Software Tools for the Design and Tuning of Parallel Supercomputer Applications

点击翻译按钮获取中文摘要

Current trends in high performance computer design indicate that parallel processing is the only way to deliver the speed needed to tackle the computational problems confronted by modern science and engineering. It is not reasonable to expect that programmers can successfully program the variety of parallel architectures being proposed without the assistance of automated and interactive parallelization tools. This project will design software tools that help the programmer develop new programs, perhaps by restructuring existing programs, that run efficiently on one or more parallel architectures. The work concerns the design and prototyping of three related tools or sets of tools: (1) a tool for parallel program performance estimation, (2) a set of tools for automatically or interactively parallelizing Lisp programs, and (3) a portable run-time environment for parallelized programs. The direct impact of this research will be that the time required to develop new programs for parallel architectures will decrease and the efficiency of programs running on parallel architectures will increase. Furthermore, because less time will be required to develop new programs it will be possible to test out new architectural ideas and new parallel processors more quickly and with less programmer effort. This may in turn lead to more rapid improvements in parallel architectures.