llama.cpp

Fix result validation in benchmark-q4_0-matmult

#987

Merged