使用融合乘加加速快速傅里叶变换计算的向量化方法  

Vectorization of accelerating fast Fourier transform computation based on fused multiply-add instruction

作　　者：刘仲[1] 陈海燕[1] 向宏卫[1] 

机构地区：[1]国防科技大学计算机学院,湖南长沙410073

出　　处：《国防科技大学学报》2015年第2期72-78,共7页Journal of National University of Defense Technology

基　　金：国家自然科学基金资助项目(61133007;61472432)

摘　　要：融合乘加指令加速快速傅里叶变换计算的向量化方法,通过变换快速傅里叶变换的蝶形单元运算流程,将传统计算方式中独立的乘法和加法操作组合成次数更少的融合乘加操作,使得时间抽取法基2快速傅里叶变换算法的蝶形单元计算的实数浮点操作由原来的10次乘(加)操作减少到6次融合乘加操作,时间抽取法基4快速傅里叶变换算法的蝶形单元计算的实数浮点操作由原来的34次乘(加)操作减少到24次融合乘加操作;优化了蝶形因子的向量访问,减少存储开销。实验结果表明,提出的方法能够显著加速快速傅里叶变换的计算,取得高效的计算性能和效率。A vectorization of accelerating fast Fourier transform computation based on fused multiply-add instruction was presented. Separate multiplication and addition operations in conventional computation were manipulated into less fused multiply-add operations by transforming process of fast Fourier transform butterfly computation, which decreased the real floating-point operations of radix-2 decimation in time fast Fourier transform butterfly computation from 10 multiplication （addition） operations to 6 multiply-add operations and decreased the real floating-point operations of radix.,4 decimation in time fast Fourier transform butterfly computation from 34 multiplication （addition） operations to 24 multiply-add operations. Vector data access on twiddle factors was optimized to reduce memory cost. Experimental results show that the presented method can greatly accelerate fast Fourier transform computation and achieve efficient performance and efficiency.

关 键 词：快速傅里叶变换 融合乘加 向量化 向量处理器 

分 类 号：P391.4[天文地球—地球物理学]