Methods for Efficient Application Execution on Multiple Computer Architectures

The overall objective of this program is to develop methods by which a set of critical software applications can be programmed to execute efficiently on a variety of existing and emerging computer architectures. The methods address both productivity in porting existing software and approaches to maximize the portability of new software. Specifically, we select a problem domain of applications and architectures of interest and propose compatible tools enabling: 1) automated software management of scratchpad memory, 2) memory-related programming language extensions, 3) a new standard for architecture-independent numerical correctness based on Unum, 4) natural representation of programs, and 5) evaluation of relative performance between architectures. REX Computing is a fabless semiconductor company focused on the development of a new general-purpose computer architecture that is positioned to derive significant benefit from investigating each of these areas. Phase I will investigate the feasibility of the tools and result in plans describing the implementation and expected benefits of each. The Phase I Option will further refine these plans and begin implementation of the tools and Phase II will continue this process with the goal of demonstrating fully functional implementations of the tools and support commercial viability of their use.