Rebase to CTK 12.3

Author: vzhurba01

README.md

@@ -7,8 +7,8 @@
 CUDA Python is supported on all platforms that CUDA is supported. Specific dependencies are as follows:
 
 * Driver: Linux (450.80.02 or later) Windows (456.38 or later)
-* CUDA Toolkit 12.0 to 12.2
-* Python 3.8 to 3.11
+* CUDA Toolkit 12.0 to 12.3
+* Python 3.9 to 3.11
 
 Only the NVRTC redistributable component is required from the CUDA Toolkit. [CUDA Toolkit Documentation](https://docs.nvidia.com/cuda/index.html) Installation Guides can be used for guidance. Note that the NVRTC component in the Toolkit can be obtained via PiPy, Conda or Local Installer.
 

cuda/_cuda/ccuda.pxd.in

@@ -392,6 +392,11 @@ cdef CUresult _cuKernelSetAttribute(CUfunction_attribute attrib, int val, CUkern
 cdef CUresult _cuKernelSetCacheConfig(CUkernel kernel, CUfunc_cache config, CUdevice dev) except ?CUDA_ERROR_NOT_FOUND nogil
 {{endif}}
 
+{{if 'cuKernelGetName' in found_functions}}
+
+cdef CUresult _cuKernelGetName(const char** name, CUkernel hfunc) except ?CUDA_ERROR_NOT_FOUND nogil
+{{endif}}
+
 {{if 'cuMemGetInfo_v2' in found_functions}}
 
 cdef CUresult _cuMemGetInfo_v2(size_t* free, size_t* total) except ?CUDA_ERROR_NOT_FOUND nogil
@@ -1002,6 +1007,11 @@ cdef CUresult _cuStreamAddCallback(CUstream hStream, CUstreamCallback callback,
 cdef CUresult _cuStreamBeginCapture_v2(CUstream hStream, CUstreamCaptureMode mode) except ?CUDA_ERROR_NOT_FOUND nogil
 {{endif}}
 
+{{if 'cuStreamBeginCaptureToGraph' in found_functions}}
+
+cdef CUresult _cuStreamBeginCaptureToGraph(CUstream hStream, CUgraph hGraph, const CUgraphNode* dependencies, const CUgraphEdgeData* dependencyData, size_t numDependencies, CUstreamCaptureMode mode) except ?CUDA_ERROR_NOT_FOUND nogil
+{{endif}}
+
 {{if 'cuThreadExchangeStreamCaptureMode' in found_functions}}
 
 cdef CUresult _cuThreadExchangeStreamCaptureMode(CUstreamCaptureMode* mode) except ?CUDA_ERROR_NOT_FOUND nogil
@@ -1022,11 +1032,21 @@ cdef CUresult _cuStreamIsCapturing(CUstream hStream, CUstreamCaptureStatus* capt
 cdef CUresult _cuStreamGetCaptureInfo_v2(CUstream hStream, CUstreamCaptureStatus* captureStatus_out, cuuint64_t* id_out, CUgraph* graph_out, const CUgraphNode** dependencies_out, size_t* numDependencies_out) except ?CUDA_ERROR_NOT_FOUND nogil
 {{endif}}
 
+{{if 'cuStreamGetCaptureInfo_v3' in found_functions}}
+
+cdef CUresult _cuStreamGetCaptureInfo_v3(CUstream hStream, CUstreamCaptureStatus* captureStatus_out, cuuint64_t* id_out, CUgraph* graph_out, const CUgraphNode** dependencies_out, const CUgraphEdgeData** edgeData_out, size_t* numDependencies_out) except ?CUDA_ERROR_NOT_FOUND nogil
+{{endif}}
+
 {{if 'cuStreamUpdateCaptureDependencies' in found_functions}}
 
 cdef CUresult _cuStreamUpdateCaptureDependencies(CUstream hStream, CUgraphNode* dependencies, size_t numDependencies, unsigned int flags) except ?CUDA_ERROR_NOT_FOUND nogil
 {{endif}}
 
+{{if 'cuStreamUpdateCaptureDependencies_v2' in found_functions}}
+
+cdef CUresult _cuStreamUpdateCaptureDependencies_v2(CUstream hStream, CUgraphNode* dependencies, const CUgraphEdgeData* dependencyData, size_t numDependencies, unsigned int flags) except ?CUDA_ERROR_NOT_FOUND nogil
+{{endif}}
+
 {{if 'cuStreamAttachMemAsync' in found_functions}}
 
 cdef CUresult _cuStreamAttachMemAsync(CUstream hStream, CUdeviceptr dptr, size_t length, unsigned int flags) except ?CUDA_ERROR_NOT_FOUND nogil
@@ -1187,6 +1207,11 @@ cdef CUresult _cuFuncSetSharedMemConfig(CUfunction hfunc, CUsharedconfig config)
 cdef CUresult _cuFuncGetModule(CUmodule* hmod, CUfunction hfunc) except ?CUDA_ERROR_NOT_FOUND nogil
 {{endif}}
 
+{{if 'cuFuncGetName' in found_functions}}
+
+cdef CUresult _cuFuncGetName(const char** name, CUfunction hfunc) except ?CUDA_ERROR_NOT_FOUND nogil
+{{endif}}
+
 {{if 'cuLaunchKernel' in found_functions}}
 
 cdef CUresult _cuLaunchKernel(CUfunction f, unsigned int gridDimX, unsigned int gridDimY, unsigned int gridDimZ, unsigned int blockDimX, unsigned int blockDimY, unsigned int blockDimZ, unsigned int sharedMemBytes, CUstream hStream, void** kernelParams, void** extra) except ?CUDA_ERROR_NOT_FOUND nogil
@@ -1487,26 +1512,51 @@ cdef CUresult _cuGraphGetRootNodes(CUgraph hGraph, CUgraphNode* rootNodes, size_
 cdef CUresult _cuGraphGetEdges(CUgraph hGraph, CUgraphNode* from_, CUgraphNode* to, size_t* numEdges) except ?CUDA_ERROR_NOT_FOUND nogil
 {{endif}}
 
+{{if 'cuGraphGetEdges_v2' in found_functions}}
+
+cdef CUresult _cuGraphGetEdges_v2(CUgraph hGraph, CUgraphNode* from_, CUgraphNode* to, CUgraphEdgeData* edgeData, size_t* numEdges) except ?CUDA_ERROR_NOT_FOUND nogil
+{{endif}}
+
 {{if 'cuGraphNodeGetDependencies' in found_functions}}
 
 cdef CUresult _cuGraphNodeGetDependencies(CUgraphNode hNode, CUgraphNode* dependencies, size_t* numDependencies) except ?CUDA_ERROR_NOT_FOUND nogil
 {{endif}}
 
+{{if 'cuGraphNodeGetDependencies_v2' in found_functions}}
+
+cdef CUresult _cuGraphNodeGetDependencies_v2(CUgraphNode hNode, CUgraphNode* dependencies, CUgraphEdgeData* edgeData, size_t* numDependencies) except ?CUDA_ERROR_NOT_FOUND nogil
+{{endif}}
+
 {{if 'cuGraphNodeGetDependentNodes' in found_functions}}
 
 cdef CUresult _cuGraphNodeGetDependentNodes(CUgraphNode hNode, CUgraphNode* dependentNodes, size_t* numDependentNodes) except ?CUDA_ERROR_NOT_FOUND nogil
 {{endif}}
 
+{{if 'cuGraphNodeGetDependentNodes_v2' in found_functions}}
+
+cdef CUresult _cuGraphNodeGetDependentNodes_v2(CUgraphNode hNode, CUgraphNode* dependentNodes, CUgraphEdgeData* edgeData, size_t* numDependentNodes) except ?CUDA_ERROR_NOT_FOUND nogil
+{{endif}}
+
 {{if 'cuGraphAddDependencies' in found_functions}}
 
 cdef CUresult _cuGraphAddDependencies(CUgraph hGraph, const CUgraphNode* from_, const CUgraphNode* to, size_t numDependencies) except ?CUDA_ERROR_NOT_FOUND nogil
 {{endif}}
 
+{{if 'cuGraphAddDependencies_v2' in found_functions}}
+
+cdef CUresult _cuGraphAddDependencies_v2(CUgraph hGraph, const CUgraphNode* from_, const CUgraphNode* to, const CUgraphEdgeData* edgeData, size_t numDependencies) except ?CUDA_ERROR_NOT_FOUND nogil
+{{endif}}
+
 {{if 'cuGraphRemoveDependencies' in found_functions}}
 
 cdef CUresult _cuGraphRemoveDependencies(CUgraph hGraph, const CUgraphNode* from_, const CUgraphNode* to, size_t numDependencies) except ?CUDA_ERROR_NOT_FOUND nogil
 {{endif}}
 
+{{if 'cuGraphRemoveDependencies_v2' in found_functions}}
+
+cdef CUresult _cuGraphRemoveDependencies_v2(CUgraph hGraph, const CUgraphNode* from_, const CUgraphNode* to, const CUgraphEdgeData* edgeData, size_t numDependencies) except ?CUDA_ERROR_NOT_FOUND nogil
+{{endif}}
+
 {{if 'cuGraphDestroyNode' in found_functions}}
 
 cdef CUresult _cuGraphDestroyNode(CUgraphNode hNode) except ?CUDA_ERROR_NOT_FOUND nogil
@@ -1657,6 +1707,11 @@ cdef CUresult _cuGraphReleaseUserObject(CUgraph graph, CUuserObject object, unsi
 cdef CUresult _cuGraphAddNode(CUgraphNode* phGraphNode, CUgraph hGraph, const CUgraphNode* dependencies, size_t numDependencies, CUgraphNodeParams* nodeParams) except ?CUDA_ERROR_NOT_FOUND nogil
 {{endif}}
 
+{{if 'cuGraphAddNode_v2' in found_functions}}
+
+cdef CUresult _cuGraphAddNode_v2(CUgraphNode* phGraphNode, CUgraph hGraph, const CUgraphNode* dependencies, const CUgraphEdgeData* dependencyData, size_t numDependencies, CUgraphNodeParams* nodeParams) except ?CUDA_ERROR_NOT_FOUND nogil
+{{endif}}
+
 {{if 'cuGraphNodeSetParams' in found_functions}}
 
 cdef CUresult _cuGraphNodeSetParams(CUgraphNode hNode, CUgraphNodeParams* nodeParams) except ?CUDA_ERROR_NOT_FOUND nogil
@@ -1667,6 +1722,11 @@ cdef CUresult _cuGraphNodeSetParams(CUgraphNode hNode, CUgraphNodeParams* nodePa
 cdef CUresult _cuGraphExecNodeSetParams(CUgraphExec hGraphExec, CUgraphNode hNode, CUgraphNodeParams* nodeParams) except ?CUDA_ERROR_NOT_FOUND nogil
 {{endif}}
 
+{{if 'cuGraphConditionalHandleCreate' in found_functions}}
+
+cdef CUresult _cuGraphConditionalHandleCreate(CUgraphConditionalHandle* pHandle_out, CUgraph hGraph, CUcontext ctx, unsigned int defaultLaunchValue, unsigned int flags) except ?CUDA_ERROR_NOT_FOUND nogil
+{{endif}}
+
 {{if 'cuOccupancyMaxActiveBlocksPerMultiprocessor' in found_functions}}
 
 cdef CUresult _cuOccupancyMaxActiveBlocksPerMultiprocessor(int* numBlocks, CUfunction func, int blockSize, size_t dynamicSMemSize) except ?CUDA_ERROR_NOT_FOUND nogil