Design and analysis of adaptive processor

Author:

Takano Shigeyuki1

Affiliation:

1. Sanyo LSI Design System Soft Co., Ltd.

Abstract

A new computation model called CACHE (Cache Architecture for Configurable Hardware Engine) is proposed in this paper. This model does not require a dedicated host processor and its software to harness the reconfiguration. Autonomous reconfiguration is performed within a working-set of application datapaths. The CACHE model has lots of side effects; caching, resource allocation and assignment, placement and routing, and defragmentation, with a processing array itself and a special register called a working-set register file. The model aims to reduce three major workloads: (1) the processor and application design workload, (2) runtime resource management and scheduling workload, and (3) reconfiguration workload. In order to reduce these workloads, processor architecture is definitely different from traditional computing model and its microprocessor architecture. There are three major ideas to construct the computing system: (1) an on-chip working-set model mainly in order to control load and store of streams, namely to control traffics introducing overheads, (2) an on-chip deadlock properties model mainly in order to manage resources and to continuously configure datapaths corresponding to a working-set window, (3) a cache memory technique to work for these models, the mechanism is equivalent to the working-set window, and the cache memory's procedure is equivalent to resource request, acquirement, and release of deadlock properties. The first model focuses onto streaming applications, for example vector and matrix operations, filters, and so on, which takes coarser grained operations such as integer operations of C-language. Regarding performance compared with DSPs, that comes from constant throughput across different scale of the applications. In addition, extended model, we call Instant model that automatically generates instance of a datapath, outperforms the DSPs. This paper shows its computation model, architecture, low-level design, and analyses about basic characteristics of the execution.

Publisher

Association for Computing Machinery (ACM)

Subject

General Computer Science

Reference56 articles.

1. Asaovic K. 1998. Vector microprocessors. Ph.D. thesis University of California Berkeley. Asaovic K. 1998. Vector microprocessors. Ph.D. thesis University of California Berkeley.

2. Bobda C. 2007. Introduction to Reconfigurable Computing: Architectures Algorithms and Applications. Springer. Bobda C. 2007. Introduction to Reconfigurable Computing: Architectures Algorithms and Applications. Springer.

Cited by 4 articles. 订阅此论文施引文献 订阅此论文施引文献,注册后可以免费订阅5篇论文的施引文献,订阅后可以查看论文全部施引文献

1. An analysis on power consumption and performance in runtime hardware reconfiguration;International Journal of Embedded Systems;2021

2. Performance Scalability of Adaptive Processor Architecture;ACM Transactions on Reconfigurable Technology and Systems;2017-04-11

3. Very Large-Scale Integrated Processor;International Journal of Networking and Computing;2013

4. Very Large-Scale Integrated Processor;2012 IEEE 26th International Parallel and Distributed Processing Symposium Workshops & PhD Forum;2012-05

同舟云学术

1.学者识别学者识别

2.学术分析学术分析

3.人才评估人才评估

"同舟云学术"是以全球学者为主线,采集、加工和组织学术论文而形成的新型学术文献查询和分析系统,可以对全球学者进行文献检索和人才价值评估。用户可以通过关注某些学科领域的顶尖人物而持续追踪该领域的学科进展和研究前沿。经过近期的数据扩容,当前同舟云学术共收录了国内外主流学术期刊6万余种,收集的期刊论文及会议论文总量共计约1.5亿篇,并以每天添加12000余篇中外论文的速度递增。我们也可以为用户提供个性化、定制化的学者数据。欢迎来电咨询!咨询电话:010-8811{复制后删除}0370

www.globalauthorid.com

TOP

Copyright © 2019-2024 北京同舟云网络信息技术有限公司
京公网安备11010802033243号  京ICP备18003416号-3