Data-Parallel Compiler Technologies for Future-Generation Multicomputers

点击翻译按钮获取中文摘要

With prior NSF support, Hatcher and Quinn have demonstrated that data-parallel SIMD programs can be complied into code that executes efficiently on both shared-memory and distributed-memory MIMD computers. In this project they develop new compliation strategies that will enable data-parallel programs to execute efficiently on future-generation multicomputers, including the Touchstone Sigma. The starting point of this work is the new C* language developed by Thinking Machines Corporation. Hatcher and Quinn extend the language to provide a machine-independent means for specifying the distribution of data among the physical processors' memories and to facilitate parallel I/O.