llama.cpp

Hexagon: Process M-tail rows on HMX instead of HVX

#22724

Merged