关键词：GameTime；数据依赖；时序分析；时序重复性
摘 要：We present a technique for automatically learning a model of the data dependencies and encoding this model into the code under analysis for processing by GameTime. Using these extensions, we show that GameTime more accurately predicts the timing for a variety of benchmarks.Unfortunately, the complexity of modern architectures and platforms has made it very difficult to obtain accurate and efficient timing estimates. To deal with this, there have been recent proposals to re-architect platforms to make execution time of instructions more repeatable. There is however no systematic formalization of what timing repeatability means. In this thesis, we also propose formal models of timing repeatability. We give an algorithmic approach to evaluate parameters of these formal models. Using GameTime along with the data-dependent extensions discussed in this thesis, we objectively evaluate the timing repeatability of a representative sample of platforms with respect to a program of interest.