MCPcopy Create free account
hub / github.com/pytorch/pytorch / library_paths

Function library_paths

torch/utils/cpp_extension.py:1178–1211  ·  view source on GitHub ↗

Get the library paths required to build a C++ or CUDA extension. Args: cuda: If `True`, includes CUDA-specific library paths. Returns: A list of library path strings.

(cuda: bool = False)

Source from the content-addressed store, hash-verified

1176
1177
1178def library_paths(cuda: bool = False) -> List[str]:
1179 """
1180 Get the library paths required to build a C++ or CUDA extension.
1181
1182 Args:
1183 cuda: If `True`, includes CUDA-specific library paths.
1184
1185 Returns:
1186 A list of library path strings.
1187 """
1188 # We need to link against libtorch.so
1189 paths = [TORCH_LIB_PATH]
1190
1191 if cuda and IS_HIP_EXTENSION:
1192 lib_dir = 'lib'
1193 paths.append(_join_rocm_home(lib_dir))
1194 if HIP_HOME is not None:
1195 paths.append(os.path.join(HIP_HOME, 'lib'))
1196 elif cuda:
1197 if IS_WINDOWS:
1198 lib_dir = os.path.join('lib', 'x64')
1199 else:
1200 lib_dir = 'lib64'
1201 if (not os.path.exists(_join_cuda_home(lib_dir)) and
1202 os.path.exists(_join_cuda_home('lib'))):
1203 # 64-bit CUDA may be installed in 'lib' (see e.g. gh-16955)
1204 # Note that it's also possible both don't exist (see
1205 # _find_cuda_home) - in that case we stay with 'lib64'.
1206 lib_dir = 'lib'
1207
1208 paths.append(_join_cuda_home(lib_dir))
1209 if CUDNN_HOME is not None:
1210 paths.append(os.path.join(CUDNN_HOME, lib_dir))
1211 return paths
1212
1213
1214def load(name,

Callers 2

CppExtensionFunction · 0.85
CUDAExtensionFunction · 0.85

Calls 4

_join_rocm_homeFunction · 0.85
_join_cuda_homeFunction · 0.85
appendMethod · 0.45
joinMethod · 0.45

Tested by

no test coverage detected

Used in the wild real call sites across dependent graphs

searching dependent graphs…