mirror/eigen
2
0
Fork 0
mirror of https://gitlab.com/libeigen/eigen.git synced 2025-01-24 14:45:14 +08:00
Code Issues Packages Projects Releases Wiki Activity

Adds missing EIGEN_STRONG_INLINE to support MSVC properly inlining small vector calculations

Browse Source 
When working with MSVC often small vector operations are not properly inlined. This behaviour is observed even on the most recent compiler versions.
This commit is contained in:
Basil Fierz 2017-10-26 22:44:28 +02:00
parent 746a6b7b81
commit 624df50945
4 changed files with 26 additions and 23 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

17
Eigen/src/Core/Dot.h
View File

@ -31,7 +31,8 @@ struct dot_nocheck
typedef scalar_conj_product_op<typename traits<T>::Scalar,typename traits<U>::Scalar> conj_prod;
typedef typename conj_prod::result_type ResScalar;
EIGEN_DEVICE_FUNC
static inline ResScalar run(const MatrixBase<T>& a, const MatrixBase<U>& b)
EIGEN_STRONG_INLINE
static ResScalar run(const MatrixBase<T>& a, const MatrixBase<U>& b)
{
return a.template binaryExpr<conj_prod>(b).sum();
}
@ -43,7 +44,8 @@ struct dot_nocheck<T, U, true>
typedef scalar_conj_product_op<typename traits<T>::Scalar,typename traits<U>::Scalar> conj_prod;
typedef typename conj_prod::result_type ResScalar;
EIGEN_DEVICE_FUNC
static inline ResScalar run(const MatrixBase<T>& a, const MatrixBase<U>& b)
EIGEN_STRONG_INLINE
static ResScalar run(const MatrixBase<T>& a, const MatrixBase<U>& b)
{
return a.transpose().template binaryExpr<conj_prod>(b).sum();
}
@ -65,6 +67,7 @@ struct dot_nocheck<T, U, true>
template<typename Derived>
template<typename OtherDerived>
EIGEN_DEVICE_FUNC
EIGEN_STRONG_INLINE
typename ScalarBinaryOpTraits<typename internal::traits<Derived>::Scalar,typename internal::traits<OtherDerived>::Scalar>::ReturnType
MatrixBase<Derived>::dot(const MatrixBase<OtherDerived>& other) const
{
@ -102,7 +105,7 @@ EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE typename NumTraits<typename internal::trai
* \sa lpNorm(), dot(), squaredNorm()
*/
template<typename Derived>
EIGEN_DEVICE_FUNC inline typename NumTraits<typename internal::traits<Derived>::Scalar>::Real MatrixBase<Derived>::norm() const
EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE typename NumTraits<typename internal::traits<Derived>::Scalar>::Real MatrixBase<Derived>::norm() const
{
return numext::sqrt(squaredNorm());
}
@ -117,7 +120,7 @@ EIGEN_DEVICE_FUNC inline typename NumTraits<typename internal::traits<Derived>::
* \sa norm(), normalize()
*/
template<typename Derived>
EIGEN_DEVICE_FUNC inline const typename MatrixBase<Derived>::PlainObject
EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE const typename MatrixBase<Derived>::PlainObject
MatrixBase<Derived>::normalized() const
{
typedef typename internal::nested_eval<Derived,2>::type _Nested;
@ -139,7 +142,7 @@ MatrixBase<Derived>::normalized() const
* \sa norm(), normalized()
*/
template<typename Derived>
EIGEN_DEVICE_FUNC inline void MatrixBase<Derived>::normalize()
EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE void MatrixBase<Derived>::normalize()
{
RealScalar z = squaredNorm();
// NOTE: after extensive benchmarking, this conditional does not impact performance, at least on recent x86 CPU
@ -160,7 +163,7 @@ EIGEN_DEVICE_FUNC inline void MatrixBase<Derived>::normalize()
* \sa stableNorm(), stableNormalize(), normalized()
*/
template<typename Derived>
EIGEN_DEVICE_FUNC inline const typename MatrixBase<Derived>::PlainObject
EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE const typename MatrixBase<Derived>::PlainObject
MatrixBase<Derived>::stableNormalized() const
{
typedef typename internal::nested_eval<Derived,3>::type _Nested;
@ -185,7 +188,7 @@ MatrixBase<Derived>::stableNormalized() const
* \sa stableNorm(), stableNormalized(), normalize()
*/
template<typename Derived>
EIGEN_DEVICE_FUNC inline void MatrixBase<Derived>::stableNormalize()
EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE void MatrixBase<Derived>::stableNormalize()
{
RealScalar w = cwiseAbs().maxCoeff();
RealScalar z = (derived()/w).squaredNorm();

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

@ -97,8 +97,8 @@ class Product : public ProductImpl<_Lhs,_Rhs,Option,
&& "if you wanted a coeff-wise or a dot product use the respective explicit functions");
}
EIGEN_DEVICE_FUNC inline Index rows() const { return m_lhs.rows(); }
EIGEN_DEVICE_FUNC inline Index cols() const { return m_rhs.cols(); }
EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE Index rows() const { return m_lhs.rows(); }
EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE Index cols() const { return m_rhs.cols(); }
EIGEN_DEVICE_FUNC const LhsNestedCleaned& lhs() const { return m_lhs; }
EIGEN_DEVICE_FUNC const RhsNestedCleaned& rhs() const { return m_rhs; }
@ -127,7 +127,7 @@ public:
using Base::derived;
typedef typename Base::Scalar Scalar;
operator const Scalar() const
EIGEN_STRONG_INLINE operator const Scalar() const
{
return internal::evaluator<ProductXpr>(derived()).coeff(0,0);
}
@ -162,7 +162,7 @@ class ProductImpl<Lhs,Rhs,Option,Dense>
public:
EIGEN_DEVICE_FUNC Scalar coeff(Index row, Index col) const
EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE Scalar coeff(Index row, Index col) const
{
EIGEN_STATIC_ASSERT(EnableCoeff, THIS_METHOD_IS_ONLY_FOR_INNER_OR_LAZY_PRODUCTS);
eigen_assert( (Option==LazyProduct) || (this->rows() == 1 && this->cols() == 1) );
@ -170,7 +170,7 @@ class ProductImpl<Lhs,Rhs,Option,Dense>
return internal::evaluator<Derived>(derived()).coeff(row,col);
}
EIGEN_DEVICE_FUNC Scalar coeff(Index i) const
EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE Scalar coeff(Index i) const
{
EIGEN_STATIC_ASSERT(EnableCoeff, THIS_METHOD_IS_ONLY_FOR_INNER_OR_LAZY_PRODUCTS);
eigen_assert( (Option==LazyProduct) || (this->rows() == 1 && this->cols() == 1) );

20
Eigen/src/Core/ProductEvaluators.h
View File

@ -32,7 +32,7 @@ struct evaluator<Product<Lhs, Rhs, Options> >
typedef Product<Lhs, Rhs, Options> XprType;
typedef product_evaluator<XprType> Base;
EIGEN_DEVICE_FUNC explicit evaluator(const XprType& xpr) : Base(xpr) {}
EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE explicit evaluator(const XprType& xpr) : Base(xpr) {}
};
// Catch "scalar * ( A * B )" and transform it to "(A*scalar) * B"
@ -55,7 +55,7 @@ struct evaluator<CwiseBinaryOp<internal::scalar_product_op<Scalar1,Scalar2>,
const Product<Lhs, Rhs, DefaultProduct> > XprType;
typedef evaluator<Product<EIGEN_SCALAR_BINARYOP_EXPR_RETURN_TYPE(Scalar1,Lhs,product), Rhs, DefaultProduct> > Base;
EIGEN_DEVICE_FUNC explicit evaluator(const XprType& xpr)
EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE explicit evaluator(const XprType& xpr)
: Base(xpr.lhs().functor().m_other * xpr.rhs().lhs() * xpr.rhs().rhs())
{}
};
@ -68,7 +68,7 @@ struct evaluator<Diagonal<const Product<Lhs, Rhs, DefaultProduct>, DiagIndex> >
typedef Diagonal<const Product<Lhs, Rhs, DefaultProduct>, DiagIndex> XprType;
typedef evaluator<Diagonal<const Product<Lhs, Rhs, LazyProduct>, DiagIndex> > Base;
EIGEN_DEVICE_FUNC explicit evaluator(const XprType& xpr)
EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE explicit evaluator(const XprType& xpr)
: Base(Diagonal<const Product<Lhs, Rhs, LazyProduct>, DiagIndex>(
Product<Lhs, Rhs, LazyProduct>(xpr.nestedExpression().lhs(), xpr.nestedExpression().rhs()),
xpr.index() ))
@ -246,19 +246,19 @@ template<typename Lhs, typename Rhs>
struct generic_product_impl<Lhs,Rhs,DenseShape,DenseShape,InnerProduct>
{
template<typename Dst>
static inline void evalTo(Dst& dst, const Lhs& lhs, const Rhs& rhs)
static EIGEN_STRONG_INLINE void evalTo(Dst& dst, const Lhs& lhs, const Rhs& rhs)
{
dst.coeffRef(0,0) = (lhs.transpose().cwiseProduct(rhs)).sum();
}
template<typename Dst>
static inline void addTo(Dst& dst, const Lhs& lhs, const Rhs& rhs)
static EIGEN_STRONG_INLINE void addTo(Dst& dst, const Lhs& lhs, const Rhs& rhs)
{
dst.coeffRef(0,0) += (lhs.transpose().cwiseProduct(rhs)).sum();
}
template<typename Dst>
static void subTo(Dst& dst, const Lhs& lhs, const Rhs& rhs)
static EIGEN_STRONG_INLINE void subTo(Dst& dst, const Lhs& lhs, const Rhs& rhs)
{ dst.coeffRef(0,0) -= (lhs.transpose().cwiseProduct(rhs)).sum(); }
};
@ -312,25 +312,25 @@ struct generic_product_impl<Lhs,Rhs,DenseShape,DenseShape,OuterProduct>
};
template<typename Dst>
static inline void evalTo(Dst& dst, const Lhs& lhs, const Rhs& rhs)
static EIGEN_STRONG_INLINE void evalTo(Dst& dst, const Lhs& lhs, const Rhs& rhs)
{
internal::outer_product_selector_run(dst, lhs, rhs, set(), is_row_major<Dst>());
}
template<typename Dst>
static inline void addTo(Dst& dst, const Lhs& lhs, const Rhs& rhs)
static EIGEN_STRONG_INLINE void addTo(Dst& dst, const Lhs& lhs, const Rhs& rhs)
{
internal::outer_product_selector_run(dst, lhs, rhs, add(), is_row_major<Dst>());
}
template<typename Dst>
static inline void subTo(Dst& dst, const Lhs& lhs, const Rhs& rhs)
static EIGEN_STRONG_INLINE void subTo(Dst& dst, const Lhs& lhs, const Rhs& rhs)
{
internal::outer_product_selector_run(dst, lhs, rhs, sub(), is_row_major<Dst>());
}
template<typename Dst>
static inline void scaleAndAddTo(Dst& dst, const Lhs& lhs, const Rhs& rhs, const Scalar& alpha)
static EIGEN_STRONG_INLINE void scaleAndAddTo(Dst& dst, const Lhs& lhs, const Rhs& rhs, const Scalar& alpha)
{
internal::outer_product_selector_run(dst, lhs, rhs, adds(alpha), is_row_major<Dst>());
}

2
Eigen/src/Core/Redux.h
View File

@ -407,7 +407,7 @@ protected:
*/
template<typename Derived>
template<typename Func>
EIGEN_DEVICE_FUNC typename internal::traits<Derived>::Scalar
EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE typename internal::traits<Derived>::Scalar
DenseBase<Derived>::redux(const Func& func) const
{
eigen_assert(this->rows()>0 && this->cols()>0 && "you are using an empty matrix");