Preparing Set-Based Analysis for Run-time Specialization
Hyun-Jun Eo, Kwangkeun Yi · 2000
We present a technique of using static analysis for estimating program’s input-dependent properties. A static analysis that is originally designed for estimating the input-independent properties of programs is transformed into one that can safely estimate the input-dependent properties at the programs ’ input occurrence. No profile is collected and no probing codes inside the running program are needed. Our idea is to defer the finish of the static analysis to the program’s run-time. By analyzing the static analysis, we identify the parts of the analysis that are sensitive to the program’s inputs, hence need to be deferred to the program’s run-time. Then by using an analysis named static value-slicing, we short-cut some of the dynamic parts so that they are solved by simple membership tests for the program’s input. This re-formulation accelerates the analysis; once the program’s input occurs the prepared dynamic parts can immediately and simultaneously start to resolve. Every step of our technique is formally defined and proven correct.