mirror of
https://github.com/wpilibsuite/allwpilib
synced 2026-07-02 02:51:42 +00:00
[wpimath] Upgrade to Eigen 3.3.9 (#2910)
It fixes some compilation errors with C++20.
This commit is contained in:
@@ -11,10 +11,6 @@
|
||||
#ifndef EIGEN_CORE_H
|
||||
#define EIGEN_CORE_H
|
||||
|
||||
#if __GNUC__ >= 9
|
||||
#pragma GCC diagnostic ignored "-Wdeprecated-copy"
|
||||
#endif
|
||||
|
||||
// first thing Eigen does: stop the compiler from committing suicide
|
||||
#include "src/Core/util/DisableStupidWarnings.h"
|
||||
|
||||
@@ -283,7 +279,10 @@
|
||||
#include <cmath>
|
||||
#include <cassert>
|
||||
#include <functional>
|
||||
#include <iosfwd>
|
||||
#include <sstream>
|
||||
#ifndef EIGEN_NO_IO
|
||||
#include <iosfwd>
|
||||
#endif
|
||||
#include <cstring>
|
||||
#include <string>
|
||||
#include <limits>
|
||||
@@ -379,7 +378,9 @@ using std::ptrdiff_t;
|
||||
|
||||
#if defined EIGEN_VECTORIZE_AVX512
|
||||
#include "src/Core/arch/SSE/PacketMath.h"
|
||||
#include "src/Core/arch/SSE/MathFunctions.h"
|
||||
#include "src/Core/arch/AVX/PacketMath.h"
|
||||
#include "src/Core/arch/AVX/MathFunctions.h"
|
||||
#include "src/Core/arch/AVX512/PacketMath.h"
|
||||
#include "src/Core/arch/AVX512/MathFunctions.h"
|
||||
#elif defined EIGEN_VECTORIZE_AVX
|
||||
|
||||
@@ -10,12 +10,13 @@
|
||||
|
||||
#include "Core"
|
||||
|
||||
#include "src/Core/util/DisableStupidWarnings.h"
|
||||
|
||||
#include "Cholesky"
|
||||
#include "Jacobi"
|
||||
#include "Householder"
|
||||
#include "LU"
|
||||
// #include "Geometry"
|
||||
|
||||
#include "src/Core/util/DisableStupidWarnings.h"
|
||||
|
||||
/** \defgroup Eigenvalues_Module Eigenvalues module
|
||||
*
|
||||
|
||||
@@ -10,12 +10,12 @@
|
||||
|
||||
#include "Core"
|
||||
|
||||
#include "src/Core/util/DisableStupidWarnings.h"
|
||||
|
||||
#include "Cholesky"
|
||||
#include "Jacobi"
|
||||
#include "Householder"
|
||||
|
||||
#include "src/Core/util/DisableStupidWarnings.h"
|
||||
|
||||
/** \defgroup QR_Module QR module
|
||||
*
|
||||
*
|
||||
|
||||
@@ -40,7 +40,7 @@ static inline void check_DenseIndex_is_signed() {
|
||||
*/
|
||||
template<typename Derived> class DenseBase
|
||||
#ifndef EIGEN_PARSED_BY_DOXYGEN
|
||||
: public DenseCoeffsBase<Derived>
|
||||
: public DenseCoeffsBase<Derived, internal::accessors_level<Derived>::value>
|
||||
#else
|
||||
: public DenseCoeffsBase<Derived,DirectWriteAccessors>
|
||||
#endif // not EIGEN_PARSED_BY_DOXYGEN
|
||||
@@ -71,7 +71,7 @@ template<typename Derived> class DenseBase
|
||||
typedef Scalar value_type;
|
||||
|
||||
typedef typename NumTraits<Scalar>::Real RealScalar;
|
||||
typedef DenseCoeffsBase<Derived> Base;
|
||||
typedef DenseCoeffsBase<Derived, internal::accessors_level<Derived>::value> Base;
|
||||
|
||||
using Base::derived;
|
||||
using Base::const_cast_derived;
|
||||
|
||||
@@ -404,7 +404,7 @@ template<typename T, int _Options> class DenseStorage<T, Dynamic, Dynamic, Dynam
|
||||
if(size != m_rows*m_cols)
|
||||
{
|
||||
internal::conditional_aligned_delete_auto<T,(_Options&DontAlign)==0>(m_data, m_rows*m_cols);
|
||||
if (size)
|
||||
if (size>0) // >0 and not simply !=0 to let the compiler knows that size cannot be negative
|
||||
m_data = internal::conditional_aligned_new_auto<T,(_Options&DontAlign)==0>(size);
|
||||
else
|
||||
m_data = 0;
|
||||
@@ -479,7 +479,7 @@ template<typename T, int _Rows, int _Options> class DenseStorage<T, Dynamic, _Ro
|
||||
if(size != _Rows*m_cols)
|
||||
{
|
||||
internal::conditional_aligned_delete_auto<T,(_Options&DontAlign)==0>(m_data, _Rows*m_cols);
|
||||
if (size)
|
||||
if (size>0) // >0 and not simply !=0 to let the compiler knows that size cannot be negative
|
||||
m_data = internal::conditional_aligned_new_auto<T,(_Options&DontAlign)==0>(size);
|
||||
else
|
||||
m_data = 0;
|
||||
@@ -553,7 +553,7 @@ template<typename T, int _Cols, int _Options> class DenseStorage<T, Dynamic, Dyn
|
||||
if(size != m_rows*_Cols)
|
||||
{
|
||||
internal::conditional_aligned_delete_auto<T,(_Options&DontAlign)==0>(m_data, _Cols*m_rows);
|
||||
if (size)
|
||||
if (size>0) // >0 and not simply !=0 to let the compiler knows that size cannot be negative
|
||||
m_data = internal::conditional_aligned_new_auto<T,(_Options&DontAlign)==0>(size);
|
||||
else
|
||||
m_data = 0;
|
||||
|
||||
@@ -351,10 +351,7 @@ template<typename Packet> EIGEN_DEVICE_FUNC inline Packet preverse(const Packet&
|
||||
/** \internal \returns \a a with real and imaginary part flipped (for complex type only) */
|
||||
template<typename Packet> EIGEN_DEVICE_FUNC inline Packet pcplxflip(const Packet& a)
|
||||
{
|
||||
// FIXME: uncomment the following in case we drop the internal imag and real functions.
|
||||
// using std::imag;
|
||||
// using std::real;
|
||||
return Packet(imag(a),real(a));
|
||||
return Packet(a.imag(),a.real());
|
||||
}
|
||||
|
||||
/**************************
|
||||
@@ -524,10 +521,10 @@ inline void palign(PacketType& first, const PacketType& second)
|
||||
#ifndef __CUDACC__
|
||||
|
||||
template<> inline std::complex<float> pmul(const std::complex<float>& a, const std::complex<float>& b)
|
||||
{ return std::complex<float>(real(a)*real(b) - imag(a)*imag(b), imag(a)*real(b) + real(a)*imag(b)); }
|
||||
{ return std::complex<float>(a.real()*b.real() - a.imag()*b.imag(), a.imag()*b.real() + a.real()*b.imag()); }
|
||||
|
||||
template<> inline std::complex<double> pmul(const std::complex<double>& a, const std::complex<double>& b)
|
||||
{ return std::complex<double>(real(a)*real(b) - imag(a)*imag(b), imag(a)*real(b) + real(a)*imag(b)); }
|
||||
{ return std::complex<double>(a.real()*b.real() - a.imag()*b.imag(), a.imag()*b.real() + a.real()*b.imag()); }
|
||||
|
||||
#endif
|
||||
|
||||
|
||||
@@ -287,7 +287,7 @@ struct abs2_impl_default<Scalar, true> // IsComplex
|
||||
EIGEN_DEVICE_FUNC
|
||||
static inline RealScalar run(const Scalar& x)
|
||||
{
|
||||
return real(x)*real(x) + imag(x)*imag(x);
|
||||
return x.real()*x.real() + x.imag()*x.imag();
|
||||
}
|
||||
};
|
||||
|
||||
@@ -313,14 +313,17 @@ struct abs2_retval
|
||||
****************************************************************************/
|
||||
|
||||
template<typename Scalar, bool IsComplex>
|
||||
struct norm1_default_impl
|
||||
struct norm1_default_impl;
|
||||
|
||||
template<typename Scalar>
|
||||
struct norm1_default_impl<Scalar,true>
|
||||
{
|
||||
typedef typename NumTraits<Scalar>::Real RealScalar;
|
||||
EIGEN_DEVICE_FUNC
|
||||
static inline RealScalar run(const Scalar& x)
|
||||
{
|
||||
EIGEN_USING_STD_MATH(abs);
|
||||
return abs(real(x)) + abs(imag(x));
|
||||
return abs(x.real()) + abs(x.imag());
|
||||
}
|
||||
};
|
||||
|
||||
@@ -662,8 +665,8 @@ struct random_default_impl<Scalar, true, false>
|
||||
{
|
||||
static inline Scalar run(const Scalar& x, const Scalar& y)
|
||||
{
|
||||
return Scalar(random(real(x), real(y)),
|
||||
random(imag(x), imag(y)));
|
||||
return Scalar(random(x.real(), y.real()),
|
||||
random(x.imag(), y.imag()));
|
||||
}
|
||||
static inline Scalar run()
|
||||
{
|
||||
@@ -916,6 +919,9 @@ inline EIGEN_MATHFUNC_RETVAL(abs2, Scalar) abs2(const Scalar& x)
|
||||
return EIGEN_MATHFUNC_IMPL(abs2, Scalar)::run(x);
|
||||
}
|
||||
|
||||
EIGEN_DEVICE_FUNC
|
||||
inline bool abs2(bool x) { return x; }
|
||||
|
||||
template<typename Scalar>
|
||||
EIGEN_DEVICE_FUNC
|
||||
inline EIGEN_MATHFUNC_RETVAL(norm1, Scalar) norm1(const Scalar& x)
|
||||
|
||||
@@ -87,17 +87,6 @@ class PermutationBase : public EigenBase<Derived>
|
||||
return derived();
|
||||
}
|
||||
|
||||
#ifndef EIGEN_PARSED_BY_DOXYGEN
|
||||
/** This is a special case of the templated operator=. Its purpose is to
|
||||
* prevent a default operator= from hiding the templated operator=.
|
||||
*/
|
||||
Derived& operator=(const PermutationBase& other)
|
||||
{
|
||||
indices() = other.indices();
|
||||
return derived();
|
||||
}
|
||||
#endif
|
||||
|
||||
/** \returns the number of rows */
|
||||
inline Index rows() const { return Index(indices().size()); }
|
||||
|
||||
@@ -333,12 +322,6 @@ class PermutationMatrix : public PermutationBase<PermutationMatrix<SizeAtCompile
|
||||
inline PermutationMatrix(const PermutationBase<OtherDerived>& other)
|
||||
: m_indices(other.indices()) {}
|
||||
|
||||
#ifndef EIGEN_PARSED_BY_DOXYGEN
|
||||
/** Standard copy constructor. Defined only to prevent a default copy constructor
|
||||
* from hiding the other templated constructor */
|
||||
inline PermutationMatrix(const PermutationMatrix& other) : m_indices(other.indices()) {}
|
||||
#endif
|
||||
|
||||
/** Generic constructor from expression of the indices. The indices
|
||||
* array has the meaning that the permutations sends each integer i to indices[i].
|
||||
*
|
||||
@@ -373,17 +356,6 @@ class PermutationMatrix : public PermutationBase<PermutationMatrix<SizeAtCompile
|
||||
return Base::operator=(tr.derived());
|
||||
}
|
||||
|
||||
#ifndef EIGEN_PARSED_BY_DOXYGEN
|
||||
/** This is a special case of the templated operator=. Its purpose is to
|
||||
* prevent a default operator= from hiding the templated operator=.
|
||||
*/
|
||||
PermutationMatrix& operator=(const PermutationMatrix& other)
|
||||
{
|
||||
m_indices = other.m_indices;
|
||||
return *this;
|
||||
}
|
||||
#endif
|
||||
|
||||
/** const version of indices(). */
|
||||
const IndicesType& indices() const { return m_indices; }
|
||||
/** \returns a reference to the stored array representing the permutation. */
|
||||
|
||||
@@ -737,8 +737,10 @@ class PlainObjectBase : public internal::dense_xpr_base<Derived>::type
|
||||
EIGEN_DEVICE_FUNC
|
||||
EIGEN_STRONG_INLINE void _init2(Index rows, Index cols, typename internal::enable_if<Base::SizeAtCompileTime!=2,T0>::type* = 0)
|
||||
{
|
||||
EIGEN_STATIC_ASSERT(bool(NumTraits<T0>::IsInteger) &&
|
||||
bool(NumTraits<T1>::IsInteger),
|
||||
const bool t0_is_integer_alike = internal::is_valid_index_type<T0>::value;
|
||||
const bool t1_is_integer_alike = internal::is_valid_index_type<T1>::value;
|
||||
EIGEN_STATIC_ASSERT(t0_is_integer_alike &&
|
||||
t1_is_integer_alike,
|
||||
FLOATING_POINT_ARGUMENT_PASSED__INTEGER_WAS_EXPECTED)
|
||||
resize(rows,cols);
|
||||
}
|
||||
@@ -773,9 +775,9 @@ class PlainObjectBase : public internal::dense_xpr_base<Derived>::type
|
||||
&& ((!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;
|
||||
EIGEN_UNUSED_VARIABLE(is_integer);
|
||||
EIGEN_STATIC_ASSERT(is_integer,
|
||||
const bool is_integer_alike = internal::is_valid_index_type<T>::value;
|
||||
EIGEN_UNUSED_VARIABLE(is_integer_alike);
|
||||
EIGEN_STATIC_ASSERT(is_integer_alike,
|
||||
FLOATING_POINT_ARGUMENT_PASSED__INTEGER_WAS_EXPECTED)
|
||||
resize(size);
|
||||
}
|
||||
|
||||
@@ -396,7 +396,7 @@ struct generic_product_impl<Lhs,Rhs,DenseShape,DenseShape,CoeffBasedProductMode>
|
||||
// but easier on the compiler side
|
||||
call_assignment_no_alias(dst, lhs.lazyProduct(rhs), internal::assign_op<typename Dst::Scalar,Scalar>());
|
||||
}
|
||||
|
||||
|
||||
template<typename Dst>
|
||||
static EIGEN_STRONG_INLINE void addTo(Dst& dst, const Lhs& lhs, const Rhs& rhs)
|
||||
{
|
||||
@@ -410,6 +410,32 @@ struct generic_product_impl<Lhs,Rhs,DenseShape,DenseShape,CoeffBasedProductMode>
|
||||
// dst.noalias() -= lhs.lazyProduct(rhs);
|
||||
call_assignment_no_alias(dst, lhs.lazyProduct(rhs), internal::sub_assign_op<typename Dst::Scalar,Scalar>());
|
||||
}
|
||||
|
||||
// Catch "dst {,+,-}= (s*A)*B" and evaluate it lazily by moving out the scalar factor:
|
||||
// dst {,+,-}= s * (A.lazyProduct(B))
|
||||
// This is a huge benefit for heap-allocated matrix types as it save one costly allocation.
|
||||
// For them, this strategy is also faster than simply by-passing the heap allocation through
|
||||
// stack allocation.
|
||||
// For fixed sizes matrices, this is less obvious, it is sometimes x2 faster, but sometimes x3 slower,
|
||||
// and the behavior depends also a lot on the compiler... so let's be conservative and enable them for dynamic-size only,
|
||||
// that is when coming from generic_product_impl<...,GemmProduct> in file GeneralMatrixMatrix.h
|
||||
template<typename Dst, typename Scalar1, typename Scalar2, typename Plain1, typename Xpr2, typename Func>
|
||||
static EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
|
||||
void eval_dynamic(Dst& dst, const CwiseBinaryOp<internal::scalar_product_op<Scalar1,Scalar2>,
|
||||
const CwiseNullaryOp<internal::scalar_constant_op<Scalar1>, Plain1>, Xpr2>& lhs, const Rhs& rhs, const Func &func)
|
||||
{
|
||||
call_assignment_no_alias(dst, lhs.lhs().functor().m_other * lhs.rhs().lazyProduct(rhs), func);
|
||||
}
|
||||
|
||||
// Here, we we always have LhsT==Lhs, but we need to make it a template type to make the above
|
||||
// overload more specialized.
|
||||
template<typename Dst, typename LhsT, typename Func>
|
||||
static EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
|
||||
void eval_dynamic(Dst& dst, const LhsT& lhs, const Rhs& rhs, const Func &func)
|
||||
{
|
||||
call_assignment_no_alias(dst, lhs.lazyProduct(rhs), func);
|
||||
}
|
||||
|
||||
|
||||
// template<typename Dst>
|
||||
// static inline void scaleAndAddTo(Dst& dst, const Lhs& lhs, const Rhs& rhs, const Scalar& alpha)
|
||||
|
||||
@@ -28,12 +28,13 @@ struct traits<Ref<_PlainObjectType, _Options, _StrideType> >
|
||||
|
||||
template<typename Derived> struct match {
|
||||
enum {
|
||||
IsVectorAtCompileTime = PlainObjectType::IsVectorAtCompileTime || Derived::IsVectorAtCompileTime,
|
||||
HasDirectAccess = internal::has_direct_access<Derived>::ret,
|
||||
StorageOrderMatch = PlainObjectType::IsVectorAtCompileTime || Derived::IsVectorAtCompileTime || ((PlainObjectType::Flags&RowMajorBit)==(Derived::Flags&RowMajorBit)),
|
||||
StorageOrderMatch = IsVectorAtCompileTime || ((PlainObjectType::Flags&RowMajorBit)==(Derived::Flags&RowMajorBit)),
|
||||
InnerStrideMatch = int(StrideType::InnerStrideAtCompileTime)==int(Dynamic)
|
||||
|| int(StrideType::InnerStrideAtCompileTime)==int(Derived::InnerStrideAtCompileTime)
|
||||
|| (int(StrideType::InnerStrideAtCompileTime)==0 && int(Derived::InnerStrideAtCompileTime)==1),
|
||||
OuterStrideMatch = Derived::IsVectorAtCompileTime
|
||||
OuterStrideMatch = IsVectorAtCompileTime
|
||||
|| int(StrideType::OuterStrideAtCompileTime)==int(Dynamic) || int(StrideType::OuterStrideAtCompileTime)==int(Derived::OuterStrideAtCompileTime),
|
||||
// NOTE, this indirection of evaluator<Derived>::Alignment is needed
|
||||
// to workaround a very strange bug in MSVC related to the instantiation
|
||||
|
||||
@@ -19,7 +19,7 @@ namespace internal {
|
||||
template<typename LhsScalar, typename RhsScalar, typename Index, int Side, int Mode, bool Conjugate, int StorageOrder>
|
||||
struct triangular_solve_vector;
|
||||
|
||||
template <typename Scalar, typename Index, int Side, int Mode, bool Conjugate, int TriStorageOrder, int OtherStorageOrder>
|
||||
template <typename Scalar, typename Index, int Side, int Mode, bool Conjugate, int TriStorageOrder, int OtherStorageOrder, int OtherInnerStride>
|
||||
struct triangular_solve_matrix;
|
||||
|
||||
// small helper struct extracting some traits on the underlying solver operation
|
||||
@@ -98,8 +98,8 @@ struct triangular_solver_selector<Lhs,Rhs,Side,Mode,NoUnrolling,Dynamic>
|
||||
BlockingType blocking(rhs.rows(), rhs.cols(), size, 1, false);
|
||||
|
||||
triangular_solve_matrix<Scalar,Index,Side,Mode,LhsProductTraits::NeedToConjugate,(int(Lhs::Flags) & RowMajorBit) ? RowMajor : ColMajor,
|
||||
(Rhs::Flags&RowMajorBit) ? RowMajor : ColMajor>
|
||||
::run(size, othersize, &actualLhs.coeffRef(0,0), actualLhs.outerStride(), &rhs.coeffRef(0,0), rhs.outerStride(), blocking);
|
||||
(Rhs::Flags&RowMajorBit) ? RowMajor : ColMajor, Rhs::InnerStrideAtCompileTime>
|
||||
::run(size, othersize, &actualLhs.coeffRef(0,0), actualLhs.outerStride(), &rhs.coeffRef(0,0), rhs.innerStride(), rhs.outerStride(), blocking);
|
||||
}
|
||||
};
|
||||
|
||||
|
||||
@@ -33,17 +33,6 @@ class TranspositionsBase
|
||||
indices() = other.indices();
|
||||
return derived();
|
||||
}
|
||||
|
||||
#ifndef EIGEN_PARSED_BY_DOXYGEN
|
||||
/** This is a special case of the templated operator=. Its purpose is to
|
||||
* prevent a default operator= from hiding the templated operator=.
|
||||
*/
|
||||
Derived& operator=(const TranspositionsBase& other)
|
||||
{
|
||||
indices() = other.indices();
|
||||
return derived();
|
||||
}
|
||||
#endif
|
||||
|
||||
/** \returns the number of transpositions */
|
||||
Index size() const { return indices().size(); }
|
||||
@@ -171,12 +160,6 @@ class Transpositions : public TranspositionsBase<Transpositions<SizeAtCompileTim
|
||||
inline Transpositions(const TranspositionsBase<OtherDerived>& other)
|
||||
: m_indices(other.indices()) {}
|
||||
|
||||
#ifndef EIGEN_PARSED_BY_DOXYGEN
|
||||
/** Standard copy constructor. Defined only to prevent a default copy constructor
|
||||
* from hiding the other templated constructor */
|
||||
inline Transpositions(const Transpositions& other) : m_indices(other.indices()) {}
|
||||
#endif
|
||||
|
||||
/** Generic constructor from expression of the transposition indices. */
|
||||
template<typename Other>
|
||||
explicit inline Transpositions(const MatrixBase<Other>& indices) : m_indices(indices)
|
||||
@@ -189,17 +172,6 @@ class Transpositions : public TranspositionsBase<Transpositions<SizeAtCompileTim
|
||||
return Base::operator=(other);
|
||||
}
|
||||
|
||||
#ifndef EIGEN_PARSED_BY_DOXYGEN
|
||||
/** This is a special case of the templated operator=. Its purpose is to
|
||||
* prevent a default operator= from hiding the templated operator=.
|
||||
*/
|
||||
Transpositions& operator=(const Transpositions& other)
|
||||
{
|
||||
m_indices = other.m_indices;
|
||||
return *this;
|
||||
}
|
||||
#endif
|
||||
|
||||
/** Constructs an uninitialized permutation matrix of given size.
|
||||
*/
|
||||
inline Transpositions(Index size) : m_indices(size)
|
||||
@@ -306,17 +278,6 @@ class TranspositionsWrapper
|
||||
return Base::operator=(other);
|
||||
}
|
||||
|
||||
#ifndef EIGEN_PARSED_BY_DOXYGEN
|
||||
/** This is a special case of the templated operator=. Its purpose is to
|
||||
* prevent a default operator= from hiding the templated operator=.
|
||||
*/
|
||||
TranspositionsWrapper& operator=(const TranspositionsWrapper& other)
|
||||
{
|
||||
m_indices = other.m_indices;
|
||||
return *this;
|
||||
}
|
||||
#endif
|
||||
|
||||
/** const version of indices(). */
|
||||
const IndicesType& indices() const { return m_indices; }
|
||||
|
||||
|
||||
@@ -768,7 +768,7 @@ struct scalar_sign_op<Scalar,true> {
|
||||
if (aa==real_type(0))
|
||||
return Scalar(0);
|
||||
aa = real_type(1)/aa;
|
||||
return Scalar(real(a)*aa, imag(a)*aa );
|
||||
return Scalar(a.real()*aa, a.imag()*aa );
|
||||
}
|
||||
//TODO
|
||||
//template <typename Packet>
|
||||
|
||||
@@ -115,7 +115,8 @@ void evaluateProductBlockingSizesHeuristic(Index& k, Index& m, Index& n, Index n
|
||||
// registers. However once the latency is hidden there is no point in
|
||||
// increasing the value of k, so we'll cap it at 320 (value determined
|
||||
// experimentally).
|
||||
const Index k_cache = (numext::mini<Index>)((l1-ksub)/kdiv, 320);
|
||||
// To avoid that k vanishes, we make k_cache at least as big as kr
|
||||
const Index k_cache = numext::maxi<Index>(kr, (numext::mini<Index>)((l1-ksub)/kdiv, 320));
|
||||
if (k_cache < k) {
|
||||
k = k_cache - (k_cache % kr);
|
||||
eigen_internal_assert(k > 0);
|
||||
@@ -648,8 +649,8 @@ public:
|
||||
// Vectorized path
|
||||
EIGEN_STRONG_INLINE void loadRhs(const RhsScalar* b, DoublePacketType& dest) const
|
||||
{
|
||||
dest.first = pset1<RealPacket>(real(*b));
|
||||
dest.second = pset1<RealPacket>(imag(*b));
|
||||
dest.first = pset1<RealPacket>(numext::real(*b));
|
||||
dest.second = pset1<RealPacket>(numext::imag(*b));
|
||||
}
|
||||
|
||||
EIGEN_STRONG_INLINE void loadRhsQuad(const RhsScalar* b, ResPacket& dest) const
|
||||
|
||||
@@ -20,8 +20,9 @@ template<typename _LhsScalar, typename _RhsScalar> class level3_blocking;
|
||||
template<
|
||||
typename Index,
|
||||
typename LhsScalar, int LhsStorageOrder, bool ConjugateLhs,
|
||||
typename RhsScalar, int RhsStorageOrder, bool ConjugateRhs>
|
||||
struct general_matrix_matrix_product<Index,LhsScalar,LhsStorageOrder,ConjugateLhs,RhsScalar,RhsStorageOrder,ConjugateRhs,RowMajor>
|
||||
typename RhsScalar, int RhsStorageOrder, bool ConjugateRhs,
|
||||
int ResInnerStride>
|
||||
struct general_matrix_matrix_product<Index,LhsScalar,LhsStorageOrder,ConjugateLhs,RhsScalar,RhsStorageOrder,ConjugateRhs,RowMajor,ResInnerStride>
|
||||
{
|
||||
typedef gebp_traits<RhsScalar,LhsScalar> Traits;
|
||||
|
||||
@@ -30,7 +31,7 @@ struct general_matrix_matrix_product<Index,LhsScalar,LhsStorageOrder,ConjugateLh
|
||||
Index rows, Index cols, Index depth,
|
||||
const LhsScalar* lhs, Index lhsStride,
|
||||
const RhsScalar* rhs, Index rhsStride,
|
||||
ResScalar* res, Index resStride,
|
||||
ResScalar* res, Index resIncr, Index resStride,
|
||||
ResScalar alpha,
|
||||
level3_blocking<RhsScalar,LhsScalar>& blocking,
|
||||
GemmParallelInfo<Index>* info = 0)
|
||||
@@ -39,8 +40,8 @@ struct general_matrix_matrix_product<Index,LhsScalar,LhsStorageOrder,ConjugateLh
|
||||
general_matrix_matrix_product<Index,
|
||||
RhsScalar, RhsStorageOrder==RowMajor ? ColMajor : RowMajor, ConjugateRhs,
|
||||
LhsScalar, LhsStorageOrder==RowMajor ? ColMajor : RowMajor, ConjugateLhs,
|
||||
ColMajor>
|
||||
::run(cols,rows,depth,rhs,rhsStride,lhs,lhsStride,res,resStride,alpha,blocking,info);
|
||||
ColMajor,ResInnerStride>
|
||||
::run(cols,rows,depth,rhs,rhsStride,lhs,lhsStride,res,resIncr,resStride,alpha,blocking,info);
|
||||
}
|
||||
};
|
||||
|
||||
@@ -49,8 +50,9 @@ struct general_matrix_matrix_product<Index,LhsScalar,LhsStorageOrder,ConjugateLh
|
||||
template<
|
||||
typename Index,
|
||||
typename LhsScalar, int LhsStorageOrder, bool ConjugateLhs,
|
||||
typename RhsScalar, int RhsStorageOrder, bool ConjugateRhs>
|
||||
struct general_matrix_matrix_product<Index,LhsScalar,LhsStorageOrder,ConjugateLhs,RhsScalar,RhsStorageOrder,ConjugateRhs,ColMajor>
|
||||
typename RhsScalar, int RhsStorageOrder, bool ConjugateRhs,
|
||||
int ResInnerStride>
|
||||
struct general_matrix_matrix_product<Index,LhsScalar,LhsStorageOrder,ConjugateLhs,RhsScalar,RhsStorageOrder,ConjugateRhs,ColMajor,ResInnerStride>
|
||||
{
|
||||
|
||||
typedef gebp_traits<LhsScalar,RhsScalar> Traits;
|
||||
@@ -59,17 +61,17 @@ typedef typename ScalarBinaryOpTraits<LhsScalar, RhsScalar>::ReturnType ResScala
|
||||
static void run(Index rows, Index cols, Index depth,
|
||||
const LhsScalar* _lhs, Index lhsStride,
|
||||
const RhsScalar* _rhs, Index rhsStride,
|
||||
ResScalar* _res, Index resStride,
|
||||
ResScalar* _res, Index resIncr, Index resStride,
|
||||
ResScalar alpha,
|
||||
level3_blocking<LhsScalar,RhsScalar>& blocking,
|
||||
GemmParallelInfo<Index>* info = 0)
|
||||
{
|
||||
typedef const_blas_data_mapper<LhsScalar, Index, LhsStorageOrder> LhsMapper;
|
||||
typedef const_blas_data_mapper<RhsScalar, Index, RhsStorageOrder> RhsMapper;
|
||||
typedef blas_data_mapper<typename Traits::ResScalar, Index, ColMajor> ResMapper;
|
||||
LhsMapper lhs(_lhs,lhsStride);
|
||||
RhsMapper rhs(_rhs,rhsStride);
|
||||
ResMapper res(_res, resStride);
|
||||
typedef blas_data_mapper<typename Traits::ResScalar, Index, ColMajor,Unaligned,ResInnerStride> ResMapper;
|
||||
LhsMapper lhs(_lhs, lhsStride);
|
||||
RhsMapper rhs(_rhs, rhsStride);
|
||||
ResMapper res(_res, resStride, resIncr);
|
||||
|
||||
Index kc = blocking.kc(); // cache block size along the K direction
|
||||
Index mc = (std::min)(rows,blocking.mc()); // cache block size along the M direction
|
||||
@@ -226,7 +228,7 @@ struct gemm_functor
|
||||
Gemm::run(rows, cols, m_lhs.cols(),
|
||||
&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(),
|
||||
(Scalar*)&(m_dest.coeffRef(row,col)), m_dest.innerStride(), m_dest.outerStride(),
|
||||
m_actualAlpha, m_blocking, info);
|
||||
}
|
||||
|
||||
@@ -428,7 +430,7 @@ struct generic_product_impl<Lhs,Rhs,DenseShape,DenseShape,GemmProduct>
|
||||
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);
|
||||
lazyproduct::eval_dynamic(dst, lhs, rhs, internal::assign_op<typename Dst::Scalar,Scalar>());
|
||||
else
|
||||
{
|
||||
dst.setZero();
|
||||
@@ -440,7 +442,7 @@ struct generic_product_impl<Lhs,Rhs,DenseShape,DenseShape,GemmProduct>
|
||||
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);
|
||||
lazyproduct::eval_dynamic(dst, lhs, rhs, internal::add_assign_op<typename Dst::Scalar,Scalar>());
|
||||
else
|
||||
scaleAndAddTo(dst,lhs, rhs, Scalar(1));
|
||||
}
|
||||
@@ -449,7 +451,7 @@ struct generic_product_impl<Lhs,Rhs,DenseShape,DenseShape,GemmProduct>
|
||||
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);
|
||||
lazyproduct::eval_dynamic(dst, lhs, rhs, internal::sub_assign_op<typename Dst::Scalar,Scalar>());
|
||||
else
|
||||
scaleAndAddTo(dst, lhs, rhs, Scalar(-1));
|
||||
}
|
||||
@@ -476,7 +478,8 @@ struct generic_product_impl<Lhs,Rhs,DenseShape,DenseShape,GemmProduct>
|
||||
Index,
|
||||
LhsScalar, (ActualLhsTypeCleaned::Flags&RowMajorBit) ? RowMajor : ColMajor, bool(LhsBlasTraits::NeedToConjugate),
|
||||
RhsScalar, (ActualRhsTypeCleaned::Flags&RowMajorBit) ? RowMajor : ColMajor, bool(RhsBlasTraits::NeedToConjugate),
|
||||
(Dest::Flags&RowMajorBit) ? RowMajor : ColMajor>,
|
||||
(Dest::Flags&RowMajorBit) ? RowMajor : ColMajor,
|
||||
Dest::InnerStrideAtCompileTime>,
|
||||
ActualLhsTypeCleaned, ActualRhsTypeCleaned, Dest, BlockingType> GemmFunctor;
|
||||
|
||||
BlockingType blocking(dst.rows(), dst.cols(), lhs.cols(), 1, true);
|
||||
|
||||
@@ -25,51 +25,54 @@ namespace internal {
|
||||
**********************************************************************/
|
||||
|
||||
// forward declarations (defined at the end of this file)
|
||||
template<typename LhsScalar, typename RhsScalar, typename Index, int mr, int nr, bool ConjLhs, bool ConjRhs, int UpLo>
|
||||
template<typename LhsScalar, typename RhsScalar, typename Index, int mr, int nr, bool ConjLhs, bool ConjRhs, int ResInnerStride, int UpLo>
|
||||
struct tribb_kernel;
|
||||
|
||||
/* Optimized matrix-matrix product evaluating only one triangular half */
|
||||
template <typename Index,
|
||||
typename LhsScalar, int LhsStorageOrder, bool ConjugateLhs,
|
||||
typename RhsScalar, int RhsStorageOrder, bool ConjugateRhs,
|
||||
int ResStorageOrder, int UpLo, int Version = Specialized>
|
||||
int ResStorageOrder, int ResInnerStride, int UpLo, int Version = Specialized>
|
||||
struct general_matrix_matrix_triangular_product;
|
||||
|
||||
// as usual if the result is row major => we transpose the product
|
||||
template <typename Index, typename LhsScalar, int LhsStorageOrder, bool ConjugateLhs,
|
||||
typename RhsScalar, int RhsStorageOrder, bool ConjugateRhs, int UpLo, int Version>
|
||||
struct general_matrix_matrix_triangular_product<Index,LhsScalar,LhsStorageOrder,ConjugateLhs,RhsScalar,RhsStorageOrder,ConjugateRhs,RowMajor,UpLo,Version>
|
||||
typename RhsScalar, int RhsStorageOrder, bool ConjugateRhs,
|
||||
int ResInnerStride, int UpLo, int Version>
|
||||
struct general_matrix_matrix_triangular_product<Index,LhsScalar,LhsStorageOrder,ConjugateLhs,RhsScalar,RhsStorageOrder,ConjugateRhs,RowMajor,ResInnerStride,UpLo,Version>
|
||||
{
|
||||
typedef typename ScalarBinaryOpTraits<LhsScalar, RhsScalar>::ReturnType ResScalar;
|
||||
static EIGEN_STRONG_INLINE void run(Index size, Index depth,const LhsScalar* lhs, Index lhsStride,
|
||||
const RhsScalar* rhs, Index rhsStride, ResScalar* res, Index resStride,
|
||||
const RhsScalar* rhs, Index rhsStride, ResScalar* res, Index resIncr, Index resStride,
|
||||
const ResScalar& alpha, level3_blocking<RhsScalar,LhsScalar>& blocking)
|
||||
{
|
||||
general_matrix_matrix_triangular_product<Index,
|
||||
RhsScalar, RhsStorageOrder==RowMajor ? ColMajor : RowMajor, ConjugateRhs,
|
||||
LhsScalar, LhsStorageOrder==RowMajor ? ColMajor : RowMajor, ConjugateLhs,
|
||||
ColMajor, UpLo==Lower?Upper:Lower>
|
||||
::run(size,depth,rhs,rhsStride,lhs,lhsStride,res,resStride,alpha,blocking);
|
||||
ColMajor, ResInnerStride, UpLo==Lower?Upper:Lower>
|
||||
::run(size,depth,rhs,rhsStride,lhs,lhsStride,res,resIncr,resStride,alpha,blocking);
|
||||
}
|
||||
};
|
||||
|
||||
template <typename Index, typename LhsScalar, int LhsStorageOrder, bool ConjugateLhs,
|
||||
typename RhsScalar, int RhsStorageOrder, bool ConjugateRhs, int UpLo, int Version>
|
||||
struct general_matrix_matrix_triangular_product<Index,LhsScalar,LhsStorageOrder,ConjugateLhs,RhsScalar,RhsStorageOrder,ConjugateRhs,ColMajor,UpLo,Version>
|
||||
typename RhsScalar, int RhsStorageOrder, bool ConjugateRhs,
|
||||
int ResInnerStride, int UpLo, int Version>
|
||||
struct general_matrix_matrix_triangular_product<Index,LhsScalar,LhsStorageOrder,ConjugateLhs,RhsScalar,RhsStorageOrder,ConjugateRhs,ColMajor,ResInnerStride,UpLo,Version>
|
||||
{
|
||||
typedef typename ScalarBinaryOpTraits<LhsScalar, RhsScalar>::ReturnType ResScalar;
|
||||
static EIGEN_STRONG_INLINE void run(Index size, Index depth,const LhsScalar* _lhs, Index lhsStride,
|
||||
const RhsScalar* _rhs, Index rhsStride, ResScalar* _res, Index resStride,
|
||||
const RhsScalar* _rhs, Index rhsStride,
|
||||
ResScalar* _res, Index resIncr, Index resStride,
|
||||
const ResScalar& alpha, level3_blocking<LhsScalar,RhsScalar>& blocking)
|
||||
{
|
||||
typedef gebp_traits<LhsScalar,RhsScalar> Traits;
|
||||
|
||||
typedef const_blas_data_mapper<LhsScalar, Index, LhsStorageOrder> LhsMapper;
|
||||
typedef const_blas_data_mapper<RhsScalar, Index, RhsStorageOrder> RhsMapper;
|
||||
typedef blas_data_mapper<typename Traits::ResScalar, Index, ColMajor> ResMapper;
|
||||
typedef blas_data_mapper<typename Traits::ResScalar, Index, ColMajor, Unaligned, ResInnerStride> ResMapper;
|
||||
LhsMapper lhs(_lhs,lhsStride);
|
||||
RhsMapper rhs(_rhs,rhsStride);
|
||||
ResMapper res(_res, resStride);
|
||||
ResMapper res(_res, resStride, resIncr);
|
||||
|
||||
Index kc = blocking.kc();
|
||||
Index mc = (std::min)(size,blocking.mc());
|
||||
@@ -87,7 +90,7 @@ struct general_matrix_matrix_triangular_product<Index,LhsScalar,LhsStorageOrder,
|
||||
gemm_pack_lhs<LhsScalar, Index, LhsMapper, Traits::mr, Traits::LhsProgress, LhsStorageOrder> pack_lhs;
|
||||
gemm_pack_rhs<RhsScalar, Index, RhsMapper, Traits::nr, RhsStorageOrder> pack_rhs;
|
||||
gebp_kernel<LhsScalar, RhsScalar, Index, ResMapper, Traits::mr, Traits::nr, ConjugateLhs, ConjugateRhs> gebp;
|
||||
tribb_kernel<LhsScalar, RhsScalar, Index, Traits::mr, Traits::nr, ConjugateLhs, ConjugateRhs, UpLo> sybb;
|
||||
tribb_kernel<LhsScalar, RhsScalar, Index, Traits::mr, Traits::nr, ConjugateLhs, ConjugateRhs, ResInnerStride, UpLo> sybb;
|
||||
|
||||
for(Index k2=0; k2<depth; k2+=kc)
|
||||
{
|
||||
@@ -110,8 +113,7 @@ struct general_matrix_matrix_triangular_product<Index,LhsScalar,LhsStorageOrder,
|
||||
gebp(res.getSubMapper(i2, 0), blockA, blockB, actual_mc, actual_kc,
|
||||
(std::min)(size,i2), alpha, -1, -1, 0, 0);
|
||||
|
||||
|
||||
sybb(_res+resStride*i2 + i2, resStride, blockA, blockB + actual_kc*i2, actual_mc, actual_kc, alpha);
|
||||
sybb(_res+resStride*i2 + resIncr*i2, resIncr, resStride, blockA, blockB + actual_kc*i2, actual_mc, actual_kc, alpha);
|
||||
|
||||
if (UpLo==Upper)
|
||||
{
|
||||
@@ -133,7 +135,7 @@ struct general_matrix_matrix_triangular_product<Index,LhsScalar,LhsStorageOrder,
|
||||
// while the triangular block overlapping the diagonal is evaluated into a
|
||||
// small temporary buffer which is then accumulated into the result using a
|
||||
// triangular traversal.
|
||||
template<typename LhsScalar, typename RhsScalar, typename Index, int mr, int nr, bool ConjLhs, bool ConjRhs, int UpLo>
|
||||
template<typename LhsScalar, typename RhsScalar, typename Index, int mr, int nr, bool ConjLhs, bool ConjRhs, int ResInnerStride, int UpLo>
|
||||
struct tribb_kernel
|
||||
{
|
||||
typedef gebp_traits<LhsScalar,RhsScalar,ConjLhs,ConjRhs> Traits;
|
||||
@@ -142,11 +144,13 @@ struct tribb_kernel
|
||||
enum {
|
||||
BlockSize = meta_least_common_multiple<EIGEN_PLAIN_ENUM_MAX(mr,nr),EIGEN_PLAIN_ENUM_MIN(mr,nr)>::ret
|
||||
};
|
||||
void operator()(ResScalar* _res, Index resStride, const LhsScalar* blockA, const RhsScalar* blockB, Index size, Index depth, const ResScalar& alpha)
|
||||
void operator()(ResScalar* _res, Index resIncr, Index resStride, const LhsScalar* blockA, const RhsScalar* blockB, Index size, Index depth, const ResScalar& alpha)
|
||||
{
|
||||
typedef blas_data_mapper<ResScalar, Index, ColMajor> ResMapper;
|
||||
ResMapper res(_res, resStride);
|
||||
gebp_kernel<LhsScalar, RhsScalar, Index, ResMapper, mr, nr, ConjLhs, ConjRhs> gebp_kernel;
|
||||
typedef blas_data_mapper<ResScalar, Index, ColMajor, Unaligned, ResInnerStride> ResMapper;
|
||||
typedef blas_data_mapper<ResScalar, Index, ColMajor, Unaligned> BufferMapper;
|
||||
ResMapper res(_res, resStride, resIncr);
|
||||
gebp_kernel<LhsScalar, RhsScalar, Index, ResMapper, mr, nr, ConjLhs, ConjRhs> gebp_kernel1;
|
||||
gebp_kernel<LhsScalar, RhsScalar, Index, BufferMapper, mr, nr, ConjLhs, ConjRhs> gebp_kernel2;
|
||||
|
||||
Matrix<ResScalar,BlockSize,BlockSize,ColMajor> buffer((internal::constructor_without_unaligned_array_assert()));
|
||||
|
||||
@@ -158,31 +162,32 @@ struct tribb_kernel
|
||||
const RhsScalar* actual_b = blockB+j*depth;
|
||||
|
||||
if(UpLo==Upper)
|
||||
gebp_kernel(res.getSubMapper(0, j), blockA, actual_b, j, depth, actualBlockSize, alpha,
|
||||
-1, -1, 0, 0);
|
||||
|
||||
gebp_kernel1(res.getSubMapper(0, j), blockA, actual_b, j, depth, actualBlockSize, alpha,
|
||||
-1, -1, 0, 0);
|
||||
|
||||
// selfadjoint micro block
|
||||
{
|
||||
Index i = j;
|
||||
buffer.setZero();
|
||||
// 1 - apply the kernel on the temporary buffer
|
||||
gebp_kernel(ResMapper(buffer.data(), BlockSize), blockA+depth*i, actual_b, actualBlockSize, depth, actualBlockSize, alpha,
|
||||
-1, -1, 0, 0);
|
||||
gebp_kernel2(BufferMapper(buffer.data(), BlockSize), blockA+depth*i, actual_b, actualBlockSize, depth, actualBlockSize, alpha,
|
||||
-1, -1, 0, 0);
|
||||
|
||||
// 2 - triangular accumulation
|
||||
for(Index j1=0; j1<actualBlockSize; ++j1)
|
||||
{
|
||||
ResScalar* r = &res(i, j + j1);
|
||||
typename ResMapper::LinearMapper r = res.getLinearMapper(i,j+j1);
|
||||
for(Index i1=UpLo==Lower ? j1 : 0;
|
||||
UpLo==Lower ? i1<actualBlockSize : i1<=j1; ++i1)
|
||||
r[i1] += buffer(i1,j1);
|
||||
r(i1) += buffer(i1,j1);
|
||||
}
|
||||
}
|
||||
|
||||
if(UpLo==Lower)
|
||||
{
|
||||
Index i = j+actualBlockSize;
|
||||
gebp_kernel(res.getSubMapper(i, j), blockA+depth*i, actual_b, size-i,
|
||||
depth, actualBlockSize, alpha, -1, -1, 0, 0);
|
||||
gebp_kernel1(res.getSubMapper(i, j), blockA+depth*i, actual_b, size-i,
|
||||
depth, actualBlockSize, alpha, -1, -1, 0, 0);
|
||||
}
|
||||
}
|
||||
}
|
||||
@@ -286,11 +291,12 @@ struct general_product_to_triangular_selector<MatrixType,ProductType,UpLo,false>
|
||||
internal::general_matrix_matrix_triangular_product<Index,
|
||||
typename Lhs::Scalar, LhsIsRowMajor ? RowMajor : ColMajor, LhsBlasTraits::NeedToConjugate,
|
||||
typename Rhs::Scalar, RhsIsRowMajor ? RowMajor : ColMajor, RhsBlasTraits::NeedToConjugate,
|
||||
IsRowMajor ? RowMajor : ColMajor, UpLo&(Lower|Upper)>
|
||||
IsRowMajor ? RowMajor : ColMajor, MatrixType::InnerStrideAtCompileTime, UpLo&(Lower|Upper)>
|
||||
::run(size, depth,
|
||||
&actualLhs.coeffRef(SkipDiag&&(UpLo&Lower)==Lower ? 1 : 0,0), actualLhs.outerStride(),
|
||||
&actualRhs.coeffRef(0,SkipDiag&&(UpLo&Upper)==Upper ? 1 : 0), actualRhs.outerStride(),
|
||||
mat.data() + (SkipDiag ? (bool(IsRowMajor) != ((UpLo&Lower)==Lower) ? 1 : mat.outerStride() ) : 0), mat.outerStride(), actualAlpha, blocking);
|
||||
mat.data() + (SkipDiag ? (bool(IsRowMajor) != ((UpLo&Lower)==Lower) ? mat.innerStride() : mat.outerStride() ) : 0),
|
||||
mat.innerStride(), mat.outerStride(), actualAlpha, blocking);
|
||||
}
|
||||
};
|
||||
|
||||
|
||||
@@ -17,7 +17,8 @@ namespace internal {
|
||||
/** \internal */
|
||||
inline void manage_multi_threading(Action action, int* v)
|
||||
{
|
||||
static EIGEN_UNUSED int m_maxThreads = -1;
|
||||
static int m_maxThreads = -1;
|
||||
EIGEN_UNUSED_VARIABLE(m_maxThreads);
|
||||
|
||||
if(action==SetAction)
|
||||
{
|
||||
@@ -150,8 +151,10 @@ void parallelize_gemm(const Functor& func, Index rows, Index cols, Index depth,
|
||||
info[i].lhs_start = r0;
|
||||
info[i].lhs_length = actualBlockRows;
|
||||
|
||||
if(transpose) func(c0, actualBlockCols, 0, rows, info);
|
||||
else func(0, rows, c0, actualBlockCols, info);
|
||||
if(transpose)
|
||||
func(c0, actualBlockCols, 0, rows, info);
|
||||
else
|
||||
func(0, rows, c0, actualBlockCols, info);
|
||||
}
|
||||
#endif
|
||||
}
|
||||
|
||||
@@ -277,20 +277,21 @@ struct symm_pack_rhs
|
||||
template <typename Scalar, typename Index,
|
||||
int LhsStorageOrder, bool LhsSelfAdjoint, bool ConjugateLhs,
|
||||
int RhsStorageOrder, bool RhsSelfAdjoint, bool ConjugateRhs,
|
||||
int ResStorageOrder>
|
||||
int ResStorageOrder, int ResInnerStride>
|
||||
struct product_selfadjoint_matrix;
|
||||
|
||||
template <typename Scalar, typename Index,
|
||||
int LhsStorageOrder, bool LhsSelfAdjoint, bool ConjugateLhs,
|
||||
int RhsStorageOrder, bool RhsSelfAdjoint, bool ConjugateRhs>
|
||||
struct product_selfadjoint_matrix<Scalar,Index,LhsStorageOrder,LhsSelfAdjoint,ConjugateLhs, RhsStorageOrder,RhsSelfAdjoint,ConjugateRhs,RowMajor>
|
||||
int RhsStorageOrder, bool RhsSelfAdjoint, bool ConjugateRhs,
|
||||
int ResInnerStride>
|
||||
struct product_selfadjoint_matrix<Scalar,Index,LhsStorageOrder,LhsSelfAdjoint,ConjugateLhs, RhsStorageOrder,RhsSelfAdjoint,ConjugateRhs,RowMajor,ResInnerStride>
|
||||
{
|
||||
|
||||
static EIGEN_STRONG_INLINE void run(
|
||||
Index rows, Index cols,
|
||||
const Scalar* lhs, Index lhsStride,
|
||||
const Scalar* rhs, Index rhsStride,
|
||||
Scalar* res, Index resStride,
|
||||
Scalar* res, Index resIncr, Index resStride,
|
||||
const Scalar& alpha, level3_blocking<Scalar,Scalar>& blocking)
|
||||
{
|
||||
product_selfadjoint_matrix<Scalar, Index,
|
||||
@@ -298,33 +299,35 @@ struct product_selfadjoint_matrix<Scalar,Index,LhsStorageOrder,LhsSelfAdjoint,Co
|
||||
RhsSelfAdjoint, NumTraits<Scalar>::IsComplex && EIGEN_LOGICAL_XOR(RhsSelfAdjoint,ConjugateRhs),
|
||||
EIGEN_LOGICAL_XOR(LhsSelfAdjoint,LhsStorageOrder==RowMajor) ? ColMajor : RowMajor,
|
||||
LhsSelfAdjoint, NumTraits<Scalar>::IsComplex && EIGEN_LOGICAL_XOR(LhsSelfAdjoint,ConjugateLhs),
|
||||
ColMajor>
|
||||
::run(cols, rows, rhs, rhsStride, lhs, lhsStride, res, resStride, alpha, blocking);
|
||||
ColMajor,ResInnerStride>
|
||||
::run(cols, rows, rhs, rhsStride, lhs, lhsStride, res, resIncr, resStride, alpha, blocking);
|
||||
}
|
||||
};
|
||||
|
||||
template <typename Scalar, typename Index,
|
||||
int LhsStorageOrder, bool ConjugateLhs,
|
||||
int RhsStorageOrder, bool ConjugateRhs>
|
||||
struct product_selfadjoint_matrix<Scalar,Index,LhsStorageOrder,true,ConjugateLhs, RhsStorageOrder,false,ConjugateRhs,ColMajor>
|
||||
int RhsStorageOrder, bool ConjugateRhs,
|
||||
int ResInnerStride>
|
||||
struct product_selfadjoint_matrix<Scalar,Index,LhsStorageOrder,true,ConjugateLhs, RhsStorageOrder,false,ConjugateRhs,ColMajor,ResInnerStride>
|
||||
{
|
||||
|
||||
static EIGEN_DONT_INLINE void run(
|
||||
Index rows, Index cols,
|
||||
const Scalar* _lhs, Index lhsStride,
|
||||
const Scalar* _rhs, Index rhsStride,
|
||||
Scalar* res, Index resStride,
|
||||
Scalar* res, Index resIncr, Index resStride,
|
||||
const Scalar& alpha, level3_blocking<Scalar,Scalar>& blocking);
|
||||
};
|
||||
|
||||
template <typename Scalar, typename Index,
|
||||
int LhsStorageOrder, bool ConjugateLhs,
|
||||
int RhsStorageOrder, bool ConjugateRhs>
|
||||
EIGEN_DONT_INLINE void product_selfadjoint_matrix<Scalar,Index,LhsStorageOrder,true,ConjugateLhs, RhsStorageOrder,false,ConjugateRhs,ColMajor>::run(
|
||||
int RhsStorageOrder, bool ConjugateRhs,
|
||||
int ResInnerStride>
|
||||
EIGEN_DONT_INLINE void product_selfadjoint_matrix<Scalar,Index,LhsStorageOrder,true,ConjugateLhs, RhsStorageOrder,false,ConjugateRhs,ColMajor,ResInnerStride>::run(
|
||||
Index rows, Index cols,
|
||||
const Scalar* _lhs, Index lhsStride,
|
||||
const Scalar* _rhs, Index rhsStride,
|
||||
Scalar* _res, Index resStride,
|
||||
Scalar* _res, Index resIncr, Index resStride,
|
||||
const Scalar& alpha, level3_blocking<Scalar,Scalar>& blocking)
|
||||
{
|
||||
Index size = rows;
|
||||
@@ -334,11 +337,11 @@ EIGEN_DONT_INLINE void product_selfadjoint_matrix<Scalar,Index,LhsStorageOrder,t
|
||||
typedef const_blas_data_mapper<Scalar, Index, LhsStorageOrder> LhsMapper;
|
||||
typedef const_blas_data_mapper<Scalar, Index, (LhsStorageOrder == RowMajor) ? ColMajor : RowMajor> LhsTransposeMapper;
|
||||
typedef const_blas_data_mapper<Scalar, Index, RhsStorageOrder> RhsMapper;
|
||||
typedef blas_data_mapper<typename Traits::ResScalar, Index, ColMajor> ResMapper;
|
||||
typedef blas_data_mapper<typename Traits::ResScalar, Index, ColMajor, Unaligned, ResInnerStride> ResMapper;
|
||||
LhsMapper lhs(_lhs,lhsStride);
|
||||
LhsTransposeMapper lhs_transpose(_lhs,lhsStride);
|
||||
RhsMapper rhs(_rhs,rhsStride);
|
||||
ResMapper res(_res, resStride);
|
||||
ResMapper res(_res, resStride, resIncr);
|
||||
|
||||
Index kc = blocking.kc(); // cache block size along the K direction
|
||||
Index mc = (std::min)(rows,blocking.mc()); // cache block size along the M direction
|
||||
@@ -398,26 +401,28 @@ EIGEN_DONT_INLINE void product_selfadjoint_matrix<Scalar,Index,LhsStorageOrder,t
|
||||
// matrix * selfadjoint product
|
||||
template <typename Scalar, typename Index,
|
||||
int LhsStorageOrder, bool ConjugateLhs,
|
||||
int RhsStorageOrder, bool ConjugateRhs>
|
||||
struct product_selfadjoint_matrix<Scalar,Index,LhsStorageOrder,false,ConjugateLhs, RhsStorageOrder,true,ConjugateRhs,ColMajor>
|
||||
int RhsStorageOrder, bool ConjugateRhs,
|
||||
int ResInnerStride>
|
||||
struct product_selfadjoint_matrix<Scalar,Index,LhsStorageOrder,false,ConjugateLhs, RhsStorageOrder,true,ConjugateRhs,ColMajor,ResInnerStride>
|
||||
{
|
||||
|
||||
static EIGEN_DONT_INLINE void run(
|
||||
Index rows, Index cols,
|
||||
const Scalar* _lhs, Index lhsStride,
|
||||
const Scalar* _rhs, Index rhsStride,
|
||||
Scalar* res, Index resStride,
|
||||
Scalar* res, Index resIncr, Index resStride,
|
||||
const Scalar& alpha, level3_blocking<Scalar,Scalar>& blocking);
|
||||
};
|
||||
|
||||
template <typename Scalar, typename Index,
|
||||
int LhsStorageOrder, bool ConjugateLhs,
|
||||
int RhsStorageOrder, bool ConjugateRhs>
|
||||
EIGEN_DONT_INLINE void product_selfadjoint_matrix<Scalar,Index,LhsStorageOrder,false,ConjugateLhs, RhsStorageOrder,true,ConjugateRhs,ColMajor>::run(
|
||||
int RhsStorageOrder, bool ConjugateRhs,
|
||||
int ResInnerStride>
|
||||
EIGEN_DONT_INLINE void product_selfadjoint_matrix<Scalar,Index,LhsStorageOrder,false,ConjugateLhs, RhsStorageOrder,true,ConjugateRhs,ColMajor,ResInnerStride>::run(
|
||||
Index rows, Index cols,
|
||||
const Scalar* _lhs, Index lhsStride,
|
||||
const Scalar* _rhs, Index rhsStride,
|
||||
Scalar* _res, Index resStride,
|
||||
Scalar* _res, Index resIncr, Index resStride,
|
||||
const Scalar& alpha, level3_blocking<Scalar,Scalar>& blocking)
|
||||
{
|
||||
Index size = cols;
|
||||
@@ -425,9 +430,9 @@ EIGEN_DONT_INLINE void product_selfadjoint_matrix<Scalar,Index,LhsStorageOrder,f
|
||||
typedef gebp_traits<Scalar,Scalar> Traits;
|
||||
|
||||
typedef const_blas_data_mapper<Scalar, Index, LhsStorageOrder> LhsMapper;
|
||||
typedef blas_data_mapper<typename Traits::ResScalar, Index, ColMajor> ResMapper;
|
||||
typedef blas_data_mapper<typename Traits::ResScalar, Index, ColMajor, Unaligned, ResInnerStride> ResMapper;
|
||||
LhsMapper lhs(_lhs,lhsStride);
|
||||
ResMapper res(_res,resStride);
|
||||
ResMapper res(_res,resStride, resIncr);
|
||||
|
||||
Index kc = blocking.kc(); // cache block size along the K direction
|
||||
Index mc = (std::min)(rows,blocking.mc()); // cache block size along the M direction
|
||||
@@ -503,12 +508,13 @@ struct selfadjoint_product_impl<Lhs,LhsMode,false,Rhs,RhsMode,false>
|
||||
NumTraits<Scalar>::IsComplex && EIGEN_LOGICAL_XOR(LhsIsUpper,bool(LhsBlasTraits::NeedToConjugate)),
|
||||
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>
|
||||
internal::traits<Dest>::Flags&RowMajorBit ? RowMajor : ColMajor,
|
||||
Dest::InnerStrideAtCompileTime>
|
||||
::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
|
||||
&dst.coeffRef(0,0), dst.innerStride(), dst.outerStride(), // result info
|
||||
actualAlpha, blocking // alpha
|
||||
);
|
||||
}
|
||||
|
||||
@@ -109,10 +109,10 @@ struct selfadjoint_product_selector<MatrixType,OtherType,UpLo,false>
|
||||
internal::general_matrix_matrix_triangular_product<Index,
|
||||
Scalar, OtherIsRowMajor ? RowMajor : ColMajor, OtherBlasTraits::NeedToConjugate && NumTraits<Scalar>::IsComplex,
|
||||
Scalar, OtherIsRowMajor ? ColMajor : RowMajor, (!OtherBlasTraits::NeedToConjugate) && NumTraits<Scalar>::IsComplex,
|
||||
IsRowMajor ? RowMajor : ColMajor, UpLo>
|
||||
IsRowMajor ? RowMajor : ColMajor, MatrixType::InnerStrideAtCompileTime, UpLo>
|
||||
::run(size, depth,
|
||||
&actualOther.coeffRef(0,0), actualOther.outerStride(), &actualOther.coeffRef(0,0), actualOther.outerStride(),
|
||||
mat.data(), mat.outerStride(), actualAlpha, blocking);
|
||||
mat.data(), mat.innerStride(), mat.outerStride(), actualAlpha, blocking);
|
||||
}
|
||||
};
|
||||
|
||||
|
||||
@@ -45,22 +45,24 @@ template <typename Scalar, typename Index,
|
||||
int Mode, bool LhsIsTriangular,
|
||||
int LhsStorageOrder, bool ConjugateLhs,
|
||||
int RhsStorageOrder, bool ConjugateRhs,
|
||||
int ResStorageOrder, int Version = Specialized>
|
||||
int ResStorageOrder, int ResInnerStride,
|
||||
int Version = Specialized>
|
||||
struct product_triangular_matrix_matrix;
|
||||
|
||||
template <typename Scalar, typename Index,
|
||||
int Mode, bool LhsIsTriangular,
|
||||
int LhsStorageOrder, bool ConjugateLhs,
|
||||
int RhsStorageOrder, bool ConjugateRhs, int Version>
|
||||
int RhsStorageOrder, bool ConjugateRhs,
|
||||
int ResInnerStride, int Version>
|
||||
struct product_triangular_matrix_matrix<Scalar,Index,Mode,LhsIsTriangular,
|
||||
LhsStorageOrder,ConjugateLhs,
|
||||
RhsStorageOrder,ConjugateRhs,RowMajor,Version>
|
||||
RhsStorageOrder,ConjugateRhs,RowMajor,ResInnerStride,Version>
|
||||
{
|
||||
static EIGEN_STRONG_INLINE void run(
|
||||
Index rows, Index cols, Index depth,
|
||||
const Scalar* lhs, Index lhsStride,
|
||||
const Scalar* rhs, Index rhsStride,
|
||||
Scalar* res, Index resStride,
|
||||
Scalar* res, Index resIncr, Index resStride,
|
||||
const Scalar& alpha, level3_blocking<Scalar,Scalar>& blocking)
|
||||
{
|
||||
product_triangular_matrix_matrix<Scalar, Index,
|
||||
@@ -70,18 +72,19 @@ struct product_triangular_matrix_matrix<Scalar,Index,Mode,LhsIsTriangular,
|
||||
ConjugateRhs,
|
||||
LhsStorageOrder==RowMajor ? ColMajor : RowMajor,
|
||||
ConjugateLhs,
|
||||
ColMajor>
|
||||
::run(cols, rows, depth, rhs, rhsStride, lhs, lhsStride, res, resStride, alpha, blocking);
|
||||
ColMajor, ResInnerStride>
|
||||
::run(cols, rows, depth, rhs, rhsStride, lhs, lhsStride, res, resIncr, resStride, alpha, blocking);
|
||||
}
|
||||
};
|
||||
|
||||
// implements col-major += alpha * op(triangular) * op(general)
|
||||
template <typename Scalar, typename Index, int Mode,
|
||||
int LhsStorageOrder, bool ConjugateLhs,
|
||||
int RhsStorageOrder, bool ConjugateRhs, int Version>
|
||||
int RhsStorageOrder, bool ConjugateRhs,
|
||||
int ResInnerStride, int Version>
|
||||
struct product_triangular_matrix_matrix<Scalar,Index,Mode,true,
|
||||
LhsStorageOrder,ConjugateLhs,
|
||||
RhsStorageOrder,ConjugateRhs,ColMajor,Version>
|
||||
RhsStorageOrder,ConjugateRhs,ColMajor,ResInnerStride,Version>
|
||||
{
|
||||
|
||||
typedef gebp_traits<Scalar,Scalar> Traits;
|
||||
@@ -95,20 +98,21 @@ struct product_triangular_matrix_matrix<Scalar,Index,Mode,true,
|
||||
Index _rows, Index _cols, Index _depth,
|
||||
const Scalar* _lhs, Index lhsStride,
|
||||
const Scalar* _rhs, Index rhsStride,
|
||||
Scalar* res, Index resStride,
|
||||
Scalar* res, Index resIncr, Index resStride,
|
||||
const Scalar& alpha, level3_blocking<Scalar,Scalar>& blocking);
|
||||
};
|
||||
|
||||
template <typename Scalar, typename Index, int Mode,
|
||||
int LhsStorageOrder, bool ConjugateLhs,
|
||||
int RhsStorageOrder, bool ConjugateRhs, int Version>
|
||||
int RhsStorageOrder, bool ConjugateRhs,
|
||||
int ResInnerStride, int Version>
|
||||
EIGEN_DONT_INLINE void product_triangular_matrix_matrix<Scalar,Index,Mode,true,
|
||||
LhsStorageOrder,ConjugateLhs,
|
||||
RhsStorageOrder,ConjugateRhs,ColMajor,Version>::run(
|
||||
RhsStorageOrder,ConjugateRhs,ColMajor,ResInnerStride,Version>::run(
|
||||
Index _rows, Index _cols, Index _depth,
|
||||
const Scalar* _lhs, Index lhsStride,
|
||||
const Scalar* _rhs, Index rhsStride,
|
||||
Scalar* _res, Index resStride,
|
||||
Scalar* _res, Index resIncr, Index resStride,
|
||||
const Scalar& alpha, level3_blocking<Scalar,Scalar>& blocking)
|
||||
{
|
||||
// strip zeros
|
||||
@@ -119,10 +123,10 @@ EIGEN_DONT_INLINE void product_triangular_matrix_matrix<Scalar,Index,Mode,true,
|
||||
|
||||
typedef const_blas_data_mapper<Scalar, Index, LhsStorageOrder> LhsMapper;
|
||||
typedef const_blas_data_mapper<Scalar, Index, RhsStorageOrder> RhsMapper;
|
||||
typedef blas_data_mapper<typename Traits::ResScalar, Index, ColMajor> ResMapper;
|
||||
typedef blas_data_mapper<typename Traits::ResScalar, Index, ColMajor, Unaligned, ResInnerStride> ResMapper;
|
||||
LhsMapper lhs(_lhs,lhsStride);
|
||||
RhsMapper rhs(_rhs,rhsStride);
|
||||
ResMapper res(_res, resStride);
|
||||
ResMapper res(_res, resStride, resIncr);
|
||||
|
||||
Index kc = blocking.kc(); // cache block size along the K direction
|
||||
Index mc = (std::min)(rows,blocking.mc()); // cache block size along the M direction
|
||||
@@ -235,10 +239,11 @@ EIGEN_DONT_INLINE void product_triangular_matrix_matrix<Scalar,Index,Mode,true,
|
||||
// implements col-major += alpha * op(general) * op(triangular)
|
||||
template <typename Scalar, typename Index, int Mode,
|
||||
int LhsStorageOrder, bool ConjugateLhs,
|
||||
int RhsStorageOrder, bool ConjugateRhs, int Version>
|
||||
int RhsStorageOrder, bool ConjugateRhs,
|
||||
int ResInnerStride, int Version>
|
||||
struct product_triangular_matrix_matrix<Scalar,Index,Mode,false,
|
||||
LhsStorageOrder,ConjugateLhs,
|
||||
RhsStorageOrder,ConjugateRhs,ColMajor,Version>
|
||||
RhsStorageOrder,ConjugateRhs,ColMajor,ResInnerStride,Version>
|
||||
{
|
||||
typedef gebp_traits<Scalar,Scalar> Traits;
|
||||
enum {
|
||||
@@ -251,20 +256,21 @@ struct product_triangular_matrix_matrix<Scalar,Index,Mode,false,
|
||||
Index _rows, Index _cols, Index _depth,
|
||||
const Scalar* _lhs, Index lhsStride,
|
||||
const Scalar* _rhs, Index rhsStride,
|
||||
Scalar* res, Index resStride,
|
||||
Scalar* res, Index resIncr, Index resStride,
|
||||
const Scalar& alpha, level3_blocking<Scalar,Scalar>& blocking);
|
||||
};
|
||||
|
||||
template <typename Scalar, typename Index, int Mode,
|
||||
int LhsStorageOrder, bool ConjugateLhs,
|
||||
int RhsStorageOrder, bool ConjugateRhs, int Version>
|
||||
int RhsStorageOrder, bool ConjugateRhs,
|
||||
int ResInnerStride, int Version>
|
||||
EIGEN_DONT_INLINE void product_triangular_matrix_matrix<Scalar,Index,Mode,false,
|
||||
LhsStorageOrder,ConjugateLhs,
|
||||
RhsStorageOrder,ConjugateRhs,ColMajor,Version>::run(
|
||||
RhsStorageOrder,ConjugateRhs,ColMajor,ResInnerStride,Version>::run(
|
||||
Index _rows, Index _cols, Index _depth,
|
||||
const Scalar* _lhs, Index lhsStride,
|
||||
const Scalar* _rhs, Index rhsStride,
|
||||
Scalar* _res, Index resStride,
|
||||
Scalar* _res, Index resIncr, Index resStride,
|
||||
const Scalar& alpha, level3_blocking<Scalar,Scalar>& blocking)
|
||||
{
|
||||
const Index PacketBytes = packet_traits<Scalar>::size*sizeof(Scalar);
|
||||
@@ -276,10 +282,10 @@ EIGEN_DONT_INLINE void product_triangular_matrix_matrix<Scalar,Index,Mode,false,
|
||||
|
||||
typedef const_blas_data_mapper<Scalar, Index, LhsStorageOrder> LhsMapper;
|
||||
typedef const_blas_data_mapper<Scalar, Index, RhsStorageOrder> RhsMapper;
|
||||
typedef blas_data_mapper<typename Traits::ResScalar, Index, ColMajor> ResMapper;
|
||||
typedef blas_data_mapper<typename Traits::ResScalar, Index, ColMajor, Unaligned, ResInnerStride> ResMapper;
|
||||
LhsMapper lhs(_lhs,lhsStride);
|
||||
RhsMapper rhs(_rhs,rhsStride);
|
||||
ResMapper res(_res, resStride);
|
||||
ResMapper res(_res, resStride, resIncr);
|
||||
|
||||
Index kc = blocking.kc(); // cache block size along the K direction
|
||||
Index mc = (std::min)(rows,blocking.mc()); // cache block size along the M direction
|
||||
@@ -433,12 +439,12 @@ struct triangular_product_impl<Mode,LhsIsTriangular,Lhs,false,Rhs,false>
|
||||
Mode, LhsIsTriangular,
|
||||
(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>
|
||||
(internal::traits<Dest >::Flags&RowMajorBit) ? RowMajor : ColMajor, Dest::InnerStrideAtCompileTime>
|
||||
::run(
|
||||
stripedRows, stripedCols, stripedDepth, // 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
|
||||
&dst.coeffRef(0,0), dst.innerStride(), dst.outerStride(), // result info
|
||||
actualAlpha, blocking
|
||||
);
|
||||
|
||||
|
||||
@@ -15,48 +15,48 @@ namespace Eigen {
|
||||
namespace internal {
|
||||
|
||||
// if the rhs is row major, let's transpose the product
|
||||
template <typename Scalar, typename Index, int Side, int Mode, bool Conjugate, int TriStorageOrder>
|
||||
struct triangular_solve_matrix<Scalar,Index,Side,Mode,Conjugate,TriStorageOrder,RowMajor>
|
||||
template <typename Scalar, typename Index, int Side, int Mode, bool Conjugate, int TriStorageOrder, int OtherInnerStride>
|
||||
struct triangular_solve_matrix<Scalar,Index,Side,Mode,Conjugate,TriStorageOrder,RowMajor,OtherInnerStride>
|
||||
{
|
||||
static void run(
|
||||
Index size, Index cols,
|
||||
const Scalar* tri, Index triStride,
|
||||
Scalar* _other, Index otherStride,
|
||||
Scalar* _other, Index otherIncr, Index otherStride,
|
||||
level3_blocking<Scalar,Scalar>& blocking)
|
||||
{
|
||||
triangular_solve_matrix<
|
||||
Scalar, Index, Side==OnTheLeft?OnTheRight:OnTheLeft,
|
||||
(Mode&UnitDiag) | ((Mode&Upper) ? Lower : Upper),
|
||||
NumTraits<Scalar>::IsComplex && Conjugate,
|
||||
TriStorageOrder==RowMajor ? ColMajor : RowMajor, ColMajor>
|
||||
::run(size, cols, tri, triStride, _other, otherStride, blocking);
|
||||
TriStorageOrder==RowMajor ? ColMajor : RowMajor, ColMajor, OtherInnerStride>
|
||||
::run(size, cols, tri, triStride, _other, otherIncr, otherStride, blocking);
|
||||
}
|
||||
};
|
||||
|
||||
/* Optimized triangular solver with multiple right hand side and the triangular matrix on the left
|
||||
*/
|
||||
template <typename Scalar, typename Index, int Mode, bool Conjugate, int TriStorageOrder>
|
||||
struct triangular_solve_matrix<Scalar,Index,OnTheLeft,Mode,Conjugate,TriStorageOrder,ColMajor>
|
||||
template <typename Scalar, typename Index, int Mode, bool Conjugate, int TriStorageOrder,int OtherInnerStride>
|
||||
struct triangular_solve_matrix<Scalar,Index,OnTheLeft,Mode,Conjugate,TriStorageOrder,ColMajor,OtherInnerStride>
|
||||
{
|
||||
static EIGEN_DONT_INLINE void run(
|
||||
Index size, Index otherSize,
|
||||
const Scalar* _tri, Index triStride,
|
||||
Scalar* _other, Index otherStride,
|
||||
Scalar* _other, Index otherIncr, Index otherStride,
|
||||
level3_blocking<Scalar,Scalar>& blocking);
|
||||
};
|
||||
template <typename Scalar, typename Index, int Mode, bool Conjugate, int TriStorageOrder>
|
||||
EIGEN_DONT_INLINE void triangular_solve_matrix<Scalar,Index,OnTheLeft,Mode,Conjugate,TriStorageOrder,ColMajor>::run(
|
||||
template <typename Scalar, typename Index, int Mode, bool Conjugate, int TriStorageOrder, int OtherInnerStride>
|
||||
EIGEN_DONT_INLINE void triangular_solve_matrix<Scalar,Index,OnTheLeft,Mode,Conjugate,TriStorageOrder,ColMajor,OtherInnerStride>::run(
|
||||
Index size, Index otherSize,
|
||||
const Scalar* _tri, Index triStride,
|
||||
Scalar* _other, Index otherStride,
|
||||
Scalar* _other, Index otherIncr, Index otherStride,
|
||||
level3_blocking<Scalar,Scalar>& blocking)
|
||||
{
|
||||
Index cols = otherSize;
|
||||
|
||||
typedef const_blas_data_mapper<Scalar, Index, TriStorageOrder> TriMapper;
|
||||
typedef blas_data_mapper<Scalar, Index, ColMajor> OtherMapper;
|
||||
typedef blas_data_mapper<Scalar, Index, ColMajor, Unaligned, OtherInnerStride> OtherMapper;
|
||||
TriMapper tri(_tri, triStride);
|
||||
OtherMapper other(_other, otherStride);
|
||||
OtherMapper other(_other, otherStride, otherIncr);
|
||||
|
||||
typedef gebp_traits<Scalar,Scalar> Traits;
|
||||
|
||||
@@ -128,19 +128,19 @@ EIGEN_DONT_INLINE void triangular_solve_matrix<Scalar,Index,OnTheLeft,Mode,Conju
|
||||
{
|
||||
Scalar b(0);
|
||||
const Scalar* l = &tri(i,s);
|
||||
Scalar* r = &other(s,j);
|
||||
typename OtherMapper::LinearMapper r = other.getLinearMapper(s,j);
|
||||
for (Index i3=0; i3<k; ++i3)
|
||||
b += conj(l[i3]) * r[i3];
|
||||
b += conj(l[i3]) * r(i3);
|
||||
|
||||
other(i,j) = (other(i,j) - b)*a;
|
||||
}
|
||||
else
|
||||
{
|
||||
Scalar b = (other(i,j) *= a);
|
||||
Scalar* r = &other(s,j);
|
||||
const Scalar* l = &tri(s,i);
|
||||
typename OtherMapper::LinearMapper r = other.getLinearMapper(s,j);
|
||||
typename TriMapper::LinearMapper l = tri.getLinearMapper(s,i);
|
||||
for (Index i3=0;i3<rs;++i3)
|
||||
r[i3] -= b * conj(l[i3]);
|
||||
r(i3) -= b * conj(l(i3));
|
||||
}
|
||||
}
|
||||
}
|
||||
@@ -185,28 +185,28 @@ EIGEN_DONT_INLINE void triangular_solve_matrix<Scalar,Index,OnTheLeft,Mode,Conju
|
||||
|
||||
/* Optimized triangular solver with multiple left hand sides and the triangular matrix on the right
|
||||
*/
|
||||
template <typename Scalar, typename Index, int Mode, bool Conjugate, int TriStorageOrder>
|
||||
struct triangular_solve_matrix<Scalar,Index,OnTheRight,Mode,Conjugate,TriStorageOrder,ColMajor>
|
||||
template <typename Scalar, typename Index, int Mode, bool Conjugate, int TriStorageOrder, int OtherInnerStride>
|
||||
struct triangular_solve_matrix<Scalar,Index,OnTheRight,Mode,Conjugate,TriStorageOrder,ColMajor,OtherInnerStride>
|
||||
{
|
||||
static EIGEN_DONT_INLINE void run(
|
||||
Index size, Index otherSize,
|
||||
const Scalar* _tri, Index triStride,
|
||||
Scalar* _other, Index otherStride,
|
||||
Scalar* _other, Index otherIncr, Index otherStride,
|
||||
level3_blocking<Scalar,Scalar>& blocking);
|
||||
};
|
||||
template <typename Scalar, typename Index, int Mode, bool Conjugate, int TriStorageOrder>
|
||||
EIGEN_DONT_INLINE void triangular_solve_matrix<Scalar,Index,OnTheRight,Mode,Conjugate,TriStorageOrder,ColMajor>::run(
|
||||
template <typename Scalar, typename Index, int Mode, bool Conjugate, int TriStorageOrder, int OtherInnerStride>
|
||||
EIGEN_DONT_INLINE void triangular_solve_matrix<Scalar,Index,OnTheRight,Mode,Conjugate,TriStorageOrder,ColMajor,OtherInnerStride>::run(
|
||||
Index size, Index otherSize,
|
||||
const Scalar* _tri, Index triStride,
|
||||
Scalar* _other, Index otherStride,
|
||||
Scalar* _other, Index otherIncr, Index otherStride,
|
||||
level3_blocking<Scalar,Scalar>& blocking)
|
||||
{
|
||||
Index rows = otherSize;
|
||||
typedef typename NumTraits<Scalar>::Real RealScalar;
|
||||
|
||||
typedef blas_data_mapper<Scalar, Index, ColMajor> LhsMapper;
|
||||
typedef blas_data_mapper<Scalar, Index, ColMajor, Unaligned, OtherInnerStride> LhsMapper;
|
||||
typedef const_blas_data_mapper<Scalar, Index, TriStorageOrder> RhsMapper;
|
||||
LhsMapper lhs(_other, otherStride);
|
||||
LhsMapper lhs(_other, otherStride, otherIncr);
|
||||
RhsMapper rhs(_tri, triStride);
|
||||
|
||||
typedef gebp_traits<Scalar,Scalar> Traits;
|
||||
@@ -297,24 +297,24 @@ EIGEN_DONT_INLINE void triangular_solve_matrix<Scalar,Index,OnTheRight,Mode,Conj
|
||||
{
|
||||
Index j = IsLower ? absolute_j2+actualPanelWidth-k-1 : absolute_j2+k;
|
||||
|
||||
Scalar* r = &lhs(i2,j);
|
||||
typename LhsMapper::LinearMapper r = lhs.getLinearMapper(i2,j);
|
||||
for (Index k3=0; k3<k; ++k3)
|
||||
{
|
||||
Scalar b = conj(rhs(IsLower ? j+1+k3 : absolute_j2+k3,j));
|
||||
Scalar* a = &lhs(i2,IsLower ? j+1+k3 : absolute_j2+k3);
|
||||
typename LhsMapper::LinearMapper a = lhs.getLinearMapper(i2,IsLower ? j+1+k3 : absolute_j2+k3);
|
||||
for (Index i=0; i<actual_mc; ++i)
|
||||
r[i] -= a[i] * b;
|
||||
r(i) -= a(i) * b;
|
||||
}
|
||||
if((Mode & UnitDiag)==0)
|
||||
{
|
||||
Scalar inv_rjj = RealScalar(1)/conj(rhs(j,j));
|
||||
for (Index i=0; i<actual_mc; ++i)
|
||||
r[i] *= inv_rjj;
|
||||
r(i) *= inv_rjj;
|
||||
}
|
||||
}
|
||||
|
||||
// pack the just computed part of lhs to A
|
||||
pack_lhs_panel(blockA, LhsMapper(_other+absolute_j2*otherStride+i2, otherStride),
|
||||
pack_lhs_panel(blockA, lhs.getSubMapper(i2,absolute_j2),
|
||||
actualPanelWidth, actual_mc,
|
||||
actual_kc, j2);
|
||||
}
|
||||
|
||||
@@ -31,7 +31,7 @@ template<
|
||||
typename Index,
|
||||
typename LhsScalar, int LhsStorageOrder, bool ConjugateLhs,
|
||||
typename RhsScalar, int RhsStorageOrder, bool ConjugateRhs,
|
||||
int ResStorageOrder>
|
||||
int ResStorageOrder, int ResInnerStride>
|
||||
struct general_matrix_matrix_product;
|
||||
|
||||
template<typename Index,
|
||||
@@ -155,13 +155,21 @@ class BlasVectorMapper {
|
||||
Scalar* m_data;
|
||||
};
|
||||
|
||||
template<typename Scalar, typename Index, int AlignmentType, int Incr=1>
|
||||
class BlasLinearMapper;
|
||||
|
||||
template<typename Scalar, typename Index, int AlignmentType>
|
||||
class BlasLinearMapper {
|
||||
class BlasLinearMapper<Scalar,Index,AlignmentType,1> {
|
||||
public:
|
||||
typedef typename packet_traits<Scalar>::type Packet;
|
||||
typedef typename packet_traits<Scalar>::half HalfPacket;
|
||||
|
||||
EIGEN_DEVICE_FUNC EIGEN_ALWAYS_INLINE BlasLinearMapper(Scalar *data) : m_data(data) {}
|
||||
EIGEN_DEVICE_FUNC EIGEN_ALWAYS_INLINE BlasLinearMapper(Scalar *data, Index incr=1)
|
||||
: m_data(data)
|
||||
{
|
||||
EIGEN_ONLY_USED_FOR_DEBUG(incr);
|
||||
eigen_assert(incr==1);
|
||||
}
|
||||
|
||||
EIGEN_DEVICE_FUNC EIGEN_ALWAYS_INLINE void prefetch(int i) const {
|
||||
internal::prefetch(&operator()(i));
|
||||
@@ -188,16 +196,25 @@ class BlasLinearMapper {
|
||||
};
|
||||
|
||||
// Lightweight helper class to access matrix coefficients.
|
||||
template<typename Scalar, typename Index, int StorageOrder, int AlignmentType = Unaligned>
|
||||
class blas_data_mapper {
|
||||
public:
|
||||
template<typename Scalar, typename Index, int StorageOrder, int AlignmentType = Unaligned, int Incr = 1>
|
||||
class blas_data_mapper;
|
||||
|
||||
template<typename Scalar, typename Index, int StorageOrder, int AlignmentType>
|
||||
class blas_data_mapper<Scalar,Index,StorageOrder,AlignmentType,1>
|
||||
{
|
||||
public:
|
||||
typedef typename packet_traits<Scalar>::type Packet;
|
||||
typedef typename packet_traits<Scalar>::half HalfPacket;
|
||||
|
||||
typedef BlasLinearMapper<Scalar, Index, AlignmentType> LinearMapper;
|
||||
typedef BlasVectorMapper<Scalar, Index> VectorMapper;
|
||||
|
||||
EIGEN_DEVICE_FUNC EIGEN_ALWAYS_INLINE blas_data_mapper(Scalar* data, Index stride) : m_data(data), m_stride(stride) {}
|
||||
EIGEN_DEVICE_FUNC EIGEN_ALWAYS_INLINE blas_data_mapper(Scalar* data, Index stride, Index incr=1)
|
||||
: m_data(data), m_stride(stride)
|
||||
{
|
||||
EIGEN_ONLY_USED_FOR_DEBUG(incr);
|
||||
eigen_assert(incr==1);
|
||||
}
|
||||
|
||||
EIGEN_DEVICE_FUNC EIGEN_ALWAYS_INLINE blas_data_mapper<Scalar, Index, StorageOrder, AlignmentType>
|
||||
getSubMapper(Index i, Index j) const {
|
||||
@@ -251,6 +268,90 @@ class blas_data_mapper {
|
||||
const Index m_stride;
|
||||
};
|
||||
|
||||
// Implementation of non-natural increment (i.e. inner-stride != 1)
|
||||
// The exposed API is not complete yet compared to the Incr==1 case
|
||||
// because some features makes less sense in this case.
|
||||
template<typename Scalar, typename Index, int AlignmentType, int Incr>
|
||||
class BlasLinearMapper
|
||||
{
|
||||
public:
|
||||
typedef typename packet_traits<Scalar>::type Packet;
|
||||
typedef typename packet_traits<Scalar>::half HalfPacket;
|
||||
|
||||
EIGEN_DEVICE_FUNC EIGEN_ALWAYS_INLINE BlasLinearMapper(Scalar *data,Index incr) : m_data(data), m_incr(incr) {}
|
||||
|
||||
EIGEN_DEVICE_FUNC EIGEN_ALWAYS_INLINE void prefetch(int i) const {
|
||||
internal::prefetch(&operator()(i));
|
||||
}
|
||||
|
||||
EIGEN_DEVICE_FUNC EIGEN_ALWAYS_INLINE Scalar& operator()(Index i) const {
|
||||
return m_data[i*m_incr.value()];
|
||||
}
|
||||
|
||||
EIGEN_DEVICE_FUNC EIGEN_ALWAYS_INLINE Packet loadPacket(Index i) const {
|
||||
return pgather<Scalar,Packet>(m_data + i*m_incr.value(), m_incr.value());
|
||||
}
|
||||
|
||||
template<typename PacketType>
|
||||
EIGEN_DEVICE_FUNC EIGEN_ALWAYS_INLINE void storePacket(Index i, const PacketType &p) const {
|
||||
pscatter<Scalar, PacketType>(m_data + i*m_incr.value(), p, m_incr.value());
|
||||
}
|
||||
|
||||
protected:
|
||||
Scalar *m_data;
|
||||
const internal::variable_if_dynamic<Index,Incr> m_incr;
|
||||
};
|
||||
|
||||
template<typename Scalar, typename Index, int StorageOrder, int AlignmentType,int Incr>
|
||||
class blas_data_mapper
|
||||
{
|
||||
public:
|
||||
typedef typename packet_traits<Scalar>::type Packet;
|
||||
typedef typename packet_traits<Scalar>::half HalfPacket;
|
||||
|
||||
typedef BlasLinearMapper<Scalar, Index, AlignmentType,Incr> LinearMapper;
|
||||
|
||||
EIGEN_DEVICE_FUNC EIGEN_ALWAYS_INLINE blas_data_mapper(Scalar* data, Index stride, Index incr) : m_data(data), m_stride(stride), m_incr(incr) {}
|
||||
|
||||
EIGEN_DEVICE_FUNC EIGEN_ALWAYS_INLINE blas_data_mapper
|
||||
getSubMapper(Index i, Index j) const {
|
||||
return blas_data_mapper(&operator()(i, j), m_stride, m_incr.value());
|
||||
}
|
||||
|
||||
EIGEN_DEVICE_FUNC EIGEN_ALWAYS_INLINE LinearMapper getLinearMapper(Index i, Index j) const {
|
||||
return LinearMapper(&operator()(i, j), m_incr.value());
|
||||
}
|
||||
|
||||
EIGEN_DEVICE_FUNC
|
||||
EIGEN_ALWAYS_INLINE Scalar& operator()(Index i, Index j) const {
|
||||
return m_data[StorageOrder==RowMajor ? j*m_incr.value() + i*m_stride : i*m_incr.value() + j*m_stride];
|
||||
}
|
||||
|
||||
EIGEN_DEVICE_FUNC EIGEN_ALWAYS_INLINE Packet loadPacket(Index i, Index j) const {
|
||||
return pgather<Scalar,Packet>(&operator()(i, j),m_incr.value());
|
||||
}
|
||||
|
||||
template <typename PacketT, int AlignmentT>
|
||||
EIGEN_DEVICE_FUNC EIGEN_ALWAYS_INLINE PacketT load(Index i, Index j) const {
|
||||
return pgather<Scalar,PacketT>(&operator()(i, j),m_incr.value());
|
||||
}
|
||||
|
||||
template<typename SubPacket>
|
||||
EIGEN_DEVICE_FUNC EIGEN_ALWAYS_INLINE void scatterPacket(Index i, Index j, const SubPacket &p) const {
|
||||
pscatter<Scalar, SubPacket>(&operator()(i, j), p, m_stride);
|
||||
}
|
||||
|
||||
template<typename SubPacket>
|
||||
EIGEN_DEVICE_FUNC EIGEN_ALWAYS_INLINE SubPacket gatherPacket(Index i, Index j) const {
|
||||
return pgather<Scalar, SubPacket>(&operator()(i, j), m_stride);
|
||||
}
|
||||
|
||||
protected:
|
||||
Scalar* EIGEN_RESTRICT m_data;
|
||||
const Index m_stride;
|
||||
const internal::variable_if_dynamic<Index,Incr> m_incr;
|
||||
};
|
||||
|
||||
// lightweight helper class to access matrix coefficients (const version)
|
||||
template<typename Scalar, typename Index, int StorageOrder>
|
||||
class const_blas_data_mapper : public blas_data_mapper<const Scalar, Index, StorageOrder> {
|
||||
|
||||
@@ -57,10 +57,10 @@
|
||||
#if __GNUC__>=6
|
||||
#pragma GCC diagnostic ignored "-Wignored-attributes"
|
||||
#endif
|
||||
#if __GNUC__>=9
|
||||
#pragma GCC diagnostic ignored "-Wdeprecated-copy"
|
||||
#if __GNUC__==7
|
||||
// See: https://gcc.gnu.org/bugzilla/show_bug.cgi?id=89325
|
||||
#pragma GCC diagnostic ignored "-Wattributes"
|
||||
#endif
|
||||
|
||||
#endif
|
||||
|
||||
#if defined __NVCC__
|
||||
@@ -83,4 +83,12 @@
|
||||
#pragma diag_suppress 2737
|
||||
#endif
|
||||
|
||||
#else
|
||||
// warnings already disabled:
|
||||
# ifndef EIGEN_WARNINGS_DISABLED_2
|
||||
# define EIGEN_WARNINGS_DISABLED_2
|
||||
# elif defined(EIGEN_INTERNAL_DEBUGGING)
|
||||
# error "Do not include \"DisableStupidWarnings.h\" recursively more than twice!"
|
||||
# endif
|
||||
|
||||
#endif // not EIGEN_WARNINGS_DISABLED
|
||||
|
||||
@@ -47,11 +47,7 @@ template<typename T> struct NumTraits;
|
||||
template<typename Derived> struct EigenBase;
|
||||
template<typename Derived> class DenseBase;
|
||||
template<typename Derived> class PlainObjectBase;
|
||||
|
||||
|
||||
template<typename Derived,
|
||||
int Level = internal::accessors_level<Derived>::value >
|
||||
class DenseCoeffsBase;
|
||||
template<typename Derived, int Level> class DenseCoeffsBase;
|
||||
|
||||
template<typename _Scalar, int _Rows, int _Cols,
|
||||
int _Options = AutoAlign |
|
||||
|
||||
@@ -13,7 +13,7 @@
|
||||
|
||||
#define EIGEN_WORLD_VERSION 3
|
||||
#define EIGEN_MAJOR_VERSION 3
|
||||
#define EIGEN_MINOR_VERSION 7
|
||||
#define EIGEN_MINOR_VERSION 9
|
||||
|
||||
#define EIGEN_VERSION_AT_LEAST(x,y,z) (EIGEN_WORLD_VERSION>x || (EIGEN_WORLD_VERSION>=x && \
|
||||
(EIGEN_MAJOR_VERSION>y || (EIGEN_MAJOR_VERSION>=y && \
|
||||
@@ -380,7 +380,8 @@
|
||||
#if EIGEN_MAX_CPP_VER>=11 && \
|
||||
((defined(__STDC_VERSION__) && (__STDC_VERSION__ >= 199901)) \
|
||||
|| (defined(__GNUC__) && defined(_GLIBCXX_USE_C99)) \
|
||||
|| (defined(_LIBCPP_VERSION) && !defined(_MSC_VER)))
|
||||
|| (defined(_LIBCPP_VERSION) && !defined(_MSC_VER)) \
|
||||
|| (EIGEN_COMP_MSVC >= 1900) )
|
||||
#define EIGEN_HAS_C99_MATH 1
|
||||
#else
|
||||
#define EIGEN_HAS_C99_MATH 0
|
||||
@@ -396,6 +397,20 @@
|
||||
#endif
|
||||
#endif
|
||||
|
||||
// Does the compiler support type_traits?
|
||||
// - full support of type traits was added only to GCC 5.1.0.
|
||||
// - 20150626 corresponds to the last release of 4.x libstdc++
|
||||
#ifndef EIGEN_HAS_TYPE_TRAITS
|
||||
#if EIGEN_MAX_CPP_VER>=11 && (EIGEN_HAS_CXX11 || EIGEN_COMP_MSVC >= 1700) \
|
||||
&& ((!EIGEN_COMP_GNUC_STRICT) || EIGEN_GNUC_AT_LEAST(5, 1)) \
|
||||
&& ((!defined(__GLIBCXX__)) || __GLIBCXX__ > 20150626)
|
||||
#define EIGEN_HAS_TYPE_TRAITS 1
|
||||
#define EIGEN_INCLUDE_TYPE_TRAITS
|
||||
#else
|
||||
#define EIGEN_HAS_TYPE_TRAITS 0
|
||||
#endif
|
||||
#endif
|
||||
|
||||
// Does the compiler support variadic templates?
|
||||
#ifndef EIGEN_HAS_VARIADIC_TEMPLATES
|
||||
#if EIGEN_MAX_CPP_VER>=11 && (__cplusplus > 199711L || EIGEN_COMP_MSVC >= 1900) \
|
||||
@@ -847,6 +862,7 @@ namespace Eigen {
|
||||
#endif
|
||||
|
||||
|
||||
|
||||
/** \internal
|
||||
* \brief Macro to manually inherit assignment operators.
|
||||
* This is necessary, because the implicitly defined assignment operator gets deleted when a custom operator= is defined.
|
||||
@@ -874,6 +890,9 @@ namespace Eigen {
|
||||
#endif
|
||||
|
||||
|
||||
|
||||
|
||||
|
||||
/**
|
||||
* Just a side note. Commenting within defines works only by documenting
|
||||
* behind the object (via '!<'). Comments cannot be multi-line and thus
|
||||
|
||||
@@ -97,6 +97,9 @@ template<> struct is_arithmetic<unsigned int> { enum { value = true }; };
|
||||
template<> struct is_arithmetic<signed long> { enum { value = true }; };
|
||||
template<> struct is_arithmetic<unsigned long> { enum { value = true }; };
|
||||
|
||||
#if EIGEN_HAS_CXX11
|
||||
using std::is_integral;
|
||||
#else
|
||||
template<typename T> struct is_integral { enum { value = false }; };
|
||||
template<> struct is_integral<bool> { enum { value = true }; };
|
||||
template<> struct is_integral<char> { enum { value = true }; };
|
||||
@@ -108,6 +111,11 @@ template<> struct is_integral<signed int> { enum { value = true }; };
|
||||
template<> struct is_integral<unsigned int> { enum { value = true }; };
|
||||
template<> struct is_integral<signed long> { enum { value = true }; };
|
||||
template<> struct is_integral<unsigned long> { enum { value = true }; };
|
||||
#if EIGEN_COMP_MSVC
|
||||
template<> struct is_integral<signed __int64> { enum { value = true }; };
|
||||
template<> struct is_integral<unsigned __int64>{ enum { value = true }; };
|
||||
#endif
|
||||
#endif
|
||||
|
||||
#if EIGEN_HAS_CXX11
|
||||
using std::make_unsigned;
|
||||
@@ -531,4 +539,30 @@ bool not_equal_strict(const double& x,const double& y) { return std::not_equal_t
|
||||
|
||||
} // end namespace Eigen
|
||||
|
||||
// Define portable (u)int{32,64} types
|
||||
#if EIGEN_HAS_CXX11
|
||||
#include <cstdint>
|
||||
namespace Eigen {
|
||||
namespace numext {
|
||||
typedef std::uint32_t uint32_t;
|
||||
typedef std::int32_t int32_t;
|
||||
typedef std::uint64_t uint64_t;
|
||||
typedef std::int64_t int64_t;
|
||||
}
|
||||
}
|
||||
#else
|
||||
// Without c++11, all compilers able to compile Eigen also
|
||||
// provides the C99 stdint.h header file.
|
||||
#include <stdint.h>
|
||||
namespace Eigen {
|
||||
namespace numext {
|
||||
typedef ::uint32_t uint32_t;
|
||||
typedef ::int32_t int32_t;
|
||||
typedef ::uint64_t uint64_t;
|
||||
typedef ::int64_t int64_t;
|
||||
}
|
||||
}
|
||||
#endif
|
||||
|
||||
|
||||
#endif // EIGEN_META_H
|
||||
|
||||
@@ -1,4 +1,8 @@
|
||||
#ifdef EIGEN_WARNINGS_DISABLED
|
||||
#ifdef EIGEN_WARNINGS_DISABLED_2
|
||||
// "DisableStupidWarnings.h" was included twice recursively: Do not reenable warnings yet!
|
||||
# undef EIGEN_WARNINGS_DISABLED_2
|
||||
|
||||
#elif defined(EIGEN_WARNINGS_DISABLED)
|
||||
#undef EIGEN_WARNINGS_DISABLED
|
||||
|
||||
#ifndef EIGEN_PERMANENTLY_DISABLE_STUPID_WARNINGS
|
||||
|
||||
@@ -34,6 +34,20 @@ inline IndexDest convert_index(const IndexSrc& idx) {
|
||||
return IndexDest(idx);
|
||||
}
|
||||
|
||||
// true if T can be considered as an integral index (i.e., and integral type or enum)
|
||||
template<typename T> struct is_valid_index_type
|
||||
{
|
||||
enum { value =
|
||||
#if EIGEN_HAS_TYPE_TRAITS
|
||||
internal::is_integral<T>::value || std::is_enum<T>::value
|
||||
#elif EIGEN_COMP_MSVC
|
||||
internal::is_integral<T>::value || __is_enum(T)
|
||||
#else
|
||||
// without C++11, we use is_convertible to Index instead of is_integral in order to treat enums as Index.
|
||||
internal::is_convertible<T,Index>::value && !internal::is_same<T,float>::value && !is_same<T,double>::value
|
||||
#endif
|
||||
};
|
||||
};
|
||||
|
||||
// promote_scalar_arg is an helper used in operation between an expression and a scalar, like:
|
||||
// expression * scalar
|
||||
|
||||
@@ -300,10 +300,13 @@ typename ComplexSchur<MatrixType>::ComplexScalar ComplexSchur<MatrixType>::compu
|
||||
ComplexScalar trace = t.coeff(0,0) + t.coeff(1,1);
|
||||
ComplexScalar eival1 = (trace + disc) / RealScalar(2);
|
||||
ComplexScalar eival2 = (trace - disc) / RealScalar(2);
|
||||
|
||||
if(numext::norm1(eival1) > numext::norm1(eival2))
|
||||
RealScalar eival1_norm = numext::norm1(eival1);
|
||||
RealScalar eival2_norm = numext::norm1(eival2);
|
||||
// A division by zero can only occur if eival1==eival2==0.
|
||||
// In this case, det==0, and all we have to do is checking that eival2_norm!=0
|
||||
if(eival1_norm > eival2_norm)
|
||||
eival2 = det / eival1;
|
||||
else
|
||||
else if(eival2_norm!=RealScalar(0))
|
||||
eival1 = det / eival2;
|
||||
|
||||
// choose the eigenvalue closest to the bottom entry of the diagonal
|
||||
|
||||
@@ -236,7 +236,7 @@ template<typename _MatrixType> class RealSchur
|
||||
typedef Matrix<Scalar,3,1> Vector3s;
|
||||
|
||||
Scalar computeNormOfT();
|
||||
Index findSmallSubdiagEntry(Index iu);
|
||||
Index findSmallSubdiagEntry(Index iu, const Scalar& considerAsZero);
|
||||
void splitOffTwoRows(Index iu, bool computeU, const Scalar& exshift);
|
||||
void computeShift(Index iu, Index iter, Scalar& exshift, Vector3s& shiftInfo);
|
||||
void initFrancisQRStep(Index il, Index iu, const Vector3s& shiftInfo, Index& im, Vector3s& firstHouseholderVector);
|
||||
@@ -302,12 +302,16 @@ RealSchur<MatrixType>& RealSchur<MatrixType>::computeFromHessenberg(const HessMa
|
||||
Index totalIter = 0; // iteration count for whole matrix
|
||||
Scalar exshift(0); // sum of exceptional shifts
|
||||
Scalar norm = computeNormOfT();
|
||||
// sub-diagonal entries smaller than considerAsZero will be treated as zero.
|
||||
// We use eps^2 to enable more precision in small eigenvalues.
|
||||
Scalar considerAsZero = numext::maxi<Scalar>( norm * numext::abs2(NumTraits<Scalar>::epsilon()),
|
||||
(std::numeric_limits<Scalar>::min)() );
|
||||
|
||||
if(norm!=Scalar(0))
|
||||
{
|
||||
while (iu >= 0)
|
||||
{
|
||||
Index il = findSmallSubdiagEntry(iu);
|
||||
Index il = findSmallSubdiagEntry(iu,considerAsZero);
|
||||
|
||||
// Check for convergence
|
||||
if (il == iu) // One root found
|
||||
@@ -364,14 +368,17 @@ inline typename MatrixType::Scalar RealSchur<MatrixType>::computeNormOfT()
|
||||
|
||||
/** \internal Look for single small sub-diagonal element and returns its index */
|
||||
template<typename MatrixType>
|
||||
inline Index RealSchur<MatrixType>::findSmallSubdiagEntry(Index iu)
|
||||
inline Index RealSchur<MatrixType>::findSmallSubdiagEntry(Index iu, const Scalar& considerAsZero)
|
||||
{
|
||||
using std::abs;
|
||||
Index res = iu;
|
||||
while (res > 0)
|
||||
{
|
||||
Scalar s = abs(m_matT.coeff(res-1,res-1)) + abs(m_matT.coeff(res,res));
|
||||
if (abs(m_matT.coeff(res,res-1)) <= NumTraits<Scalar>::epsilon() * s)
|
||||
|
||||
s = numext::maxi<Scalar>(s * NumTraits<Scalar>::epsilon(), considerAsZero);
|
||||
|
||||
if (abs(m_matT.coeff(res,res-1)) <= s)
|
||||
break;
|
||||
res--;
|
||||
}
|
||||
|
||||
@@ -605,7 +605,8 @@ template<typename SolverType> struct direct_selfadjoint_eigenvalues<SolverType,3
|
||||
EIGEN_DEVICE_FUNC
|
||||
static inline bool extract_kernel(MatrixType& mat, Ref<VectorType> res, Ref<VectorType> representative)
|
||||
{
|
||||
using std::abs;
|
||||
EIGEN_USING_STD_MATH(sqrt)
|
||||
EIGEN_USING_STD_MATH(abs)
|
||||
Index i0;
|
||||
// Find non-zero column i0 (by construction, there must exist a non zero coefficient on the diagonal):
|
||||
mat.diagonal().cwiseAbs().maxCoeff(&i0);
|
||||
@@ -616,8 +617,8 @@ template<typename SolverType> struct direct_selfadjoint_eigenvalues<SolverType,3
|
||||
VectorType c0, c1;
|
||||
n0 = (c0 = representative.cross(mat.col((i0+1)%3))).squaredNorm();
|
||||
n1 = (c1 = representative.cross(mat.col((i0+2)%3))).squaredNorm();
|
||||
if(n0>n1) res = c0/std::sqrt(n0);
|
||||
else res = c1/std::sqrt(n1);
|
||||
if(n0>n1) res = c0/sqrt(n0);
|
||||
else res = c1/sqrt(n1);
|
||||
|
||||
return true;
|
||||
}
|
||||
|
||||
@@ -519,7 +519,10 @@ void PartialPivLU<MatrixType>::compute()
|
||||
// the row permutation is stored as int indices, so just to be sure:
|
||||
eigen_assert(m_lu.rows()<NumTraits<int>::highest());
|
||||
|
||||
m_l1_norm = m_lu.cwiseAbs().colwise().sum().maxCoeff();
|
||||
if(m_lu.cols()>0)
|
||||
m_l1_norm = m_lu.cwiseAbs().colwise().sum().maxCoeff();
|
||||
else
|
||||
m_l1_norm = RealScalar(0);
|
||||
|
||||
eigen_assert(m_lu.rows() == m_lu.cols() && "PartialPivLU is only for square (and moreover invertible) matrices");
|
||||
const Index size = m_lu.rows();
|
||||
|
||||
@@ -44,7 +44,7 @@ struct compute_inverse_size4<Architecture::SSE, float, MatrixType, ResultType>
|
||||
static void run(const MatrixType& mat, ResultType& result)
|
||||
{
|
||||
ActualMatrixType matrix(mat);
|
||||
EIGEN_ALIGN16 const unsigned int _Sign_PNNP[4] = { 0x00000000, 0x80000000, 0x80000000, 0x00000000 };
|
||||
const Packet4f p4f_sign_PNNP = _mm_castsi128_ps(_mm_set_epi32(0x00000000, 0x80000000, 0x80000000, 0x00000000));
|
||||
|
||||
// Load the full matrix into registers
|
||||
__m128 _L1 = matrix.template packet<MatrixAlignment>( 0);
|
||||
@@ -139,7 +139,7 @@ struct compute_inverse_size4<Architecture::SSE, float, MatrixType, ResultType>
|
||||
iC = _mm_sub_ps(iC, _mm_mul_ps(_mm_shuffle_ps(A,A,0xB1), _mm_shuffle_ps(DC,DC,0x66)));
|
||||
|
||||
rd = _mm_shuffle_ps(rd,rd,0);
|
||||
rd = _mm_xor_ps(rd, _mm_load_ps((float*)_Sign_PNNP));
|
||||
rd = _mm_xor_ps(rd, p4f_sign_PNNP);
|
||||
|
||||
// iB = C*|B| - D*B#*A
|
||||
iB = _mm_sub_ps(_mm_mul_ps(C,_mm_shuffle_ps(dB,dB,0)), iB);
|
||||
|
||||
@@ -768,6 +768,21 @@ void BDCSVD<MatrixType>::computeSingVals(const ArrayRef& col0, const ArrayRef& d
|
||||
// measure everything relative to shift
|
||||
Map<ArrayXr> diagShifted(m_workspace.data()+4*n, n);
|
||||
diagShifted = diag - shift;
|
||||
|
||||
if(k!=actual_n-1)
|
||||
{
|
||||
// check that after the shift, f(mid) is still negative:
|
||||
RealScalar midShifted = (right - left) / RealScalar(2);
|
||||
if(shift==right)
|
||||
midShifted = -midShifted;
|
||||
RealScalar fMidShifted = secularEq(midShifted, col0, diag, perm, diagShifted, shift);
|
||||
if(fMidShifted>0)
|
||||
{
|
||||
// fMid was erroneous, fix it:
|
||||
shift = fMidShifted > Literal(0) ? left : right;
|
||||
diagShifted = diag - shift;
|
||||
}
|
||||
}
|
||||
|
||||
// initial guess
|
||||
RealScalar muPrev, muCur;
|
||||
@@ -845,11 +860,13 @@ void BDCSVD<MatrixType>::computeSingVals(const ArrayRef& col0, const ArrayRef& d
|
||||
}
|
||||
|
||||
RealScalar fLeft = secularEq(leftShifted, col0, diag, perm, diagShifted, shift);
|
||||
eigen_internal_assert(fLeft<Literal(0));
|
||||
|
||||
#if defined EIGEN_INTERNAL_DEBUGGING || defined EIGEN_BDCSVD_DEBUG_VERBOSE
|
||||
RealScalar fRight = secularEq(rightShifted, col0, diag, perm, diagShifted, shift);
|
||||
#endif
|
||||
|
||||
|
||||
#ifdef EIGEN_BDCSVD_DEBUG_VERBOSE
|
||||
if(!(fLeft * fRight<0))
|
||||
{
|
||||
@@ -858,23 +875,37 @@ void BDCSVD<MatrixType>::computeSingVals(const ArrayRef& col0, const ArrayRef& d
|
||||
}
|
||||
#endif
|
||||
eigen_internal_assert(fLeft * fRight < Literal(0));
|
||||
|
||||
while (rightShifted - leftShifted > Literal(2) * NumTraits<RealScalar>::epsilon() * numext::maxi<RealScalar>(abs(leftShifted), abs(rightShifted)))
|
||||
{
|
||||
RealScalar midShifted = (leftShifted + rightShifted) / Literal(2);
|
||||
fMid = secularEq(midShifted, col0, diag, perm, diagShifted, shift);
|
||||
if (fLeft * fMid < Literal(0))
|
||||
{
|
||||
rightShifted = midShifted;
|
||||
}
|
||||
else
|
||||
{
|
||||
leftShifted = midShifted;
|
||||
fLeft = fMid;
|
||||
}
|
||||
}
|
||||
|
||||
muCur = (leftShifted + rightShifted) / Literal(2);
|
||||
if(fLeft<Literal(0))
|
||||
{
|
||||
while (rightShifted - leftShifted > Literal(2) * NumTraits<RealScalar>::epsilon() * numext::maxi<RealScalar>(abs(leftShifted), abs(rightShifted)))
|
||||
{
|
||||
RealScalar midShifted = (leftShifted + rightShifted) / Literal(2);
|
||||
fMid = secularEq(midShifted, col0, diag, perm, diagShifted, shift);
|
||||
eigen_internal_assert((numext::isfinite)(fMid));
|
||||
|
||||
if (fLeft * fMid < Literal(0))
|
||||
{
|
||||
rightShifted = midShifted;
|
||||
}
|
||||
else
|
||||
{
|
||||
leftShifted = midShifted;
|
||||
fLeft = fMid;
|
||||
}
|
||||
}
|
||||
muCur = (leftShifted + rightShifted) / Literal(2);
|
||||
}
|
||||
else
|
||||
{
|
||||
// We have a problem as shifting on the left or right give either a positive or negative value
|
||||
// at the middle of [left,right]...
|
||||
// Instead fo abbording or entering an infinite loop,
|
||||
// let's just use the middle as the estimated zero-crossing:
|
||||
muCur = (right - left) * RealScalar(0.5);
|
||||
if(shift == right)
|
||||
muCur = -muCur;
|
||||
}
|
||||
}
|
||||
|
||||
singVals[k] = shift + muCur;
|
||||
@@ -924,7 +955,7 @@ void BDCSVD<MatrixType>::perturbCol0
|
||||
Index j = i<k ? i : perm(l-1);
|
||||
prod *= ((singVals(j)+dk) / ((diag(i)+dk))) * ((mus(j)+(shifts(j)-dk)) / ((diag(i)-dk)));
|
||||
#ifdef EIGEN_BDCSVD_DEBUG_VERBOSE
|
||||
if(i!=k && std::abs(((singVals(j)+dk)*(mus(j)+(shifts(j)-dk)))/((diag(i)+dk)*(diag(i)-dk)) - 1) > 0.9 )
|
||||
if(i!=k && numext::abs(((singVals(j)+dk)*(mus(j)+(shifts(j)-dk)))/((diag(i)+dk)*(diag(i)-dk)) - 1) > 0.9 )
|
||||
std::cout << " " << ((singVals(j)+dk)*(mus(j)+(shifts(j)-dk)))/((diag(i)+dk)*(diag(i)-dk)) << " == (" << (singVals(j)+dk) << " * " << (mus(j)+(shifts(j)-dk))
|
||||
<< ") / (" << (diag(i)+dk) << " * " << (diag(i)-dk) << ")\n";
|
||||
#endif
|
||||
@@ -934,7 +965,7 @@ void BDCSVD<MatrixType>::perturbCol0
|
||||
std::cout << "zhat(" << k << ") = sqrt( " << prod << ") ; " << (singVals(last) + dk) << " * " << mus(last) + shifts(last) << " - " << dk << "\n";
|
||||
#endif
|
||||
RealScalar tmp = sqrt(prod);
|
||||
zhat(k) = col0(k) > Literal(0) ? tmp : -tmp;
|
||||
zhat(k) = col0(k) > Literal(0) ? RealScalar(tmp) : RealScalar(-tmp);
|
||||
}
|
||||
}
|
||||
}
|
||||
|
||||
@@ -183,7 +183,7 @@ public:
|
||||
// this temporary is needed to workaround a MSVC issue
|
||||
Index diagSize = (std::max<Index>)(1,m_diagSize);
|
||||
return m_usePrescribedThreshold ? m_prescribedThreshold
|
||||
: diagSize*NumTraits<Scalar>::epsilon();
|
||||
: RealScalar(diagSize)*NumTraits<Scalar>::epsilon();
|
||||
}
|
||||
|
||||
/** \returns true if \a U (full or thin) is asked for in this SVD decomposition */
|
||||
|
||||
@@ -98,8 +98,10 @@ namespace std {
|
||||
{ return deque_base::insert(position,x); }
|
||||
void insert(const_iterator position, size_type new_size, const value_type& x)
|
||||
{ deque_base::insert(position, new_size, x); }
|
||||
#elif defined(_GLIBCXX_DEQUE) && EIGEN_GNUC_AT_LEAST(4,2)
|
||||
#elif defined(_GLIBCXX_DEQUE) && EIGEN_GNUC_AT_LEAST(4,2) && !EIGEN_GNUC_AT_LEAST(10, 1)
|
||||
// workaround GCC std::deque implementation
|
||||
// GCC 10.1 doesn't let us access _Deque_impl _M_impl anymore and we have to
|
||||
// fall-back to the default case
|
||||
void resize(size_type new_size, const value_type& x)
|
||||
{
|
||||
if (new_size < deque_base::size())
|
||||
@@ -108,7 +110,7 @@ namespace std {
|
||||
deque_base::insert(deque_base::end(), new_size - deque_base::size(), x);
|
||||
}
|
||||
#else
|
||||
// either GCC 4.1 or non-GCC
|
||||
// either non-GCC or GCC between 4.1 and 10.1
|
||||
// default implementation which should always work.
|
||||
void resize(size_type new_size, const value_type& x)
|
||||
{
|
||||
|
||||
@@ -119,7 +119,7 @@ OP(const Scalar& s) const { \
|
||||
return this->OP(Derived::PlainObject::Constant(rows(), cols(), s)); \
|
||||
} \
|
||||
EIGEN_DEVICE_FUNC friend EIGEN_STRONG_INLINE const RCmp ## COMPARATOR ## ReturnType \
|
||||
OP(const Scalar& s, const Derived& d) { \
|
||||
OP(const Scalar& s, const EIGEN_CURRENT_STORAGE_BASE_CLASS<Derived>& d) { \
|
||||
return Derived::PlainObject::Constant(d.rows(), d.cols(), s).OP(d); \
|
||||
}
|
||||
|
||||
|
||||
@@ -453,6 +453,24 @@ struct auto_diff_special_op<_DerType, false>
|
||||
void operator+() const;
|
||||
};
|
||||
|
||||
template<typename BinOp, typename A, typename B, typename RefType>
|
||||
void make_coherent_expression(CwiseBinaryOp<BinOp,A,B> xpr, const RefType &ref)
|
||||
{
|
||||
make_coherent(xpr.const_cast_derived().lhs(), ref);
|
||||
make_coherent(xpr.const_cast_derived().rhs(), ref);
|
||||
}
|
||||
|
||||
template<typename UnaryOp, typename A, typename RefType>
|
||||
void make_coherent_expression(const CwiseUnaryOp<UnaryOp,A> &xpr, const RefType &ref)
|
||||
{
|
||||
make_coherent(xpr.nestedExpression().const_cast_derived(), ref);
|
||||
}
|
||||
|
||||
// needed for compilation only
|
||||
template<typename UnaryOp, typename A, typename RefType>
|
||||
void make_coherent_expression(const CwiseNullaryOp<UnaryOp,A> &, const RefType &)
|
||||
{}
|
||||
|
||||
template<typename A_Scalar, int A_Rows, int A_Cols, int A_Options, int A_MaxRows, int A_MaxCols, typename B>
|
||||
struct make_coherent_impl<Matrix<A_Scalar, A_Rows, A_Cols, A_Options, A_MaxRows, A_MaxCols>, B> {
|
||||
typedef Matrix<A_Scalar, A_Rows, A_Cols, A_Options, A_MaxRows, A_MaxCols> A;
|
||||
@@ -462,6 +480,10 @@ struct make_coherent_impl<Matrix<A_Scalar, A_Rows, A_Cols, A_Options, A_MaxRows,
|
||||
a.resize(b.size());
|
||||
a.setZero();
|
||||
}
|
||||
else if (B::SizeAtCompileTime==Dynamic && a.size()!=0 && b.size()==0)
|
||||
{
|
||||
make_coherent_expression(b,a);
|
||||
}
|
||||
}
|
||||
};
|
||||
|
||||
@@ -474,13 +496,17 @@ struct make_coherent_impl<A, Matrix<B_Scalar, B_Rows, B_Cols, B_Options, B_MaxRo
|
||||
b.resize(a.size());
|
||||
b.setZero();
|
||||
}
|
||||
else if (A::SizeAtCompileTime==Dynamic && b.size()!=0 && a.size()==0)
|
||||
{
|
||||
make_coherent_expression(a,b);
|
||||
}
|
||||
}
|
||||
};
|
||||
|
||||
template<typename A_Scalar, int A_Rows, int A_Cols, int A_Options, int A_MaxRows, int A_MaxCols,
|
||||
typename B_Scalar, int B_Rows, int B_Cols, int B_Options, int B_MaxRows, int B_MaxCols>
|
||||
struct make_coherent_impl<Matrix<A_Scalar, A_Rows, A_Cols, A_Options, A_MaxRows, A_MaxCols>,
|
||||
Matrix<B_Scalar, B_Rows, B_Cols, B_Options, B_MaxRows, B_MaxCols> > {
|
||||
Matrix<B_Scalar, B_Rows, B_Cols, B_Options, B_MaxRows, B_MaxCols> > {
|
||||
typedef Matrix<A_Scalar, A_Rows, A_Cols, A_Options, A_MaxRows, A_MaxCols> A;
|
||||
typedef Matrix<B_Scalar, B_Rows, B_Cols, B_Options, B_MaxRows, B_MaxCols> B;
|
||||
static void run(A& a, B& b) {
|
||||
|
||||
@@ -412,7 +412,7 @@ template<typename Derived> struct MatrixExponentialReturnValue
|
||||
inline void evalTo(ResultType& result) const
|
||||
{
|
||||
const typename internal::nested_eval<Derived, 10>::type tmp(m_src);
|
||||
internal::matrix_exp_compute(tmp, result, internal::is_exp_known_type<typename Derived::Scalar>());
|
||||
internal::matrix_exp_compute(tmp, result, internal::is_exp_known_type<typename Derived::RealScalar>());
|
||||
}
|
||||
|
||||
Index rows() const { return m_src.rows(); }
|
||||
|
||||
@@ -253,18 +253,19 @@ struct matrix_sqrt_compute
|
||||
template <typename MatrixType>
|
||||
struct matrix_sqrt_compute<MatrixType, 0>
|
||||
{
|
||||
typedef typename MatrixType::PlainObject PlainType;
|
||||
template <typename ResultType>
|
||||
static void run(const MatrixType &arg, ResultType &result)
|
||||
{
|
||||
eigen_assert(arg.rows() == arg.cols());
|
||||
|
||||
// Compute Schur decomposition of arg
|
||||
const RealSchur<MatrixType> schurOfA(arg);
|
||||
const MatrixType& T = schurOfA.matrixT();
|
||||
const MatrixType& U = schurOfA.matrixU();
|
||||
const RealSchur<PlainType> schurOfA(arg);
|
||||
const PlainType& T = schurOfA.matrixT();
|
||||
const PlainType& U = schurOfA.matrixU();
|
||||
|
||||
// Compute square root of T
|
||||
MatrixType sqrtT = MatrixType::Zero(arg.rows(), arg.cols());
|
||||
PlainType sqrtT = PlainType::Zero(arg.rows(), arg.cols());
|
||||
matrix_sqrt_quasi_triangular(T, sqrtT);
|
||||
|
||||
// Compute square root of arg
|
||||
@@ -278,18 +279,19 @@ struct matrix_sqrt_compute<MatrixType, 0>
|
||||
template <typename MatrixType>
|
||||
struct matrix_sqrt_compute<MatrixType, 1>
|
||||
{
|
||||
typedef typename MatrixType::PlainObject PlainType;
|
||||
template <typename ResultType>
|
||||
static void run(const MatrixType &arg, ResultType &result)
|
||||
{
|
||||
eigen_assert(arg.rows() == arg.cols());
|
||||
|
||||
// Compute Schur decomposition of arg
|
||||
const ComplexSchur<MatrixType> schurOfA(arg);
|
||||
const MatrixType& T = schurOfA.matrixT();
|
||||
const MatrixType& U = schurOfA.matrixU();
|
||||
const ComplexSchur<PlainType> schurOfA(arg);
|
||||
const PlainType& T = schurOfA.matrixT();
|
||||
const PlainType& U = schurOfA.matrixU();
|
||||
|
||||
// Compute square root of T
|
||||
MatrixType sqrtT;
|
||||
PlainType sqrtT;
|
||||
matrix_sqrt_triangular(T, sqrtT);
|
||||
|
||||
// Compute square root of arg
|
||||
|
||||
Reference in New Issue
Block a user