]> git.djapps.eu Git - pkg/ggml/sources/llama.cpp/commit
overview / pkg / ggml / sources / llama.cpp / commit
summary | shortlog | log | commit | commitdiff | tree
(parent: e34c5af)
vulkan: optimize coopmat2 dequant functions (#10855)
authorJeff Bolz <redacted>
Sat, 21 Dec 2024 07:04:45 +0000 (01:04 -0600)
committerGitHub <redacted>
Sat, 21 Dec 2024 07:04:45 +0000 (08:04 +0100)
commita91a41364b25705dbb81ae996bc35c3440c63b35
treeeeec13a3c89fa412a3fd1ff3ff7ff59b158cc18ctree
parente34c5af43f941f0ddb92466776339897295aca11commit | diff
vulkan: optimize coopmat2 dequant functions (#10855)

Change the code to do 16b loads when possible and extract the appropriate
component late, so the code is effectively decoding a pair of elements and
then selecting one. This can allow more commoning to happen in the compiler
when neighboring elements are loaded.
ggml/src/ggml-vulkan/vulkan-shaders/dequant_funcs_cm2.comp diff | blob | history
Packaging of ggml-org/llama.cpp