Supporting parallel program debugging through control and data flow analysis

Abstract

Debugging summarizes the activities error detection and performance analysis of the software life-cycle. Both activities can be very tedious and difficult, especially if the user is confronted with parallel programs. As a result debugging tools have to provide a lot of support in order to improve execution analysis and program understanding. An example for such a tool in the domain of parallel message passing programs is CDFA, the control and data flow analyzer. Whereas most tools target at detection and display of faulty behavior, CDFA locates code that is originally responsible for incorrect behavior. This is achieved by a combination of dynamic execution data and static source code. Through visualization of control and data flow, complexity of the analysis is decreased and program comprehension is improved. With variable backtracking and program slicing, additional and powerful functionality for debugging is available.