作者: Manish Gupta , Sam Midkiff , Edith Schonberg , Ven Seshadri , David Shields
关键词: Computer science 、 Optimizing compiler 、 Programming language 、 Parallel computing 、 Compiler 、 Symbolic data analysis 、 Compiler correctness 、 Automatic parallelization 、 Functional compiler 、 Fortran 、 Software prototyping
摘要: We describe pHPF, an research prototype HPF compiler for the IBM SP series parallel machines. The accepts as input Fortran 90 and 77 programs, augmented with directives; sequential loops are automatically parallelized. supports symbolic analysis of expressions. This allows parameters such number processors to be unknown at compile-time without significantly affecting performance. Communication schedules computation guards generated in a parameterized form compile-time. Several novel optimizations improved versions well-known have been implemented pHPF exploit parallelism reduce communication costs. These include elimination redundant using data-availability analysis; collective communication; new techniques mapping scalar variables; coarse-grain wavefronting; reduction multi-dimensional shift communications. present experimental results some benchmark routines. show effectiveness generating efficient code programs.