GitHub-Proxy/eigen
1
0
Fork 0
mirror of https://gitlab.com/libeigen/eigen.git synced 2025-08-11 19:29:02 +08:00
Code Issues Packages Projects Releases Wiki Activity

bug #1317: fix performance regression with some Block expressions and clang by helping it to remove dead code.

Browse Source 
The trick is to get rid of the nested expression in the evaluator by copying only the required information (here, the strides).
This commit is contained in:
Gael Guennebaud 2016-10-01 15:37:00 +02:00
parent 8b84801f7f
commit 9d6d0dff8f
1 changed files with 29 additions and 23 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

28
Eigen/src/Core/CoreEvaluators.h
View File

@ -820,7 +820,8 @@ struct mapbase_evaluator : evaluator_base<Derived>
EIGEN_DEVICE_FUNC explicit mapbase_evaluator(const XprType& map) EIGEN_DEVICE_FUNC explicit mapbase_evaluator(const XprType& map)
: m_data(const_cast<PointerType>(map.data())), : m_data(const_cast<PointerType>(map.data())),
m_xpr(map) m_innerStride(map.innerStride()),
m_outerStride(map.outerStride())
{ {
EIGEN_STATIC_ASSERT(EIGEN_IMPLIES(evaluator<Derived>::Flags&PacketAccessBit, internal::inner_stride_at_compile_time<Derived>::ret==1), EIGEN_STATIC_ASSERT(EIGEN_IMPLIES(evaluator<Derived>::Flags&PacketAccessBit, internal::inner_stride_at_compile_time<Derived>::ret==1),
PACKET_ACCESS_REQUIRES_TO_HAVE_INNER_STRIDE_FIXED_TO_1); PACKET_ACCESS_REQUIRES_TO_HAVE_INNER_STRIDE_FIXED_TO_1);
@ -830,32 +831,32 @@ struct mapbase_evaluator : evaluator_base<Derived>
EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
CoeffReturnType coeff(Index row, Index col) const CoeffReturnType coeff(Index row, Index col) const
{ {
return m_data[col * m_xpr.colStride() + row * m_xpr.rowStride()]; return m_data[col * colStride() + row * rowStride()];
} }
EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
CoeffReturnType coeff(Index index) const CoeffReturnType coeff(Index index) const
{ {
return m_data[index * m_xpr.innerStride()]; return m_data[index * m_innerStride.value()];
} }
EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
Scalar& coeffRef(Index row, Index col) Scalar& coeffRef(Index row, Index col)
{ {
return m_data[col * m_xpr.colStride() + row * m_xpr.rowStride()]; return m_data[col * colStride() + row * rowStride()];
} }
EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
Scalar& coeffRef(Index index) Scalar& coeffRef(Index index)
{ {
return m_data[index * m_xpr.innerStride()]; return m_data[index * m_innerStride.value()];
} }
template<int LoadMode, typename PacketType> template<int LoadMode, typename PacketType>
EIGEN_STRONG_INLINE EIGEN_STRONG_INLINE
PacketType packet(Index row, Index col) const PacketType packet(Index row, Index col) const
{ {
PointerType ptr = m_data + row * m_xpr.rowStride() + col * m_xpr.colStride(); PointerType ptr = m_data + row * rowStride() + col * colStride();
return internal::ploadt<PacketType, LoadMode>(ptr); return internal::ploadt<PacketType, LoadMode>(ptr);
} }
@ -863,14 +864,14 @@ struct mapbase_evaluator : evaluator_base<Derived>
EIGEN_STRONG_INLINE EIGEN_STRONG_INLINE
PacketType packet(Index index) const PacketType packet(Index index) const
{ {
return internal::ploadt<PacketType, LoadMode>(m_data + index * m_xpr.innerStride()); return internal::ploadt<PacketType, LoadMode>(m_data + index * m_innerStride.value());
} }
template<int StoreMode, typename PacketType> template<int StoreMode, typename PacketType>
EIGEN_STRONG_INLINE EIGEN_STRONG_INLINE
void writePacket(Index row, Index col, const PacketType& x) void writePacket(Index row, Index col, const PacketType& x)
{ {
PointerType ptr = m_data + row * m_xpr.rowStride() + col * m_xpr.colStride(); PointerType ptr = m_data + row * rowStride() + col * colStride();
return internal::pstoret<Scalar, PacketType, StoreMode>(ptr, x); return internal::pstoret<Scalar, PacketType, StoreMode>(ptr, x);
} }
@ -878,12 +879,17 @@ struct mapbase_evaluator : evaluator_base<Derived>
EIGEN_STRONG_INLINE EIGEN_STRONG_INLINE
void writePacket(Index index, const PacketType& x) void writePacket(Index index, const PacketType& x)
{ {
internal::pstoret<Scalar, PacketType, StoreMode>(m_data + index * m_xpr.innerStride(), x); internal::pstoret<Scalar, PacketType, StoreMode>(m_data + index * m_innerStride.value(), x);
} }
protected: protected:
EIGEN_DEVICE_FUNC
inline Index rowStride() const { return XprType::IsRowMajor ? m_outerStride.value() : m_innerStride.value(); }
EIGEN_DEVICE_FUNC
inline Index colStride() const { return XprType::IsRowMajor ? m_innerStride.value() : m_outerStride.value(); }
PointerType m_data; PointerType m_data;
const XprType& m_xpr; const internal::variable_if_dynamic<Index, XprType::InnerStrideAtCompileTime> m_innerStride;
const internal::variable_if_dynamic<Index, XprType::OuterStrideAtCompileTime> m_outerStride;
}; };
template<typename PlainObjectType, int MapOptions, typename StrideType> template<typename PlainObjectType, int MapOptions, typename StrideType>