Locality Can Be Improved
•Improve cache behavior by “blocking”
•
•
•
•
A
B
C
for (p=0; p < n; p=p+t){ //block count
  for (r=0; r < t; r++){
    for (s=0; s < t; s++){
 for (k=0; k < t; k++){                      c[r][s] += a[p+r][p+k]*b[p+k][p+s];
      }
    }
  }
)