Interprocedural analysis based on guarded array regions

Array data flow information plays an important role for successful automatic parallelization of Fortran programs. This chapter proposes a powerful symbolic array data flow summary scheme to support array privatization and loop parallelization for programs with arbitrary control flow graphs and acyclic call graphs. Our approach summarizes array access information interprocedurally, using guarded array regions. The use of guards allows us to use the information in IF conditions to do path-sensitive data flow summary and thereby to handle difficult cases. We also provide a mechanism to overcome the disadvantages of nonclosed union and difference operations. This improves not only the exactness of summaries, but also the efficiency of the summarizing procedure. Our preliminary result on array privatization shows that our summary scheme is fast and powerful.