Principal Investigator: Michael Lang, Los Alamos National Laboratory
The aim of this project is to deliver programming model technologies to support the Ristra Next Generation Code and other ASC mission applications. This work includes adding features and the integration of the Legion programming model into higher level libraries/frameworks such as the Flexible Computational Science Infrastructure (FleCSI). Current activities are addressing fundamental scalability and performance bottlenecks and meeting the needs of dynamic multi-physics applications. Through the use of runtime-enabled (dynamic) analysis, the Legion programming system can achieve increased levels of parallelism, improved machine utilization and throughput, and aspects that help address performance portability.
Legion-centric efforts are focused on developing and integrating new capabilities such as dynamic control replication, which enables applications to be written with apparently sequential semantics and parallelize and scale to exascale systems. This and other capabilities under development are critical to complex applications that often require multiple mesh representations, different discretization strategies, and support for multiple materials in a single application. The end result of our work is development of new features, assistance and support for the integration of these features within ATDM applications, and demonstration of new capabilities for broader use cases. As we coordinate this feature development with other ECP application stakeholders through the ECP Legion project, these core features are expected to be beneficial to Combustion Pele, E3SM-MMF, and ExaFEL application projects.