Exploiting On-chip Memory Bandwidth in the VIRAM Compiler

David Judd

Katherine Yelick

Christos Kozyrakis Stanford

David Martin

David Patterson

International Workshop on Intelligent Memory Systems (IMS), 2001


Abstract

This paper explores the VIRAM architecture from the perspective of compiler writers, describing code-generation challenges and their solutions in the VIRAM compiler. It focuses on exploiting high on-chip bandwidth for memory-intensive applications, including non-contiguous and unpredictable access patterns, and on supporting the narrow data types common in media processing.