TAU targets a general complex-system computation model. It is capable of considering multiple simultaneous thread contexts on multiple nodes simultaneously, including parallel execution as with MPI. Its layered architecture takes into account system-level, software-level and communication-level events, providing abstraction layers where appropriate to facilitate efficient measurement and/or analysis. TAU's instrumentation API interfaces with each step in the compilation and execution phase to provide comprehensive data. This includes execution in a virtual environment that provides low-level access to hardware runtime execution behavior. The user is able to customize these interfaces by providing additional library modules and utilizing the associated API to gather, report and analyze data according to user-specified controls.

 
©   Cornell University  |  Center for Advanced Computing  |  Copyright Statement  |  Inclusivity Statement