7.5 LOOP SPREADING FOR SIMPLE DEPENDENT LOOPS
Consider the dependent loop shown in Listing 7.3, where s(i, j) is some statement or task to be executed.
Listing 7.3 1-D IIR digital filter algorithm
1: for i = 1:I do
2: for j = 1:J do
3: s(i, j) = f(s(i, j − 1))
4: end for
5: end for
where function evaluated by statement s(i, j) depends on s(i, j − 1). One way to distribute the tasks among the processors is to implement each iteration of the outer loop among I processors so that processor i implements all the iterations of the inner loop with its dependencies. We could increase the workload for each processor by allocating more than one iteration of the outer loop for each processor using a similar technique to the one explained in Section 7.3.
3.137.217.41