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 +1,451 @@
|
|
1
|
+
// This file is part of Eigen, a lightweight C++ template library
|
2
|
+
// for linear algebra.
|
3
|
+
//
|
4
|
+
// Copyright (C) 2008-2010 Gael Guennebaud <gael.guennebaud@inria.fr>
|
5
|
+
// Copyright (C) 2006-2008 Benoit Jacob <jacob.benoit.1@gmail.com>
|
6
|
+
//
|
7
|
+
// This Source Code Form is subject to the terms of the Mozilla
|
8
|
+
// Public License v. 2.0. If a copy of the MPL was not distributed
|
9
|
+
// with this file, You can obtain one at http://mozilla.org/MPL/2.0/.
|
10
|
+
|
11
|
+
#ifndef EIGEN_MACROS_H
|
12
|
+
#define EIGEN_MACROS_H
|
13
|
+
|
14
|
+
#define EIGEN_WORLD_VERSION 3
|
15
|
+
#define EIGEN_MAJOR_VERSION 2
|
16
|
+
#define EIGEN_MINOR_VERSION 7
|
17
|
+
|
18
|
+
#define EIGEN_VERSION_AT_LEAST(x,y,z) (EIGEN_WORLD_VERSION>x || (EIGEN_WORLD_VERSION>=x && \
|
19
|
+
(EIGEN_MAJOR_VERSION>y || (EIGEN_MAJOR_VERSION>=y && \
|
20
|
+
EIGEN_MINOR_VERSION>=z))))
|
21
|
+
#ifdef __GNUC__
|
22
|
+
#define EIGEN_GNUC_AT_LEAST(x,y) ((__GNUC__==x && __GNUC_MINOR__>=y) || __GNUC__>x)
|
23
|
+
#else
|
24
|
+
#define EIGEN_GNUC_AT_LEAST(x,y) 0
|
25
|
+
#endif
|
26
|
+
|
27
|
+
#ifdef __GNUC__
|
28
|
+
#define EIGEN_GNUC_AT_MOST(x,y) ((__GNUC__==x && __GNUC_MINOR__<=y) || __GNUC__<x)
|
29
|
+
#else
|
30
|
+
#define EIGEN_GNUC_AT_MOST(x,y) 0
|
31
|
+
#endif
|
32
|
+
|
33
|
+
#if EIGEN_GNUC_AT_MOST(4,3) && !defined(__clang__)
|
34
|
+
// see bug 89
|
35
|
+
#define EIGEN_SAFE_TO_USE_STANDARD_ASSERT_MACRO 0
|
36
|
+
#else
|
37
|
+
#define EIGEN_SAFE_TO_USE_STANDARD_ASSERT_MACRO 1
|
38
|
+
#endif
|
39
|
+
|
40
|
+
#if defined(__GNUC__) && (__GNUC__ <= 3)
|
41
|
+
#define EIGEN_GCC3_OR_OLDER 1
|
42
|
+
#else
|
43
|
+
#define EIGEN_GCC3_OR_OLDER 0
|
44
|
+
#endif
|
45
|
+
|
46
|
+
// 16 byte alignment is only useful for vectorization. Since it affects the ABI, we need to enable
|
47
|
+
// 16 byte alignment on all platforms where vectorization might be enabled. In theory we could always
|
48
|
+
// enable alignment, but it can be a cause of problems on some platforms, so we just disable it in
|
49
|
+
// certain common platform (compiler+architecture combinations) to avoid these problems.
|
50
|
+
// Only static alignment is really problematic (relies on nonstandard compiler extensions that don't
|
51
|
+
// work everywhere, for example don't work on GCC/ARM), try to keep heap alignment even
|
52
|
+
// when we have to disable static alignment.
|
53
|
+
#if defined(__GNUC__) && !(defined(__i386__) || defined(__x86_64__) || defined(__powerpc__) || defined(__ppc__) || defined(__ia64__))
|
54
|
+
#define EIGEN_GCC_AND_ARCH_DOESNT_WANT_STACK_ALIGNMENT 1
|
55
|
+
#else
|
56
|
+
#define EIGEN_GCC_AND_ARCH_DOESNT_WANT_STACK_ALIGNMENT 0
|
57
|
+
#endif
|
58
|
+
|
59
|
+
// static alignment is completely disabled with GCC 3, Sun Studio, and QCC/QNX
|
60
|
+
#if !EIGEN_GCC_AND_ARCH_DOESNT_WANT_STACK_ALIGNMENT \
|
61
|
+
&& !EIGEN_GCC3_OR_OLDER \
|
62
|
+
&& !defined(__SUNPRO_CC) \
|
63
|
+
&& !defined(__QNXNTO__)
|
64
|
+
#define EIGEN_ARCH_WANTS_STACK_ALIGNMENT 1
|
65
|
+
#else
|
66
|
+
#define EIGEN_ARCH_WANTS_STACK_ALIGNMENT 0
|
67
|
+
#endif
|
68
|
+
|
69
|
+
#ifdef EIGEN_DONT_ALIGN
|
70
|
+
#ifndef EIGEN_DONT_ALIGN_STATICALLY
|
71
|
+
#define EIGEN_DONT_ALIGN_STATICALLY
|
72
|
+
#endif
|
73
|
+
#define EIGEN_ALIGN 0
|
74
|
+
#else
|
75
|
+
#define EIGEN_ALIGN 1
|
76
|
+
#endif
|
77
|
+
|
78
|
+
// EIGEN_ALIGN_STATICALLY is the true test whether we want to align arrays on the stack or not. It takes into account both the user choice to explicitly disable
|
79
|
+
// alignment (EIGEN_DONT_ALIGN_STATICALLY) and the architecture config (EIGEN_ARCH_WANTS_STACK_ALIGNMENT). Henceforth, only EIGEN_ALIGN_STATICALLY should be used.
|
80
|
+
#if EIGEN_ARCH_WANTS_STACK_ALIGNMENT && !defined(EIGEN_DONT_ALIGN_STATICALLY)
|
81
|
+
#define EIGEN_ALIGN_STATICALLY 1
|
82
|
+
#else
|
83
|
+
#define EIGEN_ALIGN_STATICALLY 0
|
84
|
+
#ifndef EIGEN_DISABLE_UNALIGNED_ARRAY_ASSERT
|
85
|
+
#define EIGEN_DISABLE_UNALIGNED_ARRAY_ASSERT
|
86
|
+
#endif
|
87
|
+
#endif
|
88
|
+
|
89
|
+
#ifdef EIGEN_DEFAULT_TO_ROW_MAJOR
|
90
|
+
#define EIGEN_DEFAULT_MATRIX_STORAGE_ORDER_OPTION RowMajor
|
91
|
+
#else
|
92
|
+
#define EIGEN_DEFAULT_MATRIX_STORAGE_ORDER_OPTION ColMajor
|
93
|
+
#endif
|
94
|
+
|
95
|
+
#ifndef EIGEN_DEFAULT_DENSE_INDEX_TYPE
|
96
|
+
#define EIGEN_DEFAULT_DENSE_INDEX_TYPE std::ptrdiff_t
|
97
|
+
#endif
|
98
|
+
|
99
|
+
// A Clang feature extension to determine compiler features.
|
100
|
+
// We use it to determine 'cxx_rvalue_references'
|
101
|
+
#ifndef __has_feature
|
102
|
+
# define __has_feature(x) 0
|
103
|
+
#endif
|
104
|
+
|
105
|
+
// Do we support r-value references?
|
106
|
+
#if (__has_feature(cxx_rvalue_references) || \
|
107
|
+
defined(__GXX_EXPERIMENTAL_CXX0X__) || \
|
108
|
+
(defined(_MSC_VER) && _MSC_VER >= 1600))
|
109
|
+
#define EIGEN_HAVE_RVALUE_REFERENCES
|
110
|
+
#endif
|
111
|
+
|
112
|
+
|
113
|
+
// Cross compiler wrapper around LLVM's __has_builtin
|
114
|
+
#ifdef __has_builtin
|
115
|
+
# define EIGEN_HAS_BUILTIN(x) __has_builtin(x)
|
116
|
+
#else
|
117
|
+
# define EIGEN_HAS_BUILTIN(x) 0
|
118
|
+
#endif
|
119
|
+
|
120
|
+
/** Allows to disable some optimizations which might affect the accuracy of the result.
|
121
|
+
* Such optimization are enabled by default, and set EIGEN_FAST_MATH to 0 to disable them.
|
122
|
+
* They currently include:
|
123
|
+
* - single precision Cwise::sin() and Cwise::cos() when SSE vectorization is enabled.
|
124
|
+
*/
|
125
|
+
#ifndef EIGEN_FAST_MATH
|
126
|
+
#define EIGEN_FAST_MATH 1
|
127
|
+
#endif
|
128
|
+
|
129
|
+
#define EIGEN_DEBUG_VAR(x) std::cerr << #x << " = " << x << std::endl;
|
130
|
+
|
131
|
+
// concatenate two tokens
|
132
|
+
#define EIGEN_CAT2(a,b) a ## b
|
133
|
+
#define EIGEN_CAT(a,b) EIGEN_CAT2(a,b)
|
134
|
+
|
135
|
+
// convert a token to a string
|
136
|
+
#define EIGEN_MAKESTRING2(a) #a
|
137
|
+
#define EIGEN_MAKESTRING(a) EIGEN_MAKESTRING2(a)
|
138
|
+
|
139
|
+
// EIGEN_STRONG_INLINE is a stronger version of the inline, using __forceinline on MSVC,
|
140
|
+
// but it still doesn't use GCC's always_inline. This is useful in (common) situations where MSVC needs forceinline
|
141
|
+
// but GCC is still doing fine with just inline.
|
142
|
+
#if (defined _MSC_VER) || (defined __INTEL_COMPILER)
|
143
|
+
#define EIGEN_STRONG_INLINE __forceinline
|
144
|
+
#else
|
145
|
+
#define EIGEN_STRONG_INLINE inline
|
146
|
+
#endif
|
147
|
+
|
148
|
+
// EIGEN_ALWAYS_INLINE is the stronget, it has the effect of making the function inline and adding every possible
|
149
|
+
// attribute to maximize inlining. This should only be used when really necessary: in particular,
|
150
|
+
// it uses __attribute__((always_inline)) on GCC, which most of the time is useless and can severely harm compile times.
|
151
|
+
// FIXME with the always_inline attribute,
|
152
|
+
// gcc 3.4.x reports the following compilation error:
|
153
|
+
// Eval.h:91: sorry, unimplemented: inlining failed in call to 'const Eigen::Eval<Derived> Eigen::MatrixBase<Scalar, Derived>::eval() const'
|
154
|
+
// : function body not available
|
155
|
+
#if EIGEN_GNUC_AT_LEAST(4,0)
|
156
|
+
#define EIGEN_ALWAYS_INLINE __attribute__((always_inline)) inline
|
157
|
+
#else
|
158
|
+
#define EIGEN_ALWAYS_INLINE EIGEN_STRONG_INLINE
|
159
|
+
#endif
|
160
|
+
|
161
|
+
#if (defined __GNUC__)
|
162
|
+
#define EIGEN_DONT_INLINE __attribute__((noinline))
|
163
|
+
#elif (defined _MSC_VER)
|
164
|
+
#define EIGEN_DONT_INLINE __declspec(noinline)
|
165
|
+
#else
|
166
|
+
#define EIGEN_DONT_INLINE
|
167
|
+
#endif
|
168
|
+
|
169
|
+
#if (defined __GNUC__)
|
170
|
+
#define EIGEN_PERMISSIVE_EXPR __extension__
|
171
|
+
#else
|
172
|
+
#define EIGEN_PERMISSIVE_EXPR
|
173
|
+
#endif
|
174
|
+
|
175
|
+
// this macro allows to get rid of linking errors about multiply defined functions.
|
176
|
+
// - static is not very good because it prevents definitions from different object files to be merged.
|
177
|
+
// So static causes the resulting linked executable to be bloated with multiple copies of the same function.
|
178
|
+
// - inline is not perfect either as it unwantedly hints the compiler toward inlining the function.
|
179
|
+
#define EIGEN_DECLARE_FUNCTION_ALLOWING_MULTIPLE_DEFINITIONS
|
180
|
+
#define EIGEN_DEFINE_FUNCTION_ALLOWING_MULTIPLE_DEFINITIONS inline
|
181
|
+
|
182
|
+
#ifdef NDEBUG
|
183
|
+
# ifndef EIGEN_NO_DEBUG
|
184
|
+
# define EIGEN_NO_DEBUG
|
185
|
+
# endif
|
186
|
+
#endif
|
187
|
+
|
188
|
+
// eigen_plain_assert is where we implement the workaround for the assert() bug in GCC <= 4.3, see bug 89
|
189
|
+
#ifdef EIGEN_NO_DEBUG
|
190
|
+
#define eigen_plain_assert(x)
|
191
|
+
#else
|
192
|
+
#if EIGEN_SAFE_TO_USE_STANDARD_ASSERT_MACRO
|
193
|
+
namespace Eigen {
|
194
|
+
namespace internal {
|
195
|
+
inline bool copy_bool(bool b) { return b; }
|
196
|
+
}
|
197
|
+
}
|
198
|
+
#define eigen_plain_assert(x) assert(x)
|
199
|
+
#else
|
200
|
+
// work around bug 89
|
201
|
+
#include <cstdlib> // for abort
|
202
|
+
#include <iostream> // for std::cerr
|
203
|
+
|
204
|
+
namespace Eigen {
|
205
|
+
namespace internal {
|
206
|
+
// trivial function copying a bool. Must be EIGEN_DONT_INLINE, so we implement it after including Eigen headers.
|
207
|
+
// see bug 89.
|
208
|
+
namespace {
|
209
|
+
EIGEN_DONT_INLINE bool copy_bool(bool b) { return b; }
|
210
|
+
}
|
211
|
+
inline void assert_fail(const char *condition, const char *function, const char *file, int line)
|
212
|
+
{
|
213
|
+
std::cerr << "assertion failed: " << condition << " in function " << function << " at " << file << ":" << line << std::endl;
|
214
|
+
abort();
|
215
|
+
}
|
216
|
+
}
|
217
|
+
}
|
218
|
+
#define eigen_plain_assert(x) \
|
219
|
+
do { \
|
220
|
+
if(!Eigen::internal::copy_bool(x)) \
|
221
|
+
Eigen::internal::assert_fail(EIGEN_MAKESTRING(x), __PRETTY_FUNCTION__, __FILE__, __LINE__); \
|
222
|
+
} while(false)
|
223
|
+
#endif
|
224
|
+
#endif
|
225
|
+
|
226
|
+
// eigen_assert can be overridden
|
227
|
+
#ifndef eigen_assert
|
228
|
+
#define eigen_assert(x) eigen_plain_assert(x)
|
229
|
+
#endif
|
230
|
+
|
231
|
+
#ifdef EIGEN_INTERNAL_DEBUGGING
|
232
|
+
#define eigen_internal_assert(x) eigen_assert(x)
|
233
|
+
#else
|
234
|
+
#define eigen_internal_assert(x)
|
235
|
+
#endif
|
236
|
+
|
237
|
+
#ifdef EIGEN_NO_DEBUG
|
238
|
+
#define EIGEN_ONLY_USED_FOR_DEBUG(x) (void)x
|
239
|
+
#else
|
240
|
+
#define EIGEN_ONLY_USED_FOR_DEBUG(x)
|
241
|
+
#endif
|
242
|
+
|
243
|
+
#ifndef EIGEN_NO_DEPRECATED_WARNING
|
244
|
+
#if (defined __GNUC__)
|
245
|
+
#define EIGEN_DEPRECATED __attribute__((deprecated))
|
246
|
+
#elif (defined _MSC_VER)
|
247
|
+
#define EIGEN_DEPRECATED __declspec(deprecated)
|
248
|
+
#else
|
249
|
+
#define EIGEN_DEPRECATED
|
250
|
+
#endif
|
251
|
+
#else
|
252
|
+
#define EIGEN_DEPRECATED
|
253
|
+
#endif
|
254
|
+
|
255
|
+
#if (defined __GNUC__)
|
256
|
+
#define EIGEN_UNUSED __attribute__((unused))
|
257
|
+
#else
|
258
|
+
#define EIGEN_UNUSED
|
259
|
+
#endif
|
260
|
+
|
261
|
+
// Suppresses 'unused variable' warnings.
|
262
|
+
namespace Eigen {
|
263
|
+
namespace internal {
|
264
|
+
template<typename T> void ignore_unused_variable(const T&) {}
|
265
|
+
}
|
266
|
+
}
|
267
|
+
#define EIGEN_UNUSED_VARIABLE(var) Eigen::internal::ignore_unused_variable(var);
|
268
|
+
|
269
|
+
#if !defined(EIGEN_ASM_COMMENT)
|
270
|
+
#if (defined __GNUC__) && ( defined(__i386__) || defined(__x86_64__) )
|
271
|
+
#define EIGEN_ASM_COMMENT(X) __asm__("#" X)
|
272
|
+
#else
|
273
|
+
#define EIGEN_ASM_COMMENT(X)
|
274
|
+
#endif
|
275
|
+
#endif
|
276
|
+
|
277
|
+
/* EIGEN_ALIGN_TO_BOUNDARY(n) forces data to be n-byte aligned. This is used to satisfy SIMD requirements.
|
278
|
+
* However, we do that EVEN if vectorization (EIGEN_VECTORIZE) is disabled,
|
279
|
+
* so that vectorization doesn't affect binary compatibility.
|
280
|
+
*
|
281
|
+
* If we made alignment depend on whether or not EIGEN_VECTORIZE is defined, it would be impossible to link
|
282
|
+
* vectorized and non-vectorized code.
|
283
|
+
*/
|
284
|
+
#if (defined __GNUC__) || (defined __PGI) || (defined __IBMCPP__) || (defined __ARMCC_VERSION)
|
285
|
+
#define EIGEN_ALIGN_TO_BOUNDARY(n) __attribute__((aligned(n)))
|
286
|
+
#elif (defined _MSC_VER)
|
287
|
+
#define EIGEN_ALIGN_TO_BOUNDARY(n) __declspec(align(n))
|
288
|
+
#elif (defined __SUNPRO_CC)
|
289
|
+
// FIXME not sure about this one:
|
290
|
+
#define EIGEN_ALIGN_TO_BOUNDARY(n) __attribute__((aligned(n)))
|
291
|
+
#else
|
292
|
+
#error Please tell me what is the equivalent of __attribute__((aligned(n))) for your compiler
|
293
|
+
#endif
|
294
|
+
|
295
|
+
#define EIGEN_ALIGN8 EIGEN_ALIGN_TO_BOUNDARY(8)
|
296
|
+
#define EIGEN_ALIGN16 EIGEN_ALIGN_TO_BOUNDARY(16)
|
297
|
+
|
298
|
+
#if EIGEN_ALIGN_STATICALLY
|
299
|
+
#define EIGEN_USER_ALIGN_TO_BOUNDARY(n) EIGEN_ALIGN_TO_BOUNDARY(n)
|
300
|
+
#define EIGEN_USER_ALIGN16 EIGEN_ALIGN16
|
301
|
+
#else
|
302
|
+
#define EIGEN_USER_ALIGN_TO_BOUNDARY(n)
|
303
|
+
#define EIGEN_USER_ALIGN16
|
304
|
+
#endif
|
305
|
+
|
306
|
+
#ifdef EIGEN_DONT_USE_RESTRICT_KEYWORD
|
307
|
+
#define EIGEN_RESTRICT
|
308
|
+
#endif
|
309
|
+
#ifndef EIGEN_RESTRICT
|
310
|
+
#define EIGEN_RESTRICT __restrict
|
311
|
+
#endif
|
312
|
+
|
313
|
+
#ifndef EIGEN_STACK_ALLOCATION_LIMIT
|
314
|
+
// 131072 == 128 KB
|
315
|
+
#define EIGEN_STACK_ALLOCATION_LIMIT 131072
|
316
|
+
#endif
|
317
|
+
|
318
|
+
#ifndef EIGEN_DEFAULT_IO_FORMAT
|
319
|
+
#ifdef EIGEN_MAKING_DOCS
|
320
|
+
// format used in Eigen's documentation
|
321
|
+
// needed to define it here as escaping characters in CMake add_definition's argument seems very problematic.
|
322
|
+
#define EIGEN_DEFAULT_IO_FORMAT Eigen::IOFormat(3, 0, " ", "\n", "", "")
|
323
|
+
#else
|
324
|
+
#define EIGEN_DEFAULT_IO_FORMAT Eigen::IOFormat()
|
325
|
+
#endif
|
326
|
+
#endif
|
327
|
+
|
328
|
+
// just an empty macro !
|
329
|
+
#define EIGEN_EMPTY
|
330
|
+
|
331
|
+
#if defined(_MSC_VER) && (_MSC_VER < 1900) && (!defined(__INTEL_COMPILER))
|
332
|
+
#define EIGEN_INHERIT_ASSIGNMENT_EQUAL_OPERATOR(Derived) \
|
333
|
+
using Base::operator =;
|
334
|
+
#elif defined(__clang__) // workaround clang bug (see http://forum.kde.org/viewtopic.php?f=74&t=102653)
|
335
|
+
#define EIGEN_INHERIT_ASSIGNMENT_EQUAL_OPERATOR(Derived) \
|
336
|
+
using Base::operator =; \
|
337
|
+
EIGEN_STRONG_INLINE Derived& operator=(const Derived& other) { Base::operator=(other); return *this; } \
|
338
|
+
template <typename OtherDerived> \
|
339
|
+
EIGEN_STRONG_INLINE Derived& operator=(const DenseBase<OtherDerived>& other) { Base::operator=(other.derived()); return *this; }
|
340
|
+
#else
|
341
|
+
#define EIGEN_INHERIT_ASSIGNMENT_EQUAL_OPERATOR(Derived) \
|
342
|
+
using Base::operator =; \
|
343
|
+
EIGEN_STRONG_INLINE Derived& operator=(const Derived& other) \
|
344
|
+
{ \
|
345
|
+
Base::operator=(other); \
|
346
|
+
return *this; \
|
347
|
+
}
|
348
|
+
#endif
|
349
|
+
|
350
|
+
/** \internal
|
351
|
+
* \brief Macro to manually inherit assignment operators.
|
352
|
+
* This is necessary, because the implicitly defined assignment operator gets deleted when a custom operator= is defined.
|
353
|
+
*/
|
354
|
+
#define EIGEN_INHERIT_ASSIGNMENT_OPERATORS(Derived) EIGEN_INHERIT_ASSIGNMENT_EQUAL_OPERATOR(Derived)
|
355
|
+
|
356
|
+
/**
|
357
|
+
* Just a side note. Commenting within defines works only by documenting
|
358
|
+
* behind the object (via '!<'). Comments cannot be multi-line and thus
|
359
|
+
* we have these extra long lines. What is confusing doxygen over here is
|
360
|
+
* that we use '\' and basically have a bunch of typedefs with their
|
361
|
+
* documentation in a single line.
|
362
|
+
**/
|
363
|
+
|
364
|
+
#define EIGEN_GENERIC_PUBLIC_INTERFACE(Derived) \
|
365
|
+
typedef typename Eigen::internal::traits<Derived>::Scalar Scalar; /*!< \brief Numeric type, e.g. float, double, int or std::complex<float>. */ \
|
366
|
+
typedef typename Eigen::NumTraits<Scalar>::Real RealScalar; /*!< \brief The underlying numeric type for composed scalar types. \details In cases where Scalar is e.g. std::complex<T>, T were corresponding to RealScalar. */ \
|
367
|
+
typedef typename Base::CoeffReturnType CoeffReturnType; /*!< \brief The return type for coefficient access. \details Depending on whether the object allows direct coefficient access (e.g. for a MatrixXd), this type is either 'const Scalar&' or simply 'Scalar' for objects that do not allow direct coefficient access. */ \
|
368
|
+
typedef typename Eigen::internal::nested<Derived>::type Nested; \
|
369
|
+
typedef typename Eigen::internal::traits<Derived>::StorageKind StorageKind; \
|
370
|
+
typedef typename Eigen::internal::traits<Derived>::Index Index; \
|
371
|
+
enum { RowsAtCompileTime = Eigen::internal::traits<Derived>::RowsAtCompileTime, \
|
372
|
+
ColsAtCompileTime = Eigen::internal::traits<Derived>::ColsAtCompileTime, \
|
373
|
+
Flags = Eigen::internal::traits<Derived>::Flags, \
|
374
|
+
CoeffReadCost = Eigen::internal::traits<Derived>::CoeffReadCost, \
|
375
|
+
SizeAtCompileTime = Base::SizeAtCompileTime, \
|
376
|
+
MaxSizeAtCompileTime = Base::MaxSizeAtCompileTime, \
|
377
|
+
IsVectorAtCompileTime = Base::IsVectorAtCompileTime };
|
378
|
+
|
379
|
+
|
380
|
+
#define EIGEN_DENSE_PUBLIC_INTERFACE(Derived) \
|
381
|
+
typedef typename Eigen::internal::traits<Derived>::Scalar Scalar; /*!< \brief Numeric type, e.g. float, double, int or std::complex<float>. */ \
|
382
|
+
typedef typename Eigen::NumTraits<Scalar>::Real RealScalar; /*!< \brief The underlying numeric type for composed scalar types. \details In cases where Scalar is e.g. std::complex<T>, T were corresponding to RealScalar. */ \
|
383
|
+
typedef typename Base::PacketScalar PacketScalar; \
|
384
|
+
typedef typename Base::CoeffReturnType CoeffReturnType; /*!< \brief The return type for coefficient access. \details Depending on whether the object allows direct coefficient access (e.g. for a MatrixXd), this type is either 'const Scalar&' or simply 'Scalar' for objects that do not allow direct coefficient access. */ \
|
385
|
+
typedef typename Eigen::internal::nested<Derived>::type Nested; \
|
386
|
+
typedef typename Eigen::internal::traits<Derived>::StorageKind StorageKind; \
|
387
|
+
typedef typename Eigen::internal::traits<Derived>::Index Index; \
|
388
|
+
enum { RowsAtCompileTime = Eigen::internal::traits<Derived>::RowsAtCompileTime, \
|
389
|
+
ColsAtCompileTime = Eigen::internal::traits<Derived>::ColsAtCompileTime, \
|
390
|
+
MaxRowsAtCompileTime = Eigen::internal::traits<Derived>::MaxRowsAtCompileTime, \
|
391
|
+
MaxColsAtCompileTime = Eigen::internal::traits<Derived>::MaxColsAtCompileTime, \
|
392
|
+
Flags = Eigen::internal::traits<Derived>::Flags, \
|
393
|
+
CoeffReadCost = Eigen::internal::traits<Derived>::CoeffReadCost, \
|
394
|
+
SizeAtCompileTime = Base::SizeAtCompileTime, \
|
395
|
+
MaxSizeAtCompileTime = Base::MaxSizeAtCompileTime, \
|
396
|
+
IsVectorAtCompileTime = Base::IsVectorAtCompileTime }; \
|
397
|
+
using Base::derived; \
|
398
|
+
using Base::const_cast_derived;
|
399
|
+
|
400
|
+
|
401
|
+
#define EIGEN_PLAIN_ENUM_MIN(a,b) (((int)a <= (int)b) ? (int)a : (int)b)
|
402
|
+
#define EIGEN_PLAIN_ENUM_MAX(a,b) (((int)a >= (int)b) ? (int)a : (int)b)
|
403
|
+
|
404
|
+
// EIGEN_SIZE_MIN_PREFER_DYNAMIC gives the min between compile-time sizes. 0 has absolute priority, followed by 1,
|
405
|
+
// followed by Dynamic, followed by other finite values. The reason for giving Dynamic the priority over
|
406
|
+
// finite values is that min(3, Dynamic) should be Dynamic, since that could be anything between 0 and 3.
|
407
|
+
#define EIGEN_SIZE_MIN_PREFER_DYNAMIC(a,b) (((int)a == 0 || (int)b == 0) ? 0 \
|
408
|
+
: ((int)a == 1 || (int)b == 1) ? 1 \
|
409
|
+
: ((int)a == Dynamic || (int)b == Dynamic) ? Dynamic \
|
410
|
+
: ((int)a <= (int)b) ? (int)a : (int)b)
|
411
|
+
|
412
|
+
// EIGEN_SIZE_MIN_PREFER_FIXED is a variant of EIGEN_SIZE_MIN_PREFER_DYNAMIC comparing MaxSizes. The difference is that finite values
|
413
|
+
// now have priority over Dynamic, so that min(3, Dynamic) gives 3. Indeed, whatever the actual value is
|
414
|
+
// (between 0 and 3), it is not more than 3.
|
415
|
+
#define EIGEN_SIZE_MIN_PREFER_FIXED(a,b) (((int)a == 0 || (int)b == 0) ? 0 \
|
416
|
+
: ((int)a == 1 || (int)b == 1) ? 1 \
|
417
|
+
: ((int)a == Dynamic && (int)b == Dynamic) ? Dynamic \
|
418
|
+
: ((int)a == Dynamic) ? (int)b \
|
419
|
+
: ((int)b == Dynamic) ? (int)a \
|
420
|
+
: ((int)a <= (int)b) ? (int)a : (int)b)
|
421
|
+
|
422
|
+
// see EIGEN_SIZE_MIN_PREFER_DYNAMIC. No need for a separate variant for MaxSizes here.
|
423
|
+
#define EIGEN_SIZE_MAX(a,b) (((int)a == Dynamic || (int)b == Dynamic) ? Dynamic \
|
424
|
+
: ((int)a >= (int)b) ? (int)a : (int)b)
|
425
|
+
|
426
|
+
#define EIGEN_ADD_COST(a,b) int(a)==Dynamic || int(b)==Dynamic ? Dynamic : int(a)+int(b)
|
427
|
+
|
428
|
+
#define EIGEN_LOGICAL_XOR(a,b) (((a) || (b)) && !((a) && (b)))
|
429
|
+
|
430
|
+
#define EIGEN_IMPLIES(a,b) (!(a) || (b))
|
431
|
+
|
432
|
+
#define EIGEN_MAKE_CWISE_BINARY_OP(METHOD,FUNCTOR) \
|
433
|
+
template<typename OtherDerived> \
|
434
|
+
EIGEN_STRONG_INLINE const CwiseBinaryOp<FUNCTOR<Scalar>, const Derived, const OtherDerived> \
|
435
|
+
(METHOD)(const EIGEN_CURRENT_STORAGE_BASE_CLASS<OtherDerived> &other) const \
|
436
|
+
{ \
|
437
|
+
return CwiseBinaryOp<FUNCTOR<Scalar>, const Derived, const OtherDerived>(derived(), other.derived()); \
|
438
|
+
}
|
439
|
+
|
440
|
+
// the expression type of a cwise product
|
441
|
+
#define EIGEN_CWISE_PRODUCT_RETURN_TYPE(LHS,RHS) \
|
442
|
+
CwiseBinaryOp< \
|
443
|
+
internal::scalar_product_op< \
|
444
|
+
typename internal::traits<LHS>::Scalar, \
|
445
|
+
typename internal::traits<RHS>::Scalar \
|
446
|
+
>, \
|
447
|
+
const LHS, \
|
448
|
+
const RHS \
|
449
|
+
>
|
450
|
+
|
451
|
+
#endif // EIGEN_MACROS_H
|
@@ -0,0 +1,977 @@
|
|
1
|
+
// This file is part of Eigen, a lightweight C++ template library
|
2
|
+
// for linear algebra.
|
3
|
+
//
|
4
|
+
// Copyright (C) 2008-2010 Gael Guennebaud <gael.guennebaud@inria.fr>
|
5
|
+
// Copyright (C) 2008-2009 Benoit Jacob <jacob.benoit.1@gmail.com>
|
6
|
+
// Copyright (C) 2009 Kenneth Riddile <kfriddile@yahoo.com>
|
7
|
+
// Copyright (C) 2010 Hauke Heibel <hauke.heibel@gmail.com>
|
8
|
+
// Copyright (C) 2010 Thomas Capricelli <orzel@freehackers.org>
|
9
|
+
//
|
10
|
+
// This Source Code Form is subject to the terms of the Mozilla
|
11
|
+
// Public License v. 2.0. If a copy of the MPL was not distributed
|
12
|
+
// with this file, You can obtain one at http://mozilla.org/MPL/2.0/.
|
13
|
+
|
14
|
+
|
15
|
+
/*****************************************************************************
|
16
|
+
*** Platform checks for aligned malloc functions ***
|
17
|
+
*****************************************************************************/
|
18
|
+
|
19
|
+
#ifndef EIGEN_MEMORY_H
|
20
|
+
#define EIGEN_MEMORY_H
|
21
|
+
|
22
|
+
#ifndef EIGEN_MALLOC_ALREADY_ALIGNED
|
23
|
+
|
24
|
+
// Try to determine automatically if malloc is already aligned.
|
25
|
+
|
26
|
+
// On 64-bit systems, glibc's malloc returns 16-byte-aligned pointers, see:
|
27
|
+
// http://www.gnu.org/s/libc/manual/html_node/Aligned-Memory-Blocks.html
|
28
|
+
// This is true at least since glibc 2.8.
|
29
|
+
// This leaves the question how to detect 64-bit. According to this document,
|
30
|
+
// http://gcc.fyxm.net/summit/2003/Porting%20to%2064%20bit.pdf
|
31
|
+
// page 114, "[The] LP64 model [...] is used by all 64-bit UNIX ports" so it's indeed
|
32
|
+
// quite safe, at least within the context of glibc, to equate 64-bit with LP64.
|
33
|
+
#if defined(__GLIBC__) && ((__GLIBC__>=2 && __GLIBC_MINOR__ >= 8) || __GLIBC__>2) \
|
34
|
+
&& defined(__LP64__) && ! defined( __SANITIZE_ADDRESS__ )
|
35
|
+
#define EIGEN_GLIBC_MALLOC_ALREADY_ALIGNED 1
|
36
|
+
#else
|
37
|
+
#define EIGEN_GLIBC_MALLOC_ALREADY_ALIGNED 0
|
38
|
+
#endif
|
39
|
+
|
40
|
+
// FreeBSD 6 seems to have 16-byte aligned malloc
|
41
|
+
// See http://svn.freebsd.org/viewvc/base/stable/6/lib/libc/stdlib/malloc.c?view=markup
|
42
|
+
// FreeBSD 7 seems to have 16-byte aligned malloc except on ARM and MIPS architectures
|
43
|
+
// See http://svn.freebsd.org/viewvc/base/stable/7/lib/libc/stdlib/malloc.c?view=markup
|
44
|
+
#if defined(__FreeBSD__) && !defined(__arm__) && !defined(__mips__)
|
45
|
+
#define EIGEN_FREEBSD_MALLOC_ALREADY_ALIGNED 1
|
46
|
+
#else
|
47
|
+
#define EIGEN_FREEBSD_MALLOC_ALREADY_ALIGNED 0
|
48
|
+
#endif
|
49
|
+
|
50
|
+
#if defined(__APPLE__) \
|
51
|
+
|| defined(_WIN64) \
|
52
|
+
|| EIGEN_GLIBC_MALLOC_ALREADY_ALIGNED \
|
53
|
+
|| EIGEN_FREEBSD_MALLOC_ALREADY_ALIGNED
|
54
|
+
#define EIGEN_MALLOC_ALREADY_ALIGNED 1
|
55
|
+
#else
|
56
|
+
#define EIGEN_MALLOC_ALREADY_ALIGNED 0
|
57
|
+
#endif
|
58
|
+
|
59
|
+
#endif
|
60
|
+
|
61
|
+
// See bug 554 (http://eigen.tuxfamily.org/bz/show_bug.cgi?id=554)
|
62
|
+
// It seems to be unsafe to check _POSIX_ADVISORY_INFO without including unistd.h first.
|
63
|
+
// Currently, let's include it only on unix systems:
|
64
|
+
#if defined(__unix__) || defined(__unix)
|
65
|
+
#include <unistd.h>
|
66
|
+
#if ((defined __QNXNTO__) || (defined _GNU_SOURCE) || (defined __PGI) || ((defined _XOPEN_SOURCE) && (_XOPEN_SOURCE >= 600))) && (defined _POSIX_ADVISORY_INFO) && (_POSIX_ADVISORY_INFO > 0)
|
67
|
+
#define EIGEN_HAS_POSIX_MEMALIGN 1
|
68
|
+
#endif
|
69
|
+
#endif
|
70
|
+
|
71
|
+
#ifndef EIGEN_HAS_POSIX_MEMALIGN
|
72
|
+
#define EIGEN_HAS_POSIX_MEMALIGN 0
|
73
|
+
#endif
|
74
|
+
|
75
|
+
#ifdef EIGEN_VECTORIZE_SSE
|
76
|
+
#define EIGEN_HAS_MM_MALLOC 1
|
77
|
+
#else
|
78
|
+
#define EIGEN_HAS_MM_MALLOC 0
|
79
|
+
#endif
|
80
|
+
|
81
|
+
namespace Eigen {
|
82
|
+
|
83
|
+
namespace internal {
|
84
|
+
|
85
|
+
inline void throw_std_bad_alloc()
|
86
|
+
{
|
87
|
+
#ifdef EIGEN_EXCEPTIONS
|
88
|
+
throw std::bad_alloc();
|
89
|
+
#else
|
90
|
+
std::size_t huge = -1;
|
91
|
+
new int[huge];
|
92
|
+
#endif
|
93
|
+
}
|
94
|
+
|
95
|
+
/*****************************************************************************
|
96
|
+
*** Implementation of handmade aligned functions ***
|
97
|
+
*****************************************************************************/
|
98
|
+
|
99
|
+
/* ----- Hand made implementations of aligned malloc/free and realloc ----- */
|
100
|
+
|
101
|
+
/** \internal Like malloc, but the returned pointer is guaranteed to be 16-byte aligned.
|
102
|
+
* Fast, but wastes 16 additional bytes of memory. Does not throw any exception.
|
103
|
+
*/
|
104
|
+
inline void* handmade_aligned_malloc(std::size_t size)
|
105
|
+
{
|
106
|
+
void *original = std::malloc(size+16);
|
107
|
+
if (original == 0) return 0;
|
108
|
+
void *aligned = reinterpret_cast<void*>((reinterpret_cast<std::size_t>(original) & ~(std::size_t(15))) + 16);
|
109
|
+
*(reinterpret_cast<void**>(aligned) - 1) = original;
|
110
|
+
return aligned;
|
111
|
+
}
|
112
|
+
|
113
|
+
/** \internal Frees memory allocated with handmade_aligned_malloc */
|
114
|
+
inline void handmade_aligned_free(void *ptr)
|
115
|
+
{
|
116
|
+
if (ptr) std::free(*(reinterpret_cast<void**>(ptr) - 1));
|
117
|
+
}
|
118
|
+
|
119
|
+
/** \internal
|
120
|
+
* \brief Reallocates aligned memory.
|
121
|
+
* Since we know that our handmade version is based on std::realloc
|
122
|
+
* we can use std::realloc to implement efficient reallocation.
|
123
|
+
*/
|
124
|
+
inline void* handmade_aligned_realloc(void* ptr, std::size_t size, std::size_t = 0)
|
125
|
+
{
|
126
|
+
if (ptr == 0) return handmade_aligned_malloc(size);
|
127
|
+
void *original = *(reinterpret_cast<void**>(ptr) - 1);
|
128
|
+
std::ptrdiff_t previous_offset = static_cast<char *>(ptr)-static_cast<char *>(original);
|
129
|
+
original = std::realloc(original,size+16);
|
130
|
+
if (original == 0) return 0;
|
131
|
+
void *aligned = reinterpret_cast<void*>((reinterpret_cast<std::size_t>(original) & ~(std::size_t(15))) + 16);
|
132
|
+
void *previous_aligned = static_cast<char *>(original)+previous_offset;
|
133
|
+
if(aligned!=previous_aligned)
|
134
|
+
std::memmove(aligned, previous_aligned, size);
|
135
|
+
|
136
|
+
*(reinterpret_cast<void**>(aligned) - 1) = original;
|
137
|
+
return aligned;
|
138
|
+
}
|
139
|
+
|
140
|
+
/*****************************************************************************
|
141
|
+
*** Implementation of generic aligned realloc (when no realloc can be used)***
|
142
|
+
*****************************************************************************/
|
143
|
+
|
144
|
+
void* aligned_malloc(std::size_t size);
|
145
|
+
void aligned_free(void *ptr);
|
146
|
+
|
147
|
+
/** \internal
|
148
|
+
* \brief Reallocates aligned memory.
|
149
|
+
* Allows reallocation with aligned ptr types. This implementation will
|
150
|
+
* always create a new memory chunk and copy the old data.
|
151
|
+
*/
|
152
|
+
inline void* generic_aligned_realloc(void* ptr, size_t size, size_t old_size)
|
153
|
+
{
|
154
|
+
if (ptr==0)
|
155
|
+
return aligned_malloc(size);
|
156
|
+
|
157
|
+
if (size==0)
|
158
|
+
{
|
159
|
+
aligned_free(ptr);
|
160
|
+
return 0;
|
161
|
+
}
|
162
|
+
|
163
|
+
void* newptr = aligned_malloc(size);
|
164
|
+
if (newptr == 0)
|
165
|
+
{
|
166
|
+
#ifdef EIGEN_HAS_ERRNO
|
167
|
+
errno = ENOMEM; // according to the standard
|
168
|
+
#endif
|
169
|
+
return 0;
|
170
|
+
}
|
171
|
+
|
172
|
+
if (ptr != 0)
|
173
|
+
{
|
174
|
+
std::memcpy(newptr, ptr, (std::min)(size,old_size));
|
175
|
+
aligned_free(ptr);
|
176
|
+
}
|
177
|
+
|
178
|
+
return newptr;
|
179
|
+
}
|
180
|
+
|
181
|
+
/*****************************************************************************
|
182
|
+
*** Implementation of portable aligned versions of malloc/free/realloc ***
|
183
|
+
*****************************************************************************/
|
184
|
+
|
185
|
+
#ifdef EIGEN_NO_MALLOC
|
186
|
+
inline void check_that_malloc_is_allowed()
|
187
|
+
{
|
188
|
+
eigen_assert(false && "heap allocation is forbidden (EIGEN_NO_MALLOC is defined)");
|
189
|
+
}
|
190
|
+
#elif defined EIGEN_RUNTIME_NO_MALLOC
|
191
|
+
inline bool is_malloc_allowed_impl(bool update, bool new_value = false)
|
192
|
+
{
|
193
|
+
static bool value = true;
|
194
|
+
if (update == 1)
|
195
|
+
value = new_value;
|
196
|
+
return value;
|
197
|
+
}
|
198
|
+
inline bool is_malloc_allowed() { return is_malloc_allowed_impl(false); }
|
199
|
+
inline bool set_is_malloc_allowed(bool new_value) { return is_malloc_allowed_impl(true, new_value); }
|
200
|
+
inline void check_that_malloc_is_allowed()
|
201
|
+
{
|
202
|
+
eigen_assert(is_malloc_allowed() && "heap allocation is forbidden (EIGEN_RUNTIME_NO_MALLOC is defined and g_is_malloc_allowed is false)");
|
203
|
+
}
|
204
|
+
#else
|
205
|
+
inline void check_that_malloc_is_allowed()
|
206
|
+
{}
|
207
|
+
#endif
|
208
|
+
|
209
|
+
/** \internal Allocates \a size bytes. The returned pointer is guaranteed to have 16 bytes alignment.
|
210
|
+
* On allocation error, the returned pointer is null, and std::bad_alloc is thrown.
|
211
|
+
*/
|
212
|
+
inline void* aligned_malloc(size_t size)
|
213
|
+
{
|
214
|
+
check_that_malloc_is_allowed();
|
215
|
+
|
216
|
+
void *result;
|
217
|
+
#if !EIGEN_ALIGN
|
218
|
+
result = std::malloc(size);
|
219
|
+
#elif EIGEN_MALLOC_ALREADY_ALIGNED
|
220
|
+
result = std::malloc(size);
|
221
|
+
#elif EIGEN_HAS_POSIX_MEMALIGN
|
222
|
+
if(posix_memalign(&result, 16, size)) result = 0;
|
223
|
+
#elif EIGEN_HAS_MM_MALLOC
|
224
|
+
result = _mm_malloc(size, 16);
|
225
|
+
#elif defined(_MSC_VER) && (!defined(_WIN32_WCE))
|
226
|
+
result = _aligned_malloc(size, 16);
|
227
|
+
#else
|
228
|
+
result = handmade_aligned_malloc(size);
|
229
|
+
#endif
|
230
|
+
|
231
|
+
if(!result && size)
|
232
|
+
throw_std_bad_alloc();
|
233
|
+
|
234
|
+
return result;
|
235
|
+
}
|
236
|
+
|
237
|
+
/** \internal Frees memory allocated with aligned_malloc. */
|
238
|
+
inline void aligned_free(void *ptr)
|
239
|
+
{
|
240
|
+
#if !EIGEN_ALIGN
|
241
|
+
std::free(ptr);
|
242
|
+
#elif EIGEN_MALLOC_ALREADY_ALIGNED
|
243
|
+
std::free(ptr);
|
244
|
+
#elif EIGEN_HAS_POSIX_MEMALIGN
|
245
|
+
std::free(ptr);
|
246
|
+
#elif EIGEN_HAS_MM_MALLOC
|
247
|
+
_mm_free(ptr);
|
248
|
+
#elif defined(_MSC_VER) && (!defined(_WIN32_WCE))
|
249
|
+
_aligned_free(ptr);
|
250
|
+
#else
|
251
|
+
handmade_aligned_free(ptr);
|
252
|
+
#endif
|
253
|
+
}
|
254
|
+
|
255
|
+
/**
|
256
|
+
* \internal
|
257
|
+
* \brief Reallocates an aligned block of memory.
|
258
|
+
* \throws std::bad_alloc on allocation failure
|
259
|
+
**/
|
260
|
+
inline void* aligned_realloc(void *ptr, size_t new_size, size_t old_size)
|
261
|
+
{
|
262
|
+
EIGEN_UNUSED_VARIABLE(old_size);
|
263
|
+
|
264
|
+
void *result;
|
265
|
+
#if !EIGEN_ALIGN
|
266
|
+
result = std::realloc(ptr,new_size);
|
267
|
+
#elif EIGEN_MALLOC_ALREADY_ALIGNED
|
268
|
+
result = std::realloc(ptr,new_size);
|
269
|
+
#elif EIGEN_HAS_POSIX_MEMALIGN
|
270
|
+
result = generic_aligned_realloc(ptr,new_size,old_size);
|
271
|
+
#elif EIGEN_HAS_MM_MALLOC
|
272
|
+
// The defined(_mm_free) is just here to verify that this MSVC version
|
273
|
+
// implements _mm_malloc/_mm_free based on the corresponding _aligned_
|
274
|
+
// functions. This may not always be the case and we just try to be safe.
|
275
|
+
#if defined(_MSC_VER) && (!defined(_WIN32_WCE)) && defined(_mm_free)
|
276
|
+
result = _aligned_realloc(ptr,new_size,16);
|
277
|
+
#else
|
278
|
+
result = generic_aligned_realloc(ptr,new_size,old_size);
|
279
|
+
#endif
|
280
|
+
#elif defined(_MSC_VER) && (!defined(_WIN32_WCE))
|
281
|
+
result = _aligned_realloc(ptr,new_size,16);
|
282
|
+
#else
|
283
|
+
result = handmade_aligned_realloc(ptr,new_size,old_size);
|
284
|
+
#endif
|
285
|
+
|
286
|
+
if (!result && new_size)
|
287
|
+
throw_std_bad_alloc();
|
288
|
+
|
289
|
+
return result;
|
290
|
+
}
|
291
|
+
|
292
|
+
/*****************************************************************************
|
293
|
+
*** Implementation of conditionally aligned functions ***
|
294
|
+
*****************************************************************************/
|
295
|
+
|
296
|
+
/** \internal Allocates \a size bytes. If Align is true, then the returned ptr is 16-byte-aligned.
|
297
|
+
* On allocation error, the returned pointer is null, and a std::bad_alloc is thrown.
|
298
|
+
*/
|
299
|
+
template<bool Align> inline void* conditional_aligned_malloc(size_t size)
|
300
|
+
{
|
301
|
+
return aligned_malloc(size);
|
302
|
+
}
|
303
|
+
|
304
|
+
template<> inline void* conditional_aligned_malloc<false>(size_t size)
|
305
|
+
{
|
306
|
+
check_that_malloc_is_allowed();
|
307
|
+
|
308
|
+
void *result = std::malloc(size);
|
309
|
+
if(!result && size)
|
310
|
+
throw_std_bad_alloc();
|
311
|
+
return result;
|
312
|
+
}
|
313
|
+
|
314
|
+
/** \internal Frees memory allocated with conditional_aligned_malloc */
|
315
|
+
template<bool Align> inline void conditional_aligned_free(void *ptr)
|
316
|
+
{
|
317
|
+
aligned_free(ptr);
|
318
|
+
}
|
319
|
+
|
320
|
+
template<> inline void conditional_aligned_free<false>(void *ptr)
|
321
|
+
{
|
322
|
+
std::free(ptr);
|
323
|
+
}
|
324
|
+
|
325
|
+
template<bool Align> inline void* conditional_aligned_realloc(void* ptr, size_t new_size, size_t old_size)
|
326
|
+
{
|
327
|
+
return aligned_realloc(ptr, new_size, old_size);
|
328
|
+
}
|
329
|
+
|
330
|
+
template<> inline void* conditional_aligned_realloc<false>(void* ptr, size_t new_size, size_t)
|
331
|
+
{
|
332
|
+
return std::realloc(ptr, new_size);
|
333
|
+
}
|
334
|
+
|
335
|
+
/*****************************************************************************
|
336
|
+
*** Construction/destruction of array elements ***
|
337
|
+
*****************************************************************************/
|
338
|
+
|
339
|
+
/** \internal Constructs the elements of an array.
|
340
|
+
* The \a size parameter tells on how many objects to call the constructor of T.
|
341
|
+
*/
|
342
|
+
template<typename T> inline T* construct_elements_of_array(T *ptr, size_t size)
|
343
|
+
{
|
344
|
+
for (size_t i=0; i < size; ++i) ::new (ptr + i) T;
|
345
|
+
return ptr;
|
346
|
+
}
|
347
|
+
|
348
|
+
/** \internal Destructs the elements of an array.
|
349
|
+
* The \a size parameters tells on how many objects to call the destructor of T.
|
350
|
+
*/
|
351
|
+
template<typename T> inline void destruct_elements_of_array(T *ptr, size_t size)
|
352
|
+
{
|
353
|
+
// always destruct an array starting from the end.
|
354
|
+
if(ptr)
|
355
|
+
while(size) ptr[--size].~T();
|
356
|
+
}
|
357
|
+
|
358
|
+
/*****************************************************************************
|
359
|
+
*** Implementation of aligned new/delete-like functions ***
|
360
|
+
*****************************************************************************/
|
361
|
+
|
362
|
+
template<typename T>
|
363
|
+
EIGEN_ALWAYS_INLINE void check_size_for_overflow(size_t size)
|
364
|
+
{
|
365
|
+
if(size > size_t(-1) / sizeof(T))
|
366
|
+
throw_std_bad_alloc();
|
367
|
+
}
|
368
|
+
|
369
|
+
/** \internal Allocates \a size objects of type T. The returned pointer is guaranteed to have 16 bytes alignment.
|
370
|
+
* On allocation error, the returned pointer is undefined, but a std::bad_alloc is thrown.
|
371
|
+
* The default constructor of T is called.
|
372
|
+
*/
|
373
|
+
template<typename T> inline T* aligned_new(size_t size)
|
374
|
+
{
|
375
|
+
check_size_for_overflow<T>(size);
|
376
|
+
T *result = reinterpret_cast<T*>(aligned_malloc(sizeof(T)*size));
|
377
|
+
return construct_elements_of_array(result, size);
|
378
|
+
}
|
379
|
+
|
380
|
+
template<typename T, bool Align> inline T* conditional_aligned_new(size_t size)
|
381
|
+
{
|
382
|
+
check_size_for_overflow<T>(size);
|
383
|
+
T *result = reinterpret_cast<T*>(conditional_aligned_malloc<Align>(sizeof(T)*size));
|
384
|
+
return construct_elements_of_array(result, size);
|
385
|
+
}
|
386
|
+
|
387
|
+
/** \internal Deletes objects constructed with aligned_new
|
388
|
+
* The \a size parameters tells on how many objects to call the destructor of T.
|
389
|
+
*/
|
390
|
+
template<typename T> inline void aligned_delete(T *ptr, size_t size)
|
391
|
+
{
|
392
|
+
destruct_elements_of_array<T>(ptr, size);
|
393
|
+
aligned_free(ptr);
|
394
|
+
}
|
395
|
+
|
396
|
+
/** \internal Deletes objects constructed with conditional_aligned_new
|
397
|
+
* The \a size parameters tells on how many objects to call the destructor of T.
|
398
|
+
*/
|
399
|
+
template<typename T, bool Align> inline void conditional_aligned_delete(T *ptr, size_t size)
|
400
|
+
{
|
401
|
+
destruct_elements_of_array<T>(ptr, size);
|
402
|
+
conditional_aligned_free<Align>(ptr);
|
403
|
+
}
|
404
|
+
|
405
|
+
template<typename T, bool Align> inline T* conditional_aligned_realloc_new(T* pts, size_t new_size, size_t old_size)
|
406
|
+
{
|
407
|
+
check_size_for_overflow<T>(new_size);
|
408
|
+
check_size_for_overflow<T>(old_size);
|
409
|
+
if(new_size < old_size)
|
410
|
+
destruct_elements_of_array(pts+new_size, old_size-new_size);
|
411
|
+
T *result = reinterpret_cast<T*>(conditional_aligned_realloc<Align>(reinterpret_cast<void*>(pts), sizeof(T)*new_size, sizeof(T)*old_size));
|
412
|
+
if(new_size > old_size)
|
413
|
+
construct_elements_of_array(result+old_size, new_size-old_size);
|
414
|
+
return result;
|
415
|
+
}
|
416
|
+
|
417
|
+
|
418
|
+
template<typename T, bool Align> inline T* conditional_aligned_new_auto(size_t size)
|
419
|
+
{
|
420
|
+
if(size==0)
|
421
|
+
return 0; // short-cut. Also fixes Bug 884
|
422
|
+
check_size_for_overflow<T>(size);
|
423
|
+
T *result = reinterpret_cast<T*>(conditional_aligned_malloc<Align>(sizeof(T)*size));
|
424
|
+
if(NumTraits<T>::RequireInitialization)
|
425
|
+
construct_elements_of_array(result, size);
|
426
|
+
return result;
|
427
|
+
}
|
428
|
+
|
429
|
+
template<typename T, bool Align> inline T* conditional_aligned_realloc_new_auto(T* pts, size_t new_size, size_t old_size)
|
430
|
+
{
|
431
|
+
check_size_for_overflow<T>(new_size);
|
432
|
+
check_size_for_overflow<T>(old_size);
|
433
|
+
if(NumTraits<T>::RequireInitialization && (new_size < old_size))
|
434
|
+
destruct_elements_of_array(pts+new_size, old_size-new_size);
|
435
|
+
T *result = reinterpret_cast<T*>(conditional_aligned_realloc<Align>(reinterpret_cast<void*>(pts), sizeof(T)*new_size, sizeof(T)*old_size));
|
436
|
+
if(NumTraits<T>::RequireInitialization && (new_size > old_size))
|
437
|
+
construct_elements_of_array(result+old_size, new_size-old_size);
|
438
|
+
return result;
|
439
|
+
}
|
440
|
+
|
441
|
+
template<typename T, bool Align> inline void conditional_aligned_delete_auto(T *ptr, size_t size)
|
442
|
+
{
|
443
|
+
if(NumTraits<T>::RequireInitialization)
|
444
|
+
destruct_elements_of_array<T>(ptr, size);
|
445
|
+
conditional_aligned_free<Align>(ptr);
|
446
|
+
}
|
447
|
+
|
448
|
+
/****************************************************************************/
|
449
|
+
|
450
|
+
/** \internal Returns the index of the first element of the array that is well aligned for vectorization.
|
451
|
+
*
|
452
|
+
* \param array the address of the start of the array
|
453
|
+
* \param size the size of the array
|
454
|
+
*
|
455
|
+
* \note If no element of the array is well aligned, the size of the array is returned. Typically,
|
456
|
+
* for example with SSE, "well aligned" means 16-byte-aligned. If vectorization is disabled or if the
|
457
|
+
* packet size for the given scalar type is 1, then everything is considered well-aligned.
|
458
|
+
*
|
459
|
+
* \note If the scalar type is vectorizable, we rely on the following assumptions: sizeof(Scalar) is a
|
460
|
+
* power of 2, the packet size in bytes is also a power of 2, and is a multiple of sizeof(Scalar). On the
|
461
|
+
* other hand, we do not assume that the array address is a multiple of sizeof(Scalar), as that fails for
|
462
|
+
* example with Scalar=double on certain 32-bit platforms, see bug #79.
|
463
|
+
*
|
464
|
+
* There is also the variant first_aligned(const MatrixBase&) defined in DenseCoeffsBase.h.
|
465
|
+
*/
|
466
|
+
template<typename Scalar, typename Index>
|
467
|
+
static inline Index first_aligned(const Scalar* array, Index size)
|
468
|
+
{
|
469
|
+
static const Index PacketSize = packet_traits<Scalar>::size;
|
470
|
+
static const Index PacketAlignedMask = PacketSize-1;
|
471
|
+
|
472
|
+
if(PacketSize==1)
|
473
|
+
{
|
474
|
+
// Either there is no vectorization, or a packet consists of exactly 1 scalar so that all elements
|
475
|
+
// of the array have the same alignment.
|
476
|
+
return 0;
|
477
|
+
}
|
478
|
+
else if(size_t(array) & (sizeof(Scalar)-1))
|
479
|
+
{
|
480
|
+
// There is vectorization for this scalar type, but the array is not aligned to the size of a single scalar.
|
481
|
+
// Consequently, no element of the array is well aligned.
|
482
|
+
return size;
|
483
|
+
}
|
484
|
+
else
|
485
|
+
{
|
486
|
+
return std::min<Index>( (PacketSize - (Index((size_t(array)/sizeof(Scalar))) & PacketAlignedMask))
|
487
|
+
& PacketAlignedMask, size);
|
488
|
+
}
|
489
|
+
}
|
490
|
+
|
491
|
+
/** \internal Returns the smallest integer multiple of \a base and greater or equal to \a size
|
492
|
+
*/
|
493
|
+
template<typename Index>
|
494
|
+
inline static Index first_multiple(Index size, Index base)
|
495
|
+
{
|
496
|
+
return ((size+base-1)/base)*base;
|
497
|
+
}
|
498
|
+
|
499
|
+
// std::copy is much slower than memcpy, so let's introduce a smart_copy which
|
500
|
+
// use memcpy on trivial types, i.e., on types that does not require an initialization ctor.
|
501
|
+
template<typename T, bool UseMemcpy> struct smart_copy_helper;
|
502
|
+
|
503
|
+
template<typename T> void smart_copy(const T* start, const T* end, T* target)
|
504
|
+
{
|
505
|
+
smart_copy_helper<T,!NumTraits<T>::RequireInitialization>::run(start, end, target);
|
506
|
+
}
|
507
|
+
|
508
|
+
template<typename T> struct smart_copy_helper<T,true> {
|
509
|
+
static inline void run(const T* start, const T* end, T* target)
|
510
|
+
{ memcpy(target, start, std::ptrdiff_t(end)-std::ptrdiff_t(start)); }
|
511
|
+
};
|
512
|
+
|
513
|
+
template<typename T> struct smart_copy_helper<T,false> {
|
514
|
+
static inline void run(const T* start, const T* end, T* target)
|
515
|
+
{ std::copy(start, end, target); }
|
516
|
+
};
|
517
|
+
|
518
|
+
|
519
|
+
/*****************************************************************************
|
520
|
+
*** Implementation of runtime stack allocation (falling back to malloc) ***
|
521
|
+
*****************************************************************************/
|
522
|
+
|
523
|
+
// you can overwrite Eigen's default behavior regarding alloca by defining EIGEN_ALLOCA
|
524
|
+
// to the appropriate stack allocation function
|
525
|
+
#ifndef EIGEN_ALLOCA
|
526
|
+
#if (defined __linux__) || (defined __APPLE__) || (defined alloca)
|
527
|
+
#define EIGEN_ALLOCA alloca
|
528
|
+
#elif defined(_MSC_VER)
|
529
|
+
#define EIGEN_ALLOCA _alloca
|
530
|
+
#endif
|
531
|
+
#endif
|
532
|
+
|
533
|
+
// This helper class construct the allocated memory, and takes care of destructing and freeing the handled data
|
534
|
+
// at destruction time. In practice this helper class is mainly useful to avoid memory leak in case of exceptions.
|
535
|
+
template<typename T> class aligned_stack_memory_handler
|
536
|
+
{
|
537
|
+
public:
|
538
|
+
/* Creates a stack_memory_handler responsible for the buffer \a ptr of size \a size.
|
539
|
+
* Note that \a ptr can be 0 regardless of the other parameters.
|
540
|
+
* This constructor takes care of constructing/initializing the elements of the buffer if required by the scalar type T (see NumTraits<T>::RequireInitialization).
|
541
|
+
* In this case, the buffer elements will also be destructed when this handler will be destructed.
|
542
|
+
* Finally, if \a dealloc is true, then the pointer \a ptr is freed.
|
543
|
+
**/
|
544
|
+
aligned_stack_memory_handler(T* ptr, size_t size, bool dealloc)
|
545
|
+
: m_ptr(ptr), m_size(size), m_deallocate(dealloc)
|
546
|
+
{
|
547
|
+
if(NumTraits<T>::RequireInitialization && m_ptr)
|
548
|
+
Eigen::internal::construct_elements_of_array(m_ptr, size);
|
549
|
+
}
|
550
|
+
~aligned_stack_memory_handler()
|
551
|
+
{
|
552
|
+
if(NumTraits<T>::RequireInitialization && m_ptr)
|
553
|
+
Eigen::internal::destruct_elements_of_array<T>(m_ptr, m_size);
|
554
|
+
if(m_deallocate)
|
555
|
+
Eigen::internal::aligned_free(m_ptr);
|
556
|
+
}
|
557
|
+
protected:
|
558
|
+
T* m_ptr;
|
559
|
+
size_t m_size;
|
560
|
+
bool m_deallocate;
|
561
|
+
};
|
562
|
+
|
563
|
+
} // end namespace internal
|
564
|
+
|
565
|
+
/** \internal
|
566
|
+
* Declares, allocates and construct an aligned buffer named NAME of SIZE elements of type TYPE on the stack
|
567
|
+
* if SIZE is smaller than EIGEN_STACK_ALLOCATION_LIMIT, and if stack allocation is supported by the platform
|
568
|
+
* (currently, this is Linux and Visual Studio only). Otherwise the memory is allocated on the heap.
|
569
|
+
* The allocated buffer is automatically deleted when exiting the scope of this declaration.
|
570
|
+
* If BUFFER is non null, then the declared variable is simply an alias for BUFFER, and no allocation/deletion occurs.
|
571
|
+
* Here is an example:
|
572
|
+
* \code
|
573
|
+
* {
|
574
|
+
* ei_declare_aligned_stack_constructed_variable(float,data,size,0);
|
575
|
+
* // use data[0] to data[size-1]
|
576
|
+
* }
|
577
|
+
* \endcode
|
578
|
+
* The underlying stack allocation function can controlled with the EIGEN_ALLOCA preprocessor token.
|
579
|
+
*/
|
580
|
+
#ifdef EIGEN_ALLOCA
|
581
|
+
|
582
|
+
#if defined(__arm__) || defined(_WIN32)
|
583
|
+
#define EIGEN_ALIGNED_ALLOCA(SIZE) reinterpret_cast<void*>((reinterpret_cast<size_t>(EIGEN_ALLOCA(SIZE+16)) & ~(size_t(15))) + 16)
|
584
|
+
#else
|
585
|
+
#define EIGEN_ALIGNED_ALLOCA EIGEN_ALLOCA
|
586
|
+
#endif
|
587
|
+
|
588
|
+
#define ei_declare_aligned_stack_constructed_variable(TYPE,NAME,SIZE,BUFFER) \
|
589
|
+
Eigen::internal::check_size_for_overflow<TYPE>(SIZE); \
|
590
|
+
TYPE* NAME = (BUFFER)!=0 ? (BUFFER) \
|
591
|
+
: reinterpret_cast<TYPE*>( \
|
592
|
+
(sizeof(TYPE)*SIZE<=EIGEN_STACK_ALLOCATION_LIMIT) ? EIGEN_ALIGNED_ALLOCA(sizeof(TYPE)*SIZE) \
|
593
|
+
: Eigen::internal::aligned_malloc(sizeof(TYPE)*SIZE) ); \
|
594
|
+
Eigen::internal::aligned_stack_memory_handler<TYPE> EIGEN_CAT(NAME,_stack_memory_destructor)((BUFFER)==0 ? NAME : 0,SIZE,sizeof(TYPE)*SIZE>EIGEN_STACK_ALLOCATION_LIMIT)
|
595
|
+
|
596
|
+
#else
|
597
|
+
|
598
|
+
#define ei_declare_aligned_stack_constructed_variable(TYPE,NAME,SIZE,BUFFER) \
|
599
|
+
Eigen::internal::check_size_for_overflow<TYPE>(SIZE); \
|
600
|
+
TYPE* NAME = (BUFFER)!=0 ? BUFFER : reinterpret_cast<TYPE*>(Eigen::internal::aligned_malloc(sizeof(TYPE)*SIZE)); \
|
601
|
+
Eigen::internal::aligned_stack_memory_handler<TYPE> EIGEN_CAT(NAME,_stack_memory_destructor)((BUFFER)==0 ? NAME : 0,SIZE,true)
|
602
|
+
|
603
|
+
#endif
|
604
|
+
|
605
|
+
|
606
|
+
/*****************************************************************************
|
607
|
+
*** Implementation of EIGEN_MAKE_ALIGNED_OPERATOR_NEW [_IF] ***
|
608
|
+
*****************************************************************************/
|
609
|
+
|
610
|
+
#if EIGEN_ALIGN
|
611
|
+
#ifdef EIGEN_EXCEPTIONS
|
612
|
+
#define EIGEN_MAKE_ALIGNED_OPERATOR_NEW_NOTHROW(NeedsToAlign) \
|
613
|
+
void* operator new(size_t size, const std::nothrow_t&) throw() { \
|
614
|
+
try { return Eigen::internal::conditional_aligned_malloc<NeedsToAlign>(size); } \
|
615
|
+
catch (...) { return 0; } \
|
616
|
+
}
|
617
|
+
#else
|
618
|
+
#define EIGEN_MAKE_ALIGNED_OPERATOR_NEW_NOTHROW(NeedsToAlign) \
|
619
|
+
void* operator new(size_t size, const std::nothrow_t&) throw() { \
|
620
|
+
return Eigen::internal::conditional_aligned_malloc<NeedsToAlign>(size); \
|
621
|
+
}
|
622
|
+
#endif
|
623
|
+
|
624
|
+
#define EIGEN_MAKE_ALIGNED_OPERATOR_NEW_IF(NeedsToAlign) \
|
625
|
+
void *operator new(size_t size) { \
|
626
|
+
return Eigen::internal::conditional_aligned_malloc<NeedsToAlign>(size); \
|
627
|
+
} \
|
628
|
+
void *operator new[](size_t size) { \
|
629
|
+
return Eigen::internal::conditional_aligned_malloc<NeedsToAlign>(size); \
|
630
|
+
} \
|
631
|
+
void operator delete(void * ptr) throw() { Eigen::internal::conditional_aligned_free<NeedsToAlign>(ptr); } \
|
632
|
+
void operator delete[](void * ptr) throw() { Eigen::internal::conditional_aligned_free<NeedsToAlign>(ptr); } \
|
633
|
+
/* in-place new and delete. since (at least afaik) there is no actual */ \
|
634
|
+
/* memory allocated we can safely let the default implementation handle */ \
|
635
|
+
/* this particular case. */ \
|
636
|
+
static void *operator new(size_t size, void *ptr) { return ::operator new(size,ptr); } \
|
637
|
+
static void *operator new[](size_t size, void* ptr) { return ::operator new[](size,ptr); } \
|
638
|
+
void operator delete(void * memory, void *ptr) throw() { return ::operator delete(memory,ptr); } \
|
639
|
+
void operator delete[](void * memory, void *ptr) throw() { return ::operator delete[](memory,ptr); } \
|
640
|
+
/* nothrow-new (returns zero instead of std::bad_alloc) */ \
|
641
|
+
EIGEN_MAKE_ALIGNED_OPERATOR_NEW_NOTHROW(NeedsToAlign) \
|
642
|
+
void operator delete(void *ptr, const std::nothrow_t&) throw() { \
|
643
|
+
Eigen::internal::conditional_aligned_free<NeedsToAlign>(ptr); \
|
644
|
+
} \
|
645
|
+
typedef void eigen_aligned_operator_new_marker_type;
|
646
|
+
#else
|
647
|
+
#define EIGEN_MAKE_ALIGNED_OPERATOR_NEW_IF(NeedsToAlign)
|
648
|
+
#endif
|
649
|
+
|
650
|
+
#define EIGEN_MAKE_ALIGNED_OPERATOR_NEW EIGEN_MAKE_ALIGNED_OPERATOR_NEW_IF(true)
|
651
|
+
#define EIGEN_MAKE_ALIGNED_OPERATOR_NEW_IF_VECTORIZABLE_FIXED_SIZE(Scalar,Size) \
|
652
|
+
EIGEN_MAKE_ALIGNED_OPERATOR_NEW_IF(bool(((Size)!=Eigen::Dynamic) && ((sizeof(Scalar)*(Size))%16==0)))
|
653
|
+
|
654
|
+
/****************************************************************************/
|
655
|
+
|
656
|
+
/** \class aligned_allocator
|
657
|
+
* \ingroup Core_Module
|
658
|
+
*
|
659
|
+
* \brief STL compatible allocator to use with with 16 byte aligned types
|
660
|
+
*
|
661
|
+
* Example:
|
662
|
+
* \code
|
663
|
+
* // Matrix4f requires 16 bytes alignment:
|
664
|
+
* std::map< int, Matrix4f, std::less<int>,
|
665
|
+
* aligned_allocator<std::pair<const int, Matrix4f> > > my_map_mat4;
|
666
|
+
* // Vector3f does not require 16 bytes alignment, no need to use Eigen's allocator:
|
667
|
+
* std::map< int, Vector3f > my_map_vec3;
|
668
|
+
* \endcode
|
669
|
+
*
|
670
|
+
* \sa \ref TopicStlContainers.
|
671
|
+
*/
|
672
|
+
template<class T>
|
673
|
+
class aligned_allocator
|
674
|
+
{
|
675
|
+
public:
|
676
|
+
typedef size_t size_type;
|
677
|
+
typedef std::ptrdiff_t difference_type;
|
678
|
+
typedef T* pointer;
|
679
|
+
typedef const T* const_pointer;
|
680
|
+
typedef T& reference;
|
681
|
+
typedef const T& const_reference;
|
682
|
+
typedef T value_type;
|
683
|
+
|
684
|
+
template<class U>
|
685
|
+
struct rebind
|
686
|
+
{
|
687
|
+
typedef aligned_allocator<U> other;
|
688
|
+
};
|
689
|
+
|
690
|
+
pointer address( reference value ) const
|
691
|
+
{
|
692
|
+
return &value;
|
693
|
+
}
|
694
|
+
|
695
|
+
const_pointer address( const_reference value ) const
|
696
|
+
{
|
697
|
+
return &value;
|
698
|
+
}
|
699
|
+
|
700
|
+
aligned_allocator()
|
701
|
+
{
|
702
|
+
}
|
703
|
+
|
704
|
+
aligned_allocator( const aligned_allocator& )
|
705
|
+
{
|
706
|
+
}
|
707
|
+
|
708
|
+
template<class U>
|
709
|
+
aligned_allocator( const aligned_allocator<U>& )
|
710
|
+
{
|
711
|
+
}
|
712
|
+
|
713
|
+
~aligned_allocator()
|
714
|
+
{
|
715
|
+
}
|
716
|
+
|
717
|
+
size_type max_size() const
|
718
|
+
{
|
719
|
+
return (std::numeric_limits<size_type>::max)();
|
720
|
+
}
|
721
|
+
|
722
|
+
pointer allocate( size_type num, const void* hint = 0 )
|
723
|
+
{
|
724
|
+
EIGEN_UNUSED_VARIABLE(hint);
|
725
|
+
internal::check_size_for_overflow<T>(num);
|
726
|
+
return static_cast<pointer>( internal::aligned_malloc( num * sizeof(T) ) );
|
727
|
+
}
|
728
|
+
|
729
|
+
void construct( pointer p, const T& value )
|
730
|
+
{
|
731
|
+
::new( p ) T( value );
|
732
|
+
}
|
733
|
+
|
734
|
+
void destroy( pointer p )
|
735
|
+
{
|
736
|
+
p->~T();
|
737
|
+
}
|
738
|
+
|
739
|
+
void deallocate( pointer p, size_type /*num*/ )
|
740
|
+
{
|
741
|
+
internal::aligned_free( p );
|
742
|
+
}
|
743
|
+
|
744
|
+
bool operator!=(const aligned_allocator<T>& ) const
|
745
|
+
{ return false; }
|
746
|
+
|
747
|
+
bool operator==(const aligned_allocator<T>& ) const
|
748
|
+
{ return true; }
|
749
|
+
};
|
750
|
+
|
751
|
+
//---------- Cache sizes ----------
|
752
|
+
|
753
|
+
#if !defined(EIGEN_NO_CPUID)
|
754
|
+
# if defined(__GNUC__) && ( defined(__i386__) || defined(__x86_64__) )
|
755
|
+
# if defined(__PIC__) && defined(__i386__)
|
756
|
+
// Case for x86 with PIC
|
757
|
+
# define EIGEN_CPUID(abcd,func,id) \
|
758
|
+
__asm__ __volatile__ ("xchgl %%ebx, %k1;cpuid; xchgl %%ebx,%k1": "=a" (abcd[0]), "=&r" (abcd[1]), "=c" (abcd[2]), "=d" (abcd[3]) : "a" (func), "c" (id));
|
759
|
+
# elif defined(__PIC__) && defined(__x86_64__)
|
760
|
+
// Case for x64 with PIC. In theory this is only a problem with recent gcc and with medium or large code model, not with the default small code model.
|
761
|
+
// However, we cannot detect which code model is used, and the xchg overhead is negligible anyway.
|
762
|
+
# define EIGEN_CPUID(abcd,func,id) \
|
763
|
+
__asm__ __volatile__ ("xchg{q}\t{%%}rbx, %q1; cpuid; xchg{q}\t{%%}rbx, %q1": "=a" (abcd[0]), "=&r" (abcd[1]), "=c" (abcd[2]), "=d" (abcd[3]) : "0" (func), "2" (id));
|
764
|
+
# else
|
765
|
+
// Case for x86_64 or x86 w/o PIC
|
766
|
+
# define EIGEN_CPUID(abcd,func,id) \
|
767
|
+
__asm__ __volatile__ ("cpuid": "=a" (abcd[0]), "=b" (abcd[1]), "=c" (abcd[2]), "=d" (abcd[3]) : "0" (func), "2" (id) );
|
768
|
+
# endif
|
769
|
+
# elif defined(_MSC_VER)
|
770
|
+
# if (_MSC_VER > 1500) && ( defined(_M_IX86) || defined(_M_X64) )
|
771
|
+
# define EIGEN_CPUID(abcd,func,id) __cpuidex((int*)abcd,func,id)
|
772
|
+
# endif
|
773
|
+
# endif
|
774
|
+
#endif
|
775
|
+
|
776
|
+
namespace internal {
|
777
|
+
|
778
|
+
#ifdef EIGEN_CPUID
|
779
|
+
|
780
|
+
inline bool cpuid_is_vendor(int abcd[4], const int vendor[3])
|
781
|
+
{
|
782
|
+
return abcd[1]==vendor[0] && abcd[3]==vendor[1] && abcd[2]==vendor[2];
|
783
|
+
}
|
784
|
+
|
785
|
+
inline void queryCacheSizes_intel_direct(int& l1, int& l2, int& l3)
|
786
|
+
{
|
787
|
+
int abcd[4];
|
788
|
+
l1 = l2 = l3 = 0;
|
789
|
+
int cache_id = 0;
|
790
|
+
int cache_type = 0;
|
791
|
+
do {
|
792
|
+
abcd[0] = abcd[1] = abcd[2] = abcd[3] = 0;
|
793
|
+
EIGEN_CPUID(abcd,0x4,cache_id);
|
794
|
+
cache_type = (abcd[0] & 0x0F) >> 0;
|
795
|
+
if(cache_type==1||cache_type==3) // data or unified cache
|
796
|
+
{
|
797
|
+
int cache_level = (abcd[0] & 0xE0) >> 5; // A[7:5]
|
798
|
+
int ways = (abcd[1] & 0xFFC00000) >> 22; // B[31:22]
|
799
|
+
int partitions = (abcd[1] & 0x003FF000) >> 12; // B[21:12]
|
800
|
+
int line_size = (abcd[1] & 0x00000FFF) >> 0; // B[11:0]
|
801
|
+
int sets = (abcd[2]); // C[31:0]
|
802
|
+
|
803
|
+
int cache_size = (ways+1) * (partitions+1) * (line_size+1) * (sets+1);
|
804
|
+
|
805
|
+
switch(cache_level)
|
806
|
+
{
|
807
|
+
case 1: l1 = cache_size; break;
|
808
|
+
case 2: l2 = cache_size; break;
|
809
|
+
case 3: l3 = cache_size; break;
|
810
|
+
default: break;
|
811
|
+
}
|
812
|
+
}
|
813
|
+
cache_id++;
|
814
|
+
} while(cache_type>0 && cache_id<16);
|
815
|
+
}
|
816
|
+
|
817
|
+
inline void queryCacheSizes_intel_codes(int& l1, int& l2, int& l3)
|
818
|
+
{
|
819
|
+
int abcd[4];
|
820
|
+
abcd[0] = abcd[1] = abcd[2] = abcd[3] = 0;
|
821
|
+
l1 = l2 = l3 = 0;
|
822
|
+
EIGEN_CPUID(abcd,0x00000002,0);
|
823
|
+
unsigned char * bytes = reinterpret_cast<unsigned char *>(abcd)+2;
|
824
|
+
bool check_for_p2_core2 = false;
|
825
|
+
for(int i=0; i<14; ++i)
|
826
|
+
{
|
827
|
+
switch(bytes[i])
|
828
|
+
{
|
829
|
+
case 0x0A: l1 = 8; break; // 0Ah data L1 cache, 8 KB, 2 ways, 32 byte lines
|
830
|
+
case 0x0C: l1 = 16; break; // 0Ch data L1 cache, 16 KB, 4 ways, 32 byte lines
|
831
|
+
case 0x0E: l1 = 24; break; // 0Eh data L1 cache, 24 KB, 6 ways, 64 byte lines
|
832
|
+
case 0x10: l1 = 16; break; // 10h data L1 cache, 16 KB, 4 ways, 32 byte lines (IA-64)
|
833
|
+
case 0x15: l1 = 16; break; // 15h code L1 cache, 16 KB, 4 ways, 32 byte lines (IA-64)
|
834
|
+
case 0x2C: l1 = 32; break; // 2Ch data L1 cache, 32 KB, 8 ways, 64 byte lines
|
835
|
+
case 0x30: l1 = 32; break; // 30h code L1 cache, 32 KB, 8 ways, 64 byte lines
|
836
|
+
case 0x60: l1 = 16; break; // 60h data L1 cache, 16 KB, 8 ways, 64 byte lines, sectored
|
837
|
+
case 0x66: l1 = 8; break; // 66h data L1 cache, 8 KB, 4 ways, 64 byte lines, sectored
|
838
|
+
case 0x67: l1 = 16; break; // 67h data L1 cache, 16 KB, 4 ways, 64 byte lines, sectored
|
839
|
+
case 0x68: l1 = 32; break; // 68h data L1 cache, 32 KB, 4 ways, 64 byte lines, sectored
|
840
|
+
case 0x1A: l2 = 96; break; // code and data L2 cache, 96 KB, 6 ways, 64 byte lines (IA-64)
|
841
|
+
case 0x22: l3 = 512; break; // code and data L3 cache, 512 KB, 4 ways (!), 64 byte lines, dual-sectored
|
842
|
+
case 0x23: l3 = 1024; break; // code and data L3 cache, 1024 KB, 8 ways, 64 byte lines, dual-sectored
|
843
|
+
case 0x25: l3 = 2048; break; // code and data L3 cache, 2048 KB, 8 ways, 64 byte lines, dual-sectored
|
844
|
+
case 0x29: l3 = 4096; break; // code and data L3 cache, 4096 KB, 8 ways, 64 byte lines, dual-sectored
|
845
|
+
case 0x39: l2 = 128; break; // code and data L2 cache, 128 KB, 4 ways, 64 byte lines, sectored
|
846
|
+
case 0x3A: l2 = 192; break; // code and data L2 cache, 192 KB, 6 ways, 64 byte lines, sectored
|
847
|
+
case 0x3B: l2 = 128; break; // code and data L2 cache, 128 KB, 2 ways, 64 byte lines, sectored
|
848
|
+
case 0x3C: l2 = 256; break; // code and data L2 cache, 256 KB, 4 ways, 64 byte lines, sectored
|
849
|
+
case 0x3D: l2 = 384; break; // code and data L2 cache, 384 KB, 6 ways, 64 byte lines, sectored
|
850
|
+
case 0x3E: l2 = 512; break; // code and data L2 cache, 512 KB, 4 ways, 64 byte lines, sectored
|
851
|
+
case 0x40: l2 = 0; break; // no integrated L2 cache (P6 core) or L3 cache (P4 core)
|
852
|
+
case 0x41: l2 = 128; break; // code and data L2 cache, 128 KB, 4 ways, 32 byte lines
|
853
|
+
case 0x42: l2 = 256; break; // code and data L2 cache, 256 KB, 4 ways, 32 byte lines
|
854
|
+
case 0x43: l2 = 512; break; // code and data L2 cache, 512 KB, 4 ways, 32 byte lines
|
855
|
+
case 0x44: l2 = 1024; break; // code and data L2 cache, 1024 KB, 4 ways, 32 byte lines
|
856
|
+
case 0x45: l2 = 2048; break; // code and data L2 cache, 2048 KB, 4 ways, 32 byte lines
|
857
|
+
case 0x46: l3 = 4096; break; // code and data L3 cache, 4096 KB, 4 ways, 64 byte lines
|
858
|
+
case 0x47: l3 = 8192; break; // code and data L3 cache, 8192 KB, 8 ways, 64 byte lines
|
859
|
+
case 0x48: l2 = 3072; break; // code and data L2 cache, 3072 KB, 12 ways, 64 byte lines
|
860
|
+
case 0x49: if(l2!=0) l3 = 4096; else {check_for_p2_core2=true; l3 = l2 = 4096;} break;// code and data L3 cache, 4096 KB, 16 ways, 64 byte lines (P4) or L2 for core2
|
861
|
+
case 0x4A: l3 = 6144; break; // code and data L3 cache, 6144 KB, 12 ways, 64 byte lines
|
862
|
+
case 0x4B: l3 = 8192; break; // code and data L3 cache, 8192 KB, 16 ways, 64 byte lines
|
863
|
+
case 0x4C: l3 = 12288; break; // code and data L3 cache, 12288 KB, 12 ways, 64 byte lines
|
864
|
+
case 0x4D: l3 = 16384; break; // code and data L3 cache, 16384 KB, 16 ways, 64 byte lines
|
865
|
+
case 0x4E: l2 = 6144; break; // code and data L2 cache, 6144 KB, 24 ways, 64 byte lines
|
866
|
+
case 0x78: l2 = 1024; break; // code and data L2 cache, 1024 KB, 4 ways, 64 byte lines
|
867
|
+
case 0x79: l2 = 128; break; // code and data L2 cache, 128 KB, 8 ways, 64 byte lines, dual-sectored
|
868
|
+
case 0x7A: l2 = 256; break; // code and data L2 cache, 256 KB, 8 ways, 64 byte lines, dual-sectored
|
869
|
+
case 0x7B: l2 = 512; break; // code and data L2 cache, 512 KB, 8 ways, 64 byte lines, dual-sectored
|
870
|
+
case 0x7C: l2 = 1024; break; // code and data L2 cache, 1024 KB, 8 ways, 64 byte lines, dual-sectored
|
871
|
+
case 0x7D: l2 = 2048; break; // code and data L2 cache, 2048 KB, 8 ways, 64 byte lines
|
872
|
+
case 0x7E: l2 = 256; break; // code and data L2 cache, 256 KB, 8 ways, 128 byte lines, sect. (IA-64)
|
873
|
+
case 0x7F: l2 = 512; break; // code and data L2 cache, 512 KB, 2 ways, 64 byte lines
|
874
|
+
case 0x80: l2 = 512; break; // code and data L2 cache, 512 KB, 8 ways, 64 byte lines
|
875
|
+
case 0x81: l2 = 128; break; // code and data L2 cache, 128 KB, 8 ways, 32 byte lines
|
876
|
+
case 0x82: l2 = 256; break; // code and data L2 cache, 256 KB, 8 ways, 32 byte lines
|
877
|
+
case 0x83: l2 = 512; break; // code and data L2 cache, 512 KB, 8 ways, 32 byte lines
|
878
|
+
case 0x84: l2 = 1024; break; // code and data L2 cache, 1024 KB, 8 ways, 32 byte lines
|
879
|
+
case 0x85: l2 = 2048; break; // code and data L2 cache, 2048 KB, 8 ways, 32 byte lines
|
880
|
+
case 0x86: l2 = 512; break; // code and data L2 cache, 512 KB, 4 ways, 64 byte lines
|
881
|
+
case 0x87: l2 = 1024; break; // code and data L2 cache, 1024 KB, 8 ways, 64 byte lines
|
882
|
+
case 0x88: l3 = 2048; break; // code and data L3 cache, 2048 KB, 4 ways, 64 byte lines (IA-64)
|
883
|
+
case 0x89: l3 = 4096; break; // code and data L3 cache, 4096 KB, 4 ways, 64 byte lines (IA-64)
|
884
|
+
case 0x8A: l3 = 8192; break; // code and data L3 cache, 8192 KB, 4 ways, 64 byte lines (IA-64)
|
885
|
+
case 0x8D: l3 = 3072; break; // code and data L3 cache, 3072 KB, 12 ways, 128 byte lines (IA-64)
|
886
|
+
|
887
|
+
default: break;
|
888
|
+
}
|
889
|
+
}
|
890
|
+
if(check_for_p2_core2 && l2 == l3)
|
891
|
+
l3 = 0;
|
892
|
+
l1 *= 1024;
|
893
|
+
l2 *= 1024;
|
894
|
+
l3 *= 1024;
|
895
|
+
}
|
896
|
+
|
897
|
+
inline void queryCacheSizes_intel(int& l1, int& l2, int& l3, int max_std_funcs)
|
898
|
+
{
|
899
|
+
if(max_std_funcs>=4)
|
900
|
+
queryCacheSizes_intel_direct(l1,l2,l3);
|
901
|
+
else
|
902
|
+
queryCacheSizes_intel_codes(l1,l2,l3);
|
903
|
+
}
|
904
|
+
|
905
|
+
inline void queryCacheSizes_amd(int& l1, int& l2, int& l3)
|
906
|
+
{
|
907
|
+
int abcd[4];
|
908
|
+
abcd[0] = abcd[1] = abcd[2] = abcd[3] = 0;
|
909
|
+
EIGEN_CPUID(abcd,0x80000005,0);
|
910
|
+
l1 = (abcd[2] >> 24) * 1024; // C[31:24] = L1 size in KB
|
911
|
+
abcd[0] = abcd[1] = abcd[2] = abcd[3] = 0;
|
912
|
+
EIGEN_CPUID(abcd,0x80000006,0);
|
913
|
+
l2 = (abcd[2] >> 16) * 1024; // C[31;16] = l2 cache size in KB
|
914
|
+
l3 = ((abcd[3] & 0xFFFC000) >> 18) * 512 * 1024; // D[31;18] = l3 cache size in 512KB
|
915
|
+
}
|
916
|
+
#endif
|
917
|
+
|
918
|
+
/** \internal
|
919
|
+
* Queries and returns the cache sizes in Bytes of the L1, L2, and L3 data caches respectively */
|
920
|
+
inline void queryCacheSizes(int& l1, int& l2, int& l3)
|
921
|
+
{
|
922
|
+
#ifdef EIGEN_CPUID
|
923
|
+
int abcd[4];
|
924
|
+
const int GenuineIntel[] = {0x756e6547, 0x49656e69, 0x6c65746e};
|
925
|
+
const int AuthenticAMD[] = {0x68747541, 0x69746e65, 0x444d4163};
|
926
|
+
const int AMDisbetter_[] = {0x69444d41, 0x74656273, 0x21726574}; // "AMDisbetter!"
|
927
|
+
|
928
|
+
// identify the CPU vendor
|
929
|
+
EIGEN_CPUID(abcd,0x0,0);
|
930
|
+
int max_std_funcs = abcd[1];
|
931
|
+
if(cpuid_is_vendor(abcd,GenuineIntel))
|
932
|
+
queryCacheSizes_intel(l1,l2,l3,max_std_funcs);
|
933
|
+
else if(cpuid_is_vendor(abcd,AuthenticAMD) || cpuid_is_vendor(abcd,AMDisbetter_))
|
934
|
+
queryCacheSizes_amd(l1,l2,l3);
|
935
|
+
else
|
936
|
+
// by default let's use Intel's API
|
937
|
+
queryCacheSizes_intel(l1,l2,l3,max_std_funcs);
|
938
|
+
|
939
|
+
// here is the list of other vendors:
|
940
|
+
// ||cpuid_is_vendor(abcd,"VIA VIA VIA ")
|
941
|
+
// ||cpuid_is_vendor(abcd,"CyrixInstead")
|
942
|
+
// ||cpuid_is_vendor(abcd,"CentaurHauls")
|
943
|
+
// ||cpuid_is_vendor(abcd,"GenuineTMx86")
|
944
|
+
// ||cpuid_is_vendor(abcd,"TransmetaCPU")
|
945
|
+
// ||cpuid_is_vendor(abcd,"RiseRiseRise")
|
946
|
+
// ||cpuid_is_vendor(abcd,"Geode by NSC")
|
947
|
+
// ||cpuid_is_vendor(abcd,"SiS SiS SiS ")
|
948
|
+
// ||cpuid_is_vendor(abcd,"UMC UMC UMC ")
|
949
|
+
// ||cpuid_is_vendor(abcd,"NexGenDriven")
|
950
|
+
#else
|
951
|
+
l1 = l2 = l3 = -1;
|
952
|
+
#endif
|
953
|
+
}
|
954
|
+
|
955
|
+
/** \internal
|
956
|
+
* \returns the size in Bytes of the L1 data cache */
|
957
|
+
inline int queryL1CacheSize()
|
958
|
+
{
|
959
|
+
int l1(-1), l2, l3;
|
960
|
+
queryCacheSizes(l1,l2,l3);
|
961
|
+
return l1;
|
962
|
+
}
|
963
|
+
|
964
|
+
/** \internal
|
965
|
+
* \returns the size in Bytes of the L2 or L3 cache if this later is present */
|
966
|
+
inline int queryTopLevelCacheSize()
|
967
|
+
{
|
968
|
+
int l1, l2(-1), l3(-1);
|
969
|
+
queryCacheSizes(l1,l2,l3);
|
970
|
+
return (std::max)(l2,l3);
|
971
|
+
}
|
972
|
+
|
973
|
+
} // end namespace internal
|
974
|
+
|
975
|
+
} // end namespace Eigen
|
976
|
+
|
977
|
+
#endif // EIGEN_MEMORY_H
|