基于DCT预测编码的Epiphany-OpenCL大矩阵乘并行计算  

Epiphany-OpenCL Large Matrix Multiplication Parallel Computation Method Based on DCT Predictive Coding

作　　者：龙卓群[1] 王晓瑜[1] 王昌明[2] LONG Zhuo-qun Wang Xiao-yu Wang Chang-ming(School of Electronic Engineering,Xi＇ an Aeronautical University, Xi＇an 710077, China School of Electronic Engi- neering,Xidian University,Xi＇an 710077,China)

机构地区：[1]西安航空学院电子工程学院,西安710077 [2]西安电子科技大学电子工程学院,西安710077

出　　处：《自动化与仪表》2017年第7期16-21,33,共7页Automation & Instrumentation

基　　金：陕西省教育厅专项基金项目(15JK2129)

摘　　要：为提高大矩阵乘的并行计算效率和计算精度,该文提出一种基于DCT预测编码的Epiphany-OpenCL大矩阵乘并行计算方法。首先,引入DCT预测编码技术,并利用其二维数据的DCT变换值及其逆变换的二维表达式,对未编码数据的预测来消除邻近数据间在时间域以及空间域上的相关性,以达到对数据进行压缩的目的;其次,基于Epiphany进行OpenCL的并行变换编码处理流程设计,实现矩阵乘的并行化计算;最后,通过在常用编程模型和大矩阵乘法上的试验,显示所提方法具有更高的计算效率和精度,验证了所提并行计算方法的性能优势。In order to improve the parallel computing efficiency and computational precision of large matrix multipli- cation,a parallel computation method of Epiphany-OpenCL based on DCT predictive coding is proposed. Firstly,the DCT prediction encoding technology was introduced,and its two dimensional inverse DCT transform expression value of the two-dimensional data was used to predict the encoding data to eliminate the correlation in time domain and space domain of the adjacent data,which realized the compression of data;Secondly,based on Epiphany,the parallel transform coding process of OpenCL is designed to realize the parallel computation of matrix muhiplication;Finally, experiments on the common programming model and the large matrix multiplication show that the proposed method has higher computational efficiency and accuracy,and the performance advantage of the proposed parallel computing method is verified.

关 键 词：DCT预测编码 Epiphany架构 OpenCL模型 矩阵乘法 并行计算 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]