Memory: performance

Memory hierarchy can have important effect on performance

Inner loop of matrix multiply:

for (i = 0; i < 500; i++)

for (j = 0; j < 500; j++)

for (k = 0; k < 500; k++)

x[i][j] = x[i][j] + y[i][k] * z[k][j];

Running time on Silicon Graphics system with MIPS R4000 processor

and 1MB secondary cache: 77.2 seconds

If loop order reversed so i is innermost: 44.2 seconds

Only difference: order of accessing data

Other compiler optimizations: less than 10 seconds!