GitHub-Proxy/eigen
1
0
Fork 0
mirror of https://gitlab.com/libeigen/eigen.git synced 2025-09-14 18:33:16 +08:00
Code Issues Packages Projects Releases Wiki Activity

Merged eigen/eigen into default

Browse Source
This commit is contained in:
Konstantinos Margaritis 2014-09-21 14:02:51 +03:00
commit 60e093a9dc
211 changed files with 8277 additions and 9581 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

2
CMakeLists.txt
View File

@ -143,7 +143,7 @@ if(NOT MSVC)
ei_add_cxx_compiler_flag("-Wpointer-arith")
ei_add_cxx_compiler_flag("-Wwrite-strings")
ei_add_cxx_compiler_flag("-Wformat-security")
ei_add_cxx_compiler_flag("-Wshorten-64-to-32")
# ei_add_cxx_compiler_flag("-Wshorten-64-to-32")
ei_add_cxx_compiler_flag("-Wenum-conversion")
ei_add_cxx_compiler_flag("-Wc++11-extensions")

1
Eigen/Cholesky
View File

@ -21,7 +21,6 @@
* \endcode
*/
#include "src/misc/Solve.h"
#include "src/Cholesky/LLT.h"
#include "src/Cholesky/LDLT.h"
#ifdef EIGEN_USE_LAPACKE

4
Eigen/CholmodSupport
View File

@ -33,12 +33,8 @@ extern "C" {
*
*/
#include "src/misc/Solve.h"
#include "src/misc/SparseSolve.h"
#include "src/CholmodSupport/CholmodSupport.h"
#include "src/Core/util/ReenableStupidWarnings.h"
#endif // EIGEN_CHOLMODSUPPORT_MODULE_H

21
Eigen/Core
View File

@ -277,8 +277,8 @@ using std::ptrdiff_t;
*/
#include "src/Core/util/Constants.h"
#include "src/Core/util/ForwardDeclarations.h"
#include "src/Core/util/Meta.h"
#include "src/Core/util/ForwardDeclarations.h"
#include "src/Core/util/StaticAssert.h"
#include "src/Core/util/XprHelper.h"
#include "src/Core/util/Memory.h"
@ -311,19 +311,16 @@ using std::ptrdiff_t;
#include "src/Core/functors/UnaryFunctors.h"
#include "src/Core/functors/NullaryFunctors.h"
#include "src/Core/functors/StlFunctors.h"
#include "src/Core/functors/AssignmentFunctors.h"
#include "src/Core/DenseCoeffsBase.h"
#include "src/Core/DenseBase.h"
#include "src/Core/MatrixBase.h"
#include "src/Core/EigenBase.h"
#ifdef EIGEN_ENABLE_EVALUATORS
#include "src/Core/functors/AssignmentFunctors.h"
#include "src/Core/Product.h"
#include "src/Core/CoreEvaluators.h"
#include "src/Core/AssignEvaluator.h"
#include "src/Core/ProductEvaluators.h"
#endif
#ifndef EIGEN_PARSED_BY_DOXYGEN // work around Doxygen bug triggered by Assign.h r814874
// at least confirmed with Doxygen 1.5.5 and 1.5.6
@ -334,7 +331,10 @@ using std::ptrdiff_t;
#include "src/Core/util/BlasUtil.h"
#include "src/Core/DenseStorage.h"
#include "src/Core/NestByValue.h"
#include "src/Core/ForceAlignedAccess.h"
// #include "src/Core/ForceAlignedAccess.h"
// #include "src/Core/Flagged.h"
#include "src/Core/ReturnByValue.h"
#include "src/Core/NoAlias.h"
#include "src/Core/PlainObjectBase.h"
@ -347,12 +347,12 @@ using std::ptrdiff_t;
#include "src/Core/SelfCwiseBinaryOp.h"
#include "src/Core/Dot.h"
#include "src/Core/StableNorm.h"
#include "src/Core/MapBase.h"
#include "src/Core/Stride.h"
#include "src/Core/MapBase.h"
#include "src/Core/Map.h"
#include "src/Core/Ref.h"
#include "src/Core/Block.h"
#include "src/Core/VectorBlock.h"
#include "src/Core/Ref.h"
#include "src/Core/Transpose.h"
#include "src/Core/DiagonalMatrix.h"
#include "src/Core/Diagonal.h"
@ -365,14 +365,15 @@ using std::ptrdiff_t;
#include "src/Core/IO.h"
#include "src/Core/Swap.h"
#include "src/Core/CommaInitializer.h"
#include "src/Core/Flagged.h"
#include "src/Core/ProductBase.h"
#include "src/Core/GeneralProduct.h"
#include "src/Core/Solve.h"
#include "src/Core/Inverse.h"
#include "src/Core/TriangularMatrix.h"
#include "src/Core/SelfAdjointView.h"
#include "src/Core/products/GeneralBlockPanelKernel.h"
#include "src/Core/products/Parallelizer.h"
#include "src/Core/products/CoeffBasedProduct.h"
#include "src/Core/ProductEvaluators.h"
#include "src/Core/products/GeneralMatrixVector.h"
#include "src/Core/products/GeneralMatrixMatrix.h"
#include "src/Core/SolveTriangular.h"

4
Eigen/IterativeLinearSolvers
View File

@ -26,9 +26,7 @@
* \endcode
*/
#include "src/misc/Solve.h"
#include "src/misc/SparseSolve.h"
#include "src/IterativeLinearSolvers/SolveWithGuess.h"
#include "src/IterativeLinearSolvers/IterativeSolverBase.h"
#include "src/IterativeLinearSolvers/BasicPreconditioners.h"
#include "src/IterativeLinearSolvers/ConjugateGradient.h"

3
Eigen/LU
View File

@ -16,7 +16,6 @@
* \endcode
*/
#include "src/misc/Solve.h"
#include "src/misc/Kernel.h"
#include "src/misc/Image.h"
#include "src/LU/FullPivLU.h"
@ -25,7 +24,7 @@
#include "src/LU/PartialPivLU_MKL.h"
#endif
#include "src/LU/Determinant.h"
#include "src/LU/Inverse.h"
#include "src/LU/InverseImpl.h"
// Use the SSE optimized version whenever possible. At the moment the
// SSE version doesn't compile when AVX is enabled

4
Eigen/PaStiXSupport
View File

@ -35,12 +35,8 @@ extern "C" {
*
*/
#include "src/misc/Solve.h"
#include "src/misc/SparseSolve.h"
#include "src/PaStiXSupport/PaStiXSupport.h"
#include "src/Core/util/ReenableStupidWarnings.h"
#endif // EIGEN_PASTIXSUPPORT_MODULE_H

1
Eigen/QR
View File

@ -24,7 +24,6 @@
* \endcode
*/
#include "src/misc/Solve.h"
#include "src/QR/HouseholderQR.h"
#include "src/QR/FullPivHouseholderQR.h"
#include "src/QR/ColPivHouseholderQR.h"

2
Eigen/SPQRSupport
View File

@ -21,8 +21,6 @@
*
*/
#include "src/misc/Solve.h"
#include "src/misc/SparseSolve.h"
#include "src/CholmodSupport/CholmodSupport.h"
#include "src/SPQRSupport/SuiteSparseQRSupport.h"

2
Eigen/SVD
View File

@ -20,7 +20,7 @@
* \endcode
*/
#include "src/misc/Solve.h"
#include "src/SVD/SVDBase.h"
#include "src/SVD/JacobiSVD.h"
#if defined(EIGEN_USE_LAPACKE) && !defined(EIGEN_USE_LAPACKE_STRICT)
#include "src/SVD/JacobiSVD_MKL.h"

2
Eigen/SparseCholesky
View File

@ -34,8 +34,6 @@
#error The SparseCholesky module has nothing to offer in MPL2 only mode
#endif
#include "src/misc/Solve.h"
#include "src/misc/SparseSolve.h"
#include "src/SparseCholesky/SimplicialCholesky.h"
#ifndef EIGEN_MPL2_ONLY

16
Eigen/SparseCore
View File

@ -35,28 +35,30 @@ struct Sparse {};
#include "src/SparseCore/SparseUtil.h"
#include "src/SparseCore/SparseMatrixBase.h"
#include "src/SparseCore/SparseAssign.h"
#include "src/SparseCore/CompressedStorage.h"
#include "src/SparseCore/AmbiVector.h"
#include "src/SparseCore/SparseMatrix.h"
#include "src/SparseCore/MappedSparseMatrix.h"
#include "src/SparseCore/SparseVector.h"
#include "src/SparseCore/SparseBlock.h"
#include "src/SparseCore/SparseTranspose.h"
#include "src/SparseCore/SparseCwiseUnaryOp.h"
#include "src/SparseCore/SparseCwiseBinaryOp.h"
#include "src/SparseCore/SparseTranspose.h"
#include "src/SparseCore/SparseBlock.h"
#include "src/SparseCore/SparseDot.h"
#include "src/SparseCore/SparsePermutation.h"
#include "src/SparseCore/SparseRedux.h"
#include "src/SparseCore/SparseFuzzy.h"
#include "src/SparseCore/SparseView.h"
#include "src/SparseCore/SparseDiagonalProduct.h"
#include "src/SparseCore/ConservativeSparseSparseProduct.h"
#include "src/SparseCore/SparseSparseProductWithPruning.h"
#include "src/SparseCore/SparseProduct.h"
#include "src/SparseCore/SparseDenseProduct.h"
#include "src/SparseCore/SparseDiagonalProduct.h"
#include "src/SparseCore/SparseTriangularView.h"
#include "src/SparseCore/SparseSelfAdjointView.h"
#include "src/SparseCore/SparseTriangularView.h"
#include "src/SparseCore/TriangularSolver.h"
#include "src/SparseCore/SparseView.h"
#include "src/SparseCore/SparsePermutation.h"
#include "src/SparseCore/SparseFuzzy.h"
#include "src/SparseCore/SparseSolverBase.h"
#include "src/Core/util/ReenableStupidWarnings.h"

3
Eigen/SparseLU
View File

@ -20,9 +20,6 @@
* Please, see the documentation of the SparseLU class for more details.
*/
#include "src/misc/Solve.h"
#include "src/misc/SparseSolve.h"
// Ordering interface
#include "OrderingMethods"

3
Eigen/SparseQR
View File

@ -21,9 +21,6 @@
*
*/
#include "src/misc/Solve.h"
#include "src/misc/SparseSolve.h"
#include "OrderingMethods"
#include "src/SparseCore/SparseColEtree.h"
#include "src/SparseQR/SparseQR.h"

4
Eigen/SuperLUSupport
View File

@ -48,12 +48,8 @@ namespace Eigen { struct SluMatrix; }
*
*/
#include "src/misc/Solve.h"
#include "src/misc/SparseSolve.h"
#include "src/SuperLUSupport/SuperLUSupport.h"
#include "src/Core/util/ReenableStupidWarnings.h"
#endif // EIGEN_SUPERLUSUPPORT_MODULE_H

3
Eigen/UmfPackSupport
View File

@ -26,9 +26,6 @@ extern "C" {
*
*/
#include "src/misc/Solve.h"
#include "src/misc/SparseSolve.h"
#include "src/UmfPackSupport/UmfPackSupport.h"
#include "src/Core/util/ReenableStupidWarnings.h"

86
Eigen/src/Cholesky/LDLT.h
View File

@ -175,13 +175,13 @@ template<typename _MatrixType, int _UpLo> class LDLT
* \sa MatrixBase::ldlt(), SelfAdjointView::ldlt()
*/
template<typename Rhs>
inline const internal::solve_retval<LDLT, Rhs>
inline const Solve<LDLT, Rhs>
solve(const MatrixBase<Rhs>& b) const
{
eigen_assert(m_isInitialized && "LDLT is not initialized.");
eigen_assert(m_matrix.rows()==b.rows()
&& "LDLT::solve(): invalid number of rows of the right hand side matrix b");
return internal::solve_retval<LDLT, Rhs>(*this, b.derived());
return Solve<LDLT, Rhs>(*this, b.derived());
}
template<typename Derived>
@ -217,6 +217,12 @@ template<typename _MatrixType, int _UpLo> class LDLT
eigen_assert(m_isInitialized && "LDLT is not initialized.");
return Success;
}
#ifndef EIGEN_PARSED_BY_DOXYGEN
template<typename RhsType, typename DstType>
EIGEN_DEVICE_FUNC
void _solve_impl(const RhsType &rhs, DstType &dst) const;
#endif
protected:
@ -466,52 +472,46 @@ LDLT<MatrixType,_UpLo>& LDLT<MatrixType,_UpLo>::rankUpdate(const MatrixBase<Deri
return *this;
}
namespace internal {
template<typename _MatrixType, int _UpLo, typename Rhs>
struct solve_retval<LDLT<_MatrixType,_UpLo>, Rhs>
: solve_retval_base<LDLT<_MatrixType,_UpLo>, Rhs>
#ifndef EIGEN_PARSED_BY_DOXYGEN
template<typename _MatrixType, int _UpLo>
template<typename RhsType, typename DstType>
void LDLT<_MatrixType,_UpLo>::_solve_impl(const RhsType &rhs, DstType &dst) const
{
typedef LDLT<_MatrixType,_UpLo> LDLTType;
EIGEN_MAKE_SOLVE_HELPERS(LDLTType,Rhs)
eigen_assert(rhs.rows() == rows());
// dst = P b
dst = m_transpositions * rhs;
template<typename Dest> void evalTo(Dest& dst) const
// dst = L^-1 (P b)
matrixL().solveInPlace(dst);
// dst = D^-1 (L^-1 P b)
// more precisely, use pseudo-inverse of D (see bug 241)
using std::abs;
EIGEN_USING_STD_MATH(max);
const typename Diagonal<const MatrixType>::RealReturnType vecD(vectorD());
// In some previous versions, tolerance was set to the max of 1/highest and the maximal diagonal entry * epsilon
// as motivated by LAPACK's xGELSS:
// RealScalar tolerance = (max)(vectorD.array().abs().maxCoeff() *NumTraits<RealScalar>::epsilon(),RealScalar(1) / NumTraits<RealScalar>::highest());
// However, LDLT is not rank revealing, and so adjusting the tolerance wrt to the highest
// diagonal element is not well justified and to numerical issues in some cases.
// Moreover, Lapack's xSYTRS routines use 0 for the tolerance.
RealScalar tolerance = RealScalar(1) / NumTraits<RealScalar>::highest();
for (Index i = 0; i < vecD.size(); ++i)
{
eigen_assert(rhs().rows() == dec().matrixLDLT().rows());
// dst = P b
dst = dec().transpositionsP() * rhs();
// dst = L^-1 (P b)
dec().matrixL().solveInPlace(dst);
// dst = D^-1 (L^-1 P b)
// more precisely, use pseudo-inverse of D (see bug 241)
using std::abs;
EIGEN_USING_STD_MATH(max);
typedef typename LDLTType::MatrixType MatrixType;
typedef typename LDLTType::RealScalar RealScalar;
const typename Diagonal<const MatrixType>::RealReturnType vectorD(dec().vectorD());
// In some previous versions, tolerance was set to the max of 1/highest and the maximal diagonal entry * epsilon
// as motivated by LAPACK's xGELSS:
// RealScalar tolerance = (max)(vectorD.array().abs().maxCoeff() *NumTraits<RealScalar>::epsilon(),RealScalar(1) / NumTraits<RealScalar>::highest());
// However, LDLT is not rank revealing, and so adjusting the tolerance wrt to the highest
// diagonal element is not well justified and to numerical issues in some cases.
// Moreover, Lapack's xSYTRS routines use 0 for the tolerance.
RealScalar tolerance = RealScalar(1) / NumTraits<RealScalar>::highest();
for (Index i = 0; i < vectorD.size(); ++i) {
if(abs(vectorD(i)) > tolerance)
dst.row(i) /= vectorD(i);
else
dst.row(i).setZero();
}
// dst = L^-T (D^-1 L^-1 P b)
dec().matrixU().solveInPlace(dst);
// dst = P^-1 (L^-T D^-1 L^-1 P b) = A^-1 b
dst = dec().transpositionsP().transpose() * dst;
if(abs(vecD(i)) > tolerance)
dst.row(i) /= vecD(i);
else
dst.row(i).setZero();
}
};
// dst = L^-T (D^-1 L^-1 P b)
matrixU().solveInPlace(dst);
// dst = P^-1 (L^-T D^-1 L^-1 P b) = A^-1 b
dst = m_transpositions.transpose() * dst;
}
#endif
/** \internal use x = ldlt_object.solve(x);
*

32
Eigen/src/Cholesky/LLT.h
View File

@ -118,13 +118,13 @@ template<typename _MatrixType, int _UpLo> class 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());
}
template<typename Derived>
@ -161,6 +161,12 @@ template<typename _MatrixType, int _UpLo> class LLT
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:
/** \internal
@ -404,22 +410,16 @@ 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>
#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
{
typedef LLT<_MatrixType,UpLo> LLTType;
EIGEN_MAKE_SOLVE_HELPERS(LLTType,Rhs)
template<typename Dest> void evalTo(Dest& dst) const
{
dst = rhs();
dec().solveInPlace(dst);
}
};
dst = rhs;
solveInPlace(dst);
}
#endif
/** \internal use x = llt_object.solve(x);
*

76
Eigen/src/CholmodSupport/CholmodSupport.h
View File

@ -157,8 +157,12 @@ 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 };
@ -170,14 +174,14 @@ class CholmodBase : internal::noncopyable
public:
CholmodBase()
: m_cholmodFactor(0), m_info(Success), m_isInitialized(false)
: m_cholmodFactor(0), m_info(Success)
{
m_shiftOffset[0] = m_shiftOffset[1] = RealScalar(0.0);
cholmod_start(&m_cholmod);
}
CholmodBase(const MatrixType& matrix)
: m_cholmodFactor(0), m_info(Success), m_isInitialized(false)
: m_cholmodFactor(0), m_info(Success)
{
m_shiftOffset[0] = m_shiftOffset[1] = RealScalar(0.0);
cholmod_start(&m_cholmod);
@ -194,9 +198,6 @@ class CholmodBase : internal::noncopyable
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); }
/** \brief Reports whether previous computation was successful.
*
* \returns \c Success if computation was succesful,
@ -216,34 +217,6 @@ 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.
@ -290,7 +263,7 @@ 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;
@ -312,7 +285,7 @@ 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
void _solve_impl(const SparseMatrix<RhsScalar,RhsOptions,RhsIndex> &b, SparseMatrix<DestScalar,DestOptions,DestIndex> &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;
@ -357,7 +330,6 @@ class CholmodBase : internal::noncopyable
cholmod_factor* m_cholmodFactor;
RealScalar m_shiftOffset[2];
mutable ComputationInfo m_info;
bool m_isInitialized;
int m_factorizationIsOk;
int m_analysisIsOk;
};
@ -572,36 +544,6 @@ 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

18
Eigen/src/Core/Array.h
View File

@ -74,6 +74,21 @@ class Array
{
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.
*
@ -99,7 +114,7 @@ class Array
{
return Base::_set(other);
}
/** Default constructor.
*
* For fixed-size matrices, does nothing.
@ -144,7 +159,6 @@ class Array
}
#endif
#ifndef EIGEN_PARSED_BY_DOXYGEN
template<typename T>
EIGEN_DEVICE_FUNC

23
Eigen/src/Core/ArrayBase.h
View File

@ -64,8 +64,7 @@ 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;
@ -121,8 +120,14 @@ template<typename Derived> class ArrayBase
EIGEN_DEVICE_FUNC
Derived& operator=(const ArrayBase& other)
{
return internal::assign_selector<Derived,Derived>::run(derived(), other.derived());
internal::call_assignment(derived(), other.derived());
}
/** Set all the entries to \a value.
* \sa DenseBase::setConstant(), DenseBase::fill() */
EIGEN_DEVICE_FUNC
Derived& operator=(const Scalar &value)
{ Base::setConstant(value); return derived(); }
EIGEN_DEVICE_FUNC
Derived& operator+=(const Scalar& scalar);
@ -186,8 +191,7 @@ template<typename OtherDerived>
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>());
return derived();
}
@ -200,8 +204,7 @@ template<typename OtherDerived>
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>());
return derived();
}
@ -214,8 +217,7 @@ template<typename OtherDerived>
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();
}
@ -228,8 +230,7 @@ template<typename OtherDerived>
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>());
return derived();
}

12
Eigen/src/Core/ArrayWrapper.h
View File

@ -29,6 +29,11 @@ struct traits<ArrayWrapper<ExpressionType> >
: public traits<typename remove_all<typename ExpressionType::Nested>::type >
{
typedef ArrayXpr XprKind;
// Let's remove NestByRefBit
enum {
Flags0 = traits<typename remove_all<typename ExpressionType::Nested>::type >::Flags,
Flags = Flags0 & ~NestByRefBit
};
};
}
@ -39,6 +44,7 @@ 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,
@ -166,6 +172,11 @@ struct traits<MatrixWrapper<ExpressionType> >
: public traits<typename remove_all<typename ExpressionType::Nested>::type >
{
typedef MatrixXpr XprKind;
// Let's remove NestByRefBit
enum {
Flags0 = traits<typename remove_all<typename ExpressionType::Nested>::type >::Flags,
Flags = Flags0 & ~NestByRefBit
};
};
}
@ -176,6 +187,7 @@ 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,

558
Eigen/src/Core/Assign.h
View File

@ -14,485 +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(Derived::SizeAtCompileTime)
EIGEN_DEBUG_VAR(OtherDerived::CoeffReadCost)
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
};
EIGEN_DEVICE_FUNC
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>
{
EIGEN_DEVICE_FUNC
static EIGEN_STRONG_INLINE void run(Derived1 &, const Derived2 &) {}
};
template<typename Derived1, typename Derived2, int Index, int Stop>
struct assign_DefaultTraversal_InnerUnrolling
{
EIGEN_DEVICE_FUNC
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>
{
EIGEN_DEVICE_FUNC
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
{
EIGEN_DEVICE_FUNC
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>
{
EIGEN_DEVICE_FUNC
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>
{
EIGEN_DEVICE_FUNC
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;
EIGEN_DEVICE_FUNC
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>
{
EIGEN_DEVICE_FUNC
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;
EIGEN_DEVICE_FUNC
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;
EIGEN_DEVICE_FUNC
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>
{
EIGEN_DEVICE_FUNC
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 packet_traits<typename Derived1::Scalar> PacketTraits;
enum {
packetSize = PacketTraits::size,
alignable = PacketTraits::AlignedOnScalar,
dstAlignment = alignable ? Aligned : int(assign_traits<Derived1,Derived2>::DstIsAligned) ,
srcAlignment = assign_traits<Derived1,Derived2>::JointAlignment
};
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) || assign_traits<Derived1,Derived2>::DstIsAligned) ? 0
: internal::first_aligned(&dst.coeffRef(0,0), 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>
@ -506,91 +27,35 @@ 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_TEST_EVALUATORS
#ifdef EIGEN_DEBUG_ASSIGN
internal::copy_using_evaluator_traits<Derived, OtherDerived>::debug();
#endif
eigen_assert(rows() == other.rows() && cols() == other.cols());
internal::call_dense_assignment_loop(derived(),other.derived());
internal::call_assignment_no_alias(derived(),other.derived());
#else // EIGEN_TEST_EVALUATORS
#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());
#endif // EIGEN_TEST_EVALUATORS
#ifndef EIGEN_NO_DEBUG
checkTransposeAliasing(other.derived());
#endif
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> {
EIGEN_DEVICE_FUNC
static EIGEN_STRONG_INLINE Derived& run(Derived& dst, const OtherDerived& other) { return dst.lazyAssign(other.derived()); }
template<typename ActualDerived, typename ActualOtherDerived>
EIGEN_DEVICE_FUNC
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> {
EIGEN_DEVICE_FUNC
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> {
EIGEN_DEVICE_FUNC
static EIGEN_STRONG_INLINE Derived& run(Derived& dst, const OtherDerived& other) { return dst.lazyAssign(other.transpose()); }
template<typename ActualDerived, typename ActualOtherDerived>
EIGEN_DEVICE_FUNC
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> {
EIGEN_DEVICE_FUNC
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>
@ -598,7 +63,8 @@ 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>
@ -606,7 +72,8 @@ 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>
@ -614,7 +81,8 @@ 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

416
Eigen/src/Core/AssignEvaluator.h
View File

@ -2,7 +2,7 @@
// for linear algebra.
//
// Copyright (C) 2011 Benoit Jacob <jacob.benoit.1@gmail.com>
// Copyright (C) 2011-2013 Gael Guennebaud <gael.guennebaud@inria.fr>
// 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
@ -24,37 +24,46 @@ namespace internal {
// copy_using_evaluator_traits is based on assign_traits
template <typename Derived, typename OtherDerived>
template <typename DstEvaluator, typename SrcEvaluator, typename AssignFunc>
struct copy_using_evaluator_traits
{
typedef typename DstEvaluator::XprType Dst;
enum {
DstFlags = DstEvaluator::Flags,
SrcFlags = SrcEvaluator::Flags
};
public:
enum {
DstIsAligned = Derived::Flags & AlignedBit,
DstHasDirectAccess = Derived::Flags & DirectAccessBit,
SrcIsAligned = OtherDerived::Flags & AlignedBit,
JointAlignment = bool(DstIsAligned) && bool(SrcIsAligned) ? Aligned : Unaligned,
SrcEvalBeforeAssign = (evaluator_traits<OtherDerived>::HasEvalTo == 1)
DstIsAligned = DstFlags & AlignedBit,
DstHasDirectAccess = DstFlags & DirectAccessBit,
SrcIsAligned = SrcFlags & 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
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),
MaxSizeAtCompileTime = Dst::SizeAtCompileTime,
PacketSize = packet_traits<typename Dst::Scalar>::size
};
enum {
StorageOrdersAgree = (int(Derived::IsRowMajor) == int(OtherDerived::IsRowMajor)),
DstIsRowMajor = DstFlags&RowMajorBit,
SrcIsRowMajor = SrcFlags&RowMajorBit,
StorageOrdersAgree = (int(DstIsRowMajor) == int(SrcIsRowMajor)),
MightVectorize = StorageOrdersAgree
&& (int(Derived::Flags) & int(OtherDerived::Flags) & ActualPacketAccessBit),
&& (int(DstFlags) & int(SrcFlags) & ActualPacketAccessBit)
&& (functor_traits<AssignFunc>::PacketAccess),
MayInnerVectorize = MightVectorize && int(InnerSize)!=Dynamic && int(InnerSize)%int(PacketSize)==0
&& int(DstIsAligned) && int(SrcIsAligned),
MayLinearize = StorageOrdersAgree && (int(Derived::Flags) & int(OtherDerived::Flags) & LinearAccessBit),
MayLinearize = StorageOrdersAgree && (int(DstFlags) & int(SrcFlags) & 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,
@ -68,8 +77,7 @@ private:
public:
enum {
Traversal = int(SrcEvalBeforeAssign) ? int(AllAtOnceTraversal)
: int(MayInnerVectorize) ? int(InnerVectorizedTraversal)
Traversal = int(MayInnerVectorize) ? int(InnerVectorizedTraversal)
: int(MayLinearVectorize) ? int(LinearVectorizedTraversal)
: int(MaySliceVectorize) ? int(SliceVectorizedTraversal)
: int(MayLinearize) ? int(LinearTraversal)
@ -82,12 +90,12 @@ public:
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),
MayUnrollCompletely = int(Dst::SizeAtCompileTime) != Dynamic
&& int(SrcEvaluator::CoeffReadCost) != Dynamic
&& int(Dst::SizeAtCompileTime) * int(SrcEvaluator::CoeffReadCost) <= int(UnrollingLimit),
MayUnrollInner = int(InnerSize) != Dynamic
&& int(OtherDerived::CoeffReadCost) != Dynamic
&& int(InnerSize) * int(OtherDerived::CoeffReadCost) <= int(UnrollingLimit)
&& int(SrcEvaluator::CoeffReadCost) != Dynamic
&& int(InnerSize) * int(SrcEvaluator::CoeffReadCost) <= int(UnrollingLimit)
};
public:
@ -110,6 +118,12 @@ public:
#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);
EIGEN_DEBUG_VAR(DstFlags)
EIGEN_DEBUG_VAR(SrcFlags)
std::cerr.unsetf(std::ios::hex);
EIGEN_DEBUG_VAR(DstIsAligned)
EIGEN_DEBUG_VAR(SrcIsAligned)
EIGEN_DEBUG_VAR(JointAlignment)
@ -127,6 +141,7 @@ public:
EIGEN_DEBUG_VAR(MayUnrollCompletely)
EIGEN_DEBUG_VAR(MayUnrollInner)
EIGEN_DEBUG_VAR(Unrolling)
std::cerr << std::endl;
}
#endif
};
@ -142,6 +157,7 @@ public:
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;
@ -206,9 +222,10 @@ struct copy_using_evaluator_LinearTraversal_CompleteUnrolling<Kernel, Stop, Stop
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;
enum {
outer = Index / DstXprType::InnerSizeAtCompileTime,
inner = Index % DstXprType::InnerSizeAtCompileTime,
@ -235,8 +252,7 @@ struct copy_using_evaluator_innervec_InnerUnrolling
static EIGEN_STRONG_INLINE void run(Kernel &kernel, int outer)
{
kernel.template assignPacketByOuterInner<Aligned, Aligned>(outer, Index);
typedef typename Kernel::DstEvaluatorType::XprType DstXprType;
enum { NextIndex = Index + packet_traits<typename DstXprType::Scalar>::size };
enum { NextIndex = Index + packet_traits<typename Kernel::Scalar>::size };
copy_using_evaluator_innervec_InnerUnrolling<Kernel, NextIndex, Stop>::run(kernel, outer);
}
};
@ -496,25 +512,8 @@ struct dense_assignment_loop<Kernel, SliceVectorizedTraversal, NoUnrolling>
}
};
/****************************
*** All-at-once traversal ***
****************************/
// TODO: this 'AllAtOnceTraversal' should be dropped or caught earlier (Gael)
// Indeed, what to do with the kernel's functor??
template<typename Kernel>
struct dense_assignment_loop<Kernel, AllAtOnceTraversal, NoUnrolling>
{
static inline void run(Kernel & kernel)
{
// Evaluate rhs in temporary to prevent aliasing problems in a = a * a;
// TODO: Do not pass the xpr object to evalTo() (Jitse)
kernel.srcEvaluator().evalTo(kernel.dstEvaluator(), kernel.dstExpression());
}
};
/***************************************************************************
* Part 4 : Generic Assignment routine
* Part 4 : Generic dense assignment kernel
***************************************************************************/
// This class generalize the assignment of a coefficient (or packet) from one dense evaluator
@ -523,7 +522,7 @@ struct dense_assignment_loop<Kernel, AllAtOnceTraversal, NoUnrolling>
// 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>
template<typename DstEvaluatorTypeT, typename SrcEvaluatorTypeT, typename Functor, int Version = Specialized>
class generic_dense_assignment_kernel
{
protected:
@ -535,16 +534,22 @@ public:
typedef SrcEvaluatorTypeT SrcEvaluatorType;
typedef typename DstEvaluatorType::Scalar Scalar;
typedef typename DstEvaluatorType::Index Index;
typedef copy_using_evaluator_traits<DstXprType, SrcXprType> AssignmentTraits;
typedef copy_using_evaluator_traits<DstEvaluatorTypeT, SrcEvaluatorTypeT, Functor> AssignmentTraits;
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
}
Index size() const { return m_dstExpr.size(); }
Index innerSize() const { return m_dstExpr.innerSize(); }
Index outerSize() const { return m_dstExpr.outerSize(); }
Index rows() const { return m_dstExpr.rows(); }
Index cols() const { return m_dstExpr.cols(); }
Index outerStride() const { return m_dstExpr.outerStride(); }
// TODO get rid of this one:
@ -553,16 +558,19 @@ public:
DstEvaluatorType& dstEvaluator() { return m_dst; }
const SrcEvaluatorType& srcEvaluator() const { return m_src; }
/// Assign src(row,col) to dst(row,col) through the assignment functor.
void assignCoeff(Index row, Index col)
{
m_functor.assignCoeff(m_dst.coeffRef(row,col), m_src.coeff(row,col));
}
/// \sa assignCoeff(Index,Index)
void assignCoeff(Index index)
{
m_functor.assignCoeff(m_dst.coeffRef(index), m_src.coeff(index));
}
/// \sa assignCoeff(Index,Index)
void assignCoeffByOuterInner(Index outer, Index inner)
{
Index row = rowIndexByOuterInner(outer, inner);
@ -596,7 +604,7 @@ public:
typedef typename DstEvaluatorType::ExpressionTraits Traits;
return int(Traits::RowsAtCompileTime) == 1 ? 0
: int(Traits::ColsAtCompileTime) == 1 ? inner
: int(Traits::Flags)&RowMajorBit ? outer
: int(DstEvaluatorType::Flags)&RowMajorBit ? outer
: inner;
}
@ -605,7 +613,7 @@ public:
typedef typename DstEvaluatorType::ExpressionTraits Traits;
return int(Traits::ColsAtCompileTime) == 1 ? 0
: int(Traits::RowsAtCompileTime) == 1 ? inner
: int(Traits::Flags)&RowMajorBit ? inner
: int(DstEvaluatorType::Flags)&RowMajorBit ? inner
: outer;
}
@ -617,13 +625,13 @@ protected:
DstXprType& m_dstExpr;
};
/***************************************************************************
* Part 5 : Entry point for dense rectangular assignment
***************************************************************************/
template<typename DstXprType, typename SrcXprType, typename Functor>
void call_dense_assignment_loop(const DstXprType& dst, const SrcXprType& src, const Functor &func)
{
#ifdef EIGEN_DEBUG_ASSIGN
// TODO these traits should be computed from information provided by the evaluators
internal::copy_using_evaluator_traits<DstXprType, SrcXprType>::debug();
#endif
eigen_assert(dst.rows() == src.rows() && dst.cols() == src.cols());
typedef typename evaluator<DstXprType>::type DstEvaluatorType;
@ -645,195 +653,141 @@ void call_dense_assignment_loop(const DstXprType& dst, const SrcXprType& src)
}
/***************************************************************************
* Part 5 : Entry points
* Part 6 : Generic assignment
***************************************************************************/
// Based on DenseBase::LazyAssign()
// The following functions are just for testing and they are meant to be moved to operator= and the likes.
// 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;
template<typename DstXprType, template <typename> class StorageBase, typename SrcXprType>
EIGEN_STRONG_INLINE
const DstXprType& copy_using_evaluator(const NoAlias<DstXprType, StorageBase>& dst,
const EigenBase<SrcXprType>& src)
{
return noalias_copy_using_evaluator(dst.expression(), src.derived(), internal::assign_op<typename DstXprType::Scalar>());
}
// Assignement kind defined in this file:
struct Dense2Dense {};
struct EigenBase2EigenBase {};
template<typename XprType, int AssumeAliasing = evaluator_traits<XprType>::AssumeAliasing>
struct AddEvalIfAssumingAliasing;
template<typename XprType>
struct AddEvalIfAssumingAliasing<XprType, 0>
{
static const XprType& run(const XprType& xpr)
{
return xpr;
}
};
template<typename XprType>
struct AddEvalIfAssumingAliasing<XprType, 1>
{
static const EvalToTemp<XprType> run(const XprType& xpr)
{
return EvalToTemp<XprType>(xpr);
}
};
template<typename DstXprType, typename SrcXprType, typename Functor>
EIGEN_STRONG_INLINE
const DstXprType& copy_using_evaluator(const EigenBase<DstXprType>& dst, const EigenBase<SrcXprType>& src, const Functor &func)
{
return noalias_copy_using_evaluator(dst.const_cast_derived(),
AddEvalIfAssumingAliasing<SrcXprType>::run(src.derived()),
func
);
}
// this mimics operator=
template<typename DstXprType, typename SrcXprType>
EIGEN_STRONG_INLINE
const DstXprType& copy_using_evaluator(const EigenBase<DstXprType>& dst, const EigenBase<SrcXprType>& src)
{
return copy_using_evaluator(dst.const_cast_derived(), src.derived(), internal::assign_op<typename DstXprType::Scalar>());
}
template<typename DstXprType, typename SrcXprType, typename Functor>
EIGEN_STRONG_INLINE
const DstXprType& noalias_copy_using_evaluator(const PlainObjectBase<DstXprType>& dst, const EigenBase<SrcXprType>& src, const Functor &func)
{
#ifdef EIGEN_DEBUG_ASSIGN
internal::copy_using_evaluator_traits<DstXprType, SrcXprType>::debug();
#endif
#ifdef EIGEN_NO_AUTOMATIC_RESIZING
eigen_assert((dst.size()==0 || (IsVectorAtCompileTime ? (dst.size() == src.size())
: (dst.rows() == src.rows() && dst.cols() == src.cols())))
&& "Size mismatch. Automatic resizing is disabled because EIGEN_NO_AUTOMATIC_RESIZING is defined");
#else
dst.const_cast_derived().resizeLike(src.derived());
#endif
call_dense_assignment_loop(dst.const_cast_derived(), src.derived(), func);
return dst.derived();
}
template<typename DstXprType, typename SrcXprType, typename Functor>
EIGEN_STRONG_INLINE
const DstXprType& noalias_copy_using_evaluator(const EigenBase<DstXprType>& dst, const EigenBase<SrcXprType>& src, const Functor &func)
{
call_dense_assignment_loop(dst.const_cast_derived(), src.derived(), func);
return dst.derived();
}
// Based on DenseBase::swap()
// TODO: Check whether we need to do something special for swapping two
// Arrays or Matrices. (Jitse)
// Overload default assignPacket behavior for swapping them
template<typename DstEvaluatorTypeT, typename SrcEvaluatorTypeT>
class swap_kernel : public generic_dense_assignment_kernel<DstEvaluatorTypeT, SrcEvaluatorTypeT, swap_assign_op<typename DstEvaluatorTypeT::Scalar> >
{
typedef generic_dense_assignment_kernel<DstEvaluatorTypeT, SrcEvaluatorTypeT, swap_assign_op<typename DstEvaluatorTypeT::Scalar> > Base;
typedef typename DstEvaluatorTypeT::PacketScalar PacketScalar;
using Base::m_dst;
using Base::m_src;
using Base::m_functor;
public:
typedef typename Base::Scalar Scalar;
typedef typename Base::Index Index;
typedef typename Base::DstXprType DstXprType;
swap_kernel(DstEvaluatorTypeT &dst, const SrcEvaluatorTypeT &src, DstXprType& dstExpr)
: Base(dst, src, swap_assign_op<Scalar>(), dstExpr)
{}
template<int StoreMode, int LoadMode>
void assignPacket(Index row, Index col)
{
m_functor.template swapPacket<StoreMode,LoadMode,PacketScalar>(&m_dst.coeffRef(row,col), &const_cast<SrcEvaluatorTypeT&>(m_src).coeffRef(row,col));
}
template<int StoreMode, int LoadMode>
void assignPacket(Index index)
{
m_functor.template swapPacket<StoreMode,LoadMode,PacketScalar>(&m_dst.coeffRef(index), &const_cast<SrcEvaluatorTypeT&>(m_src).coeffRef(index));
}
// TODO find a simple way not to have to copy/paste this function from generic_dense_assignment_kernel, by simple I mean no CRTP (Gael)
template<int StoreMode, int LoadMode>
void assignPacketByOuterInner(Index outer, Index inner)
{
Index row = Base::rowIndexByOuterInner(outer, inner);
Index col = Base::colIndexByOuterInner(outer, inner);
assignPacket<StoreMode,LoadMode>(row, col);
}
};
template<typename DstXprType, typename SrcXprType>
void swap_using_evaluator(const DstXprType& dst, const SrcXprType& src)
{
// TODO there is too much redundancy with call_dense_assignment_loop
eigen_assert(dst.rows() == src.rows() && dst.cols() == src.cols());
typedef typename evaluator<DstXprType>::type DstEvaluatorType;
typedef typename evaluator<SrcXprType>::type SrcEvaluatorType;
DstEvaluatorType dstEvaluator(dst);
SrcEvaluatorType srcEvaluator(src);
template<typename,typename> struct AssignmentKind { typedef EigenBase2EigenBase Kind; };
template<> struct AssignmentKind<DenseShape,DenseShape> { typedef Dense2Dense Kind; };
typedef swap_kernel<DstEvaluatorType,SrcEvaluatorType> Kernel;
Kernel kernel(dstEvaluator, srcEvaluator, dst.const_cast_derived());
// 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 Scalar = typename DstXprType::Scalar>
struct Assignment;
// The only purpose of this call_assignment() function is to deal with noalias() / AssumeAliasing and automatic transposition.
// Indeed, I (Gael) think that this concept of AssumeAliasing was a mistake, and it makes thing quite complicated.
// So this intermediate function removes everything related to AssumeAliasing such that Assignment
// does not has to bother about these annoying details.
template<typename Dst, typename Src>
void call_assignment(Dst& dst, const Src& src)
{
call_assignment(dst, src, internal::assign_op<typename Dst::Scalar>());
}
template<typename Dst, typename Src>
void call_assignment(const Dst& dst, const Src& src)
{
call_assignment(dst, src, internal::assign_op<typename Dst::Scalar>());
}
// Deal with AssumeAliasing
template<typename Dst, typename Src, typename Func>
void call_assignment(Dst& dst, const Src& src, const Func& func, typename enable_if<evaluator_traits<Src>::AssumeAliasing==1, 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>
void call_assignment(Dst& dst, const Src& src, const Func& func, typename enable_if<evaluator_traits<Src>::AssumeAliasing==0, void*>::type = 0)
{
call_assignment_no_alias(dst, src, func);
}
// by-pass AssumeAliasing
// FIXME the const version should probably not be needed
// When there is no aliasing, we require that 'dst' has been properly resized
template<typename Dst, template <typename> class StorageBase, typename Src, typename Func>
void call_assignment(const NoAlias<Dst,StorageBase>& dst, const Src& src, const Func& func)
{
call_assignment_no_alias(dst.expression(), src, func);
}
template<typename Dst, template <typename> class StorageBase, typename Src, typename Func>
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>
void call_assignment_no_alias(Dst& dst, const Src& src, const Func& func)
{
enum {
NeedToTranspose = ( (int(Dst::RowsAtCompileTime) == 1 && int(Src::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(Dst::ColsAtCompileTime) == 1 && int(Src::RowsAtCompileTime) == 1))
&& int(Dst::SizeAtCompileTime) != 1
};
typename Dst::Index dstRows = NeedToTranspose ? src.cols() : src.rows();
typename Dst::Index dstCols = NeedToTranspose ? src.rows() : src.cols();
if((dst.rows()!=dstRows) || (dst.cols()!=dstCols))
dst.resize(dstRows, dstCols);
dense_assignment_loop<Kernel>::run(kernel);
}
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)
// Based on MatrixBase::operator+= (in CwiseBinaryOp.h)
template<typename DstXprType, typename SrcXprType>
void add_assign_using_evaluator(const MatrixBase<DstXprType>& dst, const MatrixBase<SrcXprType>& src)
// TODO this line is commented to allow matrix = permutation
// Actually, the "Scalar" type for a permutation matrix does not really make sense,
// perhaps it could be void, and EIGEN_CHECK_BINARY_COMPATIBILIY could allow micing void with anything...?
// EIGEN_CHECK_BINARY_COMPATIBILIY(Func,typename ActualDstTypeCleaned::Scalar,typename Src::Scalar);
Assignment<ActualDstTypeCleaned,Src,Func>::run(actualDst, src, func);
}
template<typename Dst, typename Src>
void call_assignment_no_alias(Dst& dst, const Src& src)
{
typedef typename DstXprType::Scalar Scalar;
copy_using_evaluator(dst.derived(), src.derived(), add_assign_op<Scalar>());
call_assignment_no_alias(dst, src, internal::assign_op<typename Dst::Scalar>());
}
// Based on ArrayBase::operator+=
template<typename DstXprType, typename SrcXprType>
void add_assign_using_evaluator(const ArrayBase<DstXprType>& dst, const ArrayBase<SrcXprType>& src)
// forxard declaration
template<typename Dst, typename Src> void check_for_aliasing(const Dst &dst, const Src &src);
// Generic Dense to Dense assignment
template< typename DstXprType, typename SrcXprType, typename Functor, typename Scalar>
struct Assignment<DstXprType, SrcXprType, Functor, Dense2Dense, Scalar>
{
typedef typename DstXprType::Scalar Scalar;
copy_using_evaluator(dst.derived(), src.derived(), add_assign_op<Scalar>());
}
static void run(DstXprType &dst, const SrcXprType &src, const Functor &func)
{
eigen_assert(dst.rows() == src.rows() && dst.cols() == src.cols());
#ifndef EIGEN_NO_DEBUG
internal::check_for_aliasing(dst, src);
#endif
call_dense_assignment_loop(dst, src, func);
}
};
// TODO: Add add_assign_using_evaluator for EigenBase ? (Jitse)
template<typename DstXprType, typename SrcXprType>
void subtract_assign_using_evaluator(const MatrixBase<DstXprType>& dst, const MatrixBase<SrcXprType>& src)
// Generic assignment through evalTo.
// TODO: not sure we have to keep that one, but it helps porting current code to new evaluator mechanism.
template< typename DstXprType, typename SrcXprType, typename Functor, typename Scalar>
struct Assignment<DstXprType, SrcXprType, Functor, EigenBase2EigenBase, Scalar>
{
typedef typename DstXprType::Scalar Scalar;
copy_using_evaluator(dst.derived(), src.derived(), sub_assign_op<Scalar>());
}
template<typename DstXprType, typename SrcXprType>
void subtract_assign_using_evaluator(const ArrayBase<DstXprType>& dst, const ArrayBase<SrcXprType>& src)
{
typedef typename DstXprType::Scalar Scalar;
copy_using_evaluator(dst.derived(), src.derived(), sub_assign_op<Scalar>());
}
template<typename DstXprType, typename SrcXprType>
void multiply_assign_using_evaluator(const ArrayBase<DstXprType>& dst, const ArrayBase<SrcXprType>& src)
{
typedef typename DstXprType::Scalar Scalar;
copy_using_evaluator(dst.derived(), src.derived(), mul_assign_op<Scalar>());
}
template<typename DstXprType, typename SrcXprType>
void divide_assign_using_evaluator(const ArrayBase<DstXprType>& dst, const ArrayBase<SrcXprType>& src)
{
typedef typename DstXprType::Scalar Scalar;
copy_using_evaluator(dst.derived(), src.derived(), div_assign_op<Scalar>());
}
static void run(DstXprType &dst, const SrcXprType &src, const internal::assign_op<typename DstXprType::Scalar> &/*func*/)
{
eigen_assert(dst.rows() == src.rows() && dst.cols() == src.cols());
src.evalTo(dst);
}
};
} // namespace internal

19
Eigen/src/Core/BandMatrix.h
View File

@ -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

32
Eigen/src/Core/Block.h
View File

@ -68,6 +68,7 @@ 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,
IsRowMajor = (MaxRowsAtCompileTime==1&&MaxColsAtCompileTime!=1) ? 1
: (MaxColsAtCompileTime==1&&MaxRowsAtCompileTime!=1) ? 0
@ -80,18 +81,14 @@ 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))) % EIGEN_ALIGN_BYTES) == 0)) ? AlignedBit : 0,
FlagsLinearAccessBit = (RowsAtCompileTime == 1 || ColsAtCompileTime == 1 || (InnerPanel && (traits<XprType>::Flags&LinearAccessBit))) ? LinearAccessBit : 0,
// IsAligned 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
IsAligned = 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) | FlagsLvalueBit | FlagsRowMajorBit
// FIXME DirectAccessBit should not be handled by expressions
};
};
@ -111,6 +108,8 @@ 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
*/
@ -333,6 +332,9 @@ class BlockImpl_dense<XprType,BlockRows,BlockCols, InnerPanel,true>
: public MapBase<Block<XprType, BlockRows, BlockCols, InnerPanel> >
{
typedef Block<XprType, BlockRows, BlockCols, InnerPanel> BlockType;
enum {
XprTypeIsRowMajor = (int(traits<XprType>::Flags)&RowMajorBit) != 0
};
public:
typedef MapBase<BlockType> Base;
@ -343,9 +345,8 @@ class BlockImpl_dense<XprType,BlockRows,BlockCols, InnerPanel,true>
*/
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)
@ -357,7 +358,8 @@ class BlockImpl_dense<XprType,BlockRows,BlockCols, InnerPanel,true>
*/
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)
{
init();
}
@ -368,7 +370,7 @@ class BlockImpl_dense<XprType,BlockRows,BlockCols, InnerPanel,true>
inline BlockImpl_dense(XprType& xpr,
Index startRow, Index startCol,
Index blockRows, Index blockCols)
: Base(internal::const_cast_ptr(&xpr.coeffRef(startRow,startCol)), blockRows, blockCols),
: Base(xpr.data()+xpr.innerStride()*(XprTypeIsRowMajor?startCol:startRow) + xpr.outerStride()*(XprTypeIsRowMajor?startRow:startCol), blockRows, blockCols),
m_xpr(xpr)
{
init();

32
Eigen/src/Core/BooleanRedux.h
View File

@ -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 +80,21 @@ struct any_unroller<Derived, Dynamic>
template<typename Derived>
inline bool DenseBase<Derived>::all() const
{
typedef typename internal::evaluator<Derived>::type Evaluator;
enum {
unroll = SizeAtCompileTime != Dynamic
&& CoeffReadCost != Dynamic
&& Evaluator::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 +106,21 @@ inline bool DenseBase<Derived>::all() const
template<typename Derived>
inline bool DenseBase<Derived>::any() const
{
typedef typename internal::evaluator<Derived>::type Evaluator;
enum {
unroll = SizeAtCompileTime != Dynamic
&& CoeffReadCost != Dynamic
&& Evaluator::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;
}
}

784
Eigen/src/Core/CoreEvaluators.h
View File

File diff suppressed because it is too large Load Diff

109
Eigen/src/Core/CwiseBinaryOp.h
View File

@ -1,7 +1,7 @@
// 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
@ -56,8 +56,9 @@ struct traits<CwiseBinaryOp<BinaryOp, Lhs, Rhs> >
typename Rhs::Scalar
)
>::type Scalar;
typedef typename promote_storage_type<typename traits<Lhs>::StorageKind,
typename traits<Rhs>::StorageKind>::ret StorageKind;
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>::Index,
typename traits<Rhs>::Index>::type Index;
typedef typename Lhs::Nested LhsNested;
@ -65,60 +66,36 @@ struct traits<CwiseBinaryOp<BinaryOp, Lhs, Rhs> >
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),
CoeffReadCost = LhsCoeffReadCost + RhsCoeffReadCost + functor_traits<BinaryOp>::Cost
Flags = _LhsNested::Flags & RowMajorBit
};
};
} // 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>
template<typename BinaryOp, typename LhsType, typename RhsType>
class CwiseBinaryOp : internal::no_assignment_operator,
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>
{
public:
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::nested<LhsType>::type LhsNested;
typedef typename internal::nested<RhsType>::type RhsNested;
typedef typename internal::remove_reference<LhsNested>::type _LhsNested;
typedef typename internal::remove_reference<RhsNested>::type _RhsNested;
@ -165,43 +142,13 @@ 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_DEVICE_FUNC
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_DEVICE_FUNC
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.
@ -213,8 +160,7 @@ 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>());
return derived();
}
@ -227,8 +173,7 @@ 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>());
return derived();
}

7
Eigen/src/Core/CwiseNullaryOp.h
View File

@ -35,12 +35,7 @@ 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
Flags = traits<PlainObjectType>::Flags & RowMajorBit
};
};
}

49
Eigen/src/Core/CwiseUnaryOp.h
View File

@ -1,7 +1,7 @@
// 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>
// Copyright (C) 2006-2008 Benoit Jacob <jacob.benoit.1@gmail.com>
//
// This Source Code Form is subject to the terms of the Mozilla
@ -44,10 +44,7 @@ struct traits<CwiseUnaryOp<UnaryOp, XprType> >
typedef typename XprType::Nested XprTypeNested;
typedef typename remove_reference<XprTypeNested>::type _XprTypeNested;
enum {
Flags = _XprTypeNested::Flags & (
HereditaryBits | LinearAccessBit | AlignedBit
| (functor_traits<UnaryOp>::PacketAccess ? PacketAccessBit : 0)),
CoeffReadCost = _XprTypeNested::CoeffReadCost + functor_traits<UnaryOp>::Cost
Flags = _XprTypeNested::Flags & RowMajorBit
};
};
}
@ -63,6 +60,7 @@ class CwiseUnaryOp : internal::no_assignment_operator,
typedef typename CwiseUnaryOpImpl<UnaryOp, XprType,typename internal::traits<XprType>::StorageKind>::Base Base;
EIGEN_GENERIC_PUBLIC_INTERFACE(CwiseUnaryOp)
typedef typename internal::remove_all<XprType>::type NestedExpression;
EIGEN_DEVICE_FUNC
inline CwiseUnaryOp(const XprType& xpr, const UnaryOp& func = UnaryOp())
@ -92,42 +90,13 @@ class CwiseUnaryOp : internal::no_assignment_operator,
const UnaryOp m_functor;
};
// This is the generic implementation for dense storage.
// It can be used for any expression types implementing the dense concept.
template<typename UnaryOp, typename XprType>
class CwiseUnaryOpImpl<UnaryOp,XprType,Dense>
: public internal::dense_xpr_base<CwiseUnaryOp<UnaryOp, XprType> >::type
// Generic API dispatcher
template<typename UnaryOp, typename XprType, typename StorageKind>
class CwiseUnaryOpImpl
: public internal::generic_xpr_base<CwiseUnaryOp<UnaryOp, XprType> >::type
{
public:
typedef CwiseUnaryOp<UnaryOp, XprType> Derived;
typedef typename internal::dense_xpr_base<CwiseUnaryOp<UnaryOp, XprType> >::type Base;
EIGEN_DENSE_PUBLIC_INTERFACE(Derived)
EIGEN_DEVICE_FUNC
EIGEN_STRONG_INLINE const Scalar coeff(Index rowId, Index colId) const
{
return derived().functor()(derived().nestedExpression().coeff(rowId, colId));
}
template<int LoadMode>
EIGEN_STRONG_INLINE PacketScalar packet(Index rowId, Index colId) const
{
return derived().functor().packetOp(derived().nestedExpression().template packet<LoadMode>(rowId, colId));
}
EIGEN_DEVICE_FUNC
EIGEN_STRONG_INLINE const Scalar coeff(Index index) const
{
return derived().functor()(derived().nestedExpression().coeff(index));
}
template<int LoadMode>
EIGEN_DEVICE_FUNC
EIGEN_STRONG_INLINE PacketScalar packet(Index index) const
{
return derived().functor().packetOp(derived().nestedExpression().template packet<LoadMode>(index));
}
public:
typedef typename internal::generic_xpr_base<CwiseUnaryOp<UnaryOp, XprType> >::type Base;
};
} // end namespace Eigen

37
Eigen/src/Core/CwiseUnaryView.h
View File

@ -37,8 +37,7 @@ struct traits<CwiseUnaryView<ViewOp, MatrixType> >
typedef typename MatrixType::Nested MatrixTypeNested;
typedef typename remove_all<MatrixTypeNested>::type _MatrixTypeNested;
enum {
Flags = (traits<_MatrixTypeNested>::Flags & (HereditaryBits | LvalueBit | LinearAccessBit | DirectAccessBit)),
CoeffReadCost = traits<_MatrixTypeNested>::CoeffReadCost + functor_traits<ViewOp>::Cost,
Flags = traits<_MatrixTypeNested>::Flags & (RowMajorBit | LvalueBit | DirectAccessBit), // FIXME DirectAccessBit should not be handled by expressions
MatrixTypeInnerStride = inner_stride_at_compile_time<MatrixType>::ret,
// need to cast the sizeof's from size_t to int explicitly, otherwise:
// "error: no integral type can represent all of the enumerator values
@ -62,6 +61,7 @@ class CwiseUnaryView : public CwiseUnaryViewImpl<ViewOp, MatrixType, typename in
typedef typename CwiseUnaryViewImpl<ViewOp, MatrixType,typename internal::traits<MatrixType>::StorageKind>::Base Base;
EIGEN_GENERIC_PUBLIC_INTERFACE(CwiseUnaryView)
typedef typename internal::remove_all<MatrixType>::type NestedExpression;
inline CwiseUnaryView(const MatrixType& mat, const ViewOp& func = ViewOp())
: m_matrix(mat), m_functor(func) {}
@ -88,6 +88,15 @@ class CwiseUnaryView : public CwiseUnaryViewImpl<ViewOp, MatrixType, typename in
ViewOp m_functor;
};
// Generic API dispatcher
template<typename ViewOp, typename XprType, typename StorageKind>
class CwiseUnaryViewImpl
: public internal::generic_xpr_base<CwiseUnaryView<ViewOp, XprType> >::type
{
public:
typedef typename internal::generic_xpr_base<CwiseUnaryView<ViewOp, XprType> >::type Base;
};
template<typename ViewOp, typename MatrixType>
class CwiseUnaryViewImpl<ViewOp,MatrixType,Dense>
: public internal::dense_xpr_base< CwiseUnaryView<ViewOp, MatrixType> >::type
@ -100,8 +109,8 @@ class CwiseUnaryViewImpl<ViewOp,MatrixType,Dense>
EIGEN_DENSE_PUBLIC_INTERFACE(Derived)
EIGEN_INHERIT_ASSIGNMENT_OPERATORS(CwiseUnaryViewImpl)
inline Scalar* data() { return &coeffRef(0); }
inline const Scalar* data() const { return &coeff(0); }
inline Scalar* data() { return &(this->coeffRef(0)); }
inline const Scalar* data() const { return &(this->coeff(0)); }
inline Index innerStride() const
{
@ -112,26 +121,6 @@ class CwiseUnaryViewImpl<ViewOp,MatrixType,Dense>
{
return derived().nestedExpression().outerStride() * sizeof(typename internal::traits<MatrixType>::Scalar) / sizeof(Scalar);
}
EIGEN_STRONG_INLINE CoeffReturnType coeff(Index row, Index col) const
{
return derived().functor()(derived().nestedExpression().coeff(row, col));
}
EIGEN_STRONG_INLINE CoeffReturnType coeff(Index index) const
{
return derived().functor()(derived().nestedExpression().coeff(index));
}
EIGEN_STRONG_INLINE Scalar& coeffRef(Index row, Index col)
{
return derived().functor()(const_cast_derived().nestedExpression().coeffRef(row, col));
}
EIGEN_STRONG_INLINE Scalar& coeffRef(Index index)
{
return derived().functor()(const_cast_derived().nestedExpression().coeffRef(index));
}
};
} // end namespace Eigen

38
Eigen/src/Core/DenseBase.h
View File

@ -74,16 +74,6 @@ template<typename Derived> class DenseBase
using Base::colIndexByOuterInner;
using Base::coeff;
using Base::coeffByOuterInner;
using Base::packet;
using Base::packetByOuterInner;
using Base::writePacket;
using Base::writePacketByOuterInner;
using Base::coeffRef;
using Base::coeffRefByOuterInner;
using Base::copyCoeff;
using Base::copyCoeffByOuterInner;
using Base::copyPacket;
using Base::copyPacketByOuterInner;
using Base::operator();
using Base::operator[];
using Base::x;
@ -169,11 +159,6 @@ template<typename Derived> class DenseBase
InnerSizeAtCompileTime = int(IsVectorAtCompileTime) ? int(SizeAtCompileTime)
: int(IsRowMajor) ? int(ColsAtCompileTime) : int(RowsAtCompileTime),
CoeffReadCost = internal::traits<Derived>::CoeffReadCost,
/**< This is a rough measure of how expensive it is to read one coefficient from
* this expression.
*/
InnerStrideAtCompileTime = internal::inner_stride_at_compile_time<Derived>::ret,
OuterStrideAtCompileTime = internal::outer_stride_at_compile_time<Derived>::ret
};
@ -278,7 +263,8 @@ template<typename Derived> class DenseBase
Derived& operator=(const ReturnByValue<OtherDerived>& func);
#ifndef EIGEN_PARSED_BY_DOXYGEN
/** Copies \a other into *this without evaluating other. \returns a reference to *this. */
/** Copies \a other into *this without evaluating other. \returns a reference to *this.
* \deprecated */
template<typename OtherDerived>
EIGEN_DEVICE_FUNC
Derived& lazyAssign(const DenseBase<OtherDerived>& other);
@ -287,8 +273,10 @@ template<typename Derived> class DenseBase
EIGEN_DEVICE_FUNC
CommaInitializer<Derived> operator<< (const Scalar& s);
// TODO flagged is temporarly disabled. It seems useless now
template<unsigned int Added,unsigned int Removed>
const Flagged<Derived, Added, Removed> flagged() const;
const Derived& flagged() const
{ return derived(); }
template<typename OtherDerived>
EIGEN_DEVICE_FUNC
@ -301,13 +289,6 @@ template<typename Derived> class DenseBase
ConstTransposeReturnType transpose() const;
EIGEN_DEVICE_FUNC
void transposeInPlace();
#ifndef EIGEN_NO_DEBUG
protected:
template<typename OtherDerived>
void checkTransposeAliasing(const OtherDerived& other) const;
public:
#endif
EIGEN_DEVICE_FUNC static const ConstantReturnType
Constant(Index rows, Index cols, const Scalar& value);
@ -387,7 +368,7 @@ template<typename Derived> class DenseBase
// size types on MSVC.
return typename internal::eval<Derived>::type(derived());
}
/** swaps *this with the expression \a other.
*
*/
@ -396,7 +377,8 @@ template<typename Derived> class DenseBase
void swap(const DenseBase<OtherDerived>& other,
int = OtherDerived::ThisConstantIsPrivateInPlainObjectBase)
{
SwapWrapper<Derived>(derived()).lazyAssign(other.derived());
eigen_assert(rows()==other.rows() && cols()==other.cols());
call_assignment(derived(), other.const_cast_derived(), internal::swap_assign_op<Scalar>());
}
/** swaps *this with the matrix or array \a other.
@ -406,10 +388,10 @@ template<typename Derived> class DenseBase
EIGEN_DEVICE_FUNC
void swap(PlainObjectBase<OtherDerived>& other)
{
SwapWrapper<Derived>(derived()).lazyAssign(other.derived());
eigen_assert(rows()==other.rows() && cols()==other.cols());
call_assignment(derived(), other.derived(), internal::swap_assign_op<Scalar>());
}
EIGEN_DEVICE_FUNC inline const NestByValue<Derived> nestByValue() const;
EIGEN_DEVICE_FUNC inline const ForceAlignedAccess<Derived> forceAlignedAccess() const;
EIGEN_DEVICE_FUNC inline ForceAlignedAccess<Derived> forceAlignedAccess();

169
Eigen/src/Core/DenseCoeffsBase.h
View File

@ -97,8 +97,8 @@ class DenseCoeffsBase<Derived,ReadOnlyAccessors> : public EigenBase<Derived>
EIGEN_STRONG_INLINE CoeffReturnType coeff(Index row, Index col) const
{
eigen_internal_assert(row >= 0 && row < rows()
&& col >= 0 && col < cols());
return derived().coeff(row, col);
&& col >= 0 && col < cols());
return typename internal::evaluator<Derived>::type(derived()).coeff(row,col);
}
EIGEN_DEVICE_FUNC
@ -117,7 +117,7 @@ class DenseCoeffsBase<Derived,ReadOnlyAccessors> : public EigenBase<Derived>
{
eigen_assert(row >= 0 && row < rows()
&& col >= 0 && col < cols());
return derived().coeff(row, col);
return coeff(row, col);
}
/** Short version: don't use this function, use
@ -140,7 +140,7 @@ class DenseCoeffsBase<Derived,ReadOnlyAccessors> : public EigenBase<Derived>
coeff(Index index) const
{
eigen_internal_assert(index >= 0 && index < size());
return derived().coeff(index);
return typename internal::evaluator<Derived>::type(derived()).coeff(index);
}
@ -159,7 +159,7 @@ class DenseCoeffsBase<Derived,ReadOnlyAccessors> : public EigenBase<Derived>
EIGEN_STATIC_ASSERT(Derived::IsVectorAtCompileTime,
THE_BRACKET_OPERATOR_IS_ONLY_FOR_VECTORS__USE_THE_PARENTHESIS_OPERATOR_INSTEAD)
eigen_assert(index >= 0 && index < size());
return derived().coeff(index);
return coeff(index);
}
/** \returns the coefficient at given index.
@ -177,7 +177,7 @@ class DenseCoeffsBase<Derived,ReadOnlyAccessors> : public EigenBase<Derived>
operator()(Index index) const
{
eigen_assert(index >= 0 && index < size());
return derived().coeff(index);
return coeff(index);
}
/** equivalent to operator[](0). */
@ -217,9 +217,8 @@ class DenseCoeffsBase<Derived,ReadOnlyAccessors> : public EigenBase<Derived>
template<int LoadMode>
EIGEN_STRONG_INLINE PacketReturnType packet(Index row, Index col) const
{
eigen_internal_assert(row >= 0 && row < rows()
&& col >= 0 && col < cols());
return derived().template packet<LoadMode>(row,col);
eigen_internal_assert(row >= 0 && row < rows() && col >= 0 && col < cols());
return typename internal::evaluator<Derived>::type(derived()).template packet<LoadMode>(row,col);
}
@ -245,7 +244,7 @@ class DenseCoeffsBase<Derived,ReadOnlyAccessors> : public EigenBase<Derived>
EIGEN_STRONG_INLINE PacketReturnType packet(Index index) const
{
eigen_internal_assert(index >= 0 && index < size());
return derived().template packet<LoadMode>(index);
return typename internal::evaluator<Derived>::type(derived()).template packet<LoadMode>(index);
}
protected:
@ -325,8 +324,8 @@ class DenseCoeffsBase<Derived, WriteAccessors> : public DenseCoeffsBase<Derived,
EIGEN_STRONG_INLINE Scalar& coeffRef(Index row, Index col)
{
eigen_internal_assert(row >= 0 && row < rows()
&& col >= 0 && col < cols());
return derived().coeffRef(row, col);
&& col >= 0 && col < cols());
return typename internal::evaluator<Derived>::type(derived()).coeffRef(row,col);
}
EIGEN_DEVICE_FUNC
@ -348,7 +347,7 @@ class DenseCoeffsBase<Derived, WriteAccessors> : public DenseCoeffsBase<Derived,
{
eigen_assert(row >= 0 && row < rows()
&& col >= 0 && col < cols());
return derived().coeffRef(row, col);
return coeffRef(row, col);
}
@ -372,7 +371,7 @@ class DenseCoeffsBase<Derived, WriteAccessors> : public DenseCoeffsBase<Derived,
coeffRef(Index index)
{
eigen_internal_assert(index >= 0 && index < size());
return derived().coeffRef(index);
return typename internal::evaluator<Derived>::type(derived()).coeffRef(index);
}
/** \returns a reference to the coefficient at given index.
@ -389,7 +388,7 @@ class DenseCoeffsBase<Derived, WriteAccessors> : public DenseCoeffsBase<Derived,
EIGEN_STATIC_ASSERT(Derived::IsVectorAtCompileTime,
THE_BRACKET_OPERATOR_IS_ONLY_FOR_VECTORS__USE_THE_PARENTHESIS_OPERATOR_INSTEAD)
eigen_assert(index >= 0 && index < size());
return derived().coeffRef(index);
return coeffRef(index);
}
/** \returns a reference to the coefficient at given index.
@ -406,7 +405,7 @@ class DenseCoeffsBase<Derived, WriteAccessors> : public DenseCoeffsBase<Derived,
operator()(Index index)
{
eigen_assert(index >= 0 && index < size());
return derived().coeffRef(index);
return coeffRef(index);
}
/** equivalent to operator[](0). */
@ -432,144 +431,6 @@ class DenseCoeffsBase<Derived, WriteAccessors> : public DenseCoeffsBase<Derived,
EIGEN_DEVICE_FUNC
EIGEN_STRONG_INLINE Scalar&
w() { return (*this)[3]; }
/** \internal
* Stores the given packet of coefficients, at the given row and column of this expression. It is your responsibility
* to ensure that a packet really starts there. This method is only available on expressions having the
* PacketAccessBit.
*
* The \a LoadMode parameter may have the value \a #Aligned or \a #Unaligned. Its effect is to select
* the appropriate vectorization instruction. Aligned access is faster, but is only possible for packets
* starting at an address which is a multiple of the packet size.
*/
template<int StoreMode>
EIGEN_STRONG_INLINE void writePacket
(Index row, Index col, const typename internal::packet_traits<Scalar>::type& val)
{
eigen_internal_assert(row >= 0 && row < rows()
&& col >= 0 && col < cols());
derived().template writePacket<StoreMode>(row,col,val);
}
/** \internal */
template<int StoreMode>
EIGEN_STRONG_INLINE void writePacketByOuterInner
(Index outer, Index inner, const typename internal::packet_traits<Scalar>::type& val)
{
writePacket<StoreMode>(rowIndexByOuterInner(outer, inner),
colIndexByOuterInner(outer, inner),
val);
}
/** \internal
* Stores the given packet of coefficients, at the given index in this expression. It is your responsibility
* to ensure that a packet really starts there. This method is only available on expressions having the
* PacketAccessBit and the LinearAccessBit.
*
* The \a LoadMode parameter may have the value \a Aligned or \a Unaligned. Its effect is to select
* the appropriate vectorization instruction. Aligned access is faster, but is only possible for packets
* starting at an address which is a multiple of the packet size.
*/
template<int StoreMode>
EIGEN_STRONG_INLINE void writePacket
(Index index, const typename internal::packet_traits<Scalar>::type& val)
{
eigen_internal_assert(index >= 0 && index < size());
derived().template writePacket<StoreMode>(index,val);
}
#ifndef EIGEN_PARSED_BY_DOXYGEN
/** \internal Copies the coefficient at position (row,col) of other into *this.
*
* This method is overridden in SwapWrapper, allowing swap() assignments to share 99% of their code
* with usual assignments.
*
* Outside of this internal usage, this method has probably no usefulness. It is hidden in the public API dox.
*/
template<typename OtherDerived>
EIGEN_DEVICE_FUNC
EIGEN_STRONG_INLINE void copyCoeff(Index row, Index col, const DenseBase<OtherDerived>& other)
{
eigen_internal_assert(row >= 0 && row < rows()
&& col >= 0 && col < cols());
derived().coeffRef(row, col) = other.derived().coeff(row, col);
}
/** \internal Copies the coefficient at the given index of other into *this.
*
* This method is overridden in SwapWrapper, allowing swap() assignments to share 99% of their code
* with usual assignments.
*
* Outside of this internal usage, this method has probably no usefulness. It is hidden in the public API dox.
*/
template<typename OtherDerived>
EIGEN_DEVICE_FUNC
EIGEN_STRONG_INLINE void copyCoeff(Index index, const DenseBase<OtherDerived>& other)
{
eigen_internal_assert(index >= 0 && index < size());
derived().coeffRef(index) = other.derived().coeff(index);
}
template<typename OtherDerived>
EIGEN_DEVICE_FUNC
EIGEN_STRONG_INLINE void copyCoeffByOuterInner(Index outer, Index inner, const DenseBase<OtherDerived>& other)
{
const Index row = rowIndexByOuterInner(outer,inner);
const Index col = colIndexByOuterInner(outer,inner);
// derived() is important here: copyCoeff() may be reimplemented in Derived!
derived().copyCoeff(row, col, other);
}
/** \internal Copies the packet at position (row,col) of other into *this.
*
* This method is overridden in SwapWrapper, allowing swap() assignments to share 99% of their code
* with usual assignments.
*
* Outside of this internal usage, this method has probably no usefulness. It is hidden in the public API dox.
*/
template<typename OtherDerived, int StoreMode, int LoadMode>
EIGEN_STRONG_INLINE void copyPacket(Index row, Index col, const DenseBase<OtherDerived>& other)
{
eigen_internal_assert(row >= 0 && row < rows()
&& col >= 0 && col < cols());
derived().template writePacket<StoreMode>(row, col,
other.derived().template packet<LoadMode>(row, col));
}
/** \internal Copies the packet at the given index of other into *this.
*
* This method is overridden in SwapWrapper, allowing swap() assignments to share 99% of their code
* with usual assignments.
*
* Outside of this internal usage, this method has probably no usefulness. It is hidden in the public API dox.
*/
template<typename OtherDerived, int StoreMode, int LoadMode>
EIGEN_STRONG_INLINE void copyPacket(Index index, const DenseBase<OtherDerived>& other)
{
eigen_internal_assert(index >= 0 && index < size());
derived().template writePacket<StoreMode>(index,
other.derived().template packet<LoadMode>(index));
}
/** \internal */
template<typename OtherDerived, int StoreMode, int LoadMode>
EIGEN_STRONG_INLINE void copyPacketByOuterInner(Index outer, Index inner, const DenseBase<OtherDerived>& other)
{
const Index row = rowIndexByOuterInner(outer,inner);
const Index col = colIndexByOuterInner(outer,inner);
// derived() is important here: copyCoeff() may be reimplemented in Derived!
derived().template copyPacket< OtherDerived, StoreMode, LoadMode>(row, col, other);
}
#endif
};
/** \brief Base class providing direct read-only coefficient access to matrices and arrays.

3
Eigen/src/Core/Diagonal.h
View File

@ -52,8 +52,7 @@ struct traits<Diagonal<MatrixType,DiagIndex> >
MatrixType::MaxColsAtCompileTime - EIGEN_PLAIN_ENUM_MAX( DiagIndex, 0))),
MaxColsAtCompileTime = 1,
MaskLvalueBit = is_lvalue<MatrixType>::value ? LvalueBit : 0,
Flags = (unsigned int)_MatrixTypeNested::Flags & (HereditaryBits | LinearAccessBit | MaskLvalueBit | DirectAccessBit) & ~RowMajorBit,
CoeffReadCost = _MatrixTypeNested::CoeffReadCost,
Flags = (unsigned int)_MatrixTypeNested::Flags & (RowMajorBit | MaskLvalueBit | DirectAccessBit) & ~RowMajorBit, // FIXME DirectAccessBit should not be handled by expressions
MatrixTypeOuterStride = outer_stride_at_compile_time<MatrixType>::ret,
InnerStrideAtCompileTime = MatrixTypeOuterStride == Dynamic ? Dynamic : MatrixTypeOuterStride+1,
OuterStrideAtCompileTime = 0

62
Eigen/src/Core/DiagonalMatrix.h
View File

@ -30,7 +30,7 @@ class DiagonalBase : public EigenBase<Derived>
MaxRowsAtCompileTime = DiagonalVectorType::MaxSizeAtCompileTime,
MaxColsAtCompileTime = DiagonalVectorType::MaxSizeAtCompileTime,
IsVectorAtCompileTime = 0,
Flags = 0
Flags = NoPreferredStorageOrderBit
};
typedef Matrix<Scalar, RowsAtCompileTime, ColsAtCompileTime, 0, MaxRowsAtCompileTime, MaxColsAtCompileTime> DenseMatrixType;
@ -44,18 +44,7 @@ class DiagonalBase : public EigenBase<Derived>
EIGEN_DEVICE_FUNC
DenseMatrixType toDenseMatrix() const { return derived(); }
template<typename DenseDerived>
EIGEN_DEVICE_FUNC
void evalTo(MatrixBase<DenseDerived> &other) const;
template<typename DenseDerived>
EIGEN_DEVICE_FUNC
void addTo(MatrixBase<DenseDerived> &other) const
{ other.diagonal() += diagonal(); }
template<typename DenseDerived>
EIGEN_DEVICE_FUNC
void subTo(MatrixBase<DenseDerived> &other) const
{ other.diagonal() -= diagonal(); }
EIGEN_DEVICE_FUNC
inline const DiagonalVectorType& diagonal() const { return derived().diagonal(); }
EIGEN_DEVICE_FUNC
@ -66,14 +55,12 @@ class DiagonalBase : public EigenBase<Derived>
EIGEN_DEVICE_FUNC
inline Index cols() const { return diagonal().size(); }
/** \returns the diagonal matrix product of \c *this by the matrix \a matrix.
*/
template<typename MatrixDerived>
EIGEN_DEVICE_FUNC
const DiagonalProduct<MatrixDerived, Derived, OnTheLeft>
const Product<Derived,MatrixDerived,LazyProduct>
operator*(const MatrixBase<MatrixDerived> &matrix) const
{
return DiagonalProduct<MatrixDerived, Derived, OnTheLeft>(matrix.derived(), derived());
return Product<Derived, MatrixDerived, LazyProduct>(derived(),matrix.derived());
}
EIGEN_DEVICE_FUNC
@ -97,13 +84,6 @@ class DiagonalBase : public EigenBase<Derived>
}
};
template<typename Derived>
template<typename DenseDerived>
void DiagonalBase<Derived>::evalTo(MatrixBase<DenseDerived> &other) const
{
other.setZero();
other.diagonal() = diagonal();
}
#endif
/** \class DiagonalMatrix
@ -125,10 +105,10 @@ struct traits<DiagonalMatrix<_Scalar,SizeAtCompileTime,MaxSizeAtCompileTime> >
: traits<Matrix<_Scalar,SizeAtCompileTime,SizeAtCompileTime,0,MaxSizeAtCompileTime,MaxSizeAtCompileTime> >
{
typedef Matrix<_Scalar,SizeAtCompileTime,1,0,MaxSizeAtCompileTime,1> DiagonalVectorType;
typedef Dense StorageKind;
typedef DiagonalShape StorageKind;
typedef DenseIndex Index;
enum {
Flags = LvalueBit
Flags = LvalueBit | NoPreferredStorageOrderBit
};
};
}
@ -249,13 +229,14 @@ struct traits<DiagonalWrapper<_DiagonalVectorType> >
typedef _DiagonalVectorType DiagonalVectorType;
typedef typename DiagonalVectorType::Scalar Scalar;
typedef typename DiagonalVectorType::Index Index;
typedef typename DiagonalVectorType::StorageKind StorageKind;
typedef DiagonalShape StorageKind;
typedef typename traits<DiagonalVectorType>::XprKind XprKind;
enum {
RowsAtCompileTime = DiagonalVectorType::SizeAtCompileTime,
ColsAtCompileTime = DiagonalVectorType::SizeAtCompileTime,
MaxRowsAtCompileTime = DiagonalVectorType::SizeAtCompileTime,
MaxColsAtCompileTime = DiagonalVectorType::SizeAtCompileTime,
Flags = traits<DiagonalVectorType>::Flags & LvalueBit
MaxRowsAtCompileTime = DiagonalVectorType::MaxSizeAtCompileTime,
MaxColsAtCompileTime = DiagonalVectorType::MaxSizeAtCompileTime,
Flags = (traits<DiagonalVectorType>::Flags & LvalueBit) | NoPreferredStorageOrderBit
};
};
}
@ -326,6 +307,27 @@ bool MatrixBase<Derived>::isDiagonal(const RealScalar& prec) const
return true;
}
namespace internal {
template<> struct storage_kind_to_shape<DiagonalShape> { typedef DiagonalShape Shape; };
struct Diagonal2Dense {};
template<> struct AssignmentKind<DenseShape,DiagonalShape> { typedef Diagonal2Dense Kind; };
// Diagonal matrix to Dense assignment
template< typename DstXprType, typename SrcXprType, typename Functor, typename Scalar>
struct Assignment<DstXprType, SrcXprType, Functor, Diagonal2Dense, Scalar>
{
static void run(DstXprType &dst, const SrcXprType &src, const internal::assign_op<typename DstXprType::Scalar> &/*func*/)
{
dst.setZero();
dst.diagonal() = src.diagonal();
}
};
} // namespace internal
} // end namespace Eigen
#endif // EIGEN_DIAGONALMATRIX_H

106
Eigen/src/Core/DiagonalProduct.h
View File

@ -13,116 +13,14 @@
namespace Eigen {
namespace internal {
template<typename MatrixType, typename DiagonalType, int ProductOrder>
struct traits<DiagonalProduct<MatrixType, DiagonalType, ProductOrder> >
: traits<MatrixType>
{
typedef typename scalar_product_traits<typename MatrixType::Scalar, typename DiagonalType::Scalar>::ReturnType Scalar;
enum {
RowsAtCompileTime = MatrixType::RowsAtCompileTime,
ColsAtCompileTime = MatrixType::ColsAtCompileTime,
MaxRowsAtCompileTime = MatrixType::MaxRowsAtCompileTime,
MaxColsAtCompileTime = MatrixType::MaxColsAtCompileTime,
_StorageOrder = MatrixType::Flags & RowMajorBit ? RowMajor : ColMajor,
_ScalarAccessOnDiag = !((int(_StorageOrder) == ColMajor && int(ProductOrder) == OnTheLeft)
||(int(_StorageOrder) == RowMajor && int(ProductOrder) == OnTheRight)),
_SameTypes = is_same<typename MatrixType::Scalar, typename DiagonalType::Scalar>::value,
// FIXME currently we need same types, but in the future the next rule should be the one
//_Vectorizable = bool(int(MatrixType::Flags)&PacketAccessBit) && ((!_PacketOnDiag) || (_SameTypes && bool(int(DiagonalType::DiagonalVectorType::Flags)&PacketAccessBit))),
_Vectorizable = bool(int(MatrixType::Flags)&PacketAccessBit) && _SameTypes && (_ScalarAccessOnDiag || (bool(int(DiagonalType::DiagonalVectorType::Flags)&PacketAccessBit))),
_LinearAccessMask = (RowsAtCompileTime==1 || ColsAtCompileTime==1) ? LinearAccessBit : 0,
Flags = ((HereditaryBits|_LinearAccessMask) & (unsigned int)(MatrixType::Flags)) | (_Vectorizable ? PacketAccessBit : 0) | AlignedBit,//(int(MatrixType::Flags)&int(DiagonalType::DiagonalVectorType::Flags)&AlignedBit),
CoeffReadCost = NumTraits<Scalar>::MulCost + MatrixType::CoeffReadCost + DiagonalType::DiagonalVectorType::CoeffReadCost
};
};
}
template<typename MatrixType, typename DiagonalType, int ProductOrder>
class DiagonalProduct : internal::no_assignment_operator,
public MatrixBase<DiagonalProduct<MatrixType, DiagonalType, ProductOrder> >
{
public:
typedef MatrixBase<DiagonalProduct> Base;
EIGEN_DENSE_PUBLIC_INTERFACE(DiagonalProduct)
inline DiagonalProduct(const MatrixType& matrix, const DiagonalType& diagonal)
: m_matrix(matrix), m_diagonal(diagonal)
{
eigen_assert(diagonal.diagonal().size() == (ProductOrder == OnTheLeft ? matrix.rows() : matrix.cols()));
}
EIGEN_STRONG_INLINE Index rows() const { return m_matrix.rows(); }
EIGEN_STRONG_INLINE Index cols() const { return m_matrix.cols(); }
EIGEN_STRONG_INLINE const Scalar coeff(Index row, Index col) const
{
return m_diagonal.diagonal().coeff(ProductOrder == OnTheLeft ? row : col) * m_matrix.coeff(row, col);
}
EIGEN_STRONG_INLINE const Scalar coeff(Index idx) const
{
enum {
StorageOrder = int(MatrixType::Flags) & RowMajorBit ? RowMajor : ColMajor
};
return coeff(int(StorageOrder)==ColMajor?idx:0,int(StorageOrder)==ColMajor?0:idx);
}
template<int LoadMode>
EIGEN_STRONG_INLINE PacketScalar packet(Index row, Index col) const
{
enum {
StorageOrder = Flags & RowMajorBit ? RowMajor : ColMajor
};
const Index indexInDiagonalVector = ProductOrder == OnTheLeft ? row : col;
return packet_impl<LoadMode>(row,col,indexInDiagonalVector,typename internal::conditional<
((int(StorageOrder) == RowMajor && int(ProductOrder) == OnTheLeft)
||(int(StorageOrder) == ColMajor && int(ProductOrder) == OnTheRight)), internal::true_type, internal::false_type>::type());
}
template<int LoadMode>
EIGEN_STRONG_INLINE PacketScalar packet(Index idx) const
{
enum {
StorageOrder = int(MatrixType::Flags) & RowMajorBit ? RowMajor : ColMajor
};
return packet<LoadMode>(int(StorageOrder)==ColMajor?idx:0,int(StorageOrder)==ColMajor?0:idx);
}
protected:
template<int LoadMode>
EIGEN_STRONG_INLINE PacketScalar packet_impl(Index row, Index col, Index id, internal::true_type) const
{
return internal::pmul(m_matrix.template packet<LoadMode>(row, col),
internal::pset1<PacketScalar>(m_diagonal.diagonal().coeff(id)));
}
template<int LoadMode>
EIGEN_STRONG_INLINE PacketScalar packet_impl(Index row, Index col, Index id, internal::false_type) const
{
enum {
InnerSize = (MatrixType::Flags & RowMajorBit) ? MatrixType::ColsAtCompileTime : MatrixType::RowsAtCompileTime,
DiagonalVectorPacketLoadMode = (LoadMode == Aligned && (((InnerSize%16) == 0) || (int(DiagonalType::DiagonalVectorType::Flags)&AlignedBit)==AlignedBit) ? Aligned : Unaligned)
};
return internal::pmul(m_matrix.template packet<LoadMode>(row, col),
m_diagonal.diagonal().template packet<DiagonalVectorPacketLoadMode>(id));
}
typename MatrixType::Nested m_matrix;
typename DiagonalType::Nested m_diagonal;
};
/** \returns the diagonal matrix product of \c *this by the diagonal matrix \a diagonal.
*/
template<typename Derived>
template<typename DiagonalDerived>
inline const DiagonalProduct<Derived, DiagonalDerived, OnTheRight>
inline const Product<Derived, DiagonalDerived, LazyProduct>
MatrixBase<Derived>::operator*(const DiagonalBase<DiagonalDerived> &a_diagonal) const
{
return DiagonalProduct<Derived, DiagonalDerived, OnTheRight>(derived(), a_diagonal.derived());
return Product<Derived, DiagonalDerived, LazyProduct>(derived(),a_diagonal.derived());
}
} // end namespace Eigen

7
Eigen/src/Core/Dot.h
View File

@ -113,8 +113,7 @@ template<typename Derived>
inline const typename MatrixBase<Derived>::PlainObject
MatrixBase<Derived>::normalized() const
{
typedef typename internal::nested<Derived>::type Nested;
typedef typename internal::remove_reference<Nested>::type _Nested;
typedef typename internal::nested_eval<Derived,2>::type _Nested;
_Nested n(derived());
return n / n.norm();
}
@ -206,8 +205,8 @@ template<typename OtherDerived>
bool MatrixBase<Derived>::isOrthogonal
(const MatrixBase<OtherDerived>& other, const RealScalar& prec) const
{
typename internal::nested<Derived,2>::type nested(derived());
typename internal::nested<OtherDerived,2>::type otherNested(other.derived());
typename internal::nested_eval<Derived,2>::type nested(derived());
typename internal::nested_eval<OtherDerived,2>::type otherNested(other.derived());
return numext::abs2(nested.dot(otherNested)) <= prec * prec * nested.squaredNorm() * otherNested.squaredNorm();
}

6
Eigen/src/Core/EigenBase.h
View File

@ -121,7 +121,7 @@ template<typename Derived>
template<typename OtherDerived>
Derived& DenseBase<Derived>::operator=(const EigenBase<OtherDerived> &other)
{
other.derived().evalTo(derived());
call_assignment(derived(), other.derived());
return derived();
}
@ -129,7 +129,7 @@ template<typename Derived>
template<typename OtherDerived>
Derived& DenseBase<Derived>::operator+=(const EigenBase<OtherDerived> &other)
{
other.derived().addTo(derived());
call_assignment(derived(), other.derived(), internal::add_assign_op<Scalar>());
return derived();
}
@ -137,7 +137,7 @@ template<typename Derived>
template<typename OtherDerived>
Derived& DenseBase<Derived>::operator-=(const EigenBase<OtherDerived> &other)
{
other.derived().subTo(derived());
call_assignment(derived(), other.derived(), internal::sub_assign_op<Scalar>());
return derived();
}

4
Eigen/src/Core/Fuzzy.h
View File

@ -23,8 +23,8 @@ struct isApprox_selector
static bool run(const Derived& x, const OtherDerived& y, const typename Derived::RealScalar& prec)
{
EIGEN_USING_STD_MATH(min);
typename internal::nested<Derived,2>::type nested(x);
typename internal::nested<OtherDerived,2>::type otherNested(y);
typename internal::nested_eval<Derived,2>::type nested(x);
typename internal::nested_eval<OtherDerived,2>::type otherNested(y);
return (nested - otherNested).cwiseAbs2().sum() <= prec * prec * (min)(nested.cwiseAbs2().sum(), otherNested.cwiseAbs2().sum());
}
};

422
Eigen/src/Core/GeneralProduct.h
View File

@ -13,28 +13,6 @@
namespace Eigen {
/** \class GeneralProduct
* \ingroup Core_Module
*
* \brief Expression of the product of two general matrices or vectors
*
* \param LhsNested the type used to store the left-hand side
* \param RhsNested the type used to store the right-hand side
* \param ProductMode the type of the product
*
* This class represents an expression of the product of two general matrices.
* We call a general matrix, a dense matrix with full storage. For instance,
* This excludes triangular, selfadjoint, and sparse matrices.
* It is the return type of the operator* between general matrices. Its template
* arguments are determined automatically by ProductReturnType. Therefore,
* GeneralProduct should never be used direclty. To determine the result type of a
* function which involves a matrix product, use ProductReturnType::Type.
*
* \sa ProductReturnType, MatrixBase::operator*(const MatrixBase<OtherDerived>&)
*/
template<typename Lhs, typename Rhs, int ProductType = internal::product_type<Lhs,Rhs>::value>
class GeneralProduct;
enum {
Large = 2,
Small = 3
@ -59,14 +37,14 @@ template<typename Lhs, typename Rhs> struct product_type
typedef typename remove_all<Lhs>::type _Lhs;
typedef typename remove_all<Rhs>::type _Rhs;
enum {
MaxRows = _Lhs::MaxRowsAtCompileTime,
Rows = _Lhs::RowsAtCompileTime,
MaxCols = _Rhs::MaxColsAtCompileTime,
Cols = _Rhs::ColsAtCompileTime,
MaxDepth = EIGEN_SIZE_MIN_PREFER_FIXED(_Lhs::MaxColsAtCompileTime,
_Rhs::MaxRowsAtCompileTime),
Depth = EIGEN_SIZE_MIN_PREFER_FIXED(_Lhs::ColsAtCompileTime,
_Rhs::RowsAtCompileTime)
MaxRows = traits<_Lhs>::MaxRowsAtCompileTime,
Rows = traits<_Lhs>::RowsAtCompileTime,
MaxCols = traits<_Rhs>::MaxColsAtCompileTime,
Cols = traits<_Rhs>::ColsAtCompileTime,
MaxDepth = EIGEN_SIZE_MIN_PREFER_FIXED(traits<_Lhs>::MaxColsAtCompileTime,
traits<_Rhs>::MaxRowsAtCompileTime),
Depth = EIGEN_SIZE_MIN_PREFER_FIXED(traits<_Lhs>::ColsAtCompileTime,
traits<_Rhs>::RowsAtCompileTime)
};
// the splitting into different lines of code here, introducing the _select enums and the typedef below,
@ -81,7 +59,8 @@ private:
public:
enum {
value = selector::ret
value = selector::ret,
ret = selector::ret
};
#ifdef EIGEN_DEBUG_PRODUCT
static void debug()
@ -97,6 +76,31 @@ public:
#endif
};
// template<typename Lhs, typename Rhs> struct product_tag
// {
// private:
//
// typedef typename remove_all<Lhs>::type _Lhs;
// typedef typename remove_all<Rhs>::type _Rhs;
// enum {
// Rows = _Lhs::RowsAtCompileTime,
// Cols = _Rhs::ColsAtCompileTime,
// Depth = EIGEN_SIZE_MIN_PREFER_FIXED(_Lhs::ColsAtCompileTime, _Rhs::RowsAtCompileTime)
// };
//
// enum {
// rows_select = Rows==1 ? int(Rows) : int(Large),
// cols_select = Cols==1 ? int(Cols) : int(Large),
// depth_select = Depth==1 ? int(Depth) : int(Large)
// };
// typedef product_type_selector<rows_select, cols_select, depth_select> selector;
//
// public:
// enum {
// ret = selector::ret
// };
//
// };
/* The following allows to select the kind of product at compile time
* based on the three dimensions of the product.
@ -127,54 +131,6 @@ template<> struct product_type_selector<Large,Large,Small> { enum
} // end namespace internal
/** \class ProductReturnType
* \ingroup Core_Module
*
* \brief Helper class to get the correct and optimized returned type of operator*
*
* \param Lhs the type of the left-hand side
* \param Rhs the type of the right-hand side
* \param ProductMode the type of the product (determined automatically by internal::product_mode)
*
* This class defines the typename Type representing the optimized product expression
* between two matrix expressions. In practice, using ProductReturnType<Lhs,Rhs>::Type
* is the recommended way to define the result type of a function returning an expression
* which involve a matrix product. The class Product should never be
* used directly.
*
* \sa class Product, MatrixBase::operator*(const MatrixBase<OtherDerived>&)
*/
template<typename Lhs, typename Rhs, int ProductType>
struct ProductReturnType
{
// TODO use the nested type to reduce instanciations ????
// typedef typename internal::nested<Lhs,Rhs::ColsAtCompileTime>::type LhsNested;
// typedef typename internal::nested<Rhs,Lhs::RowsAtCompileTime>::type RhsNested;
typedef GeneralProduct<Lhs/*Nested*/, Rhs/*Nested*/, ProductType> Type;
};
template<typename Lhs, typename Rhs>
struct ProductReturnType<Lhs,Rhs,CoeffBasedProductMode>
{
typedef typename internal::nested<Lhs, Rhs::ColsAtCompileTime, typename internal::plain_matrix_type<Lhs>::type >::type LhsNested;
typedef typename internal::nested<Rhs, Lhs::RowsAtCompileTime, typename internal::plain_matrix_type<Rhs>::type >::type RhsNested;
typedef CoeffBasedProduct<LhsNested, RhsNested, EvalBeforeAssigningBit | EvalBeforeNestingBit> Type;
};
template<typename Lhs, typename Rhs>
struct ProductReturnType<Lhs,Rhs,LazyCoeffBasedProductMode>
{
typedef typename internal::nested<Lhs, Rhs::ColsAtCompileTime, typename internal::plain_matrix_type<Lhs>::type >::type LhsNested;
typedef typename internal::nested<Rhs, Lhs::RowsAtCompileTime, typename internal::plain_matrix_type<Rhs>::type >::type RhsNested;
typedef CoeffBasedProduct<LhsNested, RhsNested, NestByRefBit> Type;
};
// this is a workaround for sun CC
template<typename Lhs, typename Rhs>
struct LazyProductReturnType : public ProductReturnType<Lhs,Rhs,LazyCoeffBasedProductMode>
{};
/***********************************************************************
* Implementation of Inner Vector Vector Product
***********************************************************************/
@ -186,119 +142,10 @@ struct LazyProductReturnType : public ProductReturnType<Lhs,Rhs,LazyCoeffBasedPr
// product ends up to a row-vector times col-vector product... To tackle this use
// case, we could have a specialization for Block<MatrixType,1,1> with: operator=(Scalar x);
namespace internal {
template<typename Lhs, typename Rhs>
struct traits<GeneralProduct<Lhs,Rhs,InnerProduct> >
: traits<Matrix<typename scalar_product_traits<typename Lhs::Scalar, typename Rhs::Scalar>::ReturnType,1,1> >
{};
}
template<typename Lhs, typename Rhs>
class GeneralProduct<Lhs, Rhs, InnerProduct>
: internal::no_assignment_operator,
public Matrix<typename internal::scalar_product_traits<typename Lhs::Scalar, typename Rhs::Scalar>::ReturnType,1,1>
{
typedef Matrix<typename internal::scalar_product_traits<typename Lhs::Scalar, typename Rhs::Scalar>::ReturnType,1,1> Base;
public:
GeneralProduct(const Lhs& lhs, const Rhs& rhs)
{
EIGEN_STATIC_ASSERT((internal::is_same<typename Lhs::RealScalar, typename Rhs::RealScalar>::value),
YOU_MIXED_DIFFERENT_NUMERIC_TYPES__YOU_NEED_TO_USE_THE_CAST_METHOD_OF_MATRIXBASE_TO_CAST_NUMERIC_TYPES_EXPLICITLY)
Base::coeffRef(0,0) = (lhs.transpose().cwiseProduct(rhs)).sum();
}
/** Convertion to scalar */
operator const typename Base::Scalar() const {
return Base::coeff(0,0);
}
};
/***********************************************************************
* Implementation of Outer Vector Vector Product
***********************************************************************/
namespace internal {
// Column major
template<typename ProductType, typename Dest, typename Func>
EIGEN_DONT_INLINE void outer_product_selector_run(const ProductType& prod, Dest& dest, const Func& func, const false_type&)
{
typedef typename Dest::Index Index;
// FIXME make sure lhs is sequentially stored
// FIXME not very good if rhs is real and lhs complex while alpha is real too
const Index cols = dest.cols();
for (Index j=0; j<cols; ++j)
func(dest.col(j), prod.rhs().coeff(j) * prod.lhs());
}
// Row major
template<typename ProductType, typename Dest, typename Func>
EIGEN_DONT_INLINE void outer_product_selector_run(const ProductType& prod, Dest& dest, const Func& func, const true_type&) {
typedef typename Dest::Index Index;
// FIXME make sure rhs is sequentially stored
// FIXME not very good if lhs is real and rhs complex while alpha is real too
const Index rows = dest.rows();
for (Index i=0; i<rows; ++i)
func(dest.row(i), prod.lhs().coeff(i) * prod.rhs());
}
template<typename Lhs, typename Rhs>
struct traits<GeneralProduct<Lhs,Rhs,OuterProduct> >
: traits<ProductBase<GeneralProduct<Lhs,Rhs,OuterProduct>, Lhs, Rhs> >
{};
}
template<typename Lhs, typename Rhs>
class GeneralProduct<Lhs, Rhs, OuterProduct>
: public ProductBase<GeneralProduct<Lhs,Rhs,OuterProduct>, Lhs, Rhs>
{
template<typename T> struct IsRowMajor : internal::conditional<(int(T::Flags)&RowMajorBit), internal::true_type, internal::false_type>::type {};
public:
EIGEN_PRODUCT_PUBLIC_INTERFACE(GeneralProduct)
GeneralProduct(const Lhs& lhs, const Rhs& rhs) : Base(lhs,rhs)
{
EIGEN_STATIC_ASSERT((internal::is_same<typename Lhs::RealScalar, typename Rhs::RealScalar>::value),
YOU_MIXED_DIFFERENT_NUMERIC_TYPES__YOU_NEED_TO_USE_THE_CAST_METHOD_OF_MATRIXBASE_TO_CAST_NUMERIC_TYPES_EXPLICITLY)
}
struct set { template<typename Dst, typename Src> void operator()(const Dst& dst, const Src& src) const { dst.const_cast_derived() = src; } };
struct add { template<typename Dst, typename Src> void operator()(const Dst& dst, const Src& src) const { dst.const_cast_derived() += src; } };
struct sub { template<typename Dst, typename Src> void operator()(const Dst& dst, const Src& src) const { dst.const_cast_derived() -= src; } };
struct adds {
Scalar m_scale;
adds(const Scalar& s) : m_scale(s) {}
template<typename Dst, typename Src> void operator()(const Dst& dst, const Src& src) const {
dst.const_cast_derived() += m_scale * src;
}
};
template<typename Dest>
inline void evalTo(Dest& dest) const {
internal::outer_product_selector_run(*this, dest, set(), IsRowMajor<Dest>());
}
template<typename Dest>
inline void addTo(Dest& dest) const {
internal::outer_product_selector_run(*this, dest, add(), IsRowMajor<Dest>());
}
template<typename Dest>
inline void subTo(Dest& dest) const {
internal::outer_product_selector_run(*this, dest, sub(), IsRowMajor<Dest>());
}
template<typename Dest> void scaleAndAddTo(Dest& dest, const Scalar& alpha) const
{
internal::outer_product_selector_run(*this, dest, adds(alpha), IsRowMajor<Dest>());
}
};
/***********************************************************************
* Implementation of General Matrix Vector Product
***********************************************************************/
@ -312,60 +159,13 @@ class GeneralProduct<Lhs, Rhs, OuterProduct>
*/
namespace internal {
template<typename Lhs, typename Rhs>
struct traits<GeneralProduct<Lhs,Rhs,GemvProduct> >
: traits<ProductBase<GeneralProduct<Lhs,Rhs,GemvProduct>, Lhs, Rhs> >
{};
template<int Side, int StorageOrder, bool BlasCompatible>
struct gemv_selector;
struct gemv_dense_sense_selector;
} // end namespace internal
template<typename Lhs, typename Rhs>
class GeneralProduct<Lhs, Rhs, GemvProduct>
: public ProductBase<GeneralProduct<Lhs,Rhs,GemvProduct>, Lhs, Rhs>
{
public:
EIGEN_PRODUCT_PUBLIC_INTERFACE(GeneralProduct)
typedef typename Lhs::Scalar LhsScalar;
typedef typename Rhs::Scalar RhsScalar;
GeneralProduct(const Lhs& a_lhs, const Rhs& a_rhs) : Base(a_lhs,a_rhs)
{
// EIGEN_STATIC_ASSERT((internal::is_same<typename Lhs::Scalar, typename Rhs::Scalar>::value),
// YOU_MIXED_DIFFERENT_NUMERIC_TYPES__YOU_NEED_TO_USE_THE_CAST_METHOD_OF_MATRIXBASE_TO_CAST_NUMERIC_TYPES_EXPLICITLY)
}
enum { Side = Lhs::IsVectorAtCompileTime ? OnTheLeft : OnTheRight };
typedef typename internal::conditional<int(Side)==OnTheRight,_LhsNested,_RhsNested>::type MatrixType;
template<typename Dest> void scaleAndAddTo(Dest& dst, const Scalar& alpha) const
{
eigen_assert(m_lhs.rows() == dst.rows() && m_rhs.cols() == dst.cols());
internal::gemv_selector<Side,(int(MatrixType::Flags)&RowMajorBit) ? RowMajor : ColMajor,
bool(internal::blas_traits<MatrixType>::HasUsableDirectAccess)>::run(*this, dst, alpha);
}
};
namespace internal {
// The vector is on the left => transposition
template<int StorageOrder, bool BlasCompatible>
struct gemv_selector<OnTheLeft,StorageOrder,BlasCompatible>
{
template<typename ProductType, typename Dest>
static void run(const ProductType& prod, Dest& dest, const typename ProductType::Scalar& alpha)
{
Transpose<Dest> destT(dest);
enum { OtherStorageOrder = StorageOrder == RowMajor ? ColMajor : RowMajor };
gemv_selector<OnTheRight,OtherStorageOrder,BlasCompatible>
::run(GeneralProduct<Transpose<const typename ProductType::_RhsNested>,Transpose<const typename ProductType::_LhsNested>, GemvProduct>
(prod.rhs().transpose(), prod.lhs().transpose()), destT, alpha);
}
};
template<typename Scalar,int Size,int MaxSize,bool Cond> struct gemv_static_vector_if;
template<typename Scalar,int Size,int MaxSize>
@ -402,27 +202,43 @@ struct gemv_static_vector_if<Scalar,Size,MaxSize,true>
#endif
};
template<> struct gemv_selector<OnTheRight,ColMajor,true>
// The vector is on the left => transposition
template<int StorageOrder, bool BlasCompatible>
struct gemv_dense_sense_selector<OnTheLeft,StorageOrder,BlasCompatible>
{
template<typename ProductType, typename Dest>
static inline void run(const ProductType& prod, Dest& dest, const typename ProductType::Scalar& alpha)
template<typename Lhs, typename Rhs, typename Dest>
static void run(const Lhs &lhs, const Rhs &rhs, Dest& dest, const typename Dest::Scalar& alpha)
{
typedef typename ProductType::Index Index;
typedef typename ProductType::LhsScalar LhsScalar;
typedef typename ProductType::RhsScalar RhsScalar;
typedef typename ProductType::Scalar ResScalar;
typedef typename ProductType::RealScalar RealScalar;
typedef typename ProductType::ActualLhsType ActualLhsType;
typedef typename ProductType::ActualRhsType ActualRhsType;
typedef typename ProductType::LhsBlasTraits LhsBlasTraits;
typedef typename ProductType::RhsBlasTraits RhsBlasTraits;
Transpose<Dest> destT(dest);
enum { OtherStorageOrder = StorageOrder == RowMajor ? ColMajor : RowMajor };
gemv_dense_sense_selector<OnTheRight,OtherStorageOrder,BlasCompatible>
::run(rhs.transpose(), lhs.transpose(), destT, alpha);
}
};
template<> struct gemv_dense_sense_selector<OnTheRight,ColMajor,true>
{
template<typename Lhs, typename Rhs, typename Dest>
static inline void run(const Lhs &lhs, const Rhs &rhs, Dest& dest, const typename Dest::Scalar& alpha)
{
typedef typename Dest::Index Index;
typedef typename Lhs::Scalar LhsScalar;
typedef typename Rhs::Scalar RhsScalar;
typedef typename Dest::Scalar ResScalar;
typedef typename Dest::RealScalar RealScalar;
typedef internal::blas_traits<Lhs> LhsBlasTraits;
typedef typename LhsBlasTraits::DirectLinearAccessType ActualLhsType;
typedef internal::blas_traits<Rhs> RhsBlasTraits;
typedef typename RhsBlasTraits::DirectLinearAccessType ActualRhsType;
typedef Map<Matrix<ResScalar,Dynamic,1>, Aligned> MappedDest;
ActualLhsType actualLhs = LhsBlasTraits::extract(prod.lhs());
ActualRhsType actualRhs = RhsBlasTraits::extract(prod.rhs());
ActualLhsType actualLhs = LhsBlasTraits::extract(lhs);
ActualRhsType actualRhs = RhsBlasTraits::extract(rhs);
ResScalar actualAlpha = alpha * LhsBlasTraits::extractScalarFactor(prod.lhs())
* RhsBlasTraits::extractScalarFactor(prod.rhs());
ResScalar actualAlpha = alpha * LhsBlasTraits::extractScalarFactor(lhs)
* RhsBlasTraits::extractScalarFactor(rhs);
enum {
// FIXME find a way to allow an inner stride on the result if packet_traits<Scalar>::size==1
@ -445,7 +261,7 @@ template<> struct gemv_selector<OnTheRight,ColMajor,true>
if(!evalToDest)
{
#ifdef EIGEN_DENSE_STORAGE_CTOR_PLUGIN
Index size = dest.size();
int size = dest.size();
EIGEN_DENSE_STORAGE_CTOR_PLUGIN
#endif
if(!alphaIsCompatible)
@ -475,34 +291,35 @@ template<> struct gemv_selector<OnTheRight,ColMajor,true>
}
};
template<> struct gemv_selector<OnTheRight,RowMajor,true>
template<> struct gemv_dense_sense_selector<OnTheRight,RowMajor,true>
{
template<typename ProductType, typename Dest>
static void run(const ProductType& prod, Dest& dest, const typename ProductType::Scalar& alpha)
template<typename Lhs, typename Rhs, typename Dest>
static void run(const Lhs &lhs, const Rhs &rhs, Dest& dest, const typename Dest::Scalar& alpha)
{
typedef typename ProductType::LhsScalar LhsScalar;
typedef typename ProductType::RhsScalar RhsScalar;
typedef typename ProductType::Scalar ResScalar;
typedef typename ProductType::Index Index;
typedef typename ProductType::ActualLhsType ActualLhsType;
typedef typename ProductType::ActualRhsType ActualRhsType;
typedef typename ProductType::_ActualRhsType _ActualRhsType;
typedef typename ProductType::LhsBlasTraits LhsBlasTraits;
typedef typename ProductType::RhsBlasTraits RhsBlasTraits;
typedef typename Dest::Index Index;
typedef typename Lhs::Scalar LhsScalar;
typedef typename Rhs::Scalar RhsScalar;
typedef typename Dest::Scalar ResScalar;
typedef internal::blas_traits<Lhs> LhsBlasTraits;
typedef typename LhsBlasTraits::DirectLinearAccessType ActualLhsType;
typedef internal::blas_traits<Rhs> RhsBlasTraits;
typedef typename RhsBlasTraits::DirectLinearAccessType ActualRhsType;
typedef typename internal::remove_all<ActualRhsType>::type ActualRhsTypeCleaned;
typename add_const<ActualLhsType>::type actualLhs = LhsBlasTraits::extract(prod.lhs());
typename add_const<ActualRhsType>::type actualRhs = RhsBlasTraits::extract(prod.rhs());
typename add_const<ActualLhsType>::type actualLhs = LhsBlasTraits::extract(lhs);
typename add_const<ActualRhsType>::type actualRhs = RhsBlasTraits::extract(rhs);
ResScalar actualAlpha = alpha * LhsBlasTraits::extractScalarFactor(prod.lhs())
* RhsBlasTraits::extractScalarFactor(prod.rhs());
ResScalar actualAlpha = alpha * LhsBlasTraits::extractScalarFactor(lhs)
* RhsBlasTraits::extractScalarFactor(rhs);
enum {
// FIXME find a way to allow an inner stride on the result if packet_traits<Scalar>::size==1
// on, the other hand it is good for the cache to pack the vector anyways...
DirectlyUseRhs = _ActualRhsType::InnerStrideAtCompileTime==1
DirectlyUseRhs = ActualRhsTypeCleaned::InnerStrideAtCompileTime==1
};
gemv_static_vector_if<RhsScalar,_ActualRhsType::SizeAtCompileTime,_ActualRhsType::MaxSizeAtCompileTime,!DirectlyUseRhs> static_rhs;
gemv_static_vector_if<RhsScalar,ActualRhsTypeCleaned::SizeAtCompileTime,ActualRhsTypeCleaned::MaxSizeAtCompileTime,!DirectlyUseRhs> static_rhs;
ei_declare_aligned_stack_constructed_variable(RhsScalar,actualRhsPtr,actualRhs.size(),
DirectlyUseRhs ? const_cast<RhsScalar*>(actualRhs.data()) : static_rhs.data());
@ -510,10 +327,10 @@ template<> struct gemv_selector<OnTheRight,RowMajor,true>
if(!DirectlyUseRhs)
{
#ifdef EIGEN_DENSE_STORAGE_CTOR_PLUGIN
Index size = actualRhs.size();
int size = actualRhs.size();
EIGEN_DENSE_STORAGE_CTOR_PLUGIN
#endif
Map<typename _ActualRhsType::PlainObject>(actualRhsPtr, actualRhs.size()) = actualRhs;
Map<typename ActualRhsTypeCleaned::PlainObject>(actualRhsPtr, actualRhs.size()) = actualRhs;
}
general_matrix_vector_product
@ -526,29 +343,29 @@ template<> struct gemv_selector<OnTheRight,RowMajor,true>
}
};
template<> struct gemv_selector<OnTheRight,ColMajor,false>
template<> struct gemv_dense_sense_selector<OnTheRight,ColMajor,false>
{
template<typename ProductType, typename Dest>
static void run(const ProductType& prod, Dest& dest, const typename ProductType::Scalar& alpha)
template<typename Lhs, typename Rhs, typename Dest>
static void run(const Lhs &lhs, const Rhs &rhs, Dest& dest, const typename Dest::Scalar& alpha)
{
typedef typename Dest::Index Index;
// TODO makes sure dest is sequentially stored in memory, otherwise use a temp
const Index size = prod.rhs().rows();
const Index size = rhs.rows();
for(Index k=0; k<size; ++k)
dest += (alpha*prod.rhs().coeff(k)) * prod.lhs().col(k);
dest += (alpha*rhs.coeff(k)) * lhs.col(k);
}
};
template<> struct gemv_selector<OnTheRight,RowMajor,false>
template<> struct gemv_dense_sense_selector<OnTheRight,RowMajor,false>
{
template<typename ProductType, typename Dest>
static void run(const ProductType& prod, Dest& dest, const typename ProductType::Scalar& alpha)
template<typename Lhs, typename Rhs, typename Dest>
static void run(const Lhs &lhs, const Rhs &rhs, Dest& dest, const typename Dest::Scalar& alpha)
{
typedef typename Dest::Index Index;
// TODO makes sure rhs is sequentially stored in memory, otherwise use a temp
const Index rows = prod.rows();
const Index rows = dest.rows();
for(Index i=0; i<rows; ++i)
dest.coeffRef(i) += alpha * (prod.lhs().row(i).cwiseProduct(prod.rhs().transpose())).sum();
dest.coeffRef(i) += alpha * (lhs.row(i).cwiseProduct(rhs.transpose())).sum();
}
};
@ -566,7 +383,6 @@ template<> struct gemv_selector<OnTheRight,RowMajor,false>
*/
#ifndef __CUDACC__
#ifdef EIGEN_TEST_EVALUATORS
template<typename Derived>
template<typename OtherDerived>
inline const Product<Derived, OtherDerived>
@ -597,39 +413,9 @@ MatrixBase<Derived>::operator*(const MatrixBase<OtherDerived> &other) const
return Product<Derived, OtherDerived>(derived(), other.derived());
}
#else
template<typename Derived>
template<typename OtherDerived>
inline const typename ProductReturnType<Derived, OtherDerived>::Type
MatrixBase<Derived>::operator*(const MatrixBase<OtherDerived> &other) const
{
// A note regarding the function declaration: In MSVC, this function will sometimes
// not be inlined since DenseStorage is an unwindable object for dynamic
// matrices and product types are holding a member to store the result.
// Thus it does not help tagging this function with EIGEN_STRONG_INLINE.
enum {
ProductIsValid = Derived::ColsAtCompileTime==Dynamic
|| OtherDerived::RowsAtCompileTime==Dynamic
|| int(Derived::ColsAtCompileTime)==int(OtherDerived::RowsAtCompileTime),
AreVectors = Derived::IsVectorAtCompileTime && OtherDerived::IsVectorAtCompileTime,
SameSizes = EIGEN_PREDICATE_SAME_MATRIX_SIZE(Derived,OtherDerived)
};
// note to the lost user:
// * for a dot product use: v1.dot(v2)
// * for a coeff-wise product use: v1.cwiseProduct(v2)
EIGEN_STATIC_ASSERT(ProductIsValid || !(AreVectors && SameSizes),
INVALID_VECTOR_VECTOR_PRODUCT__IF_YOU_WANTED_A_DOT_OR_COEFF_WISE_PRODUCT_YOU_MUST_USE_THE_EXPLICIT_FUNCTIONS)
EIGEN_STATIC_ASSERT(ProductIsValid || !(SameSizes && !AreVectors),
INVALID_MATRIX_PRODUCT__IF_YOU_WANTED_A_COEFF_WISE_PRODUCT_YOU_MUST_USE_THE_EXPLICIT_FUNCTION)
EIGEN_STATIC_ASSERT(ProductIsValid || SameSizes, INVALID_MATRIX_PRODUCT)
#ifdef EIGEN_DEBUG_PRODUCT
internal::product_type<Derived,OtherDerived>::debug();
#endif
return typename ProductReturnType<Derived,OtherDerived>::Type(derived(), other.derived());
}
#endif
#endif
#endif // __CUDACC__
/** \returns an expression of the matrix product of \c *this and \a other without implicit evaluation.
*
* The returned product will behave like any other expressions: the coefficients of the product will be
@ -643,7 +429,7 @@ MatrixBase<Derived>::operator*(const MatrixBase<OtherDerived> &other) const
*/
template<typename Derived>
template<typename OtherDerived>
const typename LazyProductReturnType<Derived,OtherDerived>::Type
const Product<Derived,OtherDerived,LazyProduct>
MatrixBase<Derived>::lazyProduct(const MatrixBase<OtherDerived> &other) const
{
enum {
@ -662,7 +448,7 @@ MatrixBase<Derived>::lazyProduct(const MatrixBase<OtherDerived> &other) const
INVALID_MATRIX_PRODUCT__IF_YOU_WANTED_A_COEFF_WISE_PRODUCT_YOU_MUST_USE_THE_EXPLICIT_FUNCTION)
EIGEN_STATIC_ASSERT(ProductIsValid || SameSizes, INVALID_MATRIX_PRODUCT)
return typename LazyProductReturnType<Derived,OtherDerived>::Type(derived(), other.derived());
return Product<Derived,OtherDerived,LazyProduct>(derived(), other.derived());
}
} // end namespace Eigen

130
Eigen/src/Core/Inverse.h Normal file
View File

@ -0,0 +1,130 @@
// This file is part of Eigen, a lightweight C++ template library
// for linear algebra.
//
// Copyright (C) 2014 Gael Guennebaud <gael.guennebaud@inria.fr>
//
// This Source Code Form is subject to the terms of the Mozilla
// Public License v. 2.0. If a copy of the MPL was not distributed
// with this file, You can obtain one at http://mozilla.org/MPL/2.0/.
#ifndef EIGEN_INVERSE_H
#define EIGEN_INVERSE_H
namespace Eigen {
// TODO move the general declaration in Core, and rename this file DenseInverseImpl.h, or something like this...
template<typename XprType,typename StorageKind> class InverseImpl;
namespace internal {
template<typename XprType>
struct traits<Inverse<XprType> >
: traits<typename XprType::PlainObject>
{
typedef typename XprType::PlainObject PlainObject;
typedef traits<PlainObject> BaseTraits;
enum {
Flags = BaseTraits::Flags & RowMajorBit,
CoeffReadCost = Dynamic
};
};
} // end namespace internal
/** \class Inverse
*
* \brief Expression of the inverse of another expression
*
* \tparam XprType the type of the expression we are taking the inverse
*
* This class represents an abstract expression of A.inverse()
* and most of the time this is the only way it is used.
*
*/
template<typename XprType>
class Inverse : public InverseImpl<XprType,typename internal::traits<XprType>::StorageKind>
{
public:
typedef typename XprType::Index Index;
typedef typename XprType::PlainObject PlainObject;
typedef typename internal::nested<XprType>::type XprTypeNested;
typedef typename internal::remove_all<XprTypeNested>::type XprTypeNestedCleaned;
Inverse(const XprType &xpr)
: m_xpr(xpr)
{}
EIGEN_DEVICE_FUNC Index rows() const { return m_xpr.rows(); }
EIGEN_DEVICE_FUNC Index cols() const { return m_xpr.cols(); }
EIGEN_DEVICE_FUNC const XprTypeNestedCleaned& nestedExpression() const { return m_xpr; }
protected:
XprTypeNested m_xpr;
};
/** \internal
* Specialization of the Inverse expression for dense expressions.
* Direct access to the coefficients are discared.
* FIXME this intermediate class is probably not needed anymore.
*/
template<typename XprType>
class InverseImpl<XprType,Dense>
: public MatrixBase<Inverse<XprType> >
{
typedef Inverse<XprType> Derived;
public:
typedef MatrixBase<Derived> Base;
EIGEN_DENSE_PUBLIC_INTERFACE(Derived)
typedef typename internal::remove_all<XprType>::type NestedExpression;
private:
Scalar coeff(Index row, Index col) const;
Scalar coeff(Index i) const;
};
namespace internal {
/** \internal
* \brief Default evaluator for Inverse expression.
*
* This default evaluator for Inverse expression simply evaluate the inverse into a temporary
* by a call to internal::call_assignment_no_alias.
* Therefore, inverse implementers only have to specialize Assignment<Dst,Inverse<...>, ...> for
* there own nested expression.
*
* \sa class Inverse
*/
template<typename ArgType>
struct unary_evaluator<Inverse<ArgType> >
: public evaluator<typename Inverse<ArgType>::PlainObject>::type
{
typedef Inverse<ArgType> InverseType;
typedef typename InverseType::PlainObject PlainObject;
typedef typename evaluator<PlainObject>::type Base;
typedef evaluator<InverseType> type;
typedef evaluator<InverseType> nestedType;
enum { Flags = Base::Flags | EvalBeforeNestingBit };
unary_evaluator(const InverseType& inv_xpr)
: m_result(inv_xpr.rows(), inv_xpr.cols())
{
::new (static_cast<Base*>(this)) Base(m_result);
internal::call_assignment_no_alias(m_result, inv_xpr);
}
protected:
PlainObject m_result;
};
} // end namespace internal
} // end namespace Eigen
#endif // EIGEN_INVERSE_H

15
Eigen/src/Core/Map.h
View File

@ -79,22 +79,9 @@ struct traits<Map<PlainObjectType, MapOptions, StrideType> >
OuterStrideAtCompileTime = StrideType::OuterStrideAtCompileTime == 0
? int(PlainObjectType::OuterStrideAtCompileTime)
: int(StrideType::OuterStrideAtCompileTime),
HasNoInnerStride = InnerStrideAtCompileTime == 1,
HasNoOuterStride = StrideType::OuterStrideAtCompileTime == 0,
HasNoStride = HasNoInnerStride && HasNoOuterStride,
IsAligned = bool(EIGEN_ALIGN) && ((int(MapOptions)&Aligned)==Aligned),
IsDynamicSize = PlainObjectType::SizeAtCompileTime==Dynamic,
KeepsPacketAccess = bool(HasNoInnerStride)
&& ( bool(IsDynamicSize)
|| HasNoOuterStride
|| ( OuterStrideAtCompileTime!=Dynamic
&& ((static_cast<int>(sizeof(Scalar))*OuterStrideAtCompileTime)%EIGEN_ALIGN_BYTES)==0 ) ),
Flags0 = TraitsBase::Flags & (~NestByRefBit),
Flags1 = IsAligned ? (int(Flags0) | AlignedBit) : (int(Flags0) & ~AlignedBit),
Flags2 = (bool(HasNoStride) || bool(PlainObjectType::IsVectorAtCompileTime))
? int(Flags1) : int(Flags1 & ~LinearAccessBit),
Flags3 = is_lvalue<PlainObjectType>::value ? int(Flags2) : (int(Flags2) & ~LvalueBit),
Flags = KeepsPacketAccess ? int(Flags3) : (int(Flags3) & ~PacketAccessBit)
Flags = is_lvalue<PlainObjectType>::value ? int(Flags0) : (int(Flags0) & ~LvalueBit)
};
private:
enum { Options }; // Expressions don't have Options

8
Eigen/src/Core/MapBase.h
View File

@ -12,7 +12,7 @@
#define EIGEN_MAPBASE_H
#define EIGEN_STATIC_ASSERT_INDEX_BASED_ACCESS(Derived) \
EIGEN_STATIC_ASSERT((int(internal::traits<Derived>::Flags) & LinearAccessBit) || Derived::IsVectorAtCompileTime, \
EIGEN_STATIC_ASSERT((int(internal::evaluator<Derived>::Flags) & LinearAccessBit) || Derived::IsVectorAtCompileTime, \
YOU_ARE_TRYING_TO_USE_AN_INDEX_BASED_ACCESSOR_ON_AN_EXPRESSION_THAT_DOES_NOT_SUPPORT_THAT)
namespace Eigen {
@ -161,11 +161,7 @@ template<typename Derived> class MapBase<Derived, ReadOnlyAccessors>
EIGEN_DEVICE_FUNC
void checkSanity() const
{
EIGEN_STATIC_ASSERT(EIGEN_IMPLIES(internal::traits<Derived>::Flags&PacketAccessBit,
internal::inner_stride_at_compile_time<Derived>::ret==1),
PACKET_ACCESS_REQUIRES_TO_HAVE_INNER_STRIDE_FIXED_TO_1);
eigen_assert(EIGEN_IMPLIES(internal::traits<Derived>::Flags&AlignedBit, (size_t(m_data) % EIGEN_ALIGN_BYTES) == 0)
&& "data is not aligned");
eigen_assert(EIGEN_IMPLIES(internal::traits<Derived>::IsAligned, (size_t(m_data) % EIGEN_ALIGN_BYTES) == 0) && "data is not aligned");
}
PointerType m_data;

39
Eigen/src/Core/MathFunctions.h
View File

@ -12,6 +12,15 @@
namespace Eigen {
// On WINCE, std::abs is defined for int only, so let's defined our own overloads:
// This issue has been confirmed with MSVC 2008 only, but the issue might exist for more recent versions too.
#if defined(_WIN32_WCE) && defined(_MSC_VER) && _MSC_VER<=1500
long abs(long x) { return (labs(x)); }
double abs(double x) { return (fabs(x)); }
float abs(float x) { return (fabsf(x)); }
long double abs(long double x) { return (fabsl(x)); }
#endif
namespace internal {
/** \internal \struct global_math_functions_filtering_base
@ -308,10 +317,17 @@ struct hypot_impl
using std::sqrt;
RealScalar _x = abs(x);
RealScalar _y = abs(y);
RealScalar p = (max)(_x, _y);
if(p==RealScalar(0)) return 0;
RealScalar q = (min)(_x, _y);
RealScalar qp = q/p;
Scalar p, qp;
if(_x>_y)
{
p = _x;
qp = _y / p;
}
else
{
p = _y;
qp = _x / p;
}
return p * sqrt(RealScalar(1) + qp*qp);
}
};
@ -678,6 +694,21 @@ bool (isfinite)(const std::complex<T>& x)
return isfinite(real(x)) && isfinite(imag(x));
}
// Log base 2 for 32 bits positive integers.
// Conveniently returns 0 for x==0.
inline int log2(int x)
{
eigen_assert(x>=0);
unsigned int v(x);
static const int table[32] = { 0, 9, 1, 10, 13, 21, 2, 29, 11, 14, 16, 18, 22, 25, 3, 30, 8, 12, 20, 28, 15, 17, 24, 7, 19, 27, 23, 6, 26, 5, 4, 31 };
v |= v >> 1;
v |= v >> 2;
v |= v >> 4;
v |= v >> 8;
v |= v >> 16;
return table[(v * 0x07C4ACDDU) >> 27];
}
} // end namespace numext
namespace internal {

3
Eigen/src/Core/Matrix.h
View File

@ -115,7 +115,8 @@ struct traits<Matrix<_Scalar, _Rows, _Cols, _Options, _MaxRows, _MaxCols> >
MaxRowsAtCompileTime = _MaxRows,
MaxColsAtCompileTime = _MaxCols,
Flags = compute_matrix_flags<_Scalar, _Rows, _Cols, _Options, _MaxRows, _MaxCols>::ret,
CoeffReadCost = NumTraits<Scalar>::ReadCost,
// FIXME, the following flag in only used to define NeedsToAlign in PlainObjectBase
EvaluatorFlags = compute_matrix_evaluator_flags<_Scalar, _Rows, _Cols, _Options, _MaxRows, _MaxCols>::ret,
Options = _Options,
InnerStrideAtCompileTime = 1,
OuterStrideAtCompileTime = (Options&RowMajor) ? ColsAtCompileTime : RowsAtCompileTime

27
Eigen/src/Core/MatrixBase.h
View File

@ -66,8 +66,7 @@ template<typename Derived> class MatrixBase
using Base::MaxSizeAtCompileTime;
using Base::IsVectorAtCompileTime;
using Base::Flags;
using Base::CoeffReadCost;
using Base::derived;
using Base::const_cast_derived;
using Base::rows;
@ -81,6 +80,7 @@ template<typename Derived> class MatrixBase
using Base::operator-=;
using Base::operator*=;
using Base::operator/=;
using Base::operator*;
typedef typename Base::CoeffReturnType CoeffReturnType;
typedef typename Base::ConstTransposeReturnType ConstTransposeReturnType;
@ -185,21 +185,15 @@ template<typename Derived> class MatrixBase
{ return this->lazyProduct(other); }
#else
#ifdef EIGEN_TEST_EVALUATORS
template<typename OtherDerived>
const Product<Derived,OtherDerived>
operator*(const MatrixBase<OtherDerived> &other) const;
#else
template<typename OtherDerived>
const typename ProductReturnType<Derived,OtherDerived>::Type
operator*(const MatrixBase<OtherDerived> &other) const;
#endif
#endif
template<typename OtherDerived>
EIGEN_DEVICE_FUNC
const typename LazyProductReturnType<Derived,OtherDerived>::Type
const Product<Derived,OtherDerived,LazyProduct>
lazyProduct(const MatrixBase<OtherDerived> &other) const;
template<typename OtherDerived>
@ -213,7 +207,7 @@ template<typename Derived> class MatrixBase
template<typename DiagonalDerived>
EIGEN_DEVICE_FUNC
const DiagonalProduct<Derived, DiagonalDerived, OnTheRight>
const Product<Derived, DiagonalDerived, LazyProduct>
operator*(const DiagonalBase<DiagonalDerived> &diagonal) const;
template<typename OtherDerived>
@ -333,10 +327,12 @@ template<typename Derived> class MatrixBase
NoAlias<Derived,Eigen::MatrixBase > noalias();
inline const ForceAlignedAccess<Derived> forceAlignedAccess() const;
inline ForceAlignedAccess<Derived> forceAlignedAccess();
template<bool Enable> inline typename internal::add_const_on_value_type<typename internal::conditional<Enable,ForceAlignedAccess<Derived>,Derived&>::type>::type forceAlignedAccessIf() const;
template<bool Enable> inline typename internal::conditional<Enable,ForceAlignedAccess<Derived>,Derived&>::type forceAlignedAccessIf();
// TODO forceAlignedAccess is temporarly disabled
// Need to find a nicer workaround.
inline const Derived& forceAlignedAccess() const { return derived(); }
inline Derived& forceAlignedAccess() { return derived(); }
template<bool Enable> inline const Derived& forceAlignedAccessIf() const { return derived(); }
template<bool Enable> inline Derived& forceAlignedAccessIf() { return derived(); }
Scalar trace() const;
@ -360,7 +356,8 @@ template<typename Derived> class MatrixBase
const PartialPivLU<PlainObject> lu() const;
EIGEN_DEVICE_FUNC
const internal::inverse_impl<Derived> inverse() const;
const Inverse<Derived> inverse() const;
template<typename ResultType>
void computeInverseAndDetWithCheck(
ResultType& inverse,

55
Eigen/src/Core/NoAlias.h
View File

@ -30,68 +30,35 @@ namespace Eigen {
template<typename ExpressionType, template <typename> class StorageBase>
class NoAlias
{
typedef typename ExpressionType::Scalar Scalar;
public:
typedef typename ExpressionType::Scalar Scalar;
NoAlias(ExpressionType& expression) : m_expression(expression) {}
/** Behaves like MatrixBase::lazyAssign(other)
* \sa MatrixBase::lazyAssign() */
template<typename OtherDerived>
EIGEN_DEVICE_FUNC
EIGEN_STRONG_INLINE ExpressionType& operator=(const StorageBase<OtherDerived>& other)
{ return internal::assign_selector<ExpressionType,OtherDerived,false>::run(m_expression,other.derived()); }
/** \sa MatrixBase::operator+= */
{
call_assignment_no_alias(m_expression, other.derived(), internal::assign_op<Scalar>());
return m_expression;
}
template<typename OtherDerived>
EIGEN_DEVICE_FUNC
EIGEN_STRONG_INLINE ExpressionType& operator+=(const StorageBase<OtherDerived>& other)
{
typedef SelfCwiseBinaryOp<internal::scalar_sum_op<Scalar>, ExpressionType, OtherDerived> SelfAdder;
SelfAdder tmp(m_expression);
typedef typename internal::nested<OtherDerived>::type OtherDerivedNested;
typedef typename internal::remove_all<OtherDerivedNested>::type _OtherDerivedNested;
internal::assign_selector<SelfAdder,_OtherDerivedNested,false>::run(tmp,OtherDerivedNested(other.derived()));
call_assignment_no_alias(m_expression, other.derived(), internal::add_assign_op<Scalar>());
return m_expression;
}
/** \sa MatrixBase::operator-= */
template<typename OtherDerived>
EIGEN_DEVICE_FUNC
EIGEN_STRONG_INLINE ExpressionType& operator-=(const StorageBase<OtherDerived>& other)
{
typedef SelfCwiseBinaryOp<internal::scalar_difference_op<Scalar>, ExpressionType, OtherDerived> SelfAdder;
SelfAdder tmp(m_expression);
typedef typename internal::nested<OtherDerived>::type OtherDerivedNested;
typedef typename internal::remove_all<OtherDerivedNested>::type _OtherDerivedNested;
internal::assign_selector<SelfAdder,_OtherDerivedNested,false>::run(tmp,OtherDerivedNested(other.derived()));
call_assignment_no_alias(m_expression, other.derived(), internal::sub_assign_op<Scalar>());
return m_expression;
}
#ifndef EIGEN_PARSED_BY_DOXYGEN
template<typename ProductDerived, typename Lhs, typename Rhs>
EIGEN_DEVICE_FUNC
EIGEN_STRONG_INLINE ExpressionType& operator+=(const ProductBase<ProductDerived, Lhs,Rhs>& other)
{ other.derived().addTo(m_expression); return m_expression; }
template<typename ProductDerived, typename Lhs, typename Rhs>
EIGEN_DEVICE_FUNC
EIGEN_STRONG_INLINE ExpressionType& operator-=(const ProductBase<ProductDerived, Lhs,Rhs>& other)
{ other.derived().subTo(m_expression); return m_expression; }
template<typename Lhs, typename Rhs, int NestingFlags>
EIGEN_STRONG_INLINE ExpressionType& operator+=(const CoeffBasedProduct<Lhs,Rhs,NestingFlags>& other)
{ return m_expression.derived() += CoeffBasedProduct<Lhs,Rhs,NestByRefBit>(other.lhs(), other.rhs()); }
template<typename Lhs, typename Rhs, int NestingFlags>
EIGEN_DEVICE_FUNC
EIGEN_STRONG_INLINE ExpressionType& operator-=(const CoeffBasedProduct<Lhs,Rhs,NestingFlags>& other)
{ return m_expression.derived() -= CoeffBasedProduct<Lhs,Rhs,NestByRefBit>(other.lhs(), other.rhs()); }
template<typename OtherDerived>
ExpressionType& operator=(const ReturnByValue<OtherDerived>& func)
{ return m_expression = func; }
#endif
EIGEN_DEVICE_FUNC
ExpressionType& expression() const
{

92
Eigen/src/Core/PermutationMatrix.h
View File

@ -13,7 +13,8 @@
namespace Eigen {
template<int RowCol,typename IndicesType,typename MatrixType, typename StorageKind> class PermutedImpl;
// TODO: this does not seems to be needed at all:
// template<int RowCol,typename IndicesType,typename MatrixType, typename StorageKind> class PermutedImpl;
/** \class PermutationBase
* \ingroup Core_Module
@ -60,7 +61,6 @@ class PermutationBase : public EigenBase<Derived>
typedef typename Traits::IndicesType IndicesType;
enum {
Flags = Traits::Flags,
CoeffReadCost = Traits::CoeffReadCost,
RowsAtCompileTime = Traits::RowsAtCompileTime,
ColsAtCompileTime = Traits::ColsAtCompileTime,
MaxRowsAtCompileTime = Traits::MaxRowsAtCompileTime,
@ -274,6 +274,7 @@ template<int SizeAtCompileTime, int MaxSizeAtCompileTime, typename _StorageIndex
struct traits<PermutationMatrix<SizeAtCompileTime, MaxSizeAtCompileTime, _StorageIndexType> >
: traits<Matrix<_StorageIndexType,SizeAtCompileTime,SizeAtCompileTime,0,MaxSizeAtCompileTime,MaxSizeAtCompileTime> >
{
typedef PermutationStorage StorageKind;
typedef Matrix<_StorageIndexType, SizeAtCompileTime, 1, 0, MaxSizeAtCompileTime, 1> IndicesType;
typedef typename IndicesType::Index Index;
typedef _StorageIndexType StorageIndexType;
@ -287,6 +288,8 @@ class PermutationMatrix : public PermutationBase<PermutationMatrix<SizeAtCompile
typedef internal::traits<PermutationMatrix> Traits;
public:
typedef const PermutationMatrix& Nested;
#ifndef EIGEN_PARSED_BY_DOXYGEN
typedef typename Traits::IndicesType IndicesType;
typedef typename Traits::StorageIndexType StorageIndexType;
@ -391,6 +394,7 @@ template<int SizeAtCompileTime, int MaxSizeAtCompileTime, typename _StorageIndex
struct traits<Map<PermutationMatrix<SizeAtCompileTime, MaxSizeAtCompileTime, _StorageIndexType>,_PacketAccess> >
: traits<Matrix<_StorageIndexType,SizeAtCompileTime,SizeAtCompileTime,0,MaxSizeAtCompileTime,MaxSizeAtCompileTime> >
{
typedef PermutationStorage StorageKind;
typedef Map<const Matrix<_StorageIndexType, SizeAtCompileTime, 1, 0, MaxSizeAtCompileTime, 1>, _PacketAccess> IndicesType;
typedef typename IndicesType::Index Index;
typedef _StorageIndexType StorageIndexType;
@ -462,8 +466,6 @@ class Map<PermutationMatrix<SizeAtCompileTime, MaxSizeAtCompileTime, _StorageInd
* \sa class PermutationBase, class PermutationMatrix
*/
struct PermutationStorage {};
template<typename _IndicesType> class TranspositionsWrapper;
namespace internal {
template<typename _IndicesType>
@ -477,10 +479,9 @@ struct traits<PermutationWrapper<_IndicesType> >
enum {
RowsAtCompileTime = _IndicesType::SizeAtCompileTime,
ColsAtCompileTime = _IndicesType::SizeAtCompileTime,
MaxRowsAtCompileTime = IndicesType::MaxRowsAtCompileTime,
MaxColsAtCompileTime = IndicesType::MaxColsAtCompileTime,
Flags = 0,
CoeffReadCost = _IndicesType::CoeffReadCost
MaxRowsAtCompileTime = IndicesType::MaxSizeAtCompileTime,
MaxColsAtCompileTime = IndicesType::MaxSizeAtCompileTime,
Flags = 0
};
};
}
@ -509,35 +510,39 @@ class PermutationWrapper : public PermutationBase<PermutationWrapper<_IndicesTyp
typename IndicesType::Nested m_indices;
};
// TODO: Do we need to define these operator* functions? Would it be better to have them inherited
// from MatrixBase?
/** \returns the matrix with the permutation applied to the columns.
*/
template<typename Derived, typename PermutationDerived>
inline const internal::permut_matrix_product_retval<PermutationDerived, Derived, OnTheRight>
operator*(const MatrixBase<Derived>& matrix,
const PermutationBase<PermutationDerived> &permutation)
template<typename MatrixDerived, typename PermutationDerived>
EIGEN_DEVICE_FUNC
const Product<MatrixDerived, PermutationDerived, DefaultProduct>
operator*(const MatrixBase<MatrixDerived> &matrix,
const PermutationBase<PermutationDerived>& permutation)
{
return internal::permut_matrix_product_retval
<PermutationDerived, Derived, OnTheRight>
(permutation.derived(), matrix.derived());
return Product<MatrixDerived, PermutationDerived, DefaultProduct>
(matrix.derived(), permutation.derived());
}
/** \returns the matrix with the permutation applied to the rows.
*/
template<typename Derived, typename PermutationDerived>
inline const internal::permut_matrix_product_retval
<PermutationDerived, Derived, OnTheLeft>
template<typename PermutationDerived, typename MatrixDerived>
EIGEN_DEVICE_FUNC
const Product<PermutationDerived, MatrixDerived, DefaultProduct>
operator*(const PermutationBase<PermutationDerived> &permutation,
const MatrixBase<Derived>& matrix)
const MatrixBase<MatrixDerived>& matrix)
{
return internal::permut_matrix_product_retval
<PermutationDerived, Derived, OnTheLeft>
(permutation.derived(), matrix.derived());
return Product<PermutationDerived, MatrixDerived, DefaultProduct>
(permutation.derived(), matrix.derived());
}
namespace internal {
template<typename PermutationType, typename MatrixType, int Side, bool Transposed>
struct traits<permut_matrix_product_retval<PermutationType, MatrixType, Side, Transposed> >
: traits<typename MatrixType::PlainObject>
{
typedef typename MatrixType::PlainObject ReturnType;
};
@ -617,6 +622,8 @@ struct traits<Transpose<PermutationBase<Derived> > >
} // end namespace internal
// TODO: the specificties should be handled by the evaluator,
// at the very least we should only specialize TransposeImpl
template<typename Derived>
class Transpose<PermutationBase<Derived> >
: public EigenBase<Transpose<PermutationBase<Derived> > >
@ -631,7 +638,6 @@ class Transpose<PermutationBase<Derived> >
typedef typename Derived::DenseMatrixType DenseMatrixType;
enum {
Flags = Traits::Flags,
CoeffReadCost = Traits::CoeffReadCost,
RowsAtCompileTime = Traits::RowsAtCompileTime,
ColsAtCompileTime = Traits::ColsAtCompileTime,
MaxRowsAtCompileTime = Traits::MaxRowsAtCompileTime,
@ -663,19 +669,19 @@ class Transpose<PermutationBase<Derived> >
/** \returns the matrix with the inverse permutation applied to the columns.
*/
template<typename OtherDerived> friend
inline const internal::permut_matrix_product_retval<PermutationType, OtherDerived, OnTheRight, true>
const Product<OtherDerived, Transpose, DefaultProduct>
operator*(const MatrixBase<OtherDerived>& matrix, const Transpose& trPerm)
{
return internal::permut_matrix_product_retval<PermutationType, OtherDerived, OnTheRight, true>(trPerm.m_permutation, matrix.derived());
return Product<OtherDerived, Transpose, DefaultProduct>(matrix.derived(), trPerm.derived());
}
/** \returns the matrix with the inverse permutation applied to the rows.
*/
template<typename OtherDerived>
inline const internal::permut_matrix_product_retval<PermutationType, OtherDerived, OnTheLeft, true>
const Product<Transpose, OtherDerived, DefaultProduct>
operator*(const MatrixBase<OtherDerived>& matrix) const
{
return internal::permut_matrix_product_retval<PermutationType, OtherDerived, OnTheLeft, true>(m_permutation, matrix.derived());
return Product<Transpose, OtherDerived, DefaultProduct>(*this, matrix.derived());
}
const PermutationType& nestedPermutation() const { return m_permutation; }
@ -690,6 +696,38 @@ const PermutationWrapper<const Derived> MatrixBase<Derived>::asPermutation() con
return derived();
}
namespace internal {
// TODO currently a permutation matrix expression has the form PermutationMatrix or PermutationWrapper
// or their transpose; in the future shape should be defined by the expression traits
template<int SizeAtCompileTime, int MaxSizeAtCompileTime, typename IndexType>
struct evaluator_traits<PermutationMatrix<SizeAtCompileTime, MaxSizeAtCompileTime, IndexType> >
{
typedef typename storage_kind_to_evaluator_kind<Dense>::Kind Kind;
typedef PermutationShape Shape;
static const int AssumeAliasing = 0;
};
template<typename IndicesType>
struct evaluator_traits<PermutationWrapper<IndicesType> >
{
typedef typename storage_kind_to_evaluator_kind<Dense>::Kind Kind;
typedef PermutationShape Shape;
static const int AssumeAliasing = 0;
};
template<typename Derived>
struct evaluator_traits<Transpose<PermutationBase<Derived> > >
{
typedef typename storage_kind_to_evaluator_kind<Dense>::Kind Kind;
typedef PermutationShape Shape;
static const int AssumeAliasing = 0;
};
template<> struct AssignmentKind<DenseShape,PermutationShape> { typedef EigenBase2EigenBase Kind; };
} // end namespace internal
} // end namespace Eigen
#endif // EIGEN_PERMUTATIONMATRIX_H

54
Eigen/src/Core/PlainObjectBase.h
View File

@ -128,7 +128,7 @@ class PlainObjectBase : public internal::dense_xpr_base<Derived>::type
DenseStorage<Scalar, Base::MaxSizeAtCompileTime, Base::RowsAtCompileTime, Base::ColsAtCompileTime, Options> m_storage;
public:
enum { NeedsToAlign = SizeAtCompileTime != Dynamic && (internal::traits<Derived>::Flags & AlignedBit) != 0 };
enum { NeedsToAlign = SizeAtCompileTime != Dynamic && (internal::traits<Derived>::EvaluatorFlags & AlignedBit) != 0 };
EIGEN_MAKE_ALIGNED_OPERATOR_NEW_IF(NeedsToAlign)
EIGEN_DEVICE_FUNC
@ -639,22 +639,17 @@ class PlainObjectBase : public internal::dense_xpr_base<Derived>::type
*
* \internal
*/
// aliasing is dealt once in internall::call_assignment
// so at this stage we have to assume aliasing... and resising has to be done later.
template<typename OtherDerived>
EIGEN_DEVICE_FUNC
EIGEN_STRONG_INLINE Derived& _set(const DenseBase<OtherDerived>& other)
{
_set_selector(other.derived(), typename internal::conditional<static_cast<bool>(int(OtherDerived::Flags) & EvalBeforeAssigningBit), internal::true_type, internal::false_type>::type());
internal::call_assignment(this->derived(), other.derived());
return this->derived();
return this->derived();
}
template<typename OtherDerived>
EIGEN_DEVICE_FUNC
EIGEN_STRONG_INLINE void _set_selector(const OtherDerived& other, const internal::true_type&) { _set_noalias(other.eval()); }
template<typename OtherDerived>
EIGEN_DEVICE_FUNC
EIGEN_STRONG_INLINE void _set_selector(const OtherDerived& other, const internal::false_type&) { _set_noalias(other); }
/** \internal Like _set() but additionally makes the assumption that no aliasing effect can happen (which
* is the case when creating a new matrix) so one can enforce lazy evaluation.
*
@ -669,7 +664,8 @@ class PlainObjectBase : public internal::dense_xpr_base<Derived>::type
//_resize_to_match(other);
// the 'false' below means to enforce lazy evaluation. We don't use lazyAssign() because
// it wouldn't allow to copy a row-vector into a column-vector.
return internal::assign_selector<Derived,OtherDerived,false>::run(this->derived(), other.derived());
internal::call_assignment_no_alias(this->derived(), other.derived(), internal::assign_op<Scalar>());
return this->derived();
}
template<typename T0, typename T1>
@ -704,9 +700,12 @@ class PlainObjectBase : public internal::dense_xpr_base<Derived>::type
m_storage.data()[1] = Scalar(val1);
}
// The argument is convertible to the Index type and we either have a non 1x1 Matrix, or a dynamic-sized Array,
// then the argument is meant to be the size of the object.
template<typename T>
EIGEN_DEVICE_FUNC
EIGEN_STRONG_INLINE void _init1(Index size, typename internal::enable_if<Base::SizeAtCompileTime!=1 || !internal::is_convertible<T, Scalar>::value,T>::type* = 0)
EIGEN_STRONG_INLINE void _init1(Index size, typename internal::enable_if< (Base::SizeAtCompileTime!=1 || !internal::is_convertible<T, Scalar>::value)
&& ((!internal::is_same<typename internal::traits<Derived>::XprKind,ArrayXpr>::value || Base::SizeAtCompileTime==Dynamic)),T>::type* = 0)
{
// NOTE MSVC 2008 complains if we directly put bool(NumTraits<T>::IsInteger) as the EIGEN_STATIC_ASSERT argument.
const bool is_integer = NumTraits<T>::IsInteger;
@ -714,6 +713,8 @@ class PlainObjectBase : public internal::dense_xpr_base<Derived>::type
FLOATING_POINT_ARGUMENT_PASSED__INTEGER_WAS_EXPECTED)
resize(size);
}
// We have a 1x1 matrix/array => the argument is interpreted as the value of the unique coefficient (case where scalar type can be implicitely converted)
template<typename T>
EIGEN_DEVICE_FUNC
EIGEN_STRONG_INLINE void _init1(const Scalar& val0, typename internal::enable_if<Base::SizeAtCompileTime==1 && internal::is_convertible<T, Scalar>::value,T>::type* = 0)
@ -722,6 +723,7 @@ class PlainObjectBase : public internal::dense_xpr_base<Derived>::type
m_storage.data()[0] = val0;
}
// We have a 1x1 matrix/array => the argument is interpreted as the value of the unique coefficient (case where scalar type match the index type)
template<typename T>
EIGEN_DEVICE_FUNC
EIGEN_STRONG_INLINE void _init1(const Index& val0,
@ -734,18 +736,21 @@ class PlainObjectBase : public internal::dense_xpr_base<Derived>::type
m_storage.data()[0] = Scalar(val0);
}
// Initialize a fixed size matrix from a pointer to raw data
template<typename T>
EIGEN_DEVICE_FUNC
EIGEN_STRONG_INLINE void _init1(const Scalar* data){
this->_set_noalias(ConstMapType(data));
}
// Initialize an arbitrary matrix from a dense expression
template<typename T, typename OtherDerived>
EIGEN_DEVICE_FUNC
EIGEN_STRONG_INLINE void _init1(const DenseBase<OtherDerived>& other){
this->_set_noalias(other);
}
// Initialize an arbitrary matrix from a generic Eigen expression
template<typename T, typename OtherDerived>
EIGEN_DEVICE_FUNC
EIGEN_STRONG_INLINE void _init1(const EigenBase<OtherDerived>& other){
@ -766,6 +771,31 @@ class PlainObjectBase : public internal::dense_xpr_base<Derived>::type
{
this->derived() = r;
}
// For fixed -size arrays:
template<typename T>
EIGEN_DEVICE_FUNC
EIGEN_STRONG_INLINE void _init1(const Scalar& val0,
typename internal::enable_if< Base::SizeAtCompileTime!=Dynamic
&& Base::SizeAtCompileTime!=1
&& internal::is_convertible<T, Scalar>::value
&& internal::is_same<typename internal::traits<Derived>::XprKind,ArrayXpr>::value,T>::type* = 0)
{
Base::setConstant(val0);
}
template<typename T>
EIGEN_DEVICE_FUNC
EIGEN_STRONG_INLINE void _init1(const Index& val0,
typename internal::enable_if< (!internal::is_same<Index,Scalar>::value)
&& (internal::is_same<Index,T>::value)
&& Base::SizeAtCompileTime!=Dynamic
&& Base::SizeAtCompileTime!=1
&& internal::is_convertible<T, Scalar>::value
&& internal::is_same<typename internal::traits<Derived>::XprKind,ArrayXpr>::value,T*>::type* = 0)
{
Base::setConstant(val0);
}
template<typename MatrixTypeA, typename MatrixTypeB, bool SwapPointers>
friend struct internal::matrix_swap_impl;

174
Eigen/src/Core/Product.h
View File

@ -12,8 +12,7 @@
namespace Eigen {
template<typename Lhs, typename Rhs> class Product;
template<typename Lhs, typename Rhs, typename StorageKind> class ProductImpl;
template<typename Lhs, typename Rhs, int Option, typename StorageKind> class ProductImpl;
/** \class Product
* \ingroup Core_Module
@ -24,38 +23,93 @@ template<typename Lhs, typename Rhs, typename StorageKind> class ProductImpl;
* \param Rhs the type of the right-hand side expression
*
* This class represents an expression of the product of two arbitrary matrices.
*
* The other template parameters are:
* \tparam Option can be DefaultProduct or LazyProduct
*
*/
// Use ProductReturnType to get correct traits, in particular vectorization flags
namespace internal {
template<typename Lhs, typename Rhs>
struct traits<Product<Lhs, Rhs> >
: traits<typename ProductReturnType<Lhs, Rhs>::Type>
{
// We want A+B*C to be of type Product<Matrix, Sum> and not Product<Matrix, Matrix>
// TODO: This flag should eventually go in a separate evaluator traits class
// Determine the scalar of Product<Lhs, Rhs>. This is normally the same as Lhs::Scalar times
// Rhs::Scalar, but product with permutation matrices inherit the scalar of the other factor.
template<typename Lhs, typename Rhs, typename LhsShape = typename evaluator_traits<Lhs>::Shape,
typename RhsShape = typename evaluator_traits<Rhs>::Shape >
struct product_result_scalar
{
typedef typename scalar_product_traits<typename Lhs::Scalar, typename Rhs::Scalar>::ReturnType Scalar;
};
template<typename Lhs, typename Rhs, typename RhsShape>
struct product_result_scalar<Lhs, Rhs, PermutationShape, RhsShape>
{
typedef typename Rhs::Scalar Scalar;
};
template<typename Lhs, typename Rhs, typename LhsShape>
struct product_result_scalar<Lhs, Rhs, LhsShape, PermutationShape>
{
typedef typename Lhs::Scalar Scalar;
};
template<typename Lhs, typename Rhs, int Option>
struct traits<Product<Lhs, Rhs, Option> >
{
typedef typename remove_all<Lhs>::type LhsCleaned;
typedef typename remove_all<Rhs>::type RhsCleaned;
typedef traits<LhsCleaned> LhsTraits;
typedef traits<RhsCleaned> RhsTraits;
typedef MatrixXpr XprKind;
typedef typename product_result_scalar<LhsCleaned,RhsCleaned>::Scalar Scalar;
typedef typename product_promote_storage_type<typename LhsTraits::StorageKind,
typename RhsTraits::StorageKind,
internal::product_type<Lhs,Rhs>::ret>::ret StorageKind;
typedef typename promote_index_type<typename LhsTraits::Index,
typename RhsTraits::Index>::type Index;
enum {
Flags = traits<typename ProductReturnType<Lhs, Rhs>::Type>::Flags & ~(EvalBeforeNestingBit | DirectAccessBit)
RowsAtCompileTime = LhsTraits::RowsAtCompileTime,
ColsAtCompileTime = RhsTraits::ColsAtCompileTime,
MaxRowsAtCompileTime = LhsTraits::MaxRowsAtCompileTime,
MaxColsAtCompileTime = RhsTraits::MaxColsAtCompileTime,
// FIXME: only needed by GeneralMatrixMatrixTriangular
InnerSize = EIGEN_SIZE_MIN_PREFER_FIXED(LhsTraits::ColsAtCompileTime, RhsTraits::RowsAtCompileTime),
// The storage order is somewhat arbitrary here. The correct one will be determined through the evaluator.
Flags = ( (MaxRowsAtCompileTime==1 && MaxColsAtCompileTime!=1)
|| ((LhsTraits::Flags&NoPreferredStorageOrderBit) && (RhsTraits::Flags&RowMajorBit))
|| ((RhsTraits::Flags&NoPreferredStorageOrderBit) && (LhsTraits::Flags&RowMajorBit)) )
? RowMajorBit : (MaxColsAtCompileTime==1 ? 0 : NoPreferredStorageOrderBit)
};
};
} // end namespace internal
template<typename Lhs, typename Rhs>
class Product : public ProductImpl<Lhs,Rhs,typename internal::promote_storage_type<typename internal::traits<Lhs>::StorageKind,
typename internal::traits<Rhs>::StorageKind>::ret>
template<typename _Lhs, typename _Rhs, int Option>
class Product : public ProductImpl<_Lhs,_Rhs,Option,
typename internal::product_promote_storage_type<typename internal::traits<_Lhs>::StorageKind,
typename internal::traits<_Rhs>::StorageKind,
internal::product_type<_Lhs,_Rhs>::ret>::ret>
{
public:
typedef _Lhs Lhs;
typedef _Rhs Rhs;
typedef typename ProductImpl<
Lhs, Rhs,
typename internal::promote_storage_type<typename Lhs::StorageKind,
typename Rhs::StorageKind>::ret>::Base Base;
Lhs, Rhs, Option,
typename internal::product_promote_storage_type<typename internal::traits<Lhs>::StorageKind,
typename internal::traits<Rhs>::StorageKind,
internal::product_type<Lhs,Rhs>::ret>::ret>::Base Base;
EIGEN_GENERIC_PUBLIC_INTERFACE(Product)
typedef typename Lhs::Nested LhsNested;
typedef typename Rhs::Nested RhsNested;
typedef typename internal::nested<Lhs>::type LhsNested;
typedef typename internal::nested<Rhs>::type RhsNested;
typedef typename internal::remove_all<LhsNested>::type LhsNestedCleaned;
typedef typename internal::remove_all<RhsNested>::type RhsNestedCleaned;
@ -78,14 +132,77 @@ class Product : public ProductImpl<Lhs,Rhs,typename internal::promote_storage_ty
RhsNested m_rhs;
};
template<typename Lhs, typename Rhs>
class ProductImpl<Lhs,Rhs,Dense> : public internal::dense_xpr_base<Product<Lhs,Rhs> >::type
{
typedef Product<Lhs, Rhs> Derived;
public:
namespace internal {
template<typename Lhs, typename Rhs, int Option, int ProductTag = internal::product_type<Lhs,Rhs>::ret>
class dense_product_base
: public internal::dense_xpr_base<Product<Lhs,Rhs,Option> >::type
{};
typedef typename internal::dense_xpr_base<Product<Lhs, Rhs> >::type Base;
/** Convertion to scalar for inner-products */
template<typename Lhs, typename Rhs, int Option>
class dense_product_base<Lhs, Rhs, Option, InnerProduct>
: public internal::dense_xpr_base<Product<Lhs,Rhs,Option> >::type
{
typedef Product<Lhs,Rhs,Option> ProductXpr;
typedef typename internal::dense_xpr_base<ProductXpr>::type Base;
public:
using Base::derived;
typedef typename Base::Scalar Scalar;
typedef typename Base::Index Index;
operator const Scalar() const
{
return typename internal::evaluator<ProductXpr>::type(derived()).coeff(0,0);
}
};
} // namespace internal
// Generic API dispatcher
template<typename Lhs, typename Rhs, int Option, typename StorageKind>
class ProductImpl : public internal::generic_xpr_base<Product<Lhs,Rhs,Option>, MatrixXpr, StorageKind>::type
{
public:
typedef typename internal::generic_xpr_base<Product<Lhs,Rhs,Option>, MatrixXpr, StorageKind>::type Base;
};
template<typename Lhs, typename Rhs, int Option>
class ProductImpl<Lhs,Rhs,Option,Dense>
: public internal::dense_product_base<Lhs,Rhs,Option>
{
typedef Product<Lhs, Rhs, Option> Derived;
public:
typedef typename internal::dense_product_base<Lhs, Rhs, Option> Base;
EIGEN_DENSE_PUBLIC_INTERFACE(Derived)
protected:
enum {
IsOneByOne = (RowsAtCompileTime == 1 || RowsAtCompileTime == Dynamic) &&
(ColsAtCompileTime == 1 || ColsAtCompileTime == Dynamic),
EnableCoeff = IsOneByOne || Option==LazyProduct
};
public:
Scalar coeff(Index row, Index col) const
{
EIGEN_STATIC_ASSERT(EnableCoeff, THIS_METHOD_IS_ONLY_FOR_INNER_OR_LAZY_PRODUCTS);
eigen_assert( (Option==LazyProduct) || (this->rows() == 1 && this->cols() == 1) );
return typename internal::evaluator<Derived>::type(derived()).coeff(row,col);
}
Scalar coeff(Index i) const
{
EIGEN_STATIC_ASSERT(EnableCoeff, THIS_METHOD_IS_ONLY_FOR_INNER_OR_LAZY_PRODUCTS);
eigen_assert( (Option==LazyProduct) || (this->rows() == 1 && this->cols() == 1) );
return typename internal::evaluator<Derived>::type(derived()).coeff(i);
}
};
/***************************************************************************
@ -102,6 +219,15 @@ prod(const Lhs& lhs, const Rhs& rhs)
return Product<Lhs,Rhs>(lhs,rhs);
}
/** \internal used to test the evaluator only
*/
template<typename Lhs,typename Rhs>
const Product<Lhs,Rhs,LazyProduct>
lazyprod(const Lhs& lhs, const Rhs& rhs)
{
return Product<Lhs,Rhs,LazyProduct>(lhs,rhs);
}
} // end namespace Eigen
#endif // EIGEN_PRODUCT_H

247
Eigen/src/Core/ProductBase.h
View File

@ -12,253 +12,6 @@
namespace Eigen {
/** \class ProductBase
* \ingroup Core_Module
*
*/
namespace internal {
template<typename Derived, typename _Lhs, typename _Rhs>
struct traits<ProductBase<Derived,_Lhs,_Rhs> >
{
typedef MatrixXpr XprKind;
typedef typename remove_all<_Lhs>::type Lhs;
typedef typename remove_all<_Rhs>::type Rhs;
typedef typename scalar_product_traits<typename Lhs::Scalar, typename Rhs::Scalar>::ReturnType 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;
enum {
RowsAtCompileTime = traits<Lhs>::RowsAtCompileTime,
ColsAtCompileTime = traits<Rhs>::ColsAtCompileTime,
MaxRowsAtCompileTime = traits<Lhs>::MaxRowsAtCompileTime,
MaxColsAtCompileTime = traits<Rhs>::MaxColsAtCompileTime,
Flags = (MaxRowsAtCompileTime==1 ? RowMajorBit : 0)
| EvalBeforeNestingBit | EvalBeforeAssigningBit | NestByRefBit,
// Note that EvalBeforeNestingBit and NestByRefBit
// are not used in practice because nested is overloaded for products
CoeffReadCost = 0 // FIXME why is it needed ?
};
};
}
#define EIGEN_PRODUCT_PUBLIC_INTERFACE(Derived) \
typedef ProductBase<Derived, Lhs, Rhs > Base; \
EIGEN_DENSE_PUBLIC_INTERFACE(Derived) \
typedef typename Base::LhsNested LhsNested; \
typedef typename Base::_LhsNested _LhsNested; \
typedef typename Base::LhsBlasTraits LhsBlasTraits; \
typedef typename Base::ActualLhsType ActualLhsType; \
typedef typename Base::_ActualLhsType _ActualLhsType; \
typedef typename Base::RhsNested RhsNested; \
typedef typename Base::_RhsNested _RhsNested; \
typedef typename Base::RhsBlasTraits RhsBlasTraits; \
typedef typename Base::ActualRhsType ActualRhsType; \
typedef typename Base::_ActualRhsType _ActualRhsType; \
using Base::m_lhs; \
using Base::m_rhs;
template<typename Derived, typename Lhs, typename Rhs>
class ProductBase : public MatrixBase<Derived>
{
public:
typedef MatrixBase<Derived> Base;
EIGEN_DENSE_PUBLIC_INTERFACE(ProductBase)
typedef typename Lhs::Nested LhsNested;
typedef typename internal::remove_all<LhsNested>::type _LhsNested;
typedef internal::blas_traits<_LhsNested> LhsBlasTraits;
typedef typename LhsBlasTraits::DirectLinearAccessType ActualLhsType;
typedef typename internal::remove_all<ActualLhsType>::type _ActualLhsType;
typedef typename internal::traits<Lhs>::Scalar LhsScalar;
typedef typename Rhs::Nested RhsNested;
typedef typename internal::remove_all<RhsNested>::type _RhsNested;
typedef internal::blas_traits<_RhsNested> RhsBlasTraits;
typedef typename RhsBlasTraits::DirectLinearAccessType ActualRhsType;
typedef typename internal::remove_all<ActualRhsType>::type _ActualRhsType;
typedef typename internal::traits<Rhs>::Scalar RhsScalar;
// Diagonal of a product: no need to evaluate the arguments because they are going to be evaluated only once
typedef CoeffBasedProduct<LhsNested, RhsNested, 0> FullyLazyCoeffBaseProductType;
public:
typedef typename Base::PlainObject PlainObject;
ProductBase(const Lhs& a_lhs, const Rhs& a_rhs)
: m_lhs(a_lhs), m_rhs(a_rhs)
{
eigen_assert(a_lhs.cols() == a_rhs.rows()
&& "invalid matrix product"
&& "if you wanted a coeff-wise or a dot product use the respective explicit functions");
}
inline Index rows() const { return m_lhs.rows(); }
inline Index cols() const { return m_rhs.cols(); }
template<typename Dest>
inline void evalTo(Dest& dst) const { dst.setZero(); scaleAndAddTo(dst,Scalar(1)); }
template<typename Dest>
inline void addTo(Dest& dst) const { scaleAndAddTo(dst,Scalar(1)); }
template<typename Dest>
inline void subTo(Dest& dst) const { scaleAndAddTo(dst,Scalar(-1)); }
template<typename Dest>
inline void scaleAndAddTo(Dest& dst, const Scalar& alpha) const { derived().scaleAndAddTo(dst,alpha); }
const _LhsNested& lhs() const { return m_lhs; }
const _RhsNested& rhs() const { return m_rhs; }
// Implicit conversion to the nested type (trigger the evaluation of the product)
operator const PlainObject& () const
{
m_result.resize(m_lhs.rows(), m_rhs.cols());
derived().evalTo(m_result);
return m_result;
}
const Diagonal<const FullyLazyCoeffBaseProductType,0> diagonal() const
{ return FullyLazyCoeffBaseProductType(m_lhs, m_rhs); }
template<int Index>
const Diagonal<FullyLazyCoeffBaseProductType,Index> diagonal() const
{ return FullyLazyCoeffBaseProductType(m_lhs, m_rhs); }
const Diagonal<FullyLazyCoeffBaseProductType,Dynamic> diagonal(Index index) const
{ return FullyLazyCoeffBaseProductType(m_lhs, m_rhs).diagonal(index); }
// restrict coeff accessors to 1x1 expressions. No need to care about mutators here since this isn't an Lvalue expression
typename Base::CoeffReturnType coeff(Index row, Index col) const
{
EIGEN_STATIC_ASSERT_SIZE_1x1(Derived)
eigen_assert(this->rows() == 1 && this->cols() == 1);
Matrix<Scalar,1,1> result = *this;
return result.coeff(row,col);
}
typename Base::CoeffReturnType coeff(Index i) const
{
EIGEN_STATIC_ASSERT_SIZE_1x1(Derived)
eigen_assert(this->rows() == 1 && this->cols() == 1);
Matrix<Scalar,1,1> result = *this;
return result.coeff(i);
}
const Scalar& coeffRef(Index row, Index col) const
{
EIGEN_STATIC_ASSERT_SIZE_1x1(Derived)
eigen_assert(this->rows() == 1 && this->cols() == 1);
return derived().coeffRef(row,col);
}
const Scalar& coeffRef(Index i) const
{
EIGEN_STATIC_ASSERT_SIZE_1x1(Derived)
eigen_assert(this->rows() == 1 && this->cols() == 1);
return derived().coeffRef(i);
}
protected:
LhsNested m_lhs;
RhsNested m_rhs;
mutable PlainObject m_result;
};
// here we need to overload the nested rule for products
// such that the nested type is a const reference to a plain matrix
namespace internal {
template<typename Lhs, typename Rhs, int Mode, int N, typename PlainObject>
struct nested<GeneralProduct<Lhs,Rhs,Mode>, N, PlainObject>
{
typedef PlainObject const& type;
};
}
template<typename NestedProduct>
class ScaledProduct;
// Note that these two operator* functions are not defined as member
// functions of ProductBase, because, otherwise we would have to
// define all overloads defined in MatrixBase. Furthermore, Using
// "using Base::operator*" would not work with MSVC.
//
// Also note that here we accept any compatible scalar types
template<typename Derived,typename Lhs,typename Rhs>
const ScaledProduct<Derived>
operator*(const ProductBase<Derived,Lhs,Rhs>& prod, const typename Derived::Scalar& x)
{ return ScaledProduct<Derived>(prod.derived(), x); }
template<typename Derived,typename Lhs,typename Rhs>
typename internal::enable_if<!internal::is_same<typename Derived::Scalar,typename Derived::RealScalar>::value,
const ScaledProduct<Derived> >::type
operator*(const ProductBase<Derived,Lhs,Rhs>& prod, const typename Derived::RealScalar& x)
{ return ScaledProduct<Derived>(prod.derived(), x); }
template<typename Derived,typename Lhs,typename Rhs>
const ScaledProduct<Derived>
operator*(const typename Derived::Scalar& x,const ProductBase<Derived,Lhs,Rhs>& prod)
{ return ScaledProduct<Derived>(prod.derived(), x); }
template<typename Derived,typename Lhs,typename Rhs>
typename internal::enable_if<!internal::is_same<typename Derived::Scalar,typename Derived::RealScalar>::value,
const ScaledProduct<Derived> >::type
operator*(const typename Derived::RealScalar& x,const ProductBase<Derived,Lhs,Rhs>& prod)
{ return ScaledProduct<Derived>(prod.derived(), x); }
namespace internal {
template<typename NestedProduct>
struct traits<ScaledProduct<NestedProduct> >
: traits<ProductBase<ScaledProduct<NestedProduct>,
typename NestedProduct::_LhsNested,
typename NestedProduct::_RhsNested> >
{
typedef typename traits<NestedProduct>::StorageKind StorageKind;
};
}
template<typename NestedProduct>
class ScaledProduct
: public ProductBase<ScaledProduct<NestedProduct>,
typename NestedProduct::_LhsNested,
typename NestedProduct::_RhsNested>
{
public:
typedef ProductBase<ScaledProduct<NestedProduct>,
typename NestedProduct::_LhsNested,
typename NestedProduct::_RhsNested> Base;
typedef typename Base::Scalar Scalar;
typedef typename Base::PlainObject PlainObject;
// EIGEN_PRODUCT_PUBLIC_INTERFACE(ScaledProduct)
ScaledProduct(const NestedProduct& prod, const Scalar& x)
: Base(prod.lhs(),prod.rhs()), m_prod(prod), m_alpha(x) {}
template<typename Dest>
inline void evalTo(Dest& dst) const { dst.setZero(); scaleAndAddTo(dst, Scalar(1)); }
template<typename Dest>
inline void addTo(Dest& dst) const { scaleAndAddTo(dst, Scalar(1)); }
template<typename Dest>
inline void subTo(Dest& dst) const { scaleAndAddTo(dst, Scalar(-1)); }
template<typename Dest>
inline void scaleAndAddTo(Dest& dst, const Scalar& a_alpha) const { m_prod.derived().scaleAndAddTo(dst,a_alpha * m_alpha); }
const Scalar& alpha() const { return m_alpha; }
protected:
const NestedProduct& m_prod;
Scalar m_alpha;
};
/** \internal
* Overloaded to perform an efficient C = (A*B).lazy() */
template<typename Derived>

914
Eigen/src/Core/ProductEvaluators.h
View File

File diff suppressed because it is too large Load Diff

119
Eigen/src/Core/Redux.h
View File

@ -65,6 +65,25 @@ public:
? CompleteUnrolling
: NoUnrolling
};
#ifdef EIGEN_DEBUG_ASSIGN
static void debug()
{
std::cerr << "Xpr: " << typeid(typename Derived::XprType).name() << std::endl;
std::cerr.setf(std::ios::hex, std::ios::basefield);
EIGEN_DEBUG_VAR(Derived::Flags)
std::cerr.unsetf(std::ios::hex);
EIGEN_DEBUG_VAR(InnerMaxSize)
EIGEN_DEBUG_VAR(PacketSize)
EIGEN_DEBUG_VAR(MightVectorize)
EIGEN_DEBUG_VAR(MayLinearVectorize)
EIGEN_DEBUG_VAR(MaySliceVectorize)
EIGEN_DEBUG_VAR(Traversal)
EIGEN_DEBUG_VAR(UnrollingLimit)
EIGEN_DEBUG_VAR(Unrolling)
std::cerr << std::endl;
}
#endif
};
/***************************************************************************
@ -174,7 +193,7 @@ struct redux_impl<Func, Derived, DefaultTraversal, NoUnrolling>
typedef typename Derived::Scalar Scalar;
typedef typename Derived::Index Index;
EIGEN_DEVICE_FUNC
static EIGEN_STRONG_INLINE Scalar run(const Derived& mat, const Func& func)
static EIGEN_STRONG_INLINE Scalar run(const Derived &mat, const Func& func)
{
eigen_assert(mat.rows()>0 && mat.cols()>0 && "you are using an empty matrix");
Scalar res;
@ -200,14 +219,14 @@ struct redux_impl<Func, Derived, LinearVectorizedTraversal, NoUnrolling>
typedef typename packet_traits<Scalar>::type PacketScalar;
typedef typename Derived::Index Index;
static Scalar run(const Derived& mat, const Func& func)
static Scalar run(const Derived &mat, const Func& func)
{
const Index size = mat.size();
eigen_assert(size && "you are using an empty matrix");
const Index packetSize = packet_traits<Scalar>::size;
const Index alignedStart = internal::first_aligned(mat);
enum {
alignment = bool(Derived::Flags & DirectAccessBit) || bool(Derived::Flags & AlignedBit)
alignment = (bool(Derived::Flags & DirectAccessBit) && bool(packet_traits<Scalar>::AlignedOnScalar)) || bool(Derived::Flags & AlignedBit)
? Aligned : Unaligned
};
const Index alignedSize2 = ((size-alignedStart)/(2*packetSize))*(2*packetSize);
@ -258,7 +277,7 @@ struct redux_impl<Func, Derived, SliceVectorizedTraversal, NoUnrolling>
typedef typename packet_traits<Scalar>::type PacketScalar;
typedef typename Derived::Index Index;
static Scalar run(const Derived& mat, const Func& func)
static Scalar run(const Derived &mat, const Func& func)
{
eigen_assert(mat.rows()>0 && mat.cols()>0 && "you are using an empty matrix");
const Index innerSize = mat.innerSize();
@ -300,7 +319,7 @@ struct redux_impl<Func, Derived, LinearVectorizedTraversal, CompleteUnrolling>
Size = Derived::SizeAtCompileTime,
VectorizedSize = (Size / PacketSize) * PacketSize
};
static EIGEN_STRONG_INLINE Scalar run(const Derived& mat, const Func& func)
static EIGEN_STRONG_INLINE Scalar run(const Derived &mat, const Func& func)
{
eigen_assert(mat.rows()>0 && mat.cols()>0 && "you are using an empty matrix");
if (VectorizedSize > 0) {
@ -315,6 +334,63 @@ struct redux_impl<Func, Derived, LinearVectorizedTraversal, CompleteUnrolling>
}
};
// evaluator adaptor
template<typename _XprType>
class redux_evaluator
{
public:
typedef _XprType XprType;
redux_evaluator(const XprType &xpr) : m_evaluator(xpr), m_xpr(xpr) {}
typedef typename XprType::Index Index;
typedef typename XprType::Scalar Scalar;
typedef typename XprType::CoeffReturnType CoeffReturnType;
typedef typename XprType::PacketScalar PacketScalar;
typedef typename XprType::PacketReturnType PacketReturnType;
enum {
MaxRowsAtCompileTime = XprType::MaxRowsAtCompileTime,
MaxColsAtCompileTime = XprType::MaxColsAtCompileTime,
// TODO we should not remove DirectAccessBit and rather find an elegant way to query the alignment offset at runtime from the evaluator
Flags = evaluator<XprType>::Flags & ~DirectAccessBit,
IsRowMajor = XprType::IsRowMajor,
SizeAtCompileTime = XprType::SizeAtCompileTime,
InnerSizeAtCompileTime = XprType::InnerSizeAtCompileTime,
CoeffReadCost = evaluator<XprType>::CoeffReadCost
};
Index rows() const { return m_xpr.rows(); }
Index cols() const { return m_xpr.cols(); }
Index size() const { return m_xpr.size(); }
Index innerSize() const { return m_xpr.innerSize(); }
Index outerSize() const { return m_xpr.outerSize(); }
CoeffReturnType coeff(Index row, Index col) const
{ return m_evaluator.coeff(row, col); }
CoeffReturnType coeff(Index index) const
{ return m_evaluator.coeff(index); }
template<int LoadMode>
PacketReturnType packet(Index row, Index col) const
{ return m_evaluator.template packet<LoadMode>(row, col); }
template<int LoadMode>
PacketReturnType packet(Index index) const
{ return m_evaluator.template packet<LoadMode>(index); }
CoeffReturnType coeffByOuterInner(Index outer, Index inner) const
{ return m_evaluator.coeff(IsRowMajor ? outer : inner, IsRowMajor ? inner : outer); }
template<int LoadMode>
PacketReturnType packetByOuterInner(Index outer, Index inner) const
{ return m_evaluator.template packet<LoadMode>(IsRowMajor ? outer : inner, IsRowMajor ? inner : outer); }
protected:
typename internal::evaluator<XprType>::nestedType m_evaluator;
const XprType &m_xpr;
};
} // end namespace internal
/***************************************************************************
@ -325,7 +401,7 @@ struct redux_impl<Func, Derived, LinearVectorizedTraversal, CompleteUnrolling>
/** \returns the result of a full redux operation on the whole matrix or vector using \a func
*
* The template parameter \a BinaryOp is the type of the functor \a func which must be
* an associative operator. Both current STL and TR1 functor styles are handled.
* an associative operator. Both current C++98 and C++11 functor styles are handled.
*
* \sa DenseBase::sum(), DenseBase::minCoeff(), DenseBase::maxCoeff(), MatrixBase::colwise(), MatrixBase::rowwise()
*/
@ -334,9 +410,22 @@ template<typename Func>
EIGEN_STRONG_INLINE typename internal::result_of<Func(typename internal::traits<Derived>::Scalar)>::type
DenseBase<Derived>::redux(const Func& func) const
{
typedef typename internal::remove_all<typename Derived::Nested>::type ThisNested;
return internal::redux_impl<Func, ThisNested>
::run(derived(), func);
eigen_assert(this->rows()>0 && this->cols()>0 && "you are using an empty matrix");
// FIXME, eval_nest should be handled by redux_evaluator, however:
// - it is currently difficult to provide the right Flags since they are still handled by the expressions
// - handling it here might reduce the number of template instantiations
// typedef typename internal::nested_eval<Derived,1>::type ThisNested;
// typedef typename internal::remove_all<ThisNested>::type ThisNestedCleaned;
// typedef typename internal::redux_evaluator<ThisNestedCleaned> ThisEvaluator;
//
// ThisNested thisNested(derived());
// ThisEvaluator thisEval(thisNested);
typedef typename internal::redux_evaluator<Derived> ThisEvaluator;
ThisEvaluator thisEval(derived());
return internal::redux_impl<Func, ThisEvaluator>::run(thisEval, func);
}
/** \returns the minimum of all coefficients of \c *this.
@ -346,7 +435,7 @@ template<typename Derived>
EIGEN_STRONG_INLINE typename internal::traits<Derived>::Scalar
DenseBase<Derived>::minCoeff() const
{
return this->redux(Eigen::internal::scalar_min_op<Scalar>());
return derived().redux(Eigen::internal::scalar_min_op<Scalar>());
}
/** \returns the maximum of all coefficients of \c *this.
@ -356,7 +445,7 @@ template<typename Derived>
EIGEN_STRONG_INLINE typename internal::traits<Derived>::Scalar
DenseBase<Derived>::maxCoeff() const
{
return this->redux(Eigen::internal::scalar_max_op<Scalar>());
return derived().redux(Eigen::internal::scalar_max_op<Scalar>());
}
/** \returns the sum of all coefficients of *this
@ -369,7 +458,7 @@ DenseBase<Derived>::sum() const
{
if(SizeAtCompileTime==0 || (SizeAtCompileTime==Dynamic && size()==0))
return Scalar(0);
return this->redux(Eigen::internal::scalar_sum_op<Scalar>());
return derived().redux(Eigen::internal::scalar_sum_op<Scalar>());
}
/** \returns the mean of all coefficients of *this
@ -380,7 +469,7 @@ template<typename Derived>
EIGEN_STRONG_INLINE typename internal::traits<Derived>::Scalar
DenseBase<Derived>::mean() const
{
return Scalar(this->redux(Eigen::internal::scalar_sum_op<Scalar>())) / Scalar(this->size());
return Scalar(derived().redux(Eigen::internal::scalar_sum_op<Scalar>())) / Scalar(this->size());
}
/** \returns the product of all coefficients of *this
@ -396,7 +485,7 @@ DenseBase<Derived>::prod() const
{
if(SizeAtCompileTime==0 || (SizeAtCompileTime==Dynamic && size()==0))
return Scalar(1);
return this->redux(Eigen::internal::scalar_product_op<Scalar>());
return derived().redux(Eigen::internal::scalar_product_op<Scalar>());
}
/** \returns the trace of \c *this, i.e. the sum of the coefficients on the main diagonal.

6
Eigen/src/Core/Ref.h
View File

@ -12,10 +12,6 @@
namespace Eigen {
template<typename Derived> class RefBase;
template<typename PlainObjectType, int Options = 0,
typename StrideType = typename internal::conditional<PlainObjectType::IsVectorAtCompileTime,InnerStride<1>,OuterStride<> >::type > class Ref;
/** \class Ref
* \ingroup Core_Module
*
@ -247,7 +243,7 @@ template<typename TPlainObjectType, int Options, typename StrideType> class Ref<
template<typename Expression>
void construct(const Expression& expr, internal::false_type)
{
m_object.lazyAssign(expr);
internal::call_assignment_no_alias(m_object,expr,internal::assign_op<Scalar>());
Base::construct(m_object);
}

6
Eigen/src/Core/Replicate.h
View File

@ -53,8 +53,9 @@ struct traits<Replicate<MatrixType,RowFactor,ColFactor> >
IsRowMajor = MaxRowsAtCompileTime==1 && MaxColsAtCompileTime!=1 ? 1
: MaxColsAtCompileTime==1 && MaxRowsAtCompileTime!=1 ? 0
: (MatrixType::Flags & RowMajorBit) ? 1 : 0,
Flags = (_MatrixTypeNested::Flags & HereditaryBits & ~RowMajorBit) | (IsRowMajor ? RowMajorBit : 0),
CoeffReadCost = _MatrixTypeNested::CoeffReadCost
// FIXME enable DirectAccess with negative strides?
Flags = IsRowMajor ? RowMajorBit : 0
};
};
}
@ -68,6 +69,7 @@ template<typename MatrixType,int RowFactor,int ColFactor> class Replicate
typedef typename internal::dense_xpr_base<Replicate>::type Base;
EIGEN_DENSE_PUBLIC_INTERFACE(Replicate)
typedef typename internal::remove_all<MatrixType>::type NestedExpression;
template<typename OriginalMatrixType>
inline explicit Replicate(const OriginalMatrixType& a_matrix)

34
Eigen/src/Core/ReturnByValue.h
View File

@ -38,9 +38,10 @@ struct traits<ReturnByValue<Derived> >
* So internal::nested always gives the plain return matrix type.
*
* FIXME: I don't understand why we need this specialization: isn't this taken care of by the EvalBeforeNestingBit ??
* Answer: EvalBeforeNestingBit should be deprecated since we have the evaluators
*/
template<typename Derived,int n,typename PlainObject>
struct nested<ReturnByValue<Derived>, n, PlainObject>
struct nested_eval<ReturnByValue<Derived>, n, PlainObject>
{
typedef typename traits<Derived>::ReturnType type;
};
@ -73,6 +74,7 @@ template<typename Derived> class ReturnByValue
const Unusable& coeff(Index,Index) const { return *reinterpret_cast<const Unusable*>(this); }
Unusable& coeffRef(Index) { return *reinterpret_cast<Unusable*>(this); }
Unusable& coeffRef(Index,Index) { return *reinterpret_cast<Unusable*>(this); }
#undef Unusable
#endif
};
@ -84,6 +86,36 @@ Derived& DenseBase<Derived>::operator=(const ReturnByValue<OtherDerived>& other)
return derived();
}
namespace internal {
// Expression is evaluated in a temporary; default implementation of Assignment is bypassed so that
// when a ReturnByValue expression is assigned, the evaluator is not constructed.
// TODO: Finalize port to new regime; ReturnByValue should not exist in the expression world
template<typename Derived>
struct evaluator<ReturnByValue<Derived> >
: public evaluator<typename internal::traits<Derived>::ReturnType>::type
{
typedef ReturnByValue<Derived> XprType;
typedef typename internal::traits<Derived>::ReturnType PlainObject;
typedef typename evaluator<PlainObject>::type Base;
typedef evaluator type;
typedef evaluator nestedType;
evaluator(const XprType& xpr)
: m_result(xpr.rows(), xpr.cols())
{
::new (static_cast<Base*>(this)) Base(m_result);
xpr.evalTo(m_result);
}
protected:
PlainObject m_result;
};
} // end namespace internal
} // end namespace Eigen
#endif // EIGEN_RETURNBYVALUE_H

10
Eigen/src/Core/Reverse.h
View File

@ -44,14 +44,7 @@ struct traits<Reverse<MatrixType, Direction> >
ColsAtCompileTime = MatrixType::ColsAtCompileTime,
MaxRowsAtCompileTime = MatrixType::MaxRowsAtCompileTime,
MaxColsAtCompileTime = MatrixType::MaxColsAtCompileTime,
// let's enable LinearAccess only with vectorization because of the product overhead
LinearAccess = ( (Direction==BothDirections) && (int(_MatrixTypeNested::Flags)&PacketAccessBit) )
? LinearAccessBit : 0,
Flags = int(_MatrixTypeNested::Flags) & (HereditaryBits | LvalueBit | PacketAccessBit | LinearAccess),
CoeffReadCost = _MatrixTypeNested::CoeffReadCost
Flags = _MatrixTypeNested::Flags & (RowMajorBit | LvalueBit)
};
};
@ -74,6 +67,7 @@ template<typename MatrixType, int Direction> class Reverse
typedef typename internal::dense_xpr_base<Reverse>::type Base;
EIGEN_DENSE_PUBLIC_INTERFACE(Reverse)
typedef typename internal::remove_all<MatrixType>::type NestedExpression;
using Base::IsRowMajor;
// next line is necessary because otherwise const version of operator()

5
Eigen/src/Core/Select.h
View File

@ -43,10 +43,7 @@ struct traits<Select<ConditionMatrixType, ThenMatrixType, ElseMatrixType> >
ColsAtCompileTime = ConditionMatrixType::ColsAtCompileTime,
MaxRowsAtCompileTime = ConditionMatrixType::MaxRowsAtCompileTime,
MaxColsAtCompileTime = ConditionMatrixType::MaxColsAtCompileTime,
Flags = (unsigned int)ThenMatrixType::Flags & ElseMatrixType::Flags & HereditaryBits,
CoeffReadCost = traits<typename remove_all<ConditionMatrixNested>::type>::CoeffReadCost
+ EIGEN_SIZE_MAX(traits<typename remove_all<ThenMatrixNested>::type>::CoeffReadCost,
traits<typename remove_all<ElseMatrixNested>::type>::CoeffReadCost)
Flags = (unsigned int)ThenMatrixType::Flags & ElseMatrixType::Flags & RowMajorBit
};
};
}

169
Eigen/src/Core/SelfAdjointView.h
View File

@ -35,26 +35,22 @@ struct traits<SelfAdjointView<MatrixType, UpLo> > : traits<MatrixType>
typedef typename nested<MatrixType>::type MatrixTypeNested;
typedef typename remove_all<MatrixTypeNested>::type MatrixTypeNestedCleaned;
typedef MatrixType ExpressionType;
typedef typename MatrixType::PlainObject DenseMatrixType;
typedef typename MatrixType::PlainObject FullMatrixType;
enum {
Mode = UpLo | SelfAdjoint,
Flags = MatrixTypeNestedCleaned::Flags & (HereditaryBits)
& (~(PacketAccessBit | DirectAccessBit | LinearAccessBit)), // FIXME these flags should be preserved
CoeffReadCost = MatrixTypeNestedCleaned::CoeffReadCost
& (~(PacketAccessBit | DirectAccessBit | LinearAccessBit)) // FIXME these flags should be preserved
};
};
}
template <typename Lhs, int LhsMode, bool LhsIsVector,
typename Rhs, int RhsMode, bool RhsIsVector>
struct SelfadjointProductMatrix;
// FIXME could also be called SelfAdjointWrapper to be consistent with DiagonalWrapper ??
template<typename MatrixType, unsigned int UpLo> class SelfAdjointView
: public TriangularBase<SelfAdjointView<MatrixType, UpLo> >
template<typename _MatrixType, unsigned int UpLo> class SelfAdjointView
: public TriangularBase<SelfAdjointView<_MatrixType, UpLo> >
{
public:
typedef _MatrixType MatrixType;
typedef TriangularBase<SelfAdjointView> Base;
typedef typename internal::traits<SelfAdjointView>::MatrixTypeNested MatrixTypeNested;
typedef typename internal::traits<SelfAdjointView>::MatrixTypeNestedCleaned MatrixTypeNestedCleaned;
@ -65,7 +61,8 @@ template<typename MatrixType, unsigned int UpLo> class SelfAdjointView
typedef typename MatrixType::Index Index;
enum {
Mode = internal::traits<SelfAdjointView>::Mode
Mode = internal::traits<SelfAdjointView>::Mode,
Flags = internal::traits<SelfAdjointView>::Flags
};
typedef typename MatrixType::PlainObject PlainObject;
@ -111,26 +108,29 @@ template<typename MatrixType, unsigned int UpLo> class SelfAdjointView
EIGEN_DEVICE_FUNC
MatrixTypeNestedCleaned& nestedExpression() { return *const_cast<MatrixTypeNestedCleaned*>(&m_matrix); }
/** Efficient self-adjoint matrix times vector/matrix product */
/** Efficient triangular matrix times vector/matrix product */
template<typename OtherDerived>
EIGEN_DEVICE_FUNC
SelfadjointProductMatrix<MatrixType,Mode,false,OtherDerived,0,OtherDerived::IsVectorAtCompileTime>
const Product<SelfAdjointView,OtherDerived>
operator*(const MatrixBase<OtherDerived>& rhs) const
{
return SelfadjointProductMatrix
<MatrixType,Mode,false,OtherDerived,0,OtherDerived::IsVectorAtCompileTime>
(m_matrix, rhs.derived());
return Product<SelfAdjointView,OtherDerived>(*this, rhs.derived());
}
/** Efficient vector/matrix times self-adjoint matrix product */
/** Efficient vector/matrix times triangular matrix product */
template<typename OtherDerived> friend
EIGEN_DEVICE_FUNC
SelfadjointProductMatrix<OtherDerived,0,OtherDerived::IsVectorAtCompileTime,MatrixType,Mode,false>
const Product<OtherDerived,SelfAdjointView>
operator*(const MatrixBase<OtherDerived>& lhs, const SelfAdjointView& rhs)
{
return SelfadjointProductMatrix
<OtherDerived,0,OtherDerived::IsVectorAtCompileTime,MatrixType,Mode,false>
(lhs.derived(),rhs.m_matrix);
return Product<OtherDerived,SelfAdjointView>(lhs.derived(),rhs);
}
friend EIGEN_DEVICE_FUNC
const SelfAdjointView<const CwiseUnaryOp<internal::scalar_multiple_op<Scalar>,MatrixType>,UpLo>
operator*(const Scalar& s, const SelfAdjointView& mat)
{
return (s*mat.nestedExpression()).template selfadjointView<UpLo>();
}
/** Perform a symmetric rank 2 update of the selfadjoint matrix \c *this:
@ -194,96 +194,57 @@ template<typename MatrixType, unsigned int UpLo> class SelfAdjointView
namespace internal {
template<typename Derived1, typename Derived2, int UnrollCount, bool ClearOpposite>
struct triangular_assignment_selector<Derived1, Derived2, (SelfAdjoint|Upper), UnrollCount, ClearOpposite>
// TODO currently a selfadjoint expression has the form SelfAdjointView<.,.>
// in the future selfadjoint-ness should be defined by the expression traits
// such that Transpose<SelfAdjointView<.,.> > is valid. (currently TriangularBase::transpose() is overloaded to make it work)
template<typename MatrixType, unsigned int Mode>
struct evaluator_traits<SelfAdjointView<MatrixType,Mode> >
{
enum {
col = (UnrollCount-1) / Derived1::RowsAtCompileTime,
row = (UnrollCount-1) % Derived1::RowsAtCompileTime
};
typedef typename storage_kind_to_evaluator_kind<typename MatrixType::StorageKind>::Kind Kind;
typedef SelfAdjointShape Shape;
static const int AssumeAliasing = 0;
};
EIGEN_DEVICE_FUNC
static inline void run(Derived1 &dst, const Derived2 &src)
template<int UpLo, int SetOpposite, typename DstEvaluatorTypeT, typename SrcEvaluatorTypeT, typename Functor, int Version>
class triangular_dense_assignment_kernel<UpLo,SelfAdjoint,SetOpposite,DstEvaluatorTypeT,SrcEvaluatorTypeT,Functor,Version>
: public generic_dense_assignment_kernel<DstEvaluatorTypeT, SrcEvaluatorTypeT, Functor, Version>
{
protected:
typedef generic_dense_assignment_kernel<DstEvaluatorTypeT, SrcEvaluatorTypeT, Functor, Version> Base;
typedef typename Base::DstXprType DstXprType;
typedef typename Base::SrcXprType SrcXprType;
using Base::m_dst;
using Base::m_src;
using Base::m_functor;
public:
typedef typename Base::DstEvaluatorType DstEvaluatorType;
typedef typename Base::SrcEvaluatorType SrcEvaluatorType;
typedef typename Base::Scalar Scalar;
typedef typename Base::Index Index;
typedef typename Base::AssignmentTraits AssignmentTraits;
triangular_dense_assignment_kernel(DstEvaluatorType &dst, const SrcEvaluatorType &src, const Functor &func, DstXprType& dstExpr)
: Base(dst, src, func, dstExpr)
{}
void assignCoeff(Index row, Index col)
{
triangular_assignment_selector<Derived1, Derived2, (SelfAdjoint|Upper), UnrollCount-1, ClearOpposite>::run(dst, src);
if(row == col)
dst.coeffRef(row, col) = numext::real(src.coeff(row, col));
else if(row < col)
dst.coeffRef(col, row) = numext::conj(dst.coeffRef(row, col) = src.coeff(row, col));
eigen_internal_assert(row!=col);
Scalar tmp = m_src.coeff(row,col);
m_functor.assignCoeff(m_dst.coeffRef(row,col), tmp);
m_functor.assignCoeff(m_dst.coeffRef(col,row), numext::conj(tmp));
}
};
template<typename Derived1, typename Derived2, bool ClearOpposite>
struct triangular_assignment_selector<Derived1, Derived2, SelfAdjoint|Upper, 0, ClearOpposite>
{
EIGEN_DEVICE_FUNC
static inline void run(Derived1 &, const Derived2 &) {}
};
template<typename Derived1, typename Derived2, int UnrollCount, bool ClearOpposite>
struct triangular_assignment_selector<Derived1, Derived2, (SelfAdjoint|Lower), UnrollCount, ClearOpposite>
{
enum {
col = (UnrollCount-1) / Derived1::RowsAtCompileTime,
row = (UnrollCount-1) % Derived1::RowsAtCompileTime
};
EIGEN_DEVICE_FUNC
static inline void run(Derived1 &dst, const Derived2 &src)
void assignDiagonalCoeff(Index id)
{
triangular_assignment_selector<Derived1, Derived2, (SelfAdjoint|Lower), UnrollCount-1, ClearOpposite>::run(dst, src);
if(row == col)
dst.coeffRef(row, col) = numext::real(src.coeff(row, col));
else if(row > col)
dst.coeffRef(col, row) = numext::conj(dst.coeffRef(row, col) = src.coeff(row, col));
}
};
template<typename Derived1, typename Derived2, bool ClearOpposite>
struct triangular_assignment_selector<Derived1, Derived2, SelfAdjoint|Lower, 0, ClearOpposite>
{
EIGEN_DEVICE_FUNC
static inline void run(Derived1 &, const Derived2 &) {}
};
template<typename Derived1, typename Derived2, bool ClearOpposite>
struct triangular_assignment_selector<Derived1, Derived2, SelfAdjoint|Upper, Dynamic, ClearOpposite>
{
typedef typename Derived1::Index Index;
EIGEN_DEVICE_FUNC
static inline void run(Derived1 &dst, const Derived2 &src)
{
for(Index j = 0; j < dst.cols(); ++j)
{
for(Index i = 0; i < j; ++i)
{
dst.copyCoeff(i, j, src);
dst.coeffRef(j,i) = numext::conj(dst.coeff(i,j));
}
dst.copyCoeff(j, j, src);
}
}
};
template<typename Derived1, typename Derived2, bool ClearOpposite>
struct triangular_assignment_selector<Derived1, Derived2, SelfAdjoint|Lower, Dynamic, ClearOpposite>
{
EIGEN_DEVICE_FUNC
static inline void run(Derived1 &dst, const Derived2 &src)
{
typedef typename Derived1::Index Index;
for(Index i = 0; i < dst.rows(); ++i)
{
for(Index j = 0; j < i; ++j)
{
dst.copyCoeff(i, j, src);
dst.coeffRef(j,i) = numext::conj(dst.coeff(i,j));
}
dst.copyCoeff(i, i, src);
}
Base::assignCoeff(id,id);
}
void assignOppositeCoeff(Index, Index)
{ eigen_internal_assert(false && "should never be called"); }
};
} // end namespace internal

185
Eigen/src/Core/SelfCwiseBinaryOp.h
View File

@ -12,179 +12,11 @@
namespace Eigen {
/** \class SelfCwiseBinaryOp
* \ingroup Core_Module
*
* \internal
*
* \brief Internal helper class for optimizing operators like +=, -=
*
* This is a pseudo expression class re-implementing the copyCoeff/copyPacket
* method to directly performs a +=/-= operations in an optimal way. In particular,
* this allows to make sure that the input/output data are loaded only once using
* aligned packet loads.
*
* \sa class SwapWrapper for a similar trick.
*/
namespace internal {
template<typename BinaryOp, typename Lhs, typename Rhs>
struct traits<SelfCwiseBinaryOp<BinaryOp,Lhs,Rhs> >
: traits<CwiseBinaryOp<BinaryOp,Lhs,Rhs> >
{
enum {
// Note that it is still a good idea to preserve the DirectAccessBit
// so that assign can correctly align the data.
Flags = traits<CwiseBinaryOp<BinaryOp,Lhs,Rhs> >::Flags | (Lhs::Flags&AlignedBit) | (Lhs::Flags&DirectAccessBit) | (Lhs::Flags&LvalueBit),
OuterStrideAtCompileTime = Lhs::OuterStrideAtCompileTime,
InnerStrideAtCompileTime = Lhs::InnerStrideAtCompileTime
};
};
}
template<typename BinaryOp, typename Lhs, typename Rhs> class SelfCwiseBinaryOp
: public internal::dense_xpr_base< SelfCwiseBinaryOp<BinaryOp, Lhs, Rhs> >::type
{
public:
typedef typename internal::dense_xpr_base<SelfCwiseBinaryOp>::type Base;
EIGEN_DENSE_PUBLIC_INTERFACE(SelfCwiseBinaryOp)
typedef typename internal::packet_traits<Scalar>::type Packet;
EIGEN_DEVICE_FUNC
inline SelfCwiseBinaryOp(Lhs& xpr, const BinaryOp& func = BinaryOp()) : m_matrix(xpr), m_functor(func) {}
EIGEN_DEVICE_FUNC inline Index rows() const { return m_matrix.rows(); }
EIGEN_DEVICE_FUNC inline Index cols() const { return m_matrix.cols(); }
EIGEN_DEVICE_FUNC inline Index outerStride() const { return m_matrix.outerStride(); }
EIGEN_DEVICE_FUNC inline Index innerStride() const { return m_matrix.innerStride(); }
EIGEN_DEVICE_FUNC inline const Scalar* data() const { return m_matrix.data(); }
// note that this function is needed by assign to correctly align loads/stores
// TODO make Assign use .data()
EIGEN_DEVICE_FUNC
inline Scalar& coeffRef(Index row, Index col)
{
EIGEN_STATIC_ASSERT_LVALUE(Lhs)
return m_matrix.const_cast_derived().coeffRef(row, col);
}
EIGEN_DEVICE_FUNC
inline const Scalar& coeffRef(Index row, Index col) const
{
return m_matrix.coeffRef(row, col);
}
// note that this function is needed by assign to correctly align loads/stores
// TODO make Assign use .data()
EIGEN_DEVICE_FUNC
inline Scalar& coeffRef(Index index)
{
EIGEN_STATIC_ASSERT_LVALUE(Lhs)
return m_matrix.const_cast_derived().coeffRef(index);
}
EIGEN_DEVICE_FUNC
inline const Scalar& coeffRef(Index index) const
{
return m_matrix.const_cast_derived().coeffRef(index);
}
template<typename OtherDerived>
EIGEN_DEVICE_FUNC
void copyCoeff(Index row, Index col, const DenseBase<OtherDerived>& other)
{
OtherDerived& _other = other.const_cast_derived();
eigen_internal_assert(row >= 0 && row < rows()
&& col >= 0 && col < cols());
Scalar& tmp = m_matrix.coeffRef(row,col);
tmp = m_functor(tmp, _other.coeff(row,col));
}
template<typename OtherDerived>
EIGEN_DEVICE_FUNC
void copyCoeff(Index index, const DenseBase<OtherDerived>& other)
{
OtherDerived& _other = other.const_cast_derived();
eigen_internal_assert(index >= 0 && index < m_matrix.size());
Scalar& tmp = m_matrix.coeffRef(index);
tmp = m_functor(tmp, _other.coeff(index));
}
template<typename OtherDerived, int StoreMode, int LoadMode>
void copyPacket(Index row, Index col, const DenseBase<OtherDerived>& other)
{
OtherDerived& _other = other.const_cast_derived();
eigen_internal_assert(row >= 0 && row < rows()
&& col >= 0 && col < cols());
m_matrix.template writePacket<StoreMode>(row, col,
m_functor.packetOp(m_matrix.template packet<StoreMode>(row, col),_other.template packet<LoadMode>(row, col)) );
}
template<typename OtherDerived, int StoreMode, int LoadMode>
void copyPacket(Index index, const DenseBase<OtherDerived>& other)
{
OtherDerived& _other = other.const_cast_derived();
eigen_internal_assert(index >= 0 && index < m_matrix.size());
m_matrix.template writePacket<StoreMode>(index,
m_functor.packetOp(m_matrix.template packet<StoreMode>(index),_other.template packet<LoadMode>(index)) );
}
// reimplement lazyAssign to handle complex *= real
// see CwiseBinaryOp ctor for details
template<typename RhsDerived>
EIGEN_DEVICE_FUNC
EIGEN_STRONG_INLINE SelfCwiseBinaryOp& lazyAssign(const DenseBase<RhsDerived>& rhs)
{
EIGEN_STATIC_ASSERT_SAME_MATRIX_SIZE(Lhs,RhsDerived)
EIGEN_CHECK_BINARY_COMPATIBILIY(BinaryOp,typename Lhs::Scalar,typename RhsDerived::Scalar);
#ifdef EIGEN_DEBUG_ASSIGN
internal::assign_traits<SelfCwiseBinaryOp, RhsDerived>::debug();
#endif
eigen_assert(rows() == rhs.rows() && cols() == rhs.cols());
internal::assign_impl<SelfCwiseBinaryOp, RhsDerived>::run(*this,rhs.derived());
#ifndef EIGEN_NO_DEBUG
this->checkTransposeAliasing(rhs.derived());
#endif
return *this;
}
// overloaded to honor evaluation of special matrices
// maybe another solution would be to not use SelfCwiseBinaryOp
// at first...
EIGEN_DEVICE_FUNC
SelfCwiseBinaryOp& operator=(const Rhs& _rhs)
{
typename internal::nested<Rhs>::type rhs(_rhs);
return Base::operator=(rhs);
}
EIGEN_DEVICE_FUNC
Lhs& expression() const
{
return m_matrix;
}
EIGEN_DEVICE_FUNC
const BinaryOp& functor() const
{
return m_functor;
}
protected:
Lhs& m_matrix;
const BinaryOp& m_functor;
private:
SelfCwiseBinaryOp& operator=(const SelfCwiseBinaryOp&);
};
template<typename Derived>
inline Derived& DenseBase<Derived>::operator*=(const Scalar& other)
{
typedef typename Derived::PlainObject PlainObject;
SelfCwiseBinaryOp<internal::scalar_product_op<Scalar>, Derived, typename PlainObject::ConstantReturnType> tmp(derived());
tmp = PlainObject::Constant(rows(),cols(),other);
internal::call_assignment(this->derived(), PlainObject::Constant(rows(),cols(),other), internal::mul_assign_op<Scalar>());
return derived();
}
@ -192,8 +24,7 @@ template<typename Derived>
inline Derived& ArrayBase<Derived>::operator+=(const Scalar& other)
{
typedef typename Derived::PlainObject PlainObject;
SelfCwiseBinaryOp<internal::scalar_sum_op<Scalar>, Derived, typename PlainObject::ConstantReturnType> tmp(derived());
tmp = PlainObject::Constant(rows(),cols(),other);
internal::call_assignment(this->derived(), PlainObject::Constant(rows(),cols(),other), internal::add_assign_op<Scalar>());
return derived();
}
@ -201,23 +32,15 @@ template<typename Derived>
inline Derived& ArrayBase<Derived>::operator-=(const Scalar& other)
{
typedef typename Derived::PlainObject PlainObject;
SelfCwiseBinaryOp<internal::scalar_difference_op<Scalar>, Derived, typename PlainObject::ConstantReturnType> tmp(derived());
tmp = PlainObject::Constant(rows(),cols(),other);
internal::call_assignment(this->derived(), PlainObject::Constant(rows(),cols(),other), internal::sub_assign_op<Scalar>());
return derived();
}
template<typename Derived>
inline Derived& DenseBase<Derived>::operator/=(const Scalar& other)
{
typedef typename internal::conditional<NumTraits<Scalar>::IsInteger,
internal::scalar_quotient_op<Scalar>,
internal::scalar_product_op<Scalar> >::type BinOp;
typedef typename Derived::PlainObject PlainObject;
SelfCwiseBinaryOp<BinOp, Derived, typename PlainObject::ConstantReturnType> tmp(derived());
Scalar actual_other;
if(NumTraits<Scalar>::IsInteger) actual_other = other;
else actual_other = Scalar(1)/other;
tmp = PlainObject::Constant(rows(),cols(), actual_other);
internal::call_assignment(this->derived(), PlainObject::Constant(rows(),cols(),other), internal::div_assign_op<Scalar>());
return derived();
}

152
Eigen/src/Core/Solve.h Normal file
View File

@ -0,0 +1,152 @@
// This file is part of Eigen, a lightweight C++ template library
// for linear algebra.
//
// Copyright (C) 2014 Gael Guennebaud <gael.guennebaud@inria.fr>
//
// This Source Code Form is subject to the terms of the Mozilla
// Public License v. 2.0. If a copy of the MPL was not distributed
// with this file, You can obtain one at http://mozilla.org/MPL/2.0/.
#ifndef EIGEN_SOLVE_H
#define EIGEN_SOLVE_H
namespace Eigen {
template<typename Decomposition, typename RhsType, typename StorageKind> class SolveImpl;
/** \class Solve
* \ingroup Core_Module
*
* \brief Pseudo expression representing a solving operation
*
* \tparam Decomposition the type of the matrix or decomposion object
* \tparam Rhstype the type of the right-hand side
*
* This class represents an expression of A.solve(B)
* and most of the time this is the only way it is used.
*
*/
namespace internal {
// this solve_traits class permits to determine the evaluation type with respect to storage kind (Dense vs Sparse)
template<typename Decomposition, typename RhsType,typename StorageKind> struct solve_traits;
template<typename Decomposition, typename RhsType>
struct solve_traits<Decomposition,RhsType,Dense>
{
typedef typename Decomposition::MatrixType MatrixType;
typedef Matrix<typename RhsType::Scalar,
MatrixType::ColsAtCompileTime,
RhsType::ColsAtCompileTime,
RhsType::PlainObject::Options,
MatrixType::MaxColsAtCompileTime,
RhsType::MaxColsAtCompileTime> PlainObject;
};
template<typename Decomposition, typename RhsType>
struct traits<Solve<Decomposition, RhsType> >
: traits<typename solve_traits<Decomposition,RhsType,typename internal::traits<RhsType>::StorageKind>::PlainObject>
{
typedef typename solve_traits<Decomposition,RhsType,typename internal::traits<RhsType>::StorageKind>::PlainObject PlainObject;
typedef traits<PlainObject> BaseTraits;
enum {
Flags = BaseTraits::Flags & RowMajorBit,
CoeffReadCost = Dynamic
};
};
}
template<typename Decomposition, typename RhsType>
class Solve : public SolveImpl<Decomposition,RhsType,typename internal::traits<RhsType>::StorageKind>
{
public:
typedef typename RhsType::Index Index;
typedef typename internal::traits<Solve>::PlainObject PlainObject;
Solve(const Decomposition &dec, const RhsType &rhs)
: m_dec(dec), m_rhs(rhs)
{}
EIGEN_DEVICE_FUNC Index rows() const { return m_dec.cols(); }
EIGEN_DEVICE_FUNC Index cols() const { return m_rhs.cols(); }
EIGEN_DEVICE_FUNC const Decomposition& dec() const { return m_dec; }
EIGEN_DEVICE_FUNC const RhsType& rhs() const { return m_rhs; }
protected:
const Decomposition &m_dec;
const RhsType &m_rhs;
};
// Specialization of the Solve expression for dense results
template<typename Decomposition, typename RhsType>
class SolveImpl<Decomposition,RhsType,Dense>
: public MatrixBase<Solve<Decomposition,RhsType> >
{
typedef Solve<Decomposition,RhsType> Derived;
public:
typedef MatrixBase<Solve<Decomposition,RhsType> > Base;
EIGEN_DENSE_PUBLIC_INTERFACE(Derived)
private:
Scalar coeff(Index row, Index col) const;
Scalar coeff(Index i) const;
};
// Generic API dispatcher
template<typename Decomposition, typename RhsType, typename StorageKind>
class SolveImpl : public internal::generic_xpr_base<Solve<Decomposition,RhsType>, MatrixXpr, StorageKind>::type
{
public:
typedef typename internal::generic_xpr_base<Solve<Decomposition,RhsType>, MatrixXpr, StorageKind>::type Base;
};
namespace internal {
// Evaluator of Solve -> eval into a temporary
template<typename Decomposition, typename RhsType>
struct evaluator<Solve<Decomposition,RhsType> >
: public evaluator<typename Solve<Decomposition,RhsType>::PlainObject>::type
{
typedef Solve<Decomposition,RhsType> SolveType;
typedef typename SolveType::PlainObject PlainObject;
typedef typename evaluator<PlainObject>::type Base;
typedef evaluator type;
typedef evaluator nestedType;
evaluator(const SolveType& solve)
: m_result(solve.rows(), solve.cols())
{
::new (static_cast<Base*>(this)) Base(m_result);
solve.dec()._solve_impl(solve.rhs(), m_result);
}
protected:
PlainObject m_result;
};
// Specialization for "dst = dec.solve(rhs)"
// NOTE we need to specialize it for Dense2Dense to avoid ambiguous specialization error and a Sparse2Sparse specialization must exist somewhere
template<typename DstXprType, typename DecType, typename RhsType, typename Scalar>
struct Assignment<DstXprType, Solve<DecType,RhsType>, internal::assign_op<Scalar>, Dense2Dense, Scalar>
{
typedef Solve<DecType,RhsType> SrcXprType;
static void run(DstXprType &dst, const SrcXprType &src, const internal::assign_op<Scalar> &)
{
// FIXME shall we resize dst here?
src.dec()._solve_impl(src.rhs(), dst);
}
};
} // end namepsace internal
} // end namespace Eigen
#endif // EIGEN_SOLVE_H

10
Eigen/src/Core/SolveTriangular.h
View File

@ -171,10 +171,10 @@ struct triangular_solver_selector<Lhs,Rhs,OnTheRight,Mode,CompleteUnrolling,1> {
*/
template<typename MatrixType, unsigned int Mode>
template<int Side, typename OtherDerived>
void TriangularView<MatrixType,Mode>::solveInPlace(const MatrixBase<OtherDerived>& _other) const
void TriangularViewImpl<MatrixType,Mode,Dense>::solveInPlace(const MatrixBase<OtherDerived>& _other) const
{
OtherDerived& other = _other.const_cast_derived();
eigen_assert( cols() == rows() && ((Side==OnTheLeft && cols() == other.rows()) || (Side==OnTheRight && cols() == other.cols())) );
eigen_assert( derived().cols() == derived().rows() && ((Side==OnTheLeft && derived().cols() == other.rows()) || (Side==OnTheRight && derived().cols() == other.cols())) );
eigen_assert((!(Mode & ZeroDiag)) && bool(Mode & (Upper|Lower)));
enum { copy = internal::traits<OtherDerived>::Flags & RowMajorBit && OtherDerived::IsVectorAtCompileTime };
@ -183,7 +183,7 @@ void TriangularView<MatrixType,Mode>::solveInPlace(const MatrixBase<OtherDerived
OtherCopy otherCopy(other);
internal::triangular_solver_selector<MatrixType, typename internal::remove_reference<OtherCopy>::type,
Side, Mode>::run(nestedExpression(), otherCopy);
Side, Mode>::run(derived().nestedExpression(), otherCopy);
if (copy)
other = otherCopy;
@ -213,9 +213,9 @@ void TriangularView<MatrixType,Mode>::solveInPlace(const MatrixBase<OtherDerived
template<typename Derived, unsigned int Mode>
template<int Side, typename Other>
const internal::triangular_solve_retval<Side,TriangularView<Derived,Mode>,Other>
TriangularView<Derived,Mode>::solve(const MatrixBase<Other>& other) const
TriangularViewImpl<Derived,Mode,Dense>::solve(const MatrixBase<Other>& other) const
{
return internal::triangular_solve_retval<Side,TriangularView,Other>(*this, other.derived());
return internal::triangular_solve_retval<Side,TriangularViewType,Other>(derived(), other.derived());
}
namespace internal {

22
Eigen/src/Core/StableNorm.h
View File

@ -20,7 +20,7 @@ inline void stable_norm_kernel(const ExpressionType& bl, Scalar& ssq, Scalar& sc
using std::max;
Scalar maxCoeff = bl.cwiseAbs().maxCoeff();
if (maxCoeff>scale)
if(maxCoeff>scale)
{
ssq = ssq * numext::abs2(scale/maxCoeff);
Scalar tmp = Scalar(1)/maxCoeff;
@ -29,12 +29,21 @@ inline void stable_norm_kernel(const ExpressionType& bl, Scalar& ssq, Scalar& sc
invScale = NumTraits<Scalar>::highest();
scale = Scalar(1)/invScale;
}
else if(maxCoeff>NumTraits<Scalar>::highest()) // we got a INF
{
invScale = Scalar(1);
scale = maxCoeff;
}
else
{
scale = maxCoeff;
invScale = tmp;
}
}
else if(maxCoeff!=maxCoeff) // we got a NaN
{
scale = maxCoeff;
}
// TODO if the maxCoeff is much much smaller than the current scale,
// then we can neglect this sub vector
@ -55,7 +64,7 @@ blueNorm_impl(const EigenBase<Derived>& _vec)
using std::abs;
const Derived& vec(_vec.derived());
static bool initialized = false;
static RealScalar b1, b2, s1m, s2m, overfl, rbig, relerr;
static RealScalar b1, b2, s1m, s2m, rbig, relerr;
if(!initialized)
{
int ibeta, it, iemin, iemax, iexp;
@ -84,7 +93,6 @@ blueNorm_impl(const EigenBase<Derived>& _vec)
iexp = - ((iemax+it)/2);
s2m = RealScalar(pow(RealScalar(ibeta),RealScalar(iexp))); // scaling factor for upper range
overfl = rbig*s2m; // overflow boundary for abig
eps = RealScalar(pow(double(ibeta), 1-it));
relerr = sqrt(eps); // tolerance for neglecting asml
initialized = true;
@ -101,13 +109,13 @@ blueNorm_impl(const EigenBase<Derived>& _vec)
else if(ax < b1) asml += numext::abs2(ax*s1m);
else amed += numext::abs2(ax);
}
if(amed!=amed)
return amed; // we got a NaN
if(abig > RealScalar(0))
{
abig = sqrt(abig);
if(abig > overfl)
{
return rbig;
}
if(abig > rbig) // overflow, or *this contains INF values
return abig; // return INF
if(amed > RealScalar(0))
{
abig = abig/s2m;

4
Eigen/src/Core/Stride.h
View File

@ -86,7 +86,7 @@ class Stride
/** \brief Convenience specialization of Stride to specify only an inner stride
* See class Map for some examples */
template<int Value = Dynamic>
template<int Value>
class InnerStride : public Stride<0, Value>
{
typedef Stride<0, Value> Base;
@ -98,7 +98,7 @@ class InnerStride : public Stride<0, Value>
/** \brief Convenience specialization of Stride to specify only an outer stride
* See class Map for some examples */
template<int Value = Dynamic>
template<int Value>
class OuterStride : public Stride<Value, 0>
{
typedef Stride<Value, 0> Base;

161
Eigen/src/Core/Swap.h
View File

@ -12,129 +12,54 @@
namespace Eigen {
/** \class SwapWrapper
* \ingroup Core_Module
*
* \internal
*
* \brief Internal helper class for swapping two expressions
*/
namespace internal {
template<typename ExpressionType>
struct traits<SwapWrapper<ExpressionType> > : traits<ExpressionType> {};
}
template<typename ExpressionType> class SwapWrapper
: public internal::dense_xpr_base<SwapWrapper<ExpressionType> >::type
// Overload default assignPacket behavior for swapping them
template<typename DstEvaluatorTypeT, typename SrcEvaluatorTypeT>
class generic_dense_assignment_kernel<DstEvaluatorTypeT, SrcEvaluatorTypeT, swap_assign_op<typename DstEvaluatorTypeT::Scalar>, Specialized>
: public generic_dense_assignment_kernel<DstEvaluatorTypeT, SrcEvaluatorTypeT, swap_assign_op<typename DstEvaluatorTypeT::Scalar>, BuiltIn>
{
public:
typedef typename internal::dense_xpr_base<SwapWrapper>::type Base;
EIGEN_DENSE_PUBLIC_INTERFACE(SwapWrapper)
typedef typename internal::packet_traits<Scalar>::type Packet;
EIGEN_DEVICE_FUNC
inline SwapWrapper(ExpressionType& xpr) : m_expression(xpr) {}
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(); }
typedef typename internal::conditional<
internal::is_lvalue<ExpressionType>::value,
Scalar,
const Scalar
>::type ScalarWithConstIfNotLvalue;
EIGEN_DEVICE_FUNC
inline ScalarWithConstIfNotLvalue* data() { return m_expression.data(); }
EIGEN_DEVICE_FUNC
inline const Scalar* data() const { return m_expression.data(); }
EIGEN_DEVICE_FUNC
inline Scalar& coeffRef(Index rowId, Index colId)
{
return m_expression.const_cast_derived().coeffRef(rowId, colId);
}
EIGEN_DEVICE_FUNC
inline Scalar& coeffRef(Index index)
{
return m_expression.const_cast_derived().coeffRef(index);
}
EIGEN_DEVICE_FUNC
inline Scalar& coeffRef(Index rowId, Index colId) const
{
return m_expression.coeffRef(rowId, colId);
}
EIGEN_DEVICE_FUNC
inline Scalar& coeffRef(Index index) const
{
return m_expression.coeffRef(index);
}
template<typename OtherDerived>
EIGEN_DEVICE_FUNC
void copyCoeff(Index rowId, Index colId, const DenseBase<OtherDerived>& other)
{
OtherDerived& _other = other.const_cast_derived();
eigen_internal_assert(rowId >= 0 && rowId < rows()
&& colId >= 0 && colId < cols());
Scalar tmp = m_expression.coeff(rowId, colId);
m_expression.coeffRef(rowId, colId) = _other.coeff(rowId, colId);
_other.coeffRef(rowId, colId) = tmp;
}
template<typename OtherDerived>
EIGEN_DEVICE_FUNC
void copyCoeff(Index index, const DenseBase<OtherDerived>& other)
{
OtherDerived& _other = other.const_cast_derived();
eigen_internal_assert(index >= 0 && index < m_expression.size());
Scalar tmp = m_expression.coeff(index);
m_expression.coeffRef(index) = _other.coeff(index);
_other.coeffRef(index) = tmp;
}
template<typename OtherDerived, int StoreMode, int LoadMode>
void copyPacket(Index rowId, Index colId, const DenseBase<OtherDerived>& other)
{
OtherDerived& _other = other.const_cast_derived();
eigen_internal_assert(rowId >= 0 && rowId < rows()
&& colId >= 0 && colId < cols());
Packet tmp = m_expression.template packet<StoreMode>(rowId, colId);
m_expression.template writePacket<StoreMode>(rowId, colId,
_other.template packet<LoadMode>(rowId, colId)
);
_other.template writePacket<LoadMode>(rowId, colId, tmp);
}
template<typename OtherDerived, int StoreMode, int LoadMode>
void copyPacket(Index index, const DenseBase<OtherDerived>& other)
{
OtherDerived& _other = other.const_cast_derived();
eigen_internal_assert(index >= 0 && index < m_expression.size());
Packet tmp = m_expression.template packet<StoreMode>(index);
m_expression.template writePacket<StoreMode>(index,
_other.template packet<LoadMode>(index)
);
_other.template writePacket<LoadMode>(index, tmp);
}
EIGEN_DEVICE_FUNC
ExpressionType& expression() const { return m_expression; }
protected:
ExpressionType& m_expression;
protected:
typedef generic_dense_assignment_kernel<DstEvaluatorTypeT, SrcEvaluatorTypeT, swap_assign_op<typename DstEvaluatorTypeT::Scalar>, BuiltIn> Base;
typedef typename DstEvaluatorTypeT::PacketScalar PacketScalar;
using Base::m_dst;
using Base::m_src;
using Base::m_functor;
public:
typedef typename Base::Scalar Scalar;
typedef typename Base::Index Index;
typedef typename Base::DstXprType DstXprType;
typedef swap_assign_op<Scalar> Functor;
generic_dense_assignment_kernel(DstEvaluatorTypeT &dst, const SrcEvaluatorTypeT &src, const Functor &func, DstXprType& dstExpr)
: Base(dst, src, func, dstExpr)
{}
template<int StoreMode, int LoadMode>
void assignPacket(Index row, Index col)
{
m_functor.template swapPacket<StoreMode,LoadMode,PacketScalar>(&m_dst.coeffRef(row,col), &const_cast<SrcEvaluatorTypeT&>(m_src).coeffRef(row,col));
}
template<int StoreMode, int LoadMode>
void assignPacket(Index index)
{
m_functor.template swapPacket<StoreMode,LoadMode,PacketScalar>(&m_dst.coeffRef(index), &const_cast<SrcEvaluatorTypeT&>(m_src).coeffRef(index));
}
// TODO find a simple way not to have to copy/paste this function from generic_dense_assignment_kernel, by simple I mean no CRTP (Gael)
template<int StoreMode, int LoadMode>
void assignPacketByOuterInner(Index outer, Index inner)
{
Index row = Base::rowIndexByOuterInner(outer, inner);
Index col = Base::colIndexByOuterInner(outer, inner);
assignPacket<StoreMode,LoadMode>(row, col);
}
};
} // namespace internal
} // end namespace Eigen
#endif // EIGEN_SWAP_H

84
Eigen/src/Core/Transpose.h
View File

@ -2,7 +2,7 @@
// for linear algebra.
//
// Copyright (C) 2006-2008 Benoit Jacob <jacob.benoit.1@gmail.com>
// Copyright (C) 2009-2010 Gael Guennebaud <gael.guennebaud@inria.fr>
// Copyright (C) 2009-2014 Gael Guennebaud <gael.guennebaud@inria.fr>
//
// This Source Code Form is subject to the terms of the Mozilla
// Public License v. 2.0. If a copy of the MPL was not distributed
@ -29,9 +29,10 @@ namespace Eigen {
namespace internal {
template<typename MatrixType>
struct traits<Transpose<MatrixType> > : traits<MatrixType>
struct traits<Transpose<MatrixType> >
{
typedef typename MatrixType::Scalar Scalar;
typedef typename traits<MatrixType>::Scalar Scalar;
typedef typename traits<MatrixType>::Index Index;
typedef typename nested<MatrixType>::type MatrixTypeNested;
typedef typename remove_reference<MatrixTypeNested>::type MatrixTypeNestedPlain;
typedef typename traits<MatrixType>::StorageKind StorageKind;
@ -45,7 +46,6 @@ struct traits<Transpose<MatrixType> > : traits<MatrixType>
Flags0 = MatrixTypeNestedPlain::Flags & ~(LvalueBit | NestByRefBit),
Flags1 = Flags0 | FlagsLvalueBit,
Flags = Flags1 ^ RowMajorBit,
CoeffReadCost = MatrixTypeNestedPlain::CoeffReadCost,
InnerStrideAtCompileTime = inner_stride_at_compile_time<MatrixType>::ret,
OuterStrideAtCompileTime = outer_stride_at_compile_time<MatrixType>::ret
};
@ -61,6 +61,7 @@ template<typename MatrixType> class Transpose
typedef typename TransposeImpl<MatrixType,typename internal::traits<MatrixType>::StorageKind>::Base Base;
EIGEN_GENERIC_PUBLIC_INTERFACE(Transpose)
typedef typename internal::remove_all<MatrixType>::type NestedExpression;
EIGEN_DEVICE_FUNC
inline Transpose(MatrixType& a_matrix) : m_matrix(a_matrix) {}
@ -100,12 +101,22 @@ struct TransposeImpl_base<MatrixType, false>
} // end namespace internal
// Generic API dispatcher
template<typename XprType, typename StorageKind>
class TransposeImpl
: public internal::generic_xpr_base<Transpose<XprType> >::type
{
public:
typedef typename internal::generic_xpr_base<Transpose<XprType> >::type Base;
};
template<typename MatrixType> class TransposeImpl<MatrixType,Dense>
: public internal::TransposeImpl_base<MatrixType>::type
{
public:
typedef typename internal::TransposeImpl_base<MatrixType>::type Base;
using Base::coeffRef;
EIGEN_DENSE_PUBLIC_INTERFACE(Transpose<MatrixType>)
EIGEN_INHERIT_ASSIGNMENT_OPERATORS(TransposeImpl)
@ -121,20 +132,7 @@ template<typename MatrixType> class TransposeImpl<MatrixType,Dense>
inline ScalarWithConstIfNotLvalue* data() { return derived().nestedExpression().data(); }
inline const Scalar* data() const { return derived().nestedExpression().data(); }
EIGEN_DEVICE_FUNC
inline ScalarWithConstIfNotLvalue& coeffRef(Index rowId, Index colId)
{
EIGEN_STATIC_ASSERT_LVALUE(MatrixType)
return derived().nestedExpression().const_cast_derived().coeffRef(colId, rowId);
}
EIGEN_DEVICE_FUNC
inline ScalarWithConstIfNotLvalue& coeffRef(Index index)
{
EIGEN_STATIC_ASSERT_LVALUE(MatrixType)
return derived().nestedExpression().const_cast_derived().coeffRef(index);
}
// FIXME: shall we keep the const version of coeffRef?
EIGEN_DEVICE_FUNC
inline const Scalar& coeffRef(Index rowId, Index colId) const
{
@ -146,42 +144,6 @@ template<typename MatrixType> class TransposeImpl<MatrixType,Dense>
{
return derived().nestedExpression().coeffRef(index);
}
EIGEN_DEVICE_FUNC
inline CoeffReturnType coeff(Index rowId, Index colId) const
{
return derived().nestedExpression().coeff(colId, rowId);
}
EIGEN_DEVICE_FUNC
inline CoeffReturnType coeff(Index index) const
{
return derived().nestedExpression().coeff(index);
}
template<int LoadMode>
inline const PacketScalar packet(Index rowId, Index colId) const
{
return derived().nestedExpression().template packet<LoadMode>(colId, rowId);
}
template<int LoadMode>
inline void writePacket(Index rowId, Index colId, const PacketScalar& x)
{
derived().nestedExpression().const_cast_derived().template writePacket<LoadMode>(colId, rowId, x);
}
template<int LoadMode>
inline const PacketScalar packet(Index index) const
{
return derived().nestedExpression().template packet<LoadMode>(index);
}
template<int LoadMode>
inline void writePacket(Index index, const PacketScalar& x)
{
derived().nestedExpression().const_cast_derived().template writePacket<LoadMode>(index, x);
}
};
/** \returns an expression of the transpose of *this.
@ -413,15 +375,15 @@ struct checkTransposeAliasing_impl<Derived, OtherDerived, false>
}
};
template<typename Dst, typename Src>
void check_for_aliasing(const Dst &dst, const Src &src)
{
internal::checkTransposeAliasing_impl<Dst, Src>::run(dst, src);
}
} // end namespace internal
template<typename Derived>
template<typename OtherDerived>
void DenseBase<Derived>::checkTransposeAliasing(const OtherDerived& other) const
{
internal::checkTransposeAliasing_impl<Derived, OtherDerived>::run(derived(), other);
}
#endif
#endif // EIGEN_NO_DEBUG
} // end namespace Eigen

939
Eigen/src/Core/TriangularMatrix.h
View File

File diff suppressed because it is too large Load Diff

12
Eigen/src/Core/VectorwiseOp.h
View File

@ -48,19 +48,9 @@ struct traits<PartialReduxExpr<MatrixType, MemberOp, Direction> >
ColsAtCompileTime = Direction==Horizontal ? 1 : MatrixType::ColsAtCompileTime,
MaxRowsAtCompileTime = Direction==Vertical ? 1 : MatrixType::MaxRowsAtCompileTime,
MaxColsAtCompileTime = Direction==Horizontal ? 1 : MatrixType::MaxColsAtCompileTime,
Flags0 = (unsigned int)_MatrixTypeNested::Flags & HereditaryBits,
Flags = (Flags0 & ~RowMajorBit) | (RowsAtCompileTime == 1 ? RowMajorBit : 0),
Flags = RowsAtCompileTime == 1 ? RowMajorBit : 0,
TraversalSize = Direction==Vertical ? MatrixType::RowsAtCompileTime : MatrixType::ColsAtCompileTime
};
#if EIGEN_GNUC_AT_LEAST(3,4)
typedef typename MemberOp::template Cost<InputScalar,int(TraversalSize)> CostOpType;
#else
typedef typename MemberOp::template Cost<InputScalar,TraversalSize> CostOpType;
#endif
enum {
CoeffReadCost = TraversalSize==Dynamic ? Dynamic
: TraversalSize * traits<_MatrixTypeNested>::CoeffReadCost + int(CostOpType::value)
};
};
}

44
Eigen/src/Core/Visitor.h
View File

@ -53,6 +53,33 @@ struct visitor_impl<Visitor, Derived, Dynamic>
}
};
// evaluator adaptor
template<typename XprType>
class visitor_evaluator
{
public:
visitor_evaluator(const XprType &xpr) : m_evaluator(xpr), m_xpr(xpr) {}
typedef typename XprType::Index Index;
typedef typename XprType::Scalar Scalar;
typedef typename XprType::CoeffReturnType CoeffReturnType;
enum {
RowsAtCompileTime = XprType::RowsAtCompileTime,
CoeffReadCost = internal::evaluator<XprType>::CoeffReadCost
};
Index rows() const { return m_xpr.rows(); }
Index cols() const { return m_xpr.cols(); }
Index size() const { return m_xpr.size(); }
CoeffReturnType coeff(Index row, Index col) const
{ return m_evaluator.coeff(row, col); }
protected:
typename internal::evaluator<XprType>::nestedType m_evaluator;
const XprType &m_xpr;
};
} // end namespace internal
/** Applies the visitor \a visitor to the whole coefficients of the matrix or vector.
@ -76,14 +103,17 @@ template<typename Derived>
template<typename Visitor>
void DenseBase<Derived>::visit(Visitor& visitor) const
{
enum { unroll = SizeAtCompileTime != Dynamic
&& CoeffReadCost != Dynamic
&& (SizeAtCompileTime == 1 || internal::functor_traits<Visitor>::Cost != Dynamic)
&& SizeAtCompileTime * CoeffReadCost + (SizeAtCompileTime-1) * internal::functor_traits<Visitor>::Cost
<= EIGEN_UNROLLING_LIMIT };
return internal::visitor_impl<Visitor, Derived,
typedef typename internal::visitor_evaluator<Derived> ThisEvaluator;
ThisEvaluator thisEval(derived());
enum { unroll = SizeAtCompileTime != Dynamic
&& ThisEvaluator::CoeffReadCost != Dynamic
&& (SizeAtCompileTime == 1 || internal::functor_traits<Visitor>::Cost != Dynamic)
&& SizeAtCompileTime * ThisEvaluator::CoeffReadCost + (SizeAtCompileTime-1) * internal::functor_traits<Visitor>::Cost
<= EIGEN_UNROLLING_LIMIT };
return internal::visitor_impl<Visitor, ThisEvaluator,
unroll ? int(SizeAtCompileTime) : Dynamic
>::run(derived(), visitor);
>::run(thisEval, visitor);
}
namespace internal {

4
Eigen/src/Core/arch/AVX/PacketMath.h
View File

@ -141,7 +141,7 @@ template<> EIGEN_STRONG_INLINE Packet8f pmadd(const Packet8f& a, const Packet8f&
// so let's enforce it to generate a vfmadd231ps instruction since the most common use case is to accumulate
// the result of the product.
Packet8f res = c;
asm("vfmadd231ps %[a], %[b], %[c]" : [c] "+x" (res) : [a] "x" (a), [b] "x" (b));
__asm__("vfmadd231ps %[a], %[b], %[c]" : [c] "+x" (res) : [a] "x" (a), [b] "x" (b));
return res;
#else
return _mm256_fmadd_ps(a,b,c);
@ -151,7 +151,7 @@ template<> EIGEN_STRONG_INLINE Packet4d pmadd(const Packet4d& a, const Packet4d&
#if defined(__clang__) || defined(__GNUC__)
// see above
Packet4d res = c;
asm("vfmadd231pd %[a], %[b], %[c]" : [c] "+x" (res) : [a] "x" (a), [b] "x" (b));
__asm__("vfmadd231pd %[a], %[b], %[c]" : [c] "+x" (res) : [a] "x" (a), [b] "x" (b));
return res;
#else
return _mm256_fmadd_pd(a,b,c);

4
Eigen/src/Core/arch/NEON/PacketMath.h
View File

@ -52,12 +52,12 @@ typedef uint32x4_t Packet4ui;
// arm64 does have the pld instruction. If available, let's trust the __builtin_prefetch built-in function
// which available on LLVM and GCC (at least)
#if (defined(__has_builtin) && __has_builtin(__builtin_prefetch)) || defined(__GNUC__)
#if EIGEN_HAS_BUILTIN(__builtin_prefetch) || defined(__GNUC__)
#define EIGEN_ARM_PREFETCH(ADDR) __builtin_prefetch(ADDR);
#elif defined __pld
#define EIGEN_ARM_PREFETCH(ADDR) __pld(ADDR)
#elif !defined(__aarch64__)
#define EIGEN_ARM_PREFETCH(ADDR) asm volatile ( " pld [%[addr]]\n" :: [addr] "r" (ADDR) : "cc" );
#define EIGEN_ARM_PREFETCH(ADDR) __asm__ __volatile__ ( " pld [%[addr]]\n" :: [addr] "r" (ADDR) : "cc" );
#else
// by default no explicit prefetching
#define EIGEN_ARM_PREFETCH(ADDR)

26
Eigen/src/Core/functors/AssignmentFunctors.h
View File

@ -31,7 +31,7 @@ template<typename Scalar>
struct functor_traits<assign_op<Scalar> > {
enum {
Cost = NumTraits<Scalar>::ReadCost,
PacketAccess = packet_traits<Scalar>::IsVectorized
PacketAccess = packet_traits<Scalar>::Vectorizable
};
};
@ -73,7 +73,7 @@ template<typename Scalar>
struct functor_traits<sub_assign_op<Scalar> > {
enum {
Cost = NumTraits<Scalar>::ReadCost + NumTraits<Scalar>::AddCost,
PacketAccess = packet_traits<Scalar>::HasAdd
PacketAccess = packet_traits<Scalar>::HasSub
};
};
@ -81,22 +81,24 @@ struct functor_traits<sub_assign_op<Scalar> > {
* \brief Template functor for scalar/packet assignment with multiplication
*
*/
template<typename Scalar> struct mul_assign_op {
template<typename DstScalar, typename SrcScalar=DstScalar>
struct mul_assign_op {
EIGEN_EMPTY_STRUCT_CTOR(mul_assign_op)
EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE void assignCoeff(Scalar& a, const Scalar& b) const { a *= b; }
EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE void assignCoeff(DstScalar& a, const SrcScalar& b) const { a *= b; }
template<int Alignment, typename Packet>
EIGEN_STRONG_INLINE void assignPacket(Scalar* a, const Packet& b) const
{ internal::pstoret<Scalar,Packet,Alignment>(a,internal::pmul(internal::ploadt<Packet,Alignment>(a),b)); }
EIGEN_STRONG_INLINE void assignPacket(DstScalar* a, const Packet& b) const
{ internal::pstoret<DstScalar,Packet,Alignment>(a,internal::pmul(internal::ploadt<Packet,Alignment>(a),b)); }
};
template<typename Scalar>
struct functor_traits<mul_assign_op<Scalar> > {
template<typename DstScalar, typename SrcScalar>
struct functor_traits<mul_assign_op<DstScalar,SrcScalar> > {
enum {
Cost = NumTraits<Scalar>::ReadCost + NumTraits<Scalar>::MulCost,
PacketAccess = packet_traits<Scalar>::HasMul
Cost = NumTraits<DstScalar>::ReadCost + NumTraits<DstScalar>::MulCost,
PacketAccess = is_same<DstScalar,SrcScalar>::value && packet_traits<DstScalar>::HasMul
};
};
template<typename DstScalar,typename SrcScalar> struct functor_is_product_like<mul_assign_op<DstScalar,SrcScalar> > { enum { ret = 1 }; };
/** \internal
* \brief Template functor for scalar/packet assignment with diviving
@ -115,7 +117,7 @@ template<typename Scalar>
struct functor_traits<div_assign_op<Scalar> > {
enum {
Cost = NumTraits<Scalar>::ReadCost + NumTraits<Scalar>::MulCost,
PacketAccess = packet_traits<Scalar>::HasMul
PacketAccess = packet_traits<Scalar>::HasDiv
};
};
@ -156,7 +158,7 @@ template<typename Scalar>
struct functor_traits<swap_assign_op<Scalar> > {
enum {
Cost = 3 * NumTraits<Scalar>::ReadCost,
PacketAccess = packet_traits<Scalar>::IsVectorized
PacketAccess = packet_traits<Scalar>::Vectorizable
};
};

14
Eigen/src/Core/functors/BinaryFunctors.h
View File

@ -167,9 +167,17 @@ template<typename Scalar> struct scalar_hypot_op {
EIGEN_USING_STD_MATH(max);
EIGEN_USING_STD_MATH(min);
using std::sqrt;
Scalar p = (max)(_x, _y);
Scalar q = (min)(_x, _y);
Scalar qp = q/p;
Scalar p, qp;
if(_x>_y)
{
p = _x;
qp = _y / p;
}
else
{
p = _y;
qp = _x / p;
}
return p * sqrt(Scalar(1) + qp*qp);
}
};

452
Eigen/src/Core/products/CoeffBasedProduct.h
View File

@ -1,452 +0,0 @@
// This file is part of Eigen, a lightweight C++ template library
// for linear algebra.
//
// Copyright (C) 2006-2008 Benoit Jacob <jacob.benoit.1@gmail.com>
// Copyright (C) 2008-2010 Gael Guennebaud <gael.guennebaud@inria.fr>
//
// This Source Code Form is subject to the terms of the Mozilla
// Public License v. 2.0. If a copy of the MPL was not distributed
// with this file, You can obtain one at http://mozilla.org/MPL/2.0/.
#ifndef EIGEN_COEFFBASED_PRODUCT_H
#define EIGEN_COEFFBASED_PRODUCT_H
namespace Eigen {
namespace internal {
/*********************************************************************************
* Coefficient based product implementation.
* It is designed for the following use cases:
* - small fixed sizes
* - lazy products
*********************************************************************************/
/* Since the all the dimensions of the product are small, here we can rely
* on the generic Assign mechanism to evaluate the product per coeff (or packet).
*
* Note that here the inner-loops should always be unrolled.
*/
template<int Traversal, int UnrollingIndex, typename Lhs, typename Rhs, typename RetScalar>
struct product_coeff_impl;
template<int StorageOrder, int UnrollingIndex, typename Lhs, typename Rhs, typename Packet, int LoadMode>
struct product_packet_impl;
template<typename LhsNested, typename RhsNested, int NestingFlags>
struct traits<CoeffBasedProduct<LhsNested,RhsNested,NestingFlags> >
{
typedef MatrixXpr XprKind;
typedef typename remove_all<LhsNested>::type _LhsNested;
typedef typename remove_all<RhsNested>::type _RhsNested;
typedef typename scalar_product_traits<typename _LhsNested::Scalar, typename _RhsNested::Scalar>::ReturnType Scalar;
typedef typename promote_storage_type<typename traits<_LhsNested>::StorageKind,
typename traits<_RhsNested>::StorageKind>::ret StorageKind;
typedef typename promote_index_type<typename traits<_LhsNested>::Index,
typename traits<_RhsNested>::Index>::type Index;
enum {
LhsCoeffReadCost = _LhsNested::CoeffReadCost,
RhsCoeffReadCost = _RhsNested::CoeffReadCost,
LhsFlags = _LhsNested::Flags,
RhsFlags = _RhsNested::Flags,
RowsAtCompileTime = _LhsNested::RowsAtCompileTime,
ColsAtCompileTime = _RhsNested::ColsAtCompileTime,
InnerSize = EIGEN_SIZE_MIN_PREFER_FIXED(_LhsNested::ColsAtCompileTime, _RhsNested::RowsAtCompileTime),
MaxRowsAtCompileTime = _LhsNested::MaxRowsAtCompileTime,
MaxColsAtCompileTime = _RhsNested::MaxColsAtCompileTime,
LhsRowMajor = LhsFlags & RowMajorBit,
RhsRowMajor = RhsFlags & RowMajorBit,
SameType = is_same<typename _LhsNested::Scalar,typename _RhsNested::Scalar>::value,
CanVectorizeRhs = RhsRowMajor && (RhsFlags & PacketAccessBit)
&& (ColsAtCompileTime == Dynamic
|| ( (ColsAtCompileTime % packet_traits<Scalar>::size) == 0
&& (RhsFlags&AlignedBit)
)
),
CanVectorizeLhs = (!LhsRowMajor) && (LhsFlags & PacketAccessBit)
&& (RowsAtCompileTime == Dynamic
|| ( (RowsAtCompileTime % packet_traits<Scalar>::size) == 0
&& (LhsFlags&AlignedBit)
)
),
EvalToRowMajor = (MaxRowsAtCompileTime==1&&MaxColsAtCompileTime!=1) ? 1
: (MaxColsAtCompileTime==1&&MaxRowsAtCompileTime!=1) ? 0
: (RhsRowMajor && !CanVectorizeLhs),
Flags = ((unsigned int)(LhsFlags | RhsFlags) & HereditaryBits & ~RowMajorBit)
| (EvalToRowMajor ? RowMajorBit : 0)
| NestingFlags
| (CanVectorizeLhs ? (LhsFlags & AlignedBit) : 0)
| (CanVectorizeRhs ? (RhsFlags & AlignedBit) : 0)
// TODO enable vectorization for mixed types
| (SameType && (CanVectorizeLhs || CanVectorizeRhs) ? PacketAccessBit : 0),
CoeffReadCost = InnerSize == Dynamic ? Dynamic
: InnerSize * (NumTraits<Scalar>::MulCost + LhsCoeffReadCost + RhsCoeffReadCost)
+ (InnerSize - 1) * NumTraits<Scalar>::AddCost,
/* CanVectorizeInner deserves special explanation. It does not affect the product flags. It is not used outside
* of Product. If the Product itself is not a packet-access expression, there is still a chance that the inner
* loop of the product might be vectorized. This is the meaning of CanVectorizeInner. Since it doesn't affect
* the Flags, it is safe to make this value depend on ActualPacketAccessBit, that doesn't affect the ABI.
*/
CanVectorizeInner = SameType
&& LhsRowMajor
&& (!RhsRowMajor)
&& (LhsFlags & RhsFlags & ActualPacketAccessBit)
&& (LhsFlags & RhsFlags & AlignedBit)
&& (InnerSize % packet_traits<Scalar>::size == 0)
};
};
} // end namespace internal
template<typename LhsNested, typename RhsNested, int NestingFlags>
class CoeffBasedProduct
: internal::no_assignment_operator,
public MatrixBase<CoeffBasedProduct<LhsNested, RhsNested, NestingFlags> >
{
public:
typedef MatrixBase<CoeffBasedProduct> Base;
EIGEN_DENSE_PUBLIC_INTERFACE(CoeffBasedProduct)
typedef typename Base::PlainObject PlainObject;
private:
typedef typename internal::traits<CoeffBasedProduct>::_LhsNested _LhsNested;
typedef typename internal::traits<CoeffBasedProduct>::_RhsNested _RhsNested;
enum {
PacketSize = internal::packet_traits<Scalar>::size,
InnerSize = internal::traits<CoeffBasedProduct>::InnerSize,
Unroll = CoeffReadCost != Dynamic && CoeffReadCost <= EIGEN_UNROLLING_LIMIT,
CanVectorizeInner = internal::traits<CoeffBasedProduct>::CanVectorizeInner
};
typedef internal::product_coeff_impl<CanVectorizeInner ? InnerVectorizedTraversal : DefaultTraversal,
Unroll ? InnerSize-1 : Dynamic,
_LhsNested, _RhsNested, Scalar> ScalarCoeffImpl;
typedef CoeffBasedProduct<LhsNested,RhsNested,NestByRefBit> LazyCoeffBasedProductType;
public:
EIGEN_DEVICE_FUNC
inline CoeffBasedProduct(const CoeffBasedProduct& other)
: Base(), m_lhs(other.m_lhs), m_rhs(other.m_rhs)
{}
template<typename Lhs, typename Rhs>
EIGEN_DEVICE_FUNC
inline CoeffBasedProduct(const Lhs& lhs, const Rhs& rhs)
: m_lhs(lhs), m_rhs(rhs)
{
// we don't allow taking products of matrices of different real types, as that wouldn't be vectorizable.
// We still allow to mix T and complex<T>.
EIGEN_STATIC_ASSERT((internal::scalar_product_traits<typename Lhs::RealScalar, typename Rhs::RealScalar>::Defined),
YOU_MIXED_DIFFERENT_NUMERIC_TYPES__YOU_NEED_TO_USE_THE_CAST_METHOD_OF_MATRIXBASE_TO_CAST_NUMERIC_TYPES_EXPLICITLY)
eigen_assert(lhs.cols() == rhs.rows()
&& "invalid matrix product"
&& "if you wanted a coeff-wise or a dot product use the respective explicit functions");
}
EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE Index rows() const { return m_lhs.rows(); }
EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE Index cols() const { return m_rhs.cols(); }
EIGEN_DEVICE_FUNC
EIGEN_STRONG_INLINE const Scalar coeff(Index row, Index col) const
{
Scalar res;
ScalarCoeffImpl::run(row, col, m_lhs, m_rhs, res);
return res;
}
/* Allow index-based non-packet access. It is impossible though to allow index-based packed access,
* which is why we don't set the LinearAccessBit.
*/
EIGEN_DEVICE_FUNC
EIGEN_STRONG_INLINE const Scalar coeff(Index index) const
{
Scalar res;
const Index row = RowsAtCompileTime == 1 ? 0 : index;
const Index col = RowsAtCompileTime == 1 ? index : 0;
ScalarCoeffImpl::run(row, col, m_lhs, m_rhs, res);
return res;
}
template<int LoadMode>
EIGEN_STRONG_INLINE const PacketScalar packet(Index row, Index col) const
{
PacketScalar res;
internal::product_packet_impl<Flags&RowMajorBit ? RowMajor : ColMajor,
Unroll ? InnerSize-1 : Dynamic,
_LhsNested, _RhsNested, PacketScalar, LoadMode>
::run(row, col, m_lhs, m_rhs, res);
return res;
}
// Implicit conversion to the nested type (trigger the evaluation of the product)
EIGEN_DEVICE_FUNC
EIGEN_STRONG_INLINE operator const PlainObject& () const
{
m_result.lazyAssign(*this);
return m_result;
}
EIGEN_DEVICE_FUNC const _LhsNested& lhs() const { return m_lhs; }
EIGEN_DEVICE_FUNC const _RhsNested& rhs() const { return m_rhs; }
EIGEN_DEVICE_FUNC
const Diagonal<const LazyCoeffBasedProductType,0> diagonal() const
{ return reinterpret_cast<const LazyCoeffBasedProductType&>(*this); }
template<int DiagonalIndex>
EIGEN_DEVICE_FUNC
const Diagonal<const LazyCoeffBasedProductType,DiagonalIndex> diagonal() const
{ return reinterpret_cast<const LazyCoeffBasedProductType&>(*this); }
EIGEN_DEVICE_FUNC
const Diagonal<const LazyCoeffBasedProductType,Dynamic> diagonal(Index index) const
{ return reinterpret_cast<const LazyCoeffBasedProductType&>(*this).diagonal(index); }
protected:
typename internal::add_const_on_value_type<LhsNested>::type m_lhs;
typename internal::add_const_on_value_type<RhsNested>::type m_rhs;
mutable PlainObject m_result;
};
namespace internal {
// here we need to overload the nested rule for products
// such that the nested type is a const reference to a plain matrix
template<typename Lhs, typename Rhs, int N, typename PlainObject>
struct nested<CoeffBasedProduct<Lhs,Rhs,EvalBeforeNestingBit|EvalBeforeAssigningBit>, N, PlainObject>
{
typedef PlainObject const& type;
};
/***************************************************************************
* Normal product .coeff() implementation (with meta-unrolling)
***************************************************************************/
/**************************************
*** Scalar path - no vectorization ***
**************************************/
template<int UnrollingIndex, typename Lhs, typename Rhs, typename RetScalar>
struct product_coeff_impl<DefaultTraversal, UnrollingIndex, Lhs, Rhs, RetScalar>
{
typedef typename Lhs::Index Index;
EIGEN_DEVICE_FUNC
static EIGEN_STRONG_INLINE void run(Index row, Index col, const Lhs& lhs, const Rhs& rhs, RetScalar &res)
{
product_coeff_impl<DefaultTraversal, UnrollingIndex-1, Lhs, Rhs, RetScalar>::run(row, col, lhs, rhs, res);
res += lhs.coeff(row, UnrollingIndex) * rhs.coeff(UnrollingIndex, col);
}
};
template<typename Lhs, typename Rhs, typename RetScalar>
struct product_coeff_impl<DefaultTraversal, 0, Lhs, Rhs, RetScalar>
{
typedef typename Lhs::Index Index;
EIGEN_DEVICE_FUNC
static EIGEN_STRONG_INLINE void run(Index row, Index col, const Lhs& lhs, const Rhs& rhs, RetScalar &res)
{
res = lhs.coeff(row, 0) * rhs.coeff(0, col);
}
};
template<typename Lhs, typename Rhs, typename RetScalar>
struct product_coeff_impl<DefaultTraversal, Dynamic, Lhs, Rhs, RetScalar>
{
typedef typename Lhs::Index Index;
EIGEN_DEVICE_FUNC
static EIGEN_STRONG_INLINE void run(Index row, Index col, const Lhs& lhs, const Rhs& rhs, RetScalar& res)
{
eigen_assert(lhs.cols()>0 && "you are using a non initialized matrix");
res = lhs.coeff(row, 0) * rhs.coeff(0, col);
for(Index i = 1; i < lhs.cols(); ++i)
res += lhs.coeff(row, i) * rhs.coeff(i, col);
}
};
/*******************************************
*** Scalar path with inner vectorization ***
*******************************************/
template<int UnrollingIndex, typename Lhs, typename Rhs, typename Packet>
struct product_coeff_vectorized_unroller
{
typedef typename Lhs::Index Index;
enum { PacketSize = packet_traits<typename Lhs::Scalar>::size };
static EIGEN_STRONG_INLINE void run(Index row, Index col, const Lhs& lhs, const Rhs& rhs, typename Lhs::PacketScalar &pres)
{
product_coeff_vectorized_unroller<UnrollingIndex-PacketSize, Lhs, Rhs, Packet>::run(row, col, lhs, rhs, pres);
pres = padd(pres, pmul( lhs.template packet<Aligned>(row, UnrollingIndex) , rhs.template packet<Aligned>(UnrollingIndex, col) ));
}
};
template<typename Lhs, typename Rhs, typename Packet>
struct product_coeff_vectorized_unroller<0, Lhs, Rhs, Packet>
{
typedef typename Lhs::Index Index;
static EIGEN_STRONG_INLINE void run(Index row, Index col, const Lhs& lhs, const Rhs& rhs, typename Lhs::PacketScalar &pres)
{
pres = pmul(lhs.template packet<Aligned>(row, 0) , rhs.template packet<Aligned>(0, col));
}
};
template<int UnrollingIndex, typename Lhs, typename Rhs, typename RetScalar>
struct product_coeff_impl<InnerVectorizedTraversal, UnrollingIndex, Lhs, Rhs, RetScalar>
{
typedef typename Lhs::PacketScalar Packet;
typedef typename Lhs::Index Index;
enum { PacketSize = packet_traits<typename Lhs::Scalar>::size };
static EIGEN_STRONG_INLINE void run(Index row, Index col, const Lhs& lhs, const Rhs& rhs, RetScalar &res)
{
Packet pres;
product_coeff_vectorized_unroller<UnrollingIndex+1-PacketSize, Lhs, Rhs, Packet>::run(row, col, lhs, rhs, pres);
res = predux(pres);
}
};
template<typename Lhs, typename Rhs, int LhsRows = Lhs::RowsAtCompileTime, int RhsCols = Rhs::ColsAtCompileTime>
struct product_coeff_vectorized_dyn_selector
{
typedef typename Lhs::Index Index;
static EIGEN_STRONG_INLINE void run(Index row, Index col, const Lhs& lhs, const Rhs& rhs, typename Lhs::Scalar &res)
{
res = lhs.row(row).transpose().cwiseProduct(rhs.col(col)).sum();
}
};
// NOTE the 3 following specializations are because taking .col(0) on a vector is a bit slower
// NOTE maybe they are now useless since we have a specialization for Block<Matrix>
template<typename Lhs, typename Rhs, int RhsCols>
struct product_coeff_vectorized_dyn_selector<Lhs,Rhs,1,RhsCols>
{
typedef typename Lhs::Index Index;
static EIGEN_STRONG_INLINE void run(Index /*row*/, Index col, const Lhs& lhs, const Rhs& rhs, typename Lhs::Scalar &res)
{
res = lhs.transpose().cwiseProduct(rhs.col(col)).sum();
}
};
template<typename Lhs, typename Rhs, int LhsRows>
struct product_coeff_vectorized_dyn_selector<Lhs,Rhs,LhsRows,1>
{
typedef typename Lhs::Index Index;
static EIGEN_STRONG_INLINE void run(Index row, Index /*col*/, const Lhs& lhs, const Rhs& rhs, typename Lhs::Scalar &res)
{
res = lhs.row(row).transpose().cwiseProduct(rhs).sum();
}
};
template<typename Lhs, typename Rhs>
struct product_coeff_vectorized_dyn_selector<Lhs,Rhs,1,1>
{
typedef typename Lhs::Index Index;
static EIGEN_STRONG_INLINE void run(Index /*row*/, Index /*col*/, const Lhs& lhs, const Rhs& rhs, typename Lhs::Scalar &res)
{
res = lhs.transpose().cwiseProduct(rhs).sum();
}
};
template<typename Lhs, typename Rhs, typename RetScalar>
struct product_coeff_impl<InnerVectorizedTraversal, Dynamic, Lhs, Rhs, RetScalar>
{
typedef typename Lhs::Index Index;
static EIGEN_STRONG_INLINE void run(Index row, Index col, const Lhs& lhs, const Rhs& rhs, typename Lhs::Scalar &res)
{
product_coeff_vectorized_dyn_selector<Lhs,Rhs>::run(row, col, lhs, rhs, res);
}
};
/*******************
*** Packet path ***
*******************/
template<int UnrollingIndex, typename Lhs, typename Rhs, typename Packet, int LoadMode>
struct product_packet_impl<RowMajor, UnrollingIndex, Lhs, Rhs, Packet, LoadMode>
{
typedef typename Lhs::Index Index;
static EIGEN_STRONG_INLINE void run(Index row, Index col, const Lhs& lhs, const Rhs& rhs, Packet &res)
{
product_packet_impl<RowMajor, UnrollingIndex-1, Lhs, Rhs, Packet, LoadMode>::run(row, col, lhs, rhs, res);
res = pmadd(pset1<Packet>(lhs.coeff(row, UnrollingIndex)), rhs.template packet<LoadMode>(UnrollingIndex, col), res);
}
};
template<int UnrollingIndex, typename Lhs, typename Rhs, typename Packet, int LoadMode>
struct product_packet_impl<ColMajor, UnrollingIndex, Lhs, Rhs, Packet, LoadMode>
{
typedef typename Lhs::Index Index;
static EIGEN_STRONG_INLINE void run(Index row, Index col, const Lhs& lhs, const Rhs& rhs, Packet &res)
{
product_packet_impl<ColMajor, UnrollingIndex-1, Lhs, Rhs, Packet, LoadMode>::run(row, col, lhs, rhs, res);
res = pmadd(lhs.template packet<LoadMode>(row, UnrollingIndex), pset1<Packet>(rhs.coeff(UnrollingIndex, col)), res);
}
};
template<typename Lhs, typename Rhs, typename Packet, int LoadMode>
struct product_packet_impl<RowMajor, 0, Lhs, Rhs, Packet, LoadMode>
{
typedef typename Lhs::Index Index;
static EIGEN_STRONG_INLINE void run(Index row, Index col, const Lhs& lhs, const Rhs& rhs, Packet &res)
{
res = pmul(pset1<Packet>(lhs.coeff(row, 0)),rhs.template packet<LoadMode>(0, col));
}
};
template<typename Lhs, typename Rhs, typename Packet, int LoadMode>
struct product_packet_impl<ColMajor, 0, Lhs, Rhs, Packet, LoadMode>
{
typedef typename Lhs::Index Index;
static EIGEN_STRONG_INLINE void run(Index row, Index col, const Lhs& lhs, const Rhs& rhs, Packet &res)
{
res = pmul(lhs.template packet<LoadMode>(row, 0), pset1<Packet>(rhs.coeff(0, col)));
}
};
template<typename Lhs, typename Rhs, typename Packet, int LoadMode>
struct product_packet_impl<RowMajor, Dynamic, Lhs, Rhs, Packet, LoadMode>
{
typedef typename Lhs::Index Index;
static EIGEN_STRONG_INLINE void run(Index row, Index col, const Lhs& lhs, const Rhs& rhs, Packet& res)
{
eigen_assert(lhs.cols()>0 && "you are using a non initialized matrix");
res = pmul(pset1<Packet>(lhs.coeff(row, 0)),rhs.template packet<LoadMode>(0, col));
for(Index i = 1; i < lhs.cols(); ++i)
res = pmadd(pset1<Packet>(lhs.coeff(row, i)), rhs.template packet<LoadMode>(i, col), res);
}
};
template<typename Lhs, typename Rhs, typename Packet, int LoadMode>
struct product_packet_impl<ColMajor, Dynamic, Lhs, Rhs, Packet, LoadMode>
{
typedef typename Lhs::Index Index;
static EIGEN_STRONG_INLINE void run(Index row, Index col, const Lhs& lhs, const Rhs& rhs, Packet& res)
{
eigen_assert(lhs.cols()>0 && "you are using a non initialized matrix");
res = pmul(lhs.template packet<LoadMode>(row, 0), pset1<Packet>(rhs.coeff(0, col)));
for(Index i = 1; i < lhs.cols(); ++i)
res = pmadd(lhs.template packet<LoadMode>(row, i), pset1<Packet>(rhs.coeff(i, col)), res);
}
};
} // end namespace internal
} // end namespace Eigen
#endif // EIGEN_COEFFBASED_PRODUCT_H

141
Eigen/src/Core/products/GeneralMatrixMatrix.h
View File

@ -216,8 +216,8 @@ struct gemm_functor
cols = m_rhs.cols();
Gemm::run(rows, cols, m_lhs.cols(),
/*(const Scalar*)*/&m_lhs.coeffRef(row,0), m_lhs.outerStride(),
/*(const Scalar*)*/&m_rhs.coeffRef(0,col), m_rhs.outerStride(),
&m_lhs.coeffRef(row,0), m_lhs.outerStride(),
&m_rhs.coeffRef(0,col), m_rhs.outerStride(),
(Scalar*)&(m_dest.coeffRef(row,col)), m_dest.outerStride(),
m_actualAlpha, m_blocking, info);
}
@ -367,84 +367,93 @@ class gemm_blocking_space<StorageOrder,_LhsScalar,_RhsScalar,MaxRows, MaxCols, M
} // end namespace internal
namespace internal {
template<typename Lhs, typename Rhs>
class GeneralProduct<Lhs, Rhs, GemmProduct>
: public ProductBase<GeneralProduct<Lhs,Rhs,GemmProduct>, Lhs, Rhs>
struct generic_product_impl<Lhs,Rhs,DenseShape,DenseShape,GemmProduct>
: generic_product_impl_base<Lhs,Rhs,generic_product_impl<Lhs,Rhs,DenseShape,DenseShape,GemmProduct> >
{
enum {
MaxDepthAtCompileTime = EIGEN_SIZE_MIN_PREFER_FIXED(Lhs::MaxColsAtCompileTime,Rhs::MaxRowsAtCompileTime)
};
public:
EIGEN_PRODUCT_PUBLIC_INTERFACE(GeneralProduct)
typedef typename Lhs::Scalar LhsScalar;
typedef typename Rhs::Scalar RhsScalar;
typedef Scalar ResScalar;
GeneralProduct(const Lhs& lhs, const Rhs& rhs) : Base(lhs,rhs)
typedef typename Product<Lhs,Rhs>::Scalar Scalar;
typedef typename Product<Lhs,Rhs>::Index Index;
typedef typename Lhs::Scalar LhsScalar;
typedef typename Rhs::Scalar RhsScalar;
typedef internal::blas_traits<Lhs> LhsBlasTraits;
typedef typename LhsBlasTraits::DirectLinearAccessType ActualLhsType;
typedef typename internal::remove_all<ActualLhsType>::type ActualLhsTypeCleaned;
typedef internal::blas_traits<Rhs> RhsBlasTraits;
typedef typename RhsBlasTraits::DirectLinearAccessType ActualRhsType;
typedef typename internal::remove_all<ActualRhsType>::type ActualRhsTypeCleaned;
enum {
MaxDepthAtCompileTime = EIGEN_SIZE_MIN_PREFER_FIXED(Lhs::MaxColsAtCompileTime,Rhs::MaxRowsAtCompileTime)
};
typedef generic_product_impl<Lhs,Rhs,DenseShape,DenseShape,CoeffBasedProductMode> lazyproduct;
template<typename Dst>
static void evalTo(Dst& dst, const Lhs& lhs, const Rhs& rhs)
{
if((rhs.rows()+dst.rows()+dst.cols())<20 && rhs.rows()>0)
lazyproduct::evalTo(dst, lhs, rhs);
else
{
typedef internal::scalar_product_op<LhsScalar,RhsScalar> BinOp;
EIGEN_CHECK_BINARY_COMPATIBILIY(BinOp,LhsScalar,RhsScalar);
}
template<typename Dest>
inline void evalTo(Dest& dst) const
{
if((m_rhs.rows()+dst.rows()+dst.cols())<20 && m_rhs.rows()>0)
dst.noalias() = m_lhs .lazyProduct( m_rhs );
else
{
dst.setZero();
scaleAndAddTo(dst,Scalar(1));
}
dst.setZero();
scaleAndAddTo(dst, lhs, rhs, Scalar(1));
}
}
template<typename Dest>
inline void addTo(Dest& dst) const
{
if((m_rhs.rows()+dst.rows()+dst.cols())<20 && m_rhs.rows()>0)
dst.noalias() += m_lhs .lazyProduct( m_rhs );
else
scaleAndAddTo(dst,Scalar(1));
}
template<typename Dst>
static void addTo(Dst& dst, const Lhs& lhs, const Rhs& rhs)
{
if((rhs.rows()+dst.rows()+dst.cols())<20 && rhs.rows()>0)
lazyproduct::addTo(dst, lhs, rhs);
else
scaleAndAddTo(dst,lhs, rhs, Scalar(1));
}
template<typename Dest>
inline void subTo(Dest& dst) const
{
if((m_rhs.rows()+dst.rows()+dst.cols())<20 && m_rhs.rows()>0)
dst.noalias() -= m_lhs .lazyProduct( m_rhs );
else
scaleAndAddTo(dst,Scalar(-1));
}
template<typename Dest> void scaleAndAddTo(Dest& dst, const Scalar& alpha) const
{
eigen_assert(dst.rows()==m_lhs.rows() && dst.cols()==m_rhs.cols());
template<typename Dst>
static void subTo(Dst& dst, const Lhs& lhs, const Rhs& rhs)
{
if((rhs.rows()+dst.rows()+dst.cols())<20 && rhs.rows()>0)
lazyproduct::subTo(dst, lhs, rhs);
else
scaleAndAddTo(dst, lhs, rhs, Scalar(-1));
}
template<typename Dest>
static void scaleAndAddTo(Dest& dst, const Lhs& a_lhs, const Rhs& a_rhs, const Scalar& alpha)
{
eigen_assert(dst.rows()==a_lhs.rows() && dst.cols()==a_rhs.cols());
typename internal::add_const_on_value_type<ActualLhsType>::type lhs = LhsBlasTraits::extract(m_lhs);
typename internal::add_const_on_value_type<ActualRhsType>::type rhs = RhsBlasTraits::extract(m_rhs);
typename internal::add_const_on_value_type<ActualLhsType>::type lhs = LhsBlasTraits::extract(a_lhs);
typename internal::add_const_on_value_type<ActualRhsType>::type rhs = RhsBlasTraits::extract(a_rhs);
Scalar actualAlpha = alpha * LhsBlasTraits::extractScalarFactor(m_lhs)
* RhsBlasTraits::extractScalarFactor(m_rhs);
Scalar actualAlpha = alpha * LhsBlasTraits::extractScalarFactor(a_lhs)
* RhsBlasTraits::extractScalarFactor(a_rhs);
typedef internal::gemm_blocking_space<(Dest::Flags&RowMajorBit) ? RowMajor : ColMajor,LhsScalar,RhsScalar,
Dest::MaxRowsAtCompileTime,Dest::MaxColsAtCompileTime,MaxDepthAtCompileTime> BlockingType;
typedef internal::gemm_blocking_space<(Dest::Flags&RowMajorBit) ? RowMajor : ColMajor,LhsScalar,RhsScalar,
Dest::MaxRowsAtCompileTime,Dest::MaxColsAtCompileTime,MaxDepthAtCompileTime> BlockingType;
typedef internal::gemm_functor<
Scalar, Index,
internal::general_matrix_matrix_product<
Index,
LhsScalar, (_ActualLhsType::Flags&RowMajorBit) ? RowMajor : ColMajor, bool(LhsBlasTraits::NeedToConjugate),
RhsScalar, (_ActualRhsType::Flags&RowMajorBit) ? RowMajor : ColMajor, bool(RhsBlasTraits::NeedToConjugate),
(Dest::Flags&RowMajorBit) ? RowMajor : ColMajor>,
_ActualLhsType, _ActualRhsType, Dest, BlockingType> GemmFunctor;
typedef internal::gemm_functor<
Scalar, Index,
internal::general_matrix_matrix_product<
Index,
LhsScalar, (ActualLhsTypeCleaned::Flags&RowMajorBit) ? RowMajor : ColMajor, bool(LhsBlasTraits::NeedToConjugate),
RhsScalar, (ActualRhsTypeCleaned::Flags&RowMajorBit) ? RowMajor : ColMajor, bool(RhsBlasTraits::NeedToConjugate),
(Dest::Flags&RowMajorBit) ? RowMajor : ColMajor>,
ActualLhsTypeCleaned, ActualRhsTypeCleaned, Dest, BlockingType> GemmFunctor;
BlockingType blocking(dst.rows(), dst.cols(), lhs.cols(), true);
BlockingType blocking(dst.rows(), dst.cols(), lhs.cols(), true);
internal::parallelize_gemm<(Dest::MaxRowsAtCompileTime>32 || Dest::MaxRowsAtCompileTime==Dynamic)>(GemmFunctor(lhs, rhs, dst, actualAlpha, blocking), this->rows(), this->cols(), Dest::Flags&RowMajorBit);
}
internal::parallelize_gemm<(Dest::MaxRowsAtCompileTime>32 || Dest::MaxRowsAtCompileTime==Dynamic)>
(GemmFunctor(lhs, rhs, dst, actualAlpha, blocking), a_lhs.rows(), a_rhs.cols(), Dest::Flags&RowMajorBit);
}
};
} // end namespace internal
} // end namespace Eigen
#endif // EIGEN_GENERAL_MATRIX_MATRIX_H

14
Eigen/src/Core/products/GeneralMatrixMatrixTriangular.h
View File

@ -20,7 +20,7 @@ namespace internal {
/**********************************************************************
* This file implements a general A * B product while
* evaluating only one triangular part of the product.
* This is more general version of self adjoint product (C += A A^T)
* This is a more general version of self adjoint product (C += A A^T)
* as the level 3 SYRK Blas routine.
**********************************************************************/
@ -262,14 +262,14 @@ struct general_product_to_triangular_selector<MatrixType,ProductType,UpLo,false>
};
template<typename MatrixType, unsigned int UpLo>
template<typename ProductDerived, typename _Lhs, typename _Rhs>
TriangularView<MatrixType,UpLo>& TriangularView<MatrixType,UpLo>::assignProduct(const ProductBase<ProductDerived, _Lhs,_Rhs>& prod, const Scalar& alpha)
template<typename ProductType>
TriangularView<MatrixType,UpLo>& TriangularViewImpl<MatrixType,UpLo,Dense>::_assignProduct(const ProductType& prod, const Scalar& alpha)
{
eigen_assert(m_matrix.rows() == prod.rows() && m_matrix.cols() == prod.cols());
general_product_to_triangular_selector<MatrixType, ProductDerived, UpLo, (_Lhs::ColsAtCompileTime==1) || (_Rhs::RowsAtCompileTime==1)>::run(m_matrix.const_cast_derived(), prod.derived(), alpha);
eigen_assert(derived().nestedExpression().rows() == prod.rows() && derived().cols() == prod.cols());
return *this;
general_product_to_triangular_selector<MatrixType, ProductType, UpLo, internal::traits<ProductType>::InnerSize==1>::run(derived().nestedExpression().const_cast_derived(), prod, alpha);
return derived();
}
} // end namespace Eigen

2
Eigen/src/Core/products/GeneralMatrixMatrix_MKL.h
View File

@ -53,6 +53,8 @@ template< \
int RhsStorageOrder, bool ConjugateRhs> \
struct general_matrix_matrix_product<Index,EIGTYPE,LhsStorageOrder,ConjugateLhs,EIGTYPE,RhsStorageOrder,ConjugateRhs,ColMajor> \
{ \
typedef gebp_traits<EIGTYPE,EIGTYPE> Traits; \
\
static void run(Index rows, Index cols, Index depth, \
const EIGTYPE* _lhs, Index lhsStride, \
const EIGTYPE* _rhs, Index rhsStride, \

55
Eigen/src/Core/products/SelfadjointMatrixMatrix.h
View File

@ -460,55 +460,54 @@ EIGEN_DONT_INLINE void product_selfadjoint_matrix<Scalar,Index,LhsStorageOrder,f
***************************************************************************/
namespace internal {
template<typename Lhs, int LhsMode, typename Rhs, int RhsMode>
struct traits<SelfadjointProductMatrix<Lhs,LhsMode,false,Rhs,RhsMode,false> >
: traits<ProductBase<SelfadjointProductMatrix<Lhs,LhsMode,false,Rhs,RhsMode,false>, Lhs, Rhs> >
{};
}
template<typename Lhs, int LhsMode, typename Rhs, int RhsMode>
struct SelfadjointProductMatrix<Lhs,LhsMode,false,Rhs,RhsMode,false>
: public ProductBase<SelfadjointProductMatrix<Lhs,LhsMode,false,Rhs,RhsMode,false>, Lhs, Rhs >
struct selfadjoint_product_impl<Lhs,LhsMode,false,Rhs,RhsMode,false>
{
EIGEN_PRODUCT_PUBLIC_INTERFACE(SelfadjointProductMatrix)
SelfadjointProductMatrix(const Lhs& lhs, const Rhs& rhs) : Base(lhs,rhs) {}
typedef typename Product<Lhs,Rhs>::Scalar Scalar;
typedef typename Product<Lhs,Rhs>::Index Index;
typedef internal::blas_traits<Lhs> LhsBlasTraits;
typedef typename LhsBlasTraits::DirectLinearAccessType ActualLhsType;
typedef internal::blas_traits<Rhs> RhsBlasTraits;
typedef typename RhsBlasTraits::DirectLinearAccessType ActualRhsType;
enum {
LhsIsUpper = (LhsMode&(Upper|Lower))==Upper,
LhsIsSelfAdjoint = (LhsMode&SelfAdjoint)==SelfAdjoint,
RhsIsUpper = (RhsMode&(Upper|Lower))==Upper,
RhsIsSelfAdjoint = (RhsMode&SelfAdjoint)==SelfAdjoint
};
template<typename Dest> void scaleAndAddTo(Dest& dst, const Scalar& alpha) const
template<typename Dest>
static void run(Dest &dst, const Lhs &a_lhs, const Rhs &a_rhs, const Scalar& alpha)
{
eigen_assert(dst.rows()==m_lhs.rows() && dst.cols()==m_rhs.cols());
eigen_assert(dst.rows()==a_lhs.rows() && dst.cols()==a_rhs.cols());
typename internal::add_const_on_value_type<ActualLhsType>::type lhs = LhsBlasTraits::extract(m_lhs);
typename internal::add_const_on_value_type<ActualRhsType>::type rhs = RhsBlasTraits::extract(m_rhs);
typename internal::add_const_on_value_type<ActualLhsType>::type lhs = LhsBlasTraits::extract(a_lhs);
typename internal::add_const_on_value_type<ActualRhsType>::type rhs = RhsBlasTraits::extract(a_rhs);
Scalar actualAlpha = alpha * LhsBlasTraits::extractScalarFactor(m_lhs)
* RhsBlasTraits::extractScalarFactor(m_rhs);
Scalar actualAlpha = alpha * LhsBlasTraits::extractScalarFactor(a_lhs)
* RhsBlasTraits::extractScalarFactor(a_rhs);
internal::product_selfadjoint_matrix<Scalar, Index,
EIGEN_LOGICAL_XOR(LhsIsUpper,
internal::traits<Lhs>::Flags &RowMajorBit) ? RowMajor : ColMajor, LhsIsSelfAdjoint,
EIGEN_LOGICAL_XOR(LhsIsUpper,internal::traits<Lhs>::Flags &RowMajorBit) ? RowMajor : ColMajor, LhsIsSelfAdjoint,
NumTraits<Scalar>::IsComplex && EIGEN_LOGICAL_XOR(LhsIsUpper,bool(LhsBlasTraits::NeedToConjugate)),
EIGEN_LOGICAL_XOR(RhsIsUpper,
internal::traits<Rhs>::Flags &RowMajorBit) ? RowMajor : ColMajor, RhsIsSelfAdjoint,
EIGEN_LOGICAL_XOR(RhsIsUpper,internal::traits<Rhs>::Flags &RowMajorBit) ? RowMajor : ColMajor, RhsIsSelfAdjoint,
NumTraits<Scalar>::IsComplex && EIGEN_LOGICAL_XOR(RhsIsUpper,bool(RhsBlasTraits::NeedToConjugate)),
internal::traits<Dest>::Flags&RowMajorBit ? RowMajor : ColMajor>
::run(
lhs.rows(), rhs.cols(), // sizes
&lhs.coeffRef(0,0), lhs.outerStride(), // lhs info
&rhs.coeffRef(0,0), rhs.outerStride(), // rhs info
&dst.coeffRef(0,0), dst.outerStride(), // result info
actualAlpha // alpha
lhs.rows(), rhs.cols(), // sizes
&lhs.coeffRef(0,0), lhs.outerStride(), // lhs info
&rhs.coeffRef(0,0), rhs.outerStride(), // rhs info
&dst.coeffRef(0,0), dst.outerStride(), // result info
actualAlpha // alpha
);
}
};
} // end namespace internal
} // end namespace Eigen
#endif // EIGEN_SELFADJOINT_MATRIX_MATRIX_H

81
Eigen/src/Core/products/SelfadjointMatrixVector.h
View File

@ -169,45 +169,45 @@ EIGEN_DONT_INLINE void selfadjoint_matrix_vector_product<Scalar,Index,StorageOrd
***************************************************************************/
namespace internal {
template<typename Lhs, int LhsMode, typename Rhs>
struct traits<SelfadjointProductMatrix<Lhs,LhsMode,false,Rhs,0,true> >
: traits<ProductBase<SelfadjointProductMatrix<Lhs,LhsMode,false,Rhs,0,true>, Lhs, Rhs> >
{};
}
template<typename Lhs, int LhsMode, typename Rhs>
struct SelfadjointProductMatrix<Lhs,LhsMode,false,Rhs,0,true>
: public ProductBase<SelfadjointProductMatrix<Lhs,LhsMode,false,Rhs,0,true>, Lhs, Rhs >
struct selfadjoint_product_impl<Lhs,LhsMode,false,Rhs,0,true>
{
EIGEN_PRODUCT_PUBLIC_INTERFACE(SelfadjointProductMatrix)
typedef typename Product<Lhs,Rhs>::Scalar Scalar;
typedef typename Product<Lhs,Rhs>::Index Index;
typedef internal::blas_traits<Lhs> LhsBlasTraits;
typedef typename LhsBlasTraits::DirectLinearAccessType ActualLhsType;
typedef typename internal::remove_all<ActualLhsType>::type ActualLhsTypeCleaned;
typedef internal::blas_traits<Rhs> RhsBlasTraits;
typedef typename RhsBlasTraits::DirectLinearAccessType ActualRhsType;
typedef typename internal::remove_all<ActualRhsType>::type ActualRhsTypeCleaned;
enum {
LhsUpLo = LhsMode&(Upper|Lower)
};
enum { LhsUpLo = LhsMode&(Upper|Lower) };
SelfadjointProductMatrix(const Lhs& lhs, const Rhs& rhs) : Base(lhs,rhs) {}
template<typename Dest> void scaleAndAddTo(Dest& dest, const Scalar& alpha) const
template<typename Dest>
static void run(Dest& dest, const Lhs &a_lhs, const Rhs &a_rhs, const Scalar& alpha)
{
typedef typename Dest::Scalar ResScalar;
typedef typename Base::RhsScalar RhsScalar;
typedef typename Rhs::Scalar RhsScalar;
typedef Map<Matrix<ResScalar,Dynamic,1>, Aligned> MappedDest;
eigen_assert(dest.rows()==m_lhs.rows() && dest.cols()==m_rhs.cols());
eigen_assert(dest.rows()==a_lhs.rows() && dest.cols()==a_rhs.cols());
typename internal::add_const_on_value_type<ActualLhsType>::type lhs = LhsBlasTraits::extract(m_lhs);
typename internal::add_const_on_value_type<ActualRhsType>::type rhs = RhsBlasTraits::extract(m_rhs);
typename internal::add_const_on_value_type<ActualLhsType>::type lhs = LhsBlasTraits::extract(a_lhs);
typename internal::add_const_on_value_type<ActualRhsType>::type rhs = RhsBlasTraits::extract(a_rhs);
Scalar actualAlpha = alpha * LhsBlasTraits::extractScalarFactor(m_lhs)
* RhsBlasTraits::extractScalarFactor(m_rhs);
Scalar actualAlpha = alpha * LhsBlasTraits::extractScalarFactor(a_lhs)
* RhsBlasTraits::extractScalarFactor(a_rhs);
enum {
EvalToDest = (Dest::InnerStrideAtCompileTime==1),
UseRhs = (_ActualRhsType::InnerStrideAtCompileTime==1)
UseRhs = (ActualRhsTypeCleaned::InnerStrideAtCompileTime==1)
};
internal::gemv_static_vector_if<ResScalar,Dest::SizeAtCompileTime,Dest::MaxSizeAtCompileTime,!EvalToDest> static_dest;
internal::gemv_static_vector_if<RhsScalar,_ActualRhsType::SizeAtCompileTime,_ActualRhsType::MaxSizeAtCompileTime,!UseRhs> static_rhs;
internal::gemv_static_vector_if<RhsScalar,ActualRhsTypeCleaned::SizeAtCompileTime,ActualRhsTypeCleaned::MaxSizeAtCompileTime,!UseRhs> static_rhs;
ei_declare_aligned_stack_constructed_variable(ResScalar,actualDestPtr,dest.size(),
EvalToDest ? dest.data() : static_dest.data());
@ -218,7 +218,7 @@ struct SelfadjointProductMatrix<Lhs,LhsMode,false,Rhs,0,true>
if(!EvalToDest)
{
#ifdef EIGEN_DENSE_STORAGE_CTOR_PLUGIN
Index size = dest.size();
int size = dest.size();
EIGEN_DENSE_STORAGE_CTOR_PLUGIN
#endif
MappedDest(actualDestPtr, dest.size()) = dest;
@ -227,14 +227,15 @@ struct SelfadjointProductMatrix<Lhs,LhsMode,false,Rhs,0,true>
if(!UseRhs)
{
#ifdef EIGEN_DENSE_STORAGE_CTOR_PLUGIN
Index size = rhs.size();
int size = rhs.size();
EIGEN_DENSE_STORAGE_CTOR_PLUGIN
#endif
Map<typename _ActualRhsType::PlainObject>(actualRhsPtr, rhs.size()) = rhs;
Map<typename ActualRhsTypeCleaned::PlainObject>(actualRhsPtr, rhs.size()) = rhs;
}
internal::selfadjoint_matrix_vector_product<Scalar, Index, (internal::traits<_ActualLhsType>::Flags&RowMajorBit) ? RowMajor : ColMajor, int(LhsUpLo), bool(LhsBlasTraits::NeedToConjugate), bool(RhsBlasTraits::NeedToConjugate)>::run
internal::selfadjoint_matrix_vector_product<Scalar, Index, (internal::traits<ActualLhsTypeCleaned>::Flags&RowMajorBit) ? RowMajor : ColMajor,
int(LhsUpLo), bool(LhsBlasTraits::NeedToConjugate), bool(RhsBlasTraits::NeedToConjugate)>::run
(
lhs.rows(), // size
&lhs.coeffRef(0,0), lhs.outerStride(), // lhs info
@ -248,34 +249,24 @@ struct SelfadjointProductMatrix<Lhs,LhsMode,false,Rhs,0,true>
}
};
namespace internal {
template<typename Lhs, typename Rhs, int RhsMode>
struct traits<SelfadjointProductMatrix<Lhs,0,true,Rhs,RhsMode,false> >
: traits<ProductBase<SelfadjointProductMatrix<Lhs,0,true,Rhs,RhsMode,false>, Lhs, Rhs> >
{};
}
template<typename Lhs, typename Rhs, int RhsMode>
struct SelfadjointProductMatrix<Lhs,0,true,Rhs,RhsMode,false>
: public ProductBase<SelfadjointProductMatrix<Lhs,0,true,Rhs,RhsMode,false>, Lhs, Rhs >
struct selfadjoint_product_impl<Lhs,0,true,Rhs,RhsMode,false>
{
EIGEN_PRODUCT_PUBLIC_INTERFACE(SelfadjointProductMatrix)
typedef typename Product<Lhs,Rhs>::Scalar Scalar;
enum { RhsUpLo = RhsMode&(Upper|Lower) };
enum {
RhsUpLo = RhsMode&(Upper|Lower)
};
SelfadjointProductMatrix(const Lhs& lhs, const Rhs& rhs) : Base(lhs,rhs) {}
template<typename Dest> void scaleAndAddTo(Dest& dest, const Scalar& alpha) const
template<typename Dest>
static void run(Dest& dest, const Lhs &a_lhs, const Rhs &a_rhs, const Scalar& alpha)
{
// let's simply transpose the product
Transpose<Dest> destT(dest);
SelfadjointProductMatrix<Transpose<const Rhs>, int(RhsUpLo)==Upper ? Lower : Upper, false,
Transpose<const Lhs>, 0, true>(m_rhs.transpose(), m_lhs.transpose()).scaleAndAddTo(destT, alpha);
selfadjoint_product_impl<Transpose<const Rhs>, int(RhsUpLo)==Upper ? Lower : Upper, false,
Transpose<const Lhs>, 0, true>::run(destT, a_rhs.transpose(), a_lhs.transpose(), alpha);
}
};
} // end namespace internal
} // end namespace Eigen
#endif // EIGEN_SELFADJOINT_MATRIX_VECTOR_H

43
Eigen/src/Core/products/TriangularMatrixMatrix.h
View File

@ -369,28 +369,29 @@ EIGEN_DONT_INLINE void product_triangular_matrix_matrix<Scalar,Index,Mode,false,
* Wrapper to product_triangular_matrix_matrix
***************************************************************************/
template<int Mode, bool LhsIsTriangular, typename Lhs, typename Rhs>
struct traits<TriangularProduct<Mode,LhsIsTriangular,Lhs,false,Rhs,false> >
: traits<ProductBase<TriangularProduct<Mode,LhsIsTriangular,Lhs,false,Rhs,false>, Lhs, Rhs> >
{};
} // end namespace internal
namespace internal {
template<int Mode, bool LhsIsTriangular, typename Lhs, typename Rhs>
struct TriangularProduct<Mode,LhsIsTriangular,Lhs,false,Rhs,false>
: public ProductBase<TriangularProduct<Mode,LhsIsTriangular,Lhs,false,Rhs,false>, Lhs, Rhs >
struct triangular_product_impl<Mode,LhsIsTriangular,Lhs,false,Rhs,false>
{
EIGEN_PRODUCT_PUBLIC_INTERFACE(TriangularProduct)
TriangularProduct(const Lhs& lhs, const Rhs& rhs) : Base(lhs,rhs) {}
template<typename Dest> void scaleAndAddTo(Dest& dst, const Scalar& alpha) const
template<typename Dest> static void run(Dest& dst, const Lhs &a_lhs, const Rhs &a_rhs, const typename Dest::Scalar& alpha)
{
typename internal::add_const_on_value_type<ActualLhsType>::type lhs = LhsBlasTraits::extract(m_lhs);
typename internal::add_const_on_value_type<ActualRhsType>::type rhs = RhsBlasTraits::extract(m_rhs);
typedef typename Dest::Index Index;
typedef typename Dest::Scalar Scalar;
typedef internal::blas_traits<Lhs> LhsBlasTraits;
typedef typename LhsBlasTraits::DirectLinearAccessType ActualLhsType;
typedef typename internal::remove_all<ActualLhsType>::type ActualLhsTypeCleaned;
typedef internal::blas_traits<Rhs> RhsBlasTraits;
typedef typename RhsBlasTraits::DirectLinearAccessType ActualRhsType;
typedef typename internal::remove_all<ActualRhsType>::type ActualRhsTypeCleaned;
typename internal::add_const_on_value_type<ActualLhsType>::type lhs = LhsBlasTraits::extract(a_lhs);
typename internal::add_const_on_value_type<ActualRhsType>::type rhs = RhsBlasTraits::extract(a_rhs);
Scalar actualAlpha = alpha * LhsBlasTraits::extractScalarFactor(m_lhs)
* RhsBlasTraits::extractScalarFactor(m_rhs);
Scalar actualAlpha = alpha * LhsBlasTraits::extractScalarFactor(a_lhs)
* RhsBlasTraits::extractScalarFactor(a_rhs);
typedef internal::gemm_blocking_space<(Dest::Flags&RowMajorBit) ? RowMajor : ColMajor,Scalar,Scalar,
Lhs::MaxRowsAtCompileTime, Rhs::MaxColsAtCompileTime, Lhs::MaxColsAtCompileTime,4> BlockingType;
@ -405,19 +406,21 @@ struct TriangularProduct<Mode,LhsIsTriangular,Lhs,false,Rhs,false>
internal::product_triangular_matrix_matrix<Scalar, Index,
Mode, LhsIsTriangular,
(internal::traits<_ActualLhsType>::Flags&RowMajorBit) ? RowMajor : ColMajor, LhsBlasTraits::NeedToConjugate,
(internal::traits<_ActualRhsType>::Flags&RowMajorBit) ? RowMajor : ColMajor, RhsBlasTraits::NeedToConjugate,
(internal::traits<ActualLhsTypeCleaned>::Flags&RowMajorBit) ? RowMajor : ColMajor, LhsBlasTraits::NeedToConjugate,
(internal::traits<ActualRhsTypeCleaned>::Flags&RowMajorBit) ? RowMajor : ColMajor, RhsBlasTraits::NeedToConjugate,
(internal::traits<Dest >::Flags&RowMajorBit) ? RowMajor : ColMajor>
::run(
stripedRows, stripedCols, stripedDepth, // sizes
&lhs.coeffRef(0,0), lhs.outerStride(), // lhs info
&rhs.coeffRef(0,0), rhs.outerStride(), // rhs info
&lhs.coeffRef(0,0), lhs.outerStride(), // lhs info
&rhs.coeffRef(0,0), rhs.outerStride(), // rhs info
&dst.coeffRef(0,0), dst.outerStride(), // result info
actualAlpha, blocking
);
}
};
} // end namespace internal
} // end namespace Eigen
#endif // EIGEN_TRIANGULAR_MATRIX_MATRIX_H

4
Eigen/src/Core/products/TriangularMatrixMatrix_MKL.h
View File

@ -109,7 +109,7 @@ struct product_triangular_matrix_matrix_trmm<EIGTYPE,Index,Mode,true, \
/* Non-square case - doesn't fit to MKL ?TRMM. Fall to default triangular product or call MKL ?GEMM*/ \
if (rows != depth) { \
\
int nthr = mkl_domain_get_max_threads(MKL_BLAS); \
int nthr = mkl_domain_get_max_threads(EIGEN_MKL_DOMAIN_BLAS); \
\
if (((nthr==1) && (((std::max)(rows,depth)-diagSize)/(double)diagSize < 0.5))) { \
/* Most likely no benefit to call TRMM or GEMM from MKL*/ \
@ -223,7 +223,7 @@ struct product_triangular_matrix_matrix_trmm<EIGTYPE,Index,Mode,false, \
/* Non-square case - doesn't fit to MKL ?TRMM. Fall to default triangular product or call MKL ?GEMM*/ \
if (cols != depth) { \
\
int nthr = mkl_domain_get_max_threads(MKL_BLAS); \
int nthr = mkl_domain_get_max_threads(EIGEN_MKL_DOMAIN_BLAS); \
\
if ((nthr==1) && (((std::max)(cols,depth)-diagSize)/(double)diagSize < 0.5)) { \
/* Most likely no benefit to call TRMM or GEMM from MKL*/ \

122
Eigen/src/Core/products/TriangularMatrixVector.h
View File

@ -157,83 +157,67 @@ EIGEN_DONT_INLINE void triangular_matrix_vector_product<Index,Mode,LhsScalar,Con
* Wrapper to product_triangular_vector
***************************************************************************/
template<int Mode, bool LhsIsTriangular, typename Lhs, typename Rhs>
struct traits<TriangularProduct<Mode,LhsIsTriangular,Lhs,false,Rhs,true> >
: traits<ProductBase<TriangularProduct<Mode,LhsIsTriangular,Lhs,false,Rhs,true>, Lhs, Rhs> >
{};
template<int Mode, bool LhsIsTriangular, typename Lhs, typename Rhs>
struct traits<TriangularProduct<Mode,LhsIsTriangular,Lhs,true,Rhs,false> >
: traits<ProductBase<TriangularProduct<Mode,LhsIsTriangular,Lhs,true,Rhs,false>, Lhs, Rhs> >
{};
template<int StorageOrder>
template<int Mode,int StorageOrder>
struct trmv_selector;
} // end namespace internal
namespace internal {
template<int Mode, typename Lhs, typename Rhs>
struct TriangularProduct<Mode,true,Lhs,false,Rhs,true>
: public ProductBase<TriangularProduct<Mode,true,Lhs,false,Rhs,true>, Lhs, Rhs >
struct triangular_product_impl<Mode,true,Lhs,false,Rhs,true>
{
EIGEN_PRODUCT_PUBLIC_INTERFACE(TriangularProduct)
TriangularProduct(const Lhs& lhs, const Rhs& rhs) : Base(lhs,rhs) {}
template<typename Dest> void scaleAndAddTo(Dest& dst, const Scalar& alpha) const
template<typename Dest> static void run(Dest& dst, const Lhs &lhs, const Rhs &rhs, const typename Dest::Scalar& alpha)
{
eigen_assert(dst.rows()==m_lhs.rows() && dst.cols()==m_rhs.cols());
eigen_assert(dst.rows()==lhs.rows() && dst.cols()==rhs.cols());
internal::trmv_selector<(int(internal::traits<Lhs>::Flags)&RowMajorBit) ? RowMajor : ColMajor>::run(*this, dst, alpha);
internal::trmv_selector<Mode,(int(internal::traits<Lhs>::Flags)&RowMajorBit) ? RowMajor : ColMajor>::run(lhs, rhs, dst, alpha);
}
};
template<int Mode, typename Lhs, typename Rhs>
struct TriangularProduct<Mode,false,Lhs,true,Rhs,false>
: public ProductBase<TriangularProduct<Mode,false,Lhs,true,Rhs,false>, Lhs, Rhs >
struct triangular_product_impl<Mode,false,Lhs,true,Rhs,false>
{
EIGEN_PRODUCT_PUBLIC_INTERFACE(TriangularProduct)
TriangularProduct(const Lhs& lhs, const Rhs& rhs) : Base(lhs,rhs) {}
template<typename Dest> void scaleAndAddTo(Dest& dst, const Scalar& alpha) const
template<typename Dest> static void run(Dest& dst, const Lhs &lhs, const Rhs &rhs, const typename Dest::Scalar& alpha)
{
eigen_assert(dst.rows()==m_lhs.rows() && dst.cols()==m_rhs.cols());
eigen_assert(dst.rows()==lhs.rows() && dst.cols()==rhs.cols());
typedef TriangularProduct<(Mode & (UnitDiag|ZeroDiag)) | ((Mode & Lower) ? Upper : Lower),true,Transpose<const Rhs>,false,Transpose<const Lhs>,true> TriangularProductTranspose;
Transpose<Dest> dstT(dst);
internal::trmv_selector<(int(internal::traits<Rhs>::Flags)&RowMajorBit) ? ColMajor : RowMajor>::run(
TriangularProductTranspose(m_rhs.transpose(),m_lhs.transpose()), dstT, alpha);
internal::trmv_selector<(Mode & (UnitDiag|ZeroDiag)) | ((Mode & Lower) ? Upper : Lower),
(int(internal::traits<Rhs>::Flags)&RowMajorBit) ? ColMajor : RowMajor>
::run(rhs.transpose(),lhs.transpose(), dstT, alpha);
}
};
} // end namespace internal
namespace internal {
// TODO: find a way to factorize this piece of code with gemv_selector since the logic is exactly the same.
template<> struct trmv_selector<ColMajor>
template<int Mode> struct trmv_selector<Mode,ColMajor>
{
template<int Mode, typename Lhs, typename Rhs, typename Dest>
static void run(const TriangularProduct<Mode,true,Lhs,false,Rhs,true>& prod, Dest& dest, const typename TriangularProduct<Mode,true,Lhs,false,Rhs,true>::Scalar& alpha)
template<typename Lhs, typename Rhs, typename Dest>
static void run(const Lhs &lhs, const Rhs &rhs, Dest& dest, const typename Dest::Scalar& alpha)
{
typedef TriangularProduct<Mode,true,Lhs,false,Rhs,true> ProductType;
typedef typename ProductType::Index Index;
typedef typename ProductType::LhsScalar LhsScalar;
typedef typename ProductType::RhsScalar RhsScalar;
typedef typename ProductType::Scalar ResScalar;
typedef typename ProductType::RealScalar RealScalar;
typedef typename ProductType::ActualLhsType ActualLhsType;
typedef typename ProductType::ActualRhsType ActualRhsType;
typedef typename ProductType::LhsBlasTraits LhsBlasTraits;
typedef typename ProductType::RhsBlasTraits RhsBlasTraits;
typedef typename Dest::Index Index;
typedef typename Lhs::Scalar LhsScalar;
typedef typename Rhs::Scalar RhsScalar;
typedef typename Dest::Scalar ResScalar;
typedef typename Dest::RealScalar RealScalar;
typedef internal::blas_traits<Lhs> LhsBlasTraits;
typedef typename LhsBlasTraits::DirectLinearAccessType ActualLhsType;
typedef internal::blas_traits<Rhs> RhsBlasTraits;
typedef typename RhsBlasTraits::DirectLinearAccessType ActualRhsType;
typedef Map<Matrix<ResScalar,Dynamic,1>, Aligned> MappedDest;
typename internal::add_const_on_value_type<ActualLhsType>::type actualLhs = LhsBlasTraits::extract(prod.lhs());
typename internal::add_const_on_value_type<ActualRhsType>::type actualRhs = RhsBlasTraits::extract(prod.rhs());
typename internal::add_const_on_value_type<ActualLhsType>::type actualLhs = LhsBlasTraits::extract(lhs);
typename internal::add_const_on_value_type<ActualRhsType>::type actualRhs = RhsBlasTraits::extract(rhs);
ResScalar actualAlpha = alpha * LhsBlasTraits::extractScalarFactor(prod.lhs())
* RhsBlasTraits::extractScalarFactor(prod.rhs());
ResScalar actualAlpha = alpha * LhsBlasTraits::extractScalarFactor(lhs)
* RhsBlasTraits::extractScalarFactor(rhs);
enum {
// FIXME find a way to allow an inner stride on the result if packet_traits<Scalar>::size==1
@ -288,33 +272,33 @@ template<> struct trmv_selector<ColMajor>
}
};
template<> struct trmv_selector<RowMajor>
template<int Mode> struct trmv_selector<Mode,RowMajor>
{
template<int Mode, typename Lhs, typename Rhs, typename Dest>
static void run(const TriangularProduct<Mode,true,Lhs,false,Rhs,true>& prod, Dest& dest, const typename TriangularProduct<Mode,true,Lhs,false,Rhs,true>::Scalar& alpha)
template<typename Lhs, typename Rhs, typename Dest>
static void run(const Lhs &lhs, const Rhs &rhs, Dest& dest, const typename Dest::Scalar& alpha)
{
typedef TriangularProduct<Mode,true,Lhs,false,Rhs,true> ProductType;
typedef typename ProductType::LhsScalar LhsScalar;
typedef typename ProductType::RhsScalar RhsScalar;
typedef typename ProductType::Scalar ResScalar;
typedef typename ProductType::Index Index;
typedef typename ProductType::ActualLhsType ActualLhsType;
typedef typename ProductType::ActualRhsType ActualRhsType;
typedef typename ProductType::_ActualRhsType _ActualRhsType;
typedef typename ProductType::LhsBlasTraits LhsBlasTraits;
typedef typename ProductType::RhsBlasTraits RhsBlasTraits;
typedef typename Dest::Index Index;
typedef typename Lhs::Scalar LhsScalar;
typedef typename Rhs::Scalar RhsScalar;
typedef typename Dest::Scalar ResScalar;
typedef internal::blas_traits<Lhs> LhsBlasTraits;
typedef typename LhsBlasTraits::DirectLinearAccessType ActualLhsType;
typedef internal::blas_traits<Rhs> RhsBlasTraits;
typedef typename RhsBlasTraits::DirectLinearAccessType ActualRhsType;
typedef typename internal::remove_all<ActualRhsType>::type ActualRhsTypeCleaned;
typename add_const<ActualLhsType>::type actualLhs = LhsBlasTraits::extract(prod.lhs());
typename add_const<ActualRhsType>::type actualRhs = RhsBlasTraits::extract(prod.rhs());
typename add_const<ActualLhsType>::type actualLhs = LhsBlasTraits::extract(lhs);
typename add_const<ActualRhsType>::type actualRhs = RhsBlasTraits::extract(rhs);
ResScalar actualAlpha = alpha * LhsBlasTraits::extractScalarFactor(prod.lhs())
* RhsBlasTraits::extractScalarFactor(prod.rhs());
ResScalar actualAlpha = alpha * LhsBlasTraits::extractScalarFactor(lhs)
* RhsBlasTraits::extractScalarFactor(rhs);
enum {
DirectlyUseRhs = _ActualRhsType::InnerStrideAtCompileTime==1
DirectlyUseRhs = ActualRhsTypeCleaned::InnerStrideAtCompileTime==1
};
gemv_static_vector_if<RhsScalar,_ActualRhsType::SizeAtCompileTime,_ActualRhsType::MaxSizeAtCompileTime,!DirectlyUseRhs> static_rhs;
gemv_static_vector_if<RhsScalar,ActualRhsTypeCleaned::SizeAtCompileTime,ActualRhsTypeCleaned::MaxSizeAtCompileTime,!DirectlyUseRhs> static_rhs;
ei_declare_aligned_stack_constructed_variable(RhsScalar,actualRhsPtr,actualRhs.size(),
DirectlyUseRhs ? const_cast<RhsScalar*>(actualRhs.data()) : static_rhs.data());
@ -325,7 +309,7 @@ template<> struct trmv_selector<RowMajor>
Index size = actualRhs.size();
EIGEN_DENSE_STORAGE_CTOR_PLUGIN
#endif
Map<typename _ActualRhsType::PlainObject>(actualRhsPtr, actualRhs.size()) = actualRhs;
Map<typename ActualRhsTypeCleaned::PlainObject>(actualRhsPtr, actualRhs.size()) = actualRhs;
}
internal::triangular_matrix_vector_product

30
Eigen/src/Core/util/Constants.h
View File

@ -53,14 +53,13 @@ const int Infinity = -1;
const unsigned int RowMajorBit = 0x1;
/** \ingroup flags
*
* means the expression should be evaluated by the calling expression */
const unsigned int EvalBeforeNestingBit = 0x2;
/** \ingroup flags
*
* \deprecated
* means the expression should be evaluated before any assignment */
const unsigned int EvalBeforeAssigningBit = 0x4;
const unsigned int EvalBeforeAssigningBit = 0x4; // FIXME deprecated
/** \ingroup flags
*
@ -155,6 +154,16 @@ const unsigned int AlignedBit = 0x80;
const unsigned int NestByRefBit = 0x100;
/** \ingroup flags
*
* for an expression, this means that the storage order
* can be either row-major or column-major.
* The precise choice will be decided at evaluation time or when
* combined with other expressions.
* \sa \ref RowMajorBit, \ref TopicStorageOrders */
const unsigned int NoPreferredStorageOrderBit = 0x200;
// list of flags that are inherited by default
const unsigned int HereditaryBits = RowMajorBit
| EvalBeforeNestingBit
@ -431,7 +440,7 @@ namespace Architecture
/** \internal \ingroup enums
* Enum used as template parameter in GeneralProduct. */
enum { CoeffBasedProductMode, LazyCoeffBasedProductMode, OuterProduct, InnerProduct, GemvProduct, GemmProduct };
enum { DefaultProduct=0, CoeffBasedProductMode, LazyCoeffBasedProductMode, LazyProduct, OuterProduct, InnerProduct, GemvProduct, GemmProduct };
/** \internal \ingroup enums
* Enum used in experimental parallel implementation. */
@ -440,12 +449,25 @@ enum Action {GetAction, SetAction};
/** The type used to identify a dense storage. */
struct Dense {};
/** The type used to identify a permutation storage. */
struct PermutationStorage {};
/** The type used to identify a matrix expression */
struct MatrixXpr {};
/** The type used to identify an array expression */
struct ArrayXpr {};
// An evaluator must define its shape. By default, it can be one of the following:
struct DenseShape { static std::string debugName() { return "DenseShape"; } };
struct HomogeneousShape { static std::string debugName() { return "HomogeneousShape"; } };
struct DiagonalShape { static std::string debugName() { return "DiagonalShape"; } };
struct BandShape { static std::string debugName() { return "BandShape"; } };
struct TriangularShape { static std::string debugName() { return "TriangularShape"; } };
struct SelfAdjointShape { static std::string debugName() { return "SelfAdjointShape"; } };
struct PermutationShape { static std::string debugName() { return "PermutationShape"; } };
struct SparseShape { static std::string debugName() { return "SparseShape"; } };
} // end namespace Eigen
#endif // EIGEN_CONSTANTS_H

41
Eigen/src/Core/util/ForwardDeclarations.h
View File

@ -36,6 +36,10 @@ template<typename Derived> struct accessors_level
};
};
template<typename T> struct evaluator_traits;
template< typename T> struct evaluator;
} // end namespace internal
template<typename T> struct NumTraits;
@ -87,11 +91,19 @@ template<typename NullaryOp, typename MatrixType> class CwiseNullaryOp;
template<typename UnaryOp, typename MatrixType> class CwiseUnaryOp;
template<typename ViewOp, typename MatrixType> class CwiseUnaryView;
template<typename BinaryOp, typename Lhs, typename Rhs> class CwiseBinaryOp;
template<typename BinOp, typename Lhs, typename Rhs> class SelfCwiseBinaryOp;
template<typename Derived, typename Lhs, typename Rhs> class ProductBase;
template<typename Lhs, typename Rhs> class Product;
template<typename Lhs, typename Rhs, int Mode> class GeneralProduct;
template<typename Lhs, typename Rhs, int NestingFlags> class CoeffBasedProduct;
template<typename BinOp, typename Lhs, typename Rhs> class SelfCwiseBinaryOp; // TODO deprecated
template<typename Derived, typename Lhs, typename Rhs> class ProductBase; // TODO deprecated
template<typename Decomposition, typename Rhstype> class Solve;
template<typename XprType> class Inverse;
namespace internal {
template<typename Lhs, typename Rhs> struct product_tag;
}
template<typename Lhs, typename Rhs, int Option = DefaultProduct> class Product;
template<typename Lhs, typename Rhs, int Mode> class GeneralProduct; // TODO deprecated
template<typename Lhs, typename Rhs, int NestingFlags> class CoeffBasedProduct; // TODO deprecated
template<typename Derived> class DiagonalBase;
template<typename _DiagonalVectorType> class DiagonalWrapper;
@ -109,7 +121,12 @@ template<typename Derived,
int Level = internal::accessors_level<Derived>::has_write_access ? WriteAccessors : ReadOnlyAccessors
> class MapBase;
template<int InnerStrideAtCompileTime, int OuterStrideAtCompileTime> class Stride;
template<int Value = Dynamic> class InnerStride;
template<int Value = Dynamic> class OuterStride;
template<typename MatrixType, int MapOptions=Unaligned, typename StrideType = Stride<0,0> > class Map;
template<typename Derived> class RefBase;
template<typename PlainObjectType, int Options = 0,
typename StrideType = typename internal::conditional<PlainObjectType::IsVectorAtCompileTime,InnerStride<1>,OuterStride<> >::type > class Ref;
template<typename Derived> class TriangularBase;
template<typename MatrixType, unsigned int Mode> class TriangularView;
@ -122,8 +139,6 @@ template<typename ExpressionType> class ArrayWrapper;
template<typename ExpressionType> class MatrixWrapper;
namespace internal {
template<typename DecompositionType, typename Rhs> struct solve_retval_base;
template<typename DecompositionType, typename Rhs> struct solve_retval;
template<typename DecompositionType> struct kernel_retval_base;
template<typename DecompositionType> struct kernel_retval;
template<typename DecompositionType> struct image_retval_base;
@ -136,6 +151,18 @@ template<typename _Scalar, int Rows=Dynamic, int Cols=Dynamic, int Supers=Dynami
namespace internal {
template<typename Lhs, typename Rhs> struct product_type;
/** \internal
* \class product_evaluator
* Products need their own evaluator with more template arguments allowing for
* easier partial template specializations.
*/
template< typename T,
int ProductTag = internal::product_type<typename T::Lhs,typename T::Rhs>::ret,
typename LhsShape = typename evaluator_traits<typename T::Lhs>::Shape,
typename RhsShape = typename evaluator_traits<typename T::Rhs>::Shape,
typename LhsScalar = typename traits<typename T::Lhs>::Scalar,
typename RhsScalar = typename traits<typename T::Rhs>::Scalar
> struct product_evaluator;
}
template<typename Lhs, typename Rhs,

32
Eigen/src/Core/util/MKL_support.h
View File

@ -76,6 +76,38 @@
#include <mkl_lapacke.h>
#define EIGEN_MKL_VML_THRESHOLD 128
/* MKL_DOMAIN_BLAS, etc are defined only in 10.3 update 7 */
/* MKL_BLAS, etc are not defined in 11.2 */
#ifdef MKL_DOMAIN_ALL
#define EIGEN_MKL_DOMAIN_ALL MKL_DOMAIN_ALL
#else
#define EIGEN_MKL_DOMAIN_ALL MKL_ALL
#endif
#ifdef MKL_DOMAIN_BLAS
#define EIGEN_MKL_DOMAIN_BLAS MKL_DOMAIN_BLAS
#else
#define EIGEN_MKL_DOMAIN_BLAS MKL_BLAS
#endif
#ifdef MKL_DOMAIN_FFT
#define EIGEN_MKL_DOMAIN_FFT MKL_DOMAIN_FFT
#else
#define EIGEN_MKL_DOMAIN_FFT MKL_FFT
#endif
#ifdef MKL_DOMAIN_VML
#define EIGEN_MKL_DOMAIN_VML MKL_DOMAIN_VML
#else
#define EIGEN_MKL_DOMAIN_VML MKL_VML
#endif
#ifdef MKL_DOMAIN_PARDISO
#define EIGEN_MKL_DOMAIN_PARDISO MKL_DOMAIN_PARDISO
#else
#define EIGEN_MKL_DOMAIN_PARDISO MKL_PARDISO
#endif
namespace Eigen {
typedef std::complex<double> dcomplex;

16
Eigen/src/Core/util/Macros.h
View File

@ -107,6 +107,13 @@
#define EIGEN_DEFAULT_DENSE_INDEX_TYPE std::ptrdiff_t
#endif
// Cross compiler wrapper around LLVM's __has_builtin
#ifdef __has_builtin
# define EIGEN_HAS_BUILTIN(x) __has_builtin(x)
#else
# define EIGEN_HAS_BUILTIN(x) 0
#endif
// A Clang feature extension to determine compiler features.
// We use it to determine 'cxx_rvalue_references'
#ifndef __has_feature
@ -272,7 +279,7 @@ namespace Eigen {
#if !defined(EIGEN_ASM_COMMENT)
#if (defined __GNUC__) && ( defined(__i386__) || defined(__x86_64__) )
#define EIGEN_ASM_COMMENT(X) asm("#" X)
#define EIGEN_ASM_COMMENT(X) __asm__("#" X)
#else
#define EIGEN_ASM_COMMENT(X)
#endif
@ -367,6 +374,8 @@ namespace Eigen {
* documentation in a single line.
**/
// TODO The EIGEN_DENSE_PUBLIC_INTERFACE should not exists anymore
#define EIGEN_GENERIC_PUBLIC_INTERFACE(Derived) \
typedef typename Eigen::internal::traits<Derived>::Scalar Scalar; /*!< \brief Numeric type, e.g. float, double, int or std::complex<float>. */ \
typedef typename Eigen::NumTraits<Scalar>::Real RealScalar; /*!< \brief The underlying numeric type for composed scalar types. \details In cases where Scalar is e.g. std::complex<T>, T were corresponding to RealScalar. */ \
@ -377,7 +386,6 @@ namespace Eigen {
enum { RowsAtCompileTime = Eigen::internal::traits<Derived>::RowsAtCompileTime, \
ColsAtCompileTime = Eigen::internal::traits<Derived>::ColsAtCompileTime, \
Flags = Eigen::internal::traits<Derived>::Flags, \
CoeffReadCost = Eigen::internal::traits<Derived>::CoeffReadCost, \
SizeAtCompileTime = Base::SizeAtCompileTime, \
MaxSizeAtCompileTime = Base::MaxSizeAtCompileTime, \
IsVectorAtCompileTime = Base::IsVectorAtCompileTime };
@ -396,13 +404,11 @@ namespace Eigen {
MaxRowsAtCompileTime = Eigen::internal::traits<Derived>::MaxRowsAtCompileTime, \
MaxColsAtCompileTime = Eigen::internal::traits<Derived>::MaxColsAtCompileTime, \
Flags = Eigen::internal::traits<Derived>::Flags, \
CoeffReadCost = Eigen::internal::traits<Derived>::CoeffReadCost, \
SizeAtCompileTime = Base::SizeAtCompileTime, \
MaxSizeAtCompileTime = Base::MaxSizeAtCompileTime, \
IsVectorAtCompileTime = Base::IsVectorAtCompileTime }; \
using Base::derived; \
using Base::const_cast_derived;
using Base::const_cast_derived;
#define EIGEN_PLAIN_ENUM_MIN(a,b) (((int)a <= (int)b) ? (int)a : (int)b)
#define EIGEN_PLAIN_ENUM_MAX(a,b) (((int)a >= (int)b) ? (int)a : (int)b)

12
Eigen/src/Core/util/Meta.h
View File

@ -274,18 +274,6 @@ template<typename T> struct scalar_product_traits<std::complex<T>, T>
// typedef typename scalar_product_traits<typename remove_all<ArgType0>::type, typename remove_all<ArgType1>::type>::ReturnType type;
// };
template<typename T> struct is_diagonal
{ enum { ret = false }; };
template<typename T> struct is_diagonal<DiagonalBase<T> >
{ enum { ret = true }; };
template<typename T> struct is_diagonal<DiagonalWrapper<T> >
{ enum { ret = true }; };
template<typename T, int S> struct is_diagonal<DiagonalMatrix<T,S> >
{ enum { ret = true }; };
} // end namespace internal
namespace numext {

6
Eigen/src/Core/util/StaticAssert.h
View File

@ -84,13 +84,15 @@
THIS_EXPRESSION_IS_NOT_A_LVALUE__IT_IS_READ_ONLY,
YOU_ARE_TRYING_TO_USE_AN_INDEX_BASED_ACCESSOR_ON_AN_EXPRESSION_THAT_DOES_NOT_SUPPORT_THAT,
THIS_METHOD_IS_ONLY_FOR_1x1_EXPRESSIONS,
THIS_METHOD_IS_ONLY_FOR_INNER_OR_LAZY_PRODUCTS,
THIS_METHOD_IS_ONLY_FOR_EXPRESSIONS_OF_BOOL,
THIS_METHOD_IS_ONLY_FOR_ARRAYS_NOT_MATRICES,
YOU_PASSED_A_ROW_VECTOR_BUT_A_COLUMN_VECTOR_WAS_EXPECTED,
YOU_PASSED_A_COLUMN_VECTOR_BUT_A_ROW_VECTOR_WAS_EXPECTED,
THE_INDEX_TYPE_MUST_BE_A_SIGNED_TYPE,
THE_STORAGE_ORDER_OF_BOTH_SIDES_MUST_MATCH,
OBJECT_ALLOCATED_ON_STACK_IS_TOO_BIG
OBJECT_ALLOCATED_ON_STACK_IS_TOO_BIG,
IMPLICIT_CONVERSION_TO_SCALAR_IS_FOR_INNER_PRODUCT_ONLY
};
};
@ -157,7 +159,7 @@
#define EIGEN_PREDICATE_SAME_MATRIX_SIZE(TYPE0,TYPE1) \
( \
(int(TYPE0::SizeAtCompileTime)==0 && int(TYPE1::SizeAtCompileTime)==0) \
(int(internal::size_of_xpr_at_compile_time<TYPE0>::ret)==0 && int(internal::size_of_xpr_at_compile_time<TYPE1>::ret)==0) \
|| (\
(int(TYPE0::RowsAtCompileTime)==Eigen::Dynamic \
|| int(TYPE1::RowsAtCompileTime)==Eigen::Dynamic \

178
Eigen/src/Core/util/XprHelper.h
View File

@ -127,6 +127,17 @@ template<typename _Scalar, int _Rows, int _Cols,
template<typename Scalar, int Rows, int Cols, int Options, int MaxRows, int MaxCols>
class compute_matrix_flags
{
enum { row_major_bit = Options&RowMajor ? RowMajorBit : 0 };
public:
// FIXME currently we still have to handle DirectAccessBit at the expression level to handle DenseCoeffsBase<>
// and then propagate this information to the evaluator's flags.
// However, I (Gael) think that DirectAccessBit should only matter at the evaluation stage.
enum { ret = DirectAccessBit | LvalueBit | NestByRefBit | row_major_bit };
};
template<typename Scalar, int Rows, int Cols, int Options, int MaxRows, int MaxCols>
class compute_matrix_evaluator_flags
{
enum {
row_major_bit = Options&RowMajor ? RowMajorBit : 0,
@ -156,7 +167,7 @@ class compute_matrix_flags
};
public:
enum { ret = LinearAccessBit | LvalueBit | DirectAccessBit | NestByRefBit | packet_access_bit | row_major_bit | aligned_bit };
enum { ret = LinearAccessBit | DirectAccessBit | packet_access_bit | row_major_bit | aligned_bit };
};
template<int _Rows, int _Cols> struct size_at_compile_time
@ -164,6 +175,11 @@ template<int _Rows, int _Cols> struct size_at_compile_time
enum { ret = (_Rows==Dynamic || _Cols==Dynamic) ? Dynamic : _Rows * _Cols };
};
template<typename XprType> struct size_of_xpr_at_compile_time
{
enum { ret = size_at_compile_time<traits<XprType>::RowsAtCompileTime,traits<XprType>::ColsAtCompileTime>::ret };
};
/* plain_matrix_type : the difference from eval is that plain_matrix_type is always a plain matrix type,
* whereas eval is a const reference in the case of a matrix
*/
@ -174,6 +190,10 @@ template<typename T> struct plain_matrix_type<T,Dense>
{
typedef typename plain_matrix_type_dense<T,typename traits<T>::XprKind>::type type;
};
template<typename T> struct plain_matrix_type<T,DiagonalShape>
{
typedef typename T::PlainObject type;
};
template<typename T> struct plain_matrix_type_dense<T,MatrixXpr>
{
@ -216,6 +236,11 @@ template<typename T> struct eval<T,Dense>
// > type;
};
template<typename T> struct eval<T,DiagonalShape>
{
typedef typename plain_matrix_type<T>::type type;
};
// for matrices, no need to evaluate, just use a const reference to avoid a useless copy
template<typename _Scalar, int _Rows, int _Cols, int _Options, int _MaxRows, int _MaxCols>
struct eval<Matrix<_Scalar, _Rows, _Cols, _Options, _MaxRows, _MaxCols>, Dense>
@ -294,38 +319,42 @@ struct transfer_constness
>::type type;
};
/** \internal Determines how a given expression should be nested into another one.
// When using evaluators, we never evaluate when assembling the expression!!
// TODO: get rid of this nested class since it's just an alias for ref_selector.
template<typename T, int n=1, typename PlainObject = void> struct nested
{
typedef typename ref_selector<T>::type type;
};
// However, we still need a mechanism to detect whether an expression which is evaluated multiple time
// has to be evaluated into a temporary.
// That's the purpose of this new nested_eval helper:
/** \internal Determines how a given expression should be nested when evaluated multiple times.
* For example, when you do a * (b+c), Eigen will determine how the expression b+c should be
* nested into the bigger product expression. The choice is between nesting the expression b+c as-is, or
* evaluated into the bigger product expression. The choice is between nesting the expression b+c as-is, or
* evaluating that expression b+c into a temporary variable d, and nest d so that the resulting expression is
* a*d. Evaluating can be beneficial for example if every coefficient access in the resulting expression causes
* many coefficient accesses in the nested expressions -- as is the case with matrix product for example.
*
* \param T the type of the expression being nested
* \param T the type of the expression being nested.
* \param n the number of coefficient accesses in the nested expression for each coefficient access in the bigger expression.
*
* Note that if no evaluation occur, then the constness of T is preserved.
*
* Example. Suppose that a, b, and c are of type Matrix3d. The user forms the expression a*(b+c).
* b+c is an expression "sum of matrices", which we will denote by S. In order to determine how to nest it,
* the Product expression uses: nested<S, 3>::type, which turns out to be Matrix3d because the internal logic of
* nested determined that in this case it was better to evaluate the expression b+c into a temporary. On the other hand,
* since a is of type Matrix3d, the Product expression nests it as nested<Matrix3d, 3>::type, which turns out to be
* const Matrix3d&, because the internal logic of nested determined that since a was already a matrix, there was no point
* in copying it into another matrix.
* \param PlainObject the type of the temporary if needed.
*/
template<typename T, int n=1, typename PlainObject = typename eval<T>::type> struct nested
template<typename T, int n, typename PlainObject = typename eval<T>::type> struct nested_eval
{
enum {
// for the purpose of this test, to keep it reasonably simple, we arbitrarily choose a value of Dynamic values.
// For the purpose of this test, to keep it reasonably simple, we arbitrarily choose a value of Dynamic values.
// the choice of 10000 makes it larger than any practical fixed value and even most dynamic values.
// in extreme cases where these assumptions would be wrong, we would still at worst suffer performance issues
// (poor choice of temporaries).
// it's important that this value can still be squared without integer overflowing.
// It's important that this value can still be squared without integer overflowing.
DynamicAsInteger = 10000,
ScalarReadCost = NumTraits<typename traits<T>::Scalar>::ReadCost,
ScalarReadCostAsInteger = ScalarReadCost == Dynamic ? int(DynamicAsInteger) : int(ScalarReadCost),
CoeffReadCost = traits<T>::CoeffReadCost,
CoeffReadCost = evaluator<T>::CoeffReadCost, // TODO What if an evaluator evaluate itself into a tempory?
// Then CoeffReadCost will be small but we still have to evaluate if n>1...
// The solution might be to ask the evaluator if it creates a temp. Perhaps we could even ask the number of temps?
CoeffReadCostAsInteger = CoeffReadCost == Dynamic ? int(DynamicAsInteger) : int(CoeffReadCost),
NAsInteger = n == Dynamic ? int(DynamicAsInteger) : n,
CostEvalAsInteger = (NAsInteger+1) * ScalarReadCostAsInteger + CoeffReadCostAsInteger,
@ -333,11 +362,10 @@ template<typename T, int n=1, typename PlainObject = typename eval<T>::type> str
};
typedef typename conditional<
( (int(traits<T>::Flags) & EvalBeforeNestingBit) ||
int(CostEvalAsInteger) < int(CostNoEvalAsInteger)
),
PlainObject,
typename ref_selector<T>::type
( (int(evaluator<T>::Flags) & EvalBeforeNestingBit) ||
(int(CostEvalAsInteger) < int(CostNoEvalAsInteger)) ),
PlainObject,
typename ref_selector<T>::type
>::type type;
};
@ -366,6 +394,15 @@ struct dense_xpr_base<Derived, ArrayXpr>
typedef ArrayBase<Derived> type;
};
template<typename Derived, typename XprKind = typename traits<Derived>::XprKind, typename StorageKind = typename traits<Derived>::StorageKind>
struct generic_xpr_base;
template<typename Derived, typename XprKind>
struct generic_xpr_base<Derived, XprKind, Dense>
{
typedef typename dense_xpr_base<Derived,XprKind>::type type;
};
/** \internal Helper base class to add a scalar multiple operator
* overloads for complex types */
template<typename Derived,typename Scalar,typename OtherScalar,
@ -383,13 +420,21 @@ struct special_scalar_op_base<Derived,Scalar,OtherScalar,true> : public DenseCo
const CwiseUnaryOp<scalar_multiple2_op<Scalar,OtherScalar>, Derived>
operator*(const OtherScalar& scalar) const
{
#ifdef EIGEN_SPECIAL_SCALAR_MULTIPLE_PLUGIN
EIGEN_SPECIAL_SCALAR_MULTIPLE_PLUGIN
#endif
return CwiseUnaryOp<scalar_multiple2_op<Scalar,OtherScalar>, Derived>
(*static_cast<const Derived*>(this), scalar_multiple2_op<Scalar,OtherScalar>(scalar));
}
inline friend const CwiseUnaryOp<scalar_multiple2_op<Scalar,OtherScalar>, Derived>
operator*(const OtherScalar& scalar, const Derived& matrix)
{ return static_cast<const special_scalar_op_base&>(matrix).operator*(scalar); }
{
#ifdef EIGEN_SPECIAL_SCALAR_MULTIPLE_PLUGIN
EIGEN_SPECIAL_SCALAR_MULTIPLE_PLUGIN
#endif
return static_cast<const special_scalar_op_base&>(matrix).operator*(scalar);
}
};
template<typename XprType, typename CastType> struct cast_return_type
@ -401,12 +446,59 @@ template<typename XprType, typename CastType> struct cast_return_type
const XprType&,CastType>::type type;
};
template <typename A, typename B> struct promote_storage_type;
/** \internal Specify the "storage kind" of applying a coefficient-wise
* binary operations between two expressions of kinds A and B respectively.
* The template parameter Functor permits to specialize the resulting storage kind wrt to
* the functor.
* The default rules are as follows:
* \code
* A op A -> A
* A op dense -> dense
* dense op B -> dense
* A * dense -> A
* dense * B -> B
* \endcode
*/
template <typename A, typename B, typename Functor> struct cwise_promote_storage_type;
template <typename A> struct promote_storage_type<A,A>
{
typedef A ret;
};
template <typename A, typename Functor> struct cwise_promote_storage_type<A,A,Functor> { typedef A ret; };
template <typename Functor> struct cwise_promote_storage_type<Dense,Dense,Functor> { typedef Dense ret; };
template <typename ScalarA, typename ScalarB> struct cwise_promote_storage_type<Dense,Dense,scalar_product_op<ScalarA,ScalarB> > { typedef Dense ret; };
template <typename A, typename Functor> struct cwise_promote_storage_type<A,Dense,Functor> { typedef Dense ret; };
template <typename B, typename Functor> struct cwise_promote_storage_type<Dense,B,Functor> { typedef Dense ret; };
template <typename A, typename ScalarA, typename ScalarB> struct cwise_promote_storage_type<A,Dense,scalar_product_op<ScalarA,ScalarB> > { typedef A ret; };
template <typename B, typename ScalarA, typename ScalarB> struct cwise_promote_storage_type<Dense,B,scalar_product_op<ScalarA,ScalarB> > { typedef B ret; };
/** \internal Specify the "storage kind" of multiplying an expression of kind A with kind B.
* The template parameter ProductTag permits to specialize the resulting storage kind wrt to
* some compile-time properties of the product: GemmProduct, GemvProduct, OuterProduct, InnerProduct.
* The default rules are as follows:
* \code
* K * K -> K
* dense * K -> dense
* K * dense -> dense
* diag * K -> K
* K * diag -> K
* Perm * K -> K
* K * Perm -> K
* \endcode
*/
template <typename A, typename B, int ProductTag> struct product_promote_storage_type;
template <typename A, int ProductTag> struct product_promote_storage_type<A, A, ProductTag> { typedef A ret;};
template <int ProductTag> struct product_promote_storage_type<Dense, Dense, ProductTag> { typedef Dense ret;};
template <typename A, int ProductTag> struct product_promote_storage_type<A, Dense, ProductTag> { typedef Dense ret; };
template <typename B, int ProductTag> struct product_promote_storage_type<Dense, B, ProductTag> { typedef Dense ret; };
template <typename A, int ProductTag> struct product_promote_storage_type<A, DiagonalShape, ProductTag> { typedef A ret; };
template <typename B, int ProductTag> struct product_promote_storage_type<DiagonalShape, B, ProductTag> { typedef B ret; };
template <int ProductTag> struct product_promote_storage_type<Dense, DiagonalShape, ProductTag> { typedef Dense ret; };
template <int ProductTag> struct product_promote_storage_type<DiagonalShape, Dense, ProductTag> { typedef Dense ret; };
template <typename A, int ProductTag> struct product_promote_storage_type<A, PermutationStorage, ProductTag> { typedef A ret; };
template <typename B, int ProductTag> struct product_promote_storage_type<PermutationStorage, B, ProductTag> { typedef B ret; };
template <int ProductTag> struct product_promote_storage_type<Dense, PermutationStorage, ProductTag> { typedef Dense ret; };
template <int ProductTag> struct product_promote_storage_type<PermutationStorage, Dense, ProductTag> { typedef Dense ret; };
/** \internal gives the plain matrix or array type to store a row/column/diagonal of a matrix type.
* \param Scalar optional parameter allowing to pass a different scalar type than the one of the MatrixType.
@ -464,8 +556,36 @@ struct is_lvalue
bool(traits<ExpressionType>::Flags & LvalueBit) };
};
template<typename T> struct is_diagonal
{ enum { ret = false }; };
template<typename T> struct is_diagonal<DiagonalBase<T> >
{ enum { ret = true }; };
template<typename T> struct is_diagonal<DiagonalWrapper<T> >
{ enum { ret = true }; };
template<typename T, int S> struct is_diagonal<DiagonalMatrix<T,S> >
{ enum { ret = true }; };
template<typename S1, typename S2> struct glue_shapes;
template<> struct glue_shapes<DenseShape,TriangularShape> { typedef TriangularShape type; };
} // 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)
} // end namespace Eigen
#endif // EIGEN_XPRHELPER_H

4
Eigen/src/Eigenvalues/Tridiagonalization.h
View File

@ -18,8 +18,10 @@ namespace internal {
template<typename MatrixType> struct TridiagonalizationMatrixTReturnType;
template<typename MatrixType>
struct traits<TridiagonalizationMatrixTReturnType<MatrixType> >
: public traits<typename MatrixType::PlainObject>
{
typedef typename MatrixType::PlainObject ReturnType;
typedef typename MatrixType::PlainObject ReturnType; // FIXME shall it be a BandMatrix?
enum { Flags = 0 };
};
template<typename MatrixType, typename CoeffVectorType>

2
Eigen/src/Geometry/AlignedBox.h
View File

@ -71,7 +71,7 @@ EIGEN_MAKE_ALIGNED_OPERATOR_NEW_IF_VECTORIZABLE_FIXED_SIZE(_Scalar,_AmbientDim)
template<typename Derived>
inline explicit AlignedBox(const MatrixBase<Derived>& a_p)
{
typename internal::nested<Derived,2>::type p(a_p.derived());
typename internal::nested_eval<Derived,2>::type p(a_p.derived());
m_min = p;
m_max = p;
}

124
Eigen/src/Geometry/Homogeneous.h
View File

@ -48,8 +48,7 @@ struct traits<Homogeneous<MatrixType,Direction> >
TmpFlags = _MatrixTypeNested::Flags & HereditaryBits,
Flags = ColsAtCompileTime==1 ? (TmpFlags & ~RowMajorBit)
: RowsAtCompileTime==1 ? (TmpFlags | RowMajorBit)
: TmpFlags,
CoeffReadCost = _MatrixTypeNested::CoeffReadCost
: TmpFlags
};
};
@ -63,6 +62,7 @@ template<typename MatrixType,int _Direction> class Homogeneous
{
public:
typedef MatrixType NestedExpression;
enum { Direction = _Direction };
typedef MatrixBase<Homogeneous> Base;
@ -74,37 +74,38 @@ template<typename MatrixType,int _Direction> class Homogeneous
inline Index rows() const { return m_matrix.rows() + (int(Direction)==Vertical ? 1 : 0); }
inline Index cols() const { return m_matrix.cols() + (int(Direction)==Horizontal ? 1 : 0); }
inline Scalar coeff(Index row, Index col) const
{
if( (int(Direction)==Vertical && row==m_matrix.rows())
|| (int(Direction)==Horizontal && col==m_matrix.cols()))
return 1;
return m_matrix.coeff(row, col);
}
const NestedExpression& nestedExpression() const { return m_matrix; }
template<typename Rhs>
inline const internal::homogeneous_right_product_impl<Homogeneous,Rhs>
inline const Product<Homogeneous,Rhs>
operator* (const MatrixBase<Rhs>& rhs) const
{
eigen_assert(int(Direction)==Horizontal);
return internal::homogeneous_right_product_impl<Homogeneous,Rhs>(m_matrix,rhs.derived());
return Product<Homogeneous,Rhs>(*this,rhs.derived());
}
template<typename Lhs> friend
inline const internal::homogeneous_left_product_impl<Homogeneous,Lhs>
inline const Product<Lhs,Homogeneous>
operator* (const MatrixBase<Lhs>& lhs, const Homogeneous& rhs)
{
eigen_assert(int(Direction)==Vertical);
return internal::homogeneous_left_product_impl<Homogeneous,Lhs>(lhs.derived(),rhs.m_matrix);
return Product<Lhs,Homogeneous>(lhs.derived(),rhs);
}
template<typename Scalar, int Dim, int Mode, int Options> friend
inline const internal::homogeneous_left_product_impl<Homogeneous,Transform<Scalar,Dim,Mode,Options> >
inline const Product<Transform<Scalar,Dim,Mode,Options>, Homogeneous >
operator* (const Transform<Scalar,Dim,Mode,Options>& lhs, const Homogeneous& rhs)
{
eigen_assert(int(Direction)==Vertical);
return internal::homogeneous_left_product_impl<Homogeneous,Transform<Scalar,Dim,Mode,Options> >(lhs,rhs.m_matrix);
return Product<Transform<Scalar,Dim,Mode,Options>, Homogeneous>(lhs,rhs);
}
template<typename Func>
EIGEN_STRONG_INLINE typename internal::result_of<Func(Scalar)>::type
redux(const Func& func) const
{
return func(m_matrix.redux(func), Scalar(1));
}
protected:
@ -120,7 +121,7 @@ template<typename MatrixType,int _Direction> class Homogeneous
* Example: \include MatrixBase_homogeneous.cpp
* Output: \verbinclude MatrixBase_homogeneous.out
*
* \sa class Homogeneous
* \sa VectorwiseOp::homogeneous(), class Homogeneous
*/
template<typename Derived>
inline typename MatrixBase<Derived>::HomogeneousReturnType
@ -137,7 +138,7 @@ MatrixBase<Derived>::homogeneous() const
* Example: \include VectorwiseOp_homogeneous.cpp
* Output: \verbinclude VectorwiseOp_homogeneous.out
*
* \sa MatrixBase::homogeneous() */
* \sa MatrixBase::homogeneous(), class Homogeneous */
template<typename ExpressionType, int Direction>
inline Homogeneous<ExpressionType,Direction>
VectorwiseOp<ExpressionType,Direction>::homogeneous() const
@ -300,6 +301,93 @@ struct homogeneous_right_product_impl<Homogeneous<MatrixType,Horizontal>,Rhs>
typename Rhs::Nested m_rhs;
};
template<typename ArgType,int Direction>
struct evaluator_traits<Homogeneous<ArgType,Direction> >
{
typedef typename storage_kind_to_evaluator_kind<typename ArgType::StorageKind>::Kind Kind;
typedef HomogeneousShape Shape;
static const int AssumeAliasing = 0;
};
template<> struct AssignmentKind<DenseShape,HomogeneousShape> { typedef Dense2Dense Kind; };
template<typename ArgType,int Direction>
struct unary_evaluator<Homogeneous<ArgType,Direction>, IndexBased>
: evaluator<typename Homogeneous<ArgType,Direction>::PlainObject >::type
{
typedef Homogeneous<ArgType,Direction> XprType;
typedef typename XprType::PlainObject PlainObject;
typedef typename evaluator<PlainObject>::type Base;
typedef evaluator<XprType> type;
typedef evaluator<XprType> nestedType;
unary_evaluator(const XprType& op)
: Base(), m_temp(op)
{
::new (static_cast<Base*>(this)) Base(m_temp);
}
protected:
PlainObject m_temp;
};
// dense = homogeneous
template< typename DstXprType, typename ArgType, typename Scalar>
struct Assignment<DstXprType, Homogeneous<ArgType,Vertical>, internal::assign_op<Scalar>, Dense2Dense, Scalar>
{
typedef Homogeneous<ArgType,Vertical> SrcXprType;
static void run(DstXprType &dst, const SrcXprType &src, const internal::assign_op<Scalar> &)
{
dst.template topRows<ArgType::RowsAtCompileTime>(src.nestedExpression().rows()) = src.nestedExpression();
dst.row(dst.rows()-1).setOnes();
}
};
// dense = homogeneous
template< typename DstXprType, typename ArgType, typename Scalar>
struct Assignment<DstXprType, Homogeneous<ArgType,Horizontal>, internal::assign_op<Scalar>, Dense2Dense, Scalar>
{
typedef Homogeneous<ArgType,Horizontal> SrcXprType;
static void run(DstXprType &dst, const SrcXprType &src, const internal::assign_op<Scalar> &)
{
dst.template leftCols<ArgType::ColsAtCompileTime>(src.nestedExpression().cols()) = src.nestedExpression();
dst.col(dst.cols()-1).setOnes();
}
};
template<typename LhsArg, typename Rhs, int ProductTag>
struct generic_product_impl<Homogeneous<LhsArg,Horizontal>, Rhs, HomogeneousShape, DenseShape, ProductTag>
{
template<typename Dest>
static void evalTo(Dest& dst, const Homogeneous<LhsArg,Horizontal>& lhs, const Rhs& rhs)
{
homogeneous_right_product_impl<Homogeneous<LhsArg,Horizontal>, Rhs>(lhs.nestedExpression(), rhs).evalTo(dst);
}
};
template<typename Lhs, typename RhsArg, int ProductTag>
struct generic_product_impl<Lhs, Homogeneous<RhsArg,Vertical>, DenseShape, HomogeneousShape, ProductTag>
{
template<typename Dest>
static void evalTo(Dest& dst, const Lhs& lhs, const Homogeneous<RhsArg,Vertical>& rhs)
{
homogeneous_left_product_impl<Homogeneous<RhsArg,Vertical>, Lhs>(lhs, rhs.nestedExpression()).evalTo(dst);
}
};
template<typename Scalar, int Dim, int Mode,int Options, typename RhsArg, int ProductTag>
struct generic_product_impl<Transform<Scalar,Dim,Mode,Options>, Homogeneous<RhsArg,Vertical>, DenseShape, HomogeneousShape, ProductTag>
{
typedef Transform<Scalar,Dim,Mode,Options> TransformType;
template<typename Dest>
static void evalTo(Dest& dst, const TransformType& lhs, const Homogeneous<RhsArg,Vertical>& rhs)
{
homogeneous_left_product_impl<Homogeneous<RhsArg,Vertical>, TransformType>(lhs, rhs.nestedExpression()).evalTo(dst);
}
};
} // end namespace internal
} // end namespace Eigen

23
Eigen/src/Geometry/OrthoMethods.h
View File

@ -30,8 +30,8 @@ MatrixBase<Derived>::cross(const MatrixBase<OtherDerived>& other) const
// Note that there is no need for an expression here since the compiler
// optimize such a small temporary very well (even within a complex expression)
typename internal::nested<Derived,2>::type lhs(derived());
typename internal::nested<OtherDerived,2>::type rhs(other.derived());
typename internal::nested_eval<Derived,2>::type lhs(derived());
typename internal::nested_eval<OtherDerived,2>::type rhs(other.derived());
return typename cross_product_return_type<OtherDerived>::type(
numext::conj(lhs.coeff(1) * rhs.coeff(2) - lhs.coeff(2) * rhs.coeff(1)),
numext::conj(lhs.coeff(2) * rhs.coeff(0) - lhs.coeff(0) * rhs.coeff(2)),
@ -76,8 +76,8 @@ MatrixBase<Derived>::cross3(const MatrixBase<OtherDerived>& other) const
EIGEN_STATIC_ASSERT_VECTOR_SPECIFIC_SIZE(Derived,4)
EIGEN_STATIC_ASSERT_VECTOR_SPECIFIC_SIZE(OtherDerived,4)
typedef typename internal::nested<Derived,2>::type DerivedNested;
typedef typename internal::nested<OtherDerived,2>::type OtherDerivedNested;
typedef typename internal::nested_eval<Derived,2>::type DerivedNested;
typedef typename internal::nested_eval<OtherDerived,2>::type OtherDerivedNested;
DerivedNested lhs(derived());
OtherDerivedNested rhs(other.derived());
@ -103,21 +103,24 @@ VectorwiseOp<ExpressionType,Direction>::cross(const MatrixBase<OtherDerived>& ot
EIGEN_STATIC_ASSERT_VECTOR_SPECIFIC_SIZE(OtherDerived,3)
EIGEN_STATIC_ASSERT((internal::is_same<Scalar, typename OtherDerived::Scalar>::value),
YOU_MIXED_DIFFERENT_NUMERIC_TYPES__YOU_NEED_TO_USE_THE_CAST_METHOD_OF_MATRIXBASE_TO_CAST_NUMERIC_TYPES_EXPLICITLY)
typename internal::nested_eval<ExpressionType,2>::type mat(_expression());
typename internal::nested_eval<OtherDerived,2>::type vec(other.derived());
CrossReturnType res(_expression().rows(),_expression().cols());
if(Direction==Vertical)
{
eigen_assert(CrossReturnType::RowsAtCompileTime==3 && "the matrix must have exactly 3 rows");
res.row(0) = (_expression().row(1) * other.coeff(2) - _expression().row(2) * other.coeff(1)).conjugate();
res.row(1) = (_expression().row(2) * other.coeff(0) - _expression().row(0) * other.coeff(2)).conjugate();
res.row(2) = (_expression().row(0) * other.coeff(1) - _expression().row(1) * other.coeff(0)).conjugate();
res.row(0) = (mat.row(1) * vec.coeff(2) - mat.row(2) * vec.coeff(1)).conjugate();
res.row(1) = (mat.row(2) * vec.coeff(0) - mat.row(0) * vec.coeff(2)).conjugate();
res.row(2) = (mat.row(0) * vec.coeff(1) - mat.row(1) * vec.coeff(0)).conjugate();
}
else
{
eigen_assert(CrossReturnType::ColsAtCompileTime==3 && "the matrix must have exactly 3 columns");
res.col(0) = (_expression().col(1) * other.coeff(2) - _expression().col(2) * other.coeff(1)).conjugate();
res.col(1) = (_expression().col(2) * other.coeff(0) - _expression().col(0) * other.coeff(2)).conjugate();
res.col(2) = (_expression().col(0) * other.coeff(1) - _expression().col(1) * other.coeff(0)).conjugate();
res.col(0) = (mat.col(1) * vec.coeff(2) - mat.col(2) * vec.coeff(1)).conjugate();
res.col(1) = (mat.col(2) * vec.coeff(0) - mat.col(0) * vec.coeff(2)).conjugate();
res.col(2) = (mat.col(0) * vec.coeff(1) - mat.col(1) * vec.coeff(0)).conjugate();
}
return res;
}

2
Eigen/src/Geometry/Quaternion.h
View File

@ -217,7 +217,7 @@ struct traits<Quaternion<_Scalar,_Options> >
typedef _Scalar Scalar;
typedef Matrix<_Scalar,4,1,_Options> Coefficients;
enum{
IsAligned = internal::traits<Coefficients>::Flags & AlignedBit,
IsAligned = (internal::traits<Coefficients>::EvaluatorFlags & AlignedBit) != 0,
Flags = IsAligned ? (AlignedBit | LvalueBit) : LvalueBit
};
};

19
Eigen/src/Geometry/Transform.h
View File

@ -62,6 +62,22 @@ struct transform_construct_from_matrix;
template<typename TransformType> struct transform_take_affine_part;
template<typename _Scalar, int _Dim, int _Mode, int _Options>
struct traits<Transform<_Scalar,_Dim,_Mode,_Options> >
{
typedef _Scalar Scalar;
typedef DenseIndex Index;
typedef Dense StorageKind;
enum {
Dim1 = _Dim==Dynamic ? _Dim : _Dim + 1,
RowsAtCompileTime = _Mode==Projective ? Dim1 : _Dim,
ColsAtCompileTime = Dim1,
MaxRowsAtCompileTime = RowsAtCompileTime,
MaxColsAtCompileTime = ColsAtCompileTime,
Flags = 0
};
};
} // end namespace internal
/** \geometry_module \ingroup Geometry_Module
@ -355,6 +371,9 @@ public:
inline Transform& operator=(const QTransform& other);
inline QTransform toQTransform(void) const;
#endif
Index rows() const { return int(Mode)==int(Projective) ? m_matrix.cols() : (m_matrix.cols()-1); }
Index cols() const { return m_matrix.cols(); }
/** shortcut for m_matrix(row,col);
* \sa MatrixBase::operator(Index,Index) const */

73
Eigen/src/Householder/BlockHouseholder.h
View File

@ -16,48 +16,85 @@
namespace Eigen {
namespace internal {
/** \internal */
// template<typename TriangularFactorType,typename VectorsType,typename CoeffsType>
// void make_block_householder_triangular_factor(TriangularFactorType& triFactor, const VectorsType& vectors, const CoeffsType& hCoeffs)
// {
// typedef typename TriangularFactorType::Index Index;
// typedef typename VectorsType::Scalar Scalar;
// const Index nbVecs = vectors.cols();
// eigen_assert(triFactor.rows() == nbVecs && triFactor.cols() == nbVecs && vectors.rows()>=nbVecs);
//
// for(Index i = 0; i < nbVecs; i++)
// {
// Index rs = vectors.rows() - i;
// // Warning, note that hCoeffs may alias with vectors.
// // It is then necessary to copy it before modifying vectors(i,i).
// typename CoeffsType::Scalar h = hCoeffs(i);
// // This hack permits to pass trough nested Block<> and Transpose<> expressions.
// Scalar *Vii_ptr = const_cast<Scalar*>(vectors.data() + vectors.outerStride()*i + vectors.innerStride()*i);
// Scalar Vii = *Vii_ptr;
// *Vii_ptr = Scalar(1);
// triFactor.col(i).head(i).noalias() = -h * vectors.block(i, 0, rs, i).adjoint()
// * vectors.col(i).tail(rs);
// *Vii_ptr = Vii;
// // FIXME add .noalias() once the triangular product can work inplace
// triFactor.col(i).head(i) = triFactor.block(0,0,i,i).template triangularView<Upper>()
// * triFactor.col(i).head(i);
// triFactor(i,i) = hCoeffs(i);
// }
// }
/** \internal */
// This variant avoid modifications in vectors
template<typename TriangularFactorType,typename VectorsType,typename CoeffsType>
void make_block_householder_triangular_factor(TriangularFactorType& triFactor, const VectorsType& vectors, const CoeffsType& hCoeffs)
{
typedef typename TriangularFactorType::Index Index;
typedef typename VectorsType::Scalar Scalar;
const Index nbVecs = vectors.cols();
eigen_assert(triFactor.rows() == nbVecs && triFactor.cols() == nbVecs && vectors.rows()>=nbVecs);
for(Index i = 0; i < nbVecs; i++)
for(Index i = nbVecs-1; i >=0 ; --i)
{
Index rs = vectors.rows() - i;
Scalar Vii = vectors(i,i);
vectors.const_cast_derived().coeffRef(i,i) = Scalar(1);
triFactor.col(i).head(i).noalias() = -hCoeffs(i) * vectors.block(i, 0, rs, i).adjoint()
* vectors.col(i).tail(rs);
vectors.const_cast_derived().coeffRef(i, i) = Vii;
// FIXME add .noalias() once the triangular product can work inplace
triFactor.col(i).head(i) = triFactor.block(0,0,i,i).template triangularView<Upper>()
* triFactor.col(i).head(i);
Index rs = vectors.rows() - i - 1;
Index rt = nbVecs-i-1;
if(rt>0)
{
triFactor.row(i).tail(rt).noalias() = -hCoeffs(i) * vectors.col(i).tail(rs).adjoint()
* vectors.bottomRightCorner(rs, rt).template triangularView<UnitLower>();
// FIXME add .noalias() once the triangular product can work inplace
triFactor.row(i).tail(rt) = triFactor.row(i).tail(rt) * triFactor.bottomRightCorner(rt,rt).template triangularView<Upper>();
}
triFactor(i,i) = hCoeffs(i);
}
}
/** \internal */
/** \internal
* if forward then perform mat = H0 * H1 * H2 * mat
* otherwise perform mat = H2 * H1 * H0 * mat
*/
template<typename MatrixType,typename VectorsType,typename CoeffsType>
void apply_block_householder_on_the_left(MatrixType& mat, const VectorsType& vectors, const CoeffsType& hCoeffs)
void apply_block_householder_on_the_left(MatrixType& mat, const VectorsType& vectors, const CoeffsType& hCoeffs, bool forward)
{
typedef typename MatrixType::Index Index;
enum { TFactorSize = MatrixType::ColsAtCompileTime };
Index nbVecs = vectors.cols();
Matrix<typename MatrixType::Scalar, TFactorSize, TFactorSize, ColMajor> T(nbVecs,nbVecs);
make_block_householder_triangular_factor(T, vectors, hCoeffs);
const TriangularView<const VectorsType, UnitLower>& V(vectors);
Matrix<typename MatrixType::Scalar, TFactorSize, TFactorSize, RowMajor> T(nbVecs,nbVecs);
if(forward) make_block_householder_triangular_factor(T, vectors, hCoeffs);
else make_block_householder_triangular_factor(T, vectors, hCoeffs.conjugate());
const TriangularView<const VectorsType, UnitLower> V(vectors);
// A -= V T V^* A
Matrix<typename MatrixType::Scalar,VectorsType::ColsAtCompileTime,MatrixType::ColsAtCompileTime,0,
VectorsType::MaxColsAtCompileTime,MatrixType::MaxColsAtCompileTime> tmp = V.adjoint() * mat;
// FIXME add .noalias() once the triangular product can work inplace
tmp = T.template triangularView<Upper>().adjoint() * tmp;
if(forward) tmp = T.template triangularView<Upper>() * tmp;
else tmp = T.template triangularView<Upper>().adjoint() * tmp;
mat.noalias() -= V * tmp;
}

43
Eigen/src/Householder/HouseholderSequence.h
View File

@ -73,6 +73,15 @@ struct traits<HouseholderSequence<VectorsType,CoeffsType,Side> >
};
};
struct HouseholderSequenceShape {};
template<typename VectorsType, typename CoeffsType, int Side>
struct evaluator_traits<HouseholderSequence<VectorsType,CoeffsType,Side> >
: public evaluator_traits_base<HouseholderSequence<VectorsType,CoeffsType,Side> >
{
typedef HouseholderSequenceShape Shape;
};
template<typename VectorsType, typename CoeffsType, int Side>
struct hseq_side_dependent_impl
{
@ -307,12 +316,36 @@ template<typename VectorsType, typename CoeffsType, int Side> class HouseholderS
template<typename Dest, typename Workspace>
inline void applyThisOnTheLeft(Dest& dst, Workspace& workspace) const
{
workspace.resize(dst.cols());
for(Index k = 0; k < m_length; ++k)
const Index BlockSize = 48;
// if the entries are large enough, then apply the reflectors by block
if(m_length>=BlockSize && dst.cols()>1)
{
Index actual_k = m_trans ? k : m_length-k-1;
dst.bottomRows(rows()-m_shift-actual_k)
.applyHouseholderOnTheLeft(essentialVector(actual_k), m_coeffs.coeff(actual_k), workspace.data());
for(Index i = 0; i < m_length; i+=BlockSize)
{
Index end = m_trans ? (std::min)(m_length,i+BlockSize) : m_length-i;
Index k = m_trans ? i : (std::max)(Index(0),end-BlockSize);
Index bs = end-k;
Index start = k + m_shift;
typedef Block<typename internal::remove_all<VectorsType>::type,Dynamic,Dynamic> SubVectorsType;
SubVectorsType sub_vecs1(m_vectors.const_cast_derived(), Side==OnTheRight ? k : start,
Side==OnTheRight ? start : k,
Side==OnTheRight ? bs : m_vectors.rows()-start,
Side==OnTheRight ? m_vectors.cols()-start : bs);
typename internal::conditional<Side==OnTheRight, Transpose<SubVectorsType>, SubVectorsType&>::type sub_vecs(sub_vecs1);
Block<Dest,Dynamic,Dynamic> sub_dst(dst,dst.rows()-rows()+m_shift+k,0, rows()-m_shift-k,dst.cols());
apply_block_householder_on_the_left(sub_dst, sub_vecs, m_coeffs.segment(k, bs), !m_trans);
}
}
else
{
workspace.resize(dst.cols());
for(Index k = 0; k < m_length; ++k)
{
Index actual_k = m_trans ? k : m_length-k-1;
dst.bottomRows(rows()-m_shift-actual_k)
.applyHouseholderOnTheLeft(essentialVector(actual_k), m_coeffs.coeff(actual_k), workspace.data());
}
}
}

25
Eigen/src/IterativeLinearSolvers/BasicPreconditioners.h
View File

@ -1,7 +1,7 @@
// This file is part of Eigen, a lightweight C++ template library
// for linear algebra.
//
// Copyright (C) 2011 Gael Guennebaud <gael.guennebaud@inria.fr>
// Copyright (C) 2011-2014 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
@ -80,19 +80,20 @@ class DiagonalPreconditioner
return factorize(mat);
}
/** \internal */
template<typename Rhs, typename Dest>
void _solve(const Rhs& b, Dest& x) const
void _solve_impl(const Rhs& b, Dest& x) const
{
x = m_invdiag.array() * b.array() ;
}
template<typename Rhs> inline const internal::solve_retval<DiagonalPreconditioner, Rhs>
template<typename Rhs> inline const Solve<DiagonalPreconditioner, Rhs>
solve(const MatrixBase<Rhs>& b) const
{
eigen_assert(m_isInitialized && "DiagonalPreconditioner is not initialized.");
eigen_assert(m_invdiag.size()==b.rows()
&& "DiagonalPreconditioner::solve(): invalid number of rows of the right hand side matrix b");
return internal::solve_retval<DiagonalPreconditioner, Rhs>(*this, b.derived());
return Solve<DiagonalPreconditioner, Rhs>(*this, b.derived());
}
protected:
@ -100,22 +101,6 @@ class DiagonalPreconditioner
bool m_isInitialized;
};
namespace internal {
template<typename _MatrixType, typename Rhs>
struct solve_retval<DiagonalPreconditioner<_MatrixType>, Rhs>
: solve_retval_base<DiagonalPreconditioner<_MatrixType>, Rhs>
{
typedef DiagonalPreconditioner<_MatrixType> Dec;
EIGEN_MAKE_SOLVE_HELPERS(Dec,Rhs)
template<typename Dest> void evalTo(Dest& dst) const
{
dec()._solve(rhs(),dst);
}
};
}
/** \ingroup IterativeLinearSolvers_Module
* \brief A naive preconditioner which approximates any matrix as the identity matrix

49
Eigen/src/IterativeLinearSolvers/BiCGSTAB.h
View File

@ -1,7 +1,7 @@
// This file is part of Eigen, a lightweight C++ template library
// for linear algebra.
//
// Copyright (C) 2011 Gael Guennebaud <gael.guennebaud@inria.fr>
// Copyright (C) 2011-2014 Gael Guennebaud <gael.guennebaud@inria.fr>
// Copyright (C) 2012 Désiré Nuentsa-Wakam <desire.nuentsa_wakam@inria.fr>
//
// This Source Code Form is subject to the terms of the Mozilla
@ -181,26 +181,10 @@ public:
BiCGSTAB(const MatrixType& A) : Base(A) {}
~BiCGSTAB() {}
/** \returns the solution x of \f$ A x = b \f$ using the current decomposition of A
* \a x0 as an initial solution.
*
* \sa compute()
*/
template<typename Rhs,typename Guess>
inline const internal::solve_retval_with_guess<BiCGSTAB, Rhs, Guess>
solveWithGuess(const MatrixBase<Rhs>& b, const Guess& x0) const
{
eigen_assert(m_isInitialized && "BiCGSTAB is not initialized.");
eigen_assert(Base::rows()==b.rows()
&& "BiCGSTAB::solve(): invalid number of rows of the right hand side matrix b");
return internal::solve_retval_with_guess
<BiCGSTAB, Rhs, Guess>(*this, b.derived(), x0);
}
/** \internal */
template<typename Rhs,typename Dest>
void _solveWithGuess(const Rhs& b, Dest& x) const
void _solve_with_guess_impl(const Rhs& b, Dest& x) const
{
bool failed = false;
for(int j=0; j<b.cols(); ++j)
@ -219,36 +203,19 @@ public:
}
/** \internal */
using Base::_solve_impl;
template<typename Rhs,typename Dest>
void _solve(const Rhs& b, Dest& x) const
void _solve_impl(const MatrixBase<Rhs>& b, Dest& x) const
{
// x.setZero();
x = b;
_solveWithGuess(b,x);
// x.setZero();
x = b;
_solve_with_guess_impl(b,x);
}
protected:
};
namespace internal {
template<typename _MatrixType, typename _Preconditioner, typename Rhs>
struct solve_retval<BiCGSTAB<_MatrixType, _Preconditioner>, Rhs>
: solve_retval_base<BiCGSTAB<_MatrixType, _Preconditioner>, Rhs>
{
typedef BiCGSTAB<_MatrixType, _Preconditioner> Dec;
EIGEN_MAKE_SOLVE_HELPERS(Dec,Rhs)
template<typename Dest> void evalTo(Dest& dst) const
{
dec()._solve(rhs(),dst);
}
};
} // end namespace internal
} // end namespace Eigen
#endif // EIGEN_BICGSTAB_H

Some files were not shown because too many files have changed in this diff Show More