计算机应用 ›› 2005, Vol. 25 ›› Issue (09): 2192-2195.DOI: 10.3724/SP.J.1087.2005.02192

• 典型应用 • 上一篇    下一篇

图形硬件通用计算技术的应用研究

张杨,诸昌钤,何太军   

  1. 西南交通大学计算机与通信工程学院
  • 出版日期:2005-09-01 发布日期:2011-04-11

Application research on general purpose computation on GPU

ZHANG Yang,ZHU Chang-qian,HE Tai-jun   

  1. School of Computer & Communication Engineering,Southwest Jiaotong University,Chengdu Sichuan 610031,China
  • Online:2005-09-01 Published:2011-04-11

摘要: 在通用计算的图形硬件加速研究中,综合了在OPENGL体系下的计算模型。通过实验,测试了该计算结构的性能并分析了提高计算性能的一些方法。在此基础上,介绍一种基于GPU的并行计算二维离散余弦变换方法。该方法可在GPU上通过一遍绘制,对一幅图像1至4个颜色通道,同时进行8×8大小像素块的离散余弦变换。实验表明在该实验硬件基础上,采用GPU加速的并行离散余弦变换,可比相同算法的CPU实现提高数百倍。

关键词: 图形处理器(GPU), 离散余弦变换(DCT), 可编程图形管线, 并行计算

Abstract: In the research on general purpose computation accelerated by graphics hardware, a computation model based on OPENGL was synthesized,the performance of the computation structure was tested and several ways to enhance computing performance were analyzed.Then a method of parallel 2-d DCT on GPU was presented.This method computes DCT on 8×8 pixel blocks by one pass rendering.Up to 4 color channels of a picture can be simultaneously performed during the computation.Experiment results indicate that performance of hardware accelerated DCT is hundreds of times faster than that of CPU implementation in our hardware conditions.

Key words: graphics processing unit(GPU), discrete cosine transform(DCT), programmable pipeline, parallel computation

中图分类号: