Domain Specific Parallel Adaptive Scientific Computation

本课题研究偏微分方程自适应解的特定域并行计算。并行程序的分解是在一系列注释语言的帮助下实现的，这些注释语言使用户能够对用各种编程语言编写的程序进行注释。并行计算的合成依赖于用简单的配置语言描述程序之间的相互作用以及由注释创建的程序片段。通过注释分解和通过配置合成简化了(1)确定粒度以进行高效并行处理的过程，(2)数据分布和(3)运行时优化。

This project investigates domain specific parallel computation for the adaptive solution of partial differential equations. The decomposition of parallel programs is achieved with the aid of a family of annotation languages that enable the user to annotate programs written in various programming languages. The synthesis of parallel computation relies on the description of an overall computation done in a simple configuration language in terms of interactions of programs and their fragments created by annotations. The decomposition by annotations and synthesis through configuration simplify the process of (1) determining the grain size for efficient parallel processing, (2) data distribution, and (3) run time optimization.