build separate cpu and cuda images

.travis.yml

@@ -110,25 +110,26 @@ install:
   - source activate test
   - conda install pytorch=${TORCH_VERSION} ${TOOLKIT} -c pytorch --yes
   - source script/torch.sh
-  - pip install flake8 codecov
+  - pip install flake8
+  - pip install codecov
   - pip install scipy==1.4.1
-  - python setup.py install
+  - python setup.py .[test]
 script:
   - flake8 .
   - python setup.py test
 after_success:
-  - python setup.py bdist_wheel --dist-dir=dist/torch-${TORCH_VERSION}
-  - python script/rename_wheel.py ${IDX}
+  - python setup.py bdist_wheel --dist-dir=dist
+  - ls -lah dist/
   - codecov
 deploy:
   provider: s3
-  edge: true
   region: eu-central-1
+  edge: true
   access_key_id: ${S3_ACCESS_KEY}
   secret_access_key: ${S3_SECRET_ACCESS_KEY}
   bucket: pytorch-geometric.com
-  local_dir: dist/torch-${TORCH_VERSION}
-  upload_dir: whl/torch-${TORCH_VERSION}
+  local_dir: dist
+  upload_dir: whl/torch-${TORCH_VERSION}+${IDX}
   acl: public_read
   on:
     all_branches: true

CMakeLists.txt

 cmake_minimum_required(VERSION 3.0)
 project(torchcluster)
 set(CMAKE_CXX_STANDARD 14)
-set(TORCHCLUSTER_VERSION 1.5.8)
+set(TORCHCLUSTER_VERSION 1.5.9)
 
 option(WITH_CUDA "Enable CUDA support" OFF)
 

csrc/fps.cpp

@@ -8,7 +8,11 @@
 #endif
 
 #ifdef _WIN32
-PyMODINIT_FUNC PyInit__fps(void) { return NULL; }
+#ifdef WITH_CUDA
+PyMODINIT_FUNC PyInit__fps_cuda(void) { return NULL; }
+#else
+PyMODINIT_FUNC PyInit__fps_cpu(void) { return NULL; }
+#endif
 #endif
 
 torch::Tensor fps(torch::Tensor src, torch::Tensor ptr, torch::Tensor ratio,

csrc/graclus.cpp

@@ -8,7 +8,11 @@
 #endif
 
 #ifdef _WIN32
-PyMODINIT_FUNC PyInit__graclus(void) { return NULL; }
+#ifdef WITH_CUDA
+PyMODINIT_FUNC PyInit__graclus_cuda(void) { return NULL; }
+#else
+PyMODINIT_FUNC PyInit__graclus_cpu(void) { return NULL; }
+#endif
 #endif
 
 torch::Tensor graclus(torch::Tensor rowptr, torch::Tensor col,

csrc/grid.cpp

@@ -8,7 +8,11 @@
 #endif
 
 #ifdef _WIN32
-PyMODINIT_FUNC PyInit__grid(void) { return NULL; }
+#ifdef WITH_CUDA
+PyMODINIT_FUNC PyInit__grid_cuda(void) { return NULL; }
+#else
+PyMODINIT_FUNC PyInit__grid_cpu(void) { return NULL; }
+#endif
 #endif
 
 torch::Tensor grid(torch::Tensor pos, torch::Tensor size,

csrc/knn.cpp

@@ -8,7 +8,11 @@
 #endif
 
 #ifdef _WIN32
-PyMODINIT_FUNC PyInit__knn(void) { return NULL; }
+#ifdef WITH_CUDA
+PyMODINIT_FUNC PyInit__knn_cuda(void) { return NULL; }
+#else
+PyMODINIT_FUNC PyInit__knn_cpu(void) { return NULL; }
+#endif
 #endif
 
 torch::Tensor knn(torch::Tensor x, torch::Tensor y,

csrc/nearest.cpp

@@ -6,7 +6,11 @@
 #endif
 
 #ifdef _WIN32
-PyMODINIT_FUNC PyInit__nearest(void) { return NULL; }
+#ifdef WITH_CUDA
+PyMODINIT_FUNC PyInit__nearest_cuda(void) { return NULL; }
+#else
+PyMODINIT_FUNC PyInit__nearest_cpu(void) { return NULL; }
+#endif
 #endif
 
 torch::Tensor nearest(torch::Tensor x, torch::Tensor y, torch::Tensor ptr_x,

csrc/radius.cpp

@@ -8,7 +8,11 @@
 #endif
 
 #ifdef _WIN32
-PyMODINIT_FUNC PyInit__radius(void) { return NULL; }
+#ifdef WITH_CUDA
+PyMODINIT_FUNC PyInit__radius_cuda(void) { return NULL; }
+#else
+PyMODINIT_FUNC PyInit__radius_cpu(void) { return NULL; }
+#endif
 #endif
 
 torch::Tensor radius(torch::Tensor x, torch::Tensor y,

csrc/rw.cpp

@@ -8,7 +8,11 @@
 #endif
 
 #ifdef _WIN32
-PyMODINIT_FUNC PyInit__rw(void) { return NULL; }
+#ifdef WITH_CUDA
+PyMODINIT_FUNC PyInit__rw_cuda(void) { return NULL; }
+#else
+PyMODINIT_FUNC PyInit__rw_cpu(void) { return NULL; }
+#endif
 #endif
 
 std::tuple<torch::Tensor, torch::Tensor>

csrc/sampler.cpp

@@ -4,7 +4,11 @@
 #include "cpu/sampler_cpu.h"
 
 #ifdef _WIN32
-PyMODINIT_FUNC PyInit__sampler(void) { return NULL; }
+#ifdef WITH_CUDA
+PyMODINIT_FUNC PyInit__sampler_cuda(void) { return NULL; }
+#else
+PyMODINIT_FUNC PyInit__sampler_cpu(void) { return NULL; }
+#endif
 #endif
 
 torch::Tensor neighbor_sampler(torch::Tensor start, torch::Tensor rowptr,

csrc/version.cpp

@@ -6,7 +6,11 @@
 #endif
 
 #ifdef _WIN32
-PyMODINIT_FUNC PyInit__version(void) { return NULL; }
+#ifdef WITH_CUDA
+PyMODINIT_FUNC PyInit__version_cuda(void) { return NULL; }
+#else
+PyMODINIT_FUNC PyInit__version_cpu(void) { return NULL; }
+#endif
 #endif
 
 int64_t cuda_version() {

script/cuda.sh

@@ -69,7 +69,7 @@ if [ "${TRAVIS_OS_NAME}" = "osx" ] && [ "$IDX" = "cpu" ]; then
 fi
 
 if [ "${IDX}" = "cpu" ]; then
-  export FORCE_CPU=1
+  export FORCE_ONLY_CPU=1
 else
   export FORCE_CUDA=1
 fi

script/rename_wheel.py

-import sys
-import os
-import os.path as osp
-import glob
-import shutil
-
-idx = sys.argv[1]
-assert idx in ['cpu', 'cu92', 'cu101', 'cu102', 'cu110']
-
-dist_dir = osp.join(osp.dirname(osp.abspath(__file__)), '..', 'dist')
-wheels = glob.glob(osp.join('dist', '**', '*.whl'), recursive=True)
-
-for wheel in wheels:
-    if idx in wheel:
-        continue
-
-    paths = wheel.split(osp.sep)
-    names = paths[-1].split('-')
-
-    name = '-'.join(names[:-4] + ['latest+' + idx] + names[-3:])
-    shutil.copyfile(wheel, osp.join(*paths[:-1], name))
-
-    name = '-'.join(names[:-4] + [names[-4] + '+' + idx] + names[-3:])
-    os.rename(wheel, osp.join(*paths[:-1], name))

setup.cfg

@@ -5,4 +5,4 @@ description-file = README.md
 test = pytest
 
 [tool:pytest]
-addopts = --capture=no --cov
+addopts = --cov

setup.py

 import os
-import os.path as osp
 import sys
 import glob
+import os.path as osp
+from itertools import product
 from setuptools import setup, find_packages
 
 import torch
@@ -10,44 +11,46 @@ from torch.utils.cpp_extension import BuildExtension
 from torch.utils.cpp_extension import CppExtension, CUDAExtension, CUDA_HOME
 
 WITH_CUDA = torch.cuda.is_available() and CUDA_HOME is not None
+suffices = ['cpu', 'cuda'] if WITH_CUDA else ['cpu']
 if os.getenv('FORCE_CUDA', '0') == '1':
-    WITH_CUDA = True
-if os.getenv('FORCE_CPU', '0') == '1':
-    WITH_CUDA = False
+    suffices = ['cuda', 'cpu']
+if os.getenv('FORCE_ONLY_CUDA', '0') == '1':
+    suffices = ['cuda']
+if os.getenv('FORCE_ONLY_CPU', '0') == '1':
+    suffices = ['cpu']
 
 BUILD_DOCS = os.getenv('BUILD_DOCS', '0') == '1'
 
 
 def get_extensions():
-    Extension = CppExtension
-    define_macros = []
-    extra_compile_args = {'cxx': ['-O2']}
-    extra_link_args = ['-s']
-
-    info = parallel_info()
-    if 'parallel backend: OpenMP' in info and 'OpenMP not found' not in info:
-        extra_compile_args['cxx'] += ['-DAT_PARALLEL_OPENMP']
-        if sys.platform == 'win32':
-            extra_compile_args['cxx'] += ['/openmp']
-        else:
-            extra_compile_args['cxx'] += ['-fopenmp']
-    else:
-        print('Compiling without OpenMP...')
-
-    if WITH_CUDA:
-        Extension = CUDAExtension
-        define_macros += [('WITH_CUDA', None)]
-        nvcc_flags = os.getenv('NVCC_FLAGS', '')
-        nvcc_flags = [] if nvcc_flags == '' else nvcc_flags.split(' ')
-        nvcc_flags += ['-arch=sm_35', '--expt-relaxed-constexpr', '-O2']
-        extra_compile_args['nvcc'] = nvcc_flags
+    extensions = []
 
     extensions_dir = osp.join(osp.dirname(osp.abspath(__file__)), 'csrc')
     main_files = glob.glob(osp.join(extensions_dir, '*.cpp'))
-    extensions = []
-    for main in main_files:
-        name = main.split(os.sep)[-1][:-4]
 
+    for main, suffix in product(main_files, suffices):
+        define_macros = []
+        extra_compile_args = {'cxx': ['-O2']}
+        extra_link_args = ['-s']
+
+        info = parallel_info()
+        if 'backend: OpenMP' in info and 'OpenMP not found' not in info:
+            extra_compile_args['cxx'] += ['-DAT_PARALLEL_OPENMP']
+            if sys.platform == 'win32':
+                extra_compile_args['cxx'] += ['/openmp']
+            else:
+                extra_compile_args['cxx'] += ['-fopenmp']
+        else:
+            print('Compiling without OpenMP...')
+
+        if suffix == 'cuda':
+            define_macros += [('WITH_CUDA', None)]
+            nvcc_flags = os.getenv('NVCC_FLAGS', '')
+            nvcc_flags = [] if nvcc_flags == '' else nvcc_flags.split(' ')
+            nvcc_flags += ['-arch=sm_35', '--expt-relaxed-constexpr', '-O2']
+            extra_compile_args['nvcc'] = nvcc_flags
+
+        name = main.split(os.sep)[-1][:-4]
         sources = [main]
 
         path = osp.join(extensions_dir, 'cpu', f'{name}_cpu.cpp')
@@ -58,8 +61,9 @@ def get_extensions():
         if WITH_CUDA and osp.exists(path):
             sources += [path]
 
+        Extension = CppExtension if suffix == 'cpu' else CUDAExtension
         extension = Extension(
-            'torch_cluster._' + name,
+            f'torch_cluster._{name}_{suffix}',
             sources,
             include_dirs=[extensions_dir],
             define_macros=define_macros,
@@ -77,7 +81,7 @@ tests_require = ['pytest', 'pytest-cov', 'scipy']
 
 setup(
     name='torch_cluster',
-    version='1.5.8',
+    version='1.5.9',
     author='Matthias Fey',
     author_email='matthias.fey@tu-dortmund.de',
     url='https://github.com/rusty1s/pytorch_cluster',
@@ -97,8 +101,7 @@ setup(
     extras_require={'test': tests_require},
     ext_modules=get_extensions() if not BUILD_DOCS else [],
     cmdclass={
-        'build_ext':
-        BuildExtension.with_options(no_python_abi_suffix=True, use_ninja=False)
+        'build_ext': BuildExtension.with_options(no_python_abi_suffix=True)
     },
     packages=find_packages(),
 )

torch_cluster/__init__.py

@@ -3,16 +3,18 @@ import os.path as osp
 
 import torch
 
-__version__ = '1.5.8'
+__version__ = '1.5.9'
+
+suffix = 'cuda' if torch.cuda.is_available() else 'cpu'
 
 for library in [
         '_version', '_grid', '_graclus', '_fps', '_rw', '_sampler', '_nearest',
         '_knn', '_radius'
 ]:
     torch.ops.load_library(importlib.machinery.PathFinder().find_spec(
-        library, [osp.dirname(__file__)]).origin)
+        f'{library}_{suffix}', [osp.dirname(__file__)]).origin)
 
-if torch.version.cuda is not None:  # pragma: no cover
+if torch.cuda.is_available():  # pragma: no cover
     cuda_version = torch.ops.torch_cluster.cuda_version()
 
     if cuda_version == -1: