Parallel Scheduling of a Non Procedural Specification

这项工作涉及在非过程性规范中综合一组方程的并行调度。将利用两种形式的并行性。第一种是FORALL并行性，其中多个独立的循环体可以同时运行。第二种是任务级并行，其中多个模块(可能包括forall循环)同时处于活动状态。以往从非过程性规范进行程序综合的研究主要针对顺序、数据流和分布式处理环境。这项工作集中在一个日益重要的目标机器类：共享内存MIMD配置，其中一个程序由一个任务组组成，而不是一个单一的单指令流任务。为MIMD机器生成并行程序的其他研究工作涉及并行化顺序FORTRAN程序。在这项工作中，定义了一种非常高级的问题描述语言(PS)。这种语言的使用使领域专家(而不是程序员)能够生成并行计算机程序。程序并行性的研究已经成为多指令流计算机情感使用的一个重要挑战。这一努力将增加对这一主题的理解。

This work is concerned with synthesizing a parallel schedule for a set of equations in a nonprocedural specification. Two forms of parallelism will be exploited. The first is forall parallelism, in which multiple independent loop bodies may be run concurrently. The second is task level parallelism, in which multiple modules (perhaps including forall loops) are active concurrently. Previous research in program synthesis from non procedural specification has been directed toward sequential, dataflow and distributed processing environments. This work focuses on an increasingly important target machine class: shared memory MIMD configurations, in which a program consists of a task force rather than one monolithic single instruction stream task. Other research efforts in generating parallel programs for MIMD machines have dealt with parallelizing sequential FORTRAN programs. In this work, a very high level language for problem description (PS) is defined. Use of such a language enables domain experts (as opposed to programmers) to produce parallel computer programs. The study of parallelism in programs has become an important challenge for the affective use of multiple instruction stream computers. This effort will add to the understanding of the topic.