llama.cpp
Fix data race in CUDA's "cpy" kernel (influences GGML's DUP, CONT operations).
#20507

Merged

Fix data race in CUDA's "cpy" kernel (influences GGML's DUP, CONT operations). #20507

am17an merged 2 commits into ggml-org:master from Exile333:cuda_cpy_datarace_fix

Fix datarace in CUDA's "cpy" kernel.

47c1c71e

ggerganov requested a review 136 days ago

ggerganov removed review request 136 days ago

github-actions added Nvidia GPU

github-actions added ggml

JohannesGaessler approved these changes on 2026-03-13

ORippler commented on 2026-03-13

Remove extra barrier by using more of shared memory.

79c9996f

JohannesGaessler approved these changes on 2026-03-13

am17an merged 5a32a9b8 into master 135 days ago

Exile333 deleted the cuda_cpy_datarace_fix branch 133 days ago

Reviewers

JohannesGaessler

ORippler

Assignees

No one assigned

Labels

Nvidia GPU ggml

Milestone

No milestone