Advancing scientific computation by improving scientific code development: Symbolic execution and semantic analysis

Abstract

This paper presents an implementation of a technique for automated, rigorous scientific program comprehension and error detection. The procedure analyzes fundamental semantic concepts during a symbolic execution of a user's code. Since program execution is symbolic, the analysis is general and can replace many test cases. The prototype of this procedure is demonstrated on two test cases including a 5k line of code (LOC) program. Although this technique promises a powerful tool, several challenges remain. © Springer-Verlag Berlin Heidelberg 2005.