Trace Instrumentation and Measurement (TAU)

An online trace analysis system for parallel applications depends on two important components: a parallel performance measurement system and a parallel trace analysis tool. The TAU performance system is a toolkit for performance instrumentation, measurement and analysis of parallel programs. It targets a general computational model consisting of shared-memory compute nodes where multiple contexts reside, each providing a virtual address space shared by multiple threads of execution. TAU supports a flexible instrumentation model that applies at different stages of program compilation and execution [3]. The instrumentation targets multiple code points, provides for mapping of low-level execution events to higher-level performance abstractions, and works with multi-threaded and message passing parallel computation models. Instrumentation code makes calls to the TAU measurement API. The measurement library implements performance profiling and tracing support for performance events occurring at function, method, basic block, and statement levels during program execution. Performance experiments can be composed from different measurement modules (e.g., hardware performance monitors) and measurements can be collected with respect to user-defined performance groups. The TAU data analysis and presentation utilities offer text-based and graphical tools to visualize the performance data as well as bridges to third-party software, such as Vampir [2] for sophisticated trace analysis and visualization.