The loop

for $($i $=0$; i $<$ n; i$++)$

for $($j $=0$; j $<$ n; j$++)$

a$[$i$][$j$]=$ b$[$i$][$j$]+$c$[$i$][$j$]$;

always results in a smaller number of cache misses if row$-$major order is used than it does if column$-$major order is used.