SRAD_Rodinia
ILP_Rate
ILP_Rate = ILP_16K/ILP_32
Dependency Depth:
The calculation of sum2 has the longest chain of dependent instructions.
1,2: Calculate i + Nr*j
3,4: Address Translation and Load image[i + Nr*j]
5,6: Calculate sum2 = sum2 + tmp*tmp.
Dependency Depth = 6.
ILP_32 = 32 / D.D = 32 / 6 = 5.33.
Number of instructions per iteration of the loop = 10
Taking level-4 loop unrolling:
New Number of instructions per iteration of the loop = 10 * 4 = 40
ILP_16K = (16*1024*40)/((16*1024)+(40*6)) = 39.42
ILP_Rate = 39.42 / 5.33 = 7.39.