This Small Business Innovation Research Phase I project seeks to establish the feasibility of an innovative approach for designing a toolset that automates what are perhaps the most complex tasks associated with developing software for parallel computer systems, i.e., software partitioning and mapping. The potential of existing and emerging distributed memory parallel computers to deliver cost effective solutions to computationally challenging problems is well recognized. However, the major impediment to the widespread use of these machines is the difficulty of efficiently partitioning and mapping application algorithms on these machines so as to extract a large portion of the machine's peak performance. The proposed toolset will solve this problem. The toolset will provide an ensemble of tools for assisting the system designer at each step of the application development process, from system definition to code generation for the target multiprocessor architecture. The toolset includes: 1) an automated tool for partitioning the application software; 2) a tool for automatically generating a mapping of the software on to the processors of the target architecture; and 3) a simulation tool to assess the performance of the system. All the tools in the interface will be integrated so as to provide seamless access to them by the user. Also, a sophisticated user interface will provide a uniform and intuitive interface for a system designer to interact with the different tools in the toolset and will guide the user through the sequence of steps associated with the application development process.
