Prefer device-local allocations on AMD GCN Vulkan GPUs#10

rasbid wants to merge 1 commit intomasterrasbid/llama.cpp:masterfrom

codex/modify-buffer-allocation-for-amd-gcnrasbid/llama.cpp:codex/modify-buffer-allocation-for-amd-gcnCopy head branch name to clipboard