Loops in a software application are a crucial construct that heavily drives resource consumption, both in time and memory. For example, an improper loop boundary may have severe ripple effects, resulting in resource consumption anomalies. To exacerbate this situation, code execution profilers completely hide loops from their report, to instead focus on the method context tree. The BonaFide C analyzer combines source code static analysis with profiling information to generate an extended report about loops. The analyzer aims to give hints on code parallelization: it highlights loops that may be easily parallelized. The approach is validated using the Standard Performance Evaluation Corporation (SPEC) CPU 2006 benchmark.

Reifying loops by an execution profiler in the way promoted in the paper is new and has not been considered as far as I know. Characterizing loops according to the variable they access is key to parallelizing an application that was not initially prepared for this.

On the downside, the use of Extensible Markup Language (XML) is presented as a key point of the approach, which I don’t think it should be. The use of XML cannot be more than a technical matter that should not support the (valuable) scientific claim made by the paper.

Researchers and practitioners interested in program optimization, parallelism, and collection libraries will find the paper interesting. It is pleasant and easy to read, and I recommend reading it.