在gpgpu上实现kirchhoff叠前时间偏移算法 精灵论文

《在gpgpu上实现kirchhoff叠前时间偏移算法 精灵论文》由会员上传分享，免费在线阅读，更多相关内容在学术论文-天天文库。

1、PracticalPre-StackKirchhoffTimeMigrationofSeismicProcessingonGeneralPurposeGPUSHIXiaohua,WANGXu,ZHAOChanghai,YANGHaiyan5(SchoolofComputerScience,BeihangUniversity,Beijing100083)Abstract:Inthispaper,weintroducedthreeprototypesofGPGPUsolutionsonNVidiaGeForce8800GTfora

2、practicalPre-stackKirchhoffTimeMigrationprogram.Wepresentedhowtore-designandre-implementtheoriginalCPUcodetoefficiencyGPUcode.Theprototypesaremorethanatmost7.2timesfasterthanitsCPUversiononIntel’sP43.0G.10Keywords:Kirchhoff;GPGPU;CUDA0IntroductionThemaingoalofearthe

3、xplorationistoprovidetheoilandgasindustrywithknowledgeoftheearth’ssubsurfacestructuretodetectwhereoilcanbefoundandrecovered.Todoso,large-scale15seismicsurveysoftheearthareperformed,andthedatarecordedundergoescomplexiterativeprocessingtoextractageologicalmodeloftheea

4、rth.Thedataaretheninterpretedbyexpertstohelpdecidewheretobuildoilrecoveryinfrastructure[1].Ingeneral,seismicdataprocessingisdividedintotwostepsinpractice.Thefirststepappliessignalprocessingalgorithmstonormalizethesignalovertheentiresurveyandtoincreasethesignal-to-no

5、ise20ratio.Hundredsofmathematicalalgorithmsareavailableduringthisstep,fromwhichgeophysicalspecialistwillselecttheparticularcandidatesforseismicdatabyexperience.Thesecondstep,whichisthemosttime-consumingone,isdesignedtocorrectfortheeffectsofchangingsubsurfacemediavel

6、ocityonthewavepropagationthroughtheearth.Pre-StackKirchhoffTimeMigration(PKTM)algorithmusedinthesecondstepisoneofthemostwidelyadoptedimagingmethodsintheseismic25dataprocessingindustry.Figure1showstherelationshipbetweenthesource,receiverandscatterpointaswellastheirco

7、rrespondingmigrationcurvesusedinPKTMalgorithm.Itassumesthattheenergyofasampledpointonaninputtraceisthesuperpositionofthereflectionsfromalltheundergroundscatterpointsthathavethesametraveltime.Thepurposeofthemigrationprocessingistospreadthe30pointsonaninputtracetoallp

8、ossiblescatterpointsinthe3Dspace.Eachinputtraceisindependentoftheotherswhenitismigrated,andthismakesthisproblemsuitableforparallelizationo