Contributions to High Performance Fortran Compilation
PhD dissertation, École des mines de Paris.
Report EMP CRI A-293, October 1996.
320 pages, 261 references.
Note: This thesis is mostly written in English with some parts
This thesis presents our contributions to High Performance Fortran (HPF)
compilation. HPF is a data-parallel language based on Fortran. Directives
are used to specify parallelism and data mapping onto distributed memory
parallel architectures. We aim at translating a global addressing implicit
communication HPF program into a message passing parallel model. Our
approach is based on formalizing the compilation technical problems into a
mathematical framework and on using standard algorithms to generate
optimized code. First, the HPF programming language is introduced and the
related work is discussed. Second, we analyse HPF design issues, and
suggest possible improvements to the language definition. Third, we
present our compilation techniques for parallel loops, I/O and HPF
remappings. All issues are addressed, including memory allocation and
addressing, load balancing, temporary storage management and so on.
Optimality results and experiments are presented. Finally we describe our
prototype compiler (HPFC) and the
software environment in which it has been developed
(PIPS), before concluding.