]> git.djapps.eu Git - pkg/ggml/sources/ggml/commit
overview / pkg / ggml / sources / ggml / commit
summary | shortlog | log | commit | commitdiff | tree
(parent: f1e966c)
vulkan: optimize coopmat2 dequant functions (llama/10855)
authorJeff Bolz <redacted>
Sat, 21 Dec 2024 07:04:45 +0000 (01:04 -0600)
committerGeorgi Gerganov <redacted>
Fri, 3 Jan 2025 12:00:38 +0000 (14:00 +0200)
commit2bd1868d538d09adea79f1f05c5c14eb3fc94cc4
tree1e7a3e113282ecffedc914ccbf6064e20d65e5f0tree
parentf1e966c8b1b98d93c86a0c84a736142ad4c2b778commit | diff
vulkan: optimize coopmat2 dequant functions (llama/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.
src/ggml-vulkan/vulkan-shaders/dequant_funcs_cm2.comp diff | blob | history
Packaging of ggml-org/ggml