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，更多相关内容在工程资料-天天文库。

1、PracticalPre-StackKirchhoffTimeMigrationofSeismicProcessingonGeneralPurposeGPUSHIXiaohua,WANGXu,ZHAOChanghai,YANGHaiyan(SchoolofComputerScience,BeihangUniversity,Beijing100083)Abstract:Inthispaper,wcintroducedthreeprototypesofGPGPUsolutionsonNVidiaGeForce8800GTforapracticalPre-s

2、tackKirchhoffTimeMigrationprogram.Wepresentedhowtorc-dcsignandrc-implcmcnttheoriginalCPUcodetoefficiencyGPUcode.Theprototypesarcmorethanatmost7.2timesfasterthanitsCPUversiononIntel’sP43.0G.Keywords:Kirchhoff;GPGPU;CUDA0IntroductionThemaingoalofearthexplorationistoprovidetheoilan

3、dgasindustrywithknowledgeoftheearth’ssubsurfacestructuretodetectwhereoilcanbefoundandrecovered.Todoso,large-scaleseismicsurveysoftheearthareperformed,andthedatarecordedundergoescomplexiterativeprocessingtoextractageologicalmodeloftheearth.Thedataaretheninterpretedbyexpertstohelp

4、decidewheretobuildoilrecoveryinfrastructure[1].Ingeneral，seismicdataprocessingisdividedintotwostepsinpractice.Thefirststepappliessignalprocessingalgorithmstonormalizethesignalovertheentiresurveyandtoincreasethesignal-to-noiseratio.Hundredsofmathematicalalgorithmsareavailableduri

5、ngthisstep，fromwhichgeophysicalspecialistwillselecttheparticularcandidatesforseismicdatabyexperience.Thesecondstep,whichisthemosttime-consumingone,isdesignedtocorrectfortheeffectsofchangingsubsurfacemediavelocityonthewavepropagationthroughtheearth.Pre-StackKirchhoffTimeMigration

6、(PKTM)algorithmusedinthesecondstepisoneofthemostwidelyadoptedimagingmethodsintheseismicdataprocessingindustry.Figure1showstherelationshipbetweenthesource，receiverandscatterpointaswellastheircorrespondingmigrationcurvesusedinPKTMalgorithm.Itassumesthattheenergyofasampledpointonan

7、inputtraceisthesuperpositionofthereflectionsfromalltheundergroundscatterpointsthathavethesametraveltime.Thepurposeofthemigrationprocessingistospreadthepointsonaninputtracetoallpossiblescatterpointsinthe3Dspace.Eachinputtraceisindependentoftheotherswhenitismigrated,andthismakesth

8、isproblemsuitableforparallelizationonthecluster