ADD: new track message, Entity class and Position class
This commit is contained in:
Henry Winkel
@@ -0,0 +1,688 @@
|
||||
// This file is part of Eigen, a lightweight C++ template library
|
||||
// for linear algebra.
|
||||
//
|
||||
// Copyright (C) 2008-2011 Gael Guennebaud <gael.guennebaud@inria.fr>
|
||||
// Copyright (C) 2009 Keir Mierle <mierle@gmail.com>
|
||||
// Copyright (C) 2009 Benoit Jacob <jacob.benoit.1@gmail.com>
|
||||
// Copyright (C) 2011 Timothy E. Holy <tim.holy@gmail.com >
|
||||
//
|
||||
// This Source Code Form is subject to the terms of the Mozilla
|
||||
// Public License v. 2.0. If a copy of the MPL was not distributed
|
||||
// with this file, You can obtain one at http://mozilla.org/MPL/2.0/.
|
||||
|
||||
#ifndef EIGEN_LDLT_H
|
||||
#define EIGEN_LDLT_H
|
||||
|
||||
namespace Eigen {
|
||||
|
||||
namespace internal {
|
||||
template<typename _MatrixType, int _UpLo> struct traits<LDLT<_MatrixType, _UpLo> >
|
||||
: traits<_MatrixType>
|
||||
{
|
||||
typedef MatrixXpr XprKind;
|
||||
typedef SolverStorage StorageKind;
|
||||
typedef int StorageIndex;
|
||||
enum { Flags = 0 };
|
||||
};
|
||||
|
||||
template<typename MatrixType, int UpLo> struct LDLT_Traits;
|
||||
|
||||
// PositiveSemiDef means positive semi-definite and non-zero; same for NegativeSemiDef
|
||||
enum SignMatrix { PositiveSemiDef, NegativeSemiDef, ZeroSign, Indefinite };
|
||||
}
|
||||
|
||||
/** \ingroup Cholesky_Module
|
||||
*
|
||||
* \class LDLT
|
||||
*
|
||||
* \brief Robust Cholesky decomposition of a matrix with pivoting
|
||||
*
|
||||
* \tparam _MatrixType the type of the matrix of which to compute the LDL^T Cholesky decomposition
|
||||
* \tparam _UpLo the triangular part that will be used for the decompositon: Lower (default) or Upper.
|
||||
* The other triangular part won't be read.
|
||||
*
|
||||
* Perform a robust Cholesky decomposition of a positive semidefinite or negative semidefinite
|
||||
* matrix \f$ A \f$ such that \f$ A = P^TLDL^*P \f$, where P is a permutation matrix, L
|
||||
* is lower triangular with a unit diagonal and D is a diagonal matrix.
|
||||
*
|
||||
* The decomposition uses pivoting to ensure stability, so that D will have
|
||||
* zeros in the bottom right rank(A) - n submatrix. Avoiding the square root
|
||||
* on D also stabilizes the computation.
|
||||
*
|
||||
* Remember that Cholesky decompositions are not rank-revealing. Also, do not use a Cholesky
|
||||
* decomposition to determine whether a system of equations has a solution.
|
||||
*
|
||||
* This class supports the \link InplaceDecomposition inplace decomposition \endlink mechanism.
|
||||
*
|
||||
* \sa MatrixBase::ldlt(), SelfAdjointView::ldlt(), class LLT
|
||||
*/
|
||||
template<typename _MatrixType, int _UpLo> class LDLT
|
||||
: public SolverBase<LDLT<_MatrixType, _UpLo> >
|
||||
{
|
||||
public:
|
||||
typedef _MatrixType MatrixType;
|
||||
typedef SolverBase<LDLT> Base;
|
||||
friend class SolverBase<LDLT>;
|
||||
|
||||
EIGEN_GENERIC_PUBLIC_INTERFACE(LDLT)
|
||||
enum {
|
||||
MaxRowsAtCompileTime = MatrixType::MaxRowsAtCompileTime,
|
||||
MaxColsAtCompileTime = MatrixType::MaxColsAtCompileTime,
|
||||
UpLo = _UpLo
|
||||
};
|
||||
typedef Matrix<Scalar, RowsAtCompileTime, 1, 0, MaxRowsAtCompileTime, 1> TmpMatrixType;
|
||||
|
||||
typedef Transpositions<RowsAtCompileTime, MaxRowsAtCompileTime> TranspositionType;
|
||||
typedef PermutationMatrix<RowsAtCompileTime, MaxRowsAtCompileTime> PermutationType;
|
||||
|
||||
typedef internal::LDLT_Traits<MatrixType,UpLo> Traits;
|
||||
|
||||
/** \brief Default Constructor.
|
||||
*
|
||||
* The default constructor is useful in cases in which the user intends to
|
||||
* perform decompositions via LDLT::compute(const MatrixType&).
|
||||
*/
|
||||
LDLT()
|
||||
: m_matrix(),
|
||||
m_transpositions(),
|
||||
m_sign(internal::ZeroSign),
|
||||
m_isInitialized(false)
|
||||
{}
|
||||
|
||||
/** \brief Default Constructor with memory preallocation
|
||||
*
|
||||
* Like the default constructor but with preallocation of the internal data
|
||||
* according to the specified problem \a size.
|
||||
* \sa LDLT()
|
||||
*/
|
||||
explicit LDLT(Index size)
|
||||
: m_matrix(size, size),
|
||||
m_transpositions(size),
|
||||
m_temporary(size),
|
||||
m_sign(internal::ZeroSign),
|
||||
m_isInitialized(false)
|
||||
{}
|
||||
|
||||
/** \brief Constructor with decomposition
|
||||
*
|
||||
* This calculates the decomposition for the input \a matrix.
|
||||
*
|
||||
* \sa LDLT(Index size)
|
||||
*/
|
||||
template<typename InputType>
|
||||
explicit LDLT(const EigenBase<InputType>& matrix)
|
||||
: m_matrix(matrix.rows(), matrix.cols()),
|
||||
m_transpositions(matrix.rows()),
|
||||
m_temporary(matrix.rows()),
|
||||
m_sign(internal::ZeroSign),
|
||||
m_isInitialized(false)
|
||||
{
|
||||
compute(matrix.derived());
|
||||
}
|
||||
|
||||
/** \brief Constructs a LDLT factorization from a given matrix
|
||||
*
|
||||
* This overloaded constructor is provided for \link InplaceDecomposition inplace decomposition \endlink when \c MatrixType is a Eigen::Ref.
|
||||
*
|
||||
* \sa LDLT(const EigenBase&)
|
||||
*/
|
||||
template<typename InputType>
|
||||
explicit LDLT(EigenBase<InputType>& matrix)
|
||||
: m_matrix(matrix.derived()),
|
||||
m_transpositions(matrix.rows()),
|
||||
m_temporary(matrix.rows()),
|
||||
m_sign(internal::ZeroSign),
|
||||
m_isInitialized(false)
|
||||
{
|
||||
compute(matrix.derived());
|
||||
}
|
||||
|
||||
/** Clear any existing decomposition
|
||||
* \sa rankUpdate(w,sigma)
|
||||
*/
|
||||
void setZero()
|
||||
{
|
||||
m_isInitialized = false;
|
||||
}
|
||||
|
||||
/** \returns a view of the upper triangular matrix U */
|
||||
inline typename Traits::MatrixU matrixU() const
|
||||
{
|
||||
eigen_assert(m_isInitialized && "LDLT is not initialized.");
|
||||
return Traits::getU(m_matrix);
|
||||
}
|
||||
|
||||
/** \returns a view of the lower triangular matrix L */
|
||||
inline typename Traits::MatrixL matrixL() const
|
||||
{
|
||||
eigen_assert(m_isInitialized && "LDLT is not initialized.");
|
||||
return Traits::getL(m_matrix);
|
||||
}
|
||||
|
||||
/** \returns the permutation matrix P as a transposition sequence.
|
||||
*/
|
||||
inline const TranspositionType& transpositionsP() const
|
||||
{
|
||||
eigen_assert(m_isInitialized && "LDLT is not initialized.");
|
||||
return m_transpositions;
|
||||
}
|
||||
|
||||
/** \returns the coefficients of the diagonal matrix D */
|
||||
inline Diagonal<const MatrixType> vectorD() const
|
||||
{
|
||||
eigen_assert(m_isInitialized && "LDLT is not initialized.");
|
||||
return m_matrix.diagonal();
|
||||
}
|
||||
|
||||
/** \returns true if the matrix is positive (semidefinite) */
|
||||
inline bool isPositive() const
|
||||
{
|
||||
eigen_assert(m_isInitialized && "LDLT is not initialized.");
|
||||
return m_sign == internal::PositiveSemiDef || m_sign == internal::ZeroSign;
|
||||
}
|
||||
|
||||
/** \returns true if the matrix is negative (semidefinite) */
|
||||
inline bool isNegative(void) const
|
||||
{
|
||||
eigen_assert(m_isInitialized && "LDLT is not initialized.");
|
||||
return m_sign == internal::NegativeSemiDef || m_sign == internal::ZeroSign;
|
||||
}
|
||||
|
||||
#ifdef EIGEN_PARSED_BY_DOXYGEN
|
||||
/** \returns a solution x of \f$ A x = b \f$ using the current decomposition of A.
|
||||
*
|
||||
* This function also supports in-place solves using the syntax <tt>x = decompositionObject.solve(x)</tt> .
|
||||
*
|
||||
* \note_about_checking_solutions
|
||||
*
|
||||
* More precisely, this method solves \f$ A x = b \f$ using the decomposition \f$ A = P^T L D L^* P \f$
|
||||
* by solving the systems \f$ P^T y_1 = b \f$, \f$ L y_2 = y_1 \f$, \f$ D y_3 = y_2 \f$,
|
||||
* \f$ L^* y_4 = y_3 \f$ and \f$ P x = y_4 \f$ in succession. If the matrix \f$ A \f$ is singular, then
|
||||
* \f$ D \f$ will also be singular (all the other matrices are invertible). In that case, the
|
||||
* least-square solution of \f$ D y_3 = y_2 \f$ is computed. This does not mean that this function
|
||||
* computes the least-square solution of \f$ A x = b \f$ if \f$ A \f$ is singular.
|
||||
*
|
||||
* \sa MatrixBase::ldlt(), SelfAdjointView::ldlt()
|
||||
*/
|
||||
template<typename Rhs>
|
||||
inline const Solve<LDLT, Rhs>
|
||||
solve(const MatrixBase<Rhs>& b) const;
|
||||
#endif
|
||||
|
||||
template<typename Derived>
|
||||
bool solveInPlace(MatrixBase<Derived> &bAndX) const;
|
||||
|
||||
template<typename InputType>
|
||||
LDLT& compute(const EigenBase<InputType>& matrix);
|
||||
|
||||
/** \returns an estimate of the reciprocal condition number of the matrix of
|
||||
* which \c *this is the LDLT decomposition.
|
||||
*/
|
||||
RealScalar rcond() const
|
||||
{
|
||||
eigen_assert(m_isInitialized && "LDLT is not initialized.");
|
||||
return internal::rcond_estimate_helper(m_l1_norm, *this);
|
||||
}
|
||||
|
||||
template <typename Derived>
|
||||
LDLT& rankUpdate(const MatrixBase<Derived>& w, const RealScalar& alpha=1);
|
||||
|
||||
/** \returns the internal LDLT decomposition matrix
|
||||
*
|
||||
* TODO: document the storage layout
|
||||
*/
|
||||
inline const MatrixType& matrixLDLT() const
|
||||
{
|
||||
eigen_assert(m_isInitialized && "LDLT is not initialized.");
|
||||
return m_matrix;
|
||||
}
|
||||
|
||||
MatrixType reconstructedMatrix() const;
|
||||
|
||||
/** \returns the adjoint of \c *this, that is, a const reference to the decomposition itself as the underlying matrix is self-adjoint.
|
||||
*
|
||||
* This method is provided for compatibility with other matrix decompositions, thus enabling generic code such as:
|
||||
* \code x = decomposition.adjoint().solve(b) \endcode
|
||||
*/
|
||||
const LDLT& adjoint() const { return *this; };
|
||||
|
||||
EIGEN_DEVICE_FUNC inline EIGEN_CONSTEXPR Index rows() const EIGEN_NOEXCEPT { return m_matrix.rows(); }
|
||||
EIGEN_DEVICE_FUNC inline EIGEN_CONSTEXPR Index cols() const EIGEN_NOEXCEPT { return m_matrix.cols(); }
|
||||
|
||||
/** \brief Reports whether previous computation was successful.
|
||||
*
|
||||
* \returns \c Success if computation was successful,
|
||||
* \c NumericalIssue if the factorization failed because of a zero pivot.
|
||||
*/
|
||||
ComputationInfo info() const
|
||||
{
|
||||
eigen_assert(m_isInitialized && "LDLT is not initialized.");
|
||||
return m_info;
|
||||
}
|
||||
|
||||
#ifndef EIGEN_PARSED_BY_DOXYGEN
|
||||
template<typename RhsType, typename DstType>
|
||||
void _solve_impl(const RhsType &rhs, DstType &dst) const;
|
||||
|
||||
template<bool Conjugate, typename RhsType, typename DstType>
|
||||
void _solve_impl_transposed(const RhsType &rhs, DstType &dst) const;
|
||||
#endif
|
||||
|
||||
protected:
|
||||
|
||||
static void check_template_parameters()
|
||||
{
|
||||
EIGEN_STATIC_ASSERT_NON_INTEGER(Scalar);
|
||||
}
|
||||
|
||||
/** \internal
|
||||
* Used to compute and store the Cholesky decomposition A = L D L^* = U^* D U.
|
||||
* The strict upper part is used during the decomposition, the strict lower
|
||||
* part correspond to the coefficients of L (its diagonal is equal to 1 and
|
||||
* is not stored), and the diagonal entries correspond to D.
|
||||
*/
|
||||
MatrixType m_matrix;
|
||||
RealScalar m_l1_norm;
|
||||
TranspositionType m_transpositions;
|
||||
TmpMatrixType m_temporary;
|
||||
internal::SignMatrix m_sign;
|
||||
bool m_isInitialized;
|
||||
ComputationInfo m_info;
|
||||
};
|
||||
|
||||
namespace internal {
|
||||
|
||||
template<int UpLo> struct ldlt_inplace;
|
||||
|
||||
template<> struct ldlt_inplace<Lower>
|
||||
{
|
||||
template<typename MatrixType, typename TranspositionType, typename Workspace>
|
||||
static bool unblocked(MatrixType& mat, TranspositionType& transpositions, Workspace& temp, SignMatrix& sign)
|
||||
{
|
||||
using std::abs;
|
||||
typedef typename MatrixType::Scalar Scalar;
|
||||
typedef typename MatrixType::RealScalar RealScalar;
|
||||
typedef typename TranspositionType::StorageIndex IndexType;
|
||||
eigen_assert(mat.rows()==mat.cols());
|
||||
const Index size = mat.rows();
|
||||
bool found_zero_pivot = false;
|
||||
bool ret = true;
|
||||
|
||||
if (size <= 1)
|
||||
{
|
||||
transpositions.setIdentity();
|
||||
if(size==0) sign = ZeroSign;
|
||||
else if (numext::real(mat.coeff(0,0)) > static_cast<RealScalar>(0) ) sign = PositiveSemiDef;
|
||||
else if (numext::real(mat.coeff(0,0)) < static_cast<RealScalar>(0)) sign = NegativeSemiDef;
|
||||
else sign = ZeroSign;
|
||||
return true;
|
||||
}
|
||||
|
||||
for (Index k = 0; k < size; ++k)
|
||||
{
|
||||
// Find largest diagonal element
|
||||
Index index_of_biggest_in_corner;
|
||||
mat.diagonal().tail(size-k).cwiseAbs().maxCoeff(&index_of_biggest_in_corner);
|
||||
index_of_biggest_in_corner += k;
|
||||
|
||||
transpositions.coeffRef(k) = IndexType(index_of_biggest_in_corner);
|
||||
if(k != index_of_biggest_in_corner)
|
||||
{
|
||||
// apply the transposition while taking care to consider only
|
||||
// the lower triangular part
|
||||
Index s = size-index_of_biggest_in_corner-1; // trailing size after the biggest element
|
||||
mat.row(k).head(k).swap(mat.row(index_of_biggest_in_corner).head(k));
|
||||
mat.col(k).tail(s).swap(mat.col(index_of_biggest_in_corner).tail(s));
|
||||
std::swap(mat.coeffRef(k,k),mat.coeffRef(index_of_biggest_in_corner,index_of_biggest_in_corner));
|
||||
for(Index i=k+1;i<index_of_biggest_in_corner;++i)
|
||||
{
|
||||
Scalar tmp = mat.coeffRef(i,k);
|
||||
mat.coeffRef(i,k) = numext::conj(mat.coeffRef(index_of_biggest_in_corner,i));
|
||||
mat.coeffRef(index_of_biggest_in_corner,i) = numext::conj(tmp);
|
||||
}
|
||||
if(NumTraits<Scalar>::IsComplex)
|
||||
mat.coeffRef(index_of_biggest_in_corner,k) = numext::conj(mat.coeff(index_of_biggest_in_corner,k));
|
||||
}
|
||||
|
||||
// partition the matrix:
|
||||
// A00 | - | -
|
||||
// lu = A10 | A11 | -
|
||||
// A20 | A21 | A22
|
||||
Index rs = size - k - 1;
|
||||
Block<MatrixType,Dynamic,1> A21(mat,k+1,k,rs,1);
|
||||
Block<MatrixType,1,Dynamic> A10(mat,k,0,1,k);
|
||||
Block<MatrixType,Dynamic,Dynamic> A20(mat,k+1,0,rs,k);
|
||||
|
||||
if(k>0)
|
||||
{
|
||||
temp.head(k) = mat.diagonal().real().head(k).asDiagonal() * A10.adjoint();
|
||||
mat.coeffRef(k,k) -= (A10 * temp.head(k)).value();
|
||||
if(rs>0)
|
||||
A21.noalias() -= A20 * temp.head(k);
|
||||
}
|
||||
|
||||
// In some previous versions of Eigen (e.g., 3.2.1), the scaling was omitted if the pivot
|
||||
// was smaller than the cutoff value. However, since LDLT is not rank-revealing
|
||||
// we should only make sure that we do not introduce INF or NaN values.
|
||||
// Remark that LAPACK also uses 0 as the cutoff value.
|
||||
RealScalar realAkk = numext::real(mat.coeffRef(k,k));
|
||||
bool pivot_is_valid = (abs(realAkk) > RealScalar(0));
|
||||
|
||||
if(k==0 && !pivot_is_valid)
|
||||
{
|
||||
// The entire diagonal is zero, there is nothing more to do
|
||||
// except filling the transpositions, and checking whether the matrix is zero.
|
||||
sign = ZeroSign;
|
||||
for(Index j = 0; j<size; ++j)
|
||||
{
|
||||
transpositions.coeffRef(j) = IndexType(j);
|
||||
ret = ret && (mat.col(j).tail(size-j-1).array()==Scalar(0)).all();
|
||||
}
|
||||
return ret;
|
||||
}
|
||||
|
||||
if((rs>0) && pivot_is_valid)
|
||||
A21 /= realAkk;
|
||||
else if(rs>0)
|
||||
ret = ret && (A21.array()==Scalar(0)).all();
|
||||
|
||||
if(found_zero_pivot && pivot_is_valid) ret = false; // factorization failed
|
||||
else if(!pivot_is_valid) found_zero_pivot = true;
|
||||
|
||||
if (sign == PositiveSemiDef) {
|
||||
if (realAkk < static_cast<RealScalar>(0)) sign = Indefinite;
|
||||
} else if (sign == NegativeSemiDef) {
|
||||
if (realAkk > static_cast<RealScalar>(0)) sign = Indefinite;
|
||||
} else if (sign == ZeroSign) {
|
||||
if (realAkk > static_cast<RealScalar>(0)) sign = PositiveSemiDef;
|
||||
else if (realAkk < static_cast<RealScalar>(0)) sign = NegativeSemiDef;
|
||||
}
|
||||
}
|
||||
|
||||
return ret;
|
||||
}
|
||||
|
||||
// Reference for the algorithm: Davis and Hager, "Multiple Rank
|
||||
// Modifications of a Sparse Cholesky Factorization" (Algorithm 1)
|
||||
// Trivial rearrangements of their computations (Timothy E. Holy)
|
||||
// allow their algorithm to work for rank-1 updates even if the
|
||||
// original matrix is not of full rank.
|
||||
// Here only rank-1 updates are implemented, to reduce the
|
||||
// requirement for intermediate storage and improve accuracy
|
||||
template<typename MatrixType, typename WDerived>
|
||||
static bool updateInPlace(MatrixType& mat, MatrixBase<WDerived>& w, const typename MatrixType::RealScalar& sigma=1)
|
||||
{
|
||||
using numext::isfinite;
|
||||
typedef typename MatrixType::Scalar Scalar;
|
||||
typedef typename MatrixType::RealScalar RealScalar;
|
||||
|
||||
const Index size = mat.rows();
|
||||
eigen_assert(mat.cols() == size && w.size()==size);
|
||||
|
||||
RealScalar alpha = 1;
|
||||
|
||||
// Apply the update
|
||||
for (Index j = 0; j < size; j++)
|
||||
{
|
||||
// Check for termination due to an original decomposition of low-rank
|
||||
if (!(isfinite)(alpha))
|
||||
break;
|
||||
|
||||
// Update the diagonal terms
|
||||
RealScalar dj = numext::real(mat.coeff(j,j));
|
||||
Scalar wj = w.coeff(j);
|
||||
RealScalar swj2 = sigma*numext::abs2(wj);
|
||||
RealScalar gamma = dj*alpha + swj2;
|
||||
|
||||
mat.coeffRef(j,j) += swj2/alpha;
|
||||
alpha += swj2/dj;
|
||||
|
||||
|
||||
// Update the terms of L
|
||||
Index rs = size-j-1;
|
||||
w.tail(rs) -= wj * mat.col(j).tail(rs);
|
||||
if(gamma != 0)
|
||||
mat.col(j).tail(rs) += (sigma*numext::conj(wj)/gamma)*w.tail(rs);
|
||||
}
|
||||
return true;
|
||||
}
|
||||
|
||||
template<typename MatrixType, typename TranspositionType, typename Workspace, typename WType>
|
||||
static bool update(MatrixType& mat, const TranspositionType& transpositions, Workspace& tmp, const WType& w, const typename MatrixType::RealScalar& sigma=1)
|
||||
{
|
||||
// Apply the permutation to the input w
|
||||
tmp = transpositions * w;
|
||||
|
||||
return ldlt_inplace<Lower>::updateInPlace(mat,tmp,sigma);
|
||||
}
|
||||
};
|
||||
|
||||
template<> struct ldlt_inplace<Upper>
|
||||
{
|
||||
template<typename MatrixType, typename TranspositionType, typename Workspace>
|
||||
static EIGEN_STRONG_INLINE bool unblocked(MatrixType& mat, TranspositionType& transpositions, Workspace& temp, SignMatrix& sign)
|
||||
{
|
||||
Transpose<MatrixType> matt(mat);
|
||||
return ldlt_inplace<Lower>::unblocked(matt, transpositions, temp, sign);
|
||||
}
|
||||
|
||||
template<typename MatrixType, typename TranspositionType, typename Workspace, typename WType>
|
||||
static EIGEN_STRONG_INLINE bool update(MatrixType& mat, TranspositionType& transpositions, Workspace& tmp, WType& w, const typename MatrixType::RealScalar& sigma=1)
|
||||
{
|
||||
Transpose<MatrixType> matt(mat);
|
||||
return ldlt_inplace<Lower>::update(matt, transpositions, tmp, w.conjugate(), sigma);
|
||||
}
|
||||
};
|
||||
|
||||
template<typename MatrixType> struct LDLT_Traits<MatrixType,Lower>
|
||||
{
|
||||
typedef const TriangularView<const MatrixType, UnitLower> MatrixL;
|
||||
typedef const TriangularView<const typename MatrixType::AdjointReturnType, UnitUpper> MatrixU;
|
||||
static inline MatrixL getL(const MatrixType& m) { return MatrixL(m); }
|
||||
static inline MatrixU getU(const MatrixType& m) { return MatrixU(m.adjoint()); }
|
||||
};
|
||||
|
||||
template<typename MatrixType> struct LDLT_Traits<MatrixType,Upper>
|
||||
{
|
||||
typedef const TriangularView<const typename MatrixType::AdjointReturnType, UnitLower> MatrixL;
|
||||
typedef const TriangularView<const MatrixType, UnitUpper> MatrixU;
|
||||
static inline MatrixL getL(const MatrixType& m) { return MatrixL(m.adjoint()); }
|
||||
static inline MatrixU getU(const MatrixType& m) { return MatrixU(m); }
|
||||
};
|
||||
|
||||
} // end namespace internal
|
||||
|
||||
/** Compute / recompute the LDLT decomposition A = L D L^* = U^* D U of \a matrix
|
||||
*/
|
||||
template<typename MatrixType, int _UpLo>
|
||||
template<typename InputType>
|
||||
LDLT<MatrixType,_UpLo>& LDLT<MatrixType,_UpLo>::compute(const EigenBase<InputType>& a)
|
||||
{
|
||||
check_template_parameters();
|
||||
|
||||
eigen_assert(a.rows()==a.cols());
|
||||
const Index size = a.rows();
|
||||
|
||||
m_matrix = a.derived();
|
||||
|
||||
// Compute matrix L1 norm = max abs column sum.
|
||||
m_l1_norm = RealScalar(0);
|
||||
// TODO move this code to SelfAdjointView
|
||||
for (Index col = 0; col < size; ++col) {
|
||||
RealScalar abs_col_sum;
|
||||
if (_UpLo == Lower)
|
||||
abs_col_sum = m_matrix.col(col).tail(size - col).template lpNorm<1>() + m_matrix.row(col).head(col).template lpNorm<1>();
|
||||
else
|
||||
abs_col_sum = m_matrix.col(col).head(col).template lpNorm<1>() + m_matrix.row(col).tail(size - col).template lpNorm<1>();
|
||||
if (abs_col_sum > m_l1_norm)
|
||||
m_l1_norm = abs_col_sum;
|
||||
}
|
||||
|
||||
m_transpositions.resize(size);
|
||||
m_isInitialized = false;
|
||||
m_temporary.resize(size);
|
||||
m_sign = internal::ZeroSign;
|
||||
|
||||
m_info = internal::ldlt_inplace<UpLo>::unblocked(m_matrix, m_transpositions, m_temporary, m_sign) ? Success : NumericalIssue;
|
||||
|
||||
m_isInitialized = true;
|
||||
return *this;
|
||||
}
|
||||
|
||||
/** Update the LDLT decomposition: given A = L D L^T, efficiently compute the decomposition of A + sigma w w^T.
|
||||
* \param w a vector to be incorporated into the decomposition.
|
||||
* \param sigma a scalar, +1 for updates and -1 for "downdates," which correspond to removing previously-added column vectors. Optional; default value is +1.
|
||||
* \sa setZero()
|
||||
*/
|
||||
template<typename MatrixType, int _UpLo>
|
||||
template<typename Derived>
|
||||
LDLT<MatrixType,_UpLo>& LDLT<MatrixType,_UpLo>::rankUpdate(const MatrixBase<Derived>& w, const typename LDLT<MatrixType,_UpLo>::RealScalar& sigma)
|
||||
{
|
||||
typedef typename TranspositionType::StorageIndex IndexType;
|
||||
const Index size = w.rows();
|
||||
if (m_isInitialized)
|
||||
{
|
||||
eigen_assert(m_matrix.rows()==size);
|
||||
}
|
||||
else
|
||||
{
|
||||
m_matrix.resize(size,size);
|
||||
m_matrix.setZero();
|
||||
m_transpositions.resize(size);
|
||||
for (Index i = 0; i < size; i++)
|
||||
m_transpositions.coeffRef(i) = IndexType(i);
|
||||
m_temporary.resize(size);
|
||||
m_sign = sigma>=0 ? internal::PositiveSemiDef : internal::NegativeSemiDef;
|
||||
m_isInitialized = true;
|
||||
}
|
||||
|
||||
internal::ldlt_inplace<UpLo>::update(m_matrix, m_transpositions, m_temporary, w, sigma);
|
||||
|
||||
return *this;
|
||||
}
|
||||
|
||||
#ifndef EIGEN_PARSED_BY_DOXYGEN
|
||||
template<typename _MatrixType, int _UpLo>
|
||||
template<typename RhsType, typename DstType>
|
||||
void LDLT<_MatrixType,_UpLo>::_solve_impl(const RhsType &rhs, DstType &dst) const
|
||||
{
|
||||
_solve_impl_transposed<true>(rhs, dst);
|
||||
}
|
||||
|
||||
template<typename _MatrixType,int _UpLo>
|
||||
template<bool Conjugate, typename RhsType, typename DstType>
|
||||
void LDLT<_MatrixType,_UpLo>::_solve_impl_transposed(const RhsType &rhs, DstType &dst) const
|
||||
{
|
||||
// dst = P b
|
||||
dst = m_transpositions * rhs;
|
||||
|
||||
// dst = L^-1 (P b)
|
||||
// dst = L^-*T (P b)
|
||||
matrixL().template conjugateIf<!Conjugate>().solveInPlace(dst);
|
||||
|
||||
// dst = D^-* (L^-1 P b)
|
||||
// dst = D^-1 (L^-*T P b)
|
||||
// more precisely, use pseudo-inverse of D (see bug 241)
|
||||
using std::abs;
|
||||
const typename Diagonal<const MatrixType>::RealReturnType vecD(vectorD());
|
||||
// In some previous versions, tolerance was set to the max of 1/highest (or rather numeric_limits::min())
|
||||
// and the maximal diagonal entry * epsilon as motivated by LAPACK's xGELSS:
|
||||
// RealScalar tolerance = numext::maxi(vecD.array().abs().maxCoeff() * NumTraits<RealScalar>::epsilon(),RealScalar(1) / NumTraits<RealScalar>::highest());
|
||||
// However, LDLT is not rank revealing, and so adjusting the tolerance wrt to the highest
|
||||
// diagonal element is not well justified and leads to numerical issues in some cases.
|
||||
// Moreover, Lapack's xSYTRS routines use 0 for the tolerance.
|
||||
// Using numeric_limits::min() gives us more robustness to denormals.
|
||||
RealScalar tolerance = (std::numeric_limits<RealScalar>::min)();
|
||||
for (Index i = 0; i < vecD.size(); ++i)
|
||||
{
|
||||
if(abs(vecD(i)) > tolerance)
|
||||
dst.row(i) /= vecD(i);
|
||||
else
|
||||
dst.row(i).setZero();
|
||||
}
|
||||
|
||||
// dst = L^-* (D^-* L^-1 P b)
|
||||
// dst = L^-T (D^-1 L^-*T P b)
|
||||
matrixL().transpose().template conjugateIf<Conjugate>().solveInPlace(dst);
|
||||
|
||||
// dst = P^T (L^-* D^-* L^-1 P b) = A^-1 b
|
||||
// dst = P^-T (L^-T D^-1 L^-*T P b) = A^-1 b
|
||||
dst = m_transpositions.transpose() * dst;
|
||||
}
|
||||
#endif
|
||||
|
||||
/** \internal use x = ldlt_object.solve(x);
|
||||
*
|
||||
* This is the \em in-place version of solve().
|
||||
*
|
||||
* \param bAndX represents both the right-hand side matrix b and result x.
|
||||
*
|
||||
* \returns true always! If you need to check for existence of solutions, use another decomposition like LU, QR, or SVD.
|
||||
*
|
||||
* This version avoids a copy when the right hand side matrix b is not
|
||||
* needed anymore.
|
||||
*
|
||||
* \sa LDLT::solve(), MatrixBase::ldlt()
|
||||
*/
|
||||
template<typename MatrixType,int _UpLo>
|
||||
template<typename Derived>
|
||||
bool LDLT<MatrixType,_UpLo>::solveInPlace(MatrixBase<Derived> &bAndX) const
|
||||
{
|
||||
eigen_assert(m_isInitialized && "LDLT is not initialized.");
|
||||
eigen_assert(m_matrix.rows() == bAndX.rows());
|
||||
|
||||
bAndX = this->solve(bAndX);
|
||||
|
||||
return true;
|
||||
}
|
||||
|
||||
/** \returns the matrix represented by the decomposition,
|
||||
* i.e., it returns the product: P^T L D L^* P.
|
||||
* This function is provided for debug purpose. */
|
||||
template<typename MatrixType, int _UpLo>
|
||||
MatrixType LDLT<MatrixType,_UpLo>::reconstructedMatrix() const
|
||||
{
|
||||
eigen_assert(m_isInitialized && "LDLT is not initialized.");
|
||||
const Index size = m_matrix.rows();
|
||||
MatrixType res(size,size);
|
||||
|
||||
// P
|
||||
res.setIdentity();
|
||||
res = transpositionsP() * res;
|
||||
// L^* P
|
||||
res = matrixU() * res;
|
||||
// D(L^*P)
|
||||
res = vectorD().real().asDiagonal() * res;
|
||||
// L(DL^*P)
|
||||
res = matrixL() * res;
|
||||
// P^T (LDL^*P)
|
||||
res = transpositionsP().transpose() * res;
|
||||
|
||||
return res;
|
||||
}
|
||||
|
||||
/** \cholesky_module
|
||||
* \returns the Cholesky decomposition with full pivoting without square root of \c *this
|
||||
* \sa MatrixBase::ldlt()
|
||||
*/
|
||||
template<typename MatrixType, unsigned int UpLo>
|
||||
inline const LDLT<typename SelfAdjointView<MatrixType, UpLo>::PlainObject, UpLo>
|
||||
SelfAdjointView<MatrixType, UpLo>::ldlt() const
|
||||
{
|
||||
return LDLT<PlainObject,UpLo>(m_matrix);
|
||||
}
|
||||
|
||||
/** \cholesky_module
|
||||
* \returns the Cholesky decomposition with full pivoting without square root of \c *this
|
||||
* \sa SelfAdjointView::ldlt()
|
||||
*/
|
||||
template<typename Derived>
|
||||
inline const LDLT<typename MatrixBase<Derived>::PlainObject>
|
||||
MatrixBase<Derived>::ldlt() const
|
||||
{
|
||||
return LDLT<PlainObject>(derived());
|
||||
}
|
||||
|
||||
} // end namespace Eigen
|
||||
|
||||
#endif // EIGEN_LDLT_H
|
||||
@@ -0,0 +1,558 @@
|
||||
// This file is part of Eigen, a lightweight C++ template library
|
||||
// for linear algebra.
|
||||
//
|
||||
// Copyright (C) 2008 Gael Guennebaud <gael.guennebaud@inria.fr>
|
||||
//
|
||||
// This Source Code Form is subject to the terms of the Mozilla
|
||||
// Public License v. 2.0. If a copy of the MPL was not distributed
|
||||
// with this file, You can obtain one at http://mozilla.org/MPL/2.0/.
|
||||
|
||||
#ifndef EIGEN_LLT_H
|
||||
#define EIGEN_LLT_H
|
||||
|
||||
namespace Eigen {
|
||||
|
||||
namespace internal{
|
||||
|
||||
template<typename _MatrixType, int _UpLo> struct traits<LLT<_MatrixType, _UpLo> >
|
||||
: traits<_MatrixType>
|
||||
{
|
||||
typedef MatrixXpr XprKind;
|
||||
typedef SolverStorage StorageKind;
|
||||
typedef int StorageIndex;
|
||||
enum { Flags = 0 };
|
||||
};
|
||||
|
||||
template<typename MatrixType, int UpLo> struct LLT_Traits;
|
||||
}
|
||||
|
||||
/** \ingroup Cholesky_Module
|
||||
*
|
||||
* \class LLT
|
||||
*
|
||||
* \brief Standard Cholesky decomposition (LL^T) of a matrix and associated features
|
||||
*
|
||||
* \tparam _MatrixType the type of the matrix of which we are computing the LL^T Cholesky decomposition
|
||||
* \tparam _UpLo the triangular part that will be used for the decompositon: Lower (default) or Upper.
|
||||
* The other triangular part won't be read.
|
||||
*
|
||||
* This class performs a LL^T Cholesky decomposition of a symmetric, positive definite
|
||||
* matrix A such that A = LL^* = U^*U, where L is lower triangular.
|
||||
*
|
||||
* While the Cholesky decomposition is particularly useful to solve selfadjoint problems like D^*D x = b,
|
||||
* for that purpose, we recommend the Cholesky decomposition without square root which is more stable
|
||||
* and even faster. Nevertheless, this standard Cholesky decomposition remains useful in many other
|
||||
* situations like generalised eigen problems with hermitian matrices.
|
||||
*
|
||||
* Remember that Cholesky decompositions are not rank-revealing. This LLT decomposition is only stable on positive definite matrices,
|
||||
* use LDLT instead for the semidefinite case. Also, do not use a Cholesky decomposition to determine whether a system of equations
|
||||
* has a solution.
|
||||
*
|
||||
* Example: \include LLT_example.cpp
|
||||
* Output: \verbinclude LLT_example.out
|
||||
*
|
||||
* \b Performance: for best performance, it is recommended to use a column-major storage format
|
||||
* with the Lower triangular part (the default), or, equivalently, a row-major storage format
|
||||
* with the Upper triangular part. Otherwise, you might get a 20% slowdown for the full factorization
|
||||
* step, and rank-updates can be up to 3 times slower.
|
||||
*
|
||||
* This class supports the \link InplaceDecomposition inplace decomposition \endlink mechanism.
|
||||
*
|
||||
* Note that during the decomposition, only the lower (or upper, as defined by _UpLo) triangular part of A is considered.
|
||||
* Therefore, the strict lower part does not have to store correct values.
|
||||
*
|
||||
* \sa MatrixBase::llt(), SelfAdjointView::llt(), class LDLT
|
||||
*/
|
||||
template<typename _MatrixType, int _UpLo> class LLT
|
||||
: public SolverBase<LLT<_MatrixType, _UpLo> >
|
||||
{
|
||||
public:
|
||||
typedef _MatrixType MatrixType;
|
||||
typedef SolverBase<LLT> Base;
|
||||
friend class SolverBase<LLT>;
|
||||
|
||||
EIGEN_GENERIC_PUBLIC_INTERFACE(LLT)
|
||||
enum {
|
||||
MaxColsAtCompileTime = MatrixType::MaxColsAtCompileTime
|
||||
};
|
||||
|
||||
enum {
|
||||
PacketSize = internal::packet_traits<Scalar>::size,
|
||||
AlignmentMask = int(PacketSize)-1,
|
||||
UpLo = _UpLo
|
||||
};
|
||||
|
||||
typedef internal::LLT_Traits<MatrixType,UpLo> Traits;
|
||||
|
||||
/**
|
||||
* \brief Default Constructor.
|
||||
*
|
||||
* The default constructor is useful in cases in which the user intends to
|
||||
* perform decompositions via LLT::compute(const MatrixType&).
|
||||
*/
|
||||
LLT() : m_matrix(), m_isInitialized(false) {}
|
||||
|
||||
/** \brief Default Constructor with memory preallocation
|
||||
*
|
||||
* Like the default constructor but with preallocation of the internal data
|
||||
* according to the specified problem \a size.
|
||||
* \sa LLT()
|
||||
*/
|
||||
explicit LLT(Index size) : m_matrix(size, size),
|
||||
m_isInitialized(false) {}
|
||||
|
||||
template<typename InputType>
|
||||
explicit LLT(const EigenBase<InputType>& matrix)
|
||||
: m_matrix(matrix.rows(), matrix.cols()),
|
||||
m_isInitialized(false)
|
||||
{
|
||||
compute(matrix.derived());
|
||||
}
|
||||
|
||||
/** \brief Constructs a LLT factorization from a given matrix
|
||||
*
|
||||
* This overloaded constructor is provided for \link InplaceDecomposition inplace decomposition \endlink when
|
||||
* \c MatrixType is a Eigen::Ref.
|
||||
*
|
||||
* \sa LLT(const EigenBase&)
|
||||
*/
|
||||
template<typename InputType>
|
||||
explicit LLT(EigenBase<InputType>& matrix)
|
||||
: m_matrix(matrix.derived()),
|
||||
m_isInitialized(false)
|
||||
{
|
||||
compute(matrix.derived());
|
||||
}
|
||||
|
||||
/** \returns a view of the upper triangular matrix U */
|
||||
inline typename Traits::MatrixU matrixU() const
|
||||
{
|
||||
eigen_assert(m_isInitialized && "LLT is not initialized.");
|
||||
return Traits::getU(m_matrix);
|
||||
}
|
||||
|
||||
/** \returns a view of the lower triangular matrix L */
|
||||
inline typename Traits::MatrixL matrixL() const
|
||||
{
|
||||
eigen_assert(m_isInitialized && "LLT is not initialized.");
|
||||
return Traits::getL(m_matrix);
|
||||
}
|
||||
|
||||
#ifdef EIGEN_PARSED_BY_DOXYGEN
|
||||
/** \returns the solution x of \f$ A x = b \f$ using the current decomposition of A.
|
||||
*
|
||||
* Since this LLT class assumes anyway that the matrix A is invertible, the solution
|
||||
* theoretically exists and is unique regardless of b.
|
||||
*
|
||||
* Example: \include LLT_solve.cpp
|
||||
* Output: \verbinclude LLT_solve.out
|
||||
*
|
||||
* \sa solveInPlace(), MatrixBase::llt(), SelfAdjointView::llt()
|
||||
*/
|
||||
template<typename Rhs>
|
||||
inline const Solve<LLT, Rhs>
|
||||
solve(const MatrixBase<Rhs>& b) const;
|
||||
#endif
|
||||
|
||||
template<typename Derived>
|
||||
void solveInPlace(const MatrixBase<Derived> &bAndX) const;
|
||||
|
||||
template<typename InputType>
|
||||
LLT& compute(const EigenBase<InputType>& matrix);
|
||||
|
||||
/** \returns an estimate of the reciprocal condition number of the matrix of
|
||||
* which \c *this is the Cholesky decomposition.
|
||||
*/
|
||||
RealScalar rcond() const
|
||||
{
|
||||
eigen_assert(m_isInitialized && "LLT is not initialized.");
|
||||
eigen_assert(m_info == Success && "LLT failed because matrix appears to be negative");
|
||||
return internal::rcond_estimate_helper(m_l1_norm, *this);
|
||||
}
|
||||
|
||||
/** \returns the LLT decomposition matrix
|
||||
*
|
||||
* TODO: document the storage layout
|
||||
*/
|
||||
inline const MatrixType& matrixLLT() const
|
||||
{
|
||||
eigen_assert(m_isInitialized && "LLT is not initialized.");
|
||||
return m_matrix;
|
||||
}
|
||||
|
||||
MatrixType reconstructedMatrix() const;
|
||||
|
||||
|
||||
/** \brief Reports whether previous computation was successful.
|
||||
*
|
||||
* \returns \c Success if computation was successful,
|
||||
* \c NumericalIssue if the matrix.appears not to be positive definite.
|
||||
*/
|
||||
ComputationInfo info() const
|
||||
{
|
||||
eigen_assert(m_isInitialized && "LLT is not initialized.");
|
||||
return m_info;
|
||||
}
|
||||
|
||||
/** \returns the adjoint of \c *this, that is, a const reference to the decomposition itself as the underlying matrix is self-adjoint.
|
||||
*
|
||||
* This method is provided for compatibility with other matrix decompositions, thus enabling generic code such as:
|
||||
* \code x = decomposition.adjoint().solve(b) \endcode
|
||||
*/
|
||||
const LLT& adjoint() const EIGEN_NOEXCEPT { return *this; };
|
||||
|
||||
inline EIGEN_CONSTEXPR Index rows() const EIGEN_NOEXCEPT { return m_matrix.rows(); }
|
||||
inline EIGEN_CONSTEXPR Index cols() const EIGEN_NOEXCEPT { return m_matrix.cols(); }
|
||||
|
||||
template<typename VectorType>
|
||||
LLT & rankUpdate(const VectorType& vec, const RealScalar& sigma = 1);
|
||||
|
||||
#ifndef EIGEN_PARSED_BY_DOXYGEN
|
||||
template<typename RhsType, typename DstType>
|
||||
void _solve_impl(const RhsType &rhs, DstType &dst) const;
|
||||
|
||||
template<bool Conjugate, typename RhsType, typename DstType>
|
||||
void _solve_impl_transposed(const RhsType &rhs, DstType &dst) const;
|
||||
#endif
|
||||
|
||||
protected:
|
||||
|
||||
static void check_template_parameters()
|
||||
{
|
||||
EIGEN_STATIC_ASSERT_NON_INTEGER(Scalar);
|
||||
}
|
||||
|
||||
/** \internal
|
||||
* Used to compute and store L
|
||||
* The strict upper part is not used and even not initialized.
|
||||
*/
|
||||
MatrixType m_matrix;
|
||||
RealScalar m_l1_norm;
|
||||
bool m_isInitialized;
|
||||
ComputationInfo m_info;
|
||||
};
|
||||
|
||||
namespace internal {
|
||||
|
||||
template<typename Scalar, int UpLo> struct llt_inplace;
|
||||
|
||||
template<typename MatrixType, typename VectorType>
|
||||
static Index llt_rank_update_lower(MatrixType& mat, const VectorType& vec, const typename MatrixType::RealScalar& sigma)
|
||||
{
|
||||
using std::sqrt;
|
||||
typedef typename MatrixType::Scalar Scalar;
|
||||
typedef typename MatrixType::RealScalar RealScalar;
|
||||
typedef typename MatrixType::ColXpr ColXpr;
|
||||
typedef typename internal::remove_all<ColXpr>::type ColXprCleaned;
|
||||
typedef typename ColXprCleaned::SegmentReturnType ColXprSegment;
|
||||
typedef Matrix<Scalar,Dynamic,1> TempVectorType;
|
||||
typedef typename TempVectorType::SegmentReturnType TempVecSegment;
|
||||
|
||||
Index n = mat.cols();
|
||||
eigen_assert(mat.rows()==n && vec.size()==n);
|
||||
|
||||
TempVectorType temp;
|
||||
|
||||
if(sigma>0)
|
||||
{
|
||||
// This version is based on Givens rotations.
|
||||
// It is faster than the other one below, but only works for updates,
|
||||
// i.e., for sigma > 0
|
||||
temp = sqrt(sigma) * vec;
|
||||
|
||||
for(Index i=0; i<n; ++i)
|
||||
{
|
||||
JacobiRotation<Scalar> g;
|
||||
g.makeGivens(mat(i,i), -temp(i), &mat(i,i));
|
||||
|
||||
Index rs = n-i-1;
|
||||
if(rs>0)
|
||||
{
|
||||
ColXprSegment x(mat.col(i).tail(rs));
|
||||
TempVecSegment y(temp.tail(rs));
|
||||
apply_rotation_in_the_plane(x, y, g);
|
||||
}
|
||||
}
|
||||
}
|
||||
else
|
||||
{
|
||||
temp = vec;
|
||||
RealScalar beta = 1;
|
||||
for(Index j=0; j<n; ++j)
|
||||
{
|
||||
RealScalar Ljj = numext::real(mat.coeff(j,j));
|
||||
RealScalar dj = numext::abs2(Ljj);
|
||||
Scalar wj = temp.coeff(j);
|
||||
RealScalar swj2 = sigma*numext::abs2(wj);
|
||||
RealScalar gamma = dj*beta + swj2;
|
||||
|
||||
RealScalar x = dj + swj2/beta;
|
||||
if (x<=RealScalar(0))
|
||||
return j;
|
||||
RealScalar nLjj = sqrt(x);
|
||||
mat.coeffRef(j,j) = nLjj;
|
||||
beta += swj2/dj;
|
||||
|
||||
// Update the terms of L
|
||||
Index rs = n-j-1;
|
||||
if(rs)
|
||||
{
|
||||
temp.tail(rs) -= (wj/Ljj) * mat.col(j).tail(rs);
|
||||
if(gamma != 0)
|
||||
mat.col(j).tail(rs) = (nLjj/Ljj) * mat.col(j).tail(rs) + (nLjj * sigma*numext::conj(wj)/gamma)*temp.tail(rs);
|
||||
}
|
||||
}
|
||||
}
|
||||
return -1;
|
||||
}
|
||||
|
||||
template<typename Scalar> struct llt_inplace<Scalar, Lower>
|
||||
{
|
||||
typedef typename NumTraits<Scalar>::Real RealScalar;
|
||||
template<typename MatrixType>
|
||||
static Index unblocked(MatrixType& mat)
|
||||
{
|
||||
using std::sqrt;
|
||||
|
||||
eigen_assert(mat.rows()==mat.cols());
|
||||
const Index size = mat.rows();
|
||||
for(Index k = 0; k < size; ++k)
|
||||
{
|
||||
Index rs = size-k-1; // remaining size
|
||||
|
||||
Block<MatrixType,Dynamic,1> A21(mat,k+1,k,rs,1);
|
||||
Block<MatrixType,1,Dynamic> A10(mat,k,0,1,k);
|
||||
Block<MatrixType,Dynamic,Dynamic> A20(mat,k+1,0,rs,k);
|
||||
|
||||
RealScalar x = numext::real(mat.coeff(k,k));
|
||||
if (k>0) x -= A10.squaredNorm();
|
||||
if (x<=RealScalar(0))
|
||||
return k;
|
||||
mat.coeffRef(k,k) = x = sqrt(x);
|
||||
if (k>0 && rs>0) A21.noalias() -= A20 * A10.adjoint();
|
||||
if (rs>0) A21 /= x;
|
||||
}
|
||||
return -1;
|
||||
}
|
||||
|
||||
template<typename MatrixType>
|
||||
static Index blocked(MatrixType& m)
|
||||
{
|
||||
eigen_assert(m.rows()==m.cols());
|
||||
Index size = m.rows();
|
||||
if(size<32)
|
||||
return unblocked(m);
|
||||
|
||||
Index blockSize = size/8;
|
||||
blockSize = (blockSize/16)*16;
|
||||
blockSize = (std::min)((std::max)(blockSize,Index(8)), Index(128));
|
||||
|
||||
for (Index k=0; k<size; k+=blockSize)
|
||||
{
|
||||
// partition the matrix:
|
||||
// A00 | - | -
|
||||
// lu = A10 | A11 | -
|
||||
// A20 | A21 | A22
|
||||
Index bs = (std::min)(blockSize, size-k);
|
||||
Index rs = size - k - bs;
|
||||
Block<MatrixType,Dynamic,Dynamic> A11(m,k, k, bs,bs);
|
||||
Block<MatrixType,Dynamic,Dynamic> A21(m,k+bs,k, rs,bs);
|
||||
Block<MatrixType,Dynamic,Dynamic> A22(m,k+bs,k+bs,rs,rs);
|
||||
|
||||
Index ret;
|
||||
if((ret=unblocked(A11))>=0) return k+ret;
|
||||
if(rs>0) A11.adjoint().template triangularView<Upper>().template solveInPlace<OnTheRight>(A21);
|
||||
if(rs>0) A22.template selfadjointView<Lower>().rankUpdate(A21,typename NumTraits<RealScalar>::Literal(-1)); // bottleneck
|
||||
}
|
||||
return -1;
|
||||
}
|
||||
|
||||
template<typename MatrixType, typename VectorType>
|
||||
static Index rankUpdate(MatrixType& mat, const VectorType& vec, const RealScalar& sigma)
|
||||
{
|
||||
return Eigen::internal::llt_rank_update_lower(mat, vec, sigma);
|
||||
}
|
||||
};
|
||||
|
||||
template<typename Scalar> struct llt_inplace<Scalar, Upper>
|
||||
{
|
||||
typedef typename NumTraits<Scalar>::Real RealScalar;
|
||||
|
||||
template<typename MatrixType>
|
||||
static EIGEN_STRONG_INLINE Index unblocked(MatrixType& mat)
|
||||
{
|
||||
Transpose<MatrixType> matt(mat);
|
||||
return llt_inplace<Scalar, Lower>::unblocked(matt);
|
||||
}
|
||||
template<typename MatrixType>
|
||||
static EIGEN_STRONG_INLINE Index blocked(MatrixType& mat)
|
||||
{
|
||||
Transpose<MatrixType> matt(mat);
|
||||
return llt_inplace<Scalar, Lower>::blocked(matt);
|
||||
}
|
||||
template<typename MatrixType, typename VectorType>
|
||||
static Index rankUpdate(MatrixType& mat, const VectorType& vec, const RealScalar& sigma)
|
||||
{
|
||||
Transpose<MatrixType> matt(mat);
|
||||
return llt_inplace<Scalar, Lower>::rankUpdate(matt, vec.conjugate(), sigma);
|
||||
}
|
||||
};
|
||||
|
||||
template<typename MatrixType> struct LLT_Traits<MatrixType,Lower>
|
||||
{
|
||||
typedef const TriangularView<const MatrixType, Lower> MatrixL;
|
||||
typedef const TriangularView<const typename MatrixType::AdjointReturnType, Upper> MatrixU;
|
||||
static inline MatrixL getL(const MatrixType& m) { return MatrixL(m); }
|
||||
static inline MatrixU getU(const MatrixType& m) { return MatrixU(m.adjoint()); }
|
||||
static bool inplace_decomposition(MatrixType& m)
|
||||
{ return llt_inplace<typename MatrixType::Scalar, Lower>::blocked(m)==-1; }
|
||||
};
|
||||
|
||||
template<typename MatrixType> struct LLT_Traits<MatrixType,Upper>
|
||||
{
|
||||
typedef const TriangularView<const typename MatrixType::AdjointReturnType, Lower> MatrixL;
|
||||
typedef const TriangularView<const MatrixType, Upper> MatrixU;
|
||||
static inline MatrixL getL(const MatrixType& m) { return MatrixL(m.adjoint()); }
|
||||
static inline MatrixU getU(const MatrixType& m) { return MatrixU(m); }
|
||||
static bool inplace_decomposition(MatrixType& m)
|
||||
{ return llt_inplace<typename MatrixType::Scalar, Upper>::blocked(m)==-1; }
|
||||
};
|
||||
|
||||
} // end namespace internal
|
||||
|
||||
/** Computes / recomputes the Cholesky decomposition A = LL^* = U^*U of \a matrix
|
||||
*
|
||||
* \returns a reference to *this
|
||||
*
|
||||
* Example: \include TutorialLinAlgComputeTwice.cpp
|
||||
* Output: \verbinclude TutorialLinAlgComputeTwice.out
|
||||
*/
|
||||
template<typename MatrixType, int _UpLo>
|
||||
template<typename InputType>
|
||||
LLT<MatrixType,_UpLo>& LLT<MatrixType,_UpLo>::compute(const EigenBase<InputType>& a)
|
||||
{
|
||||
check_template_parameters();
|
||||
|
||||
eigen_assert(a.rows()==a.cols());
|
||||
const Index size = a.rows();
|
||||
m_matrix.resize(size, size);
|
||||
if (!internal::is_same_dense(m_matrix, a.derived()))
|
||||
m_matrix = a.derived();
|
||||
|
||||
// Compute matrix L1 norm = max abs column sum.
|
||||
m_l1_norm = RealScalar(0);
|
||||
// TODO move this code to SelfAdjointView
|
||||
for (Index col = 0; col < size; ++col) {
|
||||
RealScalar abs_col_sum;
|
||||
if (_UpLo == Lower)
|
||||
abs_col_sum = m_matrix.col(col).tail(size - col).template lpNorm<1>() + m_matrix.row(col).head(col).template lpNorm<1>();
|
||||
else
|
||||
abs_col_sum = m_matrix.col(col).head(col).template lpNorm<1>() + m_matrix.row(col).tail(size - col).template lpNorm<1>();
|
||||
if (abs_col_sum > m_l1_norm)
|
||||
m_l1_norm = abs_col_sum;
|
||||
}
|
||||
|
||||
m_isInitialized = true;
|
||||
bool ok = Traits::inplace_decomposition(m_matrix);
|
||||
m_info = ok ? Success : NumericalIssue;
|
||||
|
||||
return *this;
|
||||
}
|
||||
|
||||
/** Performs a rank one update (or dowdate) of the current decomposition.
|
||||
* If A = LL^* before the rank one update,
|
||||
* then after it we have LL^* = A + sigma * v v^* where \a v must be a vector
|
||||
* of same dimension.
|
||||
*/
|
||||
template<typename _MatrixType, int _UpLo>
|
||||
template<typename VectorType>
|
||||
LLT<_MatrixType,_UpLo> & LLT<_MatrixType,_UpLo>::rankUpdate(const VectorType& v, const RealScalar& sigma)
|
||||
{
|
||||
EIGEN_STATIC_ASSERT_VECTOR_ONLY(VectorType);
|
||||
eigen_assert(v.size()==m_matrix.cols());
|
||||
eigen_assert(m_isInitialized);
|
||||
if(internal::llt_inplace<typename MatrixType::Scalar, UpLo>::rankUpdate(m_matrix,v,sigma)>=0)
|
||||
m_info = NumericalIssue;
|
||||
else
|
||||
m_info = Success;
|
||||
|
||||
return *this;
|
||||
}
|
||||
|
||||
#ifndef EIGEN_PARSED_BY_DOXYGEN
|
||||
template<typename _MatrixType,int _UpLo>
|
||||
template<typename RhsType, typename DstType>
|
||||
void LLT<_MatrixType,_UpLo>::_solve_impl(const RhsType &rhs, DstType &dst) const
|
||||
{
|
||||
_solve_impl_transposed<true>(rhs, dst);
|
||||
}
|
||||
|
||||
template<typename _MatrixType,int _UpLo>
|
||||
template<bool Conjugate, typename RhsType, typename DstType>
|
||||
void LLT<_MatrixType,_UpLo>::_solve_impl_transposed(const RhsType &rhs, DstType &dst) const
|
||||
{
|
||||
dst = rhs;
|
||||
|
||||
matrixL().template conjugateIf<!Conjugate>().solveInPlace(dst);
|
||||
matrixU().template conjugateIf<!Conjugate>().solveInPlace(dst);
|
||||
}
|
||||
#endif
|
||||
|
||||
/** \internal use x = llt_object.solve(x);
|
||||
*
|
||||
* This is the \em in-place version of solve().
|
||||
*
|
||||
* \param bAndX represents both the right-hand side matrix b and result x.
|
||||
*
|
||||
* This version avoids a copy when the right hand side matrix b is not needed anymore.
|
||||
*
|
||||
* \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.
|
||||
*
|
||||
* \sa LLT::solve(), MatrixBase::llt()
|
||||
*/
|
||||
template<typename MatrixType, int _UpLo>
|
||||
template<typename Derived>
|
||||
void LLT<MatrixType,_UpLo>::solveInPlace(const MatrixBase<Derived> &bAndX) const
|
||||
{
|
||||
eigen_assert(m_isInitialized && "LLT is not initialized.");
|
||||
eigen_assert(m_matrix.rows()==bAndX.rows());
|
||||
matrixL().solveInPlace(bAndX);
|
||||
matrixU().solveInPlace(bAndX);
|
||||
}
|
||||
|
||||
/** \returns the matrix represented by the decomposition,
|
||||
* i.e., it returns the product: L L^*.
|
||||
* This function is provided for debug purpose. */
|
||||
template<typename MatrixType, int _UpLo>
|
||||
MatrixType LLT<MatrixType,_UpLo>::reconstructedMatrix() const
|
||||
{
|
||||
eigen_assert(m_isInitialized && "LLT is not initialized.");
|
||||
return matrixL() * matrixL().adjoint().toDenseMatrix();
|
||||
}
|
||||
|
||||
/** \cholesky_module
|
||||
* \returns the LLT decomposition of \c *this
|
||||
* \sa SelfAdjointView::llt()
|
||||
*/
|
||||
template<typename Derived>
|
||||
inline const LLT<typename MatrixBase<Derived>::PlainObject>
|
||||
MatrixBase<Derived>::llt() const
|
||||
{
|
||||
return LLT<PlainObject>(derived());
|
||||
}
|
||||
|
||||
/** \cholesky_module
|
||||
* \returns the LLT decomposition of \c *this
|
||||
* \sa SelfAdjointView::llt()
|
||||
*/
|
||||
template<typename MatrixType, unsigned int UpLo>
|
||||
inline const LLT<typename SelfAdjointView<MatrixType, UpLo>::PlainObject, UpLo>
|
||||
SelfAdjointView<MatrixType, UpLo>::llt() const
|
||||
{
|
||||
return LLT<PlainObject,UpLo>(m_matrix);
|
||||
}
|
||||
|
||||
} // end namespace Eigen
|
||||
|
||||
#endif // EIGEN_LLT_H
|
||||
@@ -0,0 +1,99 @@
|
||||
/*
|
||||
Copyright (c) 2011, Intel Corporation. All rights reserved.
|
||||
|
||||
Redistribution and use in source and binary forms, with or without modification,
|
||||
are permitted provided that the following conditions are met:
|
||||
|
||||
* Redistributions of source code must retain the above copyright notice, this
|
||||
list of conditions and the following disclaimer.
|
||||
* Redistributions in binary form must reproduce the above copyright notice,
|
||||
this list of conditions and the following disclaimer in the documentation
|
||||
and/or other materials provided with the distribution.
|
||||
* Neither the name of Intel Corporation nor the names of its contributors may
|
||||
be used to endorse or promote products derived from this software without
|
||||
specific prior written permission.
|
||||
|
||||
THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" AND
|
||||
ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED
|
||||
WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE
|
||||
DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE LIABLE FOR
|
||||
ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES
|
||||
(INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES;
|
||||
LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON
|
||||
ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
|
||||
(INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS
|
||||
SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
|
||||
|
||||
********************************************************************************
|
||||
* Content : Eigen bindings to LAPACKe
|
||||
* LLt decomposition based on LAPACKE_?potrf function.
|
||||
********************************************************************************
|
||||
*/
|
||||
|
||||
#ifndef EIGEN_LLT_LAPACKE_H
|
||||
#define EIGEN_LLT_LAPACKE_H
|
||||
|
||||
namespace Eigen {
|
||||
|
||||
namespace internal {
|
||||
|
||||
template<typename Scalar> struct lapacke_llt;
|
||||
|
||||
#define EIGEN_LAPACKE_LLT(EIGTYPE, BLASTYPE, LAPACKE_PREFIX) \
|
||||
template<> struct lapacke_llt<EIGTYPE> \
|
||||
{ \
|
||||
template<typename MatrixType> \
|
||||
static inline Index potrf(MatrixType& m, char uplo) \
|
||||
{ \
|
||||
lapack_int matrix_order; \
|
||||
lapack_int size, lda, info, StorageOrder; \
|
||||
EIGTYPE* a; \
|
||||
eigen_assert(m.rows()==m.cols()); \
|
||||
/* Set up parameters for ?potrf */ \
|
||||
size = convert_index<lapack_int>(m.rows()); \
|
||||
StorageOrder = MatrixType::Flags&RowMajorBit?RowMajor:ColMajor; \
|
||||
matrix_order = StorageOrder==RowMajor ? LAPACK_ROW_MAJOR : LAPACK_COL_MAJOR; \
|
||||
a = &(m.coeffRef(0,0)); \
|
||||
lda = convert_index<lapack_int>(m.outerStride()); \
|
||||
\
|
||||
info = LAPACKE_##LAPACKE_PREFIX##potrf( matrix_order, uplo, size, (BLASTYPE*)a, lda ); \
|
||||
info = (info==0) ? -1 : info>0 ? info-1 : size; \
|
||||
return info; \
|
||||
} \
|
||||
}; \
|
||||
template<> struct llt_inplace<EIGTYPE, Lower> \
|
||||
{ \
|
||||
template<typename MatrixType> \
|
||||
static Index blocked(MatrixType& m) \
|
||||
{ \
|
||||
return lapacke_llt<EIGTYPE>::potrf(m, 'L'); \
|
||||
} \
|
||||
template<typename MatrixType, typename VectorType> \
|
||||
static Index rankUpdate(MatrixType& mat, const VectorType& vec, const typename MatrixType::RealScalar& sigma) \
|
||||
{ return Eigen::internal::llt_rank_update_lower(mat, vec, sigma); } \
|
||||
}; \
|
||||
template<> struct llt_inplace<EIGTYPE, Upper> \
|
||||
{ \
|
||||
template<typename MatrixType> \
|
||||
static Index blocked(MatrixType& m) \
|
||||
{ \
|
||||
return lapacke_llt<EIGTYPE>::potrf(m, 'U'); \
|
||||
} \
|
||||
template<typename MatrixType, typename VectorType> \
|
||||
static Index rankUpdate(MatrixType& mat, const VectorType& vec, const typename MatrixType::RealScalar& sigma) \
|
||||
{ \
|
||||
Transpose<MatrixType> matt(mat); \
|
||||
return llt_inplace<EIGTYPE, Lower>::rankUpdate(matt, vec.conjugate(), sigma); \
|
||||
} \
|
||||
};
|
||||
|
||||
EIGEN_LAPACKE_LLT(double, double, d)
|
||||
EIGEN_LAPACKE_LLT(float, float, s)
|
||||
EIGEN_LAPACKE_LLT(dcomplex, lapack_complex_double, z)
|
||||
EIGEN_LAPACKE_LLT(scomplex, lapack_complex_float, c)
|
||||
|
||||
} // end namespace internal
|
||||
|
||||
} // end namespace Eigen
|
||||
|
||||
#endif // EIGEN_LLT_LAPACKE_H
|
||||
@@ -0,0 +1,682 @@
|
||||
// This file is part of Eigen, a lightweight C++ template library
|
||||
// for linear algebra.
|
||||
//
|
||||
// Copyright (C) 2008-2010 Gael Guennebaud <gael.guennebaud@inria.fr>
|
||||
//
|
||||
// This Source Code Form is subject to the terms of the Mozilla
|
||||
// Public License v. 2.0. If a copy of the MPL was not distributed
|
||||
// with this file, You can obtain one at http://mozilla.org/MPL/2.0/.
|
||||
|
||||
#ifndef EIGEN_CHOLMODSUPPORT_H
|
||||
#define EIGEN_CHOLMODSUPPORT_H
|
||||
|
||||
namespace Eigen {
|
||||
|
||||
namespace internal {
|
||||
|
||||
template<typename Scalar> struct cholmod_configure_matrix;
|
||||
|
||||
template<> struct cholmod_configure_matrix<double> {
|
||||
template<typename CholmodType>
|
||||
static void run(CholmodType& mat) {
|
||||
mat.xtype = CHOLMOD_REAL;
|
||||
mat.dtype = CHOLMOD_DOUBLE;
|
||||
}
|
||||
};
|
||||
|
||||
template<> struct cholmod_configure_matrix<std::complex<double> > {
|
||||
template<typename CholmodType>
|
||||
static void run(CholmodType& mat) {
|
||||
mat.xtype = CHOLMOD_COMPLEX;
|
||||
mat.dtype = CHOLMOD_DOUBLE;
|
||||
}
|
||||
};
|
||||
|
||||
// Other scalar types are not yet supported by Cholmod
|
||||
// template<> struct cholmod_configure_matrix<float> {
|
||||
// template<typename CholmodType>
|
||||
// static void run(CholmodType& mat) {
|
||||
// mat.xtype = CHOLMOD_REAL;
|
||||
// mat.dtype = CHOLMOD_SINGLE;
|
||||
// }
|
||||
// };
|
||||
//
|
||||
// template<> struct cholmod_configure_matrix<std::complex<float> > {
|
||||
// template<typename CholmodType>
|
||||
// static void run(CholmodType& mat) {
|
||||
// mat.xtype = CHOLMOD_COMPLEX;
|
||||
// mat.dtype = CHOLMOD_SINGLE;
|
||||
// }
|
||||
// };
|
||||
|
||||
} // namespace internal
|
||||
|
||||
/** Wraps the Eigen sparse matrix \a mat into a Cholmod sparse matrix object.
|
||||
* Note that the data are shared.
|
||||
*/
|
||||
template<typename _Scalar, int _Options, typename _StorageIndex>
|
||||
cholmod_sparse viewAsCholmod(Ref<SparseMatrix<_Scalar,_Options,_StorageIndex> > mat)
|
||||
{
|
||||
cholmod_sparse res;
|
||||
res.nzmax = mat.nonZeros();
|
||||
res.nrow = mat.rows();
|
||||
res.ncol = mat.cols();
|
||||
res.p = mat.outerIndexPtr();
|
||||
res.i = mat.innerIndexPtr();
|
||||
res.x = mat.valuePtr();
|
||||
res.z = 0;
|
||||
res.sorted = 1;
|
||||
if(mat.isCompressed())
|
||||
{
|
||||
res.packed = 1;
|
||||
res.nz = 0;
|
||||
}
|
||||
else
|
||||
{
|
||||
res.packed = 0;
|
||||
res.nz = mat.innerNonZeroPtr();
|
||||
}
|
||||
|
||||
res.dtype = 0;
|
||||
res.stype = -1;
|
||||
|
||||
if (internal::is_same<_StorageIndex,int>::value)
|
||||
{
|
||||
res.itype = CHOLMOD_INT;
|
||||
}
|
||||
else if (internal::is_same<_StorageIndex,SuiteSparse_long>::value)
|
||||
{
|
||||
res.itype = CHOLMOD_LONG;
|
||||
}
|
||||
else
|
||||
{
|
||||
eigen_assert(false && "Index type not supported yet");
|
||||
}
|
||||
|
||||
// setup res.xtype
|
||||
internal::cholmod_configure_matrix<_Scalar>::run(res);
|
||||
|
||||
res.stype = 0;
|
||||
|
||||
return res;
|
||||
}
|
||||
|
||||
template<typename _Scalar, int _Options, typename _Index>
|
||||
const cholmod_sparse viewAsCholmod(const SparseMatrix<_Scalar,_Options,_Index>& mat)
|
||||
{
|
||||
cholmod_sparse res = viewAsCholmod(Ref<SparseMatrix<_Scalar,_Options,_Index> >(mat.const_cast_derived()));
|
||||
return res;
|
||||
}
|
||||
|
||||
template<typename _Scalar, int _Options, typename _Index>
|
||||
const cholmod_sparse viewAsCholmod(const SparseVector<_Scalar,_Options,_Index>& mat)
|
||||
{
|
||||
cholmod_sparse res = viewAsCholmod(Ref<SparseMatrix<_Scalar,_Options,_Index> >(mat.const_cast_derived()));
|
||||
return res;
|
||||
}
|
||||
|
||||
/** Returns a view of the Eigen sparse matrix \a mat as Cholmod sparse matrix.
|
||||
* The data are not copied but shared. */
|
||||
template<typename _Scalar, int _Options, typename _Index, unsigned int UpLo>
|
||||
cholmod_sparse viewAsCholmod(const SparseSelfAdjointView<const SparseMatrix<_Scalar,_Options,_Index>, UpLo>& mat)
|
||||
{
|
||||
cholmod_sparse res = viewAsCholmod(Ref<SparseMatrix<_Scalar,_Options,_Index> >(mat.matrix().const_cast_derived()));
|
||||
|
||||
if(UpLo==Upper) res.stype = 1;
|
||||
if(UpLo==Lower) res.stype = -1;
|
||||
// swap stype for rowmajor matrices (only works for real matrices)
|
||||
EIGEN_STATIC_ASSERT((_Options & RowMajorBit) == 0 || NumTraits<_Scalar>::IsComplex == 0, THIS_METHOD_IS_ONLY_FOR_COLUMN_MAJOR_MATRICES);
|
||||
if(_Options & RowMajorBit) res.stype *=-1;
|
||||
|
||||
return res;
|
||||
}
|
||||
|
||||
/** Returns a view of the Eigen \b dense matrix \a mat as Cholmod dense matrix.
|
||||
* The data are not copied but shared. */
|
||||
template<typename Derived>
|
||||
cholmod_dense viewAsCholmod(MatrixBase<Derived>& mat)
|
||||
{
|
||||
EIGEN_STATIC_ASSERT((internal::traits<Derived>::Flags&RowMajorBit)==0,THIS_METHOD_IS_ONLY_FOR_COLUMN_MAJOR_MATRICES);
|
||||
typedef typename Derived::Scalar Scalar;
|
||||
|
||||
cholmod_dense res;
|
||||
res.nrow = mat.rows();
|
||||
res.ncol = mat.cols();
|
||||
res.nzmax = res.nrow * res.ncol;
|
||||
res.d = Derived::IsVectorAtCompileTime ? mat.derived().size() : mat.derived().outerStride();
|
||||
res.x = (void*)(mat.derived().data());
|
||||
res.z = 0;
|
||||
|
||||
internal::cholmod_configure_matrix<Scalar>::run(res);
|
||||
|
||||
return res;
|
||||
}
|
||||
|
||||
/** Returns a view of the Cholmod sparse matrix \a cm as an Eigen sparse matrix.
|
||||
* The data are not copied but shared. */
|
||||
template<typename Scalar, int Flags, typename StorageIndex>
|
||||
MappedSparseMatrix<Scalar,Flags,StorageIndex> viewAsEigen(cholmod_sparse& cm)
|
||||
{
|
||||
return MappedSparseMatrix<Scalar,Flags,StorageIndex>
|
||||
(cm.nrow, cm.ncol, static_cast<StorageIndex*>(cm.p)[cm.ncol],
|
||||
static_cast<StorageIndex*>(cm.p), static_cast<StorageIndex*>(cm.i),static_cast<Scalar*>(cm.x) );
|
||||
}
|
||||
|
||||
namespace internal {
|
||||
|
||||
// template specializations for int and long that call the correct cholmod method
|
||||
|
||||
#define EIGEN_CHOLMOD_SPECIALIZE0(ret, name) \
|
||||
template<typename _StorageIndex> inline ret cm_ ## name (cholmod_common &Common) { return cholmod_ ## name (&Common); } \
|
||||
template<> inline ret cm_ ## name<SuiteSparse_long> (cholmod_common &Common) { return cholmod_l_ ## name (&Common); }
|
||||
|
||||
#define EIGEN_CHOLMOD_SPECIALIZE1(ret, name, t1, a1) \
|
||||
template<typename _StorageIndex> inline ret cm_ ## name (t1& a1, cholmod_common &Common) { return cholmod_ ## name (&a1, &Common); } \
|
||||
template<> inline ret cm_ ## name<SuiteSparse_long> (t1& a1, cholmod_common &Common) { return cholmod_l_ ## name (&a1, &Common); }
|
||||
|
||||
EIGEN_CHOLMOD_SPECIALIZE0(int, start)
|
||||
EIGEN_CHOLMOD_SPECIALIZE0(int, finish)
|
||||
|
||||
EIGEN_CHOLMOD_SPECIALIZE1(int, free_factor, cholmod_factor*, L)
|
||||
EIGEN_CHOLMOD_SPECIALIZE1(int, free_dense, cholmod_dense*, X)
|
||||
EIGEN_CHOLMOD_SPECIALIZE1(int, free_sparse, cholmod_sparse*, A)
|
||||
|
||||
EIGEN_CHOLMOD_SPECIALIZE1(cholmod_factor*, analyze, cholmod_sparse, A)
|
||||
|
||||
template<typename _StorageIndex> inline cholmod_dense* cm_solve (int sys, cholmod_factor& L, cholmod_dense& B, cholmod_common &Common) { return cholmod_solve (sys, &L, &B, &Common); }
|
||||
template<> inline cholmod_dense* cm_solve<SuiteSparse_long> (int sys, cholmod_factor& L, cholmod_dense& B, cholmod_common &Common) { return cholmod_l_solve (sys, &L, &B, &Common); }
|
||||
|
||||
template<typename _StorageIndex> inline cholmod_sparse* cm_spsolve (int sys, cholmod_factor& L, cholmod_sparse& B, cholmod_common &Common) { return cholmod_spsolve (sys, &L, &B, &Common); }
|
||||
template<> inline cholmod_sparse* cm_spsolve<SuiteSparse_long> (int sys, cholmod_factor& L, cholmod_sparse& B, cholmod_common &Common) { return cholmod_l_spsolve (sys, &L, &B, &Common); }
|
||||
|
||||
template<typename _StorageIndex>
|
||||
inline int cm_factorize_p (cholmod_sparse* A, double beta[2], _StorageIndex* fset, std::size_t fsize, cholmod_factor* L, cholmod_common &Common) { return cholmod_factorize_p (A, beta, fset, fsize, L, &Common); }
|
||||
template<>
|
||||
inline int cm_factorize_p<SuiteSparse_long> (cholmod_sparse* A, double beta[2], SuiteSparse_long* fset, std::size_t fsize, cholmod_factor* L, cholmod_common &Common) { return cholmod_l_factorize_p (A, beta, fset, fsize, L, &Common); }
|
||||
|
||||
#undef EIGEN_CHOLMOD_SPECIALIZE0
|
||||
#undef EIGEN_CHOLMOD_SPECIALIZE1
|
||||
|
||||
} // namespace internal
|
||||
|
||||
|
||||
enum CholmodMode {
|
||||
CholmodAuto, CholmodSimplicialLLt, CholmodSupernodalLLt, CholmodLDLt
|
||||
};
|
||||
|
||||
|
||||
/** \ingroup CholmodSupport_Module
|
||||
* \class CholmodBase
|
||||
* \brief The base class for the direct Cholesky factorization of Cholmod
|
||||
* \sa class CholmodSupernodalLLT, class CholmodSimplicialLDLT, class CholmodSimplicialLLT
|
||||
*/
|
||||
template<typename _MatrixType, int _UpLo, typename Derived>
|
||||
class CholmodBase : public SparseSolverBase<Derived>
|
||||
{
|
||||
protected:
|
||||
typedef SparseSolverBase<Derived> Base;
|
||||
using Base::derived;
|
||||
using Base::m_isInitialized;
|
||||
public:
|
||||
typedef _MatrixType MatrixType;
|
||||
enum { UpLo = _UpLo };
|
||||
typedef typename MatrixType::Scalar Scalar;
|
||||
typedef typename MatrixType::RealScalar RealScalar;
|
||||
typedef MatrixType CholMatrixType;
|
||||
typedef typename MatrixType::StorageIndex StorageIndex;
|
||||
enum {
|
||||
ColsAtCompileTime = MatrixType::ColsAtCompileTime,
|
||||
MaxColsAtCompileTime = MatrixType::MaxColsAtCompileTime
|
||||
};
|
||||
|
||||
public:
|
||||
|
||||
CholmodBase()
|
||||
: m_cholmodFactor(0), m_info(Success), m_factorizationIsOk(false), m_analysisIsOk(false)
|
||||
{
|
||||
EIGEN_STATIC_ASSERT((internal::is_same<double,RealScalar>::value), CHOLMOD_SUPPORTS_DOUBLE_PRECISION_ONLY);
|
||||
m_shiftOffset[0] = m_shiftOffset[1] = 0.0;
|
||||
internal::cm_start<StorageIndex>(m_cholmod);
|
||||
}
|
||||
|
||||
explicit CholmodBase(const MatrixType& matrix)
|
||||
: m_cholmodFactor(0), m_info(Success), m_factorizationIsOk(false), m_analysisIsOk(false)
|
||||
{
|
||||
EIGEN_STATIC_ASSERT((internal::is_same<double,RealScalar>::value), CHOLMOD_SUPPORTS_DOUBLE_PRECISION_ONLY);
|
||||
m_shiftOffset[0] = m_shiftOffset[1] = 0.0;
|
||||
internal::cm_start<StorageIndex>(m_cholmod);
|
||||
compute(matrix);
|
||||
}
|
||||
|
||||
~CholmodBase()
|
||||
{
|
||||
if(m_cholmodFactor)
|
||||
internal::cm_free_factor<StorageIndex>(m_cholmodFactor, m_cholmod);
|
||||
internal::cm_finish<StorageIndex>(m_cholmod);
|
||||
}
|
||||
|
||||
inline StorageIndex cols() const { return internal::convert_index<StorageIndex, Index>(m_cholmodFactor->n); }
|
||||
inline StorageIndex rows() const { return internal::convert_index<StorageIndex, Index>(m_cholmodFactor->n); }
|
||||
|
||||
/** \brief Reports whether previous computation was successful.
|
||||
*
|
||||
* \returns \c Success if computation was successful,
|
||||
* \c NumericalIssue if the matrix.appears to be negative.
|
||||
*/
|
||||
ComputationInfo info() const
|
||||
{
|
||||
eigen_assert(m_isInitialized && "Decomposition is not initialized.");
|
||||
return m_info;
|
||||
}
|
||||
|
||||
/** Computes the sparse Cholesky decomposition of \a matrix */
|
||||
Derived& compute(const MatrixType& matrix)
|
||||
{
|
||||
analyzePattern(matrix);
|
||||
factorize(matrix);
|
||||
return derived();
|
||||
}
|
||||
|
||||
/** Performs a symbolic decomposition on the sparsity pattern of \a matrix.
|
||||
*
|
||||
* This function is particularly useful when solving for several problems having the same structure.
|
||||
*
|
||||
* \sa factorize()
|
||||
*/
|
||||
void analyzePattern(const MatrixType& matrix)
|
||||
{
|
||||
if(m_cholmodFactor)
|
||||
{
|
||||
internal::cm_free_factor<StorageIndex>(m_cholmodFactor, m_cholmod);
|
||||
m_cholmodFactor = 0;
|
||||
}
|
||||
cholmod_sparse A = viewAsCholmod(matrix.template selfadjointView<UpLo>());
|
||||
m_cholmodFactor = internal::cm_analyze<StorageIndex>(A, m_cholmod);
|
||||
|
||||
this->m_isInitialized = true;
|
||||
this->m_info = Success;
|
||||
m_analysisIsOk = true;
|
||||
m_factorizationIsOk = false;
|
||||
}
|
||||
|
||||
/** Performs a numeric decomposition of \a matrix
|
||||
*
|
||||
* The given matrix must have the same sparsity pattern as the matrix on which the symbolic decomposition has been performed.
|
||||
*
|
||||
* \sa analyzePattern()
|
||||
*/
|
||||
void factorize(const MatrixType& matrix)
|
||||
{
|
||||
eigen_assert(m_analysisIsOk && "You must first call analyzePattern()");
|
||||
cholmod_sparse A = viewAsCholmod(matrix.template selfadjointView<UpLo>());
|
||||
internal::cm_factorize_p<StorageIndex>(&A, m_shiftOffset, 0, 0, m_cholmodFactor, m_cholmod);
|
||||
|
||||
// If the factorization failed, minor is the column at which it did. On success minor == n.
|
||||
this->m_info = (m_cholmodFactor->minor == m_cholmodFactor->n ? Success : NumericalIssue);
|
||||
m_factorizationIsOk = true;
|
||||
}
|
||||
|
||||
/** Returns a reference to the Cholmod's configuration structure to get a full control over the performed operations.
|
||||
* See the Cholmod user guide for details. */
|
||||
cholmod_common& cholmod() { return m_cholmod; }
|
||||
|
||||
#ifndef EIGEN_PARSED_BY_DOXYGEN
|
||||
/** \internal */
|
||||
template<typename Rhs,typename Dest>
|
||||
void _solve_impl(const MatrixBase<Rhs> &b, MatrixBase<Dest> &dest) const
|
||||
{
|
||||
eigen_assert(m_factorizationIsOk && "The decomposition is not in a valid state for solving, you must first call either compute() or symbolic()/numeric()");
|
||||
const Index size = m_cholmodFactor->n;
|
||||
EIGEN_UNUSED_VARIABLE(size);
|
||||
eigen_assert(size==b.rows());
|
||||
|
||||
// Cholmod needs column-major storage without inner-stride, which corresponds to the default behavior of Ref.
|
||||
Ref<const Matrix<typename Rhs::Scalar,Dynamic,Dynamic,ColMajor> > b_ref(b.derived());
|
||||
|
||||
cholmod_dense b_cd = viewAsCholmod(b_ref);
|
||||
cholmod_dense* x_cd = internal::cm_solve<StorageIndex>(CHOLMOD_A, *m_cholmodFactor, b_cd, m_cholmod);
|
||||
if(!x_cd)
|
||||
{
|
||||
this->m_info = NumericalIssue;
|
||||
return;
|
||||
}
|
||||
// TODO optimize this copy by swapping when possible (be careful with alignment, etc.)
|
||||
// NOTE Actually, the copy can be avoided by calling cholmod_solve2 instead of cholmod_solve
|
||||
dest = Matrix<Scalar,Dest::RowsAtCompileTime,Dest::ColsAtCompileTime>::Map(reinterpret_cast<Scalar*>(x_cd->x),b.rows(),b.cols());
|
||||
internal::cm_free_dense<StorageIndex>(x_cd, m_cholmod);
|
||||
}
|
||||
|
||||
/** \internal */
|
||||
template<typename RhsDerived, typename DestDerived>
|
||||
void _solve_impl(const SparseMatrixBase<RhsDerived> &b, SparseMatrixBase<DestDerived> &dest) const
|
||||
{
|
||||
eigen_assert(m_factorizationIsOk && "The decomposition is not in a valid state for solving, you must first call either compute() or symbolic()/numeric()");
|
||||
const Index size = m_cholmodFactor->n;
|
||||
EIGEN_UNUSED_VARIABLE(size);
|
||||
eigen_assert(size==b.rows());
|
||||
|
||||
// note: cs stands for Cholmod Sparse
|
||||
Ref<SparseMatrix<typename RhsDerived::Scalar,ColMajor,typename RhsDerived::StorageIndex> > b_ref(b.const_cast_derived());
|
||||
cholmod_sparse b_cs = viewAsCholmod(b_ref);
|
||||
cholmod_sparse* x_cs = internal::cm_spsolve<StorageIndex>(CHOLMOD_A, *m_cholmodFactor, b_cs, m_cholmod);
|
||||
if(!x_cs)
|
||||
{
|
||||
this->m_info = NumericalIssue;
|
||||
return;
|
||||
}
|
||||
// TODO optimize this copy by swapping when possible (be careful with alignment, etc.)
|
||||
// NOTE cholmod_spsolve in fact just calls the dense solver for blocks of 4 columns at a time (similar to Eigen's sparse solver)
|
||||
dest.derived() = viewAsEigen<typename DestDerived::Scalar,ColMajor,typename DestDerived::StorageIndex>(*x_cs);
|
||||
internal::cm_free_sparse<StorageIndex>(x_cs, m_cholmod);
|
||||
}
|
||||
#endif // EIGEN_PARSED_BY_DOXYGEN
|
||||
|
||||
|
||||
/** Sets the shift parameter that will be used to adjust the diagonal coefficients during the numerical factorization.
|
||||
*
|
||||
* During the numerical factorization, an offset term is added to the diagonal coefficients:\n
|
||||
* \c d_ii = \a offset + \c d_ii
|
||||
*
|
||||
* The default is \a offset=0.
|
||||
*
|
||||
* \returns a reference to \c *this.
|
||||
*/
|
||||
Derived& setShift(const RealScalar& offset)
|
||||
{
|
||||
m_shiftOffset[0] = double(offset);
|
||||
return derived();
|
||||
}
|
||||
|
||||
/** \returns the determinant of the underlying matrix from the current factorization */
|
||||
Scalar determinant() const
|
||||
{
|
||||
using std::exp;
|
||||
return exp(logDeterminant());
|
||||
}
|
||||
|
||||
/** \returns the log determinant of the underlying matrix from the current factorization */
|
||||
Scalar logDeterminant() const
|
||||
{
|
||||
using std::log;
|
||||
using numext::real;
|
||||
eigen_assert(m_factorizationIsOk && "The decomposition is not in a valid state for solving, you must first call either compute() or symbolic()/numeric()");
|
||||
|
||||
RealScalar logDet = 0;
|
||||
Scalar *x = static_cast<Scalar*>(m_cholmodFactor->x);
|
||||
if (m_cholmodFactor->is_super)
|
||||
{
|
||||
// Supernodal factorization stored as a packed list of dense column-major blocs,
|
||||
// as described by the following structure:
|
||||
|
||||
// super[k] == index of the first column of the j-th super node
|
||||
StorageIndex *super = static_cast<StorageIndex*>(m_cholmodFactor->super);
|
||||
// pi[k] == offset to the description of row indices
|
||||
StorageIndex *pi = static_cast<StorageIndex*>(m_cholmodFactor->pi);
|
||||
// px[k] == offset to the respective dense block
|
||||
StorageIndex *px = static_cast<StorageIndex*>(m_cholmodFactor->px);
|
||||
|
||||
Index nb_super_nodes = m_cholmodFactor->nsuper;
|
||||
for (Index k=0; k < nb_super_nodes; ++k)
|
||||
{
|
||||
StorageIndex ncols = super[k + 1] - super[k];
|
||||
StorageIndex nrows = pi[k + 1] - pi[k];
|
||||
|
||||
Map<const Array<Scalar,1,Dynamic>, 0, InnerStride<> > sk(x + px[k], ncols, InnerStride<>(nrows+1));
|
||||
logDet += sk.real().log().sum();
|
||||
}
|
||||
}
|
||||
else
|
||||
{
|
||||
// Simplicial factorization stored as standard CSC matrix.
|
||||
StorageIndex *p = static_cast<StorageIndex*>(m_cholmodFactor->p);
|
||||
Index size = m_cholmodFactor->n;
|
||||
for (Index k=0; k<size; ++k)
|
||||
logDet += log(real( x[p[k]] ));
|
||||
}
|
||||
if (m_cholmodFactor->is_ll)
|
||||
logDet *= 2.0;
|
||||
return logDet;
|
||||
};
|
||||
|
||||
template<typename Stream>
|
||||
void dumpMemory(Stream& /*s*/)
|
||||
{}
|
||||
|
||||
protected:
|
||||
mutable cholmod_common m_cholmod;
|
||||
cholmod_factor* m_cholmodFactor;
|
||||
double m_shiftOffset[2];
|
||||
mutable ComputationInfo m_info;
|
||||
int m_factorizationIsOk;
|
||||
int m_analysisIsOk;
|
||||
};
|
||||
|
||||
/** \ingroup CholmodSupport_Module
|
||||
* \class CholmodSimplicialLLT
|
||||
* \brief A simplicial direct Cholesky (LLT) factorization and solver based on Cholmod
|
||||
*
|
||||
* This class allows to solve for A.X = B sparse linear problems via a simplicial LL^T Cholesky factorization
|
||||
* using the Cholmod library.
|
||||
* This simplicial variant is equivalent to Eigen's built-in SimplicialLLT class. Therefore, it has little practical interest.
|
||||
* The sparse matrix A must be selfadjoint and positive definite. The vectors or matrices
|
||||
* X and B can be either dense or sparse.
|
||||
*
|
||||
* \tparam _MatrixType the type of the sparse matrix A, it must be a SparseMatrix<>
|
||||
* \tparam _UpLo the triangular part that will be used for the computations. It can be Lower
|
||||
* or Upper. Default is Lower.
|
||||
*
|
||||
* \implsparsesolverconcept
|
||||
*
|
||||
* This class supports all kind of SparseMatrix<>: row or column major; upper, lower, or both; compressed or non compressed.
|
||||
*
|
||||
* \warning Only double precision real and complex scalar types are supported by Cholmod.
|
||||
*
|
||||
* \sa \ref TutorialSparseSolverConcept, class CholmodSupernodalLLT, class SimplicialLLT
|
||||
*/
|
||||
template<typename _MatrixType, int _UpLo = Lower>
|
||||
class CholmodSimplicialLLT : public CholmodBase<_MatrixType, _UpLo, CholmodSimplicialLLT<_MatrixType, _UpLo> >
|
||||
{
|
||||
typedef CholmodBase<_MatrixType, _UpLo, CholmodSimplicialLLT> Base;
|
||||
using Base::m_cholmod;
|
||||
|
||||
public:
|
||||
|
||||
typedef _MatrixType MatrixType;
|
||||
|
||||
CholmodSimplicialLLT() : Base() { init(); }
|
||||
|
||||
CholmodSimplicialLLT(const MatrixType& matrix) : Base()
|
||||
{
|
||||
init();
|
||||
this->compute(matrix);
|
||||
}
|
||||
|
||||
~CholmodSimplicialLLT() {}
|
||||
protected:
|
||||
void init()
|
||||
{
|
||||
m_cholmod.final_asis = 0;
|
||||
m_cholmod.supernodal = CHOLMOD_SIMPLICIAL;
|
||||
m_cholmod.final_ll = 1;
|
||||
}
|
||||
};
|
||||
|
||||
|
||||
/** \ingroup CholmodSupport_Module
|
||||
* \class CholmodSimplicialLDLT
|
||||
* \brief A simplicial direct Cholesky (LDLT) factorization and solver based on Cholmod
|
||||
*
|
||||
* This class allows to solve for A.X = B sparse linear problems via a simplicial LDL^T Cholesky factorization
|
||||
* using the Cholmod library.
|
||||
* This simplicial variant is equivalent to Eigen's built-in SimplicialLDLT class. Therefore, it has little practical interest.
|
||||
* The sparse matrix A must be selfadjoint and positive definite. The vectors or matrices
|
||||
* X and B can be either dense or sparse.
|
||||
*
|
||||
* \tparam _MatrixType the type of the sparse matrix A, it must be a SparseMatrix<>
|
||||
* \tparam _UpLo the triangular part that will be used for the computations. It can be Lower
|
||||
* or Upper. Default is Lower.
|
||||
*
|
||||
* \implsparsesolverconcept
|
||||
*
|
||||
* This class supports all kind of SparseMatrix<>: row or column major; upper, lower, or both; compressed or non compressed.
|
||||
*
|
||||
* \warning Only double precision real and complex scalar types are supported by Cholmod.
|
||||
*
|
||||
* \sa \ref TutorialSparseSolverConcept, class CholmodSupernodalLLT, class SimplicialLDLT
|
||||
*/
|
||||
template<typename _MatrixType, int _UpLo = Lower>
|
||||
class CholmodSimplicialLDLT : public CholmodBase<_MatrixType, _UpLo, CholmodSimplicialLDLT<_MatrixType, _UpLo> >
|
||||
{
|
||||
typedef CholmodBase<_MatrixType, _UpLo, CholmodSimplicialLDLT> Base;
|
||||
using Base::m_cholmod;
|
||||
|
||||
public:
|
||||
|
||||
typedef _MatrixType MatrixType;
|
||||
|
||||
CholmodSimplicialLDLT() : Base() { init(); }
|
||||
|
||||
CholmodSimplicialLDLT(const MatrixType& matrix) : Base()
|
||||
{
|
||||
init();
|
||||
this->compute(matrix);
|
||||
}
|
||||
|
||||
~CholmodSimplicialLDLT() {}
|
||||
protected:
|
||||
void init()
|
||||
{
|
||||
m_cholmod.final_asis = 1;
|
||||
m_cholmod.supernodal = CHOLMOD_SIMPLICIAL;
|
||||
}
|
||||
};
|
||||
|
||||
/** \ingroup CholmodSupport_Module
|
||||
* \class CholmodSupernodalLLT
|
||||
* \brief A supernodal Cholesky (LLT) factorization and solver based on Cholmod
|
||||
*
|
||||
* This class allows to solve for A.X = B sparse linear problems via a supernodal LL^T Cholesky factorization
|
||||
* using the Cholmod library.
|
||||
* This supernodal variant performs best on dense enough problems, e.g., 3D FEM, or very high order 2D FEM.
|
||||
* The sparse matrix A must be selfadjoint and positive definite. The vectors or matrices
|
||||
* X and B can be either dense or sparse.
|
||||
*
|
||||
* \tparam _MatrixType the type of the sparse matrix A, it must be a SparseMatrix<>
|
||||
* \tparam _UpLo the triangular part that will be used for the computations. It can be Lower
|
||||
* or Upper. Default is Lower.
|
||||
*
|
||||
* \implsparsesolverconcept
|
||||
*
|
||||
* This class supports all kind of SparseMatrix<>: row or column major; upper, lower, or both; compressed or non compressed.
|
||||
*
|
||||
* \warning Only double precision real and complex scalar types are supported by Cholmod.
|
||||
*
|
||||
* \sa \ref TutorialSparseSolverConcept
|
||||
*/
|
||||
template<typename _MatrixType, int _UpLo = Lower>
|
||||
class CholmodSupernodalLLT : public CholmodBase<_MatrixType, _UpLo, CholmodSupernodalLLT<_MatrixType, _UpLo> >
|
||||
{
|
||||
typedef CholmodBase<_MatrixType, _UpLo, CholmodSupernodalLLT> Base;
|
||||
using Base::m_cholmod;
|
||||
|
||||
public:
|
||||
|
||||
typedef _MatrixType MatrixType;
|
||||
|
||||
CholmodSupernodalLLT() : Base() { init(); }
|
||||
|
||||
CholmodSupernodalLLT(const MatrixType& matrix) : Base()
|
||||
{
|
||||
init();
|
||||
this->compute(matrix);
|
||||
}
|
||||
|
||||
~CholmodSupernodalLLT() {}
|
||||
protected:
|
||||
void init()
|
||||
{
|
||||
m_cholmod.final_asis = 1;
|
||||
m_cholmod.supernodal = CHOLMOD_SUPERNODAL;
|
||||
}
|
||||
};
|
||||
|
||||
/** \ingroup CholmodSupport_Module
|
||||
* \class CholmodDecomposition
|
||||
* \brief A general Cholesky factorization and solver based on Cholmod
|
||||
*
|
||||
* This class allows to solve for A.X = B sparse linear problems via a LL^T or LDL^T Cholesky factorization
|
||||
* using the Cholmod library. The sparse matrix A must be selfadjoint and positive definite. The vectors or matrices
|
||||
* X and B can be either dense or sparse.
|
||||
*
|
||||
* This variant permits to change the underlying Cholesky method at runtime.
|
||||
* On the other hand, it does not provide access to the result of the factorization.
|
||||
* The default is to let Cholmod automatically choose between a simplicial and supernodal factorization.
|
||||
*
|
||||
* \tparam _MatrixType the type of the sparse matrix A, it must be a SparseMatrix<>
|
||||
* \tparam _UpLo the triangular part that will be used for the computations. It can be Lower
|
||||
* or Upper. Default is Lower.
|
||||
*
|
||||
* \implsparsesolverconcept
|
||||
*
|
||||
* This class supports all kind of SparseMatrix<>: row or column major; upper, lower, or both; compressed or non compressed.
|
||||
*
|
||||
* \warning Only double precision real and complex scalar types are supported by Cholmod.
|
||||
*
|
||||
* \sa \ref TutorialSparseSolverConcept
|
||||
*/
|
||||
template<typename _MatrixType, int _UpLo = Lower>
|
||||
class CholmodDecomposition : public CholmodBase<_MatrixType, _UpLo, CholmodDecomposition<_MatrixType, _UpLo> >
|
||||
{
|
||||
typedef CholmodBase<_MatrixType, _UpLo, CholmodDecomposition> Base;
|
||||
using Base::m_cholmod;
|
||||
|
||||
public:
|
||||
|
||||
typedef _MatrixType MatrixType;
|
||||
|
||||
CholmodDecomposition() : Base() { init(); }
|
||||
|
||||
CholmodDecomposition(const MatrixType& matrix) : Base()
|
||||
{
|
||||
init();
|
||||
this->compute(matrix);
|
||||
}
|
||||
|
||||
~CholmodDecomposition() {}
|
||||
|
||||
void setMode(CholmodMode mode)
|
||||
{
|
||||
switch(mode)
|
||||
{
|
||||
case CholmodAuto:
|
||||
m_cholmod.final_asis = 1;
|
||||
m_cholmod.supernodal = CHOLMOD_AUTO;
|
||||
break;
|
||||
case CholmodSimplicialLLt:
|
||||
m_cholmod.final_asis = 0;
|
||||
m_cholmod.supernodal = CHOLMOD_SIMPLICIAL;
|
||||
m_cholmod.final_ll = 1;
|
||||
break;
|
||||
case CholmodSupernodalLLt:
|
||||
m_cholmod.final_asis = 1;
|
||||
m_cholmod.supernodal = CHOLMOD_SUPERNODAL;
|
||||
break;
|
||||
case CholmodLDLt:
|
||||
m_cholmod.final_asis = 1;
|
||||
m_cholmod.supernodal = CHOLMOD_SIMPLICIAL;
|
||||
break;
|
||||
default:
|
||||
break;
|
||||
}
|
||||
}
|
||||
protected:
|
||||
void init()
|
||||
{
|
||||
m_cholmod.final_asis = 1;
|
||||
m_cholmod.supernodal = CHOLMOD_AUTO;
|
||||
}
|
||||
};
|
||||
|
||||
} // end namespace Eigen
|
||||
|
||||
#endif // EIGEN_CHOLMODSUPPORT_H
|
||||
@@ -0,0 +1,413 @@
|
||||
// This file is part of Eigen, a lightweight C++ template library
|
||||
// for linear algebra.
|
||||
//
|
||||
// Copyright (C) 2017 Gael Guennebaud <gael.guennebaud@inria.fr>
|
||||
//
|
||||
// This Source Code Form is subject to the terms of the Mozilla
|
||||
// Public License v. 2.0. If a copy of the MPL was not distributed
|
||||
// with this file, You can obtain one at http://mozilla.org/MPL/2.0/.
|
||||
|
||||
#ifndef EIGEN_ARITHMETIC_SEQUENCE_H
|
||||
#define EIGEN_ARITHMETIC_SEQUENCE_H
|
||||
|
||||
namespace Eigen {
|
||||
|
||||
namespace internal {
|
||||
|
||||
#if (!EIGEN_HAS_CXX11) || !((!EIGEN_COMP_GNUC) || EIGEN_COMP_GNUC>=48)
|
||||
template<typename T> struct aseq_negate {};
|
||||
|
||||
template<> struct aseq_negate<Index> {
|
||||
typedef Index type;
|
||||
};
|
||||
|
||||
template<int N> struct aseq_negate<FixedInt<N> > {
|
||||
typedef FixedInt<-N> type;
|
||||
};
|
||||
|
||||
// Compilation error in the following case:
|
||||
template<> struct aseq_negate<FixedInt<DynamicIndex> > {};
|
||||
|
||||
template<typename FirstType,typename SizeType,typename IncrType,
|
||||
bool FirstIsSymbolic=symbolic::is_symbolic<FirstType>::value,
|
||||
bool SizeIsSymbolic =symbolic::is_symbolic<SizeType>::value>
|
||||
struct aseq_reverse_first_type {
|
||||
typedef Index type;
|
||||
};
|
||||
|
||||
template<typename FirstType,typename SizeType,typename IncrType>
|
||||
struct aseq_reverse_first_type<FirstType,SizeType,IncrType,true,true> {
|
||||
typedef symbolic::AddExpr<FirstType,
|
||||
symbolic::ProductExpr<symbolic::AddExpr<SizeType,symbolic::ValueExpr<FixedInt<-1> > >,
|
||||
symbolic::ValueExpr<IncrType> >
|
||||
> type;
|
||||
};
|
||||
|
||||
template<typename SizeType,typename IncrType,typename EnableIf = void>
|
||||
struct aseq_reverse_first_type_aux {
|
||||
typedef Index type;
|
||||
};
|
||||
|
||||
template<typename SizeType,typename IncrType>
|
||||
struct aseq_reverse_first_type_aux<SizeType,IncrType,typename internal::enable_if<bool((SizeType::value+IncrType::value)|0x1)>::type> {
|
||||
typedef FixedInt<(SizeType::value-1)*IncrType::value> type;
|
||||
};
|
||||
|
||||
template<typename FirstType,typename SizeType,typename IncrType>
|
||||
struct aseq_reverse_first_type<FirstType,SizeType,IncrType,true,false> {
|
||||
typedef typename aseq_reverse_first_type_aux<SizeType,IncrType>::type Aux;
|
||||
typedef symbolic::AddExpr<FirstType,symbolic::ValueExpr<Aux> > type;
|
||||
};
|
||||
|
||||
template<typename FirstType,typename SizeType,typename IncrType>
|
||||
struct aseq_reverse_first_type<FirstType,SizeType,IncrType,false,true> {
|
||||
typedef symbolic::AddExpr<symbolic::ProductExpr<symbolic::AddExpr<SizeType,symbolic::ValueExpr<FixedInt<-1> > >,
|
||||
symbolic::ValueExpr<IncrType> >,
|
||||
symbolic::ValueExpr<> > type;
|
||||
};
|
||||
#endif
|
||||
|
||||
// Helper to cleanup the type of the increment:
|
||||
template<typename T> struct cleanup_seq_incr {
|
||||
typedef typename cleanup_index_type<T,DynamicIndex>::type type;
|
||||
};
|
||||
|
||||
}
|
||||
|
||||
//--------------------------------------------------------------------------------
|
||||
// seq(first,last,incr) and seqN(first,size,incr)
|
||||
//--------------------------------------------------------------------------------
|
||||
|
||||
template<typename FirstType=Index,typename SizeType=Index,typename IncrType=internal::FixedInt<1> >
|
||||
class ArithmeticSequence;
|
||||
|
||||
template<typename FirstType,typename SizeType,typename IncrType>
|
||||
ArithmeticSequence<typename internal::cleanup_index_type<FirstType>::type,
|
||||
typename internal::cleanup_index_type<SizeType>::type,
|
||||
typename internal::cleanup_seq_incr<IncrType>::type >
|
||||
seqN(FirstType first, SizeType size, IncrType incr);
|
||||
|
||||
/** \class ArithmeticSequence
|
||||
* \ingroup Core_Module
|
||||
*
|
||||
* This class represents an arithmetic progression \f$ a_0, a_1, a_2, ..., a_{n-1}\f$ defined by
|
||||
* its \em first value \f$ a_0 \f$, its \em size (aka length) \em n, and the \em increment (aka stride)
|
||||
* that is equal to \f$ a_{i+1}-a_{i}\f$ for any \em i.
|
||||
*
|
||||
* It is internally used as the return type of the Eigen::seq and Eigen::seqN functions, and as the input arguments
|
||||
* of DenseBase::operator()(const RowIndices&, const ColIndices&), and most of the time this is the
|
||||
* only way it is used.
|
||||
*
|
||||
* \tparam FirstType type of the first element, usually an Index,
|
||||
* but internally it can be a symbolic expression
|
||||
* \tparam SizeType type representing the size of the sequence, usually an Index
|
||||
* or a compile time integral constant. Internally, it can also be a symbolic expression
|
||||
* \tparam IncrType type of the increment, can be a runtime Index, or a compile time integral constant (default is compile-time 1)
|
||||
*
|
||||
* \sa Eigen::seq, Eigen::seqN, DenseBase::operator()(const RowIndices&, const ColIndices&), class IndexedView
|
||||
*/
|
||||
template<typename FirstType,typename SizeType,typename IncrType>
|
||||
class ArithmeticSequence
|
||||
{
|
||||
public:
|
||||
ArithmeticSequence(FirstType first, SizeType size) : m_first(first), m_size(size) {}
|
||||
ArithmeticSequence(FirstType first, SizeType size, IncrType incr) : m_first(first), m_size(size), m_incr(incr) {}
|
||||
|
||||
enum {
|
||||
SizeAtCompileTime = internal::get_fixed_value<SizeType>::value,
|
||||
IncrAtCompileTime = internal::get_fixed_value<IncrType,DynamicIndex>::value
|
||||
};
|
||||
|
||||
/** \returns the size, i.e., number of elements, of the sequence */
|
||||
Index size() const { return m_size; }
|
||||
|
||||
/** \returns the first element \f$ a_0 \f$ in the sequence */
|
||||
Index first() const { return m_first; }
|
||||
|
||||
/** \returns the value \f$ a_i \f$ at index \a i in the sequence. */
|
||||
Index operator[](Index i) const { return m_first + i * m_incr; }
|
||||
|
||||
const FirstType& firstObject() const { return m_first; }
|
||||
const SizeType& sizeObject() const { return m_size; }
|
||||
const IncrType& incrObject() const { return m_incr; }
|
||||
|
||||
protected:
|
||||
FirstType m_first;
|
||||
SizeType m_size;
|
||||
IncrType m_incr;
|
||||
|
||||
public:
|
||||
|
||||
#if EIGEN_HAS_CXX11 && ((!EIGEN_COMP_GNUC) || EIGEN_COMP_GNUC>=48)
|
||||
auto reverse() const -> decltype(Eigen::seqN(m_first+(m_size+fix<-1>())*m_incr,m_size,-m_incr)) {
|
||||
return seqN(m_first+(m_size+fix<-1>())*m_incr,m_size,-m_incr);
|
||||
}
|
||||
#else
|
||||
protected:
|
||||
typedef typename internal::aseq_negate<IncrType>::type ReverseIncrType;
|
||||
typedef typename internal::aseq_reverse_first_type<FirstType,SizeType,IncrType>::type ReverseFirstType;
|
||||
public:
|
||||
ArithmeticSequence<ReverseFirstType,SizeType,ReverseIncrType>
|
||||
reverse() const {
|
||||
return seqN(m_first+(m_size+fix<-1>())*m_incr,m_size,-m_incr);
|
||||
}
|
||||
#endif
|
||||
};
|
||||
|
||||
/** \returns an ArithmeticSequence starting at \a first, of length \a size, and increment \a incr
|
||||
*
|
||||
* \sa seqN(FirstType,SizeType), seq(FirstType,LastType,IncrType) */
|
||||
template<typename FirstType,typename SizeType,typename IncrType>
|
||||
ArithmeticSequence<typename internal::cleanup_index_type<FirstType>::type,typename internal::cleanup_index_type<SizeType>::type,typename internal::cleanup_seq_incr<IncrType>::type >
|
||||
seqN(FirstType first, SizeType size, IncrType incr) {
|
||||
return ArithmeticSequence<typename internal::cleanup_index_type<FirstType>::type,typename internal::cleanup_index_type<SizeType>::type,typename internal::cleanup_seq_incr<IncrType>::type>(first,size,incr);
|
||||
}
|
||||
|
||||
/** \returns an ArithmeticSequence starting at \a first, of length \a size, and unit increment
|
||||
*
|
||||
* \sa seqN(FirstType,SizeType,IncrType), seq(FirstType,LastType) */
|
||||
template<typename FirstType,typename SizeType>
|
||||
ArithmeticSequence<typename internal::cleanup_index_type<FirstType>::type,typename internal::cleanup_index_type<SizeType>::type >
|
||||
seqN(FirstType first, SizeType size) {
|
||||
return ArithmeticSequence<typename internal::cleanup_index_type<FirstType>::type,typename internal::cleanup_index_type<SizeType>::type>(first,size);
|
||||
}
|
||||
|
||||
#ifdef EIGEN_PARSED_BY_DOXYGEN
|
||||
|
||||
/** \returns an ArithmeticSequence starting at \a f, up (or down) to \a l, and with positive (or negative) increment \a incr
|
||||
*
|
||||
* It is essentially an alias to:
|
||||
* \code
|
||||
* seqN(f, (l-f+incr)/incr, incr);
|
||||
* \endcode
|
||||
*
|
||||
* \sa seqN(FirstType,SizeType,IncrType), seq(FirstType,LastType)
|
||||
*/
|
||||
template<typename FirstType,typename LastType, typename IncrType>
|
||||
auto seq(FirstType f, LastType l, IncrType incr);
|
||||
|
||||
/** \returns an ArithmeticSequence starting at \a f, up (or down) to \a l, and unit increment
|
||||
*
|
||||
* It is essentially an alias to:
|
||||
* \code
|
||||
* seqN(f,l-f+1);
|
||||
* \endcode
|
||||
*
|
||||
* \sa seqN(FirstType,SizeType), seq(FirstType,LastType,IncrType)
|
||||
*/
|
||||
template<typename FirstType,typename LastType>
|
||||
auto seq(FirstType f, LastType l);
|
||||
|
||||
#else // EIGEN_PARSED_BY_DOXYGEN
|
||||
|
||||
#if EIGEN_HAS_CXX11
|
||||
template<typename FirstType,typename LastType>
|
||||
auto seq(FirstType f, LastType l) -> decltype(seqN(typename internal::cleanup_index_type<FirstType>::type(f),
|
||||
( typename internal::cleanup_index_type<LastType>::type(l)
|
||||
- typename internal::cleanup_index_type<FirstType>::type(f)+fix<1>())))
|
||||
{
|
||||
return seqN(typename internal::cleanup_index_type<FirstType>::type(f),
|
||||
(typename internal::cleanup_index_type<LastType>::type(l)
|
||||
-typename internal::cleanup_index_type<FirstType>::type(f)+fix<1>()));
|
||||
}
|
||||
|
||||
template<typename FirstType,typename LastType, typename IncrType>
|
||||
auto seq(FirstType f, LastType l, IncrType incr)
|
||||
-> decltype(seqN(typename internal::cleanup_index_type<FirstType>::type(f),
|
||||
( typename internal::cleanup_index_type<LastType>::type(l)
|
||||
- typename internal::cleanup_index_type<FirstType>::type(f)+typename internal::cleanup_seq_incr<IncrType>::type(incr)
|
||||
) / typename internal::cleanup_seq_incr<IncrType>::type(incr),
|
||||
typename internal::cleanup_seq_incr<IncrType>::type(incr)))
|
||||
{
|
||||
typedef typename internal::cleanup_seq_incr<IncrType>::type CleanedIncrType;
|
||||
return seqN(typename internal::cleanup_index_type<FirstType>::type(f),
|
||||
( typename internal::cleanup_index_type<LastType>::type(l)
|
||||
-typename internal::cleanup_index_type<FirstType>::type(f)+CleanedIncrType(incr)) / CleanedIncrType(incr),
|
||||
CleanedIncrType(incr));
|
||||
}
|
||||
|
||||
#else // EIGEN_HAS_CXX11
|
||||
|
||||
template<typename FirstType,typename LastType>
|
||||
typename internal::enable_if<!(symbolic::is_symbolic<FirstType>::value || symbolic::is_symbolic<LastType>::value),
|
||||
ArithmeticSequence<typename internal::cleanup_index_type<FirstType>::type,Index> >::type
|
||||
seq(FirstType f, LastType l)
|
||||
{
|
||||
return seqN(typename internal::cleanup_index_type<FirstType>::type(f),
|
||||
Index((typename internal::cleanup_index_type<LastType>::type(l)-typename internal::cleanup_index_type<FirstType>::type(f)+fix<1>())));
|
||||
}
|
||||
|
||||
template<typename FirstTypeDerived,typename LastType>
|
||||
typename internal::enable_if<!symbolic::is_symbolic<LastType>::value,
|
||||
ArithmeticSequence<FirstTypeDerived, symbolic::AddExpr<symbolic::AddExpr<symbolic::NegateExpr<FirstTypeDerived>,symbolic::ValueExpr<> >,
|
||||
symbolic::ValueExpr<internal::FixedInt<1> > > > >::type
|
||||
seq(const symbolic::BaseExpr<FirstTypeDerived> &f, LastType l)
|
||||
{
|
||||
return seqN(f.derived(),(typename internal::cleanup_index_type<LastType>::type(l)-f.derived()+fix<1>()));
|
||||
}
|
||||
|
||||
template<typename FirstType,typename LastTypeDerived>
|
||||
typename internal::enable_if<!symbolic::is_symbolic<FirstType>::value,
|
||||
ArithmeticSequence<typename internal::cleanup_index_type<FirstType>::type,
|
||||
symbolic::AddExpr<symbolic::AddExpr<LastTypeDerived,symbolic::ValueExpr<> >,
|
||||
symbolic::ValueExpr<internal::FixedInt<1> > > > >::type
|
||||
seq(FirstType f, const symbolic::BaseExpr<LastTypeDerived> &l)
|
||||
{
|
||||
return seqN(typename internal::cleanup_index_type<FirstType>::type(f),(l.derived()-typename internal::cleanup_index_type<FirstType>::type(f)+fix<1>()));
|
||||
}
|
||||
|
||||
template<typename FirstTypeDerived,typename LastTypeDerived>
|
||||
ArithmeticSequence<FirstTypeDerived,
|
||||
symbolic::AddExpr<symbolic::AddExpr<LastTypeDerived,symbolic::NegateExpr<FirstTypeDerived> >,symbolic::ValueExpr<internal::FixedInt<1> > > >
|
||||
seq(const symbolic::BaseExpr<FirstTypeDerived> &f, const symbolic::BaseExpr<LastTypeDerived> &l)
|
||||
{
|
||||
return seqN(f.derived(),(l.derived()-f.derived()+fix<1>()));
|
||||
}
|
||||
|
||||
|
||||
template<typename FirstType,typename LastType, typename IncrType>
|
||||
typename internal::enable_if<!(symbolic::is_symbolic<FirstType>::value || symbolic::is_symbolic<LastType>::value),
|
||||
ArithmeticSequence<typename internal::cleanup_index_type<FirstType>::type,Index,typename internal::cleanup_seq_incr<IncrType>::type> >::type
|
||||
seq(FirstType f, LastType l, IncrType incr)
|
||||
{
|
||||
typedef typename internal::cleanup_seq_incr<IncrType>::type CleanedIncrType;
|
||||
return seqN(typename internal::cleanup_index_type<FirstType>::type(f),
|
||||
Index((typename internal::cleanup_index_type<LastType>::type(l)-typename internal::cleanup_index_type<FirstType>::type(f)+CleanedIncrType(incr))/CleanedIncrType(incr)), incr);
|
||||
}
|
||||
|
||||
template<typename FirstTypeDerived,typename LastType, typename IncrType>
|
||||
typename internal::enable_if<!symbolic::is_symbolic<LastType>::value,
|
||||
ArithmeticSequence<FirstTypeDerived,
|
||||
symbolic::QuotientExpr<symbolic::AddExpr<symbolic::AddExpr<symbolic::NegateExpr<FirstTypeDerived>,
|
||||
symbolic::ValueExpr<> >,
|
||||
symbolic::ValueExpr<typename internal::cleanup_seq_incr<IncrType>::type> >,
|
||||
symbolic::ValueExpr<typename internal::cleanup_seq_incr<IncrType>::type> >,
|
||||
typename internal::cleanup_seq_incr<IncrType>::type> >::type
|
||||
seq(const symbolic::BaseExpr<FirstTypeDerived> &f, LastType l, IncrType incr)
|
||||
{
|
||||
typedef typename internal::cleanup_seq_incr<IncrType>::type CleanedIncrType;
|
||||
return seqN(f.derived(),(typename internal::cleanup_index_type<LastType>::type(l)-f.derived()+CleanedIncrType(incr))/CleanedIncrType(incr), incr);
|
||||
}
|
||||
|
||||
template<typename FirstType,typename LastTypeDerived, typename IncrType>
|
||||
typename internal::enable_if<!symbolic::is_symbolic<FirstType>::value,
|
||||
ArithmeticSequence<typename internal::cleanup_index_type<FirstType>::type,
|
||||
symbolic::QuotientExpr<symbolic::AddExpr<symbolic::AddExpr<LastTypeDerived,symbolic::ValueExpr<> >,
|
||||
symbolic::ValueExpr<typename internal::cleanup_seq_incr<IncrType>::type> >,
|
||||
symbolic::ValueExpr<typename internal::cleanup_seq_incr<IncrType>::type> >,
|
||||
typename internal::cleanup_seq_incr<IncrType>::type> >::type
|
||||
seq(FirstType f, const symbolic::BaseExpr<LastTypeDerived> &l, IncrType incr)
|
||||
{
|
||||
typedef typename internal::cleanup_seq_incr<IncrType>::type CleanedIncrType;
|
||||
return seqN(typename internal::cleanup_index_type<FirstType>::type(f),
|
||||
(l.derived()-typename internal::cleanup_index_type<FirstType>::type(f)+CleanedIncrType(incr))/CleanedIncrType(incr), incr);
|
||||
}
|
||||
|
||||
template<typename FirstTypeDerived,typename LastTypeDerived, typename IncrType>
|
||||
ArithmeticSequence<FirstTypeDerived,
|
||||
symbolic::QuotientExpr<symbolic::AddExpr<symbolic::AddExpr<LastTypeDerived,
|
||||
symbolic::NegateExpr<FirstTypeDerived> >,
|
||||
symbolic::ValueExpr<typename internal::cleanup_seq_incr<IncrType>::type> >,
|
||||
symbolic::ValueExpr<typename internal::cleanup_seq_incr<IncrType>::type> >,
|
||||
typename internal::cleanup_seq_incr<IncrType>::type>
|
||||
seq(const symbolic::BaseExpr<FirstTypeDerived> &f, const symbolic::BaseExpr<LastTypeDerived> &l, IncrType incr)
|
||||
{
|
||||
typedef typename internal::cleanup_seq_incr<IncrType>::type CleanedIncrType;
|
||||
return seqN(f.derived(),(l.derived()-f.derived()+CleanedIncrType(incr))/CleanedIncrType(incr), incr);
|
||||
}
|
||||
#endif // EIGEN_HAS_CXX11
|
||||
|
||||
#endif // EIGEN_PARSED_BY_DOXYGEN
|
||||
|
||||
|
||||
#if EIGEN_HAS_CXX11 || defined(EIGEN_PARSED_BY_DOXYGEN)
|
||||
/** \cpp11
|
||||
* \returns a symbolic ArithmeticSequence representing the last \a size elements with increment \a incr.
|
||||
*
|
||||
* It is a shortcut for: \code seqN(last-(size-fix<1>)*incr, size, incr) \endcode
|
||||
*
|
||||
* \sa lastN(SizeType), seqN(FirstType,SizeType), seq(FirstType,LastType,IncrType) */
|
||||
template<typename SizeType,typename IncrType>
|
||||
auto lastN(SizeType size, IncrType incr)
|
||||
-> decltype(seqN(Eigen::last-(size-fix<1>())*incr, size, incr))
|
||||
{
|
||||
return seqN(Eigen::last-(size-fix<1>())*incr, size, incr);
|
||||
}
|
||||
|
||||
/** \cpp11
|
||||
* \returns a symbolic ArithmeticSequence representing the last \a size elements with a unit increment.
|
||||
*
|
||||
* It is a shortcut for: \code seq(last+fix<1>-size, last) \endcode
|
||||
*
|
||||
* \sa lastN(SizeType,IncrType, seqN(FirstType,SizeType), seq(FirstType,LastType) */
|
||||
template<typename SizeType>
|
||||
auto lastN(SizeType size)
|
||||
-> decltype(seqN(Eigen::last+fix<1>()-size, size))
|
||||
{
|
||||
return seqN(Eigen::last+fix<1>()-size, size);
|
||||
}
|
||||
#endif
|
||||
|
||||
namespace internal {
|
||||
|
||||
// Convert a symbolic span into a usable one (i.e., remove last/end "keywords")
|
||||
template<typename T>
|
||||
struct make_size_type {
|
||||
typedef typename internal::conditional<symbolic::is_symbolic<T>::value, Index, T>::type type;
|
||||
};
|
||||
|
||||
template<typename FirstType,typename SizeType,typename IncrType,int XprSize>
|
||||
struct IndexedViewCompatibleType<ArithmeticSequence<FirstType,SizeType,IncrType>, XprSize> {
|
||||
typedef ArithmeticSequence<Index,typename make_size_type<SizeType>::type,IncrType> type;
|
||||
};
|
||||
|
||||
template<typename FirstType,typename SizeType,typename IncrType>
|
||||
ArithmeticSequence<Index,typename make_size_type<SizeType>::type,IncrType>
|
||||
makeIndexedViewCompatible(const ArithmeticSequence<FirstType,SizeType,IncrType>& ids, Index size,SpecializedType) {
|
||||
return ArithmeticSequence<Index,typename make_size_type<SizeType>::type,IncrType>(
|
||||
eval_expr_given_size(ids.firstObject(),size),eval_expr_given_size(ids.sizeObject(),size),ids.incrObject());
|
||||
}
|
||||
|
||||
template<typename FirstType,typename SizeType,typename IncrType>
|
||||
struct get_compile_time_incr<ArithmeticSequence<FirstType,SizeType,IncrType> > {
|
||||
enum { value = get_fixed_value<IncrType,DynamicIndex>::value };
|
||||
};
|
||||
|
||||
} // end namespace internal
|
||||
|
||||
/** \namespace Eigen::indexing
|
||||
* \ingroup Core_Module
|
||||
*
|
||||
* The sole purpose of this namespace is to be able to import all functions
|
||||
* and symbols that are expected to be used within operator() for indexing
|
||||
* and slicing. If you already imported the whole Eigen namespace:
|
||||
* \code using namespace Eigen; \endcode
|
||||
* then you are already all set. Otherwise, if you don't want/cannot import
|
||||
* the whole Eigen namespace, the following line:
|
||||
* \code using namespace Eigen::indexing; \endcode
|
||||
* is equivalent to:
|
||||
* \code
|
||||
using Eigen::all;
|
||||
using Eigen::seq;
|
||||
using Eigen::seqN;
|
||||
using Eigen::lastN; // c++11 only
|
||||
using Eigen::last;
|
||||
using Eigen::lastp1;
|
||||
using Eigen::fix;
|
||||
\endcode
|
||||
*/
|
||||
namespace indexing {
|
||||
using Eigen::all;
|
||||
using Eigen::seq;
|
||||
using Eigen::seqN;
|
||||
#if EIGEN_HAS_CXX11
|
||||
using Eigen::lastN;
|
||||
#endif
|
||||
using Eigen::last;
|
||||
using Eigen::lastp1;
|
||||
using Eigen::fix;
|
||||
}
|
||||
|
||||
} // end namespace Eigen
|
||||
|
||||
#endif // EIGEN_ARITHMETIC_SEQUENCE_H
|
||||
@@ -0,0 +1,417 @@
|
||||
// This file is part of Eigen, a lightweight C++ template library
|
||||
// for linear algebra.
|
||||
//
|
||||
// Copyright (C) 2009 Gael Guennebaud <gael.guennebaud@inria.fr>
|
||||
//
|
||||
// This Source Code Form is subject to the terms of the Mozilla
|
||||
// Public License v. 2.0. If a copy of the MPL was not distributed
|
||||
// with this file, You can obtain one at http://mozilla.org/MPL/2.0/.
|
||||
|
||||
#ifndef EIGEN_ARRAY_H
|
||||
#define EIGEN_ARRAY_H
|
||||
|
||||
namespace Eigen {
|
||||
|
||||
namespace internal {
|
||||
template<typename _Scalar, int _Rows, int _Cols, int _Options, int _MaxRows, int _MaxCols>
|
||||
struct traits<Array<_Scalar, _Rows, _Cols, _Options, _MaxRows, _MaxCols> > : traits<Matrix<_Scalar, _Rows, _Cols, _Options, _MaxRows, _MaxCols> >
|
||||
{
|
||||
typedef ArrayXpr XprKind;
|
||||
typedef ArrayBase<Array<_Scalar, _Rows, _Cols, _Options, _MaxRows, _MaxCols> > XprBase;
|
||||
};
|
||||
}
|
||||
|
||||
/** \class Array
|
||||
* \ingroup Core_Module
|
||||
*
|
||||
* \brief General-purpose arrays with easy API for coefficient-wise operations
|
||||
*
|
||||
* The %Array class is very similar to the Matrix class. It provides
|
||||
* general-purpose one- and two-dimensional arrays. The difference between the
|
||||
* %Array and the %Matrix class is primarily in the API: the API for the
|
||||
* %Array class provides easy access to coefficient-wise operations, while the
|
||||
* API for the %Matrix class provides easy access to linear-algebra
|
||||
* operations.
|
||||
*
|
||||
* See documentation of class Matrix for detailed information on the template parameters
|
||||
* storage layout.
|
||||
*
|
||||
* This class can be extended with the help of the plugin mechanism described on the page
|
||||
* \ref TopicCustomizing_Plugins by defining the preprocessor symbol \c EIGEN_ARRAY_PLUGIN.
|
||||
*
|
||||
* \sa \blank \ref TutorialArrayClass, \ref TopicClassHierarchy
|
||||
*/
|
||||
template<typename _Scalar, int _Rows, int _Cols, int _Options, int _MaxRows, int _MaxCols>
|
||||
class Array
|
||||
: public PlainObjectBase<Array<_Scalar, _Rows, _Cols, _Options, _MaxRows, _MaxCols> >
|
||||
{
|
||||
public:
|
||||
|
||||
typedef PlainObjectBase<Array> Base;
|
||||
EIGEN_DENSE_PUBLIC_INTERFACE(Array)
|
||||
|
||||
enum { Options = _Options };
|
||||
typedef typename Base::PlainObject PlainObject;
|
||||
|
||||
protected:
|
||||
template <typename Derived, typename OtherDerived, bool IsVector>
|
||||
friend struct internal::conservative_resize_like_impl;
|
||||
|
||||
using Base::m_storage;
|
||||
|
||||
public:
|
||||
|
||||
using Base::base;
|
||||
using Base::coeff;
|
||||
using Base::coeffRef;
|
||||
|
||||
/**
|
||||
* The usage of
|
||||
* using Base::operator=;
|
||||
* fails on MSVC. Since the code below is working with GCC and MSVC, we skipped
|
||||
* the usage of 'using'. This should be done only for operator=.
|
||||
*/
|
||||
template<typename OtherDerived>
|
||||
EIGEN_DEVICE_FUNC
|
||||
EIGEN_STRONG_INLINE Array& operator=(const EigenBase<OtherDerived> &other)
|
||||
{
|
||||
return Base::operator=(other);
|
||||
}
|
||||
|
||||
/** Set all the entries to \a value.
|
||||
* \sa DenseBase::setConstant(), DenseBase::fill()
|
||||
*/
|
||||
/* This overload is needed because the usage of
|
||||
* using Base::operator=;
|
||||
* fails on MSVC. Since the code below is working with GCC and MSVC, we skipped
|
||||
* the usage of 'using'. This should be done only for operator=.
|
||||
*/
|
||||
EIGEN_DEVICE_FUNC
|
||||
EIGEN_STRONG_INLINE Array& operator=(const Scalar &value)
|
||||
{
|
||||
Base::setConstant(value);
|
||||
return *this;
|
||||
}
|
||||
|
||||
/** Copies the value of the expression \a other into \c *this with automatic resizing.
|
||||
*
|
||||
* *this might be resized to match the dimensions of \a other. If *this was a null matrix (not already initialized),
|
||||
* it will be initialized.
|
||||
*
|
||||
* Note that copying a row-vector into a vector (and conversely) is allowed.
|
||||
* The resizing, if any, is then done in the appropriate way so that row-vectors
|
||||
* remain row-vectors and vectors remain vectors.
|
||||
*/
|
||||
template<typename OtherDerived>
|
||||
EIGEN_DEVICE_FUNC
|
||||
EIGEN_STRONG_INLINE Array& operator=(const DenseBase<OtherDerived>& other)
|
||||
{
|
||||
return Base::_set(other);
|
||||
}
|
||||
|
||||
/** This is a special case of the templated operator=. Its purpose is to
|
||||
* prevent a default operator= from hiding the templated operator=.
|
||||
*/
|
||||
EIGEN_DEVICE_FUNC
|
||||
EIGEN_STRONG_INLINE Array& operator=(const Array& other)
|
||||
{
|
||||
return Base::_set(other);
|
||||
}
|
||||
|
||||
/** Default constructor.
|
||||
*
|
||||
* For fixed-size matrices, does nothing.
|
||||
*
|
||||
* For dynamic-size matrices, creates an empty matrix of size 0. Does not allocate any array. Such a matrix
|
||||
* is called a null matrix. This constructor is the unique way to create null matrices: resizing
|
||||
* a matrix to 0 is not supported.
|
||||
*
|
||||
* \sa resize(Index,Index)
|
||||
*/
|
||||
EIGEN_DEVICE_FUNC
|
||||
EIGEN_STRONG_INLINE Array() : Base()
|
||||
{
|
||||
Base::_check_template_params();
|
||||
EIGEN_INITIALIZE_COEFFS_IF_THAT_OPTION_IS_ENABLED
|
||||
}
|
||||
|
||||
#ifndef EIGEN_PARSED_BY_DOXYGEN
|
||||
// FIXME is it still needed ??
|
||||
/** \internal */
|
||||
EIGEN_DEVICE_FUNC
|
||||
Array(internal::constructor_without_unaligned_array_assert)
|
||||
: Base(internal::constructor_without_unaligned_array_assert())
|
||||
{
|
||||
Base::_check_template_params();
|
||||
EIGEN_INITIALIZE_COEFFS_IF_THAT_OPTION_IS_ENABLED
|
||||
}
|
||||
#endif
|
||||
|
||||
#if EIGEN_HAS_RVALUE_REFERENCES
|
||||
EIGEN_DEVICE_FUNC
|
||||
Array(Array&& other) EIGEN_NOEXCEPT_IF(std::is_nothrow_move_constructible<Scalar>::value)
|
||||
: Base(std::move(other))
|
||||
{
|
||||
Base::_check_template_params();
|
||||
}
|
||||
EIGEN_DEVICE_FUNC
|
||||
Array& operator=(Array&& other) EIGEN_NOEXCEPT_IF(std::is_nothrow_move_assignable<Scalar>::value)
|
||||
{
|
||||
Base::operator=(std::move(other));
|
||||
return *this;
|
||||
}
|
||||
#endif
|
||||
|
||||
#if EIGEN_HAS_CXX11
|
||||
/** \copydoc PlainObjectBase(const Scalar& a0, const Scalar& a1, const Scalar& a2, const Scalar& a3, const ArgTypes&... args)
|
||||
*
|
||||
* Example: \include Array_variadic_ctor_cxx11.cpp
|
||||
* Output: \verbinclude Array_variadic_ctor_cxx11.out
|
||||
*
|
||||
* \sa Array(const std::initializer_list<std::initializer_list<Scalar>>&)
|
||||
* \sa Array(const Scalar&), Array(const Scalar&,const Scalar&)
|
||||
*/
|
||||
template <typename... ArgTypes>
|
||||
EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
|
||||
Array(const Scalar& a0, const Scalar& a1, const Scalar& a2, const Scalar& a3, const ArgTypes&... args)
|
||||
: Base(a0, a1, a2, a3, args...) {}
|
||||
|
||||
/** \brief Constructs an array and initializes it from the coefficients given as initializer-lists grouped by row. \cpp11
|
||||
*
|
||||
* In the general case, the constructor takes a list of rows, each row being represented as a list of coefficients:
|
||||
*
|
||||
* Example: \include Array_initializer_list_23_cxx11.cpp
|
||||
* Output: \verbinclude Array_initializer_list_23_cxx11.out
|
||||
*
|
||||
* Each of the inner initializer lists must contain the exact same number of elements, otherwise an assertion is triggered.
|
||||
*
|
||||
* In the case of a compile-time column 1D array, implicit transposition from a single row is allowed.
|
||||
* Therefore <code> Array<int,Dynamic,1>{{1,2,3,4,5}}</code> is legal and the more verbose syntax
|
||||
* <code>Array<int,Dynamic,1>{{1},{2},{3},{4},{5}}</code> can be avoided:
|
||||
*
|
||||
* Example: \include Array_initializer_list_vector_cxx11.cpp
|
||||
* Output: \verbinclude Array_initializer_list_vector_cxx11.out
|
||||
*
|
||||
* In the case of fixed-sized arrays, the initializer list sizes must exactly match the array sizes,
|
||||
* and implicit transposition is allowed for compile-time 1D arrays only.
|
||||
*
|
||||
* \sa Array(const Scalar& a0, const Scalar& a1, const Scalar& a2, const Scalar& a3, const ArgTypes&... args)
|
||||
*/
|
||||
EIGEN_DEVICE_FUNC
|
||||
EIGEN_STRONG_INLINE Array(const std::initializer_list<std::initializer_list<Scalar>>& list) : Base(list) {}
|
||||
#endif // end EIGEN_HAS_CXX11
|
||||
|
||||
#ifndef EIGEN_PARSED_BY_DOXYGEN
|
||||
template<typename T>
|
||||
EIGEN_DEVICE_FUNC
|
||||
EIGEN_STRONG_INLINE explicit Array(const T& x)
|
||||
{
|
||||
Base::_check_template_params();
|
||||
Base::template _init1<T>(x);
|
||||
}
|
||||
|
||||
template<typename T0, typename T1>
|
||||
EIGEN_DEVICE_FUNC
|
||||
EIGEN_STRONG_INLINE Array(const T0& val0, const T1& val1)
|
||||
{
|
||||
Base::_check_template_params();
|
||||
this->template _init2<T0,T1>(val0, val1);
|
||||
}
|
||||
|
||||
#else
|
||||
/** \brief Constructs a fixed-sized array initialized with coefficients starting at \a data */
|
||||
EIGEN_DEVICE_FUNC explicit Array(const Scalar *data);
|
||||
/** Constructs a vector or row-vector with given dimension. \only_for_vectors
|
||||
*
|
||||
* Note that this is only useful for dynamic-size vectors. For fixed-size vectors,
|
||||
* it is redundant to pass the dimension here, so it makes more sense to use the default
|
||||
* constructor Array() instead.
|
||||
*/
|
||||
EIGEN_DEVICE_FUNC
|
||||
EIGEN_STRONG_INLINE explicit Array(Index dim);
|
||||
/** constructs an initialized 1x1 Array with the given coefficient
|
||||
* \sa const Scalar& a0, const Scalar& a1, const Scalar& a2, const Scalar& a3, const ArgTypes&... args */
|
||||
Array(const Scalar& value);
|
||||
/** constructs an uninitialized array with \a rows rows and \a cols columns.
|
||||
*
|
||||
* This is useful for dynamic-size arrays. For fixed-size arrays,
|
||||
* it is redundant to pass these parameters, so one should use the default constructor
|
||||
* Array() instead. */
|
||||
Array(Index rows, Index cols);
|
||||
/** constructs an initialized 2D vector with given coefficients
|
||||
* \sa Array(const Scalar& a0, const Scalar& a1, const Scalar& a2, const Scalar& a3, const ArgTypes&... args) */
|
||||
Array(const Scalar& val0, const Scalar& val1);
|
||||
#endif // end EIGEN_PARSED_BY_DOXYGEN
|
||||
|
||||
/** constructs an initialized 3D vector with given coefficients
|
||||
* \sa Array(const Scalar& a0, const Scalar& a1, const Scalar& a2, const Scalar& a3, const ArgTypes&... args)
|
||||
*/
|
||||
EIGEN_DEVICE_FUNC
|
||||
EIGEN_STRONG_INLINE Array(const Scalar& val0, const Scalar& val1, const Scalar& val2)
|
||||
{
|
||||
Base::_check_template_params();
|
||||
EIGEN_STATIC_ASSERT_VECTOR_SPECIFIC_SIZE(Array, 3)
|
||||
m_storage.data()[0] = val0;
|
||||
m_storage.data()[1] = val1;
|
||||
m_storage.data()[2] = val2;
|
||||
}
|
||||
/** constructs an initialized 4D vector with given coefficients
|
||||
* \sa Array(const Scalar& a0, const Scalar& a1, const Scalar& a2, const Scalar& a3, const ArgTypes&... args)
|
||||
*/
|
||||
EIGEN_DEVICE_FUNC
|
||||
EIGEN_STRONG_INLINE Array(const Scalar& val0, const Scalar& val1, const Scalar& val2, const Scalar& val3)
|
||||
{
|
||||
Base::_check_template_params();
|
||||
EIGEN_STATIC_ASSERT_VECTOR_SPECIFIC_SIZE(Array, 4)
|
||||
m_storage.data()[0] = val0;
|
||||
m_storage.data()[1] = val1;
|
||||
m_storage.data()[2] = val2;
|
||||
m_storage.data()[3] = val3;
|
||||
}
|
||||
|
||||
/** Copy constructor */
|
||||
EIGEN_DEVICE_FUNC
|
||||
EIGEN_STRONG_INLINE Array(const Array& other)
|
||||
: Base(other)
|
||||
{ }
|
||||
|
||||
private:
|
||||
struct PrivateType {};
|
||||
public:
|
||||
|
||||
/** \sa MatrixBase::operator=(const EigenBase<OtherDerived>&) */
|
||||
template<typename OtherDerived>
|
||||
EIGEN_DEVICE_FUNC
|
||||
EIGEN_STRONG_INLINE Array(const EigenBase<OtherDerived> &other,
|
||||
typename internal::enable_if<internal::is_convertible<typename OtherDerived::Scalar,Scalar>::value,
|
||||
PrivateType>::type = PrivateType())
|
||||
: Base(other.derived())
|
||||
{ }
|
||||
|
||||
EIGEN_DEVICE_FUNC EIGEN_CONSTEXPR
|
||||
inline Index innerStride() const EIGEN_NOEXCEPT{ return 1; }
|
||||
EIGEN_DEVICE_FUNC EIGEN_CONSTEXPR
|
||||
inline Index outerStride() const EIGEN_NOEXCEPT { return this->innerSize(); }
|
||||
|
||||
#ifdef EIGEN_ARRAY_PLUGIN
|
||||
#include EIGEN_ARRAY_PLUGIN
|
||||
#endif
|
||||
|
||||
private:
|
||||
|
||||
template<typename MatrixType, typename OtherDerived, bool SwapPointers>
|
||||
friend struct internal::matrix_swap_impl;
|
||||
};
|
||||
|
||||
/** \defgroup arraytypedefs Global array typedefs
|
||||
* \ingroup Core_Module
|
||||
*
|
||||
* %Eigen defines several typedef shortcuts for most common 1D and 2D array types.
|
||||
*
|
||||
* The general patterns are the following:
|
||||
*
|
||||
* \c ArrayRowsColsType where \c Rows and \c Cols can be \c 2,\c 3,\c 4 for fixed size square matrices or \c X for dynamic size,
|
||||
* and where \c Type can be \c i for integer, \c f for float, \c d for double, \c cf for complex float, \c cd
|
||||
* for complex double.
|
||||
*
|
||||
* For example, \c Array33d is a fixed-size 3x3 array type of doubles, and \c ArrayXXf is a dynamic-size matrix of floats.
|
||||
*
|
||||
* There are also \c ArraySizeType which are self-explanatory. For example, \c Array4cf is
|
||||
* a fixed-size 1D array of 4 complex floats.
|
||||
*
|
||||
* With \cpp11, template alias are also defined for common sizes.
|
||||
* They follow the same pattern as above except that the scalar type suffix is replaced by a
|
||||
* template parameter, i.e.:
|
||||
* - `ArrayRowsCols<Type>` where `Rows` and `Cols` can be \c 2,\c 3,\c 4, or \c X for fixed or dynamic size.
|
||||
* - `ArraySize<Type>` where `Size` can be \c 2,\c 3,\c 4 or \c X for fixed or dynamic size 1D arrays.
|
||||
*
|
||||
* \sa class Array
|
||||
*/
|
||||
|
||||
#define EIGEN_MAKE_ARRAY_TYPEDEFS(Type, TypeSuffix, Size, SizeSuffix) \
|
||||
/** \ingroup arraytypedefs */ \
|
||||
typedef Array<Type, Size, Size> Array##SizeSuffix##SizeSuffix##TypeSuffix; \
|
||||
/** \ingroup arraytypedefs */ \
|
||||
typedef Array<Type, Size, 1> Array##SizeSuffix##TypeSuffix;
|
||||
|
||||
#define EIGEN_MAKE_ARRAY_FIXED_TYPEDEFS(Type, TypeSuffix, Size) \
|
||||
/** \ingroup arraytypedefs */ \
|
||||
typedef Array<Type, Size, Dynamic> Array##Size##X##TypeSuffix; \
|
||||
/** \ingroup arraytypedefs */ \
|
||||
typedef Array<Type, Dynamic, Size> Array##X##Size##TypeSuffix;
|
||||
|
||||
#define EIGEN_MAKE_ARRAY_TYPEDEFS_ALL_SIZES(Type, TypeSuffix) \
|
||||
EIGEN_MAKE_ARRAY_TYPEDEFS(Type, TypeSuffix, 2, 2) \
|
||||
EIGEN_MAKE_ARRAY_TYPEDEFS(Type, TypeSuffix, 3, 3) \
|
||||
EIGEN_MAKE_ARRAY_TYPEDEFS(Type, TypeSuffix, 4, 4) \
|
||||
EIGEN_MAKE_ARRAY_TYPEDEFS(Type, TypeSuffix, Dynamic, X) \
|
||||
EIGEN_MAKE_ARRAY_FIXED_TYPEDEFS(Type, TypeSuffix, 2) \
|
||||
EIGEN_MAKE_ARRAY_FIXED_TYPEDEFS(Type, TypeSuffix, 3) \
|
||||
EIGEN_MAKE_ARRAY_FIXED_TYPEDEFS(Type, TypeSuffix, 4)
|
||||
|
||||
EIGEN_MAKE_ARRAY_TYPEDEFS_ALL_SIZES(int, i)
|
||||
EIGEN_MAKE_ARRAY_TYPEDEFS_ALL_SIZES(float, f)
|
||||
EIGEN_MAKE_ARRAY_TYPEDEFS_ALL_SIZES(double, d)
|
||||
EIGEN_MAKE_ARRAY_TYPEDEFS_ALL_SIZES(std::complex<float>, cf)
|
||||
EIGEN_MAKE_ARRAY_TYPEDEFS_ALL_SIZES(std::complex<double>, cd)
|
||||
|
||||
#undef EIGEN_MAKE_ARRAY_TYPEDEFS_ALL_SIZES
|
||||
#undef EIGEN_MAKE_ARRAY_TYPEDEFS
|
||||
#undef EIGEN_MAKE_ARRAY_FIXED_TYPEDEFS
|
||||
|
||||
#if EIGEN_HAS_CXX11
|
||||
|
||||
#define EIGEN_MAKE_ARRAY_TYPEDEFS(Size, SizeSuffix) \
|
||||
/** \ingroup arraytypedefs */ \
|
||||
/** \brief \cpp11 */ \
|
||||
template <typename Type> \
|
||||
using Array##SizeSuffix##SizeSuffix = Array<Type, Size, Size>; \
|
||||
/** \ingroup arraytypedefs */ \
|
||||
/** \brief \cpp11 */ \
|
||||
template <typename Type> \
|
||||
using Array##SizeSuffix = Array<Type, Size, 1>;
|
||||
|
||||
#define EIGEN_MAKE_ARRAY_FIXED_TYPEDEFS(Size) \
|
||||
/** \ingroup arraytypedefs */ \
|
||||
/** \brief \cpp11 */ \
|
||||
template <typename Type> \
|
||||
using Array##Size##X = Array<Type, Size, Dynamic>; \
|
||||
/** \ingroup arraytypedefs */ \
|
||||
/** \brief \cpp11 */ \
|
||||
template <typename Type> \
|
||||
using Array##X##Size = Array<Type, Dynamic, Size>;
|
||||
|
||||
EIGEN_MAKE_ARRAY_TYPEDEFS(2, 2)
|
||||
EIGEN_MAKE_ARRAY_TYPEDEFS(3, 3)
|
||||
EIGEN_MAKE_ARRAY_TYPEDEFS(4, 4)
|
||||
EIGEN_MAKE_ARRAY_TYPEDEFS(Dynamic, X)
|
||||
EIGEN_MAKE_ARRAY_FIXED_TYPEDEFS(2)
|
||||
EIGEN_MAKE_ARRAY_FIXED_TYPEDEFS(3)
|
||||
EIGEN_MAKE_ARRAY_FIXED_TYPEDEFS(4)
|
||||
|
||||
#undef EIGEN_MAKE_ARRAY_TYPEDEFS
|
||||
#undef EIGEN_MAKE_ARRAY_FIXED_TYPEDEFS
|
||||
|
||||
#endif // EIGEN_HAS_CXX11
|
||||
|
||||
#define EIGEN_USING_ARRAY_TYPEDEFS_FOR_TYPE_AND_SIZE(TypeSuffix, SizeSuffix) \
|
||||
using Eigen::Matrix##SizeSuffix##TypeSuffix; \
|
||||
using Eigen::Vector##SizeSuffix##TypeSuffix; \
|
||||
using Eigen::RowVector##SizeSuffix##TypeSuffix;
|
||||
|
||||
#define EIGEN_USING_ARRAY_TYPEDEFS_FOR_TYPE(TypeSuffix) \
|
||||
EIGEN_USING_ARRAY_TYPEDEFS_FOR_TYPE_AND_SIZE(TypeSuffix, 2) \
|
||||
EIGEN_USING_ARRAY_TYPEDEFS_FOR_TYPE_AND_SIZE(TypeSuffix, 3) \
|
||||
EIGEN_USING_ARRAY_TYPEDEFS_FOR_TYPE_AND_SIZE(TypeSuffix, 4) \
|
||||
EIGEN_USING_ARRAY_TYPEDEFS_FOR_TYPE_AND_SIZE(TypeSuffix, X) \
|
||||
|
||||
#define EIGEN_USING_ARRAY_TYPEDEFS \
|
||||
EIGEN_USING_ARRAY_TYPEDEFS_FOR_TYPE(i) \
|
||||
EIGEN_USING_ARRAY_TYPEDEFS_FOR_TYPE(f) \
|
||||
EIGEN_USING_ARRAY_TYPEDEFS_FOR_TYPE(d) \
|
||||
EIGEN_USING_ARRAY_TYPEDEFS_FOR_TYPE(cf) \
|
||||
EIGEN_USING_ARRAY_TYPEDEFS_FOR_TYPE(cd)
|
||||
|
||||
} // end namespace Eigen
|
||||
|
||||
#endif // EIGEN_ARRAY_H
|
||||
@@ -0,0 +1,226 @@
|
||||
// This file is part of Eigen, a lightweight C++ template library
|
||||
// for linear algebra.
|
||||
//
|
||||
// Copyright (C) 2009 Gael Guennebaud <gael.guennebaud@inria.fr>
|
||||
//
|
||||
// This Source Code Form is subject to the terms of the Mozilla
|
||||
// Public License v. 2.0. If a copy of the MPL was not distributed
|
||||
// with this file, You can obtain one at http://mozilla.org/MPL/2.0/.
|
||||
|
||||
#ifndef EIGEN_ARRAYBASE_H
|
||||
#define EIGEN_ARRAYBASE_H
|
||||
|
||||
namespace Eigen {
|
||||
|
||||
template<typename ExpressionType> class MatrixWrapper;
|
||||
|
||||
/** \class ArrayBase
|
||||
* \ingroup Core_Module
|
||||
*
|
||||
* \brief Base class for all 1D and 2D array, and related expressions
|
||||
*
|
||||
* An array is similar to a dense vector or matrix. While matrices are mathematical
|
||||
* objects with well defined linear algebra operators, an array is just a collection
|
||||
* of scalar values arranged in a one or two dimensionnal fashion. As the main consequence,
|
||||
* all operations applied to an array are performed coefficient wise. Furthermore,
|
||||
* arrays support scalar math functions of the c++ standard library (e.g., std::sin(x)), and convenient
|
||||
* constructors allowing to easily write generic code working for both scalar values
|
||||
* and arrays.
|
||||
*
|
||||
* This class is the base that is inherited by all array expression types.
|
||||
*
|
||||
* \tparam Derived is the derived type, e.g., an array or an expression type.
|
||||
*
|
||||
* This class can be extended with the help of the plugin mechanism described on the page
|
||||
* \ref TopicCustomizing_Plugins by defining the preprocessor symbol \c EIGEN_ARRAYBASE_PLUGIN.
|
||||
*
|
||||
* \sa class MatrixBase, \ref TopicClassHierarchy
|
||||
*/
|
||||
template<typename Derived> class ArrayBase
|
||||
: public DenseBase<Derived>
|
||||
{
|
||||
public:
|
||||
#ifndef EIGEN_PARSED_BY_DOXYGEN
|
||||
/** The base class for a given storage type. */
|
||||
typedef ArrayBase StorageBaseType;
|
||||
|
||||
typedef ArrayBase Eigen_BaseClassForSpecializationOfGlobalMathFuncImpl;
|
||||
|
||||
typedef typename internal::traits<Derived>::StorageKind StorageKind;
|
||||
typedef typename internal::traits<Derived>::Scalar Scalar;
|
||||
typedef typename internal::packet_traits<Scalar>::type PacketScalar;
|
||||
typedef typename NumTraits<Scalar>::Real RealScalar;
|
||||
|
||||
typedef DenseBase<Derived> Base;
|
||||
using Base::RowsAtCompileTime;
|
||||
using Base::ColsAtCompileTime;
|
||||
using Base::SizeAtCompileTime;
|
||||
using Base::MaxRowsAtCompileTime;
|
||||
using Base::MaxColsAtCompileTime;
|
||||
using Base::MaxSizeAtCompileTime;
|
||||
using Base::IsVectorAtCompileTime;
|
||||
using Base::Flags;
|
||||
|
||||
using Base::derived;
|
||||
using Base::const_cast_derived;
|
||||
using Base::rows;
|
||||
using Base::cols;
|
||||
using Base::size;
|
||||
using Base::coeff;
|
||||
using Base::coeffRef;
|
||||
using Base::lazyAssign;
|
||||
using Base::operator-;
|
||||
using Base::operator=;
|
||||
using Base::operator+=;
|
||||
using Base::operator-=;
|
||||
using Base::operator*=;
|
||||
using Base::operator/=;
|
||||
|
||||
typedef typename Base::CoeffReturnType CoeffReturnType;
|
||||
|
||||
#endif // not EIGEN_PARSED_BY_DOXYGEN
|
||||
|
||||
#ifndef EIGEN_PARSED_BY_DOXYGEN
|
||||
typedef typename Base::PlainObject PlainObject;
|
||||
|
||||
/** \internal Represents a matrix with all coefficients equal to one another*/
|
||||
typedef CwiseNullaryOp<internal::scalar_constant_op<Scalar>,PlainObject> ConstantReturnType;
|
||||
#endif // not EIGEN_PARSED_BY_DOXYGEN
|
||||
|
||||
#define EIGEN_CURRENT_STORAGE_BASE_CLASS Eigen::ArrayBase
|
||||
#define EIGEN_DOC_UNARY_ADDONS(X,Y)
|
||||
# include "../plugins/MatrixCwiseUnaryOps.h"
|
||||
# include "../plugins/ArrayCwiseUnaryOps.h"
|
||||
# include "../plugins/CommonCwiseBinaryOps.h"
|
||||
# include "../plugins/MatrixCwiseBinaryOps.h"
|
||||
# include "../plugins/ArrayCwiseBinaryOps.h"
|
||||
# ifdef EIGEN_ARRAYBASE_PLUGIN
|
||||
# include EIGEN_ARRAYBASE_PLUGIN
|
||||
# endif
|
||||
#undef EIGEN_CURRENT_STORAGE_BASE_CLASS
|
||||
#undef EIGEN_DOC_UNARY_ADDONS
|
||||
|
||||
/** Special case of the template operator=, in order to prevent the compiler
|
||||
* from generating a default operator= (issue hit with g++ 4.1)
|
||||
*/
|
||||
EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
|
||||
Derived& operator=(const ArrayBase& other)
|
||||
{
|
||||
internal::call_assignment(derived(), other.derived());
|
||||
return derived();
|
||||
}
|
||||
|
||||
/** Set all the entries to \a value.
|
||||
* \sa DenseBase::setConstant(), DenseBase::fill() */
|
||||
EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
|
||||
Derived& operator=(const Scalar &value)
|
||||
{ Base::setConstant(value); return derived(); }
|
||||
|
||||
EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
|
||||
Derived& operator+=(const Scalar& scalar);
|
||||
EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
|
||||
Derived& operator-=(const Scalar& scalar);
|
||||
|
||||
template<typename OtherDerived>
|
||||
EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
|
||||
Derived& operator+=(const ArrayBase<OtherDerived>& other);
|
||||
template<typename OtherDerived>
|
||||
EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
|
||||
Derived& operator-=(const ArrayBase<OtherDerived>& other);
|
||||
|
||||
template<typename OtherDerived>
|
||||
EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
|
||||
Derived& operator*=(const ArrayBase<OtherDerived>& other);
|
||||
|
||||
template<typename OtherDerived>
|
||||
EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
|
||||
Derived& operator/=(const ArrayBase<OtherDerived>& other);
|
||||
|
||||
public:
|
||||
EIGEN_DEVICE_FUNC
|
||||
ArrayBase<Derived>& array() { return *this; }
|
||||
EIGEN_DEVICE_FUNC
|
||||
const ArrayBase<Derived>& array() const { return *this; }
|
||||
|
||||
/** \returns an \link Eigen::MatrixBase Matrix \endlink expression of this array
|
||||
* \sa MatrixBase::array() */
|
||||
EIGEN_DEVICE_FUNC
|
||||
MatrixWrapper<Derived> matrix() { return MatrixWrapper<Derived>(derived()); }
|
||||
EIGEN_DEVICE_FUNC
|
||||
const MatrixWrapper<const Derived> matrix() const { return MatrixWrapper<const Derived>(derived()); }
|
||||
|
||||
// template<typename Dest>
|
||||
// inline void evalTo(Dest& dst) const { dst = matrix(); }
|
||||
|
||||
protected:
|
||||
EIGEN_DEFAULT_COPY_CONSTRUCTOR(ArrayBase)
|
||||
EIGEN_DEFAULT_EMPTY_CONSTRUCTOR_AND_DESTRUCTOR(ArrayBase)
|
||||
|
||||
private:
|
||||
explicit ArrayBase(Index);
|
||||
ArrayBase(Index,Index);
|
||||
template<typename OtherDerived> explicit ArrayBase(const ArrayBase<OtherDerived>&);
|
||||
protected:
|
||||
// mixing arrays and matrices is not legal
|
||||
template<typename OtherDerived> Derived& operator+=(const MatrixBase<OtherDerived>& )
|
||||
{EIGEN_STATIC_ASSERT(std::ptrdiff_t(sizeof(typename OtherDerived::Scalar))==-1,YOU_CANNOT_MIX_ARRAYS_AND_MATRICES); return *this;}
|
||||
// mixing arrays and matrices is not legal
|
||||
template<typename OtherDerived> Derived& operator-=(const MatrixBase<OtherDerived>& )
|
||||
{EIGEN_STATIC_ASSERT(std::ptrdiff_t(sizeof(typename OtherDerived::Scalar))==-1,YOU_CANNOT_MIX_ARRAYS_AND_MATRICES); return *this;}
|
||||
};
|
||||
|
||||
/** replaces \c *this by \c *this - \a other.
|
||||
*
|
||||
* \returns a reference to \c *this
|
||||
*/
|
||||
template<typename Derived>
|
||||
template<typename OtherDerived>
|
||||
EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE Derived &
|
||||
ArrayBase<Derived>::operator-=(const ArrayBase<OtherDerived> &other)
|
||||
{
|
||||
call_assignment(derived(), other.derived(), internal::sub_assign_op<Scalar,typename OtherDerived::Scalar>());
|
||||
return derived();
|
||||
}
|
||||
|
||||
/** replaces \c *this by \c *this + \a other.
|
||||
*
|
||||
* \returns a reference to \c *this
|
||||
*/
|
||||
template<typename Derived>
|
||||
template<typename OtherDerived>
|
||||
EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE Derived &
|
||||
ArrayBase<Derived>::operator+=(const ArrayBase<OtherDerived>& other)
|
||||
{
|
||||
call_assignment(derived(), other.derived(), internal::add_assign_op<Scalar,typename OtherDerived::Scalar>());
|
||||
return derived();
|
||||
}
|
||||
|
||||
/** replaces \c *this by \c *this * \a other coefficient wise.
|
||||
*
|
||||
* \returns a reference to \c *this
|
||||
*/
|
||||
template<typename Derived>
|
||||
template<typename OtherDerived>
|
||||
EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE Derived &
|
||||
ArrayBase<Derived>::operator*=(const ArrayBase<OtherDerived>& other)
|
||||
{
|
||||
call_assignment(derived(), other.derived(), internal::mul_assign_op<Scalar,typename OtherDerived::Scalar>());
|
||||
return derived();
|
||||
}
|
||||
|
||||
/** replaces \c *this by \c *this / \a other coefficient wise.
|
||||
*
|
||||
* \returns a reference to \c *this
|
||||
*/
|
||||
template<typename Derived>
|
||||
template<typename OtherDerived>
|
||||
EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE Derived &
|
||||
ArrayBase<Derived>::operator/=(const ArrayBase<OtherDerived>& other)
|
||||
{
|
||||
call_assignment(derived(), other.derived(), internal::div_assign_op<Scalar,typename OtherDerived::Scalar>());
|
||||
return derived();
|
||||
}
|
||||
|
||||
} // end namespace Eigen
|
||||
|
||||
#endif // EIGEN_ARRAYBASE_H
|
||||
@@ -0,0 +1,209 @@
|
||||
// This file is part of Eigen, a lightweight C++ template library
|
||||
// for linear algebra.
|
||||
//
|
||||
// Copyright (C) 2009-2010 Gael Guennebaud <gael.guennebaud@inria.fr>
|
||||
//
|
||||
// This Source Code Form is subject to the terms of the Mozilla
|
||||
// Public License v. 2.0. If a copy of the MPL was not distributed
|
||||
// with this file, You can obtain one at http://mozilla.org/MPL/2.0/.
|
||||
|
||||
#ifndef EIGEN_ARRAYWRAPPER_H
|
||||
#define EIGEN_ARRAYWRAPPER_H
|
||||
|
||||
namespace Eigen {
|
||||
|
||||
/** \class ArrayWrapper
|
||||
* \ingroup Core_Module
|
||||
*
|
||||
* \brief Expression of a mathematical vector or matrix as an array object
|
||||
*
|
||||
* This class is the return type of MatrixBase::array(), and most of the time
|
||||
* this is the only way it is use.
|
||||
*
|
||||
* \sa MatrixBase::array(), class MatrixWrapper
|
||||
*/
|
||||
|
||||
namespace internal {
|
||||
template<typename ExpressionType>
|
||||
struct traits<ArrayWrapper<ExpressionType> >
|
||||
: public traits<typename remove_all<typename ExpressionType::Nested>::type >
|
||||
{
|
||||
typedef ArrayXpr XprKind;
|
||||
// Let's remove NestByRefBit
|
||||
enum {
|
||||
Flags0 = traits<typename remove_all<typename ExpressionType::Nested>::type >::Flags,
|
||||
LvalueBitFlag = is_lvalue<ExpressionType>::value ? LvalueBit : 0,
|
||||
Flags = (Flags0 & ~(NestByRefBit | LvalueBit)) | LvalueBitFlag
|
||||
};
|
||||
};
|
||||
}
|
||||
|
||||
template<typename ExpressionType>
|
||||
class ArrayWrapper : public ArrayBase<ArrayWrapper<ExpressionType> >
|
||||
{
|
||||
public:
|
||||
typedef ArrayBase<ArrayWrapper> Base;
|
||||
EIGEN_DENSE_PUBLIC_INTERFACE(ArrayWrapper)
|
||||
EIGEN_INHERIT_ASSIGNMENT_OPERATORS(ArrayWrapper)
|
||||
typedef typename internal::remove_all<ExpressionType>::type NestedExpression;
|
||||
|
||||
typedef typename internal::conditional<
|
||||
internal::is_lvalue<ExpressionType>::value,
|
||||
Scalar,
|
||||
const Scalar
|
||||
>::type ScalarWithConstIfNotLvalue;
|
||||
|
||||
typedef typename internal::ref_selector<ExpressionType>::non_const_type NestedExpressionType;
|
||||
|
||||
using Base::coeffRef;
|
||||
|
||||
EIGEN_DEVICE_FUNC
|
||||
explicit EIGEN_STRONG_INLINE ArrayWrapper(ExpressionType& matrix) : m_expression(matrix) {}
|
||||
|
||||
EIGEN_DEVICE_FUNC EIGEN_CONSTEXPR
|
||||
inline Index rows() const EIGEN_NOEXCEPT { return m_expression.rows(); }
|
||||
EIGEN_DEVICE_FUNC EIGEN_CONSTEXPR
|
||||
inline Index cols() const EIGEN_NOEXCEPT { return m_expression.cols(); }
|
||||
EIGEN_DEVICE_FUNC EIGEN_CONSTEXPR
|
||||
inline Index outerStride() const EIGEN_NOEXCEPT { return m_expression.outerStride(); }
|
||||
EIGEN_DEVICE_FUNC EIGEN_CONSTEXPR
|
||||
inline Index innerStride() const EIGEN_NOEXCEPT { return m_expression.innerStride(); }
|
||||
|
||||
EIGEN_DEVICE_FUNC
|
||||
inline ScalarWithConstIfNotLvalue* data() { return m_expression.data(); }
|
||||
EIGEN_DEVICE_FUNC
|
||||
inline const Scalar* data() const { return m_expression.data(); }
|
||||
|
||||
EIGEN_DEVICE_FUNC
|
||||
inline const Scalar& coeffRef(Index rowId, Index colId) const
|
||||
{
|
||||
return m_expression.coeffRef(rowId, colId);
|
||||
}
|
||||
|
||||
EIGEN_DEVICE_FUNC
|
||||
inline const Scalar& coeffRef(Index index) const
|
||||
{
|
||||
return m_expression.coeffRef(index);
|
||||
}
|
||||
|
||||
template<typename Dest>
|
||||
EIGEN_DEVICE_FUNC
|
||||
inline void evalTo(Dest& dst) const { dst = m_expression; }
|
||||
|
||||
EIGEN_DEVICE_FUNC
|
||||
const typename internal::remove_all<NestedExpressionType>::type&
|
||||
nestedExpression() const
|
||||
{
|
||||
return m_expression;
|
||||
}
|
||||
|
||||
/** Forwards the resizing request to the nested expression
|
||||
* \sa DenseBase::resize(Index) */
|
||||
EIGEN_DEVICE_FUNC
|
||||
void resize(Index newSize) { m_expression.resize(newSize); }
|
||||
/** Forwards the resizing request to the nested expression
|
||||
* \sa DenseBase::resize(Index,Index)*/
|
||||
EIGEN_DEVICE_FUNC
|
||||
void resize(Index rows, Index cols) { m_expression.resize(rows,cols); }
|
||||
|
||||
protected:
|
||||
NestedExpressionType m_expression;
|
||||
};
|
||||
|
||||
/** \class MatrixWrapper
|
||||
* \ingroup Core_Module
|
||||
*
|
||||
* \brief Expression of an array as a mathematical vector or matrix
|
||||
*
|
||||
* This class is the return type of ArrayBase::matrix(), and most of the time
|
||||
* this is the only way it is use.
|
||||
*
|
||||
* \sa MatrixBase::matrix(), class ArrayWrapper
|
||||
*/
|
||||
|
||||
namespace internal {
|
||||
template<typename ExpressionType>
|
||||
struct traits<MatrixWrapper<ExpressionType> >
|
||||
: public traits<typename remove_all<typename ExpressionType::Nested>::type >
|
||||
{
|
||||
typedef MatrixXpr XprKind;
|
||||
// Let's remove NestByRefBit
|
||||
enum {
|
||||
Flags0 = traits<typename remove_all<typename ExpressionType::Nested>::type >::Flags,
|
||||
LvalueBitFlag = is_lvalue<ExpressionType>::value ? LvalueBit : 0,
|
||||
Flags = (Flags0 & ~(NestByRefBit | LvalueBit)) | LvalueBitFlag
|
||||
};
|
||||
};
|
||||
}
|
||||
|
||||
template<typename ExpressionType>
|
||||
class MatrixWrapper : public MatrixBase<MatrixWrapper<ExpressionType> >
|
||||
{
|
||||
public:
|
||||
typedef MatrixBase<MatrixWrapper<ExpressionType> > Base;
|
||||
EIGEN_DENSE_PUBLIC_INTERFACE(MatrixWrapper)
|
||||
EIGEN_INHERIT_ASSIGNMENT_OPERATORS(MatrixWrapper)
|
||||
typedef typename internal::remove_all<ExpressionType>::type NestedExpression;
|
||||
|
||||
typedef typename internal::conditional<
|
||||
internal::is_lvalue<ExpressionType>::value,
|
||||
Scalar,
|
||||
const Scalar
|
||||
>::type ScalarWithConstIfNotLvalue;
|
||||
|
||||
typedef typename internal::ref_selector<ExpressionType>::non_const_type NestedExpressionType;
|
||||
|
||||
using Base::coeffRef;
|
||||
|
||||
EIGEN_DEVICE_FUNC
|
||||
explicit inline MatrixWrapper(ExpressionType& matrix) : m_expression(matrix) {}
|
||||
|
||||
EIGEN_DEVICE_FUNC EIGEN_CONSTEXPR
|
||||
inline Index rows() const EIGEN_NOEXCEPT { return m_expression.rows(); }
|
||||
EIGEN_DEVICE_FUNC EIGEN_CONSTEXPR
|
||||
inline Index cols() const EIGEN_NOEXCEPT { return m_expression.cols(); }
|
||||
EIGEN_DEVICE_FUNC EIGEN_CONSTEXPR
|
||||
inline Index outerStride() const EIGEN_NOEXCEPT { return m_expression.outerStride(); }
|
||||
EIGEN_DEVICE_FUNC EIGEN_CONSTEXPR
|
||||
inline Index innerStride() const EIGEN_NOEXCEPT { return m_expression.innerStride(); }
|
||||
|
||||
EIGEN_DEVICE_FUNC
|
||||
inline ScalarWithConstIfNotLvalue* data() { return m_expression.data(); }
|
||||
EIGEN_DEVICE_FUNC
|
||||
inline const Scalar* data() const { return m_expression.data(); }
|
||||
|
||||
EIGEN_DEVICE_FUNC
|
||||
inline const Scalar& coeffRef(Index rowId, Index colId) const
|
||||
{
|
||||
return m_expression.derived().coeffRef(rowId, colId);
|
||||
}
|
||||
|
||||
EIGEN_DEVICE_FUNC
|
||||
inline const Scalar& coeffRef(Index index) const
|
||||
{
|
||||
return m_expression.coeffRef(index);
|
||||
}
|
||||
|
||||
EIGEN_DEVICE_FUNC
|
||||
const typename internal::remove_all<NestedExpressionType>::type&
|
||||
nestedExpression() const
|
||||
{
|
||||
return m_expression;
|
||||
}
|
||||
|
||||
/** Forwards the resizing request to the nested expression
|
||||
* \sa DenseBase::resize(Index) */
|
||||
EIGEN_DEVICE_FUNC
|
||||
void resize(Index newSize) { m_expression.resize(newSize); }
|
||||
/** Forwards the resizing request to the nested expression
|
||||
* \sa DenseBase::resize(Index,Index)*/
|
||||
EIGEN_DEVICE_FUNC
|
||||
void resize(Index rows, Index cols) { m_expression.resize(rows,cols); }
|
||||
|
||||
protected:
|
||||
NestedExpressionType m_expression;
|
||||
};
|
||||
|
||||
} // end namespace Eigen
|
||||
|
||||
#endif // EIGEN_ARRAYWRAPPER_H
|
||||
@@ -0,0 +1,90 @@
|
||||
// This file is part of Eigen, a lightweight C++ template library
|
||||
// for linear algebra.
|
||||
//
|
||||
// Copyright (C) 2007 Michael Olbrich <michael.olbrich@gmx.net>
|
||||
// Copyright (C) 2006-2010 Benoit Jacob <jacob.benoit.1@gmail.com>
|
||||
// Copyright (C) 2008 Gael Guennebaud <gael.guennebaud@inria.fr>
|
||||
//
|
||||
// This Source Code Form is subject to the terms of the Mozilla
|
||||
// Public License v. 2.0. If a copy of the MPL was not distributed
|
||||
// with this file, You can obtain one at http://mozilla.org/MPL/2.0/.
|
||||
|
||||
#ifndef EIGEN_ASSIGN_H
|
||||
#define EIGEN_ASSIGN_H
|
||||
|
||||
namespace Eigen {
|
||||
|
||||
template<typename Derived>
|
||||
template<typename OtherDerived>
|
||||
EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE Derived& DenseBase<Derived>
|
||||
::lazyAssign(const DenseBase<OtherDerived>& other)
|
||||
{
|
||||
enum{
|
||||
SameType = internal::is_same<typename Derived::Scalar,typename OtherDerived::Scalar>::value
|
||||
};
|
||||
|
||||
EIGEN_STATIC_ASSERT_LVALUE(Derived)
|
||||
EIGEN_STATIC_ASSERT_SAME_MATRIX_SIZE(Derived,OtherDerived)
|
||||
EIGEN_STATIC_ASSERT(SameType,YOU_MIXED_DIFFERENT_NUMERIC_TYPES__YOU_NEED_TO_USE_THE_CAST_METHOD_OF_MATRIXBASE_TO_CAST_NUMERIC_TYPES_EXPLICITLY)
|
||||
|
||||
eigen_assert(rows() == other.rows() && cols() == other.cols());
|
||||
internal::call_assignment_no_alias(derived(),other.derived());
|
||||
|
||||
return derived();
|
||||
}
|
||||
|
||||
template<typename Derived>
|
||||
template<typename OtherDerived>
|
||||
EIGEN_DEVICE_FUNC
|
||||
EIGEN_STRONG_INLINE Derived& DenseBase<Derived>::operator=(const DenseBase<OtherDerived>& other)
|
||||
{
|
||||
internal::call_assignment(derived(), other.derived());
|
||||
return derived();
|
||||
}
|
||||
|
||||
template<typename Derived>
|
||||
EIGEN_DEVICE_FUNC
|
||||
EIGEN_STRONG_INLINE Derived& DenseBase<Derived>::operator=(const DenseBase& other)
|
||||
{
|
||||
internal::call_assignment(derived(), other.derived());
|
||||
return derived();
|
||||
}
|
||||
|
||||
template<typename Derived>
|
||||
EIGEN_DEVICE_FUNC
|
||||
EIGEN_STRONG_INLINE Derived& MatrixBase<Derived>::operator=(const MatrixBase& other)
|
||||
{
|
||||
internal::call_assignment(derived(), other.derived());
|
||||
return derived();
|
||||
}
|
||||
|
||||
template<typename Derived>
|
||||
template <typename OtherDerived>
|
||||
EIGEN_DEVICE_FUNC
|
||||
EIGEN_STRONG_INLINE Derived& MatrixBase<Derived>::operator=(const DenseBase<OtherDerived>& other)
|
||||
{
|
||||
internal::call_assignment(derived(), other.derived());
|
||||
return derived();
|
||||
}
|
||||
|
||||
template<typename Derived>
|
||||
template <typename OtherDerived>
|
||||
EIGEN_DEVICE_FUNC
|
||||
EIGEN_STRONG_INLINE Derived& MatrixBase<Derived>::operator=(const EigenBase<OtherDerived>& other)
|
||||
{
|
||||
internal::call_assignment(derived(), other.derived());
|
||||
return derived();
|
||||
}
|
||||
|
||||
template<typename Derived>
|
||||
template<typename OtherDerived>
|
||||
EIGEN_DEVICE_FUNC
|
||||
EIGEN_STRONG_INLINE Derived& MatrixBase<Derived>::operator=(const ReturnByValue<OtherDerived>& other)
|
||||
{
|
||||
other.derived().evalTo(derived());
|
||||
return derived();
|
||||
}
|
||||
|
||||
} // end namespace Eigen
|
||||
|
||||
#endif // EIGEN_ASSIGN_H
|
||||
File diff suppressed because it is too large
Load Diff
Executable
+178
@@ -0,0 +1,178 @@
|
||||
/*
|
||||
Copyright (c) 2011, Intel Corporation. All rights reserved.
|
||||
Copyright (C) 2015 Gael Guennebaud <gael.guennebaud@inria.fr>
|
||||
|
||||
Redistribution and use in source and binary forms, with or without modification,
|
||||
are permitted provided that the following conditions are met:
|
||||
|
||||
* Redistributions of source code must retain the above copyright notice, this
|
||||
list of conditions and the following disclaimer.
|
||||
* Redistributions in binary form must reproduce the above copyright notice,
|
||||
this list of conditions and the following disclaimer in the documentation
|
||||
and/or other materials provided with the distribution.
|
||||
* Neither the name of Intel Corporation nor the names of its contributors may
|
||||
be used to endorse or promote products derived from this software without
|
||||
specific prior written permission.
|
||||
|
||||
THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" AND
|
||||
ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED
|
||||
WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE
|
||||
DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE LIABLE FOR
|
||||
ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES
|
||||
(INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES;
|
||||
LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON
|
||||
ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
|
||||
(INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS
|
||||
SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
|
||||
|
||||
********************************************************************************
|
||||
* Content : Eigen bindings to Intel(R) MKL
|
||||
* MKL VML support for coefficient-wise unary Eigen expressions like a=b.sin()
|
||||
********************************************************************************
|
||||
*/
|
||||
|
||||
#ifndef EIGEN_ASSIGN_VML_H
|
||||
#define EIGEN_ASSIGN_VML_H
|
||||
|
||||
namespace Eigen {
|
||||
|
||||
namespace internal {
|
||||
|
||||
template<typename Dst, typename Src>
|
||||
class vml_assign_traits
|
||||
{
|
||||
private:
|
||||
enum {
|
||||
DstHasDirectAccess = Dst::Flags & DirectAccessBit,
|
||||
SrcHasDirectAccess = Src::Flags & DirectAccessBit,
|
||||
StorageOrdersAgree = (int(Dst::IsRowMajor) == int(Src::IsRowMajor)),
|
||||
InnerSize = int(Dst::IsVectorAtCompileTime) ? int(Dst::SizeAtCompileTime)
|
||||
: int(Dst::Flags)&RowMajorBit ? int(Dst::ColsAtCompileTime)
|
||||
: int(Dst::RowsAtCompileTime),
|
||||
InnerMaxSize = int(Dst::IsVectorAtCompileTime) ? int(Dst::MaxSizeAtCompileTime)
|
||||
: int(Dst::Flags)&RowMajorBit ? int(Dst::MaxColsAtCompileTime)
|
||||
: int(Dst::MaxRowsAtCompileTime),
|
||||
MaxSizeAtCompileTime = Dst::SizeAtCompileTime,
|
||||
|
||||
MightEnableVml = StorageOrdersAgree && DstHasDirectAccess && SrcHasDirectAccess && Src::InnerStrideAtCompileTime==1 && Dst::InnerStrideAtCompileTime==1,
|
||||
MightLinearize = MightEnableVml && (int(Dst::Flags) & int(Src::Flags) & LinearAccessBit),
|
||||
VmlSize = MightLinearize ? MaxSizeAtCompileTime : InnerMaxSize,
|
||||
LargeEnough = VmlSize==Dynamic || VmlSize>=EIGEN_MKL_VML_THRESHOLD
|
||||
};
|
||||
public:
|
||||
enum {
|
||||
EnableVml = MightEnableVml && LargeEnough,
|
||||
Traversal = MightLinearize ? LinearTraversal : DefaultTraversal
|
||||
};
|
||||
};
|
||||
|
||||
#define EIGEN_PP_EXPAND(ARG) ARG
|
||||
#if !defined (EIGEN_FAST_MATH) || (EIGEN_FAST_MATH != 1)
|
||||
#define EIGEN_VMLMODE_EXPAND_xLA , VML_HA
|
||||
#else
|
||||
#define EIGEN_VMLMODE_EXPAND_xLA , VML_LA
|
||||
#endif
|
||||
|
||||
#define EIGEN_VMLMODE_EXPAND_x_
|
||||
|
||||
#define EIGEN_VMLMODE_PREFIX_xLA vm
|
||||
#define EIGEN_VMLMODE_PREFIX_x_ v
|
||||
#define EIGEN_VMLMODE_PREFIX(VMLMODE) EIGEN_CAT(EIGEN_VMLMODE_PREFIX_x,VMLMODE)
|
||||
|
||||
#define EIGEN_MKL_VML_DECLARE_UNARY_CALL(EIGENOP, VMLOP, EIGENTYPE, VMLTYPE, VMLMODE) \
|
||||
template< typename DstXprType, typename SrcXprNested> \
|
||||
struct Assignment<DstXprType, CwiseUnaryOp<scalar_##EIGENOP##_op<EIGENTYPE>, SrcXprNested>, assign_op<EIGENTYPE,EIGENTYPE>, \
|
||||
Dense2Dense, typename enable_if<vml_assign_traits<DstXprType,SrcXprNested>::EnableVml>::type> { \
|
||||
typedef CwiseUnaryOp<scalar_##EIGENOP##_op<EIGENTYPE>, SrcXprNested> SrcXprType; \
|
||||
static void run(DstXprType &dst, const SrcXprType &src, const assign_op<EIGENTYPE,EIGENTYPE> &func) { \
|
||||
resize_if_allowed(dst, src, func); \
|
||||
eigen_assert(dst.rows() == src.rows() && dst.cols() == src.cols()); \
|
||||
if(vml_assign_traits<DstXprType,SrcXprNested>::Traversal==LinearTraversal) { \
|
||||
VMLOP(dst.size(), (const VMLTYPE*)src.nestedExpression().data(), \
|
||||
(VMLTYPE*)dst.data() EIGEN_PP_EXPAND(EIGEN_VMLMODE_EXPAND_x##VMLMODE) ); \
|
||||
} else { \
|
||||
const Index outerSize = dst.outerSize(); \
|
||||
for(Index outer = 0; outer < outerSize; ++outer) { \
|
||||
const EIGENTYPE *src_ptr = src.IsRowMajor ? &(src.nestedExpression().coeffRef(outer,0)) : \
|
||||
&(src.nestedExpression().coeffRef(0, outer)); \
|
||||
EIGENTYPE *dst_ptr = dst.IsRowMajor ? &(dst.coeffRef(outer,0)) : &(dst.coeffRef(0, outer)); \
|
||||
VMLOP( dst.innerSize(), (const VMLTYPE*)src_ptr, \
|
||||
(VMLTYPE*)dst_ptr EIGEN_PP_EXPAND(EIGEN_VMLMODE_EXPAND_x##VMLMODE)); \
|
||||
} \
|
||||
} \
|
||||
} \
|
||||
}; \
|
||||
|
||||
|
||||
#define EIGEN_MKL_VML_DECLARE_UNARY_CALLS_REAL(EIGENOP, VMLOP, VMLMODE) \
|
||||
EIGEN_MKL_VML_DECLARE_UNARY_CALL(EIGENOP, EIGEN_CAT(EIGEN_VMLMODE_PREFIX(VMLMODE),s##VMLOP), float, float, VMLMODE) \
|
||||
EIGEN_MKL_VML_DECLARE_UNARY_CALL(EIGENOP, EIGEN_CAT(EIGEN_VMLMODE_PREFIX(VMLMODE),d##VMLOP), double, double, VMLMODE)
|
||||
|
||||
#define EIGEN_MKL_VML_DECLARE_UNARY_CALLS_CPLX(EIGENOP, VMLOP, VMLMODE) \
|
||||
EIGEN_MKL_VML_DECLARE_UNARY_CALL(EIGENOP, EIGEN_CAT(EIGEN_VMLMODE_PREFIX(VMLMODE),c##VMLOP), scomplex, MKL_Complex8, VMLMODE) \
|
||||
EIGEN_MKL_VML_DECLARE_UNARY_CALL(EIGENOP, EIGEN_CAT(EIGEN_VMLMODE_PREFIX(VMLMODE),z##VMLOP), dcomplex, MKL_Complex16, VMLMODE)
|
||||
|
||||
#define EIGEN_MKL_VML_DECLARE_UNARY_CALLS(EIGENOP, VMLOP, VMLMODE) \
|
||||
EIGEN_MKL_VML_DECLARE_UNARY_CALLS_REAL(EIGENOP, VMLOP, VMLMODE) \
|
||||
EIGEN_MKL_VML_DECLARE_UNARY_CALLS_CPLX(EIGENOP, VMLOP, VMLMODE)
|
||||
|
||||
|
||||
EIGEN_MKL_VML_DECLARE_UNARY_CALLS(sin, Sin, LA)
|
||||
EIGEN_MKL_VML_DECLARE_UNARY_CALLS(asin, Asin, LA)
|
||||
EIGEN_MKL_VML_DECLARE_UNARY_CALLS(sinh, Sinh, LA)
|
||||
EIGEN_MKL_VML_DECLARE_UNARY_CALLS(cos, Cos, LA)
|
||||
EIGEN_MKL_VML_DECLARE_UNARY_CALLS(acos, Acos, LA)
|
||||
EIGEN_MKL_VML_DECLARE_UNARY_CALLS(cosh, Cosh, LA)
|
||||
EIGEN_MKL_VML_DECLARE_UNARY_CALLS(tan, Tan, LA)
|
||||
EIGEN_MKL_VML_DECLARE_UNARY_CALLS(atan, Atan, LA)
|
||||
EIGEN_MKL_VML_DECLARE_UNARY_CALLS(tanh, Tanh, LA)
|
||||
// EIGEN_MKL_VML_DECLARE_UNARY_CALLS(abs, Abs, _)
|
||||
EIGEN_MKL_VML_DECLARE_UNARY_CALLS(exp, Exp, LA)
|
||||
EIGEN_MKL_VML_DECLARE_UNARY_CALLS(log, Ln, LA)
|
||||
EIGEN_MKL_VML_DECLARE_UNARY_CALLS(log10, Log10, LA)
|
||||
EIGEN_MKL_VML_DECLARE_UNARY_CALLS(sqrt, Sqrt, _)
|
||||
|
||||
EIGEN_MKL_VML_DECLARE_UNARY_CALLS_REAL(square, Sqr, _)
|
||||
EIGEN_MKL_VML_DECLARE_UNARY_CALLS_CPLX(arg, Arg, _)
|
||||
EIGEN_MKL_VML_DECLARE_UNARY_CALLS_REAL(round, Round, _)
|
||||
EIGEN_MKL_VML_DECLARE_UNARY_CALLS_REAL(floor, Floor, _)
|
||||
EIGEN_MKL_VML_DECLARE_UNARY_CALLS_REAL(ceil, Ceil, _)
|
||||
|
||||
#define EIGEN_MKL_VML_DECLARE_POW_CALL(EIGENOP, VMLOP, EIGENTYPE, VMLTYPE, VMLMODE) \
|
||||
template< typename DstXprType, typename SrcXprNested, typename Plain> \
|
||||
struct Assignment<DstXprType, CwiseBinaryOp<scalar_##EIGENOP##_op<EIGENTYPE,EIGENTYPE>, SrcXprNested, \
|
||||
const CwiseNullaryOp<internal::scalar_constant_op<EIGENTYPE>,Plain> >, assign_op<EIGENTYPE,EIGENTYPE>, \
|
||||
Dense2Dense, typename enable_if<vml_assign_traits<DstXprType,SrcXprNested>::EnableVml>::type> { \
|
||||
typedef CwiseBinaryOp<scalar_##EIGENOP##_op<EIGENTYPE,EIGENTYPE>, SrcXprNested, \
|
||||
const CwiseNullaryOp<internal::scalar_constant_op<EIGENTYPE>,Plain> > SrcXprType; \
|
||||
static void run(DstXprType &dst, const SrcXprType &src, const assign_op<EIGENTYPE,EIGENTYPE> &func) { \
|
||||
resize_if_allowed(dst, src, func); \
|
||||
eigen_assert(dst.rows() == src.rows() && dst.cols() == src.cols()); \
|
||||
VMLTYPE exponent = reinterpret_cast<const VMLTYPE&>(src.rhs().functor().m_other); \
|
||||
if(vml_assign_traits<DstXprType,SrcXprNested>::Traversal==LinearTraversal) \
|
||||
{ \
|
||||
VMLOP( dst.size(), (const VMLTYPE*)src.lhs().data(), exponent, \
|
||||
(VMLTYPE*)dst.data() EIGEN_PP_EXPAND(EIGEN_VMLMODE_EXPAND_x##VMLMODE) ); \
|
||||
} else { \
|
||||
const Index outerSize = dst.outerSize(); \
|
||||
for(Index outer = 0; outer < outerSize; ++outer) { \
|
||||
const EIGENTYPE *src_ptr = src.IsRowMajor ? &(src.lhs().coeffRef(outer,0)) : \
|
||||
&(src.lhs().coeffRef(0, outer)); \
|
||||
EIGENTYPE *dst_ptr = dst.IsRowMajor ? &(dst.coeffRef(outer,0)) : &(dst.coeffRef(0, outer)); \
|
||||
VMLOP( dst.innerSize(), (const VMLTYPE*)src_ptr, exponent, \
|
||||
(VMLTYPE*)dst_ptr EIGEN_PP_EXPAND(EIGEN_VMLMODE_EXPAND_x##VMLMODE)); \
|
||||
} \
|
||||
} \
|
||||
} \
|
||||
};
|
||||
|
||||
EIGEN_MKL_VML_DECLARE_POW_CALL(pow, vmsPowx, float, float, LA)
|
||||
EIGEN_MKL_VML_DECLARE_POW_CALL(pow, vmdPowx, double, double, LA)
|
||||
EIGEN_MKL_VML_DECLARE_POW_CALL(pow, vmcPowx, scomplex, MKL_Complex8, LA)
|
||||
EIGEN_MKL_VML_DECLARE_POW_CALL(pow, vmzPowx, dcomplex, MKL_Complex16, LA)
|
||||
|
||||
} // end namespace internal
|
||||
|
||||
} // end namespace Eigen
|
||||
|
||||
#endif // EIGEN_ASSIGN_VML_H
|
||||
@@ -0,0 +1,353 @@
|
||||
// This file is part of Eigen, a lightweight C++ template library
|
||||
// for linear algebra.
|
||||
//
|
||||
// Copyright (C) 2009 Gael Guennebaud <gael.guennebaud@inria.fr>
|
||||
//
|
||||
// This Source Code Form is subject to the terms of the Mozilla
|
||||
// Public License v. 2.0. If a copy of the MPL was not distributed
|
||||
// with this file, You can obtain one at http://mozilla.org/MPL/2.0/.
|
||||
|
||||
#ifndef EIGEN_BANDMATRIX_H
|
||||
#define EIGEN_BANDMATRIX_H
|
||||
|
||||
namespace Eigen {
|
||||
|
||||
namespace internal {
|
||||
|
||||
template<typename Derived>
|
||||
class BandMatrixBase : public EigenBase<Derived>
|
||||
{
|
||||
public:
|
||||
|
||||
enum {
|
||||
Flags = internal::traits<Derived>::Flags,
|
||||
CoeffReadCost = internal::traits<Derived>::CoeffReadCost,
|
||||
RowsAtCompileTime = internal::traits<Derived>::RowsAtCompileTime,
|
||||
ColsAtCompileTime = internal::traits<Derived>::ColsAtCompileTime,
|
||||
MaxRowsAtCompileTime = internal::traits<Derived>::MaxRowsAtCompileTime,
|
||||
MaxColsAtCompileTime = internal::traits<Derived>::MaxColsAtCompileTime,
|
||||
Supers = internal::traits<Derived>::Supers,
|
||||
Subs = internal::traits<Derived>::Subs,
|
||||
Options = internal::traits<Derived>::Options
|
||||
};
|
||||
typedef typename internal::traits<Derived>::Scalar Scalar;
|
||||
typedef Matrix<Scalar,RowsAtCompileTime,ColsAtCompileTime> DenseMatrixType;
|
||||
typedef typename DenseMatrixType::StorageIndex StorageIndex;
|
||||
typedef typename internal::traits<Derived>::CoefficientsType CoefficientsType;
|
||||
typedef EigenBase<Derived> Base;
|
||||
|
||||
protected:
|
||||
enum {
|
||||
DataRowsAtCompileTime = ((Supers!=Dynamic) && (Subs!=Dynamic))
|
||||
? 1 + Supers + Subs
|
||||
: Dynamic,
|
||||
SizeAtCompileTime = EIGEN_SIZE_MIN_PREFER_DYNAMIC(RowsAtCompileTime,ColsAtCompileTime)
|
||||
};
|
||||
|
||||
public:
|
||||
|
||||
using Base::derived;
|
||||
using Base::rows;
|
||||
using Base::cols;
|
||||
|
||||
/** \returns the number of super diagonals */
|
||||
inline Index supers() const { return derived().supers(); }
|
||||
|
||||
/** \returns the number of sub diagonals */
|
||||
inline Index subs() const { return derived().subs(); }
|
||||
|
||||
/** \returns an expression of the underlying coefficient matrix */
|
||||
inline const CoefficientsType& coeffs() const { return derived().coeffs(); }
|
||||
|
||||
/** \returns an expression of the underlying coefficient matrix */
|
||||
inline CoefficientsType& coeffs() { return derived().coeffs(); }
|
||||
|
||||
/** \returns a vector expression of the \a i -th column,
|
||||
* only the meaningful part is returned.
|
||||
* \warning the internal storage must be column major. */
|
||||
inline Block<CoefficientsType,Dynamic,1> col(Index i)
|
||||
{
|
||||
EIGEN_STATIC_ASSERT((int(Options) & int(RowMajor)) == 0, THIS_METHOD_IS_ONLY_FOR_COLUMN_MAJOR_MATRICES);
|
||||
Index start = 0;
|
||||
Index len = coeffs().rows();
|
||||
if (i<=supers())
|
||||
{
|
||||
start = supers()-i;
|
||||
len = (std::min)(rows(),std::max<Index>(0,coeffs().rows() - (supers()-i)));
|
||||
}
|
||||
else if (i>=rows()-subs())
|
||||
len = std::max<Index>(0,coeffs().rows() - (i + 1 - rows() + subs()));
|
||||
return Block<CoefficientsType,Dynamic,1>(coeffs(), start, i, len, 1);
|
||||
}
|
||||
|
||||
/** \returns a vector expression of the main diagonal */
|
||||
inline Block<CoefficientsType,1,SizeAtCompileTime> diagonal()
|
||||
{ return Block<CoefficientsType,1,SizeAtCompileTime>(coeffs(),supers(),0,1,(std::min)(rows(),cols())); }
|
||||
|
||||
/** \returns a vector expression of the main diagonal (const version) */
|
||||
inline const Block<const CoefficientsType,1,SizeAtCompileTime> diagonal() const
|
||||
{ return Block<const CoefficientsType,1,SizeAtCompileTime>(coeffs(),supers(),0,1,(std::min)(rows(),cols())); }
|
||||
|
||||
template<int Index> struct DiagonalIntReturnType {
|
||||
enum {
|
||||
ReturnOpposite = (int(Options) & int(SelfAdjoint)) && (((Index) > 0 && Supers == 0) || ((Index) < 0 && Subs == 0)),
|
||||
Conjugate = ReturnOpposite && NumTraits<Scalar>::IsComplex,
|
||||
ActualIndex = ReturnOpposite ? -Index : Index,
|
||||
DiagonalSize = (RowsAtCompileTime==Dynamic || ColsAtCompileTime==Dynamic)
|
||||
? Dynamic
|
||||
: (ActualIndex<0
|
||||
? EIGEN_SIZE_MIN_PREFER_DYNAMIC(ColsAtCompileTime, RowsAtCompileTime + ActualIndex)
|
||||
: EIGEN_SIZE_MIN_PREFER_DYNAMIC(RowsAtCompileTime, ColsAtCompileTime - ActualIndex))
|
||||
};
|
||||
typedef Block<CoefficientsType,1, DiagonalSize> BuildType;
|
||||
typedef typename internal::conditional<Conjugate,
|
||||
CwiseUnaryOp<internal::scalar_conjugate_op<Scalar>,BuildType >,
|
||||
BuildType>::type Type;
|
||||
};
|
||||
|
||||
/** \returns a vector expression of the \a N -th sub or super diagonal */
|
||||
template<int N> inline typename DiagonalIntReturnType<N>::Type diagonal()
|
||||
{
|
||||
return typename DiagonalIntReturnType<N>::BuildType(coeffs(), supers()-N, (std::max)(0,N), 1, diagonalLength(N));
|
||||
}
|
||||
|
||||
/** \returns a vector expression of the \a N -th sub or super diagonal */
|
||||
template<int N> inline const typename DiagonalIntReturnType<N>::Type diagonal() const
|
||||
{
|
||||
return typename DiagonalIntReturnType<N>::BuildType(coeffs(), supers()-N, (std::max)(0,N), 1, diagonalLength(N));
|
||||
}
|
||||
|
||||
/** \returns a vector expression of the \a i -th sub or super diagonal */
|
||||
inline Block<CoefficientsType,1,Dynamic> diagonal(Index i)
|
||||
{
|
||||
eigen_assert((i<0 && -i<=subs()) || (i>=0 && i<=supers()));
|
||||
return Block<CoefficientsType,1,Dynamic>(coeffs(), supers()-i, std::max<Index>(0,i), 1, diagonalLength(i));
|
||||
}
|
||||
|
||||
/** \returns a vector expression of the \a i -th sub or super diagonal */
|
||||
inline const Block<const CoefficientsType,1,Dynamic> diagonal(Index i) const
|
||||
{
|
||||
eigen_assert((i<0 && -i<=subs()) || (i>=0 && i<=supers()));
|
||||
return Block<const CoefficientsType,1,Dynamic>(coeffs(), supers()-i, std::max<Index>(0,i), 1, diagonalLength(i));
|
||||
}
|
||||
|
||||
template<typename Dest> inline void evalTo(Dest& dst) const
|
||||
{
|
||||
dst.resize(rows(),cols());
|
||||
dst.setZero();
|
||||
dst.diagonal() = diagonal();
|
||||
for (Index i=1; i<=supers();++i)
|
||||
dst.diagonal(i) = diagonal(i);
|
||||
for (Index i=1; i<=subs();++i)
|
||||
dst.diagonal(-i) = diagonal(-i);
|
||||
}
|
||||
|
||||
DenseMatrixType toDenseMatrix() const
|
||||
{
|
||||
DenseMatrixType res(rows(),cols());
|
||||
evalTo(res);
|
||||
return res;
|
||||
}
|
||||
|
||||
protected:
|
||||
|
||||
inline Index diagonalLength(Index i) const
|
||||
{ return i<0 ? (std::min)(cols(),rows()+i) : (std::min)(rows(),cols()-i); }
|
||||
};
|
||||
|
||||
/**
|
||||
* \class BandMatrix
|
||||
* \ingroup Core_Module
|
||||
*
|
||||
* \brief Represents a rectangular matrix with a banded storage
|
||||
*
|
||||
* \tparam _Scalar Numeric type, i.e. float, double, int
|
||||
* \tparam _Rows Number of rows, or \b Dynamic
|
||||
* \tparam _Cols Number of columns, or \b Dynamic
|
||||
* \tparam _Supers Number of super diagonal
|
||||
* \tparam _Subs Number of sub diagonal
|
||||
* \tparam _Options A combination of either \b #RowMajor or \b #ColMajor, and of \b #SelfAdjoint
|
||||
* The former controls \ref TopicStorageOrders "storage order", and defaults to
|
||||
* column-major. The latter controls whether the matrix represents a selfadjoint
|
||||
* matrix in which case either Supers of Subs have to be null.
|
||||
*
|
||||
* \sa class TridiagonalMatrix
|
||||
*/
|
||||
|
||||
template<typename _Scalar, int _Rows, int _Cols, int _Supers, int _Subs, int _Options>
|
||||
struct traits<BandMatrix<_Scalar,_Rows,_Cols,_Supers,_Subs,_Options> >
|
||||
{
|
||||
typedef _Scalar Scalar;
|
||||
typedef Dense StorageKind;
|
||||
typedef Eigen::Index StorageIndex;
|
||||
enum {
|
||||
CoeffReadCost = NumTraits<Scalar>::ReadCost,
|
||||
RowsAtCompileTime = _Rows,
|
||||
ColsAtCompileTime = _Cols,
|
||||
MaxRowsAtCompileTime = _Rows,
|
||||
MaxColsAtCompileTime = _Cols,
|
||||
Flags = LvalueBit,
|
||||
Supers = _Supers,
|
||||
Subs = _Subs,
|
||||
Options = _Options,
|
||||
DataRowsAtCompileTime = ((Supers!=Dynamic) && (Subs!=Dynamic)) ? 1 + Supers + Subs : Dynamic
|
||||
};
|
||||
typedef Matrix<Scalar, DataRowsAtCompileTime, ColsAtCompileTime, int(Options) & int(RowMajor) ? RowMajor : ColMajor> CoefficientsType;
|
||||
};
|
||||
|
||||
template<typename _Scalar, int Rows, int Cols, int Supers, int Subs, int Options>
|
||||
class BandMatrix : public BandMatrixBase<BandMatrix<_Scalar,Rows,Cols,Supers,Subs,Options> >
|
||||
{
|
||||
public:
|
||||
|
||||
typedef typename internal::traits<BandMatrix>::Scalar Scalar;
|
||||
typedef typename internal::traits<BandMatrix>::StorageIndex StorageIndex;
|
||||
typedef typename internal::traits<BandMatrix>::CoefficientsType CoefficientsType;
|
||||
|
||||
explicit inline BandMatrix(Index rows=Rows, Index cols=Cols, Index supers=Supers, Index subs=Subs)
|
||||
: m_coeffs(1+supers+subs,cols),
|
||||
m_rows(rows), m_supers(supers), m_subs(subs)
|
||||
{
|
||||
}
|
||||
|
||||
/** \returns the number of columns */
|
||||
inline EIGEN_CONSTEXPR Index rows() const { return m_rows.value(); }
|
||||
|
||||
/** \returns the number of rows */
|
||||
inline EIGEN_CONSTEXPR Index cols() const { return m_coeffs.cols(); }
|
||||
|
||||
/** \returns the number of super diagonals */
|
||||
inline EIGEN_CONSTEXPR Index supers() const { return m_supers.value(); }
|
||||
|
||||
/** \returns the number of sub diagonals */
|
||||
inline EIGEN_CONSTEXPR Index subs() const { return m_subs.value(); }
|
||||
|
||||
inline const CoefficientsType& coeffs() const { return m_coeffs; }
|
||||
inline CoefficientsType& coeffs() { return m_coeffs; }
|
||||
|
||||
protected:
|
||||
|
||||
CoefficientsType m_coeffs;
|
||||
internal::variable_if_dynamic<Index, Rows> m_rows;
|
||||
internal::variable_if_dynamic<Index, Supers> m_supers;
|
||||
internal::variable_if_dynamic<Index, Subs> m_subs;
|
||||
};
|
||||
|
||||
template<typename _CoefficientsType,int _Rows, int _Cols, int _Supers, int _Subs,int _Options>
|
||||
class BandMatrixWrapper;
|
||||
|
||||
template<typename _CoefficientsType,int _Rows, int _Cols, int _Supers, int _Subs,int _Options>
|
||||
struct traits<BandMatrixWrapper<_CoefficientsType,_Rows,_Cols,_Supers,_Subs,_Options> >
|
||||
{
|
||||
typedef typename _CoefficientsType::Scalar Scalar;
|
||||
typedef typename _CoefficientsType::StorageKind StorageKind;
|
||||
typedef typename _CoefficientsType::StorageIndex StorageIndex;
|
||||
enum {
|
||||
CoeffReadCost = internal::traits<_CoefficientsType>::CoeffReadCost,
|
||||
RowsAtCompileTime = _Rows,
|
||||
ColsAtCompileTime = _Cols,
|
||||
MaxRowsAtCompileTime = _Rows,
|
||||
MaxColsAtCompileTime = _Cols,
|
||||
Flags = LvalueBit,
|
||||
Supers = _Supers,
|
||||
Subs = _Subs,
|
||||
Options = _Options,
|
||||
DataRowsAtCompileTime = ((Supers!=Dynamic) && (Subs!=Dynamic)) ? 1 + Supers + Subs : Dynamic
|
||||
};
|
||||
typedef _CoefficientsType CoefficientsType;
|
||||
};
|
||||
|
||||
template<typename _CoefficientsType,int _Rows, int _Cols, int _Supers, int _Subs,int _Options>
|
||||
class BandMatrixWrapper : public BandMatrixBase<BandMatrixWrapper<_CoefficientsType,_Rows,_Cols,_Supers,_Subs,_Options> >
|
||||
{
|
||||
public:
|
||||
|
||||
typedef typename internal::traits<BandMatrixWrapper>::Scalar Scalar;
|
||||
typedef typename internal::traits<BandMatrixWrapper>::CoefficientsType CoefficientsType;
|
||||
typedef typename internal::traits<BandMatrixWrapper>::StorageIndex StorageIndex;
|
||||
|
||||
explicit inline BandMatrixWrapper(const CoefficientsType& coeffs, Index rows=_Rows, Index cols=_Cols, Index supers=_Supers, Index subs=_Subs)
|
||||
: m_coeffs(coeffs),
|
||||
m_rows(rows), m_supers(supers), m_subs(subs)
|
||||
{
|
||||
EIGEN_UNUSED_VARIABLE(cols);
|
||||
//internal::assert(coeffs.cols()==cols() && (supers()+subs()+1)==coeffs.rows());
|
||||
}
|
||||
|
||||
/** \returns the number of columns */
|
||||
inline EIGEN_CONSTEXPR Index rows() const { return m_rows.value(); }
|
||||
|
||||
/** \returns the number of rows */
|
||||
inline EIGEN_CONSTEXPR Index cols() const { return m_coeffs.cols(); }
|
||||
|
||||
/** \returns the number of super diagonals */
|
||||
inline EIGEN_CONSTEXPR Index supers() const { return m_supers.value(); }
|
||||
|
||||
/** \returns the number of sub diagonals */
|
||||
inline EIGEN_CONSTEXPR Index subs() const { return m_subs.value(); }
|
||||
|
||||
inline const CoefficientsType& coeffs() const { return m_coeffs; }
|
||||
|
||||
protected:
|
||||
|
||||
const CoefficientsType& m_coeffs;
|
||||
internal::variable_if_dynamic<Index, _Rows> m_rows;
|
||||
internal::variable_if_dynamic<Index, _Supers> m_supers;
|
||||
internal::variable_if_dynamic<Index, _Subs> m_subs;
|
||||
};
|
||||
|
||||
/**
|
||||
* \class TridiagonalMatrix
|
||||
* \ingroup Core_Module
|
||||
*
|
||||
* \brief Represents a tridiagonal matrix with a compact banded storage
|
||||
*
|
||||
* \tparam Scalar Numeric type, i.e. float, double, int
|
||||
* \tparam Size Number of rows and cols, or \b Dynamic
|
||||
* \tparam Options Can be 0 or \b SelfAdjoint
|
||||
*
|
||||
* \sa class BandMatrix
|
||||
*/
|
||||
template<typename Scalar, int Size, int Options>
|
||||
class TridiagonalMatrix : public BandMatrix<Scalar,Size,Size,Options&SelfAdjoint?0:1,1,Options|RowMajor>
|
||||
{
|
||||
typedef BandMatrix<Scalar,Size,Size,Options&SelfAdjoint?0:1,1,Options|RowMajor> Base;
|
||||
typedef typename Base::StorageIndex StorageIndex;
|
||||
public:
|
||||
explicit TridiagonalMatrix(Index size = Size) : Base(size,size,Options&SelfAdjoint?0:1,1) {}
|
||||
|
||||
inline typename Base::template DiagonalIntReturnType<1>::Type super()
|
||||
{ return Base::template diagonal<1>(); }
|
||||
inline const typename Base::template DiagonalIntReturnType<1>::Type super() const
|
||||
{ return Base::template diagonal<1>(); }
|
||||
inline typename Base::template DiagonalIntReturnType<-1>::Type sub()
|
||||
{ return Base::template diagonal<-1>(); }
|
||||
inline const typename Base::template DiagonalIntReturnType<-1>::Type sub() const
|
||||
{ return Base::template diagonal<-1>(); }
|
||||
protected:
|
||||
};
|
||||
|
||||
|
||||
struct BandShape {};
|
||||
|
||||
template<typename _Scalar, int _Rows, int _Cols, int _Supers, int _Subs, int _Options>
|
||||
struct evaluator_traits<BandMatrix<_Scalar,_Rows,_Cols,_Supers,_Subs,_Options> >
|
||||
: public evaluator_traits_base<BandMatrix<_Scalar,_Rows,_Cols,_Supers,_Subs,_Options> >
|
||||
{
|
||||
typedef BandShape Shape;
|
||||
};
|
||||
|
||||
template<typename _CoefficientsType,int _Rows, int _Cols, int _Supers, int _Subs,int _Options>
|
||||
struct evaluator_traits<BandMatrixWrapper<_CoefficientsType,_Rows,_Cols,_Supers,_Subs,_Options> >
|
||||
: public evaluator_traits_base<BandMatrixWrapper<_CoefficientsType,_Rows,_Cols,_Supers,_Subs,_Options> >
|
||||
{
|
||||
typedef BandShape Shape;
|
||||
};
|
||||
|
||||
template<> struct AssignmentKind<DenseShape,BandShape> { typedef EigenBase2EigenBase Kind; };
|
||||
|
||||
} // end namespace internal
|
||||
|
||||
} // end namespace Eigen
|
||||
|
||||
#endif // EIGEN_BANDMATRIX_H
|
||||
@@ -0,0 +1,448 @@
|
||||
// This file is part of Eigen, a lightweight C++ template library
|
||||
// for linear algebra.
|
||||
//
|
||||
// Copyright (C) 2008 Gael Guennebaud <gael.guennebaud@inria.fr>
|
||||
// Copyright (C) 2006-2010 Benoit Jacob <jacob.benoit.1@gmail.com>
|
||||
//
|
||||
// This Source Code Form is subject to the terms of the Mozilla
|
||||
// Public License v. 2.0. If a copy of the MPL was not distributed
|
||||
// with this file, You can obtain one at http://mozilla.org/MPL/2.0/.
|
||||
|
||||
#ifndef EIGEN_BLOCK_H
|
||||
#define EIGEN_BLOCK_H
|
||||
|
||||
namespace Eigen {
|
||||
|
||||
namespace internal {
|
||||
template<typename XprType, int BlockRows, int BlockCols, bool InnerPanel>
|
||||
struct traits<Block<XprType, BlockRows, BlockCols, InnerPanel> > : traits<XprType>
|
||||
{
|
||||
typedef typename traits<XprType>::Scalar Scalar;
|
||||
typedef typename traits<XprType>::StorageKind StorageKind;
|
||||
typedef typename traits<XprType>::XprKind XprKind;
|
||||
typedef typename ref_selector<XprType>::type XprTypeNested;
|
||||
typedef typename remove_reference<XprTypeNested>::type _XprTypeNested;
|
||||
enum{
|
||||
MatrixRows = traits<XprType>::RowsAtCompileTime,
|
||||
MatrixCols = traits<XprType>::ColsAtCompileTime,
|
||||
RowsAtCompileTime = MatrixRows == 0 ? 0 : BlockRows,
|
||||
ColsAtCompileTime = MatrixCols == 0 ? 0 : BlockCols,
|
||||
MaxRowsAtCompileTime = BlockRows==0 ? 0
|
||||
: RowsAtCompileTime != Dynamic ? int(RowsAtCompileTime)
|
||||
: int(traits<XprType>::MaxRowsAtCompileTime),
|
||||
MaxColsAtCompileTime = BlockCols==0 ? 0
|
||||
: ColsAtCompileTime != Dynamic ? int(ColsAtCompileTime)
|
||||
: int(traits<XprType>::MaxColsAtCompileTime),
|
||||
|
||||
XprTypeIsRowMajor = (int(traits<XprType>::Flags)&RowMajorBit) != 0,
|
||||
IsRowMajor = (MaxRowsAtCompileTime==1&&MaxColsAtCompileTime!=1) ? 1
|
||||
: (MaxColsAtCompileTime==1&&MaxRowsAtCompileTime!=1) ? 0
|
||||
: XprTypeIsRowMajor,
|
||||
HasSameStorageOrderAsXprType = (IsRowMajor == XprTypeIsRowMajor),
|
||||
InnerSize = IsRowMajor ? int(ColsAtCompileTime) : int(RowsAtCompileTime),
|
||||
InnerStrideAtCompileTime = HasSameStorageOrderAsXprType
|
||||
? int(inner_stride_at_compile_time<XprType>::ret)
|
||||
: int(outer_stride_at_compile_time<XprType>::ret),
|
||||
OuterStrideAtCompileTime = HasSameStorageOrderAsXprType
|
||||
? int(outer_stride_at_compile_time<XprType>::ret)
|
||||
: int(inner_stride_at_compile_time<XprType>::ret),
|
||||
|
||||
// FIXME, this traits is rather specialized for dense object and it needs to be cleaned further
|
||||
FlagsLvalueBit = is_lvalue<XprType>::value ? LvalueBit : 0,
|
||||
FlagsRowMajorBit = IsRowMajor ? RowMajorBit : 0,
|
||||
Flags = (traits<XprType>::Flags & (DirectAccessBit | (InnerPanel?CompressedAccessBit:0))) | FlagsLvalueBit | FlagsRowMajorBit,
|
||||
// FIXME DirectAccessBit should not be handled by expressions
|
||||
//
|
||||
// Alignment is needed by MapBase's assertions
|
||||
// We can sefely set it to false here. Internal alignment errors will be detected by an eigen_internal_assert in the respective evaluator
|
||||
Alignment = 0
|
||||
};
|
||||
};
|
||||
|
||||
template<typename XprType, int BlockRows=Dynamic, int BlockCols=Dynamic, bool InnerPanel = false,
|
||||
bool HasDirectAccess = internal::has_direct_access<XprType>::ret> class BlockImpl_dense;
|
||||
|
||||
} // end namespace internal
|
||||
|
||||
template<typename XprType, int BlockRows, int BlockCols, bool InnerPanel, typename StorageKind> class BlockImpl;
|
||||
|
||||
/** \class Block
|
||||
* \ingroup Core_Module
|
||||
*
|
||||
* \brief Expression of a fixed-size or dynamic-size block
|
||||
*
|
||||
* \tparam XprType the type of the expression in which we are taking a block
|
||||
* \tparam BlockRows the number of rows of the block we are taking at compile time (optional)
|
||||
* \tparam BlockCols the number of columns of the block we are taking at compile time (optional)
|
||||
* \tparam InnerPanel is true, if the block maps to a set of rows of a row major matrix or
|
||||
* to set of columns of a column major matrix (optional). The parameter allows to determine
|
||||
* at compile time whether aligned access is possible on the block expression.
|
||||
*
|
||||
* This class represents an expression of either a fixed-size or dynamic-size block. It is the return
|
||||
* type of DenseBase::block(Index,Index,Index,Index) and DenseBase::block<int,int>(Index,Index) and
|
||||
* most of the time this is the only way it is used.
|
||||
*
|
||||
* However, if you want to directly maniputate block expressions,
|
||||
* for instance if you want to write a function returning such an expression, you
|
||||
* will need to use this class.
|
||||
*
|
||||
* Here is an example illustrating the dynamic case:
|
||||
* \include class_Block.cpp
|
||||
* Output: \verbinclude class_Block.out
|
||||
*
|
||||
* \note Even though this expression has dynamic size, in the case where \a XprType
|
||||
* has fixed size, this expression inherits a fixed maximal size which means that evaluating
|
||||
* it does not cause a dynamic memory allocation.
|
||||
*
|
||||
* Here is an example illustrating the fixed-size case:
|
||||
* \include class_FixedBlock.cpp
|
||||
* Output: \verbinclude class_FixedBlock.out
|
||||
*
|
||||
* \sa DenseBase::block(Index,Index,Index,Index), DenseBase::block(Index,Index), class VectorBlock
|
||||
*/
|
||||
template<typename XprType, int BlockRows, int BlockCols, bool InnerPanel> class Block
|
||||
: public BlockImpl<XprType, BlockRows, BlockCols, InnerPanel, typename internal::traits<XprType>::StorageKind>
|
||||
{
|
||||
typedef BlockImpl<XprType, BlockRows, BlockCols, InnerPanel, typename internal::traits<XprType>::StorageKind> Impl;
|
||||
public:
|
||||
//typedef typename Impl::Base Base;
|
||||
typedef Impl Base;
|
||||
EIGEN_GENERIC_PUBLIC_INTERFACE(Block)
|
||||
EIGEN_INHERIT_ASSIGNMENT_OPERATORS(Block)
|
||||
|
||||
typedef typename internal::remove_all<XprType>::type NestedExpression;
|
||||
|
||||
/** Column or Row constructor
|
||||
*/
|
||||
EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
|
||||
Block(XprType& xpr, Index i) : Impl(xpr,i)
|
||||
{
|
||||
eigen_assert( (i>=0) && (
|
||||
((BlockRows==1) && (BlockCols==XprType::ColsAtCompileTime) && i<xpr.rows())
|
||||
||((BlockRows==XprType::RowsAtCompileTime) && (BlockCols==1) && i<xpr.cols())));
|
||||
}
|
||||
|
||||
/** Fixed-size constructor
|
||||
*/
|
||||
EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
|
||||
Block(XprType& xpr, Index startRow, Index startCol)
|
||||
: Impl(xpr, startRow, startCol)
|
||||
{
|
||||
EIGEN_STATIC_ASSERT(RowsAtCompileTime!=Dynamic && ColsAtCompileTime!=Dynamic,THIS_METHOD_IS_ONLY_FOR_FIXED_SIZE)
|
||||
eigen_assert(startRow >= 0 && BlockRows >= 0 && startRow + BlockRows <= xpr.rows()
|
||||
&& startCol >= 0 && BlockCols >= 0 && startCol + BlockCols <= xpr.cols());
|
||||
}
|
||||
|
||||
/** Dynamic-size constructor
|
||||
*/
|
||||
EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
|
||||
Block(XprType& xpr,
|
||||
Index startRow, Index startCol,
|
||||
Index blockRows, Index blockCols)
|
||||
: Impl(xpr, startRow, startCol, blockRows, blockCols)
|
||||
{
|
||||
eigen_assert((RowsAtCompileTime==Dynamic || RowsAtCompileTime==blockRows)
|
||||
&& (ColsAtCompileTime==Dynamic || ColsAtCompileTime==blockCols));
|
||||
eigen_assert(startRow >= 0 && blockRows >= 0 && startRow <= xpr.rows() - blockRows
|
||||
&& startCol >= 0 && blockCols >= 0 && startCol <= xpr.cols() - blockCols);
|
||||
}
|
||||
};
|
||||
|
||||
// The generic default implementation for dense block simplu forward to the internal::BlockImpl_dense
|
||||
// that must be specialized for direct and non-direct access...
|
||||
template<typename XprType, int BlockRows, int BlockCols, bool InnerPanel>
|
||||
class BlockImpl<XprType, BlockRows, BlockCols, InnerPanel, Dense>
|
||||
: public internal::BlockImpl_dense<XprType, BlockRows, BlockCols, InnerPanel>
|
||||
{
|
||||
typedef internal::BlockImpl_dense<XprType, BlockRows, BlockCols, InnerPanel> Impl;
|
||||
typedef typename XprType::StorageIndex StorageIndex;
|
||||
public:
|
||||
typedef Impl Base;
|
||||
EIGEN_INHERIT_ASSIGNMENT_OPERATORS(BlockImpl)
|
||||
EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE BlockImpl(XprType& xpr, Index i) : Impl(xpr,i) {}
|
||||
EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE BlockImpl(XprType& xpr, Index startRow, Index startCol) : Impl(xpr, startRow, startCol) {}
|
||||
EIGEN_DEVICE_FUNC
|
||||
EIGEN_STRONG_INLINE BlockImpl(XprType& xpr, Index startRow, Index startCol, Index blockRows, Index blockCols)
|
||||
: Impl(xpr, startRow, startCol, blockRows, blockCols) {}
|
||||
};
|
||||
|
||||
namespace internal {
|
||||
|
||||
/** \internal Internal implementation of dense Blocks in the general case. */
|
||||
template<typename XprType, int BlockRows, int BlockCols, bool InnerPanel, bool HasDirectAccess> class BlockImpl_dense
|
||||
: public internal::dense_xpr_base<Block<XprType, BlockRows, BlockCols, InnerPanel> >::type
|
||||
{
|
||||
typedef Block<XprType, BlockRows, BlockCols, InnerPanel> BlockType;
|
||||
typedef typename internal::ref_selector<XprType>::non_const_type XprTypeNested;
|
||||
public:
|
||||
|
||||
typedef typename internal::dense_xpr_base<BlockType>::type Base;
|
||||
EIGEN_DENSE_PUBLIC_INTERFACE(BlockType)
|
||||
EIGEN_INHERIT_ASSIGNMENT_OPERATORS(BlockImpl_dense)
|
||||
|
||||
// class InnerIterator; // FIXME apparently never used
|
||||
|
||||
/** Column or Row constructor
|
||||
*/
|
||||
EIGEN_DEVICE_FUNC
|
||||
inline BlockImpl_dense(XprType& xpr, Index i)
|
||||
: m_xpr(xpr),
|
||||
// It is a row if and only if BlockRows==1 and BlockCols==XprType::ColsAtCompileTime,
|
||||
// and it is a column if and only if BlockRows==XprType::RowsAtCompileTime and BlockCols==1,
|
||||
// all other cases are invalid.
|
||||
// The case a 1x1 matrix seems ambiguous, but the result is the same anyway.
|
||||
m_startRow( (BlockRows==1) && (BlockCols==XprType::ColsAtCompileTime) ? i : 0),
|
||||
m_startCol( (BlockRows==XprType::RowsAtCompileTime) && (BlockCols==1) ? i : 0),
|
||||
m_blockRows(BlockRows==1 ? 1 : xpr.rows()),
|
||||
m_blockCols(BlockCols==1 ? 1 : xpr.cols())
|
||||
{}
|
||||
|
||||
/** Fixed-size constructor
|
||||
*/
|
||||
EIGEN_DEVICE_FUNC
|
||||
inline BlockImpl_dense(XprType& xpr, Index startRow, Index startCol)
|
||||
: m_xpr(xpr), m_startRow(startRow), m_startCol(startCol),
|
||||
m_blockRows(BlockRows), m_blockCols(BlockCols)
|
||||
{}
|
||||
|
||||
/** Dynamic-size constructor
|
||||
*/
|
||||
EIGEN_DEVICE_FUNC
|
||||
inline BlockImpl_dense(XprType& xpr,
|
||||
Index startRow, Index startCol,
|
||||
Index blockRows, Index blockCols)
|
||||
: m_xpr(xpr), m_startRow(startRow), m_startCol(startCol),
|
||||
m_blockRows(blockRows), m_blockCols(blockCols)
|
||||
{}
|
||||
|
||||
EIGEN_DEVICE_FUNC inline Index rows() const { return m_blockRows.value(); }
|
||||
EIGEN_DEVICE_FUNC inline Index cols() const { return m_blockCols.value(); }
|
||||
|
||||
EIGEN_DEVICE_FUNC
|
||||
inline Scalar& coeffRef(Index rowId, Index colId)
|
||||
{
|
||||
EIGEN_STATIC_ASSERT_LVALUE(XprType)
|
||||
return m_xpr.coeffRef(rowId + m_startRow.value(), colId + m_startCol.value());
|
||||
}
|
||||
|
||||
EIGEN_DEVICE_FUNC
|
||||
inline const Scalar& coeffRef(Index rowId, Index colId) const
|
||||
{
|
||||
return m_xpr.derived().coeffRef(rowId + m_startRow.value(), colId + m_startCol.value());
|
||||
}
|
||||
|
||||
EIGEN_DEVICE_FUNC
|
||||
EIGEN_STRONG_INLINE const CoeffReturnType coeff(Index rowId, Index colId) const
|
||||
{
|
||||
return m_xpr.coeff(rowId + m_startRow.value(), colId + m_startCol.value());
|
||||
}
|
||||
|
||||
EIGEN_DEVICE_FUNC
|
||||
inline Scalar& coeffRef(Index index)
|
||||
{
|
||||
EIGEN_STATIC_ASSERT_LVALUE(XprType)
|
||||
return m_xpr.coeffRef(m_startRow.value() + (RowsAtCompileTime == 1 ? 0 : index),
|
||||
m_startCol.value() + (RowsAtCompileTime == 1 ? index : 0));
|
||||
}
|
||||
|
||||
EIGEN_DEVICE_FUNC
|
||||
inline const Scalar& coeffRef(Index index) const
|
||||
{
|
||||
return m_xpr.coeffRef(m_startRow.value() + (RowsAtCompileTime == 1 ? 0 : index),
|
||||
m_startCol.value() + (RowsAtCompileTime == 1 ? index : 0));
|
||||
}
|
||||
|
||||
EIGEN_DEVICE_FUNC
|
||||
inline const CoeffReturnType coeff(Index index) const
|
||||
{
|
||||
return m_xpr.coeff(m_startRow.value() + (RowsAtCompileTime == 1 ? 0 : index),
|
||||
m_startCol.value() + (RowsAtCompileTime == 1 ? index : 0));
|
||||
}
|
||||
|
||||
template<int LoadMode>
|
||||
inline PacketScalar packet(Index rowId, Index colId) const
|
||||
{
|
||||
return m_xpr.template packet<Unaligned>(rowId + m_startRow.value(), colId + m_startCol.value());
|
||||
}
|
||||
|
||||
template<int LoadMode>
|
||||
inline void writePacket(Index rowId, Index colId, const PacketScalar& val)
|
||||
{
|
||||
m_xpr.template writePacket<Unaligned>(rowId + m_startRow.value(), colId + m_startCol.value(), val);
|
||||
}
|
||||
|
||||
template<int LoadMode>
|
||||
inline PacketScalar packet(Index index) const
|
||||
{
|
||||
return m_xpr.template packet<Unaligned>
|
||||
(m_startRow.value() + (RowsAtCompileTime == 1 ? 0 : index),
|
||||
m_startCol.value() + (RowsAtCompileTime == 1 ? index : 0));
|
||||
}
|
||||
|
||||
template<int LoadMode>
|
||||
inline void writePacket(Index index, const PacketScalar& val)
|
||||
{
|
||||
m_xpr.template writePacket<Unaligned>
|
||||
(m_startRow.value() + (RowsAtCompileTime == 1 ? 0 : index),
|
||||
m_startCol.value() + (RowsAtCompileTime == 1 ? index : 0), val);
|
||||
}
|
||||
|
||||
#ifdef EIGEN_PARSED_BY_DOXYGEN
|
||||
/** \sa MapBase::data() */
|
||||
EIGEN_DEVICE_FUNC inline const Scalar* data() const;
|
||||
EIGEN_DEVICE_FUNC inline Index innerStride() const;
|
||||
EIGEN_DEVICE_FUNC inline Index outerStride() const;
|
||||
#endif
|
||||
|
||||
EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
|
||||
const typename internal::remove_all<XprTypeNested>::type& nestedExpression() const
|
||||
{
|
||||
return m_xpr;
|
||||
}
|
||||
|
||||
EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
|
||||
XprType& nestedExpression() { return m_xpr; }
|
||||
|
||||
EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE EIGEN_CONSTEXPR
|
||||
StorageIndex startRow() const EIGEN_NOEXCEPT
|
||||
{
|
||||
return m_startRow.value();
|
||||
}
|
||||
|
||||
EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE EIGEN_CONSTEXPR
|
||||
StorageIndex startCol() const EIGEN_NOEXCEPT
|
||||
{
|
||||
return m_startCol.value();
|
||||
}
|
||||
|
||||
protected:
|
||||
|
||||
XprTypeNested m_xpr;
|
||||
const internal::variable_if_dynamic<StorageIndex, (XprType::RowsAtCompileTime == 1 && BlockRows==1) ? 0 : Dynamic> m_startRow;
|
||||
const internal::variable_if_dynamic<StorageIndex, (XprType::ColsAtCompileTime == 1 && BlockCols==1) ? 0 : Dynamic> m_startCol;
|
||||
const internal::variable_if_dynamic<StorageIndex, RowsAtCompileTime> m_blockRows;
|
||||
const internal::variable_if_dynamic<StorageIndex, ColsAtCompileTime> m_blockCols;
|
||||
};
|
||||
|
||||
/** \internal Internal implementation of dense Blocks in the direct access case.*/
|
||||
template<typename XprType, int BlockRows, int BlockCols, bool InnerPanel>
|
||||
class BlockImpl_dense<XprType,BlockRows,BlockCols, InnerPanel,true>
|
||||
: public MapBase<Block<XprType, BlockRows, BlockCols, InnerPanel> >
|
||||
{
|
||||
typedef Block<XprType, BlockRows, BlockCols, InnerPanel> BlockType;
|
||||
typedef typename internal::ref_selector<XprType>::non_const_type XprTypeNested;
|
||||
enum {
|
||||
XprTypeIsRowMajor = (int(traits<XprType>::Flags)&RowMajorBit) != 0
|
||||
};
|
||||
public:
|
||||
|
||||
typedef MapBase<BlockType> Base;
|
||||
EIGEN_DENSE_PUBLIC_INTERFACE(BlockType)
|
||||
EIGEN_INHERIT_ASSIGNMENT_OPERATORS(BlockImpl_dense)
|
||||
|
||||
/** Column or Row constructor
|
||||
*/
|
||||
EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
|
||||
BlockImpl_dense(XprType& xpr, Index i)
|
||||
: Base(xpr.data() + i * ( ((BlockRows==1) && (BlockCols==XprType::ColsAtCompileTime) && (!XprTypeIsRowMajor))
|
||||
|| ((BlockRows==XprType::RowsAtCompileTime) && (BlockCols==1) && ( XprTypeIsRowMajor)) ? xpr.innerStride() : xpr.outerStride()),
|
||||
BlockRows==1 ? 1 : xpr.rows(),
|
||||
BlockCols==1 ? 1 : xpr.cols()),
|
||||
m_xpr(xpr),
|
||||
m_startRow( (BlockRows==1) && (BlockCols==XprType::ColsAtCompileTime) ? i : 0),
|
||||
m_startCol( (BlockRows==XprType::RowsAtCompileTime) && (BlockCols==1) ? i : 0)
|
||||
{
|
||||
init();
|
||||
}
|
||||
|
||||
/** Fixed-size constructor
|
||||
*/
|
||||
EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
|
||||
BlockImpl_dense(XprType& xpr, Index startRow, Index startCol)
|
||||
: Base(xpr.data()+xpr.innerStride()*(XprTypeIsRowMajor?startCol:startRow) + xpr.outerStride()*(XprTypeIsRowMajor?startRow:startCol)),
|
||||
m_xpr(xpr), m_startRow(startRow), m_startCol(startCol)
|
||||
{
|
||||
init();
|
||||
}
|
||||
|
||||
/** Dynamic-size constructor
|
||||
*/
|
||||
EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
|
||||
BlockImpl_dense(XprType& xpr,
|
||||
Index startRow, Index startCol,
|
||||
Index blockRows, Index blockCols)
|
||||
: Base(xpr.data()+xpr.innerStride()*(XprTypeIsRowMajor?startCol:startRow) + xpr.outerStride()*(XprTypeIsRowMajor?startRow:startCol), blockRows, blockCols),
|
||||
m_xpr(xpr), m_startRow(startRow), m_startCol(startCol)
|
||||
{
|
||||
init();
|
||||
}
|
||||
|
||||
EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
|
||||
const typename internal::remove_all<XprTypeNested>::type& nestedExpression() const EIGEN_NOEXCEPT
|
||||
{
|
||||
return m_xpr;
|
||||
}
|
||||
|
||||
EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
|
||||
XprType& nestedExpression() { return m_xpr; }
|
||||
|
||||
/** \sa MapBase::innerStride() */
|
||||
EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE EIGEN_CONSTEXPR
|
||||
Index innerStride() const EIGEN_NOEXCEPT
|
||||
{
|
||||
return internal::traits<BlockType>::HasSameStorageOrderAsXprType
|
||||
? m_xpr.innerStride()
|
||||
: m_xpr.outerStride();
|
||||
}
|
||||
|
||||
/** \sa MapBase::outerStride() */
|
||||
EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE EIGEN_CONSTEXPR
|
||||
Index outerStride() const EIGEN_NOEXCEPT
|
||||
{
|
||||
return internal::traits<BlockType>::HasSameStorageOrderAsXprType
|
||||
? m_xpr.outerStride()
|
||||
: m_xpr.innerStride();
|
||||
}
|
||||
|
||||
EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE EIGEN_CONSTEXPR
|
||||
StorageIndex startRow() const EIGEN_NOEXCEPT { return m_startRow.value(); }
|
||||
|
||||
EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE EIGEN_CONSTEXPR
|
||||
StorageIndex startCol() const EIGEN_NOEXCEPT { return m_startCol.value(); }
|
||||
|
||||
#ifndef __SUNPRO_CC
|
||||
// FIXME sunstudio is not friendly with the above friend...
|
||||
// META-FIXME there is no 'friend' keyword around here. Is this obsolete?
|
||||
protected:
|
||||
#endif
|
||||
|
||||
#ifndef EIGEN_PARSED_BY_DOXYGEN
|
||||
/** \internal used by allowAligned() */
|
||||
EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
|
||||
BlockImpl_dense(XprType& xpr, const Scalar* data, Index blockRows, Index blockCols)
|
||||
: Base(data, blockRows, blockCols), m_xpr(xpr)
|
||||
{
|
||||
init();
|
||||
}
|
||||
#endif
|
||||
|
||||
protected:
|
||||
EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
|
||||
void init()
|
||||
{
|
||||
m_outerStride = internal::traits<BlockType>::HasSameStorageOrderAsXprType
|
||||
? m_xpr.outerStride()
|
||||
: m_xpr.innerStride();
|
||||
}
|
||||
|
||||
XprTypeNested m_xpr;
|
||||
const internal::variable_if_dynamic<StorageIndex, (XprType::RowsAtCompileTime == 1 && BlockRows==1) ? 0 : Dynamic> m_startRow;
|
||||
const internal::variable_if_dynamic<StorageIndex, (XprType::ColsAtCompileTime == 1 && BlockCols==1) ? 0 : Dynamic> m_startCol;
|
||||
Index m_outerStride;
|
||||
};
|
||||
|
||||
} // end namespace internal
|
||||
|
||||
} // end namespace Eigen
|
||||
|
||||
#endif // EIGEN_BLOCK_H
|
||||
@@ -0,0 +1,162 @@
|
||||
// This file is part of Eigen, a lightweight C++ template library
|
||||
// for linear algebra.
|
||||
//
|
||||
// Copyright (C) 2008 Gael Guennebaud <gael.guennebaud@inria.fr>
|
||||
//
|
||||
// This Source Code Form is subject to the terms of the Mozilla
|
||||
// Public License v. 2.0. If a copy of the MPL was not distributed
|
||||
// with this file, You can obtain one at http://mozilla.org/MPL/2.0/.
|
||||
|
||||
#ifndef EIGEN_ALLANDANY_H
|
||||
#define EIGEN_ALLANDANY_H
|
||||
|
||||
namespace Eigen {
|
||||
|
||||
namespace internal {
|
||||
|
||||
template<typename Derived, int UnrollCount, int Rows>
|
||||
struct all_unroller
|
||||
{
|
||||
enum {
|
||||
col = (UnrollCount-1) / Rows,
|
||||
row = (UnrollCount-1) % Rows
|
||||
};
|
||||
|
||||
EIGEN_DEVICE_FUNC static inline bool run(const Derived &mat)
|
||||
{
|
||||
return all_unroller<Derived, UnrollCount-1, Rows>::run(mat) && mat.coeff(row, col);
|
||||
}
|
||||
};
|
||||
|
||||
template<typename Derived, int Rows>
|
||||
struct all_unroller<Derived, 0, Rows>
|
||||
{
|
||||
EIGEN_DEVICE_FUNC static inline bool run(const Derived &/*mat*/) { return true; }
|
||||
};
|
||||
|
||||
template<typename Derived, int Rows>
|
||||
struct all_unroller<Derived, Dynamic, Rows>
|
||||
{
|
||||
EIGEN_DEVICE_FUNC static inline bool run(const Derived &) { return false; }
|
||||
};
|
||||
|
||||
template<typename Derived, int UnrollCount, int Rows>
|
||||
struct any_unroller
|
||||
{
|
||||
enum {
|
||||
col = (UnrollCount-1) / Rows,
|
||||
row = (UnrollCount-1) % Rows
|
||||
};
|
||||
|
||||
EIGEN_DEVICE_FUNC static inline bool run(const Derived &mat)
|
||||
{
|
||||
return any_unroller<Derived, UnrollCount-1, Rows>::run(mat) || mat.coeff(row, col);
|
||||
}
|
||||
};
|
||||
|
||||
template<typename Derived, int Rows>
|
||||
struct any_unroller<Derived, 0, Rows>
|
||||
{
|
||||
EIGEN_DEVICE_FUNC static inline bool run(const Derived & /*mat*/) { return false; }
|
||||
};
|
||||
|
||||
template<typename Derived, int Rows>
|
||||
struct any_unroller<Derived, Dynamic, Rows>
|
||||
{
|
||||
EIGEN_DEVICE_FUNC static inline bool run(const Derived &) { return false; }
|
||||
};
|
||||
|
||||
} // end namespace internal
|
||||
|
||||
/** \returns true if all coefficients are true
|
||||
*
|
||||
* Example: \include MatrixBase_all.cpp
|
||||
* Output: \verbinclude MatrixBase_all.out
|
||||
*
|
||||
* \sa any(), Cwise::operator<()
|
||||
*/
|
||||
template<typename Derived>
|
||||
EIGEN_DEVICE_FUNC inline bool DenseBase<Derived>::all() const
|
||||
{
|
||||
typedef internal::evaluator<Derived> Evaluator;
|
||||
enum {
|
||||
unroll = SizeAtCompileTime != Dynamic
|
||||
&& SizeAtCompileTime * (int(Evaluator::CoeffReadCost) + int(NumTraits<Scalar>::AddCost)) <= EIGEN_UNROLLING_LIMIT
|
||||
};
|
||||
Evaluator evaluator(derived());
|
||||
if(unroll)
|
||||
return internal::all_unroller<Evaluator, unroll ? int(SizeAtCompileTime) : Dynamic, internal::traits<Derived>::RowsAtCompileTime>::run(evaluator);
|
||||
else
|
||||
{
|
||||
for(Index j = 0; j < cols(); ++j)
|
||||
for(Index i = 0; i < rows(); ++i)
|
||||
if (!evaluator.coeff(i, j)) return false;
|
||||
return true;
|
||||
}
|
||||
}
|
||||
|
||||
/** \returns true if at least one coefficient is true
|
||||
*
|
||||
* \sa all()
|
||||
*/
|
||||
template<typename Derived>
|
||||
EIGEN_DEVICE_FUNC inline bool DenseBase<Derived>::any() const
|
||||
{
|
||||
typedef internal::evaluator<Derived> Evaluator;
|
||||
enum {
|
||||
unroll = SizeAtCompileTime != Dynamic
|
||||
&& SizeAtCompileTime * (int(Evaluator::CoeffReadCost) + int(NumTraits<Scalar>::AddCost)) <= EIGEN_UNROLLING_LIMIT
|
||||
};
|
||||
Evaluator evaluator(derived());
|
||||
if(unroll)
|
||||
return internal::any_unroller<Evaluator, unroll ? int(SizeAtCompileTime) : Dynamic, internal::traits<Derived>::RowsAtCompileTime>::run(evaluator);
|
||||
else
|
||||
{
|
||||
for(Index j = 0; j < cols(); ++j)
|
||||
for(Index i = 0; i < rows(); ++i)
|
||||
if (evaluator.coeff(i, j)) return true;
|
||||
return false;
|
||||
}
|
||||
}
|
||||
|
||||
/** \returns the number of coefficients which evaluate to true
|
||||
*
|
||||
* \sa all(), any()
|
||||
*/
|
||||
template<typename Derived>
|
||||
EIGEN_DEVICE_FUNC inline Eigen::Index DenseBase<Derived>::count() const
|
||||
{
|
||||
return derived().template cast<bool>().template cast<Index>().sum();
|
||||
}
|
||||
|
||||
/** \returns true is \c *this contains at least one Not A Number (NaN).
|
||||
*
|
||||
* \sa allFinite()
|
||||
*/
|
||||
template<typename Derived>
|
||||
inline bool DenseBase<Derived>::hasNaN() const
|
||||
{
|
||||
#if EIGEN_COMP_MSVC || (defined __FAST_MATH__)
|
||||
return derived().array().isNaN().any();
|
||||
#else
|
||||
return !((derived().array()==derived().array()).all());
|
||||
#endif
|
||||
}
|
||||
|
||||
/** \returns true if \c *this contains only finite numbers, i.e., no NaN and no +/-INF values.
|
||||
*
|
||||
* \sa hasNaN()
|
||||
*/
|
||||
template<typename Derived>
|
||||
inline bool DenseBase<Derived>::allFinite() const
|
||||
{
|
||||
#if EIGEN_COMP_MSVC || (defined __FAST_MATH__)
|
||||
return derived().array().isFinite().all();
|
||||
#else
|
||||
return !((derived()-derived()).hasNaN());
|
||||
#endif
|
||||
}
|
||||
|
||||
} // end namespace Eigen
|
||||
|
||||
#endif // EIGEN_ALLANDANY_H
|
||||
@@ -0,0 +1,164 @@
|
||||
// This file is part of Eigen, a lightweight C++ template library
|
||||
// for linear algebra.
|
||||
//
|
||||
// Copyright (C) 2008 Gael Guennebaud <gael.guennebaud@inria.fr>
|
||||
// Copyright (C) 2006-2008 Benoit Jacob <jacob.benoit.1@gmail.com>
|
||||
//
|
||||
// This Source Code Form is subject to the terms of the Mozilla
|
||||
// Public License v. 2.0. If a copy of the MPL was not distributed
|
||||
// with this file, You can obtain one at http://mozilla.org/MPL/2.0/.
|
||||
|
||||
#ifndef EIGEN_COMMAINITIALIZER_H
|
||||
#define EIGEN_COMMAINITIALIZER_H
|
||||
|
||||
namespace Eigen {
|
||||
|
||||
/** \class CommaInitializer
|
||||
* \ingroup Core_Module
|
||||
*
|
||||
* \brief Helper class used by the comma initializer operator
|
||||
*
|
||||
* This class is internally used to implement the comma initializer feature. It is
|
||||
* the return type of MatrixBase::operator<<, and most of the time this is the only
|
||||
* way it is used.
|
||||
*
|
||||
* \sa \blank \ref MatrixBaseCommaInitRef "MatrixBase::operator<<", CommaInitializer::finished()
|
||||
*/
|
||||
template<typename XprType>
|
||||
struct CommaInitializer
|
||||
{
|
||||
typedef typename XprType::Scalar Scalar;
|
||||
|
||||
EIGEN_DEVICE_FUNC
|
||||
inline CommaInitializer(XprType& xpr, const Scalar& s)
|
||||
: m_xpr(xpr), m_row(0), m_col(1), m_currentBlockRows(1)
|
||||
{
|
||||
eigen_assert(m_xpr.rows() > 0 && m_xpr.cols() > 0
|
||||
&& "Cannot comma-initialize a 0x0 matrix (operator<<)");
|
||||
m_xpr.coeffRef(0,0) = s;
|
||||
}
|
||||
|
||||
template<typename OtherDerived>
|
||||
EIGEN_DEVICE_FUNC
|
||||
inline CommaInitializer(XprType& xpr, const DenseBase<OtherDerived>& other)
|
||||
: m_xpr(xpr), m_row(0), m_col(other.cols()), m_currentBlockRows(other.rows())
|
||||
{
|
||||
eigen_assert(m_xpr.rows() >= other.rows() && m_xpr.cols() >= other.cols()
|
||||
&& "Cannot comma-initialize a 0x0 matrix (operator<<)");
|
||||
m_xpr.block(0, 0, other.rows(), other.cols()) = other;
|
||||
}
|
||||
|
||||
/* Copy/Move constructor which transfers ownership. This is crucial in
|
||||
* absence of return value optimization to avoid assertions during destruction. */
|
||||
// FIXME in C++11 mode this could be replaced by a proper RValue constructor
|
||||
EIGEN_DEVICE_FUNC
|
||||
inline CommaInitializer(const CommaInitializer& o)
|
||||
: m_xpr(o.m_xpr), m_row(o.m_row), m_col(o.m_col), m_currentBlockRows(o.m_currentBlockRows) {
|
||||
// Mark original object as finished. In absence of R-value references we need to const_cast:
|
||||
const_cast<CommaInitializer&>(o).m_row = m_xpr.rows();
|
||||
const_cast<CommaInitializer&>(o).m_col = m_xpr.cols();
|
||||
const_cast<CommaInitializer&>(o).m_currentBlockRows = 0;
|
||||
}
|
||||
|
||||
/* inserts a scalar value in the target matrix */
|
||||
EIGEN_DEVICE_FUNC
|
||||
CommaInitializer& operator,(const Scalar& s)
|
||||
{
|
||||
if (m_col==m_xpr.cols())
|
||||
{
|
||||
m_row+=m_currentBlockRows;
|
||||
m_col = 0;
|
||||
m_currentBlockRows = 1;
|
||||
eigen_assert(m_row<m_xpr.rows()
|
||||
&& "Too many rows passed to comma initializer (operator<<)");
|
||||
}
|
||||
eigen_assert(m_col<m_xpr.cols()
|
||||
&& "Too many coefficients passed to comma initializer (operator<<)");
|
||||
eigen_assert(m_currentBlockRows==1);
|
||||
m_xpr.coeffRef(m_row, m_col++) = s;
|
||||
return *this;
|
||||
}
|
||||
|
||||
/* inserts a matrix expression in the target matrix */
|
||||
template<typename OtherDerived>
|
||||
EIGEN_DEVICE_FUNC
|
||||
CommaInitializer& operator,(const DenseBase<OtherDerived>& other)
|
||||
{
|
||||
if (m_col==m_xpr.cols() && (other.cols()!=0 || other.rows()!=m_currentBlockRows))
|
||||
{
|
||||
m_row+=m_currentBlockRows;
|
||||
m_col = 0;
|
||||
m_currentBlockRows = other.rows();
|
||||
eigen_assert(m_row+m_currentBlockRows<=m_xpr.rows()
|
||||
&& "Too many rows passed to comma initializer (operator<<)");
|
||||
}
|
||||
eigen_assert((m_col + other.cols() <= m_xpr.cols())
|
||||
&& "Too many coefficients passed to comma initializer (operator<<)");
|
||||
eigen_assert(m_currentBlockRows==other.rows());
|
||||
m_xpr.template block<OtherDerived::RowsAtCompileTime, OtherDerived::ColsAtCompileTime>
|
||||
(m_row, m_col, other.rows(), other.cols()) = other;
|
||||
m_col += other.cols();
|
||||
return *this;
|
||||
}
|
||||
|
||||
EIGEN_DEVICE_FUNC
|
||||
inline ~CommaInitializer()
|
||||
#if defined VERIFY_RAISES_ASSERT && (!defined EIGEN_NO_ASSERTION_CHECKING) && defined EIGEN_EXCEPTIONS
|
||||
EIGEN_EXCEPTION_SPEC(Eigen::eigen_assert_exception)
|
||||
#endif
|
||||
{
|
||||
finished();
|
||||
}
|
||||
|
||||
/** \returns the built matrix once all its coefficients have been set.
|
||||
* Calling finished is 100% optional. Its purpose is to write expressions
|
||||
* like this:
|
||||
* \code
|
||||
* quaternion.fromRotationMatrix((Matrix3f() << axis0, axis1, axis2).finished());
|
||||
* \endcode
|
||||
*/
|
||||
EIGEN_DEVICE_FUNC
|
||||
inline XprType& finished() {
|
||||
eigen_assert(((m_row+m_currentBlockRows) == m_xpr.rows() || m_xpr.cols() == 0)
|
||||
&& m_col == m_xpr.cols()
|
||||
&& "Too few coefficients passed to comma initializer (operator<<)");
|
||||
return m_xpr;
|
||||
}
|
||||
|
||||
XprType& m_xpr; // target expression
|
||||
Index m_row; // current row id
|
||||
Index m_col; // current col id
|
||||
Index m_currentBlockRows; // current block height
|
||||
};
|
||||
|
||||
/** \anchor MatrixBaseCommaInitRef
|
||||
* Convenient operator to set the coefficients of a matrix.
|
||||
*
|
||||
* The coefficients must be provided in a row major order and exactly match
|
||||
* the size of the matrix. Otherwise an assertion is raised.
|
||||
*
|
||||
* Example: \include MatrixBase_set.cpp
|
||||
* Output: \verbinclude MatrixBase_set.out
|
||||
*
|
||||
* \note According the c++ standard, the argument expressions of this comma initializer are evaluated in arbitrary order.
|
||||
*
|
||||
* \sa CommaInitializer::finished(), class CommaInitializer
|
||||
*/
|
||||
template<typename Derived>
|
||||
EIGEN_DEVICE_FUNC inline CommaInitializer<Derived> DenseBase<Derived>::operator<< (const Scalar& s)
|
||||
{
|
||||
return CommaInitializer<Derived>(*static_cast<Derived*>(this), s);
|
||||
}
|
||||
|
||||
/** \sa operator<<(const Scalar&) */
|
||||
template<typename Derived>
|
||||
template<typename OtherDerived>
|
||||
EIGEN_DEVICE_FUNC inline CommaInitializer<Derived>
|
||||
DenseBase<Derived>::operator<<(const DenseBase<OtherDerived>& other)
|
||||
{
|
||||
return CommaInitializer<Derived>(*static_cast<Derived *>(this), other);
|
||||
}
|
||||
|
||||
} // end namespace Eigen
|
||||
|
||||
#endif // EIGEN_COMMAINITIALIZER_H
|
||||
@@ -0,0 +1,175 @@
|
||||
// This file is part of Eigen, a lightweight C++ template library
|
||||
// for linear algebra.
|
||||
//
|
||||
// Copyright (C) 2016 Rasmus Munk Larsen (rmlarsen@google.com)
|
||||
//
|
||||
// This Source Code Form is subject to the terms of the Mozilla
|
||||
// Public License v. 2.0. If a copy of the MPL was not distributed
|
||||
// with this file, You can obtain one at http://mozilla.org/MPL/2.0/.
|
||||
|
||||
#ifndef EIGEN_CONDITIONESTIMATOR_H
|
||||
#define EIGEN_CONDITIONESTIMATOR_H
|
||||
|
||||
namespace Eigen {
|
||||
|
||||
namespace internal {
|
||||
|
||||
template <typename Vector, typename RealVector, bool IsComplex>
|
||||
struct rcond_compute_sign {
|
||||
static inline Vector run(const Vector& v) {
|
||||
const RealVector v_abs = v.cwiseAbs();
|
||||
return (v_abs.array() == static_cast<typename Vector::RealScalar>(0))
|
||||
.select(Vector::Ones(v.size()), v.cwiseQuotient(v_abs));
|
||||
}
|
||||
};
|
||||
|
||||
// Partial specialization to avoid elementwise division for real vectors.
|
||||
template <typename Vector>
|
||||
struct rcond_compute_sign<Vector, Vector, false> {
|
||||
static inline Vector run(const Vector& v) {
|
||||
return (v.array() < static_cast<typename Vector::RealScalar>(0))
|
||||
.select(-Vector::Ones(v.size()), Vector::Ones(v.size()));
|
||||
}
|
||||
};
|
||||
|
||||
/**
|
||||
* \returns an estimate of ||inv(matrix)||_1 given a decomposition of
|
||||
* \a matrix that implements .solve() and .adjoint().solve() methods.
|
||||
*
|
||||
* This function implements Algorithms 4.1 and 5.1 from
|
||||
* http://www.maths.manchester.ac.uk/~higham/narep/narep135.pdf
|
||||
* which also forms the basis for the condition number estimators in
|
||||
* LAPACK. Since at most 10 calls to the solve method of dec are
|
||||
* performed, the total cost is O(dims^2), as opposed to O(dims^3)
|
||||
* needed to compute the inverse matrix explicitly.
|
||||
*
|
||||
* The most common usage is in estimating the condition number
|
||||
* ||matrix||_1 * ||inv(matrix)||_1. The first term ||matrix||_1 can be
|
||||
* computed directly in O(n^2) operations.
|
||||
*
|
||||
* Supports the following decompositions: FullPivLU, PartialPivLU, LDLT, and
|
||||
* LLT.
|
||||
*
|
||||
* \sa FullPivLU, PartialPivLU, LDLT, LLT.
|
||||
*/
|
||||
template <typename Decomposition>
|
||||
typename Decomposition::RealScalar rcond_invmatrix_L1_norm_estimate(const Decomposition& dec)
|
||||
{
|
||||
typedef typename Decomposition::MatrixType MatrixType;
|
||||
typedef typename Decomposition::Scalar Scalar;
|
||||
typedef typename Decomposition::RealScalar RealScalar;
|
||||
typedef typename internal::plain_col_type<MatrixType>::type Vector;
|
||||
typedef typename internal::plain_col_type<MatrixType, RealScalar>::type RealVector;
|
||||
const bool is_complex = (NumTraits<Scalar>::IsComplex != 0);
|
||||
|
||||
eigen_assert(dec.rows() == dec.cols());
|
||||
const Index n = dec.rows();
|
||||
if (n == 0)
|
||||
return 0;
|
||||
|
||||
// Disable Index to float conversion warning
|
||||
#ifdef __INTEL_COMPILER
|
||||
#pragma warning push
|
||||
#pragma warning ( disable : 2259 )
|
||||
#endif
|
||||
Vector v = dec.solve(Vector::Ones(n) / Scalar(n));
|
||||
#ifdef __INTEL_COMPILER
|
||||
#pragma warning pop
|
||||
#endif
|
||||
|
||||
// lower_bound is a lower bound on
|
||||
// ||inv(matrix)||_1 = sup_v ||inv(matrix) v||_1 / ||v||_1
|
||||
// and is the objective maximized by the ("super-") gradient ascent
|
||||
// algorithm below.
|
||||
RealScalar lower_bound = v.template lpNorm<1>();
|
||||
if (n == 1)
|
||||
return lower_bound;
|
||||
|
||||
// Gradient ascent algorithm follows: We know that the optimum is achieved at
|
||||
// one of the simplices v = e_i, so in each iteration we follow a
|
||||
// super-gradient to move towards the optimal one.
|
||||
RealScalar old_lower_bound = lower_bound;
|
||||
Vector sign_vector(n);
|
||||
Vector old_sign_vector;
|
||||
Index v_max_abs_index = -1;
|
||||
Index old_v_max_abs_index = v_max_abs_index;
|
||||
for (int k = 0; k < 4; ++k)
|
||||
{
|
||||
sign_vector = internal::rcond_compute_sign<Vector, RealVector, is_complex>::run(v);
|
||||
if (k > 0 && !is_complex && sign_vector == old_sign_vector) {
|
||||
// Break if the solution stagnated.
|
||||
break;
|
||||
}
|
||||
// v_max_abs_index = argmax |real( inv(matrix)^T * sign_vector )|
|
||||
v = dec.adjoint().solve(sign_vector);
|
||||
v.real().cwiseAbs().maxCoeff(&v_max_abs_index);
|
||||
if (v_max_abs_index == old_v_max_abs_index) {
|
||||
// Break if the solution stagnated.
|
||||
break;
|
||||
}
|
||||
// Move to the new simplex e_j, where j = v_max_abs_index.
|
||||
v = dec.solve(Vector::Unit(n, v_max_abs_index)); // v = inv(matrix) * e_j.
|
||||
lower_bound = v.template lpNorm<1>();
|
||||
if (lower_bound <= old_lower_bound) {
|
||||
// Break if the gradient step did not increase the lower_bound.
|
||||
break;
|
||||
}
|
||||
if (!is_complex) {
|
||||
old_sign_vector = sign_vector;
|
||||
}
|
||||
old_v_max_abs_index = v_max_abs_index;
|
||||
old_lower_bound = lower_bound;
|
||||
}
|
||||
// The following calculates an independent estimate of ||matrix||_1 by
|
||||
// multiplying matrix by a vector with entries of slowly increasing
|
||||
// magnitude and alternating sign:
|
||||
// v_i = (-1)^{i} (1 + (i / (dim-1))), i = 0,...,dim-1.
|
||||
// This improvement to Hager's algorithm above is due to Higham. It was
|
||||
// added to make the algorithm more robust in certain corner cases where
|
||||
// large elements in the matrix might otherwise escape detection due to
|
||||
// exact cancellation (especially when op and op_adjoint correspond to a
|
||||
// sequence of backsubstitutions and permutations), which could cause
|
||||
// Hager's algorithm to vastly underestimate ||matrix||_1.
|
||||
Scalar alternating_sign(RealScalar(1));
|
||||
for (Index i = 0; i < n; ++i) {
|
||||
// The static_cast is needed when Scalar is a complex and RealScalar implements expression templates
|
||||
v[i] = alternating_sign * static_cast<RealScalar>(RealScalar(1) + (RealScalar(i) / (RealScalar(n - 1))));
|
||||
alternating_sign = -alternating_sign;
|
||||
}
|
||||
v = dec.solve(v);
|
||||
const RealScalar alternate_lower_bound = (2 * v.template lpNorm<1>()) / (3 * RealScalar(n));
|
||||
return numext::maxi(lower_bound, alternate_lower_bound);
|
||||
}
|
||||
|
||||
/** \brief Reciprocal condition number estimator.
|
||||
*
|
||||
* Computing a decomposition of a dense matrix takes O(n^3) operations, while
|
||||
* this method estimates the condition number quickly and reliably in O(n^2)
|
||||
* operations.
|
||||
*
|
||||
* \returns an estimate of the reciprocal condition number
|
||||
* (1 / (||matrix||_1 * ||inv(matrix)||_1)) of matrix, given ||matrix||_1 and
|
||||
* its decomposition. Supports the following decompositions: FullPivLU,
|
||||
* PartialPivLU, LDLT, and LLT.
|
||||
*
|
||||
* \sa FullPivLU, PartialPivLU, LDLT, LLT.
|
||||
*/
|
||||
template <typename Decomposition>
|
||||
typename Decomposition::RealScalar
|
||||
rcond_estimate_helper(typename Decomposition::RealScalar matrix_norm, const Decomposition& dec)
|
||||
{
|
||||
typedef typename Decomposition::RealScalar RealScalar;
|
||||
eigen_assert(dec.rows() == dec.cols());
|
||||
if (dec.rows() == 0) return NumTraits<RealScalar>::infinity();
|
||||
if (matrix_norm == RealScalar(0)) return RealScalar(0);
|
||||
if (dec.rows() == 1) return RealScalar(1);
|
||||
const RealScalar inverse_matrix_norm = rcond_invmatrix_L1_norm_estimate(dec);
|
||||
return (inverse_matrix_norm == RealScalar(0) ? RealScalar(0)
|
||||
: (RealScalar(1) / inverse_matrix_norm) / matrix_norm);
|
||||
}
|
||||
|
||||
} // namespace internal
|
||||
|
||||
} // namespace Eigen
|
||||
|
||||
#endif
|
||||
File diff suppressed because it is too large
Load Diff
@@ -0,0 +1,132 @@
|
||||
// This file is part of Eigen, a lightweight C++ template library
|
||||
// for linear algebra.
|
||||
//
|
||||
// Copyright (C) 2008-2014 Gael Guennebaud <gael.guennebaud@inria.fr>
|
||||
//
|
||||
// This Source Code Form is subject to the terms of the Mozilla
|
||||
// Public License v. 2.0. If a copy of the MPL was not distributed
|
||||
// with this file, You can obtain one at http://mozilla.org/MPL/2.0/.
|
||||
|
||||
#ifndef EIGEN_COREITERATORS_H
|
||||
#define EIGEN_COREITERATORS_H
|
||||
|
||||
namespace Eigen {
|
||||
|
||||
/* This file contains the respective InnerIterator definition of the expressions defined in Eigen/Core
|
||||
*/
|
||||
|
||||
namespace internal {
|
||||
|
||||
template<typename XprType, typename EvaluatorKind>
|
||||
class inner_iterator_selector;
|
||||
|
||||
}
|
||||
|
||||
/** \class InnerIterator
|
||||
* \brief An InnerIterator allows to loop over the element of any matrix expression.
|
||||
*
|
||||
* \warning To be used with care because an evaluator is constructed every time an InnerIterator iterator is constructed.
|
||||
*
|
||||
* TODO: add a usage example
|
||||
*/
|
||||
template<typename XprType>
|
||||
class InnerIterator
|
||||
{
|
||||
protected:
|
||||
typedef internal::inner_iterator_selector<XprType, typename internal::evaluator_traits<XprType>::Kind> IteratorType;
|
||||
typedef internal::evaluator<XprType> EvaluatorType;
|
||||
typedef typename internal::traits<XprType>::Scalar Scalar;
|
||||
public:
|
||||
/** Construct an iterator over the \a outerId -th row or column of \a xpr */
|
||||
InnerIterator(const XprType &xpr, const Index &outerId)
|
||||
: m_eval(xpr), m_iter(m_eval, outerId, xpr.innerSize())
|
||||
{}
|
||||
|
||||
/// \returns the value of the current coefficient.
|
||||
EIGEN_STRONG_INLINE Scalar value() const { return m_iter.value(); }
|
||||
/** Increment the iterator \c *this to the next non-zero coefficient.
|
||||
* Explicit zeros are not skipped over. To skip explicit zeros, see class SparseView
|
||||
*/
|
||||
EIGEN_STRONG_INLINE InnerIterator& operator++() { m_iter.operator++(); return *this; }
|
||||
EIGEN_STRONG_INLINE InnerIterator& operator+=(Index i) { m_iter.operator+=(i); return *this; }
|
||||
EIGEN_STRONG_INLINE InnerIterator operator+(Index i)
|
||||
{ InnerIterator result(*this); result+=i; return result; }
|
||||
|
||||
|
||||
/// \returns the column or row index of the current coefficient.
|
||||
EIGEN_STRONG_INLINE Index index() const { return m_iter.index(); }
|
||||
/// \returns the row index of the current coefficient.
|
||||
EIGEN_STRONG_INLINE Index row() const { return m_iter.row(); }
|
||||
/// \returns the column index of the current coefficient.
|
||||
EIGEN_STRONG_INLINE Index col() const { return m_iter.col(); }
|
||||
/// \returns \c true if the iterator \c *this still references a valid coefficient.
|
||||
EIGEN_STRONG_INLINE operator bool() const { return m_iter; }
|
||||
|
||||
protected:
|
||||
EvaluatorType m_eval;
|
||||
IteratorType m_iter;
|
||||
private:
|
||||
// If you get here, then you're not using the right InnerIterator type, e.g.:
|
||||
// SparseMatrix<double,RowMajor> A;
|
||||
// SparseMatrix<double>::InnerIterator it(A,0);
|
||||
template<typename T> InnerIterator(const EigenBase<T>&,Index outer);
|
||||
};
|
||||
|
||||
namespace internal {
|
||||
|
||||
// Generic inner iterator implementation for dense objects
|
||||
template<typename XprType>
|
||||
class inner_iterator_selector<XprType, IndexBased>
|
||||
{
|
||||
protected:
|
||||
typedef evaluator<XprType> EvaluatorType;
|
||||
typedef typename traits<XprType>::Scalar Scalar;
|
||||
enum { IsRowMajor = (XprType::Flags&RowMajorBit)==RowMajorBit };
|
||||
|
||||
public:
|
||||
EIGEN_STRONG_INLINE inner_iterator_selector(const EvaluatorType &eval, const Index &outerId, const Index &innerSize)
|
||||
: m_eval(eval), m_inner(0), m_outer(outerId), m_end(innerSize)
|
||||
{}
|
||||
|
||||
EIGEN_STRONG_INLINE Scalar value() const
|
||||
{
|
||||
return (IsRowMajor) ? m_eval.coeff(m_outer, m_inner)
|
||||
: m_eval.coeff(m_inner, m_outer);
|
||||
}
|
||||
|
||||
EIGEN_STRONG_INLINE inner_iterator_selector& operator++() { m_inner++; return *this; }
|
||||
|
||||
EIGEN_STRONG_INLINE Index index() const { return m_inner; }
|
||||
inline Index row() const { return IsRowMajor ? m_outer : index(); }
|
||||
inline Index col() const { return IsRowMajor ? index() : m_outer; }
|
||||
|
||||
EIGEN_STRONG_INLINE operator bool() const { return m_inner < m_end && m_inner>=0; }
|
||||
|
||||
protected:
|
||||
const EvaluatorType& m_eval;
|
||||
Index m_inner;
|
||||
const Index m_outer;
|
||||
const Index m_end;
|
||||
};
|
||||
|
||||
// For iterator-based evaluator, inner-iterator is already implemented as
|
||||
// evaluator<>::InnerIterator
|
||||
template<typename XprType>
|
||||
class inner_iterator_selector<XprType, IteratorBased>
|
||||
: public evaluator<XprType>::InnerIterator
|
||||
{
|
||||
protected:
|
||||
typedef typename evaluator<XprType>::InnerIterator Base;
|
||||
typedef evaluator<XprType> EvaluatorType;
|
||||
|
||||
public:
|
||||
EIGEN_STRONG_INLINE inner_iterator_selector(const EvaluatorType &eval, const Index &outerId, const Index &/*innerSize*/)
|
||||
: Base(eval, outerId)
|
||||
{}
|
||||
};
|
||||
|
||||
} // end namespace internal
|
||||
|
||||
} // end namespace Eigen
|
||||
|
||||
#endif // EIGEN_COREITERATORS_H
|
||||
@@ -0,0 +1,183 @@
|
||||
// This file is part of Eigen, a lightweight C++ template library
|
||||
// for linear algebra.
|
||||
//
|
||||
// Copyright (C) 2008-2014 Gael Guennebaud <gael.guennebaud@inria.fr>
|
||||
// Copyright (C) 2006-2008 Benoit Jacob <jacob.benoit.1@gmail.com>
|
||||
//
|
||||
// This Source Code Form is subject to the terms of the Mozilla
|
||||
// Public License v. 2.0. If a copy of the MPL was not distributed
|
||||
// with this file, You can obtain one at http://mozilla.org/MPL/2.0/.
|
||||
|
||||
#ifndef EIGEN_CWISE_BINARY_OP_H
|
||||
#define EIGEN_CWISE_BINARY_OP_H
|
||||
|
||||
namespace Eigen {
|
||||
|
||||
namespace internal {
|
||||
template<typename BinaryOp, typename Lhs, typename Rhs>
|
||||
struct traits<CwiseBinaryOp<BinaryOp, Lhs, Rhs> >
|
||||
{
|
||||
// we must not inherit from traits<Lhs> since it has
|
||||
// the potential to cause problems with MSVC
|
||||
typedef typename remove_all<Lhs>::type Ancestor;
|
||||
typedef typename traits<Ancestor>::XprKind XprKind;
|
||||
enum {
|
||||
RowsAtCompileTime = traits<Ancestor>::RowsAtCompileTime,
|
||||
ColsAtCompileTime = traits<Ancestor>::ColsAtCompileTime,
|
||||
MaxRowsAtCompileTime = traits<Ancestor>::MaxRowsAtCompileTime,
|
||||
MaxColsAtCompileTime = traits<Ancestor>::MaxColsAtCompileTime
|
||||
};
|
||||
|
||||
// even though we require Lhs and Rhs to have the same scalar type (see CwiseBinaryOp constructor),
|
||||
// we still want to handle the case when the result type is different.
|
||||
typedef typename result_of<
|
||||
BinaryOp(
|
||||
const typename Lhs::Scalar&,
|
||||
const typename Rhs::Scalar&
|
||||
)
|
||||
>::type Scalar;
|
||||
typedef typename cwise_promote_storage_type<typename traits<Lhs>::StorageKind,
|
||||
typename traits<Rhs>::StorageKind,
|
||||
BinaryOp>::ret StorageKind;
|
||||
typedef typename promote_index_type<typename traits<Lhs>::StorageIndex,
|
||||
typename traits<Rhs>::StorageIndex>::type StorageIndex;
|
||||
typedef typename Lhs::Nested LhsNested;
|
||||
typedef typename Rhs::Nested RhsNested;
|
||||
typedef typename remove_reference<LhsNested>::type _LhsNested;
|
||||
typedef typename remove_reference<RhsNested>::type _RhsNested;
|
||||
enum {
|
||||
Flags = cwise_promote_storage_order<typename traits<Lhs>::StorageKind,typename traits<Rhs>::StorageKind,_LhsNested::Flags & RowMajorBit,_RhsNested::Flags & RowMajorBit>::value
|
||||
};
|
||||
};
|
||||
} // end namespace internal
|
||||
|
||||
template<typename BinaryOp, typename Lhs, typename Rhs, typename StorageKind>
|
||||
class CwiseBinaryOpImpl;
|
||||
|
||||
/** \class CwiseBinaryOp
|
||||
* \ingroup Core_Module
|
||||
*
|
||||
* \brief Generic expression where a coefficient-wise binary operator is applied to two expressions
|
||||
*
|
||||
* \tparam BinaryOp template functor implementing the operator
|
||||
* \tparam LhsType the type of the left-hand side
|
||||
* \tparam RhsType the type of the right-hand side
|
||||
*
|
||||
* This class represents an expression where a coefficient-wise binary operator is applied to two expressions.
|
||||
* It is the return type of binary operators, by which we mean only those binary operators where
|
||||
* both the left-hand side and the right-hand side are Eigen expressions.
|
||||
* For example, the return type of matrix1+matrix2 is a CwiseBinaryOp.
|
||||
*
|
||||
* Most of the time, this is the only way that it is used, so you typically don't have to name
|
||||
* CwiseBinaryOp types explicitly.
|
||||
*
|
||||
* \sa MatrixBase::binaryExpr(const MatrixBase<OtherDerived> &,const CustomBinaryOp &) const, class CwiseUnaryOp, class CwiseNullaryOp
|
||||
*/
|
||||
template<typename BinaryOp, typename LhsType, typename RhsType>
|
||||
class CwiseBinaryOp :
|
||||
public CwiseBinaryOpImpl<
|
||||
BinaryOp, LhsType, RhsType,
|
||||
typename internal::cwise_promote_storage_type<typename internal::traits<LhsType>::StorageKind,
|
||||
typename internal::traits<RhsType>::StorageKind,
|
||||
BinaryOp>::ret>,
|
||||
internal::no_assignment_operator
|
||||
{
|
||||
public:
|
||||
|
||||
typedef typename internal::remove_all<BinaryOp>::type Functor;
|
||||
typedef typename internal::remove_all<LhsType>::type Lhs;
|
||||
typedef typename internal::remove_all<RhsType>::type Rhs;
|
||||
|
||||
typedef typename CwiseBinaryOpImpl<
|
||||
BinaryOp, LhsType, RhsType,
|
||||
typename internal::cwise_promote_storage_type<typename internal::traits<LhsType>::StorageKind,
|
||||
typename internal::traits<Rhs>::StorageKind,
|
||||
BinaryOp>::ret>::Base Base;
|
||||
EIGEN_GENERIC_PUBLIC_INTERFACE(CwiseBinaryOp)
|
||||
|
||||
typedef typename internal::ref_selector<LhsType>::type LhsNested;
|
||||
typedef typename internal::ref_selector<RhsType>::type RhsNested;
|
||||
typedef typename internal::remove_reference<LhsNested>::type _LhsNested;
|
||||
typedef typename internal::remove_reference<RhsNested>::type _RhsNested;
|
||||
|
||||
#if EIGEN_COMP_MSVC && EIGEN_HAS_CXX11
|
||||
//Required for Visual Studio or the Copy constructor will probably not get inlined!
|
||||
EIGEN_STRONG_INLINE
|
||||
CwiseBinaryOp(const CwiseBinaryOp<BinaryOp,LhsType,RhsType>&) = default;
|
||||
#endif
|
||||
|
||||
EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
|
||||
CwiseBinaryOp(const Lhs& aLhs, const Rhs& aRhs, const BinaryOp& func = BinaryOp())
|
||||
: m_lhs(aLhs), m_rhs(aRhs), m_functor(func)
|
||||
{
|
||||
EIGEN_CHECK_BINARY_COMPATIBILIY(BinaryOp,typename Lhs::Scalar,typename Rhs::Scalar);
|
||||
// require the sizes to match
|
||||
EIGEN_STATIC_ASSERT_SAME_MATRIX_SIZE(Lhs, Rhs)
|
||||
eigen_assert(aLhs.rows() == aRhs.rows() && aLhs.cols() == aRhs.cols());
|
||||
}
|
||||
|
||||
EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE EIGEN_CONSTEXPR
|
||||
Index rows() const EIGEN_NOEXCEPT {
|
||||
// return the fixed size type if available to enable compile time optimizations
|
||||
return internal::traits<typename internal::remove_all<LhsNested>::type>::RowsAtCompileTime==Dynamic ? m_rhs.rows() : m_lhs.rows();
|
||||
}
|
||||
EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE EIGEN_CONSTEXPR
|
||||
Index cols() const EIGEN_NOEXCEPT {
|
||||
// return the fixed size type if available to enable compile time optimizations
|
||||
return internal::traits<typename internal::remove_all<LhsNested>::type>::ColsAtCompileTime==Dynamic ? m_rhs.cols() : m_lhs.cols();
|
||||
}
|
||||
|
||||
/** \returns the left hand side nested expression */
|
||||
EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
|
||||
const _LhsNested& lhs() const { return m_lhs; }
|
||||
/** \returns the right hand side nested expression */
|
||||
EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
|
||||
const _RhsNested& rhs() const { return m_rhs; }
|
||||
/** \returns the functor representing the binary operation */
|
||||
EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
|
||||
const BinaryOp& functor() const { return m_functor; }
|
||||
|
||||
protected:
|
||||
LhsNested m_lhs;
|
||||
RhsNested m_rhs;
|
||||
const BinaryOp m_functor;
|
||||
};
|
||||
|
||||
// Generic API dispatcher
|
||||
template<typename BinaryOp, typename Lhs, typename Rhs, typename StorageKind>
|
||||
class CwiseBinaryOpImpl
|
||||
: public internal::generic_xpr_base<CwiseBinaryOp<BinaryOp, Lhs, Rhs> >::type
|
||||
{
|
||||
public:
|
||||
typedef typename internal::generic_xpr_base<CwiseBinaryOp<BinaryOp, Lhs, Rhs> >::type Base;
|
||||
};
|
||||
|
||||
/** replaces \c *this by \c *this - \a other.
|
||||
*
|
||||
* \returns a reference to \c *this
|
||||
*/
|
||||
template<typename Derived>
|
||||
template<typename OtherDerived>
|
||||
EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE Derived &
|
||||
MatrixBase<Derived>::operator-=(const MatrixBase<OtherDerived> &other)
|
||||
{
|
||||
call_assignment(derived(), other.derived(), internal::sub_assign_op<Scalar,typename OtherDerived::Scalar>());
|
||||
return derived();
|
||||
}
|
||||
|
||||
/** replaces \c *this by \c *this + \a other.
|
||||
*
|
||||
* \returns a reference to \c *this
|
||||
*/
|
||||
template<typename Derived>
|
||||
template<typename OtherDerived>
|
||||
EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE Derived &
|
||||
MatrixBase<Derived>::operator+=(const MatrixBase<OtherDerived>& other)
|
||||
{
|
||||
call_assignment(derived(), other.derived(), internal::add_assign_op<Scalar,typename OtherDerived::Scalar>());
|
||||
return derived();
|
||||
}
|
||||
|
||||
} // end namespace Eigen
|
||||
|
||||
#endif // EIGEN_CWISE_BINARY_OP_H
|
||||
File diff suppressed because it is too large
Load Diff
@@ -0,0 +1,197 @@
|
||||
// This file is part of Eigen, a lightweight C++ template library
|
||||
// for linear algebra.
|
||||
//
|
||||
// Copyright (C) 2008-2014 Gael Guennebaud <gael.guennebaud@inria.fr>
|
||||
// Copyright (C) 2006-2008 Benoit Jacob <jacob.benoit.1@gmail.com>
|
||||
// Copyright (C) 2016 Eugene Brevdo <ebrevdo@gmail.com>
|
||||
//
|
||||
// This Source Code Form is subject to the terms of the Mozilla
|
||||
// Public License v. 2.0. If a copy of the MPL was not distributed
|
||||
// with this file, You can obtain one at http://mozilla.org/MPL/2.0/.
|
||||
|
||||
#ifndef EIGEN_CWISE_TERNARY_OP_H
|
||||
#define EIGEN_CWISE_TERNARY_OP_H
|
||||
|
||||
namespace Eigen {
|
||||
|
||||
namespace internal {
|
||||
template <typename TernaryOp, typename Arg1, typename Arg2, typename Arg3>
|
||||
struct traits<CwiseTernaryOp<TernaryOp, Arg1, Arg2, Arg3> > {
|
||||
// we must not inherit from traits<Arg1> since it has
|
||||
// the potential to cause problems with MSVC
|
||||
typedef typename remove_all<Arg1>::type Ancestor;
|
||||
typedef typename traits<Ancestor>::XprKind XprKind;
|
||||
enum {
|
||||
RowsAtCompileTime = traits<Ancestor>::RowsAtCompileTime,
|
||||
ColsAtCompileTime = traits<Ancestor>::ColsAtCompileTime,
|
||||
MaxRowsAtCompileTime = traits<Ancestor>::MaxRowsAtCompileTime,
|
||||
MaxColsAtCompileTime = traits<Ancestor>::MaxColsAtCompileTime
|
||||
};
|
||||
|
||||
// even though we require Arg1, Arg2, and Arg3 to have the same scalar type
|
||||
// (see CwiseTernaryOp constructor),
|
||||
// we still want to handle the case when the result type is different.
|
||||
typedef typename result_of<TernaryOp(
|
||||
const typename Arg1::Scalar&, const typename Arg2::Scalar&,
|
||||
const typename Arg3::Scalar&)>::type Scalar;
|
||||
|
||||
typedef typename internal::traits<Arg1>::StorageKind StorageKind;
|
||||
typedef typename internal::traits<Arg1>::StorageIndex StorageIndex;
|
||||
|
||||
typedef typename Arg1::Nested Arg1Nested;
|
||||
typedef typename Arg2::Nested Arg2Nested;
|
||||
typedef typename Arg3::Nested Arg3Nested;
|
||||
typedef typename remove_reference<Arg1Nested>::type _Arg1Nested;
|
||||
typedef typename remove_reference<Arg2Nested>::type _Arg2Nested;
|
||||
typedef typename remove_reference<Arg3Nested>::type _Arg3Nested;
|
||||
enum { Flags = _Arg1Nested::Flags & RowMajorBit };
|
||||
};
|
||||
} // end namespace internal
|
||||
|
||||
template <typename TernaryOp, typename Arg1, typename Arg2, typename Arg3,
|
||||
typename StorageKind>
|
||||
class CwiseTernaryOpImpl;
|
||||
|
||||
/** \class CwiseTernaryOp
|
||||
* \ingroup Core_Module
|
||||
*
|
||||
* \brief Generic expression where a coefficient-wise ternary operator is
|
||||
* applied to two expressions
|
||||
*
|
||||
* \tparam TernaryOp template functor implementing the operator
|
||||
* \tparam Arg1Type the type of the first argument
|
||||
* \tparam Arg2Type the type of the second argument
|
||||
* \tparam Arg3Type the type of the third argument
|
||||
*
|
||||
* This class represents an expression where a coefficient-wise ternary
|
||||
* operator is applied to three expressions.
|
||||
* It is the return type of ternary operators, by which we mean only those
|
||||
* ternary operators where
|
||||
* all three arguments are Eigen expressions.
|
||||
* For example, the return type of betainc(matrix1, matrix2, matrix3) is a
|
||||
* CwiseTernaryOp.
|
||||
*
|
||||
* Most of the time, this is the only way that it is used, so you typically
|
||||
* don't have to name
|
||||
* CwiseTernaryOp types explicitly.
|
||||
*
|
||||
* \sa MatrixBase::ternaryExpr(const MatrixBase<Argument2> &, const
|
||||
* MatrixBase<Argument3> &, const CustomTernaryOp &) const, class CwiseBinaryOp,
|
||||
* class CwiseUnaryOp, class CwiseNullaryOp
|
||||
*/
|
||||
template <typename TernaryOp, typename Arg1Type, typename Arg2Type,
|
||||
typename Arg3Type>
|
||||
class CwiseTernaryOp : public CwiseTernaryOpImpl<
|
||||
TernaryOp, Arg1Type, Arg2Type, Arg3Type,
|
||||
typename internal::traits<Arg1Type>::StorageKind>,
|
||||
internal::no_assignment_operator
|
||||
{
|
||||
public:
|
||||
typedef typename internal::remove_all<Arg1Type>::type Arg1;
|
||||
typedef typename internal::remove_all<Arg2Type>::type Arg2;
|
||||
typedef typename internal::remove_all<Arg3Type>::type Arg3;
|
||||
|
||||
typedef typename CwiseTernaryOpImpl<
|
||||
TernaryOp, Arg1Type, Arg2Type, Arg3Type,
|
||||
typename internal::traits<Arg1Type>::StorageKind>::Base Base;
|
||||
EIGEN_GENERIC_PUBLIC_INTERFACE(CwiseTernaryOp)
|
||||
|
||||
typedef typename internal::ref_selector<Arg1Type>::type Arg1Nested;
|
||||
typedef typename internal::ref_selector<Arg2Type>::type Arg2Nested;
|
||||
typedef typename internal::ref_selector<Arg3Type>::type Arg3Nested;
|
||||
typedef typename internal::remove_reference<Arg1Nested>::type _Arg1Nested;
|
||||
typedef typename internal::remove_reference<Arg2Nested>::type _Arg2Nested;
|
||||
typedef typename internal::remove_reference<Arg3Nested>::type _Arg3Nested;
|
||||
|
||||
EIGEN_DEVICE_FUNC
|
||||
EIGEN_STRONG_INLINE CwiseTernaryOp(const Arg1& a1, const Arg2& a2,
|
||||
const Arg3& a3,
|
||||
const TernaryOp& func = TernaryOp())
|
||||
: m_arg1(a1), m_arg2(a2), m_arg3(a3), m_functor(func) {
|
||||
// require the sizes to match
|
||||
EIGEN_STATIC_ASSERT_SAME_MATRIX_SIZE(Arg1, Arg2)
|
||||
EIGEN_STATIC_ASSERT_SAME_MATRIX_SIZE(Arg1, Arg3)
|
||||
|
||||
// The index types should match
|
||||
EIGEN_STATIC_ASSERT((internal::is_same<
|
||||
typename internal::traits<Arg1Type>::StorageKind,
|
||||
typename internal::traits<Arg2Type>::StorageKind>::value),
|
||||
STORAGE_KIND_MUST_MATCH)
|
||||
EIGEN_STATIC_ASSERT((internal::is_same<
|
||||
typename internal::traits<Arg1Type>::StorageKind,
|
||||
typename internal::traits<Arg3Type>::StorageKind>::value),
|
||||
STORAGE_KIND_MUST_MATCH)
|
||||
|
||||
eigen_assert(a1.rows() == a2.rows() && a1.cols() == a2.cols() &&
|
||||
a1.rows() == a3.rows() && a1.cols() == a3.cols());
|
||||
}
|
||||
|
||||
EIGEN_DEVICE_FUNC
|
||||
EIGEN_STRONG_INLINE Index rows() const {
|
||||
// return the fixed size type if available to enable compile time
|
||||
// optimizations
|
||||
if (internal::traits<typename internal::remove_all<Arg1Nested>::type>::
|
||||
RowsAtCompileTime == Dynamic &&
|
||||
internal::traits<typename internal::remove_all<Arg2Nested>::type>::
|
||||
RowsAtCompileTime == Dynamic)
|
||||
return m_arg3.rows();
|
||||
else if (internal::traits<typename internal::remove_all<Arg1Nested>::type>::
|
||||
RowsAtCompileTime == Dynamic &&
|
||||
internal::traits<typename internal::remove_all<Arg3Nested>::type>::
|
||||
RowsAtCompileTime == Dynamic)
|
||||
return m_arg2.rows();
|
||||
else
|
||||
return m_arg1.rows();
|
||||
}
|
||||
EIGEN_DEVICE_FUNC
|
||||
EIGEN_STRONG_INLINE Index cols() const {
|
||||
// return the fixed size type if available to enable compile time
|
||||
// optimizations
|
||||
if (internal::traits<typename internal::remove_all<Arg1Nested>::type>::
|
||||
ColsAtCompileTime == Dynamic &&
|
||||
internal::traits<typename internal::remove_all<Arg2Nested>::type>::
|
||||
ColsAtCompileTime == Dynamic)
|
||||
return m_arg3.cols();
|
||||
else if (internal::traits<typename internal::remove_all<Arg1Nested>::type>::
|
||||
ColsAtCompileTime == Dynamic &&
|
||||
internal::traits<typename internal::remove_all<Arg3Nested>::type>::
|
||||
ColsAtCompileTime == Dynamic)
|
||||
return m_arg2.cols();
|
||||
else
|
||||
return m_arg1.cols();
|
||||
}
|
||||
|
||||
/** \returns the first argument nested expression */
|
||||
EIGEN_DEVICE_FUNC
|
||||
const _Arg1Nested& arg1() const { return m_arg1; }
|
||||
/** \returns the first argument nested expression */
|
||||
EIGEN_DEVICE_FUNC
|
||||
const _Arg2Nested& arg2() const { return m_arg2; }
|
||||
/** \returns the third argument nested expression */
|
||||
EIGEN_DEVICE_FUNC
|
||||
const _Arg3Nested& arg3() const { return m_arg3; }
|
||||
/** \returns the functor representing the ternary operation */
|
||||
EIGEN_DEVICE_FUNC
|
||||
const TernaryOp& functor() const { return m_functor; }
|
||||
|
||||
protected:
|
||||
Arg1Nested m_arg1;
|
||||
Arg2Nested m_arg2;
|
||||
Arg3Nested m_arg3;
|
||||
const TernaryOp m_functor;
|
||||
};
|
||||
|
||||
// Generic API dispatcher
|
||||
template <typename TernaryOp, typename Arg1, typename Arg2, typename Arg3,
|
||||
typename StorageKind>
|
||||
class CwiseTernaryOpImpl
|
||||
: public internal::generic_xpr_base<
|
||||
CwiseTernaryOp<TernaryOp, Arg1, Arg2, Arg3> >::type {
|
||||
public:
|
||||
typedef typename internal::generic_xpr_base<
|
||||
CwiseTernaryOp<TernaryOp, Arg1, Arg2, Arg3> >::type Base;
|
||||
};
|
||||
|
||||
} // end namespace Eigen
|
||||
|
||||
#endif // EIGEN_CWISE_TERNARY_OP_H
|
||||
@@ -0,0 +1,103 @@
|
||||
// This file is part of Eigen, a lightweight C++ template library
|
||||
// for linear algebra.
|
||||
//
|
||||
// Copyright (C) 2008-2014 Gael Guennebaud <gael.guennebaud@inria.fr>
|
||||
// Copyright (C) 2006-2008 Benoit Jacob <jacob.benoit.1@gmail.com>
|
||||
//
|
||||
// This Source Code Form is subject to the terms of the Mozilla
|
||||
// Public License v. 2.0. If a copy of the MPL was not distributed
|
||||
// with this file, You can obtain one at http://mozilla.org/MPL/2.0/.
|
||||
|
||||
#ifndef EIGEN_CWISE_UNARY_OP_H
|
||||
#define EIGEN_CWISE_UNARY_OP_H
|
||||
|
||||
namespace Eigen {
|
||||
|
||||
namespace internal {
|
||||
template<typename UnaryOp, typename XprType>
|
||||
struct traits<CwiseUnaryOp<UnaryOp, XprType> >
|
||||
: traits<XprType>
|
||||
{
|
||||
typedef typename result_of<
|
||||
UnaryOp(const typename XprType::Scalar&)
|
||||
>::type Scalar;
|
||||
typedef typename XprType::Nested XprTypeNested;
|
||||
typedef typename remove_reference<XprTypeNested>::type _XprTypeNested;
|
||||
enum {
|
||||
Flags = _XprTypeNested::Flags & RowMajorBit
|
||||
};
|
||||
};
|
||||
}
|
||||
|
||||
template<typename UnaryOp, typename XprType, typename StorageKind>
|
||||
class CwiseUnaryOpImpl;
|
||||
|
||||
/** \class CwiseUnaryOp
|
||||
* \ingroup Core_Module
|
||||
*
|
||||
* \brief Generic expression where a coefficient-wise unary operator is applied to an expression
|
||||
*
|
||||
* \tparam UnaryOp template functor implementing the operator
|
||||
* \tparam XprType the type of the expression to which we are applying the unary operator
|
||||
*
|
||||
* This class represents an expression where a unary operator is applied to an expression.
|
||||
* It is the return type of all operations taking exactly 1 input expression, regardless of the
|
||||
* presence of other inputs such as scalars. For example, the operator* in the expression 3*matrix
|
||||
* is considered unary, because only the right-hand side is an expression, and its
|
||||
* return type is a specialization of CwiseUnaryOp.
|
||||
*
|
||||
* Most of the time, this is the only way that it is used, so you typically don't have to name
|
||||
* CwiseUnaryOp types explicitly.
|
||||
*
|
||||
* \sa MatrixBase::unaryExpr(const CustomUnaryOp &) const, class CwiseBinaryOp, class CwiseNullaryOp
|
||||
*/
|
||||
template<typename UnaryOp, typename XprType>
|
||||
class CwiseUnaryOp : public CwiseUnaryOpImpl<UnaryOp, XprType, typename internal::traits<XprType>::StorageKind>, internal::no_assignment_operator
|
||||
{
|
||||
public:
|
||||
|
||||
typedef typename CwiseUnaryOpImpl<UnaryOp, XprType,typename internal::traits<XprType>::StorageKind>::Base Base;
|
||||
EIGEN_GENERIC_PUBLIC_INTERFACE(CwiseUnaryOp)
|
||||
typedef typename internal::ref_selector<XprType>::type XprTypeNested;
|
||||
typedef typename internal::remove_all<XprType>::type NestedExpression;
|
||||
|
||||
EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
|
||||
explicit CwiseUnaryOp(const XprType& xpr, const UnaryOp& func = UnaryOp())
|
||||
: m_xpr(xpr), m_functor(func) {}
|
||||
|
||||
EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE EIGEN_CONSTEXPR
|
||||
Index rows() const EIGEN_NOEXCEPT { return m_xpr.rows(); }
|
||||
EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE EIGEN_CONSTEXPR
|
||||
Index cols() const EIGEN_NOEXCEPT { return m_xpr.cols(); }
|
||||
|
||||
/** \returns the functor representing the unary operation */
|
||||
EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
|
||||
const UnaryOp& functor() const { return m_functor; }
|
||||
|
||||
/** \returns the nested expression */
|
||||
EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
|
||||
const typename internal::remove_all<XprTypeNested>::type&
|
||||
nestedExpression() const { return m_xpr; }
|
||||
|
||||
/** \returns the nested expression */
|
||||
EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
|
||||
typename internal::remove_all<XprTypeNested>::type&
|
||||
nestedExpression() { return m_xpr; }
|
||||
|
||||
protected:
|
||||
XprTypeNested m_xpr;
|
||||
const UnaryOp m_functor;
|
||||
};
|
||||
|
||||
// Generic API dispatcher
|
||||
template<typename UnaryOp, typename XprType, typename StorageKind>
|
||||
class CwiseUnaryOpImpl
|
||||
: public internal::generic_xpr_base<CwiseUnaryOp<UnaryOp, XprType> >::type
|
||||
{
|
||||
public:
|
||||
typedef typename internal::generic_xpr_base<CwiseUnaryOp<UnaryOp, XprType> >::type Base;
|
||||
};
|
||||
|
||||
} // end namespace Eigen
|
||||
|
||||
#endif // EIGEN_CWISE_UNARY_OP_H
|
||||
@@ -0,0 +1,132 @@
|
||||
// This file is part of Eigen, a lightweight C++ template library
|
||||
// for linear algebra.
|
||||
//
|
||||
// Copyright (C) 2009-2010 Gael Guennebaud <gael.guennebaud@inria.fr>
|
||||
//
|
||||
// This Source Code Form is subject to the terms of the Mozilla
|
||||
// Public License v. 2.0. If a copy of the MPL was not distributed
|
||||
// with this file, You can obtain one at http://mozilla.org/MPL/2.0/.
|
||||
|
||||
#ifndef EIGEN_CWISE_UNARY_VIEW_H
|
||||
#define EIGEN_CWISE_UNARY_VIEW_H
|
||||
|
||||
namespace Eigen {
|
||||
|
||||
namespace internal {
|
||||
template<typename ViewOp, typename MatrixType>
|
||||
struct traits<CwiseUnaryView<ViewOp, MatrixType> >
|
||||
: traits<MatrixType>
|
||||
{
|
||||
typedef typename result_of<
|
||||
ViewOp(const typename traits<MatrixType>::Scalar&)
|
||||
>::type Scalar;
|
||||
typedef typename MatrixType::Nested MatrixTypeNested;
|
||||
typedef typename remove_all<MatrixTypeNested>::type _MatrixTypeNested;
|
||||
enum {
|
||||
FlagsLvalueBit = is_lvalue<MatrixType>::value ? LvalueBit : 0,
|
||||
Flags = traits<_MatrixTypeNested>::Flags & (RowMajorBit | FlagsLvalueBit | DirectAccessBit), // FIXME DirectAccessBit should not be handled by expressions
|
||||
MatrixTypeInnerStride = inner_stride_at_compile_time<MatrixType>::ret,
|
||||
// need to cast the sizeof's from size_t to int explicitly, otherwise:
|
||||
// "error: no integral type can represent all of the enumerator values
|
||||
InnerStrideAtCompileTime = MatrixTypeInnerStride == Dynamic
|
||||
? int(Dynamic)
|
||||
: int(MatrixTypeInnerStride) * int(sizeof(typename traits<MatrixType>::Scalar) / sizeof(Scalar)),
|
||||
OuterStrideAtCompileTime = outer_stride_at_compile_time<MatrixType>::ret == Dynamic
|
||||
? int(Dynamic)
|
||||
: outer_stride_at_compile_time<MatrixType>::ret * int(sizeof(typename traits<MatrixType>::Scalar) / sizeof(Scalar))
|
||||
};
|
||||
};
|
||||
}
|
||||
|
||||
template<typename ViewOp, typename MatrixType, typename StorageKind>
|
||||
class CwiseUnaryViewImpl;
|
||||
|
||||
/** \class CwiseUnaryView
|
||||
* \ingroup Core_Module
|
||||
*
|
||||
* \brief Generic lvalue expression of a coefficient-wise unary operator of a matrix or a vector
|
||||
*
|
||||
* \tparam ViewOp template functor implementing the view
|
||||
* \tparam MatrixType the type of the matrix we are applying the unary operator
|
||||
*
|
||||
* This class represents a lvalue expression of a generic unary view operator of a matrix or a vector.
|
||||
* It is the return type of real() and imag(), and most of the time this is the only way it is used.
|
||||
*
|
||||
* \sa MatrixBase::unaryViewExpr(const CustomUnaryOp &) const, class CwiseUnaryOp
|
||||
*/
|
||||
template<typename ViewOp, typename MatrixType>
|
||||
class CwiseUnaryView : public CwiseUnaryViewImpl<ViewOp, MatrixType, typename internal::traits<MatrixType>::StorageKind>
|
||||
{
|
||||
public:
|
||||
|
||||
typedef typename CwiseUnaryViewImpl<ViewOp, MatrixType,typename internal::traits<MatrixType>::StorageKind>::Base Base;
|
||||
EIGEN_GENERIC_PUBLIC_INTERFACE(CwiseUnaryView)
|
||||
typedef typename internal::ref_selector<MatrixType>::non_const_type MatrixTypeNested;
|
||||
typedef typename internal::remove_all<MatrixType>::type NestedExpression;
|
||||
|
||||
explicit EIGEN_DEVICE_FUNC inline CwiseUnaryView(MatrixType& mat, const ViewOp& func = ViewOp())
|
||||
: m_matrix(mat), m_functor(func) {}
|
||||
|
||||
EIGEN_INHERIT_ASSIGNMENT_OPERATORS(CwiseUnaryView)
|
||||
|
||||
EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE EIGEN_CONSTEXPR
|
||||
Index rows() const EIGEN_NOEXCEPT { return m_matrix.rows(); }
|
||||
EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE EIGEN_CONSTEXPR
|
||||
Index cols() const EIGEN_NOEXCEPT { return m_matrix.cols(); }
|
||||
|
||||
/** \returns the functor representing unary operation */
|
||||
EIGEN_DEVICE_FUNC const ViewOp& functor() const { return m_functor; }
|
||||
|
||||
/** \returns the nested expression */
|
||||
EIGEN_DEVICE_FUNC const typename internal::remove_all<MatrixTypeNested>::type&
|
||||
nestedExpression() const { return m_matrix; }
|
||||
|
||||
/** \returns the nested expression */
|
||||
EIGEN_DEVICE_FUNC typename internal::remove_reference<MatrixTypeNested>::type&
|
||||
nestedExpression() { return m_matrix; }
|
||||
|
||||
protected:
|
||||
MatrixTypeNested m_matrix;
|
||||
ViewOp m_functor;
|
||||
};
|
||||
|
||||
// Generic API dispatcher
|
||||
template<typename ViewOp, typename XprType, typename StorageKind>
|
||||
class CwiseUnaryViewImpl
|
||||
: public internal::generic_xpr_base<CwiseUnaryView<ViewOp, XprType> >::type
|
||||
{
|
||||
public:
|
||||
typedef typename internal::generic_xpr_base<CwiseUnaryView<ViewOp, XprType> >::type Base;
|
||||
};
|
||||
|
||||
template<typename ViewOp, typename MatrixType>
|
||||
class CwiseUnaryViewImpl<ViewOp,MatrixType,Dense>
|
||||
: public internal::dense_xpr_base< CwiseUnaryView<ViewOp, MatrixType> >::type
|
||||
{
|
||||
public:
|
||||
|
||||
typedef CwiseUnaryView<ViewOp, MatrixType> Derived;
|
||||
typedef typename internal::dense_xpr_base< CwiseUnaryView<ViewOp, MatrixType> >::type Base;
|
||||
|
||||
EIGEN_DENSE_PUBLIC_INTERFACE(Derived)
|
||||
EIGEN_INHERIT_ASSIGNMENT_OPERATORS(CwiseUnaryViewImpl)
|
||||
|
||||
EIGEN_DEVICE_FUNC inline Scalar* data() { return &(this->coeffRef(0)); }
|
||||
EIGEN_DEVICE_FUNC inline const Scalar* data() const { return &(this->coeff(0)); }
|
||||
|
||||
EIGEN_DEVICE_FUNC EIGEN_CONSTEXPR inline Index innerStride() const
|
||||
{
|
||||
return derived().nestedExpression().innerStride() * sizeof(typename internal::traits<MatrixType>::Scalar) / sizeof(Scalar);
|
||||
}
|
||||
|
||||
EIGEN_DEVICE_FUNC EIGEN_CONSTEXPR inline Index outerStride() const
|
||||
{
|
||||
return derived().nestedExpression().outerStride() * sizeof(typename internal::traits<MatrixType>::Scalar) / sizeof(Scalar);
|
||||
}
|
||||
protected:
|
||||
EIGEN_DEFAULT_EMPTY_CONSTRUCTOR_AND_DESTRUCTOR(CwiseUnaryViewImpl)
|
||||
};
|
||||
|
||||
} // end namespace Eigen
|
||||
|
||||
#endif // EIGEN_CWISE_UNARY_VIEW_H
|
||||
@@ -0,0 +1,701 @@
|
||||
// This file is part of Eigen, a lightweight C++ template library
|
||||
// for linear algebra.
|
||||
//
|
||||
// Copyright (C) 2007-2010 Benoit Jacob <jacob.benoit.1@gmail.com>
|
||||
// Copyright (C) 2008-2010 Gael Guennebaud <gael.guennebaud@inria.fr>
|
||||
//
|
||||
// This Source Code Form is subject to the terms of the Mozilla
|
||||
// Public License v. 2.0. If a copy of the MPL was not distributed
|
||||
// with this file, You can obtain one at http://mozilla.org/MPL/2.0/.
|
||||
|
||||
#ifndef EIGEN_DENSEBASE_H
|
||||
#define EIGEN_DENSEBASE_H
|
||||
|
||||
namespace Eigen {
|
||||
|
||||
namespace internal {
|
||||
|
||||
// The index type defined by EIGEN_DEFAULT_DENSE_INDEX_TYPE must be a signed type.
|
||||
// This dummy function simply aims at checking that at compile time.
|
||||
static inline void check_DenseIndex_is_signed() {
|
||||
EIGEN_STATIC_ASSERT(NumTraits<DenseIndex>::IsSigned,THE_INDEX_TYPE_MUST_BE_A_SIGNED_TYPE)
|
||||
}
|
||||
|
||||
} // end namespace internal
|
||||
|
||||
/** \class DenseBase
|
||||
* \ingroup Core_Module
|
||||
*
|
||||
* \brief Base class for all dense matrices, vectors, and arrays
|
||||
*
|
||||
* This class is the base that is inherited by all dense objects (matrix, vector, arrays,
|
||||
* and related expression types). The common Eigen API for dense objects is contained in this class.
|
||||
*
|
||||
* \tparam Derived is the derived type, e.g., a matrix type or an expression.
|
||||
*
|
||||
* This class can be extended with the help of the plugin mechanism described on the page
|
||||
* \ref TopicCustomizing_Plugins by defining the preprocessor symbol \c EIGEN_DENSEBASE_PLUGIN.
|
||||
*
|
||||
* \sa \blank \ref TopicClassHierarchy
|
||||
*/
|
||||
template<typename Derived> class DenseBase
|
||||
#ifndef EIGEN_PARSED_BY_DOXYGEN
|
||||
: public DenseCoeffsBase<Derived, internal::accessors_level<Derived>::value>
|
||||
#else
|
||||
: public DenseCoeffsBase<Derived,DirectWriteAccessors>
|
||||
#endif // not EIGEN_PARSED_BY_DOXYGEN
|
||||
{
|
||||
public:
|
||||
|
||||
/** Inner iterator type to iterate over the coefficients of a row or column.
|
||||
* \sa class InnerIterator
|
||||
*/
|
||||
typedef Eigen::InnerIterator<Derived> InnerIterator;
|
||||
|
||||
typedef typename internal::traits<Derived>::StorageKind StorageKind;
|
||||
|
||||
/**
|
||||
* \brief The type used to store indices
|
||||
* \details This typedef is relevant for types that store multiple indices such as
|
||||
* PermutationMatrix or Transpositions, otherwise it defaults to Eigen::Index
|
||||
* \sa \blank \ref TopicPreprocessorDirectives, Eigen::Index, SparseMatrixBase.
|
||||
*/
|
||||
typedef typename internal::traits<Derived>::StorageIndex StorageIndex;
|
||||
|
||||
/** The numeric type of the expression' coefficients, e.g. float, double, int or std::complex<float>, etc. */
|
||||
typedef typename internal::traits<Derived>::Scalar Scalar;
|
||||
|
||||
/** The numeric type of the expression' coefficients, e.g. float, double, int or std::complex<float>, etc.
|
||||
*
|
||||
* It is an alias for the Scalar type */
|
||||
typedef Scalar value_type;
|
||||
|
||||
typedef typename NumTraits<Scalar>::Real RealScalar;
|
||||
typedef DenseCoeffsBase<Derived, internal::accessors_level<Derived>::value> Base;
|
||||
|
||||
using Base::derived;
|
||||
using Base::const_cast_derived;
|
||||
using Base::rows;
|
||||
using Base::cols;
|
||||
using Base::size;
|
||||
using Base::rowIndexByOuterInner;
|
||||
using Base::colIndexByOuterInner;
|
||||
using Base::coeff;
|
||||
using Base::coeffByOuterInner;
|
||||
using Base::operator();
|
||||
using Base::operator[];
|
||||
using Base::x;
|
||||
using Base::y;
|
||||
using Base::z;
|
||||
using Base::w;
|
||||
using Base::stride;
|
||||
using Base::innerStride;
|
||||
using Base::outerStride;
|
||||
using Base::rowStride;
|
||||
using Base::colStride;
|
||||
typedef typename Base::CoeffReturnType CoeffReturnType;
|
||||
|
||||
enum {
|
||||
|
||||
RowsAtCompileTime = internal::traits<Derived>::RowsAtCompileTime,
|
||||
/**< The number of rows at compile-time. This is just a copy of the value provided
|
||||
* by the \a Derived type. If a value is not known at compile-time,
|
||||
* it is set to the \a Dynamic constant.
|
||||
* \sa MatrixBase::rows(), MatrixBase::cols(), ColsAtCompileTime, SizeAtCompileTime */
|
||||
|
||||
ColsAtCompileTime = internal::traits<Derived>::ColsAtCompileTime,
|
||||
/**< The number of columns at compile-time. This is just a copy of the value provided
|
||||
* by the \a Derived type. If a value is not known at compile-time,
|
||||
* it is set to the \a Dynamic constant.
|
||||
* \sa MatrixBase::rows(), MatrixBase::cols(), RowsAtCompileTime, SizeAtCompileTime */
|
||||
|
||||
|
||||
SizeAtCompileTime = (internal::size_at_compile_time<internal::traits<Derived>::RowsAtCompileTime,
|
||||
internal::traits<Derived>::ColsAtCompileTime>::ret),
|
||||
/**< This is equal to the number of coefficients, i.e. the number of
|
||||
* rows times the number of columns, or to \a Dynamic if this is not
|
||||
* known at compile-time. \sa RowsAtCompileTime, ColsAtCompileTime */
|
||||
|
||||
MaxRowsAtCompileTime = internal::traits<Derived>::MaxRowsAtCompileTime,
|
||||
/**< This value is equal to the maximum possible number of rows that this expression
|
||||
* might have. If this expression might have an arbitrarily high number of rows,
|
||||
* this value is set to \a Dynamic.
|
||||
*
|
||||
* This value is useful to know when evaluating an expression, in order to determine
|
||||
* whether it is possible to avoid doing a dynamic memory allocation.
|
||||
*
|
||||
* \sa RowsAtCompileTime, MaxColsAtCompileTime, MaxSizeAtCompileTime
|
||||
*/
|
||||
|
||||
MaxColsAtCompileTime = internal::traits<Derived>::MaxColsAtCompileTime,
|
||||
/**< This value is equal to the maximum possible number of columns that this expression
|
||||
* might have. If this expression might have an arbitrarily high number of columns,
|
||||
* this value is set to \a Dynamic.
|
||||
*
|
||||
* This value is useful to know when evaluating an expression, in order to determine
|
||||
* whether it is possible to avoid doing a dynamic memory allocation.
|
||||
*
|
||||
* \sa ColsAtCompileTime, MaxRowsAtCompileTime, MaxSizeAtCompileTime
|
||||
*/
|
||||
|
||||
MaxSizeAtCompileTime = (internal::size_at_compile_time<internal::traits<Derived>::MaxRowsAtCompileTime,
|
||||
internal::traits<Derived>::MaxColsAtCompileTime>::ret),
|
||||
/**< This value is equal to the maximum possible number of coefficients that this expression
|
||||
* might have. If this expression might have an arbitrarily high number of coefficients,
|
||||
* this value is set to \a Dynamic.
|
||||
*
|
||||
* This value is useful to know when evaluating an expression, in order to determine
|
||||
* whether it is possible to avoid doing a dynamic memory allocation.
|
||||
*
|
||||
* \sa SizeAtCompileTime, MaxRowsAtCompileTime, MaxColsAtCompileTime
|
||||
*/
|
||||
|
||||
IsVectorAtCompileTime = internal::traits<Derived>::RowsAtCompileTime == 1
|
||||
|| internal::traits<Derived>::ColsAtCompileTime == 1,
|
||||
/**< This is set to true if either the number of rows or the number of
|
||||
* columns is known at compile-time to be equal to 1. Indeed, in that case,
|
||||
* we are dealing with a column-vector (if there is only one column) or with
|
||||
* a row-vector (if there is only one row). */
|
||||
|
||||
NumDimensions = int(MaxSizeAtCompileTime) == 1 ? 0 : bool(IsVectorAtCompileTime) ? 1 : 2,
|
||||
/**< This value is equal to Tensor::NumDimensions, i.e. 0 for scalars, 1 for vectors,
|
||||
* and 2 for matrices.
|
||||
*/
|
||||
|
||||
Flags = internal::traits<Derived>::Flags,
|
||||
/**< This stores expression \ref flags flags which may or may not be inherited by new expressions
|
||||
* constructed from this one. See the \ref flags "list of flags".
|
||||
*/
|
||||
|
||||
IsRowMajor = int(Flags) & RowMajorBit, /**< True if this expression has row-major storage order. */
|
||||
|
||||
InnerSizeAtCompileTime = int(IsVectorAtCompileTime) ? int(SizeAtCompileTime)
|
||||
: int(IsRowMajor) ? int(ColsAtCompileTime) : int(RowsAtCompileTime),
|
||||
|
||||
InnerStrideAtCompileTime = internal::inner_stride_at_compile_time<Derived>::ret,
|
||||
OuterStrideAtCompileTime = internal::outer_stride_at_compile_time<Derived>::ret
|
||||
};
|
||||
|
||||
typedef typename internal::find_best_packet<Scalar,SizeAtCompileTime>::type PacketScalar;
|
||||
|
||||
enum { IsPlainObjectBase = 0 };
|
||||
|
||||
/** The plain matrix type corresponding to this expression.
|
||||
* \sa PlainObject */
|
||||
typedef Matrix<typename internal::traits<Derived>::Scalar,
|
||||
internal::traits<Derived>::RowsAtCompileTime,
|
||||
internal::traits<Derived>::ColsAtCompileTime,
|
||||
AutoAlign | (internal::traits<Derived>::Flags&RowMajorBit ? RowMajor : ColMajor),
|
||||
internal::traits<Derived>::MaxRowsAtCompileTime,
|
||||
internal::traits<Derived>::MaxColsAtCompileTime
|
||||
> PlainMatrix;
|
||||
|
||||
/** The plain array type corresponding to this expression.
|
||||
* \sa PlainObject */
|
||||
typedef Array<typename internal::traits<Derived>::Scalar,
|
||||
internal::traits<Derived>::RowsAtCompileTime,
|
||||
internal::traits<Derived>::ColsAtCompileTime,
|
||||
AutoAlign | (internal::traits<Derived>::Flags&RowMajorBit ? RowMajor : ColMajor),
|
||||
internal::traits<Derived>::MaxRowsAtCompileTime,
|
||||
internal::traits<Derived>::MaxColsAtCompileTime
|
||||
> PlainArray;
|
||||
|
||||
/** \brief The plain matrix or array type corresponding to this expression.
|
||||
*
|
||||
* This is not necessarily exactly the return type of eval(). In the case of plain matrices,
|
||||
* the return type of eval() is a const reference to a matrix, not a matrix! It is however guaranteed
|
||||
* that the return type of eval() is either PlainObject or const PlainObject&.
|
||||
*/
|
||||
typedef typename internal::conditional<internal::is_same<typename internal::traits<Derived>::XprKind,MatrixXpr >::value,
|
||||
PlainMatrix, PlainArray>::type PlainObject;
|
||||
|
||||
/** \returns the number of nonzero coefficients which is in practice the number
|
||||
* of stored coefficients. */
|
||||
EIGEN_DEVICE_FUNC EIGEN_CONSTEXPR
|
||||
inline Index nonZeros() const { return size(); }
|
||||
|
||||
/** \returns the outer size.
|
||||
*
|
||||
* \note For a vector, this returns just 1. For a matrix (non-vector), this is the major dimension
|
||||
* with respect to the \ref TopicStorageOrders "storage order", i.e., the number of columns for a
|
||||
* column-major matrix, and the number of rows for a row-major matrix. */
|
||||
EIGEN_DEVICE_FUNC EIGEN_CONSTEXPR
|
||||
Index outerSize() const
|
||||
{
|
||||
return IsVectorAtCompileTime ? 1
|
||||
: int(IsRowMajor) ? this->rows() : this->cols();
|
||||
}
|
||||
|
||||
/** \returns the inner size.
|
||||
*
|
||||
* \note For a vector, this is just the size. For a matrix (non-vector), this is the minor dimension
|
||||
* with respect to the \ref TopicStorageOrders "storage order", i.e., the number of rows for a
|
||||
* column-major matrix, and the number of columns for a row-major matrix. */
|
||||
EIGEN_DEVICE_FUNC EIGEN_CONSTEXPR
|
||||
Index innerSize() const
|
||||
{
|
||||
return IsVectorAtCompileTime ? this->size()
|
||||
: int(IsRowMajor) ? this->cols() : this->rows();
|
||||
}
|
||||
|
||||
/** Only plain matrices/arrays, not expressions, may be resized; therefore the only useful resize methods are
|
||||
* Matrix::resize() and Array::resize(). The present method only asserts that the new size equals the old size, and does
|
||||
* nothing else.
|
||||
*/
|
||||
EIGEN_DEVICE_FUNC
|
||||
void resize(Index newSize)
|
||||
{
|
||||
EIGEN_ONLY_USED_FOR_DEBUG(newSize);
|
||||
eigen_assert(newSize == this->size()
|
||||
&& "DenseBase::resize() does not actually allow to resize.");
|
||||
}
|
||||
/** Only plain matrices/arrays, not expressions, may be resized; therefore the only useful resize methods are
|
||||
* Matrix::resize() and Array::resize(). The present method only asserts that the new size equals the old size, and does
|
||||
* nothing else.
|
||||
*/
|
||||
EIGEN_DEVICE_FUNC
|
||||
void resize(Index rows, Index cols)
|
||||
{
|
||||
EIGEN_ONLY_USED_FOR_DEBUG(rows);
|
||||
EIGEN_ONLY_USED_FOR_DEBUG(cols);
|
||||
eigen_assert(rows == this->rows() && cols == this->cols()
|
||||
&& "DenseBase::resize() does not actually allow to resize.");
|
||||
}
|
||||
|
||||
#ifndef EIGEN_PARSED_BY_DOXYGEN
|
||||
/** \internal Represents a matrix with all coefficients equal to one another*/
|
||||
typedef CwiseNullaryOp<internal::scalar_constant_op<Scalar>,PlainObject> ConstantReturnType;
|
||||
/** \internal \deprecated Represents a vector with linearly spaced coefficients that allows sequential access only. */
|
||||
EIGEN_DEPRECATED typedef CwiseNullaryOp<internal::linspaced_op<Scalar>,PlainObject> SequentialLinSpacedReturnType;
|
||||
/** \internal Represents a vector with linearly spaced coefficients that allows random access. */
|
||||
typedef CwiseNullaryOp<internal::linspaced_op<Scalar>,PlainObject> RandomAccessLinSpacedReturnType;
|
||||
/** \internal the return type of MatrixBase::eigenvalues() */
|
||||
typedef Matrix<typename NumTraits<typename internal::traits<Derived>::Scalar>::Real, internal::traits<Derived>::ColsAtCompileTime, 1> EigenvaluesReturnType;
|
||||
|
||||
#endif // not EIGEN_PARSED_BY_DOXYGEN
|
||||
|
||||
/** Copies \a other into *this. \returns a reference to *this. */
|
||||
template<typename OtherDerived>
|
||||
EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
|
||||
Derived& operator=(const DenseBase<OtherDerived>& other);
|
||||
|
||||
/** Special case of the template operator=, in order to prevent the compiler
|
||||
* from generating a default operator= (issue hit with g++ 4.1)
|
||||
*/
|
||||
EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
|
||||
Derived& operator=(const DenseBase& other);
|
||||
|
||||
template<typename OtherDerived>
|
||||
EIGEN_DEVICE_FUNC
|
||||
Derived& operator=(const EigenBase<OtherDerived> &other);
|
||||
|
||||
template<typename OtherDerived>
|
||||
EIGEN_DEVICE_FUNC
|
||||
Derived& operator+=(const EigenBase<OtherDerived> &other);
|
||||
|
||||
template<typename OtherDerived>
|
||||
EIGEN_DEVICE_FUNC
|
||||
Derived& operator-=(const EigenBase<OtherDerived> &other);
|
||||
|
||||
template<typename OtherDerived>
|
||||
EIGEN_DEVICE_FUNC
|
||||
Derived& operator=(const ReturnByValue<OtherDerived>& func);
|
||||
|
||||
/** \internal
|
||||
* Copies \a other into *this without evaluating other. \returns a reference to *this. */
|
||||
template<typename OtherDerived>
|
||||
/** \deprecated */
|
||||
EIGEN_DEPRECATED EIGEN_DEVICE_FUNC
|
||||
Derived& lazyAssign(const DenseBase<OtherDerived>& other);
|
||||
|
||||
EIGEN_DEVICE_FUNC
|
||||
CommaInitializer<Derived> operator<< (const Scalar& s);
|
||||
|
||||
template<unsigned int Added,unsigned int Removed>
|
||||
/** \deprecated it now returns \c *this */
|
||||
EIGEN_DEPRECATED
|
||||
const Derived& flagged() const
|
||||
{ return derived(); }
|
||||
|
||||
template<typename OtherDerived>
|
||||
EIGEN_DEVICE_FUNC
|
||||
CommaInitializer<Derived> operator<< (const DenseBase<OtherDerived>& other);
|
||||
|
||||
typedef Transpose<Derived> TransposeReturnType;
|
||||
EIGEN_DEVICE_FUNC
|
||||
TransposeReturnType transpose();
|
||||
typedef typename internal::add_const<Transpose<const Derived> >::type ConstTransposeReturnType;
|
||||
EIGEN_DEVICE_FUNC
|
||||
ConstTransposeReturnType transpose() const;
|
||||
EIGEN_DEVICE_FUNC
|
||||
void transposeInPlace();
|
||||
|
||||
EIGEN_DEVICE_FUNC static const ConstantReturnType
|
||||
Constant(Index rows, Index cols, const Scalar& value);
|
||||
EIGEN_DEVICE_FUNC static const ConstantReturnType
|
||||
Constant(Index size, const Scalar& value);
|
||||
EIGEN_DEVICE_FUNC static const ConstantReturnType
|
||||
Constant(const Scalar& value);
|
||||
|
||||
EIGEN_DEPRECATED EIGEN_DEVICE_FUNC static const RandomAccessLinSpacedReturnType
|
||||
LinSpaced(Sequential_t, Index size, const Scalar& low, const Scalar& high);
|
||||
EIGEN_DEPRECATED EIGEN_DEVICE_FUNC static const RandomAccessLinSpacedReturnType
|
||||
LinSpaced(Sequential_t, const Scalar& low, const Scalar& high);
|
||||
|
||||
EIGEN_DEVICE_FUNC static const RandomAccessLinSpacedReturnType
|
||||
LinSpaced(Index size, const Scalar& low, const Scalar& high);
|
||||
EIGEN_DEVICE_FUNC static const RandomAccessLinSpacedReturnType
|
||||
LinSpaced(const Scalar& low, const Scalar& high);
|
||||
|
||||
template<typename CustomNullaryOp> EIGEN_DEVICE_FUNC
|
||||
static const CwiseNullaryOp<CustomNullaryOp, PlainObject>
|
||||
NullaryExpr(Index rows, Index cols, const CustomNullaryOp& func);
|
||||
template<typename CustomNullaryOp> EIGEN_DEVICE_FUNC
|
||||
static const CwiseNullaryOp<CustomNullaryOp, PlainObject>
|
||||
NullaryExpr(Index size, const CustomNullaryOp& func);
|
||||
template<typename CustomNullaryOp> EIGEN_DEVICE_FUNC
|
||||
static const CwiseNullaryOp<CustomNullaryOp, PlainObject>
|
||||
NullaryExpr(const CustomNullaryOp& func);
|
||||
|
||||
EIGEN_DEVICE_FUNC static const ConstantReturnType Zero(Index rows, Index cols);
|
||||
EIGEN_DEVICE_FUNC static const ConstantReturnType Zero(Index size);
|
||||
EIGEN_DEVICE_FUNC static const ConstantReturnType Zero();
|
||||
EIGEN_DEVICE_FUNC static const ConstantReturnType Ones(Index rows, Index cols);
|
||||
EIGEN_DEVICE_FUNC static const ConstantReturnType Ones(Index size);
|
||||
EIGEN_DEVICE_FUNC static const ConstantReturnType Ones();
|
||||
|
||||
EIGEN_DEVICE_FUNC void fill(const Scalar& value);
|
||||
EIGEN_DEVICE_FUNC Derived& setConstant(const Scalar& value);
|
||||
EIGEN_DEVICE_FUNC Derived& setLinSpaced(Index size, const Scalar& low, const Scalar& high);
|
||||
EIGEN_DEVICE_FUNC Derived& setLinSpaced(const Scalar& low, const Scalar& high);
|
||||
EIGEN_DEVICE_FUNC Derived& setZero();
|
||||
EIGEN_DEVICE_FUNC Derived& setOnes();
|
||||
EIGEN_DEVICE_FUNC Derived& setRandom();
|
||||
|
||||
template<typename OtherDerived> EIGEN_DEVICE_FUNC
|
||||
bool isApprox(const DenseBase<OtherDerived>& other,
|
||||
const RealScalar& prec = NumTraits<Scalar>::dummy_precision()) const;
|
||||
EIGEN_DEVICE_FUNC
|
||||
bool isMuchSmallerThan(const RealScalar& other,
|
||||
const RealScalar& prec = NumTraits<Scalar>::dummy_precision()) const;
|
||||
template<typename OtherDerived> EIGEN_DEVICE_FUNC
|
||||
bool isMuchSmallerThan(const DenseBase<OtherDerived>& other,
|
||||
const RealScalar& prec = NumTraits<Scalar>::dummy_precision()) const;
|
||||
|
||||
EIGEN_DEVICE_FUNC bool isApproxToConstant(const Scalar& value, const RealScalar& prec = NumTraits<Scalar>::dummy_precision()) const;
|
||||
EIGEN_DEVICE_FUNC bool isConstant(const Scalar& value, const RealScalar& prec = NumTraits<Scalar>::dummy_precision()) const;
|
||||
EIGEN_DEVICE_FUNC bool isZero(const RealScalar& prec = NumTraits<Scalar>::dummy_precision()) const;
|
||||
EIGEN_DEVICE_FUNC bool isOnes(const RealScalar& prec = NumTraits<Scalar>::dummy_precision()) const;
|
||||
|
||||
inline bool hasNaN() const;
|
||||
inline bool allFinite() const;
|
||||
|
||||
EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
|
||||
Derived& operator*=(const Scalar& other);
|
||||
EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
|
||||
Derived& operator/=(const Scalar& other);
|
||||
|
||||
typedef typename internal::add_const_on_value_type<typename internal::eval<Derived>::type>::type EvalReturnType;
|
||||
/** \returns the matrix or vector obtained by evaluating this expression.
|
||||
*
|
||||
* Notice that in the case of a plain matrix or vector (not an expression) this function just returns
|
||||
* a const reference, in order to avoid a useless copy.
|
||||
*
|
||||
* \warning Be careful with eval() and the auto C++ keyword, as detailed in this \link TopicPitfalls_auto_keyword page \endlink.
|
||||
*/
|
||||
EIGEN_DEVICE_FUNC
|
||||
EIGEN_STRONG_INLINE EvalReturnType eval() const
|
||||
{
|
||||
// Even though MSVC does not honor strong inlining when the return type
|
||||
// is a dynamic matrix, we desperately need strong inlining for fixed
|
||||
// size types on MSVC.
|
||||
return typename internal::eval<Derived>::type(derived());
|
||||
}
|
||||
|
||||
/** swaps *this with the expression \a other.
|
||||
*
|
||||
*/
|
||||
template<typename OtherDerived>
|
||||
EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
|
||||
void swap(const DenseBase<OtherDerived>& other)
|
||||
{
|
||||
EIGEN_STATIC_ASSERT(!OtherDerived::IsPlainObjectBase,THIS_EXPRESSION_IS_NOT_A_LVALUE__IT_IS_READ_ONLY);
|
||||
eigen_assert(rows()==other.rows() && cols()==other.cols());
|
||||
call_assignment(derived(), other.const_cast_derived(), internal::swap_assign_op<Scalar>());
|
||||
}
|
||||
|
||||
/** swaps *this with the matrix or array \a other.
|
||||
*
|
||||
*/
|
||||
template<typename OtherDerived>
|
||||
EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
|
||||
void swap(PlainObjectBase<OtherDerived>& other)
|
||||
{
|
||||
eigen_assert(rows()==other.rows() && cols()==other.cols());
|
||||
call_assignment(derived(), other.derived(), internal::swap_assign_op<Scalar>());
|
||||
}
|
||||
|
||||
EIGEN_DEVICE_FUNC inline const NestByValue<Derived> nestByValue() const;
|
||||
EIGEN_DEVICE_FUNC inline const ForceAlignedAccess<Derived> forceAlignedAccess() const;
|
||||
EIGEN_DEVICE_FUNC inline ForceAlignedAccess<Derived> forceAlignedAccess();
|
||||
template<bool Enable> EIGEN_DEVICE_FUNC
|
||||
inline const typename internal::conditional<Enable,ForceAlignedAccess<Derived>,Derived&>::type forceAlignedAccessIf() const;
|
||||
template<bool Enable> EIGEN_DEVICE_FUNC
|
||||
inline typename internal::conditional<Enable,ForceAlignedAccess<Derived>,Derived&>::type forceAlignedAccessIf();
|
||||
|
||||
EIGEN_DEVICE_FUNC Scalar sum() const;
|
||||
EIGEN_DEVICE_FUNC Scalar mean() const;
|
||||
EIGEN_DEVICE_FUNC Scalar trace() const;
|
||||
|
||||
EIGEN_DEVICE_FUNC Scalar prod() const;
|
||||
|
||||
template<int NaNPropagation>
|
||||
EIGEN_DEVICE_FUNC typename internal::traits<Derived>::Scalar minCoeff() const;
|
||||
template<int NaNPropagation>
|
||||
EIGEN_DEVICE_FUNC typename internal::traits<Derived>::Scalar maxCoeff() const;
|
||||
|
||||
|
||||
// By default, the fastest version with undefined NaN propagation semantics is
|
||||
// used.
|
||||
// TODO(rmlarsen): Replace with default template argument when we move to
|
||||
// c++11 or beyond.
|
||||
EIGEN_DEVICE_FUNC inline typename internal::traits<Derived>::Scalar minCoeff() const {
|
||||
return minCoeff<PropagateFast>();
|
||||
}
|
||||
EIGEN_DEVICE_FUNC inline typename internal::traits<Derived>::Scalar maxCoeff() const {
|
||||
return maxCoeff<PropagateFast>();
|
||||
}
|
||||
|
||||
template<int NaNPropagation, typename IndexType>
|
||||
EIGEN_DEVICE_FUNC
|
||||
typename internal::traits<Derived>::Scalar minCoeff(IndexType* row, IndexType* col) const;
|
||||
template<int NaNPropagation, typename IndexType>
|
||||
EIGEN_DEVICE_FUNC
|
||||
typename internal::traits<Derived>::Scalar maxCoeff(IndexType* row, IndexType* col) const;
|
||||
template<int NaNPropagation, typename IndexType>
|
||||
EIGEN_DEVICE_FUNC
|
||||
typename internal::traits<Derived>::Scalar minCoeff(IndexType* index) const;
|
||||
template<int NaNPropagation, typename IndexType>
|
||||
EIGEN_DEVICE_FUNC
|
||||
typename internal::traits<Derived>::Scalar maxCoeff(IndexType* index) const;
|
||||
|
||||
// TODO(rmlarsen): Replace these methods with a default template argument.
|
||||
template<typename IndexType>
|
||||
EIGEN_DEVICE_FUNC inline
|
||||
typename internal::traits<Derived>::Scalar minCoeff(IndexType* row, IndexType* col) const {
|
||||
return minCoeff<PropagateFast>(row, col);
|
||||
}
|
||||
template<typename IndexType>
|
||||
EIGEN_DEVICE_FUNC inline
|
||||
typename internal::traits<Derived>::Scalar maxCoeff(IndexType* row, IndexType* col) const {
|
||||
return maxCoeff<PropagateFast>(row, col);
|
||||
}
|
||||
template<typename IndexType>
|
||||
EIGEN_DEVICE_FUNC inline
|
||||
typename internal::traits<Derived>::Scalar minCoeff(IndexType* index) const {
|
||||
return minCoeff<PropagateFast>(index);
|
||||
}
|
||||
template<typename IndexType>
|
||||
EIGEN_DEVICE_FUNC inline
|
||||
typename internal::traits<Derived>::Scalar maxCoeff(IndexType* index) const {
|
||||
return maxCoeff<PropagateFast>(index);
|
||||
}
|
||||
|
||||
template<typename BinaryOp>
|
||||
EIGEN_DEVICE_FUNC
|
||||
Scalar redux(const BinaryOp& func) const;
|
||||
|
||||
template<typename Visitor>
|
||||
EIGEN_DEVICE_FUNC
|
||||
void visit(Visitor& func) const;
|
||||
|
||||
/** \returns a WithFormat proxy object allowing to print a matrix the with given
|
||||
* format \a fmt.
|
||||
*
|
||||
* See class IOFormat for some examples.
|
||||
*
|
||||
* \sa class IOFormat, class WithFormat
|
||||
*/
|
||||
inline const WithFormat<Derived> format(const IOFormat& fmt) const
|
||||
{
|
||||
return WithFormat<Derived>(derived(), fmt);
|
||||
}
|
||||
|
||||
/** \returns the unique coefficient of a 1x1 expression */
|
||||
EIGEN_DEVICE_FUNC
|
||||
CoeffReturnType value() const
|
||||
{
|
||||
EIGEN_STATIC_ASSERT_SIZE_1x1(Derived)
|
||||
eigen_assert(this->rows() == 1 && this->cols() == 1);
|
||||
return derived().coeff(0,0);
|
||||
}
|
||||
|
||||
EIGEN_DEVICE_FUNC bool all() const;
|
||||
EIGEN_DEVICE_FUNC bool any() const;
|
||||
EIGEN_DEVICE_FUNC Index count() const;
|
||||
|
||||
typedef VectorwiseOp<Derived, Horizontal> RowwiseReturnType;
|
||||
typedef const VectorwiseOp<const Derived, Horizontal> ConstRowwiseReturnType;
|
||||
typedef VectorwiseOp<Derived, Vertical> ColwiseReturnType;
|
||||
typedef const VectorwiseOp<const Derived, Vertical> ConstColwiseReturnType;
|
||||
|
||||
/** \returns a VectorwiseOp wrapper of *this for broadcasting and partial reductions
|
||||
*
|
||||
* Example: \include MatrixBase_rowwise.cpp
|
||||
* Output: \verbinclude MatrixBase_rowwise.out
|
||||
*
|
||||
* \sa colwise(), class VectorwiseOp, \ref TutorialReductionsVisitorsBroadcasting
|
||||
*/
|
||||
//Code moved here due to a CUDA compiler bug
|
||||
EIGEN_DEVICE_FUNC inline ConstRowwiseReturnType rowwise() const {
|
||||
return ConstRowwiseReturnType(derived());
|
||||
}
|
||||
EIGEN_DEVICE_FUNC RowwiseReturnType rowwise();
|
||||
|
||||
/** \returns a VectorwiseOp wrapper of *this broadcasting and partial reductions
|
||||
*
|
||||
* Example: \include MatrixBase_colwise.cpp
|
||||
* Output: \verbinclude MatrixBase_colwise.out
|
||||
*
|
||||
* \sa rowwise(), class VectorwiseOp, \ref TutorialReductionsVisitorsBroadcasting
|
||||
*/
|
||||
EIGEN_DEVICE_FUNC inline ConstColwiseReturnType colwise() const {
|
||||
return ConstColwiseReturnType(derived());
|
||||
}
|
||||
EIGEN_DEVICE_FUNC ColwiseReturnType colwise();
|
||||
|
||||
typedef CwiseNullaryOp<internal::scalar_random_op<Scalar>,PlainObject> RandomReturnType;
|
||||
static const RandomReturnType Random(Index rows, Index cols);
|
||||
static const RandomReturnType Random(Index size);
|
||||
static const RandomReturnType Random();
|
||||
|
||||
template<typename ThenDerived,typename ElseDerived>
|
||||
inline EIGEN_DEVICE_FUNC const Select<Derived,ThenDerived,ElseDerived>
|
||||
select(const DenseBase<ThenDerived>& thenMatrix,
|
||||
const DenseBase<ElseDerived>& elseMatrix) const;
|
||||
|
||||
template<typename ThenDerived>
|
||||
inline EIGEN_DEVICE_FUNC const Select<Derived,ThenDerived, typename ThenDerived::ConstantReturnType>
|
||||
select(const DenseBase<ThenDerived>& thenMatrix, const typename ThenDerived::Scalar& elseScalar) const;
|
||||
|
||||
template<typename ElseDerived>
|
||||
inline EIGEN_DEVICE_FUNC const Select<Derived, typename ElseDerived::ConstantReturnType, ElseDerived >
|
||||
select(const typename ElseDerived::Scalar& thenScalar, const DenseBase<ElseDerived>& elseMatrix) const;
|
||||
|
||||
template<int p> RealScalar lpNorm() const;
|
||||
|
||||
template<int RowFactor, int ColFactor>
|
||||
EIGEN_DEVICE_FUNC
|
||||
const Replicate<Derived,RowFactor,ColFactor> replicate() const;
|
||||
/**
|
||||
* \return an expression of the replication of \c *this
|
||||
*
|
||||
* Example: \include MatrixBase_replicate_int_int.cpp
|
||||
* Output: \verbinclude MatrixBase_replicate_int_int.out
|
||||
*
|
||||
* \sa VectorwiseOp::replicate(), DenseBase::replicate<int,int>(), class Replicate
|
||||
*/
|
||||
//Code moved here due to a CUDA compiler bug
|
||||
EIGEN_DEVICE_FUNC
|
||||
const Replicate<Derived, Dynamic, Dynamic> replicate(Index rowFactor, Index colFactor) const
|
||||
{
|
||||
return Replicate<Derived, Dynamic, Dynamic>(derived(), rowFactor, colFactor);
|
||||
}
|
||||
|
||||
typedef Reverse<Derived, BothDirections> ReverseReturnType;
|
||||
typedef const Reverse<const Derived, BothDirections> ConstReverseReturnType;
|
||||
EIGEN_DEVICE_FUNC ReverseReturnType reverse();
|
||||
/** This is the const version of reverse(). */
|
||||
//Code moved here due to a CUDA compiler bug
|
||||
EIGEN_DEVICE_FUNC ConstReverseReturnType reverse() const
|
||||
{
|
||||
return ConstReverseReturnType(derived());
|
||||
}
|
||||
EIGEN_DEVICE_FUNC void reverseInPlace();
|
||||
|
||||
#ifdef EIGEN_PARSED_BY_DOXYGEN
|
||||
/** STL-like <a href="https://en.cppreference.com/w/cpp/named_req/RandomAccessIterator">RandomAccessIterator</a>
|
||||
* iterator type as returned by the begin() and end() methods.
|
||||
*/
|
||||
typedef random_access_iterator_type iterator;
|
||||
/** This is the const version of iterator (aka read-only) */
|
||||
typedef random_access_iterator_type const_iterator;
|
||||
#else
|
||||
typedef typename internal::conditional< (Flags&DirectAccessBit)==DirectAccessBit,
|
||||
internal::pointer_based_stl_iterator<Derived>,
|
||||
internal::generic_randaccess_stl_iterator<Derived>
|
||||
>::type iterator_type;
|
||||
|
||||
typedef typename internal::conditional< (Flags&DirectAccessBit)==DirectAccessBit,
|
||||
internal::pointer_based_stl_iterator<const Derived>,
|
||||
internal::generic_randaccess_stl_iterator<const Derived>
|
||||
>::type const_iterator_type;
|
||||
|
||||
// Stl-style iterators are supported only for vectors.
|
||||
|
||||
typedef typename internal::conditional< IsVectorAtCompileTime,
|
||||
iterator_type,
|
||||
void
|
||||
>::type iterator;
|
||||
|
||||
typedef typename internal::conditional< IsVectorAtCompileTime,
|
||||
const_iterator_type,
|
||||
void
|
||||
>::type const_iterator;
|
||||
#endif
|
||||
|
||||
inline iterator begin();
|
||||
inline const_iterator begin() const;
|
||||
inline const_iterator cbegin() const;
|
||||
inline iterator end();
|
||||
inline const_iterator end() const;
|
||||
inline const_iterator cend() const;
|
||||
|
||||
#define EIGEN_CURRENT_STORAGE_BASE_CLASS Eigen::DenseBase
|
||||
#define EIGEN_DOC_BLOCK_ADDONS_NOT_INNER_PANEL
|
||||
#define EIGEN_DOC_BLOCK_ADDONS_INNER_PANEL_IF(COND)
|
||||
#define EIGEN_DOC_UNARY_ADDONS(X,Y)
|
||||
# include "../plugins/CommonCwiseUnaryOps.h"
|
||||
# include "../plugins/BlockMethods.h"
|
||||
# include "../plugins/IndexedViewMethods.h"
|
||||
# include "../plugins/ReshapedMethods.h"
|
||||
# ifdef EIGEN_DENSEBASE_PLUGIN
|
||||
# include EIGEN_DENSEBASE_PLUGIN
|
||||
# endif
|
||||
#undef EIGEN_CURRENT_STORAGE_BASE_CLASS
|
||||
#undef EIGEN_DOC_BLOCK_ADDONS_NOT_INNER_PANEL
|
||||
#undef EIGEN_DOC_BLOCK_ADDONS_INNER_PANEL_IF
|
||||
#undef EIGEN_DOC_UNARY_ADDONS
|
||||
|
||||
// disable the use of evalTo for dense objects with a nice compilation error
|
||||
template<typename Dest>
|
||||
EIGEN_DEVICE_FUNC
|
||||
inline void evalTo(Dest& ) const
|
||||
{
|
||||
EIGEN_STATIC_ASSERT((internal::is_same<Dest,void>::value),THE_EVAL_EVALTO_FUNCTION_SHOULD_NEVER_BE_CALLED_FOR_DENSE_OBJECTS);
|
||||
}
|
||||
|
||||
protected:
|
||||
EIGEN_DEFAULT_COPY_CONSTRUCTOR(DenseBase)
|
||||
/** Default constructor. Do nothing. */
|
||||
EIGEN_DEVICE_FUNC DenseBase()
|
||||
{
|
||||
/* Just checks for self-consistency of the flags.
|
||||
* Only do it when debugging Eigen, as this borders on paranoia and could slow compilation down
|
||||
*/
|
||||
#ifdef EIGEN_INTERNAL_DEBUGGING
|
||||
EIGEN_STATIC_ASSERT((EIGEN_IMPLIES(MaxRowsAtCompileTime==1 && MaxColsAtCompileTime!=1, int(IsRowMajor))
|
||||
&& EIGEN_IMPLIES(MaxColsAtCompileTime==1 && MaxRowsAtCompileTime!=1, int(!IsRowMajor))),
|
||||
INVALID_STORAGE_ORDER_FOR_THIS_VECTOR_EXPRESSION)
|
||||
#endif
|
||||
}
|
||||
|
||||
private:
|
||||
EIGEN_DEVICE_FUNC explicit DenseBase(int);
|
||||
EIGEN_DEVICE_FUNC DenseBase(int,int);
|
||||
template<typename OtherDerived> EIGEN_DEVICE_FUNC explicit DenseBase(const DenseBase<OtherDerived>&);
|
||||
};
|
||||
|
||||
} // end namespace Eigen
|
||||
|
||||
#endif // EIGEN_DENSEBASE_H
|
||||
@@ -0,0 +1,685 @@
|
||||
// This file is part of Eigen, a lightweight C++ template library
|
||||
// for linear algebra.
|
||||
//
|
||||
// Copyright (C) 2006-2010 Benoit Jacob <jacob.benoit.1@gmail.com>
|
||||
//
|
||||
// This Source Code Form is subject to the terms of the Mozilla
|
||||
// Public License v. 2.0. If a copy of the MPL was not distributed
|
||||
// with this file, You can obtain one at http://mozilla.org/MPL/2.0/.
|
||||
|
||||
#ifndef EIGEN_DENSECOEFFSBASE_H
|
||||
#define EIGEN_DENSECOEFFSBASE_H
|
||||
|
||||
namespace Eigen {
|
||||
|
||||
namespace internal {
|
||||
template<typename T> struct add_const_on_value_type_if_arithmetic
|
||||
{
|
||||
typedef typename conditional<is_arithmetic<T>::value, T, typename add_const_on_value_type<T>::type>::type type;
|
||||
};
|
||||
}
|
||||
|
||||
/** \brief Base class providing read-only coefficient access to matrices and arrays.
|
||||
* \ingroup Core_Module
|
||||
* \tparam Derived Type of the derived class
|
||||
*
|
||||
* \note #ReadOnlyAccessors Constant indicating read-only access
|
||||
*
|
||||
* This class defines the \c operator() \c const function and friends, which can be used to read specific
|
||||
* entries of a matrix or array.
|
||||
*
|
||||
* \sa DenseCoeffsBase<Derived, WriteAccessors>, DenseCoeffsBase<Derived, DirectAccessors>,
|
||||
* \ref TopicClassHierarchy
|
||||
*/
|
||||
template<typename Derived>
|
||||
class DenseCoeffsBase<Derived,ReadOnlyAccessors> : public EigenBase<Derived>
|
||||
{
|
||||
public:
|
||||
typedef typename internal::traits<Derived>::StorageKind StorageKind;
|
||||
typedef typename internal::traits<Derived>::Scalar Scalar;
|
||||
typedef typename internal::packet_traits<Scalar>::type PacketScalar;
|
||||
|
||||
// Explanation for this CoeffReturnType typedef.
|
||||
// - This is the return type of the coeff() method.
|
||||
// - The LvalueBit means exactly that we can offer a coeffRef() method, which means exactly that we can get references
|
||||
// to coeffs, which means exactly that we can have coeff() return a const reference (as opposed to returning a value).
|
||||
// - The is_artihmetic check is required since "const int", "const double", etc. will cause warnings on some systems
|
||||
// while the declaration of "const T", where T is a non arithmetic type does not. Always returning "const Scalar&" is
|
||||
// not possible, since the underlying expressions might not offer a valid address the reference could be referring to.
|
||||
typedef typename internal::conditional<bool(internal::traits<Derived>::Flags&LvalueBit),
|
||||
const Scalar&,
|
||||
typename internal::conditional<internal::is_arithmetic<Scalar>::value, Scalar, const Scalar>::type
|
||||
>::type CoeffReturnType;
|
||||
|
||||
typedef typename internal::add_const_on_value_type_if_arithmetic<
|
||||
typename internal::packet_traits<Scalar>::type
|
||||
>::type PacketReturnType;
|
||||
|
||||
typedef EigenBase<Derived> Base;
|
||||
using Base::rows;
|
||||
using Base::cols;
|
||||
using Base::size;
|
||||
using Base::derived;
|
||||
|
||||
EIGEN_DEVICE_FUNC
|
||||
EIGEN_STRONG_INLINE Index rowIndexByOuterInner(Index outer, Index inner) const
|
||||
{
|
||||
return int(Derived::RowsAtCompileTime) == 1 ? 0
|
||||
: int(Derived::ColsAtCompileTime) == 1 ? inner
|
||||
: int(Derived::Flags)&RowMajorBit ? outer
|
||||
: inner;
|
||||
}
|
||||
|
||||
EIGEN_DEVICE_FUNC
|
||||
EIGEN_STRONG_INLINE Index colIndexByOuterInner(Index outer, Index inner) const
|
||||
{
|
||||
return int(Derived::ColsAtCompileTime) == 1 ? 0
|
||||
: int(Derived::RowsAtCompileTime) == 1 ? inner
|
||||
: int(Derived::Flags)&RowMajorBit ? inner
|
||||
: outer;
|
||||
}
|
||||
|
||||
/** Short version: don't use this function, use
|
||||
* \link operator()(Index,Index) const \endlink instead.
|
||||
*
|
||||
* Long version: this function is similar to
|
||||
* \link operator()(Index,Index) const \endlink, but without the assertion.
|
||||
* Use this for limiting the performance cost of debugging code when doing
|
||||
* repeated coefficient access. Only use this when it is guaranteed that the
|
||||
* parameters \a row and \a col are in range.
|
||||
*
|
||||
* If EIGEN_INTERNAL_DEBUGGING is defined, an assertion will be made, making this
|
||||
* function equivalent to \link operator()(Index,Index) const \endlink.
|
||||
*
|
||||
* \sa operator()(Index,Index) const, coeffRef(Index,Index), coeff(Index) const
|
||||
*/
|
||||
EIGEN_DEVICE_FUNC
|
||||
EIGEN_STRONG_INLINE CoeffReturnType coeff(Index row, Index col) const
|
||||
{
|
||||
eigen_internal_assert(row >= 0 && row < rows()
|
||||
&& col >= 0 && col < cols());
|
||||
return internal::evaluator<Derived>(derived()).coeff(row,col);
|
||||
}
|
||||
|
||||
EIGEN_DEVICE_FUNC
|
||||
EIGEN_STRONG_INLINE CoeffReturnType coeffByOuterInner(Index outer, Index inner) const
|
||||
{
|
||||
return coeff(rowIndexByOuterInner(outer, inner),
|
||||
colIndexByOuterInner(outer, inner));
|
||||
}
|
||||
|
||||
/** \returns the coefficient at given the given row and column.
|
||||
*
|
||||
* \sa operator()(Index,Index), operator[](Index)
|
||||
*/
|
||||
EIGEN_DEVICE_FUNC
|
||||
EIGEN_STRONG_INLINE CoeffReturnType operator()(Index row, Index col) const
|
||||
{
|
||||
eigen_assert(row >= 0 && row < rows()
|
||||
&& col >= 0 && col < cols());
|
||||
return coeff(row, col);
|
||||
}
|
||||
|
||||
/** Short version: don't use this function, use
|
||||
* \link operator[](Index) const \endlink instead.
|
||||
*
|
||||
* Long version: this function is similar to
|
||||
* \link operator[](Index) const \endlink, but without the assertion.
|
||||
* Use this for limiting the performance cost of debugging code when doing
|
||||
* repeated coefficient access. Only use this when it is guaranteed that the
|
||||
* parameter \a index is in range.
|
||||
*
|
||||
* If EIGEN_INTERNAL_DEBUGGING is defined, an assertion will be made, making this
|
||||
* function equivalent to \link operator[](Index) const \endlink.
|
||||
*
|
||||
* \sa operator[](Index) const, coeffRef(Index), coeff(Index,Index) const
|
||||
*/
|
||||
|
||||
EIGEN_DEVICE_FUNC
|
||||
EIGEN_STRONG_INLINE CoeffReturnType
|
||||
coeff(Index index) const
|
||||
{
|
||||
EIGEN_STATIC_ASSERT(internal::evaluator<Derived>::Flags & LinearAccessBit,
|
||||
THIS_COEFFICIENT_ACCESSOR_TAKING_ONE_ACCESS_IS_ONLY_FOR_EXPRESSIONS_ALLOWING_LINEAR_ACCESS)
|
||||
eigen_internal_assert(index >= 0 && index < size());
|
||||
return internal::evaluator<Derived>(derived()).coeff(index);
|
||||
}
|
||||
|
||||
|
||||
/** \returns the coefficient at given index.
|
||||
*
|
||||
* This method is allowed only for vector expressions, and for matrix expressions having the LinearAccessBit.
|
||||
*
|
||||
* \sa operator[](Index), operator()(Index,Index) const, x() const, y() const,
|
||||
* z() const, w() const
|
||||
*/
|
||||
|
||||
EIGEN_DEVICE_FUNC
|
||||
EIGEN_STRONG_INLINE CoeffReturnType
|
||||
operator[](Index index) const
|
||||
{
|
||||
EIGEN_STATIC_ASSERT(Derived::IsVectorAtCompileTime,
|
||||
THE_BRACKET_OPERATOR_IS_ONLY_FOR_VECTORS__USE_THE_PARENTHESIS_OPERATOR_INSTEAD)
|
||||
eigen_assert(index >= 0 && index < size());
|
||||
return coeff(index);
|
||||
}
|
||||
|
||||
/** \returns the coefficient at given index.
|
||||
*
|
||||
* This is synonymous to operator[](Index) const.
|
||||
*
|
||||
* This method is allowed only for vector expressions, and for matrix expressions having the LinearAccessBit.
|
||||
*
|
||||
* \sa operator[](Index), operator()(Index,Index) const, x() const, y() const,
|
||||
* z() const, w() const
|
||||
*/
|
||||
|
||||
EIGEN_DEVICE_FUNC
|
||||
EIGEN_STRONG_INLINE CoeffReturnType
|
||||
operator()(Index index) const
|
||||
{
|
||||
eigen_assert(index >= 0 && index < size());
|
||||
return coeff(index);
|
||||
}
|
||||
|
||||
/** equivalent to operator[](0). */
|
||||
|
||||
EIGEN_DEVICE_FUNC
|
||||
EIGEN_STRONG_INLINE CoeffReturnType
|
||||
x() const { return (*this)[0]; }
|
||||
|
||||
/** equivalent to operator[](1). */
|
||||
|
||||
EIGEN_DEVICE_FUNC
|
||||
EIGEN_STRONG_INLINE CoeffReturnType
|
||||
y() const
|
||||
{
|
||||
EIGEN_STATIC_ASSERT(Derived::SizeAtCompileTime==-1 || Derived::SizeAtCompileTime>=2, OUT_OF_RANGE_ACCESS);
|
||||
return (*this)[1];
|
||||
}
|
||||
|
||||
/** equivalent to operator[](2). */
|
||||
|
||||
EIGEN_DEVICE_FUNC
|
||||
EIGEN_STRONG_INLINE CoeffReturnType
|
||||
z() const
|
||||
{
|
||||
EIGEN_STATIC_ASSERT(Derived::SizeAtCompileTime==-1 || Derived::SizeAtCompileTime>=3, OUT_OF_RANGE_ACCESS);
|
||||
return (*this)[2];
|
||||
}
|
||||
|
||||
/** equivalent to operator[](3). */
|
||||
|
||||
EIGEN_DEVICE_FUNC
|
||||
EIGEN_STRONG_INLINE CoeffReturnType
|
||||
w() const
|
||||
{
|
||||
EIGEN_STATIC_ASSERT(Derived::SizeAtCompileTime==-1 || Derived::SizeAtCompileTime>=4, OUT_OF_RANGE_ACCESS);
|
||||
return (*this)[3];
|
||||
}
|
||||
|
||||
/** \internal
|
||||
* \returns the packet of coefficients starting at the given row and column. It is your responsibility
|
||||
* to ensure that a packet really starts there. This method is only available on expressions having the
|
||||
* PacketAccessBit.
|
||||
*
|
||||
* The \a LoadMode parameter may have the value \a #Aligned or \a #Unaligned. Its effect is to select
|
||||
* the appropriate vectorization instruction. Aligned access is faster, but is only possible for packets
|
||||
* starting at an address which is a multiple of the packet size.
|
||||
*/
|
||||
|
||||
template<int LoadMode>
|
||||
EIGEN_STRONG_INLINE PacketReturnType packet(Index row, Index col) const
|
||||
{
|
||||
typedef typename internal::packet_traits<Scalar>::type DefaultPacketType;
|
||||
eigen_internal_assert(row >= 0 && row < rows() && col >= 0 && col < cols());
|
||||
return internal::evaluator<Derived>(derived()).template packet<LoadMode,DefaultPacketType>(row,col);
|
||||
}
|
||||
|
||||
|
||||
/** \internal */
|
||||
template<int LoadMode>
|
||||
EIGEN_STRONG_INLINE PacketReturnType packetByOuterInner(Index outer, Index inner) const
|
||||
{
|
||||
return packet<LoadMode>(rowIndexByOuterInner(outer, inner),
|
||||
colIndexByOuterInner(outer, inner));
|
||||
}
|
||||
|
||||
/** \internal
|
||||
* \returns the packet of coefficients starting at the given index. It is your responsibility
|
||||
* to ensure that a packet really starts there. This method is only available on expressions having the
|
||||
* PacketAccessBit and the LinearAccessBit.
|
||||
*
|
||||
* The \a LoadMode parameter may have the value \a #Aligned or \a #Unaligned. Its effect is to select
|
||||
* the appropriate vectorization instruction. Aligned access is faster, but is only possible for packets
|
||||
* starting at an address which is a multiple of the packet size.
|
||||
*/
|
||||
|
||||
template<int LoadMode>
|
||||
EIGEN_STRONG_INLINE PacketReturnType packet(Index index) const
|
||||
{
|
||||
EIGEN_STATIC_ASSERT(internal::evaluator<Derived>::Flags & LinearAccessBit,
|
||||
THIS_COEFFICIENT_ACCESSOR_TAKING_ONE_ACCESS_IS_ONLY_FOR_EXPRESSIONS_ALLOWING_LINEAR_ACCESS)
|
||||
typedef typename internal::packet_traits<Scalar>::type DefaultPacketType;
|
||||
eigen_internal_assert(index >= 0 && index < size());
|
||||
return internal::evaluator<Derived>(derived()).template packet<LoadMode,DefaultPacketType>(index);
|
||||
}
|
||||
|
||||
protected:
|
||||
// explanation: DenseBase is doing "using ..." on the methods from DenseCoeffsBase.
|
||||
// But some methods are only available in the DirectAccess case.
|
||||
// So we add dummy methods here with these names, so that "using... " doesn't fail.
|
||||
// It's not private so that the child class DenseBase can access them, and it's not public
|
||||
// either since it's an implementation detail, so has to be protected.
|
||||
void coeffRef();
|
||||
void coeffRefByOuterInner();
|
||||
void writePacket();
|
||||
void writePacketByOuterInner();
|
||||
void copyCoeff();
|
||||
void copyCoeffByOuterInner();
|
||||
void copyPacket();
|
||||
void copyPacketByOuterInner();
|
||||
void stride();
|
||||
void innerStride();
|
||||
void outerStride();
|
||||
void rowStride();
|
||||
void colStride();
|
||||
};
|
||||
|
||||
/** \brief Base class providing read/write coefficient access to matrices and arrays.
|
||||
* \ingroup Core_Module
|
||||
* \tparam Derived Type of the derived class
|
||||
*
|
||||
* \note #WriteAccessors Constant indicating read/write access
|
||||
*
|
||||
* This class defines the non-const \c operator() function and friends, which can be used to write specific
|
||||
* entries of a matrix or array. This class inherits DenseCoeffsBase<Derived, ReadOnlyAccessors> which
|
||||
* defines the const variant for reading specific entries.
|
||||
*
|
||||
* \sa DenseCoeffsBase<Derived, DirectAccessors>, \ref TopicClassHierarchy
|
||||
*/
|
||||
template<typename Derived>
|
||||
class DenseCoeffsBase<Derived, WriteAccessors> : public DenseCoeffsBase<Derived, ReadOnlyAccessors>
|
||||
{
|
||||
public:
|
||||
|
||||
typedef DenseCoeffsBase<Derived, ReadOnlyAccessors> Base;
|
||||
|
||||
typedef typename internal::traits<Derived>::StorageKind StorageKind;
|
||||
typedef typename internal::traits<Derived>::Scalar Scalar;
|
||||
typedef typename internal::packet_traits<Scalar>::type PacketScalar;
|
||||
typedef typename NumTraits<Scalar>::Real RealScalar;
|
||||
|
||||
using Base::coeff;
|
||||
using Base::rows;
|
||||
using Base::cols;
|
||||
using Base::size;
|
||||
using Base::derived;
|
||||
using Base::rowIndexByOuterInner;
|
||||
using Base::colIndexByOuterInner;
|
||||
using Base::operator[];
|
||||
using Base::operator();
|
||||
using Base::x;
|
||||
using Base::y;
|
||||
using Base::z;
|
||||
using Base::w;
|
||||
|
||||
/** Short version: don't use this function, use
|
||||
* \link operator()(Index,Index) \endlink instead.
|
||||
*
|
||||
* Long version: this function is similar to
|
||||
* \link operator()(Index,Index) \endlink, but without the assertion.
|
||||
* Use this for limiting the performance cost of debugging code when doing
|
||||
* repeated coefficient access. Only use this when it is guaranteed that the
|
||||
* parameters \a row and \a col are in range.
|
||||
*
|
||||
* If EIGEN_INTERNAL_DEBUGGING is defined, an assertion will be made, making this
|
||||
* function equivalent to \link operator()(Index,Index) \endlink.
|
||||
*
|
||||
* \sa operator()(Index,Index), coeff(Index, Index) const, coeffRef(Index)
|
||||
*/
|
||||
EIGEN_DEVICE_FUNC
|
||||
EIGEN_STRONG_INLINE Scalar& coeffRef(Index row, Index col)
|
||||
{
|
||||
eigen_internal_assert(row >= 0 && row < rows()
|
||||
&& col >= 0 && col < cols());
|
||||
return internal::evaluator<Derived>(derived()).coeffRef(row,col);
|
||||
}
|
||||
|
||||
EIGEN_DEVICE_FUNC
|
||||
EIGEN_STRONG_INLINE Scalar&
|
||||
coeffRefByOuterInner(Index outer, Index inner)
|
||||
{
|
||||
return coeffRef(rowIndexByOuterInner(outer, inner),
|
||||
colIndexByOuterInner(outer, inner));
|
||||
}
|
||||
|
||||
/** \returns a reference to the coefficient at given the given row and column.
|
||||
*
|
||||
* \sa operator[](Index)
|
||||
*/
|
||||
|
||||
EIGEN_DEVICE_FUNC
|
||||
EIGEN_STRONG_INLINE Scalar&
|
||||
operator()(Index row, Index col)
|
||||
{
|
||||
eigen_assert(row >= 0 && row < rows()
|
||||
&& col >= 0 && col < cols());
|
||||
return coeffRef(row, col);
|
||||
}
|
||||
|
||||
|
||||
/** Short version: don't use this function, use
|
||||
* \link operator[](Index) \endlink instead.
|
||||
*
|
||||
* Long version: this function is similar to
|
||||
* \link operator[](Index) \endlink, but without the assertion.
|
||||
* Use this for limiting the performance cost of debugging code when doing
|
||||
* repeated coefficient access. Only use this when it is guaranteed that the
|
||||
* parameters \a row and \a col are in range.
|
||||
*
|
||||
* If EIGEN_INTERNAL_DEBUGGING is defined, an assertion will be made, making this
|
||||
* function equivalent to \link operator[](Index) \endlink.
|
||||
*
|
||||
* \sa operator[](Index), coeff(Index) const, coeffRef(Index,Index)
|
||||
*/
|
||||
|
||||
EIGEN_DEVICE_FUNC
|
||||
EIGEN_STRONG_INLINE Scalar&
|
||||
coeffRef(Index index)
|
||||
{
|
||||
EIGEN_STATIC_ASSERT(internal::evaluator<Derived>::Flags & LinearAccessBit,
|
||||
THIS_COEFFICIENT_ACCESSOR_TAKING_ONE_ACCESS_IS_ONLY_FOR_EXPRESSIONS_ALLOWING_LINEAR_ACCESS)
|
||||
eigen_internal_assert(index >= 0 && index < size());
|
||||
return internal::evaluator<Derived>(derived()).coeffRef(index);
|
||||
}
|
||||
|
||||
/** \returns a reference to the coefficient at given index.
|
||||
*
|
||||
* This method is allowed only for vector expressions, and for matrix expressions having the LinearAccessBit.
|
||||
*
|
||||
* \sa operator[](Index) const, operator()(Index,Index), x(), y(), z(), w()
|
||||
*/
|
||||
|
||||
EIGEN_DEVICE_FUNC
|
||||
EIGEN_STRONG_INLINE Scalar&
|
||||
operator[](Index index)
|
||||
{
|
||||
EIGEN_STATIC_ASSERT(Derived::IsVectorAtCompileTime,
|
||||
THE_BRACKET_OPERATOR_IS_ONLY_FOR_VECTORS__USE_THE_PARENTHESIS_OPERATOR_INSTEAD)
|
||||
eigen_assert(index >= 0 && index < size());
|
||||
return coeffRef(index);
|
||||
}
|
||||
|
||||
/** \returns a reference to the coefficient at given index.
|
||||
*
|
||||
* This is synonymous to operator[](Index).
|
||||
*
|
||||
* This method is allowed only for vector expressions, and for matrix expressions having the LinearAccessBit.
|
||||
*
|
||||
* \sa operator[](Index) const, operator()(Index,Index), x(), y(), z(), w()
|
||||
*/
|
||||
|
||||
EIGEN_DEVICE_FUNC
|
||||
EIGEN_STRONG_INLINE Scalar&
|
||||
operator()(Index index)
|
||||
{
|
||||
eigen_assert(index >= 0 && index < size());
|
||||
return coeffRef(index);
|
||||
}
|
||||
|
||||
/** equivalent to operator[](0). */
|
||||
|
||||
EIGEN_DEVICE_FUNC
|
||||
EIGEN_STRONG_INLINE Scalar&
|
||||
x() { return (*this)[0]; }
|
||||
|
||||
/** equivalent to operator[](1). */
|
||||
|
||||
EIGEN_DEVICE_FUNC
|
||||
EIGEN_STRONG_INLINE Scalar&
|
||||
y()
|
||||
{
|
||||
EIGEN_STATIC_ASSERT(Derived::SizeAtCompileTime==-1 || Derived::SizeAtCompileTime>=2, OUT_OF_RANGE_ACCESS);
|
||||
return (*this)[1];
|
||||
}
|
||||
|
||||
/** equivalent to operator[](2). */
|
||||
|
||||
EIGEN_DEVICE_FUNC
|
||||
EIGEN_STRONG_INLINE Scalar&
|
||||
z()
|
||||
{
|
||||
EIGEN_STATIC_ASSERT(Derived::SizeAtCompileTime==-1 || Derived::SizeAtCompileTime>=3, OUT_OF_RANGE_ACCESS);
|
||||
return (*this)[2];
|
||||
}
|
||||
|
||||
/** equivalent to operator[](3). */
|
||||
|
||||
EIGEN_DEVICE_FUNC
|
||||
EIGEN_STRONG_INLINE Scalar&
|
||||
w()
|
||||
{
|
||||
EIGEN_STATIC_ASSERT(Derived::SizeAtCompileTime==-1 || Derived::SizeAtCompileTime>=4, OUT_OF_RANGE_ACCESS);
|
||||
return (*this)[3];
|
||||
}
|
||||
};
|
||||
|
||||
/** \brief Base class providing direct read-only coefficient access to matrices and arrays.
|
||||
* \ingroup Core_Module
|
||||
* \tparam Derived Type of the derived class
|
||||
*
|
||||
* \note #DirectAccessors Constant indicating direct access
|
||||
*
|
||||
* This class defines functions to work with strides which can be used to access entries directly. This class
|
||||
* inherits DenseCoeffsBase<Derived, ReadOnlyAccessors> which defines functions to access entries read-only using
|
||||
* \c operator() .
|
||||
*
|
||||
* \sa \blank \ref TopicClassHierarchy
|
||||
*/
|
||||
template<typename Derived>
|
||||
class DenseCoeffsBase<Derived, DirectAccessors> : public DenseCoeffsBase<Derived, ReadOnlyAccessors>
|
||||
{
|
||||
public:
|
||||
|
||||
typedef DenseCoeffsBase<Derived, ReadOnlyAccessors> Base;
|
||||
typedef typename internal::traits<Derived>::Scalar Scalar;
|
||||
typedef typename NumTraits<Scalar>::Real RealScalar;
|
||||
|
||||
using Base::rows;
|
||||
using Base::cols;
|
||||
using Base::size;
|
||||
using Base::derived;
|
||||
|
||||
/** \returns the pointer increment between two consecutive elements within a slice in the inner direction.
|
||||
*
|
||||
* \sa outerStride(), rowStride(), colStride()
|
||||
*/
|
||||
EIGEN_DEVICE_FUNC EIGEN_CONSTEXPR
|
||||
inline Index innerStride() const
|
||||
{
|
||||
return derived().innerStride();
|
||||
}
|
||||
|
||||
/** \returns the pointer increment between two consecutive inner slices (for example, between two consecutive columns
|
||||
* in a column-major matrix).
|
||||
*
|
||||
* \sa innerStride(), rowStride(), colStride()
|
||||
*/
|
||||
EIGEN_DEVICE_FUNC EIGEN_CONSTEXPR
|
||||
inline Index outerStride() const
|
||||
{
|
||||
return derived().outerStride();
|
||||
}
|
||||
|
||||
// FIXME shall we remove it ?
|
||||
EIGEN_CONSTEXPR inline Index stride() const
|
||||
{
|
||||
return Derived::IsVectorAtCompileTime ? innerStride() : outerStride();
|
||||
}
|
||||
|
||||
/** \returns the pointer increment between two consecutive rows.
|
||||
*
|
||||
* \sa innerStride(), outerStride(), colStride()
|
||||
*/
|
||||
EIGEN_DEVICE_FUNC EIGEN_CONSTEXPR
|
||||
inline Index rowStride() const
|
||||
{
|
||||
return Derived::IsRowMajor ? outerStride() : innerStride();
|
||||
}
|
||||
|
||||
/** \returns the pointer increment between two consecutive columns.
|
||||
*
|
||||
* \sa innerStride(), outerStride(), rowStride()
|
||||
*/
|
||||
EIGEN_DEVICE_FUNC EIGEN_CONSTEXPR
|
||||
inline Index colStride() const
|
||||
{
|
||||
return Derived::IsRowMajor ? innerStride() : outerStride();
|
||||
}
|
||||
};
|
||||
|
||||
/** \brief Base class providing direct read/write coefficient access to matrices and arrays.
|
||||
* \ingroup Core_Module
|
||||
* \tparam Derived Type of the derived class
|
||||
*
|
||||
* \note #DirectWriteAccessors Constant indicating direct access
|
||||
*
|
||||
* This class defines functions to work with strides which can be used to access entries directly. This class
|
||||
* inherits DenseCoeffsBase<Derived, WriteAccessors> which defines functions to access entries read/write using
|
||||
* \c operator().
|
||||
*
|
||||
* \sa \blank \ref TopicClassHierarchy
|
||||
*/
|
||||
template<typename Derived>
|
||||
class DenseCoeffsBase<Derived, DirectWriteAccessors>
|
||||
: public DenseCoeffsBase<Derived, WriteAccessors>
|
||||
{
|
||||
public:
|
||||
|
||||
typedef DenseCoeffsBase<Derived, WriteAccessors> Base;
|
||||
typedef typename internal::traits<Derived>::Scalar Scalar;
|
||||
typedef typename NumTraits<Scalar>::Real RealScalar;
|
||||
|
||||
using Base::rows;
|
||||
using Base::cols;
|
||||
using Base::size;
|
||||
using Base::derived;
|
||||
|
||||
/** \returns the pointer increment between two consecutive elements within a slice in the inner direction.
|
||||
*
|
||||
* \sa outerStride(), rowStride(), colStride()
|
||||
*/
|
||||
EIGEN_DEVICE_FUNC EIGEN_CONSTEXPR
|
||||
inline Index innerStride() const EIGEN_NOEXCEPT
|
||||
{
|
||||
return derived().innerStride();
|
||||
}
|
||||
|
||||
/** \returns the pointer increment between two consecutive inner slices (for example, between two consecutive columns
|
||||
* in a column-major matrix).
|
||||
*
|
||||
* \sa innerStride(), rowStride(), colStride()
|
||||
*/
|
||||
EIGEN_DEVICE_FUNC EIGEN_CONSTEXPR
|
||||
inline Index outerStride() const EIGEN_NOEXCEPT
|
||||
{
|
||||
return derived().outerStride();
|
||||
}
|
||||
|
||||
// FIXME shall we remove it ?
|
||||
EIGEN_CONSTEXPR inline Index stride() const EIGEN_NOEXCEPT
|
||||
{
|
||||
return Derived::IsVectorAtCompileTime ? innerStride() : outerStride();
|
||||
}
|
||||
|
||||
/** \returns the pointer increment between two consecutive rows.
|
||||
*
|
||||
* \sa innerStride(), outerStride(), colStride()
|
||||
*/
|
||||
EIGEN_DEVICE_FUNC EIGEN_CONSTEXPR
|
||||
inline Index rowStride() const EIGEN_NOEXCEPT
|
||||
{
|
||||
return Derived::IsRowMajor ? outerStride() : innerStride();
|
||||
}
|
||||
|
||||
/** \returns the pointer increment between two consecutive columns.
|
||||
*
|
||||
* \sa innerStride(), outerStride(), rowStride()
|
||||
*/
|
||||
EIGEN_DEVICE_FUNC EIGEN_CONSTEXPR
|
||||
inline Index colStride() const EIGEN_NOEXCEPT
|
||||
{
|
||||
return Derived::IsRowMajor ? innerStride() : outerStride();
|
||||
}
|
||||
};
|
||||
|
||||
namespace internal {
|
||||
|
||||
template<int Alignment, typename Derived, bool JustReturnZero>
|
||||
struct first_aligned_impl
|
||||
{
|
||||
static EIGEN_CONSTEXPR inline Index run(const Derived&) EIGEN_NOEXCEPT
|
||||
{ return 0; }
|
||||
};
|
||||
|
||||
template<int Alignment, typename Derived>
|
||||
struct first_aligned_impl<Alignment, Derived, false>
|
||||
{
|
||||
static inline Index run(const Derived& m)
|
||||
{
|
||||
return internal::first_aligned<Alignment>(m.data(), m.size());
|
||||
}
|
||||
};
|
||||
|
||||
/** \internal \returns the index of the first element of the array stored by \a m that is properly aligned with respect to \a Alignment for vectorization.
|
||||
*
|
||||
* \tparam Alignment requested alignment in Bytes.
|
||||
*
|
||||
* There is also the variant first_aligned(const Scalar*, Integer) defined in Memory.h. See it for more
|
||||
* documentation.
|
||||
*/
|
||||
template<int Alignment, typename Derived>
|
||||
static inline Index first_aligned(const DenseBase<Derived>& m)
|
||||
{
|
||||
enum { ReturnZero = (int(evaluator<Derived>::Alignment) >= Alignment) || !(Derived::Flags & DirectAccessBit) };
|
||||
return first_aligned_impl<Alignment, Derived, ReturnZero>::run(m.derived());
|
||||
}
|
||||
|
||||
template<typename Derived>
|
||||
static inline Index first_default_aligned(const DenseBase<Derived>& m)
|
||||
{
|
||||
typedef typename Derived::Scalar Scalar;
|
||||
typedef typename packet_traits<Scalar>::type DefaultPacketType;
|
||||
return internal::first_aligned<int(unpacket_traits<DefaultPacketType>::alignment),Derived>(m);
|
||||
}
|
||||
|
||||
template<typename Derived, bool HasDirectAccess = has_direct_access<Derived>::ret>
|
||||
struct inner_stride_at_compile_time
|
||||
{
|
||||
enum { ret = traits<Derived>::InnerStrideAtCompileTime };
|
||||
};
|
||||
|
||||
template<typename Derived>
|
||||
struct inner_stride_at_compile_time<Derived, false>
|
||||
{
|
||||
enum { ret = 0 };
|
||||
};
|
||||
|
||||
template<typename Derived, bool HasDirectAccess = has_direct_access<Derived>::ret>
|
||||
struct outer_stride_at_compile_time
|
||||
{
|
||||
enum { ret = traits<Derived>::OuterStrideAtCompileTime };
|
||||
};
|
||||
|
||||
template<typename Derived>
|
||||
struct outer_stride_at_compile_time<Derived, false>
|
||||
{
|
||||
enum { ret = 0 };
|
||||
};
|
||||
|
||||
} // end namespace internal
|
||||
|
||||
} // end namespace Eigen
|
||||
|
||||
#endif // EIGEN_DENSECOEFFSBASE_H
|
||||
@@ -0,0 +1,652 @@
|
||||
// This file is part of Eigen, a lightweight C++ template library
|
||||
// for linear algebra.
|
||||
//
|
||||
// Copyright (C) 2008 Gael Guennebaud <gael.guennebaud@inria.fr>
|
||||
// Copyright (C) 2006-2009 Benoit Jacob <jacob.benoit.1@gmail.com>
|
||||
// Copyright (C) 2010-2013 Hauke Heibel <hauke.heibel@gmail.com>
|
||||
//
|
||||
// This Source Code Form is subject to the terms of the Mozilla
|
||||
// Public License v. 2.0. If a copy of the MPL was not distributed
|
||||
// with this file, You can obtain one at http://mozilla.org/MPL/2.0/.
|
||||
|
||||
#ifndef EIGEN_MATRIXSTORAGE_H
|
||||
#define EIGEN_MATRIXSTORAGE_H
|
||||
|
||||
#ifdef EIGEN_DENSE_STORAGE_CTOR_PLUGIN
|
||||
#define EIGEN_INTERNAL_DENSE_STORAGE_CTOR_PLUGIN(X) X; EIGEN_DENSE_STORAGE_CTOR_PLUGIN;
|
||||
#else
|
||||
#define EIGEN_INTERNAL_DENSE_STORAGE_CTOR_PLUGIN(X)
|
||||
#endif
|
||||
|
||||
namespace Eigen {
|
||||
|
||||
namespace internal {
|
||||
|
||||
struct constructor_without_unaligned_array_assert {};
|
||||
|
||||
template<typename T, int Size>
|
||||
EIGEN_DEVICE_FUNC
|
||||
void check_static_allocation_size()
|
||||
{
|
||||
// if EIGEN_STACK_ALLOCATION_LIMIT is defined to 0, then no limit
|
||||
#if EIGEN_STACK_ALLOCATION_LIMIT
|
||||
EIGEN_STATIC_ASSERT(Size * sizeof(T) <= EIGEN_STACK_ALLOCATION_LIMIT, OBJECT_ALLOCATED_ON_STACK_IS_TOO_BIG);
|
||||
#endif
|
||||
}
|
||||
|
||||
/** \internal
|
||||
* Static array. If the MatrixOrArrayOptions require auto-alignment, the array will be automatically aligned:
|
||||
* to 16 bytes boundary if the total size is a multiple of 16 bytes.
|
||||
*/
|
||||
template <typename T, int Size, int MatrixOrArrayOptions,
|
||||
int Alignment = (MatrixOrArrayOptions&DontAlign) ? 0
|
||||
: compute_default_alignment<T,Size>::value >
|
||||
struct plain_array
|
||||
{
|
||||
T array[Size];
|
||||
|
||||
EIGEN_DEVICE_FUNC
|
||||
plain_array()
|
||||
{
|
||||
check_static_allocation_size<T,Size>();
|
||||
}
|
||||
|
||||
EIGEN_DEVICE_FUNC
|
||||
plain_array(constructor_without_unaligned_array_assert)
|
||||
{
|
||||
check_static_allocation_size<T,Size>();
|
||||
}
|
||||
};
|
||||
|
||||
#if defined(EIGEN_DISABLE_UNALIGNED_ARRAY_ASSERT)
|
||||
#define EIGEN_MAKE_UNALIGNED_ARRAY_ASSERT(sizemask)
|
||||
#elif EIGEN_GNUC_AT_LEAST(4,7)
|
||||
// GCC 4.7 is too aggressive in its optimizations and remove the alignment test based on the fact the array is declared to be aligned.
|
||||
// See this bug report: http://gcc.gnu.org/bugzilla/show_bug.cgi?id=53900
|
||||
// Hiding the origin of the array pointer behind a function argument seems to do the trick even if the function is inlined:
|
||||
template<typename PtrType>
|
||||
EIGEN_ALWAYS_INLINE PtrType eigen_unaligned_array_assert_workaround_gcc47(PtrType array) { return array; }
|
||||
#define EIGEN_MAKE_UNALIGNED_ARRAY_ASSERT(sizemask) \
|
||||
eigen_assert((internal::UIntPtr(eigen_unaligned_array_assert_workaround_gcc47(array)) & (sizemask)) == 0 \
|
||||
&& "this assertion is explained here: " \
|
||||
"http://eigen.tuxfamily.org/dox-devel/group__TopicUnalignedArrayAssert.html" \
|
||||
" **** READ THIS WEB PAGE !!! ****");
|
||||
#else
|
||||
#define EIGEN_MAKE_UNALIGNED_ARRAY_ASSERT(sizemask) \
|
||||
eigen_assert((internal::UIntPtr(array) & (sizemask)) == 0 \
|
||||
&& "this assertion is explained here: " \
|
||||
"http://eigen.tuxfamily.org/dox-devel/group__TopicUnalignedArrayAssert.html" \
|
||||
" **** READ THIS WEB PAGE !!! ****");
|
||||
#endif
|
||||
|
||||
template <typename T, int Size, int MatrixOrArrayOptions>
|
||||
struct plain_array<T, Size, MatrixOrArrayOptions, 8>
|
||||
{
|
||||
EIGEN_ALIGN_TO_BOUNDARY(8) T array[Size];
|
||||
|
||||
EIGEN_DEVICE_FUNC
|
||||
plain_array()
|
||||
{
|
||||
EIGEN_MAKE_UNALIGNED_ARRAY_ASSERT(7);
|
||||
check_static_allocation_size<T,Size>();
|
||||
}
|
||||
|
||||
EIGEN_DEVICE_FUNC
|
||||
plain_array(constructor_without_unaligned_array_assert)
|
||||
{
|
||||
check_static_allocation_size<T,Size>();
|
||||
}
|
||||
};
|
||||
|
||||
template <typename T, int Size, int MatrixOrArrayOptions>
|
||||
struct plain_array<T, Size, MatrixOrArrayOptions, 16>
|
||||
{
|
||||
EIGEN_ALIGN_TO_BOUNDARY(16) T array[Size];
|
||||
|
||||
EIGEN_DEVICE_FUNC
|
||||
plain_array()
|
||||
{
|
||||
EIGEN_MAKE_UNALIGNED_ARRAY_ASSERT(15);
|
||||
check_static_allocation_size<T,Size>();
|
||||
}
|
||||
|
||||
EIGEN_DEVICE_FUNC
|
||||
plain_array(constructor_without_unaligned_array_assert)
|
||||
{
|
||||
check_static_allocation_size<T,Size>();
|
||||
}
|
||||
};
|
||||
|
||||
template <typename T, int Size, int MatrixOrArrayOptions>
|
||||
struct plain_array<T, Size, MatrixOrArrayOptions, 32>
|
||||
{
|
||||
EIGEN_ALIGN_TO_BOUNDARY(32) T array[Size];
|
||||
|
||||
EIGEN_DEVICE_FUNC
|
||||
plain_array()
|
||||
{
|
||||
EIGEN_MAKE_UNALIGNED_ARRAY_ASSERT(31);
|
||||
check_static_allocation_size<T,Size>();
|
||||
}
|
||||
|
||||
EIGEN_DEVICE_FUNC
|
||||
plain_array(constructor_without_unaligned_array_assert)
|
||||
{
|
||||
check_static_allocation_size<T,Size>();
|
||||
}
|
||||
};
|
||||
|
||||
template <typename T, int Size, int MatrixOrArrayOptions>
|
||||
struct plain_array<T, Size, MatrixOrArrayOptions, 64>
|
||||
{
|
||||
EIGEN_ALIGN_TO_BOUNDARY(64) T array[Size];
|
||||
|
||||
EIGEN_DEVICE_FUNC
|
||||
plain_array()
|
||||
{
|
||||
EIGEN_MAKE_UNALIGNED_ARRAY_ASSERT(63);
|
||||
check_static_allocation_size<T,Size>();
|
||||
}
|
||||
|
||||
EIGEN_DEVICE_FUNC
|
||||
plain_array(constructor_without_unaligned_array_assert)
|
||||
{
|
||||
check_static_allocation_size<T,Size>();
|
||||
}
|
||||
};
|
||||
|
||||
template <typename T, int MatrixOrArrayOptions, int Alignment>
|
||||
struct plain_array<T, 0, MatrixOrArrayOptions, Alignment>
|
||||
{
|
||||
T array[1];
|
||||
EIGEN_DEVICE_FUNC plain_array() {}
|
||||
EIGEN_DEVICE_FUNC plain_array(constructor_without_unaligned_array_assert) {}
|
||||
};
|
||||
|
||||
struct plain_array_helper {
|
||||
template<typename T, int Size, int MatrixOrArrayOptions, int Alignment>
|
||||
EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
|
||||
static void copy(const plain_array<T, Size, MatrixOrArrayOptions, Alignment>& src, const Eigen::Index size,
|
||||
plain_array<T, Size, MatrixOrArrayOptions, Alignment>& dst) {
|
||||
smart_copy(src.array, src.array + size, dst.array);
|
||||
}
|
||||
|
||||
template<typename T, int Size, int MatrixOrArrayOptions, int Alignment>
|
||||
EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
|
||||
static void swap(plain_array<T, Size, MatrixOrArrayOptions, Alignment>& a, const Eigen::Index a_size,
|
||||
plain_array<T, Size, MatrixOrArrayOptions, Alignment>& b, const Eigen::Index b_size) {
|
||||
if (a_size < b_size) {
|
||||
std::swap_ranges(b.array, b.array + a_size, a.array);
|
||||
smart_move(b.array + a_size, b.array + b_size, a.array + a_size);
|
||||
} else if (a_size > b_size) {
|
||||
std::swap_ranges(a.array, a.array + b_size, b.array);
|
||||
smart_move(a.array + b_size, a.array + a_size, b.array + b_size);
|
||||
} else {
|
||||
std::swap_ranges(a.array, a.array + a_size, b.array);
|
||||
}
|
||||
}
|
||||
};
|
||||
|
||||
} // end namespace internal
|
||||
|
||||
/** \internal
|
||||
*
|
||||
* \class DenseStorage
|
||||
* \ingroup Core_Module
|
||||
*
|
||||
* \brief Stores the data of a matrix
|
||||
*
|
||||
* This class stores the data of fixed-size, dynamic-size or mixed matrices
|
||||
* in a way as compact as possible.
|
||||
*
|
||||
* \sa Matrix
|
||||
*/
|
||||
template<typename T, int Size, int _Rows, int _Cols, int _Options> class DenseStorage;
|
||||
|
||||
// purely fixed-size matrix
|
||||
template<typename T, int Size, int _Rows, int _Cols, int _Options> class DenseStorage
|
||||
{
|
||||
internal::plain_array<T,Size,_Options> m_data;
|
||||
public:
|
||||
EIGEN_DEVICE_FUNC DenseStorage() {
|
||||
EIGEN_INTERNAL_DENSE_STORAGE_CTOR_PLUGIN(Index size = Size)
|
||||
}
|
||||
EIGEN_DEVICE_FUNC
|
||||
explicit DenseStorage(internal::constructor_without_unaligned_array_assert)
|
||||
: m_data(internal::constructor_without_unaligned_array_assert()) {}
|
||||
#if !EIGEN_HAS_CXX11 || defined(EIGEN_DENSE_STORAGE_CTOR_PLUGIN)
|
||||
EIGEN_DEVICE_FUNC
|
||||
DenseStorage(const DenseStorage& other) : m_data(other.m_data) {
|
||||
EIGEN_INTERNAL_DENSE_STORAGE_CTOR_PLUGIN(Index size = Size)
|
||||
}
|
||||
#else
|
||||
EIGEN_DEVICE_FUNC DenseStorage(const DenseStorage&) = default;
|
||||
#endif
|
||||
#if !EIGEN_HAS_CXX11
|
||||
EIGEN_DEVICE_FUNC
|
||||
DenseStorage& operator=(const DenseStorage& other)
|
||||
{
|
||||
if (this != &other) m_data = other.m_data;
|
||||
return *this;
|
||||
}
|
||||
#else
|
||||
EIGEN_DEVICE_FUNC DenseStorage& operator=(const DenseStorage&) = default;
|
||||
#endif
|
||||
#if EIGEN_HAS_RVALUE_REFERENCES
|
||||
#if !EIGEN_HAS_CXX11
|
||||
EIGEN_DEVICE_FUNC DenseStorage(DenseStorage&& other) EIGEN_NOEXCEPT
|
||||
: m_data(std::move(other.m_data))
|
||||
{
|
||||
}
|
||||
EIGEN_DEVICE_FUNC DenseStorage& operator=(DenseStorage&& other) EIGEN_NOEXCEPT
|
||||
{
|
||||
if (this != &other)
|
||||
m_data = std::move(other.m_data);
|
||||
return *this;
|
||||
}
|
||||
#else
|
||||
EIGEN_DEVICE_FUNC DenseStorage(DenseStorage&&) = default;
|
||||
EIGEN_DEVICE_FUNC DenseStorage& operator=(DenseStorage&&) = default;
|
||||
#endif
|
||||
#endif
|
||||
EIGEN_DEVICE_FUNC DenseStorage(Index size, Index rows, Index cols) {
|
||||
EIGEN_INTERNAL_DENSE_STORAGE_CTOR_PLUGIN({})
|
||||
eigen_internal_assert(size==rows*cols && rows==_Rows && cols==_Cols);
|
||||
EIGEN_UNUSED_VARIABLE(size);
|
||||
EIGEN_UNUSED_VARIABLE(rows);
|
||||
EIGEN_UNUSED_VARIABLE(cols);
|
||||
}
|
||||
EIGEN_DEVICE_FUNC void swap(DenseStorage& other) {
|
||||
numext::swap(m_data, other.m_data);
|
||||
}
|
||||
EIGEN_DEVICE_FUNC static EIGEN_CONSTEXPR Index rows(void) EIGEN_NOEXCEPT {return _Rows;}
|
||||
EIGEN_DEVICE_FUNC static EIGEN_CONSTEXPR Index cols(void) EIGEN_NOEXCEPT {return _Cols;}
|
||||
EIGEN_DEVICE_FUNC void conservativeResize(Index,Index,Index) {}
|
||||
EIGEN_DEVICE_FUNC void resize(Index,Index,Index) {}
|
||||
EIGEN_DEVICE_FUNC const T *data() const { return m_data.array; }
|
||||
EIGEN_DEVICE_FUNC T *data() { return m_data.array; }
|
||||
};
|
||||
|
||||
// null matrix
|
||||
template<typename T, int _Rows, int _Cols, int _Options> class DenseStorage<T, 0, _Rows, _Cols, _Options>
|
||||
{
|
||||
public:
|
||||
EIGEN_DEVICE_FUNC DenseStorage() {}
|
||||
EIGEN_DEVICE_FUNC explicit DenseStorage(internal::constructor_without_unaligned_array_assert) {}
|
||||
EIGEN_DEVICE_FUNC DenseStorage(const DenseStorage&) {}
|
||||
EIGEN_DEVICE_FUNC DenseStorage& operator=(const DenseStorage&) { return *this; }
|
||||
EIGEN_DEVICE_FUNC DenseStorage(Index,Index,Index) {}
|
||||
EIGEN_DEVICE_FUNC void swap(DenseStorage& ) {}
|
||||
EIGEN_DEVICE_FUNC static EIGEN_CONSTEXPR Index rows(void) EIGEN_NOEXCEPT {return _Rows;}
|
||||
EIGEN_DEVICE_FUNC static EIGEN_CONSTEXPR Index cols(void) EIGEN_NOEXCEPT {return _Cols;}
|
||||
EIGEN_DEVICE_FUNC void conservativeResize(Index,Index,Index) {}
|
||||
EIGEN_DEVICE_FUNC void resize(Index,Index,Index) {}
|
||||
EIGEN_DEVICE_FUNC const T *data() const { return 0; }
|
||||
EIGEN_DEVICE_FUNC T *data() { return 0; }
|
||||
};
|
||||
|
||||
// more specializations for null matrices; these are necessary to resolve ambiguities
|
||||
template<typename T, int _Options> class DenseStorage<T, 0, Dynamic, Dynamic, _Options>
|
||||
: public DenseStorage<T, 0, 0, 0, _Options> { };
|
||||
|
||||
template<typename T, int _Rows, int _Options> class DenseStorage<T, 0, _Rows, Dynamic, _Options>
|
||||
: public DenseStorage<T, 0, 0, 0, _Options> { };
|
||||
|
||||
template<typename T, int _Cols, int _Options> class DenseStorage<T, 0, Dynamic, _Cols, _Options>
|
||||
: public DenseStorage<T, 0, 0, 0, _Options> { };
|
||||
|
||||
// dynamic-size matrix with fixed-size storage
|
||||
template<typename T, int Size, int _Options> class DenseStorage<T, Size, Dynamic, Dynamic, _Options>
|
||||
{
|
||||
internal::plain_array<T,Size,_Options> m_data;
|
||||
Index m_rows;
|
||||
Index m_cols;
|
||||
public:
|
||||
EIGEN_DEVICE_FUNC DenseStorage() : m_rows(0), m_cols(0) {}
|
||||
EIGEN_DEVICE_FUNC explicit DenseStorage(internal::constructor_without_unaligned_array_assert)
|
||||
: m_data(internal::constructor_without_unaligned_array_assert()), m_rows(0), m_cols(0) {}
|
||||
EIGEN_DEVICE_FUNC DenseStorage(const DenseStorage& other)
|
||||
: m_data(internal::constructor_without_unaligned_array_assert()), m_rows(other.m_rows), m_cols(other.m_cols)
|
||||
{
|
||||
internal::plain_array_helper::copy(other.m_data, m_rows * m_cols, m_data);
|
||||
}
|
||||
EIGEN_DEVICE_FUNC DenseStorage& operator=(const DenseStorage& other)
|
||||
{
|
||||
if (this != &other)
|
||||
{
|
||||
m_rows = other.m_rows;
|
||||
m_cols = other.m_cols;
|
||||
internal::plain_array_helper::copy(other.m_data, m_rows * m_cols, m_data);
|
||||
}
|
||||
return *this;
|
||||
}
|
||||
EIGEN_DEVICE_FUNC DenseStorage(Index, Index rows, Index cols) : m_rows(rows), m_cols(cols) {}
|
||||
EIGEN_DEVICE_FUNC void swap(DenseStorage& other)
|
||||
{
|
||||
internal::plain_array_helper::swap(m_data, m_rows * m_cols, other.m_data, other.m_rows * other.m_cols);
|
||||
numext::swap(m_rows,other.m_rows);
|
||||
numext::swap(m_cols,other.m_cols);
|
||||
}
|
||||
EIGEN_DEVICE_FUNC Index rows() const {return m_rows;}
|
||||
EIGEN_DEVICE_FUNC Index cols() const {return m_cols;}
|
||||
EIGEN_DEVICE_FUNC void conservativeResize(Index, Index rows, Index cols) { m_rows = rows; m_cols = cols; }
|
||||
EIGEN_DEVICE_FUNC void resize(Index, Index rows, Index cols) { m_rows = rows; m_cols = cols; }
|
||||
EIGEN_DEVICE_FUNC const T *data() const { return m_data.array; }
|
||||
EIGEN_DEVICE_FUNC T *data() { return m_data.array; }
|
||||
};
|
||||
|
||||
// dynamic-size matrix with fixed-size storage and fixed width
|
||||
template<typename T, int Size, int _Cols, int _Options> class DenseStorage<T, Size, Dynamic, _Cols, _Options>
|
||||
{
|
||||
internal::plain_array<T,Size,_Options> m_data;
|
||||
Index m_rows;
|
||||
public:
|
||||
EIGEN_DEVICE_FUNC DenseStorage() : m_rows(0) {}
|
||||
EIGEN_DEVICE_FUNC explicit DenseStorage(internal::constructor_without_unaligned_array_assert)
|
||||
: m_data(internal::constructor_without_unaligned_array_assert()), m_rows(0) {}
|
||||
EIGEN_DEVICE_FUNC DenseStorage(const DenseStorage& other)
|
||||
: m_data(internal::constructor_without_unaligned_array_assert()), m_rows(other.m_rows)
|
||||
{
|
||||
internal::plain_array_helper::copy(other.m_data, m_rows * _Cols, m_data);
|
||||
}
|
||||
|
||||
EIGEN_DEVICE_FUNC DenseStorage& operator=(const DenseStorage& other)
|
||||
{
|
||||
if (this != &other)
|
||||
{
|
||||
m_rows = other.m_rows;
|
||||
internal::plain_array_helper::copy(other.m_data, m_rows * _Cols, m_data);
|
||||
}
|
||||
return *this;
|
||||
}
|
||||
EIGEN_DEVICE_FUNC DenseStorage(Index, Index rows, Index) : m_rows(rows) {}
|
||||
EIGEN_DEVICE_FUNC void swap(DenseStorage& other)
|
||||
{
|
||||
internal::plain_array_helper::swap(m_data, m_rows * _Cols, other.m_data, other.m_rows * _Cols);
|
||||
numext::swap(m_rows, other.m_rows);
|
||||
}
|
||||
EIGEN_DEVICE_FUNC Index rows(void) const EIGEN_NOEXCEPT {return m_rows;}
|
||||
EIGEN_DEVICE_FUNC EIGEN_CONSTEXPR Index cols(void) const EIGEN_NOEXCEPT {return _Cols;}
|
||||
EIGEN_DEVICE_FUNC void conservativeResize(Index, Index rows, Index) { m_rows = rows; }
|
||||
EIGEN_DEVICE_FUNC void resize(Index, Index rows, Index) { m_rows = rows; }
|
||||
EIGEN_DEVICE_FUNC const T *data() const { return m_data.array; }
|
||||
EIGEN_DEVICE_FUNC T *data() { return m_data.array; }
|
||||
};
|
||||
|
||||
// dynamic-size matrix with fixed-size storage and fixed height
|
||||
template<typename T, int Size, int _Rows, int _Options> class DenseStorage<T, Size, _Rows, Dynamic, _Options>
|
||||
{
|
||||
internal::plain_array<T,Size,_Options> m_data;
|
||||
Index m_cols;
|
||||
public:
|
||||
EIGEN_DEVICE_FUNC DenseStorage() : m_cols(0) {}
|
||||
EIGEN_DEVICE_FUNC explicit DenseStorage(internal::constructor_without_unaligned_array_assert)
|
||||
: m_data(internal::constructor_without_unaligned_array_assert()), m_cols(0) {}
|
||||
EIGEN_DEVICE_FUNC DenseStorage(const DenseStorage& other)
|
||||
: m_data(internal::constructor_without_unaligned_array_assert()), m_cols(other.m_cols)
|
||||
{
|
||||
internal::plain_array_helper::copy(other.m_data, _Rows * m_cols, m_data);
|
||||
}
|
||||
EIGEN_DEVICE_FUNC DenseStorage& operator=(const DenseStorage& other)
|
||||
{
|
||||
if (this != &other)
|
||||
{
|
||||
m_cols = other.m_cols;
|
||||
internal::plain_array_helper::copy(other.m_data, _Rows * m_cols, m_data);
|
||||
}
|
||||
return *this;
|
||||
}
|
||||
EIGEN_DEVICE_FUNC DenseStorage(Index, Index, Index cols) : m_cols(cols) {}
|
||||
EIGEN_DEVICE_FUNC void swap(DenseStorage& other) {
|
||||
internal::plain_array_helper::swap(m_data, _Rows * m_cols, other.m_data, _Rows * other.m_cols);
|
||||
numext::swap(m_cols, other.m_cols);
|
||||
}
|
||||
EIGEN_DEVICE_FUNC EIGEN_CONSTEXPR Index rows(void) const EIGEN_NOEXCEPT {return _Rows;}
|
||||
EIGEN_DEVICE_FUNC Index cols(void) const EIGEN_NOEXCEPT {return m_cols;}
|
||||
EIGEN_DEVICE_FUNC void conservativeResize(Index, Index, Index cols) { m_cols = cols; }
|
||||
EIGEN_DEVICE_FUNC void resize(Index, Index, Index cols) { m_cols = cols; }
|
||||
EIGEN_DEVICE_FUNC const T *data() const { return m_data.array; }
|
||||
EIGEN_DEVICE_FUNC T *data() { return m_data.array; }
|
||||
};
|
||||
|
||||
// purely dynamic matrix.
|
||||
template<typename T, int _Options> class DenseStorage<T, Dynamic, Dynamic, Dynamic, _Options>
|
||||
{
|
||||
T *m_data;
|
||||
Index m_rows;
|
||||
Index m_cols;
|
||||
public:
|
||||
EIGEN_DEVICE_FUNC DenseStorage() : m_data(0), m_rows(0), m_cols(0) {}
|
||||
EIGEN_DEVICE_FUNC explicit DenseStorage(internal::constructor_without_unaligned_array_assert)
|
||||
: m_data(0), m_rows(0), m_cols(0) {}
|
||||
EIGEN_DEVICE_FUNC DenseStorage(Index size, Index rows, Index cols)
|
||||
: m_data(internal::conditional_aligned_new_auto<T,(_Options&DontAlign)==0>(size)), m_rows(rows), m_cols(cols)
|
||||
{
|
||||
EIGEN_INTERNAL_DENSE_STORAGE_CTOR_PLUGIN({})
|
||||
eigen_internal_assert(size==rows*cols && rows>=0 && cols >=0);
|
||||
}
|
||||
EIGEN_DEVICE_FUNC DenseStorage(const DenseStorage& other)
|
||||
: m_data(internal::conditional_aligned_new_auto<T,(_Options&DontAlign)==0>(other.m_rows*other.m_cols))
|
||||
, m_rows(other.m_rows)
|
||||
, m_cols(other.m_cols)
|
||||
{
|
||||
EIGEN_INTERNAL_DENSE_STORAGE_CTOR_PLUGIN(Index size = m_rows*m_cols)
|
||||
internal::smart_copy(other.m_data, other.m_data+other.m_rows*other.m_cols, m_data);
|
||||
}
|
||||
EIGEN_DEVICE_FUNC DenseStorage& operator=(const DenseStorage& other)
|
||||
{
|
||||
if (this != &other)
|
||||
{
|
||||
DenseStorage tmp(other);
|
||||
this->swap(tmp);
|
||||
}
|
||||
return *this;
|
||||
}
|
||||
#if EIGEN_HAS_RVALUE_REFERENCES
|
||||
EIGEN_DEVICE_FUNC
|
||||
DenseStorage(DenseStorage&& other) EIGEN_NOEXCEPT
|
||||
: m_data(std::move(other.m_data))
|
||||
, m_rows(std::move(other.m_rows))
|
||||
, m_cols(std::move(other.m_cols))
|
||||
{
|
||||
other.m_data = nullptr;
|
||||
other.m_rows = 0;
|
||||
other.m_cols = 0;
|
||||
}
|
||||
EIGEN_DEVICE_FUNC
|
||||
DenseStorage& operator=(DenseStorage&& other) EIGEN_NOEXCEPT
|
||||
{
|
||||
numext::swap(m_data, other.m_data);
|
||||
numext::swap(m_rows, other.m_rows);
|
||||
numext::swap(m_cols, other.m_cols);
|
||||
return *this;
|
||||
}
|
||||
#endif
|
||||
EIGEN_DEVICE_FUNC ~DenseStorage() { internal::conditional_aligned_delete_auto<T,(_Options&DontAlign)==0>(m_data, m_rows*m_cols); }
|
||||
EIGEN_DEVICE_FUNC void swap(DenseStorage& other)
|
||||
{
|
||||
numext::swap(m_data,other.m_data);
|
||||
numext::swap(m_rows,other.m_rows);
|
||||
numext::swap(m_cols,other.m_cols);
|
||||
}
|
||||
EIGEN_DEVICE_FUNC Index rows(void) const EIGEN_NOEXCEPT {return m_rows;}
|
||||
EIGEN_DEVICE_FUNC Index cols(void) const EIGEN_NOEXCEPT {return m_cols;}
|
||||
void conservativeResize(Index size, Index rows, Index cols)
|
||||
{
|
||||
m_data = internal::conditional_aligned_realloc_new_auto<T,(_Options&DontAlign)==0>(m_data, size, m_rows*m_cols);
|
||||
m_rows = rows;
|
||||
m_cols = cols;
|
||||
}
|
||||
EIGEN_DEVICE_FUNC void resize(Index size, Index rows, Index cols)
|
||||
{
|
||||
if(size != m_rows*m_cols)
|
||||
{
|
||||
internal::conditional_aligned_delete_auto<T,(_Options&DontAlign)==0>(m_data, m_rows*m_cols);
|
||||
if (size>0) // >0 and not simply !=0 to let the compiler knows that size cannot be negative
|
||||
m_data = internal::conditional_aligned_new_auto<T,(_Options&DontAlign)==0>(size);
|
||||
else
|
||||
m_data = 0;
|
||||
EIGEN_INTERNAL_DENSE_STORAGE_CTOR_PLUGIN({})
|
||||
}
|
||||
m_rows = rows;
|
||||
m_cols = cols;
|
||||
}
|
||||
EIGEN_DEVICE_FUNC const T *data() const { return m_data; }
|
||||
EIGEN_DEVICE_FUNC T *data() { return m_data; }
|
||||
};
|
||||
|
||||
// matrix with dynamic width and fixed height (so that matrix has dynamic size).
|
||||
template<typename T, int _Rows, int _Options> class DenseStorage<T, Dynamic, _Rows, Dynamic, _Options>
|
||||
{
|
||||
T *m_data;
|
||||
Index m_cols;
|
||||
public:
|
||||
EIGEN_DEVICE_FUNC DenseStorage() : m_data(0), m_cols(0) {}
|
||||
explicit DenseStorage(internal::constructor_without_unaligned_array_assert) : m_data(0), m_cols(0) {}
|
||||
EIGEN_DEVICE_FUNC DenseStorage(Index size, Index rows, Index cols) : m_data(internal::conditional_aligned_new_auto<T,(_Options&DontAlign)==0>(size)), m_cols(cols)
|
||||
{
|
||||
EIGEN_INTERNAL_DENSE_STORAGE_CTOR_PLUGIN({})
|
||||
eigen_internal_assert(size==rows*cols && rows==_Rows && cols >=0);
|
||||
EIGEN_UNUSED_VARIABLE(rows);
|
||||
}
|
||||
EIGEN_DEVICE_FUNC DenseStorage(const DenseStorage& other)
|
||||
: m_data(internal::conditional_aligned_new_auto<T,(_Options&DontAlign)==0>(_Rows*other.m_cols))
|
||||
, m_cols(other.m_cols)
|
||||
{
|
||||
EIGEN_INTERNAL_DENSE_STORAGE_CTOR_PLUGIN(Index size = m_cols*_Rows)
|
||||
internal::smart_copy(other.m_data, other.m_data+_Rows*m_cols, m_data);
|
||||
}
|
||||
EIGEN_DEVICE_FUNC DenseStorage& operator=(const DenseStorage& other)
|
||||
{
|
||||
if (this != &other)
|
||||
{
|
||||
DenseStorage tmp(other);
|
||||
this->swap(tmp);
|
||||
}
|
||||
return *this;
|
||||
}
|
||||
#if EIGEN_HAS_RVALUE_REFERENCES
|
||||
EIGEN_DEVICE_FUNC
|
||||
DenseStorage(DenseStorage&& other) EIGEN_NOEXCEPT
|
||||
: m_data(std::move(other.m_data))
|
||||
, m_cols(std::move(other.m_cols))
|
||||
{
|
||||
other.m_data = nullptr;
|
||||
other.m_cols = 0;
|
||||
}
|
||||
EIGEN_DEVICE_FUNC
|
||||
DenseStorage& operator=(DenseStorage&& other) EIGEN_NOEXCEPT
|
||||
{
|
||||
numext::swap(m_data, other.m_data);
|
||||
numext::swap(m_cols, other.m_cols);
|
||||
return *this;
|
||||
}
|
||||
#endif
|
||||
EIGEN_DEVICE_FUNC ~DenseStorage() { internal::conditional_aligned_delete_auto<T,(_Options&DontAlign)==0>(m_data, _Rows*m_cols); }
|
||||
EIGEN_DEVICE_FUNC void swap(DenseStorage& other) {
|
||||
numext::swap(m_data,other.m_data);
|
||||
numext::swap(m_cols,other.m_cols);
|
||||
}
|
||||
EIGEN_DEVICE_FUNC static EIGEN_CONSTEXPR Index rows(void) EIGEN_NOEXCEPT {return _Rows;}
|
||||
EIGEN_DEVICE_FUNC Index cols(void) const EIGEN_NOEXCEPT {return m_cols;}
|
||||
EIGEN_DEVICE_FUNC void conservativeResize(Index size, Index, Index cols)
|
||||
{
|
||||
m_data = internal::conditional_aligned_realloc_new_auto<T,(_Options&DontAlign)==0>(m_data, size, _Rows*m_cols);
|
||||
m_cols = cols;
|
||||
}
|
||||
EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE void resize(Index size, Index, Index cols)
|
||||
{
|
||||
if(size != _Rows*m_cols)
|
||||
{
|
||||
internal::conditional_aligned_delete_auto<T,(_Options&DontAlign)==0>(m_data, _Rows*m_cols);
|
||||
if (size>0) // >0 and not simply !=0 to let the compiler knows that size cannot be negative
|
||||
m_data = internal::conditional_aligned_new_auto<T,(_Options&DontAlign)==0>(size);
|
||||
else
|
||||
m_data = 0;
|
||||
EIGEN_INTERNAL_DENSE_STORAGE_CTOR_PLUGIN({})
|
||||
}
|
||||
m_cols = cols;
|
||||
}
|
||||
EIGEN_DEVICE_FUNC const T *data() const { return m_data; }
|
||||
EIGEN_DEVICE_FUNC T *data() { return m_data; }
|
||||
};
|
||||
|
||||
// matrix with dynamic height and fixed width (so that matrix has dynamic size).
|
||||
template<typename T, int _Cols, int _Options> class DenseStorage<T, Dynamic, Dynamic, _Cols, _Options>
|
||||
{
|
||||
T *m_data;
|
||||
Index m_rows;
|
||||
public:
|
||||
EIGEN_DEVICE_FUNC DenseStorage() : m_data(0), m_rows(0) {}
|
||||
explicit DenseStorage(internal::constructor_without_unaligned_array_assert) : m_data(0), m_rows(0) {}
|
||||
EIGEN_DEVICE_FUNC DenseStorage(Index size, Index rows, Index cols) : m_data(internal::conditional_aligned_new_auto<T,(_Options&DontAlign)==0>(size)), m_rows(rows)
|
||||
{
|
||||
EIGEN_INTERNAL_DENSE_STORAGE_CTOR_PLUGIN({})
|
||||
eigen_internal_assert(size==rows*cols && rows>=0 && cols == _Cols);
|
||||
EIGEN_UNUSED_VARIABLE(cols);
|
||||
}
|
||||
EIGEN_DEVICE_FUNC DenseStorage(const DenseStorage& other)
|
||||
: m_data(internal::conditional_aligned_new_auto<T,(_Options&DontAlign)==0>(other.m_rows*_Cols))
|
||||
, m_rows(other.m_rows)
|
||||
{
|
||||
EIGEN_INTERNAL_DENSE_STORAGE_CTOR_PLUGIN(Index size = m_rows*_Cols)
|
||||
internal::smart_copy(other.m_data, other.m_data+other.m_rows*_Cols, m_data);
|
||||
}
|
||||
EIGEN_DEVICE_FUNC DenseStorage& operator=(const DenseStorage& other)
|
||||
{
|
||||
if (this != &other)
|
||||
{
|
||||
DenseStorage tmp(other);
|
||||
this->swap(tmp);
|
||||
}
|
||||
return *this;
|
||||
}
|
||||
#if EIGEN_HAS_RVALUE_REFERENCES
|
||||
EIGEN_DEVICE_FUNC
|
||||
DenseStorage(DenseStorage&& other) EIGEN_NOEXCEPT
|
||||
: m_data(std::move(other.m_data))
|
||||
, m_rows(std::move(other.m_rows))
|
||||
{
|
||||
other.m_data = nullptr;
|
||||
other.m_rows = 0;
|
||||
}
|
||||
EIGEN_DEVICE_FUNC
|
||||
DenseStorage& operator=(DenseStorage&& other) EIGEN_NOEXCEPT
|
||||
{
|
||||
numext::swap(m_data, other.m_data);
|
||||
numext::swap(m_rows, other.m_rows);
|
||||
return *this;
|
||||
}
|
||||
#endif
|
||||
EIGEN_DEVICE_FUNC ~DenseStorage() { internal::conditional_aligned_delete_auto<T,(_Options&DontAlign)==0>(m_data, _Cols*m_rows); }
|
||||
EIGEN_DEVICE_FUNC void swap(DenseStorage& other) {
|
||||
numext::swap(m_data,other.m_data);
|
||||
numext::swap(m_rows,other.m_rows);
|
||||
}
|
||||
EIGEN_DEVICE_FUNC Index rows(void) const EIGEN_NOEXCEPT {return m_rows;}
|
||||
EIGEN_DEVICE_FUNC static EIGEN_CONSTEXPR Index cols(void) {return _Cols;}
|
||||
void conservativeResize(Index size, Index rows, Index)
|
||||
{
|
||||
m_data = internal::conditional_aligned_realloc_new_auto<T,(_Options&DontAlign)==0>(m_data, size, m_rows*_Cols);
|
||||
m_rows = rows;
|
||||
}
|
||||
EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE void resize(Index size, Index rows, Index)
|
||||
{
|
||||
if(size != m_rows*_Cols)
|
||||
{
|
||||
internal::conditional_aligned_delete_auto<T,(_Options&DontAlign)==0>(m_data, _Cols*m_rows);
|
||||
if (size>0) // >0 and not simply !=0 to let the compiler knows that size cannot be negative
|
||||
m_data = internal::conditional_aligned_new_auto<T,(_Options&DontAlign)==0>(size);
|
||||
else
|
||||
m_data = 0;
|
||||
EIGEN_INTERNAL_DENSE_STORAGE_CTOR_PLUGIN({})
|
||||
}
|
||||
m_rows = rows;
|
||||
}
|
||||
EIGEN_DEVICE_FUNC const T *data() const { return m_data; }
|
||||
EIGEN_DEVICE_FUNC T *data() { return m_data; }
|
||||
};
|
||||
|
||||
} // end namespace Eigen
|
||||
|
||||
#endif // EIGEN_MATRIX_H
|
||||
@@ -0,0 +1,258 @@
|
||||
// This file is part of Eigen, a lightweight C++ template library
|
||||
// for linear algebra.
|
||||
//
|
||||
// Copyright (C) 2007-2009 Benoit Jacob <jacob.benoit.1@gmail.com>
|
||||
// Copyright (C) 2009-2010 Gael Guennebaud <gael.guennebaud@inria.fr>
|
||||
//
|
||||
// This Source Code Form is subject to the terms of the Mozilla
|
||||
// Public License v. 2.0. If a copy of the MPL was not distributed
|
||||
// with this file, You can obtain one at http://mozilla.org/MPL/2.0/.
|
||||
|
||||
#ifndef EIGEN_DIAGONAL_H
|
||||
#define EIGEN_DIAGONAL_H
|
||||
|
||||
namespace Eigen {
|
||||
|
||||
/** \class Diagonal
|
||||
* \ingroup Core_Module
|
||||
*
|
||||
* \brief Expression of a diagonal/subdiagonal/superdiagonal in a matrix
|
||||
*
|
||||
* \param MatrixType the type of the object in which we are taking a sub/main/super diagonal
|
||||
* \param DiagIndex the index of the sub/super diagonal. The default is 0 and it means the main diagonal.
|
||||
* A positive value means a superdiagonal, a negative value means a subdiagonal.
|
||||
* You can also use DynamicIndex so the index can be set at runtime.
|
||||
*
|
||||
* The matrix is not required to be square.
|
||||
*
|
||||
* This class represents an expression of the main diagonal, or any sub/super diagonal
|
||||
* of a square matrix. It is the return type of MatrixBase::diagonal() and MatrixBase::diagonal(Index) and most of the
|
||||
* time this is the only way it is used.
|
||||
*
|
||||
* \sa MatrixBase::diagonal(), MatrixBase::diagonal(Index)
|
||||
*/
|
||||
|
||||
namespace internal {
|
||||
template<typename MatrixType, int DiagIndex>
|
||||
struct traits<Diagonal<MatrixType,DiagIndex> >
|
||||
: traits<MatrixType>
|
||||
{
|
||||
typedef typename ref_selector<MatrixType>::type MatrixTypeNested;
|
||||
typedef typename remove_reference<MatrixTypeNested>::type _MatrixTypeNested;
|
||||
typedef typename MatrixType::StorageKind StorageKind;
|
||||
enum {
|
||||
RowsAtCompileTime = (int(DiagIndex) == DynamicIndex || int(MatrixType::SizeAtCompileTime) == Dynamic) ? Dynamic
|
||||
: (EIGEN_PLAIN_ENUM_MIN(MatrixType::RowsAtCompileTime - EIGEN_PLAIN_ENUM_MAX(-DiagIndex, 0),
|
||||
MatrixType::ColsAtCompileTime - EIGEN_PLAIN_ENUM_MAX( DiagIndex, 0))),
|
||||
ColsAtCompileTime = 1,
|
||||
MaxRowsAtCompileTime = int(MatrixType::MaxSizeAtCompileTime) == Dynamic ? Dynamic
|
||||
: DiagIndex == DynamicIndex ? EIGEN_SIZE_MIN_PREFER_FIXED(MatrixType::MaxRowsAtCompileTime,
|
||||
MatrixType::MaxColsAtCompileTime)
|
||||
: (EIGEN_PLAIN_ENUM_MIN(MatrixType::MaxRowsAtCompileTime - EIGEN_PLAIN_ENUM_MAX(-DiagIndex, 0),
|
||||
MatrixType::MaxColsAtCompileTime - EIGEN_PLAIN_ENUM_MAX( DiagIndex, 0))),
|
||||
MaxColsAtCompileTime = 1,
|
||||
MaskLvalueBit = is_lvalue<MatrixType>::value ? LvalueBit : 0,
|
||||
Flags = (unsigned int)_MatrixTypeNested::Flags & (RowMajorBit | MaskLvalueBit | DirectAccessBit) & ~RowMajorBit, // FIXME DirectAccessBit should not be handled by expressions
|
||||
MatrixTypeOuterStride = outer_stride_at_compile_time<MatrixType>::ret,
|
||||
InnerStrideAtCompileTime = MatrixTypeOuterStride == Dynamic ? Dynamic : MatrixTypeOuterStride+1,
|
||||
OuterStrideAtCompileTime = 0
|
||||
};
|
||||
};
|
||||
}
|
||||
|
||||
template<typename MatrixType, int _DiagIndex> class Diagonal
|
||||
: public internal::dense_xpr_base< Diagonal<MatrixType,_DiagIndex> >::type
|
||||
{
|
||||
public:
|
||||
|
||||
enum { DiagIndex = _DiagIndex };
|
||||
typedef typename internal::dense_xpr_base<Diagonal>::type Base;
|
||||
EIGEN_DENSE_PUBLIC_INTERFACE(Diagonal)
|
||||
|
||||
EIGEN_DEVICE_FUNC
|
||||
explicit inline Diagonal(MatrixType& matrix, Index a_index = DiagIndex) : m_matrix(matrix), m_index(a_index)
|
||||
{
|
||||
eigen_assert( a_index <= m_matrix.cols() && -a_index <= m_matrix.rows() );
|
||||
}
|
||||
|
||||
EIGEN_INHERIT_ASSIGNMENT_OPERATORS(Diagonal)
|
||||
|
||||
EIGEN_DEVICE_FUNC
|
||||
inline Index rows() const
|
||||
{
|
||||
return m_index.value()<0 ? numext::mini<Index>(m_matrix.cols(),m_matrix.rows()+m_index.value())
|
||||
: numext::mini<Index>(m_matrix.rows(),m_matrix.cols()-m_index.value());
|
||||
}
|
||||
|
||||
EIGEN_DEVICE_FUNC EIGEN_CONSTEXPR
|
||||
inline Index cols() const EIGEN_NOEXCEPT { return 1; }
|
||||
|
||||
EIGEN_DEVICE_FUNC EIGEN_CONSTEXPR
|
||||
inline Index innerStride() const EIGEN_NOEXCEPT {
|
||||
return m_matrix.outerStride() + 1;
|
||||
}
|
||||
|
||||
EIGEN_DEVICE_FUNC EIGEN_CONSTEXPR
|
||||
inline Index outerStride() const EIGEN_NOEXCEPT { return 0; }
|
||||
|
||||
typedef typename internal::conditional<
|
||||
internal::is_lvalue<MatrixType>::value,
|
||||
Scalar,
|
||||
const Scalar
|
||||
>::type ScalarWithConstIfNotLvalue;
|
||||
|
||||
EIGEN_DEVICE_FUNC
|
||||
inline ScalarWithConstIfNotLvalue* data() { return &(m_matrix.coeffRef(rowOffset(), colOffset())); }
|
||||
EIGEN_DEVICE_FUNC
|
||||
inline const Scalar* data() const { return &(m_matrix.coeffRef(rowOffset(), colOffset())); }
|
||||
|
||||
EIGEN_DEVICE_FUNC
|
||||
inline Scalar& coeffRef(Index row, Index)
|
||||
{
|
||||
EIGEN_STATIC_ASSERT_LVALUE(MatrixType)
|
||||
return m_matrix.coeffRef(row+rowOffset(), row+colOffset());
|
||||
}
|
||||
|
||||
EIGEN_DEVICE_FUNC
|
||||
inline const Scalar& coeffRef(Index row, Index) const
|
||||
{
|
||||
return m_matrix.coeffRef(row+rowOffset(), row+colOffset());
|
||||
}
|
||||
|
||||
EIGEN_DEVICE_FUNC
|
||||
inline CoeffReturnType coeff(Index row, Index) const
|
||||
{
|
||||
return m_matrix.coeff(row+rowOffset(), row+colOffset());
|
||||
}
|
||||
|
||||
EIGEN_DEVICE_FUNC
|
||||
inline Scalar& coeffRef(Index idx)
|
||||
{
|
||||
EIGEN_STATIC_ASSERT_LVALUE(MatrixType)
|
||||
return m_matrix.coeffRef(idx+rowOffset(), idx+colOffset());
|
||||
}
|
||||
|
||||
EIGEN_DEVICE_FUNC
|
||||
inline const Scalar& coeffRef(Index idx) const
|
||||
{
|
||||
return m_matrix.coeffRef(idx+rowOffset(), idx+colOffset());
|
||||
}
|
||||
|
||||
EIGEN_DEVICE_FUNC
|
||||
inline CoeffReturnType coeff(Index idx) const
|
||||
{
|
||||
return m_matrix.coeff(idx+rowOffset(), idx+colOffset());
|
||||
}
|
||||
|
||||
EIGEN_DEVICE_FUNC
|
||||
inline const typename internal::remove_all<typename MatrixType::Nested>::type&
|
||||
nestedExpression() const
|
||||
{
|
||||
return m_matrix;
|
||||
}
|
||||
|
||||
EIGEN_DEVICE_FUNC
|
||||
inline Index index() const
|
||||
{
|
||||
return m_index.value();
|
||||
}
|
||||
|
||||
protected:
|
||||
typename internal::ref_selector<MatrixType>::non_const_type m_matrix;
|
||||
const internal::variable_if_dynamicindex<Index, DiagIndex> m_index;
|
||||
|
||||
private:
|
||||
// some compilers may fail to optimize std::max etc in case of compile-time constants...
|
||||
EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE EIGEN_CONSTEXPR
|
||||
Index absDiagIndex() const EIGEN_NOEXCEPT { return m_index.value()>0 ? m_index.value() : -m_index.value(); }
|
||||
EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE EIGEN_CONSTEXPR
|
||||
Index rowOffset() const EIGEN_NOEXCEPT { return m_index.value()>0 ? 0 : -m_index.value(); }
|
||||
EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE EIGEN_CONSTEXPR
|
||||
Index colOffset() const EIGEN_NOEXCEPT { return m_index.value()>0 ? m_index.value() : 0; }
|
||||
// trigger a compile-time error if someone try to call packet
|
||||
template<int LoadMode> typename MatrixType::PacketReturnType packet(Index) const;
|
||||
template<int LoadMode> typename MatrixType::PacketReturnType packet(Index,Index) const;
|
||||
};
|
||||
|
||||
/** \returns an expression of the main diagonal of the matrix \c *this
|
||||
*
|
||||
* \c *this is not required to be square.
|
||||
*
|
||||
* Example: \include MatrixBase_diagonal.cpp
|
||||
* Output: \verbinclude MatrixBase_diagonal.out
|
||||
*
|
||||
* \sa class Diagonal */
|
||||
template<typename Derived>
|
||||
EIGEN_DEVICE_FUNC inline typename MatrixBase<Derived>::DiagonalReturnType
|
||||
MatrixBase<Derived>::diagonal()
|
||||
{
|
||||
return DiagonalReturnType(derived());
|
||||
}
|
||||
|
||||
/** This is the const version of diagonal(). */
|
||||
template<typename Derived>
|
||||
EIGEN_DEVICE_FUNC inline typename MatrixBase<Derived>::ConstDiagonalReturnType
|
||||
MatrixBase<Derived>::diagonal() const
|
||||
{
|
||||
return ConstDiagonalReturnType(derived());
|
||||
}
|
||||
|
||||
/** \returns an expression of the \a DiagIndex-th sub or super diagonal of the matrix \c *this
|
||||
*
|
||||
* \c *this is not required to be square.
|
||||
*
|
||||
* The template parameter \a DiagIndex represent a super diagonal if \a DiagIndex > 0
|
||||
* and a sub diagonal otherwise. \a DiagIndex == 0 is equivalent to the main diagonal.
|
||||
*
|
||||
* Example: \include MatrixBase_diagonal_int.cpp
|
||||
* Output: \verbinclude MatrixBase_diagonal_int.out
|
||||
*
|
||||
* \sa MatrixBase::diagonal(), class Diagonal */
|
||||
template<typename Derived>
|
||||
EIGEN_DEVICE_FUNC inline typename MatrixBase<Derived>::DiagonalDynamicIndexReturnType
|
||||
MatrixBase<Derived>::diagonal(Index index)
|
||||
{
|
||||
return DiagonalDynamicIndexReturnType(derived(), index);
|
||||
}
|
||||
|
||||
/** This is the const version of diagonal(Index). */
|
||||
template<typename Derived>
|
||||
EIGEN_DEVICE_FUNC inline typename MatrixBase<Derived>::ConstDiagonalDynamicIndexReturnType
|
||||
MatrixBase<Derived>::diagonal(Index index) const
|
||||
{
|
||||
return ConstDiagonalDynamicIndexReturnType(derived(), index);
|
||||
}
|
||||
|
||||
/** \returns an expression of the \a DiagIndex-th sub or super diagonal of the matrix \c *this
|
||||
*
|
||||
* \c *this is not required to be square.
|
||||
*
|
||||
* The template parameter \a DiagIndex represent a super diagonal if \a DiagIndex > 0
|
||||
* and a sub diagonal otherwise. \a DiagIndex == 0 is equivalent to the main diagonal.
|
||||
*
|
||||
* Example: \include MatrixBase_diagonal_template_int.cpp
|
||||
* Output: \verbinclude MatrixBase_diagonal_template_int.out
|
||||
*
|
||||
* \sa MatrixBase::diagonal(), class Diagonal */
|
||||
template<typename Derived>
|
||||
template<int Index_>
|
||||
EIGEN_DEVICE_FUNC
|
||||
inline typename MatrixBase<Derived>::template DiagonalIndexReturnType<Index_>::Type
|
||||
MatrixBase<Derived>::diagonal()
|
||||
{
|
||||
return typename DiagonalIndexReturnType<Index_>::Type(derived());
|
||||
}
|
||||
|
||||
/** This is the const version of diagonal<int>(). */
|
||||
template<typename Derived>
|
||||
template<int Index_>
|
||||
EIGEN_DEVICE_FUNC
|
||||
inline typename MatrixBase<Derived>::template ConstDiagonalIndexReturnType<Index_>::Type
|
||||
MatrixBase<Derived>::diagonal() const
|
||||
{
|
||||
return typename ConstDiagonalIndexReturnType<Index_>::Type(derived());
|
||||
}
|
||||
|
||||
} // end namespace Eigen
|
||||
|
||||
#endif // EIGEN_DIAGONAL_H
|
||||
@@ -0,0 +1,391 @@
|
||||
// This file is part of Eigen, a lightweight C++ template library
|
||||
// for linear algebra.
|
||||
//
|
||||
// Copyright (C) 2009 Gael Guennebaud <gael.guennebaud@inria.fr>
|
||||
// Copyright (C) 2007-2009 Benoit Jacob <jacob.benoit.1@gmail.com>
|
||||
//
|
||||
// This Source Code Form is subject to the terms of the Mozilla
|
||||
// Public License v. 2.0. If a copy of the MPL was not distributed
|
||||
// with this file, You can obtain one at http://mozilla.org/MPL/2.0/.
|
||||
|
||||
#ifndef EIGEN_DIAGONALMATRIX_H
|
||||
#define EIGEN_DIAGONALMATRIX_H
|
||||
|
||||
namespace Eigen {
|
||||
|
||||
#ifndef EIGEN_PARSED_BY_DOXYGEN
|
||||
template<typename Derived>
|
||||
class DiagonalBase : public EigenBase<Derived>
|
||||
{
|
||||
public:
|
||||
typedef typename internal::traits<Derived>::DiagonalVectorType DiagonalVectorType;
|
||||
typedef typename DiagonalVectorType::Scalar Scalar;
|
||||
typedef typename DiagonalVectorType::RealScalar RealScalar;
|
||||
typedef typename internal::traits<Derived>::StorageKind StorageKind;
|
||||
typedef typename internal::traits<Derived>::StorageIndex StorageIndex;
|
||||
|
||||
enum {
|
||||
RowsAtCompileTime = DiagonalVectorType::SizeAtCompileTime,
|
||||
ColsAtCompileTime = DiagonalVectorType::SizeAtCompileTime,
|
||||
MaxRowsAtCompileTime = DiagonalVectorType::MaxSizeAtCompileTime,
|
||||
MaxColsAtCompileTime = DiagonalVectorType::MaxSizeAtCompileTime,
|
||||
IsVectorAtCompileTime = 0,
|
||||
Flags = NoPreferredStorageOrderBit
|
||||
};
|
||||
|
||||
typedef Matrix<Scalar, RowsAtCompileTime, ColsAtCompileTime, 0, MaxRowsAtCompileTime, MaxColsAtCompileTime> DenseMatrixType;
|
||||
typedef DenseMatrixType DenseType;
|
||||
typedef DiagonalMatrix<Scalar,DiagonalVectorType::SizeAtCompileTime,DiagonalVectorType::MaxSizeAtCompileTime> PlainObject;
|
||||
|
||||
EIGEN_DEVICE_FUNC
|
||||
inline const Derived& derived() const { return *static_cast<const Derived*>(this); }
|
||||
EIGEN_DEVICE_FUNC
|
||||
inline Derived& derived() { return *static_cast<Derived*>(this); }
|
||||
|
||||
EIGEN_DEVICE_FUNC
|
||||
DenseMatrixType toDenseMatrix() const { return derived(); }
|
||||
|
||||
EIGEN_DEVICE_FUNC
|
||||
inline const DiagonalVectorType& diagonal() const { return derived().diagonal(); }
|
||||
EIGEN_DEVICE_FUNC
|
||||
inline DiagonalVectorType& diagonal() { return derived().diagonal(); }
|
||||
|
||||
EIGEN_DEVICE_FUNC
|
||||
inline Index rows() const { return diagonal().size(); }
|
||||
EIGEN_DEVICE_FUNC
|
||||
inline Index cols() const { return diagonal().size(); }
|
||||
|
||||
template<typename MatrixDerived>
|
||||
EIGEN_DEVICE_FUNC
|
||||
const Product<Derived,MatrixDerived,LazyProduct>
|
||||
operator*(const MatrixBase<MatrixDerived> &matrix) const
|
||||
{
|
||||
return Product<Derived, MatrixDerived, LazyProduct>(derived(),matrix.derived());
|
||||
}
|
||||
|
||||
typedef DiagonalWrapper<const CwiseUnaryOp<internal::scalar_inverse_op<Scalar>, const DiagonalVectorType> > InverseReturnType;
|
||||
EIGEN_DEVICE_FUNC
|
||||
inline const InverseReturnType
|
||||
inverse() const
|
||||
{
|
||||
return InverseReturnType(diagonal().cwiseInverse());
|
||||
}
|
||||
|
||||
EIGEN_DEVICE_FUNC
|
||||
inline const DiagonalWrapper<const EIGEN_EXPR_BINARYOP_SCALAR_RETURN_TYPE(DiagonalVectorType,Scalar,product) >
|
||||
operator*(const Scalar& scalar) const
|
||||
{
|
||||
return DiagonalWrapper<const EIGEN_EXPR_BINARYOP_SCALAR_RETURN_TYPE(DiagonalVectorType,Scalar,product) >(diagonal() * scalar);
|
||||
}
|
||||
EIGEN_DEVICE_FUNC
|
||||
friend inline const DiagonalWrapper<const EIGEN_SCALAR_BINARYOP_EXPR_RETURN_TYPE(Scalar,DiagonalVectorType,product) >
|
||||
operator*(const Scalar& scalar, const DiagonalBase& other)
|
||||
{
|
||||
return DiagonalWrapper<const EIGEN_SCALAR_BINARYOP_EXPR_RETURN_TYPE(Scalar,DiagonalVectorType,product) >(scalar * other.diagonal());
|
||||
}
|
||||
|
||||
template<typename OtherDerived>
|
||||
EIGEN_DEVICE_FUNC
|
||||
#ifdef EIGEN_PARSED_BY_DOXYGEN
|
||||
inline unspecified_expression_type
|
||||
#else
|
||||
inline const DiagonalWrapper<const EIGEN_CWISE_BINARY_RETURN_TYPE(DiagonalVectorType,typename OtherDerived::DiagonalVectorType,sum) >
|
||||
#endif
|
||||
operator+(const DiagonalBase<OtherDerived>& other) const
|
||||
{
|
||||
return (diagonal() + other.diagonal()).asDiagonal();
|
||||
}
|
||||
|
||||
template<typename OtherDerived>
|
||||
EIGEN_DEVICE_FUNC
|
||||
#ifdef EIGEN_PARSED_BY_DOXYGEN
|
||||
inline unspecified_expression_type
|
||||
#else
|
||||
inline const DiagonalWrapper<const EIGEN_CWISE_BINARY_RETURN_TYPE(DiagonalVectorType,typename OtherDerived::DiagonalVectorType,difference) >
|
||||
#endif
|
||||
operator-(const DiagonalBase<OtherDerived>& other) const
|
||||
{
|
||||
return (diagonal() - other.diagonal()).asDiagonal();
|
||||
}
|
||||
};
|
||||
|
||||
#endif
|
||||
|
||||
/** \class DiagonalMatrix
|
||||
* \ingroup Core_Module
|
||||
*
|
||||
* \brief Represents a diagonal matrix with its storage
|
||||
*
|
||||
* \param _Scalar the type of coefficients
|
||||
* \param SizeAtCompileTime the dimension of the matrix, or Dynamic
|
||||
* \param MaxSizeAtCompileTime the dimension of the matrix, or Dynamic. This parameter is optional and defaults
|
||||
* to SizeAtCompileTime. Most of the time, you do not need to specify it.
|
||||
*
|
||||
* \sa class DiagonalWrapper
|
||||
*/
|
||||
|
||||
namespace internal {
|
||||
template<typename _Scalar, int SizeAtCompileTime, int MaxSizeAtCompileTime>
|
||||
struct traits<DiagonalMatrix<_Scalar,SizeAtCompileTime,MaxSizeAtCompileTime> >
|
||||
: traits<Matrix<_Scalar,SizeAtCompileTime,SizeAtCompileTime,0,MaxSizeAtCompileTime,MaxSizeAtCompileTime> >
|
||||
{
|
||||
typedef Matrix<_Scalar,SizeAtCompileTime,1,0,MaxSizeAtCompileTime,1> DiagonalVectorType;
|
||||
typedef DiagonalShape StorageKind;
|
||||
enum {
|
||||
Flags = LvalueBit | NoPreferredStorageOrderBit
|
||||
};
|
||||
};
|
||||
}
|
||||
template<typename _Scalar, int SizeAtCompileTime, int MaxSizeAtCompileTime>
|
||||
class DiagonalMatrix
|
||||
: public DiagonalBase<DiagonalMatrix<_Scalar,SizeAtCompileTime,MaxSizeAtCompileTime> >
|
||||
{
|
||||
public:
|
||||
#ifndef EIGEN_PARSED_BY_DOXYGEN
|
||||
typedef typename internal::traits<DiagonalMatrix>::DiagonalVectorType DiagonalVectorType;
|
||||
typedef const DiagonalMatrix& Nested;
|
||||
typedef _Scalar Scalar;
|
||||
typedef typename internal::traits<DiagonalMatrix>::StorageKind StorageKind;
|
||||
typedef typename internal::traits<DiagonalMatrix>::StorageIndex StorageIndex;
|
||||
#endif
|
||||
|
||||
protected:
|
||||
|
||||
DiagonalVectorType m_diagonal;
|
||||
|
||||
public:
|
||||
|
||||
/** const version of diagonal(). */
|
||||
EIGEN_DEVICE_FUNC
|
||||
inline const DiagonalVectorType& diagonal() const { return m_diagonal; }
|
||||
/** \returns a reference to the stored vector of diagonal coefficients. */
|
||||
EIGEN_DEVICE_FUNC
|
||||
inline DiagonalVectorType& diagonal() { return m_diagonal; }
|
||||
|
||||
/** Default constructor without initialization */
|
||||
EIGEN_DEVICE_FUNC
|
||||
inline DiagonalMatrix() {}
|
||||
|
||||
/** Constructs a diagonal matrix with given dimension */
|
||||
EIGEN_DEVICE_FUNC
|
||||
explicit inline DiagonalMatrix(Index dim) : m_diagonal(dim) {}
|
||||
|
||||
/** 2D constructor. */
|
||||
EIGEN_DEVICE_FUNC
|
||||
inline DiagonalMatrix(const Scalar& x, const Scalar& y) : m_diagonal(x,y) {}
|
||||
|
||||
/** 3D constructor. */
|
||||
EIGEN_DEVICE_FUNC
|
||||
inline DiagonalMatrix(const Scalar& x, const Scalar& y, const Scalar& z) : m_diagonal(x,y,z) {}
|
||||
|
||||
#if EIGEN_HAS_CXX11
|
||||
/** \brief Construct a diagonal matrix with fixed size from an arbitrary number of coefficients. \cpp11
|
||||
*
|
||||
* There exists C++98 anologue constructors for fixed-size diagonal matrices having 2 or 3 coefficients.
|
||||
*
|
||||
* \warning To construct a diagonal matrix of fixed size, the number of values passed to this
|
||||
* constructor must match the fixed dimension of \c *this.
|
||||
*
|
||||
* \sa DiagonalMatrix(const Scalar&, const Scalar&)
|
||||
* \sa DiagonalMatrix(const Scalar&, const Scalar&, const Scalar&)
|
||||
*/
|
||||
template <typename... ArgTypes>
|
||||
EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
|
||||
DiagonalMatrix(const Scalar& a0, const Scalar& a1, const Scalar& a2, const ArgTypes&... args)
|
||||
: m_diagonal(a0, a1, a2, args...) {}
|
||||
|
||||
/** \brief Constructs a DiagonalMatrix and initializes it by elements given by an initializer list of initializer
|
||||
* lists \cpp11
|
||||
*/
|
||||
EIGEN_DEVICE_FUNC
|
||||
explicit EIGEN_STRONG_INLINE DiagonalMatrix(const std::initializer_list<std::initializer_list<Scalar>>& list)
|
||||
: m_diagonal(list) {}
|
||||
#endif // EIGEN_HAS_CXX11
|
||||
|
||||
/** Copy constructor. */
|
||||
template<typename OtherDerived>
|
||||
EIGEN_DEVICE_FUNC
|
||||
inline DiagonalMatrix(const DiagonalBase<OtherDerived>& other) : m_diagonal(other.diagonal()) {}
|
||||
|
||||
#ifndef EIGEN_PARSED_BY_DOXYGEN
|
||||
/** copy constructor. prevent a default copy constructor from hiding the other templated constructor */
|
||||
inline DiagonalMatrix(const DiagonalMatrix& other) : m_diagonal(other.diagonal()) {}
|
||||
#endif
|
||||
|
||||
/** generic constructor from expression of the diagonal coefficients */
|
||||
template<typename OtherDerived>
|
||||
EIGEN_DEVICE_FUNC
|
||||
explicit inline DiagonalMatrix(const MatrixBase<OtherDerived>& other) : m_diagonal(other)
|
||||
{}
|
||||
|
||||
/** Copy operator. */
|
||||
template<typename OtherDerived>
|
||||
EIGEN_DEVICE_FUNC
|
||||
DiagonalMatrix& operator=(const DiagonalBase<OtherDerived>& other)
|
||||
{
|
||||
m_diagonal = other.diagonal();
|
||||
return *this;
|
||||
}
|
||||
|
||||
#ifndef EIGEN_PARSED_BY_DOXYGEN
|
||||
/** This is a special case of the templated operator=. Its purpose is to
|
||||
* prevent a default operator= from hiding the templated operator=.
|
||||
*/
|
||||
EIGEN_DEVICE_FUNC
|
||||
DiagonalMatrix& operator=(const DiagonalMatrix& other)
|
||||
{
|
||||
m_diagonal = other.diagonal();
|
||||
return *this;
|
||||
}
|
||||
#endif
|
||||
|
||||
/** Resizes to given size. */
|
||||
EIGEN_DEVICE_FUNC
|
||||
inline void resize(Index size) { m_diagonal.resize(size); }
|
||||
/** Sets all coefficients to zero. */
|
||||
EIGEN_DEVICE_FUNC
|
||||
inline void setZero() { m_diagonal.setZero(); }
|
||||
/** Resizes and sets all coefficients to zero. */
|
||||
EIGEN_DEVICE_FUNC
|
||||
inline void setZero(Index size) { m_diagonal.setZero(size); }
|
||||
/** Sets this matrix to be the identity matrix of the current size. */
|
||||
EIGEN_DEVICE_FUNC
|
||||
inline void setIdentity() { m_diagonal.setOnes(); }
|
||||
/** Sets this matrix to be the identity matrix of the given size. */
|
||||
EIGEN_DEVICE_FUNC
|
||||
inline void setIdentity(Index size) { m_diagonal.setOnes(size); }
|
||||
};
|
||||
|
||||
/** \class DiagonalWrapper
|
||||
* \ingroup Core_Module
|
||||
*
|
||||
* \brief Expression of a diagonal matrix
|
||||
*
|
||||
* \param _DiagonalVectorType the type of the vector of diagonal coefficients
|
||||
*
|
||||
* This class is an expression of a diagonal matrix, but not storing its own vector of diagonal coefficients,
|
||||
* instead wrapping an existing vector expression. It is the return type of MatrixBase::asDiagonal()
|
||||
* and most of the time this is the only way that it is used.
|
||||
*
|
||||
* \sa class DiagonalMatrix, class DiagonalBase, MatrixBase::asDiagonal()
|
||||
*/
|
||||
|
||||
namespace internal {
|
||||
template<typename _DiagonalVectorType>
|
||||
struct traits<DiagonalWrapper<_DiagonalVectorType> >
|
||||
{
|
||||
typedef _DiagonalVectorType DiagonalVectorType;
|
||||
typedef typename DiagonalVectorType::Scalar Scalar;
|
||||
typedef typename DiagonalVectorType::StorageIndex StorageIndex;
|
||||
typedef DiagonalShape StorageKind;
|
||||
typedef typename traits<DiagonalVectorType>::XprKind XprKind;
|
||||
enum {
|
||||
RowsAtCompileTime = DiagonalVectorType::SizeAtCompileTime,
|
||||
ColsAtCompileTime = DiagonalVectorType::SizeAtCompileTime,
|
||||
MaxRowsAtCompileTime = DiagonalVectorType::MaxSizeAtCompileTime,
|
||||
MaxColsAtCompileTime = DiagonalVectorType::MaxSizeAtCompileTime,
|
||||
Flags = (traits<DiagonalVectorType>::Flags & LvalueBit) | NoPreferredStorageOrderBit
|
||||
};
|
||||
};
|
||||
}
|
||||
|
||||
template<typename _DiagonalVectorType>
|
||||
class DiagonalWrapper
|
||||
: public DiagonalBase<DiagonalWrapper<_DiagonalVectorType> >, internal::no_assignment_operator
|
||||
{
|
||||
public:
|
||||
#ifndef EIGEN_PARSED_BY_DOXYGEN
|
||||
typedef _DiagonalVectorType DiagonalVectorType;
|
||||
typedef DiagonalWrapper Nested;
|
||||
#endif
|
||||
|
||||
/** Constructor from expression of diagonal coefficients to wrap. */
|
||||
EIGEN_DEVICE_FUNC
|
||||
explicit inline DiagonalWrapper(DiagonalVectorType& a_diagonal) : m_diagonal(a_diagonal) {}
|
||||
|
||||
/** \returns a const reference to the wrapped expression of diagonal coefficients. */
|
||||
EIGEN_DEVICE_FUNC
|
||||
const DiagonalVectorType& diagonal() const { return m_diagonal; }
|
||||
|
||||
protected:
|
||||
typename DiagonalVectorType::Nested m_diagonal;
|
||||
};
|
||||
|
||||
/** \returns a pseudo-expression of a diagonal matrix with *this as vector of diagonal coefficients
|
||||
*
|
||||
* \only_for_vectors
|
||||
*
|
||||
* Example: \include MatrixBase_asDiagonal.cpp
|
||||
* Output: \verbinclude MatrixBase_asDiagonal.out
|
||||
*
|
||||
* \sa class DiagonalWrapper, class DiagonalMatrix, diagonal(), isDiagonal()
|
||||
**/
|
||||
template<typename Derived>
|
||||
EIGEN_DEVICE_FUNC inline const DiagonalWrapper<const Derived>
|
||||
MatrixBase<Derived>::asDiagonal() const
|
||||
{
|
||||
return DiagonalWrapper<const Derived>(derived());
|
||||
}
|
||||
|
||||
/** \returns true if *this is approximately equal to a diagonal matrix,
|
||||
* within the precision given by \a prec.
|
||||
*
|
||||
* Example: \include MatrixBase_isDiagonal.cpp
|
||||
* Output: \verbinclude MatrixBase_isDiagonal.out
|
||||
*
|
||||
* \sa asDiagonal()
|
||||
*/
|
||||
template<typename Derived>
|
||||
bool MatrixBase<Derived>::isDiagonal(const RealScalar& prec) const
|
||||
{
|
||||
if(cols() != rows()) return false;
|
||||
RealScalar maxAbsOnDiagonal = static_cast<RealScalar>(-1);
|
||||
for(Index j = 0; j < cols(); ++j)
|
||||
{
|
||||
RealScalar absOnDiagonal = numext::abs(coeff(j,j));
|
||||
if(absOnDiagonal > maxAbsOnDiagonal) maxAbsOnDiagonal = absOnDiagonal;
|
||||
}
|
||||
for(Index j = 0; j < cols(); ++j)
|
||||
for(Index i = 0; i < j; ++i)
|
||||
{
|
||||
if(!internal::isMuchSmallerThan(coeff(i, j), maxAbsOnDiagonal, prec)) return false;
|
||||
if(!internal::isMuchSmallerThan(coeff(j, i), maxAbsOnDiagonal, prec)) return false;
|
||||
}
|
||||
return true;
|
||||
}
|
||||
|
||||
namespace internal {
|
||||
|
||||
template<> struct storage_kind_to_shape<DiagonalShape> { typedef DiagonalShape Shape; };
|
||||
|
||||
struct Diagonal2Dense {};
|
||||
|
||||
template<> struct AssignmentKind<DenseShape,DiagonalShape> { typedef Diagonal2Dense Kind; };
|
||||
|
||||
// Diagonal matrix to Dense assignment
|
||||
template< typename DstXprType, typename SrcXprType, typename Functor>
|
||||
struct Assignment<DstXprType, SrcXprType, Functor, Diagonal2Dense>
|
||||
{
|
||||
static void run(DstXprType &dst, const SrcXprType &src, const internal::assign_op<typename DstXprType::Scalar,typename SrcXprType::Scalar> &/*func*/)
|
||||
{
|
||||
Index dstRows = src.rows();
|
||||
Index dstCols = src.cols();
|
||||
if((dst.rows()!=dstRows) || (dst.cols()!=dstCols))
|
||||
dst.resize(dstRows, dstCols);
|
||||
|
||||
dst.setZero();
|
||||
dst.diagonal() = src.diagonal();
|
||||
}
|
||||
|
||||
static void run(DstXprType &dst, const SrcXprType &src, const internal::add_assign_op<typename DstXprType::Scalar,typename SrcXprType::Scalar> &/*func*/)
|
||||
{ dst.diagonal() += src.diagonal(); }
|
||||
|
||||
static void run(DstXprType &dst, const SrcXprType &src, const internal::sub_assign_op<typename DstXprType::Scalar,typename SrcXprType::Scalar> &/*func*/)
|
||||
{ dst.diagonal() -= src.diagonal(); }
|
||||
};
|
||||
|
||||
} // namespace internal
|
||||
|
||||
} // end namespace Eigen
|
||||
|
||||
#endif // EIGEN_DIAGONALMATRIX_H
|
||||
@@ -0,0 +1,28 @@
|
||||
// This file is part of Eigen, a lightweight C++ template library
|
||||
// for linear algebra.
|
||||
//
|
||||
// Copyright (C) 2008 Gael Guennebaud <gael.guennebaud@inria.fr>
|
||||
// Copyright (C) 2007-2009 Benoit Jacob <jacob.benoit.1@gmail.com>
|
||||
//
|
||||
// This Source Code Form is subject to the terms of the Mozilla
|
||||
// Public License v. 2.0. If a copy of the MPL was not distributed
|
||||
// with this file, You can obtain one at http://mozilla.org/MPL/2.0/.
|
||||
|
||||
#ifndef EIGEN_DIAGONALPRODUCT_H
|
||||
#define EIGEN_DIAGONALPRODUCT_H
|
||||
|
||||
namespace Eigen {
|
||||
|
||||
/** \returns the diagonal matrix product of \c *this by the diagonal matrix \a diagonal.
|
||||
*/
|
||||
template<typename Derived>
|
||||
template<typename DiagonalDerived>
|
||||
EIGEN_DEVICE_FUNC inline const Product<Derived, DiagonalDerived, LazyProduct>
|
||||
MatrixBase<Derived>::operator*(const DiagonalBase<DiagonalDerived> &a_diagonal) const
|
||||
{
|
||||
return Product<Derived, DiagonalDerived, LazyProduct>(derived(),a_diagonal.derived());
|
||||
}
|
||||
|
||||
} // end namespace Eigen
|
||||
|
||||
#endif // EIGEN_DIAGONALPRODUCT_H
|
||||
@@ -0,0 +1,318 @@
|
||||
// This file is part of Eigen, a lightweight C++ template library
|
||||
// for linear algebra.
|
||||
//
|
||||
// Copyright (C) 2006-2008, 2010 Benoit Jacob <jacob.benoit.1@gmail.com>
|
||||
//
|
||||
// This Source Code Form is subject to the terms of the Mozilla
|
||||
// Public License v. 2.0. If a copy of the MPL was not distributed
|
||||
// with this file, You can obtain one at http://mozilla.org/MPL/2.0/.
|
||||
|
||||
#ifndef EIGEN_DOT_H
|
||||
#define EIGEN_DOT_H
|
||||
|
||||
namespace Eigen {
|
||||
|
||||
namespace internal {
|
||||
|
||||
// helper function for dot(). The problem is that if we put that in the body of dot(), then upon calling dot
|
||||
// with mismatched types, the compiler emits errors about failing to instantiate cwiseProduct BEFORE
|
||||
// looking at the static assertions. Thus this is a trick to get better compile errors.
|
||||
template<typename T, typename U,
|
||||
// the NeedToTranspose condition here is taken straight from Assign.h
|
||||
bool NeedToTranspose = T::IsVectorAtCompileTime
|
||||
&& U::IsVectorAtCompileTime
|
||||
&& ((int(T::RowsAtCompileTime) == 1 && int(U::ColsAtCompileTime) == 1)
|
||||
| // FIXME | instead of || to please GCC 4.4.0 stupid warning "suggest parentheses around &&".
|
||||
// revert to || as soon as not needed anymore.
|
||||
(int(T::ColsAtCompileTime) == 1 && int(U::RowsAtCompileTime) == 1))
|
||||
>
|
||||
struct dot_nocheck
|
||||
{
|
||||
typedef scalar_conj_product_op<typename traits<T>::Scalar,typename traits<U>::Scalar> conj_prod;
|
||||
typedef typename conj_prod::result_type ResScalar;
|
||||
EIGEN_DEVICE_FUNC
|
||||
EIGEN_STRONG_INLINE
|
||||
static ResScalar run(const MatrixBase<T>& a, const MatrixBase<U>& b)
|
||||
{
|
||||
return a.template binaryExpr<conj_prod>(b).sum();
|
||||
}
|
||||
};
|
||||
|
||||
template<typename T, typename U>
|
||||
struct dot_nocheck<T, U, true>
|
||||
{
|
||||
typedef scalar_conj_product_op<typename traits<T>::Scalar,typename traits<U>::Scalar> conj_prod;
|
||||
typedef typename conj_prod::result_type ResScalar;
|
||||
EIGEN_DEVICE_FUNC
|
||||
EIGEN_STRONG_INLINE
|
||||
static ResScalar run(const MatrixBase<T>& a, const MatrixBase<U>& b)
|
||||
{
|
||||
return a.transpose().template binaryExpr<conj_prod>(b).sum();
|
||||
}
|
||||
};
|
||||
|
||||
} // end namespace internal
|
||||
|
||||
/** \fn MatrixBase::dot
|
||||
* \returns the dot product of *this with other.
|
||||
*
|
||||
* \only_for_vectors
|
||||
*
|
||||
* \note If the scalar type is complex numbers, then this function returns the hermitian
|
||||
* (sesquilinear) dot product, conjugate-linear in the first variable and linear in the
|
||||
* second variable.
|
||||
*
|
||||
* \sa squaredNorm(), norm()
|
||||
*/
|
||||
template<typename Derived>
|
||||
template<typename OtherDerived>
|
||||
EIGEN_DEVICE_FUNC
|
||||
EIGEN_STRONG_INLINE
|
||||
typename ScalarBinaryOpTraits<typename internal::traits<Derived>::Scalar,typename internal::traits<OtherDerived>::Scalar>::ReturnType
|
||||
MatrixBase<Derived>::dot(const MatrixBase<OtherDerived>& other) const
|
||||
{
|
||||
EIGEN_STATIC_ASSERT_VECTOR_ONLY(Derived)
|
||||
EIGEN_STATIC_ASSERT_VECTOR_ONLY(OtherDerived)
|
||||
EIGEN_STATIC_ASSERT_SAME_VECTOR_SIZE(Derived,OtherDerived)
|
||||
#if !(defined(EIGEN_NO_STATIC_ASSERT) && defined(EIGEN_NO_DEBUG))
|
||||
typedef internal::scalar_conj_product_op<Scalar,typename OtherDerived::Scalar> func;
|
||||
EIGEN_CHECK_BINARY_COMPATIBILIY(func,Scalar,typename OtherDerived::Scalar);
|
||||
#endif
|
||||
|
||||
eigen_assert(size() == other.size());
|
||||
|
||||
return internal::dot_nocheck<Derived,OtherDerived>::run(*this, other);
|
||||
}
|
||||
|
||||
//---------- implementation of L2 norm and related functions ----------
|
||||
|
||||
/** \returns, for vectors, the squared \em l2 norm of \c *this, and for matrices the squared Frobenius norm.
|
||||
* In both cases, it consists in the sum of the square of all the matrix entries.
|
||||
* For vectors, this is also equals to the dot product of \c *this with itself.
|
||||
*
|
||||
* \sa dot(), norm(), lpNorm()
|
||||
*/
|
||||
template<typename Derived>
|
||||
EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE typename NumTraits<typename internal::traits<Derived>::Scalar>::Real MatrixBase<Derived>::squaredNorm() const
|
||||
{
|
||||
return numext::real((*this).cwiseAbs2().sum());
|
||||
}
|
||||
|
||||
/** \returns, for vectors, the \em l2 norm of \c *this, and for matrices the Frobenius norm.
|
||||
* In both cases, it consists in the square root of the sum of the square of all the matrix entries.
|
||||
* For vectors, this is also equals to the square root of the dot product of \c *this with itself.
|
||||
*
|
||||
* \sa lpNorm(), dot(), squaredNorm()
|
||||
*/
|
||||
template<typename Derived>
|
||||
EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE typename NumTraits<typename internal::traits<Derived>::Scalar>::Real MatrixBase<Derived>::norm() const
|
||||
{
|
||||
return numext::sqrt(squaredNorm());
|
||||
}
|
||||
|
||||
/** \returns an expression of the quotient of \c *this by its own norm.
|
||||
*
|
||||
* \warning If the input vector is too small (i.e., this->norm()==0),
|
||||
* then this function returns a copy of the input.
|
||||
*
|
||||
* \only_for_vectors
|
||||
*
|
||||
* \sa norm(), normalize()
|
||||
*/
|
||||
template<typename Derived>
|
||||
EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE const typename MatrixBase<Derived>::PlainObject
|
||||
MatrixBase<Derived>::normalized() const
|
||||
{
|
||||
typedef typename internal::nested_eval<Derived,2>::type _Nested;
|
||||
_Nested n(derived());
|
||||
RealScalar z = n.squaredNorm();
|
||||
// NOTE: after extensive benchmarking, this conditional does not impact performance, at least on recent x86 CPU
|
||||
if(z>RealScalar(0))
|
||||
return n / numext::sqrt(z);
|
||||
else
|
||||
return n;
|
||||
}
|
||||
|
||||
/** Normalizes the vector, i.e. divides it by its own norm.
|
||||
*
|
||||
* \only_for_vectors
|
||||
*
|
||||
* \warning If the input vector is too small (i.e., this->norm()==0), then \c *this is left unchanged.
|
||||
*
|
||||
* \sa norm(), normalized()
|
||||
*/
|
||||
template<typename Derived>
|
||||
EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE void MatrixBase<Derived>::normalize()
|
||||
{
|
||||
RealScalar z = squaredNorm();
|
||||
// NOTE: after extensive benchmarking, this conditional does not impact performance, at least on recent x86 CPU
|
||||
if(z>RealScalar(0))
|
||||
derived() /= numext::sqrt(z);
|
||||
}
|
||||
|
||||
/** \returns an expression of the quotient of \c *this by its own norm while avoiding underflow and overflow.
|
||||
*
|
||||
* \only_for_vectors
|
||||
*
|
||||
* This method is analogue to the normalized() method, but it reduces the risk of
|
||||
* underflow and overflow when computing the norm.
|
||||
*
|
||||
* \warning If the input vector is too small (i.e., this->norm()==0),
|
||||
* then this function returns a copy of the input.
|
||||
*
|
||||
* \sa stableNorm(), stableNormalize(), normalized()
|
||||
*/
|
||||
template<typename Derived>
|
||||
EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE const typename MatrixBase<Derived>::PlainObject
|
||||
MatrixBase<Derived>::stableNormalized() const
|
||||
{
|
||||
typedef typename internal::nested_eval<Derived,3>::type _Nested;
|
||||
_Nested n(derived());
|
||||
RealScalar w = n.cwiseAbs().maxCoeff();
|
||||
RealScalar z = (n/w).squaredNorm();
|
||||
if(z>RealScalar(0))
|
||||
return n / (numext::sqrt(z)*w);
|
||||
else
|
||||
return n;
|
||||
}
|
||||
|
||||
/** Normalizes the vector while avoid underflow and overflow
|
||||
*
|
||||
* \only_for_vectors
|
||||
*
|
||||
* This method is analogue to the normalize() method, but it reduces the risk of
|
||||
* underflow and overflow when computing the norm.
|
||||
*
|
||||
* \warning If the input vector is too small (i.e., this->norm()==0), then \c *this is left unchanged.
|
||||
*
|
||||
* \sa stableNorm(), stableNormalized(), normalize()
|
||||
*/
|
||||
template<typename Derived>
|
||||
EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE void MatrixBase<Derived>::stableNormalize()
|
||||
{
|
||||
RealScalar w = cwiseAbs().maxCoeff();
|
||||
RealScalar z = (derived()/w).squaredNorm();
|
||||
if(z>RealScalar(0))
|
||||
derived() /= numext::sqrt(z)*w;
|
||||
}
|
||||
|
||||
//---------- implementation of other norms ----------
|
||||
|
||||
namespace internal {
|
||||
|
||||
template<typename Derived, int p>
|
||||
struct lpNorm_selector
|
||||
{
|
||||
typedef typename NumTraits<typename traits<Derived>::Scalar>::Real RealScalar;
|
||||
EIGEN_DEVICE_FUNC
|
||||
static inline RealScalar run(const MatrixBase<Derived>& m)
|
||||
{
|
||||
EIGEN_USING_STD(pow)
|
||||
return pow(m.cwiseAbs().array().pow(p).sum(), RealScalar(1)/p);
|
||||
}
|
||||
};
|
||||
|
||||
template<typename Derived>
|
||||
struct lpNorm_selector<Derived, 1>
|
||||
{
|
||||
EIGEN_DEVICE_FUNC
|
||||
static inline typename NumTraits<typename traits<Derived>::Scalar>::Real run(const MatrixBase<Derived>& m)
|
||||
{
|
||||
return m.cwiseAbs().sum();
|
||||
}
|
||||
};
|
||||
|
||||
template<typename Derived>
|
||||
struct lpNorm_selector<Derived, 2>
|
||||
{
|
||||
EIGEN_DEVICE_FUNC
|
||||
static inline typename NumTraits<typename traits<Derived>::Scalar>::Real run(const MatrixBase<Derived>& m)
|
||||
{
|
||||
return m.norm();
|
||||
}
|
||||
};
|
||||
|
||||
template<typename Derived>
|
||||
struct lpNorm_selector<Derived, Infinity>
|
||||
{
|
||||
typedef typename NumTraits<typename traits<Derived>::Scalar>::Real RealScalar;
|
||||
EIGEN_DEVICE_FUNC
|
||||
static inline RealScalar run(const MatrixBase<Derived>& m)
|
||||
{
|
||||
if(Derived::SizeAtCompileTime==0 || (Derived::SizeAtCompileTime==Dynamic && m.size()==0))
|
||||
return RealScalar(0);
|
||||
return m.cwiseAbs().maxCoeff();
|
||||
}
|
||||
};
|
||||
|
||||
} // end namespace internal
|
||||
|
||||
/** \returns the \b coefficient-wise \f$ \ell^p \f$ norm of \c *this, that is, returns the p-th root of the sum of the p-th powers of the absolute values
|
||||
* of the coefficients of \c *this. If \a p is the special value \a Eigen::Infinity, this function returns the \f$ \ell^\infty \f$
|
||||
* norm, that is the maximum of the absolute values of the coefficients of \c *this.
|
||||
*
|
||||
* In all cases, if \c *this is empty, then the value 0 is returned.
|
||||
*
|
||||
* \note For matrices, this function does not compute the <a href="https://en.wikipedia.org/wiki/Operator_norm">operator-norm</a>. That is, if \c *this is a matrix, then its coefficients are interpreted as a 1D vector. Nonetheless, you can easily compute the 1-norm and \f$\infty\f$-norm matrix operator norms using \link TutorialReductionsVisitorsBroadcastingReductionsNorm partial reductions \endlink.
|
||||
*
|
||||
* \sa norm()
|
||||
*/
|
||||
template<typename Derived>
|
||||
template<int p>
|
||||
#ifndef EIGEN_PARSED_BY_DOXYGEN
|
||||
EIGEN_DEVICE_FUNC inline typename NumTraits<typename internal::traits<Derived>::Scalar>::Real
|
||||
#else
|
||||
EIGEN_DEVICE_FUNC MatrixBase<Derived>::RealScalar
|
||||
#endif
|
||||
MatrixBase<Derived>::lpNorm() const
|
||||
{
|
||||
return internal::lpNorm_selector<Derived, p>::run(*this);
|
||||
}
|
||||
|
||||
//---------- implementation of isOrthogonal / isUnitary ----------
|
||||
|
||||
/** \returns true if *this is approximately orthogonal to \a other,
|
||||
* within the precision given by \a prec.
|
||||
*
|
||||
* Example: \include MatrixBase_isOrthogonal.cpp
|
||||
* Output: \verbinclude MatrixBase_isOrthogonal.out
|
||||
*/
|
||||
template<typename Derived>
|
||||
template<typename OtherDerived>
|
||||
bool MatrixBase<Derived>::isOrthogonal
|
||||
(const MatrixBase<OtherDerived>& other, const RealScalar& prec) const
|
||||
{
|
||||
typename internal::nested_eval<Derived,2>::type nested(derived());
|
||||
typename internal::nested_eval<OtherDerived,2>::type otherNested(other.derived());
|
||||
return numext::abs2(nested.dot(otherNested)) <= prec * prec * nested.squaredNorm() * otherNested.squaredNorm();
|
||||
}
|
||||
|
||||
/** \returns true if *this is approximately an unitary matrix,
|
||||
* within the precision given by \a prec. In the case where the \a Scalar
|
||||
* type is real numbers, a unitary matrix is an orthogonal matrix, whence the name.
|
||||
*
|
||||
* \note This can be used to check whether a family of vectors forms an orthonormal basis.
|
||||
* Indeed, \c m.isUnitary() returns true if and only if the columns (equivalently, the rows) of m form an
|
||||
* orthonormal basis.
|
||||
*
|
||||
* Example: \include MatrixBase_isUnitary.cpp
|
||||
* Output: \verbinclude MatrixBase_isUnitary.out
|
||||
*/
|
||||
template<typename Derived>
|
||||
bool MatrixBase<Derived>::isUnitary(const RealScalar& prec) const
|
||||
{
|
||||
typename internal::nested_eval<Derived,1>::type self(derived());
|
||||
for(Index i = 0; i < cols(); ++i)
|
||||
{
|
||||
if(!internal::isApprox(self.col(i).squaredNorm(), static_cast<RealScalar>(1), prec))
|
||||
return false;
|
||||
for(Index j = 0; j < i; ++j)
|
||||
if(!internal::isMuchSmallerThan(self.col(i).dot(self.col(j)), static_cast<Scalar>(1), prec))
|
||||
return false;
|
||||
}
|
||||
return true;
|
||||
}
|
||||
|
||||
} // end namespace Eigen
|
||||
|
||||
#endif // EIGEN_DOT_H
|
||||
@@ -0,0 +1,160 @@
|
||||
// This file is part of Eigen, a lightweight C++ template library
|
||||
// for linear algebra.
|
||||
//
|
||||
// Copyright (C) 2009 Benoit Jacob <jacob.benoit.1@gmail.com>
|
||||
// Copyright (C) 2009 Gael Guennebaud <gael.guennebaud@inria.fr>
|
||||
//
|
||||
// This Source Code Form is subject to the terms of the Mozilla
|
||||
// Public License v. 2.0. If a copy of the MPL was not distributed
|
||||
// with this file, You can obtain one at http://mozilla.org/MPL/2.0/.
|
||||
|
||||
#ifndef EIGEN_EIGENBASE_H
|
||||
#define EIGEN_EIGENBASE_H
|
||||
|
||||
namespace Eigen {
|
||||
|
||||
/** \class EigenBase
|
||||
* \ingroup Core_Module
|
||||
*
|
||||
* Common base class for all classes T such that MatrixBase has an operator=(T) and a constructor MatrixBase(T).
|
||||
*
|
||||
* In other words, an EigenBase object is an object that can be copied into a MatrixBase.
|
||||
*
|
||||
* Besides MatrixBase-derived classes, this also includes special matrix classes such as diagonal matrices, etc.
|
||||
*
|
||||
* Notice that this class is trivial, it is only used to disambiguate overloaded functions.
|
||||
*
|
||||
* \sa \blank \ref TopicClassHierarchy
|
||||
*/
|
||||
template<typename Derived> struct EigenBase
|
||||
{
|
||||
// typedef typename internal::plain_matrix_type<Derived>::type PlainObject;
|
||||
|
||||
/** \brief The interface type of indices
|
||||
* \details To change this, \c \#define the preprocessor symbol \c EIGEN_DEFAULT_DENSE_INDEX_TYPE.
|
||||
* \sa StorageIndex, \ref TopicPreprocessorDirectives.
|
||||
* DEPRECATED: Since Eigen 3.3, its usage is deprecated. Use Eigen::Index instead.
|
||||
* Deprecation is not marked with a doxygen comment because there are too many existing usages to add the deprecation attribute.
|
||||
*/
|
||||
typedef Eigen::Index Index;
|
||||
|
||||
// FIXME is it needed?
|
||||
typedef typename internal::traits<Derived>::StorageKind StorageKind;
|
||||
|
||||
/** \returns a reference to the derived object */
|
||||
EIGEN_DEVICE_FUNC
|
||||
Derived& derived() { return *static_cast<Derived*>(this); }
|
||||
/** \returns a const reference to the derived object */
|
||||
EIGEN_DEVICE_FUNC
|
||||
const Derived& derived() const { return *static_cast<const Derived*>(this); }
|
||||
|
||||
EIGEN_DEVICE_FUNC
|
||||
inline Derived& const_cast_derived() const
|
||||
{ return *static_cast<Derived*>(const_cast<EigenBase*>(this)); }
|
||||
EIGEN_DEVICE_FUNC
|
||||
inline const Derived& const_derived() const
|
||||
{ return *static_cast<const Derived*>(this); }
|
||||
|
||||
/** \returns the number of rows. \sa cols(), RowsAtCompileTime */
|
||||
EIGEN_DEVICE_FUNC EIGEN_CONSTEXPR
|
||||
inline Index rows() const EIGEN_NOEXCEPT { return derived().rows(); }
|
||||
/** \returns the number of columns. \sa rows(), ColsAtCompileTime*/
|
||||
EIGEN_DEVICE_FUNC EIGEN_CONSTEXPR
|
||||
inline Index cols() const EIGEN_NOEXCEPT { return derived().cols(); }
|
||||
/** \returns the number of coefficients, which is rows()*cols().
|
||||
* \sa rows(), cols(), SizeAtCompileTime. */
|
||||
EIGEN_DEVICE_FUNC EIGEN_CONSTEXPR
|
||||
inline Index size() const EIGEN_NOEXCEPT { return rows() * cols(); }
|
||||
|
||||
/** \internal Don't use it, but do the equivalent: \code dst = *this; \endcode */
|
||||
template<typename Dest>
|
||||
EIGEN_DEVICE_FUNC
|
||||
inline void evalTo(Dest& dst) const
|
||||
{ derived().evalTo(dst); }
|
||||
|
||||
/** \internal Don't use it, but do the equivalent: \code dst += *this; \endcode */
|
||||
template<typename Dest>
|
||||
EIGEN_DEVICE_FUNC
|
||||
inline void addTo(Dest& dst) const
|
||||
{
|
||||
// This is the default implementation,
|
||||
// derived class can reimplement it in a more optimized way.
|
||||
typename Dest::PlainObject res(rows(),cols());
|
||||
evalTo(res);
|
||||
dst += res;
|
||||
}
|
||||
|
||||
/** \internal Don't use it, but do the equivalent: \code dst -= *this; \endcode */
|
||||
template<typename Dest>
|
||||
EIGEN_DEVICE_FUNC
|
||||
inline void subTo(Dest& dst) const
|
||||
{
|
||||
// This is the default implementation,
|
||||
// derived class can reimplement it in a more optimized way.
|
||||
typename Dest::PlainObject res(rows(),cols());
|
||||
evalTo(res);
|
||||
dst -= res;
|
||||
}
|
||||
|
||||
/** \internal Don't use it, but do the equivalent: \code dst.applyOnTheRight(*this); \endcode */
|
||||
template<typename Dest>
|
||||
EIGEN_DEVICE_FUNC inline void applyThisOnTheRight(Dest& dst) const
|
||||
{
|
||||
// This is the default implementation,
|
||||
// derived class can reimplement it in a more optimized way.
|
||||
dst = dst * this->derived();
|
||||
}
|
||||
|
||||
/** \internal Don't use it, but do the equivalent: \code dst.applyOnTheLeft(*this); \endcode */
|
||||
template<typename Dest>
|
||||
EIGEN_DEVICE_FUNC inline void applyThisOnTheLeft(Dest& dst) const
|
||||
{
|
||||
// This is the default implementation,
|
||||
// derived class can reimplement it in a more optimized way.
|
||||
dst = this->derived() * dst;
|
||||
}
|
||||
|
||||
};
|
||||
|
||||
/***************************************************************************
|
||||
* Implementation of matrix base methods
|
||||
***************************************************************************/
|
||||
|
||||
/** \brief Copies the generic expression \a other into *this.
|
||||
*
|
||||
* \details The expression must provide a (templated) evalTo(Derived& dst) const
|
||||
* function which does the actual job. In practice, this allows any user to write
|
||||
* its own special matrix without having to modify MatrixBase
|
||||
*
|
||||
* \returns a reference to *this.
|
||||
*/
|
||||
template<typename Derived>
|
||||
template<typename OtherDerived>
|
||||
EIGEN_DEVICE_FUNC
|
||||
Derived& DenseBase<Derived>::operator=(const EigenBase<OtherDerived> &other)
|
||||
{
|
||||
call_assignment(derived(), other.derived());
|
||||
return derived();
|
||||
}
|
||||
|
||||
template<typename Derived>
|
||||
template<typename OtherDerived>
|
||||
EIGEN_DEVICE_FUNC
|
||||
Derived& DenseBase<Derived>::operator+=(const EigenBase<OtherDerived> &other)
|
||||
{
|
||||
call_assignment(derived(), other.derived(), internal::add_assign_op<Scalar,typename OtherDerived::Scalar>());
|
||||
return derived();
|
||||
}
|
||||
|
||||
template<typename Derived>
|
||||
template<typename OtherDerived>
|
||||
EIGEN_DEVICE_FUNC
|
||||
Derived& DenseBase<Derived>::operator-=(const EigenBase<OtherDerived> &other)
|
||||
{
|
||||
call_assignment(derived(), other.derived(), internal::sub_assign_op<Scalar,typename OtherDerived::Scalar>());
|
||||
return derived();
|
||||
}
|
||||
|
||||
} // end namespace Eigen
|
||||
|
||||
#endif // EIGEN_EIGENBASE_H
|
||||
@@ -0,0 +1,150 @@
|
||||
// This file is part of Eigen, a lightweight C++ template library
|
||||
// for linear algebra.
|
||||
//
|
||||
// Copyright (C) 2009-2010 Gael Guennebaud <gael.guennebaud@inria.fr>
|
||||
//
|
||||
// This Source Code Form is subject to the terms of the Mozilla
|
||||
// Public License v. 2.0. If a copy of the MPL was not distributed
|
||||
// with this file, You can obtain one at http://mozilla.org/MPL/2.0/.
|
||||
|
||||
#ifndef EIGEN_FORCEALIGNEDACCESS_H
|
||||
#define EIGEN_FORCEALIGNEDACCESS_H
|
||||
|
||||
namespace Eigen {
|
||||
|
||||
/** \class ForceAlignedAccess
|
||||
* \ingroup Core_Module
|
||||
*
|
||||
* \brief Enforce aligned packet loads and stores regardless of what is requested
|
||||
*
|
||||
* \param ExpressionType the type of the object of which we are forcing aligned packet access
|
||||
*
|
||||
* This class is the return type of MatrixBase::forceAlignedAccess()
|
||||
* and most of the time this is the only way it is used.
|
||||
*
|
||||
* \sa MatrixBase::forceAlignedAccess()
|
||||
*/
|
||||
|
||||
namespace internal {
|
||||
template<typename ExpressionType>
|
||||
struct traits<ForceAlignedAccess<ExpressionType> > : public traits<ExpressionType>
|
||||
{};
|
||||
}
|
||||
|
||||
template<typename ExpressionType> class ForceAlignedAccess
|
||||
: public internal::dense_xpr_base< ForceAlignedAccess<ExpressionType> >::type
|
||||
{
|
||||
public:
|
||||
|
||||
typedef typename internal::dense_xpr_base<ForceAlignedAccess>::type Base;
|
||||
EIGEN_DENSE_PUBLIC_INTERFACE(ForceAlignedAccess)
|
||||
|
||||
EIGEN_DEVICE_FUNC explicit inline ForceAlignedAccess(const ExpressionType& matrix) : m_expression(matrix) {}
|
||||
|
||||
EIGEN_DEVICE_FUNC EIGEN_CONSTEXPR
|
||||
inline Index rows() const EIGEN_NOEXCEPT { return m_expression.rows(); }
|
||||
EIGEN_DEVICE_FUNC EIGEN_CONSTEXPR
|
||||
inline Index cols() const EIGEN_NOEXCEPT { return m_expression.cols(); }
|
||||
EIGEN_DEVICE_FUNC EIGEN_CONSTEXPR
|
||||
inline Index outerStride() const EIGEN_NOEXCEPT { return m_expression.outerStride(); }
|
||||
EIGEN_DEVICE_FUNC EIGEN_CONSTEXPR
|
||||
inline Index innerStride() const EIGEN_NOEXCEPT { return m_expression.innerStride(); }
|
||||
|
||||
EIGEN_DEVICE_FUNC inline const CoeffReturnType coeff(Index row, Index col) const
|
||||
{
|
||||
return m_expression.coeff(row, col);
|
||||
}
|
||||
|
||||
EIGEN_DEVICE_FUNC inline Scalar& coeffRef(Index row, Index col)
|
||||
{
|
||||
return m_expression.const_cast_derived().coeffRef(row, col);
|
||||
}
|
||||
|
||||
EIGEN_DEVICE_FUNC inline const CoeffReturnType coeff(Index index) const
|
||||
{
|
||||
return m_expression.coeff(index);
|
||||
}
|
||||
|
||||
EIGEN_DEVICE_FUNC inline Scalar& coeffRef(Index index)
|
||||
{
|
||||
return m_expression.const_cast_derived().coeffRef(index);
|
||||
}
|
||||
|
||||
template<int LoadMode>
|
||||
inline const PacketScalar packet(Index row, Index col) const
|
||||
{
|
||||
return m_expression.template packet<Aligned>(row, col);
|
||||
}
|
||||
|
||||
template<int LoadMode>
|
||||
inline void writePacket(Index row, Index col, const PacketScalar& x)
|
||||
{
|
||||
m_expression.const_cast_derived().template writePacket<Aligned>(row, col, x);
|
||||
}
|
||||
|
||||
template<int LoadMode>
|
||||
inline const PacketScalar packet(Index index) const
|
||||
{
|
||||
return m_expression.template packet<Aligned>(index);
|
||||
}
|
||||
|
||||
template<int LoadMode>
|
||||
inline void writePacket(Index index, const PacketScalar& x)
|
||||
{
|
||||
m_expression.const_cast_derived().template writePacket<Aligned>(index, x);
|
||||
}
|
||||
|
||||
EIGEN_DEVICE_FUNC operator const ExpressionType&() const { return m_expression; }
|
||||
|
||||
protected:
|
||||
const ExpressionType& m_expression;
|
||||
|
||||
private:
|
||||
ForceAlignedAccess& operator=(const ForceAlignedAccess&);
|
||||
};
|
||||
|
||||
/** \returns an expression of *this with forced aligned access
|
||||
* \sa forceAlignedAccessIf(),class ForceAlignedAccess
|
||||
*/
|
||||
template<typename Derived>
|
||||
inline const ForceAlignedAccess<Derived>
|
||||
MatrixBase<Derived>::forceAlignedAccess() const
|
||||
{
|
||||
return ForceAlignedAccess<Derived>(derived());
|
||||
}
|
||||
|
||||
/** \returns an expression of *this with forced aligned access
|
||||
* \sa forceAlignedAccessIf(), class ForceAlignedAccess
|
||||
*/
|
||||
template<typename Derived>
|
||||
inline ForceAlignedAccess<Derived>
|
||||
MatrixBase<Derived>::forceAlignedAccess()
|
||||
{
|
||||
return ForceAlignedAccess<Derived>(derived());
|
||||
}
|
||||
|
||||
/** \returns an expression of *this with forced aligned access if \a Enable is true.
|
||||
* \sa forceAlignedAccess(), class ForceAlignedAccess
|
||||
*/
|
||||
template<typename Derived>
|
||||
template<bool Enable>
|
||||
inline typename internal::add_const_on_value_type<typename internal::conditional<Enable,ForceAlignedAccess<Derived>,Derived&>::type>::type
|
||||
MatrixBase<Derived>::forceAlignedAccessIf() const
|
||||
{
|
||||
return derived(); // FIXME This should not work but apparently is never used
|
||||
}
|
||||
|
||||
/** \returns an expression of *this with forced aligned access if \a Enable is true.
|
||||
* \sa forceAlignedAccess(), class ForceAlignedAccess
|
||||
*/
|
||||
template<typename Derived>
|
||||
template<bool Enable>
|
||||
inline typename internal::conditional<Enable,ForceAlignedAccess<Derived>,Derived&>::type
|
||||
MatrixBase<Derived>::forceAlignedAccessIf()
|
||||
{
|
||||
return derived(); // FIXME This should not work but apparently is never used
|
||||
}
|
||||
|
||||
} // end namespace Eigen
|
||||
|
||||
#endif // EIGEN_FORCEALIGNEDACCESS_H
|
||||
@@ -0,0 +1,155 @@
|
||||
// This file is part of Eigen, a lightweight C++ template library
|
||||
// for linear algebra.
|
||||
//
|
||||
// Copyright (C) 2006-2008 Benoit Jacob <jacob.benoit.1@gmail.com>
|
||||
// Copyright (C) 2008 Gael Guennebaud <gael.guennebaud@inria.fr>
|
||||
//
|
||||
// This Source Code Form is subject to the terms of the Mozilla
|
||||
// Public License v. 2.0. If a copy of the MPL was not distributed
|
||||
// with this file, You can obtain one at http://mozilla.org/MPL/2.0/.
|
||||
|
||||
#ifndef EIGEN_FUZZY_H
|
||||
#define EIGEN_FUZZY_H
|
||||
|
||||
namespace Eigen {
|
||||
|
||||
namespace internal
|
||||
{
|
||||
|
||||
template<typename Derived, typename OtherDerived, bool is_integer = NumTraits<typename Derived::Scalar>::IsInteger>
|
||||
struct isApprox_selector
|
||||
{
|
||||
EIGEN_DEVICE_FUNC
|
||||
static bool run(const Derived& x, const OtherDerived& y, const typename Derived::RealScalar& prec)
|
||||
{
|
||||
typename internal::nested_eval<Derived,2>::type nested(x);
|
||||
typename internal::nested_eval<OtherDerived,2>::type otherNested(y);
|
||||
return (nested - otherNested).cwiseAbs2().sum() <= prec * prec * numext::mini(nested.cwiseAbs2().sum(), otherNested.cwiseAbs2().sum());
|
||||
}
|
||||
};
|
||||
|
||||
template<typename Derived, typename OtherDerived>
|
||||
struct isApprox_selector<Derived, OtherDerived, true>
|
||||
{
|
||||
EIGEN_DEVICE_FUNC
|
||||
static bool run(const Derived& x, const OtherDerived& y, const typename Derived::RealScalar&)
|
||||
{
|
||||
return x.matrix() == y.matrix();
|
||||
}
|
||||
};
|
||||
|
||||
template<typename Derived, typename OtherDerived, bool is_integer = NumTraits<typename Derived::Scalar>::IsInteger>
|
||||
struct isMuchSmallerThan_object_selector
|
||||
{
|
||||
EIGEN_DEVICE_FUNC
|
||||
static bool run(const Derived& x, const OtherDerived& y, const typename Derived::RealScalar& prec)
|
||||
{
|
||||
return x.cwiseAbs2().sum() <= numext::abs2(prec) * y.cwiseAbs2().sum();
|
||||
}
|
||||
};
|
||||
|
||||
template<typename Derived, typename OtherDerived>
|
||||
struct isMuchSmallerThan_object_selector<Derived, OtherDerived, true>
|
||||
{
|
||||
EIGEN_DEVICE_FUNC
|
||||
static bool run(const Derived& x, const OtherDerived&, const typename Derived::RealScalar&)
|
||||
{
|
||||
return x.matrix() == Derived::Zero(x.rows(), x.cols()).matrix();
|
||||
}
|
||||
};
|
||||
|
||||
template<typename Derived, bool is_integer = NumTraits<typename Derived::Scalar>::IsInteger>
|
||||
struct isMuchSmallerThan_scalar_selector
|
||||
{
|
||||
EIGEN_DEVICE_FUNC
|
||||
static bool run(const Derived& x, const typename Derived::RealScalar& y, const typename Derived::RealScalar& prec)
|
||||
{
|
||||
return x.cwiseAbs2().sum() <= numext::abs2(prec * y);
|
||||
}
|
||||
};
|
||||
|
||||
template<typename Derived>
|
||||
struct isMuchSmallerThan_scalar_selector<Derived, true>
|
||||
{
|
||||
EIGEN_DEVICE_FUNC
|
||||
static bool run(const Derived& x, const typename Derived::RealScalar&, const typename Derived::RealScalar&)
|
||||
{
|
||||
return x.matrix() == Derived::Zero(x.rows(), x.cols()).matrix();
|
||||
}
|
||||
};
|
||||
|
||||
} // end namespace internal
|
||||
|
||||
|
||||
/** \returns \c true if \c *this is approximately equal to \a other, within the precision
|
||||
* determined by \a prec.
|
||||
*
|
||||
* \note The fuzzy compares are done multiplicatively. Two vectors \f$ v \f$ and \f$ w \f$
|
||||
* are considered to be approximately equal within precision \f$ p \f$ if
|
||||
* \f[ \Vert v - w \Vert \leqslant p\,\min(\Vert v\Vert, \Vert w\Vert). \f]
|
||||
* For matrices, the comparison is done using the Hilbert-Schmidt norm (aka Frobenius norm
|
||||
* L2 norm).
|
||||
*
|
||||
* \note Because of the multiplicativeness of this comparison, one can't use this function
|
||||
* to check whether \c *this is approximately equal to the zero matrix or vector.
|
||||
* Indeed, \c isApprox(zero) returns false unless \c *this itself is exactly the zero matrix
|
||||
* or vector. If you want to test whether \c *this is zero, use internal::isMuchSmallerThan(const
|
||||
* RealScalar&, RealScalar) instead.
|
||||
*
|
||||
* \sa internal::isMuchSmallerThan(const RealScalar&, RealScalar) const
|
||||
*/
|
||||
template<typename Derived>
|
||||
template<typename OtherDerived>
|
||||
EIGEN_DEVICE_FUNC bool DenseBase<Derived>::isApprox(
|
||||
const DenseBase<OtherDerived>& other,
|
||||
const RealScalar& prec
|
||||
) const
|
||||
{
|
||||
return internal::isApprox_selector<Derived, OtherDerived>::run(derived(), other.derived(), prec);
|
||||
}
|
||||
|
||||
/** \returns \c true if the norm of \c *this is much smaller than \a other,
|
||||
* within the precision determined by \a prec.
|
||||
*
|
||||
* \note The fuzzy compares are done multiplicatively. A vector \f$ v \f$ is
|
||||
* considered to be much smaller than \f$ x \f$ within precision \f$ p \f$ if
|
||||
* \f[ \Vert v \Vert \leqslant p\,\vert x\vert. \f]
|
||||
*
|
||||
* For matrices, the comparison is done using the Hilbert-Schmidt norm. For this reason,
|
||||
* the value of the reference scalar \a other should come from the Hilbert-Schmidt norm
|
||||
* of a reference matrix of same dimensions.
|
||||
*
|
||||
* \sa isApprox(), isMuchSmallerThan(const DenseBase<OtherDerived>&, RealScalar) const
|
||||
*/
|
||||
template<typename Derived>
|
||||
EIGEN_DEVICE_FUNC bool DenseBase<Derived>::isMuchSmallerThan(
|
||||
const typename NumTraits<Scalar>::Real& other,
|
||||
const RealScalar& prec
|
||||
) const
|
||||
{
|
||||
return internal::isMuchSmallerThan_scalar_selector<Derived>::run(derived(), other, prec);
|
||||
}
|
||||
|
||||
/** \returns \c true if the norm of \c *this is much smaller than the norm of \a other,
|
||||
* within the precision determined by \a prec.
|
||||
*
|
||||
* \note The fuzzy compares are done multiplicatively. A vector \f$ v \f$ is
|
||||
* considered to be much smaller than a vector \f$ w \f$ within precision \f$ p \f$ if
|
||||
* \f[ \Vert v \Vert \leqslant p\,\Vert w\Vert. \f]
|
||||
* For matrices, the comparison is done using the Hilbert-Schmidt norm.
|
||||
*
|
||||
* \sa isApprox(), isMuchSmallerThan(const RealScalar&, RealScalar) const
|
||||
*/
|
||||
template<typename Derived>
|
||||
template<typename OtherDerived>
|
||||
EIGEN_DEVICE_FUNC bool DenseBase<Derived>::isMuchSmallerThan(
|
||||
const DenseBase<OtherDerived>& other,
|
||||
const RealScalar& prec
|
||||
) const
|
||||
{
|
||||
return internal::isMuchSmallerThan_object_selector<Derived, OtherDerived>::run(derived(), other.derived(), prec);
|
||||
}
|
||||
|
||||
} // end namespace Eigen
|
||||
|
||||
#endif // EIGEN_FUZZY_H
|
||||
@@ -0,0 +1,465 @@
|
||||
// This file is part of Eigen, a lightweight C++ template library
|
||||
// for linear algebra.
|
||||
//
|
||||
// Copyright (C) 2006-2008 Benoit Jacob <jacob.benoit.1@gmail.com>
|
||||
// Copyright (C) 2008-2011 Gael Guennebaud <gael.guennebaud@inria.fr>
|
||||
//
|
||||
// This Source Code Form is subject to the terms of the Mozilla
|
||||
// Public License v. 2.0. If a copy of the MPL was not distributed
|
||||
// with this file, You can obtain one at http://mozilla.org/MPL/2.0/.
|
||||
|
||||
#ifndef EIGEN_GENERAL_PRODUCT_H
|
||||
#define EIGEN_GENERAL_PRODUCT_H
|
||||
|
||||
namespace Eigen {
|
||||
|
||||
enum {
|
||||
Large = 2,
|
||||
Small = 3
|
||||
};
|
||||
|
||||
// Define the threshold value to fallback from the generic matrix-matrix product
|
||||
// implementation (heavy) to the lightweight coeff-based product one.
|
||||
// See generic_product_impl<Lhs,Rhs,DenseShape,DenseShape,GemmProduct>
|
||||
// in products/GeneralMatrixMatrix.h for more details.
|
||||
// TODO This threshold should also be used in the compile-time selector below.
|
||||
#ifndef EIGEN_GEMM_TO_COEFFBASED_THRESHOLD
|
||||
// This default value has been obtained on a Haswell architecture.
|
||||
#define EIGEN_GEMM_TO_COEFFBASED_THRESHOLD 20
|
||||
#endif
|
||||
|
||||
namespace internal {
|
||||
|
||||
template<int Rows, int Cols, int Depth> struct product_type_selector;
|
||||
|
||||
template<int Size, int MaxSize> struct product_size_category
|
||||
{
|
||||
enum {
|
||||
#ifndef EIGEN_GPU_COMPILE_PHASE
|
||||
is_large = MaxSize == Dynamic ||
|
||||
Size >= EIGEN_CACHEFRIENDLY_PRODUCT_THRESHOLD ||
|
||||
(Size==Dynamic && MaxSize>=EIGEN_CACHEFRIENDLY_PRODUCT_THRESHOLD),
|
||||
#else
|
||||
is_large = 0,
|
||||
#endif
|
||||
value = is_large ? Large
|
||||
: Size == 1 ? 1
|
||||
: Small
|
||||
};
|
||||
};
|
||||
|
||||
template<typename Lhs, typename Rhs> struct product_type
|
||||
{
|
||||
typedef typename remove_all<Lhs>::type _Lhs;
|
||||
typedef typename remove_all<Rhs>::type _Rhs;
|
||||
enum {
|
||||
MaxRows = traits<_Lhs>::MaxRowsAtCompileTime,
|
||||
Rows = traits<_Lhs>::RowsAtCompileTime,
|
||||
MaxCols = traits<_Rhs>::MaxColsAtCompileTime,
|
||||
Cols = traits<_Rhs>::ColsAtCompileTime,
|
||||
MaxDepth = EIGEN_SIZE_MIN_PREFER_FIXED(traits<_Lhs>::MaxColsAtCompileTime,
|
||||
traits<_Rhs>::MaxRowsAtCompileTime),
|
||||
Depth = EIGEN_SIZE_MIN_PREFER_FIXED(traits<_Lhs>::ColsAtCompileTime,
|
||||
traits<_Rhs>::RowsAtCompileTime)
|
||||
};
|
||||
|
||||
// the splitting into different lines of code here, introducing the _select enums and the typedef below,
|
||||
// is to work around an internal compiler error with gcc 4.1 and 4.2.
|
||||
private:
|
||||
enum {
|
||||
rows_select = product_size_category<Rows,MaxRows>::value,
|
||||
cols_select = product_size_category<Cols,MaxCols>::value,
|
||||
depth_select = product_size_category<Depth,MaxDepth>::value
|
||||
};
|
||||
typedef product_type_selector<rows_select, cols_select, depth_select> selector;
|
||||
|
||||
public:
|
||||
enum {
|
||||
value = selector::ret,
|
||||
ret = selector::ret
|
||||
};
|
||||
#ifdef EIGEN_DEBUG_PRODUCT
|
||||
static void debug()
|
||||
{
|
||||
EIGEN_DEBUG_VAR(Rows);
|
||||
EIGEN_DEBUG_VAR(Cols);
|
||||
EIGEN_DEBUG_VAR(Depth);
|
||||
EIGEN_DEBUG_VAR(rows_select);
|
||||
EIGEN_DEBUG_VAR(cols_select);
|
||||
EIGEN_DEBUG_VAR(depth_select);
|
||||
EIGEN_DEBUG_VAR(value);
|
||||
}
|
||||
#endif
|
||||
};
|
||||
|
||||
/* The following allows to select the kind of product at compile time
|
||||
* based on the three dimensions of the product.
|
||||
* This is a compile time mapping from {1,Small,Large}^3 -> {product types} */
|
||||
// FIXME I'm not sure the current mapping is the ideal one.
|
||||
template<int M, int N> struct product_type_selector<M,N,1> { enum { ret = OuterProduct }; };
|
||||
template<int M> struct product_type_selector<M, 1, 1> { enum { ret = LazyCoeffBasedProductMode }; };
|
||||
template<int N> struct product_type_selector<1, N, 1> { enum { ret = LazyCoeffBasedProductMode }; };
|
||||
template<int Depth> struct product_type_selector<1, 1, Depth> { enum { ret = InnerProduct }; };
|
||||
template<> struct product_type_selector<1, 1, 1> { enum { ret = InnerProduct }; };
|
||||
template<> struct product_type_selector<Small,1, Small> { enum { ret = CoeffBasedProductMode }; };
|
||||
template<> struct product_type_selector<1, Small,Small> { enum { ret = CoeffBasedProductMode }; };
|
||||
template<> struct product_type_selector<Small,Small,Small> { enum { ret = CoeffBasedProductMode }; };
|
||||
template<> struct product_type_selector<Small, Small, 1> { enum { ret = LazyCoeffBasedProductMode }; };
|
||||
template<> struct product_type_selector<Small, Large, 1> { enum { ret = LazyCoeffBasedProductMode }; };
|
||||
template<> struct product_type_selector<Large, Small, 1> { enum { ret = LazyCoeffBasedProductMode }; };
|
||||
template<> struct product_type_selector<1, Large,Small> { enum { ret = CoeffBasedProductMode }; };
|
||||
template<> struct product_type_selector<1, Large,Large> { enum { ret = GemvProduct }; };
|
||||
template<> struct product_type_selector<1, Small,Large> { enum { ret = CoeffBasedProductMode }; };
|
||||
template<> struct product_type_selector<Large,1, Small> { enum { ret = CoeffBasedProductMode }; };
|
||||
template<> struct product_type_selector<Large,1, Large> { enum { ret = GemvProduct }; };
|
||||
template<> struct product_type_selector<Small,1, Large> { enum { ret = CoeffBasedProductMode }; };
|
||||
template<> struct product_type_selector<Small,Small,Large> { enum { ret = GemmProduct }; };
|
||||
template<> struct product_type_selector<Large,Small,Large> { enum { ret = GemmProduct }; };
|
||||
template<> struct product_type_selector<Small,Large,Large> { enum { ret = GemmProduct }; };
|
||||
template<> struct product_type_selector<Large,Large,Large> { enum { ret = GemmProduct }; };
|
||||
template<> struct product_type_selector<Large,Small,Small> { enum { ret = CoeffBasedProductMode }; };
|
||||
template<> struct product_type_selector<Small,Large,Small> { enum { ret = CoeffBasedProductMode }; };
|
||||
template<> struct product_type_selector<Large,Large,Small> { enum { ret = GemmProduct }; };
|
||||
|
||||
} // end namespace internal
|
||||
|
||||
/***********************************************************************
|
||||
* Implementation of Inner Vector Vector Product
|
||||
***********************************************************************/
|
||||
|
||||
// FIXME : maybe the "inner product" could return a Scalar
|
||||
// instead of a 1x1 matrix ??
|
||||
// Pro: more natural for the user
|
||||
// Cons: this could be a problem if in a meta unrolled algorithm a matrix-matrix
|
||||
// product ends up to a row-vector times col-vector product... To tackle this use
|
||||
// case, we could have a specialization for Block<MatrixType,1,1> with: operator=(Scalar x);
|
||||
|
||||
/***********************************************************************
|
||||
* Implementation of Outer Vector Vector Product
|
||||
***********************************************************************/
|
||||
|
||||
/***********************************************************************
|
||||
* Implementation of General Matrix Vector Product
|
||||
***********************************************************************/
|
||||
|
||||
/* According to the shape/flags of the matrix we have to distinghish 3 different cases:
|
||||
* 1 - the matrix is col-major, BLAS compatible and M is large => call fast BLAS-like colmajor routine
|
||||
* 2 - the matrix is row-major, BLAS compatible and N is large => call fast BLAS-like rowmajor routine
|
||||
* 3 - all other cases are handled using a simple loop along the outer-storage direction.
|
||||
* Therefore we need a lower level meta selector.
|
||||
* Furthermore, if the matrix is the rhs, then the product has to be transposed.
|
||||
*/
|
||||
namespace internal {
|
||||
|
||||
template<int Side, int StorageOrder, bool BlasCompatible>
|
||||
struct gemv_dense_selector;
|
||||
|
||||
} // end namespace internal
|
||||
|
||||
namespace internal {
|
||||
|
||||
template<typename Scalar,int Size,int MaxSize,bool Cond> struct gemv_static_vector_if;
|
||||
|
||||
template<typename Scalar,int Size,int MaxSize>
|
||||
struct gemv_static_vector_if<Scalar,Size,MaxSize,false>
|
||||
{
|
||||
EIGEN_STRONG_INLINE EIGEN_DEVICE_FUNC Scalar* data() { eigen_internal_assert(false && "should never be called"); return 0; }
|
||||
};
|
||||
|
||||
template<typename Scalar,int Size>
|
||||
struct gemv_static_vector_if<Scalar,Size,Dynamic,true>
|
||||
{
|
||||
EIGEN_STRONG_INLINE EIGEN_DEVICE_FUNC Scalar* data() { return 0; }
|
||||
};
|
||||
|
||||
template<typename Scalar,int Size,int MaxSize>
|
||||
struct gemv_static_vector_if<Scalar,Size,MaxSize,true>
|
||||
{
|
||||
enum {
|
||||
ForceAlignment = internal::packet_traits<Scalar>::Vectorizable,
|
||||
PacketSize = internal::packet_traits<Scalar>::size
|
||||
};
|
||||
#if EIGEN_MAX_STATIC_ALIGN_BYTES!=0
|
||||
internal::plain_array<Scalar,EIGEN_SIZE_MIN_PREFER_FIXED(Size,MaxSize),0,EIGEN_PLAIN_ENUM_MIN(AlignedMax,PacketSize)> m_data;
|
||||
EIGEN_STRONG_INLINE Scalar* data() { return m_data.array; }
|
||||
#else
|
||||
// Some architectures cannot align on the stack,
|
||||
// => let's manually enforce alignment by allocating more data and return the address of the first aligned element.
|
||||
internal::plain_array<Scalar,EIGEN_SIZE_MIN_PREFER_FIXED(Size,MaxSize)+(ForceAlignment?EIGEN_MAX_ALIGN_BYTES:0),0> m_data;
|
||||
EIGEN_STRONG_INLINE Scalar* data() {
|
||||
return ForceAlignment
|
||||
? reinterpret_cast<Scalar*>((internal::UIntPtr(m_data.array) & ~(std::size_t(EIGEN_MAX_ALIGN_BYTES-1))) + EIGEN_MAX_ALIGN_BYTES)
|
||||
: m_data.array;
|
||||
}
|
||||
#endif
|
||||
};
|
||||
|
||||
// The vector is on the left => transposition
|
||||
template<int StorageOrder, bool BlasCompatible>
|
||||
struct gemv_dense_selector<OnTheLeft,StorageOrder,BlasCompatible>
|
||||
{
|
||||
template<typename Lhs, typename Rhs, typename Dest>
|
||||
static void run(const Lhs &lhs, const Rhs &rhs, Dest& dest, const typename Dest::Scalar& alpha)
|
||||
{
|
||||
Transpose<Dest> destT(dest);
|
||||
enum { OtherStorageOrder = StorageOrder == RowMajor ? ColMajor : RowMajor };
|
||||
gemv_dense_selector<OnTheRight,OtherStorageOrder,BlasCompatible>
|
||||
::run(rhs.transpose(), lhs.transpose(), destT, alpha);
|
||||
}
|
||||
};
|
||||
|
||||
template<> struct gemv_dense_selector<OnTheRight,ColMajor,true>
|
||||
{
|
||||
template<typename Lhs, typename Rhs, typename Dest>
|
||||
static inline void run(const Lhs &lhs, const Rhs &rhs, Dest& dest, const typename Dest::Scalar& alpha)
|
||||
{
|
||||
typedef typename Lhs::Scalar LhsScalar;
|
||||
typedef typename Rhs::Scalar RhsScalar;
|
||||
typedef typename Dest::Scalar ResScalar;
|
||||
typedef typename Dest::RealScalar RealScalar;
|
||||
|
||||
typedef internal::blas_traits<Lhs> LhsBlasTraits;
|
||||
typedef typename LhsBlasTraits::DirectLinearAccessType ActualLhsType;
|
||||
typedef internal::blas_traits<Rhs> RhsBlasTraits;
|
||||
typedef typename RhsBlasTraits::DirectLinearAccessType ActualRhsType;
|
||||
|
||||
typedef Map<Matrix<ResScalar,Dynamic,1>, EIGEN_PLAIN_ENUM_MIN(AlignedMax,internal::packet_traits<ResScalar>::size)> MappedDest;
|
||||
|
||||
ActualLhsType actualLhs = LhsBlasTraits::extract(lhs);
|
||||
ActualRhsType actualRhs = RhsBlasTraits::extract(rhs);
|
||||
|
||||
ResScalar actualAlpha = combine_scalar_factors(alpha, lhs, rhs);
|
||||
|
||||
// make sure Dest is a compile-time vector type (bug 1166)
|
||||
typedef typename conditional<Dest::IsVectorAtCompileTime, Dest, typename Dest::ColXpr>::type ActualDest;
|
||||
|
||||
enum {
|
||||
// FIXME find a way to allow an inner stride on the result if packet_traits<Scalar>::size==1
|
||||
// on, the other hand it is good for the cache to pack the vector anyways...
|
||||
EvalToDestAtCompileTime = (ActualDest::InnerStrideAtCompileTime==1),
|
||||
ComplexByReal = (NumTraits<LhsScalar>::IsComplex) && (!NumTraits<RhsScalar>::IsComplex),
|
||||
MightCannotUseDest = ((!EvalToDestAtCompileTime) || ComplexByReal) && (ActualDest::MaxSizeAtCompileTime!=0)
|
||||
};
|
||||
|
||||
typedef const_blas_data_mapper<LhsScalar,Index,ColMajor> LhsMapper;
|
||||
typedef const_blas_data_mapper<RhsScalar,Index,RowMajor> RhsMapper;
|
||||
RhsScalar compatibleAlpha = get_factor<ResScalar,RhsScalar>::run(actualAlpha);
|
||||
|
||||
if(!MightCannotUseDest)
|
||||
{
|
||||
// shortcut if we are sure to be able to use dest directly,
|
||||
// this ease the compiler to generate cleaner and more optimzized code for most common cases
|
||||
general_matrix_vector_product
|
||||
<Index,LhsScalar,LhsMapper,ColMajor,LhsBlasTraits::NeedToConjugate,RhsScalar,RhsMapper,RhsBlasTraits::NeedToConjugate>::run(
|
||||
actualLhs.rows(), actualLhs.cols(),
|
||||
LhsMapper(actualLhs.data(), actualLhs.outerStride()),
|
||||
RhsMapper(actualRhs.data(), actualRhs.innerStride()),
|
||||
dest.data(), 1,
|
||||
compatibleAlpha);
|
||||
}
|
||||
else
|
||||
{
|
||||
gemv_static_vector_if<ResScalar,ActualDest::SizeAtCompileTime,ActualDest::MaxSizeAtCompileTime,MightCannotUseDest> static_dest;
|
||||
|
||||
const bool alphaIsCompatible = (!ComplexByReal) || (numext::imag(actualAlpha)==RealScalar(0));
|
||||
const bool evalToDest = EvalToDestAtCompileTime && alphaIsCompatible;
|
||||
|
||||
ei_declare_aligned_stack_constructed_variable(ResScalar,actualDestPtr,dest.size(),
|
||||
evalToDest ? dest.data() : static_dest.data());
|
||||
|
||||
if(!evalToDest)
|
||||
{
|
||||
#ifdef EIGEN_DENSE_STORAGE_CTOR_PLUGIN
|
||||
Index size = dest.size();
|
||||
EIGEN_DENSE_STORAGE_CTOR_PLUGIN
|
||||
#endif
|
||||
if(!alphaIsCompatible)
|
||||
{
|
||||
MappedDest(actualDestPtr, dest.size()).setZero();
|
||||
compatibleAlpha = RhsScalar(1);
|
||||
}
|
||||
else
|
||||
MappedDest(actualDestPtr, dest.size()) = dest;
|
||||
}
|
||||
|
||||
general_matrix_vector_product
|
||||
<Index,LhsScalar,LhsMapper,ColMajor,LhsBlasTraits::NeedToConjugate,RhsScalar,RhsMapper,RhsBlasTraits::NeedToConjugate>::run(
|
||||
actualLhs.rows(), actualLhs.cols(),
|
||||
LhsMapper(actualLhs.data(), actualLhs.outerStride()),
|
||||
RhsMapper(actualRhs.data(), actualRhs.innerStride()),
|
||||
actualDestPtr, 1,
|
||||
compatibleAlpha);
|
||||
|
||||
if (!evalToDest)
|
||||
{
|
||||
if(!alphaIsCompatible)
|
||||
dest.matrix() += actualAlpha * MappedDest(actualDestPtr, dest.size());
|
||||
else
|
||||
dest = MappedDest(actualDestPtr, dest.size());
|
||||
}
|
||||
}
|
||||
}
|
||||
};
|
||||
|
||||
template<> struct gemv_dense_selector<OnTheRight,RowMajor,true>
|
||||
{
|
||||
template<typename Lhs, typename Rhs, typename Dest>
|
||||
static void run(const Lhs &lhs, const Rhs &rhs, Dest& dest, const typename Dest::Scalar& alpha)
|
||||
{
|
||||
typedef typename Lhs::Scalar LhsScalar;
|
||||
typedef typename Rhs::Scalar RhsScalar;
|
||||
typedef typename Dest::Scalar ResScalar;
|
||||
|
||||
typedef internal::blas_traits<Lhs> LhsBlasTraits;
|
||||
typedef typename LhsBlasTraits::DirectLinearAccessType ActualLhsType;
|
||||
typedef internal::blas_traits<Rhs> RhsBlasTraits;
|
||||
typedef typename RhsBlasTraits::DirectLinearAccessType ActualRhsType;
|
||||
typedef typename internal::remove_all<ActualRhsType>::type ActualRhsTypeCleaned;
|
||||
|
||||
typename add_const<ActualLhsType>::type actualLhs = LhsBlasTraits::extract(lhs);
|
||||
typename add_const<ActualRhsType>::type actualRhs = RhsBlasTraits::extract(rhs);
|
||||
|
||||
ResScalar actualAlpha = combine_scalar_factors(alpha, lhs, rhs);
|
||||
|
||||
enum {
|
||||
// FIXME find a way to allow an inner stride on the result if packet_traits<Scalar>::size==1
|
||||
// on, the other hand it is good for the cache to pack the vector anyways...
|
||||
DirectlyUseRhs = ActualRhsTypeCleaned::InnerStrideAtCompileTime==1 || ActualRhsTypeCleaned::MaxSizeAtCompileTime==0
|
||||
};
|
||||
|
||||
gemv_static_vector_if<RhsScalar,ActualRhsTypeCleaned::SizeAtCompileTime,ActualRhsTypeCleaned::MaxSizeAtCompileTime,!DirectlyUseRhs> static_rhs;
|
||||
|
||||
ei_declare_aligned_stack_constructed_variable(RhsScalar,actualRhsPtr,actualRhs.size(),
|
||||
DirectlyUseRhs ? const_cast<RhsScalar*>(actualRhs.data()) : static_rhs.data());
|
||||
|
||||
if(!DirectlyUseRhs)
|
||||
{
|
||||
#ifdef EIGEN_DENSE_STORAGE_CTOR_PLUGIN
|
||||
Index size = actualRhs.size();
|
||||
EIGEN_DENSE_STORAGE_CTOR_PLUGIN
|
||||
#endif
|
||||
Map<typename ActualRhsTypeCleaned::PlainObject>(actualRhsPtr, actualRhs.size()) = actualRhs;
|
||||
}
|
||||
|
||||
typedef const_blas_data_mapper<LhsScalar,Index,RowMajor> LhsMapper;
|
||||
typedef const_blas_data_mapper<RhsScalar,Index,ColMajor> RhsMapper;
|
||||
general_matrix_vector_product
|
||||
<Index,LhsScalar,LhsMapper,RowMajor,LhsBlasTraits::NeedToConjugate,RhsScalar,RhsMapper,RhsBlasTraits::NeedToConjugate>::run(
|
||||
actualLhs.rows(), actualLhs.cols(),
|
||||
LhsMapper(actualLhs.data(), actualLhs.outerStride()),
|
||||
RhsMapper(actualRhsPtr, 1),
|
||||
dest.data(), dest.col(0).innerStride(), //NOTE if dest is not a vector at compile-time, then dest.innerStride() might be wrong. (bug 1166)
|
||||
actualAlpha);
|
||||
}
|
||||
};
|
||||
|
||||
template<> struct gemv_dense_selector<OnTheRight,ColMajor,false>
|
||||
{
|
||||
template<typename Lhs, typename Rhs, typename Dest>
|
||||
static void run(const Lhs &lhs, const Rhs &rhs, Dest& dest, const typename Dest::Scalar& alpha)
|
||||
{
|
||||
EIGEN_STATIC_ASSERT((!nested_eval<Lhs,1>::Evaluate),EIGEN_INTERNAL_COMPILATION_ERROR_OR_YOU_MADE_A_PROGRAMMING_MISTAKE);
|
||||
// TODO if rhs is large enough it might be beneficial to make sure that dest is sequentially stored in memory, otherwise use a temp
|
||||
typename nested_eval<Rhs,1>::type actual_rhs(rhs);
|
||||
const Index size = rhs.rows();
|
||||
for(Index k=0; k<size; ++k)
|
||||
dest += (alpha*actual_rhs.coeff(k)) * lhs.col(k);
|
||||
}
|
||||
};
|
||||
|
||||
template<> struct gemv_dense_selector<OnTheRight,RowMajor,false>
|
||||
{
|
||||
template<typename Lhs, typename Rhs, typename Dest>
|
||||
static void run(const Lhs &lhs, const Rhs &rhs, Dest& dest, const typename Dest::Scalar& alpha)
|
||||
{
|
||||
EIGEN_STATIC_ASSERT((!nested_eval<Lhs,1>::Evaluate),EIGEN_INTERNAL_COMPILATION_ERROR_OR_YOU_MADE_A_PROGRAMMING_MISTAKE);
|
||||
typename nested_eval<Rhs,Lhs::RowsAtCompileTime>::type actual_rhs(rhs);
|
||||
const Index rows = dest.rows();
|
||||
for(Index i=0; i<rows; ++i)
|
||||
dest.coeffRef(i) += alpha * (lhs.row(i).cwiseProduct(actual_rhs.transpose())).sum();
|
||||
}
|
||||
};
|
||||
|
||||
} // end namespace internal
|
||||
|
||||
/***************************************************************************
|
||||
* Implementation of matrix base methods
|
||||
***************************************************************************/
|
||||
|
||||
/** \returns the matrix product of \c *this and \a other.
|
||||
*
|
||||
* \note If instead of the matrix product you want the coefficient-wise product, see Cwise::operator*().
|
||||
*
|
||||
* \sa lazyProduct(), operator*=(const MatrixBase&), Cwise::operator*()
|
||||
*/
|
||||
template<typename Derived>
|
||||
template<typename OtherDerived>
|
||||
EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
|
||||
const Product<Derived, OtherDerived>
|
||||
MatrixBase<Derived>::operator*(const MatrixBase<OtherDerived> &other) const
|
||||
{
|
||||
// A note regarding the function declaration: In MSVC, this function will sometimes
|
||||
// not be inlined since DenseStorage is an unwindable object for dynamic
|
||||
// matrices and product types are holding a member to store the result.
|
||||
// Thus it does not help tagging this function with EIGEN_STRONG_INLINE.
|
||||
enum {
|
||||
ProductIsValid = Derived::ColsAtCompileTime==Dynamic
|
||||
|| OtherDerived::RowsAtCompileTime==Dynamic
|
||||
|| int(Derived::ColsAtCompileTime)==int(OtherDerived::RowsAtCompileTime),
|
||||
AreVectors = Derived::IsVectorAtCompileTime && OtherDerived::IsVectorAtCompileTime,
|
||||
SameSizes = EIGEN_PREDICATE_SAME_MATRIX_SIZE(Derived,OtherDerived)
|
||||
};
|
||||
// note to the lost user:
|
||||
// * for a dot product use: v1.dot(v2)
|
||||
// * for a coeff-wise product use: v1.cwiseProduct(v2)
|
||||
EIGEN_STATIC_ASSERT(ProductIsValid || !(AreVectors && SameSizes),
|
||||
INVALID_VECTOR_VECTOR_PRODUCT__IF_YOU_WANTED_A_DOT_OR_COEFF_WISE_PRODUCT_YOU_MUST_USE_THE_EXPLICIT_FUNCTIONS)
|
||||
EIGEN_STATIC_ASSERT(ProductIsValid || !(SameSizes && !AreVectors),
|
||||
INVALID_MATRIX_PRODUCT__IF_YOU_WANTED_A_COEFF_WISE_PRODUCT_YOU_MUST_USE_THE_EXPLICIT_FUNCTION)
|
||||
EIGEN_STATIC_ASSERT(ProductIsValid || SameSizes, INVALID_MATRIX_PRODUCT)
|
||||
#ifdef EIGEN_DEBUG_PRODUCT
|
||||
internal::product_type<Derived,OtherDerived>::debug();
|
||||
#endif
|
||||
|
||||
return Product<Derived, OtherDerived>(derived(), other.derived());
|
||||
}
|
||||
|
||||
/** \returns an expression of the matrix product of \c *this and \a other without implicit evaluation.
|
||||
*
|
||||
* The returned product will behave like any other expressions: the coefficients of the product will be
|
||||
* computed once at a time as requested. This might be useful in some extremely rare cases when only
|
||||
* a small and no coherent fraction of the result's coefficients have to be computed.
|
||||
*
|
||||
* \warning This version of the matrix product can be much much slower. So use it only if you know
|
||||
* what you are doing and that you measured a true speed improvement.
|
||||
*
|
||||
* \sa operator*(const MatrixBase&)
|
||||
*/
|
||||
template<typename Derived>
|
||||
template<typename OtherDerived>
|
||||
EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
|
||||
const Product<Derived,OtherDerived,LazyProduct>
|
||||
MatrixBase<Derived>::lazyProduct(const MatrixBase<OtherDerived> &other) const
|
||||
{
|
||||
enum {
|
||||
ProductIsValid = Derived::ColsAtCompileTime==Dynamic
|
||||
|| OtherDerived::RowsAtCompileTime==Dynamic
|
||||
|| int(Derived::ColsAtCompileTime)==int(OtherDerived::RowsAtCompileTime),
|
||||
AreVectors = Derived::IsVectorAtCompileTime && OtherDerived::IsVectorAtCompileTime,
|
||||
SameSizes = EIGEN_PREDICATE_SAME_MATRIX_SIZE(Derived,OtherDerived)
|
||||
};
|
||||
// note to the lost user:
|
||||
// * for a dot product use: v1.dot(v2)
|
||||
// * for a coeff-wise product use: v1.cwiseProduct(v2)
|
||||
EIGEN_STATIC_ASSERT(ProductIsValid || !(AreVectors && SameSizes),
|
||||
INVALID_VECTOR_VECTOR_PRODUCT__IF_YOU_WANTED_A_DOT_OR_COEFF_WISE_PRODUCT_YOU_MUST_USE_THE_EXPLICIT_FUNCTIONS)
|
||||
EIGEN_STATIC_ASSERT(ProductIsValid || !(SameSizes && !AreVectors),
|
||||
INVALID_MATRIX_PRODUCT__IF_YOU_WANTED_A_COEFF_WISE_PRODUCT_YOU_MUST_USE_THE_EXPLICIT_FUNCTION)
|
||||
EIGEN_STATIC_ASSERT(ProductIsValid || SameSizes, INVALID_MATRIX_PRODUCT)
|
||||
|
||||
return Product<Derived,OtherDerived,LazyProduct>(derived(), other.derived());
|
||||
}
|
||||
|
||||
} // end namespace Eigen
|
||||
|
||||
#endif // EIGEN_PRODUCT_H
|
||||
File diff suppressed because it is too large
Load Diff
@@ -0,0 +1,194 @@
|
||||
// This file is part of Eigen, a lightweight C++ template library
|
||||
// for linear algebra.
|
||||
//
|
||||
// Copyright (C) 2010-2016 Gael Guennebaud <gael.guennebaud@inria.fr>
|
||||
// Copyright (C) 2010 Benoit Jacob <jacob.benoit.1@gmail.com>
|
||||
//
|
||||
// This Source Code Form is subject to the terms of the Mozilla
|
||||
// Public License v. 2.0. If a copy of the MPL was not distributed
|
||||
// with this file, You can obtain one at http://mozilla.org/MPL/2.0/.
|
||||
|
||||
#ifndef EIGEN_GLOBAL_FUNCTIONS_H
|
||||
#define EIGEN_GLOBAL_FUNCTIONS_H
|
||||
|
||||
#ifdef EIGEN_PARSED_BY_DOXYGEN
|
||||
|
||||
#define EIGEN_ARRAY_DECLARE_GLOBAL_UNARY(NAME,FUNCTOR,DOC_OP,DOC_DETAILS) \
|
||||
/** \returns an expression of the coefficient-wise DOC_OP of \a x
|
||||
|
||||
DOC_DETAILS
|
||||
|
||||
\sa <a href="group__CoeffwiseMathFunctions.html#cwisetable_##NAME">Math functions</a>, class CwiseUnaryOp
|
||||
*/ \
|
||||
template<typename Derived> \
|
||||
inline const Eigen::CwiseUnaryOp<Eigen::internal::FUNCTOR<typename Derived::Scalar>, const Derived> \
|
||||
NAME(const Eigen::ArrayBase<Derived>& x);
|
||||
|
||||
#else
|
||||
|
||||
#define EIGEN_ARRAY_DECLARE_GLOBAL_UNARY(NAME,FUNCTOR,DOC_OP,DOC_DETAILS) \
|
||||
template<typename Derived> \
|
||||
inline const Eigen::CwiseUnaryOp<Eigen::internal::FUNCTOR<typename Derived::Scalar>, const Derived> \
|
||||
(NAME)(const Eigen::ArrayBase<Derived>& x) { \
|
||||
return Eigen::CwiseUnaryOp<Eigen::internal::FUNCTOR<typename Derived::Scalar>, const Derived>(x.derived()); \
|
||||
}
|
||||
|
||||
#endif // EIGEN_PARSED_BY_DOXYGEN
|
||||
|
||||
#define EIGEN_ARRAY_DECLARE_GLOBAL_EIGEN_UNARY(NAME,FUNCTOR) \
|
||||
\
|
||||
template<typename Derived> \
|
||||
struct NAME##_retval<ArrayBase<Derived> > \
|
||||
{ \
|
||||
typedef const Eigen::CwiseUnaryOp<Eigen::internal::FUNCTOR<typename Derived::Scalar>, const Derived> type; \
|
||||
}; \
|
||||
template<typename Derived> \
|
||||
struct NAME##_impl<ArrayBase<Derived> > \
|
||||
{ \
|
||||
static inline typename NAME##_retval<ArrayBase<Derived> >::type run(const Eigen::ArrayBase<Derived>& x) \
|
||||
{ \
|
||||
return typename NAME##_retval<ArrayBase<Derived> >::type(x.derived()); \
|
||||
} \
|
||||
};
|
||||
|
||||
namespace Eigen
|
||||
{
|
||||
EIGEN_ARRAY_DECLARE_GLOBAL_UNARY(real,scalar_real_op,real part,\sa ArrayBase::real)
|
||||
EIGEN_ARRAY_DECLARE_GLOBAL_UNARY(imag,scalar_imag_op,imaginary part,\sa ArrayBase::imag)
|
||||
EIGEN_ARRAY_DECLARE_GLOBAL_UNARY(conj,scalar_conjugate_op,complex conjugate,\sa ArrayBase::conjugate)
|
||||
EIGEN_ARRAY_DECLARE_GLOBAL_UNARY(inverse,scalar_inverse_op,inverse,\sa ArrayBase::inverse)
|
||||
EIGEN_ARRAY_DECLARE_GLOBAL_UNARY(sin,scalar_sin_op,sine,\sa ArrayBase::sin)
|
||||
EIGEN_ARRAY_DECLARE_GLOBAL_UNARY(cos,scalar_cos_op,cosine,\sa ArrayBase::cos)
|
||||
EIGEN_ARRAY_DECLARE_GLOBAL_UNARY(tan,scalar_tan_op,tangent,\sa ArrayBase::tan)
|
||||
EIGEN_ARRAY_DECLARE_GLOBAL_UNARY(atan,scalar_atan_op,arc-tangent,\sa ArrayBase::atan)
|
||||
EIGEN_ARRAY_DECLARE_GLOBAL_UNARY(asin,scalar_asin_op,arc-sine,\sa ArrayBase::asin)
|
||||
EIGEN_ARRAY_DECLARE_GLOBAL_UNARY(acos,scalar_acos_op,arc-consine,\sa ArrayBase::acos)
|
||||
EIGEN_ARRAY_DECLARE_GLOBAL_UNARY(sinh,scalar_sinh_op,hyperbolic sine,\sa ArrayBase::sinh)
|
||||
EIGEN_ARRAY_DECLARE_GLOBAL_UNARY(cosh,scalar_cosh_op,hyperbolic cosine,\sa ArrayBase::cosh)
|
||||
EIGEN_ARRAY_DECLARE_GLOBAL_UNARY(tanh,scalar_tanh_op,hyperbolic tangent,\sa ArrayBase::tanh)
|
||||
#if EIGEN_HAS_CXX11_MATH
|
||||
EIGEN_ARRAY_DECLARE_GLOBAL_UNARY(asinh,scalar_asinh_op,inverse hyperbolic sine,\sa ArrayBase::asinh)
|
||||
EIGEN_ARRAY_DECLARE_GLOBAL_UNARY(acosh,scalar_acosh_op,inverse hyperbolic cosine,\sa ArrayBase::acosh)
|
||||
EIGEN_ARRAY_DECLARE_GLOBAL_UNARY(atanh,scalar_atanh_op,inverse hyperbolic tangent,\sa ArrayBase::atanh)
|
||||
#endif
|
||||
EIGEN_ARRAY_DECLARE_GLOBAL_UNARY(logistic,scalar_logistic_op,logistic function,\sa ArrayBase::logistic)
|
||||
EIGEN_ARRAY_DECLARE_GLOBAL_UNARY(lgamma,scalar_lgamma_op,natural logarithm of the gamma function,\sa ArrayBase::lgamma)
|
||||
EIGEN_ARRAY_DECLARE_GLOBAL_UNARY(digamma,scalar_digamma_op,derivative of lgamma,\sa ArrayBase::digamma)
|
||||
EIGEN_ARRAY_DECLARE_GLOBAL_UNARY(erf,scalar_erf_op,error function,\sa ArrayBase::erf)
|
||||
EIGEN_ARRAY_DECLARE_GLOBAL_UNARY(erfc,scalar_erfc_op,complement error function,\sa ArrayBase::erfc)
|
||||
EIGEN_ARRAY_DECLARE_GLOBAL_UNARY(ndtri,scalar_ndtri_op,inverse normal distribution function,\sa ArrayBase::ndtri)
|
||||
EIGEN_ARRAY_DECLARE_GLOBAL_UNARY(exp,scalar_exp_op,exponential,\sa ArrayBase::exp)
|
||||
EIGEN_ARRAY_DECLARE_GLOBAL_UNARY(expm1,scalar_expm1_op,exponential of a value minus 1,\sa ArrayBase::expm1)
|
||||
EIGEN_ARRAY_DECLARE_GLOBAL_UNARY(log,scalar_log_op,natural logarithm,\sa Eigen::log10 DOXCOMMA ArrayBase::log)
|
||||
EIGEN_ARRAY_DECLARE_GLOBAL_UNARY(log1p,scalar_log1p_op,natural logarithm of 1 plus the value,\sa ArrayBase::log1p)
|
||||
EIGEN_ARRAY_DECLARE_GLOBAL_UNARY(log10,scalar_log10_op,base 10 logarithm,\sa Eigen::log DOXCOMMA ArrayBase::log10)
|
||||
EIGEN_ARRAY_DECLARE_GLOBAL_UNARY(log2,scalar_log2_op,base 2 logarithm,\sa Eigen::log DOXCOMMA ArrayBase::log2)
|
||||
EIGEN_ARRAY_DECLARE_GLOBAL_UNARY(abs,scalar_abs_op,absolute value,\sa ArrayBase::abs DOXCOMMA MatrixBase::cwiseAbs)
|
||||
EIGEN_ARRAY_DECLARE_GLOBAL_UNARY(abs2,scalar_abs2_op,squared absolute value,\sa ArrayBase::abs2 DOXCOMMA MatrixBase::cwiseAbs2)
|
||||
EIGEN_ARRAY_DECLARE_GLOBAL_UNARY(arg,scalar_arg_op,complex argument,\sa ArrayBase::arg DOXCOMMA MatrixBase::cwiseArg)
|
||||
EIGEN_ARRAY_DECLARE_GLOBAL_UNARY(sqrt,scalar_sqrt_op,square root,\sa ArrayBase::sqrt DOXCOMMA MatrixBase::cwiseSqrt)
|
||||
EIGEN_ARRAY_DECLARE_GLOBAL_UNARY(rsqrt,scalar_rsqrt_op,reciprocal square root,\sa ArrayBase::rsqrt)
|
||||
EIGEN_ARRAY_DECLARE_GLOBAL_UNARY(square,scalar_square_op,square (power 2),\sa Eigen::abs2 DOXCOMMA Eigen::pow DOXCOMMA ArrayBase::square)
|
||||
EIGEN_ARRAY_DECLARE_GLOBAL_UNARY(cube,scalar_cube_op,cube (power 3),\sa Eigen::pow DOXCOMMA ArrayBase::cube)
|
||||
EIGEN_ARRAY_DECLARE_GLOBAL_UNARY(rint,scalar_rint_op,nearest integer,\sa Eigen::floor DOXCOMMA Eigen::ceil DOXCOMMA ArrayBase::round)
|
||||
EIGEN_ARRAY_DECLARE_GLOBAL_UNARY(round,scalar_round_op,nearest integer,\sa Eigen::floor DOXCOMMA Eigen::ceil DOXCOMMA ArrayBase::round)
|
||||
EIGEN_ARRAY_DECLARE_GLOBAL_UNARY(floor,scalar_floor_op,nearest integer not greater than the giben value,\sa Eigen::ceil DOXCOMMA ArrayBase::floor)
|
||||
EIGEN_ARRAY_DECLARE_GLOBAL_UNARY(ceil,scalar_ceil_op,nearest integer not less than the giben value,\sa Eigen::floor DOXCOMMA ArrayBase::ceil)
|
||||
EIGEN_ARRAY_DECLARE_GLOBAL_UNARY(isnan,scalar_isnan_op,not-a-number test,\sa Eigen::isinf DOXCOMMA Eigen::isfinite DOXCOMMA ArrayBase::isnan)
|
||||
EIGEN_ARRAY_DECLARE_GLOBAL_UNARY(isinf,scalar_isinf_op,infinite value test,\sa Eigen::isnan DOXCOMMA Eigen::isfinite DOXCOMMA ArrayBase::isinf)
|
||||
EIGEN_ARRAY_DECLARE_GLOBAL_UNARY(isfinite,scalar_isfinite_op,finite value test,\sa Eigen::isinf DOXCOMMA Eigen::isnan DOXCOMMA ArrayBase::isfinite)
|
||||
EIGEN_ARRAY_DECLARE_GLOBAL_UNARY(sign,scalar_sign_op,sign (or 0),\sa ArrayBase::sign)
|
||||
|
||||
/** \returns an expression of the coefficient-wise power of \a x to the given constant \a exponent.
|
||||
*
|
||||
* \tparam ScalarExponent is the scalar type of \a exponent. It must be compatible with the scalar type of the given expression (\c Derived::Scalar).
|
||||
*
|
||||
* \sa ArrayBase::pow()
|
||||
*
|
||||
* \relates ArrayBase
|
||||
*/
|
||||
#ifdef EIGEN_PARSED_BY_DOXYGEN
|
||||
template<typename Derived,typename ScalarExponent>
|
||||
inline const CwiseBinaryOp<internal::scalar_pow_op<Derived::Scalar,ScalarExponent>,Derived,Constant<ScalarExponent> >
|
||||
pow(const Eigen::ArrayBase<Derived>& x, const ScalarExponent& exponent);
|
||||
#else
|
||||
template <typename Derived,typename ScalarExponent>
|
||||
EIGEN_DEVICE_FUNC inline
|
||||
EIGEN_MSVC10_WORKAROUND_BINARYOP_RETURN_TYPE(
|
||||
const EIGEN_EXPR_BINARYOP_SCALAR_RETURN_TYPE(Derived,typename internal::promote_scalar_arg<typename Derived::Scalar
|
||||
EIGEN_COMMA ScalarExponent EIGEN_COMMA
|
||||
EIGEN_SCALAR_BINARY_SUPPORTED(pow,typename Derived::Scalar,ScalarExponent)>::type,pow))
|
||||
pow(const Eigen::ArrayBase<Derived>& x, const ScalarExponent& exponent)
|
||||
{
|
||||
typedef typename internal::promote_scalar_arg<typename Derived::Scalar,ScalarExponent,
|
||||
EIGEN_SCALAR_BINARY_SUPPORTED(pow,typename Derived::Scalar,ScalarExponent)>::type PromotedExponent;
|
||||
return EIGEN_EXPR_BINARYOP_SCALAR_RETURN_TYPE(Derived,PromotedExponent,pow)(x.derived(),
|
||||
typename internal::plain_constant_type<Derived,PromotedExponent>::type(x.derived().rows(), x.derived().cols(), internal::scalar_constant_op<PromotedExponent>(exponent)));
|
||||
}
|
||||
#endif
|
||||
|
||||
/** \returns an expression of the coefficient-wise power of \a x to the given array of \a exponents.
|
||||
*
|
||||
* This function computes the coefficient-wise power.
|
||||
*
|
||||
* Example: \include Cwise_array_power_array.cpp
|
||||
* Output: \verbinclude Cwise_array_power_array.out
|
||||
*
|
||||
* \sa ArrayBase::pow()
|
||||
*
|
||||
* \relates ArrayBase
|
||||
*/
|
||||
template<typename Derived,typename ExponentDerived>
|
||||
inline const Eigen::CwiseBinaryOp<Eigen::internal::scalar_pow_op<typename Derived::Scalar, typename ExponentDerived::Scalar>, const Derived, const ExponentDerived>
|
||||
pow(const Eigen::ArrayBase<Derived>& x, const Eigen::ArrayBase<ExponentDerived>& exponents)
|
||||
{
|
||||
return Eigen::CwiseBinaryOp<Eigen::internal::scalar_pow_op<typename Derived::Scalar, typename ExponentDerived::Scalar>, const Derived, const ExponentDerived>(
|
||||
x.derived(),
|
||||
exponents.derived()
|
||||
);
|
||||
}
|
||||
|
||||
/** \returns an expression of the coefficient-wise power of the scalar \a x to the given array of \a exponents.
|
||||
*
|
||||
* This function computes the coefficient-wise power between a scalar and an array of exponents.
|
||||
*
|
||||
* \tparam Scalar is the scalar type of \a x. It must be compatible with the scalar type of the given array expression (\c Derived::Scalar).
|
||||
*
|
||||
* Example: \include Cwise_scalar_power_array.cpp
|
||||
* Output: \verbinclude Cwise_scalar_power_array.out
|
||||
*
|
||||
* \sa ArrayBase::pow()
|
||||
*
|
||||
* \relates ArrayBase
|
||||
*/
|
||||
#ifdef EIGEN_PARSED_BY_DOXYGEN
|
||||
template<typename Scalar,typename Derived>
|
||||
inline const CwiseBinaryOp<internal::scalar_pow_op<Scalar,Derived::Scalar>,Constant<Scalar>,Derived>
|
||||
pow(const Scalar& x,const Eigen::ArrayBase<Derived>& x);
|
||||
#else
|
||||
template <typename Scalar, typename Derived>
|
||||
EIGEN_DEVICE_FUNC inline
|
||||
EIGEN_MSVC10_WORKAROUND_BINARYOP_RETURN_TYPE(
|
||||
const EIGEN_SCALAR_BINARYOP_EXPR_RETURN_TYPE(typename internal::promote_scalar_arg<typename Derived::Scalar
|
||||
EIGEN_COMMA Scalar EIGEN_COMMA
|
||||
EIGEN_SCALAR_BINARY_SUPPORTED(pow,Scalar,typename Derived::Scalar)>::type,Derived,pow))
|
||||
pow(const Scalar& x, const Eigen::ArrayBase<Derived>& exponents) {
|
||||
typedef typename internal::promote_scalar_arg<typename Derived::Scalar,Scalar,
|
||||
EIGEN_SCALAR_BINARY_SUPPORTED(pow,Scalar,typename Derived::Scalar)>::type PromotedScalar;
|
||||
return EIGEN_SCALAR_BINARYOP_EXPR_RETURN_TYPE(PromotedScalar,Derived,pow)(
|
||||
typename internal::plain_constant_type<Derived,PromotedScalar>::type(exponents.derived().rows(), exponents.derived().cols(), internal::scalar_constant_op<PromotedScalar>(x)), exponents.derived());
|
||||
}
|
||||
#endif
|
||||
|
||||
|
||||
namespace internal
|
||||
{
|
||||
EIGEN_ARRAY_DECLARE_GLOBAL_EIGEN_UNARY(real,scalar_real_op)
|
||||
EIGEN_ARRAY_DECLARE_GLOBAL_EIGEN_UNARY(imag,scalar_imag_op)
|
||||
EIGEN_ARRAY_DECLARE_GLOBAL_EIGEN_UNARY(abs2,scalar_abs2_op)
|
||||
}
|
||||
}
|
||||
|
||||
// TODO: cleanly disable those functions that are not supported on Array (numext::real_ref, internal::random, internal::isApprox...)
|
||||
|
||||
#endif // EIGEN_GLOBAL_FUNCTIONS_H
|
||||
@@ -0,0 +1,258 @@
|
||||
// This file is part of Eigen, a lightweight C++ template library
|
||||
// for linear algebra.
|
||||
//
|
||||
// Copyright (C) 2006-2008 Benoit Jacob <jacob.benoit.1@gmail.com>
|
||||
// Copyright (C) 2008 Gael Guennebaud <gael.guennebaud@inria.fr>
|
||||
//
|
||||
// This Source Code Form is subject to the terms of the Mozilla
|
||||
// Public License v. 2.0. If a copy of the MPL was not distributed
|
||||
// with this file, You can obtain one at http://mozilla.org/MPL/2.0/.
|
||||
|
||||
#ifndef EIGEN_IO_H
|
||||
#define EIGEN_IO_H
|
||||
|
||||
namespace Eigen {
|
||||
|
||||
enum { DontAlignCols = 1 };
|
||||
enum { StreamPrecision = -1,
|
||||
FullPrecision = -2 };
|
||||
|
||||
namespace internal {
|
||||
template<typename Derived>
|
||||
std::ostream & print_matrix(std::ostream & s, const Derived& _m, const IOFormat& fmt);
|
||||
}
|
||||
|
||||
/** \class IOFormat
|
||||
* \ingroup Core_Module
|
||||
*
|
||||
* \brief Stores a set of parameters controlling the way matrices are printed
|
||||
*
|
||||
* List of available parameters:
|
||||
* - \b precision number of digits for floating point values, or one of the special constants \c StreamPrecision and \c FullPrecision.
|
||||
* The default is the special value \c StreamPrecision which means to use the
|
||||
* stream's own precision setting, as set for instance using \c cout.precision(3). The other special value
|
||||
* \c FullPrecision means that the number of digits will be computed to match the full precision of each floating-point
|
||||
* type.
|
||||
* - \b flags an OR-ed combination of flags, the default value is 0, the only currently available flag is \c DontAlignCols which
|
||||
* allows to disable the alignment of columns, resulting in faster code.
|
||||
* - \b coeffSeparator string printed between two coefficients of the same row
|
||||
* - \b rowSeparator string printed between two rows
|
||||
* - \b rowPrefix string printed at the beginning of each row
|
||||
* - \b rowSuffix string printed at the end of each row
|
||||
* - \b matPrefix string printed at the beginning of the matrix
|
||||
* - \b matSuffix string printed at the end of the matrix
|
||||
* - \b fill character printed to fill the empty space in aligned columns
|
||||
*
|
||||
* Example: \include IOFormat.cpp
|
||||
* Output: \verbinclude IOFormat.out
|
||||
*
|
||||
* \sa DenseBase::format(), class WithFormat
|
||||
*/
|
||||
struct IOFormat
|
||||
{
|
||||
/** Default constructor, see class IOFormat for the meaning of the parameters */
|
||||
IOFormat(int _precision = StreamPrecision, int _flags = 0,
|
||||
const std::string& _coeffSeparator = " ",
|
||||
const std::string& _rowSeparator = "\n", const std::string& _rowPrefix="", const std::string& _rowSuffix="",
|
||||
const std::string& _matPrefix="", const std::string& _matSuffix="", const char _fill=' ')
|
||||
: matPrefix(_matPrefix), matSuffix(_matSuffix), rowPrefix(_rowPrefix), rowSuffix(_rowSuffix), rowSeparator(_rowSeparator),
|
||||
rowSpacer(""), coeffSeparator(_coeffSeparator), fill(_fill), precision(_precision), flags(_flags)
|
||||
{
|
||||
// TODO check if rowPrefix, rowSuffix or rowSeparator contains a newline
|
||||
// don't add rowSpacer if columns are not to be aligned
|
||||
if((flags & DontAlignCols))
|
||||
return;
|
||||
int i = int(matSuffix.length())-1;
|
||||
while (i>=0 && matSuffix[i]!='\n')
|
||||
{
|
||||
rowSpacer += ' ';
|
||||
i--;
|
||||
}
|
||||
}
|
||||
std::string matPrefix, matSuffix;
|
||||
std::string rowPrefix, rowSuffix, rowSeparator, rowSpacer;
|
||||
std::string coeffSeparator;
|
||||
char fill;
|
||||
int precision;
|
||||
int flags;
|
||||
};
|
||||
|
||||
/** \class WithFormat
|
||||
* \ingroup Core_Module
|
||||
*
|
||||
* \brief Pseudo expression providing matrix output with given format
|
||||
*
|
||||
* \tparam ExpressionType the type of the object on which IO stream operations are performed
|
||||
*
|
||||
* This class represents an expression with stream operators controlled by a given IOFormat.
|
||||
* It is the return type of DenseBase::format()
|
||||
* and most of the time this is the only way it is used.
|
||||
*
|
||||
* See class IOFormat for some examples.
|
||||
*
|
||||
* \sa DenseBase::format(), class IOFormat
|
||||
*/
|
||||
template<typename ExpressionType>
|
||||
class WithFormat
|
||||
{
|
||||
public:
|
||||
|
||||
WithFormat(const ExpressionType& matrix, const IOFormat& format)
|
||||
: m_matrix(matrix), m_format(format)
|
||||
{}
|
||||
|
||||
friend std::ostream & operator << (std::ostream & s, const WithFormat& wf)
|
||||
{
|
||||
return internal::print_matrix(s, wf.m_matrix.eval(), wf.m_format);
|
||||
}
|
||||
|
||||
protected:
|
||||
typename ExpressionType::Nested m_matrix;
|
||||
IOFormat m_format;
|
||||
};
|
||||
|
||||
namespace internal {
|
||||
|
||||
// NOTE: This helper is kept for backward compatibility with previous code specializing
|
||||
// this internal::significant_decimals_impl structure. In the future we should directly
|
||||
// call digits10() which has been introduced in July 2016 in 3.3.
|
||||
template<typename Scalar>
|
||||
struct significant_decimals_impl
|
||||
{
|
||||
static inline int run()
|
||||
{
|
||||
return NumTraits<Scalar>::digits10();
|
||||
}
|
||||
};
|
||||
|
||||
/** \internal
|
||||
* print the matrix \a _m to the output stream \a s using the output format \a fmt */
|
||||
template<typename Derived>
|
||||
std::ostream & print_matrix(std::ostream & s, const Derived& _m, const IOFormat& fmt)
|
||||
{
|
||||
using internal::is_same;
|
||||
using internal::conditional;
|
||||
|
||||
if(_m.size() == 0)
|
||||
{
|
||||
s << fmt.matPrefix << fmt.matSuffix;
|
||||
return s;
|
||||
}
|
||||
|
||||
typename Derived::Nested m = _m;
|
||||
typedef typename Derived::Scalar Scalar;
|
||||
typedef typename
|
||||
conditional<
|
||||
is_same<Scalar, char>::value ||
|
||||
is_same<Scalar, unsigned char>::value ||
|
||||
is_same<Scalar, numext::int8_t>::value ||
|
||||
is_same<Scalar, numext::uint8_t>::value,
|
||||
int,
|
||||
typename conditional<
|
||||
is_same<Scalar, std::complex<char> >::value ||
|
||||
is_same<Scalar, std::complex<unsigned char> >::value ||
|
||||
is_same<Scalar, std::complex<numext::int8_t> >::value ||
|
||||
is_same<Scalar, std::complex<numext::uint8_t> >::value,
|
||||
std::complex<int>,
|
||||
const Scalar&
|
||||
>::type
|
||||
>::type PrintType;
|
||||
|
||||
Index width = 0;
|
||||
|
||||
std::streamsize explicit_precision;
|
||||
if(fmt.precision == StreamPrecision)
|
||||
{
|
||||
explicit_precision = 0;
|
||||
}
|
||||
else if(fmt.precision == FullPrecision)
|
||||
{
|
||||
if (NumTraits<Scalar>::IsInteger)
|
||||
{
|
||||
explicit_precision = 0;
|
||||
}
|
||||
else
|
||||
{
|
||||
explicit_precision = significant_decimals_impl<Scalar>::run();
|
||||
}
|
||||
}
|
||||
else
|
||||
{
|
||||
explicit_precision = fmt.precision;
|
||||
}
|
||||
|
||||
std::streamsize old_precision = 0;
|
||||
if(explicit_precision) old_precision = s.precision(explicit_precision);
|
||||
|
||||
bool align_cols = !(fmt.flags & DontAlignCols);
|
||||
if(align_cols)
|
||||
{
|
||||
// compute the largest width
|
||||
for(Index j = 0; j < m.cols(); ++j)
|
||||
for(Index i = 0; i < m.rows(); ++i)
|
||||
{
|
||||
std::stringstream sstr;
|
||||
sstr.copyfmt(s);
|
||||
sstr << static_cast<PrintType>(m.coeff(i,j));
|
||||
width = std::max<Index>(width, Index(sstr.str().length()));
|
||||
}
|
||||
}
|
||||
std::streamsize old_width = s.width();
|
||||
char old_fill_character = s.fill();
|
||||
s << fmt.matPrefix;
|
||||
for(Index i = 0; i < m.rows(); ++i)
|
||||
{
|
||||
if (i)
|
||||
s << fmt.rowSpacer;
|
||||
s << fmt.rowPrefix;
|
||||
if(width) {
|
||||
s.fill(fmt.fill);
|
||||
s.width(width);
|
||||
}
|
||||
s << static_cast<PrintType>(m.coeff(i, 0));
|
||||
for(Index j = 1; j < m.cols(); ++j)
|
||||
{
|
||||
s << fmt.coeffSeparator;
|
||||
if(width) {
|
||||
s.fill(fmt.fill);
|
||||
s.width(width);
|
||||
}
|
||||
s << static_cast<PrintType>(m.coeff(i, j));
|
||||
}
|
||||
s << fmt.rowSuffix;
|
||||
if( i < m.rows() - 1)
|
||||
s << fmt.rowSeparator;
|
||||
}
|
||||
s << fmt.matSuffix;
|
||||
if(explicit_precision) s.precision(old_precision);
|
||||
if(width) {
|
||||
s.fill(old_fill_character);
|
||||
s.width(old_width);
|
||||
}
|
||||
return s;
|
||||
}
|
||||
|
||||
} // end namespace internal
|
||||
|
||||
/** \relates DenseBase
|
||||
*
|
||||
* Outputs the matrix, to the given stream.
|
||||
*
|
||||
* If you wish to print the matrix with a format different than the default, use DenseBase::format().
|
||||
*
|
||||
* It is also possible to change the default format by defining EIGEN_DEFAULT_IO_FORMAT before including Eigen headers.
|
||||
* If not defined, this will automatically be defined to Eigen::IOFormat(), that is the Eigen::IOFormat with default parameters.
|
||||
*
|
||||
* \sa DenseBase::format()
|
||||
*/
|
||||
template<typename Derived>
|
||||
std::ostream & operator <<
|
||||
(std::ostream & s,
|
||||
const DenseBase<Derived> & m)
|
||||
{
|
||||
return internal::print_matrix(s, m.eval(), EIGEN_DEFAULT_IO_FORMAT);
|
||||
}
|
||||
|
||||
} // end namespace Eigen
|
||||
|
||||
#endif // EIGEN_IO_H
|
||||
@@ -0,0 +1,237 @@
|
||||
// This file is part of Eigen, a lightweight C++ template library
|
||||
// for linear algebra.
|
||||
//
|
||||
// Copyright (C) 2017 Gael Guennebaud <gael.guennebaud@inria.fr>
|
||||
//
|
||||
// This Source Code Form is subject to the terms of the Mozilla
|
||||
// Public License v. 2.0. If a copy of the MPL was not distributed
|
||||
// with this file, You can obtain one at http://mozilla.org/MPL/2.0/.
|
||||
|
||||
#ifndef EIGEN_INDEXED_VIEW_H
|
||||
#define EIGEN_INDEXED_VIEW_H
|
||||
|
||||
namespace Eigen {
|
||||
|
||||
namespace internal {
|
||||
|
||||
template<typename XprType, typename RowIndices, typename ColIndices>
|
||||
struct traits<IndexedView<XprType, RowIndices, ColIndices> >
|
||||
: traits<XprType>
|
||||
{
|
||||
enum {
|
||||
RowsAtCompileTime = int(array_size<RowIndices>::value),
|
||||
ColsAtCompileTime = int(array_size<ColIndices>::value),
|
||||
MaxRowsAtCompileTime = RowsAtCompileTime != Dynamic ? int(RowsAtCompileTime) : Dynamic,
|
||||
MaxColsAtCompileTime = ColsAtCompileTime != Dynamic ? int(ColsAtCompileTime) : Dynamic,
|
||||
|
||||
XprTypeIsRowMajor = (int(traits<XprType>::Flags)&RowMajorBit) != 0,
|
||||
IsRowMajor = (MaxRowsAtCompileTime==1&&MaxColsAtCompileTime!=1) ? 1
|
||||
: (MaxColsAtCompileTime==1&&MaxRowsAtCompileTime!=1) ? 0
|
||||
: XprTypeIsRowMajor,
|
||||
|
||||
RowIncr = int(get_compile_time_incr<RowIndices>::value),
|
||||
ColIncr = int(get_compile_time_incr<ColIndices>::value),
|
||||
InnerIncr = IsRowMajor ? ColIncr : RowIncr,
|
||||
OuterIncr = IsRowMajor ? RowIncr : ColIncr,
|
||||
|
||||
HasSameStorageOrderAsXprType = (IsRowMajor == XprTypeIsRowMajor),
|
||||
XprInnerStride = HasSameStorageOrderAsXprType ? int(inner_stride_at_compile_time<XprType>::ret) : int(outer_stride_at_compile_time<XprType>::ret),
|
||||
XprOuterstride = HasSameStorageOrderAsXprType ? int(outer_stride_at_compile_time<XprType>::ret) : int(inner_stride_at_compile_time<XprType>::ret),
|
||||
|
||||
InnerSize = XprTypeIsRowMajor ? ColsAtCompileTime : RowsAtCompileTime,
|
||||
IsBlockAlike = InnerIncr==1 && OuterIncr==1,
|
||||
IsInnerPannel = HasSameStorageOrderAsXprType && is_same<AllRange<InnerSize>,typename conditional<XprTypeIsRowMajor,ColIndices,RowIndices>::type>::value,
|
||||
|
||||
InnerStrideAtCompileTime = InnerIncr<0 || InnerIncr==DynamicIndex || XprInnerStride==Dynamic ? Dynamic : XprInnerStride * InnerIncr,
|
||||
OuterStrideAtCompileTime = OuterIncr<0 || OuterIncr==DynamicIndex || XprOuterstride==Dynamic ? Dynamic : XprOuterstride * OuterIncr,
|
||||
|
||||
ReturnAsScalar = is_same<RowIndices,SingleRange>::value && is_same<ColIndices,SingleRange>::value,
|
||||
ReturnAsBlock = (!ReturnAsScalar) && IsBlockAlike,
|
||||
ReturnAsIndexedView = (!ReturnAsScalar) && (!ReturnAsBlock),
|
||||
|
||||
// FIXME we deal with compile-time strides if and only if we have DirectAccessBit flag,
|
||||
// but this is too strict regarding negative strides...
|
||||
DirectAccessMask = (int(InnerIncr)!=UndefinedIncr && int(OuterIncr)!=UndefinedIncr && InnerIncr>=0 && OuterIncr>=0) ? DirectAccessBit : 0,
|
||||
FlagsRowMajorBit = IsRowMajor ? RowMajorBit : 0,
|
||||
FlagsLvalueBit = is_lvalue<XprType>::value ? LvalueBit : 0,
|
||||
FlagsLinearAccessBit = (RowsAtCompileTime == 1 || ColsAtCompileTime == 1) ? LinearAccessBit : 0,
|
||||
Flags = (traits<XprType>::Flags & (HereditaryBits | DirectAccessMask )) | FlagsLvalueBit | FlagsRowMajorBit | FlagsLinearAccessBit
|
||||
};
|
||||
|
||||
typedef Block<XprType,RowsAtCompileTime,ColsAtCompileTime,IsInnerPannel> BlockType;
|
||||
};
|
||||
|
||||
}
|
||||
|
||||
template<typename XprType, typename RowIndices, typename ColIndices, typename StorageKind>
|
||||
class IndexedViewImpl;
|
||||
|
||||
|
||||
/** \class IndexedView
|
||||
* \ingroup Core_Module
|
||||
*
|
||||
* \brief Expression of a non-sequential sub-matrix defined by arbitrary sequences of row and column indices
|
||||
*
|
||||
* \tparam XprType the type of the expression in which we are taking the intersections of sub-rows and sub-columns
|
||||
* \tparam RowIndices the type of the object defining the sequence of row indices
|
||||
* \tparam ColIndices the type of the object defining the sequence of column indices
|
||||
*
|
||||
* This class represents an expression of a sub-matrix (or sub-vector) defined as the intersection
|
||||
* of sub-sets of rows and columns, that are themself defined by generic sequences of row indices \f$ \{r_0,r_1,..r_{m-1}\} \f$
|
||||
* and column indices \f$ \{c_0,c_1,..c_{n-1} \}\f$. Let \f$ A \f$ be the nested matrix, then the resulting matrix \f$ B \f$ has \c m
|
||||
* rows and \c n columns, and its entries are given by: \f$ B(i,j) = A(r_i,c_j) \f$.
|
||||
*
|
||||
* The \c RowIndices and \c ColIndices types must be compatible with the following API:
|
||||
* \code
|
||||
* <integral type> operator[](Index) const;
|
||||
* Index size() const;
|
||||
* \endcode
|
||||
*
|
||||
* Typical supported types thus include:
|
||||
* - std::vector<int>
|
||||
* - std::valarray<int>
|
||||
* - std::array<int>
|
||||
* - Plain C arrays: int[N]
|
||||
* - Eigen::ArrayXi
|
||||
* - decltype(ArrayXi::LinSpaced(...))
|
||||
* - Any view/expressions of the previous types
|
||||
* - Eigen::ArithmeticSequence
|
||||
* - Eigen::internal::AllRange (helper for Eigen::all)
|
||||
* - Eigen::internal::SingleRange (helper for single index)
|
||||
* - etc.
|
||||
*
|
||||
* In typical usages of %Eigen, this class should never be used directly. It is the return type of
|
||||
* DenseBase::operator()(const RowIndices&, const ColIndices&).
|
||||
*
|
||||
* \sa class Block
|
||||
*/
|
||||
template<typename XprType, typename RowIndices, typename ColIndices>
|
||||
class IndexedView : public IndexedViewImpl<XprType, RowIndices, ColIndices, typename internal::traits<XprType>::StorageKind>
|
||||
{
|
||||
public:
|
||||
typedef typename IndexedViewImpl<XprType, RowIndices, ColIndices, typename internal::traits<XprType>::StorageKind>::Base Base;
|
||||
EIGEN_GENERIC_PUBLIC_INTERFACE(IndexedView)
|
||||
EIGEN_INHERIT_ASSIGNMENT_OPERATORS(IndexedView)
|
||||
|
||||
typedef typename internal::ref_selector<XprType>::non_const_type MatrixTypeNested;
|
||||
typedef typename internal::remove_all<XprType>::type NestedExpression;
|
||||
|
||||
template<typename T0, typename T1>
|
||||
IndexedView(XprType& xpr, const T0& rowIndices, const T1& colIndices)
|
||||
: m_xpr(xpr), m_rowIndices(rowIndices), m_colIndices(colIndices)
|
||||
{}
|
||||
|
||||
/** \returns number of rows */
|
||||
Index rows() const { return internal::size(m_rowIndices); }
|
||||
|
||||
/** \returns number of columns */
|
||||
Index cols() const { return internal::size(m_colIndices); }
|
||||
|
||||
/** \returns the nested expression */
|
||||
const typename internal::remove_all<XprType>::type&
|
||||
nestedExpression() const { return m_xpr; }
|
||||
|
||||
/** \returns the nested expression */
|
||||
typename internal::remove_reference<XprType>::type&
|
||||
nestedExpression() { return m_xpr; }
|
||||
|
||||
/** \returns a const reference to the object storing/generating the row indices */
|
||||
const RowIndices& rowIndices() const { return m_rowIndices; }
|
||||
|
||||
/** \returns a const reference to the object storing/generating the column indices */
|
||||
const ColIndices& colIndices() const { return m_colIndices; }
|
||||
|
||||
protected:
|
||||
MatrixTypeNested m_xpr;
|
||||
RowIndices m_rowIndices;
|
||||
ColIndices m_colIndices;
|
||||
};
|
||||
|
||||
|
||||
// Generic API dispatcher
|
||||
template<typename XprType, typename RowIndices, typename ColIndices, typename StorageKind>
|
||||
class IndexedViewImpl
|
||||
: public internal::generic_xpr_base<IndexedView<XprType, RowIndices, ColIndices> >::type
|
||||
{
|
||||
public:
|
||||
typedef typename internal::generic_xpr_base<IndexedView<XprType, RowIndices, ColIndices> >::type Base;
|
||||
};
|
||||
|
||||
namespace internal {
|
||||
|
||||
|
||||
template<typename ArgType, typename RowIndices, typename ColIndices>
|
||||
struct unary_evaluator<IndexedView<ArgType, RowIndices, ColIndices>, IndexBased>
|
||||
: evaluator_base<IndexedView<ArgType, RowIndices, ColIndices> >
|
||||
{
|
||||
typedef IndexedView<ArgType, RowIndices, ColIndices> XprType;
|
||||
|
||||
enum {
|
||||
CoeffReadCost = evaluator<ArgType>::CoeffReadCost /* TODO + cost of row/col index */,
|
||||
|
||||
FlagsLinearAccessBit = (traits<XprType>::RowsAtCompileTime == 1 || traits<XprType>::ColsAtCompileTime == 1) ? LinearAccessBit : 0,
|
||||
|
||||
FlagsRowMajorBit = traits<XprType>::FlagsRowMajorBit,
|
||||
|
||||
Flags = (evaluator<ArgType>::Flags & (HereditaryBits & ~RowMajorBit /*| LinearAccessBit | DirectAccessBit*/)) | FlagsLinearAccessBit | FlagsRowMajorBit,
|
||||
|
||||
Alignment = 0
|
||||
};
|
||||
|
||||
EIGEN_DEVICE_FUNC explicit unary_evaluator(const XprType& xpr) : m_argImpl(xpr.nestedExpression()), m_xpr(xpr)
|
||||
{
|
||||
EIGEN_INTERNAL_CHECK_COST_VALUE(CoeffReadCost);
|
||||
}
|
||||
|
||||
typedef typename XprType::Scalar Scalar;
|
||||
typedef typename XprType::CoeffReturnType CoeffReturnType;
|
||||
|
||||
EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
|
||||
CoeffReturnType coeff(Index row, Index col) const
|
||||
{
|
||||
return m_argImpl.coeff(m_xpr.rowIndices()[row], m_xpr.colIndices()[col]);
|
||||
}
|
||||
|
||||
EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
|
||||
Scalar& coeffRef(Index row, Index col)
|
||||
{
|
||||
return m_argImpl.coeffRef(m_xpr.rowIndices()[row], m_xpr.colIndices()[col]);
|
||||
}
|
||||
|
||||
EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
|
||||
Scalar& coeffRef(Index index)
|
||||
{
|
||||
EIGEN_STATIC_ASSERT_LVALUE(XprType)
|
||||
Index row = XprType::RowsAtCompileTime == 1 ? 0 : index;
|
||||
Index col = XprType::RowsAtCompileTime == 1 ? index : 0;
|
||||
return m_argImpl.coeffRef( m_xpr.rowIndices()[row], m_xpr.colIndices()[col]);
|
||||
}
|
||||
|
||||
EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
|
||||
const Scalar& coeffRef(Index index) const
|
||||
{
|
||||
Index row = XprType::RowsAtCompileTime == 1 ? 0 : index;
|
||||
Index col = XprType::RowsAtCompileTime == 1 ? index : 0;
|
||||
return m_argImpl.coeffRef( m_xpr.rowIndices()[row], m_xpr.colIndices()[col]);
|
||||
}
|
||||
|
||||
EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
|
||||
const CoeffReturnType coeff(Index index) const
|
||||
{
|
||||
Index row = XprType::RowsAtCompileTime == 1 ? 0 : index;
|
||||
Index col = XprType::RowsAtCompileTime == 1 ? index : 0;
|
||||
return m_argImpl.coeff( m_xpr.rowIndices()[row], m_xpr.colIndices()[col]);
|
||||
}
|
||||
|
||||
protected:
|
||||
|
||||
evaluator<ArgType> m_argImpl;
|
||||
const XprType& m_xpr;
|
||||
|
||||
};
|
||||
|
||||
} // end namespace internal
|
||||
|
||||
} // end namespace Eigen
|
||||
|
||||
#endif // EIGEN_INDEXED_VIEW_H
|
||||
@@ -0,0 +1,117 @@
|
||||
// This file is part of Eigen, a lightweight C++ template library
|
||||
// for linear algebra.
|
||||
//
|
||||
// Copyright (C) 2014-2019 Gael Guennebaud <gael.guennebaud@inria.fr>
|
||||
//
|
||||
// This Source Code Form is subject to the terms of the Mozilla
|
||||
// Public License v. 2.0. If a copy of the MPL was not distributed
|
||||
// with this file, You can obtain one at http://mozilla.org/MPL/2.0/.
|
||||
|
||||
#ifndef EIGEN_INVERSE_H
|
||||
#define EIGEN_INVERSE_H
|
||||
|
||||
namespace Eigen {
|
||||
|
||||
template<typename XprType,typename StorageKind> class InverseImpl;
|
||||
|
||||
namespace internal {
|
||||
|
||||
template<typename XprType>
|
||||
struct traits<Inverse<XprType> >
|
||||
: traits<typename XprType::PlainObject>
|
||||
{
|
||||
typedef typename XprType::PlainObject PlainObject;
|
||||
typedef traits<PlainObject> BaseTraits;
|
||||
enum {
|
||||
Flags = BaseTraits::Flags & RowMajorBit
|
||||
};
|
||||
};
|
||||
|
||||
} // end namespace internal
|
||||
|
||||
/** \class Inverse
|
||||
*
|
||||
* \brief Expression of the inverse of another expression
|
||||
*
|
||||
* \tparam XprType the type of the expression we are taking the inverse
|
||||
*
|
||||
* This class represents an abstract expression of A.inverse()
|
||||
* and most of the time this is the only way it is used.
|
||||
*
|
||||
*/
|
||||
template<typename XprType>
|
||||
class Inverse : public InverseImpl<XprType,typename internal::traits<XprType>::StorageKind>
|
||||
{
|
||||
public:
|
||||
typedef typename XprType::StorageIndex StorageIndex;
|
||||
typedef typename XprType::Scalar Scalar;
|
||||
typedef typename internal::ref_selector<XprType>::type XprTypeNested;
|
||||
typedef typename internal::remove_all<XprTypeNested>::type XprTypeNestedCleaned;
|
||||
typedef typename internal::ref_selector<Inverse>::type Nested;
|
||||
typedef typename internal::remove_all<XprType>::type NestedExpression;
|
||||
|
||||
explicit EIGEN_DEVICE_FUNC Inverse(const XprType &xpr)
|
||||
: m_xpr(xpr)
|
||||
{}
|
||||
|
||||
EIGEN_DEVICE_FUNC EIGEN_CONSTEXPR Index rows() const EIGEN_NOEXCEPT { return m_xpr.cols(); }
|
||||
EIGEN_DEVICE_FUNC EIGEN_CONSTEXPR Index cols() const EIGEN_NOEXCEPT { return m_xpr.rows(); }
|
||||
|
||||
EIGEN_DEVICE_FUNC const XprTypeNestedCleaned& nestedExpression() const { return m_xpr; }
|
||||
|
||||
protected:
|
||||
XprTypeNested m_xpr;
|
||||
};
|
||||
|
||||
// Generic API dispatcher
|
||||
template<typename XprType, typename StorageKind>
|
||||
class InverseImpl
|
||||
: public internal::generic_xpr_base<Inverse<XprType> >::type
|
||||
{
|
||||
public:
|
||||
typedef typename internal::generic_xpr_base<Inverse<XprType> >::type Base;
|
||||
typedef typename XprType::Scalar Scalar;
|
||||
private:
|
||||
|
||||
Scalar coeff(Index row, Index col) const;
|
||||
Scalar coeff(Index i) const;
|
||||
};
|
||||
|
||||
namespace internal {
|
||||
|
||||
/** \internal
|
||||
* \brief Default evaluator for Inverse expression.
|
||||
*
|
||||
* This default evaluator for Inverse expression simply evaluate the inverse into a temporary
|
||||
* by a call to internal::call_assignment_no_alias.
|
||||
* Therefore, inverse implementers only have to specialize Assignment<Dst,Inverse<...>, ...> for
|
||||
* there own nested expression.
|
||||
*
|
||||
* \sa class Inverse
|
||||
*/
|
||||
template<typename ArgType>
|
||||
struct unary_evaluator<Inverse<ArgType> >
|
||||
: public evaluator<typename Inverse<ArgType>::PlainObject>
|
||||
{
|
||||
typedef Inverse<ArgType> InverseType;
|
||||
typedef typename InverseType::PlainObject PlainObject;
|
||||
typedef evaluator<PlainObject> Base;
|
||||
|
||||
enum { Flags = Base::Flags | EvalBeforeNestingBit };
|
||||
|
||||
unary_evaluator(const InverseType& inv_xpr)
|
||||
: m_result(inv_xpr.rows(), inv_xpr.cols())
|
||||
{
|
||||
::new (static_cast<Base*>(this)) Base(m_result);
|
||||
internal::call_assignment_no_alias(m_result, inv_xpr);
|
||||
}
|
||||
|
||||
protected:
|
||||
PlainObject m_result;
|
||||
};
|
||||
|
||||
} // end namespace internal
|
||||
|
||||
} // end namespace Eigen
|
||||
|
||||
#endif // EIGEN_INVERSE_H
|
||||
@@ -0,0 +1,171 @@
|
||||
// This file is part of Eigen, a lightweight C++ template library
|
||||
// for linear algebra.
|
||||
//
|
||||
// Copyright (C) 2007-2010 Benoit Jacob <jacob.benoit.1@gmail.com>
|
||||
// Copyright (C) 2008 Gael Guennebaud <gael.guennebaud@inria.fr>
|
||||
//
|
||||
// This Source Code Form is subject to the terms of the Mozilla
|
||||
// Public License v. 2.0. If a copy of the MPL was not distributed
|
||||
// with this file, You can obtain one at http://mozilla.org/MPL/2.0/.
|
||||
|
||||
#ifndef EIGEN_MAP_H
|
||||
#define EIGEN_MAP_H
|
||||
|
||||
namespace Eigen {
|
||||
|
||||
namespace internal {
|
||||
template<typename PlainObjectType, int MapOptions, typename StrideType>
|
||||
struct traits<Map<PlainObjectType, MapOptions, StrideType> >
|
||||
: public traits<PlainObjectType>
|
||||
{
|
||||
typedef traits<PlainObjectType> TraitsBase;
|
||||
enum {
|
||||
PlainObjectTypeInnerSize = ((traits<PlainObjectType>::Flags&RowMajorBit)==RowMajorBit)
|
||||
? PlainObjectType::ColsAtCompileTime
|
||||
: PlainObjectType::RowsAtCompileTime,
|
||||
|
||||
InnerStrideAtCompileTime = StrideType::InnerStrideAtCompileTime == 0
|
||||
? int(PlainObjectType::InnerStrideAtCompileTime)
|
||||
: int(StrideType::InnerStrideAtCompileTime),
|
||||
OuterStrideAtCompileTime = StrideType::OuterStrideAtCompileTime == 0
|
||||
? (InnerStrideAtCompileTime==Dynamic || PlainObjectTypeInnerSize==Dynamic
|
||||
? Dynamic
|
||||
: int(InnerStrideAtCompileTime) * int(PlainObjectTypeInnerSize))
|
||||
: int(StrideType::OuterStrideAtCompileTime),
|
||||
Alignment = int(MapOptions)&int(AlignedMask),
|
||||
Flags0 = TraitsBase::Flags & (~NestByRefBit),
|
||||
Flags = is_lvalue<PlainObjectType>::value ? int(Flags0) : (int(Flags0) & ~LvalueBit)
|
||||
};
|
||||
private:
|
||||
enum { Options }; // Expressions don't have Options
|
||||
};
|
||||
}
|
||||
|
||||
/** \class Map
|
||||
* \ingroup Core_Module
|
||||
*
|
||||
* \brief A matrix or vector expression mapping an existing array of data.
|
||||
*
|
||||
* \tparam PlainObjectType the equivalent matrix type of the mapped data
|
||||
* \tparam MapOptions specifies the pointer alignment in bytes. It can be: \c #Aligned128, \c #Aligned64, \c #Aligned32, \c #Aligned16, \c #Aligned8 or \c #Unaligned.
|
||||
* The default is \c #Unaligned.
|
||||
* \tparam StrideType optionally specifies strides. By default, Map assumes the memory layout
|
||||
* of an ordinary, contiguous array. This can be overridden by specifying strides.
|
||||
* The type passed here must be a specialization of the Stride template, see examples below.
|
||||
*
|
||||
* This class represents a matrix or vector expression mapping an existing array of data.
|
||||
* It can be used to let Eigen interface without any overhead with non-Eigen data structures,
|
||||
* such as plain C arrays or structures from other libraries. By default, it assumes that the
|
||||
* data is laid out contiguously in memory. You can however override this by explicitly specifying
|
||||
* inner and outer strides.
|
||||
*
|
||||
* Here's an example of simply mapping a contiguous array as a \ref TopicStorageOrders "column-major" matrix:
|
||||
* \include Map_simple.cpp
|
||||
* Output: \verbinclude Map_simple.out
|
||||
*
|
||||
* If you need to map non-contiguous arrays, you can do so by specifying strides:
|
||||
*
|
||||
* Here's an example of mapping an array as a vector, specifying an inner stride, that is, the pointer
|
||||
* increment between two consecutive coefficients. Here, we're specifying the inner stride as a compile-time
|
||||
* fixed value.
|
||||
* \include Map_inner_stride.cpp
|
||||
* Output: \verbinclude Map_inner_stride.out
|
||||
*
|
||||
* Here's an example of mapping an array while specifying an outer stride. Here, since we're mapping
|
||||
* as a column-major matrix, 'outer stride' means the pointer increment between two consecutive columns.
|
||||
* Here, we're specifying the outer stride as a runtime parameter. Note that here \c OuterStride<> is
|
||||
* a short version of \c OuterStride<Dynamic> because the default template parameter of OuterStride
|
||||
* is \c Dynamic
|
||||
* \include Map_outer_stride.cpp
|
||||
* Output: \verbinclude Map_outer_stride.out
|
||||
*
|
||||
* For more details and for an example of specifying both an inner and an outer stride, see class Stride.
|
||||
*
|
||||
* \b Tip: to change the array of data mapped by a Map object, you can use the C++
|
||||
* placement new syntax:
|
||||
*
|
||||
* Example: \include Map_placement_new.cpp
|
||||
* Output: \verbinclude Map_placement_new.out
|
||||
*
|
||||
* This class is the return type of PlainObjectBase::Map() but can also be used directly.
|
||||
*
|
||||
* \sa PlainObjectBase::Map(), \ref TopicStorageOrders
|
||||
*/
|
||||
template<typename PlainObjectType, int MapOptions, typename StrideType> class Map
|
||||
: public MapBase<Map<PlainObjectType, MapOptions, StrideType> >
|
||||
{
|
||||
public:
|
||||
|
||||
typedef MapBase<Map> Base;
|
||||
EIGEN_DENSE_PUBLIC_INTERFACE(Map)
|
||||
|
||||
typedef typename Base::PointerType PointerType;
|
||||
typedef PointerType PointerArgType;
|
||||
EIGEN_DEVICE_FUNC
|
||||
inline PointerType cast_to_pointer_type(PointerArgType ptr) { return ptr; }
|
||||
|
||||
EIGEN_DEVICE_FUNC EIGEN_CONSTEXPR
|
||||
inline Index innerStride() const
|
||||
{
|
||||
return StrideType::InnerStrideAtCompileTime != 0 ? m_stride.inner() : 1;
|
||||
}
|
||||
|
||||
EIGEN_DEVICE_FUNC EIGEN_CONSTEXPR
|
||||
inline Index outerStride() const
|
||||
{
|
||||
return StrideType::OuterStrideAtCompileTime != 0 ? m_stride.outer()
|
||||
: internal::traits<Map>::OuterStrideAtCompileTime != Dynamic ? Index(internal::traits<Map>::OuterStrideAtCompileTime)
|
||||
: IsVectorAtCompileTime ? (this->size() * innerStride())
|
||||
: int(Flags)&RowMajorBit ? (this->cols() * innerStride())
|
||||
: (this->rows() * innerStride());
|
||||
}
|
||||
|
||||
/** Constructor in the fixed-size case.
|
||||
*
|
||||
* \param dataPtr pointer to the array to map
|
||||
* \param stride optional Stride object, passing the strides.
|
||||
*/
|
||||
EIGEN_DEVICE_FUNC
|
||||
explicit inline Map(PointerArgType dataPtr, const StrideType& stride = StrideType())
|
||||
: Base(cast_to_pointer_type(dataPtr)), m_stride(stride)
|
||||
{
|
||||
PlainObjectType::Base::_check_template_params();
|
||||
}
|
||||
|
||||
/** Constructor in the dynamic-size vector case.
|
||||
*
|
||||
* \param dataPtr pointer to the array to map
|
||||
* \param size the size of the vector expression
|
||||
* \param stride optional Stride object, passing the strides.
|
||||
*/
|
||||
EIGEN_DEVICE_FUNC
|
||||
inline Map(PointerArgType dataPtr, Index size, const StrideType& stride = StrideType())
|
||||
: Base(cast_to_pointer_type(dataPtr), size), m_stride(stride)
|
||||
{
|
||||
PlainObjectType::Base::_check_template_params();
|
||||
}
|
||||
|
||||
/** Constructor in the dynamic-size matrix case.
|
||||
*
|
||||
* \param dataPtr pointer to the array to map
|
||||
* \param rows the number of rows of the matrix expression
|
||||
* \param cols the number of columns of the matrix expression
|
||||
* \param stride optional Stride object, passing the strides.
|
||||
*/
|
||||
EIGEN_DEVICE_FUNC
|
||||
inline Map(PointerArgType dataPtr, Index rows, Index cols, const StrideType& stride = StrideType())
|
||||
: Base(cast_to_pointer_type(dataPtr), rows, cols), m_stride(stride)
|
||||
{
|
||||
PlainObjectType::Base::_check_template_params();
|
||||
}
|
||||
|
||||
EIGEN_INHERIT_ASSIGNMENT_OPERATORS(Map)
|
||||
|
||||
protected:
|
||||
StrideType m_stride;
|
||||
};
|
||||
|
||||
|
||||
} // end namespace Eigen
|
||||
|
||||
#endif // EIGEN_MAP_H
|
||||
@@ -0,0 +1,310 @@
|
||||
// This file is part of Eigen, a lightweight C++ template library
|
||||
// for linear algebra.
|
||||
//
|
||||
// Copyright (C) 2007-2010 Benoit Jacob <jacob.benoit.1@gmail.com>
|
||||
// Copyright (C) 2008 Gael Guennebaud <gael.guennebaud@inria.fr>
|
||||
//
|
||||
// This Source Code Form is subject to the terms of the Mozilla
|
||||
// Public License v. 2.0. If a copy of the MPL was not distributed
|
||||
// with this file, You can obtain one at http://mozilla.org/MPL/2.0/.
|
||||
|
||||
#ifndef EIGEN_MAPBASE_H
|
||||
#define EIGEN_MAPBASE_H
|
||||
|
||||
#define EIGEN_STATIC_ASSERT_INDEX_BASED_ACCESS(Derived) \
|
||||
EIGEN_STATIC_ASSERT((int(internal::evaluator<Derived>::Flags) & LinearAccessBit) || Derived::IsVectorAtCompileTime, \
|
||||
YOU_ARE_TRYING_TO_USE_AN_INDEX_BASED_ACCESSOR_ON_AN_EXPRESSION_THAT_DOES_NOT_SUPPORT_THAT)
|
||||
|
||||
namespace Eigen {
|
||||
|
||||
/** \ingroup Core_Module
|
||||
*
|
||||
* \brief Base class for dense Map and Block expression with direct access
|
||||
*
|
||||
* This base class provides the const low-level accessors (e.g. coeff, coeffRef) of dense
|
||||
* Map and Block objects with direct access.
|
||||
* Typical users do not have to directly deal with this class.
|
||||
*
|
||||
* This class can be extended by through the macro plugin \c EIGEN_MAPBASE_PLUGIN.
|
||||
* See \link TopicCustomizing_Plugins customizing Eigen \endlink for details.
|
||||
*
|
||||
* The \c Derived class has to provide the following two methods describing the memory layout:
|
||||
* \code Index innerStride() const; \endcode
|
||||
* \code Index outerStride() const; \endcode
|
||||
*
|
||||
* \sa class Map, class Block
|
||||
*/
|
||||
template<typename Derived> class MapBase<Derived, ReadOnlyAccessors>
|
||||
: public internal::dense_xpr_base<Derived>::type
|
||||
{
|
||||
public:
|
||||
|
||||
typedef typename internal::dense_xpr_base<Derived>::type Base;
|
||||
enum {
|
||||
RowsAtCompileTime = internal::traits<Derived>::RowsAtCompileTime,
|
||||
ColsAtCompileTime = internal::traits<Derived>::ColsAtCompileTime,
|
||||
InnerStrideAtCompileTime = internal::traits<Derived>::InnerStrideAtCompileTime,
|
||||
SizeAtCompileTime = Base::SizeAtCompileTime
|
||||
};
|
||||
|
||||
typedef typename internal::traits<Derived>::StorageKind StorageKind;
|
||||
typedef typename internal::traits<Derived>::Scalar Scalar;
|
||||
typedef typename internal::packet_traits<Scalar>::type PacketScalar;
|
||||
typedef typename NumTraits<Scalar>::Real RealScalar;
|
||||
typedef typename internal::conditional<
|
||||
bool(internal::is_lvalue<Derived>::value),
|
||||
Scalar *,
|
||||
const Scalar *>::type
|
||||
PointerType;
|
||||
|
||||
using Base::derived;
|
||||
// using Base::RowsAtCompileTime;
|
||||
// using Base::ColsAtCompileTime;
|
||||
// using Base::SizeAtCompileTime;
|
||||
using Base::MaxRowsAtCompileTime;
|
||||
using Base::MaxColsAtCompileTime;
|
||||
using Base::MaxSizeAtCompileTime;
|
||||
using Base::IsVectorAtCompileTime;
|
||||
using Base::Flags;
|
||||
using Base::IsRowMajor;
|
||||
|
||||
using Base::rows;
|
||||
using Base::cols;
|
||||
using Base::size;
|
||||
using Base::coeff;
|
||||
using Base::coeffRef;
|
||||
using Base::lazyAssign;
|
||||
using Base::eval;
|
||||
|
||||
using Base::innerStride;
|
||||
using Base::outerStride;
|
||||
using Base::rowStride;
|
||||
using Base::colStride;
|
||||
|
||||
// bug 217 - compile error on ICC 11.1
|
||||
using Base::operator=;
|
||||
|
||||
typedef typename Base::CoeffReturnType CoeffReturnType;
|
||||
|
||||
/** \copydoc DenseBase::rows() */
|
||||
EIGEN_DEVICE_FUNC EIGEN_CONSTEXPR
|
||||
inline Index rows() const EIGEN_NOEXCEPT { return m_rows.value(); }
|
||||
/** \copydoc DenseBase::cols() */
|
||||
EIGEN_DEVICE_FUNC EIGEN_CONSTEXPR
|
||||
inline Index cols() const EIGEN_NOEXCEPT { return m_cols.value(); }
|
||||
|
||||
/** Returns a pointer to the first coefficient of the matrix or vector.
|
||||
*
|
||||
* \note When addressing this data, make sure to honor the strides returned by innerStride() and outerStride().
|
||||
*
|
||||
* \sa innerStride(), outerStride()
|
||||
*/
|
||||
EIGEN_DEVICE_FUNC inline const Scalar* data() const { return m_data; }
|
||||
|
||||
/** \copydoc PlainObjectBase::coeff(Index,Index) const */
|
||||
EIGEN_DEVICE_FUNC
|
||||
inline const Scalar& coeff(Index rowId, Index colId) const
|
||||
{
|
||||
return m_data[colId * colStride() + rowId * rowStride()];
|
||||
}
|
||||
|
||||
/** \copydoc PlainObjectBase::coeff(Index) const */
|
||||
EIGEN_DEVICE_FUNC
|
||||
inline const Scalar& coeff(Index index) const
|
||||
{
|
||||
EIGEN_STATIC_ASSERT_INDEX_BASED_ACCESS(Derived)
|
||||
return m_data[index * innerStride()];
|
||||
}
|
||||
|
||||
/** \copydoc PlainObjectBase::coeffRef(Index,Index) const */
|
||||
EIGEN_DEVICE_FUNC
|
||||
inline const Scalar& coeffRef(Index rowId, Index colId) const
|
||||
{
|
||||
return this->m_data[colId * colStride() + rowId * rowStride()];
|
||||
}
|
||||
|
||||
/** \copydoc PlainObjectBase::coeffRef(Index) const */
|
||||
EIGEN_DEVICE_FUNC
|
||||
inline const Scalar& coeffRef(Index index) const
|
||||
{
|
||||
EIGEN_STATIC_ASSERT_INDEX_BASED_ACCESS(Derived)
|
||||
return this->m_data[index * innerStride()];
|
||||
}
|
||||
|
||||
/** \internal */
|
||||
template<int LoadMode>
|
||||
inline PacketScalar packet(Index rowId, Index colId) const
|
||||
{
|
||||
return internal::ploadt<PacketScalar, LoadMode>
|
||||
(m_data + (colId * colStride() + rowId * rowStride()));
|
||||
}
|
||||
|
||||
/** \internal */
|
||||
template<int LoadMode>
|
||||
inline PacketScalar packet(Index index) const
|
||||
{
|
||||
EIGEN_STATIC_ASSERT_INDEX_BASED_ACCESS(Derived)
|
||||
return internal::ploadt<PacketScalar, LoadMode>(m_data + index * innerStride());
|
||||
}
|
||||
|
||||
/** \internal Constructor for fixed size matrices or vectors */
|
||||
EIGEN_DEVICE_FUNC
|
||||
explicit inline MapBase(PointerType dataPtr) : m_data(dataPtr), m_rows(RowsAtCompileTime), m_cols(ColsAtCompileTime)
|
||||
{
|
||||
EIGEN_STATIC_ASSERT_FIXED_SIZE(Derived)
|
||||
checkSanity<Derived>();
|
||||
}
|
||||
|
||||
/** \internal Constructor for dynamically sized vectors */
|
||||
EIGEN_DEVICE_FUNC
|
||||
inline MapBase(PointerType dataPtr, Index vecSize)
|
||||
: m_data(dataPtr),
|
||||
m_rows(RowsAtCompileTime == Dynamic ? vecSize : Index(RowsAtCompileTime)),
|
||||
m_cols(ColsAtCompileTime == Dynamic ? vecSize : Index(ColsAtCompileTime))
|
||||
{
|
||||
EIGEN_STATIC_ASSERT_VECTOR_ONLY(Derived)
|
||||
eigen_assert(vecSize >= 0);
|
||||
eigen_assert(dataPtr == 0 || SizeAtCompileTime == Dynamic || SizeAtCompileTime == vecSize);
|
||||
checkSanity<Derived>();
|
||||
}
|
||||
|
||||
/** \internal Constructor for dynamically sized matrices */
|
||||
EIGEN_DEVICE_FUNC
|
||||
inline MapBase(PointerType dataPtr, Index rows, Index cols)
|
||||
: m_data(dataPtr), m_rows(rows), m_cols(cols)
|
||||
{
|
||||
eigen_assert( (dataPtr == 0)
|
||||
|| ( rows >= 0 && (RowsAtCompileTime == Dynamic || RowsAtCompileTime == rows)
|
||||
&& cols >= 0 && (ColsAtCompileTime == Dynamic || ColsAtCompileTime == cols)));
|
||||
checkSanity<Derived>();
|
||||
}
|
||||
|
||||
#ifdef EIGEN_MAPBASE_PLUGIN
|
||||
#include EIGEN_MAPBASE_PLUGIN
|
||||
#endif
|
||||
|
||||
protected:
|
||||
EIGEN_DEFAULT_COPY_CONSTRUCTOR(MapBase)
|
||||
EIGEN_DEFAULT_EMPTY_CONSTRUCTOR_AND_DESTRUCTOR(MapBase)
|
||||
|
||||
template<typename T>
|
||||
EIGEN_DEVICE_FUNC
|
||||
void checkSanity(typename internal::enable_if<(internal::traits<T>::Alignment>0),void*>::type = 0) const
|
||||
{
|
||||
#if EIGEN_MAX_ALIGN_BYTES>0
|
||||
// innerStride() is not set yet when this function is called, so we optimistically assume the lowest plausible value:
|
||||
const Index minInnerStride = InnerStrideAtCompileTime == Dynamic ? 1 : Index(InnerStrideAtCompileTime);
|
||||
EIGEN_ONLY_USED_FOR_DEBUG(minInnerStride);
|
||||
eigen_assert(( ((internal::UIntPtr(m_data) % internal::traits<Derived>::Alignment) == 0)
|
||||
|| (cols() * rows() * minInnerStride * sizeof(Scalar)) < internal::traits<Derived>::Alignment ) && "data is not aligned");
|
||||
#endif
|
||||
}
|
||||
|
||||
template<typename T>
|
||||
EIGEN_DEVICE_FUNC
|
||||
void checkSanity(typename internal::enable_if<internal::traits<T>::Alignment==0,void*>::type = 0) const
|
||||
{}
|
||||
|
||||
PointerType m_data;
|
||||
const internal::variable_if_dynamic<Index, RowsAtCompileTime> m_rows;
|
||||
const internal::variable_if_dynamic<Index, ColsAtCompileTime> m_cols;
|
||||
};
|
||||
|
||||
/** \ingroup Core_Module
|
||||
*
|
||||
* \brief Base class for non-const dense Map and Block expression with direct access
|
||||
*
|
||||
* This base class provides the non-const low-level accessors (e.g. coeff and coeffRef) of
|
||||
* dense Map and Block objects with direct access.
|
||||
* It inherits MapBase<Derived, ReadOnlyAccessors> which defines the const variant for reading specific entries.
|
||||
*
|
||||
* \sa class Map, class Block
|
||||
*/
|
||||
template<typename Derived> class MapBase<Derived, WriteAccessors>
|
||||
: public MapBase<Derived, ReadOnlyAccessors>
|
||||
{
|
||||
typedef MapBase<Derived, ReadOnlyAccessors> ReadOnlyMapBase;
|
||||
public:
|
||||
|
||||
typedef MapBase<Derived, ReadOnlyAccessors> Base;
|
||||
|
||||
typedef typename Base::Scalar Scalar;
|
||||
typedef typename Base::PacketScalar PacketScalar;
|
||||
typedef typename Base::StorageIndex StorageIndex;
|
||||
typedef typename Base::PointerType PointerType;
|
||||
|
||||
using Base::derived;
|
||||
using Base::rows;
|
||||
using Base::cols;
|
||||
using Base::size;
|
||||
using Base::coeff;
|
||||
using Base::coeffRef;
|
||||
|
||||
using Base::innerStride;
|
||||
using Base::outerStride;
|
||||
using Base::rowStride;
|
||||
using Base::colStride;
|
||||
|
||||
typedef typename internal::conditional<
|
||||
internal::is_lvalue<Derived>::value,
|
||||
Scalar,
|
||||
const Scalar
|
||||
>::type ScalarWithConstIfNotLvalue;
|
||||
|
||||
EIGEN_DEVICE_FUNC
|
||||
inline const Scalar* data() const { return this->m_data; }
|
||||
EIGEN_DEVICE_FUNC
|
||||
inline ScalarWithConstIfNotLvalue* data() { return this->m_data; } // no const-cast here so non-const-correct code will give a compile error
|
||||
|
||||
EIGEN_DEVICE_FUNC
|
||||
inline ScalarWithConstIfNotLvalue& coeffRef(Index row, Index col)
|
||||
{
|
||||
return this->m_data[col * colStride() + row * rowStride()];
|
||||
}
|
||||
|
||||
EIGEN_DEVICE_FUNC
|
||||
inline ScalarWithConstIfNotLvalue& coeffRef(Index index)
|
||||
{
|
||||
EIGEN_STATIC_ASSERT_INDEX_BASED_ACCESS(Derived)
|
||||
return this->m_data[index * innerStride()];
|
||||
}
|
||||
|
||||
template<int StoreMode>
|
||||
inline void writePacket(Index row, Index col, const PacketScalar& val)
|
||||
{
|
||||
internal::pstoret<Scalar, PacketScalar, StoreMode>
|
||||
(this->m_data + (col * colStride() + row * rowStride()), val);
|
||||
}
|
||||
|
||||
template<int StoreMode>
|
||||
inline void writePacket(Index index, const PacketScalar& val)
|
||||
{
|
||||
EIGEN_STATIC_ASSERT_INDEX_BASED_ACCESS(Derived)
|
||||
internal::pstoret<Scalar, PacketScalar, StoreMode>
|
||||
(this->m_data + index * innerStride(), val);
|
||||
}
|
||||
|
||||
EIGEN_DEVICE_FUNC explicit inline MapBase(PointerType dataPtr) : Base(dataPtr) {}
|
||||
EIGEN_DEVICE_FUNC inline MapBase(PointerType dataPtr, Index vecSize) : Base(dataPtr, vecSize) {}
|
||||
EIGEN_DEVICE_FUNC inline MapBase(PointerType dataPtr, Index rows, Index cols) : Base(dataPtr, rows, cols) {}
|
||||
|
||||
EIGEN_DEVICE_FUNC
|
||||
Derived& operator=(const MapBase& other)
|
||||
{
|
||||
ReadOnlyMapBase::Base::operator=(other);
|
||||
return derived();
|
||||
}
|
||||
|
||||
// In theory we could simply refer to Base:Base::operator=, but MSVC does not like Base::Base,
|
||||
// see bugs 821 and 920.
|
||||
using ReadOnlyMapBase::Base::operator=;
|
||||
protected:
|
||||
EIGEN_DEFAULT_COPY_CONSTRUCTOR(MapBase)
|
||||
EIGEN_DEFAULT_EMPTY_CONSTRUCTOR_AND_DESTRUCTOR(MapBase)
|
||||
};
|
||||
|
||||
#undef EIGEN_STATIC_ASSERT_INDEX_BASED_ACCESS
|
||||
|
||||
} // end namespace Eigen
|
||||
|
||||
#endif // EIGEN_MAPBASE_H
|
||||
File diff suppressed because it is too large
Load Diff
@@ -0,0 +1,200 @@
|
||||
// This file is part of Eigen, a lightweight C++ template library
|
||||
// for linear algebra.
|
||||
//
|
||||
// Copyright (C) 2014 Pedro Gonnet (pedro.gonnet@gmail.com)
|
||||
// Copyright (C) 2016 Gael Guennebaud <gael.guennebaud@inria.fr>
|
||||
//
|
||||
// This Source Code Form is subject to the terms of the Mozilla
|
||||
// Public License v. 2.0. If a copy of the MPL was not distributed
|
||||
// with this file, You can obtain one at http://mozilla.org/MPL/2.0/.
|
||||
|
||||
#ifndef EIGEN_MATHFUNCTIONSIMPL_H
|
||||
#define EIGEN_MATHFUNCTIONSIMPL_H
|
||||
|
||||
namespace Eigen {
|
||||
|
||||
namespace internal {
|
||||
|
||||
/** \internal \returns the hyperbolic tan of \a a (coeff-wise)
|
||||
Doesn't do anything fancy, just a 13/6-degree rational interpolant which
|
||||
is accurate up to a couple of ulps in the (approximate) range [-8, 8],
|
||||
outside of which tanh(x) = +/-1 in single precision. The input is clamped
|
||||
to the range [-c, c]. The value c is chosen as the smallest value where
|
||||
the approximation evaluates to exactly 1. In the reange [-0.0004, 0.0004]
|
||||
the approxmation tanh(x) ~= x is used for better accuracy as x tends to zero.
|
||||
|
||||
This implementation works on both scalars and packets.
|
||||
*/
|
||||
template<typename T>
|
||||
T generic_fast_tanh_float(const T& a_x)
|
||||
{
|
||||
// Clamp the inputs to the range [-c, c]
|
||||
#ifdef EIGEN_VECTORIZE_FMA
|
||||
const T plus_clamp = pset1<T>(7.99881172180175781f);
|
||||
const T minus_clamp = pset1<T>(-7.99881172180175781f);
|
||||
#else
|
||||
const T plus_clamp = pset1<T>(7.90531110763549805f);
|
||||
const T minus_clamp = pset1<T>(-7.90531110763549805f);
|
||||
#endif
|
||||
const T tiny = pset1<T>(0.0004f);
|
||||
const T x = pmax(pmin(a_x, plus_clamp), minus_clamp);
|
||||
const T tiny_mask = pcmp_lt(pabs(a_x), tiny);
|
||||
// The monomial coefficients of the numerator polynomial (odd).
|
||||
const T alpha_1 = pset1<T>(4.89352455891786e-03f);
|
||||
const T alpha_3 = pset1<T>(6.37261928875436e-04f);
|
||||
const T alpha_5 = pset1<T>(1.48572235717979e-05f);
|
||||
const T alpha_7 = pset1<T>(5.12229709037114e-08f);
|
||||
const T alpha_9 = pset1<T>(-8.60467152213735e-11f);
|
||||
const T alpha_11 = pset1<T>(2.00018790482477e-13f);
|
||||
const T alpha_13 = pset1<T>(-2.76076847742355e-16f);
|
||||
|
||||
// The monomial coefficients of the denominator polynomial (even).
|
||||
const T beta_0 = pset1<T>(4.89352518554385e-03f);
|
||||
const T beta_2 = pset1<T>(2.26843463243900e-03f);
|
||||
const T beta_4 = pset1<T>(1.18534705686654e-04f);
|
||||
const T beta_6 = pset1<T>(1.19825839466702e-06f);
|
||||
|
||||
// Since the polynomials are odd/even, we need x^2.
|
||||
const T x2 = pmul(x, x);
|
||||
|
||||
// Evaluate the numerator polynomial p.
|
||||
T p = pmadd(x2, alpha_13, alpha_11);
|
||||
p = pmadd(x2, p, alpha_9);
|
||||
p = pmadd(x2, p, alpha_7);
|
||||
p = pmadd(x2, p, alpha_5);
|
||||
p = pmadd(x2, p, alpha_3);
|
||||
p = pmadd(x2, p, alpha_1);
|
||||
p = pmul(x, p);
|
||||
|
||||
// Evaluate the denominator polynomial q.
|
||||
T q = pmadd(x2, beta_6, beta_4);
|
||||
q = pmadd(x2, q, beta_2);
|
||||
q = pmadd(x2, q, beta_0);
|
||||
|
||||
// Divide the numerator by the denominator.
|
||||
return pselect(tiny_mask, x, pdiv(p, q));
|
||||
}
|
||||
|
||||
template<typename RealScalar>
|
||||
EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
|
||||
RealScalar positive_real_hypot(const RealScalar& x, const RealScalar& y)
|
||||
{
|
||||
// IEEE IEC 6059 special cases.
|
||||
if ((numext::isinf)(x) || (numext::isinf)(y))
|
||||
return NumTraits<RealScalar>::infinity();
|
||||
if ((numext::isnan)(x) || (numext::isnan)(y))
|
||||
return NumTraits<RealScalar>::quiet_NaN();
|
||||
|
||||
EIGEN_USING_STD(sqrt);
|
||||
RealScalar p, qp;
|
||||
p = numext::maxi(x,y);
|
||||
if(p==RealScalar(0)) return RealScalar(0);
|
||||
qp = numext::mini(y,x) / p;
|
||||
return p * sqrt(RealScalar(1) + qp*qp);
|
||||
}
|
||||
|
||||
template<typename Scalar>
|
||||
struct hypot_impl
|
||||
{
|
||||
typedef typename NumTraits<Scalar>::Real RealScalar;
|
||||
static EIGEN_DEVICE_FUNC
|
||||
inline RealScalar run(const Scalar& x, const Scalar& y)
|
||||
{
|
||||
EIGEN_USING_STD(abs);
|
||||
return positive_real_hypot<RealScalar>(abs(x), abs(y));
|
||||
}
|
||||
};
|
||||
|
||||
// Generic complex sqrt implementation that correctly handles corner cases
|
||||
// according to https://en.cppreference.com/w/cpp/numeric/complex/sqrt
|
||||
template<typename T>
|
||||
EIGEN_DEVICE_FUNC std::complex<T> complex_sqrt(const std::complex<T>& z) {
|
||||
// Computes the principal sqrt of the input.
|
||||
//
|
||||
// For a complex square root of the number x + i*y. We want to find real
|
||||
// numbers u and v such that
|
||||
// (u + i*v)^2 = x + i*y <=>
|
||||
// u^2 - v^2 + i*2*u*v = x + i*v.
|
||||
// By equating the real and imaginary parts we get:
|
||||
// u^2 - v^2 = x
|
||||
// 2*u*v = y.
|
||||
//
|
||||
// For x >= 0, this has the numerically stable solution
|
||||
// u = sqrt(0.5 * (x + sqrt(x^2 + y^2)))
|
||||
// v = y / (2 * u)
|
||||
// and for x < 0,
|
||||
// v = sign(y) * sqrt(0.5 * (-x + sqrt(x^2 + y^2)))
|
||||
// u = y / (2 * v)
|
||||
//
|
||||
// Letting w = sqrt(0.5 * (|x| + |z|)),
|
||||
// if x == 0: u = w, v = sign(y) * w
|
||||
// if x > 0: u = w, v = y / (2 * w)
|
||||
// if x < 0: u = |y| / (2 * w), v = sign(y) * w
|
||||
|
||||
const T x = numext::real(z);
|
||||
const T y = numext::imag(z);
|
||||
const T zero = T(0);
|
||||
const T w = numext::sqrt(T(0.5) * (numext::abs(x) + numext::hypot(x, y)));
|
||||
|
||||
return
|
||||
(numext::isinf)(y) ? std::complex<T>(NumTraits<T>::infinity(), y)
|
||||
: x == zero ? std::complex<T>(w, y < zero ? -w : w)
|
||||
: x > zero ? std::complex<T>(w, y / (2 * w))
|
||||
: std::complex<T>(numext::abs(y) / (2 * w), y < zero ? -w : w );
|
||||
}
|
||||
|
||||
// Generic complex rsqrt implementation.
|
||||
template<typename T>
|
||||
EIGEN_DEVICE_FUNC std::complex<T> complex_rsqrt(const std::complex<T>& z) {
|
||||
// Computes the principal reciprocal sqrt of the input.
|
||||
//
|
||||
// For a complex reciprocal square root of the number z = x + i*y. We want to
|
||||
// find real numbers u and v such that
|
||||
// (u + i*v)^2 = 1 / (x + i*y) <=>
|
||||
// u^2 - v^2 + i*2*u*v = x/|z|^2 - i*v/|z|^2.
|
||||
// By equating the real and imaginary parts we get:
|
||||
// u^2 - v^2 = x/|z|^2
|
||||
// 2*u*v = y/|z|^2.
|
||||
//
|
||||
// For x >= 0, this has the numerically stable solution
|
||||
// u = sqrt(0.5 * (x + |z|)) / |z|
|
||||
// v = -y / (2 * u * |z|)
|
||||
// and for x < 0,
|
||||
// v = -sign(y) * sqrt(0.5 * (-x + |z|)) / |z|
|
||||
// u = -y / (2 * v * |z|)
|
||||
//
|
||||
// Letting w = sqrt(0.5 * (|x| + |z|)),
|
||||
// if x == 0: u = w / |z|, v = -sign(y) * w / |z|
|
||||
// if x > 0: u = w / |z|, v = -y / (2 * w * |z|)
|
||||
// if x < 0: u = |y| / (2 * w * |z|), v = -sign(y) * w / |z|
|
||||
|
||||
const T x = numext::real(z);
|
||||
const T y = numext::imag(z);
|
||||
const T zero = T(0);
|
||||
|
||||
const T abs_z = numext::hypot(x, y);
|
||||
const T w = numext::sqrt(T(0.5) * (numext::abs(x) + abs_z));
|
||||
const T woz = w / abs_z;
|
||||
// Corner cases consistent with 1/sqrt(z) on gcc/clang.
|
||||
return
|
||||
abs_z == zero ? std::complex<T>(NumTraits<T>::infinity(), NumTraits<T>::quiet_NaN())
|
||||
: ((numext::isinf)(x) || (numext::isinf)(y)) ? std::complex<T>(zero, zero)
|
||||
: x == zero ? std::complex<T>(woz, y < zero ? woz : -woz)
|
||||
: x > zero ? std::complex<T>(woz, -y / (2 * w * abs_z))
|
||||
: std::complex<T>(numext::abs(y) / (2 * w * abs_z), y < zero ? woz : -woz );
|
||||
}
|
||||
|
||||
template<typename T>
|
||||
EIGEN_DEVICE_FUNC std::complex<T> complex_log(const std::complex<T>& z) {
|
||||
// Computes complex log.
|
||||
T a = numext::abs(z);
|
||||
EIGEN_USING_STD(atan2);
|
||||
T b = atan2(z.imag(), z.real());
|
||||
return std::complex<T>(numext::log(a), b);
|
||||
}
|
||||
|
||||
} // end namespace internal
|
||||
|
||||
} // end namespace Eigen
|
||||
|
||||
#endif // EIGEN_MATHFUNCTIONSIMPL_H
|
||||
@@ -0,0 +1,565 @@
|
||||
// This file is part of Eigen, a lightweight C++ template library
|
||||
// for linear algebra.
|
||||
//
|
||||
// Copyright (C) 2006-2010 Benoit Jacob <jacob.benoit.1@gmail.com>
|
||||
// Copyright (C) 2008-2009 Gael Guennebaud <gael.guennebaud@inria.fr>
|
||||
//
|
||||
// This Source Code Form is subject to the terms of the Mozilla
|
||||
// Public License v. 2.0. If a copy of the MPL was not distributed
|
||||
// with this file, You can obtain one at http://mozilla.org/MPL/2.0/.
|
||||
|
||||
#ifndef EIGEN_MATRIX_H
|
||||
#define EIGEN_MATRIX_H
|
||||
|
||||
namespace Eigen {
|
||||
|
||||
namespace internal {
|
||||
template<typename _Scalar, int _Rows, int _Cols, int _Options, int _MaxRows, int _MaxCols>
|
||||
struct traits<Matrix<_Scalar, _Rows, _Cols, _Options, _MaxRows, _MaxCols> >
|
||||
{
|
||||
private:
|
||||
enum { size = internal::size_at_compile_time<_Rows,_Cols>::ret };
|
||||
typedef typename find_best_packet<_Scalar,size>::type PacketScalar;
|
||||
enum {
|
||||
row_major_bit = _Options&RowMajor ? RowMajorBit : 0,
|
||||
is_dynamic_size_storage = _MaxRows==Dynamic || _MaxCols==Dynamic,
|
||||
max_size = is_dynamic_size_storage ? Dynamic : _MaxRows*_MaxCols,
|
||||
default_alignment = compute_default_alignment<_Scalar,max_size>::value,
|
||||
actual_alignment = ((_Options&DontAlign)==0) ? default_alignment : 0,
|
||||
required_alignment = unpacket_traits<PacketScalar>::alignment,
|
||||
packet_access_bit = (packet_traits<_Scalar>::Vectorizable && (EIGEN_UNALIGNED_VECTORIZE || (actual_alignment>=required_alignment))) ? PacketAccessBit : 0
|
||||
};
|
||||
|
||||
public:
|
||||
typedef _Scalar Scalar;
|
||||
typedef Dense StorageKind;
|
||||
typedef Eigen::Index StorageIndex;
|
||||
typedef MatrixXpr XprKind;
|
||||
enum {
|
||||
RowsAtCompileTime = _Rows,
|
||||
ColsAtCompileTime = _Cols,
|
||||
MaxRowsAtCompileTime = _MaxRows,
|
||||
MaxColsAtCompileTime = _MaxCols,
|
||||
Flags = compute_matrix_flags<_Scalar, _Rows, _Cols, _Options, _MaxRows, _MaxCols>::ret,
|
||||
Options = _Options,
|
||||
InnerStrideAtCompileTime = 1,
|
||||
OuterStrideAtCompileTime = (Options&RowMajor) ? ColsAtCompileTime : RowsAtCompileTime,
|
||||
|
||||
// FIXME, the following flag in only used to define NeedsToAlign in PlainObjectBase
|
||||
EvaluatorFlags = LinearAccessBit | DirectAccessBit | packet_access_bit | row_major_bit,
|
||||
Alignment = actual_alignment
|
||||
};
|
||||
};
|
||||
}
|
||||
|
||||
/** \class Matrix
|
||||
* \ingroup Core_Module
|
||||
*
|
||||
* \brief The matrix class, also used for vectors and row-vectors
|
||||
*
|
||||
* The %Matrix class is the work-horse for all \em dense (\ref dense "note") matrices and vectors within Eigen.
|
||||
* Vectors are matrices with one column, and row-vectors are matrices with one row.
|
||||
*
|
||||
* The %Matrix class encompasses \em both fixed-size and dynamic-size objects (\ref fixedsize "note").
|
||||
*
|
||||
* The first three template parameters are required:
|
||||
* \tparam _Scalar Numeric type, e.g. float, double, int or std::complex<float>.
|
||||
* User defined scalar types are supported as well (see \ref user_defined_scalars "here").
|
||||
* \tparam _Rows Number of rows, or \b Dynamic
|
||||
* \tparam _Cols Number of columns, or \b Dynamic
|
||||
*
|
||||
* The remaining template parameters are optional -- in most cases you don't have to worry about them.
|
||||
* \tparam _Options A combination of either \b #RowMajor or \b #ColMajor, and of either
|
||||
* \b #AutoAlign or \b #DontAlign.
|
||||
* The former controls \ref TopicStorageOrders "storage order", and defaults to column-major. The latter controls alignment, which is required
|
||||
* for vectorization. It defaults to aligning matrices except for fixed sizes that aren't a multiple of the packet size.
|
||||
* \tparam _MaxRows Maximum number of rows. Defaults to \a _Rows (\ref maxrows "note").
|
||||
* \tparam _MaxCols Maximum number of columns. Defaults to \a _Cols (\ref maxrows "note").
|
||||
*
|
||||
* Eigen provides a number of typedefs covering the usual cases. Here are some examples:
|
||||
*
|
||||
* \li \c Matrix2d is a 2x2 square matrix of doubles (\c Matrix<double, 2, 2>)
|
||||
* \li \c Vector4f is a vector of 4 floats (\c Matrix<float, 4, 1>)
|
||||
* \li \c RowVector3i is a row-vector of 3 ints (\c Matrix<int, 1, 3>)
|
||||
*
|
||||
* \li \c MatrixXf is a dynamic-size matrix of floats (\c Matrix<float, Dynamic, Dynamic>)
|
||||
* \li \c VectorXf is a dynamic-size vector of floats (\c Matrix<float, Dynamic, 1>)
|
||||
*
|
||||
* \li \c Matrix2Xf is a partially fixed-size (dynamic-size) matrix of floats (\c Matrix<float, 2, Dynamic>)
|
||||
* \li \c MatrixX3d is a partially dynamic-size (fixed-size) matrix of double (\c Matrix<double, Dynamic, 3>)
|
||||
*
|
||||
* See \link matrixtypedefs this page \endlink for a complete list of predefined \em %Matrix and \em Vector typedefs.
|
||||
*
|
||||
* You can access elements of vectors and matrices using normal subscripting:
|
||||
*
|
||||
* \code
|
||||
* Eigen::VectorXd v(10);
|
||||
* v[0] = 0.1;
|
||||
* v[1] = 0.2;
|
||||
* v(0) = 0.3;
|
||||
* v(1) = 0.4;
|
||||
*
|
||||
* Eigen::MatrixXi m(10, 10);
|
||||
* m(0, 1) = 1;
|
||||
* m(0, 2) = 2;
|
||||
* m(0, 3) = 3;
|
||||
* \endcode
|
||||
*
|
||||
* This class can be extended with the help of the plugin mechanism described on the page
|
||||
* \ref TopicCustomizing_Plugins by defining the preprocessor symbol \c EIGEN_MATRIX_PLUGIN.
|
||||
*
|
||||
* <i><b>Some notes:</b></i>
|
||||
*
|
||||
* <dl>
|
||||
* <dt><b>\anchor dense Dense versus sparse:</b></dt>
|
||||
* <dd>This %Matrix class handles dense, not sparse matrices and vectors. For sparse matrices and vectors, see the Sparse module.
|
||||
*
|
||||
* Dense matrices and vectors are plain usual arrays of coefficients. All the coefficients are stored, in an ordinary contiguous array.
|
||||
* This is unlike Sparse matrices and vectors where the coefficients are stored as a list of nonzero coefficients.</dd>
|
||||
*
|
||||
* <dt><b>\anchor fixedsize Fixed-size versus dynamic-size:</b></dt>
|
||||
* <dd>Fixed-size means that the numbers of rows and columns are known are compile-time. In this case, Eigen allocates the array
|
||||
* of coefficients as a fixed-size array, as a class member. This makes sense for very small matrices, typically up to 4x4, sometimes up
|
||||
* to 16x16. Larger matrices should be declared as dynamic-size even if one happens to know their size at compile-time.
|
||||
*
|
||||
* Dynamic-size means that the numbers of rows or columns are not necessarily known at compile-time. In this case they are runtime
|
||||
* variables, and the array of coefficients is allocated dynamically on the heap.
|
||||
*
|
||||
* Note that \em dense matrices, be they Fixed-size or Dynamic-size, <em>do not</em> expand dynamically in the sense of a std::map.
|
||||
* If you want this behavior, see the Sparse module.</dd>
|
||||
*
|
||||
* <dt><b>\anchor maxrows _MaxRows and _MaxCols:</b></dt>
|
||||
* <dd>In most cases, one just leaves these parameters to the default values.
|
||||
* These parameters mean the maximum size of rows and columns that the matrix may have. They are useful in cases
|
||||
* when the exact numbers of rows and columns are not known are compile-time, but it is known at compile-time that they cannot
|
||||
* exceed a certain value. This happens when taking dynamic-size blocks inside fixed-size matrices: in this case _MaxRows and _MaxCols
|
||||
* are the dimensions of the original matrix, while _Rows and _Cols are Dynamic.</dd>
|
||||
* </dl>
|
||||
*
|
||||
* <i><b>ABI and storage layout</b></i>
|
||||
*
|
||||
* The table below summarizes the ABI of some possible Matrix instances which is fixed thorough the lifetime of Eigen 3.
|
||||
* <table class="manual">
|
||||
* <tr><th>Matrix type</th><th>Equivalent C structure</th></tr>
|
||||
* <tr><td>\code Matrix<T,Dynamic,Dynamic> \endcode</td><td>\code
|
||||
* struct {
|
||||
* T *data; // with (size_t(data)%EIGEN_MAX_ALIGN_BYTES)==0
|
||||
* Eigen::Index rows, cols;
|
||||
* };
|
||||
* \endcode</td></tr>
|
||||
* <tr class="alt"><td>\code
|
||||
* Matrix<T,Dynamic,1>
|
||||
* Matrix<T,1,Dynamic> \endcode</td><td>\code
|
||||
* struct {
|
||||
* T *data; // with (size_t(data)%EIGEN_MAX_ALIGN_BYTES)==0
|
||||
* Eigen::Index size;
|
||||
* };
|
||||
* \endcode</td></tr>
|
||||
* <tr><td>\code Matrix<T,Rows,Cols> \endcode</td><td>\code
|
||||
* struct {
|
||||
* T data[Rows*Cols]; // with (size_t(data)%A(Rows*Cols*sizeof(T)))==0
|
||||
* };
|
||||
* \endcode</td></tr>
|
||||
* <tr class="alt"><td>\code Matrix<T,Dynamic,Dynamic,0,MaxRows,MaxCols> \endcode</td><td>\code
|
||||
* struct {
|
||||
* T data[MaxRows*MaxCols]; // with (size_t(data)%A(MaxRows*MaxCols*sizeof(T)))==0
|
||||
* Eigen::Index rows, cols;
|
||||
* };
|
||||
* \endcode</td></tr>
|
||||
* </table>
|
||||
* Note that in this table Rows, Cols, MaxRows and MaxCols are all positive integers. A(S) is defined to the largest possible power-of-two
|
||||
* smaller to EIGEN_MAX_STATIC_ALIGN_BYTES.
|
||||
*
|
||||
* \see MatrixBase for the majority of the API methods for matrices, \ref TopicClassHierarchy,
|
||||
* \ref TopicStorageOrders
|
||||
*/
|
||||
|
||||
template<typename _Scalar, int _Rows, int _Cols, int _Options, int _MaxRows, int _MaxCols>
|
||||
class Matrix
|
||||
: public PlainObjectBase<Matrix<_Scalar, _Rows, _Cols, _Options, _MaxRows, _MaxCols> >
|
||||
{
|
||||
public:
|
||||
|
||||
/** \brief Base class typedef.
|
||||
* \sa PlainObjectBase
|
||||
*/
|
||||
typedef PlainObjectBase<Matrix> Base;
|
||||
|
||||
enum { Options = _Options };
|
||||
|
||||
EIGEN_DENSE_PUBLIC_INTERFACE(Matrix)
|
||||
|
||||
typedef typename Base::PlainObject PlainObject;
|
||||
|
||||
using Base::base;
|
||||
using Base::coeffRef;
|
||||
|
||||
/**
|
||||
* \brief Assigns matrices to each other.
|
||||
*
|
||||
* \note This is a special case of the templated operator=. Its purpose is
|
||||
* to prevent a default operator= from hiding the templated operator=.
|
||||
*
|
||||
* \callgraph
|
||||
*/
|
||||
EIGEN_DEVICE_FUNC
|
||||
EIGEN_STRONG_INLINE Matrix& operator=(const Matrix& other)
|
||||
{
|
||||
return Base::_set(other);
|
||||
}
|
||||
|
||||
/** \internal
|
||||
* \brief Copies the value of the expression \a other into \c *this with automatic resizing.
|
||||
*
|
||||
* *this might be resized to match the dimensions of \a other. If *this was a null matrix (not already initialized),
|
||||
* it will be initialized.
|
||||
*
|
||||
* Note that copying a row-vector into a vector (and conversely) is allowed.
|
||||
* The resizing, if any, is then done in the appropriate way so that row-vectors
|
||||
* remain row-vectors and vectors remain vectors.
|
||||
*/
|
||||
template<typename OtherDerived>
|
||||
EIGEN_DEVICE_FUNC
|
||||
EIGEN_STRONG_INLINE Matrix& operator=(const DenseBase<OtherDerived>& other)
|
||||
{
|
||||
return Base::_set(other);
|
||||
}
|
||||
|
||||
/* Here, doxygen failed to copy the brief information when using \copydoc */
|
||||
|
||||
/**
|
||||
* \brief Copies the generic expression \a other into *this.
|
||||
* \copydetails DenseBase::operator=(const EigenBase<OtherDerived> &other)
|
||||
*/
|
||||
template<typename OtherDerived>
|
||||
EIGEN_DEVICE_FUNC
|
||||
EIGEN_STRONG_INLINE Matrix& operator=(const EigenBase<OtherDerived> &other)
|
||||
{
|
||||
return Base::operator=(other);
|
||||
}
|
||||
|
||||
template<typename OtherDerived>
|
||||
EIGEN_DEVICE_FUNC
|
||||
EIGEN_STRONG_INLINE Matrix& operator=(const ReturnByValue<OtherDerived>& func)
|
||||
{
|
||||
return Base::operator=(func);
|
||||
}
|
||||
|
||||
/** \brief Default constructor.
|
||||
*
|
||||
* For fixed-size matrices, does nothing.
|
||||
*
|
||||
* For dynamic-size matrices, creates an empty matrix of size 0. Does not allocate any array. Such a matrix
|
||||
* is called a null matrix. This constructor is the unique way to create null matrices: resizing
|
||||
* a matrix to 0 is not supported.
|
||||
*
|
||||
* \sa resize(Index,Index)
|
||||
*/
|
||||
EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
|
||||
Matrix() : Base()
|
||||
{
|
||||
Base::_check_template_params();
|
||||
EIGEN_INITIALIZE_COEFFS_IF_THAT_OPTION_IS_ENABLED
|
||||
}
|
||||
|
||||
// FIXME is it still needed
|
||||
EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
|
||||
explicit Matrix(internal::constructor_without_unaligned_array_assert)
|
||||
: Base(internal::constructor_without_unaligned_array_assert())
|
||||
{ Base::_check_template_params(); EIGEN_INITIALIZE_COEFFS_IF_THAT_OPTION_IS_ENABLED }
|
||||
|
||||
#if EIGEN_HAS_RVALUE_REFERENCES
|
||||
EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
|
||||
Matrix(Matrix&& other) EIGEN_NOEXCEPT_IF(std::is_nothrow_move_constructible<Scalar>::value)
|
||||
: Base(std::move(other))
|
||||
{
|
||||
Base::_check_template_params();
|
||||
}
|
||||
EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
|
||||
Matrix& operator=(Matrix&& other) EIGEN_NOEXCEPT_IF(std::is_nothrow_move_assignable<Scalar>::value)
|
||||
{
|
||||
Base::operator=(std::move(other));
|
||||
return *this;
|
||||
}
|
||||
#endif
|
||||
|
||||
#if EIGEN_HAS_CXX11
|
||||
/** \copydoc PlainObjectBase(const Scalar&, const Scalar&, const Scalar&, const Scalar&, const ArgTypes&... args)
|
||||
*
|
||||
* Example: \include Matrix_variadic_ctor_cxx11.cpp
|
||||
* Output: \verbinclude Matrix_variadic_ctor_cxx11.out
|
||||
*
|
||||
* \sa Matrix(const std::initializer_list<std::initializer_list<Scalar>>&)
|
||||
*/
|
||||
template <typename... ArgTypes>
|
||||
EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
|
||||
Matrix(const Scalar& a0, const Scalar& a1, const Scalar& a2, const Scalar& a3, const ArgTypes&... args)
|
||||
: Base(a0, a1, a2, a3, args...) {}
|
||||
|
||||
/** \brief Constructs a Matrix and initializes it from the coefficients given as initializer-lists grouped by row. \cpp11
|
||||
*
|
||||
* In the general case, the constructor takes a list of rows, each row being represented as a list of coefficients:
|
||||
*
|
||||
* Example: \include Matrix_initializer_list_23_cxx11.cpp
|
||||
* Output: \verbinclude Matrix_initializer_list_23_cxx11.out
|
||||
*
|
||||
* Each of the inner initializer lists must contain the exact same number of elements, otherwise an assertion is triggered.
|
||||
*
|
||||
* In the case of a compile-time column vector, implicit transposition from a single row is allowed.
|
||||
* Therefore <code>VectorXd{{1,2,3,4,5}}</code> is legal and the more verbose syntax
|
||||
* <code>RowVectorXd{{1},{2},{3},{4},{5}}</code> can be avoided:
|
||||
*
|
||||
* Example: \include Matrix_initializer_list_vector_cxx11.cpp
|
||||
* Output: \verbinclude Matrix_initializer_list_vector_cxx11.out
|
||||
*
|
||||
* In the case of fixed-sized matrices, the initializer list sizes must exactly match the matrix sizes,
|
||||
* and implicit transposition is allowed for compile-time vectors only.
|
||||
*
|
||||
* \sa Matrix(const Scalar& a0, const Scalar& a1, const Scalar& a2, const Scalar& a3, const ArgTypes&... args)
|
||||
*/
|
||||
EIGEN_DEVICE_FUNC
|
||||
explicit EIGEN_STRONG_INLINE Matrix(const std::initializer_list<std::initializer_list<Scalar>>& list) : Base(list) {}
|
||||
#endif // end EIGEN_HAS_CXX11
|
||||
|
||||
#ifndef EIGEN_PARSED_BY_DOXYGEN
|
||||
|
||||
// This constructor is for both 1x1 matrices and dynamic vectors
|
||||
template<typename T>
|
||||
EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
|
||||
explicit Matrix(const T& x)
|
||||
{
|
||||
Base::_check_template_params();
|
||||
Base::template _init1<T>(x);
|
||||
}
|
||||
|
||||
template<typename T0, typename T1>
|
||||
EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
|
||||
Matrix(const T0& x, const T1& y)
|
||||
{
|
||||
Base::_check_template_params();
|
||||
Base::template _init2<T0,T1>(x, y);
|
||||
}
|
||||
|
||||
|
||||
#else
|
||||
/** \brief Constructs a fixed-sized matrix initialized with coefficients starting at \a data */
|
||||
EIGEN_DEVICE_FUNC
|
||||
explicit Matrix(const Scalar *data);
|
||||
|
||||
/** \brief Constructs a vector or row-vector with given dimension. \only_for_vectors
|
||||
*
|
||||
* This is useful for dynamic-size vectors. For fixed-size vectors,
|
||||
* it is redundant to pass these parameters, so one should use the default constructor
|
||||
* Matrix() instead.
|
||||
*
|
||||
* \warning This constructor is disabled for fixed-size \c 1x1 matrices. For instance,
|
||||
* calling Matrix<double,1,1>(1) will call the initialization constructor: Matrix(const Scalar&).
|
||||
* For fixed-size \c 1x1 matrices it is therefore recommended to use the default
|
||||
* constructor Matrix() instead, especially when using one of the non standard
|
||||
* \c EIGEN_INITIALIZE_MATRICES_BY_{ZERO,\c NAN} macros (see \ref TopicPreprocessorDirectives).
|
||||
*/
|
||||
EIGEN_STRONG_INLINE explicit Matrix(Index dim);
|
||||
/** \brief Constructs an initialized 1x1 matrix with the given coefficient
|
||||
* \sa Matrix(const Scalar&, const Scalar&, const Scalar&, const Scalar&, const ArgTypes&...) */
|
||||
Matrix(const Scalar& x);
|
||||
/** \brief Constructs an uninitialized matrix with \a rows rows and \a cols columns.
|
||||
*
|
||||
* This is useful for dynamic-size matrices. For fixed-size matrices,
|
||||
* it is redundant to pass these parameters, so one should use the default constructor
|
||||
* Matrix() instead.
|
||||
*
|
||||
* \warning This constructor is disabled for fixed-size \c 1x2 and \c 2x1 vectors. For instance,
|
||||
* calling Matrix2f(2,1) will call the initialization constructor: Matrix(const Scalar& x, const Scalar& y).
|
||||
* For fixed-size \c 1x2 or \c 2x1 vectors it is therefore recommended to use the default
|
||||
* constructor Matrix() instead, especially when using one of the non standard
|
||||
* \c EIGEN_INITIALIZE_MATRICES_BY_{ZERO,\c NAN} macros (see \ref TopicPreprocessorDirectives).
|
||||
*/
|
||||
EIGEN_DEVICE_FUNC
|
||||
Matrix(Index rows, Index cols);
|
||||
|
||||
/** \brief Constructs an initialized 2D vector with given coefficients
|
||||
* \sa Matrix(const Scalar&, const Scalar&, const Scalar&, const Scalar&, const ArgTypes&...) */
|
||||
Matrix(const Scalar& x, const Scalar& y);
|
||||
#endif // end EIGEN_PARSED_BY_DOXYGEN
|
||||
|
||||
/** \brief Constructs an initialized 3D vector with given coefficients
|
||||
* \sa Matrix(const Scalar&, const Scalar&, const Scalar&, const Scalar&, const ArgTypes&...)
|
||||
*/
|
||||
EIGEN_DEVICE_FUNC
|
||||
EIGEN_STRONG_INLINE Matrix(const Scalar& x, const Scalar& y, const Scalar& z)
|
||||
{
|
||||
Base::_check_template_params();
|
||||
EIGEN_STATIC_ASSERT_VECTOR_SPECIFIC_SIZE(Matrix, 3)
|
||||
m_storage.data()[0] = x;
|
||||
m_storage.data()[1] = y;
|
||||
m_storage.data()[2] = z;
|
||||
}
|
||||
/** \brief Constructs an initialized 4D vector with given coefficients
|
||||
* \sa Matrix(const Scalar&, const Scalar&, const Scalar&, const Scalar&, const ArgTypes&...)
|
||||
*/
|
||||
EIGEN_DEVICE_FUNC
|
||||
EIGEN_STRONG_INLINE Matrix(const Scalar& x, const Scalar& y, const Scalar& z, const Scalar& w)
|
||||
{
|
||||
Base::_check_template_params();
|
||||
EIGEN_STATIC_ASSERT_VECTOR_SPECIFIC_SIZE(Matrix, 4)
|
||||
m_storage.data()[0] = x;
|
||||
m_storage.data()[1] = y;
|
||||
m_storage.data()[2] = z;
|
||||
m_storage.data()[3] = w;
|
||||
}
|
||||
|
||||
|
||||
/** \brief Copy constructor */
|
||||
EIGEN_DEVICE_FUNC
|
||||
EIGEN_STRONG_INLINE Matrix(const Matrix& other) : Base(other)
|
||||
{ }
|
||||
|
||||
/** \brief Copy constructor for generic expressions.
|
||||
* \sa MatrixBase::operator=(const EigenBase<OtherDerived>&)
|
||||
*/
|
||||
template<typename OtherDerived>
|
||||
EIGEN_DEVICE_FUNC
|
||||
EIGEN_STRONG_INLINE Matrix(const EigenBase<OtherDerived> &other)
|
||||
: Base(other.derived())
|
||||
{ }
|
||||
|
||||
EIGEN_DEVICE_FUNC EIGEN_CONSTEXPR
|
||||
inline Index innerStride() const EIGEN_NOEXCEPT { return 1; }
|
||||
EIGEN_DEVICE_FUNC EIGEN_CONSTEXPR
|
||||
inline Index outerStride() const EIGEN_NOEXCEPT { return this->innerSize(); }
|
||||
|
||||
/////////// Geometry module ///////////
|
||||
|
||||
template<typename OtherDerived>
|
||||
EIGEN_DEVICE_FUNC
|
||||
explicit Matrix(const RotationBase<OtherDerived,ColsAtCompileTime>& r);
|
||||
template<typename OtherDerived>
|
||||
EIGEN_DEVICE_FUNC
|
||||
Matrix& operator=(const RotationBase<OtherDerived,ColsAtCompileTime>& r);
|
||||
|
||||
// allow to extend Matrix outside Eigen
|
||||
#ifdef EIGEN_MATRIX_PLUGIN
|
||||
#include EIGEN_MATRIX_PLUGIN
|
||||
#endif
|
||||
|
||||
protected:
|
||||
template <typename Derived, typename OtherDerived, bool IsVector>
|
||||
friend struct internal::conservative_resize_like_impl;
|
||||
|
||||
using Base::m_storage;
|
||||
};
|
||||
|
||||
/** \defgroup matrixtypedefs Global matrix typedefs
|
||||
*
|
||||
* \ingroup Core_Module
|
||||
*
|
||||
* %Eigen defines several typedef shortcuts for most common matrix and vector types.
|
||||
*
|
||||
* The general patterns are the following:
|
||||
*
|
||||
* \c MatrixSizeType where \c Size can be \c 2,\c 3,\c 4 for fixed size square matrices or \c X for dynamic size,
|
||||
* and where \c Type can be \c i for integer, \c f for float, \c d for double, \c cf for complex float, \c cd
|
||||
* for complex double.
|
||||
*
|
||||
* For example, \c Matrix3d is a fixed-size 3x3 matrix type of doubles, and \c MatrixXf is a dynamic-size matrix of floats.
|
||||
*
|
||||
* There are also \c VectorSizeType and \c RowVectorSizeType which are self-explanatory. For example, \c Vector4cf is
|
||||
* a fixed-size vector of 4 complex floats.
|
||||
*
|
||||
* With \cpp11, template alias are also defined for common sizes.
|
||||
* They follow the same pattern as above except that the scalar type suffix is replaced by a
|
||||
* template parameter, i.e.:
|
||||
* - `MatrixSize<Type>` where `Size` can be \c 2,\c 3,\c 4 for fixed size square matrices or \c X for dynamic size.
|
||||
* - `MatrixXSize<Type>` and `MatrixSizeX<Type>` where `Size` can be \c 2,\c 3,\c 4 for hybrid dynamic/fixed matrices.
|
||||
* - `VectorSize<Type>` and `RowVectorSize<Type>` for column and row vectors.
|
||||
*
|
||||
* With \cpp11, you can also use fully generic column and row vector types: `Vector<Type,Size>` and `RowVector<Type,Size>`.
|
||||
*
|
||||
* \sa class Matrix
|
||||
*/
|
||||
|
||||
#define EIGEN_MAKE_TYPEDEFS(Type, TypeSuffix, Size, SizeSuffix) \
|
||||
/** \ingroup matrixtypedefs */ \
|
||||
typedef Matrix<Type, Size, Size> Matrix##SizeSuffix##TypeSuffix; \
|
||||
/** \ingroup matrixtypedefs */ \
|
||||
typedef Matrix<Type, Size, 1> Vector##SizeSuffix##TypeSuffix; \
|
||||
/** \ingroup matrixtypedefs */ \
|
||||
typedef Matrix<Type, 1, Size> RowVector##SizeSuffix##TypeSuffix;
|
||||
|
||||
#define EIGEN_MAKE_FIXED_TYPEDEFS(Type, TypeSuffix, Size) \
|
||||
/** \ingroup matrixtypedefs */ \
|
||||
typedef Matrix<Type, Size, Dynamic> Matrix##Size##X##TypeSuffix; \
|
||||
/** \ingroup matrixtypedefs */ \
|
||||
typedef Matrix<Type, Dynamic, Size> Matrix##X##Size##TypeSuffix;
|
||||
|
||||
#define EIGEN_MAKE_TYPEDEFS_ALL_SIZES(Type, TypeSuffix) \
|
||||
EIGEN_MAKE_TYPEDEFS(Type, TypeSuffix, 2, 2) \
|
||||
EIGEN_MAKE_TYPEDEFS(Type, TypeSuffix, 3, 3) \
|
||||
EIGEN_MAKE_TYPEDEFS(Type, TypeSuffix, 4, 4) \
|
||||
EIGEN_MAKE_TYPEDEFS(Type, TypeSuffix, Dynamic, X) \
|
||||
EIGEN_MAKE_FIXED_TYPEDEFS(Type, TypeSuffix, 2) \
|
||||
EIGEN_MAKE_FIXED_TYPEDEFS(Type, TypeSuffix, 3) \
|
||||
EIGEN_MAKE_FIXED_TYPEDEFS(Type, TypeSuffix, 4)
|
||||
|
||||
EIGEN_MAKE_TYPEDEFS_ALL_SIZES(int, i)
|
||||
EIGEN_MAKE_TYPEDEFS_ALL_SIZES(float, f)
|
||||
EIGEN_MAKE_TYPEDEFS_ALL_SIZES(double, d)
|
||||
EIGEN_MAKE_TYPEDEFS_ALL_SIZES(std::complex<float>, cf)
|
||||
EIGEN_MAKE_TYPEDEFS_ALL_SIZES(std::complex<double>, cd)
|
||||
|
||||
#undef EIGEN_MAKE_TYPEDEFS_ALL_SIZES
|
||||
#undef EIGEN_MAKE_TYPEDEFS
|
||||
#undef EIGEN_MAKE_FIXED_TYPEDEFS
|
||||
|
||||
#if EIGEN_HAS_CXX11
|
||||
|
||||
#define EIGEN_MAKE_TYPEDEFS(Size, SizeSuffix) \
|
||||
/** \ingroup matrixtypedefs */ \
|
||||
/** \brief \cpp11 */ \
|
||||
template <typename Type> \
|
||||
using Matrix##SizeSuffix = Matrix<Type, Size, Size>; \
|
||||
/** \ingroup matrixtypedefs */ \
|
||||
/** \brief \cpp11 */ \
|
||||
template <typename Type> \
|
||||
using Vector##SizeSuffix = Matrix<Type, Size, 1>; \
|
||||
/** \ingroup matrixtypedefs */ \
|
||||
/** \brief \cpp11 */ \
|
||||
template <typename Type> \
|
||||
using RowVector##SizeSuffix = Matrix<Type, 1, Size>;
|
||||
|
||||
#define EIGEN_MAKE_FIXED_TYPEDEFS(Size) \
|
||||
/** \ingroup matrixtypedefs */ \
|
||||
/** \brief \cpp11 */ \
|
||||
template <typename Type> \
|
||||
using Matrix##Size##X = Matrix<Type, Size, Dynamic>; \
|
||||
/** \ingroup matrixtypedefs */ \
|
||||
/** \brief \cpp11 */ \
|
||||
template <typename Type> \
|
||||
using Matrix##X##Size = Matrix<Type, Dynamic, Size>;
|
||||
|
||||
EIGEN_MAKE_TYPEDEFS(2, 2)
|
||||
EIGEN_MAKE_TYPEDEFS(3, 3)
|
||||
EIGEN_MAKE_TYPEDEFS(4, 4)
|
||||
EIGEN_MAKE_TYPEDEFS(Dynamic, X)
|
||||
EIGEN_MAKE_FIXED_TYPEDEFS(2)
|
||||
EIGEN_MAKE_FIXED_TYPEDEFS(3)
|
||||
EIGEN_MAKE_FIXED_TYPEDEFS(4)
|
||||
|
||||
/** \ingroup matrixtypedefs
|
||||
* \brief \cpp11 */
|
||||
template <typename Type, int Size>
|
||||
using Vector = Matrix<Type, Size, 1>;
|
||||
|
||||
/** \ingroup matrixtypedefs
|
||||
* \brief \cpp11 */
|
||||
template <typename Type, int Size>
|
||||
using RowVector = Matrix<Type, 1, Size>;
|
||||
|
||||
#undef EIGEN_MAKE_TYPEDEFS
|
||||
#undef EIGEN_MAKE_FIXED_TYPEDEFS
|
||||
|
||||
#endif // EIGEN_HAS_CXX11
|
||||
|
||||
} // end namespace Eigen
|
||||
|
||||
#endif // EIGEN_MATRIX_H
|
||||
@@ -0,0 +1,547 @@
|
||||
// This file is part of Eigen, a lightweight C++ template library
|
||||
// for linear algebra.
|
||||
//
|
||||
// Copyright (C) 2006-2009 Benoit Jacob <jacob.benoit.1@gmail.com>
|
||||
// Copyright (C) 2008 Gael Guennebaud <gael.guennebaud@inria.fr>
|
||||
//
|
||||
// This Source Code Form is subject to the terms of the Mozilla
|
||||
// Public License v. 2.0. If a copy of the MPL was not distributed
|
||||
// with this file, You can obtain one at http://mozilla.org/MPL/2.0/.
|
||||
|
||||
#ifndef EIGEN_MATRIXBASE_H
|
||||
#define EIGEN_MATRIXBASE_H
|
||||
|
||||
namespace Eigen {
|
||||
|
||||
/** \class MatrixBase
|
||||
* \ingroup Core_Module
|
||||
*
|
||||
* \brief Base class for all dense matrices, vectors, and expressions
|
||||
*
|
||||
* This class is the base that is inherited by all matrix, vector, and related expression
|
||||
* types. Most of the Eigen API is contained in this class, and its base classes. Other important
|
||||
* classes for the Eigen API are Matrix, and VectorwiseOp.
|
||||
*
|
||||
* Note that some methods are defined in other modules such as the \ref LU_Module LU module
|
||||
* for all functions related to matrix inversions.
|
||||
*
|
||||
* \tparam Derived is the derived type, e.g. a matrix type, or an expression, etc.
|
||||
*
|
||||
* When writing a function taking Eigen objects as argument, if you want your function
|
||||
* to take as argument any matrix, vector, or expression, just let it take a
|
||||
* MatrixBase argument. As an example, here is a function printFirstRow which, given
|
||||
* a matrix, vector, or expression \a x, prints the first row of \a x.
|
||||
*
|
||||
* \code
|
||||
template<typename Derived>
|
||||
void printFirstRow(const Eigen::MatrixBase<Derived>& x)
|
||||
{
|
||||
cout << x.row(0) << endl;
|
||||
}
|
||||
* \endcode
|
||||
*
|
||||
* This class can be extended with the help of the plugin mechanism described on the page
|
||||
* \ref TopicCustomizing_Plugins by defining the preprocessor symbol \c EIGEN_MATRIXBASE_PLUGIN.
|
||||
*
|
||||
* \sa \blank \ref TopicClassHierarchy
|
||||
*/
|
||||
template<typename Derived> class MatrixBase
|
||||
: public DenseBase<Derived>
|
||||
{
|
||||
public:
|
||||
#ifndef EIGEN_PARSED_BY_DOXYGEN
|
||||
typedef MatrixBase StorageBaseType;
|
||||
typedef typename internal::traits<Derived>::StorageKind StorageKind;
|
||||
typedef typename internal::traits<Derived>::StorageIndex StorageIndex;
|
||||
typedef typename internal::traits<Derived>::Scalar Scalar;
|
||||
typedef typename internal::packet_traits<Scalar>::type PacketScalar;
|
||||
typedef typename NumTraits<Scalar>::Real RealScalar;
|
||||
|
||||
typedef DenseBase<Derived> Base;
|
||||
using Base::RowsAtCompileTime;
|
||||
using Base::ColsAtCompileTime;
|
||||
using Base::SizeAtCompileTime;
|
||||
using Base::MaxRowsAtCompileTime;
|
||||
using Base::MaxColsAtCompileTime;
|
||||
using Base::MaxSizeAtCompileTime;
|
||||
using Base::IsVectorAtCompileTime;
|
||||
using Base::Flags;
|
||||
|
||||
using Base::derived;
|
||||
using Base::const_cast_derived;
|
||||
using Base::rows;
|
||||
using Base::cols;
|
||||
using Base::size;
|
||||
using Base::coeff;
|
||||
using Base::coeffRef;
|
||||
using Base::lazyAssign;
|
||||
using Base::eval;
|
||||
using Base::operator-;
|
||||
using Base::operator+=;
|
||||
using Base::operator-=;
|
||||
using Base::operator*=;
|
||||
using Base::operator/=;
|
||||
|
||||
typedef typename Base::CoeffReturnType CoeffReturnType;
|
||||
typedef typename Base::ConstTransposeReturnType ConstTransposeReturnType;
|
||||
typedef typename Base::RowXpr RowXpr;
|
||||
typedef typename Base::ColXpr ColXpr;
|
||||
#endif // not EIGEN_PARSED_BY_DOXYGEN
|
||||
|
||||
|
||||
|
||||
#ifndef EIGEN_PARSED_BY_DOXYGEN
|
||||
/** type of the equivalent square matrix */
|
||||
typedef Matrix<Scalar,EIGEN_SIZE_MAX(RowsAtCompileTime,ColsAtCompileTime),
|
||||
EIGEN_SIZE_MAX(RowsAtCompileTime,ColsAtCompileTime)> SquareMatrixType;
|
||||
#endif // not EIGEN_PARSED_BY_DOXYGEN
|
||||
|
||||
/** \returns the size of the main diagonal, which is min(rows(),cols()).
|
||||
* \sa rows(), cols(), SizeAtCompileTime. */
|
||||
EIGEN_DEVICE_FUNC
|
||||
inline Index diagonalSize() const { return (numext::mini)(rows(),cols()); }
|
||||
|
||||
typedef typename Base::PlainObject PlainObject;
|
||||
|
||||
#ifndef EIGEN_PARSED_BY_DOXYGEN
|
||||
/** \internal Represents a matrix with all coefficients equal to one another*/
|
||||
typedef CwiseNullaryOp<internal::scalar_constant_op<Scalar>,PlainObject> ConstantReturnType;
|
||||
/** \internal the return type of MatrixBase::adjoint() */
|
||||
typedef typename internal::conditional<NumTraits<Scalar>::IsComplex,
|
||||
CwiseUnaryOp<internal::scalar_conjugate_op<Scalar>, ConstTransposeReturnType>,
|
||||
ConstTransposeReturnType
|
||||
>::type AdjointReturnType;
|
||||
/** \internal Return type of eigenvalues() */
|
||||
typedef Matrix<std::complex<RealScalar>, internal::traits<Derived>::ColsAtCompileTime, 1, ColMajor> EigenvaluesReturnType;
|
||||
/** \internal the return type of identity */
|
||||
typedef CwiseNullaryOp<internal::scalar_identity_op<Scalar>,PlainObject> IdentityReturnType;
|
||||
/** \internal the return type of unit vectors */
|
||||
typedef Block<const CwiseNullaryOp<internal::scalar_identity_op<Scalar>, SquareMatrixType>,
|
||||
internal::traits<Derived>::RowsAtCompileTime,
|
||||
internal::traits<Derived>::ColsAtCompileTime> BasisReturnType;
|
||||
#endif // not EIGEN_PARSED_BY_DOXYGEN
|
||||
|
||||
#define EIGEN_CURRENT_STORAGE_BASE_CLASS Eigen::MatrixBase
|
||||
#define EIGEN_DOC_UNARY_ADDONS(X,Y)
|
||||
# include "../plugins/CommonCwiseBinaryOps.h"
|
||||
# include "../plugins/MatrixCwiseUnaryOps.h"
|
||||
# include "../plugins/MatrixCwiseBinaryOps.h"
|
||||
# ifdef EIGEN_MATRIXBASE_PLUGIN
|
||||
# include EIGEN_MATRIXBASE_PLUGIN
|
||||
# endif
|
||||
#undef EIGEN_CURRENT_STORAGE_BASE_CLASS
|
||||
#undef EIGEN_DOC_UNARY_ADDONS
|
||||
|
||||
/** Special case of the template operator=, in order to prevent the compiler
|
||||
* from generating a default operator= (issue hit with g++ 4.1)
|
||||
*/
|
||||
EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
|
||||
Derived& operator=(const MatrixBase& other);
|
||||
|
||||
// We cannot inherit here via Base::operator= since it is causing
|
||||
// trouble with MSVC.
|
||||
|
||||
template <typename OtherDerived>
|
||||
EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
|
||||
Derived& operator=(const DenseBase<OtherDerived>& other);
|
||||
|
||||
template <typename OtherDerived>
|
||||
EIGEN_DEVICE_FUNC
|
||||
Derived& operator=(const EigenBase<OtherDerived>& other);
|
||||
|
||||
template<typename OtherDerived>
|
||||
EIGEN_DEVICE_FUNC
|
||||
Derived& operator=(const ReturnByValue<OtherDerived>& other);
|
||||
|
||||
template<typename OtherDerived>
|
||||
EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
|
||||
Derived& operator+=(const MatrixBase<OtherDerived>& other);
|
||||
template<typename OtherDerived>
|
||||
EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
|
||||
Derived& operator-=(const MatrixBase<OtherDerived>& other);
|
||||
|
||||
template<typename OtherDerived>
|
||||
EIGEN_DEVICE_FUNC
|
||||
const Product<Derived,OtherDerived>
|
||||
operator*(const MatrixBase<OtherDerived> &other) const;
|
||||
|
||||
template<typename OtherDerived>
|
||||
EIGEN_DEVICE_FUNC
|
||||
const Product<Derived,OtherDerived,LazyProduct>
|
||||
lazyProduct(const MatrixBase<OtherDerived> &other) const;
|
||||
|
||||
template<typename OtherDerived>
|
||||
Derived& operator*=(const EigenBase<OtherDerived>& other);
|
||||
|
||||
template<typename OtherDerived>
|
||||
void applyOnTheLeft(const EigenBase<OtherDerived>& other);
|
||||
|
||||
template<typename OtherDerived>
|
||||
void applyOnTheRight(const EigenBase<OtherDerived>& other);
|
||||
|
||||
template<typename DiagonalDerived>
|
||||
EIGEN_DEVICE_FUNC
|
||||
const Product<Derived, DiagonalDerived, LazyProduct>
|
||||
operator*(const DiagonalBase<DiagonalDerived> &diagonal) const;
|
||||
|
||||
template<typename OtherDerived>
|
||||
EIGEN_DEVICE_FUNC
|
||||
typename ScalarBinaryOpTraits<typename internal::traits<Derived>::Scalar,typename internal::traits<OtherDerived>::Scalar>::ReturnType
|
||||
dot(const MatrixBase<OtherDerived>& other) const;
|
||||
|
||||
EIGEN_DEVICE_FUNC RealScalar squaredNorm() const;
|
||||
EIGEN_DEVICE_FUNC RealScalar norm() const;
|
||||
RealScalar stableNorm() const;
|
||||
RealScalar blueNorm() const;
|
||||
RealScalar hypotNorm() const;
|
||||
EIGEN_DEVICE_FUNC const PlainObject normalized() const;
|
||||
EIGEN_DEVICE_FUNC const PlainObject stableNormalized() const;
|
||||
EIGEN_DEVICE_FUNC void normalize();
|
||||
EIGEN_DEVICE_FUNC void stableNormalize();
|
||||
|
||||
EIGEN_DEVICE_FUNC const AdjointReturnType adjoint() const;
|
||||
EIGEN_DEVICE_FUNC void adjointInPlace();
|
||||
|
||||
typedef Diagonal<Derived> DiagonalReturnType;
|
||||
EIGEN_DEVICE_FUNC
|
||||
DiagonalReturnType diagonal();
|
||||
|
||||
typedef typename internal::add_const<Diagonal<const Derived> >::type ConstDiagonalReturnType;
|
||||
EIGEN_DEVICE_FUNC
|
||||
ConstDiagonalReturnType diagonal() const;
|
||||
|
||||
template<int Index> struct DiagonalIndexReturnType { typedef Diagonal<Derived,Index> Type; };
|
||||
template<int Index> struct ConstDiagonalIndexReturnType { typedef const Diagonal<const Derived,Index> Type; };
|
||||
|
||||
template<int Index>
|
||||
EIGEN_DEVICE_FUNC
|
||||
typename DiagonalIndexReturnType<Index>::Type diagonal();
|
||||
|
||||
template<int Index>
|
||||
EIGEN_DEVICE_FUNC
|
||||
typename ConstDiagonalIndexReturnType<Index>::Type diagonal() const;
|
||||
|
||||
typedef Diagonal<Derived,DynamicIndex> DiagonalDynamicIndexReturnType;
|
||||
typedef typename internal::add_const<Diagonal<const Derived,DynamicIndex> >::type ConstDiagonalDynamicIndexReturnType;
|
||||
|
||||
EIGEN_DEVICE_FUNC
|
||||
DiagonalDynamicIndexReturnType diagonal(Index index);
|
||||
EIGEN_DEVICE_FUNC
|
||||
ConstDiagonalDynamicIndexReturnType diagonal(Index index) const;
|
||||
|
||||
template<unsigned int Mode> struct TriangularViewReturnType { typedef TriangularView<Derived, Mode> Type; };
|
||||
template<unsigned int Mode> struct ConstTriangularViewReturnType { typedef const TriangularView<const Derived, Mode> Type; };
|
||||
|
||||
template<unsigned int Mode>
|
||||
EIGEN_DEVICE_FUNC
|
||||
typename TriangularViewReturnType<Mode>::Type triangularView();
|
||||
template<unsigned int Mode>
|
||||
EIGEN_DEVICE_FUNC
|
||||
typename ConstTriangularViewReturnType<Mode>::Type triangularView() const;
|
||||
|
||||
template<unsigned int UpLo> struct SelfAdjointViewReturnType { typedef SelfAdjointView<Derived, UpLo> Type; };
|
||||
template<unsigned int UpLo> struct ConstSelfAdjointViewReturnType { typedef const SelfAdjointView<const Derived, UpLo> Type; };
|
||||
|
||||
template<unsigned int UpLo>
|
||||
EIGEN_DEVICE_FUNC
|
||||
typename SelfAdjointViewReturnType<UpLo>::Type selfadjointView();
|
||||
template<unsigned int UpLo>
|
||||
EIGEN_DEVICE_FUNC
|
||||
typename ConstSelfAdjointViewReturnType<UpLo>::Type selfadjointView() const;
|
||||
|
||||
const SparseView<Derived> sparseView(const Scalar& m_reference = Scalar(0),
|
||||
const typename NumTraits<Scalar>::Real& m_epsilon = NumTraits<Scalar>::dummy_precision()) const;
|
||||
EIGEN_DEVICE_FUNC static const IdentityReturnType Identity();
|
||||
EIGEN_DEVICE_FUNC static const IdentityReturnType Identity(Index rows, Index cols);
|
||||
EIGEN_DEVICE_FUNC static const BasisReturnType Unit(Index size, Index i);
|
||||
EIGEN_DEVICE_FUNC static const BasisReturnType Unit(Index i);
|
||||
EIGEN_DEVICE_FUNC static const BasisReturnType UnitX();
|
||||
EIGEN_DEVICE_FUNC static const BasisReturnType UnitY();
|
||||
EIGEN_DEVICE_FUNC static const BasisReturnType UnitZ();
|
||||
EIGEN_DEVICE_FUNC static const BasisReturnType UnitW();
|
||||
|
||||
EIGEN_DEVICE_FUNC
|
||||
const DiagonalWrapper<const Derived> asDiagonal() const;
|
||||
const PermutationWrapper<const Derived> asPermutation() const;
|
||||
|
||||
EIGEN_DEVICE_FUNC
|
||||
Derived& setIdentity();
|
||||
EIGEN_DEVICE_FUNC
|
||||
Derived& setIdentity(Index rows, Index cols);
|
||||
EIGEN_DEVICE_FUNC Derived& setUnit(Index i);
|
||||
EIGEN_DEVICE_FUNC Derived& setUnit(Index newSize, Index i);
|
||||
|
||||
bool isIdentity(const RealScalar& prec = NumTraits<Scalar>::dummy_precision()) const;
|
||||
bool isDiagonal(const RealScalar& prec = NumTraits<Scalar>::dummy_precision()) const;
|
||||
|
||||
bool isUpperTriangular(const RealScalar& prec = NumTraits<Scalar>::dummy_precision()) const;
|
||||
bool isLowerTriangular(const RealScalar& prec = NumTraits<Scalar>::dummy_precision()) const;
|
||||
|
||||
template<typename OtherDerived>
|
||||
bool isOrthogonal(const MatrixBase<OtherDerived>& other,
|
||||
const RealScalar& prec = NumTraits<Scalar>::dummy_precision()) const;
|
||||
bool isUnitary(const RealScalar& prec = NumTraits<Scalar>::dummy_precision()) const;
|
||||
|
||||
/** \returns true if each coefficients of \c *this and \a other are all exactly equal.
|
||||
* \warning When using floating point scalar values you probably should rather use a
|
||||
* fuzzy comparison such as isApprox()
|
||||
* \sa isApprox(), operator!= */
|
||||
template<typename OtherDerived>
|
||||
EIGEN_DEVICE_FUNC inline bool operator==(const MatrixBase<OtherDerived>& other) const
|
||||
{ return cwiseEqual(other).all(); }
|
||||
|
||||
/** \returns true if at least one pair of coefficients of \c *this and \a other are not exactly equal to each other.
|
||||
* \warning When using floating point scalar values you probably should rather use a
|
||||
* fuzzy comparison such as isApprox()
|
||||
* \sa isApprox(), operator== */
|
||||
template<typename OtherDerived>
|
||||
EIGEN_DEVICE_FUNC inline bool operator!=(const MatrixBase<OtherDerived>& other) const
|
||||
{ return cwiseNotEqual(other).any(); }
|
||||
|
||||
NoAlias<Derived,Eigen::MatrixBase > EIGEN_DEVICE_FUNC noalias();
|
||||
|
||||
// TODO forceAlignedAccess is temporarily disabled
|
||||
// Need to find a nicer workaround.
|
||||
inline const Derived& forceAlignedAccess() const { return derived(); }
|
||||
inline Derived& forceAlignedAccess() { return derived(); }
|
||||
template<bool Enable> inline const Derived& forceAlignedAccessIf() const { return derived(); }
|
||||
template<bool Enable> inline Derived& forceAlignedAccessIf() { return derived(); }
|
||||
|
||||
EIGEN_DEVICE_FUNC Scalar trace() const;
|
||||
|
||||
template<int p> EIGEN_DEVICE_FUNC RealScalar lpNorm() const;
|
||||
|
||||
EIGEN_DEVICE_FUNC MatrixBase<Derived>& matrix() { return *this; }
|
||||
EIGEN_DEVICE_FUNC const MatrixBase<Derived>& matrix() const { return *this; }
|
||||
|
||||
/** \returns an \link Eigen::ArrayBase Array \endlink expression of this matrix
|
||||
* \sa ArrayBase::matrix() */
|
||||
EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE ArrayWrapper<Derived> array() { return ArrayWrapper<Derived>(derived()); }
|
||||
/** \returns a const \link Eigen::ArrayBase Array \endlink expression of this matrix
|
||||
* \sa ArrayBase::matrix() */
|
||||
EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE const ArrayWrapper<const Derived> array() const { return ArrayWrapper<const Derived>(derived()); }
|
||||
|
||||
/////////// LU module ///////////
|
||||
|
||||
inline const FullPivLU<PlainObject> fullPivLu() const;
|
||||
inline const PartialPivLU<PlainObject> partialPivLu() const;
|
||||
|
||||
inline const PartialPivLU<PlainObject> lu() const;
|
||||
|
||||
EIGEN_DEVICE_FUNC
|
||||
inline const Inverse<Derived> inverse() const;
|
||||
|
||||
template<typename ResultType>
|
||||
inline void computeInverseAndDetWithCheck(
|
||||
ResultType& inverse,
|
||||
typename ResultType::Scalar& determinant,
|
||||
bool& invertible,
|
||||
const RealScalar& absDeterminantThreshold = NumTraits<Scalar>::dummy_precision()
|
||||
) const;
|
||||
|
||||
template<typename ResultType>
|
||||
inline void computeInverseWithCheck(
|
||||
ResultType& inverse,
|
||||
bool& invertible,
|
||||
const RealScalar& absDeterminantThreshold = NumTraits<Scalar>::dummy_precision()
|
||||
) const;
|
||||
|
||||
EIGEN_DEVICE_FUNC
|
||||
Scalar determinant() const;
|
||||
|
||||
/////////// Cholesky module ///////////
|
||||
|
||||
inline const LLT<PlainObject> llt() const;
|
||||
inline const LDLT<PlainObject> ldlt() const;
|
||||
|
||||
/////////// QR module ///////////
|
||||
|
||||
inline const HouseholderQR<PlainObject> householderQr() const;
|
||||
inline const ColPivHouseholderQR<PlainObject> colPivHouseholderQr() const;
|
||||
inline const FullPivHouseholderQR<PlainObject> fullPivHouseholderQr() const;
|
||||
inline const CompleteOrthogonalDecomposition<PlainObject> completeOrthogonalDecomposition() const;
|
||||
|
||||
/////////// Eigenvalues module ///////////
|
||||
|
||||
inline EigenvaluesReturnType eigenvalues() const;
|
||||
inline RealScalar operatorNorm() const;
|
||||
|
||||
/////////// SVD module ///////////
|
||||
|
||||
inline JacobiSVD<PlainObject> jacobiSvd(unsigned int computationOptions = 0) const;
|
||||
inline BDCSVD<PlainObject> bdcSvd(unsigned int computationOptions = 0) const;
|
||||
|
||||
/////////// Geometry module ///////////
|
||||
|
||||
#ifndef EIGEN_PARSED_BY_DOXYGEN
|
||||
/// \internal helper struct to form the return type of the cross product
|
||||
template<typename OtherDerived> struct cross_product_return_type {
|
||||
typedef typename ScalarBinaryOpTraits<typename internal::traits<Derived>::Scalar,typename internal::traits<OtherDerived>::Scalar>::ReturnType Scalar;
|
||||
typedef Matrix<Scalar,MatrixBase::RowsAtCompileTime,MatrixBase::ColsAtCompileTime> type;
|
||||
};
|
||||
#endif // EIGEN_PARSED_BY_DOXYGEN
|
||||
template<typename OtherDerived>
|
||||
EIGEN_DEVICE_FUNC
|
||||
#ifndef EIGEN_PARSED_BY_DOXYGEN
|
||||
inline typename cross_product_return_type<OtherDerived>::type
|
||||
#else
|
||||
inline PlainObject
|
||||
#endif
|
||||
cross(const MatrixBase<OtherDerived>& other) const;
|
||||
|
||||
template<typename OtherDerived>
|
||||
EIGEN_DEVICE_FUNC
|
||||
inline PlainObject cross3(const MatrixBase<OtherDerived>& other) const;
|
||||
|
||||
EIGEN_DEVICE_FUNC
|
||||
inline PlainObject unitOrthogonal(void) const;
|
||||
|
||||
EIGEN_DEVICE_FUNC
|
||||
inline Matrix<Scalar,3,1> eulerAngles(Index a0, Index a1, Index a2) const;
|
||||
|
||||
// put this as separate enum value to work around possible GCC 4.3 bug (?)
|
||||
enum { HomogeneousReturnTypeDirection = ColsAtCompileTime==1&&RowsAtCompileTime==1 ? ((internal::traits<Derived>::Flags&RowMajorBit)==RowMajorBit ? Horizontal : Vertical)
|
||||
: ColsAtCompileTime==1 ? Vertical : Horizontal };
|
||||
typedef Homogeneous<Derived, HomogeneousReturnTypeDirection> HomogeneousReturnType;
|
||||
EIGEN_DEVICE_FUNC
|
||||
inline HomogeneousReturnType homogeneous() const;
|
||||
|
||||
enum {
|
||||
SizeMinusOne = SizeAtCompileTime==Dynamic ? Dynamic : SizeAtCompileTime-1
|
||||
};
|
||||
typedef Block<const Derived,
|
||||
internal::traits<Derived>::ColsAtCompileTime==1 ? SizeMinusOne : 1,
|
||||
internal::traits<Derived>::ColsAtCompileTime==1 ? 1 : SizeMinusOne> ConstStartMinusOne;
|
||||
typedef EIGEN_EXPR_BINARYOP_SCALAR_RETURN_TYPE(ConstStartMinusOne,Scalar,quotient) HNormalizedReturnType;
|
||||
EIGEN_DEVICE_FUNC
|
||||
inline const HNormalizedReturnType hnormalized() const;
|
||||
|
||||
////////// Householder module ///////////
|
||||
|
||||
EIGEN_DEVICE_FUNC
|
||||
void makeHouseholderInPlace(Scalar& tau, RealScalar& beta);
|
||||
template<typename EssentialPart>
|
||||
EIGEN_DEVICE_FUNC
|
||||
void makeHouseholder(EssentialPart& essential,
|
||||
Scalar& tau, RealScalar& beta) const;
|
||||
template<typename EssentialPart>
|
||||
EIGEN_DEVICE_FUNC
|
||||
void applyHouseholderOnTheLeft(const EssentialPart& essential,
|
||||
const Scalar& tau,
|
||||
Scalar* workspace);
|
||||
template<typename EssentialPart>
|
||||
EIGEN_DEVICE_FUNC
|
||||
void applyHouseholderOnTheRight(const EssentialPart& essential,
|
||||
const Scalar& tau,
|
||||
Scalar* workspace);
|
||||
|
||||
///////// Jacobi module /////////
|
||||
|
||||
template<typename OtherScalar>
|
||||
EIGEN_DEVICE_FUNC
|
||||
void applyOnTheLeft(Index p, Index q, const JacobiRotation<OtherScalar>& j);
|
||||
template<typename OtherScalar>
|
||||
EIGEN_DEVICE_FUNC
|
||||
void applyOnTheRight(Index p, Index q, const JacobiRotation<OtherScalar>& j);
|
||||
|
||||
///////// SparseCore module /////////
|
||||
|
||||
template<typename OtherDerived>
|
||||
EIGEN_STRONG_INLINE const typename SparseMatrixBase<OtherDerived>::template CwiseProductDenseReturnType<Derived>::Type
|
||||
cwiseProduct(const SparseMatrixBase<OtherDerived> &other) const
|
||||
{
|
||||
return other.cwiseProduct(derived());
|
||||
}
|
||||
|
||||
///////// MatrixFunctions module /////////
|
||||
|
||||
typedef typename internal::stem_function<Scalar>::type StemFunction;
|
||||
#define EIGEN_MATRIX_FUNCTION(ReturnType, Name, Description) \
|
||||
/** \returns an expression of the matrix Description of \c *this. \brief This function requires the <a href="unsupported/group__MatrixFunctions__Module.html"> unsupported MatrixFunctions module</a>. To compute the coefficient-wise Description use ArrayBase::##Name . */ \
|
||||
const ReturnType<Derived> Name() const;
|
||||
#define EIGEN_MATRIX_FUNCTION_1(ReturnType, Name, Description, Argument) \
|
||||
/** \returns an expression of the matrix Description of \c *this. \brief This function requires the <a href="unsupported/group__MatrixFunctions__Module.html"> unsupported MatrixFunctions module</a>. To compute the coefficient-wise Description use ArrayBase::##Name . */ \
|
||||
const ReturnType<Derived> Name(Argument) const;
|
||||
|
||||
EIGEN_MATRIX_FUNCTION(MatrixExponentialReturnValue, exp, exponential)
|
||||
/** \brief Helper function for the <a href="unsupported/group__MatrixFunctions__Module.html"> unsupported MatrixFunctions module</a>.*/
|
||||
const MatrixFunctionReturnValue<Derived> matrixFunction(StemFunction f) const;
|
||||
EIGEN_MATRIX_FUNCTION(MatrixFunctionReturnValue, cosh, hyperbolic cosine)
|
||||
EIGEN_MATRIX_FUNCTION(MatrixFunctionReturnValue, sinh, hyperbolic sine)
|
||||
#if EIGEN_HAS_CXX11_MATH
|
||||
EIGEN_MATRIX_FUNCTION(MatrixFunctionReturnValue, atanh, inverse hyperbolic cosine)
|
||||
EIGEN_MATRIX_FUNCTION(MatrixFunctionReturnValue, acosh, inverse hyperbolic cosine)
|
||||
EIGEN_MATRIX_FUNCTION(MatrixFunctionReturnValue, asinh, inverse hyperbolic sine)
|
||||
#endif
|
||||
EIGEN_MATRIX_FUNCTION(MatrixFunctionReturnValue, cos, cosine)
|
||||
EIGEN_MATRIX_FUNCTION(MatrixFunctionReturnValue, sin, sine)
|
||||
EIGEN_MATRIX_FUNCTION(MatrixSquareRootReturnValue, sqrt, square root)
|
||||
EIGEN_MATRIX_FUNCTION(MatrixLogarithmReturnValue, log, logarithm)
|
||||
EIGEN_MATRIX_FUNCTION_1(MatrixPowerReturnValue, pow, power to \c p, const RealScalar& p)
|
||||
EIGEN_MATRIX_FUNCTION_1(MatrixComplexPowerReturnValue, pow, power to \c p, const std::complex<RealScalar>& p)
|
||||
|
||||
protected:
|
||||
EIGEN_DEFAULT_COPY_CONSTRUCTOR(MatrixBase)
|
||||
EIGEN_DEFAULT_EMPTY_CONSTRUCTOR_AND_DESTRUCTOR(MatrixBase)
|
||||
|
||||
private:
|
||||
EIGEN_DEVICE_FUNC explicit MatrixBase(int);
|
||||
EIGEN_DEVICE_FUNC MatrixBase(int,int);
|
||||
template<typename OtherDerived> EIGEN_DEVICE_FUNC explicit MatrixBase(const MatrixBase<OtherDerived>&);
|
||||
protected:
|
||||
// mixing arrays and matrices is not legal
|
||||
template<typename OtherDerived> Derived& operator+=(const ArrayBase<OtherDerived>& )
|
||||
{EIGEN_STATIC_ASSERT(std::ptrdiff_t(sizeof(typename OtherDerived::Scalar))==-1,YOU_CANNOT_MIX_ARRAYS_AND_MATRICES); return *this;}
|
||||
// mixing arrays and matrices is not legal
|
||||
template<typename OtherDerived> Derived& operator-=(const ArrayBase<OtherDerived>& )
|
||||
{EIGEN_STATIC_ASSERT(std::ptrdiff_t(sizeof(typename OtherDerived::Scalar))==-1,YOU_CANNOT_MIX_ARRAYS_AND_MATRICES); return *this;}
|
||||
};
|
||||
|
||||
|
||||
/***************************************************************************
|
||||
* Implementation of matrix base methods
|
||||
***************************************************************************/
|
||||
|
||||
/** replaces \c *this by \c *this * \a other.
|
||||
*
|
||||
* \returns a reference to \c *this
|
||||
*
|
||||
* Example: \include MatrixBase_applyOnTheRight.cpp
|
||||
* Output: \verbinclude MatrixBase_applyOnTheRight.out
|
||||
*/
|
||||
template<typename Derived>
|
||||
template<typename OtherDerived>
|
||||
inline Derived&
|
||||
MatrixBase<Derived>::operator*=(const EigenBase<OtherDerived> &other)
|
||||
{
|
||||
other.derived().applyThisOnTheRight(derived());
|
||||
return derived();
|
||||
}
|
||||
|
||||
/** replaces \c *this by \c *this * \a other. It is equivalent to MatrixBase::operator*=().
|
||||
*
|
||||
* Example: \include MatrixBase_applyOnTheRight.cpp
|
||||
* Output: \verbinclude MatrixBase_applyOnTheRight.out
|
||||
*/
|
||||
template<typename Derived>
|
||||
template<typename OtherDerived>
|
||||
inline void MatrixBase<Derived>::applyOnTheRight(const EigenBase<OtherDerived> &other)
|
||||
{
|
||||
other.derived().applyThisOnTheRight(derived());
|
||||
}
|
||||
|
||||
/** replaces \c *this by \a other * \c *this.
|
||||
*
|
||||
* Example: \include MatrixBase_applyOnTheLeft.cpp
|
||||
* Output: \verbinclude MatrixBase_applyOnTheLeft.out
|
||||
*/
|
||||
template<typename Derived>
|
||||
template<typename OtherDerived>
|
||||
inline void MatrixBase<Derived>::applyOnTheLeft(const EigenBase<OtherDerived> &other)
|
||||
{
|
||||
other.derived().applyThisOnTheLeft(derived());
|
||||
}
|
||||
|
||||
} // end namespace Eigen
|
||||
|
||||
#endif // EIGEN_MATRIXBASE_H
|
||||
@@ -0,0 +1,85 @@
|
||||
// This file is part of Eigen, a lightweight C++ template library
|
||||
// for linear algebra.
|
||||
//
|
||||
// Copyright (C) 2008 Gael Guennebaud <gael.guennebaud@inria.fr>
|
||||
// Copyright (C) 2006-2008 Benoit Jacob <jacob.benoit.1@gmail.com>
|
||||
//
|
||||
// This Source Code Form is subject to the terms of the Mozilla
|
||||
// Public License v. 2.0. If a copy of the MPL was not distributed
|
||||
// with this file, You can obtain one at http://mozilla.org/MPL/2.0/.
|
||||
|
||||
#ifndef EIGEN_NESTBYVALUE_H
|
||||
#define EIGEN_NESTBYVALUE_H
|
||||
|
||||
namespace Eigen {
|
||||
|
||||
namespace internal {
|
||||
template<typename ExpressionType>
|
||||
struct traits<NestByValue<ExpressionType> > : public traits<ExpressionType>
|
||||
{
|
||||
enum {
|
||||
Flags = traits<ExpressionType>::Flags & ~NestByRefBit
|
||||
};
|
||||
};
|
||||
}
|
||||
|
||||
/** \class NestByValue
|
||||
* \ingroup Core_Module
|
||||
*
|
||||
* \brief Expression which must be nested by value
|
||||
*
|
||||
* \tparam ExpressionType the type of the object of which we are requiring nesting-by-value
|
||||
*
|
||||
* This class is the return type of MatrixBase::nestByValue()
|
||||
* and most of the time this is the only way it is used.
|
||||
*
|
||||
* \sa MatrixBase::nestByValue()
|
||||
*/
|
||||
template<typename ExpressionType> class NestByValue
|
||||
: public internal::dense_xpr_base< NestByValue<ExpressionType> >::type
|
||||
{
|
||||
public:
|
||||
|
||||
typedef typename internal::dense_xpr_base<NestByValue>::type Base;
|
||||
EIGEN_DENSE_PUBLIC_INTERFACE(NestByValue)
|
||||
|
||||
EIGEN_DEVICE_FUNC explicit inline NestByValue(const ExpressionType& matrix) : m_expression(matrix) {}
|
||||
|
||||
EIGEN_DEVICE_FUNC EIGEN_CONSTEXPR inline Index rows() const EIGEN_NOEXCEPT { return m_expression.rows(); }
|
||||
EIGEN_DEVICE_FUNC EIGEN_CONSTEXPR inline Index cols() const EIGEN_NOEXCEPT { return m_expression.cols(); }
|
||||
|
||||
EIGEN_DEVICE_FUNC operator const ExpressionType&() const { return m_expression; }
|
||||
|
||||
EIGEN_DEVICE_FUNC const ExpressionType& nestedExpression() const { return m_expression; }
|
||||
|
||||
protected:
|
||||
const ExpressionType m_expression;
|
||||
};
|
||||
|
||||
/** \returns an expression of the temporary version of *this.
|
||||
*/
|
||||
template<typename Derived>
|
||||
EIGEN_DEVICE_FUNC inline const NestByValue<Derived>
|
||||
DenseBase<Derived>::nestByValue() const
|
||||
{
|
||||
return NestByValue<Derived>(derived());
|
||||
}
|
||||
|
||||
namespace internal {
|
||||
|
||||
// Evaluator of Solve -> eval into a temporary
|
||||
template<typename ArgType>
|
||||
struct evaluator<NestByValue<ArgType> >
|
||||
: public evaluator<ArgType>
|
||||
{
|
||||
typedef evaluator<ArgType> Base;
|
||||
|
||||
EIGEN_DEVICE_FUNC explicit evaluator(const NestByValue<ArgType>& xpr)
|
||||
: Base(xpr.nestedExpression())
|
||||
{}
|
||||
};
|
||||
}
|
||||
|
||||
} // end namespace Eigen
|
||||
|
||||
#endif // EIGEN_NESTBYVALUE_H
|
||||
@@ -0,0 +1,109 @@
|
||||
// This file is part of Eigen, a lightweight C++ template library
|
||||
// for linear algebra.
|
||||
//
|
||||
// Copyright (C) 2009 Gael Guennebaud <gael.guennebaud@inria.fr>
|
||||
//
|
||||
// This Source Code Form is subject to the terms of the Mozilla
|
||||
// Public License v. 2.0. If a copy of the MPL was not distributed
|
||||
// with this file, You can obtain one at http://mozilla.org/MPL/2.0/.
|
||||
|
||||
#ifndef EIGEN_NOALIAS_H
|
||||
#define EIGEN_NOALIAS_H
|
||||
|
||||
namespace Eigen {
|
||||
|
||||
/** \class NoAlias
|
||||
* \ingroup Core_Module
|
||||
*
|
||||
* \brief Pseudo expression providing an operator = assuming no aliasing
|
||||
*
|
||||
* \tparam ExpressionType the type of the object on which to do the lazy assignment
|
||||
*
|
||||
* This class represents an expression with special assignment operators
|
||||
* assuming no aliasing between the target expression and the source expression.
|
||||
* More precisely it alloas to bypass the EvalBeforeAssignBit flag of the source expression.
|
||||
* It is the return type of MatrixBase::noalias()
|
||||
* and most of the time this is the only way it is used.
|
||||
*
|
||||
* \sa MatrixBase::noalias()
|
||||
*/
|
||||
template<typename ExpressionType, template <typename> class StorageBase>
|
||||
class NoAlias
|
||||
{
|
||||
public:
|
||||
typedef typename ExpressionType::Scalar Scalar;
|
||||
|
||||
EIGEN_DEVICE_FUNC
|
||||
explicit NoAlias(ExpressionType& expression) : m_expression(expression) {}
|
||||
|
||||
template<typename OtherDerived>
|
||||
EIGEN_DEVICE_FUNC
|
||||
EIGEN_STRONG_INLINE ExpressionType& operator=(const StorageBase<OtherDerived>& other)
|
||||
{
|
||||
call_assignment_no_alias(m_expression, other.derived(), internal::assign_op<Scalar,typename OtherDerived::Scalar>());
|
||||
return m_expression;
|
||||
}
|
||||
|
||||
template<typename OtherDerived>
|
||||
EIGEN_DEVICE_FUNC
|
||||
EIGEN_STRONG_INLINE ExpressionType& operator+=(const StorageBase<OtherDerived>& other)
|
||||
{
|
||||
call_assignment_no_alias(m_expression, other.derived(), internal::add_assign_op<Scalar,typename OtherDerived::Scalar>());
|
||||
return m_expression;
|
||||
}
|
||||
|
||||
template<typename OtherDerived>
|
||||
EIGEN_DEVICE_FUNC
|
||||
EIGEN_STRONG_INLINE ExpressionType& operator-=(const StorageBase<OtherDerived>& other)
|
||||
{
|
||||
call_assignment_no_alias(m_expression, other.derived(), internal::sub_assign_op<Scalar,typename OtherDerived::Scalar>());
|
||||
return m_expression;
|
||||
}
|
||||
|
||||
EIGEN_DEVICE_FUNC
|
||||
ExpressionType& expression() const
|
||||
{
|
||||
return m_expression;
|
||||
}
|
||||
|
||||
protected:
|
||||
ExpressionType& m_expression;
|
||||
};
|
||||
|
||||
/** \returns a pseudo expression of \c *this with an operator= assuming
|
||||
* no aliasing between \c *this and the source expression.
|
||||
*
|
||||
* More precisely, noalias() allows to bypass the EvalBeforeAssignBit flag.
|
||||
* Currently, even though several expressions may alias, only product
|
||||
* expressions have this flag. Therefore, noalias() is only useful when
|
||||
* the source expression contains a matrix product.
|
||||
*
|
||||
* Here are some examples where noalias is useful:
|
||||
* \code
|
||||
* D.noalias() = A * B;
|
||||
* D.noalias() += A.transpose() * B;
|
||||
* D.noalias() -= 2 * A * B.adjoint();
|
||||
* \endcode
|
||||
*
|
||||
* On the other hand the following example will lead to a \b wrong result:
|
||||
* \code
|
||||
* A.noalias() = A * B;
|
||||
* \endcode
|
||||
* because the result matrix A is also an operand of the matrix product. Therefore,
|
||||
* there is no alternative than evaluating A * B in a temporary, that is the default
|
||||
* behavior when you write:
|
||||
* \code
|
||||
* A = A * B;
|
||||
* \endcode
|
||||
*
|
||||
* \sa class NoAlias
|
||||
*/
|
||||
template<typename Derived>
|
||||
NoAlias<Derived,MatrixBase> EIGEN_DEVICE_FUNC MatrixBase<Derived>::noalias()
|
||||
{
|
||||
return NoAlias<Derived, Eigen::MatrixBase >(derived());
|
||||
}
|
||||
|
||||
} // end namespace Eigen
|
||||
|
||||
#endif // EIGEN_NOALIAS_H
|
||||
@@ -0,0 +1,335 @@
|
||||
// This file is part of Eigen, a lightweight C++ template library
|
||||
// for linear algebra.
|
||||
//
|
||||
// Copyright (C) 2006-2010 Benoit Jacob <jacob.benoit.1@gmail.com>
|
||||
//
|
||||
// This Source Code Form is subject to the terms of the Mozilla
|
||||
// Public License v. 2.0. If a copy of the MPL was not distributed
|
||||
// with this file, You can obtain one at http://mozilla.org/MPL/2.0/.
|
||||
|
||||
#ifndef EIGEN_NUMTRAITS_H
|
||||
#define EIGEN_NUMTRAITS_H
|
||||
|
||||
namespace Eigen {
|
||||
|
||||
namespace internal {
|
||||
|
||||
// default implementation of digits10(), based on numeric_limits if specialized,
|
||||
// 0 for integer types, and log10(epsilon()) otherwise.
|
||||
template< typename T,
|
||||
bool use_numeric_limits = std::numeric_limits<T>::is_specialized,
|
||||
bool is_integer = NumTraits<T>::IsInteger>
|
||||
struct default_digits10_impl
|
||||
{
|
||||
EIGEN_DEVICE_FUNC EIGEN_CONSTEXPR
|
||||
static int run() { return std::numeric_limits<T>::digits10; }
|
||||
};
|
||||
|
||||
template<typename T>
|
||||
struct default_digits10_impl<T,false,false> // Floating point
|
||||
{
|
||||
EIGEN_DEVICE_FUNC EIGEN_CONSTEXPR
|
||||
static int run() {
|
||||
using std::log10;
|
||||
using std::ceil;
|
||||
typedef typename NumTraits<T>::Real Real;
|
||||
return int(ceil(-log10(NumTraits<Real>::epsilon())));
|
||||
}
|
||||
};
|
||||
|
||||
template<typename T>
|
||||
struct default_digits10_impl<T,false,true> // Integer
|
||||
{
|
||||
EIGEN_DEVICE_FUNC EIGEN_CONSTEXPR
|
||||
static int run() { return 0; }
|
||||
};
|
||||
|
||||
|
||||
// default implementation of digits(), based on numeric_limits if specialized,
|
||||
// 0 for integer types, and log2(epsilon()) otherwise.
|
||||
template< typename T,
|
||||
bool use_numeric_limits = std::numeric_limits<T>::is_specialized,
|
||||
bool is_integer = NumTraits<T>::IsInteger>
|
||||
struct default_digits_impl
|
||||
{
|
||||
EIGEN_DEVICE_FUNC EIGEN_CONSTEXPR
|
||||
static int run() { return std::numeric_limits<T>::digits; }
|
||||
};
|
||||
|
||||
template<typename T>
|
||||
struct default_digits_impl<T,false,false> // Floating point
|
||||
{
|
||||
EIGEN_DEVICE_FUNC EIGEN_CONSTEXPR
|
||||
static int run() {
|
||||
using std::log;
|
||||
using std::ceil;
|
||||
typedef typename NumTraits<T>::Real Real;
|
||||
return int(ceil(-log(NumTraits<Real>::epsilon())/log(static_cast<Real>(2))));
|
||||
}
|
||||
};
|
||||
|
||||
template<typename T>
|
||||
struct default_digits_impl<T,false,true> // Integer
|
||||
{
|
||||
EIGEN_DEVICE_FUNC EIGEN_CONSTEXPR
|
||||
static int run() { return 0; }
|
||||
};
|
||||
|
||||
} // end namespace internal
|
||||
|
||||
namespace numext {
|
||||
/** \internal bit-wise cast without changing the underlying bit representation. */
|
||||
|
||||
// TODO: Replace by std::bit_cast (available in C++20)
|
||||
template <typename Tgt, typename Src>
|
||||
EIGEN_STRONG_INLINE EIGEN_DEVICE_FUNC Tgt bit_cast(const Src& src) {
|
||||
#if EIGEN_HAS_TYPE_TRAITS
|
||||
// The behaviour of memcpy is not specified for non-trivially copyable types
|
||||
EIGEN_STATIC_ASSERT(std::is_trivially_copyable<Src>::value, THIS_TYPE_IS_NOT_SUPPORTED);
|
||||
EIGEN_STATIC_ASSERT(std::is_trivially_copyable<Tgt>::value && std::is_default_constructible<Tgt>::value,
|
||||
THIS_TYPE_IS_NOT_SUPPORTED);
|
||||
#endif
|
||||
|
||||
EIGEN_STATIC_ASSERT(sizeof(Src) == sizeof(Tgt), THIS_TYPE_IS_NOT_SUPPORTED);
|
||||
Tgt tgt;
|
||||
EIGEN_USING_STD(memcpy)
|
||||
memcpy(&tgt, &src, sizeof(Tgt));
|
||||
return tgt;
|
||||
}
|
||||
} // namespace numext
|
||||
|
||||
/** \class NumTraits
|
||||
* \ingroup Core_Module
|
||||
*
|
||||
* \brief Holds information about the various numeric (i.e. scalar) types allowed by Eigen.
|
||||
*
|
||||
* \tparam T the numeric type at hand
|
||||
*
|
||||
* This class stores enums, typedefs and static methods giving information about a numeric type.
|
||||
*
|
||||
* The provided data consists of:
|
||||
* \li A typedef \c Real, giving the "real part" type of \a T. If \a T is already real,
|
||||
* then \c Real is just a typedef to \a T. If \a T is \c std::complex<U> then \c Real
|
||||
* is a typedef to \a U.
|
||||
* \li A typedef \c NonInteger, giving the type that should be used for operations producing non-integral values,
|
||||
* such as quotients, square roots, etc. If \a T is a floating-point type, then this typedef just gives
|
||||
* \a T again. Note however that many Eigen functions such as internal::sqrt simply refuse to
|
||||
* take integers. Outside of a few cases, Eigen doesn't do automatic type promotion. Thus, this typedef is
|
||||
* only intended as a helper for code that needs to explicitly promote types.
|
||||
* \li A typedef \c Literal giving the type to use for numeric literals such as "2" or "0.5". For instance, for \c std::complex<U>, Literal is defined as \c U.
|
||||
* Of course, this type must be fully compatible with \a T. In doubt, just use \a T here.
|
||||
* \li A typedef \a Nested giving the type to use to nest a value inside of the expression tree. If you don't know what
|
||||
* this means, just use \a T here.
|
||||
* \li An enum value \a IsComplex. It is equal to 1 if \a T is a \c std::complex
|
||||
* type, and to 0 otherwise.
|
||||
* \li An enum value \a IsInteger. It is equal to \c 1 if \a T is an integer type such as \c int,
|
||||
* and to \c 0 otherwise.
|
||||
* \li Enum values ReadCost, AddCost and MulCost representing a rough estimate of the number of CPU cycles needed
|
||||
* to by move / add / mul instructions respectively, assuming the data is already stored in CPU registers.
|
||||
* Stay vague here. No need to do architecture-specific stuff. If you don't know what this means, just use \c Eigen::HugeCost.
|
||||
* \li An enum value \a IsSigned. It is equal to \c 1 if \a T is a signed type and to 0 if \a T is unsigned.
|
||||
* \li An enum value \a RequireInitialization. It is equal to \c 1 if the constructor of the numeric type \a T must
|
||||
* be called, and to 0 if it is safe not to call it. Default is 0 if \a T is an arithmetic type, and 1 otherwise.
|
||||
* \li An epsilon() function which, unlike <a href="http://en.cppreference.com/w/cpp/types/numeric_limits/epsilon">std::numeric_limits::epsilon()</a>,
|
||||
* it returns a \a Real instead of a \a T.
|
||||
* \li A dummy_precision() function returning a weak epsilon value. It is mainly used as a default
|
||||
* value by the fuzzy comparison operators.
|
||||
* \li highest() and lowest() functions returning the highest and lowest possible values respectively.
|
||||
* \li digits() function returning the number of radix digits (non-sign digits for integers, mantissa for floating-point). This is
|
||||
* the analogue of <a href="http://en.cppreference.com/w/cpp/types/numeric_limits/digits">std::numeric_limits<T>::digits</a>
|
||||
* which is used as the default implementation if specialized.
|
||||
* \li digits10() function returning the number of decimal digits that can be represented without change. This is
|
||||
* the analogue of <a href="http://en.cppreference.com/w/cpp/types/numeric_limits/digits10">std::numeric_limits<T>::digits10</a>
|
||||
* which is used as the default implementation if specialized.
|
||||
* \li min_exponent() and max_exponent() functions returning the highest and lowest possible values, respectively,
|
||||
* such that the radix raised to the power exponent-1 is a normalized floating-point number. These are equivalent to
|
||||
* <a href="http://en.cppreference.com/w/cpp/types/numeric_limits/min_exponent">std::numeric_limits<T>::min_exponent</a>/
|
||||
* <a href="http://en.cppreference.com/w/cpp/types/numeric_limits/max_exponent">std::numeric_limits<T>::max_exponent</a>.
|
||||
* \li infinity() function returning a representation of positive infinity, if available.
|
||||
* \li quiet_NaN function returning a non-signaling "not-a-number", if available.
|
||||
*/
|
||||
|
||||
template<typename T> struct GenericNumTraits
|
||||
{
|
||||
enum {
|
||||
IsInteger = std::numeric_limits<T>::is_integer,
|
||||
IsSigned = std::numeric_limits<T>::is_signed,
|
||||
IsComplex = 0,
|
||||
RequireInitialization = internal::is_arithmetic<T>::value ? 0 : 1,
|
||||
ReadCost = 1,
|
||||
AddCost = 1,
|
||||
MulCost = 1
|
||||
};
|
||||
|
||||
typedef T Real;
|
||||
typedef typename internal::conditional<
|
||||
IsInteger,
|
||||
typename internal::conditional<sizeof(T)<=2, float, double>::type,
|
||||
T
|
||||
>::type NonInteger;
|
||||
typedef T Nested;
|
||||
typedef T Literal;
|
||||
|
||||
EIGEN_DEVICE_FUNC EIGEN_CONSTEXPR
|
||||
static inline Real epsilon()
|
||||
{
|
||||
return numext::numeric_limits<T>::epsilon();
|
||||
}
|
||||
|
||||
EIGEN_DEVICE_FUNC EIGEN_CONSTEXPR
|
||||
static inline int digits10()
|
||||
{
|
||||
return internal::default_digits10_impl<T>::run();
|
||||
}
|
||||
|
||||
EIGEN_DEVICE_FUNC EIGEN_CONSTEXPR
|
||||
static inline int digits()
|
||||
{
|
||||
return internal::default_digits_impl<T>::run();
|
||||
}
|
||||
|
||||
EIGEN_DEVICE_FUNC EIGEN_CONSTEXPR
|
||||
static inline int min_exponent()
|
||||
{
|
||||
return numext::numeric_limits<T>::min_exponent;
|
||||
}
|
||||
|
||||
EIGEN_DEVICE_FUNC EIGEN_CONSTEXPR
|
||||
static inline int max_exponent()
|
||||
{
|
||||
return numext::numeric_limits<T>::max_exponent;
|
||||
}
|
||||
|
||||
EIGEN_DEVICE_FUNC EIGEN_CONSTEXPR
|
||||
static inline Real dummy_precision()
|
||||
{
|
||||
// make sure to override this for floating-point types
|
||||
return Real(0);
|
||||
}
|
||||
|
||||
EIGEN_DEVICE_FUNC EIGEN_CONSTEXPR
|
||||
static inline T highest() {
|
||||
return (numext::numeric_limits<T>::max)();
|
||||
}
|
||||
|
||||
EIGEN_DEVICE_FUNC EIGEN_CONSTEXPR
|
||||
static inline T lowest() {
|
||||
return IsInteger ? (numext::numeric_limits<T>::min)()
|
||||
: static_cast<T>(-(numext::numeric_limits<T>::max)());
|
||||
}
|
||||
|
||||
EIGEN_DEVICE_FUNC EIGEN_CONSTEXPR
|
||||
static inline T infinity() {
|
||||
return numext::numeric_limits<T>::infinity();
|
||||
}
|
||||
|
||||
EIGEN_DEVICE_FUNC EIGEN_CONSTEXPR
|
||||
static inline T quiet_NaN() {
|
||||
return numext::numeric_limits<T>::quiet_NaN();
|
||||
}
|
||||
};
|
||||
|
||||
template<typename T> struct NumTraits : GenericNumTraits<T>
|
||||
{};
|
||||
|
||||
template<> struct NumTraits<float>
|
||||
: GenericNumTraits<float>
|
||||
{
|
||||
EIGEN_DEVICE_FUNC EIGEN_CONSTEXPR
|
||||
static inline float dummy_precision() { return 1e-5f; }
|
||||
};
|
||||
|
||||
template<> struct NumTraits<double> : GenericNumTraits<double>
|
||||
{
|
||||
EIGEN_DEVICE_FUNC EIGEN_CONSTEXPR
|
||||
static inline double dummy_precision() { return 1e-12; }
|
||||
};
|
||||
|
||||
template<> struct NumTraits<long double>
|
||||
: GenericNumTraits<long double>
|
||||
{
|
||||
EIGEN_CONSTEXPR
|
||||
static inline long double dummy_precision() { return 1e-15l; }
|
||||
};
|
||||
|
||||
template<typename _Real> struct NumTraits<std::complex<_Real> >
|
||||
: GenericNumTraits<std::complex<_Real> >
|
||||
{
|
||||
typedef _Real Real;
|
||||
typedef typename NumTraits<_Real>::Literal Literal;
|
||||
enum {
|
||||
IsComplex = 1,
|
||||
RequireInitialization = NumTraits<_Real>::RequireInitialization,
|
||||
ReadCost = 2 * NumTraits<_Real>::ReadCost,
|
||||
AddCost = 2 * NumTraits<Real>::AddCost,
|
||||
MulCost = 4 * NumTraits<Real>::MulCost + 2 * NumTraits<Real>::AddCost
|
||||
};
|
||||
|
||||
EIGEN_DEVICE_FUNC EIGEN_CONSTEXPR
|
||||
static inline Real epsilon() { return NumTraits<Real>::epsilon(); }
|
||||
EIGEN_DEVICE_FUNC EIGEN_CONSTEXPR
|
||||
static inline Real dummy_precision() { return NumTraits<Real>::dummy_precision(); }
|
||||
EIGEN_DEVICE_FUNC EIGEN_CONSTEXPR
|
||||
static inline int digits10() { return NumTraits<Real>::digits10(); }
|
||||
};
|
||||
|
||||
template<typename Scalar, int Rows, int Cols, int Options, int MaxRows, int MaxCols>
|
||||
struct NumTraits<Array<Scalar, Rows, Cols, Options, MaxRows, MaxCols> >
|
||||
{
|
||||
typedef Array<Scalar, Rows, Cols, Options, MaxRows, MaxCols> ArrayType;
|
||||
typedef typename NumTraits<Scalar>::Real RealScalar;
|
||||
typedef Array<RealScalar, Rows, Cols, Options, MaxRows, MaxCols> Real;
|
||||
typedef typename NumTraits<Scalar>::NonInteger NonIntegerScalar;
|
||||
typedef Array<NonIntegerScalar, Rows, Cols, Options, MaxRows, MaxCols> NonInteger;
|
||||
typedef ArrayType & Nested;
|
||||
typedef typename NumTraits<Scalar>::Literal Literal;
|
||||
|
||||
enum {
|
||||
IsComplex = NumTraits<Scalar>::IsComplex,
|
||||
IsInteger = NumTraits<Scalar>::IsInteger,
|
||||
IsSigned = NumTraits<Scalar>::IsSigned,
|
||||
RequireInitialization = 1,
|
||||
ReadCost = ArrayType::SizeAtCompileTime==Dynamic ? HugeCost : ArrayType::SizeAtCompileTime * int(NumTraits<Scalar>::ReadCost),
|
||||
AddCost = ArrayType::SizeAtCompileTime==Dynamic ? HugeCost : ArrayType::SizeAtCompileTime * int(NumTraits<Scalar>::AddCost),
|
||||
MulCost = ArrayType::SizeAtCompileTime==Dynamic ? HugeCost : ArrayType::SizeAtCompileTime * int(NumTraits<Scalar>::MulCost)
|
||||
};
|
||||
|
||||
EIGEN_DEVICE_FUNC EIGEN_CONSTEXPR
|
||||
static inline RealScalar epsilon() { return NumTraits<RealScalar>::epsilon(); }
|
||||
EIGEN_DEVICE_FUNC EIGEN_CONSTEXPR
|
||||
static inline RealScalar dummy_precision() { return NumTraits<RealScalar>::dummy_precision(); }
|
||||
|
||||
EIGEN_CONSTEXPR
|
||||
static inline int digits10() { return NumTraits<Scalar>::digits10(); }
|
||||
};
|
||||
|
||||
template<> struct NumTraits<std::string>
|
||||
: GenericNumTraits<std::string>
|
||||
{
|
||||
enum {
|
||||
RequireInitialization = 1,
|
||||
ReadCost = HugeCost,
|
||||
AddCost = HugeCost,
|
||||
MulCost = HugeCost
|
||||
};
|
||||
|
||||
EIGEN_CONSTEXPR
|
||||
static inline int digits10() { return 0; }
|
||||
|
||||
private:
|
||||
static inline std::string epsilon();
|
||||
static inline std::string dummy_precision();
|
||||
static inline std::string lowest();
|
||||
static inline std::string highest();
|
||||
static inline std::string infinity();
|
||||
static inline std::string quiet_NaN();
|
||||
};
|
||||
|
||||
// Empty specialization for void to allow template specialization based on NumTraits<T>::Real with T==void and SFINAE.
|
||||
template<> struct NumTraits<void> {};
|
||||
|
||||
template<> struct NumTraits<bool> : GenericNumTraits<bool> {};
|
||||
|
||||
} // end namespace Eigen
|
||||
|
||||
#endif // EIGEN_NUMTRAITS_H
|
||||
@@ -0,0 +1,232 @@
|
||||
// This file is part of Eigen, a lightweight C++ template library
|
||||
// for linear algebra.
|
||||
//
|
||||
// Copyright (C) 2011-2018 Gael Guennebaud <gael.guennebaud@inria.fr>
|
||||
//
|
||||
// This Source Code Form is subject to the terms of the Mozilla
|
||||
// Public License v. 2.0. If a copy of the MPL was not distributed
|
||||
// with this file, You can obtain one at http://mozilla.org/MPL/2.0/.
|
||||
|
||||
#ifndef EIGEN_PARTIALREDUX_H
|
||||
#define EIGEN_PARTIALREDUX_H
|
||||
|
||||
namespace Eigen {
|
||||
|
||||
namespace internal {
|
||||
|
||||
|
||||
/***************************************************************************
|
||||
*
|
||||
* This file provides evaluators for partial reductions.
|
||||
* There are two modes:
|
||||
*
|
||||
* - scalar path: simply calls the respective function on the column or row.
|
||||
* -> nothing special here, all the tricky part is handled by the return
|
||||
* types of VectorwiseOp's members. They embed the functor calling the
|
||||
* respective DenseBase's member function.
|
||||
*
|
||||
* - vectorized path: implements a packet-wise reductions followed by
|
||||
* some (optional) processing of the outcome, e.g., division by n for mean.
|
||||
*
|
||||
* For the vectorized path let's observe that the packet-size and outer-unrolling
|
||||
* are both decided by the assignement logic. So all we have to do is to decide
|
||||
* on the inner unrolling.
|
||||
*
|
||||
* For the unrolling, we can reuse "internal::redux_vec_unroller" from Redux.h,
|
||||
* but be need to be careful to specify correct increment.
|
||||
*
|
||||
***************************************************************************/
|
||||
|
||||
|
||||
/* logic deciding a strategy for unrolling of vectorized paths */
|
||||
template<typename Func, typename Evaluator>
|
||||
struct packetwise_redux_traits
|
||||
{
|
||||
enum {
|
||||
OuterSize = int(Evaluator::IsRowMajor) ? Evaluator::RowsAtCompileTime : Evaluator::ColsAtCompileTime,
|
||||
Cost = OuterSize == Dynamic ? HugeCost
|
||||
: OuterSize * Evaluator::CoeffReadCost + (OuterSize-1) * functor_traits<Func>::Cost,
|
||||
Unrolling = Cost <= EIGEN_UNROLLING_LIMIT ? CompleteUnrolling : NoUnrolling
|
||||
};
|
||||
|
||||
};
|
||||
|
||||
/* Value to be returned when size==0 , by default let's return 0 */
|
||||
template<typename PacketType,typename Func>
|
||||
EIGEN_DEVICE_FUNC
|
||||
PacketType packetwise_redux_empty_value(const Func& ) { return pset1<PacketType>(0); }
|
||||
|
||||
/* For products the default is 1 */
|
||||
template<typename PacketType,typename Scalar>
|
||||
EIGEN_DEVICE_FUNC
|
||||
PacketType packetwise_redux_empty_value(const scalar_product_op<Scalar,Scalar>& ) { return pset1<PacketType>(1); }
|
||||
|
||||
/* Perform the actual reduction */
|
||||
template<typename Func, typename Evaluator,
|
||||
int Unrolling = packetwise_redux_traits<Func, Evaluator>::Unrolling
|
||||
>
|
||||
struct packetwise_redux_impl;
|
||||
|
||||
/* Perform the actual reduction with unrolling */
|
||||
template<typename Func, typename Evaluator>
|
||||
struct packetwise_redux_impl<Func, Evaluator, CompleteUnrolling>
|
||||
{
|
||||
typedef redux_novec_unroller<Func,Evaluator, 0, Evaluator::SizeAtCompileTime> Base;
|
||||
typedef typename Evaluator::Scalar Scalar;
|
||||
|
||||
template<typename PacketType>
|
||||
EIGEN_DEVICE_FUNC static EIGEN_STRONG_INLINE
|
||||
PacketType run(const Evaluator &eval, const Func& func, Index /*size*/)
|
||||
{
|
||||
return redux_vec_unroller<Func, Evaluator, 0, packetwise_redux_traits<Func, Evaluator>::OuterSize>::template run<PacketType>(eval,func);
|
||||
}
|
||||
};
|
||||
|
||||
/* Add a specialization of redux_vec_unroller for size==0 at compiletime.
|
||||
* This specialization is not required for general reductions, which is
|
||||
* why it is defined here.
|
||||
*/
|
||||
template<typename Func, typename Evaluator, int Start>
|
||||
struct redux_vec_unroller<Func, Evaluator, Start, 0>
|
||||
{
|
||||
template<typename PacketType>
|
||||
EIGEN_DEVICE_FUNC
|
||||
static EIGEN_STRONG_INLINE PacketType run(const Evaluator &, const Func& f)
|
||||
{
|
||||
return packetwise_redux_empty_value<PacketType>(f);
|
||||
}
|
||||
};
|
||||
|
||||
/* Perform the actual reduction for dynamic sizes */
|
||||
template<typename Func, typename Evaluator>
|
||||
struct packetwise_redux_impl<Func, Evaluator, NoUnrolling>
|
||||
{
|
||||
typedef typename Evaluator::Scalar Scalar;
|
||||
typedef typename redux_traits<Func, Evaluator>::PacketType PacketScalar;
|
||||
|
||||
template<typename PacketType>
|
||||
EIGEN_DEVICE_FUNC
|
||||
static PacketType run(const Evaluator &eval, const Func& func, Index size)
|
||||
{
|
||||
if(size==0)
|
||||
return packetwise_redux_empty_value<PacketType>(func);
|
||||
|
||||
const Index size4 = (size-1)&(~3);
|
||||
PacketType p = eval.template packetByOuterInner<Unaligned,PacketType>(0,0);
|
||||
Index i = 1;
|
||||
// This loop is optimized for instruction pipelining:
|
||||
// - each iteration generates two independent instructions
|
||||
// - thanks to branch prediction and out-of-order execution we have independent instructions across loops
|
||||
for(; i<size4; i+=4)
|
||||
p = func.packetOp(p,
|
||||
func.packetOp(
|
||||
func.packetOp(eval.template packetByOuterInner<Unaligned,PacketType>(i+0,0),eval.template packetByOuterInner<Unaligned,PacketType>(i+1,0)),
|
||||
func.packetOp(eval.template packetByOuterInner<Unaligned,PacketType>(i+2,0),eval.template packetByOuterInner<Unaligned,PacketType>(i+3,0))));
|
||||
for(; i<size; ++i)
|
||||
p = func.packetOp(p, eval.template packetByOuterInner<Unaligned,PacketType>(i,0));
|
||||
return p;
|
||||
}
|
||||
};
|
||||
|
||||
template< typename ArgType, typename MemberOp, int Direction>
|
||||
struct evaluator<PartialReduxExpr<ArgType, MemberOp, Direction> >
|
||||
: evaluator_base<PartialReduxExpr<ArgType, MemberOp, Direction> >
|
||||
{
|
||||
typedef PartialReduxExpr<ArgType, MemberOp, Direction> XprType;
|
||||
typedef typename internal::nested_eval<ArgType,1>::type ArgTypeNested;
|
||||
typedef typename internal::add_const_on_value_type<ArgTypeNested>::type ConstArgTypeNested;
|
||||
typedef typename internal::remove_all<ArgTypeNested>::type ArgTypeNestedCleaned;
|
||||
typedef typename ArgType::Scalar InputScalar;
|
||||
typedef typename XprType::Scalar Scalar;
|
||||
enum {
|
||||
TraversalSize = Direction==int(Vertical) ? int(ArgType::RowsAtCompileTime) : int(ArgType::ColsAtCompileTime)
|
||||
};
|
||||
typedef typename MemberOp::template Cost<int(TraversalSize)> CostOpType;
|
||||
enum {
|
||||
CoeffReadCost = TraversalSize==Dynamic ? HugeCost
|
||||
: TraversalSize==0 ? 1
|
||||
: int(TraversalSize) * int(evaluator<ArgType>::CoeffReadCost) + int(CostOpType::value),
|
||||
|
||||
_ArgFlags = evaluator<ArgType>::Flags,
|
||||
|
||||
_Vectorizable = bool(int(_ArgFlags)&PacketAccessBit)
|
||||
&& bool(MemberOp::Vectorizable)
|
||||
&& (Direction==int(Vertical) ? bool(_ArgFlags&RowMajorBit) : (_ArgFlags&RowMajorBit)==0)
|
||||
&& (TraversalSize!=0),
|
||||
|
||||
Flags = (traits<XprType>::Flags&RowMajorBit)
|
||||
| (evaluator<ArgType>::Flags&(HereditaryBits&(~RowMajorBit)))
|
||||
| (_Vectorizable ? PacketAccessBit : 0)
|
||||
| LinearAccessBit,
|
||||
|
||||
Alignment = 0 // FIXME this will need to be improved once PartialReduxExpr is vectorized
|
||||
};
|
||||
|
||||
EIGEN_DEVICE_FUNC explicit evaluator(const XprType xpr)
|
||||
: m_arg(xpr.nestedExpression()), m_functor(xpr.functor())
|
||||
{
|
||||
EIGEN_INTERNAL_CHECK_COST_VALUE(TraversalSize==Dynamic ? HugeCost : (TraversalSize==0 ? 1 : int(CostOpType::value)));
|
||||
EIGEN_INTERNAL_CHECK_COST_VALUE(CoeffReadCost);
|
||||
}
|
||||
|
||||
typedef typename XprType::CoeffReturnType CoeffReturnType;
|
||||
|
||||
EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
|
||||
const Scalar coeff(Index i, Index j) const
|
||||
{
|
||||
return coeff(Direction==Vertical ? j : i);
|
||||
}
|
||||
|
||||
EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
|
||||
const Scalar coeff(Index index) const
|
||||
{
|
||||
return m_functor(m_arg.template subVector<DirectionType(Direction)>(index));
|
||||
}
|
||||
|
||||
template<int LoadMode,typename PacketType>
|
||||
EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
|
||||
PacketType packet(Index i, Index j) const
|
||||
{
|
||||
return packet<LoadMode,PacketType>(Direction==Vertical ? j : i);
|
||||
}
|
||||
|
||||
template<int LoadMode,typename PacketType>
|
||||
EIGEN_STRONG_INLINE EIGEN_DEVICE_FUNC
|
||||
PacketType packet(Index idx) const
|
||||
{
|
||||
enum { PacketSize = internal::unpacket_traits<PacketType>::size };
|
||||
typedef Block<const ArgTypeNestedCleaned,
|
||||
Direction==Vertical ? int(ArgType::RowsAtCompileTime) : int(PacketSize),
|
||||
Direction==Vertical ? int(PacketSize) : int(ArgType::ColsAtCompileTime),
|
||||
true /* InnerPanel */> PanelType;
|
||||
|
||||
PanelType panel(m_arg,
|
||||
Direction==Vertical ? 0 : idx,
|
||||
Direction==Vertical ? idx : 0,
|
||||
Direction==Vertical ? m_arg.rows() : Index(PacketSize),
|
||||
Direction==Vertical ? Index(PacketSize) : m_arg.cols());
|
||||
|
||||
// FIXME
|
||||
// See bug 1612, currently if PacketSize==1 (i.e. complex<double> with 128bits registers) then the storage-order of panel get reversed
|
||||
// and methods like packetByOuterInner do not make sense anymore in this context.
|
||||
// So let's just by pass "vectorization" in this case:
|
||||
if(PacketSize==1)
|
||||
return internal::pset1<PacketType>(coeff(idx));
|
||||
|
||||
typedef typename internal::redux_evaluator<PanelType> PanelEvaluator;
|
||||
PanelEvaluator panel_eval(panel);
|
||||
typedef typename MemberOp::BinaryOp BinaryOp;
|
||||
PacketType p = internal::packetwise_redux_impl<BinaryOp,PanelEvaluator>::template run<PacketType>(panel_eval,m_functor.binaryFunc(),m_arg.outerSize());
|
||||
return p;
|
||||
}
|
||||
|
||||
protected:
|
||||
ConstArgTypeNested m_arg;
|
||||
const MemberOp m_functor;
|
||||
};
|
||||
|
||||
} // end namespace internal
|
||||
|
||||
} // end namespace Eigen
|
||||
|
||||
#endif // EIGEN_PARTIALREDUX_H
|
||||
@@ -0,0 +1,605 @@
|
||||
// This file is part of Eigen, a lightweight C++ template library
|
||||
// for linear algebra.
|
||||
//
|
||||
// Copyright (C) 2009 Benoit Jacob <jacob.benoit.1@gmail.com>
|
||||
// Copyright (C) 2009-2015 Gael Guennebaud <gael.guennebaud@inria.fr>
|
||||
//
|
||||
// This Source Code Form is subject to the terms of the Mozilla
|
||||
// Public License v. 2.0. If a copy of the MPL was not distributed
|
||||
// with this file, You can obtain one at http://mozilla.org/MPL/2.0/.
|
||||
|
||||
#ifndef EIGEN_PERMUTATIONMATRIX_H
|
||||
#define EIGEN_PERMUTATIONMATRIX_H
|
||||
|
||||
namespace Eigen {
|
||||
|
||||
namespace internal {
|
||||
|
||||
enum PermPermProduct_t {PermPermProduct};
|
||||
|
||||
} // end namespace internal
|
||||
|
||||
/** \class PermutationBase
|
||||
* \ingroup Core_Module
|
||||
*
|
||||
* \brief Base class for permutations
|
||||
*
|
||||
* \tparam Derived the derived class
|
||||
*
|
||||
* This class is the base class for all expressions representing a permutation matrix,
|
||||
* internally stored as a vector of integers.
|
||||
* The convention followed here is that if \f$ \sigma \f$ is a permutation, the corresponding permutation matrix
|
||||
* \f$ P_\sigma \f$ is such that if \f$ (e_1,\ldots,e_p) \f$ is the canonical basis, we have:
|
||||
* \f[ P_\sigma(e_i) = e_{\sigma(i)}. \f]
|
||||
* This convention ensures that for any two permutations \f$ \sigma, \tau \f$, we have:
|
||||
* \f[ P_{\sigma\circ\tau} = P_\sigma P_\tau. \f]
|
||||
*
|
||||
* Permutation matrices are square and invertible.
|
||||
*
|
||||
* Notice that in addition to the member functions and operators listed here, there also are non-member
|
||||
* operator* to multiply any kind of permutation object with any kind of matrix expression (MatrixBase)
|
||||
* on either side.
|
||||
*
|
||||
* \sa class PermutationMatrix, class PermutationWrapper
|
||||
*/
|
||||
template<typename Derived>
|
||||
class PermutationBase : public EigenBase<Derived>
|
||||
{
|
||||
typedef internal::traits<Derived> Traits;
|
||||
typedef EigenBase<Derived> Base;
|
||||
public:
|
||||
|
||||
#ifndef EIGEN_PARSED_BY_DOXYGEN
|
||||
typedef typename Traits::IndicesType IndicesType;
|
||||
enum {
|
||||
Flags = Traits::Flags,
|
||||
RowsAtCompileTime = Traits::RowsAtCompileTime,
|
||||
ColsAtCompileTime = Traits::ColsAtCompileTime,
|
||||
MaxRowsAtCompileTime = Traits::MaxRowsAtCompileTime,
|
||||
MaxColsAtCompileTime = Traits::MaxColsAtCompileTime
|
||||
};
|
||||
typedef typename Traits::StorageIndex StorageIndex;
|
||||
typedef Matrix<StorageIndex,RowsAtCompileTime,ColsAtCompileTime,0,MaxRowsAtCompileTime,MaxColsAtCompileTime>
|
||||
DenseMatrixType;
|
||||
typedef PermutationMatrix<IndicesType::SizeAtCompileTime,IndicesType::MaxSizeAtCompileTime,StorageIndex>
|
||||
PlainPermutationType;
|
||||
typedef PlainPermutationType PlainObject;
|
||||
using Base::derived;
|
||||
typedef Inverse<Derived> InverseReturnType;
|
||||
typedef void Scalar;
|
||||
#endif
|
||||
|
||||
/** Copies the other permutation into *this */
|
||||
template<typename OtherDerived>
|
||||
Derived& operator=(const PermutationBase<OtherDerived>& other)
|
||||
{
|
||||
indices() = other.indices();
|
||||
return derived();
|
||||
}
|
||||
|
||||
/** Assignment from the Transpositions \a tr */
|
||||
template<typename OtherDerived>
|
||||
Derived& operator=(const TranspositionsBase<OtherDerived>& tr)
|
||||
{
|
||||
setIdentity(tr.size());
|
||||
for(Index k=size()-1; k>=0; --k)
|
||||
applyTranspositionOnTheRight(k,tr.coeff(k));
|
||||
return derived();
|
||||
}
|
||||
|
||||
/** \returns the number of rows */
|
||||
inline EIGEN_DEVICE_FUNC Index rows() const { return Index(indices().size()); }
|
||||
|
||||
/** \returns the number of columns */
|
||||
inline EIGEN_DEVICE_FUNC Index cols() const { return Index(indices().size()); }
|
||||
|
||||
/** \returns the size of a side of the respective square matrix, i.e., the number of indices */
|
||||
inline EIGEN_DEVICE_FUNC Index size() const { return Index(indices().size()); }
|
||||
|
||||
#ifndef EIGEN_PARSED_BY_DOXYGEN
|
||||
template<typename DenseDerived>
|
||||
void evalTo(MatrixBase<DenseDerived>& other) const
|
||||
{
|
||||
other.setZero();
|
||||
for (Index i=0; i<rows(); ++i)
|
||||
other.coeffRef(indices().coeff(i),i) = typename DenseDerived::Scalar(1);
|
||||
}
|
||||
#endif
|
||||
|
||||
/** \returns a Matrix object initialized from this permutation matrix. Notice that it
|
||||
* is inefficient to return this Matrix object by value. For efficiency, favor using
|
||||
* the Matrix constructor taking EigenBase objects.
|
||||
*/
|
||||
DenseMatrixType toDenseMatrix() const
|
||||
{
|
||||
return derived();
|
||||
}
|
||||
|
||||
/** const version of indices(). */
|
||||
const IndicesType& indices() const { return derived().indices(); }
|
||||
/** \returns a reference to the stored array representing the permutation. */
|
||||
IndicesType& indices() { return derived().indices(); }
|
||||
|
||||
/** Resizes to given size.
|
||||
*/
|
||||
inline void resize(Index newSize)
|
||||
{
|
||||
indices().resize(newSize);
|
||||
}
|
||||
|
||||
/** Sets *this to be the identity permutation matrix */
|
||||
void setIdentity()
|
||||
{
|
||||
StorageIndex n = StorageIndex(size());
|
||||
for(StorageIndex i = 0; i < n; ++i)
|
||||
indices().coeffRef(i) = i;
|
||||
}
|
||||
|
||||
/** Sets *this to be the identity permutation matrix of given size.
|
||||
*/
|
||||
void setIdentity(Index newSize)
|
||||
{
|
||||
resize(newSize);
|
||||
setIdentity();
|
||||
}
|
||||
|
||||
/** Multiplies *this by the transposition \f$(ij)\f$ on the left.
|
||||
*
|
||||
* \returns a reference to *this.
|
||||
*
|
||||
* \warning This is much slower than applyTranspositionOnTheRight(Index,Index):
|
||||
* this has linear complexity and requires a lot of branching.
|
||||
*
|
||||
* \sa applyTranspositionOnTheRight(Index,Index)
|
||||
*/
|
||||
Derived& applyTranspositionOnTheLeft(Index i, Index j)
|
||||
{
|
||||
eigen_assert(i>=0 && j>=0 && i<size() && j<size());
|
||||
for(Index k = 0; k < size(); ++k)
|
||||
{
|
||||
if(indices().coeff(k) == i) indices().coeffRef(k) = StorageIndex(j);
|
||||
else if(indices().coeff(k) == j) indices().coeffRef(k) = StorageIndex(i);
|
||||
}
|
||||
return derived();
|
||||
}
|
||||
|
||||
/** Multiplies *this by the transposition \f$(ij)\f$ on the right.
|
||||
*
|
||||
* \returns a reference to *this.
|
||||
*
|
||||
* This is a fast operation, it only consists in swapping two indices.
|
||||
*
|
||||
* \sa applyTranspositionOnTheLeft(Index,Index)
|
||||
*/
|
||||
Derived& applyTranspositionOnTheRight(Index i, Index j)
|
||||
{
|
||||
eigen_assert(i>=0 && j>=0 && i<size() && j<size());
|
||||
std::swap(indices().coeffRef(i), indices().coeffRef(j));
|
||||
return derived();
|
||||
}
|
||||
|
||||
/** \returns the inverse permutation matrix.
|
||||
*
|
||||
* \note \blank \note_try_to_help_rvo
|
||||
*/
|
||||
inline InverseReturnType inverse() const
|
||||
{ return InverseReturnType(derived()); }
|
||||
/** \returns the tranpose permutation matrix.
|
||||
*
|
||||
* \note \blank \note_try_to_help_rvo
|
||||
*/
|
||||
inline InverseReturnType transpose() const
|
||||
{ return InverseReturnType(derived()); }
|
||||
|
||||
/**** multiplication helpers to hopefully get RVO ****/
|
||||
|
||||
|
||||
#ifndef EIGEN_PARSED_BY_DOXYGEN
|
||||
protected:
|
||||
template<typename OtherDerived>
|
||||
void assignTranspose(const PermutationBase<OtherDerived>& other)
|
||||
{
|
||||
for (Index i=0; i<rows();++i) indices().coeffRef(other.indices().coeff(i)) = i;
|
||||
}
|
||||
template<typename Lhs,typename Rhs>
|
||||
void assignProduct(const Lhs& lhs, const Rhs& rhs)
|
||||
{
|
||||
eigen_assert(lhs.cols() == rhs.rows());
|
||||
for (Index i=0; i<rows();++i) indices().coeffRef(i) = lhs.indices().coeff(rhs.indices().coeff(i));
|
||||
}
|
||||
#endif
|
||||
|
||||
public:
|
||||
|
||||
/** \returns the product permutation matrix.
|
||||
*
|
||||
* \note \blank \note_try_to_help_rvo
|
||||
*/
|
||||
template<typename Other>
|
||||
inline PlainPermutationType operator*(const PermutationBase<Other>& other) const
|
||||
{ return PlainPermutationType(internal::PermPermProduct, derived(), other.derived()); }
|
||||
|
||||
/** \returns the product of a permutation with another inverse permutation.
|
||||
*
|
||||
* \note \blank \note_try_to_help_rvo
|
||||
*/
|
||||
template<typename Other>
|
||||
inline PlainPermutationType operator*(const InverseImpl<Other,PermutationStorage>& other) const
|
||||
{ return PlainPermutationType(internal::PermPermProduct, *this, other.eval()); }
|
||||
|
||||
/** \returns the product of an inverse permutation with another permutation.
|
||||
*
|
||||
* \note \blank \note_try_to_help_rvo
|
||||
*/
|
||||
template<typename Other> friend
|
||||
inline PlainPermutationType operator*(const InverseImpl<Other, PermutationStorage>& other, const PermutationBase& perm)
|
||||
{ return PlainPermutationType(internal::PermPermProduct, other.eval(), perm); }
|
||||
|
||||
/** \returns the determinant of the permutation matrix, which is either 1 or -1 depending on the parity of the permutation.
|
||||
*
|
||||
* This function is O(\c n) procedure allocating a buffer of \c n booleans.
|
||||
*/
|
||||
Index determinant() const
|
||||
{
|
||||
Index res = 1;
|
||||
Index n = size();
|
||||
Matrix<bool,RowsAtCompileTime,1,0,MaxRowsAtCompileTime> mask(n);
|
||||
mask.fill(false);
|
||||
Index r = 0;
|
||||
while(r < n)
|
||||
{
|
||||
// search for the next seed
|
||||
while(r<n && mask[r]) r++;
|
||||
if(r>=n)
|
||||
break;
|
||||
// we got one, let's follow it until we are back to the seed
|
||||
Index k0 = r++;
|
||||
mask.coeffRef(k0) = true;
|
||||
for(Index k=indices().coeff(k0); k!=k0; k=indices().coeff(k))
|
||||
{
|
||||
mask.coeffRef(k) = true;
|
||||
res = -res;
|
||||
}
|
||||
}
|
||||
return res;
|
||||
}
|
||||
|
||||
protected:
|
||||
|
||||
};
|
||||
|
||||
namespace internal {
|
||||
template<int SizeAtCompileTime, int MaxSizeAtCompileTime, typename _StorageIndex>
|
||||
struct traits<PermutationMatrix<SizeAtCompileTime, MaxSizeAtCompileTime, _StorageIndex> >
|
||||
: traits<Matrix<_StorageIndex,SizeAtCompileTime,SizeAtCompileTime,0,MaxSizeAtCompileTime,MaxSizeAtCompileTime> >
|
||||
{
|
||||
typedef PermutationStorage StorageKind;
|
||||
typedef Matrix<_StorageIndex, SizeAtCompileTime, 1, 0, MaxSizeAtCompileTime, 1> IndicesType;
|
||||
typedef _StorageIndex StorageIndex;
|
||||
typedef void Scalar;
|
||||
};
|
||||
}
|
||||
|
||||
/** \class PermutationMatrix
|
||||
* \ingroup Core_Module
|
||||
*
|
||||
* \brief Permutation matrix
|
||||
*
|
||||
* \tparam SizeAtCompileTime the number of rows/cols, or Dynamic
|
||||
* \tparam MaxSizeAtCompileTime the maximum number of rows/cols, or Dynamic. This optional parameter defaults to SizeAtCompileTime. Most of the time, you should not have to specify it.
|
||||
* \tparam _StorageIndex the integer type of the indices
|
||||
*
|
||||
* This class represents a permutation matrix, internally stored as a vector of integers.
|
||||
*
|
||||
* \sa class PermutationBase, class PermutationWrapper, class DiagonalMatrix
|
||||
*/
|
||||
template<int SizeAtCompileTime, int MaxSizeAtCompileTime, typename _StorageIndex>
|
||||
class PermutationMatrix : public PermutationBase<PermutationMatrix<SizeAtCompileTime, MaxSizeAtCompileTime, _StorageIndex> >
|
||||
{
|
||||
typedef PermutationBase<PermutationMatrix> Base;
|
||||
typedef internal::traits<PermutationMatrix> Traits;
|
||||
public:
|
||||
|
||||
typedef const PermutationMatrix& Nested;
|
||||
|
||||
#ifndef EIGEN_PARSED_BY_DOXYGEN
|
||||
typedef typename Traits::IndicesType IndicesType;
|
||||
typedef typename Traits::StorageIndex StorageIndex;
|
||||
#endif
|
||||
|
||||
inline PermutationMatrix()
|
||||
{}
|
||||
|
||||
/** Constructs an uninitialized permutation matrix of given size.
|
||||
*/
|
||||
explicit inline PermutationMatrix(Index size) : m_indices(size)
|
||||
{
|
||||
eigen_internal_assert(size <= NumTraits<StorageIndex>::highest());
|
||||
}
|
||||
|
||||
/** Copy constructor. */
|
||||
template<typename OtherDerived>
|
||||
inline PermutationMatrix(const PermutationBase<OtherDerived>& other)
|
||||
: m_indices(other.indices()) {}
|
||||
|
||||
/** Generic constructor from expression of the indices. The indices
|
||||
* array has the meaning that the permutations sends each integer i to indices[i].
|
||||
*
|
||||
* \warning It is your responsibility to check that the indices array that you passes actually
|
||||
* describes a permutation, i.e., each value between 0 and n-1 occurs exactly once, where n is the
|
||||
* array's size.
|
||||
*/
|
||||
template<typename Other>
|
||||
explicit inline PermutationMatrix(const MatrixBase<Other>& indices) : m_indices(indices)
|
||||
{}
|
||||
|
||||
/** Convert the Transpositions \a tr to a permutation matrix */
|
||||
template<typename Other>
|
||||
explicit PermutationMatrix(const TranspositionsBase<Other>& tr)
|
||||
: m_indices(tr.size())
|
||||
{
|
||||
*this = tr;
|
||||
}
|
||||
|
||||
/** Copies the other permutation into *this */
|
||||
template<typename Other>
|
||||
PermutationMatrix& operator=(const PermutationBase<Other>& other)
|
||||
{
|
||||
m_indices = other.indices();
|
||||
return *this;
|
||||
}
|
||||
|
||||
/** Assignment from the Transpositions \a tr */
|
||||
template<typename Other>
|
||||
PermutationMatrix& operator=(const TranspositionsBase<Other>& tr)
|
||||
{
|
||||
return Base::operator=(tr.derived());
|
||||
}
|
||||
|
||||
/** const version of indices(). */
|
||||
const IndicesType& indices() const { return m_indices; }
|
||||
/** \returns a reference to the stored array representing the permutation. */
|
||||
IndicesType& indices() { return m_indices; }
|
||||
|
||||
|
||||
/**** multiplication helpers to hopefully get RVO ****/
|
||||
|
||||
#ifndef EIGEN_PARSED_BY_DOXYGEN
|
||||
template<typename Other>
|
||||
PermutationMatrix(const InverseImpl<Other,PermutationStorage>& other)
|
||||
: m_indices(other.derived().nestedExpression().size())
|
||||
{
|
||||
eigen_internal_assert(m_indices.size() <= NumTraits<StorageIndex>::highest());
|
||||
StorageIndex end = StorageIndex(m_indices.size());
|
||||
for (StorageIndex i=0; i<end;++i)
|
||||
m_indices.coeffRef(other.derived().nestedExpression().indices().coeff(i)) = i;
|
||||
}
|
||||
template<typename Lhs,typename Rhs>
|
||||
PermutationMatrix(internal::PermPermProduct_t, const Lhs& lhs, const Rhs& rhs)
|
||||
: m_indices(lhs.indices().size())
|
||||
{
|
||||
Base::assignProduct(lhs,rhs);
|
||||
}
|
||||
#endif
|
||||
|
||||
protected:
|
||||
|
||||
IndicesType m_indices;
|
||||
};
|
||||
|
||||
|
||||
namespace internal {
|
||||
template<int SizeAtCompileTime, int MaxSizeAtCompileTime, typename _StorageIndex, int _PacketAccess>
|
||||
struct traits<Map<PermutationMatrix<SizeAtCompileTime, MaxSizeAtCompileTime, _StorageIndex>,_PacketAccess> >
|
||||
: traits<Matrix<_StorageIndex,SizeAtCompileTime,SizeAtCompileTime,0,MaxSizeAtCompileTime,MaxSizeAtCompileTime> >
|
||||
{
|
||||
typedef PermutationStorage StorageKind;
|
||||
typedef Map<const Matrix<_StorageIndex, SizeAtCompileTime, 1, 0, MaxSizeAtCompileTime, 1>, _PacketAccess> IndicesType;
|
||||
typedef _StorageIndex StorageIndex;
|
||||
typedef void Scalar;
|
||||
};
|
||||
}
|
||||
|
||||
template<int SizeAtCompileTime, int MaxSizeAtCompileTime, typename _StorageIndex, int _PacketAccess>
|
||||
class Map<PermutationMatrix<SizeAtCompileTime, MaxSizeAtCompileTime, _StorageIndex>,_PacketAccess>
|
||||
: public PermutationBase<Map<PermutationMatrix<SizeAtCompileTime, MaxSizeAtCompileTime, _StorageIndex>,_PacketAccess> >
|
||||
{
|
||||
typedef PermutationBase<Map> Base;
|
||||
typedef internal::traits<Map> Traits;
|
||||
public:
|
||||
|
||||
#ifndef EIGEN_PARSED_BY_DOXYGEN
|
||||
typedef typename Traits::IndicesType IndicesType;
|
||||
typedef typename IndicesType::Scalar StorageIndex;
|
||||
#endif
|
||||
|
||||
inline Map(const StorageIndex* indicesPtr)
|
||||
: m_indices(indicesPtr)
|
||||
{}
|
||||
|
||||
inline Map(const StorageIndex* indicesPtr, Index size)
|
||||
: m_indices(indicesPtr,size)
|
||||
{}
|
||||
|
||||
/** Copies the other permutation into *this */
|
||||
template<typename Other>
|
||||
Map& operator=(const PermutationBase<Other>& other)
|
||||
{ return Base::operator=(other.derived()); }
|
||||
|
||||
/** Assignment from the Transpositions \a tr */
|
||||
template<typename Other>
|
||||
Map& operator=(const TranspositionsBase<Other>& tr)
|
||||
{ return Base::operator=(tr.derived()); }
|
||||
|
||||
#ifndef EIGEN_PARSED_BY_DOXYGEN
|
||||
/** This is a special case of the templated operator=. Its purpose is to
|
||||
* prevent a default operator= from hiding the templated operator=.
|
||||
*/
|
||||
Map& operator=(const Map& other)
|
||||
{
|
||||
m_indices = other.m_indices;
|
||||
return *this;
|
||||
}
|
||||
#endif
|
||||
|
||||
/** const version of indices(). */
|
||||
const IndicesType& indices() const { return m_indices; }
|
||||
/** \returns a reference to the stored array representing the permutation. */
|
||||
IndicesType& indices() { return m_indices; }
|
||||
|
||||
protected:
|
||||
|
||||
IndicesType m_indices;
|
||||
};
|
||||
|
||||
template<typename _IndicesType> class TranspositionsWrapper;
|
||||
namespace internal {
|
||||
template<typename _IndicesType>
|
||||
struct traits<PermutationWrapper<_IndicesType> >
|
||||
{
|
||||
typedef PermutationStorage StorageKind;
|
||||
typedef void Scalar;
|
||||
typedef typename _IndicesType::Scalar StorageIndex;
|
||||
typedef _IndicesType IndicesType;
|
||||
enum {
|
||||
RowsAtCompileTime = _IndicesType::SizeAtCompileTime,
|
||||
ColsAtCompileTime = _IndicesType::SizeAtCompileTime,
|
||||
MaxRowsAtCompileTime = IndicesType::MaxSizeAtCompileTime,
|
||||
MaxColsAtCompileTime = IndicesType::MaxSizeAtCompileTime,
|
||||
Flags = 0
|
||||
};
|
||||
};
|
||||
}
|
||||
|
||||
/** \class PermutationWrapper
|
||||
* \ingroup Core_Module
|
||||
*
|
||||
* \brief Class to view a vector of integers as a permutation matrix
|
||||
*
|
||||
* \tparam _IndicesType the type of the vector of integer (can be any compatible expression)
|
||||
*
|
||||
* This class allows to view any vector expression of integers as a permutation matrix.
|
||||
*
|
||||
* \sa class PermutationBase, class PermutationMatrix
|
||||
*/
|
||||
template<typename _IndicesType>
|
||||
class PermutationWrapper : public PermutationBase<PermutationWrapper<_IndicesType> >
|
||||
{
|
||||
typedef PermutationBase<PermutationWrapper> Base;
|
||||
typedef internal::traits<PermutationWrapper> Traits;
|
||||
public:
|
||||
|
||||
#ifndef EIGEN_PARSED_BY_DOXYGEN
|
||||
typedef typename Traits::IndicesType IndicesType;
|
||||
#endif
|
||||
|
||||
inline PermutationWrapper(const IndicesType& indices)
|
||||
: m_indices(indices)
|
||||
{}
|
||||
|
||||
/** const version of indices(). */
|
||||
const typename internal::remove_all<typename IndicesType::Nested>::type&
|
||||
indices() const { return m_indices; }
|
||||
|
||||
protected:
|
||||
|
||||
typename IndicesType::Nested m_indices;
|
||||
};
|
||||
|
||||
|
||||
/** \returns the matrix with the permutation applied to the columns.
|
||||
*/
|
||||
template<typename MatrixDerived, typename PermutationDerived>
|
||||
EIGEN_DEVICE_FUNC
|
||||
const Product<MatrixDerived, PermutationDerived, AliasFreeProduct>
|
||||
operator*(const MatrixBase<MatrixDerived> &matrix,
|
||||
const PermutationBase<PermutationDerived>& permutation)
|
||||
{
|
||||
return Product<MatrixDerived, PermutationDerived, AliasFreeProduct>
|
||||
(matrix.derived(), permutation.derived());
|
||||
}
|
||||
|
||||
/** \returns the matrix with the permutation applied to the rows.
|
||||
*/
|
||||
template<typename PermutationDerived, typename MatrixDerived>
|
||||
EIGEN_DEVICE_FUNC
|
||||
const Product<PermutationDerived, MatrixDerived, AliasFreeProduct>
|
||||
operator*(const PermutationBase<PermutationDerived> &permutation,
|
||||
const MatrixBase<MatrixDerived>& matrix)
|
||||
{
|
||||
return Product<PermutationDerived, MatrixDerived, AliasFreeProduct>
|
||||
(permutation.derived(), matrix.derived());
|
||||
}
|
||||
|
||||
|
||||
template<typename PermutationType>
|
||||
class InverseImpl<PermutationType, PermutationStorage>
|
||||
: public EigenBase<Inverse<PermutationType> >
|
||||
{
|
||||
typedef typename PermutationType::PlainPermutationType PlainPermutationType;
|
||||
typedef internal::traits<PermutationType> PermTraits;
|
||||
protected:
|
||||
InverseImpl() {}
|
||||
public:
|
||||
typedef Inverse<PermutationType> InverseType;
|
||||
using EigenBase<Inverse<PermutationType> >::derived;
|
||||
|
||||
#ifndef EIGEN_PARSED_BY_DOXYGEN
|
||||
typedef typename PermutationType::DenseMatrixType DenseMatrixType;
|
||||
enum {
|
||||
RowsAtCompileTime = PermTraits::RowsAtCompileTime,
|
||||
ColsAtCompileTime = PermTraits::ColsAtCompileTime,
|
||||
MaxRowsAtCompileTime = PermTraits::MaxRowsAtCompileTime,
|
||||
MaxColsAtCompileTime = PermTraits::MaxColsAtCompileTime
|
||||
};
|
||||
#endif
|
||||
|
||||
#ifndef EIGEN_PARSED_BY_DOXYGEN
|
||||
template<typename DenseDerived>
|
||||
void evalTo(MatrixBase<DenseDerived>& other) const
|
||||
{
|
||||
other.setZero();
|
||||
for (Index i=0; i<derived().rows();++i)
|
||||
other.coeffRef(i, derived().nestedExpression().indices().coeff(i)) = typename DenseDerived::Scalar(1);
|
||||
}
|
||||
#endif
|
||||
|
||||
/** \return the equivalent permutation matrix */
|
||||
PlainPermutationType eval() const { return derived(); }
|
||||
|
||||
DenseMatrixType toDenseMatrix() const { return derived(); }
|
||||
|
||||
/** \returns the matrix with the inverse permutation applied to the columns.
|
||||
*/
|
||||
template<typename OtherDerived> friend
|
||||
const Product<OtherDerived, InverseType, AliasFreeProduct>
|
||||
operator*(const MatrixBase<OtherDerived>& matrix, const InverseType& trPerm)
|
||||
{
|
||||
return Product<OtherDerived, InverseType, AliasFreeProduct>(matrix.derived(), trPerm.derived());
|
||||
}
|
||||
|
||||
/** \returns the matrix with the inverse permutation applied to the rows.
|
||||
*/
|
||||
template<typename OtherDerived>
|
||||
const Product<InverseType, OtherDerived, AliasFreeProduct>
|
||||
operator*(const MatrixBase<OtherDerived>& matrix) const
|
||||
{
|
||||
return Product<InverseType, OtherDerived, AliasFreeProduct>(derived(), matrix.derived());
|
||||
}
|
||||
};
|
||||
|
||||
template<typename Derived>
|
||||
const PermutationWrapper<const Derived> MatrixBase<Derived>::asPermutation() const
|
||||
{
|
||||
return derived();
|
||||
}
|
||||
|
||||
namespace internal {
|
||||
|
||||
template<> struct AssignmentKind<DenseShape,PermutationShape> { typedef EigenBase2EigenBase Kind; };
|
||||
|
||||
} // end namespace internal
|
||||
|
||||
} // end namespace Eigen
|
||||
|
||||
#endif // EIGEN_PERMUTATIONMATRIX_H
|
||||
File diff suppressed because it is too large
Load Diff
@@ -0,0 +1,191 @@
|
||||
// This file is part of Eigen, a lightweight C++ template library
|
||||
// for linear algebra.
|
||||
//
|
||||
// Copyright (C) 2008-2011 Gael Guennebaud <gael.guennebaud@inria.fr>
|
||||
//
|
||||
// This Source Code Form is subject to the terms of the Mozilla
|
||||
// Public License v. 2.0. If a copy of the MPL was not distributed
|
||||
// with this file, You can obtain one at http://mozilla.org/MPL/2.0/.
|
||||
|
||||
#ifndef EIGEN_PRODUCT_H
|
||||
#define EIGEN_PRODUCT_H
|
||||
|
||||
namespace Eigen {
|
||||
|
||||
template<typename Lhs, typename Rhs, int Option, typename StorageKind> class ProductImpl;
|
||||
|
||||
namespace internal {
|
||||
|
||||
template<typename Lhs, typename Rhs, int Option>
|
||||
struct traits<Product<Lhs, Rhs, Option> >
|
||||
{
|
||||
typedef typename remove_all<Lhs>::type LhsCleaned;
|
||||
typedef typename remove_all<Rhs>::type RhsCleaned;
|
||||
typedef traits<LhsCleaned> LhsTraits;
|
||||
typedef traits<RhsCleaned> RhsTraits;
|
||||
|
||||
typedef MatrixXpr XprKind;
|
||||
|
||||
typedef typename ScalarBinaryOpTraits<typename traits<LhsCleaned>::Scalar, typename traits<RhsCleaned>::Scalar>::ReturnType Scalar;
|
||||
typedef typename product_promote_storage_type<typename LhsTraits::StorageKind,
|
||||
typename RhsTraits::StorageKind,
|
||||
internal::product_type<Lhs,Rhs>::ret>::ret StorageKind;
|
||||
typedef typename promote_index_type<typename LhsTraits::StorageIndex,
|
||||
typename RhsTraits::StorageIndex>::type StorageIndex;
|
||||
|
||||
enum {
|
||||
RowsAtCompileTime = LhsTraits::RowsAtCompileTime,
|
||||
ColsAtCompileTime = RhsTraits::ColsAtCompileTime,
|
||||
MaxRowsAtCompileTime = LhsTraits::MaxRowsAtCompileTime,
|
||||
MaxColsAtCompileTime = RhsTraits::MaxColsAtCompileTime,
|
||||
|
||||
// FIXME: only needed by GeneralMatrixMatrixTriangular
|
||||
InnerSize = EIGEN_SIZE_MIN_PREFER_FIXED(LhsTraits::ColsAtCompileTime, RhsTraits::RowsAtCompileTime),
|
||||
|
||||
// The storage order is somewhat arbitrary here. The correct one will be determined through the evaluator.
|
||||
Flags = (MaxRowsAtCompileTime==1 && MaxColsAtCompileTime!=1) ? RowMajorBit
|
||||
: (MaxColsAtCompileTime==1 && MaxRowsAtCompileTime!=1) ? 0
|
||||
: ( ((LhsTraits::Flags&NoPreferredStorageOrderBit) && (RhsTraits::Flags&RowMajorBit))
|
||||
|| ((RhsTraits::Flags&NoPreferredStorageOrderBit) && (LhsTraits::Flags&RowMajorBit)) ) ? RowMajorBit
|
||||
: NoPreferredStorageOrderBit
|
||||
};
|
||||
};
|
||||
|
||||
} // end namespace internal
|
||||
|
||||
/** \class Product
|
||||
* \ingroup Core_Module
|
||||
*
|
||||
* \brief Expression of the product of two arbitrary matrices or vectors
|
||||
*
|
||||
* \tparam _Lhs the type of the left-hand side expression
|
||||
* \tparam _Rhs the type of the right-hand side expression
|
||||
*
|
||||
* This class represents an expression of the product of two arbitrary matrices.
|
||||
*
|
||||
* The other template parameters are:
|
||||
* \tparam Option can be DefaultProduct, AliasFreeProduct, or LazyProduct
|
||||
*
|
||||
*/
|
||||
template<typename _Lhs, typename _Rhs, int Option>
|
||||
class Product : public ProductImpl<_Lhs,_Rhs,Option,
|
||||
typename internal::product_promote_storage_type<typename internal::traits<_Lhs>::StorageKind,
|
||||
typename internal::traits<_Rhs>::StorageKind,
|
||||
internal::product_type<_Lhs,_Rhs>::ret>::ret>
|
||||
{
|
||||
public:
|
||||
|
||||
typedef _Lhs Lhs;
|
||||
typedef _Rhs Rhs;
|
||||
|
||||
typedef typename ProductImpl<
|
||||
Lhs, Rhs, Option,
|
||||
typename internal::product_promote_storage_type<typename internal::traits<Lhs>::StorageKind,
|
||||
typename internal::traits<Rhs>::StorageKind,
|
||||
internal::product_type<Lhs,Rhs>::ret>::ret>::Base Base;
|
||||
EIGEN_GENERIC_PUBLIC_INTERFACE(Product)
|
||||
|
||||
typedef typename internal::ref_selector<Lhs>::type LhsNested;
|
||||
typedef typename internal::ref_selector<Rhs>::type RhsNested;
|
||||
typedef typename internal::remove_all<LhsNested>::type LhsNestedCleaned;
|
||||
typedef typename internal::remove_all<RhsNested>::type RhsNestedCleaned;
|
||||
|
||||
EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
|
||||
Product(const Lhs& lhs, const Rhs& rhs) : m_lhs(lhs), m_rhs(rhs)
|
||||
{
|
||||
eigen_assert(lhs.cols() == rhs.rows()
|
||||
&& "invalid matrix product"
|
||||
&& "if you wanted a coeff-wise or a dot product use the respective explicit functions");
|
||||
}
|
||||
|
||||
EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE EIGEN_CONSTEXPR
|
||||
Index rows() const EIGEN_NOEXCEPT { return m_lhs.rows(); }
|
||||
EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE EIGEN_CONSTEXPR
|
||||
Index cols() const EIGEN_NOEXCEPT { return m_rhs.cols(); }
|
||||
|
||||
EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
|
||||
const LhsNestedCleaned& lhs() const { return m_lhs; }
|
||||
EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
|
||||
const RhsNestedCleaned& rhs() const { return m_rhs; }
|
||||
|
||||
protected:
|
||||
|
||||
LhsNested m_lhs;
|
||||
RhsNested m_rhs;
|
||||
};
|
||||
|
||||
namespace internal {
|
||||
|
||||
template<typename Lhs, typename Rhs, int Option, int ProductTag = internal::product_type<Lhs,Rhs>::ret>
|
||||
class dense_product_base
|
||||
: public internal::dense_xpr_base<Product<Lhs,Rhs,Option> >::type
|
||||
{};
|
||||
|
||||
/** Conversion to scalar for inner-products */
|
||||
template<typename Lhs, typename Rhs, int Option>
|
||||
class dense_product_base<Lhs, Rhs, Option, InnerProduct>
|
||||
: public internal::dense_xpr_base<Product<Lhs,Rhs,Option> >::type
|
||||
{
|
||||
typedef Product<Lhs,Rhs,Option> ProductXpr;
|
||||
typedef typename internal::dense_xpr_base<ProductXpr>::type Base;
|
||||
public:
|
||||
using Base::derived;
|
||||
typedef typename Base::Scalar Scalar;
|
||||
|
||||
EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE operator const Scalar() const
|
||||
{
|
||||
return internal::evaluator<ProductXpr>(derived()).coeff(0,0);
|
||||
}
|
||||
};
|
||||
|
||||
} // namespace internal
|
||||
|
||||
// Generic API dispatcher
|
||||
template<typename Lhs, typename Rhs, int Option, typename StorageKind>
|
||||
class ProductImpl : public internal::generic_xpr_base<Product<Lhs,Rhs,Option>, MatrixXpr, StorageKind>::type
|
||||
{
|
||||
public:
|
||||
typedef typename internal::generic_xpr_base<Product<Lhs,Rhs,Option>, MatrixXpr, StorageKind>::type Base;
|
||||
};
|
||||
|
||||
template<typename Lhs, typename Rhs, int Option>
|
||||
class ProductImpl<Lhs,Rhs,Option,Dense>
|
||||
: public internal::dense_product_base<Lhs,Rhs,Option>
|
||||
{
|
||||
typedef Product<Lhs, Rhs, Option> Derived;
|
||||
|
||||
public:
|
||||
|
||||
typedef typename internal::dense_product_base<Lhs, Rhs, Option> Base;
|
||||
EIGEN_DENSE_PUBLIC_INTERFACE(Derived)
|
||||
protected:
|
||||
enum {
|
||||
IsOneByOne = (RowsAtCompileTime == 1 || RowsAtCompileTime == Dynamic) &&
|
||||
(ColsAtCompileTime == 1 || ColsAtCompileTime == Dynamic),
|
||||
EnableCoeff = IsOneByOne || Option==LazyProduct
|
||||
};
|
||||
|
||||
public:
|
||||
|
||||
EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE Scalar coeff(Index row, Index col) const
|
||||
{
|
||||
EIGEN_STATIC_ASSERT(EnableCoeff, THIS_METHOD_IS_ONLY_FOR_INNER_OR_LAZY_PRODUCTS);
|
||||
eigen_assert( (Option==LazyProduct) || (this->rows() == 1 && this->cols() == 1) );
|
||||
|
||||
return internal::evaluator<Derived>(derived()).coeff(row,col);
|
||||
}
|
||||
|
||||
EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE Scalar coeff(Index i) const
|
||||
{
|
||||
EIGEN_STATIC_ASSERT(EnableCoeff, THIS_METHOD_IS_ONLY_FOR_INNER_OR_LAZY_PRODUCTS);
|
||||
eigen_assert( (Option==LazyProduct) || (this->rows() == 1 && this->cols() == 1) );
|
||||
|
||||
return internal::evaluator<Derived>(derived()).coeff(i);
|
||||
}
|
||||
|
||||
|
||||
};
|
||||
|
||||
} // end namespace Eigen
|
||||
|
||||
#endif // EIGEN_PRODUCT_H
|
||||
File diff suppressed because it is too large
Load Diff
@@ -0,0 +1,218 @@
|
||||
// This file is part of Eigen, a lightweight C++ template library
|
||||
// for linear algebra.
|
||||
//
|
||||
// Copyright (C) 2008 Gael Guennebaud <gael.guennebaud@inria.fr>
|
||||
//
|
||||
// This Source Code Form is subject to the terms of the Mozilla
|
||||
// Public License v. 2.0. If a copy of the MPL was not distributed
|
||||
// with this file, You can obtain one at http://mozilla.org/MPL/2.0/.
|
||||
|
||||
#ifndef EIGEN_RANDOM_H
|
||||
#define EIGEN_RANDOM_H
|
||||
|
||||
namespace Eigen {
|
||||
|
||||
namespace internal {
|
||||
|
||||
template<typename Scalar> struct scalar_random_op {
|
||||
EIGEN_EMPTY_STRUCT_CTOR(scalar_random_op)
|
||||
inline const Scalar operator() () const { return random<Scalar>(); }
|
||||
};
|
||||
|
||||
template<typename Scalar>
|
||||
struct functor_traits<scalar_random_op<Scalar> >
|
||||
{ enum { Cost = 5 * NumTraits<Scalar>::MulCost, PacketAccess = false, IsRepeatable = false }; };
|
||||
|
||||
} // end namespace internal
|
||||
|
||||
/** \returns a random matrix expression
|
||||
*
|
||||
* Numbers are uniformly spread through their whole definition range for integer types,
|
||||
* and in the [-1:1] range for floating point scalar types.
|
||||
*
|
||||
* The parameters \a rows and \a cols are the number of rows and of columns of
|
||||
* the returned matrix. Must be compatible with this MatrixBase type.
|
||||
*
|
||||
* \not_reentrant
|
||||
*
|
||||
* This variant is meant to be used for dynamic-size matrix types. For fixed-size types,
|
||||
* it is redundant to pass \a rows and \a cols as arguments, so Random() should be used
|
||||
* instead.
|
||||
*
|
||||
*
|
||||
* Example: \include MatrixBase_random_int_int.cpp
|
||||
* Output: \verbinclude MatrixBase_random_int_int.out
|
||||
*
|
||||
* This expression has the "evaluate before nesting" flag so that it will be evaluated into
|
||||
* a temporary matrix whenever it is nested in a larger expression. This prevents unexpected
|
||||
* behavior with expressions involving random matrices.
|
||||
*
|
||||
* See DenseBase::NullaryExpr(Index, const CustomNullaryOp&) for an example using C++11 random generators.
|
||||
*
|
||||
* \sa DenseBase::setRandom(), DenseBase::Random(Index), DenseBase::Random()
|
||||
*/
|
||||
template<typename Derived>
|
||||
inline const typename DenseBase<Derived>::RandomReturnType
|
||||
DenseBase<Derived>::Random(Index rows, Index cols)
|
||||
{
|
||||
return NullaryExpr(rows, cols, internal::scalar_random_op<Scalar>());
|
||||
}
|
||||
|
||||
/** \returns a random vector expression
|
||||
*
|
||||
* Numbers are uniformly spread through their whole definition range for integer types,
|
||||
* and in the [-1:1] range for floating point scalar types.
|
||||
*
|
||||
* The parameter \a size is the size of the returned vector.
|
||||
* Must be compatible with this MatrixBase type.
|
||||
*
|
||||
* \only_for_vectors
|
||||
* \not_reentrant
|
||||
*
|
||||
* This variant is meant to be used for dynamic-size vector types. For fixed-size types,
|
||||
* it is redundant to pass \a size as argument, so Random() should be used
|
||||
* instead.
|
||||
*
|
||||
* Example: \include MatrixBase_random_int.cpp
|
||||
* Output: \verbinclude MatrixBase_random_int.out
|
||||
*
|
||||
* This expression has the "evaluate before nesting" flag so that it will be evaluated into
|
||||
* a temporary vector whenever it is nested in a larger expression. This prevents unexpected
|
||||
* behavior with expressions involving random matrices.
|
||||
*
|
||||
* \sa DenseBase::setRandom(), DenseBase::Random(Index,Index), DenseBase::Random()
|
||||
*/
|
||||
template<typename Derived>
|
||||
inline const typename DenseBase<Derived>::RandomReturnType
|
||||
DenseBase<Derived>::Random(Index size)
|
||||
{
|
||||
return NullaryExpr(size, internal::scalar_random_op<Scalar>());
|
||||
}
|
||||
|
||||
/** \returns a fixed-size random matrix or vector expression
|
||||
*
|
||||
* Numbers are uniformly spread through their whole definition range for integer types,
|
||||
* and in the [-1:1] range for floating point scalar types.
|
||||
*
|
||||
* This variant is only for fixed-size MatrixBase types. For dynamic-size types, you
|
||||
* need to use the variants taking size arguments.
|
||||
*
|
||||
* Example: \include MatrixBase_random.cpp
|
||||
* Output: \verbinclude MatrixBase_random.out
|
||||
*
|
||||
* This expression has the "evaluate before nesting" flag so that it will be evaluated into
|
||||
* a temporary matrix whenever it is nested in a larger expression. This prevents unexpected
|
||||
* behavior with expressions involving random matrices.
|
||||
*
|
||||
* \not_reentrant
|
||||
*
|
||||
* \sa DenseBase::setRandom(), DenseBase::Random(Index,Index), DenseBase::Random(Index)
|
||||
*/
|
||||
template<typename Derived>
|
||||
inline const typename DenseBase<Derived>::RandomReturnType
|
||||
DenseBase<Derived>::Random()
|
||||
{
|
||||
return NullaryExpr(RowsAtCompileTime, ColsAtCompileTime, internal::scalar_random_op<Scalar>());
|
||||
}
|
||||
|
||||
/** Sets all coefficients in this expression to random values.
|
||||
*
|
||||
* Numbers are uniformly spread through their whole definition range for integer types,
|
||||
* and in the [-1:1] range for floating point scalar types.
|
||||
*
|
||||
* \not_reentrant
|
||||
*
|
||||
* Example: \include MatrixBase_setRandom.cpp
|
||||
* Output: \verbinclude MatrixBase_setRandom.out
|
||||
*
|
||||
* \sa class CwiseNullaryOp, setRandom(Index), setRandom(Index,Index)
|
||||
*/
|
||||
template<typename Derived>
|
||||
EIGEN_DEVICE_FUNC inline Derived& DenseBase<Derived>::setRandom()
|
||||
{
|
||||
return *this = Random(rows(), cols());
|
||||
}
|
||||
|
||||
/** Resizes to the given \a newSize, and sets all coefficients in this expression to random values.
|
||||
*
|
||||
* Numbers are uniformly spread through their whole definition range for integer types,
|
||||
* and in the [-1:1] range for floating point scalar types.
|
||||
*
|
||||
* \only_for_vectors
|
||||
* \not_reentrant
|
||||
*
|
||||
* Example: \include Matrix_setRandom_int.cpp
|
||||
* Output: \verbinclude Matrix_setRandom_int.out
|
||||
*
|
||||
* \sa DenseBase::setRandom(), setRandom(Index,Index), class CwiseNullaryOp, DenseBase::Random()
|
||||
*/
|
||||
template<typename Derived>
|
||||
EIGEN_STRONG_INLINE Derived&
|
||||
PlainObjectBase<Derived>::setRandom(Index newSize)
|
||||
{
|
||||
resize(newSize);
|
||||
return setRandom();
|
||||
}
|
||||
|
||||
/** Resizes to the given size, and sets all coefficients in this expression to random values.
|
||||
*
|
||||
* Numbers are uniformly spread through their whole definition range for integer types,
|
||||
* and in the [-1:1] range for floating point scalar types.
|
||||
*
|
||||
* \not_reentrant
|
||||
*
|
||||
* \param rows the new number of rows
|
||||
* \param cols the new number of columns
|
||||
*
|
||||
* Example: \include Matrix_setRandom_int_int.cpp
|
||||
* Output: \verbinclude Matrix_setRandom_int_int.out
|
||||
*
|
||||
* \sa DenseBase::setRandom(), setRandom(Index), class CwiseNullaryOp, DenseBase::Random()
|
||||
*/
|
||||
template<typename Derived>
|
||||
EIGEN_STRONG_INLINE Derived&
|
||||
PlainObjectBase<Derived>::setRandom(Index rows, Index cols)
|
||||
{
|
||||
resize(rows, cols);
|
||||
return setRandom();
|
||||
}
|
||||
|
||||
/** Resizes to the given size, changing only the number of columns, and sets all
|
||||
* coefficients in this expression to random values. For the parameter of type
|
||||
* NoChange_t, just pass the special value \c NoChange.
|
||||
*
|
||||
* Numbers are uniformly spread through their whole definition range for integer types,
|
||||
* and in the [-1:1] range for floating point scalar types.
|
||||
*
|
||||
* \not_reentrant
|
||||
*
|
||||
* \sa DenseBase::setRandom(), setRandom(Index), setRandom(Index, NoChange_t), class CwiseNullaryOp, DenseBase::Random()
|
||||
*/
|
||||
template<typename Derived>
|
||||
EIGEN_STRONG_INLINE Derived&
|
||||
PlainObjectBase<Derived>::setRandom(NoChange_t, Index cols)
|
||||
{
|
||||
return setRandom(rows(), cols);
|
||||
}
|
||||
|
||||
/** Resizes to the given size, changing only the number of rows, and sets all
|
||||
* coefficients in this expression to random values. For the parameter of type
|
||||
* NoChange_t, just pass the special value \c NoChange.
|
||||
*
|
||||
* Numbers are uniformly spread through their whole definition range for integer types,
|
||||
* and in the [-1:1] range for floating point scalar types.
|
||||
*
|
||||
* \not_reentrant
|
||||
*
|
||||
* \sa DenseBase::setRandom(), setRandom(Index), setRandom(NoChange_t, Index), class CwiseNullaryOp, DenseBase::Random()
|
||||
*/
|
||||
template<typename Derived>
|
||||
EIGEN_STRONG_INLINE Derived&
|
||||
PlainObjectBase<Derived>::setRandom(Index rows, NoChange_t)
|
||||
{
|
||||
return setRandom(rows, cols());
|
||||
}
|
||||
|
||||
} // end namespace Eigen
|
||||
|
||||
#endif // EIGEN_RANDOM_H
|
||||
@@ -0,0 +1,515 @@
|
||||
// This file is part of Eigen, a lightweight C++ template library
|
||||
// for linear algebra.
|
||||
//
|
||||
// Copyright (C) 2008 Gael Guennebaud <gael.guennebaud@inria.fr>
|
||||
// Copyright (C) 2006-2008 Benoit Jacob <jacob.benoit.1@gmail.com>
|
||||
//
|
||||
// This Source Code Form is subject to the terms of the Mozilla
|
||||
// Public License v. 2.0. If a copy of the MPL was not distributed
|
||||
// with this file, You can obtain one at http://mozilla.org/MPL/2.0/.
|
||||
|
||||
#ifndef EIGEN_REDUX_H
|
||||
#define EIGEN_REDUX_H
|
||||
|
||||
namespace Eigen {
|
||||
|
||||
namespace internal {
|
||||
|
||||
// TODO
|
||||
// * implement other kind of vectorization
|
||||
// * factorize code
|
||||
|
||||
/***************************************************************************
|
||||
* Part 1 : the logic deciding a strategy for vectorization and unrolling
|
||||
***************************************************************************/
|
||||
|
||||
template<typename Func, typename Evaluator>
|
||||
struct redux_traits
|
||||
{
|
||||
public:
|
||||
typedef typename find_best_packet<typename Evaluator::Scalar,Evaluator::SizeAtCompileTime>::type PacketType;
|
||||
enum {
|
||||
PacketSize = unpacket_traits<PacketType>::size,
|
||||
InnerMaxSize = int(Evaluator::IsRowMajor)
|
||||
? Evaluator::MaxColsAtCompileTime
|
||||
: Evaluator::MaxRowsAtCompileTime,
|
||||
OuterMaxSize = int(Evaluator::IsRowMajor)
|
||||
? Evaluator::MaxRowsAtCompileTime
|
||||
: Evaluator::MaxColsAtCompileTime,
|
||||
SliceVectorizedWork = int(InnerMaxSize)==Dynamic ? Dynamic
|
||||
: int(OuterMaxSize)==Dynamic ? (int(InnerMaxSize)>=int(PacketSize) ? Dynamic : 0)
|
||||
: (int(InnerMaxSize)/int(PacketSize)) * int(OuterMaxSize)
|
||||
};
|
||||
|
||||
enum {
|
||||
MightVectorize = (int(Evaluator::Flags)&ActualPacketAccessBit)
|
||||
&& (functor_traits<Func>::PacketAccess),
|
||||
MayLinearVectorize = bool(MightVectorize) && (int(Evaluator::Flags)&LinearAccessBit),
|
||||
MaySliceVectorize = bool(MightVectorize) && (int(SliceVectorizedWork)==Dynamic || int(SliceVectorizedWork)>=3)
|
||||
};
|
||||
|
||||
public:
|
||||
enum {
|
||||
Traversal = int(MayLinearVectorize) ? int(LinearVectorizedTraversal)
|
||||
: int(MaySliceVectorize) ? int(SliceVectorizedTraversal)
|
||||
: int(DefaultTraversal)
|
||||
};
|
||||
|
||||
public:
|
||||
enum {
|
||||
Cost = Evaluator::SizeAtCompileTime == Dynamic ? HugeCost
|
||||
: int(Evaluator::SizeAtCompileTime) * int(Evaluator::CoeffReadCost) + (Evaluator::SizeAtCompileTime-1) * functor_traits<Func>::Cost,
|
||||
UnrollingLimit = EIGEN_UNROLLING_LIMIT * (int(Traversal) == int(DefaultTraversal) ? 1 : int(PacketSize))
|
||||
};
|
||||
|
||||
public:
|
||||
enum {
|
||||
Unrolling = Cost <= UnrollingLimit ? CompleteUnrolling : NoUnrolling
|
||||
};
|
||||
|
||||
#ifdef EIGEN_DEBUG_ASSIGN
|
||||
static void debug()
|
||||
{
|
||||
std::cerr << "Xpr: " << typeid(typename Evaluator::XprType).name() << std::endl;
|
||||
std::cerr.setf(std::ios::hex, std::ios::basefield);
|
||||
EIGEN_DEBUG_VAR(Evaluator::Flags)
|
||||
std::cerr.unsetf(std::ios::hex);
|
||||
EIGEN_DEBUG_VAR(InnerMaxSize)
|
||||
EIGEN_DEBUG_VAR(OuterMaxSize)
|
||||
EIGEN_DEBUG_VAR(SliceVectorizedWork)
|
||||
EIGEN_DEBUG_VAR(PacketSize)
|
||||
EIGEN_DEBUG_VAR(MightVectorize)
|
||||
EIGEN_DEBUG_VAR(MayLinearVectorize)
|
||||
EIGEN_DEBUG_VAR(MaySliceVectorize)
|
||||
std::cerr << "Traversal" << " = " << Traversal << " (" << demangle_traversal(Traversal) << ")" << std::endl;
|
||||
EIGEN_DEBUG_VAR(UnrollingLimit)
|
||||
std::cerr << "Unrolling" << " = " << Unrolling << " (" << demangle_unrolling(Unrolling) << ")" << std::endl;
|
||||
std::cerr << std::endl;
|
||||
}
|
||||
#endif
|
||||
};
|
||||
|
||||
/***************************************************************************
|
||||
* Part 2 : unrollers
|
||||
***************************************************************************/
|
||||
|
||||
/*** no vectorization ***/
|
||||
|
||||
template<typename Func, typename Evaluator, int Start, int Length>
|
||||
struct redux_novec_unroller
|
||||
{
|
||||
enum {
|
||||
HalfLength = Length/2
|
||||
};
|
||||
|
||||
typedef typename Evaluator::Scalar Scalar;
|
||||
|
||||
EIGEN_DEVICE_FUNC
|
||||
static EIGEN_STRONG_INLINE Scalar run(const Evaluator &eval, const Func& func)
|
||||
{
|
||||
return func(redux_novec_unroller<Func, Evaluator, Start, HalfLength>::run(eval,func),
|
||||
redux_novec_unroller<Func, Evaluator, Start+HalfLength, Length-HalfLength>::run(eval,func));
|
||||
}
|
||||
};
|
||||
|
||||
template<typename Func, typename Evaluator, int Start>
|
||||
struct redux_novec_unroller<Func, Evaluator, Start, 1>
|
||||
{
|
||||
enum {
|
||||
outer = Start / Evaluator::InnerSizeAtCompileTime,
|
||||
inner = Start % Evaluator::InnerSizeAtCompileTime
|
||||
};
|
||||
|
||||
typedef typename Evaluator::Scalar Scalar;
|
||||
|
||||
EIGEN_DEVICE_FUNC
|
||||
static EIGEN_STRONG_INLINE Scalar run(const Evaluator &eval, const Func&)
|
||||
{
|
||||
return eval.coeffByOuterInner(outer, inner);
|
||||
}
|
||||
};
|
||||
|
||||
// This is actually dead code and will never be called. It is required
|
||||
// to prevent false warnings regarding failed inlining though
|
||||
// for 0 length run() will never be called at all.
|
||||
template<typename Func, typename Evaluator, int Start>
|
||||
struct redux_novec_unroller<Func, Evaluator, Start, 0>
|
||||
{
|
||||
typedef typename Evaluator::Scalar Scalar;
|
||||
EIGEN_DEVICE_FUNC
|
||||
static EIGEN_STRONG_INLINE Scalar run(const Evaluator&, const Func&) { return Scalar(); }
|
||||
};
|
||||
|
||||
/*** vectorization ***/
|
||||
|
||||
template<typename Func, typename Evaluator, int Start, int Length>
|
||||
struct redux_vec_unroller
|
||||
{
|
||||
template<typename PacketType>
|
||||
EIGEN_DEVICE_FUNC
|
||||
static EIGEN_STRONG_INLINE PacketType run(const Evaluator &eval, const Func& func)
|
||||
{
|
||||
enum {
|
||||
PacketSize = unpacket_traits<PacketType>::size,
|
||||
HalfLength = Length/2
|
||||
};
|
||||
|
||||
return func.packetOp(
|
||||
redux_vec_unroller<Func, Evaluator, Start, HalfLength>::template run<PacketType>(eval,func),
|
||||
redux_vec_unroller<Func, Evaluator, Start+HalfLength, Length-HalfLength>::template run<PacketType>(eval,func) );
|
||||
}
|
||||
};
|
||||
|
||||
template<typename Func, typename Evaluator, int Start>
|
||||
struct redux_vec_unroller<Func, Evaluator, Start, 1>
|
||||
{
|
||||
template<typename PacketType>
|
||||
EIGEN_DEVICE_FUNC
|
||||
static EIGEN_STRONG_INLINE PacketType run(const Evaluator &eval, const Func&)
|
||||
{
|
||||
enum {
|
||||
PacketSize = unpacket_traits<PacketType>::size,
|
||||
index = Start * PacketSize,
|
||||
outer = index / int(Evaluator::InnerSizeAtCompileTime),
|
||||
inner = index % int(Evaluator::InnerSizeAtCompileTime),
|
||||
alignment = Evaluator::Alignment
|
||||
};
|
||||
return eval.template packetByOuterInner<alignment,PacketType>(outer, inner);
|
||||
}
|
||||
};
|
||||
|
||||
/***************************************************************************
|
||||
* Part 3 : implementation of all cases
|
||||
***************************************************************************/
|
||||
|
||||
template<typename Func, typename Evaluator,
|
||||
int Traversal = redux_traits<Func, Evaluator>::Traversal,
|
||||
int Unrolling = redux_traits<Func, Evaluator>::Unrolling
|
||||
>
|
||||
struct redux_impl;
|
||||
|
||||
template<typename Func, typename Evaluator>
|
||||
struct redux_impl<Func, Evaluator, DefaultTraversal, NoUnrolling>
|
||||
{
|
||||
typedef typename Evaluator::Scalar Scalar;
|
||||
|
||||
template<typename XprType>
|
||||
EIGEN_DEVICE_FUNC static EIGEN_STRONG_INLINE
|
||||
Scalar run(const Evaluator &eval, const Func& func, const XprType& xpr)
|
||||
{
|
||||
eigen_assert(xpr.rows()>0 && xpr.cols()>0 && "you are using an empty matrix");
|
||||
Scalar res;
|
||||
res = eval.coeffByOuterInner(0, 0);
|
||||
for(Index i = 1; i < xpr.innerSize(); ++i)
|
||||
res = func(res, eval.coeffByOuterInner(0, i));
|
||||
for(Index i = 1; i < xpr.outerSize(); ++i)
|
||||
for(Index j = 0; j < xpr.innerSize(); ++j)
|
||||
res = func(res, eval.coeffByOuterInner(i, j));
|
||||
return res;
|
||||
}
|
||||
};
|
||||
|
||||
template<typename Func, typename Evaluator>
|
||||
struct redux_impl<Func,Evaluator, DefaultTraversal, CompleteUnrolling>
|
||||
: redux_novec_unroller<Func,Evaluator, 0, Evaluator::SizeAtCompileTime>
|
||||
{
|
||||
typedef redux_novec_unroller<Func,Evaluator, 0, Evaluator::SizeAtCompileTime> Base;
|
||||
typedef typename Evaluator::Scalar Scalar;
|
||||
template<typename XprType>
|
||||
EIGEN_DEVICE_FUNC static EIGEN_STRONG_INLINE
|
||||
Scalar run(const Evaluator &eval, const Func& func, const XprType& /*xpr*/)
|
||||
{
|
||||
return Base::run(eval,func);
|
||||
}
|
||||
};
|
||||
|
||||
template<typename Func, typename Evaluator>
|
||||
struct redux_impl<Func, Evaluator, LinearVectorizedTraversal, NoUnrolling>
|
||||
{
|
||||
typedef typename Evaluator::Scalar Scalar;
|
||||
typedef typename redux_traits<Func, Evaluator>::PacketType PacketScalar;
|
||||
|
||||
template<typename XprType>
|
||||
static Scalar run(const Evaluator &eval, const Func& func, const XprType& xpr)
|
||||
{
|
||||
const Index size = xpr.size();
|
||||
|
||||
const Index packetSize = redux_traits<Func, Evaluator>::PacketSize;
|
||||
const int packetAlignment = unpacket_traits<PacketScalar>::alignment;
|
||||
enum {
|
||||
alignment0 = (bool(Evaluator::Flags & DirectAccessBit) && bool(packet_traits<Scalar>::AlignedOnScalar)) ? int(packetAlignment) : int(Unaligned),
|
||||
alignment = EIGEN_PLAIN_ENUM_MAX(alignment0, Evaluator::Alignment)
|
||||
};
|
||||
const Index alignedStart = internal::first_default_aligned(xpr);
|
||||
const Index alignedSize2 = ((size-alignedStart)/(2*packetSize))*(2*packetSize);
|
||||
const Index alignedSize = ((size-alignedStart)/(packetSize))*(packetSize);
|
||||
const Index alignedEnd2 = alignedStart + alignedSize2;
|
||||
const Index alignedEnd = alignedStart + alignedSize;
|
||||
Scalar res;
|
||||
if(alignedSize)
|
||||
{
|
||||
PacketScalar packet_res0 = eval.template packet<alignment,PacketScalar>(alignedStart);
|
||||
if(alignedSize>packetSize) // we have at least two packets to partly unroll the loop
|
||||
{
|
||||
PacketScalar packet_res1 = eval.template packet<alignment,PacketScalar>(alignedStart+packetSize);
|
||||
for(Index index = alignedStart + 2*packetSize; index < alignedEnd2; index += 2*packetSize)
|
||||
{
|
||||
packet_res0 = func.packetOp(packet_res0, eval.template packet<alignment,PacketScalar>(index));
|
||||
packet_res1 = func.packetOp(packet_res1, eval.template packet<alignment,PacketScalar>(index+packetSize));
|
||||
}
|
||||
|
||||
packet_res0 = func.packetOp(packet_res0,packet_res1);
|
||||
if(alignedEnd>alignedEnd2)
|
||||
packet_res0 = func.packetOp(packet_res0, eval.template packet<alignment,PacketScalar>(alignedEnd2));
|
||||
}
|
||||
res = func.predux(packet_res0);
|
||||
|
||||
for(Index index = 0; index < alignedStart; ++index)
|
||||
res = func(res,eval.coeff(index));
|
||||
|
||||
for(Index index = alignedEnd; index < size; ++index)
|
||||
res = func(res,eval.coeff(index));
|
||||
}
|
||||
else // too small to vectorize anything.
|
||||
// since this is dynamic-size hence inefficient anyway for such small sizes, don't try to optimize.
|
||||
{
|
||||
res = eval.coeff(0);
|
||||
for(Index index = 1; index < size; ++index)
|
||||
res = func(res,eval.coeff(index));
|
||||
}
|
||||
|
||||
return res;
|
||||
}
|
||||
};
|
||||
|
||||
// NOTE: for SliceVectorizedTraversal we simply bypass unrolling
|
||||
template<typename Func, typename Evaluator, int Unrolling>
|
||||
struct redux_impl<Func, Evaluator, SliceVectorizedTraversal, Unrolling>
|
||||
{
|
||||
typedef typename Evaluator::Scalar Scalar;
|
||||
typedef typename redux_traits<Func, Evaluator>::PacketType PacketType;
|
||||
|
||||
template<typename XprType>
|
||||
EIGEN_DEVICE_FUNC static Scalar run(const Evaluator &eval, const Func& func, const XprType& xpr)
|
||||
{
|
||||
eigen_assert(xpr.rows()>0 && xpr.cols()>0 && "you are using an empty matrix");
|
||||
const Index innerSize = xpr.innerSize();
|
||||
const Index outerSize = xpr.outerSize();
|
||||
enum {
|
||||
packetSize = redux_traits<Func, Evaluator>::PacketSize
|
||||
};
|
||||
const Index packetedInnerSize = ((innerSize)/packetSize)*packetSize;
|
||||
Scalar res;
|
||||
if(packetedInnerSize)
|
||||
{
|
||||
PacketType packet_res = eval.template packet<Unaligned,PacketType>(0,0);
|
||||
for(Index j=0; j<outerSize; ++j)
|
||||
for(Index i=(j==0?packetSize:0); i<packetedInnerSize; i+=Index(packetSize))
|
||||
packet_res = func.packetOp(packet_res, eval.template packetByOuterInner<Unaligned,PacketType>(j,i));
|
||||
|
||||
res = func.predux(packet_res);
|
||||
for(Index j=0; j<outerSize; ++j)
|
||||
for(Index i=packetedInnerSize; i<innerSize; ++i)
|
||||
res = func(res, eval.coeffByOuterInner(j,i));
|
||||
}
|
||||
else // too small to vectorize anything.
|
||||
// since this is dynamic-size hence inefficient anyway for such small sizes, don't try to optimize.
|
||||
{
|
||||
res = redux_impl<Func, Evaluator, DefaultTraversal, NoUnrolling>::run(eval, func, xpr);
|
||||
}
|
||||
|
||||
return res;
|
||||
}
|
||||
};
|
||||
|
||||
template<typename Func, typename Evaluator>
|
||||
struct redux_impl<Func, Evaluator, LinearVectorizedTraversal, CompleteUnrolling>
|
||||
{
|
||||
typedef typename Evaluator::Scalar Scalar;
|
||||
|
||||
typedef typename redux_traits<Func, Evaluator>::PacketType PacketType;
|
||||
enum {
|
||||
PacketSize = redux_traits<Func, Evaluator>::PacketSize,
|
||||
Size = Evaluator::SizeAtCompileTime,
|
||||
VectorizedSize = (int(Size) / int(PacketSize)) * int(PacketSize)
|
||||
};
|
||||
|
||||
template<typename XprType>
|
||||
EIGEN_DEVICE_FUNC static EIGEN_STRONG_INLINE
|
||||
Scalar run(const Evaluator &eval, const Func& func, const XprType &xpr)
|
||||
{
|
||||
EIGEN_ONLY_USED_FOR_DEBUG(xpr)
|
||||
eigen_assert(xpr.rows()>0 && xpr.cols()>0 && "you are using an empty matrix");
|
||||
if (VectorizedSize > 0) {
|
||||
Scalar res = func.predux(redux_vec_unroller<Func, Evaluator, 0, Size / PacketSize>::template run<PacketType>(eval,func));
|
||||
if (VectorizedSize != Size)
|
||||
res = func(res,redux_novec_unroller<Func, Evaluator, VectorizedSize, Size-VectorizedSize>::run(eval,func));
|
||||
return res;
|
||||
}
|
||||
else {
|
||||
return redux_novec_unroller<Func, Evaluator, 0, Size>::run(eval,func);
|
||||
}
|
||||
}
|
||||
};
|
||||
|
||||
// evaluator adaptor
|
||||
template<typename _XprType>
|
||||
class redux_evaluator : public internal::evaluator<_XprType>
|
||||
{
|
||||
typedef internal::evaluator<_XprType> Base;
|
||||
public:
|
||||
typedef _XprType XprType;
|
||||
EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
|
||||
explicit redux_evaluator(const XprType &xpr) : Base(xpr) {}
|
||||
|
||||
typedef typename XprType::Scalar Scalar;
|
||||
typedef typename XprType::CoeffReturnType CoeffReturnType;
|
||||
typedef typename XprType::PacketScalar PacketScalar;
|
||||
|
||||
enum {
|
||||
MaxRowsAtCompileTime = XprType::MaxRowsAtCompileTime,
|
||||
MaxColsAtCompileTime = XprType::MaxColsAtCompileTime,
|
||||
// TODO we should not remove DirectAccessBit and rather find an elegant way to query the alignment offset at runtime from the evaluator
|
||||
Flags = Base::Flags & ~DirectAccessBit,
|
||||
IsRowMajor = XprType::IsRowMajor,
|
||||
SizeAtCompileTime = XprType::SizeAtCompileTime,
|
||||
InnerSizeAtCompileTime = XprType::InnerSizeAtCompileTime
|
||||
};
|
||||
|
||||
EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
|
||||
CoeffReturnType coeffByOuterInner(Index outer, Index inner) const
|
||||
{ return Base::coeff(IsRowMajor ? outer : inner, IsRowMajor ? inner : outer); }
|
||||
|
||||
template<int LoadMode, typename PacketType>
|
||||
EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
|
||||
PacketType packetByOuterInner(Index outer, Index inner) const
|
||||
{ return Base::template packet<LoadMode,PacketType>(IsRowMajor ? outer : inner, IsRowMajor ? inner : outer); }
|
||||
|
||||
};
|
||||
|
||||
} // end namespace internal
|
||||
|
||||
/***************************************************************************
|
||||
* Part 4 : public API
|
||||
***************************************************************************/
|
||||
|
||||
|
||||
/** \returns the result of a full redux operation on the whole matrix or vector using \a func
|
||||
*
|
||||
* The template parameter \a BinaryOp is the type of the functor \a func which must be
|
||||
* an associative operator. Both current C++98 and C++11 functor styles are handled.
|
||||
*
|
||||
* \warning the matrix must be not empty, otherwise an assertion is triggered.
|
||||
*
|
||||
* \sa DenseBase::sum(), DenseBase::minCoeff(), DenseBase::maxCoeff(), MatrixBase::colwise(), MatrixBase::rowwise()
|
||||
*/
|
||||
template<typename Derived>
|
||||
template<typename Func>
|
||||
EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE typename internal::traits<Derived>::Scalar
|
||||
DenseBase<Derived>::redux(const Func& func) const
|
||||
{
|
||||
eigen_assert(this->rows()>0 && this->cols()>0 && "you are using an empty matrix");
|
||||
|
||||
typedef typename internal::redux_evaluator<Derived> ThisEvaluator;
|
||||
ThisEvaluator thisEval(derived());
|
||||
|
||||
// The initial expression is passed to the reducer as an additional argument instead of
|
||||
// passing it as a member of redux_evaluator to help
|
||||
return internal::redux_impl<Func, ThisEvaluator>::run(thisEval, func, derived());
|
||||
}
|
||||
|
||||
/** \returns the minimum of all coefficients of \c *this.
|
||||
* In case \c *this contains NaN, NaNPropagation determines the behavior:
|
||||
* NaNPropagation == PropagateFast : undefined
|
||||
* NaNPropagation == PropagateNaN : result is NaN
|
||||
* NaNPropagation == PropagateNumbers : result is minimum of elements that are not NaN
|
||||
* \warning the matrix must be not empty, otherwise an assertion is triggered.
|
||||
*/
|
||||
template<typename Derived>
|
||||
template<int NaNPropagation>
|
||||
EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE typename internal::traits<Derived>::Scalar
|
||||
DenseBase<Derived>::minCoeff() const
|
||||
{
|
||||
return derived().redux(Eigen::internal::scalar_min_op<Scalar,Scalar, NaNPropagation>());
|
||||
}
|
||||
|
||||
/** \returns the maximum of all coefficients of \c *this.
|
||||
* In case \c *this contains NaN, NaNPropagation determines the behavior:
|
||||
* NaNPropagation == PropagateFast : undefined
|
||||
* NaNPropagation == PropagateNaN : result is NaN
|
||||
* NaNPropagation == PropagateNumbers : result is maximum of elements that are not NaN
|
||||
* \warning the matrix must be not empty, otherwise an assertion is triggered.
|
||||
*/
|
||||
template<typename Derived>
|
||||
template<int NaNPropagation>
|
||||
EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE typename internal::traits<Derived>::Scalar
|
||||
DenseBase<Derived>::maxCoeff() const
|
||||
{
|
||||
return derived().redux(Eigen::internal::scalar_max_op<Scalar,Scalar, NaNPropagation>());
|
||||
}
|
||||
|
||||
/** \returns the sum of all coefficients of \c *this
|
||||
*
|
||||
* If \c *this is empty, then the value 0 is returned.
|
||||
*
|
||||
* \sa trace(), prod(), mean()
|
||||
*/
|
||||
template<typename Derived>
|
||||
EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE typename internal::traits<Derived>::Scalar
|
||||
DenseBase<Derived>::sum() const
|
||||
{
|
||||
if(SizeAtCompileTime==0 || (SizeAtCompileTime==Dynamic && size()==0))
|
||||
return Scalar(0);
|
||||
return derived().redux(Eigen::internal::scalar_sum_op<Scalar,Scalar>());
|
||||
}
|
||||
|
||||
/** \returns the mean of all coefficients of *this
|
||||
*
|
||||
* \sa trace(), prod(), sum()
|
||||
*/
|
||||
template<typename Derived>
|
||||
EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE typename internal::traits<Derived>::Scalar
|
||||
DenseBase<Derived>::mean() const
|
||||
{
|
||||
#ifdef __INTEL_COMPILER
|
||||
#pragma warning push
|
||||
#pragma warning ( disable : 2259 )
|
||||
#endif
|
||||
return Scalar(derived().redux(Eigen::internal::scalar_sum_op<Scalar,Scalar>())) / Scalar(this->size());
|
||||
#ifdef __INTEL_COMPILER
|
||||
#pragma warning pop
|
||||
#endif
|
||||
}
|
||||
|
||||
/** \returns the product of all coefficients of *this
|
||||
*
|
||||
* Example: \include MatrixBase_prod.cpp
|
||||
* Output: \verbinclude MatrixBase_prod.out
|
||||
*
|
||||
* \sa sum(), mean(), trace()
|
||||
*/
|
||||
template<typename Derived>
|
||||
EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE typename internal::traits<Derived>::Scalar
|
||||
DenseBase<Derived>::prod() const
|
||||
{
|
||||
if(SizeAtCompileTime==0 || (SizeAtCompileTime==Dynamic && size()==0))
|
||||
return Scalar(1);
|
||||
return derived().redux(Eigen::internal::scalar_product_op<Scalar>());
|
||||
}
|
||||
|
||||
/** \returns the trace of \c *this, i.e. the sum of the coefficients on the main diagonal.
|
||||
*
|
||||
* \c *this can be any matrix, not necessarily square.
|
||||
*
|
||||
* \sa diagonal(), sum()
|
||||
*/
|
||||
template<typename Derived>
|
||||
EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE typename internal::traits<Derived>::Scalar
|
||||
MatrixBase<Derived>::trace() const
|
||||
{
|
||||
return derived().diagonal().sum();
|
||||
}
|
||||
|
||||
} // end namespace Eigen
|
||||
|
||||
#endif // EIGEN_REDUX_H
|
||||
@@ -0,0 +1,381 @@
|
||||
// This file is part of Eigen, a lightweight C++ template library
|
||||
// for linear algebra.
|
||||
//
|
||||
// Copyright (C) 2012 Gael Guennebaud <gael.guennebaud@inria.fr>
|
||||
//
|
||||
// This Source Code Form is subject to the terms of the Mozilla
|
||||
// Public License v. 2.0. If a copy of the MPL was not distributed
|
||||
// with this file, You can obtain one at http://mozilla.org/MPL/2.0/.
|
||||
|
||||
#ifndef EIGEN_REF_H
|
||||
#define EIGEN_REF_H
|
||||
|
||||
namespace Eigen {
|
||||
|
||||
namespace internal {
|
||||
|
||||
template<typename _PlainObjectType, int _Options, typename _StrideType>
|
||||
struct traits<Ref<_PlainObjectType, _Options, _StrideType> >
|
||||
: public traits<Map<_PlainObjectType, _Options, _StrideType> >
|
||||
{
|
||||
typedef _PlainObjectType PlainObjectType;
|
||||
typedef _StrideType StrideType;
|
||||
enum {
|
||||
Options = _Options,
|
||||
Flags = traits<Map<_PlainObjectType, _Options, _StrideType> >::Flags | NestByRefBit,
|
||||
Alignment = traits<Map<_PlainObjectType, _Options, _StrideType> >::Alignment
|
||||
};
|
||||
|
||||
template<typename Derived> struct match {
|
||||
enum {
|
||||
IsVectorAtCompileTime = PlainObjectType::IsVectorAtCompileTime || Derived::IsVectorAtCompileTime,
|
||||
HasDirectAccess = internal::has_direct_access<Derived>::ret,
|
||||
StorageOrderMatch = IsVectorAtCompileTime || ((PlainObjectType::Flags&RowMajorBit)==(Derived::Flags&RowMajorBit)),
|
||||
InnerStrideMatch = int(StrideType::InnerStrideAtCompileTime)==int(Dynamic)
|
||||
|| int(StrideType::InnerStrideAtCompileTime)==int(Derived::InnerStrideAtCompileTime)
|
||||
|| (int(StrideType::InnerStrideAtCompileTime)==0 && int(Derived::InnerStrideAtCompileTime)==1),
|
||||
OuterStrideMatch = IsVectorAtCompileTime
|
||||
|| int(StrideType::OuterStrideAtCompileTime)==int(Dynamic) || int(StrideType::OuterStrideAtCompileTime)==int(Derived::OuterStrideAtCompileTime),
|
||||
// NOTE, this indirection of evaluator<Derived>::Alignment is needed
|
||||
// to workaround a very strange bug in MSVC related to the instantiation
|
||||
// of has_*ary_operator in evaluator<CwiseNullaryOp>.
|
||||
// This line is surprisingly very sensitive. For instance, simply adding parenthesis
|
||||
// as "DerivedAlignment = (int(evaluator<Derived>::Alignment))," will make MSVC fail...
|
||||
DerivedAlignment = int(evaluator<Derived>::Alignment),
|
||||
AlignmentMatch = (int(traits<PlainObjectType>::Alignment)==int(Unaligned)) || (DerivedAlignment >= int(Alignment)), // FIXME the first condition is not very clear, it should be replaced by the required alignment
|
||||
ScalarTypeMatch = internal::is_same<typename PlainObjectType::Scalar, typename Derived::Scalar>::value,
|
||||
MatchAtCompileTime = HasDirectAccess && StorageOrderMatch && InnerStrideMatch && OuterStrideMatch && AlignmentMatch && ScalarTypeMatch
|
||||
};
|
||||
typedef typename internal::conditional<MatchAtCompileTime,internal::true_type,internal::false_type>::type type;
|
||||
};
|
||||
|
||||
};
|
||||
|
||||
template<typename Derived>
|
||||
struct traits<RefBase<Derived> > : public traits<Derived> {};
|
||||
|
||||
}
|
||||
|
||||
template<typename Derived> class RefBase
|
||||
: public MapBase<Derived>
|
||||
{
|
||||
typedef typename internal::traits<Derived>::PlainObjectType PlainObjectType;
|
||||
typedef typename internal::traits<Derived>::StrideType StrideType;
|
||||
|
||||
public:
|
||||
|
||||
typedef MapBase<Derived> Base;
|
||||
EIGEN_DENSE_PUBLIC_INTERFACE(RefBase)
|
||||
|
||||
EIGEN_DEVICE_FUNC EIGEN_CONSTEXPR inline Index innerStride() const
|
||||
{
|
||||
return StrideType::InnerStrideAtCompileTime != 0 ? m_stride.inner() : 1;
|
||||
}
|
||||
|
||||
EIGEN_DEVICE_FUNC EIGEN_CONSTEXPR inline Index outerStride() const
|
||||
{
|
||||
return StrideType::OuterStrideAtCompileTime != 0 ? m_stride.outer()
|
||||
: IsVectorAtCompileTime ? this->size()
|
||||
: int(Flags)&RowMajorBit ? this->cols()
|
||||
: this->rows();
|
||||
}
|
||||
|
||||
EIGEN_DEVICE_FUNC RefBase()
|
||||
: Base(0,RowsAtCompileTime==Dynamic?0:RowsAtCompileTime,ColsAtCompileTime==Dynamic?0:ColsAtCompileTime),
|
||||
// Stride<> does not allow default ctor for Dynamic strides, so let' initialize it with dummy values:
|
||||
m_stride(StrideType::OuterStrideAtCompileTime==Dynamic?0:StrideType::OuterStrideAtCompileTime,
|
||||
StrideType::InnerStrideAtCompileTime==Dynamic?0:StrideType::InnerStrideAtCompileTime)
|
||||
{}
|
||||
|
||||
EIGEN_INHERIT_ASSIGNMENT_OPERATORS(RefBase)
|
||||
|
||||
protected:
|
||||
|
||||
typedef Stride<StrideType::OuterStrideAtCompileTime,StrideType::InnerStrideAtCompileTime> StrideBase;
|
||||
|
||||
// Resolves inner stride if default 0.
|
||||
static EIGEN_DEVICE_FUNC EIGEN_CONSTEXPR Index resolveInnerStride(Index inner) {
|
||||
return inner == 0 ? 1 : inner;
|
||||
}
|
||||
|
||||
// Resolves outer stride if default 0.
|
||||
static EIGEN_DEVICE_FUNC EIGEN_CONSTEXPR Index resolveOuterStride(Index inner, Index outer, Index rows, Index cols, bool isVectorAtCompileTime, bool isRowMajor) {
|
||||
return outer == 0 ? isVectorAtCompileTime ? inner * rows * cols : isRowMajor ? inner * cols : inner * rows : outer;
|
||||
}
|
||||
|
||||
// Returns true if construction is valid, false if there is a stride mismatch,
|
||||
// and fails if there is a size mismatch.
|
||||
template<typename Expression>
|
||||
EIGEN_DEVICE_FUNC bool construct(Expression& expr)
|
||||
{
|
||||
// Check matrix sizes. If this is a compile-time vector, we do allow
|
||||
// implicitly transposing.
|
||||
EIGEN_STATIC_ASSERT(
|
||||
EIGEN_PREDICATE_SAME_MATRIX_SIZE(PlainObjectType, Expression)
|
||||
// If it is a vector, the transpose sizes might match.
|
||||
|| ( PlainObjectType::IsVectorAtCompileTime
|
||||
&& ((int(PlainObjectType::RowsAtCompileTime)==Eigen::Dynamic
|
||||
|| int(Expression::ColsAtCompileTime)==Eigen::Dynamic
|
||||
|| int(PlainObjectType::RowsAtCompileTime)==int(Expression::ColsAtCompileTime))
|
||||
&& (int(PlainObjectType::ColsAtCompileTime)==Eigen::Dynamic
|
||||
|| int(Expression::RowsAtCompileTime)==Eigen::Dynamic
|
||||
|| int(PlainObjectType::ColsAtCompileTime)==int(Expression::RowsAtCompileTime)))),
|
||||
YOU_MIXED_MATRICES_OF_DIFFERENT_SIZES
|
||||
)
|
||||
|
||||
// Determine runtime rows and columns.
|
||||
Index rows = expr.rows();
|
||||
Index cols = expr.cols();
|
||||
if(PlainObjectType::RowsAtCompileTime==1)
|
||||
{
|
||||
eigen_assert(expr.rows()==1 || expr.cols()==1);
|
||||
rows = 1;
|
||||
cols = expr.size();
|
||||
}
|
||||
else if(PlainObjectType::ColsAtCompileTime==1)
|
||||
{
|
||||
eigen_assert(expr.rows()==1 || expr.cols()==1);
|
||||
rows = expr.size();
|
||||
cols = 1;
|
||||
}
|
||||
// Verify that the sizes are valid.
|
||||
eigen_assert(
|
||||
(PlainObjectType::RowsAtCompileTime == Dynamic) || (PlainObjectType::RowsAtCompileTime == rows));
|
||||
eigen_assert(
|
||||
(PlainObjectType::ColsAtCompileTime == Dynamic) || (PlainObjectType::ColsAtCompileTime == cols));
|
||||
|
||||
|
||||
// If this is a vector, we might be transposing, which means that stride should swap.
|
||||
const bool transpose = PlainObjectType::IsVectorAtCompileTime && (rows != expr.rows());
|
||||
// If the storage format differs, we also need to swap the stride.
|
||||
const bool row_major = ((PlainObjectType::Flags)&RowMajorBit) != 0;
|
||||
const bool expr_row_major = (Expression::Flags&RowMajorBit) != 0;
|
||||
const bool storage_differs = (row_major != expr_row_major);
|
||||
|
||||
const bool swap_stride = (transpose != storage_differs);
|
||||
|
||||
// Determine expr's actual strides, resolving any defaults if zero.
|
||||
const Index expr_inner_actual = resolveInnerStride(expr.innerStride());
|
||||
const Index expr_outer_actual = resolveOuterStride(expr_inner_actual,
|
||||
expr.outerStride(),
|
||||
expr.rows(),
|
||||
expr.cols(),
|
||||
Expression::IsVectorAtCompileTime != 0,
|
||||
expr_row_major);
|
||||
|
||||
// If this is a column-major row vector or row-major column vector, the inner-stride
|
||||
// is arbitrary, so set it to either the compile-time inner stride or 1.
|
||||
const bool row_vector = (rows == 1);
|
||||
const bool col_vector = (cols == 1);
|
||||
const Index inner_stride =
|
||||
( (!row_major && row_vector) || (row_major && col_vector) ) ?
|
||||
( StrideType::InnerStrideAtCompileTime > 0 ? Index(StrideType::InnerStrideAtCompileTime) : 1)
|
||||
: swap_stride ? expr_outer_actual : expr_inner_actual;
|
||||
|
||||
// If this is a column-major column vector or row-major row vector, the outer-stride
|
||||
// is arbitrary, so set it to either the compile-time outer stride or vector size.
|
||||
const Index outer_stride =
|
||||
( (!row_major && col_vector) || (row_major && row_vector) ) ?
|
||||
( StrideType::OuterStrideAtCompileTime > 0 ? Index(StrideType::OuterStrideAtCompileTime) : rows * cols * inner_stride)
|
||||
: swap_stride ? expr_inner_actual : expr_outer_actual;
|
||||
|
||||
// Check if given inner/outer strides are compatible with compile-time strides.
|
||||
const bool inner_valid = (StrideType::InnerStrideAtCompileTime == Dynamic)
|
||||
|| (resolveInnerStride(Index(StrideType::InnerStrideAtCompileTime)) == inner_stride);
|
||||
if (!inner_valid) {
|
||||
return false;
|
||||
}
|
||||
|
||||
const bool outer_valid = (StrideType::OuterStrideAtCompileTime == Dynamic)
|
||||
|| (resolveOuterStride(
|
||||
inner_stride,
|
||||
Index(StrideType::OuterStrideAtCompileTime),
|
||||
rows, cols, PlainObjectType::IsVectorAtCompileTime != 0,
|
||||
row_major)
|
||||
== outer_stride);
|
||||
if (!outer_valid) {
|
||||
return false;
|
||||
}
|
||||
|
||||
::new (static_cast<Base*>(this)) Base(expr.data(), rows, cols);
|
||||
::new (&m_stride) StrideBase(
|
||||
(StrideType::OuterStrideAtCompileTime == 0) ? 0 : outer_stride,
|
||||
(StrideType::InnerStrideAtCompileTime == 0) ? 0 : inner_stride );
|
||||
return true;
|
||||
}
|
||||
|
||||
StrideBase m_stride;
|
||||
};
|
||||
|
||||
/** \class Ref
|
||||
* \ingroup Core_Module
|
||||
*
|
||||
* \brief A matrix or vector expression mapping an existing expression
|
||||
*
|
||||
* \tparam PlainObjectType the equivalent matrix type of the mapped data
|
||||
* \tparam Options specifies the pointer alignment in bytes. It can be: \c #Aligned128, , \c #Aligned64, \c #Aligned32, \c #Aligned16, \c #Aligned8 or \c #Unaligned.
|
||||
* The default is \c #Unaligned.
|
||||
* \tparam StrideType optionally specifies strides. By default, Ref implies a contiguous storage along the inner dimension (inner stride==1),
|
||||
* but accepts a variable outer stride (leading dimension).
|
||||
* This can be overridden by specifying strides.
|
||||
* The type passed here must be a specialization of the Stride template, see examples below.
|
||||
*
|
||||
* This class provides a way to write non-template functions taking Eigen objects as parameters while limiting the number of copies.
|
||||
* A Ref<> object can represent either a const expression or a l-value:
|
||||
* \code
|
||||
* // in-out argument:
|
||||
* void foo1(Ref<VectorXf> x);
|
||||
*
|
||||
* // read-only const argument:
|
||||
* void foo2(const Ref<const VectorXf>& x);
|
||||
* \endcode
|
||||
*
|
||||
* In the in-out case, the input argument must satisfy the constraints of the actual Ref<> type, otherwise a compilation issue will be triggered.
|
||||
* By default, a Ref<VectorXf> can reference any dense vector expression of float having a contiguous memory layout.
|
||||
* Likewise, a Ref<MatrixXf> can reference any column-major dense matrix expression of float whose column's elements are contiguously stored with
|
||||
* the possibility to have a constant space in-between each column, i.e. the inner stride must be equal to 1, but the outer stride (or leading dimension)
|
||||
* can be greater than the number of rows.
|
||||
*
|
||||
* In the const case, if the input expression does not match the above requirement, then it is evaluated into a temporary before being passed to the function.
|
||||
* Here are some examples:
|
||||
* \code
|
||||
* MatrixXf A;
|
||||
* VectorXf a;
|
||||
* foo1(a.head()); // OK
|
||||
* foo1(A.col()); // OK
|
||||
* foo1(A.row()); // Compilation error because here innerstride!=1
|
||||
* foo2(A.row()); // Compilation error because A.row() is a 1xN object while foo2 is expecting a Nx1 object
|
||||
* foo2(A.row().transpose()); // The row is copied into a contiguous temporary
|
||||
* foo2(2*a); // The expression is evaluated into a temporary
|
||||
* foo2(A.col().segment(2,4)); // No temporary
|
||||
* \endcode
|
||||
*
|
||||
* The range of inputs that can be referenced without temporary can be enlarged using the last two template parameters.
|
||||
* Here is an example accepting an innerstride!=1:
|
||||
* \code
|
||||
* // in-out argument:
|
||||
* void foo3(Ref<VectorXf,0,InnerStride<> > x);
|
||||
* foo3(A.row()); // OK
|
||||
* \endcode
|
||||
* The downside here is that the function foo3 might be significantly slower than foo1 because it won't be able to exploit vectorization, and will involve more
|
||||
* expensive address computations even if the input is contiguously stored in memory. To overcome this issue, one might propose to overload internally calling a
|
||||
* template function, e.g.:
|
||||
* \code
|
||||
* // in the .h:
|
||||
* void foo(const Ref<MatrixXf>& A);
|
||||
* void foo(const Ref<MatrixXf,0,Stride<> >& A);
|
||||
*
|
||||
* // in the .cpp:
|
||||
* template<typename TypeOfA> void foo_impl(const TypeOfA& A) {
|
||||
* ... // crazy code goes here
|
||||
* }
|
||||
* void foo(const Ref<MatrixXf>& A) { foo_impl(A); }
|
||||
* void foo(const Ref<MatrixXf,0,Stride<> >& A) { foo_impl(A); }
|
||||
* \endcode
|
||||
*
|
||||
* See also the following stackoverflow questions for further references:
|
||||
* - <a href="http://stackoverflow.com/questions/21132538/correct-usage-of-the-eigenref-class">Correct usage of the Eigen::Ref<> class</a>
|
||||
*
|
||||
* \sa PlainObjectBase::Map(), \ref TopicStorageOrders
|
||||
*/
|
||||
template<typename PlainObjectType, int Options, typename StrideType> class Ref
|
||||
: public RefBase<Ref<PlainObjectType, Options, StrideType> >
|
||||
{
|
||||
private:
|
||||
typedef internal::traits<Ref> Traits;
|
||||
template<typename Derived>
|
||||
EIGEN_DEVICE_FUNC inline Ref(const PlainObjectBase<Derived>& expr,
|
||||
typename internal::enable_if<bool(Traits::template match<Derived>::MatchAtCompileTime),Derived>::type* = 0);
|
||||
public:
|
||||
|
||||
typedef RefBase<Ref> Base;
|
||||
EIGEN_DENSE_PUBLIC_INTERFACE(Ref)
|
||||
|
||||
|
||||
#ifndef EIGEN_PARSED_BY_DOXYGEN
|
||||
template<typename Derived>
|
||||
EIGEN_DEVICE_FUNC inline Ref(PlainObjectBase<Derived>& expr,
|
||||
typename internal::enable_if<bool(Traits::template match<Derived>::MatchAtCompileTime),Derived>::type* = 0)
|
||||
{
|
||||
EIGEN_STATIC_ASSERT(bool(Traits::template match<Derived>::MatchAtCompileTime), STORAGE_LAYOUT_DOES_NOT_MATCH);
|
||||
// Construction must pass since we will not create temprary storage in the non-const case.
|
||||
const bool success = Base::construct(expr.derived());
|
||||
EIGEN_UNUSED_VARIABLE(success)
|
||||
eigen_assert(success);
|
||||
}
|
||||
template<typename Derived>
|
||||
EIGEN_DEVICE_FUNC inline Ref(const DenseBase<Derived>& expr,
|
||||
typename internal::enable_if<bool(Traits::template match<Derived>::MatchAtCompileTime),Derived>::type* = 0)
|
||||
#else
|
||||
/** Implicit constructor from any dense expression */
|
||||
template<typename Derived>
|
||||
inline Ref(DenseBase<Derived>& expr)
|
||||
#endif
|
||||
{
|
||||
EIGEN_STATIC_ASSERT(bool(internal::is_lvalue<Derived>::value), THIS_EXPRESSION_IS_NOT_A_LVALUE__IT_IS_READ_ONLY);
|
||||
EIGEN_STATIC_ASSERT(bool(Traits::template match<Derived>::MatchAtCompileTime), STORAGE_LAYOUT_DOES_NOT_MATCH);
|
||||
EIGEN_STATIC_ASSERT(!Derived::IsPlainObjectBase,THIS_EXPRESSION_IS_NOT_A_LVALUE__IT_IS_READ_ONLY);
|
||||
// Construction must pass since we will not create temporary storage in the non-const case.
|
||||
const bool success = Base::construct(expr.const_cast_derived());
|
||||
EIGEN_UNUSED_VARIABLE(success)
|
||||
eigen_assert(success);
|
||||
}
|
||||
|
||||
EIGEN_INHERIT_ASSIGNMENT_OPERATORS(Ref)
|
||||
|
||||
};
|
||||
|
||||
// this is the const ref version
|
||||
template<typename TPlainObjectType, int Options, typename StrideType> class Ref<const TPlainObjectType, Options, StrideType>
|
||||
: public RefBase<Ref<const TPlainObjectType, Options, StrideType> >
|
||||
{
|
||||
typedef internal::traits<Ref> Traits;
|
||||
public:
|
||||
|
||||
typedef RefBase<Ref> Base;
|
||||
EIGEN_DENSE_PUBLIC_INTERFACE(Ref)
|
||||
|
||||
template<typename Derived>
|
||||
EIGEN_DEVICE_FUNC inline Ref(const DenseBase<Derived>& expr,
|
||||
typename internal::enable_if<bool(Traits::template match<Derived>::ScalarTypeMatch),Derived>::type* = 0)
|
||||
{
|
||||
// std::cout << match_helper<Derived>::HasDirectAccess << "," << match_helper<Derived>::OuterStrideMatch << "," << match_helper<Derived>::InnerStrideMatch << "\n";
|
||||
// std::cout << int(StrideType::OuterStrideAtCompileTime) << " - " << int(Derived::OuterStrideAtCompileTime) << "\n";
|
||||
// std::cout << int(StrideType::InnerStrideAtCompileTime) << " - " << int(Derived::InnerStrideAtCompileTime) << "\n";
|
||||
construct(expr.derived(), typename Traits::template match<Derived>::type());
|
||||
}
|
||||
|
||||
EIGEN_DEVICE_FUNC inline Ref(const Ref& other) : Base(other) {
|
||||
// copy constructor shall not copy the m_object, to avoid unnecessary malloc and copy
|
||||
}
|
||||
|
||||
template<typename OtherRef>
|
||||
EIGEN_DEVICE_FUNC inline Ref(const RefBase<OtherRef>& other) {
|
||||
construct(other.derived(), typename Traits::template match<OtherRef>::type());
|
||||
}
|
||||
|
||||
protected:
|
||||
|
||||
template<typename Expression>
|
||||
EIGEN_DEVICE_FUNC void construct(const Expression& expr,internal::true_type)
|
||||
{
|
||||
// Check if we can use the underlying expr's storage directly, otherwise call the copy version.
|
||||
if (!Base::construct(expr)) {
|
||||
construct(expr, internal::false_type());
|
||||
}
|
||||
}
|
||||
|
||||
template<typename Expression>
|
||||
EIGEN_DEVICE_FUNC void construct(const Expression& expr, internal::false_type)
|
||||
{
|
||||
internal::call_assignment_no_alias(m_object,expr,internal::assign_op<Scalar,Scalar>());
|
||||
Base::construct(m_object);
|
||||
}
|
||||
|
||||
protected:
|
||||
TPlainObjectType m_object;
|
||||
};
|
||||
|
||||
} // end namespace Eigen
|
||||
|
||||
#endif // EIGEN_REF_H
|
||||
@@ -0,0 +1,142 @@
|
||||
// This file is part of Eigen, a lightweight C++ template library
|
||||
// for linear algebra.
|
||||
//
|
||||
// Copyright (C) 2009-2010 Gael Guennebaud <gael.guennebaud@inria.fr>
|
||||
//
|
||||
// This Source Code Form is subject to the terms of the Mozilla
|
||||
// Public License v. 2.0. If a copy of the MPL was not distributed
|
||||
// with this file, You can obtain one at http://mozilla.org/MPL/2.0/.
|
||||
|
||||
#ifndef EIGEN_REPLICATE_H
|
||||
#define EIGEN_REPLICATE_H
|
||||
|
||||
namespace Eigen {
|
||||
|
||||
namespace internal {
|
||||
template<typename MatrixType,int RowFactor,int ColFactor>
|
||||
struct traits<Replicate<MatrixType,RowFactor,ColFactor> >
|
||||
: traits<MatrixType>
|
||||
{
|
||||
typedef typename MatrixType::Scalar Scalar;
|
||||
typedef typename traits<MatrixType>::StorageKind StorageKind;
|
||||
typedef typename traits<MatrixType>::XprKind XprKind;
|
||||
typedef typename ref_selector<MatrixType>::type MatrixTypeNested;
|
||||
typedef typename remove_reference<MatrixTypeNested>::type _MatrixTypeNested;
|
||||
enum {
|
||||
RowsAtCompileTime = RowFactor==Dynamic || int(MatrixType::RowsAtCompileTime)==Dynamic
|
||||
? Dynamic
|
||||
: RowFactor * MatrixType::RowsAtCompileTime,
|
||||
ColsAtCompileTime = ColFactor==Dynamic || int(MatrixType::ColsAtCompileTime)==Dynamic
|
||||
? Dynamic
|
||||
: ColFactor * MatrixType::ColsAtCompileTime,
|
||||
//FIXME we don't propagate the max sizes !!!
|
||||
MaxRowsAtCompileTime = RowsAtCompileTime,
|
||||
MaxColsAtCompileTime = ColsAtCompileTime,
|
||||
IsRowMajor = MaxRowsAtCompileTime==1 && MaxColsAtCompileTime!=1 ? 1
|
||||
: MaxColsAtCompileTime==1 && MaxRowsAtCompileTime!=1 ? 0
|
||||
: (MatrixType::Flags & RowMajorBit) ? 1 : 0,
|
||||
|
||||
// FIXME enable DirectAccess with negative strides?
|
||||
Flags = IsRowMajor ? RowMajorBit : 0
|
||||
};
|
||||
};
|
||||
}
|
||||
|
||||
/**
|
||||
* \class Replicate
|
||||
* \ingroup Core_Module
|
||||
*
|
||||
* \brief Expression of the multiple replication of a matrix or vector
|
||||
*
|
||||
* \tparam MatrixType the type of the object we are replicating
|
||||
* \tparam RowFactor number of repetitions at compile time along the vertical direction, can be Dynamic.
|
||||
* \tparam ColFactor number of repetitions at compile time along the horizontal direction, can be Dynamic.
|
||||
*
|
||||
* This class represents an expression of the multiple replication of a matrix or vector.
|
||||
* It is the return type of DenseBase::replicate() and most of the time
|
||||
* this is the only way it is used.
|
||||
*
|
||||
* \sa DenseBase::replicate()
|
||||
*/
|
||||
template<typename MatrixType,int RowFactor,int ColFactor> class Replicate
|
||||
: public internal::dense_xpr_base< Replicate<MatrixType,RowFactor,ColFactor> >::type
|
||||
{
|
||||
typedef typename internal::traits<Replicate>::MatrixTypeNested MatrixTypeNested;
|
||||
typedef typename internal::traits<Replicate>::_MatrixTypeNested _MatrixTypeNested;
|
||||
public:
|
||||
|
||||
typedef typename internal::dense_xpr_base<Replicate>::type Base;
|
||||
EIGEN_DENSE_PUBLIC_INTERFACE(Replicate)
|
||||
typedef typename internal::remove_all<MatrixType>::type NestedExpression;
|
||||
|
||||
template<typename OriginalMatrixType>
|
||||
EIGEN_DEVICE_FUNC
|
||||
inline explicit Replicate(const OriginalMatrixType& matrix)
|
||||
: m_matrix(matrix), m_rowFactor(RowFactor), m_colFactor(ColFactor)
|
||||
{
|
||||
EIGEN_STATIC_ASSERT((internal::is_same<typename internal::remove_const<MatrixType>::type,OriginalMatrixType>::value),
|
||||
THE_MATRIX_OR_EXPRESSION_THAT_YOU_PASSED_DOES_NOT_HAVE_THE_EXPECTED_TYPE)
|
||||
eigen_assert(RowFactor!=Dynamic && ColFactor!=Dynamic);
|
||||
}
|
||||
|
||||
template<typename OriginalMatrixType>
|
||||
EIGEN_DEVICE_FUNC
|
||||
inline Replicate(const OriginalMatrixType& matrix, Index rowFactor, Index colFactor)
|
||||
: m_matrix(matrix), m_rowFactor(rowFactor), m_colFactor(colFactor)
|
||||
{
|
||||
EIGEN_STATIC_ASSERT((internal::is_same<typename internal::remove_const<MatrixType>::type,OriginalMatrixType>::value),
|
||||
THE_MATRIX_OR_EXPRESSION_THAT_YOU_PASSED_DOES_NOT_HAVE_THE_EXPECTED_TYPE)
|
||||
}
|
||||
|
||||
EIGEN_DEVICE_FUNC EIGEN_CONSTEXPR
|
||||
inline Index rows() const { return m_matrix.rows() * m_rowFactor.value(); }
|
||||
EIGEN_DEVICE_FUNC EIGEN_CONSTEXPR
|
||||
inline Index cols() const { return m_matrix.cols() * m_colFactor.value(); }
|
||||
|
||||
EIGEN_DEVICE_FUNC
|
||||
const _MatrixTypeNested& nestedExpression() const
|
||||
{
|
||||
return m_matrix;
|
||||
}
|
||||
|
||||
protected:
|
||||
MatrixTypeNested m_matrix;
|
||||
const internal::variable_if_dynamic<Index, RowFactor> m_rowFactor;
|
||||
const internal::variable_if_dynamic<Index, ColFactor> m_colFactor;
|
||||
};
|
||||
|
||||
/**
|
||||
* \return an expression of the replication of \c *this
|
||||
*
|
||||
* Example: \include MatrixBase_replicate.cpp
|
||||
* Output: \verbinclude MatrixBase_replicate.out
|
||||
*
|
||||
* \sa VectorwiseOp::replicate(), DenseBase::replicate(Index,Index), class Replicate
|
||||
*/
|
||||
template<typename Derived>
|
||||
template<int RowFactor, int ColFactor>
|
||||
EIGEN_DEVICE_FUNC const Replicate<Derived,RowFactor,ColFactor>
|
||||
DenseBase<Derived>::replicate() const
|
||||
{
|
||||
return Replicate<Derived,RowFactor,ColFactor>(derived());
|
||||
}
|
||||
|
||||
/**
|
||||
* \return an expression of the replication of each column (or row) of \c *this
|
||||
*
|
||||
* Example: \include DirectionWise_replicate_int.cpp
|
||||
* Output: \verbinclude DirectionWise_replicate_int.out
|
||||
*
|
||||
* \sa VectorwiseOp::replicate(), DenseBase::replicate(), class Replicate
|
||||
*/
|
||||
template<typename ExpressionType, int Direction>
|
||||
EIGEN_DEVICE_FUNC const typename VectorwiseOp<ExpressionType,Direction>::ReplicateReturnType
|
||||
VectorwiseOp<ExpressionType,Direction>::replicate(Index factor) const
|
||||
{
|
||||
return typename VectorwiseOp<ExpressionType,Direction>::ReplicateReturnType
|
||||
(_expression(),Direction==Vertical?factor:1,Direction==Horizontal?factor:1);
|
||||
}
|
||||
|
||||
} // end namespace Eigen
|
||||
|
||||
#endif // EIGEN_REPLICATE_H
|
||||
@@ -0,0 +1,454 @@
|
||||
// This file is part of Eigen, a lightweight C++ template library
|
||||
// for linear algebra.
|
||||
//
|
||||
// Copyright (C) 2008-2017 Gael Guennebaud <gael.guennebaud@inria.fr>
|
||||
// Copyright (C) 2014 yoco <peter.xiau@gmail.com>
|
||||
//
|
||||
// This Source Code Form is subject to the terms of the Mozilla
|
||||
// Public License v. 2.0. If a copy of the MPL was not distributed
|
||||
// with this file, You can obtain one at http://mozilla.org/MPL/2.0/.
|
||||
|
||||
#ifndef EIGEN_RESHAPED_H
|
||||
#define EIGEN_RESHAPED_H
|
||||
|
||||
namespace Eigen {
|
||||
|
||||
/** \class Reshaped
|
||||
* \ingroup Core_Module
|
||||
*
|
||||
* \brief Expression of a fixed-size or dynamic-size reshape
|
||||
*
|
||||
* \tparam XprType the type of the expression in which we are taking a reshape
|
||||
* \tparam Rows the number of rows of the reshape we are taking at compile time (optional)
|
||||
* \tparam Cols the number of columns of the reshape we are taking at compile time (optional)
|
||||
* \tparam Order can be ColMajor or RowMajor, default is ColMajor.
|
||||
*
|
||||
* This class represents an expression of either a fixed-size or dynamic-size reshape.
|
||||
* It is the return type of DenseBase::reshaped(NRowsType,NColsType) and
|
||||
* most of the time this is the only way it is used.
|
||||
*
|
||||
* However, in C++98, if you want to directly maniputate reshaped expressions,
|
||||
* for instance if you want to write a function returning such an expression, you
|
||||
* will need to use this class. In C++11, it is advised to use the \em auto
|
||||
* keyword for such use cases.
|
||||
*
|
||||
* Here is an example illustrating the dynamic case:
|
||||
* \include class_Reshaped.cpp
|
||||
* Output: \verbinclude class_Reshaped.out
|
||||
*
|
||||
* Here is an example illustrating the fixed-size case:
|
||||
* \include class_FixedReshaped.cpp
|
||||
* Output: \verbinclude class_FixedReshaped.out
|
||||
*
|
||||
* \sa DenseBase::reshaped(NRowsType,NColsType)
|
||||
*/
|
||||
|
||||
namespace internal {
|
||||
|
||||
template<typename XprType, int Rows, int Cols, int Order>
|
||||
struct traits<Reshaped<XprType, Rows, Cols, Order> > : traits<XprType>
|
||||
{
|
||||
typedef typename traits<XprType>::Scalar Scalar;
|
||||
typedef typename traits<XprType>::StorageKind StorageKind;
|
||||
typedef typename traits<XprType>::XprKind XprKind;
|
||||
enum{
|
||||
MatrixRows = traits<XprType>::RowsAtCompileTime,
|
||||
MatrixCols = traits<XprType>::ColsAtCompileTime,
|
||||
RowsAtCompileTime = Rows,
|
||||
ColsAtCompileTime = Cols,
|
||||
MaxRowsAtCompileTime = Rows,
|
||||
MaxColsAtCompileTime = Cols,
|
||||
XpxStorageOrder = ((int(traits<XprType>::Flags) & RowMajorBit) == RowMajorBit) ? RowMajor : ColMajor,
|
||||
ReshapedStorageOrder = (RowsAtCompileTime == 1 && ColsAtCompileTime != 1) ? RowMajor
|
||||
: (ColsAtCompileTime == 1 && RowsAtCompileTime != 1) ? ColMajor
|
||||
: XpxStorageOrder,
|
||||
HasSameStorageOrderAsXprType = (ReshapedStorageOrder == XpxStorageOrder),
|
||||
InnerSize = (ReshapedStorageOrder==int(RowMajor)) ? int(ColsAtCompileTime) : int(RowsAtCompileTime),
|
||||
InnerStrideAtCompileTime = HasSameStorageOrderAsXprType
|
||||
? int(inner_stride_at_compile_time<XprType>::ret)
|
||||
: Dynamic,
|
||||
OuterStrideAtCompileTime = Dynamic,
|
||||
|
||||
HasDirectAccess = internal::has_direct_access<XprType>::ret
|
||||
&& (Order==int(XpxStorageOrder))
|
||||
&& ((evaluator<XprType>::Flags&LinearAccessBit)==LinearAccessBit),
|
||||
|
||||
MaskPacketAccessBit = (InnerSize == Dynamic || (InnerSize % packet_traits<Scalar>::size) == 0)
|
||||
&& (InnerStrideAtCompileTime == 1)
|
||||
? PacketAccessBit : 0,
|
||||
//MaskAlignedBit = ((OuterStrideAtCompileTime!=Dynamic) && (((OuterStrideAtCompileTime * int(sizeof(Scalar))) % 16) == 0)) ? AlignedBit : 0,
|
||||
FlagsLinearAccessBit = (RowsAtCompileTime == 1 || ColsAtCompileTime == 1) ? LinearAccessBit : 0,
|
||||
FlagsLvalueBit = is_lvalue<XprType>::value ? LvalueBit : 0,
|
||||
FlagsRowMajorBit = (ReshapedStorageOrder==int(RowMajor)) ? RowMajorBit : 0,
|
||||
FlagsDirectAccessBit = HasDirectAccess ? DirectAccessBit : 0,
|
||||
Flags0 = traits<XprType>::Flags & ( (HereditaryBits & ~RowMajorBit) | MaskPacketAccessBit),
|
||||
|
||||
Flags = (Flags0 | FlagsLinearAccessBit | FlagsLvalueBit | FlagsRowMajorBit | FlagsDirectAccessBit)
|
||||
};
|
||||
};
|
||||
|
||||
template<typename XprType, int Rows, int Cols, int Order, bool HasDirectAccess> class ReshapedImpl_dense;
|
||||
|
||||
} // end namespace internal
|
||||
|
||||
template<typename XprType, int Rows, int Cols, int Order, typename StorageKind> class ReshapedImpl;
|
||||
|
||||
template<typename XprType, int Rows, int Cols, int Order> class Reshaped
|
||||
: public ReshapedImpl<XprType, Rows, Cols, Order, typename internal::traits<XprType>::StorageKind>
|
||||
{
|
||||
typedef ReshapedImpl<XprType, Rows, Cols, Order, typename internal::traits<XprType>::StorageKind> Impl;
|
||||
public:
|
||||
//typedef typename Impl::Base Base;
|
||||
typedef Impl Base;
|
||||
EIGEN_GENERIC_PUBLIC_INTERFACE(Reshaped)
|
||||
EIGEN_INHERIT_ASSIGNMENT_OPERATORS(Reshaped)
|
||||
|
||||
/** Fixed-size constructor
|
||||
*/
|
||||
EIGEN_DEVICE_FUNC
|
||||
inline Reshaped(XprType& xpr)
|
||||
: Impl(xpr)
|
||||
{
|
||||
EIGEN_STATIC_ASSERT(RowsAtCompileTime!=Dynamic && ColsAtCompileTime!=Dynamic,THIS_METHOD_IS_ONLY_FOR_FIXED_SIZE)
|
||||
eigen_assert(Rows * Cols == xpr.rows() * xpr.cols());
|
||||
}
|
||||
|
||||
/** Dynamic-size constructor
|
||||
*/
|
||||
EIGEN_DEVICE_FUNC
|
||||
inline Reshaped(XprType& xpr,
|
||||
Index reshapeRows, Index reshapeCols)
|
||||
: Impl(xpr, reshapeRows, reshapeCols)
|
||||
{
|
||||
eigen_assert((RowsAtCompileTime==Dynamic || RowsAtCompileTime==reshapeRows)
|
||||
&& (ColsAtCompileTime==Dynamic || ColsAtCompileTime==reshapeCols));
|
||||
eigen_assert(reshapeRows * reshapeCols == xpr.rows() * xpr.cols());
|
||||
}
|
||||
};
|
||||
|
||||
// The generic default implementation for dense reshape simply forward to the internal::ReshapedImpl_dense
|
||||
// that must be specialized for direct and non-direct access...
|
||||
template<typename XprType, int Rows, int Cols, int Order>
|
||||
class ReshapedImpl<XprType, Rows, Cols, Order, Dense>
|
||||
: public internal::ReshapedImpl_dense<XprType, Rows, Cols, Order,internal::traits<Reshaped<XprType,Rows,Cols,Order> >::HasDirectAccess>
|
||||
{
|
||||
typedef internal::ReshapedImpl_dense<XprType, Rows, Cols, Order,internal::traits<Reshaped<XprType,Rows,Cols,Order> >::HasDirectAccess> Impl;
|
||||
public:
|
||||
typedef Impl Base;
|
||||
EIGEN_INHERIT_ASSIGNMENT_OPERATORS(ReshapedImpl)
|
||||
EIGEN_DEVICE_FUNC inline ReshapedImpl(XprType& xpr) : Impl(xpr) {}
|
||||
EIGEN_DEVICE_FUNC inline ReshapedImpl(XprType& xpr, Index reshapeRows, Index reshapeCols)
|
||||
: Impl(xpr, reshapeRows, reshapeCols) {}
|
||||
};
|
||||
|
||||
namespace internal {
|
||||
|
||||
/** \internal Internal implementation of dense Reshaped in the general case. */
|
||||
template<typename XprType, int Rows, int Cols, int Order>
|
||||
class ReshapedImpl_dense<XprType,Rows,Cols,Order,false>
|
||||
: public internal::dense_xpr_base<Reshaped<XprType, Rows, Cols, Order> >::type
|
||||
{
|
||||
typedef Reshaped<XprType, Rows, Cols, Order> ReshapedType;
|
||||
public:
|
||||
|
||||
typedef typename internal::dense_xpr_base<ReshapedType>::type Base;
|
||||
EIGEN_DENSE_PUBLIC_INTERFACE(ReshapedType)
|
||||
EIGEN_INHERIT_ASSIGNMENT_OPERATORS(ReshapedImpl_dense)
|
||||
|
||||
typedef typename internal::ref_selector<XprType>::non_const_type MatrixTypeNested;
|
||||
typedef typename internal::remove_all<XprType>::type NestedExpression;
|
||||
|
||||
class InnerIterator;
|
||||
|
||||
/** Fixed-size constructor
|
||||
*/
|
||||
EIGEN_DEVICE_FUNC
|
||||
inline ReshapedImpl_dense(XprType& xpr)
|
||||
: m_xpr(xpr), m_rows(Rows), m_cols(Cols)
|
||||
{}
|
||||
|
||||
/** Dynamic-size constructor
|
||||
*/
|
||||
EIGEN_DEVICE_FUNC
|
||||
inline ReshapedImpl_dense(XprType& xpr, Index nRows, Index nCols)
|
||||
: m_xpr(xpr), m_rows(nRows), m_cols(nCols)
|
||||
{}
|
||||
|
||||
EIGEN_DEVICE_FUNC Index rows() const { return m_rows; }
|
||||
EIGEN_DEVICE_FUNC Index cols() const { return m_cols; }
|
||||
|
||||
#ifdef EIGEN_PARSED_BY_DOXYGEN
|
||||
/** \sa MapBase::data() */
|
||||
EIGEN_DEVICE_FUNC inline const Scalar* data() const;
|
||||
EIGEN_DEVICE_FUNC inline Index innerStride() const;
|
||||
EIGEN_DEVICE_FUNC inline Index outerStride() const;
|
||||
#endif
|
||||
|
||||
/** \returns the nested expression */
|
||||
EIGEN_DEVICE_FUNC
|
||||
const typename internal::remove_all<XprType>::type&
|
||||
nestedExpression() const { return m_xpr; }
|
||||
|
||||
/** \returns the nested expression */
|
||||
EIGEN_DEVICE_FUNC
|
||||
typename internal::remove_reference<XprType>::type&
|
||||
nestedExpression() { return m_xpr; }
|
||||
|
||||
protected:
|
||||
|
||||
MatrixTypeNested m_xpr;
|
||||
const internal::variable_if_dynamic<Index, Rows> m_rows;
|
||||
const internal::variable_if_dynamic<Index, Cols> m_cols;
|
||||
};
|
||||
|
||||
|
||||
/** \internal Internal implementation of dense Reshaped in the direct access case. */
|
||||
template<typename XprType, int Rows, int Cols, int Order>
|
||||
class ReshapedImpl_dense<XprType, Rows, Cols, Order, true>
|
||||
: public MapBase<Reshaped<XprType, Rows, Cols, Order> >
|
||||
{
|
||||
typedef Reshaped<XprType, Rows, Cols, Order> ReshapedType;
|
||||
typedef typename internal::ref_selector<XprType>::non_const_type XprTypeNested;
|
||||
public:
|
||||
|
||||
typedef MapBase<ReshapedType> Base;
|
||||
EIGEN_DENSE_PUBLIC_INTERFACE(ReshapedType)
|
||||
EIGEN_INHERIT_ASSIGNMENT_OPERATORS(ReshapedImpl_dense)
|
||||
|
||||
/** Fixed-size constructor
|
||||
*/
|
||||
EIGEN_DEVICE_FUNC
|
||||
inline ReshapedImpl_dense(XprType& xpr)
|
||||
: Base(xpr.data()), m_xpr(xpr)
|
||||
{}
|
||||
|
||||
/** Dynamic-size constructor
|
||||
*/
|
||||
EIGEN_DEVICE_FUNC
|
||||
inline ReshapedImpl_dense(XprType& xpr, Index nRows, Index nCols)
|
||||
: Base(xpr.data(), nRows, nCols),
|
||||
m_xpr(xpr)
|
||||
{}
|
||||
|
||||
EIGEN_DEVICE_FUNC
|
||||
const typename internal::remove_all<XprTypeNested>::type& nestedExpression() const
|
||||
{
|
||||
return m_xpr;
|
||||
}
|
||||
|
||||
EIGEN_DEVICE_FUNC
|
||||
XprType& nestedExpression() { return m_xpr; }
|
||||
|
||||
/** \sa MapBase::innerStride() */
|
||||
EIGEN_DEVICE_FUNC EIGEN_CONSTEXPR
|
||||
inline Index innerStride() const
|
||||
{
|
||||
return m_xpr.innerStride();
|
||||
}
|
||||
|
||||
/** \sa MapBase::outerStride() */
|
||||
EIGEN_DEVICE_FUNC EIGEN_CONSTEXPR
|
||||
inline Index outerStride() const
|
||||
{
|
||||
return ((Flags&RowMajorBit)==RowMajorBit) ? this->cols() : this->rows();
|
||||
}
|
||||
|
||||
protected:
|
||||
|
||||
XprTypeNested m_xpr;
|
||||
};
|
||||
|
||||
// Evaluators
|
||||
template<typename ArgType, int Rows, int Cols, int Order, bool HasDirectAccess> struct reshaped_evaluator;
|
||||
|
||||
template<typename ArgType, int Rows, int Cols, int Order>
|
||||
struct evaluator<Reshaped<ArgType, Rows, Cols, Order> >
|
||||
: reshaped_evaluator<ArgType, Rows, Cols, Order, traits<Reshaped<ArgType,Rows,Cols,Order> >::HasDirectAccess>
|
||||
{
|
||||
typedef Reshaped<ArgType, Rows, Cols, Order> XprType;
|
||||
typedef typename XprType::Scalar Scalar;
|
||||
// TODO: should check for smaller packet types
|
||||
typedef typename packet_traits<Scalar>::type PacketScalar;
|
||||
|
||||
enum {
|
||||
CoeffReadCost = evaluator<ArgType>::CoeffReadCost,
|
||||
HasDirectAccess = traits<XprType>::HasDirectAccess,
|
||||
|
||||
// RowsAtCompileTime = traits<XprType>::RowsAtCompileTime,
|
||||
// ColsAtCompileTime = traits<XprType>::ColsAtCompileTime,
|
||||
// MaxRowsAtCompileTime = traits<XprType>::MaxRowsAtCompileTime,
|
||||
// MaxColsAtCompileTime = traits<XprType>::MaxColsAtCompileTime,
|
||||
//
|
||||
// InnerStrideAtCompileTime = traits<XprType>::HasSameStorageOrderAsXprType
|
||||
// ? int(inner_stride_at_compile_time<ArgType>::ret)
|
||||
// : Dynamic,
|
||||
// OuterStrideAtCompileTime = Dynamic,
|
||||
|
||||
FlagsLinearAccessBit = (traits<XprType>::RowsAtCompileTime == 1 || traits<XprType>::ColsAtCompileTime == 1 || HasDirectAccess) ? LinearAccessBit : 0,
|
||||
FlagsRowMajorBit = (traits<XprType>::ReshapedStorageOrder==int(RowMajor)) ? RowMajorBit : 0,
|
||||
FlagsDirectAccessBit = HasDirectAccess ? DirectAccessBit : 0,
|
||||
Flags0 = evaluator<ArgType>::Flags & (HereditaryBits & ~RowMajorBit),
|
||||
Flags = Flags0 | FlagsLinearAccessBit | FlagsRowMajorBit | FlagsDirectAccessBit,
|
||||
|
||||
PacketAlignment = unpacket_traits<PacketScalar>::alignment,
|
||||
Alignment = evaluator<ArgType>::Alignment
|
||||
};
|
||||
typedef reshaped_evaluator<ArgType, Rows, Cols, Order, HasDirectAccess> reshaped_evaluator_type;
|
||||
EIGEN_DEVICE_FUNC explicit evaluator(const XprType& xpr) : reshaped_evaluator_type(xpr)
|
||||
{
|
||||
EIGEN_INTERNAL_CHECK_COST_VALUE(CoeffReadCost);
|
||||
}
|
||||
};
|
||||
|
||||
template<typename ArgType, int Rows, int Cols, int Order>
|
||||
struct reshaped_evaluator<ArgType, Rows, Cols, Order, /* HasDirectAccess */ false>
|
||||
: evaluator_base<Reshaped<ArgType, Rows, Cols, Order> >
|
||||
{
|
||||
typedef Reshaped<ArgType, Rows, Cols, Order> XprType;
|
||||
|
||||
enum {
|
||||
CoeffReadCost = evaluator<ArgType>::CoeffReadCost /* TODO + cost of index computations */,
|
||||
|
||||
Flags = (evaluator<ArgType>::Flags & (HereditaryBits /*| LinearAccessBit | DirectAccessBit*/)),
|
||||
|
||||
Alignment = 0
|
||||
};
|
||||
|
||||
EIGEN_DEVICE_FUNC explicit reshaped_evaluator(const XprType& xpr) : m_argImpl(xpr.nestedExpression()), m_xpr(xpr)
|
||||
{
|
||||
EIGEN_INTERNAL_CHECK_COST_VALUE(CoeffReadCost);
|
||||
}
|
||||
|
||||
typedef typename XprType::Scalar Scalar;
|
||||
typedef typename XprType::CoeffReturnType CoeffReturnType;
|
||||
|
||||
typedef std::pair<Index, Index> RowCol;
|
||||
|
||||
inline RowCol index_remap(Index rowId, Index colId) const
|
||||
{
|
||||
if(Order==ColMajor)
|
||||
{
|
||||
const Index nth_elem_idx = colId * m_xpr.rows() + rowId;
|
||||
return RowCol(nth_elem_idx % m_xpr.nestedExpression().rows(),
|
||||
nth_elem_idx / m_xpr.nestedExpression().rows());
|
||||
}
|
||||
else
|
||||
{
|
||||
const Index nth_elem_idx = colId + rowId * m_xpr.cols();
|
||||
return RowCol(nth_elem_idx / m_xpr.nestedExpression().cols(),
|
||||
nth_elem_idx % m_xpr.nestedExpression().cols());
|
||||
}
|
||||
}
|
||||
|
||||
EIGEN_DEVICE_FUNC
|
||||
inline Scalar& coeffRef(Index rowId, Index colId)
|
||||
{
|
||||
EIGEN_STATIC_ASSERT_LVALUE(XprType)
|
||||
const RowCol row_col = index_remap(rowId, colId);
|
||||
return m_argImpl.coeffRef(row_col.first, row_col.second);
|
||||
}
|
||||
|
||||
EIGEN_DEVICE_FUNC
|
||||
inline const Scalar& coeffRef(Index rowId, Index colId) const
|
||||
{
|
||||
const RowCol row_col = index_remap(rowId, colId);
|
||||
return m_argImpl.coeffRef(row_col.first, row_col.second);
|
||||
}
|
||||
|
||||
EIGEN_DEVICE_FUNC
|
||||
EIGEN_STRONG_INLINE const CoeffReturnType coeff(Index rowId, Index colId) const
|
||||
{
|
||||
const RowCol row_col = index_remap(rowId, colId);
|
||||
return m_argImpl.coeff(row_col.first, row_col.second);
|
||||
}
|
||||
|
||||
EIGEN_DEVICE_FUNC
|
||||
inline Scalar& coeffRef(Index index)
|
||||
{
|
||||
EIGEN_STATIC_ASSERT_LVALUE(XprType)
|
||||
const RowCol row_col = index_remap(Rows == 1 ? 0 : index,
|
||||
Rows == 1 ? index : 0);
|
||||
return m_argImpl.coeffRef(row_col.first, row_col.second);
|
||||
|
||||
}
|
||||
|
||||
EIGEN_DEVICE_FUNC
|
||||
inline const Scalar& coeffRef(Index index) const
|
||||
{
|
||||
const RowCol row_col = index_remap(Rows == 1 ? 0 : index,
|
||||
Rows == 1 ? index : 0);
|
||||
return m_argImpl.coeffRef(row_col.first, row_col.second);
|
||||
}
|
||||
|
||||
EIGEN_DEVICE_FUNC
|
||||
inline const CoeffReturnType coeff(Index index) const
|
||||
{
|
||||
const RowCol row_col = index_remap(Rows == 1 ? 0 : index,
|
||||
Rows == 1 ? index : 0);
|
||||
return m_argImpl.coeff(row_col.first, row_col.second);
|
||||
}
|
||||
#if 0
|
||||
EIGEN_DEVICE_FUNC
|
||||
template<int LoadMode>
|
||||
inline PacketScalar packet(Index rowId, Index colId) const
|
||||
{
|
||||
const RowCol row_col = index_remap(rowId, colId);
|
||||
return m_argImpl.template packet<Unaligned>(row_col.first, row_col.second);
|
||||
|
||||
}
|
||||
|
||||
template<int LoadMode>
|
||||
EIGEN_DEVICE_FUNC
|
||||
inline void writePacket(Index rowId, Index colId, const PacketScalar& val)
|
||||
{
|
||||
const RowCol row_col = index_remap(rowId, colId);
|
||||
m_argImpl.const_cast_derived().template writePacket<Unaligned>
|
||||
(row_col.first, row_col.second, val);
|
||||
}
|
||||
|
||||
template<int LoadMode>
|
||||
EIGEN_DEVICE_FUNC
|
||||
inline PacketScalar packet(Index index) const
|
||||
{
|
||||
const RowCol row_col = index_remap(RowsAtCompileTime == 1 ? 0 : index,
|
||||
RowsAtCompileTime == 1 ? index : 0);
|
||||
return m_argImpl.template packet<Unaligned>(row_col.first, row_col.second);
|
||||
}
|
||||
|
||||
template<int LoadMode>
|
||||
EIGEN_DEVICE_FUNC
|
||||
inline void writePacket(Index index, const PacketScalar& val)
|
||||
{
|
||||
const RowCol row_col = index_remap(RowsAtCompileTime == 1 ? 0 : index,
|
||||
RowsAtCompileTime == 1 ? index : 0);
|
||||
return m_argImpl.template packet<Unaligned>(row_col.first, row_col.second, val);
|
||||
}
|
||||
#endif
|
||||
protected:
|
||||
|
||||
evaluator<ArgType> m_argImpl;
|
||||
const XprType& m_xpr;
|
||||
|
||||
};
|
||||
|
||||
template<typename ArgType, int Rows, int Cols, int Order>
|
||||
struct reshaped_evaluator<ArgType, Rows, Cols, Order, /* HasDirectAccess */ true>
|
||||
: mapbase_evaluator<Reshaped<ArgType, Rows, Cols, Order>,
|
||||
typename Reshaped<ArgType, Rows, Cols, Order>::PlainObject>
|
||||
{
|
||||
typedef Reshaped<ArgType, Rows, Cols, Order> XprType;
|
||||
typedef typename XprType::Scalar Scalar;
|
||||
|
||||
EIGEN_DEVICE_FUNC explicit reshaped_evaluator(const XprType& xpr)
|
||||
: mapbase_evaluator<XprType, typename XprType::PlainObject>(xpr)
|
||||
{
|
||||
// TODO: for the 3.4 release, this should be turned to an internal assertion, but let's keep it as is for the beta lifetime
|
||||
eigen_assert(((internal::UIntPtr(xpr.data()) % EIGEN_PLAIN_ENUM_MAX(1,evaluator<XprType>::Alignment)) == 0) && "data is not aligned");
|
||||
}
|
||||
};
|
||||
|
||||
} // end namespace internal
|
||||
|
||||
} // end namespace Eigen
|
||||
|
||||
#endif // EIGEN_RESHAPED_H
|
||||
@@ -0,0 +1,119 @@
|
||||
// This file is part of Eigen, a lightweight C++ template library
|
||||
// for linear algebra.
|
||||
//
|
||||
// Copyright (C) 2009-2010 Gael Guennebaud <gael.guennebaud@inria.fr>
|
||||
// Copyright (C) 2009-2010 Benoit Jacob <jacob.benoit.1@gmail.com>
|
||||
//
|
||||
// This Source Code Form is subject to the terms of the Mozilla
|
||||
// Public License v. 2.0. If a copy of the MPL was not distributed
|
||||
// with this file, You can obtain one at http://mozilla.org/MPL/2.0/.
|
||||
|
||||
#ifndef EIGEN_RETURNBYVALUE_H
|
||||
#define EIGEN_RETURNBYVALUE_H
|
||||
|
||||
namespace Eigen {
|
||||
|
||||
namespace internal {
|
||||
|
||||
template<typename Derived>
|
||||
struct traits<ReturnByValue<Derived> >
|
||||
: public traits<typename traits<Derived>::ReturnType>
|
||||
{
|
||||
enum {
|
||||
// We're disabling the DirectAccess because e.g. the constructor of
|
||||
// the Block-with-DirectAccess expression requires to have a coeffRef method.
|
||||
// Also, we don't want to have to implement the stride stuff.
|
||||
Flags = (traits<typename traits<Derived>::ReturnType>::Flags
|
||||
| EvalBeforeNestingBit) & ~DirectAccessBit
|
||||
};
|
||||
};
|
||||
|
||||
/* The ReturnByValue object doesn't even have a coeff() method.
|
||||
* So the only way that nesting it in an expression can work, is by evaluating it into a plain matrix.
|
||||
* So internal::nested always gives the plain return matrix type.
|
||||
*
|
||||
* FIXME: I don't understand why we need this specialization: isn't this taken care of by the EvalBeforeNestingBit ??
|
||||
* Answer: EvalBeforeNestingBit should be deprecated since we have the evaluators
|
||||
*/
|
||||
template<typename Derived,int n,typename PlainObject>
|
||||
struct nested_eval<ReturnByValue<Derived>, n, PlainObject>
|
||||
{
|
||||
typedef typename traits<Derived>::ReturnType type;
|
||||
};
|
||||
|
||||
} // end namespace internal
|
||||
|
||||
/** \class ReturnByValue
|
||||
* \ingroup Core_Module
|
||||
*
|
||||
*/
|
||||
template<typename Derived> class ReturnByValue
|
||||
: public internal::dense_xpr_base< ReturnByValue<Derived> >::type, internal::no_assignment_operator
|
||||
{
|
||||
public:
|
||||
typedef typename internal::traits<Derived>::ReturnType ReturnType;
|
||||
|
||||
typedef typename internal::dense_xpr_base<ReturnByValue>::type Base;
|
||||
EIGEN_DENSE_PUBLIC_INTERFACE(ReturnByValue)
|
||||
|
||||
template<typename Dest>
|
||||
EIGEN_DEVICE_FUNC
|
||||
inline void evalTo(Dest& dst) const
|
||||
{ static_cast<const Derived*>(this)->evalTo(dst); }
|
||||
EIGEN_DEVICE_FUNC EIGEN_CONSTEXPR
|
||||
inline Index rows() const EIGEN_NOEXCEPT { return static_cast<const Derived*>(this)->rows(); }
|
||||
EIGEN_DEVICE_FUNC EIGEN_CONSTEXPR
|
||||
inline Index cols() const EIGEN_NOEXCEPT { return static_cast<const Derived*>(this)->cols(); }
|
||||
|
||||
#ifndef EIGEN_PARSED_BY_DOXYGEN
|
||||
#define Unusable YOU_ARE_TRYING_TO_ACCESS_A_SINGLE_COEFFICIENT_IN_A_SPECIAL_EXPRESSION_WHERE_THAT_IS_NOT_ALLOWED_BECAUSE_THAT_WOULD_BE_INEFFICIENT
|
||||
class Unusable{
|
||||
Unusable(const Unusable&) {}
|
||||
Unusable& operator=(const Unusable&) {return *this;}
|
||||
};
|
||||
const Unusable& coeff(Index) const { return *reinterpret_cast<const Unusable*>(this); }
|
||||
const Unusable& coeff(Index,Index) const { return *reinterpret_cast<const Unusable*>(this); }
|
||||
Unusable& coeffRef(Index) { return *reinterpret_cast<Unusable*>(this); }
|
||||
Unusable& coeffRef(Index,Index) { return *reinterpret_cast<Unusable*>(this); }
|
||||
#undef Unusable
|
||||
#endif
|
||||
};
|
||||
|
||||
template<typename Derived>
|
||||
template<typename OtherDerived>
|
||||
EIGEN_DEVICE_FUNC Derived& DenseBase<Derived>::operator=(const ReturnByValue<OtherDerived>& other)
|
||||
{
|
||||
other.evalTo(derived());
|
||||
return derived();
|
||||
}
|
||||
|
||||
namespace internal {
|
||||
|
||||
// Expression is evaluated in a temporary; default implementation of Assignment is bypassed so that
|
||||
// when a ReturnByValue expression is assigned, the evaluator is not constructed.
|
||||
// TODO: Finalize port to new regime; ReturnByValue should not exist in the expression world
|
||||
|
||||
template<typename Derived>
|
||||
struct evaluator<ReturnByValue<Derived> >
|
||||
: public evaluator<typename internal::traits<Derived>::ReturnType>
|
||||
{
|
||||
typedef ReturnByValue<Derived> XprType;
|
||||
typedef typename internal::traits<Derived>::ReturnType PlainObject;
|
||||
typedef evaluator<PlainObject> Base;
|
||||
|
||||
EIGEN_DEVICE_FUNC explicit evaluator(const XprType& xpr)
|
||||
: m_result(xpr.rows(), xpr.cols())
|
||||
{
|
||||
::new (static_cast<Base*>(this)) Base(m_result);
|
||||
xpr.evalTo(m_result);
|
||||
}
|
||||
|
||||
protected:
|
||||
PlainObject m_result;
|
||||
};
|
||||
|
||||
} // end namespace internal
|
||||
|
||||
} // end namespace Eigen
|
||||
|
||||
#endif // EIGEN_RETURNBYVALUE_H
|
||||
@@ -0,0 +1,217 @@
|
||||
// This file is part of Eigen, a lightweight C++ template library
|
||||
// for linear algebra.
|
||||
//
|
||||
// Copyright (C) 2006-2008 Benoit Jacob <jacob.benoit.1@gmail.com>
|
||||
// Copyright (C) 2009 Ricard Marxer <email@ricardmarxer.com>
|
||||
// Copyright (C) 2009-2010 Gael Guennebaud <gael.guennebaud@inria.fr>
|
||||
//
|
||||
// This Source Code Form is subject to the terms of the Mozilla
|
||||
// Public License v. 2.0. If a copy of the MPL was not distributed
|
||||
// with this file, You can obtain one at http://mozilla.org/MPL/2.0/.
|
||||
|
||||
#ifndef EIGEN_REVERSE_H
|
||||
#define EIGEN_REVERSE_H
|
||||
|
||||
namespace Eigen {
|
||||
|
||||
namespace internal {
|
||||
|
||||
template<typename MatrixType, int Direction>
|
||||
struct traits<Reverse<MatrixType, Direction> >
|
||||
: traits<MatrixType>
|
||||
{
|
||||
typedef typename MatrixType::Scalar Scalar;
|
||||
typedef typename traits<MatrixType>::StorageKind StorageKind;
|
||||
typedef typename traits<MatrixType>::XprKind XprKind;
|
||||
typedef typename ref_selector<MatrixType>::type MatrixTypeNested;
|
||||
typedef typename remove_reference<MatrixTypeNested>::type _MatrixTypeNested;
|
||||
enum {
|
||||
RowsAtCompileTime = MatrixType::RowsAtCompileTime,
|
||||
ColsAtCompileTime = MatrixType::ColsAtCompileTime,
|
||||
MaxRowsAtCompileTime = MatrixType::MaxRowsAtCompileTime,
|
||||
MaxColsAtCompileTime = MatrixType::MaxColsAtCompileTime,
|
||||
Flags = _MatrixTypeNested::Flags & (RowMajorBit | LvalueBit)
|
||||
};
|
||||
};
|
||||
|
||||
template<typename PacketType, bool ReversePacket> struct reverse_packet_cond
|
||||
{
|
||||
static inline PacketType run(const PacketType& x) { return preverse(x); }
|
||||
};
|
||||
|
||||
template<typename PacketType> struct reverse_packet_cond<PacketType,false>
|
||||
{
|
||||
static inline PacketType run(const PacketType& x) { return x; }
|
||||
};
|
||||
|
||||
} // end namespace internal
|
||||
|
||||
/** \class Reverse
|
||||
* \ingroup Core_Module
|
||||
*
|
||||
* \brief Expression of the reverse of a vector or matrix
|
||||
*
|
||||
* \tparam MatrixType the type of the object of which we are taking the reverse
|
||||
* \tparam Direction defines the direction of the reverse operation, can be Vertical, Horizontal, or BothDirections
|
||||
*
|
||||
* This class represents an expression of the reverse of a vector.
|
||||
* It is the return type of MatrixBase::reverse() and VectorwiseOp::reverse()
|
||||
* and most of the time this is the only way it is used.
|
||||
*
|
||||
* \sa MatrixBase::reverse(), VectorwiseOp::reverse()
|
||||
*/
|
||||
template<typename MatrixType, int Direction> class Reverse
|
||||
: public internal::dense_xpr_base< Reverse<MatrixType, Direction> >::type
|
||||
{
|
||||
public:
|
||||
|
||||
typedef typename internal::dense_xpr_base<Reverse>::type Base;
|
||||
EIGEN_DENSE_PUBLIC_INTERFACE(Reverse)
|
||||
typedef typename internal::remove_all<MatrixType>::type NestedExpression;
|
||||
using Base::IsRowMajor;
|
||||
|
||||
protected:
|
||||
enum {
|
||||
PacketSize = internal::packet_traits<Scalar>::size,
|
||||
IsColMajor = !IsRowMajor,
|
||||
ReverseRow = (Direction == Vertical) || (Direction == BothDirections),
|
||||
ReverseCol = (Direction == Horizontal) || (Direction == BothDirections),
|
||||
OffsetRow = ReverseRow && IsColMajor ? PacketSize : 1,
|
||||
OffsetCol = ReverseCol && IsRowMajor ? PacketSize : 1,
|
||||
ReversePacket = (Direction == BothDirections)
|
||||
|| ((Direction == Vertical) && IsColMajor)
|
||||
|| ((Direction == Horizontal) && IsRowMajor)
|
||||
};
|
||||
typedef internal::reverse_packet_cond<PacketScalar,ReversePacket> reverse_packet;
|
||||
public:
|
||||
|
||||
EIGEN_DEVICE_FUNC explicit inline Reverse(const MatrixType& matrix) : m_matrix(matrix) { }
|
||||
|
||||
EIGEN_INHERIT_ASSIGNMENT_OPERATORS(Reverse)
|
||||
|
||||
EIGEN_DEVICE_FUNC EIGEN_CONSTEXPR
|
||||
inline Index rows() const EIGEN_NOEXCEPT { return m_matrix.rows(); }
|
||||
EIGEN_DEVICE_FUNC EIGEN_CONSTEXPR
|
||||
inline Index cols() const EIGEN_NOEXCEPT { return m_matrix.cols(); }
|
||||
|
||||
EIGEN_DEVICE_FUNC inline Index innerStride() const
|
||||
{
|
||||
return -m_matrix.innerStride();
|
||||
}
|
||||
|
||||
EIGEN_DEVICE_FUNC const typename internal::remove_all<typename MatrixType::Nested>::type&
|
||||
nestedExpression() const
|
||||
{
|
||||
return m_matrix;
|
||||
}
|
||||
|
||||
protected:
|
||||
typename MatrixType::Nested m_matrix;
|
||||
};
|
||||
|
||||
/** \returns an expression of the reverse of *this.
|
||||
*
|
||||
* Example: \include MatrixBase_reverse.cpp
|
||||
* Output: \verbinclude MatrixBase_reverse.out
|
||||
*
|
||||
*/
|
||||
template<typename Derived>
|
||||
EIGEN_DEVICE_FUNC inline typename DenseBase<Derived>::ReverseReturnType
|
||||
DenseBase<Derived>::reverse()
|
||||
{
|
||||
return ReverseReturnType(derived());
|
||||
}
|
||||
|
||||
|
||||
//reverse const overload moved DenseBase.h due to a CUDA compiler bug
|
||||
|
||||
/** This is the "in place" version of reverse: it reverses \c *this.
|
||||
*
|
||||
* In most cases it is probably better to simply use the reversed expression
|
||||
* of a matrix. However, when reversing the matrix data itself is really needed,
|
||||
* then this "in-place" version is probably the right choice because it provides
|
||||
* the following additional benefits:
|
||||
* - less error prone: doing the same operation with .reverse() requires special care:
|
||||
* \code m = m.reverse().eval(); \endcode
|
||||
* - this API enables reverse operations without the need for a temporary
|
||||
* - it allows future optimizations (cache friendliness, etc.)
|
||||
*
|
||||
* \sa VectorwiseOp::reverseInPlace(), reverse() */
|
||||
template<typename Derived>
|
||||
EIGEN_DEVICE_FUNC inline void DenseBase<Derived>::reverseInPlace()
|
||||
{
|
||||
if(cols()>rows())
|
||||
{
|
||||
Index half = cols()/2;
|
||||
leftCols(half).swap(rightCols(half).reverse());
|
||||
if((cols()%2)==1)
|
||||
{
|
||||
Index half2 = rows()/2;
|
||||
col(half).head(half2).swap(col(half).tail(half2).reverse());
|
||||
}
|
||||
}
|
||||
else
|
||||
{
|
||||
Index half = rows()/2;
|
||||
topRows(half).swap(bottomRows(half).reverse());
|
||||
if((rows()%2)==1)
|
||||
{
|
||||
Index half2 = cols()/2;
|
||||
row(half).head(half2).swap(row(half).tail(half2).reverse());
|
||||
}
|
||||
}
|
||||
}
|
||||
|
||||
namespace internal {
|
||||
|
||||
template<int Direction>
|
||||
struct vectorwise_reverse_inplace_impl;
|
||||
|
||||
template<>
|
||||
struct vectorwise_reverse_inplace_impl<Vertical>
|
||||
{
|
||||
template<typename ExpressionType>
|
||||
static void run(ExpressionType &xpr)
|
||||
{
|
||||
const int HalfAtCompileTime = ExpressionType::RowsAtCompileTime==Dynamic?Dynamic:ExpressionType::RowsAtCompileTime/2;
|
||||
Index half = xpr.rows()/2;
|
||||
xpr.topRows(fix<HalfAtCompileTime>(half))
|
||||
.swap(xpr.bottomRows(fix<HalfAtCompileTime>(half)).colwise().reverse());
|
||||
}
|
||||
};
|
||||
|
||||
template<>
|
||||
struct vectorwise_reverse_inplace_impl<Horizontal>
|
||||
{
|
||||
template<typename ExpressionType>
|
||||
static void run(ExpressionType &xpr)
|
||||
{
|
||||
const int HalfAtCompileTime = ExpressionType::ColsAtCompileTime==Dynamic?Dynamic:ExpressionType::ColsAtCompileTime/2;
|
||||
Index half = xpr.cols()/2;
|
||||
xpr.leftCols(fix<HalfAtCompileTime>(half))
|
||||
.swap(xpr.rightCols(fix<HalfAtCompileTime>(half)).rowwise().reverse());
|
||||
}
|
||||
};
|
||||
|
||||
} // end namespace internal
|
||||
|
||||
/** This is the "in place" version of VectorwiseOp::reverse: it reverses each column or row of \c *this.
|
||||
*
|
||||
* In most cases it is probably better to simply use the reversed expression
|
||||
* of a matrix. However, when reversing the matrix data itself is really needed,
|
||||
* then this "in-place" version is probably the right choice because it provides
|
||||
* the following additional benefits:
|
||||
* - less error prone: doing the same operation with .reverse() requires special care:
|
||||
* \code m = m.reverse().eval(); \endcode
|
||||
* - this API enables reverse operations without the need for a temporary
|
||||
*
|
||||
* \sa DenseBase::reverseInPlace(), reverse() */
|
||||
template<typename ExpressionType, int Direction>
|
||||
EIGEN_DEVICE_FUNC void VectorwiseOp<ExpressionType,Direction>::reverseInPlace()
|
||||
{
|
||||
internal::vectorwise_reverse_inplace_impl<Direction>::run(m_matrix);
|
||||
}
|
||||
|
||||
} // end namespace Eigen
|
||||
|
||||
#endif // EIGEN_REVERSE_H
|
||||
@@ -0,0 +1,164 @@
|
||||
// This file is part of Eigen, a lightweight C++ template library
|
||||
// for linear algebra.
|
||||
//
|
||||
// Copyright (C) 2008-2010 Gael Guennebaud <gael.guennebaud@inria.fr>
|
||||
//
|
||||
// This Source Code Form is subject to the terms of the Mozilla
|
||||
// Public License v. 2.0. If a copy of the MPL was not distributed
|
||||
// with this file, You can obtain one at http://mozilla.org/MPL/2.0/.
|
||||
|
||||
#ifndef EIGEN_SELECT_H
|
||||
#define EIGEN_SELECT_H
|
||||
|
||||
namespace Eigen {
|
||||
|
||||
/** \class Select
|
||||
* \ingroup Core_Module
|
||||
*
|
||||
* \brief Expression of a coefficient wise version of the C++ ternary operator ?:
|
||||
*
|
||||
* \param ConditionMatrixType the type of the \em condition expression which must be a boolean matrix
|
||||
* \param ThenMatrixType the type of the \em then expression
|
||||
* \param ElseMatrixType the type of the \em else expression
|
||||
*
|
||||
* This class represents an expression of a coefficient wise version of the C++ ternary operator ?:.
|
||||
* It is the return type of DenseBase::select() and most of the time this is the only way it is used.
|
||||
*
|
||||
* \sa DenseBase::select(const DenseBase<ThenDerived>&, const DenseBase<ElseDerived>&) const
|
||||
*/
|
||||
|
||||
namespace internal {
|
||||
template<typename ConditionMatrixType, typename ThenMatrixType, typename ElseMatrixType>
|
||||
struct traits<Select<ConditionMatrixType, ThenMatrixType, ElseMatrixType> >
|
||||
: traits<ThenMatrixType>
|
||||
{
|
||||
typedef typename traits<ThenMatrixType>::Scalar Scalar;
|
||||
typedef Dense StorageKind;
|
||||
typedef typename traits<ThenMatrixType>::XprKind XprKind;
|
||||
typedef typename ConditionMatrixType::Nested ConditionMatrixNested;
|
||||
typedef typename ThenMatrixType::Nested ThenMatrixNested;
|
||||
typedef typename ElseMatrixType::Nested ElseMatrixNested;
|
||||
enum {
|
||||
RowsAtCompileTime = ConditionMatrixType::RowsAtCompileTime,
|
||||
ColsAtCompileTime = ConditionMatrixType::ColsAtCompileTime,
|
||||
MaxRowsAtCompileTime = ConditionMatrixType::MaxRowsAtCompileTime,
|
||||
MaxColsAtCompileTime = ConditionMatrixType::MaxColsAtCompileTime,
|
||||
Flags = (unsigned int)ThenMatrixType::Flags & ElseMatrixType::Flags & RowMajorBit
|
||||
};
|
||||
};
|
||||
}
|
||||
|
||||
template<typename ConditionMatrixType, typename ThenMatrixType, typename ElseMatrixType>
|
||||
class Select : public internal::dense_xpr_base< Select<ConditionMatrixType, ThenMatrixType, ElseMatrixType> >::type,
|
||||
internal::no_assignment_operator
|
||||
{
|
||||
public:
|
||||
|
||||
typedef typename internal::dense_xpr_base<Select>::type Base;
|
||||
EIGEN_DENSE_PUBLIC_INTERFACE(Select)
|
||||
|
||||
inline EIGEN_DEVICE_FUNC
|
||||
Select(const ConditionMatrixType& a_conditionMatrix,
|
||||
const ThenMatrixType& a_thenMatrix,
|
||||
const ElseMatrixType& a_elseMatrix)
|
||||
: m_condition(a_conditionMatrix), m_then(a_thenMatrix), m_else(a_elseMatrix)
|
||||
{
|
||||
eigen_assert(m_condition.rows() == m_then.rows() && m_condition.rows() == m_else.rows());
|
||||
eigen_assert(m_condition.cols() == m_then.cols() && m_condition.cols() == m_else.cols());
|
||||
}
|
||||
|
||||
inline EIGEN_DEVICE_FUNC EIGEN_CONSTEXPR
|
||||
Index rows() const EIGEN_NOEXCEPT { return m_condition.rows(); }
|
||||
inline EIGEN_DEVICE_FUNC EIGEN_CONSTEXPR
|
||||
Index cols() const EIGEN_NOEXCEPT { return m_condition.cols(); }
|
||||
|
||||
inline EIGEN_DEVICE_FUNC
|
||||
const Scalar coeff(Index i, Index j) const
|
||||
{
|
||||
if (m_condition.coeff(i,j))
|
||||
return m_then.coeff(i,j);
|
||||
else
|
||||
return m_else.coeff(i,j);
|
||||
}
|
||||
|
||||
inline EIGEN_DEVICE_FUNC
|
||||
const Scalar coeff(Index i) const
|
||||
{
|
||||
if (m_condition.coeff(i))
|
||||
return m_then.coeff(i);
|
||||
else
|
||||
return m_else.coeff(i);
|
||||
}
|
||||
|
||||
inline EIGEN_DEVICE_FUNC const ConditionMatrixType& conditionMatrix() const
|
||||
{
|
||||
return m_condition;
|
||||
}
|
||||
|
||||
inline EIGEN_DEVICE_FUNC const ThenMatrixType& thenMatrix() const
|
||||
{
|
||||
return m_then;
|
||||
}
|
||||
|
||||
inline EIGEN_DEVICE_FUNC const ElseMatrixType& elseMatrix() const
|
||||
{
|
||||
return m_else;
|
||||
}
|
||||
|
||||
protected:
|
||||
typename ConditionMatrixType::Nested m_condition;
|
||||
typename ThenMatrixType::Nested m_then;
|
||||
typename ElseMatrixType::Nested m_else;
|
||||
};
|
||||
|
||||
|
||||
/** \returns a matrix where each coefficient (i,j) is equal to \a thenMatrix(i,j)
|
||||
* if \c *this(i,j), and \a elseMatrix(i,j) otherwise.
|
||||
*
|
||||
* Example: \include MatrixBase_select.cpp
|
||||
* Output: \verbinclude MatrixBase_select.out
|
||||
*
|
||||
* \sa class Select
|
||||
*/
|
||||
template<typename Derived>
|
||||
template<typename ThenDerived,typename ElseDerived>
|
||||
inline EIGEN_DEVICE_FUNC const Select<Derived,ThenDerived,ElseDerived>
|
||||
DenseBase<Derived>::select(const DenseBase<ThenDerived>& thenMatrix,
|
||||
const DenseBase<ElseDerived>& elseMatrix) const
|
||||
{
|
||||
return Select<Derived,ThenDerived,ElseDerived>(derived(), thenMatrix.derived(), elseMatrix.derived());
|
||||
}
|
||||
|
||||
/** Version of DenseBase::select(const DenseBase&, const DenseBase&) with
|
||||
* the \em else expression being a scalar value.
|
||||
*
|
||||
* \sa DenseBase::select(const DenseBase<ThenDerived>&, const DenseBase<ElseDerived>&) const, class Select
|
||||
*/
|
||||
template<typename Derived>
|
||||
template<typename ThenDerived>
|
||||
inline EIGEN_DEVICE_FUNC const Select<Derived,ThenDerived, typename ThenDerived::ConstantReturnType>
|
||||
DenseBase<Derived>::select(const DenseBase<ThenDerived>& thenMatrix,
|
||||
const typename ThenDerived::Scalar& elseScalar) const
|
||||
{
|
||||
return Select<Derived,ThenDerived,typename ThenDerived::ConstantReturnType>(
|
||||
derived(), thenMatrix.derived(), ThenDerived::Constant(rows(),cols(),elseScalar));
|
||||
}
|
||||
|
||||
/** Version of DenseBase::select(const DenseBase&, const DenseBase&) with
|
||||
* the \em then expression being a scalar value.
|
||||
*
|
||||
* \sa DenseBase::select(const DenseBase<ThenDerived>&, const DenseBase<ElseDerived>&) const, class Select
|
||||
*/
|
||||
template<typename Derived>
|
||||
template<typename ElseDerived>
|
||||
inline EIGEN_DEVICE_FUNC const Select<Derived, typename ElseDerived::ConstantReturnType, ElseDerived >
|
||||
DenseBase<Derived>::select(const typename ElseDerived::Scalar& thenScalar,
|
||||
const DenseBase<ElseDerived>& elseMatrix) const
|
||||
{
|
||||
return Select<Derived,typename ElseDerived::ConstantReturnType,ElseDerived>(
|
||||
derived(), ElseDerived::Constant(rows(),cols(),thenScalar), elseMatrix.derived());
|
||||
}
|
||||
|
||||
} // end namespace Eigen
|
||||
|
||||
#endif // EIGEN_SELECT_H
|
||||
@@ -0,0 +1,365 @@
|
||||
// This file is part of Eigen, a lightweight C++ template library
|
||||
// for linear algebra.
|
||||
//
|
||||
// Copyright (C) 2009 Gael Guennebaud <gael.guennebaud@inria.fr>
|
||||
//
|
||||
// This Source Code Form is subject to the terms of the Mozilla
|
||||
// Public License v. 2.0. If a copy of the MPL was not distributed
|
||||
// with this file, You can obtain one at http://mozilla.org/MPL/2.0/.
|
||||
|
||||
#ifndef EIGEN_SELFADJOINTMATRIX_H
|
||||
#define EIGEN_SELFADJOINTMATRIX_H
|
||||
|
||||
namespace Eigen {
|
||||
|
||||
/** \class SelfAdjointView
|
||||
* \ingroup Core_Module
|
||||
*
|
||||
*
|
||||
* \brief Expression of a selfadjoint matrix from a triangular part of a dense matrix
|
||||
*
|
||||
* \param MatrixType the type of the dense matrix storing the coefficients
|
||||
* \param TriangularPart can be either \c #Lower or \c #Upper
|
||||
*
|
||||
* This class is an expression of a sefladjoint matrix from a triangular part of a matrix
|
||||
* with given dense storage of the coefficients. It is the return type of MatrixBase::selfadjointView()
|
||||
* and most of the time this is the only way that it is used.
|
||||
*
|
||||
* \sa class TriangularBase, MatrixBase::selfadjointView()
|
||||
*/
|
||||
|
||||
namespace internal {
|
||||
template<typename MatrixType, unsigned int UpLo>
|
||||
struct traits<SelfAdjointView<MatrixType, UpLo> > : traits<MatrixType>
|
||||
{
|
||||
typedef typename ref_selector<MatrixType>::non_const_type MatrixTypeNested;
|
||||
typedef typename remove_all<MatrixTypeNested>::type MatrixTypeNestedCleaned;
|
||||
typedef MatrixType ExpressionType;
|
||||
typedef typename MatrixType::PlainObject FullMatrixType;
|
||||
enum {
|
||||
Mode = UpLo | SelfAdjoint,
|
||||
FlagsLvalueBit = is_lvalue<MatrixType>::value ? LvalueBit : 0,
|
||||
Flags = MatrixTypeNestedCleaned::Flags & (HereditaryBits|FlagsLvalueBit)
|
||||
& (~(PacketAccessBit | DirectAccessBit | LinearAccessBit)) // FIXME these flags should be preserved
|
||||
};
|
||||
};
|
||||
}
|
||||
|
||||
|
||||
template<typename _MatrixType, unsigned int UpLo> class SelfAdjointView
|
||||
: public TriangularBase<SelfAdjointView<_MatrixType, UpLo> >
|
||||
{
|
||||
public:
|
||||
|
||||
typedef _MatrixType MatrixType;
|
||||
typedef TriangularBase<SelfAdjointView> Base;
|
||||
typedef typename internal::traits<SelfAdjointView>::MatrixTypeNested MatrixTypeNested;
|
||||
typedef typename internal::traits<SelfAdjointView>::MatrixTypeNestedCleaned MatrixTypeNestedCleaned;
|
||||
typedef MatrixTypeNestedCleaned NestedExpression;
|
||||
|
||||
/** \brief The type of coefficients in this matrix */
|
||||
typedef typename internal::traits<SelfAdjointView>::Scalar Scalar;
|
||||
typedef typename MatrixType::StorageIndex StorageIndex;
|
||||
typedef typename internal::remove_all<typename MatrixType::ConjugateReturnType>::type MatrixConjugateReturnType;
|
||||
typedef SelfAdjointView<typename internal::add_const<MatrixType>::type, UpLo> ConstSelfAdjointView;
|
||||
|
||||
enum {
|
||||
Mode = internal::traits<SelfAdjointView>::Mode,
|
||||
Flags = internal::traits<SelfAdjointView>::Flags,
|
||||
TransposeMode = ((int(Mode) & int(Upper)) ? Lower : 0) | ((int(Mode) & int(Lower)) ? Upper : 0)
|
||||
};
|
||||
typedef typename MatrixType::PlainObject PlainObject;
|
||||
|
||||
EIGEN_DEVICE_FUNC
|
||||
explicit inline SelfAdjointView(MatrixType& matrix) : m_matrix(matrix)
|
||||
{
|
||||
EIGEN_STATIC_ASSERT(UpLo==Lower || UpLo==Upper,SELFADJOINTVIEW_ACCEPTS_UPPER_AND_LOWER_MODE_ONLY);
|
||||
}
|
||||
|
||||
EIGEN_DEVICE_FUNC EIGEN_CONSTEXPR
|
||||
inline Index rows() const EIGEN_NOEXCEPT { return m_matrix.rows(); }
|
||||
EIGEN_DEVICE_FUNC EIGEN_CONSTEXPR
|
||||
inline Index cols() const EIGEN_NOEXCEPT { return m_matrix.cols(); }
|
||||
EIGEN_DEVICE_FUNC EIGEN_CONSTEXPR
|
||||
inline Index outerStride() const EIGEN_NOEXCEPT { return m_matrix.outerStride(); }
|
||||
EIGEN_DEVICE_FUNC EIGEN_CONSTEXPR
|
||||
inline Index innerStride() const EIGEN_NOEXCEPT { return m_matrix.innerStride(); }
|
||||
|
||||
/** \sa MatrixBase::coeff()
|
||||
* \warning the coordinates must fit into the referenced triangular part
|
||||
*/
|
||||
EIGEN_DEVICE_FUNC
|
||||
inline Scalar coeff(Index row, Index col) const
|
||||
{
|
||||
Base::check_coordinates_internal(row, col);
|
||||
return m_matrix.coeff(row, col);
|
||||
}
|
||||
|
||||
/** \sa MatrixBase::coeffRef()
|
||||
* \warning the coordinates must fit into the referenced triangular part
|
||||
*/
|
||||
EIGEN_DEVICE_FUNC
|
||||
inline Scalar& coeffRef(Index row, Index col)
|
||||
{
|
||||
EIGEN_STATIC_ASSERT_LVALUE(SelfAdjointView);
|
||||
Base::check_coordinates_internal(row, col);
|
||||
return m_matrix.coeffRef(row, col);
|
||||
}
|
||||
|
||||
/** \internal */
|
||||
EIGEN_DEVICE_FUNC
|
||||
const MatrixTypeNestedCleaned& _expression() const { return m_matrix; }
|
||||
|
||||
EIGEN_DEVICE_FUNC
|
||||
const MatrixTypeNestedCleaned& nestedExpression() const { return m_matrix; }
|
||||
EIGEN_DEVICE_FUNC
|
||||
MatrixTypeNestedCleaned& nestedExpression() { return m_matrix; }
|
||||
|
||||
/** Efficient triangular matrix times vector/matrix product */
|
||||
template<typename OtherDerived>
|
||||
EIGEN_DEVICE_FUNC
|
||||
const Product<SelfAdjointView,OtherDerived>
|
||||
operator*(const MatrixBase<OtherDerived>& rhs) const
|
||||
{
|
||||
return Product<SelfAdjointView,OtherDerived>(*this, rhs.derived());
|
||||
}
|
||||
|
||||
/** Efficient vector/matrix times triangular matrix product */
|
||||
template<typename OtherDerived> friend
|
||||
EIGEN_DEVICE_FUNC
|
||||
const Product<OtherDerived,SelfAdjointView>
|
||||
operator*(const MatrixBase<OtherDerived>& lhs, const SelfAdjointView& rhs)
|
||||
{
|
||||
return Product<OtherDerived,SelfAdjointView>(lhs.derived(),rhs);
|
||||
}
|
||||
|
||||
friend EIGEN_DEVICE_FUNC
|
||||
const SelfAdjointView<const EIGEN_SCALAR_BINARYOP_EXPR_RETURN_TYPE(Scalar,MatrixType,product),UpLo>
|
||||
operator*(const Scalar& s, const SelfAdjointView& mat)
|
||||
{
|
||||
return (s*mat.nestedExpression()).template selfadjointView<UpLo>();
|
||||
}
|
||||
|
||||
/** Perform a symmetric rank 2 update of the selfadjoint matrix \c *this:
|
||||
* \f$ this = this + \alpha u v^* + conj(\alpha) v u^* \f$
|
||||
* \returns a reference to \c *this
|
||||
*
|
||||
* The vectors \a u and \c v \b must be column vectors, however they can be
|
||||
* a adjoint expression without any overhead. Only the meaningful triangular
|
||||
* part of the matrix is updated, the rest is left unchanged.
|
||||
*
|
||||
* \sa rankUpdate(const MatrixBase<DerivedU>&, Scalar)
|
||||
*/
|
||||
template<typename DerivedU, typename DerivedV>
|
||||
EIGEN_DEVICE_FUNC
|
||||
SelfAdjointView& rankUpdate(const MatrixBase<DerivedU>& u, const MatrixBase<DerivedV>& v, const Scalar& alpha = Scalar(1));
|
||||
|
||||
/** Perform a symmetric rank K update of the selfadjoint matrix \c *this:
|
||||
* \f$ this = this + \alpha ( u u^* ) \f$ where \a u is a vector or matrix.
|
||||
*
|
||||
* \returns a reference to \c *this
|
||||
*
|
||||
* Note that to perform \f$ this = this + \alpha ( u^* u ) \f$ you can simply
|
||||
* call this function with u.adjoint().
|
||||
*
|
||||
* \sa rankUpdate(const MatrixBase<DerivedU>&, const MatrixBase<DerivedV>&, Scalar)
|
||||
*/
|
||||
template<typename DerivedU>
|
||||
EIGEN_DEVICE_FUNC
|
||||
SelfAdjointView& rankUpdate(const MatrixBase<DerivedU>& u, const Scalar& alpha = Scalar(1));
|
||||
|
||||
/** \returns an expression of a triangular view extracted from the current selfadjoint view of a given triangular part
|
||||
*
|
||||
* The parameter \a TriMode can have the following values: \c #Upper, \c #StrictlyUpper, \c #UnitUpper,
|
||||
* \c #Lower, \c #StrictlyLower, \c #UnitLower.
|
||||
*
|
||||
* If \c TriMode references the same triangular part than \c *this, then this method simply return a \c TriangularView of the nested expression,
|
||||
* otherwise, the nested expression is first transposed, thus returning a \c TriangularView<Transpose<MatrixType>> object.
|
||||
*
|
||||
* \sa MatrixBase::triangularView(), class TriangularView
|
||||
*/
|
||||
template<unsigned int TriMode>
|
||||
EIGEN_DEVICE_FUNC
|
||||
typename internal::conditional<(TriMode&(Upper|Lower))==(UpLo&(Upper|Lower)),
|
||||
TriangularView<MatrixType,TriMode>,
|
||||
TriangularView<typename MatrixType::AdjointReturnType,TriMode> >::type
|
||||
triangularView() const
|
||||
{
|
||||
typename internal::conditional<(TriMode&(Upper|Lower))==(UpLo&(Upper|Lower)), MatrixType&, typename MatrixType::ConstTransposeReturnType>::type tmp1(m_matrix);
|
||||
typename internal::conditional<(TriMode&(Upper|Lower))==(UpLo&(Upper|Lower)), MatrixType&, typename MatrixType::AdjointReturnType>::type tmp2(tmp1);
|
||||
return typename internal::conditional<(TriMode&(Upper|Lower))==(UpLo&(Upper|Lower)),
|
||||
TriangularView<MatrixType,TriMode>,
|
||||
TriangularView<typename MatrixType::AdjointReturnType,TriMode> >::type(tmp2);
|
||||
}
|
||||
|
||||
typedef SelfAdjointView<const MatrixConjugateReturnType,UpLo> ConjugateReturnType;
|
||||
/** \sa MatrixBase::conjugate() const */
|
||||
EIGEN_DEVICE_FUNC
|
||||
inline const ConjugateReturnType conjugate() const
|
||||
{ return ConjugateReturnType(m_matrix.conjugate()); }
|
||||
|
||||
/** \returns an expression of the complex conjugate of \c *this if Cond==true,
|
||||
* returns \c *this otherwise.
|
||||
*/
|
||||
template<bool Cond>
|
||||
EIGEN_DEVICE_FUNC
|
||||
inline typename internal::conditional<Cond,ConjugateReturnType,ConstSelfAdjointView>::type
|
||||
conjugateIf() const
|
||||
{
|
||||
typedef typename internal::conditional<Cond,ConjugateReturnType,ConstSelfAdjointView>::type ReturnType;
|
||||
return ReturnType(m_matrix.template conjugateIf<Cond>());
|
||||
}
|
||||
|
||||
typedef SelfAdjointView<const typename MatrixType::AdjointReturnType,TransposeMode> AdjointReturnType;
|
||||
/** \sa MatrixBase::adjoint() const */
|
||||
EIGEN_DEVICE_FUNC
|
||||
inline const AdjointReturnType adjoint() const
|
||||
{ return AdjointReturnType(m_matrix.adjoint()); }
|
||||
|
||||
typedef SelfAdjointView<typename MatrixType::TransposeReturnType,TransposeMode> TransposeReturnType;
|
||||
/** \sa MatrixBase::transpose() */
|
||||
EIGEN_DEVICE_FUNC
|
||||
inline TransposeReturnType transpose()
|
||||
{
|
||||
EIGEN_STATIC_ASSERT_LVALUE(MatrixType)
|
||||
typename MatrixType::TransposeReturnType tmp(m_matrix);
|
||||
return TransposeReturnType(tmp);
|
||||
}
|
||||
|
||||
typedef SelfAdjointView<const typename MatrixType::ConstTransposeReturnType,TransposeMode> ConstTransposeReturnType;
|
||||
/** \sa MatrixBase::transpose() const */
|
||||
EIGEN_DEVICE_FUNC
|
||||
inline const ConstTransposeReturnType transpose() const
|
||||
{
|
||||
return ConstTransposeReturnType(m_matrix.transpose());
|
||||
}
|
||||
|
||||
/** \returns a const expression of the main diagonal of the matrix \c *this
|
||||
*
|
||||
* This method simply returns the diagonal of the nested expression, thus by-passing the SelfAdjointView decorator.
|
||||
*
|
||||
* \sa MatrixBase::diagonal(), class Diagonal */
|
||||
EIGEN_DEVICE_FUNC
|
||||
typename MatrixType::ConstDiagonalReturnType diagonal() const
|
||||
{
|
||||
return typename MatrixType::ConstDiagonalReturnType(m_matrix);
|
||||
}
|
||||
|
||||
/////////// Cholesky module ///////////
|
||||
|
||||
const LLT<PlainObject, UpLo> llt() const;
|
||||
const LDLT<PlainObject, UpLo> ldlt() const;
|
||||
|
||||
/////////// Eigenvalue module ///////////
|
||||
|
||||
/** Real part of #Scalar */
|
||||
typedef typename NumTraits<Scalar>::Real RealScalar;
|
||||
/** Return type of eigenvalues() */
|
||||
typedef Matrix<RealScalar, internal::traits<MatrixType>::ColsAtCompileTime, 1> EigenvaluesReturnType;
|
||||
|
||||
EIGEN_DEVICE_FUNC
|
||||
EigenvaluesReturnType eigenvalues() const;
|
||||
EIGEN_DEVICE_FUNC
|
||||
RealScalar operatorNorm() const;
|
||||
|
||||
protected:
|
||||
MatrixTypeNested m_matrix;
|
||||
};
|
||||
|
||||
|
||||
// template<typename OtherDerived, typename MatrixType, unsigned int UpLo>
|
||||
// internal::selfadjoint_matrix_product_returntype<OtherDerived,SelfAdjointView<MatrixType,UpLo> >
|
||||
// operator*(const MatrixBase<OtherDerived>& lhs, const SelfAdjointView<MatrixType,UpLo>& rhs)
|
||||
// {
|
||||
// return internal::matrix_selfadjoint_product_returntype<OtherDerived,SelfAdjointView<MatrixType,UpLo> >(lhs.derived(),rhs);
|
||||
// }
|
||||
|
||||
// selfadjoint to dense matrix
|
||||
|
||||
namespace internal {
|
||||
|
||||
// TODO currently a selfadjoint expression has the form SelfAdjointView<.,.>
|
||||
// in the future selfadjoint-ness should be defined by the expression traits
|
||||
// such that Transpose<SelfAdjointView<.,.> > is valid. (currently TriangularBase::transpose() is overloaded to make it work)
|
||||
template<typename MatrixType, unsigned int Mode>
|
||||
struct evaluator_traits<SelfAdjointView<MatrixType,Mode> >
|
||||
{
|
||||
typedef typename storage_kind_to_evaluator_kind<typename MatrixType::StorageKind>::Kind Kind;
|
||||
typedef SelfAdjointShape Shape;
|
||||
};
|
||||
|
||||
template<int UpLo, int SetOpposite, typename DstEvaluatorTypeT, typename SrcEvaluatorTypeT, typename Functor, int Version>
|
||||
class triangular_dense_assignment_kernel<UpLo,SelfAdjoint,SetOpposite,DstEvaluatorTypeT,SrcEvaluatorTypeT,Functor,Version>
|
||||
: public generic_dense_assignment_kernel<DstEvaluatorTypeT, SrcEvaluatorTypeT, Functor, Version>
|
||||
{
|
||||
protected:
|
||||
typedef generic_dense_assignment_kernel<DstEvaluatorTypeT, SrcEvaluatorTypeT, Functor, Version> Base;
|
||||
typedef typename Base::DstXprType DstXprType;
|
||||
typedef typename Base::SrcXprType SrcXprType;
|
||||
using Base::m_dst;
|
||||
using Base::m_src;
|
||||
using Base::m_functor;
|
||||
public:
|
||||
|
||||
typedef typename Base::DstEvaluatorType DstEvaluatorType;
|
||||
typedef typename Base::SrcEvaluatorType SrcEvaluatorType;
|
||||
typedef typename Base::Scalar Scalar;
|
||||
typedef typename Base::AssignmentTraits AssignmentTraits;
|
||||
|
||||
|
||||
EIGEN_DEVICE_FUNC triangular_dense_assignment_kernel(DstEvaluatorType &dst, const SrcEvaluatorType &src, const Functor &func, DstXprType& dstExpr)
|
||||
: Base(dst, src, func, dstExpr)
|
||||
{}
|
||||
|
||||
EIGEN_DEVICE_FUNC void assignCoeff(Index row, Index col)
|
||||
{
|
||||
eigen_internal_assert(row!=col);
|
||||
Scalar tmp = m_src.coeff(row,col);
|
||||
m_functor.assignCoeff(m_dst.coeffRef(row,col), tmp);
|
||||
m_functor.assignCoeff(m_dst.coeffRef(col,row), numext::conj(tmp));
|
||||
}
|
||||
|
||||
EIGEN_DEVICE_FUNC void assignDiagonalCoeff(Index id)
|
||||
{
|
||||
Base::assignCoeff(id,id);
|
||||
}
|
||||
|
||||
EIGEN_DEVICE_FUNC void assignOppositeCoeff(Index, Index)
|
||||
{ eigen_internal_assert(false && "should never be called"); }
|
||||
};
|
||||
|
||||
} // end namespace internal
|
||||
|
||||
/***************************************************************************
|
||||
* Implementation of MatrixBase methods
|
||||
***************************************************************************/
|
||||
|
||||
/** This is the const version of MatrixBase::selfadjointView() */
|
||||
template<typename Derived>
|
||||
template<unsigned int UpLo>
|
||||
EIGEN_DEVICE_FUNC typename MatrixBase<Derived>::template ConstSelfAdjointViewReturnType<UpLo>::Type
|
||||
MatrixBase<Derived>::selfadjointView() const
|
||||
{
|
||||
return typename ConstSelfAdjointViewReturnType<UpLo>::Type(derived());
|
||||
}
|
||||
|
||||
/** \returns an expression of a symmetric/self-adjoint view extracted from the upper or lower triangular part of the current matrix
|
||||
*
|
||||
* The parameter \a UpLo can be either \c #Upper or \c #Lower
|
||||
*
|
||||
* Example: \include MatrixBase_selfadjointView.cpp
|
||||
* Output: \verbinclude MatrixBase_selfadjointView.out
|
||||
*
|
||||
* \sa class SelfAdjointView
|
||||
*/
|
||||
template<typename Derived>
|
||||
template<unsigned int UpLo>
|
||||
EIGEN_DEVICE_FUNC typename MatrixBase<Derived>::template SelfAdjointViewReturnType<UpLo>::Type
|
||||
MatrixBase<Derived>::selfadjointView()
|
||||
{
|
||||
return typename SelfAdjointViewReturnType<UpLo>::Type(derived());
|
||||
}
|
||||
|
||||
} // end namespace Eigen
|
||||
|
||||
#endif // EIGEN_SELFADJOINTMATRIX_H
|
||||
@@ -0,0 +1,47 @@
|
||||
// This file is part of Eigen, a lightweight C++ template library
|
||||
// for linear algebra.
|
||||
//
|
||||
// Copyright (C) 2009-2010 Gael Guennebaud <gael.guennebaud@inria.fr>
|
||||
//
|
||||
// This Source Code Form is subject to the terms of the Mozilla
|
||||
// Public License v. 2.0. If a copy of the MPL was not distributed
|
||||
// with this file, You can obtain one at http://mozilla.org/MPL/2.0/.
|
||||
|
||||
#ifndef EIGEN_SELFCWISEBINARYOP_H
|
||||
#define EIGEN_SELFCWISEBINARYOP_H
|
||||
|
||||
namespace Eigen {
|
||||
|
||||
// TODO generalize the scalar type of 'other'
|
||||
|
||||
template<typename Derived>
|
||||
EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE Derived& DenseBase<Derived>::operator*=(const Scalar& other)
|
||||
{
|
||||
internal::call_assignment(this->derived(), PlainObject::Constant(rows(),cols(),other), internal::mul_assign_op<Scalar,Scalar>());
|
||||
return derived();
|
||||
}
|
||||
|
||||
template<typename Derived>
|
||||
EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE Derived& ArrayBase<Derived>::operator+=(const Scalar& other)
|
||||
{
|
||||
internal::call_assignment(this->derived(), PlainObject::Constant(rows(),cols(),other), internal::add_assign_op<Scalar,Scalar>());
|
||||
return derived();
|
||||
}
|
||||
|
||||
template<typename Derived>
|
||||
EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE Derived& ArrayBase<Derived>::operator-=(const Scalar& other)
|
||||
{
|
||||
internal::call_assignment(this->derived(), PlainObject::Constant(rows(),cols(),other), internal::sub_assign_op<Scalar,Scalar>());
|
||||
return derived();
|
||||
}
|
||||
|
||||
template<typename Derived>
|
||||
EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE Derived& DenseBase<Derived>::operator/=(const Scalar& other)
|
||||
{
|
||||
internal::call_assignment(this->derived(), PlainObject::Constant(rows(),cols(),other), internal::div_assign_op<Scalar,Scalar>());
|
||||
return derived();
|
||||
}
|
||||
|
||||
} // end namespace Eigen
|
||||
|
||||
#endif // EIGEN_SELFCWISEBINARYOP_H
|
||||
@@ -0,0 +1,188 @@
|
||||
// This file is part of Eigen, a lightweight C++ template library
|
||||
// for linear algebra.
|
||||
//
|
||||
// Copyright (C) 2014 Gael Guennebaud <gael.guennebaud@inria.fr>
|
||||
//
|
||||
// This Source Code Form is subject to the terms of the Mozilla
|
||||
// Public License v. 2.0. If a copy of the MPL was not distributed
|
||||
// with this file, You can obtain one at http://mozilla.org/MPL/2.0/.
|
||||
|
||||
#ifndef EIGEN_SOLVE_H
|
||||
#define EIGEN_SOLVE_H
|
||||
|
||||
namespace Eigen {
|
||||
|
||||
template<typename Decomposition, typename RhsType, typename StorageKind> class SolveImpl;
|
||||
|
||||
/** \class Solve
|
||||
* \ingroup Core_Module
|
||||
*
|
||||
* \brief Pseudo expression representing a solving operation
|
||||
*
|
||||
* \tparam Decomposition the type of the matrix or decomposition object
|
||||
* \tparam Rhstype the type of the right-hand side
|
||||
*
|
||||
* This class represents an expression of A.solve(B)
|
||||
* and most of the time this is the only way it is used.
|
||||
*
|
||||
*/
|
||||
namespace internal {
|
||||
|
||||
// this solve_traits class permits to determine the evaluation type with respect to storage kind (Dense vs Sparse)
|
||||
template<typename Decomposition, typename RhsType,typename StorageKind> struct solve_traits;
|
||||
|
||||
template<typename Decomposition, typename RhsType>
|
||||
struct solve_traits<Decomposition,RhsType,Dense>
|
||||
{
|
||||
typedef typename make_proper_matrix_type<typename RhsType::Scalar,
|
||||
Decomposition::ColsAtCompileTime,
|
||||
RhsType::ColsAtCompileTime,
|
||||
RhsType::PlainObject::Options,
|
||||
Decomposition::MaxColsAtCompileTime,
|
||||
RhsType::MaxColsAtCompileTime>::type PlainObject;
|
||||
};
|
||||
|
||||
template<typename Decomposition, typename RhsType>
|
||||
struct traits<Solve<Decomposition, RhsType> >
|
||||
: traits<typename solve_traits<Decomposition,RhsType,typename internal::traits<RhsType>::StorageKind>::PlainObject>
|
||||
{
|
||||
typedef typename solve_traits<Decomposition,RhsType,typename internal::traits<RhsType>::StorageKind>::PlainObject PlainObject;
|
||||
typedef typename promote_index_type<typename Decomposition::StorageIndex, typename RhsType::StorageIndex>::type StorageIndex;
|
||||
typedef traits<PlainObject> BaseTraits;
|
||||
enum {
|
||||
Flags = BaseTraits::Flags & RowMajorBit,
|
||||
CoeffReadCost = HugeCost
|
||||
};
|
||||
};
|
||||
|
||||
}
|
||||
|
||||
|
||||
template<typename Decomposition, typename RhsType>
|
||||
class Solve : public SolveImpl<Decomposition,RhsType,typename internal::traits<RhsType>::StorageKind>
|
||||
{
|
||||
public:
|
||||
typedef typename internal::traits<Solve>::PlainObject PlainObject;
|
||||
typedef typename internal::traits<Solve>::StorageIndex StorageIndex;
|
||||
|
||||
Solve(const Decomposition &dec, const RhsType &rhs)
|
||||
: m_dec(dec), m_rhs(rhs)
|
||||
{}
|
||||
|
||||
EIGEN_DEVICE_FUNC EIGEN_CONSTEXPR Index rows() const EIGEN_NOEXCEPT { return m_dec.cols(); }
|
||||
EIGEN_DEVICE_FUNC EIGEN_CONSTEXPR Index cols() const EIGEN_NOEXCEPT { return m_rhs.cols(); }
|
||||
|
||||
EIGEN_DEVICE_FUNC const Decomposition& dec() const { return m_dec; }
|
||||
EIGEN_DEVICE_FUNC const RhsType& rhs() const { return m_rhs; }
|
||||
|
||||
protected:
|
||||
const Decomposition &m_dec;
|
||||
const RhsType &m_rhs;
|
||||
};
|
||||
|
||||
|
||||
// Specialization of the Solve expression for dense results
|
||||
template<typename Decomposition, typename RhsType>
|
||||
class SolveImpl<Decomposition,RhsType,Dense>
|
||||
: public MatrixBase<Solve<Decomposition,RhsType> >
|
||||
{
|
||||
typedef Solve<Decomposition,RhsType> Derived;
|
||||
|
||||
public:
|
||||
|
||||
typedef MatrixBase<Solve<Decomposition,RhsType> > Base;
|
||||
EIGEN_DENSE_PUBLIC_INTERFACE(Derived)
|
||||
|
||||
private:
|
||||
|
||||
Scalar coeff(Index row, Index col) const;
|
||||
Scalar coeff(Index i) const;
|
||||
};
|
||||
|
||||
// Generic API dispatcher
|
||||
template<typename Decomposition, typename RhsType, typename StorageKind>
|
||||
class SolveImpl : public internal::generic_xpr_base<Solve<Decomposition,RhsType>, MatrixXpr, StorageKind>::type
|
||||
{
|
||||
public:
|
||||
typedef typename internal::generic_xpr_base<Solve<Decomposition,RhsType>, MatrixXpr, StorageKind>::type Base;
|
||||
};
|
||||
|
||||
namespace internal {
|
||||
|
||||
// Evaluator of Solve -> eval into a temporary
|
||||
template<typename Decomposition, typename RhsType>
|
||||
struct evaluator<Solve<Decomposition,RhsType> >
|
||||
: public evaluator<typename Solve<Decomposition,RhsType>::PlainObject>
|
||||
{
|
||||
typedef Solve<Decomposition,RhsType> SolveType;
|
||||
typedef typename SolveType::PlainObject PlainObject;
|
||||
typedef evaluator<PlainObject> Base;
|
||||
|
||||
enum { Flags = Base::Flags | EvalBeforeNestingBit };
|
||||
|
||||
EIGEN_DEVICE_FUNC explicit evaluator(const SolveType& solve)
|
||||
: m_result(solve.rows(), solve.cols())
|
||||
{
|
||||
::new (static_cast<Base*>(this)) Base(m_result);
|
||||
solve.dec()._solve_impl(solve.rhs(), m_result);
|
||||
}
|
||||
|
||||
protected:
|
||||
PlainObject m_result;
|
||||
};
|
||||
|
||||
// Specialization for "dst = dec.solve(rhs)"
|
||||
// NOTE we need to specialize it for Dense2Dense to avoid ambiguous specialization error and a Sparse2Sparse specialization must exist somewhere
|
||||
template<typename DstXprType, typename DecType, typename RhsType, typename Scalar>
|
||||
struct Assignment<DstXprType, Solve<DecType,RhsType>, internal::assign_op<Scalar,Scalar>, Dense2Dense>
|
||||
{
|
||||
typedef Solve<DecType,RhsType> SrcXprType;
|
||||
static void run(DstXprType &dst, const SrcXprType &src, const internal::assign_op<Scalar,Scalar> &)
|
||||
{
|
||||
Index dstRows = src.rows();
|
||||
Index dstCols = src.cols();
|
||||
if((dst.rows()!=dstRows) || (dst.cols()!=dstCols))
|
||||
dst.resize(dstRows, dstCols);
|
||||
|
||||
src.dec()._solve_impl(src.rhs(), dst);
|
||||
}
|
||||
};
|
||||
|
||||
// Specialization for "dst = dec.transpose().solve(rhs)"
|
||||
template<typename DstXprType, typename DecType, typename RhsType, typename Scalar>
|
||||
struct Assignment<DstXprType, Solve<Transpose<const DecType>,RhsType>, internal::assign_op<Scalar,Scalar>, Dense2Dense>
|
||||
{
|
||||
typedef Solve<Transpose<const DecType>,RhsType> SrcXprType;
|
||||
static void run(DstXprType &dst, const SrcXprType &src, const internal::assign_op<Scalar,Scalar> &)
|
||||
{
|
||||
Index dstRows = src.rows();
|
||||
Index dstCols = src.cols();
|
||||
if((dst.rows()!=dstRows) || (dst.cols()!=dstCols))
|
||||
dst.resize(dstRows, dstCols);
|
||||
|
||||
src.dec().nestedExpression().template _solve_impl_transposed<false>(src.rhs(), dst);
|
||||
}
|
||||
};
|
||||
|
||||
// Specialization for "dst = dec.adjoint().solve(rhs)"
|
||||
template<typename DstXprType, typename DecType, typename RhsType, typename Scalar>
|
||||
struct Assignment<DstXprType, Solve<CwiseUnaryOp<internal::scalar_conjugate_op<typename DecType::Scalar>, const Transpose<const DecType> >,RhsType>,
|
||||
internal::assign_op<Scalar,Scalar>, Dense2Dense>
|
||||
{
|
||||
typedef Solve<CwiseUnaryOp<internal::scalar_conjugate_op<typename DecType::Scalar>, const Transpose<const DecType> >,RhsType> SrcXprType;
|
||||
static void run(DstXprType &dst, const SrcXprType &src, const internal::assign_op<Scalar,Scalar> &)
|
||||
{
|
||||
Index dstRows = src.rows();
|
||||
Index dstCols = src.cols();
|
||||
if((dst.rows()!=dstRows) || (dst.cols()!=dstCols))
|
||||
dst.resize(dstRows, dstCols);
|
||||
|
||||
src.dec().nestedExpression().nestedExpression().template _solve_impl_transposed<true>(src.rhs(), dst);
|
||||
}
|
||||
};
|
||||
|
||||
} // end namespace internal
|
||||
|
||||
} // end namespace Eigen
|
||||
|
||||
#endif // EIGEN_SOLVE_H
|
||||
@@ -0,0 +1,235 @@
|
||||
// This file is part of Eigen, a lightweight C++ template library
|
||||
// for linear algebra.
|
||||
//
|
||||
// Copyright (C) 2008-2009 Gael Guennebaud <gael.guennebaud@inria.fr>
|
||||
//
|
||||
// This Source Code Form is subject to the terms of the Mozilla
|
||||
// Public License v. 2.0. If a copy of the MPL was not distributed
|
||||
// with this file, You can obtain one at http://mozilla.org/MPL/2.0/.
|
||||
|
||||
#ifndef EIGEN_SOLVETRIANGULAR_H
|
||||
#define EIGEN_SOLVETRIANGULAR_H
|
||||
|
||||
namespace Eigen {
|
||||
|
||||
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, int OtherInnerStride>
|
||||
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 EIGEN_DEVICE_FUNC 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;
|
||||
|
||||
ei_declare_aligned_stack_constructed_variable(RhsScalar,actualRhs,rhs.size(),
|
||||
(useRhsDirectly ? rhs.data() : 0));
|
||||
|
||||
if(!useRhsDirectly)
|
||||
MappedRhs(actualRhs,rhs.size()) = rhs;
|
||||
|
||||
triangular_solve_vector<LhsScalar, RhsScalar, 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());
|
||||
}
|
||||
};
|
||||
|
||||
// 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 blas_traits<Lhs> LhsProductTraits;
|
||||
typedef typename LhsProductTraits::DirectLinearAccessType ActualLhsType;
|
||||
|
||||
static EIGEN_DEVICE_FUNC void run(const Lhs& lhs, Rhs& rhs)
|
||||
{
|
||||
typename internal::add_const_on_value_type<ActualLhsType>::type actualLhs = LhsProductTraits::extract(lhs);
|
||||
|
||||
const Index size = lhs.rows();
|
||||
const Index othersize = Side==OnTheLeft? rhs.cols() : rhs.rows();
|
||||
|
||||
typedef internal::gemm_blocking_space<(Rhs::Flags&RowMajorBit) ? RowMajor : ColMajor,Scalar,Scalar,
|
||||
Rhs::MaxRowsAtCompileTime, Rhs::MaxColsAtCompileTime, Lhs::MaxRowsAtCompileTime,4> BlockingType;
|
||||
|
||||
BlockingType blocking(rhs.rows(), rhs.cols(), size, 1, false);
|
||||
|
||||
triangular_solve_matrix<Scalar,Index,Side,Mode,LhsProductTraits::NeedToConjugate,(int(Lhs::Flags) & RowMajorBit) ? RowMajor : ColMajor,
|
||||
(Rhs::Flags&RowMajorBit) ? RowMajor : ColMajor, Rhs::InnerStrideAtCompileTime>
|
||||
::run(size, othersize, &actualLhs.coeffRef(0,0), actualLhs.outerStride(), &rhs.coeffRef(0,0), rhs.innerStride(), rhs.outerStride(), blocking);
|
||||
}
|
||||
};
|
||||
|
||||
/***************************************************************************
|
||||
* meta-unrolling implementation
|
||||
***************************************************************************/
|
||||
|
||||
template<typename Lhs, typename Rhs, int Mode, int LoopIndex, int Size,
|
||||
bool Stop = LoopIndex==Size>
|
||||
struct triangular_solver_unroller;
|
||||
|
||||
template<typename Lhs, typename Rhs, int Mode, int LoopIndex, int Size>
|
||||
struct triangular_solver_unroller<Lhs,Rhs,Mode,LoopIndex,Size,false> {
|
||||
enum {
|
||||
IsLower = ((Mode&Lower)==Lower),
|
||||
DiagIndex = IsLower ? LoopIndex : Size - LoopIndex - 1,
|
||||
StartIndex = IsLower ? 0 : DiagIndex+1
|
||||
};
|
||||
static EIGEN_DEVICE_FUNC void run(const Lhs& lhs, Rhs& rhs)
|
||||
{
|
||||
if (LoopIndex>0)
|
||||
rhs.coeffRef(DiagIndex) -= lhs.row(DiagIndex).template segment<LoopIndex>(StartIndex).transpose()
|
||||
.cwiseProduct(rhs.template segment<LoopIndex>(StartIndex)).sum();
|
||||
|
||||
if(!(Mode & UnitDiag))
|
||||
rhs.coeffRef(DiagIndex) /= lhs.coeff(DiagIndex,DiagIndex);
|
||||
|
||||
triangular_solver_unroller<Lhs,Rhs,Mode,LoopIndex+1,Size>::run(lhs,rhs);
|
||||
}
|
||||
};
|
||||
|
||||
template<typename Lhs, typename Rhs, int Mode, int LoopIndex, int Size>
|
||||
struct triangular_solver_unroller<Lhs,Rhs,Mode,LoopIndex,Size,true> {
|
||||
static EIGEN_DEVICE_FUNC void run(const Lhs&, Rhs&) {}
|
||||
};
|
||||
|
||||
template<typename Lhs, typename Rhs, int Mode>
|
||||
struct triangular_solver_selector<Lhs,Rhs,OnTheLeft,Mode,CompleteUnrolling,1> {
|
||||
static EIGEN_DEVICE_FUNC 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 EIGEN_DEVICE_FUNC void run(const Lhs& lhs, Rhs& rhs)
|
||||
{
|
||||
Transpose<const Lhs> trLhs(lhs);
|
||||
Transpose<Rhs> trRhs(rhs);
|
||||
|
||||
triangular_solver_unroller<Transpose<const Lhs>,Transpose<Rhs>,
|
||||
((Mode&Upper)==Upper ? Lower : Upper) | (Mode&UnitDiag),
|
||||
0,Rhs::SizeAtCompileTime>::run(trLhs,trRhs);
|
||||
}
|
||||
};
|
||||
|
||||
} // end namespace internal
|
||||
|
||||
/***************************************************************************
|
||||
* TriangularView methods
|
||||
***************************************************************************/
|
||||
|
||||
#ifndef EIGEN_PARSED_BY_DOXYGEN
|
||||
template<typename MatrixType, unsigned int Mode>
|
||||
template<int Side, typename OtherDerived>
|
||||
EIGEN_DEVICE_FUNC void TriangularViewImpl<MatrixType,Mode,Dense>::solveInPlace(const MatrixBase<OtherDerived>& _other) const
|
||||
{
|
||||
OtherDerived& other = _other.const_cast_derived();
|
||||
eigen_assert( derived().cols() == derived().rows() && ((Side==OnTheLeft && derived().cols() == other.rows()) || (Side==OnTheRight && derived().cols() == other.cols())) );
|
||||
eigen_assert((!(int(Mode) & int(ZeroDiag))) && bool(int(Mode) & (int(Upper) | int(Lower))));
|
||||
// If solving for a 0x0 matrix, nothing to do, simply return.
|
||||
if (derived().cols() == 0)
|
||||
return;
|
||||
|
||||
enum { copy = (internal::traits<OtherDerived>::Flags & RowMajorBit) && OtherDerived::IsVectorAtCompileTime && OtherDerived::SizeAtCompileTime!=1};
|
||||
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(derived().nestedExpression(), otherCopy);
|
||||
|
||||
if (copy)
|
||||
other = otherCopy;
|
||||
}
|
||||
|
||||
template<typename Derived, unsigned int Mode>
|
||||
template<int Side, typename Other>
|
||||
const internal::triangular_solve_retval<Side,TriangularView<Derived,Mode>,Other>
|
||||
TriangularViewImpl<Derived,Mode,Dense>::solve(const MatrixBase<Other>& other) const
|
||||
{
|
||||
return internal::triangular_solve_retval<Side,TriangularViewType,Other>(derived(), other.derived());
|
||||
}
|
||||
#endif
|
||||
|
||||
namespace internal {
|
||||
|
||||
|
||||
template<int Side, typename TriangularType, typename Rhs>
|
||||
struct traits<triangular_solve_retval<Side, TriangularType, Rhs> >
|
||||
{
|
||||
typedef typename internal::plain_matrix_type_column_major<Rhs>::type ReturnType;
|
||||
};
|
||||
|
||||
template<int Side, typename TriangularType, typename Rhs> struct triangular_solve_retval
|
||||
: public ReturnByValue<triangular_solve_retval<Side, TriangularType, Rhs> >
|
||||
{
|
||||
typedef typename remove_all<typename Rhs::Nested>::type RhsNestedCleaned;
|
||||
typedef ReturnByValue<triangular_solve_retval> Base;
|
||||
|
||||
triangular_solve_retval(const TriangularType& tri, const Rhs& rhs)
|
||||
: m_triangularMatrix(tri), m_rhs(rhs)
|
||||
{}
|
||||
|
||||
inline EIGEN_CONSTEXPR Index rows() const EIGEN_NOEXCEPT { return m_rhs.rows(); }
|
||||
inline EIGEN_CONSTEXPR Index cols() const EIGEN_NOEXCEPT { return m_rhs.cols(); }
|
||||
|
||||
template<typename Dest> inline void evalTo(Dest& dst) const
|
||||
{
|
||||
if(!is_same_dense(dst,m_rhs))
|
||||
dst = m_rhs;
|
||||
m_triangularMatrix.template solveInPlace<Side>(dst);
|
||||
}
|
||||
|
||||
protected:
|
||||
const TriangularType& m_triangularMatrix;
|
||||
typename Rhs::Nested m_rhs;
|
||||
};
|
||||
|
||||
} // namespace internal
|
||||
|
||||
} // end namespace Eigen
|
||||
|
||||
#endif // EIGEN_SOLVETRIANGULAR_H
|
||||
@@ -0,0 +1,168 @@
|
||||
// This file is part of Eigen, a lightweight C++ template library
|
||||
// for linear algebra.
|
||||
//
|
||||
// Copyright (C) 2015 Gael Guennebaud <gael.guennebaud@inria.fr>
|
||||
//
|
||||
// This Source Code Form is subject to the terms of the Mozilla
|
||||
// Public License v. 2.0. If a copy of the MPL was not distributed
|
||||
// with this file, You can obtain one at http://mozilla.org/MPL/2.0/.
|
||||
|
||||
#ifndef EIGEN_SOLVERBASE_H
|
||||
#define EIGEN_SOLVERBASE_H
|
||||
|
||||
namespace Eigen {
|
||||
|
||||
namespace internal {
|
||||
|
||||
template<typename Derived>
|
||||
struct solve_assertion {
|
||||
template<bool Transpose_, typename Rhs>
|
||||
static void run(const Derived& solver, const Rhs& b) { solver.template _check_solve_assertion<Transpose_>(b); }
|
||||
};
|
||||
|
||||
template<typename Derived>
|
||||
struct solve_assertion<Transpose<Derived> >
|
||||
{
|
||||
typedef Transpose<Derived> type;
|
||||
|
||||
template<bool Transpose_, typename Rhs>
|
||||
static void run(const type& transpose, const Rhs& b)
|
||||
{
|
||||
internal::solve_assertion<typename internal::remove_all<Derived>::type>::template run<true>(transpose.nestedExpression(), b);
|
||||
}
|
||||
};
|
||||
|
||||
template<typename Scalar, typename Derived>
|
||||
struct solve_assertion<CwiseUnaryOp<Eigen::internal::scalar_conjugate_op<Scalar>, const Transpose<Derived> > >
|
||||
{
|
||||
typedef CwiseUnaryOp<Eigen::internal::scalar_conjugate_op<Scalar>, const Transpose<Derived> > type;
|
||||
|
||||
template<bool Transpose_, typename Rhs>
|
||||
static void run(const type& adjoint, const Rhs& b)
|
||||
{
|
||||
internal::solve_assertion<typename internal::remove_all<Transpose<Derived> >::type>::template run<true>(adjoint.nestedExpression(), b);
|
||||
}
|
||||
};
|
||||
} // end namespace internal
|
||||
|
||||
/** \class SolverBase
|
||||
* \brief A base class for matrix decomposition and solvers
|
||||
*
|
||||
* \tparam Derived the actual type of the decomposition/solver.
|
||||
*
|
||||
* Any matrix decomposition inheriting this base class provide the following API:
|
||||
*
|
||||
* \code
|
||||
* MatrixType A, b, x;
|
||||
* DecompositionType dec(A);
|
||||
* x = dec.solve(b); // solve A * x = b
|
||||
* x = dec.transpose().solve(b); // solve A^T * x = b
|
||||
* x = dec.adjoint().solve(b); // solve A' * x = b
|
||||
* \endcode
|
||||
*
|
||||
* \warning Currently, any other usage of transpose() and adjoint() are not supported and will produce compilation errors.
|
||||
*
|
||||
* \sa class PartialPivLU, class FullPivLU, class HouseholderQR, class ColPivHouseholderQR, class FullPivHouseholderQR, class CompleteOrthogonalDecomposition, class LLT, class LDLT, class SVDBase
|
||||
*/
|
||||
template<typename Derived>
|
||||
class SolverBase : public EigenBase<Derived>
|
||||
{
|
||||
public:
|
||||
|
||||
typedef EigenBase<Derived> Base;
|
||||
typedef typename internal::traits<Derived>::Scalar Scalar;
|
||||
typedef Scalar CoeffReturnType;
|
||||
|
||||
template<typename Derived_>
|
||||
friend struct internal::solve_assertion;
|
||||
|
||||
enum {
|
||||
RowsAtCompileTime = internal::traits<Derived>::RowsAtCompileTime,
|
||||
ColsAtCompileTime = internal::traits<Derived>::ColsAtCompileTime,
|
||||
SizeAtCompileTime = (internal::size_at_compile_time<internal::traits<Derived>::RowsAtCompileTime,
|
||||
internal::traits<Derived>::ColsAtCompileTime>::ret),
|
||||
MaxRowsAtCompileTime = internal::traits<Derived>::MaxRowsAtCompileTime,
|
||||
MaxColsAtCompileTime = internal::traits<Derived>::MaxColsAtCompileTime,
|
||||
MaxSizeAtCompileTime = (internal::size_at_compile_time<internal::traits<Derived>::MaxRowsAtCompileTime,
|
||||
internal::traits<Derived>::MaxColsAtCompileTime>::ret),
|
||||
IsVectorAtCompileTime = internal::traits<Derived>::MaxRowsAtCompileTime == 1
|
||||
|| internal::traits<Derived>::MaxColsAtCompileTime == 1,
|
||||
NumDimensions = int(MaxSizeAtCompileTime) == 1 ? 0 : bool(IsVectorAtCompileTime) ? 1 : 2
|
||||
};
|
||||
|
||||
/** Default constructor */
|
||||
SolverBase()
|
||||
{}
|
||||
|
||||
~SolverBase()
|
||||
{}
|
||||
|
||||
using Base::derived;
|
||||
|
||||
/** \returns an expression of the solution x of \f$ A x = b \f$ using the current decomposition of A.
|
||||
*/
|
||||
template<typename Rhs>
|
||||
inline const Solve<Derived, Rhs>
|
||||
solve(const MatrixBase<Rhs>& b) const
|
||||
{
|
||||
internal::solve_assertion<typename internal::remove_all<Derived>::type>::template run<false>(derived(), b);
|
||||
return Solve<Derived, Rhs>(derived(), b.derived());
|
||||
}
|
||||
|
||||
/** \internal the return type of transpose() */
|
||||
typedef typename internal::add_const<Transpose<const Derived> >::type ConstTransposeReturnType;
|
||||
/** \returns an expression of the transposed of the factored matrix.
|
||||
*
|
||||
* A typical usage is to solve for the transposed problem A^T x = b:
|
||||
* \code x = dec.transpose().solve(b); \endcode
|
||||
*
|
||||
* \sa adjoint(), solve()
|
||||
*/
|
||||
inline ConstTransposeReturnType transpose() const
|
||||
{
|
||||
return ConstTransposeReturnType(derived());
|
||||
}
|
||||
|
||||
/** \internal the return type of adjoint() */
|
||||
typedef typename internal::conditional<NumTraits<Scalar>::IsComplex,
|
||||
CwiseUnaryOp<internal::scalar_conjugate_op<Scalar>, ConstTransposeReturnType>,
|
||||
ConstTransposeReturnType
|
||||
>::type AdjointReturnType;
|
||||
/** \returns an expression of the adjoint of the factored matrix
|
||||
*
|
||||
* A typical usage is to solve for the adjoint problem A' x = b:
|
||||
* \code x = dec.adjoint().solve(b); \endcode
|
||||
*
|
||||
* For real scalar types, this function is equivalent to transpose().
|
||||
*
|
||||
* \sa transpose(), solve()
|
||||
*/
|
||||
inline AdjointReturnType adjoint() const
|
||||
{
|
||||
return AdjointReturnType(derived().transpose());
|
||||
}
|
||||
|
||||
protected:
|
||||
|
||||
template<bool Transpose_, typename Rhs>
|
||||
void _check_solve_assertion(const Rhs& b) const {
|
||||
EIGEN_ONLY_USED_FOR_DEBUG(b);
|
||||
eigen_assert(derived().m_isInitialized && "Solver is not initialized.");
|
||||
eigen_assert((Transpose_?derived().cols():derived().rows())==b.rows() && "SolverBase::solve(): invalid number of rows of the right hand side matrix b");
|
||||
}
|
||||
};
|
||||
|
||||
namespace internal {
|
||||
|
||||
template<typename Derived>
|
||||
struct generic_xpr_base<Derived, MatrixXpr, SolverStorage>
|
||||
{
|
||||
typedef SolverBase<Derived> type;
|
||||
|
||||
};
|
||||
|
||||
} // end namespace internal
|
||||
|
||||
} // end namespace Eigen
|
||||
|
||||
#endif // EIGEN_SOLVERBASE_H
|
||||
@@ -0,0 +1,251 @@
|
||||
// This file is part of Eigen, a lightweight C++ template library
|
||||
// for linear algebra.
|
||||
//
|
||||
// Copyright (C) 2009 Gael Guennebaud <gael.guennebaud@inria.fr>
|
||||
//
|
||||
// This Source Code Form is subject to the terms of the Mozilla
|
||||
// Public License v. 2.0. If a copy of the MPL was not distributed
|
||||
// with this file, You can obtain one at http://mozilla.org/MPL/2.0/.
|
||||
|
||||
#ifndef EIGEN_STABLENORM_H
|
||||
#define EIGEN_STABLENORM_H
|
||||
|
||||
namespace Eigen {
|
||||
|
||||
namespace internal {
|
||||
|
||||
template<typename ExpressionType, typename Scalar>
|
||||
inline void stable_norm_kernel(const ExpressionType& bl, Scalar& ssq, Scalar& scale, Scalar& invScale)
|
||||
{
|
||||
Scalar maxCoeff = bl.cwiseAbs().maxCoeff();
|
||||
|
||||
if(maxCoeff>scale)
|
||||
{
|
||||
ssq = ssq * numext::abs2(scale/maxCoeff);
|
||||
Scalar tmp = Scalar(1)/maxCoeff;
|
||||
if(tmp > NumTraits<Scalar>::highest())
|
||||
{
|
||||
invScale = NumTraits<Scalar>::highest();
|
||||
scale = Scalar(1)/invScale;
|
||||
}
|
||||
else if(maxCoeff>NumTraits<Scalar>::highest()) // we got a INF
|
||||
{
|
||||
invScale = Scalar(1);
|
||||
scale = maxCoeff;
|
||||
}
|
||||
else
|
||||
{
|
||||
scale = maxCoeff;
|
||||
invScale = tmp;
|
||||
}
|
||||
}
|
||||
else if(maxCoeff!=maxCoeff) // we got a NaN
|
||||
{
|
||||
scale = maxCoeff;
|
||||
}
|
||||
|
||||
// TODO if the maxCoeff is much much smaller than the current scale,
|
||||
// then we can neglect this sub vector
|
||||
if(scale>Scalar(0)) // if scale==0, then bl is 0
|
||||
ssq += (bl*invScale).squaredNorm();
|
||||
}
|
||||
|
||||
template<typename VectorType, typename RealScalar>
|
||||
void stable_norm_impl_inner_step(const VectorType &vec, RealScalar& ssq, RealScalar& scale, RealScalar& invScale)
|
||||
{
|
||||
typedef typename VectorType::Scalar Scalar;
|
||||
const Index blockSize = 4096;
|
||||
|
||||
typedef typename internal::nested_eval<VectorType,2>::type VectorTypeCopy;
|
||||
typedef typename internal::remove_all<VectorTypeCopy>::type VectorTypeCopyClean;
|
||||
const VectorTypeCopy copy(vec);
|
||||
|
||||
enum {
|
||||
CanAlign = ( (int(VectorTypeCopyClean::Flags)&DirectAccessBit)
|
||||
|| (int(internal::evaluator<VectorTypeCopyClean>::Alignment)>0) // FIXME Alignment)>0 might not be enough
|
||||
) && (blockSize*sizeof(Scalar)*2<EIGEN_STACK_ALLOCATION_LIMIT)
|
||||
&& (EIGEN_MAX_STATIC_ALIGN_BYTES>0) // if we cannot allocate on the stack, then let's not bother about this optimization
|
||||
};
|
||||
typedef typename internal::conditional<CanAlign, Ref<const Matrix<Scalar,Dynamic,1,0,blockSize,1>, internal::evaluator<VectorTypeCopyClean>::Alignment>,
|
||||
typename VectorTypeCopyClean::ConstSegmentReturnType>::type SegmentWrapper;
|
||||
Index n = vec.size();
|
||||
|
||||
Index bi = internal::first_default_aligned(copy);
|
||||
if (bi>0)
|
||||
internal::stable_norm_kernel(copy.head(bi), ssq, scale, invScale);
|
||||
for (; bi<n; bi+=blockSize)
|
||||
internal::stable_norm_kernel(SegmentWrapper(copy.segment(bi,numext::mini(blockSize, n - bi))), ssq, scale, invScale);
|
||||
}
|
||||
|
||||
template<typename VectorType>
|
||||
typename VectorType::RealScalar
|
||||
stable_norm_impl(const VectorType &vec, typename enable_if<VectorType::IsVectorAtCompileTime>::type* = 0 )
|
||||
{
|
||||
using std::sqrt;
|
||||
using std::abs;
|
||||
|
||||
Index n = vec.size();
|
||||
|
||||
if(n==1)
|
||||
return abs(vec.coeff(0));
|
||||
|
||||
typedef typename VectorType::RealScalar RealScalar;
|
||||
RealScalar scale(0);
|
||||
RealScalar invScale(1);
|
||||
RealScalar ssq(0); // sum of squares
|
||||
|
||||
stable_norm_impl_inner_step(vec, ssq, scale, invScale);
|
||||
|
||||
return scale * sqrt(ssq);
|
||||
}
|
||||
|
||||
template<typename MatrixType>
|
||||
typename MatrixType::RealScalar
|
||||
stable_norm_impl(const MatrixType &mat, typename enable_if<!MatrixType::IsVectorAtCompileTime>::type* = 0 )
|
||||
{
|
||||
using std::sqrt;
|
||||
|
||||
typedef typename MatrixType::RealScalar RealScalar;
|
||||
RealScalar scale(0);
|
||||
RealScalar invScale(1);
|
||||
RealScalar ssq(0); // sum of squares
|
||||
|
||||
for(Index j=0; j<mat.outerSize(); ++j)
|
||||
stable_norm_impl_inner_step(mat.innerVector(j), ssq, scale, invScale);
|
||||
return scale * sqrt(ssq);
|
||||
}
|
||||
|
||||
template<typename Derived>
|
||||
inline typename NumTraits<typename traits<Derived>::Scalar>::Real
|
||||
blueNorm_impl(const EigenBase<Derived>& _vec)
|
||||
{
|
||||
typedef typename Derived::RealScalar RealScalar;
|
||||
using std::pow;
|
||||
using std::sqrt;
|
||||
using std::abs;
|
||||
|
||||
// This program calculates the machine-dependent constants
|
||||
// bl, b2, slm, s2m, relerr overfl
|
||||
// from the "basic" machine-dependent numbers
|
||||
// nbig, ibeta, it, iemin, iemax, rbig.
|
||||
// The following define the basic machine-dependent constants.
|
||||
// For portability, the PORT subprograms "ilmaeh" and "rlmach"
|
||||
// are used. For any specific computer, each of the assignment
|
||||
// statements can be replaced
|
||||
static const int ibeta = std::numeric_limits<RealScalar>::radix; // base for floating-point numbers
|
||||
static const int it = NumTraits<RealScalar>::digits(); // number of base-beta digits in mantissa
|
||||
static const int iemin = NumTraits<RealScalar>::min_exponent(); // minimum exponent
|
||||
static const int iemax = NumTraits<RealScalar>::max_exponent(); // maximum exponent
|
||||
static const RealScalar rbig = NumTraits<RealScalar>::highest(); // largest floating-point number
|
||||
static const RealScalar b1 = RealScalar(pow(RealScalar(ibeta),RealScalar(-((1-iemin)/2)))); // lower boundary of midrange
|
||||
static const RealScalar b2 = RealScalar(pow(RealScalar(ibeta),RealScalar((iemax + 1 - it)/2))); // upper boundary of midrange
|
||||
static const RealScalar s1m = RealScalar(pow(RealScalar(ibeta),RealScalar((2-iemin)/2))); // scaling factor for lower range
|
||||
static const RealScalar s2m = RealScalar(pow(RealScalar(ibeta),RealScalar(- ((iemax+it)/2)))); // scaling factor for upper range
|
||||
static const RealScalar eps = RealScalar(pow(double(ibeta), 1-it));
|
||||
static const RealScalar relerr = sqrt(eps); // tolerance for neglecting asml
|
||||
|
||||
const Derived& vec(_vec.derived());
|
||||
Index n = vec.size();
|
||||
RealScalar ab2 = b2 / RealScalar(n);
|
||||
RealScalar asml = RealScalar(0);
|
||||
RealScalar amed = RealScalar(0);
|
||||
RealScalar abig = RealScalar(0);
|
||||
|
||||
for(Index j=0; j<vec.outerSize(); ++j)
|
||||
{
|
||||
for(typename Derived::InnerIterator iter(vec, j); iter; ++iter)
|
||||
{
|
||||
RealScalar ax = abs(iter.value());
|
||||
if(ax > ab2) abig += numext::abs2(ax*s2m);
|
||||
else if(ax < b1) asml += numext::abs2(ax*s1m);
|
||||
else amed += numext::abs2(ax);
|
||||
}
|
||||
}
|
||||
if(amed!=amed)
|
||||
return amed; // we got a NaN
|
||||
if(abig > RealScalar(0))
|
||||
{
|
||||
abig = sqrt(abig);
|
||||
if(abig > rbig) // overflow, or *this contains INF values
|
||||
return abig; // return INF
|
||||
if(amed > RealScalar(0))
|
||||
{
|
||||
abig = abig/s2m;
|
||||
amed = sqrt(amed);
|
||||
}
|
||||
else
|
||||
return abig/s2m;
|
||||
}
|
||||
else if(asml > RealScalar(0))
|
||||
{
|
||||
if (amed > RealScalar(0))
|
||||
{
|
||||
abig = sqrt(amed);
|
||||
amed = sqrt(asml) / s1m;
|
||||
}
|
||||
else
|
||||
return sqrt(asml)/s1m;
|
||||
}
|
||||
else
|
||||
return sqrt(amed);
|
||||
asml = numext::mini(abig, amed);
|
||||
abig = numext::maxi(abig, amed);
|
||||
if(asml <= abig*relerr)
|
||||
return abig;
|
||||
else
|
||||
return abig * sqrt(RealScalar(1) + numext::abs2(asml/abig));
|
||||
}
|
||||
|
||||
} // end namespace internal
|
||||
|
||||
/** \returns the \em l2 norm of \c *this avoiding underflow and overflow.
|
||||
* This version use a blockwise two passes algorithm:
|
||||
* 1 - find the absolute largest coefficient \c s
|
||||
* 2 - compute \f$ s \Vert \frac{*this}{s} \Vert \f$ in a standard way
|
||||
*
|
||||
* For architecture/scalar types supporting vectorization, this version
|
||||
* is faster than blueNorm(). Otherwise the blueNorm() is much faster.
|
||||
*
|
||||
* \sa norm(), blueNorm(), hypotNorm()
|
||||
*/
|
||||
template<typename Derived>
|
||||
inline typename NumTraits<typename internal::traits<Derived>::Scalar>::Real
|
||||
MatrixBase<Derived>::stableNorm() const
|
||||
{
|
||||
return internal::stable_norm_impl(derived());
|
||||
}
|
||||
|
||||
/** \returns the \em l2 norm of \c *this using the Blue's algorithm.
|
||||
* A Portable Fortran Program to Find the Euclidean Norm of a Vector,
|
||||
* ACM TOMS, Vol 4, Issue 1, 1978.
|
||||
*
|
||||
* For architecture/scalar types without vectorization, this version
|
||||
* is much faster than stableNorm(). Otherwise the stableNorm() is faster.
|
||||
*
|
||||
* \sa norm(), stableNorm(), hypotNorm()
|
||||
*/
|
||||
template<typename Derived>
|
||||
inline typename NumTraits<typename internal::traits<Derived>::Scalar>::Real
|
||||
MatrixBase<Derived>::blueNorm() const
|
||||
{
|
||||
return internal::blueNorm_impl(*this);
|
||||
}
|
||||
|
||||
/** \returns the \em l2 norm of \c *this avoiding undeflow and overflow.
|
||||
* This version use a concatenation of hypot() calls, and it is very slow.
|
||||
*
|
||||
* \sa norm(), stableNorm()
|
||||
*/
|
||||
template<typename Derived>
|
||||
inline typename NumTraits<typename internal::traits<Derived>::Scalar>::Real
|
||||
MatrixBase<Derived>::hypotNorm() const
|
||||
{
|
||||
if(size()==1)
|
||||
return numext::abs(coeff(0,0));
|
||||
else
|
||||
return this->cwiseAbs().redux(internal::scalar_hypot_op<RealScalar>());
|
||||
}
|
||||
|
||||
} // end namespace Eigen
|
||||
|
||||
#endif // EIGEN_STABLENORM_H
|
||||
@@ -0,0 +1,463 @@
|
||||
// This file is part of Eigen, a lightweight C++ template library
|
||||
// for linear algebra.
|
||||
//
|
||||
// Copyright (C) 2018 Gael Guennebaud <gael.guennebaud@inria.fr>
|
||||
//
|
||||
// This Source Code Form is subject to the terms of the Mozilla
|
||||
// Public License v. 2.0. If a copy of the MPL was not distributed
|
||||
// with this file, You can obtain one at http://mozilla.org/MPL/2.0/.
|
||||
|
||||
#ifndef EIGEN_STLITERATORS_H
|
||||
#define EIGEN_STLITERATORS_H
|
||||
|
||||
namespace Eigen {
|
||||
|
||||
namespace internal {
|
||||
|
||||
template<typename IteratorType>
|
||||
struct indexed_based_stl_iterator_traits;
|
||||
|
||||
template<typename Derived>
|
||||
class indexed_based_stl_iterator_base
|
||||
{
|
||||
protected:
|
||||
typedef indexed_based_stl_iterator_traits<Derived> traits;
|
||||
typedef typename traits::XprType XprType;
|
||||
typedef indexed_based_stl_iterator_base<typename traits::non_const_iterator> non_const_iterator;
|
||||
typedef indexed_based_stl_iterator_base<typename traits::const_iterator> const_iterator;
|
||||
typedef typename internal::conditional<internal::is_const<XprType>::value,non_const_iterator,const_iterator>::type other_iterator;
|
||||
// NOTE: in C++03 we cannot declare friend classes through typedefs because we need to write friend class:
|
||||
friend class indexed_based_stl_iterator_base<typename traits::const_iterator>;
|
||||
friend class indexed_based_stl_iterator_base<typename traits::non_const_iterator>;
|
||||
public:
|
||||
typedef Index difference_type;
|
||||
typedef std::random_access_iterator_tag iterator_category;
|
||||
|
||||
indexed_based_stl_iterator_base() EIGEN_NO_THROW : mp_xpr(0), m_index(0) {}
|
||||
indexed_based_stl_iterator_base(XprType& xpr, Index index) EIGEN_NO_THROW : mp_xpr(&xpr), m_index(index) {}
|
||||
|
||||
indexed_based_stl_iterator_base(const non_const_iterator& other) EIGEN_NO_THROW
|
||||
: mp_xpr(other.mp_xpr), m_index(other.m_index)
|
||||
{}
|
||||
|
||||
indexed_based_stl_iterator_base& operator=(const non_const_iterator& other)
|
||||
{
|
||||
mp_xpr = other.mp_xpr;
|
||||
m_index = other.m_index;
|
||||
return *this;
|
||||
}
|
||||
|
||||
Derived& operator++() { ++m_index; return derived(); }
|
||||
Derived& operator--() { --m_index; return derived(); }
|
||||
|
||||
Derived operator++(int) { Derived prev(derived()); operator++(); return prev;}
|
||||
Derived operator--(int) { Derived prev(derived()); operator--(); return prev;}
|
||||
|
||||
friend Derived operator+(const indexed_based_stl_iterator_base& a, Index b) { Derived ret(a.derived()); ret += b; return ret; }
|
||||
friend Derived operator-(const indexed_based_stl_iterator_base& a, Index b) { Derived ret(a.derived()); ret -= b; return ret; }
|
||||
friend Derived operator+(Index a, const indexed_based_stl_iterator_base& b) { Derived ret(b.derived()); ret += a; return ret; }
|
||||
friend Derived operator-(Index a, const indexed_based_stl_iterator_base& b) { Derived ret(b.derived()); ret -= a; return ret; }
|
||||
|
||||
Derived& operator+=(Index b) { m_index += b; return derived(); }
|
||||
Derived& operator-=(Index b) { m_index -= b; return derived(); }
|
||||
|
||||
difference_type operator-(const indexed_based_stl_iterator_base& other) const
|
||||
{
|
||||
eigen_assert(mp_xpr == other.mp_xpr);
|
||||
return m_index - other.m_index;
|
||||
}
|
||||
|
||||
difference_type operator-(const other_iterator& other) const
|
||||
{
|
||||
eigen_assert(mp_xpr == other.mp_xpr);
|
||||
return m_index - other.m_index;
|
||||
}
|
||||
|
||||
bool operator==(const indexed_based_stl_iterator_base& other) const { eigen_assert(mp_xpr == other.mp_xpr); return m_index == other.m_index; }
|
||||
bool operator!=(const indexed_based_stl_iterator_base& other) const { eigen_assert(mp_xpr == other.mp_xpr); return m_index != other.m_index; }
|
||||
bool operator< (const indexed_based_stl_iterator_base& other) const { eigen_assert(mp_xpr == other.mp_xpr); return m_index < other.m_index; }
|
||||
bool operator<=(const indexed_based_stl_iterator_base& other) const { eigen_assert(mp_xpr == other.mp_xpr); return m_index <= other.m_index; }
|
||||
bool operator> (const indexed_based_stl_iterator_base& other) const { eigen_assert(mp_xpr == other.mp_xpr); return m_index > other.m_index; }
|
||||
bool operator>=(const indexed_based_stl_iterator_base& other) const { eigen_assert(mp_xpr == other.mp_xpr); return m_index >= other.m_index; }
|
||||
|
||||
bool operator==(const other_iterator& other) const { eigen_assert(mp_xpr == other.mp_xpr); return m_index == other.m_index; }
|
||||
bool operator!=(const other_iterator& other) const { eigen_assert(mp_xpr == other.mp_xpr); return m_index != other.m_index; }
|
||||
bool operator< (const other_iterator& other) const { eigen_assert(mp_xpr == other.mp_xpr); return m_index < other.m_index; }
|
||||
bool operator<=(const other_iterator& other) const { eigen_assert(mp_xpr == other.mp_xpr); return m_index <= other.m_index; }
|
||||
bool operator> (const other_iterator& other) const { eigen_assert(mp_xpr == other.mp_xpr); return m_index > other.m_index; }
|
||||
bool operator>=(const other_iterator& other) const { eigen_assert(mp_xpr == other.mp_xpr); return m_index >= other.m_index; }
|
||||
|
||||
protected:
|
||||
|
||||
Derived& derived() { return static_cast<Derived&>(*this); }
|
||||
const Derived& derived() const { return static_cast<const Derived&>(*this); }
|
||||
|
||||
XprType *mp_xpr;
|
||||
Index m_index;
|
||||
};
|
||||
|
||||
template<typename Derived>
|
||||
class indexed_based_stl_reverse_iterator_base
|
||||
{
|
||||
protected:
|
||||
typedef indexed_based_stl_iterator_traits<Derived> traits;
|
||||
typedef typename traits::XprType XprType;
|
||||
typedef indexed_based_stl_reverse_iterator_base<typename traits::non_const_iterator> non_const_iterator;
|
||||
typedef indexed_based_stl_reverse_iterator_base<typename traits::const_iterator> const_iterator;
|
||||
typedef typename internal::conditional<internal::is_const<XprType>::value,non_const_iterator,const_iterator>::type other_iterator;
|
||||
// NOTE: in C++03 we cannot declare friend classes through typedefs because we need to write friend class:
|
||||
friend class indexed_based_stl_reverse_iterator_base<typename traits::const_iterator>;
|
||||
friend class indexed_based_stl_reverse_iterator_base<typename traits::non_const_iterator>;
|
||||
public:
|
||||
typedef Index difference_type;
|
||||
typedef std::random_access_iterator_tag iterator_category;
|
||||
|
||||
indexed_based_stl_reverse_iterator_base() : mp_xpr(0), m_index(0) {}
|
||||
indexed_based_stl_reverse_iterator_base(XprType& xpr, Index index) : mp_xpr(&xpr), m_index(index) {}
|
||||
|
||||
indexed_based_stl_reverse_iterator_base(const non_const_iterator& other)
|
||||
: mp_xpr(other.mp_xpr), m_index(other.m_index)
|
||||
{}
|
||||
|
||||
indexed_based_stl_reverse_iterator_base& operator=(const non_const_iterator& other)
|
||||
{
|
||||
mp_xpr = other.mp_xpr;
|
||||
m_index = other.m_index;
|
||||
return *this;
|
||||
}
|
||||
|
||||
Derived& operator++() { --m_index; return derived(); }
|
||||
Derived& operator--() { ++m_index; return derived(); }
|
||||
|
||||
Derived operator++(int) { Derived prev(derived()); operator++(); return prev;}
|
||||
Derived operator--(int) { Derived prev(derived()); operator--(); return prev;}
|
||||
|
||||
friend Derived operator+(const indexed_based_stl_reverse_iterator_base& a, Index b) { Derived ret(a.derived()); ret += b; return ret; }
|
||||
friend Derived operator-(const indexed_based_stl_reverse_iterator_base& a, Index b) { Derived ret(a.derived()); ret -= b; return ret; }
|
||||
friend Derived operator+(Index a, const indexed_based_stl_reverse_iterator_base& b) { Derived ret(b.derived()); ret += a; return ret; }
|
||||
friend Derived operator-(Index a, const indexed_based_stl_reverse_iterator_base& b) { Derived ret(b.derived()); ret -= a; return ret; }
|
||||
|
||||
Derived& operator+=(Index b) { m_index -= b; return derived(); }
|
||||
Derived& operator-=(Index b) { m_index += b; return derived(); }
|
||||
|
||||
difference_type operator-(const indexed_based_stl_reverse_iterator_base& other) const
|
||||
{
|
||||
eigen_assert(mp_xpr == other.mp_xpr);
|
||||
return other.m_index - m_index;
|
||||
}
|
||||
|
||||
difference_type operator-(const other_iterator& other) const
|
||||
{
|
||||
eigen_assert(mp_xpr == other.mp_xpr);
|
||||
return other.m_index - m_index;
|
||||
}
|
||||
|
||||
bool operator==(const indexed_based_stl_reverse_iterator_base& other) const { eigen_assert(mp_xpr == other.mp_xpr); return m_index == other.m_index; }
|
||||
bool operator!=(const indexed_based_stl_reverse_iterator_base& other) const { eigen_assert(mp_xpr == other.mp_xpr); return m_index != other.m_index; }
|
||||
bool operator< (const indexed_based_stl_reverse_iterator_base& other) const { eigen_assert(mp_xpr == other.mp_xpr); return m_index > other.m_index; }
|
||||
bool operator<=(const indexed_based_stl_reverse_iterator_base& other) const { eigen_assert(mp_xpr == other.mp_xpr); return m_index >= other.m_index; }
|
||||
bool operator> (const indexed_based_stl_reverse_iterator_base& other) const { eigen_assert(mp_xpr == other.mp_xpr); return m_index < other.m_index; }
|
||||
bool operator>=(const indexed_based_stl_reverse_iterator_base& other) const { eigen_assert(mp_xpr == other.mp_xpr); return m_index <= other.m_index; }
|
||||
|
||||
bool operator==(const other_iterator& other) const { eigen_assert(mp_xpr == other.mp_xpr); return m_index == other.m_index; }
|
||||
bool operator!=(const other_iterator& other) const { eigen_assert(mp_xpr == other.mp_xpr); return m_index != other.m_index; }
|
||||
bool operator< (const other_iterator& other) const { eigen_assert(mp_xpr == other.mp_xpr); return m_index > other.m_index; }
|
||||
bool operator<=(const other_iterator& other) const { eigen_assert(mp_xpr == other.mp_xpr); return m_index >= other.m_index; }
|
||||
bool operator> (const other_iterator& other) const { eigen_assert(mp_xpr == other.mp_xpr); return m_index < other.m_index; }
|
||||
bool operator>=(const other_iterator& other) const { eigen_assert(mp_xpr == other.mp_xpr); return m_index <= other.m_index; }
|
||||
|
||||
protected:
|
||||
|
||||
Derived& derived() { return static_cast<Derived&>(*this); }
|
||||
const Derived& derived() const { return static_cast<const Derived&>(*this); }
|
||||
|
||||
XprType *mp_xpr;
|
||||
Index m_index;
|
||||
};
|
||||
|
||||
template<typename XprType>
|
||||
class pointer_based_stl_iterator
|
||||
{
|
||||
enum { is_lvalue = internal::is_lvalue<XprType>::value };
|
||||
typedef pointer_based_stl_iterator<typename internal::remove_const<XprType>::type> non_const_iterator;
|
||||
typedef pointer_based_stl_iterator<typename internal::add_const<XprType>::type> const_iterator;
|
||||
typedef typename internal::conditional<internal::is_const<XprType>::value,non_const_iterator,const_iterator>::type other_iterator;
|
||||
// NOTE: in C++03 we cannot declare friend classes through typedefs because we need to write friend class:
|
||||
friend class pointer_based_stl_iterator<typename internal::add_const<XprType>::type>;
|
||||
friend class pointer_based_stl_iterator<typename internal::remove_const<XprType>::type>;
|
||||
public:
|
||||
typedef Index difference_type;
|
||||
typedef typename XprType::Scalar value_type;
|
||||
typedef std::random_access_iterator_tag iterator_category;
|
||||
typedef typename internal::conditional<bool(is_lvalue), value_type*, const value_type*>::type pointer;
|
||||
typedef typename internal::conditional<bool(is_lvalue), value_type&, const value_type&>::type reference;
|
||||
|
||||
|
||||
pointer_based_stl_iterator() EIGEN_NO_THROW : m_ptr(0) {}
|
||||
pointer_based_stl_iterator(XprType& xpr, Index index) EIGEN_NO_THROW : m_incr(xpr.innerStride())
|
||||
{
|
||||
m_ptr = xpr.data() + index * m_incr.value();
|
||||
}
|
||||
|
||||
pointer_based_stl_iterator(const non_const_iterator& other) EIGEN_NO_THROW
|
||||
: m_ptr(other.m_ptr), m_incr(other.m_incr)
|
||||
{}
|
||||
|
||||
pointer_based_stl_iterator& operator=(const non_const_iterator& other) EIGEN_NO_THROW
|
||||
{
|
||||
m_ptr = other.m_ptr;
|
||||
m_incr.setValue(other.m_incr);
|
||||
return *this;
|
||||
}
|
||||
|
||||
reference operator*() const { return *m_ptr; }
|
||||
reference operator[](Index i) const { return *(m_ptr+i*m_incr.value()); }
|
||||
pointer operator->() const { return m_ptr; }
|
||||
|
||||
pointer_based_stl_iterator& operator++() { m_ptr += m_incr.value(); return *this; }
|
||||
pointer_based_stl_iterator& operator--() { m_ptr -= m_incr.value(); return *this; }
|
||||
|
||||
pointer_based_stl_iterator operator++(int) { pointer_based_stl_iterator prev(*this); operator++(); return prev;}
|
||||
pointer_based_stl_iterator operator--(int) { pointer_based_stl_iterator prev(*this); operator--(); return prev;}
|
||||
|
||||
friend pointer_based_stl_iterator operator+(const pointer_based_stl_iterator& a, Index b) { pointer_based_stl_iterator ret(a); ret += b; return ret; }
|
||||
friend pointer_based_stl_iterator operator-(const pointer_based_stl_iterator& a, Index b) { pointer_based_stl_iterator ret(a); ret -= b; return ret; }
|
||||
friend pointer_based_stl_iterator operator+(Index a, const pointer_based_stl_iterator& b) { pointer_based_stl_iterator ret(b); ret += a; return ret; }
|
||||
friend pointer_based_stl_iterator operator-(Index a, const pointer_based_stl_iterator& b) { pointer_based_stl_iterator ret(b); ret -= a; return ret; }
|
||||
|
||||
pointer_based_stl_iterator& operator+=(Index b) { m_ptr += b*m_incr.value(); return *this; }
|
||||
pointer_based_stl_iterator& operator-=(Index b) { m_ptr -= b*m_incr.value(); return *this; }
|
||||
|
||||
difference_type operator-(const pointer_based_stl_iterator& other) const {
|
||||
return (m_ptr - other.m_ptr)/m_incr.value();
|
||||
}
|
||||
|
||||
difference_type operator-(const other_iterator& other) const {
|
||||
return (m_ptr - other.m_ptr)/m_incr.value();
|
||||
}
|
||||
|
||||
bool operator==(const pointer_based_stl_iterator& other) const { return m_ptr == other.m_ptr; }
|
||||
bool operator!=(const pointer_based_stl_iterator& other) const { return m_ptr != other.m_ptr; }
|
||||
bool operator< (const pointer_based_stl_iterator& other) const { return m_ptr < other.m_ptr; }
|
||||
bool operator<=(const pointer_based_stl_iterator& other) const { return m_ptr <= other.m_ptr; }
|
||||
bool operator> (const pointer_based_stl_iterator& other) const { return m_ptr > other.m_ptr; }
|
||||
bool operator>=(const pointer_based_stl_iterator& other) const { return m_ptr >= other.m_ptr; }
|
||||
|
||||
bool operator==(const other_iterator& other) const { return m_ptr == other.m_ptr; }
|
||||
bool operator!=(const other_iterator& other) const { return m_ptr != other.m_ptr; }
|
||||
bool operator< (const other_iterator& other) const { return m_ptr < other.m_ptr; }
|
||||
bool operator<=(const other_iterator& other) const { return m_ptr <= other.m_ptr; }
|
||||
bool operator> (const other_iterator& other) const { return m_ptr > other.m_ptr; }
|
||||
bool operator>=(const other_iterator& other) const { return m_ptr >= other.m_ptr; }
|
||||
|
||||
protected:
|
||||
|
||||
pointer m_ptr;
|
||||
internal::variable_if_dynamic<Index, XprType::InnerStrideAtCompileTime> m_incr;
|
||||
};
|
||||
|
||||
template<typename _XprType>
|
||||
struct indexed_based_stl_iterator_traits<generic_randaccess_stl_iterator<_XprType> >
|
||||
{
|
||||
typedef _XprType XprType;
|
||||
typedef generic_randaccess_stl_iterator<typename internal::remove_const<XprType>::type> non_const_iterator;
|
||||
typedef generic_randaccess_stl_iterator<typename internal::add_const<XprType>::type> const_iterator;
|
||||
};
|
||||
|
||||
template<typename XprType>
|
||||
class generic_randaccess_stl_iterator : public indexed_based_stl_iterator_base<generic_randaccess_stl_iterator<XprType> >
|
||||
{
|
||||
public:
|
||||
typedef typename XprType::Scalar value_type;
|
||||
|
||||
protected:
|
||||
|
||||
enum {
|
||||
has_direct_access = (internal::traits<XprType>::Flags & DirectAccessBit) ? 1 : 0,
|
||||
is_lvalue = internal::is_lvalue<XprType>::value
|
||||
};
|
||||
|
||||
typedef indexed_based_stl_iterator_base<generic_randaccess_stl_iterator> Base;
|
||||
using Base::m_index;
|
||||
using Base::mp_xpr;
|
||||
|
||||
// TODO currently const Transpose/Reshape expressions never returns const references,
|
||||
// so lets return by value too.
|
||||
//typedef typename internal::conditional<bool(has_direct_access), const value_type&, const value_type>::type read_only_ref_t;
|
||||
typedef const value_type read_only_ref_t;
|
||||
|
||||
public:
|
||||
|
||||
typedef typename internal::conditional<bool(is_lvalue), value_type *, const value_type *>::type pointer;
|
||||
typedef typename internal::conditional<bool(is_lvalue), value_type&, read_only_ref_t>::type reference;
|
||||
|
||||
generic_randaccess_stl_iterator() : Base() {}
|
||||
generic_randaccess_stl_iterator(XprType& xpr, Index index) : Base(xpr,index) {}
|
||||
generic_randaccess_stl_iterator(const typename Base::non_const_iterator& other) : Base(other) {}
|
||||
using Base::operator=;
|
||||
|
||||
reference operator*() const { return (*mp_xpr)(m_index); }
|
||||
reference operator[](Index i) const { return (*mp_xpr)(m_index+i); }
|
||||
pointer operator->() const { return &((*mp_xpr)(m_index)); }
|
||||
};
|
||||
|
||||
template<typename _XprType, DirectionType Direction>
|
||||
struct indexed_based_stl_iterator_traits<subvector_stl_iterator<_XprType,Direction> >
|
||||
{
|
||||
typedef _XprType XprType;
|
||||
typedef subvector_stl_iterator<typename internal::remove_const<XprType>::type, Direction> non_const_iterator;
|
||||
typedef subvector_stl_iterator<typename internal::add_const<XprType>::type, Direction> const_iterator;
|
||||
};
|
||||
|
||||
template<typename XprType, DirectionType Direction>
|
||||
class subvector_stl_iterator : public indexed_based_stl_iterator_base<subvector_stl_iterator<XprType,Direction> >
|
||||
{
|
||||
protected:
|
||||
|
||||
enum { is_lvalue = internal::is_lvalue<XprType>::value };
|
||||
|
||||
typedef indexed_based_stl_iterator_base<subvector_stl_iterator> Base;
|
||||
using Base::m_index;
|
||||
using Base::mp_xpr;
|
||||
|
||||
typedef typename internal::conditional<Direction==Vertical,typename XprType::ColXpr,typename XprType::RowXpr>::type SubVectorType;
|
||||
typedef typename internal::conditional<Direction==Vertical,typename XprType::ConstColXpr,typename XprType::ConstRowXpr>::type ConstSubVectorType;
|
||||
|
||||
|
||||
public:
|
||||
typedef typename internal::conditional<bool(is_lvalue), SubVectorType, ConstSubVectorType>::type reference;
|
||||
typedef typename reference::PlainObject value_type;
|
||||
|
||||
private:
|
||||
class subvector_stl_iterator_ptr
|
||||
{
|
||||
public:
|
||||
subvector_stl_iterator_ptr(const reference &subvector) : m_subvector(subvector) {}
|
||||
reference* operator->() { return &m_subvector; }
|
||||
private:
|
||||
reference m_subvector;
|
||||
};
|
||||
public:
|
||||
|
||||
typedef subvector_stl_iterator_ptr pointer;
|
||||
|
||||
subvector_stl_iterator() : Base() {}
|
||||
subvector_stl_iterator(XprType& xpr, Index index) : Base(xpr,index) {}
|
||||
|
||||
reference operator*() const { return (*mp_xpr).template subVector<Direction>(m_index); }
|
||||
reference operator[](Index i) const { return (*mp_xpr).template subVector<Direction>(m_index+i); }
|
||||
pointer operator->() const { return (*mp_xpr).template subVector<Direction>(m_index); }
|
||||
};
|
||||
|
||||
template<typename _XprType, DirectionType Direction>
|
||||
struct indexed_based_stl_iterator_traits<subvector_stl_reverse_iterator<_XprType,Direction> >
|
||||
{
|
||||
typedef _XprType XprType;
|
||||
typedef subvector_stl_reverse_iterator<typename internal::remove_const<XprType>::type, Direction> non_const_iterator;
|
||||
typedef subvector_stl_reverse_iterator<typename internal::add_const<XprType>::type, Direction> const_iterator;
|
||||
};
|
||||
|
||||
template<typename XprType, DirectionType Direction>
|
||||
class subvector_stl_reverse_iterator : public indexed_based_stl_reverse_iterator_base<subvector_stl_reverse_iterator<XprType,Direction> >
|
||||
{
|
||||
protected:
|
||||
|
||||
enum { is_lvalue = internal::is_lvalue<XprType>::value };
|
||||
|
||||
typedef indexed_based_stl_reverse_iterator_base<subvector_stl_reverse_iterator> Base;
|
||||
using Base::m_index;
|
||||
using Base::mp_xpr;
|
||||
|
||||
typedef typename internal::conditional<Direction==Vertical,typename XprType::ColXpr,typename XprType::RowXpr>::type SubVectorType;
|
||||
typedef typename internal::conditional<Direction==Vertical,typename XprType::ConstColXpr,typename XprType::ConstRowXpr>::type ConstSubVectorType;
|
||||
|
||||
|
||||
public:
|
||||
typedef typename internal::conditional<bool(is_lvalue), SubVectorType, ConstSubVectorType>::type reference;
|
||||
typedef typename reference::PlainObject value_type;
|
||||
|
||||
private:
|
||||
class subvector_stl_reverse_iterator_ptr
|
||||
{
|
||||
public:
|
||||
subvector_stl_reverse_iterator_ptr(const reference &subvector) : m_subvector(subvector) {}
|
||||
reference* operator->() { return &m_subvector; }
|
||||
private:
|
||||
reference m_subvector;
|
||||
};
|
||||
public:
|
||||
|
||||
typedef subvector_stl_reverse_iterator_ptr pointer;
|
||||
|
||||
subvector_stl_reverse_iterator() : Base() {}
|
||||
subvector_stl_reverse_iterator(XprType& xpr, Index index) : Base(xpr,index) {}
|
||||
|
||||
reference operator*() const { return (*mp_xpr).template subVector<Direction>(m_index); }
|
||||
reference operator[](Index i) const { return (*mp_xpr).template subVector<Direction>(m_index+i); }
|
||||
pointer operator->() const { return (*mp_xpr).template subVector<Direction>(m_index); }
|
||||
};
|
||||
|
||||
} // namespace internal
|
||||
|
||||
|
||||
/** returns an iterator to the first element of the 1D vector or array
|
||||
* \only_for_vectors
|
||||
* \sa end(), cbegin()
|
||||
*/
|
||||
template<typename Derived>
|
||||
inline typename DenseBase<Derived>::iterator DenseBase<Derived>::begin()
|
||||
{
|
||||
EIGEN_STATIC_ASSERT_VECTOR_ONLY(Derived);
|
||||
return iterator(derived(), 0);
|
||||
}
|
||||
|
||||
/** const version of begin() */
|
||||
template<typename Derived>
|
||||
inline typename DenseBase<Derived>::const_iterator DenseBase<Derived>::begin() const
|
||||
{
|
||||
return cbegin();
|
||||
}
|
||||
|
||||
/** returns a read-only const_iterator to the first element of the 1D vector or array
|
||||
* \only_for_vectors
|
||||
* \sa cend(), begin()
|
||||
*/
|
||||
template<typename Derived>
|
||||
inline typename DenseBase<Derived>::const_iterator DenseBase<Derived>::cbegin() const
|
||||
{
|
||||
EIGEN_STATIC_ASSERT_VECTOR_ONLY(Derived);
|
||||
return const_iterator(derived(), 0);
|
||||
}
|
||||
|
||||
/** returns an iterator to the element following the last element of the 1D vector or array
|
||||
* \only_for_vectors
|
||||
* \sa begin(), cend()
|
||||
*/
|
||||
template<typename Derived>
|
||||
inline typename DenseBase<Derived>::iterator DenseBase<Derived>::end()
|
||||
{
|
||||
EIGEN_STATIC_ASSERT_VECTOR_ONLY(Derived);
|
||||
return iterator(derived(), size());
|
||||
}
|
||||
|
||||
/** const version of end() */
|
||||
template<typename Derived>
|
||||
inline typename DenseBase<Derived>::const_iterator DenseBase<Derived>::end() const
|
||||
{
|
||||
return cend();
|
||||
}
|
||||
|
||||
/** returns a read-only const_iterator to the element following the last element of the 1D vector or array
|
||||
* \only_for_vectors
|
||||
* \sa begin(), cend()
|
||||
*/
|
||||
template<typename Derived>
|
||||
inline typename DenseBase<Derived>::const_iterator DenseBase<Derived>::cend() const
|
||||
{
|
||||
EIGEN_STATIC_ASSERT_VECTOR_ONLY(Derived);
|
||||
return const_iterator(derived(), size());
|
||||
}
|
||||
|
||||
} // namespace Eigen
|
||||
|
||||
#endif // EIGEN_STLITERATORS_H
|
||||
@@ -0,0 +1,116 @@
|
||||
// This file is part of Eigen, a lightweight C++ template library
|
||||
// for linear algebra.
|
||||
//
|
||||
// Copyright (C) 2010 Benoit Jacob <jacob.benoit.1@gmail.com>
|
||||
//
|
||||
// This Source Code Form is subject to the terms of the Mozilla
|
||||
// Public License v. 2.0. If a copy of the MPL was not distributed
|
||||
// with this file, You can obtain one at http://mozilla.org/MPL/2.0/.
|
||||
|
||||
#ifndef EIGEN_STRIDE_H
|
||||
#define EIGEN_STRIDE_H
|
||||
|
||||
namespace Eigen {
|
||||
|
||||
/** \class Stride
|
||||
* \ingroup Core_Module
|
||||
*
|
||||
* \brief Holds strides information for Map
|
||||
*
|
||||
* This class holds the strides information for mapping arrays with strides with class Map.
|
||||
*
|
||||
* It holds two values: the inner stride and the outer stride.
|
||||
*
|
||||
* The inner stride is the pointer increment between two consecutive entries within a given row of a
|
||||
* row-major matrix or within a given column of a column-major matrix.
|
||||
*
|
||||
* The outer stride is the pointer increment between two consecutive rows of a row-major matrix or
|
||||
* between two consecutive columns of a column-major matrix.
|
||||
*
|
||||
* These two values can be passed either at compile-time as template parameters, or at runtime as
|
||||
* arguments to the constructor.
|
||||
*
|
||||
* Indeed, this class takes two template parameters:
|
||||
* \tparam _OuterStrideAtCompileTime the outer stride, or Dynamic if you want to specify it at runtime.
|
||||
* \tparam _InnerStrideAtCompileTime the inner stride, or Dynamic if you want to specify it at runtime.
|
||||
*
|
||||
* Here is an example:
|
||||
* \include Map_general_stride.cpp
|
||||
* Output: \verbinclude Map_general_stride.out
|
||||
*
|
||||
* Both strides can be negative, however, a negative stride of -1 cannot be specified at compiletime
|
||||
* because of the ambiguity with Dynamic which is defined to -1 (historically, negative strides were
|
||||
* not allowed).
|
||||
*
|
||||
* \sa class InnerStride, class OuterStride, \ref TopicStorageOrders
|
||||
*/
|
||||
template<int _OuterStrideAtCompileTime, int _InnerStrideAtCompileTime>
|
||||
class Stride
|
||||
{
|
||||
public:
|
||||
typedef Eigen::Index Index; ///< \deprecated since Eigen 3.3
|
||||
enum {
|
||||
InnerStrideAtCompileTime = _InnerStrideAtCompileTime,
|
||||
OuterStrideAtCompileTime = _OuterStrideAtCompileTime
|
||||
};
|
||||
|
||||
/** Default constructor, for use when strides are fixed at compile time */
|
||||
EIGEN_DEVICE_FUNC
|
||||
Stride()
|
||||
: m_outer(OuterStrideAtCompileTime), m_inner(InnerStrideAtCompileTime)
|
||||
{
|
||||
// FIXME: for Eigen 4 we should use DynamicIndex instead of Dynamic.
|
||||
// FIXME: for Eigen 4 we should also unify this API with fix<>
|
||||
eigen_assert(InnerStrideAtCompileTime != Dynamic && OuterStrideAtCompileTime != Dynamic);
|
||||
}
|
||||
|
||||
/** Constructor allowing to pass the strides at runtime */
|
||||
EIGEN_DEVICE_FUNC
|
||||
Stride(Index outerStride, Index innerStride)
|
||||
: m_outer(outerStride), m_inner(innerStride)
|
||||
{
|
||||
}
|
||||
|
||||
/** Copy constructor */
|
||||
EIGEN_DEVICE_FUNC
|
||||
Stride(const Stride& other)
|
||||
: m_outer(other.outer()), m_inner(other.inner())
|
||||
{}
|
||||
|
||||
/** \returns the outer stride */
|
||||
EIGEN_DEVICE_FUNC EIGEN_CONSTEXPR
|
||||
inline Index outer() const { return m_outer.value(); }
|
||||
/** \returns the inner stride */
|
||||
EIGEN_DEVICE_FUNC EIGEN_CONSTEXPR
|
||||
inline Index inner() const { return m_inner.value(); }
|
||||
|
||||
protected:
|
||||
internal::variable_if_dynamic<Index, OuterStrideAtCompileTime> m_outer;
|
||||
internal::variable_if_dynamic<Index, InnerStrideAtCompileTime> m_inner;
|
||||
};
|
||||
|
||||
/** \brief Convenience specialization of Stride to specify only an inner stride
|
||||
* See class Map for some examples */
|
||||
template<int Value>
|
||||
class InnerStride : public Stride<0, Value>
|
||||
{
|
||||
typedef Stride<0, Value> Base;
|
||||
public:
|
||||
EIGEN_DEVICE_FUNC InnerStride() : Base() {}
|
||||
EIGEN_DEVICE_FUNC InnerStride(Index v) : Base(0, v) {} // FIXME making this explicit could break valid code
|
||||
};
|
||||
|
||||
/** \brief Convenience specialization of Stride to specify only an outer stride
|
||||
* See class Map for some examples */
|
||||
template<int Value>
|
||||
class OuterStride : public Stride<Value, 0>
|
||||
{
|
||||
typedef Stride<Value, 0> Base;
|
||||
public:
|
||||
EIGEN_DEVICE_FUNC OuterStride() : Base() {}
|
||||
EIGEN_DEVICE_FUNC OuterStride(Index v) : Base(v,0) {} // FIXME making this explicit could break valid code
|
||||
};
|
||||
|
||||
} // end namespace Eigen
|
||||
|
||||
#endif // EIGEN_STRIDE_H
|
||||
@@ -0,0 +1,68 @@
|
||||
// This file is part of Eigen, a lightweight C++ template library
|
||||
// for linear algebra.
|
||||
//
|
||||
// Copyright (C) 2006-2008 Benoit Jacob <jacob.benoit.1@gmail.com>
|
||||
//
|
||||
// This Source Code Form is subject to the terms of the Mozilla
|
||||
// Public License v. 2.0. If a copy of the MPL was not distributed
|
||||
// with this file, You can obtain one at http://mozilla.org/MPL/2.0/.
|
||||
|
||||
#ifndef EIGEN_SWAP_H
|
||||
#define EIGEN_SWAP_H
|
||||
|
||||
namespace Eigen {
|
||||
|
||||
namespace internal {
|
||||
|
||||
// Overload default assignPacket behavior for swapping them
|
||||
template<typename DstEvaluatorTypeT, typename SrcEvaluatorTypeT>
|
||||
class generic_dense_assignment_kernel<DstEvaluatorTypeT, SrcEvaluatorTypeT, swap_assign_op<typename DstEvaluatorTypeT::Scalar>, Specialized>
|
||||
: public generic_dense_assignment_kernel<DstEvaluatorTypeT, SrcEvaluatorTypeT, swap_assign_op<typename DstEvaluatorTypeT::Scalar>, BuiltIn>
|
||||
{
|
||||
protected:
|
||||
typedef generic_dense_assignment_kernel<DstEvaluatorTypeT, SrcEvaluatorTypeT, swap_assign_op<typename DstEvaluatorTypeT::Scalar>, BuiltIn> Base;
|
||||
using Base::m_dst;
|
||||
using Base::m_src;
|
||||
using Base::m_functor;
|
||||
|
||||
public:
|
||||
typedef typename Base::Scalar Scalar;
|
||||
typedef typename Base::DstXprType DstXprType;
|
||||
typedef swap_assign_op<Scalar> Functor;
|
||||
|
||||
EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
|
||||
generic_dense_assignment_kernel(DstEvaluatorTypeT &dst, const SrcEvaluatorTypeT &src, const Functor &func, DstXprType& dstExpr)
|
||||
: Base(dst, src, func, dstExpr)
|
||||
{}
|
||||
|
||||
template<int StoreMode, int LoadMode, typename PacketType>
|
||||
EIGEN_STRONG_INLINE void assignPacket(Index row, Index col)
|
||||
{
|
||||
PacketType tmp = m_src.template packet<LoadMode,PacketType>(row,col);
|
||||
const_cast<SrcEvaluatorTypeT&>(m_src).template writePacket<LoadMode>(row,col, m_dst.template packet<StoreMode,PacketType>(row,col));
|
||||
m_dst.template writePacket<StoreMode>(row,col,tmp);
|
||||
}
|
||||
|
||||
template<int StoreMode, int LoadMode, typename PacketType>
|
||||
EIGEN_STRONG_INLINE void assignPacket(Index index)
|
||||
{
|
||||
PacketType tmp = m_src.template packet<LoadMode,PacketType>(index);
|
||||
const_cast<SrcEvaluatorTypeT&>(m_src).template writePacket<LoadMode>(index, m_dst.template packet<StoreMode,PacketType>(index));
|
||||
m_dst.template writePacket<StoreMode>(index,tmp);
|
||||
}
|
||||
|
||||
// TODO find a simple way not to have to copy/paste this function from generic_dense_assignment_kernel, by simple I mean no CRTP (Gael)
|
||||
template<int StoreMode, int LoadMode, typename PacketType>
|
||||
EIGEN_STRONG_INLINE void assignPacketByOuterInner(Index outer, Index inner)
|
||||
{
|
||||
Index row = Base::rowIndexByOuterInner(outer, inner);
|
||||
Index col = Base::colIndexByOuterInner(outer, inner);
|
||||
assignPacket<StoreMode,LoadMode,PacketType>(row, col);
|
||||
}
|
||||
};
|
||||
|
||||
} // namespace internal
|
||||
|
||||
} // end namespace Eigen
|
||||
|
||||
#endif // EIGEN_SWAP_H
|
||||
@@ -0,0 +1,464 @@
|
||||
// This file is part of Eigen, a lightweight C++ template library
|
||||
// for linear algebra.
|
||||
//
|
||||
// Copyright (C) 2006-2008 Benoit Jacob <jacob.benoit.1@gmail.com>
|
||||
// Copyright (C) 2009-2014 Gael Guennebaud <gael.guennebaud@inria.fr>
|
||||
//
|
||||
// This Source Code Form is subject to the terms of the Mozilla
|
||||
// Public License v. 2.0. If a copy of the MPL was not distributed
|
||||
// with this file, You can obtain one at http://mozilla.org/MPL/2.0/.
|
||||
|
||||
#ifndef EIGEN_TRANSPOSE_H
|
||||
#define EIGEN_TRANSPOSE_H
|
||||
|
||||
namespace Eigen {
|
||||
|
||||
namespace internal {
|
||||
template<typename MatrixType>
|
||||
struct traits<Transpose<MatrixType> > : public traits<MatrixType>
|
||||
{
|
||||
typedef typename ref_selector<MatrixType>::type MatrixTypeNested;
|
||||
typedef typename remove_reference<MatrixTypeNested>::type MatrixTypeNestedPlain;
|
||||
enum {
|
||||
RowsAtCompileTime = MatrixType::ColsAtCompileTime,
|
||||
ColsAtCompileTime = MatrixType::RowsAtCompileTime,
|
||||
MaxRowsAtCompileTime = MatrixType::MaxColsAtCompileTime,
|
||||
MaxColsAtCompileTime = MatrixType::MaxRowsAtCompileTime,
|
||||
FlagsLvalueBit = is_lvalue<MatrixType>::value ? LvalueBit : 0,
|
||||
Flags0 = traits<MatrixTypeNestedPlain>::Flags & ~(LvalueBit | NestByRefBit),
|
||||
Flags1 = Flags0 | FlagsLvalueBit,
|
||||
Flags = Flags1 ^ RowMajorBit,
|
||||
InnerStrideAtCompileTime = inner_stride_at_compile_time<MatrixType>::ret,
|
||||
OuterStrideAtCompileTime = outer_stride_at_compile_time<MatrixType>::ret
|
||||
};
|
||||
};
|
||||
}
|
||||
|
||||
template<typename MatrixType, typename StorageKind> class TransposeImpl;
|
||||
|
||||
/** \class Transpose
|
||||
* \ingroup Core_Module
|
||||
*
|
||||
* \brief Expression of the transpose of a matrix
|
||||
*
|
||||
* \tparam MatrixType the type of the object of which we are taking the transpose
|
||||
*
|
||||
* This class represents an expression of the transpose of a matrix.
|
||||
* It is the return type of MatrixBase::transpose() and MatrixBase::adjoint()
|
||||
* and most of the time this is the only way it is used.
|
||||
*
|
||||
* \sa MatrixBase::transpose(), MatrixBase::adjoint()
|
||||
*/
|
||||
template<typename MatrixType> class Transpose
|
||||
: public TransposeImpl<MatrixType,typename internal::traits<MatrixType>::StorageKind>
|
||||
{
|
||||
public:
|
||||
|
||||
typedef typename internal::ref_selector<MatrixType>::non_const_type MatrixTypeNested;
|
||||
|
||||
typedef typename TransposeImpl<MatrixType,typename internal::traits<MatrixType>::StorageKind>::Base Base;
|
||||
EIGEN_GENERIC_PUBLIC_INTERFACE(Transpose)
|
||||
typedef typename internal::remove_all<MatrixType>::type NestedExpression;
|
||||
|
||||
EIGEN_DEVICE_FUNC
|
||||
explicit EIGEN_STRONG_INLINE Transpose(MatrixType& matrix) : m_matrix(matrix) {}
|
||||
|
||||
EIGEN_INHERIT_ASSIGNMENT_OPERATORS(Transpose)
|
||||
|
||||
EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE EIGEN_CONSTEXPR
|
||||
Index rows() const EIGEN_NOEXCEPT { return m_matrix.cols(); }
|
||||
EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE EIGEN_CONSTEXPR
|
||||
Index cols() const EIGEN_NOEXCEPT { return m_matrix.rows(); }
|
||||
|
||||
/** \returns the nested expression */
|
||||
EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
|
||||
const typename internal::remove_all<MatrixTypeNested>::type&
|
||||
nestedExpression() const { return m_matrix; }
|
||||
|
||||
/** \returns the nested expression */
|
||||
EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
|
||||
typename internal::remove_reference<MatrixTypeNested>::type&
|
||||
nestedExpression() { return m_matrix; }
|
||||
|
||||
/** \internal */
|
||||
EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
|
||||
void resize(Index nrows, Index ncols) {
|
||||
m_matrix.resize(ncols,nrows);
|
||||
}
|
||||
|
||||
protected:
|
||||
typename internal::ref_selector<MatrixType>::non_const_type m_matrix;
|
||||
};
|
||||
|
||||
namespace internal {
|
||||
|
||||
template<typename MatrixType, bool HasDirectAccess = has_direct_access<MatrixType>::ret>
|
||||
struct TransposeImpl_base
|
||||
{
|
||||
typedef typename dense_xpr_base<Transpose<MatrixType> >::type type;
|
||||
};
|
||||
|
||||
template<typename MatrixType>
|
||||
struct TransposeImpl_base<MatrixType, false>
|
||||
{
|
||||
typedef typename dense_xpr_base<Transpose<MatrixType> >::type type;
|
||||
};
|
||||
|
||||
} // end namespace internal
|
||||
|
||||
// Generic API dispatcher
|
||||
template<typename XprType, typename StorageKind>
|
||||
class TransposeImpl
|
||||
: public internal::generic_xpr_base<Transpose<XprType> >::type
|
||||
{
|
||||
public:
|
||||
typedef typename internal::generic_xpr_base<Transpose<XprType> >::type Base;
|
||||
};
|
||||
|
||||
template<typename MatrixType> class TransposeImpl<MatrixType,Dense>
|
||||
: public internal::TransposeImpl_base<MatrixType>::type
|
||||
{
|
||||
public:
|
||||
|
||||
typedef typename internal::TransposeImpl_base<MatrixType>::type Base;
|
||||
using Base::coeffRef;
|
||||
EIGEN_DENSE_PUBLIC_INTERFACE(Transpose<MatrixType>)
|
||||
EIGEN_INHERIT_ASSIGNMENT_OPERATORS(TransposeImpl)
|
||||
|
||||
EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
|
||||
Index innerStride() const { return derived().nestedExpression().innerStride(); }
|
||||
EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
|
||||
Index outerStride() const { return derived().nestedExpression().outerStride(); }
|
||||
|
||||
typedef typename internal::conditional<
|
||||
internal::is_lvalue<MatrixType>::value,
|
||||
Scalar,
|
||||
const Scalar
|
||||
>::type ScalarWithConstIfNotLvalue;
|
||||
|
||||
EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
|
||||
ScalarWithConstIfNotLvalue* data() { return derived().nestedExpression().data(); }
|
||||
EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
|
||||
const Scalar* data() const { return derived().nestedExpression().data(); }
|
||||
|
||||
// FIXME: shall we keep the const version of coeffRef?
|
||||
EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
|
||||
const Scalar& coeffRef(Index rowId, Index colId) const
|
||||
{
|
||||
return derived().nestedExpression().coeffRef(colId, rowId);
|
||||
}
|
||||
|
||||
EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
|
||||
const Scalar& coeffRef(Index index) const
|
||||
{
|
||||
return derived().nestedExpression().coeffRef(index);
|
||||
}
|
||||
protected:
|
||||
EIGEN_DEFAULT_EMPTY_CONSTRUCTOR_AND_DESTRUCTOR(TransposeImpl)
|
||||
};
|
||||
|
||||
/** \returns an expression of the transpose of *this.
|
||||
*
|
||||
* Example: \include MatrixBase_transpose.cpp
|
||||
* Output: \verbinclude MatrixBase_transpose.out
|
||||
*
|
||||
* \warning If you want to replace a matrix by its own transpose, do \b NOT do this:
|
||||
* \code
|
||||
* m = m.transpose(); // bug!!! caused by aliasing effect
|
||||
* \endcode
|
||||
* Instead, use the transposeInPlace() method:
|
||||
* \code
|
||||
* m.transposeInPlace();
|
||||
* \endcode
|
||||
* which gives Eigen good opportunities for optimization, or alternatively you can also do:
|
||||
* \code
|
||||
* m = m.transpose().eval();
|
||||
* \endcode
|
||||
*
|
||||
* \sa transposeInPlace(), adjoint() */
|
||||
template<typename Derived>
|
||||
EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
|
||||
Transpose<Derived>
|
||||
DenseBase<Derived>::transpose()
|
||||
{
|
||||
return TransposeReturnType(derived());
|
||||
}
|
||||
|
||||
/** This is the const version of transpose().
|
||||
*
|
||||
* Make sure you read the warning for transpose() !
|
||||
*
|
||||
* \sa transposeInPlace(), adjoint() */
|
||||
template<typename Derived>
|
||||
EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
|
||||
typename DenseBase<Derived>::ConstTransposeReturnType
|
||||
DenseBase<Derived>::transpose() const
|
||||
{
|
||||
return ConstTransposeReturnType(derived());
|
||||
}
|
||||
|
||||
/** \returns an expression of the adjoint (i.e. conjugate transpose) of *this.
|
||||
*
|
||||
* Example: \include MatrixBase_adjoint.cpp
|
||||
* Output: \verbinclude MatrixBase_adjoint.out
|
||||
*
|
||||
* \warning If you want to replace a matrix by its own adjoint, do \b NOT do this:
|
||||
* \code
|
||||
* m = m.adjoint(); // bug!!! caused by aliasing effect
|
||||
* \endcode
|
||||
* Instead, use the adjointInPlace() method:
|
||||
* \code
|
||||
* m.adjointInPlace();
|
||||
* \endcode
|
||||
* which gives Eigen good opportunities for optimization, or alternatively you can also do:
|
||||
* \code
|
||||
* m = m.adjoint().eval();
|
||||
* \endcode
|
||||
*
|
||||
* \sa adjointInPlace(), transpose(), conjugate(), class Transpose, class internal::scalar_conjugate_op */
|
||||
template<typename Derived>
|
||||
EIGEN_DEVICE_FUNC inline const typename MatrixBase<Derived>::AdjointReturnType
|
||||
MatrixBase<Derived>::adjoint() const
|
||||
{
|
||||
return AdjointReturnType(this->transpose());
|
||||
}
|
||||
|
||||
/***************************************************************************
|
||||
* "in place" transpose implementation
|
||||
***************************************************************************/
|
||||
|
||||
namespace internal {
|
||||
|
||||
template<typename MatrixType,
|
||||
bool IsSquare = (MatrixType::RowsAtCompileTime == MatrixType::ColsAtCompileTime) && MatrixType::RowsAtCompileTime!=Dynamic,
|
||||
bool MatchPacketSize =
|
||||
(int(MatrixType::RowsAtCompileTime) == int(internal::packet_traits<typename MatrixType::Scalar>::size))
|
||||
&& (internal::evaluator<MatrixType>::Flags&PacketAccessBit) >
|
||||
struct inplace_transpose_selector;
|
||||
|
||||
template<typename MatrixType>
|
||||
struct inplace_transpose_selector<MatrixType,true,false> { // square matrix
|
||||
static void run(MatrixType& m) {
|
||||
m.matrix().template triangularView<StrictlyUpper>().swap(m.matrix().transpose().template triangularView<StrictlyUpper>());
|
||||
}
|
||||
};
|
||||
|
||||
template<typename MatrixType>
|
||||
struct inplace_transpose_selector<MatrixType,true,true> { // PacketSize x PacketSize
|
||||
static void run(MatrixType& m) {
|
||||
typedef typename MatrixType::Scalar Scalar;
|
||||
typedef typename internal::packet_traits<typename MatrixType::Scalar>::type Packet;
|
||||
const Index PacketSize = internal::packet_traits<Scalar>::size;
|
||||
const Index Alignment = internal::evaluator<MatrixType>::Alignment;
|
||||
PacketBlock<Packet> A;
|
||||
for (Index i=0; i<PacketSize; ++i)
|
||||
A.packet[i] = m.template packetByOuterInner<Alignment>(i,0);
|
||||
internal::ptranspose(A);
|
||||
for (Index i=0; i<PacketSize; ++i)
|
||||
m.template writePacket<Alignment>(m.rowIndexByOuterInner(i,0), m.colIndexByOuterInner(i,0), A.packet[i]);
|
||||
}
|
||||
};
|
||||
|
||||
|
||||
template <typename MatrixType, Index Alignment>
|
||||
void BlockedInPlaceTranspose(MatrixType& m) {
|
||||
typedef typename MatrixType::Scalar Scalar;
|
||||
typedef typename internal::packet_traits<typename MatrixType::Scalar>::type Packet;
|
||||
const Index PacketSize = internal::packet_traits<Scalar>::size;
|
||||
eigen_assert(m.rows() == m.cols());
|
||||
int row_start = 0;
|
||||
for (; row_start + PacketSize <= m.rows(); row_start += PacketSize) {
|
||||
for (int col_start = row_start; col_start + PacketSize <= m.cols(); col_start += PacketSize) {
|
||||
PacketBlock<Packet> A;
|
||||
if (row_start == col_start) {
|
||||
for (Index i=0; i<PacketSize; ++i)
|
||||
A.packet[i] = m.template packetByOuterInner<Alignment>(row_start + i,col_start);
|
||||
internal::ptranspose(A);
|
||||
for (Index i=0; i<PacketSize; ++i)
|
||||
m.template writePacket<Alignment>(m.rowIndexByOuterInner(row_start + i, col_start), m.colIndexByOuterInner(row_start + i,col_start), A.packet[i]);
|
||||
} else {
|
||||
PacketBlock<Packet> B;
|
||||
for (Index i=0; i<PacketSize; ++i) {
|
||||
A.packet[i] = m.template packetByOuterInner<Alignment>(row_start + i,col_start);
|
||||
B.packet[i] = m.template packetByOuterInner<Alignment>(col_start + i, row_start);
|
||||
}
|
||||
internal::ptranspose(A);
|
||||
internal::ptranspose(B);
|
||||
for (Index i=0; i<PacketSize; ++i) {
|
||||
m.template writePacket<Alignment>(m.rowIndexByOuterInner(row_start + i, col_start), m.colIndexByOuterInner(row_start + i,col_start), B.packet[i]);
|
||||
m.template writePacket<Alignment>(m.rowIndexByOuterInner(col_start + i, row_start), m.colIndexByOuterInner(col_start + i,row_start), A.packet[i]);
|
||||
}
|
||||
}
|
||||
}
|
||||
}
|
||||
for (Index row = row_start; row < m.rows(); ++row) {
|
||||
m.matrix().row(row).head(row).swap(
|
||||
m.matrix().col(row).head(row).transpose());
|
||||
}
|
||||
}
|
||||
|
||||
template<typename MatrixType,bool MatchPacketSize>
|
||||
struct inplace_transpose_selector<MatrixType,false,MatchPacketSize> { // non square or dynamic matrix
|
||||
static void run(MatrixType& m) {
|
||||
typedef typename MatrixType::Scalar Scalar;
|
||||
if (m.rows() == m.cols()) {
|
||||
const Index PacketSize = internal::packet_traits<Scalar>::size;
|
||||
if (!NumTraits<Scalar>::IsComplex && m.rows() >= PacketSize) {
|
||||
if ((m.rows() % PacketSize) == 0)
|
||||
BlockedInPlaceTranspose<MatrixType,internal::evaluator<MatrixType>::Alignment>(m);
|
||||
else
|
||||
BlockedInPlaceTranspose<MatrixType,Unaligned>(m);
|
||||
}
|
||||
else {
|
||||
m.matrix().template triangularView<StrictlyUpper>().swap(m.matrix().transpose().template triangularView<StrictlyUpper>());
|
||||
}
|
||||
} else {
|
||||
m = m.transpose().eval();
|
||||
}
|
||||
}
|
||||
};
|
||||
|
||||
|
||||
} // end namespace internal
|
||||
|
||||
/** This is the "in place" version of transpose(): it replaces \c *this by its own transpose.
|
||||
* Thus, doing
|
||||
* \code
|
||||
* m.transposeInPlace();
|
||||
* \endcode
|
||||
* has the same effect on m as doing
|
||||
* \code
|
||||
* m = m.transpose().eval();
|
||||
* \endcode
|
||||
* and is faster and also safer because in the latter line of code, forgetting the eval() results
|
||||
* in a bug caused by \ref TopicAliasing "aliasing".
|
||||
*
|
||||
* Notice however that this method is only useful if you want to replace a matrix by its own transpose.
|
||||
* If you just need the transpose of a matrix, use transpose().
|
||||
*
|
||||
* \note if the matrix is not square, then \c *this must be a resizable matrix.
|
||||
* This excludes (non-square) fixed-size matrices, block-expressions and maps.
|
||||
*
|
||||
* \sa transpose(), adjoint(), adjointInPlace() */
|
||||
template<typename Derived>
|
||||
EIGEN_DEVICE_FUNC inline void DenseBase<Derived>::transposeInPlace()
|
||||
{
|
||||
eigen_assert((rows() == cols() || (RowsAtCompileTime == Dynamic && ColsAtCompileTime == Dynamic))
|
||||
&& "transposeInPlace() called on a non-square non-resizable matrix");
|
||||
internal::inplace_transpose_selector<Derived>::run(derived());
|
||||
}
|
||||
|
||||
/***************************************************************************
|
||||
* "in place" adjoint implementation
|
||||
***************************************************************************/
|
||||
|
||||
/** This is the "in place" version of adjoint(): it replaces \c *this by its own transpose.
|
||||
* Thus, doing
|
||||
* \code
|
||||
* m.adjointInPlace();
|
||||
* \endcode
|
||||
* has the same effect on m as doing
|
||||
* \code
|
||||
* m = m.adjoint().eval();
|
||||
* \endcode
|
||||
* and is faster and also safer because in the latter line of code, forgetting the eval() results
|
||||
* in a bug caused by aliasing.
|
||||
*
|
||||
* Notice however that this method is only useful if you want to replace a matrix by its own adjoint.
|
||||
* If you just need the adjoint of a matrix, use adjoint().
|
||||
*
|
||||
* \note if the matrix is not square, then \c *this must be a resizable matrix.
|
||||
* This excludes (non-square) fixed-size matrices, block-expressions and maps.
|
||||
*
|
||||
* \sa transpose(), adjoint(), transposeInPlace() */
|
||||
template<typename Derived>
|
||||
EIGEN_DEVICE_FUNC inline void MatrixBase<Derived>::adjointInPlace()
|
||||
{
|
||||
derived() = adjoint().eval();
|
||||
}
|
||||
|
||||
#ifndef EIGEN_NO_DEBUG
|
||||
|
||||
// The following is to detect aliasing problems in most common cases.
|
||||
|
||||
namespace internal {
|
||||
|
||||
template<bool DestIsTransposed, typename OtherDerived>
|
||||
struct check_transpose_aliasing_compile_time_selector
|
||||
{
|
||||
enum { ret = bool(blas_traits<OtherDerived>::IsTransposed) != DestIsTransposed };
|
||||
};
|
||||
|
||||
template<bool DestIsTransposed, typename BinOp, typename DerivedA, typename DerivedB>
|
||||
struct check_transpose_aliasing_compile_time_selector<DestIsTransposed,CwiseBinaryOp<BinOp,DerivedA,DerivedB> >
|
||||
{
|
||||
enum { ret = bool(blas_traits<DerivedA>::IsTransposed) != DestIsTransposed
|
||||
|| bool(blas_traits<DerivedB>::IsTransposed) != DestIsTransposed
|
||||
};
|
||||
};
|
||||
|
||||
template<typename Scalar, bool DestIsTransposed, typename OtherDerived>
|
||||
struct check_transpose_aliasing_run_time_selector
|
||||
{
|
||||
static bool run(const Scalar* dest, const OtherDerived& src)
|
||||
{
|
||||
return (bool(blas_traits<OtherDerived>::IsTransposed) != DestIsTransposed) && (dest!=0 && dest==(const Scalar*)extract_data(src));
|
||||
}
|
||||
};
|
||||
|
||||
template<typename Scalar, bool DestIsTransposed, typename BinOp, typename DerivedA, typename DerivedB>
|
||||
struct check_transpose_aliasing_run_time_selector<Scalar,DestIsTransposed,CwiseBinaryOp<BinOp,DerivedA,DerivedB> >
|
||||
{
|
||||
static bool run(const Scalar* dest, const CwiseBinaryOp<BinOp,DerivedA,DerivedB>& src)
|
||||
{
|
||||
return ((blas_traits<DerivedA>::IsTransposed != DestIsTransposed) && (dest!=0 && dest==(const Scalar*)extract_data(src.lhs())))
|
||||
|| ((blas_traits<DerivedB>::IsTransposed != DestIsTransposed) && (dest!=0 && dest==(const Scalar*)extract_data(src.rhs())));
|
||||
}
|
||||
};
|
||||
|
||||
// the following selector, checkTransposeAliasing_impl, based on MightHaveTransposeAliasing,
|
||||
// is because when the condition controlling the assert is known at compile time, ICC emits a warning.
|
||||
// This is actually a good warning: in expressions that don't have any transposing, the condition is
|
||||
// known at compile time to be false, and using that, we can avoid generating the code of the assert again
|
||||
// and again for all these expressions that don't need it.
|
||||
|
||||
template<typename Derived, typename OtherDerived,
|
||||
bool MightHaveTransposeAliasing
|
||||
= check_transpose_aliasing_compile_time_selector
|
||||
<blas_traits<Derived>::IsTransposed,OtherDerived>::ret
|
||||
>
|
||||
struct checkTransposeAliasing_impl
|
||||
{
|
||||
static void run(const Derived& dst, const OtherDerived& other)
|
||||
{
|
||||
eigen_assert((!check_transpose_aliasing_run_time_selector
|
||||
<typename Derived::Scalar,blas_traits<Derived>::IsTransposed,OtherDerived>
|
||||
::run(extract_data(dst), other))
|
||||
&& "aliasing detected during transposition, use transposeInPlace() "
|
||||
"or evaluate the rhs into a temporary using .eval()");
|
||||
|
||||
}
|
||||
};
|
||||
|
||||
template<typename Derived, typename OtherDerived>
|
||||
struct checkTransposeAliasing_impl<Derived, OtherDerived, false>
|
||||
{
|
||||
static void run(const Derived&, const OtherDerived&)
|
||||
{
|
||||
}
|
||||
};
|
||||
|
||||
template<typename Dst, typename Src>
|
||||
void check_for_aliasing(const Dst &dst, const Src &src)
|
||||
{
|
||||
if((!Dst::IsVectorAtCompileTime) && dst.rows()>1 && dst.cols()>1)
|
||||
internal::checkTransposeAliasing_impl<Dst, Src>::run(dst, src);
|
||||
}
|
||||
|
||||
} // end namespace internal
|
||||
|
||||
#endif // EIGEN_NO_DEBUG
|
||||
|
||||
} // end namespace Eigen
|
||||
|
||||
#endif // EIGEN_TRANSPOSE_H
|
||||
@@ -0,0 +1,386 @@
|
||||
// This file is part of Eigen, a lightweight C++ template library
|
||||
// for linear algebra.
|
||||
//
|
||||
// Copyright (C) 2010-2011 Gael Guennebaud <gael.guennebaud@inria.fr>
|
||||
//
|
||||
// This Source Code Form is subject to the terms of the Mozilla
|
||||
// Public License v. 2.0. If a copy of the MPL was not distributed
|
||||
// with this file, You can obtain one at http://mozilla.org/MPL/2.0/.
|
||||
|
||||
#ifndef EIGEN_TRANSPOSITIONS_H
|
||||
#define EIGEN_TRANSPOSITIONS_H
|
||||
|
||||
namespace Eigen {
|
||||
|
||||
template<typename Derived>
|
||||
class TranspositionsBase
|
||||
{
|
||||
typedef internal::traits<Derived> Traits;
|
||||
|
||||
public:
|
||||
|
||||
typedef typename Traits::IndicesType IndicesType;
|
||||
typedef typename IndicesType::Scalar StorageIndex;
|
||||
typedef Eigen::Index Index; ///< \deprecated since Eigen 3.3
|
||||
|
||||
EIGEN_DEVICE_FUNC
|
||||
Derived& derived() { return *static_cast<Derived*>(this); }
|
||||
EIGEN_DEVICE_FUNC
|
||||
const Derived& derived() const { return *static_cast<const Derived*>(this); }
|
||||
|
||||
/** Copies the \a other transpositions into \c *this */
|
||||
template<typename OtherDerived>
|
||||
Derived& operator=(const TranspositionsBase<OtherDerived>& other)
|
||||
{
|
||||
indices() = other.indices();
|
||||
return derived();
|
||||
}
|
||||
|
||||
/** \returns the number of transpositions */
|
||||
EIGEN_DEVICE_FUNC
|
||||
Index size() const { return indices().size(); }
|
||||
/** \returns the number of rows of the equivalent permutation matrix */
|
||||
EIGEN_DEVICE_FUNC
|
||||
Index rows() const { return indices().size(); }
|
||||
/** \returns the number of columns of the equivalent permutation matrix */
|
||||
EIGEN_DEVICE_FUNC
|
||||
Index cols() const { return indices().size(); }
|
||||
|
||||
/** Direct access to the underlying index vector */
|
||||
EIGEN_DEVICE_FUNC
|
||||
inline const StorageIndex& coeff(Index i) const { return indices().coeff(i); }
|
||||
/** Direct access to the underlying index vector */
|
||||
inline StorageIndex& coeffRef(Index i) { return indices().coeffRef(i); }
|
||||
/** Direct access to the underlying index vector */
|
||||
inline const StorageIndex& operator()(Index i) const { return indices()(i); }
|
||||
/** Direct access to the underlying index vector */
|
||||
inline StorageIndex& operator()(Index i) { return indices()(i); }
|
||||
/** Direct access to the underlying index vector */
|
||||
inline const StorageIndex& operator[](Index i) const { return indices()(i); }
|
||||
/** Direct access to the underlying index vector */
|
||||
inline StorageIndex& operator[](Index i) { return indices()(i); }
|
||||
|
||||
/** const version of indices(). */
|
||||
EIGEN_DEVICE_FUNC
|
||||
const IndicesType& indices() const { return derived().indices(); }
|
||||
/** \returns a reference to the stored array representing the transpositions. */
|
||||
EIGEN_DEVICE_FUNC
|
||||
IndicesType& indices() { return derived().indices(); }
|
||||
|
||||
/** Resizes to given size. */
|
||||
inline void resize(Index newSize)
|
||||
{
|
||||
indices().resize(newSize);
|
||||
}
|
||||
|
||||
/** Sets \c *this to represents an identity transformation */
|
||||
void setIdentity()
|
||||
{
|
||||
for(StorageIndex i = 0; i < indices().size(); ++i)
|
||||
coeffRef(i) = i;
|
||||
}
|
||||
|
||||
// FIXME: do we want such methods ?
|
||||
// might be useful when the target matrix expression is complex, e.g.:
|
||||
// object.matrix().block(..,..,..,..) = trans * object.matrix().block(..,..,..,..);
|
||||
/*
|
||||
template<typename MatrixType>
|
||||
void applyForwardToRows(MatrixType& mat) const
|
||||
{
|
||||
for(Index k=0 ; k<size() ; ++k)
|
||||
if(m_indices(k)!=k)
|
||||
mat.row(k).swap(mat.row(m_indices(k)));
|
||||
}
|
||||
|
||||
template<typename MatrixType>
|
||||
void applyBackwardToRows(MatrixType& mat) const
|
||||
{
|
||||
for(Index k=size()-1 ; k>=0 ; --k)
|
||||
if(m_indices(k)!=k)
|
||||
mat.row(k).swap(mat.row(m_indices(k)));
|
||||
}
|
||||
*/
|
||||
|
||||
/** \returns the inverse transformation */
|
||||
inline Transpose<TranspositionsBase> inverse() const
|
||||
{ return Transpose<TranspositionsBase>(derived()); }
|
||||
|
||||
/** \returns the tranpose transformation */
|
||||
inline Transpose<TranspositionsBase> transpose() const
|
||||
{ return Transpose<TranspositionsBase>(derived()); }
|
||||
|
||||
protected:
|
||||
};
|
||||
|
||||
namespace internal {
|
||||
template<int SizeAtCompileTime, int MaxSizeAtCompileTime, typename _StorageIndex>
|
||||
struct traits<Transpositions<SizeAtCompileTime,MaxSizeAtCompileTime,_StorageIndex> >
|
||||
: traits<PermutationMatrix<SizeAtCompileTime,MaxSizeAtCompileTime,_StorageIndex> >
|
||||
{
|
||||
typedef Matrix<_StorageIndex, SizeAtCompileTime, 1, 0, MaxSizeAtCompileTime, 1> IndicesType;
|
||||
typedef TranspositionsStorage StorageKind;
|
||||
};
|
||||
}
|
||||
|
||||
/** \class Transpositions
|
||||
* \ingroup Core_Module
|
||||
*
|
||||
* \brief Represents a sequence of transpositions (row/column interchange)
|
||||
*
|
||||
* \tparam SizeAtCompileTime the number of transpositions, or Dynamic
|
||||
* \tparam MaxSizeAtCompileTime the maximum number of transpositions, or Dynamic. This optional parameter defaults to SizeAtCompileTime. Most of the time, you should not have to specify it.
|
||||
*
|
||||
* This class represents a permutation transformation as a sequence of \em n transpositions
|
||||
* \f$[T_{n-1} \ldots T_{i} \ldots T_{0}]\f$. It is internally stored as a vector of integers \c indices.
|
||||
* Each transposition \f$ T_{i} \f$ applied on the left of a matrix (\f$ T_{i} M\f$) interchanges
|
||||
* the rows \c i and \c indices[i] of the matrix \c M.
|
||||
* A transposition applied on the right (e.g., \f$ M T_{i}\f$) yields a column interchange.
|
||||
*
|
||||
* Compared to the class PermutationMatrix, such a sequence of transpositions is what is
|
||||
* computed during a decomposition with pivoting, and it is faster when applying the permutation in-place.
|
||||
*
|
||||
* To apply a sequence of transpositions to a matrix, simply use the operator * as in the following example:
|
||||
* \code
|
||||
* Transpositions tr;
|
||||
* MatrixXf mat;
|
||||
* mat = tr * mat;
|
||||
* \endcode
|
||||
* In this example, we detect that the matrix appears on both side, and so the transpositions
|
||||
* are applied in-place without any temporary or extra copy.
|
||||
*
|
||||
* \sa class PermutationMatrix
|
||||
*/
|
||||
|
||||
template<int SizeAtCompileTime, int MaxSizeAtCompileTime, typename _StorageIndex>
|
||||
class Transpositions : public TranspositionsBase<Transpositions<SizeAtCompileTime,MaxSizeAtCompileTime,_StorageIndex> >
|
||||
{
|
||||
typedef internal::traits<Transpositions> Traits;
|
||||
public:
|
||||
|
||||
typedef TranspositionsBase<Transpositions> Base;
|
||||
typedef typename Traits::IndicesType IndicesType;
|
||||
typedef typename IndicesType::Scalar StorageIndex;
|
||||
|
||||
inline Transpositions() {}
|
||||
|
||||
/** Copy constructor. */
|
||||
template<typename OtherDerived>
|
||||
inline Transpositions(const TranspositionsBase<OtherDerived>& other)
|
||||
: m_indices(other.indices()) {}
|
||||
|
||||
/** Generic constructor from expression of the transposition indices. */
|
||||
template<typename Other>
|
||||
explicit inline Transpositions(const MatrixBase<Other>& indices) : m_indices(indices)
|
||||
{}
|
||||
|
||||
/** Copies the \a other transpositions into \c *this */
|
||||
template<typename OtherDerived>
|
||||
Transpositions& operator=(const TranspositionsBase<OtherDerived>& other)
|
||||
{
|
||||
return Base::operator=(other);
|
||||
}
|
||||
|
||||
/** Constructs an uninitialized permutation matrix of given size.
|
||||
*/
|
||||
inline Transpositions(Index size) : m_indices(size)
|
||||
{}
|
||||
|
||||
/** const version of indices(). */
|
||||
EIGEN_DEVICE_FUNC
|
||||
const IndicesType& indices() const { return m_indices; }
|
||||
/** \returns a reference to the stored array representing the transpositions. */
|
||||
EIGEN_DEVICE_FUNC
|
||||
IndicesType& indices() { return m_indices; }
|
||||
|
||||
protected:
|
||||
|
||||
IndicesType m_indices;
|
||||
};
|
||||
|
||||
|
||||
namespace internal {
|
||||
template<int SizeAtCompileTime, int MaxSizeAtCompileTime, typename _StorageIndex, int _PacketAccess>
|
||||
struct traits<Map<Transpositions<SizeAtCompileTime,MaxSizeAtCompileTime,_StorageIndex>,_PacketAccess> >
|
||||
: traits<PermutationMatrix<SizeAtCompileTime,MaxSizeAtCompileTime,_StorageIndex> >
|
||||
{
|
||||
typedef Map<const Matrix<_StorageIndex,SizeAtCompileTime,1,0,MaxSizeAtCompileTime,1>, _PacketAccess> IndicesType;
|
||||
typedef _StorageIndex StorageIndex;
|
||||
typedef TranspositionsStorage StorageKind;
|
||||
};
|
||||
}
|
||||
|
||||
template<int SizeAtCompileTime, int MaxSizeAtCompileTime, typename _StorageIndex, int PacketAccess>
|
||||
class Map<Transpositions<SizeAtCompileTime,MaxSizeAtCompileTime,_StorageIndex>,PacketAccess>
|
||||
: public TranspositionsBase<Map<Transpositions<SizeAtCompileTime,MaxSizeAtCompileTime,_StorageIndex>,PacketAccess> >
|
||||
{
|
||||
typedef internal::traits<Map> Traits;
|
||||
public:
|
||||
|
||||
typedef TranspositionsBase<Map> Base;
|
||||
typedef typename Traits::IndicesType IndicesType;
|
||||
typedef typename IndicesType::Scalar StorageIndex;
|
||||
|
||||
explicit inline Map(const StorageIndex* indicesPtr)
|
||||
: m_indices(indicesPtr)
|
||||
{}
|
||||
|
||||
inline Map(const StorageIndex* indicesPtr, Index size)
|
||||
: m_indices(indicesPtr,size)
|
||||
{}
|
||||
|
||||
/** Copies the \a other transpositions into \c *this */
|
||||
template<typename OtherDerived>
|
||||
Map& operator=(const TranspositionsBase<OtherDerived>& other)
|
||||
{
|
||||
return Base::operator=(other);
|
||||
}
|
||||
|
||||
#ifndef EIGEN_PARSED_BY_DOXYGEN
|
||||
/** This is a special case of the templated operator=. Its purpose is to
|
||||
* prevent a default operator= from hiding the templated operator=.
|
||||
*/
|
||||
Map& operator=(const Map& other)
|
||||
{
|
||||
m_indices = other.m_indices;
|
||||
return *this;
|
||||
}
|
||||
#endif
|
||||
|
||||
/** const version of indices(). */
|
||||
EIGEN_DEVICE_FUNC
|
||||
const IndicesType& indices() const { return m_indices; }
|
||||
|
||||
/** \returns a reference to the stored array representing the transpositions. */
|
||||
EIGEN_DEVICE_FUNC
|
||||
IndicesType& indices() { return m_indices; }
|
||||
|
||||
protected:
|
||||
|
||||
IndicesType m_indices;
|
||||
};
|
||||
|
||||
namespace internal {
|
||||
template<typename _IndicesType>
|
||||
struct traits<TranspositionsWrapper<_IndicesType> >
|
||||
: traits<PermutationWrapper<_IndicesType> >
|
||||
{
|
||||
typedef TranspositionsStorage StorageKind;
|
||||
};
|
||||
}
|
||||
|
||||
template<typename _IndicesType>
|
||||
class TranspositionsWrapper
|
||||
: public TranspositionsBase<TranspositionsWrapper<_IndicesType> >
|
||||
{
|
||||
typedef internal::traits<TranspositionsWrapper> Traits;
|
||||
public:
|
||||
|
||||
typedef TranspositionsBase<TranspositionsWrapper> Base;
|
||||
typedef typename Traits::IndicesType IndicesType;
|
||||
typedef typename IndicesType::Scalar StorageIndex;
|
||||
|
||||
explicit inline TranspositionsWrapper(IndicesType& indices)
|
||||
: m_indices(indices)
|
||||
{}
|
||||
|
||||
/** Copies the \a other transpositions into \c *this */
|
||||
template<typename OtherDerived>
|
||||
TranspositionsWrapper& operator=(const TranspositionsBase<OtherDerived>& other)
|
||||
{
|
||||
return Base::operator=(other);
|
||||
}
|
||||
|
||||
/** const version of indices(). */
|
||||
EIGEN_DEVICE_FUNC
|
||||
const IndicesType& indices() const { return m_indices; }
|
||||
|
||||
/** \returns a reference to the stored array representing the transpositions. */
|
||||
EIGEN_DEVICE_FUNC
|
||||
IndicesType& indices() { return m_indices; }
|
||||
|
||||
protected:
|
||||
|
||||
typename IndicesType::Nested m_indices;
|
||||
};
|
||||
|
||||
|
||||
|
||||
/** \returns the \a matrix with the \a transpositions applied to the columns.
|
||||
*/
|
||||
template<typename MatrixDerived, typename TranspositionsDerived>
|
||||
EIGEN_DEVICE_FUNC
|
||||
const Product<MatrixDerived, TranspositionsDerived, AliasFreeProduct>
|
||||
operator*(const MatrixBase<MatrixDerived> &matrix,
|
||||
const TranspositionsBase<TranspositionsDerived>& transpositions)
|
||||
{
|
||||
return Product<MatrixDerived, TranspositionsDerived, AliasFreeProduct>
|
||||
(matrix.derived(), transpositions.derived());
|
||||
}
|
||||
|
||||
/** \returns the \a matrix with the \a transpositions applied to the rows.
|
||||
*/
|
||||
template<typename TranspositionsDerived, typename MatrixDerived>
|
||||
EIGEN_DEVICE_FUNC
|
||||
const Product<TranspositionsDerived, MatrixDerived, AliasFreeProduct>
|
||||
operator*(const TranspositionsBase<TranspositionsDerived> &transpositions,
|
||||
const MatrixBase<MatrixDerived>& matrix)
|
||||
{
|
||||
return Product<TranspositionsDerived, MatrixDerived, AliasFreeProduct>
|
||||
(transpositions.derived(), matrix.derived());
|
||||
}
|
||||
|
||||
// Template partial specialization for transposed/inverse transpositions
|
||||
|
||||
namespace internal {
|
||||
|
||||
template<typename Derived>
|
||||
struct traits<Transpose<TranspositionsBase<Derived> > >
|
||||
: traits<Derived>
|
||||
{};
|
||||
|
||||
} // end namespace internal
|
||||
|
||||
template<typename TranspositionsDerived>
|
||||
class Transpose<TranspositionsBase<TranspositionsDerived> >
|
||||
{
|
||||
typedef TranspositionsDerived TranspositionType;
|
||||
typedef typename TranspositionType::IndicesType IndicesType;
|
||||
public:
|
||||
|
||||
explicit Transpose(const TranspositionType& t) : m_transpositions(t) {}
|
||||
|
||||
EIGEN_DEVICE_FUNC EIGEN_CONSTEXPR
|
||||
Index size() const EIGEN_NOEXCEPT { return m_transpositions.size(); }
|
||||
EIGEN_DEVICE_FUNC EIGEN_CONSTEXPR
|
||||
Index rows() const EIGEN_NOEXCEPT { return m_transpositions.size(); }
|
||||
EIGEN_DEVICE_FUNC EIGEN_CONSTEXPR
|
||||
Index cols() const EIGEN_NOEXCEPT { return m_transpositions.size(); }
|
||||
|
||||
/** \returns the \a matrix with the inverse transpositions applied to the columns.
|
||||
*/
|
||||
template<typename OtherDerived> friend
|
||||
const Product<OtherDerived, Transpose, AliasFreeProduct>
|
||||
operator*(const MatrixBase<OtherDerived>& matrix, const Transpose& trt)
|
||||
{
|
||||
return Product<OtherDerived, Transpose, AliasFreeProduct>(matrix.derived(), trt);
|
||||
}
|
||||
|
||||
/** \returns the \a matrix with the inverse transpositions applied to the rows.
|
||||
*/
|
||||
template<typename OtherDerived>
|
||||
const Product<Transpose, OtherDerived, AliasFreeProduct>
|
||||
operator*(const MatrixBase<OtherDerived>& matrix) const
|
||||
{
|
||||
return Product<Transpose, OtherDerived, AliasFreeProduct>(*this, matrix.derived());
|
||||
}
|
||||
|
||||
EIGEN_DEVICE_FUNC
|
||||
const TranspositionType& nestedExpression() const { return m_transpositions; }
|
||||
|
||||
protected:
|
||||
const TranspositionType& m_transpositions;
|
||||
};
|
||||
|
||||
} // end namespace Eigen
|
||||
|
||||
#endif // EIGEN_TRANSPOSITIONS_H
|
||||
File diff suppressed because it is too large
Load Diff
@@ -0,0 +1,96 @@
|
||||
// This file is part of Eigen, a lightweight C++ template library
|
||||
// for linear algebra.
|
||||
//
|
||||
// Copyright (C) 2008-2010 Gael Guennebaud <gael.guennebaud@inria.fr>
|
||||
// Copyright (C) 2006-2008 Benoit Jacob <jacob.benoit.1@gmail.com>
|
||||
//
|
||||
// This Source Code Form is subject to the terms of the Mozilla
|
||||
// Public License v. 2.0. If a copy of the MPL was not distributed
|
||||
// with this file, You can obtain one at http://mozilla.org/MPL/2.0/.
|
||||
|
||||
#ifndef EIGEN_VECTORBLOCK_H
|
||||
#define EIGEN_VECTORBLOCK_H
|
||||
|
||||
namespace Eigen {
|
||||
|
||||
namespace internal {
|
||||
template<typename VectorType, int Size>
|
||||
struct traits<VectorBlock<VectorType, Size> >
|
||||
: public traits<Block<VectorType,
|
||||
traits<VectorType>::Flags & RowMajorBit ? 1 : Size,
|
||||
traits<VectorType>::Flags & RowMajorBit ? Size : 1> >
|
||||
{
|
||||
};
|
||||
}
|
||||
|
||||
/** \class VectorBlock
|
||||
* \ingroup Core_Module
|
||||
*
|
||||
* \brief Expression of a fixed-size or dynamic-size sub-vector
|
||||
*
|
||||
* \tparam VectorType the type of the object in which we are taking a sub-vector
|
||||
* \tparam Size size of the sub-vector we are taking at compile time (optional)
|
||||
*
|
||||
* This class represents an expression of either a fixed-size or dynamic-size sub-vector.
|
||||
* It is the return type of DenseBase::segment(Index,Index) and DenseBase::segment<int>(Index) and
|
||||
* most of the time this is the only way it is used.
|
||||
*
|
||||
* However, if you want to directly manipulate sub-vector expressions,
|
||||
* for instance if you want to write a function returning such an expression, you
|
||||
* will need to use this class.
|
||||
*
|
||||
* Here is an example illustrating the dynamic case:
|
||||
* \include class_VectorBlock.cpp
|
||||
* Output: \verbinclude class_VectorBlock.out
|
||||
*
|
||||
* \note Even though this expression has dynamic size, in the case where \a VectorType
|
||||
* has fixed size, this expression inherits a fixed maximal size which means that evaluating
|
||||
* it does not cause a dynamic memory allocation.
|
||||
*
|
||||
* Here is an example illustrating the fixed-size case:
|
||||
* \include class_FixedVectorBlock.cpp
|
||||
* Output: \verbinclude class_FixedVectorBlock.out
|
||||
*
|
||||
* \sa class Block, DenseBase::segment(Index,Index,Index,Index), DenseBase::segment(Index,Index)
|
||||
*/
|
||||
template<typename VectorType, int Size> class VectorBlock
|
||||
: public Block<VectorType,
|
||||
internal::traits<VectorType>::Flags & RowMajorBit ? 1 : Size,
|
||||
internal::traits<VectorType>::Flags & RowMajorBit ? Size : 1>
|
||||
{
|
||||
typedef Block<VectorType,
|
||||
internal::traits<VectorType>::Flags & RowMajorBit ? 1 : Size,
|
||||
internal::traits<VectorType>::Flags & RowMajorBit ? Size : 1> Base;
|
||||
enum {
|
||||
IsColVector = !(internal::traits<VectorType>::Flags & RowMajorBit)
|
||||
};
|
||||
public:
|
||||
EIGEN_DENSE_PUBLIC_INTERFACE(VectorBlock)
|
||||
|
||||
using Base::operator=;
|
||||
|
||||
/** Dynamic-size constructor
|
||||
*/
|
||||
EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
|
||||
VectorBlock(VectorType& vector, Index start, Index size)
|
||||
: Base(vector,
|
||||
IsColVector ? start : 0, IsColVector ? 0 : start,
|
||||
IsColVector ? size : 1, IsColVector ? 1 : size)
|
||||
{
|
||||
EIGEN_STATIC_ASSERT_VECTOR_ONLY(VectorBlock);
|
||||
}
|
||||
|
||||
/** Fixed-size constructor
|
||||
*/
|
||||
EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
|
||||
VectorBlock(VectorType& vector, Index start)
|
||||
: Base(vector, IsColVector ? start : 0, IsColVector ? 0 : start)
|
||||
{
|
||||
EIGEN_STATIC_ASSERT_VECTOR_ONLY(VectorBlock);
|
||||
}
|
||||
};
|
||||
|
||||
|
||||
} // end namespace Eigen
|
||||
|
||||
#endif // EIGEN_VECTORBLOCK_H
|
||||
@@ -0,0 +1,784 @@
|
||||
// This file is part of Eigen, a lightweight C++ template library
|
||||
// for linear algebra.
|
||||
//
|
||||
// Copyright (C) 2008-2019 Gael Guennebaud <gael.guennebaud@inria.fr>
|
||||
// Copyright (C) 2006-2008 Benoit Jacob <jacob.benoit.1@gmail.com>
|
||||
//
|
||||
// This Source Code Form is subject to the terms of the Mozilla
|
||||
// Public License v. 2.0. If a copy of the MPL was not distributed
|
||||
// with this file, You can obtain one at http://mozilla.org/MPL/2.0/.
|
||||
|
||||
#ifndef EIGEN_PARTIAL_REDUX_H
|
||||
#define EIGEN_PARTIAL_REDUX_H
|
||||
|
||||
namespace Eigen {
|
||||
|
||||
/** \class PartialReduxExpr
|
||||
* \ingroup Core_Module
|
||||
*
|
||||
* \brief Generic expression of a partially reduxed matrix
|
||||
*
|
||||
* \tparam MatrixType the type of the matrix we are applying the redux operation
|
||||
* \tparam MemberOp type of the member functor
|
||||
* \tparam Direction indicates the direction of the redux (#Vertical or #Horizontal)
|
||||
*
|
||||
* This class represents an expression of a partial redux operator of a matrix.
|
||||
* It is the return type of some VectorwiseOp functions,
|
||||
* and most of the time this is the only way it is used.
|
||||
*
|
||||
* \sa class VectorwiseOp
|
||||
*/
|
||||
|
||||
template< typename MatrixType, typename MemberOp, int Direction>
|
||||
class PartialReduxExpr;
|
||||
|
||||
namespace internal {
|
||||
template<typename MatrixType, typename MemberOp, int Direction>
|
||||
struct traits<PartialReduxExpr<MatrixType, MemberOp, Direction> >
|
||||
: traits<MatrixType>
|
||||
{
|
||||
typedef typename MemberOp::result_type Scalar;
|
||||
typedef typename traits<MatrixType>::StorageKind StorageKind;
|
||||
typedef typename traits<MatrixType>::XprKind XprKind;
|
||||
typedef typename MatrixType::Scalar InputScalar;
|
||||
enum {
|
||||
RowsAtCompileTime = Direction==Vertical ? 1 : MatrixType::RowsAtCompileTime,
|
||||
ColsAtCompileTime = Direction==Horizontal ? 1 : MatrixType::ColsAtCompileTime,
|
||||
MaxRowsAtCompileTime = Direction==Vertical ? 1 : MatrixType::MaxRowsAtCompileTime,
|
||||
MaxColsAtCompileTime = Direction==Horizontal ? 1 : MatrixType::MaxColsAtCompileTime,
|
||||
Flags = RowsAtCompileTime == 1 ? RowMajorBit : 0,
|
||||
TraversalSize = Direction==Vertical ? MatrixType::RowsAtCompileTime : MatrixType::ColsAtCompileTime
|
||||
};
|
||||
};
|
||||
}
|
||||
|
||||
template< typename MatrixType, typename MemberOp, int Direction>
|
||||
class PartialReduxExpr : public internal::dense_xpr_base< PartialReduxExpr<MatrixType, MemberOp, Direction> >::type,
|
||||
internal::no_assignment_operator
|
||||
{
|
||||
public:
|
||||
|
||||
typedef typename internal::dense_xpr_base<PartialReduxExpr>::type Base;
|
||||
EIGEN_DENSE_PUBLIC_INTERFACE(PartialReduxExpr)
|
||||
|
||||
EIGEN_DEVICE_FUNC
|
||||
explicit PartialReduxExpr(const MatrixType& mat, const MemberOp& func = MemberOp())
|
||||
: m_matrix(mat), m_functor(func) {}
|
||||
|
||||
EIGEN_DEVICE_FUNC EIGEN_CONSTEXPR
|
||||
Index rows() const EIGEN_NOEXCEPT { return (Direction==Vertical ? 1 : m_matrix.rows()); }
|
||||
EIGEN_DEVICE_FUNC EIGEN_CONSTEXPR
|
||||
Index cols() const EIGEN_NOEXCEPT { return (Direction==Horizontal ? 1 : m_matrix.cols()); }
|
||||
|
||||
EIGEN_DEVICE_FUNC
|
||||
typename MatrixType::Nested nestedExpression() const { return m_matrix; }
|
||||
|
||||
EIGEN_DEVICE_FUNC
|
||||
const MemberOp& functor() const { return m_functor; }
|
||||
|
||||
protected:
|
||||
typename MatrixType::Nested m_matrix;
|
||||
const MemberOp m_functor;
|
||||
};
|
||||
|
||||
template<typename A,typename B> struct partial_redux_dummy_func;
|
||||
|
||||
#define EIGEN_MAKE_PARTIAL_REDUX_FUNCTOR(MEMBER,COST,VECTORIZABLE,BINARYOP) \
|
||||
template <typename ResultType,typename Scalar> \
|
||||
struct member_##MEMBER { \
|
||||
EIGEN_EMPTY_STRUCT_CTOR(member_##MEMBER) \
|
||||
typedef ResultType result_type; \
|
||||
typedef BINARYOP<Scalar,Scalar> BinaryOp; \
|
||||
template<int Size> struct Cost { enum { value = COST }; }; \
|
||||
enum { Vectorizable = VECTORIZABLE }; \
|
||||
template<typename XprType> \
|
||||
EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE \
|
||||
ResultType operator()(const XprType& mat) const \
|
||||
{ return mat.MEMBER(); } \
|
||||
BinaryOp binaryFunc() const { return BinaryOp(); } \
|
||||
}
|
||||
|
||||
#define EIGEN_MEMBER_FUNCTOR(MEMBER,COST) \
|
||||
EIGEN_MAKE_PARTIAL_REDUX_FUNCTOR(MEMBER,COST,0,partial_redux_dummy_func)
|
||||
|
||||
namespace internal {
|
||||
|
||||
EIGEN_MEMBER_FUNCTOR(norm, (Size+5) * NumTraits<Scalar>::MulCost + (Size-1)*NumTraits<Scalar>::AddCost);
|
||||
EIGEN_MEMBER_FUNCTOR(stableNorm, (Size+5) * NumTraits<Scalar>::MulCost + (Size-1)*NumTraits<Scalar>::AddCost);
|
||||
EIGEN_MEMBER_FUNCTOR(blueNorm, (Size+5) * NumTraits<Scalar>::MulCost + (Size-1)*NumTraits<Scalar>::AddCost);
|
||||
EIGEN_MEMBER_FUNCTOR(hypotNorm, (Size-1) * functor_traits<scalar_hypot_op<Scalar> >::Cost );
|
||||
EIGEN_MEMBER_FUNCTOR(all, (Size-1)*NumTraits<Scalar>::AddCost);
|
||||
EIGEN_MEMBER_FUNCTOR(any, (Size-1)*NumTraits<Scalar>::AddCost);
|
||||
EIGEN_MEMBER_FUNCTOR(count, (Size-1)*NumTraits<Scalar>::AddCost);
|
||||
|
||||
EIGEN_MAKE_PARTIAL_REDUX_FUNCTOR(sum, (Size-1)*NumTraits<Scalar>::AddCost, 1, internal::scalar_sum_op);
|
||||
EIGEN_MAKE_PARTIAL_REDUX_FUNCTOR(minCoeff, (Size-1)*NumTraits<Scalar>::AddCost, 1, internal::scalar_min_op);
|
||||
EIGEN_MAKE_PARTIAL_REDUX_FUNCTOR(maxCoeff, (Size-1)*NumTraits<Scalar>::AddCost, 1, internal::scalar_max_op);
|
||||
EIGEN_MAKE_PARTIAL_REDUX_FUNCTOR(prod, (Size-1)*NumTraits<Scalar>::MulCost, 1, internal::scalar_product_op);
|
||||
|
||||
template <int p, typename ResultType,typename Scalar>
|
||||
struct member_lpnorm {
|
||||
typedef ResultType result_type;
|
||||
enum { Vectorizable = 0 };
|
||||
template<int Size> struct Cost
|
||||
{ enum { value = (Size+5) * NumTraits<Scalar>::MulCost + (Size-1)*NumTraits<Scalar>::AddCost }; };
|
||||
EIGEN_DEVICE_FUNC member_lpnorm() {}
|
||||
template<typename XprType>
|
||||
EIGEN_DEVICE_FUNC inline ResultType operator()(const XprType& mat) const
|
||||
{ return mat.template lpNorm<p>(); }
|
||||
};
|
||||
|
||||
template <typename BinaryOpT, typename Scalar>
|
||||
struct member_redux {
|
||||
typedef BinaryOpT BinaryOp;
|
||||
typedef typename result_of<
|
||||
BinaryOp(const Scalar&,const Scalar&)
|
||||
>::type result_type;
|
||||
|
||||
enum { Vectorizable = functor_traits<BinaryOp>::PacketAccess };
|
||||
template<int Size> struct Cost { enum { value = (Size-1) * functor_traits<BinaryOp>::Cost }; };
|
||||
EIGEN_DEVICE_FUNC explicit member_redux(const BinaryOp func) : m_functor(func) {}
|
||||
template<typename Derived>
|
||||
EIGEN_DEVICE_FUNC inline result_type operator()(const DenseBase<Derived>& mat) const
|
||||
{ return mat.redux(m_functor); }
|
||||
const BinaryOp& binaryFunc() const { return m_functor; }
|
||||
const BinaryOp m_functor;
|
||||
};
|
||||
}
|
||||
|
||||
/** \class VectorwiseOp
|
||||
* \ingroup Core_Module
|
||||
*
|
||||
* \brief Pseudo expression providing broadcasting and partial reduction operations
|
||||
*
|
||||
* \tparam ExpressionType the type of the object on which to do partial reductions
|
||||
* \tparam Direction indicates whether to operate on columns (#Vertical) or rows (#Horizontal)
|
||||
*
|
||||
* This class represents a pseudo expression with broadcasting and partial reduction features.
|
||||
* It is the return type of DenseBase::colwise() and DenseBase::rowwise()
|
||||
* and most of the time this is the only way it is explicitly used.
|
||||
*
|
||||
* To understand the logic of rowwise/colwise expression, let's consider a generic case `A.colwise().foo()`
|
||||
* where `foo` is any method of `VectorwiseOp`. This expression is equivalent to applying `foo()` to each
|
||||
* column of `A` and then re-assemble the outputs in a matrix expression:
|
||||
* \code [A.col(0).foo(), A.col(1).foo(), ..., A.col(A.cols()-1).foo()] \endcode
|
||||
*
|
||||
* Example: \include MatrixBase_colwise.cpp
|
||||
* Output: \verbinclude MatrixBase_colwise.out
|
||||
*
|
||||
* The begin() and end() methods are obviously exceptions to the previous rule as they
|
||||
* return STL-compatible begin/end iterators to the rows or columns of the nested expression.
|
||||
* Typical use cases include for-range-loop and calls to STL algorithms:
|
||||
*
|
||||
* Example: \include MatrixBase_colwise_iterator_cxx11.cpp
|
||||
* Output: \verbinclude MatrixBase_colwise_iterator_cxx11.out
|
||||
*
|
||||
* For a partial reduction on an empty input, some rules apply.
|
||||
* For the sake of clarity, let's consider a vertical reduction:
|
||||
* - If the number of columns is zero, then a 1x0 row-major vector expression is returned.
|
||||
* - Otherwise, if the number of rows is zero, then
|
||||
* - a row vector of zeros is returned for sum-like reductions (sum, squaredNorm, norm, etc.)
|
||||
* - a row vector of ones is returned for a product reduction (e.g., <code>MatrixXd(n,0).colwise().prod()</code>)
|
||||
* - an assert is triggered for all other reductions (minCoeff,maxCoeff,redux(bin_op))
|
||||
*
|
||||
* \sa DenseBase::colwise(), DenseBase::rowwise(), class PartialReduxExpr
|
||||
*/
|
||||
template<typename ExpressionType, int Direction> class VectorwiseOp
|
||||
{
|
||||
public:
|
||||
|
||||
typedef typename ExpressionType::Scalar Scalar;
|
||||
typedef typename ExpressionType::RealScalar RealScalar;
|
||||
typedef Eigen::Index Index; ///< \deprecated since Eigen 3.3
|
||||
typedef typename internal::ref_selector<ExpressionType>::non_const_type ExpressionTypeNested;
|
||||
typedef typename internal::remove_all<ExpressionTypeNested>::type ExpressionTypeNestedCleaned;
|
||||
|
||||
template<template<typename OutScalar,typename InputScalar> class Functor,
|
||||
typename ReturnScalar=Scalar> struct ReturnType
|
||||
{
|
||||
typedef PartialReduxExpr<ExpressionType,
|
||||
Functor<ReturnScalar,Scalar>,
|
||||
Direction
|
||||
> Type;
|
||||
};
|
||||
|
||||
template<typename BinaryOp> struct ReduxReturnType
|
||||
{
|
||||
typedef PartialReduxExpr<ExpressionType,
|
||||
internal::member_redux<BinaryOp,Scalar>,
|
||||
Direction
|
||||
> Type;
|
||||
};
|
||||
|
||||
enum {
|
||||
isVertical = (Direction==Vertical) ? 1 : 0,
|
||||
isHorizontal = (Direction==Horizontal) ? 1 : 0
|
||||
};
|
||||
|
||||
protected:
|
||||
|
||||
template<typename OtherDerived> struct ExtendedType {
|
||||
typedef Replicate<OtherDerived,
|
||||
isVertical ? 1 : ExpressionType::RowsAtCompileTime,
|
||||
isHorizontal ? 1 : ExpressionType::ColsAtCompileTime> Type;
|
||||
};
|
||||
|
||||
/** \internal
|
||||
* Replicates a vector to match the size of \c *this */
|
||||
template<typename OtherDerived>
|
||||
EIGEN_DEVICE_FUNC
|
||||
typename ExtendedType<OtherDerived>::Type
|
||||
extendedTo(const DenseBase<OtherDerived>& other) const
|
||||
{
|
||||
EIGEN_STATIC_ASSERT(EIGEN_IMPLIES(isVertical, OtherDerived::MaxColsAtCompileTime==1),
|
||||
YOU_PASSED_A_ROW_VECTOR_BUT_A_COLUMN_VECTOR_WAS_EXPECTED)
|
||||
EIGEN_STATIC_ASSERT(EIGEN_IMPLIES(isHorizontal, OtherDerived::MaxRowsAtCompileTime==1),
|
||||
YOU_PASSED_A_COLUMN_VECTOR_BUT_A_ROW_VECTOR_WAS_EXPECTED)
|
||||
return typename ExtendedType<OtherDerived>::Type
|
||||
(other.derived(),
|
||||
isVertical ? 1 : m_matrix.rows(),
|
||||
isHorizontal ? 1 : m_matrix.cols());
|
||||
}
|
||||
|
||||
template<typename OtherDerived> struct OppositeExtendedType {
|
||||
typedef Replicate<OtherDerived,
|
||||
isHorizontal ? 1 : ExpressionType::RowsAtCompileTime,
|
||||
isVertical ? 1 : ExpressionType::ColsAtCompileTime> Type;
|
||||
};
|
||||
|
||||
/** \internal
|
||||
* Replicates a vector in the opposite direction to match the size of \c *this */
|
||||
template<typename OtherDerived>
|
||||
EIGEN_DEVICE_FUNC
|
||||
typename OppositeExtendedType<OtherDerived>::Type
|
||||
extendedToOpposite(const DenseBase<OtherDerived>& other) const
|
||||
{
|
||||
EIGEN_STATIC_ASSERT(EIGEN_IMPLIES(isHorizontal, OtherDerived::MaxColsAtCompileTime==1),
|
||||
YOU_PASSED_A_ROW_VECTOR_BUT_A_COLUMN_VECTOR_WAS_EXPECTED)
|
||||
EIGEN_STATIC_ASSERT(EIGEN_IMPLIES(isVertical, OtherDerived::MaxRowsAtCompileTime==1),
|
||||
YOU_PASSED_A_COLUMN_VECTOR_BUT_A_ROW_VECTOR_WAS_EXPECTED)
|
||||
return typename OppositeExtendedType<OtherDerived>::Type
|
||||
(other.derived(),
|
||||
isHorizontal ? 1 : m_matrix.rows(),
|
||||
isVertical ? 1 : m_matrix.cols());
|
||||
}
|
||||
|
||||
public:
|
||||
EIGEN_DEVICE_FUNC
|
||||
explicit inline VectorwiseOp(ExpressionType& matrix) : m_matrix(matrix) {}
|
||||
|
||||
/** \internal */
|
||||
EIGEN_DEVICE_FUNC
|
||||
inline const ExpressionType& _expression() const { return m_matrix; }
|
||||
|
||||
#ifdef EIGEN_PARSED_BY_DOXYGEN
|
||||
/** STL-like <a href="https://en.cppreference.com/w/cpp/named_req/RandomAccessIterator">RandomAccessIterator</a>
|
||||
* iterator type over the columns or rows as returned by the begin() and end() methods.
|
||||
*/
|
||||
random_access_iterator_type iterator;
|
||||
/** This is the const version of iterator (aka read-only) */
|
||||
random_access_iterator_type const_iterator;
|
||||
#else
|
||||
typedef internal::subvector_stl_iterator<ExpressionType, DirectionType(Direction)> iterator;
|
||||
typedef internal::subvector_stl_iterator<const ExpressionType, DirectionType(Direction)> const_iterator;
|
||||
typedef internal::subvector_stl_reverse_iterator<ExpressionType, DirectionType(Direction)> reverse_iterator;
|
||||
typedef internal::subvector_stl_reverse_iterator<const ExpressionType, DirectionType(Direction)> const_reverse_iterator;
|
||||
#endif
|
||||
|
||||
/** returns an iterator to the first row (rowwise) or column (colwise) of the nested expression.
|
||||
* \sa end(), cbegin()
|
||||
*/
|
||||
iterator begin() { return iterator (m_matrix, 0); }
|
||||
/** const version of begin() */
|
||||
const_iterator begin() const { return const_iterator(m_matrix, 0); }
|
||||
/** const version of begin() */
|
||||
const_iterator cbegin() const { return const_iterator(m_matrix, 0); }
|
||||
|
||||
/** returns a reverse iterator to the last row (rowwise) or column (colwise) of the nested expression.
|
||||
* \sa rend(), crbegin()
|
||||
*/
|
||||
reverse_iterator rbegin() { return reverse_iterator (m_matrix, m_matrix.template subVectors<DirectionType(Direction)>()-1); }
|
||||
/** const version of rbegin() */
|
||||
const_reverse_iterator rbegin() const { return const_reverse_iterator (m_matrix, m_matrix.template subVectors<DirectionType(Direction)>()-1); }
|
||||
/** const version of rbegin() */
|
||||
const_reverse_iterator crbegin() const { return const_reverse_iterator (m_matrix, m_matrix.template subVectors<DirectionType(Direction)>()-1); }
|
||||
|
||||
/** returns an iterator to the row (resp. column) following the last row (resp. column) of the nested expression
|
||||
* \sa begin(), cend()
|
||||
*/
|
||||
iterator end() { return iterator (m_matrix, m_matrix.template subVectors<DirectionType(Direction)>()); }
|
||||
/** const version of end() */
|
||||
const_iterator end() const { return const_iterator(m_matrix, m_matrix.template subVectors<DirectionType(Direction)>()); }
|
||||
/** const version of end() */
|
||||
const_iterator cend() const { return const_iterator(m_matrix, m_matrix.template subVectors<DirectionType(Direction)>()); }
|
||||
|
||||
/** returns a reverse iterator to the row (resp. column) before the first row (resp. column) of the nested expression
|
||||
* \sa begin(), cend()
|
||||
*/
|
||||
reverse_iterator rend() { return reverse_iterator (m_matrix, -1); }
|
||||
/** const version of rend() */
|
||||
const_reverse_iterator rend() const { return const_reverse_iterator (m_matrix, -1); }
|
||||
/** const version of rend() */
|
||||
const_reverse_iterator crend() const { return const_reverse_iterator (m_matrix, -1); }
|
||||
|
||||
/** \returns a row or column vector expression of \c *this reduxed by \a func
|
||||
*
|
||||
* The template parameter \a BinaryOp is the type of the functor
|
||||
* of the custom redux operator. Note that func must be an associative operator.
|
||||
*
|
||||
* \warning the size along the reduction direction must be strictly positive,
|
||||
* otherwise an assertion is triggered.
|
||||
*
|
||||
* \sa class VectorwiseOp, DenseBase::colwise(), DenseBase::rowwise()
|
||||
*/
|
||||
template<typename BinaryOp>
|
||||
EIGEN_DEVICE_FUNC
|
||||
const typename ReduxReturnType<BinaryOp>::Type
|
||||
redux(const BinaryOp& func = BinaryOp()) const
|
||||
{
|
||||
eigen_assert(redux_length()>0 && "you are using an empty matrix");
|
||||
return typename ReduxReturnType<BinaryOp>::Type(_expression(), internal::member_redux<BinaryOp,Scalar>(func));
|
||||
}
|
||||
|
||||
typedef typename ReturnType<internal::member_minCoeff>::Type MinCoeffReturnType;
|
||||
typedef typename ReturnType<internal::member_maxCoeff>::Type MaxCoeffReturnType;
|
||||
typedef PartialReduxExpr<const CwiseUnaryOp<internal::scalar_abs2_op<Scalar>, const ExpressionTypeNestedCleaned>,internal::member_sum<RealScalar,RealScalar>,Direction> SquaredNormReturnType;
|
||||
typedef CwiseUnaryOp<internal::scalar_sqrt_op<RealScalar>, const SquaredNormReturnType> NormReturnType;
|
||||
typedef typename ReturnType<internal::member_blueNorm,RealScalar>::Type BlueNormReturnType;
|
||||
typedef typename ReturnType<internal::member_stableNorm,RealScalar>::Type StableNormReturnType;
|
||||
typedef typename ReturnType<internal::member_hypotNorm,RealScalar>::Type HypotNormReturnType;
|
||||
typedef typename ReturnType<internal::member_sum>::Type SumReturnType;
|
||||
typedef EIGEN_EXPR_BINARYOP_SCALAR_RETURN_TYPE(SumReturnType,Scalar,quotient) MeanReturnType;
|
||||
typedef typename ReturnType<internal::member_all>::Type AllReturnType;
|
||||
typedef typename ReturnType<internal::member_any>::Type AnyReturnType;
|
||||
typedef PartialReduxExpr<ExpressionType, internal::member_count<Index,Scalar>, Direction> CountReturnType;
|
||||
typedef typename ReturnType<internal::member_prod>::Type ProdReturnType;
|
||||
typedef Reverse<const ExpressionType, Direction> ConstReverseReturnType;
|
||||
typedef Reverse<ExpressionType, Direction> ReverseReturnType;
|
||||
|
||||
template<int p> struct LpNormReturnType {
|
||||
typedef PartialReduxExpr<ExpressionType, internal::member_lpnorm<p,RealScalar,Scalar>,Direction> Type;
|
||||
};
|
||||
|
||||
/** \returns a row (or column) vector expression of the smallest coefficient
|
||||
* of each column (or row) of the referenced expression.
|
||||
*
|
||||
* \warning the size along the reduction direction must be strictly positive,
|
||||
* otherwise an assertion is triggered.
|
||||
*
|
||||
* \warning the result is undefined if \c *this contains NaN.
|
||||
*
|
||||
* Example: \include PartialRedux_minCoeff.cpp
|
||||
* Output: \verbinclude PartialRedux_minCoeff.out
|
||||
*
|
||||
* \sa DenseBase::minCoeff() */
|
||||
EIGEN_DEVICE_FUNC
|
||||
const MinCoeffReturnType minCoeff() const
|
||||
{
|
||||
eigen_assert(redux_length()>0 && "you are using an empty matrix");
|
||||
return MinCoeffReturnType(_expression());
|
||||
}
|
||||
|
||||
/** \returns a row (or column) vector expression of the largest coefficient
|
||||
* of each column (or row) of the referenced expression.
|
||||
*
|
||||
* \warning the size along the reduction direction must be strictly positive,
|
||||
* otherwise an assertion is triggered.
|
||||
*
|
||||
* \warning the result is undefined if \c *this contains NaN.
|
||||
*
|
||||
* Example: \include PartialRedux_maxCoeff.cpp
|
||||
* Output: \verbinclude PartialRedux_maxCoeff.out
|
||||
*
|
||||
* \sa DenseBase::maxCoeff() */
|
||||
EIGEN_DEVICE_FUNC
|
||||
const MaxCoeffReturnType maxCoeff() const
|
||||
{
|
||||
eigen_assert(redux_length()>0 && "you are using an empty matrix");
|
||||
return MaxCoeffReturnType(_expression());
|
||||
}
|
||||
|
||||
/** \returns a row (or column) vector expression of the squared norm
|
||||
* of each column (or row) of the referenced expression.
|
||||
* This is a vector with real entries, even if the original matrix has complex entries.
|
||||
*
|
||||
* Example: \include PartialRedux_squaredNorm.cpp
|
||||
* Output: \verbinclude PartialRedux_squaredNorm.out
|
||||
*
|
||||
* \sa DenseBase::squaredNorm() */
|
||||
EIGEN_DEVICE_FUNC
|
||||
const SquaredNormReturnType squaredNorm() const
|
||||
{ return SquaredNormReturnType(m_matrix.cwiseAbs2()); }
|
||||
|
||||
/** \returns a row (or column) vector expression of the norm
|
||||
* of each column (or row) of the referenced expression.
|
||||
* This is a vector with real entries, even if the original matrix has complex entries.
|
||||
*
|
||||
* Example: \include PartialRedux_norm.cpp
|
||||
* Output: \verbinclude PartialRedux_norm.out
|
||||
*
|
||||
* \sa DenseBase::norm() */
|
||||
EIGEN_DEVICE_FUNC
|
||||
const NormReturnType norm() const
|
||||
{ return NormReturnType(squaredNorm()); }
|
||||
|
||||
/** \returns a row (or column) vector expression of the norm
|
||||
* of each column (or row) of the referenced expression.
|
||||
* This is a vector with real entries, even if the original matrix has complex entries.
|
||||
*
|
||||
* Example: \include PartialRedux_norm.cpp
|
||||
* Output: \verbinclude PartialRedux_norm.out
|
||||
*
|
||||
* \sa DenseBase::norm() */
|
||||
template<int p>
|
||||
EIGEN_DEVICE_FUNC
|
||||
const typename LpNormReturnType<p>::Type lpNorm() const
|
||||
{ return typename LpNormReturnType<p>::Type(_expression()); }
|
||||
|
||||
|
||||
/** \returns a row (or column) vector expression of the norm
|
||||
* of each column (or row) of the referenced expression, using
|
||||
* Blue's algorithm.
|
||||
* This is a vector with real entries, even if the original matrix has complex entries.
|
||||
*
|
||||
* \sa DenseBase::blueNorm() */
|
||||
EIGEN_DEVICE_FUNC
|
||||
const BlueNormReturnType blueNorm() const
|
||||
{ return BlueNormReturnType(_expression()); }
|
||||
|
||||
|
||||
/** \returns a row (or column) vector expression of the norm
|
||||
* of each column (or row) of the referenced expression, avoiding
|
||||
* underflow and overflow.
|
||||
* This is a vector with real entries, even if the original matrix has complex entries.
|
||||
*
|
||||
* \sa DenseBase::stableNorm() */
|
||||
EIGEN_DEVICE_FUNC
|
||||
const StableNormReturnType stableNorm() const
|
||||
{ return StableNormReturnType(_expression()); }
|
||||
|
||||
|
||||
/** \returns a row (or column) vector expression of the norm
|
||||
* of each column (or row) of the referenced expression, avoiding
|
||||
* underflow and overflow using a concatenation of hypot() calls.
|
||||
* This is a vector with real entries, even if the original matrix has complex entries.
|
||||
*
|
||||
* \sa DenseBase::hypotNorm() */
|
||||
EIGEN_DEVICE_FUNC
|
||||
const HypotNormReturnType hypotNorm() const
|
||||
{ return HypotNormReturnType(_expression()); }
|
||||
|
||||
/** \returns a row (or column) vector expression of the sum
|
||||
* of each column (or row) of the referenced expression.
|
||||
*
|
||||
* Example: \include PartialRedux_sum.cpp
|
||||
* Output: \verbinclude PartialRedux_sum.out
|
||||
*
|
||||
* \sa DenseBase::sum() */
|
||||
EIGEN_DEVICE_FUNC
|
||||
const SumReturnType sum() const
|
||||
{ return SumReturnType(_expression()); }
|
||||
|
||||
/** \returns a row (or column) vector expression of the mean
|
||||
* of each column (or row) of the referenced expression.
|
||||
*
|
||||
* \sa DenseBase::mean() */
|
||||
EIGEN_DEVICE_FUNC
|
||||
const MeanReturnType mean() const
|
||||
{ return sum() / Scalar(Direction==Vertical?m_matrix.rows():m_matrix.cols()); }
|
||||
|
||||
/** \returns a row (or column) vector expression representing
|
||||
* whether \b all coefficients of each respective column (or row) are \c true.
|
||||
* This expression can be assigned to a vector with entries of type \c bool.
|
||||
*
|
||||
* \sa DenseBase::all() */
|
||||
EIGEN_DEVICE_FUNC
|
||||
const AllReturnType all() const
|
||||
{ return AllReturnType(_expression()); }
|
||||
|
||||
/** \returns a row (or column) vector expression representing
|
||||
* whether \b at \b least one coefficient of each respective column (or row) is \c true.
|
||||
* This expression can be assigned to a vector with entries of type \c bool.
|
||||
*
|
||||
* \sa DenseBase::any() */
|
||||
EIGEN_DEVICE_FUNC
|
||||
const AnyReturnType any() const
|
||||
{ return AnyReturnType(_expression()); }
|
||||
|
||||
/** \returns a row (or column) vector expression representing
|
||||
* the number of \c true coefficients of each respective column (or row).
|
||||
* This expression can be assigned to a vector whose entries have the same type as is used to
|
||||
* index entries of the original matrix; for dense matrices, this is \c std::ptrdiff_t .
|
||||
*
|
||||
* Example: \include PartialRedux_count.cpp
|
||||
* Output: \verbinclude PartialRedux_count.out
|
||||
*
|
||||
* \sa DenseBase::count() */
|
||||
EIGEN_DEVICE_FUNC
|
||||
const CountReturnType count() const
|
||||
{ return CountReturnType(_expression()); }
|
||||
|
||||
/** \returns a row (or column) vector expression of the product
|
||||
* of each column (or row) of the referenced expression.
|
||||
*
|
||||
* Example: \include PartialRedux_prod.cpp
|
||||
* Output: \verbinclude PartialRedux_prod.out
|
||||
*
|
||||
* \sa DenseBase::prod() */
|
||||
EIGEN_DEVICE_FUNC
|
||||
const ProdReturnType prod() const
|
||||
{ return ProdReturnType(_expression()); }
|
||||
|
||||
|
||||
/** \returns a matrix expression
|
||||
* where each column (or row) are reversed.
|
||||
*
|
||||
* Example: \include Vectorwise_reverse.cpp
|
||||
* Output: \verbinclude Vectorwise_reverse.out
|
||||
*
|
||||
* \sa DenseBase::reverse() */
|
||||
EIGEN_DEVICE_FUNC
|
||||
const ConstReverseReturnType reverse() const
|
||||
{ return ConstReverseReturnType( _expression() ); }
|
||||
|
||||
/** \returns a writable matrix expression
|
||||
* where each column (or row) are reversed.
|
||||
*
|
||||
* \sa reverse() const */
|
||||
EIGEN_DEVICE_FUNC
|
||||
ReverseReturnType reverse()
|
||||
{ return ReverseReturnType( _expression() ); }
|
||||
|
||||
typedef Replicate<ExpressionType,(isVertical?Dynamic:1),(isHorizontal?Dynamic:1)> ReplicateReturnType;
|
||||
EIGEN_DEVICE_FUNC
|
||||
const ReplicateReturnType replicate(Index factor) const;
|
||||
|
||||
/**
|
||||
* \return an expression of the replication of each column (or row) of \c *this
|
||||
*
|
||||
* Example: \include DirectionWise_replicate.cpp
|
||||
* Output: \verbinclude DirectionWise_replicate.out
|
||||
*
|
||||
* \sa VectorwiseOp::replicate(Index), DenseBase::replicate(), class Replicate
|
||||
*/
|
||||
// NOTE implemented here because of sunstudio's compilation errors
|
||||
// isVertical*Factor+isHorizontal instead of (isVertical?Factor:1) to handle CUDA bug with ternary operator
|
||||
template<int Factor> const Replicate<ExpressionType,isVertical*Factor+isHorizontal,isHorizontal*Factor+isVertical>
|
||||
EIGEN_DEVICE_FUNC
|
||||
replicate(Index factor = Factor) const
|
||||
{
|
||||
return Replicate<ExpressionType,(isVertical?Factor:1),(isHorizontal?Factor:1)>
|
||||
(_expression(),isVertical?factor:1,isHorizontal?factor:1);
|
||||
}
|
||||
|
||||
/////////// Artithmetic operators ///////////
|
||||
|
||||
/** Copies the vector \a other to each subvector of \c *this */
|
||||
template<typename OtherDerived>
|
||||
EIGEN_DEVICE_FUNC
|
||||
ExpressionType& operator=(const DenseBase<OtherDerived>& other)
|
||||
{
|
||||
EIGEN_STATIC_ASSERT_VECTOR_ONLY(OtherDerived)
|
||||
EIGEN_STATIC_ASSERT_SAME_XPR_KIND(ExpressionType, OtherDerived)
|
||||
//eigen_assert((m_matrix.isNull()) == (other.isNull())); FIXME
|
||||
return m_matrix = extendedTo(other.derived());
|
||||
}
|
||||
|
||||
/** Adds the vector \a other to each subvector of \c *this */
|
||||
template<typename OtherDerived>
|
||||
EIGEN_DEVICE_FUNC
|
||||
ExpressionType& operator+=(const DenseBase<OtherDerived>& other)
|
||||
{
|
||||
EIGEN_STATIC_ASSERT_VECTOR_ONLY(OtherDerived)
|
||||
EIGEN_STATIC_ASSERT_SAME_XPR_KIND(ExpressionType, OtherDerived)
|
||||
return m_matrix += extendedTo(other.derived());
|
||||
}
|
||||
|
||||
/** Substracts the vector \a other to each subvector of \c *this */
|
||||
template<typename OtherDerived>
|
||||
EIGEN_DEVICE_FUNC
|
||||
ExpressionType& operator-=(const DenseBase<OtherDerived>& other)
|
||||
{
|
||||
EIGEN_STATIC_ASSERT_VECTOR_ONLY(OtherDerived)
|
||||
EIGEN_STATIC_ASSERT_SAME_XPR_KIND(ExpressionType, OtherDerived)
|
||||
return m_matrix -= extendedTo(other.derived());
|
||||
}
|
||||
|
||||
/** Multiples each subvector of \c *this by the vector \a other */
|
||||
template<typename OtherDerived>
|
||||
EIGEN_DEVICE_FUNC
|
||||
ExpressionType& operator*=(const DenseBase<OtherDerived>& other)
|
||||
{
|
||||
EIGEN_STATIC_ASSERT_VECTOR_ONLY(OtherDerived)
|
||||
EIGEN_STATIC_ASSERT_ARRAYXPR(ExpressionType)
|
||||
EIGEN_STATIC_ASSERT_SAME_XPR_KIND(ExpressionType, OtherDerived)
|
||||
m_matrix *= extendedTo(other.derived());
|
||||
return m_matrix;
|
||||
}
|
||||
|
||||
/** Divides each subvector of \c *this by the vector \a other */
|
||||
template<typename OtherDerived>
|
||||
EIGEN_DEVICE_FUNC
|
||||
ExpressionType& operator/=(const DenseBase<OtherDerived>& other)
|
||||
{
|
||||
EIGEN_STATIC_ASSERT_VECTOR_ONLY(OtherDerived)
|
||||
EIGEN_STATIC_ASSERT_ARRAYXPR(ExpressionType)
|
||||
EIGEN_STATIC_ASSERT_SAME_XPR_KIND(ExpressionType, OtherDerived)
|
||||
m_matrix /= extendedTo(other.derived());
|
||||
return m_matrix;
|
||||
}
|
||||
|
||||
/** Returns the expression of the sum of the vector \a other to each subvector of \c *this */
|
||||
template<typename OtherDerived> EIGEN_STRONG_INLINE EIGEN_DEVICE_FUNC
|
||||
CwiseBinaryOp<internal::scalar_sum_op<Scalar,typename OtherDerived::Scalar>,
|
||||
const ExpressionTypeNestedCleaned,
|
||||
const typename ExtendedType<OtherDerived>::Type>
|
||||
operator+(const DenseBase<OtherDerived>& other) const
|
||||
{
|
||||
EIGEN_STATIC_ASSERT_VECTOR_ONLY(OtherDerived)
|
||||
EIGEN_STATIC_ASSERT_SAME_XPR_KIND(ExpressionType, OtherDerived)
|
||||
return m_matrix + extendedTo(other.derived());
|
||||
}
|
||||
|
||||
/** Returns the expression of the difference between each subvector of \c *this and the vector \a other */
|
||||
template<typename OtherDerived>
|
||||
EIGEN_DEVICE_FUNC
|
||||
CwiseBinaryOp<internal::scalar_difference_op<Scalar,typename OtherDerived::Scalar>,
|
||||
const ExpressionTypeNestedCleaned,
|
||||
const typename ExtendedType<OtherDerived>::Type>
|
||||
operator-(const DenseBase<OtherDerived>& other) const
|
||||
{
|
||||
EIGEN_STATIC_ASSERT_VECTOR_ONLY(OtherDerived)
|
||||
EIGEN_STATIC_ASSERT_SAME_XPR_KIND(ExpressionType, OtherDerived)
|
||||
return m_matrix - extendedTo(other.derived());
|
||||
}
|
||||
|
||||
/** Returns the expression where each subvector is the product of the vector \a other
|
||||
* by the corresponding subvector of \c *this */
|
||||
template<typename OtherDerived> EIGEN_STRONG_INLINE EIGEN_DEVICE_FUNC
|
||||
CwiseBinaryOp<internal::scalar_product_op<Scalar>,
|
||||
const ExpressionTypeNestedCleaned,
|
||||
const typename ExtendedType<OtherDerived>::Type>
|
||||
EIGEN_DEVICE_FUNC
|
||||
operator*(const DenseBase<OtherDerived>& other) const
|
||||
{
|
||||
EIGEN_STATIC_ASSERT_VECTOR_ONLY(OtherDerived)
|
||||
EIGEN_STATIC_ASSERT_ARRAYXPR(ExpressionType)
|
||||
EIGEN_STATIC_ASSERT_SAME_XPR_KIND(ExpressionType, OtherDerived)
|
||||
return m_matrix * extendedTo(other.derived());
|
||||
}
|
||||
|
||||
/** Returns the expression where each subvector is the quotient of the corresponding
|
||||
* subvector of \c *this by the vector \a other */
|
||||
template<typename OtherDerived>
|
||||
EIGEN_DEVICE_FUNC
|
||||
CwiseBinaryOp<internal::scalar_quotient_op<Scalar>,
|
||||
const ExpressionTypeNestedCleaned,
|
||||
const typename ExtendedType<OtherDerived>::Type>
|
||||
operator/(const DenseBase<OtherDerived>& other) const
|
||||
{
|
||||
EIGEN_STATIC_ASSERT_VECTOR_ONLY(OtherDerived)
|
||||
EIGEN_STATIC_ASSERT_ARRAYXPR(ExpressionType)
|
||||
EIGEN_STATIC_ASSERT_SAME_XPR_KIND(ExpressionType, OtherDerived)
|
||||
return m_matrix / extendedTo(other.derived());
|
||||
}
|
||||
|
||||
/** \returns an expression where each column (or row) of the referenced matrix are normalized.
|
||||
* The referenced matrix is \b not modified.
|
||||
* \sa MatrixBase::normalized(), normalize()
|
||||
*/
|
||||
EIGEN_DEVICE_FUNC
|
||||
CwiseBinaryOp<internal::scalar_quotient_op<Scalar>,
|
||||
const ExpressionTypeNestedCleaned,
|
||||
const typename OppositeExtendedType<NormReturnType>::Type>
|
||||
normalized() const { return m_matrix.cwiseQuotient(extendedToOpposite(this->norm())); }
|
||||
|
||||
|
||||
/** Normalize in-place each row or columns of the referenced matrix.
|
||||
* \sa MatrixBase::normalize(), normalized()
|
||||
*/
|
||||
EIGEN_DEVICE_FUNC void normalize() {
|
||||
m_matrix = this->normalized();
|
||||
}
|
||||
|
||||
EIGEN_DEVICE_FUNC inline void reverseInPlace();
|
||||
|
||||
/////////// Geometry module ///////////
|
||||
|
||||
typedef Homogeneous<ExpressionType,Direction> HomogeneousReturnType;
|
||||
EIGEN_DEVICE_FUNC
|
||||
HomogeneousReturnType homogeneous() const;
|
||||
|
||||
typedef typename ExpressionType::PlainObject CrossReturnType;
|
||||
template<typename OtherDerived>
|
||||
EIGEN_DEVICE_FUNC
|
||||
const CrossReturnType cross(const MatrixBase<OtherDerived>& other) const;
|
||||
|
||||
enum {
|
||||
HNormalized_Size = Direction==Vertical ? internal::traits<ExpressionType>::RowsAtCompileTime
|
||||
: internal::traits<ExpressionType>::ColsAtCompileTime,
|
||||
HNormalized_SizeMinusOne = HNormalized_Size==Dynamic ? Dynamic : HNormalized_Size-1
|
||||
};
|
||||
typedef Block<const ExpressionType,
|
||||
Direction==Vertical ? int(HNormalized_SizeMinusOne)
|
||||
: int(internal::traits<ExpressionType>::RowsAtCompileTime),
|
||||
Direction==Horizontal ? int(HNormalized_SizeMinusOne)
|
||||
: int(internal::traits<ExpressionType>::ColsAtCompileTime)>
|
||||
HNormalized_Block;
|
||||
typedef Block<const ExpressionType,
|
||||
Direction==Vertical ? 1 : int(internal::traits<ExpressionType>::RowsAtCompileTime),
|
||||
Direction==Horizontal ? 1 : int(internal::traits<ExpressionType>::ColsAtCompileTime)>
|
||||
HNormalized_Factors;
|
||||
typedef CwiseBinaryOp<internal::scalar_quotient_op<typename internal::traits<ExpressionType>::Scalar>,
|
||||
const HNormalized_Block,
|
||||
const Replicate<HNormalized_Factors,
|
||||
Direction==Vertical ? HNormalized_SizeMinusOne : 1,
|
||||
Direction==Horizontal ? HNormalized_SizeMinusOne : 1> >
|
||||
HNormalizedReturnType;
|
||||
|
||||
EIGEN_DEVICE_FUNC
|
||||
const HNormalizedReturnType hnormalized() const;
|
||||
|
||||
# ifdef EIGEN_VECTORWISEOP_PLUGIN
|
||||
# include EIGEN_VECTORWISEOP_PLUGIN
|
||||
# endif
|
||||
|
||||
protected:
|
||||
Index redux_length() const
|
||||
{
|
||||
return Direction==Vertical ? m_matrix.rows() : m_matrix.cols();
|
||||
}
|
||||
ExpressionTypeNested m_matrix;
|
||||
};
|
||||
|
||||
//const colwise moved to DenseBase.h due to CUDA compiler bug
|
||||
|
||||
|
||||
/** \returns a writable VectorwiseOp wrapper of *this providing additional partial reduction operations
|
||||
*
|
||||
* \sa rowwise(), class VectorwiseOp, \ref TutorialReductionsVisitorsBroadcasting
|
||||
*/
|
||||
template<typename Derived>
|
||||
EIGEN_DEVICE_FUNC inline typename DenseBase<Derived>::ColwiseReturnType
|
||||
DenseBase<Derived>::colwise()
|
||||
{
|
||||
return ColwiseReturnType(derived());
|
||||
}
|
||||
|
||||
//const rowwise moved to DenseBase.h due to CUDA compiler bug
|
||||
|
||||
|
||||
/** \returns a writable VectorwiseOp wrapper of *this providing additional partial reduction operations
|
||||
*
|
||||
* \sa colwise(), class VectorwiseOp, \ref TutorialReductionsVisitorsBroadcasting
|
||||
*/
|
||||
template<typename Derived>
|
||||
EIGEN_DEVICE_FUNC inline typename DenseBase<Derived>::RowwiseReturnType
|
||||
DenseBase<Derived>::rowwise()
|
||||
{
|
||||
return RowwiseReturnType(derived());
|
||||
}
|
||||
|
||||
} // end namespace Eigen
|
||||
|
||||
#endif // EIGEN_PARTIAL_REDUX_H
|
||||
@@ -0,0 +1,381 @@
|
||||
// This file is part of Eigen, a lightweight C++ template library
|
||||
// for linear algebra.
|
||||
//
|
||||
// Copyright (C) 2008 Gael Guennebaud <gael.guennebaud@inria.fr>
|
||||
//
|
||||
// This Source Code Form is subject to the terms of the Mozilla
|
||||
// Public License v. 2.0. If a copy of the MPL was not distributed
|
||||
// with this file, You can obtain one at http://mozilla.org/MPL/2.0/.
|
||||
|
||||
#ifndef EIGEN_VISITOR_H
|
||||
#define EIGEN_VISITOR_H
|
||||
|
||||
namespace Eigen {
|
||||
|
||||
namespace internal {
|
||||
|
||||
template<typename Visitor, typename Derived, int UnrollCount>
|
||||
struct visitor_impl
|
||||
{
|
||||
enum {
|
||||
col = (UnrollCount-1) / Derived::RowsAtCompileTime,
|
||||
row = (UnrollCount-1) % Derived::RowsAtCompileTime
|
||||
};
|
||||
|
||||
EIGEN_DEVICE_FUNC
|
||||
static inline void run(const Derived &mat, Visitor& visitor)
|
||||
{
|
||||
visitor_impl<Visitor, Derived, UnrollCount-1>::run(mat, visitor);
|
||||
visitor(mat.coeff(row, col), row, col);
|
||||
}
|
||||
};
|
||||
|
||||
template<typename Visitor, typename Derived>
|
||||
struct visitor_impl<Visitor, Derived, 1>
|
||||
{
|
||||
EIGEN_DEVICE_FUNC
|
||||
static inline void run(const Derived &mat, Visitor& visitor)
|
||||
{
|
||||
return visitor.init(mat.coeff(0, 0), 0, 0);
|
||||
}
|
||||
};
|
||||
|
||||
// This specialization enables visitors on empty matrices at compile-time
|
||||
template<typename Visitor, typename Derived>
|
||||
struct visitor_impl<Visitor, Derived, 0> {
|
||||
EIGEN_DEVICE_FUNC
|
||||
static inline void run(const Derived &/*mat*/, Visitor& /*visitor*/)
|
||||
{}
|
||||
};
|
||||
|
||||
template<typename Visitor, typename Derived>
|
||||
struct visitor_impl<Visitor, Derived, Dynamic>
|
||||
{
|
||||
EIGEN_DEVICE_FUNC
|
||||
static inline void run(const Derived& mat, Visitor& visitor)
|
||||
{
|
||||
visitor.init(mat.coeff(0,0), 0, 0);
|
||||
for(Index i = 1; i < mat.rows(); ++i)
|
||||
visitor(mat.coeff(i, 0), i, 0);
|
||||
for(Index j = 1; j < mat.cols(); ++j)
|
||||
for(Index i = 0; i < mat.rows(); ++i)
|
||||
visitor(mat.coeff(i, j), i, j);
|
||||
}
|
||||
};
|
||||
|
||||
// evaluator adaptor
|
||||
template<typename XprType>
|
||||
class visitor_evaluator
|
||||
{
|
||||
public:
|
||||
EIGEN_DEVICE_FUNC
|
||||
explicit visitor_evaluator(const XprType &xpr) : m_evaluator(xpr), m_xpr(xpr) {}
|
||||
|
||||
typedef typename XprType::Scalar Scalar;
|
||||
typedef typename XprType::CoeffReturnType CoeffReturnType;
|
||||
|
||||
enum {
|
||||
RowsAtCompileTime = XprType::RowsAtCompileTime,
|
||||
CoeffReadCost = internal::evaluator<XprType>::CoeffReadCost
|
||||
};
|
||||
|
||||
EIGEN_DEVICE_FUNC EIGEN_CONSTEXPR Index rows() const EIGEN_NOEXCEPT { return m_xpr.rows(); }
|
||||
EIGEN_DEVICE_FUNC EIGEN_CONSTEXPR Index cols() const EIGEN_NOEXCEPT { return m_xpr.cols(); }
|
||||
EIGEN_DEVICE_FUNC EIGEN_CONSTEXPR Index size() const EIGEN_NOEXCEPT { return m_xpr.size(); }
|
||||
|
||||
EIGEN_DEVICE_FUNC CoeffReturnType coeff(Index row, Index col) const
|
||||
{ return m_evaluator.coeff(row, col); }
|
||||
|
||||
protected:
|
||||
internal::evaluator<XprType> m_evaluator;
|
||||
const XprType &m_xpr;
|
||||
};
|
||||
} // end namespace internal
|
||||
|
||||
/** Applies the visitor \a visitor to the whole coefficients of the matrix or vector.
|
||||
*
|
||||
* The template parameter \a Visitor is the type of the visitor and provides the following interface:
|
||||
* \code
|
||||
* struct MyVisitor {
|
||||
* // called for the first coefficient
|
||||
* void init(const Scalar& value, Index i, Index j);
|
||||
* // called for all other coefficients
|
||||
* void operator() (const Scalar& value, Index i, Index j);
|
||||
* };
|
||||
* \endcode
|
||||
*
|
||||
* \note compared to one or two \em for \em loops, visitors offer automatic
|
||||
* unrolling for small fixed size matrix.
|
||||
*
|
||||
* \note if the matrix is empty, then the visitor is left unchanged.
|
||||
*
|
||||
* \sa minCoeff(Index*,Index*), maxCoeff(Index*,Index*), DenseBase::redux()
|
||||
*/
|
||||
template<typename Derived>
|
||||
template<typename Visitor>
|
||||
EIGEN_DEVICE_FUNC
|
||||
void DenseBase<Derived>::visit(Visitor& visitor) const
|
||||
{
|
||||
if(size()==0)
|
||||
return;
|
||||
|
||||
typedef typename internal::visitor_evaluator<Derived> ThisEvaluator;
|
||||
ThisEvaluator thisEval(derived());
|
||||
|
||||
enum {
|
||||
unroll = SizeAtCompileTime != Dynamic
|
||||
&& SizeAtCompileTime * int(ThisEvaluator::CoeffReadCost) + (SizeAtCompileTime-1) * int(internal::functor_traits<Visitor>::Cost) <= EIGEN_UNROLLING_LIMIT
|
||||
};
|
||||
return internal::visitor_impl<Visitor, ThisEvaluator, unroll ? int(SizeAtCompileTime) : Dynamic>::run(thisEval, visitor);
|
||||
}
|
||||
|
||||
namespace internal {
|
||||
|
||||
/** \internal
|
||||
* \brief Base class to implement min and max visitors
|
||||
*/
|
||||
template <typename Derived>
|
||||
struct coeff_visitor
|
||||
{
|
||||
// default initialization to avoid countless invalid maybe-uninitialized warnings by gcc
|
||||
EIGEN_DEVICE_FUNC
|
||||
coeff_visitor() : row(-1), col(-1), res(0) {}
|
||||
typedef typename Derived::Scalar Scalar;
|
||||
Index row, col;
|
||||
Scalar res;
|
||||
EIGEN_DEVICE_FUNC
|
||||
inline void init(const Scalar& value, Index i, Index j)
|
||||
{
|
||||
res = value;
|
||||
row = i;
|
||||
col = j;
|
||||
}
|
||||
};
|
||||
|
||||
/** \internal
|
||||
* \brief Visitor computing the min coefficient with its value and coordinates
|
||||
*
|
||||
* \sa DenseBase::minCoeff(Index*, Index*)
|
||||
*/
|
||||
template <typename Derived, int NaNPropagation>
|
||||
struct min_coeff_visitor : coeff_visitor<Derived>
|
||||
{
|
||||
typedef typename Derived::Scalar Scalar;
|
||||
EIGEN_DEVICE_FUNC
|
||||
void operator() (const Scalar& value, Index i, Index j)
|
||||
{
|
||||
if(value < this->res)
|
||||
{
|
||||
this->res = value;
|
||||
this->row = i;
|
||||
this->col = j;
|
||||
}
|
||||
}
|
||||
};
|
||||
|
||||
template <typename Derived>
|
||||
struct min_coeff_visitor<Derived, PropagateNumbers> : coeff_visitor<Derived>
|
||||
{
|
||||
typedef typename Derived::Scalar Scalar;
|
||||
EIGEN_DEVICE_FUNC
|
||||
void operator() (const Scalar& value, Index i, Index j)
|
||||
{
|
||||
if((numext::isnan)(this->res) || (!(numext::isnan)(value) && value < this->res))
|
||||
{
|
||||
this->res = value;
|
||||
this->row = i;
|
||||
this->col = j;
|
||||
}
|
||||
}
|
||||
};
|
||||
|
||||
template <typename Derived>
|
||||
struct min_coeff_visitor<Derived, PropagateNaN> : coeff_visitor<Derived>
|
||||
{
|
||||
typedef typename Derived::Scalar Scalar;
|
||||
EIGEN_DEVICE_FUNC
|
||||
void operator() (const Scalar& value, Index i, Index j)
|
||||
{
|
||||
if((numext::isnan)(value) || value < this->res)
|
||||
{
|
||||
this->res = value;
|
||||
this->row = i;
|
||||
this->col = j;
|
||||
}
|
||||
}
|
||||
};
|
||||
|
||||
template<typename Scalar, int NaNPropagation>
|
||||
struct functor_traits<min_coeff_visitor<Scalar, NaNPropagation> > {
|
||||
enum {
|
||||
Cost = NumTraits<Scalar>::AddCost
|
||||
};
|
||||
};
|
||||
|
||||
/** \internal
|
||||
* \brief Visitor computing the max coefficient with its value and coordinates
|
||||
*
|
||||
* \sa DenseBase::maxCoeff(Index*, Index*)
|
||||
*/
|
||||
template <typename Derived, int NaNPropagation>
|
||||
struct max_coeff_visitor : coeff_visitor<Derived>
|
||||
{
|
||||
typedef typename Derived::Scalar Scalar;
|
||||
EIGEN_DEVICE_FUNC
|
||||
void operator() (const Scalar& value, Index i, Index j)
|
||||
{
|
||||
if(value > this->res)
|
||||
{
|
||||
this->res = value;
|
||||
this->row = i;
|
||||
this->col = j;
|
||||
}
|
||||
}
|
||||
};
|
||||
|
||||
template <typename Derived>
|
||||
struct max_coeff_visitor<Derived, PropagateNumbers> : coeff_visitor<Derived>
|
||||
{
|
||||
typedef typename Derived::Scalar Scalar;
|
||||
EIGEN_DEVICE_FUNC
|
||||
void operator() (const Scalar& value, Index i, Index j)
|
||||
{
|
||||
if((numext::isnan)(this->res) || (!(numext::isnan)(value) && value > this->res))
|
||||
{
|
||||
this->res = value;
|
||||
this->row = i;
|
||||
this->col = j;
|
||||
}
|
||||
}
|
||||
};
|
||||
|
||||
template <typename Derived>
|
||||
struct max_coeff_visitor<Derived, PropagateNaN> : coeff_visitor<Derived>
|
||||
{
|
||||
typedef typename Derived::Scalar Scalar;
|
||||
EIGEN_DEVICE_FUNC
|
||||
void operator() (const Scalar& value, Index i, Index j)
|
||||
{
|
||||
if((numext::isnan)(value) || value > this->res)
|
||||
{
|
||||
this->res = value;
|
||||
this->row = i;
|
||||
this->col = j;
|
||||
}
|
||||
}
|
||||
};
|
||||
|
||||
template<typename Scalar, int NaNPropagation>
|
||||
struct functor_traits<max_coeff_visitor<Scalar, NaNPropagation> > {
|
||||
enum {
|
||||
Cost = NumTraits<Scalar>::AddCost
|
||||
};
|
||||
};
|
||||
|
||||
} // end namespace internal
|
||||
|
||||
/** \fn DenseBase<Derived>::minCoeff(IndexType* rowId, IndexType* colId) const
|
||||
* \returns the minimum of all coefficients of *this and puts in *row and *col its location.
|
||||
*
|
||||
* In case \c *this contains NaN, NaNPropagation determines the behavior:
|
||||
* NaNPropagation == PropagateFast : undefined
|
||||
* NaNPropagation == PropagateNaN : result is NaN
|
||||
* NaNPropagation == PropagateNumbers : result is maximum of elements that are not NaN
|
||||
* \warning the matrix must be not empty, otherwise an assertion is triggered.
|
||||
*
|
||||
* \sa DenseBase::minCoeff(Index*), DenseBase::maxCoeff(Index*,Index*), DenseBase::visit(), DenseBase::minCoeff()
|
||||
*/
|
||||
template<typename Derived>
|
||||
template<int NaNPropagation, typename IndexType>
|
||||
EIGEN_DEVICE_FUNC
|
||||
typename internal::traits<Derived>::Scalar
|
||||
DenseBase<Derived>::minCoeff(IndexType* rowId, IndexType* colId) const
|
||||
{
|
||||
eigen_assert(this->rows()>0 && this->cols()>0 && "you are using an empty matrix");
|
||||
|
||||
internal::min_coeff_visitor<Derived, NaNPropagation> minVisitor;
|
||||
this->visit(minVisitor);
|
||||
*rowId = minVisitor.row;
|
||||
if (colId) *colId = minVisitor.col;
|
||||
return minVisitor.res;
|
||||
}
|
||||
|
||||
/** \returns the minimum of all coefficients of *this and puts in *index its location.
|
||||
*
|
||||
* In case \c *this contains NaN, NaNPropagation determines the behavior:
|
||||
* NaNPropagation == PropagateFast : undefined
|
||||
* NaNPropagation == PropagateNaN : result is NaN
|
||||
* NaNPropagation == PropagateNumbers : result is maximum of elements that are not NaN
|
||||
* \warning the matrix must be not empty, otherwise an assertion is triggered.
|
||||
*
|
||||
* \sa DenseBase::minCoeff(IndexType*,IndexType*), DenseBase::maxCoeff(IndexType*,IndexType*), DenseBase::visit(), DenseBase::minCoeff()
|
||||
*/
|
||||
template<typename Derived>
|
||||
template<int NaNPropagation, typename IndexType>
|
||||
EIGEN_DEVICE_FUNC
|
||||
typename internal::traits<Derived>::Scalar
|
||||
DenseBase<Derived>::minCoeff(IndexType* index) const
|
||||
{
|
||||
eigen_assert(this->rows()>0 && this->cols()>0 && "you are using an empty matrix");
|
||||
|
||||
EIGEN_STATIC_ASSERT_VECTOR_ONLY(Derived)
|
||||
internal::min_coeff_visitor<Derived, NaNPropagation> minVisitor;
|
||||
this->visit(minVisitor);
|
||||
*index = IndexType((RowsAtCompileTime==1) ? minVisitor.col : minVisitor.row);
|
||||
return minVisitor.res;
|
||||
}
|
||||
|
||||
/** \fn DenseBase<Derived>::maxCoeff(IndexType* rowId, IndexType* colId) const
|
||||
* \returns the maximum of all coefficients of *this and puts in *row and *col its location.
|
||||
*
|
||||
* In case \c *this contains NaN, NaNPropagation determines the behavior:
|
||||
* NaNPropagation == PropagateFast : undefined
|
||||
* NaNPropagation == PropagateNaN : result is NaN
|
||||
* NaNPropagation == PropagateNumbers : result is maximum of elements that are not NaN
|
||||
* \warning the matrix must be not empty, otherwise an assertion is triggered.
|
||||
*
|
||||
* \sa DenseBase::minCoeff(IndexType*,IndexType*), DenseBase::visit(), DenseBase::maxCoeff()
|
||||
*/
|
||||
template<typename Derived>
|
||||
template<int NaNPropagation, typename IndexType>
|
||||
EIGEN_DEVICE_FUNC
|
||||
typename internal::traits<Derived>::Scalar
|
||||
DenseBase<Derived>::maxCoeff(IndexType* rowPtr, IndexType* colPtr) const
|
||||
{
|
||||
eigen_assert(this->rows()>0 && this->cols()>0 && "you are using an empty matrix");
|
||||
|
||||
internal::max_coeff_visitor<Derived, NaNPropagation> maxVisitor;
|
||||
this->visit(maxVisitor);
|
||||
*rowPtr = maxVisitor.row;
|
||||
if (colPtr) *colPtr = maxVisitor.col;
|
||||
return maxVisitor.res;
|
||||
}
|
||||
|
||||
/** \returns the maximum of all coefficients of *this and puts in *index its location.
|
||||
*
|
||||
* In case \c *this contains NaN, NaNPropagation determines the behavior:
|
||||
* NaNPropagation == PropagateFast : undefined
|
||||
* NaNPropagation == PropagateNaN : result is NaN
|
||||
* NaNPropagation == PropagateNumbers : result is maximum of elements that are not NaN
|
||||
* \warning the matrix must be not empty, otherwise an assertion is triggered.
|
||||
*
|
||||
* \sa DenseBase::maxCoeff(IndexType*,IndexType*), DenseBase::minCoeff(IndexType*,IndexType*), DenseBase::visitor(), DenseBase::maxCoeff()
|
||||
*/
|
||||
template<typename Derived>
|
||||
template<int NaNPropagation, typename IndexType>
|
||||
EIGEN_DEVICE_FUNC
|
||||
typename internal::traits<Derived>::Scalar
|
||||
DenseBase<Derived>::maxCoeff(IndexType* index) const
|
||||
{
|
||||
eigen_assert(this->rows()>0 && this->cols()>0 && "you are using an empty matrix");
|
||||
|
||||
EIGEN_STATIC_ASSERT_VECTOR_ONLY(Derived)
|
||||
internal::max_coeff_visitor<Derived, NaNPropagation> maxVisitor;
|
||||
this->visit(maxVisitor);
|
||||
*index = (RowsAtCompileTime==1) ? maxVisitor.col : maxVisitor.row;
|
||||
return maxVisitor.res;
|
||||
}
|
||||
|
||||
} // end namespace Eigen
|
||||
|
||||
#endif // EIGEN_VISITOR_H
|
||||
@@ -0,0 +1,372 @@
|
||||
// This file is part of Eigen, a lightweight C++ template library
|
||||
// for linear algebra.
|
||||
//
|
||||
// Copyright (C) 2014 Benoit Steiner (benoit.steiner.goog@gmail.com)
|
||||
//
|
||||
// This Source Code Form is subject to the terms of the Mozilla
|
||||
// Public License v. 2.0. If a copy of the MPL was not distributed
|
||||
// with this file, You can obtain one at http://mozilla.org/MPL/2.0/.
|
||||
|
||||
#ifndef EIGEN_COMPLEX_AVX_H
|
||||
#define EIGEN_COMPLEX_AVX_H
|
||||
|
||||
namespace Eigen {
|
||||
|
||||
namespace internal {
|
||||
|
||||
//---------- float ----------
|
||||
struct Packet4cf
|
||||
{
|
||||
EIGEN_STRONG_INLINE Packet4cf() {}
|
||||
EIGEN_STRONG_INLINE explicit Packet4cf(const __m256& a) : v(a) {}
|
||||
__m256 v;
|
||||
};
|
||||
|
||||
#ifndef EIGEN_VECTORIZE_AVX512
|
||||
template<> struct packet_traits<std::complex<float> > : default_packet_traits
|
||||
{
|
||||
typedef Packet4cf type;
|
||||
typedef Packet2cf half;
|
||||
enum {
|
||||
Vectorizable = 1,
|
||||
AlignedOnScalar = 1,
|
||||
size = 4,
|
||||
HasHalfPacket = 1,
|
||||
|
||||
HasAdd = 1,
|
||||
HasSub = 1,
|
||||
HasMul = 1,
|
||||
HasDiv = 1,
|
||||
HasNegate = 1,
|
||||
HasSqrt = 1,
|
||||
HasAbs = 0,
|
||||
HasAbs2 = 0,
|
||||
HasMin = 0,
|
||||
HasMax = 0,
|
||||
HasSetLinear = 0
|
||||
};
|
||||
};
|
||||
#endif
|
||||
|
||||
template<> struct unpacket_traits<Packet4cf> {
|
||||
typedef std::complex<float> type;
|
||||
typedef Packet2cf half;
|
||||
typedef Packet8f as_real;
|
||||
enum {
|
||||
size=4,
|
||||
alignment=Aligned32,
|
||||
vectorizable=true,
|
||||
masked_load_available=false,
|
||||
masked_store_available=false
|
||||
};
|
||||
};
|
||||
|
||||
template<> EIGEN_STRONG_INLINE Packet4cf padd<Packet4cf>(const Packet4cf& a, const Packet4cf& b) { return Packet4cf(_mm256_add_ps(a.v,b.v)); }
|
||||
template<> EIGEN_STRONG_INLINE Packet4cf psub<Packet4cf>(const Packet4cf& a, const Packet4cf& b) { return Packet4cf(_mm256_sub_ps(a.v,b.v)); }
|
||||
template<> EIGEN_STRONG_INLINE Packet4cf pnegate(const Packet4cf& a)
|
||||
{
|
||||
return Packet4cf(pnegate(a.v));
|
||||
}
|
||||
template<> EIGEN_STRONG_INLINE Packet4cf pconj(const Packet4cf& a)
|
||||
{
|
||||
const __m256 mask = _mm256_castsi256_ps(_mm256_setr_epi32(0x00000000,0x80000000,0x00000000,0x80000000,0x00000000,0x80000000,0x00000000,0x80000000));
|
||||
return Packet4cf(_mm256_xor_ps(a.v,mask));
|
||||
}
|
||||
|
||||
template<> EIGEN_STRONG_INLINE Packet4cf pmul<Packet4cf>(const Packet4cf& a, const Packet4cf& b)
|
||||
{
|
||||
__m256 tmp1 = _mm256_mul_ps(_mm256_moveldup_ps(a.v), b.v);
|
||||
__m256 tmp2 = _mm256_mul_ps(_mm256_movehdup_ps(a.v), _mm256_permute_ps(b.v, _MM_SHUFFLE(2,3,0,1)));
|
||||
__m256 result = _mm256_addsub_ps(tmp1, tmp2);
|
||||
return Packet4cf(result);
|
||||
}
|
||||
|
||||
template <>
|
||||
EIGEN_STRONG_INLINE Packet4cf pcmp_eq(const Packet4cf& a, const Packet4cf& b) {
|
||||
__m256 eq = _mm256_cmp_ps(a.v, b.v, _CMP_EQ_OQ);
|
||||
return Packet4cf(_mm256_and_ps(eq, _mm256_permute_ps(eq, 0xb1)));
|
||||
}
|
||||
|
||||
template<> EIGEN_STRONG_INLINE Packet4cf ptrue<Packet4cf>(const Packet4cf& a) { return Packet4cf(ptrue(Packet8f(a.v))); }
|
||||
template<> EIGEN_STRONG_INLINE Packet4cf pand <Packet4cf>(const Packet4cf& a, const Packet4cf& b) { return Packet4cf(_mm256_and_ps(a.v,b.v)); }
|
||||
template<> EIGEN_STRONG_INLINE Packet4cf por <Packet4cf>(const Packet4cf& a, const Packet4cf& b) { return Packet4cf(_mm256_or_ps(a.v,b.v)); }
|
||||
template<> EIGEN_STRONG_INLINE Packet4cf pxor <Packet4cf>(const Packet4cf& a, const Packet4cf& b) { return Packet4cf(_mm256_xor_ps(a.v,b.v)); }
|
||||
template<> EIGEN_STRONG_INLINE Packet4cf pandnot<Packet4cf>(const Packet4cf& a, const Packet4cf& b) { return Packet4cf(_mm256_andnot_ps(b.v,a.v)); }
|
||||
|
||||
template<> EIGEN_STRONG_INLINE Packet4cf pload <Packet4cf>(const std::complex<float>* from) { EIGEN_DEBUG_ALIGNED_LOAD return Packet4cf(pload<Packet8f>(&numext::real_ref(*from))); }
|
||||
template<> EIGEN_STRONG_INLINE Packet4cf ploadu<Packet4cf>(const std::complex<float>* from) { EIGEN_DEBUG_UNALIGNED_LOAD return Packet4cf(ploadu<Packet8f>(&numext::real_ref(*from))); }
|
||||
|
||||
|
||||
template<> EIGEN_STRONG_INLINE Packet4cf pset1<Packet4cf>(const std::complex<float>& from)
|
||||
{
|
||||
return Packet4cf(_mm256_castpd_ps(_mm256_broadcast_sd((const double*)(const void*)&from)));
|
||||
}
|
||||
|
||||
template<> EIGEN_STRONG_INLINE Packet4cf ploaddup<Packet4cf>(const std::complex<float>* from)
|
||||
{
|
||||
// FIXME The following might be optimized using _mm256_movedup_pd
|
||||
Packet2cf a = ploaddup<Packet2cf>(from);
|
||||
Packet2cf b = ploaddup<Packet2cf>(from+1);
|
||||
return Packet4cf(_mm256_insertf128_ps(_mm256_castps128_ps256(a.v), b.v, 1));
|
||||
}
|
||||
|
||||
template<> EIGEN_STRONG_INLINE void pstore <std::complex<float> >(std::complex<float>* to, const Packet4cf& from) { EIGEN_DEBUG_ALIGNED_STORE pstore(&numext::real_ref(*to), from.v); }
|
||||
template<> EIGEN_STRONG_INLINE void pstoreu<std::complex<float> >(std::complex<float>* to, const Packet4cf& from) { EIGEN_DEBUG_UNALIGNED_STORE pstoreu(&numext::real_ref(*to), from.v); }
|
||||
|
||||
template<> EIGEN_DEVICE_FUNC inline Packet4cf pgather<std::complex<float>, Packet4cf>(const std::complex<float>* from, Index stride)
|
||||
{
|
||||
return Packet4cf(_mm256_set_ps(std::imag(from[3*stride]), std::real(from[3*stride]),
|
||||
std::imag(from[2*stride]), std::real(from[2*stride]),
|
||||
std::imag(from[1*stride]), std::real(from[1*stride]),
|
||||
std::imag(from[0*stride]), std::real(from[0*stride])));
|
||||
}
|
||||
|
||||
template<> EIGEN_DEVICE_FUNC inline void pscatter<std::complex<float>, Packet4cf>(std::complex<float>* to, const Packet4cf& from, Index stride)
|
||||
{
|
||||
__m128 low = _mm256_extractf128_ps(from.v, 0);
|
||||
to[stride*0] = std::complex<float>(_mm_cvtss_f32(_mm_shuffle_ps(low, low, 0)),
|
||||
_mm_cvtss_f32(_mm_shuffle_ps(low, low, 1)));
|
||||
to[stride*1] = std::complex<float>(_mm_cvtss_f32(_mm_shuffle_ps(low, low, 2)),
|
||||
_mm_cvtss_f32(_mm_shuffle_ps(low, low, 3)));
|
||||
|
||||
__m128 high = _mm256_extractf128_ps(from.v, 1);
|
||||
to[stride*2] = std::complex<float>(_mm_cvtss_f32(_mm_shuffle_ps(high, high, 0)),
|
||||
_mm_cvtss_f32(_mm_shuffle_ps(high, high, 1)));
|
||||
to[stride*3] = std::complex<float>(_mm_cvtss_f32(_mm_shuffle_ps(high, high, 2)),
|
||||
_mm_cvtss_f32(_mm_shuffle_ps(high, high, 3)));
|
||||
|
||||
}
|
||||
|
||||
template<> EIGEN_STRONG_INLINE std::complex<float> pfirst<Packet4cf>(const Packet4cf& a)
|
||||
{
|
||||
return pfirst(Packet2cf(_mm256_castps256_ps128(a.v)));
|
||||
}
|
||||
|
||||
template<> EIGEN_STRONG_INLINE Packet4cf preverse(const Packet4cf& a) {
|
||||
__m128 low = _mm256_extractf128_ps(a.v, 0);
|
||||
__m128 high = _mm256_extractf128_ps(a.v, 1);
|
||||
__m128d lowd = _mm_castps_pd(low);
|
||||
__m128d highd = _mm_castps_pd(high);
|
||||
low = _mm_castpd_ps(_mm_shuffle_pd(lowd,lowd,0x1));
|
||||
high = _mm_castpd_ps(_mm_shuffle_pd(highd,highd,0x1));
|
||||
__m256 result = _mm256_setzero_ps();
|
||||
result = _mm256_insertf128_ps(result, low, 1);
|
||||
result = _mm256_insertf128_ps(result, high, 0);
|
||||
return Packet4cf(result);
|
||||
}
|
||||
|
||||
template<> EIGEN_STRONG_INLINE std::complex<float> predux<Packet4cf>(const Packet4cf& a)
|
||||
{
|
||||
return predux(padd(Packet2cf(_mm256_extractf128_ps(a.v,0)),
|
||||
Packet2cf(_mm256_extractf128_ps(a.v,1))));
|
||||
}
|
||||
|
||||
template<> EIGEN_STRONG_INLINE std::complex<float> predux_mul<Packet4cf>(const Packet4cf& a)
|
||||
{
|
||||
return predux_mul(pmul(Packet2cf(_mm256_extractf128_ps(a.v, 0)),
|
||||
Packet2cf(_mm256_extractf128_ps(a.v, 1))));
|
||||
}
|
||||
|
||||
EIGEN_MAKE_CONJ_HELPER_CPLX_REAL(Packet4cf,Packet8f)
|
||||
|
||||
template<> EIGEN_STRONG_INLINE Packet4cf pdiv<Packet4cf>(const Packet4cf& a, const Packet4cf& b)
|
||||
{
|
||||
Packet4cf num = pmul(a, pconj(b));
|
||||
__m256 tmp = _mm256_mul_ps(b.v, b.v);
|
||||
__m256 tmp2 = _mm256_shuffle_ps(tmp,tmp,0xB1);
|
||||
__m256 denom = _mm256_add_ps(tmp, tmp2);
|
||||
return Packet4cf(_mm256_div_ps(num.v, denom));
|
||||
}
|
||||
|
||||
template<> EIGEN_STRONG_INLINE Packet4cf pcplxflip<Packet4cf>(const Packet4cf& x)
|
||||
{
|
||||
return Packet4cf(_mm256_shuffle_ps(x.v, x.v, _MM_SHUFFLE(2, 3, 0 ,1)));
|
||||
}
|
||||
|
||||
//---------- double ----------
|
||||
struct Packet2cd
|
||||
{
|
||||
EIGEN_STRONG_INLINE Packet2cd() {}
|
||||
EIGEN_STRONG_INLINE explicit Packet2cd(const __m256d& a) : v(a) {}
|
||||
__m256d v;
|
||||
};
|
||||
|
||||
#ifndef EIGEN_VECTORIZE_AVX512
|
||||
template<> struct packet_traits<std::complex<double> > : default_packet_traits
|
||||
{
|
||||
typedef Packet2cd type;
|
||||
typedef Packet1cd half;
|
||||
enum {
|
||||
Vectorizable = 1,
|
||||
AlignedOnScalar = 0,
|
||||
size = 2,
|
||||
HasHalfPacket = 1,
|
||||
|
||||
HasAdd = 1,
|
||||
HasSub = 1,
|
||||
HasMul = 1,
|
||||
HasDiv = 1,
|
||||
HasNegate = 1,
|
||||
HasSqrt = 1,
|
||||
HasAbs = 0,
|
||||
HasAbs2 = 0,
|
||||
HasMin = 0,
|
||||
HasMax = 0,
|
||||
HasSetLinear = 0
|
||||
};
|
||||
};
|
||||
#endif
|
||||
|
||||
template<> struct unpacket_traits<Packet2cd> {
|
||||
typedef std::complex<double> type;
|
||||
typedef Packet1cd half;
|
||||
typedef Packet4d as_real;
|
||||
enum {
|
||||
size=2,
|
||||
alignment=Aligned32,
|
||||
vectorizable=true,
|
||||
masked_load_available=false,
|
||||
masked_store_available=false
|
||||
};
|
||||
};
|
||||
|
||||
template<> EIGEN_STRONG_INLINE Packet2cd padd<Packet2cd>(const Packet2cd& a, const Packet2cd& b) { return Packet2cd(_mm256_add_pd(a.v,b.v)); }
|
||||
template<> EIGEN_STRONG_INLINE Packet2cd psub<Packet2cd>(const Packet2cd& a, const Packet2cd& b) { return Packet2cd(_mm256_sub_pd(a.v,b.v)); }
|
||||
template<> EIGEN_STRONG_INLINE Packet2cd pnegate(const Packet2cd& a) { return Packet2cd(pnegate(a.v)); }
|
||||
template<> EIGEN_STRONG_INLINE Packet2cd pconj(const Packet2cd& a)
|
||||
{
|
||||
const __m256d mask = _mm256_castsi256_pd(_mm256_set_epi32(0x80000000,0x0,0x0,0x0,0x80000000,0x0,0x0,0x0));
|
||||
return Packet2cd(_mm256_xor_pd(a.v,mask));
|
||||
}
|
||||
|
||||
template<> EIGEN_STRONG_INLINE Packet2cd pmul<Packet2cd>(const Packet2cd& a, const Packet2cd& b)
|
||||
{
|
||||
__m256d tmp1 = _mm256_shuffle_pd(a.v,a.v,0x0);
|
||||
__m256d even = _mm256_mul_pd(tmp1, b.v);
|
||||
__m256d tmp2 = _mm256_shuffle_pd(a.v,a.v,0xF);
|
||||
__m256d tmp3 = _mm256_shuffle_pd(b.v,b.v,0x5);
|
||||
__m256d odd = _mm256_mul_pd(tmp2, tmp3);
|
||||
return Packet2cd(_mm256_addsub_pd(even, odd));
|
||||
}
|
||||
|
||||
template <>
|
||||
EIGEN_STRONG_INLINE Packet2cd pcmp_eq(const Packet2cd& a, const Packet2cd& b) {
|
||||
__m256d eq = _mm256_cmp_pd(a.v, b.v, _CMP_EQ_OQ);
|
||||
return Packet2cd(pand(eq, _mm256_permute_pd(eq, 0x5)));
|
||||
}
|
||||
|
||||
template<> EIGEN_STRONG_INLINE Packet2cd ptrue<Packet2cd>(const Packet2cd& a) { return Packet2cd(ptrue(Packet4d(a.v))); }
|
||||
template<> EIGEN_STRONG_INLINE Packet2cd pand <Packet2cd>(const Packet2cd& a, const Packet2cd& b) { return Packet2cd(_mm256_and_pd(a.v,b.v)); }
|
||||
template<> EIGEN_STRONG_INLINE Packet2cd por <Packet2cd>(const Packet2cd& a, const Packet2cd& b) { return Packet2cd(_mm256_or_pd(a.v,b.v)); }
|
||||
template<> EIGEN_STRONG_INLINE Packet2cd pxor <Packet2cd>(const Packet2cd& a, const Packet2cd& b) { return Packet2cd(_mm256_xor_pd(a.v,b.v)); }
|
||||
template<> EIGEN_STRONG_INLINE Packet2cd pandnot<Packet2cd>(const Packet2cd& a, const Packet2cd& b) { return Packet2cd(_mm256_andnot_pd(b.v,a.v)); }
|
||||
|
||||
template<> EIGEN_STRONG_INLINE Packet2cd pload <Packet2cd>(const std::complex<double>* from)
|
||||
{ EIGEN_DEBUG_ALIGNED_LOAD return Packet2cd(pload<Packet4d>((const double*)from)); }
|
||||
template<> EIGEN_STRONG_INLINE Packet2cd ploadu<Packet2cd>(const std::complex<double>* from)
|
||||
{ EIGEN_DEBUG_UNALIGNED_LOAD return Packet2cd(ploadu<Packet4d>((const double*)from)); }
|
||||
|
||||
template<> EIGEN_STRONG_INLINE Packet2cd pset1<Packet2cd>(const std::complex<double>& from)
|
||||
{
|
||||
// in case casting to a __m128d* is really not safe, then we can still fallback to this version: (much slower though)
|
||||
// return Packet2cd(_mm256_loadu2_m128d((const double*)&from,(const double*)&from));
|
||||
return Packet2cd(_mm256_broadcast_pd((const __m128d*)(const void*)&from));
|
||||
}
|
||||
|
||||
template<> EIGEN_STRONG_INLINE Packet2cd ploaddup<Packet2cd>(const std::complex<double>* from) { return pset1<Packet2cd>(*from); }
|
||||
|
||||
template<> EIGEN_STRONG_INLINE void pstore <std::complex<double> >(std::complex<double> * to, const Packet2cd& from) { EIGEN_DEBUG_ALIGNED_STORE pstore((double*)to, from.v); }
|
||||
template<> EIGEN_STRONG_INLINE void pstoreu<std::complex<double> >(std::complex<double> * to, const Packet2cd& from) { EIGEN_DEBUG_UNALIGNED_STORE pstoreu((double*)to, from.v); }
|
||||
|
||||
template<> EIGEN_DEVICE_FUNC inline Packet2cd pgather<std::complex<double>, Packet2cd>(const std::complex<double>* from, Index stride)
|
||||
{
|
||||
return Packet2cd(_mm256_set_pd(std::imag(from[1*stride]), std::real(from[1*stride]),
|
||||
std::imag(from[0*stride]), std::real(from[0*stride])));
|
||||
}
|
||||
|
||||
template<> EIGEN_DEVICE_FUNC inline void pscatter<std::complex<double>, Packet2cd>(std::complex<double>* to, const Packet2cd& from, Index stride)
|
||||
{
|
||||
__m128d low = _mm256_extractf128_pd(from.v, 0);
|
||||
to[stride*0] = std::complex<double>(_mm_cvtsd_f64(low), _mm_cvtsd_f64(_mm_shuffle_pd(low, low, 1)));
|
||||
__m128d high = _mm256_extractf128_pd(from.v, 1);
|
||||
to[stride*1] = std::complex<double>(_mm_cvtsd_f64(high), _mm_cvtsd_f64(_mm_shuffle_pd(high, high, 1)));
|
||||
}
|
||||
|
||||
template<> EIGEN_STRONG_INLINE std::complex<double> pfirst<Packet2cd>(const Packet2cd& a)
|
||||
{
|
||||
__m128d low = _mm256_extractf128_pd(a.v, 0);
|
||||
EIGEN_ALIGN16 double res[2];
|
||||
_mm_store_pd(res, low);
|
||||
return std::complex<double>(res[0],res[1]);
|
||||
}
|
||||
|
||||
template<> EIGEN_STRONG_INLINE Packet2cd preverse(const Packet2cd& a) {
|
||||
__m256d result = _mm256_permute2f128_pd(a.v, a.v, 1);
|
||||
return Packet2cd(result);
|
||||
}
|
||||
|
||||
template<> EIGEN_STRONG_INLINE std::complex<double> predux<Packet2cd>(const Packet2cd& a)
|
||||
{
|
||||
return predux(padd(Packet1cd(_mm256_extractf128_pd(a.v,0)),
|
||||
Packet1cd(_mm256_extractf128_pd(a.v,1))));
|
||||
}
|
||||
|
||||
template<> EIGEN_STRONG_INLINE std::complex<double> predux_mul<Packet2cd>(const Packet2cd& a)
|
||||
{
|
||||
return predux(pmul(Packet1cd(_mm256_extractf128_pd(a.v,0)),
|
||||
Packet1cd(_mm256_extractf128_pd(a.v,1))));
|
||||
}
|
||||
|
||||
EIGEN_MAKE_CONJ_HELPER_CPLX_REAL(Packet2cd,Packet4d)
|
||||
|
||||
template<> EIGEN_STRONG_INLINE Packet2cd pdiv<Packet2cd>(const Packet2cd& a, const Packet2cd& b)
|
||||
{
|
||||
Packet2cd num = pmul(a, pconj(b));
|
||||
__m256d tmp = _mm256_mul_pd(b.v, b.v);
|
||||
__m256d denom = _mm256_hadd_pd(tmp, tmp);
|
||||
return Packet2cd(_mm256_div_pd(num.v, denom));
|
||||
}
|
||||
|
||||
template<> EIGEN_STRONG_INLINE Packet2cd pcplxflip<Packet2cd>(const Packet2cd& x)
|
||||
{
|
||||
return Packet2cd(_mm256_shuffle_pd(x.v, x.v, 0x5));
|
||||
}
|
||||
|
||||
EIGEN_DEVICE_FUNC inline void
|
||||
ptranspose(PacketBlock<Packet4cf,4>& kernel) {
|
||||
__m256d P0 = _mm256_castps_pd(kernel.packet[0].v);
|
||||
__m256d P1 = _mm256_castps_pd(kernel.packet[1].v);
|
||||
__m256d P2 = _mm256_castps_pd(kernel.packet[2].v);
|
||||
__m256d P3 = _mm256_castps_pd(kernel.packet[3].v);
|
||||
|
||||
__m256d T0 = _mm256_shuffle_pd(P0, P1, 15);
|
||||
__m256d T1 = _mm256_shuffle_pd(P0, P1, 0);
|
||||
__m256d T2 = _mm256_shuffle_pd(P2, P3, 15);
|
||||
__m256d T3 = _mm256_shuffle_pd(P2, P3, 0);
|
||||
|
||||
kernel.packet[1].v = _mm256_castpd_ps(_mm256_permute2f128_pd(T0, T2, 32));
|
||||
kernel.packet[3].v = _mm256_castpd_ps(_mm256_permute2f128_pd(T0, T2, 49));
|
||||
kernel.packet[0].v = _mm256_castpd_ps(_mm256_permute2f128_pd(T1, T3, 32));
|
||||
kernel.packet[2].v = _mm256_castpd_ps(_mm256_permute2f128_pd(T1, T3, 49));
|
||||
}
|
||||
|
||||
EIGEN_DEVICE_FUNC inline void
|
||||
ptranspose(PacketBlock<Packet2cd,2>& kernel) {
|
||||
__m256d tmp = _mm256_permute2f128_pd(kernel.packet[0].v, kernel.packet[1].v, 0+(2<<4));
|
||||
kernel.packet[1].v = _mm256_permute2f128_pd(kernel.packet[0].v, kernel.packet[1].v, 1+(3<<4));
|
||||
kernel.packet[0].v = tmp;
|
||||
}
|
||||
|
||||
template<> EIGEN_STRONG_INLINE Packet2cd psqrt<Packet2cd>(const Packet2cd& a) {
|
||||
return psqrt_complex<Packet2cd>(a);
|
||||
}
|
||||
|
||||
template<> EIGEN_STRONG_INLINE Packet4cf psqrt<Packet4cf>(const Packet4cf& a) {
|
||||
return psqrt_complex<Packet4cf>(a);
|
||||
}
|
||||
|
||||
} // end namespace internal
|
||||
|
||||
} // end namespace Eigen
|
||||
|
||||
#endif // EIGEN_COMPLEX_AVX_H
|
||||
@@ -0,0 +1,228 @@
|
||||
// This file is part of Eigen, a lightweight C++ template library
|
||||
// for linear algebra.
|
||||
//
|
||||
// Copyright (C) 2014 Pedro Gonnet (pedro.gonnet@gmail.com)
|
||||
//
|
||||
// This Source Code Form is subject to the terms of the Mozilla
|
||||
// Public License v. 2.0. If a copy of the MPL was not distributed
|
||||
// with this file, You can obtain one at http://mozilla.org/MPL/2.0/.
|
||||
|
||||
#ifndef EIGEN_MATH_FUNCTIONS_AVX_H
|
||||
#define EIGEN_MATH_FUNCTIONS_AVX_H
|
||||
|
||||
/* The sin and cos functions of this file are loosely derived from
|
||||
* Julien Pommier's sse math library: http://gruntthepeon.free.fr/ssemath/
|
||||
*/
|
||||
|
||||
namespace Eigen {
|
||||
|
||||
namespace internal {
|
||||
|
||||
template <>
|
||||
EIGEN_DEFINE_FUNCTION_ALLOWING_MULTIPLE_DEFINITIONS EIGEN_UNUSED Packet8f
|
||||
psin<Packet8f>(const Packet8f& _x) {
|
||||
return psin_float(_x);
|
||||
}
|
||||
|
||||
template <>
|
||||
EIGEN_DEFINE_FUNCTION_ALLOWING_MULTIPLE_DEFINITIONS EIGEN_UNUSED Packet8f
|
||||
pcos<Packet8f>(const Packet8f& _x) {
|
||||
return pcos_float(_x);
|
||||
}
|
||||
|
||||
template <>
|
||||
EIGEN_DEFINE_FUNCTION_ALLOWING_MULTIPLE_DEFINITIONS EIGEN_UNUSED Packet8f
|
||||
plog<Packet8f>(const Packet8f& _x) {
|
||||
return plog_float(_x);
|
||||
}
|
||||
|
||||
template <>
|
||||
EIGEN_DEFINE_FUNCTION_ALLOWING_MULTIPLE_DEFINITIONS EIGEN_UNUSED Packet4d
|
||||
plog<Packet4d>(const Packet4d& _x) {
|
||||
return plog_double(_x);
|
||||
}
|
||||
|
||||
template <>
|
||||
EIGEN_DEFINE_FUNCTION_ALLOWING_MULTIPLE_DEFINITIONS EIGEN_UNUSED Packet8f
|
||||
plog2<Packet8f>(const Packet8f& _x) {
|
||||
return plog2_float(_x);
|
||||
}
|
||||
|
||||
template <>
|
||||
EIGEN_DEFINE_FUNCTION_ALLOWING_MULTIPLE_DEFINITIONS EIGEN_UNUSED Packet4d
|
||||
plog2<Packet4d>(const Packet4d& _x) {
|
||||
return plog2_double(_x);
|
||||
}
|
||||
|
||||
template<> EIGEN_DEFINE_FUNCTION_ALLOWING_MULTIPLE_DEFINITIONS EIGEN_UNUSED
|
||||
Packet8f plog1p<Packet8f>(const Packet8f& _x) {
|
||||
return generic_plog1p(_x);
|
||||
}
|
||||
|
||||
template<> EIGEN_DEFINE_FUNCTION_ALLOWING_MULTIPLE_DEFINITIONS EIGEN_UNUSED
|
||||
Packet8f pexpm1<Packet8f>(const Packet8f& _x) {
|
||||
return generic_expm1(_x);
|
||||
}
|
||||
|
||||
// Exponential function. Works by writing "x = m*log(2) + r" where
|
||||
// "m = floor(x/log(2)+1/2)" and "r" is the remainder. The result is then
|
||||
// "exp(x) = 2^m*exp(r)" where exp(r) is in the range [-1,1).
|
||||
template <>
|
||||
EIGEN_DEFINE_FUNCTION_ALLOWING_MULTIPLE_DEFINITIONS EIGEN_UNUSED Packet8f
|
||||
pexp<Packet8f>(const Packet8f& _x) {
|
||||
return pexp_float(_x);
|
||||
}
|
||||
|
||||
// Hyperbolic Tangent function.
|
||||
template <>
|
||||
EIGEN_DEFINE_FUNCTION_ALLOWING_MULTIPLE_DEFINITIONS EIGEN_UNUSED Packet8f
|
||||
ptanh<Packet8f>(const Packet8f& _x) {
|
||||
return internal::generic_fast_tanh_float(_x);
|
||||
}
|
||||
|
||||
// Exponential function for doubles.
|
||||
template <>
|
||||
EIGEN_DEFINE_FUNCTION_ALLOWING_MULTIPLE_DEFINITIONS EIGEN_UNUSED Packet4d
|
||||
pexp<Packet4d>(const Packet4d& _x) {
|
||||
return pexp_double(_x);
|
||||
}
|
||||
|
||||
// Functions for sqrt.
|
||||
// The EIGEN_FAST_MATH version uses the _mm_rsqrt_ps approximation and one step
|
||||
// of Newton's method, at a cost of 1-2 bits of precision as opposed to the
|
||||
// exact solution. It does not handle +inf, or denormalized numbers correctly.
|
||||
// The main advantage of this approach is not just speed, but also the fact that
|
||||
// it can be inlined and pipelined with other computations, further reducing its
|
||||
// effective latency. This is similar to Quake3's fast inverse square root.
|
||||
// For detail see here: http://www.beyond3d.com/content/articles/8/
|
||||
#if EIGEN_FAST_MATH
|
||||
template <>
|
||||
EIGEN_DEFINE_FUNCTION_ALLOWING_MULTIPLE_DEFINITIONS EIGEN_UNUSED
|
||||
Packet8f psqrt<Packet8f>(const Packet8f& _x) {
|
||||
Packet8f minus_half_x = pmul(_x, pset1<Packet8f>(-0.5f));
|
||||
Packet8f denormal_mask = pandnot(
|
||||
pcmp_lt(_x, pset1<Packet8f>((std::numeric_limits<float>::min)())),
|
||||
pcmp_lt(_x, pzero(_x)));
|
||||
|
||||
// Compute approximate reciprocal sqrt.
|
||||
Packet8f x = _mm256_rsqrt_ps(_x);
|
||||
// Do a single step of Newton's iteration.
|
||||
x = pmul(x, pmadd(minus_half_x, pmul(x,x), pset1<Packet8f>(1.5f)));
|
||||
// Flush results for denormals to zero.
|
||||
return pandnot(pmul(_x,x), denormal_mask);
|
||||
}
|
||||
|
||||
#else
|
||||
|
||||
template <> EIGEN_DEFINE_FUNCTION_ALLOWING_MULTIPLE_DEFINITIONS EIGEN_UNUSED
|
||||
Packet8f psqrt<Packet8f>(const Packet8f& _x) {
|
||||
return _mm256_sqrt_ps(_x);
|
||||
}
|
||||
|
||||
#endif
|
||||
|
||||
template <> EIGEN_DEFINE_FUNCTION_ALLOWING_MULTIPLE_DEFINITIONS EIGEN_UNUSED
|
||||
Packet4d psqrt<Packet4d>(const Packet4d& _x) {
|
||||
return _mm256_sqrt_pd(_x);
|
||||
}
|
||||
|
||||
#if EIGEN_FAST_MATH
|
||||
template<> EIGEN_DEFINE_FUNCTION_ALLOWING_MULTIPLE_DEFINITIONS EIGEN_UNUSED
|
||||
Packet8f prsqrt<Packet8f>(const Packet8f& _x) {
|
||||
_EIGEN_DECLARE_CONST_Packet8f_FROM_INT(inf, 0x7f800000);
|
||||
_EIGEN_DECLARE_CONST_Packet8f(one_point_five, 1.5f);
|
||||
_EIGEN_DECLARE_CONST_Packet8f(minus_half, -0.5f);
|
||||
_EIGEN_DECLARE_CONST_Packet8f_FROM_INT(flt_min, 0x00800000);
|
||||
|
||||
Packet8f neg_half = pmul(_x, p8f_minus_half);
|
||||
|
||||
// select only the inverse sqrt of positive normal inputs (denormals are
|
||||
// flushed to zero and cause infs as well).
|
||||
Packet8f lt_min_mask = _mm256_cmp_ps(_x, p8f_flt_min, _CMP_LT_OQ);
|
||||
Packet8f inf_mask = _mm256_cmp_ps(_x, p8f_inf, _CMP_EQ_OQ);
|
||||
Packet8f not_normal_finite_mask = _mm256_or_ps(lt_min_mask, inf_mask);
|
||||
|
||||
// Compute an approximate result using the rsqrt intrinsic.
|
||||
Packet8f y_approx = _mm256_rsqrt_ps(_x);
|
||||
|
||||
// Do a single step of Newton-Raphson iteration to improve the approximation.
|
||||
// This uses the formula y_{n+1} = y_n * (1.5 - y_n * (0.5 * x) * y_n).
|
||||
// It is essential to evaluate the inner term like this because forming
|
||||
// y_n^2 may over- or underflow.
|
||||
Packet8f y_newton = pmul(y_approx, pmadd(y_approx, pmul(neg_half, y_approx), p8f_one_point_five));
|
||||
|
||||
// Select the result of the Newton-Raphson step for positive normal arguments.
|
||||
// For other arguments, choose the output of the intrinsic. This will
|
||||
// return rsqrt(+inf) = 0, rsqrt(x) = NaN if x < 0, and rsqrt(x) = +inf if
|
||||
// x is zero or a positive denormalized float (equivalent to flushing positive
|
||||
// denormalized inputs to zero).
|
||||
return pselect<Packet8f>(not_normal_finite_mask, y_approx, y_newton);
|
||||
}
|
||||
|
||||
#else
|
||||
template <> EIGEN_DEFINE_FUNCTION_ALLOWING_MULTIPLE_DEFINITIONS EIGEN_UNUSED
|
||||
Packet8f prsqrt<Packet8f>(const Packet8f& _x) {
|
||||
_EIGEN_DECLARE_CONST_Packet8f(one, 1.0f);
|
||||
return _mm256_div_ps(p8f_one, _mm256_sqrt_ps(_x));
|
||||
}
|
||||
#endif
|
||||
|
||||
template <> EIGEN_DEFINE_FUNCTION_ALLOWING_MULTIPLE_DEFINITIONS EIGEN_UNUSED
|
||||
Packet4d prsqrt<Packet4d>(const Packet4d& _x) {
|
||||
_EIGEN_DECLARE_CONST_Packet4d(one, 1.0);
|
||||
return _mm256_div_pd(p4d_one, _mm256_sqrt_pd(_x));
|
||||
}
|
||||
|
||||
F16_PACKET_FUNCTION(Packet8f, Packet8h, psin)
|
||||
F16_PACKET_FUNCTION(Packet8f, Packet8h, pcos)
|
||||
F16_PACKET_FUNCTION(Packet8f, Packet8h, plog)
|
||||
F16_PACKET_FUNCTION(Packet8f, Packet8h, plog2)
|
||||
F16_PACKET_FUNCTION(Packet8f, Packet8h, plog1p)
|
||||
F16_PACKET_FUNCTION(Packet8f, Packet8h, pexpm1)
|
||||
F16_PACKET_FUNCTION(Packet8f, Packet8h, pexp)
|
||||
F16_PACKET_FUNCTION(Packet8f, Packet8h, ptanh)
|
||||
F16_PACKET_FUNCTION(Packet8f, Packet8h, psqrt)
|
||||
F16_PACKET_FUNCTION(Packet8f, Packet8h, prsqrt)
|
||||
|
||||
template <>
|
||||
EIGEN_STRONG_INLINE Packet8h pfrexp(const Packet8h& a, Packet8h& exponent) {
|
||||
Packet8f fexponent;
|
||||
const Packet8h out = float2half(pfrexp<Packet8f>(half2float(a), fexponent));
|
||||
exponent = float2half(fexponent);
|
||||
return out;
|
||||
}
|
||||
|
||||
template <>
|
||||
EIGEN_STRONG_INLINE Packet8h pldexp(const Packet8h& a, const Packet8h& exponent) {
|
||||
return float2half(pldexp<Packet8f>(half2float(a), half2float(exponent)));
|
||||
}
|
||||
|
||||
BF16_PACKET_FUNCTION(Packet8f, Packet8bf, psin)
|
||||
BF16_PACKET_FUNCTION(Packet8f, Packet8bf, pcos)
|
||||
BF16_PACKET_FUNCTION(Packet8f, Packet8bf, plog)
|
||||
BF16_PACKET_FUNCTION(Packet8f, Packet8bf, plog2)
|
||||
BF16_PACKET_FUNCTION(Packet8f, Packet8bf, plog1p)
|
||||
BF16_PACKET_FUNCTION(Packet8f, Packet8bf, pexpm1)
|
||||
BF16_PACKET_FUNCTION(Packet8f, Packet8bf, pexp)
|
||||
BF16_PACKET_FUNCTION(Packet8f, Packet8bf, ptanh)
|
||||
BF16_PACKET_FUNCTION(Packet8f, Packet8bf, psqrt)
|
||||
BF16_PACKET_FUNCTION(Packet8f, Packet8bf, prsqrt)
|
||||
|
||||
template <>
|
||||
EIGEN_STRONG_INLINE Packet8bf pfrexp(const Packet8bf& a, Packet8bf& exponent) {
|
||||
Packet8f fexponent;
|
||||
const Packet8bf out = F32ToBf16(pfrexp<Packet8f>(Bf16ToF32(a), fexponent));
|
||||
exponent = F32ToBf16(fexponent);
|
||||
return out;
|
||||
}
|
||||
|
||||
template <>
|
||||
EIGEN_STRONG_INLINE Packet8bf pldexp(const Packet8bf& a, const Packet8bf& exponent) {
|
||||
return F32ToBf16(pldexp<Packet8f>(Bf16ToF32(a), Bf16ToF32(exponent)));
|
||||
}
|
||||
|
||||
} // end namespace internal
|
||||
|
||||
} // end namespace Eigen
|
||||
|
||||
#endif // EIGEN_MATH_FUNCTIONS_AVX_H
|
||||
File diff suppressed because it is too large
Load Diff
@@ -0,0 +1,115 @@
|
||||
// This file is part of Eigen, a lightweight C++ template library
|
||||
// for linear algebra.
|
||||
//
|
||||
// Copyright (C) 2015 Benoit Steiner <benoit.steiner.goog@gmail.com>
|
||||
//
|
||||
// This Source Code Form is subject to the terms of the Mozilla
|
||||
// Public License v. 2.0. If a copy of the MPL was not distributed
|
||||
// with this file, You can obtain one at http://mozilla.org/MPL/2.0/.
|
||||
|
||||
#ifndef EIGEN_TYPE_CASTING_AVX_H
|
||||
#define EIGEN_TYPE_CASTING_AVX_H
|
||||
|
||||
namespace Eigen {
|
||||
|
||||
namespace internal {
|
||||
|
||||
// For now we use SSE to handle integers, so we can't use AVX instructions to cast
|
||||
// from int to float
|
||||
template <>
|
||||
struct type_casting_traits<float, int> {
|
||||
enum {
|
||||
VectorizedCast = 0,
|
||||
SrcCoeffRatio = 1,
|
||||
TgtCoeffRatio = 1
|
||||
};
|
||||
};
|
||||
|
||||
template <>
|
||||
struct type_casting_traits<int, float> {
|
||||
enum {
|
||||
VectorizedCast = 0,
|
||||
SrcCoeffRatio = 1,
|
||||
TgtCoeffRatio = 1
|
||||
};
|
||||
};
|
||||
|
||||
|
||||
#ifndef EIGEN_VECTORIZE_AVX512
|
||||
|
||||
template <>
|
||||
struct type_casting_traits<Eigen::half, float> {
|
||||
enum {
|
||||
VectorizedCast = 1,
|
||||
SrcCoeffRatio = 1,
|
||||
TgtCoeffRatio = 1
|
||||
};
|
||||
};
|
||||
|
||||
|
||||
template <>
|
||||
struct type_casting_traits<float, Eigen::half> {
|
||||
enum {
|
||||
VectorizedCast = 1,
|
||||
SrcCoeffRatio = 1,
|
||||
TgtCoeffRatio = 1
|
||||
};
|
||||
};
|
||||
|
||||
template <>
|
||||
struct type_casting_traits<bfloat16, float> {
|
||||
enum {
|
||||
VectorizedCast = 1,
|
||||
SrcCoeffRatio = 1,
|
||||
TgtCoeffRatio = 1
|
||||
};
|
||||
};
|
||||
|
||||
template <>
|
||||
struct type_casting_traits<float, bfloat16> {
|
||||
enum {
|
||||
VectorizedCast = 1,
|
||||
SrcCoeffRatio = 1,
|
||||
TgtCoeffRatio = 1
|
||||
};
|
||||
};
|
||||
|
||||
#endif // EIGEN_VECTORIZE_AVX512
|
||||
|
||||
template<> EIGEN_STRONG_INLINE Packet8i pcast<Packet8f, Packet8i>(const Packet8f& a) {
|
||||
return _mm256_cvttps_epi32(a);
|
||||
}
|
||||
|
||||
template<> EIGEN_STRONG_INLINE Packet8f pcast<Packet8i, Packet8f>(const Packet8i& a) {
|
||||
return _mm256_cvtepi32_ps(a);
|
||||
}
|
||||
|
||||
template<> EIGEN_STRONG_INLINE Packet8i preinterpret<Packet8i,Packet8f>(const Packet8f& a) {
|
||||
return _mm256_castps_si256(a);
|
||||
}
|
||||
|
||||
template<> EIGEN_STRONG_INLINE Packet8f preinterpret<Packet8f,Packet8i>(const Packet8i& a) {
|
||||
return _mm256_castsi256_ps(a);
|
||||
}
|
||||
|
||||
template<> EIGEN_STRONG_INLINE Packet8f pcast<Packet8h, Packet8f>(const Packet8h& a) {
|
||||
return half2float(a);
|
||||
}
|
||||
|
||||
template<> EIGEN_STRONG_INLINE Packet8f pcast<Packet8bf, Packet8f>(const Packet8bf& a) {
|
||||
return Bf16ToF32(a);
|
||||
}
|
||||
|
||||
template<> EIGEN_STRONG_INLINE Packet8h pcast<Packet8f, Packet8h>(const Packet8f& a) {
|
||||
return float2half(a);
|
||||
}
|
||||
|
||||
template<> EIGEN_STRONG_INLINE Packet8bf pcast<Packet8f, Packet8bf>(const Packet8f& a) {
|
||||
return F32ToBf16(a);
|
||||
}
|
||||
|
||||
} // end namespace internal
|
||||
|
||||
} // end namespace Eigen
|
||||
|
||||
#endif // EIGEN_TYPE_CASTING_AVX_H
|
||||
@@ -0,0 +1,422 @@
|
||||
// This file is part of Eigen, a lightweight C++ template library
|
||||
// for linear algebra.
|
||||
//
|
||||
// Copyright (C) 2018 Gael Guennebaud <gael.guennebaud@inria.fr>
|
||||
//
|
||||
// This Source Code Form is subject to the terms of the Mozilla
|
||||
// Public License v. 2.0. If a copy of the MPL was not distributed
|
||||
// with this file, You can obtain one at http://mozilla.org/MPL/2.0/.
|
||||
|
||||
#ifndef EIGEN_COMPLEX_AVX512_H
|
||||
#define EIGEN_COMPLEX_AVX512_H
|
||||
|
||||
namespace Eigen {
|
||||
|
||||
namespace internal {
|
||||
|
||||
//---------- float ----------
|
||||
struct Packet8cf
|
||||
{
|
||||
EIGEN_STRONG_INLINE Packet8cf() {}
|
||||
EIGEN_STRONG_INLINE explicit Packet8cf(const __m512& a) : v(a) {}
|
||||
__m512 v;
|
||||
};
|
||||
|
||||
template<> struct packet_traits<std::complex<float> > : default_packet_traits
|
||||
{
|
||||
typedef Packet8cf type;
|
||||
typedef Packet4cf half;
|
||||
enum {
|
||||
Vectorizable = 1,
|
||||
AlignedOnScalar = 1,
|
||||
size = 8,
|
||||
HasHalfPacket = 1,
|
||||
|
||||
HasAdd = 1,
|
||||
HasSub = 1,
|
||||
HasMul = 1,
|
||||
HasDiv = 1,
|
||||
HasNegate = 1,
|
||||
HasSqrt = 1,
|
||||
HasAbs = 0,
|
||||
HasAbs2 = 0,
|
||||
HasMin = 0,
|
||||
HasMax = 0,
|
||||
HasSetLinear = 0
|
||||
};
|
||||
};
|
||||
|
||||
template<> struct unpacket_traits<Packet8cf> {
|
||||
typedef std::complex<float> type;
|
||||
typedef Packet4cf half;
|
||||
typedef Packet16f as_real;
|
||||
enum {
|
||||
size = 8,
|
||||
alignment=unpacket_traits<Packet16f>::alignment,
|
||||
vectorizable=true,
|
||||
masked_load_available=false,
|
||||
masked_store_available=false
|
||||
};
|
||||
};
|
||||
|
||||
template<> EIGEN_STRONG_INLINE Packet8cf ptrue<Packet8cf>(const Packet8cf& a) { return Packet8cf(ptrue(Packet16f(a.v))); }
|
||||
template<> EIGEN_STRONG_INLINE Packet8cf padd<Packet8cf>(const Packet8cf& a, const Packet8cf& b) { return Packet8cf(_mm512_add_ps(a.v,b.v)); }
|
||||
template<> EIGEN_STRONG_INLINE Packet8cf psub<Packet8cf>(const Packet8cf& a, const Packet8cf& b) { return Packet8cf(_mm512_sub_ps(a.v,b.v)); }
|
||||
template<> EIGEN_STRONG_INLINE Packet8cf pnegate(const Packet8cf& a)
|
||||
{
|
||||
return Packet8cf(pnegate(a.v));
|
||||
}
|
||||
template<> EIGEN_STRONG_INLINE Packet8cf pconj(const Packet8cf& a)
|
||||
{
|
||||
const __m512 mask = _mm512_castsi512_ps(_mm512_setr_epi32(
|
||||
0x00000000,0x80000000,0x00000000,0x80000000,0x00000000,0x80000000,0x00000000,0x80000000,
|
||||
0x00000000,0x80000000,0x00000000,0x80000000,0x00000000,0x80000000,0x00000000,0x80000000));
|
||||
return Packet8cf(pxor(a.v,mask));
|
||||
}
|
||||
|
||||
template<> EIGEN_STRONG_INLINE Packet8cf pmul<Packet8cf>(const Packet8cf& a, const Packet8cf& b)
|
||||
{
|
||||
__m512 tmp2 = _mm512_mul_ps(_mm512_movehdup_ps(a.v), _mm512_permute_ps(b.v, _MM_SHUFFLE(2,3,0,1)));
|
||||
return Packet8cf(_mm512_fmaddsub_ps(_mm512_moveldup_ps(a.v), b.v, tmp2));
|
||||
}
|
||||
|
||||
template<> EIGEN_STRONG_INLINE Packet8cf pand <Packet8cf>(const Packet8cf& a, const Packet8cf& b) { return Packet8cf(pand(a.v,b.v)); }
|
||||
template<> EIGEN_STRONG_INLINE Packet8cf por <Packet8cf>(const Packet8cf& a, const Packet8cf& b) { return Packet8cf(por(a.v,b.v)); }
|
||||
template<> EIGEN_STRONG_INLINE Packet8cf pxor <Packet8cf>(const Packet8cf& a, const Packet8cf& b) { return Packet8cf(pxor(a.v,b.v)); }
|
||||
template<> EIGEN_STRONG_INLINE Packet8cf pandnot<Packet8cf>(const Packet8cf& a, const Packet8cf& b) { return Packet8cf(pandnot(a.v,b.v)); }
|
||||
|
||||
template <>
|
||||
EIGEN_STRONG_INLINE Packet8cf pcmp_eq(const Packet8cf& a, const Packet8cf& b) {
|
||||
__m512 eq = pcmp_eq<Packet16f>(a.v, b.v);
|
||||
return Packet8cf(pand(eq, _mm512_permute_ps(eq, 0xB1)));
|
||||
}
|
||||
|
||||
template<> EIGEN_STRONG_INLINE Packet8cf pload <Packet8cf>(const std::complex<float>* from) { EIGEN_DEBUG_ALIGNED_LOAD return Packet8cf(pload<Packet16f>(&numext::real_ref(*from))); }
|
||||
template<> EIGEN_STRONG_INLINE Packet8cf ploadu<Packet8cf>(const std::complex<float>* from) { EIGEN_DEBUG_UNALIGNED_LOAD return Packet8cf(ploadu<Packet16f>(&numext::real_ref(*from))); }
|
||||
|
||||
|
||||
template<> EIGEN_STRONG_INLINE Packet8cf pset1<Packet8cf>(const std::complex<float>& from)
|
||||
{
|
||||
return Packet8cf(_mm512_castpd_ps(pload1<Packet8d>((const double*)(const void*)&from)));
|
||||
}
|
||||
|
||||
template<> EIGEN_STRONG_INLINE Packet8cf ploaddup<Packet8cf>(const std::complex<float>* from)
|
||||
{
|
||||
return Packet8cf( _mm512_castpd_ps( ploaddup<Packet8d>((const double*)(const void*)from )) );
|
||||
}
|
||||
template<> EIGEN_STRONG_INLINE Packet8cf ploadquad<Packet8cf>(const std::complex<float>* from)
|
||||
{
|
||||
return Packet8cf( _mm512_castpd_ps( ploadquad<Packet8d>((const double*)(const void*)from )) );
|
||||
}
|
||||
|
||||
template<> EIGEN_STRONG_INLINE void pstore <std::complex<float> >(std::complex<float>* to, const Packet8cf& from) { EIGEN_DEBUG_ALIGNED_STORE pstore(&numext::real_ref(*to), from.v); }
|
||||
template<> EIGEN_STRONG_INLINE void pstoreu<std::complex<float> >(std::complex<float>* to, const Packet8cf& from) { EIGEN_DEBUG_UNALIGNED_STORE pstoreu(&numext::real_ref(*to), from.v); }
|
||||
|
||||
template<> EIGEN_DEVICE_FUNC inline Packet8cf pgather<std::complex<float>, Packet8cf>(const std::complex<float>* from, Index stride)
|
||||
{
|
||||
return Packet8cf(_mm512_castpd_ps(pgather<double,Packet8d>((const double*)(const void*)from, stride)));
|
||||
}
|
||||
|
||||
template<> EIGEN_DEVICE_FUNC inline void pscatter<std::complex<float>, Packet8cf>(std::complex<float>* to, const Packet8cf& from, Index stride)
|
||||
{
|
||||
pscatter((double*)(void*)to, _mm512_castps_pd(from.v), stride);
|
||||
}
|
||||
|
||||
template<> EIGEN_STRONG_INLINE std::complex<float> pfirst<Packet8cf>(const Packet8cf& a)
|
||||
{
|
||||
return pfirst(Packet2cf(_mm512_castps512_ps128(a.v)));
|
||||
}
|
||||
|
||||
template<> EIGEN_STRONG_INLINE Packet8cf preverse(const Packet8cf& a) {
|
||||
return Packet8cf(_mm512_castsi512_ps(
|
||||
_mm512_permutexvar_epi64( _mm512_set_epi32(0, 0, 0, 1, 0, 2, 0, 3, 0, 4, 0, 5, 0, 6, 0, 7),
|
||||
_mm512_castps_si512(a.v))));
|
||||
}
|
||||
|
||||
template<> EIGEN_STRONG_INLINE std::complex<float> predux<Packet8cf>(const Packet8cf& a)
|
||||
{
|
||||
return predux(padd(Packet4cf(extract256<0>(a.v)),
|
||||
Packet4cf(extract256<1>(a.v))));
|
||||
}
|
||||
|
||||
template<> EIGEN_STRONG_INLINE std::complex<float> predux_mul<Packet8cf>(const Packet8cf& a)
|
||||
{
|
||||
return predux_mul(pmul(Packet4cf(extract256<0>(a.v)),
|
||||
Packet4cf(extract256<1>(a.v))));
|
||||
}
|
||||
|
||||
template <>
|
||||
EIGEN_STRONG_INLINE Packet4cf predux_half_dowto4<Packet8cf>(const Packet8cf& a) {
|
||||
__m256 lane0 = extract256<0>(a.v);
|
||||
__m256 lane1 = extract256<1>(a.v);
|
||||
__m256 res = _mm256_add_ps(lane0, lane1);
|
||||
return Packet4cf(res);
|
||||
}
|
||||
|
||||
EIGEN_MAKE_CONJ_HELPER_CPLX_REAL(Packet8cf,Packet16f)
|
||||
|
||||
template<> EIGEN_STRONG_INLINE Packet8cf pdiv<Packet8cf>(const Packet8cf& a, const Packet8cf& b)
|
||||
{
|
||||
Packet8cf num = pmul(a, pconj(b));
|
||||
__m512 tmp = _mm512_mul_ps(b.v, b.v);
|
||||
__m512 tmp2 = _mm512_shuffle_ps(tmp,tmp,0xB1);
|
||||
__m512 denom = _mm512_add_ps(tmp, tmp2);
|
||||
return Packet8cf(_mm512_div_ps(num.v, denom));
|
||||
}
|
||||
|
||||
template<> EIGEN_STRONG_INLINE Packet8cf pcplxflip<Packet8cf>(const Packet8cf& x)
|
||||
{
|
||||
return Packet8cf(_mm512_shuffle_ps(x.v, x.v, _MM_SHUFFLE(2, 3, 0 ,1)));
|
||||
}
|
||||
|
||||
//---------- double ----------
|
||||
struct Packet4cd
|
||||
{
|
||||
EIGEN_STRONG_INLINE Packet4cd() {}
|
||||
EIGEN_STRONG_INLINE explicit Packet4cd(const __m512d& a) : v(a) {}
|
||||
__m512d v;
|
||||
};
|
||||
|
||||
template<> struct packet_traits<std::complex<double> > : default_packet_traits
|
||||
{
|
||||
typedef Packet4cd type;
|
||||
typedef Packet2cd half;
|
||||
enum {
|
||||
Vectorizable = 1,
|
||||
AlignedOnScalar = 0,
|
||||
size = 4,
|
||||
HasHalfPacket = 1,
|
||||
|
||||
HasAdd = 1,
|
||||
HasSub = 1,
|
||||
HasMul = 1,
|
||||
HasDiv = 1,
|
||||
HasNegate = 1,
|
||||
HasSqrt = 1,
|
||||
HasAbs = 0,
|
||||
HasAbs2 = 0,
|
||||
HasMin = 0,
|
||||
HasMax = 0,
|
||||
HasSetLinear = 0
|
||||
};
|
||||
};
|
||||
|
||||
template<> struct unpacket_traits<Packet4cd> {
|
||||
typedef std::complex<double> type;
|
||||
typedef Packet2cd half;
|
||||
typedef Packet8d as_real;
|
||||
enum {
|
||||
size = 4,
|
||||
alignment = unpacket_traits<Packet8d>::alignment,
|
||||
vectorizable=true,
|
||||
masked_load_available=false,
|
||||
masked_store_available=false
|
||||
};
|
||||
};
|
||||
|
||||
template<> EIGEN_STRONG_INLINE Packet4cd padd<Packet4cd>(const Packet4cd& a, const Packet4cd& b) { return Packet4cd(_mm512_add_pd(a.v,b.v)); }
|
||||
template<> EIGEN_STRONG_INLINE Packet4cd psub<Packet4cd>(const Packet4cd& a, const Packet4cd& b) { return Packet4cd(_mm512_sub_pd(a.v,b.v)); }
|
||||
template<> EIGEN_STRONG_INLINE Packet4cd pnegate(const Packet4cd& a) { return Packet4cd(pnegate(a.v)); }
|
||||
template<> EIGEN_STRONG_INLINE Packet4cd pconj(const Packet4cd& a)
|
||||
{
|
||||
const __m512d mask = _mm512_castsi512_pd(
|
||||
_mm512_set_epi32(0x80000000,0x0,0x0,0x0,0x80000000,0x0,0x0,0x0,
|
||||
0x80000000,0x0,0x0,0x0,0x80000000,0x0,0x0,0x0));
|
||||
return Packet4cd(pxor(a.v,mask));
|
||||
}
|
||||
|
||||
template<> EIGEN_STRONG_INLINE Packet4cd pmul<Packet4cd>(const Packet4cd& a, const Packet4cd& b)
|
||||
{
|
||||
__m512d tmp1 = _mm512_shuffle_pd(a.v,a.v,0x0);
|
||||
__m512d tmp2 = _mm512_shuffle_pd(a.v,a.v,0xFF);
|
||||
__m512d tmp3 = _mm512_shuffle_pd(b.v,b.v,0x55);
|
||||
__m512d odd = _mm512_mul_pd(tmp2, tmp3);
|
||||
return Packet4cd(_mm512_fmaddsub_pd(tmp1, b.v, odd));
|
||||
}
|
||||
|
||||
template<> EIGEN_STRONG_INLINE Packet4cd ptrue<Packet4cd>(const Packet4cd& a) { return Packet4cd(ptrue(Packet8d(a.v))); }
|
||||
template<> EIGEN_STRONG_INLINE Packet4cd pand <Packet4cd>(const Packet4cd& a, const Packet4cd& b) { return Packet4cd(pand(a.v,b.v)); }
|
||||
template<> EIGEN_STRONG_INLINE Packet4cd por <Packet4cd>(const Packet4cd& a, const Packet4cd& b) { return Packet4cd(por(a.v,b.v)); }
|
||||
template<> EIGEN_STRONG_INLINE Packet4cd pxor <Packet4cd>(const Packet4cd& a, const Packet4cd& b) { return Packet4cd(pxor(a.v,b.v)); }
|
||||
template<> EIGEN_STRONG_INLINE Packet4cd pandnot<Packet4cd>(const Packet4cd& a, const Packet4cd& b) { return Packet4cd(pandnot(a.v,b.v)); }
|
||||
|
||||
template <>
|
||||
EIGEN_STRONG_INLINE Packet4cd pcmp_eq(const Packet4cd& a, const Packet4cd& b) {
|
||||
__m512d eq = pcmp_eq<Packet8d>(a.v, b.v);
|
||||
return Packet4cd(pand(eq, _mm512_permute_pd(eq, 0x55)));
|
||||
}
|
||||
|
||||
template<> EIGEN_STRONG_INLINE Packet4cd pload <Packet4cd>(const std::complex<double>* from)
|
||||
{ EIGEN_DEBUG_ALIGNED_LOAD return Packet4cd(pload<Packet8d>((const double*)from)); }
|
||||
template<> EIGEN_STRONG_INLINE Packet4cd ploadu<Packet4cd>(const std::complex<double>* from)
|
||||
{ EIGEN_DEBUG_UNALIGNED_LOAD return Packet4cd(ploadu<Packet8d>((const double*)from)); }
|
||||
|
||||
template<> EIGEN_STRONG_INLINE Packet4cd pset1<Packet4cd>(const std::complex<double>& from)
|
||||
{
|
||||
#ifdef EIGEN_VECTORIZE_AVX512DQ
|
||||
return Packet4cd(_mm512_broadcast_f64x2(pset1<Packet1cd>(from).v));
|
||||
#else
|
||||
return Packet4cd(_mm512_castps_pd(_mm512_broadcast_f32x4( _mm_castpd_ps(pset1<Packet1cd>(from).v))));
|
||||
#endif
|
||||
}
|
||||
|
||||
template<> EIGEN_STRONG_INLINE Packet4cd ploaddup<Packet4cd>(const std::complex<double>* from) {
|
||||
return Packet4cd(_mm512_insertf64x4(
|
||||
_mm512_castpd256_pd512(ploaddup<Packet2cd>(from).v), ploaddup<Packet2cd>(from+1).v, 1));
|
||||
}
|
||||
|
||||
template<> EIGEN_STRONG_INLINE void pstore <std::complex<double> >(std::complex<double> * to, const Packet4cd& from) { EIGEN_DEBUG_ALIGNED_STORE pstore((double*)to, from.v); }
|
||||
template<> EIGEN_STRONG_INLINE void pstoreu<std::complex<double> >(std::complex<double> * to, const Packet4cd& from) { EIGEN_DEBUG_UNALIGNED_STORE pstoreu((double*)to, from.v); }
|
||||
|
||||
template<> EIGEN_DEVICE_FUNC inline Packet4cd pgather<std::complex<double>, Packet4cd>(const std::complex<double>* from, Index stride)
|
||||
{
|
||||
return Packet4cd(_mm512_insertf64x4(_mm512_castpd256_pd512(
|
||||
_mm256_insertf128_pd(_mm256_castpd128_pd256(ploadu<Packet1cd>(from+0*stride).v), ploadu<Packet1cd>(from+1*stride).v,1)),
|
||||
_mm256_insertf128_pd(_mm256_castpd128_pd256(ploadu<Packet1cd>(from+2*stride).v), ploadu<Packet1cd>(from+3*stride).v,1), 1));
|
||||
}
|
||||
|
||||
template<> EIGEN_DEVICE_FUNC inline void pscatter<std::complex<double>, Packet4cd>(std::complex<double>* to, const Packet4cd& from, Index stride)
|
||||
{
|
||||
__m512i fromi = _mm512_castpd_si512(from.v);
|
||||
double* tod = (double*)(void*)to;
|
||||
_mm_storeu_pd(tod+0*stride, _mm_castsi128_pd(_mm512_extracti32x4_epi32(fromi,0)) );
|
||||
_mm_storeu_pd(tod+2*stride, _mm_castsi128_pd(_mm512_extracti32x4_epi32(fromi,1)) );
|
||||
_mm_storeu_pd(tod+4*stride, _mm_castsi128_pd(_mm512_extracti32x4_epi32(fromi,2)) );
|
||||
_mm_storeu_pd(tod+6*stride, _mm_castsi128_pd(_mm512_extracti32x4_epi32(fromi,3)) );
|
||||
}
|
||||
|
||||
template<> EIGEN_STRONG_INLINE std::complex<double> pfirst<Packet4cd>(const Packet4cd& a)
|
||||
{
|
||||
__m128d low = extract128<0>(a.v);
|
||||
EIGEN_ALIGN16 double res[2];
|
||||
_mm_store_pd(res, low);
|
||||
return std::complex<double>(res[0],res[1]);
|
||||
}
|
||||
|
||||
template<> EIGEN_STRONG_INLINE Packet4cd preverse(const Packet4cd& a) {
|
||||
return Packet4cd(_mm512_shuffle_f64x2(a.v, a.v, (shuffle_mask<3,2,1,0>::mask)));
|
||||
}
|
||||
|
||||
template<> EIGEN_STRONG_INLINE std::complex<double> predux<Packet4cd>(const Packet4cd& a)
|
||||
{
|
||||
return predux(padd(Packet2cd(_mm512_extractf64x4_pd(a.v,0)),
|
||||
Packet2cd(_mm512_extractf64x4_pd(a.v,1))));
|
||||
}
|
||||
|
||||
template<> EIGEN_STRONG_INLINE std::complex<double> predux_mul<Packet4cd>(const Packet4cd& a)
|
||||
{
|
||||
return predux_mul(pmul(Packet2cd(_mm512_extractf64x4_pd(a.v,0)),
|
||||
Packet2cd(_mm512_extractf64x4_pd(a.v,1))));
|
||||
}
|
||||
|
||||
template<> struct conj_helper<Packet4cd, Packet4cd, false,true>
|
||||
{
|
||||
EIGEN_STRONG_INLINE Packet4cd pmadd(const Packet4cd& x, const Packet4cd& y, const Packet4cd& c) const
|
||||
{ return padd(pmul(x,y),c); }
|
||||
|
||||
EIGEN_STRONG_INLINE Packet4cd pmul(const Packet4cd& a, const Packet4cd& b) const
|
||||
{
|
||||
return internal::pmul(a, pconj(b));
|
||||
}
|
||||
};
|
||||
|
||||
template<> struct conj_helper<Packet4cd, Packet4cd, true,false>
|
||||
{
|
||||
EIGEN_STRONG_INLINE Packet4cd pmadd(const Packet4cd& x, const Packet4cd& y, const Packet4cd& c) const
|
||||
{ return padd(pmul(x,y),c); }
|
||||
|
||||
EIGEN_STRONG_INLINE Packet4cd pmul(const Packet4cd& a, const Packet4cd& b) const
|
||||
{
|
||||
return internal::pmul(pconj(a), b);
|
||||
}
|
||||
};
|
||||
|
||||
template<> struct conj_helper<Packet4cd, Packet4cd, true,true>
|
||||
{
|
||||
EIGEN_STRONG_INLINE Packet4cd pmadd(const Packet4cd& x, const Packet4cd& y, const Packet4cd& c) const
|
||||
{ return padd(pmul(x,y),c); }
|
||||
|
||||
EIGEN_STRONG_INLINE Packet4cd pmul(const Packet4cd& a, const Packet4cd& b) const
|
||||
{
|
||||
return pconj(internal::pmul(a, b));
|
||||
}
|
||||
};
|
||||
|
||||
EIGEN_MAKE_CONJ_HELPER_CPLX_REAL(Packet4cd,Packet8d)
|
||||
|
||||
template<> EIGEN_STRONG_INLINE Packet4cd pdiv<Packet4cd>(const Packet4cd& a, const Packet4cd& b)
|
||||
{
|
||||
Packet4cd num = pmul(a, pconj(b));
|
||||
__m512d tmp = _mm512_mul_pd(b.v, b.v);
|
||||
__m512d denom = padd(_mm512_permute_pd(tmp,0x55), tmp);
|
||||
return Packet4cd(_mm512_div_pd(num.v, denom));
|
||||
}
|
||||
|
||||
template<> EIGEN_STRONG_INLINE Packet4cd pcplxflip<Packet4cd>(const Packet4cd& x)
|
||||
{
|
||||
return Packet4cd(_mm512_permute_pd(x.v,0x55));
|
||||
}
|
||||
|
||||
EIGEN_DEVICE_FUNC inline void
|
||||
ptranspose(PacketBlock<Packet8cf,4>& kernel) {
|
||||
PacketBlock<Packet8d,4> pb;
|
||||
|
||||
pb.packet[0] = _mm512_castps_pd(kernel.packet[0].v);
|
||||
pb.packet[1] = _mm512_castps_pd(kernel.packet[1].v);
|
||||
pb.packet[2] = _mm512_castps_pd(kernel.packet[2].v);
|
||||
pb.packet[3] = _mm512_castps_pd(kernel.packet[3].v);
|
||||
ptranspose(pb);
|
||||
kernel.packet[0].v = _mm512_castpd_ps(pb.packet[0]);
|
||||
kernel.packet[1].v = _mm512_castpd_ps(pb.packet[1]);
|
||||
kernel.packet[2].v = _mm512_castpd_ps(pb.packet[2]);
|
||||
kernel.packet[3].v = _mm512_castpd_ps(pb.packet[3]);
|
||||
}
|
||||
|
||||
EIGEN_DEVICE_FUNC inline void
|
||||
ptranspose(PacketBlock<Packet8cf,8>& kernel) {
|
||||
PacketBlock<Packet8d,8> pb;
|
||||
|
||||
pb.packet[0] = _mm512_castps_pd(kernel.packet[0].v);
|
||||
pb.packet[1] = _mm512_castps_pd(kernel.packet[1].v);
|
||||
pb.packet[2] = _mm512_castps_pd(kernel.packet[2].v);
|
||||
pb.packet[3] = _mm512_castps_pd(kernel.packet[3].v);
|
||||
pb.packet[4] = _mm512_castps_pd(kernel.packet[4].v);
|
||||
pb.packet[5] = _mm512_castps_pd(kernel.packet[5].v);
|
||||
pb.packet[6] = _mm512_castps_pd(kernel.packet[6].v);
|
||||
pb.packet[7] = _mm512_castps_pd(kernel.packet[7].v);
|
||||
ptranspose(pb);
|
||||
kernel.packet[0].v = _mm512_castpd_ps(pb.packet[0]);
|
||||
kernel.packet[1].v = _mm512_castpd_ps(pb.packet[1]);
|
||||
kernel.packet[2].v = _mm512_castpd_ps(pb.packet[2]);
|
||||
kernel.packet[3].v = _mm512_castpd_ps(pb.packet[3]);
|
||||
kernel.packet[4].v = _mm512_castpd_ps(pb.packet[4]);
|
||||
kernel.packet[5].v = _mm512_castpd_ps(pb.packet[5]);
|
||||
kernel.packet[6].v = _mm512_castpd_ps(pb.packet[6]);
|
||||
kernel.packet[7].v = _mm512_castpd_ps(pb.packet[7]);
|
||||
}
|
||||
|
||||
EIGEN_DEVICE_FUNC inline void
|
||||
ptranspose(PacketBlock<Packet4cd,4>& kernel) {
|
||||
__m512d T0 = _mm512_shuffle_f64x2(kernel.packet[0].v, kernel.packet[1].v, (shuffle_mask<0,1,0,1>::mask)); // [a0 a1 b0 b1]
|
||||
__m512d T1 = _mm512_shuffle_f64x2(kernel.packet[0].v, kernel.packet[1].v, (shuffle_mask<2,3,2,3>::mask)); // [a2 a3 b2 b3]
|
||||
__m512d T2 = _mm512_shuffle_f64x2(kernel.packet[2].v, kernel.packet[3].v, (shuffle_mask<0,1,0,1>::mask)); // [c0 c1 d0 d1]
|
||||
__m512d T3 = _mm512_shuffle_f64x2(kernel.packet[2].v, kernel.packet[3].v, (shuffle_mask<2,3,2,3>::mask)); // [c2 c3 d2 d3]
|
||||
|
||||
kernel.packet[3] = Packet4cd(_mm512_shuffle_f64x2(T1, T3, (shuffle_mask<1,3,1,3>::mask))); // [a3 b3 c3 d3]
|
||||
kernel.packet[2] = Packet4cd(_mm512_shuffle_f64x2(T1, T3, (shuffle_mask<0,2,0,2>::mask))); // [a2 b2 c2 d2]
|
||||
kernel.packet[1] = Packet4cd(_mm512_shuffle_f64x2(T0, T2, (shuffle_mask<1,3,1,3>::mask))); // [a1 b1 c1 d1]
|
||||
kernel.packet[0] = Packet4cd(_mm512_shuffle_f64x2(T0, T2, (shuffle_mask<0,2,0,2>::mask))); // [a0 b0 c0 d0]
|
||||
}
|
||||
|
||||
template<> EIGEN_STRONG_INLINE Packet4cd psqrt<Packet4cd>(const Packet4cd& a) {
|
||||
return psqrt_complex<Packet4cd>(a);
|
||||
}
|
||||
|
||||
template<> EIGEN_STRONG_INLINE Packet8cf psqrt<Packet8cf>(const Packet8cf& a) {
|
||||
return psqrt_complex<Packet8cf>(a);
|
||||
}
|
||||
|
||||
} // end namespace internal
|
||||
} // end namespace Eigen
|
||||
|
||||
#endif // EIGEN_COMPLEX_AVX512_H
|
||||
@@ -0,0 +1,362 @@
|
||||
// This file is part of Eigen, a lightweight C++ template library
|
||||
// for linear algebra.
|
||||
//
|
||||
// Copyright (C) 2016 Pedro Gonnet (pedro.gonnet@gmail.com)
|
||||
//
|
||||
// This Source Code Form is subject to the terms of the Mozilla
|
||||
// Public License v. 2.0. If a copy of the MPL was not distributed
|
||||
// with this file, You can obtain one at http://mozilla.org/MPL/2.0/.
|
||||
|
||||
#ifndef THIRD_PARTY_EIGEN3_EIGEN_SRC_CORE_ARCH_AVX512_MATHFUNCTIONS_H_
|
||||
#define THIRD_PARTY_EIGEN3_EIGEN_SRC_CORE_ARCH_AVX512_MATHFUNCTIONS_H_
|
||||
|
||||
namespace Eigen {
|
||||
|
||||
namespace internal {
|
||||
|
||||
// Disable the code for older versions of gcc that don't support many of the required avx512 instrinsics.
|
||||
#if EIGEN_GNUC_AT_LEAST(5, 3) || EIGEN_COMP_CLANG || EIGEN_COMP_MSVC >= 1923
|
||||
|
||||
#define _EIGEN_DECLARE_CONST_Packet16f(NAME, X) \
|
||||
const Packet16f p16f_##NAME = pset1<Packet16f>(X)
|
||||
|
||||
#define _EIGEN_DECLARE_CONST_Packet16f_FROM_INT(NAME, X) \
|
||||
const Packet16f p16f_##NAME = preinterpret<Packet16f,Packet16i>(pset1<Packet16i>(X))
|
||||
|
||||
#define _EIGEN_DECLARE_CONST_Packet8d(NAME, X) \
|
||||
const Packet8d p8d_##NAME = pset1<Packet8d>(X)
|
||||
|
||||
#define _EIGEN_DECLARE_CONST_Packet8d_FROM_INT64(NAME, X) \
|
||||
const Packet8d p8d_##NAME = _mm512_castsi512_pd(_mm512_set1_epi64(X))
|
||||
|
||||
#define _EIGEN_DECLARE_CONST_Packet16bf(NAME, X) \
|
||||
const Packet16bf p16bf_##NAME = pset1<Packet16bf>(X)
|
||||
|
||||
#define _EIGEN_DECLARE_CONST_Packet16bf_FROM_INT(NAME, X) \
|
||||
const Packet16bf p16bf_##NAME = preinterpret<Packet16bf,Packet16i>(pset1<Packet16i>(X))
|
||||
|
||||
template <>
|
||||
EIGEN_DEFINE_FUNCTION_ALLOWING_MULTIPLE_DEFINITIONS EIGEN_UNUSED Packet16f
|
||||
plog<Packet16f>(const Packet16f& _x) {
|
||||
return plog_float(_x);
|
||||
}
|
||||
|
||||
template <>
|
||||
EIGEN_DEFINE_FUNCTION_ALLOWING_MULTIPLE_DEFINITIONS EIGEN_UNUSED Packet8d
|
||||
plog<Packet8d>(const Packet8d& _x) {
|
||||
return plog_double(_x);
|
||||
}
|
||||
|
||||
F16_PACKET_FUNCTION(Packet16f, Packet16h, plog)
|
||||
BF16_PACKET_FUNCTION(Packet16f, Packet16bf, plog)
|
||||
|
||||
template <>
|
||||
EIGEN_DEFINE_FUNCTION_ALLOWING_MULTIPLE_DEFINITIONS EIGEN_UNUSED Packet16f
|
||||
plog2<Packet16f>(const Packet16f& _x) {
|
||||
return plog2_float(_x);
|
||||
}
|
||||
|
||||
template <>
|
||||
EIGEN_DEFINE_FUNCTION_ALLOWING_MULTIPLE_DEFINITIONS EIGEN_UNUSED Packet8d
|
||||
plog2<Packet8d>(const Packet8d& _x) {
|
||||
return plog2_double(_x);
|
||||
}
|
||||
|
||||
F16_PACKET_FUNCTION(Packet16f, Packet16h, plog2)
|
||||
BF16_PACKET_FUNCTION(Packet16f, Packet16bf, plog2)
|
||||
|
||||
// Exponential function. Works by writing "x = m*log(2) + r" where
|
||||
// "m = floor(x/log(2)+1/2)" and "r" is the remainder. The result is then
|
||||
// "exp(x) = 2^m*exp(r)" where exp(r) is in the range [-1,1).
|
||||
template <>
|
||||
EIGEN_DEFINE_FUNCTION_ALLOWING_MULTIPLE_DEFINITIONS EIGEN_UNUSED Packet16f
|
||||
pexp<Packet16f>(const Packet16f& _x) {
|
||||
_EIGEN_DECLARE_CONST_Packet16f(1, 1.0f);
|
||||
_EIGEN_DECLARE_CONST_Packet16f(half, 0.5f);
|
||||
_EIGEN_DECLARE_CONST_Packet16f(127, 127.0f);
|
||||
|
||||
_EIGEN_DECLARE_CONST_Packet16f(exp_hi, 88.3762626647950f);
|
||||
_EIGEN_DECLARE_CONST_Packet16f(exp_lo, -88.3762626647949f);
|
||||
|
||||
_EIGEN_DECLARE_CONST_Packet16f(cephes_LOG2EF, 1.44269504088896341f);
|
||||
|
||||
_EIGEN_DECLARE_CONST_Packet16f(cephes_exp_p0, 1.9875691500E-4f);
|
||||
_EIGEN_DECLARE_CONST_Packet16f(cephes_exp_p1, 1.3981999507E-3f);
|
||||
_EIGEN_DECLARE_CONST_Packet16f(cephes_exp_p2, 8.3334519073E-3f);
|
||||
_EIGEN_DECLARE_CONST_Packet16f(cephes_exp_p3, 4.1665795894E-2f);
|
||||
_EIGEN_DECLARE_CONST_Packet16f(cephes_exp_p4, 1.6666665459E-1f);
|
||||
_EIGEN_DECLARE_CONST_Packet16f(cephes_exp_p5, 5.0000001201E-1f);
|
||||
|
||||
// Clamp x.
|
||||
Packet16f x = pmax(pmin(_x, p16f_exp_hi), p16f_exp_lo);
|
||||
|
||||
// Express exp(x) as exp(m*ln(2) + r), start by extracting
|
||||
// m = floor(x/ln(2) + 0.5).
|
||||
Packet16f m = _mm512_floor_ps(pmadd(x, p16f_cephes_LOG2EF, p16f_half));
|
||||
|
||||
// Get r = x - m*ln(2). Note that we can do this without losing more than one
|
||||
// ulp precision due to the FMA instruction.
|
||||
_EIGEN_DECLARE_CONST_Packet16f(nln2, -0.6931471805599453f);
|
||||
Packet16f r = _mm512_fmadd_ps(m, p16f_nln2, x);
|
||||
Packet16f r2 = pmul(r, r);
|
||||
Packet16f r3 = pmul(r2, r);
|
||||
|
||||
// Evaluate the polynomial approximant,improved by instruction-level parallelism.
|
||||
Packet16f y, y1, y2;
|
||||
y = pmadd(p16f_cephes_exp_p0, r, p16f_cephes_exp_p1);
|
||||
y1 = pmadd(p16f_cephes_exp_p3, r, p16f_cephes_exp_p4);
|
||||
y2 = padd(r, p16f_1);
|
||||
y = pmadd(y, r, p16f_cephes_exp_p2);
|
||||
y1 = pmadd(y1, r, p16f_cephes_exp_p5);
|
||||
y = pmadd(y, r3, y1);
|
||||
y = pmadd(y, r2, y2);
|
||||
|
||||
// Build emm0 = 2^m.
|
||||
Packet16i emm0 = _mm512_cvttps_epi32(padd(m, p16f_127));
|
||||
emm0 = _mm512_slli_epi32(emm0, 23);
|
||||
|
||||
// Return 2^m * exp(r).
|
||||
return pmax(pmul(y, _mm512_castsi512_ps(emm0)), _x);
|
||||
}
|
||||
|
||||
template <>
|
||||
EIGEN_DEFINE_FUNCTION_ALLOWING_MULTIPLE_DEFINITIONS EIGEN_UNUSED Packet8d
|
||||
pexp<Packet8d>(const Packet8d& _x) {
|
||||
return pexp_double(_x);
|
||||
}
|
||||
|
||||
F16_PACKET_FUNCTION(Packet16f, Packet16h, pexp)
|
||||
BF16_PACKET_FUNCTION(Packet16f, Packet16bf, pexp)
|
||||
|
||||
template <>
|
||||
EIGEN_STRONG_INLINE Packet16h pfrexp(const Packet16h& a, Packet16h& exponent) {
|
||||
Packet16f fexponent;
|
||||
const Packet16h out = float2half(pfrexp<Packet16f>(half2float(a), fexponent));
|
||||
exponent = float2half(fexponent);
|
||||
return out;
|
||||
}
|
||||
|
||||
template <>
|
||||
EIGEN_STRONG_INLINE Packet16h pldexp(const Packet16h& a, const Packet16h& exponent) {
|
||||
return float2half(pldexp<Packet16f>(half2float(a), half2float(exponent)));
|
||||
}
|
||||
|
||||
template <>
|
||||
EIGEN_STRONG_INLINE Packet16bf pfrexp(const Packet16bf& a, Packet16bf& exponent) {
|
||||
Packet16f fexponent;
|
||||
const Packet16bf out = F32ToBf16(pfrexp<Packet16f>(Bf16ToF32(a), fexponent));
|
||||
exponent = F32ToBf16(fexponent);
|
||||
return out;
|
||||
}
|
||||
|
||||
template <>
|
||||
EIGEN_STRONG_INLINE Packet16bf pldexp(const Packet16bf& a, const Packet16bf& exponent) {
|
||||
return F32ToBf16(pldexp<Packet16f>(Bf16ToF32(a), Bf16ToF32(exponent)));
|
||||
}
|
||||
|
||||
// Functions for sqrt.
|
||||
// The EIGEN_FAST_MATH version uses the _mm_rsqrt_ps approximation and one step
|
||||
// of Newton's method, at a cost of 1-2 bits of precision as opposed to the
|
||||
// exact solution. The main advantage of this approach is not just speed, but
|
||||
// also the fact that it can be inlined and pipelined with other computations,
|
||||
// further reducing its effective latency.
|
||||
#if EIGEN_FAST_MATH
|
||||
template <>
|
||||
EIGEN_DEFINE_FUNCTION_ALLOWING_MULTIPLE_DEFINITIONS EIGEN_UNUSED Packet16f
|
||||
psqrt<Packet16f>(const Packet16f& _x) {
|
||||
Packet16f neg_half = pmul(_x, pset1<Packet16f>(-.5f));
|
||||
__mmask16 denormal_mask = _mm512_kand(
|
||||
_mm512_cmp_ps_mask(_x, pset1<Packet16f>((std::numeric_limits<float>::min)()),
|
||||
_CMP_LT_OQ),
|
||||
_mm512_cmp_ps_mask(_x, _mm512_setzero_ps(), _CMP_GE_OQ));
|
||||
|
||||
Packet16f x = _mm512_rsqrt14_ps(_x);
|
||||
|
||||
// Do a single step of Newton's iteration.
|
||||
x = pmul(x, pmadd(neg_half, pmul(x, x), pset1<Packet16f>(1.5f)));
|
||||
|
||||
// Flush results for denormals to zero.
|
||||
return _mm512_mask_blend_ps(denormal_mask, pmul(_x,x), _mm512_setzero_ps());
|
||||
}
|
||||
|
||||
template <>
|
||||
EIGEN_DEFINE_FUNCTION_ALLOWING_MULTIPLE_DEFINITIONS EIGEN_UNUSED Packet8d
|
||||
psqrt<Packet8d>(const Packet8d& _x) {
|
||||
Packet8d neg_half = pmul(_x, pset1<Packet8d>(-.5));
|
||||
__mmask16 denormal_mask = _mm512_kand(
|
||||
_mm512_cmp_pd_mask(_x, pset1<Packet8d>((std::numeric_limits<double>::min)()),
|
||||
_CMP_LT_OQ),
|
||||
_mm512_cmp_pd_mask(_x, _mm512_setzero_pd(), _CMP_GE_OQ));
|
||||
|
||||
Packet8d x = _mm512_rsqrt14_pd(_x);
|
||||
|
||||
// Do a single step of Newton's iteration.
|
||||
x = pmul(x, pmadd(neg_half, pmul(x, x), pset1<Packet8d>(1.5)));
|
||||
|
||||
// Do a second step of Newton's iteration.
|
||||
x = pmul(x, pmadd(neg_half, pmul(x, x), pset1<Packet8d>(1.5)));
|
||||
|
||||
return _mm512_mask_blend_pd(denormal_mask, pmul(_x,x), _mm512_setzero_pd());
|
||||
}
|
||||
#else
|
||||
template <>
|
||||
EIGEN_STRONG_INLINE Packet16f psqrt<Packet16f>(const Packet16f& x) {
|
||||
return _mm512_sqrt_ps(x);
|
||||
}
|
||||
|
||||
template <>
|
||||
EIGEN_STRONG_INLINE Packet8d psqrt<Packet8d>(const Packet8d& x) {
|
||||
return _mm512_sqrt_pd(x);
|
||||
}
|
||||
#endif
|
||||
|
||||
F16_PACKET_FUNCTION(Packet16f, Packet16h, psqrt)
|
||||
BF16_PACKET_FUNCTION(Packet16f, Packet16bf, psqrt)
|
||||
|
||||
// prsqrt for float.
|
||||
#if defined(EIGEN_VECTORIZE_AVX512ER)
|
||||
|
||||
template <>
|
||||
EIGEN_STRONG_INLINE Packet16f prsqrt<Packet16f>(const Packet16f& x) {
|
||||
return _mm512_rsqrt28_ps(x);
|
||||
}
|
||||
#elif EIGEN_FAST_MATH
|
||||
|
||||
template <>
|
||||
EIGEN_DEFINE_FUNCTION_ALLOWING_MULTIPLE_DEFINITIONS EIGEN_UNUSED Packet16f
|
||||
prsqrt<Packet16f>(const Packet16f& _x) {
|
||||
_EIGEN_DECLARE_CONST_Packet16f_FROM_INT(inf, 0x7f800000);
|
||||
_EIGEN_DECLARE_CONST_Packet16f(one_point_five, 1.5f);
|
||||
_EIGEN_DECLARE_CONST_Packet16f(minus_half, -0.5f);
|
||||
|
||||
Packet16f neg_half = pmul(_x, p16f_minus_half);
|
||||
|
||||
// Identity infinite, negative and denormal arguments.
|
||||
__mmask16 inf_mask = _mm512_cmp_ps_mask(_x, p16f_inf, _CMP_EQ_OQ);
|
||||
__mmask16 not_pos_mask = _mm512_cmp_ps_mask(_x, _mm512_setzero_ps(), _CMP_LE_OQ);
|
||||
__mmask16 not_finite_pos_mask = not_pos_mask | inf_mask;
|
||||
|
||||
// Compute an approximate result using the rsqrt intrinsic, forcing +inf
|
||||
// for denormals for consistency with AVX and SSE implementations.
|
||||
Packet16f y_approx = _mm512_rsqrt14_ps(_x);
|
||||
|
||||
// Do a single step of Newton-Raphson iteration to improve the approximation.
|
||||
// This uses the formula y_{n+1} = y_n * (1.5 - y_n * (0.5 * x) * y_n).
|
||||
// It is essential to evaluate the inner term like this because forming
|
||||
// y_n^2 may over- or underflow.
|
||||
Packet16f y_newton = pmul(y_approx, pmadd(y_approx, pmul(neg_half, y_approx), p16f_one_point_five));
|
||||
|
||||
// Select the result of the Newton-Raphson step for positive finite arguments.
|
||||
// For other arguments, choose the output of the intrinsic. This will
|
||||
// return rsqrt(+inf) = 0, rsqrt(x) = NaN if x < 0, and rsqrt(0) = +inf.
|
||||
return _mm512_mask_blend_ps(not_finite_pos_mask, y_newton, y_approx);
|
||||
}
|
||||
#else
|
||||
|
||||
template <>
|
||||
EIGEN_STRONG_INLINE Packet16f prsqrt<Packet16f>(const Packet16f& x) {
|
||||
_EIGEN_DECLARE_CONST_Packet16f(one, 1.0f);
|
||||
return _mm512_div_ps(p16f_one, _mm512_sqrt_ps(x));
|
||||
}
|
||||
#endif
|
||||
|
||||
F16_PACKET_FUNCTION(Packet16f, Packet16h, prsqrt)
|
||||
BF16_PACKET_FUNCTION(Packet16f, Packet16bf, prsqrt)
|
||||
|
||||
// prsqrt for double.
|
||||
#if EIGEN_FAST_MATH
|
||||
template <>
|
||||
EIGEN_DEFINE_FUNCTION_ALLOWING_MULTIPLE_DEFINITIONS EIGEN_UNUSED Packet8d
|
||||
prsqrt<Packet8d>(const Packet8d& _x) {
|
||||
_EIGEN_DECLARE_CONST_Packet8d(one_point_five, 1.5);
|
||||
_EIGEN_DECLARE_CONST_Packet8d(minus_half, -0.5);
|
||||
_EIGEN_DECLARE_CONST_Packet8d_FROM_INT64(inf, 0x7ff0000000000000LL);
|
||||
|
||||
Packet8d neg_half = pmul(_x, p8d_minus_half);
|
||||
|
||||
// Identity infinite, negative and denormal arguments.
|
||||
__mmask8 inf_mask = _mm512_cmp_pd_mask(_x, p8d_inf, _CMP_EQ_OQ);
|
||||
__mmask8 not_pos_mask = _mm512_cmp_pd_mask(_x, _mm512_setzero_pd(), _CMP_LE_OQ);
|
||||
__mmask8 not_finite_pos_mask = not_pos_mask | inf_mask;
|
||||
|
||||
// Compute an approximate result using the rsqrt intrinsic, forcing +inf
|
||||
// for denormals for consistency with AVX and SSE implementations.
|
||||
#if defined(EIGEN_VECTORIZE_AVX512ER)
|
||||
Packet8d y_approx = _mm512_rsqrt28_pd(_x);
|
||||
#else
|
||||
Packet8d y_approx = _mm512_rsqrt14_pd(_x);
|
||||
#endif
|
||||
// Do one or two steps of Newton-Raphson's to improve the approximation, depending on the
|
||||
// starting accuracy (either 2^-14 or 2^-28, depending on whether AVX512ER is available).
|
||||
// The Newton-Raphson algorithm has quadratic convergence and roughly doubles the number
|
||||
// of correct digits for each step.
|
||||
// This uses the formula y_{n+1} = y_n * (1.5 - y_n * (0.5 * x) * y_n).
|
||||
// It is essential to evaluate the inner term like this because forming
|
||||
// y_n^2 may over- or underflow.
|
||||
Packet8d y_newton = pmul(y_approx, pmadd(neg_half, pmul(y_approx, y_approx), p8d_one_point_five));
|
||||
#if !defined(EIGEN_VECTORIZE_AVX512ER)
|
||||
y_newton = pmul(y_newton, pmadd(y_newton, pmul(neg_half, y_newton), p8d_one_point_five));
|
||||
#endif
|
||||
// Select the result of the Newton-Raphson step for positive finite arguments.
|
||||
// For other arguments, choose the output of the intrinsic. This will
|
||||
// return rsqrt(+inf) = 0, rsqrt(x) = NaN if x < 0, and rsqrt(0) = +inf.
|
||||
return _mm512_mask_blend_pd(not_finite_pos_mask, y_newton, y_approx);
|
||||
}
|
||||
#else
|
||||
template <>
|
||||
EIGEN_STRONG_INLINE Packet8d prsqrt<Packet8d>(const Packet8d& x) {
|
||||
_EIGEN_DECLARE_CONST_Packet8d(one, 1.0f);
|
||||
return _mm512_div_pd(p8d_one, _mm512_sqrt_pd(x));
|
||||
}
|
||||
#endif
|
||||
|
||||
template<> EIGEN_DEFINE_FUNCTION_ALLOWING_MULTIPLE_DEFINITIONS EIGEN_UNUSED
|
||||
Packet16f plog1p<Packet16f>(const Packet16f& _x) {
|
||||
return generic_plog1p(_x);
|
||||
}
|
||||
|
||||
F16_PACKET_FUNCTION(Packet16f, Packet16h, plog1p)
|
||||
BF16_PACKET_FUNCTION(Packet16f, Packet16bf, plog1p)
|
||||
|
||||
template<> EIGEN_DEFINE_FUNCTION_ALLOWING_MULTIPLE_DEFINITIONS EIGEN_UNUSED
|
||||
Packet16f pexpm1<Packet16f>(const Packet16f& _x) {
|
||||
return generic_expm1(_x);
|
||||
}
|
||||
|
||||
F16_PACKET_FUNCTION(Packet16f, Packet16h, pexpm1)
|
||||
BF16_PACKET_FUNCTION(Packet16f, Packet16bf, pexpm1)
|
||||
|
||||
#endif
|
||||
|
||||
|
||||
template <>
|
||||
EIGEN_DEFINE_FUNCTION_ALLOWING_MULTIPLE_DEFINITIONS EIGEN_UNUSED Packet16f
|
||||
psin<Packet16f>(const Packet16f& _x) {
|
||||
return psin_float(_x);
|
||||
}
|
||||
|
||||
template <>
|
||||
EIGEN_DEFINE_FUNCTION_ALLOWING_MULTIPLE_DEFINITIONS EIGEN_UNUSED Packet16f
|
||||
pcos<Packet16f>(const Packet16f& _x) {
|
||||
return pcos_float(_x);
|
||||
}
|
||||
|
||||
template <>
|
||||
EIGEN_DEFINE_FUNCTION_ALLOWING_MULTIPLE_DEFINITIONS EIGEN_UNUSED Packet16f
|
||||
ptanh<Packet16f>(const Packet16f& _x) {
|
||||
return internal::generic_fast_tanh_float(_x);
|
||||
}
|
||||
|
||||
F16_PACKET_FUNCTION(Packet16f, Packet16h, psin)
|
||||
F16_PACKET_FUNCTION(Packet16f, Packet16h, pcos)
|
||||
F16_PACKET_FUNCTION(Packet16f, Packet16h, ptanh)
|
||||
|
||||
BF16_PACKET_FUNCTION(Packet16f, Packet16bf, psin)
|
||||
BF16_PACKET_FUNCTION(Packet16f, Packet16bf, pcos)
|
||||
BF16_PACKET_FUNCTION(Packet16f, Packet16bf, ptanh)
|
||||
|
||||
} // end namespace internal
|
||||
|
||||
} // end namespace Eigen
|
||||
|
||||
#endif // THIRD_PARTY_EIGEN3_EIGEN_SRC_CORE_ARCH_AVX512_MATHFUNCTIONS_H_
|
||||
File diff suppressed because it is too large
Load Diff
@@ -0,0 +1,89 @@
|
||||
// This file is part of Eigen, a lightweight C++ template library
|
||||
// for linear algebra.
|
||||
//
|
||||
// Copyright (C) 2019 Rasmus Munk Larsen <rmlarsen@google.com>
|
||||
//
|
||||
// This Source Code Form is subject to the terms of the Mozilla
|
||||
// Public License v. 2.0. If a copy of the MPL was not distributed
|
||||
// with this file, You can obtain one at http://mozilla.org/MPL/2.0/.
|
||||
|
||||
#ifndef EIGEN_TYPE_CASTING_AVX512_H
|
||||
#define EIGEN_TYPE_CASTING_AVX512_H
|
||||
|
||||
namespace Eigen {
|
||||
|
||||
namespace internal {
|
||||
|
||||
template<> EIGEN_STRONG_INLINE Packet16i pcast<Packet16f, Packet16i>(const Packet16f& a) {
|
||||
return _mm512_cvttps_epi32(a);
|
||||
}
|
||||
|
||||
template<> EIGEN_STRONG_INLINE Packet16f pcast<Packet16i, Packet16f>(const Packet16i& a) {
|
||||
return _mm512_cvtepi32_ps(a);
|
||||
}
|
||||
|
||||
template<> EIGEN_STRONG_INLINE Packet16i preinterpret<Packet16i, Packet16f>(const Packet16f& a) {
|
||||
return _mm512_castps_si512(a);
|
||||
}
|
||||
|
||||
template<> EIGEN_STRONG_INLINE Packet16f preinterpret<Packet16f, Packet16i>(const Packet16i& a) {
|
||||
return _mm512_castsi512_ps(a);
|
||||
}
|
||||
|
||||
template <>
|
||||
struct type_casting_traits<half, float> {
|
||||
enum {
|
||||
VectorizedCast = 1,
|
||||
SrcCoeffRatio = 1,
|
||||
TgtCoeffRatio = 1
|
||||
};
|
||||
};
|
||||
|
||||
template<> EIGEN_STRONG_INLINE Packet16f pcast<Packet16h, Packet16f>(const Packet16h& a) {
|
||||
return half2float(a);
|
||||
}
|
||||
|
||||
template <>
|
||||
struct type_casting_traits<float, half> {
|
||||
enum {
|
||||
VectorizedCast = 1,
|
||||
SrcCoeffRatio = 1,
|
||||
TgtCoeffRatio = 1
|
||||
};
|
||||
};
|
||||
|
||||
template<> EIGEN_STRONG_INLINE Packet16h pcast<Packet16f, Packet16h>(const Packet16f& a) {
|
||||
return float2half(a);
|
||||
}
|
||||
|
||||
template <>
|
||||
struct type_casting_traits<bfloat16, float> {
|
||||
enum {
|
||||
VectorizedCast = 1,
|
||||
SrcCoeffRatio = 1,
|
||||
TgtCoeffRatio = 1
|
||||
};
|
||||
};
|
||||
|
||||
template<> EIGEN_STRONG_INLINE Packet16f pcast<Packet16bf, Packet16f>(const Packet16bf& a) {
|
||||
return Bf16ToF32(a);
|
||||
}
|
||||
|
||||
template <>
|
||||
struct type_casting_traits<float, bfloat16> {
|
||||
enum {
|
||||
VectorizedCast = 1,
|
||||
SrcCoeffRatio = 1,
|
||||
TgtCoeffRatio = 1
|
||||
};
|
||||
};
|
||||
|
||||
template<> EIGEN_STRONG_INLINE Packet16bf pcast<Packet16f, Packet16bf>(const Packet16f& a) {
|
||||
return F32ToBf16(a);
|
||||
}
|
||||
|
||||
} // end namespace internal
|
||||
|
||||
} // end namespace Eigen
|
||||
|
||||
#endif // EIGEN_TYPE_CASTING_AVX512_H
|
||||
@@ -0,0 +1,417 @@
|
||||
// This file is part of Eigen, a lightweight C++ template library
|
||||
// for linear algebra.
|
||||
//
|
||||
// Copyright (C) 2010 Gael Guennebaud <gael.guennebaud@inria.fr>
|
||||
// Copyright (C) 2010-2016 Konstantinos Margaritis <markos@freevec.org>
|
||||
//
|
||||
// This Source Code Form is subject to the terms of the Mozilla
|
||||
// Public License v. 2.0. If a copy of the MPL was not distributed
|
||||
// with this file, You can obtain one at http://mozilla.org/MPL/2.0/.
|
||||
|
||||
#ifndef EIGEN_COMPLEX32_ALTIVEC_H
|
||||
#define EIGEN_COMPLEX32_ALTIVEC_H
|
||||
|
||||
namespace Eigen {
|
||||
|
||||
namespace internal {
|
||||
|
||||
static Packet4ui p4ui_CONJ_XOR = vec_mergeh((Packet4ui)p4i_ZERO, (Packet4ui)p4f_MZERO);//{ 0x00000000, 0x80000000, 0x00000000, 0x80000000 };
|
||||
#ifdef __VSX__
|
||||
#if defined(_BIG_ENDIAN)
|
||||
static Packet2ul p2ul_CONJ_XOR1 = (Packet2ul) vec_sld((Packet4ui) p2d_MZERO, (Packet4ui) p2l_ZERO, 8);//{ 0x8000000000000000, 0x0000000000000000 };
|
||||
static Packet2ul p2ul_CONJ_XOR2 = (Packet2ul) vec_sld((Packet4ui) p2l_ZERO, (Packet4ui) p2d_MZERO, 8);//{ 0x8000000000000000, 0x0000000000000000 };
|
||||
#else
|
||||
static Packet2ul p2ul_CONJ_XOR1 = (Packet2ul) vec_sld((Packet4ui) p2l_ZERO, (Packet4ui) p2d_MZERO, 8);//{ 0x8000000000000000, 0x0000000000000000 };
|
||||
static Packet2ul p2ul_CONJ_XOR2 = (Packet2ul) vec_sld((Packet4ui) p2d_MZERO, (Packet4ui) p2l_ZERO, 8);//{ 0x8000000000000000, 0x0000000000000000 };
|
||||
#endif
|
||||
#endif
|
||||
|
||||
//---------- float ----------
|
||||
struct Packet2cf
|
||||
{
|
||||
EIGEN_STRONG_INLINE explicit Packet2cf() {}
|
||||
EIGEN_STRONG_INLINE explicit Packet2cf(const Packet4f& a) : v(a) {}
|
||||
|
||||
EIGEN_STRONG_INLINE Packet2cf pmul(const Packet2cf& a, const Packet2cf& b)
|
||||
{
|
||||
Packet4f v1, v2;
|
||||
|
||||
// Permute and multiply the real parts of a and b
|
||||
v1 = vec_perm(a.v, a.v, p16uc_PSET32_WODD);
|
||||
// Get the imaginary parts of a
|
||||
v2 = vec_perm(a.v, a.v, p16uc_PSET32_WEVEN);
|
||||
// multiply a_re * b
|
||||
v1 = vec_madd(v1, b.v, p4f_ZERO);
|
||||
// multiply a_im * b and get the conjugate result
|
||||
v2 = vec_madd(v2, b.v, p4f_ZERO);
|
||||
v2 = reinterpret_cast<Packet4f>(pxor(v2, reinterpret_cast<Packet4f>(p4ui_CONJ_XOR)));
|
||||
// permute back to a proper order
|
||||
v2 = vec_perm(v2, v2, p16uc_COMPLEX32_REV);
|
||||
|
||||
return Packet2cf(padd<Packet4f>(v1, v2));
|
||||
}
|
||||
|
||||
EIGEN_STRONG_INLINE Packet2cf& operator*=(const Packet2cf& b) {
|
||||
v = pmul(Packet2cf(*this), b).v;
|
||||
return *this;
|
||||
}
|
||||
EIGEN_STRONG_INLINE Packet2cf operator*(const Packet2cf& b) const {
|
||||
return Packet2cf(*this) *= b;
|
||||
}
|
||||
|
||||
EIGEN_STRONG_INLINE Packet2cf& operator+=(const Packet2cf& b) {
|
||||
v = padd(v, b.v);
|
||||
return *this;
|
||||
}
|
||||
EIGEN_STRONG_INLINE Packet2cf operator+(const Packet2cf& b) const {
|
||||
return Packet2cf(*this) += b;
|
||||
}
|
||||
EIGEN_STRONG_INLINE Packet2cf& operator-=(const Packet2cf& b) {
|
||||
v = psub(v, b.v);
|
||||
return *this;
|
||||
}
|
||||
EIGEN_STRONG_INLINE Packet2cf operator-(const Packet2cf& b) const {
|
||||
return Packet2cf(*this) -= b;
|
||||
}
|
||||
EIGEN_STRONG_INLINE Packet2cf operator-(void) const {
|
||||
return Packet2cf(-v);
|
||||
}
|
||||
|
||||
Packet4f v;
|
||||
};
|
||||
|
||||
template<> struct packet_traits<std::complex<float> > : default_packet_traits
|
||||
{
|
||||
typedef Packet2cf type;
|
||||
typedef Packet2cf half;
|
||||
typedef Packet4f as_real;
|
||||
enum {
|
||||
Vectorizable = 1,
|
||||
AlignedOnScalar = 1,
|
||||
size = 2,
|
||||
HasHalfPacket = 0,
|
||||
|
||||
HasAdd = 1,
|
||||
HasSub = 1,
|
||||
HasMul = 1,
|
||||
HasDiv = 1,
|
||||
HasNegate = 1,
|
||||
HasAbs = 0,
|
||||
HasAbs2 = 0,
|
||||
HasMin = 0,
|
||||
HasMax = 0,
|
||||
#ifdef __VSX__
|
||||
HasBlend = 1,
|
||||
#endif
|
||||
HasSetLinear = 0
|
||||
};
|
||||
};
|
||||
|
||||
template<> struct unpacket_traits<Packet2cf> { typedef std::complex<float> type; enum {size=2, alignment=Aligned16, vectorizable=true, masked_load_available=false, masked_store_available=false}; typedef Packet2cf half; typedef Packet4f as_real; };
|
||||
|
||||
template<> EIGEN_STRONG_INLINE Packet2cf pset1<Packet2cf>(const std::complex<float>& from)
|
||||
{
|
||||
Packet2cf res;
|
||||
if((std::ptrdiff_t(&from) % 16) == 0)
|
||||
res.v = pload<Packet4f>((const float *)&from);
|
||||
else
|
||||
res.v = ploadu<Packet4f>((const float *)&from);
|
||||
res.v = vec_perm(res.v, res.v, p16uc_PSET64_HI);
|
||||
return res;
|
||||
}
|
||||
|
||||
template<> EIGEN_STRONG_INLINE Packet2cf pload<Packet2cf>(const std::complex<float>* from) { return Packet2cf(pload<Packet4f>((const float *) from)); }
|
||||
template<> EIGEN_STRONG_INLINE Packet2cf ploadu<Packet2cf>(const std::complex<float>* from) { return Packet2cf(ploadu<Packet4f>((const float*) from)); }
|
||||
template<> EIGEN_STRONG_INLINE Packet2cf ploaddup<Packet2cf>(const std::complex<float>* from) { return pset1<Packet2cf>(*from); }
|
||||
|
||||
template<> EIGEN_STRONG_INLINE void pstore <std::complex<float> >(std::complex<float> * to, const Packet2cf& from) { pstore((float*)to, from.v); }
|
||||
template<> EIGEN_STRONG_INLINE void pstoreu<std::complex<float> >(std::complex<float> * to, const Packet2cf& from) { pstoreu((float*)to, from.v); }
|
||||
|
||||
EIGEN_STRONG_INLINE Packet2cf pload2(const std::complex<float>* from0, const std::complex<float>* from1)
|
||||
{
|
||||
Packet4f res0, res1;
|
||||
#ifdef __VSX__
|
||||
__asm__ ("lxsdx %x0,%y1" : "=wa" (res0) : "Z" (*from0));
|
||||
__asm__ ("lxsdx %x0,%y1" : "=wa" (res1) : "Z" (*from1));
|
||||
#ifdef _BIG_ENDIAN
|
||||
__asm__ ("xxpermdi %x0, %x1, %x2, 0" : "=wa" (res0) : "wa" (res0), "wa" (res1));
|
||||
#else
|
||||
__asm__ ("xxpermdi %x0, %x2, %x1, 0" : "=wa" (res0) : "wa" (res0), "wa" (res1));
|
||||
#endif
|
||||
#else
|
||||
*reinterpret_cast<std::complex<float> *>(&res0) = *from0;
|
||||
*reinterpret_cast<std::complex<float> *>(&res1) = *from1;
|
||||
res0 = vec_perm(res0, res1, p16uc_TRANSPOSE64_HI);
|
||||
#endif
|
||||
return Packet2cf(res0);
|
||||
}
|
||||
|
||||
template<> EIGEN_DEVICE_FUNC inline Packet2cf pgather<std::complex<float>, Packet2cf>(const std::complex<float>* from, Index stride)
|
||||
{
|
||||
EIGEN_ALIGN16 std::complex<float> af[2];
|
||||
af[0] = from[0*stride];
|
||||
af[1] = from[1*stride];
|
||||
return pload<Packet2cf>(af);
|
||||
}
|
||||
template<> EIGEN_DEVICE_FUNC inline void pscatter<std::complex<float>, Packet2cf>(std::complex<float>* to, const Packet2cf& from, Index stride)
|
||||
{
|
||||
EIGEN_ALIGN16 std::complex<float> af[2];
|
||||
pstore<std::complex<float> >((std::complex<float> *) af, from);
|
||||
to[0*stride] = af[0];
|
||||
to[1*stride] = af[1];
|
||||
}
|
||||
|
||||
template<> EIGEN_STRONG_INLINE Packet2cf padd<Packet2cf>(const Packet2cf& a, const Packet2cf& b) { return Packet2cf(a.v + b.v); }
|
||||
template<> EIGEN_STRONG_INLINE Packet2cf psub<Packet2cf>(const Packet2cf& a, const Packet2cf& b) { return Packet2cf(a.v - b.v); }
|
||||
template<> EIGEN_STRONG_INLINE Packet2cf pnegate(const Packet2cf& a) { return Packet2cf(pnegate(a.v)); }
|
||||
template<> EIGEN_STRONG_INLINE Packet2cf pconj(const Packet2cf& a) { return Packet2cf(pxor<Packet4f>(a.v, reinterpret_cast<Packet4f>(p4ui_CONJ_XOR))); }
|
||||
|
||||
template<> EIGEN_STRONG_INLINE Packet2cf pand <Packet2cf>(const Packet2cf& a, const Packet2cf& b) { return Packet2cf(pand<Packet4f>(a.v, b.v)); }
|
||||
template<> EIGEN_STRONG_INLINE Packet2cf por <Packet2cf>(const Packet2cf& a, const Packet2cf& b) { return Packet2cf(por<Packet4f>(a.v, b.v)); }
|
||||
template<> EIGEN_STRONG_INLINE Packet2cf pxor <Packet2cf>(const Packet2cf& a, const Packet2cf& b) { return Packet2cf(pxor<Packet4f>(a.v, b.v)); }
|
||||
template<> EIGEN_STRONG_INLINE Packet2cf pandnot<Packet2cf>(const Packet2cf& a, const Packet2cf& b) { return Packet2cf(pandnot<Packet4f>(a.v, b.v)); }
|
||||
|
||||
template<> EIGEN_STRONG_INLINE void prefetch<std::complex<float> >(const std::complex<float> * addr) { EIGEN_PPC_PREFETCH(addr); }
|
||||
|
||||
template<> EIGEN_STRONG_INLINE std::complex<float> pfirst<Packet2cf>(const Packet2cf& a)
|
||||
{
|
||||
EIGEN_ALIGN16 std::complex<float> res[2];
|
||||
pstore((float *)&res, a.v);
|
||||
|
||||
return res[0];
|
||||
}
|
||||
|
||||
template<> EIGEN_STRONG_INLINE Packet2cf preverse(const Packet2cf& a)
|
||||
{
|
||||
Packet4f rev_a;
|
||||
rev_a = vec_perm(a.v, a.v, p16uc_COMPLEX32_REV2);
|
||||
return Packet2cf(rev_a);
|
||||
}
|
||||
|
||||
template<> EIGEN_STRONG_INLINE std::complex<float> predux<Packet2cf>(const Packet2cf& a)
|
||||
{
|
||||
Packet4f b;
|
||||
b = vec_sld(a.v, a.v, 8);
|
||||
b = padd<Packet4f>(a.v, b);
|
||||
return pfirst<Packet2cf>(Packet2cf(b));
|
||||
}
|
||||
|
||||
template<> EIGEN_STRONG_INLINE std::complex<float> predux_mul<Packet2cf>(const Packet2cf& a)
|
||||
{
|
||||
Packet4f b;
|
||||
Packet2cf prod;
|
||||
b = vec_sld(a.v, a.v, 8);
|
||||
prod = pmul<Packet2cf>(a, Packet2cf(b));
|
||||
|
||||
return pfirst<Packet2cf>(prod);
|
||||
}
|
||||
|
||||
EIGEN_MAKE_CONJ_HELPER_CPLX_REAL(Packet2cf,Packet4f)
|
||||
|
||||
template<> EIGEN_STRONG_INLINE Packet2cf pdiv<Packet2cf>(const Packet2cf& a, const Packet2cf& b)
|
||||
{
|
||||
// TODO optimize it for AltiVec
|
||||
Packet2cf res = pmul(a, pconj(b));
|
||||
Packet4f s = pmul<Packet4f>(b.v, b.v);
|
||||
return Packet2cf(pdiv(res.v, padd<Packet4f>(s, vec_perm(s, s, p16uc_COMPLEX32_REV))));
|
||||
}
|
||||
|
||||
template<> EIGEN_STRONG_INLINE Packet2cf pcplxflip<Packet2cf>(const Packet2cf& x)
|
||||
{
|
||||
return Packet2cf(vec_perm(x.v, x.v, p16uc_COMPLEX32_REV));
|
||||
}
|
||||
|
||||
EIGEN_STRONG_INLINE void ptranspose(PacketBlock<Packet2cf,2>& kernel)
|
||||
{
|
||||
Packet4f tmp = vec_perm(kernel.packet[0].v, kernel.packet[1].v, p16uc_TRANSPOSE64_HI);
|
||||
kernel.packet[1].v = vec_perm(kernel.packet[0].v, kernel.packet[1].v, p16uc_TRANSPOSE64_LO);
|
||||
kernel.packet[0].v = tmp;
|
||||
}
|
||||
|
||||
template<> EIGEN_STRONG_INLINE Packet2cf pcmp_eq(const Packet2cf& a, const Packet2cf& b) {
|
||||
Packet4f eq = reinterpret_cast<Packet4f>(vec_cmpeq(a.v,b.v));
|
||||
return Packet2cf(vec_and(eq, vec_perm(eq, eq, p16uc_COMPLEX32_REV)));
|
||||
}
|
||||
|
||||
#ifdef __VSX__
|
||||
template<> EIGEN_STRONG_INLINE Packet2cf pblend(const Selector<2>& ifPacket, const Packet2cf& thenPacket, const Packet2cf& elsePacket) {
|
||||
Packet2cf result;
|
||||
result.v = reinterpret_cast<Packet4f>(pblend<Packet2d>(ifPacket, reinterpret_cast<Packet2d>(thenPacket.v), reinterpret_cast<Packet2d>(elsePacket.v)));
|
||||
return result;
|
||||
}
|
||||
#endif
|
||||
|
||||
template<> EIGEN_STRONG_INLINE Packet2cf psqrt<Packet2cf>(const Packet2cf& a)
|
||||
{
|
||||
return psqrt_complex<Packet2cf>(a);
|
||||
}
|
||||
|
||||
//---------- double ----------
|
||||
#ifdef __VSX__
|
||||
struct Packet1cd
|
||||
{
|
||||
EIGEN_STRONG_INLINE Packet1cd() {}
|
||||
EIGEN_STRONG_INLINE explicit Packet1cd(const Packet2d& a) : v(a) {}
|
||||
|
||||
EIGEN_STRONG_INLINE Packet1cd pmul(const Packet1cd& a, const Packet1cd& b)
|
||||
{
|
||||
Packet2d a_re, a_im, v1, v2;
|
||||
|
||||
// Permute and multiply the real parts of a and b
|
||||
a_re = vec_perm(a.v, a.v, p16uc_PSET64_HI);
|
||||
// Get the imaginary parts of a
|
||||
a_im = vec_perm(a.v, a.v, p16uc_PSET64_LO);
|
||||
// multiply a_re * b
|
||||
v1 = vec_madd(a_re, b.v, p2d_ZERO);
|
||||
// multiply a_im * b and get the conjugate result
|
||||
v2 = vec_madd(a_im, b.v, p2d_ZERO);
|
||||
v2 = reinterpret_cast<Packet2d>(vec_sld(reinterpret_cast<Packet4ui>(v2), reinterpret_cast<Packet4ui>(v2), 8));
|
||||
v2 = pxor(v2, reinterpret_cast<Packet2d>(p2ul_CONJ_XOR1));
|
||||
|
||||
return Packet1cd(padd<Packet2d>(v1, v2));
|
||||
}
|
||||
|
||||
EIGEN_STRONG_INLINE Packet1cd& operator*=(const Packet1cd& b) {
|
||||
v = pmul(Packet1cd(*this), b).v;
|
||||
return *this;
|
||||
}
|
||||
EIGEN_STRONG_INLINE Packet1cd operator*(const Packet1cd& b) const {
|
||||
return Packet1cd(*this) *= b;
|
||||
}
|
||||
|
||||
EIGEN_STRONG_INLINE Packet1cd& operator+=(const Packet1cd& b) {
|
||||
v = padd(v, b.v);
|
||||
return *this;
|
||||
}
|
||||
EIGEN_STRONG_INLINE Packet1cd operator+(const Packet1cd& b) const {
|
||||
return Packet1cd(*this) += b;
|
||||
}
|
||||
EIGEN_STRONG_INLINE Packet1cd& operator-=(const Packet1cd& b) {
|
||||
v = psub(v, b.v);
|
||||
return *this;
|
||||
}
|
||||
EIGEN_STRONG_INLINE Packet1cd operator-(const Packet1cd& b) const {
|
||||
return Packet1cd(*this) -= b;
|
||||
}
|
||||
EIGEN_STRONG_INLINE Packet1cd operator-(void) const {
|
||||
return Packet1cd(-v);
|
||||
}
|
||||
|
||||
Packet2d v;
|
||||
};
|
||||
|
||||
template<> struct packet_traits<std::complex<double> > : default_packet_traits
|
||||
{
|
||||
typedef Packet1cd type;
|
||||
typedef Packet1cd half;
|
||||
typedef Packet2d as_real;
|
||||
enum {
|
||||
Vectorizable = 1,
|
||||
AlignedOnScalar = 0,
|
||||
size = 1,
|
||||
HasHalfPacket = 0,
|
||||
|
||||
HasAdd = 1,
|
||||
HasSub = 1,
|
||||
HasMul = 1,
|
||||
HasDiv = 1,
|
||||
HasNegate = 1,
|
||||
HasAbs = 0,
|
||||
HasAbs2 = 0,
|
||||
HasMin = 0,
|
||||
HasMax = 0,
|
||||
HasSetLinear = 0
|
||||
};
|
||||
};
|
||||
|
||||
template<> struct unpacket_traits<Packet1cd> { typedef std::complex<double> type; enum {size=1, alignment=Aligned16, vectorizable=true, masked_load_available=false, masked_store_available=false}; typedef Packet1cd half; typedef Packet2d as_real; };
|
||||
|
||||
template<> EIGEN_STRONG_INLINE Packet1cd pload <Packet1cd>(const std::complex<double>* from) { return Packet1cd(pload<Packet2d>((const double*)from)); }
|
||||
template<> EIGEN_STRONG_INLINE Packet1cd ploadu<Packet1cd>(const std::complex<double>* from) { return Packet1cd(ploadu<Packet2d>((const double*)from)); }
|
||||
template<> EIGEN_STRONG_INLINE void pstore <std::complex<double> >(std::complex<double> * to, const Packet1cd& from) { pstore((double*)to, from.v); }
|
||||
template<> EIGEN_STRONG_INLINE void pstoreu<std::complex<double> >(std::complex<double> * to, const Packet1cd& from) { pstoreu((double*)to, from.v); }
|
||||
|
||||
template<> EIGEN_STRONG_INLINE Packet1cd pset1<Packet1cd>(const std::complex<double>& from)
|
||||
{ /* here we really have to use unaligned loads :( */ return ploadu<Packet1cd>(&from); }
|
||||
|
||||
template<> EIGEN_DEVICE_FUNC inline Packet1cd pgather<std::complex<double>, Packet1cd>(const std::complex<double>* from, Index)
|
||||
{
|
||||
return pload<Packet1cd>(from);
|
||||
}
|
||||
template<> EIGEN_DEVICE_FUNC inline void pscatter<std::complex<double>, Packet1cd>(std::complex<double>* to, const Packet1cd& from, Index)
|
||||
{
|
||||
pstore<std::complex<double> >(to, from);
|
||||
}
|
||||
|
||||
template<> EIGEN_STRONG_INLINE Packet1cd padd<Packet1cd>(const Packet1cd& a, const Packet1cd& b) { return Packet1cd(a.v + b.v); }
|
||||
template<> EIGEN_STRONG_INLINE Packet1cd psub<Packet1cd>(const Packet1cd& a, const Packet1cd& b) { return Packet1cd(a.v - b.v); }
|
||||
template<> EIGEN_STRONG_INLINE Packet1cd pnegate(const Packet1cd& a) { return Packet1cd(pnegate(Packet2d(a.v))); }
|
||||
template<> EIGEN_STRONG_INLINE Packet1cd pconj(const Packet1cd& a) { return Packet1cd(pxor(a.v, reinterpret_cast<Packet2d>(p2ul_CONJ_XOR2))); }
|
||||
|
||||
template<> EIGEN_STRONG_INLINE Packet1cd pand <Packet1cd>(const Packet1cd& a, const Packet1cd& b) { return Packet1cd(pand(a.v,b.v)); }
|
||||
template<> EIGEN_STRONG_INLINE Packet1cd por <Packet1cd>(const Packet1cd& a, const Packet1cd& b) { return Packet1cd(por(a.v,b.v)); }
|
||||
template<> EIGEN_STRONG_INLINE Packet1cd pxor <Packet1cd>(const Packet1cd& a, const Packet1cd& b) { return Packet1cd(pxor(a.v,b.v)); }
|
||||
template<> EIGEN_STRONG_INLINE Packet1cd pandnot<Packet1cd>(const Packet1cd& a, const Packet1cd& b) { return Packet1cd(pandnot(a.v, b.v)); }
|
||||
|
||||
template<> EIGEN_STRONG_INLINE Packet1cd ploaddup<Packet1cd>(const std::complex<double>* from) { return pset1<Packet1cd>(*from); }
|
||||
|
||||
template<> EIGEN_STRONG_INLINE void prefetch<std::complex<double> >(const std::complex<double> * addr) { EIGEN_PPC_PREFETCH(addr); }
|
||||
|
||||
template<> EIGEN_STRONG_INLINE std::complex<double> pfirst<Packet1cd>(const Packet1cd& a)
|
||||
{
|
||||
EIGEN_ALIGN16 std::complex<double> res[2];
|
||||
pstore<std::complex<double> >(res, a);
|
||||
|
||||
return res[0];
|
||||
}
|
||||
|
||||
template<> EIGEN_STRONG_INLINE Packet1cd preverse(const Packet1cd& a) { return a; }
|
||||
|
||||
template<> EIGEN_STRONG_INLINE std::complex<double> predux<Packet1cd>(const Packet1cd& a) { return pfirst(a); }
|
||||
|
||||
template<> EIGEN_STRONG_INLINE std::complex<double> predux_mul<Packet1cd>(const Packet1cd& a) { return pfirst(a); }
|
||||
|
||||
EIGEN_MAKE_CONJ_HELPER_CPLX_REAL(Packet1cd,Packet2d)
|
||||
|
||||
template<> EIGEN_STRONG_INLINE Packet1cd pdiv<Packet1cd>(const Packet1cd& a, const Packet1cd& b)
|
||||
{
|
||||
// TODO optimize it for AltiVec
|
||||
Packet1cd res = pmul(a,pconj(b));
|
||||
Packet2d s = pmul<Packet2d>(b.v, b.v);
|
||||
return Packet1cd(pdiv(res.v, padd<Packet2d>(s, vec_perm(s, s, p16uc_REVERSE64))));
|
||||
}
|
||||
|
||||
EIGEN_STRONG_INLINE Packet1cd pcplxflip/*<Packet1cd>*/(const Packet1cd& x)
|
||||
{
|
||||
return Packet1cd(preverse(Packet2d(x.v)));
|
||||
}
|
||||
|
||||
EIGEN_STRONG_INLINE void ptranspose(PacketBlock<Packet1cd,2>& kernel)
|
||||
{
|
||||
Packet2d tmp = vec_perm(kernel.packet[0].v, kernel.packet[1].v, p16uc_TRANSPOSE64_HI);
|
||||
kernel.packet[1].v = vec_perm(kernel.packet[0].v, kernel.packet[1].v, p16uc_TRANSPOSE64_LO);
|
||||
kernel.packet[0].v = tmp;
|
||||
}
|
||||
|
||||
template<> EIGEN_STRONG_INLINE Packet1cd pcmp_eq(const Packet1cd& a, const Packet1cd& b) {
|
||||
// Compare real and imaginary parts of a and b to get the mask vector:
|
||||
// [re(a)==re(b), im(a)==im(b)]
|
||||
Packet2d eq = reinterpret_cast<Packet2d>(vec_cmpeq(a.v,b.v));
|
||||
// Swap real/imag elements in the mask in to get:
|
||||
// [im(a)==im(b), re(a)==re(b)]
|
||||
Packet2d eq_swapped = reinterpret_cast<Packet2d>(vec_sld(reinterpret_cast<Packet4ui>(eq), reinterpret_cast<Packet4ui>(eq), 8));
|
||||
// Return re(a)==re(b) & im(a)==im(b) by computing bitwise AND of eq and eq_swapped
|
||||
return Packet1cd(vec_and(eq, eq_swapped));
|
||||
}
|
||||
|
||||
template<> EIGEN_STRONG_INLINE Packet1cd psqrt<Packet1cd>(const Packet1cd& a)
|
||||
{
|
||||
return psqrt_complex<Packet1cd>(a);
|
||||
}
|
||||
|
||||
#endif // __VSX__
|
||||
} // end namespace internal
|
||||
|
||||
} // end namespace Eigen
|
||||
|
||||
#endif // EIGEN_COMPLEX32_ALTIVEC_H
|
||||
@@ -0,0 +1,90 @@
|
||||
// This file is part of Eigen, a lightweight C++ template library
|
||||
// for linear algebra.
|
||||
//
|
||||
// Copyright (C) 2007 Julien Pommier
|
||||
// Copyright (C) 2009 Gael Guennebaud <gael.guennebaud@inria.fr>
|
||||
// Copyright (C) 2016 Konstantinos Margaritis <markos@freevec.org>
|
||||
//
|
||||
// This Source Code Form is subject to the terms of the Mozilla
|
||||
// Public License v. 2.0. If a copy of the MPL was not distributed
|
||||
// with this file, You can obtain one at http://mozilla.org/MPL/2.0/.
|
||||
|
||||
#ifndef EIGEN_MATH_FUNCTIONS_ALTIVEC_H
|
||||
#define EIGEN_MATH_FUNCTIONS_ALTIVEC_H
|
||||
|
||||
namespace Eigen {
|
||||
|
||||
namespace internal {
|
||||
|
||||
template<> EIGEN_DEFINE_FUNCTION_ALLOWING_MULTIPLE_DEFINITIONS EIGEN_UNUSED
|
||||
Packet4f plog<Packet4f>(const Packet4f& _x)
|
||||
{
|
||||
return plog_float(_x);
|
||||
}
|
||||
|
||||
template<> EIGEN_DEFINE_FUNCTION_ALLOWING_MULTIPLE_DEFINITIONS EIGEN_UNUSED
|
||||
Packet4f pexp<Packet4f>(const Packet4f& _x)
|
||||
{
|
||||
return pexp_float(_x);
|
||||
}
|
||||
|
||||
template<> EIGEN_DEFINE_FUNCTION_ALLOWING_MULTIPLE_DEFINITIONS EIGEN_UNUSED
|
||||
Packet4f psin<Packet4f>(const Packet4f& _x)
|
||||
{
|
||||
return psin_float(_x);
|
||||
}
|
||||
|
||||
template<> EIGEN_DEFINE_FUNCTION_ALLOWING_MULTIPLE_DEFINITIONS EIGEN_UNUSED
|
||||
Packet4f pcos<Packet4f>(const Packet4f& _x)
|
||||
{
|
||||
return pcos_float(_x);
|
||||
}
|
||||
|
||||
#ifndef EIGEN_COMP_CLANG
|
||||
template<> EIGEN_DEFINE_FUNCTION_ALLOWING_MULTIPLE_DEFINITIONS EIGEN_UNUSED
|
||||
Packet4f prsqrt<Packet4f>(const Packet4f& x)
|
||||
{
|
||||
return vec_rsqrt(x);
|
||||
}
|
||||
#endif
|
||||
|
||||
#ifdef __VSX__
|
||||
#ifndef EIGEN_COMP_CLANG
|
||||
template<> EIGEN_DEFINE_FUNCTION_ALLOWING_MULTIPLE_DEFINITIONS EIGEN_UNUSED
|
||||
Packet2d prsqrt<Packet2d>(const Packet2d& x)
|
||||
{
|
||||
return vec_rsqrt(x);
|
||||
}
|
||||
#endif
|
||||
|
||||
template<> EIGEN_DEFINE_FUNCTION_ALLOWING_MULTIPLE_DEFINITIONS EIGEN_UNUSED
|
||||
Packet4f psqrt<Packet4f>(const Packet4f& x)
|
||||
{
|
||||
return vec_sqrt(x);
|
||||
}
|
||||
|
||||
template<> EIGEN_DEFINE_FUNCTION_ALLOWING_MULTIPLE_DEFINITIONS EIGEN_UNUSED
|
||||
Packet2d psqrt<Packet2d>(const Packet2d& x)
|
||||
{
|
||||
return vec_sqrt(x);
|
||||
}
|
||||
|
||||
template<> EIGEN_DEFINE_FUNCTION_ALLOWING_MULTIPLE_DEFINITIONS EIGEN_UNUSED
|
||||
Packet2d pexp<Packet2d>(const Packet2d& _x)
|
||||
{
|
||||
return pexp_double(_x);
|
||||
}
|
||||
#endif
|
||||
|
||||
// Hyperbolic Tangent function.
|
||||
template <>
|
||||
EIGEN_DEFINE_FUNCTION_ALLOWING_MULTIPLE_DEFINITIONS EIGEN_UNUSED Packet4f
|
||||
ptanh<Packet4f>(const Packet4f& x) {
|
||||
return internal::generic_fast_tanh_float(x);
|
||||
}
|
||||
|
||||
} // end namespace internal
|
||||
|
||||
} // end namespace Eigen
|
||||
|
||||
#endif // EIGEN_MATH_FUNCTIONS_ALTIVEC_H
|
||||
File diff suppressed because it is too large
Load Diff
@@ -0,0 +1,221 @@
|
||||
//#define EIGEN_POWER_USE_PREFETCH // Use prefetching in gemm routines
|
||||
#ifdef EIGEN_POWER_USE_PREFETCH
|
||||
#define EIGEN_POWER_PREFETCH(p) prefetch(p)
|
||||
#else
|
||||
#define EIGEN_POWER_PREFETCH(p)
|
||||
#endif
|
||||
|
||||
namespace Eigen {
|
||||
|
||||
namespace internal {
|
||||
|
||||
template<typename Scalar, typename Packet, typename DataMapper, typename Index, const Index accRows>
|
||||
EIGEN_STRONG_INLINE void gemm_extra_col(
|
||||
const DataMapper& res,
|
||||
const Scalar* lhs_base,
|
||||
const Scalar* rhs_base,
|
||||
Index depth,
|
||||
Index strideA,
|
||||
Index offsetA,
|
||||
Index row,
|
||||
Index col,
|
||||
Index remaining_rows,
|
||||
Index remaining_cols,
|
||||
const Packet& pAlpha);
|
||||
|
||||
template<typename Scalar, typename Packet, typename DataMapper, typename Index, const Index accRows, const Index accCols>
|
||||
EIGEN_STRONG_INLINE void gemm_extra_row(
|
||||
const DataMapper& res,
|
||||
const Scalar* lhs_base,
|
||||
const Scalar* rhs_base,
|
||||
Index depth,
|
||||
Index strideA,
|
||||
Index offsetA,
|
||||
Index row,
|
||||
Index col,
|
||||
Index rows,
|
||||
Index cols,
|
||||
Index remaining_rows,
|
||||
const Packet& pAlpha,
|
||||
const Packet& pMask);
|
||||
|
||||
template<typename Scalar, typename Packet, typename DataMapper, typename Index, const Index accCols>
|
||||
EIGEN_STRONG_INLINE void gemm_unrolled_col(
|
||||
const DataMapper& res,
|
||||
const Scalar* lhs_base,
|
||||
const Scalar* rhs_base,
|
||||
Index depth,
|
||||
Index strideA,
|
||||
Index offsetA,
|
||||
Index& row,
|
||||
Index rows,
|
||||
Index col,
|
||||
Index remaining_cols,
|
||||
const Packet& pAlpha);
|
||||
|
||||
template<typename Packet>
|
||||
EIGEN_ALWAYS_INLINE Packet bmask(const int remaining_rows);
|
||||
|
||||
template<typename Scalar, typename Packet, typename Packetc, typename DataMapper, typename Index, const Index accRows, const Index accCols, bool ConjugateLhs, bool ConjugateRhs, bool LhsIsReal, bool RhsIsReal>
|
||||
EIGEN_STRONG_INLINE void gemm_complex_extra_col(
|
||||
const DataMapper& res,
|
||||
const Scalar* lhs_base,
|
||||
const Scalar* rhs_base,
|
||||
Index depth,
|
||||
Index strideA,
|
||||
Index offsetA,
|
||||
Index strideB,
|
||||
Index row,
|
||||
Index col,
|
||||
Index remaining_rows,
|
||||
Index remaining_cols,
|
||||
const Packet& pAlphaReal,
|
||||
const Packet& pAlphaImag);
|
||||
|
||||
template<typename Scalar, typename Packet, typename Packetc, typename DataMapper, typename Index, const Index accRows, const Index accCols, bool ConjugateLhs, bool ConjugateRhs, bool LhsIsReal, bool RhsIsReal>
|
||||
EIGEN_STRONG_INLINE void gemm_complex_extra_row(
|
||||
const DataMapper& res,
|
||||
const Scalar* lhs_base,
|
||||
const Scalar* rhs_base,
|
||||
Index depth,
|
||||
Index strideA,
|
||||
Index offsetA,
|
||||
Index strideB,
|
||||
Index row,
|
||||
Index col,
|
||||
Index rows,
|
||||
Index cols,
|
||||
Index remaining_rows,
|
||||
const Packet& pAlphaReal,
|
||||
const Packet& pAlphaImag,
|
||||
const Packet& pMask);
|
||||
|
||||
template<typename Scalar, typename Packet, typename Packetc, typename DataMapper, typename Index, const Index accCols, bool ConjugateLhs, bool ConjugateRhs, bool LhsIsReal, bool RhsIsReal>
|
||||
EIGEN_STRONG_INLINE void gemm_complex_unrolled_col(
|
||||
const DataMapper& res,
|
||||
const Scalar* lhs_base,
|
||||
const Scalar* rhs_base,
|
||||
Index depth,
|
||||
Index strideA,
|
||||
Index offsetA,
|
||||
Index strideB,
|
||||
Index& row,
|
||||
Index rows,
|
||||
Index col,
|
||||
Index remaining_cols,
|
||||
const Packet& pAlphaReal,
|
||||
const Packet& pAlphaImag);
|
||||
|
||||
template<typename Scalar, typename Packet>
|
||||
EIGEN_ALWAYS_INLINE Packet ploadLhs(const Scalar* lhs);
|
||||
|
||||
template<typename DataMapper, typename Packet, typename Index, const Index accCols, int N, int StorageOrder>
|
||||
EIGEN_ALWAYS_INLINE void bload(PacketBlock<Packet,4>& acc, const DataMapper& res, Index row, Index col);
|
||||
|
||||
template<typename DataMapper, typename Packet, typename Index, const Index accCols, int N, int StorageOrder>
|
||||
EIGEN_ALWAYS_INLINE void bload(PacketBlock<Packet,8>& acc, const DataMapper& res, Index row, Index col);
|
||||
|
||||
template<typename Packet>
|
||||
EIGEN_ALWAYS_INLINE void bscale(PacketBlock<Packet,4>& acc, PacketBlock<Packet,4>& accZ, const Packet& pAlpha);
|
||||
|
||||
template<typename Packet, int N>
|
||||
EIGEN_ALWAYS_INLINE void bscalec(PacketBlock<Packet,N>& aReal, PacketBlock<Packet,N>& aImag, const Packet& bReal, const Packet& bImag, PacketBlock<Packet,N>& cReal, PacketBlock<Packet,N>& cImag);
|
||||
|
||||
const static Packet16uc p16uc_SETCOMPLEX32_FIRST = { 0, 1, 2, 3,
|
||||
16, 17, 18, 19,
|
||||
4, 5, 6, 7,
|
||||
20, 21, 22, 23};
|
||||
|
||||
const static Packet16uc p16uc_SETCOMPLEX32_SECOND = { 8, 9, 10, 11,
|
||||
24, 25, 26, 27,
|
||||
12, 13, 14, 15,
|
||||
28, 29, 30, 31};
|
||||
//[a,b],[ai,bi] = [a,ai] - This is equivalent to p16uc_GETREAL64
|
||||
const static Packet16uc p16uc_SETCOMPLEX64_FIRST = { 0, 1, 2, 3, 4, 5, 6, 7,
|
||||
16, 17, 18, 19, 20, 21, 22, 23};
|
||||
|
||||
//[a,b],[ai,bi] = [b,bi] - This is equivalent to p16uc_GETIMAG64
|
||||
const static Packet16uc p16uc_SETCOMPLEX64_SECOND = { 8, 9, 10, 11, 12, 13, 14, 15,
|
||||
24, 25, 26, 27, 28, 29, 30, 31};
|
||||
|
||||
|
||||
// Grab two decouples real/imaginary PacketBlocks and return two coupled (real/imaginary pairs) PacketBlocks.
|
||||
template<typename Packet, typename Packetc>
|
||||
EIGEN_ALWAYS_INLINE void bcouple_common(PacketBlock<Packet,4>& taccReal, PacketBlock<Packet,4>& taccImag, PacketBlock<Packetc, 4>& acc1, PacketBlock<Packetc, 4>& acc2)
|
||||
{
|
||||
acc1.packet[0].v = vec_perm(taccReal.packet[0], taccImag.packet[0], p16uc_SETCOMPLEX32_FIRST);
|
||||
acc1.packet[1].v = vec_perm(taccReal.packet[1], taccImag.packet[1], p16uc_SETCOMPLEX32_FIRST);
|
||||
acc1.packet[2].v = vec_perm(taccReal.packet[2], taccImag.packet[2], p16uc_SETCOMPLEX32_FIRST);
|
||||
acc1.packet[3].v = vec_perm(taccReal.packet[3], taccImag.packet[3], p16uc_SETCOMPLEX32_FIRST);
|
||||
|
||||
acc2.packet[0].v = vec_perm(taccReal.packet[0], taccImag.packet[0], p16uc_SETCOMPLEX32_SECOND);
|
||||
acc2.packet[1].v = vec_perm(taccReal.packet[1], taccImag.packet[1], p16uc_SETCOMPLEX32_SECOND);
|
||||
acc2.packet[2].v = vec_perm(taccReal.packet[2], taccImag.packet[2], p16uc_SETCOMPLEX32_SECOND);
|
||||
acc2.packet[3].v = vec_perm(taccReal.packet[3], taccImag.packet[3], p16uc_SETCOMPLEX32_SECOND);
|
||||
}
|
||||
|
||||
template<typename Packet, typename Packetc>
|
||||
EIGEN_ALWAYS_INLINE void bcouple(PacketBlock<Packet,4>& taccReal, PacketBlock<Packet,4>& taccImag, PacketBlock<Packetc,8>& tRes, PacketBlock<Packetc, 4>& acc1, PacketBlock<Packetc, 4>& acc2)
|
||||
{
|
||||
bcouple_common<Packet, Packetc>(taccReal, taccImag, acc1, acc2);
|
||||
|
||||
acc1.packet[0] = padd<Packetc>(tRes.packet[0], acc1.packet[0]);
|
||||
acc1.packet[1] = padd<Packetc>(tRes.packet[1], acc1.packet[1]);
|
||||
acc1.packet[2] = padd<Packetc>(tRes.packet[2], acc1.packet[2]);
|
||||
acc1.packet[3] = padd<Packetc>(tRes.packet[3], acc1.packet[3]);
|
||||
|
||||
acc2.packet[0] = padd<Packetc>(tRes.packet[4], acc2.packet[0]);
|
||||
acc2.packet[1] = padd<Packetc>(tRes.packet[5], acc2.packet[1]);
|
||||
acc2.packet[2] = padd<Packetc>(tRes.packet[6], acc2.packet[2]);
|
||||
acc2.packet[3] = padd<Packetc>(tRes.packet[7], acc2.packet[3]);
|
||||
}
|
||||
|
||||
template<typename Packet, typename Packetc>
|
||||
EIGEN_ALWAYS_INLINE void bcouple_common(PacketBlock<Packet,1>& taccReal, PacketBlock<Packet,1>& taccImag, PacketBlock<Packetc, 1>& acc1, PacketBlock<Packetc, 1>& acc2)
|
||||
{
|
||||
acc1.packet[0].v = vec_perm(taccReal.packet[0], taccImag.packet[0], p16uc_SETCOMPLEX32_FIRST);
|
||||
|
||||
acc2.packet[0].v = vec_perm(taccReal.packet[0], taccImag.packet[0], p16uc_SETCOMPLEX32_SECOND);
|
||||
}
|
||||
|
||||
template<typename Packet, typename Packetc>
|
||||
EIGEN_ALWAYS_INLINE void bcouple(PacketBlock<Packet,1>& taccReal, PacketBlock<Packet,1>& taccImag, PacketBlock<Packetc,2>& tRes, PacketBlock<Packetc, 1>& acc1, PacketBlock<Packetc, 1>& acc2)
|
||||
{
|
||||
bcouple_common<Packet, Packetc>(taccReal, taccImag, acc1, acc2);
|
||||
|
||||
acc1.packet[0] = padd<Packetc>(tRes.packet[0], acc1.packet[0]);
|
||||
|
||||
acc2.packet[0] = padd<Packetc>(tRes.packet[1], acc2.packet[0]);
|
||||
}
|
||||
|
||||
template<>
|
||||
EIGEN_ALWAYS_INLINE void bcouple_common<Packet2d, Packet1cd>(PacketBlock<Packet2d,4>& taccReal, PacketBlock<Packet2d,4>& taccImag, PacketBlock<Packet1cd, 4>& acc1, PacketBlock<Packet1cd, 4>& acc2)
|
||||
{
|
||||
acc1.packet[0].v = vec_perm(taccReal.packet[0], taccImag.packet[0], p16uc_SETCOMPLEX64_FIRST);
|
||||
acc1.packet[1].v = vec_perm(taccReal.packet[1], taccImag.packet[1], p16uc_SETCOMPLEX64_FIRST);
|
||||
acc1.packet[2].v = vec_perm(taccReal.packet[2], taccImag.packet[2], p16uc_SETCOMPLEX64_FIRST);
|
||||
acc1.packet[3].v = vec_perm(taccReal.packet[3], taccImag.packet[3], p16uc_SETCOMPLEX64_FIRST);
|
||||
|
||||
acc2.packet[0].v = vec_perm(taccReal.packet[0], taccImag.packet[0], p16uc_SETCOMPLEX64_SECOND);
|
||||
acc2.packet[1].v = vec_perm(taccReal.packet[1], taccImag.packet[1], p16uc_SETCOMPLEX64_SECOND);
|
||||
acc2.packet[2].v = vec_perm(taccReal.packet[2], taccImag.packet[2], p16uc_SETCOMPLEX64_SECOND);
|
||||
acc2.packet[3].v = vec_perm(taccReal.packet[3], taccImag.packet[3], p16uc_SETCOMPLEX64_SECOND);
|
||||
}
|
||||
|
||||
template<>
|
||||
EIGEN_ALWAYS_INLINE void bcouple_common<Packet2d, Packet1cd>(PacketBlock<Packet2d,1>& taccReal, PacketBlock<Packet2d,1>& taccImag, PacketBlock<Packet1cd, 1>& acc1, PacketBlock<Packet1cd, 1>& acc2)
|
||||
{
|
||||
acc1.packet[0].v = vec_perm(taccReal.packet[0], taccImag.packet[0], p16uc_SETCOMPLEX64_FIRST);
|
||||
|
||||
acc2.packet[0].v = vec_perm(taccReal.packet[0], taccImag.packet[0], p16uc_SETCOMPLEX64_SECOND);
|
||||
}
|
||||
|
||||
// This is necessary because ploadRhs for double returns a pair of vectors when MMA is enabled.
|
||||
template<typename Scalar, typename Packet>
|
||||
EIGEN_ALWAYS_INLINE Packet ploadRhs(const Scalar* rhs)
|
||||
{
|
||||
return ploadu<Packet>(rhs);
|
||||
}
|
||||
|
||||
} // end namespace internal
|
||||
} // end namespace Eigen
|
||||
@@ -0,0 +1,629 @@
|
||||
// This file is part of Eigen, a lightweight C++ template library
|
||||
// for linear algebra.
|
||||
//
|
||||
// Copyright (C) 2020 Everton Constantino (everton.constantino@ibm.com)
|
||||
// Copyright (C) 2021 Chip Kerchner (chip.kerchner@ibm.com)
|
||||
//
|
||||
// This Source Code Form is subject to the terms of the Mozilla
|
||||
// Public License v. 2.0. If a copy of the MPL was not distributed
|
||||
// with this file, You can obtain one at http://mozilla.org/MPL/2.0/.
|
||||
|
||||
#ifndef EIGEN_MATRIX_PRODUCT_MMA_ALTIVEC_H
|
||||
#define EIGEN_MATRIX_PRODUCT_MMA_ALTIVEC_H
|
||||
|
||||
#pragma GCC target("cpu=power10")
|
||||
|
||||
#ifdef __has_builtin
|
||||
#if !__has_builtin(__builtin_vsx_assemble_pair)
|
||||
#define __builtin_vsx_assemble_pair __builtin_mma_assemble_pair
|
||||
#endif
|
||||
#endif
|
||||
|
||||
namespace Eigen {
|
||||
|
||||
namespace internal {
|
||||
|
||||
template<typename Scalar, typename Packet>
|
||||
EIGEN_ALWAYS_INLINE void bsetzeroMMA(__vector_quad* acc)
|
||||
{
|
||||
__builtin_mma_xxsetaccz(acc);
|
||||
}
|
||||
|
||||
template<typename DataMapper, typename Index, typename Packet, const Index accCols>
|
||||
EIGEN_ALWAYS_INLINE void storeAccumulator(Index i, Index j, const DataMapper& data, const Packet& alpha, __vector_quad* acc)
|
||||
{
|
||||
PacketBlock<Packet, 4> result;
|
||||
__builtin_mma_disassemble_acc(&result.packet, acc);
|
||||
|
||||
PacketBlock<Packet, 4> tRes;
|
||||
bload<DataMapper, Packet, Index, accCols, 0, ColMajor>(tRes, data, i, j);
|
||||
|
||||
bscale<Packet>(tRes, result, alpha);
|
||||
|
||||
data.template storePacketBlock<Packet, 4>(i, j, tRes);
|
||||
}
|
||||
|
||||
template<typename DataMapper, typename Index, typename Packet, typename Packetc, const Index accColsC, int N>
|
||||
EIGEN_ALWAYS_INLINE void storeComplexAccumulator(Index i, Index j, const DataMapper& data, const Packet& alphaReal, const Packet& alphaImag, __vector_quad* accReal, __vector_quad* accImag)
|
||||
{
|
||||
PacketBlock<Packet, 4> resultReal, resultImag;
|
||||
__builtin_mma_disassemble_acc(&resultReal.packet, accReal);
|
||||
__builtin_mma_disassemble_acc(&resultImag.packet, accImag);
|
||||
|
||||
PacketBlock<Packetc, 8> tRes;
|
||||
bload<DataMapper, Packetc, Index, accColsC, N, ColMajor>(tRes, data, i, j);
|
||||
|
||||
PacketBlock<Packet,4> taccReal, taccImag;
|
||||
bscalec<Packet,4>(resultReal, resultImag, alphaReal, alphaImag, taccReal, taccImag);
|
||||
|
||||
PacketBlock<Packetc, 4> acc1, acc2;
|
||||
bcouple<Packet, Packetc>(taccReal, taccImag, tRes, acc1, acc2);
|
||||
|
||||
data.template storePacketBlock<Packetc, 4>(i + N*accColsC, j, acc1);
|
||||
data.template storePacketBlock<Packetc, 4>(i + (N+1)*accColsC, j, acc2);
|
||||
}
|
||||
|
||||
// Defaults to float32, since Eigen still supports C++03 we can't use default template arguments
|
||||
template<typename LhsPacket, typename RhsPacket, bool NegativeAccumulate>
|
||||
EIGEN_ALWAYS_INLINE void pgerMMA(__vector_quad* acc, const RhsPacket& a, const LhsPacket& b)
|
||||
{
|
||||
if(NegativeAccumulate)
|
||||
{
|
||||
__builtin_mma_xvf32gernp(acc, (__vector unsigned char)a, (__vector unsigned char)b);
|
||||
} else {
|
||||
__builtin_mma_xvf32gerpp(acc, (__vector unsigned char)a, (__vector unsigned char)b);
|
||||
}
|
||||
}
|
||||
|
||||
template<typename LhsPacket, typename RhsPacket, bool NegativeAccumulate>
|
||||
EIGEN_ALWAYS_INLINE void pgerMMA(__vector_quad* acc, const PacketBlock<Packet2d,2>& a, const Packet2d& b)
|
||||
{
|
||||
__vector_pair* a0 = (__vector_pair *)(&a.packet[0]);
|
||||
if(NegativeAccumulate)
|
||||
{
|
||||
__builtin_mma_xvf64gernp(acc, *a0, (__vector unsigned char)b);
|
||||
} else {
|
||||
__builtin_mma_xvf64gerpp(acc, *a0, (__vector unsigned char)b);
|
||||
}
|
||||
}
|
||||
|
||||
template<typename LhsPacket, typename RhsPacket, bool NegativeAccumulate>
|
||||
EIGEN_ALWAYS_INLINE void pgerMMA(__vector_quad* acc, const __vector_pair& a, const Packet2d& b)
|
||||
{
|
||||
if(NegativeAccumulate)
|
||||
{
|
||||
__builtin_mma_xvf64gernp(acc, (__vector_pair)a, (__vector unsigned char)b);
|
||||
} else {
|
||||
__builtin_mma_xvf64gerpp(acc, (__vector_pair)a, (__vector unsigned char)b);
|
||||
}
|
||||
}
|
||||
|
||||
template<typename LhsPacket, typename RhsPacket, bool NegativeAccumulate>
|
||||
EIGEN_ALWAYS_INLINE void pgerMMA(__vector_quad*, const __vector_pair&, const Packet4f&)
|
||||
{
|
||||
// Just for compilation
|
||||
}
|
||||
|
||||
template<typename Scalar, typename Packet, typename RhsPacket, bool ConjugateLhs, bool ConjugateRhs, bool LhsIsReal, bool RhsIsReal>
|
||||
EIGEN_ALWAYS_INLINE void pgercMMA(__vector_quad* accReal, __vector_quad* accImag, const Packet& lhsV, const Packet& lhsVi, const RhsPacket& rhsV, const RhsPacket& rhsVi)
|
||||
{
|
||||
pgerMMA<Packet, RhsPacket, false>(accReal, rhsV, lhsV);
|
||||
if(LhsIsReal) {
|
||||
pgerMMA<Packet, RhsPacket, ConjugateRhs>(accImag, rhsVi, lhsV);
|
||||
} else {
|
||||
if(!RhsIsReal) {
|
||||
pgerMMA<Packet, RhsPacket, ConjugateLhs == ConjugateRhs>(accReal, rhsVi, lhsVi);
|
||||
pgerMMA<Packet, RhsPacket, ConjugateRhs>(accImag, rhsVi, lhsV);
|
||||
} else {
|
||||
EIGEN_UNUSED_VARIABLE(rhsVi);
|
||||
}
|
||||
pgerMMA<Packet, RhsPacket, ConjugateLhs>(accImag, rhsV, lhsVi);
|
||||
}
|
||||
}
|
||||
|
||||
// This is necessary because ploadRhs for double returns a pair of vectors when MMA is enabled.
|
||||
template<typename Scalar, typename Packet>
|
||||
EIGEN_ALWAYS_INLINE void ploadRhsMMA(const Scalar* rhs, Packet& rhsV)
|
||||
{
|
||||
rhsV = ploadRhs<Scalar, Packet>((const Scalar*)(rhs));
|
||||
}
|
||||
|
||||
template<>
|
||||
EIGEN_ALWAYS_INLINE void ploadRhsMMA<double, PacketBlock<Packet2d, 2> >(const double* rhs, PacketBlock<Packet2d, 2>& rhsV)
|
||||
{
|
||||
rhsV.packet[0] = ploadRhs<double, Packet2d>((const double *)((Packet2d *)rhs ));
|
||||
rhsV.packet[1] = ploadRhs<double, Packet2d>((const double *)(((Packet2d *)rhs) + 1));
|
||||
}
|
||||
|
||||
template<>
|
||||
EIGEN_ALWAYS_INLINE void ploadRhsMMA<double, __vector_pair>(const double* rhs, __vector_pair& rhsV)
|
||||
{
|
||||
#if EIGEN_COMP_LLVM
|
||||
__builtin_vsx_assemble_pair(&rhsV,
|
||||
(__vector unsigned char)(ploadRhs<double, Packet2d>((const double *)(((Packet2d *)rhs) + 1))),
|
||||
(__vector unsigned char)(ploadRhs<double, Packet2d>((const double *)((Packet2d *)rhs ))));
|
||||
#else
|
||||
__asm__ ("lxvp %x0,%1" : "=wa" (rhsV) : "Y" (*rhs));
|
||||
#endif
|
||||
}
|
||||
|
||||
template<>
|
||||
EIGEN_ALWAYS_INLINE void ploadRhsMMA(const float*, __vector_pair&)
|
||||
{
|
||||
// Just for compilation
|
||||
}
|
||||
|
||||
// PEEL_MMA loop factor.
|
||||
#define PEEL_MMA 7
|
||||
|
||||
#define MICRO_MMA_UNROLL(func) \
|
||||
func(0) func(1) func(2) func(3) func(4) func(5) func(6) func(7)
|
||||
|
||||
#define MICRO_MMA_LOAD_ONE(iter) \
|
||||
if (unroll_factor > iter) { \
|
||||
lhsV##iter = ploadLhs<Scalar, Packet>(lhs_ptr##iter); \
|
||||
lhs_ptr##iter += accCols; \
|
||||
} else { \
|
||||
EIGEN_UNUSED_VARIABLE(lhsV##iter); \
|
||||
}
|
||||
|
||||
#define MICRO_MMA_WORK_ONE(iter, type, peel) \
|
||||
if (unroll_factor > iter) { \
|
||||
pgerMMA<Packet, type, false>(&accZero##iter, rhsV##peel, lhsV##iter); \
|
||||
}
|
||||
|
||||
#define MICRO_MMA_TYPE_PEEL(func, func2, type, peel) \
|
||||
if (PEEL_MMA > peel) { \
|
||||
Packet lhsV0, lhsV1, lhsV2, lhsV3, lhsV4, lhsV5, lhsV6, lhsV7; \
|
||||
ploadRhsMMA<Scalar, type>(rhs_ptr + (accRows * peel), rhsV##peel); \
|
||||
MICRO_MMA_UNROLL(func2); \
|
||||
func(0,type,peel) func(1,type,peel) func(2,type,peel) func(3,type,peel) \
|
||||
func(4,type,peel) func(5,type,peel) func(6,type,peel) func(7,type,peel) \
|
||||
} else { \
|
||||
EIGEN_UNUSED_VARIABLE(rhsV##peel); \
|
||||
}
|
||||
|
||||
#define MICRO_MMA_UNROLL_TYPE_PEEL(func, func2, type) \
|
||||
type rhsV0, rhsV1, rhsV2, rhsV3, rhsV4, rhsV5, rhsV6, rhsV7, rhsV8, rhsV9; \
|
||||
MICRO_MMA_TYPE_PEEL(func,func2,type,0); MICRO_MMA_TYPE_PEEL(func,func2,type,1); \
|
||||
MICRO_MMA_TYPE_PEEL(func,func2,type,2); MICRO_MMA_TYPE_PEEL(func,func2,type,3); \
|
||||
MICRO_MMA_TYPE_PEEL(func,func2,type,4); MICRO_MMA_TYPE_PEEL(func,func2,type,5); \
|
||||
MICRO_MMA_TYPE_PEEL(func,func2,type,6); MICRO_MMA_TYPE_PEEL(func,func2,type,7); \
|
||||
MICRO_MMA_TYPE_PEEL(func,func2,type,8); MICRO_MMA_TYPE_PEEL(func,func2,type,9);
|
||||
|
||||
#define MICRO_MMA_UNROLL_TYPE_ONE(func, func2, type) \
|
||||
type rhsV0; \
|
||||
MICRO_MMA_TYPE_PEEL(func,func2,type,0);
|
||||
|
||||
#define MICRO_MMA_ONE_PEEL \
|
||||
if (sizeof(Scalar) == sizeof(float)) { \
|
||||
MICRO_MMA_UNROLL_TYPE_PEEL(MICRO_MMA_WORK_ONE, MICRO_MMA_LOAD_ONE, RhsPacket); \
|
||||
} else { \
|
||||
MICRO_MMA_UNROLL_TYPE_PEEL(MICRO_MMA_WORK_ONE, MICRO_MMA_LOAD_ONE, __vector_pair); \
|
||||
} \
|
||||
rhs_ptr += (accRows * PEEL_MMA);
|
||||
|
||||
#define MICRO_MMA_ONE \
|
||||
if (sizeof(Scalar) == sizeof(float)) { \
|
||||
MICRO_MMA_UNROLL_TYPE_ONE(MICRO_MMA_WORK_ONE, MICRO_MMA_LOAD_ONE, RhsPacket); \
|
||||
} else { \
|
||||
MICRO_MMA_UNROLL_TYPE_ONE(MICRO_MMA_WORK_ONE, MICRO_MMA_LOAD_ONE, __vector_pair); \
|
||||
} \
|
||||
rhs_ptr += accRows;
|
||||
|
||||
#define MICRO_MMA_DST_PTR_ONE(iter) \
|
||||
if (unroll_factor > iter) { \
|
||||
bsetzeroMMA<Scalar, Packet>(&accZero##iter); \
|
||||
} else { \
|
||||
EIGEN_UNUSED_VARIABLE(accZero##iter); \
|
||||
}
|
||||
|
||||
#define MICRO_MMA_DST_PTR MICRO_MMA_UNROLL(MICRO_MMA_DST_PTR_ONE)
|
||||
|
||||
#define MICRO_MMA_SRC_PTR_ONE(iter) \
|
||||
if (unroll_factor > iter) { \
|
||||
lhs_ptr##iter = lhs_base + ( (row/accCols) + iter )*strideA*accCols + accCols*offsetA; \
|
||||
} else { \
|
||||
EIGEN_UNUSED_VARIABLE(lhs_ptr##iter); \
|
||||
}
|
||||
|
||||
#define MICRO_MMA_SRC_PTR MICRO_MMA_UNROLL(MICRO_MMA_SRC_PTR_ONE)
|
||||
|
||||
#define MICRO_MMA_PREFETCH_ONE(iter) \
|
||||
if (unroll_factor > iter) { \
|
||||
EIGEN_POWER_PREFETCH(lhs_ptr##iter); \
|
||||
}
|
||||
|
||||
#define MICRO_MMA_PREFETCH MICRO_MMA_UNROLL(MICRO_MMA_PREFETCH_ONE)
|
||||
|
||||
#define MICRO_MMA_STORE_ONE(iter) \
|
||||
if (unroll_factor > iter) { \
|
||||
storeAccumulator<DataMapper, Index, Packet, accCols>(row + iter*accCols, col, res, pAlpha, &accZero##iter); \
|
||||
}
|
||||
|
||||
#define MICRO_MMA_STORE MICRO_MMA_UNROLL(MICRO_MMA_STORE_ONE)
|
||||
|
||||
template<int unroll_factor, typename Scalar, typename Packet, typename RhsPacket, typename DataMapper, typename Index, const Index accRows, const Index accCols>
|
||||
EIGEN_STRONG_INLINE void gemm_unrolled_MMA_iteration(
|
||||
const DataMapper& res,
|
||||
const Scalar* lhs_base,
|
||||
const Scalar* rhs_base,
|
||||
Index depth,
|
||||
Index strideA,
|
||||
Index offsetA,
|
||||
Index& row,
|
||||
Index col,
|
||||
const Packet& pAlpha)
|
||||
{
|
||||
const Scalar* rhs_ptr = rhs_base;
|
||||
const Scalar* lhs_ptr0 = NULL, * lhs_ptr1 = NULL, * lhs_ptr2 = NULL, * lhs_ptr3 = NULL, * lhs_ptr4 = NULL, * lhs_ptr5 = NULL, * lhs_ptr6 = NULL, * lhs_ptr7 = NULL;
|
||||
__vector_quad accZero0, accZero1, accZero2, accZero3, accZero4, accZero5, accZero6, accZero7;
|
||||
|
||||
MICRO_MMA_SRC_PTR
|
||||
MICRO_MMA_DST_PTR
|
||||
|
||||
Index k = 0;
|
||||
for(; k + PEEL_MMA <= depth; k+= PEEL_MMA)
|
||||
{
|
||||
EIGEN_POWER_PREFETCH(rhs_ptr);
|
||||
MICRO_MMA_PREFETCH
|
||||
MICRO_MMA_ONE_PEEL
|
||||
}
|
||||
for(; k < depth; k++)
|
||||
{
|
||||
MICRO_MMA_ONE
|
||||
}
|
||||
MICRO_MMA_STORE
|
||||
|
||||
row += unroll_factor*accCols;
|
||||
}
|
||||
|
||||
template<typename Scalar, typename Index, typename Packet, typename RhsPacket, typename DataMapper, const Index accRows, const Index accCols>
|
||||
void gemmMMA(const DataMapper& res, const Scalar* blockA, const Scalar* blockB, Index rows, Index depth, Index cols, Scalar alpha, Index strideA, Index strideB, Index offsetA, Index offsetB)
|
||||
{
|
||||
const Index remaining_rows = rows % accCols;
|
||||
const Index remaining_cols = cols % accRows;
|
||||
|
||||
if( strideA == -1 ) strideA = depth;
|
||||
if( strideB == -1 ) strideB = depth;
|
||||
|
||||
const Packet pAlpha = pset1<Packet>(alpha);
|
||||
const Packet pMask = bmask<Packet>((const int)(remaining_rows));
|
||||
|
||||
Index col = 0;
|
||||
for(; col + accRows <= cols; col += accRows)
|
||||
{
|
||||
const Scalar* rhs_base = blockB + col*strideB + accRows*offsetB;
|
||||
const Scalar* lhs_base = blockA;
|
||||
|
||||
Index row = 0;
|
||||
#define MAX_MMA_UNROLL 7
|
||||
while(row + MAX_MMA_UNROLL*accCols <= rows) {
|
||||
gemm_unrolled_MMA_iteration<MAX_MMA_UNROLL, Scalar, Packet, RhsPacket, DataMapper, Index, accRows, accCols>(res, lhs_base, rhs_base, depth, strideA, offsetA, row, col, pAlpha);
|
||||
}
|
||||
switch( (rows-row)/accCols ) {
|
||||
#if MAX_MMA_UNROLL > 7
|
||||
case 7:
|
||||
gemm_unrolled_MMA_iteration<7, Scalar, Packet, RhsPacket, DataMapper, Index, accRows, accCols>(res, lhs_base, rhs_base, depth, strideA, offsetA, row, col, pAlpha);
|
||||
break;
|
||||
#endif
|
||||
#if MAX_MMA_UNROLL > 6
|
||||
case 6:
|
||||
gemm_unrolled_MMA_iteration<6, Scalar, Packet, RhsPacket, DataMapper, Index, accRows, accCols>(res, lhs_base, rhs_base, depth, strideA, offsetA, row, col, pAlpha);
|
||||
break;
|
||||
#endif
|
||||
#if MAX_MMA_UNROLL > 5
|
||||
case 5:
|
||||
gemm_unrolled_MMA_iteration<5, Scalar, Packet, RhsPacket, DataMapper, Index, accRows, accCols>(res, lhs_base, rhs_base, depth, strideA, offsetA, row, col, pAlpha);
|
||||
break;
|
||||
#endif
|
||||
#if MAX_MMA_UNROLL > 4
|
||||
case 4:
|
||||
gemm_unrolled_MMA_iteration<4, Scalar, Packet, RhsPacket, DataMapper, Index, accRows, accCols>(res, lhs_base, rhs_base, depth, strideA, offsetA, row, col, pAlpha);
|
||||
break;
|
||||
#endif
|
||||
#if MAX_MMA_UNROLL > 3
|
||||
case 3:
|
||||
gemm_unrolled_MMA_iteration<3, Scalar, Packet, RhsPacket, DataMapper, Index, accRows, accCols>(res, lhs_base, rhs_base, depth, strideA, offsetA, row, col, pAlpha);
|
||||
break;
|
||||
#endif
|
||||
#if MAX_MMA_UNROLL > 2
|
||||
case 2:
|
||||
gemm_unrolled_MMA_iteration<2, Scalar, Packet, RhsPacket, DataMapper, Index, accRows, accCols>(res, lhs_base, rhs_base, depth, strideA, offsetA, row, col, pAlpha);
|
||||
break;
|
||||
#endif
|
||||
#if MAX_MMA_UNROLL > 1
|
||||
case 1:
|
||||
gemm_unrolled_MMA_iteration<1, Scalar, Packet, RhsPacket, DataMapper, Index, accRows, accCols>(res, lhs_base, rhs_base, depth, strideA, offsetA, row, col, pAlpha);
|
||||
break;
|
||||
#endif
|
||||
default:
|
||||
break;
|
||||
}
|
||||
#undef MAX_MMA_UNROLL
|
||||
|
||||
if(remaining_rows > 0)
|
||||
{
|
||||
gemm_extra_row<Scalar, Packet, DataMapper, Index, accRows, accCols>(res, lhs_base, rhs_base, depth, strideA, offsetA, row, col, rows, cols, remaining_rows, pAlpha, pMask);
|
||||
}
|
||||
}
|
||||
|
||||
if(remaining_cols > 0)
|
||||
{
|
||||
const Scalar* rhs_base = blockB + col*strideB + remaining_cols*offsetB;
|
||||
const Scalar* lhs_base = blockA;
|
||||
|
||||
for(; col < cols; col++)
|
||||
{
|
||||
Index row = 0;
|
||||
|
||||
gemm_unrolled_col<Scalar, Packet, DataMapper, Index, accCols>(res, lhs_base, rhs_base, depth, strideA, offsetA, row, rows, col, remaining_cols, pAlpha);
|
||||
|
||||
if (remaining_rows > 0)
|
||||
{
|
||||
gemm_extra_col<Scalar, Packet, DataMapper, Index, accRows>(res, lhs_base, rhs_base, depth, strideA, offsetA, row, col, remaining_rows, remaining_cols, pAlpha);
|
||||
}
|
||||
rhs_base++;
|
||||
}
|
||||
}
|
||||
}
|
||||
|
||||
#define accColsC (accCols / 2)
|
||||
#define advanceRows ((LhsIsReal) ? 1 : 2)
|
||||
#define advanceCols ((RhsIsReal) ? 1 : 2)
|
||||
|
||||
// PEEL_COMPLEX_MMA loop factor.
|
||||
#define PEEL_COMPLEX_MMA 7
|
||||
|
||||
#define MICRO_COMPLEX_MMA_UNROLL(func) \
|
||||
func(0) func(1) func(2) func(3) func(4)
|
||||
|
||||
#define MICRO_COMPLEX_MMA_LOAD_ONE(iter) \
|
||||
if (unroll_factor > iter) { \
|
||||
lhsV##iter = ploadLhs<Scalar, Packet>(lhs_ptr_real##iter); \
|
||||
lhs_ptr_real##iter += accCols; \
|
||||
if(!LhsIsReal) { \
|
||||
lhsVi##iter = ploadLhs<Scalar, Packet>(lhs_ptr_imag##iter); \
|
||||
lhs_ptr_imag##iter += accCols; \
|
||||
} else { \
|
||||
EIGEN_UNUSED_VARIABLE(lhsVi##iter); \
|
||||
} \
|
||||
} else { \
|
||||
EIGEN_UNUSED_VARIABLE(lhsV##iter); \
|
||||
EIGEN_UNUSED_VARIABLE(lhsVi##iter); \
|
||||
}
|
||||
|
||||
#define MICRO_COMPLEX_MMA_WORK_ONE(iter, type, peel) \
|
||||
if (unroll_factor > iter) { \
|
||||
pgercMMA<Scalar, Packet, type, ConjugateLhs, ConjugateRhs, LhsIsReal, RhsIsReal>(&accReal##iter, &accImag##iter, lhsV##iter, lhsVi##iter, rhsV##peel, rhsVi##peel); \
|
||||
}
|
||||
|
||||
#define MICRO_COMPLEX_MMA_TYPE_PEEL(func, func2, type, peel) \
|
||||
if (PEEL_COMPLEX_MMA > peel) { \
|
||||
Packet lhsV0, lhsV1, lhsV2, lhsV3, lhsV4; \
|
||||
Packet lhsVi0, lhsVi1, lhsVi2, lhsVi3, lhsVi4; \
|
||||
ploadRhsMMA<Scalar, type>(rhs_ptr_real + (accRows * peel), rhsV##peel); \
|
||||
if(!RhsIsReal) { \
|
||||
ploadRhsMMA<Scalar, type>(rhs_ptr_imag + (accRows * peel), rhsVi##peel); \
|
||||
} else { \
|
||||
EIGEN_UNUSED_VARIABLE(rhsVi##peel); \
|
||||
} \
|
||||
MICRO_COMPLEX_MMA_UNROLL(func2); \
|
||||
func(0,type,peel) func(1,type,peel) func(2,type,peel) func(3,type,peel) func(4,type,peel) \
|
||||
} else { \
|
||||
EIGEN_UNUSED_VARIABLE(rhsV##peel); \
|
||||
EIGEN_UNUSED_VARIABLE(rhsVi##peel); \
|
||||
}
|
||||
|
||||
#define MICRO_COMPLEX_MMA_UNROLL_TYPE_PEEL(func, func2, type) \
|
||||
type rhsV0, rhsV1, rhsV2, rhsV3, rhsV4, rhsV5, rhsV6, rhsV7, rhsV8, rhsV9; \
|
||||
type rhsVi0, rhsVi1, rhsVi2, rhsVi3, rhsVi4, rhsVi5, rhsVi6, rhsVi7, rhsVi8, rhsVi9; \
|
||||
MICRO_COMPLEX_MMA_TYPE_PEEL(func,func2,type,0); MICRO_COMPLEX_MMA_TYPE_PEEL(func,func2,type,1); \
|
||||
MICRO_COMPLEX_MMA_TYPE_PEEL(func,func2,type,2); MICRO_COMPLEX_MMA_TYPE_PEEL(func,func2,type,3); \
|
||||
MICRO_COMPLEX_MMA_TYPE_PEEL(func,func2,type,4); MICRO_COMPLEX_MMA_TYPE_PEEL(func,func2,type,5); \
|
||||
MICRO_COMPLEX_MMA_TYPE_PEEL(func,func2,type,6); MICRO_COMPLEX_MMA_TYPE_PEEL(func,func2,type,7); \
|
||||
MICRO_COMPLEX_MMA_TYPE_PEEL(func,func2,type,8); MICRO_COMPLEX_MMA_TYPE_PEEL(func,func2,type,9);
|
||||
|
||||
#define MICRO_COMPLEX_MMA_UNROLL_TYPE_ONE(func, func2, type) \
|
||||
type rhsV0, rhsVi0; \
|
||||
MICRO_COMPLEX_MMA_TYPE_PEEL(func,func2,type,0);
|
||||
|
||||
#define MICRO_COMPLEX_MMA_ONE_PEEL \
|
||||
if (sizeof(Scalar) == sizeof(float)) { \
|
||||
MICRO_COMPLEX_MMA_UNROLL_TYPE_PEEL(MICRO_COMPLEX_MMA_WORK_ONE, MICRO_COMPLEX_MMA_LOAD_ONE, RhsPacket); \
|
||||
} else { \
|
||||
MICRO_COMPLEX_MMA_UNROLL_TYPE_PEEL(MICRO_COMPLEX_MMA_WORK_ONE, MICRO_COMPLEX_MMA_LOAD_ONE, __vector_pair); \
|
||||
} \
|
||||
rhs_ptr_real += (accRows * PEEL_COMPLEX_MMA); \
|
||||
if(!RhsIsReal) rhs_ptr_imag += (accRows * PEEL_COMPLEX_MMA);
|
||||
|
||||
#define MICRO_COMPLEX_MMA_ONE \
|
||||
if (sizeof(Scalar) == sizeof(float)) { \
|
||||
MICRO_COMPLEX_MMA_UNROLL_TYPE_ONE(MICRO_COMPLEX_MMA_WORK_ONE, MICRO_COMPLEX_MMA_LOAD_ONE, RhsPacket); \
|
||||
} else { \
|
||||
MICRO_COMPLEX_MMA_UNROLL_TYPE_ONE(MICRO_COMPLEX_MMA_WORK_ONE, MICRO_COMPLEX_MMA_LOAD_ONE, __vector_pair); \
|
||||
} \
|
||||
rhs_ptr_real += accRows; \
|
||||
if(!RhsIsReal) rhs_ptr_imag += accRows;
|
||||
|
||||
#define MICRO_COMPLEX_MMA_DST_PTR_ONE(iter) \
|
||||
if (unroll_factor > iter) { \
|
||||
bsetzeroMMA<Scalar, Packet>(&accReal##iter); \
|
||||
bsetzeroMMA<Scalar, Packet>(&accImag##iter); \
|
||||
} else { \
|
||||
EIGEN_UNUSED_VARIABLE(accReal##iter); \
|
||||
EIGEN_UNUSED_VARIABLE(accImag##iter); \
|
||||
}
|
||||
|
||||
#define MICRO_COMPLEX_MMA_DST_PTR MICRO_COMPLEX_MMA_UNROLL(MICRO_COMPLEX_MMA_DST_PTR_ONE)
|
||||
|
||||
#define MICRO_COMPLEX_MMA_SRC_PTR_ONE(iter) \
|
||||
if (unroll_factor > iter) { \
|
||||
lhs_ptr_real##iter = lhs_base + ( ((advanceRows*row)/accCols) + iter*advanceRows )*strideA*accCols + accCols*offsetA; \
|
||||
if(!LhsIsReal) { \
|
||||
lhs_ptr_imag##iter = lhs_ptr_real##iter + accCols*strideA; \
|
||||
} else { \
|
||||
EIGEN_UNUSED_VARIABLE(lhs_ptr_imag##iter); \
|
||||
} \
|
||||
} else { \
|
||||
EIGEN_UNUSED_VARIABLE(lhs_ptr_real##iter); \
|
||||
EIGEN_UNUSED_VARIABLE(lhs_ptr_imag##iter); \
|
||||
}
|
||||
|
||||
#define MICRO_COMPLEX_MMA_SRC_PTR MICRO_COMPLEX_MMA_UNROLL(MICRO_COMPLEX_MMA_SRC_PTR_ONE)
|
||||
|
||||
#define MICRO_COMPLEX_MMA_PREFETCH_ONE(iter) \
|
||||
if (unroll_factor > iter) { \
|
||||
EIGEN_POWER_PREFETCH(lhs_ptr_real##iter); \
|
||||
if(!LhsIsReal) { \
|
||||
EIGEN_POWER_PREFETCH(lhs_ptr_imag##iter); \
|
||||
} \
|
||||
}
|
||||
|
||||
#define MICRO_COMPLEX_MMA_PREFETCH MICRO_COMPLEX_MMA_UNROLL(MICRO_COMPLEX_MMA_PREFETCH_ONE)
|
||||
|
||||
#define MICRO_COMPLEX_MMA_STORE_ONE(iter) \
|
||||
if (unroll_factor > iter) { \
|
||||
storeComplexAccumulator<DataMapper, Index, Packet, Packetc, accColsC, 0>(row + iter*accCols, col, res, pAlphaReal, pAlphaImag, &accReal##iter, &accImag##iter); \
|
||||
}
|
||||
|
||||
#define MICRO_COMPLEX_MMA_STORE MICRO_COMPLEX_MMA_UNROLL(MICRO_COMPLEX_MMA_STORE_ONE)
|
||||
|
||||
template<int unroll_factor, typename Scalar, typename Packet, typename Packetc, typename RhsPacket, typename DataMapper, typename Index, const Index accRows, const Index accCols, bool ConjugateLhs, bool ConjugateRhs, bool LhsIsReal, bool RhsIsReal>
|
||||
EIGEN_STRONG_INLINE void gemm_complex_unrolled_MMA_iteration(
|
||||
const DataMapper& res,
|
||||
const Scalar* lhs_base,
|
||||
const Scalar* rhs_base,
|
||||
Index depth,
|
||||
Index strideA,
|
||||
Index offsetA,
|
||||
Index strideB,
|
||||
Index& row,
|
||||
Index col,
|
||||
const Packet& pAlphaReal,
|
||||
const Packet& pAlphaImag)
|
||||
{
|
||||
const Scalar* rhs_ptr_real = rhs_base;
|
||||
const Scalar* rhs_ptr_imag;
|
||||
if(!RhsIsReal) {
|
||||
rhs_ptr_imag = rhs_base + accRows*strideB;
|
||||
} else {
|
||||
EIGEN_UNUSED_VARIABLE(rhs_ptr_imag);
|
||||
}
|
||||
const Scalar* lhs_ptr_real0 = NULL, * lhs_ptr_imag0 = NULL, * lhs_ptr_real1 = NULL, * lhs_ptr_imag1 = NULL;
|
||||
const Scalar* lhs_ptr_real2 = NULL, * lhs_ptr_imag2 = NULL, * lhs_ptr_real3 = NULL, * lhs_ptr_imag3 = NULL;
|
||||
const Scalar* lhs_ptr_real4 = NULL, * lhs_ptr_imag4 = NULL;
|
||||
__vector_quad accReal0, accImag0, accReal1, accImag1, accReal2, accImag2, accReal3, accImag3, accReal4, accImag4;
|
||||
|
||||
MICRO_COMPLEX_MMA_SRC_PTR
|
||||
MICRO_COMPLEX_MMA_DST_PTR
|
||||
|
||||
Index k = 0;
|
||||
for(; k + PEEL_COMPLEX_MMA <= depth; k+= PEEL_COMPLEX_MMA)
|
||||
{
|
||||
EIGEN_POWER_PREFETCH(rhs_ptr_real);
|
||||
if(!RhsIsReal) {
|
||||
EIGEN_POWER_PREFETCH(rhs_ptr_imag);
|
||||
}
|
||||
MICRO_COMPLEX_MMA_PREFETCH
|
||||
MICRO_COMPLEX_MMA_ONE_PEEL
|
||||
}
|
||||
for(; k < depth; k++)
|
||||
{
|
||||
MICRO_COMPLEX_MMA_ONE
|
||||
}
|
||||
MICRO_COMPLEX_MMA_STORE
|
||||
|
||||
row += unroll_factor*accCols;
|
||||
}
|
||||
|
||||
template<typename LhsScalar, typename RhsScalar, typename Scalarc, typename Scalar, typename Index, typename Packet, typename Packetc, typename RhsPacket, typename DataMapper, const Index accRows, const Index accCols, bool ConjugateLhs, bool ConjugateRhs, bool LhsIsReal, bool RhsIsReal>
|
||||
void gemm_complexMMA(const DataMapper& res, const LhsScalar* blockAc, const RhsScalar* blockBc, Index rows, Index depth, Index cols, Scalarc alpha, Index strideA, Index strideB, Index offsetA, Index offsetB)
|
||||
{
|
||||
const Index remaining_rows = rows % accCols;
|
||||
const Index remaining_cols = cols % accRows;
|
||||
|
||||
if( strideA == -1 ) strideA = depth;
|
||||
if( strideB == -1 ) strideB = depth;
|
||||
|
||||
const Packet pAlphaReal = pset1<Packet>(alpha.real());
|
||||
const Packet pAlphaImag = pset1<Packet>(alpha.imag());
|
||||
const Packet pMask = bmask<Packet>((const int)(remaining_rows));
|
||||
|
||||
const Scalar* blockA = (Scalar *) blockAc;
|
||||
const Scalar* blockB = (Scalar *) blockBc;
|
||||
|
||||
Index col = 0;
|
||||
for(; col + accRows <= cols; col += accRows)
|
||||
{
|
||||
const Scalar* rhs_base = blockB + advanceCols*col*strideB + accRows*offsetB;
|
||||
const Scalar* lhs_base = blockA;
|
||||
Index row = 0;
|
||||
|
||||
#define MAX_COMPLEX_MMA_UNROLL 4
|
||||
while(row + MAX_COMPLEX_MMA_UNROLL*accCols <= rows) {
|
||||
gemm_complex_unrolled_MMA_iteration<MAX_COMPLEX_MMA_UNROLL, Scalar, Packet, Packetc, RhsPacket, DataMapper, Index, accRows, accCols, ConjugateLhs, ConjugateRhs, LhsIsReal, RhsIsReal>(res, lhs_base, rhs_base, depth, strideA, offsetA, strideB, row, col, pAlphaReal, pAlphaImag);
|
||||
}
|
||||
switch( (rows-row)/accCols ) {
|
||||
#if MAX_COMPLEX_MMA_UNROLL > 4
|
||||
case 4:
|
||||
gemm_complex_unrolled_MMA_iteration<4, Scalar, Packet, Packetc, RhsPacket, DataMapper, Index, accRows, accCols, ConjugateLhs, ConjugateRhs, LhsIsReal, RhsIsReal>(res, lhs_base, rhs_base, depth, strideA, offsetA, strideB, row, col, pAlphaReal, pAlphaImag);
|
||||
break;
|
||||
#endif
|
||||
#if MAX_COMPLEX_MMA_UNROLL > 3
|
||||
case 3:
|
||||
gemm_complex_unrolled_MMA_iteration<3, Scalar, Packet, Packetc, RhsPacket, DataMapper, Index, accRows, accCols, ConjugateLhs, ConjugateRhs, LhsIsReal, RhsIsReal>(res, lhs_base, rhs_base, depth, strideA, offsetA, strideB, row, col, pAlphaReal, pAlphaImag);
|
||||
break;
|
||||
#endif
|
||||
#if MAX_COMPLEX_MMA_UNROLL > 2
|
||||
case 2:
|
||||
gemm_complex_unrolled_MMA_iteration<2, Scalar, Packet, Packetc, RhsPacket, DataMapper, Index, accRows, accCols, ConjugateLhs, ConjugateRhs, LhsIsReal, RhsIsReal>(res, lhs_base, rhs_base, depth, strideA, offsetA, strideB, row, col, pAlphaReal, pAlphaImag);
|
||||
break;
|
||||
#endif
|
||||
#if MAX_COMPLEX_MMA_UNROLL > 1
|
||||
case 1:
|
||||
gemm_complex_unrolled_MMA_iteration<1, Scalar, Packet, Packetc, RhsPacket, DataMapper, Index, accRows, accCols, ConjugateLhs, ConjugateRhs, LhsIsReal, RhsIsReal>(res, lhs_base, rhs_base, depth, strideA, offsetA, strideB, row, col, pAlphaReal, pAlphaImag);
|
||||
break;
|
||||
#endif
|
||||
default:
|
||||
break;
|
||||
}
|
||||
#undef MAX_COMPLEX_MMA_UNROLL
|
||||
|
||||
if(remaining_rows > 0)
|
||||
{
|
||||
gemm_complex_extra_row<Scalar, Packet, Packetc, DataMapper, Index, accRows, accCols, ConjugateLhs, ConjugateRhs, LhsIsReal, RhsIsReal>(res, lhs_base, rhs_base, depth, strideA, offsetA, strideB, row, col, rows, cols, remaining_rows, pAlphaReal, pAlphaImag, pMask);
|
||||
}
|
||||
}
|
||||
|
||||
if(remaining_cols > 0)
|
||||
{
|
||||
const Scalar* rhs_base = blockB + advanceCols*col*strideB + remaining_cols*offsetB;
|
||||
const Scalar* lhs_base = blockA;
|
||||
|
||||
for(; col < cols; col++)
|
||||
{
|
||||
Index row = 0;
|
||||
|
||||
gemm_complex_unrolled_col<Scalar, Packet, Packetc, DataMapper, Index, accCols, ConjugateLhs, ConjugateRhs, LhsIsReal, RhsIsReal>(res, lhs_base, rhs_base, depth, strideA, offsetA, strideB, row, rows, col, remaining_cols, pAlphaReal, pAlphaImag);
|
||||
|
||||
if (remaining_rows > 0)
|
||||
{
|
||||
gemm_complex_extra_col<Scalar, Packet, Packetc, DataMapper, Index, accRows, accCols, ConjugateLhs, ConjugateRhs, LhsIsReal, RhsIsReal>(res, lhs_base, rhs_base, depth, strideA, offsetA, strideB, row, col, remaining_rows, remaining_cols, pAlphaReal, pAlphaImag);
|
||||
}
|
||||
rhs_base++;
|
||||
}
|
||||
}
|
||||
}
|
||||
|
||||
#undef accColsC
|
||||
#undef advanceRows
|
||||
#undef advanceCols
|
||||
|
||||
#pragma GCC reset_options
|
||||
} // end namespace internal
|
||||
|
||||
} // end namespace Eigen
|
||||
|
||||
#endif // EIGEN_MATRIX_PRODUCT_MMA_ALTIVEC_H
|
||||
|
||||
+2711
File diff suppressed because it is too large
Load Diff
@@ -0,0 +1,258 @@
|
||||
// This file is part of Eigen, a lightweight C++ template library
|
||||
// for linear algebra.
|
||||
//
|
||||
// Copyright (C) 2014 Benoit Steiner <benoit.steiner.goog@gmail.com>
|
||||
// Copyright (C) 2021 C. Antonio Sanchez <cantonios@google.com>
|
||||
//
|
||||
// This Source Code Form is subject to the terms of the Mozilla
|
||||
// Public License v. 2.0. If a copy of the MPL was not distributed
|
||||
// with this file, You can obtain one at http://mozilla.org/MPL/2.0/.
|
||||
|
||||
#ifndef EIGEN_COMPLEX_CUDA_H
|
||||
#define EIGEN_COMPLEX_CUDA_H
|
||||
|
||||
// clang-format off
|
||||
// Many std::complex methods such as operator+, operator-, operator* and
|
||||
// operator/ are not constexpr. Due to this, GCC and older versions of clang do
|
||||
// not treat them as device functions and thus Eigen functors making use of
|
||||
// these operators fail to compile. Here, we manually specialize these
|
||||
// operators and functors for complex types when building for CUDA to enable
|
||||
// their use on-device.
|
||||
|
||||
#if defined(EIGEN_CUDACC) && defined(EIGEN_GPU_COMPILE_PHASE)
|
||||
|
||||
// ICC already specializes std::complex<float> and std::complex<double>
|
||||
// operators, preventing us from making them device functions here.
|
||||
// This will lead to silent runtime errors if the operators are used on device.
|
||||
//
|
||||
// To allow std::complex operator use on device, define _OVERRIDE_COMPLEX_SPECIALIZATION_
|
||||
// prior to first inclusion of <complex>. This prevents ICC from adding
|
||||
// its own specializations, so our custom ones below can be used instead.
|
||||
#if !(defined(EIGEN_COMP_ICC) && defined(_USE_COMPLEX_SPECIALIZATION_))
|
||||
|
||||
// Import Eigen's internal operator specializations.
|
||||
#define EIGEN_USING_STD_COMPLEX_OPERATORS \
|
||||
using Eigen::complex_operator_detail::operator+; \
|
||||
using Eigen::complex_operator_detail::operator-; \
|
||||
using Eigen::complex_operator_detail::operator*; \
|
||||
using Eigen::complex_operator_detail::operator/; \
|
||||
using Eigen::complex_operator_detail::operator+=; \
|
||||
using Eigen::complex_operator_detail::operator-=; \
|
||||
using Eigen::complex_operator_detail::operator*=; \
|
||||
using Eigen::complex_operator_detail::operator/=; \
|
||||
using Eigen::complex_operator_detail::operator==; \
|
||||
using Eigen::complex_operator_detail::operator!=;
|
||||
|
||||
namespace Eigen {
|
||||
|
||||
// Specialized std::complex overloads.
|
||||
namespace complex_operator_detail {
|
||||
|
||||
template<typename T>
|
||||
EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
|
||||
std::complex<T> complex_multiply(const std::complex<T>& a, const std::complex<T>& b) {
|
||||
const T a_real = numext::real(a);
|
||||
const T a_imag = numext::imag(a);
|
||||
const T b_real = numext::real(b);
|
||||
const T b_imag = numext::imag(b);
|
||||
return std::complex<T>(
|
||||
a_real * b_real - a_imag * b_imag,
|
||||
a_imag * b_real + a_real * b_imag);
|
||||
}
|
||||
|
||||
template<typename T>
|
||||
EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
|
||||
std::complex<T> complex_divide_fast(const std::complex<T>& a, const std::complex<T>& b) {
|
||||
const T a_real = numext::real(a);
|
||||
const T a_imag = numext::imag(a);
|
||||
const T b_real = numext::real(b);
|
||||
const T b_imag = numext::imag(b);
|
||||
const T norm = (b_real * b_real + b_imag * b_imag);
|
||||
return std::complex<T>((a_real * b_real + a_imag * b_imag) / norm,
|
||||
(a_imag * b_real - a_real * b_imag) / norm);
|
||||
}
|
||||
|
||||
template<typename T>
|
||||
EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
|
||||
std::complex<T> complex_divide_stable(const std::complex<T>& a, const std::complex<T>& b) {
|
||||
const T a_real = numext::real(a);
|
||||
const T a_imag = numext::imag(a);
|
||||
const T b_real = numext::real(b);
|
||||
const T b_imag = numext::imag(b);
|
||||
// Smith's complex division (https://arxiv.org/pdf/1210.4539.pdf),
|
||||
// guards against over/under-flow.
|
||||
const bool scale_imag = numext::abs(b_imag) <= numext::abs(b_real);
|
||||
const T rscale = scale_imag ? T(1) : b_real / b_imag;
|
||||
const T iscale = scale_imag ? b_imag / b_real : T(1);
|
||||
const T denominator = b_real * rscale + b_imag * iscale;
|
||||
return std::complex<T>((a_real * rscale + a_imag * iscale) / denominator,
|
||||
(a_imag * rscale - a_real * iscale) / denominator);
|
||||
}
|
||||
|
||||
template<typename T>
|
||||
EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
|
||||
std::complex<T> complex_divide(const std::complex<T>& a, const std::complex<T>& b) {
|
||||
#if EIGEN_FAST_MATH
|
||||
return complex_divide_fast(a, b);
|
||||
#else
|
||||
return complex_divide_stable(a, b);
|
||||
#endif
|
||||
}
|
||||
|
||||
// NOTE: We cannot specialize compound assignment operators with Scalar T,
|
||||
// (i.e. operator@=(const T&), for @=+,-,*,/)
|
||||
// since they are already specialized for float/double/long double within
|
||||
// the standard <complex> header. We also do not specialize the stream
|
||||
// operators.
|
||||
#define EIGEN_CREATE_STD_COMPLEX_OPERATOR_SPECIALIZATIONS(T) \
|
||||
\
|
||||
EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE \
|
||||
std::complex<T> operator+(const std::complex<T>& a) { return a; } \
|
||||
\
|
||||
EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE \
|
||||
std::complex<T> operator-(const std::complex<T>& a) { \
|
||||
return std::complex<T>(-numext::real(a), -numext::imag(a)); \
|
||||
} \
|
||||
\
|
||||
EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE \
|
||||
std::complex<T> operator+(const std::complex<T>& a, const std::complex<T>& b) { \
|
||||
return std::complex<T>(numext::real(a) + numext::real(b), numext::imag(a) + numext::imag(b)); \
|
||||
} \
|
||||
\
|
||||
EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE \
|
||||
std::complex<T> operator+(const std::complex<T>& a, const T& b) { \
|
||||
return std::complex<T>(numext::real(a) + b, numext::imag(a)); \
|
||||
} \
|
||||
\
|
||||
EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE \
|
||||
std::complex<T> operator+(const T& a, const std::complex<T>& b) { \
|
||||
return std::complex<T>(a + numext::real(b), numext::imag(b)); \
|
||||
} \
|
||||
\
|
||||
EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE \
|
||||
std::complex<T> operator-(const std::complex<T>& a, const std::complex<T>& b) { \
|
||||
return std::complex<T>(numext::real(a) - numext::real(b), numext::imag(a) - numext::imag(b)); \
|
||||
} \
|
||||
\
|
||||
EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE \
|
||||
std::complex<T> operator-(const std::complex<T>& a, const T& b) { \
|
||||
return std::complex<T>(numext::real(a) - b, numext::imag(a)); \
|
||||
} \
|
||||
\
|
||||
EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE \
|
||||
std::complex<T> operator-(const T& a, const std::complex<T>& b) { \
|
||||
return std::complex<T>(a - numext::real(b), -numext::imag(b)); \
|
||||
} \
|
||||
\
|
||||
EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE \
|
||||
std::complex<T> operator*(const std::complex<T>& a, const std::complex<T>& b) { \
|
||||
return complex_multiply(a, b); \
|
||||
} \
|
||||
\
|
||||
EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE \
|
||||
std::complex<T> operator*(const std::complex<T>& a, const T& b) { \
|
||||
return std::complex<T>(numext::real(a) * b, numext::imag(a) * b); \
|
||||
} \
|
||||
\
|
||||
EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE \
|
||||
std::complex<T> operator*(const T& a, const std::complex<T>& b) { \
|
||||
return std::complex<T>(a * numext::real(b), a * numext::imag(b)); \
|
||||
} \
|
||||
\
|
||||
EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE \
|
||||
std::complex<T> operator/(const std::complex<T>& a, const std::complex<T>& b) { \
|
||||
return complex_divide(a, b); \
|
||||
} \
|
||||
\
|
||||
EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE \
|
||||
std::complex<T> operator/(const std::complex<T>& a, const T& b) { \
|
||||
return std::complex<T>(numext::real(a) / b, numext::imag(a) / b); \
|
||||
} \
|
||||
\
|
||||
EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE \
|
||||
std::complex<T> operator/(const T& a, const std::complex<T>& b) { \
|
||||
return complex_divide(std::complex<T>(a, 0), b); \
|
||||
} \
|
||||
\
|
||||
EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE \
|
||||
std::complex<T>& operator+=(std::complex<T>& a, const std::complex<T>& b) { \
|
||||
numext::real_ref(a) += numext::real(b); \
|
||||
numext::imag_ref(a) += numext::imag(b); \
|
||||
return a; \
|
||||
} \
|
||||
\
|
||||
EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE \
|
||||
std::complex<T>& operator-=(std::complex<T>& a, const std::complex<T>& b) { \
|
||||
numext::real_ref(a) -= numext::real(b); \
|
||||
numext::imag_ref(a) -= numext::imag(b); \
|
||||
return a; \
|
||||
} \
|
||||
\
|
||||
EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE \
|
||||
std::complex<T>& operator*=(std::complex<T>& a, const std::complex<T>& b) { \
|
||||
a = complex_multiply(a, b); \
|
||||
return a; \
|
||||
} \
|
||||
\
|
||||
EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE \
|
||||
std::complex<T>& operator/=(std::complex<T>& a, const std::complex<T>& b) { \
|
||||
a = complex_divide(a, b); \
|
||||
return a; \
|
||||
} \
|
||||
\
|
||||
EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE \
|
||||
bool operator==(const std::complex<T>& a, const std::complex<T>& b) { \
|
||||
return numext::real(a) == numext::real(b) && numext::imag(a) == numext::imag(b); \
|
||||
} \
|
||||
\
|
||||
EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE \
|
||||
bool operator==(const std::complex<T>& a, const T& b) { \
|
||||
return numext::real(a) == b && numext::imag(a) == 0; \
|
||||
} \
|
||||
\
|
||||
EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE \
|
||||
bool operator==(const T& a, const std::complex<T>& b) { \
|
||||
return a == numext::real(b) && 0 == numext::imag(b); \
|
||||
} \
|
||||
\
|
||||
EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE \
|
||||
bool operator!=(const std::complex<T>& a, const std::complex<T>& b) { \
|
||||
return !(a == b); \
|
||||
} \
|
||||
\
|
||||
EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE \
|
||||
bool operator!=(const std::complex<T>& a, const T& b) { \
|
||||
return !(a == b); \
|
||||
} \
|
||||
\
|
||||
EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE \
|
||||
bool operator!=(const T& a, const std::complex<T>& b) { \
|
||||
return !(a == b); \
|
||||
}
|
||||
|
||||
// Do not specialize for long double, since that reduces to double on device.
|
||||
EIGEN_CREATE_STD_COMPLEX_OPERATOR_SPECIALIZATIONS(float)
|
||||
EIGEN_CREATE_STD_COMPLEX_OPERATOR_SPECIALIZATIONS(double)
|
||||
|
||||
#undef EIGEN_CREATE_STD_COMPLEX_OPERATOR_SPECIALIZATIONS
|
||||
|
||||
|
||||
} // namespace complex_operator_detail
|
||||
|
||||
EIGEN_USING_STD_COMPLEX_OPERATORS
|
||||
|
||||
namespace numext {
|
||||
EIGEN_USING_STD_COMPLEX_OPERATORS
|
||||
} // namespace numext
|
||||
|
||||
namespace internal {
|
||||
EIGEN_USING_STD_COMPLEX_OPERATORS
|
||||
|
||||
} // namespace internal
|
||||
} // namespace Eigen
|
||||
|
||||
#endif // !(EIGEN_COMP_ICC && _USE_COMPLEX_SPECIALIZATION_)
|
||||
|
||||
#endif // EIGEN_CUDACC && EIGEN_GPU_COMPILE_PHASE
|
||||
|
||||
#endif // EIGEN_COMPLEX_CUDA_H
|
||||
@@ -0,0 +1,700 @@
|
||||
/* Copyright 2017 The TensorFlow Authors. All Rights Reserved.
|
||||
|
||||
Licensed under the Apache License, Version 2.0 (the "License");
|
||||
you may not use this file except in compliance with the License.
|
||||
You may obtain a copy of the License at
|
||||
|
||||
http://www.apache.org/licenses/LICENSE-2.0
|
||||
|
||||
Unless required by applicable law or agreed to in writing, software
|
||||
distributed under the License is distributed on an "AS IS" BASIS,
|
||||
WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
|
||||
See the License for the specific language governing permissions and
|
||||
limitations under the License.
|
||||
==============================================================================*/
|
||||
|
||||
#ifndef EIGEN_BFLOAT16_H
|
||||
#define EIGEN_BFLOAT16_H
|
||||
|
||||
#define BF16_PACKET_FUNCTION(PACKET_F, PACKET_BF16, METHOD) \
|
||||
template <> \
|
||||
EIGEN_DEFINE_FUNCTION_ALLOWING_MULTIPLE_DEFINITIONS EIGEN_UNUSED \
|
||||
PACKET_BF16 METHOD<PACKET_BF16>(const PACKET_BF16& _x) { \
|
||||
return F32ToBf16(METHOD<PACKET_F>(Bf16ToF32(_x))); \
|
||||
}
|
||||
|
||||
namespace Eigen {
|
||||
|
||||
struct bfloat16;
|
||||
|
||||
namespace bfloat16_impl {
|
||||
|
||||
// Make our own __bfloat16_raw definition.
|
||||
struct __bfloat16_raw {
|
||||
EIGEN_DEVICE_FUNC EIGEN_CONSTEXPR __bfloat16_raw() : value(0) {}
|
||||
explicit EIGEN_DEVICE_FUNC EIGEN_CONSTEXPR __bfloat16_raw(unsigned short raw) : value(raw) {}
|
||||
unsigned short value;
|
||||
};
|
||||
|
||||
EIGEN_STRONG_INLINE EIGEN_DEVICE_FUNC EIGEN_CONSTEXPR __bfloat16_raw raw_uint16_to_bfloat16(unsigned short value);
|
||||
template <bool AssumeArgumentIsNormalOrInfinityOrZero>
|
||||
EIGEN_STRONG_INLINE EIGEN_DEVICE_FUNC __bfloat16_raw float_to_bfloat16_rtne(float ff);
|
||||
// Forward declarations of template specializations, to avoid Visual C++ 2019 errors, saying:
|
||||
// > error C2908: explicit specialization; 'float_to_bfloat16_rtne' has already been instantiated
|
||||
template <>
|
||||
EIGEN_STRONG_INLINE EIGEN_DEVICE_FUNC __bfloat16_raw float_to_bfloat16_rtne<false>(float ff);
|
||||
template <>
|
||||
EIGEN_STRONG_INLINE EIGEN_DEVICE_FUNC __bfloat16_raw float_to_bfloat16_rtne<true>(float ff);
|
||||
EIGEN_STRONG_INLINE EIGEN_DEVICE_FUNC float bfloat16_to_float(__bfloat16_raw h);
|
||||
|
||||
struct bfloat16_base : public __bfloat16_raw {
|
||||
EIGEN_DEVICE_FUNC EIGEN_CONSTEXPR bfloat16_base() {}
|
||||
EIGEN_DEVICE_FUNC EIGEN_CONSTEXPR bfloat16_base(const __bfloat16_raw& h) : __bfloat16_raw(h) {}
|
||||
};
|
||||
|
||||
} // namespace bfloat16_impl
|
||||
|
||||
// Class definition.
|
||||
struct bfloat16 : public bfloat16_impl::bfloat16_base {
|
||||
|
||||
typedef bfloat16_impl::__bfloat16_raw __bfloat16_raw;
|
||||
|
||||
EIGEN_DEVICE_FUNC EIGEN_CONSTEXPR bfloat16() {}
|
||||
|
||||
EIGEN_DEVICE_FUNC EIGEN_CONSTEXPR bfloat16(const __bfloat16_raw& h) : bfloat16_impl::bfloat16_base(h) {}
|
||||
|
||||
explicit EIGEN_DEVICE_FUNC EIGEN_CONSTEXPR bfloat16(bool b)
|
||||
: bfloat16_impl::bfloat16_base(bfloat16_impl::raw_uint16_to_bfloat16(b ? 0x3f80 : 0)) {}
|
||||
|
||||
template<class T>
|
||||
explicit EIGEN_DEVICE_FUNC EIGEN_CONSTEXPR bfloat16(T val)
|
||||
: bfloat16_impl::bfloat16_base(bfloat16_impl::float_to_bfloat16_rtne<internal::is_integral<T>::value>(static_cast<float>(val))) {}
|
||||
|
||||
explicit EIGEN_DEVICE_FUNC bfloat16(float f)
|
||||
: bfloat16_impl::bfloat16_base(bfloat16_impl::float_to_bfloat16_rtne<false>(f)) {}
|
||||
|
||||
// Following the convention of numpy, converting between complex and
|
||||
// float will lead to loss of imag value.
|
||||
template<typename RealScalar>
|
||||
explicit EIGEN_DEVICE_FUNC EIGEN_CONSTEXPR bfloat16(const std::complex<RealScalar>& val)
|
||||
: bfloat16_impl::bfloat16_base(bfloat16_impl::float_to_bfloat16_rtne<false>(static_cast<float>(val.real()))) {}
|
||||
|
||||
EIGEN_DEVICE_FUNC operator float() const { // NOLINT: Allow implicit conversion to float, because it is lossless.
|
||||
return bfloat16_impl::bfloat16_to_float(*this);
|
||||
}
|
||||
};
|
||||
} // namespace Eigen
|
||||
|
||||
namespace std {
|
||||
template<>
|
||||
struct numeric_limits<Eigen::bfloat16> {
|
||||
static const bool is_specialized = true;
|
||||
static const bool is_signed = true;
|
||||
static const bool is_integer = false;
|
||||
static const bool is_exact = false;
|
||||
static const bool has_infinity = true;
|
||||
static const bool has_quiet_NaN = true;
|
||||
static const bool has_signaling_NaN = true;
|
||||
static const float_denorm_style has_denorm = std::denorm_absent;
|
||||
static const bool has_denorm_loss = false;
|
||||
static const std::float_round_style round_style = numeric_limits<float>::round_style;
|
||||
static const bool is_iec559 = false;
|
||||
static const bool is_bounded = true;
|
||||
static const bool is_modulo = false;
|
||||
static const int digits = 8;
|
||||
static const int digits10 = 2;
|
||||
static const int max_digits10 = 4;
|
||||
static const int radix = 2;
|
||||
static const int min_exponent = numeric_limits<float>::min_exponent;
|
||||
static const int min_exponent10 = numeric_limits<float>::min_exponent10;
|
||||
static const int max_exponent = numeric_limits<float>::max_exponent;
|
||||
static const int max_exponent10 = numeric_limits<float>::max_exponent10;
|
||||
static const bool traps = numeric_limits<float>::traps;
|
||||
static const bool tinyness_before = numeric_limits<float>::tinyness_before;
|
||||
|
||||
static Eigen::bfloat16 (min)() { return Eigen::bfloat16_impl::raw_uint16_to_bfloat16(0x0080); }
|
||||
static Eigen::bfloat16 lowest() { return Eigen::bfloat16_impl::raw_uint16_to_bfloat16(0xff7f); }
|
||||
static Eigen::bfloat16 (max)() { return Eigen::bfloat16_impl::raw_uint16_to_bfloat16(0x7f7f); }
|
||||
static Eigen::bfloat16 epsilon() { return Eigen::bfloat16_impl::raw_uint16_to_bfloat16(0x3c00); }
|
||||
static Eigen::bfloat16 round_error() { return Eigen::bfloat16(0x3f00); }
|
||||
static Eigen::bfloat16 infinity() { return Eigen::bfloat16_impl::raw_uint16_to_bfloat16(0x7f80); }
|
||||
static Eigen::bfloat16 quiet_NaN() { return Eigen::bfloat16_impl::raw_uint16_to_bfloat16(0x7fc0); }
|
||||
static Eigen::bfloat16 signaling_NaN() { return Eigen::bfloat16_impl::raw_uint16_to_bfloat16(0x7f81); }
|
||||
static Eigen::bfloat16 denorm_min() { return Eigen::bfloat16_impl::raw_uint16_to_bfloat16(0x0001); }
|
||||
};
|
||||
|
||||
// If std::numeric_limits<T> is specialized, should also specialize
|
||||
// std::numeric_limits<const T>, std::numeric_limits<volatile T>, and
|
||||
// std::numeric_limits<const volatile T>
|
||||
// https://stackoverflow.com/a/16519653/
|
||||
template<>
|
||||
struct numeric_limits<const Eigen::bfloat16> : numeric_limits<Eigen::bfloat16> {};
|
||||
template<>
|
||||
struct numeric_limits<volatile Eigen::bfloat16> : numeric_limits<Eigen::bfloat16> {};
|
||||
template<>
|
||||
struct numeric_limits<const volatile Eigen::bfloat16> : numeric_limits<Eigen::bfloat16> {};
|
||||
} // namespace std
|
||||
|
||||
namespace Eigen {
|
||||
|
||||
namespace bfloat16_impl {
|
||||
|
||||
// We need to distinguish ‘clang as the CUDA compiler’ from ‘clang as the host compiler,
|
||||
// invoked by NVCC’ (e.g. on MacOS). The former needs to see both host and device implementation
|
||||
// of the functions, while the latter can only deal with one of them.
|
||||
#if !defined(EIGEN_HAS_NATIVE_BF16) || (EIGEN_COMP_CLANG && !EIGEN_COMP_NVCC) // Emulate support for bfloat16 floats
|
||||
|
||||
#if EIGEN_COMP_CLANG && defined(EIGEN_CUDACC)
|
||||
// We need to provide emulated *host-side* BF16 operators for clang.
|
||||
#pragma push_macro("EIGEN_DEVICE_FUNC")
|
||||
#undef EIGEN_DEVICE_FUNC
|
||||
#if defined(EIGEN_HAS_CUDA_BF16) && defined(EIGEN_HAS_NATIVE_BF16)
|
||||
#define EIGEN_DEVICE_FUNC __host__
|
||||
#else // both host and device need emulated ops.
|
||||
#define EIGEN_DEVICE_FUNC __host__ __device__
|
||||
#endif
|
||||
#endif
|
||||
|
||||
// Definitions for CPUs, mostly working through conversion
|
||||
// to/from fp32.
|
||||
|
||||
EIGEN_STRONG_INLINE EIGEN_DEVICE_FUNC bfloat16 operator + (const bfloat16& a, const bfloat16& b) {
|
||||
return bfloat16(float(a) + float(b));
|
||||
}
|
||||
EIGEN_STRONG_INLINE EIGEN_DEVICE_FUNC bfloat16 operator + (const bfloat16& a, const int& b) {
|
||||
return bfloat16(float(a) + static_cast<float>(b));
|
||||
}
|
||||
EIGEN_STRONG_INLINE EIGEN_DEVICE_FUNC bfloat16 operator + (const int& a, const bfloat16& b) {
|
||||
return bfloat16(static_cast<float>(a) + float(b));
|
||||
}
|
||||
EIGEN_STRONG_INLINE EIGEN_DEVICE_FUNC bfloat16 operator * (const bfloat16& a, const bfloat16& b) {
|
||||
return bfloat16(float(a) * float(b));
|
||||
}
|
||||
EIGEN_STRONG_INLINE EIGEN_DEVICE_FUNC bfloat16 operator - (const bfloat16& a, const bfloat16& b) {
|
||||
return bfloat16(float(a) - float(b));
|
||||
}
|
||||
EIGEN_STRONG_INLINE EIGEN_DEVICE_FUNC bfloat16 operator / (const bfloat16& a, const bfloat16& b) {
|
||||
return bfloat16(float(a) / float(b));
|
||||
}
|
||||
EIGEN_STRONG_INLINE EIGEN_DEVICE_FUNC bfloat16 operator - (const bfloat16& a) {
|
||||
bfloat16 result;
|
||||
result.value = a.value ^ 0x8000;
|
||||
return result;
|
||||
}
|
||||
EIGEN_STRONG_INLINE EIGEN_DEVICE_FUNC bfloat16& operator += (bfloat16& a, const bfloat16& b) {
|
||||
a = bfloat16(float(a) + float(b));
|
||||
return a;
|
||||
}
|
||||
EIGEN_STRONG_INLINE EIGEN_DEVICE_FUNC bfloat16& operator *= (bfloat16& a, const bfloat16& b) {
|
||||
a = bfloat16(float(a) * float(b));
|
||||
return a;
|
||||
}
|
||||
EIGEN_STRONG_INLINE EIGEN_DEVICE_FUNC bfloat16& operator -= (bfloat16& a, const bfloat16& b) {
|
||||
a = bfloat16(float(a) - float(b));
|
||||
return a;
|
||||
}
|
||||
EIGEN_STRONG_INLINE EIGEN_DEVICE_FUNC bfloat16& operator /= (bfloat16& a, const bfloat16& b) {
|
||||
a = bfloat16(float(a) / float(b));
|
||||
return a;
|
||||
}
|
||||
EIGEN_STRONG_INLINE EIGEN_DEVICE_FUNC bfloat16 operator++(bfloat16& a) {
|
||||
a += bfloat16(1);
|
||||
return a;
|
||||
}
|
||||
EIGEN_STRONG_INLINE EIGEN_DEVICE_FUNC bfloat16 operator--(bfloat16& a) {
|
||||
a -= bfloat16(1);
|
||||
return a;
|
||||
}
|
||||
EIGEN_STRONG_INLINE EIGEN_DEVICE_FUNC bfloat16 operator++(bfloat16& a, int) {
|
||||
bfloat16 original_value = a;
|
||||
++a;
|
||||
return original_value;
|
||||
}
|
||||
EIGEN_STRONG_INLINE EIGEN_DEVICE_FUNC bfloat16 operator--(bfloat16& a, int) {
|
||||
bfloat16 original_value = a;
|
||||
--a;
|
||||
return original_value;
|
||||
}
|
||||
EIGEN_STRONG_INLINE EIGEN_DEVICE_FUNC bool operator == (const bfloat16& a, const bfloat16& b) {
|
||||
return numext::equal_strict(float(a),float(b));
|
||||
}
|
||||
EIGEN_STRONG_INLINE EIGEN_DEVICE_FUNC bool operator != (const bfloat16& a, const bfloat16& b) {
|
||||
return numext::not_equal_strict(float(a), float(b));
|
||||
}
|
||||
EIGEN_STRONG_INLINE EIGEN_DEVICE_FUNC bool operator < (const bfloat16& a, const bfloat16& b) {
|
||||
return float(a) < float(b);
|
||||
}
|
||||
EIGEN_STRONG_INLINE EIGEN_DEVICE_FUNC bool operator <= (const bfloat16& a, const bfloat16& b) {
|
||||
return float(a) <= float(b);
|
||||
}
|
||||
EIGEN_STRONG_INLINE EIGEN_DEVICE_FUNC bool operator > (const bfloat16& a, const bfloat16& b) {
|
||||
return float(a) > float(b);
|
||||
}
|
||||
EIGEN_STRONG_INLINE EIGEN_DEVICE_FUNC bool operator >= (const bfloat16& a, const bfloat16& b) {
|
||||
return float(a) >= float(b);
|
||||
}
|
||||
|
||||
#if EIGEN_COMP_CLANG && defined(EIGEN_CUDACC)
|
||||
#pragma pop_macro("EIGEN_DEVICE_FUNC")
|
||||
#endif
|
||||
#endif // Emulate support for bfloat16 floats
|
||||
|
||||
// Division by an index. Do it in full float precision to avoid accuracy
|
||||
// issues in converting the denominator to bfloat16.
|
||||
EIGEN_STRONG_INLINE EIGEN_DEVICE_FUNC bfloat16 operator / (const bfloat16& a, Index b) {
|
||||
return bfloat16(static_cast<float>(a) / static_cast<float>(b));
|
||||
}
|
||||
|
||||
EIGEN_STRONG_INLINE EIGEN_DEVICE_FUNC __bfloat16_raw truncate_to_bfloat16(const float v) {
|
||||
__bfloat16_raw output;
|
||||
if (Eigen::numext::isnan EIGEN_NOT_A_MACRO(v)) {
|
||||
output.value = std::signbit(v) ? 0xFFC0: 0x7FC0;
|
||||
return output;
|
||||
}
|
||||
const uint16_t* p = reinterpret_cast<const uint16_t*>(&v);
|
||||
#if defined(__BYTE_ORDER__) && __BYTE_ORDER__ == __ORDER_BIG_ENDIAN__
|
||||
output.value = p[0];
|
||||
#else
|
||||
output.value = p[1];
|
||||
#endif
|
||||
return output;
|
||||
}
|
||||
|
||||
EIGEN_STRONG_INLINE EIGEN_DEVICE_FUNC EIGEN_CONSTEXPR __bfloat16_raw raw_uint16_to_bfloat16(numext::uint16_t value) {
|
||||
return __bfloat16_raw(value);
|
||||
}
|
||||
|
||||
EIGEN_STRONG_INLINE EIGEN_DEVICE_FUNC EIGEN_CONSTEXPR numext::uint16_t raw_bfloat16_as_uint16(const __bfloat16_raw& bf) {
|
||||
return bf.value;
|
||||
}
|
||||
|
||||
// float_to_bfloat16_rtne template specialization that does not make any
|
||||
// assumption about the value of its function argument (ff).
|
||||
template <>
|
||||
EIGEN_STRONG_INLINE EIGEN_DEVICE_FUNC __bfloat16_raw float_to_bfloat16_rtne<false>(float ff) {
|
||||
#if (defined(EIGEN_HAS_CUDA_BF16) && defined(EIGEN_HAS_HIP_BF16))
|
||||
// Nothing to do here
|
||||
#else
|
||||
__bfloat16_raw output;
|
||||
|
||||
if (Eigen::numext::isnan EIGEN_NOT_A_MACRO(ff)) {
|
||||
// If the value is a NaN, squash it to a qNaN with msb of fraction set,
|
||||
// this makes sure after truncation we don't end up with an inf.
|
||||
//
|
||||
// qNaN magic: All exponent bits set + most significant bit of fraction
|
||||
// set.
|
||||
output.value = std::signbit(ff) ? 0xFFC0: 0x7FC0;
|
||||
} else {
|
||||
// Fast rounding algorithm that rounds a half value to nearest even. This
|
||||
// reduces expected error when we convert a large number of floats. Here
|
||||
// is how it works:
|
||||
//
|
||||
// Definitions:
|
||||
// To convert a float 32 to bfloat16, a float 32 can be viewed as 32 bits
|
||||
// with the following tags:
|
||||
//
|
||||
// Sign | Exp (8 bits) | Frac (23 bits)
|
||||
// S EEEEEEEE FFFFFFLRTTTTTTTTTTTTTTT
|
||||
//
|
||||
// S: Sign bit.
|
||||
// E: Exponent bits.
|
||||
// F: First 6 bits of fraction.
|
||||
// L: Least significant bit of resulting bfloat16 if we truncate away the
|
||||
// rest of the float32. This is also the 7th bit of fraction
|
||||
// R: Rounding bit, 8th bit of fraction.
|
||||
// T: Sticky bits, rest of fraction, 15 bits.
|
||||
//
|
||||
// To round half to nearest even, there are 3 cases where we want to round
|
||||
// down (simply truncate the result of the bits away, which consists of
|
||||
// rounding bit and sticky bits) and two cases where we want to round up
|
||||
// (truncate then add one to the result).
|
||||
//
|
||||
// The fast converting algorithm simply adds lsb (L) to 0x7fff (15 bits of
|
||||
// 1s) as the rounding bias, adds the rounding bias to the input, then
|
||||
// truncates the last 16 bits away.
|
||||
//
|
||||
// To understand how it works, we can analyze this algorithm case by case:
|
||||
//
|
||||
// 1. L = 0, R = 0:
|
||||
// Expect: round down, this is less than half value.
|
||||
//
|
||||
// Algorithm:
|
||||
// - Rounding bias: 0x7fff + 0 = 0x7fff
|
||||
// - Adding rounding bias to input may create any carry, depending on
|
||||
// whether there is any value set to 1 in T bits.
|
||||
// - R may be set to 1 if there is a carry.
|
||||
// - L remains 0.
|
||||
// - Note that this case also handles Inf and -Inf, where all fraction
|
||||
// bits, including L, R and Ts are all 0. The output remains Inf after
|
||||
// this algorithm.
|
||||
//
|
||||
// 2. L = 1, R = 0:
|
||||
// Expect: round down, this is less than half value.
|
||||
//
|
||||
// Algorithm:
|
||||
// - Rounding bias: 0x7fff + 1 = 0x8000
|
||||
// - Adding rounding bias to input doesn't change sticky bits but
|
||||
// adds 1 to rounding bit.
|
||||
// - L remains 1.
|
||||
//
|
||||
// 3. L = 0, R = 1, all of T are 0:
|
||||
// Expect: round down, this is exactly at half, the result is already
|
||||
// even (L=0).
|
||||
//
|
||||
// Algorithm:
|
||||
// - Rounding bias: 0x7fff + 0 = 0x7fff
|
||||
// - Adding rounding bias to input sets all sticky bits to 1, but
|
||||
// doesn't create a carry.
|
||||
// - R remains 1.
|
||||
// - L remains 0.
|
||||
//
|
||||
// 4. L = 1, R = 1:
|
||||
// Expect: round up, this is exactly at half, the result needs to be
|
||||
// round to the next even number.
|
||||
//
|
||||
// Algorithm:
|
||||
// - Rounding bias: 0x7fff + 1 = 0x8000
|
||||
// - Adding rounding bias to input doesn't change sticky bits, but
|
||||
// creates a carry from rounding bit.
|
||||
// - The carry sets L to 0, creates another carry bit and propagate
|
||||
// forward to F bits.
|
||||
// - If all the F bits are 1, a carry then propagates to the exponent
|
||||
// bits, which then creates the minimum value with the next exponent
|
||||
// value. Note that we won't have the case where exponents are all 1,
|
||||
// since that's either a NaN (handled in the other if condition) or inf
|
||||
// (handled in case 1).
|
||||
//
|
||||
// 5. L = 0, R = 1, any of T is 1:
|
||||
// Expect: round up, this is greater than half.
|
||||
//
|
||||
// Algorithm:
|
||||
// - Rounding bias: 0x7fff + 0 = 0x7fff
|
||||
// - Adding rounding bias to input creates a carry from sticky bits,
|
||||
// sets rounding bit to 0, then create another carry.
|
||||
// - The second carry sets L to 1.
|
||||
//
|
||||
// Examples:
|
||||
//
|
||||
// Exact half value that is already even:
|
||||
// Input:
|
||||
// Sign | Exp (8 bit) | Frac (first 7 bit) | Frac (last 16 bit)
|
||||
// S E E E E E E E E F F F F F F L RTTTTTTTTTTTTTTT
|
||||
// 0 0 0 0 0 0 0 0 0 0 0 0 0 0 1 0 1000000000000000
|
||||
//
|
||||
// This falls into case 3. We truncate the rest of 16 bits and no
|
||||
// carry is created into F and L:
|
||||
//
|
||||
// Output:
|
||||
// Sign | Exp (8 bit) | Frac (first 7 bit)
|
||||
// S E E E E E E E E F F F F F F L
|
||||
// 0 0 0 0 0 0 0 0 0 0 0 0 0 0 1 0
|
||||
//
|
||||
// Exact half value, round to next even number:
|
||||
// Input:
|
||||
// Sign | Exp (8 bit) | Frac (first 7 bit) | Frac (last 16 bit)
|
||||
// S E E E E E E E E F F F F F F L RTTTTTTTTTTTTTTT
|
||||
// 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 1 1000000000000000
|
||||
//
|
||||
// This falls into case 4. We create a carry from R and T,
|
||||
// which then propagates into L and F:
|
||||
//
|
||||
// Output:
|
||||
// Sign | Exp (8 bit) | Frac (first 7 bit)
|
||||
// S E E E E E E E E F F F F F F L
|
||||
// 0 0 0 0 0 0 0 0 0 0 0 0 0 0 1 0
|
||||
//
|
||||
//
|
||||
// Max denormal value round to min normal value:
|
||||
// Input:
|
||||
// Sign | Exp (8 bit) | Frac (first 7 bit) | Frac (last 16 bit)
|
||||
// S E E E E E E E E F F F F F F L RTTTTTTTTTTTTTTT
|
||||
// 0 0 0 0 0 0 0 0 0 1 1 1 1 1 1 1 1111111111111111
|
||||
//
|
||||
// This falls into case 4. We create a carry from R and T,
|
||||
// propagate into L and F, which then propagates into exponent
|
||||
// bits:
|
||||
//
|
||||
// Output:
|
||||
// Sign | Exp (8 bit) | Frac (first 7 bit)
|
||||
// S E E E E E E E E F F F F F F L
|
||||
// 0 0 0 0 0 0 0 0 1 0 0 0 0 0 0 0
|
||||
//
|
||||
// Max normal value round to Inf:
|
||||
// Input:
|
||||
// Sign | Exp (8 bit) | Frac (first 7 bit) | Frac (last 16 bit)
|
||||
// S E E E E E E E E F F F F F F L RTTTTTTTTTTTTTTT
|
||||
// 0 1 1 1 1 1 1 1 0 1 1 1 1 1 1 1 1111111111111111
|
||||
//
|
||||
// This falls into case 4. We create a carry from R and T,
|
||||
// propagate into L and F, which then propagates into exponent
|
||||
// bits:
|
||||
//
|
||||
// Sign | Exp (8 bit) | Frac (first 7 bit)
|
||||
// S E E E E E E E E F F F F F F L
|
||||
// 0 1 1 1 1 1 1 1 1 0 0 0 0 0 0 0
|
||||
|
||||
// At this point, ff must be either a normal float, or +/-infinity.
|
||||
output = float_to_bfloat16_rtne<true>(ff);
|
||||
}
|
||||
return output;
|
||||
#endif
|
||||
}
|
||||
|
||||
// float_to_bfloat16_rtne template specialization that assumes that its function
|
||||
// argument (ff) is either a normal floating point number, or +/-infinity, or
|
||||
// zero. Used to improve the runtime performance of conversion from an integer
|
||||
// type to bfloat16.
|
||||
template <>
|
||||
EIGEN_STRONG_INLINE EIGEN_DEVICE_FUNC __bfloat16_raw float_to_bfloat16_rtne<true>(float ff) {
|
||||
#if (defined(EIGEN_HAS_CUDA_BF16) && defined(EIGEN_HAS_HIP_BF16))
|
||||
// Nothing to do here
|
||||
#else
|
||||
numext::uint32_t input = numext::bit_cast<numext::uint32_t>(ff);
|
||||
__bfloat16_raw output;
|
||||
|
||||
// Least significant bit of resulting bfloat.
|
||||
numext::uint32_t lsb = (input >> 16) & 1;
|
||||
numext::uint32_t rounding_bias = 0x7fff + lsb;
|
||||
input += rounding_bias;
|
||||
output.value = static_cast<numext::uint16_t>(input >> 16);
|
||||
return output;
|
||||
#endif
|
||||
}
|
||||
|
||||
EIGEN_STRONG_INLINE EIGEN_DEVICE_FUNC float bfloat16_to_float(__bfloat16_raw h) {
|
||||
float result = 0;
|
||||
unsigned short* q = reinterpret_cast<unsigned short*>(&result);
|
||||
#if defined(__BYTE_ORDER__) && __BYTE_ORDER__ == __ORDER_BIG_ENDIAN__
|
||||
q[0] = h.value;
|
||||
#else
|
||||
q[1] = h.value;
|
||||
#endif
|
||||
return result;
|
||||
}
|
||||
// --- standard functions ---
|
||||
|
||||
EIGEN_STRONG_INLINE EIGEN_DEVICE_FUNC bool (isinf)(const bfloat16& a) {
|
||||
EIGEN_USING_STD(isinf);
|
||||
return (isinf)(float(a));
|
||||
}
|
||||
EIGEN_STRONG_INLINE EIGEN_DEVICE_FUNC bool (isnan)(const bfloat16& a) {
|
||||
EIGEN_USING_STD(isnan);
|
||||
return (isnan)(float(a));
|
||||
}
|
||||
EIGEN_STRONG_INLINE EIGEN_DEVICE_FUNC bool (isfinite)(const bfloat16& a) {
|
||||
return !(isinf EIGEN_NOT_A_MACRO (a)) && !(isnan EIGEN_NOT_A_MACRO (a));
|
||||
}
|
||||
|
||||
EIGEN_STRONG_INLINE EIGEN_DEVICE_FUNC bfloat16 abs(const bfloat16& a) {
|
||||
bfloat16 result;
|
||||
result.value = a.value & 0x7FFF;
|
||||
return result;
|
||||
}
|
||||
EIGEN_STRONG_INLINE EIGEN_DEVICE_FUNC bfloat16 exp(const bfloat16& a) {
|
||||
return bfloat16(::expf(float(a)));
|
||||
}
|
||||
EIGEN_STRONG_INLINE EIGEN_DEVICE_FUNC bfloat16 expm1(const bfloat16& a) {
|
||||
return bfloat16(numext::expm1(float(a)));
|
||||
}
|
||||
EIGEN_STRONG_INLINE EIGEN_DEVICE_FUNC bfloat16 log(const bfloat16& a) {
|
||||
return bfloat16(::logf(float(a)));
|
||||
}
|
||||
EIGEN_STRONG_INLINE EIGEN_DEVICE_FUNC bfloat16 log1p(const bfloat16& a) {
|
||||
return bfloat16(numext::log1p(float(a)));
|
||||
}
|
||||
EIGEN_STRONG_INLINE EIGEN_DEVICE_FUNC bfloat16 log10(const bfloat16& a) {
|
||||
return bfloat16(::log10f(float(a)));
|
||||
}
|
||||
EIGEN_STRONG_INLINE EIGEN_DEVICE_FUNC bfloat16 log2(const bfloat16& a) {
|
||||
return bfloat16(static_cast<float>(EIGEN_LOG2E) * ::logf(float(a)));
|
||||
}
|
||||
EIGEN_STRONG_INLINE EIGEN_DEVICE_FUNC bfloat16 sqrt(const bfloat16& a) {
|
||||
return bfloat16(::sqrtf(float(a)));
|
||||
}
|
||||
EIGEN_STRONG_INLINE EIGEN_DEVICE_FUNC bfloat16 pow(const bfloat16& a, const bfloat16& b) {
|
||||
return bfloat16(::powf(float(a), float(b)));
|
||||
}
|
||||
EIGEN_STRONG_INLINE EIGEN_DEVICE_FUNC bfloat16 sin(const bfloat16& a) {
|
||||
return bfloat16(::sinf(float(a)));
|
||||
}
|
||||
EIGEN_STRONG_INLINE EIGEN_DEVICE_FUNC bfloat16 cos(const bfloat16& a) {
|
||||
return bfloat16(::cosf(float(a)));
|
||||
}
|
||||
EIGEN_STRONG_INLINE EIGEN_DEVICE_FUNC bfloat16 tan(const bfloat16& a) {
|
||||
return bfloat16(::tanf(float(a)));
|
||||
}
|
||||
EIGEN_STRONG_INLINE EIGEN_DEVICE_FUNC bfloat16 asin(const bfloat16& a) {
|
||||
return bfloat16(::asinf(float(a)));
|
||||
}
|
||||
EIGEN_STRONG_INLINE EIGEN_DEVICE_FUNC bfloat16 acos(const bfloat16& a) {
|
||||
return bfloat16(::acosf(float(a)));
|
||||
}
|
||||
EIGEN_STRONG_INLINE EIGEN_DEVICE_FUNC bfloat16 atan(const bfloat16& a) {
|
||||
return bfloat16(::atanf(float(a)));
|
||||
}
|
||||
EIGEN_STRONG_INLINE EIGEN_DEVICE_FUNC bfloat16 sinh(const bfloat16& a) {
|
||||
return bfloat16(::sinhf(float(a)));
|
||||
}
|
||||
EIGEN_STRONG_INLINE EIGEN_DEVICE_FUNC bfloat16 cosh(const bfloat16& a) {
|
||||
return bfloat16(::coshf(float(a)));
|
||||
}
|
||||
EIGEN_STRONG_INLINE EIGEN_DEVICE_FUNC bfloat16 tanh(const bfloat16& a) {
|
||||
return bfloat16(::tanhf(float(a)));
|
||||
}
|
||||
#if EIGEN_HAS_CXX11_MATH
|
||||
EIGEN_STRONG_INLINE EIGEN_DEVICE_FUNC bfloat16 asinh(const bfloat16& a) {
|
||||
return bfloat16(::asinhf(float(a)));
|
||||
}
|
||||
EIGEN_STRONG_INLINE EIGEN_DEVICE_FUNC bfloat16 acosh(const bfloat16& a) {
|
||||
return bfloat16(::acoshf(float(a)));
|
||||
}
|
||||
EIGEN_STRONG_INLINE EIGEN_DEVICE_FUNC bfloat16 atanh(const bfloat16& a) {
|
||||
return bfloat16(::atanhf(float(a)));
|
||||
}
|
||||
#endif
|
||||
EIGEN_STRONG_INLINE EIGEN_DEVICE_FUNC bfloat16 floor(const bfloat16& a) {
|
||||
return bfloat16(::floorf(float(a)));
|
||||
}
|
||||
EIGEN_STRONG_INLINE EIGEN_DEVICE_FUNC bfloat16 ceil(const bfloat16& a) {
|
||||
return bfloat16(::ceilf(float(a)));
|
||||
}
|
||||
EIGEN_STRONG_INLINE EIGEN_DEVICE_FUNC bfloat16 rint(const bfloat16& a) {
|
||||
return bfloat16(::rintf(float(a)));
|
||||
}
|
||||
EIGEN_STRONG_INLINE EIGEN_DEVICE_FUNC bfloat16 round(const bfloat16& a) {
|
||||
return bfloat16(::roundf(float(a)));
|
||||
}
|
||||
EIGEN_STRONG_INLINE EIGEN_DEVICE_FUNC bfloat16 fmod(const bfloat16& a, const bfloat16& b) {
|
||||
return bfloat16(::fmodf(float(a), float(b)));
|
||||
}
|
||||
|
||||
EIGEN_STRONG_INLINE EIGEN_DEVICE_FUNC bfloat16 (min)(const bfloat16& a, const bfloat16& b) {
|
||||
const float f1 = static_cast<float>(a);
|
||||
const float f2 = static_cast<float>(b);
|
||||
return f2 < f1 ? b : a;
|
||||
}
|
||||
EIGEN_STRONG_INLINE EIGEN_DEVICE_FUNC bfloat16 (max)(const bfloat16& a, const bfloat16& b) {
|
||||
const float f1 = static_cast<float>(a);
|
||||
const float f2 = static_cast<float>(b);
|
||||
return f1 < f2 ? b : a;
|
||||
}
|
||||
|
||||
EIGEN_STRONG_INLINE EIGEN_DEVICE_FUNC bfloat16 fmin(const bfloat16& a, const bfloat16& b) {
|
||||
const float f1 = static_cast<float>(a);
|
||||
const float f2 = static_cast<float>(b);
|
||||
return bfloat16(::fminf(f1, f2));
|
||||
}
|
||||
EIGEN_STRONG_INLINE EIGEN_DEVICE_FUNC bfloat16 fmax(const bfloat16& a, const bfloat16& b) {
|
||||
const float f1 = static_cast<float>(a);
|
||||
const float f2 = static_cast<float>(b);
|
||||
return bfloat16(::fmaxf(f1, f2));
|
||||
}
|
||||
|
||||
#ifndef EIGEN_NO_IO
|
||||
EIGEN_ALWAYS_INLINE std::ostream& operator << (std::ostream& os, const bfloat16& v) {
|
||||
os << static_cast<float>(v);
|
||||
return os;
|
||||
}
|
||||
#endif
|
||||
|
||||
} // namespace bfloat16_impl
|
||||
|
||||
namespace internal {
|
||||
|
||||
template<>
|
||||
struct random_default_impl<bfloat16, false, false>
|
||||
{
|
||||
static inline bfloat16 run(const bfloat16& x, const bfloat16& y)
|
||||
{
|
||||
return x + (y-x) * bfloat16(float(std::rand()) / float(RAND_MAX));
|
||||
}
|
||||
static inline bfloat16 run()
|
||||
{
|
||||
return run(bfloat16(-1.f), bfloat16(1.f));
|
||||
}
|
||||
};
|
||||
|
||||
template<> struct is_arithmetic<bfloat16> { enum { value = true }; };
|
||||
|
||||
} // namespace internal
|
||||
|
||||
template<> struct NumTraits<Eigen::bfloat16>
|
||||
: GenericNumTraits<Eigen::bfloat16>
|
||||
{
|
||||
enum {
|
||||
IsSigned = true,
|
||||
IsInteger = false,
|
||||
IsComplex = false,
|
||||
RequireInitialization = false
|
||||
};
|
||||
|
||||
EIGEN_DEVICE_FUNC EIGEN_CONSTEXPR static EIGEN_STRONG_INLINE Eigen::bfloat16 epsilon() {
|
||||
return bfloat16_impl::raw_uint16_to_bfloat16(0x3c00);
|
||||
}
|
||||
EIGEN_DEVICE_FUNC EIGEN_CONSTEXPR static EIGEN_STRONG_INLINE Eigen::bfloat16 dummy_precision() {
|
||||
return bfloat16_impl::raw_uint16_to_bfloat16(0x3D4D); // bfloat16(5e-2f);
|
||||
|
||||
}
|
||||
EIGEN_DEVICE_FUNC EIGEN_CONSTEXPR static EIGEN_STRONG_INLINE Eigen::bfloat16 highest() {
|
||||
return bfloat16_impl::raw_uint16_to_bfloat16(0x7F7F);
|
||||
}
|
||||
EIGEN_DEVICE_FUNC EIGEN_CONSTEXPR static EIGEN_STRONG_INLINE Eigen::bfloat16 lowest() {
|
||||
return bfloat16_impl::raw_uint16_to_bfloat16(0xFF7F);
|
||||
}
|
||||
EIGEN_DEVICE_FUNC EIGEN_CONSTEXPR static EIGEN_STRONG_INLINE Eigen::bfloat16 infinity() {
|
||||
return bfloat16_impl::raw_uint16_to_bfloat16(0x7f80);
|
||||
}
|
||||
EIGEN_DEVICE_FUNC EIGEN_CONSTEXPR static EIGEN_STRONG_INLINE Eigen::bfloat16 quiet_NaN() {
|
||||
return bfloat16_impl::raw_uint16_to_bfloat16(0x7fc0);
|
||||
}
|
||||
};
|
||||
|
||||
} // namespace Eigen
|
||||
|
||||
namespace Eigen {
|
||||
namespace numext {
|
||||
|
||||
template<>
|
||||
EIGEN_DEVICE_FUNC EIGEN_ALWAYS_INLINE
|
||||
bool (isnan)(const Eigen::bfloat16& h) {
|
||||
return (bfloat16_impl::isnan)(h);
|
||||
}
|
||||
|
||||
template<>
|
||||
EIGEN_DEVICE_FUNC EIGEN_ALWAYS_INLINE
|
||||
bool (isinf)(const Eigen::bfloat16& h) {
|
||||
return (bfloat16_impl::isinf)(h);
|
||||
}
|
||||
|
||||
template<>
|
||||
EIGEN_DEVICE_FUNC EIGEN_ALWAYS_INLINE
|
||||
bool (isfinite)(const Eigen::bfloat16& h) {
|
||||
return (bfloat16_impl::isfinite)(h);
|
||||
}
|
||||
|
||||
template <>
|
||||
EIGEN_STRONG_INLINE EIGEN_DEVICE_FUNC Eigen::bfloat16 bit_cast<Eigen::bfloat16, uint16_t>(const uint16_t& src) {
|
||||
return Eigen::bfloat16(Eigen::bfloat16_impl::raw_uint16_to_bfloat16(src));
|
||||
}
|
||||
|
||||
template <>
|
||||
EIGEN_STRONG_INLINE EIGEN_DEVICE_FUNC uint16_t bit_cast<uint16_t, Eigen::bfloat16>(const Eigen::bfloat16& src) {
|
||||
return Eigen::bfloat16_impl::raw_bfloat16_as_uint16(src);
|
||||
}
|
||||
|
||||
} // namespace numext
|
||||
} // namespace Eigen
|
||||
|
||||
#if EIGEN_HAS_STD_HASH
|
||||
namespace std {
|
||||
template <>
|
||||
struct hash<Eigen::bfloat16> {
|
||||
EIGEN_STRONG_INLINE std::size_t operator()(const Eigen::bfloat16& a) const {
|
||||
return static_cast<std::size_t>(Eigen::numext::bit_cast<Eigen::numext::uint16_t>(a));
|
||||
}
|
||||
};
|
||||
} // namespace std
|
||||
#endif
|
||||
|
||||
|
||||
#endif // EIGEN_BFLOAT16_H
|
||||
@@ -0,0 +1,117 @@
|
||||
|
||||
// This file is part of Eigen, a lightweight C++ template library
|
||||
// for linear algebra.
|
||||
//
|
||||
// Copyright (C) 2017 Gael Guennebaud <gael.guennebaud@inria.fr>
|
||||
//
|
||||
// This Source Code Form is subject to the terms of the Mozilla
|
||||
// Public License v. 2.0. If a copy of the MPL was not distributed
|
||||
// with this file, You can obtain one at http://mozilla.org/MPL/2.0/.
|
||||
|
||||
#ifndef EIGEN_ARCH_CONJ_HELPER_H
|
||||
#define EIGEN_ARCH_CONJ_HELPER_H
|
||||
|
||||
#define EIGEN_MAKE_CONJ_HELPER_CPLX_REAL(PACKET_CPLX, PACKET_REAL) \
|
||||
template <> \
|
||||
struct conj_helper<PACKET_REAL, PACKET_CPLX, false, false> { \
|
||||
EIGEN_STRONG_INLINE PACKET_CPLX pmadd(const PACKET_REAL& x, \
|
||||
const PACKET_CPLX& y, \
|
||||
const PACKET_CPLX& c) const { \
|
||||
return padd(c, this->pmul(x, y)); \
|
||||
} \
|
||||
EIGEN_STRONG_INLINE PACKET_CPLX pmul(const PACKET_REAL& x, \
|
||||
const PACKET_CPLX& y) const { \
|
||||
return PACKET_CPLX(Eigen::internal::pmul<PACKET_REAL>(x, y.v)); \
|
||||
} \
|
||||
}; \
|
||||
\
|
||||
template <> \
|
||||
struct conj_helper<PACKET_CPLX, PACKET_REAL, false, false> { \
|
||||
EIGEN_STRONG_INLINE PACKET_CPLX pmadd(const PACKET_CPLX& x, \
|
||||
const PACKET_REAL& y, \
|
||||
const PACKET_CPLX& c) const { \
|
||||
return padd(c, this->pmul(x, y)); \
|
||||
} \
|
||||
EIGEN_STRONG_INLINE PACKET_CPLX pmul(const PACKET_CPLX& x, \
|
||||
const PACKET_REAL& y) const { \
|
||||
return PACKET_CPLX(Eigen::internal::pmul<PACKET_REAL>(x.v, y)); \
|
||||
} \
|
||||
};
|
||||
|
||||
namespace Eigen {
|
||||
namespace internal {
|
||||
|
||||
template<bool Conjugate> struct conj_if;
|
||||
|
||||
template<> struct conj_if<true> {
|
||||
template<typename T>
|
||||
EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE T operator()(const T& x) const { return numext::conj(x); }
|
||||
template<typename T>
|
||||
EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE T pconj(const T& x) const { return internal::pconj(x); }
|
||||
};
|
||||
|
||||
template<> struct conj_if<false> {
|
||||
template<typename T>
|
||||
EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE const T& operator()(const T& x) const { return x; }
|
||||
template<typename T>
|
||||
EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE const T& pconj(const T& x) const { return x; }
|
||||
};
|
||||
|
||||
// Generic Implementation, assume scalars since the packet-version is
|
||||
// specialized below.
|
||||
template<typename LhsType, typename RhsType, bool ConjLhs, bool ConjRhs>
|
||||
struct conj_helper {
|
||||
typedef typename ScalarBinaryOpTraits<LhsType, RhsType>::ReturnType ResultType;
|
||||
|
||||
EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE ResultType
|
||||
pmadd(const LhsType& x, const RhsType& y, const ResultType& c) const
|
||||
{ return this->pmul(x, y) + c; }
|
||||
|
||||
EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE ResultType
|
||||
pmul(const LhsType& x, const RhsType& y) const
|
||||
{ return conj_if<ConjLhs>()(x) * conj_if<ConjRhs>()(y); }
|
||||
};
|
||||
|
||||
template<typename LhsScalar, typename RhsScalar>
|
||||
struct conj_helper<LhsScalar, RhsScalar, true, true> {
|
||||
typedef typename ScalarBinaryOpTraits<LhsScalar,RhsScalar>::ReturnType ResultType;
|
||||
|
||||
EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE ResultType
|
||||
pmadd(const LhsScalar& x, const RhsScalar& y, const ResultType& c) const
|
||||
{ return this->pmul(x, y) + c; }
|
||||
|
||||
// We save a conjuation by using the identity conj(a)*conj(b) = conj(a*b).
|
||||
EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE ResultType
|
||||
pmul(const LhsScalar& x, const RhsScalar& y) const
|
||||
{ return numext::conj(x * y); }
|
||||
};
|
||||
|
||||
// Implementation with equal type, use packet operations.
|
||||
template<typename Packet, bool ConjLhs, bool ConjRhs>
|
||||
struct conj_helper<Packet, Packet, ConjLhs, ConjRhs>
|
||||
{
|
||||
typedef Packet ResultType;
|
||||
EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE Packet pmadd(const Packet& x, const Packet& y, const Packet& c) const
|
||||
{ return Eigen::internal::pmadd(conj_if<ConjLhs>().pconj(x), conj_if<ConjRhs>().pconj(y), c); }
|
||||
|
||||
|
||||
EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE Packet pmul(const Packet& x, const Packet& y) const
|
||||
{ return Eigen::internal::pmul(conj_if<ConjLhs>().pconj(x), conj_if<ConjRhs>().pconj(y)); }
|
||||
};
|
||||
|
||||
template<typename Packet>
|
||||
struct conj_helper<Packet, Packet, true, true>
|
||||
{
|
||||
typedef Packet ResultType;
|
||||
|
||||
EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE Packet pmadd(const Packet& x, const Packet& y, const Packet& c) const
|
||||
{ return Eigen::internal::pmadd(pconj(x), pconj(y), c); }
|
||||
// We save a conjuation by using the identity conj(a)*conj(b) = conj(a*b).
|
||||
EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE Packet pmul(const Packet& x, const Packet& y) const
|
||||
{ return pconj(Eigen::internal::pmul(x, y)); }
|
||||
};
|
||||
|
||||
} // namespace internal
|
||||
} // namespace Eigen
|
||||
|
||||
#endif // EIGEN_ARCH_CONJ_HELPER_H
|
||||
File diff suppressed because it is too large
Load Diff
@@ -0,0 +1,110 @@
|
||||
// This file is part of Eigen, a lightweight C++ template library
|
||||
// for linear algebra.
|
||||
//
|
||||
// Copyright (C) 2019 Gael Guennebaud <gael.guennebaud@inria.fr>
|
||||
//
|
||||
// This Source Code Form is subject to the terms of the Mozilla
|
||||
// Public License v. 2.0. If a copy of the MPL was not distributed
|
||||
// with this file, You can obtain one at http://mozilla.org/MPL/2.0/.
|
||||
|
||||
#ifndef EIGEN_ARCH_GENERIC_PACKET_MATH_FUNCTIONS_FWD_H
|
||||
#define EIGEN_ARCH_GENERIC_PACKET_MATH_FUNCTIONS_FWD_H
|
||||
|
||||
namespace Eigen {
|
||||
namespace internal {
|
||||
|
||||
// Forward declarations of the generic math functions
|
||||
// implemented in GenericPacketMathFunctions.h
|
||||
// This is needed to workaround a circular dependency.
|
||||
|
||||
/***************************************************************************
|
||||
* Some generic implementations to be used by implementors
|
||||
***************************************************************************/
|
||||
|
||||
/** Default implementation of pfrexp.
|
||||
* It is expected to be called by implementers of template<> pfrexp.
|
||||
*/
|
||||
template<typename Packet> EIGEN_STRONG_INLINE EIGEN_DEVICE_FUNC
|
||||
Packet pfrexp_generic(const Packet& a, Packet& exponent);
|
||||
|
||||
// Extracts the biased exponent value from Packet p, and casts the results to
|
||||
// a floating-point Packet type. Used by pfrexp_generic. Override this if
|
||||
// there is no unpacket_traits<Packet>::integer_packet.
|
||||
template<typename Packet> EIGEN_STRONG_INLINE EIGEN_DEVICE_FUNC
|
||||
Packet pfrexp_generic_get_biased_exponent(const Packet& p);
|
||||
|
||||
/** Default implementation of pldexp.
|
||||
* It is expected to be called by implementers of template<> pldexp.
|
||||
*/
|
||||
template<typename Packet> EIGEN_STRONG_INLINE EIGEN_DEVICE_FUNC
|
||||
Packet pldexp_generic(const Packet& a, const Packet& exponent);
|
||||
|
||||
/** \internal \returns log(x) for single precision float */
|
||||
template <typename Packet>
|
||||
EIGEN_DEFINE_FUNCTION_ALLOWING_MULTIPLE_DEFINITIONS
|
||||
EIGEN_UNUSED
|
||||
Packet plog_float(const Packet _x);
|
||||
|
||||
/** \internal \returns log2(x) for single precision float */
|
||||
template <typename Packet>
|
||||
EIGEN_DEFINE_FUNCTION_ALLOWING_MULTIPLE_DEFINITIONS
|
||||
EIGEN_UNUSED
|
||||
Packet plog2_float(const Packet _x);
|
||||
|
||||
/** \internal \returns log(x) for single precision float */
|
||||
template <typename Packet>
|
||||
EIGEN_DEFINE_FUNCTION_ALLOWING_MULTIPLE_DEFINITIONS
|
||||
EIGEN_UNUSED
|
||||
Packet plog_double(const Packet _x);
|
||||
|
||||
/** \internal \returns log2(x) for single precision float */
|
||||
template <typename Packet>
|
||||
EIGEN_DEFINE_FUNCTION_ALLOWING_MULTIPLE_DEFINITIONS
|
||||
EIGEN_UNUSED
|
||||
Packet plog2_double(const Packet _x);
|
||||
|
||||
/** \internal \returns log(1 + x) */
|
||||
template<typename Packet>
|
||||
Packet generic_plog1p(const Packet& x);
|
||||
|
||||
/** \internal \returns exp(x)-1 */
|
||||
template<typename Packet>
|
||||
Packet generic_expm1(const Packet& x);
|
||||
|
||||
/** \internal \returns exp(x) for single precision float */
|
||||
template <typename Packet>
|
||||
EIGEN_DEFINE_FUNCTION_ALLOWING_MULTIPLE_DEFINITIONS
|
||||
EIGEN_UNUSED
|
||||
Packet pexp_float(const Packet _x);
|
||||
|
||||
/** \internal \returns exp(x) for double precision real numbers */
|
||||
template <typename Packet>
|
||||
EIGEN_DEFINE_FUNCTION_ALLOWING_MULTIPLE_DEFINITIONS
|
||||
EIGEN_UNUSED
|
||||
Packet pexp_double(const Packet _x);
|
||||
|
||||
/** \internal \returns sin(x) for single precision float */
|
||||
template<typename Packet>
|
||||
EIGEN_DEFINE_FUNCTION_ALLOWING_MULTIPLE_DEFINITIONS
|
||||
EIGEN_UNUSED
|
||||
Packet psin_float(const Packet& x);
|
||||
|
||||
/** \internal \returns cos(x) for single precision float */
|
||||
template<typename Packet>
|
||||
EIGEN_DEFINE_FUNCTION_ALLOWING_MULTIPLE_DEFINITIONS
|
||||
EIGEN_UNUSED
|
||||
Packet pcos_float(const Packet& x);
|
||||
|
||||
/** \internal \returns sqrt(x) for complex types */
|
||||
template<typename Packet>
|
||||
EIGEN_DEFINE_FUNCTION_ALLOWING_MULTIPLE_DEFINITIONS
|
||||
EIGEN_UNUSED
|
||||
Packet psqrt_complex(const Packet& a);
|
||||
|
||||
template <typename Packet, int N> struct ppolevl;
|
||||
|
||||
|
||||
} // end namespace internal
|
||||
} // end namespace Eigen
|
||||
|
||||
#endif // EIGEN_ARCH_GENERIC_PACKET_MATH_FUNCTIONS_FWD_H
|
||||
@@ -0,0 +1,942 @@
|
||||
// This file is part of Eigen, a lightweight C++ template library
|
||||
// for linear algebra.
|
||||
//
|
||||
// This Source Code Form is subject to the terms of the Mozilla
|
||||
// Public License v. 2.0. If a copy of the MPL was not distributed
|
||||
// with this file, You can obtain one at http://mozilla.org/MPL/2.0/.
|
||||
//
|
||||
// The conversion routines are Copyright (c) Fabian Giesen, 2016.
|
||||
// The original license follows:
|
||||
//
|
||||
// Copyright (c) Fabian Giesen, 2016
|
||||
// All rights reserved.
|
||||
// Redistribution and use in source and binary forms, with or without
|
||||
// modification, are permitted.
|
||||
// THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
|
||||
// "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
|
||||
// LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
|
||||
// A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
|
||||
// HOLDER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
|
||||
// SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
|
||||
// LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
|
||||
// DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
|
||||
// THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
|
||||
// (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
|
||||
// OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
|
||||
|
||||
|
||||
// Standard 16-bit float type, mostly useful for GPUs. Defines a new
|
||||
// type Eigen::half (inheriting either from CUDA's or HIP's __half struct) with
|
||||
// operator overloads such that it behaves basically as an arithmetic
|
||||
// type. It will be quite slow on CPUs (so it is recommended to stay
|
||||
// in fp32 for CPUs, except for simple parameter conversions, I/O
|
||||
// to disk and the likes), but fast on GPUs.
|
||||
|
||||
|
||||
#ifndef EIGEN_HALF_H
|
||||
#define EIGEN_HALF_H
|
||||
|
||||
#include <sstream>
|
||||
|
||||
#if defined(EIGEN_HAS_GPU_FP16) || defined(EIGEN_HAS_ARM64_FP16_SCALAR_ARITHMETIC)
|
||||
// When compiling with GPU support, the "__half_raw" base class as well as
|
||||
// some other routines are defined in the GPU compiler header files
|
||||
// (cuda_fp16.h, hip_fp16.h), and they are not tagged constexpr
|
||||
// As a consequence, we get compile failures when compiling Eigen with
|
||||
// GPU support. Hence the need to disable EIGEN_CONSTEXPR when building
|
||||
// Eigen with GPU support
|
||||
#pragma push_macro("EIGEN_CONSTEXPR")
|
||||
#undef EIGEN_CONSTEXPR
|
||||
#define EIGEN_CONSTEXPR
|
||||
#endif
|
||||
|
||||
#define F16_PACKET_FUNCTION(PACKET_F, PACKET_F16, METHOD) \
|
||||
template <> \
|
||||
EIGEN_STRONG_INLINE EIGEN_DEVICE_FUNC EIGEN_UNUSED \
|
||||
PACKET_F16 METHOD<PACKET_F16>(const PACKET_F16& _x) { \
|
||||
return float2half(METHOD<PACKET_F>(half2float(_x))); \
|
||||
}
|
||||
|
||||
namespace Eigen {
|
||||
|
||||
struct half;
|
||||
|
||||
namespace half_impl {
|
||||
|
||||
// We want to use the __half_raw struct from the HIP header file only during the device compile phase.
|
||||
// This is required because of a quirk in the way TensorFlow GPU builds are done.
|
||||
// When compiling TensorFlow source code with GPU support, files that
|
||||
// * contain GPU kernels (i.e. *.cu.cc files) are compiled via hipcc
|
||||
// * do not contain GPU kernels ( i.e. *.cc files) are compiled via gcc (typically)
|
||||
//
|
||||
// Tensorflow uses the Eigen::half type as its FP16 type, and there are functions that
|
||||
// * are defined in a file that gets compiled via hipcc AND
|
||||
// * have Eigen::half as a pass-by-value argument AND
|
||||
// * are called in a file that gets compiled via gcc
|
||||
//
|
||||
// In the scenario described above the caller and callee will see different versions
|
||||
// of the Eigen::half base class __half_raw, and they will be compiled by different compilers
|
||||
//
|
||||
// There appears to be an ABI mismatch between gcc and clang (which is called by hipcc) that results in
|
||||
// the callee getting corrupted values for the Eigen::half argument.
|
||||
//
|
||||
// Making the host side compile phase of hipcc use the same Eigen::half impl, as the gcc compile, resolves
|
||||
// this error, and hence the following convoluted #if condition
|
||||
#if !defined(EIGEN_HAS_GPU_FP16) || !defined(EIGEN_GPU_COMPILE_PHASE)
|
||||
// Make our own __half_raw definition that is similar to CUDA's.
|
||||
struct __half_raw {
|
||||
#if (defined(EIGEN_HAS_GPU_FP16) && !defined(EIGEN_GPU_COMPILE_PHASE))
|
||||
// Eigen::half can be used as the datatype for shared memory declarations (in Eigen and TF)
|
||||
// The element type for shared memory cannot have non-trivial constructors
|
||||
// and hence the following special casing (which skips the zero-initilization).
|
||||
// Note that this check gets done even in the host compilation phase, and
|
||||
// hence the need for this
|
||||
EIGEN_DEVICE_FUNC __half_raw() {}
|
||||
#else
|
||||
EIGEN_DEVICE_FUNC EIGEN_CONSTEXPR __half_raw() : x(0) {}
|
||||
#endif
|
||||
#if defined(EIGEN_HAS_ARM64_FP16_SCALAR_ARITHMETIC)
|
||||
explicit EIGEN_DEVICE_FUNC EIGEN_CONSTEXPR __half_raw(numext::uint16_t raw) : x(numext::bit_cast<__fp16>(raw)) {
|
||||
}
|
||||
__fp16 x;
|
||||
#else
|
||||
explicit EIGEN_DEVICE_FUNC EIGEN_CONSTEXPR __half_raw(numext::uint16_t raw) : x(raw) {}
|
||||
numext::uint16_t x;
|
||||
#endif
|
||||
};
|
||||
|
||||
#elif defined(EIGEN_HAS_HIP_FP16)
|
||||
// Nothing to do here
|
||||
// HIP fp16 header file has a definition for __half_raw
|
||||
#elif defined(EIGEN_HAS_CUDA_FP16)
|
||||
#if EIGEN_CUDA_SDK_VER < 90000
|
||||
// In CUDA < 9.0, __half is the equivalent of CUDA 9's __half_raw
|
||||
typedef __half __half_raw;
|
||||
#endif // defined(EIGEN_HAS_CUDA_FP16)
|
||||
#elif defined(SYCL_DEVICE_ONLY)
|
||||
typedef cl::sycl::half __half_raw;
|
||||
#endif
|
||||
|
||||
EIGEN_STRONG_INLINE EIGEN_DEVICE_FUNC EIGEN_CONSTEXPR __half_raw raw_uint16_to_half(numext::uint16_t x);
|
||||
EIGEN_STRONG_INLINE EIGEN_DEVICE_FUNC __half_raw float_to_half_rtne(float ff);
|
||||
EIGEN_STRONG_INLINE EIGEN_DEVICE_FUNC float half_to_float(__half_raw h);
|
||||
|
||||
struct half_base : public __half_raw {
|
||||
EIGEN_DEVICE_FUNC EIGEN_CONSTEXPR half_base() {}
|
||||
EIGEN_DEVICE_FUNC EIGEN_CONSTEXPR half_base(const __half_raw& h) : __half_raw(h) {}
|
||||
|
||||
#if defined(EIGEN_HAS_GPU_FP16)
|
||||
#if defined(EIGEN_HAS_HIP_FP16)
|
||||
EIGEN_DEVICE_FUNC EIGEN_CONSTEXPR half_base(const __half& h) { x = __half_as_ushort(h); }
|
||||
#elif defined(EIGEN_HAS_CUDA_FP16)
|
||||
#if EIGEN_CUDA_SDK_VER >= 90000
|
||||
EIGEN_DEVICE_FUNC EIGEN_CONSTEXPR half_base(const __half& h) : __half_raw(*(__half_raw*)&h) {}
|
||||
#endif
|
||||
#endif
|
||||
#endif
|
||||
};
|
||||
|
||||
} // namespace half_impl
|
||||
|
||||
// Class definition.
|
||||
struct half : public half_impl::half_base {
|
||||
|
||||
// Writing this out as separate #if-else blocks to make the code easier to follow
|
||||
// The same applies to most #if-else blocks in this file
|
||||
#if !defined(EIGEN_HAS_GPU_FP16) || !defined(EIGEN_GPU_COMPILE_PHASE)
|
||||
// Use the same base class for the following two scenarios
|
||||
// * when compiling without GPU support enabled
|
||||
// * during host compile phase when compiling with GPU support enabled
|
||||
typedef half_impl::__half_raw __half_raw;
|
||||
#elif defined(EIGEN_HAS_HIP_FP16)
|
||||
// Nothing to do here
|
||||
// HIP fp16 header file has a definition for __half_raw
|
||||
#elif defined(EIGEN_HAS_CUDA_FP16)
|
||||
// Note that EIGEN_CUDA_SDK_VER is set to 0 even when compiling with HIP, so
|
||||
// (EIGEN_CUDA_SDK_VER < 90000) is true even for HIP! So keeping this within
|
||||
// #if defined(EIGEN_HAS_CUDA_FP16) is needed
|
||||
#if defined(EIGEN_CUDA_SDK_VER) && EIGEN_CUDA_SDK_VER < 90000
|
||||
typedef half_impl::__half_raw __half_raw;
|
||||
#endif
|
||||
#endif
|
||||
|
||||
EIGEN_DEVICE_FUNC EIGEN_CONSTEXPR half() {}
|
||||
|
||||
EIGEN_DEVICE_FUNC EIGEN_CONSTEXPR half(const __half_raw& h) : half_impl::half_base(h) {}
|
||||
|
||||
#if defined(EIGEN_HAS_GPU_FP16)
|
||||
#if defined(EIGEN_HAS_HIP_FP16)
|
||||
EIGEN_DEVICE_FUNC EIGEN_CONSTEXPR half(const __half& h) : half_impl::half_base(h) {}
|
||||
#elif defined(EIGEN_HAS_CUDA_FP16)
|
||||
#if defined(EIGEN_CUDA_SDK_VER) && EIGEN_CUDA_SDK_VER >= 90000
|
||||
EIGEN_DEVICE_FUNC EIGEN_CONSTEXPR half(const __half& h) : half_impl::half_base(h) {}
|
||||
#endif
|
||||
#endif
|
||||
#endif
|
||||
|
||||
|
||||
explicit EIGEN_DEVICE_FUNC EIGEN_CONSTEXPR half(bool b)
|
||||
: half_impl::half_base(half_impl::raw_uint16_to_half(b ? 0x3c00 : 0)) {}
|
||||
template<class T>
|
||||
explicit EIGEN_DEVICE_FUNC half(T val)
|
||||
: half_impl::half_base(half_impl::float_to_half_rtne(static_cast<float>(val))) {}
|
||||
explicit EIGEN_DEVICE_FUNC half(float f)
|
||||
: half_impl::half_base(half_impl::float_to_half_rtne(f)) {}
|
||||
|
||||
// Following the convention of numpy, converting between complex and
|
||||
// float will lead to loss of imag value.
|
||||
template<typename RealScalar>
|
||||
explicit EIGEN_DEVICE_FUNC half(std::complex<RealScalar> c)
|
||||
: half_impl::half_base(half_impl::float_to_half_rtne(static_cast<float>(c.real()))) {}
|
||||
|
||||
EIGEN_DEVICE_FUNC operator float() const { // NOLINT: Allow implicit conversion to float, because it is lossless.
|
||||
return half_impl::half_to_float(*this);
|
||||
}
|
||||
|
||||
#if defined(EIGEN_HAS_GPU_FP16) && !defined(EIGEN_GPU_COMPILE_PHASE)
|
||||
EIGEN_DEVICE_FUNC operator __half() const {
|
||||
::__half_raw hr;
|
||||
hr.x = x;
|
||||
return __half(hr);
|
||||
}
|
||||
#endif
|
||||
};
|
||||
|
||||
} // end namespace Eigen
|
||||
|
||||
namespace std {
|
||||
template<>
|
||||
struct numeric_limits<Eigen::half> {
|
||||
static const bool is_specialized = true;
|
||||
static const bool is_signed = true;
|
||||
static const bool is_integer = false;
|
||||
static const bool is_exact = false;
|
||||
static const bool has_infinity = true;
|
||||
static const bool has_quiet_NaN = true;
|
||||
static const bool has_signaling_NaN = true;
|
||||
static const float_denorm_style has_denorm = denorm_present;
|
||||
static const bool has_denorm_loss = false;
|
||||
static const std::float_round_style round_style = std::round_to_nearest;
|
||||
static const bool is_iec559 = false;
|
||||
static const bool is_bounded = false;
|
||||
static const bool is_modulo = false;
|
||||
static const int digits = 11;
|
||||
static const int digits10 = 3; // according to http://half.sourceforge.net/structstd_1_1numeric__limits_3_01half__float_1_1half_01_4.html
|
||||
static const int max_digits10 = 5; // according to http://half.sourceforge.net/structstd_1_1numeric__limits_3_01half__float_1_1half_01_4.html
|
||||
static const int radix = 2;
|
||||
static const int min_exponent = -13;
|
||||
static const int min_exponent10 = -4;
|
||||
static const int max_exponent = 16;
|
||||
static const int max_exponent10 = 4;
|
||||
static const bool traps = true;
|
||||
static const bool tinyness_before = false;
|
||||
|
||||
static Eigen::half (min)() { return Eigen::half_impl::raw_uint16_to_half(0x400); }
|
||||
static Eigen::half lowest() { return Eigen::half_impl::raw_uint16_to_half(0xfbff); }
|
||||
static Eigen::half (max)() { return Eigen::half_impl::raw_uint16_to_half(0x7bff); }
|
||||
static Eigen::half epsilon() { return Eigen::half_impl::raw_uint16_to_half(0x0800); }
|
||||
static Eigen::half round_error() { return Eigen::half(0.5); }
|
||||
static Eigen::half infinity() { return Eigen::half_impl::raw_uint16_to_half(0x7c00); }
|
||||
static Eigen::half quiet_NaN() { return Eigen::half_impl::raw_uint16_to_half(0x7e00); }
|
||||
static Eigen::half signaling_NaN() { return Eigen::half_impl::raw_uint16_to_half(0x7d00); }
|
||||
static Eigen::half denorm_min() { return Eigen::half_impl::raw_uint16_to_half(0x1); }
|
||||
};
|
||||
|
||||
// If std::numeric_limits<T> is specialized, should also specialize
|
||||
// std::numeric_limits<const T>, std::numeric_limits<volatile T>, and
|
||||
// std::numeric_limits<const volatile T>
|
||||
// https://stackoverflow.com/a/16519653/
|
||||
template<>
|
||||
struct numeric_limits<const Eigen::half> : numeric_limits<Eigen::half> {};
|
||||
template<>
|
||||
struct numeric_limits<volatile Eigen::half> : numeric_limits<Eigen::half> {};
|
||||
template<>
|
||||
struct numeric_limits<const volatile Eigen::half> : numeric_limits<Eigen::half> {};
|
||||
} // end namespace std
|
||||
|
||||
namespace Eigen {
|
||||
|
||||
namespace half_impl {
|
||||
|
||||
#if (defined(EIGEN_HAS_CUDA_FP16) && defined(EIGEN_CUDA_ARCH) && \
|
||||
EIGEN_CUDA_ARCH >= 530) || \
|
||||
(defined(EIGEN_HAS_HIP_FP16) && defined(HIP_DEVICE_COMPILE))
|
||||
// Note: We deliberatly do *not* define this to 1 even if we have Arm's native
|
||||
// fp16 type since GPU halfs are rather different from native CPU halfs.
|
||||
// TODO: Rename to something like EIGEN_HAS_NATIVE_GPU_FP16
|
||||
#define EIGEN_HAS_NATIVE_FP16
|
||||
#endif
|
||||
|
||||
// Intrinsics for native fp16 support. Note that on current hardware,
|
||||
// these are no faster than fp32 arithmetic (you need to use the half2
|
||||
// versions to get the ALU speed increased), but you do save the
|
||||
// conversion steps back and forth.
|
||||
|
||||
#if defined(EIGEN_HAS_NATIVE_FP16)
|
||||
EIGEN_STRONG_INLINE __device__ half operator + (const half& a, const half& b) {
|
||||
#if defined(EIGEN_CUDA_SDK_VER) && EIGEN_CUDA_SDK_VER >= 90000
|
||||
return __hadd(::__half(a), ::__half(b));
|
||||
#else
|
||||
return __hadd(a, b);
|
||||
#endif
|
||||
}
|
||||
EIGEN_STRONG_INLINE __device__ half operator * (const half& a, const half& b) {
|
||||
return __hmul(a, b);
|
||||
}
|
||||
EIGEN_STRONG_INLINE __device__ half operator - (const half& a, const half& b) {
|
||||
return __hsub(a, b);
|
||||
}
|
||||
EIGEN_STRONG_INLINE __device__ half operator / (const half& a, const half& b) {
|
||||
#if defined(EIGEN_CUDA_SDK_VER) && EIGEN_CUDA_SDK_VER >= 90000
|
||||
return __hdiv(a, b);
|
||||
#else
|
||||
float num = __half2float(a);
|
||||
float denom = __half2float(b);
|
||||
return __float2half(num / denom);
|
||||
#endif
|
||||
}
|
||||
EIGEN_STRONG_INLINE __device__ half operator - (const half& a) {
|
||||
return __hneg(a);
|
||||
}
|
||||
EIGEN_STRONG_INLINE __device__ half& operator += (half& a, const half& b) {
|
||||
a = a + b;
|
||||
return a;
|
||||
}
|
||||
EIGEN_STRONG_INLINE __device__ half& operator *= (half& a, const half& b) {
|
||||
a = a * b;
|
||||
return a;
|
||||
}
|
||||
EIGEN_STRONG_INLINE __device__ half& operator -= (half& a, const half& b) {
|
||||
a = a - b;
|
||||
return a;
|
||||
}
|
||||
EIGEN_STRONG_INLINE __device__ half& operator /= (half& a, const half& b) {
|
||||
a = a / b;
|
||||
return a;
|
||||
}
|
||||
EIGEN_STRONG_INLINE __device__ bool operator == (const half& a, const half& b) {
|
||||
return __heq(a, b);
|
||||
}
|
||||
EIGEN_STRONG_INLINE __device__ bool operator != (const half& a, const half& b) {
|
||||
return __hne(a, b);
|
||||
}
|
||||
EIGEN_STRONG_INLINE __device__ bool operator < (const half& a, const half& b) {
|
||||
return __hlt(a, b);
|
||||
}
|
||||
EIGEN_STRONG_INLINE __device__ bool operator <= (const half& a, const half& b) {
|
||||
return __hle(a, b);
|
||||
}
|
||||
EIGEN_STRONG_INLINE __device__ bool operator > (const half& a, const half& b) {
|
||||
return __hgt(a, b);
|
||||
}
|
||||
EIGEN_STRONG_INLINE __device__ bool operator >= (const half& a, const half& b) {
|
||||
return __hge(a, b);
|
||||
}
|
||||
#endif
|
||||
|
||||
#if defined(EIGEN_HAS_ARM64_FP16_SCALAR_ARITHMETIC)
|
||||
EIGEN_STRONG_INLINE EIGEN_DEVICE_FUNC half operator + (const half& a, const half& b) {
|
||||
return half(vaddh_f16(a.x, b.x));
|
||||
}
|
||||
EIGEN_STRONG_INLINE EIGEN_DEVICE_FUNC half operator * (const half& a, const half& b) {
|
||||
return half(vmulh_f16(a.x, b.x));
|
||||
}
|
||||
EIGEN_STRONG_INLINE EIGEN_DEVICE_FUNC half operator - (const half& a, const half& b) {
|
||||
return half(vsubh_f16(a.x, b.x));
|
||||
}
|
||||
EIGEN_STRONG_INLINE EIGEN_DEVICE_FUNC half operator / (const half& a, const half& b) {
|
||||
return half(vdivh_f16(a.x, b.x));
|
||||
}
|
||||
EIGEN_STRONG_INLINE EIGEN_DEVICE_FUNC half operator - (const half& a) {
|
||||
return half(vnegh_f16(a.x));
|
||||
}
|
||||
EIGEN_STRONG_INLINE EIGEN_DEVICE_FUNC half& operator += (half& a, const half& b) {
|
||||
a = half(vaddh_f16(a.x, b.x));
|
||||
return a;
|
||||
}
|
||||
EIGEN_STRONG_INLINE EIGEN_DEVICE_FUNC half& operator *= (half& a, const half& b) {
|
||||
a = half(vmulh_f16(a.x, b.x));
|
||||
return a;
|
||||
}
|
||||
EIGEN_STRONG_INLINE EIGEN_DEVICE_FUNC half& operator -= (half& a, const half& b) {
|
||||
a = half(vsubh_f16(a.x, b.x));
|
||||
return a;
|
||||
}
|
||||
EIGEN_STRONG_INLINE EIGEN_DEVICE_FUNC half& operator /= (half& a, const half& b) {
|
||||
a = half(vdivh_f16(a.x, b.x));
|
||||
return a;
|
||||
}
|
||||
EIGEN_STRONG_INLINE EIGEN_DEVICE_FUNC bool operator == (const half& a, const half& b) {
|
||||
return vceqh_f16(a.x, b.x);
|
||||
}
|
||||
EIGEN_STRONG_INLINE EIGEN_DEVICE_FUNC bool operator != (const half& a, const half& b) {
|
||||
return !vceqh_f16(a.x, b.x);
|
||||
}
|
||||
EIGEN_STRONG_INLINE EIGEN_DEVICE_FUNC bool operator < (const half& a, const half& b) {
|
||||
return vclth_f16(a.x, b.x);
|
||||
}
|
||||
EIGEN_STRONG_INLINE EIGEN_DEVICE_FUNC bool operator <= (const half& a, const half& b) {
|
||||
return vcleh_f16(a.x, b.x);
|
||||
}
|
||||
EIGEN_STRONG_INLINE EIGEN_DEVICE_FUNC bool operator > (const half& a, const half& b) {
|
||||
return vcgth_f16(a.x, b.x);
|
||||
}
|
||||
EIGEN_STRONG_INLINE EIGEN_DEVICE_FUNC bool operator >= (const half& a, const half& b) {
|
||||
return vcgeh_f16(a.x, b.x);
|
||||
}
|
||||
// We need to distinguish ‘clang as the CUDA compiler’ from ‘clang as the host compiler,
|
||||
// invoked by NVCC’ (e.g. on MacOS). The former needs to see both host and device implementation
|
||||
// of the functions, while the latter can only deal with one of them.
|
||||
#elif !defined(EIGEN_HAS_NATIVE_FP16) || (EIGEN_COMP_CLANG && !EIGEN_COMP_NVCC) // Emulate support for half floats
|
||||
|
||||
#if EIGEN_COMP_CLANG && defined(EIGEN_CUDACC)
|
||||
// We need to provide emulated *host-side* FP16 operators for clang.
|
||||
#pragma push_macro("EIGEN_DEVICE_FUNC")
|
||||
#undef EIGEN_DEVICE_FUNC
|
||||
#if defined(EIGEN_HAS_CUDA_FP16) && defined(EIGEN_HAS_NATIVE_FP16)
|
||||
#define EIGEN_DEVICE_FUNC __host__
|
||||
#else // both host and device need emulated ops.
|
||||
#define EIGEN_DEVICE_FUNC __host__ __device__
|
||||
#endif
|
||||
#endif
|
||||
|
||||
// Definitions for CPUs and older HIP+CUDA, mostly working through conversion
|
||||
// to/from fp32.
|
||||
EIGEN_STRONG_INLINE EIGEN_DEVICE_FUNC half operator + (const half& a, const half& b) {
|
||||
return half(float(a) + float(b));
|
||||
}
|
||||
EIGEN_STRONG_INLINE EIGEN_DEVICE_FUNC half operator * (const half& a, const half& b) {
|
||||
return half(float(a) * float(b));
|
||||
}
|
||||
EIGEN_STRONG_INLINE EIGEN_DEVICE_FUNC half operator - (const half& a, const half& b) {
|
||||
return half(float(a) - float(b));
|
||||
}
|
||||
EIGEN_STRONG_INLINE EIGEN_DEVICE_FUNC half operator / (const half& a, const half& b) {
|
||||
return half(float(a) / float(b));
|
||||
}
|
||||
EIGEN_STRONG_INLINE EIGEN_DEVICE_FUNC half operator - (const half& a) {
|
||||
half result;
|
||||
result.x = a.x ^ 0x8000;
|
||||
return result;
|
||||
}
|
||||
EIGEN_STRONG_INLINE EIGEN_DEVICE_FUNC half& operator += (half& a, const half& b) {
|
||||
a = half(float(a) + float(b));
|
||||
return a;
|
||||
}
|
||||
EIGEN_STRONG_INLINE EIGEN_DEVICE_FUNC half& operator *= (half& a, const half& b) {
|
||||
a = half(float(a) * float(b));
|
||||
return a;
|
||||
}
|
||||
EIGEN_STRONG_INLINE EIGEN_DEVICE_FUNC half& operator -= (half& a, const half& b) {
|
||||
a = half(float(a) - float(b));
|
||||
return a;
|
||||
}
|
||||
EIGEN_STRONG_INLINE EIGEN_DEVICE_FUNC half& operator /= (half& a, const half& b) {
|
||||
a = half(float(a) / float(b));
|
||||
return a;
|
||||
}
|
||||
EIGEN_STRONG_INLINE EIGEN_DEVICE_FUNC bool operator == (const half& a, const half& b) {
|
||||
return numext::equal_strict(float(a),float(b));
|
||||
}
|
||||
EIGEN_STRONG_INLINE EIGEN_DEVICE_FUNC bool operator != (const half& a, const half& b) {
|
||||
return numext::not_equal_strict(float(a), float(b));
|
||||
}
|
||||
EIGEN_STRONG_INLINE EIGEN_DEVICE_FUNC bool operator < (const half& a, const half& b) {
|
||||
return float(a) < float(b);
|
||||
}
|
||||
EIGEN_STRONG_INLINE EIGEN_DEVICE_FUNC bool operator <= (const half& a, const half& b) {
|
||||
return float(a) <= float(b);
|
||||
}
|
||||
EIGEN_STRONG_INLINE EIGEN_DEVICE_FUNC bool operator > (const half& a, const half& b) {
|
||||
return float(a) > float(b);
|
||||
}
|
||||
EIGEN_STRONG_INLINE EIGEN_DEVICE_FUNC bool operator >= (const half& a, const half& b) {
|
||||
return float(a) >= float(b);
|
||||
}
|
||||
|
||||
#if defined(__clang__) && defined(__CUDA__)
|
||||
#pragma pop_macro("EIGEN_DEVICE_FUNC")
|
||||
#endif
|
||||
#endif // Emulate support for half floats
|
||||
|
||||
// Division by an index. Do it in full float precision to avoid accuracy
|
||||
// issues in converting the denominator to half.
|
||||
EIGEN_STRONG_INLINE EIGEN_DEVICE_FUNC half operator / (const half& a, Index b) {
|
||||
return half(static_cast<float>(a) / static_cast<float>(b));
|
||||
}
|
||||
|
||||
EIGEN_STRONG_INLINE EIGEN_DEVICE_FUNC half operator++(half& a) {
|
||||
a += half(1);
|
||||
return a;
|
||||
}
|
||||
|
||||
EIGEN_STRONG_INLINE EIGEN_DEVICE_FUNC half operator--(half& a) {
|
||||
a -= half(1);
|
||||
return a;
|
||||
}
|
||||
|
||||
EIGEN_STRONG_INLINE EIGEN_DEVICE_FUNC half operator++(half& a, int) {
|
||||
half original_value = a;
|
||||
++a;
|
||||
return original_value;
|
||||
}
|
||||
|
||||
EIGEN_STRONG_INLINE EIGEN_DEVICE_FUNC half operator--(half& a, int) {
|
||||
half original_value = a;
|
||||
--a;
|
||||
return original_value;
|
||||
}
|
||||
|
||||
// Conversion routines, including fallbacks for the host or older CUDA.
|
||||
// Note that newer Intel CPUs (Haswell or newer) have vectorized versions of
|
||||
// these in hardware. If we need more performance on older/other CPUs, they are
|
||||
// also possible to vectorize directly.
|
||||
|
||||
EIGEN_STRONG_INLINE EIGEN_DEVICE_FUNC EIGEN_CONSTEXPR __half_raw raw_uint16_to_half(numext::uint16_t x) {
|
||||
// We cannot simply do a "return __half_raw(x)" here, because __half_raw is union type
|
||||
// in the hip_fp16 header file, and that will trigger a compile error
|
||||
// On the other hand, having anything but a return statement also triggers a compile error
|
||||
// because this is constexpr function.
|
||||
// Fortunately, since we need to disable EIGEN_CONSTEXPR for GPU anyway, we can get out
|
||||
// of this catch22 by having separate bodies for GPU / non GPU
|
||||
#if defined(EIGEN_HAS_GPU_FP16)
|
||||
__half_raw h;
|
||||
h.x = x;
|
||||
return h;
|
||||
#else
|
||||
return __half_raw(x);
|
||||
#endif
|
||||
}
|
||||
|
||||
EIGEN_STRONG_INLINE EIGEN_DEVICE_FUNC numext::uint16_t raw_half_as_uint16(const __half_raw& h) {
|
||||
// HIP/CUDA/Default have a member 'x' of type uint16_t.
|
||||
// For ARM64 native half, the member 'x' is of type __fp16, so we need to bit-cast.
|
||||
// For SYCL, cl::sycl::half is _Float16, so cast directly.
|
||||
#if defined(EIGEN_HAS_ARM64_FP16_SCALAR_ARITHMETIC)
|
||||
return numext::bit_cast<numext::uint16_t>(h.x);
|
||||
#elif defined(SYCL_DEVICE_ONLY)
|
||||
return numext::bit_cast<numext::uint16_t>(h);
|
||||
#else
|
||||
return h.x;
|
||||
#endif
|
||||
}
|
||||
|
||||
union float32_bits {
|
||||
unsigned int u;
|
||||
float f;
|
||||
};
|
||||
|
||||
EIGEN_STRONG_INLINE EIGEN_DEVICE_FUNC __half_raw float_to_half_rtne(float ff) {
|
||||
#if (defined(EIGEN_HAS_CUDA_FP16) && defined(EIGEN_CUDA_ARCH) && EIGEN_CUDA_ARCH >= 300) || \
|
||||
(defined(EIGEN_HAS_HIP_FP16) && defined(EIGEN_HIP_DEVICE_COMPILE))
|
||||
__half tmp_ff = __float2half(ff);
|
||||
return *(__half_raw*)&tmp_ff;
|
||||
|
||||
#elif defined(EIGEN_HAS_FP16_C)
|
||||
__half_raw h;
|
||||
h.x = _cvtss_sh(ff, 0);
|
||||
return h;
|
||||
|
||||
#elif defined(EIGEN_HAS_ARM64_FP16_SCALAR_ARITHMETIC)
|
||||
__half_raw h;
|
||||
h.x = static_cast<__fp16>(ff);
|
||||
return h;
|
||||
|
||||
#else
|
||||
float32_bits f; f.f = ff;
|
||||
|
||||
const float32_bits f32infty = { 255 << 23 };
|
||||
const float32_bits f16max = { (127 + 16) << 23 };
|
||||
const float32_bits denorm_magic = { ((127 - 15) + (23 - 10) + 1) << 23 };
|
||||
unsigned int sign_mask = 0x80000000u;
|
||||
__half_raw o;
|
||||
o.x = static_cast<numext::uint16_t>(0x0u);
|
||||
|
||||
unsigned int sign = f.u & sign_mask;
|
||||
f.u ^= sign;
|
||||
|
||||
// NOTE all the integer compares in this function can be safely
|
||||
// compiled into signed compares since all operands are below
|
||||
// 0x80000000. Important if you want fast straight SSE2 code
|
||||
// (since there's no unsigned PCMPGTD).
|
||||
|
||||
if (f.u >= f16max.u) { // result is Inf or NaN (all exponent bits set)
|
||||
o.x = (f.u > f32infty.u) ? 0x7e00 : 0x7c00; // NaN->qNaN and Inf->Inf
|
||||
} else { // (De)normalized number or zero
|
||||
if (f.u < (113 << 23)) { // resulting FP16 is subnormal or zero
|
||||
// use a magic value to align our 10 mantissa bits at the bottom of
|
||||
// the float. as long as FP addition is round-to-nearest-even this
|
||||
// just works.
|
||||
f.f += denorm_magic.f;
|
||||
|
||||
// and one integer subtract of the bias later, we have our final float!
|
||||
o.x = static_cast<numext::uint16_t>(f.u - denorm_magic.u);
|
||||
} else {
|
||||
unsigned int mant_odd = (f.u >> 13) & 1; // resulting mantissa is odd
|
||||
|
||||
// update exponent, rounding bias part 1
|
||||
// Equivalent to `f.u += ((unsigned int)(15 - 127) << 23) + 0xfff`, but
|
||||
// without arithmetic overflow.
|
||||
f.u += 0xc8000fffU;
|
||||
// rounding bias part 2
|
||||
f.u += mant_odd;
|
||||
// take the bits!
|
||||
o.x = static_cast<numext::uint16_t>(f.u >> 13);
|
||||
}
|
||||
}
|
||||
|
||||
o.x |= static_cast<numext::uint16_t>(sign >> 16);
|
||||
return o;
|
||||
#endif
|
||||
}
|
||||
|
||||
EIGEN_STRONG_INLINE EIGEN_DEVICE_FUNC float half_to_float(__half_raw h) {
|
||||
#if (defined(EIGEN_HAS_CUDA_FP16) && defined(EIGEN_CUDA_ARCH) && EIGEN_CUDA_ARCH >= 300) || \
|
||||
(defined(EIGEN_HAS_HIP_FP16) && defined(EIGEN_HIP_DEVICE_COMPILE))
|
||||
return __half2float(h);
|
||||
#elif defined(EIGEN_HAS_FP16_C)
|
||||
return _cvtsh_ss(h.x);
|
||||
#elif defined(EIGEN_HAS_ARM64_FP16_SCALAR_ARITHMETIC)
|
||||
return static_cast<float>(h.x);
|
||||
#else
|
||||
const float32_bits magic = { 113 << 23 };
|
||||
const unsigned int shifted_exp = 0x7c00 << 13; // exponent mask after shift
|
||||
float32_bits o;
|
||||
|
||||
o.u = (h.x & 0x7fff) << 13; // exponent/mantissa bits
|
||||
unsigned int exp = shifted_exp & o.u; // just the exponent
|
||||
o.u += (127 - 15) << 23; // exponent adjust
|
||||
|
||||
// handle exponent special cases
|
||||
if (exp == shifted_exp) { // Inf/NaN?
|
||||
o.u += (128 - 16) << 23; // extra exp adjust
|
||||
} else if (exp == 0) { // Zero/Denormal?
|
||||
o.u += 1 << 23; // extra exp adjust
|
||||
o.f -= magic.f; // renormalize
|
||||
}
|
||||
|
||||
o.u |= (h.x & 0x8000) << 16; // sign bit
|
||||
return o.f;
|
||||
#endif
|
||||
}
|
||||
|
||||
// --- standard functions ---
|
||||
|
||||
EIGEN_STRONG_INLINE EIGEN_DEVICE_FUNC bool (isinf)(const half& a) {
|
||||
#ifdef EIGEN_HAS_ARM64_FP16_SCALAR_ARITHMETIC
|
||||
return (numext::bit_cast<numext::uint16_t>(a.x) & 0x7fff) == 0x7c00;
|
||||
#else
|
||||
return (a.x & 0x7fff) == 0x7c00;
|
||||
#endif
|
||||
}
|
||||
EIGEN_STRONG_INLINE EIGEN_DEVICE_FUNC bool (isnan)(const half& a) {
|
||||
#if (defined(EIGEN_HAS_CUDA_FP16) && defined(EIGEN_CUDA_ARCH) && EIGEN_CUDA_ARCH >= 530) || \
|
||||
(defined(EIGEN_HAS_HIP_FP16) && defined(EIGEN_HIP_DEVICE_COMPILE))
|
||||
return __hisnan(a);
|
||||
#elif defined(EIGEN_HAS_ARM64_FP16_SCALAR_ARITHMETIC)
|
||||
return (numext::bit_cast<numext::uint16_t>(a.x) & 0x7fff) > 0x7c00;
|
||||
#else
|
||||
return (a.x & 0x7fff) > 0x7c00;
|
||||
#endif
|
||||
}
|
||||
EIGEN_STRONG_INLINE EIGEN_DEVICE_FUNC bool (isfinite)(const half& a) {
|
||||
return !(isinf EIGEN_NOT_A_MACRO (a)) && !(isnan EIGEN_NOT_A_MACRO (a));
|
||||
}
|
||||
|
||||
EIGEN_STRONG_INLINE EIGEN_DEVICE_FUNC half abs(const half& a) {
|
||||
#if defined(EIGEN_HAS_ARM64_FP16_SCALAR_ARITHMETIC)
|
||||
return half(vabsh_f16(a.x));
|
||||
#else
|
||||
half result;
|
||||
result.x = a.x & 0x7FFF;
|
||||
return result;
|
||||
#endif
|
||||
}
|
||||
EIGEN_STRONG_INLINE EIGEN_DEVICE_FUNC half exp(const half& a) {
|
||||
#if (EIGEN_CUDA_SDK_VER >= 80000 && defined EIGEN_CUDA_ARCH && EIGEN_CUDA_ARCH >= 530) || \
|
||||
defined(EIGEN_HIP_DEVICE_COMPILE)
|
||||
return half(hexp(a));
|
||||
#else
|
||||
return half(::expf(float(a)));
|
||||
#endif
|
||||
}
|
||||
EIGEN_STRONG_INLINE EIGEN_DEVICE_FUNC half expm1(const half& a) {
|
||||
return half(numext::expm1(float(a)));
|
||||
}
|
||||
EIGEN_STRONG_INLINE EIGEN_DEVICE_FUNC half log(const half& a) {
|
||||
#if (defined(EIGEN_HAS_CUDA_FP16) && EIGEN_CUDA_SDK_VER >= 80000 && defined(EIGEN_CUDA_ARCH) && EIGEN_CUDA_ARCH >= 530) || \
|
||||
(defined(EIGEN_HAS_HIP_FP16) && defined(EIGEN_HIP_DEVICE_COMPILE))
|
||||
return half(::hlog(a));
|
||||
#else
|
||||
return half(::logf(float(a)));
|
||||
#endif
|
||||
}
|
||||
EIGEN_STRONG_INLINE EIGEN_DEVICE_FUNC half log1p(const half& a) {
|
||||
return half(numext::log1p(float(a)));
|
||||
}
|
||||
EIGEN_STRONG_INLINE EIGEN_DEVICE_FUNC half log10(const half& a) {
|
||||
return half(::log10f(float(a)));
|
||||
}
|
||||
EIGEN_STRONG_INLINE EIGEN_DEVICE_FUNC half log2(const half& a) {
|
||||
return half(static_cast<float>(EIGEN_LOG2E) * ::logf(float(a)));
|
||||
}
|
||||
|
||||
EIGEN_STRONG_INLINE EIGEN_DEVICE_FUNC half sqrt(const half& a) {
|
||||
#if (EIGEN_CUDA_SDK_VER >= 80000 && defined EIGEN_CUDA_ARCH && EIGEN_CUDA_ARCH >= 530) || \
|
||||
defined(EIGEN_HIP_DEVICE_COMPILE)
|
||||
return half(hsqrt(a));
|
||||
#else
|
||||
return half(::sqrtf(float(a)));
|
||||
#endif
|
||||
}
|
||||
EIGEN_STRONG_INLINE EIGEN_DEVICE_FUNC half pow(const half& a, const half& b) {
|
||||
return half(::powf(float(a), float(b)));
|
||||
}
|
||||
EIGEN_STRONG_INLINE EIGEN_DEVICE_FUNC half sin(const half& a) {
|
||||
return half(::sinf(float(a)));
|
||||
}
|
||||
EIGEN_STRONG_INLINE EIGEN_DEVICE_FUNC half cos(const half& a) {
|
||||
return half(::cosf(float(a)));
|
||||
}
|
||||
EIGEN_STRONG_INLINE EIGEN_DEVICE_FUNC half tan(const half& a) {
|
||||
return half(::tanf(float(a)));
|
||||
}
|
||||
EIGEN_STRONG_INLINE EIGEN_DEVICE_FUNC half tanh(const half& a) {
|
||||
return half(::tanhf(float(a)));
|
||||
}
|
||||
EIGEN_STRONG_INLINE EIGEN_DEVICE_FUNC half asin(const half& a) {
|
||||
return half(::asinf(float(a)));
|
||||
}
|
||||
EIGEN_STRONG_INLINE EIGEN_DEVICE_FUNC half acos(const half& a) {
|
||||
return half(::acosf(float(a)));
|
||||
}
|
||||
EIGEN_STRONG_INLINE EIGEN_DEVICE_FUNC half floor(const half& a) {
|
||||
#if (EIGEN_CUDA_SDK_VER >= 80000 && defined EIGEN_CUDA_ARCH && EIGEN_CUDA_ARCH >= 300) || \
|
||||
defined(EIGEN_HIP_DEVICE_COMPILE)
|
||||
return half(hfloor(a));
|
||||
#else
|
||||
return half(::floorf(float(a)));
|
||||
#endif
|
||||
}
|
||||
EIGEN_STRONG_INLINE EIGEN_DEVICE_FUNC half ceil(const half& a) {
|
||||
#if (EIGEN_CUDA_SDK_VER >= 80000 && defined EIGEN_CUDA_ARCH && EIGEN_CUDA_ARCH >= 300) || \
|
||||
defined(EIGEN_HIP_DEVICE_COMPILE)
|
||||
return half(hceil(a));
|
||||
#else
|
||||
return half(::ceilf(float(a)));
|
||||
#endif
|
||||
}
|
||||
EIGEN_STRONG_INLINE EIGEN_DEVICE_FUNC half rint(const half& a) {
|
||||
return half(::rintf(float(a)));
|
||||
}
|
||||
EIGEN_STRONG_INLINE EIGEN_DEVICE_FUNC half round(const half& a) {
|
||||
return half(::roundf(float(a)));
|
||||
}
|
||||
EIGEN_STRONG_INLINE EIGEN_DEVICE_FUNC half fmod(const half& a, const half& b) {
|
||||
return half(::fmodf(float(a), float(b)));
|
||||
}
|
||||
|
||||
EIGEN_STRONG_INLINE EIGEN_DEVICE_FUNC half (min)(const half& a, const half& b) {
|
||||
#if (defined(EIGEN_HAS_CUDA_FP16) && defined(EIGEN_CUDA_ARCH) && EIGEN_CUDA_ARCH >= 530) || \
|
||||
(defined(EIGEN_HAS_HIP_FP16) && defined(EIGEN_HIP_DEVICE_COMPILE))
|
||||
return __hlt(b, a) ? b : a;
|
||||
#else
|
||||
const float f1 = static_cast<float>(a);
|
||||
const float f2 = static_cast<float>(b);
|
||||
return f2 < f1 ? b : a;
|
||||
#endif
|
||||
}
|
||||
EIGEN_STRONG_INLINE EIGEN_DEVICE_FUNC half (max)(const half& a, const half& b) {
|
||||
#if (defined(EIGEN_HAS_CUDA_FP16) && defined(EIGEN_CUDA_ARCH) && EIGEN_CUDA_ARCH >= 530) || \
|
||||
(defined(EIGEN_HAS_HIP_FP16) && defined(EIGEN_HIP_DEVICE_COMPILE))
|
||||
return __hlt(a, b) ? b : a;
|
||||
#else
|
||||
const float f1 = static_cast<float>(a);
|
||||
const float f2 = static_cast<float>(b);
|
||||
return f1 < f2 ? b : a;
|
||||
#endif
|
||||
}
|
||||
|
||||
#ifndef EIGEN_NO_IO
|
||||
EIGEN_ALWAYS_INLINE std::ostream& operator << (std::ostream& os, const half& v) {
|
||||
os << static_cast<float>(v);
|
||||
return os;
|
||||
}
|
||||
#endif
|
||||
|
||||
} // end namespace half_impl
|
||||
|
||||
// import Eigen::half_impl::half into Eigen namespace
|
||||
// using half_impl::half;
|
||||
|
||||
namespace internal {
|
||||
|
||||
template<>
|
||||
struct random_default_impl<half, false, false>
|
||||
{
|
||||
static inline half run(const half& x, const half& y)
|
||||
{
|
||||
return x + (y-x) * half(float(std::rand()) / float(RAND_MAX));
|
||||
}
|
||||
static inline half run()
|
||||
{
|
||||
return run(half(-1.f), half(1.f));
|
||||
}
|
||||
};
|
||||
|
||||
template<> struct is_arithmetic<half> { enum { value = true }; };
|
||||
|
||||
} // end namespace internal
|
||||
|
||||
template<> struct NumTraits<Eigen::half>
|
||||
: GenericNumTraits<Eigen::half>
|
||||
{
|
||||
enum {
|
||||
IsSigned = true,
|
||||
IsInteger = false,
|
||||
IsComplex = false,
|
||||
RequireInitialization = false
|
||||
};
|
||||
|
||||
EIGEN_DEVICE_FUNC EIGEN_CONSTEXPR static EIGEN_STRONG_INLINE Eigen::half epsilon() {
|
||||
return half_impl::raw_uint16_to_half(0x0800);
|
||||
}
|
||||
EIGEN_DEVICE_FUNC EIGEN_CONSTEXPR static EIGEN_STRONG_INLINE Eigen::half dummy_precision() {
|
||||
return half_impl::raw_uint16_to_half(0x211f); // Eigen::half(1e-2f);
|
||||
}
|
||||
EIGEN_DEVICE_FUNC EIGEN_CONSTEXPR static EIGEN_STRONG_INLINE Eigen::half highest() {
|
||||
return half_impl::raw_uint16_to_half(0x7bff);
|
||||
}
|
||||
EIGEN_DEVICE_FUNC EIGEN_CONSTEXPR static EIGEN_STRONG_INLINE Eigen::half lowest() {
|
||||
return half_impl::raw_uint16_to_half(0xfbff);
|
||||
}
|
||||
EIGEN_DEVICE_FUNC EIGEN_CONSTEXPR static EIGEN_STRONG_INLINE Eigen::half infinity() {
|
||||
return half_impl::raw_uint16_to_half(0x7c00);
|
||||
}
|
||||
EIGEN_DEVICE_FUNC EIGEN_CONSTEXPR static EIGEN_STRONG_INLINE Eigen::half quiet_NaN() {
|
||||
return half_impl::raw_uint16_to_half(0x7e00);
|
||||
}
|
||||
};
|
||||
|
||||
} // end namespace Eigen
|
||||
|
||||
#if defined(EIGEN_HAS_GPU_FP16) || defined(EIGEN_HAS_ARM64_FP16_SCALAR_ARITHMETIC)
|
||||
#pragma pop_macro("EIGEN_CONSTEXPR")
|
||||
#endif
|
||||
|
||||
namespace Eigen {
|
||||
namespace numext {
|
||||
|
||||
#if defined(EIGEN_GPU_COMPILE_PHASE)
|
||||
|
||||
template <>
|
||||
EIGEN_DEVICE_FUNC EIGEN_ALWAYS_INLINE bool(isnan)(const Eigen::half& h) {
|
||||
return (half_impl::isnan)(h);
|
||||
}
|
||||
|
||||
template <>
|
||||
EIGEN_DEVICE_FUNC EIGEN_ALWAYS_INLINE bool(isinf)(const Eigen::half& h) {
|
||||
return (half_impl::isinf)(h);
|
||||
}
|
||||
|
||||
template <>
|
||||
EIGEN_DEVICE_FUNC EIGEN_ALWAYS_INLINE bool(isfinite)(const Eigen::half& h) {
|
||||
return (half_impl::isfinite)(h);
|
||||
}
|
||||
|
||||
#endif
|
||||
|
||||
template <>
|
||||
EIGEN_STRONG_INLINE EIGEN_DEVICE_FUNC Eigen::half bit_cast<Eigen::half, uint16_t>(const uint16_t& src) {
|
||||
return Eigen::half(Eigen::half_impl::raw_uint16_to_half(src));
|
||||
}
|
||||
|
||||
template <>
|
||||
EIGEN_STRONG_INLINE EIGEN_DEVICE_FUNC uint16_t bit_cast<uint16_t, Eigen::half>(const Eigen::half& src) {
|
||||
return Eigen::half_impl::raw_half_as_uint16(src);
|
||||
}
|
||||
|
||||
} // namespace numext
|
||||
} // namespace Eigen
|
||||
|
||||
// Add the missing shfl* intrinsics.
|
||||
// The __shfl* functions are only valid on HIP or _CUDA_ARCH_ >= 300.
|
||||
// CUDA defines them for (__CUDA_ARCH__ >= 300 || !defined(__CUDA_ARCH__))
|
||||
//
|
||||
// HIP and CUDA prior to SDK 9.0 define
|
||||
// __shfl, __shfl_up, __shfl_down, __shfl_xor for int and float
|
||||
// CUDA since 9.0 deprecates those and instead defines
|
||||
// __shfl_sync, __shfl_up_sync, __shfl_down_sync, __shfl_xor_sync,
|
||||
// with native support for __half and __nv_bfloat16
|
||||
//
|
||||
// Note that the following are __device__ - only functions.
|
||||
#if (defined(EIGEN_CUDACC) && (!defined(EIGEN_CUDA_ARCH) || EIGEN_CUDA_ARCH >= 300)) \
|
||||
|| defined(EIGEN_HIPCC)
|
||||
|
||||
#if defined(EIGEN_HAS_CUDA_FP16) && EIGEN_CUDA_SDK_VER >= 90000
|
||||
|
||||
__device__ EIGEN_STRONG_INLINE Eigen::half __shfl_sync(unsigned mask, Eigen::half var, int srcLane, int width=warpSize) {
|
||||
const __half h = var;
|
||||
return static_cast<Eigen::half>(__shfl_sync(mask, h, srcLane, width));
|
||||
}
|
||||
|
||||
__device__ EIGEN_STRONG_INLINE Eigen::half __shfl_up_sync(unsigned mask, Eigen::half var, unsigned int delta, int width=warpSize) {
|
||||
const __half h = var;
|
||||
return static_cast<Eigen::half>(__shfl_up_sync(mask, h, delta, width));
|
||||
}
|
||||
|
||||
__device__ EIGEN_STRONG_INLINE Eigen::half __shfl_down_sync(unsigned mask, Eigen::half var, unsigned int delta, int width=warpSize) {
|
||||
const __half h = var;
|
||||
return static_cast<Eigen::half>(__shfl_down_sync(mask, h, delta, width));
|
||||
}
|
||||
|
||||
__device__ EIGEN_STRONG_INLINE Eigen::half __shfl_xor_sync(unsigned mask, Eigen::half var, int laneMask, int width=warpSize) {
|
||||
const __half h = var;
|
||||
return static_cast<Eigen::half>(__shfl_xor_sync(mask, h, laneMask, width));
|
||||
}
|
||||
|
||||
#else // HIP or CUDA SDK < 9.0
|
||||
|
||||
__device__ EIGEN_STRONG_INLINE Eigen::half __shfl(Eigen::half var, int srcLane, int width=warpSize) {
|
||||
const int ivar = static_cast<int>(Eigen::numext::bit_cast<Eigen::numext::uint16_t>(var));
|
||||
return Eigen::numext::bit_cast<Eigen::half>(static_cast<Eigen::numext::uint16_t>(__shfl(ivar, srcLane, width)));
|
||||
}
|
||||
|
||||
__device__ EIGEN_STRONG_INLINE Eigen::half __shfl_up(Eigen::half var, unsigned int delta, int width=warpSize) {
|
||||
const int ivar = static_cast<int>(Eigen::numext::bit_cast<Eigen::numext::uint16_t>(var));
|
||||
return Eigen::numext::bit_cast<Eigen::half>(static_cast<Eigen::numext::uint16_t>(__shfl_up(ivar, delta, width)));
|
||||
}
|
||||
|
||||
__device__ EIGEN_STRONG_INLINE Eigen::half __shfl_down(Eigen::half var, unsigned int delta, int width=warpSize) {
|
||||
const int ivar = static_cast<int>(Eigen::numext::bit_cast<Eigen::numext::uint16_t>(var));
|
||||
return Eigen::numext::bit_cast<Eigen::half>(static_cast<Eigen::numext::uint16_t>(__shfl_down(ivar, delta, width)));
|
||||
}
|
||||
|
||||
__device__ EIGEN_STRONG_INLINE Eigen::half __shfl_xor(Eigen::half var, int laneMask, int width=warpSize) {
|
||||
const int ivar = static_cast<int>(Eigen::numext::bit_cast<Eigen::numext::uint16_t>(var));
|
||||
return Eigen::numext::bit_cast<Eigen::half>(static_cast<Eigen::numext::uint16_t>(__shfl_xor(ivar, laneMask, width)));
|
||||
}
|
||||
|
||||
#endif // HIP vs CUDA
|
||||
#endif // __shfl*
|
||||
|
||||
// ldg() has an overload for __half_raw, but we also need one for Eigen::half.
|
||||
#if (defined(EIGEN_CUDACC) && (!defined(EIGEN_CUDA_ARCH) || EIGEN_CUDA_ARCH >= 350)) \
|
||||
|| defined(EIGEN_HIPCC)
|
||||
EIGEN_STRONG_INLINE __device__ Eigen::half __ldg(const Eigen::half* ptr) {
|
||||
return Eigen::half_impl::raw_uint16_to_half(__ldg(reinterpret_cast<const Eigen::numext::uint16_t*>(ptr)));
|
||||
}
|
||||
#endif // __ldg
|
||||
|
||||
#if EIGEN_HAS_STD_HASH
|
||||
namespace std {
|
||||
template <>
|
||||
struct hash<Eigen::half> {
|
||||
EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE std::size_t operator()(const Eigen::half& a) const {
|
||||
return static_cast<std::size_t>(Eigen::numext::bit_cast<Eigen::numext::uint16_t>(a));
|
||||
}
|
||||
};
|
||||
} // end namespace std
|
||||
#endif
|
||||
|
||||
#endif // EIGEN_HALF_H
|
||||
@@ -0,0 +1,49 @@
|
||||
// This file is part of Eigen, a lightweight C++ template library
|
||||
// for linear algebra.
|
||||
//
|
||||
// Copyright (C) 2008-2010 Gael Guennebaud <gael.guennebaud@inria.fr>
|
||||
// Copyright (C) 2006-2008 Benoit Jacob <jacob.benoit.1@gmail.com>
|
||||
//
|
||||
// This Source Code Form is subject to the terms of the Mozilla
|
||||
// Public License v. 2.0. If a copy of the MPL was not distributed
|
||||
// with this file, You can obtain one at http://mozilla.org/MPL/2.0/.
|
||||
|
||||
|
||||
/* All the parameters defined in this file can be specialized in the
|
||||
* architecture specific files, and/or by the user.
|
||||
* More to come... */
|
||||
|
||||
#ifndef EIGEN_DEFAULT_SETTINGS_H
|
||||
#define EIGEN_DEFAULT_SETTINGS_H
|
||||
|
||||
/** Defines the maximal loop size to enable meta unrolling of loops.
|
||||
* Note that the value here is expressed in Eigen's own notion of "number of FLOPS",
|
||||
* it does not correspond to the number of iterations or the number of instructions
|
||||
*/
|
||||
#ifndef EIGEN_UNROLLING_LIMIT
|
||||
#define EIGEN_UNROLLING_LIMIT 110
|
||||
#endif
|
||||
|
||||
/** Defines the threshold between a "small" and a "large" matrix.
|
||||
* This threshold is mainly used to select the proper product implementation.
|
||||
*/
|
||||
#ifndef EIGEN_CACHEFRIENDLY_PRODUCT_THRESHOLD
|
||||
#define EIGEN_CACHEFRIENDLY_PRODUCT_THRESHOLD 8
|
||||
#endif
|
||||
|
||||
/** Defines the maximal width of the blocks used in the triangular product and solver
|
||||
* for vectors (level 2 blas xTRMV and xTRSV). The default is 8.
|
||||
*/
|
||||
#ifndef EIGEN_TUNE_TRIANGULAR_PANEL_WIDTH
|
||||
#define EIGEN_TUNE_TRIANGULAR_PANEL_WIDTH 8
|
||||
#endif
|
||||
|
||||
|
||||
/** Defines the default number of registers available for that architecture.
|
||||
* Currently it must be 8 or 16. Other values will fail.
|
||||
*/
|
||||
#ifndef EIGEN_ARCH_DEFAULT_NUMBER_OF_REGISTERS
|
||||
#define EIGEN_ARCH_DEFAULT_NUMBER_OF_REGISTERS 8
|
||||
#endif
|
||||
|
||||
#endif // EIGEN_DEFAULT_SETTINGS_H
|
||||
@@ -0,0 +1,120 @@
|
||||
// This file is part of Eigen, a lightweight C++ template library
|
||||
// for linear algebra.
|
||||
//
|
||||
// Copyright (C) 2016 Benoit Steiner <benoit.steiner.goog@gmail.com>
|
||||
// Copyright (C) 2019 Rasmus Munk Larsen <rmlarsen@google.com>
|
||||
//
|
||||
// This Source Code Form is subject to the terms of the Mozilla
|
||||
// Public License v. 2.0. If a copy of the MPL was not distributed
|
||||
// with this file, You can obtain one at http://mozilla.org/MPL/2.0/.
|
||||
|
||||
#ifndef EIGEN_GENERIC_TYPE_CASTING_H
|
||||
#define EIGEN_GENERIC_TYPE_CASTING_H
|
||||
|
||||
namespace Eigen {
|
||||
|
||||
namespace internal {
|
||||
|
||||
template<>
|
||||
struct scalar_cast_op<float, Eigen::half> {
|
||||
EIGEN_EMPTY_STRUCT_CTOR(scalar_cast_op)
|
||||
typedef Eigen::half result_type;
|
||||
EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE Eigen::half operator() (const float& a) const {
|
||||
#if (defined(EIGEN_HAS_CUDA_FP16) && defined(EIGEN_CUDA_ARCH) && EIGEN_CUDA_ARCH >= 300) || \
|
||||
(defined(EIGEN_HAS_HIP_FP16) && defined(EIGEN_HIP_DEVICE_COMPILE))
|
||||
return __float2half(a);
|
||||
#else
|
||||
return Eigen::half(a);
|
||||
#endif
|
||||
}
|
||||
};
|
||||
|
||||
template<>
|
||||
struct functor_traits<scalar_cast_op<float, Eigen::half> >
|
||||
{ enum { Cost = NumTraits<float>::AddCost, PacketAccess = false }; };
|
||||
|
||||
|
||||
template<>
|
||||
struct scalar_cast_op<int, Eigen::half> {
|
||||
EIGEN_EMPTY_STRUCT_CTOR(scalar_cast_op)
|
||||
typedef Eigen::half result_type;
|
||||
EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE Eigen::half operator() (const int& a) const {
|
||||
#if (defined(EIGEN_HAS_CUDA_FP16) && defined(EIGEN_CUDA_ARCH) && EIGEN_CUDA_ARCH >= 300) || \
|
||||
(defined(EIGEN_HAS_HIP_FP16) && defined(EIGEN_HIP_DEVICE_COMPILE))
|
||||
return __float2half(static_cast<float>(a));
|
||||
#else
|
||||
return Eigen::half(static_cast<float>(a));
|
||||
#endif
|
||||
}
|
||||
};
|
||||
|
||||
template<>
|
||||
struct functor_traits<scalar_cast_op<int, Eigen::half> >
|
||||
{ enum { Cost = NumTraits<float>::AddCost, PacketAccess = false }; };
|
||||
|
||||
|
||||
template<>
|
||||
struct scalar_cast_op<Eigen::half, float> {
|
||||
EIGEN_EMPTY_STRUCT_CTOR(scalar_cast_op)
|
||||
typedef float result_type;
|
||||
EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE float operator() (const Eigen::half& a) const {
|
||||
#if (defined(EIGEN_HAS_CUDA_FP16) && defined(EIGEN_CUDA_ARCH) && EIGEN_CUDA_ARCH >= 300) || \
|
||||
(defined(EIGEN_HAS_HIP_FP16) && defined(EIGEN_HIP_DEVICE_COMPILE))
|
||||
return __half2float(a);
|
||||
#else
|
||||
return static_cast<float>(a);
|
||||
#endif
|
||||
}
|
||||
};
|
||||
|
||||
template<>
|
||||
struct functor_traits<scalar_cast_op<Eigen::half, float> >
|
||||
{ enum { Cost = NumTraits<float>::AddCost, PacketAccess = false }; };
|
||||
|
||||
|
||||
template<>
|
||||
struct scalar_cast_op<float, Eigen::bfloat16> {
|
||||
EIGEN_EMPTY_STRUCT_CTOR(scalar_cast_op)
|
||||
typedef Eigen::bfloat16 result_type;
|
||||
EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE Eigen::bfloat16 operator() (const float& a) const {
|
||||
return Eigen::bfloat16(a);
|
||||
}
|
||||
};
|
||||
|
||||
template<>
|
||||
struct functor_traits<scalar_cast_op<float, Eigen::bfloat16> >
|
||||
{ enum { Cost = NumTraits<float>::AddCost, PacketAccess = false }; };
|
||||
|
||||
|
||||
template<>
|
||||
struct scalar_cast_op<int, Eigen::bfloat16> {
|
||||
EIGEN_EMPTY_STRUCT_CTOR(scalar_cast_op)
|
||||
typedef Eigen::bfloat16 result_type;
|
||||
EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE Eigen::bfloat16 operator() (const int& a) const {
|
||||
return Eigen::bfloat16(static_cast<float>(a));
|
||||
}
|
||||
};
|
||||
|
||||
template<>
|
||||
struct functor_traits<scalar_cast_op<int, Eigen::bfloat16> >
|
||||
{ enum { Cost = NumTraits<float>::AddCost, PacketAccess = false }; };
|
||||
|
||||
|
||||
template<>
|
||||
struct scalar_cast_op<Eigen::bfloat16, float> {
|
||||
EIGEN_EMPTY_STRUCT_CTOR(scalar_cast_op)
|
||||
typedef float result_type;
|
||||
EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE float operator() (const Eigen::bfloat16& a) const {
|
||||
return static_cast<float>(a);
|
||||
}
|
||||
};
|
||||
|
||||
template<>
|
||||
struct functor_traits<scalar_cast_op<Eigen::bfloat16, float> >
|
||||
{ enum { Cost = NumTraits<float>::AddCost, PacketAccess = false }; };
|
||||
|
||||
|
||||
}
|
||||
}
|
||||
|
||||
#endif // EIGEN_GENERIC_TYPE_CASTING_H
|
||||
@@ -0,0 +1,103 @@
|
||||
// This file is part of Eigen, a lightweight C++ template library
|
||||
// for linear algebra.
|
||||
//
|
||||
// Copyright (C) 2014 Benoit Steiner <benoit.steiner.goog@gmail.com>
|
||||
//
|
||||
// This Source Code Form is subject to the terms of the Mozilla
|
||||
// Public License v. 2.0. If a copy of the MPL was not distributed
|
||||
// with this file, You can obtain one at http://mozilla.org/MPL/2.0/.
|
||||
|
||||
#ifndef EIGEN_MATH_FUNCTIONS_GPU_H
|
||||
#define EIGEN_MATH_FUNCTIONS_GPU_H
|
||||
|
||||
namespace Eigen {
|
||||
|
||||
namespace internal {
|
||||
|
||||
// Make sure this is only available when targeting a GPU: we don't want to
|
||||
// introduce conflicts between these packet_traits definitions and the ones
|
||||
// we'll use on the host side (SSE, AVX, ...)
|
||||
#if defined(EIGEN_GPUCC) && defined(EIGEN_USE_GPU)
|
||||
template<> EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
|
||||
float4 plog<float4>(const float4& a)
|
||||
{
|
||||
return make_float4(logf(a.x), logf(a.y), logf(a.z), logf(a.w));
|
||||
}
|
||||
|
||||
template<> EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
|
||||
double2 plog<double2>(const double2& a)
|
||||
{
|
||||
using ::log;
|
||||
return make_double2(log(a.x), log(a.y));
|
||||
}
|
||||
|
||||
template<> EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
|
||||
float4 plog1p<float4>(const float4& a)
|
||||
{
|
||||
return make_float4(log1pf(a.x), log1pf(a.y), log1pf(a.z), log1pf(a.w));
|
||||
}
|
||||
|
||||
template<> EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
|
||||
double2 plog1p<double2>(const double2& a)
|
||||
{
|
||||
return make_double2(log1p(a.x), log1p(a.y));
|
||||
}
|
||||
|
||||
template<> EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
|
||||
float4 pexp<float4>(const float4& a)
|
||||
{
|
||||
return make_float4(expf(a.x), expf(a.y), expf(a.z), expf(a.w));
|
||||
}
|
||||
|
||||
template<> EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
|
||||
double2 pexp<double2>(const double2& a)
|
||||
{
|
||||
using ::exp;
|
||||
return make_double2(exp(a.x), exp(a.y));
|
||||
}
|
||||
|
||||
template<> EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
|
||||
float4 pexpm1<float4>(const float4& a)
|
||||
{
|
||||
return make_float4(expm1f(a.x), expm1f(a.y), expm1f(a.z), expm1f(a.w));
|
||||
}
|
||||
|
||||
template<> EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
|
||||
double2 pexpm1<double2>(const double2& a)
|
||||
{
|
||||
return make_double2(expm1(a.x), expm1(a.y));
|
||||
}
|
||||
|
||||
template<> EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
|
||||
float4 psqrt<float4>(const float4& a)
|
||||
{
|
||||
return make_float4(sqrtf(a.x), sqrtf(a.y), sqrtf(a.z), sqrtf(a.w));
|
||||
}
|
||||
|
||||
template<> EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
|
||||
double2 psqrt<double2>(const double2& a)
|
||||
{
|
||||
using ::sqrt;
|
||||
return make_double2(sqrt(a.x), sqrt(a.y));
|
||||
}
|
||||
|
||||
template<> EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
|
||||
float4 prsqrt<float4>(const float4& a)
|
||||
{
|
||||
return make_float4(rsqrtf(a.x), rsqrtf(a.y), rsqrtf(a.z), rsqrtf(a.w));
|
||||
}
|
||||
|
||||
template<> EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
|
||||
double2 prsqrt<double2>(const double2& a)
|
||||
{
|
||||
return make_double2(rsqrt(a.x), rsqrt(a.y));
|
||||
}
|
||||
|
||||
|
||||
#endif
|
||||
|
||||
} // end namespace internal
|
||||
|
||||
} // end namespace Eigen
|
||||
|
||||
#endif // EIGEN_MATH_FUNCTIONS_GPU_H
|
||||
File diff suppressed because it is too large
Load Diff
Some files were not shown because too many files have changed in this diff Show More
Reference in New Issue
Block a user