Rewrite the code for psum1 (Figure 5.1) so that it does not need to repeatedly retrieve the value of p[i] from memory. You do not need to use loop unrolling. We measured the resulting code to have a CPE of 3.00, limited by the latency of floating-point addition.

Figure 5.1

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 /* Compute prefix sum of vector a */ void psum1 (float all, float p[], long n) { } long i; p[0] = a[0]; for (i=1;i