Collaborative Research: S12-SSI: Task-Based Environment for Scientific Simulation at Extreme Scale (TESSE)

This project, TESSE, is developing a new-generation programming environment that uniquely address the needs of emerging computational models in chemistry and other fields, and allowing these models to reap the benefits of the computer hardware of tomorrow. The project thus is bringing closer the ability to predict properties of matter and design matter transformations of importance to the technological leadership and energy security of the US. TESSE's impact has scientific, cyberinfrastructure, and interdisciplinary educational aspects. TESSE helps carry widely used scientific applications (NWChem, MADNESS, MPQC, etc.) into a whole new generation of many-core and accelerated architectures, while at the same time dramatically improving programmer productivity. In terms of cyberinfrastructure, TESSE's PaRSEC runtime system delivers capabilities that many libraries and advanced applications will require for high performance on large scale and hybridized systems. The goals of this project, Task-based Environment for Scientific Simulation at Extreme Scale (TESSE), are to design and demonstrate via substantial scientific simulations within chemistry and other disciplines a prototype software framework that provides a groundbreaking response to the twin problems of portable performance and programmer productivity for advanced scientific applications on emerging massively-parallel, hybrid, many-core systems. TESSE will create a viable foundation for a new generation of science codes, one which enables even more rapid exploration of new physical models, provides greatly enhanced performance portability through directed acyclic graph (DAG) scheduling and auto-tuned kernels, and works towards full interoperability between major chemistry packages through compatible runtimes and data structures. TESSE will mature to become a standard, widely available, community-based and broadly-applicable parallel programming environment complementing and rivaling MPI/OpenMP. This is needed due to the widely appreciated shortfalls of existing mainstream programming models and the already huge successes of the existing DAG-based runtimes that are the foundation of the next generation of NSF- and DOE-supported (Sca)LAPACK high-performance linear algebra libraries.