CUDA Fortran for Scientists and Engineers

《CUDA Fortran for Scientists and Engineers》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、CUDAFortranforScientistsandEngineersGregRuetschMassimilianoFaticaNVIDIACorporation2701SanTomasExpressway,SantaClara,CA95050October3,2011iiContentsPreface..............................................viICUDAFortranProgramming11Introduction31.1ParallelComputation..........................

2、.........41.2BasicConcepts.......................................51.2.1ArstCUDAFortranprogram..........................51.2.2Extendingtolargerarrays.............................81.2.3Multidimensionalarrays..............................111.3DeterminingCUDAHardwareFeaturesandLimits......

3、.............121.3.1Singleanddoubleprecision............................151.4ErrorHandling.......................................161.5CompilingCUDAFortranCode..............................171.6SystemandEnvironmentManagement..........................182PerformanceMeasurementandMetrics212

4、.1MeasuringKernelExecutionTime............................212.1.1Host-devicesynchronizationandCPUtimers..................212.1.2TimingviaCUDAevents.............................222.1.3Command-lineproler...............................242.2Instruction,Bandwidth,andLatencyBoundKernels.....

5、.............252.3MemoryBandwidth....................................272.3.1Theoreticalbandwidth...............................282.3.2Eectivebandwidth................................282.3.3Throughputvs.eectivebandwidth.......................29iiiivCONTENTS3Optimization313.1TransfersBe

6、tweenHostandDevice............................313.1.1Pinnedmemory...................................323.1.2ExplicittransfersusingcudaMemcpy()......................353.1.3Asynchronousdatatransfers(AdvancedTopic).................353.2DeviceMemory.......................................433.

7、2.1Coalescedaccesstoglobalmemory........................443.2.2Localmemory...................................523.2.3Constantmemory.................................533.3On-chipMemory......................................563.3.1L1cache.......................................563.3.