Dependence Analysis and Parallelizing Transformations.

摘要: The first impression one gets when phrases like dependence analysis and automatic parallelization are mentioned, is that of loop programs array variables. This not surprising, since loops the classic repetitive structure in any programming language, clearly, this where spend a significant amount their time. Second, because early impetus on high performance computing for large numerical applications (eg. FORTRAN programs) supercomputers, there has been long research effort parallelizing such programs. area active over quarter century, number well known texts topic readily available. Any approach to must be based following fundamental notions: (i) detection which statements iterations (possibly multiply nested possibly imperfect) depend each other, precise sense them produces value consumed by other; (ii) hence determining operations can executed parallel; (iii) transforming program so choice rendered explicit. problem called analysis, second constitutes additional necessary choose parallelization, third or restructuring. In all generality, these extremely difficult problems. Nevertheless, certain classes elegant powerful methods Therefore, rather than giving “yet another survey” vast field, we present chapter, somewhat less approach,