Compiler Based Disjoint Eager Execution

Faster computers allow better performance on existing programs and allow new, previously unknown applications to be developed. Instruction level parallel (ILP) processing improves performance by performing many tiny, independent calculations simultaneously. Ordinary programs contain many decisions and branches that axe often unpredictable and make the program difficult to parallelize. Disjoint eager execution (DEE) is an approach to automatically parallelizing programs that allocates computer resources to performing calculations that axe most likely to be needed. Simulations have shown that DEE has the potential to overcome many of the negative effects of branches in ordinary programs. This project will develop a compiler based DEE for an ILP machine with many resources. To overcome the engineering challenges, we have been provided with an advanced ILP compiler and simulator by IBM Research. We will measure the speedup from DEE and other effects on program performance. We expect that the combination of a sophisticated DEE compiler and a simple, but fast computer will greatly reduce the effects of branches.