llama.cpp

hexagon: add fp16 support for binary ops: add,sub,mul,div

#20139

Merged