How to pass user-defined struct in C with OpenMP to accelerator?

Question

I have a struct defined, in which I have a dynamically allocated array, and I need to transfer this struct from the host to the accelerator (in my case it would be some nvidia GPU) through some OpenMP directives (in a C-code).

The struct looks as follows:

struct my_grid {
    double **a,  
           **b,
           **c;
};

It is initialized as follows:

initialize_my_grid( struct my_grid *my_gd, 
                    struct grid_config grid_cfg ) {
    my_gd->a = matrix_double( grid_cfg.N1.value, grid_cfg.N2.value );
    my_gd->b = matrix_double( grid_cfg.N1.value, grid_cfg.N2.value );
    my_gd->c = matrix_double( grid_cfg.N1.value, grid_cfg.N2.value ); 
}

The additional struct grid_config only contains some scalars, the relevant part looks as follows:

struct grid_config {
    struct my_variable_size_t
        N1, N2;
}
struct my_variable_size_t {
    size_t value;

The memory allocation function is as follows (omitting the NULL-error check for easier readability):

double **matrix_double( size_t n_rows, size_t n_cols ) {
    size_t ii;
    double *arr;

    // (c)allocate pointers to rows
    arr = calloc( (size_t)n_rows, sizeof(double*) );

    // (c)allocate rows and set pointer to cols (?)
    arr[0] = calloc( (size_t)(n_rows*n_cols), sizeof(double) );
    for ( ii=1 ; ii<n_rows ; ++ii )
        arr[ii] = arr[ii-1] + n_cols;  
}

In the relevant loop which I'd like to run on the accelerator (the GPU), I thought I could do something like

#pragma omp target teams distribute private(jj,kk) 
for ( jj=2 ; jj<grid_cfg.N1.value-2 ; jj+=2 ) {   
    for ( kk=2 ; kk<grid_cfg.N2.value-2 ; kk+=2 ) {
        wf_gd->a[jj  ][kk  ] += wf_gd->b[jj+1][kk  ];  // plus additional calculations
     }
}

But that does not work as I get compiler errors like Unsupported OpenACC construct Deep copy -- wf_gd when compiling with cc.

First of all, I don't understand why the compiler is mentioning OpenACC and then I also like to know if it is possible to use these kind of data structure or if I need to get rid of it for this case?

EDIT: I tried to compile it on a CRAY machine which is important for the compiler message but should not play a role for the general question (how to pass such a construction to the accelerator - if it is possible at all). Unfortunately I can access that machine now and therefore not give further information about the version of the compiler I used - but again, that should not be related to the general question (at least in my understanding).