Detecting more independent loops across hierarchical structures

Detecting coarse-grain parallelism is an important issue for high speed execution on multiprocessor systems. It is especially advantageous to detect loops which are executable in parallel. This paper describes a practical algorithm to detect more parallel loops across hierarchical structures. This paper shows that without any loop transformation a program does not have much coarse-grain parallelism between different loops. Moreover, much more parallelism can be obtained if loop fission and loop peeling are applied to a program in combination with such optimization techniques as inline expansion, constant propagation, branch elimination, and scalar expansion