Add intra_rank to all communicators #177
Merged
Conversation
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
keisukefukuda
changed the title
|
Overall this patch is simple and great. But one more, I want to see a small test that assures diff --git a/tests/chainermn_tests/communicator_tests/test_communicator.py b/tests/chainermn_tests/communicator_tests/test_communicator.py
index e06e459..814e5d3 100644
--- a/tests/chainermn_tests/communicator_tests/test_communicator.py
+++ b/tests/chainermn_tests/communicator_tests/test_communicator.py
@@ -195,6 +195,11 @@ def check_send_recv(param, use_gpu):
def check_collective_communication(param, use_gpu):
communicator = create_communicator(param, use_gpu)
+ assert 0 <= communicator.intra_rank
+ assert communicator.intra_rank < communicator.intra_size
+ assert 0 <= communicator.inter_rank
+ assert communicator.inter_rank < communicator.inter_size
+
model = ExampleModel()
if use_gpu:
model.to_gpu() |
kuenishi
requested changes
Dec 25, 2017
kuenishi
reviewed
Dec 25, 2017
| @@ -36,8 +36,27 @@ def __init__(self, obj): | |||
|
|
|||
| class CommunicatorBase(object): | |||
|
|
|||
| def __init__(self, mpi_comm): | |||
| def __init__(self, mpi_comm, use_nccl=False): | |||
There was a problem hiding this comment.
If you change this default value to False, PureNcclCommunicator.__init__(), superclass initialization must be updated, too.
super(PureNcclCommunicator, self).__init__(mpi_comm)
self._init_ranks()There was a problem hiding this comment.
And I added tests to check the values of intra_rank and inter_rank, in test_node_aware_communicator_base.py.
kuenishi
approved these changes
Dec 25, 2017
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
In the current communicator class hierarchy, not all communicator have
intra_rankattribute due to a historical reason.The attribute is currently used to determine if ChainerMN is using GPUs because
intra_rankwas implemented to determine GPUs to use in a local physical node. However, GPU ID andintra_rankare different concepts and the usage is inappropriate.This PR merges
NodeAwareCommunicatorintoCommunicatorBasesoCommunicatorBaseclass and all of its ancestors haveintra_rankattribute.