llama.cpp

47373271 - HIP: Enable support for ggml_backend_cuda_register_host_buffer (#15615)