llama.cpp
ggml-cuda: fix divergent barrier in f16 flash attention
#27870
Merged

ggml-cuda: fix divergent barrier in f16 flash attention #27870

siavashnorouzi
siavashnorouzi siavashnorouzi requested a review 85 days ago
github-actions github-actions added ggml
github-actions github-actions added CUDA
siavashnorouzi
siavashnorouzi
JohannesGaessler
JohannesGaessler commented on 2026-09-04
JohannesGaessler JohannesGaessler assigned JohannesGaessler JohannesGaessler 78 days ago
siavashnorouzi ggml-cuda: fix divergent barrier in f16 flash attention
672f59d8
siavashnorouzi ggml-cuda: avoid duplicate metadata pointer setup
c4b9b7e1
siavashnorouzi siavashnorouzi force pushed to c4b9b7e1 78 days ago
JohannesGaessler
JohannesGaessler approved these changes on 2026-09-05
JohannesGaessler JohannesGaessler added merge ready
ggerganov ggerganov merged b74f590e into master 76 days ago
siavashnorouzi

Login to write a write a comment.

Login via GitHub

Reviewers
Assignees
Labels
Milestone