Static Path Execution Frequency Inference

Knowledge of the frequency with which the control flow paths of a program tend
to executed can prove to be extremely useful for a variety of different
analyses, especially for supporting optimizing transformations. In current
practice, this information is sometimes mined from execution traces, but more
often is simply unavailable. In this paper, we show how machine learning
techniques can be used to statically estimate the run-time frequency with which
control flow paths are taken.