Analyzing Program Behavior of SPECint2000 Benchmark Suite using Principal Components Analysis

Reducing simulation time during the phase of design of a microprocessor has been one of the key issues discussed in the community lately. This report tries to throw light on program behavior of the whole SPECint 2000 benchmark suite including reference and train input sets using a statistical technique called Principal Components Analysis. Analyzing the results of Principal Components Analysis and the raw data collected for the program-input pairs together helps to identify the similarity between benchmarks. It also helps to choose few program-input pairs and hence reduce simulation time. The project also talks about how we can associate program-input pairs to characteristics that are closely related to the performance of a microprocessor