MatrixDiagonalPartitioned.h

Go to the documentation of this file.

 
#ifndef GOOSEFEM_MATRIXDIAGONALPARTITIONED_H
#define GOOSEFEM_MATRIXDIAGONALPARTITIONED_H
 
#include "MatrixDiagonal.h"
#include "Mesh.h"
#include "config.h"
 
namespace GooseFEM {
 
class MatrixDiagonalPartitioned : public MatrixPartitionedBase<MatrixDiagonalPartitioned>,
                                  public MatrixDiagonalBase<MatrixDiagonalPartitioned> {
private:
    friend MatrixBase<MatrixDiagonalPartitioned>;
    friend MatrixPartitionedBase<MatrixDiagonalPartitioned>;
    friend MatrixDiagonalBase<MatrixDiagonalPartitioned>;
 
public:
    MatrixDiagonalPartitioned() = default;
 
    MatrixDiagonalPartitioned(
        const array_type::tensor<size_t, 2>& conn,
        const array_type::tensor<size_t, 2>& dofs,
        const array_type::tensor<size_t, 1>& iip)
    {
        m_conn = conn;
        m_dofs = dofs;
        m_nelem = m_conn.shape(0);
        m_nne = m_conn.shape(1);
        m_nnode = m_dofs.shape(0);
        m_ndim = m_dofs.shape(1);
        m_ndof = xt::amax(m_dofs)() + 1;
 
        GOOSEFEM_ASSERT(is_unique(iip));
        GOOSEFEM_ASSERT(xt::amax(conn)() + 1 <= m_nnode);
        GOOSEFEM_ASSERT(m_ndof <= m_nnode * m_ndim);
        GOOSEFEM_ASSERT(xt::amax(iip)() <= xt::amax(dofs)());
 
        m_iip = iip;
        m_iiu = xt::setdiff1d(dofs, iip);
        m_nnp = m_iip.size();
        m_nnu = m_iiu.size();
        m_part = Mesh::Reorder({m_iiu, m_iip}).apply(m_dofs);
        m_Auu = xt::empty<double>({m_nnu});
        m_App = xt::empty<double>({m_nnp});
        m_inv_uu = xt::empty<double>({m_nnu});
    }
 
private:
    template <class T>
    void assemble_impl(const T& elemmat)
    {
        GOOSEFEM_ASSERT(xt::has_shape(elemmat, {m_nelem, m_nne * m_ndim, m_nne * m_ndim}));
        GOOSEFEM_ASSERT(Element::isDiagonal(elemmat));
 
        m_Auu.fill(0.0);
        m_App.fill(0.0);
 
        for (size_t e = 0; e < m_nelem; ++e) {
            for (size_t m = 0; m < m_nne; ++m) {
                for (size_t i = 0; i < m_ndim; ++i) {
 
                    size_t d = m_part(m_conn(e, m), i);
 
                    if (d < m_nnu) {
                        m_Auu(d) += elemmat(e, m * m_ndim + i, m * m_ndim + i);
                    }
                    else {
                        m_App(d - m_nnu) += elemmat(e, m * m_ndim + i, m * m_ndim + i);
                    }
                }
            }
        }
 
        m_changed = true;
    }
 
    template <class T>
    void todense_impl(T& ret) const
    {
        ret.fill(0.0);
 
#pragma omp parallel for
        for (size_t d = 0; d < m_nnu; ++d) {
            ret(m_iiu(d), m_iiu(d)) = m_Auu(d);
        }
 
#pragma omp parallel for
        for (size_t d = 0; d < m_nnp; ++d) {
            ret(m_iip(d), m_iip(d)) = m_App(d);
        }
    }
 
public:
    void set(const array_type::tensor<double, 1>& A)
    {
        GOOSEFEM_ASSERT(A.size() == m_ndof);
 
#pragma omp parallel for
        for (size_t d = 0; d < m_nnu; ++d) {
            m_Auu(d) = A(m_iiu(d));
        }
 
#pragma omp parallel for
        for (size_t d = 0; d < m_nnp; ++d) {
            m_App(d) = A(m_iip(d));
        }
 
        m_changed = true;
    }
 
    array_type::tensor<double, 1> data() const
    {
        array_type::tensor<double, 1> ret = xt::zeros<double>({m_ndof});
 
#pragma omp parallel for
        for (size_t d = 0; d < m_nnu; ++d) {
            ret(m_iiu(d)) = m_Auu(d);
        }
 
#pragma omp parallel for
        for (size_t d = 0; d < m_nnp; ++d) {
            ret(m_iip(d)) = m_App(d);
        }
 
        return ret;
    }
 
    const array_type::tensor<double, 1>& data_uu() const
    {
        return m_Auu;
    }
 
    const array_type::tensor<double, 1>& data_pp() const
    {
        return m_App;
    }
 
    [[deprecated]] array_type::tensor<double, 1> Todiagonal() const
    {
        return this->data();
    }
 
private:
    template <class T>
    void dot_nodevec_impl(const T& x, T& b) const
    {
        GOOSEFEM_ASSERT(xt::has_shape(x, {m_nnode, m_ndim}));
        GOOSEFEM_ASSERT(xt::has_shape(b, {m_nnode, m_ndim}));
 
#pragma omp parallel for
        for (size_t m = 0; m < m_nnode; ++m) {
            for (size_t i = 0; i < m_ndim; ++i) {
 
                size_t d = m_part(m, i);
 
                if (d < m_nnu) {
                    b(m, i) = m_Auu(d) * x(m, i);
                }
                else {
                    b(m, i) = m_App(d - m_nnu) * x(m, i);
                }
            }
        }
    }
 
    template <class T>
    void dot_dofval_impl(const T& x, T& b) const
    {
        GOOSEFEM_ASSERT(x.size() == m_ndof);
        GOOSEFEM_ASSERT(b.size() == m_ndof);
 
#pragma omp parallel for
        for (size_t d = 0; d < m_nnu; ++d) {
            b(m_iiu(d)) = m_Auu(d) * x(m_iiu(d));
        }
 
#pragma omp parallel for
        for (size_t d = 0; d < m_nnp; ++d) {
            b(m_iip(d)) = m_App(d) * x(m_iip(d));
        }
    }
 
public:
    array_type::tensor<double, 1>
    Dot_u(const array_type::tensor<double, 1>& x_u, const array_type::tensor<double, 1>& x_p) const
    {
        array_type::tensor<double, 1> b_u = xt::empty<double>({m_nnu});
        this->dot_u(x_u, x_p, b_u);
        return b_u;
    }
 
    void dot_u(
        const array_type::tensor<double, 1>& x_u,
        const array_type::tensor<double, 1>& x_p,
        array_type::tensor<double, 1>& b_u) const
    {
        UNUSED(x_p);
 
        GOOSEFEM_ASSERT(x_u.size() == m_nnu);
        GOOSEFEM_ASSERT(x_p.size() == m_nnp);
        GOOSEFEM_ASSERT(b_u.size() == m_nnu);
 
#pragma omp parallel for
        for (size_t d = 0; d < m_nnu; ++d) {
            b_u(d) = m_Auu(d) * x_u(d);
        }
    }
 
    array_type::tensor<double, 1>
    Dot_p(const array_type::tensor<double, 1>& x_u, const array_type::tensor<double, 1>& x_p) const
    {
        array_type::tensor<double, 1> b_p = xt::empty<double>({m_nnp});
        this->dot_p(x_u, x_p, b_p);
        return b_p;
    }
 
    void dot_p(
        const array_type::tensor<double, 1>& x_u,
        const array_type::tensor<double, 1>& x_p,
        array_type::tensor<double, 1>& b_p) const
    {
        UNUSED(x_u);
 
        GOOSEFEM_ASSERT(x_u.size() == m_nnu);
        GOOSEFEM_ASSERT(x_p.size() == m_nnp);
        GOOSEFEM_ASSERT(b_p.size() == m_nnp);
 
#pragma omp parallel for
        for (size_t d = 0; d < m_nnp; ++d) {
            b_p(d) = m_App(d) * x_p(d);
        }
    }
 
private:
    template <class T>
    void solve_nodevec_impl(const T& b, T& x)
    {
        GOOSEFEM_ASSERT(xt::has_shape(b, {m_nnode, m_ndim}));
        GOOSEFEM_ASSERT(xt::has_shape(x, {m_nnode, m_ndim}));
 
        this->factorize();
 
#pragma omp parallel for
        for (size_t m = 0; m < m_nnode; ++m) {
            for (size_t i = 0; i < m_ndim; ++i) {
                if (m_part(m, i) < m_nnu) {
                    x(m, i) = m_inv_uu(m_part(m, i)) * b(m, i);
                }
            }
        }
    }
 
    template <class T>
    void solve_dofval_impl(const T& b, T& x)
    {
        GOOSEFEM_ASSERT(b.size() == m_ndof);
        GOOSEFEM_ASSERT(x.size() == m_ndof);
 
        this->factorize();
 
#pragma omp parallel for
        for (size_t d = 0; d < m_nnu; ++d) {
            x(m_iiu(d)) = m_inv_uu(d) * b(m_iiu(d));
        }
    }
 
public:
    array_type::tensor<double, 1>
    Solve_u(const array_type::tensor<double, 1>& b_u, const array_type::tensor<double, 1>& x_p)
    {
        array_type::tensor<double, 1> x_u = xt::empty<double>({m_nnu});
        this->solve_u(b_u, x_p, x_u);
        return x_u;
    }
 
    void solve_u(
        const array_type::tensor<double, 1>& b_u,
        const array_type::tensor<double, 1>& x_p,
        array_type::tensor<double, 1>& x_u)
    {
        UNUSED(x_p);
 
        GOOSEFEM_ASSERT(b_u.size() == m_nnu);
        GOOSEFEM_ASSERT(x_p.size() == m_nnp);
        GOOSEFEM_ASSERT(x_u.size() == m_nnu);
 
        this->factorize();
 
#pragma omp parallel for
        for (size_t d = 0; d < m_nnu; ++d) {
            x_u(d) = m_inv_uu(d) * b_u(d);
        }
    }
 
private:
    template <class T>
    void reaction_nodevec_impl(const T& x, T& b) const
    {
        GOOSEFEM_ASSERT(xt::has_shape(x, {m_nnode, m_ndim}));
        GOOSEFEM_ASSERT(xt::has_shape(b, {m_nnode, m_ndim}));
 
#pragma omp parallel for
        for (size_t m = 0; m < m_nnode; ++m) {
            for (size_t i = 0; i < m_ndim; ++i) {
                if (m_part(m, i) >= m_nnu) {
                    b(m, i) = m_App(m_part(m, i) - m_nnu) * x(m, i);
                }
            }
        }
    }
 
    template <class T>
    void reaction_dofval_impl(const T& x, T& b) const
    {
        GOOSEFEM_ASSERT(x.size() == m_ndof);
        GOOSEFEM_ASSERT(b.size() == m_ndof);
 
#pragma omp parallel for
        for (size_t d = 0; d < m_nnp; ++d) {
            b(m_iip(d)) = m_App(d) * x(m_iip(d));
        }
    }
 
    void reaction_p_impl(
        const array_type::tensor<double, 1>& x_u,
        const array_type::tensor<double, 1>& x_p,
        array_type::tensor<double, 1>& b_p) const
    {
        UNUSED(x_u);
 
        GOOSEFEM_ASSERT(x_u.size() == m_nnu);
        GOOSEFEM_ASSERT(x_p.size() == m_nnp);
        GOOSEFEM_ASSERT(b_p.size() == m_nnp);
 
#pragma omp parallel for
        for (size_t d = 0; d < m_nnp; ++d) {
            b_p(d) = m_App(d) * x_p(d);
        }
    }
 
private:
    // The diagonal matrix, and its inverse (re-used to solve different RHS)
    array_type::tensor<double, 1> m_Auu;
    array_type::tensor<double, 1> m_App;
    array_type::tensor<double, 1> m_inv_uu;
 
    // Bookkeeping
    array_type::tensor<size_t, 2> m_part; // DOF-numbers per node, renumbered  [nnode, ndim]
    array_type::tensor<size_t, 1> m_iiu; // DOF-numbers that are unknown      [nnu]
    array_type::tensor<size_t, 1> m_iip; // DOF-numbers that are prescribed   [nnp]
 
    // Dimensions
    size_t m_nnu; // number of unknown DOFs
    size_t m_nnp; // number of prescribed DOFs
 
    // Compute inverse (automatically evaluated by "solve")
    void factorize()
    {
        if (!m_changed) {
            return;
        }
 
#pragma omp parallel for
        for (size_t d = 0; d < m_nnu; ++d) {
            m_inv_uu(d) = 1.0 / m_Auu(d);
        }
 
        m_changed = false;
    }
};
 
} // namespace GooseFEM
 
#endif