ruby-eigen 0.0.9 → 0.0.10.pre1
Sign up to get free protection for your applications and to get access to all the features.
- checksums.yaml +4 -4
- data/LICENSE +22 -0
- data/README.md +21 -0
- data/ext/eigen/eigen3/COPYING.BSD +26 -0
- data/ext/eigen/eigen3/COPYING.MPL2 +373 -0
- data/ext/eigen/eigen3/COPYING.README +18 -0
- data/ext/eigen/eigen3/Eigen/Array +11 -0
- data/ext/eigen/eigen3/Eigen/Cholesky +32 -0
- data/ext/eigen/eigen3/Eigen/CholmodSupport +45 -0
- data/ext/eigen/eigen3/Eigen/Core +376 -0
- data/ext/eigen/eigen3/Eigen/Dense +7 -0
- data/ext/eigen/eigen3/Eigen/Eigen +2 -0
- data/ext/eigen/eigen3/Eigen/Eigen2Support +95 -0
- data/ext/eigen/eigen3/Eigen/Eigenvalues +48 -0
- data/ext/eigen/eigen3/Eigen/Geometry +63 -0
- data/ext/eigen/eigen3/Eigen/Householder +23 -0
- data/ext/eigen/eigen3/Eigen/IterativeLinearSolvers +40 -0
- data/ext/eigen/eigen3/Eigen/Jacobi +26 -0
- data/ext/eigen/eigen3/Eigen/LU +41 -0
- data/ext/eigen/eigen3/Eigen/LeastSquares +32 -0
- data/ext/eigen/eigen3/Eigen/MetisSupport +28 -0
- data/ext/eigen/eigen3/Eigen/PaStiXSupport +46 -0
- data/ext/eigen/eigen3/Eigen/PardisoSupport +30 -0
- data/ext/eigen/eigen3/Eigen/QR +45 -0
- data/ext/eigen/eigen3/Eigen/QtAlignedMalloc +34 -0
- data/ext/eigen/eigen3/Eigen/SPQRSupport +29 -0
- data/ext/eigen/eigen3/Eigen/SVD +37 -0
- data/ext/eigen/eigen3/Eigen/Sparse +27 -0
- data/ext/eigen/eigen3/Eigen/SparseCore +64 -0
- data/ext/eigen/eigen3/Eigen/SparseLU +49 -0
- data/ext/eigen/eigen3/Eigen/SparseQR +33 -0
- data/ext/eigen/eigen3/Eigen/StdDeque +27 -0
- data/ext/eigen/eigen3/Eigen/StdList +26 -0
- data/ext/eigen/eigen3/Eigen/StdVector +27 -0
- data/ext/eigen/eigen3/Eigen/SuperLUSupport +59 -0
- data/ext/eigen/eigen3/Eigen/UmfPackSupport +36 -0
- data/ext/eigen/eigen3/Eigen/src/Cholesky/LDLT.h +611 -0
- data/ext/eigen/eigen3/Eigen/src/Cholesky/LLT.h +498 -0
- data/ext/eigen/eigen3/Eigen/src/Cholesky/LLT_MKL.h +102 -0
- data/ext/eigen/eigen3/Eigen/src/CholmodSupport/CholmodSupport.h +607 -0
- data/ext/eigen/eigen3/Eigen/src/Core/Array.h +323 -0
- data/ext/eigen/eigen3/Eigen/src/Core/ArrayBase.h +226 -0
- data/ext/eigen/eigen3/Eigen/src/Core/ArrayWrapper.h +264 -0
- data/ext/eigen/eigen3/Eigen/src/Core/Assign.h +590 -0
- data/ext/eigen/eigen3/Eigen/src/Core/Assign_MKL.h +224 -0
- data/ext/eigen/eigen3/Eigen/src/Core/BandMatrix.h +334 -0
- data/ext/eigen/eigen3/Eigen/src/Core/Block.h +406 -0
- data/ext/eigen/eigen3/Eigen/src/Core/BooleanRedux.h +154 -0
- data/ext/eigen/eigen3/Eigen/src/Core/CommaInitializer.h +154 -0
- data/ext/eigen/eigen3/Eigen/src/Core/CoreIterators.h +61 -0
- data/ext/eigen/eigen3/Eigen/src/Core/CwiseBinaryOp.h +230 -0
- data/ext/eigen/eigen3/Eigen/src/Core/CwiseNullaryOp.h +864 -0
- data/ext/eigen/eigen3/Eigen/src/Core/CwiseUnaryOp.h +126 -0
- data/ext/eigen/eigen3/Eigen/src/Core/CwiseUnaryView.h +139 -0
- data/ext/eigen/eigen3/Eigen/src/Core/DenseBase.h +521 -0
- data/ext/eigen/eigen3/Eigen/src/Core/DenseCoeffsBase.h +754 -0
- data/ext/eigen/eigen3/Eigen/src/Core/DenseStorage.h +434 -0
- data/ext/eigen/eigen3/Eigen/src/Core/Diagonal.h +237 -0
- data/ext/eigen/eigen3/Eigen/src/Core/DiagonalMatrix.h +313 -0
- data/ext/eigen/eigen3/Eigen/src/Core/DiagonalProduct.h +131 -0
- data/ext/eigen/eigen3/Eigen/src/Core/Dot.h +263 -0
- data/ext/eigen/eigen3/Eigen/src/Core/EigenBase.h +131 -0
- data/ext/eigen/eigen3/Eigen/src/Core/Flagged.h +140 -0
- data/ext/eigen/eigen3/Eigen/src/Core/ForceAlignedAccess.h +146 -0
- data/ext/eigen/eigen3/Eigen/src/Core/Functors.h +1026 -0
- data/ext/eigen/eigen3/Eigen/src/Core/Fuzzy.h +150 -0
- data/ext/eigen/eigen3/Eigen/src/Core/GeneralProduct.h +635 -0
- data/ext/eigen/eigen3/Eigen/src/Core/GenericPacketMath.h +350 -0
- data/ext/eigen/eigen3/Eigen/src/Core/GlobalFunctions.h +92 -0
- data/ext/eigen/eigen3/Eigen/src/Core/IO.h +250 -0
- data/ext/eigen/eigen3/Eigen/src/Core/Map.h +192 -0
- data/ext/eigen/eigen3/Eigen/src/Core/MapBase.h +247 -0
- data/ext/eigen/eigen3/Eigen/src/Core/MathFunctions.h +768 -0
- data/ext/eigen/eigen3/Eigen/src/Core/Matrix.h +420 -0
- data/ext/eigen/eigen3/Eigen/src/Core/MatrixBase.h +563 -0
- data/ext/eigen/eigen3/Eigen/src/Core/NestByValue.h +111 -0
- data/ext/eigen/eigen3/Eigen/src/Core/NoAlias.h +134 -0
- data/ext/eigen/eigen3/Eigen/src/Core/NumTraits.h +150 -0
- data/ext/eigen/eigen3/Eigen/src/Core/PermutationMatrix.h +721 -0
- data/ext/eigen/eigen3/Eigen/src/Core/PlainObjectBase.h +822 -0
- data/ext/eigen/eigen3/Eigen/src/Core/ProductBase.h +290 -0
- data/ext/eigen/eigen3/Eigen/src/Core/Random.h +152 -0
- data/ext/eigen/eigen3/Eigen/src/Core/Redux.h +409 -0
- data/ext/eigen/eigen3/Eigen/src/Core/Ref.h +278 -0
- data/ext/eigen/eigen3/Eigen/src/Core/Replicate.h +177 -0
- data/ext/eigen/eigen3/Eigen/src/Core/ReturnByValue.h +99 -0
- data/ext/eigen/eigen3/Eigen/src/Core/Reverse.h +224 -0
- data/ext/eigen/eigen3/Eigen/src/Core/Select.h +162 -0
- data/ext/eigen/eigen3/Eigen/src/Core/SelfAdjointView.h +314 -0
- data/ext/eigen/eigen3/Eigen/src/Core/SelfCwiseBinaryOp.h +191 -0
- data/ext/eigen/eigen3/Eigen/src/Core/SolveTriangular.h +260 -0
- data/ext/eigen/eigen3/Eigen/src/Core/StableNorm.h +203 -0
- data/ext/eigen/eigen3/Eigen/src/Core/Stride.h +108 -0
- data/ext/eigen/eigen3/Eigen/src/Core/Swap.h +126 -0
- data/ext/eigen/eigen3/Eigen/src/Core/Transpose.h +419 -0
- data/ext/eigen/eigen3/Eigen/src/Core/Transpositions.h +436 -0
- data/ext/eigen/eigen3/Eigen/src/Core/TriangularMatrix.h +839 -0
- data/ext/eigen/eigen3/Eigen/src/Core/VectorBlock.h +95 -0
- data/ext/eigen/eigen3/Eigen/src/Core/VectorwiseOp.h +642 -0
- data/ext/eigen/eigen3/Eigen/src/Core/Visitor.h +237 -0
- data/ext/eigen/eigen3/Eigen/src/Core/arch/AltiVec/Complex.h +217 -0
- data/ext/eigen/eigen3/Eigen/src/Core/arch/AltiVec/PacketMath.h +501 -0
- data/ext/eigen/eigen3/Eigen/src/Core/arch/Default/Settings.h +49 -0
- data/ext/eigen/eigen3/Eigen/src/Core/arch/NEON/Complex.h +253 -0
- data/ext/eigen/eigen3/Eigen/src/Core/arch/NEON/PacketMath.h +420 -0
- data/ext/eigen/eigen3/Eigen/src/Core/arch/SSE/Complex.h +442 -0
- data/ext/eigen/eigen3/Eigen/src/Core/arch/SSE/MathFunctions.h +475 -0
- data/ext/eigen/eigen3/Eigen/src/Core/arch/SSE/PacketMath.h +649 -0
- data/ext/eigen/eigen3/Eigen/src/Core/products/CoeffBasedProduct.h +476 -0
- data/ext/eigen/eigen3/Eigen/src/Core/products/GeneralBlockPanelKernel.h +1341 -0
- data/ext/eigen/eigen3/Eigen/src/Core/products/GeneralMatrixMatrix.h +427 -0
- data/ext/eigen/eigen3/Eigen/src/Core/products/GeneralMatrixMatrixTriangular.h +278 -0
- data/ext/eigen/eigen3/Eigen/src/Core/products/GeneralMatrixMatrixTriangular_MKL.h +146 -0
- data/ext/eigen/eigen3/Eigen/src/Core/products/GeneralMatrixMatrix_MKL.h +118 -0
- data/ext/eigen/eigen3/Eigen/src/Core/products/GeneralMatrixVector.h +566 -0
- data/ext/eigen/eigen3/Eigen/src/Core/products/GeneralMatrixVector_MKL.h +131 -0
- data/ext/eigen/eigen3/Eigen/src/Core/products/Parallelizer.h +162 -0
- data/ext/eigen/eigen3/Eigen/src/Core/products/SelfadjointMatrixMatrix.h +436 -0
- data/ext/eigen/eigen3/Eigen/src/Core/products/SelfadjointMatrixMatrix_MKL.h +295 -0
- data/ext/eigen/eigen3/Eigen/src/Core/products/SelfadjointMatrixVector.h +281 -0
- data/ext/eigen/eigen3/Eigen/src/Core/products/SelfadjointMatrixVector_MKL.h +114 -0
- data/ext/eigen/eigen3/Eigen/src/Core/products/SelfadjointProduct.h +123 -0
- data/ext/eigen/eigen3/Eigen/src/Core/products/SelfadjointRank2Update.h +93 -0
- data/ext/eigen/eigen3/Eigen/src/Core/products/TriangularMatrixMatrix.h +427 -0
- data/ext/eigen/eigen3/Eigen/src/Core/products/TriangularMatrixMatrix_MKL.h +309 -0
- data/ext/eigen/eigen3/Eigen/src/Core/products/TriangularMatrixVector.h +348 -0
- data/ext/eigen/eigen3/Eigen/src/Core/products/TriangularMatrixVector_MKL.h +247 -0
- data/ext/eigen/eigen3/Eigen/src/Core/products/TriangularSolverMatrix.h +332 -0
- data/ext/eigen/eigen3/Eigen/src/Core/products/TriangularSolverMatrix_MKL.h +155 -0
- data/ext/eigen/eigen3/Eigen/src/Core/products/TriangularSolverVector.h +139 -0
- data/ext/eigen/eigen3/Eigen/src/Core/util/BlasUtil.h +264 -0
- data/ext/eigen/eigen3/Eigen/src/Core/util/Constants.h +451 -0
- data/ext/eigen/eigen3/Eigen/src/Core/util/DisableStupidWarnings.h +40 -0
- data/ext/eigen/eigen3/Eigen/src/Core/util/ForwardDeclarations.h +302 -0
- data/ext/eigen/eigen3/Eigen/src/Core/util/MKL_support.h +158 -0
- data/ext/eigen/eigen3/Eigen/src/Core/util/Macros.h +451 -0
- data/ext/eigen/eigen3/Eigen/src/Core/util/Memory.h +977 -0
- data/ext/eigen/eigen3/Eigen/src/Core/util/Meta.h +243 -0
- data/ext/eigen/eigen3/Eigen/src/Core/util/NonMPL2.h +3 -0
- data/ext/eigen/eigen3/Eigen/src/Core/util/ReenableStupidWarnings.h +14 -0
- data/ext/eigen/eigen3/Eigen/src/Core/util/StaticAssert.h +208 -0
- data/ext/eigen/eigen3/Eigen/src/Core/util/XprHelper.h +469 -0
- data/ext/eigen/eigen3/Eigen/src/Eigen2Support/Block.h +126 -0
- data/ext/eigen/eigen3/Eigen/src/Eigen2Support/Cwise.h +192 -0
- data/ext/eigen/eigen3/Eigen/src/Eigen2Support/CwiseOperators.h +298 -0
- data/ext/eigen/eigen3/Eigen/src/Eigen2Support/Geometry/AlignedBox.h +159 -0
- data/ext/eigen/eigen3/Eigen/src/Eigen2Support/Geometry/All.h +115 -0
- data/ext/eigen/eigen3/Eigen/src/Eigen2Support/Geometry/AngleAxis.h +214 -0
- data/ext/eigen/eigen3/Eigen/src/Eigen2Support/Geometry/Hyperplane.h +254 -0
- data/ext/eigen/eigen3/Eigen/src/Eigen2Support/Geometry/ParametrizedLine.h +141 -0
- data/ext/eigen/eigen3/Eigen/src/Eigen2Support/Geometry/Quaternion.h +495 -0
- data/ext/eigen/eigen3/Eigen/src/Eigen2Support/Geometry/Rotation2D.h +145 -0
- data/ext/eigen/eigen3/Eigen/src/Eigen2Support/Geometry/RotationBase.h +123 -0
- data/ext/eigen/eigen3/Eigen/src/Eigen2Support/Geometry/Scaling.h +167 -0
- data/ext/eigen/eigen3/Eigen/src/Eigen2Support/Geometry/Transform.h +786 -0
- data/ext/eigen/eigen3/Eigen/src/Eigen2Support/Geometry/Translation.h +184 -0
- data/ext/eigen/eigen3/Eigen/src/Eigen2Support/LU.h +120 -0
- data/ext/eigen/eigen3/Eigen/src/Eigen2Support/Lazy.h +71 -0
- data/ext/eigen/eigen3/Eigen/src/Eigen2Support/LeastSquares.h +169 -0
- data/ext/eigen/eigen3/Eigen/src/Eigen2Support/Macros.h +20 -0
- data/ext/eigen/eigen3/Eigen/src/Eigen2Support/MathFunctions.h +57 -0
- data/ext/eigen/eigen3/Eigen/src/Eigen2Support/Memory.h +45 -0
- data/ext/eigen/eigen3/Eigen/src/Eigen2Support/Meta.h +75 -0
- data/ext/eigen/eigen3/Eigen/src/Eigen2Support/Minor.h +117 -0
- data/ext/eigen/eigen3/Eigen/src/Eigen2Support/QR.h +67 -0
- data/ext/eigen/eigen3/Eigen/src/Eigen2Support/SVD.h +637 -0
- data/ext/eigen/eigen3/Eigen/src/Eigen2Support/TriangularSolver.h +42 -0
- data/ext/eigen/eigen3/Eigen/src/Eigen2Support/VectorBlock.h +94 -0
- data/ext/eigen/eigen3/Eigen/src/Eigenvalues/ComplexEigenSolver.h +341 -0
- data/ext/eigen/eigen3/Eigen/src/Eigenvalues/ComplexSchur.h +456 -0
- data/ext/eigen/eigen3/Eigen/src/Eigenvalues/ComplexSchur_MKL.h +94 -0
- data/ext/eigen/eigen3/Eigen/src/Eigenvalues/EigenSolver.h +607 -0
- data/ext/eigen/eigen3/Eigen/src/Eigenvalues/GeneralizedEigenSolver.h +350 -0
- data/ext/eigen/eigen3/Eigen/src/Eigenvalues/GeneralizedSelfAdjointEigenSolver.h +227 -0
- data/ext/eigen/eigen3/Eigen/src/Eigenvalues/HessenbergDecomposition.h +373 -0
- data/ext/eigen/eigen3/Eigen/src/Eigenvalues/MatrixBaseEigenvalues.h +160 -0
- data/ext/eigen/eigen3/Eigen/src/Eigenvalues/RealQZ.h +624 -0
- data/ext/eigen/eigen3/Eigen/src/Eigenvalues/RealSchur.h +525 -0
- data/ext/eigen/eigen3/Eigen/src/Eigenvalues/RealSchur_MKL.h +83 -0
- data/ext/eigen/eigen3/Eigen/src/Eigenvalues/SelfAdjointEigenSolver.h +801 -0
- data/ext/eigen/eigen3/Eigen/src/Eigenvalues/SelfAdjointEigenSolver_MKL.h +92 -0
- data/ext/eigen/eigen3/Eigen/src/Eigenvalues/Tridiagonalization.h +557 -0
- data/ext/eigen/eigen3/Eigen/src/Geometry/AlignedBox.h +392 -0
- data/ext/eigen/eigen3/Eigen/src/Geometry/AngleAxis.h +233 -0
- data/ext/eigen/eigen3/Eigen/src/Geometry/EulerAngles.h +104 -0
- data/ext/eigen/eigen3/Eigen/src/Geometry/Homogeneous.h +307 -0
- data/ext/eigen/eigen3/Eigen/src/Geometry/Hyperplane.h +280 -0
- data/ext/eigen/eigen3/Eigen/src/Geometry/OrthoMethods.h +218 -0
- data/ext/eigen/eigen3/Eigen/src/Geometry/ParametrizedLine.h +195 -0
- data/ext/eigen/eigen3/Eigen/src/Geometry/Quaternion.h +776 -0
- data/ext/eigen/eigen3/Eigen/src/Geometry/Rotation2D.h +160 -0
- data/ext/eigen/eigen3/Eigen/src/Geometry/RotationBase.h +206 -0
- data/ext/eigen/eigen3/Eigen/src/Geometry/Scaling.h +166 -0
- data/ext/eigen/eigen3/Eigen/src/Geometry/Transform.h +1455 -0
- data/ext/eigen/eigen3/Eigen/src/Geometry/Translation.h +206 -0
- data/ext/eigen/eigen3/Eigen/src/Geometry/Umeyama.h +177 -0
- data/ext/eigen/eigen3/Eigen/src/Geometry/arch/Geometry_SSE.h +115 -0
- data/ext/eigen/eigen3/Eigen/src/Householder/BlockHouseholder.h +68 -0
- data/ext/eigen/eigen3/Eigen/src/Householder/Householder.h +171 -0
- data/ext/eigen/eigen3/Eigen/src/Householder/HouseholderSequence.h +441 -0
- data/ext/eigen/eigen3/Eigen/src/IterativeLinearSolvers/BasicPreconditioners.h +149 -0
- data/ext/eigen/eigen3/Eigen/src/IterativeLinearSolvers/BiCGSTAB.h +263 -0
- data/ext/eigen/eigen3/Eigen/src/IterativeLinearSolvers/ConjugateGradient.h +256 -0
- data/ext/eigen/eigen3/Eigen/src/IterativeLinearSolvers/IterativeSolverBase.h +282 -0
- data/ext/eigen/eigen3/Eigen/src/Jacobi/Jacobi.h +433 -0
- data/ext/eigen/eigen3/Eigen/src/LU/Determinant.h +101 -0
- data/ext/eigen/eigen3/Eigen/src/LU/FullPivLU.h +751 -0
- data/ext/eigen/eigen3/Eigen/src/LU/Inverse.h +400 -0
- data/ext/eigen/eigen3/Eigen/src/LU/PartialPivLU.h +509 -0
- data/ext/eigen/eigen3/Eigen/src/LU/PartialPivLU_MKL.h +85 -0
- data/ext/eigen/eigen3/Eigen/src/LU/arch/Inverse_SSE.h +329 -0
- data/ext/eigen/eigen3/Eigen/src/MetisSupport/MetisSupport.h +137 -0
- data/ext/eigen/eigen3/Eigen/src/OrderingMethods/Amd.h +444 -0
- data/ext/eigen/eigen3/Eigen/src/OrderingMethods/Eigen_Colamd.h +1850 -0
- data/ext/eigen/eigen3/Eigen/src/PaStiXSupport/PaStiXSupport.h +721 -0
- data/ext/eigen/eigen3/Eigen/src/PardisoSupport/PardisoSupport.h +592 -0
- data/ext/eigen/eigen3/Eigen/src/QR/ColPivHouseholderQR.h +580 -0
- data/ext/eigen/eigen3/Eigen/src/QR/ColPivHouseholderQR_MKL.h +99 -0
- data/ext/eigen/eigen3/Eigen/src/QR/FullPivHouseholderQR.h +622 -0
- data/ext/eigen/eigen3/Eigen/src/QR/HouseholderQR.h +388 -0
- data/ext/eigen/eigen3/Eigen/src/QR/HouseholderQR_MKL.h +71 -0
- data/ext/eigen/eigen3/Eigen/src/SPQRSupport/SuiteSparseQRSupport.h +338 -0
- data/ext/eigen/eigen3/Eigen/src/SVD/JacobiSVD.h +976 -0
- data/ext/eigen/eigen3/Eigen/src/SVD/JacobiSVD_MKL.h +92 -0
- data/ext/eigen/eigen3/Eigen/src/SVD/UpperBidiagonalization.h +148 -0
- data/ext/eigen/eigen3/Eigen/src/SparseCholesky/SimplicialCholesky.h +671 -0
- data/ext/eigen/eigen3/Eigen/src/SparseCholesky/SimplicialCholesky_impl.h +199 -0
- data/ext/eigen/eigen3/Eigen/src/SparseCore/AmbiVector.h +373 -0
- data/ext/eigen/eigen3/Eigen/src/SparseCore/CompressedStorage.h +233 -0
- data/ext/eigen/eigen3/Eigen/src/SparseCore/ConservativeSparseSparseProduct.h +245 -0
- data/ext/eigen/eigen3/Eigen/src/SparseCore/MappedSparseMatrix.h +181 -0
- data/ext/eigen/eigen3/Eigen/src/SparseCore/SparseBlock.h +537 -0
- data/ext/eigen/eigen3/Eigen/src/SparseCore/SparseColEtree.h +206 -0
- data/ext/eigen/eigen3/Eigen/src/SparseCore/SparseCwiseBinaryOp.h +325 -0
- data/ext/eigen/eigen3/Eigen/src/SparseCore/SparseCwiseUnaryOp.h +163 -0
- data/ext/eigen/eigen3/Eigen/src/SparseCore/SparseDenseProduct.h +311 -0
- data/ext/eigen/eigen3/Eigen/src/SparseCore/SparseDiagonalProduct.h +196 -0
- data/ext/eigen/eigen3/Eigen/src/SparseCore/SparseDot.h +101 -0
- data/ext/eigen/eigen3/Eigen/src/SparseCore/SparseFuzzy.h +26 -0
- data/ext/eigen/eigen3/Eigen/src/SparseCore/SparseMatrix.h +1262 -0
- data/ext/eigen/eigen3/Eigen/src/SparseCore/SparseMatrixBase.h +461 -0
- data/ext/eigen/eigen3/Eigen/src/SparseCore/SparsePermutation.h +148 -0
- data/ext/eigen/eigen3/Eigen/src/SparseCore/SparseProduct.h +188 -0
- data/ext/eigen/eigen3/Eigen/src/SparseCore/SparseRedux.h +45 -0
- data/ext/eigen/eigen3/Eigen/src/SparseCore/SparseSelfAdjointView.h +507 -0
- data/ext/eigen/eigen3/Eigen/src/SparseCore/SparseSparseProductWithPruning.h +150 -0
- data/ext/eigen/eigen3/Eigen/src/SparseCore/SparseTranspose.h +63 -0
- data/ext/eigen/eigen3/Eigen/src/SparseCore/SparseTriangularView.h +179 -0
- data/ext/eigen/eigen3/Eigen/src/SparseCore/SparseUtil.h +172 -0
- data/ext/eigen/eigen3/Eigen/src/SparseCore/SparseVector.h +448 -0
- data/ext/eigen/eigen3/Eigen/src/SparseCore/SparseView.h +99 -0
- data/ext/eigen/eigen3/Eigen/src/SparseCore/TriangularSolver.h +334 -0
- data/ext/eigen/eigen3/Eigen/src/SparseLU/SparseLU.h +806 -0
- data/ext/eigen/eigen3/Eigen/src/SparseLU/SparseLUImpl.h +66 -0
- data/ext/eigen/eigen3/Eigen/src/SparseLU/SparseLU_Memory.h +227 -0
- data/ext/eigen/eigen3/Eigen/src/SparseLU/SparseLU_Structs.h +111 -0
- data/ext/eigen/eigen3/Eigen/src/SparseLU/SparseLU_SupernodalMatrix.h +298 -0
- data/ext/eigen/eigen3/Eigen/src/SparseLU/SparseLU_Utils.h +80 -0
- data/ext/eigen/eigen3/Eigen/src/SparseLU/SparseLU_column_bmod.h +180 -0
- data/ext/eigen/eigen3/Eigen/src/SparseLU/SparseLU_column_dfs.h +177 -0
- data/ext/eigen/eigen3/Eigen/src/SparseLU/SparseLU_copy_to_ucol.h +106 -0
- data/ext/eigen/eigen3/Eigen/src/SparseLU/SparseLU_gemm_kernel.h +279 -0
- data/ext/eigen/eigen3/Eigen/src/SparseLU/SparseLU_heap_relax_snode.h +127 -0
- data/ext/eigen/eigen3/Eigen/src/SparseLU/SparseLU_kernel_bmod.h +130 -0
- data/ext/eigen/eigen3/Eigen/src/SparseLU/SparseLU_panel_bmod.h +223 -0
- data/ext/eigen/eigen3/Eigen/src/SparseLU/SparseLU_panel_dfs.h +258 -0
- data/ext/eigen/eigen3/Eigen/src/SparseLU/SparseLU_pivotL.h +137 -0
- data/ext/eigen/eigen3/Eigen/src/SparseLU/SparseLU_pruneL.h +135 -0
- data/ext/eigen/eigen3/Eigen/src/SparseLU/SparseLU_relax_snode.h +83 -0
- data/ext/eigen/eigen3/Eigen/src/SparseQR/SparseQR.h +714 -0
- data/ext/eigen/eigen3/Eigen/src/StlSupport/StdDeque.h +134 -0
- data/ext/eigen/eigen3/Eigen/src/StlSupport/StdList.h +114 -0
- data/ext/eigen/eigen3/Eigen/src/StlSupport/StdVector.h +126 -0
- data/ext/eigen/eigen3/Eigen/src/StlSupport/details.h +84 -0
- data/ext/eigen/eigen3/Eigen/src/SuperLUSupport/SuperLUSupport.h +1026 -0
- data/ext/eigen/eigen3/Eigen/src/UmfPackSupport/UmfPackSupport.h +474 -0
- data/ext/eigen/eigen3/Eigen/src/misc/Image.h +84 -0
- data/ext/eigen/eigen3/Eigen/src/misc/Kernel.h +81 -0
- data/ext/eigen/eigen3/Eigen/src/misc/Solve.h +76 -0
- data/ext/eigen/eigen3/Eigen/src/misc/SparseSolve.h +128 -0
- data/ext/eigen/eigen3/Eigen/src/misc/blas.h +658 -0
- data/ext/eigen/eigen3/Eigen/src/plugins/ArrayCwiseBinaryOps.h +253 -0
- data/ext/eigen/eigen3/Eigen/src/plugins/ArrayCwiseUnaryOps.h +187 -0
- data/ext/eigen/eigen3/Eigen/src/plugins/BlockMethods.h +935 -0
- data/ext/eigen/eigen3/Eigen/src/plugins/CommonCwiseBinaryOps.h +46 -0
- data/ext/eigen/eigen3/Eigen/src/plugins/CommonCwiseUnaryOps.h +172 -0
- data/ext/eigen/eigen3/Eigen/src/plugins/MatrixCwiseBinaryOps.h +143 -0
- data/ext/eigen/eigen3/Eigen/src/plugins/MatrixCwiseUnaryOps.h +52 -0
- data/ext/eigen/eigen3/signature_of_eigen3_matrix_library +1 -0
- data/ext/eigen/eigen_wrap.cxx +19420 -10396
- data/ext/eigen/extconf.rb +37 -2
- data/lib/eigen.rb +146 -3
- metadata +294 -7
@@ -0,0 +1,427 @@
|
|
1
|
+
// This file is part of Eigen, a lightweight C++ template library
|
2
|
+
// for linear algebra.
|
3
|
+
//
|
4
|
+
// Copyright (C) 2008-2009 Gael Guennebaud <gael.guennebaud@inria.fr>
|
5
|
+
//
|
6
|
+
// This Source Code Form is subject to the terms of the Mozilla
|
7
|
+
// Public License v. 2.0. If a copy of the MPL was not distributed
|
8
|
+
// with this file, You can obtain one at http://mozilla.org/MPL/2.0/.
|
9
|
+
|
10
|
+
#ifndef EIGEN_GENERAL_MATRIX_MATRIX_H
|
11
|
+
#define EIGEN_GENERAL_MATRIX_MATRIX_H
|
12
|
+
|
13
|
+
namespace Eigen {
|
14
|
+
|
15
|
+
namespace internal {
|
16
|
+
|
17
|
+
template<typename _LhsScalar, typename _RhsScalar> class level3_blocking;
|
18
|
+
|
19
|
+
/* Specialization for a row-major destination matrix => simple transposition of the product */
|
20
|
+
template<
|
21
|
+
typename Index,
|
22
|
+
typename LhsScalar, int LhsStorageOrder, bool ConjugateLhs,
|
23
|
+
typename RhsScalar, int RhsStorageOrder, bool ConjugateRhs>
|
24
|
+
struct general_matrix_matrix_product<Index,LhsScalar,LhsStorageOrder,ConjugateLhs,RhsScalar,RhsStorageOrder,ConjugateRhs,RowMajor>
|
25
|
+
{
|
26
|
+
typedef typename scalar_product_traits<LhsScalar, RhsScalar>::ReturnType ResScalar;
|
27
|
+
static EIGEN_STRONG_INLINE void run(
|
28
|
+
Index rows, Index cols, Index depth,
|
29
|
+
const LhsScalar* lhs, Index lhsStride,
|
30
|
+
const RhsScalar* rhs, Index rhsStride,
|
31
|
+
ResScalar* res, Index resStride,
|
32
|
+
ResScalar alpha,
|
33
|
+
level3_blocking<RhsScalar,LhsScalar>& blocking,
|
34
|
+
GemmParallelInfo<Index>* info = 0)
|
35
|
+
{
|
36
|
+
// transpose the product such that the result is column major
|
37
|
+
general_matrix_matrix_product<Index,
|
38
|
+
RhsScalar, RhsStorageOrder==RowMajor ? ColMajor : RowMajor, ConjugateRhs,
|
39
|
+
LhsScalar, LhsStorageOrder==RowMajor ? ColMajor : RowMajor, ConjugateLhs,
|
40
|
+
ColMajor>
|
41
|
+
::run(cols,rows,depth,rhs,rhsStride,lhs,lhsStride,res,resStride,alpha,blocking,info);
|
42
|
+
}
|
43
|
+
};
|
44
|
+
|
45
|
+
/* Specialization for a col-major destination matrix
|
46
|
+
* => Blocking algorithm following Goto's paper */
|
47
|
+
template<
|
48
|
+
typename Index,
|
49
|
+
typename LhsScalar, int LhsStorageOrder, bool ConjugateLhs,
|
50
|
+
typename RhsScalar, int RhsStorageOrder, bool ConjugateRhs>
|
51
|
+
struct general_matrix_matrix_product<Index,LhsScalar,LhsStorageOrder,ConjugateLhs,RhsScalar,RhsStorageOrder,ConjugateRhs,ColMajor>
|
52
|
+
{
|
53
|
+
|
54
|
+
typedef typename scalar_product_traits<LhsScalar, RhsScalar>::ReturnType ResScalar;
|
55
|
+
static void run(Index rows, Index cols, Index depth,
|
56
|
+
const LhsScalar* _lhs, Index lhsStride,
|
57
|
+
const RhsScalar* _rhs, Index rhsStride,
|
58
|
+
ResScalar* res, Index resStride,
|
59
|
+
ResScalar alpha,
|
60
|
+
level3_blocking<LhsScalar,RhsScalar>& blocking,
|
61
|
+
GemmParallelInfo<Index>* info = 0)
|
62
|
+
{
|
63
|
+
const_blas_data_mapper<LhsScalar, Index, LhsStorageOrder> lhs(_lhs,lhsStride);
|
64
|
+
const_blas_data_mapper<RhsScalar, Index, RhsStorageOrder> rhs(_rhs,rhsStride);
|
65
|
+
|
66
|
+
typedef gebp_traits<LhsScalar,RhsScalar> Traits;
|
67
|
+
|
68
|
+
Index kc = blocking.kc(); // cache block size along the K direction
|
69
|
+
Index mc = (std::min)(rows,blocking.mc()); // cache block size along the M direction
|
70
|
+
//Index nc = blocking.nc(); // cache block size along the N direction
|
71
|
+
|
72
|
+
gemm_pack_lhs<LhsScalar, Index, Traits::mr, Traits::LhsProgress, LhsStorageOrder> pack_lhs;
|
73
|
+
gemm_pack_rhs<RhsScalar, Index, Traits::nr, RhsStorageOrder> pack_rhs;
|
74
|
+
gebp_kernel<LhsScalar, RhsScalar, Index, Traits::mr, Traits::nr, ConjugateLhs, ConjugateRhs> gebp;
|
75
|
+
|
76
|
+
#ifdef EIGEN_HAS_OPENMP
|
77
|
+
if(info)
|
78
|
+
{
|
79
|
+
// this is the parallel version!
|
80
|
+
Index tid = omp_get_thread_num();
|
81
|
+
Index threads = omp_get_num_threads();
|
82
|
+
|
83
|
+
std::size_t sizeA = kc*mc;
|
84
|
+
std::size_t sizeW = kc*Traits::WorkSpaceFactor;
|
85
|
+
ei_declare_aligned_stack_constructed_variable(LhsScalar, blockA, sizeA, 0);
|
86
|
+
ei_declare_aligned_stack_constructed_variable(RhsScalar, w, sizeW, 0);
|
87
|
+
|
88
|
+
RhsScalar* blockB = blocking.blockB();
|
89
|
+
eigen_internal_assert(blockB!=0);
|
90
|
+
|
91
|
+
// For each horizontal panel of the rhs, and corresponding vertical panel of the lhs...
|
92
|
+
for(Index k=0; k<depth; k+=kc)
|
93
|
+
{
|
94
|
+
const Index actual_kc = (std::min)(k+kc,depth)-k; // => rows of B', and cols of the A'
|
95
|
+
|
96
|
+
// In order to reduce the chance that a thread has to wait for the other,
|
97
|
+
// let's start by packing A'.
|
98
|
+
pack_lhs(blockA, &lhs(0,k), lhsStride, actual_kc, mc);
|
99
|
+
|
100
|
+
// Pack B_k to B' in a parallel fashion:
|
101
|
+
// each thread packs the sub block B_k,j to B'_j where j is the thread id.
|
102
|
+
|
103
|
+
// However, before copying to B'_j, we have to make sure that no other thread is still using it,
|
104
|
+
// i.e., we test that info[tid].users equals 0.
|
105
|
+
// Then, we set info[tid].users to the number of threads to mark that all other threads are going to use it.
|
106
|
+
while(info[tid].users!=0) {}
|
107
|
+
info[tid].users += threads;
|
108
|
+
|
109
|
+
pack_rhs(blockB+info[tid].rhs_start*actual_kc, &rhs(k,info[tid].rhs_start), rhsStride, actual_kc, info[tid].rhs_length);
|
110
|
+
|
111
|
+
// Notify the other threads that the part B'_j is ready to go.
|
112
|
+
info[tid].sync = k;
|
113
|
+
|
114
|
+
// Computes C_i += A' * B' per B'_j
|
115
|
+
for(Index shift=0; shift<threads; ++shift)
|
116
|
+
{
|
117
|
+
Index j = (tid+shift)%threads;
|
118
|
+
|
119
|
+
// At this point we have to make sure that B'_j has been updated by the thread j,
|
120
|
+
// we use testAndSetOrdered to mimic a volatile access.
|
121
|
+
// However, no need to wait for the B' part which has been updated by the current thread!
|
122
|
+
if(shift>0)
|
123
|
+
while(info[j].sync!=k) {}
|
124
|
+
|
125
|
+
gebp(res+info[j].rhs_start*resStride, resStride, blockA, blockB+info[j].rhs_start*actual_kc, mc, actual_kc, info[j].rhs_length, alpha, -1,-1,0,0, w);
|
126
|
+
}
|
127
|
+
|
128
|
+
// Then keep going as usual with the remaining A'
|
129
|
+
for(Index i=mc; i<rows; i+=mc)
|
130
|
+
{
|
131
|
+
const Index actual_mc = (std::min)(i+mc,rows)-i;
|
132
|
+
|
133
|
+
// pack A_i,k to A'
|
134
|
+
pack_lhs(blockA, &lhs(i,k), lhsStride, actual_kc, actual_mc);
|
135
|
+
|
136
|
+
// C_i += A' * B'
|
137
|
+
gebp(res+i, resStride, blockA, blockB, actual_mc, actual_kc, cols, alpha, -1,-1,0,0, w);
|
138
|
+
}
|
139
|
+
|
140
|
+
// Release all the sub blocks B'_j of B' for the current thread,
|
141
|
+
// i.e., we simply decrement the number of users by 1
|
142
|
+
for(Index j=0; j<threads; ++j)
|
143
|
+
#pragma omp atomic
|
144
|
+
--(info[j].users);
|
145
|
+
}
|
146
|
+
}
|
147
|
+
else
|
148
|
+
#endif // EIGEN_HAS_OPENMP
|
149
|
+
{
|
150
|
+
EIGEN_UNUSED_VARIABLE(info);
|
151
|
+
|
152
|
+
// this is the sequential version!
|
153
|
+
std::size_t sizeA = kc*mc;
|
154
|
+
std::size_t sizeB = kc*cols;
|
155
|
+
std::size_t sizeW = kc*Traits::WorkSpaceFactor;
|
156
|
+
|
157
|
+
ei_declare_aligned_stack_constructed_variable(LhsScalar, blockA, sizeA, blocking.blockA());
|
158
|
+
ei_declare_aligned_stack_constructed_variable(RhsScalar, blockB, sizeB, blocking.blockB());
|
159
|
+
ei_declare_aligned_stack_constructed_variable(RhsScalar, blockW, sizeW, blocking.blockW());
|
160
|
+
|
161
|
+
// For each horizontal panel of the rhs, and corresponding panel of the lhs...
|
162
|
+
// (==GEMM_VAR1)
|
163
|
+
for(Index k2=0; k2<depth; k2+=kc)
|
164
|
+
{
|
165
|
+
const Index actual_kc = (std::min)(k2+kc,depth)-k2;
|
166
|
+
|
167
|
+
// OK, here we have selected one horizontal panel of rhs and one vertical panel of lhs.
|
168
|
+
// => Pack rhs's panel into a sequential chunk of memory (L2 caching)
|
169
|
+
// Note that this panel will be read as many times as the number of blocks in the lhs's
|
170
|
+
// vertical panel which is, in practice, a very low number.
|
171
|
+
pack_rhs(blockB, &rhs(k2,0), rhsStride, actual_kc, cols);
|
172
|
+
|
173
|
+
// For each mc x kc block of the lhs's vertical panel...
|
174
|
+
// (==GEPP_VAR1)
|
175
|
+
for(Index i2=0; i2<rows; i2+=mc)
|
176
|
+
{
|
177
|
+
const Index actual_mc = (std::min)(i2+mc,rows)-i2;
|
178
|
+
|
179
|
+
// We pack the lhs's block into a sequential chunk of memory (L1 caching)
|
180
|
+
// Note that this block will be read a very high number of times, which is equal to the number of
|
181
|
+
// micro vertical panel of the large rhs's panel (e.g., cols/4 times).
|
182
|
+
pack_lhs(blockA, &lhs(i2,k2), lhsStride, actual_kc, actual_mc);
|
183
|
+
|
184
|
+
// Everything is packed, we can now call the block * panel kernel:
|
185
|
+
gebp(res+i2, resStride, blockA, blockB, actual_mc, actual_kc, cols, alpha, -1, -1, 0, 0, blockW);
|
186
|
+
}
|
187
|
+
}
|
188
|
+
}
|
189
|
+
}
|
190
|
+
|
191
|
+
};
|
192
|
+
|
193
|
+
/*********************************************************************************
|
194
|
+
* Specialization of GeneralProduct<> for "large" GEMM, i.e.,
|
195
|
+
* implementation of the high level wrapper to general_matrix_matrix_product
|
196
|
+
**********************************************************************************/
|
197
|
+
|
198
|
+
template<typename Lhs, typename Rhs>
|
199
|
+
struct traits<GeneralProduct<Lhs,Rhs,GemmProduct> >
|
200
|
+
: traits<ProductBase<GeneralProduct<Lhs,Rhs,GemmProduct>, Lhs, Rhs> >
|
201
|
+
{};
|
202
|
+
|
203
|
+
template<typename Scalar, typename Index, typename Gemm, typename Lhs, typename Rhs, typename Dest, typename BlockingType>
|
204
|
+
struct gemm_functor
|
205
|
+
{
|
206
|
+
gemm_functor(const Lhs& lhs, const Rhs& rhs, Dest& dest, const Scalar& actualAlpha,
|
207
|
+
BlockingType& blocking)
|
208
|
+
: m_lhs(lhs), m_rhs(rhs), m_dest(dest), m_actualAlpha(actualAlpha), m_blocking(blocking)
|
209
|
+
{}
|
210
|
+
|
211
|
+
void initParallelSession() const
|
212
|
+
{
|
213
|
+
m_blocking.allocateB();
|
214
|
+
}
|
215
|
+
|
216
|
+
void operator() (Index row, Index rows, Index col=0, Index cols=-1, GemmParallelInfo<Index>* info=0) const
|
217
|
+
{
|
218
|
+
if(cols==-1)
|
219
|
+
cols = m_rhs.cols();
|
220
|
+
|
221
|
+
Gemm::run(rows, cols, m_lhs.cols(),
|
222
|
+
/*(const Scalar*)*/&m_lhs.coeffRef(row,0), m_lhs.outerStride(),
|
223
|
+
/*(const Scalar*)*/&m_rhs.coeffRef(0,col), m_rhs.outerStride(),
|
224
|
+
(Scalar*)&(m_dest.coeffRef(row,col)), m_dest.outerStride(),
|
225
|
+
m_actualAlpha, m_blocking, info);
|
226
|
+
}
|
227
|
+
|
228
|
+
protected:
|
229
|
+
const Lhs& m_lhs;
|
230
|
+
const Rhs& m_rhs;
|
231
|
+
Dest& m_dest;
|
232
|
+
Scalar m_actualAlpha;
|
233
|
+
BlockingType& m_blocking;
|
234
|
+
};
|
235
|
+
|
236
|
+
template<int StorageOrder, typename LhsScalar, typename RhsScalar, int MaxRows, int MaxCols, int MaxDepth, int KcFactor=1,
|
237
|
+
bool FiniteAtCompileTime = MaxRows!=Dynamic && MaxCols!=Dynamic && MaxDepth != Dynamic> class gemm_blocking_space;
|
238
|
+
|
239
|
+
template<typename _LhsScalar, typename _RhsScalar>
|
240
|
+
class level3_blocking
|
241
|
+
{
|
242
|
+
typedef _LhsScalar LhsScalar;
|
243
|
+
typedef _RhsScalar RhsScalar;
|
244
|
+
|
245
|
+
protected:
|
246
|
+
LhsScalar* m_blockA;
|
247
|
+
RhsScalar* m_blockB;
|
248
|
+
RhsScalar* m_blockW;
|
249
|
+
|
250
|
+
DenseIndex m_mc;
|
251
|
+
DenseIndex m_nc;
|
252
|
+
DenseIndex m_kc;
|
253
|
+
|
254
|
+
public:
|
255
|
+
|
256
|
+
level3_blocking()
|
257
|
+
: m_blockA(0), m_blockB(0), m_blockW(0), m_mc(0), m_nc(0), m_kc(0)
|
258
|
+
{}
|
259
|
+
|
260
|
+
inline DenseIndex mc() const { return m_mc; }
|
261
|
+
inline DenseIndex nc() const { return m_nc; }
|
262
|
+
inline DenseIndex kc() const { return m_kc; }
|
263
|
+
|
264
|
+
inline LhsScalar* blockA() { return m_blockA; }
|
265
|
+
inline RhsScalar* blockB() { return m_blockB; }
|
266
|
+
inline RhsScalar* blockW() { return m_blockW; }
|
267
|
+
};
|
268
|
+
|
269
|
+
template<int StorageOrder, typename _LhsScalar, typename _RhsScalar, int MaxRows, int MaxCols, int MaxDepth, int KcFactor>
|
270
|
+
class gemm_blocking_space<StorageOrder,_LhsScalar,_RhsScalar,MaxRows, MaxCols, MaxDepth, KcFactor, true>
|
271
|
+
: public level3_blocking<
|
272
|
+
typename conditional<StorageOrder==RowMajor,_RhsScalar,_LhsScalar>::type,
|
273
|
+
typename conditional<StorageOrder==RowMajor,_LhsScalar,_RhsScalar>::type>
|
274
|
+
{
|
275
|
+
enum {
|
276
|
+
Transpose = StorageOrder==RowMajor,
|
277
|
+
ActualRows = Transpose ? MaxCols : MaxRows,
|
278
|
+
ActualCols = Transpose ? MaxRows : MaxCols
|
279
|
+
};
|
280
|
+
typedef typename conditional<Transpose,_RhsScalar,_LhsScalar>::type LhsScalar;
|
281
|
+
typedef typename conditional<Transpose,_LhsScalar,_RhsScalar>::type RhsScalar;
|
282
|
+
typedef gebp_traits<LhsScalar,RhsScalar> Traits;
|
283
|
+
enum {
|
284
|
+
SizeA = ActualRows * MaxDepth,
|
285
|
+
SizeB = ActualCols * MaxDepth,
|
286
|
+
SizeW = MaxDepth * Traits::WorkSpaceFactor
|
287
|
+
};
|
288
|
+
|
289
|
+
EIGEN_ALIGN16 LhsScalar m_staticA[SizeA];
|
290
|
+
EIGEN_ALIGN16 RhsScalar m_staticB[SizeB];
|
291
|
+
EIGEN_ALIGN16 RhsScalar m_staticW[SizeW];
|
292
|
+
|
293
|
+
public:
|
294
|
+
|
295
|
+
gemm_blocking_space(DenseIndex /*rows*/, DenseIndex /*cols*/, DenseIndex /*depth*/)
|
296
|
+
{
|
297
|
+
this->m_mc = ActualRows;
|
298
|
+
this->m_nc = ActualCols;
|
299
|
+
this->m_kc = MaxDepth;
|
300
|
+
this->m_blockA = m_staticA;
|
301
|
+
this->m_blockB = m_staticB;
|
302
|
+
this->m_blockW = m_staticW;
|
303
|
+
}
|
304
|
+
|
305
|
+
inline void allocateA() {}
|
306
|
+
inline void allocateB() {}
|
307
|
+
inline void allocateW() {}
|
308
|
+
inline void allocateAll() {}
|
309
|
+
};
|
310
|
+
|
311
|
+
template<int StorageOrder, typename _LhsScalar, typename _RhsScalar, int MaxRows, int MaxCols, int MaxDepth, int KcFactor>
|
312
|
+
class gemm_blocking_space<StorageOrder,_LhsScalar,_RhsScalar,MaxRows, MaxCols, MaxDepth, KcFactor, false>
|
313
|
+
: public level3_blocking<
|
314
|
+
typename conditional<StorageOrder==RowMajor,_RhsScalar,_LhsScalar>::type,
|
315
|
+
typename conditional<StorageOrder==RowMajor,_LhsScalar,_RhsScalar>::type>
|
316
|
+
{
|
317
|
+
enum {
|
318
|
+
Transpose = StorageOrder==RowMajor
|
319
|
+
};
|
320
|
+
typedef typename conditional<Transpose,_RhsScalar,_LhsScalar>::type LhsScalar;
|
321
|
+
typedef typename conditional<Transpose,_LhsScalar,_RhsScalar>::type RhsScalar;
|
322
|
+
typedef gebp_traits<LhsScalar,RhsScalar> Traits;
|
323
|
+
|
324
|
+
DenseIndex m_sizeA;
|
325
|
+
DenseIndex m_sizeB;
|
326
|
+
DenseIndex m_sizeW;
|
327
|
+
|
328
|
+
public:
|
329
|
+
|
330
|
+
gemm_blocking_space(DenseIndex rows, DenseIndex cols, DenseIndex depth)
|
331
|
+
{
|
332
|
+
this->m_mc = Transpose ? cols : rows;
|
333
|
+
this->m_nc = Transpose ? rows : cols;
|
334
|
+
this->m_kc = depth;
|
335
|
+
|
336
|
+
computeProductBlockingSizes<LhsScalar,RhsScalar,KcFactor>(this->m_kc, this->m_mc, this->m_nc);
|
337
|
+
m_sizeA = this->m_mc * this->m_kc;
|
338
|
+
m_sizeB = this->m_kc * this->m_nc;
|
339
|
+
m_sizeW = this->m_kc*Traits::WorkSpaceFactor;
|
340
|
+
}
|
341
|
+
|
342
|
+
void allocateA()
|
343
|
+
{
|
344
|
+
if(this->m_blockA==0)
|
345
|
+
this->m_blockA = aligned_new<LhsScalar>(m_sizeA);
|
346
|
+
}
|
347
|
+
|
348
|
+
void allocateB()
|
349
|
+
{
|
350
|
+
if(this->m_blockB==0)
|
351
|
+
this->m_blockB = aligned_new<RhsScalar>(m_sizeB);
|
352
|
+
}
|
353
|
+
|
354
|
+
void allocateW()
|
355
|
+
{
|
356
|
+
if(this->m_blockW==0)
|
357
|
+
this->m_blockW = aligned_new<RhsScalar>(m_sizeW);
|
358
|
+
}
|
359
|
+
|
360
|
+
void allocateAll()
|
361
|
+
{
|
362
|
+
allocateA();
|
363
|
+
allocateB();
|
364
|
+
allocateW();
|
365
|
+
}
|
366
|
+
|
367
|
+
~gemm_blocking_space()
|
368
|
+
{
|
369
|
+
aligned_delete(this->m_blockA, m_sizeA);
|
370
|
+
aligned_delete(this->m_blockB, m_sizeB);
|
371
|
+
aligned_delete(this->m_blockW, m_sizeW);
|
372
|
+
}
|
373
|
+
};
|
374
|
+
|
375
|
+
} // end namespace internal
|
376
|
+
|
377
|
+
template<typename Lhs, typename Rhs>
|
378
|
+
class GeneralProduct<Lhs, Rhs, GemmProduct>
|
379
|
+
: public ProductBase<GeneralProduct<Lhs,Rhs,GemmProduct>, Lhs, Rhs>
|
380
|
+
{
|
381
|
+
enum {
|
382
|
+
MaxDepthAtCompileTime = EIGEN_SIZE_MIN_PREFER_FIXED(Lhs::MaxColsAtCompileTime,Rhs::MaxRowsAtCompileTime)
|
383
|
+
};
|
384
|
+
public:
|
385
|
+
EIGEN_PRODUCT_PUBLIC_INTERFACE(GeneralProduct)
|
386
|
+
|
387
|
+
typedef typename Lhs::Scalar LhsScalar;
|
388
|
+
typedef typename Rhs::Scalar RhsScalar;
|
389
|
+
typedef Scalar ResScalar;
|
390
|
+
|
391
|
+
GeneralProduct(const Lhs& lhs, const Rhs& rhs) : Base(lhs,rhs)
|
392
|
+
{
|
393
|
+
typedef internal::scalar_product_op<LhsScalar,RhsScalar> BinOp;
|
394
|
+
EIGEN_CHECK_BINARY_COMPATIBILIY(BinOp,LhsScalar,RhsScalar);
|
395
|
+
}
|
396
|
+
|
397
|
+
template<typename Dest> void scaleAndAddTo(Dest& dst, const Scalar& alpha) const
|
398
|
+
{
|
399
|
+
eigen_assert(dst.rows()==m_lhs.rows() && dst.cols()==m_rhs.cols());
|
400
|
+
|
401
|
+
typename internal::add_const_on_value_type<ActualLhsType>::type lhs = LhsBlasTraits::extract(m_lhs);
|
402
|
+
typename internal::add_const_on_value_type<ActualRhsType>::type rhs = RhsBlasTraits::extract(m_rhs);
|
403
|
+
|
404
|
+
Scalar actualAlpha = alpha * LhsBlasTraits::extractScalarFactor(m_lhs)
|
405
|
+
* RhsBlasTraits::extractScalarFactor(m_rhs);
|
406
|
+
|
407
|
+
typedef internal::gemm_blocking_space<(Dest::Flags&RowMajorBit) ? RowMajor : ColMajor,LhsScalar,RhsScalar,
|
408
|
+
Dest::MaxRowsAtCompileTime,Dest::MaxColsAtCompileTime,MaxDepthAtCompileTime> BlockingType;
|
409
|
+
|
410
|
+
typedef internal::gemm_functor<
|
411
|
+
Scalar, Index,
|
412
|
+
internal::general_matrix_matrix_product<
|
413
|
+
Index,
|
414
|
+
LhsScalar, (_ActualLhsType::Flags&RowMajorBit) ? RowMajor : ColMajor, bool(LhsBlasTraits::NeedToConjugate),
|
415
|
+
RhsScalar, (_ActualRhsType::Flags&RowMajorBit) ? RowMajor : ColMajor, bool(RhsBlasTraits::NeedToConjugate),
|
416
|
+
(Dest::Flags&RowMajorBit) ? RowMajor : ColMajor>,
|
417
|
+
_ActualLhsType, _ActualRhsType, Dest, BlockingType> GemmFunctor;
|
418
|
+
|
419
|
+
BlockingType blocking(dst.rows(), dst.cols(), lhs.cols());
|
420
|
+
|
421
|
+
internal::parallelize_gemm<(Dest::MaxRowsAtCompileTime>32 || Dest::MaxRowsAtCompileTime==Dynamic)>(GemmFunctor(lhs, rhs, dst, actualAlpha, blocking), this->rows(), this->cols(), Dest::Flags&RowMajorBit);
|
422
|
+
}
|
423
|
+
};
|
424
|
+
|
425
|
+
} // end namespace Eigen
|
426
|
+
|
427
|
+
#endif // EIGEN_GENERAL_MATRIX_MATRIX_H
|
@@ -0,0 +1,278 @@
|
|
1
|
+
// This file is part of Eigen, a lightweight C++ template library
|
2
|
+
// for linear algebra.
|
3
|
+
//
|
4
|
+
// Copyright (C) 2009-2010 Gael Guennebaud <gael.guennebaud@inria.fr>
|
5
|
+
//
|
6
|
+
// This Source Code Form is subject to the terms of the Mozilla
|
7
|
+
// Public License v. 2.0. If a copy of the MPL was not distributed
|
8
|
+
// with this file, You can obtain one at http://mozilla.org/MPL/2.0/.
|
9
|
+
|
10
|
+
#ifndef EIGEN_GENERAL_MATRIX_MATRIX_TRIANGULAR_H
|
11
|
+
#define EIGEN_GENERAL_MATRIX_MATRIX_TRIANGULAR_H
|
12
|
+
|
13
|
+
namespace Eigen {
|
14
|
+
|
15
|
+
template<typename Scalar, typename Index, int StorageOrder, int UpLo, bool ConjLhs, bool ConjRhs>
|
16
|
+
struct selfadjoint_rank1_update;
|
17
|
+
|
18
|
+
namespace internal {
|
19
|
+
|
20
|
+
/**********************************************************************
|
21
|
+
* This file implements a general A * B product while
|
22
|
+
* evaluating only one triangular part of the product.
|
23
|
+
* This is more general version of self adjoint product (C += A A^T)
|
24
|
+
* as the level 3 SYRK Blas routine.
|
25
|
+
**********************************************************************/
|
26
|
+
|
27
|
+
// forward declarations (defined at the end of this file)
|
28
|
+
template<typename LhsScalar, typename RhsScalar, typename Index, int mr, int nr, bool ConjLhs, bool ConjRhs, int UpLo>
|
29
|
+
struct tribb_kernel;
|
30
|
+
|
31
|
+
/* Optimized matrix-matrix product evaluating only one triangular half */
|
32
|
+
template <typename Index,
|
33
|
+
typename LhsScalar, int LhsStorageOrder, bool ConjugateLhs,
|
34
|
+
typename RhsScalar, int RhsStorageOrder, bool ConjugateRhs,
|
35
|
+
int ResStorageOrder, int UpLo, int Version = Specialized>
|
36
|
+
struct general_matrix_matrix_triangular_product;
|
37
|
+
|
38
|
+
// as usual if the result is row major => we transpose the product
|
39
|
+
template <typename Index, typename LhsScalar, int LhsStorageOrder, bool ConjugateLhs,
|
40
|
+
typename RhsScalar, int RhsStorageOrder, bool ConjugateRhs, int UpLo, int Version>
|
41
|
+
struct general_matrix_matrix_triangular_product<Index,LhsScalar,LhsStorageOrder,ConjugateLhs,RhsScalar,RhsStorageOrder,ConjugateRhs,RowMajor,UpLo,Version>
|
42
|
+
{
|
43
|
+
typedef typename scalar_product_traits<LhsScalar, RhsScalar>::ReturnType ResScalar;
|
44
|
+
static EIGEN_STRONG_INLINE void run(Index size, Index depth,const LhsScalar* lhs, Index lhsStride,
|
45
|
+
const RhsScalar* rhs, Index rhsStride, ResScalar* res, Index resStride, const ResScalar& alpha)
|
46
|
+
{
|
47
|
+
general_matrix_matrix_triangular_product<Index,
|
48
|
+
RhsScalar, RhsStorageOrder==RowMajor ? ColMajor : RowMajor, ConjugateRhs,
|
49
|
+
LhsScalar, LhsStorageOrder==RowMajor ? ColMajor : RowMajor, ConjugateLhs,
|
50
|
+
ColMajor, UpLo==Lower?Upper:Lower>
|
51
|
+
::run(size,depth,rhs,rhsStride,lhs,lhsStride,res,resStride,alpha);
|
52
|
+
}
|
53
|
+
};
|
54
|
+
|
55
|
+
template <typename Index, typename LhsScalar, int LhsStorageOrder, bool ConjugateLhs,
|
56
|
+
typename RhsScalar, int RhsStorageOrder, bool ConjugateRhs, int UpLo, int Version>
|
57
|
+
struct general_matrix_matrix_triangular_product<Index,LhsScalar,LhsStorageOrder,ConjugateLhs,RhsScalar,RhsStorageOrder,ConjugateRhs,ColMajor,UpLo,Version>
|
58
|
+
{
|
59
|
+
typedef typename scalar_product_traits<LhsScalar, RhsScalar>::ReturnType ResScalar;
|
60
|
+
static EIGEN_STRONG_INLINE void run(Index size, Index depth,const LhsScalar* _lhs, Index lhsStride,
|
61
|
+
const RhsScalar* _rhs, Index rhsStride, ResScalar* res, Index resStride, const ResScalar& alpha)
|
62
|
+
{
|
63
|
+
const_blas_data_mapper<LhsScalar, Index, LhsStorageOrder> lhs(_lhs,lhsStride);
|
64
|
+
const_blas_data_mapper<RhsScalar, Index, RhsStorageOrder> rhs(_rhs,rhsStride);
|
65
|
+
|
66
|
+
typedef gebp_traits<LhsScalar,RhsScalar> Traits;
|
67
|
+
|
68
|
+
Index kc = depth; // cache block size along the K direction
|
69
|
+
Index mc = size; // cache block size along the M direction
|
70
|
+
Index nc = size; // cache block size along the N direction
|
71
|
+
computeProductBlockingSizes<LhsScalar,RhsScalar>(kc, mc, nc);
|
72
|
+
// !!! mc must be a multiple of nr:
|
73
|
+
if(mc > Traits::nr)
|
74
|
+
mc = (mc/Traits::nr)*Traits::nr;
|
75
|
+
|
76
|
+
std::size_t sizeW = kc*Traits::WorkSpaceFactor;
|
77
|
+
std::size_t sizeB = sizeW + kc*size;
|
78
|
+
ei_declare_aligned_stack_constructed_variable(LhsScalar, blockA, kc*mc, 0);
|
79
|
+
ei_declare_aligned_stack_constructed_variable(RhsScalar, allocatedBlockB, sizeB, 0);
|
80
|
+
RhsScalar* blockB = allocatedBlockB + sizeW;
|
81
|
+
|
82
|
+
gemm_pack_lhs<LhsScalar, Index, Traits::mr, Traits::LhsProgress, LhsStorageOrder> pack_lhs;
|
83
|
+
gemm_pack_rhs<RhsScalar, Index, Traits::nr, RhsStorageOrder> pack_rhs;
|
84
|
+
gebp_kernel <LhsScalar, RhsScalar, Index, Traits::mr, Traits::nr, ConjugateLhs, ConjugateRhs> gebp;
|
85
|
+
tribb_kernel<LhsScalar, RhsScalar, Index, Traits::mr, Traits::nr, ConjugateLhs, ConjugateRhs, UpLo> sybb;
|
86
|
+
|
87
|
+
for(Index k2=0; k2<depth; k2+=kc)
|
88
|
+
{
|
89
|
+
const Index actual_kc = (std::min)(k2+kc,depth)-k2;
|
90
|
+
|
91
|
+
// note that the actual rhs is the transpose/adjoint of mat
|
92
|
+
pack_rhs(blockB, &rhs(k2,0), rhsStride, actual_kc, size);
|
93
|
+
|
94
|
+
for(Index i2=0; i2<size; i2+=mc)
|
95
|
+
{
|
96
|
+
const Index actual_mc = (std::min)(i2+mc,size)-i2;
|
97
|
+
|
98
|
+
pack_lhs(blockA, &lhs(i2, k2), lhsStride, actual_kc, actual_mc);
|
99
|
+
|
100
|
+
// the selected actual_mc * size panel of res is split into three different part:
|
101
|
+
// 1 - before the diagonal => processed with gebp or skipped
|
102
|
+
// 2 - the actual_mc x actual_mc symmetric block => processed with a special kernel
|
103
|
+
// 3 - after the diagonal => processed with gebp or skipped
|
104
|
+
if (UpLo==Lower)
|
105
|
+
gebp(res+i2, resStride, blockA, blockB, actual_mc, actual_kc, (std::min)(size,i2), alpha,
|
106
|
+
-1, -1, 0, 0, allocatedBlockB);
|
107
|
+
|
108
|
+
sybb(res+resStride*i2 + i2, resStride, blockA, blockB + actual_kc*i2, actual_mc, actual_kc, alpha, allocatedBlockB);
|
109
|
+
|
110
|
+
if (UpLo==Upper)
|
111
|
+
{
|
112
|
+
Index j2 = i2+actual_mc;
|
113
|
+
gebp(res+resStride*j2+i2, resStride, blockA, blockB+actual_kc*j2, actual_mc, actual_kc, (std::max)(Index(0), size-j2), alpha,
|
114
|
+
-1, -1, 0, 0, allocatedBlockB);
|
115
|
+
}
|
116
|
+
}
|
117
|
+
}
|
118
|
+
}
|
119
|
+
};
|
120
|
+
|
121
|
+
// Optimized packed Block * packed Block product kernel evaluating only one given triangular part
|
122
|
+
// This kernel is built on top of the gebp kernel:
|
123
|
+
// - the current destination block is processed per panel of actual_mc x BlockSize
|
124
|
+
// where BlockSize is set to the minimal value allowing gebp to be as fast as possible
|
125
|
+
// - then, as usual, each panel is split into three parts along the diagonal,
|
126
|
+
// the sub blocks above and below the diagonal are processed as usual,
|
127
|
+
// while the triangular block overlapping the diagonal is evaluated into a
|
128
|
+
// small temporary buffer which is then accumulated into the result using a
|
129
|
+
// triangular traversal.
|
130
|
+
template<typename LhsScalar, typename RhsScalar, typename Index, int mr, int nr, bool ConjLhs, bool ConjRhs, int UpLo>
|
131
|
+
struct tribb_kernel
|
132
|
+
{
|
133
|
+
typedef gebp_traits<LhsScalar,RhsScalar,ConjLhs,ConjRhs> Traits;
|
134
|
+
typedef typename Traits::ResScalar ResScalar;
|
135
|
+
|
136
|
+
enum {
|
137
|
+
BlockSize = EIGEN_PLAIN_ENUM_MAX(mr,nr)
|
138
|
+
};
|
139
|
+
void operator()(ResScalar* res, Index resStride, const LhsScalar* blockA, const RhsScalar* blockB, Index size, Index depth, const ResScalar& alpha, RhsScalar* workspace)
|
140
|
+
{
|
141
|
+
gebp_kernel<LhsScalar, RhsScalar, Index, mr, nr, ConjLhs, ConjRhs> gebp_kernel;
|
142
|
+
Matrix<ResScalar,BlockSize,BlockSize,ColMajor> buffer;
|
143
|
+
|
144
|
+
// let's process the block per panel of actual_mc x BlockSize,
|
145
|
+
// again, each is split into three parts, etc.
|
146
|
+
for (Index j=0; j<size; j+=BlockSize)
|
147
|
+
{
|
148
|
+
Index actualBlockSize = std::min<Index>(BlockSize,size - j);
|
149
|
+
const RhsScalar* actual_b = blockB+j*depth;
|
150
|
+
|
151
|
+
if(UpLo==Upper)
|
152
|
+
gebp_kernel(res+j*resStride, resStride, blockA, actual_b, j, depth, actualBlockSize, alpha,
|
153
|
+
-1, -1, 0, 0, workspace);
|
154
|
+
|
155
|
+
// selfadjoint micro block
|
156
|
+
{
|
157
|
+
Index i = j;
|
158
|
+
buffer.setZero();
|
159
|
+
// 1 - apply the kernel on the temporary buffer
|
160
|
+
gebp_kernel(buffer.data(), BlockSize, blockA+depth*i, actual_b, actualBlockSize, depth, actualBlockSize, alpha,
|
161
|
+
-1, -1, 0, 0, workspace);
|
162
|
+
// 2 - triangular accumulation
|
163
|
+
for(Index j1=0; j1<actualBlockSize; ++j1)
|
164
|
+
{
|
165
|
+
ResScalar* r = res + (j+j1)*resStride + i;
|
166
|
+
for(Index i1=UpLo==Lower ? j1 : 0;
|
167
|
+
UpLo==Lower ? i1<actualBlockSize : i1<=j1; ++i1)
|
168
|
+
r[i1] += buffer(i1,j1);
|
169
|
+
}
|
170
|
+
}
|
171
|
+
|
172
|
+
if(UpLo==Lower)
|
173
|
+
{
|
174
|
+
Index i = j+actualBlockSize;
|
175
|
+
gebp_kernel(res+j*resStride+i, resStride, blockA+depth*i, actual_b, size-i, depth, actualBlockSize, alpha,
|
176
|
+
-1, -1, 0, 0, workspace);
|
177
|
+
}
|
178
|
+
}
|
179
|
+
}
|
180
|
+
};
|
181
|
+
|
182
|
+
} // end namespace internal
|
183
|
+
|
184
|
+
// high level API
|
185
|
+
|
186
|
+
template<typename MatrixType, typename ProductType, int UpLo, bool IsOuterProduct>
|
187
|
+
struct general_product_to_triangular_selector;
|
188
|
+
|
189
|
+
|
190
|
+
template<typename MatrixType, typename ProductType, int UpLo>
|
191
|
+
struct general_product_to_triangular_selector<MatrixType,ProductType,UpLo,true>
|
192
|
+
{
|
193
|
+
static void run(MatrixType& mat, const ProductType& prod, const typename MatrixType::Scalar& alpha)
|
194
|
+
{
|
195
|
+
typedef typename MatrixType::Scalar Scalar;
|
196
|
+
typedef typename MatrixType::Index Index;
|
197
|
+
|
198
|
+
typedef typename internal::remove_all<typename ProductType::LhsNested>::type Lhs;
|
199
|
+
typedef internal::blas_traits<Lhs> LhsBlasTraits;
|
200
|
+
typedef typename LhsBlasTraits::DirectLinearAccessType ActualLhs;
|
201
|
+
typedef typename internal::remove_all<ActualLhs>::type _ActualLhs;
|
202
|
+
typename internal::add_const_on_value_type<ActualLhs>::type actualLhs = LhsBlasTraits::extract(prod.lhs());
|
203
|
+
|
204
|
+
typedef typename internal::remove_all<typename ProductType::RhsNested>::type Rhs;
|
205
|
+
typedef internal::blas_traits<Rhs> RhsBlasTraits;
|
206
|
+
typedef typename RhsBlasTraits::DirectLinearAccessType ActualRhs;
|
207
|
+
typedef typename internal::remove_all<ActualRhs>::type _ActualRhs;
|
208
|
+
typename internal::add_const_on_value_type<ActualRhs>::type actualRhs = RhsBlasTraits::extract(prod.rhs());
|
209
|
+
|
210
|
+
Scalar actualAlpha = alpha * LhsBlasTraits::extractScalarFactor(prod.lhs().derived()) * RhsBlasTraits::extractScalarFactor(prod.rhs().derived());
|
211
|
+
|
212
|
+
enum {
|
213
|
+
StorageOrder = (internal::traits<MatrixType>::Flags&RowMajorBit) ? RowMajor : ColMajor,
|
214
|
+
UseLhsDirectly = _ActualLhs::InnerStrideAtCompileTime==1,
|
215
|
+
UseRhsDirectly = _ActualRhs::InnerStrideAtCompileTime==1
|
216
|
+
};
|
217
|
+
|
218
|
+
internal::gemv_static_vector_if<Scalar,Lhs::SizeAtCompileTime,Lhs::MaxSizeAtCompileTime,!UseLhsDirectly> static_lhs;
|
219
|
+
ei_declare_aligned_stack_constructed_variable(Scalar, actualLhsPtr, actualLhs.size(),
|
220
|
+
(UseLhsDirectly ? const_cast<Scalar*>(actualLhs.data()) : static_lhs.data()));
|
221
|
+
if(!UseLhsDirectly) Map<typename _ActualLhs::PlainObject>(actualLhsPtr, actualLhs.size()) = actualLhs;
|
222
|
+
|
223
|
+
internal::gemv_static_vector_if<Scalar,Rhs::SizeAtCompileTime,Rhs::MaxSizeAtCompileTime,!UseRhsDirectly> static_rhs;
|
224
|
+
ei_declare_aligned_stack_constructed_variable(Scalar, actualRhsPtr, actualRhs.size(),
|
225
|
+
(UseRhsDirectly ? const_cast<Scalar*>(actualRhs.data()) : static_rhs.data()));
|
226
|
+
if(!UseRhsDirectly) Map<typename _ActualRhs::PlainObject>(actualRhsPtr, actualRhs.size()) = actualRhs;
|
227
|
+
|
228
|
+
|
229
|
+
selfadjoint_rank1_update<Scalar,Index,StorageOrder,UpLo,
|
230
|
+
LhsBlasTraits::NeedToConjugate && NumTraits<Scalar>::IsComplex,
|
231
|
+
RhsBlasTraits::NeedToConjugate && NumTraits<Scalar>::IsComplex>
|
232
|
+
::run(actualLhs.size(), mat.data(), mat.outerStride(), actualLhsPtr, actualRhsPtr, actualAlpha);
|
233
|
+
}
|
234
|
+
};
|
235
|
+
|
236
|
+
template<typename MatrixType, typename ProductType, int UpLo>
|
237
|
+
struct general_product_to_triangular_selector<MatrixType,ProductType,UpLo,false>
|
238
|
+
{
|
239
|
+
static void run(MatrixType& mat, const ProductType& prod, const typename MatrixType::Scalar& alpha)
|
240
|
+
{
|
241
|
+
typedef typename MatrixType::Index Index;
|
242
|
+
|
243
|
+
typedef typename internal::remove_all<typename ProductType::LhsNested>::type Lhs;
|
244
|
+
typedef internal::blas_traits<Lhs> LhsBlasTraits;
|
245
|
+
typedef typename LhsBlasTraits::DirectLinearAccessType ActualLhs;
|
246
|
+
typedef typename internal::remove_all<ActualLhs>::type _ActualLhs;
|
247
|
+
typename internal::add_const_on_value_type<ActualLhs>::type actualLhs = LhsBlasTraits::extract(prod.lhs());
|
248
|
+
|
249
|
+
typedef typename internal::remove_all<typename ProductType::RhsNested>::type Rhs;
|
250
|
+
typedef internal::blas_traits<Rhs> RhsBlasTraits;
|
251
|
+
typedef typename RhsBlasTraits::DirectLinearAccessType ActualRhs;
|
252
|
+
typedef typename internal::remove_all<ActualRhs>::type _ActualRhs;
|
253
|
+
typename internal::add_const_on_value_type<ActualRhs>::type actualRhs = RhsBlasTraits::extract(prod.rhs());
|
254
|
+
|
255
|
+
typename ProductType::Scalar actualAlpha = alpha * LhsBlasTraits::extractScalarFactor(prod.lhs().derived()) * RhsBlasTraits::extractScalarFactor(prod.rhs().derived());
|
256
|
+
|
257
|
+
internal::general_matrix_matrix_triangular_product<Index,
|
258
|
+
typename Lhs::Scalar, _ActualLhs::Flags&RowMajorBit ? RowMajor : ColMajor, LhsBlasTraits::NeedToConjugate,
|
259
|
+
typename Rhs::Scalar, _ActualRhs::Flags&RowMajorBit ? RowMajor : ColMajor, RhsBlasTraits::NeedToConjugate,
|
260
|
+
MatrixType::Flags&RowMajorBit ? RowMajor : ColMajor, UpLo>
|
261
|
+
::run(mat.cols(), actualLhs.cols(),
|
262
|
+
&actualLhs.coeffRef(0,0), actualLhs.outerStride(), &actualRhs.coeffRef(0,0), actualRhs.outerStride(),
|
263
|
+
mat.data(), mat.outerStride(), actualAlpha);
|
264
|
+
}
|
265
|
+
};
|
266
|
+
|
267
|
+
template<typename MatrixType, unsigned int UpLo>
|
268
|
+
template<typename ProductDerived, typename _Lhs, typename _Rhs>
|
269
|
+
TriangularView<MatrixType,UpLo>& TriangularView<MatrixType,UpLo>::assignProduct(const ProductBase<ProductDerived, _Lhs,_Rhs>& prod, const Scalar& alpha)
|
270
|
+
{
|
271
|
+
general_product_to_triangular_selector<MatrixType, ProductDerived, UpLo, (_Lhs::ColsAtCompileTime==1) || (_Rhs::RowsAtCompileTime==1)>::run(m_matrix.const_cast_derived(), prod.derived(), alpha);
|
272
|
+
|
273
|
+
return *this;
|
274
|
+
}
|
275
|
+
|
276
|
+
} // end namespace Eigen
|
277
|
+
|
278
|
+
#endif // EIGEN_GENERAL_MATRIX_MATRIX_TRIANGULAR_H
|