// This file is part of Eigen, a lightweight C++ template library // for linear algebra. // // Copyright (C) 2007 Michael Olbrich // Copyright (C) 2006-2010 Benoit Jacob // Copyright (C) 2008 Gael Guennebaud // // Eigen is free software; you can redistribute it and/or // modify it under the terms of the GNU Lesser General Public // License as published by the Free Software Foundation; either // version 3 of the License, or (at your option) any later version. // // Alternatively, you can redistribute it and/or // modify it under the terms of the GNU General Public License as // published by the Free Software Foundation; either version 2 of // the License, or (at your option) any later version. // // Eigen is distributed in the hope that it will be useful, but WITHOUT ANY // WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS // FOR A PARTICULAR PURPOSE. See the GNU Lesser General Public License or the // GNU General Public License for more details. // // You should have received a copy of the GNU Lesser General Public // License and a copy of the GNU General Public License along with // Eigen. If not, see . #ifndef EIGEN_ASSIGN_H #define EIGEN_ASSIGN_H /*************************************************************************** * Part 1 : the logic deciding a strategy for traversal and unrolling * ***************************************************************************/ template struct ei_assign_traits { public: enum { DstIsAligned = Derived::Flags & AlignedBit, DstHasDirectAccess = Derived::Flags & DirectAccessBit, SrcIsAligned = OtherDerived::Flags & AlignedBit, JointAlignment = DstIsAligned && SrcIsAligned ? Aligned : Unaligned }; private: enum { InnerSize = int(Derived::IsVectorAtCompileTime) ? int(Derived::SizeAtCompileTime) : int(Derived::Flags)&RowMajorBit ? int(Derived::ColsAtCompileTime) : int(Derived::RowsAtCompileTime), InnerMaxSize = int(Derived::IsVectorAtCompileTime) ? int(Derived::MaxSizeAtCompileTime) : int(Derived::Flags)&RowMajorBit ? int(Derived::MaxColsAtCompileTime) : int(Derived::MaxRowsAtCompileTime), MaxSizeAtCompileTime = Derived::SizeAtCompileTime, PacketSize = ei_packet_traits::size }; enum { StorageOrdersAgree = (int(Derived::IsRowMajor) == int(OtherDerived::IsRowMajor)), MightVectorize = StorageOrdersAgree && (int(Derived::Flags) & int(OtherDerived::Flags) & ActualPacketAccessBit), MayInnerVectorize = MightVectorize && int(InnerSize)!=Dynamic && int(InnerSize)%int(PacketSize)==0 && int(DstIsAligned) && int(SrcIsAligned), MayLinearize = StorageOrdersAgree && (int(Derived::Flags) & int(OtherDerived::Flags) & LinearAccessBit), MayLinearVectorize = MightVectorize && MayLinearize && DstHasDirectAccess && (DstIsAligned || MaxSizeAtCompileTime == Dynamic), /* If the destination isn't aligned, we have to do runtime checks and we don't unroll, so it's only good for large enough sizes. */ MaySliceVectorize = MightVectorize && DstHasDirectAccess && (int(InnerMaxSize)==Dynamic || int(InnerMaxSize)>=3*PacketSize) /* slice vectorization can be slow, so we only want it if the slices are big, which is indicated by InnerMaxSize rather than InnerSize, think of the case of a dynamic block in a fixed-size matrix */ }; public: enum { Traversal = int(MayInnerVectorize) ? int(InnerVectorizedTraversal) : int(MayLinearVectorize) ? int(LinearVectorizedTraversal) : int(MaySliceVectorize) ? int(SliceVectorizedTraversal) : int(MayLinearize) ? int(LinearTraversal) : int(DefaultTraversal), Vectorized = int(Traversal) == InnerVectorizedTraversal || int(Traversal) == LinearVectorizedTraversal || int(Traversal) == SliceVectorizedTraversal }; private: enum { UnrollingLimit = EIGEN_UNROLLING_LIMIT * (Vectorized ? int(PacketSize) : 1), MayUnrollCompletely = int(Derived::SizeAtCompileTime) != Dynamic && int(OtherDerived::CoeffReadCost) != Dynamic && int(Derived::SizeAtCompileTime) * int(OtherDerived::CoeffReadCost) <= int(UnrollingLimit), MayUnrollInner = int(InnerSize) != Dynamic && int(OtherDerived::CoeffReadCost) != Dynamic && int(InnerSize) * int(OtherDerived::CoeffReadCost) <= int(UnrollingLimit) }; public: enum { Unrolling = (int(Traversal) == int(InnerVectorizedTraversal) || int(Traversal) == int(DefaultTraversal)) ? ( int(MayUnrollCompletely) ? int(CompleteUnrolling) : int(MayUnrollInner) ? int(InnerUnrolling) : int(NoUnrolling) ) : int(Traversal) == int(LinearVectorizedTraversal) ? ( int(MayUnrollCompletely) && int(DstIsAligned) ? int(CompleteUnrolling) : int(NoUnrolling) ) : int(Traversal) == int(LinearTraversal) ? ( int(MayUnrollCompletely) ? int(CompleteUnrolling) : int(NoUnrolling) ) : int(NoUnrolling) }; #ifdef EIGEN_DEBUG_ASSIGN static void debug() { EIGEN_DEBUG_VAR(DstIsAligned) EIGEN_DEBUG_VAR(SrcIsAligned) EIGEN_DEBUG_VAR(JointAlignment) EIGEN_DEBUG_VAR(InnerSize) EIGEN_DEBUG_VAR(InnerMaxSize) EIGEN_DEBUG_VAR(PacketSize) EIGEN_DEBUG_VAR(StorageOrdersAgree) EIGEN_DEBUG_VAR(MightVectorize) EIGEN_DEBUG_VAR(MayLinearize) EIGEN_DEBUG_VAR(MayInnerVectorize) EIGEN_DEBUG_VAR(MayLinearVectorize) EIGEN_DEBUG_VAR(MaySliceVectorize) EIGEN_DEBUG_VAR(Traversal) EIGEN_DEBUG_VAR(UnrollingLimit) EIGEN_DEBUG_VAR(MayUnrollCompletely) EIGEN_DEBUG_VAR(MayUnrollInner) EIGEN_DEBUG_VAR(Unrolling) } #endif }; /*************************************************************************** * Part 2 : meta-unrollers ***************************************************************************/ /************************ *** Default traversal *** ************************/ template struct ei_assign_DefaultTraversal_CompleteUnrolling { enum { outer = Index / Derived1::InnerSizeAtCompileTime, inner = Index % Derived1::InnerSizeAtCompileTime }; EIGEN_STRONG_INLINE static void run(Derived1 &dst, const Derived2 &src) { dst.copyCoeffByOuterInner(outer, inner, src); ei_assign_DefaultTraversal_CompleteUnrolling::run(dst, src); } }; template struct ei_assign_DefaultTraversal_CompleteUnrolling { EIGEN_STRONG_INLINE static void run(Derived1 &, const Derived2 &) {} }; template struct ei_assign_DefaultTraversal_InnerUnrolling { EIGEN_STRONG_INLINE static void run(Derived1 &dst, const Derived2 &src, int outer) { dst.copyCoeffByOuterInner(outer, Index, src); ei_assign_DefaultTraversal_InnerUnrolling::run(dst, src, outer); } }; template struct ei_assign_DefaultTraversal_InnerUnrolling { EIGEN_STRONG_INLINE static void run(Derived1 &, const Derived2 &, int) {} }; /*********************** *** Linear traversal *** ***********************/ template struct ei_assign_LinearTraversal_CompleteUnrolling { EIGEN_STRONG_INLINE static void run(Derived1 &dst, const Derived2 &src) { dst.copyCoeff(Index, src); ei_assign_LinearTraversal_CompleteUnrolling::run(dst, src); } }; template struct ei_assign_LinearTraversal_CompleteUnrolling { EIGEN_STRONG_INLINE static void run(Derived1 &, const Derived2 &) {} }; /************************** *** Inner vectorization *** **************************/ template struct ei_assign_innervec_CompleteUnrolling { enum { outer = Index / Derived1::InnerSizeAtCompileTime, inner = Index % Derived1::InnerSizeAtCompileTime, JointAlignment = ei_assign_traits::JointAlignment }; EIGEN_STRONG_INLINE static void run(Derived1 &dst, const Derived2 &src) { dst.template copyPacketByOuterInner(outer, inner, src); ei_assign_innervec_CompleteUnrolling::size, Stop>::run(dst, src); } }; template struct ei_assign_innervec_CompleteUnrolling { EIGEN_STRONG_INLINE static void run(Derived1 &, const Derived2 &) {} }; template struct ei_assign_innervec_InnerUnrolling { EIGEN_STRONG_INLINE static void run(Derived1 &dst, const Derived2 &src, int outer) { dst.template copyPacketByOuterInner(outer, Index, src); ei_assign_innervec_InnerUnrolling::size, Stop>::run(dst, src, outer); } }; template struct ei_assign_innervec_InnerUnrolling { EIGEN_STRONG_INLINE static void run(Derived1 &, const Derived2 &, int) {} }; /*************************************************************************** * Part 3 : implementation of all cases ***************************************************************************/ template::Traversal, int Unrolling = ei_assign_traits::Unrolling> struct ei_assign_impl; /************************ *** Default traversal *** ************************/ template struct ei_assign_impl { inline static void run(Derived1 &, const Derived2 &) { } }; template struct ei_assign_impl { typedef typename Derived1::Index Index; inline static void run(Derived1 &dst, const Derived2 &src) { const Index innerSize = dst.innerSize(); const Index outerSize = dst.outerSize(); for(Index outer = 0; outer < outerSize; ++outer) for(Index inner = 0; inner < innerSize; ++inner) dst.copyCoeffByOuterInner(outer, inner, src); } }; template struct ei_assign_impl { EIGEN_STRONG_INLINE static void run(Derived1 &dst, const Derived2 &src) { ei_assign_DefaultTraversal_CompleteUnrolling ::run(dst, src); } }; template struct ei_assign_impl { typedef typename Derived1::Index Index; EIGEN_STRONG_INLINE static void run(Derived1 &dst, const Derived2 &src) { const Index outerSize = dst.outerSize(); for(Index outer = 0; outer < outerSize; ++outer) ei_assign_DefaultTraversal_InnerUnrolling ::run(dst, src, outer); } }; /*********************** *** Linear traversal *** ***********************/ template struct ei_assign_impl { typedef typename Derived1::Index Index; inline static void run(Derived1 &dst, const Derived2 &src) { const Index size = dst.size(); for(Index i = 0; i < size; ++i) dst.copyCoeff(i, src); } }; template struct ei_assign_impl { EIGEN_STRONG_INLINE static void run(Derived1 &dst, const Derived2 &src) { ei_assign_LinearTraversal_CompleteUnrolling ::run(dst, src); } }; /************************** *** Inner vectorization *** **************************/ template struct ei_assign_impl { typedef typename Derived1::Index Index; inline static void run(Derived1 &dst, const Derived2 &src) { const Index innerSize = dst.innerSize(); const Index outerSize = dst.outerSize(); const Index packetSize = ei_packet_traits::size; for(Index outer = 0; outer < outerSize; ++outer) for(Index inner = 0; inner < innerSize; inner+=packetSize) dst.template copyPacketByOuterInner(outer, inner, src); } }; template struct ei_assign_impl { EIGEN_STRONG_INLINE static void run(Derived1 &dst, const Derived2 &src) { ei_assign_innervec_CompleteUnrolling ::run(dst, src); } }; template struct ei_assign_impl { typedef typename Derived1::Index Index; EIGEN_STRONG_INLINE static void run(Derived1 &dst, const Derived2 &src) { const Index outerSize = dst.outerSize(); for(Index outer = 0; outer < outerSize; ++outer) ei_assign_innervec_InnerUnrolling ::run(dst, src, outer); } }; /*************************** *** Linear vectorization *** ***************************/ template struct ei_unaligned_assign_impl { template static EIGEN_STRONG_INLINE void run(const Derived&, OtherDerived&, typename Derived::Index, typename Derived::Index) {} }; template <> struct ei_unaligned_assign_impl { // MSVC must not inline this functions. If it does, it fails to optimize the // packet access path. #ifdef _MSC_VER template static EIGEN_DONT_INLINE void run(const Derived& src, OtherDerived& dst, typename Derived::Index start, typename Derived::Index end) #else template static EIGEN_STRONG_INLINE void run(const Derived& src, OtherDerived& dst, typename Derived::Index start, typename Derived::Index end) #endif { for (typename Derived::Index index = start; index < end; ++index) dst.copyCoeff(index, src); } }; template struct ei_assign_impl { typedef typename Derived1::Index Index; EIGEN_STRONG_INLINE static void run(Derived1 &dst, const Derived2 &src) { const Index size = dst.size(); typedef ei_packet_traits PacketTraits; enum { packetSize = PacketTraits::size, dstAlignment = PacketTraits::AlignedOnScalar ? Aligned : int(ei_assign_traits::DstIsAligned) , srcAlignment = ei_assign_traits::JointAlignment }; const Index alignedStart = ei_assign_traits::DstIsAligned ? 0 : ei_first_aligned(&dst.coeffRef(0), size); const Index alignedEnd = alignedStart + ((size-alignedStart)/packetSize)*packetSize; ei_unaligned_assign_impl::DstIsAligned!=0>::run(src,dst,0,alignedStart); for(Index index = alignedStart; index < alignedEnd; index += packetSize) { dst.template copyPacket(index, src); } ei_unaligned_assign_impl<>::run(src,dst,alignedEnd,size); } }; template struct ei_assign_impl { typedef typename Derived1::Index Index; EIGEN_STRONG_INLINE static void run(Derived1 &dst, const Derived2 &src) { enum { size = Derived1::SizeAtCompileTime, packetSize = ei_packet_traits::size, alignedSize = (size/packetSize)*packetSize }; ei_assign_innervec_CompleteUnrolling::run(dst, src); ei_assign_DefaultTraversal_CompleteUnrolling::run(dst, src); } }; /************************** *** Slice vectorization *** ***************************/ template struct ei_assign_impl { typedef typename Derived1::Index Index; inline static void run(Derived1 &dst, const Derived2 &src) { typedef ei_packet_traits PacketTraits; enum { packetSize = PacketTraits::size, alignable = PacketTraits::AlignedOnScalar, dstAlignment = alignable ? Aligned : int(ei_assign_traits::DstIsAligned) , srcAlignment = ei_assign_traits::JointAlignment }; const Index packetAlignedMask = packetSize - 1; const Index innerSize = dst.innerSize(); const Index outerSize = dst.outerSize(); const Index alignedStep = alignable ? (packetSize - dst.outerStride() % packetSize) & packetAlignedMask : 0; Index alignedStart = ((!alignable) || ei_assign_traits::DstIsAligned) ? 0 : ei_first_aligned(&dst.coeffRef(0,0), innerSize); for(Index outer = 0; outer < outerSize; ++outer) { const Index alignedEnd = alignedStart + ((innerSize-alignedStart) & ~packetAlignedMask); // do the non-vectorizable part of the assignment for(Index inner = 0; inner(outer, inner, src); // do the non-vectorizable part of the assignment for(Index inner = alignedEnd; inner((alignedStart+alignedStep)%packetSize, innerSize); } } }; /*************************************************************************** * Part 4 : implementation of DenseBase methods ***************************************************************************/ template template EIGEN_STRONG_INLINE Derived& DenseBase ::lazyAssign(const DenseBase& other) { enum{ SameType = ei_is_same_type::ret }; EIGEN_STATIC_ASSERT_SAME_MATRIX_SIZE(Derived,OtherDerived) EIGEN_STATIC_ASSERT(SameType,YOU_MIXED_DIFFERENT_NUMERIC_TYPES__YOU_NEED_TO_USE_THE_CAST_METHOD_OF_MATRIXBASE_TO_CAST_NUMERIC_TYPES_EXPLICITLY) #ifdef EIGEN_DEBUG_ASSIGN ei_assign_traits::debug(); #endif ei_assert(rows() == other.rows() && cols() == other.cols()); ei_assign_impl::Traversal) : int(InvalidTraversal)>::run(derived(),other.derived()); #ifndef EIGEN_NO_DEBUG checkTransposeAliasing(other.derived()); #endif return derived(); } template struct ei_assign_selector; template struct ei_assign_selector { EIGEN_STRONG_INLINE static Derived& run(Derived& dst, const OtherDerived& other) { return dst.lazyAssign(other.derived()); } }; template struct ei_assign_selector { EIGEN_STRONG_INLINE static Derived& run(Derived& dst, const OtherDerived& other) { return dst.lazyAssign(other.eval()); } }; template struct ei_assign_selector { EIGEN_STRONG_INLINE static Derived& run(Derived& dst, const OtherDerived& other) { return dst.lazyAssign(other.transpose()); } }; template struct ei_assign_selector { EIGEN_STRONG_INLINE static Derived& run(Derived& dst, const OtherDerived& other) { return dst.lazyAssign(other.transpose().eval()); } }; template template EIGEN_STRONG_INLINE Derived& DenseBase::operator=(const DenseBase& other) { return ei_assign_selector::run(derived(), other.derived()); } template EIGEN_STRONG_INLINE Derived& DenseBase::operator=(const DenseBase& other) { return ei_assign_selector::run(derived(), other.derived()); } template EIGEN_STRONG_INLINE Derived& MatrixBase::operator=(const MatrixBase& other) { return ei_assign_selector::run(derived(), other.derived()); } template template EIGEN_STRONG_INLINE Derived& MatrixBase::operator=(const DenseBase& other) { return ei_assign_selector::run(derived(), other.derived()); } template template EIGEN_STRONG_INLINE Derived& MatrixBase::operator=(const EigenBase& other) { other.derived().evalTo(derived()); return derived(); } template template EIGEN_STRONG_INLINE Derived& MatrixBase::operator=(const ReturnByValue& other) { other.evalTo(derived()); return derived(); } #endif // EIGEN_ASSIGN_H