Universal Mechanisms for Data-Parallel Architectures
| Sorted by Date | Classified by Publication Type | Classified by Project |
Karthikeyan Sankaralingam, Stephen W. Keckler, William R. Mark, and Doug Burger. Universal Mechanisms for Data-Parallel Architectures. In Proceedings of the 36th Annual International Symposium on Microarchitecture, pp. 303-314, December 2003.
Download
[PDF]
Abstract
Data-parallel programs are both growing in importanceand increasing in diversity, resulting in specialized processors targeted at specific classes of these programs. Thispaper presents a classification scheme for data-parallel programattributes, and proposes micro-architectural mechanisms to supportapplications with diverse behavior using a single reconfigurablearchitecture.We focus on the following four broad kinds of data-parallel programs--- DSP/multimedia, scientific, networking, and real-time graphicsworkloads. While all of these programs exhibit high computationalintensity, coarse-grain regular control behavior, and some regularmemory access behavior, they show wide variance in the computationrequirements, fine grain control behavior, and the frequency of othertypes of memory accesses. Based on this study of applicationattributes, this paper proposes a set of general micro-architecturalmechanisms that enable a baseline architecture to be dynamicallytailored to the demands of a particular application. These mechanismsprovide efficient execution across a spectrum of data-parallelapplications and can be applied to diverse architectures ranging fromvector cores to conventional superscalar cores. Our results using a baseline TRIPS processor show that the configurability of the architecture to theapplication demands provides harmonic mean performance improvement of5%--55% over scalable yet less flexible architectures, and performs competitively against specialized architectures.
Additional Information
This is a test of the extra info broadcasting system.
BibTeX
@InProceedings{micro03, author = "Karthikeyan Sankaralingam and Stephen W. Keckler and William R. Mark and Doug Burger", title = "{Universal Mechanisms for Data-Parallel Architectures}", booktitle = "Proceedings of the 36th Annual International Symposium on Microarchitecture", year = 2003, pages = "303-314", month = {December}, abstract={Data-parallel programs are both growing in importance and increasing in diversity, resulting in specialized processors targeted at specific classes of these programs. This paper presents a classification scheme for data-parallel program attributes, and proposes micro-architectural mechanisms to support applications with diverse behavior using a single reconfigurable architecture. We focus on the following four broad kinds of data-parallel programs --- DSP/multimedia, scientific, networking, and real-time graphics workloads. While all of these programs exhibit high computational intensity, coarse-grain regular control behavior, and some regular memory access behavior, they show wide variance in the computation requirements, fine grain control behavior, and the frequency of other types of memory accesses. Based on this study of application attributes, this paper proposes a set of general micro-architectural mechanisms that enable a baseline architecture to be dynamically tailored to the demands of a particular application. These mechanisms provide efficient execution across a spectrum of data-parallel applications and can be applied to diverse architectures ranging from vector cores to conventional superscalar cores. Our results using a baseline TRIPS processor show that the configurability of the architecture to the application demands provides harmonic mean performance improvement of 5\%--55\% over scalable yet less flexible architectures, and performs competitively against specialized architectures. }, bib_dl_pdf = "http://www.cs.wisc.edu/~karu/docs/papers/utexas/micro03-dlp.pdf", bib_pubtype = {Refereed Conference}, bib_rescat = {proj-trips}, bib_extra_info = {This is a test of the extra info broadcasting system.} }
Generated by bib.pl (written by Patrick Riley ) on Sun Sep 26, 2021 16:14:29 time=1207019082