Example #1
0
def test_nccl_init():
    n = nccl()
    uid = n.get_unique_id()

    n.init(1, uid, 0)

    assert 0 == n.user_rank()
    assert 0 == n.cu_device()

    n.destroy()
Example #2
0
def _func_init_nccl(sessionId, uniqueId):
    """
    Initialize ncclComm_t on worker
    :param workerId: int ID of the current worker running the function
    :param nWorkers: int Number of workers in the cluster
    :param uniqueId: array[byte] The NCCL unique Id generated from the
                     client.
    """
    wid = worker_state(sessionId)["wid"]
    nWorkers = worker_state(sessionId)["nworkers"]

    n = nccl()
    n.init(nWorkers, uniqueId, wid)
    worker_state(sessionId)["nccl"] = n