Check CUDA runtime API result

In CUDA example program, check the result of CUDA runtime API is always a good practice:

#define CUDA_SAFE_CALL(call)  \                                             
do {\                                                                  
    cudaError_t err = call;\                                           
    if (cudaSuccess != err) {\                                         
        fprintf (stderr, "Cuda error in file '%s' in line %i : %s.",\ 
                 __FILE__, __LINE__, cudaGetErrorString(err) );\       
        exit(EXIT_FAILURE);\                                          
    }\                                                                
} while (0)

This can let we know the CUDA wrong state as early as possible.

References:
Is cudaSafeCall no longer needed?.