diff --git a/Eigen/src/Core/Matrix.h b/Eigen/src/Core/Matrix.h index 888729ee7..22d8bd8a7 100644 --- a/Eigen/src/Core/Matrix.h +++ b/Eigen/src/Core/Matrix.h @@ -40,7 +40,10 @@ * \param _Cols Number of columns, or \b Dynamic * * The remaining template parameters are optional -- in most cases you don't have to worry about them. - * \param _StorageOrder Either \b RowMajor or \b ColMajor. The default is \b ColMajor. + * \param _Options A combination of either \b Matrix_RowMajor or \b Matrix_ColMajor, and of either + * \b Matrix_AutoAlign or \b Matrix_DontAlign. + * The former controls storage order, and defaults to column-major. The latter controls alignment, which is required + * for vectorization. It defaults to aligning matrices except for fixed sizes that aren't a multiple of the packet size. * \param _MaxRows Maximum number of rows. Defaults to \a _Rows (\ref maxrows "note"). * \param _MaxCols Maximum number of columns. Defaults to \a _Cols (\ref maxrows "note"). * @@ -85,7 +88,7 @@ * to 16x16. Larger matrices should be declared as dynamic-size even if one happens to know their size at compile-time. * * Dynamic-size means that the numbers of rows or columns are not necessarily known at compile-time. In this case they are runtime - * variables, and the array of coefficients is allocated dynamically, typically on the heap (\ref alloca "note"). + * variables, and the array of coefficients is allocated dynamically on the heap. * * Note that \em dense matrices, be they Fixed-size or Dynamic-size, do not expand dynamically in the sense of a std::map. * If you want this behavior, see the Sparse module. @@ -97,15 +100,10 @@ * exceed a certain value. This happens when taking dynamic-size blocks inside fixed-size matrices: in this case _MaxRows and _MaxCols * are the dimensions of the original matrix, while _Rows and _Cols are Dynamic. * - *
\anchor alloca Usage of alloca():
- *
On the Linux platform, for small enough arrays, Eigen will avoid heap allocation and instead will use alloca() to perform a dynamic - * allocation on the stack.
- * - * * \see MatrixBase for the majority of the API methods for matrices */ -template -struct ei_traits > +template +struct ei_traits > { typedef _Scalar Scalar; enum { @@ -113,27 +111,34 @@ struct ei_traits::ret, + Flags = ei_compute_matrix_flags<_Scalar, _Rows, _Cols, _Options, _MaxRows, _MaxCols>::ret, CoeffReadCost = NumTraits::ReadCost, SupportedAccessPatterns = RandomAccessPattern }; }; -template +template +struct ei_matrix_with_aligned_operator_new : WithAlignedOperatorNew {}; + +template +struct ei_matrix_with_aligned_operator_new {}; + +template class Matrix - : public MatrixBase > - , public ei_with_aligned_operator_new<_Scalar,ei_size_at_compile_time<_Rows,_Cols>::ret> + : public MatrixBase > + , public ei_matrix_with_aligned_operator_new<_Options> { public: EIGEN_GENERIC_PUBLIC_INTERFACE(Matrix) - enum { StorageOrder = _StorageOrder }; + enum { Options = _Options }; friend class Eigen::Map; typedef class Eigen::Map UnalignedMapType; friend class Eigen::Map; typedef class Eigen::Map AlignedMapType; protected: - ei_matrix_storage m_storage; + ei_matrix_storage m_storage; public: diff --git a/Eigen/src/Core/MatrixStorage.h b/Eigen/src/Core/MatrixStorage.h index 9027acee7..7d869579b 100644 --- a/Eigen/src/Core/MatrixStorage.h +++ b/Eigen/src/Core/MatrixStorage.h @@ -26,6 +26,27 @@ #ifndef EIGEN_MATRIXSTORAGE_H #define EIGEN_MATRIXSTORAGE_H +/** \internal + * Static array automatically aligned if the total byte size is a multiple of 16 and the matrix options require auto alignment + */ +template struct ei_matrix_array +{ + EIGEN_ALIGN_128 T array[Size]; + + ei_matrix_array() + { + ei_assert((reinterpret_cast(array) & 0xf) == 0 + && "this assertion is explained here: http://eigen.tuxfamily.org/api/UnalignedArrayAssert.html **** READ THIS WEB PAGE !!! ****"); + } +}; + +template struct ei_matrix_array +{ + T array[Size]; +}; + /** \internal * * \class ei_matrix_storage @@ -37,12 +58,12 @@ * * \sa Matrix */ -template class ei_matrix_storage; +template class ei_matrix_storage; // purely fixed-size matrix -template class ei_matrix_storage +template class ei_matrix_storage { - ei_aligned_array m_data; + ei_matrix_array m_data; public: inline explicit ei_matrix_storage() {} inline ei_matrix_storage(int,int,int) {} @@ -55,9 +76,9 @@ template class ei_matrix_storage }; // dynamic-size matrix with fixed-size storage -template class ei_matrix_storage +template class ei_matrix_storage { - ei_aligned_array m_data; + ei_matrix_array m_data; int m_rows; int m_cols; public: @@ -78,9 +99,9 @@ template class ei_matrix_storage class ei_matrix_storage +template class ei_matrix_storage { - ei_aligned_array m_data; + ei_matrix_array m_data; int m_rows; public: inline explicit ei_matrix_storage() : m_rows(0) {} @@ -98,9 +119,9 @@ template class ei_matrix_storage class ei_matrix_storage +template class ei_matrix_storage { - ei_aligned_array m_data; + ei_matrix_array m_data; int m_cols; public: inline explicit ei_matrix_storage() : m_cols(0) {} @@ -118,7 +139,7 @@ template class ei_matrix_storage class ei_matrix_storage +template class ei_matrix_storage { T *m_data; int m_rows; @@ -147,7 +168,7 @@ template class ei_matrix_storage }; // matrix with dynamic width and fixed height (so that matrix has dynamic size). -template class ei_matrix_storage +template class ei_matrix_storage { T *m_data; int m_cols; @@ -172,7 +193,7 @@ template class ei_matrix_storage class ei_matrix_storage +template class ei_matrix_storage { T *m_data; int m_rows; diff --git a/Eigen/src/Core/util/Constants.h b/Eigen/src/Core/util/Constants.h index 203f3b294..a535e2417 100644 --- a/Eigen/src/Core/util/Constants.h +++ b/Eigen/src/Core/util/Constants.h @@ -223,8 +223,15 @@ enum { }; enum { - ColMajor = 0, - RowMajor = RowMajorBit + Matrix_ColMajor = 0, + Matrix_RowMajor = 0x1, // it is only a coincidence that this is equal to RowMajorBit -- don't rely on that + /** \internal Don't require alignment for the matrix itself (the array of coefficients, if dynamically allocated, may still be + requested to be aligned) */ + ColMajor = Matrix_ColMajor, // deprecated + RowMajor = Matrix_RowMajor, // deprecated + Matrix_DontAlign = 0, + /** \internal Align the matrix itself */ + Matrix_AutoAlign = 0x2 }; enum { diff --git a/Eigen/src/Core/util/ForwardDeclarations.h b/Eigen/src/Core/util/ForwardDeclarations.h index b61316dfc..bb4567ad0 100644 --- a/Eigen/src/Core/util/ForwardDeclarations.h +++ b/Eigen/src/Core/util/ForwardDeclarations.h @@ -28,7 +28,8 @@ template struct ei_traits; template struct NumTraits; -template class Matrix; template class Flagged; diff --git a/Eigen/src/Core/util/Macros.h b/Eigen/src/Core/util/Macros.h index 422bc7bd1..3160ce7f6 100644 --- a/Eigen/src/Core/util/Macros.h +++ b/Eigen/src/Core/util/Macros.h @@ -28,6 +28,12 @@ #undef minor +#ifdef EIGEN_DEFAULT_TO_ROW_MAJOR +#define EIGEN_DEFAULT_MATRIX_STORAGE_ORDER_OPTION Matrix_RowMajor +#else +#define EIGEN_DEFAULT_MATRIX_STORAGE_ORDER_OPTION Matrix_ColMajor +#endif + /** \internal Defines the maximal loop size to enable meta unrolling of loops. * Note that the value here is expressed in Eigen's own notion of "number of FLOPS", * it does not correspond to the number of iterations or the number of instructions diff --git a/Eigen/src/Core/util/Memory.h b/Eigen/src/Core/util/Memory.h index 35b6b4ab9..0e49ffeae 100644 --- a/Eigen/src/Core/util/Memory.h +++ b/Eigen/src/Core/util/Memory.h @@ -31,25 +31,6 @@ extern "C" int posix_memalign (void **, size_t, size_t) throw (); #endif -/** \internal - * Static array automatically aligned if the total byte size is a multiple of 16 - */ -template struct ei_aligned_array -{ - EIGEN_ALIGN_128 T array[Size]; - - ei_aligned_array() - { - ei_assert((reinterpret_cast(array) & 0xf) == 0 - && "this assertion is explained here: http://eigen.tuxfamily.org/api/UnalignedArrayAssert.html **** READ THIS WEB PAGE !!! ****"); - } -}; - -template struct ei_aligned_array -{ - T array[Size]; -}; - struct ei_byte_forcing_aligned_malloc { unsigned char c; // sizeof must be 1. diff --git a/Eigen/src/Core/util/XprHelper.h b/Eigen/src/Core/util/XprHelper.h index ae8703958..739eb1108 100644 --- a/Eigen/src/Core/util/XprHelper.h +++ b/Eigen/src/Core/util/XprHelper.h @@ -85,22 +85,16 @@ template struct ei_unpacket_traits enum {size=1}; }; - -template +template class ei_compute_matrix_flags { enum { - row_major_bit = (Rows != 1 && Cols != 1) // if this is not a vector, - // then the storage order really matters, - // so let us strictly honor the user's choice. - ? StorageOrder - : Cols > 1 ? RowMajorBit : 0, + row_major_bit = Options&Matrix_RowMajor ? RowMajorBit : 0, inner_max_size = row_major_bit ? MaxCols : MaxRows, is_big = inner_max_size == Dynamic, - is_packet_size_multiple = (Cols * Rows)%ei_packet_traits::size==0, - packet_access_bit = ei_packet_traits::size > 1 - && (is_big || is_packet_size_multiple) ? PacketAccessBit : 0, - aligned_bit = packet_access_bit && (is_big || is_packet_size_multiple) ? AlignedBit : 0 + is_packet_size_multiple = (Cols*Rows) % ei_packet_traits::size == 0, + aligned_bit = ((Options&Matrix_AutoAlign) && (is_big || is_packet_size_multiple)) ? AlignedBit : 0, + packet_access_bit = ei_packet_traits::size > 1 && aligned_bit ? PacketAccessBit : 0 }; public: @@ -123,7 +117,7 @@ template struct ei_eval typedef Matrix::Scalar, ei_traits::RowsAtCompileTime, ei_traits::ColsAtCompileTime, - ei_traits::Flags&RowMajorBit ? RowMajor : ColMajor, + Matrix_AutoAlign | (ei_traits::Flags&RowMajorBit ? Matrix_RowMajor : Matrix_ColMajor), ei_traits::MaxRowsAtCompileTime, ei_traits::MaxColsAtCompileTime > type; @@ -144,7 +138,7 @@ template struct ei_plain_matrix_type typedef Matrix::Scalar, ei_traits::RowsAtCompileTime, ei_traits::ColsAtCompileTime, - ei_traits::Flags&RowMajorBit ? RowMajor : ColMajor, + Matrix_AutoAlign | (ei_traits::Flags&RowMajorBit ? Matrix_RowMajor : Matrix_ColMajor), ei_traits::MaxRowsAtCompileTime, ei_traits::MaxColsAtCompileTime > type; @@ -157,7 +151,7 @@ template struct ei_plain_matrix_type_column_major typedef Matrix::Scalar, ei_traits::RowsAtCompileTime, ei_traits::ColsAtCompileTime, - ColMajor, + Matrix_AutoAlign | Matrix_ColMajor, ei_traits::MaxRowsAtCompileTime, ei_traits::MaxColsAtCompileTime > type; diff --git a/Eigen/src/LU/LU.h b/Eigen/src/LU/LU.h index 1ac14034f..82b408d3c 100644 --- a/Eigen/src/LU/LU.h +++ b/Eigen/src/LU/LU.h @@ -76,7 +76,7 @@ template class LU MatrixType::ColsAtCompileTime, // the number of rows in the "kernel matrix" is the number of cols of the original matrix // so that the product "matrix * kernel = zero" makes sense Dynamic, // we don't know at compile-time the dimension of the kernel - MatrixType::StorageOrder, + MatrixType::Options, MatrixType::MaxColsAtCompileTime, // see explanation for 2nd template parameter MatrixType::MaxColsAtCompileTime // the kernel is a subspace of the domain space, whose dimension is the number // of columns of the original matrix @@ -86,7 +86,7 @@ template class LU MatrixType::RowsAtCompileTime, // the image is a subspace of the destination space, whose dimension is the number // of rows of the original matrix Dynamic, // we don't know at compile time the dimension of the image (the rank) - MatrixType::StorageOrder, + MatrixType::Options, MatrixType::MaxRowsAtCompileTime, // the image matrix will consist of columns from the original matrix, MatrixType::MaxColsAtCompileTime // so it has the same number of rows and at most as many columns. > ImageResultType; @@ -436,7 +436,7 @@ void LU::computeKernel(KernelMatrixType *result) const * independent vectors in Ker U. */ - Matrix y(-m_lu.corner(TopRight, m_rank, dimker)); @@ -504,7 +504,7 @@ bool LU::solve( // Step 2 Matrix l(rows, rows); l.setZero(); @@ -523,7 +523,7 @@ bool LU::solve( return false; } Matrix d(c.corner(TopLeft, m_rank, c.cols())); m_lu.corner(TopLeft, m_rank, m_rank) diff --git a/bench/btl/cmake/FindEigen2.cmake b/bench/btl/cmake/FindEigen2.cmake index 49dfd1e6d..2579a6c9c 100644 --- a/bench/btl/cmake/FindEigen2.cmake +++ b/bench/btl/cmake/FindEigen2.cmake @@ -1,11 +1,9 @@ -# - Try to find eigen2 headers +# - Try to find Eigen2 lib # Once done this will define # -# EIGEN2_FOUND - system has eigen2 lib -# EIGEN2_INCLUDE_DIR - the eigen2 include directory -# -# Copyright (C) 2008 Gael Guennebaud -# Adapted from FindEigen.cmake: +# EIGEN2_FOUND - system has eigen lib +# EIGEN2_INCLUDE_DIR - the eigen include directory + # Copyright (c) 2006, 2007 Montel Laurent, # Redistribution and use is allowed according to the terms of the BSD license. # For details see the accompanying COPYING-CMAKE-SCRIPTS file. @@ -19,12 +17,14 @@ else (EIGEN2_INCLUDE_DIR) find_path(EIGEN2_INCLUDE_DIR NAMES Eigen/Core PATHS - ${Eigen_SOURCE_DIR}/ ${INCLUDE_INSTALL_DIR} + ${KDE4_INCLUDE_DIR} + PATH_SUFFIXES eigen2 ) include(FindPackageHandleStandardArgs) -find_package_handle_standard_args(Eigen2 DEFAULT_MSG EIGEN2_INCLUDE_DIR) +find_package_handle_standard_args(Eigen2 DEFAULT_MSG EIGEN2_INCLUDE_DIR ) + mark_as_advanced(EIGEN2_INCLUDE_DIR) diff --git a/test/determinant.cpp b/test/determinant.cpp index 68e91e4c1..bc647d25d 100644 --- a/test/determinant.cpp +++ b/test/determinant.cpp @@ -38,8 +38,8 @@ template void determinant(const MatrixType& m) m2.setRandom(); typedef typename MatrixType::Scalar Scalar; Scalar x = ei_random(); - VERIFY(ei_isApprox(MatrixType::Identity(size, size).determinant(), Scalar(1))); - VERIFY(ei_isApprox((m1*m2).determinant(), m1.determinant() * m2.determinant())); + VERIFY_IS_APPROX(MatrixType::Identity(size, size).determinant(), Scalar(1)); + VERIFY_IS_APPROX((m1*m2).determinant(), m1.determinant() * m2.determinant()); if(size==1) return; int i = ei_random(0, size-1); int j; @@ -48,18 +48,18 @@ template void determinant(const MatrixType& m) } while(j==i); m2 = m1; m2.row(i).swap(m2.row(j)); - VERIFY(ei_isApprox(m2.determinant(), -m1.determinant())); + VERIFY_IS_APPROX(m2.determinant(), -m1.determinant()); m2 = m1; m2.col(i).swap(m2.col(j)); - VERIFY(ei_isApprox(m2.determinant(), -m1.determinant())); - VERIFY(ei_isApprox(m2.determinant(), m2.transpose().determinant())); - VERIFY(ei_isApprox(ei_conj(m2.determinant()), m2.adjoint().determinant())); + VERIFY_IS_APPROX(m2.determinant(), -m1.determinant()); + VERIFY_IS_APPROX(m2.determinant(), m2.transpose().determinant()); + VERIFY_IS_APPROX(ei_conj(m2.determinant()), m2.adjoint().determinant()); m2 = m1; m2.row(i) += x*m2.row(j); - VERIFY(ei_isApprox(m2.determinant(), m1.determinant())); + VERIFY_IS_APPROX(m2.determinant(), m1.determinant()); m2 = m1; m2.row(i) *= x; - VERIFY(ei_isApprox(m2.determinant(), m1.determinant() * x)); + VERIFY_IS_APPROX(m2.determinant(), m1.determinant() * x); } void test_determinant() @@ -72,4 +72,5 @@ void test_determinant() CALL_SUBTEST( determinant(Matrix, 10, 10>()) ); CALL_SUBTEST( determinant(MatrixXd(20, 20)) ); } + CALL_SUBTEST( determinant(MatrixXd(200, 200)) ); } diff --git a/test/unalignedassert.cpp b/test/unalignedassert.cpp index 4f9495a51..0b5bf0c77 100644 --- a/test/unalignedassert.cpp +++ b/test/unalignedassert.cpp @@ -67,6 +67,12 @@ struct Good8 : Eigen::WithAlignedOperatorNew Matrix4f m; }; +struct Good9 +{ + Matrix m; // good: no alignment requested + float f; +}; + template void check_unalignedassert_good() { @@ -80,7 +86,7 @@ void check_unalignedassert_good() template void check_unalignedassert_bad() { - float buf[1000]; + float buf[sizeof(T)+16]; float *unaligned = buf; while((reinterpret_cast(unaligned)&0xf)==0) ++unaligned; // make sure unaligned is really unaligned T *x = new(static_cast(unaligned)) T; @@ -97,6 +103,7 @@ void unalignedassert() VERIFY_RAISES_ASSERT(check_unalignedassert_bad()); check_unalignedassert_good(); check_unalignedassert_good(); + check_unalignedassert_good(); } void test_unalignedassert()