]> git.djapps.eu Git - pkg/ggml/sources/ggml/commit
overview / pkg / ggml / sources / ggml / commit
summary | shortlog | log | commit | commitdiff | tree
(parent: 2f8c72b)
vulkan: Handle GPUs with less shared memory (llama/10468)
authorJeff Bolz <redacted>
Wed, 27 Nov 2024 07:30:27 +0000 (01:30 -0600)
committerGeorgi Gerganov <redacted>
Tue, 3 Dec 2024 19:05:37 +0000 (21:05 +0200)
commit48fe3adff0a8c2314d73f99aea6d676fcb770cad
treeb53609619671cc3457a9a0957e37bfc1091a840ctree
parent2f8c72b8a6073b0d0c86a47bdc0cb009b43fb0c6commit | diff
vulkan: Handle GPUs with less shared memory (llama/10468)

There have been reports of failure to compile on systems with <= 32KB
of shared memory (e.g. #10037). This change makes the large tile size
fall back to a smaller size if necessary, and makes mul_mat_id fall
back to CPU if there's only 16KB of shared memory.
src/ggml-vulkan/ggml-vulkan.cpp diff | blob | history
Packaging of ggml-org/ggml