Data scheduling on processor-in-memory arrays based on data placement and data movement.pdfVIP

下载本文档

4
0
约5.1万字
约 30页
2017-04-11 发布于江苏
举报
版权申诉

Data scheduling on processor-in-memory arrays based on data placement and data movement.pdf

1、原创力文档（book118）网站文档一经付费（服务费），不意味着购买了该文档的版权，仅供个人/单位学习、研究之用，不得用于商业用途，未经授权，严禁复制、发行、汇编、翻译或者网络传播等，侵权必究。。
2、本站所有内容均由合作方或网友上传，本站不对文档的完整性、权威性及其观点立场正确性做任何保证或承诺！文档内容仅供研究参考，付费前请自行鉴别。如您付费，意味着您自己接受本站规则且自行承担风险，本站不退款、不进行额外附加服务；查看《如何避免下载的几个坑》。如果您已付费下载过本站文档，您可以点击这里二次下载。
3、如文档侵犯商业秘密、侵犯著作权、侵犯人身权等，请点击“版权申诉”（推荐），也可以打举报电话：400-050-0827(电话支持时间：9:00-18:30)。
4、该文档为VIP文档，如果想要下载，成为VIP会员后，下载免费。
5、成为VIP后，下载本文档将扣除1次下载权益。下载后，不支持退款、换文档。如有疑问请联系我们。
6、成为VIP后，您将拥有八大权益，权益包括：VIP文档下载权益、阅读免打扰、文档格式转换、高级专利检索、专属身份标志、高级客服、多端互通、版权登记。
7、VIP文档为合作方或网友上传，每下载1次，网站将根据用户上传文档的质量评分、类型等，对文档贡献者给予高额补贴、流量扶持。如果你也想贡献VIP文档。上传文档

Data scheduling on processor-in-memory arrays based on data placement and data movement

Data Scheduling on Processor-In-MemoryArrays Based on Data Placement and DataMovementYi Tian Edwin H.-M. Sha Chantana ChantrapornchaiPeter M. KoggeDept. of Computer Science and EngineeringUniversity of Notre DameNotre Dame, IN 46556AbstractIn the study of PetaFlop project, Processor-In-Memory array was proposed to bea target architecture in achieving 1015 oating point operations per second computingperformance. However, one of the major obstacles to achieve the fast computing wasinterprocessor communications, which lengthen the total execution time of an appli-cation. A good data scheduling, consisting of nding initial data placement and datamovement during the run-time, can give a signicant reduction in the total commu-nication cost and the execution time of the application. In this paper, we proposeecient algorithms for the data scheduling problem. Experimental results show theeectiveness of the proposed approaches. Compared with default data distributionmethods such as row-wise or column-wise distributions, the average improvement forthe tested benchmarks can be up to 30%. This work was partially supported by NSF MIP 95-01006 and NSF ACS 96-12028.1 1 IntroductionSuper parallel architectures are studied in the PetaFlop project in order to achieve 1015 oat-ing point operations per second. As one of the design point teams for the PetaFlop project,we propose the Processor-In-Memory (PIM) architecture to achieve the desired PetaFlopperformance. The PIM technology is used to avoid the \traditional Von Neuman problemso as to enhance the computing capability [8]. PIM, the combination on one chip of bothdense DRAM memory and signicant amounts of logic, can place computing either rightnext to, or even inside of, the memory macros, where there are huge amounts of raw mem-ory bandwidth. One of the biggest obstacles encountered during the project developmentis how to nd the best way to place data and move data in the PIM array in order tominimize communications [20]