diff --git a/paddle/fluid/operators/dgc_op.h b/paddle/fluid/operators/dgc_op.h
index ac9c440076257a11e265acc4e18047c06c797609..b1bf5e2778167ba9a8e39803c4369be12823259b 100644
--- a/paddle/fluid/operators/dgc_op.h
+++ b/paddle/fluid/operators/dgc_op.h
@@ -15,9 +15,11 @@ limitations under the License. */
 #pragma once
 #include <vector>
 #include "dgc/dgc.h"
+
 #include "paddle/fluid/framework/eigen.h"
 #include "paddle/fluid/memory/malloc.h"
-#include "paddle/fluid/operators/elementwise/elementwise_add_op.h"
+#include "paddle/fluid/operators/elementwise/elementwise_op_function.h"
+#include "paddle/phi/kernels/funcs/elementwise_functor.h"
 
 namespace paddle {
 namespace operators {
@@ -153,18 +155,18 @@ class DGCOpKernel : public framework::OpKernel<T> {
       u_out_e.device(eigen_ctx) = m * (u_e + grad_out_e);
 
       // v = u + v + g
-      ElementwiseComputeEx<AddFunctor<T>, DeviceContext, T>(
-          ctx, u, v, 0, AddFunctor<T>(), v_out);
+      ElementwiseComputeEx<phi::funcs::AddFunctor<T>, DeviceContext, T>(
+          ctx, u, v, 0, phi::funcs::AddFunctor<T>(), v_out);
 
-      ElementwiseComputeEx<AddFunctor<T>, DeviceContext, T>(
-          ctx, g, v, 0, AddFunctor<T>(), v_out);
+      ElementwiseComputeEx<phi::funcs::AddFunctor<T>, DeviceContext, T>(
+          ctx, g, v, 0, phi::funcs::AddFunctor<T>(), v_out);
     } else {
       // u = m * u + g
       u_out_e.device(eigen_ctx) = m * u_e + grad_out_e;
 
       // v = u + v
-      ElementwiseComputeEx<AddFunctor<T>, DeviceContext, T>(
-          ctx, u, v, 0, AddFunctor<T>(), v_out);
+      ElementwiseComputeEx<phi::funcs::AddFunctor<T>, DeviceContext, T>(
+          ctx, u, v, 0, phi::funcs::AddFunctor<T>(), v_out);
     }
 
     T* v_out_data = v_out->mutable_data<T>(ctx.GetPlace());
diff --git a/paddle/fluid/operators/elementwise/elementwise_add_op.cc b/paddle/fluid/operators/elementwise/elementwise_add_op.cc
index 53037c1fa653648044e2dc0981ec5c63351e7c15..ed9b98a128a2166907c70e46fafa31d3b9c1cb35 100644
--- a/paddle/fluid/operators/elementwise/elementwise_add_op.cc
+++ b/paddle/fluid/operators/elementwise/elementwise_add_op.cc
@@ -12,8 +12,6 @@ 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_add_op.h"
-
 #include <string>
 
 #include "paddle/fluid/operators/elementwise/elementwise_op.h"
@@ -125,17 +123,6 @@ REGISTER_OPERATOR(
     paddle::framework::EmptyGradOpMaker<paddle::framework::OpDesc>,
     paddle::framework::EmptyGradOpMaker<paddle::imperative::OpBase>);
 
-REGISTER_OP_CPU_KERNEL(
-    grad_add,
-    ops::ElementwiseAddKernel<paddle::platform::CPUDeviceContext, float>,
-    ops::ElementwiseAddKernel<paddle::platform::CPUDeviceContext, double>,
-    ops::ElementwiseAddKernel<paddle::platform::CPUDeviceContext, int>,
-    ops::ElementwiseAddKernel<paddle::platform::CPUDeviceContext, int64_t>,
-    ops::ElementwiseAddKernel<paddle::platform::CPUDeviceContext,
-                              paddle::platform::complex<float>>,
-    ops::ElementwiseAddKernel<paddle::platform::CPUDeviceContext,
-                              paddle::platform::complex<double>>);
-
 REGISTER_OP_VERSION(elementwise_add)
     .AddCheckpoint(
         R"ROC(Register elementwise_add for adding the attribute of
diff --git a/paddle/fluid/operators/elementwise/elementwise_add_op.h b/paddle/fluid/operators/elementwise/elementwise_add_op.h
deleted file mode 100644
index d77d4ed036394e50f1cf6affcfa7045636df6ac0..0000000000000000000000000000000000000000
--- a/paddle/fluid/operators/elementwise/elementwise_add_op.h
+++ /dev/null
@@ -1,66 +0,0 @@
-/* Copyright (c) 2016 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
-#ifdef __xpu__
-#include <memory>
-#include <string>
-#include "paddle/fluid/operators/elementwise/elementwise_op.h"
-#include "paddle/fluid/operators/elementwise/elementwise_op_broadcast.cu.h"
-#include "paddle/fluid/operators/elementwise/elementwise_xpu.h"
-#include "paddle/fluid/platform/device/device_wrapper.h"
-#else
-#include <algorithm>
-#include <utility>
-#include "paddle/fluid/operators/elementwise/elementwise_op.h"
-
-// only can include the headers in paddle/phi/include dirs
-#include "paddle/phi/kernels/elementwise_add_grad_kernel.h"
-#include "paddle/phi/kernels/elementwise_add_kernel.h"
-#endif
-
-namespace paddle {
-namespace operators {
-
-template <typename DeviceContext, typename T>
-class ElementwiseAddKernel : public framework::OpKernel<T> {
- public:
-  void Compute(const framework::ExecutionContext& ctx) const override {
-#ifdef __xpu__
-    std::vector<const framework::Tensor*> ins;
-    std::vector<framework::Tensor*> outs;
-    int axis = PackTensorsIntoVector<T>(ctx, &ins, &outs);
-    const auto& xpu_ctx =
-        ctx.template device_context<paddle::platform::XPUDeviceContext>();
-    paddle::operators::LaunchElementwiseCudaKernel<ElementwiseType::kBinary, T,
-                                                   T, kps::AddFunctor<T>, 1>(
-        xpu_ctx, ins, &outs, axis, kps::AddFunctor<T>());
-#else
-    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());
-
-    auto &dev_ctx = ctx.device_context<DeviceContext>();
-    int axis = ctx.Attr<int>("axis");
-    phi::AddRawKernel<T>(
-        static_cast<const typename framework::ConvertToPhiContext<
-            DeviceContext>::TYPE &>(dev_ctx),
-        *x, *y, axis, z);
-#endif
-  }
-};
-
-}  // namespace operators
-}  // namespace paddle
diff --git a/paddle/fluid/operators/elementwise/elementwise_add_op.kps b/paddle/fluid/operators/elementwise/elementwise_add_op.kps
deleted file mode 100644
index ecd52a310acdbebd648da5ed0f1abddb298224ef..0000000000000000000000000000000000000000
--- a/paddle/fluid/operators/elementwise/elementwise_add_op.kps
+++ /dev/null
@@ -1,61 +0,0 @@
-/* Copyright (c) 2022 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. */
-
-#ifdef PADDLE_WITH_XPU_KP
-
-// Please do not modify the following code
-#if defined(__CUDA_ARCH__)
-#undef __CUDA_ARCH__
-#endif
-
-#if defined(__CUDACC__)
-#undef __CUDACC__
-#endif
-
-#if defined(__CUDA__)
-#undef __CUDA__
-#endif
-
-#if defined(__NVCC__)
-#undef __NVCC__
-#endif
-
-#include <xpu/runtime.h>                // NOLINT
-#include "xpu/kernel/cluster_header.h"  // NOLINT
-#include "xpu/kernel/debug.h"           // NOLINT
-#include "xpu/kernel/math.h"            // NOLINT
-
-#include "paddle/fluid/operators/elementwise/elementwise_add_op.h"
-#else
-#include "paddle/fluid/operators/elementwise/elementwise_add_op.h"
-#include "paddle/phi/kernels/gpu/elementwise_grad.h"
-#endif
-
-namespace ops = paddle::operators;
-namespace plat = paddle::platform;
-
-#ifdef PADDLE_WITH_XPU_KP
-REGISTER_OP_KERNEL(elementwise_add, KP, plat::XPUPlace,
-                   ops::ElementwiseAddKernel<plat::XPUDeviceContext, float>);
-#else
-REGISTER_OP_CUDA_KERNEL(
-    grad_add, ops::ElementwiseAddKernel<plat::CUDADeviceContext, float>,
-    ops::ElementwiseAddKernel<plat::CUDADeviceContext, double>,
-    ops::ElementwiseAddKernel<plat::CUDADeviceContext, int>,
-    ops::ElementwiseAddKernel<plat::CUDADeviceContext, int64_t>,
-    ops::ElementwiseAddKernel<plat::CUDADeviceContext, plat::float16>,
-    ops::ElementwiseAddKernel<plat::CUDADeviceContext, plat::bfloat16>,
-    ops::ElementwiseAddKernel<plat::CUDADeviceContext, plat::complex<float>>,
-    ops::ElementwiseAddKernel<plat::CUDADeviceContext, plat::complex<double>>);
-#endif
diff --git a/paddle/fluid/operators/elementwise/elementwise_add_op_npu.cc b/paddle/fluid/operators/elementwise/elementwise_add_op_npu.cc
index d35e3f6641b4548e8176c1f2ee14a01a6f3824e8..178aa329577b7104bf91bd329e8a4e71ef5437a3 100644
--- a/paddle/fluid/operators/elementwise/elementwise_add_op_npu.cc
+++ b/paddle/fluid/operators/elementwise/elementwise_add_op_npu.cc
@@ -16,7 +16,6 @@ limitations under the License. */
 #include <string>
 
 #include "paddle/fluid/framework/tensor_util.h"
-#include "paddle/fluid/operators/elementwise/elementwise_add_op.h"
 #include "paddle/fluid/operators/elementwise/elementwise_npu.h"
 #include "paddle/fluid/platform/device/npu/npu_op_runner.h"
 
diff --git a/paddle/fluid/operators/elementwise/elementwise_add_op_xpu.cc b/paddle/fluid/operators/elementwise/elementwise_add_op_xpu.cc
index feb73abf3ff08ced63ba84badb7920a0dcdc07a2..22a5de4c6094167ee4e87d5d2ab67d341f897bbe 100644
--- a/paddle/fluid/operators/elementwise/elementwise_add_op_xpu.cc
+++ b/paddle/fluid/operators/elementwise/elementwise_add_op_xpu.cc
@@ -13,7 +13,6 @@ See the License for the specific language governing permissions and
 limitations under the License. */
 
 #ifdef PADDLE_WITH_XPU
-#include "paddle/fluid/operators/elementwise/elementwise_add_op.h"
 #include <memory>
 #include <string>
 #include "paddle/fluid/operators/elementwise/elementwise_op.h"
diff --git a/paddle/fluid/operators/fused/attn_gemm.h b/paddle/fluid/operators/fused/attn_gemm.h
index c4e73c6bf97fdb13177a122c1b8cb433a21e30fb..9542f0742ea349bcb61cc8dbc2cd8942d16d16e0 100644
--- a/paddle/fluid/operators/fused/attn_gemm.h
+++ b/paddle/fluid/operators/fused/attn_gemm.h
@@ -14,9 +14,10 @@ limitations under the License. */
 #include "paddle/fluid/platform/float16.h"
 #include "paddle/phi/kernels/funcs/blas/blas.h"
 
-#include "paddle/fluid/operators/elementwise/elementwise_op_broadcast.cu.h"
 #include "paddle/fluid/operators/kernel_primitives/kernel_primitives.h"
 #include "paddle/fluid/operators/reduce_ops/reduce_op.cu.h"
+#include "paddle/phi/kernels/funcs/broadcast_function.h"
+#include "paddle/phi/kernels/funcs/elementwise_functor.h"
 
 namespace paddle {
 namespace operators {
@@ -67,9 +68,8 @@ class AttnMatMul {
       ins.emplace_back(bias);
       outs.emplace_back(bias_out);
       int elewise_add_axis = -1;
-      paddle::operators::LaunchElementwiseCudaKernel<ElementwiseType::kBinary,
-                                                     T, T>(
-          dev_ctx_, ins, &outs, elewise_add_axis, AddFunctor<T>());
+      phi::funcs::BroadcastKernel<phi::ElementwiseType::kBinary, T, T>(
+          dev_ctx_, ins, &outs, elewise_add_axis, phi::funcs::AddFunctor<T>());
     }
   }
 
diff --git a/paddle/fluid/operators/fused/fmha_ref.h b/paddle/fluid/operators/fused/fmha_ref.h
index 3d75d127ab60aec17d7d089498752bf2e6d075e0..0e9fba73933b7fd7c976d090f9297ccba6c5444b 100644
--- a/paddle/fluid/operators/fused/fmha_ref.h
+++ b/paddle/fluid/operators/fused/fmha_ref.h
@@ -12,12 +12,12 @@ limitations under the License. */
 #pragma once
 
 #include "paddle/fluid/operators/dropout_impl.cu.h"
-#include "paddle/fluid/operators/elementwise/elementwise_add_op.h"
-#include "paddle/fluid/operators/elementwise/elementwise_op_broadcast.cu.h"
 #include "paddle/fluid/operators/fused/fused_softmax_mask.cu.h"
 #include "paddle/fluid/operators/transpose_op.cu.h"
+#include "paddle/phi/kernels/funcs/broadcast_function.h"
 #include "paddle/phi/kernels/funcs/concat_and_split_functor.h"
 #include "paddle/phi/kernels/funcs/elementwise_base.h"
+#include "paddle/phi/kernels/funcs/elementwise_functor.h"
 #include "paddle/phi/kernels/funcs/functors.h"
 #include "paddle/phi/kernels/gpudnn/softmax_gpudnn.h"
 
@@ -160,9 +160,9 @@ class FMHARef {
         ins.emplace_back(src_mask_tensor);
         outs.emplace_back(src_mask_out_tensor);
         int elewise_add_axis = -1;
-        paddle::operators::LaunchElementwiseCudaKernel<ElementwiseType::kBinary,
-                                                       T, T>(
-            dev_ctx_, ins, &outs, elewise_add_axis, AddFunctor<T>());
+        phi::funcs::BroadcastKernel<phi::ElementwiseType::kBinary, T, T>(
+            dev_ctx_, ins, &outs, elewise_add_axis,
+            phi::funcs::AddFunctor<T>());
 
         phi::SoftmaxForwardCUDAKernelDriver<T>(
             dev_ctx_, *src_mask_out_tensor, softmax_axis, softmax_out_tensor);
diff --git a/paddle/fluid/operators/fused/fused_attention_op.cu b/paddle/fluid/operators/fused/fused_attention_op.cu
index d26577f06fe683fb1528c61b4401b9e578c90c9f..ec8a4d962e8084f8c8dfb74f9b789ef5d94bb00c 100644
--- a/paddle/fluid/operators/fused/fused_attention_op.cu
+++ b/paddle/fluid/operators/fused/fused_attention_op.cu
@@ -19,7 +19,8 @@ limitations under the License. */
 #include "paddle/fluid/platform/device/gpu/gpu_device_function.h"
 #include "paddle/fluid/platform/device/gpu/gpu_dnn.h"
 
-#include "paddle/fluid/operators/elementwise/elementwise_add_op.h"
+#include "paddle/phi/kernels/funcs/broadcast_function.h"
+#include "paddle/phi/kernels/funcs/elementwise_functor.h"
 #include "paddle/phi/kernels/funcs/math_function.h"
 
 #include "paddle/fluid/operators/fused/attention_layer_norm.h"
@@ -543,10 +544,9 @@ class FusedAttentionGradKernel : public framework::OpKernel<T> {
     ins.emplace_back(d_x);
     outs.emplace_back(d_x);
     int elewise_add_axis = -1;
-    paddle::operators::LaunchElementwiseCudaKernel<ElementwiseType::kBinary, T,
-                                                   T>(
+    phi::funcs::BroadcastKernel<phi::ElementwiseType::kBinary, T, T>(
         ctx.cuda_device_context(), ins, &outs, elewise_add_axis,
-        AddFunctor<T>());
+        phi::funcs::AddFunctor<T>());
   }
 };
 
diff --git a/paddle/fluid/operators/fused/fused_feedforward_op.cu b/paddle/fluid/operators/fused/fused_feedforward_op.cu
index c38d9f7d4bcbd25b3111b35a918de0f4ebdabefb..2eb9885286dab8702cd96519a76b4bb69ed6aff6 100644
--- a/paddle/fluid/operators/fused/fused_feedforward_op.cu
+++ b/paddle/fluid/operators/fused/fused_feedforward_op.cu
@@ -17,9 +17,10 @@ limitations under the License. */
 #include "paddle/fluid/operators/matmul_v2_op.h"
 #include "paddle/phi/kernels/funcs/blas/blas.h"
 
-#include "paddle/fluid/operators/elementwise/elementwise_add_op.h"
 #include "paddle/fluid/operators/fused/fused_dropout_helper.h"
 #include "paddle/fluid/operators/layer_norm_kernel.cu.h"
+#include "paddle/phi/kernels/funcs/broadcast_function.h"
+#include "paddle/phi/kernels/funcs/elementwise_functor.h"
 
 #if defined(PADDLE_WITH_NCCL) || defined(PADDLE_WITH_RCCL)
 #include "paddle/fluid/platform/collective_helper.h"
@@ -345,9 +346,8 @@ class FusedFeedForwardGradKernel : public framework::OpKernel<T> {
     ins[1] = d_x;
     outs[0] = d_x;
     int elewise_add_axis = -1;
-    paddle::operators::LaunchElementwiseCudaKernel<ElementwiseType::kBinary, T,
-                                                   T>(
-        ctx, ins, &outs, elewise_add_axis, AddFunctor<T>());
+    phi::funcs::BroadcastKernel<phi::ElementwiseType::kBinary, T, T>(
+        ctx, ins, &outs, elewise_add_axis, phi::funcs::AddFunctor<T>());
   }
 
   void Compute(const framework::ExecutionContext& context) const override {
diff --git a/paddle/fluid/operators/fused/fused_multi_transformer_op.cu b/paddle/fluid/operators/fused/fused_multi_transformer_op.cu
index fdd0208c3d316e9386eba2a60dde85fbab6005ae..fe93d323c59bcd83b567b60fa00333fb0275b115 100644
--- a/paddle/fluid/operators/fused/fused_multi_transformer_op.cu
+++ b/paddle/fluid/operators/fused/fused_multi_transformer_op.cu
@@ -24,7 +24,6 @@ limitations under the License. */
 #include "paddle/fluid/platform/device/gpu/gpu_device_function.h"
 #include "paddle/fluid/platform/device/gpu/gpu_dnn.h"
 
-#include "paddle/fluid/operators/elementwise/elementwise_add_op.h"
 #include "paddle/phi/kernels/funcs/math_function.h"
 
 #include "paddle/fluid/operators/fused/attention_layer_norm.h"
diff --git a/paddle/phi/kernels/cpu/elementwise_add_kernel.cc b/paddle/phi/kernels/cpu/elementwise_add_kernel.cc
index 607026454724937db841150bdf96d6ea8b976a4c..5019b9f57062874b5f78a002dbf1cdd411bc4e9c 100644
--- a/paddle/phi/kernels/cpu/elementwise_add_kernel.cc
+++ b/paddle/phi/kernels/cpu/elementwise_add_kernel.cc
@@ -34,6 +34,14 @@ void AddKernel(const Context& dev_ctx,
   AddRawKernel<T>(dev_ctx, x, y, axis, out);
 }
 
+template <typename T, typename Context>
+void GradAddKernel(const Context& dev_ctx,
+                   const DenseTensor& x,
+                   const DenseTensor& y,
+                   DenseTensor* out) {
+  AddKernel<T>(dev_ctx, x, y, out);
+}
+
 }  // namespace phi
 
 using complex64 = ::phi::dtype::complex<float>;
@@ -65,3 +73,15 @@ PD_REGISTER_KERNEL(add,
                    int64_t,
                    complex64,
                    complex128) {}
+
+PD_REGISTER_KERNEL(grad_add,
+                   CPU,
+                   ALL_LAYOUT,
+                   phi::GradAddKernel,
+                   float,
+                   double,
+                   int16_t,
+                   int,
+                   int64_t,
+                   complex64,
+                   complex128) {}
diff --git a/paddle/phi/kernels/kps/elementwise_add_kernel.cu b/paddle/phi/kernels/kps/elementwise_add_kernel.cu
index 8f7d45771d9d0694b4ed116c6a4c36955fea60d9..98e39ada32b8be0fa0efd3f5f711f13881383167 100644
--- a/paddle/phi/kernels/kps/elementwise_add_kernel.cu
+++ b/paddle/phi/kernels/kps/elementwise_add_kernel.cu
@@ -33,6 +33,14 @@ void AddKernel(const Context& dev_ctx,
   AddRawKernel<T>(dev_ctx, x, y, axis, out);
 }
 
+template <typename T, typename Context>
+void GradAddKernel(const Context& dev_ctx,
+                   const DenseTensor& x,
+                   const DenseTensor& y,
+                   DenseTensor* out) {
+  AddKernel<T>(dev_ctx, x, y, out);
+}
+
 }  // namespace phi
 
 #ifdef PADDLE_WITH_XPU_KP
@@ -71,4 +79,18 @@ PD_REGISTER_KERNEL(add,
                    phi::dtype::bfloat16,
                    complex64,
                    complex128) {}
+
+PD_REGISTER_KERNEL(grad_add,
+                   KPS,
+                   ALL_LAYOUT,
+                   phi::GradAddKernel,
+                   float,
+                   double,
+                   int16_t,
+                   int,
+                   int64_t,
+                   phi::dtype::float16,
+                   phi::dtype::bfloat16,
+                   complex64,
+                   complex128) {}
 #endif
diff --git a/paddle/phi/ops/compat/elementwise_sig.cc b/paddle/phi/ops/compat/elementwise_sig.cc
index c760c966b0647d0ab68834f20bba01e5eb791f82..17fb1858373d9581e62ed89cd3e05abe744362fe 100644
--- a/paddle/phi/ops/compat/elementwise_sig.cc
+++ b/paddle/phi/ops/compat/elementwise_sig.cc
@@ -25,6 +25,11 @@ KernelSignature ElementwiseAddOpArgumentMapping(
   return KernelSignature("add_raw", {"X", "Y"}, {"axis"}, {"Out"});
 }
 
+KernelSignature ElementwiseGradAddOpArgumentMapping(
+    const ArgumentMappingContext& ctx) {
+  return KernelSignature("grad_add", {"X", "Y"}, {}, {"Out"});
+}
+
 KernelSignature ElementwiseSubOpArgumentMapping(
     const ArgumentMappingContext& ctx) {
   int axis = paddle::any_cast<int>(ctx.Attr("axis"));
@@ -317,3 +322,4 @@ PD_REGISTER_ARG_MAPPING_FN(elementwise_heaviside_grad,
                            phi::ElementwiseHeavisideGradOpArgumentMapping);
 PD_REGISTER_ARG_MAPPING_FN(elementwise_pow_grad,
                            phi::ElementwisePowGradOpArgumentMapping);
+PD_REGISTER_ARG_MAPPING_FN(grad_add, phi::ElementwiseGradAddOpArgumentMapping);