Towards Exascale Applications
Task-based programming models require new debugging strategies.
Temanejo allows to visualize parallel applications as a graph of inter-dependent units of work, so called tasks, and to step through an application taskwise rather than concentrating on threads or processes. This is achieved with the help of the communication library Ayudame, which is distributed together with Temanejo. We aim at supporting any programming model that can meaningfully define the concepts of task and dependencies between them. Notably, this includes GPGPU‘s, OpenMP and MPI, i.e. the working horses in high-performance computing.
TEMANEJO allows to:
- connect to a task parallel application and
- analyze and
- interact with
the task dependency graph while the program is running.
TEMANEJO works with:
- and more programming models to come.
TEMANEJO is continuously being developed in the projects TEXT, H4H and MontBlanc2.
The source and documentation is available at:
Dr. José Gracia
Höchstleistungsrechenzentrum Universität Stuttgart
Nobelstraße 19, 70569 Stuttgart, Germany