Reservoir Labs, Inc. — Department of Energy STTR Phase I: Calculations undertaken by physicists for solving fundamental problems such as Lattice Qua

Description

Calculations undertaken by physicists for solving fundamental problems such as Lattice Quantum Chromo Dynamics (LQCD) are prohibitively demanding without expertly optimized software. Such software is very complex and costly to write and maintain without automated tools, especially for advanced heterogeneous architectures. PUMA-V provides these automated tools, an optimization and visualization toolchain to meet this complexity challenge and to increases user productivity. The Polyhedral User Mapping Assistant and Visualizer (PUMA-V) addresses this problem starting from existing high-level technology for automatic, mixed static/dynamic, mapping of dense multi-linear algebra algorithms. Analytical optimizations balance objectives modeling parallelization, locality, contiguity, and data layout within a unified framework based on a sound mathematical model. Thanks to automatic mapping technology, the problem of automatic efficient code generation can be attacked starting from high-level productivity languages. PUMA-V will leverage the existing USQCD software stack which revolves around templated C++ code. Users of the system will be able to visualize complex optimization tradeoffs between conflicting metrics such as the amount of parallelism in the program or the amount of locality in the program as the code is transformed by the toolchain for execution on heterogeneous hardware such as clusters of GPUs. The overall objective of PUMA-V will be the development of this set of tools for automated source-to-source transformation, program optimization enhanced by tools for visualization of these transformations and capabilities for human-in-the-loop interaction. Commercial Applications and Other Benefits: The commercial applications of this technology exist in for applications in the oil and gas industry (seismic imaging, optimizing reverse time migration simulations), electronics industry (modeling electromagnetic propagation), and materials industry (modeling the properties of new materials or of processes such as combustion). The resulting benefits include more effective US energy production, new products, and energy efficiency.