An Efficient Approach to Data Flow Analysis in a Multiple Pass Global Optimizer

Abstract

Data flow analysis is a time-consuming part of the optimization process. As transformations are made in a multiple pass global optimizer, the data flow information must be updated to reflect these changes. Various approaches have been used, including complete recalculation as well as partial recalculation over the affected area. The approach presented here has been designed for maximum efficiency. Data flow information is completely calculated only once, using an interval analysis method which is slightly faster than a purely iterative approach, and which allows partial recomputation when appropriate. A minimal set of data flow information is computed, keeping the computation and update cost low. Following each set of transformations, the data flow information is updated based on knowledge of the effect of each change. This approach saves considerable time over complete recalculation, and proper ordering of the various optimizations minimizes the amount of update required. © 1988, ACM. All rights reserved.