New submission from Dariusz Trawinski <dtraw...@gmail.com>:
Currently, in order to share numpy array between processes via multiprocessing.SharedMemory object, it is required to copy the memory content with: input = np.ones((1,10,10,10)) shm = shared_memory.SharedMemory(create=True, size=input.nbytes) write_array = np.ndarray(input.shape, dtype=input.dtype,buffer=shm.buf) write_array1[:] = input[:] In result the original numpy array is duplicated in RAM. It also adds extra cpu cycles to copy the content. I would like to recommend adding an option to create shared memory object by pointing it to existing memoryview object, beside current method of using shared memory name. Is that doable? ---------- components: C API messages: 362754 nosy: Dariusz Trawinski priority: normal severity: normal status: open title: create multiprocessing.SharedMemory by pointing to existing memoryview type: performance versions: Python 3.8 _______________________________________ Python tracker <rep...@bugs.python.org> <https://bugs.python.org/issue39767> _______________________________________ _______________________________________________ Python-bugs-list mailing list Unsubscribe: https://mail.python.org/mailman/options/python-bugs-list/archive%40mail-archive.com