Add reduce functions for FDTensor

.github/PULL_REQUEST_TEMPLATE.md

@@ -0,0 +1,9 @@
+<!-- Demo: https://github.com/PaddlePaddle/Paddle/pull/24810 -->
+### PR types
+<!-- One of [ New features | Bug fixes | Function optimization | Performance optimization | Breaking changes | Others ] -->
+
+### PR changes
+<!-- One of [ OPs | APIs | Docs | Others ] -->
+
+### Describe
+<!-- Describe what this PR does -->

CMakeLists.txt

@@ -47,6 +47,9 @@ option(TRT_DIRECTORY "If build tensorrt backend, need to define path of tensorrt
 option(ENABLE_VISION "Whether to enable vision models usage." OFF)
 option(ENABLE_VISION_VISUALIZE "Whether to enable visualize vision model result toolbox." ON)
 option(ENABLE_TEXT "Whether to enable text models usage." OFF)
+option(WITH_TESTING "Whether to compile with unittest." OFF)
+# TODO(zhoushunjie): Will remove it later.
+option(ENABLE_FDTENSOR_FUNC "Whether to compile with function of FDTensor." OFF)
 
 # Please don't open this flag now, some bugs exists.
 option(ENABLE_OPENCV_CUDA "Whether to enable opencv with cuda, this will allow process image with GPU." OFF)
@@ -56,6 +59,12 @@ option(ENABLE_DEBUG "Whether to enable print debug information, this may reduce
 option(WITH_VISION_EXAMPLES "Whether to build fastdeply with vision examples" OFF)
 option(WITH_TEXT_EXAMPLES "Whether to build fastdeply with text examples" OFF)
 
+# config GIT_URL with github mirrors to speed up dependent repos clone
+option(GIT_URL "Git URL to clone dependent repos" ${GIT_URL})
+if(NOT GIT_URL)
+    set(GIT_URL "https://github.com")
+endif()
+
 # Check for 32bit system
 if(WIN32)
   if(NOT CMAKE_CL_64)
@@ -108,13 +117,14 @@ endif()
 
 add_definitions(-DFASTDEPLOY_LIB)
 file(GLOB_RECURSE ALL_DEPLOY_SRCS ${PROJECT_SOURCE_DIR}/${CSRCS_DIR_NAME}/fastdeploy/*.cc)
+file(GLOB_RECURSE FDTENSOR_FUNC_SRCS ${PROJECT_SOURCE_DIR}/${CSRCS_DIR_NAME}/fastdeploy/function/*.cc)
 file(GLOB_RECURSE DEPLOY_ORT_SRCS ${PROJECT_SOURCE_DIR}/${CSRCS_DIR_NAME}/fastdeploy/backends/ort/*.cc)
 file(GLOB_RECURSE DEPLOY_PADDLE_SRCS ${PROJECT_SOURCE_DIR}/${CSRCS_DIR_NAME}/fastdeploy/backends/paddle/*.cc)
 file(GLOB_RECURSE DEPLOY_TRT_SRCS ${PROJECT_SOURCE_DIR}/${CSRCS_DIR_NAME}/fastdeploy/backends/tensorrt/*.cc ${PROJECT_SOURCE_DIR}/${CSRCS_DIR_NAME}/fastdeploy/backends/tensorrt/*.cpp)
 file(GLOB_RECURSE DEPLOY_VISION_SRCS ${PROJECT_SOURCE_DIR}/${CSRCS_DIR_NAME}/fastdeploy/vision/*.cc)
 file(GLOB_RECURSE DEPLOY_TEXT_SRCS ${PROJECT_SOURCE_DIR}/${CSRCS_DIR_NAME}/fastdeploy/text/*.cc)
 file(GLOB_RECURSE DEPLOY_PYBIND_SRCS ${PROJECT_SOURCE_DIR}/${CSRCS_DIR_NAME}/fastdeploy/pybind/*.cc ${PROJECT_SOURCE_DIR}/${CSRCS_DIR_NAME}/fastdeploy/*_pybind.cc)
-list(REMOVE_ITEM ALL_DEPLOY_SRCS ${DEPLOY_ORT_SRCS} ${DEPLOY_PADDLE_SRCS} ${DEPLOY_TRT_SRCS} ${DEPLOY_VISION_SRCS} ${DEPLOY_TEXT_SRCS})
+list(REMOVE_ITEM ALL_DEPLOY_SRCS ${DEPLOY_ORT_SRCS} ${DEPLOY_PADDLE_SRCS} ${DEPLOY_TRT_SRCS} ${DEPLOY_VISION_SRCS} ${DEPLOY_TEXT_SRCS} ${FDTENSOR_FUNC_SRCS})
 
 set(DEPEND_LIBS "")
 
@@ -223,6 +233,11 @@ if(ENABLE_TEXT)
   include(external/faster_tokenizer.cmake)
 endif()
 
+if (ENABLE_FDTENSOR_FUNC)
+  add_definitions(-DENABLE_FDTENSOR_FUNC)
+  list(APPEND ALL_DEPLOY_SRCS ${FDTENSOR_FUNC_SRCS})
+endif()
+
 configure_file(${PROJECT_SOURCE_DIR}/${CSRCS_DIR_NAME}/fastdeploy/core/config.h.in ${PROJECT_SOURCE_DIR}/${CSRCS_DIR_NAME}/fastdeploy/core/config.h)
 configure_file(${PROJECT_SOURCE_DIR}/${CSRCS_DIR_NAME}/fastdeploy/pybind/main.cc.in ${PROJECT_SOURCE_DIR}/${CSRCS_DIR_NAME}/fastdeploy/pybind/main.cc)
 configure_file(${PROJECT_SOURCE_DIR}/FastDeploy.cmake.in ${PROJECT_SOURCE_DIR}/FastDeploy.cmake @ONLY)
@@ -231,6 +246,8 @@ configure_file(${PROJECT_SOURCE_DIR}/fastdeploy/c_lib_wrap.py.in ${PROJECT_SOURC
 list(REMOVE_ITEM ALL_DEPLOY_SRCS ${DEPLOY_PYBIND_SRCS})
 
 add_library(${LIBRARY_NAME} SHARED ${ALL_DEPLOY_SRCS})
+add_dependencies(${LIBRARY_NAME} extern_eigen3)
+
 redefine_file_macro(${LIBRARY_NAME})
 set_target_properties(${LIBRARY_NAME} PROPERTIES COMPILE_FLAGS "-fvisibility=hidden")
 if(NOT APPLE)
@@ -276,6 +293,14 @@ if (WITH_TEXT_EXAMPLES AND EXISTS ${PROJECT_SOURCE_DIR}/examples)
   endif()
 endif()
 
+if (WITH_TESTING AND EXISTS ${PROJECT_SOURCE_DIR}/tests)
+  add_definitions(-DWITH_TESTING)
+  include(external/gtest.cmake)
+  include(external/gflags.cmake)
+  include(external/glog.cmake)
+  add_subdirectory(tests)
+endif()
+
 include(external/summary.cmake)
 fastdeploy_summary()
 if(WIN32)

csrcs/fastdeploy/function/eigen.cc

@@ -0,0 +1,32 @@
+// 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 "fastdeploy/function/eigen.h"
+
+namespace fastdeploy {
+
+std::shared_ptr<EigenDeviceWrapper> EigenDeviceWrapper::instance_ = nullptr;
+
+std::shared_ptr<EigenDeviceWrapper> EigenDeviceWrapper::GetInstance() {
+  if (instance_ == nullptr) {
+    instance_ = std::make_shared<EigenDeviceWrapper>();
+  }
+  return instance_;
+}
+
+const Eigen::DefaultDevice* EigenDeviceWrapper::GetDevice() const {
+  return &device_;
+}
+
+}  // namespace fastdeploy

csrcs/fastdeploy/function/eigen.h

@@ -0,0 +1,109 @@
+// 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 <algorithm>
+#include <memory>
+#include <vector>
+#include "fastdeploy/core/fd_tensor.h"
+#include "unsupported/Eigen/CXX11/Tensor"
+
+namespace fastdeploy {
+// EigenDim converts shape into Eigen::DSizes.
+template <int D>
+struct EigenDim {
+  using Type = Eigen::DSizes<Eigen::DenseIndex, D>;
+
+  static Type From(const std::vector<int64_t>& dims) {
+    Type ret;
+    for (int64_t d = 0; d < dims.size(); d++) {
+      ret[d] = dims[d];
+    }
+    return ret;
+  }
+};
+
+// Interpret FDTensor as EigenTensor and EigenConstTensor.
+template <typename T, size_t D, int MajorType = Eigen::RowMajor,
+          typename IndexType = Eigen::DenseIndex>
+struct EigenTensor {
+  using Type = Eigen::TensorMap<Eigen::Tensor<T, D, MajorType, IndexType>>;
+
+  using ConstType =
+      Eigen::TensorMap<Eigen::Tensor<const T, D, MajorType, IndexType>>;
+
+  static Type From(FDTensor& tensor,
+                   const std::vector<int64_t>& dims) {  // NOLINT
+    return Type(reinterpret_cast<T*>(tensor.Data()), EigenDim<D>::From(dims));
+  }
+
+  static Type From(FDTensor& tensor) {  // NOLINT
+    return From(tensor, tensor.shape);
+  }  // NOLINT
+
+  static ConstType From(const FDTensor& tensor,
+                        const std::vector<int64_t>& dims) {
+    return ConstType(reinterpret_cast<const T*>(tensor.Data()),
+                     EigenDim<D>::From(dims));
+  }
+
+  static ConstType From(const FDTensor& tensor) {
+    return From(tensor, tensor.shape);
+  }
+};
+
+template <typename T, int MajorType = Eigen::RowMajor,
+          typename IndexType = Eigen::DenseIndex>
+struct EigenScalar {
+  // Scalar tensor (implemented as a rank-0 tensor) of scalar type T.
+  using Type = Eigen::TensorMap<
+      Eigen::TensorFixedSize<T, Eigen::Sizes<>, MajorType, IndexType>>;
+  using ConstType = Eigen::TensorMap<
+      Eigen::TensorFixedSize<const T, Eigen::Sizes<>, MajorType, IndexType>>;
+
+  static Type From(FDTensor& tensor) {
+    return Type(reinterpret_cast<T*>(tensor.Data()));
+  }  // NOLINT
+
+  static ConstType From(const FDTensor& tensor) {
+    return ConstType(reinterpret_cast<const T*>(tensor.Data()));
+  }
+};
+
+template <typename T, int MajorType = Eigen::RowMajor,
+          typename IndexType = Eigen::DenseIndex>
+struct EigenVector : public EigenTensor<T, 1, MajorType, IndexType> {
+  // Flatten reshapes a Tensor into an EigenVector.
+  static typename EigenVector::Type Flatten(FDTensor& tensor) {  // NOLINT
+    return EigenVector::From(tensor, {tensor.Numel()});
+  }
+
+  static typename EigenVector::ConstType Flatten(
+      const FDTensor& tensor) {  // NOLINT
+    return EigenVector::From(tensor, {tensor.Numel()});
+  }
+};
+
+class EigenDeviceWrapper {
+ public:
+  static std::shared_ptr<EigenDeviceWrapper> GetInstance();
+  const Eigen::DefaultDevice* GetDevice() const;
+
+ private:
+  Eigen::DefaultDevice device_;
+  static std::shared_ptr<EigenDeviceWrapper> instance_;
+};
+
+}  // namespace fastdeploy