Add windows 10 support .

CMakeLists.txt

 cmake_minimum_required(VERSION 3.13 FATAL_ERROR)
 project(SparseConv LANGUAGES CXX CUDA VERSION 1.0)
 
-option(SPCONV_BuildTests "Build the unit tests when BUILD_TESTING is enabled." ON)
+option(SPCONV_BuildTests "Build the unit tests when BUILD_TESTING is enabled." OFF)
 set(CMAKE_CXX_EXTENSIONS OFF) # avoid gnu++11 be added to CXX flags
 
 set(CUDA_TOOLKIT_ROOT_DIR "${CMAKE_CUDA_COMPILER}")

README.md

@@ -8,7 +8,7 @@ The GPU Indice Generation algorithm is a unofficial implementation of paper [SEC
 
 This project only support CUDA 9.0+. If you are using cuda 8.0, please update it to 9.0.
 
-## Install 
+## Install on Ubuntu 16.04/18.04
 
 0. Use ```git clone xxx.git --recursive``` to clone this repo.
 
@@ -16,7 +16,21 @@ This project only support CUDA 9.0+. If you are using cuda 8.0, please update it
 
 2. Download cmake >= 3.13.2, then add cmake executables to PATH.
 
-3. Ensure you have install pytorch 1.0 in your environment, run ```python setup.py bdist_wheel``` (don't use ```python setup.py install```).
+3. Ensure you have installed pytorch 1.0 in your environment, run ```python setup.py bdist_wheel``` (don't use ```python setup.py install```).
+
+4. Run ```cd ./dist```, use pip to install generated whl file.
+
+## Install on Windows 10 with CUDA 10 and python 3.6 (python 3.7 may have problem, see [this](https://github.com/pytorch/pytorch/issues/17233))
+
+Since install newest driver and CUDA is very simple on windows, please use CUDA 10 on windows. 
+
+0. Use ```git clone xxx.git --recursive``` to clone this repo.
+
+1. Download compressed files from boost official website and copy headers (i.e. boost_1_69/boost) to spconv/include.
+
+2. Download and install cmake >= 3.13.2, select add cmake to User or System PATH.
+
+3. Ensure you have installed pytorch 1.0 in your environment, run ```python setup.py bdist_wheel``` (don't use ```python setup.py install```).
 
 4. Run ```cd ./dist```, use pip to install generated whl file.
 

include/spconv/box_iou.h

@@ -54,7 +54,7 @@ rbbox_iou(py::array_t<DType> box_corners, py::array_t<DType> qbox_corners,
   auto standup_iou_r = standup_iou.template unchecked<2>();
   auto N = box_corners_r.shape(0);
   auto K = qbox_corners_r.shape(0);
-  py::array_t<DType> overlaps = zeros<DType>({N, K});
+  py::array_t<DType> overlaps = zeros<DType>({int(N), int(K)});
   auto overlaps_rw = overlaps.template mutable_unchecked<2>();
   if (N == 0 || K == 0) {
     return overlaps;

include/spconv/geometry.h

@@ -166,7 +166,8 @@ Index getIndicePairsConv(tv::TensorView<const Index> indicesIn,
     kernelVolume *= kernelSize[i];
   }
   Index numValidPoints = 0;
-  Index validPoints[kernelVolume * (NDim + 1)];
+  std::vector<Index> validPoints_(kernelVolume * (NDim + 1));
+  Index* validPoints = validPoints_.data();
   Index *pointPtr = nullptr;
   for (int j = 0; j < numActIn; ++j) {
     batchIdx = indicesIn(j, 0);
@@ -216,7 +217,8 @@ Index getIndicePairsDeConv(tv::TensorView<const Index> indicesIn,
     kernelVolume *= kernelSize[i];
   }
   Index numValidPoints = 0;
-  Index validPoints[kernelVolume * (NDim + 1)];
+  std::vector<Index> validPoints_(kernelVolume * (NDim + 1));
+  Index* validPoints = validPoints_.data();
   Index *pointPtr = nullptr;
   for (int j = 0; j < numActIn; ++j) {
     batchIdx = indicesIn(j, 0);
@@ -265,7 +267,9 @@ Index getIndicePairsSubM(tv::TensorView<const Index> indicesIn,
     kernelVolume *= kernelSize[i];
   }
   Index numValidPoints = 0;
-  Index validPoints[kernelVolume * (NDim + 1)];
+  // Index validPoints[kernelVolume * (NDim + 1)];
+  std::vector<Index> validPoints_(kernelVolume * (NDim + 1));
+  Index* validPoints = validPoints_.data();
   Index *pointPtr = nullptr;
   Index index = 0;
   for (int j = 0; j < numActIn; ++j) {

include/spconv/nms.h

@@ -34,9 +34,9 @@ std::vector<int> non_max_suppression_cpu(py::array_t<DType> boxes,
   auto ndets = boxes.shape(0);
   auto boxes_r = boxes.template unchecked<2>();
   auto order_r = order.template unchecked<1>();
-  auto suppressed = zeros<int>({ndets});
+  auto suppressed = zeros<int>({int(ndets)});
   auto suppressed_rw = suppressed.template mutable_unchecked<1>();
-  auto area = zeros<DType>({ndets});
+  auto area = zeros<DType>({int(ndets)});
   auto area_rw = area.template mutable_unchecked<1>();
   // get areas
   for (int i = 0; i < ndets; ++i) {
@@ -82,7 +82,7 @@ std::vector<int> rotate_non_max_suppression_cpu(py::array_t<DType> box_corners,
   auto ndets = box_corners.shape(0);
   auto box_corners_r = box_corners.template unchecked<3>();
   auto order_r = order.template unchecked<1>();
-  auto suppressed = zeros<int>({ndets});
+  auto suppressed = zeros<int>({int(ndets)});
   auto suppressed_rw = suppressed.template mutable_unchecked<1>();
   auto standup_iou_r = standup_iou.template unchecked<2>();
   std::vector<int> keep;

setup.py

@@ -17,7 +17,6 @@ LIBTORCH_ROOT = str(Path(torch.__file__).parent)
 
 PYTHON_VERSION = "{}.{}".format(sys.version_info.major, sys.version_info.minor)
 
-
 class CMakeExtension(Extension):
     def __init__(self, name, sourcedir='', library_dirs=[]):
         Extension.__init__(self, name, sources=[], library_dirs=library_dirs)
@@ -33,30 +32,35 @@ class CMakeBuild(build_ext):
                                ", ".join(e.name for e in self.extensions))
 
         if platform.system() == "Windows":
-            raise NotImplementedError
+            cmake_version = LooseVersion(re.search(r'version\s*([\d.]+)', out.decode()).group(1))
+            if cmake_version < '3.13.0':
+                raise RuntimeError("CMake >= 3.13.0 is required on Windows")
 
         for ext in self.extensions:
             self.build_extension(ext)
 
     def build_extension(self, ext):
         extdir = os.path.abspath(os.path.dirname(self.get_ext_fullpath(ext.name)))
-        print(extdir)
-        cmake_args = ['-DCMAKE_LIBRARY_OUTPUT_DIRECTORY=' + extdir + "/spconv",
-                      '-DCMAKE_PREFIX_PATH=' + LIBTORCH_ROOT,
+        cmake_args = [# '-G "Visual Studio 15 2017 Win64"',
+                      '-DCMAKE_PREFIX_PATH={}'.format(LIBTORCH_ROOT),
                       '-DPYBIND11_PYTHON_VERSION={}'.format(PYTHON_VERSION),
                       '-DSPCONV_BuildTests=OFF',
                       '-DCMAKE_CUDA_FLAGS="--expt-relaxed-constexpr"']
-
+        assert self.debug is False, "pytorch ops don't support debug build."
         cfg = 'Debug' if self.debug else 'Release'
         # cfg = 'Debug'
         build_args = ['--config', cfg]
         print(cfg)
         if platform.system() == "Windows":
-            cmake_args += ['-DCMAKE_LIBRARY_OUTPUT_DIRECTORY_{}={}'.format(cfg.upper(), extdir)]
+            cmake_args += ['-DCMAKE_BUILD_TYPE=' + cfg]
+            cmake_args += ['-DCMAKE_LIBRARY_OUTPUT_DIRECTORY_{}={}'.format(cfg.upper(), str(Path(extdir) / "spconv"))]
+            # cmake_args += ['-DCMAKE_ARCHIVE_OUTPUT_DIRECTORY_{}={}'.format(cfg.upper(), str(Path(extdir) / "spconv"))]
+            cmake_args += ['-DCMAKE_RUNTIME_OUTPUT_DIRECTORY_{}={}'.format(cfg.upper(), str(Path(extdir) / "spconv"))]
             if sys.maxsize > 2**32:
                 cmake_args += ['-A', 'x64']
             build_args += ['--', '/m']
         else:
+            cmake_args += ['-DCMAKE_LIBRARY_OUTPUT_DIRECTORY={}'.format(extdir + str(Path(extdir) / "spconv"))]
             cmake_args += ['-DCMAKE_BUILD_TYPE=' + cfg]
             build_args += ['--', '-j4']
 
@@ -65,7 +69,7 @@ class CMakeBuild(build_ext):
                                                               self.distribution.get_version())
         if not os.path.exists(self.build_temp):
             os.makedirs(self.build_temp)
-        
+        print("|||||||||", cmake_args)
         subprocess.check_call(['cmake', ext.sourcedir] + cmake_args, cwd=self.build_temp, env=env)
         subprocess.check_call(['cmake', '--build', '.'] + build_args, cwd=self.build_temp)
 

spconv/__init__.py

@@ -14,6 +14,7 @@
 
 from pathlib import Path
 
+import platform
 import numpy as np
 import torch
 from spconv import utils
@@ -23,7 +24,10 @@ from spconv.conv import SparseInverseConv2d, SparseInverseConv3d
 from spconv.modules import SparseModule, SparseSequential
 from spconv.pool import SparseMaxPool2d, SparseMaxPool3d
 
-_LIB_PATH = str(Path(__file__).parent / "libspconv.so")
+_LIB_FILE_NAME = "libspconv.so"
+if platform.system() == "Windows":
+    _LIB_FILE_NAME = "spconv.dll"
+_LIB_PATH = str(Path(__file__).parent / _LIB_FILE_NAME)
 torch.ops.load_library(_LIB_PATH)
 
 def scatter_nd(indices, updates, shape):

src/utils/CMakeLists.txt

-add_library(spconv_nms SHARED nms.cu)
+add_library(spconv_nms STATIC nms.cu)
 set_target_properties(spconv_nms PROPERTIES VERSION ${PROJECT_VERSION})
 set_target_properties(spconv_nms PROPERTIES SOVERSION 1)
 target_include_directories(spconv_nms PRIVATE ${ALL_INCLUDE})
@@ -17,6 +17,6 @@ set_property(TARGET spconv_utils PROPERTY CXX_STANDARD 14)
 set_property(TARGET spconv_utils PROPERTY CUDA_STANDARD 14)
 set_target_properties(spconv_utils PROPERTIES PREFIX "${PYTHON_MODULE_PREFIX}"
                                          SUFFIX "${PYTHON_MODULE_EXTENSION}")
-target_link_libraries(spconv_utils ${CUDA_CUDART} pybind11::module -Wl,--no-as-needed spconv_nms)
+target_link_libraries(spconv_utils ${CUDA_CUDART} pybind11::module spconv_nms)
 
 install (TARGETS spconv_utils DESTINATION lib)