, *x, *y, *out, *dout, axis, dx, dy, PowGradDX<T>(), PowGradDY<T>());
+  }
+};
+}  // namespace operators
+}  // namespace paddle
diff --git a/paddle/fluid/operators/elementwise/elementwise_pow_op_npu.cc b/paddle/fluid/operators/elementwise/elementwise_pow_op_npu.cc
index c8fbd456123c831d98264d162603e21ccc3f9ee0..a2d22767471299796592f1bac69ef497fa8bb671 100644
--- a/paddle/fluid/operators/elementwise/elementwise_pow_op_npu.cc
+++ b/paddle/fluid/operators/elementwise/elementwise_pow_op_npu.cc
@@ -16,6 +16,7 @@ limitations under the License. */
 #include <string>
 
 #include "paddle/fluid/operators/elementwise/elementwise_npu.h"
+#include "paddle/fluid/operators/elementwise/elementwise_pow_op.h"
 #include "paddle/fluid/platform/device/npu/npu_op_runner.h"
 
 namespace paddle {
diff --git a/paddle/phi/kernels/cpu/elementwise_grad_kernel.cc b/paddle/phi/kernels/cpu/elementwise_grad_kernel.cc
index 1548272f8622c4817718311b4db57b0f00583738..b617064987507339743ab145ef94bdf15bdbdca2 100644
--- a/paddle/phi/kernels/cpu/elementwise_grad_kernel.cc
+++ b/paddle/phi/kernels/cpu/elementwise_grad_kernel.cc
@@ -323,11 +323,3 @@ PD_REGISTER_KERNEL(minimum_grad,
                    int,
                    int64_t,
                    phi::dtype::bfloat16) {}
-PD_REGISTER_KERNEL(elementwise_pow_grad,
-                   CPU,
-                   ALL_LAYOUT,
-                   phi::ElementwisePowGradKernel,
-                   float,
-                   double,
-                   int,
-                   int64_t) {}
diff --git a/paddle/phi/kernels/cpu/elementwise_kernel.cc b/paddle/phi/kernels/cpu/elementwise_kernel.cc
index 4ca41de7bb64a9f444e4b57713ccb46adbf96e4e..1de40cb9466b295e5656ae231cb754990e004195 100644
--- a/paddle/phi/kernels/cpu/elementwise_kernel.cc
+++ b/paddle/phi/kernels/cpu/elementwise_kernel.cc
@@ -113,36 +113,6 @@ void ModuloRawKernel(const Context& dev_ctx,
   }
 }
 
-template <typename T, typename Context>
-void FloorDivideRawKernel(const Context& dev_ctx,
-                          const DenseTensor& x,
-                          const DenseTensor& y,
-                          int axis,
-                          DenseTensor* out) {
-  // allocate memory for out
-  dev_ctx.template Alloc<T>(out);
-  auto x_dims = x.dims();
-  auto y_dims = y.dims();
-  if (x_dims.size() >= y_dims.size()) {
-    funcs::ElementwiseCompute<funcs::FloorDivideFunctor<T>, T>(
-        dev_ctx, x, y, axis, funcs::FloorDivideFunctor<T>(), out);
-  } else {
-    funcs::ElementwiseCompute<funcs::InverseFloorDivideFunctor<T>, T>(
-        dev_ctx, x, y, axis, funcs::InverseFloorDivideFunctor<T>(), out);
-  }
-}
-
-template <typename T, typename Context>
-void ElementwisePowRawKernel(const Context& dev_ctx,
-                             const DenseTensor& x,
-                             const DenseTensor& y,
-                             int axis,
-                             DenseTensor* out) {
-  // allocate memory for out
-  dev_ctx.template Alloc<T>(out);
-  funcs::ElementwiseCompute<funcs::ElementwisePowFunctor<T>, T>(
-      dev_ctx, x, y, axis, funcs::ElementwisePowFunctor<T>(), out);
-}
 // Create the definition of Add
 DEFINE_CPU_ELEMENTWISE_OP(Add)
 
@@ -237,17 +207,3 @@ PD_REGISTER_KERNEL(modulo_raw,
                    double,
                    int,
                    int64_t) {}
-PD_REGISTER_KERNEL(floor_divide_raw,
-                   CPU,
-                   ALL_LAYOUT,
-                   phi::FloorDivideRawKernel,
-                   int,
-                   int64_t) {}
-PD_REGISTER_KERNEL(elementwise_pow_raw,
-                   CPU,
-                   ALL_LAYOUT,
-                   phi::ElementwisePowRawKernel,
-                   float,
-                   double,
-                   int,
-                   int64_t) {}
diff --git a/paddle/phi/kernels/elementwise_grad_kernel.h b/paddle/phi/kernels/elementwise_grad_kernel.h
index 979bb61c2e3cab464e3ed6b9895e8e4914ef8583..95832013caff5d366f264123262d3c05fb36b35f 100644
--- a/paddle/phi/kernels/elementwise_grad_kernel.h
+++ b/paddle/phi/kernels/elementwise_grad_kernel.h
@@ -159,13 +159,4 @@ void MinimumGradKernel(const Context& dev_ctx,
                        int axis,
                        DenseTensor* dx,
                        DenseTensor* dy);
-
-template <typename T, typename Context>
-void ElementwisePowGradKernel(const Context& dev_ctx,
-                              const DenseTensor& x,
-                              const DenseTensor& y,
-                              const DenseTensor& dout,
-                              int axis,
-                              DenseTensor* dx,
-                              DenseTensor* dy);
 }  // namespace phi
diff --git a/paddle/phi/kernels/elementwise_kernel.cc b/paddle/phi/kernels/elementwise_kernel.cc
index 4194631ea266b70365ba5e0576451ff1d9d80ff2..019d4fed5b28eaed72a370bfb51c1a75807964fd 100644
--- a/paddle/phi/kernels/elementwise_kernel.cc
+++ b/paddle/phi/kernels/elementwise_kernel.cc
@@ -81,25 +81,6 @@ void ModuloKernel(const Context& dev_ctx,
   int axis = -1;
   ModuloRawKernel<T>(dev_ctx, x, y, axis, out);
 }
-
-template <typename T, typename Context>
-void FloorDivideKernel(const Context& dev_ctx,
-                       const DenseTensor& x,
-                       const DenseTensor& y,
-                       DenseTensor* out) {
-  int axis = -1;
-  FloorDivideRawKernel<T>(dev_ctx, x, y, axis, out);
-}
-
-template <typename T, typename Context>
-void ElementwisePowKernel(const Context& dev_ctx,
-                          const DenseTensor& x,
-                          const DenseTensor& y,
-                          DenseTensor* out) {
-  int axis = -1;
-  ElementwisePowRawKernel<T>(dev_ctx, x, y, axis, out);
-}
-
 }  // namespace phi
 
 using complex64 = ::phi::dtype::complex<float>;
@@ -170,16 +151,6 @@ PD_REGISTER_KERNEL(minimum,
                    phi::dtype::bfloat16) {}
 PD_REGISTER_KERNEL(
     modulo, CPU, ALL_LAYOUT, phi::ModuloKernel, float, double, int, int64_t) {}
-PD_REGISTER_KERNEL(
-    floor_divide, CPU, ALL_LAYOUT, phi::FloorDivideKernel, int, int64_t) {}
-PD_REGISTER_KERNEL(elementwise_pow,
-                   CPU,
-                   ALL_LAYOUT,
-                   phi::ElementwisePowKernel,
-                   float,
-                   double,
-                   int,
-                   int64_t) {}
 
 #if defined(PADDLE_WITH_CUDA) || defined(PADDLE_WITH_HIP)
 
@@ -255,14 +226,4 @@ PD_REGISTER_KERNEL(minimum,
                    phi::dtype::bfloat16) {}
 PD_REGISTER_KERNEL(
     modulo, GPU, ALL_LAYOUT, phi::ModuloKernel, float, double, int, int64_t) {}
-PD_REGISTER_KERNEL(
-    floor_divide, GPU, ALL_LAYOUT, phi::FloorDivideKernel, int, int64_t) {}
-PD_REGISTER_KERNEL(elementwise_pow,
-                   GPU,
-                   ALL_LAYOUT,
-                   phi::ElementwisePowKernel,
-                   float,
-                   double,
-                   int,
-                   int64_t) {}
 #endif
diff --git a/paddle/phi/kernels/elementwise_kernel.h b/paddle/phi/kernels/elementwise_kernel.h
index 09b6b02e372571095de200b856a2a0103ef05182..f9c9c7f7139f3ad2298ba0ebcae026746ab450dc 100644
--- a/paddle/phi/kernels/elementwise_kernel.h
+++ b/paddle/phi/kernels/elementwise_kernel.h
@@ -124,32 +124,6 @@ void ModuloKernel(const Context& dev_ctx,
                   const DenseTensor& y,
                   DenseTensor* out);
 
-template <typename T, typename Context>
-void FloorDivideRawKernel(const Context& dev_ctx,
-                          const DenseTensor& x,
-                          const DenseTensor& y,
-                          int axis,
-                          DenseTensor* out);
-
-template <typename T, typename Context>
-void FloorDivideKernel(const Context& dev_ctx,
-                       const DenseTensor& x,
-                       const DenseTensor& y,
-                       DenseTensor* out);
-
-template <typename T, typename Context>
-void ElementwisePowRawKernel(const Context& dev_ctx,
-                             const DenseTensor& x,
-                             const DenseTensor& y,
-                             int axis,
-                             DenseTensor* out);
-
-template <typename T, typename Context>
-void ElementwisePowKernel(const Context& dev_ctx,
-                          const DenseTensor& x,
-                          const DenseTensor& y,
-                          DenseTensor* out);
-
 template <typename T, typename Context>
 DenseTensor Add(const Context& dev_ctx,
                 const DenseTensor& x,
@@ -226,27 +200,4 @@ DenseTensor Modulo(const Context& dev_ctx,
   ModuloKernel<T, Context>(dev_ctx, x, y, &dense_out);
   return dense_out;
 }
-
-template <typename T, typename Context>
-DenseTensor FloorDivide(const Context& dev_ctx,
-                        const DenseTensor& x,
-                        const DenseTensor& y) {
-  DenseTensor dense_out;
-  MetaTensor meta_out(&dense_out);
-  ElementwiseInferMeta(x, y, &meta_out);
-  FloorDivideKernel<T, Context>(dev_ctx, x, y, &dense_out);
-  return dense_out;
-}
-
-template <typename T, typename Context>
-DenseTensor ElementwisePow(const Context& dev_ctx,
-                           const DenseTensor& x,
-                           const DenseTensor& y) {
-  DenseTensor dense_out;
-  MetaTensor meta_out(&dense_out);
-  ElementwiseInferMeta(x, y, &meta_out);
-  ElementwisePowKernel<T, Context>(dev_ctx, x, y, &dense_out);
-  return dense_out;
-}
-
 }  // namespace phi
diff --git a/paddle/phi/kernels/funcs/elementwise_functor.h b/paddle/phi/kernels/funcs/elementwise_functor.h
index 0ea5ff0e82e7659de57732745c4a51e007233375..1e39cf55035129cc0df92e689cb44a32aeba5562 100644
--- a/paddle/phi/kernels/funcs/elementwise_functor.h
+++ b/paddle/phi/kernels/funcs/elementwise_functor.h
@@ -538,40 +538,5 @@ struct InverseModuloFunctor<
     return res;
   }
 };
-
-template <typename T>
-struct FloorDivideFunctor {
-  inline HOSTDEVICE T operator()(const T a, const T b) const {
-    PADDLE_ENFORCE(b != 0, DIV_ERROR_INFO);
-    return static_cast<T>(std::trunc(a / b));
-  }
-};
-
-template <typename T>
-struct InverseFloorDivideFunctor {
-  inline HOSTDEVICE T operator()(const T a, const T b) const {
-    PADDLE_ENFORCE(a != 0, DIV_ERROR_INFO);
-    return static_cast<T>(std::trunc(b / a));
-  }
-};
-
-template <typename T>
-struct ElementwisePowFunctor {
-  inline HOSTDEVICE T operator()(const T a, const T b) const {
-// TODO(wujionghao): A potential speed improvement is supporting different
-// types in C++.
-#if defined(__CUDA_ARCH__) || defined(__HIPCC__)
-    // On CUDAPlace, std::pow(3, 1) calls pow(float, float), and
-    // it will return a float number like 2.99... , which floor to 2
-    // when cast to int by default and it is wrong.
-    // Use llrint to cast it to the nearest integer, which is 3.
-    if (std::is_integral<T>::value) {
-      return std::llrint(
-          std::pow(static_cast<double>(a), static_cast<double>(b)));
-    }
-#endif
-    return std::pow(a, b);
-  }
-};
 }  // namespace funcs
 }  // namespace phi
diff --git a/paddle/phi/kernels/gpu/elementwise_grad_kernel.cu b/paddle/phi/kernels/gpu/elementwise_grad_kernel.cu
index 3750e4b2bd61e1310077510ca6c19dc100d7e935..52819fd3dee8c5a0314f4e95e563e3f3db32a136 100644
--- a/paddle/phi/kernels/gpu/elementwise_grad_kernel.cu
+++ b/paddle/phi/kernels/gpu/elementwise_grad_kernel.cu
@@ -382,11 +382,3 @@ PD_REGISTER_KERNEL(minimum_grad,
                    int64_t,
                    phi::dtype::float16,
                    phi::dtype::bfloat16) {}
-PD_REGISTER_KERNEL(elementwise_pow_grad,
-                   GPU,
-                   ALL_LAYOUT,
-                   phi::ElementwisePowGradKernel,
-                   float,
-                   double,
-                   int,
-                   int64_t) {}
diff --git a/paddle/phi/kernels/gpu/elementwise_kernel.cu b/paddle/phi/kernels/gpu/elementwise_kernel.cu
index 73964d31a34eebaa257095839900f82c127d28ce..bd6995cb1353c71ad9a0cbda857c25f0beeccf7e 100644
--- a/paddle/phi/kernels/gpu/elementwise_kernel.cu
+++ b/paddle/phi/kernels/gpu/elementwise_kernel.cu
@@ -55,10 +55,6 @@ DEFINE_CUDA_ELEMENTWISE_OP(Maximum)
 DEFINE_CUDA_ELEMENTWISE_OP(Minimum)
 // Create the definition of Modulo
 DEFINE_CUDA_ELEMENTWISE_OP(Modulo)
-// Create the definition of FloorDivide
-DEFINE_CUDA_ELEMENTWISE_OP(FloorDivide)
-// Create the definition of Pow
-DEFINE_CUDA_ELEMENTWISE_OP(ElementwisePow)
 
 }  // namespace phi
 
@@ -152,17 +148,3 @@ PD_REGISTER_KERNEL(modulo_raw,
                    double,
                    int,
                    int64_t) {}
-PD_REGISTER_KERNEL(floor_divide_raw,
-                   GPU,
-                   ALL_LAYOUT,
-                   phi::FloorDivideRawKernel,
-                   int,
-                   int64_t) {}
-PD_REGISTER_KERNEL(elementwise_pow_raw,
-                   GPU,
-                   ALL_LAYOUT,
-                   phi::ElementwisePowRawKernel,
-                   float,
-                   double,
-                   int,
-                   int64_t) {}
diff --git a/paddle/phi/kernels/impl/elementwise_grad_kernel_impl.h b/paddle/phi/kernels/impl/elementwise_grad_kernel_impl.h
index aba4a5f5fbd43c5baaa770cbee9d3f3f45c13783..07e5bf9ae040e10e1f339b0100ba7f8daccd438f 100644
--- a/paddle/phi/kernels/impl/elementwise_grad_kernel_impl.h
+++ b/paddle/phi/kernels/impl/elementwise_grad_kernel_impl.h
@@ -666,44 +666,4 @@ struct MinGradDy {
     return dout * static_cast<T>(x >= y);
   }
 };
-
-template <typename T>
-struct PowGradDX {
-  HOSTDEVICE T operator()(T x, T y, T out, T dout) const {
-#if defined(__CUDA_ARCH__) || defined(__HIPCC__)
-    if (std::is_integral<T>::value) {
-      return dout * y *
-             std::pow(static_cast<double>(x), static_cast<double>(y - 1));
-    }
-#endif
-    return dout * y * std::pow(x, y - 1);
-  }
-};
-
-template <typename T, typename Enable = void>
-struct PowGradDY {
-  HOSTDEVICE T operator()(T x, T y, T out, T dout) const {
-#if defined(__CUDA_ARCH__) || defined(__HIPCC__)
-    if (std::is_integral<T>::value) {
-      return dout * std::log(static_cast<double>(x)) *
-             std::pow(static_cast<double>(x), static_cast<double>(y));
-    }
-#endif
-    return dout * std::log(x) * std::pow(x, y);
-  }
-};
-
-template <typename T, typename Context>
-void ElementwisePowGradKernel(const Context& dev_ctx,
-                              const DenseTensor& x,
-                              const DenseTensor& y,
-                              const DenseTensor& dout,
-                              int axis,
-                              DenseTensor* dx,
-                              DenseTensor* dy) {
-  funcs::ElementwiseGradPreProcess(dout, dx);
-  phi::funcs::ElemwiseGradCompute<Context, T, PowGradDX<T>, PowGradDY<T>>(
-      dev_ctx, x, y, dout, dout, axis, dx, dy, PowGradDX<T>(), PowGradDY<T>());
-}
-
 }  // namespace phi
diff --git a/paddle/phi/ops/compat/elementwise_sig.cc b/paddle/phi/ops/compat/elementwise_sig.cc
index cf6f9d4dfb2f11660361ac6bbdef72d4c308c348..7f00af6f9af86f35709c2d120a7a47917d8d8431 100644
--- a/paddle/phi/ops/compat/elementwise_sig.cc
+++ b/paddle/phi/ops/compat/elementwise_sig.cc
@@ -82,24 +82,6 @@ KernelSignature ElementwiseModOpArgumentMapping(
   return KernelSignature("modulo_raw", {"X", "Y"}, {"axis"}, {"Out"});
 }
 
-KernelSignature ElementwiseFloorDivOpArgumentMapping(
-    const ArgumentMappingContext& ctx) {
-  int axis = paddle::any_cast<int>(ctx.Attr("axis"));
-  if (axis == -1) {
-    return KernelSignature("floor_divide", {"X", "Y"}, {}, {"Out"});
-  }
-  return KernelSignature("floor_divide_raw", {"X", "Y"}, {"axis"}, {"Out"});
-}
-
-KernelSignature ElementwisePowOpArgumentMapping(
-    const ArgumentMappingContext& ctx) {
-  int axis = paddle::any_cast<int>(ctx.Attr("axis"));
-  if (axis == -1) {
-    return KernelSignature("elementwise_pow", {"X", "Y"}, {}, {"Out"});
-  }
-  return KernelSignature("elementwise_pow_raw", {"X", "Y"}, {"axis"}, {"Out"});
-}
-
 KernelSignature ElementwiseAddGradOpArgumentMapping(
     const ArgumentMappingContext& ctx) {
   return KernelSignature("add_grad",
@@ -218,13 +200,6 @@ KernelSignature ElementwiseMinGradOpArgumentMapping(
                          {"axis"},
                          {GradVarName("X"), GradVarName("Y")});
 }
-KernelSignature ElementwisePowGradOpArgumentMapping(
-    const ArgumentMappingContext& ctx) {
-  return KernelSignature("elementwise_pow_grad",
-                         {"X", "Y", GradVarName("Out")},
-                         {"axis"},
-                         {GradVarName("X"), GradVarName("Y")});
-}
 }  // namespace phi
 
 PD_REGISTER_BASE_KERNEL_NAME(elementwise_add, add);
@@ -234,7 +209,6 @@ PD_REGISTER_BASE_KERNEL_NAME(elementwise_div, divide);
 PD_REGISTER_BASE_KERNEL_NAME(elementwise_max, maximum);
 PD_REGISTER_BASE_KERNEL_NAME(elementwise_min, minimum);
 PD_REGISTER_BASE_KERNEL_NAME(elementwise_mod, modulo);
-PD_REGISTER_BASE_KERNEL_NAME(elementwise_floordiv, floor_divide);
 PD_REGISTER_BASE_KERNEL_NAME(elementwise_add_grad, add_grad);
 PD_REGISTER_BASE_KERNEL_NAME(elementwise_add_grad_grad, add_double_grad);
 PD_REGISTER_BASE_KERNEL_NAME(elementwise_add_triple_grad, add_triple_grad);
@@ -266,10 +240,6 @@ PD_REGISTER_ARG_MAPPING_FN(elementwise_min,
                            phi::ElementwiseMinOpArgumentMapping);
 PD_REGISTER_ARG_MAPPING_FN(elementwise_mod,
                            phi::ElementwiseModOpArgumentMapping);
-PD_REGISTER_ARG_MAPPING_FN(elementwise_floordiv,
-                           phi::ElementwiseFloorDivOpArgumentMapping);
-PD_REGISTER_ARG_MAPPING_FN(elementwise_pow,
-                           phi::ElementwisePowOpArgumentMapping);
 PD_REGISTER_ARG_MAPPING_FN(elementwise_add_grad,
                            phi::ElementwiseAddGradOpArgumentMapping);
 PD_REGISTER_ARG_MAPPING_FN(elementwise_add_grad_grad,
@@ -302,5 +272,3 @@ PD_REGISTER_ARG_MAPPING_FN(elementwise_max_grad,
                            phi::ElementwiseMaxGradOpArgumentMapping);
 PD_REGISTER_ARG_MAPPING_FN(elementwise_min_grad,
                            phi::ElementwiseMinGradOpArgumentMapping);
-PD_REGISTER_ARG_MAPPING_FN(elementwise_pow_grad,
-                           phi::ElementwisePowGradOpArgumentMapping);