大点数FFT在同构多核系统中的映射实现  

The Implementation of Large FFT on Homogeneous Multi-core Systems

作　　者：曲冬明 王晓蕾 张多利[1] 宋宇鲲[1] 杜高明[1] QU Dong-ming;WANG Xiao-lei;ZHANG Duo-li;SONG Yu-kun;DU Gao-ming(School of Electronic Science and Applied Physics,Hefei University of Technology,Hefei 230009,China)

机构地区：[1]合肥工业大学电子科学与应用物理学院,安徽合肥230009

出　　处：《微电子学与计算机》2018年第12期90-94,共5页Microelectronics & Computer

摘　　要：大点数FFT被广泛地应用于图像处理、雷达信号处理、卫星通信、生物医学等高吞吐、高实时性的应用中,同时,在过去几年中多核系统的发展一直是主流方向.高效、低资源消耗、且便于多核系统实现的大点数FFT映射方案极有研究价值.本文介绍一种在同构多核系统中实现大点数FFT的映射方案.该方案在片上存储消耗和资源运算节点(PE)的负载之间取得了很好的折中.本文采用了2维FFT原理和基2时间抽取(DIT)FFT方法将大点数FFT分成若干较小规模,同时送到多个资源运算节点(PE)中并行计算.为了减少处理器执行时间和编程任务量,本文巧妙的将每级蝶型运算的源数据和结果数据分别访问相同的存储区.同时,本文设计了一种流水线结构来实现多个大点数FFT计算,提高了运算的并行性.Large FFT is widely applied to high throughput and high real-time applications such as image processing, radar signal processing,satellite communications and biomedicine.At the same time,the development of multi-core system has been the mainstream in the past few years.So it is valuable to study the implementation of large FFT on the multi-core systems which is high efficient and low resource consumption Extremely.This paper presents one implementation of large FFT on homogeneous multi-core system and achieves a good balance between the on-chip memory consumption and the payload in the process elements (PE).This paper adopts the 2-diamension FFT principle and radix-2decimation-in-time (DIT)FFT method,which divides the large FFT into several shorter ones and computes each part with the process elements (PE)on the homogeneous multi-core system.In order to reduce the time of processor execution and reduce programming tasks,this paper make the source data and the results of data of each butterfly access to the same storage area skillfully.In addition,a pipeline structure is introduced for multiple FFTs computing at the same time aiming at improving the parallelism.

关 键 词：多核系统 浮点处理器 大点数FFT 映射方案 并行计算 

分 类 号：TN402[电子电信—微电子学与固体电子学]