Using knowledge-based systems for research on parallelizing compilers

The main function of parallelizing compilers is to analyze sequential progr ams, in particular the loop structure, to detect hidden parallelism and aut omatically restructure sequential programs into parallel subtasks that are executed on a multiprocessor. This article describes the design and impleme ntation of an efficient parallelizing compiler to parallelize loops and ach ieve high speedup rates on multiprocessor systems. It is well known that th e execution efficiency of a loop can be enhanced if the loop is executed in parallel or partially parallel, such as in a DOALL or DOACROSS loop. This article also reviews a practical parallel loop detector (PPD) that is imple mented in our PFPC on finding the parallelism in loops. The PPD can extract the potential DOALL and DOACROSS loops in a program by verifying array sub scripts. In addition, a new model by using knowledge-based approach is prop osed to exploit more loop parallelisms in this paper. The knowledge-based a pproach integrates existing loop transformations and loop scheduling algori thms to make good use of their ability to extract loop parallelisms. Two ru le-based systems, called the KPLT and IPLS, are then developed using repert ory grid analysis and attribute-ordering tables respectively, to construct the knowledge bases. These systems can choose an appropriate transform and loop schedule, and then apply the resulting methods to perform loop paralle lization and obtain a high speedup rate. For example, the IPLS system can c hoose an appropriate loop schedule for running on multiprocessor systems. F inally, a runtime technique based on the inspector/executor scheme is propo sed in this article for finding available parallelism on loops. Our inspect or can determine the wavefronts of a loop with any complex indirected array -indexing pattern by building a DEF-USE table. The inspector is fully paral lel without any synchronization. Experimental results show that the new met hod can resolve any complex data dependence patterns where no previous rese arch can. One of the ultimate goals is to construct a high-performance and portable FORTRAN parallelizing compiler on shared-memory multiprocessors. W e believe that our research may provide more insight into the development o f a highperformance parallelizing compiler Copyright (C) 2001 John Wiley & Sons, Ltd.