Efficient Implementation of Dynamic Parallel Programs

点击翻译按钮获取中文摘要

The project is studying the feasibility of using high level programming constructs which directly express the parallelism in an algorithm. Although parallelizing compilers for sequential programming languages provide good results for some numerical codes, namely those with regular communication patterns, such compilers are not effective in extracting parallelism in dynamic programs. The methodology used in this project is to express a computation as a collection of encapsulated, asynchronous fine- grained objects, or actors, which interact by sending each other messages. The project is studying ways of reducing the costs associated with managing fine-grained objects and messages. Moreover, because efficient execution on concurrent computer architectures also requires architecture-specific placement (and migration) strategies, the project is studying methods to allow code for placement to be independently expressed and combined with code for the logical operations in an algorithm. The advantage of such a separation is to allow portability of the algorithmic code across a wide range of concurrent architectures. ***