Quick construct of symmetric/SPD IJ matrix

[iaae]

To benchmark Hypre Linear IJ matrix interface further, I made an equivalent symmetric assembly of IJ matrix using the function below:

static HYPRE_IJMatrix* p_matrix;
static HYPER_Int TotalEq;

HYPRESOL_API void hyperij_set_symmetry( )
{
int nrow = 1, nj = 1;
HYPRE_Int* cols = new HYPRE_Int[totalEq];
HYPRE_Int* cols_per_row = new HYPRE_Int[totalEq];
HYPRE_Int* id_rows = new HYPRE_Int[totalEq];
HYPRE_Complex* val = new HYPRE_Complex[totalEq];
HYPRE_Int irow, ncol;

for (int i = 0; i < totalEq; i++) id_rows[i] = i;

HYPRE_IJMatrixGetRowCounts(*p_matrix, totalEq, id_rows, cols_per_row);

for (int i = 1; i < totalEq; i++) {//copy the lower tri component to the upper tri
ncol = cols_per_row[i];
HYPRE_IJMatrixGetValues(*p_matrix, -nrow, &ncol, &id_rows[i], cols, val);
for (int j = 0; j < ncol; j++) {
irow = cols[j];
if (i!=irow)
HYPRE_IJMatrixSetValues(*p_matrix, nrow, &nj, &irow, &i, &val[j]);
}
}

delete[] id_rows;
delete[] cols_per_row;
delete[] cols;
delete[] val;
}

So, after the IJ matrix assembly with only the lower tri components, I make the call

HYPRE_IJMatrixAssemble(*p_matrix);
hyperij_set_symmetry();

It is not working during solving. I did test the whole entry testing and it worked well, even under multi-threaded processes. I tested this hyperij_set_symmetry() on the full entry matrix and it validated the functionality.

Is there re-initialization needed after IJMatrixAssembly()?

thanks!