paralleling BiCGSTAB using RowMajor sparse matrix still run in siquential

Question

I am using BiCGSTAB to solve a sparse system of equations. I have used MPFR for multiprecision and the following code, but the code is running sequentially. I followed the Make Eigen run in parallel.

using MatrixXmp = Eigen::Matrix<mpfr::mpreal, Eigen::Dynamic, Eigen::Dynamic>;
using VectorXmp = Eigen::Matrix<mpfr::mpreal, Eigen::Dynamic, 1>;
...

omp_set_num_threads(num_threads);
Eigen::setNbThreads(num_threads);
mpfr::mpreal::set_default_prec(128);

Eigen::BiCGSTAB<Eigen::SparseMatrix<mpfr::mpreal, Eigen::RowMajor>> solver;
solver.setTolerance(tol);
solver.setMaxIterations(maxiter);
solver.compute(A);
xx = solver.solve(b);

compile:

g++ -o solver main.cpp -lmpfr -lgmp -fopenmp

I check top for cpu usage. My OS is Ubuntu 20.04. I have seen this post which is almost the same but I still have the sequential running issue.

Edit: I fill matrix A like this:

Eigen::SparseMatrix<mpfr::mpreal, Eigen::RowMajor> A(num_rows, num_rows);
A.reserve(Eigen::RowVectorXi::Constant(num_elements, 3));

for (int i = 0; i < num_elements; i++)
{
    A.insert(rows[i], cols[i]) = data[i];
}
A.makeCompressed();