PaddlePaddle · chenwhql · Nov 3, 2021 · Nov 2, 2021
@@ -40,7 +40,7 @@ limitations under the License. */
 #include "paddle/fluid/platform/variant.h"
 #include "paddle/utils/flat_hash_map.h"
 
-#include "paddle/pten/api/include/core.h"
+#include "paddle/pten/include/core.h"
 
 namespace paddle {
 namespace framework {

diff --git a/paddle/fluid/framework/pten_utils.h b/paddle/fluid/framework/pten_utils.h
@@ -24,8 +24,8 @@ limitations under the License. */
 #include "paddle/fluid/imperative/type_defs.h"
 #include "paddle/fluid/platform/macros.h"
 #include "paddle/fluid/platform/place.h"
-#include "paddle/pten/api/include/core.h"
-#include "paddle/pten/hapi/lib/utils/tensor_utils.h"
+#include "paddle/pten/api/lib/utils/tensor_utils.h"
+#include "paddle/pten/include/core.h"
 #include "paddle/utils/flat_hash_map.h"
 #include "paddle/utils/small_vector.h"
 

diff --git a/paddle/fluid/imperative/prepared_operator.h b/paddle/fluid/imperative/prepared_operator.h
@@ -27,7 +27,7 @@
 #include "paddle/fluid/imperative/layer.h"
 #include "paddle/fluid/imperative/type_defs.h"
 
-#include "paddle/pten/api/include/core.h"
+#include "paddle/pten/include/core.h"
 
 DECLARE_bool(use_mkldnn);
 

diff --git a/paddle/fluid/operators/dot_op.h b/paddle/fluid/operators/dot_op.h
@@ -20,9 +20,9 @@
 #include "paddle/fluid/platform/for_range.h"
 
 // only can include the headers in paddle/pten/api dirs
-#include "paddle/pten/api/include/core.h"
-#include "paddle/pten/api/include/linalg.h"
-#include "paddle/pten/hapi/lib/utils/tensor_utils.h"
+#include "paddle/pten/api/lib/utils/tensor_utils.h"
+#include "paddle/pten/include/core.h"
+#include "paddle/pten/include/linalg.h"
 
 namespace paddle {
 namespace operators {

diff --git a/paddle/fluid/operators/fill_any_like_op.h b/paddle/fluid/operators/fill_any_like_op.h
@@ -19,8 +19,8 @@ limitations under the License. */
 #include "paddle/fluid/framework/op_registry.h"
 #include "paddle/fluid/framework/pten_utils.h"
 
-#include "paddle/pten/api/include/core.h"
-#include "paddle/pten/api/include/creation.h"
+#include "paddle/pten/include/core.h"
+#include "paddle/pten/include/creation.h"
 
 namespace paddle {
 namespace operators {

diff --git a/paddle/fluid/operators/matmul_v2_op.h b/paddle/fluid/operators/matmul_v2_op.h
@@ -26,9 +26,9 @@ limitations under the License. */
 #include "paddle/fluid/operators/reduce_ops/reduce_sum_op.h"
 
 // only can include the headers in paddle/pten/api dirs
-#include "paddle/pten/api/include/core.h"
-#include "paddle/pten/api/include/linalg.h"
-#include "paddle/pten/hapi/lib/utils/tensor_utils.h"
+#include "paddle/pten/api/lib/utils/tensor_utils.h"
+#include "paddle/pten/include/core.h"
+#include "paddle/pten/include/linalg.h"
 
 #if defined(__NVCC__) || defined(__HIPCC__)
 #include "paddle/fluid/operators/reduce_ops/cub_reduce.h"

diff --git a/paddle/fluid/operators/mean_op.h b/paddle/fluid/operators/mean_op.h
@@ -18,9 +18,9 @@ limitations under the License. */
 #include "paddle/fluid/framework/pten_utils.h"
 
 // only can include the headers in paddle/top/api dirs
-#include "paddle/pten/api/include/core.h"
-#include "paddle/pten/api/include/math.h"
-#include "paddle/pten/hapi/lib/utils/tensor_utils.h"
+#include "paddle/pten/api/lib/utils/tensor_utils.h"
+#include "paddle/pten/include/core.h"
+#include "paddle/pten/include/math.h"
 
 namespace paddle {
 namespace operators {

diff --git a/paddle/fluid/operators/scale_op.h b/paddle/fluid/operators/scale_op.h
@@ -18,9 +18,9 @@ limitations under the License. */
 #include "paddle/fluid/framework/pten_utils.h"
 
 // only can include the headers in paddle/top/api dirs
-#include "paddle/pten/api/include/core.h"
-#include "paddle/pten/api/include/math.h"
-#include "paddle/pten/hapi/lib/utils/tensor_utils.h"
+#include "paddle/pten/api/lib/utils/tensor_utils.h"
+#include "paddle/pten/include/core.h"
+#include "paddle/pten/include/math.h"
 
 namespace paddle {
 namespace operators {

diff --git a/paddle/fluid/operators/sign_op.h b/paddle/fluid/operators/sign_op.h
@@ -20,8 +20,8 @@ limitations under the License. */
 #include "paddle/fluid/operators/eigen/eigen_function.h"
 
 // only can include the headers in paddle/pten/api dirs
-#include "paddle/pten/api/include/core.h"
-#include "paddle/pten/api/include/math.h"
+#include "paddle/pten/include/core.h"
+#include "paddle/pten/include/math.h"
 
 namespace paddle {
 namespace operators {

diff --git a/paddle/pten/CMakeLists.txt b/paddle/pten/CMakeLists.txt
@@ -1,7 +1,6 @@
-# pten api
+# pten (low level) api headers: include
+# pten (high level) api
 add_subdirectory(api)
-# pten high level api
-add_subdirectory(hapi)
 # pten core components
 add_subdirectory(core)
 # pten kernels for diff device
@@ -10,3 +9,13 @@ add_subdirectory(kernels)
 add_subdirectory(infershape)
 # pten tests
 add_subdirectory(tests)
+
+# make an unity target for compile deps
+set(PTEN_DEPS convert_utils dense_tensor kernel_factory kernel_context)
+set(PTEN_DEPS ${PTEN_DEPS} math_cpu linalg_cpu creation_cpu manipulation_cpu)
+set(PTEN_DEPS ${PTEN_DEPS} unary binary)
+if(WITH_GPU OR WITH_ROCM)
+  set(PTEN_DEPS ${PTEN_DEPS} math_cuda linalg_cuda creation_cuda manipulation_cuda)
+endif()
+
+cc_library(pten SRCS all.cc DEPS ${PTEN_DEPS})
diff --git a/paddle/pten/hapi/all.cc → paddle/pten/all.cc b/paddle/pten/hapi/all.cc → paddle/pten/all.cc
@@ -12,8 +12,6 @@ WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
 See the License for the specific language governing permissions and
 limitations under the License. */
 
-#include "paddle/pten/hapi/all.h"
+#include "paddle/pten/all.h"
 
-namespace paddle {
-namespace experimental {}  // namespace experimental
-}  // namespace paddle
+namespace pten {}  // namespace pten
diff --git a/paddle/pten/hapi/all.h → paddle/pten/all.h b/paddle/pten/hapi/all.h → paddle/pten/all.h
@@ -14,9 +14,10 @@ limitations under the License. */
 
 #pragma once
 
-// user apis
-#include "paddle/pten/hapi/include/creation.h"
-#include "paddle/pten/hapi/include/linalg.h"
-#include "paddle/pten/hapi/include/manipulation.h"
-#include "paddle/pten/hapi/include/math.h"
-#include "paddle/pten/hapi/include/tensor.h"
+// develop apis
+#include "paddle/pten/include/core.h"
+#include "paddle/pten/include/creation.h"
+#include "paddle/pten/include/infershape.h"
+#include "paddle/pten/include/linalg.h"
+#include "paddle/pten/include/manipulation.h"
+#include "paddle/pten/include/math.h"
diff --git a/paddle/pten/api/CMakeLists.txt b/paddle/pten/api/CMakeLists.txt
@@ -1,8 +1,3 @@
-set(PTEN_DEPS convert_utils dense_tensor kernel_factory kernel_context)
-set(PTEN_DEPS ${PTEN_DEPS} math_cpu linalg_cpu creation_cpu manipulation_cpu)
-set(PTEN_DEPS ${PTEN_DEPS} unary binary)
-if(WITH_GPU OR WITH_ROCM)
-  set(PTEN_DEPS ${PTEN_DEPS} math_cuda linalg_cuda creation_cuda manipulation_cuda)
-endif()
+add_subdirectory(lib)
 
-cc_library(pten SRCS all.cc DEPS ${PTEN_DEPS})
+cc_library(pten_hapi SRCS all.cc DEPS linalg_api math_api creation_api)
diff --git a/paddle/pten/api/all.cc b/paddle/pten/api/all.cc
@@ -14,4 +14,6 @@ limitations under the License. */
 
 #include "paddle/pten/api/all.h"
 
-namespace pten {}  // namespace pten
+namespace paddle {
+namespace experimental {}  // namespace experimental
+}  // namespace paddle
diff --git a/paddle/pten/api/all.h b/paddle/pten/api/all.h
@@ -14,10 +14,9 @@ limitations under the License. */
 
 #pragma once
 
-// develop apis
-#include "paddle/pten/api/include/core.h"
+// user apis
 #include "paddle/pten/api/include/creation.h"
-#include "paddle/pten/api/include/infershape.h"
 #include "paddle/pten/api/include/linalg.h"
 #include "paddle/pten/api/include/manipulation.h"
 #include "paddle/pten/api/include/math.h"
+#include "paddle/pten/api/include/tensor.h"
diff --git a/paddle/pten/hapi/include/backend_set.h → paddle/pten/api/include/backend_set.h b/paddle/pten/hapi/include/backend_set.h → paddle/pten/api/include/backend_set.h
diff --git a/paddle/pten/api/include/creation.h b/paddle/pten/api/include/creation.h
@@ -14,26 +14,20 @@
 
 #pragma once
 
-#include "paddle/pten/api/include/infershape.h"
-#include "paddle/pten/hapi/lib/utils/allocator.h"
-#include "paddle/pten/kernels/cpu/creation.h"
-#include "paddle/pten/kernels/cuda/creation.h"
+#include "paddle/pten/api/include/tensor.h"
+#include "paddle/pten/common/data_type.h"
+#include "paddle/pten/common/scalar.h"
 
-namespace pten {
+namespace paddle {
+namespace experimental {
 
-// TODO(YuanRisheng) This function name should be same as User API name.
-// TODO(zyfncg) Automatic code generation
-template <typename T, typename ContextT>
-DenseTensor FillAnyLike(const ContextT& dev_ctx,
-                        const DenseTensor& x,
-                        const Scalar& val) {
-  auto out_meta = UnchangedInferShape(x.meta());
-  const auto allocator =
-      std::make_shared<paddle::experimental::DefaultAllocator>(
-          dev_ctx.GetPlace());
-  pten::DenseTensor dense_out(allocator, out_meta);
-  FillAnyLike<T>(dev_ctx, x, val, &dense_out);
-  return dense_out;
-}
+Tensor full_like(const Tensor& x,
+                 const Scalar& value,
+                 DataType dtype = DataType::UNDEFINED);
 
-}  // namespace pten
+Tensor ones_like(const Tensor& x, DataType dtype = DataType::UNDEFINED);
+
+Tensor zeros_like(const Tensor& x, DataType dtype = DataType::UNDEFINED);
+
+}  // namespace experimental
+}  // namespace paddle
diff --git a/paddle/pten/api/include/linalg.h b/paddle/pten/api/include/linalg.h
@@ -14,25 +14,17 @@
 
 #pragma once
 
-// See Note: [ How do we organize the kernel directory ]
-#include "paddle/pten/api/include/infershape.h"
-#include "paddle/pten/hapi/lib/utils/allocator.h"
-#include "paddle/pten/kernels/cpu/linalg.h"
-#include "paddle/pten/kernels/cuda/linalg.h"
+#include "paddle/pten/api/include/tensor.h"
 
-namespace pten {
+namespace paddle {
+namespace experimental {
 
-template <typename T, typename ContextT>
-DenseTensor Dot(const ContextT& dev_ctx,
-                const DenseTensor& x,
-                const DenseTensor& y) {
-  auto out_meta = DotInferShape(x.meta(), y.meta());
-  const auto allocator =
-      std::make_shared<paddle::experimental::DefaultAllocator>(
-          dev_ctx.GetPlace());
-  pten::DenseTensor dense_out(allocator, out_meta);
-  Dot<T>(dev_ctx, x, y, &dense_out);
-  return dense_out;
-}
+Tensor dot(const Tensor& x, const Tensor& y);
 
-}  // namespace pten
+Tensor matmul(const Tensor& x,
+              const Tensor& y,
+              bool transpose_x,
+              bool transpose_y);
+
+}  // namespace experimental
+}  // namespace paddle
diff --git a/paddle/pten/api/include/manipulation.h b/paddle/pten/api/include/manipulation.h
@@ -1,39 +1,25 @@
-// 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.
+/* 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/include/infershape.h"
-#include "paddle/pten/hapi/lib/utils/allocator.h"
-#include "paddle/pten/kernels/cpu/manipulation.h"
-#include "paddle/pten/kernels/cuda/manipulation.h"
-
-namespace pten {
-
-template <typename T, typename ContextT>
-DenseTensor Flatten(const ContextT& dev_ctx,
-                    const DenseTensor& x,
-                    int start_axis,
-                    int stop_axis) {
-  auto out_meta = FlattenInferShape(x.meta(), start_axis, stop_axis);
-  const auto allocator =
-      std::make_shared<paddle::experimental::DefaultAllocator>(
-          dev_ctx.GetPlace());
-  pten::DenseTensor dense_out(allocator, out_meta);
-  Flatten<T>(dev_ctx, x, start_axis, stop_axis, &dense_out);
-  return dense_out;
-}
-
-}  // namespace pten
+#include "paddle/pten/api/include/tensor.h"
+
+namespace paddle {
+namespace experimental {
+
+Tensor flatten(const Tensor& x, int start_axis, int stop_axis);
+
+}  // namespace experimental
+}  // namespace paddle
diff --git a/paddle/pten/api/include/math.h b/paddle/pten/api/include/math.h
@@ -14,63 +14,14 @@ limitations under the License. */
 
 #pragma once
 
-// See Note: [ How do we organize the kernel directory ]
-#include "paddle/pten/api/include/infershape.h"
-#include "paddle/pten/hapi/lib/utils/allocator.h"
-#include "paddle/pten/kernels/cpu/math.h"
-#include "paddle/pten/kernels/cuda/math.h"
+#include "paddle/pten/api/include/tensor.h"
 
-namespace pten {
+namespace paddle {
+namespace experimental {
 
-template <typename T, typename ContextT>
-DenseTensor Sign(const ContextT& dev_ctx, const DenseTensor& x) {
-  auto out_meta = UnchangedInferShape(x.meta());
-  const auto allocator =
-      std::make_shared<paddle::experimental::DefaultAllocator>(
-          dev_ctx.GetPlace());
-  pten::DenseTensor dense_out(allocator, out_meta);
-  Sign<T>(dev_ctx, x, &dense_out);
-  return dense_out;
-}
+// TODO(chenweihang): add scale API
+// TODO(chenweihang): move mean API into stat.h/cc
+Tensor mean(const Tensor& x);
 
-template <typename T, typename ContextT>
-DenseTensor Mean(const ContextT& dev_ctx, const DenseTensor& x) {
-  auto out_meta = ReductionInferShape(x.meta());
-  const auto allocator =
-      std::make_shared<paddle::experimental::DefaultAllocator>(
-          dev_ctx.GetPlace());
-  pten::DenseTensor dense_out(allocator, out_meta);
-  Mean<T>(dev_ctx, x, &dense_out);
-  return dense_out;
-}
-
-template <typename T, typename ContextT>
-DenseTensor Scale(const ContextT& dev_ctx,
-                  const DenseTensor& x,
-                  float scale,
-                  float bias,
-                  bool bias_after_scale) {
-  auto out_meta = UnchangedInferShape(x.meta());
-  const auto allocator =
-      std::make_shared<paddle::experimental::DefaultAllocator>(
-          dev_ctx.GetPlace());
-  pten::DenseTensor dense_out(allocator, out_meta);
-  Scale<T>(dev_ctx, x, scale, bias, bias_after_scale, &dense_out);
-  return dense_out;
-}
-
-template <typename T, typename ContextT>
-DenseTensor Scale(const ContextT& dev_ctx,
-                  const DenseTensor& x,
-                  const DenseTensor& scale,
-                  float bias,
-                  bool bias_after_scale) {
-  auto out_meta = UnchangedInferShape(x.meta());
-  const auto allocator =
-      std::make_shared<paddle::experimental::DefaultAllocator>(
-          dev_ctx.GetPlace());
-  pten::DenseTensor dense_out(allocator, out_meta);
-  ScaleHost<T>(dev_ctx, x, scale, bias, bias_after_scale, &dense_out);
-  return dense_out;
-}
-}  // namespace pten
+}  // namespace experimental
+}  // namespace paddle
diff --git a/paddle/pten/hapi/include/tensor.h → paddle/pten/api/include/tensor.h b/paddle/pten/hapi/include/tensor.h → paddle/pten/api/include/tensor.h
diff --git a/paddle/pten/hapi/lib/CMakeLists.txt → paddle/pten/api/lib/CMakeLists.txt b/paddle/pten/hapi/lib/CMakeLists.txt → paddle/pten/api/lib/CMakeLists.txt