Reduction of Quadratic Equations for Data Dependence Tests

摘要

A number of data dependence tests have been proposed for detecting parallelism in sequential programs. Despite the fact that nonlinear subscripts appear in real-life programs sometimes, most dependence tests restrict array subscripts to be linear function of loop index variables. The result of the experiment from Eispack. Linpack and Lapack have shown us that quadratic expressions that it is one of nonlinear subscripts can be expressed in a special form. This paper presents a technique that breaks down a quadratic equation into two linear equations if it satisfies some conditions, then it improve accuracy and efficiency of data dependence analysis.