摘要
针对传统基于模块描述建模方法在性能估算和验证的精确性、灵活性和设计迭代方面的不足,围绕资源分配、性能参数标注和仲裁算法指定等,提出基于事务数据流的系统芯片(SoC)性能建模方法.该方法通过加入Sys-temC的时序控制机制的TDFLib的C++泛型库,以及描述SoC系统模型的性能建模语言(PML)的域特定语言来实现.PML源文件可生成包含TDFLib调用的C++代码,再经过编译并与可复用仿真程序框架链接得到可执行的系统性能分析模型.该模型在SystemC内核控制下进行周期精确的仿真,运行结果保存在MySQL数据库中.结果表明该性能建模方法能够提高架构设计与分析的效率.
Traditional modeling method based on module descriptions is neither accurate nor flexible and also expansive in design iteration.A transaction dataflow based performance modeling（TBPM） methodology for SoC was proposed with data flow specification,resource allocation,performance parameter annotation,and arbitration algorithm specification.The TBPM methodology was implemented by a C＋＋ generic library named TDFLib containing SystemC timing controlling mechanism,and a domain specific language named PML to describe SoC performance model.The PML source code could be used to generate C＋＋ code that includes TDFLib API call,and then the C＋＋ code was compiled and linked with reusable simulation framework to construct an executable model.Cycle-accurate simulation was performed with this model under control of SystemC kernel,and all runtime status was dumped into MySQL database for post analysis.A experiment demonstrated that the TBPM methodology can improve efficiency of architectural design and analysis.
出处
《浙江大学学报（工学版）》
EI
CAS
CSCD
北大核心
2011年第2期314-322,共9页
Journal of Zhejiang University：Engineering Science
基金
国家"863"高技术研究发展计划资助项目(2005AA1Z1260)
国家自然科学基金资助项目(60720106003)
关键词
集成电路
事务数据流
性能分析建模
设计空间探查
架构设计
域特定语言
integrated circuits
transaction dataflow
performance analysis modeling
design space exploration
architecture design
domain specific language
作者简介
孟昕（1981-），男，浙江临安人，博士生，从事数字超大规模集电路设计研究工作．E—mail：mengxin@vlsi．zju．edu．cn
通信联系人：沈海斌，男，副教授．E-mail：shb@vlsi．zju．edu．cn
