llama.cpp

hexagon: make vmem and buffer-size configurable

#22487

Merged