A Source Code Analysis Method with Parallel Acceleration for Mining MPI Application Communication Counts

摘要

MPI(Message Passing Interface) applications are widely used to solve complex problems in the field of science and engineering. In addition to calculation information concerns, communication information is also an important focal point for MPI application analysis. The communication information is generally obtained through manual reviewing the source code, which is impractical for complex applications. Another way is to develop an instrumentation library and compile the target application with the library. Despite the communication information can be accurately obtained by this type of methods, the target application must be executed. To address these issues, in this paper, a source code analysis method is proposed to get the MPI applications communication information, which does not need to execute the target application and can work with general computers. Furthermore, a parallel strategy is presented to accelerate the source code analysis process. Two frequently-used communication patterns are used to evaluate the proposed method. The experimental results show that the method proposed in this paper not only can be ran on general computers, but also reduce the analysis overhead, and can also achieve efficiency advantages in the MPI applications with large message size.