[Phi]Change Copy from Kernel to basic component utils (#43622)

* perfect copy

* deal with conflict

* deal with conflict

* fix compile bugs

* fix unittest bugs

* change code format

* deal with conflict

* modify code by review

* fix ce bugs

* fix ce bugs

* add lo

* perfect code format

* deal with conflicts

paddle/fluid/eager/tests/data_structure_tests/eager_tensor_test.cc

@@ -21,13 +21,6 @@
 #include "paddle/phi/common/layout.h"
 #include "paddle/phi/core/kernel_registry.h"
 
-PD_DECLARE_KERNEL(copy, CPU, ALL_LAYOUT);
-PD_DECLARE_KERNEL(copy_sr, CPU, ALL_LAYOUT);
-#if defined(PADDLE_WITH_CUDA) || defined(PADDLE_WITH_HIP)
-PD_DECLARE_KERNEL(copy, GPU, ALL_LAYOUT);
-PD_DECLARE_KERNEL(copy_sr, GPU, ALL_LAYOUT);
-#endif
-
 namespace eager_test {
 using AbstractAutogradMeta = paddle::experimental::AbstractAutogradMeta;
 class AutogradMetaTest : public AbstractAutogradMeta {
@@ -212,7 +205,8 @@ TEST(EagerVariable, Constructor) {
 TEST(EagerVariable, DataLayout) {
   paddle::experimental::Tensor tensor;
   phi::DenseTensorMeta meta =
-      phi::DenseTensorMeta(phi::DataType::FLOAT32, phi::make_ddim({1, 1, 1, 1}),
+      phi::DenseTensorMeta(phi::DataType::FLOAT32,
+                           phi::make_ddim({1, 1, 1, 1}),
                            paddle::experimental::DataLayout::UNDEFINED);
   std::shared_ptr<phi::DenseTensor> dt = std::make_shared<phi::DenseTensor>(
       std::make_unique<paddle::experimental::DefaultAllocator>(

paddle/fluid/eager/tests/task_tests/backward_test.cc

@@ -30,7 +30,6 @@
 #include "paddle/phi/core/tensor_meta.h"
 
 PD_DECLARE_KERNEL(full, CPU, ALL_LAYOUT);
-PD_DECLARE_KERNEL(copy, CPU, ALL_LAYOUT);
 PD_DECLARE_KERNEL(add, CPU, ALL_LAYOUT);
 
 namespace egr {
@@ -44,9 +43,12 @@ TEST(Backward, SingleNodeEmptyGrad) {
 
   // Create Target Tensor
   paddle::experimental::Tensor target_tensor =
-      egr_utils_api::CreateTensorWithValue(
-          ddim, paddle::platform::CPUPlace(), phi::DataType::FLOAT32,
-          phi::DataLayout::NCHW, 1.0 /*value*/, false /*is_leaf*/);
+      egr_utils_api::CreateTensorWithValue(ddim,
+                                           paddle::platform::CPUPlace(),
+                                           phi::DataType::FLOAT32,
+                                           phi::DataLayout::NCHW,
+                                           1.0 /*value*/,
+                                           false /*is_leaf*/);
 
   paddle::experimental::Tensor leaf_tensor;
   {
@@ -92,17 +94,24 @@ TEST(Backward, SingleNodeCustomGrad) {
   paddle::framework::DDim ddim = phi::make_ddim({4, 16, 16, 32});
 
   // Create Target Tensor
-  paddle::experimental::Tensor tensor = egr_utils_api::CreateTensorWithValue(
-      ddim, paddle::platform::CPUPlace(), phi::DataType::FLOAT32,
-      phi::DataLayout::NCHW, 1.0 /*value*/, false /*is_leaf*/);
+  paddle::experimental::Tensor tensor =
+      egr_utils_api::CreateTensorWithValue(ddim,
+                                           paddle::platform::CPUPlace(),
+                                           phi::DataType::FLOAT32,
+                                           phi::DataLayout::NCHW,
+                                           1.0 /*value*/,
+                                           false /*is_leaf*/);
   target_tensors.emplace_back(std::move(tensor));
 
   std::vector<paddle::experimental::Tensor> grad_tensors;
   // Create Grad Tensor
   paddle::experimental::Tensor grad_tensor =
-      egr_utils_api::CreateTensorWithValue(
-          ddim, paddle::platform::CPUPlace(), phi::DataType::FLOAT32,
-          phi::DataLayout::NCHW, 10.0 /*value*/, false /*is_leaf*/);
+      egr_utils_api::CreateTensorWithValue(ddim,
+                                           paddle::platform::CPUPlace(),
+                                           phi::DataType::FLOAT32,
+                                           phi::DataLayout::NCHW,
+                                           10.0 /*value*/,
+                                           false /*is_leaf*/);
   grad_tensors.emplace_back(std::move(grad_tensor));
 
   paddle::experimental::Tensor leaf_tensor;
@@ -157,9 +166,13 @@ TEST(Backward, LinearNodes) {
   paddle::framework::DDim ddim = phi::make_ddim({4, 16, 16, 32});
 
   // Create Target Tensor
-  paddle::experimental::Tensor tensor = egr_utils_api::CreateTensorWithValue(
-      ddim, paddle::platform::CPUPlace(), phi::DataType::FLOAT32,
-      phi::DataLayout::NCHW, 1.0 /*value*/, false /*is_leaf*/);
+  paddle::experimental::Tensor tensor =
+      egr_utils_api::CreateTensorWithValue(ddim,
+                                           paddle::platform::CPUPlace(),
+                                           phi::DataType::FLOAT32,
+                                           phi::DataLayout::NCHW,
+                                           1.0 /*value*/,
+                                           false /*is_leaf*/);
   target_tensors.emplace_back(std::move(tensor));
 
   paddle::experimental::Tensor leaf_tensor;
@@ -229,25 +242,39 @@ TEST(Backward, WithAccumulation) {
 
   // Create Target Tensor
   std::vector<paddle::experimental::Tensor> target_tensors;
-  paddle::experimental::Tensor tensor0 = egr_utils_api::CreateTensorWithValue(
-      ddim, paddle::platform::CPUPlace(), phi::DataType::FLOAT32,
-      phi::DataLayout::NCHW, 1.0 /*value*/, false /*is_leaf*/);
-  paddle::experimental::Tensor tensor1 = egr_utils_api::CreateTensorWithValue(
-      ddim, paddle::platform::CPUPlace(), phi::DataType::FLOAT32,
-      phi::DataLayout::NCHW, 1.0 /*value*/, false /*is_leaf*/);
+  paddle::experimental::Tensor tensor0 =
+      egr_utils_api::CreateTensorWithValue(ddim,
+                                           paddle::platform::CPUPlace(),
+                                           phi::DataType::FLOAT32,
+                                           phi::DataLayout::NCHW,
+                                           1.0 /*value*/,
+                                           false /*is_leaf*/);
+  paddle::experimental::Tensor tensor1 =
+      egr_utils_api::CreateTensorWithValue(ddim,
+                                           paddle::platform::CPUPlace(),
+                                           phi::DataType::FLOAT32,
+                                           phi::DataLayout::NCHW,
+                                           1.0 /*value*/,
+                                           false /*is_leaf*/);
   target_tensors.emplace_back(std::move(tensor0));
   target_tensors.emplace_back(std::move(tensor1));
 
   // Create Grad Tensor
   std::vector<paddle::experimental::Tensor> grad_tensors;
   paddle::experimental::Tensor grad_tensor0 =
-      egr_utils_api::CreateTensorWithValue(
-          ddim, paddle::platform::CPUPlace(), phi::DataType::FLOAT32,
-          phi::DataLayout::NCHW, 5.0 /*value*/, false /*is_leaf*/);
+      egr_utils_api::CreateTensorWithValue(ddim,
+                                           paddle::platform::CPUPlace(),
+                                           phi::DataType::FLOAT32,
+                                           phi::DataLayout::NCHW,
+                                           5.0 /*value*/,
+                                           false /*is_leaf*/);
   paddle::experimental::Tensor grad_tensor1 =
-      egr_utils_api::CreateTensorWithValue(
-          ddim, paddle::platform::CPUPlace(), phi::DataType::FLOAT32,
-          phi::DataLayout::NCHW, 10.0 /*value*/, false /*is_leaf*/);
+      egr_utils_api::CreateTensorWithValue(ddim,
+                                           paddle::platform::CPUPlace(),
+                                           phi::DataType::FLOAT32,
+                                           phi::DataLayout::NCHW,
+                                           10.0 /*value*/,
+                                           false /*is_leaf*/);
   grad_tensors.emplace_back(std::move(grad_tensor0));
   grad_tensors.emplace_back(std::move(grad_tensor1));
 

paddle/fluid/eager/tests/task_tests/grad_test.cc

@@ -29,7 +29,6 @@
 #include "paddle/phi/core/tensor_meta.h"
 
 PD_DECLARE_KERNEL(full, CPU, ALL_LAYOUT);
-PD_DECLARE_KERNEL(copy, CPU, ALL_LAYOUT);
 PD_DECLARE_KERNEL(add, CPU, ALL_LAYOUT);
 
 namespace egr {
@@ -43,15 +42,21 @@ TEST(Grad, SingleNodeEmptyGrad) {
 
   // Create Target Tensor (output)
   paddle::experimental::Tensor output_tensor =
-      egr_utils_api::CreateTensorWithValue(
-          ddim, paddle::platform::CPUPlace(), phi::DataType::FLOAT32,
-          phi::DataLayout::NCHW, 1.0 /*value*/, false /*is_leaf*/);
+      egr_utils_api::CreateTensorWithValue(ddim,
+                                           paddle::platform::CPUPlace(),
+                                           phi::DataType::FLOAT32,
+                                           phi::DataLayout::NCHW,
+                                           1.0 /*value*/,
+                                           false /*is_leaf*/);
 
   // Create input tensor
   const paddle::experimental::Tensor leaf_tensor =
-      egr_utils_api::CreateTensorWithValue(
-          ddim, paddle::platform::CPUPlace(), phi::DataType::FLOAT32,
-          phi::DataLayout::NCHW, 1.0 /*value*/, true /*is_leaf*/);
+      egr_utils_api::CreateTensorWithValue(ddim,
+                                           paddle::platform::CPUPlace(),
+                                           phi::DataType::FLOAT32,
+                                           phi::DataLayout::NCHW,
+                                           1.0 /*value*/,
+                                           true /*is_leaf*/);
 
   {
     // Create Scale Node
@@ -103,23 +108,33 @@ TEST(Grad, SingleNodeCustomGrad) {
   paddle::framework::DDim ddim = phi::make_ddim({4, 16, 16, 32});
 
   // Create Target Tensor
-  paddle::experimental::Tensor tensor = egr_utils_api::CreateTensorWithValue(
-      ddim, paddle::platform::CPUPlace(), phi::DataType::FLOAT32,
-      phi::DataLayout::NCHW, 1.0 /*value*/, false /*is_leaf*/);
+  paddle::experimental::Tensor tensor =
+      egr_utils_api::CreateTensorWithValue(ddim,
+                                           paddle::platform::CPUPlace(),
+                                           phi::DataType::FLOAT32,
+                                           phi::DataLayout::NCHW,
+                                           1.0 /*value*/,
+                                           false /*is_leaf*/);
   target_tensors.emplace_back(std::move(tensor));
 
   std::vector<paddle::experimental::Tensor> grad_tensors;
   // Create Grad Tensor
   paddle::experimental::Tensor grad_tensor =
-      egr_utils_api::CreateTensorWithValue(
-          ddim, paddle::platform::CPUPlace(), phi::DataType::FLOAT32,
-          phi::DataLayout::NCHW, 10.0 /*value*/, false /*is_leaf*/);
+      egr_utils_api::CreateTensorWithValue(ddim,
+                                           paddle::platform::CPUPlace(),
+                                           phi::DataType::FLOAT32,
+                                           phi::DataLayout::NCHW,
+                                           10.0 /*value*/,
+                                           false /*is_leaf*/);
   grad_tensors.emplace_back(std::move(grad_tensor));
 
   paddle::experimental::Tensor leaf_tensor =
-      egr_utils_api::CreateTensorWithValue(
-          ddim, paddle::platform::CPUPlace(), phi::DataType::FLOAT32,
-          phi::DataLayout::NCHW, 1.0 /*value*/, true /*is_leaf*/);
+      egr_utils_api::CreateTensorWithValue(ddim,
+                                           paddle::platform::CPUPlace(),
+                                           phi::DataType::FLOAT32,
+                                           phi::DataLayout::NCHW,
+                                           1.0 /*value*/,
+                                           true /*is_leaf*/);
 
   {
     // Create Scale Node
@@ -172,15 +187,22 @@ TEST(Grad, LinearNodes) {
   paddle::framework::DDim ddim = phi::make_ddim({4, 16, 16, 32});
 
   // Create Target Tensor
-  paddle::experimental::Tensor tensor = egr_utils_api::CreateTensorWithValue(
-      ddim, paddle::platform::CPUPlace(), phi::DataType::FLOAT32,
-      phi::DataLayout::NCHW, 1.0 /*value*/, false /*is_leaf*/);
+  paddle::experimental::Tensor tensor =
+      egr_utils_api::CreateTensorWithValue(ddim,
+                                           paddle::platform::CPUPlace(),
+                                           phi::DataType::FLOAT32,
+                                           phi::DataLayout::NCHW,
+                                           1.0 /*value*/,
+                                           false /*is_leaf*/);
   target_tensors.emplace_back(std::move(tensor));
 
   paddle::experimental::Tensor leaf_tensor =
-      egr_utils_api::CreateTensorWithValue(
-          ddim, paddle::platform::CPUPlace(), phi::DataType::FLOAT32,
-          phi::DataLayout::NCHW, 1.0 /*value*/, true /*is_leaf*/);
+      egr_utils_api::CreateTensorWithValue(ddim,
+                                           paddle::platform::CPUPlace(),
+                                           phi::DataType::FLOAT32,
+                                           phi::DataLayout::NCHW,
+                                           1.0 /*value*/,
+                                           true /*is_leaf*/);
   {
     // Create Node0
     auto node0_ptr = std::make_shared<GradNodeScale>(1, 1);
@@ -247,25 +269,39 @@ TEST(Grad, WithAccumulation) {
 
   // Create Target Tensor
   std::vector<paddle::experimental::Tensor> target_tensors;
-  paddle::experimental::Tensor tensor0 = egr_utils_api::CreateTensorWithValue(
-      ddim, paddle::platform::CPUPlace(), phi::DataType::FLOAT32,
-      phi::DataLayout::NCHW, 1.0 /*value*/, false /*is_leaf*/);
-  paddle::experimental::Tensor tensor1 = egr_utils_api::CreateTensorWithValue(
-      ddim, paddle::platform::CPUPlace(), phi::DataType::FLOAT32,
-      phi::DataLayout::NCHW, 1.0 /*value*/, false /*is_leaf*/);
+  paddle::experimental::Tensor tensor0 =
+      egr_utils_api::CreateTensorWithValue(ddim,
+                                           paddle::platform::CPUPlace(),
+                                           phi::DataType::FLOAT32,
+                                           phi::DataLayout::NCHW,
+                                           1.0 /*value*/,
+                                           false /*is_leaf*/);
+  paddle::experimental::Tensor tensor1 =
+      egr_utils_api::CreateTensorWithValue(ddim,
+                                           paddle::platform::CPUPlace(),
+                                           phi::DataType::FLOAT32,
+                                           phi::DataLayout::NCHW,
+                                           1.0 /*value*/,
+                                           false /*is_leaf*/);
   target_tensors.emplace_back(std::move(tensor0));
   target_tensors.emplace_back(std::move(tensor1));
 
   // Create Grad Tensor
   std::vector<paddle::experimental::Tensor> grad_tensors;
   paddle::experimental::Tensor grad_tensor0 =
-      egr_utils_api::CreateTensorWithValue(
-          ddim, paddle::platform::CPUPlace(), phi::DataType::FLOAT32,
-          phi::DataLayout::NCHW, 5.0 /*value*/, false /*is_leaf*/);
+      egr_utils_api::CreateTensorWithValue(ddim,
+                                           paddle::platform::CPUPlace(),
+                                           phi::DataType::FLOAT32,
+                                           phi::DataLayout::NCHW,
+                                           5.0 /*value*/,
+                                           false /*is_leaf*/);
   paddle::experimental::Tensor grad_tensor1 =
-      egr_utils_api::CreateTensorWithValue(
-          ddim, paddle::platform::CPUPlace(), phi::DataType::FLOAT32,
-          phi::DataLayout::NCHW, 10.0 /*value*/, false /*is_leaf*/);
+      egr_utils_api::CreateTensorWithValue(ddim,
+                                           paddle::platform::CPUPlace(),
+                                           phi::DataType::FLOAT32,
+                                           phi::DataLayout::NCHW,
+                                           10.0 /*value*/,
+                                           false /*is_leaf*/);
   grad_tensors.emplace_back(std::move(grad_tensor0));
   grad_tensors.emplace_back(std::move(grad_tensor1));
 

paddle/fluid/jit/layer_test.cc

@@ -21,7 +21,7 @@
 #include "paddle/fluid/imperative/tracer.h"
 #include "paddle/phi/core/dense_tensor.h"
 #include "paddle/phi/core/kernel_registry.h"
-#include "paddle/phi/kernels/copy_kernel.h"
+#include "paddle/phi/core/tensor_utils.h"
 #include "paddle/phi/kernels/funcs/math_function.h"
 
 #include "paddle/fluid/jit/layer.h"

paddle/phi/api/lib/data_transform.cc

@@ -19,8 +19,8 @@ limitations under the License. */
 #include "paddle/phi/api/lib/utils/allocator.h"
 #include "paddle/phi/backends/all_context.h"
 #include "paddle/phi/core/kernel_registry.h"
+#include "paddle/phi/core/tensor_utils.h"
 #include "paddle/phi/kernels/cast_kernel.h"
-#include "paddle/phi/kernels/copy_kernel.h"
 #include "paddle/phi/kernels/transfer_layout_kernel.h"
 
 #include "paddle/fluid/framework/tensor_util.h"

paddle/phi/api/lib/tensor_copy.cc

@@ -20,6 +20,7 @@ limitations under the License. */
 #include "paddle/phi/core/compat/convert_utils.h"
 #include "paddle/phi/core/kernel_registry.h"
 #include "paddle/phi/core/meta_tensor.h"
+#include "paddle/phi/core/tensor_utils.h"
 #include "paddle/phi/infermeta/unary.h"
 
 namespace paddle {
@@ -31,10 +32,7 @@ void copy(const Tensor& src, const Place& place, bool blocking, Tensor* dst) {
       kernel_key_set.backend_set | BackendSet(phi::TransToPhiBackend(place));
   auto kernel_key = kernel_key_set.GetHighestPriorityKernelKey();
 
-  VLOG(6) << "copy API kernel key: " << kernel_key;
-  auto kernel = phi::KernelFactory::Instance().SelectKernelOrThrowError(
-      "copy", kernel_key);
-  VLOG(6) << "copy API kernel: " << kernel;
+  VLOG(6) << "start copy. ";
 
   auto target_place = phi::TransToPhiPlace(kernel_key.backend());
   auto& pool = paddle::experimental::DeviceContextPool::Instance();
@@ -47,14 +45,9 @@ void copy(const Tensor& src, const Place& place, bool blocking, Tensor* dst) {
   phi::MetaTensor meta_out(kernel_out);
   phi::UnchangedInferMeta(*dense_x, &meta_out);
 
-  using kernel_signature = void (*)(const platform::DeviceContext&,
-                                    const phi::DenseTensor&,
-                                    phi::Place,
-                                    bool,
-                                    phi::DenseTensor*);
+  phi::Copy(*dev_ctx, *dense_x, place, blocking, kernel_out);
 
-  auto* kernel_fn = kernel.GetVariadicKernelFn<kernel_signature>();
-  (*kernel_fn)(*dev_ctx, *dense_x, place, blocking, kernel_out);
+  VLOG(6) << "copy finished. ";
 }
 
 }  // namespace experimental

paddle/phi/api/lib/tensor_method.cc

@@ -19,9 +19,11 @@ limitations under the License. */
 #include "paddle/phi/core/compat/convert_utils.h"
 #include "paddle/phi/core/tensor_base.h"
 
+#include "paddle/phi/api/include/context_pool.h"
 #include "paddle/phi/api/include/sparse_api.h"
 #include "paddle/phi/api/lib/api_gen_utils.h"
 #include "paddle/phi/api/lib/kernel_dispatch.h"
+#include "paddle/phi/core/tensor_utils.h"
 #include "paddle/phi/infermeta/unary.h"
 // clang-format off
 
@@ -113,9 +115,15 @@ void Tensor::copy_(const Tensor &src,
     // Deep Copy AutoGrad info from src to self.
     *autograd_meta_ = *(src.autograd_meta_);
   }
-
+  kernel_key_set.backend_set =
+      kernel_key_set.backend_set |
+      BackendSet(phi::TransToPhiBackend(target_place));
   auto kernel_key = kernel_key_set.GetHighestPriorityKernelKey();
-  auto *dev_ctx = GetDeviceContextByBackend(kernel_key.backend());
+  auto place = phi::TransToPhiPlace(kernel_key.backend());
+  auto& pool = paddle::experimental::DeviceContextPool::Instance();
+  auto* dev_ctx = pool.GetMutable(
+      place.GetType() == target_place.GetType() ? target_place : place);
+
   Backend kernel_backend = Backend::UNDEFINED;
   DataLayout kernel_layout = DataLayout::UNDEFINED;
   DataType kernel_data_type = DataType::UNDEFINED;
@@ -135,49 +143,29 @@ void Tensor::copy_(const Tensor &src,
   }
 
   if (kernel_type == KernelType::DENSE_TENSOR_KENREL) {
-    auto kernel = phi::KernelFactory::Instance().SelectKernelOrThrowError(
-        "copy", {kernel_backend, kernel_layout, kernel_data_type});
-    VLOG(6) << "copy API kernel key: " << kernel_key;
-    VLOG(6) << "copy API kernel: " << kernel;
-    using kernel_signature = void (*)(const platform::DeviceContext &,
-                                      const phi::DenseTensor &,
-                                      phi::Place,
-                                      bool,
-                                      phi::DenseTensor *);
     SetKernelOutput(kernel_backend, this);
     phi::MetaTensor meta_out(impl_.get());
     phi::UnchangedInferMeta(
         MakeMetaTensor(
             *(std::static_pointer_cast<phi::DenseTensor>(src.impl_))),
         &meta_out);
-    auto *kernel_fn = kernel.GetVariadicKernelFn<kernel_signature>();
-    (*kernel_fn)(*dev_ctx,
-                 (*(std::static_pointer_cast<phi::DenseTensor>(src.impl_))),
-                 target_place,
-                 blocking,
-                 static_cast<phi::DenseTensor *>(impl_.get()));
+    phi::Copy(*dev_ctx,
+              (*(std::static_pointer_cast<phi::DenseTensor>(src.impl_))),
+              target_place,
+              blocking,
+              static_cast<phi::DenseTensor *>(impl_.get()));
   } else if (kernel_type == KernelType::SELECTED_ROWS_KENREL) {
-    auto kernel = phi::KernelFactory::Instance().SelectKernelOrThrowError(
-        "copy_sr", {kernel_backend, kernel_layout, kernel_data_type});
-    VLOG(6) << "copy API kernel key: " << kernel_key;
-    VLOG(6) << "copy API kernel: " << kernel;
-    using kernel_signature = void (*)(const platform::DeviceContext &,
-                                      const phi::SelectedRows &,
-                                      phi::Place,
-                                      bool,
-                                      phi::SelectedRows *);
     SetSelectedRowsKernelOutput(kernel_backend, this);
     phi::MetaTensor meta_out(impl_.get());
     phi::UnchangedInferMeta(
         MakeMetaTensor(
             *(std::static_pointer_cast<phi::SelectedRows>(src.impl_))),
         &meta_out);
-    auto *kernel_fn = kernel.GetVariadicKernelFn<kernel_signature>();
-    (*kernel_fn)(*dev_ctx,
-                 (*(std::static_pointer_cast<phi::SelectedRows>(src.impl_))),
-                 target_place,
-                 blocking,
-                 static_cast<phi::SelectedRows *>(impl_.get()));
+    phi::Copy(*dev_ctx,
+              (*(std::static_pointer_cast<phi::SelectedRows>(src.impl_))),
+              target_place,
+              blocking,
+              static_cast<phi::SelectedRows *>(impl_.get()));
   } else if (kernel_type == KernelType::SPARSE_COO_KERNEL) {
     auto kernel = phi::KernelFactory::Instance().SelectKernelOrThrowError(
         "copy_sparse_coo", {kernel_backend, kernel_layout, kernel_data_type});

paddle/phi/core/CMakeLists.txt

@@ -81,3 +81,43 @@ if(WITH_MKLDNN)
   add_dependencies(dense_tensor mkldnn)
   add_dependencies(tensor_base mkldnn)
 endif()
+
+if(WITH_GPU)
+  nv_library(
+    phi_tensor_utils
+    SRCS tensor_utils.cc
+    DEPS cpu_context
+         gpu_context
+         dense_tensor
+         selected_rows
+         malloc
+         memcpy
+         device_context)
+elseif(WITH_ROCM)
+  hip_library(
+    phi_tensor_utils
+    SRCS tensor_utils.cc
+    DEPS cpu_context
+         gpu_context
+         dense_tensor
+         selected_rows
+         malloc
+         memcpy
+         device_context)
+elseif(WITH_XPU_KP)
+  xpu_library(
+    phi_tensor_utils
+    SRCS tensor_utils.cc
+    DEPS cpu_context
+         xpu_context
+         dense_tensor
+         selected_rows
+         malloc
+         memcpy
+         device_context)
+else()
+  cc_library(
+    phi_tensor_utils
+    SRCS tensor_utils.cc
+    DEPS cpu_context dense_tensor selected_rows malloc memcpy device_context)
+endif()

paddle/phi/kernels/gpu/copy_kernel.cu → paddle/phi/core/tensor_utils.cc

@@ -12,7 +12,7 @@ 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/phi/kernels/copy_kernel.h"
+#include "paddle/phi/core/tensor_utils.h"
 
 #include "paddle/phi/backends/gpu/gpu_context.h"
 #include "paddle/phi/common/data_type.h"
@@ -43,30 +43,49 @@ void Copy(const Context& dev_ctx,
   void* dst_ptr = nullptr;
   if (paddle::platform::is_cpu_place(dst_place)) {
     dst_ptr = dev_ctx.HostAlloc(dst, src.dtype());
-  } else if (paddle::platform::is_cuda_pinned_place(dst_place)) {
-    // now we only can use mutable_data to Alloc pinned memory here,
-    // dev_ctx can not alloc pinned memory now
-    dst_ptr = dst->mutable_data(dst_place, src.dtype());
-  } else {
+#if defined(PADDLE_WITH_CUDA) || defined(PADDLE_WITH_HIP)
+  } else if (paddle::platform::is_gpu_place(dst_place) ||
+             paddle::platform::is_cuda_pinned_place(dst_place)) {
     dst_ptr = dev_ctx.Alloc(
         dst, src.dtype(), 0, paddle::platform::is_cuda_pinned_place(dst_place));
+#endif
+
+#ifdef PADDLE_WITH_XPU
+  } else if (paddle::platform::is_xpu_place(dst_place)) {
+    dst_ptr = dev_ctx.Alloc(dst, src.dtype());
+#endif
+  }
+
+  auto size = src.numel() * paddle::experimental::SizeOf(src.dtype());
+  if (UNLIKELY(size) == 0) {
+    return;
   }
 
+  PADDLE_ENFORCE_EQ(
+      dst->place(),
+      dst_place,
+      phi::errors::Unavailable(
+          "The Dst Tensor's place and dst_place do not match, Tensor's place "
+          "place is %s, dst_place is %s.",
+          dst->place(),
+          dst_place));
+
   if (src_ptr == dst_ptr && src_place == dst_place) {
     VLOG(3) << "Skip copy the same data async from " << src_place << " to "
             << dst_place;
     return;
   }
   VLOG(4) << "src:" << src_ptr << ", dst:" << dst_ptr;
-
   CHECK(dst->layout() == src.layout());
 
-  auto size = src.numel() * paddle::experimental::SizeOf(src.dtype());
-
-  if ((paddle::platform::is_cpu_place(src_place) ||
-       paddle::platform::is_cuda_pinned_place(src_place)) &&  // NOLINT
-      (paddle::platform::is_cpu_place(dst_place) ||
-       paddle::platform::is_cuda_pinned_place(dst_place))) {
+  if (paddle::platform::is_cpu_place(src_place) &&
+      paddle::platform::is_cpu_place(dst_place)) {
+    paddle::memory::Copy(src_place, dst_ptr, src_place, src_ptr, size);
+#if defined(PADDLE_WITH_CUDA) || defined(PADDLE_WITH_HIP)
+  } else if ((paddle::platform::is_cpu_place(src_place) ||
+              paddle::platform::is_cuda_pinned_place(src_place)) &&  // NOLINT
+             (paddle::platform::is_cpu_place(dst_place) ||
+              paddle::platform::is_cuda_pinned_place(dst_place))) {
     paddle::memory::Copy(dst_place, dst_ptr, src_place, src_ptr, size, nullptr);
   } else if (paddle::platform::is_gpu_place(src_place) &&  // NOLINT
              paddle::platform::is_cpu_place(dst_place)) {
@@ -176,13 +195,87 @@ void Copy(const Context& dev_ctx,
                  : reinterpret_cast<const phi::GPUContext&>(dev_ctx).stream();
     paddle::memory::Copy(
         dst_cuda_pinned_place, dst_ptr, src_gpu_place, src_ptr, size, stream);
+#endif
+  }
+#ifdef PADDLE_WITH_XPU
+  else if (paddle::platform::is_xpu_place(src_place) &&  // NOLINT
+           paddle::platform::is_cpu_place(dst_place)) {
+    paddle::memory::Copy(dst_place, dst_ptr, src_place, src_ptr, size);
+  } else if (paddle::platform::is_cpu_place(src_place) &&
+             paddle::platform::is_xpu_place(dst_place)) {
+    paddle::memory::Copy(dst_place, dst_ptr, src_place, src_ptr, size);
+  } else if (paddle::platform::is_xpu_place(src_place) &&
+             paddle::platform::is_xpu_place(dst_place)) {
+    if (src_ptr == dst_ptr) {
+      VLOG(3) << "Skip copy the same data async from " << src_place << " to "
+              << dst_place;
+      return;
+    }
+    paddle::memory::Copy(dst_place, dst_ptr, src_place, src_ptr, size);
   } else {
-    PADDLE_THROW(phi::errors::InvalidArgument(
-        "Place type error. Please check the place of src and dst Tensor."));
+    PADDLE_THROW(phi::errors::Unimplemented(
+        "Copy from %s to %s is not supported.", src_place, dst_place));
   }
+#endif
 }
 
-}  // namespace phi
+template <typename Context>
+void Copy(const Context& dev_ctx,
+          const SelectedRows& src,
+          Place dst_place,
+          bool blocking,
+          SelectedRows* dst) {
+  if (src.value().Holder() != dst->value().Holder() ||
+      src.value().data() != dst->value().data()) {
+    dst->set_rows(src.rows());
+    dst->set_height(src.height());
+  }
+  Copy<Context>(
+      dev_ctx, src.value(), dst_place, blocking, dst->mutable_value());
+}
+
+template void Copy(const CPUContext& dev_ctx,
+                   const DenseTensor& src,
+                   Place dst_place,
+                   bool blocking,
+                   DenseTensor* dst);
+
+template void Copy(const DeviceContext& dev_ctx,
+                   const DenseTensor& src,
+                   Place dst_place,
+                   bool blocking,
+                   DenseTensor* dst);
+
+template void Copy(const CPUContext& dev_ctx,
+                   const SelectedRows& src,
+                   Place dst_place,
+                   bool blocking,
+                   SelectedRows* dst);
+template void Copy(const DeviceContext& dev_ctx,
+                   const SelectedRows& src,
+                   Place dst_place,
+                   bool blocking,
+                   SelectedRows* dst);
 
-PD_REGISTER_GENERAL_KERNEL(
-    copy, GPU, ALL_LAYOUT, phi::Copy<phi::GPUContext>, ALL_DTYPE) {}
+#if defined(PADDLE_WITH_CUDA) || defined(PADDLE_WITH_HIP)
+template void Copy(const GPUContext& dev_ctx,
+                   const DenseTensor& src,
+                   Place dst_place,
+                   bool blocking,
+                   DenseTensor* dst);
+template void Copy(const GPUContext& dev_ctx,
+                   const SelectedRows& src,
+                   Place dst_place,
+                   bool blocking,
+                   SelectedRows* dst);
+#endif
+
+#ifdef PADDLE_WITH_XPU
+template void Copy(const XPUContext& dev_ctx,
+                   const DenseTensor& src,
+                   Place dst_place,
+                   bool blocking,
+                   DenseTensor* dst);
+#endif
+
+}  // namespace phi

paddle/phi/core/tensor_utils.h

@@ -15,6 +15,7 @@ limitations under the License. */
 #pragma once
 
 #include "paddle/phi/core/dense_tensor.h"
+#include "paddle/phi/core/selected_rows.h"
 #include "paddle/phi/core/tensor_meta.h"
 
 namespace phi {
@@ -70,4 +71,18 @@ class DenseTensorUtils {
   }
 };
 
+template <typename Context>
+void Copy(const Context& dev_ctx,
+          const DenseTensor& src,
+          Place dst_place,
+          bool blocking,
+          DenseTensor* dst);
+
+template <typename Context>
+void Copy(const Context& dev_ctx,
+          const SelectedRows& src,
+          Place dst_place,
+          bool blocking,
+          SelectedRows* dst);
+
 }  // namespace phi

paddle/phi/kernels/CMakeLists.txt

@@ -29,7 +29,8 @@ set(COMMON_KERNEL_DEPS
     arg_map_context
     convert_utils
     lod_utils
-    custom_kernel)
+    custom_kernel
+    phi_tensor_utils)
 set(COMMON_KERNEL_DEPS
     ${COMMON_KERNEL_DEPS}
     eigen_function

paddle/phi/kernels/assign_kernel.cc

@@ -16,7 +16,7 @@
 
 #include "paddle/fluid/framework/tensor_util.h"
 #include "paddle/phi/core/kernel_registry.h"
-#include "paddle/phi/kernels/copy_kernel.h"
+#include "paddle/phi/core/tensor_utils.h"
 #include "paddle/utils/optional.h"
 
 namespace phi {

paddle/phi/kernels/autotune/auto_tune_test.cu

@@ -20,8 +20,8 @@
 #include "paddle/phi/backends/gpu/gpu_launch_config.h"
 #include "paddle/phi/core/dense_tensor.h"
 #include "paddle/phi/core/tensor_meta.h"
+#include "paddle/phi/core/tensor_utils.h"
 #include "paddle/phi/kernels/autotune/auto_tune_base.h"
-#include "paddle/phi/kernels/copy_kernel.h"
 #include "paddle/phi/kernels/funcs/aligned_vector.h"
 
 namespace tune = phi::autotune;

paddle/phi/kernels/copy_kernel.h

-/* Copyright (c) 2021 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/phi/core/dense_tensor.h"
-
-namespace phi {
-
-template <typename Context>
-void Copy(const Context& dev_ctx,
-          const DenseTensor& src,
-          Place dst_place,
-          bool blocking,
-          DenseTensor* dst);
-}  // namespace phi

paddle/phi/kernels/cpu/adam_kernel.cc

@@ -20,7 +20,7 @@
 #include "paddle/fluid/operators/jit/kernels.h"
 #include "paddle/phi/backends/cpu/cpu_context.h"
 #include "paddle/phi/core/kernel_registry.h"
-#include "paddle/phi/kernels/copy_kernel.h"
+#include "paddle/phi/core/tensor_utils.h"
 #include "paddle/phi/kernels/funcs/adam_functors.h"
 
 DECLARE_int32(inner_op_parallelism);

paddle/phi/kernels/cpu/copy_kernel.cc

-/* Copyright (c) 2021 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/phi/kernels/copy_kernel.h"
-
-#include "paddle/phi/backends/cpu/cpu_context.h"
-#include "paddle/phi/common/data_type.h"
-#include "paddle/phi/core/compat/convert_utils.h"
-#include "paddle/phi/core/kernel_registry.h"
-
-// See Note [ Why still include the fluid headers? ]
-#include "paddle/fluid/memory/memcpy.h"
-
-namespace phi {
-
-// NOTE(chenweihang): blocking is useless in cpu kernel
-template <typename Context>
-void Copy(const Context& dev_ctx,
-          const DenseTensor& src,
-          Place dst_place,
-          bool blocking,
-          DenseTensor* dst) {
-  auto* src_ptr = src.data();
-  const auto& src_place = src.place();
-
-  VLOG(3) << "TensorCopy " << src.dims() << " from " << src.place() << " to "
-          << src_place;
-
-  dst->Resize(src.dims());
-  auto* dst_ptr = dev_ctx.HostAlloc(dst, src.dtype());
-
-  if (src_ptr == dst_ptr) {
-    VLOG(3) << "Skip copy the same data async from " << src_place << " to "
-            << src_place;
-    return;
-  }
-  VLOG(4) << "src:" << src_ptr << ", dst:" << dst_ptr;
-  CHECK(dst->layout() == src.layout());
-
-  auto size = src.numel() * paddle::experimental::SizeOf(src.dtype());
-
-  if (paddle::platform::is_cpu_place(src_place)) {
-    paddle::memory::Copy(src_place, dst_ptr, src_place, src_ptr, size);
-  }
-}
-
-}  // namespace phi
-
-PD_REGISTER_GENERAL_KERNEL(
-    copy, CPU, ALL_LAYOUT, phi::Copy<phi::CPUContext>, ALL_DTYPE) {}

paddle/phi/kernels/cpu/cross_entropy_grad_kernel.cc

@@ -16,8 +16,8 @@ limitations under the License. */
 
 #include "paddle/phi/backends/cpu/cpu_context.h"
 #include "paddle/phi/core/kernel_registry.h"
+#include "paddle/phi/core/tensor_utils.h"
 #include "paddle/phi/core/visit_type.h"
-#include "paddle/phi/kernels/copy_kernel.h"
 #include "paddle/phi/kernels/funcs/axis_utils.h"
 #include "paddle/phi/kernels/funcs/eigen/common.h"
 

paddle/phi/kernels/cpu/cross_entropy_kernel.cc

@@ -17,7 +17,7 @@ limitations under the License. */
 #include "paddle/fluid/operators/math/cross_entropy.h"
 #include "paddle/phi/backends/cpu/cpu_context.h"
 #include "paddle/phi/core/kernel_registry.h"
-#include "paddle/phi/kernels/copy_kernel.h"
+#include "paddle/phi/core/tensor_utils.h"
 #include "paddle/phi/kernels/funcs/axis_utils.h"
 #include "paddle/phi/kernels/funcs/math_function.h"
 #include "paddle/phi/kernels/softmax_kernel.h"

paddle/phi/kernels/cpu/elementwise_divide_grad_kernel.cc

@@ -16,7 +16,7 @@
 
 #include "paddle/phi/backends/cpu/cpu_context.h"
 #include "paddle/phi/core/kernel_registry.h"
-#include "paddle/phi/kernels/copy_kernel.h"
+#include "paddle/phi/core/tensor_utils.h"
 #include "paddle/phi/kernels/cpu/elementwise_grad.h"
 #include "paddle/phi/kernels/funcs/elementwise_functor.h"
 #include "paddle/phi/kernels/impl/elementwise_grad_kernel_impl.h"

paddle/phi/kernels/cpu/elementwise_grad_kernel.cc

@@ -16,7 +16,7 @@
 
 #include "paddle/phi/backends/cpu/cpu_context.h"
 #include "paddle/phi/core/kernel_registry.h"
-#include "paddle/phi/kernels/copy_kernel.h"
+#include "paddle/phi/core/tensor_utils.h"
 #include "paddle/phi/kernels/cpu/elementwise_grad.h"
 #include "paddle/phi/kernels/funcs/elementwise_functor.h"
 #include "paddle/phi/kernels/impl/elementwise_grad_kernel_impl.h"

paddle/phi/kernels/cpu/elementwise_multiply_grad_kernel.cc

@@ -16,7 +16,7 @@
 
 #include "paddle/phi/backends/cpu/cpu_context.h"
 #include "paddle/phi/core/kernel_registry.h"
-#include "paddle/phi/kernels/copy_kernel.h"
+#include "paddle/phi/core/tensor_utils.h"
 #include "paddle/phi/kernels/cpu/elementwise_grad.h"
 #include "paddle/phi/kernels/funcs/elementwise_functor.h"
 #include "paddle/phi/kernels/impl/elementwise_grad_kernel_impl.h"

paddle/phi/kernels/cpu/index_select_impl.h

@@ -15,7 +15,7 @@
 #pragma once
 
 #include "paddle/phi/core/dense_tensor.h"
-#include "paddle/phi/kernels/copy_kernel.h"
+#include "paddle/phi/core/tensor_utils.h"
 #include "paddle/phi/kernels/funcs/blas/blas.h"
 #include "paddle/phi/kernels/funcs/eigen/common.h"
 #include "paddle/phi/kernels/funcs/math_function.h"

paddle/phi/kernels/cpu/kthvalue_grad_kernel.cc

@@ -16,7 +16,7 @@
 
 #include "paddle/phi/backends/cpu/cpu_context.h"
 #include "paddle/phi/core/kernel_registry.h"
-#include "paddle/phi/kernels/copy_kernel.h"
+#include "paddle/phi/core/tensor_utils.h"
 #include "paddle/phi/kernels/funcs/eigen/common.h"
 #include "paddle/phi/kernels/funcs/math_function.h"
 

paddle/phi/kernels/cpu/layer_norm_grad_kernel.cc

@@ -22,7 +22,7 @@
 #endif
 #include "paddle/phi/backends/cpu/cpu_context.h"
 #include "paddle/phi/core/kernel_registry.h"
-#include "paddle/phi/kernels/copy_kernel.h"
+#include "paddle/phi/core/tensor_utils.h"
 #include "paddle/phi/kernels/funcs/blas/blas.h"
 #include "paddle/phi/kernels/funcs/elementwise_base.h"
 #include "paddle/phi/kernels/funcs/elementwise_functor.h"

paddle/phi/kernels/cpu/mode_grad_kernel.cc

@@ -16,7 +16,7 @@
 
 #include "paddle/phi/backends/cpu/cpu_context.h"
 #include "paddle/phi/core/kernel_registry.h"
-#include "paddle/phi/kernels/copy_kernel.h"
+#include "paddle/phi/core/tensor_utils.h"
 #include "paddle/phi/kernels/funcs/mode.h"
 
 namespace phi {

paddle/phi/kernels/cpu/put_along_axis_grad_kernel.cc

@@ -19,7 +19,7 @@
 #include "paddle/fluid/platform/place.h"
 #include "paddle/phi/backends/cpu/cpu_context.h"
 #include "paddle/phi/core/kernel_registry.h"
-#include "paddle/phi/kernels/copy_kernel.h"
+#include "paddle/phi/core/tensor_utils.h"
 
 namespace phi {
 

paddle/phi/kernels/cpu/put_along_axis_kernel.cc

@@ -19,7 +19,7 @@
 #include "paddle/fluid/platform/place.h"
 #include "paddle/phi/backends/cpu/cpu_context.h"
 #include "paddle/phi/core/kernel_registry.h"
-#include "paddle/phi/kernels/copy_kernel.h"
+#include "paddle/phi/core/tensor_utils.h"
 
 namespace phi {
 

paddle/phi/kernels/cpu/rnn_functor.h

@@ -17,7 +17,7 @@
 #include "paddle/fluid/framework/generator.h"
 #include "paddle/fluid/operators/utils.h"
 #include "paddle/phi/core/dense_tensor.h"
-#include "paddle/phi/kernels/copy_kernel.h"
+#include "paddle/phi/core/tensor_utils.h"
 #include "paddle/phi/kernels/empty_kernel.h"
 #include "paddle/phi/kernels/funcs/eigen/common.h"
 #include "paddle/phi/kernels/funcs/eigen/eigen_function.h"

paddle/phi/kernels/cpu/rnn_grad_kernel.cc

@@ -16,7 +16,7 @@
 
 #include "paddle/phi/backends/cpu/cpu_context.h"
 #include "paddle/phi/core/kernel_registry.h"
-#include "paddle/phi/kernels/copy_kernel.h"
+#include "paddle/phi/core/tensor_utils.h"
 #include "paddle/phi/kernels/cpu/rnn_functor.h"
 #include "paddle/phi/kernels/funcs/activation_functor.h"
 #include "paddle/phi/kernels/funcs/blas/blas.h"

paddle/phi/kernels/cpu/rnn_kernel.cc

@@ -16,7 +16,7 @@
 
 #include "paddle/phi/backends/cpu/cpu_context.h"
 #include "paddle/phi/core/kernel_registry.h"
-#include "paddle/phi/kernels/copy_kernel.h"
+#include "paddle/phi/core/tensor_utils.h"
 #include "paddle/phi/kernels/cpu/rnn_functor.h"
 #include "paddle/phi/kernels/empty_kernel.h"
 #include "paddle/phi/kernels/full_kernel.h"

paddle/phi/kernels/cpu/scatter_grad_kernel.cc

@@ -16,7 +16,7 @@
 
 #include "paddle/phi/backends/cpu/cpu_context.h"
 #include "paddle/phi/core/kernel_registry.h"
-#include "paddle/phi/kernels/copy_kernel.h"
+#include "paddle/phi/core/tensor_utils.h"
 #include "paddle/phi/kernels/funcs/gather.h"
 #include "paddle/phi/kernels/funcs/scatter.h"
 

paddle/phi/kernels/cpu/scatter_kernel.cc

@@ -16,7 +16,7 @@
 
 #include "paddle/phi/backends/cpu/cpu_context.h"
 #include "paddle/phi/core/kernel_registry.h"
-#include "paddle/phi/kernels/copy_kernel.h"
+#include "paddle/phi/core/tensor_utils.h"
 #include "paddle/phi/kernels/funcs/scatter.h"
 
 namespace phi {

paddle/phi/kernels/cpu/scatter_nd_add_grad_kernel.cc

@@ -16,7 +16,7 @@
 
 #include "paddle/phi/backends/cpu/cpu_context.h"
 #include "paddle/phi/core/kernel_registry.h"
-#include "paddle/phi/kernels/copy_kernel.h"
+#include "paddle/phi/core/tensor_utils.h"
 #include "paddle/phi/kernels/funcs/gather.h"
 
 namespace phi {

paddle/phi/kernels/cpu/scatter_nd_add_kernel.cc

@@ -16,7 +16,7 @@
 
 #include "paddle/phi/backends/cpu/cpu_context.h"
 #include "paddle/phi/core/kernel_registry.h"
-#include "paddle/phi/kernels/copy_kernel.h"
+#include "paddle/phi/core/tensor_utils.h"
 #include "paddle/phi/kernels/funcs/scatter.h"
 
 namespace phi {

paddle/phi/kernels/cpu/viterbi_decode_kernel.cc

@@ -21,7 +21,7 @@
 
 #include "paddle/phi/backends/cpu/cpu_context.h"
 #include "paddle/phi/core/kernel_registry.h"
-#include "paddle/phi/kernels/copy_kernel.h"
+#include "paddle/phi/core/tensor_utils.h"
 #include "paddle/phi/kernels/empty_kernel.h"
 #include "paddle/phi/kernels/funcs/compare_functors.h"
 #include "paddle/phi/kernels/funcs/concat_and_split_functor.h"

paddle/phi/kernels/flatten_grad_kernel.cc

@@ -16,7 +16,7 @@
 
 #include "paddle/phi/backends/all_context.h"
 #include "paddle/phi/core/kernel_registry.h"
-#include "paddle/phi/kernels/copy_kernel.h"
+#include "paddle/phi/core/tensor_utils.h"
 
 namespace phi {
 

paddle/phi/kernels/flatten_kernel.cc

@@ -16,8 +16,8 @@
 
 #include "paddle/phi/backends/all_context.h"
 #include "paddle/phi/core/kernel_registry.h"
+#include "paddle/phi/core/tensor_utils.h"
 #include "paddle/phi/infermeta/unary.h"
-#include "paddle/phi/kernels/copy_kernel.h"
 #include "paddle/phi/kernels/funcs/common_shape.h"
 
 namespace phi {

paddle/phi/kernels/funcs/mode.h

@@ -35,7 +35,7 @@
 
 #include "paddle/phi/backends/gpu/gpu_context.h"
 #include "paddle/phi/core/dense_tensor.h"
-#include "paddle/phi/kernels/copy_kernel.h"
+#include "paddle/phi/core/tensor_utils.h"
 #include "paddle/phi/kernels/funcs/eigen/common.h"
 #include "paddle/phi/kernels/funcs/eigen/eigen_function.h"
 #include "paddle/phi/kernels/funcs/math_function.h"

paddle/phi/kernels/funcs/strided_slice.h

@@ -20,7 +20,7 @@
 #include "paddle/phi/core/ddim.h"
 #include "paddle/phi/core/dense_tensor.h"
 #include "paddle/phi/core/enforce.h"
-#include "paddle/phi/kernels/copy_kernel.h"
+#include "paddle/phi/core/tensor_utils.h"
 #include "paddle/phi/kernels/funcs/eigen/common.h"
 #include "paddle/phi/kernels/funcs/eigen/eigen_function.h"
 #include "paddle/phi/kernels/funcs/math_function.h"

paddle/phi/kernels/gpu/adam_kernel.cu

@@ -24,7 +24,7 @@
 #include "paddle/phi/common/amp_type_traits.h"
 #include "paddle/phi/common/float16.h"
 #include "paddle/phi/core/kernel_registry.h"
-#include "paddle/phi/kernels/copy_kernel.h"
+#include "paddle/phi/core/tensor_utils.h"
 #include "paddle/phi/kernels/funcs/adam_functors.h"
 #include "paddle/phi/kernels/funcs/for_range.h"
 

paddle/phi/kernels/gpu/adamw_kernel.cu

@@ -24,7 +24,7 @@
 #include "paddle/phi/common/amp_type_traits.h"
 #include "paddle/phi/common/float16.h"
 #include "paddle/phi/core/kernel_registry.h"
-#include "paddle/phi/kernels/copy_kernel.h"
+#include "paddle/phi/core/tensor_utils.h"
 #include "paddle/phi/kernels/funcs/adam_functors.h"
 #include "paddle/phi/kernels/funcs/for_range.h"
 

paddle/phi/kernels/gpu/arange_kernel.cu

@@ -16,7 +16,7 @@
 
 #include "paddle/phi/backends/gpu/gpu_context.h"
 #include "paddle/phi/core/kernel_registry.h"
-#include "paddle/phi/kernels/copy_kernel.h"
+#include "paddle/phi/core/tensor_utils.h"
 #include "paddle/phi/kernels/funcs/range_function.h"
 
 namespace phi {

paddle/phi/kernels/gpu/cross_entropy_grad_kernel.cu

@@ -28,8 +28,8 @@ namespace cub = hipcub;
 #include "paddle/fluid/platform/device/gpu/gpu_dnn.h"
 #include "paddle/phi/common/amp_type_traits.h"
 #include "paddle/phi/core/kernel_registry.h"
+#include "paddle/phi/core/tensor_utils.h"
 #include "paddle/phi/core/visit_type.h"
-#include "paddle/phi/kernels/copy_kernel.h"
 #include "paddle/phi/kernels/funcs/axis_utils.h"
 #include "paddle/phi/kernels/funcs/for_range.h"
 #include "paddle/phi/kernels/funcs/math_function.h"

paddle/phi/kernels/gpu/cross_entropy_kernel.cu

@@ -28,8 +28,8 @@ namespace cub = hipcub;
 #include "paddle/fluid/platform/device/gpu/gpu_dnn.h"
 #include "paddle/phi/common/amp_type_traits.h"
 #include "paddle/phi/core/kernel_registry.h"
+#include "paddle/phi/core/tensor_utils.h"
 #include "paddle/phi/core/visit_type.h"
-#include "paddle/phi/kernels/copy_kernel.h"
 #include "paddle/phi/kernels/funcs/axis_utils.h"
 #include "paddle/phi/kernels/funcs/for_range.h"
 #include "paddle/phi/kernels/funcs/math_function.h"

paddle/phi/kernels/gpu/elementwise_add_grad_kernel.cu

@@ -19,7 +19,7 @@
 #include "paddle/phi/common/complex.h"
 #include "paddle/phi/common/float16.h"
 #include "paddle/phi/core/kernel_registry.h"
-#include "paddle/phi/kernels/copy_kernel.h"
+#include "paddle/phi/core/tensor_utils.h"
 #include "paddle/phi/kernels/funcs/elementwise_functor.h"
 #include "paddle/phi/kernels/gpu/elementwise_grad.h"
 #include "paddle/phi/kernels/impl/elementwise_grad_kernel_impl.h"

paddle/phi/kernels/gpu/elementwise_grad.h

@@ -15,7 +15,7 @@ limitations under the License. */
 #pragma once
 
 #include "paddle/phi/common/place.h"
-#include "paddle/phi/kernels/copy_kernel.h"
+#include "paddle/phi/core/tensor_utils.h"
 #include "paddle/phi/kernels/funcs/broadcast_function.h"
 #include "paddle/phi/kernels/funcs/elementwise_grad_base.h"
 #include "paddle/phi/kernels/funcs/reduce_function.h"

paddle/phi/kernels/gpu/elementwise_grad_kernel.cu

@@ -19,7 +19,7 @@
 #include "paddle/phi/common/complex.h"
 #include "paddle/phi/common/float16.h"
 #include "paddle/phi/core/kernel_registry.h"
-#include "paddle/phi/kernels/copy_kernel.h"
+#include "paddle/phi/core/tensor_utils.h"
 #include "paddle/phi/kernels/funcs/elementwise_functor.h"
 #include "paddle/phi/kernels/gpu/elementwise_grad.h"
 #include "paddle/phi/kernels/impl/elementwise_grad_kernel_impl.h"

paddle/phi/kernels/gpu/elementwise_subtract_grad_kernel.cu

@@ -17,7 +17,7 @@
 #include "paddle/phi/common/complex.h"
 #include "paddle/phi/common/float16.h"
 #include "paddle/phi/core/kernel_registry.h"
-#include "paddle/phi/kernels/copy_kernel.h"
+#include "paddle/phi/core/tensor_utils.h"
 #include "paddle/phi/kernels/elementwise_grad_kernel.h"
 #include "paddle/phi/kernels/funcs/elementwise_functor.h"
 #include "paddle/phi/kernels/gpu/elementwise_grad.h"

paddle/phi/kernels/gpu/instance_norm_grad_kernel.cu

@@ -17,7 +17,7 @@
 #include "paddle/phi/backends/gpu/gpu_context.h"
 #include "paddle/phi/common/layout.h"
 #include "paddle/phi/core/kernel_registry.h"
-#include "paddle/phi/kernels/copy_kernel.h"
+#include "paddle/phi/core/tensor_utils.h"
 #include "paddle/phi/kernels/funcs/math_function.h"
 #include "paddle/phi/kernels/funcs/norm_utils.h"
 #include "paddle/phi/kernels/gpu/instance_norm_utils.h"

paddle/phi/kernels/gpu/linspace_kernel.cu

@@ -17,7 +17,7 @@
 #include "paddle/fluid/platform/device/gpu/gpu_primitives.h"
 #include "paddle/phi/backends/gpu/gpu_context.h"
 #include "paddle/phi/core/kernel_registry.h"
-#include "paddle/phi/kernels/copy_kernel.h"
+#include "paddle/phi/core/tensor_utils.h"
 #include "paddle/phi/kernels/funcs/math_function.h"
 
 namespace phi {

paddle/phi/kernels/gpu/logspace_kernel.cu

@@ -16,7 +16,7 @@
 
 #include "paddle/phi/backends/gpu/gpu_context.h"
 #include "paddle/phi/core/kernel_registry.h"
-#include "paddle/phi/kernels/copy_kernel.h"
+#include "paddle/phi/core/tensor_utils.h"
 #include "paddle/phi/kernels/funcs/data_type_transform.h"
 #include "paddle/phi/kernels/funcs/math_function.h"
 

paddle/phi/kernels/gpu/psroi_pool_grad_kernel.cu

@@ -19,7 +19,7 @@
 #include "paddle/fluid/platform/device/gpu/gpu_primitives.h"
 #include "paddle/phi/common/place.h"
 #include "paddle/phi/core/kernel_registry.h"
-#include "paddle/phi/kernels/copy_kernel.h"
+#include "paddle/phi/core/tensor_utils.h"
 #include "paddle/phi/kernels/funcs/math_function.h"
 #include "paddle/phi/kernels/psroi_pool_kernel.h"
 

paddle/phi/kernels/gpu/psroi_pool_kernel.cu

@@ -20,7 +20,7 @@
 #include "paddle/fluid/memory/memory.h"
 #include "paddle/phi/common/place.h"
 #include "paddle/phi/core/kernel_registry.h"
-#include "paddle/phi/kernels/copy_kernel.h"
+#include "paddle/phi/core/tensor_utils.h"
 
 namespace phi {
 

paddle/phi/kernels/gpu/put_along_axis_grad_kernel.cu

@@ -19,7 +19,7 @@
 #include "paddle/fluid/platform/place.h"
 #include "paddle/phi/backends/gpu/gpu_context.h"
 #include "paddle/phi/core/kernel_registry.h"
-#include "paddle/phi/kernels/copy_kernel.h"
+#include "paddle/phi/core/tensor_utils.h"
 
 namespace phi {
 

paddle/phi/kernels/gpu/put_along_axis_kernel.cu

@@ -19,7 +19,7 @@
 #include "paddle/fluid/platform/place.h"
 #include "paddle/phi/backends/gpu/gpu_context.h"
 #include "paddle/phi/core/kernel_registry.h"
-#include "paddle/phi/kernels/copy_kernel.h"
+#include "paddle/phi/core/tensor_utils.h"
 
 namespace phi {
 

paddle/phi/kernels/gpu/scatter_grad_kernel.cu

@@ -16,7 +16,7 @@
 
 #include "paddle/phi/backends/gpu/gpu_context.h"
 #include "paddle/phi/core/kernel_registry.h"
-#include "paddle/phi/kernels/copy_kernel.h"
+#include "paddle/phi/core/tensor_utils.h"
 #include "paddle/phi/kernels/funcs/gather.cu.h"
 #include "paddle/phi/kernels/funcs/scatter.cu.h"
 

paddle/phi/kernels/gpu/scatter_kernel.cu

@@ -16,7 +16,7 @@
 
 #include "paddle/phi/backends/gpu/gpu_context.h"
 #include "paddle/phi/core/kernel_registry.h"
-#include "paddle/phi/kernels/copy_kernel.h"
+#include "paddle/phi/core/tensor_utils.h"
 #include "paddle/phi/kernels/funcs/scatter.cu.h"
 
 namespace phi {

paddle/phi/kernels/gpu/scatter_nd_add_grad_kernel.cu

@@ -16,7 +16,7 @@
 
 #include "paddle/phi/backends/gpu/gpu_context.h"
 #include "paddle/phi/core/kernel_registry.h"
-#include "paddle/phi/kernels/copy_kernel.h"
+#include "paddle/phi/core/tensor_utils.h"
 #include "paddle/phi/kernels/funcs/gather.cu.h"
 
 namespace phi {

paddle/phi/kernels/gpu/scatter_nd_add_kernel.cu

@@ -16,7 +16,7 @@
 
 #include "paddle/phi/backends/gpu/gpu_context.h"
 #include "paddle/phi/core/kernel_registry.h"
-#include "paddle/phi/kernels/copy_kernel.h"
+#include "paddle/phi/core/tensor_utils.h"
 #include "paddle/phi/kernels/funcs/scatter.cu.h"
 
 namespace phi {

paddle/phi/kernels/gpu/sigmoid_cross_entropy_with_logits.h

@@ -22,7 +22,7 @@
 #include "paddle/phi/backends/gpu/gpu_helper.h"
 #include "paddle/phi/core/hostdevice.h"
 #include "paddle/phi/core/kernel_registry.h"
-#include "paddle/phi/kernels/copy_kernel.h"
+#include "paddle/phi/core/tensor_utils.h"
 #include "paddle/phi/kernels/funcs/elementwise_base.h"
 #include "paddle/phi/kernels/gpu/reduce.h"
 

paddle/phi/kernels/gpu/top_k_kernel.cu

@@ -17,7 +17,7 @@
 #include "paddle/fluid/operators/top_k_function_cuda.h"
 #include "paddle/phi/backends/gpu/gpu_context.h"
 #include "paddle/phi/core/kernel_registry.h"
-#include "paddle/phi/kernels/copy_kernel.h"
+#include "paddle/phi/core/tensor_utils.h"
 #include "paddle/phi/kernels/funcs/gather.cu.h"
 #include "paddle/phi/kernels/funcs/math_function.h"
 

paddle/phi/kernels/gpu/unique_kernel.cu

@@ -28,7 +28,7 @@
 #include "paddle/fluid/framework/tensor_util.h"  // TensorToVector()
 #include "paddle/phi/backends/gpu/gpu_context.h"
 #include "paddle/phi/core/kernel_registry.h"
-#include "paddle/phi/kernels/copy_kernel.h"
+#include "paddle/phi/core/tensor_utils.h"
 #include "paddle/phi/kernels/funcs/unique_functor.h"
 
 namespace phi {

paddle/phi/kernels/gpu/viterbi_decode_kernel.cu

@@ -33,7 +33,7 @@ namespace cub = hipcub;
 #include "paddle/fluid/operators/elementwise/elementwise_op_function.h"
 #include "paddle/phi/backends/gpu/gpu_context.h"
 #include "paddle/phi/core/kernel_registry.h"
-#include "paddle/phi/kernels/copy_kernel.h"
+#include "paddle/phi/core/tensor_utils.h"
 #include "paddle/phi/kernels/empty_kernel.h"
 #include "paddle/phi/kernels/funcs/compare_functors.h"
 #include "paddle/phi/kernels/funcs/concat_and_split_functor.h"

paddle/phi/kernels/impl/cholesky_solve_grad_kernel_impl.h

@@ -17,7 +17,6 @@
 #include "paddle/phi/kernels/cholesky_solve_grad_kernel.h"
 #include "paddle/phi/kernels/cholesky_solve_kernel.h"
 #include "paddle/phi/kernels/complex_kernel.h"
-#include "paddle/phi/kernels/copy_kernel.h"
 #include "paddle/phi/kernels/elementwise_add_kernel.h"
 #include "paddle/phi/kernels/empty_kernel.h"
 #include "paddle/phi/kernels/expand_kernel.h"

paddle/phi/kernels/impl/cholesky_solve_kernel_impl.h

@@ -16,7 +16,6 @@
 
 #include "paddle/phi/kernels/cholesky_solve_kernel.h"
 #include "paddle/phi/kernels/complex_kernel.h"
-#include "paddle/phi/kernels/copy_kernel.h"
 #include "paddle/phi/kernels/empty_kernel.h"
 #include "paddle/phi/kernels/expand_kernel.h"
 #include "paddle/phi/kernels/funcs/common_shape.h"

paddle/phi/kernels/impl/determinant_grad_kernel_impl.h

@@ -14,7 +14,7 @@
 
 #pragma once
 
-#include "paddle/phi/kernels/copy_kernel.h"
+#include "paddle/phi/core/tensor_utils.h"
 #include "paddle/phi/kernels/determinant_grad_kernel.h"
 #include "paddle/phi/kernels/elementwise_multiply_kernel.h"
 #include "paddle/phi/kernels/empty_kernel.h"

paddle/phi/kernels/impl/elementwise_grad_kernel_impl.h

@@ -16,7 +16,7 @@ limitations under the License. */
 
 #include "paddle/phi/common/complex.h"
 #include "paddle/phi/core/dense_tensor.h"
-#include "paddle/phi/kernels/copy_kernel.h"
+#include "paddle/phi/core/tensor_utils.h"
 #include "paddle/phi/kernels/funcs/broadcast_function.h"
 #include "paddle/phi/kernels/funcs/eigen/common.h"
 #include "paddle/phi/kernels/funcs/elementwise_functor.h"

paddle/phi/kernels/impl/expand_as_grad_kernel_impl.h

@@ -14,7 +14,7 @@
 
 #pragma once
 
-#include "paddle/phi/kernels/copy_kernel.h"
+#include "paddle/phi/core/tensor_utils.h"
 #include "paddle/phi/kernels/impl/expand_as_kernel_impl.h"
 
 namespace phi {

paddle/phi/kernels/impl/expand_grad_kernel_impl.h

@@ -14,7 +14,7 @@
 
 #pragma once
 
-#include "paddle/phi/kernels/copy_kernel.h"
+#include "paddle/phi/core/tensor_utils.h"
 #include "paddle/phi/kernels/funcs/eigen/common.h"
 #include "paddle/phi/kernels/funcs/eigen/eigen_function.h"
 #include "paddle/phi/kernels/impl/expand_kernel_impl.h"

paddle/phi/kernels/impl/meshgrid_kernel_impl.h

@@ -16,7 +16,7 @@
 
 #include "paddle/fluid/framework/tensor_util.h"
 #include "paddle/phi/core/dense_tensor.h"
-#include "paddle/phi/kernels/copy_kernel.h"
+#include "paddle/phi/core/tensor_utils.h"
 #include "paddle/phi/kernels/funcs/eigen/common.h"
 #include "paddle/phi/kernels/funcs/eigen/eigen_function.h"
 #include "paddle/phi/kernels/meshgrid_kernel.h"

paddle/phi/kernels/impl/set_value_grad_kernel_impl.h

@@ -16,7 +16,7 @@
 
 #include "paddle/phi/common/int_array.h"
 #include "paddle/phi/core/dense_tensor.h"
-#include "paddle/phi/kernels/copy_kernel.h"
+#include "paddle/phi/core/tensor_utils.h"
 #include "paddle/phi/kernels/full_kernel.h"
 #include "paddle/phi/kernels/funcs/eigen/common.h"
 #include "paddle/phi/kernels/funcs/eigen/eigen_function.h"

paddle/phi/kernels/impl/set_value_kernel_impl.h

@@ -17,7 +17,7 @@
 #include "paddle/phi/common/int_array.h"
 #include "paddle/phi/common/scalar.h"
 #include "paddle/phi/core/dense_tensor.h"
-#include "paddle/phi/kernels/copy_kernel.h"
+#include "paddle/phi/core/tensor_utils.h"
 #include "paddle/phi/kernels/empty_kernel.h"
 #include "paddle/phi/kernels/funcs/broadcast_function.h"
 #include "paddle/phi/kernels/funcs/eigen/common.h"

paddle/phi/kernels/impl/size_kernel_impl.h

@@ -14,7 +14,7 @@
 
 #pragma once
 
-#include "paddle/phi/kernels/copy_kernel.h"
+#include "paddle/phi/core/tensor_utils.h"
 
 namespace phi {
 

paddle/phi/kernels/impl/squeeze_grad_kernel_impl.h

@@ -14,7 +14,7 @@
 
 #pragma once
 #include "paddle/phi/core/dense_tensor.h"
-#include "paddle/phi/kernels/copy_kernel.h"
+#include "paddle/phi/core/tensor_utils.h"
 
 namespace phi {
 template <typename T, typename Context>

paddle/phi/kernels/impl/squeeze_kernel_impl.h

@@ -14,7 +14,7 @@
 
 #pragma once
 #include "paddle/phi/core/dense_tensor.h"
-#include "paddle/phi/kernels/copy_kernel.h"
+#include "paddle/phi/core/tensor_utils.h"
 #include "paddle/phi/kernels/funcs/unsqueeze.h"
 
 namespace phi {

paddle/phi/kernels/impl/tile_grad_kernel_impl.h

@@ -16,7 +16,7 @@
 #include <type_traits>
 #include <vector>
 
-#include "paddle/phi/kernels/copy_kernel.h"
+#include "paddle/phi/core/tensor_utils.h"
 #include "paddle/phi/kernels/funcs/eigen/common.h"
 #include "paddle/phi/kernels/funcs/eigen/eigen_function.h"
 #include "paddle/phi/kernels/tile_grad_kernel.h"

paddle/phi/kernels/impl/triangular_solve_grad_kernel_impl.h

@@ -14,7 +14,7 @@
 
 #pragma once
 
-#include "paddle/phi/kernels/copy_kernel.h"
+#include "paddle/phi/core/tensor_utils.h"
 #include "paddle/phi/kernels/empty_kernel.h"
 #include "paddle/phi/kernels/funcs/blas/blas.h"
 #include "paddle/phi/kernels/funcs/common_shape.h"

paddle/phi/kernels/impl/unsqueeze_grad_kernel_impl.h

@@ -14,7 +14,7 @@
 
 #pragma once
 #include "paddle/phi/core/dense_tensor.h"
-#include "paddle/phi/kernels/copy_kernel.h"
+#include "paddle/phi/core/tensor_utils.h"
 
 namespace phi {
 template <typename T, typename Context>

paddle/phi/kernels/impl/unsqueeze_kernel_impl.h

@@ -14,7 +14,7 @@
 
 #pragma once
 #include "paddle/phi/core/dense_tensor.h"
-#include "paddle/phi/kernels/copy_kernel.h"
+#include "paddle/phi/core/tensor_utils.h"
 #include "paddle/phi/kernels/funcs/unsqueeze.h"
 
 namespace phi {

paddle/phi/kernels/impl/warpctc_kernel_impl.h

@@ -20,7 +20,7 @@
 #include "paddle/fluid/operators/math/sequence_scale.h"
 #include "paddle/phi/backends/dynload/warpctc.h"
 #include "paddle/phi/core/dense_tensor.h"
-#include "paddle/phi/kernels/copy_kernel.h"
+#include "paddle/phi/core/tensor_utils.h"
 #include "paddle/phi/kernels/empty_kernel.h"
 #include "paddle/phi/kernels/funcs/math_function.h"
 #include "paddle/utils/optional.h"

paddle/phi/kernels/reshape_grad_kernel.cc

@@ -16,7 +16,7 @@
 
 #include "paddle/phi/backends/all_context.h"
 #include "paddle/phi/core/kernel_registry.h"
-#include "paddle/phi/kernels/copy_kernel.h"
+#include "paddle/phi/core/tensor_utils.h"
 
 namespace phi {
 

paddle/phi/kernels/reshape_kernel.cc

@@ -16,8 +16,8 @@
 
 #include "paddle/phi/backends/all_context.h"
 #include "paddle/phi/core/kernel_registry.h"
+#include "paddle/phi/core/tensor_utils.h"
 #include "paddle/phi/infermeta/unary.h"
-#include "paddle/phi/kernels/copy_kernel.h"
 #include "paddle/phi/kernels/funcs/common_shape.h"
 
 namespace phi {

paddle/phi/kernels/reverse_kernel.cc

@@ -16,7 +16,7 @@
 
 #include "paddle/phi/core/enforce.h"
 #include "paddle/phi/core/kernel_registry.h"
-#include "paddle/phi/kernels/copy_kernel.h"
+#include "paddle/phi/core/tensor_utils.h"
 
 namespace phi {
 

paddle/phi/kernels/selected_rows/copy_kernel.cc

-/* 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. */
-
-#include "paddle/phi/kernels/selected_rows/copy_kernel.h"
-
-#include "paddle/phi/backends/cpu/cpu_context.h"
-#include "paddle/phi/backends/gpu/gpu_context.h"
-#include "paddle/phi/common/bfloat16.h"
-#include "paddle/phi/core/kernel_registry.h"
-#include "paddle/phi/kernels/copy_kernel.h"
-namespace phi {
-namespace sr {
-
-template <typename Context>
-void Copy(const Context& dev_ctx,
-          const SelectedRows& src,
-          Place dst_place,
-          bool blocking,
-          SelectedRows* dst) {
-  if (src.value().Holder() != dst->value().Holder() ||
-      src.value().data() != dst->value().data()) {
-    dst->set_rows(src.rows());
-    dst->set_height(src.height());
-  }
-  phi::Copy<Context>(
-      dev_ctx, src.value(), dst_place, blocking, dst->mutable_value());
-}
-
-}  // namespace sr
-}  // namespace phi
-
-PD_REGISTER_GENERAL_KERNEL(
-    copy_sr, CPU, ALL_LAYOUT, phi::sr::Copy<phi::CPUContext>, ALL_DTYPE) {}
-
-#if defined(PADDLE_WITH_CUDA) || defined(PADDLE_WITH_HIP)
-PD_REGISTER_GENERAL_KERNEL(
-    copy_sr, GPU, ALL_LAYOUT, phi::sr::Copy<phi::GPUContext>, ALL_DTYPE) {}
-#endif

paddle/phi/kernels/selected_rows/copy_kernel.h

-// 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.
-
-#pragma once
-
-#include "paddle/phi/core/selected_rows.h"
-#include "paddle/phi/core/sparse_csr_tensor.h"
-
-namespace phi {
-namespace sr {
-
-template <typename Context>
-void Copy(const Context& dev_ctx,
-          const SelectedRows& src,
-          Place dst_place,
-          bool blocking,
-          SelectedRows* dst);
-
-}  // namespace sr
-}  // namespace phi

paddle/phi/kernels/selected_rows/cpu/adam_kernel.cc

@@ -19,7 +19,7 @@
 #include "paddle/fluid/operators/math/selected_rows_functor.h"
 #include "paddle/phi/backends/cpu/cpu_context.h"
 #include "paddle/phi/core/kernel_registry.h"
-#include "paddle/phi/kernels/copy_kernel.h"
+#include "paddle/phi/core/tensor_utils.h"
 #include "paddle/phi/kernels/funcs/adam_functors.h"
 
 namespace phi {

paddle/phi/kernels/selected_rows/gpu/adam_kernel.cu

@@ -20,7 +20,7 @@
 #include "paddle/phi/common/amp_type_traits.h"
 #include "paddle/phi/common/float16.h"
 #include "paddle/phi/core/kernel_registry.h"
-#include "paddle/phi/kernels/copy_kernel.h"
+#include "paddle/phi/core/tensor_utils.h"
 #include "paddle/phi/kernels/funcs/adam_functors.h"
 #include "paddle/phi/kernels/funcs/for_range.h"
 

paddle/phi/kernels/selected_rows/gpu/adamw_kernel.cu

@@ -24,7 +24,7 @@
 #include "paddle/phi/common/amp_type_traits.h"
 #include "paddle/phi/common/float16.h"
 #include "paddle/phi/core/kernel_registry.h"
-#include "paddle/phi/kernels/copy_kernel.h"
+#include "paddle/phi/core/tensor_utils.h"
 #include "paddle/phi/kernels/funcs/adam_functors.h"
 #include "paddle/phi/kernels/funcs/for_range.h"
 

paddle/phi/kernels/sparse/copy_kernel.cc

@@ -19,7 +19,7 @@ limitations under the License. */
 #include "paddle/phi/core/kernel_registry.h"
 #include "paddle/phi/core/sparse_coo_tensor.h"
 #include "paddle/phi/core/sparse_csr_tensor.h"
-#include "paddle/phi/kernels/copy_kernel.h"
+#include "paddle/phi/core/tensor_utils.h"
 
 namespace phi {
 namespace sparse {

paddle/phi/kernels/sparse/cpu/convolution_grad_kernel.cc

@@ -15,7 +15,6 @@ limitations under the License. */
 #include "paddle/phi/kernels/sparse/convolution_grad_kernel.h"
 
 #include "paddle/phi/core/visit_type.h"
-#include "paddle/phi/kernels/copy_kernel.h"
 #include "paddle/phi/kernels/funcs/blas/blas.h"
 #include "paddle/phi/kernels/funcs/math_function.h"
 #include "paddle/phi/kernels/sparse/cpu/convolution.h"

paddle/phi/kernels/sparse/cpu/full_kernel.cc

@@ -17,7 +17,7 @@ limitations under the License. */
 #include "paddle/phi/backends/cpu/cpu_context.h"
 #include "paddle/phi/core/dense_tensor.h"
 #include "paddle/phi/core/kernel_registry.h"
-#include "paddle/phi/kernels/copy_kernel.h"
+#include "paddle/phi/core/tensor_utils.h"
 #include "paddle/phi/kernels/funcs/eigen/common.h"
 #include "paddle/phi/kernels/funcs/eigen/eigen_function.h"
 

paddle/phi/kernels/sparse/cpu/sparse_mask_kernel.cc

@@ -18,8 +18,8 @@ limitations under the License. */
 #include "paddle/phi/core/ddim.h"
 #include "paddle/phi/core/enforce.h"
 #include "paddle/phi/core/kernel_registry.h"
+#include "paddle/phi/core/tensor_utils.h"
 #include "paddle/phi/core/visit_type.h"
-#include "paddle/phi/kernels/copy_kernel.h"
 #include "paddle/phi/kernels/empty_kernel.h"
 #include "paddle/phi/kernels/funcs/math_function.h"
 #include "paddle/phi/kernels/funcs/sparse/flatten_indices.h"

paddle/phi/kernels/sparse/cpu/sparse_pool_grad_kernel.cc

@@ -15,8 +15,8 @@ limitations under the License. */
 #include "paddle/phi/kernels/sparse/sparse_pool_grad_kernel.h"
 
 #include "paddle/phi/core/kernel_registry.h"
+#include "paddle/phi/core/tensor_utils.h"
 #include "paddle/phi/core/visit_type.h"
-#include "paddle/phi/kernels/copy_kernel.h"
 #include "paddle/phi/kernels/empty_kernel.h"
 #include "paddle/phi/kernels/funcs/pooling.h"
 #include "paddle/phi/kernels/funcs/sparse/convolution.h"

paddle/phi/kernels/sparse/empty_kernel.cc

@@ -17,7 +17,7 @@ limitations under the License. */
 #include "paddle/phi/backends/cpu/cpu_context.h"
 #include "paddle/phi/backends/gpu/gpu_context.h"
 #include "paddle/phi/core/kernel_registry.h"
-#include "paddle/phi/kernels/copy_kernel.h"
+#include "paddle/phi/core/tensor_utils.h"
 
 namespace phi {
 namespace sparse {

paddle/phi/kernels/sparse/gpu/convolution.cu.h

@@ -23,7 +23,7 @@ limitations under the License. */
 #include "paddle/phi/backends/gpu/gpu_context.h"
 #include "paddle/phi/backends/gpu/gpu_info.h"
 #include "paddle/phi/backends/gpu/gpu_launch_config.h"
-#include "paddle/phi/kernels/copy_kernel.h"
+#include "paddle/phi/core/tensor_utils.h"
 #include "paddle/phi/kernels/funcs/index_impl.cu.h"
 #include "paddle/phi/kernels/funcs/math_function.h"
 #include "paddle/phi/kernels/funcs/sparse/utils.cu.h"

paddle/phi/kernels/sparse/gpu/convolution_grad_kernel.cu

@@ -20,8 +20,8 @@ limitations under the License. */
 #include "paddle/phi/backends/gpu/gpu_launch_config.h"
 #include "paddle/phi/core/kernel_registry.h"
 #include "paddle/phi/core/tensor_meta.h"
+#include "paddle/phi/core/tensor_utils.h"
 #include "paddle/phi/core/visit_type.h"
-#include "paddle/phi/kernels/copy_kernel.h"
 #include "paddle/phi/kernels/funcs/blas/blas.h"
 #include "paddle/phi/kernels/funcs/math_function.h"
 #include "paddle/phi/kernels/funcs/scatter.cu.h"

paddle/phi/kernels/sparse/gpu/full_kernel.cu

@@ -17,7 +17,7 @@ limitations under the License. */
 #include "paddle/phi/backends/gpu/gpu_context.h"
 #include "paddle/phi/core/dense_tensor.h"
 #include "paddle/phi/core/kernel_registry.h"
-#include "paddle/phi/kernels/copy_kernel.h"
+#include "paddle/phi/core/tensor_utils.h"
 #include "paddle/phi/kernels/funcs/elementwise_base.h"
 
 namespace phi {

paddle/phi/kernels/sparse/gpu/matmul_grad_kernel.cu

@@ -18,7 +18,7 @@ limitations under the License. */
 
 #include "paddle/phi/backends/gpu/gpu_context.h"
 #include "paddle/phi/core/kernel_registry.h"
-#include "paddle/phi/kernels/copy_kernel.h"
+#include "paddle/phi/core/tensor_utils.h"
 #include "paddle/phi/kernels/empty_kernel.h"
 #include "paddle/phi/kernels/funcs/sparse/sparse_blas.h"
 #include "paddle/phi/kernels/sparse/empty_kernel.h"

paddle/phi/kernels/sparse/gpu/matmul_kernel.cu

@@ -23,7 +23,7 @@ limitations under the License. */
 #include "paddle/phi/core/meta_tensor.h"
 #include "paddle/phi/core/sparse_coo_tensor.h"
 #include "paddle/phi/core/sparse_csr_tensor.h"
-#include "paddle/phi/kernels/copy_kernel.h"
+#include "paddle/phi/core/tensor_utils.h"
 #include "paddle/phi/kernels/empty_kernel.h"
 #include "paddle/phi/kernels/funcs/sparse/sparse_blas.h"
 #include "paddle/phi/kernels/sparse/empty_kernel.h"