Collaborative Research: CSR---AES: A Framework for Optimizing Scientific Applications

点击翻译按钮获取中文摘要

Title:Collaborative Research: CSR---AES: A Framework for Optimizing Scientific ApplicationsThe Design Optimizer for Scientific Applications (DOSA) frameworkallows the programmer or compiler writer to explore alternativedesigns and optimize for speed (or power) at design-time and use itsrun-time optimizer as an automatic application composition system(ACS) that constructs an efficient application that dynamically adaptsto changes in the underlying execution environment based on the kernelmodel, architecture, system features, available resources, andperformance feedback. DOSA allows design-time exploration andautomatic run-time optimizations using continuous performanceoptimizations (CPO) so that application programmers and compilerwriters are relieved from the challenging task of optimizing thecomputation in order to achieve high performance. As an illustrationof the DOSA framework, one complex, full application is optimized forIBM Cell. The innovative performance optimization techniques for thememory hierarchy use new techniques for reducing I/O complexity, datalayout, data remapping, and in-memory processing, and are supported byDOSA, the semi-automatic design framework and dynamic run-timesystem. This framework allows rapid, high-level performance estimationand detailed low-level simulation by incorporating high-levelperformance models into the model-integrated computing framework. Therun-time system dynamically improves application performance using thecomponent library, the models, and the run-time optimizer. Theapplication studies are chosen for their broad impact to traditionaland emerging scientific areas such as bioinformatics, computationalbiology, and medical applications, as well as for national security.The project especially encourages the participation by women,minorities, and underrepresented groups.