Visualizing application behavior on superscalar processors

摘要: The advent of superscalar processors with out-of-order execution makes it increasingly difficult to determine how well an application is utilizing the processor and adapt improve its performance. We describe a visualization system for analysis behavior on processors. Our provides overview-plus-detail display application's execution. A timeline view pipeline performance data shows overall utilization pipeline. This information displayed using multiple time scales, enabling user drill down from high-level overview focus region hundreds cycles. interest in detail animated cycle-by-cycle instructions are reordered executed functional units being utilized. Additional context views correlate instuctions this detailed relevant source code application. allows discover root cause poor make changes can be easily configured variety models configurations. demonstrate both MXS MMIX models.