Integrating Static and Dynamic Analysis Techniques for Detecting Dynamic Errors in MPI Programs

Message passing interface is the de-facto standard for programming high performance computing applications and it is ready for scaling to extreme scale system with millions of nodes and billions of cores. With this huge number of components MPI will be error prone. Many types of errors can occur with MPI implementation such as deadlock. Testing tools can assist application developers in the detection of such errors. This thesis provides an integration of static and dynamic analysis to enable a precise testing of MPI applications. This paper presents a new algorithm that is able to detect deadlock for point –to-point communication in two phases. By using static phase the overhead in this dynamic phase will limited. The experimental results show that our verification tool verifies several benchmarks and finds deadlock.