Llama.cpp supports Vulkan, which is supported by all GPU vendors that care about standards and interoperability.

The default should be open and portable APIs, not needlessly furthering a hegemony that is detrimental to us all.