Multiloop Parallelisation Using Unrolling and Fission

Figure 3

Mapping/scheduling example for loop unrolling and fission without data dependencies between iterations: (a) Task graph containing two parallel loops. (b) Task graph after unrolling the two loops for three iterations. (c) Mapping and scheduling solution before unrolling and fission, overall processing time for 3 iterations is 90 time units. (d) Mapping and scheduling solution after unrolling and fission, overall processing time is 50 time units. (e) Execution time of one iteration of the loop body for different processing elements. Higher interloop and intraloop parallelism are achieved by unrolling two loops.