Skip to content
Hide whitespace changes
Inline Side-by-side
external/eigen3/Eigen/src/Cholesky/LLT.h
View file @ a394b22a
......@@ -10,7 +10,7 @@
#ifndef EIGEN_LLT_H
#define EIGEN_LLT_H
namespace Eigen {
namespace Eigen {
namespace internal{
template<typename MatrixType, int UpLo> struct LLT_Traits;
......@@ -22,8 +22,8 @@ template<typename MatrixType, int UpLo> struct LLT_Traits;
*
* \brief Standard Cholesky decomposition (LL^T) of a matrix and associated features
*
* \param MatrixType the type of the matrix of which we are computing the LL^T Cholesky decomposition
* \param UpLo the triangular part that will be used for the decompositon: Lower (default) or Upper.
* \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
......@@ -40,8 +40,10 @@ template<typename MatrixType, int UpLo> struct LLT_Traits;
*
* Example: \include LLT_example.cpp
* Output: \verbinclude LLT_example.out
*
* \sa MatrixBase::llt(), class LDLT
*
* This class supports the \link InplaceDecomposition inplace decomposition \endlink mechanism.
*
* \sa MatrixBase::llt(), SelfAdjointView::llt(), class LDLT
*/
/* HEY THIS DOX IS DISABLED BECAUSE THERE's A BUG EITHER HERE OR IN LDLT ABOUT THAT (OR BOTH)
* Note that during the decomposition, only the upper triangular part of A is considered. Therefore,
......@@ -54,12 +56,12 @@ template<typename _MatrixType, int _UpLo> class LLT
enum {
RowsAtCompileTime = MatrixType::RowsAtCompileTime,
ColsAtCompileTime = MatrixType::ColsAtCompileTime,
Options = MatrixType::Options,
MaxColsAtCompileTime = MatrixType::MaxColsAtCompileTime
};
typedef typename MatrixType::Scalar Scalar;
typedef typename NumTraits<typename MatrixType::Scalar>::Real RealScalar;
typedef typename MatrixType::Index Index;
typedef Eigen::Index Index; ///< \deprecated since Eigen 3.3
typedef typename MatrixType::StorageIndex StorageIndex;
enum {
PacketSize = internal::packet_traits<Scalar>::size,
......@@ -83,14 +85,30 @@ template<typename _MatrixType, int _UpLo> class LLT
* according to the specified problem \a size.
* \sa LLT()
*/
LLT(Index size) : m_matrix(size, size),
explicit LLT(Index size) : m_matrix(size, size),
m_isInitialized(false) {}
LLT(const MatrixType& matrix)
template<typename InputType>
explicit LLT(const EigenBase<InputType>& matrix)
: m_matrix(matrix.rows(), matrix.cols()),
m_isInitialized(false)
{
compute(matrix);
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 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 */
......@@ -115,33 +133,33 @@ template<typename _MatrixType, int _UpLo> class LLT
* Example: \include LLT_solve.cpp
* Output: \verbinclude LLT_solve.out
*
* \sa solveInPlace(), MatrixBase::llt()
* \sa solveInPlace(), MatrixBase::llt(), SelfAdjointView::llt()
*/
template<typename Rhs>
inline const internal::solve_retval<LLT, Rhs>
inline const Solve<LLT, Rhs>
solve(const MatrixBase<Rhs>& b) const
{
eigen_assert(m_isInitialized && "LLT is not initialized.");
eigen_assert(m_matrix.rows()==b.rows()
&& "LLT::solve(): invalid number of rows of the right hand side matrix b");
return internal::solve_retval<LLT, Rhs>(*this, b.derived());
return Solve<LLT, Rhs>(*this, b.derived());
}
#ifdef EIGEN2_SUPPORT
template<typename OtherDerived, typename ResultType>
bool solve(const MatrixBase<OtherDerived>& b, ResultType *result) const
{
*result = this->solve(b);
return true;
}
bool isPositiveDefinite() const { return true; }
#endif
template<typename Derived>
void solveInPlace(MatrixBase<Derived> &bAndX) const;
LLT& compute(const MatrixType& matrix);
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
*
......@@ -167,24 +185,38 @@ template<typename _MatrixType, int _UpLo> class LLT
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 { return *this; };
inline Index rows() const { return m_matrix.rows(); }
inline Index cols() const { 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>
EIGEN_DEVICE_FUNC
void _solve_impl(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;
};
......@@ -194,12 +226,11 @@ namespace internal {
template<typename Scalar, int UpLo> struct llt_inplace;
template<typename MatrixType, typename VectorType>
static typename MatrixType::Index llt_rank_update_lower(MatrixType& mat, const VectorType& vec, const typename MatrixType::RealScalar& sigma)
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::Index Index;
typedef typename MatrixType::ColXpr ColXpr;
typedef typename internal::remove_all<ColXpr>::type ColXprCleaned;
typedef typename ColXprCleaned::SegmentReturnType ColXprSegment;
......@@ -268,11 +299,10 @@ template<typename Scalar> struct llt_inplace<Scalar, Lower>
{
typedef typename NumTraits<Scalar>::Real RealScalar;
template<typename MatrixType>
static typename MatrixType::Index unblocked(MatrixType& mat)
static Index unblocked(MatrixType& mat)
{
using std::sqrt;
typedef typename MatrixType::Index Index;
eigen_assert(mat.rows()==mat.cols());
const Index size = mat.rows();
for(Index k = 0; k < size; ++k)
......@@ -295,9 +325,8 @@ template<typename Scalar> struct llt_inplace<Scalar, Lower>
}
template<typename MatrixType>
static typename MatrixType::Index blocked(MatrixType& m)
static Index blocked(MatrixType& m)
{
typedef typename MatrixType::Index Index;
eigen_assert(m.rows()==m.cols());
Index size = m.rows();
if(size<32)
......@@ -322,36 +351,36 @@ template<typename Scalar> struct llt_inplace<Scalar, Lower>
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,-1); // bottleneck
if(rs>0) A22.template selfadjointView<Lower>().rankUpdate(A21,typename NumTraits<RealScalar>::Literal(-1)); // bottleneck
}
return -1;
}
template<typename MatrixType, typename VectorType>
static typename MatrixType::Index rankUpdate(MatrixType& mat, const VectorType& vec, const RealScalar& sigma)
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 typename MatrixType::Index unblocked(MatrixType& mat)
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 typename MatrixType::Index blocked(MatrixType& mat)
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 typename MatrixType::Index rankUpdate(MatrixType& mat, const VectorType& vec, const RealScalar& sigma)
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);
......@@ -362,8 +391,8 @@ 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 m; }
static inline MatrixU getU(const MatrixType& m) { return m.adjoint(); }
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; }
};
......@@ -372,8 +401,8 @@ 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 m.adjoint(); }
static inline MatrixU getU(const MatrixType& m) { return m; }
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; }
};
......@@ -388,14 +417,28 @@ template<typename MatrixType> struct LLT_Traits<MatrixType,Upper>
* Output: \verbinclude TutorialLinAlgComputeTwice.out
*/
template<typename MatrixType, int _UpLo>
LLT<MatrixType,_UpLo>& LLT<MatrixType,_UpLo>::compute(const MatrixType& a)
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);
m_matrix = a;
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);
......@@ -423,33 +466,24 @@ LLT<_MatrixType,_UpLo> LLT<_MatrixType,_UpLo>::rankUpdate(const VectorType& v, c
return *this;
}
namespace internal {
template<typename _MatrixType, int UpLo, typename Rhs>
struct solve_retval<LLT<_MatrixType, UpLo>, Rhs>
: solve_retval_base<LLT<_MatrixType, UpLo>, Rhs>
{
typedef LLT<_MatrixType,UpLo> LLTType;
EIGEN_MAKE_SOLVE_HELPERS(LLTType,Rhs)
template<typename Dest> void evalTo(Dest& dst) const
{
dst = rhs();
dec().solveInPlace(dst);
}
};
#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
{
dst = rhs;
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.
*
* \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.
* This version avoids a copy when the right hand side matrix b is not needed anymore.
*
* \sa LLT::solve(), MatrixBase::llt()
*/
......@@ -475,6 +509,7 @@ MatrixType LLT<MatrixType,_UpLo>::reconstructedMatrix() const
/** \cholesky_module
* \returns the LLT decomposition of \c *this
* \sa SelfAdjointView::llt()
*/
template<typename Derived>
inline const LLT<typename MatrixBase<Derived>::PlainObject>
......@@ -485,6 +520,7 @@ MatrixBase<Derived>::llt() const
/** \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>
......
external/eigen3/Eigen/src/Cholesky/LLT_MKL.h → external/eigen3/Eigen/src/Cholesky/LLT_LAPACKE.h
View file @ a394b22a
......@@ -25,41 +25,38 @@
SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
********************************************************************************
* Content : Eigen bindings to Intel(R) MKL
* Content : Eigen bindings to LAPACKe
* LLt decomposition based on LAPACKE_?potrf function.
********************************************************************************
*/
#ifndef EIGEN_LLT_MKL_H
#define EIGEN_LLT_MKL_H
#include "Eigen/src/Core/util/MKL_support.h"
#include <iostream>
#ifndef EIGEN_LLT_LAPACKE_H
#define EIGEN_LLT_LAPACKE_H
namespace Eigen {
namespace internal {
template<typename Scalar> struct mkl_llt;
template<typename Scalar> struct lapacke_llt;
#define EIGEN_MKL_LLT(EIGTYPE, MKLTYPE, MKLPREFIX) \
template<> struct mkl_llt<EIGTYPE> \
#define EIGEN_LAPACKE_LLT(EIGTYPE, BLASTYPE, LAPACKE_PREFIX) \
template<> struct lapacke_llt<EIGTYPE> \
{ \
template<typename MatrixType> \
static inline typename MatrixType::Index potrf(MatrixType& m, char uplo) \
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 = m.rows(); \
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 = m.outerStride(); \
lda = convert_index<lapack_int>(m.outerStride()); \
\
info = LAPACKE_##MKLPREFIX##potrf( matrix_order, uplo, size, (MKLTYPE*)a, lda ); \
info = LAPACKE_##LAPACKE_PREFIX##potrf( matrix_order, uplo, size, (BLASTYPE*)a, lda ); \
info = (info==0) ? -1 : info>0 ? info-1 : size; \
return info; \
} \
......@@ -67,36 +64,36 @@ template<> struct mkl_llt<EIGTYPE> \
template<> struct llt_inplace<EIGTYPE, Lower> \
{ \
template<typename MatrixType> \
static typename MatrixType::Index blocked(MatrixType& m) \
static Index blocked(MatrixType& m) \
{ \
return mkl_llt<EIGTYPE>::potrf(m, 'L'); \
return lapacke_llt<EIGTYPE>::potrf(m, 'L'); \
} \
template<typename MatrixType, typename VectorType> \
static typename MatrixType::Index rankUpdate(MatrixType& mat, const VectorType& vec, const typename MatrixType::RealScalar& sigma) \
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 typename MatrixType::Index blocked(MatrixType& m) \
static Index blocked(MatrixType& m) \
{ \
return mkl_llt<EIGTYPE>::potrf(m, 'U'); \
return lapacke_llt<EIGTYPE>::potrf(m, 'U'); \
} \
template<typename MatrixType, typename VectorType> \
static typename MatrixType::Index rankUpdate(MatrixType& mat, const VectorType& vec, const typename MatrixType::RealScalar& sigma) \
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_MKL_LLT(double, double, d)
EIGEN_MKL_LLT(float, float, s)
EIGEN_MKL_LLT(dcomplex, MKL_Complex16, z)
EIGEN_MKL_LLT(scomplex, MKL_Complex8, c)
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_MKL_H
#endif // EIGEN_LLT_LAPACKE_H
external/eigen3/Eigen/src/CholmodSupport/CMakeLists.txt deleted 100644 → 0
View file @ 701c0225
FILE(GLOB Eigen_CholmodSupport_SRCS "*.h")
INSTALL(FILES
${Eigen_CholmodSupport_SRCS}
DESTINATION ${INCLUDE_INSTALL_DIR}/Eigen/src/CholmodSupport COMPONENT Devel
)
external/eigen3/Eigen/src/CholmodSupport/CholmodSupport.h
View file @ a394b22a
......@@ -14,46 +14,52 @@ namespace Eigen {
namespace internal {
template<typename Scalar, typename CholmodType>
void cholmod_configure_matrix(CholmodType& mat)
{
if (internal::is_same<Scalar,float>::value)
{
mat.xtype = CHOLMOD_REAL;
mat.dtype = CHOLMOD_SINGLE;
}
else if (internal::is_same<Scalar,double>::value)
{
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;
}
else if (internal::is_same<Scalar,std::complex<float> >::value)
{
mat.xtype = CHOLMOD_COMPLEX;
mat.dtype = CHOLMOD_SINGLE;
}
else if (internal::is_same<Scalar,std::complex<double> >::value)
{
};
template<> struct cholmod_configure_matrix<std::complex<double> > {
template<typename CholmodType>
static void run(CholmodType& mat) {
mat.xtype = CHOLMOD_COMPLEX;
mat.dtype = CHOLMOD_DOUBLE;
}
else
{
eigen_assert(false && "Scalar type not supported by CHOLMOD");
}
}
};
// Other scalar types are not yet suppotred 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 _Index>
cholmod_sparse viewAsCholmod(SparseMatrix<_Scalar,_Options,_Index>& mat)
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.nrow = mat.rows();
res.ncol = mat.cols();
res.p = mat.outerIndexPtr();
res.i = mat.innerIndexPtr();
......@@ -74,11 +80,11 @@ cholmod_sparse viewAsCholmod(SparseMatrix<_Scalar,_Options,_Index>& mat)
res.dtype = 0;
res.stype = -1;
if (internal::is_same<_Index,int>::value)
if (internal::is_same<_StorageIndex,int>::value)
{
res.itype = CHOLMOD_INT;
}
else if (internal::is_same<_Index,SuiteSparse_long>::value)
else if (internal::is_same<_StorageIndex,long>::value)
{
res.itype = CHOLMOD_LONG;
}
......@@ -88,7 +94,7 @@ cholmod_sparse viewAsCholmod(SparseMatrix<_Scalar,_Options,_Index>& mat)
}
// setup res.xtype
internal::cholmod_configure_matrix<_Scalar>(res);
internal::cholmod_configure_matrix<_Scalar>::run(res);
res.stype = 0;
......@@ -98,16 +104,23 @@ cholmod_sparse viewAsCholmod(SparseMatrix<_Scalar,_Options,_Index>& mat)
template<typename _Scalar, int _Options, typename _Index>
const cholmod_sparse viewAsCholmod(const SparseMatrix<_Scalar,_Options,_Index>& mat)
{
cholmod_sparse res = viewAsCholmod(mat.const_cast_derived());
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<SparseMatrix<_Scalar,_Options,_Index>, UpLo>& mat)
cholmod_sparse viewAsCholmod(const SparseSelfAdjointView<const SparseMatrix<_Scalar,_Options,_Index>, UpLo>& mat)
{
cholmod_sparse res = viewAsCholmod(mat.matrix().const_cast_derived());
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;
......@@ -131,19 +144,19 @@ cholmod_dense viewAsCholmod(MatrixBase<Derived>& mat)
res.x = (void*)(mat.derived().data());
res.z = 0;
internal::cholmod_configure_matrix<Scalar>(res);
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 Index>
MappedSparseMatrix<Scalar,Flags,Index> viewAsEigen(cholmod_sparse& cm)
template<typename Scalar, int Flags, typename StorageIndex>
MappedSparseMatrix<Scalar,Flags,StorageIndex> viewAsEigen(cholmod_sparse& cm)
{
return MappedSparseMatrix<Scalar,Flags,Index>
(cm.nrow, cm.ncol, static_cast<Index*>(cm.p)[cm.ncol],
static_cast<Index*>(cm.p), static_cast<Index*>(cm.i),static_cast<Scalar*>(cm.x) );
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) );
}
enum CholmodMode {
......@@ -157,29 +170,39 @@ enum CholmodMode {
* \sa class CholmodSupernodalLLT, class CholmodSimplicialLDLT, class CholmodSimplicialLLT
*/
template<typename _MatrixType, int _UpLo, typename Derived>
class CholmodBase : internal::noncopyable
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::Index Index;
typedef typename MatrixType::StorageIndex StorageIndex;
enum {
ColsAtCompileTime = MatrixType::ColsAtCompileTime,
MaxColsAtCompileTime = MatrixType::MaxColsAtCompileTime
};
public:
CholmodBase()
: m_cholmodFactor(0), m_info(Success), m_isInitialized(false)
: m_cholmodFactor(0), m_info(Success), m_factorizationIsOk(false), m_analysisIsOk(false)
{
m_shiftOffset[0] = m_shiftOffset[1] = RealScalar(0.0);
EIGEN_STATIC_ASSERT((internal::is_same<double,RealScalar>::value), CHOLMOD_SUPPORTS_DOUBLE_PRECISION_ONLY);
m_shiftOffset[0] = m_shiftOffset[1] = 0.0;
cholmod_start(&m_cholmod);
}
CholmodBase(const MatrixType& matrix)
: m_cholmodFactor(0), m_info(Success), m_isInitialized(false)
explicit CholmodBase(const MatrixType& matrix)
: m_cholmodFactor(0), m_info(Success), m_factorizationIsOk(false), m_analysisIsOk(false)
{
m_shiftOffset[0] = m_shiftOffset[1] = RealScalar(0.0);
EIGEN_STATIC_ASSERT((internal::is_same<double,RealScalar>::value), CHOLMOD_SUPPORTS_DOUBLE_PRECISION_ONLY);
m_shiftOffset[0] = m_shiftOffset[1] = 0.0;
cholmod_start(&m_cholmod);
compute(matrix);
}
......@@ -191,11 +214,8 @@ class CholmodBase : internal::noncopyable
cholmod_finish(&m_cholmod);
}
inline Index cols() const { return m_cholmodFactor->n; }
inline Index rows() const { return m_cholmodFactor->n; }
Derived& derived() { return *static_cast<Derived*>(this); }
const Derived& derived() const { return *static_cast<const Derived*>(this); }
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.
*
......@@ -216,34 +236,6 @@ class CholmodBase : internal::noncopyable
return derived();
}
/** \returns the solution x of \f$ A x = b \f$ using the current decomposition of A.
*
* \sa compute()
*/
template<typename Rhs>
inline const internal::solve_retval<CholmodBase, Rhs>
solve(const MatrixBase<Rhs>& b) const
{
eigen_assert(m_isInitialized && "LLT is not initialized.");
eigen_assert(rows()==b.rows()
&& "CholmodDecomposition::solve(): invalid number of rows of the right hand side matrix b");
return internal::solve_retval<CholmodBase, Rhs>(*this, b.derived());
}
/** \returns the solution x of \f$ A x = b \f$ using the current decomposition of A.
*
* \sa compute()
*/
template<typename Rhs>
inline const internal::sparse_solve_retval<CholmodBase, Rhs>
solve(const SparseMatrixBase<Rhs>& b) const
{
eigen_assert(m_isInitialized && "LLT is not initialized.");
eigen_assert(rows()==b.rows()
&& "CholmodDecomposition::solve(): invalid number of rows of the right hand side matrix b");
return internal::sparse_solve_retval<CholmodBase, Rhs>(*this, b.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.
......@@ -277,7 +269,7 @@ class CholmodBase : internal::noncopyable
eigen_assert(m_analysisIsOk && "You must first call analyzePattern()");
cholmod_sparse A = viewAsCholmod(matrix.template selfadjointView<UpLo>());
cholmod_factorize_p(&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;
......@@ -290,20 +282,22 @@ class CholmodBase : internal::noncopyable
#ifndef EIGEN_PARSED_BY_DOXYGEN
/** \internal */
template<typename Rhs,typename Dest>
void _solve(const MatrixBase<Rhs> &b, MatrixBase<Dest> &dest) const
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 stoarge without inner-stride, which corresponds to the default behavior of Ref.
Ref<const Matrix<typename Rhs::Scalar,Dynamic,Dynamic,ColMajor> > b_ref(b.derived());
// note: cd stands for Cholmod Dense
Rhs& b_ref(b.const_cast_derived());
cholmod_dense b_cd = viewAsCholmod(b_ref);
cholmod_dense* x_cd = cholmod_solve(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.)
dest = Matrix<Scalar,Dest::RowsAtCompileTime,Dest::ColsAtCompileTime>::Map(reinterpret_cast<Scalar*>(x_cd->x),b.rows(),b.cols());
......@@ -311,8 +305,8 @@ class CholmodBase : internal::noncopyable
}
/** \internal */
template<typename RhsScalar, int RhsOptions, typename RhsIndex, typename DestScalar, int DestOptions, typename DestIndex>
void _solve(const SparseMatrix<RhsScalar,RhsOptions,RhsIndex> &b, SparseMatrix<DestScalar,DestOptions,DestIndex> &dest) const
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;
......@@ -320,14 +314,16 @@ class CholmodBase : internal::noncopyable
eigen_assert(size==b.rows());
// note: cs stands for Cholmod Sparse
cholmod_sparse b_cs = viewAsCholmod(b);
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 = cholmod_spsolve(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.)
dest = viewAsEigen<DestScalar,DestOptions,DestIndex>(*x_cs);
dest.derived() = viewAsEigen<typename DestDerived::Scalar,ColMajor,typename DestDerived::StorageIndex>(*x_cs);
cholmod_free_sparse(&x_cs, &m_cholmod);
}
#endif // EIGEN_PARSED_BY_DOXYGEN
......@@ -344,10 +340,61 @@ class CholmodBase : internal::noncopyable
*/
Derived& setShift(const RealScalar& offset)
{
m_shiftOffset[0] = 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*/)
{}
......@@ -355,9 +402,8 @@ class CholmodBase : internal::noncopyable
protected:
mutable cholmod_common m_cholmod;
cholmod_factor* m_cholmodFactor;
RealScalar m_shiftOffset[2];
double m_shiftOffset[2];
mutable ComputationInfo m_info;
bool m_isInitialized;
int m_factorizationIsOk;
int m_analysisIsOk;
};
......@@ -376,9 +422,13 @@ class CholmodBase : internal::noncopyable
* \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.
*
* \sa \ref TutorialSparseDirectSolvers, class CholmodSupernodalLLT, class SimplicialLLT
* \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> >
......@@ -395,7 +445,7 @@ class CholmodSimplicialLLT : public CholmodBase<_MatrixType, _UpLo, CholmodSimpl
CholmodSimplicialLLT(const MatrixType& matrix) : Base()
{
init();
Base::compute(matrix);
this->compute(matrix);
}
~CholmodSimplicialLLT() {}
......@@ -423,9 +473,13 @@ class CholmodSimplicialLLT : public CholmodBase<_MatrixType, _UpLo, CholmodSimpl
* \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.
*
* \sa \ref TutorialSparseDirectSolvers, class CholmodSupernodalLLT, class SimplicialLDLT
* \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> >
......@@ -442,7 +496,7 @@ class CholmodSimplicialLDLT : public CholmodBase<_MatrixType, _UpLo, CholmodSimp
CholmodSimplicialLDLT(const MatrixType& matrix) : Base()
{
init();
Base::compute(matrix);
this->compute(matrix);
}
~CholmodSimplicialLDLT() {}
......@@ -468,9 +522,13 @@ class CholmodSimplicialLDLT : public CholmodBase<_MatrixType, _UpLo, CholmodSimp
* \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.
*
* \sa \ref TutorialSparseDirectSolvers
* \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> >
......@@ -487,7 +545,7 @@ class CholmodSupernodalLLT : public CholmodBase<_MatrixType, _UpLo, CholmodSuper
CholmodSupernodalLLT(const MatrixType& matrix) : Base()
{
init();
Base::compute(matrix);
this->compute(matrix);
}
~CholmodSupernodalLLT() {}
......@@ -515,9 +573,13 @@ class CholmodSupernodalLLT : public CholmodBase<_MatrixType, _UpLo, CholmodSuper
* \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.
*
* \sa \ref TutorialSparseDirectSolvers
* \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> >
......@@ -534,7 +596,7 @@ class CholmodDecomposition : public CholmodBase<_MatrixType, _UpLo, CholmodDecom
CholmodDecomposition(const MatrixType& matrix) : Base()
{
init();
Base::compute(matrix);
this->compute(matrix);
}
~CholmodDecomposition() {}
......@@ -572,36 +634,6 @@ class CholmodDecomposition : public CholmodBase<_MatrixType, _UpLo, CholmodDecom
}
};
namespace internal {
template<typename _MatrixType, int _UpLo, typename Derived, typename Rhs>
struct solve_retval<CholmodBase<_MatrixType,_UpLo,Derived>, Rhs>
: solve_retval_base<CholmodBase<_MatrixType,_UpLo,Derived>, Rhs>
{
typedef CholmodBase<_MatrixType,_UpLo,Derived> Dec;
EIGEN_MAKE_SOLVE_HELPERS(Dec,Rhs)
template<typename Dest> void evalTo(Dest& dst) const
{
dec()._solve(rhs(),dst);
}
};
template<typename _MatrixType, int _UpLo, typename Derived, typename Rhs>
struct sparse_solve_retval<CholmodBase<_MatrixType,_UpLo,Derived>, Rhs>
: sparse_solve_retval_base<CholmodBase<_MatrixType,_UpLo,Derived>, Rhs>
{
typedef CholmodBase<_MatrixType,_UpLo,Derived> Dec;
EIGEN_MAKE_SPARSE_SOLVE_HELPERS(Dec,Rhs)
template<typename Dest> void evalTo(Dest& dst) const
{
dec()._solve(rhs(),dst);
}
};
} // end namespace internal
} // end namespace Eigen
#endif // EIGEN_CHOLMODSUPPORT_H
external/eigen3/Eigen/src/Core/Array.h
View file @ a394b22a
......@@ -12,7 +12,16 @@
namespace Eigen {
/** \class Array
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
......@@ -24,20 +33,14 @@ namespace Eigen {
* 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 TopicCustomizingEigen by defining the preprocessor symbol \c EIGEN_ARRAY_PLUGIN.
* \ref TopicCustomizing_Plugins by defining the preprocessor symbol \c EIGEN_ARRAY_PLUGIN.
*
* \sa \ref TutorialArrayClass, \ref TopicClassHierarchy
* \sa \blank \ref TutorialArrayClass, \ref TopicClassHierarchy
*/
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;
};
}
template<typename _Scalar, int _Rows, int _Cols, int _Options, int _MaxRows, int _MaxCols>
class Array
: public PlainObjectBase<Array<_Scalar, _Rows, _Cols, _Options, _MaxRows, _MaxCols> >
......@@ -69,11 +72,27 @@ class Array
* 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),
......@@ -84,7 +103,8 @@ class Array
* remain row-vectors and vectors remain vectors.
*/
template<typename OtherDerived>
EIGEN_STRONG_INLINE Array& operator=(const ArrayBase<OtherDerived>& other)
EIGEN_DEVICE_FUNC
EIGEN_STRONG_INLINE Array& operator=(const DenseBase<OtherDerived>& other)
{
return Base::_set(other);
}
......@@ -92,11 +112,12 @@ class Array
/** 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.
......@@ -107,6 +128,7 @@ class Array
*
* \sa resize(Index,Index)
*/
EIGEN_DEVICE_FUNC
EIGEN_STRONG_INLINE Array() : Base()
{
Base::_check_template_params();
......@@ -116,6 +138,7 @@ class Array
#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())
{
......@@ -124,56 +147,64 @@ class Array
}
#endif
#ifdef EIGEN_HAVE_RVALUE_REFERENCES
Array(Array&& other)
#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();
if (RowsAtCompileTime!=Dynamic && ColsAtCompileTime!=Dynamic)
Base::_set_noalias(other);
}
Array& operator=(Array&& other)
EIGEN_DEVICE_FUNC
Array& operator=(Array&& other) EIGEN_NOEXCEPT_IF(std::is_nothrow_move_assignable<Scalar>::value)
{
other.swap(*this);
return *this;
}
#endif
/** 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 Matrix() instead.
*/
EIGEN_STRONG_INLINE explicit Array(Index dim)
: Base(dim, RowsAtCompileTime == 1 ? 1 : dim, ColsAtCompileTime == 1 ? 1 : dim)
#ifndef EIGEN_PARSED_BY_DOXYGEN
template<typename T>
EIGEN_DEVICE_FUNC
EIGEN_STRONG_INLINE explicit Array(const T& x)
{
Base::_check_template_params();
EIGEN_STATIC_ASSERT_VECTOR_ONLY(Array)
eigen_assert(dim >= 0);
eigen_assert(SizeAtCompileTime == Dynamic || SizeAtCompileTime == dim);
EIGEN_INITIALIZE_COEFFS_IF_THAT_OPTION_IS_ENABLED
Base::template _init1<T>(x);
}
#ifndef EIGEN_PARSED_BY_DOXYGEN
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
/** constructs an uninitialized matrix with \a rows rows and \a cols columns.
/** \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 */
Array(const Scalar& value);
/** constructs an uninitialized array with \a rows rows and \a cols columns.
*
* This is useful for dynamic-size matrices. For fixed-size matrices,
* 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
* Matrix() instead. */
* Array() instead. */
Array(Index rows, Index cols);
/** constructs an initialized 2D vector with given coefficients */
Array(const Scalar& val0, const Scalar& val1);
#endif
/** constructs an initialized 3D vector with given coefficients */
EIGEN_DEVICE_FUNC
EIGEN_STRONG_INLINE Array(const Scalar& val0, const Scalar& val1, const Scalar& val2)
{
Base::_check_template_params();
......@@ -183,6 +214,7 @@ class Array
m_storage.data()[2] = val2;
}
/** constructs an initialized 4D vector with given coefficients */
EIGEN_DEVICE_FUNC
EIGEN_STRONG_INLINE Array(const Scalar& val0, const Scalar& val1, const Scalar& val2, const Scalar& val3)
{
Base::_check_template_params();
......@@ -193,51 +225,27 @@ class Array
m_storage.data()[3] = val3;
}
explicit Array(const Scalar *data);
/** Constructor copying the value of the expression \a other */
template<typename OtherDerived>
EIGEN_STRONG_INLINE Array(const ArrayBase<OtherDerived>& other)
: Base(other.rows() * other.cols(), other.rows(), other.cols())
{
Base::_check_template_params();
Base::_set_noalias(other);
}
/** Copy constructor */
EIGEN_DEVICE_FUNC
EIGEN_STRONG_INLINE Array(const Array& other)
: Base(other.rows() * other.cols(), other.rows(), other.cols())
{
Base::_check_template_params();
Base::_set_noalias(other);
}
/** Copy constructor with in-place evaluation */
template<typename OtherDerived>
EIGEN_STRONG_INLINE Array(const ReturnByValue<OtherDerived>& other)
{
Base::_check_template_params();
Base::resize(other.rows(), other.cols());
other.evalTo(*this);
}
: Base(other)
{ }
/** \sa MatrixBase::operator=(const EigenBase<OtherDerived>&) */
template<typename OtherDerived>
EIGEN_STRONG_INLINE Array(const EigenBase<OtherDerived> &other)
: Base(other.derived().rows() * other.derived().cols(), other.derived().rows(), other.derived().cols())
{
Base::_check_template_params();
Base::_resize_to_match(other);
*this = other;
}
private:
struct PrivateType {};
public:
/** Override MatrixBase::swap() since for dynamic-sized matrices of same type it is enough to swap the
* data pointers.
*/
/** \sa MatrixBase::operator=(const EigenBase<OtherDerived>&) */
template<typename OtherDerived>
void swap(ArrayBase<OtherDerived> const & other)
{ this->_swap(other.derived()); }
inline Index innerStride() const { return 1; }
inline Index outerStride() const { return this->innerSize(); }
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 inline Index innerStride() const { return 1; }
EIGEN_DEVICE_FUNC inline Index outerStride() const { return this->innerSize(); }
#ifdef EIGEN_ARRAY_PLUGIN
#include EIGEN_ARRAY_PLUGIN
......
external/eigen3/Eigen/src/Core/ArrayBase.h
View file @ a394b22a
......@@ -32,7 +32,7 @@ template<typename ExpressionType> class MatrixWrapper;
* \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 TopicCustomizingEigen by defining the preprocessor symbol \c EIGEN_ARRAYBASE_PLUGIN.
* \ref TopicCustomizing_Plugins by defining the preprocessor symbol \c EIGEN_ARRAYBASE_PLUGIN.
*
* \sa class MatrixBase, \ref TopicClassHierarchy
*/
......@@ -47,13 +47,11 @@ template<typename Derived> class ArrayBase
typedef ArrayBase Eigen_BaseClassForSpecializationOfGlobalMathFuncImpl;
typedef typename internal::traits<Derived>::StorageKind StorageKind;
typedef typename internal::traits<Derived>::Index Index;
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::operator*;
using Base::RowsAtCompileTime;
using Base::ColsAtCompileTime;
using Base::SizeAtCompileTime;
......@@ -62,8 +60,7 @@ template<typename Derived> class ArrayBase
using Base::MaxSizeAtCompileTime;
using Base::IsVectorAtCompileTime;
using Base::Flags;
using Base::CoeffReadCost;
using Base::derived;
using Base::const_cast_derived;
using Base::rows;
......@@ -83,25 +80,14 @@ template<typename Derived> class ArrayBase
#endif // not EIGEN_PARSED_BY_DOXYGEN
#ifndef EIGEN_PARSED_BY_DOXYGEN
/** \internal the plain matrix type corresponding to this expression. Note that 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 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
> PlainObject;
typedef typename Base::PlainObject PlainObject;
/** \internal Represents a matrix with all coefficients equal to one another*/
typedef CwiseNullaryOp<internal::scalar_constant_op<Scalar>,Derived> ConstantReturnType;
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/CommonCwiseUnaryOps.h"
# include "../plugins/MatrixCwiseUnaryOps.h"
# include "../plugins/ArrayCwiseUnaryOps.h"
......@@ -112,44 +98,62 @@ template<typename Derived> class ArrayBase
# 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)
{
return internal::assign_selector<Derived,Derived>::run(derived(), other.derived());
internal::call_assignment(derived(), other.derived());
return derived();
}
Derived& operator+=(const Scalar& scalar)
{ return *this = derived() + scalar; }
Derived& operator-=(const Scalar& scalar)
{ return *this = derived() - scalar; }
/** 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() */
MatrixWrapper<Derived> matrix() { return derived(); }
const MatrixWrapper<const Derived> matrix() const { return derived(); }
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_DEVICE_FUNC
ArrayBase() : Base() {}
private:
......@@ -171,11 +175,10 @@ template<typename Derived> class ArrayBase
*/
template<typename Derived>
template<typename OtherDerived>
EIGEN_STRONG_INLINE Derived &
EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE Derived &
ArrayBase<Derived>::operator-=(const ArrayBase<OtherDerived> &other)
{
SelfCwiseBinaryOp<internal::scalar_difference_op<Scalar>, Derived, OtherDerived> tmp(derived());
tmp = other.derived();
call_assignment(derived(), other.derived(), internal::sub_assign_op<Scalar,typename OtherDerived::Scalar>());
return derived();
}
......@@ -185,11 +188,10 @@ ArrayBase<Derived>::operator-=(const ArrayBase<OtherDerived> &other)
*/
template<typename Derived>
template<typename OtherDerived>
EIGEN_STRONG_INLINE Derived &
EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE Derived &
ArrayBase<Derived>::operator+=(const ArrayBase<OtherDerived>& other)
{
SelfCwiseBinaryOp<internal::scalar_sum_op<Scalar>, Derived, OtherDerived> tmp(derived());
tmp = other.derived();
call_assignment(derived(), other.derived(), internal::add_assign_op<Scalar,typename OtherDerived::Scalar>());
return derived();
}
......@@ -199,11 +201,10 @@ ArrayBase<Derived>::operator+=(const ArrayBase<OtherDerived>& other)
*/
template<typename Derived>
template<typename OtherDerived>
EIGEN_STRONG_INLINE Derived &
EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE Derived &
ArrayBase<Derived>::operator*=(const ArrayBase<OtherDerived>& other)
{
SelfCwiseBinaryOp<internal::scalar_product_op<Scalar>, Derived, OtherDerived> tmp(derived());
tmp = other.derived();
call_assignment(derived(), other.derived(), internal::mul_assign_op<Scalar,typename OtherDerived::Scalar>());
return derived();
}
......@@ -213,11 +214,10 @@ ArrayBase<Derived>::operator*=(const ArrayBase<OtherDerived>& other)
*/
template<typename Derived>
template<typename OtherDerived>
EIGEN_STRONG_INLINE Derived &
EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE Derived &
ArrayBase<Derived>::operator/=(const ArrayBase<OtherDerived>& other)
{
SelfCwiseBinaryOp<internal::scalar_quotient_op<Scalar>, Derived, OtherDerived> tmp(derived());
tmp = other.derived();
call_assignment(derived(), other.derived(), internal::div_assign_op<Scalar,typename OtherDerived::Scalar>());
return derived();
}
......
external/eigen3/Eigen/src/Core/ArrayWrapper.h
View file @ a394b22a
......@@ -32,7 +32,8 @@ struct traits<ArrayWrapper<ExpressionType> >
// Let's remove NestByRefBit
enum {
Flags0 = traits<typename remove_all<typename ExpressionType::Nested>::type >::Flags,
Flags = Flags0 & ~NestByRefBit
LvalueBitFlag = is_lvalue<ExpressionType>::value ? LvalueBit : 0,
Flags = (Flags0 & ~(NestByRefBit | LvalueBit)) | LvalueBitFlag
};
};
}
......@@ -44,6 +45,7 @@ class ArrayWrapper : public ArrayBase<ArrayWrapper<ExpressionType> >
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,
......@@ -51,76 +53,45 @@ class ArrayWrapper : public ArrayBase<ArrayWrapper<ExpressionType> >
const Scalar
>::type ScalarWithConstIfNotLvalue;
typedef typename internal::nested<ExpressionType>::type NestedExpressionType;
typedef typename internal::ref_selector<ExpressionType>::non_const_type NestedExpressionType;
inline ArrayWrapper(ExpressionType& matrix) : m_expression(matrix) {}
using Base::coeffRef;
EIGEN_DEVICE_FUNC
explicit EIGEN_STRONG_INLINE ArrayWrapper(ExpressionType& matrix) : m_expression(matrix) {}
EIGEN_DEVICE_FUNC
inline Index rows() const { return m_expression.rows(); }
EIGEN_DEVICE_FUNC
inline Index cols() const { return m_expression.cols(); }
EIGEN_DEVICE_FUNC
inline Index outerStride() const { return m_expression.outerStride(); }
EIGEN_DEVICE_FUNC
inline Index innerStride() const { return m_expression.innerStride(); }
inline ScalarWithConstIfNotLvalue* data() { return m_expression.const_cast_derived().data(); }
EIGEN_DEVICE_FUNC
inline ScalarWithConstIfNotLvalue* data() { return m_expression.data(); }
EIGEN_DEVICE_FUNC
inline const Scalar* data() const { return m_expression.data(); }
inline CoeffReturnType coeff(Index rowId, Index colId) const
{
return m_expression.coeff(rowId, colId);
}
inline Scalar& coeffRef(Index rowId, Index colId)
{
return m_expression.const_cast_derived().coeffRef(rowId, colId);
}
EIGEN_DEVICE_FUNC
inline const Scalar& coeffRef(Index rowId, Index colId) const
{
return m_expression.const_cast_derived().coeffRef(rowId, colId);
}
inline CoeffReturnType coeff(Index index) const
{
return m_expression.coeff(index);
}
inline Scalar& coeffRef(Index index)
{
return m_expression.const_cast_derived().coeffRef(index);
return m_expression.coeffRef(rowId, colId);
}
EIGEN_DEVICE_FUNC
inline const Scalar& coeffRef(Index index) const
{
return m_expression.const_cast_derived().coeffRef(index);
}
template<int LoadMode>
inline const PacketScalar packet(Index rowId, Index colId) const
{
return m_expression.template packet<LoadMode>(rowId, colId);
}
template<int LoadMode>
inline void writePacket(Index rowId, Index colId, const PacketScalar& val)
{
m_expression.const_cast_derived().template writePacket<LoadMode>(rowId, colId, val);
}
template<int LoadMode>
inline const PacketScalar packet(Index index) const
{
return m_expression.template packet<LoadMode>(index);
}
template<int LoadMode>
inline void writePacket(Index index, const PacketScalar& val)
{
m_expression.const_cast_derived().template writePacket<LoadMode>(index, val);
return m_expression.coeffRef(index);
}
template<typename Dest>
EIGEN_DEVICE_FUNC
inline void evalTo(Dest& dst) const { dst = m_expression; }
const typename internal::remove_all<NestedExpressionType>::type&
EIGEN_DEVICE_FUNC
nestedExpression() const
{
return m_expression;
......@@ -128,10 +99,12 @@ class ArrayWrapper : public ArrayBase<ArrayWrapper<ExpressionType> >
/** Forwards the resizing request to the nested expression
* \sa DenseBase::resize(Index) */
void resize(Index newSize) { m_expression.const_cast_derived().resize(newSize); }
EIGEN_DEVICE_FUNC
void resize(Index newSize) { m_expression.resize(newSize); }
/** Forwards the resizing request to the nested expression
* \sa DenseBase::resize(Index,Index)*/
void resize(Index nbRows, Index nbCols) { m_expression.const_cast_derived().resize(nbRows,nbCols); }
EIGEN_DEVICE_FUNC
void resize(Index rows, Index cols) { m_expression.resize(rows,cols); }
protected:
NestedExpressionType m_expression;
......@@ -157,7 +130,8 @@ struct traits<MatrixWrapper<ExpressionType> >
// Let's remove NestByRefBit
enum {
Flags0 = traits<typename remove_all<typename ExpressionType::Nested>::type >::Flags,
Flags = Flags0 & ~NestByRefBit
LvalueBitFlag = is_lvalue<ExpressionType>::value ? LvalueBit : 0,
Flags = (Flags0 & ~(NestByRefBit | LvalueBit)) | LvalueBitFlag
};
};
}
......@@ -169,6 +143,7 @@ class MatrixWrapper : public MatrixBase<MatrixWrapper<ExpressionType> >
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,
......@@ -176,72 +151,40 @@ class MatrixWrapper : public MatrixBase<MatrixWrapper<ExpressionType> >
const Scalar
>::type ScalarWithConstIfNotLvalue;
typedef typename internal::nested<ExpressionType>::type NestedExpressionType;
typedef typename internal::ref_selector<ExpressionType>::non_const_type NestedExpressionType;
inline MatrixWrapper(ExpressionType& a_matrix) : m_expression(a_matrix) {}
using Base::coeffRef;
EIGEN_DEVICE_FUNC
explicit inline MatrixWrapper(ExpressionType& matrix) : m_expression(matrix) {}
EIGEN_DEVICE_FUNC
inline Index rows() const { return m_expression.rows(); }
EIGEN_DEVICE_FUNC
inline Index cols() const { return m_expression.cols(); }
EIGEN_DEVICE_FUNC
inline Index outerStride() const { return m_expression.outerStride(); }
EIGEN_DEVICE_FUNC
inline Index innerStride() const { return m_expression.innerStride(); }
inline ScalarWithConstIfNotLvalue* data() { return m_expression.const_cast_derived().data(); }
EIGEN_DEVICE_FUNC
inline ScalarWithConstIfNotLvalue* data() { return m_expression.data(); }
EIGEN_DEVICE_FUNC
inline const Scalar* data() const { return m_expression.data(); }
inline CoeffReturnType coeff(Index rowId, Index colId) const
{
return m_expression.coeff(rowId, colId);
}
inline Scalar& coeffRef(Index rowId, Index colId)
{
return m_expression.const_cast_derived().coeffRef(rowId, colId);
}
EIGEN_DEVICE_FUNC
inline const Scalar& coeffRef(Index rowId, Index colId) const
{
return m_expression.derived().coeffRef(rowId, colId);
}
inline CoeffReturnType coeff(Index index) const
{
return m_expression.coeff(index);
}
inline Scalar& coeffRef(Index index)
{
return m_expression.const_cast_derived().coeffRef(index);
}
EIGEN_DEVICE_FUNC
inline const Scalar& coeffRef(Index index) const
{
return m_expression.const_cast_derived().coeffRef(index);
}
template<int LoadMode>
inline const PacketScalar packet(Index rowId, Index colId) const
{
return m_expression.template packet<LoadMode>(rowId, colId);
}
template<int LoadMode>
inline void writePacket(Index rowId, Index colId, const PacketScalar& val)
{
m_expression.const_cast_derived().template writePacket<LoadMode>(rowId, colId, val);
}
template<int LoadMode>
inline const PacketScalar packet(Index index) const
{
return m_expression.template packet<LoadMode>(index);
}
template<int LoadMode>
inline void writePacket(Index index, const PacketScalar& val)
{
m_expression.const_cast_derived().template writePacket<LoadMode>(index, val);
return m_expression.coeffRef(index);
}
EIGEN_DEVICE_FUNC
const typename internal::remove_all<NestedExpressionType>::type&
nestedExpression() const
{
......@@ -250,10 +193,12 @@ class MatrixWrapper : public MatrixBase<MatrixWrapper<ExpressionType> >
/** Forwards the resizing request to the nested expression
* \sa DenseBase::resize(Index) */
void resize(Index newSize) { m_expression.const_cast_derived().resize(newSize); }
EIGEN_DEVICE_FUNC
void resize(Index newSize) { m_expression.resize(newSize); }
/** Forwards the resizing request to the nested expression
* \sa DenseBase::resize(Index,Index)*/
void resize(Index nbRows, Index nbCols) { m_expression.const_cast_derived().resize(nbRows,nbCols); }
EIGEN_DEVICE_FUNC
void resize(Index rows, Index cols) { m_expression.resize(rows,cols); }
protected:
NestedExpressionType m_expression;
......
external/eigen3/Eigen/src/Core/Assign.h
View file @ a394b22a
......@@ -14,478 +14,6 @@
namespace Eigen {
namespace internal {
/***************************************************************************
* Part 1 : the logic deciding a strategy for traversal and unrolling *
***************************************************************************/
template <typename Derived, typename OtherDerived>
struct assign_traits
{
public:
enum {
DstIsAligned = Derived::Flags & AlignedBit,
DstHasDirectAccess = Derived::Flags & DirectAccessBit,
SrcIsAligned = OtherDerived::Flags & AlignedBit,
JointAlignment = bool(DstIsAligned) && bool(SrcIsAligned) ? Aligned : Unaligned
};
private:
enum {
InnerSize = int(Derived::IsVectorAtCompileTime) ? int(Derived::SizeAtCompileTime)
: int(Derived::Flags)&RowMajorBit ? int(Derived::ColsAtCompileTime)
: int(Derived::RowsAtCompileTime),
InnerMaxSize = int(Derived::IsVectorAtCompileTime) ? int(Derived::MaxSizeAtCompileTime)
: int(Derived::Flags)&RowMajorBit ? int(Derived::MaxColsAtCompileTime)
: int(Derived::MaxRowsAtCompileTime),
MaxSizeAtCompileTime = Derived::SizeAtCompileTime,
PacketSize = packet_traits<typename Derived::Scalar>::size
};
enum {
StorageOrdersAgree = (int(Derived::IsRowMajor) == int(OtherDerived::IsRowMajor)),
MightVectorize = StorageOrdersAgree
&& (int(Derived::Flags) & int(OtherDerived::Flags) & ActualPacketAccessBit),
MayInnerVectorize = MightVectorize && int(InnerSize)!=Dynamic && int(InnerSize)%int(PacketSize)==0
&& int(DstIsAligned) && int(SrcIsAligned),
MayLinearize = StorageOrdersAgree && (int(Derived::Flags) & int(OtherDerived::Flags) & LinearAccessBit),
MayLinearVectorize = MightVectorize && MayLinearize && DstHasDirectAccess
&& (DstIsAligned || MaxSizeAtCompileTime == Dynamic),
/* If the destination isn't aligned, we have to do runtime checks and we don't unroll,
so it's only good for large enough sizes. */
MaySliceVectorize = MightVectorize && DstHasDirectAccess
&& (int(InnerMaxSize)==Dynamic || int(InnerMaxSize)>=3*PacketSize)
/* slice vectorization can be slow, so we only want it if the slices are big, which is
indicated by InnerMaxSize rather than InnerSize, think of the case of a dynamic block
in a fixed-size matrix */
};
public:
enum {
Traversal = int(MayInnerVectorize) ? int(InnerVectorizedTraversal)
: int(MayLinearVectorize) ? int(LinearVectorizedTraversal)
: int(MaySliceVectorize) ? int(SliceVectorizedTraversal)
: int(MayLinearize) ? int(LinearTraversal)
: int(DefaultTraversal),
Vectorized = int(Traversal) == InnerVectorizedTraversal
|| int(Traversal) == LinearVectorizedTraversal
|| int(Traversal) == SliceVectorizedTraversal
};
private:
enum {
UnrollingLimit = EIGEN_UNROLLING_LIMIT * (Vectorized ? int(PacketSize) : 1),
MayUnrollCompletely = int(Derived::SizeAtCompileTime) != Dynamic
&& int(OtherDerived::CoeffReadCost) != Dynamic
&& int(Derived::SizeAtCompileTime) * int(OtherDerived::CoeffReadCost) <= int(UnrollingLimit),
MayUnrollInner = int(InnerSize) != Dynamic
&& int(OtherDerived::CoeffReadCost) != Dynamic
&& int(InnerSize) * int(OtherDerived::CoeffReadCost) <= int(UnrollingLimit)
};
public:
enum {
Unrolling = (int(Traversal) == int(InnerVectorizedTraversal) || int(Traversal) == int(DefaultTraversal))
? (
int(MayUnrollCompletely) ? int(CompleteUnrolling)
: int(MayUnrollInner) ? int(InnerUnrolling)
: int(NoUnrolling)
)
: int(Traversal) == int(LinearVectorizedTraversal)
? ( bool(MayUnrollCompletely) && bool(DstIsAligned) ? int(CompleteUnrolling) : int(NoUnrolling) )
: int(Traversal) == int(LinearTraversal)
? ( bool(MayUnrollCompletely) ? int(CompleteUnrolling) : int(NoUnrolling) )
: int(NoUnrolling)
};
#ifdef EIGEN_DEBUG_ASSIGN
static void debug()
{
EIGEN_DEBUG_VAR(DstIsAligned)
EIGEN_DEBUG_VAR(SrcIsAligned)
EIGEN_DEBUG_VAR(JointAlignment)
EIGEN_DEBUG_VAR(InnerSize)
EIGEN_DEBUG_VAR(InnerMaxSize)
EIGEN_DEBUG_VAR(PacketSize)
EIGEN_DEBUG_VAR(StorageOrdersAgree)
EIGEN_DEBUG_VAR(MightVectorize)
EIGEN_DEBUG_VAR(MayLinearize)
EIGEN_DEBUG_VAR(MayInnerVectorize)
EIGEN_DEBUG_VAR(MayLinearVectorize)
EIGEN_DEBUG_VAR(MaySliceVectorize)
EIGEN_DEBUG_VAR(Traversal)
EIGEN_DEBUG_VAR(UnrollingLimit)
EIGEN_DEBUG_VAR(MayUnrollCompletely)
EIGEN_DEBUG_VAR(MayUnrollInner)
EIGEN_DEBUG_VAR(Unrolling)
}
#endif
};
/***************************************************************************
* Part 2 : meta-unrollers
***************************************************************************/
/************************
*** Default traversal ***
************************/
template<typename Derived1, typename Derived2, int Index, int Stop>
struct assign_DefaultTraversal_CompleteUnrolling
{
enum {
outer = Index / Derived1::InnerSizeAtCompileTime,
inner = Index % Derived1::InnerSizeAtCompileTime
};
static EIGEN_STRONG_INLINE void run(Derived1 &dst, const Derived2 &src)
{
dst.copyCoeffByOuterInner(outer, inner, src);
assign_DefaultTraversal_CompleteUnrolling<Derived1, Derived2, Index+1, Stop>::run(dst, src);
}
};
template<typename Derived1, typename Derived2, int Stop>
struct assign_DefaultTraversal_CompleteUnrolling<Derived1, Derived2, Stop, Stop>
{
static EIGEN_STRONG_INLINE void run(Derived1 &, const Derived2 &) {}
};
template<typename Derived1, typename Derived2, int Index, int Stop>
struct assign_DefaultTraversal_InnerUnrolling
{
static EIGEN_STRONG_INLINE void run(Derived1 &dst, const Derived2 &src, typename Derived1::Index outer)
{
dst.copyCoeffByOuterInner(outer, Index, src);
assign_DefaultTraversal_InnerUnrolling<Derived1, Derived2, Index+1, Stop>::run(dst, src, outer);
}
};
template<typename Derived1, typename Derived2, int Stop>
struct assign_DefaultTraversal_InnerUnrolling<Derived1, Derived2, Stop, Stop>
{
static EIGEN_STRONG_INLINE void run(Derived1 &, const Derived2 &, typename Derived1::Index) {}
};
/***********************
*** Linear traversal ***
***********************/
template<typename Derived1, typename Derived2, int Index, int Stop>
struct assign_LinearTraversal_CompleteUnrolling
{
static EIGEN_STRONG_INLINE void run(Derived1 &dst, const Derived2 &src)
{
dst.copyCoeff(Index, src);
assign_LinearTraversal_CompleteUnrolling<Derived1, Derived2, Index+1, Stop>::run(dst, src);
}
};
template<typename Derived1, typename Derived2, int Stop>
struct assign_LinearTraversal_CompleteUnrolling<Derived1, Derived2, Stop, Stop>
{
static EIGEN_STRONG_INLINE void run(Derived1 &, const Derived2 &) {}
};
/**************************
*** Inner vectorization ***
**************************/
template<typename Derived1, typename Derived2, int Index, int Stop>
struct assign_innervec_CompleteUnrolling
{
enum {
outer = Index / Derived1::InnerSizeAtCompileTime,
inner = Index % Derived1::InnerSizeAtCompileTime,
JointAlignment = assign_traits<Derived1,Derived2>::JointAlignment
};
static EIGEN_STRONG_INLINE void run(Derived1 &dst, const Derived2 &src)
{
dst.template copyPacketByOuterInner<Derived2, Aligned, JointAlignment>(outer, inner, src);
assign_innervec_CompleteUnrolling<Derived1, Derived2,
Index+packet_traits<typename Derived1::Scalar>::size, Stop>::run(dst, src);
}
};
template<typename Derived1, typename Derived2, int Stop>
struct assign_innervec_CompleteUnrolling<Derived1, Derived2, Stop, Stop>
{
static EIGEN_STRONG_INLINE void run(Derived1 &, const Derived2 &) {}
};
template<typename Derived1, typename Derived2, int Index, int Stop>
struct assign_innervec_InnerUnrolling
{
static EIGEN_STRONG_INLINE void run(Derived1 &dst, const Derived2 &src, typename Derived1::Index outer)
{
dst.template copyPacketByOuterInner<Derived2, Aligned, Aligned>(outer, Index, src);
assign_innervec_InnerUnrolling<Derived1, Derived2,
Index+packet_traits<typename Derived1::Scalar>::size, Stop>::run(dst, src, outer);
}
};
template<typename Derived1, typename Derived2, int Stop>
struct assign_innervec_InnerUnrolling<Derived1, Derived2, Stop, Stop>
{
static EIGEN_STRONG_INLINE void run(Derived1 &, const Derived2 &, typename Derived1::Index) {}
};
/***************************************************************************
* Part 3 : implementation of all cases
***************************************************************************/
template<typename Derived1, typename Derived2,
int Traversal = assign_traits<Derived1, Derived2>::Traversal,
int Unrolling = assign_traits<Derived1, Derived2>::Unrolling,
int Version = Specialized>
struct assign_impl;
/************************
*** Default traversal ***
************************/
template<typename Derived1, typename Derived2, int Unrolling, int Version>
struct assign_impl<Derived1, Derived2, InvalidTraversal, Unrolling, Version>
{
static inline void run(Derived1 &, const Derived2 &) { }
};
template<typename Derived1, typename Derived2, int Version>
struct assign_impl<Derived1, Derived2, DefaultTraversal, NoUnrolling, Version>
{
typedef typename Derived1::Index Index;
static inline void run(Derived1 &dst, const Derived2 &src)
{
const Index innerSize = dst.innerSize();
const Index outerSize = dst.outerSize();
for(Index outer = 0; outer < outerSize; ++outer)
for(Index inner = 0; inner < innerSize; ++inner)
dst.copyCoeffByOuterInner(outer, inner, src);
}
};
template<typename Derived1, typename Derived2, int Version>
struct assign_impl<Derived1, Derived2, DefaultTraversal, CompleteUnrolling, Version>
{
static EIGEN_STRONG_INLINE void run(Derived1 &dst, const Derived2 &src)
{
assign_DefaultTraversal_CompleteUnrolling<Derived1, Derived2, 0, Derived1::SizeAtCompileTime>
::run(dst, src);
}
};
template<typename Derived1, typename Derived2, int Version>
struct assign_impl<Derived1, Derived2, DefaultTraversal, InnerUnrolling, Version>
{
typedef typename Derived1::Index Index;
static EIGEN_STRONG_INLINE void run(Derived1 &dst, const Derived2 &src)
{
const Index outerSize = dst.outerSize();
for(Index outer = 0; outer < outerSize; ++outer)
assign_DefaultTraversal_InnerUnrolling<Derived1, Derived2, 0, Derived1::InnerSizeAtCompileTime>
::run(dst, src, outer);
}
};
/***********************
*** Linear traversal ***
***********************/
template<typename Derived1, typename Derived2, int Version>
struct assign_impl<Derived1, Derived2, LinearTraversal, NoUnrolling, Version>
{
typedef typename Derived1::Index Index;
static inline void run(Derived1 &dst, const Derived2 &src)
{
const Index size = dst.size();
for(Index i = 0; i < size; ++i)
dst.copyCoeff(i, src);
}
};
template<typename Derived1, typename Derived2, int Version>
struct assign_impl<Derived1, Derived2, LinearTraversal, CompleteUnrolling, Version>
{
static EIGEN_STRONG_INLINE void run(Derived1 &dst, const Derived2 &src)
{
assign_LinearTraversal_CompleteUnrolling<Derived1, Derived2, 0, Derived1::SizeAtCompileTime>
::run(dst, src);
}
};
/**************************
*** Inner vectorization ***
**************************/
template<typename Derived1, typename Derived2, int Version>
struct assign_impl<Derived1, Derived2, InnerVectorizedTraversal, NoUnrolling, Version>
{
typedef typename Derived1::Index Index;
static inline void run(Derived1 &dst, const Derived2 &src)
{
const Index innerSize = dst.innerSize();
const Index outerSize = dst.outerSize();
const Index packetSize = packet_traits<typename Derived1::Scalar>::size;
for(Index outer = 0; outer < outerSize; ++outer)
for(Index inner = 0; inner < innerSize; inner+=packetSize)
dst.template copyPacketByOuterInner<Derived2, Aligned, Aligned>(outer, inner, src);
}
};
template<typename Derived1, typename Derived2, int Version>
struct assign_impl<Derived1, Derived2, InnerVectorizedTraversal, CompleteUnrolling, Version>
{
static EIGEN_STRONG_INLINE void run(Derived1 &dst, const Derived2 &src)
{
assign_innervec_CompleteUnrolling<Derived1, Derived2, 0, Derived1::SizeAtCompileTime>
::run(dst, src);
}
};
template<typename Derived1, typename Derived2, int Version>
struct assign_impl<Derived1, Derived2, InnerVectorizedTraversal, InnerUnrolling, Version>
{
typedef typename Derived1::Index Index;
static EIGEN_STRONG_INLINE void run(Derived1 &dst, const Derived2 &src)
{
const Index outerSize = dst.outerSize();
for(Index outer = 0; outer < outerSize; ++outer)
assign_innervec_InnerUnrolling<Derived1, Derived2, 0, Derived1::InnerSizeAtCompileTime>
::run(dst, src, outer);
}
};
/***************************
*** Linear vectorization ***
***************************/
template <bool IsAligned = false>
struct unaligned_assign_impl
{
template <typename Derived, typename OtherDerived>
static EIGEN_STRONG_INLINE void run(const Derived&, OtherDerived&, typename Derived::Index, typename Derived::Index) {}
};
template <>
struct unaligned_assign_impl<false>
{
// MSVC must not inline this functions. If it does, it fails to optimize the
// packet access path.
#ifdef _MSC_VER
template <typename Derived, typename OtherDerived>
static EIGEN_DONT_INLINE void run(const Derived& src, OtherDerived& dst, typename Derived::Index start, typename Derived::Index end)
#else
template <typename Derived, typename OtherDerived>
static EIGEN_STRONG_INLINE void run(const Derived& src, OtherDerived& dst, typename Derived::Index start, typename Derived::Index end)
#endif
{
for (typename Derived::Index index = start; index < end; ++index)
dst.copyCoeff(index, src);
}
};
template<typename Derived1, typename Derived2, int Version>
struct assign_impl<Derived1, Derived2, LinearVectorizedTraversal, NoUnrolling, Version>
{
typedef typename Derived1::Index Index;
static EIGEN_STRONG_INLINE void run(Derived1 &dst, const Derived2 &src)
{
const Index size = dst.size();
typedef packet_traits<typename Derived1::Scalar> PacketTraits;
enum {
packetSize = PacketTraits::size,
dstAlignment = PacketTraits::AlignedOnScalar ? Aligned : int(assign_traits<Derived1,Derived2>::DstIsAligned) ,
srcAlignment = assign_traits<Derived1,Derived2>::JointAlignment
};
const Index alignedStart = assign_traits<Derived1,Derived2>::DstIsAligned ? 0
: internal::first_aligned(&dst.coeffRef(0), size);
const Index alignedEnd = alignedStart + ((size-alignedStart)/packetSize)*packetSize;
unaligned_assign_impl<assign_traits<Derived1,Derived2>::DstIsAligned!=0>::run(src,dst,0,alignedStart);
for(Index index = alignedStart; index < alignedEnd; index += packetSize)
{
dst.template copyPacket<Derived2, dstAlignment, srcAlignment>(index, src);
}
unaligned_assign_impl<>::run(src,dst,alignedEnd,size);
}
};
template<typename Derived1, typename Derived2, int Version>
struct assign_impl<Derived1, Derived2, LinearVectorizedTraversal, CompleteUnrolling, Version>
{
typedef typename Derived1::Index Index;
static EIGEN_STRONG_INLINE void run(Derived1 &dst, const Derived2 &src)
{
enum { size = Derived1::SizeAtCompileTime,
packetSize = packet_traits<typename Derived1::Scalar>::size,
alignedSize = (size/packetSize)*packetSize };
assign_innervec_CompleteUnrolling<Derived1, Derived2, 0, alignedSize>::run(dst, src);
assign_DefaultTraversal_CompleteUnrolling<Derived1, Derived2, alignedSize, size>::run(dst, src);
}
};
/**************************
*** Slice vectorization ***
***************************/
template<typename Derived1, typename Derived2, int Version>
struct assign_impl<Derived1, Derived2, SliceVectorizedTraversal, NoUnrolling, Version>
{
typedef typename Derived1::Index Index;
static inline void run(Derived1 &dst, const Derived2 &src)
{
typedef typename Derived1::Scalar Scalar;
typedef packet_traits<Scalar> PacketTraits;
enum {
packetSize = PacketTraits::size,
alignable = PacketTraits::AlignedOnScalar,
dstIsAligned = assign_traits<Derived1,Derived2>::DstIsAligned,
dstAlignment = alignable ? Aligned : int(dstIsAligned),
srcAlignment = assign_traits<Derived1,Derived2>::JointAlignment
};
const Scalar *dst_ptr = &dst.coeffRef(0,0);
if((!bool(dstIsAligned)) && (size_t(dst_ptr) % sizeof(Scalar))>0)
{
// the pointer is not aligend-on scalar, so alignment is not possible
return assign_impl<Derived1,Derived2,DefaultTraversal,NoUnrolling>::run(dst, src);
}
const Index packetAlignedMask = packetSize - 1;
const Index innerSize = dst.innerSize();
const Index outerSize = dst.outerSize();
const Index alignedStep = alignable ? (packetSize - dst.outerStride() % packetSize) & packetAlignedMask : 0;
Index alignedStart = ((!alignable) || bool(dstIsAligned)) ? 0 : internal::first_aligned(dst_ptr, innerSize);
for(Index outer = 0; outer < outerSize; ++outer)
{
const Index alignedEnd = alignedStart + ((innerSize-alignedStart) & ~packetAlignedMask);
// do the non-vectorizable part of the assignment
for(Index inner = 0; inner<alignedStart ; ++inner)
dst.copyCoeffByOuterInner(outer, inner, src);
// do the vectorizable part of the assignment
for(Index inner = alignedStart; inner<alignedEnd; inner+=packetSize)
dst.template copyPacketByOuterInner<Derived2, dstAlignment, Unaligned>(outer, inner, src);
// do the non-vectorizable part of the assignment
for(Index inner = alignedEnd; inner<innerSize ; ++inner)
dst.copyCoeffByOuterInner(outer, inner, src);
alignedStart = std::min<Index>((alignedStart+alignedStep)%packetSize, innerSize);
}
}
};
} // end namespace internal
/***************************************************************************
* Part 4 : implementation of DenseBase methods
***************************************************************************/
template<typename Derived>
template<typename OtherDerived>
EIGEN_STRONG_INLINE Derived& DenseBase<Derived>
......@@ -499,90 +27,62 @@ EIGEN_STRONG_INLINE Derived& DenseBase<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)
#ifdef EIGEN_DEBUG_ASSIGN
internal::assign_traits<Derived, OtherDerived>::debug();
#endif
eigen_assert(rows() == other.rows() && cols() == other.cols());
internal::assign_impl<Derived, OtherDerived, int(SameType) ? int(internal::assign_traits<Derived, OtherDerived>::Traversal)
: int(InvalidTraversal)>::run(derived(),other.derived());
#ifndef EIGEN_NO_DEBUG
checkTransposeAliasing(other.derived());
#endif
internal::call_assignment_no_alias(derived(),other.derived());
return derived();
}
namespace internal {
template<typename Derived, typename OtherDerived,
bool EvalBeforeAssigning = (int(internal::traits<OtherDerived>::Flags) & EvalBeforeAssigningBit) != 0,
bool NeedToTranspose = ((int(Derived::RowsAtCompileTime) == 1 && int(OtherDerived::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(Derived::ColsAtCompileTime) == 1 && int(OtherDerived::RowsAtCompileTime) == 1))
&& int(Derived::SizeAtCompileTime) != 1>
struct assign_selector;
template<typename Derived, typename OtherDerived>
struct assign_selector<Derived,OtherDerived,false,false> {
static EIGEN_STRONG_INLINE Derived& run(Derived& dst, const OtherDerived& other) { return dst.lazyAssign(other.derived()); }
template<typename ActualDerived, typename ActualOtherDerived>
static EIGEN_STRONG_INLINE Derived& evalTo(ActualDerived& dst, const ActualOtherDerived& other) { other.evalTo(dst); return dst; }
};
template<typename Derived, typename OtherDerived>
struct assign_selector<Derived,OtherDerived,true,false> {
static EIGEN_STRONG_INLINE Derived& run(Derived& dst, const OtherDerived& other) { return dst.lazyAssign(other.eval()); }
};
template<typename Derived, typename OtherDerived>
struct assign_selector<Derived,OtherDerived,false,true> {
static EIGEN_STRONG_INLINE Derived& run(Derived& dst, const OtherDerived& other) { return dst.lazyAssign(other.transpose()); }
template<typename ActualDerived, typename ActualOtherDerived>
static EIGEN_STRONG_INLINE Derived& evalTo(ActualDerived& dst, const ActualOtherDerived& other) { Transpose<ActualDerived> dstTrans(dst); other.evalTo(dstTrans); return dst; }
};
template<typename Derived, typename OtherDerived>
struct assign_selector<Derived,OtherDerived,true,true> {
static EIGEN_STRONG_INLINE Derived& run(Derived& dst, const OtherDerived& other) { return dst.lazyAssign(other.transpose().eval()); }
};
} // end namespace internal
template<typename Derived>
template<typename OtherDerived>
EIGEN_DEVICE_FUNC
EIGEN_STRONG_INLINE Derived& DenseBase<Derived>::operator=(const DenseBase<OtherDerived>& other)
{
return internal::assign_selector<Derived,OtherDerived>::run(derived(), other.derived());
internal::call_assignment(derived(), other.derived());
return derived();
}
template<typename Derived>
EIGEN_DEVICE_FUNC
EIGEN_STRONG_INLINE Derived& DenseBase<Derived>::operator=(const DenseBase& other)
{
return internal::assign_selector<Derived,Derived>::run(derived(), other.derived());
internal::call_assignment(derived(), other.derived());
return derived();
}
template<typename Derived>
EIGEN_DEVICE_FUNC
EIGEN_STRONG_INLINE Derived& MatrixBase<Derived>::operator=(const MatrixBase& other)
{
return internal::assign_selector<Derived,Derived>::run(derived(), other.derived());
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)
{
return internal::assign_selector<Derived,OtherDerived>::run(derived(), other.derived());
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)
{
return internal::assign_selector<Derived,OtherDerived,false>::evalTo(derived(), other.derived());
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)
{
return internal::assign_selector<Derived,OtherDerived,false>::evalTo(derived(), other.derived());
other.derived().evalTo(derived());
return derived();
}
} // end namespace Eigen
......
external/eigen3/Eigen/src/Core/AssignEvaluator.h 0 → 100644
View file @ a394b22a
// This file is part of Eigen, a lightweight C++ template library
// for linear algebra.
//
// Copyright (C) 2011 Benoit Jacob <jacob.benoit.1@gmail.com>
// Copyright (C) 2011-2014 Gael Guennebaud <gael.guennebaud@inria.fr>
// Copyright (C) 2011-2012 Jitse Niesen <jitse@maths.leeds.ac.uk>
//
// 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_EVALUATOR_H
#define EIGEN_ASSIGN_EVALUATOR_H
namespace Eigen {
// This implementation is based on Assign.h
namespace internal {
/***************************************************************************
* Part 1 : the logic deciding a strategy for traversal and unrolling *
***************************************************************************/
// copy_using_evaluator_traits is based on assign_traits
template <typename DstEvaluator, typename SrcEvaluator, typename AssignFunc>
struct copy_using_evaluator_traits
{
typedef typename DstEvaluator::XprType Dst;
typedef typename Dst::Scalar DstScalar;
enum {
DstFlags = DstEvaluator::Flags,
SrcFlags = SrcEvaluator::Flags
};
public:
enum {
DstAlignment = DstEvaluator::Alignment,
SrcAlignment = SrcEvaluator::Alignment,
DstHasDirectAccess = DstFlags & DirectAccessBit,
JointAlignment = EIGEN_PLAIN_ENUM_MIN(DstAlignment,SrcAlignment)
};
private:
enum {
InnerSize = int(Dst::IsVectorAtCompileTime) ? int(Dst::SizeAtCompileTime)
: int(DstFlags)&RowMajorBit ? int(Dst::ColsAtCompileTime)
: int(Dst::RowsAtCompileTime),
InnerMaxSize = int(Dst::IsVectorAtCompileTime) ? int(Dst::MaxSizeAtCompileTime)
: int(DstFlags)&RowMajorBit ? int(Dst::MaxColsAtCompileTime)
: int(Dst::MaxRowsAtCompileTime),
OuterStride = int(outer_stride_at_compile_time<Dst>::ret),
MaxSizeAtCompileTime = Dst::SizeAtCompileTime
};
// TODO distinguish between linear traversal and inner-traversals
typedef typename find_best_packet<DstScalar,Dst::SizeAtCompileTime>::type LinearPacketType;
typedef typename find_best_packet<DstScalar,InnerSize>::type InnerPacketType;
enum {
LinearPacketSize = unpacket_traits<LinearPacketType>::size,
InnerPacketSize = unpacket_traits<InnerPacketType>::size
};
public:
enum {
LinearRequiredAlignment = unpacket_traits<LinearPacketType>::alignment,
InnerRequiredAlignment = unpacket_traits<InnerPacketType>::alignment
};
private:
enum {
DstIsRowMajor = DstFlags&RowMajorBit,
SrcIsRowMajor = SrcFlags&RowMajorBit,
StorageOrdersAgree = (int(DstIsRowMajor) == int(SrcIsRowMajor)),
MightVectorize = bool(StorageOrdersAgree)
&& (int(DstFlags) & int(SrcFlags) & ActualPacketAccessBit)
&& bool(functor_traits<AssignFunc>::PacketAccess),
MayInnerVectorize = MightVectorize
&& int(InnerSize)!=Dynamic && int(InnerSize)%int(InnerPacketSize)==0
&& int(OuterStride)!=Dynamic && int(OuterStride)%int(InnerPacketSize)==0
&& (EIGEN_UNALIGNED_VECTORIZE || int(JointAlignment)>=int(InnerRequiredAlignment)),
MayLinearize = bool(StorageOrdersAgree) && (int(DstFlags) & int(SrcFlags) & LinearAccessBit),
MayLinearVectorize = bool(MightVectorize) && MayLinearize && DstHasDirectAccess
&& (EIGEN_UNALIGNED_VECTORIZE || (int(DstAlignment)>=int(LinearRequiredAlignment)) || MaxSizeAtCompileTime == Dynamic),
/* If the destination isn't aligned, we have to do runtime checks and we don't unroll,
so it's only good for large enough sizes. */
MaySliceVectorize = bool(MightVectorize) && bool(DstHasDirectAccess)
&& (int(InnerMaxSize)==Dynamic || int(InnerMaxSize)>=(EIGEN_UNALIGNED_VECTORIZE?InnerPacketSize:(3*InnerPacketSize)))
/* slice vectorization can be slow, so we only want it if the slices are big, which is
indicated by InnerMaxSize rather than InnerSize, think of the case of a dynamic block
in a fixed-size matrix
However, with EIGEN_UNALIGNED_VECTORIZE and unrolling, slice vectorization is still worth it */
};
public:
enum {
Traversal = int(MayLinearVectorize) && (LinearPacketSize>InnerPacketSize) ? int(LinearVectorizedTraversal)
: int(MayInnerVectorize) ? int(InnerVectorizedTraversal)
: int(MayLinearVectorize) ? int(LinearVectorizedTraversal)
: int(MaySliceVectorize) ? int(SliceVectorizedTraversal)
: int(MayLinearize) ? int(LinearTraversal)
: int(DefaultTraversal),
Vectorized = int(Traversal) == InnerVectorizedTraversal
|| int(Traversal) == LinearVectorizedTraversal
|| int(Traversal) == SliceVectorizedTraversal
};
typedef typename conditional<int(Traversal)==LinearVectorizedTraversal, LinearPacketType, InnerPacketType>::type PacketType;
private:
enum {
ActualPacketSize = int(Traversal)==LinearVectorizedTraversal ? LinearPacketSize
: Vectorized ? InnerPacketSize
: 1,
UnrollingLimit = EIGEN_UNROLLING_LIMIT * ActualPacketSize,
MayUnrollCompletely = int(Dst::SizeAtCompileTime) != Dynamic
&& int(Dst::SizeAtCompileTime) * (int(DstEvaluator::CoeffReadCost)+int(SrcEvaluator::CoeffReadCost)) <= int(UnrollingLimit),
MayUnrollInner = int(InnerSize) != Dynamic
&& int(InnerSize) * (int(DstEvaluator::CoeffReadCost)+int(SrcEvaluator::CoeffReadCost)) <= int(UnrollingLimit)
};
public:
enum {
Unrolling = (int(Traversal) == int(InnerVectorizedTraversal) || int(Traversal) == int(DefaultTraversal))
? (
int(MayUnrollCompletely) ? int(CompleteUnrolling)
: int(MayUnrollInner) ? int(InnerUnrolling)
: int(NoUnrolling)
)
: int(Traversal) == int(LinearVectorizedTraversal)
? ( bool(MayUnrollCompletely) && ( EIGEN_UNALIGNED_VECTORIZE || (int(DstAlignment)>=int(LinearRequiredAlignment)))
? int(CompleteUnrolling)
: int(NoUnrolling) )
: int(Traversal) == int(LinearTraversal)
? ( bool(MayUnrollCompletely) ? int(CompleteUnrolling)
: int(NoUnrolling) )
#if EIGEN_UNALIGNED_VECTORIZE
: int(Traversal) == int(SliceVectorizedTraversal)
? ( bool(MayUnrollInner) ? int(InnerUnrolling)
: int(NoUnrolling) )
#endif
: int(NoUnrolling)
};
#ifdef EIGEN_DEBUG_ASSIGN
static void debug()
{
std::cerr << "DstXpr: " << typeid(typename DstEvaluator::XprType).name() << std::endl;
std::cerr << "SrcXpr: " << typeid(typename SrcEvaluator::XprType).name() << std::endl;
std::cerr.setf(std::ios::hex, std::ios::basefield);
std::cerr << "DstFlags" << " = " << DstFlags << " (" << demangle_flags(DstFlags) << " )" << std::endl;
std::cerr << "SrcFlags" << " = " << SrcFlags << " (" << demangle_flags(SrcFlags) << " )" << std::endl;
std::cerr.unsetf(std::ios::hex);
EIGEN_DEBUG_VAR(DstAlignment)
EIGEN_DEBUG_VAR(SrcAlignment)
EIGEN_DEBUG_VAR(LinearRequiredAlignment)
EIGEN_DEBUG_VAR(InnerRequiredAlignment)
EIGEN_DEBUG_VAR(JointAlignment)
EIGEN_DEBUG_VAR(InnerSize)
EIGEN_DEBUG_VAR(InnerMaxSize)
EIGEN_DEBUG_VAR(LinearPacketSize)
EIGEN_DEBUG_VAR(InnerPacketSize)
EIGEN_DEBUG_VAR(ActualPacketSize)
EIGEN_DEBUG_VAR(StorageOrdersAgree)
EIGEN_DEBUG_VAR(MightVectorize)
EIGEN_DEBUG_VAR(MayLinearize)
EIGEN_DEBUG_VAR(MayInnerVectorize)
EIGEN_DEBUG_VAR(MayLinearVectorize)
EIGEN_DEBUG_VAR(MaySliceVectorize)
std::cerr << "Traversal" << " = " << Traversal << " (" << demangle_traversal(Traversal) << ")" << std::endl;
EIGEN_DEBUG_VAR(SrcEvaluator::CoeffReadCost)
EIGEN_DEBUG_VAR(UnrollingLimit)
EIGEN_DEBUG_VAR(MayUnrollCompletely)
EIGEN_DEBUG_VAR(MayUnrollInner)
std::cerr << "Unrolling" << " = " << Unrolling << " (" << demangle_unrolling(Unrolling) << ")" << std::endl;
std::cerr << std::endl;
}
#endif
};
/***************************************************************************
* Part 2 : meta-unrollers
***************************************************************************/
/************************
*** Default traversal ***
************************/
template<typename Kernel, int Index, int Stop>
struct copy_using_evaluator_DefaultTraversal_CompleteUnrolling
{
// FIXME: this is not very clean, perhaps this information should be provided by the kernel?
typedef typename Kernel::DstEvaluatorType DstEvaluatorType;
typedef typename DstEvaluatorType::XprType DstXprType;
enum {
outer = Index / DstXprType::InnerSizeAtCompileTime,
inner = Index % DstXprType::InnerSizeAtCompileTime
};
EIGEN_DEVICE_FUNC static EIGEN_STRONG_INLINE void run(Kernel &kernel)
{
kernel.assignCoeffByOuterInner(outer, inner);
copy_using_evaluator_DefaultTraversal_CompleteUnrolling<Kernel, Index+1, Stop>::run(kernel);
}
};
template<typename Kernel, int Stop>
struct copy_using_evaluator_DefaultTraversal_CompleteUnrolling<Kernel, Stop, Stop>
{
EIGEN_DEVICE_FUNC static EIGEN_STRONG_INLINE void run(Kernel&) { }
};
template<typename Kernel, int Index_, int Stop>
struct copy_using_evaluator_DefaultTraversal_InnerUnrolling
{
EIGEN_DEVICE_FUNC static EIGEN_STRONG_INLINE void run(Kernel &kernel, Index outer)
{
kernel.assignCoeffByOuterInner(outer, Index_);
copy_using_evaluator_DefaultTraversal_InnerUnrolling<Kernel, Index_+1, Stop>::run(kernel, outer);
}
};
template<typename Kernel, int Stop>
struct copy_using_evaluator_DefaultTraversal_InnerUnrolling<Kernel, Stop, Stop>
{
EIGEN_DEVICE_FUNC static EIGEN_STRONG_INLINE void run(Kernel&, Index) { }
};
/***********************
*** Linear traversal ***
***********************/
template<typename Kernel, int Index, int Stop>
struct copy_using_evaluator_LinearTraversal_CompleteUnrolling
{
EIGEN_DEVICE_FUNC static EIGEN_STRONG_INLINE void run(Kernel& kernel)
{
kernel.assignCoeff(Index);
copy_using_evaluator_LinearTraversal_CompleteUnrolling<Kernel, Index+1, Stop>::run(kernel);
}
};
template<typename Kernel, int Stop>
struct copy_using_evaluator_LinearTraversal_CompleteUnrolling<Kernel, Stop, Stop>
{
EIGEN_DEVICE_FUNC static EIGEN_STRONG_INLINE void run(Kernel&) { }
};
/**************************
*** Inner vectorization ***
**************************/
template<typename Kernel, int Index, int Stop>
struct copy_using_evaluator_innervec_CompleteUnrolling
{
// FIXME: this is not very clean, perhaps this information should be provided by the kernel?
typedef typename Kernel::DstEvaluatorType DstEvaluatorType;
typedef typename DstEvaluatorType::XprType DstXprType;
typedef typename Kernel::PacketType PacketType;
enum {
outer = Index / DstXprType::InnerSizeAtCompileTime,
inner = Index % DstXprType::InnerSizeAtCompileTime,
SrcAlignment = Kernel::AssignmentTraits::SrcAlignment,
DstAlignment = Kernel::AssignmentTraits::DstAlignment
};
EIGEN_DEVICE_FUNC static EIGEN_STRONG_INLINE void run(Kernel &kernel)
{
kernel.template assignPacketByOuterInner<DstAlignment, SrcAlignment, PacketType>(outer, inner);
enum { NextIndex = Index + unpacket_traits<PacketType>::size };
copy_using_evaluator_innervec_CompleteUnrolling<Kernel, NextIndex, Stop>::run(kernel);
}
};
template<typename Kernel, int Stop>
struct copy_using_evaluator_innervec_CompleteUnrolling<Kernel, Stop, Stop>
{
EIGEN_DEVICE_FUNC static EIGEN_STRONG_INLINE void run(Kernel&) { }
};
template<typename Kernel, int Index_, int Stop, int SrcAlignment, int DstAlignment>
struct copy_using_evaluator_innervec_InnerUnrolling
{
typedef typename Kernel::PacketType PacketType;
EIGEN_DEVICE_FUNC static EIGEN_STRONG_INLINE void run(Kernel &kernel, Index outer)
{
kernel.template assignPacketByOuterInner<DstAlignment, SrcAlignment, PacketType>(outer, Index_);
enum { NextIndex = Index_ + unpacket_traits<PacketType>::size };
copy_using_evaluator_innervec_InnerUnrolling<Kernel, NextIndex, Stop, SrcAlignment, DstAlignment>::run(kernel, outer);
}
};
template<typename Kernel, int Stop, int SrcAlignment, int DstAlignment>
struct copy_using_evaluator_innervec_InnerUnrolling<Kernel, Stop, Stop, SrcAlignment, DstAlignment>
{
EIGEN_DEVICE_FUNC static EIGEN_STRONG_INLINE void run(Kernel &, Index) { }
};
/***************************************************************************
* Part 3 : implementation of all cases
***************************************************************************/
// dense_assignment_loop is based on assign_impl
template<typename Kernel,
int Traversal = Kernel::AssignmentTraits::Traversal,
int Unrolling = Kernel::AssignmentTraits::Unrolling>
struct dense_assignment_loop;
/************************
*** Default traversal ***
************************/
template<typename Kernel>
struct dense_assignment_loop<Kernel, DefaultTraversal, NoUnrolling>
{
EIGEN_DEVICE_FUNC static void EIGEN_STRONG_INLINE run(Kernel &kernel)
{
for(Index outer = 0; outer < kernel.outerSize(); ++outer) {
for(Index inner = 0; inner < kernel.innerSize(); ++inner) {
kernel.assignCoeffByOuterInner(outer, inner);
}
}
}
};
template<typename Kernel>
struct dense_assignment_loop<Kernel, DefaultTraversal, CompleteUnrolling>
{
EIGEN_DEVICE_FUNC static EIGEN_STRONG_INLINE void run(Kernel &kernel)
{
typedef typename Kernel::DstEvaluatorType::XprType DstXprType;
copy_using_evaluator_DefaultTraversal_CompleteUnrolling<Kernel, 0, DstXprType::SizeAtCompileTime>::run(kernel);
}
};
template<typename Kernel>
struct dense_assignment_loop<Kernel, DefaultTraversal, InnerUnrolling>
{
EIGEN_DEVICE_FUNC static EIGEN_STRONG_INLINE void run(Kernel &kernel)
{
typedef typename Kernel::DstEvaluatorType::XprType DstXprType;
const Index outerSize = kernel.outerSize();
for(Index outer = 0; outer < outerSize; ++outer)
copy_using_evaluator_DefaultTraversal_InnerUnrolling<Kernel, 0, DstXprType::InnerSizeAtCompileTime>::run(kernel, outer);
}
};
/***************************
*** Linear vectorization ***
***************************/
// The goal of unaligned_dense_assignment_loop is simply to factorize the handling
// of the non vectorizable beginning and ending parts
template <bool IsAligned = false>
struct unaligned_dense_assignment_loop
{
// if IsAligned = true, then do nothing
template <typename Kernel>
EIGEN_DEVICE_FUNC static EIGEN_STRONG_INLINE void run(Kernel&, Index, Index) {}
};
template <>
struct unaligned_dense_assignment_loop<false>
{
// MSVC must not inline this functions. If it does, it fails to optimize the
// packet access path.
// FIXME check which version exhibits this issue
#if EIGEN_COMP_MSVC
template <typename Kernel>
static EIGEN_DONT_INLINE void run(Kernel &kernel,
Index start,
Index end)
#else
template <typename Kernel>
EIGEN_DEVICE_FUNC static EIGEN_STRONG_INLINE void run(Kernel &kernel,
Index start,
Index end)
#endif
{
for (Index index = start; index < end; ++index)
kernel.assignCoeff(index);
}
};
template<typename Kernel>
struct dense_assignment_loop<Kernel, LinearVectorizedTraversal, NoUnrolling>
{
EIGEN_DEVICE_FUNC static EIGEN_STRONG_INLINE void run(Kernel &kernel)
{
const Index size = kernel.size();
typedef typename Kernel::Scalar Scalar;
typedef typename Kernel::PacketType PacketType;
enum {
requestedAlignment = Kernel::AssignmentTraits::LinearRequiredAlignment,
packetSize = unpacket_traits<PacketType>::size,
dstIsAligned = int(Kernel::AssignmentTraits::DstAlignment)>=int(requestedAlignment),
dstAlignment = packet_traits<Scalar>::AlignedOnScalar ? int(requestedAlignment)
: int(Kernel::AssignmentTraits::DstAlignment),
srcAlignment = Kernel::AssignmentTraits::JointAlignment
};
const Index alignedStart = dstIsAligned ? 0 : internal::first_aligned<requestedAlignment>(kernel.dstDataPtr(), size);
const Index alignedEnd = alignedStart + ((size-alignedStart)/packetSize)*packetSize;
unaligned_dense_assignment_loop<dstIsAligned!=0>::run(kernel, 0, alignedStart);
for(Index index = alignedStart; index < alignedEnd; index += packetSize)
kernel.template assignPacket<dstAlignment, srcAlignment, PacketType>(index);
unaligned_dense_assignment_loop<>::run(kernel, alignedEnd, size);
}
};
template<typename Kernel>
struct dense_assignment_loop<Kernel, LinearVectorizedTraversal, CompleteUnrolling>
{
EIGEN_DEVICE_FUNC static EIGEN_STRONG_INLINE void run(Kernel &kernel)
{
typedef typename Kernel::DstEvaluatorType::XprType DstXprType;
typedef typename Kernel::PacketType PacketType;
enum { size = DstXprType::SizeAtCompileTime,
packetSize =unpacket_traits<PacketType>::size,
alignedSize = (size/packetSize)*packetSize };
copy_using_evaluator_innervec_CompleteUnrolling<Kernel, 0, alignedSize>::run(kernel);
copy_using_evaluator_DefaultTraversal_CompleteUnrolling<Kernel, alignedSize, size>::run(kernel);
}
};
/**************************
*** Inner vectorization ***
**************************/
template<typename Kernel>
struct dense_assignment_loop<Kernel, InnerVectorizedTraversal, NoUnrolling>
{
typedef typename Kernel::PacketType PacketType;
enum {
SrcAlignment = Kernel::AssignmentTraits::SrcAlignment,
DstAlignment = Kernel::AssignmentTraits::DstAlignment
};
EIGEN_DEVICE_FUNC static EIGEN_STRONG_INLINE void run(Kernel &kernel)
{
const Index innerSize = kernel.innerSize();
const Index outerSize = kernel.outerSize();
const Index packetSize = unpacket_traits<PacketType>::size;
for(Index outer = 0; outer < outerSize; ++outer)
for(Index inner = 0; inner < innerSize; inner+=packetSize)
kernel.template assignPacketByOuterInner<DstAlignment, SrcAlignment, PacketType>(outer, inner);
}
};
template<typename Kernel>
struct dense_assignment_loop<Kernel, InnerVectorizedTraversal, CompleteUnrolling>
{
EIGEN_DEVICE_FUNC static EIGEN_STRONG_INLINE void run(Kernel &kernel)
{
typedef typename Kernel::DstEvaluatorType::XprType DstXprType;
copy_using_evaluator_innervec_CompleteUnrolling<Kernel, 0, DstXprType::SizeAtCompileTime>::run(kernel);
}
};
template<typename Kernel>
struct dense_assignment_loop<Kernel, InnerVectorizedTraversal, InnerUnrolling>
{
EIGEN_DEVICE_FUNC static EIGEN_STRONG_INLINE void run(Kernel &kernel)
{
typedef typename Kernel::DstEvaluatorType::XprType DstXprType;
typedef typename Kernel::AssignmentTraits Traits;
const Index outerSize = kernel.outerSize();
for(Index outer = 0; outer < outerSize; ++outer)
copy_using_evaluator_innervec_InnerUnrolling<Kernel, 0, DstXprType::InnerSizeAtCompileTime,
Traits::SrcAlignment, Traits::DstAlignment>::run(kernel, outer);
}
};
/***********************
*** Linear traversal ***
***********************/
template<typename Kernel>
struct dense_assignment_loop<Kernel, LinearTraversal, NoUnrolling>
{
EIGEN_DEVICE_FUNC static EIGEN_STRONG_INLINE void run(Kernel &kernel)
{
const Index size = kernel.size();
for(Index i = 0; i < size; ++i)
kernel.assignCoeff(i);
}
};
template<typename Kernel>
struct dense_assignment_loop<Kernel, LinearTraversal, CompleteUnrolling>
{
EIGEN_DEVICE_FUNC static EIGEN_STRONG_INLINE void run(Kernel &kernel)
{
typedef typename Kernel::DstEvaluatorType::XprType DstXprType;
copy_using_evaluator_LinearTraversal_CompleteUnrolling<Kernel, 0, DstXprType::SizeAtCompileTime>::run(kernel);
}
};
/**************************
*** Slice vectorization ***
***************************/
template<typename Kernel>
struct dense_assignment_loop<Kernel, SliceVectorizedTraversal, NoUnrolling>
{
EIGEN_DEVICE_FUNC static EIGEN_STRONG_INLINE void run(Kernel &kernel)
{
typedef typename Kernel::Scalar Scalar;
typedef typename Kernel::PacketType PacketType;
enum {
packetSize = unpacket_traits<PacketType>::size,
requestedAlignment = int(Kernel::AssignmentTraits::InnerRequiredAlignment),
alignable = packet_traits<Scalar>::AlignedOnScalar || int(Kernel::AssignmentTraits::DstAlignment)>=sizeof(Scalar),
dstIsAligned = int(Kernel::AssignmentTraits::DstAlignment)>=int(requestedAlignment),
dstAlignment = alignable ? int(requestedAlignment)
: int(Kernel::AssignmentTraits::DstAlignment)
};
const Scalar *dst_ptr = kernel.dstDataPtr();
if((!bool(dstIsAligned)) && (UIntPtr(dst_ptr) % sizeof(Scalar))>0)
{
// the pointer is not aligend-on scalar, so alignment is not possible
return dense_assignment_loop<Kernel,DefaultTraversal,NoUnrolling>::run(kernel);
}
const Index packetAlignedMask = packetSize - 1;
const Index innerSize = kernel.innerSize();
const Index outerSize = kernel.outerSize();
const Index alignedStep = alignable ? (packetSize - kernel.outerStride() % packetSize) & packetAlignedMask : 0;
Index alignedStart = ((!alignable) || bool(dstIsAligned)) ? 0 : internal::first_aligned<requestedAlignment>(dst_ptr, innerSize);
for(Index outer = 0; outer < outerSize; ++outer)
{
const Index alignedEnd = alignedStart + ((innerSize-alignedStart) & ~packetAlignedMask);
// do the non-vectorizable part of the assignment
for(Index inner = 0; inner<alignedStart ; ++inner)
kernel.assignCoeffByOuterInner(outer, inner);
// do the vectorizable part of the assignment
for(Index inner = alignedStart; inner<alignedEnd; inner+=packetSize)
kernel.template assignPacketByOuterInner<dstAlignment, Unaligned, PacketType>(outer, inner);
// do the non-vectorizable part of the assignment
for(Index inner = alignedEnd; inner<innerSize ; ++inner)
kernel.assignCoeffByOuterInner(outer, inner);
alignedStart = numext::mini((alignedStart+alignedStep)%packetSize, innerSize);
}
}
};
#if EIGEN_UNALIGNED_VECTORIZE
template<typename Kernel>
struct dense_assignment_loop<Kernel, SliceVectorizedTraversal, InnerUnrolling>
{
EIGEN_DEVICE_FUNC static EIGEN_STRONG_INLINE void run(Kernel &kernel)
{
typedef typename Kernel::DstEvaluatorType::XprType DstXprType;
typedef typename Kernel::PacketType PacketType;
enum { size = DstXprType::InnerSizeAtCompileTime,
packetSize =unpacket_traits<PacketType>::size,
vectorizableSize = (size/packetSize)*packetSize };
for(Index outer = 0; outer < kernel.outerSize(); ++outer)
{
copy_using_evaluator_innervec_InnerUnrolling<Kernel, 0, vectorizableSize, 0, 0>::run(kernel, outer);
copy_using_evaluator_DefaultTraversal_InnerUnrolling<Kernel, vectorizableSize, size>::run(kernel, outer);
}
}
};
#endif
/***************************************************************************
* Part 4 : Generic dense assignment kernel
***************************************************************************/
// This class generalize the assignment of a coefficient (or packet) from one dense evaluator
// to another dense writable evaluator.
// It is parametrized by the two evaluators, and the actual assignment functor.
// This abstraction level permits to keep the evaluation loops as simple and as generic as possible.
// One can customize the assignment using this generic dense_assignment_kernel with different
// functors, or by completely overloading it, by-passing a functor.
template<typename DstEvaluatorTypeT, typename SrcEvaluatorTypeT, typename Functor, int Version = Specialized>
class generic_dense_assignment_kernel
{
protected:
typedef typename DstEvaluatorTypeT::XprType DstXprType;
typedef typename SrcEvaluatorTypeT::XprType SrcXprType;
public:
typedef DstEvaluatorTypeT DstEvaluatorType;
typedef SrcEvaluatorTypeT SrcEvaluatorType;
typedef typename DstEvaluatorType::Scalar Scalar;
typedef copy_using_evaluator_traits<DstEvaluatorTypeT, SrcEvaluatorTypeT, Functor> AssignmentTraits;
typedef typename AssignmentTraits::PacketType PacketType;
EIGEN_DEVICE_FUNC generic_dense_assignment_kernel(DstEvaluatorType &dst, const SrcEvaluatorType &src, const Functor &func, DstXprType& dstExpr)
: m_dst(dst), m_src(src), m_functor(func), m_dstExpr(dstExpr)
{
#ifdef EIGEN_DEBUG_ASSIGN
AssignmentTraits::debug();
#endif
}
EIGEN_DEVICE_FUNC Index size() const { return m_dstExpr.size(); }
EIGEN_DEVICE_FUNC Index innerSize() const { return m_dstExpr.innerSize(); }
EIGEN_DEVICE_FUNC Index outerSize() const { return m_dstExpr.outerSize(); }
EIGEN_DEVICE_FUNC Index rows() const { return m_dstExpr.rows(); }
EIGEN_DEVICE_FUNC Index cols() const { return m_dstExpr.cols(); }
EIGEN_DEVICE_FUNC Index outerStride() const { return m_dstExpr.outerStride(); }
EIGEN_DEVICE_FUNC DstEvaluatorType& dstEvaluator() { return m_dst; }
EIGEN_DEVICE_FUNC const SrcEvaluatorType& srcEvaluator() const { return m_src; }
/// Assign src(row,col) to dst(row,col) through the assignment functor.
EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE void assignCoeff(Index row, Index col)
{
m_functor.assignCoeff(m_dst.coeffRef(row,col), m_src.coeff(row,col));
}
/// \sa assignCoeff(Index,Index)
EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE void assignCoeff(Index index)
{
m_functor.assignCoeff(m_dst.coeffRef(index), m_src.coeff(index));
}
/// \sa assignCoeff(Index,Index)
EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE void assignCoeffByOuterInner(Index outer, Index inner)
{
Index row = rowIndexByOuterInner(outer, inner);
Index col = colIndexByOuterInner(outer, inner);
assignCoeff(row, col);
}
template<int StoreMode, int LoadMode, typename PacketType>
EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE void assignPacket(Index row, Index col)
{
m_functor.template assignPacket<StoreMode>(&m_dst.coeffRef(row,col), m_src.template packet<LoadMode,PacketType>(row,col));
}
template<int StoreMode, int LoadMode, typename PacketType>
EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE void assignPacket(Index index)
{
m_functor.template assignPacket<StoreMode>(&m_dst.coeffRef(index), m_src.template packet<LoadMode,PacketType>(index));
}
template<int StoreMode, int LoadMode, typename PacketType>
EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE void assignPacketByOuterInner(Index outer, Index inner)
{
Index row = rowIndexByOuterInner(outer, inner);
Index col = colIndexByOuterInner(outer, inner);
assignPacket<StoreMode,LoadMode,PacketType>(row, col);
}
EIGEN_DEVICE_FUNC static EIGEN_STRONG_INLINE Index rowIndexByOuterInner(Index outer, Index inner)
{
typedef typename DstEvaluatorType::ExpressionTraits Traits;
return int(Traits::RowsAtCompileTime) == 1 ? 0
: int(Traits::ColsAtCompileTime) == 1 ? inner
: int(DstEvaluatorType::Flags)&RowMajorBit ? outer
: inner;
}
EIGEN_DEVICE_FUNC static EIGEN_STRONG_INLINE Index colIndexByOuterInner(Index outer, Index inner)
{
typedef typename DstEvaluatorType::ExpressionTraits Traits;
return int(Traits::ColsAtCompileTime) == 1 ? 0
: int(Traits::RowsAtCompileTime) == 1 ? inner
: int(DstEvaluatorType::Flags)&RowMajorBit ? inner
: outer;
}
EIGEN_DEVICE_FUNC const Scalar* dstDataPtr() const
{
return m_dstExpr.data();
}
protected:
DstEvaluatorType& m_dst;
const SrcEvaluatorType& m_src;
const Functor &m_functor;
// TODO find a way to avoid the needs of the original expression
DstXprType& m_dstExpr;
};
/***************************************************************************
* Part 5 : Entry point for dense rectangular assignment
***************************************************************************/
template<typename DstXprType,typename SrcXprType, typename Functor>
EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
void resize_if_allowed(DstXprType &dst, const SrcXprType& src, const Functor &/*func*/)
{
EIGEN_ONLY_USED_FOR_DEBUG(dst);
EIGEN_ONLY_USED_FOR_DEBUG(src);
eigen_assert(dst.rows() == src.rows() && dst.cols() == src.cols());
}
template<typename DstXprType,typename SrcXprType, typename T1, typename T2>
EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
void resize_if_allowed(DstXprType &dst, const SrcXprType& src, const internal::assign_op<T1,T2> &/*func*/)
{
Index dstRows = src.rows();
Index dstCols = src.cols();
if(((dst.rows()!=dstRows) || (dst.cols()!=dstCols)))
dst.resize(dstRows, dstCols);
eigen_assert(dst.rows() == dstRows && dst.cols() == dstCols);
}
template<typename DstXprType, typename SrcXprType, typename Functor>
EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE void call_dense_assignment_loop(DstXprType& dst, const SrcXprType& src, const Functor &func)
{
typedef evaluator<DstXprType> DstEvaluatorType;
typedef evaluator<SrcXprType> SrcEvaluatorType;
SrcEvaluatorType srcEvaluator(src);
// NOTE To properly handle A = (A*A.transpose())/s with A rectangular,
// we need to resize the destination after the source evaluator has been created.
resize_if_allowed(dst, src, func);
DstEvaluatorType dstEvaluator(dst);
typedef generic_dense_assignment_kernel<DstEvaluatorType,SrcEvaluatorType,Functor> Kernel;
Kernel kernel(dstEvaluator, srcEvaluator, func, dst.const_cast_derived());
dense_assignment_loop<Kernel>::run(kernel);
}
template<typename DstXprType, typename SrcXprType>
EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE void call_dense_assignment_loop(DstXprType& dst, const SrcXprType& src)
{
call_dense_assignment_loop(dst, src, internal::assign_op<typename DstXprType::Scalar,typename SrcXprType::Scalar>());
}
/***************************************************************************
* Part 6 : Generic assignment
***************************************************************************/
// Based on the respective shapes of the destination and source,
// the class AssignmentKind determine the kind of assignment mechanism.
// AssignmentKind must define a Kind typedef.
template<typename DstShape, typename SrcShape> struct AssignmentKind;
// Assignement kind defined in this file:
struct Dense2Dense {};
struct EigenBase2EigenBase {};
template<typename,typename> struct AssignmentKind { typedef EigenBase2EigenBase Kind; };
template<> struct AssignmentKind<DenseShape,DenseShape> { typedef Dense2Dense Kind; };
// This is the main assignment class
template< typename DstXprType, typename SrcXprType, typename Functor,
typename Kind = typename AssignmentKind< typename evaluator_traits<DstXprType>::Shape , typename evaluator_traits<SrcXprType>::Shape >::Kind,
typename EnableIf = void>
struct Assignment;
// The only purpose of this call_assignment() function is to deal with noalias() / "assume-aliasing" and automatic transposition.
// Indeed, I (Gael) think that this concept of "assume-aliasing" was a mistake, and it makes thing quite complicated.
// So this intermediate function removes everything related to "assume-aliasing" such that Assignment
// does not has to bother about these annoying details.
template<typename Dst, typename Src>
EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
void call_assignment(Dst& dst, const Src& src)
{
call_assignment(dst, src, internal::assign_op<typename Dst::Scalar,typename Src::Scalar>());
}
template<typename Dst, typename Src>
EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
void call_assignment(const Dst& dst, const Src& src)
{
call_assignment(dst, src, internal::assign_op<typename Dst::Scalar,typename Src::Scalar>());
}
// Deal with "assume-aliasing"
template<typename Dst, typename Src, typename Func>
EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
void call_assignment(Dst& dst, const Src& src, const Func& func, typename enable_if< evaluator_assume_aliasing<Src>::value, void*>::type = 0)
{
typename plain_matrix_type<Src>::type tmp(src);
call_assignment_no_alias(dst, tmp, func);
}
template<typename Dst, typename Src, typename Func>
EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
void call_assignment(Dst& dst, const Src& src, const Func& func, typename enable_if<!evaluator_assume_aliasing<Src>::value, void*>::type = 0)
{
call_assignment_no_alias(dst, src, func);
}
// by-pass "assume-aliasing"
// When there is no aliasing, we require that 'dst' has been properly resized
template<typename Dst, template <typename> class StorageBase, typename Src, typename Func>
EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
void call_assignment(NoAlias<Dst,StorageBase>& dst, const Src& src, const Func& func)
{
call_assignment_no_alias(dst.expression(), src, func);
}
template<typename Dst, typename Src, typename Func>
EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
void call_assignment_no_alias(Dst& dst, const Src& src, const Func& func)
{
enum {
NeedToTranspose = ( (int(Dst::RowsAtCompileTime) == 1 && int(Src::ColsAtCompileTime) == 1)
|| (int(Dst::ColsAtCompileTime) == 1 && int(Src::RowsAtCompileTime) == 1)
) && int(Dst::SizeAtCompileTime) != 1
};
typedef typename internal::conditional<NeedToTranspose, Transpose<Dst>, Dst>::type ActualDstTypeCleaned;
typedef typename internal::conditional<NeedToTranspose, Transpose<Dst>, Dst&>::type ActualDstType;
ActualDstType actualDst(dst);
// TODO check whether this is the right place to perform these checks:
EIGEN_STATIC_ASSERT_LVALUE(Dst)
EIGEN_STATIC_ASSERT_SAME_MATRIX_SIZE(ActualDstTypeCleaned,Src)
EIGEN_CHECK_BINARY_COMPATIBILIY(Func,typename ActualDstTypeCleaned::Scalar,typename Src::Scalar);
Assignment<ActualDstTypeCleaned,Src,Func>::run(actualDst, src, func);
}
template<typename Dst, typename Src>
EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
void call_assignment_no_alias(Dst& dst, const Src& src)
{
call_assignment_no_alias(dst, src, internal::assign_op<typename Dst::Scalar,typename Src::Scalar>());
}
template<typename Dst, typename Src, typename Func>
EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
void call_assignment_no_alias_no_transpose(Dst& dst, const Src& src, const Func& func)
{
// TODO check whether this is the right place to perform these checks:
EIGEN_STATIC_ASSERT_LVALUE(Dst)
EIGEN_STATIC_ASSERT_SAME_MATRIX_SIZE(Dst,Src)
EIGEN_CHECK_BINARY_COMPATIBILIY(Func,typename Dst::Scalar,typename Src::Scalar);
Assignment<Dst,Src,Func>::run(dst, src, func);
}
template<typename Dst, typename Src>
EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
void call_assignment_no_alias_no_transpose(Dst& dst, const Src& src)
{
call_assignment_no_alias_no_transpose(dst, src, internal::assign_op<typename Dst::Scalar,typename Src::Scalar>());
}
// forward declaration
template<typename Dst, typename Src> void check_for_aliasing(const Dst &dst, const Src &src);
// Generic Dense to Dense assignment
// Note that the last template argument "Weak" is needed to make it possible to perform
// both partial specialization+SFINAE without ambiguous specialization
template< typename DstXprType, typename SrcXprType, typename Functor, typename Weak>
struct Assignment<DstXprType, SrcXprType, Functor, Dense2Dense, Weak>
{
EIGEN_DEVICE_FUNC
static EIGEN_STRONG_INLINE void run(DstXprType &dst, const SrcXprType &src, const Functor &func)
{
#ifndef EIGEN_NO_DEBUG
internal::check_for_aliasing(dst, src);
#endif
call_dense_assignment_loop(dst, src, func);
}
};
// Generic assignment through evalTo.
// TODO: not sure we have to keep that one, but it helps porting current code to new evaluator mechanism.
// Note that the last template argument "Weak" is needed to make it possible to perform
// both partial specialization+SFINAE without ambiguous specialization
template< typename DstXprType, typename SrcXprType, typename Functor, typename Weak>
struct Assignment<DstXprType, SrcXprType, Functor, EigenBase2EigenBase, Weak>
{
EIGEN_DEVICE_FUNC
static EIGEN_STRONG_INLINE 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);
eigen_assert(dst.rows() == src.rows() && dst.cols() == src.cols());
src.evalTo(dst);
}
// NOTE The following two functions are templated to avoid their instanciation if not needed
// This is needed because some expressions supports evalTo only and/or have 'void' as scalar type.
template<typename SrcScalarType>
EIGEN_DEVICE_FUNC
static EIGEN_STRONG_INLINE void run(DstXprType &dst, const SrcXprType &src, const internal::add_assign_op<typename DstXprType::Scalar,SrcScalarType> &/*func*/)
{
Index dstRows = src.rows();
Index dstCols = src.cols();
if((dst.rows()!=dstRows) || (dst.cols()!=dstCols))
dst.resize(dstRows, dstCols);
eigen_assert(dst.rows() == src.rows() && dst.cols() == src.cols());
src.addTo(dst);
}
template<typename SrcScalarType>
EIGEN_DEVICE_FUNC
static EIGEN_STRONG_INLINE void run(DstXprType &dst, const SrcXprType &src, const internal::sub_assign_op<typename DstXprType::Scalar,SrcScalarType> &/*func*/)
{
Index dstRows = src.rows();
Index dstCols = src.cols();
if((dst.rows()!=dstRows) || (dst.cols()!=dstCols))
dst.resize(dstRows, dstCols);
eigen_assert(dst.rows() == src.rows() && dst.cols() == src.cols());
src.subTo(dst);
}
};
} // namespace internal
} // end namespace Eigen
#endif // EIGEN_ASSIGN_EVALUATOR_H
external/eigen3/Eigen/src/Core/Assign_MKL.h 100644 → 100755
View file @ a394b22a
/*
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:
......@@ -37,17 +38,13 @@ namespace Eigen {
namespace internal {
template<typename Op> struct vml_call
{ enum { IsSupported = 0 }; };
template<typename Dst, typename Src, typename UnaryOp>
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)
......@@ -57,165 +54,120 @@ class vml_assign_traits
: int(Dst::MaxRowsAtCompileTime),
MaxSizeAtCompileTime = Dst::SizeAtCompileTime,
MightEnableVml = vml_call<UnaryOp>::IsSupported && StorageOrdersAgree && DstHasDirectAccess && SrcHasDirectAccess
&& Src::InnerStrideAtCompileTime==1 && Dst::InnerStrideAtCompileTime==1,
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,
MayEnableVml = MightEnableVml && LargeEnough,
MayLinearize = MayEnableVml && MightLinearize
LargeEnough = VmlSize==Dynamic || VmlSize>=EIGEN_MKL_VML_THRESHOLD
};
public:
enum {
Traversal = MayLinearize ? LinearVectorizedTraversal
: MayEnableVml ? InnerVectorizedTraversal
: DefaultTraversal
EnableVml = MightEnableVml && LargeEnough,
Traversal = MightLinearize ? LinearTraversal : DefaultTraversal
};
};
template<typename Derived1, typename Derived2, typename UnaryOp, int Traversal, int Unrolling,
int VmlTraversal = vml_assign_traits<Derived1, Derived2, UnaryOp>::Traversal >
struct vml_assign_impl
: assign_impl<Derived1, Eigen::CwiseUnaryOp<UnaryOp, Derived2>,Traversal,Unrolling,BuiltIn>
{
};
template<typename Derived1, typename Derived2, typename UnaryOp, int Traversal, int Unrolling>
struct vml_assign_impl<Derived1, Derived2, UnaryOp, Traversal, Unrolling, InnerVectorizedTraversal>
{
typedef typename Derived1::Scalar Scalar;
typedef typename Derived1::Index Index;
static inline void run(Derived1& dst, const CwiseUnaryOp<UnaryOp, Derived2>& src)
{
// in case we want to (or have to) skip VML at runtime we can call:
// assign_impl<Derived1,Eigen::CwiseUnaryOp<UnaryOp, Derived2>,Traversal,Unrolling,BuiltIn>::run(dst,src);
const Index innerSize = dst.innerSize();
const Index outerSize = dst.outerSize();
for(Index outer = 0; outer < outerSize; ++outer) {
const Scalar *src_ptr = src.IsRowMajor ? &(src.nestedExpression().coeffRef(outer,0)) :
&(src.nestedExpression().coeffRef(0, outer));
Scalar *dst_ptr = dst.IsRowMajor ? &(dst.coeffRef(outer,0)) : &(dst.coeffRef(0, outer));
vml_call<UnaryOp>::run(src.functor(), innerSize, src_ptr, dst_ptr );
}
}
};
template<typename Derived1, typename Derived2, typename UnaryOp, int Traversal, int Unrolling>
struct vml_assign_impl<Derived1, Derived2, UnaryOp, Traversal, Unrolling, LinearVectorizedTraversal>
{
static inline void run(Derived1& dst, const CwiseUnaryOp<UnaryOp, Derived2>& src)
{
// in case we want to (or have to) skip VML at runtime we can call:
// assign_impl<Derived1,Eigen::CwiseUnaryOp<UnaryOp, Derived2>,Traversal,Unrolling,BuiltIn>::run(dst,src);
vml_call<UnaryOp>::run(src.functor(), dst.size(), src.nestedExpression().data(), dst.data() );
}
};
// Macroses
#define EIGEN_MKL_VML_SPECIALIZE_ASSIGN(TRAVERSAL,UNROLLING) \
template<typename Derived1, typename Derived2, typename UnaryOp> \
struct assign_impl<Derived1, Eigen::CwiseUnaryOp<UnaryOp, Derived2>, TRAVERSAL, UNROLLING, Specialized> { \
static inline void run(Derived1 &dst, const Eigen::CwiseUnaryOp<UnaryOp, Derived2> &src) { \
vml_assign_impl<Derived1,Derived2,UnaryOp,TRAVERSAL,UNROLLING>::run(dst, src); \
} \
};
EIGEN_MKL_VML_SPECIALIZE_ASSIGN(DefaultTraversal,NoUnrolling)
EIGEN_MKL_VML_SPECIALIZE_ASSIGN(DefaultTraversal,CompleteUnrolling)
EIGEN_MKL_VML_SPECIALIZE_ASSIGN(DefaultTraversal,InnerUnrolling)
EIGEN_MKL_VML_SPECIALIZE_ASSIGN(LinearTraversal,NoUnrolling)
EIGEN_MKL_VML_SPECIALIZE_ASSIGN(LinearTraversal,CompleteUnrolling)
EIGEN_MKL_VML_SPECIALIZE_ASSIGN(InnerVectorizedTraversal,NoUnrolling)
EIGEN_MKL_VML_SPECIALIZE_ASSIGN(InnerVectorizedTraversal,CompleteUnrolling)
EIGEN_MKL_VML_SPECIALIZE_ASSIGN(InnerVectorizedTraversal,InnerUnrolling)
EIGEN_MKL_VML_SPECIALIZE_ASSIGN(LinearVectorizedTraversal,CompleteUnrolling)
EIGEN_MKL_VML_SPECIALIZE_ASSIGN(LinearVectorizedTraversal,NoUnrolling)
EIGEN_MKL_VML_SPECIALIZE_ASSIGN(SliceVectorizedTraversal,NoUnrolling)
#define EIGEN_PP_EXPAND(ARG) ARG
#if !defined (EIGEN_FAST_MATH) || (EIGEN_FAST_MATH != 1)
#define EIGEN_MKL_VML_MODE VML_HA
#define EIGEN_VMLMODE_EXPAND_LA , VML_HA
#else
#define EIGEN_MKL_VML_MODE VML_LA
#define EIGEN_VMLMODE_EXPAND_LA , VML_LA
#endif
#define EIGEN_MKL_VML_DECLARE_UNARY_CALL(EIGENOP, VMLOP, EIGENTYPE, VMLTYPE) \
template<> struct vml_call< scalar_##EIGENOP##_op<EIGENTYPE> > { \
enum { IsSupported = 1 }; \
static inline void run( const scalar_##EIGENOP##_op<EIGENTYPE>& /*func*/, \
int size, const EIGENTYPE* src, EIGENTYPE* dst) { \
VMLOP(size, (const VMLTYPE*)src, (VMLTYPE*)dst); \
} \
#define EIGEN_VMLMODE_EXPAND__
#define EIGEN_VMLMODE_PREFIX_LA vm
#define EIGEN_VMLMODE_PREFIX__ v
#define EIGEN_VMLMODE_PREFIX(VMLMODE) EIGEN_CAT(EIGEN_VMLMODE_PREFIX_,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*/) { \
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_##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_##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*/) { \
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_##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_##VMLMODE)); \
} \
} \
} \
};
#define EIGEN_MKL_VML_DECLARE_UNARY_CALL_LA(EIGENOP, VMLOP, EIGENTYPE, VMLTYPE) \
template<> struct vml_call< scalar_##EIGENOP##_op<EIGENTYPE> > { \
enum { IsSupported = 1 }; \
static inline void run( const scalar_##EIGENOP##_op<EIGENTYPE>& /*func*/, \
int size, const EIGENTYPE* src, EIGENTYPE* dst) { \
MKL_INT64 vmlMode = EIGEN_MKL_VML_MODE; \
VMLOP(size, (const VMLTYPE*)src, (VMLTYPE*)dst, vmlMode); \
} \
};
#define EIGEN_MKL_VML_DECLARE_POW_CALL(EIGENOP, VMLOP, EIGENTYPE, VMLTYPE) \
template<> struct vml_call< scalar_##EIGENOP##_op<EIGENTYPE> > { \
enum { IsSupported = 1 }; \
static inline void run( const scalar_##EIGENOP##_op<EIGENTYPE>& func, \
int size, const EIGENTYPE* src, EIGENTYPE* dst) { \
EIGENTYPE exponent = func.m_exponent; \
MKL_INT64 vmlMode = EIGEN_MKL_VML_MODE; \
VMLOP(&size, (const VMLTYPE*)src, (const VMLTYPE*)&exponent, \
(VMLTYPE*)dst, &vmlMode); \
} \
};
#define EIGEN_MKL_VML_DECLARE_UNARY_CALLS_REAL(EIGENOP, VMLOP) \
EIGEN_MKL_VML_DECLARE_UNARY_CALL(EIGENOP, vs##VMLOP, float, float) \
EIGEN_MKL_VML_DECLARE_UNARY_CALL(EIGENOP, vd##VMLOP, double, double)
#define EIGEN_MKL_VML_DECLARE_UNARY_CALLS_COMPLEX(EIGENOP, VMLOP) \
EIGEN_MKL_VML_DECLARE_UNARY_CALL(EIGENOP, vc##VMLOP, scomplex, MKL_Complex8) \
EIGEN_MKL_VML_DECLARE_UNARY_CALL(EIGENOP, vz##VMLOP, dcomplex, MKL_Complex16)
#define EIGEN_MKL_VML_DECLARE_UNARY_CALLS(EIGENOP, VMLOP) \
EIGEN_MKL_VML_DECLARE_UNARY_CALLS_REAL(EIGENOP, VMLOP) \
EIGEN_MKL_VML_DECLARE_UNARY_CALLS_COMPLEX(EIGENOP, VMLOP)
#define EIGEN_MKL_VML_DECLARE_UNARY_CALLS_REAL_LA(EIGENOP, VMLOP) \
EIGEN_MKL_VML_DECLARE_UNARY_CALL_LA(EIGENOP, vms##VMLOP, float, float) \
EIGEN_MKL_VML_DECLARE_UNARY_CALL_LA(EIGENOP, vmd##VMLOP, double, double)
#define EIGEN_MKL_VML_DECLARE_UNARY_CALLS_COMPLEX_LA(EIGENOP, VMLOP) \
EIGEN_MKL_VML_DECLARE_UNARY_CALL_LA(EIGENOP, vmc##VMLOP, scomplex, MKL_Complex8) \
EIGEN_MKL_VML_DECLARE_UNARY_CALL_LA(EIGENOP, vmz##VMLOP, dcomplex, MKL_Complex16)
#define EIGEN_MKL_VML_DECLARE_UNARY_CALLS_LA(EIGENOP, VMLOP) \
EIGEN_MKL_VML_DECLARE_UNARY_CALLS_REAL_LA(EIGENOP, VMLOP) \
EIGEN_MKL_VML_DECLARE_UNARY_CALLS_COMPLEX_LA(EIGENOP, VMLOP)
EIGEN_MKL_VML_DECLARE_UNARY_CALLS_LA(sin, Sin)
EIGEN_MKL_VML_DECLARE_UNARY_CALLS_LA(asin, Asin)
EIGEN_MKL_VML_DECLARE_UNARY_CALLS_LA(cos, Cos)
EIGEN_MKL_VML_DECLARE_UNARY_CALLS_LA(acos, Acos)
EIGEN_MKL_VML_DECLARE_UNARY_CALLS_LA(tan, Tan)
//EIGEN_MKL_VML_DECLARE_UNARY_CALLS(abs, Abs)
EIGEN_MKL_VML_DECLARE_UNARY_CALLS_LA(exp, Exp)
EIGEN_MKL_VML_DECLARE_UNARY_CALLS_LA(log, Ln)
EIGEN_MKL_VML_DECLARE_UNARY_CALLS_LA(sqrt, Sqrt)
EIGEN_MKL_VML_DECLARE_UNARY_CALLS_REAL(square, Sqr)
// The vm*powx functions are not avaibale in the windows version of MKL.
#ifndef _WIN32
EIGEN_MKL_VML_DECLARE_POW_CALL(pow, vmspowx_, float, float)
EIGEN_MKL_VML_DECLARE_POW_CALL(pow, vmdpowx_, double, double)
EIGEN_MKL_VML_DECLARE_POW_CALL(pow, vmcpowx_, scomplex, MKL_Complex8)
EIGEN_MKL_VML_DECLARE_POW_CALL(pow, vmzpowx_, dcomplex, MKL_Complex16)
#endif
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
......
external/eigen3/Eigen/src/Core/BandMatrix.h
View file @ a394b22a
......@@ -32,7 +32,7 @@ class BandMatrixBase : public EigenBase<Derived>
};
typedef typename internal::traits<Derived>::Scalar Scalar;
typedef Matrix<Scalar,RowsAtCompileTime,ColsAtCompileTime> DenseMatrixType;
typedef typename DenseMatrixType::Index Index;
typedef typename DenseMatrixType::StorageIndex StorageIndex;
typedef typename internal::traits<Derived>::CoefficientsType CoefficientsType;
typedef EigenBase<Derived> Base;
......@@ -161,15 +161,15 @@ class BandMatrixBase : public EigenBase<Derived>
*
* \brief Represents a rectangular matrix with a banded storage
*
* \param _Scalar Numeric type, i.e. float, double, int
* \param Rows Number of rows, or \b Dynamic
* \param Cols Number of columns, or \b Dynamic
* \param Supers Number of super diagonal
* \param Subs Number of sub diagonal
* \param _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.
* \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
*/
......@@ -179,7 +179,7 @@ struct traits<BandMatrix<_Scalar,_Rows,_Cols,_Supers,_Subs,_Options> >
{
typedef _Scalar Scalar;
typedef Dense StorageKind;
typedef DenseIndex Index;
typedef Eigen::Index StorageIndex;
enum {
CoeffReadCost = NumTraits<Scalar>::ReadCost,
RowsAtCompileTime = _Rows,
......@@ -201,10 +201,10 @@ class BandMatrix : public BandMatrixBase<BandMatrix<_Scalar,Rows,Cols,Supers,Sub
public:
typedef typename internal::traits<BandMatrix>::Scalar Scalar;
typedef typename internal::traits<BandMatrix>::Index Index;
typedef typename internal::traits<BandMatrix>::StorageIndex StorageIndex;
typedef typename internal::traits<BandMatrix>::CoefficientsType CoefficientsType;
inline BandMatrix(Index rows=Rows, Index cols=Cols, Index supers=Supers, Index subs=Subs)
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)
{
......@@ -241,7 +241,7 @@ struct traits<BandMatrixWrapper<_CoefficientsType,_Rows,_Cols,_Supers,_Subs,_Opt
{
typedef typename _CoefficientsType::Scalar Scalar;
typedef typename _CoefficientsType::StorageKind StorageKind;
typedef typename _CoefficientsType::Index Index;
typedef typename _CoefficientsType::StorageIndex StorageIndex;
enum {
CoeffReadCost = internal::traits<_CoefficientsType>::CoeffReadCost,
RowsAtCompileTime = _Rows,
......@@ -264,9 +264,9 @@ class BandMatrixWrapper : public BandMatrixBase<BandMatrixWrapper<_CoefficientsT
typedef typename internal::traits<BandMatrixWrapper>::Scalar Scalar;
typedef typename internal::traits<BandMatrixWrapper>::CoefficientsType CoefficientsType;
typedef typename internal::traits<BandMatrixWrapper>::Index Index;
typedef typename internal::traits<BandMatrixWrapper>::StorageIndex StorageIndex;
inline BandMatrixWrapper(const CoefficientsType& coeffs, Index rows=_Rows, Index cols=_Cols, Index supers=_Supers, Index subs=_Subs)
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)
{
......@@ -302,9 +302,9 @@ class BandMatrixWrapper : public BandMatrixBase<BandMatrixWrapper<_CoefficientsT
*
* \brief Represents a tridiagonal matrix with a compact banded storage
*
* \param _Scalar Numeric type, i.e. float, double, int
* \param Size Number of rows and cols, or \b Dynamic
* \param _Options Can be 0 or \b SelfAdjoint
* \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
*/
......@@ -312,9 +312,9 @@ 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::Index Index;
typedef typename Base::StorageIndex StorageIndex;
public:
TridiagonalMatrix(Index size = Size) : Base(size,size,Options&SelfAdjoint?0:1,1) {}
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>(); }
......@@ -327,6 +327,25 @@ class TridiagonalMatrix : public BandMatrix<Scalar,Size,Size,Options&SelfAdjoint
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
......
external/eigen3/Eigen/src/Core/Block.h
View file @ a394b22a
......@@ -13,38 +13,6 @@
namespace Eigen {
/** \class Block
* \ingroup Core_Module
*
* \brief Expression of a fixed-size or dynamic-size block
*
* \param XprType the type of the expression in which we are taking a block
* \param BlockRows the number of rows of the block we are taking at compile time (optional)
* \param BlockCols the number of columns of the block we are taking at compile time (optional)
*
* 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
*/
namespace internal {
template<typename XprType, int BlockRows, int BlockCols, bool InnerPanel>
struct traits<Block<XprType, BlockRows, BlockCols, InnerPanel> > : traits<XprType>
......@@ -52,7 +20,7 @@ struct traits<Block<XprType, BlockRows, BlockCols, InnerPanel> > : traits<XprTyp
typedef typename traits<XprType>::Scalar Scalar;
typedef typename traits<XprType>::StorageKind StorageKind;
typedef typename traits<XprType>::XprKind XprKind;
typedef typename nested<XprType>::type XprTypeNested;
typedef typename ref_selector<XprType>::type XprTypeNested;
typedef typename remove_reference<XprTypeNested>::type _XprTypeNested;
enum{
MatrixRows = traits<XprType>::RowsAtCompileTime,
......@@ -65,10 +33,10 @@ struct traits<Block<XprType, BlockRows, BlockCols, InnerPanel> > : traits<XprTyp
MaxColsAtCompileTime = BlockCols==0 ? 0
: ColsAtCompileTime != Dynamic ? int(ColsAtCompileTime)
: int(traits<XprType>::MaxColsAtCompileTime),
XprTypeIsRowMajor = (int(traits<XprType>::Flags)&RowMajorBit) != 0,
IsDense = is_same<StorageKind,Dense>::value,
IsRowMajor = (IsDense&&MaxRowsAtCompileTime==1&&MaxColsAtCompileTime!=1) ? 1
: (IsDense&&MaxColsAtCompileTime==1&&MaxRowsAtCompileTime!=1) ? 0
IsRowMajor = (MaxRowsAtCompileTime==1&&MaxColsAtCompileTime!=1) ? 1
: (MaxColsAtCompileTime==1&&MaxRowsAtCompileTime!=1) ? 0
: XprTypeIsRowMajor,
HasSameStorageOrderAsXprType = (IsRowMajor == XprTypeIsRowMajor),
InnerSize = IsRowMajor ? int(ColsAtCompileTime) : int(RowsAtCompileTime),
......@@ -78,18 +46,16 @@ struct traits<Block<XprType, BlockRows, BlockCols, InnerPanel> > : traits<XprTyp
OuterStrideAtCompileTime = HasSameStorageOrderAsXprType
? int(outer_stride_at_compile_time<XprType>::ret)
: int(inner_stride_at_compile_time<XprType>::ret),
MaskPacketAccessBit = (InnerSize == Dynamic || (InnerSize % packet_traits<Scalar>::size) == 0)
&& (InnerStrideAtCompileTime == 1)
? PacketAccessBit : 0,
MaskAlignedBit = (InnerPanel && (OuterStrideAtCompileTime!=Dynamic) && (((OuterStrideAtCompileTime * int(sizeof(Scalar))) % 16) == 0)) ? AlignedBit : 0,
FlagsLinearAccessBit = (RowsAtCompileTime == 1 || ColsAtCompileTime == 1 || (InnerPanel && (traits<XprType>::Flags&LinearAccessBit))) ? LinearAccessBit : 0,
// 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,
Flags0 = traits<XprType>::Flags & ( (HereditaryBits & ~RowMajorBit) |
DirectAccessBit |
MaskPacketAccessBit |
MaskAlignedBit),
Flags = Flags0 | FlagsLinearAccessBit | FlagsLvalueBit | FlagsRowMajorBit
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
};
};
......@@ -100,6 +66,40 @@ template<typename XprType, int BlockRows=Dynamic, int BlockCols=Dynamic, bool In
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>
{
......@@ -109,9 +109,12 @@ template<typename XprType, int BlockRows, int BlockCols, bool InnerPanel> class
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
inline Block(XprType& xpr, Index i) : Impl(xpr,i)
{
eigen_assert( (i>=0) && (
......@@ -121,25 +124,27 @@ template<typename XprType, int BlockRows, int BlockCols, bool InnerPanel> class
/** Fixed-size constructor
*/
inline Block(XprType& xpr, Index a_startRow, Index a_startCol)
: Impl(xpr, a_startRow, a_startCol)
EIGEN_DEVICE_FUNC
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(a_startRow >= 0 && BlockRows >= 1 && a_startRow + BlockRows <= xpr.rows()
&& a_startCol >= 0 && BlockCols >= 1 && a_startCol + BlockCols <= xpr.cols());
eigen_assert(startRow >= 0 && BlockRows >= 0 && startRow + BlockRows <= xpr.rows()
&& startCol >= 0 && BlockCols >= 0 && startCol + BlockCols <= xpr.cols());
}
/** Dynamic-size constructor
*/
EIGEN_DEVICE_FUNC
inline Block(XprType& xpr,
Index a_startRow, Index a_startCol,
Index startRow, Index startCol,
Index blockRows, Index blockCols)
: Impl(xpr, a_startRow, a_startCol, blockRows, blockCols)
: Impl(xpr, startRow, startCol, blockRows, blockCols)
{
eigen_assert((RowsAtCompileTime==Dynamic || RowsAtCompileTime==blockRows)
&& (ColsAtCompileTime==Dynamic || ColsAtCompileTime==blockCols));
eigen_assert(a_startRow >= 0 && blockRows >= 0 && a_startRow <= xpr.rows() - blockRows
&& a_startCol >= 0 && blockCols >= 0 && a_startCol <= xpr.cols() - blockCols);
eigen_assert(startRow >= 0 && blockRows >= 0 && startRow <= xpr.rows() - blockRows
&& startCol >= 0 && blockCols >= 0 && startCol <= xpr.cols() - blockCols);
}
};
......@@ -150,14 +155,15 @@ 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::Index Index;
typedef typename XprType::StorageIndex StorageIndex;
public:
typedef Impl Base;
EIGEN_INHERIT_ASSIGNMENT_OPERATORS(BlockImpl)
inline BlockImpl(XprType& xpr, Index i) : Impl(xpr,i) {}
inline BlockImpl(XprType& xpr, Index a_startRow, Index a_startCol) : Impl(xpr, a_startRow, a_startCol) {}
inline BlockImpl(XprType& xpr, Index a_startRow, Index a_startCol, Index blockRows, Index blockCols)
: Impl(xpr, a_startRow, a_startCol, blockRows, blockCols) {}
EIGEN_DEVICE_FUNC inline BlockImpl(XprType& xpr, Index i) : Impl(xpr,i) {}
EIGEN_DEVICE_FUNC inline BlockImpl(XprType& xpr, Index startRow, Index startCol) : Impl(xpr, startRow, startCol) {}
EIGEN_DEVICE_FUNC
inline BlockImpl(XprType& xpr, Index startRow, Index startCol, Index blockRows, Index blockCols)
: Impl(xpr, startRow, startCol, blockRows, blockCols) {}
};
namespace internal {
......@@ -167,16 +173,18 @@ template<typename XprType, int BlockRows, int BlockCols, bool InnerPanel, bool H
: 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;
// 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,
......@@ -191,75 +199,76 @@ template<typename XprType, int BlockRows, int BlockCols, bool InnerPanel, bool H
/** Fixed-size constructor
*/
inline BlockImpl_dense(XprType& xpr, Index a_startRow, Index a_startCol)
: m_xpr(xpr), m_startRow(a_startRow), m_startCol(a_startCol),
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 a_startRow, Index a_startCol,
Index startRow, Index startCol,
Index blockRows, Index blockCols)
: m_xpr(xpr), m_startRow(a_startRow), m_startCol(a_startCol),
: m_xpr(xpr), m_startRow(startRow), m_startCol(startCol),
m_blockRows(blockRows), m_blockCols(blockCols)
{}
inline Index rows() const { return m_blockRows.value(); }
inline Index cols() const { return m_blockCols.value(); }
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.const_cast_derived()
.coeffRef(rowId + m_startRow.value(), colId + m_startCol.value());
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());
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.const_cast_derived()
.coeffRef(m_startRow.value() + (RowsAtCompileTime == 1 ? 0 : index),
m_startCol.value() + (RowsAtCompileTime == 1 ? index : 0));
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.const_cast_derived()
.coeffRef(m_startRow.value() + (RowsAtCompileTime == 1 ? 0 : index),
m_startCol.value() + (RowsAtCompileTime == 1 ? index : 0));
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));
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());
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.const_cast_derived().template writePacket<Unaligned>
(rowId + m_startRow.value(), colId + m_startCol.value(), val);
m_xpr.template writePacket<Unaligned>(rowId + m_startRow.value(), colId + m_startCol.value(), val);
}
template<int LoadMode>
......@@ -273,40 +282,46 @@ template<typename XprType, int BlockRows, int BlockCols, bool InnerPanel, bool H
template<int LoadMode>
inline void writePacket(Index index, const PacketScalar& val)
{
m_xpr.const_cast_derived().template writePacket<Unaligned>
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() */
inline const Scalar* data() const;
inline Index innerStride() const;
inline Index outerStride() const;
EIGEN_DEVICE_FUNC inline const Scalar* data() const;
EIGEN_DEVICE_FUNC inline Index innerStride() const;
EIGEN_DEVICE_FUNC inline Index outerStride() const;
#endif
const typename internal::remove_all<typename XprType::Nested>::type& nestedExpression() const
EIGEN_DEVICE_FUNC
const typename internal::remove_all<XprTypeNested>::type& nestedExpression() const
{
return m_xpr;
}
EIGEN_DEVICE_FUNC
XprType& nestedExpression() { return m_xpr; }
Index startRow() const
EIGEN_DEVICE_FUNC
StorageIndex startRow() const
{
return m_startRow.value();
}
Index startCol() const
EIGEN_DEVICE_FUNC
StorageIndex startCol() const
{
return m_startCol.value();
}
protected:
const typename XprType::Nested m_xpr;
const internal::variable_if_dynamic<Index, XprType::RowsAtCompileTime == 1 ? 0 : Dynamic> m_startRow;
const internal::variable_if_dynamic<Index, XprType::ColsAtCompileTime == 1 ? 0 : Dynamic> m_startCol;
const internal::variable_if_dynamic<Index, RowsAtCompileTime> m_blockRows;
const internal::variable_if_dynamic<Index, ColsAtCompileTime> m_blockCols;
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.*/
......@@ -315,6 +330,10 @@ 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;
......@@ -323,42 +342,52 @@ class BlockImpl_dense<XprType,BlockRows,BlockCols, InnerPanel,true>
/** Column or Row constructor
*/
EIGEN_DEVICE_FUNC
inline BlockImpl_dense(XprType& xpr, Index i)
: Base(internal::const_cast_ptr(&xpr.coeffRef(
(BlockRows==1) && (BlockCols==XprType::ColsAtCompileTime) ? i : 0,
(BlockRows==XprType::RowsAtCompileTime) && (BlockCols==1) ? i : 0)),
: 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_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
inline BlockImpl_dense(XprType& xpr, Index startRow, Index startCol)
: Base(internal::const_cast_ptr(&xpr.coeffRef(startRow,startCol))), m_xpr(xpr)
: 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
inline BlockImpl_dense(XprType& xpr,
Index startRow, Index startCol,
Index blockRows, Index blockCols)
: Base(internal::const_cast_ptr(&xpr.coeffRef(startRow,startCol)), blockRows, blockCols),
m_xpr(xpr)
: 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();
}
const typename internal::remove_all<typename XprType::Nested>::type& nestedExpression() const
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
inline Index innerStride() const
{
return internal::traits<BlockType>::HasSameStorageOrderAsXprType
......@@ -367,11 +396,24 @@ class BlockImpl_dense<XprType,BlockRows,BlockCols, InnerPanel,true>
}
/** \sa MapBase::outerStride() */
EIGEN_DEVICE_FUNC
inline Index outerStride() const
{
return m_outerStride;
}
EIGEN_DEVICE_FUNC
StorageIndex startRow() const
{
return m_startRow.value();
}
EIGEN_DEVICE_FUNC
StorageIndex startCol() const
{
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?
......@@ -380,6 +422,7 @@ class BlockImpl_dense<XprType,BlockRows,BlockCols, InnerPanel,true>
#ifndef EIGEN_PARSED_BY_DOXYGEN
/** \internal used by allowAligned() */
EIGEN_DEVICE_FUNC
inline BlockImpl_dense(XprType& xpr, const Scalar* data, Index blockRows, Index blockCols)
: Base(data, blockRows, blockCols), m_xpr(xpr)
{
......@@ -388,6 +431,7 @@ class BlockImpl_dense<XprType,BlockRows,BlockCols, InnerPanel,true>
#endif
protected:
EIGEN_DEVICE_FUNC
void init()
{
m_outerStride = internal::traits<BlockType>::HasSameStorageOrderAsXprType
......@@ -395,7 +439,9 @@ class BlockImpl_dense<XprType,BlockRows,BlockCols, InnerPanel,true>
: m_xpr.innerStride();
}
typename XprType::Nested m_xpr;
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;
};
......
external/eigen3/Eigen/src/Core/BooleanRedux.h
View file @ a394b22a
......@@ -17,9 +17,10 @@ namespace internal {
template<typename Derived, int UnrollCount>
struct all_unroller
{
typedef typename Derived::ExpressionTraits Traits;
enum {
col = (UnrollCount-1) / Derived::RowsAtCompileTime,
row = (UnrollCount-1) % Derived::RowsAtCompileTime
col = (UnrollCount-1) / Traits::RowsAtCompileTime,
row = (UnrollCount-1) % Traits::RowsAtCompileTime
};
static inline bool run(const Derived &mat)
......@@ -43,11 +44,12 @@ struct all_unroller<Derived, Dynamic>
template<typename Derived, int UnrollCount>
struct any_unroller
{
typedef typename Derived::ExpressionTraits Traits;
enum {
col = (UnrollCount-1) / Derived::RowsAtCompileTime,
row = (UnrollCount-1) % Derived::RowsAtCompileTime
col = (UnrollCount-1) / Traits::RowsAtCompileTime,
row = (UnrollCount-1) % Traits::RowsAtCompileTime
};
static inline bool run(const Derived &mat)
{
return any_unroller<Derived, UnrollCount-1>::run(mat) || mat.coeff(row, col);
......@@ -78,19 +80,19 @@ struct any_unroller<Derived, Dynamic>
template<typename Derived>
inline bool DenseBase<Derived>::all() const
{
typedef internal::evaluator<Derived> Evaluator;
enum {
unroll = SizeAtCompileTime != Dynamic
&& CoeffReadCost != Dynamic
&& NumTraits<Scalar>::AddCost != Dynamic
&& SizeAtCompileTime * (CoeffReadCost + NumTraits<Scalar>::AddCost) <= EIGEN_UNROLLING_LIMIT
&& SizeAtCompileTime * (Evaluator::CoeffReadCost + NumTraits<Scalar>::AddCost) <= EIGEN_UNROLLING_LIMIT
};
Evaluator evaluator(derived());
if(unroll)
return internal::all_unroller<Derived, unroll ? int(SizeAtCompileTime) : Dynamic>::run(derived());
return internal::all_unroller<Evaluator, unroll ? int(SizeAtCompileTime) : Dynamic>::run(evaluator);
else
{
for(Index j = 0; j < cols(); ++j)
for(Index i = 0; i < rows(); ++i)
if (!coeff(i, j)) return false;
if (!evaluator.coeff(i, j)) return false;
return true;
}
}
......@@ -102,19 +104,19 @@ inline bool DenseBase<Derived>::all() const
template<typename Derived>
inline bool DenseBase<Derived>::any() const
{
typedef internal::evaluator<Derived> Evaluator;
enum {
unroll = SizeAtCompileTime != Dynamic
&& CoeffReadCost != Dynamic
&& NumTraits<Scalar>::AddCost != Dynamic
&& SizeAtCompileTime * (CoeffReadCost + NumTraits<Scalar>::AddCost) <= EIGEN_UNROLLING_LIMIT
&& SizeAtCompileTime * (Evaluator::CoeffReadCost + NumTraits<Scalar>::AddCost) <= EIGEN_UNROLLING_LIMIT
};
Evaluator evaluator(derived());
if(unroll)
return internal::any_unroller<Derived, unroll ? int(SizeAtCompileTime) : Dynamic>::run(derived());
return internal::any_unroller<Evaluator, unroll ? int(SizeAtCompileTime) : Dynamic>::run(evaluator);
else
{
for(Index j = 0; j < cols(); ++j)
for(Index i = 0; i < rows(); ++i)
if (coeff(i, j)) return true;
if (evaluator.coeff(i, j)) return true;
return false;
}
}
......@@ -124,7 +126,7 @@ inline bool DenseBase<Derived>::any() const
* \sa all(), any()
*/
template<typename Derived>
inline typename DenseBase<Derived>::Index DenseBase<Derived>::count() const
inline Eigen::Index DenseBase<Derived>::count() const
{
return derived().template cast<bool>().template cast<Index>().sum();
}
......@@ -136,7 +138,11 @@ inline typename DenseBase<Derived>::Index DenseBase<Derived>::count() const
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.
......@@ -146,7 +152,11 @@ inline bool DenseBase<Derived>::hasNaN() const
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
......
external/eigen3/Eigen/src/Core/CMakeLists.txt deleted 100644 → 0
View file @ 701c0225
FILE(GLOB Eigen_Core_SRCS "*.h")
INSTALL(FILES
${Eigen_Core_SRCS}
DESTINATION ${INCLUDE_INSTALL_DIR}/Eigen/src/Core COMPONENT Devel
)
ADD_SUBDIRECTORY(products)
ADD_SUBDIRECTORY(util)
ADD_SUBDIRECTORY(arch)
external/eigen3/Eigen/src/Core/CommaInitializer.h
View file @ a394b22a
......@@ -22,14 +22,14 @@ namespace Eigen {
* the return type of MatrixBase::operator<<, and most of the time this is the only
* way it is used.
*
* \sa \ref MatrixBaseCommaInitRef "MatrixBase::operator<<", CommaInitializer::finished()
* \sa \blank \ref MatrixBaseCommaInitRef "MatrixBase::operator<<", CommaInitializer::finished()
*/
template<typename XprType>
struct CommaInitializer
{
typedef typename XprType::Scalar Scalar;
typedef typename XprType::Index Index;
EIGEN_DEVICE_FUNC
inline CommaInitializer(XprType& xpr, const Scalar& s)
: m_xpr(xpr), m_row(0), m_col(1), m_currentBlockRows(1)
{
......@@ -37,6 +37,7 @@ struct CommaInitializer
}
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())
{
......@@ -46,6 +47,7 @@ struct CommaInitializer
/* 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:
......@@ -55,6 +57,7 @@ struct CommaInitializer
}
/* inserts a scalar value in the target matrix */
EIGEN_DEVICE_FUNC
CommaInitializer& operator,(const Scalar& s)
{
if (m_col==m_xpr.cols())
......@@ -74,11 +77,10 @@ struct CommaInitializer
/* inserts a matrix expression in the target matrix */
template<typename OtherDerived>
EIGEN_DEVICE_FUNC
CommaInitializer& operator,(const DenseBase<OtherDerived>& other)
{
if(other.cols()==0 || other.rows()==0)
return *this;
if (m_col==m_xpr.cols())
if (m_col==m_xpr.cols() && (other.cols()!=0 || other.rows()!=m_currentBlockRows))
{
m_row+=m_currentBlockRows;
m_col = 0;
......@@ -86,24 +88,22 @@ struct CommaInitializer
eigen_assert(m_row+m_currentBlockRows<=m_xpr.rows()
&& "Too many rows passed to comma initializer (operator<<)");
}
eigen_assert(m_col<m_xpr.cols()
eigen_assert((m_col + other.cols() <= m_xpr.cols())
&& "Too many coefficients passed to comma initializer (operator<<)");
eigen_assert(m_currentBlockRows==other.rows());
if (OtherDerived::SizeAtCompileTime != Dynamic)
m_xpr.template block<OtherDerived::RowsAtCompileTime != Dynamic ? OtherDerived::RowsAtCompileTime : 1,
OtherDerived::ColsAtCompileTime != Dynamic ? OtherDerived::ColsAtCompileTime : 1>
(m_row, m_col) = other;
else
m_xpr.block(m_row, m_col, other.rows(), other.cols()) = other;
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
{
eigen_assert((m_row+m_currentBlockRows) == m_xpr.rows()
&& m_col == m_xpr.cols()
&& "Too few coefficients passed to comma initializer (operator<<)");
finished();
}
/** \returns the built matrix once all its coefficients have been set.
......@@ -113,9 +113,15 @@ struct CommaInitializer
* quaternion.fromRotationMatrix((Matrix3f() << axis0, axis1, axis2).finished());
* \endcode
*/
inline XprType& finished() { return m_xpr; }
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
XprType& m_xpr; // target expression
Index m_row; // current row id
Index m_col; // current col id
Index m_currentBlockRows; // current block height
......
external/eigen3/Eigen/src/Core/ConditionEstimator.h 0 → 100644
View file @ a394b22a
// 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 RealScalar(1);
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
external/eigen3/Eigen/src/Core/CoreEvaluators.h 0 → 100644
View file @ a394b22a
// This file is part of Eigen, a lightweight C++ template library
// for linear algebra.
//
// Copyright (C) 2011 Benoit Jacob <jacob.benoit.1@gmail.com>
// Copyright (C) 2011-2014 Gael Guennebaud <gael.guennebaud@inria.fr>
// Copyright (C) 2011-2012 Jitse Niesen <jitse@maths.leeds.ac.uk>
//
// 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_COREEVALUATORS_H
#define EIGEN_COREEVALUATORS_H
namespace Eigen {
namespace internal {
// This class returns the evaluator kind from the expression storage kind.
// Default assumes index based accessors
template<typename StorageKind>
struct storage_kind_to_evaluator_kind {
typedef IndexBased Kind;
};
// This class returns the evaluator shape from the expression storage kind.
// It can be Dense, Sparse, Triangular, Diagonal, SelfAdjoint, Band, etc.
template<typename StorageKind> struct storage_kind_to_shape;
template<> struct storage_kind_to_shape<Dense> { typedef DenseShape Shape; };
template<> struct storage_kind_to_shape<SolverStorage> { typedef SolverShape Shape; };
template<> struct storage_kind_to_shape<PermutationStorage> { typedef PermutationShape Shape; };
template<> struct storage_kind_to_shape<TranspositionsStorage> { typedef TranspositionsShape Shape; };
// Evaluators have to be specialized with respect to various criteria such as:
// - storage/structure/shape
// - scalar type
// - etc.
// Therefore, we need specialization of evaluator providing additional template arguments for each kind of evaluators.
// We currently distinguish the following kind of evaluators:
// - unary_evaluator for expressions taking only one arguments (CwiseUnaryOp, CwiseUnaryView, Transpose, MatrixWrapper, ArrayWrapper, Reverse, Replicate)
// - binary_evaluator for expression taking two arguments (CwiseBinaryOp)
// - ternary_evaluator for expression taking three arguments (CwiseTernaryOp)
// - product_evaluator for linear algebra products (Product); special case of binary_evaluator because it requires additional tags for dispatching.
// - mapbase_evaluator for Map, Block, Ref
// - block_evaluator for Block (special dispatching to a mapbase_evaluator or unary_evaluator)
template< typename T,
typename Arg1Kind = typename evaluator_traits<typename T::Arg1>::Kind,
typename Arg2Kind = typename evaluator_traits<typename T::Arg2>::Kind,
typename Arg3Kind = typename evaluator_traits<typename T::Arg3>::Kind,
typename Arg1Scalar = typename traits<typename T::Arg1>::Scalar,
typename Arg2Scalar = typename traits<typename T::Arg2>::Scalar,
typename Arg3Scalar = typename traits<typename T::Arg3>::Scalar> struct ternary_evaluator;
template< typename T,
typename LhsKind = typename evaluator_traits<typename T::Lhs>::Kind,
typename RhsKind = typename evaluator_traits<typename T::Rhs>::Kind,
typename LhsScalar = typename traits<typename T::Lhs>::Scalar,
typename RhsScalar = typename traits<typename T::Rhs>::Scalar> struct binary_evaluator;
template< typename T,
typename Kind = typename evaluator_traits<typename T::NestedExpression>::Kind,
typename Scalar = typename T::Scalar> struct unary_evaluator;
// evaluator_traits<T> contains traits for evaluator<T>
template<typename T>
struct evaluator_traits_base
{
// by default, get evaluator kind and shape from storage
typedef typename storage_kind_to_evaluator_kind<typename traits<T>::StorageKind>::Kind Kind;
typedef typename storage_kind_to_shape<typename traits<T>::StorageKind>::Shape Shape;
};
// Default evaluator traits
template<typename T>
struct evaluator_traits : public evaluator_traits_base<T>
{
};
template<typename T, typename Shape = typename evaluator_traits<T>::Shape >
struct evaluator_assume_aliasing {
static const bool value = false;
};
// By default, we assume a unary expression:
template<typename T>
struct evaluator : public unary_evaluator<T>
{
typedef unary_evaluator<T> Base;
EIGEN_DEVICE_FUNC explicit evaluator(const T& xpr) : Base(xpr) {}
};
// TODO: Think about const-correctness
template<typename T>
struct evaluator<const T>
: evaluator<T>
{
EIGEN_DEVICE_FUNC
explicit evaluator(const T& xpr) : evaluator<T>(xpr) {}
};
// ---------- base class for all evaluators ----------
template<typename ExpressionType>
struct evaluator_base : public noncopyable
{
// TODO that's not very nice to have to propagate all these traits. They are currently only needed to handle outer,inner indices.
typedef traits<ExpressionType> ExpressionTraits;
enum {
Alignment = 0
};
};
// -------------------- Matrix and Array --------------------
//
// evaluator<PlainObjectBase> is a common base class for the
// Matrix and Array evaluators.
// Here we directly specialize evaluator. This is not really a unary expression, and it is, by definition, dense,
// so no need for more sophisticated dispatching.
template<typename Derived>
struct evaluator<PlainObjectBase<Derived> >
: evaluator_base<Derived>
{
typedef PlainObjectBase<Derived> PlainObjectType;
typedef typename PlainObjectType::Scalar Scalar;
typedef typename PlainObjectType::CoeffReturnType CoeffReturnType;
enum {
IsRowMajor = PlainObjectType::IsRowMajor,
IsVectorAtCompileTime = PlainObjectType::IsVectorAtCompileTime,
RowsAtCompileTime = PlainObjectType::RowsAtCompileTime,
ColsAtCompileTime = PlainObjectType::ColsAtCompileTime,
CoeffReadCost = NumTraits<Scalar>::ReadCost,
Flags = traits<Derived>::EvaluatorFlags,
Alignment = traits<Derived>::Alignment
};
EIGEN_DEVICE_FUNC evaluator()
: m_data(0),
m_outerStride(IsVectorAtCompileTime ? 0
: int(IsRowMajor) ? ColsAtCompileTime
: RowsAtCompileTime)
{
EIGEN_INTERNAL_CHECK_COST_VALUE(CoeffReadCost);
}
EIGEN_DEVICE_FUNC explicit evaluator(const PlainObjectType& m)
: m_data(m.data()), m_outerStride(IsVectorAtCompileTime ? 0 : m.outerStride())
{
EIGEN_INTERNAL_CHECK_COST_VALUE(CoeffReadCost);
}
EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
CoeffReturnType coeff(Index row, Index col) const
{
if (IsRowMajor)
return m_data[row * m_outerStride.value() + col];
else
return m_data[row + col * m_outerStride.value()];
}
EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
CoeffReturnType coeff(Index index) const
{
return m_data[index];
}
EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
Scalar& coeffRef(Index row, Index col)
{
if (IsRowMajor)
return const_cast<Scalar*>(m_data)[row * m_outerStride.value() + col];
else
return const_cast<Scalar*>(m_data)[row + col * m_outerStride.value()];
}
EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
Scalar& coeffRef(Index index)
{
return const_cast<Scalar*>(m_data)[index];
}
template<int LoadMode, typename PacketType>
EIGEN_STRONG_INLINE
PacketType packet(Index row, Index col) const
{
if (IsRowMajor)
return ploadt<PacketType, LoadMode>(m_data + row * m_outerStride.value() + col);
else
return ploadt<PacketType, LoadMode>(m_data + row + col * m_outerStride.value());
}
template<int LoadMode, typename PacketType>
EIGEN_STRONG_INLINE
PacketType packet(Index index) const
{
return ploadt<PacketType, LoadMode>(m_data + index);
}
template<int StoreMode,typename PacketType>
EIGEN_STRONG_INLINE
void writePacket(Index row, Index col, const PacketType& x)
{
if (IsRowMajor)
return pstoret<Scalar, PacketType, StoreMode>
(const_cast<Scalar*>(m_data) + row * m_outerStride.value() + col, x);
else
return pstoret<Scalar, PacketType, StoreMode>
(const_cast<Scalar*>(m_data) + row + col * m_outerStride.value(), x);
}
template<int StoreMode, typename PacketType>
EIGEN_STRONG_INLINE
void writePacket(Index index, const PacketType& x)
{
return pstoret<Scalar, PacketType, StoreMode>(const_cast<Scalar*>(m_data) + index, x);
}
protected:
const Scalar *m_data;
// We do not need to know the outer stride for vectors
variable_if_dynamic<Index, IsVectorAtCompileTime ? 0
: int(IsRowMajor) ? ColsAtCompileTime
: RowsAtCompileTime> m_outerStride;
};
template<typename Scalar, int Rows, int Cols, int Options, int MaxRows, int MaxCols>
struct evaluator<Matrix<Scalar, Rows, Cols, Options, MaxRows, MaxCols> >
: evaluator<PlainObjectBase<Matrix<Scalar, Rows, Cols, Options, MaxRows, MaxCols> > >
{
typedef Matrix<Scalar, Rows, Cols, Options, MaxRows, MaxCols> XprType;
EIGEN_DEVICE_FUNC evaluator() {}
EIGEN_DEVICE_FUNC explicit evaluator(const XprType& m)
: evaluator<PlainObjectBase<XprType> >(m)
{ }
};
template<typename Scalar, int Rows, int Cols, int Options, int MaxRows, int MaxCols>
struct evaluator<Array<Scalar, Rows, Cols, Options, MaxRows, MaxCols> >
: evaluator<PlainObjectBase<Array<Scalar, Rows, Cols, Options, MaxRows, MaxCols> > >
{
typedef Array<Scalar, Rows, Cols, Options, MaxRows, MaxCols> XprType;
EIGEN_DEVICE_FUNC evaluator() {}
EIGEN_DEVICE_FUNC explicit evaluator(const XprType& m)
: evaluator<PlainObjectBase<XprType> >(m)
{ }
};
// -------------------- Transpose --------------------
template<typename ArgType>
struct unary_evaluator<Transpose<ArgType>, IndexBased>
: evaluator_base<Transpose<ArgType> >
{
typedef Transpose<ArgType> XprType;
enum {
CoeffReadCost = evaluator<ArgType>::CoeffReadCost,
Flags = evaluator<ArgType>::Flags ^ RowMajorBit,
Alignment = evaluator<ArgType>::Alignment
};
EIGEN_DEVICE_FUNC explicit unary_evaluator(const XprType& t) : m_argImpl(t.nestedExpression()) {}
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(col, row);
}
EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
CoeffReturnType coeff(Index index) const
{
return m_argImpl.coeff(index);
}
EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
Scalar& coeffRef(Index row, Index col)
{
return m_argImpl.coeffRef(col, row);
}
EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
typename XprType::Scalar& coeffRef(Index index)
{
return m_argImpl.coeffRef(index);
}
template<int LoadMode, typename PacketType>
EIGEN_STRONG_INLINE
PacketType packet(Index row, Index col) const
{
return m_argImpl.template packet<LoadMode,PacketType>(col, row);
}
template<int LoadMode, typename PacketType>
EIGEN_STRONG_INLINE
PacketType packet(Index index) const
{
return m_argImpl.template packet<LoadMode,PacketType>(index);
}
template<int StoreMode, typename PacketType>
EIGEN_STRONG_INLINE
void writePacket(Index row, Index col, const PacketType& x)
{
m_argImpl.template writePacket<StoreMode,PacketType>(col, row, x);
}
template<int StoreMode, typename PacketType>
EIGEN_STRONG_INLINE
void writePacket(Index index, const PacketType& x)
{
m_argImpl.template writePacket<StoreMode,PacketType>(index, x);
}
protected:
evaluator<ArgType> m_argImpl;
};
// -------------------- CwiseNullaryOp --------------------
// Like Matrix and Array, this is not really a unary expression, so we directly specialize evaluator.
// Likewise, there is not need to more sophisticated dispatching here.
template<typename Scalar,typename NullaryOp,
bool has_nullary = has_nullary_operator<NullaryOp>::value,
bool has_unary = has_unary_operator<NullaryOp>::value,
bool has_binary = has_binary_operator<NullaryOp>::value>
struct nullary_wrapper
{
template <typename IndexType>
EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE Scalar operator()(const NullaryOp& op, IndexType i, IndexType j) const { return op(i,j); }
template <typename IndexType>
EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE Scalar operator()(const NullaryOp& op, IndexType i) const { return op(i); }
template <typename T, typename IndexType> EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE T packetOp(const NullaryOp& op, IndexType i, IndexType j) const { return op.template packetOp<T>(i,j); }
template <typename T, typename IndexType> EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE T packetOp(const NullaryOp& op, IndexType i) const { return op.template packetOp<T>(i); }
};
template<typename Scalar,typename NullaryOp>
struct nullary_wrapper<Scalar,NullaryOp,true,false,false>
{
template <typename IndexType>
EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE Scalar operator()(const NullaryOp& op, IndexType=0, IndexType=0) const { return op(); }
template <typename T, typename IndexType> EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE T packetOp(const NullaryOp& op, IndexType=0, IndexType=0) const { return op.template packetOp<T>(); }
};
template<typename Scalar,typename NullaryOp>
struct nullary_wrapper<Scalar,NullaryOp,false,false,true>
{
template <typename IndexType>
EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE Scalar operator()(const NullaryOp& op, IndexType i, IndexType j=0) const { return op(i,j); }
template <typename T, typename IndexType> EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE T packetOp(const NullaryOp& op, IndexType i, IndexType j=0) const { return op.template packetOp<T>(i,j); }
};
// We need the following specialization for vector-only functors assigned to a runtime vector,
// for instance, using linspace and assigning a RowVectorXd to a MatrixXd or even a row of a MatrixXd.
// In this case, i==0 and j is used for the actual iteration.
template<typename Scalar,typename NullaryOp>
struct nullary_wrapper<Scalar,NullaryOp,false,true,false>
{
template <typename IndexType>
EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE Scalar operator()(const NullaryOp& op, IndexType i, IndexType j) const {
eigen_assert(i==0 || j==0);
return op(i+j);
}
template <typename T, typename IndexType> EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE T packetOp(const NullaryOp& op, IndexType i, IndexType j) const {
eigen_assert(i==0 || j==0);
return op.template packetOp<T>(i+j);
}
template <typename IndexType>
EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE Scalar operator()(const NullaryOp& op, IndexType i) const { return op(i); }
template <typename T, typename IndexType>
EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE T packetOp(const NullaryOp& op, IndexType i) const { return op.template packetOp<T>(i); }
};
template<typename Scalar,typename NullaryOp>
struct nullary_wrapper<Scalar,NullaryOp,false,false,false> {};
#if 0 && EIGEN_COMP_MSVC>0
// Disable this ugly workaround. This is now handled in traits<Ref>::match,
// but this piece of code might still become handly if some other weird compilation
// erros pop up again.
// MSVC exhibits a weird compilation error when
// compiling:
// Eigen::MatrixXf A = MatrixXf::Random(3,3);
// Ref<const MatrixXf> R = 2.f*A;
// and that has_*ary_operator<scalar_constant_op<float>> have not been instantiated yet.
// The "problem" is that evaluator<2.f*A> is instantiated by traits<Ref>::match<2.f*A>
// and at that time has_*ary_operator<T> returns true regardless of T.
// Then nullary_wrapper is badly instantiated as nullary_wrapper<.,.,true,true,true>.
// The trick is thus to defer the proper instantiation of nullary_wrapper when coeff(),
// and packet() are really instantiated as implemented below:
// This is a simple wrapper around Index to enforce the re-instantiation of
// has_*ary_operator when needed.
template<typename T> struct nullary_wrapper_workaround_msvc {
nullary_wrapper_workaround_msvc(const T&);
operator T()const;
};
template<typename Scalar,typename NullaryOp>
struct nullary_wrapper<Scalar,NullaryOp,true,true,true>
{
template <typename IndexType>
EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE Scalar operator()(const NullaryOp& op, IndexType i, IndexType j) const {
return nullary_wrapper<Scalar,NullaryOp,
has_nullary_operator<NullaryOp,nullary_wrapper_workaround_msvc<IndexType> >::value,
has_unary_operator<NullaryOp,nullary_wrapper_workaround_msvc<IndexType> >::value,
has_binary_operator<NullaryOp,nullary_wrapper_workaround_msvc<IndexType> >::value>().operator()(op,i,j);
}
template <typename IndexType>
EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE Scalar operator()(const NullaryOp& op, IndexType i) const {
return nullary_wrapper<Scalar,NullaryOp,
has_nullary_operator<NullaryOp,nullary_wrapper_workaround_msvc<IndexType> >::value,
has_unary_operator<NullaryOp,nullary_wrapper_workaround_msvc<IndexType> >::value,
has_binary_operator<NullaryOp,nullary_wrapper_workaround_msvc<IndexType> >::value>().operator()(op,i);
}
template <typename T, typename IndexType>
EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE T packetOp(const NullaryOp& op, IndexType i, IndexType j) const {
return nullary_wrapper<Scalar,NullaryOp,
has_nullary_operator<NullaryOp,nullary_wrapper_workaround_msvc<IndexType> >::value,
has_unary_operator<NullaryOp,nullary_wrapper_workaround_msvc<IndexType> >::value,
has_binary_operator<NullaryOp,nullary_wrapper_workaround_msvc<IndexType> >::value>().template packetOp<T>(op,i,j);
}
template <typename T, typename IndexType>
EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE T packetOp(const NullaryOp& op, IndexType i) const {
return nullary_wrapper<Scalar,NullaryOp,
has_nullary_operator<NullaryOp,nullary_wrapper_workaround_msvc<IndexType> >::value,
has_unary_operator<NullaryOp,nullary_wrapper_workaround_msvc<IndexType> >::value,
has_binary_operator<NullaryOp,nullary_wrapper_workaround_msvc<IndexType> >::value>().template packetOp<T>(op,i);
}
};
#endif // MSVC workaround
template<typename NullaryOp, typename PlainObjectType>
struct evaluator<CwiseNullaryOp<NullaryOp,PlainObjectType> >
: evaluator_base<CwiseNullaryOp<NullaryOp,PlainObjectType> >
{
typedef CwiseNullaryOp<NullaryOp,PlainObjectType> XprType;
typedef typename internal::remove_all<PlainObjectType>::type PlainObjectTypeCleaned;
enum {
CoeffReadCost = internal::functor_traits<NullaryOp>::Cost,
Flags = (evaluator<PlainObjectTypeCleaned>::Flags
& ( HereditaryBits
| (functor_has_linear_access<NullaryOp>::ret ? LinearAccessBit : 0)
| (functor_traits<NullaryOp>::PacketAccess ? PacketAccessBit : 0)))
| (functor_traits<NullaryOp>::IsRepeatable ? 0 : EvalBeforeNestingBit),
Alignment = AlignedMax
};
EIGEN_DEVICE_FUNC explicit evaluator(const XprType& n)
: m_functor(n.functor()), m_wrapper()
{
EIGEN_INTERNAL_CHECK_COST_VALUE(CoeffReadCost);
}
typedef typename XprType::CoeffReturnType CoeffReturnType;
template <typename IndexType>
EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
CoeffReturnType coeff(IndexType row, IndexType col) const
{
return m_wrapper(m_functor, row, col);
}
template <typename IndexType>
EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
CoeffReturnType coeff(IndexType index) const
{
return m_wrapper(m_functor,index);
}
template<int LoadMode, typename PacketType, typename IndexType>
EIGEN_STRONG_INLINE
PacketType packet(IndexType row, IndexType col) const
{
return m_wrapper.template packetOp<PacketType>(m_functor, row, col);
}
template<int LoadMode, typename PacketType, typename IndexType>
EIGEN_STRONG_INLINE
PacketType packet(IndexType index) const
{
return m_wrapper.template packetOp<PacketType>(m_functor, index);
}
protected:
const NullaryOp m_functor;
const internal::nullary_wrapper<CoeffReturnType,NullaryOp> m_wrapper;
};
// -------------------- CwiseUnaryOp --------------------
template<typename UnaryOp, typename ArgType>
struct unary_evaluator<CwiseUnaryOp<UnaryOp, ArgType>, IndexBased >
: evaluator_base<CwiseUnaryOp<UnaryOp, ArgType> >
{
typedef CwiseUnaryOp<UnaryOp, ArgType> XprType;
enum {
CoeffReadCost = evaluator<ArgType>::CoeffReadCost + functor_traits<UnaryOp>::Cost,
Flags = evaluator<ArgType>::Flags
& (HereditaryBits | LinearAccessBit | (functor_traits<UnaryOp>::PacketAccess ? PacketAccessBit : 0)),
Alignment = evaluator<ArgType>::Alignment
};
EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
explicit unary_evaluator(const XprType& op)
: m_functor(op.functor()),
m_argImpl(op.nestedExpression())
{
EIGEN_INTERNAL_CHECK_COST_VALUE(functor_traits<UnaryOp>::Cost);
EIGEN_INTERNAL_CHECK_COST_VALUE(CoeffReadCost);
}
typedef typename XprType::CoeffReturnType CoeffReturnType;
EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
CoeffReturnType coeff(Index row, Index col) const
{
return m_functor(m_argImpl.coeff(row, col));
}
EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
CoeffReturnType coeff(Index index) const
{
return m_functor(m_argImpl.coeff(index));
}
template<int LoadMode, typename PacketType>
EIGEN_STRONG_INLINE
PacketType packet(Index row, Index col) const
{
return m_functor.packetOp(m_argImpl.template packet<LoadMode, PacketType>(row, col));
}
template<int LoadMode, typename PacketType>
EIGEN_STRONG_INLINE
PacketType packet(Index index) const
{
return m_functor.packetOp(m_argImpl.template packet<LoadMode, PacketType>(index));
}
protected:
const UnaryOp m_functor;
evaluator<ArgType> m_argImpl;
};
// -------------------- CwiseTernaryOp --------------------
// this is a ternary expression
template<typename TernaryOp, typename Arg1, typename Arg2, typename Arg3>
struct evaluator<CwiseTernaryOp<TernaryOp, Arg1, Arg2, Arg3> >
: public ternary_evaluator<CwiseTernaryOp<TernaryOp, Arg1, Arg2, Arg3> >
{
typedef CwiseTernaryOp<TernaryOp, Arg1, Arg2, Arg3> XprType;
typedef ternary_evaluator<CwiseTernaryOp<TernaryOp, Arg1, Arg2, Arg3> > Base;
EIGEN_DEVICE_FUNC explicit evaluator(const XprType& xpr) : Base(xpr) {}
};
template<typename TernaryOp, typename Arg1, typename Arg2, typename Arg3>
struct ternary_evaluator<CwiseTernaryOp<TernaryOp, Arg1, Arg2, Arg3>, IndexBased, IndexBased>
: evaluator_base<CwiseTernaryOp<TernaryOp, Arg1, Arg2, Arg3> >
{
typedef CwiseTernaryOp<TernaryOp, Arg1, Arg2, Arg3> XprType;
enum {
CoeffReadCost = evaluator<Arg1>::CoeffReadCost + evaluator<Arg2>::CoeffReadCost + evaluator<Arg3>::CoeffReadCost + functor_traits<TernaryOp>::Cost,
Arg1Flags = evaluator<Arg1>::Flags,
Arg2Flags = evaluator<Arg2>::Flags,
Arg3Flags = evaluator<Arg3>::Flags,
SameType = is_same<typename Arg1::Scalar,typename Arg2::Scalar>::value && is_same<typename Arg1::Scalar,typename Arg3::Scalar>::value,
StorageOrdersAgree = (int(Arg1Flags)&RowMajorBit)==(int(Arg2Flags)&RowMajorBit) && (int(Arg1Flags)&RowMajorBit)==(int(Arg3Flags)&RowMajorBit),
Flags0 = (int(Arg1Flags) | int(Arg2Flags) | int(Arg3Flags)) & (
HereditaryBits
| (int(Arg1Flags) & int(Arg2Flags) & int(Arg3Flags) &
( (StorageOrdersAgree ? LinearAccessBit : 0)
| (functor_traits<TernaryOp>::PacketAccess && StorageOrdersAgree && SameType ? PacketAccessBit : 0)
)
)
),
Flags = (Flags0 & ~RowMajorBit) | (Arg1Flags & RowMajorBit),
Alignment = EIGEN_PLAIN_ENUM_MIN(
EIGEN_PLAIN_ENUM_MIN(evaluator<Arg1>::Alignment, evaluator<Arg2>::Alignment),
evaluator<Arg3>::Alignment)
};
EIGEN_DEVICE_FUNC explicit ternary_evaluator(const XprType& xpr)
: m_functor(xpr.functor()),
m_arg1Impl(xpr.arg1()),
m_arg2Impl(xpr.arg2()),
m_arg3Impl(xpr.arg3())
{
EIGEN_INTERNAL_CHECK_COST_VALUE(functor_traits<TernaryOp>::Cost);
EIGEN_INTERNAL_CHECK_COST_VALUE(CoeffReadCost);
}
typedef typename XprType::CoeffReturnType CoeffReturnType;
EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
CoeffReturnType coeff(Index row, Index col) const
{
return m_functor(m_arg1Impl.coeff(row, col), m_arg2Impl.coeff(row, col), m_arg3Impl.coeff(row, col));
}
EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
CoeffReturnType coeff(Index index) const
{
return m_functor(m_arg1Impl.coeff(index), m_arg2Impl.coeff(index), m_arg3Impl.coeff(index));
}
template<int LoadMode, typename PacketType>
EIGEN_STRONG_INLINE
PacketType packet(Index row, Index col) const
{
return m_functor.packetOp(m_arg1Impl.template packet<LoadMode,PacketType>(row, col),
m_arg2Impl.template packet<LoadMode,PacketType>(row, col),
m_arg3Impl.template packet<LoadMode,PacketType>(row, col));
}
template<int LoadMode, typename PacketType>
EIGEN_STRONG_INLINE
PacketType packet(Index index) const
{
return m_functor.packetOp(m_arg1Impl.template packet<LoadMode,PacketType>(index),
m_arg2Impl.template packet<LoadMode,PacketType>(index),
m_arg3Impl.template packet<LoadMode,PacketType>(index));
}
protected:
const TernaryOp m_functor;
evaluator<Arg1> m_arg1Impl;
evaluator<Arg2> m_arg2Impl;
evaluator<Arg3> m_arg3Impl;
};
// -------------------- CwiseBinaryOp --------------------
// this is a binary expression
template<typename BinaryOp, typename Lhs, typename Rhs>
struct evaluator<CwiseBinaryOp<BinaryOp, Lhs, Rhs> >
: public binary_evaluator<CwiseBinaryOp<BinaryOp, Lhs, Rhs> >
{
typedef CwiseBinaryOp<BinaryOp, Lhs, Rhs> XprType;
typedef binary_evaluator<CwiseBinaryOp<BinaryOp, Lhs, Rhs> > Base;
EIGEN_DEVICE_FUNC explicit evaluator(const XprType& xpr) : Base(xpr) {}
};
template<typename BinaryOp, typename Lhs, typename Rhs>
struct binary_evaluator<CwiseBinaryOp<BinaryOp, Lhs, Rhs>, IndexBased, IndexBased>
: evaluator_base<CwiseBinaryOp<BinaryOp, Lhs, Rhs> >
{
typedef CwiseBinaryOp<BinaryOp, Lhs, Rhs> XprType;
enum {
CoeffReadCost = evaluator<Lhs>::CoeffReadCost + evaluator<Rhs>::CoeffReadCost + functor_traits<BinaryOp>::Cost,
LhsFlags = evaluator<Lhs>::Flags,
RhsFlags = evaluator<Rhs>::Flags,
SameType = is_same<typename Lhs::Scalar,typename Rhs::Scalar>::value,
StorageOrdersAgree = (int(LhsFlags)&RowMajorBit)==(int(RhsFlags)&RowMajorBit),
Flags0 = (int(LhsFlags) | int(RhsFlags)) & (
HereditaryBits
| (int(LhsFlags) & int(RhsFlags) &
( (StorageOrdersAgree ? LinearAccessBit : 0)
| (functor_traits<BinaryOp>::PacketAccess && StorageOrdersAgree && SameType ? PacketAccessBit : 0)
)
)
),
Flags = (Flags0 & ~RowMajorBit) | (LhsFlags & RowMajorBit),
Alignment = EIGEN_PLAIN_ENUM_MIN(evaluator<Lhs>::Alignment,evaluator<Rhs>::Alignment)
};
EIGEN_DEVICE_FUNC explicit binary_evaluator(const XprType& xpr)
: m_functor(xpr.functor()),
m_lhsImpl(xpr.lhs()),
m_rhsImpl(xpr.rhs())
{
EIGEN_INTERNAL_CHECK_COST_VALUE(functor_traits<BinaryOp>::Cost);
EIGEN_INTERNAL_CHECK_COST_VALUE(CoeffReadCost);
}
typedef typename XprType::CoeffReturnType CoeffReturnType;
EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
CoeffReturnType coeff(Index row, Index col) const
{
return m_functor(m_lhsImpl.coeff(row, col), m_rhsImpl.coeff(row, col));
}
EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
CoeffReturnType coeff(Index index) const
{
return m_functor(m_lhsImpl.coeff(index), m_rhsImpl.coeff(index));
}
template<int LoadMode, typename PacketType>
EIGEN_STRONG_INLINE
PacketType packet(Index row, Index col) const
{
return m_functor.packetOp(m_lhsImpl.template packet<LoadMode,PacketType>(row, col),
m_rhsImpl.template packet<LoadMode,PacketType>(row, col));
}
template<int LoadMode, typename PacketType>
EIGEN_STRONG_INLINE
PacketType packet(Index index) const
{
return m_functor.packetOp(m_lhsImpl.template packet<LoadMode,PacketType>(index),
m_rhsImpl.template packet<LoadMode,PacketType>(index));
}
protected:
const BinaryOp m_functor;
evaluator<Lhs> m_lhsImpl;
evaluator<Rhs> m_rhsImpl;
};
// -------------------- CwiseUnaryView --------------------
template<typename UnaryOp, typename ArgType>
struct unary_evaluator<CwiseUnaryView<UnaryOp, ArgType>, IndexBased>
: evaluator_base<CwiseUnaryView<UnaryOp, ArgType> >
{
typedef CwiseUnaryView<UnaryOp, ArgType> XprType;
enum {
CoeffReadCost = evaluator<ArgType>::CoeffReadCost + functor_traits<UnaryOp>::Cost,
Flags = (evaluator<ArgType>::Flags & (HereditaryBits | LinearAccessBit | DirectAccessBit)),
Alignment = 0 // FIXME it is not very clear why alignment is necessarily lost...
};
EIGEN_DEVICE_FUNC explicit unary_evaluator(const XprType& op)
: m_unaryOp(op.functor()),
m_argImpl(op.nestedExpression())
{
EIGEN_INTERNAL_CHECK_COST_VALUE(functor_traits<UnaryOp>::Cost);
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_unaryOp(m_argImpl.coeff(row, col));
}
EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
CoeffReturnType coeff(Index index) const
{
return m_unaryOp(m_argImpl.coeff(index));
}
EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
Scalar& coeffRef(Index row, Index col)
{
return m_unaryOp(m_argImpl.coeffRef(row, col));
}
EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
Scalar& coeffRef(Index index)
{
return m_unaryOp(m_argImpl.coeffRef(index));
}
protected:
const UnaryOp m_unaryOp;
evaluator<ArgType> m_argImpl;
};
// -------------------- Map --------------------
// FIXME perhaps the PlainObjectType could be provided by Derived::PlainObject ?
// but that might complicate template specialization
template<typename Derived, typename PlainObjectType>
struct mapbase_evaluator;
template<typename Derived, typename PlainObjectType>
struct mapbase_evaluator : evaluator_base<Derived>
{
typedef Derived XprType;
typedef typename XprType::PointerType PointerType;
typedef typename XprType::Scalar Scalar;
typedef typename XprType::CoeffReturnType CoeffReturnType;
enum {
IsRowMajor = XprType::RowsAtCompileTime,
ColsAtCompileTime = XprType::ColsAtCompileTime,
CoeffReadCost = NumTraits<Scalar>::ReadCost
};
EIGEN_DEVICE_FUNC explicit mapbase_evaluator(const XprType& map)
: m_data(const_cast<PointerType>(map.data())),
m_innerStride(map.innerStride()),
m_outerStride(map.outerStride())
{
EIGEN_STATIC_ASSERT(EIGEN_IMPLIES(evaluator<Derived>::Flags&PacketAccessBit, internal::inner_stride_at_compile_time<Derived>::ret==1),
PACKET_ACCESS_REQUIRES_TO_HAVE_INNER_STRIDE_FIXED_TO_1);
EIGEN_INTERNAL_CHECK_COST_VALUE(CoeffReadCost);
}
EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
CoeffReturnType coeff(Index row, Index col) const
{
return m_data[col * colStride() + row * rowStride()];
}
EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
CoeffReturnType coeff(Index index) const
{
return m_data[index * m_innerStride.value()];
}
EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
Scalar& coeffRef(Index row, Index col)
{
return m_data[col * colStride() + row * rowStride()];
}
EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
Scalar& coeffRef(Index index)
{
return m_data[index * m_innerStride.value()];
}
template<int LoadMode, typename PacketType>
EIGEN_STRONG_INLINE
PacketType packet(Index row, Index col) const
{
PointerType ptr = m_data + row * rowStride() + col * colStride();
return internal::ploadt<PacketType, LoadMode>(ptr);
}
template<int LoadMode, typename PacketType>
EIGEN_STRONG_INLINE
PacketType packet(Index index) const
{
return internal::ploadt<PacketType, LoadMode>(m_data + index * m_innerStride.value());
}
template<int StoreMode, typename PacketType>
EIGEN_STRONG_INLINE
void writePacket(Index row, Index col, const PacketType& x)
{
PointerType ptr = m_data + row * rowStride() + col * colStride();
return internal::pstoret<Scalar, PacketType, StoreMode>(ptr, x);
}
template<int StoreMode, typename PacketType>
EIGEN_STRONG_INLINE
void writePacket(Index index, const PacketType& x)
{
internal::pstoret<Scalar, PacketType, StoreMode>(m_data + index * m_innerStride.value(), x);
}
protected:
EIGEN_DEVICE_FUNC
inline Index rowStride() const { return XprType::IsRowMajor ? m_outerStride.value() : m_innerStride.value(); }
EIGEN_DEVICE_FUNC
inline Index colStride() const { return XprType::IsRowMajor ? m_innerStride.value() : m_outerStride.value(); }
PointerType m_data;
const internal::variable_if_dynamic<Index, XprType::InnerStrideAtCompileTime> m_innerStride;
const internal::variable_if_dynamic<Index, XprType::OuterStrideAtCompileTime> m_outerStride;
};
template<typename PlainObjectType, int MapOptions, typename StrideType>
struct evaluator<Map<PlainObjectType, MapOptions, StrideType> >
: public mapbase_evaluator<Map<PlainObjectType, MapOptions, StrideType>, PlainObjectType>
{
typedef Map<PlainObjectType, MapOptions, StrideType> XprType;
typedef typename XprType::Scalar Scalar;
// TODO: should check for smaller packet types once we can handle multi-sized packet types
typedef typename packet_traits<Scalar>::type PacketScalar;
enum {
InnerStrideAtCompileTime = StrideType::InnerStrideAtCompileTime == 0
? int(PlainObjectType::InnerStrideAtCompileTime)
: int(StrideType::InnerStrideAtCompileTime),
OuterStrideAtCompileTime = StrideType::OuterStrideAtCompileTime == 0
? int(PlainObjectType::OuterStrideAtCompileTime)
: int(StrideType::OuterStrideAtCompileTime),
HasNoInnerStride = InnerStrideAtCompileTime == 1,
HasNoOuterStride = StrideType::OuterStrideAtCompileTime == 0,
HasNoStride = HasNoInnerStride && HasNoOuterStride,
IsDynamicSize = PlainObjectType::SizeAtCompileTime==Dynamic,
PacketAccessMask = bool(HasNoInnerStride) ? ~int(0) : ~int(PacketAccessBit),
LinearAccessMask = bool(HasNoStride) || bool(PlainObjectType::IsVectorAtCompileTime) ? ~int(0) : ~int(LinearAccessBit),
Flags = int( evaluator<PlainObjectType>::Flags) & (LinearAccessMask&PacketAccessMask),
Alignment = int(MapOptions)&int(AlignedMask)
};
EIGEN_DEVICE_FUNC explicit evaluator(const XprType& map)
: mapbase_evaluator<XprType, PlainObjectType>(map)
{ }
};
// -------------------- Ref --------------------
template<typename PlainObjectType, int RefOptions, typename StrideType>
struct evaluator<Ref<PlainObjectType, RefOptions, StrideType> >
: public mapbase_evaluator<Ref<PlainObjectType, RefOptions, StrideType>, PlainObjectType>
{
typedef Ref<PlainObjectType, RefOptions, StrideType> XprType;
enum {
Flags = evaluator<Map<PlainObjectType, RefOptions, StrideType> >::Flags,
Alignment = evaluator<Map<PlainObjectType, RefOptions, StrideType> >::Alignment
};
EIGEN_DEVICE_FUNC explicit evaluator(const XprType& ref)
: mapbase_evaluator<XprType, PlainObjectType>(ref)
{ }
};
// -------------------- Block --------------------
template<typename ArgType, int BlockRows, int BlockCols, bool InnerPanel,
bool HasDirectAccess = internal::has_direct_access<ArgType>::ret> struct block_evaluator;
template<typename ArgType, int BlockRows, int BlockCols, bool InnerPanel>
struct evaluator<Block<ArgType, BlockRows, BlockCols, InnerPanel> >
: block_evaluator<ArgType, BlockRows, BlockCols, InnerPanel>
{
typedef Block<ArgType, BlockRows, BlockCols, InnerPanel> XprType;
typedef typename XprType::Scalar Scalar;
// TODO: should check for smaller packet types once we can handle multi-sized packet types
typedef typename packet_traits<Scalar>::type PacketScalar;
enum {
CoeffReadCost = evaluator<ArgType>::CoeffReadCost,
RowsAtCompileTime = traits<XprType>::RowsAtCompileTime,
ColsAtCompileTime = traits<XprType>::ColsAtCompileTime,
MaxRowsAtCompileTime = traits<XprType>::MaxRowsAtCompileTime,
MaxColsAtCompileTime = traits<XprType>::MaxColsAtCompileTime,
ArgTypeIsRowMajor = (int(evaluator<ArgType>::Flags)&RowMajorBit) != 0,
IsRowMajor = (MaxRowsAtCompileTime==1 && MaxColsAtCompileTime!=1) ? 1
: (MaxColsAtCompileTime==1 && MaxRowsAtCompileTime!=1) ? 0
: ArgTypeIsRowMajor,
HasSameStorageOrderAsArgType = (IsRowMajor == ArgTypeIsRowMajor),
InnerSize = IsRowMajor ? int(ColsAtCompileTime) : int(RowsAtCompileTime),
InnerStrideAtCompileTime = HasSameStorageOrderAsArgType
? int(inner_stride_at_compile_time<ArgType>::ret)
: int(outer_stride_at_compile_time<ArgType>::ret),
OuterStrideAtCompileTime = HasSameStorageOrderAsArgType
? int(outer_stride_at_compile_time<ArgType>::ret)
: int(inner_stride_at_compile_time<ArgType>::ret),
MaskPacketAccessBit = (InnerStrideAtCompileTime == 1) ? PacketAccessBit : 0,
FlagsLinearAccessBit = (RowsAtCompileTime == 1 || ColsAtCompileTime == 1 || (InnerPanel && (evaluator<ArgType>::Flags&LinearAccessBit))) ? LinearAccessBit : 0,
FlagsRowMajorBit = XprType::Flags&RowMajorBit,
Flags0 = evaluator<ArgType>::Flags & ( (HereditaryBits & ~RowMajorBit) |
DirectAccessBit |
MaskPacketAccessBit),
Flags = Flags0 | FlagsLinearAccessBit | FlagsRowMajorBit,
PacketAlignment = unpacket_traits<PacketScalar>::alignment,
Alignment0 = (InnerPanel && (OuterStrideAtCompileTime!=Dynamic) && (((OuterStrideAtCompileTime * int(sizeof(Scalar))) % int(PacketAlignment)) == 0)) ? int(PacketAlignment) : 0,
Alignment = EIGEN_PLAIN_ENUM_MIN(evaluator<ArgType>::Alignment, Alignment0)
};
typedef block_evaluator<ArgType, BlockRows, BlockCols, InnerPanel> block_evaluator_type;
EIGEN_DEVICE_FUNC explicit evaluator(const XprType& block) : block_evaluator_type(block)
{
EIGEN_INTERNAL_CHECK_COST_VALUE(CoeffReadCost);
}
};
// no direct-access => dispatch to a unary evaluator
template<typename ArgType, int BlockRows, int BlockCols, bool InnerPanel>
struct block_evaluator<ArgType, BlockRows, BlockCols, InnerPanel, /*HasDirectAccess*/ false>
: unary_evaluator<Block<ArgType, BlockRows, BlockCols, InnerPanel> >
{
typedef Block<ArgType, BlockRows, BlockCols, InnerPanel> XprType;
EIGEN_DEVICE_FUNC explicit block_evaluator(const XprType& block)
: unary_evaluator<XprType>(block)
{}
};
template<typename ArgType, int BlockRows, int BlockCols, bool InnerPanel>
struct unary_evaluator<Block<ArgType, BlockRows, BlockCols, InnerPanel>, IndexBased>
: evaluator_base<Block<ArgType, BlockRows, BlockCols, InnerPanel> >
{
typedef Block<ArgType, BlockRows, BlockCols, InnerPanel> XprType;
EIGEN_DEVICE_FUNC explicit unary_evaluator(const XprType& block)
: m_argImpl(block.nestedExpression()),
m_startRow(block.startRow()),
m_startCol(block.startCol())
{ }
typedef typename XprType::Scalar Scalar;
typedef typename XprType::CoeffReturnType CoeffReturnType;
enum {
RowsAtCompileTime = XprType::RowsAtCompileTime
};
EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
CoeffReturnType coeff(Index row, Index col) const
{
return m_argImpl.coeff(m_startRow.value() + row, m_startCol.value() + col);
}
EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
CoeffReturnType coeff(Index index) const
{
return coeff(RowsAtCompileTime == 1 ? 0 : index, RowsAtCompileTime == 1 ? index : 0);
}
EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
Scalar& coeffRef(Index row, Index col)
{
return m_argImpl.coeffRef(m_startRow.value() + row, m_startCol.value() + col);
}
EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
Scalar& coeffRef(Index index)
{
return coeffRef(RowsAtCompileTime == 1 ? 0 : index, RowsAtCompileTime == 1 ? index : 0);
}
template<int LoadMode, typename PacketType>
EIGEN_STRONG_INLINE
PacketType packet(Index row, Index col) const
{
return m_argImpl.template packet<LoadMode,PacketType>(m_startRow.value() + row, m_startCol.value() + col);
}
template<int LoadMode, typename PacketType>
EIGEN_STRONG_INLINE
PacketType packet(Index index) const
{
return packet<LoadMode,PacketType>(RowsAtCompileTime == 1 ? 0 : index,
RowsAtCompileTime == 1 ? index : 0);
}
template<int StoreMode, typename PacketType>
EIGEN_STRONG_INLINE
void writePacket(Index row, Index col, const PacketType& x)
{
return m_argImpl.template writePacket<StoreMode,PacketType>(m_startRow.value() + row, m_startCol.value() + col, x);
}
template<int StoreMode, typename PacketType>
EIGEN_STRONG_INLINE
void writePacket(Index index, const PacketType& x)
{
return writePacket<StoreMode,PacketType>(RowsAtCompileTime == 1 ? 0 : index,
RowsAtCompileTime == 1 ? index : 0,
x);
}
protected:
evaluator<ArgType> m_argImpl;
const variable_if_dynamic<Index, (ArgType::RowsAtCompileTime == 1 && BlockRows==1) ? 0 : Dynamic> m_startRow;
const variable_if_dynamic<Index, (ArgType::ColsAtCompileTime == 1 && BlockCols==1) ? 0 : Dynamic> m_startCol;
};
// TODO: This evaluator does not actually use the child evaluator;
// all action is via the data() as returned by the Block expression.
template<typename ArgType, int BlockRows, int BlockCols, bool InnerPanel>
struct block_evaluator<ArgType, BlockRows, BlockCols, InnerPanel, /* HasDirectAccess */ true>
: mapbase_evaluator<Block<ArgType, BlockRows, BlockCols, InnerPanel>,
typename Block<ArgType, BlockRows, BlockCols, InnerPanel>::PlainObject>
{
typedef Block<ArgType, BlockRows, BlockCols, InnerPanel> XprType;
typedef typename XprType::Scalar Scalar;
EIGEN_DEVICE_FUNC explicit block_evaluator(const XprType& block)
: mapbase_evaluator<XprType, typename XprType::PlainObject>(block)
{
// TODO: for the 3.3 release, this should be turned to an internal assertion, but let's keep it as is for the beta lifetime
eigen_assert(((internal::UIntPtr(block.data()) % EIGEN_PLAIN_ENUM_MAX(1,evaluator<XprType>::Alignment)) == 0) && "data is not aligned");
}
};
// -------------------- Select --------------------
// NOTE shall we introduce a ternary_evaluator?
// TODO enable vectorization for Select
template<typename ConditionMatrixType, typename ThenMatrixType, typename ElseMatrixType>
struct evaluator<Select<ConditionMatrixType, ThenMatrixType, ElseMatrixType> >
: evaluator_base<Select<ConditionMatrixType, ThenMatrixType, ElseMatrixType> >
{
typedef Select<ConditionMatrixType, ThenMatrixType, ElseMatrixType> XprType;
enum {
CoeffReadCost = evaluator<ConditionMatrixType>::CoeffReadCost
+ EIGEN_PLAIN_ENUM_MAX(evaluator<ThenMatrixType>::CoeffReadCost,
evaluator<ElseMatrixType>::CoeffReadCost),
Flags = (unsigned int)evaluator<ThenMatrixType>::Flags & evaluator<ElseMatrixType>::Flags & HereditaryBits,
Alignment = EIGEN_PLAIN_ENUM_MIN(evaluator<ThenMatrixType>::Alignment, evaluator<ElseMatrixType>::Alignment)
};
EIGEN_DEVICE_FUNC explicit evaluator(const XprType& select)
: m_conditionImpl(select.conditionMatrix()),
m_thenImpl(select.thenMatrix()),
m_elseImpl(select.elseMatrix())
{
EIGEN_INTERNAL_CHECK_COST_VALUE(CoeffReadCost);
}
typedef typename XprType::CoeffReturnType CoeffReturnType;
EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
CoeffReturnType coeff(Index row, Index col) const
{
if (m_conditionImpl.coeff(row, col))
return m_thenImpl.coeff(row, col);
else
return m_elseImpl.coeff(row, col);
}
EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
CoeffReturnType coeff(Index index) const
{
if (m_conditionImpl.coeff(index))
return m_thenImpl.coeff(index);
else
return m_elseImpl.coeff(index);
}
protected:
evaluator<ConditionMatrixType> m_conditionImpl;
evaluator<ThenMatrixType> m_thenImpl;
evaluator<ElseMatrixType> m_elseImpl;
};
// -------------------- Replicate --------------------
template<typename ArgType, int RowFactor, int ColFactor>
struct unary_evaluator<Replicate<ArgType, RowFactor, ColFactor> >
: evaluator_base<Replicate<ArgType, RowFactor, ColFactor> >
{
typedef Replicate<ArgType, RowFactor, ColFactor> XprType;
typedef typename XprType::CoeffReturnType CoeffReturnType;
enum {
Factor = (RowFactor==Dynamic || ColFactor==Dynamic) ? Dynamic : RowFactor*ColFactor
};
typedef typename internal::nested_eval<ArgType,Factor>::type ArgTypeNested;
typedef typename internal::remove_all<ArgTypeNested>::type ArgTypeNestedCleaned;
enum {
CoeffReadCost = evaluator<ArgTypeNestedCleaned>::CoeffReadCost,
LinearAccessMask = XprType::IsVectorAtCompileTime ? LinearAccessBit : 0,
Flags = (evaluator<ArgTypeNestedCleaned>::Flags & (HereditaryBits|LinearAccessMask) & ~RowMajorBit) | (traits<XprType>::Flags & RowMajorBit),
Alignment = evaluator<ArgTypeNestedCleaned>::Alignment
};
EIGEN_DEVICE_FUNC explicit unary_evaluator(const XprType& replicate)
: m_arg(replicate.nestedExpression()),
m_argImpl(m_arg),
m_rows(replicate.nestedExpression().rows()),
m_cols(replicate.nestedExpression().cols())
{}
EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
CoeffReturnType coeff(Index row, Index col) const
{
// try to avoid using modulo; this is a pure optimization strategy
const Index actual_row = internal::traits<XprType>::RowsAtCompileTime==1 ? 0
: RowFactor==1 ? row
: row % m_rows.value();
const Index actual_col = internal::traits<XprType>::ColsAtCompileTime==1 ? 0
: ColFactor==1 ? col
: col % m_cols.value();
return m_argImpl.coeff(actual_row, actual_col);
}
EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
CoeffReturnType coeff(Index index) const
{
// try to avoid using modulo; this is a pure optimization strategy
const Index actual_index = internal::traits<XprType>::RowsAtCompileTime==1
? (ColFactor==1 ? index : index%m_cols.value())
: (RowFactor==1 ? index : index%m_rows.value());
return m_argImpl.coeff(actual_index);
}
template<int LoadMode, typename PacketType>
EIGEN_STRONG_INLINE
PacketType packet(Index row, Index col) const
{
const Index actual_row = internal::traits<XprType>::RowsAtCompileTime==1 ? 0
: RowFactor==1 ? row
: row % m_rows.value();
const Index actual_col = internal::traits<XprType>::ColsAtCompileTime==1 ? 0
: ColFactor==1 ? col
: col % m_cols.value();
return m_argImpl.template packet<LoadMode,PacketType>(actual_row, actual_col);
}
template<int LoadMode, typename PacketType>
EIGEN_STRONG_INLINE
PacketType packet(Index index) const
{
const Index actual_index = internal::traits<XprType>::RowsAtCompileTime==1
? (ColFactor==1 ? index : index%m_cols.value())
: (RowFactor==1 ? index : index%m_rows.value());
return m_argImpl.template packet<LoadMode,PacketType>(actual_index);
}
protected:
const ArgTypeNested m_arg;
evaluator<ArgTypeNestedCleaned> m_argImpl;
const variable_if_dynamic<Index, ArgType::RowsAtCompileTime> m_rows;
const variable_if_dynamic<Index, ArgType::ColsAtCompileTime> m_cols;
};
// -------------------- PartialReduxExpr --------------------
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::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<InputScalar,int(TraversalSize)> CostOpType;
enum {
CoeffReadCost = TraversalSize==Dynamic ? HugeCost
: TraversalSize * evaluator<ArgType>::CoeffReadCost + int(CostOpType::value),
Flags = (traits<XprType>::Flags&RowMajorBit) | (evaluator<ArgType>::Flags&(HereditaryBits&(~RowMajorBit))) | 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 : 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
{
if (Direction==Vertical)
return m_functor(m_arg.col(j));
else
return m_functor(m_arg.row(i));
}
EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
const Scalar coeff(Index index) const
{
if (Direction==Vertical)
return m_functor(m_arg.col(index));
else
return m_functor(m_arg.row(index));
}
protected:
typename internal::add_const_on_value_type<ArgTypeNested>::type m_arg;
const MemberOp m_functor;
};
// -------------------- MatrixWrapper and ArrayWrapper --------------------
//
// evaluator_wrapper_base<T> is a common base class for the
// MatrixWrapper and ArrayWrapper evaluators.
template<typename XprType>
struct evaluator_wrapper_base
: evaluator_base<XprType>
{
typedef typename remove_all<typename XprType::NestedExpressionType>::type ArgType;
enum {
CoeffReadCost = evaluator<ArgType>::CoeffReadCost,
Flags = evaluator<ArgType>::Flags,
Alignment = evaluator<ArgType>::Alignment
};
EIGEN_DEVICE_FUNC explicit evaluator_wrapper_base(const ArgType& arg) : m_argImpl(arg) {}
typedef typename ArgType::Scalar Scalar;
typedef typename ArgType::CoeffReturnType CoeffReturnType;
EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
CoeffReturnType coeff(Index row, Index col) const
{
return m_argImpl.coeff(row, col);
}
EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
CoeffReturnType coeff(Index index) const
{
return m_argImpl.coeff(index);
}
EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
Scalar& coeffRef(Index row, Index col)
{
return m_argImpl.coeffRef(row, col);
}
EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
Scalar& coeffRef(Index index)
{
return m_argImpl.coeffRef(index);
}
template<int LoadMode, typename PacketType>
EIGEN_STRONG_INLINE
PacketType packet(Index row, Index col) const
{
return m_argImpl.template packet<LoadMode,PacketType>(row, col);
}
template<int LoadMode, typename PacketType>
EIGEN_STRONG_INLINE
PacketType packet(Index index) const
{
return m_argImpl.template packet<LoadMode,PacketType>(index);
}
template<int StoreMode, typename PacketType>
EIGEN_STRONG_INLINE
void writePacket(Index row, Index col, const PacketType& x)
{
m_argImpl.template writePacket<StoreMode>(row, col, x);
}
template<int StoreMode, typename PacketType>
EIGEN_STRONG_INLINE
void writePacket(Index index, const PacketType& x)
{
m_argImpl.template writePacket<StoreMode>(index, x);
}
protected:
evaluator<ArgType> m_argImpl;
};
template<typename TArgType>
struct unary_evaluator<MatrixWrapper<TArgType> >
: evaluator_wrapper_base<MatrixWrapper<TArgType> >
{
typedef MatrixWrapper<TArgType> XprType;
EIGEN_DEVICE_FUNC explicit unary_evaluator(const XprType& wrapper)
: evaluator_wrapper_base<MatrixWrapper<TArgType> >(wrapper.nestedExpression())
{ }
};
template<typename TArgType>
struct unary_evaluator<ArrayWrapper<TArgType> >
: evaluator_wrapper_base<ArrayWrapper<TArgType> >
{
typedef ArrayWrapper<TArgType> XprType;
EIGEN_DEVICE_FUNC explicit unary_evaluator(const XprType& wrapper)
: evaluator_wrapper_base<ArrayWrapper<TArgType> >(wrapper.nestedExpression())
{ }
};
// -------------------- Reverse --------------------
// defined in Reverse.h:
template<typename PacketType, bool ReversePacket> struct reverse_packet_cond;
template<typename ArgType, int Direction>
struct unary_evaluator<Reverse<ArgType, Direction> >
: evaluator_base<Reverse<ArgType, Direction> >
{
typedef Reverse<ArgType, Direction> XprType;
typedef typename XprType::Scalar Scalar;
typedef typename XprType::CoeffReturnType CoeffReturnType;
enum {
IsRowMajor = XprType::IsRowMajor,
IsColMajor = !IsRowMajor,
ReverseRow = (Direction == Vertical) || (Direction == BothDirections),
ReverseCol = (Direction == Horizontal) || (Direction == BothDirections),
ReversePacket = (Direction == BothDirections)
|| ((Direction == Vertical) && IsColMajor)
|| ((Direction == Horizontal) && IsRowMajor),
CoeffReadCost = evaluator<ArgType>::CoeffReadCost,
// let's enable LinearAccess only with vectorization because of the product overhead
// FIXME enable DirectAccess with negative strides?
Flags0 = evaluator<ArgType>::Flags,
LinearAccess = ( (Direction==BothDirections) && (int(Flags0)&PacketAccessBit) )
|| ((ReverseRow && XprType::ColsAtCompileTime==1) || (ReverseCol && XprType::RowsAtCompileTime==1))
? LinearAccessBit : 0,
Flags = int(Flags0) & (HereditaryBits | PacketAccessBit | LinearAccess),
Alignment = 0 // FIXME in some rare cases, Alignment could be preserved, like a Vector4f.
};
EIGEN_DEVICE_FUNC explicit unary_evaluator(const XprType& reverse)
: m_argImpl(reverse.nestedExpression()),
m_rows(ReverseRow ? reverse.nestedExpression().rows() : 1),
m_cols(ReverseCol ? reverse.nestedExpression().cols() : 1)
{ }
EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
CoeffReturnType coeff(Index row, Index col) const
{
return m_argImpl.coeff(ReverseRow ? m_rows.value() - row - 1 : row,
ReverseCol ? m_cols.value() - col - 1 : col);
}
EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
CoeffReturnType coeff(Index index) const
{
return m_argImpl.coeff(m_rows.value() * m_cols.value() - index - 1);
}
EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
Scalar& coeffRef(Index row, Index col)
{
return m_argImpl.coeffRef(ReverseRow ? m_rows.value() - row - 1 : row,
ReverseCol ? m_cols.value() - col - 1 : col);
}
EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
Scalar& coeffRef(Index index)
{
return m_argImpl.coeffRef(m_rows.value() * m_cols.value() - index - 1);
}
template<int LoadMode, typename PacketType>
EIGEN_STRONG_INLINE
PacketType packet(Index row, Index col) const
{
enum {
PacketSize = unpacket_traits<PacketType>::size,
OffsetRow = ReverseRow && IsColMajor ? PacketSize : 1,
OffsetCol = ReverseCol && IsRowMajor ? PacketSize : 1
};
typedef internal::reverse_packet_cond<PacketType,ReversePacket> reverse_packet;
return reverse_packet::run(m_argImpl.template packet<LoadMode,PacketType>(
ReverseRow ? m_rows.value() - row - OffsetRow : row,
ReverseCol ? m_cols.value() - col - OffsetCol : col));
}
template<int LoadMode, typename PacketType>
EIGEN_STRONG_INLINE
PacketType packet(Index index) const
{
enum { PacketSize = unpacket_traits<PacketType>::size };
return preverse(m_argImpl.template packet<LoadMode,PacketType>(m_rows.value() * m_cols.value() - index - PacketSize));
}
template<int LoadMode, typename PacketType>
EIGEN_STRONG_INLINE
void writePacket(Index row, Index col, const PacketType& x)
{
// FIXME we could factorize some code with packet(i,j)
enum {
PacketSize = unpacket_traits<PacketType>::size,
OffsetRow = ReverseRow && IsColMajor ? PacketSize : 1,
OffsetCol = ReverseCol && IsRowMajor ? PacketSize : 1
};
typedef internal::reverse_packet_cond<PacketType,ReversePacket> reverse_packet;
m_argImpl.template writePacket<LoadMode>(
ReverseRow ? m_rows.value() - row - OffsetRow : row,
ReverseCol ? m_cols.value() - col - OffsetCol : col,
reverse_packet::run(x));
}
template<int LoadMode, typename PacketType>
EIGEN_STRONG_INLINE
void writePacket(Index index, const PacketType& x)
{
enum { PacketSize = unpacket_traits<PacketType>::size };
m_argImpl.template writePacket<LoadMode>
(m_rows.value() * m_cols.value() - index - PacketSize, preverse(x));
}
protected:
evaluator<ArgType> m_argImpl;
// If we do not reverse rows, then we do not need to know the number of rows; same for columns
// Nonetheless, in this case it is important to set to 1 such that the coeff(index) method works fine for vectors.
const variable_if_dynamic<Index, ReverseRow ? ArgType::RowsAtCompileTime : 1> m_rows;
const variable_if_dynamic<Index, ReverseCol ? ArgType::ColsAtCompileTime : 1> m_cols;
};
// -------------------- Diagonal --------------------
template<typename ArgType, int DiagIndex>
struct evaluator<Diagonal<ArgType, DiagIndex> >
: evaluator_base<Diagonal<ArgType, DiagIndex> >
{
typedef Diagonal<ArgType, DiagIndex> XprType;
enum {
CoeffReadCost = evaluator<ArgType>::CoeffReadCost,
Flags = (unsigned int)(evaluator<ArgType>::Flags & (HereditaryBits | DirectAccessBit) & ~RowMajorBit) | LinearAccessBit,
Alignment = 0
};
EIGEN_DEVICE_FUNC explicit evaluator(const XprType& diagonal)
: m_argImpl(diagonal.nestedExpression()),
m_index(diagonal.index())
{ }
typedef typename XprType::Scalar Scalar;
typedef typename XprType::CoeffReturnType CoeffReturnType;
EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
CoeffReturnType coeff(Index row, Index) const
{
return m_argImpl.coeff(row + rowOffset(), row + colOffset());
}
EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
CoeffReturnType coeff(Index index) const
{
return m_argImpl.coeff(index + rowOffset(), index + colOffset());
}
EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
Scalar& coeffRef(Index row, Index)
{
return m_argImpl.coeffRef(row + rowOffset(), row + colOffset());
}
EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
Scalar& coeffRef(Index index)
{
return m_argImpl.coeffRef(index + rowOffset(), index + colOffset());
}
protected:
evaluator<ArgType> m_argImpl;
const internal::variable_if_dynamicindex<Index, XprType::DiagIndex> m_index;
private:
EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE Index rowOffset() const { return m_index.value() > 0 ? 0 : -m_index.value(); }
EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE Index colOffset() const { return m_index.value() > 0 ? m_index.value() : 0; }
};
//----------------------------------------------------------------------
// deprecated code
//----------------------------------------------------------------------
// -------------------- EvalToTemp --------------------
// expression class for evaluating nested expression to a temporary
template<typename ArgType> class EvalToTemp;
template<typename ArgType>
struct traits<EvalToTemp<ArgType> >
: public traits<ArgType>
{ };
template<typename ArgType>
class EvalToTemp
: public dense_xpr_base<EvalToTemp<ArgType> >::type
{
public:
typedef typename dense_xpr_base<EvalToTemp>::type Base;
EIGEN_GENERIC_PUBLIC_INTERFACE(EvalToTemp)
explicit EvalToTemp(const ArgType& arg)
: m_arg(arg)
{ }
const ArgType& arg() const
{
return m_arg;
}
Index rows() const
{
return m_arg.rows();
}
Index cols() const
{
return m_arg.cols();
}
private:
const ArgType& m_arg;
};
template<typename ArgType>
struct evaluator<EvalToTemp<ArgType> >
: public evaluator<typename ArgType::PlainObject>
{
typedef EvalToTemp<ArgType> XprType;
typedef typename ArgType::PlainObject PlainObject;
typedef evaluator<PlainObject> Base;
EIGEN_DEVICE_FUNC explicit evaluator(const XprType& xpr)
: m_result(xpr.arg())
{
::new (static_cast<Base*>(this)) Base(m_result);
}
// This constructor is used when nesting an EvalTo evaluator in another evaluator
EIGEN_DEVICE_FUNC evaluator(const ArgType& arg)
: m_result(arg)
{
::new (static_cast<Base*>(this)) Base(m_result);
}
protected:
PlainObject m_result;
};
} // namespace internal
} // end namespace Eigen
#endif // EIGEN_COREEVALUATORS_H
external/eigen3/Eigen/src/Core/CoreIterators.h
View file @ a394b22a
// 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) 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
......@@ -15,47 +15,113 @@ namespace Eigen {
/* This file contains the respective InnerIterator definition of the expressions defined in Eigen/Core
*/
/** \ingroup SparseCore_Module
* \class InnerIterator
* \brief An InnerIterator allows to loop over the element of a sparse (or dense) matrix or expression
*
* todo
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; }
/// \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 version for dense matrix and expressions
template<typename Derived> class DenseBase<Derived>::InnerIterator
// Generic inner iterator implementation for dense objects
template<typename XprType>
class inner_iterator_selector<XprType, IndexBased>
{
protected:
typedef typename Derived::Scalar Scalar;
typedef typename Derived::Index Index;
enum { IsRowMajor = (Derived::Flags&RowMajorBit)==RowMajorBit };
public:
EIGEN_STRONG_INLINE InnerIterator(const Derived& expr, Index outer)
: m_expression(expr), m_inner(0), m_outer(outer), m_end(expr.innerSize())
{}
EIGEN_STRONG_INLINE Scalar value() const
{
return (IsRowMajor) ? m_expression.coeff(m_outer, m_inner)
: m_expression.coeff(m_inner, m_outer);
}
EIGEN_STRONG_INLINE InnerIterator& 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 Derived& m_expression;
Index m_inner;
const Index m_outer;
const Index m_end;
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
external/eigen3/Eigen/src/Core/CwiseBinaryOp.h
View file @ a394b22a
// This file is part of Eigen, a lightweight C++ template library
// for linear algebra.
//
// Copyright (C) 2008-2009 Gael Guennebaud <gael.guennebaud@inria.fr>
// 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
......@@ -13,26 +13,6 @@
namespace Eigen {
/** \class CwiseBinaryOp
* \ingroup Core_Module
*
* \brief Generic expression where a coefficient-wise binary operator is applied to two expressions
*
* \param BinaryOp template functor implementing the operator
* \param Lhs the type of the left-hand side
* \param Rhs 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
*/
namespace internal {
template<typename BinaryOp, typename Lhs, typename Rhs>
struct traits<CwiseBinaryOp<BinaryOp, Lhs, Rhs> >
......@@ -52,77 +32,75 @@ struct traits<CwiseBinaryOp<BinaryOp, Lhs, Rhs> >
// we still want to handle the case when the result type is different.
typedef typename result_of<
BinaryOp(
typename Lhs::Scalar,
typename Rhs::Scalar
const typename Lhs::Scalar&,
const typename Rhs::Scalar&
)
>::type Scalar;
typedef typename promote_storage_type<typename traits<Lhs>::StorageKind,
typename traits<Rhs>::StorageKind>::ret StorageKind;
typedef typename promote_index_type<typename traits<Lhs>::Index,
typename traits<Rhs>::Index>::type Index;
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 {
LhsCoeffReadCost = _LhsNested::CoeffReadCost,
RhsCoeffReadCost = _RhsNested::CoeffReadCost,
LhsFlags = _LhsNested::Flags,
RhsFlags = _RhsNested::Flags,
SameType = is_same<typename _LhsNested::Scalar,typename _RhsNested::Scalar>::value,
StorageOrdersAgree = (int(Lhs::Flags)&RowMajorBit)==(int(Rhs::Flags)&RowMajorBit),
Flags0 = (int(LhsFlags) | int(RhsFlags)) & (
HereditaryBits
| (int(LhsFlags) & int(RhsFlags) &
( AlignedBit
| (StorageOrdersAgree ? LinearAccessBit : 0)
| (functor_traits<BinaryOp>::PacketAccess && StorageOrdersAgree && SameType ? PacketAccessBit : 0)
)
)
),
Flags = (Flags0 & ~RowMajorBit) | (LhsFlags & RowMajorBit),
Cost0 = EIGEN_ADD_COST(LhsCoeffReadCost,RhsCoeffReadCost),
CoeffReadCost = EIGEN_ADD_COST(Cost0,functor_traits<BinaryOp>::Cost)
Flags = cwise_promote_storage_order<typename traits<Lhs>::StorageKind,typename traits<Rhs>::StorageKind,_LhsNested::Flags & RowMajorBit,_RhsNested::Flags & RowMajorBit>::value
};
};
} // end namespace internal
// we require Lhs and Rhs to have the same scalar type. Currently there is no example of a binary functor
// that would take two operands of different types. If there were such an example, then this check should be
// moved to the BinaryOp functors, on a per-case basis. This would however require a change in the BinaryOp functors, as
// currently they take only one typename Scalar template parameter.
// It is tempting to always allow mixing different types but remember that this is often impossible in the vectorized paths.
// So allowing mixing different types gives very unexpected errors when enabling vectorization, when the user tries to
// add together a float matrix and a double matrix.
#define EIGEN_CHECK_BINARY_COMPATIBILIY(BINOP,LHS,RHS) \
EIGEN_STATIC_ASSERT((internal::functor_is_product_like<BINOP>::ret \
? int(internal::scalar_product_traits<LHS, RHS>::Defined) \
: int(internal::is_same<LHS, RHS>::value)), \
YOU_MIXED_DIFFERENT_NUMERIC_TYPES__YOU_NEED_TO_USE_THE_CAST_METHOD_OF_MATRIXBASE_TO_CAST_NUMERIC_TYPES_EXPLICITLY)
template<typename BinaryOp, typename Lhs, typename Rhs, typename StorageKind>
class CwiseBinaryOpImpl;
template<typename BinaryOp, typename Lhs, typename Rhs>
class CwiseBinaryOp : internal::no_assignment_operator,
/** \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, Lhs, Rhs,
typename internal::promote_storage_type<typename internal::traits<Lhs>::StorageKind,
typename internal::traits<Rhs>::StorageKind>::ret>
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, Lhs, Rhs,
typename internal::promote_storage_type<typename internal::traits<Lhs>::StorageKind,
typename internal::traits<Rhs>::StorageKind>::ret>::Base Base;
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::nested<Lhs>::type LhsNested;
typedef typename internal::nested<Rhs>::type RhsNested;
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;
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)
{
......@@ -132,6 +110,7 @@ class CwiseBinaryOp : internal::no_assignment_operator,
eigen_assert(aLhs.rows() == aRhs.rows() && aLhs.cols() == aRhs.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<LhsNested>::type>::RowsAtCompileTime==Dynamic)
......@@ -139,6 +118,7 @@ class CwiseBinaryOp : internal::no_assignment_operator,
else
return m_lhs.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<LhsNested>::type>::ColsAtCompileTime==Dynamic)
......@@ -148,10 +128,13 @@ class CwiseBinaryOp : internal::no_assignment_operator,
}
/** \returns the left hand side nested expression */
EIGEN_DEVICE_FUNC
const _LhsNested& lhs() const { return m_lhs; }
/** \returns the right hand side nested expression */
EIGEN_DEVICE_FUNC
const _RhsNested& rhs() const { return m_rhs; }
/** \returns the functor representing the binary operation */
EIGEN_DEVICE_FUNC
const BinaryOp& functor() const { return m_functor; }
protected:
......@@ -160,41 +143,13 @@ class CwiseBinaryOp : internal::no_assignment_operator,
const BinaryOp m_functor;
};
template<typename BinaryOp, typename Lhs, typename Rhs>
class CwiseBinaryOpImpl<BinaryOp, Lhs, Rhs, Dense>
: public internal::dense_xpr_base<CwiseBinaryOp<BinaryOp, Lhs, Rhs> >::type
// Generic API dispatcher
template<typename BinaryOp, typename Lhs, typename Rhs, typename StorageKind>
class CwiseBinaryOpImpl
: public internal::generic_xpr_base<CwiseBinaryOp<BinaryOp, Lhs, Rhs> >::type
{
typedef CwiseBinaryOp<BinaryOp, Lhs, Rhs> Derived;
public:
typedef typename internal::dense_xpr_base<CwiseBinaryOp<BinaryOp, Lhs, Rhs> >::type Base;
EIGEN_DENSE_PUBLIC_INTERFACE( Derived )
EIGEN_STRONG_INLINE const Scalar coeff(Index rowId, Index colId) const
{
return derived().functor()(derived().lhs().coeff(rowId, colId),
derived().rhs().coeff(rowId, colId));
}
template<int LoadMode>
EIGEN_STRONG_INLINE PacketScalar packet(Index rowId, Index colId) const
{
return derived().functor().packetOp(derived().lhs().template packet<LoadMode>(rowId, colId),
derived().rhs().template packet<LoadMode>(rowId, colId));
}
EIGEN_STRONG_INLINE const Scalar coeff(Index index) const
{
return derived().functor()(derived().lhs().coeff(index),
derived().rhs().coeff(index));
}
template<int LoadMode>
EIGEN_STRONG_INLINE PacketScalar packet(Index index) const
{
return derived().functor().packetOp(derived().lhs().template packet<LoadMode>(index),
derived().rhs().template packet<LoadMode>(index));
}
public:
typedef typename internal::generic_xpr_base<CwiseBinaryOp<BinaryOp, Lhs, Rhs> >::type Base;
};
/** replaces \c *this by \c *this - \a other.
......@@ -206,8 +161,7 @@ template<typename OtherDerived>
EIGEN_STRONG_INLINE Derived &
MatrixBase<Derived>::operator-=(const MatrixBase<OtherDerived> &other)
{
SelfCwiseBinaryOp<internal::scalar_difference_op<Scalar>, Derived, OtherDerived> tmp(derived());
tmp = other.derived();
call_assignment(derived(), other.derived(), internal::sub_assign_op<Scalar,typename OtherDerived::Scalar>());
return derived();
}
......@@ -220,11 +174,11 @@ template<typename OtherDerived>
EIGEN_STRONG_INLINE Derived &
MatrixBase<Derived>::operator+=(const MatrixBase<OtherDerived>& other)
{
SelfCwiseBinaryOp<internal::scalar_sum_op<Scalar>, Derived, OtherDerived> tmp(derived());
tmp = other.derived();
call_assignment(derived(), other.derived(), internal::add_assign_op<Scalar,typename OtherDerived::Scalar>());
return derived();
}
} // end namespace Eigen
#endif // EIGEN_CWISE_BINARY_OP_H
external/eigen3/Eigen/src/Core/CwiseNullaryOp.h
View file @ a394b22a
......@@ -12,13 +12,24 @@
namespace Eigen {
namespace internal {
template<typename NullaryOp, typename PlainObjectType>
struct traits<CwiseNullaryOp<NullaryOp, PlainObjectType> > : traits<PlainObjectType>
{
enum {
Flags = traits<PlainObjectType>::Flags & RowMajorBit
};
};
} // namespace internal
/** \class CwiseNullaryOp
* \ingroup Core_Module
*
* \brief Generic expression of a matrix where all coefficients are defined by a functor
*
* \param NullaryOp template functor implementing the operator
* \param PlainObjectType the underlying plain matrix/array type
* \tparam NullaryOp template functor implementing the operator
* \tparam PlainObjectType the underlying plain matrix/array type
*
* This class represents an expression of a generic nullary operator.
* It is the return type of the Ones(), Zero(), Constant(), Identity() and Random() methods,
......@@ -27,68 +38,49 @@ namespace Eigen {
* However, if you want to write a function returning such an expression, you
* will need to use this class.
*
* \sa class CwiseUnaryOp, class CwiseBinaryOp, DenseBase::NullaryExpr()
* The functor NullaryOp must expose one of the following method:
<table class="manual">
<tr ><td>\c operator()() </td><td>if the procedural generation does not depend on the coefficient entries (e.g., random numbers)</td></tr>
<tr class="alt"><td>\c operator()(Index i)</td><td>if the procedural generation makes sense for vectors only and that it depends on the coefficient index \c i (e.g., linspace) </td></tr>
<tr ><td>\c operator()(Index i,Index j)</td><td>if the procedural generation depends on the matrix coordinates \c i, \c j (e.g., to generate a checkerboard with 0 and 1)</td></tr>
</table>
* It is also possible to expose the last two operators if the generation makes sense for matrices but can be optimized for vectors.
*
* See DenseBase::NullaryExpr(Index,const CustomNullaryOp&) for an example binding
* C++11 random number generators.
*
* A nullary expression can also be used to implement custom sophisticated matrix manipulations
* that cannot be covered by the existing set of natively supported matrix manipulations.
* See this \ref TopicCustomizing_NullaryExpr "page" for some examples and additional explanations
* on the behavior of CwiseNullaryOp.
*
* \sa class CwiseUnaryOp, class CwiseBinaryOp, DenseBase::NullaryExpr
*/
namespace internal {
template<typename NullaryOp, typename PlainObjectType>
struct traits<CwiseNullaryOp<NullaryOp, PlainObjectType> > : traits<PlainObjectType>
{
enum {
Flags = (traits<PlainObjectType>::Flags
& ( HereditaryBits
| (functor_has_linear_access<NullaryOp>::ret ? LinearAccessBit : 0)
| (functor_traits<NullaryOp>::PacketAccess ? PacketAccessBit : 0)))
| (functor_traits<NullaryOp>::IsRepeatable ? 0 : EvalBeforeNestingBit),
CoeffReadCost = functor_traits<NullaryOp>::Cost
};
};
}
template<typename NullaryOp, typename PlainObjectType>
class CwiseNullaryOp : internal::no_assignment_operator,
public internal::dense_xpr_base< CwiseNullaryOp<NullaryOp, PlainObjectType> >::type
class CwiseNullaryOp : public internal::dense_xpr_base< CwiseNullaryOp<NullaryOp, PlainObjectType> >::type, internal::no_assignment_operator
{
public:
typedef typename internal::dense_xpr_base<CwiseNullaryOp>::type Base;
EIGEN_DENSE_PUBLIC_INTERFACE(CwiseNullaryOp)
CwiseNullaryOp(Index nbRows, Index nbCols, const NullaryOp& func = NullaryOp())
: m_rows(nbRows), m_cols(nbCols), m_functor(func)
EIGEN_DEVICE_FUNC
CwiseNullaryOp(Index rows, Index cols, const NullaryOp& func = NullaryOp())
: m_rows(rows), m_cols(cols), m_functor(func)
{
eigen_assert(nbRows >= 0
&& (RowsAtCompileTime == Dynamic || RowsAtCompileTime == nbRows)
&& nbCols >= 0
&& (ColsAtCompileTime == Dynamic || ColsAtCompileTime == nbCols));
eigen_assert(rows >= 0
&& (RowsAtCompileTime == Dynamic || RowsAtCompileTime == rows)
&& cols >= 0
&& (ColsAtCompileTime == Dynamic || ColsAtCompileTime == cols));
}
EIGEN_DEVICE_FUNC
EIGEN_STRONG_INLINE Index rows() const { return m_rows.value(); }
EIGEN_DEVICE_FUNC
EIGEN_STRONG_INLINE Index cols() const { return m_cols.value(); }
EIGEN_STRONG_INLINE const Scalar coeff(Index rowId, Index colId) const
{
return m_functor(rowId, colId);
}
template<int LoadMode>
EIGEN_STRONG_INLINE PacketScalar packet(Index rowId, Index colId) const
{
return m_functor.packetOp(rowId, colId);
}
EIGEN_STRONG_INLINE const Scalar coeff(Index index) const
{
return m_functor(index);
}
template<int LoadMode>
EIGEN_STRONG_INLINE PacketScalar packet(Index index) const
{
return m_functor.packetOp(index);
}
/** \returns the functor representing the nullary operation */
EIGEN_DEVICE_FUNC
const NullaryOp& functor() const { return m_functor; }
protected:
......@@ -113,10 +105,10 @@ class CwiseNullaryOp : internal::no_assignment_operator,
*/
template<typename Derived>
template<typename CustomNullaryOp>
EIGEN_STRONG_INLINE const CwiseNullaryOp<CustomNullaryOp, Derived>
EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE const CwiseNullaryOp<CustomNullaryOp, typename DenseBase<Derived>::PlainObject>
DenseBase<Derived>::NullaryExpr(Index rows, Index cols, const CustomNullaryOp& func)
{
return CwiseNullaryOp<CustomNullaryOp, Derived>(rows, cols, func);
return CwiseNullaryOp<CustomNullaryOp, PlainObject>(rows, cols, func);
}
/** \returns an expression of a matrix defined by a custom functor \a func
......@@ -132,16 +124,19 @@ DenseBase<Derived>::NullaryExpr(Index rows, Index cols, const CustomNullaryOp& f
*
* The template parameter \a CustomNullaryOp is the type of the functor.
*
* Here is an example with C++11 random generators: \include random_cpp11.cpp
* Output: \verbinclude random_cpp11.out
*
* \sa class CwiseNullaryOp
*/
template<typename Derived>
template<typename CustomNullaryOp>
EIGEN_STRONG_INLINE const CwiseNullaryOp<CustomNullaryOp, Derived>
EIGEN_STRONG_INLINE const CwiseNullaryOp<CustomNullaryOp, typename DenseBase<Derived>::PlainObject>
DenseBase<Derived>::NullaryExpr(Index size, const CustomNullaryOp& func)
{
EIGEN_STATIC_ASSERT_VECTOR_ONLY(Derived)
if(RowsAtCompileTime == 1) return CwiseNullaryOp<CustomNullaryOp, Derived>(1, size, func);
else return CwiseNullaryOp<CustomNullaryOp, Derived>(size, 1, func);
if(RowsAtCompileTime == 1) return CwiseNullaryOp<CustomNullaryOp, PlainObject>(1, size, func);
else return CwiseNullaryOp<CustomNullaryOp, PlainObject>(size, 1, func);
}
/** \returns an expression of a matrix defined by a custom functor \a func
......@@ -155,19 +150,19 @@ DenseBase<Derived>::NullaryExpr(Index size, const CustomNullaryOp& func)
*/
template<typename Derived>
template<typename CustomNullaryOp>
EIGEN_STRONG_INLINE const CwiseNullaryOp<CustomNullaryOp, Derived>
EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE const CwiseNullaryOp<CustomNullaryOp, typename DenseBase<Derived>::PlainObject>
DenseBase<Derived>::NullaryExpr(const CustomNullaryOp& func)
{
return CwiseNullaryOp<CustomNullaryOp, Derived>(RowsAtCompileTime, ColsAtCompileTime, func);
return CwiseNullaryOp<CustomNullaryOp, PlainObject>(RowsAtCompileTime, ColsAtCompileTime, func);
}
/** \returns an expression of a constant matrix of value \a value
*
* The parameters \a nbRows and \a nbCols are the number of rows and of columns of
* The parameters \a rows and \a cols are the number of rows and of columns of
* the returned matrix. Must be compatible with this DenseBase type.
*
* This variant is meant to be used for dynamic-size matrix types. For fixed-size types,
* it is redundant to pass \a nbRows and \a nbCols as arguments, so Zero() should be used
* it is redundant to pass \a rows and \a cols as arguments, so Zero() should be used
* instead.
*
* The template parameter \a CustomNullaryOp is the type of the functor.
......@@ -176,9 +171,9 @@ DenseBase<Derived>::NullaryExpr(const CustomNullaryOp& func)
*/
template<typename Derived>
EIGEN_STRONG_INLINE const typename DenseBase<Derived>::ConstantReturnType
DenseBase<Derived>::Constant(Index nbRows, Index nbCols, const Scalar& value)
DenseBase<Derived>::Constant(Index rows, Index cols, const Scalar& value)
{
return DenseBase<Derived>::NullaryExpr(nbRows, nbCols, internal::scalar_constant_op<Scalar>(value));
return DenseBase<Derived>::NullaryExpr(rows, cols, internal::scalar_constant_op<Scalar>(value));
}
/** \returns an expression of a constant matrix of value \a value
......@@ -197,7 +192,7 @@ DenseBase<Derived>::Constant(Index nbRows, Index nbCols, const Scalar& value)
* \sa class CwiseNullaryOp
*/
template<typename Derived>
EIGEN_STRONG_INLINE const typename DenseBase<Derived>::ConstantReturnType
EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE const typename DenseBase<Derived>::ConstantReturnType
DenseBase<Derived>::Constant(Index size, const Scalar& value)
{
return DenseBase<Derived>::NullaryExpr(size, internal::scalar_constant_op<Scalar>(value));
......@@ -213,53 +208,40 @@ DenseBase<Derived>::Constant(Index size, const Scalar& value)
* \sa class CwiseNullaryOp
*/
template<typename Derived>
EIGEN_STRONG_INLINE const typename DenseBase<Derived>::ConstantReturnType
EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE const typename DenseBase<Derived>::ConstantReturnType
DenseBase<Derived>::Constant(const Scalar& value)
{
EIGEN_STATIC_ASSERT_FIXED_SIZE(Derived)
return DenseBase<Derived>::NullaryExpr(RowsAtCompileTime, ColsAtCompileTime, internal::scalar_constant_op<Scalar>(value));
}
/**
* \brief Sets a linearly space vector.
*
* The function generates 'size' equally spaced values in the closed interval [low,high].
* This particular version of LinSpaced() uses sequential access, i.e. vector access is
* assumed to be a(0), a(1), ..., a(size). This assumption allows for better vectorization
* and yields faster code than the random access version.
*
* When size is set to 1, a vector of length 1 containing 'high' is returned.
*
* \only_for_vectors
*
* Example: \include DenseBase_LinSpaced_seq.cpp
* Output: \verbinclude DenseBase_LinSpaced_seq.out
/** \deprecated because of accuracy loss. In Eigen 3.3, it is an alias for LinSpaced(Index,const Scalar&,const Scalar&)
*
* \sa setLinSpaced(Index,const Scalar&,const Scalar&), LinSpaced(Index,Scalar,Scalar), CwiseNullaryOp
* \sa LinSpaced(Index,Scalar,Scalar), setLinSpaced(Index,const Scalar&,const Scalar&)
*/
template<typename Derived>
EIGEN_STRONG_INLINE const typename DenseBase<Derived>::SequentialLinSpacedReturnType
EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE const typename DenseBase<Derived>::RandomAccessLinSpacedReturnType
DenseBase<Derived>::LinSpaced(Sequential_t, Index size, const Scalar& low, const Scalar& high)
{
EIGEN_STATIC_ASSERT_VECTOR_ONLY(Derived)
return DenseBase<Derived>::NullaryExpr(size, internal::linspaced_op<Scalar,false>(low,high,size));
return DenseBase<Derived>::NullaryExpr(size, internal::linspaced_op<Scalar,PacketScalar>(low,high,size));
}
/**
* \copydoc DenseBase::LinSpaced(Sequential_t, Index, const Scalar&, const Scalar&)
* Special version for fixed size types which does not require the size parameter.
/** \deprecated because of accuracy loss. In Eigen 3.3, it is an alias for LinSpaced(const Scalar&,const Scalar&)
*
* \sa LinSpaced(Scalar,Scalar)
*/
template<typename Derived>
EIGEN_STRONG_INLINE const typename DenseBase<Derived>::SequentialLinSpacedReturnType
EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE const typename DenseBase<Derived>::RandomAccessLinSpacedReturnType
DenseBase<Derived>::LinSpaced(Sequential_t, const Scalar& low, const Scalar& high)
{
EIGEN_STATIC_ASSERT_VECTOR_ONLY(Derived)
EIGEN_STATIC_ASSERT_FIXED_SIZE(Derived)
return DenseBase<Derived>::NullaryExpr(Derived::SizeAtCompileTime, internal::linspaced_op<Scalar,false>(low,high,Derived::SizeAtCompileTime));
return DenseBase<Derived>::NullaryExpr(Derived::SizeAtCompileTime, internal::linspaced_op<Scalar,PacketScalar>(low,high,Derived::SizeAtCompileTime));
}
/**
* \brief Sets a linearly space vector.
* \brief Sets a linearly spaced vector.
*
* The function generates 'size' equally spaced values in the closed interval [low,high].
* When size is set to 1, a vector of length 1 containing 'high' is returned.
......@@ -269,14 +251,24 @@ DenseBase<Derived>::LinSpaced(Sequential_t, const Scalar& low, const Scalar& hig
* Example: \include DenseBase_LinSpaced.cpp
* Output: \verbinclude DenseBase_LinSpaced.out
*
* \sa setLinSpaced(Index,const Scalar&,const Scalar&), LinSpaced(Sequential_t,Index,const Scalar&,const Scalar&,Index), CwiseNullaryOp
* For integer scalar types, an even spacing is possible if and only if the length of the range,
* i.e., \c high-low is a scalar multiple of \c size-1, or if \c size is a scalar multiple of the
* number of values \c high-low+1 (meaning each value can be repeated the same number of time).
* If one of these two considions is not satisfied, then \c high is lowered to the largest value
* satisfying one of this constraint.
* Here are some examples:
*
* Example: \include DenseBase_LinSpacedInt.cpp
* Output: \verbinclude DenseBase_LinSpacedInt.out
*
* \sa setLinSpaced(Index,const Scalar&,const Scalar&), CwiseNullaryOp
*/
template<typename Derived>
EIGEN_STRONG_INLINE const typename DenseBase<Derived>::RandomAccessLinSpacedReturnType
EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE const typename DenseBase<Derived>::RandomAccessLinSpacedReturnType
DenseBase<Derived>::LinSpaced(Index size, const Scalar& low, const Scalar& high)
{
EIGEN_STATIC_ASSERT_VECTOR_ONLY(Derived)
return DenseBase<Derived>::NullaryExpr(size, internal::linspaced_op<Scalar,true>(low,high,size));
return DenseBase<Derived>::NullaryExpr(size, internal::linspaced_op<Scalar,PacketScalar>(low,high,size));
}
/**
......@@ -284,22 +276,23 @@ DenseBase<Derived>::LinSpaced(Index size, const Scalar& low, const Scalar& high)
* Special version for fixed size types which does not require the size parameter.
*/
template<typename Derived>
EIGEN_STRONG_INLINE const typename DenseBase<Derived>::RandomAccessLinSpacedReturnType
EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE const typename DenseBase<Derived>::RandomAccessLinSpacedReturnType
DenseBase<Derived>::LinSpaced(const Scalar& low, const Scalar& high)
{
EIGEN_STATIC_ASSERT_VECTOR_ONLY(Derived)
EIGEN_STATIC_ASSERT_FIXED_SIZE(Derived)
return DenseBase<Derived>::NullaryExpr(Derived::SizeAtCompileTime, internal::linspaced_op<Scalar,true>(low,high,Derived::SizeAtCompileTime));
return DenseBase<Derived>::NullaryExpr(Derived::SizeAtCompileTime, internal::linspaced_op<Scalar,PacketScalar>(low,high,Derived::SizeAtCompileTime));
}
/** \returns true if all coefficients in this matrix are approximately equal to \a val, to within precision \a prec */
template<typename Derived>
bool DenseBase<Derived>::isApproxToConstant
EIGEN_DEVICE_FUNC bool DenseBase<Derived>::isApproxToConstant
(const Scalar& val, const RealScalar& prec) const
{
typename internal::nested_eval<Derived,1>::type self(derived());
for(Index j = 0; j < cols(); ++j)
for(Index i = 0; i < rows(); ++i)
if(!internal::isApprox(this->coeff(i, j), val, prec))
if(!internal::isApprox(self.coeff(i, j), val, prec))
return false;
return true;
}
......@@ -308,7 +301,7 @@ bool DenseBase<Derived>::isApproxToConstant
*
* \returns true if all coefficients in this matrix are approximately equal to \a value, to within precision \a prec */
template<typename Derived>
bool DenseBase<Derived>::isConstant
EIGEN_DEVICE_FUNC bool DenseBase<Derived>::isConstant
(const Scalar& val, const RealScalar& prec) const
{
return isApproxToConstant(val, prec);
......@@ -319,22 +312,22 @@ bool DenseBase<Derived>::isConstant
* \sa setConstant(), Constant(), class CwiseNullaryOp
*/
template<typename Derived>
EIGEN_STRONG_INLINE void DenseBase<Derived>::fill(const Scalar& val)
EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE void DenseBase<Derived>::fill(const Scalar& val)
{
setConstant(val);
}
/** Sets all coefficients in this expression to \a value.
/** Sets all coefficients in this expression to value \a val.
*
* \sa fill(), setConstant(Index,const Scalar&), setConstant(Index,Index,const Scalar&), setZero(), setOnes(), Constant(), class CwiseNullaryOp, setZero(), setOnes()
*/
template<typename Derived>
EIGEN_STRONG_INLINE Derived& DenseBase<Derived>::setConstant(const Scalar& val)
EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE Derived& DenseBase<Derived>::setConstant(const Scalar& val)
{
return derived() = Constant(rows(), cols(), val);
}
/** Resizes to the given \a size, and sets all coefficients in this expression to the given \a value.
/** Resizes to the given \a size, and sets all coefficients in this expression to the given value \a val.
*
* \only_for_vectors
*
......@@ -344,17 +337,17 @@ EIGEN_STRONG_INLINE Derived& DenseBase<Derived>::setConstant(const Scalar& val)
* \sa MatrixBase::setConstant(const Scalar&), setConstant(Index,Index,const Scalar&), class CwiseNullaryOp, MatrixBase::Constant(const Scalar&)
*/
template<typename Derived>
EIGEN_STRONG_INLINE Derived&
EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE Derived&
PlainObjectBase<Derived>::setConstant(Index size, const Scalar& val)
{
resize(size);
return setConstant(val);
}
/** Resizes to the given size, and sets all coefficients in this expression to the given \a value.
/** Resizes to the given size, and sets all coefficients in this expression to the given value \a val.
*
* \param nbRows the new number of rows
* \param nbCols the new number of columns
* \param rows the new number of rows
* \param cols the new number of columns
* \param val the value to which all coefficients are set
*
* Example: \include Matrix_setConstant_int_int.cpp
......@@ -363,15 +356,15 @@ PlainObjectBase<Derived>::setConstant(Index size, const Scalar& val)
* \sa MatrixBase::setConstant(const Scalar&), setConstant(Index,const Scalar&), class CwiseNullaryOp, MatrixBase::Constant(const Scalar&)
*/
template<typename Derived>
EIGEN_STRONG_INLINE Derived&
PlainObjectBase<Derived>::setConstant(Index nbRows, Index nbCols, const Scalar& val)
EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE Derived&
PlainObjectBase<Derived>::setConstant(Index rows, Index cols, const Scalar& val)
{
resize(nbRows, nbCols);
resize(rows, cols);
return setConstant(val);
}
/**
* \brief Sets a linearly space vector.
* \brief Sets a linearly spaced vector.
*
* The function generates 'size' equally spaced values in the closed interval [low,high].
* When size is set to 1, a vector of length 1 containing 'high' is returned.
......@@ -381,27 +374,33 @@ PlainObjectBase<Derived>::setConstant(Index nbRows, Index nbCols, const Scalar&
* Example: \include DenseBase_setLinSpaced.cpp
* Output: \verbinclude DenseBase_setLinSpaced.out
*
* \sa CwiseNullaryOp
* For integer scalar types, do not miss the explanations on the definition
* of \link LinSpaced(Index,const Scalar&,const Scalar&) even spacing \endlink.
*
* \sa LinSpaced(Index,const Scalar&,const Scalar&), CwiseNullaryOp
*/
template<typename Derived>
EIGEN_STRONG_INLINE Derived& DenseBase<Derived>::setLinSpaced(Index newSize, const Scalar& low, const Scalar& high)
EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE Derived& DenseBase<Derived>::setLinSpaced(Index newSize, const Scalar& low, const Scalar& high)
{
EIGEN_STATIC_ASSERT_VECTOR_ONLY(Derived)
return derived() = Derived::NullaryExpr(newSize, internal::linspaced_op<Scalar,false>(low,high,newSize));
return derived() = Derived::NullaryExpr(newSize, internal::linspaced_op<Scalar,PacketScalar>(low,high,newSize));
}
/**
* \brief Sets a linearly space vector.
* \brief Sets a linearly spaced vector.
*
* The function fill *this with equally spaced values in the closed interval [low,high].
* The function fills \c *this with equally spaced values in the closed interval [low,high].
* When size is set to 1, a vector of length 1 containing 'high' is returned.
*
* \only_for_vectors
*
* \sa setLinSpaced(Index, const Scalar&, const Scalar&), CwiseNullaryOp
* For integer scalar types, do not miss the explanations on the definition
* of \link LinSpaced(Index,const Scalar&,const Scalar&) even spacing \endlink.
*
* \sa LinSpaced(Index,const Scalar&,const Scalar&), setLinSpaced(Index, const Scalar&, const Scalar&), CwiseNullaryOp
*/
template<typename Derived>
EIGEN_STRONG_INLINE Derived& DenseBase<Derived>::setLinSpaced(const Scalar& low, const Scalar& high)
EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE Derived& DenseBase<Derived>::setLinSpaced(const Scalar& low, const Scalar& high)
{
EIGEN_STATIC_ASSERT_VECTOR_ONLY(Derived)
return setLinSpaced(size(), low, high);
......@@ -424,10 +423,10 @@ EIGEN_STRONG_INLINE Derived& DenseBase<Derived>::setLinSpaced(const Scalar& low,
* \sa Zero(), Zero(Index)
*/
template<typename Derived>
EIGEN_STRONG_INLINE const typename DenseBase<Derived>::ConstantReturnType
DenseBase<Derived>::Zero(Index nbRows, Index nbCols)
EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE const typename DenseBase<Derived>::ConstantReturnType
DenseBase<Derived>::Zero(Index rows, Index cols)
{
return Constant(nbRows, nbCols, Scalar(0));
return Constant(rows, cols, Scalar(0));
}
/** \returns an expression of a zero vector.
......@@ -447,7 +446,7 @@ DenseBase<Derived>::Zero(Index nbRows, Index nbCols)
* \sa Zero(), Zero(Index,Index)
*/
template<typename Derived>
EIGEN_STRONG_INLINE const typename DenseBase<Derived>::ConstantReturnType
EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE const typename DenseBase<Derived>::ConstantReturnType
DenseBase<Derived>::Zero(Index size)
{
return Constant(size, Scalar(0));
......@@ -464,7 +463,7 @@ DenseBase<Derived>::Zero(Index size)
* \sa Zero(Index), Zero(Index,Index)
*/
template<typename Derived>
EIGEN_STRONG_INLINE const typename DenseBase<Derived>::ConstantReturnType
EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE const typename DenseBase<Derived>::ConstantReturnType
DenseBase<Derived>::Zero()
{
return Constant(Scalar(0));
......@@ -479,11 +478,12 @@ DenseBase<Derived>::Zero()
* \sa class CwiseNullaryOp, Zero()
*/
template<typename Derived>
bool DenseBase<Derived>::isZero(const RealScalar& prec) const
EIGEN_DEVICE_FUNC bool DenseBase<Derived>::isZero(const RealScalar& prec) const
{
typename internal::nested_eval<Derived,1>::type self(derived());
for(Index j = 0; j < cols(); ++j)
for(Index i = 0; i < rows(); ++i)
if(!internal::isMuchSmallerThan(this->coeff(i, j), static_cast<Scalar>(1), prec))
if(!internal::isMuchSmallerThan(self.coeff(i, j), static_cast<Scalar>(1), prec))
return false;
return true;
}
......@@ -496,7 +496,7 @@ bool DenseBase<Derived>::isZero(const RealScalar& prec) const
* \sa class CwiseNullaryOp, Zero()
*/
template<typename Derived>
EIGEN_STRONG_INLINE Derived& DenseBase<Derived>::setZero()
EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE Derived& DenseBase<Derived>::setZero()
{
return setConstant(Scalar(0));
}
......@@ -511,7 +511,7 @@ EIGEN_STRONG_INLINE Derived& DenseBase<Derived>::setZero()
* \sa DenseBase::setZero(), setZero(Index,Index), class CwiseNullaryOp, DenseBase::Zero()
*/
template<typename Derived>
EIGEN_STRONG_INLINE Derived&
EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE Derived&
PlainObjectBase<Derived>::setZero(Index newSize)
{
resize(newSize);
......@@ -520,8 +520,8 @@ PlainObjectBase<Derived>::setZero(Index newSize)
/** Resizes to the given size, and sets all coefficients in this expression to zero.
*
* \param nbRows the new number of rows
* \param nbCols the new number of columns
* \param rows the new number of rows
* \param cols the new number of columns
*
* Example: \include Matrix_setZero_int_int.cpp
* Output: \verbinclude Matrix_setZero_int_int.out
......@@ -529,10 +529,10 @@ PlainObjectBase<Derived>::setZero(Index newSize)
* \sa DenseBase::setZero(), setZero(Index), class CwiseNullaryOp, DenseBase::Zero()
*/
template<typename Derived>
EIGEN_STRONG_INLINE Derived&
PlainObjectBase<Derived>::setZero(Index nbRows, Index nbCols)
EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE Derived&
PlainObjectBase<Derived>::setZero(Index rows, Index cols)
{
resize(nbRows, nbCols);
resize(rows, cols);
return setConstant(Scalar(0));
}
......@@ -540,7 +540,7 @@ PlainObjectBase<Derived>::setZero(Index nbRows, Index nbCols)
/** \returns an expression of a matrix where all coefficients equal one.
*
* The parameters \a nbRows and \a nbCols are the number of rows and of columns of
* 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.
*
* This variant is meant to be used for dynamic-size matrix types. For fixed-size types,
......@@ -553,10 +553,10 @@ PlainObjectBase<Derived>::setZero(Index nbRows, Index nbCols)
* \sa Ones(), Ones(Index), isOnes(), class Ones
*/
template<typename Derived>
EIGEN_STRONG_INLINE const typename DenseBase<Derived>::ConstantReturnType
DenseBase<Derived>::Ones(Index nbRows, Index nbCols)
EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE const typename DenseBase<Derived>::ConstantReturnType
DenseBase<Derived>::Ones(Index rows, Index cols)
{
return Constant(nbRows, nbCols, Scalar(1));
return Constant(rows, cols, Scalar(1));
}
/** \returns an expression of a vector where all coefficients equal one.
......@@ -576,7 +576,7 @@ DenseBase<Derived>::Ones(Index nbRows, Index nbCols)
* \sa Ones(), Ones(Index,Index), isOnes(), class Ones
*/
template<typename Derived>
EIGEN_STRONG_INLINE const typename DenseBase<Derived>::ConstantReturnType
EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE const typename DenseBase<Derived>::ConstantReturnType
DenseBase<Derived>::Ones(Index newSize)
{
return Constant(newSize, Scalar(1));
......@@ -593,7 +593,7 @@ DenseBase<Derived>::Ones(Index newSize)
* \sa Ones(Index), Ones(Index,Index), isOnes(), class Ones
*/
template<typename Derived>
EIGEN_STRONG_INLINE const typename DenseBase<Derived>::ConstantReturnType
EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE const typename DenseBase<Derived>::ConstantReturnType
DenseBase<Derived>::Ones()
{
return Constant(Scalar(1));
......@@ -608,7 +608,7 @@ DenseBase<Derived>::Ones()
* \sa class CwiseNullaryOp, Ones()
*/
template<typename Derived>
bool DenseBase<Derived>::isOnes
EIGEN_DEVICE_FUNC bool DenseBase<Derived>::isOnes
(const RealScalar& prec) const
{
return isApproxToConstant(Scalar(1), prec);
......@@ -622,7 +622,7 @@ bool DenseBase<Derived>::isOnes
* \sa class CwiseNullaryOp, Ones()
*/
template<typename Derived>
EIGEN_STRONG_INLINE Derived& DenseBase<Derived>::setOnes()
EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE Derived& DenseBase<Derived>::setOnes()
{
return setConstant(Scalar(1));
}
......@@ -637,7 +637,7 @@ EIGEN_STRONG_INLINE Derived& DenseBase<Derived>::setOnes()
* \sa MatrixBase::setOnes(), setOnes(Index,Index), class CwiseNullaryOp, MatrixBase::Ones()
*/
template<typename Derived>
EIGEN_STRONG_INLINE Derived&
EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE Derived&
PlainObjectBase<Derived>::setOnes(Index newSize)
{
resize(newSize);
......@@ -646,8 +646,8 @@ PlainObjectBase<Derived>::setOnes(Index newSize)
/** Resizes to the given size, and sets all coefficients in this expression to one.
*
* \param nbRows the new number of rows
* \param nbCols the new number of columns
* \param rows the new number of rows
* \param cols the new number of columns
*
* Example: \include Matrix_setOnes_int_int.cpp
* Output: \verbinclude Matrix_setOnes_int_int.out
......@@ -655,10 +655,10 @@ PlainObjectBase<Derived>::setOnes(Index newSize)
* \sa MatrixBase::setOnes(), setOnes(Index), class CwiseNullaryOp, MatrixBase::Ones()
*/
template<typename Derived>
EIGEN_STRONG_INLINE Derived&
PlainObjectBase<Derived>::setOnes(Index nbRows, Index nbCols)
EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE Derived&
PlainObjectBase<Derived>::setOnes(Index rows, Index cols)
{
resize(nbRows, nbCols);
resize(rows, cols);
return setConstant(Scalar(1));
}
......@@ -666,7 +666,7 @@ PlainObjectBase<Derived>::setOnes(Index nbRows, Index nbCols)
/** \returns an expression of the identity matrix (not necessarily square).
*
* The parameters \a nbRows and \a nbCols are the number of rows and of columns of
* 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.
*
* This variant is meant to be used for dynamic-size matrix types. For fixed-size types,
......@@ -679,10 +679,10 @@ PlainObjectBase<Derived>::setOnes(Index nbRows, Index nbCols)
* \sa Identity(), setIdentity(), isIdentity()
*/
template<typename Derived>
EIGEN_STRONG_INLINE const typename MatrixBase<Derived>::IdentityReturnType
MatrixBase<Derived>::Identity(Index nbRows, Index nbCols)
EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE const typename MatrixBase<Derived>::IdentityReturnType
MatrixBase<Derived>::Identity(Index rows, Index cols)
{
return DenseBase<Derived>::NullaryExpr(nbRows, nbCols, internal::scalar_identity_op<Scalar>());
return DenseBase<Derived>::NullaryExpr(rows, cols, internal::scalar_identity_op<Scalar>());
}
/** \returns an expression of the identity matrix (not necessarily square).
......@@ -696,7 +696,7 @@ MatrixBase<Derived>::Identity(Index nbRows, Index nbCols)
* \sa Identity(Index,Index), setIdentity(), isIdentity()
*/
template<typename Derived>
EIGEN_STRONG_INLINE const typename MatrixBase<Derived>::IdentityReturnType
EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE const typename MatrixBase<Derived>::IdentityReturnType
MatrixBase<Derived>::Identity()
{
EIGEN_STATIC_ASSERT_FIXED_SIZE(Derived)
......@@ -716,18 +716,19 @@ template<typename Derived>
bool MatrixBase<Derived>::isIdentity
(const RealScalar& prec) const
{
typename internal::nested_eval<Derived,1>::type self(derived());
for(Index j = 0; j < cols(); ++j)
{
for(Index i = 0; i < rows(); ++i)
{
if(i == j)
{
if(!internal::isApprox(this->coeff(i, j), static_cast<Scalar>(1), prec))
if(!internal::isApprox(self.coeff(i, j), static_cast<Scalar>(1), prec))
return false;
}
else
{
if(!internal::isMuchSmallerThan(this->coeff(i, j), static_cast<RealScalar>(1), prec))
if(!internal::isMuchSmallerThan(self.coeff(i, j), static_cast<RealScalar>(1), prec))
return false;
}
}
......@@ -740,6 +741,7 @@ namespace internal {
template<typename Derived, bool Big = (Derived::SizeAtCompileTime>=16)>
struct setIdentity_impl
{
EIGEN_DEVICE_FUNC
static EIGEN_STRONG_INLINE Derived& run(Derived& m)
{
return m = Derived::Identity(m.rows(), m.cols());
......@@ -749,11 +751,11 @@ struct setIdentity_impl
template<typename Derived>
struct setIdentity_impl<Derived, true>
{
typedef typename Derived::Index Index;
EIGEN_DEVICE_FUNC
static EIGEN_STRONG_INLINE Derived& run(Derived& m)
{
m.setZero();
const Index size = (std::min)(m.rows(), m.cols());
const Index size = numext::mini(m.rows(), m.cols());
for(Index i = 0; i < size; ++i) m.coeffRef(i,i) = typename Derived::Scalar(1);
return m;
}
......@@ -769,15 +771,15 @@ struct setIdentity_impl<Derived, true>
* \sa class CwiseNullaryOp, Identity(), Identity(Index,Index), isIdentity()
*/
template<typename Derived>
EIGEN_STRONG_INLINE Derived& MatrixBase<Derived>::setIdentity()
EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE Derived& MatrixBase<Derived>::setIdentity()
{
return internal::setIdentity_impl<Derived>::run(derived());
}
/** \brief Resizes to the given size, and writes the identity expression (not necessarily square) into *this.
*
* \param nbRows the new number of rows
* \param nbCols the new number of columns
* \param rows the new number of rows
* \param cols the new number of columns
*
* Example: \include Matrix_setIdentity_int_int.cpp
* Output: \verbinclude Matrix_setIdentity_int_int.out
......@@ -785,9 +787,9 @@ EIGEN_STRONG_INLINE Derived& MatrixBase<Derived>::setIdentity()
* \sa MatrixBase::setIdentity(), class CwiseNullaryOp, MatrixBase::Identity()
*/
template<typename Derived>
EIGEN_STRONG_INLINE Derived& MatrixBase<Derived>::setIdentity(Index nbRows, Index nbCols)
EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE Derived& MatrixBase<Derived>::setIdentity(Index rows, Index cols)
{
derived().resize(nbRows, nbCols);
derived().resize(rows, cols);
return setIdentity();
}
......@@ -798,7 +800,7 @@ EIGEN_STRONG_INLINE Derived& MatrixBase<Derived>::setIdentity(Index nbRows, Inde
* \sa MatrixBase::Unit(Index), MatrixBase::UnitX(), MatrixBase::UnitY(), MatrixBase::UnitZ(), MatrixBase::UnitW()
*/
template<typename Derived>
EIGEN_STRONG_INLINE const typename MatrixBase<Derived>::BasisReturnType MatrixBase<Derived>::Unit(Index newSize, Index i)
EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE const typename MatrixBase<Derived>::BasisReturnType MatrixBase<Derived>::Unit(Index newSize, Index i)
{
EIGEN_STATIC_ASSERT_VECTOR_ONLY(Derived)
return BasisReturnType(SquareMatrixType::Identity(newSize,newSize), i);
......@@ -813,7 +815,7 @@ EIGEN_STRONG_INLINE const typename MatrixBase<Derived>::BasisReturnType MatrixBa
* \sa MatrixBase::Unit(Index,Index), MatrixBase::UnitX(), MatrixBase::UnitY(), MatrixBase::UnitZ(), MatrixBase::UnitW()
*/
template<typename Derived>
EIGEN_STRONG_INLINE const typename MatrixBase<Derived>::BasisReturnType MatrixBase<Derived>::Unit(Index i)
EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE const typename MatrixBase<Derived>::BasisReturnType MatrixBase<Derived>::Unit(Index i)
{
EIGEN_STATIC_ASSERT_VECTOR_ONLY(Derived)
return BasisReturnType(SquareMatrixType::Identity(),i);
......@@ -826,7 +828,7 @@ EIGEN_STRONG_INLINE const typename MatrixBase<Derived>::BasisReturnType MatrixBa
* \sa MatrixBase::Unit(Index,Index), MatrixBase::Unit(Index), MatrixBase::UnitY(), MatrixBase::UnitZ(), MatrixBase::UnitW()
*/
template<typename Derived>
EIGEN_STRONG_INLINE const typename MatrixBase<Derived>::BasisReturnType MatrixBase<Derived>::UnitX()
EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE const typename MatrixBase<Derived>::BasisReturnType MatrixBase<Derived>::UnitX()
{ return Derived::Unit(0); }
/** \returns an expression of the Y axis unit vector (0,1{,0}^*)
......@@ -836,7 +838,7 @@ EIGEN_STRONG_INLINE const typename MatrixBase<Derived>::BasisReturnType MatrixBa
* \sa MatrixBase::Unit(Index,Index), MatrixBase::Unit(Index), MatrixBase::UnitY(), MatrixBase::UnitZ(), MatrixBase::UnitW()
*/
template<typename Derived>
EIGEN_STRONG_INLINE const typename MatrixBase<Derived>::BasisReturnType MatrixBase<Derived>::UnitY()
EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE const typename MatrixBase<Derived>::BasisReturnType MatrixBase<Derived>::UnitY()
{ return Derived::Unit(1); }
/** \returns an expression of the Z axis unit vector (0,0,1{,0}^*)
......@@ -846,7 +848,7 @@ EIGEN_STRONG_INLINE const typename MatrixBase<Derived>::BasisReturnType MatrixBa
* \sa MatrixBase::Unit(Index,Index), MatrixBase::Unit(Index), MatrixBase::UnitY(), MatrixBase::UnitZ(), MatrixBase::UnitW()
*/
template<typename Derived>
EIGEN_STRONG_INLINE const typename MatrixBase<Derived>::BasisReturnType MatrixBase<Derived>::UnitZ()
EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE const typename MatrixBase<Derived>::BasisReturnType MatrixBase<Derived>::UnitZ()
{ return Derived::Unit(2); }
/** \returns an expression of the W axis unit vector (0,0,0,1)
......@@ -856,7 +858,7 @@ EIGEN_STRONG_INLINE const typename MatrixBase<Derived>::BasisReturnType MatrixBa
* \sa MatrixBase::Unit(Index,Index), MatrixBase::Unit(Index), MatrixBase::UnitY(), MatrixBase::UnitZ(), MatrixBase::UnitW()
*/
template<typename Derived>
EIGEN_STRONG_INLINE const typename MatrixBase<Derived>::BasisReturnType MatrixBase<Derived>::UnitW()
EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE const typename MatrixBase<Derived>::BasisReturnType MatrixBase<Derived>::UnitW()
{ return Derived::Unit(3); }
} // end namespace Eigen
......