UCP/RNDV: Avoid strict key for fence between PUT and AM#11636
Open
tvegas1 wants to merge 1 commit into
Open
Conversation
This file contains hidden or 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
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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.
What?
Add
UCT_FENCE_FLAG_AM_ONLYand use it inrndv/put/mtypebefore ATP.Why?
RNDV
put/mtypeonly needs prior PUT data to be visible before the remote ATP AM callback. It does not need ordering between PUTs before and after the ATP.Today the fence arms
fence_beat, but SEND/AM does not consume it, so the next unrelated PUT pays the strict/atomic rkey cost. Impact on setup with PCIe relaxed ordering and two 90GB/s lanes:UCX_RNDV_FRAG_SIZE=cuda:16M ucx_perftest -t ucp_put_bw -s $((64*1024*1024)) -O64 -m cudaHow?
Pass
UCT_FENCE_FLAG_AM_ONLYfor the fenced ATP path. IB RC/DC keeps the fence call but does not armfence_beat; other transports keep their existing fence behavior.RDMA may use relaxed ordering and does not consume RX CQEs. The following SEND/AM does consume an RX CQE, whose PCIe write is non-relaxed, so prior relaxed PCIe writes are visible before the AM callback runs.