AVS标准中整数DCT变换的CUDA并行算法  

Parallel Algorithm of Integer DCT in the AVS standard based on CUDA

作　　者：孟小华[1,2] 刘坚强[1] 

机构地区：[1]暨南大学计算机科学系 [2]暨南大学天体测量,动力学与空间科学中法联合实验室,广州510632

出　　处：《微计算机应用》2011年第11期40-46,共7页Microcomputer Applications

摘　　要：随着图形处理器(GPU)的处理能力的不断增强,图形处理器越来越多的运用在计算密集型的数据处理中。AVS标准视频压缩算法中一些步骤存在典型的并行特性,高清、超清视频压缩的串行算法执行时间开销较大,难以满足实时编码的需要,因此利用GPU的并行处理能力和CUDA的编程框架对AVS标准中的整数DCT变换算法进行了并行实现。经过实验测试,并行算法与串行算法相比具有较高的加速比。With the growth of GPU＇ s processing capability, GPU is more increasingly applied in calculation - intensive data process.Part of steps in the AVS video compression algorithm standard have relatively high parallel characteristics, it will consume much time to execute HD/ultra HD video compression according to serial algorithms ,which can not easy to meet the real - time needs ,using the CUDA parallel processing power and graphics hardware GPU programmablility , has realized the parallel algorithms of integer DCT transform in the AVS standards. After experimental testing and compared with serial programme ,it has relatively high speedup.

关 键 词：图形处理器 AVS CUDA 并行算法 加速比 

分 类 号：TN919.81[电子电信—通信与信息系统]