应用定制可重构流水线的动态存储管理
发布时间：2018-07-16 18:47
【摘要】：可重构体系结构因为丰富的硬件资源和可重构互连网络,可以提供接近专用集成电路的性能和接近通用处理器的灵活性.首先介绍了一种面向领域应用的应用定制可重构流水线体系结构,可带来计算性能的显著提升;其次介绍了面向应用定制可重构流水线的存储管理方法,存储问题对可重构体系结构性能发挥至关重要.该方法基于配置信息动态管理数据输入、地址计算与访存操作分离,可以实现多种寻址方式,使在硬件上运行的硬件流水线高效流动.实验证明,相对于设置专用访存单元的可重构系统,节省了计算资源同时保证了流水线效率.
[Abstract]:Because of the abundant hardware resources and reconfigurable interconnection network, the reconfigurable architecture can provide close to the performance of ASIC and the flexibility of universal processor. Firstly, a domain oriented application customization reconfigurable pipeline architecture is introduced, which can significantly improve computing performance. Secondly, the storage management method for application customization reconfigurable pipeline is introduced. Storage problems are critical to the performance of reconfigurable architectures. This method is based on configuration information dynamic management data input, address calculation and memory access operation separation, can achieve a variety of addressing methods, so that the hardware running on the hardware pipeline efficient flow. Experiments show that compared with the reconfigurable system with special memory access unit, the computational resources are saved and the pipeline efficiency is guaranteed.
【作者单位】： 国防科学技术大学计算机学院;
【基金】：国家自然科学基金面上项目(61076020)
【分类号】：TP333
本文编号：2127303
[Abstract]:Because of the abundant hardware resources and reconfigurable interconnection network, the reconfigurable architecture can provide close to the performance of ASIC and the flexibility of universal processor. Firstly, a domain oriented application customization reconfigurable pipeline architecture is introduced, which can significantly improve computing performance. Secondly, the storage management method for application customization reconfigurable pipeline is introduced. Storage problems are critical to the performance of reconfigurable architectures. This method is based on configuration information dynamic management data input, address calculation and memory access operation separation, can achieve a variety of addressing methods, so that the hardware running on the hardware pipeline efficient flow. Experiments show that compared with the reconfigurable system with special memory access unit, the computational resources are saved and the pipeline efficiency is guaranteed.
【作者单位】： 国防科学技术大学计算机学院;
【基金】：国家自然科学基金面上项目(61076020)
【分类号】：TP333
【相似文献】
相关期刊论文 前3条
1 赖兆磬;潘明;许勇;张辉;;嵌入式五级流水线CPU核的设计与实现[J];微计算机信息;2008年29期
2 何虎,孙义和;面向寄存器的流水线处理器建模及验证方法[J];半导体学报;2003年01期
3 ;[J];;年期
相关重要报纸文章 前1条
1 MIPS科技公司工程总监 VidyaRajagopalan;MIPS:性能与面积完美结合[N];中国计算机报;2007年
相关硕士学位论文 前1条
1 王英茂;基于流水线处理的RISC微控制器软核的设计[D];中南大学;2007年
，本文编号：2127303
本文链接：https://www.wllwen.com/kejilunwen/jisuanjikexuelunwen/2127303.html
