采用SIMT结构GPU的二维离散哈尔小波变换的优化  

Optimization of Two-dimensional Discrete Harr Wavelet Transform Using GPU with SIMT Structure

作　　者：李一芒 孙凤全 LI Yimang;SUN Fengquan(School of Mechanical Engineering,Changzhou University,Changzhou 213000,China)

机构地区：[1]常州大学机械工程学院,江苏常州213000

出　　处：《计算机测量与控制》2023年第2期185-189,222,共6页Computer Measurement &Control

摘　　要：为了解决CPU环境下小波变换在运行时对高分辨率图片处理速度较慢的问题,利用GPU有大量可编程核心的特点,针对二维离散哈尔小波变换进行了在SIMT(单指令多线程)体系结构GPU环境下的并行推导,同时调整GPU的逻辑布局,将数据分割,更改了数据同步方式,并且采用了虚拟寻址,将速度进一步提升到了0.92 ms,比CPU环境下效率提升51.1%,比SIMD架构效率提升16.3%,效果显著,满足实时性要求。In order to solve the slow problem that wavelet transform has high resolution image in CPU environment in runtime,by using the characteristic of a large number of programmable core of GPU,a two-dimensional discrete Haar wavelet transform is deduced in parallel in single instruction multithreading(SIMT)under the environment of the GPU architecture,at the same time,the GPU logical layout of the GPU is adjusted,and the data is segregated.The data synchronization mode is changed,the virtual addressing is used to further increase the processing time to 0.92 ms,the efficiency of the GPU environment is 51.1%higher than that of the CPU environment,and is 16.3%higher than that in SIMD architecture.The effect is remarkable and meets the real-time requirements.

关 键 词：哈尔小波 GPU SIMT 优化 

分 类 号：TP391.4[自动化与计算机技术—计算机应用技术]