Add fused buffer scaling and unpack/pack kernels on GPU. #2973
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Outdated suggestions cannot be applied.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.
Checklist before submitting
Description
For a Horovod allreduce operation using NCCL, the batched fusion buffer pack, prescaling, postscaling, and batched fusion buffer unpack are all individual CUDA kernel launches. This PR introduces a fused batched memcpy and scaling CUDA kernel to perform a (un)pack and scaling in a single kernel launch. This fusion reduces the number of kernels Horovod launches and improves performance by removing extra read/writes of GPU buffer memory between the existing individual pack and scaling kernels.