In this research supported by two grants from the National Science Foundation, we will develop efficient compilation and runtime techniques for optimized code generation for distributed memory multicomputers, and distributed shared memory multiprocessors. We will produce the next generation PARADIGM compiler to handle general block, cyclic and block-cyclic data distributions in regular applications. We will develop techniques using global dataflow analysis to perform redundant array access elimination. We will develop techniques for generating code for one-way communication such as get/put communication. We will investigate techniques for handling regular and irregular codes in a uniform framework based on the PILAR library. We will also develop techniques for cache memory management in distributed shared memory machines such as the SGI Origin 2000 and the HP/Convex Exemplar SPP 2000 system. Finally, we will develop compiler and runtime support for parallel I/O in parallel computing. Another future direction of the PARADIGM compiler is to develop techniques to automatically parallelize MATLAB programs for parallel machines, and distributed embedded systems.