| ¡¡ | Chinese Journal of Computers Full Text |
| Title | A Metadata-Driven Optimization for Sampling Simulation |
| Authors | YAN Qiang ZHANG Wei-Hua LIU Li-Li ZANG Bin-Yu ZHU Chuan-Qi |
| Address | (Parallel Processing Institute, Fudan University, Shanghai 201203) |
| Year | 2008 |
| Issue | No.11(1986¡ª1994) |
| Abstract & Background | Abstract This paper analyzes the disadvantage of mainstream sampling simulation techniques using fixed-length samples and proposes a metadata-driven optimization for sampling simulation, BigLoopSP. In the approach, the compiler selects candidate loops and annotates the boundaries of those loops as metadata. Those metadata are used to divide the execution into varied-length candidate samples, for which each candidate sample corresponds to one iteration of the chosen loop. Since the program execution exhibits dynamic behaviors, the approach combines the knowledge from the metadata and the dynamic profiles to guide phase partition and selects simulation points for those phases. This approach effectively reduces the number of representative samples while preserving the good quality of them. So, compared with those mainstream sampling simulation techniques, such as SimPoint, our approach achieves better accuracy and reduces more simulation time (a speedup of 2.63X over SimPoint). Keywords compiler; simulator; metadata; sampling simulation. |