SHF: EAGER: Platform-Agnostic Supercomputing from Scientific Metaprogramming

University Of Utah, Salt Lake City UT

Investigators

Abstract

Porting scientific codes to new high-performance platforms is a time consuming and error-prone process. The research funded by this EAGER award will make scientific codes more portable by raising the level of abstraction. The Principal Investigators will develop a domain-specific language embedded in C++ for solving partial differential equations. They will evaluate the effectiveness of their approach by applying their language to the problem of reacting flow simulations, which is a problem that is important for modeling zero-emission power plants. Computation is an increasingly important tool for scientists, but scientists must hand-tune their code for each new machine to take advantage of the latest high-performance hardware. This research represents a new approach to implementing scientific codes on high-performance machines. The approach allows domain experts to write code at a high level of abstraction while gaining portability across parallel platforms. This research has the potential to improve the productivity of computational scientists by transforming the way that scientific codes are developed.