# use already compiled GPU op and not the others. # Also, if cxx is not available, we need to disable all GPU code. set_cuda_disabled() compile_cuda_ndarray = False elif not config.device.startswith('gpu') and config.force_device: # We where asked to NEVER use the GPU set_cuda_disabled() compile_cuda_ndarray = False else: # Add the theano cache directory's cuda_ndarray subdirectory to the # list of places that are hard-coded into compiled modules' runtime # library search list. This works in conjunction with # nvcc_compiler.NVCC_compiler.compile_str which adds this folder during # compilation with -L and also adds -lcuda_ndarray when compiling # modules. nvcc_compiler.add_standard_rpath(cuda_ndarray_loc) compile_cuda_ndarray = not try_import() if compile_cuda_ndarray and cuda_available: get_lock() try: # Retry to load again in case someone else compiled it # while we waited for the lock if not try_import(): try: if not nvcc_compiler.is_nvcc_available(): set_cuda_disabled() if cuda_available: code = open(os.path.join(cuda_path, "cuda_ndarray.cu")).read()
# use already compiled GPU op and not the others. # Also, if cxx is not available, we need to disable all GPU code. set_cuda_disabled() compile_cuda_ndarray = False elif not config.device.startswith('gpu') and config.force_device: # We where asked to NEVER use the GPU set_cuda_disabled() compile_cuda_ndarray = False else: # Add the theano cache directory's cuda_ndarray subdirectory to the # list of places that are hard-coded into compiled modules' runtime # library search list. This works in conjunction with # nvcc_compiler.NVCC_compiler.compile_str which adds this folder during # compilation with -L and also adds -lcuda_ndarray when compiling # modules. nvcc_compiler.add_standard_rpath(cuda_ndarray_loc) compile_cuda_ndarray = not try_import() if compile_cuda_ndarray and cuda_available: get_lock() try: # Retry to load again in case someone else compiled it # while we waited for the lock if not try_import(): try: if not nvcc_compiler.is_nvcc_available(): set_cuda_disabled() if cuda_available: code = open(os.path.join(cuda_path,