ggml : document occupancy heuristics in cuda_op_mean call to reduce_rows kernal#18212
Merged
am17an merged 1 commit intoggml-org:masterfrom Dec 20, 2025
Merged
Conversation
github-actions
Bot
added
Nvidia GPU
Contributor
|
it would be better if you just link the original PR which has the complete discussion |
…row count and column size, derived from historical commit context
Aadeshveer
force-pushed
the
docs-reduce-rows-heuristic
branch
from
0edb2f7 to
473a670
Compare
Contributor
Author
|
I've replaced the explanation with a link to PR #15132 |
am17an
approved these changes
Dec 20, 2025
Anico2
added a commit
to Anico2/llama.cpp
that referenced
this pull request
Jan 15, 2026
…row count and column size, derived from historical commit context (ggml-org#18212)
blime4
referenced
this pull request
in blime4/llama.cpp
Feb 5, 2026
…row count and column size, derived from historical commit context (#18212)
Seunghhon
pushed a commit
to Seunghhon/llama.cpp
that referenced
this pull request
Apr 26, 2026
…row count and column size, derived from historical commit context (ggml-org#18212)
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
2 participants
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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.
Added comments to
ggml_cuda_op_meanto explain the logic behind the thread block size selection (512 vs 128 vs 32) for callingreduce_rowskernalAs a newcomer to the codebase, I spent a significant amount of time trying to decipher why these specific "magic numbers" were chosen and how the
nsmcheck works. I dug through the commit history to find the original optimization context regarding latency hiding and scheduler overhead.This inline documentation documents those heuristics so future contributors don't have to repeat that archaeology work.