A Demand-Driven Pointer-Range Analysis Technique for Data Transmission Optimization

Abstract

The goal of range analysis is to determine a program variable's minimum and maximum value at runtime and it becomes more complex to calculate the range space when the variable is a pointer. In this paper, we analyzed the optimization problem of data transmission in parallelization for heterogeneous structure and distributed memory structure. On the basis of symbolic range analysis, we proposed a demand-driven pointer-range analysis technique for data transmission optimization. At first, we introduced the analysis framework of this technique and the representations of pointer range. Then we described the algorithm of the demand-driven pointer-range analysis. The experimental results with various benchmarks demonstrate that our technique can bring about significant performance improvement.