引用本文: | 刘磊,杨子煜,沈剑良,等.一种应用定制指令集可重构结构及FFT算法映射优化.[J].国防科技大学学报,2012,34(6):39-45.[点击复制] |
LIU Lei,YANG Ziyu,SHEN Jianliang,et al.An application specific instruction set reconfigurable architecture and the mapping of FFT on it [J].Journal of National University of Defense Technology,2012,34(6):39-45[点击复制] |
|
|
|
本文已被:浏览 8013次 下载 6408次 |
一种应用定制指令集可重构结构及FFT算法映射优化 |
刘磊, 杨子煜, 沈剑良, 李思昆 |
(国防科技大学 计算机学院,湖南 长沙 410073)
|
摘要: |
现代无线通信应用对FFT计算吞吐率与灵活性需求越来越高,针对传统方案实现FFT计算时难以兼顾性能与灵活性的问题,提出一种应用定制指令集可重构结构ASRA,实现了FFT算法在该结构上的映射优化。ASRA在静态多发射处理器内紧耦合应用定制的混合粒度可重构硬件作为扩展功能单元簇,通过运行时重构动态切换扩展指令集。ASRA采用多体便笺存储器、多端口便笺管理单元及可重构互连构成片上缓存系统,结合多体并行访问、循环级乒乓交替、读/写流水化等技术有效提高了访存带宽;静态多发射和运行时语境管理机制支持核心循环的硬件自动流水执行和软流水执行,开发了指令级、数据级和循环级等多层次并行性。实验结果表明,ASRA大幅提升了FFT计算吞吐率,且支持的FFT计算参数更加灵活,而增加的面积开销相对较小。 |
关键词: 应用定制 运行时重构 傅里叶变换 |
DOI: |
投稿日期:2012-07-02 |
基金项目:国家自然科学基金资助项目(61076020,61133007) |
|
An application specific instruction set reconfigurable architecture and the mapping of FFT on it |
LIU Lei, YANG Ziyu, SHEN Jianliang, LI Sikun |
(College of Computer, National University of Defense Technology, Changsha 410073,China)
|
Abstract: |
To meet the stringent requirements on both high-throughput and high-flexibility for FFT computation in modern wireless communication applications, an application specific reconfigurable architecture called ASRA is presented. ASRA is a VLIW-like static multi-issue processor with custom multi-grained reconfigurable fabric as extensible function units. The run-time context manager in ASRA offers multi-grained Custom Instructions selection and is bound to the appropriate reconfigurable fabric while considering run-time changing scenarios. Mapping of the FFT algorithm on ASRA is studied and optimized. ASRA employs on-chip scratchpad as fast local memory. The read/write operation and butterfly operation is pipelined to support hardware pipeline execution of a loop kernel. Experiment results show that ASRA achieves a high performance improvement and a good trade-off between area and performance. |
Keywords: application specific instruction set run-time reconfigurable fast Fourier transform |
|
|