图形处理器CUDA编程模型的应用研究  

Compute Unified Device Architecture on Graphics Processors

作　　者：钱悦[1] 

机构地区：[1]同济大学软件学院,上海200124

出　　处：《计算机与数字工程》2008年第12期177-180,共4页Computer & Digital Engineering

摘　　要：由于图形处理器(GPU)最近几年的快速发展,基于GPU的通用计算已经成为一个新的研究领域。通过对nVIDIA公司最新的通用计算GPU编程模型—CUDA的研究,阐明了CUDA应用程序的结构和它本身特征,讨论和分析了CUDA编程方法与普通CPU编程的差别,并以H.264数字视频编解码中,以消除宏块边界锯齿为主要目的的去块滤波模块为实例,详细描述了CUDA编程的方法和特点,最后通过与CPU编程实现的去块滤波模块的性能比较,揭示了CUDA在计算能力上的优势,为进一步优化编解码器性能和GPU通用计算提供了新的方法和思路。With the development of GPU in recent years, researchers begin to start an evolving field of general purpose computation on graphics processors（GPGPU） . This paper analyzes the features of Compute Unified Device Architecture （CUDA）, which is a new technology of GPGPU developed by nVIDIA corp., demonstrates the structure of GPU, and studies the difference between CUDA and CPU programming. This paper gives a demonstration of the development method of CUDA-based application by an example of deblocking filter in H. 264 video compression. The comparison between CUDA and CPU in deblocking filter shows the advance of the CUDA programming and this provides new methods and ideas to optimiz the video codecs and GPGPU

关 键 词：图形处理器 通用计算 计算机统一设备架构 去块滤波 

分 类 号：TP301.6[自动化与计算机技术—计算机系统结构]