Transaction level statistical analysis for efficient micro-architectural power and performance studies

In general, we lack in EDA industry tools and automated solutions in μArchitectural domain. In this paper, we elaborate on our attempt to advance performance simulation based statistical analysis techniques. On one hand, we utilize the content knowledge of μArchitectural specification (e.g., explicit specification of the major transactions), and on the other hand, the statistical compact representation of the simulation trace. We name the compact statistical modeling of transaction level performance simulation traces, Magenta (Modeling Agent for Transactional Analysis). As demonstrated by industrial case studies, Magenta can effectively capture all the sample flows that are represented in the simulation trace that exhibit the transaction of interest in terms of μArchitectural events in a statistical event dependency graph. Our industrial experience shows that Magenta is an effective statistical model for μArchitectural performance verification and power/performance trade-off.