make detailed comment in cuda_shift

mmaction/models/backbones/cuda_shift/setup.py

@@ -7,15 +7,18 @@
 
 
 def _find_cuda_home():
-    # guess rule 3 of torch.utils.cpp_extension
+    # set directory path to CUDA which supports `nvcc` command as CUDA_HOME
+    # other than 3 Guess in PyTorch
     nvcc = subprocess.check_output(['which', 'nvcc']).decode().rstrip('\r\n')
     cuda_home = os.path.dirname(os.path.dirname(nvcc))
     print(f'find cuda home:{cuda_home}')
     return cuda_home
 
 
-# remember to overwrite PyTorch auto-detected cuda home which
-# may not be our expected
+# overwrite PyTorch auto-detected CUDA_HOME which may not be our expected,
+# because PyTorch determines the CUDA_HOME using this priority:
+# os.environ['CUDA_HOME'] > path to CUDA supporting `nvcc` > '/usr/local/cuda',
+# although The first two guess should be the same in most cases.
 torch.utils.cpp_extension.CUDA_HOME = _find_cuda_home()
 CUDA_HOME = torch.utils.cpp_extension.CUDA_HOME
 
@@ -34,11 +37,10 @@ def _find_cuda_home():
 else:
     raise ValueError('CUDA is not available')
 
-extra_compile_args = dict(cxx=[])  # ['-fopenmp', '-std=c99']
+extra_compile_args = dict(cxx=[])
 extra_compile_args['nvcc'] = []
 
 this_file = os.path.dirname(os.path.realpath(__file__))
-print(this_file)
 sources = [os.path.join(this_file, fname) for fname in sources]
 headers = [os.path.join(this_file, fname) for fname in headers]
 extra_objects = [os.path.join(this_file, fname) for fname in extra_objects]