Windows (#40)

* build on windos

* WIN32 to MSVC

* CI

* fix CI

* verbose pytest

* cast to 64bit int

---------

Co-authored-by: tenteroring <tenteroring@tenteroring.org>

.github/workflows/build-windows.yml

@@ -0,0 +1,36 @@
+name: Build(Windows)
+
+on:
+  push:
+    branches: [ master ]
+    paths-ignore: '**.md'
+  pull_request:
+    branches: [ master ]
+    paths-ignore: '**.md'
+
+  # Allows you to run this workflow manually from the Actions tab
+  workflow_dispatch:
+
+jobs:
+  build:
+    runs-on: windows-latest
+
+    steps:
+      - uses: actions/checkout@v4
+
+      - uses: microsoft/setup-msbuild@v2
+
+      - uses: crazy-max/ghaction-chocolatey@v3
+        with:
+          args: install cmake eigen
+
+      - uses: actions/setup-python@v5
+        with:
+          python-version: '3.12'
+          cache: 'pip'
+      
+      - run: pip install numpy scipy pytest
+      
+      - run: pip install .
+
+      - run: python -m pytest src/test/python_test.py -v

CMakeLists.txt

@@ -82,6 +82,11 @@ if(BUILD_WITH_MARCH_NATIVE)
   add_compile_options(-march=native)
 endif()
 
+if(MSVC)
+  add_compile_definitions(_USE_MATH_DEFINES)
+  add_compile_options(/openmp:llvm)
+endif()
+
 ##############
 ## Coverage ##
 ##############
@@ -127,6 +132,9 @@ if(BUILD_PYTHON_BINDINGS)
   find_package(pybind11 CONFIG REQUIRED)
 
   pybind11_add_module(small_gicp_python
+    src/small_gicp/registration/registration.cpp
+    src/small_gicp/registration/registration_helper.cpp
+
     src/python/pointcloud.cpp
     src/python/kdtree.cpp
     src/python/voxelmap.cpp
@@ -138,8 +146,15 @@ if(BUILD_PYTHON_BINDINGS)
 
     src/python/python.cpp
   )
+  target_include_directories(small_gicp_python PUBLIC
+    $<BUILD_INTERFACE:${CMAKE_CURRENT_SOURCE_DIR}/include>
+    $<INSTALL_INTERFACE:include>
+  )
+  target_link_libraries(small_gicp_python PUBLIC
+    Eigen3::Eigen
+    $<TARGET_NAME_IF_EXISTS:OpenMP::OpenMP_CXX>
+  )
   set_target_properties(small_gicp_python PROPERTIES OUTPUT_NAME small_gicp)
-  target_link_libraries(small_gicp_python PRIVATE small_gicp)
 endif()
 
 ###############
@@ -293,7 +308,7 @@ if(BUILD_PYTHON_BINDINGS)
   elseif(NOT DEFINED PYTHON_INSTALL_DIR)
     set(PYTHON_INSTALL_DIR lib/python${Python_VERSION_MAJOR}.${Python_VERSION_MINOR}/site-packages)
   endif()
-  install(TARGETS small_gicp_python small_gicp
+  install(TARGETS small_gicp_python
     LIBRARY DESTINATION ${PYTHON_INSTALL_DIR}
     COMPONENT python
   )

include/small_gicp/ann/incremental_voxelmap.hpp

@@ -141,7 +141,7 @@ public:
 
   inline size_t calc_index(const size_t voxel_id, const size_t point_id) const { return (voxel_id << point_id_bits) | point_id; }
   inline size_t voxel_id(const size_t i) const { return i >> point_id_bits; }                ///< Extract the point ID from an index
-  inline size_t point_id(const size_t i) const { return i & ((1ul << point_id_bits) - 1); }  ///< Extract the voxel ID from an index
+  inline size_t point_id(const size_t i) const { return i & ((1ull << point_id_bits) - 1); }  ///< Extract the voxel ID from an index
 
 public:
   static_assert(sizeof(size_t) == 8, "size_t must be 64-bit");

include/small_gicp/benchmark/read_points.hpp

@@ -50,7 +50,7 @@ static void write_points(const std::string& filename, const std::vector<Eigen::V
 /// @param filename  Filename
 /// @return          Points
 static std::vector<Eigen::Vector4f> read_ply(const std::string& filename) {
-  std::ifstream ifs(filename);
+  std::ifstream ifs(filename, std::ios::binary);
   if (!ifs) {
     std::cerr << "error: failed to open " << filename << std::endl;
     return {};

include/small_gicp/registration/reduction_omp.hpp

@@ -30,7 +30,7 @@ struct ParallelReductionOMP {
     std::vector<double> es(num_threads, 0.0);
 
 #pragma omp parallel for num_threads(num_threads) schedule(guided, 8)
-    for (size_t i = 0; i < factors.size(); i++) {
+    for (std::int64_t i = 0; i < factors.size(); i++) {
       Eigen::Matrix<double, 6, 6> H;
       Eigen::Matrix<double, 6, 1> b;
       double e;
@@ -59,7 +59,7 @@ struct ParallelReductionOMP {
     double sum_e = 0.0;
 
 #pragma omp parallel for num_threads(num_threads) schedule(guided, 8) reduction(+ : sum_e)
-    for (size_t i = 0; i < factors.size(); i++) {
+    for (std::int64_t i = 0; i < factors.size(); i++) {
       sum_e += factors[i].error(target, source, T);
     }
     return sum_e;

include/small_gicp/util/downsampling.hpp

@@ -40,9 +40,9 @@ std::shared_ptr<OutputPointCloud> voxelgrid_sampling(const InputPointCloud& poin
 
     // Compute voxel coord bits (0|1bit, z|21bit, y|21bit, x|21bit)
     const std::uint64_t bits =                                 //
-      ((coord[0] & coord_bit_mask) << (coord_bit_size * 0)) |  //
-      ((coord[1] & coord_bit_mask) << (coord_bit_size * 1)) |  //
-      ((coord[2] & coord_bit_mask) << (coord_bit_size * 2));
+      (static_cast<std::uint64_t>(coord[0] & coord_bit_mask) << (coord_bit_size * 0)) |  //
+      (static_cast<std::uint64_t>(coord[1] & coord_bit_mask) << (coord_bit_size * 1)) |  //
+      (static_cast<std::uint64_t>(coord[2] & coord_bit_mask) << (coord_bit_size * 2));
     coord_pt[i] = {bits, i};
   }
 

include/small_gicp/util/downsampling_omp.hpp

@@ -31,7 +31,7 @@ std::shared_ptr<OutputPointCloud> voxelgrid_sampling_omp(const InputPointCloud&
 
   std::vector<std::pair<std::uint64_t, size_t>> coord_pt(traits::size(points));
 #pragma omp parallel for num_threads(num_threads) schedule(guided, 32)
-  for (size_t i = 0; i < traits::size(points); i++) {
+  for (std::int64_t i = 0; i < traits::size(points); i++) {
     const Eigen::Array4i coord = fast_floor(traits::point(points, i) * inv_leaf_size) + coord_offset;
     if ((coord < 0).any() || (coord > coord_bit_mask).any()) {
       std::cerr << "warning: voxel coord is out of range!!" << std::endl;
@@ -40,9 +40,9 @@ std::shared_ptr<OutputPointCloud> voxelgrid_sampling_omp(const InputPointCloud&
     }
     // Compute voxel coord bits (0|1bit, z|21bit, y|21bit, x|21bit)
     const std::uint64_t bits =                                 //
-      ((coord[0] & coord_bit_mask) << (coord_bit_size * 0)) |  //
-      ((coord[1] & coord_bit_mask) << (coord_bit_size * 1)) |  //
-      ((coord[2] & coord_bit_mask) << (coord_bit_size * 2));
+      (static_cast<std::uint64_t>(coord[0] & coord_bit_mask) << (coord_bit_size * 0)) |  //
+      (static_cast<std::uint64_t>(coord[1] & coord_bit_mask) << (coord_bit_size * 1)) |  //
+      (static_cast<std::uint64_t>(coord[2] & coord_bit_mask) << (coord_bit_size * 2));
     coord_pt[i] = {bits, i};
   }
 
@@ -61,7 +61,7 @@ std::shared_ptr<OutputPointCloud> voxelgrid_sampling_omp(const InputPointCloud&
   std::atomic_uint64_t num_points = 0;
 
 #pragma omp parallel for num_threads(num_threads) schedule(guided, 4)
-  for (size_t block_begin = 0; block_begin < traits::size(points); block_begin += block_size) {
+  for (std::int64_t block_begin = 0; block_begin < traits::size(points); block_begin += block_size) {
     std::vector<Eigen::Vector4d> sub_points;
     sub_points.reserve(block_size);
 

include/small_gicp/util/downsampling_tbb.hpp

@@ -41,9 +41,9 @@ std::shared_ptr<OutputPointCloud> voxelgrid_sampling_tbb(const InputPointCloud&
 
       // Compute voxel coord bits (0|1bit, z|21bit, y|21bit, x|21bit)
       const std::uint64_t bits =                                 //
-        ((coord[0] & coord_bit_mask) << (coord_bit_size * 0)) |  //
-        ((coord[1] & coord_bit_mask) << (coord_bit_size * 1)) |  //
-        ((coord[2] & coord_bit_mask) << (coord_bit_size * 2));
+        (static_cast<std::uint64_t>(coord[0] & coord_bit_mask) << (coord_bit_size * 0)) |  //
+        (static_cast<std::uint64_t>(coord[1] & coord_bit_mask) << (coord_bit_size * 1)) |  //
+        (static_cast<std::uint64_t>(coord[2] & coord_bit_mask) << (coord_bit_size * 2));
       coord_pt[i] = {bits, i};
     }
   });

include/small_gicp/util/normal_estimation_omp.hpp

@@ -20,7 +20,7 @@ template <typename Setter, typename PointCloud, typename KdTree>
 void estimate_local_features_omp(PointCloud& cloud, KdTree& kdtree, int num_neighbors, int num_threads) {
   traits::resize(cloud, traits::size(cloud));
 #pragma omp parallel for num_threads(num_threads)
-  for (size_t i = 0; i < traits::size(cloud); i++) {
+  for (std::int64_t i = 0; i < traits::size(cloud); i++) {
     estimate_local_features<Setter>(cloud, kdtree, num_neighbors, i);
   }
 }

pyproject.toml

@@ -28,6 +28,7 @@ wheel.py-api = "cp312" # Build stable ABI wheels for CPython 3.12+
 
 [tool.scikit-build.cmake.define]
 BUILD_PYTHON_BINDINGS = true
+BUILD_HELPER = false
 BUILD_SHARED_LIBS = false
 
 [tool.cibuildwheel]