nvidia cuda compute unified device...

Version 2.0

June 2008

NVIDIA CUDA

Compute Unified Device Architecture

Reference Manual

Contents

1 RuntimeApiReference 1

1.1 DeviceManagement RT . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2

1.1.1 cudaGetDeviceCount . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3

1.1.2 cudaSetDevice . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4

1.1.3 cudaGetDevice . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5

1.1.4 cudaGetDeviceProperties . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6

1.1.5 cudaChooseDevice . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8

1.2 ThreadManagement RT . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9

1.2.1 cudaThreadSynchronize . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10

1.2.2 cudaThreadExit . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11

1.3 StreamManagement RT . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12

1.3.1 cudaStreamCreate . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13

1.3.2 cudaStreamQuery . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14

1.3.3 cudaStreamSynchronize . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15

1.3.4 cudaStreamDestroy . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16

1.4 EventManagement RT . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17

1.4.1 cudaEventCreate . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18

1.4.2 cudaEventRecord . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19

1.4.3 cudaEventQuery . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 20

1.4.4 cudaEventSynchronize . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21

1.4.5 cudaEventDestroy . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22

1.4.6 cudaEventElapsedTime . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23

1.5 MemoryManagement RT . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 24

1.5.1 cudaMalloc . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25

1.5.2 cudaMallocPitch . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 26

1.5.3 cudaFree . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 27

1.5.4 cudaMallocArray . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 28

1.5.5 cudaFreeArray . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 29

1.5.6 cudaMallocHost . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 30

1.5.7 cudaFreeHost . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31

1.5.8 cudaMemset . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 32

1.5.9 cudaMemset2D . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 33

i

1.5.10 cudaMemcpy . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 34

1.5.11 cudaMemcpy2D . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 35

1.5.12 cudaMemcpyToArray . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 36

1.5.13 cudaMemcpy2DToArray . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 37

1.5.14 cudaMemcpyFromArray . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 38

1.5.15 cudaMemcpy2DFromArray . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 39

1.5.16 cudaMemcpyArrayToArray . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 40

1.5.17 cudaMemcpy2DArrayToArray . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 41

1.5.18 cudaMemcpyToSymbol . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 42

1.5.19 cudaMemcpyFromSymbol . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 43

1.5.20 cudaGetSymbolAddress . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 44

1.5.21 cudaGetSymbolSize . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 45

1.5.22 cudaMalloc3D . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 46

1.5.23 cudaMalloc3DArray . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 48

1.5.24 cudaMemset3D . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 50

1.5.25 cudaMemcpy3D . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 52

1.6 TextureReferenceManagement RT . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 54

1.6.1 LowLevelApi . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 55

1.6.2 HighLevelApi . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 63

1.7 ExecutionControl RT . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 68

1.7.1 cudaConfigureCall . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 69

1.7.2 cudaLaunch . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 70

1.7.3 cudaSetupArgument . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 71

1.8 OpenGlInteroperability RT . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 72

1.8.1 cudaGLSetGLDevice . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 73

1.8.2 cudaGLRegisterBufferObject . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 74

1.8.3 cudaGLMapBufferObject . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 75

1.8.4 cudaGLUnmapBufferObject . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 76

1.8.5 cudaGLUnregisterBufferObject . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 77

1.9 Direct3dInteroperability RT . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 78

1.9.1 cudaD3D9GetDevice . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 79

1.9.2 cudaD3D9SetDirect3DDevice . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 80

1.9.3 cudaD3D9GetDirect3DDevice . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 81

1.9.4 cudaD3D9RegisterResource . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 82

ii

1.9.5 cudaD3D9UnregisterResource . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 84

1.9.6 cudaD3D9MapResources . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 85

1.9.7 cudaD3D9UnmapResources . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 86

1.9.8 cudaD3D9ResourceSetMapFlags . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 87

1.9.9 cudaD3D9ResourceGetSurfaceDimensions . . . . . . . . . . . . . . . . . . . . . . . . . 89

1.9.10 cudaD3D9ResourceGetMappedPointer . . . . . . . . . . . . . . . . . . . . . . . . . . 90

1.9.11 cudaD3D9ResourceGetMappedSize . . . . . . . . . . . . . . . . . . . . . . . . . . . . 91

1.9.12 cudaD3D9ResourceGetMappedPitch . . . . . . . . . . . . . . . . . . . . . . . . . . . . 92

1.10 ErrorHandling RT . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 94

1.10.1 cudaGetLastError . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 95

1.10.2 cudaGetErrorString . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 97

2 DriverApiReference 98

2.1 Initialization . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 99

2.1.1 cuInit . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 100

2.2 DeviceManagement . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 101

2.2.1 cuDeviceComputeCapability . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 102

2.2.2 cuDeviceGet . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 103

2.2.3 cuDeviceGetAttribute . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 104

2.2.4 cuDeviceGetCount . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 106

2.2.5 cuDeviceGetName . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 107

2.2.6 cuDeviceGetProperties . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 108

2.2.7 cuDeviceTotalMem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 110

2.3 ContextManagement . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 111

2.3.1 cuCtxAttach . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 112

2.3.2 cuCtxCreate . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 113

2.3.3 cuCtxDestroy . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 115

2.3.4 cuCtxDetach . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 116

2.3.5 cuCtxGetDevice . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 117

2.3.6 cuCtxPopCurrent . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 118

2.3.7 cuCtxPushCurrent . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 119

2.3.8 cuCtxSynchronize . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 120

2.4 ModuleManagement . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 121

2.4.1 cuModuleGetFunction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 122

iii

2.4.2 cuModuleGetGlobal . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 123

2.4.3 cuModuleGetTexRef . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 124

2.4.4 cuModuleLoad . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 125

2.4.5 cuModuleLoadData . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 126

2.4.6 cuModuleLoadFatBinary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 127

2.4.7 cuModuleUnload . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 128

2.5 StreamManagement . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 129

2.5.1 cuStreamCreate . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 130

2.5.2 cuStreamDestroy . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 131

2.5.3 cuStreamQuery . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 132

2.5.4 cuStreamSynchronize . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 133

2.6 EventManagement . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 134

2.6.1 cuEventCreate . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 135

2.6.2 cuEventDestroy . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 136

2.6.3 cuEventElapsedTime . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 137

2.6.4 cuEventQuery . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 138

2.6.5 cuEventRecord . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 139

2.6.6 cuEventSynchronize . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 140

2.7 ExecutionControl . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 141

2.7.1 cuLaunch . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 142

2.7.2 cuLaunchGrid . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 143

2.7.3 cuParamSetSize . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 144

2.7.4 cuParamSetTexRef . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 145

2.7.5 cuParamSetf . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 146

2.7.6 cuParamSeti . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 147

2.7.7 cuParamSetv . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 148

2.7.8 cuFuncSetBlockShape . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 149

2.7.9 cuFuncSetSharedSize . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 150

2.8 MemoryManagement . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 151

2.8.1 cuArrayCreate . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 152

2.8.2 cuArray3DCreate . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 154

2.8.3 cuArrayDestroy . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 156

2.8.4 cuArrayGetDescriptor . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 157

2.8.5 cuArray3DGetDescriptor . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 158

iv

2.8.6 cuMemAlloc . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 159

2.8.7 cuMemAllocHost . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 160

2.8.8 cuMemAllocPitch . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 161

2.8.9 cuMemFree . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 163

2.8.10 cuMemFreeHost . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 164

2.8.11 cuMemGetAddressRange . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 165

2.8.12 cuMemGetInfo . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 166

2.8.13 cuMemcpy2D . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 167

2.8.14 cuMemcpy3D . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 170

2.8.15 cuMemcpyAtoA . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 173

2.8.16 cuMemcpyAtoD . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 174

2.8.17 cuMemcpyAtoH . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 175

2.8.18 cuMemcpyDtoA . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 176

2.8.19 cuMemcpyDtoD . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 177

2.8.20 cuMemcpyDtoH . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 178

2.8.21 cuMemcpyHtoA . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 179

2.8.22 cuMemcpyHtoD . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 180

2.8.23 cuMemset . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 181

2.8.24 cuMemset2D . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 182

2.9 TextureReferenceManagement . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 183

2.9.1 cuTexRefCreate . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 184

2.9.2 cuTexRefDestroy . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 185

2.9.3 cuTexRefGetAddress . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 186

2.9.4 cuTexRefGetAddressMode . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 187

2.9.5 cuTexRefGetArray . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 188

2.9.6 cuTexRefGetFilterMode . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 189

2.9.7 cuTexRefGetFlags . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 190

2.9.8 cuTexRefGetFormat . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 191

2.9.9 cuTexRefSetAddress . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 192

2.9.10 cuTexRefSetAddressMode . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 193

2.9.11 cuTexRefSetArray . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 194

2.9.12 cuTexRefSetFilterMode . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 195

2.9.13 cuTexRefSetFlags . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 196

2.9.14 cuTexRefSetFormat . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 197

v

2.10 OpenGlInteroperability . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 198

2.10.1 cuGLCtxCreate . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 199

2.10.2 cuGLInit . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 200

2.10.3 cuGLMapBufferObject . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 201

2.10.4 cuGLRegisterBufferObject . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 202

2.10.5 cuGLUnmapBufferObject . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 203

2.10.6 cuGLUnregisterBufferObject . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 204

2.11 Direct3dInteroperability . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 205

2.11.1 cuD3D9GetDevice . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 206

2.11.2 cuD3D9CtxCreate . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 207

2.11.3 cuD3D9GetDirect3DDevice . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 208

2.11.4 cuD3D9RegisterResource . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 209

2.11.5 cuD3D9UnregisterResource . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 211

2.11.6 cuD3D9MapResources . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 212

2.11.7 cuD3D9UnmapResources . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 213

2.11.8 cuD3D9ResourceSetMapFlags . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 214

2.11.9 cuD3D9ResourceGetSurfaceDimensions . . . . . . . . . . . . . . . . . . . . . . . . . . 216

2.11.10 cuD3D9ResourceGetMappedPointer . . . . . . . . . . . . . . . . . . . . . . . . . . . . 217

2.11.11 cuD3D9ResourceGetMappedSize . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 219

2.11.12 cuD3D9ResourceGetMappedPitch . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 220

3 AtomicFunctions 222

3.1 ArithmeticFunctions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 223

3.1.1 atomicAdd . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 224

3.1.2 atomicSub . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 225

3.1.3 atomicExch . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 226

3.1.4 atomicMin . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 227

3.1.5 atomicMax . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 228

3.1.6 atomicInc . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 229

3.1.7 atomicDec . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 230

3.1.8 atomicCAS . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 231

3.2 BitwiseFunctions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 232

3.2.1 atomicAnd . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 233

3.2.2 atomicOr . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 234

vi

3.2.3 atomicXor . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 235

vii

1 RuntimeApiReference

NAME

Runtime API Reference

DESCRIPTION

There are two levels for the runtime API.

The low-level API (cuda_runtime_api.h) is a C-style interface that does not require compiling with nvcc.

The high-level API (cuda_runtime.h) is a C++-style interface built on top of the low-level API. It wrapssome of the low level API routines, using overloading, references and default arguments. These wrappers canbe used from C++ code and can be compiled with any C++ compiler. The high-level API also has someCUDA-specific wrappers that wrap low-level routines that deal with symbols, textures, and device functions.These wrappers require the use of nvcc because they depend on code being generated by the compiler. Forexample, the execution configuration syntax to invoke kernels is only available in source code compiled withnvcc.

SEE ALSO

Device Management, Thread Management, Stream Management, Event Management, Execution Management,Memory Management, Texture Reference Management, OpenGL Interoperability, Direct3D Interoperability, ErrorHandling

1

1.1 DeviceManagement RT

NAME

Device Management

DESCRIPTION

This section describes the CUDA runtime application programming interface.

cudaGetDeviceCount

cudaSetDevice

cudaGetDevice

cudaGetDeviceProperties

cudaChooseDevice

SEE ALSO


2

1.1.1 cudaGetDeviceCount

NAME

cudaGetDeviceCount - returns the number of compute-capable devices

SYNOPSIS

cudaError_t cudaGetDeviceCount( int* count )

DESCRIPTION

Returns in *count the number of devices with compute capability greater or equal to 1.0 that are availablefor execution. If there is no such device, cudaGetDeviceCount() returns 1 and device 0 only supports deviceemulation mode. Since this device will be able to emulate all hardware features, this device will report majorand minor compute capability versions of 9999.

RETURN VALUE

Relevant return values:

cudaSuccess

Note that this function may also return error codes from previous, asynchronous launches.

SEE ALSO

cudaGetDevice, cudaSetDevice, cudaGetDeviceProperties, cudaChooseDevice

3

1.1.2 cudaSetDevice

NAME

cudaSetDevice - sets device to be used for GPU executions

SYNOPSIS

cudaError_t cudaSetDevice( int dev )

DESCRIPTION

Records dev as the device on which the active host thread executes the device code.

RETURN VALUE


cudaSuccess

cudaErrorInvalidDevice


SEE ALSO

cudaGetDeviceCount, cudaGetDevice, cudaGetDeviceProperties, cudaChooseDevice

4

1.1.3 cudaGetDevice

NAME

cudaGetDevice - returns which device is currently being used

SYNOPSIS

cudaError_t cudaGetDevice( int* dev )

DESCRIPTION

Returns in *dev the device on which the active host thread executes the device code.

RETURN VALUE


cudaSuccess


SEE ALSO

cudaGetDeviceCount, cudaSetDevice, cudaGetDeviceProperties, cudaChooseDevice

5

1.1.4 cudaGetDeviceProperties

NAME

cudaGetDeviceProperties - returns information on the compute-device

SYNOPSIS

cudaError_t cudaGetDeviceProperties( struct cudaDeviceProp* prop, int dev )

DESCRIPTION

Returns in *prop the properties of device dev. The cudaDeviceProp structure is defined as:

struct cudaDeviceProp {

char name[256];

size_t totalGlobalMem;

size_t sharedMemPerBlock;

int regsPerBlock;

int warpSize;

size_t memPitch;

int maxThreadsPerBlock;

int maxThreadsDim[3];

int maxGridSize[3];

size_t totalConstMem;

int major;

int minor;

int clockRate;

size_t textureAlignment;

int deviceOverlap;

int multiProcessorCount;

}

where:

name

is an ASCII string identifying the device;

totalGlobalMem

is the total amount of global memory available on the device in bytes;

sharedMemPerBlock

is the maximum amount of shared memory available to a thread block in bytes; this amount is sharedby all thread blocks simultaneously resident on a multiprocessor;

regsPerBlock

is the maximum number of 32-bit registers available to a thread block; this number is shared by allthread blocks simultaneously resident on a multiprocessor;

warpSize

is the warp size in threads;

6

memPitch

is the maximum pitch in bytes allowed by the memory copy functions that involve memory regionsallocated through cudaMallocPitch();

maxThreadsPerBlock

is the maximum number of threads per block;

maxThreadsDim[3]

is the maximum sizes of each dimension of a block;

maxGridSize[3]

is the maximum sizes of each dimension of a grid;

totalConstMem

is the total amount of constant memory available on the device in bytes;

major, minor

are the major and minor revision numbers defining the device’s compute capability;

clockRate

is the clock frequency in kilohertz;

textureAlignment

is the alignment requirement; texture base addresses that are aligned to textureAlignment bytes donot need an offset applied to texture fetches;

deviceOverlap

is 1 if the device can concurrently copy memory between host and device while executing a kernel, or0 if not;

multiProcessorCount

is the number of multiprocessors on the device.

RETURN VALUE


cudaSuccess



SEE ALSO

cudaGetDeviceCount, cudaGetDevice, cudaSetDevice, cudaChooseDevice

7

1.1.5 cudaChooseDevice

NAME

cudaChooseDevice - select compute-device which best matches criteria

SYNOPSIS

cudaError_t cudaChooseDevice( int* dev, const struct cudaDeviceProp* prop )

DESCRIPTION

Returns in *dev the device which properties best match *prop.

RETURN VALUE


cudaSuccess

cudaErrorInvalidValue


SEE ALSO

cudaGetDeviceCount, cudaGetDevice, cudaSetDevice, cudaGetDeviceProperties

8

1.2 ThreadManagement RT

NAME

Thread Management

DESCRIPTION


cudaThreadSynchronize

cudaThreadExit

SEE ALSO


9

1.2.1 cudaThreadSynchronize

NAME

cudaThreadSynchronize - wait for compute-device to finish

SYNOPSIS

cudaError_t cudaThreadSynchronize(void)

DESCRIPTION

Blocks until the device has completed all preceding requested tasks. cudaThreadSynchronize() returnsan error if one of the preceding tasks failed.

RETURN VALUE


cudaSuccess


SEE ALSO

cudaThreadExit

10

1.2.2 cudaThreadExit

NAME

cudaThreadExit - exit and clean-up from CUDA launches

SYNOPSIS

cudaError_t cudaThreadExit(void)

DESCRIPTION

Explicitly cleans up all runtime-related resources associated with the calling host thread. Any subsequentAPI call reinitializes the runtime. cudaThreadExit() is implicitly called on host thread exit.

RETURN VALUE


cudaSuccess


SEE ALSO

cudaThreadSynchronize

11

1.3 StreamManagement RT

NAME

Stream Management

DESCRIPTION


cudaStreamCreate

cudaStreamQuery

cudaStreamSynchronize

cudaStreamDestroy

SEE ALSO


12

1.3.1 cudaStreamCreate

NAME

cudaStreamCreate - create an async stream

SYNOPSIS

cudaError_t cudaStreamCreate( cudaStream_t* stream )

DESCRIPTION

Creates a stream.

RETURN VALUE


cudaSuccess



SEE ALSO

cudaStreamQuery, cudaStreamSynchronize, cudaStreamDestroy

13

1.3.2 cudaStreamQuery

NAME

cudaStreamQuery - queries a stream for completion-status

SYNOPSIS

cudaError_t cudaStreamQuery(cudaStream_t stream)

DESCRIPTION

Returns cudaSuccess if all operations in the stream have completed, or cudaErrorNotReady if not.

RETURN VALUE


cudaSuccess

cudaErrorNotReady

cudaErrorInvalidResourceHandle


SEE ALSO

cudaStreamCreate, cudaStreamDestroy, cudaStreamSynchronize

14

1.3.3 cudaStreamSynchronize

NAME

cudaStreamSynchronize - waits for stream tasks to complete

SYNOPSIS

cudaError_t cudaStreamSynchronize( cudaStream_t stream )

DESCRIPTION

Blocks until the device has completed all operations in the stream.

RETURN VALUE


cudaSuccess



SEE ALSO

cudaStreamCreate, cudaStreamDestroy, cudaStreamQuery

15

1.3.4 cudaStreamDestroy

NAME

cudaStreamDestroy - destroys and cleans-up a stream object

SYNOPSIS

cudaError_t cudaStreamDestroy( cudaStream_t stream )

DESCRIPTION

Destroys a stream object.

RETURN VALUE


cudaSuccess



SEE ALSO

cudaStreamCreate, cudaStreamSychronize, cudaStreamDestroy

16

1.4 EventManagement RT

NAME

Event Management

DESCRIPTION


cudaEventCreate

cudaEventRecord

cudaEventQuery

cudaEventSynchronize

cudaEventDestroy

cudaEventElapsedTime

SEE ALSO


17

1.4.1 cudaEventCreate

NAME

cudaEventCreate - creates an event-object

SYNOPSIS

cudaError_t cudaEventCreate( cudaEvent_t* event )

DESCRIPTION

Creates an event object.

RETURN VALUE


cudaSuccess

cudaErrorInitializationError

cudaErrorPriorLaunchFailure


cudaErrorMemoryAllocation


SEE ALSO

cudaEventRecord, cudaEventQuery, cudaEventSynchronzie, cudaEventDestroy, cudaEventElapsedTime

18

1.4.2 cudaEventRecord

NAME

cudaEventRecord - records an event

SYNOPSIS

cudaError_t cudaEventRecord( cudaEvent_t event, CUstream stream )

DESCRIPTION

Records an event. If stream is non-zero, the event is recorded after all preceding operations in the streamhave been completed; otherwise, it is recorded after all preceding operations in the CUDA context have beencompleted. Since this operation is asynchronous, cudaEventQuery() and/or cudaEventSynchronize()must be used to determine when the event has actually been recorded.

If cudaEventRecord() has previously been called and the event has not been recorded yet, this functionreturns cudaErrorInvalidValue.

RETURN VALUE


cudaSuccess






SEE ALSO

cudaEventCreate, cudaEventQuery, cudaEventSynchronize, cudaEventDestroy, cudaEventElapsedTime

19

1.4.3 cudaEventQuery

NAME

cudaEventQuery - query if an event has been recorded

SYNOPSIS

cudaError_t cudaEventQuery( cudaEvent_t event )

DESCRIPTION

Returns cudaSuccess if the event has actually been recorded, or cudaErrorNotReady if not. If cud-aEventRecord() has not been called on this event, the function returns cudaErrorInvalidValue.

RETURN VALUE


cudaSuccess

cudaErrorNotReady






SEE ALSO

cudaEventCreate, cudaEventRecord, cudaEventSynchronize, cudaEventDestroy, cudaEventElapsedTime

20

1.4.4 cudaEventSynchronize

NAME

cudaEventSynchronize - wait for an event to be recorded

SYNOPSIS

cudaError_t cudaEventSynchronize( cudaEvent_t event )

DESCRIPTION

Blocks until the event has actually been recorded. If cudaEventRecord() has not been called on thisevent, the function returns cudaErrorInvalidValue.

RETURN VALUE


cudaSuccess






SEE ALSO

cudaEventCreate, cudaEventRecord, cudaEventQuery, cudaEventDestroy, cudaEventElapsedTime

21

1.4.5 cudaEventDestroy

NAME

cudaEventDestroy - destroys an event-object

SYNOPSIS

cudaError_t cudaEventDestroy( cudaEvent_t event )

DESCRIPTION

Destroys the event-object.

RETURN VALUE


cudaSuccess





SEE ALSO

cudaEventCreate, cudaEventQuery, cudaEventSynchronize, cudaEventRecord, cudaEventElapsedTime

22

1.4.6 cudaEventElapsedTime

NAME

cudaEventElapsedTime - computes the elapsed time between events

SYNOPSIS

cudaError_t cudaEventElapsedTime( float* time, cudaEvent_t start, cudaEvent_t end );

DESCRIPTION

Computes the elapsed time between two events (in milliseconds with a resolution of around 0.5 microseconds).If either event has not been recorded yet, this function returns cudaErrorInvalidValue. If either eventhas been recorded with a non-zero stream, the result is undefined.

RETURN VALUE


cudaSuccess







SEE ALSO

cudaEventCreate, cudaEventQuery, cudaEventSynchronize, cudaEventDestroy, cudaEventRecord

23

1.5 MemoryManagement RT

NAME

Memory Management

DESCRIPTION


cudaMalloc

cudaMallocPitch

cudaFree

cudaMallocArray

cudaFreeArray

cudaMallocHost

cudaFreeHost

cudaMemset

cudaMemset2D

cudaMemcpy

cudaMemcpy2D

cudaMemcpyToArray

cudaMemcpy2DToArray

cudaMemcpyFromArray

cudaMemcpy2DFromArray

cudaMemcpyArrayToArray

cudaMemcpy2DArrayToArray

cudaMemcpyToSymbol

cudaMemcpyFromSymbol

cudaGetSymbolAddress

cudaGetSymbolSize

SEE ALSO


24

1.5.1 cudaMalloc

NAME

cudaMalloc - allocate memory on the GPU

SYNOPSIS

cudaError_t cudaMalloc( void** devPtr, size_t count )

DESCRIPTION

Allocates count bytes of linear memory on the device and returns in *devPtr a pointer to the allocatedmemory. The allocated memory is suitably aligned for any kind of variable. The memory is not cleared.cudaMalloc() returns cudaErrorMemoryAllocation in case of failure.

RETURN VALUE


cudaSuccess



SEE ALSO

cudaMallocPitch, cudaFree, cudaMallocArray, cudaFreeArray, cudaMallocHost, cudaFreeHost

25

1.5.2 cudaMallocPitch

NAME

cudaMallocPitch - allocates memory on the GPU

SYNOPSIS

cudaError_t cudaMallocPitch( void** devPtr, size_t* pitch, size_t widthInBytes, size_t height

)

DESCRIPTION

Allocates at least widthInBytes*height bytes of linear memory on the device and returns in *devPtra pointer to the allocated memory. The function may pad the allocation to ensure that correspondingpointers in any given row will continue to meet the alignment requirements for coalescing as the address isupdated from row to row. The pitch returned in *pitch by cudaMallocPitch() is the width in bytes ofthe allocation. The intended usage of pitch is as a separate parameter of the allocation, used to computeaddresses within the 2D array. Given the row and column of an array element of type T, the address iscomputed as

T* pElement = (T*)((char*)BaseAddress + Row * pitch) + Column;

For allocations of 2D arrays, it is recommended that programmers consider performing pitch allocationsusing cudaMallocPitch(). Due to pitch alignment restrictions in the hardware, this is especially true ifthe application will be performing 2D memory copies between different regions of device memory (whetherlinear memory or CUDA arrays).

RETURN VALUE


cudaSuccess



SEE ALSO

cudaMalloc, cudaFree, cudaMallocArray, cudaFreeArray, cudaMallocHost, cudaFreeHost

26

1.5.3 cudaFree

NAME

cudaFree - frees memory on the GPU

SYNOPSIS

cudaError_t cudaFree(void* devPtr)

DESCRIPTION

Frees the memory space pointed to by devPtr, which must have been returned by a previous call to cud-aMalloc() or cudaMallocPitch(). Otherwise, or if cudaFree(devPtr) has already been called before,an error is returned. If devPtr is 0, no operation is performed. cudaFree() returns cudaErrorInvalid-DevicePointer in case of failure.

RETURN VALUE


cudaSuccess

cudaErrorInvalidDevicePointer



SEE ALSO

cudaMalloc, cudaMallocPitch, cudaMallocArray, cudaFreeArray, cudaMallocHost, cudaFreeHost

27

1.5.4 cudaMallocArray

NAME

cudaMallocArray - allocate an array on the GPU

SYNOPSIS

cudaError_t cudaMallocArray( struct cudaArray** array, const struct cudaChannelFormatDesc*

desc, size_t width, size_t height )

DESCRIPTION

Allocates a CUDA array according to the cudaChannelFormatDesc structure desc and returns a handleto the new CUDA array in *array. The cudaChannelFormatDesc is defined as:

struct cudaChannelFormatDesc {

int x, y, z, w;

enum cudaChannelFormatKind f;

};

where cudaChannelFormatKind is one of cudaChannelFormatKindSigned, cudaChannelFormatKin-dUnsigned, cudaChannelFormatKindFloat.

RETURN VALUE


cudaSuccess



SEE ALSO

cudaMalloc, cudaMallocPitch, cudaFree, cudaFreeArray, cudaMallocHost, cudaFreeHost

28

1.5.5 cudaFreeArray

NAME

cudaFreeArray - frees an array on the GPU

SYNOPSIS

cudaError_t cudaFreeArray( struct cudaArray* array )

DESCRIPTION

Frees the CUDA array array. If array is 0, no operation is performed.

RETURN VALUE


cudaSuccess



SEE ALSO

cudaMalloc, cudaMallocPitch, cudaFree, cudaMallocArray, cudaMallocHost, cudaFreeHost

29

1.5.6 cudaMallocHost

NAME

cudaMallocHost - allocates page-locked memory on the host

SYNOPSIS

cudaError_t cudaMallocHost( void** hostPtr, size_t size )

DESCRIPTION

Allocates size bytes of host memory that is page-locked and accessible to the device. The driver tracksthe virtual memory ranges allocated with this function and automatically accelerates calls to functions suchas cudaMemcpy*(). Since the memory can be accessed directly by the device, it can be read or writtenwith much higher bandwidth than pageable memory obtained with functions such as malloc(). Allocatingexcessive amounts of memory with cudaMallocHost() may degrade system performance, since it reducesthe amount of memory available to the system for paging. As a result, this function is best used sparinglyto allocate staging areas for data exchange between host and device.

RETURN VALUE


cudaSuccess



SEE ALSO

cudaMalloc, cudaMallocPitch, cudaFree, cudaMallocArray, cudaFreeArray, cudaFreeHost

30

1.5.7 cudaFreeHost

NAME

cudaFreeHost - frees page-locked memory

SYNOPSIS

cudaError_t cudaFreeHost( void* hostPtr )

DESCRIPTION

Frees the memory space pointed to by hostPtr, which must have been returned by a previous call tocudaMallocHost().

RETURN VALUE


cudaSuccess



SEE ALSO

cudaMalloc, cudaMallocPitch, cudaFree, cudaMallocArray, cudaFreeArray, cudaMallocHost

31

1.5.8 cudaMemset

NAME

cudaMemset - initializes or sets GPU memory to a value

SYNOPSIS

cudaError_t cudaMemset( void* devPtr, int value, size_t count )

DESCRIPTION

Fills the first count bytes of the memory area pointed to by devPtr with the constant byte value value.

RETURN VALUE


cudaSuccess




SEE ALSO

cudaMemset2D, cudaMemset3D

32

1.5.9 cudaMemset2D

NAME

cudaMemset2D - initializes or sets GPU memory to a value

SYNOPSIS

cudaError_t cudaMemset2D( void* dstPtr, size_t pitch, int value, size_t width, size_t height

)

DESCRIPTION

Sets to the specified value value a matrix (height rows of width bytes each) pointed to by dstPtr. pitchis the width in memory in bytes of the 2D array pointed to by dstPtr, including any padding added tothe end of each row. This function performs fastest when the pitch is one that has been passed back bycudaMallocPitch().

RETURN VALUE


cudaSuccess




SEE ALSO

cudaMemset, cudaMemset3D

33

1.5.10 cudaMemcpy

NAME

cudaMemcpy - copies data between GPU and host

SYNOPSIS

cudaError_t cudaMemcpy( void* dst, const void* src, size_t count, enum cudaMemcpyKind kind

)

cudaError_t cudaMemcpyAsync( void* dst, const void* src, size_t count, enum cudaMemcpyKind

kind, cudaStream_t stream )

DESCRIPTION

Copies count bytes from the memory area pointed to by src to the memory area pointed to by dst,where kind is one of cudaMemcpyHostToHost, cudaMemcpyHostToDevice, cudaMemcpyDevice-ToHost, or cudaMemcpyDeviceToDevice, and specifies the direction of the copy. The memory areasmay not overlap. Calling cudaMemcpy() with dst and src pointers that do not match the direction of thecopy results in an undefined behavior.

cudaMemcpyAsync() is asynchronous and can optionally be associated to a stream by passing a non-zerostream argument. It only works on page-locked host memory and returns an error if a pointer to pageablememory is passed as input.

RETURN VALUE


cudaSuccess



cudaErrorInvalidMemcpyDirection


SEE ALSO

cudaMemcpy2D, cudaMemcpyToArray, cudaMemcpy2DToArray, cudaMemcpyFromArray, cudaMemcpy2DFromArray,cudaMemcpyArrayToArray, cudaMemcpy2DArrayToArray, cudaMemcpyToSymbol, cudaMemcpyFromSym-

bol

34

1.5.11 cudaMemcpy2D

NAME

cudaMemcpy2D - copies data between host and device

SYNOPSIS

cudaError_t cudaMemcpy2D( void* dst, size_t dpitch, const void* src, size_t spitch, size_t

width, size_t height, enum cudaMemcpyKind kind )

cudaError_t cudaMemcpy2DAsync( void* dst, size_t dpitch, const void* src, size_t spitch, size_t

width, size_t height, enum cudaMemcpyKind kind, cudaStream_t stream )

DESCRIPTION

Copies a matrix (height rows of width bytes each) from the memory area pointed to by src to the memoryarea pointed to by dst, where kind is one of cudaMemcpyHostToHost, cudaMemcpyHostToDevice,cudaMemcpyDeviceToHost, or cudaMemcpyDeviceToDevice, and specifies the direction of the copy.dpitch and spitch are the widths in memory in bytes of the 2D arrays pointed to by dst and src, includingany padding added to the end of each row. The memory areas may not overlap. Calling cudaMemcpy2D()with dst and src pointers that do not match the direction of the copy results in an undefined behavior.cudaMemcpy2D() returns an error if dpitch or spitch is greater than the maximum allowed.

cudaMemcpy2DAsync() is asynchronous and can optionally be associated to a stream by passing a non-zero stream argument. It only works on page-locked host memory and returns an error if a pointer topageable memory is passed as input.

RETURN VALUE


cudaSuccess


cudaErrorInvalidPitchValue




SEE ALSO

cudaMemcpy, cudaMemcpyToArray, cudaMemcpy2DToArray, cudaMemcpyFromArray, cudaMemcpy2DFromArray,cudaMemcpyArrayToArray, cudaMemcpy2DArrayToArray, cudaMemcpyToSymbol, cudaMemcpyFromSym-

bol

35

1.5.12 cudaMemcpyToArray

NAME

cudaMemcpyToArray - copies data between host and device

SYNOPSIS

cudaError_t cudaMemcpyToArray(struct cudaArray* dstArray, size_t dstX, size_t dstY, const void*

src, size_t count, enum cudaMemcpyKind kind)

cudaError_t cudaMemcpyToArrayAsync(struct cudaArray* dstArray, size_t dstX, size_t dstY, const

void* src, size_t count, enum cudaMemcpyKind kind, cudaStream_t stream)

DESCRIPTION

Copies count bytes from the memory area pointed to by src to the CUDA array dstArray starting at theupper left corner (dstX, dstY), where kind is one of cudaMemcpyHostToHost, cudaMemcpyHost-ToDevice, cudaMemcpyDeviceToHost, or cudaMemcpyDeviceToDevice, and specifies the directionof the copy.

cudaMemcpyToArrayAsync() is asynchronous and can optionally be associated to a stream by passinga non-zero stream argument. It only works on page-locked host memory and returns an error if a pointer topageable memory is passed as input.

RETURN VALUE


cudaSuccess





SEE ALSO

cudaMemcpy, cudaMemcpy2D, cudaMemcpy2DToArray, cudaMemcpyFromArray, cudaMemcpy2DFromArray,cudaMemcpyArrayToArray, cudaMemcpy2DArrayToArray, cudaMemcpyToSymbol, cudaMemcpyFromSym-

bol

36

1.5.13 cudaMemcpy2DToArray

NAME

cudaMemcpy2DToArray - copies data between host and device

SYNOPSIS

cudaError_t cudaMemcpy2DToArray(struct cudaArray* dstArray, size_t dstX, size_t dstY, const

void* src, size_t spitch, size_t width, size_t height, enum cudaMemcpyKind kind); cudaError_t

cudaMemcpy2DToArrayAsync(struct cudaArray* dstArray, size_t dstX, size_t dstY, const void*

src, size_t spitch, size_t width, size_t height, enum cudaMemcpyKind kind, cudaStream_t stream);

DESCRIPTION

Copies a matrix (height rows of width bytes each) from the memory area pointed to by src to the CUDAarray dstArray starting at the upper left corner (dstX, dstY), where kind is one of cudaMemcpyHost-ToHost, cudaMemcpyHostToDevice, cudaMemcpyDeviceToHost, or cudaMemcpyDeviceToDe-vice, and specifies the direction of the copy. spitch is the width in memory in bytes of the 2D array pointedto by src, including any padding added to the end of each row. cudaMemcpy2D() returns an error ifspitch is greater than the maximum allowed.

cudaMemcpy2DToArrayAsync() is asynchronous and can optionally be associated to a stream by pass-ing a non-zero stream argument. It only works on page-locked host memory and returns an error if a pointerto pageable memory is passed as input.

RETURN VALUE


cudaSuccess






SEE ALSO

cudaMemcpy, cudaMemcpy2D, cudaMemcpyToArray, cudaMemcpyFromArray, cudaMemcpy2DFromArray,cudaMemcpyArrayToArray, cudaMemcpy2DArrayToArray, cudaMemcpyToSymbol, cudaMemcpyFromSym-

bol

37

1.5.14 cudaMemcpyFromArray

NAME

cudaMemcpyFromArray - copies data between host and device

SYNOPSIS

cudaError_t cudaMemcpyFromArray(void* dst, const struct cudaArray* srcArray, size_t srcX, size_t

srcY, size_t count, enum cudaMemcpyKind kind)

cudaError_t cudaMemcpyFromArrayAsync(void* dst, const struct cudaArray* srcArray, size_t srcX,

size_t srcY, size_t count, enum cudaMemcpyKind kind, cudaStream_t stream)

DESCRIPTION

Copies count bytes from the CUDA array srcArray starting at the upper left corner (srcX, srcY) to thememory area pointed to by dst, where kind is one of cudaMemcpyHostToHost, cudaMemcpyHost-ToDevice, cudaMemcpyDeviceToHost, or cudaMemcpyDeviceToDevice, and specifies the directionof the copy.

cudaMemcpyFromArrayAsync() is asynchronous and can optionally be associated to a stream by passinga non-zero stream argument. It only works on page-locked host memory and returns an error if a pointer topageable memory is passed as input.

RETURN VALUE


cudaSuccess





SEE ALSO

cudaMemcpy, cudaMemcpy2D, cudaMemcpyToArray, cudaMemcpy2DToArray, cudaMemcpy2DFromArray,cudaMemcpyArrayToArray, cudaMemcpy2DArrayToArray, cudaMemcpyToSymbol, cudaMemcpyFromSym-

bol

38

1.5.15 cudaMemcpy2DFromArray

NAME

cudaMemcpy2DFromArray - copies data between host and device

SYNOPSIS

cudaError_t cudaMemcpy2DFromArray(void* dst, size_t dpitch, const struct cudaArray* srcArray,

size_t srcX, size_t srcY, size_t width, size_t height, enum cudaMemcpyKind kind)

cudaError_t cudaMemcpy2DFromArrayAsync(void* dst, size_t dpitch, const struct cudaArray* srcArray,

size_t srcX, size_t srcY, size_t width, size_t height, enum cudaMemcpyKind kind, cudaStream_t

stream)

DESCRIPTION

Copies a matrix (height rows of width bytes each) from the CUDA array srcArray starting at the upperleft corner (srcX, srcY) to the memory area pointed to by dst, where kind is one of cudaMemcpyHost-ToHost, cudaMemcpyHostToDevice, cudaMemcpyDeviceToHost, or cudaMemcpyDeviceToDe-vice, and specifies the direction of the copy. dpitch is the width in memory in bytes of the 2D array pointedto by dst, including any padding added to the end of each row. cudaMemcpy2D() returns an error ifdpitch is greater than the maximum allowed.

cudaMemcpy2DFromArrayAsync() is asynchronous and can optionally be associated to a stream bypassing a non-zero stream argument. It only works on page-locked host memory and returns an error if apointer to pageable memory is passed as input.

RETURN VALUE


cudaSuccess






SEE ALSO

cudaMemcpy, cudaMemcpy2D, cudaMemcpyToArray, cudaMemcpy2DToArray, cudaMemcpyFromArray, cu-

daMemcpyArrayToArray, cudaMemcpy2DArrayToArray, cudaMemcpyToSymbol, cudaMemcpyFromSymbol

39

1.5.16 cudaMemcpyArrayToArray

NAME

cudaMemcpyArrayToArray - copies data between host and device

SYNOPSIS

cudaError_t cudaMemcpyArrayToArray(struct cudaArray* dstArray, size_t dstX, size_t dstY, const

struct cudaArray* srcArray, size_t srcX, size_t srcY, size_t count, enum cudaMemcpyKind kind)

DESCRIPTION

Copies count bytes from the CUDA array srcArray starting at the upper left corner (srcX, srcY) to theCUDA array dstArray starting at the upper left corner (dstX, dstY), where kind is one of cudaMem-cpyHostToHost, cudaMemcpyHostToDevice, cudaMemcpyDeviceToHost, or cudaMemcpyDe-viceToDevice, and specifies the direction of the copy.

RETURN VALUE


cudaSuccess




SEE ALSO


daMemcpy2DFromArray, cudaMemcpy2DArrayToArray, cudaMemcpyToSymbol, cudaMemcpyFromSymbol

40

1.5.17 cudaMemcpy2DArrayToArray

NAME

cudaMemcpy2DArrayToArray - copies data between host and device

SYNOPSIS

cudaError_t cudaMemcpy2DArrayToArray(struct cudaArray* dstArray, size_t dstX, size_t dstY,

const struct cudaArray* srcArray, size_t srcX, size_t srcY, size_t width, size_t height, enum

cudaMemcpyKind kind)

DESCRIPTION

Copies a matrix (height rows of width bytes each) from the CUDA array srcArray starting at the upperleft corner (srcX, srcY) to the CUDA array dstArray starting at the upper left corner (dstX, dstY),where kind is one of cudaMemcpyHostToHost, cudaMemcpyHostToDevice, cudaMemcpyDevice-ToHost, or cudaMemcpyDeviceToDevice, and specifies the direction of the copy.

RETURN VALUE


cudaSuccess




SEE ALSO


daMemcpy2DFromArray, cudaMemcpyArrayToArray, cudaMemcpyToSymbol, cudaMemcpyFromSymbol

41

1.5.18 cudaMemcpyToSymbol

NAME

cudaMemcpyToSymbol - copies data from host memory to GPU

SYNOPSIS

template < class T >

cudaError_t cudaMemcpyToSymbol( const T& symbol, const void* src, size_t count, size_t offset,

enum cudaMemcpyKind kind)

DESCRIPTION

Copies count bytes from the memory area pointed to by src to the memory area pointed to by offset

bytes from the start of symbol symbol. The memory areas may not overlap. symbol can either be avariable that resides in global or constant memory space, or it can be a character string, naming a vari-able that resides in global or constant memory space. kind can be either cudaMemcpyHostToDevice orcudaMemcpyDeviceToDevice.

RETURN VALUE


cudaSuccess


cudaErrorInvalidSymbol




SEE ALSO


daMemcpy2DFromArray, cudaMemcpyArrayToArray, cudaMemcpy2DArrayToArray, cudaMemcpyFromSym-

bol

42

1.5.19 cudaMemcpyFromSymbol

NAME

cudaMemcpyFromSymbol - copies data from GPU to host memory

SYNOPSIS


cudaError_t cudaMemcpyFromSymbol( void *dst, const T& symbol, size_t count, size_t offset,

enum cudaMemcpyKind kind)

DESCRIPTION

Copies count bytes from the memory area pointed to by offset bytes from the start of symbol symbolto the memory area pointed to by dst. The memory areas may not overlap. symbol can either be avariable that resides in global or constant memory space, or it can be a character string, naming a vari-able that resides in global or constant memory space. kind can be either cudaMemcpyDeviceToHost orcudaMemcpyDeviceToDevice.

RETURN VALUE


cudaSuccess






SEE ALSO


daMemcpy2DFromArray, cudaMemcpyArrayToArray, cudaMemcpy2DArrayToArray, cudaMemcpyToSym-

bol

43

1.5.20 cudaGetSymbolAddress

NAME

cudaGetSymbolAddress - finds the address associated with a CUDA symbol

SYNOPSIS


cudaError_t cudaGetSymbolAddress(void** devPtr, const T& symbol)

DESCRIPTION

Returns in *devPtr the address of symbol symbol on the device. symbol can either be a variable thatresides in global memory space, or it can be a character string, naming a variable that resides in globalmemory space. If symbol cannot be found, or if symbol is not declared in global memory space, *devPtris unchanged and an error is returned. cudaGetSymbolAddress() returns cudaErrorInvalidSymbol incase of failure

RETURN VALUE


cudaSuccess


cudaErrorAddressOfConstant


SEE ALSO

cudaGetSymbolSize

44

1.5.21 cudaGetSymbolSize

NAME

cudaGetSymbolSize - finds the size of the object associated with a CUDA symbol

SYNOPSIS


cudaError_t cudaGetSymbolSize(size_t* size, const T& symbol)

DESCRIPTION

Returns in *size the size of symbol symbol. symbol can either be a variable that resides in global orconstant memory space, or it can be a character string, naming a variable that resides in global or constantmemory space. If symbol cannot be found, or if symbol is not declared in global or constant memory space,*size is unchanged and an error is returned. cudaGetSymbolSize() returns cudaErrorInvalidSymbolin case of failure.

RETURN VALUE


cudaSuccess



SEE ALSO

cudaGetSymbolAddress

45

1.5.22 cudaMalloc3D

NAME

cudaMalloc3D - allocates logical 1D, 2D, or 3D memory objects on the GPU

SYNOPSIS

struct cudaPitchedPtr {

void *ptr;

size_t pitch;

size_t xsize;

size_t ysize;

};

struct cudaExtent {

size_t width;

size_t height;

size_t depth;

};

cudaError_t cudaMalloc3D( struct cudaPitchedPtr* pitchDevPtr, struct cudaExtent extent )

DESCRIPTION

Allocates at least width*height*depth bytes of linear memory on the device and returns a pitchedDevPtrin which ptr is a pointer to the allocated memory. The function may pad the allocation to ensure hardwarealignment requirements are met. The pitch returned in the pitch field of the pitchedDevPtr is the widthin bytes of the allocation.

The returned cudaPitchedPtr contains additional fields xsize and ysize, the logical width and height of theallocation, which are equivalent to the width and height extent parameters provided by the programmerduring allocation.

For allocations of 2D, 3D objects, it is highly recommended that programmers perform allocations usingcudaMalloc3D() or cudaMallocPitch(). Due to alignment restrictions in the hardware, this is especiallytrue if the application will be performing memory copies involving 2D or 3D objects (whether linear memoryor CUDA arrays).

RETURN VALUE


cudaSuccess



46

SEE ALSO

cudaMallocPitch, cudaFree, cudaMemcpy3D, cudaMemset3D, cudaMalloc3DArray, cudaMallocArray, cud-

aFreeArray, cudaMallocHost, cudaFreeHost

47

1.5.23 cudaMalloc3DArray

NAME

cudaMalloc3DArray - allocate an array on the GPU

SYNOPSIS

struct cudaExtent { size_t width; size_t height; size_t depth; };

cudaError_t cudaMalloc3DArray( struct cudaArray** arrayPtr, const struct cudaChannelFormatDesc*

desc, struct cudaExtent extent )

DESCRIPTION

Allocates a CUDA array according to the cudaChannelFormatDesc structure desc and returns a handleto the new CUDA array in *arrayPtr. The cudaChannelFormatDesc is defined as:


int x, y, z, w;


};


cudaMalloc3DArray is able to allocate 1D, 2D, or 3D arrays.

• A 1D array is allocated if the height and depth extent are both zero. For 1D arrays valid extents are{(1, 8192), 0, 0} .

• A 2D array is allocated if only the depth extent is zero. For 2D arrays valid extents are {(1, 65536),(1, 32768), 0}.

• A 3D array is allocate if all three extents are non-zero. For 3D arrays valid extents are {(1, 2048), (1,2048), (1, 2048)}.

Note: That because of the differing extent limits it may be advantageous to use a degenerate array (withunused dimensions set to one) of higher dimensionality. For instance, a degenerate 2D array allows forsignificantly more linear storage than a 1D array.

RETURN VALUE


cudaSuccess



48

SEE ALSO

cudaMalloc3D, cudaMalloc, cudaMallocPitch, cudaFree, cudaFreeArray, cudaMallocHost, cudaFreeHost

49

1.5.24 cudaMemset3D

NAME

cudaMemset3D - initializes or sets GPU memory to a value

SYNOPSIS

struct cudaPitchedPtr {

void *ptr;

size_t pitch;

size_t xsize;

size_t ysize;

};

struct cudaExtent {

size_t width;

size_t height;

size_t depth;

};

cudaError_t cudaMemset3D( struct cudaPitchedPtr dstPitchPtr, int value, struct cudaExtent extent

)

DESCRIPTION

Initializes each element of a 3D array to the specified value value. The object to initialize is defined bydstPitchPtr. The pitch field of dstPitchPtr is the width in memory in bytes of the 3D array pointed toby dstPitchPtr, including any padding added to the end of each row. The xsize field specifies the logicalwidth of each row in bytes, while the ysize field specifies the height of each 2D slice in rows.

The extents of the initialized region are specified as a width in bytes, a height in rows, and a depth in slices.

Extents with width greater than or equal to the xsize of dstPitchPtr may perform significantly faster thanextents narrower than the xsize. Secondarily, extents with height equal to the ysize of dstPitchPtr willperform faster than when the hieght is shorter than the ysize.

This function performs fastest when the dstPitchPtr has been allocated by cudaMalloc3D().

RETURN VALUE


cudaSuccess




50

SEE ALSO

cudaMemset, cudaMemset2D, cudaMalloc3D

51

1.5.25 cudaMemcpy3D

NAME

cudaMemcpy3D - copies data between between 3D objects

SYNOPSIS

struct cudaExtent {

size_t width, height, depth;

};

struct cudaPos {

size_t x, y, z;

};

struct cudaMemcpy3DParms {

struct cudaArray *srcArray;

struct cudaPos srcPos;

struct cudaPitchedPtr srcPtr;

struct cudaArray *dstArray;

struct cudaPos dstPos;

struct cudaPitchedPtr dstPtr;

struct cudaExtent extent;

enum cudaMemcpyKind kind;

};

cudaError_t cudaMemcpy3D( const struct cudaMemcpy3DParms *p )

cudaError_t cudaMemcpy3DAsync( const struct cudaMemcpy3DParms *p, cudaStream_t stream )

DESCRIPTION

cudaMemcpy3D() copies data betwen two 3D objects. The source and destination objects may be ineither host memory, device memory, or a CUDA array. The source, destination, extent, and kind of copyperformed is specified by the cudaMemcpy3DParms struct which should be initialized to zero before use:

cudaMemcpy3DParms myParms = {0};

The struct passed to cudaMemcpy3D() must specify one of srcArray or srcPtr and one of dstArray ordstPtr. Passing more than one non-zero source or destination will cause cudaMemcpy3D() to return anerror.

The srcPos and dstPos fields are optional offsets into the source and destination objects and are defined inunits of each object’s elements. The element for a host or device pointer is assumed to be unsigned char.For CUDA arrays, positions must be in the range [0, 2048) for any dimension.

The extent field defines the dimensions of the transferred area in elements. If a CUDA array is participatingin the copy the extent is defined in terms of that array’s elements. If no CUDA array is participating in thecopy then the extents are defined in elements of unsigned char.

The kind field defines the direction of the copy. It must be one of cudaMemcpyHostToHost, cudaMem-cpyHostToDevice, cudaMemcpyDeviceToHost, or cudaMemcpyDeviceToDevice.

52

If the source and destination are both arrays cudaMemcpy3D() will return an error if they do not havethe same element size.

The source and destination object may not overlap. If overlapping source and destination objects are specifiedundefined behavior will result.

cudaMemcpy3D() returns an error if the pitch of srcPtr or dstPtr is greater than the maximum allowed.The pitch of a cudaPitchedPtr allocated with cudaMalloc3D() will always be valid.

cudaMemcpy3DAsync() is an asynchronous copy operation and can optionally be associated to a streamby passing a non-zero stream argument. If either the source or destination is a host object it must beallocated in page-locked memory returned from cudaMallocHost(). It will return an error if a pointer tomemory not allocated with cudaMallocHost() is passed as input.

RETURN VALUE

cudaSuccess

SEE ALSO

cudaMalloc3D, cudaMalloc3DArray, cudaMemset3D, cudaMemcpy, cudaMemcpyToArray, cudaMemcpy2DToArray,cudaMemcpyFromArray, cudaMemcpy2DFromArray, cudaMemcpyArrayToArray, cudaMemcpy2DArrayToArray,cudaMemcpyToSymbol, cudaMemcpyFromSymbol

53

1.6 TextureReferenceManagement RT

NAME

Texture Reference Management

DESCRIPTION


Low-Level API

High-Level API

SEE ALSO


54

1.6.1 LowLevelApi

NAME

Low-Level Texture API

DESCRIPTION

This section describes the low-level CUDA run-time application programming interface for textures

cudaCreateChannelDesc

cudaGetChannelDesc

cudaGetTextureReference

cudaBindTexture

cudaBindTextureToArray

cudaUnbindTexture

cudaGetTextureAlignmentOffset

SEE ALSO

High-Level API

55


NAME

cudaCreateChannelDesc - Low-level texture API

SYNOPSIS

struct cudaChannelFormatDesc cudaCreateChannelDesc(int x, int y, int z, int w, enum cudaChannelFormatKind

f);

DESCRIPTION

Returns a channel descriptor with format f and number of bits of each component x, y, z, and w. ThecudaChannelFormatDesc is defined as:


int x, y, z, w;


};


RETURN VALUE


cudaSuccess


SEE ALSO

cudaGetChannelDesc, cudaGetTextureReference, cudaBindTexture, cudaBindTextureToArray, cudaUnbind-

Texture, cudaGetTextureAlignmentOffset

56

cudaGetChannelDesc

NAME

cudaGetChannelDesc - Low-level texture API

SYNOPSIS

cudaError_t cudaGetChannelDesc(struct cudaChannelFormatDesc* desc, const struct cudaArray*

array);

DESCRIPTION

Returns in *desc the channel descriptor of the CUDA array array.

RETURN VALUE


cudaSuccess



SEE ALSO

cudaCreateChannelDesc, cudaGetTextureReference, cudaBindTexture, cudaBindTextureToArray, cudaUn-

bindTexture, cudaGetTextureAlignmentOffset

57

cudaGetTextureReference

NAME

cudaGetTextureReference - Low-level texture API

SYNOPSIS

cudaError_t cudaGetTextureReference( struct textureReference** texRef, const char* symbol)

DESCRIPTION

Returns in *texRef the structure associated to the texture reference defined by symbol symbol.

RETURN VALUE


cudaSuccess

cudaErrorInvalidTexture


SEE ALSO

cudaCreateChannelDesc, cudaGetChannelDesc, cudaBindTexture, cudaBindTextureToArray, cudaUnbind-


58

cudaBindTexture

NAME

cudaBindTexture - Low-level texture API

SYNOPSIS

cudaError_t cudaBindTexture(size_t* offset, const struct textureReference* texRef, const void*

devPtr, const struct cudaChannelFormatDesc* desc, size_t size = UINT_MAX);

DESCRIPTION

Binds size bytes of the memory area pointed to by devPtr to the texture reference texRef. desc describeshow the memory is interpreted when fetching values from the texture. Any memory previously bound totexRef is unbound.

Since the hardware enforces an alignment requirement on texture base addresses, cudaBindTexture()returns in *offset a byte offset that must be applied to texture fetches in order to read from the desiredmemory. This offset must be divided by the texel size and passed to kernels that read from the textureso they can be applied to the tex1Dfetch() function. If the device memory pointer was returned fromcudaMalloc(), the offset is guaranteed to be 0 and NULL may be passed as the offset parameter.

RETURN VALUE


cudaSuccess





SEE ALSO

cudaCreateChannelDesc, cudaGetChannelDesc, cudaGetTextureReference, cudaBindTextureToArray, cud-

aUnbindTexture, cudaGetTextureAlignmentOffset

59


NAME

cudaBindTextureToArray - Low-level texture API

SYNOPSIS

cudaError_t cudaBindTextureToArray( const struct textureReference* texRef, const struct cudaArray*

array, const struct cudaChannelFormatDesc* desc);

DESCRIPTION

Binds the CUDA array array to the texture reference texRef. desc describes how the memory is interpretedwhen fetching values from the texture. Any CUDA array previously bound to texRef is unbound.

RETURN VALUE


cudaSuccess





SEE ALSO

cudaCreateChannelDesc, cudaGetChannelDesc, cudaGetTextureReference, cudaBindTexture, cudaUnbind-


60

cudaUnbindTexture

NAME

cudaUnbindTexture - Low-level texture API

SYNOPSIS

cudaError_t cudaUnbindTexture( const struct textureReference* texRef);

DESCRIPTION

Unbinds the texture bound to texture reference texRef.

RETURN VALUE


cudaSuccess


SEE ALSO

cudaCreateChannelDesc, cudaGetChannelDesc, cudaGetTextureReference, cudaBindTexture, cudaBindTex-

tureToArray, cudaGetTextureAlignmentOffset

61

cudaGetTextureAlignmentOffset

NAME

cudaGetTextureAlignmentOffset - Low-level texture API

SYNOPSIS

cudaError_t cudaGetTextureAlignmentOffset(size_t* offset, const struct textureReference* texRef);

DESCRIPTION

Returns in *offset the offset that was returned when texture reference texRef was bound.

RETURN VALUE


cudaSuccess


cudaErrorInvalidTextureBinding


SEE ALSO

cudaCreateChannelDesc, cudaGetChannelDesc, cudaGetTextureReference, cudaBindTexture, cudaBindTex-

tureToArray, cudaUnbindTexture

62

1.6.2 HighLevelApi

NAME

High-Level Texture API

DESCRIPTION

This section describes the high-level CUDA run-time application programming interface for textures


cudaBindTexture


cudaUnbindTexture

SEE ALSO

Low-Level API

63

cudaCreateChannelDesc HL

NAME

cudaCreateChannelDesc - High-level texture API

SYNOPSIS


struct cudaChannelFormatDesc cudaCreateChannelDesc <T >();

DESCRIPTION

Returns a channel descriptor with format f and number of bits of each component x, y, z, and w. ThecudaChannelFormatDesc is defined as:


int x, y, z, w;


};


RETURN VALUE


cudaSuccess


SEE ALSO

64

cudaBindTexture HL

NAME

cudaBindTexture - High-level texture API

SYNOPSIS

template < class T, int dim, enum cudaTextureReadMode readMode >

static __inline__ __host__ cudaError_t cudaBindTexture(size_t* offset, const struct texture

< T, dim, readMode >& texRef, const void* devPtr, const struct cudaChannelFormatDesc& desc,

size_t size = UINT_MAX)

DESCRIPTION

Binds size bytes of the memory area pointed to by devPtr to texture reference texRef. desc describeshow the memory is interpreted when fetching values from the texture. The offset parameter is an optionalbyte offset as with the low-level cudaBindTexture() function. Any memory previously bound to texRefis unbound.

template E<lt> class T, int dim, enum cudaTextureReadMode readMode E<gt>

static __inline__ __host__ cudaError_t cudaBindTexture(

size_t* offset,

const struct texture E<lt> T, dim, readMode E<gt>& texRef,

const void* devPtr,

size_t size = UINT_MAX);

binds size bytes of the memory area pointed to by devPtr to texture reference texRef. The channel descriptoris inherited from the texture reference type. The offset parameter is an optional byte offset as with the low-level cudaBindTexture() function described

RETURN VALUE


cudaSuccess





SEE ALSO

65

cudaBindTextureToArray HL

NAME

cudaBindTextureToArray - High-level texture API

SYNOPSIS


static __inline__ __host__ cudaError_t cudaBindTextureToArray( const struct texture < T, dim,

readMode >& texRef, const struct cudaArray* cuArray, const struct cudaChannelFormatDesc& desc)

DESCRIPTION

Binds the CUDA array array to texture reference texRef. desc describes how the memory is interpretedwhen fetching values from the texture. Any CUDA array previously bound to texRef is unbound.

template E<lt> class T, int dim, enum cudaTextureReadMode readMode E<gt>

static __inline__ __host__ cudaError_t cudaBindTextureToArray(

const struct texture E<lt> T, dim, readMode E<gt>& texRef,

const struct cudaArray* cuArray);

binds the CUDA array array to texture reference texRef. The channel descriptor is inherited from theCUDA array. Any CUDA array previously bound to texRef is unbound.

RETURN VALUE


cudaSuccess





SEE ALSO

cudaCreateChannelDesc_HL

66

cudaUnbindTexture HL

NAME

cudaUnbindTexture - High-level texture API

SYNOPSIS


static __inline__ __host__ cudaError_t cudaUnbindTexture(const struct texture < T, dim, readMode

>& texRef)

DESCRIPTION

Unbinds the texture bound to texture reference texRef.

RETURN VALUE


cudaSuccess


SEE ALSO

67

1.7 ExecutionControl RT

NAME

Execution Control

DESCRIPTION


cudaConfigureCall

cudaLaunch

cudaSetupArgument

SEE ALSO


68

1.7.1 cudaConfigureCall

NAME

cudaConfigureCall - configure a device-launch

SYNOPSIS

cudaError_t cudaConfigureCall(dim3 gridDim, dim3 blockDim, size_t sharedMem = 0, int tokens

= 0)

DESCRIPTION

Specifies the grid and block dimensions for the device call to be executed similar to the execution configurationsyntax. cudaConfigureCall() is stack based. Each call pushes data on top of an execution stack. Thisdata contains the dimension for the grid and thread blocks, together with any arguments for the call.

RETURN VALUE


cudaSuccess

cudaErrorInvalidConfiguration


SEE ALSO

cudaLaunch, cudaSetupArgument

69

1.7.2 cudaLaunch

NAME

cudaLaunch - launches a device function

SYNOPSIS

template < class T > cudaError_t cudaLaunch(T entry)

DESCRIPTION

Launches the function entry on the device. entry can either be a function that executes on the device,or it can be a character string, naming a function that executes on the device. entry must be declared asa __global__ function. cudaLaunch() must be preceded by a call to cudaConfigureCall() since itpops the data that was pushed by cudaConfigureCall() from the execution stack.

RETURN VALUE


cudaSuccess

cudaErrorInvalidDeviceFunction



SEE ALSO

cudaConfigureCall, cudaSetupArgument

70

1.7.3 cudaSetupArgument

NAME

cudaSetupArgument - configure a device-launch

SYNOPSIS

cudaError_t cudaSetupArgument(void* arg, size_t count, size_t offset)

template < class T > cudaError_t cudaSetupArgument(T arg, size_t offset)

DESCRIPTION

Pushes count bytes of the argument pointed to by arg at offset bytes from the start of the parameterpassing area, which starts at offset 0. The arguments are stored in the top of the execution stack. cudaSe-tupArgument() must be preceded by a call to cudaConfigureCall().

RETURN VALUE


cudaSuccess


SEE ALSO

cudaConfigureCall, cudaLaunch

71

1.8 OpenGlInteroperability RT

NAME

OpenGL Interoperability

DESCRIPTION


cudaGLSetGLDevice

cudaGLRegisterBufferObject

cudaGLMapBufferObject

cudaGLUnmapBufferObject

cudaGLUnregisterBufferObject

SEE ALSO


72

1.8.1 cudaGLSetGLDevice

NAME

cudaGLSetGLDevice - sets the CUDA device for use with GL Interopability

SYNOPSIS

cudaError_t cudaGLSetGLDevice(int device);

DESCRIPTION

Records dev as the device on which the active host thread executes the device code. Records the thread asusing GL Interopability.

RETURN VALUE


cudaSuccess



SEE ALSO

cudaGLRegisterBufferObject, cudaGLMapBufferObject, cudaGLUnmapBufferObject, cudaGLUnregisterBuffer-

Object

73

1.8.2 cudaGLRegisterBufferObject

NAME

cudaGLRegisterBufferObject - OpenGL interoperability

SYNOPSIS

cudaError_t cudaGLRegisterBufferObject(GLuint bufferObj)

DESCRIPTION

Registers the buffer object of ID bufferObj for access by CUDA. This function must be called before CUDAcan map the buffer object. While it is registered, the buffer object cannot be used by any OpenGL commandsexcept as a data source for OpenGL drawing commands.

RETURN VALUE


cudaSuccess

cudaErrorNotInitialized


SEE ALSO

cudaGLSetGLDevice, cudaGLMapBufferObject, cudaGLUnmapBufferObject, cudaGLUnregisterBufferObject

74

1.8.3 cudaGLMapBufferObject

NAME

cudaGLMapBufferObject - OpenGL interoperability

SYNOPSIS

cudaError_t cudaGLMapBufferObject(void** devPtr, GLuint bufferObj);

DESCRIPTION

Maps the buffer object of ID bufferObj into the address space of CUDA and returns in *devPtr the basepointer of the resulting mapping.

RETURN VALUE


cudaSuccess

cudaErrorMapBufferObjectFailed


SEE ALSO

cudaGLSetGLDevice, cudaGLRegisterBufferObject, cudaGLUnmapBufferObject, cudaGLUnregisterBuffer-

Object

75

1.8.4 cudaGLUnmapBufferObject

NAME

cudaGLUnmapBufferObject - OpenGL interoperability

SYNOPSIS

cudaError_t cudaGLUnmapBufferObject(GLuint bufferObj);

DESCRIPTION

Unmaps the buffer object of ID bufferObj for access by CUDA.

RETURN VALUE


cudaSuccess


cudaErrorUnmapBufferObjectFailed


SEE ALSO

cudaGLSetGLDevice, cudaGLRegisterBufferObject, cudaGLMapBufferObject, cudaGLUnregisterBufferOb-

ject

76

1.8.5 cudaGLUnregisterBufferObject

NAME

cudaGLUnregisterBufferObject - OpenGL interoperability

SYNOPSIS

cudaError_t cudaGLUnregisterBufferObject(GLuint bufferObj);

DESCRIPTION

Unregisters the buffer object of ID bufferObj for access by CUDA.

RETURN VALUE


cudaSuccess


SEE ALSO

cudaGLSetGLDevice, cudaGLRegisterBufferObject, cudaGLMapBufferObject, cudaGLUnmapBufferObject

77

1.9 Direct3dInteroperability RT

NAME

Direct3D Interoperability

DESCRIPTION


cudaD3D9GetDevice

cudaD3D9SetDirect3DDevice

cudaD3D9GetDirect3DDevice

cudaD3D9RegisterResource

cudaD3D9UnregisterResource

cudaD3D9MapResources

cudaD3D9UnmapResources

cudaD3D9ResourceGetSurfaceDimensions

cudaD3D9ResourceSetMapFlags

cudaD3D9ResourceGetMappedPointer

cudaD3D9ResourceGetMappedSize

cudaD3D9ResourceGetMappedPitch

As of CUDA 2.0 the following functions are deprecated. They should not be used in new development.

cudaD3D9Begin

cudaD3D9End

cudaD3D9RegisterVertexBuffer

cudaD3D9MapVertexBuffer

cudaD3D9UnmapVertexBuffer

cudaD3D9UnregisterVertexBuffer

SEE ALSO


78

1.9.1 cudaD3D9GetDevice

NAME

cudaD3D9GetDevice - gets the device number for an adapter

SYNOPSIS

cudaError_t cudaD3D9GetDevice(int* dev, const char* adapterName);

DESCRIPTION

Returns in *dev the CUDA-compatible device corresponding to the adapter name adapterName obtainedfrom EnumDisplayDevices or IDirect3D9::GetAdapterIdentifier(). If no device on the adapter withname adapterName is CUDA-compatible then the call will fail.

RETURN VALUE


cudaSuccess


cudaErrorUnknown


SEE ALSO

cudaD3D9SetDirect3DDevice, cudaD3D9GetDirect3DDevice, cudaD3D9RegisterResource, cudaD3D9UnregisterResource,cudaD3D9MapResources, cudaD3D9UnmapResources, cudaD3D9ResourceGetSurfaceDimensions, cudaD3D9ResourceSetMapFlags

cudaD3D9ResourceGetMappedPointer, cudaD3D9ResourceGetMappedSize, cudaD3D9ResourceGetMappedPitch

79

1.9.2 cudaD3D9SetDirect3DDevice

NAME

cudaD3D9SetDirect3DDevice - sets the Direct3D device to use for interoperability in this thread

SYNOPSIS

cudaError_t cudaD3D9SetDirect3DDevice(IDirect3DDevice9* pDxDevice);

DESCRIPTION

Records pDxDevice as the Direct3D device to use for Direct3D interoperability on this host thread. Inorder to use Direct3D interoperability, this call must be made before any other CUDA runtime calls on thisthread.

Successful context creation on pDxDevice will increase the internal reference count on pDxDevice. Thisreference count will be decremented upon destruction of this context through cudaThreadExit.

RETURN VALUE


cudaSuccess




SEE ALSO

cudaD3D9GetDevice, cudaD3D9GetDirect3DDevice, cudaD3D9RegisterResource, cudaD3D9UnregisterResource,cudaD3D9MapResources, cudaD3D9UnmapResources, cudaD3D9ResourceGetSurfaceDimensions, cudaD3D9ResourceSetMapFlags


80

1.9.3 cudaD3D9GetDirect3DDevice

NAME

cudaD3D9GetDirect3DDevice - get the Direct3D device against which the current CUDA context wascreated

SYNOPSIS

cudaError_t cudaD3D9GetDirect3DDevice(IDirect3DDevice9** ppDxDevice);

DESCRIPTION

Returns in *ppDxDevice the Direct3D device against which this CUDA context was created in cu-daD3D9SetDirect3DDevice.

RETURN VALUE


cudaSuccess

cudaErrorUnknown


SEE ALSO

cudaD3D9GetDevice, cudaD3D9SetDirect3DDevice, cudaD3D9RegisterResource, cudaD3D9UnregisterResource,cudaD3D9MapResources, cudaD3D9UnmapResources, cudaD3D9ResourceGetSurfaceDimensions, cudaD3D9ResourceSetMapFlags


81

1.9.4 cudaD3D9RegisterResource

NAME

cudaD3D9RegisterResource - register a Direct3D resource for access by CUDA

SYNOPSIS

cudaError_t cudaD3D9RegisterResource(IDirect3DResource9* pResource, unsigned int Flags);

DESCRIPTION

Registers the Direct3D resource pResource for access by CUDA.

If this call is successful then the application will be able to map and unmap this resource until it is unreg-istered through cudaD3D9UnregisterResource. Also on success, this call will increase the internal referencecount on pResource. This reference count will be decremented when this resource is unregistered throughcudaD3D9UnregisterResource.

This call is potentially high-overhead and should not be called every frame in interactive applications.

The type of pResource must be one of the following.

• IDirect3DVertexBuffer9: No notes.

• IDirect3DIndexBuffer9: No notes.

• IDirect3DSurface9: Only stand-alone objects of type IDirect3DSurface9 may be explicitly shared.In particular, individual mipmap levels and faces of cube maps may not be registered directly. To accessindividual surfaces associated with a texture, one must register the base texture object.

• IDirect3DBaseTexture9: When a texture is registered all surfaces associated with the all mipmaplevels of all faces of the texture will be accessible to CUDA.

The Flags argument specifies the mechanism through which CUDA will access the Direct3D resource. Thefollowing value is allowed.

• cudaD3D9RegisterFlagsNone: Specifies that CUDA will access this resource through a void*. Thepointer, size, and pitch for each subresource of this resource may be queried through cudaD3D9ResourceGetMappedPointer

cudaD3D9ResourceGetMappedSize, and cudaD3D9ResourceGetMappedPitch respectively. This optionis valid for all resource types.

Not all Direct3D resources of the above types may be used for interoperability with CUDA. The followingare some limitations.

• The primary rendertarget may not be registered with CUDA.

• Resources allocated as shared may not be registered with CUDA.

• Any resources allocated in D3DPOOL_SYSTEMMEM may not be registered with CUDA.

• Textures which are not of a format which is 1, 2, or 4 channels of 8, 16, or 32-bit integer or floating-pointdata cannot be shared.

82

• Surfaces of depth or stencil formats cannot be shared.

If Direct3D interoperability is not initialized on this context then is returned. If pResource is of incorrecttype (e.g, is a non-stand-alone IDirect3DSurface9) or is already registered then cudaErrorInvalidHan-dle is returned. If pResource cannot be registered then cudaErrorUnknown is returned.

RETURN VALUE


cudaSuccess


cudaErrorInvalidHandle

cudaErrorUnknown


SEE ALSO

cudaD3D9GetDevice, cudaD3D9SetDirect3DDevice, cudaD3D9GetDirect3DDevice, cudaD3D9UnregisterResource,cudaD3D9MapResources, cudaD3D9UnmapResources, cudaD3D9ResourceGetSurfaceDimensions, cudaD3D9ResourceSetMapFlags


83

1.9.5 cudaD3D9UnregisterResource

NAME

cudaD3D9UnregisterResource - unregister a Direct3D resource

SYNOPSIS

cudaError_t cudaD3D9UnregisterResource(IDirect3DResource9* pResource);

DESCRIPTION

Unregisters the Direct3D resource pResource so it is not accessable by CUDA unless registered again.

If pResource is not registered then cudaErrorInvalidHandle is returned.

RETURN VALUE


cudaSuccess


cudaErrorUnknown


SEE ALSO

cudaD3D9GetDevice, cudaD3D9SetDirect3DDevice, cudaD3D9GetDirect3DDevice, cudaD3D9RegisterResource,cudaD3D9MapResources, cudaD3D9UnmapResources, cudaD3D9ResourceGetSurfaceDimensions, cudaD3D9ResourceSetMapFlags


84

1.9.6 cudaD3D9MapResources

NAME

cudaD3D9MapResources - map Direct3D resources for access by CUDA

SYNOPSIS

cudaError_t cudaD3D9MapResources(unsigned int count, IDirect3DResource9 **ppResources);

DESCRIPTION

Maps the count Direct3D resources in ppResources for access by CUDA.

The resources in ppResources may be accessed in CUDA kernels until they are unmapped. Direct3Dshould not access any resources while they are mapped by CUDA. If an application does so the results areundefined.

This function provides the synchronization guarantee that any Direct3D calls issued before cudaD3D9MapResourceswill complete before any CUDA kernels issued after cudaD3D9MapResources begin.

If any of ppResources have not been registered for use with CUDA or if ppResources contains anyduplicate entries then cudaErrorInvalidHandle is returned. If any of ppResources are presently mappedfor access by CUDA then cudaErrorUnknown is returned.

RETURN VALUE


cudaSuccess


cudaErrorUnknown


SEE ALSO

cudaD3D9GetDevice, cudaD3D9SetDirect3DDevice, cudaD3D9GetDirect3DDevice, cudaD3D9RegisterResource,cudaD3D9UnregisterResource, cudaD3D9UnmapResources, cudaD3D9ResourceGetSurfaceDimensions, cu-

daD3D9ResourceSetMapFlags, cudaD3D9ResourceGetMappedPointer, cudaD3D9ResourceGetMappedSize, cu-

daD3D9ResourceGetMappedPitch

85

1.9.7 cudaD3D9UnmapResources

NAME

cudaD3D9UnmapResources - unmap Direct3D resources

SYNOPSIS

cudaError_t cudaD3D9UnmapResources(unsigned int count, IDirect3DResource9** ppResources);

DESCRIPTION

Unmaps the count Direct3D resources in ppResources.

This function provides the synchronization guarantee that any CUDA kernels issued before cudaD3D9UnmapResourceswill complete before any Direct3D calls issued after cudaD3D9UnmapResources begin.

If any of ppResources have not been registered for use with CUDA or if ppResources contains anyduplicate entries then cudaErrorInvalidHandle is returned. If any of ppResources are not presentlymapped for access by CUDA then cudaErrorUnknown is returned.

RETURN VALUE


cudaSuccess


cudaErrorUnknown


SEE ALSO

cudaD3D9GetDevice, cudaD3D9SetDirect3DDevice, cudaD3D9GetDirect3DDevice, cudaD3D9RegisterResource,cudaD3D9UnregisterResource, cudaD3D9MapResources, cudaD3D9ResourceGetSurfaceDimensions, cudaD3D9ResourceSetMapFlags


86

1.9.8 cudaD3D9ResourceSetMapFlags

NAME

cudaD3D9ResourceSetMapFlags - set usage flags for mapping a Direct3D resource

SYNOPSIS

cudaError_t cudaD3D9ResourceSetMapFlags(IDirect3DResource9 *pResource, unsigned int Flags);

DESCRIPTION

Set flags for mapping the Direct3D resource pResource.

Changes to flags will take effect the next time pResource is mapped. The Flags argument may be any ofthe following.

• cudaD3D9MapFlagsNone: Specifies no hints about how this resource will be used. It is thereforeassumed that this resource will be read from and written to by CUDA kernels. This is the defaultvalue.

• cudaD3D9MapFlagsReadOnly: Specifies that CUDA kernels which access this resource will notwrite to this resource.

• cudaD3D9MapFlagsWriteDiscard: Specifies that CUDA kernels which access this resource willnot read from this resource and will write over the entire contents of the resource, so none of the datapreviously stored in the resource will be preserved.

If pResource has not been registered for use with CUDA then cudaErrorInvalidHandle is returned. IfpResource is presently mapped for access by CUDA then cudaErrorUnknown is returned.

RETURN VALUE


cudaSuccess



cudaErrorUnknown


87

SEE ALSO

cudaD3D9GetDevice, cudaD3D9SetDirect3DDevice, cudaD3D9GetDirect3DDevice, cudaD3D9RegisterResource,cudaD3D9UnregisterResource, cudaD3D9MapResources, cudaD3D9UnmapResources, cudaD3D9ResourceGetSurfaceDimensions


88

1.9.9 cudaD3D9ResourceGetSurfaceDimensions

NAME

cudaD3D9ResourceGetSurfaceDimensions - get the dimensions of a registered surface

SYNOPSIS

cudaError_t cudaD3D9ResourceGetSurfaceDimensions(size_t* pWidth, size_t* pHeight, size_t *pDepth,

IDirect3DResource9* pResource, unsigned int Face, unsigned int Level);

DESCRIPTION

Returns in *pWidth, *pHeight, and *pDepth the dimensions of the subresource of the mapped Direct3Dresource pResource which corresponds to Face and Level.

Because anti-aliased surfaces may have multiple samples per pixel it is possible that the dimensions of aresource will be an integer factor larger than the dimensions reported by the Direct3D runtime.

The parameters pWidth, pHeight, and pDepth are optional. For 2D surfaces, the value returned in*pDepth will be 0.

If pResource is not of type IDirect3DBaseTexture9 or IDirect3DSurface9 or if pResource has notbeen registered for use with CUDA then cudaErrorInvalidHandle is returned.

For usage requirements of Face and Level parameters see cudaD3D9ResourceGetMappedPointer.

RETURN VALUE


cudaSuccess




SEE ALSO

cudaD3D9GetDevice, cudaD3D9SetDirect3DDevice, cudaD3D9GetDirect3DDevice, cudaD3D9RegisterResource,cudaD3D9UnregisterResource, cudaD3D9MapResources, cudaD3D9UnmapResources, cudaD3D9ResourceSetMapFlags,cudaD3D9ResourceGetMappedPointer, cudaD3D9ResourceGetMappedSize, cudaD3D9ResourceGetMappedPitch

89

1.9.10 cudaD3D9ResourceGetMappedPointer

NAME

cudaD3D9ResourceGetMappedPointer - get a pointer through which to access a subresource of aDirect3D resource which has been mapped for access by CUDA

SYNOPSIS

cudaError_t cudaD3D9ResourceGetMappedPointer(void** pPointer, IDirect3DResource9* pResource,

unsigned int Face, unsigned int Level);

DESCRIPTION

Returns in *pPointer the base pointer of the subresource of the mapped Direct3D resource pResourcewhich corresponds to Face and Level. The value set in pPointer may change every time that pResourceis mapped.

If pResource is not registered then cudaErrorInvalidHandle is returned. If pResource was not reg-istered with usage flags cudaD3D9RegisterFlagsNone then cudaErrorInvalidHandle is returned. IfpResource is not mapped then cudaErrorUnknown is returned.

If pResource is of type IDirect3DCubeTexture9 then Face must one of the values enumerated by typeD3DCUBEMAP_FACES. For all other types Face must be 0. If Face is invalid then cudaErrorIn-validValue is returned.

If pResource is of type IDirect3DBaseTexture9 then Level must correspond to a valid mipmap level.Only mipmap level 0 is supported for now. For all other types Level must be 0. If Level is invalid thencudaErrorInvalidValue is returned.

RETURN VALUE


cudaSuccess



cudaErrorUnknown


SEE ALSO


cudaD3D9ResourceSetMapFlags, cudaD3D9ResourceGetMappedSize, cudaD3D9ResourceGetMappedPitch

90

1.9.11 cudaD3D9ResourceGetMappedSize

NAME

cudaD3D9ResourceGetMappedSize - get the size of a subresource of a Direct3D resource which hasbeen mapped for access by CUDA

SYNOPSIS

cudaError_t cudaD3D9ResourceGetMappedSize(size_t* pSize, IDirect3DResource9* pResource, unsigned

int Face, unsigned int Level);

DESCRIPTION

Returns in *pSize the size of the subresource of the mapped Direct3D resource pResource which corre-sponds to Face and Level. The value set in pSize may change every time that pResource is mapped.

If pResource has not been registered for use with CUDA then cudaErrorInvalidHandle is returned.If pResource was not registered with usage flags cudaD3D9RegisterFlagsNone then cudaErrorIn-validHandle is returned. If pResource is not mapped for access by CUDA then cudaErrorUnknown isreturned.


RETURN VALUE


cudaSuccess



cudaErrorUnknown


SEE ALSO


cudaD3D9ResourceSetMapFlags, cudaD3D9ResourceGetMappedPointer, cudaD3D9ResourceGetMappedPitch

91

1.9.12 cudaD3D9ResourceGetMappedPitch

NAME

cudaD3D9ResourceGetMappedPitch - get the pitch of a subresource of a Direct3D resource which hasbeen mapped for access by CUDA

SYNOPSIS

cudaError_t cudaD3D9ResourceGetMappedPitch(size_t* pPitch, size_t* pPitchSlice, IDirect3DResource9*

pResource, unsigned int Face, unsigned int Level);

DESCRIPTION

Returns in *pPitch and *pPitchSlice the pitch and Z-slice pitch of the subresource of the mapped Direct3Dresource pResource which corresponds to Face and Level. The values set in pPitch and pPitchSlicemay change every time that pResource is mapped.

The pitch and Z-slice pitch values may be used to compute the location of a sample on a surface as follows.

y*pitch + (bytes per pixel)*x

For a 3D surface the byte offset of the sample of at position x,y,z from the base pointer of the surface is

z*slicePitch + y*pitch + (bytes per pixel)*x

Both parameters pPitch and pPitchSlice are optional and may be set to NULL.

For a 2D surface the byte offset of the sample of at position x,y from the base pointer of the surface is

If pResource is not of type IDirect3DBaseTexture9 or one of its sub-types or if pResource has notbeen registered for use with CUDA then cudaErrorInvalidHandle is returned. If pResource was notregistered with usage flags cudaD3D9RegisterFlagsNone then cudaErrorInvalidHandle is returned.If pResource is not mapped for access by CUDA then cudaErrorUnknown is returned.


RETURN VALUE


cudaSuccess



cudaErrorUnknown


92

SEE ALSO


cudaD3D9ResourceSetMapFlags, cudaD3D9ResourceGetMappedPointer, cudaD3D9ResourceGetMappedSize

93

1.10 ErrorHandling RT

NAME

Error Handling

DESCRIPTION


cudaGetLastError

cudaGetErrorString

SEE ALSO


94

1.10.1 cudaGetLastError

NAME

cudaGetLastError - returns the last error from a run-time call

SYNOPSIS

cudaError_t cudaGetLastError( void )

DESCRIPTION

Returns the last error that was returned from any of the runtime calls in the same host thread and resets itto cudaSuccess.

RETURN VALUE


cudaSuccess


cudaErrorLaunchFailure


cudaErrorLaunchTimeout

cudaErrorLaunchOutOfResources

cudaErrorInvalidDeviceFunction






cudaErrorInvalidTextureBinding

cudaErrorInvalidChannelDescriptor

cudaErrorTextureFetchFailed

cudaErrorTextureNotBound

cudaErrorSynchronizationError

cudaErrorUnknown


cudaErrorNotReady

Note that this function may also return error codes from previous asynchronous launches.

95

SEE ALSO

cudaGetErrorString, cudaError

96

1.10.2 cudaGetErrorString

NAME

cudaGetErrorString - returns the message string from an error

SYNOPSIS

const char* cudaGetErrorString(cudaError_t error);

DESCRIPTION

Returns a message string from an error code.

RETURN VALUE

char* pointer to a NULL-terminated string

SEE ALSO

cudaGetLastError

97

2 DriverApiReference

NAME

Driver API Reference

DESCRIPTION

This section describes the low-level CUDA driver application programming interface.

SEE ALSO

Initialization, DeviceManagement, ContextManagement, ModuleManagement, StreamManagement, Event-

Management, ExecutionControl, MemoryManagement, TextureReferenceManagement, OpenGlInteroperabil-

ity, Direct3dInteroperability

98

2.1 Initialization

NAME

Driver Initialization

DESCRIPTION


cuInit

SEE ALSO




99

2.1.1 cuInit

NAME

cuInit - initialize the CUDA driver API

SYNOPSIS

CUresult cuInit( unsigned int Flags );

DESCRIPTION

Initializes the driver API and must be called before any other function from the driver API. Currently, theFlags parameters must be 0. If cuInit() has not been called, any function from the driver API will returnCUDA_ERROR_NOT_INITIALIZED.

RETURN VALUE


CUDA_SUCCESS

CUDA_ERROR_INVALID_VALUE

CUDA_ERROR_NO_DEVICE


SEE ALSO

100

2.2 DeviceManagement

NAME

Device Management

DESCRIPTION


cuDeviceComputeCapability

cuDeviceGet

cuDeviceGetAttribute

cuDeviceGetCount

cuDeviceGetName

cuDeviceGetProperties

cuDeviceTotalMem

SEE ALSO




101

2.2.1 cuDeviceComputeCapability

NAME

cuDeviceComputeCapability - returns the compute capability of the device

SYNOPSIS

CUresult cuDeviceComputeCapability(int* major, int* minor, CUdevice dev);

DESCRIPTION

Returns in *major and *minor the major and minor revision numbers that define the compute capabilityof device dev.

RETURN VALUE


CUDA_SUCCESS

CUDA_ERROR_DEINITIALIZED

CUDA_ERROR_NOT_INITIALIZED

CUDA_ERROR_INVALID_CONTEXT


CUDA_ERROR_INVALID_DEVICE


SEE ALSO

cuDeviceGetAttribute, cuDeviceGetCount, cuDeviceGetName, cuDeviceGet, cuDeviceGetProperties, cuDe-

viceTotalMem

102

2.2.2 cuDeviceGet

NAME

cuDeviceGet - returns a device-handle

SYNOPSIS

CUresult cuDeviceGet(CUdevice* dev, int ordinal);

DESCRIPTION

Returns in *dev a device handle given an ordinal in the range [0, cuDeviceGetCount()-1].

RETURN VALUE


CUDA_SUCCESS







SEE ALSO

cuDeviceComputeCapability, cuDeviceGetAttribute, cuDeviceGetCount, cuDeviceGetName, cuDeviceGetProp-

erties, cuDeviceTotalMem

103

2.2.3 cuDeviceGetAttribute

NAME

cuDeviceGetAttribute - returns information about the device

SYNOPSIS

CUresult cuDeviceGetAttribute(int* value, CUdevice_attribute attrib, CUdevice dev);

DESCRIPTION

Returns in *value the integer value of the attribute attrib on device dev. The supported attributes are:

• CU_DEVICE_ATTRIBUTE_MAX_THREADS_PER_BLOCK: maximum number of threadsper block;

• CU_DEVICE_ATTRIBUTE_MAX_BLOCK_DIM_X: maximum x-dimension of a block;

• CU_DEVICE_ATTRIBUTE_MAX_BLOCK_DIM_Y: maximum y-dimension of a block;

• CU_DEVICE_ATTRIBUTE_MAX_BLOCK_DIM_Z: maximum z-dimension of a block;

• CU_DEVICE_ATTRIBUTE_MAX_GRID_DIM_X: maximum x-dimension of a grid;

• CU_DEVICE_ATTRIBUTE_MAX_GRID_DIM_Y: maximum y-dimension of a grid;

• CU_DEVICE_ATTRIBUTE_MAX_GRID_DIM_Z: maximum z-dimension of a grid;

• CU_DEVICE_ATTRIBUTE_MAX_SHARED_MEMORY_PER_BLOCK: maximum amountof shared memory available to a thread block in bytes; this amount is shared by all thread blocks si-multaneously resident on a multiprocessor;

• CU_DEVICE_ATTRIBUTE_TOTAL_CONSTANT_MEMORY: total amount of constantmemory available on the device in bytes;

• CU_DEVICE_ATTRIBUTE_WARP_SIZE: warp size in threads;

• CU_DEVICE_ATTRIBUTE_MAX_PITCH: maximum pitch in bytes allowed by the memorycopy functions that involve memory regions allocated through cuMemAllocPitch();

• CU_DEVICE_ATTRIBUTE_MAX_REGISTERS_PER_BLOCK: maximum number of32-bit registers available to a thread block; this number is shared by all thread blocks simultaneouslyresident on a multiprocessor;

• CU_DEVICE_ATTRIBUTE_CLOCK_RATE: clock frequency in kilohertz;

• CU_DEVICE_ATTRIBUTE_TEXTURE_ALIGNMENT: alignment requirement; texturebase addresses aligned to textureAlign bytes do not need an offset applied to texture fetches;

• CU_DEVICE_ATTRIBUTE_GPU_OVERLAP: 1 if the device can concurrently copy mem-ory between host and device while executing a kernel, or 0 if not;

• CU_DEVICE_ATTRIBUTE_MULTIPROCESSOR_COUNT: number of multiprocessorson the device.

104

RETURN VALUE


CUDA_SUCCESS







SEE ALSO

cuDeviceComputeCapability, cuDeviceGetCount, cuDeviceGetName, cuDeviceGet, cuDeviceGetProperties,cuDeviceTotalMem

105

2.2.4 cuDeviceGetCount

NAME

cuDeviceGetCount - returns the number of compute-capable devices

SYNOPSIS

CUresult cuDeviceGetCount(int* count);

DESCRIPTION

Returns in *count the number of devices with compute capability greater or equal to 1.0 that are availablefor execution. If there is no such device, cuDeviceGetCount() returns 0.

RETURN VALUE


CUDA_SUCCESS






SEE ALSO

cuDeviceComputeCapability, cuDeviceGetAttribute, cuDeviceGetName, cuDeviceGet, cuDeviceGetProperties,cuDeviceTotalMem

106

2.2.5 cuDeviceGetName

NAME

cuDeviceGetName - returns an identifier string

SYNOPSIS

CUresult cuDeviceGetName(char* name, int len, CUdevice dev);

DESCRIPTION

Returns an ASCII string identifying the device dev in the NULL-terminated string pointed to by name.len specifies the maximum length of the string that may be returned.

RETURN VALUE


CUDA_SUCCESS







SEE ALSO

cuDeviceComputeCapability, cuDeviceGetAttribute, cuDeviceGetCount, cuDeviceGet, cuDeviceGetProper-

ties, cuDeviceTotalMem

107

2.2.6 cuDeviceGetProperties

NAME

cuDeviceGetProperties - get device properties

SYNOPSIS

CUresult cuDeviceGetProperties(CUdevprop* prop, CUdevice dev);

DESCRIPTION

Returns in *prop the properties of device dev. The CUdevprop structure is defined as:

typedef struct CUdevprop_st {

int maxThreadsPerBlock;

int maxThreadsDim[3];

int maxGridSize[3];

int sharedMemPerBlock;

int totalConstantMemory;

int SIMDWidth;

int memPitch;

int regsPerBlock;

int clockRate;

int textureAlign

} CUdevprop;

where:

• maxThreadsPerBlock is the maximum number of threads per block;

• maxThreadsDim[3] is the maximum sizes of each dimension of a block;

• maxGridSize[3] is the maximum sizes of each dimension of a grid;

• sharedMemPerBlock is the total amount of shared memory available per block in bytes;

• totalConstantMemory is the total amount of constant memory available on the device in bytes;

• SIMDWidth is the warp size;

• memPitch is the maximum pitch allowed by the memory copy functions that involve memory regionsallocated through cuMemAllocPitch();

• regsPerBlock is the total number of registers available per block;

• clockRate is the clock frequency in kilohertz;

• textureAlign is the alignment requirement; texture base addresses that are aligned to textureAlignbytes do not need an offset applied to texture fetches.

108

RETURN VALUE


CUDA_SUCCESS







SEE ALSO

cuDeviceComputeCapability, cuDeviceGetAttribute, cuDeviceGetCount, cuDeviceGetName, cuDeviceGet, cuD-

eviceTotalMem

109

2.2.7 cuDeviceTotalMem

NAME

cuDeviceTotalMem - returns the total amount of memory on the device

SYNOPSIS

CUresult cuDeviceTotalMem( unsigned int* bytes, CUdevice dev );

DESCRIPTION

Returns in *bytes the total amount of memory available on the device dev in bytes.

RETURN VALUE


CUDA_SUCCESS







SEE ALSO

cuDeviceComputeCapability, cuDeviceGetAttribute, cuDeviceGetCount, cuDeviceGetName, cuDeviceGet, cuD-

eviceGetProperties

110

2.3 ContextManagement

NAME

Context Management

DESCRIPTION


cuCtxAttach

cuCtxCreate

cuCtxDestroy

cuCtxDetach

cuCtxGetDevice

cuCtxPopCurrent

cuCtxPushCurrent

cuCtxSynchronize

SEE ALSO




111

2.3.1 cuCtxAttach

NAME

cuCtxAttach - increment context usage-count

SYNOPSIS

CUresult cuCtxAttach(CUcontext* pCtx, unsigned int Flags);

DESCRIPTION

Increments the usage count of the context and passes back a context handle in *pCtx that must be passed tocuCtxDetach() when the application is done with the context. cuCtxAttach() fails if there is no contextcurrent to the thread.

Currently, the Flags parameter must be 0.

RETURN VALUE


CUDA_SUCCESS






SEE ALSO

cuCtxCreate, cuCtxDetach, cuCtxGetDevice, cuCtxSynchronize

112

2.3.2 cuCtxCreate

NAME

cuCtxCreate - create a CUDA context

SYNOPSIS

CUresult cuCtxCreate(CUcontext* pCtx, unsigned int Flags, CUdevice dev);

DESCRIPTION

Creates a new CUDA context and associates it with the calling thread. The Flags parameter is describedbelow. The context is created with a usage count of 1 and the caller of cuCtxCreate() must call cuCtxDe-stroy() or cuCtxDetach() when done using the context. If a context is already current to the thread, it issupplanted by the newly created context and may be restored by a subsequent call to cuCtxPopCurrent().

The two LSBs of the Flags parameter can be used to control how the OS thread which owns the CUDAcontext at the time of an API call interacts with the OS scheduler when waiting for results from the GPU.

• CU_CTX_SCHED_AUTO: The default value if the Flags parameter is zero, uses a heuristicbased on the number of active CUDA contexts in the process C and the number of logical processorsin the system P. If C > P then CUDA will yield to other OS threads when waiting for the GPU,otherwise CUDA will not yield while waiting for results and actively spin on the processor.

• CU_CTX_SCHED_SPIN: Instruct CUDA to actively spin when waiting for results from theGPU. This can decrease latency when waiting for the GPU, but may lower the performance of CPUthreads if they are performing work in parallel with the CUDA thread.

• CU_CTX_SCHED_YIELD: Instruct CUDA to yield its thread when waiting for results from theGPU. This can increase latency when waiting for the GPU, but can increase the performance of CPUthreads performing work in parallel with the GPU.

RETURN VALUE


CUDA_SUCCESS






CUDA_ERROR_OUT_OF_MEMORY

CUDA_ERROR_UNKNOWN


113

SEE ALSO

cuCtxAttach, cuCtxDetach, cuCtxDestroy, cuCtxPushCurrent, cuCtxPopCurrent

114

2.3.3 cuCtxDestroy

NAME

cuCtxDestroy - destroy the current context context or a floating CuDA context

SYNOPSIS

CUresult cuCtxDestroy(CUcontext ctx);

DESCRIPTION

Destroys the given CUDA context. If the context usage count is not equal to 1, or the context is current toany CPU thread other than the current one, this function fails. Floating contexts (detached from a CPUthread via cuCtxPopCurrent()) may be destroyed by this function.

RETURN VALUE


CUDA_SUCCESS






SEE ALSO

cuCtxCreate, cuCtxAttach, cuCtxDetach, cuCtxPushCurrent, cuCtxPopCurrent

115

2.3.4 cuCtxDetach

NAME

cuCtxDetach - decrement a context’s usage-count

SYNOPSIS

CUresult cuCtxDetach(CUcontext ctx);

DESCRIPTION

Decrements the usage count of the context, and destroys the context if the usage count goes to 0. The contextmust be a handle that was passed back by cuCtxCreate() or cuCtxAttach(), and must be current to thecalling thread.

RETURN VALUE


CUDA_SUCCESS





SEE ALSO

cuCtxCreate, cuCtxAttach, cuCtxDestroy, cuCtxPushCurrent, cuCtxPopCurrent

116

2.3.5 cuCtxGetDevice

NAME

cuCtxGetDevice - return device-ID for current context

SYNOPSIS

CUresult cuCtxGetDevice(CUdevice* device);

DESCRIPTION

Returns in *device the ordinal of the current context’s device.

RETURN VALUE


CUDA_SUCCESS






SEE ALSO

cuCtxCreate, cuCtxAttach, cuCtxDetach, cuCtxSynchronize

117

2.3.6 cuCtxPopCurrent

NAME

cuCtxPopCurrent - pops the current CUDA context from the current CPU thread

SYNOPSIS

CUresult cuCtxPopCurrent(CUcontext *pctx);

DESCRIPTION

Pops the current CUDA context from the CPU thread. The CUDA context must have a usage countof 1. CUDA contexts have a usage count of 1 upon creation; the usage count may be incremented withcuCtxAttach() and decremented with cuCtxDetach().

If successful, cuCtxPopCurrent() passes back the context handle in *pctx. The context may then bemade current to a different CPU thread by calling cuCtxPushCurrent().

Floating contexts may be destroyed by calling cuCtxDestroy().

If a context was current to the CPU thread before cuCtxCreate or cuCtxPushCurrent was called, thisfunction makes that context current to the CPU thread again.

RETURN VALUE


CUDA_SUCCESS





SEE ALSO

cuCtxCreate, cuCtxAttach, cuCtxDetach, cuCtxDestroy, cuCtxPushCurrent

118

2.3.7 cuCtxPushCurrent

NAME

cuCtxPushCurrent - attach floating context to CPU thread

SYNOPSIS

CUresult cuCtxPushCurrent(CUcontext ctx);

DESCRIPTION

Pushes the given context onto the CPU thread’s stack of current contexts. The specified context becomesthe CPU thread’s current context, so all CUDA functions that operate on the current context are affected.

The previous current context may be made current again by calling cuCtxDestroy() or cuCtxPopCur-rent().

The context must be "floating," i.e. not attached to any thread. Contexts are made to float by callingcuCtxPopCurrent().

RETURN VALUE


CUDA_SUCCESS






SEE ALSO

cuCtxCreate, cuCtxAttach, cuCtxDetach, cuCtxDestroy, cuCtxPopCurrent

119

2.3.8 cuCtxSynchronize

NAME

cuCtxSynchronize - block for a context’s tasks to complete

SYNOPSIS

CUresult cuCtxSynchronize(void);

DESCRIPTION

Blocks until the device has completed all preceding requested tasks. cuCtxSynchronize() returns an errorif one of the preceding tasks failed.

RETURN VALUE


CUDA_SUCCESS





SEE ALSO

cuCtxCreate, cuCtxAttach, cuCtxDetach, cuCtxGetDevice

120

2.4 ModuleManagement

NAME

Module Management

DESCRIPTION


cuModuleGetFunction

cuModuleGetGlobal

cuModuleGetTexRef

cuModuleLoad

cuModuleLoadData

cuModuleLoadFatBinary

cuModuleUnload

SEE ALSO




121

2.4.1 cuModuleGetFunction

NAME

cuModuleGetFunction - returns a function handle

SYNOPSIS

CUresult cuModuleGetFunction(CUfunction* func, CUmodule mod, const char* funcname);

DESCRIPTION

Returns in *func the handle of the function of name funcname located in module mod. If no function ofthat name exists, cuModuleGetFunction() returns CUDA_ERROR_NOT_FOUND.

RETURN VALUE


CUDA_SUCCESS





CUDA_ERROR_NOT_FOUND


SEE ALSO

cuModuleLoad, cuModuleLoadData, cuModuleLoadFatBinary, cuModuleUnload, cuModuleGetGlobal, cuMod-

uleGetTexRef

122

2.4.2 cuModuleGetGlobal

NAME

cuModuleGetGlobal - returns a global pointer from a module

SYNOPSIS

CUresult cuModuleGetGlobal(CUdeviceptr* devPtr, unsigned int* bytes, CUmodule mod, const char*

globalname);

DESCRIPTION

Returns in *devPtr and *bytes the base pointer and size of the global of name globalname located in mod-ule mod. If no variable of that name exists, cuModuleGetGlobal() returns CUDA_ERROR_NOT_FOUND.Both parameters devPtr and bytes are optional. If one of them is null, it is ignored.

RETURN VALUE


CUDA_SUCCESS







SEE ALSO

cuModuleLoad, cuModuleLoadData, cuModuleLoadFatBinary, cuModuleUnload, cuModuleGetFunction, cu-

ModuleGetTexRef

123

2.4.3 cuModuleGetTexRef

NAME

cuModuleGetTexRef - gets a handle to a texture-reference

SYNOPSIS

CUresult cuModuleGetTexRef(CUtexref* texRef, CUmodule hmod, const char* texrefname);

DESCRIPTION

Returns in *texref the handle of the texture reference of name texrefname in the module mod. If no tex-ture reference of that name exists, cuModuleGetTexRef() returns CUDA_ERROR_NOT_FOUND.This texture reference handle should not be destroyed, since it will be destroyed when the module is unloaded.

RETURN VALUE


CUDA_SUCCESS







SEE ALSO

cuModuleLoad, cuModuleLoadData, cuModuleLoadFatBinary, cuModuleUnload, cuModuleGetFunction, cu-

ModuleGetGlobal

124

2.4.4 cuModuleLoad

NAME

cuModuleLoad - loads a compute module

SYNOPSIS

CUresult cuModuleLoad(CUmodule* mod, const char* filename);

DESCRIPTION

Takes a file name filename and loads the corresponding module mod into the current context. The CUDAdriver API does not attempt to lazily allocate the resources needed by a module; if the memory for functionsand data (constant and global) needed by the module cannot be allocated, cuModuleLoad() fails. The fileshould be a cubin file as output by nvcc.

RETURN VALUE


CUDA_SUCCESS







CUDA_ERROR_FILE_NOT_FOUND


SEE ALSO

cuModuleLoadData, cuModuleLoadFatBinary, cuModuleUnload, cuModuleGetFunction, cuModuleGetGlobal,cuModuleGetTexRef

125

2.4.5 cuModuleLoadData

NAME

cuModuleLoadData - loads a module’s data

SYNOPSIS

CUresult cuModuleLoadData(CUmodule* mod, const void* image);

DESCRIPTION

Takes a pointer image and loads the corresponding module mod into the current context. The pointer maybe obtained by mapping a cubin file, passing a cubin file as a text string, or incorporating a cubin objectinto the executable resources and using operation system calls such as WindowsâĂŹ FindResource() toobtain the pointer.

RETURN VALUE


CUDA_SUCCESS







SEE ALSO

cuModuleLoad, cuModuleLoadFatBinary, cuModuleUnload, cuModuleGetFunction, cuModuleGetGlobal, cu-

ModuleGetTexRef

126

2.4.6 cuModuleLoadFatBinary

NAME

cuModuleLoadFatBinary - loads a fat-binary object

SYNOPSIS

CUresult cuModuleLoadFatBinary(CUmodule* mod, const void* fatBin);

DESCRIPTION

Takes a pointer fatBin and loads the corresponding module mod into the current context. The pointerrepresents a fat binary object, which is a collection of different cubin files, all representing the same devicecode but compiled and optimized for different architectures. There is currently no documented API forconstructing and using fat binary objects by programmers, and therefore this function is an internal functionin this version of CUDA. More information can be found in the nvcc document.

RETURN VALUE


CUDA_SUCCESS







CUDA_ERROR_NO_BINARY_FOR_GPU


SEE ALSO

cuModuleLoad, cuModuleLoadData, cuModuleUnload, cuModuleGetFunction, cuModuleGetGlobal, cuMod-

uleGetTexRef

127

2.4.7 cuModuleUnload

NAME

cuModuleUnload - unloads a module

SYNOPSIS

CUresult cuModuleUnload(CUmodule mod);

DESCRIPTION

Unloads a module mod from the current context.

RETURN VALUE


CUDA_SUCCESS






SEE ALSO

cuModuleLoad, cuModuleLoadData, cuModuleLoadFatBinary, cuModuleGetFunction, cuModuleGetGlobal, cu-

ModuleGetTexRef

128

2.5 StreamManagement

NAME

Stream Management

DESCRIPTION


cuStreamCreate

cuStreamDestroy

cuStreamQuery

cuStreamSynchronize

SEE ALSO




129

2.5.1 cuStreamCreate

NAME

cuStreamCreate - create a stream

SYNOPSIS

CUresult cuStreamCreate(CUstream* stream, unsigned int flags);

DESCRIPTION

Creates a stream. At present, flags is required to be 0.

RETURN VALUE


CUDA_SUCCESS







SEE ALSO

cuStreamQuery, cuStreamSynchronize, cuStreamDestroy

130

2.5.2 cuStreamDestroy

NAME

cuStreamDestroy - destroys a stream

SYNOPSIS

CUresult cuStreamDestroy(CUstream stream);

DESCRIPTION

Destroys the stream.

RETURN VALUE


CUDA_SUCCESS




CUDA_ERROR_INVALID_HANDLE


SEE ALSO

cuStreamCreate, cuStreamQuery, cuStreamSynchronize

131

2.5.3 cuStreamQuery

NAME

cuStreamQuery - determine status of a compute stream

SYNOPSIS

CUresult cuStreamQuery(CUstream stream);

DESCRIPTION

Returns CUDA_SUCCESS if all operations in the stream have completed, or CUDA_ERROR_NOT_READYif not.

RETURN VALUE


CUDA_SUCCESS





CUDA_ERROR_NOT_READY


SEE ALSO

cuStreamCreate, cuStreamSynchronize, cuStreamDestroy

132

2.5.4 cuStreamSynchronize

NAME

cuStreamSynchronize - block until a stream’s tasks are completed

SYNOPSIS

CUresult cuStreamSynchronize(CUstream stream);

DESCRIPTION

Blocks until the device has completed all operations in the stream.

RETURN VALUE


CUDA_SUCCESS






SEE ALSO

cuStreamCreate, cuStreamQuery, cuStreamDestroy

133

2.6 EventManagement

NAME

Event Management

DESCRIPTION


cuEventCreate

cuEventDestroy

cuEventElapsedTime

cuEventQuery

cuEventRecord

cuEventSynchronize

SEE ALSO




134

2.6.1 cuEventCreate

NAME

cuEventCreate - creates an event

SYNOPSIS

CUresult cuEventCreate(CUevent* event, unsigned int flags);

DESCRIPTION

Creates an event. At present, flags is required to be 0.

RETURN VALUE


CUDA_SUCCESS







SEE ALSO

cuEventRecord, cuEventQuery, cuEventSynchronize, cuEventDestroy, cuEventElapsedTime

135

2.6.2 cuEventDestroy

NAME

cuEventDestroy - destroys an event

SYNOPSIS

CUresult cuEventDestroy(CUevent event);

DESCRIPTION

Destroys the event.

RETURN VALUE


CUDA_SUCCESS






SEE ALSO

cuEventCreate, cuEventRecord, cuEventQuery, cuEventSynchronize, cuEventElapsedTime

136

2.6.3 cuEventElapsedTime

NAME

cuEventElapsedTime - computes the elapsed time between two events

SYNOPSIS

CUresult cuEventDestroy(float* time, CUevent start, CUevent end);

DESCRIPTION

Computes the elapsed time between two events (in milliseconds with a resolution of around 0.5 microseconds).If either event has not been recorded yet, this function returns CUDA_ERROR_INVALID_VALUE.If either event has been recorded with a non-zero stream, the result is undefined.

RETURN VALUE


CUDA_SUCCESS







SEE ALSO

cuEventCreate, cuEventRecord, cuEventQuery, cuEventSynchronize, cuEventDestroy

137

2.6.4 cuEventQuery

NAME

cuEventQuery - queries an event’s status

SYNOPSIS

CUresult cuEventQuery(CUevent event);

DESCRIPTION

Returns CUDA_SUCCESS if the event has actually been recorded, or CUDA_ERROR_NOT_READYif not. If cuEventRecord() has not been called on this event, the function returns CUDA_ERROR_INVALID_VALUE

RETURN VALUE


CUDA_SUCCESS





CUDA_ERROR_NOT_READY


SEE ALSO

cuEventCreate, cuEventRecord, cuEventSynchronize, cuEventDestroy, cuEventElapsedTime

138

2.6.5 cuEventRecord

NAME

cuEventRecord - records an event

SYNOPSIS

CUresult cuEventRecord(CUevent event, CUstream stream);

DESCRIPTION

Records an event. If stream is non-zero, the event is recorded after all preceding operations in the streamhave been completed; otherwise, it is recorded after all preceding operations in the CUDA context have beencompleted. Since this operation is asynchronous, cuEventQuery() and/or cuEventSynchronize() mustbe used to determine when the event has actually been recorded.

If cuEventRecord() has previously been called and the event has not been recorded yet, this functionreturns CUDA_ERROR_INVALID_VALUE.

RETURN VALUE


CUDA_SUCCESS







SEE ALSO

cuEventCreate, cuEventQuery, cuEventSynchronize, cuEventDestroy, cuEventElapsedTime

139

2.6.6 cuEventSynchronize

NAME

cuEventSynchronize - waits for an event to complete

SYNOPSIS

CUresult cuEventSynchronize(CUevent event);

DESCRIPTION

Blocks until the event has actually been recorded. If cuEventRecord() has not been called on this event,the function returns CUDA_ERROR_INVALID_VALUE.

RETURN VALUE


CUDA_SUCCESS






SEE ALSO

cuEventCreate, cuEventRecord, cuEventQuery, cuEventDestroy, cuEventElapsedTime

140

2.7 ExecutionControl

NAME

Execution Control

DESCRIPTION


cuLaunch

cuLaunchGrid

cuParamSetSize

cuParamSetTexRef

cuParamSetf

cuParamSeti

cuParamSetv

cuFuncSetBlockShape

cuFuncSetSharedSize

SEE ALSO




141

2.7.1 cuLaunch

NAME

cuLaunch - launches a CUDA function

SYNOPSIS

CUresult cuLaunch(CUfunction func);

DESCRIPTION

Invokes the kernel func on a 1ÃŮ1 grid of blocks. The block contains the number of threads specified by aprevious call to cuFuncSetBlockShape().

RETURN VALUE


CUDA_SUCCESS





CUDA_ERROR_LAUNCH_INCOMPATIBLE_TEXTURING


SEE ALSO

cuFuncSetBlockShape, cuFuncSetSharedSize, cuParamSetSize, cuParamSeti, cuParamSetf, cuParamSetv, cu-

ParamSetTexRef, cuLaunchGrid

142

2.7.2 cuLaunchGrid

NAME

cuLaunchGrid - launches a CUDA function

SYNOPSIS

CUresult cuLaunchGrid(CUfunction func, int grid_width, int grid_height);

CUresult cuLaunchGridAsync(CUfunction func, int grid_width, int grid_height, CUstream stream);

DESCRIPTION

Invokes the kernel on a grid_width x grid_height grid of blocks. Each block contains the number ofthreads specified by a previous call to cuFuncSetBlockShape().

cuLaunchGridAsync() can optionally be associated to a stream by passing a non-zero stream argument.

RETURN VALUE


CUDA_SUCCESS





CUDA_ERROR_LAUNCH_INCOMPATIBLE_TEXTURING


SEE ALSO


ParamSetTexRef, cuLaunch

143

2.7.3 cuParamSetSize

NAME

cuParamSetSize - sets the parameter-size for the function

SYNOPSIS

CUresult cuParamSetSize(CUfunction func, unsigned int numbytes);

DESCRIPTION

Sets through numbytes the total size in bytes needed by the function parameters of function func.

RETURN VALUE


CUDA_SUCCESS






SEE ALSO

cuFuncSetBlockShape, cuFuncSetSharedSize, cuParamSeti, cuParamSetf, cuParamSetv, cuParamSetTexRef,cuLaunch, cuLaunchGrid

144

2.7.4 cuParamSetTexRef

NAME

cuParamSetTexRef - adds a texture-reference to the function’s argument list

SYNOPSIS

CUresult cuParamSetTexRef(CUfunction func, int texunit, CUtexref texRef);

DESCRIPTION

Makes the CUDA array or linear memory bound to the texture reference texRef available to a deviceprogram as a texture. In this version of CUDA, the texture reference must be obtained via cuModuleGet-TexRef() and the texunit parameter must be set to CU_PARAM_TR_DEFAULT.

RETURN VALUE


CUDA_SUCCESS






SEE ALSO


Launch, cuLaunchGrid

145

2.7.5 cuParamSetf

NAME

cuParamSetf - adds a floating-point parameter to the function’s argument list

SYNOPSIS

CUresult cuParamSetf(CUfunction func, int offset, float value);

DESCRIPTION

Sets a floating point parameter that will be specified the next time the kernel corresponding to func will beinvoked. offset is a byte offset.

RETURN VALUE


CUDA_SUCCESS






SEE ALSO

cuFuncSetBlockShape, cuFuncSetSharedSize, cuParamSetSize, cuParamSeti, cuParamSetv, cuParamSetTexRef,cuLaunch, cuLaunchGrid

146

2.7.6 cuParamSeti

NAME

cuParamSeti - adds an integer parameter to the function’s argument list

SYNOPSIS

CUresult cuParamSeti(CUfunction func, int offset, unsigned int value);

DESCRIPTION

Sets an integer parameter that will be specified the next time the kernel corresponding to func will beinvoked. offset is a byte offset.

RETURN VALUE


CUDA_SUCCESS






SEE ALSO

cuFuncSetBlockShape, cuFuncSetSharedSize, cuParamSetSize, cuParamSetf, cuParamSetv, cuParamSetTexRef,cuLaunch, cuLaunchGrid

147

2.7.7 cuParamSetv

NAME

cuParamSetv - adds arbitrary data to the function’s argument list

SYNOPSIS

CUresult cuParamSetv(CUfunction func, int offset, void* ptr, unsigned int numbytes);

DESCRIPTION

Copies an arbitrary amount of data into the parameter space of the kernel corresponding to func. offset isa byte offset.

RETURN VALUE


CUDA_SUCCESS






SEE ALSO

cuFuncSetBlockShape, cuFuncSetSharedSize, cuParamSetSize, cuParamSeti, cuParamSetf, cuParamSetTexRef,cuLaunch, cuLaunchGrid

148

2.7.8 cuFuncSetBlockShape

NAME

cuFuncSetBlockShape - sets the block-dimensions for the function

SYNOPSIS

CUresult cuFuncSetBlockShape(CUfunction func, int x, int y, int z);

DESCRIPTION

Specifies the X, Y and Z dimensions of the thread blocks that are created when the kernel given by func islaunched.

RETURN VALUE


CUDA_SUCCESS







SEE ALSO

cuFuncSetSharedSize, cuParamSetSize, cuParamSeti, cuParamSetf, cuParamSetv, cuParamSetTexRef, cu-


149

2.7.9 cuFuncSetSharedSize

NAME

cuFuncSetSharedSize - sets the shared-memory size for the function

SYNOPSIS

CUresult cuFuncSetSharedSize(CUfunction func, unsigned int bytes);

DESCRIPTION

Sets through bytes the amount of shared memory that will be available to each thread block when the kernelgiven by func is launched.

RETURN VALUE


CUDA_SUCCESS







SEE ALSO

cuFuncSetBlockShape, cuParamSetSize, cuParamSeti, cuParamSetf, cuParamSetv, cuParamSetTexRef, cu-


150

2.8 MemoryManagement

NAME

Memory Management

DESCRIPTION


cuArrayCreate

cuArray3DCreate

cuArrayDestroy

cuArrayGetDescriptor

cuArray3DGetDescriptor

cuMemAlloc

cuMemAllocHost

cuMemAllocPitch

cuMemFree

cuMemFreeHost

cuMemGetAddressRange

cuMemGetInfo

cuMemcpy2D

cuMemcpy3D

cuMemcpyAtoA

cuMemcpyAtoD

cuMemcpyAtoH

cuMemcpyDtoA

cuMemcpyDtoD

cuMemcpyDtoH

cuMemcpyHtoA

cuMemcpyHtoD

cuMemset

cuMemset2D

SEE ALSO




151

2.8.1 cuArrayCreate

NAME

cuArrayCreate - creates a 1D or 2D CUDA array

SYNOPSIS

CUresult cuArrayCreate(CUarray* array, const CUDA_ARRAY_DESCRIPTOR* desc);

DESCRIPTION

Creates a CUDA array according to the CUDA_ARRAY_DESCRIPTOR structure desc and returns ahandle to the new CUDA array in *array. The CUDA_ARRAY_DESCRIPTOR structure is definedas such:

typedef struct {

unsigned int Width;

unsigned int Height;

CUarray_format Format;

unsigned int NumChannels;

} CUDA_ARRAY_DESCRIPTOR;

where:

• Width and Height are the width and height of the CUDA array (in elements); the CUDA array isone-dimensional if height is 0, two-dimensional, otherwise;

• NumChannels specifies the number of packed components per CUDA array element.; it may be 1, 2or 4;

• Format specifies the format of the elements; CUarray_format is defined as such:

typedef enum CUarray_format_enum {

CU_AD_FORMAT_UNSIGNED_INT8 = 0x01,



CU_AD_FORMAT_SIGNED_INT8 = 0x08,


CU_AD_FORMAT_SIGNED_INT32 = 0x0a,

CU_AD_FORMAT_HALF = 0x10,

CU_AD_FORMAT_FLOAT = 0x20

} CUarray_format;

Here are examples of CUDA array descriptions:

• Description for a CUDA array of 2048 floats:

CUDA_ARRAY_DESCRIPTOR desc;

desc.Format = CU_AD_FORMAT_FLOAT;

152

desc.NumChannels = 1;

desc.Width = 2048;

desc.Height = 1;

• Description for a 64 x 64 CUDA array of floats:




desc.Width = 64;

desc.Height = 64;

• Description for a width x height CUDA array of 64-bit, 4x16-bit float16’s:


desc.FormatFlags = CU_AD_FORMAT_HALF;


desc.Width = width;

desc.Height = height;

• Description for a width x height CUDA array of 16-bit elements, each of which is two 8-bit unsignedchars:

CUDA_ARRAY_DESCRIPTOR arrayDesc;

desc.FormatFlags = CU_AD_FORMAT_UNSIGNED_INTS;


desc.Width = width;


RETURN VALUE


CUDA_SUCCESS






CUDA_ERROR_UNKNOWN


SEE ALSO

cuMemGetInfo, cuMemAlloc, cuMemAllocPitch, cuMemFree, cuMemAllocHost, cuMemFreeHost, cuMemGe-

tAddressRange, cuArrayGetDescriptor, cuArrayDestroy, cuMemset, cuMemset2D

153

2.8.2 cuArray3DCreate

NAME

cuArray3DCreate - creates a CUDA array

SYNOPSIS

CUresult cuArray3DCreate(CUarray* array, const CUDA_ARRAY3D_DESCRIPTOR* desc);

DESCRIPTION

Creates a CUDA array according to the CUDA_ARRAY3D_DESCRIPTOR structure desc and re-turns a handle to the new CUDA array in *array. The CUDA_ARRAY3D_DESCRIPTOR structureis defined as such:

typedef struct {

unsigned int Width;


unsigned int Depth;

CUarray_format Format;

unsigned int NumChannels;

unsigned int Flags;

} CUDA_ARRAY3D_DESCRIPTOR;

where:

• Width, Height and Depth are the width, height and depth of the CUDA array (in elements); theCUDA array is one-dimensional if height and depth are 0, two-dimensional if depth is 0, and three-dimensional otherwise;

• NumChannels specifies the number of packed components per CUDA array element.; it may be 1, 2or 4;

• Format specifies the format of the elements; CUarray_format is defined as such:

typedef enum CUarray_format_enum {






CU_AD_FORMAT_SIGNED_INT32 = 0x0a,

CU_AD_FORMAT_HALF = 0x10,

CU_AD_FORMAT_FLOAT = 0x20

} CUarray_format;

• Flags provides for future features. For now, it must be set to 0.

Here are examples of CUDA array descriptions:

154

• Description for a CUDA array of 2048 floats:

CUDA_ARRAY3D_DESCRIPTOR desc;



desc.Width = 2048;

desc.Height = 0;

desc.Depth = 0;

• Description for a 64 x 64 CUDA array of floats:

CUDA_ARRAY3D_DESCRIPTOR desc;



desc.Width = 64;

desc.Height = 64;

desc.Depth = 0;

• Description for a width x height x depth CUDA array of 64-bit, 4x16-bit float16’s:


desc.FormatFlags = CU_AD_FORMAT_HALF;


desc.Width = width;


desc.Depth = depth;

RETURN VALUE


CUDA_SUCCESS






CUDA_ERROR_UNKNOWN


SEE ALSO


tAddressRange, cuArray3DGetDescriptor, cuArrayDestroy

155

2.8.3 cuArrayDestroy

NAME

cuArrayDestroy - destroys a CUDA array

SYNOPSIS

CUresult cuArrayDestroy(CUarray array);

DESCRIPTION

Destroys the CUDA array array.

RETURN VALUE


CUDA_SUCCESS





CUDA_ERROR_ARRAY_IS_MAPPED


SEE ALSO


tAddressRange, cuArrayCreate, cuArrayGetDescriptor, cuMemset, cuMemset2D

156

2.8.4 cuArrayGetDescriptor

NAME

cuArrayGetDescriptor - get a 1D or 2D CUDA array descriptor

SYNOPSIS

CUresult cuArrayGetDescriptor(CUDA_ARRAY_DESCRIPTOR* arrayDesc, CUarray array)

DESCRIPTION

Returns in *arrayDesc a descriptor of the format and dimensions of the 1D or 2D CUDA array array. It isuseful for subroutines that have been passed a CUDA array, but need to know the CUDA array parametersfor validation or other purposes.

RETURN VALUE


CUDA_SUCCESS







SEE ALSO

cuArrayCreate, cuArray3DCreate, cuArray3DGetDescriptor, cuArrayDestroy

157

2.8.5 cuArray3DGetDescriptor

NAME

cuArray3DGetDescriptor - get a 3D CUDA array descriptor

SYNOPSIS

CUresult cuArray3DGetDescriptor(CUDA_ARRAY3D_DESC *arrayDesc, CUarray array);

DESCRIPTION

Returns in *arrayDesc a descriptor containing information on the format and dimensions of the CUDAarray array. It is useful for subroutines that have been passed a CUDA array, but need to know the CUDAarray parameters for validation or other purposes.

RETURN VALUE


CUDA_SUCCESS







SEE ALSO

cuArrayCreate, cuArray3DCreate, cuArrayGetDescriptor, cuArrayDestroy

158

2.8.6 cuMemAlloc

NAME

cuMemAlloc - allocates device memory

SYNOPSIS

CUresult cuMemAlloc(CUdeviceptr* devPtr, unsigned int count);

DESCRIPTION

Allocates count bytes of linear memory on the device and returns in *devPtr a pointer to the allocatedmemory. The allocated memory is suitably aligned for any kind of variable. The memory is not cleared. Ifcount is 0, cuMemAlloc() returns CUDA_ERROR_INVALID_VALUE.

RETURN VALUE


CUDA_SUCCESS







SEE ALSO

cuMemGetInfo, cuMemAllocPitch, cuMemFree, cuMemAllocHost, cuMemFreeHost, cuMemGetAddressRange,cuArrayCreate, cuArrayGetDescriptor, cuArrayDestroy, cuMemset, cuMemset2D

159

2.8.7 cuMemAllocHost

NAME

cuMemAllocHost - allocates page-locked host memory

SYNOPSIS

CUresult cuMemAllocHost(void** hostPtr, unsigned int count);

DESCRIPTION

Allocates count bytes of host memory that is page-locked and accessible to the device. The driver tracksthe virtual memory ranges allocated with this function and automatically accelerates calls to functions suchas cuMemcpy(). Since the memory can be accessed directly by the device, it can be read or writtenwith much higher bandwidth than pageable memory obtained with functions such as malloc(). Allocatingexcessive amounts of memory with cuMemAllocHost() may degrade system performance, since it reducesthe amount of memory available to the system for paging. As a result, this function is best used sparinglyto allocate staging areas for data exchange between host and device.

RETURN VALUE


CUDA_SUCCESS







SEE ALSO

cuMemGetInfo, cuMemAlloc, cuMemAllocPitch, cuMemFree, cuMemFreeHost, cuMemGetAddressRange, cuAr-

rayCreate, cuArrayGetDescriptor, cuArrayDestroy, cuMemset, cuMemset2D

160

2.8.8 cuMemAllocPitch

NAME

cuMemAllocPitch - allocates device memory

SYNOPSIS

CUresult cuMemAllocPitch(CUdeviceptr* devPtr, unsigned int* pitch, unsigned int widthInBytes,

unsigned int height, unsigned int elementSizeBytes);

DESCRIPTION

Allocates at least widthInBytes*height bytes of linear memory on the device and returns in *devPtra pointer to the allocated memory. The function may pad the allocation to ensure that correspondingpointers in any given row will continue to meet the alignment requirements for coalescing as the address isupdated from row to row. elementSizeBytes specifies the size of the largest reads and writes that will beperformed on the memory range. elementSizeBytes may be 4, 8 or 16 (since coalesced memory transactionsare not possible on other data sizes). If elementSizeBytes is smaller than the actual read/write size ofa kernel, the kernel will run correctly, but possibly at reduced speed. The pitch returned in *pitch bycuMemAllocPitch() is the width in bytes of the allocation. The intended usage of pitch is as a separateparameter of the allocation, used to compute addresses within the 2D array. Given the row and column ofan array element of type T, the address is computed as

T* pElement = (T*)((char*)BaseAddress + Row * Pitch) + Column;

The pitch returned by cuMemAllocPitch() is guaranteed to work with cuMemcpy2D() under all cir-cumstances. For allocations of 2D arrays, it is recommended that programmers consider performing pitchallocations using cuMemAllocPitch(). Due to alignment restrictions in the hardware, this is especiallytrue if the application will be performing 2D memory copies between different regions of device memory(whether linear memory or CUDA arrays).

RETURN VALUE


CUDA_SUCCESS







161

SEE ALSO

cuMemGetInfo, cuMemAlloc, cuMemFree, cuMemAllocHost, cuMemFreeHost, cuMemGetAddressRange, cuAr-

rayCreate, cuArrayGetDescriptor, cuArrayDestroy, cuMemset, cuMemset2D

162

2.8.9 cuMemFree

NAME

cuMemFree - frees device memory

SYNOPSIS

CUresult cuMemFree(CUdeviceptr devPtr);

DESCRIPTION

Frees the memory space pointed to by devPtr, which must have been returned by a previous call tocuMemMalloc() or cuMemMallocPitch().

RETURN VALUE


CUDA_SUCCESS






SEE ALSO

cuMemGetInfo, cuMemAlloc, cuMemAllocPitch, cuMemAllocHost, cuMemFreeHost, cuMemGetAddressRange,cuArrayCreate, cuArrayGetDescriptor, cuArrayDestroy, cuMemset, cuMemset2D

163

2.8.10 cuMemFreeHost

NAME

cuMemFreeHost - frees page-locked host memory

SYNOPSIS

CUresult cuMemFreeHost(void* hostPtr);

DESCRIPTION

Frees the memory space pointed to by hostPtr, which must have been returned by a previous call tocuMemAllocHost().

RETURN VALUE


CUDA_SUCCESS






SEE ALSO

cuMemGetInfo, cuMemAlloc, cuMemAllocPitch, cuMemFree, cuMemAllocHost, cuMemGetAddressRange,cuArrayCreate, cuArrayGetDescriptor, cuArrayDestroy, cuMemset, cuMemset2D

164

2.8.11 cuMemGetAddressRange

NAME

cuMemGetAddressRange - get information on memory allocations

SYNOPSIS

CUresult cuMemGetAddressRange(CUdeviceptr* basePtr, unsigned int* size, CUdeviceptr devPtr);

DESCRIPTION

Returns the base address in *basePtr and size and *size of the allocation by cuMemAlloc() or cuMemAl-locPitch() that contains the input pointer devPtr. Both parameters basePtr and size are optional. Ifone of them is null, it is ignored.

RETURN VALUE


CUDA_SUCCESS






SEE ALSO

cuMemGetInfo, cuMemAlloc, cuMemAllocPitch, cuMemFree, cuMemAllocHost, cuMemFreeHost, cuArray-

Create, cuArrayGetDescriptor, cuArrayDestroy, cuMemset, cuMemset2D

165

2.8.12 cuMemGetInfo

NAME

cuMemGetInfo - gets free and total memory

SYNOPSIS

CUresult cuMemGetInfo(unsigned int* free, unsigned int* total);

DESCRIPTION

Returns in *free and *total respectively, the free and total amount of memory available for allocation bythe CUDA context, in bytes.

RETURN VALUE


CUDA_SUCCESS






SEE ALSO

cuMemAlloc, cuMemAllocPitch, cuMemFree, cuMemAllocHost, cuMemFreeHost, cuMemGetAddressRange,cuArrayCreate, cuArrayGetDescriptor, cuArrayDestroy, cuMemset, cuMemset2D

166

2.8.13 cuMemcpy2D

NAME

cuMemcpy2D - copies memory for 2D arrays

SYNOPSIS

CUresult cuMemcpy2D(const CUDA_MEMCPY2D* copyParam);

CUresult cuMemcpy2DUnaligned(const CUDA_MEMCPY2D* copyParam);

CUresult cuMemcpy2DAsync(const CUDA_MEMCPY2D* copyParam, CUstream stream);

DESCRIPTION

Perform a 2D memory copy according to the parameters specified in copyParam. The CUDA_MEMCPY2Dstructure is defined as such:

typedef struct CUDA_MEMCPY2D_st {

unsigned int srcXInBytes, srcY;

CUmemorytype srcMemoryType;

const void *srcHost;

CUdeviceptr srcDevice;

CUarray srcArray;

unsigned int srcPitch;

unsigned int dstXInBytes, dstY;

CUmemorytype dstMemoryType;

void *dstHost;

CUdeviceptr dstDevice;

CUarray dstArray;

unsigned int dstPitch;

unsigned int WidthInBytes;


} CUDA_MEMCPY2D;

where:

• srcMemoryType and dstMemoryType specify the type of memory of the source and destination,respectively; CUmemorytype_enum is defined as such:

typedef enum CUmemorytype_enum {

CU_MEMORYTYPE_HOST = 0x01,

CU_MEMORYTYPE_DEVICE = 0x02,

CU_MEMORYTYPE_ARRAY = 0x03

} CUmemorytype;

If srcMemoryType is CU_MEMORYTYPE_HOST, srcHost and srcPitch specify the (host)base address of the source data and the bytes per row to apply. srcArray is ignored.

If srcMemoryType is CU_MEMORYTYPE_DEVICE, srcDevice and srcPitch specify the(device) base address of the source data and the bytes per row to apply. srcArray is ignored.

167

If srcMemoryType is CU_MEMORYTYPE_ARRAY, srcArray specifies the handle of thesource data. srcHost, srcDevice and srcPitch are ignored.

If dstMemoryType is CU_MEMORYTYPE_HOST, dstHost and dstPitch specify the (host)base address of the destination data and the bytes per row to apply. dstArray is ignored.

If dstMemoryType is CU_MEMORYTYPE_DEVICE, dstDevice and dstPitch specify the(device) base address of the destination data and the bytes per row to apply. dstArray is ignored.

If dstMemoryType is CU_MEMORYTYPE_ARRAY, dstArray specifies the handle of thedestination data. dstHost, dstDevice and dstPitch are ignored.

• srcXInBytes and srcY specify the base address of the source data for the copy.

For host pointers, the starting address is

void* Start = (void*)((char*)srcHost+srcY*srcPitch + srcXInBytes);

For device pointers, the starting address is

CUdeviceptr Start = srcDevice+srcY*srcPitch+srcXInBytes;

For CUDA arrays, srcXInBytes must be evenly divisible by the array element size.

• dstXInBytes and dstY specify the base address of the destination data for the copy.

For host pointers, the base address is

void* dstStart = (void*)((char*)dstHost+dstY*dstPitch + dstXInBytes);


CUdeviceptr dstStart = dstDevice+dstY*dstPitch+dstXInBytes;

For CUDA arrays, dstXInBytes must be evenly divisible by the array element size.

• WidthInBytes and Height specify the width (in bytes) and height of the 2D copy being performed.Any pitches must be greater than or equal to WidthInBytes.

cuMemcpy2D() returns an error if any pitch is greater than the maximum allowed (CU_DEVICE_ATTRIBUTE_MAX_PITCHcuMemAllocPitch() passes back pitches that always work with cuMemcpy2D(). On intra-devicememory copies (device ? device, CUDA array ? device, CUDA array ? CUDA array), cuMem-cpy2D() may fail for pitches not computed by cuMemAllocPitch(). cuMemcpy2DUnaligned()does not have this restriction, but may run significantly slower in the cases where cuMemcpy2D()would have returned an error code.

cuMemcpy2DAsync() is asynchronous and can optionally be associated to a stream by passing anon-zero stream argument. It only works on page-locked host memory and returns an error if a pointerto pageable memory is passed as input.

RETURN VALUE


CUDA_SUCCESS


168





SEE ALSO

cuMemcpyHtoD, cuMemcpyDtoH, cuMemcpyDtoD, cuMemcpyDtoA, cuMemcpyAtoD, cuMemcpyAtoH, cuMem-

cpyHtoA, cuMemcpyAtoA, cuMemcpy3D

169

2.8.14 cuMemcpy3D

NAME

cuMemcpy3D - copies memory for 3D arrays

SYNOPSIS

CUresult cuMemcpy3D(const CUDA_MEMCPY3D* copyParam);

CUresult cuMemcpy3DAsync(const CUDA_MEMCPY3D* copyParam, CUstream stream);

DESCRIPTION

Perform a 3D memory copy according to the parameters specified in copyParam. The CUDA_MEMCPY3Dstructure is defined as such:

typedef struct CUDA_MEMCPY3D_st {

unsigned int srcXInBytes, srcY, srcZ;

unsigned int srcLOD;

CUmemorytype srcMemoryType;

const void *srcHost;

CUdeviceptr srcDevice;

CUarray srcArray;

unsigned int srcPitch; // ignored when src is array

unsigned int srcHeight; // ignored when src is array; may be 0 if Depth==1

unsigned int dstXInBytes, dstY, dstZ;

unsigned int dstLOD;

CUmemorytype dstMemoryType;

void *dstHost;

CUdeviceptr dstDevice;

CUarray dstArray;

unsigned int dstPitch; // ignored when dst is array

unsigned int dstHeight; // ignored when dst is array; may be 0 if Depth==1

unsigned int WidthInBytes;


unsigned int Depth;

} CUDA_MEMCPY3D;

CUresult CUDAAPI cuMemcpy3D( const CUDA_MEMCPY3D *pCopy );

where:

• srcMemoryType and dstMemoryType specify the type of memory of the source and destination,respectively; CUmemorytype_enum is defined as such:

typedef enum CUmemorytype_enum {

170

CU_MEMORYTYPE_HOST = 0x01,

CU_MEMORYTYPE_DEVICE = 0x02,

CU_MEMORYTYPE_ARRAY = 0x03

} CUmemorytype;

If srcMemoryType is CU_MEMORYTYPE_HOST, srcHost, srcPitch and srcHeight spec-ify the (host) base address of the source data, the bytes per row, and the height of each 2D slice of the3D array. srcArray is ignored.

If srcMemoryType is CU_MEMORYTYPE_DEVICE, srcDevice, srcPitch and srcHeightspecify the (device) base address of the source data, the bytes per row, and the height of each 2D sliceof the 3D array. srcArray is ignored.

If srcMemoryType is CU_MEMORYTYPE_ARRAY, srcArray specifies the handle of thesource data. srcHost, srcDevice, srcPitch and srcHeight are ignored.

If dstMemoryType is CU_MEMORYTYPE_HOST, dstHost and dstPitch specify the (host)base address of the destination data, the bytes per row, and the height of each 2D slice of the 3D array.dstArray is ignored.

If dstMemoryType is CU_MEMORYTYPE_DEVICE, dstDevice and dstPitch specify the(device) base address of the destination data, the bytes per row, and the height of each 2D slice of the3D array. dstArray is ignored.

If dstMemoryType is CU_MEMORYTYPE_ARRAY, dstArray specifies the handle of thedestination data. dstHost, dstDevice, dstPitch and dstHeight are ignored.

• srcXInBytes, srcY and srcZ specify the base address of the source data for the copy.

For host pointers, the starting address is

void* Start = (void*)((char*)srcHost+(srcZ*srcHeight+srcY)*srcPitch + srcXInBytes);


CUdeviceptr Start = srcDevice+(srcZ*srcHeight+srcY)*srcPitch+srcXInBytes;

For CUDA arrays, srcXInBytes must be evenly divisible by the array element size.

• dstXInBytes, dstY and dstZ specify the base address of the destination data for the copy.

For host pointers, the base address is

void* dstStart = (void*)((char*)dstHost+(dstZ*dstHeight+dstY)*dstPitch + dstXInBytes);


CUdeviceptr dstStart = dstDevice+(dstZ*dstHeight+dstY)*dstPitch+dstXInBytes;

For CUDA arrays, dstXInBytes must be evenly divisible by the array element size.

• WidthInBytes, Height and Depth specify the width (in bytes), height and depth of the 3D copybeing performed. Any pitches must be greater than or equal to WidthInBytes.

cuMemcpy3D() returns an error if any pitch is greater than the maximum allowed (CU_DEVICE_ATTRIBUTE_MAX_PITCH

cuMemcpy3DAsync() is asynchronous and can optionally be associated to a stream by passing anon-zero stream argument. It only works on page-locked host memory and returns an error if a pointerto pageable memory is passed as input.

The srcLOD and dstLOD members of the CUDA_MEMCPY3D structure must be set to 0.

171

RETURN VALUE

CUDA_SUCCESS

SEE ALSO


cpyHtoA, cuMemcpyAtoA, cuMemcpy2D, cuMemcpy2DAsync

172

2.8.15 cuMemcpyAtoA

NAME

cuMemcpyAtoA - copies memory from Array to Array

SYNOPSIS

CUresult cuMemcpyAtoA(CUarray dstArray, unsigned int dstIndex, CUarray srcArray, unsigned int

srcIndex, unsigned int count);

DESCRIPTION

Copies from one 1D CUDA array to another. dstArray and srcArray specify the handles of the destinationand source CUDA arrays for the copy, respectively. dstIndex and srcIndex specify the destination andsource indices into the CUDA array. These values are in the range [0, Width-1] for the CUDA array; theyare not byte offsets. count is the number of bytes to be copied. The size of the elements in the CUDA arraysneed not be the same format, but the elements must be the same size; and count must be evenly divisibleby that size.

RETURN VALUE


CUDA_SUCCESS






SEE ALSO


cpyHtoA, cuMemcpy2D

173

2.8.16 cuMemcpyAtoD

NAME

cuMemcpyAtoD - copies memory from Array to Device

SYNOPSIS

CUresult cuMemcpyAtoD(CUdeviceptr dstDevPtr, CUarray srcArray, unsigned int srcIndex, unsigned

int count);

DESCRIPTION

Copies from a 1D CUDA array to device memory. dstDevPtr specifies the base pointer of the destinationand must be naturally aligned with the CUDA array elements. srcArray and srcIndex specify the CUDAarray handle and the index (in array elements) of the array element where the copy is to begin. countspecifies the number of bytes to copy and must be evenly divisible by the array element size.

RETURN VALUE


CUDA_SUCCESS






SEE ALSO

cuMemcpyHtoD, cuMemcpyDtoH, cuMemcpyDtoD, cuMemcpyDtoA, cuMemcpyAtoH, cuMemcpyHtoA, cuMem-

cpyAtoA, cuMemcpy2D

174

2.8.17 cuMemcpyAtoH

NAME

cuMempcyAtoH - copies memory from Array to Host

SYNOPSIS

CUresult cuMemcpyAtoH(void* dstHostPtr, CUarray srcArray, unsigned int srcIndex, unsigned int

count);

CUresult cuMemcpyAtoHAsync(void* dstHostPtr, CUarray srcArray, unsigned int srcIndex, unsigned

int count, CUstream stream);

DESCRIPTION

Copies from a 1D CUDA array to host memory. dstHostPtr specifies the base pointer of the destination.srcArray and srcIndex specify the CUDA array handle and starting index of the source data. countspecifies the number of bytes to copy.

cuMemcpyAtoHAsync() is asynchronous and can optionally be associated to a stream by passing a non-zero stream argument. It only works on page-locked host memory and returns an error if a pointer topageable memory is passed as input.

RETURN VALUE


CUDA_SUCCESS






SEE ALSO

cuMemcpyHtoD, cuMemcpyDtoH, cuMemcpyDtoD, cuMemcpyDtoA, cuMemcpyAtoD, cuMemcpyHtoA, cuMem-

cpyAtoA, cuMemcpy2D

175

2.8.18 cuMemcpyDtoA

NAME

cuMemcpyDtoA - copies memory from Device to Array

SYNOPSIS

CUresult cuMemcpyDtoA(CUarray dstArray, unsigned int dstIndex, CUdeviceptr srcDevPtr, unsigned

int count);

DESCRIPTION

Copies from device memory to a 1D CUDA array. dstArray and dstIndex specify the CUDA array handleand starting index of the destination data. srcDevPtr specifies the base pointer of the source. countspecifies the number of bytes to copy.

RETURN VALUE


CUDA_SUCCESS






SEE ALSO

cuMemcpyHtoD, cuMemcpyDtoH, cuMemcpyDtoD, cuMemcpyAtoD, cuMemcpyAtoH, cuMemcpyHtoA, cuMem-

cpyAtoA, cuMemcpy2D

176

2.8.19 cuMemcpyDtoD

NAME

cuMemcpyDtoD - copies memory from Device to Device

SYNOPSIS

CUresult cuMemcpyDtoD(CUdeviceptr dstDevPtr, CUdeviceptr srcDevPtr, unsigned int count);

DESCRIPTION

Copies from device memory to device memory. dstDevice and srcDevPtr are the base pointers of thedestination and source, respectively. count specifies the number of bytes to copy.

RETURN VALUE


CUDA_SUCCESS






SEE ALSO

cuMemcpyHtoD, cuMemcpyDtoH, cuMemcpyDtoA, cuMemcpyAtoD, cuMemcpyAtoH, cuMemcpyHtoA, cuMem-

cpyAtoA, cuMemcpy2D

177

2.8.20 cuMemcpyDtoH

NAME

cuMemcpyDtoH - copies memory from Device to Host

SYNOPSIS

CUresult cuMemcpyDtoH(void* dstHostPtr, CUdeviceptr srcDevPtr, unsigned int count);

CUresult cuMemcpyDtoHAsync(void* dstHostPtr, CUdeviceptr srcDevPtr, unsigned int count, CUstream

stream);

DESCRIPTION

Copies from device to host memory. dstHostPtr and srcDevPtrspecify the base addresses of the sourceand destination, respectively. countspecifies the number of bytes to copy.

MemcpyDtoHAsync() is asynchronous and can optionally be associated to a stream by passing a non-zerostream argument. It only works on page-locked host memory and returns an error if a pointer to pageablememory is passed as input.

RETURN VALUE


CUDA_SUCCESS






SEE ALSO

cuMemcpyHtoD, cuMemcpyDtoD, cuMemcpyDtoA, cuMemcpyAtoD, cuMemcpyAtoH, cuMemcpyHtoA, cuMem-

cpyAtoA, cuMemcpy2D

178

2.8.21 cuMemcpyHtoA

NAME

cuMemcpyHtoA - copies memory from Host to Array

SYNOPSIS

CUresult cuMemcpyHtoA(CUarray dstArray, unsigned int dstIndex, const void *srcHostPtr, unsigned

int count);

CUresult cuMemcpyHtoAAsync(CUarray dstArray, unsigned int dstIndex, const void *srcHostPtr,

unsigned int count, CUstream stream);

DESCRIPTION

Copies from host memory to a 1D CUDA array. dstArray and dstIndex specify the CUDA array handleand starting index of the destination data. srcHostPtr specify the base addresse of the source. countspecifies the number of bytes to copy.

cuMemcpyHtoAAsync() is asynchronous and can optionally be associated to a stream by passing a non-zero stream argument. It only works on page-locked host memory and returns an error if a pointer topageable memory is passed as input.

RETURN VALUE


CUDA_SUCCESS






SEE ALSO


cpyAtoA, cuMemcpy2D

179

2.8.22 cuMemcpyHtoD

NAME

cuMemcpyHtoD - copy memory from Host to Device

SYNOPSIS

CUresult cuMemcpyHtoD(CUdeviceptr dstDevPtr, const void *srcHostPtr, unsigned int count);

CUresult cuMemcpyHtoDAsync(CUdeviceptr dstDevPtr, const void *srcHostPtr, unsigned int count,

CUstream stream);

DESCRIPTION

Copies from host memory to device memory. dstDevPtr and srcHostPtr specify the base addresses ofthe destination and source, respectively. count specifies the number of bytes to copy.

cuMemcpyHtoDAsync() is asynchronous and can optionally be associated to a stream by passing a non-zero stream argument. It only works on page-locked host memory and returns an error if a pointer topageable memory is passed as input.

RETURN VALUE


CUDA_SUCCESS






SEE ALSO

cuMemcpyDtoH, cuMemcpyDtoD, cuMemcpyDtoA, cuMemcpyAtoD, cuMemcpyAtoH, cuMemcpyHtoA, cuMem-

cpyAtoA, cuMemcpy2D

180

2.8.23 cuMemset

NAME

cuMemset - initializes device memory

SYNOPSIS

CUresult cuMemsetD8(CUdeviceptr dstDevPtr, unsigned char value, unsigned int count );

CUresult cuMemsetD16(CUdeviceptr dstDevPtr, unsigned short value, unsigned int count );

CUresult cuMemsetD32(CUdeviceptr dstDevPtr, unsigned int value, unsigned int count );

DESCRIPTION

Sets the memory range of count 8-, 16-, or 32-bit values to the specified value value.

RETURN VALUE


CUDA_SUCCESS






SEE ALSO


tAddressRange, cuArrayCreate, cuArrayGetDescriptor, cuArrayDestroy, cuMemset2D

181

2.8.24 cuMemset2D

NAME

cuMemset2D - initializes device memory

SYNOPSIS

CUresult cuMemsetD2D8(CUdeviceptr dstDevPtr, unsigned int dstPitch, unsigned char value, unsigned

int width, unsigned int height );

CUresult cuMemsetD2D16(CUdeviceptr dstDevPtr, unsigned int dstPitch, unsigned short value,

unsigned int width, unsigned int height );

CUresult cuMemsetD2D32(CUdeviceptr dstDevPtr, unsigned int dstPitch, unsigned int value, unsigned

int width, unsigned int height );

DESCRIPTION

Sets the 2D memory range of width 8-, 16-, or 32-bit values to the specified value value. height specifiesthe number of rows to set, and dstPitch specifies the number of bytes between each row. These functionsperform fastest when the pitch is one that has been passed back by cuMemAllocPitch().

RETURN VALUE


CUDA_SUCCESS






SEE ALSO


tAddressRange, cuArrayCreate, cuArrayGetDescriptor, cuArrayDestroy, cuMemset

182

2.9 TextureReferenceManagement

NAME

Texture Reference Management

DESCRIPTION


cuTexRefCreate

cuTexRefDestroy

cuTexRefGetAddress

cuTexRefGetAddressMode

cuTexRefGetArray

cuTexRefGetFilterMode

cuTexRefGetFlags

cuTexRefGetFormat

cuTexRefSetAddress

cuTexRefSetAddressMode

cuTexRefSetArray

cuTexRefSetFilterMode

cuTexRefSetFlags

cuTexRefSetFormat

SEE ALSO




183

2.9.1 cuTexRefCreate

NAME

cuTexRefCreate - creates a texture-reference

SYNOPSIS

CUresult cuTexRefCreate(CUtexref* texRef);

DESCRIPTION

Creates a texture reference and returns its handle in *texRef. Once created, the application must callcuTexRefSetArray() or cuTexRefSetAddress() to associate the reference with allocated memory. Othertexture reference functions are used to specify the format and interpretation (addressing, filtering, etc.) tobe used when the memory is read through this texture reference. To associate the texture reference with atexture ordinal for a given function, the application should call cuParamSetTexRef().

RETURN VALUE


CUDA_SUCCESS






SEE ALSO

cuTexRefDestroy, cuTexRefSetArray, cuTexRefSetAddress, cuTexRefSetFormat, cuTexRefSetAddressMode,cuTexRefSetFilterMode, cuTexRefSetFlags, cuTexRefGetAddress, cuTexRefGetArray, cuTexRefGetAddress-

Mode, cuTexRefGetFilterMode, cuTexRefGetFormat, cuTexRefGetFlags

184

2.9.2 cuTexRefDestroy

NAME

cuTexRefDestroy - destroys a texture-reference

SYNOPSIS

CUresult cuTexRefDestroy(CUtexref texRef);

DESCRIPTION

Destroys the texture reference.

RETURN VALUE


CUDA_SUCCESS






SEE ALSO

cuTexRefCreate, cuTexRefSetArray, cuTexRefSetAddress, cuTexRefSetFormat, cuTexRefSetAddressMode, cu-

TexRefSetFilterMode, cuTexRefSetFlags, cuTexRefGetAddress, cuTexRefGetArray, cuTexRefGetAddressMode,cuTexRefGetFilterMode, cuTexRefGetFormat, cuTexRefGetFlags

185

2.9.3 cuTexRefGetAddress

NAME

cuTexRefGetAddress - gets the address associated with a texture-reference

SYNOPSIS

CUresult cuTexRefGetAddress(CUdeviceptr* devPtr, CUtexref texRef);

DESCRIPTION

Returns in *devPtr the base address bound to the texture reference texRef, or returns CUDA_ERROR_INVALID_VALUEif the texture reference is not bound to any device memory range.

RETURN VALUE


CUDA_SUCCESS






SEE ALSO

cuTexRefCreate, cuTexRefDestroy, cuTexRefSetArray, cuTexRefSetAddress, cuTexRefSetFormat, cuTexRef-

SetAddressMode, cuTexRefSetFilterMode, cuTexRefSetFlags, cuTexRefGetArray, cuTexRefGetAddressMode,cuTexRefGetFilterMode, cuTexRefGetFormat, cuTexRefGetFlags

186

2.9.4 cuTexRefGetAddressMode

NAME

cuTexRefGetAddressMode - gets the addressing mode used by a texture-reference

SYNOPSIS

CUresult cuTexRefGetAddressMode(CUaddress_mode* mode, CUtexref texRef, int dim);

DESCRIPTION

Returns in *mode the addressing mode corresponding to the dimension dim of the texture reference texRef.Currently the only valid values for dim are 0 and 1.

RETURN VALUE


CUDA_SUCCESS






SEE ALSO


SetAddressMode, cuTexRefSetFilterMode, cuTexRefSetFlags, cuTexRefGetAddress, cuTexRefGetArray, cu-

TexRefGetFilterMode, cuTexRefGetFormat, cuTexRefGetFlags

187

2.9.5 cuTexRefGetArray

NAME

cuTexRefGetArray - gets the array bound to a texture-reference

SYNOPSIS

CUresult cuTexRefGetArray(CUarray* array, CUtexref texRef);

DESCRIPTION

Returns in *array the CUDA array bound by the texture reference texRef, or returns CUDA_ERROR_INVALID_VALUEif the texture reference is not bound to any CUDA array.

RETURN VALUE


CUDA_SUCCESS






SEE ALSO


SetAddressMode, cuTexRefSetFilterMode, cuTexRefSetFlags, cuTexRefGetAddress, cuTexRefGetAddressMode,cuTexRefGetFilterMode, cuTexRefGetFormat, cuTexRefGetFlags

188

2.9.6 cuTexRefGetFilterMode

NAME

cuTexRefGetFilterMode - gets the filter-mode used by a texture-reference

SYNOPSIS

CUresult cuTexRefGetFilterMode(CUfilter_mode* mode, CUtexref texRef);

DESCRIPTION

Returns in *mode the filtering mode of the texture reference texRef.

RETURN VALUE


CUDA_SUCCESS






SEE ALSO



TexRefGetAddressMode, cuTexRefGetFormat, cuTexRefGetFlags

189

2.9.7 cuTexRefGetFlags

NAME

cuTexRefGetFlags - gets the flags used by a texture-reference

SYNOPSIS

CUresult cuTexRefGetFlags(unsigned int* flags, CUtexref texRef);

DESCRIPTION

Returns in *flags the flags of the texture reference texRef.

RETURN VALUE


CUDA_SUCCESS






SEE ALSO



TexRefGetAddressMode, cuTexRefGetFilterMode, cuTexRefGetFormat

190

2.9.8 cuTexRefGetFormat

NAME

cuTexRefGetFormat - gets the format used by a texture-reference

SYNOPSIS

CUresult cuTexRefGetFormat(CUarray_format* format, int* numPackedComponents, CUtexref texRef);

DESCRIPTION

Returns in *format and *numPackedComponents the format and number of components of the CUDAarray bound to the texture reference texRef. If format or numPackedComponents is null, it will beignored.

RETURN VALUE


CUDA_SUCCESS






SEE ALSO



TexRefGetAddressMode, cuTexRefGetFilterMode, cuTexRefGetFlags

191

2.9.9 cuTexRefSetAddress

NAME

cuTexRefSetAddress - binds an address as a texture-reference

SYNOPSIS

CUresult cuTexRefSetAddress(unsigned int* byteOffset, CUtexref texRef, CUdeviceptr devPtr,

int bytes);

DESCRIPTION

Binds a linear address range to the texture reference texRef. Any previous address or CUDA array stateassociated with the texture reference is superseded by this function. Any memory previously bound totexRef is unbound.

Since the hardware enforces an alignment requirement on texture base addresses, cuTexRefSetAddress()passes back a byte offset in *byteOffset that must be applied to texture fetches in order to read from thedesired memory. This offset must be divided by the texel size and passed to kernels that read from thetexture so they can be applied to the tex1Dfetch() function.

If the device memory pointer was returned from cuMemAlloc(), the offset is guaranteed to be 0 and NULLmay be passed as the ByteOffset parameter.

RETURN VALUE


CUDA_SUCCESS






SEE ALSO

cuTexRefCreate, cuTexRefDestroy, cuTexRefSetArray, cuTexRefSetFormat, cuTexRefSetAddressMode, cu-


192

2.9.10 cuTexRefSetAddressMode

NAME

cuTexRefSetAddressMode - set the addressing mode for a texture-reference

SYNOPSIS

CUresult cuTexRefSetAddressMode(CUtexref texRef, int dim, CUaddress_mode mode);

DESCRIPTION

Specifies the addressing mode mode for the given dimension of the texture reference texRef. If dim iszero, the addressing mode is applied to the first parameter of the functions used to fetch from the texture;if dim is 1, the second, and so on. CUaddress_mode is defined as such:

typedef enum CUaddress_mode_enum {

CU_TR_ADDRESS_MODE_WRAP = 0,

CU_TR_ADDRESS_MODE_CLAMP = 1,

CU_TR_ADDRESS_MODE_MIRROR = 2,

} CUaddress_mode;

Note that this call has no effect if texRef is bound to linear memory.

RETURN VALUE


CUDA_SUCCESS






SEE ALSO


SetFilterMode, cuTexRefSetFlags, cuTexRefGetAddress, cuTexRefGetArray, cuTexRefGetAddressMode, cu-


193

2.9.11 cuTexRefSetArray

NAME

cuTexRefSetArray - binds an array to a texture-reference

SYNOPSIS

CUresult cuTexRefSetArray(CUtexref texRef, CUarray array, unsigned int flags);

DESCRIPTION

Binds the CUDA array array to the texture reference texRef. Any previous address or CUDA ar-ray state associated with the texture reference is superseded by this function. flags must be set toCU_TRSA_OVERRIDE_FORMAT. Any CUDA array previously bound to texRef is unbound.

RETURN VALUE


CUDA_SUCCESS






SEE ALSO

cuTexRefCreate, cuTexRefDestroy, cuTexRefSetAddress, cuTexRefSetFormat, cuTexRefSetAddressMode, cu-


194

2.9.12 cuTexRefSetFilterMode

NAME

cuTexRefSetFilterMode - sets the mode for a texture-reference

SYNOPSIS

CUresult cuTexRefSetFilterMode(CUtexref texRef, CUfilter_mode mode);

DESCRIPTION

Specifies the filtering mode mode to be used when reading memory through the texture reference texRef.CUfilter_mode_enum is defined as such:

typedef enum CUfilter_mode_enum {

CU_TR_FILTER_MODE_POINT = 0,

CU_TR_FILTER_MODE_LINEAR = 1

} CUfilter_mode;

Note that this call has no effect if texRef is bound to linear memory.

RETURN VALUE


CUDA_SUCCESS






SEE ALSO


SetAddressMode, cuTexRefSetFlags, cuTexRefGetAddress, cuTexRefGetArray, cuTexRefGetAddressMode, cu-


195

2.9.13 cuTexRefSetFlags

NAME

cuTexRefSetFlags - sets flags for a texture-reference

SYNOPSIS

CUresult cuTexRefSetFlags(CUtexref texRef, unsigned int Flags);

DESCRIPTION

Specifies optional flags to control the behavior of data returned through the texture reference. The validflags are:

• CU_TRSF_READ_AS_INTEGER, which suppresses the default behavior of having the texturepromote integer data to floating point data in the range [0, 1];

• CU_TRSF_NORMALIZED_COORDINATES, which suppresses the default behavior of hav-ing the texture coordinates range from [0, Dim) where Dim is the width or height of the CUDA array.Instead, the texture coordinates [0, 1.0) reference the entire breadth of the array dimension

RETURN VALUE


CUDA_SUCCESS






SEE ALSO


SetAddressMode, cuTexRefSetFilterMode, cuTexRefGetAddress, cuTexRefGetArray, cuTexRefGetAddressMode,cuTexRefGetFilterMode, cuTexRefGetFormat, cuTexRefGetFlags

196

2.9.14 cuTexRefSetFormat

NAME

cuTexRefSetFormat - sets the format for a texture-reference

SYNOPSIS

CUresult cuTexRefSetFormat(CUtexref texRef, CUarray_format format, int numPackedComponents)

DESCRIPTION

Specifies the format of the data to be read by the texture reference texRef. format and numPackedCom-ponents are exactly analogous to the Format and NumChannels members of the CUDA_ARRAY_DESCRIPTORstructure: They specify the format of each component and the number of components per array element.

RETURN VALUE


CUDA_SUCCESS






SEE ALSO

cuTexRefCreate, cuTexRefDestroy, cuTexRefSetArray, cuTexRefSetAddress, cuTexRefSetAddressMode, cu-


197

2.10 OpenGlInteroperability

NAME

OpenGL Interoperability

DESCRIPTION


cuGLCtxCreate

cuGLInit

cuGLMapBufferObject

cuGLRegisterBufferObject

cuGLUnmapBufferObject

cuGLUnregisterBufferObject

SEE ALSO


Management, ExecutionControl, MemoryManagement, TextureReferenceManagement, OpenGLInteroperabil-


198

2.10.1 cuGLCtxCreate

NAME

cuGLCtxCreate - create a CUDA context for interoperability with OpenGL

SYNOPSIS

CUresult cuGLCtxCreate(CUcontext *pCtx, unsigned int Flags, CUdevice device);

DESCRIPTION

Creates a new CUDA context, initializes OpenGL interoperability, and associates the CUDA context withthe calling thread. It must be called before performing any other OpenGL interoperability operations. Itmay fail if the needed OpenGL driver facilities are not available. For usage of the Flags parameter, seecuCtxCreate.

RETURN VALUE


CUDA_SUCCESS







SEE ALSO

cuCtxCreate, cuGLInit cuGLRegisterBufferObject, cuGLMapBufferObject, cuGLUnmapBufferObject, cuGLUn-

registerBufferObject

199

2.10.2 cuGLInit

NAME

cuGLInit - initializes GL interoperability

SYNOPSIS

CUresult cuGLInit(void);

DESCRIPTION

Initializes OpenGL interoperability. It must be called before performing any other OpenGL interoperabilityoperations. It may fail if the needed OpenGL driver facilities are not available.

RETURN VALUE


CUDA_SUCCESS




CUDA_ERROR_UNKNOWN


SEE ALSO

cuGLCtxCreate, cuGLRegisterBufferObject, cuGLMapBufferObject, cuGLUnmapBufferObject, cuGLUnreg-

isterBufferObject

200

2.10.3 cuGLMapBufferObject

NAME

cuGLMapBufferObject - maps a GL buffer object

SYNOPSIS

CUresult cuGLMapBufferObject(CUdeviceptr* devPtr, unsigned int* size, GLuint bufferObj);

DESCRIPTION

Maps the buffer object of ID bufferObj into the address space of the current CUDA context and returnsin *devPtr and *size the base pointer and size of the resulting mapping.

RETURN VALUE


CUDA_SUCCESS





CUDA_ERROR_MAP_FAILED


SEE ALSO

cuGLCtxCreate, cuGLInit, cuGLRegisterBufferObject, cuGLUnmapBufferObject, cuGLUnregisterBufferOb-

ject

201

2.10.4 cuGLRegisterBufferObject

NAME

cuGLRegisterBufferObject - registers a GL buffer object

SYNOPSIS

CUresult cuGLRegisterBufferObject(GLuint bufferObj);

DESCRIPTION

Registers the buffer object of ID bufferObj for access by CUDA. This function must be called before CUDAcan map the buffer object. While it is registered, the buffer object cannot be used by any OpenGL commandsexcept as a data source for OpenGL drawing commands.

RETURN VALUE


CUDA_SUCCESS




CUDA_ERROR_ALREADY_MAPPED


SEE ALSO

cuGLCtxCreate, cuGLInit, cuGLMapBufferObject, cuGLUnmapBufferObject, cuGLUnregisterBufferObject

202

2.10.5 cuGLUnmapBufferObject

NAME

cuGLUnmapBufferObject - unmaps a GL buffer object

SYNOPSIS

CUresult cuGLUnmapBufferObject(GLuint bufferObj);

DESCRIPTION

Unmaps the buffer object of ID bufferObj for access by CUDA.

RETURN VALUE


CUDA_SUCCESS






SEE ALSO

cuGLCtxCreate, cuGLInit, cuGLRegisterBufferObject, cuGLMapBufferObject, cuGLUnregisterBufferObject

203

2.10.6 cuGLUnregisterBufferObject

NAME

cuGLUnregisterBufferObject - unregister a GL buffer object

SYNOPSIS

CUresult cuGLUnregisterBufferObject(GLuint bufferObj);

DESCRIPTION

Unregisters the buffer object of ID bufferObj for access by CUDA.

RETURN VALUE


CUDA_SUCCESS






SEE ALSO

cuGLCtxCreate, cuGLInit, cuGLRegisterBufferObject, cuGLMapBufferObject, cuGLUnmapBufferObject

204

2.11 Direct3dInteroperability

NAME

Direct3D Interoperability

DESCRIPTION

This section describes Direct3D interoperability in the the low-level CUDA driver application programminginterface.

cuD3D9GetDevice

cuD3D9CtxCreate

cuD3D9GetDirect3DDevice

cuD3D9RegisterResource

cuD3D9UnregisterResource

cuD3D9MapResources

cuD3D9UnmapResources

cuD3D9ResourceGetSurfaceDimensions

cuD3D9ResourceSetMapFlags

cuD3D9ResourceGetMappedPointer

cuD3D9ResourceGetMappedSize

cuD3D9ResourceGetMappedPitch

As of CUDA 2.0 the following functions are deprecated. They should not be used in new development.

cuD3D9Begin

cuD3D9End

cuD3D9MapVertexBuffer

cuD3D9RegisterVertexBuffer

cuD3D9UnmapVertexBuffer

cuD3D9UnregisterVertexBuffer

SEE ALSO


Management, ExecutionControl, MemoryManagement, TextureReferenceManagement, OpenGLInteroperabil-


205

2.11.1 cuD3D9GetDevice

NAME

cuD3D9GetDevice - gets the device number for an adapter

SYNOPSIS

CUresult cuD3D9GetDevice(CUdevice* dev, const char* adapterName);

DESCRIPTION

Returns in *dev the CUDA-compatible device corresponding to the adapter name adapterName obtainedfrom EnumDisplayDevices or IDirect3D9::GetAdapterIdentifier(). If no device on the adapter withname adapterName is CUDA-compatible then the call will fail.

RETURN VALUE


CUDA_SUCCESS




CUDA_ERROR_UNKNOWN


SEE ALSO

cuD3D9CtxCreate, cuD3D9GetDirect3DDevice, cuD3D9RegisterResource, cuD3D9UnregisterResource, cuD3D9MapResources,cuD3D9UnmapResources, cuD3D9ResourceGetSurfaceDimensions, cuD3D9ResourceSetMapFlags, cuD3D9ResourceGetMappe

cuD3D9ResourceGetMappedSize, cuD3D9ResourceGetMappedPitch

206

2.11.2 cuD3D9CtxCreate

NAME

cuD3D9CtxCreate - create a CUDA context for interoperability with Direct3D

SYNOPSIS

CUresult cuD3D9CtxCreate(CUcontext* pCtx, CUdevice* pCuDevice, unsigned int Flags, IDirect3DDevice9*

pDxDevice);

DESCRIPTION

Creates a new CUDA context, enables interoperability for that context with the Direct3D device pDxDe-vice, and associates the created CUDA context with the calling thread. The CUcontext created will bereturned in *pCtx. If pCuDevice is non-NULL then the CUdevice on which this CUDA context wascreated will be returned in *pCuDevice. For usage of the Flags parameter, see cuCtxCreate. Direct3Dresources from this device may be registered and mapped through the lifetime of this CUDA context.

This context will function only until its Direct3D device is destroyed. On success, this call will increase theinternal reference count on pDxDevice. This reference count will be decremented upon destruction of thiscontext through cuCtxDestroy.

RETURN VALUE


CUDA_SUCCESS





CUDA_ERROR_UNKNOWN


SEE ALSO

cuD3D9GetDevice, cuD3D9GetDirect3DDevice, cuD3D9RegisterResource, cuD3D9UnregisterResource, cuD3D9MapResources,cuD3D9UnmapResources, cuD3D9ResourceGetSurfaceDimensions, cuD3D9ResourceSetMapFlags, cuD3D9ResourceGetMappe


207

2.11.3 cuD3D9GetDirect3DDevice

NAME

cuD3D9GetDirect3DDevice - get the Direct3D device against which the current CUDA context wascreated

SYNOPSIS

CUresult cuD3D9GetDirect3DDevice(IDirect3DDevice9** ppDxDevice);

DESCRIPTION

Returns in *ppDxDevice the Direct3D device against which this CUDA context was created in cuD3D9CtxCreate.

RETURN VALUE


CUDA_SUCCESS





SEE ALSO

cuD3D9GetDevice, cuD3D9CtxCreate, cuD3D9RegisterResource, cuD3D9UnregisterResource, cuD3D9MapResources,cuD3D9UnmapResources, cuD3D9ResourceGetSurfaceDimensions, cuD3D9ResourceSetMapFlags, cuD3D9ResourceGetMappe


208

2.11.4 cuD3D9RegisterResource

NAME

cuD3D9RegisterResource - register a Direct3D resource for access by CUDA

SYNOPSIS

CUresult cuD3D9RegisterResource(IDirect3DResource9* pResource, unsigned int Flags);

DESCRIPTION

Registers the Direct3D resource pResource for access by CUDA.

If this call is successful then the application will be able to map and unmap this resource until it is un-registered through cuD3D9UnregisterResource. Also on success, this call will increase the internal referencecount on pResource. This reference count will be decremented when this resource is unregistered throughcuD3D9UnregisterResource.

This call is potentially high-overhead and should not be called every frame in interactive applications.

The type of pResource must be one of the following.

• IDirect3DVertexBuffer9: No notes.

• IDirect3DIndexBuffer9: No notes.

• IDirect3DSurface9: Only stand-alone objects of type IDirect3DSurface9 may be explicitly shared.In particular, individual mipmap levels and faces of cube maps may not be registered directly. To accessindividual surfaces associated with a texture, one must register the base texture object.

• IDirect3DBaseTexture9: When a texture is registered all surfaces associated with the all mipmaplevels of all faces of the texture will be accessible to CUDA.

The Flags argument specifies the mechanism through which CUDA will access the Direct3D resource. Thefollowing value is allowed.

• CU_D3D9_REGISTER_FLAGS_NONE: Specifies that CUDA will access this resource througha CUdeviceptr. The pointer, size, and pitch for each subresource of this resource may be queriedthrough cuD3D9ResourceGetMappedPointer, cuD3D9ResourceGetMappedSize, and cuD3D9ResourceGetMappedPitch

respectively. This option is valid for all resource types.

Not all Direct3D resources of the above types may be used for interoperability with CUDA. The followingare some limitations.

• The primary rendertarget may not be registered with CUDA.

• Resources allocated as shared may not be registered with CUDA.

• Any resources allocated in D3DPOOL_SYSTEMMEM may not be registered with CUDA.

• Textures which are not of a format which is 1, 2, or 4 channels of 8, 16, or 32-bit integer or floating-pointdata cannot be shared.

209

• Surfaces of depth or stencil formats cannot be shared.

If Direct3D interoperability is not initialized on this context then CUDA_ERROR_INVALID_CONTEXTis returned. If pResource is of incorrect type (e.g, is a non-stand-alone IDirect3DSurface9) or is alreadyregistered then CUDA_ERROR_INVALID_HANDLE is returned. If pResource cannot be regis-tered then CUDA_ERROR_UNKNOWN is returned.

RETURN VALUE


CUDA_SUCCESS







CUDA_ERROR_UNKNOWN


SEE ALSO

cuD3D9GetDevice, cuD3D9CtxCreate, cuD3D9GetDirect3DDevice, cuD3D9UnregisterResource, cuD3D9MapResources,cuD3D9UnmapResources, cuD3D9ResourceGetSurfaceDimensions, cuD3D9ResourceSetMapFlags, cuD3D9ResourceGetMappe


210

2.11.5 cuD3D9UnregisterResource

NAME

cuD3D9UnregisterResource - unregister a Direct3D resource

SYNOPSIS

CUresult cuD3D9UnregisterResource(IDirect3DResource9* pResource);

DESCRIPTION

Unregisters the Direct3D resource pResource so it is not accessable by CUDA unless registered again.

If pResource is not registered then CUDA_ERROR_INVALID_HANDLE is returned.

RETURN VALUE


CUDA_SUCCESS





CUDA_ERROR_UNKNOWN


SEE ALSO

cuD3D9GetDevice, cuD3D9CtxCreate, cuD3D9GetDirect3DDevice, cuD3D9RegisterResource, cuD3D9MapResources,cuD3D9UnmapResources, cuD3D9ResourceGetSurfaceDimensions, cuD3D9ResourceSetMapFlags, cuD3D9ResourceGetMappe


211

2.11.6 cuD3D9MapResources

NAME

cuD3D9MapResources - map Direct3D resources for access by CUDA

SYNOPSIS

CUresult cuD3D9MapResources(unsigned int count, IDirect3DResource9 **ppResources);

DESCRIPTION

Maps the count Direct3D resources in ppResources for access by CUDA.

The resources in ppResources may be accessed in CUDA kernels until they are unmapped. Direct3Dshould not access any resources while they are mapped by CUDA. If an application does so the results areundefined.

This function provides the synchronization guarantee that any Direct3D calls issued before cuD3D9MapResourceswill complete before any CUDA kernels issued after cuD3D9MapResources begin.

If any of ppResources have not been registered for use with CUDA or if ppResources contains anyduplicate entries then CUDA_ERROR_INVALID_HANDLE is returned. If any of ppResourcesare presently mapped for access by CUDA then CUDA_ERROR_ALREADY_MAPPED is returned.

RETURN VALUE


CUDA_SUCCESS






CUDA_ERROR_UNKNOWN


SEE ALSO

cuD3D9GetDevice, cuD3D9CtxCreate, cuD3D9GetDirect3DDevice, cuD3D9RegisterResource, cuD3D9UnregisterResource,cuD3D9UnmapResources, cuD3D9ResourceGetSurfaceDimensions, cuD3D9ResourceSetMapFlags, cuD3D9ResourceGetMappe


212

2.11.7 cuD3D9UnmapResources

NAME

cuD3D9UnmapResources - unmap Direct3D resources

SYNOPSIS

CUresult cuD3D9UnmapResources(unsigned int count, IDirect3DResource9** ppResources);

DESCRIPTION

Unmaps the count Direct3D resources in ppResources.

This function provides the synchronization guarantee that any CUDA kernels issued before cuD3D9UnmapResourceswill complete before any Direct3D calls issued after cuD3D9UnmapResources begin.

If any of ppResources have not been registered for use with CUDA or if ppResources contains anyduplicate entries then CUDA_ERROR_INVALID_HANDLE is returned. If any of ppResourcesare not presently mapped for access by CUDA then CUDA_ERROR_NOT_MAPPED is returned.

RETURN VALUE


CUDA_SUCCESS





CUDA_ERROR_NOT_MAPPED

CUDA_ERROR_UNKNOWN


SEE ALSO

cuD3D9GetDevice, cuD3D9CtxCreate, cuD3D9GetDirect3DDevice, cuD3D9RegisterResource, cuD3D9UnregisterResource,cuD3D9MapResources, cuD3D9ResourceGetSurfaceDimensions, cuD3D9ResourceSetMapFlags, cuD3D9ResourceGetMappedPointer


213

2.11.8 cuD3D9ResourceSetMapFlags

NAME

cuD3D9ResourceSetMapFlags - set usage flags for mapping a Direct3D resource

SYNOPSIS

CUresult cuD3D9ResourceSetMapFlags(IDirect3DResource9 *pResource, unsigned int Flags);

DESCRIPTION

Set flags for mapping the Direct3D resource pResource.

Changes to flags will take effect the next time pResource is mapped. The Flags argument may be any ofthe following.

• CU_D3D9_MAPRESOURCE_FLAGS_NONE: Specifies no hints about how this resourcewill be used. It is therefore assumed that this resource will be read from and written to by CUDAkernels. This is the default value.

• CU_D3D9_MAPRESOURCE_FLAGS_READONLY: Specifies that CUDA kernels whichaccess this resource will not write to this resource.

• CU_D3D9_MAPRESOURCE_FLAGS_WRITEDISCARD: Specifies that CUDA kernelswhich access this resource will not read from this resource and will write over the entire contentsof the resource, so none of the data previously stored in the resource will be preserved.

If pResource has not been registered for use with CUDA then CUDA_ERROR_INVALID_HANDLEis returned. If pResource is presently mapped for access by CUDA then CUDA_ERROR_ALREADY_MAPPEDis returned.

RETURN VALUE


CUDA_SUCCESS








214

SEE ALSO

cuD3D9GetDevice, cuD3D9CtxCreate, cuD3D9GetDirect3DDevice, cuD3D9RegisterResource, cuD3D9UnregisterResource,cuD3D9MapResources, cuD3D9UnmapResources, cuD3D9ResourceGetSurfaceDimensions, cuD3D9ResourceGetMappedPointer


215

2.11.9 cuD3D9ResourceGetSurfaceDimensions

NAME

cuD3D9ResourceGetSurfaceDimensions - get the dimensions of a registered surface

SYNOPSIS

CUresult cuD3D9ResourceGetSurfaceDimensions(unsigned int* pWidth, unsigned int* pHeight, unsigned

int *pDepth, IDirect3DResource9* pResource, unsigned int Face, unsigned int Level);

DESCRIPTION

Returns in *pWidth, *pHeight, and *pDepth the dimensions of the subresource of the mapped Direct3Dresource pResource which corresponds to Face and Level.

Because anti-aliased surfaces may have multiple samples per pixel it is possible that the dimensions of aresource will be an integer factor larger than the dimensions reported by the Direct3D runtime.

The parameters pWidth, pHeight, and pDepth are optional. For 2D surfaces, the value returned in*pDepth will be 0.

If pResource is not of type IDirect3DBaseTexture9 or IDirect3DSurface9 or if pResource has notbeen registered for use with CUDA then CUDA_ERROR_INVALID_HANDLE is returned.

For usage requirements of Face and Level parameters see cuD3D9ResourceGetMappedPointer.

RETURN VALUE


CUDA_SUCCESS







SEE ALSO

cuD3D9GetDevice, cuD3D9CtxCreate, cuD3D9GetDirect3DDevice, cuD3D9RegisterResource, cuD3D9UnregisterResource,cuD3D9MapResources, cuD3D9UnmapResources, cuD3D9ResourceSetMapFlags, cuD3D9ResourceGetMappedPointer,cuD3D9ResourceGetMappedSize, cuD3D9ResourceGetMappedPitch

216

2.11.10 cuD3D9ResourceGetMappedPointer

NAME

cuD3D9ResourceGetMappedPointer - get a pointer through which to access a subresource of a Direct3Dresource which has been mapped for access by CUDA

SYNOPSIS

CUresult cuD3D9ResourceGetMappedPointer(CUdeviceptr* pDevPtr, IDirect3DResource9* pResource,

unsigned int Face, unsigned int Level);

DESCRIPTION

Returns in *pDevPtr the base pointer of the subresource of the mapped Direct3D resource pResourcewhich corresponds to Face and Level. The value set in pDevPtr may change every time that pResourceis mapped.

If pResource is not registered then CUDA_ERROR_INVALID_HANDLE is returned. If pRe-source was not registered with usage flags CU_D3D9_REGISTER_FLAGS_NONE then CUDA_ERROR_INVALID_HANDLEis returned. If pResource is not mapped then CUDA_ERROR_NOT_MAPPED is returned.

If pResource is of type IDirect3DCubeTexture9 then Face must one of the values enumerated by typeD3DCUBEMAP_FACES. For all other types Face must be 0. If Face is invalid then CUDA_ERROR_INVALID_VALUEis returned.

If pResource is of type IDirect3DBaseTexture9 then Level must correspond to a valid mipmap level.At present only mipmap level 0 is supported. For all other types Level must be 0. If Level is invalid thenCUDA_ERROR_INVALID_VALUE is returned.

RETURN VALUE


CUDA_SUCCESS








217

SEE ALSO

cuD3D9GetDevice, cuD3D9CtxCreate, cuD3D9GetDirect3DDevice, cuD3D9RegisterResource, cuD3D9UnregisterResource,cuD3D9MapResources, cuD3D9UnmapResources, cuD3D9ResourceGetSurfaceDimensions, cuD3D9ResourceSetMapFlags,cuD3D9ResourceGetMappedSize, cuD3D9ResourceGetMappedPitch

218

2.11.11 cuD3D9ResourceGetMappedSize

NAME

cuD3D9ResourceGetMappedSize - get the size of a subresource of a Direct3D resource which has beenmapped for access by CUDA

SYNOPSIS

CUresult cuD3D9ResourceGetMappedSize(unsigned int* pSize, IDirect3DResource9* pResource, unsigned

int Face, unsigned int Level);

DESCRIPTION

Returns in *pSize the size of the subresource of the mapped Direct3D resource pResource which corre-sponds to Face and Level. The value set in pSize may change every time that pResource is mapped.

If pResource has not been registered for use with CUDA then CUDA_ERROR_INVALID_HANDLEis returned. If pResource was not registered with usage flags CU_D3D9_REGISTER_FLAGS_NONEthen CUDA_ERROR_INVALID_HANDLE is returned. If pResource is not mapped for access byCUDA then CUDA_ERROR_NOT_MAPPED is returned.


RETURN VALUE


CUDA_SUCCESS








SEE ALSO

cuD3D9GetDevice, cuD3D9CtxCreate, cuD3D9GetDirect3DDevice, cuD3D9RegisterResource, cuD3D9UnregisterResource,cuD3D9MapResources, cuD3D9UnmapResources, cuD3D9ResourceGetSurfaceDimensions, cuD3D9ResourceSetMapFlags,cuD3D9ResourceGetMappedPointer, cuD3D9ResourceGetMappedPitch

219

2.11.12 cuD3D9ResourceGetMappedPitch

NAME

cuD3D9ResourceGetMappedPitch - get the pitch of a subresource of a Direct3D resource which hasbeen mapped for access by CUDA

SYNOPSIS

CUresult cuD3D9ResourceGetMappedPitch(unsigned int* pPitch, unsigned int* pPitchSlice, IDirect3DResource9*

pResource, unsigned int Face, unsigned int Level);

DESCRIPTION

Returns in *pPitch and *pPitchSlice the pitch and Z-slice pitch of the subresource of the mapped Direct3Dresource pResource which corresponds to Face and Level. The values set in pPitch and pPitchSlicemay change every time that pResource is mapped.

The pitch and Z-slice pitch values may be used to compute the location of a sample on a surface as follows.

y*pitch + (bytes per pixel)*x

For a 3D surface the byte offset of the sample of at position x,y,z from the base pointer of the surface is

z*slicePitch + y*pitch + (bytes per pixel)*x

Both parameters pPitch and pPitchSlice are optional and may be set to NULL.

For a 2D surface the byte offset of the sample of at position x,y from the base pointer of the surface is

If pResource is not of type IDirect3DBaseTexture9 or one of its sub-types or if pResource hasnot been registered for use with CUDA then CUDA_ERROR_INVALID_HANDLE is returned.If pResource was not registered with usage flags CU_D3D9_REGISTER_FLAGS_NONE thenCUDA_ERROR_INVALID_HANDLE is returned. If pResource is not mapped for access by CUDAthen CUDA_ERROR_NOT_MAPPED is returned.


RETURN VALUE


CUDA_SUCCESS








220

SEE ALSO

cuD3D9GetDevice, cuD3D9CtxCreate, cuD3D9GetDirect3DDevice, cuD3D9RegisterResource, cuD3D9UnregisterResource,cuD3D9MapResources, cuD3D9UnmapResources, cuD3D9ResourceGetSurfaceDimensions, cuD3D9ResourceSetMapFlags,cuD3D9ResourceGetMappedPointer, cuD3D9ResourceGetMappedSize

221

3 AtomicFunctions

NAME

Atomic Functions

DESCRIPTION

Atomic functions can only be used in device functions.

NOTES

32-bit atomic operations are only supported on devices of compute capability 1.1 and higher. 64-bit atomicoperations are only supported on devices of compute capability 1.2 and higher.

SEE ALSO

ArithmeticFunctions, BitwiseFunctions

222

3.1 ArithmeticFunctions

NAME

Arithmetic Functions

DESCRIPTION

This section describes the atomic arithmetic functions.

atomicAdd

atomicSub

atomicExch

atomicMin

atomicMax

atomicInc

atomicDec

atomicCAS

SEE ALSO

BitwiseFunctions

223

3.1.1 atomicAdd

NAME

atomicAdd - atomic addition

SYNOPSIS

int atomicAdd(int* address, int val);

unsigned int atomicAdd(unsigned int* address, unsigned int val);

unsigned long long int atomicAdd(unsigned long long int* address, unsigned long long int val);

DESCRIPTION

Reads the 32- or 64-bit word old located at the address address in global memory, computes (old + val),and stores the result back to global memory at the same address. These three operations are performed inone atomic transaction. The function returns old.

NOTES


SEE ALSO

atomicSub, atomicExch, atomicMin, atomicMax, atomicInc, atomicDec, atomicCAS

224

3.1.2 atomicSub

NAME

atomicSub - atomic subtraction

SYNOPSIS

int atomicSub(int* address, int val);

unsigned int atomicSub(unsigned int* address, unsigned int val);

DESCRIPTION

Reads the 32-bit word old located at the address address in global memory, computes (old - val), andstores the result back to global memory at the same address. These three operations are performed in oneatomic transaction. The function returns old.

NOTES

Atomic operations are only supported on devices of compute capability 1.1 and higher.

SEE ALSO

atomicAdd, atomicExch, atomicMin, atomicMax, atomicInc, atomicDec, atomicCAS

225

3.1.3 atomicExch

NAME

atomicExch - atomic exchange

SYNOPSIS

int atomicExch(int* address, int val);

unsigned int atomicExch(unsigned int* address, unsigned int val);

unsigned long long int atomicExch(unsigned long long int* address, unsigned long long int val);

DESCRIPTION

Reads the 32- or 64-bit word old located at the address address in global memory and stores val back toglobal memory at the same address. These two operations are performed in one atomic transaction. Thefunction returns old.

NOTES


SEE ALSO

atomicAdd, atomicSub, atomicMin, atomicMax, atomicInc, atomicDec, atomicCAS

226

3.1.4 atomicMin

NAME

atomicMin - atomic minimum

SYNOPSIS

int atomicMin(int* address, int val);

unsigned int atomicMin(unsigned int* address, unsigned int val);

DESCRIPTION

Reads the 32-bit word old located at the address address in global memory, computes the minimum ofold and val, and stores the result back to global memory at the same address. These three operations areperformed in one atomic transaction. The function returns b<old>.

NOTES


SEE ALSO

atomicAdd, atomicSub, atomicExch, atomicMax, atomicInc, atomicDec, atomicCAS

227

3.1.5 atomicMax

NAME

atomicMax - atomic maximum

SYNOPSIS

int atomicMax(int* address, int val);

unsigned int atomicMax(unsigned int* address, unsigned int val);

DESCRIPTION

Reads the 32-bit word old located at the address address in global memory, computes the maximum ofold and val, and stores the result back to global memory at the same address. These three operations areperformed in one atomic transaction. The function returns old.

NOTES


SEE ALSO

atomicAdd, atomicSub, atomicExch, atomicMin, atomicInc, atomicDec, atomicCAS

228

3.1.6 atomicInc

NAME

atomicInc - atomic increment

SYNOPSIS

unsigned int atomicInc(unsigned int* address, unsigned int val);

DESCRIPTION

Reads the 32-bit word old located at the address address in global memory, computes ((old >= val) ? 0: (old+1)), and stores the result back to global memory at the same address. These three operations areperformed in one atomic transaction. The function returns old.

NOTES


SEE ALSO

atomicAdd, atomicSub, atomicExch, atomicMin, atomicMax, atomicDec, atomicCAS

229

3.1.7 atomicDec

NAME

atomicDec - atomic decrement

SYNOPSIS

unsigned int atomicDec(unsigned int* address, unsigned int val);

DESCRIPTION

Reads the 32-bit word old located at the address address in global memory, computes (((old == 0) , andstores the result back to global memory at the same address. These three operations are performed in oneatomic transaction. The function returns old.

NOTES


SEE ALSO

atomicAdd, atomicSub, atomicExch, atomicMin, atomicMax, atomicInc, atomicCAS

230

3.1.8 atomicCAS

NAME

atomicCAS - atomic compare-and-swap

SYNOPSIS

int atomicCAS(int* address, int compare, int val);

unsigned int atomicCAS(unsigned int* address, unsigned int compare, unsigned int val);

unsigned long long int atomicCAS(unsigned long long int* address, unsigned long long int compare,

unsigned long long int val);

DESCRIPTION

Reads the 32- or 64-bit word old located at the address address in global memory, computes (old ==compare ? val : old), and stores the result back to global memory at the same address. These threeoperations are performed in one atomic transaction. The function returns old (Compare And Swap).

NOTES


SEE ALSO

atomicAdd, atomicSub, atomicExch, atomicMin, atomicMax, atomicInc, atomicDec

231

3.2 BitwiseFunctions

NAME

Bitwise Functions

DESCRIPTION

This section describes the atomic bitwise functions.

atomicAnd

atomicOr

atomicXor

SEE ALSO

ArithmeticFunctions

232

3.2.1 atomicAnd

NAME

atomicAnd - atomic bitwise-and

SYNOPSIS

int atomicAnd(int* address, int val);

unsigned int atomicAnd(unsigned int* address, unsigned int val);

DESCRIPTION

Reads the 32-bit word old located at the address address in global memory, computes (old & val), andstores the result back to global memory at the same address. These three operations are performed in oneatomic transaction. The function returns old.

NOTES


SEE ALSO

atomicOr, atomicXor

233

3.2.2 atomicOr

NAME

atomicOr - atomic bitwise-or

SYNOPSIS

int atomicOr(int* address, int val);

unsigned int atomicOr(unsigned int* address, unsigned int val);

DESCRIPTION

Reads the 32-bit word old located at the address address in global memory, computes (old , and storesthe result back to global memory at the same address. These three operations are performed in one atomictransaction. The function returns old.

NOTES


SEE ALSO

atomicAnd, atomicXor

234

3.2.3 atomicXor

NAME

atomicXor - atomic bitwise-xor

SYNOPSIS

int atomicXor(int* address, int val);

unsigned int atomicXor(unsigned int* address, unsigned int val);

DESCRIPTION

Reads the 32-bit word old located at the address address in global memory, computes (old ˆ val), andstores the result back to global memory at the same address. These three operations are performed in oneatomic transaction. The function returns old.

NOTES


SEE ALSO

atomicAnd, atomicOr

235

Index

ArithmeticFunctions, 223atomicAdd, 224atomicAnd, 233atomicCAS, 231atomicDec, 230atomicExch, 226AtomicFunctions, 222atomicInc, 229atomicMax, 228atomicMin, 227atomicOr, 234atomicSub, 225atomicXor, 235

BitwiseFunctions, 232

ContextManagement, 111cuArray3DCreate, 154cuArray3DGetDescriptor, 158cuArrayCreate, 152cuArrayDestroy, 156cuArrayGetDescriptor, 157cuCtxAttach, 112cuCtxCreate, 113cuCtxDestroy, 115cuCtxDetach, 116cuCtxGetDevice, 117cuCtxPopCurrent, 118cuCtxPushCurrent, 119cuCtxSynchronize, 120cuD3D9CtxCreate, 207cuD3D9GetDevice, 206cuD3D9GetDirect3DDevice, 208cuD3D9MapResources, 212cuD3D9RegisterResource, 209cuD3D9ResourceGetMappedPitch, 220cuD3D9ResourceGetMappedPointer, 217cuD3D9ResourceGetMappedSize, 219cuD3D9ResourceGetSurfaceDimensions, 216cuD3D9ResourceSetMapFlags, 214cuD3D9UnmapResources, 213cuD3D9UnregisterResource, 211cudaBindTexture, 59cudaBindTexture HL, 65cudaBindTextureToArray, 60cudaBindTextureToArray HL, 66cudaChooseDevice, 8cudaConfigureCall, 69cudaCreateChannelDesc, 56cudaCreateChannelDesc HL, 64cudaD3D9GetDevice, 79

cudaD3D9GetDirect3DDevice, 81cudaD3D9MapResources, 85cudaD3D9RegisterResource, 82cudaD3D9ResourceGetMappedPitch, 92cudaD3D9ResourceGetMappedPointer, 90cudaD3D9ResourceGetMappedSize, 91cudaD3D9ResourceGetSurfaceDimensions, 89cudaD3D9ResourceSetMapFlags, 87cudaD3D9SetDirect3DDevice, 80cudaD3D9UnmapResources, 86cudaD3D9UnregisterResource, 84cudaEventCreate, 18cudaEventDestroy, 22cudaEventElapsedTime, 23cudaEventQuery, 20cudaEventRecord, 19cudaEventSynchronize, 21cudaFree, 27cudaFreeArray, 29cudaFreeHost, 31cudaGetChannelDesc, 57cudaGetDevice, 5cudaGetDeviceCount, 3cudaGetDeviceProperties, 6cudaGetErrorString, 97cudaGetLastError, 95cudaGetSymbolAddress, 44cudaGetSymbolSize, 45cudaGetTextureAlignmentOffset, 62cudaGetTextureReference, 58cudaGLMapBufferObject, 75cudaGLRegisterBufferObject, 74cudaGLSetGLDevice, 73cudaGLUnmapBufferObject, 76cudaGLUnregisterBufferObject, 77cudaLaunch, 70cudaMalloc, 25cudaMalloc3D, 46cudaMalloc3DArray, 48cudaMallocArray, 28cudaMallocHost, 30cudaMallocPitch, 26cudaMemcpy, 34cudaMemcpy2D, 35cudaMemcpy2DArrayToArray, 41cudaMemcpy2DFromArray, 39cudaMemcpy2DToArray, 37cudaMemcpy3D, 52cudaMemcpyArrayToArray, 40cudaMemcpyFromArray, 38cudaMemcpyFromSymbol, 43

236

cudaMemcpyToArray, 36cudaMemcpyToSymbol, 42cudaMemset, 32cudaMemset2D, 33cudaMemset3D, 50cudaSetDevice, 4cudaSetupArgument, 71cudaStreamCreate, 13cudaStreamDestroy, 16cudaStreamQuery, 14cudaStreamSynchronize, 15cudaThreadExit, 11cudaThreadSynchronize, 10cudaUnbindTexture, 61cudaUnbindTexture HL, 67cuDeviceComputeCapability, 102cuDeviceGet, 103cuDeviceGetAttribute, 104cuDeviceGetCount, 106cuDeviceGetName, 107cuDeviceGetProperties, 108cuDeviceTotalMem, 110cuEventCreate, 135cuEventDestroy, 136cuEventElapsedTime, 137cuEventQuery, 138cuEventRecord, 139cuEventSynchronize, 140cuFuncSetBlockShape, 149cuFuncSetSharedSize, 150cuGLCtxCreate, 199cuGLInit, 200cuGLMapBufferObject, 201cuGLRegisterBufferObject, 202cuGLUnmapBufferObject, 203cuGLUnregisterBufferObject, 204cuInit, 100cuLaunch, 142cuLaunchGrid, 143cuMemAlloc, 159cuMemAllocHost, 160cuMemAllocPitch, 161cuMemcpy2D, 167cuMemcpy3D, 170cuMemcpyAtoA, 173cuMemcpyAtoD, 174cuMemcpyAtoH, 175cuMemcpyDtoA, 176cuMemcpyDtoD, 177cuMemcpyDtoH, 178cuMemcpyHtoA, 179cuMemcpyHtoD, 180cuMemFree, 163cuMemFreeHost, 164

cuMemGetAddressRange, 165cuMemGetInfo, 166cuMemset, 181cuMemset2D, 182cuModuleGetFunction, 122cuModuleGetGlobal, 123cuModuleGetTexRef, 124cuModuleLoad, 125cuModuleLoadData, 126cuModuleLoadFatBinary, 127cuModuleUnload, 128cuParamSetf, 146cuParamSeti, 147cuParamSetSize, 144cuParamSetTexRef, 145cuParamSetv, 148cuStreamCreate, 130cuStreamDestroy, 131cuStreamQuery, 132cuStreamSynchronize, 133cuTexRefCreate, 184cuTexRefDestroy, 185cuTexRefGetAddress, 186cuTexRefGetAddressMode, 187cuTexRefGetArray, 188cuTexRefGetFilterMode, 189cuTexRefGetFlags, 190cuTexRefGetFormat, 191cuTexRefSetAddress, 192cuTexRefSetAddressMode, 193cuTexRefSetArray, 194cuTexRefSetFilterMode, 195cuTexRefSetFlags, 196cuTexRefSetFormat, 197

DeviceManagement, 101DeviceManagement RT, 2Direct3dInteroperability, 205Direct3dInteroperability RT, 78DriverApiReference, 98

ErrorHandling RT, 94EventManagement, 134EventManagement RT, 17ExecutionControl, 141ExecutionControl RT, 68

HighLevelApi, 63

Initialization, 99

LowLevelApi, 55

MemoryManagement, 151MemoryManagement RT, 24

237

ModuleManagement, 121

OpenGlInteroperability, 198OpenGlInteroperability RT, 72

RuntimeApiReference, 1

StreamManagement, 129StreamManagement RT, 12

TextureReferenceManagement, 183TextureReferenceManagement RT, 54ThreadManagement RT, 9

238

nvidia cuda compute unified device...

Documents