// This file is part of Eigen, a lightweight C++ template library
// for linear algebra.
//
// Copyright (C) 2008-2009 Gael Guennebaud <gael.guennebaud@inria.fr>
//
// Eigen is free software; you can redistribute it and/or
// modify it under the terms of the GNU Lesser General Public
// License as published by the Free Software Foundation; either
// version 3 of the License, or (at your option) any later version.
//
// Alternatively, you can redistribute it and/or
// modify it under the terms of the GNU General Public License as
// published by the Free Software Foundation; either version 2 of
// the License, or (at your option) any later version.
//
// Eigen is distributed in the hope that it will be useful, but WITHOUT ANY
// WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS
// FOR A PARTICULAR PURPOSE. See the GNU Lesser General Public License or the
// GNU General Public License for more details.
//
// You should have received a copy of the GNU Lesser General Public
// License and a copy of the GNU General Public License along with
// Eigen. If not, see <http://www.gnu.org/licenses/>.

#ifndef EIGEN_SOLVETRIANGULAR_H
#define EIGEN_SOLVETRIANGULAR_H

namespace internal {

// Forward declarations:
// The following two routines are implemented in the products/TriangularSolver*.h files
template<typename LhsScalar, typename RhsScalar, typename Index, int Side, int Mode, bool Conjugate, int StorageOrder>
struct triangular_solve_vector;

template <typename Scalar, typename Index, int Side, int Mode, bool Conjugate, int TriStorageOrder, int OtherStorageOrder>
struct triangular_solve_matrix;

// small helper struct extracting some traits on the underlying solver operation
template<typename Lhs, typename Rhs, int Side>
class trsolve_traits
{
  private:
    enum {
      RhsIsVectorAtCompileTime = (Side==OnTheLeft ? Rhs::ColsAtCompileTime : Rhs::RowsAtCompileTime)==1
    };
  public:
    enum {
      Unrolling   = (RhsIsVectorAtCompileTime && Rhs::SizeAtCompileTime != Dynamic && Rhs::SizeAtCompileTime <= 8)
                  ? CompleteUnrolling : NoUnrolling,
      RhsVectors  = RhsIsVectorAtCompileTime ? 1 : Dynamic
    };
};

template<typename Lhs, typename Rhs,
  int Side, // can be OnTheLeft/OnTheRight
  int Mode, // can be Upper/Lower | UnitDiag
  int Unrolling = trsolve_traits<Lhs,Rhs,Side>::Unrolling,
  int RhsVectors = trsolve_traits<Lhs,Rhs,Side>::RhsVectors
  >
struct triangular_solver_selector;

template<typename Lhs, typename Rhs, int Side, int Mode>
struct triangular_solver_selector<Lhs,Rhs,Side,Mode,NoUnrolling,1>
{
  typedef typename Lhs::Scalar LhsScalar;
  typedef typename Rhs::Scalar RhsScalar;
  typedef blas_traits<Lhs> LhsProductTraits;
  typedef typename LhsProductTraits::ExtractType ActualLhsType;
  typedef Map<Matrix<RhsScalar,Dynamic,1>, Aligned> MappedRhs;
  static void run(const Lhs& lhs, Rhs& rhs)
  {
    ActualLhsType actualLhs = LhsProductTraits::extract(lhs);

    // FIXME find a way to allow an inner stride if packet_traits<Scalar>::size==1

    bool useRhsDirectly = Rhs::InnerStrideAtCompileTime==1 || rhs.innerStride()==1;
    RhsScalar* actualRhs;
    if(useRhsDirectly)
    {
      actualRhs = &rhs.coeffRef(0);
    }
    else
    {
      actualRhs = ei_aligned_stack_new(RhsScalar,rhs.size());
      MappedRhs(actualRhs,rhs.size()) = rhs;
    }

    triangular_solve_vector<LhsScalar, RhsScalar, typename Lhs::Index, Side, Mode, LhsProductTraits::NeedToConjugate,
                            (int(Lhs::Flags) & RowMajorBit) ? RowMajor : ColMajor>
      ::run(actualLhs.cols(), actualLhs.data(), actualLhs.outerStride(), actualRhs);

    if(!useRhsDirectly)
    {
      rhs = MappedRhs(actualRhs, rhs.size());
      ei_aligned_stack_delete(RhsScalar, actualRhs, rhs.size());
    }
  }
};

// the rhs is a matrix
template<typename Lhs, typename Rhs, int Side, int Mode>
struct triangular_solver_selector<Lhs,Rhs,Side,Mode,NoUnrolling,Dynamic>
{
  typedef typename Rhs::Scalar Scalar;
  typedef typename Rhs::Index Index;
  typedef blas_traits<Lhs> LhsProductTraits;
  typedef typename LhsProductTraits::DirectLinearAccessType ActualLhsType;
  static void run(const Lhs& lhs, Rhs& rhs)
  {
    const ActualLhsType actualLhs = LhsProductTraits::extract(lhs);
    triangular_solve_matrix<Scalar,Index,Side,Mode,LhsProductTraits::NeedToConjugate,(int(Lhs::Flags) & RowMajorBit) ? RowMajor : ColMajor,
                               (Rhs::Flags&RowMajorBit) ? RowMajor : ColMajor>
      ::run(lhs.rows(), Side==OnTheLeft? rhs.cols() : rhs.rows(), &actualLhs.coeffRef(0,0), actualLhs.outerStride(), &rhs.coeffRef(0,0), rhs.outerStride());
  }
};

/***************************************************************************
* meta-unrolling implementation
***************************************************************************/

template<typename Lhs, typename Rhs, int Mode, int Index, int Size,
         bool Stop = Index==Size>
struct triangular_solver_unroller;

template<typename Lhs, typename Rhs, int Mode, int Index, int Size>
struct triangular_solver_unroller<Lhs,Rhs,Mode,Index,Size,false> {
  enum {
    IsLower = ((Mode&Lower)==Lower),
    I = IsLower ? Index : Size - Index - 1,
    S = IsLower ? 0     : I+1
  };
  static void run(const Lhs& lhs, Rhs& rhs)
  {
    if (Index>0)
      rhs.coeffRef(I) -= lhs.row(I).template segment<Index>(S).transpose()
                         .cwiseProduct(rhs.template segment<Index>(S)).sum();

    if(!(Mode & UnitDiag))
      rhs.coeffRef(I) /= lhs.coeff(I,I);

    triangular_solver_unroller<Lhs,Rhs,Mode,Index+1,Size>::run(lhs,rhs);
  }
};

template<typename Lhs, typename Rhs, int Mode, int Index, int Size>
struct triangular_solver_unroller<Lhs,Rhs,Mode,Index,Size,true> {
  static void run(const Lhs&, Rhs&) {}
};

template<typename Lhs, typename Rhs, int Mode>
struct triangular_solver_selector<Lhs,Rhs,OnTheLeft,Mode,CompleteUnrolling,1> {
  static void run(const Lhs& lhs, Rhs& rhs)
  { triangular_solver_unroller<Lhs,Rhs,Mode,0,Rhs::SizeAtCompileTime>::run(lhs,rhs); }
};

template<typename Lhs, typename Rhs, int Mode>
struct triangular_solver_selector<Lhs,Rhs,OnTheRight,Mode,CompleteUnrolling,1> {
  static void run(const Lhs& lhs, Rhs& rhs)
  {
    Transpose<Lhs> trLhs(lhs);
    Transpose<Rhs> trRhs(rhs);
    
    triangular_solver_unroller<Transpose<Lhs>,Transpose<Rhs>,
                              ((Mode&Upper)==Upper ? Lower : Upper) | (Mode&UnitDiag),
                              0,Rhs::SizeAtCompileTime>::run(trLhs,trRhs);
  }
};

} // end namespace internal

/***************************************************************************
* TriangularView methods
***************************************************************************/

/** "in-place" version of TriangularView::solve() where the result is written in \a other
  *
  *
  *
  * \warning The parameter is only marked 'const' to make the C++ compiler accept a temporary expression here.
  * This function will const_cast it, so constness isn't honored here.
  *
  * See TriangularView:solve() for the details.
  */
template<typename MatrixType, unsigned int Mode>
template<int Side, typename OtherDerived>
void TriangularView<MatrixType,Mode>::solveInPlace(const MatrixBase<OtherDerived>& _other) const
{
  OtherDerived& other = _other.const_cast_derived();
  eigen_assert(cols() == rows());
  eigen_assert( (Side==OnTheLeft && cols() == other.rows()) || (Side==OnTheRight && cols() == other.cols()) );
  eigen_assert(!(Mode & ZeroDiag));
  eigen_assert(Mode & (Upper|Lower));

  enum { copy = internal::traits<OtherDerived>::Flags & RowMajorBit  && OtherDerived::IsVectorAtCompileTime };
  typedef typename internal::conditional<copy,
    typename internal::plain_matrix_type_column_major<OtherDerived>::type, OtherDerived&>::type OtherCopy;
  OtherCopy otherCopy(other);

  internal::triangular_solver_selector<MatrixType, typename internal::remove_reference<OtherCopy>::type,
    Side, Mode>::run(nestedExpression(), otherCopy);

  if (copy)
    other = otherCopy;
}

/** \returns the product of the inverse of \c *this with \a other, \a *this being triangular.
  *
  *
  *
  * This function computes the inverse-matrix matrix product inverse(\c *this) * \a other.
  * The matrix \c *this must be triangular and invertible (i.e., all the coefficients of the
  * diagonal must be non zero). It works as a forward (resp. backward) substitution if \c *this
  * is an upper (resp. lower) triangular matrix.
  *
  * It is required that \c *this be marked as either an upper or a lower triangular matrix, which
  * can be done by marked(), and that is automatically the case with expressions such as those returned
  * by extract().
  *
  * Example: \include MatrixBase_marked.cpp
  * Output: \verbinclude MatrixBase_marked.out
  *
  * This function is essentially a wrapper to the faster solveTriangularInPlace() function creating
  * a temporary copy of \a other, calling solveTriangularInPlace() on the copy and returning it.
  * Therefore, if \a other is not needed anymore, it is quite faster to call solveTriangularInPlace()
  * instead of solveTriangular().
  *
  * For users coming from BLAS, this function (and more specifically solveTriangularInPlace()) offer
  * all the operations supported by the \c *TRSV and \c *TRSM BLAS routines.
  *
  * \b Tips: to perform a \em "right-inverse-multiply" you can simply transpose the operation, e.g.:
  * \code
  * M * T^1  <=>  T.transpose().solveInPlace(M.transpose());
  * \endcode
  *
  * \sa TriangularView::solveInPlace()
  */
template<typename Derived, unsigned int Mode>
template<int Side, typename RhsDerived>
typename internal::plain_matrix_type_column_major<RhsDerived>::type
TriangularView<Derived,Mode>::solve(const MatrixBase<RhsDerived>& rhs) const
{
  typename internal::plain_matrix_type_column_major<RhsDerived>::type res(rhs);
  solveInPlace<Side>(res);
  return res;
}

#endif // EIGEN_SOLVETRIANGULAR_H