A Dynamic Data Dependence Analysis Approach for Software Pipelining

摘要

This paper presents a run-time pointer aliasing disambiguation method for software pipelining techniques. By combining hardware with software, the method is better than run-time checking method or run-time compensation method, which is capable of dealing with irreversible code, and has limited compensation code space without serious rerollability problem. The new method solves pointer aliasing problem efficiently and makes it possible to obtain potential instruction-level parallel speedup. In this paper instruction-level parallel speedups of the new method are analyzed in detail. Three theoretical speedups, i.e., general speedup, probabilistic speedup and mean speedup with probability, are given, which will be helpful for studying and evaluating instruction-level parallelism of the new method.