Add API and unit test for reshape (#37232)

* reshape kernel refactor

* fix compile bugs when run ci

* support xpu for reshape

* fix bugs when run unittest in kunlun ci

* fix compile bugs when run kunlun

* perfect code according to suggestion

* add api and unit test for reshape

paddle/fluid/framework/operator.cc

@@ -1884,9 +1884,14 @@ void OperatorWithKernel::BuildPtenKernelContext(
       } else if (attr_defs[i].type_index == std::type_index(typeid(bool))) {
         pt_kernel_context_->EmplaceBackAttr(BOOST_GET_CONST(bool, attr));
       } else if (attr_defs[i].type_index ==
-                 std::type_index(typeid(std::vector<int>))) {
-        pt_kernel_context_->EmplaceBackAttr(
-            BOOST_GET_CONST(std::vector<int>, attr));
+                     std::type_index(typeid(std::vector<int64_t>)) &&
+                 std::type_index(attr.type()) ==
+                     std::type_index(typeid(std::vector<int>))) {
+        // Emplace Back Attr according to the type of Pten_Kernel args.
+        const auto& vector_int_attr = BOOST_GET_CONST(std::vector<int>, attr);
+        const std::vector<int64_t> vector_int64_attr(vector_int_attr.begin(),
+                                                     vector_int_attr.end());
+        pt_kernel_context_->EmplaceBackAttr(vector_int64_attr);
       } else {
         PADDLE_THROW(platform::errors::Unimplemented(
             "unsupported cast op attribute `%s` when construct "

paddle/fluid/imperative/prepared_operator.cc

@@ -373,8 +373,14 @@ static void BuildDygraphPtenKernelContext(
       } else if (attr_defs[i].type_index == std::type_index(typeid(bool))) {
         kernel_ctx->EmplaceBackAttr(BOOST_GET_CONST(bool, attr));
       } else if (attr_defs[i].type_index ==
-                 std::type_index(typeid(std::vector<int>))) {
-        kernel_ctx->EmplaceBackAttr(BOOST_GET_CONST(std::vector<int>, attr));
+                     std::type_index(typeid(std::vector<int64_t>)) &&
+                 std::type_index(attr.type()) ==
+                     std::type_index(typeid(std::vector<int>))) {
+        // Emplace Back Attr according to the type of Pten_Kernel args.
+        const auto& vector_int_attr = BOOST_GET_CONST(std::vector<int>, attr);
+        const std::vector<int64_t> vector_int64_attr(vector_int_attr.begin(),
+                                                     vector_int_attr.end());
+        kernel_ctx->EmplaceBackAttr(vector_int64_attr);
       } else {
         PADDLE_THROW(platform::errors::Unimplemented(
             "unsupported cast op attribute `%s` when construct "

paddle/fluid/operators/reshape_op.cc

@@ -459,7 +459,9 @@ class ReshapeKernel {
       }
 #endif
     } else {
-      auto &shape_vec = ctx.Attr<std::vector<int>>("shape");
+      auto &shape_attr = ctx.Attr<std::vector<int>>("shape");
+      const std::vector<int64_t> shape_vec(shape_attr.begin(),
+                                           shape_attr.end());
       if (platform::is_cpu_place(ctx.GetPlace())) {
         auto &dev_ctx = ctx.device_context<platform::CPUDeviceContext>();
         pten::ReshapeFromVectorVal(dev_ctx, *pt_x.get(), shape_vec, pt_out);

paddle/pten/api/include/manipulation.h

@@ -21,5 +21,6 @@ namespace experimental {
 
 PD_DLL_DECL Tensor flatten(const Tensor& x, int start_axis, int stop_axis);
 
+PD_DLL_DECL Tensor reshape(const Tensor& x, const std::vector<int64_t>& shape);
 }  // namespace experimental
 }  // namespace paddle

paddle/pten/api/lib/manipulation.cc

@@ -59,6 +59,40 @@ PD_DLL_DECL Tensor flatten(const Tensor& x, int start_axis, int stop_axis) {
 
   return out;
 }
+
+PD_DLL_DECL Tensor reshape(const Tensor& x, const std::vector<int64_t>& shape) {
+  // 1. Get kernel signature and kernel
+  auto kernel_key_set = ParseKernelKeyByInputArgs(x);
+  auto kernel_key = kernel_key_set.GetHigestPriorityKernelKey();
+  auto kernel = pten::KernelFactory::Instance().SelectKernelOrThrowError(
+      "reshape2", kernel_key);
+
+  // 2. Get Device Context
+  auto* dev_ctx = GetDeviceContextByBackend(kernel_key.backend());
+  auto kernel_context = pten::KernelContext(dev_ctx);
+
+  // 3. Auto data transform
+  auto dense_x = std::dynamic_pointer_cast<pten::DenseTensor>(x.impl());
+  kernel_context.EmplaceBackInput(dense_x);
+  kernel_context.EmplaceBackAttr(shape);
+
+  // 4. InferShape
+  auto out_meta = InferShapeFromVecValue(dense_x->meta(), shape);
+
+  // 5. Prepare outputs
+  Tensor out;
+  const auto allocator = std::make_shared<DefaultAllocator>(
+      pten::TransToFluidPlace(kernel_key.backend()));
+  auto dense_out = std::make_shared<pten::DenseTensor>(allocator, out_meta);
+  kernel_context.EmplaceBackOutput(dense_out);
+  out.set_impl(dense_out);
+
+  // 6. Call kernel
+  kernel(&kernel_context);
+
+  return out;
+}
+
 }  // namespace experimental
 }  // namespace paddle
 

paddle/pten/core/kernel_utils.h

@@ -208,7 +208,6 @@ struct KernelImpl<Return (*)(Args...), kernel_fn> {
   PT_SPECIALIZE_KernelCallHelper_FOR_ATTRIBUTE(paddle::platform::float16);
   PT_SPECIALIZE_KernelCallHelper_FOR_ATTRIBUTE(const Scalar&);
   PT_SPECIALIZE_KernelCallHelper_FOR_ATTRIBUTE(const std::vector<int64_t>&);
-  PT_SPECIALIZE_KernelCallHelper_FOR_ATTRIBUTE(const std::vector<int>&);
 
   /* Output Helpers */
 

paddle/pten/include/manipulation.h

@@ -40,7 +40,7 @@ DenseTensor Flatten(const ContextT& dev_ctx,
 template <typename T, typename ContextT>
 DenseTensor Reshape(const ContextT& dev_ctx,
                     const DenseTensor& x,
-                    const std::vector<int>& shape) {
+                    const std::vector<int64_t>& shape) {
   auto out_meta = InferShapeFromVecValue(x.meta(), shape);
   const auto allocator =
       std::make_shared<paddle::experimental::DefaultAllocator>(

paddle/pten/infermeta/unary.cc

@@ -83,7 +83,7 @@ DenseTensorMeta FullLikeInferShape(const DenseTensorMeta& x_meta,
 }
 
 static paddle::framework::DDim ValidateShape(
-    const std::vector<int> shape, const paddle::framework::DDim& in_dims) {
+    const std::vector<int64_t> shape, const paddle::framework::DDim& in_dims) {
   const int64_t in_size = paddle::framework::product(in_dims);
   auto in_dims_vec = paddle::framework::vectorize(in_dims);
   bool all_positive = std::all_of(in_dims_vec.cbegin(),
@@ -203,7 +203,7 @@ static paddle::framework::DDim ValidateShape(
 }
 
 DenseTensorMeta InferShapeFromVecValue(const DenseTensorMeta& x_meta,
-                                       const std::vector<int>& shape) {
+                                       const std::vector<int64_t>& shape) {
   PADDLE_ENFORCE_EQ(!shape.empty(),
                     true,
                     paddle::platform::errors::InvalidArgument(

paddle/pten/infermeta/unary.h

@@ -46,5 +46,5 @@ DenseTensorMeta FullLikeInferShape(const DenseTensorMeta& x_meta,
                                    DataLayout layout);
 
 DenseTensorMeta InferShapeFromVecValue(const DenseTensorMeta& x_meta,
-                                       const std::vector<int>& shape);
+                                       const std::vector<int64_t>& shape);
 }  // namespace pten

paddle/pten/kernels/cpu/manipulation.cc

@@ -46,7 +46,7 @@ void FlattenWithXShape(const CPUContext& dev_ctx,
 
 void ReshapeFromVectorVal(const CPUContext& dev_ctx,
                           const DenseTensor& x,
-                          const std::vector<int>& shape,
+                          const std::vector<int64_t>& shape,
                           DenseTensor* out) {
   auto out_meta = InferShapeFromVecValue(x.meta(), shape);
   if (&x == out) {
@@ -59,7 +59,7 @@ void ReshapeFromVectorVal(const CPUContext& dev_ctx,
 
 void ReshapeFromVectorValWithXShape(const CPUContext& dev_ctx,
                                     const DenseTensor& x,
-                                    const std::vector<int>& shape,
+                                    const std::vector<int64_t>& shape,
                                     DenseTensor* xshape,
                                     DenseTensor* out) {
   ReshapeFromVectorVal(dev_ctx, x, shape, out);
@@ -71,8 +71,10 @@ void ReshapeFromDT(const CPUContext& dev_ctx,
                    const DenseTensor& shape,
                    DenseTensor* out) {
   auto* shape_data = shape.data<int>();
-  auto vector_shape = std::vector<int>(shape_data, shape_data + shape.numel());
+  auto vector_shape =
+      std::vector<int64_t>(shape_data, shape_data + shape.numel());
   ReshapeFromVectorVal(dev_ctx, x, vector_shape, out);
+  out->set_lod(x.lod());
 }
 
 void ReshapeFromDTWithXShape(const CPUContext& dev_ctx,
@@ -88,7 +90,7 @@ void ReshapeFromVectorDT(const CPUContext& dev_ctx,
                          const DenseTensor& x,
                          const std::vector<DenseTensor>& shape,
                          DenseTensor* out) {
-  std::vector<int> vector_shape;
+  std::vector<int64_t> vector_shape;
   for (auto& tensor : shape) {
     PADDLE_ENFORCE_EQ(
         tensor.dims(),

paddle/pten/kernels/cpu/manipulation.h

@@ -36,7 +36,7 @@ void ReshapeFromDT(const CPUContext& dev_ctx,
 
 void ReshapeFromVectorVal(const CPUContext& dev_ctx,
                           const DenseTensor& x,
-                          const std::vector<int>& shape,
+                          const std::vector<int64_t>& shape,
                           DenseTensor* out);
 
 void ReshapeFromVectorDT(const CPUContext& dev_ctx,
@@ -52,7 +52,7 @@ void ReshapeFromDTWithXShape(const CPUContext& dev_ctx,
 
 void ReshapeFromVectorValWithXShape(const CPUContext& dev_ctx,
                                     const DenseTensor& x,
-                                    const std::vector<int>& shape,
+                                    const std::vector<int64_t>& shape,
                                     DenseTensor* xshape,
                                     DenseTensor* out);
 

paddle/pten/kernels/cuda/manipulation.cu

@@ -46,7 +46,7 @@ void FlattenWithXShape(const CUDAContext& dev_ctx,
 
 void ReshapeFromVectorVal(const CUDAContext& dev_ctx,
                           const DenseTensor& x,
-                          const std::vector<int>& shape,
+                          const std::vector<int64_t>& shape,
                           DenseTensor* out) {
   auto out_meta = InferShapeFromVecValue(x.meta(), shape);
   if (&x == out) {
@@ -60,7 +60,7 @@ void ReshapeFromVectorVal(const CUDAContext& dev_ctx,
 
 void ReshapeFromVectorValWithXShape(const CUDAContext& dev_ctx,
                                     const DenseTensor& x,
-                                    const std::vector<int>& shape,
+                                    const std::vector<int64_t>& shape,
                                     DenseTensor* xshape,
                                     DenseTensor* out) {
   ReshapeFromVectorVal(dev_ctx, x, shape, out);
@@ -72,8 +72,10 @@ void ReshapeFromDT(const CUDAContext& dev_ctx,
                    const DenseTensor& shape,
                    DenseTensor* out) {
   auto* shape_data = shape.data<int>();
-  auto vector_shape = std::vector<int>(shape_data, shape_data + shape.numel());
+  auto vector_shape =
+      std::vector<int64_t>(shape_data, shape_data + shape.numel());
   ReshapeFromVectorVal(dev_ctx, x, vector_shape, out);
+  out->set_lod(x.lod());
 }
 
 void ReshapeFromDTWithXShape(const CUDAContext& dev_ctx,
@@ -89,7 +91,7 @@ void ReshapeFromVectorDT(const CUDAContext& dev_ctx,
                          const DenseTensor& x,
                          const std::vector<DenseTensor>& shape,
                          DenseTensor* out) {
-  std::vector<int> vector_shape;
+  std::vector<int64_t> vector_shape;
   for (auto& tensor : shape) {
     PADDLE_ENFORCE_EQ(
         tensor.dims(),

paddle/pten/kernels/cuda/manipulation.h

@@ -40,7 +40,7 @@ void ReshapeFromDT(const CUDAContext& dev_ctx,
 
 void ReshapeFromVectorVal(const CUDAContext& dev_ctx,
                           const DenseTensor& x,
-                          const std::vector<int>& shape,
+                          const std::vector<int64_t>& shape,
                           DenseTensor* out);
 
 void ReshapeFromVectorDT(const CUDAContext& dev_ctx,
@@ -56,7 +56,7 @@ void ReshapeFromDTWithXShape(const CUDAContext& dev_ctx,
 
 void ReshapeFromVectorValWithXShape(const CUDAContext& dev_ctx,
                                     const DenseTensor& x,
-                                    const std::vector<int>& shape,
+                                    const std::vector<int64_t>& shape,
                                     DenseTensor* xshape,
                                     DenseTensor* out);
 

paddle/pten/kernels/xpu/manipulation.cc

@@ -53,7 +53,7 @@ void FlattenWithXShape(const XPUContext& dev_ctx,
 
 void ReshapeFromVectorVal(const XPUContext& dev_ctx,
                           const DenseTensor& x,
-                          const std::vector<int>& shape,
+                          const std::vector<int64_t>& shape,
                           DenseTensor* out) {
   auto out_meta = InferShapeFromVecValue(x.meta(), shape);
   if (&x == out) {
@@ -69,7 +69,8 @@ void ReshapeFromDT(const XPUContext& dev_ctx,
                    const DenseTensor& shape,
                    DenseTensor* out) {
   auto* shape_data = shape.data<int>();
-  auto vector_shape = std::vector<int>(shape_data, shape_data + shape.numel());
+  auto vector_shape =
+      std::vector<int64_t>(shape_data, shape_data + shape.numel());
   ReshapeFromVectorVal(dev_ctx, x, vector_shape, out);
 }
 
@@ -77,7 +78,7 @@ void ReshapeFromVectorDT(const XPUContext& dev_ctx,
                          const DenseTensor& x,
                          const std::vector<DenseTensor>& shape,
                          DenseTensor* out) {
-  std::vector<int> vector_shape;
+  std::vector<int64_t> vector_shape;
   for (auto& tensor : shape) {
     PADDLE_ENFORCE_EQ(
         tensor.dims(),

paddle/pten/kernels/xpu/manipulation.h

@@ -40,7 +40,7 @@ void ReshapeFromDT(const XPUContext& dev_ctx,
 
 void ReshapeFromVectorVal(const XPUContext& dev_ctx,
                           const DenseTensor& x,
-                          const std::vector<int>& shape,
+                          const std::vector<int64_t>& shape,
                           DenseTensor* out);
 
 void ReshapeFromVectorDT(const XPUContext& dev_ctx,

paddle/pten/tests/api/CMakeLists.txt

@@ -14,3 +14,4 @@ cc_test(test_matmul_api SRCS test_matmul_api.cc DEPS pten_tensor pten_api pten_a
 cc_test(test_fill_api SRCS test_fill_api.cc DEPS pten_tensor pten_api pten_api_utils)
 cc_test(test_flatten_api SRCS test_flatten_api.cc DEPS pten_tensor pten_api pten_api_utils)
 cc_test(test_elementwise_api SRCS test_elementwise_api.cc DEPS pten_tensor pten_api pten_api_utils)
+cc_test(test_reshape_api SRCS test_reshape_api.cc DEPS pten_tensor pten_api pten_api_utils)

paddle/pten/tests/api/test_reshape_api.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 <gtest/gtest.h>
+#include <memory>
+
+#include "paddle/pten/api/include/manipulation.h"
+
+#include "paddle/pten/api/lib/utils/allocator.h"
+#include "paddle/pten/core/dense_tensor.h"
+#include "paddle/pten/core/kernel_registry.h"
+
+PT_DECLARE_MODULE(ManipulationCPU);
+
+#if defined(PADDLE_WITH_CUDA) || defined(PADDLE_WITH_HIP)
+PT_DECLARE_MODULE(ManipulationCUDA);
+#endif
+
+namespace framework = paddle::framework;
+using DDim = paddle::framework::DDim;
+
+// TODO(chenweihang): Remove this test after the API is used in the dygraph
+TEST(API, reshape) {
+  // 1. create tensor
+  const auto alloc = std::make_shared<paddle::experimental::DefaultAllocator>(
+      paddle::platform::CPUPlace());
+  auto dense_x = std::make_shared<pten::DenseTensor>(
+      alloc,
+      pten::DenseTensorMeta(pten::DataType::FLOAT32,
+                            framework::make_ddim({3, 2, 2, 3}),
+                            pten::DataLayout::NCHW));
+  auto* dense_x_data = dense_x->mutable_data<float>();
+
+  for (int i = 0; i < dense_x->numel(); i++) {
+    dense_x_data[i] = i;
+  }
+
+  paddle::experimental::Tensor x(dense_x);
+  std::vector<int64_t> shape{12, 3};
+  // 2. test API
+  auto out = paddle::experimental::reshape(x, shape);
+  // 3. check result
+  std::vector<int64_t> expect_shape = {12, 3};
+  ASSERT_EQ(out.shape()[0], expect_shape[0]);
+  ASSERT_EQ(out.shape()[1], expect_shape[1]);
+  ASSERT_EQ(out.numel(), 36);
+  ASSERT_EQ(out.is_cpu(), true);
+  ASSERT_EQ(out.type(), pten::DataType::FLOAT32);
+  ASSERT_EQ(out.layout(), pten::DataLayout::NCHW);
+  ASSERT_EQ(out.initialized(), true);
+  bool value_equal = true;
+  auto dense_out = std::dynamic_pointer_cast<pten::DenseTensor>(out.impl());
+  auto* dense_out_data = dense_out->data<float>();
+  for (int i = 0; i < dense_x->numel(); i++) {
+    if (std::abs(dense_x_data[i] - dense_out_data[i]) > 1e-6f)
+      value_equal = false;
+  }
+  ASSERT_EQ(value_equal, true);
+}

paddle/pten/tests/kernels/CMakeLists.txt

@@ -5,3 +5,4 @@ cc_test(test_flatten_dev_api SRCS test_flatten_dev_api.cc DEPS pten pten_api_uti
 cc_test(test_mean_dev_api SRCS test_mean_dev_api.cc DEPS pten pten_api_utils)
 cc_test(test_scale_dev_api SRCS test_scale_dev_api.cc DEPS pten pten_api_utils)
 cc_test(test_elementwise_dev_api SRCS test_elementwise_dev_api.cc DEPS pten pten_api_utils)
+cc_test(test_reshape_dev_api SRCS test_reshape_dev_api.cc DEPS pten pten_api_utils)

paddle/pten/tests/kernels/test_reshape_dev_api.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 <gtest/gtest.h>
+#include <memory>
+
+#include "paddle/pten/include/manipulation.h"
+
+#include "paddle/pten/api/lib/utils/allocator.h"
+#include "paddle/pten/core/dense_tensor.h"
+#include "paddle/pten/core/kernel_registry.h"
+
+namespace framework = paddle::framework;
+using DDim = paddle::framework::DDim;
+
+// TODO(chenweihang): Remove this test after the API is used in the dygraph
+TEST(DEV_API, reshape) {
+  // 1. create tensor
+  const auto alloc = std::make_shared<paddle::experimental::DefaultAllocator>(
+      paddle::platform::CPUPlace());
+  pten::DenseTensor dense_x(
+      alloc,
+      pten::DenseTensorMeta(pten::DataType::FLOAT32,
+                            framework::make_ddim({3, 2, 2, 3}),
+                            pten::DataLayout::NCHW));
+  auto* dense_x_data = dense_x.mutable_data<float>();
+
+  for (int i = 0; i < dense_x.numel(); i++) {
+    dense_x_data[i] = i;
+  }
+  paddle::platform::DeviceContextPool& pool =
+      paddle::platform::DeviceContextPool::Instance();
+  auto* dev_ctx = pool.Get(paddle::platform::CPUPlace());
+  std::vector<int64_t> shape{12, 3};
+
+  // 2. test API
+  auto out = pten::Reshape<float>(
+      *(static_cast<paddle::platform::CPUDeviceContext*>(dev_ctx)),
+      dense_x,
+      shape);
+  // 3. check result
+  std::vector<int64_t> expect_shape = {12, 3};
+  ASSERT_EQ(out.dims()[0], expect_shape[0]);
+  ASSERT_EQ(out.dims()[1], expect_shape[1]);
+  ASSERT_EQ(out.numel(), 36);
+  ASSERT_EQ(out.meta().type, pten::DataType::FLOAT32);
+  ASSERT_EQ(out.meta().layout, pten::DataLayout::NCHW);
+
+  bool value_equal = true;
+  auto* dense_out_data = out.data<float>();
+  for (int i = 0; i < dense_x.numel(); i++) {
+    if (std::abs(dense_x_data[i] - dense_out_data[i]) > 1e-6f)
+      value_equal = false;
+  }
+  ASSERT_EQ(value_equal, true);
+}