OpenCL_Best_Practices_Guide英文文献资料

《OpenCL_Best_Practices_Guide英文文献资料》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、OptimizationOpenCLBestPracticesGuideMay27,2010OpenCLBestPracticesGuideREVISIONSJuly2009(OriginalRelease)April2010May2010iiMay27,2010TableofContentsPreface............................................................................................................................vWhatIsT

2、hisDocument?vWhoShouldReadThisGuide?vRecommendationsandBestPracticesvContentsSummaryviChapter1.HeterogeneousComputingwithOpenCL.....................................................11.1DifferencesBetweenHostandDevice11.2WhatRunsonanOpenCL-EnabledDevice?21.3MaximumPerformanceBenefit3Chap

3、ter2.PerformanceMetrics.....................................................................................52.1Timing52.1.1UsingCPUTimers52.1.2UsingOpenCLGPUTimers62.2Bandwidth62.2.1TheoreticalBandwidthCalculation62.2.2EffectiveBandwidthCalculation72.2.3ThroughputReportedbytheOpenCLVi

4、sualProfiler7Chapter3.MemoryOptimizations..................................................................................93.1DataTransferBetweenHostandDevice93.1.1PinnedMemory93.1.2AsynchronousTransfers123.1.3OverlappingTransfersandDeviceComputation123.2DeviceMemorySpaces163.2.1Coale

5、scedAccesstoGlobalMemory173.2.1.1ASimpleAccessPattern183.2.1.2ASequentialbutMisalignedAccessPattern183.2.1.3EffectsofMisalignedAccesses193.2.1.4StridedAccesses213.2.2SharedMemory223.2.2.1SharedMemoryandMemoryBanks223.2.2.2SharedMemoryinMatrixMultiplication(C=AB)23T3.2.2.3SharedMemoryin

6、MatrixMultiplication(C=AA)27April30,2010iiiOpenCLBestPracticesGuide3.2.2.4SharedMemoryUsebyKernelArguments293.2.3LocalMemory293.2.4TextureMemory303.2.4.1TexturedFetchvs.GlobalMemoryRead303.2.4.2AdditionalTextureCapabilities303.2.5ConstantMemory313.2.6Registers313.2.6.1RegisterPressure3

7、1Chapter4.NDRangeOptimizations..............................................................................334.1Occupancy334.2CalculatingOccupancy334.3HidingRegisterDependencies354.4ThreadandBlockHeuristics354.5EffectsofSharedMemory36Chapter5.InstructionOptimizations................