A parallelization method for DO-loop based on equivalence classification

In this paper, a new method for DO-loop parallelization based on the new concept allocation-dependence and equivalence classification of iteration space is proposed. This method has many advantages:

⊙ It is a general, unified method for DO-loop parallelization.

⊙ It is used in coarse grain parallel partitioning on MIMD and SPMD.

⊙ While partitioning iteration space, it also does the data partition and computation partition such that these partitions are independent each other.

⊙ It can extract the potential parallelism of program accurately.

⊙ Combining with task-level parallelization vectorization and pipeline, it can extract parallelism thoroughly.