]> git.djapps.eu Git - pkg/ggml/sources/llama.cpp/commit
overview / pkg / ggml / sources / llama.cpp / commit
summary | shortlog | log | commit | commitdiff | tree
(parent: 0c6ee1c)
CUDA: better coalesce data-access for contiguous concat (#22330)
authorOliver Simons <redacted>
Sun, 26 Apr 2026 07:21:45 +0000 (09:21 +0200)
committerGitHub <redacted>
Sun, 26 Apr 2026 07:21:45 +0000 (09:21 +0200)
commitb1a5bd4e0c19ba8e82eea716a8362c30918b9560
tree4d502e704c55f2c298246f6638c44316c3f7a101tree
parent0c6ee1cadee96d3ba6e06afb0c001d33f413a0b0commit | diff
CUDA: better coalesce data-access for contiguous concat (#22330)

Also, distribute all elements across CTAs evenly instead of launching
one CTA per dim
ggml/src/ggml-cuda/concat.cu diff | blob | history
Packaging of ggml-org/llama.cpp