Specialized Interprocedural Analysis For Scientific Applications

Scientific applications follow some well-known execution patterns. The patterns, such as using function pointers for model configuration, pose challenges for general purpose tools. Many analyses are too expensive to employ in this case. However, we hypothesize that it is possible to use the patterns to run expensive analysis on specific portions of the code, reducing the cost and improving the benefit. We propose to build a customizable compiler for optimizing scientific applications. The compiler will run analyses that are typically considered too expensive. This analysis would form the basis for accurate optimizations there by increasing performance of such scientific applications