Scholarlink

This chapter introduces a novel cooperative coevolution framework for an effective Design Space Exploration (DSE) method that is used to search for optimal multiprocessor mappings associated with some signal processing and multimedia applications. The introduction section first gives an overview about these tasks that are specified by Synchronous Dataflow (SDF) models and their subsequent extension in the form of the Multi-Mode Dataflow (MMDF) models. The former problem of SDF mapping to a given set of heterogeneous processors is known to be NP-hard and widely studied in the design automation. The latter problem of MMDF mapping that allows a finite number of behaviours (modes), with each mode represented by an SDF graph, represents the setting of modern embedded applications that are more complex with dynamic mode changes over time. The multiprocessor mapping of an MMDF is far more challenging as the design space increases with the number of modes. The next section formulates the problem of MMDF mapping. This is then followed by a presentation of the cooperative coevolution framework to effectively explore the design space by a new problem-specific decomposition strategy whereby the solutions of node mapping for each individual mode are assigned to an individual population. The specific implementation of the Cooperative Coevolutionary GA (CCGA) that is aimed for improved search effectiveness includes (1) a problem-specific local search operator to supplement to the global search of CCGA and (2) a fitness approximation method and a hybrid fitness evaluation strategy that reduce the time consumption of fitness evaluation significantly. Computational studies are carried out to demonstrate the advantage of the proposed CCGA-based DSE method over the previous GA-based method. The proposed method can obtain an optimization result with better quality (2–3 times) using less optimization time (1/2–1/3). This chapter closes with further discussions of more complex settings found in the real-world that includes settings whereby the number of modes can be uncertain or bigger than what is considered in current studies.

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Design Space Exploration for Multimode Dataflow Mapping

  • 章节

  • 发表日期：

    2025年

查看全文
  • Xin Yao,
  • Siang Yew Chong

摘要

This chapter introduces a novel cooperative coevolution framework for an effective Design Space Exploration (DSE) method that is used to search for optimal multiprocessor mappings associated with some signal processing and multimedia applications. The introduction section first gives an overview about these tasks that are specified by Synchronous Dataflow (SDF) models and their subsequent extension in the form of the Multi-Mode Dataflow (MMDF) models. The former problem of SDF mapping to a given set of heterogeneous processors is known to be NP-hard and widely studied in the design automation. The latter problem of MMDF mapping that allows a finite number of behaviours (modes), with each mode represented by an SDF graph, represents the setting of modern embedded applications that are more complex with dynamic mode changes over time. The multiprocessor mapping of an MMDF is far more challenging as the design space increases with the number of modes. The next section formulates the problem of MMDF mapping. This is then followed by a presentation of the cooperative coevolution framework to effectively explore the design space by a new problem-specific decomposition strategy whereby the solutions of node mapping for each individual mode are assigned to an individual population. The specific implementation of the Cooperative Coevolutionary GA (CCGA) that is aimed for improved search effectiveness includes (1) a problem-specific local search operator to supplement to the global search of CCGA and (2) a fitness approximation method and a hybrid fitness evaluation strategy that reduce the time consumption of fitness evaluation significantly. Computational studies are carried out to demonstrate the advantage of the proposed CCGA-based DSE method over the previous GA-based method. The proposed method can obtain an optimization result with better quality (2–3 times) using less optimization time (1/2–1/3). This chapter closes with further discussions of more complex settings found in the real-world that includes settings whereby the number of modes can be uncertain or bigger than what is considered in current studies.