Research on Program Execution Time Assessment of LLVM IR Program

摘要

Program performance assessment is one of important methods to optimize a system design. The assessment of program execution time is always key topic in computer structure. A good assessment can provide important measure basis for hardware/software automatic partitioning in reconfigurable computer compiler. This paper analyzes the characteristics of IR program and proposes an IR layer program classification algorithm for assessing the program execution time. The method can provide an accurate measure and can be constructed into a BP neural network assessment system. The experimental results show that the proposed assessment model has lower assessment deviation compared to other models under the same conditions. The BP neural network model can be trained.