Execução de funções parciais em linguagem de programação C

Abstract

Currently the computer science community has several powerful programs for performing dynamic analysis. Examples of these analyzers are: Valgrind, useful for performing several types of checks related to the running of programs; and aprof, used to aid in the search for asymptotic inefficiencies. However, these tools expect to receive as an input an executable program. This fact complicates the analysis of specific code snippets, such as functions. Considering this difficulty, in this dissertation we present a set of techniques that, together, enable us to execute isolated functions written in C language automatically. In particular, we focus on methods that deal with access to arrays by indexing. Besides that, it is our goal that, when executed, a target function does not contain invalid memory accesses caused by the data automatically generated to test it. With the ideas presented here, we developed a tool, amazonc, which enable us to execute functions in C isolated from the rest of the code. Experiments performed with the kernels of the Polybench benchmarks suite reinforce the effectiveness of the method presented. Together with aprof, amazonc was able to reconstruct the asymptotic complexity curves of all core functions in that benchmark. In addition, the process of reconstructing these curves did not lead to memory errors, as Valgrind has certified.