llama.cpp

ggml-cpu: optimise assembly calls for hsum on s390x

#14037

Merged