chDrv.h contains some useful facilities for driver API developers:
the chCUDADevice() class simplifies management of devices and contexts,
and the loadModuleFromFile() method simplifies the creation of a module
from a .cubin or .ptx file.
In addition, the chGetErrorString() function passes back a read-only
string corresponding to an error value. Besides implementing this
function for the driver API’s CUresult type, a specialization of
chGetErrorString() also wraps the CUDA runtime’s cudaGetErrorString()
function.
class chCUDADevice{public: chCUDADevice(); virtual ~chCUDADevice(); CUresult Initialize( int ordinal, list<string>& moduleList, unsigned int Flags = 0, unsigned int numOptions = 0, CUjit_option *options = NULL, void **optionValues = NULL ); CUresult loadModuleFromFile( CUmodule *pModule, string fileName, unsigned int numOptions = 0, CUjit_option *options = NULL, void **optionValues = NULL ); CUdevice device() const { return m_device; } CUcontext context() const { return m_context; } CUmodule module( string s ) const { return (*m_modules.find(s)).second; } private: CUdevice m_device; CUcontext m_context; map<string, CUmodule> m_modules; };chCUDADevice class. (source on GitHub)