E. Forbes, Zhenqian Zhang, Randy Widialaksono, Brandon H. Dwiel, Rangeen Basu Roy Chowdhury, V. Srinivasan, S. Lipa, E. Rotenberg, W. R. Davis, P. Franzon
{"title":"Under 100-cycle thread migration latency in a single-ISA heterogeneous multi-core processor","authors":"E. Forbes, Zhenqian Zhang, Randy Widialaksono, Brandon H. Dwiel, Rangeen Basu Roy Chowdhury, V. Srinivasan, S. Lipa, E. Rotenberg, W. R. Davis, P. Franzon","doi":"10.1109/HOTCHIPS.2015.7477478","DOIUrl":null,"url":null,"abstract":"This article consists of a single slide from the authors' conference presentation. Single-ISA Heterogeneous Multi-core: General purpose cores with different microarchitectures, tuned for different energy/performance points. Performance and energy of a program can be optimized by migrating among the core types as program characteristics change. Prior research has shown as much as a 50% improvement in energy when migrating every 1,000 cycles versus every 10,000 cycles. Such fine-grained thread migration requires very low migration overhead. We propose hardware support for fast thread migration. To migrate a thread, committed register values and the program counter must be moved from the source core to the destination core.","PeriodicalId":6666,"journal":{"name":"2015 IEEE Hot Chips 27 Symposium (HCS)","volume":"11 1","pages":"1-1"},"PeriodicalIF":0.0000,"publicationDate":"2015-08-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":"8","resultStr":null,"platform":"Semanticscholar","paperid":null,"PeriodicalName":"2015 IEEE Hot Chips 27 Symposium (HCS)","FirstCategoryId":"1085","ListUrlMain":"https://doi.org/10.1109/HOTCHIPS.2015.7477478","RegionNum":0,"RegionCategory":null,"ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":null,"EPubDate":"","PubModel":"","JCR":"","JCRName":"","Score":null,"Total":0}
引用次数: 8
Abstract
This article consists of a single slide from the authors' conference presentation. Single-ISA Heterogeneous Multi-core: General purpose cores with different microarchitectures, tuned for different energy/performance points. Performance and energy of a program can be optimized by migrating among the core types as program characteristics change. Prior research has shown as much as a 50% improvement in energy when migrating every 1,000 cycles versus every 10,000 cycles. Such fine-grained thread migration requires very low migration overhead. We propose hardware support for fast thread migration. To migrate a thread, committed register values and the program counter must be moved from the source core to the destination core.