add checks for init

mindspore/communication/management.py

@@ -62,8 +62,8 @@ def init(backend_name=None):
     """
     if _is_role_pserver() or _is_role_sched():
         return
+    device_target = context.get_context("device_target")
     if backend_name is None:
-        device_target = context.get_context("device_target")
         if device_target == "Ascend":
             backend_name = "hccl"
         elif device_target == "GPU":
@@ -74,6 +74,8 @@ def init(backend_name=None):
         raise TypeError("Backend name must be a string, but got {}".format(type(backend_name)))
 
     if backend_name == "hccl":
+        if device_target != "Ascend":
+            raise RuntimeError("Device target should be 'Ascend' to init hccl, but got {}".format(device_target))
         init_hccl()
         GlobalComm.BACKEND = Backend("hccl")
         GlobalComm.WORLD_COMM_GROUP = HCCL_WORLD_COMM_GROUP

mindspore/context.py

@@ -380,7 +380,8 @@ def set_auto_parallel_context(**kwargs):
         full_batch (bool): Whether to load the whole batch on each device. Default: False.
         enable_parallel_optimizer (bool): This is a developing feature, which shards the weight update  computation in
                        data parallel training in the benefit of time and memory saving.
-        all_reduce_fusion_config (list): Set allreduce fusion strategy by parameters indices.
+        all_reduce_fusion_config (list): Set allreduce fusion strategy by parameters indices. Only support ReduceOp.SUM
+                       and HCCL_WORLD_GROUP/NCCL_WORLD_GROUP.
 
     Raises:
         ValueError: If input key is not attribute in auto parallel context.

tests/ut/python/communication/test_comm.py

@@ -33,6 +33,7 @@ from mindspore.ops.operations.comm_ops import Broadcast
 
 tag = 0
 
+context.set_context(device_target="Ascend")
 init("hccl")