vulkan: prefer host-visible memory buffers on UMA devices#22930

0cc4m merged 2 commits into

ggml-org:masterggml-org/llama.cpp:masterfrom

winstonma:vulkan-uma-allocationwinstonma/llama.cpp:vulkan-uma-allocationCopy head branch name to clipboard