Visual Amortization Analysis of Recompilation Strategies

Dynamic recompilation tries to produce more efficient code by exploiting runtime information. Virtual machines like the Jikes RVM use recompilation heuristics to decide how to recompile the program, i.e. what parts are recompiled at what level of optimization. In this paper we present our post-mortem amortization analysis based on improved call stack sampling. Our tool presents the results of the analysis as an interactive visualizations to help both virtual machine implementors improve their recompilation strategies, as well as programmers assess whether these recompilation strategies pay off not only for their application as a whole, but also for individual methods.