Experiences with a parallel algorithm for data flow analysis

Lee, Yong-Fong; Ryder, Barbara G.; Marlowe, Thomas J.

doi:10.1007/bf00127842

Cited by 11 publications

(1 citation statement)

References 36 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Our rst experiments were with a family of parallel hybrid algorithms, which combine xed point iteration within regions and an elimination-like propagation between regions. Using the parallel hybrid algorithm on the reaching de nitions problem, we already have demonstrated the potential of using a parallel analysis technique on a distributed memory machine, the Intel iPSC/2 23,24]. The regions used in hybrid algorithms are single-entry clusters of one or more strongly connected components.…”

Section: Introductionmentioning

confidence: 99%

Region analysis: a parallel elimination method for data flow analysis

Lee

Ryder

Fiuczynski

1995

IIEEE Trans. Software Eng.

View full text Add to dashboard Cite

| Parallel data ow analysis methods o er the promise of calculating detailed semantic information about a program at compile-time more e ciently than sequential techniques. Previous work on parallel elimination methods 1, 2] has been hampered by the lack of control over interval size; this can prohibit e ective parallel execution of these methods. To overcome this problem, we have designed the region analysis method, a new elimination method for data ow analysis. Region analysis emphasizes ow graph partitioning to enable better load balancing in a more e ective parallel algorithm. In this paper, we present the design of region analysis and the empirical results we have obtained that indicate (1) the prevalence of large intervals in ow graphs derived from real programs, and (2) the performance improvement of region analysis over parallel Allen-Cocke interval analysis. Our implementation analyzed programs from the Perfect Benchmarks 3] and netlib 4] running on a Sequent Symmetry S81.Index Terms | Data ow analysis, elimination algorithms, interval analysis, parallel algorithms, program optimization. 1 In 20], Reps proves hardness results for some interprocedural data ow problems, assuming the \meet-overall-valid-paths" precision, and uses them to assert the non-existence of fast (NC-class) parallel algorithms for these problems. Nevertheless, he also points out that for the imprecise version of interprocedural data ow analysis, there does exist a fast (NC-class) parallel algorithm. Moreover, the worst case time complexity of useful data ow techniques are not observed in practice. 2 The available expressions problem involves information necessary for common subexpression elimination 19].

show abstract

Section: Introductionmentioning

confidence: 99%