Within the large scale scientific computing arena, there is a need to transfer dynamic access optimization (DAO) techniques, developed at Lawrence Livermore National Laboratory (LLNL) and the University of Utah, into new computing technology that utilizes Multi-Adaptive Processors (MAPs) based on Field Programmable Gate Arrays (FPGAs). This project will develop memory optimization technology to efficiently deliver data to and distribute them within the MAPs. This memory accessing technology should make reconfigurable symmetric multi-processor (SMP) architectures a cost-effective solution for large-scale scientific computing. Phase I will focus on, implementing (1) DAO techniques for the MAP that are appropriate for applications with regular access patterns and (2) the compiler modifications necessary to use them. The efficacy of the approach will be demonstrated for at least one significant application and several small computationally intensive kernels. Commercial Applications and Other Benefits as described by awardee: The use of FPGA technology on DOE critical missions offers the potential for rapid prototyping and simultation of logic solutions of targeted algorithms, orders of magnitude speed-ups for major computational algorithms, major reduction in the cost of a computer system through the reduced processor counts required for reduction, and reduction in operating costs and floor space for the computing system.
