Hierarchical Phase Analysis for Sampling Simulations

Weihua Zhang1, Qiang Yan1, Binyu Zang1 and Pen-Chung Yew2

1Parallel Processing Institute, Fudan University
2University of Minnesota

Since most programs exhibit cyclic behavior patterns, representative sampling simulation has been widely used to reduce simulation time. Generally, it selects a small representative portion of program execution intervals for detailed simulation, while functionally simulating the remaining portion. Although the granularity of those execution intervals has been one of the most important parameters in a representative sampling simulation method, to our best knowledge, there has been very little research on it. Most researchers simply applied a fine-grained strategy to reduce detailed simulation time. However, in this paper, we show that coarse-grained phase analysis should be involved in a representative sampling method instead of only applying a fine-grained phase analysis. The reasons are that: First, a coarse-grained phase analysis can be applied to optimize the functional simulation time, which has become the most time-dominant portion in a fine-grained representative sampling simulation method; Furthermore, through selecting finer-grained simulation points within coarser-grained phases, the detailed simulation time can be reduced. Based on the above observation, we design and implement a new multi-level phase analysis framework for representative sampling simulation. Experimental results show such a framework is very effective. It achieves a speedup of 9X over the SimPoint method while still maintaining comparable accuracy.

Long Abstract

Back to Program