aboutsummaryrefslogtreecommitdiffhomepage
diff options
context:
space:
mode:
Diffstat
-rw-r--r--Eigen/Core1
-rw-r--r--Eigen/src/Core/CoreEvaluators.h58
-rw-r--r--Eigen/src/Core/PartialReduxEvaluator.h224
-rw-r--r--Eigen/src/Core/VectorwiseOp.h78
-rw-r--r--test/vectorwiseop.cpp1
5 files changed, 270 insertions, 92 deletions
diff --git a/Eigen/Core b/Eigen/Core
index 6fd32dd82..a4596e73b 100644
--- a/Eigen/Core
+++ b/Eigen/Core
@@ -306,6 +306,7 @@ using std::ptrdiff_t;
#include "src/Core/BooleanRedux.h"
#include "src/Core/Select.h"
#include "src/Core/VectorwiseOp.h"
+#include "src/Core/PartialReduxEvaluator.h"
#include "src/Core/Random.h"
#include "src/Core/Replicate.h"
#include "src/Core/Reverse.h"
diff --git a/Eigen/src/Core/CoreEvaluators.h b/Eigen/src/Core/CoreEvaluators.h
index 264446f65..d5da5cdec 100644
--- a/Eigen/src/Core/CoreEvaluators.h
+++ b/Eigen/src/Core/CoreEvaluators.h
@@ -1325,64 +1325,6 @@ protected:
const variable_if_dynamic<Index, ArgType::ColsAtCompileTime> m_cols;
};
-
-// -------------------- PartialReduxExpr --------------------
-
-template< typename ArgType, typename MemberOp, int Direction>
-struct evaluator<PartialReduxExpr<ArgType, MemberOp, Direction> >
- : evaluator_base<PartialReduxExpr<ArgType, MemberOp, Direction> >
-{
- typedef PartialReduxExpr<ArgType, MemberOp, Direction> XprType;
- typedef typename internal::nested_eval<ArgType,1>::type ArgTypeNested;
- typedef typename internal::remove_all<ArgTypeNested>::type ArgTypeNestedCleaned;
- typedef typename ArgType::Scalar InputScalar;
- typedef typename XprType::Scalar Scalar;
- enum {
- TraversalSize = Direction==int(Vertical) ? int(ArgType::RowsAtCompileTime) : int(ArgType::ColsAtCompileTime)
- };
- typedef typename MemberOp::template Cost<InputScalar,int(TraversalSize)> CostOpType;
- enum {
- CoeffReadCost = TraversalSize==Dynamic ? HugeCost
- : TraversalSize * evaluator<ArgType>::CoeffReadCost + int(CostOpType::value),
-
- Flags = (traits<XprType>::Flags&RowMajorBit) | (evaluator<ArgType>::Flags&(HereditaryBits&(~RowMajorBit))) | LinearAccessBit,
-
- Alignment = 0 // FIXME this will need to be improved once PartialReduxExpr is vectorized
- };
-
- EIGEN_DEVICE_FUNC explicit evaluator(const XprType xpr)
- : m_arg(xpr.nestedExpression()), m_functor(xpr.functor())
- {
- EIGEN_INTERNAL_CHECK_COST_VALUE(TraversalSize==Dynamic ? HugeCost : int(CostOpType::value));
- EIGEN_INTERNAL_CHECK_COST_VALUE(CoeffReadCost);
- }
-
- typedef typename XprType::CoeffReturnType CoeffReturnType;
-
- EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
- const Scalar coeff(Index i, Index j) const
- {
- if (Direction==Vertical)
- return m_functor(m_arg.col(j));
- else
- return m_functor(m_arg.row(i));
- }
-
- EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
- const Scalar coeff(Index index) const
- {
- if (Direction==Vertical)
- return m_functor(m_arg.col(index));
- else
- return m_functor(m_arg.row(index));
- }
-
-protected:
- typename internal::add_const_on_value_type<ArgTypeNested>::type m_arg;
- const MemberOp m_functor;
-};
-
-
// -------------------- MatrixWrapper and ArrayWrapper --------------------
//
// evaluator_wrapper_base<T> is a common base class for the
diff --git a/Eigen/src/Core/PartialReduxEvaluator.h b/Eigen/src/Core/PartialReduxEvaluator.h
new file mode 100644
index 000000000..0bf8a50e0
--- /dev/null
+++ b/Eigen/src/Core/PartialReduxEvaluator.h
@@ -0,0 +1,224 @@
+// This file is part of Eigen, a lightweight C++ template library
+// for linear algebra.
+//
+// Copyright (C) 2011-2018 Gael Guennebaud <gael.guennebaud@inria.fr>
+//
+// This Source Code Form is subject to the terms of the Mozilla
+// Public License v. 2.0. If a copy of the MPL was not distributed
+// with this file, You can obtain one at http://mozilla.org/MPL/2.0/.
+
+#ifndef EIGEN_PARTIALREDUX_H
+#define EIGEN_PARTIALREDUX_H
+
+namespace Eigen {
+
+namespace internal {
+
+
+/***************************************************************************
+*
+* This file provides evaluators for partial reductions.
+* There are two modes:
+*
+* - scalar path: simply calls the respective function on the column or row.
+* -> nothing special here, all the tricky part is handled by the return
+* types of VectorwiseOp's members. They embed the functor calling the
+* respective DenseBase's member function.
+*
+* - vectorized path: implements a packet-wise reductions followed by
+* some (optional) processing of the outcome, e.g., division by n for mean.
+*
+* For the vectorized path let's observe that the packet-size and outer-unrolling
+* are both decided by the assignement logic. So all we have to do is to decide
+* on the inner unrolling.
+*
+* For the unrolling, we can reuse "internal::redux_vec_unroller" from Redux.h,
+* but be need to be careful to specify correct increment.
+*
+***************************************************************************/
+
+
+/* logic deciding a strategy for unrolling of vectorized paths */
+template<typename Func, typename Evaluator>
+struct packetwise_redux_traits
+{
+ enum {
+ OuterSize = int(Evaluator::IsRowMajor) ? Evaluator::RowsAtCompileTime : Evaluator::ColsAtCompileTime,
+ Cost = OuterSize == Dynamic ? HugeCost
+ : OuterSize * Evaluator::CoeffReadCost + (OuterSize-1) * functor_traits<Func>::Cost,
+ Unrolling = Cost <= EIGEN_UNROLLING_LIMIT ? CompleteUnrolling : NoUnrolling
+ };
+
+};
+
+/* Value to be returned when size==0 , by default let's return 0 */
+template<typename PacketType,typename Func>
+EIGEN_DEVICE_FUNC
+PacketType packetwise_redux_empty_value(const Func& ) { return pset1<PacketType>(0); }
+
+/* For products the default is 1 */
+template<typename PacketType,typename Scalar>
+EIGEN_DEVICE_FUNC
+PacketType packetwise_redux_empty_value(const scalar_product_op<Scalar,Scalar>& ) { return pset1<PacketType>(1); }
+
+/* Perform the actual reduction */
+template<typename Func, typename Evaluator,
+ int Unrolling = packetwise_redux_traits<Func, Evaluator>::Unrolling
+>
+struct packetwise_redux_impl;
+
+/* Perform the actual reduction with unrolling */
+template<typename Func, typename Evaluator>
+struct packetwise_redux_impl<Func, Evaluator, CompleteUnrolling>
+{
+ typedef redux_novec_unroller<Func,Evaluator, 0, Evaluator::SizeAtCompileTime> Base;
+ typedef typename Evaluator::Scalar Scalar;
+
+ template<typename PacketType>
+ EIGEN_DEVICE_FUNC static EIGEN_STRONG_INLINE
+ PacketType run(const Evaluator &eval, const Func& func, Index /*size*/)
+ {
+ return redux_vec_unroller<Func, Evaluator, 0, packetwise_redux_traits<Func, Evaluator>::OuterSize>::template run<PacketType>(eval,func);
+ }
+};
+
+/* Add a specialization of redux_vec_unroller for size==0 at compiletime.
+ * This specialization is not required for general reductions, which is
+ * why it is defined here.
+ */
+template<typename Func, typename Evaluator, int Start>
+struct redux_vec_unroller<Func, Evaluator, Start, 0>
+{
+ template<typename PacketType>
+ EIGEN_DEVICE_FUNC
+ static EIGEN_STRONG_INLINE PacketType run(const Evaluator &, const Func& f)
+ {
+ return packetwise_redux_empty_value<PacketType>(f);
+ }
+};
+
+/* Perform the actual reduction for dynamic sizes */
+template<typename Func, typename Evaluator>
+struct packetwise_redux_impl<Func, Evaluator, NoUnrolling>
+{
+ typedef typename Evaluator::Scalar Scalar;
+ typedef typename redux_traits<Func, Evaluator>::PacketType PacketScalar;
+
+ template<typename PacketType>
+ EIGEN_DEVICE_FUNC
+ static PacketType run(const Evaluator &eval, const Func& func, Index size)
+ {
+ if(size==0)
+ return packetwise_redux_empty_value<PacketType>(func);
+
+ const Index size4 = (size-1)&(~3);
+ PacketType p = eval.template packetByOuterInner<Unaligned,PacketType>(0,0);
+ Index i = 1;
+ // This loop is optimized for instruction pipelining:
+ // - each iteration generates two independent instructions
+ // - thanks to branch prediction and out-of-order execution we have independent instructions across loops
+ for(; i<size4; i+=4)
+ p = func.packetOp(p,
+ func.packetOp(
+ func.packetOp(eval.template packetByOuterInner<Unaligned,PacketType>(i+0,0),eval.template packetByOuterInner<Unaligned,PacketType>(i+1,0)),
+ func.packetOp(eval.template packetByOuterInner<Unaligned,PacketType>(i+2,0),eval.template packetByOuterInner<Unaligned,PacketType>(i+3,0))));
+ for(; i<size; ++i)
+ p = func.packetOp(p, eval.template packetByOuterInner<Unaligned,PacketType>(i,0));
+ return p;
+ }
+};
+
+template< typename ArgType, typename MemberOp, int Direction>
+struct evaluator<PartialReduxExpr<ArgType, MemberOp, Direction> >
+ : evaluator_base<PartialReduxExpr<ArgType, MemberOp, Direction> >
+{
+ typedef PartialReduxExpr<ArgType, MemberOp, Direction> XprType;
+ typedef typename internal::nested_eval<ArgType,1>::type ArgTypeNested;
+ typedef typename internal::remove_all<ArgTypeNested>::type ArgTypeNestedCleaned;
+ typedef typename ArgType::Scalar InputScalar;
+ typedef typename XprType::Scalar Scalar;
+ enum {
+ TraversalSize = Direction==int(Vertical) ? int(ArgType::RowsAtCompileTime) : int(ArgType::ColsAtCompileTime)
+ };
+ typedef typename MemberOp::template Cost<int(TraversalSize)> CostOpType;
+ enum {
+ CoeffReadCost = TraversalSize==Dynamic ? HugeCost
+ : TraversalSize==0 ? 1
+ : TraversalSize * evaluator<ArgType>::CoeffReadCost + int(CostOpType::value),
+
+ _ArgFlags = evaluator<ArgType>::Flags,
+
+ _Vectorizable = bool(int(_ArgFlags)&PacketAccessBit)
+ && bool(MemberOp::Vectorizable)
+ && (Direction==int(Vertical) ? bool(_ArgFlags&RowMajorBit) : (_ArgFlags&RowMajorBit)==0)
+ && (TraversalSize!=0),
+
+ Flags = (traits<XprType>::Flags&RowMajorBit)
+ | (evaluator<ArgType>::Flags&(HereditaryBits&(~RowMajorBit)))
+ | (_Vectorizable ? PacketAccessBit : 0)
+ | LinearAccessBit,
+
+ Alignment = 0 // FIXME this will need to be improved once PartialReduxExpr is vectorized
+ };
+
+ EIGEN_DEVICE_FUNC explicit evaluator(const XprType xpr)
+ : m_arg(xpr.nestedExpression()), m_functor(xpr.functor())
+ {
+ EIGEN_INTERNAL_CHECK_COST_VALUE(TraversalSize==Dynamic ? HugeCost : (TraversalSize==0 ? 1 : int(CostOpType::value)));
+ EIGEN_INTERNAL_CHECK_COST_VALUE(CoeffReadCost);
+ }
+
+ typedef typename XprType::CoeffReturnType CoeffReturnType;
+
+ EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
+ const Scalar coeff(Index i, Index j) const
+ {
+ return coeff(Direction==Vertical ? j : i);
+ }
+
+ EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
+ const Scalar coeff(Index index) const
+ {
+ return m_functor(m_arg.template subVector<DirectionType(Direction)>(index));
+ }
+
+ template<int LoadMode,typename PacketType>
+ EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
+ PacketType packet(Index i, Index j) const
+ {
+ return packet<LoadMode,PacketType>(Direction==Vertical ? j : i);
+ }
+
+ template<int LoadMode,typename PacketType>
+ EIGEN_STRONG_INLINE EIGEN_DEVICE_FUNC
+ PacketType packet(Index idx) const
+ {
+ enum { PacketSize = internal::unpacket_traits<PacketType>::size };
+ typedef Block<const ArgType,
+ Direction==Vertical ? int(ArgType::RowsAtCompileTime) : int(PacketSize),
+ Direction==Vertical ? int(PacketSize) : int(ArgType::ColsAtCompileTime),
+ true /* InnerPanel */> PanelType;
+
+ PanelType panel(m_arg,
+ Direction==Vertical ? 0 : idx,
+ Direction==Vertical ? idx : 0,
+ Direction==Vertical ? m_arg.rows() : Index(PacketSize),
+ Direction==Vertical ? Index(PacketSize) : m_arg.cols());
+
+ typedef typename internal::redux_evaluator<PanelType> PanelEvaluator;
+ PanelEvaluator panel_eval(panel);
+ typedef typename MemberOp::BinaryOp BinaryOp;
+ PacketType p = internal::packetwise_redux_impl<BinaryOp,PanelEvaluator>::template run<PacketType>(panel_eval,m_functor.binaryFunc(),m_arg.outerSize());
+ return p;
+ }
+
+protected:
+ typename internal::add_const_on_value_type<ArgTypeNested>::type m_arg;
+ const MemberOp m_functor;
+};
+
+} // end namespace internal
+
+} // end namespace Eigen
+
+#endif // EIGEN_PARTIALREDUX_H
diff --git a/Eigen/src/Core/VectorwiseOp.h b/Eigen/src/Core/VectorwiseOp.h
index 2a72c3cdd..a88b6e736 100644
--- a/Eigen/src/Core/VectorwiseOp.h
+++ b/Eigen/src/Core/VectorwiseOp.h
@@ -81,39 +81,46 @@ class PartialReduxExpr : public internal::dense_xpr_base< PartialReduxExpr<Matri
const MemberOp m_functor;
};
-#define EIGEN_MEMBER_FUNCTOR(MEMBER,COST) \
- template <typename ResultType> \
- struct member_##MEMBER { \
- EIGEN_EMPTY_STRUCT_CTOR(member_##MEMBER) \
- typedef ResultType result_type; \
- template<typename Scalar, int Size> struct Cost \
- { enum { value = COST }; }; \
- template<typename XprType> \
- EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE \
- ResultType operator()(const XprType& mat) const \
- { return mat.MEMBER(); } \
+template<typename A,typename B> struct partial_redux_dummy_func;
+
+#define EIGEN_MAKE_PARTIAL_REDUX_FUNCTOR(MEMBER,COST,VECTORIZABLE,BINARYOP) \
+ template <typename ResultType,typename Scalar> \
+ struct member_##MEMBER { \
+ EIGEN_EMPTY_STRUCT_CTOR(member_##MEMBER) \
+ typedef ResultType result_type; \
+ typedef BINARYOP<Scalar,Scalar> BinaryOp; \
+ template<int Size> struct Cost { enum { value = COST }; }; \
+ enum { Vectorizable = VECTORIZABLE }; \
+ template<typename XprType> \
+ EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE \
+ ResultType operator()(const XprType& mat) const \
+ { return mat.MEMBER(); } \
+ BinaryOp binaryFunc() const { return BinaryOp(); } \
}
+#define EIGEN_MEMBER_FUNCTOR(MEMBER,COST) \
+ EIGEN_MAKE_PARTIAL_REDUX_FUNCTOR(MEMBER,COST,0,partial_redux_dummy_func)
+
namespace internal {
-EIGEN_MEMBER_FUNCTOR(squaredNorm, Size * NumTraits<Scalar>::MulCost + (Size-1)*NumTraits<Scalar>::AddCost);
EIGEN_MEMBER_FUNCTOR(norm, (Size+5) * NumTraits<Scalar>::MulCost + (Size-1)*NumTraits<Scalar>::AddCost);
EIGEN_MEMBER_FUNCTOR(stableNorm, (Size+5) * NumTraits<Scalar>::MulCost + (Size-1)*NumTraits<Scalar>::AddCost);
EIGEN_MEMBER_FUNCTOR(blueNorm, (Size+5) * NumTraits<Scalar>::MulCost + (Size-1)*NumTraits<Scalar>::AddCost);
EIGEN_MEMBER_FUNCTOR(hypotNorm, (Size-1) * functor_traits<scalar_hypot_op<Scalar> >::Cost );
-EIGEN_MEMBER_FUNCTOR(sum, (Size-1)*NumTraits<Scalar>::AddCost);
-EIGEN_MEMBER_FUNCTOR(mean, (Size-1)*NumTraits<Scalar>::AddCost + NumTraits<Scalar>::MulCost);
-EIGEN_MEMBER_FUNCTOR(minCoeff, (Size-1)*NumTraits<Scalar>::AddCost);
-EIGEN_MEMBER_FUNCTOR(maxCoeff, (Size-1)*NumTraits<Scalar>::AddCost);
EIGEN_MEMBER_FUNCTOR(all, (Size-1)*NumTraits<Scalar>::AddCost);
EIGEN_MEMBER_FUNCTOR(any, (Size-1)*NumTraits<Scalar>::AddCost);
EIGEN_MEMBER_FUNCTOR(count, (Size-1)*NumTraits<Scalar>::AddCost);
-EIGEN_MEMBER_FUNCTOR(prod, (Size-1)*NumTraits<Scalar>::MulCost);
-template <int p, typename ResultType>
+EIGEN_MAKE_PARTIAL_REDUX_FUNCTOR(sum, (Size-1)*NumTraits<Scalar>::AddCost, 1, internal::scalar_sum_op);
+EIGEN_MAKE_PARTIAL_REDUX_FUNCTOR(minCoeff, (Size-1)*NumTraits<Scalar>::AddCost, 1, internal::scalar_min_op);
+EIGEN_MAKE_PARTIAL_REDUX_FUNCTOR(maxCoeff, (Size-1)*NumTraits<Scalar>::AddCost, 1, internal::scalar_max_op);
+EIGEN_MAKE_PARTIAL_REDUX_FUNCTOR(prod, (Size-1)*NumTraits<Scalar>::MulCost, 1, internal::scalar_product_op);
+
+template <int p, typename ResultType,typename Scalar>
struct member_lpnorm {
typedef ResultType result_type;
- template<typename Scalar, int Size> struct Cost
+ enum { Vectorizable = 0 };
+ template<int Size> struct Cost
{ enum { value = (Size+5) * NumTraits<Scalar>::MulCost + (Size-1)*NumTraits<Scalar>::AddCost }; };
EIGEN_DEVICE_FUNC member_lpnorm() {}
template<typename XprType>
@@ -121,17 +128,20 @@ struct member_lpnorm {
{ return mat.template lpNorm<p>(); }
};
-template <typename BinaryOp, typename Scalar>
+template <typename BinaryOpT, typename Scalar>
struct member_redux {
+ typedef BinaryOpT BinaryOp;
typedef typename result_of<
BinaryOp(const Scalar&,const Scalar&)
>::type result_type;
- template<typename _Scalar, int Size> struct Cost
- { enum { value = (Size-1) * functor_traits<BinaryOp>::Cost }; };
+
+ enum { Vectorizable = functor_traits<BinaryOp>::PacketAccess };
+ template<int Size> struct Cost { enum { value = (Size-1) * functor_traits<BinaryOp>::Cost }; };
EIGEN_DEVICE_FUNC explicit member_redux(const BinaryOp func) : m_functor(func) {}
template<typename Derived>
EIGEN_DEVICE_FUNC inline result_type operator()(const DenseBase<Derived>& mat) const
{ return mat.redux(m_functor); }
+ const BinaryOp& binaryFunc() const { return m_functor; }
const BinaryOp m_functor;
};
}
@@ -175,11 +185,11 @@ template<typename ExpressionType, int Direction> class VectorwiseOp
typedef typename internal::ref_selector<ExpressionType>::non_const_type ExpressionTypeNested;
typedef typename internal::remove_all<ExpressionTypeNested>::type ExpressionTypeNestedCleaned;
- template<template<typename _Scalar> class Functor,
- typename Scalar_=Scalar> struct ReturnType
+ template<template<typename OutScalar,typename InputScalar> class Functor,
+ typename ReturnScalar=Scalar> struct ReturnType
{
typedef PartialReduxExpr<ExpressionType,
- Functor<Scalar_>,
+ Functor<ReturnScalar,Scalar>,
Direction
> Type;
};
@@ -294,22 +304,22 @@ template<typename ExpressionType, int Direction> class VectorwiseOp
typedef typename ReturnType<internal::member_minCoeff>::Type MinCoeffReturnType;
typedef typename ReturnType<internal::member_maxCoeff>::Type MaxCoeffReturnType;
- typedef typename ReturnType<internal::member_squaredNorm,RealScalar>::Type SquaredNormReturnType;
- typedef typename ReturnType<internal::member_norm,RealScalar>::Type NormReturnType;
+ typedef PartialReduxExpr<const CwiseUnaryOp<internal::scalar_abs2_op<Scalar>, const ExpressionTypeNestedCleaned>,internal::member_sum<RealScalar,RealScalar>,Direction> SquaredNormReturnType;
+ typedef CwiseUnaryOp<internal::scalar_sqrt_op<RealScalar>, const SquaredNormReturnType> NormReturnType;
typedef typename ReturnType<internal::member_blueNorm,RealScalar>::Type BlueNormReturnType;
typedef typename ReturnType<internal::member_stableNorm,RealScalar>::Type StableNormReturnType;
typedef typename ReturnType<internal::member_hypotNorm,RealScalar>::Type HypotNormReturnType;
typedef typename ReturnType<internal::member_sum>::Type SumReturnType;
- typedef typename ReturnType<internal::member_mean>::Type MeanReturnType;
+ typedef EIGEN_EXPR_BINARYOP_SCALAR_RETURN_TYPE(SumReturnType,Scalar,quotient) MeanReturnType;
typedef typename ReturnType<internal::member_all>::Type AllReturnType;
typedef typename ReturnType<internal::member_any>::Type AnyReturnType;
- typedef PartialReduxExpr<ExpressionType, internal::member_count<Index>, Direction> CountReturnType;
+ typedef PartialReduxExpr<ExpressionType, internal::member_count<Index,Scalar>, Direction> CountReturnType;
typedef typename ReturnType<internal::member_prod>::Type ProdReturnType;
typedef Reverse<const ExpressionType, Direction> ConstReverseReturnType;
typedef Reverse<ExpressionType, Direction> ReverseReturnType;
template<int p> struct LpNormReturnType {
- typedef PartialReduxExpr<ExpressionType, internal::member_lpnorm<p,RealScalar>,Direction> Type;
+ typedef PartialReduxExpr<ExpressionType, internal::member_lpnorm<p,RealScalar,Scalar>,Direction> Type;
};
/** \returns a row (or column) vector expression of the smallest coefficient
@@ -348,7 +358,7 @@ template<typename ExpressionType, int Direction> class VectorwiseOp
* \sa DenseBase::squaredNorm() */
EIGEN_DEVICE_FUNC
const SquaredNormReturnType squaredNorm() const
- { return SquaredNormReturnType(_expression()); }
+ { return SquaredNormReturnType(m_matrix.cwiseAbs2()); }
/** \returns a row (or column) vector expression of the norm
* of each column (or row) of the referenced expression.
@@ -360,7 +370,7 @@ template<typename ExpressionType, int Direction> class VectorwiseOp
* \sa DenseBase::norm() */
EIGEN_DEVICE_FUNC
const NormReturnType norm() const
- { return NormReturnType(_expression()); }
+ { return NormReturnType(squaredNorm()); }
/** \returns a row (or column) vector expression of the norm
* of each column (or row) of the referenced expression.
@@ -425,7 +435,7 @@ template<typename ExpressionType, int Direction> class VectorwiseOp
* \sa DenseBase::mean() */
EIGEN_DEVICE_FUNC
const MeanReturnType mean() const
- { return MeanReturnType(_expression()); }
+ { return sum() / Scalar(Direction==Vertical?m_matrix.rows():m_matrix.cols()); }
/** \returns a row (or column) vector expression representing
* whether \b all coefficients of each respective column (or row) are \c true.
@@ -630,7 +640,7 @@ template<typename ExpressionType, int Direction> class VectorwiseOp
EIGEN_DEVICE_FUNC
CwiseBinaryOp<internal::scalar_quotient_op<Scalar>,
const ExpressionTypeNestedCleaned,
- const typename OppositeExtendedType<typename ReturnType<internal::member_norm,RealScalar>::Type>::Type>
+ const typename OppositeExtendedType<NormReturnType>::Type>
normalized() const { return m_matrix.cwiseQuotient(extendedToOpposite(this->norm())); }
diff --git a/test/vectorwiseop.cpp b/test/vectorwiseop.cpp
index 96a9bb0ee..a6745cb85 100644
--- a/test/vectorwiseop.cpp
+++ b/test/vectorwiseop.cpp
@@ -256,6 +256,7 @@ EIGEN_DECLARE_TEST(vectorwiseop)
CALL_SUBTEST_2( vectorwiseop_array(Array<double, 3, 2>()) );
CALL_SUBTEST_3( vectorwiseop_array(ArrayXXf(3, 4)) );
CALL_SUBTEST_4( vectorwiseop_matrix(Matrix4cf()) );
+ CALL_SUBTEST_5( vectorwiseop_matrix(Matrix4f()) );
CALL_SUBTEST_5( vectorwiseop_matrix(Matrix<float,4,5>()) );
CALL_SUBTEST_6( vectorwiseop_matrix(MatrixXd(internal::random<int>(1,EIGEN_TEST_MAX_SIZE), internal::random<int>(1,EIGEN_TEST_MAX_SIZE))) );
CALL_SUBTEST_7( vectorwiseop_matrix(VectorXd(internal::random<int>(1,EIGEN_TEST_MAX_SIZE))) );
generated by cgit v1.2.3 (git 2.39.1) at 2024-06-04 18:12:46 +0000