Selection of An Intermediate Representationfor Program Analysis and Optimization

Abstract

An Intermediate Representation (IR) is an important part of a compiler. Selectingthe right IR can significantly improve not only analyses and optimizations processesof acompiler but also reduce overall time of compiler design. There are many IRs foundtoday but selecting the right IR for compiler is difficult job because different IRs havedifferent properties. In this dissertation, two important IRs, Static Single Assignment(SSA) and Program Dependence Graph (PDG), are studiedand presented comparativeanalyses betweenPDG and three flavors of SSA form:minimal, pruned and semi-pruned.SelectedIRsare implemented in theMachine SUIF compiler infrastructure. PDG pass isimplemented in this workbuthas usedMachine SUIF Static Single Assignment Libraryof Machine SUIF for SSA form.Selected IRs are tested and analyzed with benchmarkprograms. The results showed that the comparative study presented in this work is veryuseful to the compiler designer for selecting appropriate IR.