Multi-Chain Prefetching Effective Exploitation of Inter-Chain.pdfVIP

Appears in Proceedings of the 10th Annual International Conference on Parallel Architectures and Compilation Techniques, Barcelona, Spain , September 2001. Multi-Chain Prefetching: Effective Exploitation of Inter-Chain Memory Parallelism for Pointer-Chasing Codes Nicholas Kohout Seungryul Choi Dongkeun Kim, Donald Yeung Intel Corp. Computer Science Dept. Electrical Computer Eng. Dept. Univ. of Maryland, College Park Univ. of Maryland, College Park nicholas.j.kohout@ choi@ fdongkeun,yeungg@ Abstract 1. Introduction Pointer-chasing applications tend to traverse composed data Prefetching, whether using software [11, 7, 1], hardware [3, structures consisting of multiple independent pointer chains. 13, 5], or hybrid [2, 4] techniques, has proven successful at hid- While the traversal of any single pointer chain leads to the se- ing memory latency for applications that employ regular data rialization of memory operations, the traversal of independent structures (e.g. arrays). Unfortunately, these techniques are far pointer chains provides a source of memory parallelism. This pa- less successful for applications that employ linked data structures per presents multi-chain prefetching, a technique that utilizes off- (LDSs) due to the memory serialization effects associated with line analysis and a hardware prefetch engine to prefetch multiple LDS traversal, known as the pointer chasing problem . The m