Commit a6547fce authored by Yaroslav Gevorkov's avatar Yaroslav Gevorkov
Browse files

eigen update

parent 6caeec18
......@@ -12,7 +12,7 @@ if(NOT CMAKE_BUILD_TYPE AND NOT CMAKE_CONFIGURATION_TYPES)
"MinSizeRel" "RelWithDebInfo")
endif()
find_package(Eigen3 3.3.4 NO_MODULE)
find_package(Eigen3 3.3.7 NO_MODULE)
find_package(OpenMP)
include_directories(include)
......
include(RegexUtils)
test_escape_string_as_regex()
file(GLOB Eigen_directory_files "*")
escape_string_as_regex(ESCAPED_CMAKE_CURRENT_SOURCE_DIR "${CMAKE_CURRENT_SOURCE_DIR}")
foreach(f ${Eigen_directory_files})
if(NOT f MATCHES "\\.txt" AND NOT f MATCHES "${ESCAPED_CMAKE_CURRENT_SOURCE_DIR}/[.].+" AND NOT f MATCHES "${ESCAPED_CMAKE_CURRENT_SOURCE_DIR}/src")
list(APPEND Eigen_directory_files_to_install ${f})
endif()
endforeach(f ${Eigen_directory_files})
install(FILES
${Eigen_directory_files_to_install}
DESTINATION ${INCLUDE_INSTALL_DIR}/Eigen COMPONENT Devel
)
install(DIRECTORY src DESTINATION ${INCLUDE_INSTALL_DIR}/Eigen COMPONENT Devel FILES_MATCHING PATTERN "*.h")
......@@ -9,6 +9,7 @@
#define EIGEN_CHOLESKY_MODULE_H
#include "Core"
#include "Jacobi"
#include "src/Core/util/DisableStupidWarnings.h"
......@@ -31,7 +32,11 @@
#include "src/Cholesky/LLT.h"
#include "src/Cholesky/LDLT.h"
#ifdef EIGEN_USE_LAPACKE
#ifdef EIGEN_USE_MKL
#include "mkl_lapacke.h"
#else
#include "src/misc/lapacke.h"
#endif
#include "src/Cholesky/LLT_LAPACKE.h"
#endif
......
......@@ -14,6 +14,22 @@
// first thing Eigen does: stop the compiler from committing suicide
#include "src/Core/util/DisableStupidWarnings.h"
#if defined(__CUDACC__) && !defined(EIGEN_NO_CUDA)
#define EIGEN_CUDACC __CUDACC__
#endif
#if defined(__CUDA_ARCH__) && !defined(EIGEN_NO_CUDA)
#define EIGEN_CUDA_ARCH __CUDA_ARCH__
#endif
#if defined(__CUDACC_VER_MAJOR__) && (__CUDACC_VER_MAJOR__ >= 9)
#define EIGEN_CUDACC_VER ((__CUDACC_VER_MAJOR__ * 10000) + (__CUDACC_VER_MINOR__ * 100))
#elif defined(__CUDACC_VER__)
#define EIGEN_CUDACC_VER __CUDACC_VER__
#else
#define EIGEN_CUDACC_VER 0
#endif
// Handle NVCC/CUDA/SYCL
#if defined(__CUDACC__) || defined(__SYCL_DEVICE_ONLY__)
// Do not try asserts on CUDA and SYCL!
......@@ -37,9 +53,9 @@
#endif
#define EIGEN_DEVICE_FUNC __host__ __device__
// We need math_functions.hpp to ensure that that EIGEN_USING_STD_MATH macro
// We need cuda_runtime.h to ensure that that EIGEN_USING_STD_MATH macro
// works properly on the device side
#include <math_functions.hpp>
#include <cuda_runtime.h>
#else
#define EIGEN_DEVICE_FUNC
#endif
......@@ -155,6 +171,9 @@
#ifdef __AVX512DQ__
#define EIGEN_VECTORIZE_AVX512DQ
#endif
#ifdef __AVX512ER__
#define EIGEN_VECTORIZE_AVX512ER
#endif
#endif
// include files
......@@ -229,7 +248,7 @@
#if defined __CUDACC__
#define EIGEN_VECTORIZE_CUDA
#include <vector_types.h>
#if defined __CUDACC_VER__ && __CUDACC_VER__ >= 70500
#if EIGEN_CUDACC_VER >= 70500
#define EIGEN_HAS_CUDA_FP16
#endif
#endif
......@@ -352,6 +371,7 @@ using std::ptrdiff_t;
#include "src/Core/MathFunctions.h"
#include "src/Core/GenericPacketMath.h"
#include "src/Core/MathFunctionsImpl.h"
#include "src/Core/arch/Default/ConjHelper.h"
#if defined EIGEN_VECTORIZE_AVX512
#include "src/Core/arch/SSE/PacketMath.h"
......@@ -367,6 +387,7 @@ using std::ptrdiff_t;
#include "src/Core/arch/AVX/MathFunctions.h"
#include "src/Core/arch/AVX/Complex.h"
#include "src/Core/arch/AVX/TypeCasting.h"
#include "src/Core/arch/SSE/TypeCasting.h"
#elif defined EIGEN_VECTORIZE_SSE
#include "src/Core/arch/SSE/PacketMath.h"
#include "src/Core/arch/SSE/MathFunctions.h"
......
......@@ -45,7 +45,11 @@
#include "src/Eigenvalues/GeneralizedEigenSolver.h"
#include "src/Eigenvalues/MatrixBaseEigenvalues.h"
#ifdef EIGEN_USE_LAPACKE
#ifdef EIGEN_USE_MKL
#include "mkl_lapacke.h"
#else
#include "src/misc/lapacke.h"
#endif
#include "src/Eigenvalues/RealSchur_LAPACKE.h"
#include "src/Eigenvalues/ComplexSchur_LAPACKE.h"
#include "src/Eigenvalues/SelfAdjointEigenSolver_LAPACKE.h"
......
......@@ -28,7 +28,11 @@
#include "src/LU/FullPivLU.h"
#include "src/LU/PartialPivLU.h"
#ifdef EIGEN_USE_LAPACKE
#ifdef EIGEN_USE_MKL
#include "mkl_lapacke.h"
#else
#include "src/misc/lapacke.h"
#endif
#include "src/LU/PartialPivLU_LAPACKE.h"
#endif
#include "src/LU/Determinant.h"
......
......@@ -36,7 +36,11 @@
#include "src/QR/ColPivHouseholderQR.h"
#include "src/QR/CompleteOrthogonalDecomposition.h"
#ifdef EIGEN_USE_LAPACKE
#ifdef EIGEN_USE_MKL
#include "mkl_lapacke.h"
#else
#include "src/misc/lapacke.h"
#endif
#include "src/QR/HouseholderQR_LAPACKE.h"
#include "src/QR/ColPivHouseholderQR_LAPACKE.h"
#endif
......
......@@ -27,7 +27,7 @@ void qFree(void *ptr)
void *qRealloc(void *ptr, std::size_t size)
{
void* newPtr = Eigen::internal::aligned_malloc(size);
memcpy(newPtr, ptr, size);
std::memcpy(newPtr, ptr, size);
Eigen::internal::aligned_free(ptr);
return newPtr;
}
......
......@@ -37,7 +37,11 @@
#include "src/SVD/JacobiSVD.h"
#include "src/SVD/BDCSVD.h"
#if defined(EIGEN_USE_LAPACKE) && !defined(EIGEN_USE_LAPACKE_STRICT)
#ifdef EIGEN_USE_MKL
#include "mkl_lapacke.h"
#else
#include "src/misc/lapacke.h"
#endif
#include "src/SVD/JacobiSVD_LAPACKE.h"
#endif
......
......@@ -248,7 +248,7 @@ template<typename _MatrixType, int _UpLo> class LDLT
/** \brief Reports whether previous computation was successful.
*
* \returns \c Success if computation was succesful,
* \c NumericalIssue if the matrix.appears to be negative.
* \c NumericalIssue if the factorization failed because of a zero pivot.
*/
ComputationInfo info() const
{
......@@ -305,7 +305,8 @@ template<> struct ldlt_inplace<Lower>
if (size <= 1)
{
transpositions.setIdentity();
if (numext::real(mat.coeff(0,0)) > static_cast<RealScalar>(0) ) sign = PositiveSemiDef;
if(size==0) sign = ZeroSign;
else if (numext::real(mat.coeff(0,0)) > static_cast<RealScalar>(0) ) sign = PositiveSemiDef;
else if (numext::real(mat.coeff(0,0)) < static_cast<RealScalar>(0)) sign = NegativeSemiDef;
else sign = ZeroSign;
return true;
......@@ -376,6 +377,8 @@ template<> struct ldlt_inplace<Lower>
if((rs>0) && pivot_is_valid)
A21 /= realAkk;
else if(rs>0)
ret = ret && (A21.array()==Scalar(0)).all();
if(found_zero_pivot && pivot_is_valid) ret = false; // factorization failed
else if(!pivot_is_valid) found_zero_pivot = true;
......@@ -568,13 +571,14 @@ void LDLT<_MatrixType,_UpLo>::_solve_impl(const RhsType &rhs, DstType &dst) cons
// more precisely, use pseudo-inverse of D (see bug 241)
using std::abs;
const typename Diagonal<const MatrixType>::RealReturnType vecD(vectorD());
// In some previous versions, tolerance was set to the max of 1/highest and the maximal diagonal entry * epsilon
// as motivated by LAPACK's xGELSS:
// In some previous versions, tolerance was set to the max of 1/highest (or rather numeric_limits::min())
// and the maximal diagonal entry * epsilon as motivated by LAPACK's xGELSS:
// RealScalar tolerance = numext::maxi(vecD.array().abs().maxCoeff() * NumTraits<RealScalar>::epsilon(),RealScalar(1) / NumTraits<RealScalar>::highest());
// However, LDLT is not rank revealing, and so adjusting the tolerance wrt to the highest
// diagonal element is not well justified and leads to numerical issues in some cases.
// Moreover, Lapack's xSYTRS routines use 0 for the tolerance.
RealScalar tolerance = RealScalar(1) / NumTraits<RealScalar>::highest();
// Using numeric_limits::min() gives us more robustness to denormals.
RealScalar tolerance = (std::numeric_limits<RealScalar>::min)();
for (Index i = 0; i < vecD.size(); ++i)
{
......
......@@ -24,7 +24,7 @@ template<typename MatrixType, int UpLo> struct LLT_Traits;
*
* \tparam _MatrixType the type of the matrix of which we are computing the LL^T Cholesky decomposition
* \tparam _UpLo the triangular part that will be used for the decompositon: Lower (default) or Upper.
* The other triangular part won't be read.
* The other triangular part won't be read.
*
* This class performs a LL^T Cholesky decomposition of a symmetric, positive definite
* matrix A such that A = LL^* = U^*U, where L is lower triangular.
......@@ -41,14 +41,18 @@ template<typename MatrixType, int UpLo> struct LLT_Traits;
* Example: \include LLT_example.cpp
* Output: \verbinclude LLT_example.out
*
* \b Performance: for best performance, it is recommended to use a column-major storage format
* with the Lower triangular part (the default), or, equivalently, a row-major storage format
* with the Upper triangular part. Otherwise, you might get a 20% slowdown for the full factorization
* step, and rank-updates can be up to 3 times slower.
*
* This class supports the \link InplaceDecomposition inplace decomposition \endlink mechanism.
*
* Note that during the decomposition, only the lower (or upper, as defined by _UpLo) triangular part of A is considered.
* Therefore, the strict lower part does not have to store correct values.
*
* \sa MatrixBase::llt(), SelfAdjointView::llt(), class LDLT
*/
/* HEY THIS DOX IS DISABLED BECAUSE THERE's A BUG EITHER HERE OR IN LDLT ABOUT THAT (OR BOTH)
* Note that during the decomposition, only the upper triangular part of A is considered. Therefore,
* the strict lower part does not have to store correct values.
*/
template<typename _MatrixType, int _UpLo> class LLT
{
public:
......@@ -146,7 +150,7 @@ template<typename _MatrixType, int _UpLo> class LLT
}
template<typename Derived>
void solveInPlace(MatrixBase<Derived> &bAndX) const;
void solveInPlace(const MatrixBase<Derived> &bAndX) const;
template<typename InputType>
LLT& compute(const EigenBase<InputType>& matrix);
......@@ -177,7 +181,7 @@ template<typename _MatrixType, int _UpLo> class LLT
/** \brief Reports whether previous computation was successful.
*
* \returns \c Success if computation was succesful,
* \c NumericalIssue if the matrix.appears to be negative.
* \c NumericalIssue if the matrix.appears not to be positive definite.
*/
ComputationInfo info() const
{
......@@ -425,7 +429,8 @@ LLT<MatrixType,_UpLo>& LLT<MatrixType,_UpLo>::compute(const EigenBase<InputType>
eigen_assert(a.rows()==a.cols());
const Index size = a.rows();
m_matrix.resize(size, size);
m_matrix = a.derived();
if (!internal::is_same_dense(m_matrix, a.derived()))
m_matrix = a.derived();
// Compute matrix L1 norm = max abs column sum.
m_l1_norm = RealScalar(0);
......@@ -485,11 +490,14 @@ void LLT<_MatrixType,_UpLo>::_solve_impl(const RhsType &rhs, DstType &dst) const
*
* This version avoids a copy when the right hand side matrix b is not needed anymore.
*
* \warning The parameter is only marked 'const' to make the C++ compiler accept a temporary expression here.
* This function will const_cast it, so constness isn't honored here.
*
* \sa LLT::solve(), MatrixBase::llt()
*/
template<typename MatrixType, int _UpLo>
template<typename Derived>
void LLT<MatrixType,_UpLo>::solveInPlace(MatrixBase<Derived> &bAndX) const
void LLT<MatrixType,_UpLo>::solveInPlace(const MatrixBase<Derived> &bAndX) const
{
eigen_assert(m_isInitialized && "LLT is not initialized.");
eigen_assert(m_matrix.rows()==bAndX.rows());
......
......@@ -153,8 +153,6 @@ class Array
: Base(std::move(other))
{
Base::_check_template_params();
if (RowsAtCompileTime!=Dynamic && ColsAtCompileTime!=Dynamic)
Base::_set_noalias(other);
}
EIGEN_DEVICE_FUNC
Array& operator=(Array&& other) EIGEN_NOEXCEPT_IF(std::is_nothrow_move_assignable<Scalar>::value)
......
......@@ -39,7 +39,7 @@ public:
enum {
DstAlignment = DstEvaluator::Alignment,
SrcAlignment = SrcEvaluator::Alignment,
DstHasDirectAccess = DstFlags & DirectAccessBit,
DstHasDirectAccess = (DstFlags & DirectAccessBit) == DirectAccessBit,
JointAlignment = EIGEN_PLAIN_ENUM_MIN(DstAlignment,SrcAlignment)
};
......@@ -83,7 +83,7 @@ private:
&& int(OuterStride)!=Dynamic && int(OuterStride)%int(InnerPacketSize)==0
&& (EIGEN_UNALIGNED_VECTORIZE || int(JointAlignment)>=int(InnerRequiredAlignment)),
MayLinearize = bool(StorageOrdersAgree) && (int(DstFlags) & int(SrcFlags) & LinearAccessBit),
MayLinearVectorize = bool(MightVectorize) && MayLinearize && DstHasDirectAccess
MayLinearVectorize = bool(MightVectorize) && bool(MayLinearize) && bool(DstHasDirectAccess)
&& (EIGEN_UNALIGNED_VECTORIZE || (int(DstAlignment)>=int(LinearRequiredAlignment)) || MaxSizeAtCompileTime == Dynamic),
/* If the destination isn't aligned, we have to do runtime checks and we don't unroll,
so it's only good for large enough sizes. */
......
......@@ -84,7 +84,8 @@ class vml_assign_traits
struct Assignment<DstXprType, CwiseUnaryOp<scalar_##EIGENOP##_op<EIGENTYPE>, SrcXprNested>, assign_op<EIGENTYPE,EIGENTYPE>, \
Dense2Dense, typename enable_if<vml_assign_traits<DstXprType,SrcXprNested>::EnableVml>::type> { \
typedef CwiseUnaryOp<scalar_##EIGENOP##_op<EIGENTYPE>, SrcXprNested> SrcXprType; \
static void run(DstXprType &dst, const SrcXprType &src, const assign_op<EIGENTYPE,EIGENTYPE> &/*func*/) { \
static void run(DstXprType &dst, const SrcXprType &src, const assign_op<EIGENTYPE,EIGENTYPE> &func) { \
resize_if_allowed(dst, src, func); \
eigen_assert(dst.rows() == src.rows() && dst.cols() == src.cols()); \
if(vml_assign_traits<DstXprType,SrcXprNested>::Traversal==LinearTraversal) { \
VMLOP(dst.size(), (const VMLTYPE*)src.nestedExpression().data(), \
......@@ -144,7 +145,8 @@ EIGEN_MKL_VML_DECLARE_UNARY_CALLS_REAL(ceil, Ceil, _)
Dense2Dense, typename enable_if<vml_assign_traits<DstXprType,SrcXprNested>::EnableVml>::type> { \
typedef CwiseBinaryOp<scalar_##EIGENOP##_op<EIGENTYPE,EIGENTYPE>, SrcXprNested, \
const CwiseNullaryOp<internal::scalar_constant_op<EIGENTYPE>,Plain> > SrcXprType; \
static void run(DstXprType &dst, const SrcXprType &src, const assign_op<EIGENTYPE,EIGENTYPE> &/*func*/) { \
static void run(DstXprType &dst, const SrcXprType &src, const assign_op<EIGENTYPE,EIGENTYPE> &func) { \
resize_if_allowed(dst, src, func); \
eigen_assert(dst.rows() == src.rows() && dst.cols() == src.cols()); \
VMLTYPE exponent = reinterpret_cast<const VMLTYPE&>(src.rhs().functor().m_other); \
if(vml_assign_traits<DstXprType,SrcXprNested>::Traversal==LinearTraversal) \
......
......@@ -160,7 +160,7 @@ rcond_estimate_helper(typename Decomposition::RealScalar matrix_norm, const Deco
{
typedef typename Decomposition::RealScalar RealScalar;
eigen_assert(dec.rows() == dec.cols());
if (dec.rows() == 0) return RealScalar(1);
if (dec.rows() == 0) return NumTraits<RealScalar>::infinity();
if (matrix_norm == RealScalar(0)) return RealScalar(0);
if (dec.rows() == 1) return RealScalar(1);
const RealScalar inverse_matrix_norm = rcond_invmatrix_L1_norm_estimate(dec);
......
......@@ -977,7 +977,7 @@ struct evaluator<Block<ArgType, BlockRows, BlockCols, InnerPanel> >
OuterStrideAtCompileTime = HasSameStorageOrderAsArgType
? int(outer_stride_at_compile_time<ArgType>::ret)
: int(inner_stride_at_compile_time<ArgType>::ret),
MaskPacketAccessBit = (InnerStrideAtCompileTime == 1) ? PacketAccessBit : 0,
MaskPacketAccessBit = (InnerStrideAtCompileTime == 1 || HasSameStorageOrderAsArgType) ? PacketAccessBit : 0,
FlagsLinearAccessBit = (RowsAtCompileTime == 1 || ColsAtCompileTime == 1 || (InnerPanel && (evaluator<ArgType>::Flags&LinearAccessBit))) ? LinearAccessBit : 0,
FlagsRowMajorBit = XprType::Flags&RowMajorBit,
......@@ -987,7 +987,9 @@ struct evaluator<Block<ArgType, BlockRows, BlockCols, InnerPanel> >
Flags = Flags0 | FlagsLinearAccessBit | FlagsRowMajorBit,
PacketAlignment = unpacket_traits<PacketScalar>::alignment,
Alignment0 = (InnerPanel && (OuterStrideAtCompileTime!=Dynamic) && (((OuterStrideAtCompileTime * int(sizeof(Scalar))) % int(PacketAlignment)) == 0)) ? int(PacketAlignment) : 0,
Alignment0 = (InnerPanel && (OuterStrideAtCompileTime!=Dynamic)
&& (OuterStrideAtCompileTime!=0)
&& (((OuterStrideAtCompileTime * int(sizeof(Scalar))) % int(PacketAlignment)) == 0)) ? int(PacketAlignment) : 0,
Alignment = EIGEN_PLAIN_ENUM_MIN(evaluator<ArgType>::Alignment, Alignment0)
};
typedef block_evaluator<ArgType, BlockRows, BlockCols, InnerPanel> block_evaluator_type;
......@@ -1018,14 +1020,16 @@ struct unary_evaluator<Block<ArgType, BlockRows, BlockCols, InnerPanel>, IndexBa
EIGEN_DEVICE_FUNC explicit unary_evaluator(const XprType& block)
: m_argImpl(block.nestedExpression()),
m_startRow(block.startRow()),
m_startCol(block.startCol())
m_startCol(block.startCol()),
m_linear_offset(InnerPanel?(XprType::IsRowMajor ? block.startRow()*block.cols() : block.startCol()*block.rows()):0)
{ }
typedef typename XprType::Scalar Scalar;
typedef typename XprType::CoeffReturnType CoeffReturnType;
enum {
RowsAtCompileTime = XprType::RowsAtCompileTime
RowsAtCompileTime = XprType::RowsAtCompileTime,
ForwardLinearAccess = InnerPanel && bool(evaluator<ArgType>::Flags&LinearAccessBit)
};
EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
......@@ -1037,7 +1041,10 @@ struct unary_evaluator<Block<ArgType, BlockRows, BlockCols, InnerPanel>, IndexBa
EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
CoeffReturnType coeff(Index index) const
{
return coeff(RowsAtCompileTime == 1 ? 0 : index, RowsAtCompileTime == 1 ? index : 0);
if (ForwardLinearAccess)
return m_argImpl.coeff(m_linear_offset.value() + index);
else
return coeff(RowsAtCompileTime == 1 ? 0 : index, RowsAtCompileTime == 1 ? index : 0);
}
EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
......@@ -1049,7 +1056,10 @@ struct unary_evaluator<Block<ArgType, BlockRows, BlockCols, InnerPanel>, IndexBa
EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
Scalar& coeffRef(Index index)
{
return coeffRef(RowsAtCompileTime == 1 ? 0 : index, RowsAtCompileTime == 1 ? index : 0);
if (ForwardLinearAccess)
return m_argImpl.coeffRef(m_linear_offset.value() + index);
else
return coeffRef(RowsAtCompileTime == 1 ? 0 : index, RowsAtCompileTime == 1 ? index : 0);
}
template<int LoadMode, typename PacketType>
......@@ -1063,8 +1073,11 @@ struct unary_evaluator<Block<ArgType, BlockRows, BlockCols, InnerPanel>, IndexBa
EIGEN_STRONG_INLINE
PacketType packet(Index index) const
{
return packet<LoadMode,PacketType>(RowsAtCompileTime == 1 ? 0 : index,
RowsAtCompileTime == 1 ? index : 0);
if (ForwardLinearAccess)
return m_argImpl.template packet<LoadMode,PacketType>(m_linear_offset.value() + index);
else
return packet<LoadMode,PacketType>(RowsAtCompileTime == 1 ? 0 : index,
RowsAtCompileTime == 1 ? index : 0);
}
template<int StoreMode, typename PacketType>
......@@ -1078,15 +1091,19 @@ struct unary_evaluator<Block<ArgType, BlockRows, BlockCols, InnerPanel>, IndexBa
EIGEN_STRONG_INLINE
void writePacket(Index index, const PacketType& x)
{
return writePacket<StoreMode,PacketType>(RowsAtCompileTime == 1 ? 0 : index,
RowsAtCompileTime == 1 ? index : 0,
x);
if (ForwardLinearAccess)
return m_argImpl.template writePacket<StoreMode,PacketType>(m_linear_offset.value() + index, x);
else
return writePacket<StoreMode,PacketType>(RowsAtCompileTime == 1 ? 0 : index,
RowsAtCompileTime == 1 ? index : 0,
x);
}
protected:
evaluator<ArgType> m_argImpl;
const variable_if_dynamic<Index, (ArgType::RowsAtCompileTime == 1 && BlockRows==1) ? 0 : Dynamic> m_startRow;
const variable_if_dynamic<Index, (ArgType::ColsAtCompileTime == 1 && BlockCols==1) ? 0 : Dynamic> m_startCol;
const variable_if_dynamic<Index, InnerPanel ? Dynamic : 0> m_linear_offset;
};
// TODO: This evaluator does not actually use the child evaluator;
......
......@@ -70,7 +70,10 @@ template<typename MatrixType, int _DiagIndex> class Diagonal
EIGEN_DENSE_PUBLIC_INTERFACE(Diagonal)
EIGEN_DEVICE_FUNC
explicit inline Diagonal(MatrixType& matrix, Index a_index = DiagIndex) : m_matrix(matrix), m_index(a_index) {}
explicit inline Diagonal(MatrixType& matrix, Index a_index = DiagIndex) : m_matrix(matrix), m_index(a_index)
{
eigen_assert( a_index <= m_matrix.cols() && -a_index <= m_matrix.rows() );
}
EIGEN_INHERIT_ASSIGNMENT_OPERATORS(Diagonal)
......
......@@ -31,7 +31,8 @@ struct dot_nocheck
typedef scalar_conj_product_op<typename traits<T>::Scalar,typename traits<U>::Scalar> conj_prod;
typedef typename conj_prod::result_type ResScalar;
EIGEN_DEVICE_FUNC
static inline ResScalar run(const MatrixBase<T>& a, const MatrixBase<U>& b)
EIGEN_STRONG_INLINE
static ResScalar run(const MatrixBase<T>& a, const MatrixBase<U>& b)
{
return a.template binaryExpr<conj_prod>(b).sum();
}
......@@ -43,7 +44,8 @@ struct dot_nocheck<T, U, true>
typedef scalar_conj_product_op<typename traits<T>::Scalar,typename traits<U>::Scalar> conj_prod;
typedef typename conj_prod::result_type ResScalar;
EIGEN_DEVICE_FUNC
static inline ResScalar run(const MatrixBase<T>& a, const MatrixBase<U>& b)
EIGEN_STRONG_INLINE
static ResScalar run(const MatrixBase<T>& a, const MatrixBase<U>& b)
{
return a.transpose().template binaryExpr<conj_prod>(b).sum();
}
......@@ -65,6 +67,7 @@ struct dot_nocheck<T, U, true>
template<typename Derived>
template<typename OtherDerived>
EIGEN_DEVICE_FUNC
EIGEN_STRONG_INLINE
typename ScalarBinaryOpTraits<typename internal::traits<Derived>::Scalar,typename internal::traits<OtherDerived>::Scalar>::ReturnType
MatrixBase<Derived>::dot(const MatrixBase<OtherDerived>& other) const
{
......@@ -102,7 +105,7 @@ EIGEN_STRONG_INLINE typename NumTraits<typename internal::traits<Derived>::Scala
* \sa lpNorm(), dot(), squaredNorm()
*/
template<typename Derived>
inline typename NumTraits<typename internal::traits<Derived>::Scalar>::Real MatrixBase<Derived>::norm() const
EIGEN_STRONG_INLINE typename NumTraits<typename internal::traits<Derived>::Scalar>::Real MatrixBase<Derived>::norm() const
{
return numext::sqrt(squaredNorm());
}
......@@ -117,7 +120,7 @@ inline typename NumTraits<typename internal::traits<Derived>::Scalar>::Real Matr
* \sa norm(), normalize()
*/
template<typename Derived>
inline const typename MatrixBase<Derived>::PlainObject
EIGEN_STRONG_INLINE const typename MatrixBase<Derived>::PlainObject
MatrixBase<Derived>::normalized() const
{
typedef typename internal::nested_eval<Derived,2>::type _Nested;
......@@ -139,7 +142,7 @@ MatrixBase<Derived>::normalized() const
* \sa norm(), normalized()
*/
template<typename Derived>
inline void MatrixBase<Derived>::normalize()
EIGEN_STRONG_INLINE void MatrixBase<Derived>::normalize()
{
RealScalar z = squaredNorm();
// NOTE: after extensive benchmarking, this conditional does not impact performance, at least on recent x86 CPU
......@@ -160,7 +163,7 @@ inline void MatrixBase<Derived>::normalize()
* \sa stableNorm(), stableNormalize(), normalized()
*/
template<typename Derived>
inline const typename MatrixBase<Derived>::PlainObject
EIGEN_STRONG_INLINE const typename MatrixBase<Derived>::PlainObject
MatrixBase<Derived>::stableNormalized() const
{
typedef typename internal::nested_eval<Derived,3>::type _Nested;
......@@ -185,7 +188,7 @@ MatrixBase<Derived>::stableNormalized() const
* \sa stableNorm(), stableNormalized(), normalize()
*/
template<typename Derived>
inline void MatrixBase<Derived>::stableNormalize()
EIGEN_STRONG_INLINE void MatrixBase<Derived>::stableNormalize()
{
RealScalar w = cwiseAbs().maxCoeff();
RealScalar z = (derived()/w).squaredNorm();
......
......@@ -24,12 +24,17 @@ template<int Rows, int Cols, int Depth> struct product_type_selector;
template<int Size, int MaxSize> struct product_size_category
{
enum { is_large = MaxSize == Dynamic ||
Size >= EIGEN_CACHEFRIENDLY_PRODUCT_THRESHOLD ||
(Size==Dynamic && MaxSize>=EIGEN_CACHEFRIENDLY_PRODUCT_THRESHOLD),
value = is_large ? Large
: Size == 1 ? 1
: Small
enum {
#ifndef EIGEN_CUDA_ARCH
is_large = MaxSize == Dynamic ||
Size >= EIGEN_CACHEFRIENDLY_PRODUCT_THRESHOLD ||
(Size==Dynamic && MaxSize>=EIGEN_CACHEFRIENDLY_PRODUCT_THRESHOLD),
#else
is_large = 0,
#endif
value = is_large ? Large
: Size == 1 ? 1
: Small
};
};
......@@ -379,8 +384,6 @@ template<> struct gemv_dense_selector<OnTheRight,RowMajor,false>
*
* \sa lazyProduct(), operator*=(const MatrixBase&), Cwise::operator*()
*/
#ifndef __CUDACC__
template<typename Derived>
template<typename OtherDerived>
inline const Product<Derived, OtherDerived>
......@@ -412,8 +415,6 @@ MatrixBase<Derived>::operator*(const MatrixBase<OtherDerived> &other) const
return Product<Derived, OtherDerived>(derived(), other.derived());
}
#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
......
......@@ -20,11 +20,17 @@ struct traits<Map<PlainObjectType, MapOptions, StrideType> >
{
typedef traits<PlainObjectType> TraitsBase;
enum {
PlainObjectTypeInnerSize = ((traits<PlainObjectType>::Flags&RowMajorBit)==RowMajorBit)
? PlainObjectType::ColsAtCompileTime
: PlainObjectType::RowsAtCompileTime,
InnerStrideAtCompileTime = StrideType::InnerStrideAtCompileTime == 0
? int(PlainObjectType::InnerStrideAtCompileTime)
: int(StrideType::InnerStrideAtCompileTime),
OuterStrideAtCompileTime = StrideType::OuterStrideAtCompileTime == 0
? int(PlainObjectType::OuterStrideAtCompileTime)
? (InnerStrideAtCompileTime==Dynamic || PlainObjectTypeInnerSize==Dynamic
? Dynamic
: int(InnerStrideAtCompileTime) * int(PlainObjectTypeInnerSize))
: int(StrideType::OuterStrideAtCompileTime),
Alignment = int(MapOptions)&int(AlignedMask),
Flags0 = TraitsBase::Flags & (~NestByRefBit),
......@@ -107,10 +113,11 @@ template<typename PlainObjectType, int MapOptions, typename StrideType> class Ma
EIGEN_DEVICE_FUNC
inline Index outerStride() const
{
return StrideType::OuterStrideAtCompileTime != 0 ? m_stride.outer()