OPENMP的使用

如何加到cmake里

include(FindOpenMP)
if(OPENMP_FOUND)
  set(CMAKE_C_FLAGS "${CMAKE_C_FLAGS} ${OpenMP_C_FLAGS}")
  set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} ${OpenMP_CXX_FLAGS}")
  set(CMAKE_EXE_LINKER_FLAGS "${CMAKE_EXE_LINKER_FLAGS} ${OpenMP_EXE_LINKER_FLAGS}")
else(OPENMP_FOUND)
  message("ERROR: OpenMP could not be found.")
endif(OPENMP_FOUND)

# NanoGICP
add_library(nano_gicp STATIC
)
target_link_libraries(nano_gicp ${PCL_LIBRARIES} ${OpenMP_LIBS} nanoflann)
target_include_directories(nano_gicp PUBLIC include ${PCL_INCLUDE_DIRS} ${EIGEN3_INCLUDE_DIR})

# Odometry Node
add_executable(dlo_odom_node src/dlo/odom_node.cc src/dlo/odom.cc)
add_dependencies(dlo_odom_node ${catkin_EXPORTED_TARGETS})
target_compile_options(dlo_odom_node PRIVATE ${OpenMP_FLAGS})
target_link_libraries(dlo_odom_node ${catkin_LIBRARIES} ${PCL_LIBRARIES} ${OpenMP_LIBS} Threads::Threads nano_gicp)

以下指令是啥意思

  std::vector<int> k_indices(1);
  std::vector<float> k_sq_dists(1);

#pragma omp parallel for num_threads(num_threads_) firstprivate(k_indices, k_sq_dists) schedule(guided, 8)
  for (int i = 0; i < input_->size(); i++) {
    PointTarget pt;
    pt.getVector4fMap() = trans_f * input_->at(i).getVector4fMap();

    target_kdtree_->nearestKSearch(pt, 1, k_indices, k_sq_dists);

  double sum_errors = 0.0;
  std::vector<Eigen::Matrix<double, 6, 6>, Eigen::aligned_allocator<Eigen::Matrix<double, 6, 6>>> Hs(num_threads_);
  std::vector<Eigen::Matrix<double, 6, 1>, Eigen::aligned_allocator<Eigen::Matrix<double, 6, 1>>> bs(num_threads_);
  for (int i = 0; i < num_threads_; i++) {
    Hs[i].setZero();
    bs[i].setZero();
  }

#pragma omp parallel for num_threads(num_threads_) reduction(+ : sum_errors) schedule(guided, 8)
  for (int i = 0; i < input_->size(); i++) {
    int target_index = correspondences_[i];
    if (target_index < 0) {
      continue;
    }

    const Eigen::Vector4d mean_A = input_->at(i).getVector4fMap().template cast<double>();
    const auto& cov_A = source_covs_[i];

    const Eigen::Vector4d mean_B = target_->at(target_index).getVector4fMap().template cast<double>();
    const auto& cov_B = target_covs_[target_index];

    const Eigen::Vector4d transed_mean_A = trans * mean_A;
    const Eigen::Vector4d error = mean_B - transed_mean_A;

    sum_errors += error.transpose() * mahalanobis_[i] * error;

#pragma omp parallel for num_threads(num_threads_) schedule(guided, 8)
  for (int i = 0; i < cloud->size(); i++) {
    std::vector<int> k_indices;
    std::vector<float> k_sq_distances;
    kdtree.nearestKSearch(cloud->at(i), k_correspondences_, k_indices, k_sq_distances);

20230905更新:

报错cout’ not specified in enclosing ‘parallel’

#pragma omp parallel for \
  default(none) \
  shared(output) \
  firstprivate(nn_indices, nn_dists) \
  num_threads(threads_) \
  schedule(dynamic, chunk_size_)

增加成

#pragma omp parallel for \
  default(none) \
  shared(output) \
  shared(std::cout) \
  firstprivate(nn_indices, nn_dists) \
  num_threads(threads_) \
  schedule(dynamic, chunk_size_)

 

OpenMP: ‘std::cout’ not specified in enclosing parallel_choubu1703的博客-CSDN博客