CAREER: Compiler-Assisted Data Adaptation

University Of Rochester, Rochester NY

Investigators

Abstract

Data management has become a core function of computing, strongly influencing not only performance but also energy consumption, program security, fault tolerance, and other program properties. While many programs use a large amount of data, they do not access all data at all times. Hence, data adaptation can improve cache efficiency by identifying and compacting precisely those data required at a given point of computation. Such fine-grained and dynamic data management may significantly alleviate the bottleneck on communication and strengthen the control of data. This research studies fine-grained data adaptation through a compiler-based approach. It develops three compiler-assisted techniques. First, program monitoring analyzes selected data; then, locality analysis identifies their group locality; and finally, data transformation changes their layout while maintaining program correctness using run-time data maps. The new system improves cache optimization for data-intensive applications on conventional machines and memory management for general-purpose programs on embedded devices. In parallel to research, the investigator develops data-analysis tools to support teaching of memory issues in computer organization, compiler, and programming courses. The investigator also develops an advanced course that focuses on memory optimization techniques at all system levels.