llama.cpp

ggml : move FP16 <-> FP32 stuff to ggml-impl.h

#3861

Merged