[pten]Remove pten/include dir files (#38878)

* move dot_dev api into dot_kernel.h

* add infermate header

* modify to dotkerel in dot_op.h

* mvoe conj dev api into complex_kernel.h

* move sign dev api into  sign_kernel.h

* move scale dev api into kernel.h and remove infermete.h

* rm paddle/pten/include/math.h

* rm paddle/pten/include/math.h

* rm include dir

* rm paddle/pten/include/math.h

* fix conflict with develop branch

* rm devContext in conj_op.h

* add the missing complex_kernel header

paddle/fluid/eager/accumulation/accumulation_node.cc

@@ -18,7 +18,6 @@
 
 #include "paddle/pten/api/all.h"
 #include "paddle/pten/core/dense_tensor.h"
-#include "paddle/pten/include/core.h"
 
 #include "paddle/fluid/platform/device_context.h"
 #include "paddle/fluid/platform/enforce.h"

paddle/fluid/eager/accumulation/gradient_accumulation.cc

@@ -28,7 +28,6 @@
 #include "paddle/fluid/platform/float16.h"
 #include "paddle/pten/api/all.h"
 #include "paddle/pten/core/convert_utils.h"
-#include "paddle/pten/include/core.h"
 #include "unsupported/Eigen/CXX11/Tensor"
 #ifdef PADDLE_WITH_XPU
 #include "xpu/refactor/math.h"

paddle/fluid/eager/api/generated/eager_generated/backwards/scale_node.cc

@@ -16,7 +16,7 @@
 #include "paddle/fluid/eager/api/utils/global_utils.h"
 #include "paddle/fluid/eager/eager_tensor.h"
 
-#include "paddle/pten/api/all.h"
+#include "paddle/pten/kernels/scale_kernel.h"
 
 #include "paddle/fluid/platform/device_context.h"
 #include "paddle/fluid/platform/enforce.h"
@@ -33,28 +33,28 @@ static void ScaleDeviceDispatch(const pten::DenseTensor& dense_tensor,
                                 pten::DenseTensor* dense_out) {
   switch (dense_tensor.dtype()) {
     case pten::DataType::FLOAT64: {
-      pten::Scale<double, DeviceContext>(
+      pten::ScaleKernel<double, DeviceContext>(
           dev_ctx, dense_tensor /* tensor */, scale /* scale */,
           bias /* bias */, bias_after_scale /* bias_after_scale */,
           dense_out /* out tensor */);
       break;
     }
     case pten::DataType::FLOAT32: {
-      pten::Scale<float, DeviceContext>(dev_ctx, dense_tensor /* tensor */,
-                                        scale /* scale */, bias /* bias */,
-                                        bias_after_scale /* bias_after_scale */,
+      pten::ScaleKernel<float, DeviceContext>(
+          dev_ctx, dense_tensor /* tensor */, scale /* scale */,
+          bias /* bias */, bias_after_scale /* bias_after_scale */,
           dense_out /* out tensor */);
       break;
     }
     case pten::DataType::INT64: {
-      pten::Scale<int64_t, DeviceContext>(
+      pten::ScaleKernel<int64_t, DeviceContext>(
           dev_ctx, dense_tensor /* tensor */, scale /* scale */,
           bias /* bias */, bias_after_scale /* bias_after_scale */,
           dense_out /* out tensor */);
       break;
     }
     case pten::DataType::INT32: {
-      pten::Scale<int32_t, DeviceContext>(
+      pten::ScaleKernel<int32_t, DeviceContext>(
           dev_ctx, dense_tensor /* tensor */, scale /* scale */,
           bias /* bias */, bias_after_scale /* bias_after_scale */,
           dense_out /* out tensor */);

paddle/fluid/eager/api/generated/eager_generated/forwards/scale.cc

@@ -29,7 +29,6 @@
 #include "paddle/fluid/eager/utils.h"
 
 #include "paddle/pten/api/all.h"
-#include "paddle/pten/include/core.h"
 
 namespace egr {
 

paddle/fluid/eager/eager_tensor.h

@@ -18,7 +18,6 @@
 #include "paddle/fluid/framework/tensor.h"
 #include "paddle/fluid/framework/variable.h"
 // pten deps
-#include "paddle/pten/all.h"
 #include "paddle/pten/api/all.h"
 #include "paddle/pten/api/lib/api_declare.h"
 #include "paddle/pten/api/lib/utils/tensor_utils.h"

paddle/fluid/eager/grad_node_info.h

@@ -16,7 +16,6 @@
 
 #include "paddle/fluid/eager/eager_tensor.h"
 #include "paddle/pten/api/all.h"
-#include "paddle/pten/include/core.h"
 
 namespace egr {
 /**

paddle/fluid/eager/legacy/infer_var_type_context.h

@@ -26,7 +26,6 @@
 #include "paddle/fluid/framework/var_type_inference.h"
 #include "paddle/fluid/framework/var_type_traits.h"
 #include "paddle/pten/api/all.h"
-#include "paddle/pten/include/core.h"
 
 namespace egr {
 namespace legacy {

paddle/fluid/eager/legacy/prepared_operator.h

@@ -25,8 +25,6 @@
 #include "paddle/fluid/framework/operator.h"
 #include "paddle/fluid/framework/type_defs.h"
 
-#include "paddle/pten/include/core.h"
-
 DECLARE_bool(use_mkldnn);
 
 namespace paddle {

paddle/fluid/eager/legacy/tensor_helper.h

@@ -17,7 +17,6 @@
 #include <vector>
 #include "paddle/fluid/eager/eager_tensor.h"
 #include "paddle/pten/api/all.h"
-#include "paddle/pten/include/core.h"
 namespace egr {
 namespace legacy {
 

paddle/fluid/framework/data_device_transform_test.cu

@@ -24,7 +24,6 @@ limitations under the License. */
 #include "paddle/fluid/platform/init.h"
 
 #include "paddle/fluid/framework/pten_utils.h"
-#include "paddle/pten/include/core.h"
 
 namespace paddle {
 namespace framework {

paddle/fluid/framework/operator.h

@@ -41,7 +41,8 @@ limitations under the License. */
 #include "paddle/utils/flat_hash_map.h"
 
 #include "paddle/pten/core/arg_map_context.h"
-#include "paddle/pten/include/core.h"
+#include "paddle/pten/core/kernel_context.h"
+#include "paddle/pten/core/kernel_factory.h"
 
 namespace paddle {
 namespace framework {

paddle/fluid/imperative/layer.h

@@ -37,7 +37,6 @@
 #include "paddle/fluid/imperative/variable_wrapper.h"
 #include "paddle/fluid/platform/enforce.h"
 #include "paddle/fluid/platform/macros.h"
-#include "paddle/pten/include/core.h"
 namespace paddle {
 namespace framework {
 class Variable;

paddle/fluid/imperative/op_base.h

@@ -25,7 +25,6 @@
 #include "paddle/fluid/imperative/type_defs.h"
 #include "paddle/fluid/imperative/variable_wrapper.h"
 #include "paddle/fluid/platform/place.h"
-#include "paddle/pten/include/core.h"
 
 namespace paddle {
 namespace imperative {

paddle/fluid/imperative/prepared_operator.h

@@ -27,8 +27,6 @@
 #include "paddle/fluid/imperative/layer.h"
 #include "paddle/fluid/imperative/type_defs.h"
 
-#include "paddle/pten/include/core.h"
-
 DECLARE_bool(use_mkldnn);
 
 namespace paddle {

paddle/fluid/operators/cast_op.h

@@ -19,7 +19,6 @@ limitations under the License. */
 #include "paddle/fluid/platform/transform.h"
 
 #include "paddle/pten/api/lib/utils/tensor_utils.h"
-#include "paddle/pten/include/core.h"
 #include "paddle/pten/kernels/cast_kernel.h"
 
 namespace paddle {

paddle/fluid/operators/conj_op.h

@@ -19,7 +19,6 @@
 
 // only can include the headers in paddle/pten/api dirs
 #include "paddle/pten/api/lib/utils/tensor_utils.h"
-#include "paddle/pten/include/core.h"
 #include "paddle/pten/kernels/complex_kernel.h"
 
 namespace paddle {
@@ -39,7 +38,7 @@ class ConjKernel : public framework::OpKernel<T> {
     auto pt_out = paddle::experimental::MakePtenDenseTensor(*out);
 
     // call new kernel
-    pten::ConjKernel<T, DeviceContext>(dev_ctx, *pt_x.get(), pt_out.get());
+    pten::ConjKernel<T>(dev_ctx, *pt_x.get(), pt_out.get());
   }
 };
 

paddle/fluid/operators/dot_op.h

@@ -21,7 +21,6 @@
 
 // only can include the headers in paddle/pten/api dirs
 #include "paddle/pten/api/lib/utils/tensor_utils.h"
-#include "paddle/pten/include/core.h"
 #include "paddle/pten/kernels/dot_grad_kernel.h"
 #include "paddle/pten/kernels/dot_kernel.h"
 

paddle/fluid/operators/elementwise/elementwise_add_op.h

@@ -18,7 +18,6 @@ limitations under the License. */
 #include <utility>
 #include "paddle/fluid/operators/elementwise/elementwise_op.h"
 
-// only can include the headers in paddle/pten/include dirs
 #include "paddle/pten/kernels/math_kernel.h"
 
 namespace paddle {

paddle/fluid/operators/elementwise/elementwise_mul_op.h

@@ -18,7 +18,6 @@ limitations under the License. */
 #include "paddle/fluid/operators/elementwise/elementwise_op.h"
 #include "paddle/fluid/platform/cpu_info.h"
 
-// only can include the headers in paddle/pten/include dirs
 #include "paddle/pten/kernels/math_kernel.h"
 
 namespace paddle {

paddle/fluid/operators/elementwise/elementwise_op_function.h

@@ -29,7 +29,6 @@ limitations under the License. */
 #include "paddle/fluid/platform/device/gpu/gpu_info.h"
 #include "paddle/fluid/platform/transform.h"
 
-// only can include the headers in paddle/pten/include dirs
 #include "paddle/pten/api/lib/utils/tensor_utils.h"
 #include "paddle/pten/kernels/cpu/elementwise.h"
 

paddle/fluid/operators/elementwise/elementwise_op_impl.cu.h

@@ -22,7 +22,6 @@ limitations under the License. */
 
 // only can include the headers in paddle/top/api dirs
 #include "paddle/pten/api/lib/utils/tensor_utils.h"
-#include "paddle/pten/include/core.h"
 #include "paddle/pten/kernels/gpu/elementwise.h"
 
 namespace paddle {

paddle/fluid/operators/elementwise/elementwise_sub_op.h

@@ -16,7 +16,6 @@ limitations under the License. */
 
 #include "paddle/fluid/operators/elementwise/elementwise_op.h"
 
-// only can include the headers in paddle/pten/include dirs
 #include "paddle/pten/kernels/math_kernel.h"
 namespace paddle {
 namespace operators {

paddle/fluid/operators/fill_any_like_op.h

@@ -19,7 +19,6 @@ limitations under the License. */
 #include "paddle/fluid/framework/op_registry.h"
 #include "paddle/fluid/framework/pten_utils.h"
 
-#include "paddle/pten/include/core.h"
 #include "paddle/pten/kernels/full_kernel.h"
 
 namespace paddle {

paddle/fluid/operators/flatten_op.h

@@ -20,7 +20,6 @@ limitations under the License. */
 #include "paddle/fluid/operators/math/math_function.h"
 #include "paddle/fluid/operators/math/pooling.h"
 #include "paddle/fluid/platform/device_context.h"
-#include "paddle/pten/include/core.h"
 #include "paddle/pten/kernels/empty_kernel.h"
 #include "paddle/pten/kernels/flatten_grad_kernel.h"
 #include "paddle/pten/kernels/flatten_kernel.h"

paddle/fluid/operators/matmul_v2_op.h

@@ -27,7 +27,6 @@ limitations under the License. */
 
 // only can include the headers in paddle/pten/api dirs
 #include "paddle/pten/api/lib/utils/tensor_utils.h"
-#include "paddle/pten/include/core.h"
 #include "paddle/pten/kernels/matmul_grad_kernel.h"
 #include "paddle/pten/kernels/matmul_kernel.h"
 

paddle/fluid/operators/reduce_ops/reduce_op.h

@@ -26,8 +26,6 @@ limitations under the License. */
 
 // only can include the headers in paddle/pten/api dirs
 #include "paddle/pten/api/lib/utils/tensor_utils.h"
-#include "paddle/pten/include/core.h"
-#include "paddle/pten/include/math.h"
 #include "paddle/pten/kernels/cpu/reduce.h"
 
 #if defined(__HIPCC__) || defined(__NVCC__)

paddle/fluid/operators/reshape_op.cc

@@ -20,7 +20,6 @@ limitations under the License. */
 // only can include the headers in paddle/pten/api dirs
 #include "paddle/pten/api/lib/utils/tensor_utils.h"
 #include "paddle/pten/common/scalar_array.h"
-#include "paddle/pten/include/core.h"
 #include "paddle/pten/kernels/reshape_grad_kernel.h"
 #include "paddle/pten/kernels/reshape_kernel.h"
 namespace paddle {

paddle/fluid/operators/scale_op.h

@@ -19,7 +19,6 @@ limitations under the License. */
 
 // only can include the headers in paddle/top/api dirs
 #include "paddle/pten/api/lib/utils/tensor_utils.h"
-#include "paddle/pten/include/core.h"
 #include "paddle/pten/kernels/scale_kernel.h"
 
 namespace paddle {
@@ -70,7 +69,7 @@ class ScaleKernel : public framework::OpKernel<T> {
     auto pt_out = paddle::experimental::MakePtenDenseTensor(*out);
 
     // call new kernel
-    pten::Scale<T>(dev_ctx, *pt_x.get(), scale, bias, bias_after_scale,
+    pten::ScaleKernel<T>(dev_ctx, *pt_x.get(), scale, bias, bias_after_scale,
                          pt_out.get());
   }
 };

paddle/fluid/operators/sign_op.h

@@ -19,7 +19,6 @@ limitations under the License. */
 #include "paddle/fluid/framework/pten_utils.h"
 #include "paddle/fluid/operators/eigen/eigen_function.h"
 
-#include "paddle/pten/include/core.h"
 #include "paddle/pten/kernels/sign_kernel.h"
 
 namespace paddle {

paddle/fluid/pybind/eager.cc

@@ -26,7 +26,6 @@ limitations under the License. */
 #include "paddle/pten/common/data_type.h"
 #include "paddle/pten/core/convert_utils.h"
 #include "paddle/pten/core/dense_tensor.h"
-#include "paddle/pten/include/core.h"
 #include "pybind11/numpy.h"
 #include "pybind11/pybind11.h"
 #pragma GCC diagnostic ignored "-Wmissing-field-initializers"

paddle/fluid/pybind/eager_functions.cc

@@ -34,7 +34,6 @@ limitations under the License. */
 #include "paddle/pten/common/data_type.h"
 #include "paddle/pten/core/convert_utils.h"
 #include "paddle/pten/core/dense_tensor.h"
-#include "paddle/pten/include/core.h"
 
 namespace paddle {
 namespace pybind {

paddle/fluid/pybind/eager_method.cc

@@ -31,7 +31,6 @@ limitations under the License. */
 #include "paddle/pten/common/data_type.h"
 #include "paddle/pten/core/convert_utils.h"
 #include "paddle/pten/core/dense_tensor.h"
-#include "paddle/pten/include/core.h"
 namespace paddle {
 namespace pybind {
 

paddle/fluid/pybind/eager_properties.cc

@@ -28,7 +28,6 @@ limitations under the License. */
 #include "paddle/pten/common/data_type.h"
 #include "paddle/pten/core/convert_utils.h"
 #include "paddle/pten/core/dense_tensor.h"
-#include "paddle/pten/include/core.h"
 #pragma GCC diagnostic ignored "-Wwrite-strings"
 
 namespace paddle {

paddle/fluid/pybind/eager_utils.cc

@@ -26,7 +26,6 @@ limitations under the License. */
 #include "paddle/pten/common/data_type.h"
 #include "paddle/pten/core/convert_utils.h"
 #include "paddle/pten/core/dense_tensor.h"
-#include "paddle/pten/include/core.h"
 
 namespace paddle {
 namespace pybind {

paddle/pten/CMakeLists.txt

@@ -29,4 +29,4 @@ get_property(pten_kernels GLOBAL PROPERTY PTEN_KERNELS)
 message(STATUS "All standard pten kernels: ${pten_kernels}")
 set(PTEN_DEPS ${PTEN_DEPS} ${pten_kernels})
 
-cc_library(pten SRCS all.cc DEPS ${PTEN_DEPS})
+cc_library(pten DEPS ${PTEN_DEPS})

paddle/pten/all.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/pten/all.h"
-
-namespace pten {}  // namespace pten

paddle/pten/all.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
-
-// developer apis
-#include "paddle/pten/include/core.h"
-#include "paddle/pten/include/infermeta.h"
-#include "paddle/pten/include/math.h"

paddle/pten/api/lib/utils.cc

@@ -22,8 +22,7 @@ limitations under the License. */
 #include "paddle/pten/api/lib/kernel_dispatch.h"
 #include "paddle/pten/api/lib/utils/storage.h"
 #include "paddle/pten/core/kernel_registry.h"
-#include "paddle/pten/include/core.h"
-#include "paddle/pten/include/infermeta.h"
+#include "paddle/pten/infermeta/unary.h"
 
 PT_DECLARE_KERNEL(copy, CPU, ALL_LAYOUT);
 

paddle/pten/include/core.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
-
-// See Note: [ How do we organize the kernel directory ]
-#include "paddle/pten/core/convert_utils.h"
-#include "paddle/pten/core/dense_tensor.h"
-#include "paddle/pten/core/kernel_context.h"
-#include "paddle/pten/core/kernel_factory.h"
-#include "paddle/pten/core/tensor_meta.h"

paddle/pten/include/infermeta.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
-
-// See Note: [ How do we organize the kernel directory ]
-#include "paddle/pten/infermeta/binary.h"
-#include "paddle/pten/infermeta/multiary.h"
-#include "paddle/pten/infermeta/nullary.h"
-#include "paddle/pten/infermeta/unary.h"

paddle/pten/include/math.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
-
-// See Note: [ How do we organize the kernel directory ]
-#include "paddle/pten/api/lib/utils/storage.h"
-#include "paddle/pten/include/infermeta.h"
-#include "paddle/pten/kernels/scale_kernel.h"
-
-namespace pten {
-
-template <typename T, typename ContextT>
-DenseTensor Scale(const ContextT& dev_ctx,
-                  const DenseTensor& x,
-                  const Scalar& scale,
-                  float bias,
-                  bool bias_after_scale) {
-  auto out_meta = UnchangedInferMeta(x.meta());
-  pten::DenseTensor dense_out(
-      pten::make_intrusive<paddle::experimental::SharedStorage>(
-          dev_ctx.GetPlace()),
-      std::move(out_meta));
-  Scale<T, ContextT>(dev_ctx, x, scale, bias, bias_after_scale, &dense_out);
-  return dense_out;
-}
-
-}  // namespace pten

paddle/pten/kernels/complex_kernel.h

@@ -15,9 +15,6 @@ limitations under the License. */
 #pragma once
 
 #include "paddle/pten/core/dense_tensor.h"
-#include "paddle/pten/include/infermeta.h"
-#include "paddle/pten/kernels/empty_kernel.h"
-
 #include "paddle/pten/infermeta/unary.h"
 #include "paddle/pten/kernels/empty_kernel.h"
 

paddle/pten/kernels/cpu/scale_kernel.cc

@@ -13,18 +13,48 @@ See the License for the specific language governing permissions and
 limitations under the License. */
 
 #include "paddle/pten/kernels/scale_kernel.h"
-#include "paddle/pten/kernels/impl/scale_kernel_impl.h"
 
 #include "paddle/pten/backends/cpu/cpu_context.h"
+#include "paddle/pten/common/scalar.h"
+#include "paddle/pten/core/dense_tensor.h"
 #include "paddle/pten/core/kernel_registry.h"
+#include "paddle/pten/kernels/funcs/eigen/common.h"
 
 // See Note [ Why still include the fluid headers? ]
+#include "paddle/fluid/operators/eigen/eigen_function.h"
 #include "paddle/fluid/platform/bfloat16.h"
+namespace pten {
+
+template <typename T, typename Context>
+void ScaleKernel(const Context& dev_ctx,
+                 const DenseTensor& x,
+                 const Scalar& scale,
+                 float bias,
+                 bool bias_after_scale,
+                 DenseTensor* out) {
+  // calc
+  out->mutable_data<T>();
+  auto eigen_out = pten::EigenVector<T>::Flatten(*out);
+  auto eigen_x = pten::EigenVector<T>::Flatten(x);
+  auto& dev = *dev_ctx.eigen_device();
+  // TODO(chenweihang): now the eigen function here need the dtype of scale,
+  // eigen_x, bias should be same, so here need cast for two scalar arg,
+  // maybe we declare that the type of scale and bias is T?
+  paddle::operators::EigenScale<std::decay_t<decltype(dev)>, T>::Eval(
+      dev,
+      eigen_out,
+      eigen_x,
+      scale.to<T>(),
+      static_cast<T>(bias),
+      bias_after_scale);
+}
+
+}  // namespace pten
 
 PT_REGISTER_CTX_KERNEL(scale,
                        CPU,
                        ALL_LAYOUT,
-                       pten::Scale,
+                       pten::ScaleKernel,
                        float,
                        double,
                        paddle::platform::bfloat16,

paddle/pten/kernels/flatten_kernel.h

@@ -15,7 +15,7 @@ limitations under the License. */
 #pragma once
 
 #include "paddle/pten/core/dense_tensor.h"
-#include "paddle/pten/include/infermeta.h"
+#include "paddle/pten/infermeta/unary.h"
 #include "paddle/pten/kernels/empty_kernel.h"
 
 namespace pten {

paddle/pten/kernels/gpu/scale_kernel.cu

@@ -44,7 +44,7 @@ struct ScaleFunctor {
 };
 
 template <typename T, typename ContextT>
-void Scale(const ContextT& dev_ctx,
+void ScaleKernel(const ContextT& dev_ctx,
                  const DenseTensor& x,
                  const Scalar& scale,
                  float bias,
@@ -67,7 +67,7 @@ void Scale(const ContextT& dev_ctx,
 PT_REGISTER_CTX_KERNEL(scale,
                        GPU,
                        ALL_LAYOUT,
-                       pten::Scale,
+                       pten::ScaleKernel,
                        float,
                        double,
                        paddle::platform::float16,

paddle/pten/kernels/impl/matmul_grad_kernel_impl.h

@@ -14,8 +14,7 @@ limitations under the License. */
 
 #pragma once
 
-// #include "paddle/pten/kernels/complex_kernel.h"
-#include "paddle/pten/include/math.h"
+#include "paddle/pten/kernels/complex_kernel.h"
 #include "paddle/pten/kernels/empty_kernel.h"
 #include "paddle/pten/kernels/impl/dot_grad_kernel_impl.h"
 #include "paddle/pten/kernels/impl/matmul_kernel_impl.h"

paddle/pten/kernels/impl/scale_kernel_impl.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/pten/common/scalar.h"
-#include "paddle/pten/core/dense_tensor.h"
-#include "paddle/pten/kernels/funcs/eigen/common.h"
-
-// See Note [ Why still include the fluid headers? ]
-#include "paddle/fluid/operators/eigen/eigen_function.h"
-
-namespace pten {
-
-template <typename T, typename Context>
-void Scale(const Context& dev_ctx,
-           const DenseTensor& x,
-           const Scalar& scale,
-           float bias,
-           bool bias_after_scale,
-           DenseTensor* out) {
-  // calc
-  out->mutable_data<T>();
-  auto eigen_out = pten::EigenVector<T>::Flatten(*out);
-  auto eigen_x = pten::EigenVector<T>::Flatten(x);
-  auto& dev = *dev_ctx.eigen_device();
-  // TODO(chenweihang): now the eigen function here need the dtype of scale,
-  // eigen_x, bias should be same, so here need cast for two scalar arg,
-  // maybe we declare that the type of scale and bias is T?
-  paddle::operators::EigenScale<std::decay_t<decltype(dev)>, T>::Eval(
-      dev,
-      eigen_out,
-      eigen_x,
-      scale.to<T>(),
-      static_cast<T>(bias),
-      bias_after_scale);
-}
-
-}  // namespace pten

paddle/pten/kernels/math_kernel.h

@@ -16,7 +16,8 @@ limitations under the License. */
 
 #include "paddle/pten/api/lib/utils/storage.h"
 #include "paddle/pten/core/dense_tensor.h"
-#include "paddle/pten/include/infermeta.h"
+#include "paddle/pten/infermeta/binary.h"
+#include "paddle/pten/infermeta/unary.h"
 #include "paddle/pten/kernels/empty_kernel.h"
 
 namespace pten {

paddle/pten/kernels/reshape_kernel.h

@@ -16,7 +16,7 @@ limitations under the License. */
 
 #include "paddle/pten/common/scalar_array.h"
 #include "paddle/pten/core/dense_tensor.h"
-#include "paddle/pten/include/infermeta.h"
+#include "paddle/pten/infermeta/unary.h"
 #include "paddle/pten/kernels/empty_kernel.h"
 
 namespace pten {

paddle/pten/kernels/scale_kernel.h

@@ -16,15 +16,29 @@ limitations under the License. */
 
 #include "paddle/pten/common/scalar.h"
 #include "paddle/pten/core/dense_tensor.h"
-
+#include "paddle/pten/infermeta/unary.h"
+#include "paddle/pten/kernels/empty_kernel.h"
 namespace pten {
 
 template <typename T, typename Context>
-void Scale(const Context& dev_ctx,
+void ScaleKernel(const Context& dev_ctx,
                  const DenseTensor& x,
                  const Scalar& scale,
                  float bias,
                  bool bias_after_scale,
                  DenseTensor* out);
 
+template <typename T, typename ContextT>
+DenseTensor Scale(const ContextT& dev_ctx,
+                  const DenseTensor& x,
+                  const Scalar& scale,
+                  float bias,
+                  bool bias_after_scale) {
+  auto out_meta = UnchangedInferMeta(x.meta());
+  auto dense_out = pten::Empty<T, ContextT>(dev_ctx, std::move(out_meta));
+  ScaleKernel<T, ContextT>(
+      dev_ctx, x, scale, bias, bias_after_scale, &dense_out);
+  return dense_out;
+}
+
 }  // namespace pten

paddle/pten/kernels/sign_kernel.h

@@ -15,7 +15,7 @@ limitations under the License. */
 #pragma once
 
 #include "paddle/pten/core/dense_tensor.h"
-#include "paddle/pten/include/infermeta.h"
+#include "paddle/pten/infermeta/unary.h"
 #include "paddle/pten/kernels/empty_kernel.h"
 
 namespace pten {

paddle/pten/tests/api/scale_api.h

@@ -23,8 +23,7 @@
 #include "paddle/pten/common/scalar.h"
 #include "paddle/pten/common/scalar_array.h"
 #include "paddle/pten/core/kernel_registry.h"
-#include "paddle/pten/include/core.h"
-#include "paddle/pten/include/infermeta.h"
+#include "paddle/pten/infermeta/unary.h"
 #include "paddle/pten/kernels/scale_kernel.h"
 
 namespace paddle {
@@ -92,42 +91,42 @@ static void ScaleCPU(DataType kernel_dtype,
                      pten::DenseTensor* dense_out) {
   switch (kernel_dtype) {
     case pten::DataType::FLOAT64: {
-      pten::Scale<double>(
+      pten::ScaleKernel<double>(
           dev_ctx, x, pten::Scalar(scale), bias, bias_after_scale, dense_out);
       break;
     }
     case pten::DataType::FLOAT32: {
-      pten::Scale<float>(
+      pten::ScaleKernel<float>(
           dev_ctx, x, pten::Scalar(scale), bias, bias_after_scale, dense_out);
       break;
     }
     case pten::DataType::BFLOAT16: {
-      pten::Scale<paddle::platform::bfloat16>(
+      pten::ScaleKernel<paddle::platform::bfloat16>(
           dev_ctx, x, pten::Scalar(scale), bias, bias_after_scale, dense_out);
       break;
     }
     case pten::DataType::INT64: {
-      pten::Scale<int64_t>(
+      pten::ScaleKernel<int64_t>(
           dev_ctx, x, pten::Scalar(scale), bias, bias_after_scale, dense_out);
       break;
     }
     case pten::DataType::INT32: {
-      pten::Scale<int32_t>(
+      pten::ScaleKernel<int32_t>(
           dev_ctx, x, pten::Scalar(scale), bias, bias_after_scale, dense_out);
       break;
     }
     case pten::DataType::INT16: {
-      pten::Scale<int16_t>(
+      pten::ScaleKernel<int16_t>(
           dev_ctx, x, pten::Scalar(scale), bias, bias_after_scale, dense_out);
       break;
     }
     case pten::DataType::INT8: {
-      pten::Scale<int8_t>(
+      pten::ScaleKernel<int8_t>(
           dev_ctx, x, pten::Scalar(scale), bias, bias_after_scale, dense_out);
       break;
     }
     case pten::DataType::UINT8: {
-      pten::Scale<uint8_t>(
+      pten::ScaleKernel<uint8_t>(
           dev_ctx, x, pten::Scalar(scale), bias, bias_after_scale, dense_out);
       break;
     }
@@ -151,42 +150,42 @@ static void ScaleGPU(DataType kernel_dtype,
                      pten::DenseTensor* dense_out) {
   switch (kernel_dtype) {
     case pten::DataType::FLOAT64: {
-      pten::Scale<double>(
+      pten::ScaleKernel<double>(
           dev_ctx, x, pten::Scalar(scale), bias, bias_after_scale, dense_out);
       break;
     }
     case pten::DataType::FLOAT32: {
-      pten::Scale<float>(
+      pten::ScaleKernel<float>(
           dev_ctx, x, pten::Scalar(scale), bias, bias_after_scale, dense_out);
       break;
     }
     case pten::DataType::FLOAT16: {
-      pten::Scale<paddle::platform::float16>(
+      pten::ScaleKernel<paddle::platform::float16>(
           dev_ctx, x, pten::Scalar(scale), bias, bias_after_scale, dense_out);
       break;
     }
     case pten::DataType::INT64: {
-      pten::Scale<int64_t>(
+      pten::ScaleKernel<int64_t>(
           dev_ctx, x, pten::Scalar(scale), bias, bias_after_scale, dense_out);
       break;
     }
     case pten::DataType::INT32: {
-      pten::Scale<int32_t>(
+      pten::ScaleKernel<int32_t>(
           dev_ctx, x, pten::Scalar(scale), bias, bias_after_scale, dense_out);
       break;
     }
     case pten::DataType::INT16: {
-      pten::Scale<int16_t>(
+      pten::ScaleKernel<int16_t>(
           dev_ctx, x, pten::Scalar(scale), bias, bias_after_scale, dense_out);
       break;
     }
     case pten::DataType::INT8: {
-      pten::Scale<int8_t>(
+      pten::ScaleKernel<int8_t>(
           dev_ctx, x, pten::Scalar(scale), bias, bias_after_scale, dense_out);
       break;
     }
     case pten::DataType::UINT8: {
-      pten::Scale<uint8_t>(
+      pten::ScaleKernel<uint8_t>(
           dev_ctx, x, pten::Scalar(scale), bias, bias_after_scale, dense_out);
       break;
     }

paddle/pten/tests/kernels/test_scale_dev_api.cc

@@ -15,7 +15,7 @@ limitations under the License. */
 #include <gtest/gtest.h>
 #include <memory>
 
-#include "paddle/pten/include/math.h"
+#include "paddle/pten/kernels/scale_kernel.h"
 
 #include "paddle/pten/api/lib/utils/allocator.h"
 #include "paddle/pten/core/dense_tensor.h"

python/paddle/utils/code_gen/api_gen.py

@@ -345,8 +345,10 @@ def source_include(header_file_path):
 #include "paddle/pten/api/lib/kernel_dispatch.h"
 #include "paddle/pten/api/lib/utils/storage.h"
 #include "paddle/pten/core/kernel_registry.h"
-#include "paddle/pten/include/core.h"
-#include "paddle/pten/include/infermeta.h"
+#include "paddle/pten/infermeta/binary.h"
+#include "paddle/pten/infermeta/multiary.h"
+#include "paddle/pten/infermeta/nullary.h"
+#include "paddle/pten/infermeta/unary.h"
 #include "paddle/pten/kernels/declarations.h"
 """