Added triangular assignement, e.g.:

m.upper() = a+b;
only updates the upper triangular part of m.
Note that:
 m = (a+b).upper();
updates all coefficients of m (but half of the additions
will be skiped)

Updated back/forward substitution to better use Eigen's capability.
This commit is contained in:
Gael Guennebaud 2008-04-26 19:20:26 +00:00
parent 4c92150676
commit b4c974d059
4 changed files with 120 additions and 59 deletions

View File

@ -52,6 +52,7 @@ namespace Eigen {
#include "src/Core/Swap.h"
#include "src/Core/CommaInitializer.h"
#include "src/Core/Triangular.h"
#include "src/Core/TriangularAssign.h"
} // namespace Eigen

View File

@ -103,7 +103,8 @@ bool Vectorize = (Derived::Flags & OtherDerived::Flags & VectorizableBit)
&& ( (Derived::Flags & OtherDerived::Flags & Like1DArrayBit)
||((Derived::Flags&RowMajorBit)
? Derived::ColsAtCompileTime!=Dynamic && (Derived::ColsAtCompileTime%ei_packet_traits<typename Derived::Scalar>::size==0)
: Derived::RowsAtCompileTime!=Dynamic && (Derived::RowsAtCompileTime%ei_packet_traits<typename Derived::Scalar>::size==0)) )>
: Derived::RowsAtCompileTime!=Dynamic && (Derived::RowsAtCompileTime%ei_packet_traits<typename Derived::Scalar>::size==0)) ),
bool TriangularAssign = false>
struct ei_assignment_impl;
template<typename Derived>
@ -112,7 +113,9 @@ Derived& MatrixBase<Derived>
::lazyAssign(const MatrixBase<OtherDerived>& other)
{
// std::cout << "lazyAssign = " << Derived::Flags << " " << OtherDerived::Flags << "\n";
ei_assignment_impl<Derived,OtherDerived>::execute(derived(),other.derived());
ei_assignment_impl<Derived, OtherDerived,
Derived::Flags & (NullLowerBit | NullUpperBit)>
::execute(derived(),other.derived());
return derived();
}

View File

@ -96,49 +96,6 @@ template<int Mode, typename MatrixType> class Triangular
return Triangular<(Upper | Lower) xor Mode, Transpose<MatrixType> >((m_matrix.transpose()));
}
#if 0
template<typename OtherDerived>
Triangular& operator=(const MatrixBase<OtherDerived>& other);
/** Overloaded to provide optimal evaluation loops */
template<typename OtherDerived>
Triangular& operator +=(const MatrixBase<OtherDerived>& other)
{
return *this = m_matrix + other;
}
/** Overloaded to provide optimal evaluation loops */
template<typename OtherDerived>
Triangular& operator *=(const MatrixBase<OtherDerived>& other)
{
return *this = this->lazyProduct(other).eval();
}
/** Optimized triangular matrix - matrix product */
template<typename OtherDerived>
TriangularProduct<Mode, MatrixType, OtherDerived> lazyProduct(const MatrixBase<Scalar, OtherDerived>& other) const
{
return TriangularProduct<Mode,MatrixType,OtherDerived>(m_matrix, other.ref());
}
/** Optimized triangular matrix - matrix product */
template<typename OtherDerived>
Eval<TriangularProduct<Mode, MatrixType, OtherDerived> > operator * (const MatrixBase<Scalar, OtherDerived>& other) const
{
return this->lazyProduct(other).eval();
}
/** Optimized matrix - triangular matrix product */
template<typename OtherDerived>
friend Eval<Transpose<TriangularProduct<0x1 xor Mode, Transpose<MatRef>, Transpose<OtherDerived> > > >
operator * (const MatrixBase<Scalar, OtherDerived>& other, const Triangular<Mode,MatrixType>& tri)
{
return tri.transpose().lazyProduct(other.transpose()).transpose().eval();
}
#endif
/** \returns the product of the inverse of *this with \a other.
*
* This function computes the inverse-matrix matrix product inv(*this) \a other
@ -159,36 +116,32 @@ template<int Mode, typename MatrixType> class Triangular
{
// forward substitution
if (Flags & UnitDiagBit)
res.col(c)[0] = other.col(c)[0];
res(0,c) = other(0,c);
else
res.col(c)[0] = other.col(c)[0]/_coeff(0, 0);
res(0,c) = other(0,c)/_coeff(0, 0);
for (int i=1 ; i<_rows() ; ++i)
{
Scalar tmp = other.col(c)[i];
for (int j = 0 ; j < i ; ++j)
tmp -= _coeff(i,j) * res.col(c)[j];
Scalar tmp = other(i,c) - ((this->row(i).start(i)).transpose() * res.col(c).start(i))(0,0);
if (Flags & UnitDiagBit)
res.col(c)[i] = tmp;
res(i,c) = tmp;
else
res.col(c)[i] = tmp/_coeff(i,i);
res(i,c) = tmp/_coeff(i,i);
}
}
else
{
// backward substitution
if (Flags & UnitDiagBit)
res.col(c)[_cols()-1] = other.col(c)[_cols()-1];
res(_cols()-1,c) = other(_cols()-1,c);
else
res.col(c)[_cols()-1] = other.col(c)[_cols()-1]/_coeff(_rows()-1, _cols()-1);
res(_cols()-1,c) = other(_cols()-1, c)/_coeff(_rows()-1, _cols()-1);
for (int i=_rows()-2 ; i>=0 ; --i)
{
Scalar tmp = other.col(c)[i];
for (int j = i+1 ; j < _cols() ; ++j)
tmp -= _coeff(i,j) * res.col(c)[j];
Scalar tmp = other(i,c) - ((this->row(i).end(_cols()-i-1)).transpose() * res.col(c).end(_cols()-i-1))(0,0);
if (Flags & UnitDiagBit)
res.col(c)[i] = tmp;
res(i,c) = tmp;
else
res.col(c)[i] = tmp/_coeff(i,i);
res(i,c) = tmp/_coeff(i,i);
}
}
}

View File

@ -0,0 +1,104 @@
// This file is part of Eigen, a lightweight C++ template library
// for linear algebra. Eigen itself is part of the KDE project.
//
// Copyright (C) 2008 Gael Guennebaud <g.gael@free.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_TRIANGULAR_ASSIGN_H
#define EIGEN_TRIANGULAR_ASSIGN_H
template<typename Derived1, typename Derived2, int UnrollCount, int Mode>
struct ei_triangular_assign_unroller
{
enum {
col = (UnrollCount-1) / Derived1::RowsAtCompileTime,
row = (UnrollCount-1) % Derived1::RowsAtCompileTime
};
static void run(Derived1 &dst, const Derived2 &src)
{
ei_triangular_assign_unroller<Derived1, Derived2,
(Mode & Lower) ?
((row==col) ? UnrollCount-1-row : UnrollCount-1)
: ((row==0) ? UnrollCount-1-Derived1::ColsAtCompileTime+col : UnrollCount-1),
Mode>::run(dst, src);
dst.coeffRef(row, col) = src.coeff(row, col);
}
};
template<typename Derived1, typename Derived2, int Mode>
struct ei_triangular_assign_unroller<Derived1, Derived2, 1, Mode>
{
static void run(Derived1 &dst, const Derived2 &src)
{
dst.coeffRef(0, 0) = src.coeff(0, 0);
}
};
// prevent buggy user code from causing an infinite recursion
template<typename Derived1, typename Derived2, int Mode>
struct ei_triangular_assign_unroller<Derived1, Derived2, 0, Mode>
{
static void run(Derived1 &, const Derived2 &) {}
};
template<typename Derived1, typename Derived2, int Mode>
struct ei_triangular_assign_unroller<Derived1, Derived2, Dynamic, Mode>
{
static void run(Derived1 &, const Derived2 &) {}
};
template <typename Derived, typename OtherDerived, bool DummyVectorize>
struct ei_assignment_impl<Derived, OtherDerived, DummyVectorize, true>
{
static void execute(Derived & dst, const OtherDerived & src)
{
assert(src.rows()==src.cols());
assert(dst.rows() == src.rows() && dst.cols() == src.cols());
const bool unroll = Derived::SizeAtCompileTime * OtherDerived::CoeffReadCost <= EIGEN_UNROLLING_LIMIT;
if(unroll)
{
ei_triangular_assign_unroller
<Derived, OtherDerived, unroll ? Derived::SizeAtCompileTime : Dynamic, Derived::Flags>::run
(dst.derived(), src.derived());
}
else
{
if (Derived::Flags & Lower)
{
for(int j = 0; j < dst.cols(); j++)
for(int i = j; i < dst.rows(); i++)
dst.coeffRef(i, j) = src.coeff(i, j);
}
else
{
for(int j = 0; j < dst.cols(); j++)
for(int i = 0; i <= j; i++)
dst.coeffRef(i, j) = src.coeff(i, j);
}
}
}
};
#endif // EIGEN_TRIANGULAR_ASSIGN_H