{"title":"Optimal loop scheduling with register constraints using flow graphs","authors":"Jan Müller, D. Fimmel, R. Merker","doi":"10.1109/ISPAN.2004.1300478","DOIUrl":null,"url":null,"abstract":"We present a novel loop scheduling approach using a generalized flow graph model of the resource constraints. From this model we derive a new flow graph to incorporate register constraints. Our linear programming implementation produces an optimum loop schedule, respecting the constraints on functional units and registers in a single optimization problem. Moreover, the iteration interval is treated as a rational number, and the approach supports heterogeneous processor architectures and pipelined functional units. Compared to earlier approaches, the solution can reduce the problem complexity and solution time, and provide faster loop schedules.","PeriodicalId":198404,"journal":{"name":"7th International Symposium on Parallel Architectures, Algorithms and Networks, 2004. Proceedings.","volume":null,"pages":null},"PeriodicalIF":0.0000,"publicationDate":"2004-05-10","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":"10","resultStr":null,"platform":"Semanticscholar","paperid":null,"PeriodicalName":"7th International Symposium on Parallel Architectures, Algorithms and Networks, 2004. Proceedings.","FirstCategoryId":"1085","ListUrlMain":"https://doi.org/10.1109/ISPAN.2004.1300478","RegionNum":0,"RegionCategory":null,"ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":null,"EPubDate":"","PubModel":"","JCR":"","JCRName":"","Score":null,"Total":0}
引用次数: 10
Abstract
We present a novel loop scheduling approach using a generalized flow graph model of the resource constraints. From this model we derive a new flow graph to incorporate register constraints. Our linear programming implementation produces an optimum loop schedule, respecting the constraints on functional units and registers in a single optimization problem. Moreover, the iteration interval is treated as a rational number, and the approach supports heterogeneous processor architectures and pipelined functional units. Compared to earlier approaches, the solution can reduce the problem complexity and solution time, and provide faster loop schedules.