SHF: Small: Collaborative Research: Lighthouse: Resource-Aware Advisor for High-Performance Linear Algebra

点击翻译按钮获取中文摘要

This research will study ways to ease the production anduse of high-performance matrix algebra software. Matrix algebracalculations constitute the most time-consuming part of simulationsin diverse fields, and lowering the runtimes of those computationscan have a significant impact on overall application performance. Theprocess of converting matrix algebra from algorithm to high-qualityimplementation is, however, a complex one. At each step, the codedeveloper is confronted with a myriad of possibilities, many requiringexpertise in numerical computation, mathematical software, compilers,and computer architecture. In response to these difficulties, the PIshave developed a prototype taxonomy implementation entitled Lighthouse,which is a guide to the linear system solver routines from the softwarepackage LAPACK. It is the first framework that combines a matrix algebrasoftware ontology with code generation and tuning capabilities. Itsinterface is designed for users across a spectrum of disciplines, careerlevels, and programming experience.The PIs will dramatically extend the Lighthouse framework in a number of new directions. First, they will construct a general taxonomy of software that can be used to buildhighly-optimized mathematical applications. The taxonomy will initiallyprovide an organized ontology of software components for high-performancematrix algebra and later other numerical software from a variety ofproblem domains. It will serve as a guide to practitioners seekingto learn what is available for their mathematical programming tasks,how to use it, and how the various parts fit together. Second, the PIswill apply a combination of source code analysis and machine learningtechniques to fully automate the generation of parameterized models that,given representative inputs and a simple architecture description, can beevaluated to identify methods from various libraries that best reflectthe user's resource and performance requirements. This automation iscritical for ensuring that the taxonomy is comprehensive enough to beuseful and that it accurately reflects the features and performanceof the latest versions of numerical libraries. Finally, the PIs willadvance the state-of-the-art in tuning tools by improving some of thetools included in the taxonomy, broadening their ranges of functionalityin terms of problem domains and languages. This project will producethe following impacts: greater performance by applications, enablingboth more discovery with available computing resources and greaterproductivity of application programmers; greater understanding of theinteraction between architecture and algorithms; and an educational toolfor future computational scientists.