CAREER: QoS Architecture for General Purpose Network Computing.

职业：通用网络计算的 QoS 架构。

• 批准号: 9875742
• 负责人: David Yau
• 金额: $ 20.57万
• 依托单位: Purdue Research Foundation
• 依托单位国家: 美国
• 项目类别: Continuing Grant
• 财政年份: 1999
• 资助国家: 美国
• 起止时间: 1999-03-15 至 2005-06-30
• 项目状态: 已结题
• 来源: https://www.nsf.gov/awardsearch/showAward?AWD_ID=9875742&HistoricalAwards=false
• 关键词: CAREER; QoS; Architecture; General; Purpose

9876117CAREER: Compiler Optimizations for Instruction BandwidthWilson C. HsiehThis project reexamines how compiler optimization affects one of thebottlenecks in commercial applications: the cost of fetchinginstructions. Standard compiler optimizations trade instructionbandwidth for faster instruction sequences, which works well forapplications that have good instruction locality. Unfortunately,recent studies have shown that instruction bandwidth is aperformance-limiting factor in many commercial applications, such asdatabases, circuit simulators, and multimedia programs. In addition,technology trends in processor architectures will also increaseapplications' demands for instruction bandwidth. The proposedresearch will explore several solutions to these problems. First, twonew compiler optimizations to improve utilization of the instructioncache, instruction cache scheduling and code factoring, will beinvestigated. Trace analysis of commercial applications will be usedto quantify their benefits. Second, traditional compileroptimizations will be reexamined and redesigned as necessary, with thegoal of making them sensitive to instruction cache utilization.Finally, architectural changes to support explicit cooperation betweenthe compiler and processor to make more efficient use of theinstruction cache will be explored. This research will provide abetter understanding of how compilers should optimize large,commercial applications, which is a field that has remained relativelyunexamined in the research community.

9876117职业：指令带宽的编译器优化Wilson C. Hsieh 该项目重新审视编译器优化如何影响商业应用中的瓶颈之一：获取指令的成本。 标准编译器优化以指令带宽换取更快的指令序列，这对于具有良好指令局部性的应用程序非常有效。 不幸的是，最近的研究表明，指令带宽是许多商业应用（例如数据库、电路模拟器和多媒体程序）中的性能限制因素。 此外，处理器架构的技术趋势也将增加应用程序对指令带宽的需求。 拟议的研究将探索这些问题的几种解决方案。 首先，将研究两个新的编译器优化，以提高指令缓存的利用率，即指令缓存调度和代码分解。 商业应用的痕量分析将用于量化其效益。 其次，传统的编译器优化将根据需要进行重新检查和重新设计，目标是使它们对指令缓存利用率敏感。最后，将探索支持编译器和处理器之间显式合作的架构更改，以更有效地利用指令缓存。 这项研究将更好地理解编译器应如何优化大型商业应用程序，这是一个在研究界相对未经审查的领域。