Speculative alias analysis for executable code

Optimizations performed at link time or directly applied to final program executables have received increased attention in recent years. Such low-level optimizations can benefit greatly from pointer alias information. However as almost all existing alias analyses are formulated in terms of source language constructs, they turn out to be of limited utility at the machine code level. This paper describes two different approaches to high-quality, low-cost, speculative may-alias analysis, to be applied in the context of link-time or executable code optimizers. The key idea behind our proposals is the introduction of unsafe speculations at analysis-time, which increases alias precision on important portions of code, and keeps the analysis reasonably cost-efficient. Experimental results indicate that introducing speculation at analysis-time is clearly beneficial: precision increases up to 83% in average, against a baseline precision of 16%. Furthermore, the percentage of dynamic misspeculations is typically about 2%, which shows that our technique can be used even for scenarios where speculation recovery is expensive.