Enabling dependence analysis in C

grantor: University of TorontoDependence analysis is a fundamental tool used in many compiler transformations which optimize and parallelize scientific code written for high-performance vector and parallel computer architectures. Implementing dependence analysis for the C programming language is difficult because of complications caused by nonstandard control flow and use of pointers to reference arrays. In order to enable dependence analysis in C, code can be preprocessed to convert loops which violate FORTRAN-like conventions into a canonical form which can be processed successfully by the dependence analyzer. We developed two algorithms to enable such processing. Loop control flow normalization (LCFN) normalizes loop control flow and array reference subscripts. Pointer array access normalization (PAN) recovers implicit array references through pointers. A prototype implementation of the LCFN and FAN methods was built using the Parafrase-2 compiler. Experimental results generated using the SPEC95 and NAS benchmark suites showed that these techniques can successfully enable dependence analysis.M.Sc