diff --git a/paddle/fluid/operators/concat_op.cc b/paddle/fluid/operators/concat_op.cc
index 9eba127a9b3ceace225e3d3dcf867df518c4477e..68a4d09f3b92dcaa81390014640ddfa1afeb31dc 100644
--- a/paddle/fluid/operators/concat_op.cc
+++ b/paddle/fluid/operators/concat_op.cc
@@ -253,19 +253,7 @@ REGISTER_OPERATOR(concat_grad, ops::ConcatOpGrad,
                   ops::ConcatDoubleGradOpMaker<paddle::framework::OpDesc>,
                   ops::ConcatDoubleGradOpMaker<paddle::imperative::OpBase>,
                   ops::ConcatOpGradNoNeedBufferVarInferer);
-REGISTER_OP_CPU_KERNEL(
-    concat, ops::ConcatKernel<paddle::platform::CPUDeviceContext, double>,
-    ops::ConcatKernel<paddle::platform::CPUDeviceContext, float>,
-    ops::ConcatKernel<paddle::platform::CPUDeviceContext, bool>,
-    ops::ConcatKernel<paddle::platform::CPUDeviceContext, int64_t>,
-    ops::ConcatKernel<paddle::platform::CPUDeviceContext,
-                      paddle::platform::float16>,
-    ops::ConcatKernel<paddle::platform::CPUDeviceContext, int>,
-    ops::ConcatKernel<paddle::platform::CPUDeviceContext, uint8_t>,
-    ops::ConcatKernel<paddle::platform::CPUDeviceContext,
-                      paddle::platform::complex<float>>,
-    ops::ConcatKernel<paddle::platform::CPUDeviceContext,
-                      paddle::platform::complex<double>>);
+
 REGISTER_OP_CPU_KERNEL(
     concat_grad,
     ops::ConcatGradKernel<paddle::platform::CPUDeviceContext, double>,
diff --git a/paddle/fluid/operators/concat_op.cu.cc b/paddle/fluid/operators/concat_op.cu.cc
index 2be763298575490932f6cd3867dd59cee0ebeba7..3aee2c6504fdd7bf2a803f8757a974728aa262dd 100644
--- a/paddle/fluid/operators/concat_op.cu.cc
+++ b/paddle/fluid/operators/concat_op.cu.cc
@@ -18,18 +18,7 @@ limitations under the License. */
 
 namespace ops = paddle::operators;
 namespace plat = paddle::platform;
-REGISTER_OP_CUDA_KERNEL(
-    concat, ops::ConcatKernel<paddle::platform::CUDADeviceContext, double>,
-    ops::ConcatKernel<paddle::platform::CUDADeviceContext, float>,
-    ops::ConcatKernel<paddle::platform::CUDADeviceContext, bool>,
-    ops::ConcatKernel<paddle::platform::CUDADeviceContext, plat::float16>,
-    ops::ConcatKernel<paddle::platform::CUDADeviceContext, int64_t>,
-    ops::ConcatKernel<paddle::platform::CUDADeviceContext, int>,
-    ops::ConcatKernel<paddle::platform::CUDADeviceContext, uint8_t>,
-    ops::ConcatKernel<paddle::platform::CUDADeviceContext,
-                      plat::complex<float>>,
-    ops::ConcatKernel<paddle::platform::CUDADeviceContext,
-                      plat::complex<double>>);
+
 REGISTER_OP_CUDA_KERNEL(
     concat_grad,
     ops::ConcatGradKernel<paddle::platform::CUDADeviceContext, double>,
diff --git a/paddle/fluid/operators/concat_op.h b/paddle/fluid/operators/concat_op.h
index 1d9c10bdb8cc6a698a4a1b6ab376e90b67eb2a03..7b53b9df6f95134f3aaafa7c34bef71eaf805d3c 100644
--- a/paddle/fluid/operators/concat_op.h
+++ b/paddle/fluid/operators/concat_op.h
@@ -39,54 +39,6 @@ static inline int64_t ComputeAxis(int64_t axis, int64_t rank) {
   }
   return axis > 0 ? axis : 0;
 }
-
-template <typename DeviceContext, typename T>
-class ConcatKernel : public framework::OpKernel<T> {
- public:
-  void Compute(const framework::ExecutionContext& ctx) const override {
-    auto ins = ctx.MultiInput<framework::LoDTensor>("X");
-    framework::LoDTensor* out = ctx.Output<framework::LoDTensor>("Out");
-    PADDLE_ENFORCE_NOT_NULL(ins[0],
-                            platform::errors::NotFound(
-                                "The first input tensor is not initalized."));
-    auto axis = ctx.Attr<int>("axis");
-    bool need_resize_out_dims = false;
-    if (ctx.HasInput("AxisTensor")) {
-      auto* axis_tensor = ctx.Input<framework::Tensor>("AxisTensor");
-      axis = GetDataFromTensor<int>(axis_tensor)[0];
-      need_resize_out_dims = true;
-    }
-    axis = ComputeAxis(static_cast<int64_t>(axis),
-                       static_cast<int64_t>(ins[0]->dims().size()));
-
-    if (need_resize_out_dims) {
-      const size_t n = ins.size();
-      std::vector<framework::DDim> ins_dims(n);
-      for (size_t i = 0; i < n; i++) {
-        ins_dims[i] = ins[i]->dims();
-      }
-
-      framework::DDim out_dims =
-          pten::funcs::ComputeAndCheckShape(true, ins_dims, axis);
-      out->Resize(out_dims);
-    }
-    auto place = ctx.GetPlace();
-    out->mutable_data<T>(place);
-
-    // call new kernel
-    auto& dev_ctx = ctx.device_context<DeviceContext>();
-    std::vector<pten::DenseTensor> pt_ins;
-    for (auto& in : ins) {
-      pt_ins.push_back(*in);
-    }
-
-    pten::ConcatKernel<T>(
-        static_cast<const typename paddle::framework::ConvertToPtenContext<
-            DeviceContext>::TYPE&>(dev_ctx),
-        pt_ins, axis, out);
-  }
-};
-
 template <typename DeviceContext, typename T>
 class ConcatGradKernel : public framework::OpKernel<T> {
  public:
diff --git a/paddle/fluid/operators/tensor_array_to_tensor_op.cc b/paddle/fluid/operators/tensor_array_to_tensor_op.cc
index eb20e1c2cd2748a5ab4db28df0c4798837c7bf21..fa49f254d972a38ad54922c8a303654dedc36682 100644
--- a/paddle/fluid/operators/tensor_array_to_tensor_op.cc
+++ b/paddle/fluid/operators/tensor_array_to_tensor_op.cc
@@ -299,7 +299,7 @@ class TensorArrayToTensorGradOpMaker : public framework::SingleGradOpMaker<T> {
 
 }  // namespace operators
 }  // namespace paddle
-USE_OP(concat);
+USE_OP_ITSELF(concat);
 
 namespace ops = paddle::operators;
 REGISTER_OPERATOR(