预览加载中,请您耐心等待几秒...
1/2
2/2

在线预览结束,喜欢就下载吧,查找使用更方便

如果您无法下载资料,请参考说明:

1、部分资料下载需要金币,请确保您的账户上有足够的金币

2、已购买过的文档,再次下载不重复扣费

3、资料包下载后请先用软件解压,在使用对应软件打开

基于CUDA的MPEG2编码技术的研究的中期报告 Title:ResearchonCUDA-basedMPEG2EncodingTechnology Abstract: Inthisresearch,wearestudyingtheMPEG2encodingtechnologywhichisbasedonNVIDIA’sCUDAplatform.Wehavecompletedtheinitialresearchwork,andinthismid-termreport,wewillintroducethemainprogressofourresearch. Firstly,weresearchedtheencodingtechnologyandtheCUDAprogrammingmodel.TheMPEG2encodingtechnologyadoptsahybridencodingalgorithm,whichisrelativelycomplex.Weanalyzedthealgorithmanddivideditintoseveralmodules,suchasmotionestimation,discretecosinetransform(DCT)andquantization.ThenweresearchedhowtoacceleratethesemodulesusingCUDAprogramming,whichisaparallelcomputingplatformdevelopedbyNVIDIA. Secondly,weimplementedthemotionestimationmoduleusingCUDA,whichisoneofthemostcomputation-intensivemodulesofMPEG2encoding.Weoptimizedthesearchalgorithmandrealizedthedataparallelismofthecomputingprocess.WecomparedtheperformancebetweenCPUandGPU,andtheresultshowedthatourCUDAimplementationcouldachieve38timesspeedupcomparedwiththeCPUimplementation. Thirdly,westartedtoimplementtheDCTandquantizationmodulesusingCUDA.IntheDCTmodule,weusedthedistributedcomputingofCUDAtospeeduptheprocess.Inthequantizationmodule,weutilizedthesharedmemoryofCUDAanddesignedanewquantizationmatrixtoachievebettercompressionperformance. Finally,weevaluatedtheperformanceofourCUDAimplementationofMPEG2encodingtechnologywithvarioustestdatasets.TheexperimentresultdemonstratedthattheCUDAimplementationcouldachievehighcompressionqualityandsignificantlyincreasethespeedoftheencodingprocess. Inconclusion,ourresearchdemonstratesthatNVIDIA’sCUDAplatformprovidesafeasiblemethodforacceleratingcomplexalgorithmssuchasMPEG2encoding.Inthenextfewmonths,weplantocompletetheimplementationoftheremainingmodulesofMPEG2encodingandevaluatetheperformanceofthecompleteCUDA-basedMPEG2encodingsystem. Keywords:MPEG2encoding,CUDA,motionestimation,DCT,quantization,performanceevaluation.