Optimal Loop Parallelization under Register Constraints

Abstract

: This report deals with the interaction between instruction scheduling and register allocation, in the case of straight line code and in the case of loops. This problem is at the heart of code optimization in microprocessors with instruction-level parallelism. Usual solutions use heuristics based on a decoupled approach. We propose here a formulation by linear integer programming, that allows dependence, resource and register constraints to be integrated in the same framework . By varying the parameters, all kinds of optimization problems can be solved exactly (maximization of the throughput, minimization of the number of registers). We report on examples of computation timings that turn out to be prohibitive in some specific cases, but tractable on average. Key-words: loop scheduling, register allocation, linear integer programming (R'esum'e : tsvp) This work was partially supported by ESPRIT Project COMPARE Unit de recherche INRIA Rocquencourt Domaine de Voluceau, Rocquencourt, ...