Question

You are given the following two pseudo-codes and asked to write C codes to compare their speeds and memory requirements for reasonably large arrays. Any meaningful performance differences between these two loops? Explain your answer.

LOOP A:

For I = 1, 1000
For J = 1, 1000
     c(J) = c(J) + a(I, J)*b(J)

LOOP B:

For J = 1, 1000
For I = 1, 1000
     c(J) = c(J) + a(I, J)*b(J)

Answer 1

observation:

Loop A traverses row wise

Loop B traverses column wise

since i and j both are size of 1000, we can assume this to be 1000X1000 matrix.

Traversing symmetric matrix row wise is far better than traversing it column wise.

reason:

2D matrix is represented as 1D array, row by row.

row-column uses better cpu-caches than column wise.

LOOP A is better than Loop B