Abstract de la publi numéro 163

Static pointer analysis by aliasing becomes more and more significant for low-level optimisation while superscalar processors exploit more and more instruction-level parallelism. This paper describes an original method to compute static pointer dependencies. Static pointer approximations are extracted from the source language program by the so-called Abstract Interpretation technique and used for improving the efficiency of memory access reordering by allowing some ready loads to be launched in presence of former blocking stores. Our approach is implemented using the SUIF compiler and performance results are obtained thanks to the SimpleScalar simulator.