MSCCL Multithreaded regression alternative state management #1352
+40
−20
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Outdated suggestions cannot be applied.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.
Details
Do not mention proprietary info or link to internal work items in this PR.
Work item: Internal
What were the changes?
Address MT MSCCL issue and reenable MSCCL in MT mode through enabling MSCCL single process mode
This is change to the original implementation that hands off the state from thread-local to rank appropriately, and also uses
vector
instead ofunordered_map
.Why were the changes made?
Severe regression when multiple devices are used per thread in rccl-tests allreduce.
How was the outcome achieved?
Set the device id correctly so that scratch memory and sync buffers are allocated on the right device
Additional Details:
Single-threaded (original | fix)
Multi-threaded (original | fix)
Multi-process (original | fix)
Approval Checklist
Do not approve until these items are satisfied.