llama.cpp

opencl: use flat variants of gemv for very large M

#24006

Merged