llama.cpp

60a76588 - opencl: allow mixed f16/f32 `add` (#15140)