metal : improve F32, F16 and BF16 mat-vec multiplication#16057
Merged
metal : improve F32, F16 and BF16 mat-vec multiplication#16057
Conversation
github-actions
Bot
added
ggml
ggerganov
force-pushed
the
gg/metal-mul-mv-opt-2
branch
from
5fbb485 to
64c6dcb
Compare
blime4
referenced
this pull request
in blime4/llama.cpp
Feb 5, 2026
* metal : improve F32, F16 and BF16 mat-vec multiplication ggml-ci * metal : make the NSG a function constant in mul_mv kernels ggml-ci
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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.
Better threadgroup utilization for the floating-point mat-vec kernels:
Also some TG gains for MoE models of any quantization since they use F32 matrix multiplication in the FFN: