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Outdated suggestions cannot be applied.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.
Issue #, if available:
None
Description of changes:
This PR adds HMEM copy callbacks that use GDRCopy to perform memory copies. This allows aws-ofi-nccl to support providers that copy data between device and host buffers from using
cudaMemcpy
. Because NCCL kernels are running on the GPU, submitting memcpy work to CUDA can (and usually does) result in a deadlock. GDRCopy can be used for such copies without introducing the risk of a deadlock. This change required the introduction of a GDRCopy buffer registration management layer (called hcopy) and a refactor of the MR handle in aws-ofi-nccl so that it can hold both the OFI and GDRCopy registration information for the buffer. GDRCopy callbacks are enabled at compile time by configuring with--with-gdrcopy=...
.Note, this PR is based on top of #145.
By submitting this pull request, I confirm that my contribution is made under the terms of the Apache 2.0 license.