perf: COALESCE=1 multi-expert coalesced pread for contiguous shard runs - #1398
Draft
KyleSanderson wants to merge 2 commits into
Draft
KyleSanderson wants to merge 2 commits into
KyleSanderson wants to merge 2 commits into
Conversation
When a PIPE miss batch contains experts that are contiguous in the same shard file, issue a single pread for the combined range instead of one per expert. Gated by COALESCE=1 env flag (default off); zero behavior change when off. A/B on GLM-5.2 real workload (--ram 8, PIPE=1 DIRECT=1 IDOT_TEAM=1): prefill: 63.4s -> 58.3s (-8%) decode: 50.9s -> 43.6s (-14%) disk service: prefill -8%, decode -16% No regression in output correctness or expert-matmul IPC.
KyleSanderson
marked this pull request as draft
September 8, 2026 04:43
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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.
Summary
When a PIPE miss batch contains experts that are contiguous in the same shard file,
issue a single pread for the combined range instead of one per expert.
Gated by COALESCE=1 env flag (default off); zero behavior change when off.
A/B on GLM-5.2 real workload (--ram 8, PIPE=1 DIRECT=1 IDOT_TEAM=1):
prefill: 63.4s -> 58.3s (-8%)
decode: 50.9s -> 43.6s (-14%)
disk service: prefill -8%, decode -16%
No regression in output correctness or expert-matmul IPC.
Validation
make -C c checkmake -C c cuda-test(if applicable)Compatibility