minimal breaking of nvprof for uvm

break.cu

                                                                                                                        
__global__ void foo(float* array)                                                                                       
{                                                                                                                       
    int i = 0;                                                                                                          
    array[i] = 0.0;                                                                                                     
}                                                                                                                       
                                                                                                                        
#define ARRAY_SIZE 500                                                                                                  
int main (void)                                                                                                         
{                                                                                                                       
    float* array;                                                                                                       
    cudaMallocManaged(&array, ARRAY_SIZE * sizeof(float));                                                              
    foo<<<1,1>>>(array);                                                                                                
    cudaDeviceSynchronize();                                                                                            
    return 0;                                                                                                           
}                                                                                                                       
~