Compare commits

..

75 Commits

Author SHA1 Message Date
Gael Guennebaud
a65053d80b bump to 3.0.2 2011-08-26 14:56:26 +02:00
root
adcb220db3 fix linking issue with msvc 2011-08-26 15:22:48 +02:00
Gael Guennebaud
b21f9c3573 fix bug #330: Index to int conversion warning
(transplanted from 8414be739b
)
2011-08-23 11:02:10 +02:00
Gael Guennebaud
fe228fc50b mv the mpreal copy in its own folder
(transplanted from ea4a1960f0
)
2011-08-19 15:08:29 +02:00
Gael Guennebaud
4ab20b4cae update to latest mpreal and fix a min/max issue in mprel.h
(transplanted from 79ad55a901
)
2011-08-19 15:03:45 +02:00
Gael Guennebaud
5d5cf478ab oops EIGEN_DEFINE_STL_VECTOR_SPECIALIZATION now perfroms full specialization,
no need for the typename keywords
(transplanted from b3f5fbbd9a
)
2011-08-22 10:48:04 +02:00
Gael Guennebaud
55149df4e8 fix bug #262: Compilation error of stdvector_overload test with GCC 4.6
Now our aligned allocator is automatically activatived only when the user
did not specified an allocator (or specified the default std::allocator).
(transplanted from b85c89c313
)
2011-08-22 10:12:10 +02:00
Gael Guennebaud
b2d10249b4 fix linking issue
(transplanted from ca7d3dca79
)
2011-08-12 22:38:53 +02:00
Thomas Capricelli
bdf0b0c47e fix a bug where some rotations were not initialized
They actually were in the original minpack code, this is a bug introduced
by our migration.
Reported on #322 and
http://forum.kde.org/viewtopic.php?f=74&t=96197#p201158
2011-08-04 05:02:47 +02:00
Thomas Capricelli
ea7923c6f9 wa2 was computed twice because of a confustion between changesets
746c787a76
 and ee0e39284c
.
Reported on forum:
http://forum.kde.org/viewtopic.php?f=74&t=96197#p201158
2011-08-04 03:25:29 +02:00
Gael Guennebaud
49b6e9143e protect calls to min and max with parentheses to make Eigen compatible with default windows.h 2011-07-21 11:19:36 +02:00
Gael Guennebaud
f096553344 fix bug #320 (pretty gdb printer on mingw)
(transplanted from d4bd8bddb5
)
2011-07-20 11:15:42 +02:00
Gael Guennebaud
433b353013 fix bug #316 - SelfAdjointEigenSolver::compute does not handle matrices of size (1,1) correctly
(transplanted from 5fdebc2fa5
)
2011-07-09 07:15:14 +02:00
Thomas Capricelli
3cb088c39f fix few warnings reported by clang 2011-07-07 22:19:43 +02:00
Gael Guennebaud
a99ea69b32 fix constness of intersection methods (bug #309)
(transplanted from c98cd5e564
)
2011-06-27 13:15:01 +02:00
Thomas Capricelli
d03bbcbcbc fix typo in doc for ParametrizedLine 2011-06-23 00:34:30 +02:00
Tim Holy
fae2aa3fd9 Relatively straightforward changes to wording of documentation, focusing particularly on the sparse and (to a lesser extent) geometry pages.
(transplanted from 16a2d896bc
)
2011-06-20 22:47:58 -05:00
Tim Holy
13a17d968f A first tiny test commit: fix a spelling error in the documentation.
(transplanted from 4a95badf74
)
2011-06-19 14:39:19 -05:00
Gael Guennebaud
135ba535a4 fix documentation of norm
(transplanted from a55c27a15f
)
2011-06-18 08:30:34 +02:00
Gael Guennebaud
bbbf0559fe remove the use of non standard long long
(transplanted from 40287d2fd9
)
2011-06-14 10:56:47 +02:00
Gael Guennebaud
c91fed1eec fix aligned_allocator::allocate interface
(transplanted from f82b3ea241
)
2011-06-14 08:50:25 +02:00
Thomas Capricelli
f59b08f3bd fix typo in constant name 2011-06-12 23:53:46 +02:00
Gael Guennebaud
9155002901 fix compilation with MinGW
(transplanted from 5bc4abc45e
)
2011-06-01 12:16:21 +02:00
Gael Guennebaud
46f4bd9ed4 fix aligned_stack_memory_handler for null pointers
(transplanted from 6441e8727b
)
2011-04-21 09:00:55 +02:00
Gael Guennebaud
ebad34db21 Added tag 3.0.1 for changeset c0f867ed10 2011-05-30 15:23:33 +02:00
Gael Guennebaud
c0f867ed10 bump to 3.0.1 2011-05-30 15:15:37 +02:00
Gael Guennebaud
d225bbe534 do not directly call std::ceil
(transplanted from 9464745385
)
2011-05-28 16:46:38 +02:00
Jitse Niesen
a6f8da7c48 Fix typo ('using namespace' instead of 'using').
(transplanted from d23845c4cc
)
2011-05-26 09:52:36 +01:00
Gael Guennebaud
33efb8ed62 Simplify the use of custom scalar types, the rule is to never directly call a standard math function using std:: but rather put a using std::foo before and simply call foo:
using std::max;
max(a,b);
(transplanted from 87ac09daa8
)
2011-05-25 08:41:45 +02:00
Gael Guennebaud
63e5cf525f work around an ICE with ICC 12 2011-05-29 11:23:31 +02:00
Gael Guennebaud
3cd1641dac fix bug #278: geometry tutorial 2011-05-28 22:12:15 +02:00
Gael Guennebaud
4fe4ab8fc0 finish to fix bug #270: we have to use EIGEN_ALIGN_STATICALLY and not EIGEN_DONT_ALIGN_STATICALLY...
(transplanted from 7b46d7ed0f
)
2011-05-28 11:38:53 +02:00
Gael Guennebaud
d7d76bf4ca bug #225: add a unit test for memory leak
(transplanted from 5541bcb769
)
2011-05-23 14:20:49 +02:00
Gael Guennebaud
cf76a50a34 bug #271: fix copy/paste mistakes in doc 2011-05-23 13:39:26 +02:00
Gael Guennebaud
ee46ae9ba7 clean a bit previous patch (ctor vs static_cast and a few bits)
(transplanted from da644fb0c3e0b7fcda03ba27a02061c084809b9f)
2011-05-23 13:34:04 +02:00
David H. Bailey
b3c3627c72 fix implicit scalar conversions (needed to support fancy scalar types, see bug #276)
(transplanted from d61f1eae804a5dc4924f167c00fbde31c1bef7ea)
2011-05-23 11:20:13 +02:00
Gael Guennebaud
e3a521be6b backport 7209d6a126
(fix gemv_static_vector_if on architectures that cannot aligned on the stack (e.g., ARM NEON))
2011-05-21 22:19:12 +02:00
Gael Guennebaud
4c7d57490c clean several other assertion checking tests
(transplanted from 96464f8563
)
2011-05-20 09:59:15 +02:00
Gael Guennebaud
fe21e084b4 fix vectorization_logic when EIGEN_GCC_AND_ARCH_DOESNT_WANT_STACK_ALIGNMENT
(transplanted from 501bc602ec
)
2011-05-19 21:52:40 +02:00
Gael Guennebaud
282fd7a2da NEON: fix plset
(transplanted from f2837aebc4
)
2011-05-18 21:12:08 +02:00
Gael Guennebaud
7d28c618a0 add unit test for plset
(transplanted from 8170ef0b2d
)
2011-05-18 21:11:03 +02:00
Gael Guennebaud
f07fca2c80 NEON: disable unaligned assertion checking for non vectorized types
(transplanted from 7f2a88c91f
)
2011-05-18 14:11:40 +02:00
Gael Guennebaud
99ab2411e5 NEON: fix ploaddup
(transplanted from 85c137ccd4
)
2011-05-18 08:15:47 +02:00
Gael Guennebaud
ffefe1bd2e fix trmm for some unusual trapezoidal cases (a dense set of columns or rows is zero)
(transplanted from 568478ffe5
)
2011-03-28 17:41:46 +02:00
Gael Guennebaud
55574053d0 fix bug #267: alloca is not aligned on arm
(transplanted from 179d42bb2b
)
2011-05-17 21:30:12 +02:00
Gael Guennebaud
ffee1d1c87 fix 228 (ei_aligned_stack_delete does not exist anymore)
(transplanted from 5fda8cdfb3
)
2011-03-21 21:59:42 +01:00
Gael Guennebaud
adf5992767 port sparse LLT/LDLT to new stack allocation API
(transplanted from 535a61ede8
)
2011-03-20 17:10:43 +01:00
Gael Guennebaud
19e7c672bb clean a bit the stack allocation mechanism
(transplanted from b8ecda5c66
)
2011-03-19 10:27:47 +01:00
Gael Guennebaud
99a6178e6a test the new stack allocation mechanism
(transplanted from bbb4b35dfc
)
2011-03-19 08:51:38 +01:00
Gael Guennebaud
c3342b0bb4 fix memory leak when a custom scalar throw an exception
(transplanted from 290205dfc0
)
2011-03-19 01:06:50 +01:00
John Tytgat
84c8b6d5c5 fix bug #260: broken Qt support for Transform 2011-05-11 22:31:36 +02:00
Jitse Niesen
18a8034348 Get rid of wrong "subscript above bounds" warning (bug #149). 2011-05-07 18:44:11 +01:00
Gael Guennebaud
697e1656ce add missing .data() members to MatrixWrapper and ArrayWrapper
(transplanted from fb76452cbc
)
2011-05-06 21:15:05 +02:00
Gael Guennebaud
c2a23c3e24 fix compilation on ARM NEON (missing AlignedOnScalar)
(transplanted from 97b6d26f5b
)
2011-05-06 09:03:48 +02:00
Thomas Capricelli
6d0e3154d7 better fix for gcc 4.6.0 / ptrdiff_t, as suggested by Benoit 2011-05-05 18:48:40 +02:00
Thomas Capricelli
7b122ed158 backport of a18a1be42d
Fix compilation with gcc-4.6.0, patch provided by Anton Gladky <gladky.anton@gmail.com>,
working on debian packaging.
2011-05-05 00:48:13 +02:00
Jitse Niesen
d9232a96aa Bail out if preprocessor symbol Success is defined (bug #253). 2011-05-04 14:28:01 +01:00
Jitse Niesen
4ecf67f5e4 Backport of a96c849c20
: Document enums in Contants.h (bug #248).
2011-05-03 17:18:10 +01:00
Gael Guennebaud
860d66c0f1 fix bug #258: asin/acos copy paste mistake
(transplanted from 1947da39ab
)
2011-05-02 13:26:44 +02:00
Mathieu Gautier
ba3aafa85f Quaternion : add Flags on Quaternion's traits with the LvalueBit set if needed
Quaternion : change PacketAccess to IsAligned to mimic other traits
test : add a test and 4 failtest on Map<const Quaternion> based on Eigen::Map ones
(transplanted from 2b5868ee7e71398e35d495d447b02e0be54f53da)
2011-04-12 14:49:50 +02:00
Thomas Capricelli
b478521ecd eigen_gen_docs : be nice with the server : dont use -j3 2011-04-19 17:41:23 +02:00
Thomas Capricelli
e8fa6dde01 adapt eigen_gen_docs for the 3.0 branch. Also, create the 'build' dir if
not present.
2011-04-19 17:36:56 +02:00
Gael Guennebaud
134b83c310 fix bug #250: compilation error with gcc 4.6 (STL header files no longer include cstddef)
(transplanted from e87f653924
)
2011-04-19 16:34:25 +02:00
Gael Guennebaud
b0e810fb3f fix bug #242: vectorization was wrongly enabled on MSVC 2005
(transplanted from 67d50f539b
)
2011-04-19 15:25:00 +02:00
Eamon Nerbonne
dee686f762 WIN32 isn't defined ?? but _WIN32 is. 2011-04-19 14:37:04 +02:00
Jitse Niesen
90cacfa610 Make MapBase(PointerType) constructor explicit (fixes bug #251).
Backport of changeset 0b40b36d10
.
2011-04-19 12:56:41 +01:00
Benoit Jacob
de21678aab fix unaligned-array-assert link 2011-04-18 06:35:54 -04:00
Jitse Niesen
a700d3c506 Backport of c9b5531d6c
: Normalize eigenvectors (bug #249).
2011-04-15 17:41:12 +01:00
Jitse Niesen
fc4684fe97 Backport of 70d5837e00
: Correct typo in QuickReference doc.
2011-04-01 16:59:45 +01:00
Adam Szalkowski
c088ee78c8 fix bug #239: the essential part was left uninitialized in some cases
(transplanted from 969e92261d
)
2011-03-31 09:54:52 +02:00
Jitse Niesen
e53539435d Backport of changeset c6ad2deead
. Fixes bug #232.
2011-03-24 10:45:24 +00:00
Benoit Jacob
1e8b834ceb fix typos 2011-03-21 06:45:57 -04:00
Benoit Jacob
3c510db6bf Added tag 3.0.0 for changeset 72ffb63165 2011-03-19 11:43:21 -04:00
Gael Guennebaud
72ffb63165 fix compilation for old but not so old versions of glew 2011-03-18 10:26:21 +01:00
Benoit Jacob
67e24b85a4 bump 2011-03-18 05:13:34 -04:00
342 changed files with 10905 additions and 17940 deletions

11
.hgeol
View File

@@ -1,8 +1,3 @@
[patterns]
scripts/*.in = LF
debug/msvc/*.dat = CRLF
unsupported/test/mpreal/*.* = CRLF
** = native
[repository]
native = LF
[patterns]
**.* = native
eigen_autoexp_part.dat = CRLF

View File

@@ -103,8 +103,6 @@ endif()
add_definitions("-DEIGEN_PERMANENTLY_DISABLE_STUPID_WARNINGS")
set(EIGEN_TEST_MAX_SIZE "320" CACHE STRING "Maximal matrix/vector size, default is 320")
if(CMAKE_COMPILER_IS_GNUCXX)
set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -Wnon-virtual-dtor -Wno-long-long -ansi -Wundef -Wcast-align -Wchar-subscripts -Wall -W -Wpointer-arith -Wwrite-strings -Wformat-security -fexceptions -fno-check-new -fno-common -fstrict-aliasing")
set(CMAKE_CXX_FLAGS_DEBUG "-g3")
@@ -281,21 +279,9 @@ install(FILES
)
if(EIGEN_BUILD_PKGCONFIG)
SET(path_separator ":")
STRING(REPLACE ${path_separator} ";" pkg_config_libdir_search "$ENV{PKG_CONFIG_LIBDIR}")
message(STATUS "searching for 'pkgconfig' directory in PKG_CONFIG_LIBDIR ( $ENV{PKG_CONFIG_LIBDIR} ), ${CMAKE_INSTALL_PREFIX}/share, and ${CMAKE_INSTALL_PREFIX}/lib")
FIND_PATH(pkg_config_libdir pkgconfig ${pkg_config_libdir_search} ${CMAKE_INSTALL_PREFIX}/share ${CMAKE_INSTALL_PREFIX}/lib ${pkg_config_libdir_search})
if(pkg_config_libdir)
SET(pkg_config_install_dir ${pkg_config_libdir})
message(STATUS "found ${pkg_config_libdir}/pkgconfig" )
else(pkg_config_libdir)
SET(pkg_config_install_dir ${CMAKE_INSTALL_PREFIX}/share)
message(STATUS "pkgconfig not found; installing in ${pkg_config_install_dir}" )
endif(pkg_config_libdir)
configure_file(eigen3.pc.in eigen3.pc)
install(FILES ${CMAKE_CURRENT_BINARY_DIR}/eigen3.pc
DESTINATION ${pkg_config_install_dir}/pkgconfig
DESTINATION share/pkgconfig
)
endif(EIGEN_BUILD_PKGCONFIG)
@@ -303,9 +289,44 @@ add_subdirectory(Eigen)
add_subdirectory(doc EXCLUDE_FROM_ALL)
include(EigenConfigureTesting)
# fixme, not sure this line is still needed:
add_custom_target(buildtests)
add_custom_target(check COMMAND "ctest")
add_dependencies(check buildtests)
# CMake/Ctest does not allow us to change the build command,
# so we have to workaround by directly editing the generated DartConfiguration.tcl file
# save CMAKE_MAKE_PROGRAM
set(CMAKE_MAKE_PROGRAM_SAVE ${CMAKE_MAKE_PROGRAM})
# and set a fake one
set(CMAKE_MAKE_PROGRAM "@EIGEN_MAKECOMMAND_PLACEHOLDER@")
include(CTest)
enable_testing() # must be called from the root CMakeLists, see man page
include(EigenTesting)
ei_init_testing()
# overwrite default DartConfiguration.tcl
# The worarounds are different for each version of the MSVC IDE
if(MSVC_IDE)
if(MSVC_VERSION EQUAL 1600) # MSVC 2010
set(EIGEN_MAKECOMMAND_PLACEHOLDER "${CMAKE_MAKE_PROGRAM_SAVE} buildtests.vcxproj /p:Configuration=\${CTEST_CONFIGURATION_TYPE} \n # ")
else() # MSVC 2008 (TODO check MSVC 2005)
set(EIGEN_MAKECOMMAND_PLACEHOLDER "${CMAKE_MAKE_PROGRAM_SAVE} /project buildtests")
endif()
else()
# for make and nmake
set(EIGEN_MAKECOMMAND_PLACEHOLDER "${CMAKE_MAKE_PROGRAM_SAVE} buildtests")
endif()
configure_file(${CMAKE_BINARY_DIR}/DartConfiguration.tcl ${CMAKE_BINARY_DIR}/DartConfiguration.tcl)
# restore default CMAKE_MAKE_PROGRAM
set(CMAKE_MAKE_PROGRAM ${CMAKE_MAKE_PROGRAM_SAVE})
# un-set temporary variables so that it is like they never existed.
# CMake 2.6.3 introduces the more logical unset() syntax for this.
set(CMAKE_MAKE_PROGRAM_SAVE)
set(EIGEN_MAKECOMMAND_PLACEHOLDER)
configure_file(${CMAKE_SOURCE_DIR}/CTestCustom.cmake.in ${CMAKE_BINARY_DIR}/CTestCustom.cmake)
if(EIGEN_LEAVE_TEST_IN_ALL_TARGET)
@@ -314,13 +335,15 @@ else()
add_subdirectory(test EXCLUDE_FROM_ALL)
endif()
if(EIGEN_LEAVE_TEST_IN_ALL_TARGET)
add_subdirectory(blas)
add_subdirectory(lapack)
else()
add_subdirectory(blas EXCLUDE_FROM_ALL)
add_subdirectory(lapack EXCLUDE_FROM_ALL)
endif()
if(NOT MSVC)
if(EIGEN_LEAVE_TEST_IN_ALL_TARGET)
add_subdirectory(blas)
add_subdirectory(lapack)
else()
add_subdirectory(blas EXCLUDE_FROM_ALL)
add_subdirectory(lapack EXCLUDE_FROM_ALL)
endif()
endif(NOT MSVC)
add_subdirectory(unsupported)

View File

@@ -1,34 +0,0 @@
#ifndef EIGEN_CHOLMODSUPPORT_MODULE_H
#define EIGEN_CHOLMODSUPPORT_MODULE_H
#include "SparseCore"
#include "src/Core/util/DisableStupidWarnings.h"
extern "C" {
#include <cholmod.h>
}
namespace Eigen {
/** \ingroup Support_modules
* \defgroup CholmodSupport_Module CholmodSupport module
*
*
* \code
* #include <Eigen/CholmodSupport>
* \endcode
*/
#include "src/misc/Solve.h"
#include "src/misc/SparseSolve.h"
#include "src/CholmodSupport/CholmodSupport.h"
} // namespace Eigen
#include "src/Core/util/ReenableStupidWarnings.h"
#endif // EIGEN_CHOLMODSUPPORT_MODULE_H

View File

@@ -167,7 +167,7 @@
#include <intrin.h>
#endif
#if defined(_CPPUNWIND) || defined(__EXCEPTIONS)
#if (defined(_CPPUNWIND) || defined(__EXCEPTIONS)) && !defined(EIGEN_NO_EXCEPTIONS)
#define EIGEN_EXCEPTIONS
#endif
@@ -175,6 +175,9 @@
#include <new>
#endif
// defined in bits/termios.h
#undef B0
/** \brief Namespace containing all symbols from the %Eigen library. */
namespace Eigen {
@@ -244,10 +247,6 @@ using std::ptrdiff_t;
* \endcode
*/
/** \defgroup Support_modules Support modules [category]
* Category of modules which add support for external libraries.
*/
#include "src/Core/util/Constants.h"
#include "src/Core/util/ForwardDeclarations.h"
#include "src/Core/util/Meta.h"
@@ -319,7 +318,7 @@ using std::ptrdiff_t;
#include "src/Core/CommaInitializer.h"
#include "src/Core/Flagged.h"
#include "src/Core/ProductBase.h"
#include "src/Core/GeneralProduct.h"
#include "src/Core/Product.h"
#include "src/Core/TriangularMatrix.h"
#include "src/Core/SelfAdjointView.h"
#include "src/Core/SolveTriangular.h"
@@ -348,12 +347,6 @@ using std::ptrdiff_t;
#include "src/Core/ArrayBase.h"
#include "src/Core/ArrayWrapper.h"
#ifdef EIGEN_ENABLE_EVALUATORS
#include "src/Core/Product.h"
#include "src/Core/CoreEvaluators.h"
#include "src/Core/AssignEvaluator.h"
#endif
} // namespace Eigen
#include "src/Core/GlobalFunctions.h"

View File

@@ -33,8 +33,7 @@
namespace Eigen {
/** \ingroup Support_modules
* \defgroup Eigen2Support_Module Eigen2 support module
/** \defgroup Eigen2Support_Module Eigen2 support module
* This module provides a couple of deprecated functions improving the compatibility with Eigen2.
*
* To use it, define EIGEN2_SUPPORT before including any Eigen header
@@ -64,24 +63,6 @@ namespace Eigen {
// Eigen2 used to include iostream
#include<iostream>
#define EIGEN_USING_MATRIX_TYPEDEFS_FOR_TYPE_AND_SIZE(TypeSuffix, SizeSuffix) \
using Eigen::Matrix##SizeSuffix##TypeSuffix; \
using Eigen::Vector##SizeSuffix##TypeSuffix; \
using Eigen::RowVector##SizeSuffix##TypeSuffix;
#define EIGEN_USING_MATRIX_TYPEDEFS_FOR_TYPE(TypeSuffix) \
EIGEN_USING_MATRIX_TYPEDEFS_FOR_TYPE_AND_SIZE(TypeSuffix, 2) \
EIGEN_USING_MATRIX_TYPEDEFS_FOR_TYPE_AND_SIZE(TypeSuffix, 3) \
EIGEN_USING_MATRIX_TYPEDEFS_FOR_TYPE_AND_SIZE(TypeSuffix, 4) \
EIGEN_USING_MATRIX_TYPEDEFS_FOR_TYPE_AND_SIZE(TypeSuffix, X) \
#define EIGEN_USING_MATRIX_TYPEDEFS \
EIGEN_USING_MATRIX_TYPEDEFS_FOR_TYPE(i) \
EIGEN_USING_MATRIX_TYPEDEFS_FOR_TYPE(f) \
EIGEN_USING_MATRIX_TYPEDEFS_FOR_TYPE(d) \
EIGEN_USING_MATRIX_TYPEDEFS_FOR_TYPE(cf) \
EIGEN_USING_MATRIX_TYPEDEFS_FOR_TYPE(cd)
#define USING_PART_OF_NAMESPACE_EIGEN \
EIGEN_USING_MATRIX_TYPEDEFS \
using Eigen::Matrix; \

View File

@@ -9,7 +9,6 @@
#include "Jacobi"
#include "Householder"
#include "LU"
#include "Geometry"
namespace Eigen {

View File

@@ -1,37 +0,0 @@
#ifndef EIGEN_ITERATIVELINEARSOLVERS_MODULE_H
#define EIGEN_ITERATIVELINEARSOLVERS_MODULE_H
#include "SparseCore"
#include "src/Core/util/DisableStupidWarnings.h"
namespace Eigen {
/** \ingroup Sparse_modules
* \defgroup IterativeLinearSolvers_Module IterativeLinearSolvers module
*
* This module currently provides iterative methods to solve problems of the form \c A \c x = \c b, where \c A is a squared matrix, usually very large and sparse.
* Those solvers are accessible via the following classes:
* - ConjugateGradient for selfadjoint (hermitian) matrices,
* - BiCGSTAB for general square matrices.
*
* Such problems can also be solved using the direct sparse decomposition modules: SparseCholesky, CholmodSupport, UmfPackSupport, SuperLUSupport.
*
* \code
* #include <Eigen/IterativeLinearSolvers>
* \endcode
*/
#include "src/misc/Solve.h"
#include "src/misc/SparseSolve.h"
#include "src/IterativeLinearSolvers/IterativeSolverBase.h"
#include "src/IterativeLinearSolvers/BasicPreconditioners.h"
#include "src/IterativeLinearSolvers/ConjugateGradient.h"
#include "src/IterativeLinearSolvers/BiCGSTAB.h"
} // namespace Eigen
#include "src/Core/util/ReenableStupidWarnings.h"
#endif // EIGEN_ITERATIVELINEARSOLVERS_MODULE_H

View File

@@ -1,27 +0,0 @@
#ifndef EIGEN_ORDERINGMETHODS_MODULE_H
#define EIGEN_ORDERINGMETHODS_MODULE_H
#include "SparseCore"
#include "src/Core/util/DisableStupidWarnings.h"
namespace Eigen {
/** \ingroup Sparse_modules
* \defgroup OrderingMethods_Module OrderingMethods module
*
* This module is currently for internal use only.
*
*
* \code
* #include <Eigen/OrderingMethods>
* \endcode
*/
#include "src/OrderingMethods/Amd.h"
} // namespace Eigen
#include "src/Core/util/ReenableStupidWarnings.h"
#endif // EIGEN_ORDERINGMETHODS_MODULE_H

View File

@@ -1,27 +1,69 @@
#ifndef EIGEN_SPARSE_MODULE_H
#define EIGEN_SPARSE_MODULE_H
#include "Core"
#include "src/Core/util/DisableStupidWarnings.h"
#include <vector>
#include <map>
#include <cstdlib>
#include <cstring>
#include <algorithm>
#ifdef EIGEN2_SUPPORT
#define EIGEN_YES_I_KNOW_SPARSE_MODULE_IS_NOT_STABLE_YET
#endif
#ifndef EIGEN_YES_I_KNOW_SPARSE_MODULE_IS_NOT_STABLE_YET
#error The sparse module API is not stable yet. To use it anyway, please define the EIGEN_YES_I_KNOW_SPARSE_MODULE_IS_NOT_STABLE_YET preprocessor token.
#endif
namespace Eigen {
/** \defgroup Sparse_modules Sparse modules
/** \defgroup Sparse_Module Sparse module
*
* Meta-module including all related modules:
* - SparseCore
* - OrderingMethods
* - SparseCholesky
* - IterativeLinearSolvers
*
*
* See the \ref TutorialSparse "Sparse tutorial"
*
* \code
* #include <Eigen/Sparse>
* \endcode
*/
/** The type used to identify a general sparse storage. */
struct Sparse {};
#include "src/Sparse/SparseUtil.h"
#include "src/Sparse/SparseMatrixBase.h"
#include "src/Sparse/CompressedStorage.h"
#include "src/Sparse/AmbiVector.h"
#include "src/Sparse/SparseMatrix.h"
#include "src/Sparse/DynamicSparseMatrix.h"
#include "src/Sparse/MappedSparseMatrix.h"
#include "src/Sparse/SparseVector.h"
#include "src/Sparse/CoreIterators.h"
#include "src/Sparse/SparseBlock.h"
#include "src/Sparse/SparseTranspose.h"
#include "src/Sparse/SparseCwiseUnaryOp.h"
#include "src/Sparse/SparseCwiseBinaryOp.h"
#include "src/Sparse/SparseDot.h"
#include "src/Sparse/SparseAssign.h"
#include "src/Sparse/SparseRedux.h"
#include "src/Sparse/SparseFuzzy.h"
#include "src/Sparse/SparseProduct.h"
#include "src/Sparse/SparseSparseProduct.h"
#include "src/Sparse/SparseDenseProduct.h"
#include "src/Sparse/SparseDiagonalProduct.h"
#include "src/Sparse/SparseTriangularView.h"
#include "src/Sparse/SparseSelfAdjointView.h"
#include "src/Sparse/TriangularSolver.h"
#include "src/Sparse/SparseView.h"
} // namespace Eigen
#include "SparseCore"
#include "OrderingMethods"
#include "SparseCholesky"
#include "IterativeLinearSolvers"
#include "src/Core/util/ReenableStupidWarnings.h"
#endif // EIGEN_SPARSE_MODULE_H

View File

@@ -1,34 +0,0 @@
#ifndef EIGEN_SPARSECHOLESKY_MODULE_H
#define EIGEN_SPARSECHOLESKY_MODULE_H
#include "SparseCore"
#include "src/Core/util/DisableStupidWarnings.h"
namespace Eigen {
/** \ingroup Sparse_modules
* \defgroup SparseCholesky_Module SparseCholesky module
*
* This module currently provides two variants of the direct sparse Cholesky decomposition for selfadjoint (hermitian) matrices.
* Those decompositions are accessible via the following classes:
* - SimplicialLLt,
* - SimplicialLDLt
*
* Such problems can also be solved using the ConjugateGradient solver from the IterativeLinearSolvers module.
*
* \code
* #include <Eigen/SparseCholesky>
* \endcode
*/
#include "src/misc/Solve.h"
#include "src/misc/SparseSolve.h"
#include "src/SparseCholesky/SimplicialCholesky.h"
} // namespace Eigen
#include "src/Core/util/ReenableStupidWarnings.h"
#endif // EIGEN_SPARSECHOLESKY_MODULE_H

View File

@@ -1,65 +0,0 @@
#ifndef EIGEN_SPARSECORE_MODULE_H
#define EIGEN_SPARSECORE_MODULE_H
#include "Core"
#include "src/Core/util/DisableStupidWarnings.h"
#include <vector>
#include <map>
#include <cstdlib>
#include <cstring>
#include <algorithm>
namespace Eigen {
/** \ingroup Sparse_modules
* \defgroup SparseCore_Module SparseCore module
*
* This module provides a sparse matrix representation, and basic associatd matrix manipulations
* and operations.
*
* See the \ref TutorialSparse "Sparse tutorial"
*
* \code
* #include <Eigen/SparseCore>
* \endcode
*
* This module depends on: Core.
*/
/** The type used to identify a general sparse storage. */
struct Sparse {};
#include "src/SparseCore/SparseUtil.h"
#include "src/SparseCore/SparseMatrixBase.h"
#include "src/SparseCore/CompressedStorage.h"
#include "src/SparseCore/AmbiVector.h"
#include "src/SparseCore/SparseMatrix.h"
#include "src/SparseCore/MappedSparseMatrix.h"
#include "src/SparseCore/SparseVector.h"
#include "src/SparseCore/CoreIterators.h"
#include "src/SparseCore/SparseBlock.h"
#include "src/SparseCore/SparseTranspose.h"
#include "src/SparseCore/SparseCwiseUnaryOp.h"
#include "src/SparseCore/SparseCwiseBinaryOp.h"
#include "src/SparseCore/SparseDot.h"
#include "src/SparseCore/SparseAssign.h"
#include "src/SparseCore/SparseRedux.h"
#include "src/SparseCore/SparseFuzzy.h"
#include "src/SparseCore/ConservativeSparseSparseProduct.h"
#include "src/SparseCore/SparseSparseProductWithPruning.h"
#include "src/SparseCore/SparseProduct.h"
#include "src/SparseCore/SparseDenseProduct.h"
#include "src/SparseCore/SparseDiagonalProduct.h"
#include "src/SparseCore/SparseTriangularView.h"
#include "src/SparseCore/SparseSelfAdjointView.h"
#include "src/SparseCore/TriangularSolver.h"
#include "src/SparseCore/SparseView.h"
} // namespace Eigen
#include "src/Core/util/ReenableStupidWarnings.h"
#endif // EIGEN_SPARSECORE_MODULE_H

View File

@@ -1,53 +0,0 @@
#ifndef EIGEN_SUPERLUSUPPORT_MODULE_H
#define EIGEN_SUPERLUSUPPORT_MODULE_H
#include "SparseCore"
#include "src/Core/util/DisableStupidWarnings.h"
#ifdef EMPTY
#define EIGEN_EMPTY_WAS_ALREADY_DEFINED
#endif
typedef int int_t;
#include <slu_Cnames.h>
#include <supermatrix.h>
#include <slu_util.h>
// slu_util.h defines a preprocessor token named EMPTY which is really polluting,
// so we remove it in favor of a SUPERLU_EMPTY token.
// If EMPTY was already, defined then we don't undef it.
#if defined(EIGEN_EMPTY_WAS_ALREADY_DEFINED)
# undef EIGEN_EMPTY_WAS_ALREADY_DEFINED
#elif defined(EMPTY)
# undef EMPTY
#endif
#define SUPERLU_EMPTY (-1)
namespace Eigen { struct SluMatrix; }
namespace Eigen {
/** \ingroup Support_modules
* \defgroup SuperLUSupport_Module SuperLUSupport module
*
* \warning When including this module, you have to use SUPERLU_EMPTY instead of EMPTY which is no longer defined because it is too polluting.
*
* \code
* #include <Eigen/SuperLUSupport>
* \endcode
*/
#include "src/misc/Solve.h"
#include "src/misc/SparseSolve.h"
#include "src/SuperLUSupport/SuperLUSupport.h"
} // namespace Eigen
#include "src/Core/util/ReenableStupidWarnings.h"
#endif // EIGEN_SUPERLUSUPPORT_MODULE_H

View File

@@ -1,34 +0,0 @@
#ifndef EIGEN_UMFPACKSUPPORT_MODULE_H
#define EIGEN_UMFPACKSUPPORT_MODULE_H
#include "SparseCore"
#include "src/Core/util/DisableStupidWarnings.h"
extern "C" {
#include <umfpack.h>
}
namespace Eigen {
/** \ingroup Support_modules
* \defgroup UmfPackSupport_Module UmfPackSupport module
*
*
*
*
* \code
* #include <Eigen/UmfPackSupport>
* \endcode
*/
#include "src/misc/Solve.h"
#include "src/misc/SparseSolve.h"
#include "src/UmfPackSupport/UmfPackSupport.h"
} // namespace Eigen
#include "src/Core/util/ReenableStupidWarnings.h"
#endif // EIGEN_UMFPACKSUPPORT_MODULE_H

View File

@@ -31,7 +31,7 @@ namespace internal {
template<typename MatrixType, int UpLo> struct LDLT_Traits;
}
/** \ingroup Cholesky_Module
/** \ingroup cholesky_Module
*
* \class LDLT
*
@@ -158,19 +158,10 @@ template<typename _MatrixType, int _UpLo> class LDLT
}
/** \returns a solution x of \f$ A x = b \f$ using the current decomposition of A.
*
* This function also supports in-place solves using the syntax <tt>x = decompositionObject.solve(x)</tt> .
*
* \note_about_checking_solutions
*
* More precisely, this method solves \f$ A x = b \f$ using the decomposition \f$ A = P^T L D L^* P \f$
* by solving the systems \f$ P^T y_1 = b \f$, \f$ L y_2 = y_1 \f$, \f$ D y_3 = y_2 \f$,
* \f$ L^* y_4 = y_3 \f$ and \f$ P x = y_4 \f$ in succession. If the matrix \f$ A \f$ is singular, then
* \f$ D \f$ will also be singular (all the other matrices are invertible). In that case, the
* least-square solution of \f$ D y_3 = y_2 \f$ is computed. This does not mean that this function
* computes the least-square solution of \f$ A x = b \f$ is \f$ A \f$ is singular.
*
* \sa MatrixBase::ldlt()
* \sa solveInPlace(), MatrixBase::ldlt()
*/
template<typename Rhs>
inline const internal::solve_retval<LDLT, Rhs>
@@ -385,21 +376,7 @@ struct solve_retval<LDLT<_MatrixType,_UpLo>, Rhs>
dec().matrixL().solveInPlace(dst);
// dst = D^-1 (L^-1 P b)
// more precisely, use pseudo-inverse of D (see bug 241)
using std::abs;
using std::max;
typedef typename LDLTType::MatrixType MatrixType;
typedef typename LDLTType::Scalar Scalar;
typedef typename LDLTType::RealScalar RealScalar;
const Diagonal<const MatrixType> vectorD = dec().vectorD();
RealScalar tolerance = (max)(vectorD.array().abs().maxCoeff() * NumTraits<Scalar>::epsilon(),
RealScalar(1) / NumTraits<RealScalar>::highest()); // motivated by LAPACK's xGELSS
for (Index i = 0; i < vectorD.size(); ++i) {
if(abs(vectorD(i)) > tolerance)
dst.row(i) /= vectorD(i);
else
dst.row(i).setZero();
}
dst = dec().vectorD().asDiagonal().inverse() * dst;
// dst = L^-T (D^-1 L^-1 P b)
dec().matrixU().solveInPlace(dst);

View File

@@ -29,7 +29,7 @@ namespace internal{
template<typename MatrixType, int UpLo> struct LLT_Traits;
}
/** \ingroup Cholesky_Module
/** \ingroup cholesky_Module
*
* \class LLT
*
@@ -49,9 +49,6 @@ template<typename MatrixType, int UpLo> struct LLT_Traits;
* use LDLT instead for the semidefinite case. Also, do not use a Cholesky decomposition to determine whether a system of equations
* has a solution.
*
* Example: \include LLT_example.cpp
* Output: \verbinclude LLT_example.out
*
* \sa MatrixBase::llt(), class LDLT
*/
/* HEY THIS DOX IS DISABLED BECAUSE THERE's A BUG EITHER HERE OR IN LDLT ABOUT THAT (OR BOTH)
@@ -181,9 +178,6 @@ template<typename _MatrixType, int _UpLo> class LLT
inline Index rows() const { return m_matrix.rows(); }
inline Index cols() const { return m_matrix.cols(); }
template<typename VectorType>
void rankUpdate(const VectorType& vec);
protected:
/** \internal
* Used to compute and store L
@@ -260,35 +254,6 @@ template<> struct llt_inplace<Lower>
}
return -1;
}
template<typename MatrixType, typename VectorType>
static void rankUpdate(MatrixType& mat, const VectorType& vec)
{
typedef typename MatrixType::ColXpr ColXpr;
typedef typename internal::remove_all<ColXpr>::type ColXprCleaned;
typedef typename ColXprCleaned::SegmentReturnType ColXprSegment;
typedef typename MatrixType::Scalar Scalar;
typedef Matrix<Scalar,Dynamic,1> TempVectorType;
typedef typename TempVectorType::SegmentReturnType TempVecSegment;
int n = mat.cols();
eigen_assert(mat.rows()==n && vec.size()==n);
TempVectorType temp(vec);
for(int i=0; i<n; ++i)
{
JacobiRotation<Scalar> g;
g.makeGivens(mat(i,i), -temp(i), &mat(i,i));
int rs = n-i-1;
if(rs>0)
{
ColXprSegment x(mat.col(i).tail(rs));
TempVecSegment y(temp.tail(rs));
apply_rotation_in_the_plane(x, y, g);
}
}
}
};
template<> struct llt_inplace<Upper>
@@ -305,12 +270,6 @@ template<> struct llt_inplace<Upper>
Transpose<MatrixType> matt(mat);
return llt_inplace<Lower>::blocked(matt);
}
template<typename MatrixType, typename VectorType>
static void rankUpdate(MatrixType& mat, const VectorType& vec)
{
Transpose<MatrixType> matt(mat);
return llt_inplace<Lower>::rankUpdate(matt, vec.conjugate());
}
};
template<typename MatrixType> struct LLT_Traits<MatrixType,Lower>
@@ -337,10 +296,8 @@ template<typename MatrixType> struct LLT_Traits<MatrixType,Upper>
/** Computes / recomputes the Cholesky decomposition A = LL^* = U^*U of \a matrix
*
* \returns a reference to *this
*
* Example: \include TutorialLinAlgComputeTwice.cpp
* Output: \verbinclude TutorialLinAlgComputeTwice.out
* \returns a reference to *this
*/
template<typename MatrixType, int _UpLo>
LLT<MatrixType,_UpLo>& LLT<MatrixType,_UpLo>::compute(const MatrixType& a)
@@ -357,20 +314,6 @@ LLT<MatrixType,_UpLo>& LLT<MatrixType,_UpLo>::compute(const MatrixType& a)
return *this;
}
/** Performs a rank one update of the current decomposition.
* If A = LL^* before the rank one update,
* then after it we have LL^* = A + vv^* where \a v must be a vector
* of same dimension.
*
*/
template<typename MatrixType, int _UpLo>
template<typename VectorType>
void LLT<MatrixType,_UpLo>::rankUpdate(const VectorType& v)
{
EIGEN_STATIC_ASSERT_VECTOR_ONLY(VectorType);
internal::llt_inplace<UpLo>::rankUpdate(m_matrix,v);
}
namespace internal {
template<typename _MatrixType, int UpLo, typename Rhs>
struct solve_retval<LLT<_MatrixType, UpLo>, Rhs>
@@ -441,4 +384,3 @@ SelfAdjointView<MatrixType, UpLo>::llt() const
}
#endif // EIGEN_LLT_H

View File

@@ -1,6 +0,0 @@
FILE(GLOB Eigen_CholmodSupport_SRCS "*.h")
INSTALL(FILES
${Eigen_CholmodSupport_SRCS}
DESTINATION ${INCLUDE_INSTALL_DIR}/Eigen/src/CholmodSupport COMPONENT Devel
)

View File

@@ -68,8 +68,10 @@ class Array
friend struct internal::conservative_resize_like_impl;
using Base::m_storage;
public:
enum { NeedsToAlign = (!(Options&DontAlign))
&& SizeAtCompileTime!=Dynamic && ((static_cast<int>(sizeof(Scalar))*SizeAtCompileTime)%16)==0 };
EIGEN_MAKE_ALIGNED_OPERATOR_NEW_IF(NeedsToAlign)
using Base::base;
using Base::coeff;

View File

@@ -128,12 +128,6 @@ class ArrayWrapper : public ArrayBase<ArrayWrapper<ExpressionType> >
template<typename Dest>
inline void evalTo(Dest& dst) const { dst = m_expression; }
const typename internal::remove_all<NestedExpressionType>::type&
nestedExpression() const
{
return m_expression;
}
protected:
const NestedExpressionType m_expression;
};
@@ -238,12 +232,6 @@ class MatrixWrapper : public MatrixBase<MatrixWrapper<ExpressionType> >
m_expression.const_cast_derived().template writePacket<LoadMode>(index, x);
}
const typename internal::remove_all<NestedExpressionType>::type&
nestedExpression() const
{
return m_expression;
}
protected:
const NestedExpressionType m_expression;
};

View File

@@ -251,22 +251,21 @@ struct assign_innervec_InnerUnrolling<Derived1, Derived2, Stop, Stop>
template<typename Derived1, typename Derived2,
int Traversal = assign_traits<Derived1, Derived2>::Traversal,
int Unrolling = assign_traits<Derived1, Derived2>::Unrolling,
int Version = Specialized>
int Unrolling = assign_traits<Derived1, Derived2>::Unrolling>
struct assign_impl;
/************************
*** Default traversal ***
************************/
template<typename Derived1, typename Derived2, int Unrolling, int Version>
struct assign_impl<Derived1, Derived2, InvalidTraversal, Unrolling, Version>
template<typename Derived1, typename Derived2, int Unrolling>
struct assign_impl<Derived1, Derived2, InvalidTraversal, Unrolling>
{
inline static void run(Derived1 &, const Derived2 &) { }
};
template<typename Derived1, typename Derived2, int Version>
struct assign_impl<Derived1, Derived2, DefaultTraversal, NoUnrolling, Version>
template<typename Derived1, typename Derived2>
struct assign_impl<Derived1, Derived2, DefaultTraversal, NoUnrolling>
{
typedef typename Derived1::Index Index;
inline static void run(Derived1 &dst, const Derived2 &src)
@@ -279,8 +278,8 @@ struct assign_impl<Derived1, Derived2, DefaultTraversal, NoUnrolling, Version>
}
};
template<typename Derived1, typename Derived2, int Version>
struct assign_impl<Derived1, Derived2, DefaultTraversal, CompleteUnrolling, Version>
template<typename Derived1, typename Derived2>
struct assign_impl<Derived1, Derived2, DefaultTraversal, CompleteUnrolling>
{
EIGEN_STRONG_INLINE static void run(Derived1 &dst, const Derived2 &src)
{
@@ -289,8 +288,8 @@ struct assign_impl<Derived1, Derived2, DefaultTraversal, CompleteUnrolling, Vers
}
};
template<typename Derived1, typename Derived2, int Version>
struct assign_impl<Derived1, Derived2, DefaultTraversal, InnerUnrolling, Version>
template<typename Derived1, typename Derived2>
struct assign_impl<Derived1, Derived2, DefaultTraversal, InnerUnrolling>
{
typedef typename Derived1::Index Index;
EIGEN_STRONG_INLINE static void run(Derived1 &dst, const Derived2 &src)
@@ -306,8 +305,8 @@ struct assign_impl<Derived1, Derived2, DefaultTraversal, InnerUnrolling, Version
*** Linear traversal ***
***********************/
template<typename Derived1, typename Derived2, int Version>
struct assign_impl<Derived1, Derived2, LinearTraversal, NoUnrolling, Version>
template<typename Derived1, typename Derived2>
struct assign_impl<Derived1, Derived2, LinearTraversal, NoUnrolling>
{
typedef typename Derived1::Index Index;
inline static void run(Derived1 &dst, const Derived2 &src)
@@ -318,8 +317,8 @@ struct assign_impl<Derived1, Derived2, LinearTraversal, NoUnrolling, Version>
}
};
template<typename Derived1, typename Derived2, int Version>
struct assign_impl<Derived1, Derived2, LinearTraversal, CompleteUnrolling, Version>
template<typename Derived1, typename Derived2>
struct assign_impl<Derived1, Derived2, LinearTraversal, CompleteUnrolling>
{
EIGEN_STRONG_INLINE static void run(Derived1 &dst, const Derived2 &src)
{
@@ -332,8 +331,8 @@ struct assign_impl<Derived1, Derived2, LinearTraversal, CompleteUnrolling, Versi
*** Inner vectorization ***
**************************/
template<typename Derived1, typename Derived2, int Version>
struct assign_impl<Derived1, Derived2, InnerVectorizedTraversal, NoUnrolling, Version>
template<typename Derived1, typename Derived2>
struct assign_impl<Derived1, Derived2, InnerVectorizedTraversal, NoUnrolling>
{
typedef typename Derived1::Index Index;
inline static void run(Derived1 &dst, const Derived2 &src)
@@ -347,8 +346,8 @@ struct assign_impl<Derived1, Derived2, InnerVectorizedTraversal, NoUnrolling, Ve
}
};
template<typename Derived1, typename Derived2, int Version>
struct assign_impl<Derived1, Derived2, InnerVectorizedTraversal, CompleteUnrolling, Version>
template<typename Derived1, typename Derived2>
struct assign_impl<Derived1, Derived2, InnerVectorizedTraversal, CompleteUnrolling>
{
EIGEN_STRONG_INLINE static void run(Derived1 &dst, const Derived2 &src)
{
@@ -357,8 +356,8 @@ struct assign_impl<Derived1, Derived2, InnerVectorizedTraversal, CompleteUnrolli
}
};
template<typename Derived1, typename Derived2, int Version>
struct assign_impl<Derived1, Derived2, InnerVectorizedTraversal, InnerUnrolling, Version>
template<typename Derived1, typename Derived2>
struct assign_impl<Derived1, Derived2, InnerVectorizedTraversal, InnerUnrolling>
{
typedef typename Derived1::Index Index;
EIGEN_STRONG_INLINE static void run(Derived1 &dst, const Derived2 &src)
@@ -399,8 +398,8 @@ struct unaligned_assign_impl<false>
}
};
template<typename Derived1, typename Derived2, int Version>
struct assign_impl<Derived1, Derived2, LinearVectorizedTraversal, NoUnrolling, Version>
template<typename Derived1, typename Derived2>
struct assign_impl<Derived1, Derived2, LinearVectorizedTraversal, NoUnrolling>
{
typedef typename Derived1::Index Index;
EIGEN_STRONG_INLINE static void run(Derived1 &dst, const Derived2 &src)
@@ -427,8 +426,8 @@ struct assign_impl<Derived1, Derived2, LinearVectorizedTraversal, NoUnrolling, V
}
};
template<typename Derived1, typename Derived2, int Version>
struct assign_impl<Derived1, Derived2, LinearVectorizedTraversal, CompleteUnrolling, Version>
template<typename Derived1, typename Derived2>
struct assign_impl<Derived1, Derived2, LinearVectorizedTraversal, CompleteUnrolling>
{
typedef typename Derived1::Index Index;
EIGEN_STRONG_INLINE static void run(Derived1 &dst, const Derived2 &src)
@@ -446,8 +445,8 @@ struct assign_impl<Derived1, Derived2, LinearVectorizedTraversal, CompleteUnroll
*** Slice vectorization ***
***************************/
template<typename Derived1, typename Derived2, int Version>
struct assign_impl<Derived1, Derived2, SliceVectorizedTraversal, NoUnrolling, Version>
template<typename Derived1, typename Derived2>
struct assign_impl<Derived1, Derived2, SliceVectorizedTraversal, NoUnrolling>
{
typedef typename Derived1::Index Index;
inline static void run(Derived1 &dst, const Derived2 &src)

View File

@@ -1,682 +0,0 @@
// This file is part of Eigen, a lightweight C++ template library
// for linear algebra.
//
// Copyright (C) 2011 Benoit Jacob <jacob.benoit.1@gmail.com>
// Copyright (C) 2011 Gael Guennebaud <gael.guennebaud@inria.fr>
// Copyright (C) 2011 Jitse Niesen <jitse@maths.leeds.ac.uk>
//
// Eigen is free software; you can redistribute it and/or
// modify it under the terms of the GNU Lesser General Public
// License as published by the Free Software Foundation; either
// version 3 of the License, or (at your option) any later version.
//
// Alternatively, you can redistribute it and/or
// modify it under the terms of the GNU General Public License as
// published by the Free Software Foundation; either version 2 of
// the License, or (at your option) any later version.
//
// Eigen is distributed in the hope that it will be useful, but WITHOUT ANY
// WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS
// FOR A PARTICULAR PURPOSE. See the GNU Lesser General Public License or the
// GNU General Public License for more details.
//
// You should have received a copy of the GNU Lesser General Public
// License and a copy of the GNU General Public License along with
// Eigen. If not, see <http://www.gnu.org/licenses/>.
#ifndef EIGEN_ASSIGN_EVALUATOR_H
#define EIGEN_ASSIGN_EVALUATOR_H
// This implementation is based on Assign.h
namespace internal {
/***************************************************************************
* Part 1 : the logic deciding a strategy for traversal and unrolling *
***************************************************************************/
// copy_using_evaluator_traits is based on assign_traits
// (actually, it's identical)
template <typename Derived, typename OtherDerived>
struct copy_using_evaluator_traits
{
public:
enum {
DstIsAligned = Derived::Flags & AlignedBit,
DstHasDirectAccess = Derived::Flags & DirectAccessBit,
SrcIsAligned = OtherDerived::Flags & AlignedBit,
JointAlignment = bool(DstIsAligned) && bool(SrcIsAligned) ? Aligned : Unaligned
};
private:
enum {
InnerSize = int(Derived::IsVectorAtCompileTime) ? int(Derived::SizeAtCompileTime)
: int(Derived::Flags)&RowMajorBit ? int(Derived::ColsAtCompileTime)
: int(Derived::RowsAtCompileTime),
InnerMaxSize = int(Derived::IsVectorAtCompileTime) ? int(Derived::MaxSizeAtCompileTime)
: int(Derived::Flags)&RowMajorBit ? int(Derived::MaxColsAtCompileTime)
: int(Derived::MaxRowsAtCompileTime),
MaxSizeAtCompileTime = Derived::SizeAtCompileTime,
PacketSize = packet_traits<typename Derived::Scalar>::size
};
enum {
StorageOrdersAgree = (int(Derived::IsRowMajor) == int(OtherDerived::IsRowMajor)),
MightVectorize = StorageOrdersAgree
&& (int(Derived::Flags) & int(OtherDerived::Flags) & ActualPacketAccessBit),
MayInnerVectorize = MightVectorize && int(InnerSize)!=Dynamic && int(InnerSize)%int(PacketSize)==0
&& int(DstIsAligned) && int(SrcIsAligned),
MayLinearize = StorageOrdersAgree && (int(Derived::Flags) & int(OtherDerived::Flags) & LinearAccessBit),
MayLinearVectorize = MightVectorize && MayLinearize && DstHasDirectAccess
&& (DstIsAligned || MaxSizeAtCompileTime == Dynamic),
/* If the destination isn't aligned, we have to do runtime checks and we don't unroll,
so it's only good for large enough sizes. */
MaySliceVectorize = MightVectorize && DstHasDirectAccess
&& (int(InnerMaxSize)==Dynamic || int(InnerMaxSize)>=3*PacketSize)
/* slice vectorization can be slow, so we only want it if the slices are big, which is
indicated by InnerMaxSize rather than InnerSize, think of the case of a dynamic block
in a fixed-size matrix */
};
public:
enum {
Traversal = int(MayInnerVectorize) ? int(InnerVectorizedTraversal)
: int(MayLinearVectorize) ? int(LinearVectorizedTraversal)
: int(MaySliceVectorize) ? int(SliceVectorizedTraversal)
: int(MayLinearize) ? int(LinearTraversal)
: int(DefaultTraversal),
Vectorized = int(Traversal) == InnerVectorizedTraversal
|| int(Traversal) == LinearVectorizedTraversal
|| int(Traversal) == SliceVectorizedTraversal
};
private:
enum {
UnrollingLimit = EIGEN_UNROLLING_LIMIT * (Vectorized ? int(PacketSize) : 1),
MayUnrollCompletely = int(Derived::SizeAtCompileTime) != Dynamic
&& int(OtherDerived::CoeffReadCost) != Dynamic
&& int(Derived::SizeAtCompileTime) * int(OtherDerived::CoeffReadCost) <= int(UnrollingLimit),
MayUnrollInner = int(InnerSize) != Dynamic
&& int(OtherDerived::CoeffReadCost) != Dynamic
&& int(InnerSize) * int(OtherDerived::CoeffReadCost) <= int(UnrollingLimit)
};
public:
enum {
Unrolling = (int(Traversal) == int(InnerVectorizedTraversal) || int(Traversal) == int(DefaultTraversal))
? (
int(MayUnrollCompletely) ? int(CompleteUnrolling)
: int(MayUnrollInner) ? int(InnerUnrolling)
: int(NoUnrolling)
)
: int(Traversal) == int(LinearVectorizedTraversal)
? ( bool(MayUnrollCompletely) && bool(DstIsAligned) ? int(CompleteUnrolling)
: int(NoUnrolling) )
: int(Traversal) == int(LinearTraversal)
? ( bool(MayUnrollCompletely) ? int(CompleteUnrolling)
: int(NoUnrolling) )
: int(NoUnrolling)
};
#ifdef EIGEN_DEBUG_ASSIGN
static void debug()
{
EIGEN_DEBUG_VAR(DstIsAligned)
EIGEN_DEBUG_VAR(SrcIsAligned)
EIGEN_DEBUG_VAR(JointAlignment)
EIGEN_DEBUG_VAR(InnerSize)
EIGEN_DEBUG_VAR(InnerMaxSize)
EIGEN_DEBUG_VAR(PacketSize)
EIGEN_DEBUG_VAR(StorageOrdersAgree)
EIGEN_DEBUG_VAR(MightVectorize)
EIGEN_DEBUG_VAR(MayLinearize)
EIGEN_DEBUG_VAR(MayInnerVectorize)
EIGEN_DEBUG_VAR(MayLinearVectorize)
EIGEN_DEBUG_VAR(MaySliceVectorize)
EIGEN_DEBUG_VAR(Traversal)
EIGEN_DEBUG_VAR(UnrollingLimit)
EIGEN_DEBUG_VAR(MayUnrollCompletely)
EIGEN_DEBUG_VAR(MayUnrollInner)
EIGEN_DEBUG_VAR(Unrolling)
}
#endif
};
/***************************************************************************
* Part 2 : meta-unrollers
***************************************************************************/
/************************
*** Default traversal ***
************************/
template<typename DstEvaluatorType, typename SrcEvaluatorType, int Index, int Stop>
struct copy_using_evaluator_DefaultTraversal_CompleteUnrolling
{
typedef typename DstEvaluatorType::XprType DstXprType;
enum {
outer = Index / DstXprType::InnerSizeAtCompileTime,
inner = Index % DstXprType::InnerSizeAtCompileTime
};
EIGEN_STRONG_INLINE static void run(DstEvaluatorType &dstEvaluator,
SrcEvaluatorType &srcEvaluator)
{
dstEvaluator.copyCoeffByOuterInner(outer, inner, srcEvaluator);
copy_using_evaluator_DefaultTraversal_CompleteUnrolling
<DstEvaluatorType, SrcEvaluatorType, Index+1, Stop>
::run(dstEvaluator, srcEvaluator);
}
};
template<typename DstEvaluatorType, typename SrcEvaluatorType, int Stop>
struct copy_using_evaluator_DefaultTraversal_CompleteUnrolling<DstEvaluatorType, SrcEvaluatorType, Stop, Stop>
{
EIGEN_STRONG_INLINE static void run(DstEvaluatorType&, SrcEvaluatorType&) { }
};
template<typename DstEvaluatorType, typename SrcEvaluatorType, int Index, int Stop>
struct copy_using_evaluator_DefaultTraversal_InnerUnrolling
{
EIGEN_STRONG_INLINE static void run(DstEvaluatorType &dstEvaluator,
SrcEvaluatorType &srcEvaluator,
int outer)
{
dstEvaluator.copyCoeffByOuterInner(outer, Index, srcEvaluator);
copy_using_evaluator_DefaultTraversal_InnerUnrolling
<DstEvaluatorType, SrcEvaluatorType, Index+1, Stop>
::run(dstEvaluator, srcEvaluator, outer);
}
};
template<typename DstEvaluatorType, typename SrcEvaluatorType, int Stop>
struct copy_using_evaluator_DefaultTraversal_InnerUnrolling<DstEvaluatorType, SrcEvaluatorType, Stop, Stop>
{
EIGEN_STRONG_INLINE static void run(DstEvaluatorType&, SrcEvaluatorType&, int) { }
};
/***********************
*** Linear traversal ***
***********************/
template<typename DstEvaluatorType, typename SrcEvaluatorType, int Index, int Stop>
struct copy_using_evaluator_LinearTraversal_CompleteUnrolling
{
EIGEN_STRONG_INLINE static void run(DstEvaluatorType &dstEvaluator,
SrcEvaluatorType &srcEvaluator)
{
dstEvaluator.copyCoeff(Index, srcEvaluator);
copy_using_evaluator_LinearTraversal_CompleteUnrolling
<DstEvaluatorType, SrcEvaluatorType, Index+1, Stop>
::run(dstEvaluator, srcEvaluator);
}
};
template<typename DstEvaluatorType, typename SrcEvaluatorType, int Stop>
struct copy_using_evaluator_LinearTraversal_CompleteUnrolling<DstEvaluatorType, SrcEvaluatorType, Stop, Stop>
{
EIGEN_STRONG_INLINE static void run(DstEvaluatorType&, SrcEvaluatorType&) { }
};
/**************************
*** Inner vectorization ***
**************************/
template<typename DstEvaluatorType, typename SrcEvaluatorType, int Index, int Stop>
struct copy_using_evaluator_innervec_CompleteUnrolling
{
typedef typename DstEvaluatorType::XprType DstXprType;
typedef typename SrcEvaluatorType::XprType SrcXprType;
enum {
outer = Index / DstXprType::InnerSizeAtCompileTime,
inner = Index % DstXprType::InnerSizeAtCompileTime,
JointAlignment = copy_using_evaluator_traits<DstXprType,SrcXprType>::JointAlignment
};
EIGEN_STRONG_INLINE static void run(DstEvaluatorType &dstEvaluator,
SrcEvaluatorType &srcEvaluator)
{
dstEvaluator.template copyPacketByOuterInner<Aligned, JointAlignment>(outer, inner, srcEvaluator);
enum { NextIndex = Index + packet_traits<typename DstXprType::Scalar>::size };
copy_using_evaluator_innervec_CompleteUnrolling
<DstEvaluatorType, SrcEvaluatorType, NextIndex, Stop>
::run(dstEvaluator, srcEvaluator);
}
};
template<typename DstEvaluatorType, typename SrcEvaluatorType, int Stop>
struct copy_using_evaluator_innervec_CompleteUnrolling<DstEvaluatorType, SrcEvaluatorType, Stop, Stop>
{
EIGEN_STRONG_INLINE static void run(DstEvaluatorType&, SrcEvaluatorType&) { }
};
template<typename DstEvaluatorType, typename SrcEvaluatorType, int Index, int Stop>
struct copy_using_evaluator_innervec_InnerUnrolling
{
EIGEN_STRONG_INLINE static void run(DstEvaluatorType &dstEvaluator,
SrcEvaluatorType &srcEvaluator,
int outer)
{
dstEvaluator.template copyPacketByOuterInner<Aligned, Aligned>(outer, Index, srcEvaluator);
typedef typename DstEvaluatorType::XprType DstXprType;
enum { NextIndex = Index + packet_traits<typename DstXprType::Scalar>::size };
copy_using_evaluator_innervec_InnerUnrolling
<DstEvaluatorType, SrcEvaluatorType, NextIndex, Stop>
::run(dstEvaluator, srcEvaluator, outer);
}
};
template<typename DstEvaluatorType, typename SrcEvaluatorType, int Stop>
struct copy_using_evaluator_innervec_InnerUnrolling<DstEvaluatorType, SrcEvaluatorType, Stop, Stop>
{
EIGEN_STRONG_INLINE static void run(DstEvaluatorType&, SrcEvaluatorType&, int) { }
};
/***************************************************************************
* Part 3 : implementation of all cases
***************************************************************************/
// copy_using_evaluator_impl is based on assign_impl
template<typename DstXprType, typename SrcXprType,
int Traversal = copy_using_evaluator_traits<DstXprType, SrcXprType>::Traversal,
int Unrolling = copy_using_evaluator_traits<DstXprType, SrcXprType>::Unrolling>
struct copy_using_evaluator_impl;
/************************
*** Default traversal ***
************************/
template<typename DstXprType, typename SrcXprType>
struct copy_using_evaluator_impl<DstXprType, SrcXprType, DefaultTraversal, NoUnrolling>
{
static void run(DstXprType& dst, const SrcXprType& src)
{
typedef typename evaluator<DstXprType>::type DstEvaluatorType;
typedef typename evaluator<SrcXprType>::type SrcEvaluatorType;
typedef typename DstXprType::Index Index;
DstEvaluatorType dstEvaluator(dst);
SrcEvaluatorType srcEvaluator(src);
for(Index outer = 0; outer < dst.outerSize(); ++outer) {
for(Index inner = 0; inner < dst.innerSize(); ++inner) {
dstEvaluator.copyCoeffByOuterInner(outer, inner, srcEvaluator);
}
}
}
};
template<typename DstXprType, typename SrcXprType>
struct copy_using_evaluator_impl<DstXprType, SrcXprType, DefaultTraversal, CompleteUnrolling>
{
EIGEN_STRONG_INLINE static void run(DstXprType &dst, const SrcXprType &src)
{
typedef typename evaluator<DstXprType>::type DstEvaluatorType;
typedef typename evaluator<SrcXprType>::type SrcEvaluatorType;
DstEvaluatorType dstEvaluator(dst);
SrcEvaluatorType srcEvaluator(src);
copy_using_evaluator_DefaultTraversal_CompleteUnrolling
<DstEvaluatorType, SrcEvaluatorType, 0, DstXprType::SizeAtCompileTime>
::run(dstEvaluator, srcEvaluator);
}
};
template<typename DstXprType, typename SrcXprType>
struct copy_using_evaluator_impl<DstXprType, SrcXprType, DefaultTraversal, InnerUnrolling>
{
typedef typename DstXprType::Index Index;
EIGEN_STRONG_INLINE static void run(DstXprType &dst, const SrcXprType &src)
{
typedef typename evaluator<DstXprType>::type DstEvaluatorType;
typedef typename evaluator<SrcXprType>::type SrcEvaluatorType;
DstEvaluatorType dstEvaluator(dst);
SrcEvaluatorType srcEvaluator(src);
const Index outerSize = dst.outerSize();
for(Index outer = 0; outer < outerSize; ++outer)
copy_using_evaluator_DefaultTraversal_InnerUnrolling
<DstEvaluatorType, SrcEvaluatorType, 0, DstXprType::InnerSizeAtCompileTime>
::run(dstEvaluator, srcEvaluator, outer);
}
};
/***************************
*** Linear vectorization ***
***************************/
template <bool IsAligned = false>
struct unaligned_copy_using_evaluator_impl
{
// if IsAligned = true, then do nothing
template <typename SrcEvaluatorType, typename DstEvaluatorType>
static EIGEN_STRONG_INLINE void run(const SrcEvaluatorType&, DstEvaluatorType&,
typename SrcEvaluatorType::Index, typename SrcEvaluatorType::Index) {}
};
template <>
struct unaligned_copy_using_evaluator_impl<false>
{
// MSVC must not inline this functions. If it does, it fails to optimize the
// packet access path.
#ifdef _MSC_VER
template <typename DstEvaluatorType, typename SrcEvaluatorType>
static EIGEN_DONT_INLINE void run(DstEvaluatorType &dstEvaluator,
const SrcEvaluatorType &srcEvaluator,
typename DstEvaluatorType::Index start,
typename DstEvaluatorType::Index end)
#else
template <typename DstEvaluatorType, typename SrcEvaluatorType>
static EIGEN_STRONG_INLINE void run(DstEvaluatorType &dstEvaluator,
const SrcEvaluatorType &srcEvaluator,
typename DstEvaluatorType::Index start,
typename DstEvaluatorType::Index end)
#endif
{
for (typename DstEvaluatorType::Index index = start; index < end; ++index)
dstEvaluator.copyCoeff(index, srcEvaluator);
}
};
template<typename DstXprType, typename SrcXprType>
struct copy_using_evaluator_impl<DstXprType, SrcXprType, LinearVectorizedTraversal, NoUnrolling>
{
EIGEN_STRONG_INLINE static void run(DstXprType &dst, const SrcXprType &src)
{
typedef typename evaluator<DstXprType>::type DstEvaluatorType;
typedef typename evaluator<SrcXprType>::type SrcEvaluatorType;
typedef typename DstXprType::Index Index;
DstEvaluatorType dstEvaluator(dst);
SrcEvaluatorType srcEvaluator(src);
const Index size = dst.size();
typedef packet_traits<typename DstXprType::Scalar> PacketTraits;
enum {
packetSize = PacketTraits::size,
dstIsAligned = int(copy_using_evaluator_traits<DstXprType,SrcXprType>::DstIsAligned),
dstAlignment = PacketTraits::AlignedOnScalar ? Aligned : dstIsAligned,
srcAlignment = copy_using_evaluator_traits<DstXprType,SrcXprType>::JointAlignment
};
const Index alignedStart = dstIsAligned ? 0 : first_aligned(&dstEvaluator.coeffRef(0), size);
const Index alignedEnd = alignedStart + ((size-alignedStart)/packetSize)*packetSize;
unaligned_copy_using_evaluator_impl<dstIsAligned!=0>::run(dstEvaluator, srcEvaluator, 0, alignedStart);
for(Index index = alignedStart; index < alignedEnd; index += packetSize)
{
dstEvaluator.template copyPacket<dstAlignment, srcAlignment>(index, srcEvaluator);
}
unaligned_copy_using_evaluator_impl<>::run(dstEvaluator, srcEvaluator, alignedEnd, size);
}
};
template<typename DstXprType, typename SrcXprType>
struct copy_using_evaluator_impl<DstXprType, SrcXprType, LinearVectorizedTraversal, CompleteUnrolling>
{
typedef typename DstXprType::Index Index;
EIGEN_STRONG_INLINE static void run(DstXprType &dst, const SrcXprType &src)
{
typedef typename evaluator<DstXprType>::type DstEvaluatorType;
typedef typename evaluator<SrcXprType>::type SrcEvaluatorType;
DstEvaluatorType dstEvaluator(dst);
SrcEvaluatorType srcEvaluator(src);
enum { size = DstXprType::SizeAtCompileTime,
packetSize = packet_traits<typename DstXprType::Scalar>::size,
alignedSize = (size/packetSize)*packetSize };
copy_using_evaluator_innervec_CompleteUnrolling
<DstEvaluatorType, SrcEvaluatorType, 0, alignedSize>
::run(dstEvaluator, srcEvaluator);
copy_using_evaluator_DefaultTraversal_CompleteUnrolling
<DstEvaluatorType, SrcEvaluatorType, alignedSize, size>
::run(dstEvaluator, srcEvaluator);
}
};
/**************************
*** Inner vectorization ***
**************************/
template<typename DstXprType, typename SrcXprType>
struct copy_using_evaluator_impl<DstXprType, SrcXprType, InnerVectorizedTraversal, NoUnrolling>
{
inline static void run(DstXprType &dst, const SrcXprType &src)
{
typedef typename evaluator<DstXprType>::type DstEvaluatorType;
typedef typename evaluator<SrcXprType>::type SrcEvaluatorType;
typedef typename DstXprType::Index Index;
DstEvaluatorType dstEvaluator(dst);
SrcEvaluatorType srcEvaluator(src);
const Index innerSize = dst.innerSize();
const Index outerSize = dst.outerSize();
const Index packetSize = packet_traits<typename DstXprType::Scalar>::size;
for(Index outer = 0; outer < outerSize; ++outer)
for(Index inner = 0; inner < innerSize; inner+=packetSize) {
dstEvaluator.template copyPacketByOuterInner<Aligned, Aligned>(outer, inner, srcEvaluator);
}
}
};
template<typename DstXprType, typename SrcXprType>
struct copy_using_evaluator_impl<DstXprType, SrcXprType, InnerVectorizedTraversal, CompleteUnrolling>
{
EIGEN_STRONG_INLINE static void run(DstXprType &dst, const SrcXprType &src)
{
typedef typename evaluator<DstXprType>::type DstEvaluatorType;
typedef typename evaluator<SrcXprType>::type SrcEvaluatorType;
DstEvaluatorType dstEvaluator(dst);
SrcEvaluatorType srcEvaluator(src);
copy_using_evaluator_innervec_CompleteUnrolling
<DstEvaluatorType, SrcEvaluatorType, 0, DstXprType::SizeAtCompileTime>
::run(dstEvaluator, srcEvaluator);
}
};
template<typename DstXprType, typename SrcXprType>
struct copy_using_evaluator_impl<DstXprType, SrcXprType, InnerVectorizedTraversal, InnerUnrolling>
{
typedef typename DstXprType::Index Index;
EIGEN_STRONG_INLINE static void run(DstXprType &dst, const SrcXprType &src)
{
typedef typename evaluator<DstXprType>::type DstEvaluatorType;
typedef typename evaluator<SrcXprType>::type SrcEvaluatorType;
DstEvaluatorType dstEvaluator(dst);
SrcEvaluatorType srcEvaluator(src);
const Index outerSize = dst.outerSize();
for(Index outer = 0; outer < outerSize; ++outer)
copy_using_evaluator_innervec_InnerUnrolling
<DstEvaluatorType, SrcEvaluatorType, 0, DstXprType::InnerSizeAtCompileTime>
::run(dstEvaluator, srcEvaluator, outer);
}
};
/***********************
*** Linear traversal ***
***********************/
template<typename DstXprType, typename SrcXprType>
struct copy_using_evaluator_impl<DstXprType, SrcXprType, LinearTraversal, NoUnrolling>
{
inline static void run(DstXprType &dst, const SrcXprType &src)
{
typedef typename evaluator<DstXprType>::type DstEvaluatorType;
typedef typename evaluator<SrcXprType>::type SrcEvaluatorType;
typedef typename DstXprType::Index Index;
DstEvaluatorType dstEvaluator(dst);
SrcEvaluatorType srcEvaluator(src);
const Index size = dst.size();
for(Index i = 0; i < size; ++i)
dstEvaluator.copyCoeff(i, srcEvaluator);
}
};
template<typename DstXprType, typename SrcXprType>
struct copy_using_evaluator_impl<DstXprType, SrcXprType, LinearTraversal, CompleteUnrolling>
{
EIGEN_STRONG_INLINE static void run(DstXprType &dst, const SrcXprType &src)
{
typedef typename evaluator<DstXprType>::type DstEvaluatorType;
typedef typename evaluator<SrcXprType>::type SrcEvaluatorType;
DstEvaluatorType dstEvaluator(dst);
SrcEvaluatorType srcEvaluator(src);
copy_using_evaluator_LinearTraversal_CompleteUnrolling
<DstEvaluatorType, SrcEvaluatorType, 0, DstXprType::SizeAtCompileTime>
::run(dstEvaluator, srcEvaluator);
}
};
/**************************
*** Slice vectorization ***
***************************/
template<typename DstXprType, typename SrcXprType>
struct copy_using_evaluator_impl<DstXprType, SrcXprType, SliceVectorizedTraversal, NoUnrolling>
{
inline static void run(DstXprType &dst, const SrcXprType &src)
{
typedef typename evaluator<DstXprType>::type DstEvaluatorType;
typedef typename evaluator<SrcXprType>::type SrcEvaluatorType;
typedef typename DstXprType::Index Index;
DstEvaluatorType dstEvaluator(dst);
SrcEvaluatorType srcEvaluator(src);
typedef packet_traits<typename DstXprType::Scalar> PacketTraits;
enum {
packetSize = PacketTraits::size,
alignable = PacketTraits::AlignedOnScalar,
dstAlignment = alignable ? Aligned : int(copy_using_evaluator_traits<DstXprType,SrcXprType>::DstIsAligned) ,
srcAlignment = copy_using_evaluator_traits<DstXprType,SrcXprType>::JointAlignment
};
const Index packetAlignedMask = packetSize - 1;
const Index innerSize = dst.innerSize();
const Index outerSize = dst.outerSize();
const Index alignedStep = alignable ? (packetSize - dst.outerStride() % packetSize) & packetAlignedMask : 0;
Index alignedStart = ((!alignable) || copy_using_evaluator_traits<DstXprType,SrcXprType>::DstIsAligned) ? 0
: first_aligned(&dstEvaluator.coeffRef(0,0), innerSize);
for(Index outer = 0; outer < outerSize; ++outer)
{
const Index alignedEnd = alignedStart + ((innerSize-alignedStart) & ~packetAlignedMask);
// do the non-vectorizable part of the assignment
for(Index inner = 0; inner<alignedStart ; ++inner) {
dstEvaluator.copyCoeffByOuterInner(outer, inner, srcEvaluator);
}
// do the vectorizable part of the assignment
for(Index inner = alignedStart; inner<alignedEnd; inner+=packetSize) {
dstEvaluator.template copyPacketByOuterInner<dstAlignment, srcAlignment>(outer, inner, srcEvaluator);
}
// do the non-vectorizable part of the assignment
for(Index inner = alignedEnd; inner<innerSize ; ++inner) {
dstEvaluator.copyCoeffByOuterInner(outer, inner, srcEvaluator);
}
alignedStart = std::min<Index>((alignedStart+alignedStep)%packetSize, innerSize);
}
}
};
/***************************************************************************
* Part 4 : Entry points
***************************************************************************/
// Based on DenseBase::LazyAssign()
template<typename DstXprType, typename SrcXprType>
const DstXprType& copy_using_evaluator(const DstXprType& dst, const SrcXprType& src)
{
#ifdef EIGEN_DEBUG_ASSIGN
internal::copy_using_evaluator_traits<DstXprType, SrcXprType>::debug();
#endif
copy_using_evaluator_impl<DstXprType, SrcXprType>::run(const_cast<DstXprType&>(dst), src);
return dst;
}
// Based on DenseBase::swap()
// TODO: Chech whether we need to do something special for swapping two
// Arrays or Matrices.
template<typename DstXprType, typename SrcXprType>
void swap_using_evaluator(const DstXprType& dst, const SrcXprType& src)
{
copy_using_evaluator(SwapWrapper<DstXprType>(const_cast<DstXprType&>(dst)), src);
}
// Based on MatrixBase::operator+= (in CwiseBinaryOp.h)
template<typename DstXprType, typename SrcXprType>
void add_assign_using_evaluator(const MatrixBase<DstXprType>& dst, const MatrixBase<SrcXprType>& src)
{
typedef typename DstXprType::Scalar Scalar;
SelfCwiseBinaryOp<internal::scalar_sum_op<Scalar>, DstXprType, SrcXprType> tmp(dst.const_cast_derived());
copy_using_evaluator(tmp, src.derived());
}
// Based on ArrayBase::operator+=
template<typename DstXprType, typename SrcXprType>
void add_assign_using_evaluator(const ArrayBase<DstXprType>& dst, const ArrayBase<SrcXprType>& src)
{
typedef typename DstXprType::Scalar Scalar;
SelfCwiseBinaryOp<internal::scalar_sum_op<Scalar>, DstXprType, SrcXprType> tmp(dst.const_cast_derived());
copy_using_evaluator(tmp, src.derived());
}
// TODO: Add add_assign_using_evaluator for EigenBase ?
template<typename DstXprType, typename SrcXprType>
void subtract_assign_using_evaluator(const MatrixBase<DstXprType>& dst, const MatrixBase<SrcXprType>& src)
{
typedef typename DstXprType::Scalar Scalar;
SelfCwiseBinaryOp<internal::scalar_difference_op<Scalar>, DstXprType, SrcXprType> tmp(dst.const_cast_derived());
copy_using_evaluator(tmp, src.derived());
}
template<typename DstXprType, typename SrcXprType>
void subtract_assign_using_evaluator(const ArrayBase<DstXprType>& dst, const ArrayBase<SrcXprType>& src)
{
typedef typename DstXprType::Scalar Scalar;
SelfCwiseBinaryOp<internal::scalar_difference_op<Scalar>, DstXprType, SrcXprType> tmp(dst.const_cast_derived());
copy_using_evaluator(tmp, src.derived());
}
template<typename DstXprType, typename SrcXprType>
void multiply_assign_using_evaluator(const ArrayBase<DstXprType>& dst, const ArrayBase<SrcXprType>& src)
{
typedef typename DstXprType::Scalar Scalar;
SelfCwiseBinaryOp<internal::scalar_product_op<Scalar>, DstXprType, SrcXprType> tmp(dst.const_cast_derived());
copy_using_evaluator(tmp, src.derived());
}
template<typename DstXprType, typename SrcXprType>
void divide_assign_using_evaluator(const ArrayBase<DstXprType>& dst, const ArrayBase<SrcXprType>& src)
{
typedef typename DstXprType::Scalar Scalar;
SelfCwiseBinaryOp<internal::scalar_quotient_op<Scalar>, DstXprType, SrcXprType> tmp(dst.const_cast_derived());
copy_using_evaluator(tmp, src.derived());
}
} // namespace internal
#endif // EIGEN_ASSIGN_EVALUATOR_H

View File

@@ -94,7 +94,7 @@ struct traits<Block<XprType, BlockRows, BlockCols, InnerPanel, HasDirectAccess>
MaskPacketAccessBit = (InnerSize == Dynamic || (InnerSize % packet_traits<Scalar>::size) == 0)
&& (InnerStrideAtCompileTime == 1)
? PacketAccessBit : 0,
MaskAlignedBit = (InnerPanel && (OuterStrideAtCompileTime!=Dynamic) && (((OuterStrideAtCompileTime * sizeof(Scalar)) % 16) == 0)) ? AlignedBit : 0,
MaskAlignedBit = (InnerPanel && (OuterStrideAtCompileTime!=Dynamic) && ((OuterStrideAtCompileTime % packet_traits<Scalar>::size) == 0)) ? AlignedBit : 0,
FlagsLinearAccessBit = (RowsAtCompileTime == 1 || ColsAtCompileTime == 1) ? LinearAccessBit : 0,
FlagsLvalueBit = is_lvalue<XprType>::value ? LvalueBit : 0,
FlagsRowMajorBit = IsRowMajor ? RowMajorBit : 0,
@@ -242,21 +242,6 @@ template<typename XprType, int BlockRows, int BlockCols, bool InnerPanel, bool H
inline Index outerStride() const;
#endif
const typename internal::remove_all<typename XprType::Nested>::type& nestedExpression() const
{
return m_xpr;
}
Index startRow() const
{
return m_startRow.value();
}
Index startCol() const
{
return m_startCol.value();
}
protected:
const typename XprType::Nested m_xpr;
@@ -319,11 +304,6 @@ class Block<XprType,BlockRows,BlockCols, InnerPanel,true>
init();
}
const typename internal::remove_all<typename XprType::Nested>::type& nestedExpression() const
{
return m_xpr;
}
/** \sa MapBase::innerStride() */
inline Index innerStride() const
{

File diff suppressed because it is too large Load Diff

View File

@@ -101,9 +101,6 @@ class CwiseNullaryOp : internal::no_assignment_operator,
return m_functor.packetOp(index);
}
/** \returns the functor representing the nullary operation */
const NullaryOp& functor() const { return m_functor; }
protected:
const internal::variable_if_dynamic<Index, RowsAtCompileTime> m_rows;
const internal::variable_if_dynamic<Index, ColsAtCompileTime> m_cols;

View File

@@ -128,16 +128,6 @@ template<typename T, int _Rows, int _Cols, int _Options> class DenseStorage<T, 0
inline T *data() { return 0; }
};
// more specializations for null matrices; these are necessary to resolve ambiguities
template<typename T, int _Options> class DenseStorage<T, 0, Dynamic, Dynamic, _Options>
: public DenseStorage<T, 0, 0, 0, _Options> { };
template<typename T, int _Rows, int _Options> class DenseStorage<T, 0, _Rows, Dynamic, _Options>
: public DenseStorage<T, 0, 0, 0, _Options> { };
template<typename T, int _Cols, int _Options> class DenseStorage<T, 0, Dynamic, _Cols, _Options>
: public DenseStorage<T, 0, 0, 0, _Options> { };
// dynamic-size matrix with fixed-size storage
template<typename T, int Size, int _Options> class DenseStorage<T, Size, Dynamic, Dynamic, _Options>
{

View File

@@ -133,17 +133,6 @@ template<typename MatrixType, int DiagIndex> class Diagonal
return m_matrix.coeff(index+rowOffset(), index+colOffset());
}
const typename internal::remove_all<typename MatrixType::Nested>::type&
nestedExpression() const
{
return m_matrix;
}
int index() const
{
return m_index.value();
}
protected:
const typename MatrixType::Nested m_matrix;
const internal::variable_if_dynamic<Index, DiagIndex> m_index;

View File

@@ -220,38 +220,6 @@ struct functor_traits<scalar_quotient_op<Scalar> > {
};
};
/** \internal
* \brief Template functor to compute the and of two booleans
*
* \sa class CwiseBinaryOp, ArrayBase::operator&&
*/
struct scalar_boolean_and_op {
EIGEN_EMPTY_STRUCT_CTOR(scalar_boolean_and_op)
EIGEN_STRONG_INLINE bool operator() (const bool& a, const bool& b) const { return a && b; }
};
template<> struct functor_traits<scalar_boolean_and_op> {
enum {
Cost = NumTraits<bool>::AddCost,
PacketAccess = false
};
};
/** \internal
* \brief Template functor to compute the or of two booleans
*
* \sa class CwiseBinaryOp, ArrayBase::operator||
*/
struct scalar_boolean_or_op {
EIGEN_EMPTY_STRUCT_CTOR(scalar_boolean_or_op)
EIGEN_STRONG_INLINE bool operator() (const bool& a, const bool& b) const { return a || b; }
};
template<> struct functor_traits<scalar_boolean_or_op> {
enum {
Cost = NumTraits<bool>::AddCost,
PacketAccess = false
};
};
// unary functors:
/** \internal

View File

@@ -94,7 +94,7 @@ struct isMuchSmallerThan_scalar_selector<Derived, true>
*
* \note The fuzzy compares are done multiplicatively. Two vectors \f$ v \f$ and \f$ w \f$
* are considered to be approximately equal within precision \f$ p \f$ if
* \f[ \Vert v - w \Vert \leqslant p\,\min(\Vert v\Vert, \Vert w\Vert). \f]
* \f[ \Vert v - w \Vert \leqslant p\,\(min)(\Vert v\Vert, \Vert w\Vert). \f]
* For matrices, the comparison is done using the Hilbert-Schmidt norm (aka Frobenius norm
* L2 norm).
*

View File

@@ -1,624 +0,0 @@
// This file is part of Eigen, a lightweight C++ template library
// for linear algebra.
//
// Copyright (C) 2006-2008 Benoit Jacob <jacob.benoit.1@gmail.com>
// Copyright (C) 2008-2011 Gael Guennebaud <gael.guennebaud@inria.fr>
//
// Eigen is free software; you can redistribute it and/or
// modify it under the terms of the GNU Lesser General Public
// License as published by the Free Software Foundation; either
// version 3 of the License, or (at your option) any later version.
//
// Alternatively, you can redistribute it and/or
// modify it under the terms of the GNU General Public License as
// published by the Free Software Foundation; either version 2 of
// the License, or (at your option) any later version.
//
// Eigen is distributed in the hope that it will be useful, but WITHOUT ANY
// WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS
// FOR A PARTICULAR PURPOSE. See the GNU Lesser General Public License or the
// GNU General Public License for more details.
//
// You should have received a copy of the GNU Lesser General Public
// License and a copy of the GNU General Public License along with
// Eigen. If not, see <http://www.gnu.org/licenses/>.
#ifndef EIGEN_GENERAL_PRODUCT_H
#define EIGEN_GENERAL_PRODUCT_H
/** \class GeneralProduct
* \ingroup Core_Module
*
* \brief Expression of the product of two general matrices or vectors
*
* \param LhsNested the type used to store the left-hand side
* \param RhsNested the type used to store the right-hand side
* \param ProductMode the type of the product
*
* This class represents an expression of the product of two general matrices.
* We call a general matrix, a dense matrix with full storage. For instance,
* This excludes triangular, selfadjoint, and sparse matrices.
* It is the return type of the operator* between general matrices. Its template
* arguments are determined automatically by ProductReturnType. Therefore,
* GeneralProduct should never be used direclty. To determine the result type of a
* function which involves a matrix product, use ProductReturnType::Type.
*
* \sa ProductReturnType, MatrixBase::operator*(const MatrixBase<OtherDerived>&)
*/
template<typename Lhs, typename Rhs, int ProductType = internal::product_type<Lhs,Rhs>::value>
class GeneralProduct;
enum {
Large = 2,
Small = 3
};
namespace internal {
template<int Rows, int Cols, int Depth> struct product_type_selector;
template<int Size, int MaxSize> struct product_size_category
{
enum { is_large = MaxSize == Dynamic ||
Size >= EIGEN_CACHEFRIENDLY_PRODUCT_THRESHOLD,
value = is_large ? Large
: Size == 1 ? 1
: Small
};
};
template<typename Lhs, typename Rhs> struct product_type
{
typedef typename remove_all<Lhs>::type _Lhs;
typedef typename remove_all<Rhs>::type _Rhs;
enum {
MaxRows = _Lhs::MaxRowsAtCompileTime,
Rows = _Lhs::RowsAtCompileTime,
MaxCols = _Rhs::MaxColsAtCompileTime,
Cols = _Rhs::ColsAtCompileTime,
MaxDepth = EIGEN_SIZE_MIN_PREFER_FIXED(_Lhs::MaxColsAtCompileTime,
_Rhs::MaxRowsAtCompileTime),
Depth = EIGEN_SIZE_MIN_PREFER_FIXED(_Lhs::ColsAtCompileTime,
_Rhs::RowsAtCompileTime),
LargeThreshold = EIGEN_CACHEFRIENDLY_PRODUCT_THRESHOLD
};
// the splitting into different lines of code here, introducing the _select enums and the typedef below,
// is to work around an internal compiler error with gcc 4.1 and 4.2.
private:
enum {
rows_select = product_size_category<Rows,MaxRows>::value,
cols_select = product_size_category<Cols,MaxCols>::value,
depth_select = product_size_category<Depth,MaxDepth>::value
};
typedef product_type_selector<rows_select, cols_select, depth_select> selector;
public:
enum {
value = selector::ret
};
#ifdef EIGEN_DEBUG_PRODUCT
static void debug()
{
EIGEN_DEBUG_VAR(Rows);
EIGEN_DEBUG_VAR(Cols);
EIGEN_DEBUG_VAR(Depth);
EIGEN_DEBUG_VAR(rows_select);
EIGEN_DEBUG_VAR(cols_select);
EIGEN_DEBUG_VAR(depth_select);
EIGEN_DEBUG_VAR(value);
}
#endif
};
/* The following allows to select the kind of product at compile time
* based on the three dimensions of the product.
* This is a compile time mapping from {1,Small,Large}^3 -> {product types} */
// FIXME I'm not sure the current mapping is the ideal one.
template<int M, int N> struct product_type_selector<M,N,1> { enum { ret = OuterProduct }; };
template<int Depth> struct product_type_selector<1, 1, Depth> { enum { ret = InnerProduct }; };
template<> struct product_type_selector<1, 1, 1> { enum { ret = InnerProduct }; };
template<> struct product_type_selector<Small,1, Small> { enum { ret = CoeffBasedProductMode }; };
template<> struct product_type_selector<1, Small,Small> { enum { ret = CoeffBasedProductMode }; };
template<> struct product_type_selector<Small,Small,Small> { enum { ret = CoeffBasedProductMode }; };
template<> struct product_type_selector<Small, Small, 1> { enum { ret = LazyCoeffBasedProductMode }; };
template<> struct product_type_selector<Small, Large, 1> { enum { ret = LazyCoeffBasedProductMode }; };
template<> struct product_type_selector<Large, Small, 1> { enum { ret = LazyCoeffBasedProductMode }; };
template<> struct product_type_selector<1, Large,Small> { enum { ret = CoeffBasedProductMode }; };
template<> struct product_type_selector<1, Large,Large> { enum { ret = GemvProduct }; };
template<> struct product_type_selector<1, Small,Large> { enum { ret = CoeffBasedProductMode }; };
template<> struct product_type_selector<Large,1, Small> { enum { ret = CoeffBasedProductMode }; };
template<> struct product_type_selector<Large,1, Large> { enum { ret = GemvProduct }; };
template<> struct product_type_selector<Small,1, Large> { enum { ret = CoeffBasedProductMode }; };
template<> struct product_type_selector<Small,Small,Large> { enum { ret = GemmProduct }; };
template<> struct product_type_selector<Large,Small,Large> { enum { ret = GemmProduct }; };
template<> struct product_type_selector<Small,Large,Large> { enum { ret = GemmProduct }; };
template<> struct product_type_selector<Large,Large,Large> { enum { ret = GemmProduct }; };
template<> struct product_type_selector<Large,Small,Small> { enum { ret = GemmProduct }; };
template<> struct product_type_selector<Small,Large,Small> { enum { ret = GemmProduct }; };
template<> struct product_type_selector<Large,Large,Small> { enum { ret = GemmProduct }; };
} // end namespace internal
/** \class ProductReturnType
* \ingroup Core_Module
*
* \brief Helper class to get the correct and optimized returned type of operator*
*
* \param Lhs the type of the left-hand side
* \param Rhs the type of the right-hand side
* \param ProductMode the type of the product (determined automatically by internal::product_mode)
*
* This class defines the typename Type representing the optimized product expression
* between two matrix expressions. In practice, using ProductReturnType<Lhs,Rhs>::Type
* is the recommended way to define the result type of a function returning an expression
* which involve a matrix product. The class Product should never be
* used directly.
*
* \sa class Product, MatrixBase::operator*(const MatrixBase<OtherDerived>&)
*/
template<typename Lhs, typename Rhs, int ProductType>
struct ProductReturnType
{
// TODO use the nested type to reduce instanciations ????
// typedef typename internal::nested<Lhs,Rhs::ColsAtCompileTime>::type LhsNested;
// typedef typename internal::nested<Rhs,Lhs::RowsAtCompileTime>::type RhsNested;
typedef GeneralProduct<Lhs/*Nested*/, Rhs/*Nested*/, ProductType> Type;
};
template<typename Lhs, typename Rhs>
struct ProductReturnType<Lhs,Rhs,CoeffBasedProductMode>
{
typedef typename internal::nested<Lhs, Rhs::ColsAtCompileTime, typename internal::plain_matrix_type<Lhs>::type >::type LhsNested;
typedef typename internal::nested<Rhs, Lhs::RowsAtCompileTime, typename internal::plain_matrix_type<Rhs>::type >::type RhsNested;
typedef CoeffBasedProduct<LhsNested, RhsNested, EvalBeforeAssigningBit | EvalBeforeNestingBit> Type;
};
template<typename Lhs, typename Rhs>
struct ProductReturnType<Lhs,Rhs,LazyCoeffBasedProductMode>
{
typedef typename internal::nested<Lhs, Rhs::ColsAtCompileTime, typename internal::plain_matrix_type<Lhs>::type >::type LhsNested;
typedef typename internal::nested<Rhs, Lhs::RowsAtCompileTime, typename internal::plain_matrix_type<Rhs>::type >::type RhsNested;
typedef CoeffBasedProduct<LhsNested, RhsNested, NestByRefBit> Type;
};
// this is a workaround for sun CC
template<typename Lhs, typename Rhs>
struct LazyProductReturnType : public ProductReturnType<Lhs,Rhs,LazyCoeffBasedProductMode>
{};
/***********************************************************************
* Implementation of Inner Vector Vector Product
***********************************************************************/
// FIXME : maybe the "inner product" could return a Scalar
// instead of a 1x1 matrix ??
// Pro: more natural for the user
// Cons: this could be a problem if in a meta unrolled algorithm a matrix-matrix
// product ends up to a row-vector times col-vector product... To tackle this use
// case, we could have a specialization for Block<MatrixType,1,1> with: operator=(Scalar x);
namespace internal {
template<typename Lhs, typename Rhs>
struct traits<GeneralProduct<Lhs,Rhs,InnerProduct> >
: traits<Matrix<typename scalar_product_traits<typename Lhs::Scalar, typename Rhs::Scalar>::ReturnType,1,1> >
{};
}
template<typename Lhs, typename Rhs>
class GeneralProduct<Lhs, Rhs, InnerProduct>
: internal::no_assignment_operator,
public Matrix<typename internal::scalar_product_traits<typename Lhs::Scalar, typename Rhs::Scalar>::ReturnType,1,1>
{
typedef Matrix<typename internal::scalar_product_traits<typename Lhs::Scalar, typename Rhs::Scalar>::ReturnType,1,1> Base;
public:
GeneralProduct(const Lhs& lhs, const Rhs& rhs)
{
EIGEN_STATIC_ASSERT((internal::is_same<typename Lhs::RealScalar, typename Rhs::RealScalar>::value),
YOU_MIXED_DIFFERENT_NUMERIC_TYPES__YOU_NEED_TO_USE_THE_CAST_METHOD_OF_MATRIXBASE_TO_CAST_NUMERIC_TYPES_EXPLICITLY)
Base::coeffRef(0,0) = (lhs.transpose().cwiseProduct(rhs)).sum();
}
/** Convertion to scalar */
operator const typename Base::Scalar() const {
return Base::coeff(0,0);
}
};
/***********************************************************************
* Implementation of Outer Vector Vector Product
***********************************************************************/
namespace internal {
template<int StorageOrder> struct outer_product_selector;
template<typename Lhs, typename Rhs>
struct traits<GeneralProduct<Lhs,Rhs,OuterProduct> >
: traits<ProductBase<GeneralProduct<Lhs,Rhs,OuterProduct>, Lhs, Rhs> >
{};
}
template<typename Lhs, typename Rhs>
class GeneralProduct<Lhs, Rhs, OuterProduct>
: public ProductBase<GeneralProduct<Lhs,Rhs,OuterProduct>, Lhs, Rhs>
{
public:
EIGEN_PRODUCT_PUBLIC_INTERFACE(GeneralProduct)
GeneralProduct(const Lhs& lhs, const Rhs& rhs) : Base(lhs,rhs)
{
EIGEN_STATIC_ASSERT((internal::is_same<typename Lhs::RealScalar, typename Rhs::RealScalar>::value),
YOU_MIXED_DIFFERENT_NUMERIC_TYPES__YOU_NEED_TO_USE_THE_CAST_METHOD_OF_MATRIXBASE_TO_CAST_NUMERIC_TYPES_EXPLICITLY)
}
template<typename Dest> void scaleAndAddTo(Dest& dest, Scalar alpha) const
{
internal::outer_product_selector<(int(Dest::Flags)&RowMajorBit) ? RowMajor : ColMajor>::run(*this, dest, alpha);
}
};
namespace internal {
template<> struct outer_product_selector<ColMajor> {
template<typename ProductType, typename Dest>
static EIGEN_DONT_INLINE void run(const ProductType& prod, Dest& dest, typename ProductType::Scalar alpha) {
typedef typename Dest::Index Index;
// FIXME make sure lhs is sequentially stored
// FIXME not very good if rhs is real and lhs complex while alpha is real too
const Index cols = dest.cols();
for (Index j=0; j<cols; ++j)
dest.col(j) += (alpha * prod.rhs().coeff(j)) * prod.lhs();
}
};
template<> struct outer_product_selector<RowMajor> {
template<typename ProductType, typename Dest>
static EIGEN_DONT_INLINE void run(const ProductType& prod, Dest& dest, typename ProductType::Scalar alpha) {
typedef typename Dest::Index Index;
// FIXME make sure rhs is sequentially stored
// FIXME not very good if lhs is real and rhs complex while alpha is real too
const Index rows = dest.rows();
for (Index i=0; i<rows; ++i)
dest.row(i) += (alpha * prod.lhs().coeff(i)) * prod.rhs();
}
};
} // end namespace internal
/***********************************************************************
* Implementation of General Matrix Vector Product
***********************************************************************/
/* According to the shape/flags of the matrix we have to distinghish 3 different cases:
* 1 - the matrix is col-major, BLAS compatible and M is large => call fast BLAS-like colmajor routine
* 2 - the matrix is row-major, BLAS compatible and N is large => call fast BLAS-like rowmajor routine
* 3 - all other cases are handled using a simple loop along the outer-storage direction.
* Therefore we need a lower level meta selector.
* Furthermore, if the matrix is the rhs, then the product has to be transposed.
*/
namespace internal {
template<typename Lhs, typename Rhs>
struct traits<GeneralProduct<Lhs,Rhs,GemvProduct> >
: traits<ProductBase<GeneralProduct<Lhs,Rhs,GemvProduct>, Lhs, Rhs> >
{};
template<int Side, int StorageOrder, bool BlasCompatible>
struct gemv_selector;
} // end namespace internal
template<typename Lhs, typename Rhs>
class GeneralProduct<Lhs, Rhs, GemvProduct>
: public ProductBase<GeneralProduct<Lhs,Rhs,GemvProduct>, Lhs, Rhs>
{
public:
EIGEN_PRODUCT_PUBLIC_INTERFACE(GeneralProduct)
typedef typename Lhs::Scalar LhsScalar;
typedef typename Rhs::Scalar RhsScalar;
GeneralProduct(const Lhs& lhs, const Rhs& rhs) : Base(lhs,rhs)
{
// EIGEN_STATIC_ASSERT((internal::is_same<typename Lhs::Scalar, typename Rhs::Scalar>::value),
// YOU_MIXED_DIFFERENT_NUMERIC_TYPES__YOU_NEED_TO_USE_THE_CAST_METHOD_OF_MATRIXBASE_TO_CAST_NUMERIC_TYPES_EXPLICITLY)
}
enum { Side = Lhs::IsVectorAtCompileTime ? OnTheLeft : OnTheRight };
typedef typename internal::conditional<int(Side)==OnTheRight,_LhsNested,_RhsNested>::type MatrixType;
template<typename Dest> void scaleAndAddTo(Dest& dst, Scalar alpha) const
{
eigen_assert(m_lhs.rows() == dst.rows() && m_rhs.cols() == dst.cols());
internal::gemv_selector<Side,(int(MatrixType::Flags)&RowMajorBit) ? RowMajor : ColMajor,
bool(internal::blas_traits<MatrixType>::HasUsableDirectAccess)>::run(*this, dst, alpha);
}
};
namespace internal {
// The vector is on the left => transposition
template<int StorageOrder, bool BlasCompatible>
struct gemv_selector<OnTheLeft,StorageOrder,BlasCompatible>
{
template<typename ProductType, typename Dest>
static void run(const ProductType& prod, Dest& dest, typename ProductType::Scalar alpha)
{
Transpose<Dest> destT(dest);
enum { OtherStorageOrder = StorageOrder == RowMajor ? ColMajor : RowMajor };
gemv_selector<OnTheRight,OtherStorageOrder,BlasCompatible>
::run(GeneralProduct<Transpose<const typename ProductType::_RhsNested>,Transpose<const typename ProductType::_LhsNested>, GemvProduct>
(prod.rhs().transpose(), prod.lhs().transpose()), destT, alpha);
}
};
template<typename Scalar,int Size,int MaxSize,bool Cond> struct gemv_static_vector_if;
template<typename Scalar,int Size,int MaxSize>
struct gemv_static_vector_if<Scalar,Size,MaxSize,false>
{
EIGEN_STRONG_INLINE Scalar* data() { eigen_internal_assert(false && "should never be called"); return 0; }
};
template<typename Scalar,int Size>
struct gemv_static_vector_if<Scalar,Size,Dynamic,true>
{
EIGEN_STRONG_INLINE Scalar* data() { return 0; }
};
template<typename Scalar,int Size,int MaxSize>
struct gemv_static_vector_if<Scalar,Size,MaxSize,true>
{
#if EIGEN_ALIGN_STATICALLY
internal::plain_array<Scalar,EIGEN_SIZE_MIN_PREFER_FIXED(Size,MaxSize),0> m_data;
EIGEN_STRONG_INLINE Scalar* data() { return m_data.array; }
#else
// Some architectures cannot align on the stack,
// => let's manually enforce alignment by allocating more data and return the address of the first aligned element.
enum {
ForceAlignment = internal::packet_traits<Scalar>::Vectorizable,
PacketSize = internal::packet_traits<Scalar>::size
};
internal::plain_array<Scalar,EIGEN_SIZE_MIN_PREFER_FIXED(Size,MaxSize)+(ForceAlignment?PacketSize:0),0> m_data;
EIGEN_STRONG_INLINE Scalar* data() {
return ForceAlignment
? reinterpret_cast<Scalar*>((reinterpret_cast<size_t>(m_data.array) & ~(size_t(15))) + 16)
: m_data.array;
}
#endif
};
template<> struct gemv_selector<OnTheRight,ColMajor,true>
{
template<typename ProductType, typename Dest>
static inline void run(const ProductType& prod, Dest& dest, typename ProductType::Scalar alpha)
{
typedef typename ProductType::Index Index;
typedef typename ProductType::LhsScalar LhsScalar;
typedef typename ProductType::RhsScalar RhsScalar;
typedef typename ProductType::Scalar ResScalar;
typedef typename ProductType::RealScalar RealScalar;
typedef typename ProductType::ActualLhsType ActualLhsType;
typedef typename ProductType::ActualRhsType ActualRhsType;
typedef typename ProductType::LhsBlasTraits LhsBlasTraits;
typedef typename ProductType::RhsBlasTraits RhsBlasTraits;
typedef Map<Matrix<ResScalar,Dynamic,1>, Aligned> MappedDest;
const ActualLhsType actualLhs = LhsBlasTraits::extract(prod.lhs());
const ActualRhsType actualRhs = RhsBlasTraits::extract(prod.rhs());
ResScalar actualAlpha = alpha * LhsBlasTraits::extractScalarFactor(prod.lhs())
* RhsBlasTraits::extractScalarFactor(prod.rhs());
enum {
// FIXME find a way to allow an inner stride on the result if packet_traits<Scalar>::size==1
// on, the other hand it is good for the cache to pack the vector anyways...
EvalToDestAtCompileTime = Dest::InnerStrideAtCompileTime==1,
ComplexByReal = (NumTraits<LhsScalar>::IsComplex) && (!NumTraits<RhsScalar>::IsComplex),
MightCannotUseDest = (Dest::InnerStrideAtCompileTime!=1) || ComplexByReal
};
gemv_static_vector_if<ResScalar,Dest::SizeAtCompileTime,Dest::MaxSizeAtCompileTime,MightCannotUseDest> static_dest;
bool alphaIsCompatible = (!ComplexByReal) || (imag(actualAlpha)==RealScalar(0));
bool evalToDest = EvalToDestAtCompileTime && alphaIsCompatible;
RhsScalar compatibleAlpha = get_factor<ResScalar,RhsScalar>::run(actualAlpha);
ei_declare_aligned_stack_constructed_variable(ResScalar,actualDestPtr,dest.size(),
evalToDest ? dest.data() : static_dest.data());
if(!evalToDest)
{
#ifdef EIGEN_DENSE_STORAGE_CTOR_PLUGIN
int size = dest.size();
EIGEN_DENSE_STORAGE_CTOR_PLUGIN
#endif
if(!alphaIsCompatible)
{
MappedDest(actualDestPtr, dest.size()).setZero();
compatibleAlpha = RhsScalar(1);
}
else
MappedDest(actualDestPtr, dest.size()) = dest;
}
general_matrix_vector_product
<Index,LhsScalar,ColMajor,LhsBlasTraits::NeedToConjugate,RhsScalar,RhsBlasTraits::NeedToConjugate>::run(
actualLhs.rows(), actualLhs.cols(),
&actualLhs.coeffRef(0,0), actualLhs.outerStride(),
actualRhs.data(), actualRhs.innerStride(),
actualDestPtr, 1,
compatibleAlpha);
if (!evalToDest)
{
if(!alphaIsCompatible)
dest += actualAlpha * MappedDest(actualDestPtr, dest.size());
else
dest = MappedDest(actualDestPtr, dest.size());
}
}
};
template<> struct gemv_selector<OnTheRight,RowMajor,true>
{
template<typename ProductType, typename Dest>
static void run(const ProductType& prod, Dest& dest, typename ProductType::Scalar alpha)
{
typedef typename ProductType::LhsScalar LhsScalar;
typedef typename ProductType::RhsScalar RhsScalar;
typedef typename ProductType::Scalar ResScalar;
typedef typename ProductType::Index Index;
typedef typename ProductType::ActualLhsType ActualLhsType;
typedef typename ProductType::ActualRhsType ActualRhsType;
typedef typename ProductType::_ActualRhsType _ActualRhsType;
typedef typename ProductType::LhsBlasTraits LhsBlasTraits;
typedef typename ProductType::RhsBlasTraits RhsBlasTraits;
typename add_const<ActualLhsType>::type actualLhs = LhsBlasTraits::extract(prod.lhs());
typename add_const<ActualRhsType>::type actualRhs = RhsBlasTraits::extract(prod.rhs());
ResScalar actualAlpha = alpha * LhsBlasTraits::extractScalarFactor(prod.lhs())
* RhsBlasTraits::extractScalarFactor(prod.rhs());
enum {
// FIXME find a way to allow an inner stride on the result if packet_traits<Scalar>::size==1
// on, the other hand it is good for the cache to pack the vector anyways...
DirectlyUseRhs = _ActualRhsType::InnerStrideAtCompileTime==1
};
gemv_static_vector_if<RhsScalar,_ActualRhsType::SizeAtCompileTime,_ActualRhsType::MaxSizeAtCompileTime,!DirectlyUseRhs> static_rhs;
ei_declare_aligned_stack_constructed_variable(RhsScalar,actualRhsPtr,actualRhs.size(),
DirectlyUseRhs ? const_cast<RhsScalar*>(actualRhs.data()) : static_rhs.data());
if(!DirectlyUseRhs)
{
#ifdef EIGEN_DENSE_STORAGE_CTOR_PLUGIN
int size = actualRhs.size();
EIGEN_DENSE_STORAGE_CTOR_PLUGIN
#endif
Map<typename _ActualRhsType::PlainObject>(actualRhsPtr, actualRhs.size()) = actualRhs;
}
general_matrix_vector_product
<Index,LhsScalar,RowMajor,LhsBlasTraits::NeedToConjugate,RhsScalar,RhsBlasTraits::NeedToConjugate>::run(
actualLhs.rows(), actualLhs.cols(),
&actualLhs.coeffRef(0,0), actualLhs.outerStride(),
actualRhsPtr, 1,
&dest.coeffRef(0,0), dest.innerStride(),
actualAlpha);
}
};
template<> struct gemv_selector<OnTheRight,ColMajor,false>
{
template<typename ProductType, typename Dest>
static void run(const ProductType& prod, Dest& dest, typename ProductType::Scalar alpha)
{
typedef typename Dest::Index Index;
// TODO makes sure dest is sequentially stored in memory, otherwise use a temp
const Index size = prod.rhs().rows();
for(Index k=0; k<size; ++k)
dest += (alpha*prod.rhs().coeff(k)) * prod.lhs().col(k);
}
};
template<> struct gemv_selector<OnTheRight,RowMajor,false>
{
template<typename ProductType, typename Dest>
static void run(const ProductType& prod, Dest& dest, typename ProductType::Scalar alpha)
{
typedef typename Dest::Index Index;
// TODO makes sure rhs is sequentially stored in memory, otherwise use a temp
const Index rows = prod.rows();
for(Index i=0; i<rows; ++i)
dest.coeffRef(i) += alpha * (prod.lhs().row(i).cwiseProduct(prod.rhs().transpose())).sum();
}
};
} // end namespace internal
/***************************************************************************
* Implementation of matrix base methods
***************************************************************************/
/** \returns the matrix product of \c *this and \a other.
*
* \note If instead of the matrix product you want the coefficient-wise product, see Cwise::operator*().
*
* \sa lazyProduct(), operator*=(const MatrixBase&), Cwise::operator*()
*/
template<typename Derived>
template<typename OtherDerived>
inline const typename ProductReturnType<Derived,OtherDerived>::Type
MatrixBase<Derived>::operator*(const MatrixBase<OtherDerived> &other) const
{
// A note regarding the function declaration: In MSVC, this function will sometimes
// not be inlined since DenseStorage is an unwindable object for dynamic
// matrices and product types are holding a member to store the result.
// Thus it does not help tagging this function with EIGEN_STRONG_INLINE.
enum {
ProductIsValid = Derived::ColsAtCompileTime==Dynamic
|| OtherDerived::RowsAtCompileTime==Dynamic
|| int(Derived::ColsAtCompileTime)==int(OtherDerived::RowsAtCompileTime),
AreVectors = Derived::IsVectorAtCompileTime && OtherDerived::IsVectorAtCompileTime,
SameSizes = EIGEN_PREDICATE_SAME_MATRIX_SIZE(Derived,OtherDerived)
};
// note to the lost user:
// * for a dot product use: v1.dot(v2)
// * for a coeff-wise product use: v1.cwiseProduct(v2)
EIGEN_STATIC_ASSERT(ProductIsValid || !(AreVectors && SameSizes),
INVALID_VECTOR_VECTOR_PRODUCT__IF_YOU_WANTED_A_DOT_OR_COEFF_WISE_PRODUCT_YOU_MUST_USE_THE_EXPLICIT_FUNCTIONS)
EIGEN_STATIC_ASSERT(ProductIsValid || !(SameSizes && !AreVectors),
INVALID_MATRIX_PRODUCT__IF_YOU_WANTED_A_COEFF_WISE_PRODUCT_YOU_MUST_USE_THE_EXPLICIT_FUNCTION)
EIGEN_STATIC_ASSERT(ProductIsValid || SameSizes, INVALID_MATRIX_PRODUCT)
#ifdef EIGEN_DEBUG_PRODUCT
internal::product_type<Derived,OtherDerived>::debug();
#endif
return typename ProductReturnType<Derived,OtherDerived>::Type(derived(), other.derived());
}
/** \returns an expression of the matrix product of \c *this and \a other without implicit evaluation.
*
* The returned product will behave like any other expressions: the coefficients of the product will be
* computed once at a time as requested. This might be useful in some extremely rare cases when only
* a small and no coherent fraction of the result's coefficients have to be computed.
*
* \warning This version of the matrix product can be much much slower. So use it only if you know
* what you are doing and that you measured a true speed improvement.
*
* \sa operator*(const MatrixBase&)
*/
template<typename Derived>
template<typename OtherDerived>
const typename LazyProductReturnType<Derived,OtherDerived>::Type
MatrixBase<Derived>::lazyProduct(const MatrixBase<OtherDerived> &other) const
{
enum {
ProductIsValid = Derived::ColsAtCompileTime==Dynamic
|| OtherDerived::RowsAtCompileTime==Dynamic
|| int(Derived::ColsAtCompileTime)==int(OtherDerived::RowsAtCompileTime),
AreVectors = Derived::IsVectorAtCompileTime && OtherDerived::IsVectorAtCompileTime,
SameSizes = EIGEN_PREDICATE_SAME_MATRIX_SIZE(Derived,OtherDerived)
};
// note to the lost user:
// * for a dot product use: v1.dot(v2)
// * for a coeff-wise product use: v1.cwiseProduct(v2)
EIGEN_STATIC_ASSERT(ProductIsValid || !(AreVectors && SameSizes),
INVALID_VECTOR_VECTOR_PRODUCT__IF_YOU_WANTED_A_DOT_OR_COEFF_WISE_PRODUCT_YOU_MUST_USE_THE_EXPLICIT_FUNCTIONS)
EIGEN_STATIC_ASSERT(ProductIsValid || !(SameSizes && !AreVectors),
INVALID_MATRIX_PRODUCT__IF_YOU_WANTED_A_COEFF_WISE_PRODUCT_YOU_MUST_USE_THE_EXPLICIT_FUNCTION)
EIGEN_STATIC_ASSERT(ProductIsValid || SameSizes, INVALID_MATRIX_PRODUCT)
return typename LazyProductReturnType<Derived,OtherDerived>::Type(derived(), other.derived());
}
#endif // EIGEN_PRODUCT_H

View File

@@ -34,7 +34,7 @@
* \tparam PlainObjectType the equivalent matrix type of the mapped data
* \tparam MapOptions specifies whether the pointer is \c #Aligned, or \c #Unaligned.
* The default is \c #Unaligned.
* \tparam StrideType optionally specifies strides. By default, Map assumes the memory layout
* \tparam StrideType optionnally specifies strides. By default, Map assumes the memory layout
* of an ordinary, contiguous array. This can be overridden by specifying strides.
* The type passed here must be a specialization of the Stride template, see examples below.
*
@@ -72,9 +72,9 @@
* Example: \include Map_placement_new.cpp
* Output: \verbinclude Map_placement_new.out
*
* This class is the return type of PlainObjectBase::Map() but can also be used directly.
* This class is the return type of Matrix::Map() but can also be used directly.
*
* \sa PlainObjectBase::Map(), \ref TopicStorageOrders
* \sa Matrix::Map(), \ref TopicStorageOrders
*/
namespace internal {

View File

@@ -170,8 +170,8 @@ template<typename Derived> class MapBase<Derived, ReadOnlyAccessors>
EIGEN_STATIC_ASSERT(EIGEN_IMPLIES(internal::traits<Derived>::Flags&PacketAccessBit,
internal::inner_stride_at_compile_time<Derived>::ret==1),
PACKET_ACCESS_REQUIRES_TO_HAVE_INNER_STRIDE_FIXED_TO_1);
eigen_assert(EIGEN_IMPLIES(internal::traits<Derived>::Flags&AlignedBit, (size_t(m_data) % 16) == 0)
&& "data is not aligned");
eigen_assert(EIGEN_IMPLIES(internal::traits<Derived>::Flags&AlignedBit, (size_t(m_data) % (sizeof(Scalar)*internal::packet_traits<Scalar>::size)) == 0)
&& "data is not aligned");
}
PointerType m_data;

View File

@@ -309,7 +309,8 @@ struct abs2_impl<std::complex<RealScalar> >
{
static inline RealScalar run(const std::complex<RealScalar>& x)
{
return real(x)*real(x) + imag(x)*imag(x);
using std::norm;
return norm(x);
}
};

View File

@@ -153,6 +153,10 @@ class Matrix
typedef typename Base::PlainObject PlainObject;
enum { NeedsToAlign = (!(Options&DontAlign))
&& SizeAtCompileTime!=Dynamic && ((static_cast<int>(sizeof(Scalar))*SizeAtCompileTime)%16)==0 };
EIGEN_MAKE_ALIGNED_OPERATOR_NEW_IF(NeedsToAlign)
using Base::base;
using Base::coeffRef;
@@ -411,6 +415,25 @@ EIGEN_MAKE_TYPEDEFS_ALL_SIZES(std::complex<double>, cd)
#undef EIGEN_MAKE_TYPEDEFS_ALL_SIZES
#undef EIGEN_MAKE_TYPEDEFS
#undef EIGEN_MAKE_FIXED_TYPEDEFS
#undef EIGEN_MAKE_TYPEDEFS_LARGE
#define EIGEN_USING_MATRIX_TYPEDEFS_FOR_TYPE_AND_SIZE(TypeSuffix, SizeSuffix) \
using Eigen::Matrix##SizeSuffix##TypeSuffix; \
using Eigen::Vector##SizeSuffix##TypeSuffix; \
using Eigen::RowVector##SizeSuffix##TypeSuffix;
#define EIGEN_USING_MATRIX_TYPEDEFS_FOR_TYPE(TypeSuffix) \
EIGEN_USING_MATRIX_TYPEDEFS_FOR_TYPE_AND_SIZE(TypeSuffix, 2) \
EIGEN_USING_MATRIX_TYPEDEFS_FOR_TYPE_AND_SIZE(TypeSuffix, 3) \
EIGEN_USING_MATRIX_TYPEDEFS_FOR_TYPE_AND_SIZE(TypeSuffix, 4) \
EIGEN_USING_MATRIX_TYPEDEFS_FOR_TYPE_AND_SIZE(TypeSuffix, X) \
#define EIGEN_USING_MATRIX_TYPEDEFS \
EIGEN_USING_MATRIX_TYPEDEFS_FOR_TYPE(i) \
EIGEN_USING_MATRIX_TYPEDEFS_FOR_TYPE(f) \
EIGEN_USING_MATRIX_TYPEDEFS_FOR_TYPE(d) \
EIGEN_USING_MATRIX_TYPEDEFS_FOR_TYPE(cf) \
EIGEN_USING_MATRIX_TYPEDEFS_FOR_TYPE(cd)
#endif // EIGEN_MATRIX_H

View File

@@ -465,8 +465,6 @@ template<typename Derived> class MatrixBase
const MatrixFunctionReturnValue<Derived> sinh() const;
const MatrixFunctionReturnValue<Derived> cos() const;
const MatrixFunctionReturnValue<Derived> sin() const;
const MatrixSquareRootReturnValue<Derived> sqrt() const;
const MatrixLogarithmReturnValue<Derived> log() const;
#ifdef EIGEN2_SUPPORT
template<typename ProductDerived, typename Lhs, typename Rhs>

View File

@@ -34,19 +34,6 @@
namespace internal {
template<typename Index>
EIGEN_ALWAYS_INLINE void check_rows_cols_for_overflow(Index rows, Index cols)
{
// http://hg.mozilla.org/mozilla-central/file/6c8a909977d3/xpcom/ds/CheckedInt.h#l242
// we assume Index is signed
Index max_index = (size_t(1) << (8 * sizeof(Index) - 1)) - 1; // assume Index is signed
bool error = (rows < 0 || cols < 0) ? true
: (rows == 0 || cols == 0) ? false
: (rows > max_index / cols);
if (error)
throw_std_bad_alloc();
}
template <typename Derived, typename OtherDerived = Derived, bool IsVector = static_cast<bool>(Derived::IsVectorAtCompileTime)> struct conservative_resize_like_impl;
template<typename MatrixTypeA, typename MatrixTypeB, bool SwapPointers> struct matrix_swap_impl;
@@ -97,12 +84,14 @@ class PlainObjectBase : public internal::dense_xpr_base<Derived>::type
template<typename StrideType> struct StridedConstMapType { typedef Eigen::Map<const Derived, Unaligned, StrideType> type; };
template<typename StrideType> struct StridedAlignedMapType { typedef Eigen::Map<Derived, Aligned, StrideType> type; };
template<typename StrideType> struct StridedConstAlignedMapType { typedef Eigen::Map<const Derived, Aligned, StrideType> type; };
protected:
DenseStorage<Scalar, Base::MaxSizeAtCompileTime, Base::RowsAtCompileTime, Base::ColsAtCompileTime, Options> m_storage;
public:
enum { NeedsToAlign = SizeAtCompileTime != Dynamic && (internal::traits<Derived>::Flags & AlignedBit) != 0 };
enum { NeedsToAlign = (!(Options&DontAlign))
&& SizeAtCompileTime!=Dynamic && ((static_cast<int>(sizeof(Scalar))*SizeAtCompileTime)%16)==0 };
EIGEN_MAKE_ALIGNED_OPERATOR_NEW_IF(NeedsToAlign)
Base& base() { return *static_cast<Base*>(this); }
@@ -211,13 +200,11 @@ class PlainObjectBase : public internal::dense_xpr_base<Derived>::type
EIGEN_STRONG_INLINE void resize(Index rows, Index cols)
{
#ifdef EIGEN_INITIALIZE_MATRICES_BY_ZERO
internal::check_rows_cols_for_overflow(rows, cols);
Index size = rows*cols;
bool size_changed = size != this->size();
m_storage.resize(size, rows, cols);
if(size_changed) EIGEN_INITIALIZE_BY_ZERO_IF_THAT_OPTION_IS_ENABLED
#else
internal::check_rows_cols_for_overflow(rows, cols);
m_storage.resize(rows*cols, rows, cols);
#endif
}
@@ -286,7 +273,6 @@ class PlainObjectBase : public internal::dense_xpr_base<Derived>::type
EIGEN_STRONG_INLINE void resizeLike(const EigenBase<OtherDerived>& _other)
{
const OtherDerived& other = _other.derived();
internal::check_rows_cols_for_overflow(other.rows(), other.cols());
const Index othersize = other.rows()*other.cols();
if(RowsAtCompileTime == 1)
{
@@ -431,7 +417,6 @@ class PlainObjectBase : public internal::dense_xpr_base<Derived>::type
: m_storage(other.derived().rows() * other.derived().cols(), other.derived().rows(), other.derived().cols())
{
_check_template_params();
internal::check_rows_cols_for_overflow(other.derived().rows(), other.derived().cols());
Base::operator=(other.derived());
}
@@ -440,6 +425,9 @@ class PlainObjectBase : public internal::dense_xpr_base<Derived>::type
* while the AlignedMap() functions return aligned Map objects and thus should be called only with 16-byte-aligned
* \a data pointers.
*
* These methods do not allow to specify strides. If you need to specify strides, you have to
* use the Map class directly.
*
* \see class Map
*/
//@{
@@ -594,12 +582,8 @@ class PlainObjectBase : public internal::dense_xpr_base<Derived>::type
template<typename T0, typename T1>
EIGEN_STRONG_INLINE void _init2(Index rows, Index cols, typename internal::enable_if<Base::SizeAtCompileTime!=2,T0>::type* = 0)
{
EIGEN_STATIC_ASSERT(bool(NumTraits<T0>::IsInteger) &&
bool(NumTraits<T1>::IsInteger),
FLOATING_POINT_ARGUMENT_PASSED__INTEGER_WAS_EXPECTED)
eigen_assert(rows >= 0 && (RowsAtCompileTime == Dynamic || RowsAtCompileTime == rows)
&& cols >= 0 && (ColsAtCompileTime == Dynamic || ColsAtCompileTime == cols));
internal::check_rows_cols_for_overflow(rows, cols);
m_storage.resize(rows*cols,rows,cols);
EIGEN_INITIALIZE_BY_ZERO_IF_THAT_OPTION_IS_ENABLED
}
@@ -657,7 +641,6 @@ struct internal::conservative_resize_like_impl
if ( ( Derived::IsRowMajor && _this.cols() == cols) || // row-major and we change only the number of rows
(!Derived::IsRowMajor && _this.rows() == rows) ) // column-major and we change only the number of columns
{
internal::check_rows_cols_for_overflow(rows, cols);
_this.derived().m_storage.conservativeResize(rows*cols,rows,cols);
}
else

View File

@@ -1,7 +1,8 @@
// This file is part of Eigen, a lightweight C++ template library
// for linear algebra.
//
// Copyright (C) 2008-2011 Gael Guennebaud <gael.guennebaud@inria.fr>
// Copyright (C) 2006-2008 Benoit Jacob <jacob.benoit.1@gmail.com>
// Copyright (C) 2008 Gael Guennebaud <gael.guennebaud@inria.fr>
//
// Eigen is free software; you can redistribute it and/or
// modify it under the terms of the GNU Lesser General Public
@@ -25,103 +26,600 @@
#ifndef EIGEN_PRODUCT_H
#define EIGEN_PRODUCT_H
template<typename Lhs, typename Rhs> class Product;
template<typename Lhs, typename Rhs, typename StorageKind> class ProductImpl;
/** \class Product
/** \class GeneralProduct
* \ingroup Core_Module
*
* \brief Expression of the product of two arbitrary matrices or vectors
* \brief Expression of the product of two general matrices or vectors
*
* \param Lhs the type of the left-hand side expression
* \param Rhs the type of the right-hand side expression
* \param LhsNested the type used to store the left-hand side
* \param RhsNested the type used to store the right-hand side
* \param ProductMode the type of the product
*
* This class represents an expression of the product of two arbitrary matrices.
* This class represents an expression of the product of two general matrices.
* We call a general matrix, a dense matrix with full storage. For instance,
* This excludes triangular, selfadjoint, and sparse matrices.
* It is the return type of the operator* between general matrices. Its template
* arguments are determined automatically by ProductReturnType. Therefore,
* GeneralProduct should never be used direclty. To determine the result type of a
* function which involves a matrix product, use ProductReturnType::Type.
*
* \sa ProductReturnType, MatrixBase::operator*(const MatrixBase<OtherDerived>&)
*/
template<typename Lhs, typename Rhs, int ProductType = internal::product_type<Lhs,Rhs>::value>
class GeneralProduct;
enum {
Large = 2,
Small = 3
};
namespace internal {
template<typename Lhs, typename Rhs>
struct traits<Product<Lhs, Rhs> >
template<int Rows, int Cols, int Depth> struct product_type_selector;
template<int Size, int MaxSize> struct product_size_category
{
typedef MatrixXpr XprKind;
typedef typename remove_all<Lhs>::type LhsCleaned;
typedef typename remove_all<Rhs>::type RhsCleaned;
typedef typename scalar_product_traits<typename traits<LhsCleaned>::Scalar, typename traits<RhsCleaned>::Scalar>::ReturnType Scalar;
typedef typename promote_storage_type<typename traits<LhsCleaned>::StorageKind,
typename traits<RhsCleaned>::StorageKind>::ret StorageKind;
typedef typename promote_index_type<typename traits<LhsCleaned>::Index,
typename traits<RhsCleaned>::Index>::type Index;
enum {
RowsAtCompileTime = LhsCleaned::RowsAtCompileTime,
ColsAtCompileTime = RhsCleaned::ColsAtCompileTime,
MaxRowsAtCompileTime = LhsCleaned::MaxRowsAtCompileTime,
MaxColsAtCompileTime = RhsCleaned::MaxColsAtCompileTime,
Flags = (MaxRowsAtCompileTime==1 ? RowMajorBit : 0), // TODO should be no storage order
CoeffReadCost = 0 // TODO CoeffReadCost should not be part of the expression traits
enum { is_large = MaxSize == Dynamic ||
Size >= EIGEN_CACHEFRIENDLY_PRODUCT_THRESHOLD,
value = is_large ? Large
: Size == 1 ? 1
: Small
};
};
template<typename Lhs, typename Rhs> struct product_type
{
typedef typename remove_all<Lhs>::type _Lhs;
typedef typename remove_all<Rhs>::type _Rhs;
enum {
MaxRows = _Lhs::MaxRowsAtCompileTime,
Rows = _Lhs::RowsAtCompileTime,
MaxCols = _Rhs::MaxColsAtCompileTime,
Cols = _Rhs::ColsAtCompileTime,
MaxDepth = EIGEN_SIZE_MIN_PREFER_FIXED(_Lhs::MaxColsAtCompileTime,
_Rhs::MaxRowsAtCompileTime),
Depth = EIGEN_SIZE_MIN_PREFER_FIXED(_Lhs::ColsAtCompileTime,
_Rhs::RowsAtCompileTime),
LargeThreshold = EIGEN_CACHEFRIENDLY_PRODUCT_THRESHOLD
};
// the splitting into different lines of code here, introducing the _select enums and the typedef below,
// is to work around an internal compiler error with gcc 4.1 and 4.2.
private:
enum {
rows_select = product_size_category<Rows,MaxRows>::value,
cols_select = product_size_category<Cols,MaxCols>::value,
depth_select = product_size_category<Depth,MaxDepth>::value
};
typedef product_type_selector<rows_select, cols_select, depth_select> selector;
public:
enum {
value = selector::ret
};
#ifdef EIGEN_DEBUG_PRODUCT
static void debug()
{
EIGEN_DEBUG_VAR(Rows);
EIGEN_DEBUG_VAR(Cols);
EIGEN_DEBUG_VAR(Depth);
EIGEN_DEBUG_VAR(rows_select);
EIGEN_DEBUG_VAR(cols_select);
EIGEN_DEBUG_VAR(depth_select);
EIGEN_DEBUG_VAR(value);
}
#endif
};
/* The following allows to select the kind of product at compile time
* based on the three dimensions of the product.
* This is a compile time mapping from {1,Small,Large}^3 -> {product types} */
// FIXME I'm not sure the current mapping is the ideal one.
template<int M, int N> struct product_type_selector<M,N,1> { enum { ret = OuterProduct }; };
template<int Depth> struct product_type_selector<1, 1, Depth> { enum { ret = InnerProduct }; };
template<> struct product_type_selector<1, 1, 1> { enum { ret = InnerProduct }; };
template<> struct product_type_selector<Small,1, Small> { enum { ret = CoeffBasedProductMode }; };
template<> struct product_type_selector<1, Small,Small> { enum { ret = CoeffBasedProductMode }; };
template<> struct product_type_selector<Small,Small,Small> { enum { ret = CoeffBasedProductMode }; };
template<> struct product_type_selector<Small, Small, 1> { enum { ret = LazyCoeffBasedProductMode }; };
template<> struct product_type_selector<Small, Large, 1> { enum { ret = LazyCoeffBasedProductMode }; };
template<> struct product_type_selector<Large, Small, 1> { enum { ret = LazyCoeffBasedProductMode }; };
template<> struct product_type_selector<1, Large,Small> { enum { ret = CoeffBasedProductMode }; };
template<> struct product_type_selector<1, Large,Large> { enum { ret = GemvProduct }; };
template<> struct product_type_selector<1, Small,Large> { enum { ret = CoeffBasedProductMode }; };
template<> struct product_type_selector<Large,1, Small> { enum { ret = CoeffBasedProductMode }; };
template<> struct product_type_selector<Large,1, Large> { enum { ret = GemvProduct }; };
template<> struct product_type_selector<Small,1, Large> { enum { ret = CoeffBasedProductMode }; };
template<> struct product_type_selector<Small,Small,Large> { enum { ret = GemmProduct }; };
template<> struct product_type_selector<Large,Small,Large> { enum { ret = GemmProduct }; };
template<> struct product_type_selector<Small,Large,Large> { enum { ret = GemmProduct }; };
template<> struct product_type_selector<Large,Large,Large> { enum { ret = GemmProduct }; };
template<> struct product_type_selector<Large,Small,Small> { enum { ret = GemmProduct }; };
template<> struct product_type_selector<Small,Large,Small> { enum { ret = GemmProduct }; };
template<> struct product_type_selector<Large,Large,Small> { enum { ret = GemmProduct }; };
} // end namespace internal
/** \class ProductReturnType
* \ingroup Core_Module
*
* \brief Helper class to get the correct and optimized returned type of operator*
*
* \param Lhs the type of the left-hand side
* \param Rhs the type of the right-hand side
* \param ProductMode the type of the product (determined automatically by internal::product_mode)
*
* This class defines the typename Type representing the optimized product expression
* between two matrix expressions. In practice, using ProductReturnType<Lhs,Rhs>::Type
* is the recommended way to define the result type of a function returning an expression
* which involve a matrix product. The class Product should never be
* used directly.
*
* \sa class Product, MatrixBase::operator*(const MatrixBase<OtherDerived>&)
*/
template<typename Lhs, typename Rhs, int ProductType>
struct ProductReturnType
{
// TODO use the nested type to reduce instanciations ????
// typedef typename internal::nested<Lhs,Rhs::ColsAtCompileTime>::type LhsNested;
// typedef typename internal::nested<Rhs,Lhs::RowsAtCompileTime>::type RhsNested;
typedef GeneralProduct<Lhs/*Nested*/, Rhs/*Nested*/, ProductType> Type;
};
template<typename Lhs, typename Rhs>
class Product : public ProductImpl<Lhs,Rhs,typename internal::promote_storage_type<typename internal::traits<Lhs>::StorageKind,
typename internal::traits<Rhs>::StorageKind>::ret>
struct ProductReturnType<Lhs,Rhs,CoeffBasedProductMode>
{
typedef typename internal::nested<Lhs, Rhs::ColsAtCompileTime, typename internal::plain_matrix_type<Lhs>::type >::type LhsNested;
typedef typename internal::nested<Rhs, Lhs::RowsAtCompileTime, typename internal::plain_matrix_type<Rhs>::type >::type RhsNested;
typedef CoeffBasedProduct<LhsNested, RhsNested, EvalBeforeAssigningBit | EvalBeforeNestingBit> Type;
};
template<typename Lhs, typename Rhs>
struct ProductReturnType<Lhs,Rhs,LazyCoeffBasedProductMode>
{
typedef typename internal::nested<Lhs, Rhs::ColsAtCompileTime, typename internal::plain_matrix_type<Lhs>::type >::type LhsNested;
typedef typename internal::nested<Rhs, Lhs::RowsAtCompileTime, typename internal::plain_matrix_type<Rhs>::type >::type RhsNested;
typedef CoeffBasedProduct<LhsNested, RhsNested, NestByRefBit> Type;
};
// this is a workaround for sun CC
template<typename Lhs, typename Rhs>
struct LazyProductReturnType : public ProductReturnType<Lhs,Rhs,LazyCoeffBasedProductMode>
{};
/***********************************************************************
* Implementation of Inner Vector Vector Product
***********************************************************************/
// FIXME : maybe the "inner product" could return a Scalar
// instead of a 1x1 matrix ??
// Pro: more natural for the user
// Cons: this could be a problem if in a meta unrolled algorithm a matrix-matrix
// product ends up to a row-vector times col-vector product... To tackle this use
// case, we could have a specialization for Block<MatrixType,1,1> with: operator=(Scalar x);
namespace internal {
template<typename Lhs, typename Rhs>
struct traits<GeneralProduct<Lhs,Rhs,InnerProduct> >
: traits<Matrix<typename scalar_product_traits<typename Lhs::Scalar, typename Rhs::Scalar>::ReturnType,1,1> >
{};
}
template<typename Lhs, typename Rhs>
class GeneralProduct<Lhs, Rhs, InnerProduct>
: internal::no_assignment_operator,
public Matrix<typename internal::scalar_product_traits<typename Lhs::Scalar, typename Rhs::Scalar>::ReturnType,1,1>
{
typedef Matrix<typename internal::scalar_product_traits<typename Lhs::Scalar, typename Rhs::Scalar>::ReturnType,1,1> Base;
public:
typedef typename ProductImpl<
Lhs, Rhs,
typename internal::promote_storage_type<typename Lhs::StorageKind,
typename Rhs::StorageKind>::ret>::Base Base;
EIGEN_GENERIC_PUBLIC_INTERFACE(Product)
typedef typename Lhs::Nested LhsNested;
typedef typename Rhs::Nested RhsNested;
typedef typename internal::remove_all<LhsNested>::type LhsNestedCleaned;
typedef typename internal::remove_all<RhsNested>::type RhsNestedCleaned;
Product(const Lhs& lhs, const Rhs& rhs) : m_lhs(lhs), m_rhs(rhs)
GeneralProduct(const Lhs& lhs, const Rhs& rhs)
{
eigen_assert(lhs.cols() == rhs.rows()
&& "invalid matrix product"
&& "if you wanted a coeff-wise or a dot product use the respective explicit functions");
EIGEN_STATIC_ASSERT((internal::is_same<typename Lhs::RealScalar, typename Rhs::RealScalar>::value),
YOU_MIXED_DIFFERENT_NUMERIC_TYPES__YOU_NEED_TO_USE_THE_CAST_METHOD_OF_MATRIXBASE_TO_CAST_NUMERIC_TYPES_EXPLICITLY)
Base::coeffRef(0,0) = (lhs.transpose().cwiseProduct(rhs)).sum();
}
inline Index rows() const { return m_lhs.rows(); }
inline Index cols() const { return m_rhs.cols(); }
const LhsNestedCleaned& lhs() const { return m_lhs; }
const RhsNestedCleaned& rhs() const { return m_rhs; }
protected:
const LhsNested m_lhs;
const RhsNested m_rhs;
/** Convertion to scalar */
operator const typename Base::Scalar() const {
return Base::coeff(0,0);
}
};
/***********************************************************************
* Implementation of Outer Vector Vector Product
***********************************************************************/
namespace internal {
template<int StorageOrder> struct outer_product_selector;
template<typename Lhs, typename Rhs>
class ProductImpl<Lhs,Rhs,Dense> : public internal::dense_xpr_base<Product<Lhs,Rhs> >::type
{
typedef Product<Lhs, Rhs> Derived;
public:
struct traits<GeneralProduct<Lhs,Rhs,OuterProduct> >
: traits<ProductBase<GeneralProduct<Lhs,Rhs,OuterProduct>, Lhs, Rhs> >
{};
typedef typename internal::dense_xpr_base<Product<Lhs, Rhs> >::type Base;
EIGEN_DENSE_PUBLIC_INTERFACE(Derived)
}
template<typename Lhs, typename Rhs>
class GeneralProduct<Lhs, Rhs, OuterProduct>
: public ProductBase<GeneralProduct<Lhs,Rhs,OuterProduct>, Lhs, Rhs>
{
public:
EIGEN_PRODUCT_PUBLIC_INTERFACE(GeneralProduct)
GeneralProduct(const Lhs& lhs, const Rhs& rhs) : Base(lhs,rhs)
{
EIGEN_STATIC_ASSERT((internal::is_same<typename Lhs::RealScalar, typename Rhs::RealScalar>::value),
YOU_MIXED_DIFFERENT_NUMERIC_TYPES__YOU_NEED_TO_USE_THE_CAST_METHOD_OF_MATRIXBASE_TO_CAST_NUMERIC_TYPES_EXPLICITLY)
}
template<typename Dest> void scaleAndAddTo(Dest& dest, Scalar alpha) const
{
internal::outer_product_selector<(int(Dest::Flags)&RowMajorBit) ? RowMajor : ColMajor>::run(*this, dest, alpha);
}
};
namespace internal {
template<> struct outer_product_selector<ColMajor> {
template<typename ProductType, typename Dest>
static EIGEN_DONT_INLINE void run(const ProductType& prod, Dest& dest, typename ProductType::Scalar alpha) {
typedef typename Dest::Index Index;
// FIXME make sure lhs is sequentially stored
// FIXME not very good if rhs is real and lhs complex while alpha is real too
const Index cols = dest.cols();
for (Index j=0; j<cols; ++j)
dest.col(j) += (alpha * prod.rhs().coeff(j)) * prod.lhs();
}
};
template<> struct outer_product_selector<RowMajor> {
template<typename ProductType, typename Dest>
static EIGEN_DONT_INLINE void run(const ProductType& prod, Dest& dest, typename ProductType::Scalar alpha) {
typedef typename Dest::Index Index;
// FIXME make sure rhs is sequentially stored
// FIXME not very good if lhs is real and rhs complex while alpha is real too
const Index rows = dest.rows();
for (Index i=0; i<rows; ++i)
dest.row(i) += (alpha * prod.lhs().coeff(i)) * prod.rhs();
}
};
} // end namespace internal
/***********************************************************************
* Implementation of General Matrix Vector Product
***********************************************************************/
/* According to the shape/flags of the matrix we have to distinghish 3 different cases:
* 1 - the matrix is col-major, BLAS compatible and M is large => call fast BLAS-like colmajor routine
* 2 - the matrix is row-major, BLAS compatible and N is large => call fast BLAS-like rowmajor routine
* 3 - all other cases are handled using a simple loop along the outer-storage direction.
* Therefore we need a lower level meta selector.
* Furthermore, if the matrix is the rhs, then the product has to be transposed.
*/
namespace internal {
template<typename Lhs, typename Rhs>
struct traits<GeneralProduct<Lhs,Rhs,GemvProduct> >
: traits<ProductBase<GeneralProduct<Lhs,Rhs,GemvProduct>, Lhs, Rhs> >
{};
template<int Side, int StorageOrder, bool BlasCompatible>
struct gemv_selector;
} // end namespace internal
template<typename Lhs, typename Rhs>
class GeneralProduct<Lhs, Rhs, GemvProduct>
: public ProductBase<GeneralProduct<Lhs,Rhs,GemvProduct>, Lhs, Rhs>
{
public:
EIGEN_PRODUCT_PUBLIC_INTERFACE(GeneralProduct)
typedef typename Lhs::Scalar LhsScalar;
typedef typename Rhs::Scalar RhsScalar;
GeneralProduct(const Lhs& lhs, const Rhs& rhs) : Base(lhs,rhs)
{
// EIGEN_STATIC_ASSERT((internal::is_same<typename Lhs::Scalar, typename Rhs::Scalar>::value),
// YOU_MIXED_DIFFERENT_NUMERIC_TYPES__YOU_NEED_TO_USE_THE_CAST_METHOD_OF_MATRIXBASE_TO_CAST_NUMERIC_TYPES_EXPLICITLY)
}
enum { Side = Lhs::IsVectorAtCompileTime ? OnTheLeft : OnTheRight };
typedef typename internal::conditional<int(Side)==OnTheRight,_LhsNested,_RhsNested>::type MatrixType;
template<typename Dest> void scaleAndAddTo(Dest& dst, Scalar alpha) const
{
eigen_assert(m_lhs.rows() == dst.rows() && m_rhs.cols() == dst.cols());
internal::gemv_selector<Side,(int(MatrixType::Flags)&RowMajorBit) ? RowMajor : ColMajor,
bool(internal::blas_traits<MatrixType>::HasUsableDirectAccess)>::run(*this, dst, alpha);
}
};
namespace internal {
// The vector is on the left => transposition
template<int StorageOrder, bool BlasCompatible>
struct gemv_selector<OnTheLeft,StorageOrder,BlasCompatible>
{
template<typename ProductType, typename Dest>
static void run(const ProductType& prod, Dest& dest, typename ProductType::Scalar alpha)
{
Transpose<Dest> destT(dest);
enum { OtherStorageOrder = StorageOrder == RowMajor ? ColMajor : RowMajor };
gemv_selector<OnTheRight,OtherStorageOrder,BlasCompatible>
::run(GeneralProduct<Transpose<const typename ProductType::_RhsNested>,Transpose<const typename ProductType::_LhsNested>, GemvProduct>
(prod.rhs().transpose(), prod.lhs().transpose()), destT, alpha);
}
};
template<typename Scalar,int Size,int MaxSize,bool Cond> struct gemv_static_vector_if;
template<typename Scalar,int Size,int MaxSize>
struct gemv_static_vector_if<Scalar,Size,MaxSize,false>
{
EIGEN_STRONG_INLINE Scalar* data() { eigen_internal_assert(false && "should never be called"); return 0; }
};
template<typename Scalar,int Size>
struct gemv_static_vector_if<Scalar,Size,Dynamic,true>
{
EIGEN_STRONG_INLINE Scalar* data() { return 0; }
};
template<typename Scalar,int Size,int MaxSize>
struct gemv_static_vector_if<Scalar,Size,MaxSize,true>
{
#if EIGEN_ALIGN_STATICALLY
internal::plain_array<Scalar,EIGEN_SIZE_MIN_PREFER_FIXED(Size,MaxSize),0> m_data;
EIGEN_STRONG_INLINE Scalar* data() { return m_data.array; }
#else
// Some architectures cannot align on the stack,
// => let's manually enforce alignment by allocating more data and return the address of the first aligned element.
enum {
ForceAlignment = internal::packet_traits<Scalar>::Vectorizable,
PacketSize = internal::packet_traits<Scalar>::size
};
internal::plain_array<Scalar,EIGEN_SIZE_MIN_PREFER_FIXED(Size,MaxSize)+(ForceAlignment?PacketSize:0),0> m_data;
EIGEN_STRONG_INLINE Scalar* data() {
return ForceAlignment
? reinterpret_cast<Scalar*>((reinterpret_cast<size_t>(m_data.array) & ~(size_t(15))) + 16)
: m_data.array;
}
#endif
};
template<> struct gemv_selector<OnTheRight,ColMajor,true>
{
template<typename ProductType, typename Dest>
static inline void run(const ProductType& prod, Dest& dest, typename ProductType::Scalar alpha)
{
typedef typename ProductType::Index Index;
typedef typename ProductType::LhsScalar LhsScalar;
typedef typename ProductType::RhsScalar RhsScalar;
typedef typename ProductType::Scalar ResScalar;
typedef typename ProductType::RealScalar RealScalar;
typedef typename ProductType::ActualLhsType ActualLhsType;
typedef typename ProductType::ActualRhsType ActualRhsType;
typedef typename ProductType::LhsBlasTraits LhsBlasTraits;
typedef typename ProductType::RhsBlasTraits RhsBlasTraits;
typedef Map<Matrix<ResScalar,Dynamic,1>, Aligned> MappedDest;
const ActualLhsType actualLhs = LhsBlasTraits::extract(prod.lhs());
const ActualRhsType actualRhs = RhsBlasTraits::extract(prod.rhs());
ResScalar actualAlpha = alpha * LhsBlasTraits::extractScalarFactor(prod.lhs())
* RhsBlasTraits::extractScalarFactor(prod.rhs());
enum {
// FIXME find a way to allow an inner stride on the result if packet_traits<Scalar>::size==1
// on, the other hand it is good for the cache to pack the vector anyways...
EvalToDestAtCompileTime = Dest::InnerStrideAtCompileTime==1,
ComplexByReal = (NumTraits<LhsScalar>::IsComplex) && (!NumTraits<RhsScalar>::IsComplex),
MightCannotUseDest = (Dest::InnerStrideAtCompileTime!=1) || ComplexByReal
};
gemv_static_vector_if<ResScalar,Dest::SizeAtCompileTime,Dest::MaxSizeAtCompileTime,MightCannotUseDest> static_dest;
// this is written like this (i.e., with a ?:) to workaround an ICE with ICC 12
bool alphaIsCompatible = (!ComplexByReal) ? true : (imag(actualAlpha)==RealScalar(0));
bool evalToDest = EvalToDestAtCompileTime && alphaIsCompatible;
RhsScalar compatibleAlpha = get_factor<ResScalar,RhsScalar>::run(actualAlpha);
ei_declare_aligned_stack_constructed_variable(ResScalar,actualDestPtr,dest.size(),
evalToDest ? dest.data() : static_dest.data());
if(!evalToDest)
{
#ifdef EIGEN_DENSE_STORAGE_CTOR_PLUGIN
int size = dest.size();
EIGEN_DENSE_STORAGE_CTOR_PLUGIN
#endif
if(!alphaIsCompatible)
{
MappedDest(actualDestPtr, dest.size()).setZero();
compatibleAlpha = RhsScalar(1);
}
else
MappedDest(actualDestPtr, dest.size()) = dest;
}
general_matrix_vector_product
<Index,LhsScalar,ColMajor,LhsBlasTraits::NeedToConjugate,RhsScalar,RhsBlasTraits::NeedToConjugate>::run(
actualLhs.rows(), actualLhs.cols(),
&actualLhs.coeffRef(0,0), actualLhs.outerStride(),
actualRhs.data(), actualRhs.innerStride(),
actualDestPtr, 1,
compatibleAlpha);
if (!evalToDest)
{
if(!alphaIsCompatible)
dest += actualAlpha * MappedDest(actualDestPtr, dest.size());
else
dest = MappedDest(actualDestPtr, dest.size());
}
}
};
template<> struct gemv_selector<OnTheRight,RowMajor,true>
{
template<typename ProductType, typename Dest>
static void run(const ProductType& prod, Dest& dest, typename ProductType::Scalar alpha)
{
typedef typename ProductType::LhsScalar LhsScalar;
typedef typename ProductType::RhsScalar RhsScalar;
typedef typename ProductType::Scalar ResScalar;
typedef typename ProductType::Index Index;
typedef typename ProductType::ActualLhsType ActualLhsType;
typedef typename ProductType::ActualRhsType ActualRhsType;
typedef typename ProductType::_ActualRhsType _ActualRhsType;
typedef typename ProductType::LhsBlasTraits LhsBlasTraits;
typedef typename ProductType::RhsBlasTraits RhsBlasTraits;
typename add_const<ActualLhsType>::type actualLhs = LhsBlasTraits::extract(prod.lhs());
typename add_const<ActualRhsType>::type actualRhs = RhsBlasTraits::extract(prod.rhs());
ResScalar actualAlpha = alpha * LhsBlasTraits::extractScalarFactor(prod.lhs())
* RhsBlasTraits::extractScalarFactor(prod.rhs());
enum {
// FIXME find a way to allow an inner stride on the result if packet_traits<Scalar>::size==1
// on, the other hand it is good for the cache to pack the vector anyways...
DirectlyUseRhs = _ActualRhsType::InnerStrideAtCompileTime==1
};
gemv_static_vector_if<RhsScalar,_ActualRhsType::SizeAtCompileTime,_ActualRhsType::MaxSizeAtCompileTime,!DirectlyUseRhs> static_rhs;
ei_declare_aligned_stack_constructed_variable(RhsScalar,actualRhsPtr,actualRhs.size(),
DirectlyUseRhs ? const_cast<RhsScalar*>(actualRhs.data()) : static_rhs.data());
if(!DirectlyUseRhs)
{
#ifdef EIGEN_DENSE_STORAGE_CTOR_PLUGIN
int size = actualRhs.size();
EIGEN_DENSE_STORAGE_CTOR_PLUGIN
#endif
Map<typename _ActualRhsType::PlainObject>(actualRhsPtr, actualRhs.size()) = actualRhs;
}
general_matrix_vector_product
<Index,LhsScalar,RowMajor,LhsBlasTraits::NeedToConjugate,RhsScalar,RhsBlasTraits::NeedToConjugate>::run(
actualLhs.rows(), actualLhs.cols(),
&actualLhs.coeffRef(0,0), actualLhs.outerStride(),
actualRhsPtr, 1,
&dest.coeffRef(0,0), dest.innerStride(),
actualAlpha);
}
};
template<> struct gemv_selector<OnTheRight,ColMajor,false>
{
template<typename ProductType, typename Dest>
static void run(const ProductType& prod, Dest& dest, typename ProductType::Scalar alpha)
{
typedef typename Dest::Index Index;
// TODO makes sure dest is sequentially stored in memory, otherwise use a temp
const Index size = prod.rhs().rows();
for(Index k=0; k<size; ++k)
dest += (alpha*prod.rhs().coeff(k)) * prod.lhs().col(k);
}
};
template<> struct gemv_selector<OnTheRight,RowMajor,false>
{
template<typename ProductType, typename Dest>
static void run(const ProductType& prod, Dest& dest, typename ProductType::Scalar alpha)
{
typedef typename Dest::Index Index;
// TODO makes sure rhs is sequentially stored in memory, otherwise use a temp
const Index rows = prod.rows();
for(Index i=0; i<rows; ++i)
dest.coeffRef(i) += alpha * (prod.lhs().row(i).cwiseProduct(prod.rhs().transpose())).sum();
}
};
} // end namespace internal
/***************************************************************************
* Implementation of matrix base methods
***************************************************************************/
/** \internal used to test the evaluator only
/** \returns the matrix product of \c *this and \a other.
*
* \note If instead of the matrix product you want the coefficient-wise product, see Cwise::operator*().
*
* \sa lazyProduct(), operator*=(const MatrixBase&), Cwise::operator*()
*/
template<typename Lhs,typename Rhs>
const Product<Lhs,Rhs>
prod(const Lhs& lhs, const Rhs& rhs)
template<typename Derived>
template<typename OtherDerived>
inline const typename ProductReturnType<Derived,OtherDerived>::Type
MatrixBase<Derived>::operator*(const MatrixBase<OtherDerived> &other) const
{
return Product<Lhs,Rhs>(lhs,rhs);
// A note regarding the function declaration: In MSVC, this function will sometimes
// not be inlined since DenseStorage is an unwindable object for dynamic
// matrices and product types are holding a member to store the result.
// Thus it does not help tagging this function with EIGEN_STRONG_INLINE.
enum {
ProductIsValid = Derived::ColsAtCompileTime==Dynamic
|| OtherDerived::RowsAtCompileTime==Dynamic
|| int(Derived::ColsAtCompileTime)==int(OtherDerived::RowsAtCompileTime),
AreVectors = Derived::IsVectorAtCompileTime && OtherDerived::IsVectorAtCompileTime,
SameSizes = EIGEN_PREDICATE_SAME_MATRIX_SIZE(Derived,OtherDerived)
};
// note to the lost user:
// * for a dot product use: v1.dot(v2)
// * for a coeff-wise product use: v1.cwiseProduct(v2)
EIGEN_STATIC_ASSERT(ProductIsValid || !(AreVectors && SameSizes),
INVALID_VECTOR_VECTOR_PRODUCT__IF_YOU_WANTED_A_DOT_OR_COEFF_WISE_PRODUCT_YOU_MUST_USE_THE_EXPLICIT_FUNCTIONS)
EIGEN_STATIC_ASSERT(ProductIsValid || !(SameSizes && !AreVectors),
INVALID_MATRIX_PRODUCT__IF_YOU_WANTED_A_COEFF_WISE_PRODUCT_YOU_MUST_USE_THE_EXPLICIT_FUNCTION)
EIGEN_STATIC_ASSERT(ProductIsValid || SameSizes, INVALID_MATRIX_PRODUCT)
#ifdef EIGEN_DEBUG_PRODUCT
internal::product_type<Derived,OtherDerived>::debug();
#endif
return typename ProductReturnType<Derived,OtherDerived>::Type(derived(), other.derived());
}
/** \returns an expression of the matrix product of \c *this and \a other without implicit evaluation.
*
* The returned product will behave like any other expressions: the coefficients of the product will be
* computed once at a time as requested. This might be useful in some extremely rare cases when only
* a small and no coherent fraction of the result's coefficients have to be computed.
*
* \warning This version of the matrix product can be much much slower. So use it only if you know
* what you are doing and that you measured a true speed improvement.
*
* \sa operator*(const MatrixBase&)
*/
template<typename Derived>
template<typename OtherDerived>
const typename LazyProductReturnType<Derived,OtherDerived>::Type
MatrixBase<Derived>::lazyProduct(const MatrixBase<OtherDerived> &other) const
{
enum {
ProductIsValid = Derived::ColsAtCompileTime==Dynamic
|| OtherDerived::RowsAtCompileTime==Dynamic
|| int(Derived::ColsAtCompileTime)==int(OtherDerived::RowsAtCompileTime),
AreVectors = Derived::IsVectorAtCompileTime && OtherDerived::IsVectorAtCompileTime,
SameSizes = EIGEN_PREDICATE_SAME_MATRIX_SIZE(Derived,OtherDerived)
};
// note to the lost user:
// * for a dot product use: v1.dot(v2)
// * for a coeff-wise product use: v1.cwiseProduct(v2)
EIGEN_STATIC_ASSERT(ProductIsValid || !(AreVectors && SameSizes),
INVALID_VECTOR_VECTOR_PRODUCT__IF_YOU_WANTED_A_DOT_OR_COEFF_WISE_PRODUCT_YOU_MUST_USE_THE_EXPLICIT_FUNCTIONS)
EIGEN_STATIC_ASSERT(ProductIsValid || !(SameSizes && !AreVectors),
INVALID_MATRIX_PRODUCT__IF_YOU_WANTED_A_COEFF_WISE_PRODUCT_YOU_MUST_USE_THE_EXPLICIT_FUNCTION)
EIGEN_STATIC_ASSERT(ProductIsValid || SameSizes, INVALID_MATRIX_PRODUCT)
return typename LazyProductReturnType<Derived,OtherDerived>::Type(derived(), other.derived());
}
#endif // EIGEN_PRODUCT_H

View File

@@ -256,16 +256,16 @@ class ScaledProduct
: Base(prod.lhs(),prod.rhs()), m_prod(prod), m_alpha(x) {}
template<typename Dest>
inline void evalTo(Dest& dst) const { dst.setZero(); scaleAndAddTo(dst, Scalar(1)); }
inline void evalTo(Dest& dst) const { dst.setZero(); scaleAndAddTo(dst,m_alpha); }
template<typename Dest>
inline void addTo(Dest& dst) const { scaleAndAddTo(dst, Scalar(1)); }
inline void addTo(Dest& dst) const { scaleAndAddTo(dst,m_alpha); }
template<typename Dest>
inline void subTo(Dest& dst) const { scaleAndAddTo(dst, Scalar(-1)); }
inline void subTo(Dest& dst) const { scaleAndAddTo(dst,-m_alpha); }
template<typename Dest>
inline void scaleAndAddTo(Dest& dst,Scalar alpha) const { m_prod.derived().scaleAndAddTo(dst,alpha * m_alpha); }
inline void scaleAndAddTo(Dest& dst,Scalar alpha) const { m_prod.derived().scaleAndAddTo(dst,alpha); }
const Scalar& alpha() const { return m_alpha; }

View File

@@ -219,28 +219,15 @@ struct redux_impl<Func, Derived, LinearVectorizedTraversal, NoUnrolling>
alignment = bool(Derived::Flags & DirectAccessBit) || bool(Derived::Flags & AlignedBit)
? Aligned : Unaligned
};
const Index alignedSize2 = ((size-alignedStart)/(2*packetSize))*(2*packetSize);
const Index alignedSize = ((size-alignedStart)/(packetSize))*(packetSize);
const Index alignedEnd2 = alignedStart + alignedSize2;
const Index alignedEnd = alignedStart + alignedSize;
const Index alignedSize = ((size-alignedStart)/packetSize)*packetSize;
const Index alignedEnd = alignedStart + alignedSize;
Scalar res;
if(alignedSize)
{
PacketScalar packet_res0 = mat.template packet<alignment>(alignedStart);
if(alignedSize>packetSize) // we have at least two packets to partly unroll the loop
{
PacketScalar packet_res1 = mat.template packet<alignment>(alignedStart+packetSize);
for(Index index = alignedStart + 2*packetSize; index < alignedEnd2; index += 2*packetSize)
{
packet_res0 = func.packetOp(packet_res0, mat.template packet<alignment>(index));
packet_res1 = func.packetOp(packet_res1, mat.template packet<alignment>(index+packetSize));
}
packet_res0 = func.packetOp(packet_res0,packet_res1);
if(alignedEnd>alignedEnd2)
packet_res0 = func.packetOp(packet_res0, mat.template packet<alignment>(alignedEnd2));
}
res = func.predux(packet_res0);
PacketScalar packet_res = mat.template packet<alignment>(alignedStart);
for(Index index = alignedStart + packetSize; index < alignedEnd; index += packetSize)
packet_res = func.packetOp(packet_res, mat.template packet<alignment>(index));
res = func.predux(packet_res);
for(Index index = 0; index < alignedStart; ++index)
res = func(res,mat.coeff(index));

View File

@@ -122,10 +122,6 @@ template<typename MatrixType,int RowFactor,int ColFactor> class Replicate
return m_matrix.template packet<LoadMode>(actual_row, actual_col);
}
const typename internal::remove_all<typename MatrixType::Nested>::type& nestedExpression() const
{
return m_matrix;
}
protected:
const typename MatrixType::Nested m_matrix;

View File

@@ -183,12 +183,6 @@ template<typename MatrixType, int Direction> class Reverse
m_matrix.const_cast_derived().template writePacket<LoadMode>(m_matrix.size() - index - PacketSize, internal::preverse(x));
}
const typename internal::remove_all<typename MatrixType::Nested>::type&
nestedExpression() const
{
return m_matrix;
}
protected:
const typename MatrixType::Nested m_matrix;
};

View File

@@ -101,21 +101,6 @@ class Select : internal::no_assignment_operator,
return m_else.coeff(i);
}
const ConditionMatrixType& conditionMatrix() const
{
return m_condition;
}
const ThenMatrixType& thenMatrix() const
{
return m_then;
}
const ElseMatrixType& elseMatrix() const
{
return m_else;
}
protected:
const typename ConditionMatrixType::Nested m_condition;
const typename ThenMatrixType::Nested m_then;

View File

@@ -163,16 +163,6 @@ template<typename BinaryOp, typename Lhs, typename Rhs> class SelfCwiseBinaryOp
return Base::operator=(rhs);
}
Lhs& expression() const
{
return m_matrix;
}
const BinaryOp& functor() const
{
return m_functor;
}
protected:
Lhs& m_matrix;
const BinaryOp& m_functor;

View File

@@ -119,8 +119,6 @@ template<typename ExpressionType> class SwapWrapper
_other.template writePacket<LoadMode>(index, tmp);
}
ExpressionType& expression() const { return m_expression; }
protected:
ExpressionType& m_expression;
};

View File

@@ -237,10 +237,7 @@ template<typename ExpressionType, int Direction> class VectorwiseOp
typename ExtendedType<OtherDerived>::Type
extendedTo(const DenseBase<OtherDerived>& other) const
{
EIGEN_STATIC_ASSERT(EIGEN_IMPLIES(Direction==Vertical, OtherDerived::MaxColsAtCompileTime==1),
YOU_PASSED_A_ROW_VECTOR_BUT_A_COLUMN_VECTOR_WAS_EXPECTED)
EIGEN_STATIC_ASSERT(EIGEN_IMPLIES(Direction==Horizontal, OtherDerived::MaxRowsAtCompileTime==1),
YOU_PASSED_A_COLUMN_VECTOR_BUT_A_ROW_VECTOR_WAS_EXPECTED)
EIGEN_STATIC_ASSERT_VECTOR_ONLY(OtherDerived);
return typename ExtendedType<OtherDerived>::Type
(other.derived(),
Direction==Vertical ? 1 : m_matrix.rows(),
@@ -421,9 +418,10 @@ template<typename ExpressionType, int Direction> class VectorwiseOp
ExpressionType& operator=(const DenseBase<OtherDerived>& other)
{
EIGEN_STATIC_ASSERT_VECTOR_ONLY(OtherDerived)
EIGEN_STATIC_ASSERT_SAME_XPR_KIND(ExpressionType, OtherDerived)
//eigen_assert((m_matrix.isNull()) == (other.isNull())); FIXME
return const_cast<ExpressionType&>(m_matrix = extendedTo(other.derived()));
for(Index j=0; j<subVectors(); ++j)
subVector(j) = other;
return const_cast<ExpressionType&>(m_matrix);
}
/** Adds the vector \a other to each subvector of \c *this */
@@ -431,8 +429,9 @@ template<typename ExpressionType, int Direction> class VectorwiseOp
ExpressionType& operator+=(const DenseBase<OtherDerived>& other)
{
EIGEN_STATIC_ASSERT_VECTOR_ONLY(OtherDerived)
EIGEN_STATIC_ASSERT_SAME_XPR_KIND(ExpressionType, OtherDerived)
return const_cast<ExpressionType&>(m_matrix += extendedTo(other.derived()));
for(Index j=0; j<subVectors(); ++j)
subVector(j) += other.derived();
return const_cast<ExpressionType&>(m_matrix);
}
/** Substracts the vector \a other to each subvector of \c *this */
@@ -440,29 +439,8 @@ template<typename ExpressionType, int Direction> class VectorwiseOp
ExpressionType& operator-=(const DenseBase<OtherDerived>& other)
{
EIGEN_STATIC_ASSERT_VECTOR_ONLY(OtherDerived)
EIGEN_STATIC_ASSERT_SAME_XPR_KIND(ExpressionType, OtherDerived)
return const_cast<ExpressionType&>(m_matrix -= extendedTo(other.derived()));
}
/** Multiples each subvector of \c *this by the vector \a other */
template<typename OtherDerived>
ExpressionType& operator*=(const DenseBase<OtherDerived>& other)
{
EIGEN_STATIC_ASSERT_VECTOR_ONLY(OtherDerived)
EIGEN_STATIC_ASSERT_ARRAYXPR(ExpressionType)
EIGEN_STATIC_ASSERT_SAME_XPR_KIND(ExpressionType, OtherDerived)
m_matrix *= extendedTo(other.derived());
return const_cast<ExpressionType&>(m_matrix);
}
/** Divides each subvector of \c *this by the vector \a other */
template<typename OtherDerived>
ExpressionType& operator/=(const DenseBase<OtherDerived>& other)
{
EIGEN_STATIC_ASSERT_VECTOR_ONLY(OtherDerived)
EIGEN_STATIC_ASSERT_ARRAYXPR(ExpressionType)
EIGEN_STATIC_ASSERT_SAME_XPR_KIND(ExpressionType, OtherDerived)
m_matrix /= extendedTo(other.derived());
for(Index j=0; j<subVectors(); ++j)
subVector(j) -= other.derived();
return const_cast<ExpressionType&>(m_matrix);
}
@@ -473,8 +451,7 @@ template<typename ExpressionType, int Direction> class VectorwiseOp
const typename ExtendedType<OtherDerived>::Type>
operator+(const DenseBase<OtherDerived>& other) const
{
EIGEN_STATIC_ASSERT_VECTOR_ONLY(OtherDerived)
EIGEN_STATIC_ASSERT_SAME_XPR_KIND(ExpressionType, OtherDerived)
EIGEN_STATIC_ASSERT_VECTOR_ONLY(OtherDerived);
return m_matrix + extendedTo(other.derived());
}
@@ -485,39 +462,10 @@ template<typename ExpressionType, int Direction> class VectorwiseOp
const typename ExtendedType<OtherDerived>::Type>
operator-(const DenseBase<OtherDerived>& other) const
{
EIGEN_STATIC_ASSERT_VECTOR_ONLY(OtherDerived)
EIGEN_STATIC_ASSERT_SAME_XPR_KIND(ExpressionType, OtherDerived)
EIGEN_STATIC_ASSERT_VECTOR_ONLY(OtherDerived);
return m_matrix - extendedTo(other.derived());
}
/** Returns the expression where each subvector is the product of the vector \a other
* by the corresponding subvector of \c *this */
template<typename OtherDerived> EIGEN_STRONG_INLINE
CwiseBinaryOp<internal::scalar_product_op<Scalar>,
const ExpressionTypeNestedCleaned,
const typename ExtendedType<OtherDerived>::Type>
operator*(const DenseBase<OtherDerived>& other) const
{
EIGEN_STATIC_ASSERT_VECTOR_ONLY(OtherDerived)
EIGEN_STATIC_ASSERT_ARRAYXPR(ExpressionType)
EIGEN_STATIC_ASSERT_SAME_XPR_KIND(ExpressionType, OtherDerived)
return m_matrix * extendedTo(other.derived());
}
/** Returns the expression where each subvector is the quotient of the corresponding
* subvector of \c *this by the vector \a other */
template<typename OtherDerived>
CwiseBinaryOp<internal::scalar_quotient_op<Scalar>,
const ExpressionTypeNestedCleaned,
const typename ExtendedType<OtherDerived>::Type>
operator/(const DenseBase<OtherDerived>& other) const
{
EIGEN_STATIC_ASSERT_VECTOR_ONLY(OtherDerived)
EIGEN_STATIC_ASSERT_ARRAYXPR(ExpressionType)
EIGEN_STATIC_ASSERT_SAME_XPR_KIND(ExpressionType, OtherDerived)
return m_matrix / extendedTo(other.derived());
}
/////////// Geometry module ///////////
#if EIGEN2_SUPPORT_STAGE > STAGE20_RESOLVE_API_CONFLICTS
@@ -561,7 +509,7 @@ template<typename ExpressionType, int Direction> class VectorwiseOp
* Example: \include MatrixBase_colwise.cpp
* Output: \verbinclude MatrixBase_colwise.out
*
* \sa rowwise(), class VectorwiseOp, \ref TutorialReductionsVisitorsBroadcasting
* \sa rowwise(), class VectorwiseOp
*/
template<typename Derived>
inline const typename DenseBase<Derived>::ConstColwiseReturnType
@@ -572,7 +520,7 @@ DenseBase<Derived>::colwise() const
/** \returns a writable VectorwiseOp wrapper of *this providing additional partial reduction operations
*
* \sa rowwise(), class VectorwiseOp, \ref TutorialReductionsVisitorsBroadcasting
* \sa rowwise(), class VectorwiseOp
*/
template<typename Derived>
inline typename DenseBase<Derived>::ColwiseReturnType
@@ -586,7 +534,7 @@ DenseBase<Derived>::colwise()
* Example: \include MatrixBase_rowwise.cpp
* Output: \verbinclude MatrixBase_rowwise.out
*
* \sa colwise(), class VectorwiseOp, \ref TutorialReductionsVisitorsBroadcasting
* \sa colwise(), class VectorwiseOp
*/
template<typename Derived>
inline const typename DenseBase<Derived>::ConstRowwiseReturnType
@@ -597,7 +545,7 @@ DenseBase<Derived>::rowwise() const
/** \returns a writable VectorwiseOp wrapper of *this providing additional partial reduction operations
*
* \sa colwise(), class VectorwiseOp, \ref TutorialReductionsVisitorsBroadcasting
* \sa colwise(), class VectorwiseOp
*/
template<typename Derived>
inline typename DenseBase<Derived>::RowwiseReturnType

View File

@@ -27,8 +27,8 @@
namespace internal {
static uint32x4_t p4ui_CONJ_XOR = EIGEN_INIT_NEON_PACKET4(0x00000000, 0x80000000, 0x00000000, 0x80000000);
static uint32x2_t p2ui_CONJ_XOR = EIGEN_INIT_NEON_PACKET2(0x00000000, 0x80000000);
static uint32x4_t p4ui_CONJ_XOR = { 0x00000000, 0x80000000, 0x00000000, 0x80000000 };
static uint32x2_t p2ui_CONJ_XOR = { 0x00000000, 0x80000000 };
//---------- float ----------
struct Packet2cf

View File

@@ -52,16 +52,6 @@ typedef uint32x4_t Packet4ui;
#define _EIGEN_DECLARE_CONST_Packet4i(NAME,X) \
const Packet4i p4i_##NAME = pset1<Packet4i>(X)
#if defined(__llvm__) && !defined(__clang__)
//Special treatment for Apple's llvm-gcc, its NEON packet types are unions
#define EIGEN_INIT_NEON_PACKET2(X, Y) {{X, Y}}
#define EIGEN_INIT_NEON_PACKET4(X, Y, Z, W) {{X, Y, Z, W}}
#else
//Default initializer for packets
#define EIGEN_INIT_NEON_PACKET2(X, Y) {X, Y}
#define EIGEN_INIT_NEON_PACKET4(X, Y, Z, W) {X, Y, Z, W}
#endif
#ifndef __pld
#define __pld(x) asm volatile ( " pld [%[addr]]\n" :: [addr] "r" (x) : "cc" );
#endif
@@ -94,7 +84,7 @@ template<> struct packet_traits<int> : default_packet_traits
};
};
#if EIGEN_GNUC_AT_MOST(4,4) && !defined(__llvm__)
#if EIGEN_GNUC_AT_MOST(4,4)
// workaround gcc 4.2, 4.3 and 4.4 compilatin issue
EIGEN_STRONG_INLINE float32x4_t vld1q_f32(const float* x) { return ::vld1q_f32((const float32_t*)x); }
EIGEN_STRONG_INLINE float32x2_t vld1_f32 (const float* x) { return ::vld1_f32 ((const float32_t*)x); }
@@ -110,12 +100,12 @@ template<> EIGEN_STRONG_INLINE Packet4i pset1<Packet4i>(const int& from) {
template<> EIGEN_STRONG_INLINE Packet4f plset<float>(const float& a)
{
Packet4f countdown = EIGEN_INIT_NEON_PACKET4(0, 1, 2, 3);
Packet4f countdown = { 0, 1, 2, 3 };
return vaddq_f32(pset1<Packet4f>(a), countdown);
}
template<> EIGEN_STRONG_INLINE Packet4i plset<int>(const int& a)
{
Packet4i countdown = EIGEN_INIT_NEON_PACKET4(0, 1, 2, 3);
Packet4i countdown = { 0, 1, 2, 3 };
return vaddq_s32(pset1<Packet4i>(a), countdown);
}
@@ -405,29 +395,25 @@ template<> EIGEN_STRONG_INLINE int predux_max<Packet4i>(const Packet4i& a)
return s[0];
}
// this PALIGN_NEON business is to work around a bug in LLVM Clang 3.0 causing incorrect compilation errors,
// see bug 347 and this LLVM bug: http://llvm.org/bugs/show_bug.cgi?id=11074
#define PALIGN_NEON(Offset,Type,Command) \
template<>\
struct palign_impl<Offset,Type>\
{\
EIGEN_STRONG_INLINE static void run(Type& first, const Type& second)\
{\
if (Offset!=0)\
first = Command(first, second, Offset);\
}\
};\
template<int Offset>
struct palign_impl<Offset,Packet4f>
{
EIGEN_STRONG_INLINE static void run(Packet4f& first, const Packet4f& second)
{
if (Offset!=0)
first = vextq_f32(first, second, Offset);
}
};
PALIGN_NEON(0,Packet4f,vextq_f32)
PALIGN_NEON(1,Packet4f,vextq_f32)
PALIGN_NEON(2,Packet4f,vextq_f32)
PALIGN_NEON(3,Packet4f,vextq_f32)
PALIGN_NEON(0,Packet4i,vextq_s32)
PALIGN_NEON(1,Packet4i,vextq_s32)
PALIGN_NEON(2,Packet4i,vextq_s32)
PALIGN_NEON(3,Packet4i,vextq_s32)
#undef PALIGN_NEON
template<int Offset>
struct palign_impl<Offset,Packet4i>
{
EIGEN_STRONG_INLINE static void run(Packet4i& first, const Packet4i& second)
{
if (Offset!=0)
first = vextq_s32(first, second, Offset);
}
};
} // end namespace internal

View File

@@ -110,18 +110,9 @@ template<> struct unpacket_traits<Packet4f> { typedef float type; enum {size=4}
template<> struct unpacket_traits<Packet2d> { typedef double type; enum {size=2}; };
template<> struct unpacket_traits<Packet4i> { typedef int type; enum {size=4}; };
#if defined(_MSC_VER) && (_MSC_VER==1500)
// Workaround MSVC 9 internal compiler error.
// TODO: It has been detected with win64 builds (amd64), so let's check whether it also happens in 32bits+SSE mode
// TODO: let's check whether there does not exist a better fix, like adding a pset0() function. (it crashed on pset1(0)).
template<> EIGEN_STRONG_INLINE Packet4f pset1<Packet4f>(const float& from) { return _mm_set_ps(from,from,from,from); }
template<> EIGEN_STRONG_INLINE Packet2d pset1<Packet2d>(const double& from) { return _mm_set_pd(from,from); }
template<> EIGEN_STRONG_INLINE Packet4i pset1<Packet4i>(const int& from) { return _mm_set_epi32(from,from,from,from); }
#else
template<> EIGEN_STRONG_INLINE Packet4f pset1<Packet4f>(const float& from) { return _mm_set1_ps(from); }
template<> EIGEN_STRONG_INLINE Packet2d pset1<Packet2d>(const double& from) { return _mm_set1_pd(from); }
template<> EIGEN_STRONG_INLINE Packet4i pset1<Packet4i>(const int& from) { return _mm_set1_epi32(from); }
#endif
template<> EIGEN_STRONG_INLINE Packet4f plset<float>(const float& a) { return _mm_add_ps(pset1<Packet4f>(a), _mm_set_ps(3,2,1,0)); }
template<> EIGEN_STRONG_INLINE Packet2d plset<double>(const double& a) { return _mm_add_pd(pset1<Packet2d>(a),_mm_set_pd(1,0)); }

View File

@@ -118,14 +118,14 @@ inline void computeProductBlockingSizes(std::ptrdiff_t& k, std::ptrdiff_t& m, st
// FIXME (a bit overkill maybe ?)
template<typename CJ, typename A, typename B, typename C, typename T> struct gebp_madd_selector {
EIGEN_ALWAYS_INLINE static void run(const CJ& cj, A& a, B& b, C& c, T& /*t*/)
EIGEN_STRONG_INLINE EIGEN_ALWAYS_INLINE_ATTRIB static void run(const CJ& cj, A& a, B& b, C& c, T& /*t*/)
{
c = cj.pmadd(a,b,c);
}
};
template<typename CJ, typename T> struct gebp_madd_selector<CJ,T,T,T,T> {
EIGEN_ALWAYS_INLINE static void run(const CJ& cj, T& a, T& b, T& c, T& t)
EIGEN_STRONG_INLINE EIGEN_ALWAYS_INLINE_ATTRIB static void run(const CJ& cj, T& a, T& b, T& c, T& t)
{
t = b; t = cj.pmul(a,t); c = padd(c,t);
}
@@ -536,7 +536,7 @@ struct gebp_kernel
ResPacketSize = Traits::ResPacketSize
};
EIGEN_DONT_INLINE EIGEN_FLATTEN_ATTRIB
EIGEN_FLATTEN_ATTRIB
void operator()(ResScalar* res, Index resStride, const LhsScalar* blockA, const RhsScalar* blockB, Index rows, Index depth, Index cols, ResScalar alpha,
Index strideA=-1, Index strideB=-1, Index offsetA=0, Index offsetB=0, RhsScalar* unpackedB = 0)
{
@@ -598,64 +598,64 @@ struct gebp_kernel
if(nr==2)
{
LhsPacket A0, A1;
RhsPacket B_0;
RhsPacket B0;
RhsPacket T0;
EIGEN_ASM_COMMENT("mybegin2");
traits.loadLhs(&blA[0*LhsProgress], A0);
traits.loadLhs(&blA[1*LhsProgress], A1);
traits.loadRhs(&blB[0*RhsProgress], B_0);
traits.madd(A0,B_0,C0,T0);
traits.madd(A1,B_0,C4,B_0);
traits.loadRhs(&blB[1*RhsProgress], B_0);
traits.madd(A0,B_0,C1,T0);
traits.madd(A1,B_0,C5,B_0);
traits.loadRhs(&blB[0*RhsProgress], B0);
traits.madd(A0,B0,C0,T0);
traits.madd(A1,B0,C4,B0);
traits.loadRhs(&blB[1*RhsProgress], B0);
traits.madd(A0,B0,C1,T0);
traits.madd(A1,B0,C5,B0);
traits.loadLhs(&blA[2*LhsProgress], A0);
traits.loadLhs(&blA[3*LhsProgress], A1);
traits.loadRhs(&blB[2*RhsProgress], B_0);
traits.madd(A0,B_0,C0,T0);
traits.madd(A1,B_0,C4,B_0);
traits.loadRhs(&blB[3*RhsProgress], B_0);
traits.madd(A0,B_0,C1,T0);
traits.madd(A1,B_0,C5,B_0);
traits.loadRhs(&blB[2*RhsProgress], B0);
traits.madd(A0,B0,C0,T0);
traits.madd(A1,B0,C4,B0);
traits.loadRhs(&blB[3*RhsProgress], B0);
traits.madd(A0,B0,C1,T0);
traits.madd(A1,B0,C5,B0);
traits.loadLhs(&blA[4*LhsProgress], A0);
traits.loadLhs(&blA[5*LhsProgress], A1);
traits.loadRhs(&blB[4*RhsProgress], B_0);
traits.madd(A0,B_0,C0,T0);
traits.madd(A1,B_0,C4,B_0);
traits.loadRhs(&blB[5*RhsProgress], B_0);
traits.madd(A0,B_0,C1,T0);
traits.madd(A1,B_0,C5,B_0);
traits.loadRhs(&blB[4*RhsProgress], B0);
traits.madd(A0,B0,C0,T0);
traits.madd(A1,B0,C4,B0);
traits.loadRhs(&blB[5*RhsProgress], B0);
traits.madd(A0,B0,C1,T0);
traits.madd(A1,B0,C5,B0);
traits.loadLhs(&blA[6*LhsProgress], A0);
traits.loadLhs(&blA[7*LhsProgress], A1);
traits.loadRhs(&blB[6*RhsProgress], B_0);
traits.madd(A0,B_0,C0,T0);
traits.madd(A1,B_0,C4,B_0);
traits.loadRhs(&blB[7*RhsProgress], B_0);
traits.madd(A0,B_0,C1,T0);
traits.madd(A1,B_0,C5,B_0);
traits.loadRhs(&blB[6*RhsProgress], B0);
traits.madd(A0,B0,C0,T0);
traits.madd(A1,B0,C4,B0);
traits.loadRhs(&blB[7*RhsProgress], B0);
traits.madd(A0,B0,C1,T0);
traits.madd(A1,B0,C5,B0);
EIGEN_ASM_COMMENT("myend");
}
else
{
EIGEN_ASM_COMMENT("mybegin4");
LhsPacket A0, A1;
RhsPacket B_0, B1, B2, B3;
RhsPacket B0, B1, B2, B3;
RhsPacket T0;
traits.loadLhs(&blA[0*LhsProgress], A0);
traits.loadLhs(&blA[1*LhsProgress], A1);
traits.loadRhs(&blB[0*RhsProgress], B_0);
traits.loadRhs(&blB[0*RhsProgress], B0);
traits.loadRhs(&blB[1*RhsProgress], B1);
traits.madd(A0,B_0,C0,T0);
traits.madd(A0,B0,C0,T0);
traits.loadRhs(&blB[2*RhsProgress], B2);
traits.madd(A1,B_0,C4,B_0);
traits.madd(A1,B0,C4,B0);
traits.loadRhs(&blB[3*RhsProgress], B3);
traits.loadRhs(&blB[4*RhsProgress], B_0);
traits.loadRhs(&blB[4*RhsProgress], B0);
traits.madd(A0,B1,C1,T0);
traits.madd(A1,B1,C5,B1);
traits.loadRhs(&blB[5*RhsProgress], B1);
@@ -667,9 +667,9 @@ EIGEN_ASM_COMMENT("mybegin4");
traits.madd(A1,B3,C7,B3);
traits.loadLhs(&blA[3*LhsProgress], A1);
traits.loadRhs(&blB[7*RhsProgress], B3);
traits.madd(A0,B_0,C0,T0);
traits.madd(A1,B_0,C4,B_0);
traits.loadRhs(&blB[8*RhsProgress], B_0);
traits.madd(A0,B0,C0,T0);
traits.madd(A1,B0,C4,B0);
traits.loadRhs(&blB[8*RhsProgress], B0);
traits.madd(A0,B1,C1,T0);
traits.madd(A1,B1,C5,B1);
traits.loadRhs(&blB[9*RhsProgress], B1);
@@ -682,9 +682,9 @@ EIGEN_ASM_COMMENT("mybegin4");
traits.loadLhs(&blA[5*LhsProgress], A1);
traits.loadRhs(&blB[11*RhsProgress], B3);
traits.madd(A0,B_0,C0,T0);
traits.madd(A1,B_0,C4,B_0);
traits.loadRhs(&blB[12*RhsProgress], B_0);
traits.madd(A0,B0,C0,T0);
traits.madd(A1,B0,C4,B0);
traits.loadRhs(&blB[12*RhsProgress], B0);
traits.madd(A0,B1,C1,T0);
traits.madd(A1,B1,C5,B1);
traits.loadRhs(&blB[13*RhsProgress], B1);
@@ -696,8 +696,8 @@ EIGEN_ASM_COMMENT("mybegin4");
traits.madd(A1,B3,C7,B3);
traits.loadLhs(&blA[7*LhsProgress], A1);
traits.loadRhs(&blB[15*RhsProgress], B3);
traits.madd(A0,B_0,C0,T0);
traits.madd(A1,B_0,C4,B_0);
traits.madd(A0,B0,C0,T0);
traits.madd(A1,B0,C4,B0);
traits.madd(A0,B1,C1,T0);
traits.madd(A1,B1,C5,B1);
traits.madd(A0,B2,C2,T0);
@@ -715,32 +715,32 @@ EIGEN_ASM_COMMENT("mybegin4");
if(nr==2)
{
LhsPacket A0, A1;
RhsPacket B_0;
RhsPacket B0;
RhsPacket T0;
traits.loadLhs(&blA[0*LhsProgress], A0);
traits.loadLhs(&blA[1*LhsProgress], A1);
traits.loadRhs(&blB[0*RhsProgress], B_0);
traits.madd(A0,B_0,C0,T0);
traits.madd(A1,B_0,C4,B_0);
traits.loadRhs(&blB[1*RhsProgress], B_0);
traits.madd(A0,B_0,C1,T0);
traits.madd(A1,B_0,C5,B_0);
traits.loadRhs(&blB[0*RhsProgress], B0);
traits.madd(A0,B0,C0,T0);
traits.madd(A1,B0,C4,B0);
traits.loadRhs(&blB[1*RhsProgress], B0);
traits.madd(A0,B0,C1,T0);
traits.madd(A1,B0,C5,B0);
}
else
{
LhsPacket A0, A1;
RhsPacket B_0, B1, B2, B3;
RhsPacket B0, B1, B2, B3;
RhsPacket T0;
traits.loadLhs(&blA[0*LhsProgress], A0);
traits.loadLhs(&blA[1*LhsProgress], A1);
traits.loadRhs(&blB[0*RhsProgress], B_0);
traits.loadRhs(&blB[0*RhsProgress], B0);
traits.loadRhs(&blB[1*RhsProgress], B1);
traits.madd(A0,B_0,C0,T0);
traits.madd(A0,B0,C0,T0);
traits.loadRhs(&blB[2*RhsProgress], B2);
traits.madd(A1,B_0,C4,B_0);
traits.madd(A1,B0,C4,B0);
traits.loadRhs(&blB[3*RhsProgress], B3);
traits.madd(A0,B1,C1,T0);
traits.madd(A1,B1,C5,B1);
@@ -827,42 +827,42 @@ EIGEN_ASM_COMMENT("mybegin4");
if(nr==2)
{
LhsPacket A0;
RhsPacket B_0, B1;
RhsPacket B0, B1;
traits.loadLhs(&blA[0*LhsProgress], A0);
traits.loadRhs(&blB[0*RhsProgress], B_0);
traits.loadRhs(&blB[0*RhsProgress], B0);
traits.loadRhs(&blB[1*RhsProgress], B1);
traits.madd(A0,B_0,C0,B_0);
traits.loadRhs(&blB[2*RhsProgress], B_0);
traits.madd(A0,B0,C0,B0);
traits.loadRhs(&blB[2*RhsProgress], B0);
traits.madd(A0,B1,C1,B1);
traits.loadLhs(&blA[1*LhsProgress], A0);
traits.loadRhs(&blB[3*RhsProgress], B1);
traits.madd(A0,B_0,C0,B_0);
traits.loadRhs(&blB[4*RhsProgress], B_0);
traits.madd(A0,B0,C0,B0);
traits.loadRhs(&blB[4*RhsProgress], B0);
traits.madd(A0,B1,C1,B1);
traits.loadLhs(&blA[2*LhsProgress], A0);
traits.loadRhs(&blB[5*RhsProgress], B1);
traits.madd(A0,B_0,C0,B_0);
traits.loadRhs(&blB[6*RhsProgress], B_0);
traits.madd(A0,B0,C0,B0);
traits.loadRhs(&blB[6*RhsProgress], B0);
traits.madd(A0,B1,C1,B1);
traits.loadLhs(&blA[3*LhsProgress], A0);
traits.loadRhs(&blB[7*RhsProgress], B1);
traits.madd(A0,B_0,C0,B_0);
traits.madd(A0,B0,C0,B0);
traits.madd(A0,B1,C1,B1);
}
else
{
LhsPacket A0;
RhsPacket B_0, B1, B2, B3;
RhsPacket B0, B1, B2, B3;
traits.loadLhs(&blA[0*LhsProgress], A0);
traits.loadRhs(&blB[0*RhsProgress], B_0);
traits.loadRhs(&blB[0*RhsProgress], B0);
traits.loadRhs(&blB[1*RhsProgress], B1);
traits.madd(A0,B_0,C0,B_0);
traits.madd(A0,B0,C0,B0);
traits.loadRhs(&blB[2*RhsProgress], B2);
traits.loadRhs(&blB[3*RhsProgress], B3);
traits.loadRhs(&blB[4*RhsProgress], B_0);
traits.loadRhs(&blB[4*RhsProgress], B0);
traits.madd(A0,B1,C1,B1);
traits.loadRhs(&blB[5*RhsProgress], B1);
traits.madd(A0,B2,C2,B2);
@@ -870,8 +870,8 @@ EIGEN_ASM_COMMENT("mybegin4");
traits.madd(A0,B3,C3,B3);
traits.loadLhs(&blA[1*LhsProgress], A0);
traits.loadRhs(&blB[7*RhsProgress], B3);
traits.madd(A0,B_0,C0,B_0);
traits.loadRhs(&blB[8*RhsProgress], B_0);
traits.madd(A0,B0,C0,B0);
traits.loadRhs(&blB[8*RhsProgress], B0);
traits.madd(A0,B1,C1,B1);
traits.loadRhs(&blB[9*RhsProgress], B1);
traits.madd(A0,B2,C2,B2);
@@ -880,8 +880,8 @@ EIGEN_ASM_COMMENT("mybegin4");
traits.loadLhs(&blA[2*LhsProgress], A0);
traits.loadRhs(&blB[11*RhsProgress], B3);
traits.madd(A0,B_0,C0,B_0);
traits.loadRhs(&blB[12*RhsProgress], B_0);
traits.madd(A0,B0,C0,B0);
traits.loadRhs(&blB[12*RhsProgress], B0);
traits.madd(A0,B1,C1,B1);
traits.loadRhs(&blB[13*RhsProgress], B1);
traits.madd(A0,B2,C2,B2);
@@ -890,7 +890,7 @@ EIGEN_ASM_COMMENT("mybegin4");
traits.loadLhs(&blA[3*LhsProgress], A0);
traits.loadRhs(&blB[15*RhsProgress], B3);
traits.madd(A0,B_0,C0,B_0);
traits.madd(A0,B0,C0,B0);
traits.madd(A0,B1,C1,B1);
traits.madd(A0,B2,C2,B2);
traits.madd(A0,B3,C3,B3);
@@ -905,26 +905,26 @@ EIGEN_ASM_COMMENT("mybegin4");
if(nr==2)
{
LhsPacket A0;
RhsPacket B_0, B1;
RhsPacket B0, B1;
traits.loadLhs(&blA[0*LhsProgress], A0);
traits.loadRhs(&blB[0*RhsProgress], B_0);
traits.loadRhs(&blB[0*RhsProgress], B0);
traits.loadRhs(&blB[1*RhsProgress], B1);
traits.madd(A0,B_0,C0,B_0);
traits.madd(A0,B0,C0,B0);
traits.madd(A0,B1,C1,B1);
}
else
{
LhsPacket A0;
RhsPacket B_0, B1, B2, B3;
RhsPacket B0, B1, B2, B3;
traits.loadLhs(&blA[0*LhsProgress], A0);
traits.loadRhs(&blB[0*RhsProgress], B_0);
traits.loadRhs(&blB[0*RhsProgress], B0);
traits.loadRhs(&blB[1*RhsProgress], B1);
traits.loadRhs(&blB[2*RhsProgress], B2);
traits.loadRhs(&blB[3*RhsProgress], B3);
traits.madd(A0,B_0,C0,B_0);
traits.madd(A0,B0,C0,B0);
traits.madd(A0,B1,C1,B1);
traits.madd(A0,B2,C2,B2);
traits.madd(A0,B3,C3,B3);
@@ -971,26 +971,26 @@ EIGEN_ASM_COMMENT("mybegin4");
if(nr==2)
{
LhsScalar A0;
RhsScalar B_0, B1;
RhsScalar B0, B1;
A0 = blA[k];
B_0 = blB[0];
B0 = blB[0];
B1 = blB[1];
MADD(cj,A0,B_0,C0,B_0);
MADD(cj,A0,B0,C0,B0);
MADD(cj,A0,B1,C1,B1);
}
else
{
LhsScalar A0;
RhsScalar B_0, B1, B2, B3;
RhsScalar B0, B1, B2, B3;
A0 = blA[k];
B_0 = blB[0];
B0 = blB[0];
B1 = blB[1];
B2 = blB[2];
B3 = blB[3];
MADD(cj,A0,B_0,C0,B_0);
MADD(cj,A0,B0,C0,B0);
MADD(cj,A0,B1,C1,B1);
MADD(cj,A0,B2,C2,B2);
MADD(cj,A0,B3,C3,B3);
@@ -1027,14 +1027,14 @@ EIGEN_ASM_COMMENT("mybegin4");
for(Index k=0; k<depth; k++)
{
LhsPacket A0, A1;
RhsPacket B_0;
RhsPacket B0;
RhsPacket T0;
traits.loadLhs(&blA[0*LhsProgress], A0);
traits.loadLhs(&blA[1*LhsProgress], A1);
traits.loadRhs(&blB[0*RhsProgress], B_0);
traits.madd(A0,B_0,C0,T0);
traits.madd(A1,B_0,C4,B_0);
traits.loadRhs(&blB[0*RhsProgress], B0);
traits.madd(A0,B0,C0,T0);
traits.madd(A1,B0,C4,B0);
blB += RhsProgress;
blA += 2*LhsProgress;
@@ -1066,10 +1066,10 @@ EIGEN_ASM_COMMENT("mybegin4");
for(Index k=0; k<depth; k++)
{
LhsPacket A0;
RhsPacket B_0;
RhsPacket B0;
traits.loadLhs(blA, A0);
traits.loadRhs(blB, B_0);
traits.madd(A0, B_0, C0, B_0);
traits.loadRhs(blB, B0);
traits.madd(A0, B0, C0, B0);
blB += RhsProgress;
blA += LhsProgress;
}
@@ -1091,8 +1091,8 @@ EIGEN_ASM_COMMENT("mybegin4");
for(Index k=0; k<depth; k++)
{
LhsScalar A0 = blA[k];
RhsScalar B_0 = blB[k];
MADD(cj, A0, B_0, C0, B_0);
RhsScalar B0 = blB[k];
MADD(cj, A0, B0, C0, B0);
}
res[(j2+0)*resStride + i] += alpha*C0;
}

View File

@@ -99,7 +99,7 @@ struct product_triangular_matrix_matrix<Scalar,Index,Mode,true,
typedef gebp_traits<Scalar,Scalar> Traits;
enum {
SmallPanelWidth = 2 * EIGEN_PLAIN_ENUM_MAX(Traits::mr,Traits::nr),
SmallPanelWidth = EIGEN_PLAIN_ENUM_MAX(Traits::mr,Traits::nr),
IsLower = (Mode&Lower) == Lower,
SetDiag = (Mode&(ZeroDiag|UnitDiag)) ? 0 : 1
};

View File

@@ -36,16 +36,12 @@ struct product_triangular_matrix_vector<Index,Mode,LhsScalar,ConjLhs,RhsScalar,C
typedef typename scalar_product_traits<LhsScalar, RhsScalar>::ReturnType ResScalar;
enum {
IsLower = ((Mode&Lower)==Lower),
HasUnitDiag = (Mode & UnitDiag)==UnitDiag,
HasZeroDiag = (Mode & ZeroDiag)==ZeroDiag
HasUnitDiag = (Mode & UnitDiag)==UnitDiag
};
static EIGEN_DONT_INLINE void run(Index _rows, Index _cols, const LhsScalar* _lhs, Index lhsStride,
static EIGEN_DONT_INLINE void run(Index rows, Index cols, const LhsScalar* _lhs, Index lhsStride,
const RhsScalar* _rhs, Index rhsIncr, ResScalar* _res, Index resIncr, ResScalar alpha)
{
static const Index PanelWidth = EIGEN_TUNE_TRIANGULAR_PANEL_WIDTH;
Index size = (std::min)(_rows,_cols);
Index rows = IsLower ? _rows : (std::min)(_rows,_cols);
Index cols = IsLower ? (std::min)(_rows,_cols) : _cols;
typedef Map<const Matrix<LhsScalar,Dynamic,Dynamic,ColMajor>, 0, OuterStride<> > LhsMap;
const LhsMap lhs(_lhs,rows,cols,OuterStride<>(lhsStride));
@@ -58,20 +54,20 @@ struct product_triangular_matrix_vector<Index,Mode,LhsScalar,ConjLhs,RhsScalar,C
typedef Map<Matrix<ResScalar,Dynamic,1> > ResMap;
ResMap res(_res,rows);
for (Index pi=0; pi<size; pi+=PanelWidth)
for (Index pi=0; pi<cols; pi+=PanelWidth)
{
Index actualPanelWidth = (std::min)(PanelWidth, size-pi);
Index actualPanelWidth = (std::min)(PanelWidth, cols-pi);
for (Index k=0; k<actualPanelWidth; ++k)
{
Index i = pi + k;
Index s = IsLower ? ((HasUnitDiag||HasZeroDiag) ? i+1 : i ) : pi;
Index s = IsLower ? (HasUnitDiag ? i+1 : i ) : pi;
Index r = IsLower ? actualPanelWidth-k : k+1;
if ((!(HasUnitDiag||HasZeroDiag)) || (--r)>0)
if ((!HasUnitDiag) || (--r)>0)
res.segment(s,r) += (alpha * cjRhs.coeff(i)) * cjLhs.col(i).segment(s,r);
if (HasUnitDiag)
res.coeffRef(i) += alpha * cjRhs.coeff(i);
}
Index r = IsLower ? rows - pi - actualPanelWidth : pi;
Index r = IsLower ? cols - pi - actualPanelWidth : pi;
if (r>0)
{
Index s = IsLower ? pi+actualPanelWidth : 0;
@@ -82,14 +78,6 @@ struct product_triangular_matrix_vector<Index,Mode,LhsScalar,ConjLhs,RhsScalar,C
&res.coeffRef(s), resIncr, alpha);
}
}
if((!IsLower) && cols>size)
{
general_matrix_vector_product<Index,LhsScalar,ColMajor,ConjLhs,RhsScalar,ConjRhs>::run(
rows, cols-size,
&lhs.coeffRef(0,size), lhsStride,
&rhs.coeffRef(size), rhsIncr,
_res, resIncr, alpha);
}
}
};
@@ -99,16 +87,12 @@ struct product_triangular_matrix_vector<Index,Mode,LhsScalar,ConjLhs,RhsScalar,C
typedef typename scalar_product_traits<LhsScalar, RhsScalar>::ReturnType ResScalar;
enum {
IsLower = ((Mode&Lower)==Lower),
HasUnitDiag = (Mode & UnitDiag)==UnitDiag,
HasZeroDiag = (Mode & ZeroDiag)==ZeroDiag
HasUnitDiag = (Mode & UnitDiag)==UnitDiag
};
static void run(Index _rows, Index _cols, const LhsScalar* _lhs, Index lhsStride,
static void run(Index rows, Index cols, const LhsScalar* _lhs, Index lhsStride,
const RhsScalar* _rhs, Index rhsIncr, ResScalar* _res, Index resIncr, ResScalar alpha)
{
static const Index PanelWidth = EIGEN_TUNE_TRIANGULAR_PANEL_WIDTH;
Index diagSize = (std::min)(_rows,_cols);
Index rows = IsLower ? _rows : diagSize;
Index cols = IsLower ? diagSize : _cols;
typedef Map<const Matrix<LhsScalar,Dynamic,Dynamic,RowMajor>, 0, OuterStride<> > LhsMap;
const LhsMap lhs(_lhs,rows,cols,OuterStride<>(lhsStride));
@@ -121,15 +105,15 @@ struct product_triangular_matrix_vector<Index,Mode,LhsScalar,ConjLhs,RhsScalar,C
typedef Map<Matrix<ResScalar,Dynamic,1>, 0, InnerStride<> > ResMap;
ResMap res(_res,rows,InnerStride<>(resIncr));
for (Index pi=0; pi<diagSize; pi+=PanelWidth)
for (Index pi=0; pi<cols; pi+=PanelWidth)
{
Index actualPanelWidth = (std::min)(PanelWidth, diagSize-pi);
Index actualPanelWidth = (std::min)(PanelWidth, cols-pi);
for (Index k=0; k<actualPanelWidth; ++k)
{
Index i = pi + k;
Index s = IsLower ? pi : ((HasUnitDiag||HasZeroDiag) ? i+1 : i);
Index s = IsLower ? pi : (HasUnitDiag ? i+1 : i);
Index r = IsLower ? k+1 : actualPanelWidth-k;
if ((!(HasUnitDiag||HasZeroDiag)) || (--r)>0)
if ((!HasUnitDiag) || (--r)>0)
res.coeffRef(i) += alpha * (cjLhs.row(i).segment(s,r).cwiseProduct(cjRhs.segment(s,r).transpose())).sum();
if (HasUnitDiag)
res.coeffRef(i) += alpha * cjRhs.coeff(i);
@@ -145,14 +129,6 @@ struct product_triangular_matrix_vector<Index,Mode,LhsScalar,ConjLhs,RhsScalar,C
&res.coeffRef(pi), resIncr, alpha);
}
}
if(IsLower && rows>diagSize)
{
general_matrix_vector_product<Index,LhsScalar,RowMajor,ConjLhs,RhsScalar,ConjRhs>::run(
rows-diagSize, cols,
&lhs.coeffRef(diagSize,0), lhsStride,
&rhs.coeffRef(0), rhsIncr,
&res.coeffRef(diagSize), resIncr, alpha);
}
}
};
@@ -204,7 +180,7 @@ struct TriangularProduct<Mode,false,Lhs,true,Rhs,false>
{
eigen_assert(dst.rows()==m_lhs.rows() && dst.cols()==m_rhs.cols());
typedef TriangularProduct<(Mode & (UnitDiag|ZeroDiag)) | ((Mode & Lower) ? Upper : Lower),true,Transpose<const Rhs>,false,Transpose<const Lhs>,true> TriangularProductTranspose;
typedef TriangularProduct<(Mode & UnitDiag) | ((Mode & Lower) ? Upper : Lower),true,Transpose<const Rhs>,false,Transpose<const Lhs>,true> TriangularProductTranspose;
Transpose<Dest> dstT(dst);
internal::trmv_selector<(int(internal::traits<Rhs>::Flags)&RowMajorBit) ? ColMajor : RowMajor>::run(
TriangularProductTranspose(m_rhs.transpose(),m_lhs.transpose()), dstT, alpha);

View File

@@ -75,20 +75,12 @@ struct triangular_solve_matrix<Scalar,Index,OnTheLeft,Mode,Conjugate,TriStorageO
ei_declare_aligned_stack_constructed_variable(Scalar, blockA, kc*mc, 0);
ei_declare_aligned_stack_constructed_variable(Scalar, allocatedBlockB, sizeB, 0);
Scalar* blockB = allocatedBlockB + sizeW;
Scalar* blockW = allocatedBlockB;
conj_if<Conjugate> conj;
gebp_kernel<Scalar, Scalar, Index, Traits::mr, Traits::nr, Conjugate, false> gebp_kernel;
gemm_pack_lhs<Scalar, Index, Traits::mr, Traits::LhsProgress, TriStorageOrder> pack_lhs;
gemm_pack_rhs<Scalar, Index, Traits::nr, ColMajor, false, true> pack_rhs;
// the goal here is to subdivise the Rhs panels such that we keep some cache
// coherence when accessing the rhs elements
std::ptrdiff_t l1, l2;
manage_caching_sizes(GetAction, &l1, &l2);
Index subcols = cols>0 ? l2/(4 * sizeof(Scalar) * otherStride) : 0;
subcols = std::max<Index>((subcols/Traits::nr)*Traits::nr, Traits::nr);
for(Index k2=IsLower ? 0 : size;
IsLower ? k2<size : k2>0;
IsLower ? k2+=kc : k2-=kc)
@@ -100,18 +92,16 @@ struct triangular_solve_matrix<Scalar,Index,OnTheLeft,Mode,Conjugate,TriStorageO
// A11 (the triangular part) and A21 the remaining rectangular part.
// Then the high level algorithm is:
// - B = R1 => general block copy (done during the next step)
// - R1 = A11^-1 B => tricky part
// - R1 = L1^-1 B => tricky part
// - update B from the new R1 => actually this has to be performed continuously during the above step
// - R2 -= A21 * B => GEPP
// - R2 = L2 * B => GEPP
// The tricky part: compute R1 = A11^-1 B while updating B from R1
// The idea is to split A11 into multiple small vertical panels.
// Each panel can be split into a small triangular part T1k which is processed without optimization,
// and the remaining small part T2k which is processed using gebp with appropriate block strides
for(Index j2=0; j2<cols; j2+=subcols)
// The tricky part: compute R1 = L1^-1 B while updating B from R1
// The idea is to split L1 into multiple small vertical panels.
// Each panel can be split into a small triangular part A1 which is processed without optimization,
// and the remaining small part A2 which is processed using gebp with appropriate block strides
{
Index actual_cols = (std::min)(cols-j2,subcols);
// for each small vertical panels [T1k^T, T2k^T]^T of lhs
// for each small vertical panels of lhs
for (Index k1=0; k1<actual_kc; k1+=SmallPanelWidth)
{
Index actualPanelWidth = std::min<Index>(actual_kc-k1, SmallPanelWidth);
@@ -124,7 +114,7 @@ struct triangular_solve_matrix<Scalar,Index,OnTheLeft,Mode,Conjugate,TriStorageO
Index rs = actualPanelWidth - k - 1; // remaining size
Scalar a = (Mode & UnitDiag) ? Scalar(1) : Scalar(1)/conj(tri(i,i));
for (Index j=j2; j<j2+actual_cols; ++j)
for (Index j=0; j<cols; ++j)
{
if (TriStorageOrder==RowMajor)
{
@@ -153,7 +143,7 @@ struct triangular_solve_matrix<Scalar,Index,OnTheLeft,Mode,Conjugate,TriStorageO
Index blockBOffset = IsLower ? k1 : lengthTarget;
// update the respective rows of B from other
pack_rhs(blockB+actual_kc*j2, &other(startBlock,j2), otherStride, actualPanelWidth, actual_cols, actual_kc, blockBOffset);
pack_rhs(blockB, _other+startBlock, otherStride, actualPanelWidth, cols, actual_kc, blockBOffset);
// GEBP
if (lengthTarget>0)
@@ -162,13 +152,13 @@ struct triangular_solve_matrix<Scalar,Index,OnTheLeft,Mode,Conjugate,TriStorageO
pack_lhs(blockA, &tri(startTarget,startBlock), triStride, actualPanelWidth, lengthTarget);
gebp_kernel(&other(startTarget,j2), otherStride, blockA, blockB+actual_kc*j2, lengthTarget, actualPanelWidth, actual_cols, Scalar(-1),
actualPanelWidth, actual_kc, 0, blockBOffset, blockW);
gebp_kernel(_other+startTarget, otherStride, blockA, blockB, lengthTarget, actualPanelWidth, cols, Scalar(-1),
actualPanelWidth, actual_kc, 0, blockBOffset);
}
}
}
// R2 -= A21 * B => GEPP
// R2 = A2 * B => GEPP
{
Index start = IsLower ? k2+kc : 0;
Index end = IsLower ? size : k2-kc;

View File

@@ -200,6 +200,8 @@ enum {
Aligned=1
};
enum { ConditionalJumpCost = 5 };
/** \ingroup enums
* Enum used by DenseBase::corner() in Eigen2 compatibility mode. */
// FIXME after the corner() API change, this was not needed anymore, except by AlignedBox
@@ -221,6 +223,8 @@ enum DirectionType {
BothDirections
};
enum ProductEvaluationMode { NormalProduct, CacheFriendlyProduct };
/** \internal \ingroup enums
* Enum to specify how to traverse the entries of a matrix. */
enum {
@@ -253,13 +257,6 @@ enum {
CompleteUnrolling
};
/** \internal \ingroup enums
* Enum to specify whether to use the default (built-in) implementation or the specialization. */
enum {
Specialized,
BuiltIn
};
/** \ingroup enums
* Enum containing possible values for the \p _Options template parameter of
* Matrix, Array and BandMatrix. */
@@ -386,10 +383,7 @@ enum ComputationInfo {
/** The provided data did not satisfy the prerequisites. */
NumericalIssue = 1,
/** Iterative procedure did not converge. */
NoConvergence = 2,
/** The inputs are invalid, or the algorithm has been properly called.
* When assertions are enabled, such errors trigger an assert. */
InvalidInput = 3
NoConvergence = 2
};
/** \ingroup enums

View File

@@ -133,7 +133,6 @@ template<typename ExpressionType> class WithFormat;
template<typename MatrixType> struct CommaInitializer;
template<typename Derived> class ReturnByValue;
template<typename ExpressionType> class ArrayWrapper;
template<typename ExpressionType> class MatrixWrapper;
namespace internal {
template<typename DecompositionType, typename Rhs> struct solve_retval_base;
@@ -283,8 +282,6 @@ template<typename MatrixType,int Direction> class Homogeneous;
// MatrixFunctions module
template<typename Derived> struct MatrixExponentialReturnValue;
template<typename Derived> class MatrixFunctionReturnValue;
template<typename Derived> class MatrixSquareRootReturnValue;
template<typename Derived> class MatrixLogarithmReturnValue;
namespace internal {
template <typename Scalar>

View File

@@ -28,7 +28,7 @@
#define EIGEN_WORLD_VERSION 3
#define EIGEN_MAJOR_VERSION 0
#define EIGEN_MINOR_VERSION 91
#define EIGEN_MINOR_VERSION 2
#define EIGEN_VERSION_AT_LEAST(x,y,z) (EIGEN_WORLD_VERSION>x || (EIGEN_WORLD_VERSION>=x && \
(EIGEN_MAJOR_VERSION>y || (EIGEN_MAJOR_VERSION>=y && \
@@ -45,7 +45,7 @@
#define EIGEN_GNUC_AT_MOST(x,y) 0
#endif
#if EIGEN_GNUC_AT_MOST(4,3) && !defined(__clang__)
#if EIGEN_GNUC_AT_MOST(4,3)
// see bug 89
#define EIGEN_SAFE_TO_USE_STANDARD_ASSERT_MACRO 0
#else
@@ -130,34 +130,31 @@
#define EIGEN_MAKESTRING2(a) #a
#define EIGEN_MAKESTRING(a) EIGEN_MAKESTRING2(a)
// EIGEN_ALWAYS_INLINE_ATTRIB should be use in the declaration of function
// which should be inlined even in debug mode.
// FIXME with the always_inline attribute,
// gcc 3.4.x reports the following compilation error:
// Eval.h:91: sorry, unimplemented: inlining failed in call to 'const Eigen::Eval<Derived> Eigen::MatrixBase<Scalar, Derived>::eval() const'
// : function body not available
#if EIGEN_GNUC_AT_LEAST(4,0)
#define EIGEN_ALWAYS_INLINE_ATTRIB __attribute__((always_inline))
#else
#define EIGEN_ALWAYS_INLINE_ATTRIB
#endif
#if EIGEN_GNUC_AT_LEAST(4,1) && !defined(__clang__) && !defined(__INTEL_COMPILER)
#define EIGEN_FLATTEN_ATTRIB __attribute__((flatten))
#else
#define EIGEN_FLATTEN_ATTRIB
#endif
// EIGEN_STRONG_INLINE is a stronger version of the inline, using __forceinline on MSVC,
// but it still doesn't use GCC's always_inline. This is useful in (common) situations where MSVC needs forceinline
// but GCC is still doing fine with just inline.
// EIGEN_FORCE_INLINE means "inline as much as possible"
#if (defined _MSC_VER) || (defined __INTEL_COMPILER)
#define EIGEN_STRONG_INLINE __forceinline
#else
#define EIGEN_STRONG_INLINE inline
#endif
// EIGEN_ALWAYS_INLINE is the stronget, it has the effect of making the function inline and adding every possible
// attribute to maximize inlining. This should only be used when really necessary: in particular,
// it uses __attribute__((always_inline)) on GCC, which most of the time is useless and can severely harm compile times.
// FIXME with the always_inline attribute,
// gcc 3.4.x reports the following compilation error:
// Eval.h:91: sorry, unimplemented: inlining failed in call to 'const Eigen::Eval<Derived> Eigen::MatrixBase<Scalar, Derived>::eval() const'
// : function body not available
#if EIGEN_GNUC_AT_LEAST(4,0)
#define EIGEN_ALWAYS_INLINE __attribute__((always_inline)) inline
#else
#define EIGEN_ALWAYS_INLINE EIGEN_STRONG_INLINE
#endif
#if (defined __GNUC__)
#define EIGEN_DONT_INLINE __attribute__((noinline))
#elif (defined _MSC_VER)
@@ -234,16 +231,12 @@
#define EIGEN_ONLY_USED_FOR_DEBUG(x)
#endif
#ifndef EIGEN_NO_DEPRECATED_WARNING
#if (defined __GNUC__)
#define EIGEN_DEPRECATED __attribute__((deprecated))
#elif (defined _MSC_VER)
#define EIGEN_DEPRECATED __declspec(deprecated)
#else
#define EIGEN_DEPRECATED
#endif
#if (defined __GNUC__)
#define EIGEN_DEPRECATED __attribute__((deprecated))
#elif (defined _MSC_VER)
#define EIGEN_DEPRECATED __declspec(deprecated)
#else
#define EIGEN_DEPRECATED
#define EIGEN_DEPRECATED
#endif
#if (defined __GNUC__)
@@ -256,7 +249,7 @@
#define EIGEN_UNUSED_VARIABLE(var) (void)var;
#if (defined __GNUC__)
#define EIGEN_ASM_COMMENT(X) asm("#" X)
#define EIGEN_ASM_COMMENT(X) asm("#"X)
#else
#define EIGEN_ASM_COMMENT(X)
#endif

View File

@@ -82,16 +82,6 @@
namespace internal {
inline void throw_std_bad_alloc()
{
#ifdef EIGEN_EXCEPTIONS
throw std::bad_alloc();
#else
std::size_t huge = -1;
new int[huge];
#endif
}
/*****************************************************************************
*** Implementation of handmade aligned functions ***
*****************************************************************************/
@@ -202,7 +192,7 @@ inline void check_that_malloc_is_allowed()
#endif
/** \internal Allocates \a size bytes. The returned pointer is guaranteed to have 16 bytes alignment.
* On allocation error, the returned pointer is null, and std::bad_alloc is thrown.
* On allocation error, the returned pointer is null, and if exceptions are enabled then a std::bad_alloc is thrown.
*/
inline void* aligned_malloc(size_t size)
{
@@ -223,9 +213,10 @@ inline void* aligned_malloc(size_t size)
result = handmade_aligned_malloc(size);
#endif
if(!result && size)
throw_std_bad_alloc();
#ifdef EIGEN_EXCEPTIONS
if(result == 0)
throw std::bad_alloc();
#endif
return result;
}
@@ -250,7 +241,7 @@ inline void aligned_free(void *ptr)
/**
* \internal
* \brief Reallocates an aligned block of memory.
* \throws std::bad_alloc on allocation failure
* \throws std::bad_alloc if EIGEN_EXCEPTIONS are defined.
**/
inline void* aligned_realloc(void *ptr, size_t new_size, size_t old_size)
{
@@ -278,9 +269,10 @@ inline void* aligned_realloc(void *ptr, size_t new_size, size_t old_size)
result = handmade_aligned_realloc(ptr,new_size,old_size);
#endif
if (!result && new_size)
throw_std_bad_alloc();
#ifdef EIGEN_EXCEPTIONS
if (result==0 && new_size!=0)
throw std::bad_alloc();
#endif
return result;
}
@@ -289,7 +281,7 @@ inline void* aligned_realloc(void *ptr, size_t new_size, size_t old_size)
*****************************************************************************/
/** \internal Allocates \a size bytes. If Align is true, then the returned ptr is 16-byte-aligned.
* On allocation error, the returned pointer is null, and a std::bad_alloc is thrown.
* On allocation error, the returned pointer is null, and if exceptions are enabled then a std::bad_alloc is thrown.
*/
template<bool Align> inline void* conditional_aligned_malloc(size_t size)
{
@@ -301,8 +293,9 @@ template<> inline void* conditional_aligned_malloc<false>(size_t size)
check_that_malloc_is_allowed();
void *result = std::malloc(size);
if(!result && size)
throw_std_bad_alloc();
#ifdef EIGEN_EXCEPTIONS
if(!result) throw std::bad_alloc();
#endif
return result;
}
@@ -354,27 +347,18 @@ template<typename T> inline void destruct_elements_of_array(T *ptr, size_t size)
*** Implementation of aligned new/delete-like functions ***
*****************************************************************************/
template<typename T>
EIGEN_ALWAYS_INLINE void check_size_for_overflow(size_t size)
{
if(size > size_t(-1) / sizeof(T))
throw_std_bad_alloc();
}
/** \internal Allocates \a size objects of type T. The returned pointer is guaranteed to have 16 bytes alignment.
* On allocation error, the returned pointer is undefined, but a std::bad_alloc is thrown.
* On allocation error, the returned pointer is undefined, but if exceptions are enabled then a std::bad_alloc is thrown.
* The default constructor of T is called.
*/
template<typename T> inline T* aligned_new(size_t size)
{
check_size_for_overflow<T>(size);
T *result = reinterpret_cast<T*>(aligned_malloc(sizeof(T)*size));
return construct_elements_of_array(result, size);
}
template<typename T, bool Align> inline T* conditional_aligned_new(size_t size)
{
check_size_for_overflow<T>(size);
T *result = reinterpret_cast<T*>(conditional_aligned_malloc<Align>(sizeof(T)*size));
return construct_elements_of_array(result, size);
}
@@ -399,8 +383,6 @@ template<typename T, bool Align> inline void conditional_aligned_delete(T *ptr,
template<typename T, bool Align> inline T* conditional_aligned_realloc_new(T* pts, size_t new_size, size_t old_size)
{
check_size_for_overflow<T>(new_size);
check_size_for_overflow<T>(old_size);
if(new_size < old_size)
destruct_elements_of_array(pts+new_size, old_size-new_size);
T *result = reinterpret_cast<T*>(conditional_aligned_realloc<Align>(reinterpret_cast<void*>(pts), sizeof(T)*new_size, sizeof(T)*old_size));
@@ -412,7 +394,6 @@ template<typename T, bool Align> inline T* conditional_aligned_realloc_new(T* pt
template<typename T, bool Align> inline T* conditional_aligned_new_auto(size_t size)
{
check_size_for_overflow<T>(size);
T *result = reinterpret_cast<T*>(conditional_aligned_malloc<Align>(sizeof(T)*size));
if(NumTraits<T>::RequireInitialization)
construct_elements_of_array(result, size);
@@ -421,8 +402,6 @@ template<typename T, bool Align> inline T* conditional_aligned_new_auto(size_t s
template<typename T, bool Align> inline T* conditional_aligned_realloc_new_auto(T* pts, size_t new_size, size_t old_size)
{
check_size_for_overflow<T>(new_size);
check_size_for_overflow<T>(old_size);
if(NumTraits<T>::RequireInitialization && (new_size < old_size))
destruct_elements_of_array(pts+new_size, old_size-new_size);
T *result = reinterpret_cast<T*>(conditional_aligned_realloc<Align>(reinterpret_cast<void*>(pts), sizeof(T)*new_size, sizeof(T)*old_size));
@@ -483,27 +462,6 @@ inline static Index first_aligned(const Scalar* array, Index size)
}
}
// std::copy is much slower than memcpy, so let's introduce a smart_copy which
// use memcpy on trivial types, i.e., on types that does not require an initialization ctor.
template<typename T, bool UseMemcpy> struct smart_copy_helper;
template<typename T> void smart_copy(const T* start, const T* end, T* target)
{
smart_copy_helper<T,!NumTraits<T>::RequireInitialization>::run(start, end, target);
}
template<typename T> struct smart_copy_helper<T,true> {
inline static void run(const T* start, const T* end, T* target)
{ memcpy(target, start, std::ptrdiff_t(end)-std::ptrdiff_t(start)); }
};
template<typename T> struct smart_copy_helper<T,false> {
inline static void run(const T* start, const T* end, T* target)
{ std::copy(start, end, target); }
};
} // end namespace internal
/*****************************************************************************
@@ -559,7 +517,7 @@ template<typename T> class aligned_stack_memory_handler
* if SIZE is smaller than EIGEN_STACK_ALLOCATION_LIMIT, and if stack allocation is supported by the platform
* (currently, this is Linux and Visual Studio only). Otherwise the memory is allocated on the heap.
* The allocated buffer is automatically deleted when exiting the scope of this declaration.
* If BUFFER is non null, then the declared variable is simply an alias for BUFFER, and no allocation/deletion occurs.
* If BUFFER is non nul, then the declared variable is simply an alias for BUFFER, and no allocation/deletion occurs.
* Here is an example:
* \code
* {
@@ -578,7 +536,6 @@ template<typename T> class aligned_stack_memory_handler
#endif
#define ei_declare_aligned_stack_constructed_variable(TYPE,NAME,SIZE,BUFFER) \
Eigen::internal::check_size_for_overflow<TYPE>(SIZE); \
TYPE* NAME = (BUFFER)!=0 ? (BUFFER) \
: reinterpret_cast<TYPE*>( \
(sizeof(TYPE)*SIZE<=EIGEN_STACK_ALLOCATION_LIMIT) ? EIGEN_ALIGNED_ALLOCA(sizeof(TYPE)*SIZE) \
@@ -588,7 +545,6 @@ template<typename T> class aligned_stack_memory_handler
#else
#define ei_declare_aligned_stack_constructed_variable(TYPE,NAME,SIZE,BUFFER) \
Eigen::internal::check_size_for_overflow<TYPE>(SIZE); \
TYPE* NAME = (BUFFER)!=0 ? BUFFER : reinterpret_cast<TYPE*>(Eigen::internal::aligned_malloc(sizeof(TYPE)*SIZE)); \
Eigen::internal::aligned_stack_memory_handler<TYPE> EIGEN_CAT(NAME,_stack_memory_destructor)((BUFFER)==0 ? NAME : 0,SIZE,true)
@@ -713,7 +669,6 @@ public:
pointer allocate( size_type num, const void* hint = 0 )
{
EIGEN_UNUSED_VARIABLE(hint);
internal::check_size_for_overflow<T>(num);
return static_cast<pointer>( internal::aligned_malloc( num * sizeof(T) ) );
}

View File

@@ -70,7 +70,6 @@
YOU_CALLED_A_DYNAMIC_SIZE_METHOD_ON_A_FIXED_SIZE_MATRIX_OR_VECTOR,
UNALIGNED_LOAD_AND_STORE_OPERATIONS_UNIMPLEMENTED_ON_ALTIVEC,
THIS_FUNCTION_IS_NOT_FOR_INTEGER_NUMERIC_TYPES,
FLOATING_POINT_ARGUMENT_PASSED__INTEGER_WAS_EXPECTED,
NUMERIC_TYPE_MUST_BE_REAL,
COEFFICIENT_WRITE_ACCESS_TO_SELFADJOINT_NOT_SUPPORTED,
WRITING_TO_TRIANGULAR_PART_WITH_UNIT_DIAGONAL_IS_NOT_SUPPORTED,
@@ -96,12 +95,7 @@
YOU_PERFORMED_AN_INVALID_TRANSFORMATION_CONVERSION,
THIS_EXPRESSION_IS_NOT_A_LVALUE__IT_IS_READ_ONLY,
YOU_ARE_TRYING_TO_USE_AN_INDEX_BASED_ACCESSOR_ON_AN_EXPRESSION_THAT_DOES_NOT_SUPPORT_THAT,
THIS_METHOD_IS_ONLY_FOR_1x1_EXPRESSIONS,
THIS_METHOD_IS_ONLY_FOR_EXPRESSIONS_OF_BOOL,
THIS_METHOD_IS_ONLY_FOR_ARRAYS_NOT_MATRICES,
YOU_PASSED_A_ROW_VECTOR_BUT_A_COLUMN_VECTOR_WAS_EXPECTED,
YOU_PASSED_A_COLUMN_VECTOR_BUT_A_ROW_VECTOR_WAS_EXPECTED,
THE_INDEX_TYPE_MUST_BE_A_SIGNED_TYPE
THIS_METHOD_IS_ONLY_FOR_1x1_EXPRESSIONS
};
};
@@ -201,15 +195,4 @@
EIGEN_STATIC_ASSERT(internal::is_lvalue<Derived>::value, \
THIS_EXPRESSION_IS_NOT_A_LVALUE__IT_IS_READ_ONLY)
#define EIGEN_STATIC_ASSERT_ARRAYXPR(Derived) \
EIGEN_STATIC_ASSERT((internal::is_same<typename internal::traits<Derived>::XprKind, ArrayXpr>::value), \
THIS_METHOD_IS_ONLY_FOR_ARRAYS_NOT_MATRICES)
#define EIGEN_STATIC_ASSERT_SAME_XPR_KIND(Derived1, Derived2) \
EIGEN_STATIC_ASSERT((internal::is_same<typename internal::traits<Derived1>::XprKind, \
typename internal::traits<Derived2>::XprKind \
>::value), \
YOU_CANNOT_MIX_ARRAYS_AND_MATRICES)
#endif // EIGEN_STATIC_ASSERT_H

View File

@@ -125,9 +125,10 @@ class compute_matrix_flags
aligned_bit =
(
((Options&DontAlign)==0)
&& packet_traits<Scalar>::Vectorizable
&& (
#if EIGEN_ALIGN_STATICALLY
((!is_dynamic_size_storage) && (((MaxCols*MaxRows*sizeof(Scalar)) % 16) == 0))
((!is_dynamic_size_storage) && (((MaxCols*MaxRows) % packet_traits<Scalar>::size) == 0))
#else
0
#endif

View File

@@ -51,14 +51,14 @@ EIGEN_MAKE_ALIGNED_OPERATOR_NEW_IF_VECTORIZABLE_FIXED_SIZE(_Scalar,_AmbientDim==
{ if (AmbientDimAtCompileTime!=Dynamic) setNull(); }
/** Constructs a null box with \a _dim the dimension of the ambient space. */
inline explicit AlignedBox(int _dim) : m_min(_dim), m_max(_dim)
inline explicit AlignedBox(int _dim) : m_(min)(_dim), m_(max)(_dim)
{ setNull(); }
/** Constructs a box with extremities \a _min and \a _max. */
inline AlignedBox(const VectorType& _min, const VectorType& _max) : m_min(_min), m_max(_max) {}
inline AlignedBox(const VectorType& _min, const VectorType& _max) : m_(min)(_min), m_(max)(_max) {}
/** Constructs a box containing a single point \a p. */
inline explicit AlignedBox(const VectorType& p) : m_min(p), m_max(p) {}
inline explicit AlignedBox(const VectorType& p) : m_(min)(p), m_(max)(p) {}
~AlignedBox() {}
@@ -71,8 +71,8 @@ EIGEN_MAKE_ALIGNED_OPERATOR_NEW_IF_VECTORIZABLE_FIXED_SIZE(_Scalar,_AmbientDim==
/** Makes \c *this a null/empty box. */
inline void setNull()
{
m_min.setConstant( (std::numeric_limits<Scalar>::max)());
m_max.setConstant(-(std::numeric_limits<Scalar>::max)());
m_min.setConstant( std::numeric_limits<Scalar>::(max)());
m_max.setConstant(-std::numeric_limits<Scalar>::(max)());
}
/** \returns the minimal corner */
@@ -90,19 +90,19 @@ EIGEN_MAKE_ALIGNED_OPERATOR_NEW_IF_VECTORIZABLE_FIXED_SIZE(_Scalar,_AmbientDim==
/** \returns true if the box \a b is entirely inside the box \c *this. */
inline bool contains(const AlignedBox& b) const
{ return (m_min.cwise()<=(b.min)()).all() && ((b.max)().cwise()<=m_max).all(); }
{ return (m_min.cwise()<=b.(min)()).all() && (b.(max)().cwise()<=m_max).all(); }
/** Extends \c *this such that it contains the point \a p and returns a reference to \c *this. */
inline AlignedBox& extend(const VectorType& p)
{ m_min = (m_min.cwise().min)(p); m_max = (m_max.cwise().max)(p); return *this; }
{ m_min = m_min.cwise().(min)(p); m_max = m_max.cwise().(max)(p); return *this; }
/** Extends \c *this such that it contains the box \a b and returns a reference to \c *this. */
inline AlignedBox& extend(const AlignedBox& b)
{ m_min = (m_min.cwise().min)(b.m_min); m_max = (m_max.cwise().max)(b.m_max); return *this; }
{ m_min = m_min.cwise().(min)(b.m_min); m_max = m_max.cwise().(max)(b.m_max); return *this; }
/** Clamps \c *this by the box \a b and returns a reference to \c *this. */
inline AlignedBox& clamp(const AlignedBox& b)
{ m_min = (m_min.cwise().max)(b.m_min); m_max = (m_max.cwise().min)(b.m_max); return *this; }
{ m_min = m_min.cwise().(max)(b.m_min); m_max = m_max.cwise().(min)(b.m_max); return *this; }
/** Translate \c *this by the vector \a t and returns a reference to \c *this. */
inline AlignedBox& translate(const VectorType& t)
@@ -138,8 +138,8 @@ EIGEN_MAKE_ALIGNED_OPERATOR_NEW_IF_VECTORIZABLE_FIXED_SIZE(_Scalar,_AmbientDim==
template<typename OtherScalarType>
inline explicit AlignedBox(const AlignedBox<OtherScalarType,AmbientDimAtCompileTime>& other)
{
m_min = (other.min)().template cast<Scalar>();
m_max = (other.max)().template cast<Scalar>();
m_min = other.(min)().template cast<Scalar>();
m_max = other.(max)().template cast<Scalar>();
}
/** \returns \c true if \c *this is approximately equal to \a other, within the precision

View File

@@ -112,4 +112,4 @@
#undef Hyperplane
#undef ParametrizedLine
#endif // EIGEN2_GEOMETRY_MODULE_H
#endif // EIGEN2_GEOMETRY_MODULE_H

View File

@@ -227,6 +227,46 @@ template<typename _MatrixType> class ComplexSchur
friend struct internal::complex_schur_reduce_to_hessenberg<MatrixType, NumTraits<Scalar>::IsComplex>;
};
namespace internal {
/** Computes the principal value of the square root of the complex \a z. */
template<typename RealScalar>
std::complex<RealScalar> sqrt(const std::complex<RealScalar> &z)
{
RealScalar t, tre, tim;
t = abs(z);
if (abs(real(z)) <= abs(imag(z)))
{
// No cancellation in these formulas
tre = sqrt(RealScalar(0.5)*(t + real(z)));
tim = sqrt(RealScalar(0.5)*(t - real(z)));
}
else
{
// Stable computation of the above formulas
if (z.real() > RealScalar(0))
{
tre = t + z.real();
tim = abs(imag(z))*sqrt(RealScalar(0.5)/tre);
tre = sqrt(RealScalar(0.5)*tre);
}
else
{
tim = t - z.real();
tre = abs(imag(z))*sqrt(RealScalar(0.5)/tim);
tim = sqrt(RealScalar(0.5)*tim);
}
}
if(z.imag() < RealScalar(0))
tim = -tim;
return (std::complex<RealScalar>(tre,tim));
}
} // end namespace internal
/** If m_matT(i+1,i) is neglegible in floating point arithmetic
* compared to m_matT(i,i) and m_matT(j,j), then set it to zero and
* return true, else return false. */
@@ -262,7 +302,7 @@ typename ComplexSchur<MatrixType>::ComplexScalar ComplexSchur<MatrixType>::compu
ComplexScalar b = t.coeff(0,1) * t.coeff(1,0);
ComplexScalar c = t.coeff(0,0) - t.coeff(1,1);
ComplexScalar disc = sqrt(c*c + RealScalar(4)*b);
ComplexScalar disc = internal::sqrt(c*c + RealScalar(4)*b);
ComplexScalar det = t.coeff(0,0) * t.coeff(1,1) - b;
ComplexScalar trace = t.coeff(0,0) + t.coeff(1,1);
ComplexScalar eival1 = (trace + disc) / RealScalar(2);

View File

@@ -32,10 +32,6 @@
template<typename _MatrixType>
class GeneralizedSelfAdjointEigenSolver;
namespace internal {
template<typename SolverType,int Size,bool IsComplex> struct direct_selfadjoint_eigenvalues;
}
/** \eigenvalues_module \ingroup Eigenvalues_Module
*
*
@@ -90,7 +86,7 @@ template<typename _MatrixType> class SelfAdjointEigenSolver
Options = MatrixType::Options,
MaxColsAtCompileTime = MatrixType::MaxColsAtCompileTime
};
/** \brief Scalar type for matrices of type \p _MatrixType. */
typedef typename MatrixType::Scalar Scalar;
typedef typename MatrixType::Index Index;
@@ -102,8 +98,6 @@ template<typename _MatrixType> class SelfAdjointEigenSolver
* complex.
*/
typedef typename NumTraits<Scalar>::Real RealScalar;
friend struct internal::direct_selfadjoint_eigenvalues<SelfAdjointEigenSolver,Size,NumTraits<Scalar>::IsComplex>;
/** \brief Type for vector of eigenvalues as returned by eigenvalues().
*
@@ -204,22 +198,6 @@ template<typename _MatrixType> class SelfAdjointEigenSolver
* \sa SelfAdjointEigenSolver(const MatrixType&, int)
*/
SelfAdjointEigenSolver& compute(const MatrixType& matrix, int options = ComputeEigenvectors);
/** \brief Computes eigendecomposition of given matrix using a direct algorithm
*
* This is a variant of compute(const MatrixType&, int options) which
* directly solves the underlying polynomial equation.
*
* Currently only 3x3 matrices for which the sizes are known at compile time are supported (e.g., Matrix3d).
*
* This method is usually significantly faster than the QR algorithm
* but it might also be less accurate. It is also worth noting that
* for 3x3 matrices it involves trigonometric operations which are
* not necessarily available for all scalar types.
*
* \sa compute(const MatrixType&, int options)
*/
SelfAdjointEigenSolver& computeDirect(const MatrixType& matrix, int options = ComputeEigenvectors);
/** \brief Returns the eigenvectors of given matrix.
*
@@ -242,7 +220,6 @@ template<typename _MatrixType> class SelfAdjointEigenSolver
const MatrixType& eigenvectors() const
{
eigen_assert(m_isInitialized && "SelfAdjointEigenSolver is not initialized.");
eigen_assert(info() == Success && "Eigenvalue computation did not converge.");
eigen_assert(m_eigenvectorsOk && "The eigenvectors have not been computed together with the eigenvalues.");
return m_eivec;
}
@@ -265,7 +242,6 @@ template<typename _MatrixType> class SelfAdjointEigenSolver
const RealVectorType& eigenvalues() const
{
eigen_assert(m_isInitialized && "SelfAdjointEigenSolver is not initialized.");
eigen_assert(info() == Success && "Eigenvalue computation did not converge.");
return m_eivalues;
}
@@ -290,7 +266,6 @@ template<typename _MatrixType> class SelfAdjointEigenSolver
MatrixType operatorSqrt() const
{
eigen_assert(m_isInitialized && "SelfAdjointEigenSolver is not initialized.");
eigen_assert(info() == Success && "Eigenvalue computation did not converge.");
eigen_assert(m_eigenvectorsOk && "The eigenvectors have not been computed together with the eigenvalues.");
return m_eivec * m_eivalues.cwiseSqrt().asDiagonal() * m_eivec.adjoint();
}
@@ -316,7 +291,6 @@ template<typename _MatrixType> class SelfAdjointEigenSolver
MatrixType operatorInverseSqrt() const
{
eigen_assert(m_isInitialized && "SelfAdjointEigenSolver is not initialized.");
eigen_assert(info() == Success && "Eigenvalue computation did not converge.");
eigen_assert(m_eigenvectorsOk && "The eigenvectors have not been computed together with the eigenvalues.");
return m_eivec * m_eivalues.cwiseInverse().cwiseSqrt().asDiagonal() * m_eivec.adjoint();
}
@@ -333,8 +307,7 @@ template<typename _MatrixType> class SelfAdjointEigenSolver
/** \brief Maximum number of iterations.
*
* The algorithm terminates if it does not converge within m_maxIterations * n iterations, where n
* denotes the size of the matrix. This value is currently set to 30 (copied from LAPACK).
* Maximum number of iterations allowed for an eigenvalue to converge.
*/
static const int m_maxIterations = 30;
@@ -434,7 +407,7 @@ SelfAdjointEigenSolver<MatrixType>& SelfAdjointEigenSolver<MatrixType>
Index end = n-1;
Index start = 0;
Index iter = 0; // total number of iterations
Index iter = 0; // number of iterations we are working on one element
while (end>0)
{
@@ -445,14 +418,15 @@ SelfAdjointEigenSolver<MatrixType>& SelfAdjointEigenSolver<MatrixType>
// find the largest unreduced block
while (end>0 && m_subdiag[end-1]==0)
{
iter = 0;
end--;
}
if (end<=0)
break;
// if we spent too many iterations, we give up
// if we spent too many iterations on the current element, we give up
iter++;
if(iter > m_maxIterations * n) break;
if(iter > m_maxIterations) break;
start = end - 1;
while (start>0 && m_subdiag[start-1]!=0)
@@ -461,7 +435,7 @@ SelfAdjointEigenSolver<MatrixType>& SelfAdjointEigenSolver<MatrixType>
internal::tridiagonal_qr_step<MatrixType::Flags&RowMajorBit ? RowMajor : ColMajor>(diag.data(), m_subdiag.data(), start, end, computeEigenvectors ? m_eivec.data() : (Scalar*)0, n);
}
if (iter <= m_maxIterations * n)
if (iter <= m_maxIterations)
m_info = Success;
else
m_info = NoConvergence;
@@ -492,264 +466,6 @@ SelfAdjointEigenSolver<MatrixType>& SelfAdjointEigenSolver<MatrixType>
return *this;
}
namespace internal {
template<typename SolverType,int Size,bool IsComplex> struct direct_selfadjoint_eigenvalues
{
inline static void run(SolverType& eig, const typename SolverType::MatrixType& A, int options)
{ eig.compute(A,options); }
};
template<typename SolverType> struct direct_selfadjoint_eigenvalues<SolverType,3,false>
{
typedef typename SolverType::MatrixType MatrixType;
typedef typename SolverType::RealVectorType VectorType;
typedef typename SolverType::Scalar Scalar;
inline static void computeRoots(const MatrixType& m, VectorType& roots)
{
using std::sqrt;
using std::atan2;
using std::cos;
using std::sin;
const Scalar s_inv3 = 1.0/3.0;
const Scalar s_sqrt3 = sqrt(Scalar(3.0));
// The characteristic equation is x^3 - c2*x^2 + c1*x - c0 = 0. The
// eigenvalues are the roots to this equation, all guaranteed to be
// real-valued, because the matrix is symmetric.
Scalar c0 = m(0,0)*m(1,1)*m(2,2) + Scalar(2)*m(1,0)*m(2,0)*m(2,1) - m(0,0)*m(2,1)*m(2,1) - m(1,1)*m(2,0)*m(2,0) - m(2,2)*m(1,0)*m(1,0);
Scalar c1 = m(0,0)*m(1,1) - m(1,0)*m(1,0) + m(0,0)*m(2,2) - m(2,0)*m(2,0) + m(1,1)*m(2,2) - m(2,1)*m(2,1);
Scalar c2 = m(0,0) + m(1,1) + m(2,2);
// Construct the parameters used in classifying the roots of the equation
// and in solving the equation for the roots in closed form.
Scalar c2_over_3 = c2*s_inv3;
Scalar a_over_3 = (c1 - c2*c2_over_3)*s_inv3;
if (a_over_3 > Scalar(0))
a_over_3 = Scalar(0);
Scalar half_b = Scalar(0.5)*(c0 + c2_over_3*(Scalar(2)*c2_over_3*c2_over_3 - c1));
Scalar q = half_b*half_b + a_over_3*a_over_3*a_over_3;
if (q > Scalar(0))
q = Scalar(0);
// Compute the eigenvalues by solving for the roots of the polynomial.
Scalar rho = sqrt(-a_over_3);
Scalar theta = atan2(sqrt(-q),half_b)*s_inv3;
Scalar cos_theta = cos(theta);
Scalar sin_theta = sin(theta);
roots(0) = c2_over_3 + Scalar(2)*rho*cos_theta;
roots(1) = c2_over_3 - rho*(cos_theta + s_sqrt3*sin_theta);
roots(2) = c2_over_3 - rho*(cos_theta - s_sqrt3*sin_theta);
// Sort in increasing order.
if (roots(0) >= roots(1))
std::swap(roots(0),roots(1));
if (roots(1) >= roots(2))
{
std::swap(roots(1),roots(2));
if (roots(0) >= roots(1))
std::swap(roots(0),roots(1));
}
}
inline static void run(SolverType& solver, const MatrixType& mat, int options)
{
using std::sqrt;
eigen_assert(mat.cols() == 3 && mat.cols() == mat.rows());
eigen_assert((options&~(EigVecMask|GenEigMask))==0
&& (options&EigVecMask)!=EigVecMask
&& "invalid option parameter");
bool computeEigenvectors = (options&ComputeEigenvectors)==ComputeEigenvectors;
MatrixType& eivecs = solver.m_eivec;
VectorType& eivals = solver.m_eivalues;
// map the matrix coefficients to [-1:1] to avoid over- and underflow.
Scalar scale = mat.cwiseAbs().maxCoeff();
MatrixType scaledMat = mat / scale;
// compute the eigenvalues
computeRoots(scaledMat,eivals);
// compute the eigen vectors
if(computeEigenvectors)
{
Scalar safeNorm2 = Eigen::NumTraits<Scalar>::epsilon();
safeNorm2 *= safeNorm2;
if((eivals(2)-eivals(0))<=Eigen::NumTraits<Scalar>::epsilon())
{
eivecs.setIdentity();
}
else
{
scaledMat = scaledMat.template selfadjointView<Lower>();
MatrixType tmp;
tmp = scaledMat;
Scalar d0 = eivals(2) - eivals(1);
Scalar d1 = eivals(1) - eivals(0);
int k = d0 > d1 ? 2 : 0;
d0 = d0 > d1 ? d1 : d0;
tmp.diagonal().array () -= eivals(k);
VectorType cross;
Scalar n;
n = (cross = tmp.row(0).cross(tmp.row(1))).squaredNorm();
if(n>safeNorm2)
eivecs.col(k) = cross / sqrt(n);
else
{
n = (cross = tmp.row(0).cross(tmp.row(2))).squaredNorm();
if(n>safeNorm2)
eivecs.col(k) = cross / sqrt(n);
else
{
n = (cross = tmp.row(1).cross(tmp.row(2))).squaredNorm();
if(n>safeNorm2)
eivecs.col(k) = cross / sqrt(n);
else
{
// the input matrix and/or the eigenvaues probably contains some inf/NaN,
// => exit
// scale back to the original size.
eivals *= scale;
solver.m_info = NumericalIssue;
solver.m_isInitialized = true;
solver.m_eigenvectorsOk = computeEigenvectors;
return;
}
}
}
tmp = scaledMat;
tmp.diagonal().array() -= eivals(1);
if(d0<=Eigen::NumTraits<Scalar>::epsilon())
eivecs.col(1) = eivecs.col(k).unitOrthogonal();
else
{
n = (cross = eivecs.col(k).cross(tmp.row(0).normalized())).squaredNorm();
if(n>safeNorm2)
eivecs.col(1) = cross / sqrt(n);
else
{
n = (cross = eivecs.col(k).cross(tmp.row(1))).squaredNorm();
if(n>safeNorm2)
eivecs.col(1) = cross / sqrt(n);
else
{
n = (cross = eivecs.col(k).cross(tmp.row(2))).squaredNorm();
if(n>safeNorm2)
eivecs.col(1) = cross / sqrt(n);
else
{
// we should never reach this point,
// if so the last two eigenvalues are likely to ve very closed to each other
eivecs.col(1) = eivecs.col(k).unitOrthogonal();
}
}
}
// make sure that eivecs[1] is orthogonal to eivecs[2]
Scalar d = eivecs.col(1).dot(eivecs.col(k));
eivecs.col(1) = (eivecs.col(1) - d * eivecs.col(k)).normalized();
}
eivecs.col(k==2 ? 0 : 2) = eivecs.col(k).cross(eivecs.col(1)).normalized();
}
}
// Rescale back to the original size.
eivals *= scale;
solver.m_info = Success;
solver.m_isInitialized = true;
solver.m_eigenvectorsOk = computeEigenvectors;
}
};
// 2x2 direct eigenvalues decomposition, code from Hauke Heibel
template<typename SolverType> struct direct_selfadjoint_eigenvalues<SolverType,2,false>
{
typedef typename SolverType::MatrixType MatrixType;
typedef typename SolverType::RealVectorType VectorType;
typedef typename SolverType::Scalar Scalar;
inline static void computeRoots(const MatrixType& m, VectorType& roots)
{
using std::sqrt;
const Scalar t0 = Scalar(0.5) * sqrt( abs2(m(0,0)-m(1,1)) + Scalar(4)*m(1,0)*m(1,0));
const Scalar t1 = Scalar(0.5) * (m(0,0) + m(1,1));
roots(0) = t1 - t0;
roots(1) = t1 + t0;
}
inline static void run(SolverType& solver, const MatrixType& mat, int options)
{
eigen_assert(mat.cols() == 2 && mat.cols() == mat.rows());
eigen_assert((options&~(EigVecMask|GenEigMask))==0
&& (options&EigVecMask)!=EigVecMask
&& "invalid option parameter");
bool computeEigenvectors = (options&ComputeEigenvectors)==ComputeEigenvectors;
MatrixType& eivecs = solver.m_eivec;
VectorType& eivals = solver.m_eivalues;
// map the matrix coefficients to [-1:1] to avoid over- and underflow.
Scalar scale = mat.cwiseAbs().maxCoeff();
scale = (std::max)(scale,Scalar(1));
MatrixType scaledMat = mat / scale;
// Compute the eigenvalues
computeRoots(scaledMat,eivals);
// compute the eigen vectors
if(computeEigenvectors)
{
scaledMat.diagonal().array () -= eivals(1);
Scalar a2 = abs2(scaledMat(0,0));
Scalar c2 = abs2(scaledMat(1,1));
Scalar b2 = abs2(scaledMat(1,0));
if(a2>c2)
{
eivecs.col(1) << -scaledMat(1,0), scaledMat(0,0);
eivecs.col(1) /= sqrt(a2+b2);
}
else
{
eivecs.col(1) << -scaledMat(1,1), scaledMat(1,0);
eivecs.col(1) /= sqrt(c2+b2);
}
eivecs.col(0) << eivecs.col(1).unitOrthogonal();
}
// Rescale back to the original size.
eivals *= scale;
solver.m_info = Success;
solver.m_isInitialized = true;
solver.m_eigenvectorsOk = computeEigenvectors;
}
};
}
template<typename MatrixType>
SelfAdjointEigenSolver<MatrixType>& SelfAdjointEigenSolver<MatrixType>
::computeDirect(const MatrixType& matrix, int options)
{
internal::direct_selfadjoint_eigenvalues<SelfAdjointEigenSolver,Size,NumTraits<Scalar>::IsComplex>::run(*this,matrix,options);
return *this;
}
namespace internal {
template<int StorageOrder,typename RealScalar, typename Scalar, typename Index>
static void tridiagonal_qr_step(RealScalar* diag, RealScalar* subdiag, Index start, Index end, Scalar* matrixQ, Index n)

View File

@@ -349,38 +349,4 @@ inline Scalar AlignedBox<Scalar,AmbientDim>::squaredExteriorDistance(const Align
return dist2;
}
/** \defgroup alignedboxtypedefs Global aligned box typedefs
*
* \ingroup Geometry_Module
*
* Eigen defines several typedef shortcuts for most common aligned box types.
*
* The general patterns are the following:
*
* \c AlignedBoxSizeType where \c Size can be \c 1, \c 2,\c 3,\c 4 for fixed size boxes or \c X for dynamic size,
* and where \c Type can be \c i for integer, \c f for float, \c d for double.
*
* For example, \c AlignedBox3d is a fixed-size 3x3 aligned box type of doubles, and \c AlignedBoxXf is a dynamic-size aligned box of floats.
*
* \sa class AlignedBox
*/
#define EIGEN_MAKE_TYPEDEFS(Type, TypeSuffix, Size, SizeSuffix) \
/** \ingroup alignedboxtypedefs */ \
typedef AlignedBox<Type, Size> AlignedBox##SizeSuffix##TypeSuffix;
#define EIGEN_MAKE_TYPEDEFS_ALL_SIZES(Type, TypeSuffix) \
EIGEN_MAKE_TYPEDEFS(Type, TypeSuffix, 1, 1) \
EIGEN_MAKE_TYPEDEFS(Type, TypeSuffix, 2, 2) \
EIGEN_MAKE_TYPEDEFS(Type, TypeSuffix, 3, 3) \
EIGEN_MAKE_TYPEDEFS(Type, TypeSuffix, 4, 4) \
EIGEN_MAKE_TYPEDEFS(Type, TypeSuffix, Dynamic, X)
EIGEN_MAKE_TYPEDEFS_ALL_SIZES(int, i)
EIGEN_MAKE_TYPEDEFS_ALL_SIZES(float, f)
EIGEN_MAKE_TYPEDEFS_ALL_SIZES(double, d)
#undef EIGEN_MAKE_TYPEDEFS_ALL_SIZES
#undef EIGEN_MAKE_TYPEDEFS
#endif // EIGEN_ALIGNEDBOX_H

View File

@@ -182,9 +182,10 @@ public:
template<typename NewScalarType>
inline typename internal::cast_return_type<Derived,Quaternion<NewScalarType> >::type cast() const
{
return typename internal::cast_return_type<Derived,Quaternion<NewScalarType> >::type(derived());
return typename internal::cast_return_type<Derived,Quaternion<NewScalarType> >::type(
coeffs().template cast<NewScalarType>());
}
#ifdef EIGEN_QUATERNIONBASE_PLUGIN
# include EIGEN_QUATERNIONBASE_PLUGIN
#endif
@@ -224,25 +225,22 @@ struct traits<Quaternion<_Scalar,_Options> >
typedef _Scalar Scalar;
typedef Matrix<_Scalar,4,1,_Options> Coefficients;
enum{
IsAligned = internal::traits<Coefficients>::Flags & AlignedBit,
IsAligned = bool(EIGEN_ALIGN) && ((int(_Options)&Aligned)==Aligned),
Flags = IsAligned ? (AlignedBit | LvalueBit) : LvalueBit
};
};
}
template<typename _Scalar, int _Options>
class Quaternion : public QuaternionBase<Quaternion<_Scalar,_Options> >
{
class Quaternion : public QuaternionBase<Quaternion<_Scalar,_Options> >{
typedef QuaternionBase<Quaternion<_Scalar,_Options> > Base;
enum { IsAligned = internal::traits<Quaternion>::IsAligned };
public:
typedef _Scalar Scalar;
EIGEN_INHERIT_ASSIGNMENT_EQUAL_OPERATOR(Quaternion)
using Base::operator*=;
typedef typename internal::traits<Quaternion>::Coefficients Coefficients;
typedef typename internal::traits<Quaternion<Scalar,_Options> >::Coefficients Coefficients;
typedef typename Base::AngleAxisType AngleAxisType;
/** Default constructor leaving the quaternion uninitialized. */
@@ -273,16 +271,9 @@ public:
template<typename Derived>
explicit inline Quaternion(const MatrixBase<Derived>& other) { *this = other; }
/** Explicit copy constructor with scalar conversion */
template<typename OtherScalar, int OtherOptions>
explicit inline Quaternion(const Quaternion<OtherScalar, OtherOptions>& other)
{ m_coeffs = other.coeffs().template cast<Scalar>(); }
inline Coefficients& coeffs() { return m_coeffs;}
inline const Coefficients& coeffs() const { return m_coeffs;}
EIGEN_MAKE_ALIGNED_OPERATOR_NEW_IF(IsAligned)
protected:
Coefficients m_coeffs;

View File

@@ -279,9 +279,6 @@ public:
template<typename OtherDerived>
inline explicit Transform(const EigenBase<OtherDerived>& other)
{
EIGEN_STATIC_ASSERT((internal::is_same<Scalar,typename OtherDerived::Scalar>::value),
YOU_MIXED_DIFFERENT_NUMERIC_TYPES__YOU_NEED_TO_USE_THE_CAST_METHOD_OF_MATRIXBASE_TO_CAST_NUMERIC_TYPES_EXPLICITLY);
check_template_params();
internal::transform_construct_from_matrix<OtherDerived,Mode,Options,Dim,HDim>::run(this, other.derived());
}
@@ -290,9 +287,6 @@ public:
template<typename OtherDerived>
inline Transform& operator=(const EigenBase<OtherDerived>& other)
{
EIGEN_STATIC_ASSERT((internal::is_same<Scalar,typename OtherDerived::Scalar>::value),
YOU_MIXED_DIFFERENT_NUMERIC_TYPES__YOU_NEED_TO_USE_THE_CAST_METHOD_OF_MATRIXBASE_TO_CAST_NUMERIC_TYPES_EXPLICITLY);
internal::transform_construct_from_matrix<OtherDerived,Mode,Options,Dim,HDim>::run(this, other.derived());
return *this;
}

View File

@@ -54,8 +54,6 @@ public:
typedef Matrix<Scalar,Dim,Dim> LinearMatrixType;
/** corresponding affine transformation type */
typedef Transform<Scalar,Dim,Affine> AffineTransformType;
/** corresponding isometric transformation type */
typedef Transform<Scalar,Dim,Isometry> IsometryTransformType;
protected:
@@ -116,8 +114,8 @@ public:
/** Concatenates a translation and a rotation */
template<typename Derived>
inline IsometryTransformType operator*(const RotationBase<Derived,Dim>& r) const
{ return *this * IsometryTransformType(r); }
inline AffineTransformType operator*(const RotationBase<Derived,Dim>& r) const
{ return *this * r.toRotationMatrix(); }
/** \returns the concatenation of a linear transformation \a l with the translation \a t */
// its a nightmare to define a templated friend function outside its declaration

View File

@@ -237,20 +237,13 @@ template<typename VectorsType, typename CoeffsType, int Side> class HouseholderS
ConjugateReturnType inverse() const { return adjoint(); }
/** \internal */
template<typename DestType> inline void evalTo(DestType& dst) const
template<typename DestType> void evalTo(DestType& dst) const
{
Matrix<Scalar, DestType::RowsAtCompileTime, 1,
AutoAlign|ColMajor, DestType::MaxRowsAtCompileTime, 1> workspace(rows());
evalTo(dst, workspace);
}
/** \internal */
template<typename Dest, typename Workspace>
void evalTo(Dest& dst, Workspace& workspace) const
{
workspace.resize(rows());
Index vecs = m_length;
if( internal::is_same<typename internal::remove_all<VectorsType>::type,Dest>::value
// FIXME find a way to pass this temporary if the user wants to
Matrix<Scalar, DestType::RowsAtCompileTime, 1,
AutoAlign|ColMajor, DestType::MaxRowsAtCompileTime, 1> temp(rows());
if( internal::is_same<typename internal::remove_all<VectorsType>::type,DestType>::value
&& internal::extract_data(dst) == internal::extract_data(m_vectors))
{
// in-place
@@ -261,10 +254,10 @@ template<typename VectorsType, typename CoeffsType, int Side> class HouseholderS
Index cornerSize = rows() - k - m_shift;
if(m_trans)
dst.bottomRightCorner(cornerSize, cornerSize)
.applyHouseholderOnTheRight(essentialVector(k), m_coeffs.coeff(k), workspace.data());
.applyHouseholderOnTheRight(essentialVector(k), m_coeffs.coeff(k), &temp.coeffRef(0));
else
dst.bottomRightCorner(cornerSize, cornerSize)
.applyHouseholderOnTheLeft(essentialVector(k), m_coeffs.coeff(k), workspace.data());
.applyHouseholderOnTheLeft(essentialVector(k), m_coeffs.coeff(k), &temp.coeffRef(0));
// clear the off diagonal vector
dst.col(k).tail(rows()-k-1).setZero();
@@ -281,10 +274,10 @@ template<typename VectorsType, typename CoeffsType, int Side> class HouseholderS
Index cornerSize = rows() - k - m_shift;
if(m_trans)
dst.bottomRightCorner(cornerSize, cornerSize)
.applyHouseholderOnTheRight(essentialVector(k), m_coeffs.coeff(k), &workspace.coeffRef(0));
.applyHouseholderOnTheRight(essentialVector(k), m_coeffs.coeff(k), &temp.coeffRef(0));
else
dst.bottomRightCorner(cornerSize, cornerSize)
.applyHouseholderOnTheLeft(essentialVector(k), m_coeffs.coeff(k), &workspace.coeffRef(0));
.applyHouseholderOnTheLeft(essentialVector(k), m_coeffs.coeff(k), &temp.coeffRef(0));
}
}
}
@@ -292,40 +285,24 @@ template<typename VectorsType, typename CoeffsType, int Side> class HouseholderS
/** \internal */
template<typename Dest> inline void applyThisOnTheRight(Dest& dst) const
{
Matrix<Scalar,1,Dest::RowsAtCompileTime,RowMajor,1,Dest::MaxRowsAtCompileTime> workspace(dst.rows());
applyThisOnTheRight(dst, workspace);
}
/** \internal */
template<typename Dest, typename Workspace>
inline void applyThisOnTheRight(Dest& dst, Workspace& workspace) const
{
workspace.resize(dst.rows());
Matrix<Scalar,1,Dest::RowsAtCompileTime> temp(dst.rows());
for(Index k = 0; k < m_length; ++k)
{
Index actual_k = m_trans ? m_length-k-1 : k;
dst.rightCols(rows()-m_shift-actual_k)
.applyHouseholderOnTheRight(essentialVector(actual_k), m_coeffs.coeff(actual_k), workspace.data());
.applyHouseholderOnTheRight(essentialVector(actual_k), m_coeffs.coeff(actual_k), &temp.coeffRef(0));
}
}
/** \internal */
template<typename Dest> inline void applyThisOnTheLeft(Dest& dst) const
{
Matrix<Scalar,1,Dest::ColsAtCompileTime,RowMajor,1,Dest::MaxColsAtCompileTime> workspace(dst.cols());
applyThisOnTheLeft(dst, workspace);
}
/** \internal */
template<typename Dest, typename Workspace>
inline void applyThisOnTheLeft(Dest& dst, Workspace& workspace) const
{
workspace.resize(dst.cols());
Matrix<Scalar,1,Dest::ColsAtCompileTime> temp(dst.cols());
for(Index k = 0; k < m_length; ++k)
{
Index actual_k = m_trans ? k : m_length-k-1;
dst.bottomRows(rows()-m_shift-actual_k)
.applyHouseholderOnTheLeft(essentialVector(actual_k), m_coeffs.coeff(actual_k), workspace.data());
.applyHouseholderOnTheLeft(essentialVector(actual_k), m_coeffs.coeff(actual_k), &temp.coeffRef(0));
}
}

View File

@@ -1,141 +0,0 @@
// This file is part of Eigen, a lightweight C++ template library
// for linear algebra.
//
// Copyright (C) 2011 Gael Guennebaud <gael.guennebaud@inria.fr>
//
// Eigen is free software; you can redistribute it and/or
// modify it under the terms of the GNU Lesser General Public
// License as published by the Free Software Foundation; either
// version 3 of the License, or (at your option) any later version.
//
// Alternatively, you can redistribute it and/or
// modify it under the terms of the GNU General Public License as
// published by the Free Software Foundation; either version 2 of
// the License, or (at your option) any later version.
//
// Eigen is distributed in the hope that it will be useful, but WITHOUT ANY
// WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS
// FOR A PARTICULAR PURPOSE. See the GNU Lesser General Public License or the
// GNU General Public License for more details.
//
// You should have received a copy of the GNU Lesser General Public
// License and a copy of the GNU General Public License along with
// Eigen. If not, see <http://www.gnu.org/licenses/>.
#ifndef EIGEN_BASIC_PRECONDITIONERS_H
#define EIGEN_BASIC_PRECONDITIONERS_H
/** \ingroup IterativeLinearSolvers_Module
* \brief A preconditioner based on the digonal entries
*
* This class allows to approximately solve for A.x = b problems assuming A is a diagonal matrix.
* In other words, this preconditioner neglects all off diagonal entries and, in Eigen's language, solves for:
* \code
* A.diagonal().asDiagonal() . x = b
* \endcode
*
* \tparam _Scalar the type of the scalar.
*
* This preconditioner is suitable for both selfadjoint and general problems.
* The diagonal entries are pre-inverted and stored into a dense vector.
*
* \note A variant that has yet to be implemented would attempt to preserve the norm of each column.
*
*/
template <typename _Scalar>
class DiagonalPreconditioner
{
typedef _Scalar Scalar;
typedef Matrix<Scalar,Dynamic,1> Vector;
typedef typename Vector::Index Index;
public:
typedef Matrix<Scalar,Dynamic,Dynamic> MatrixType;
DiagonalPreconditioner() : m_isInitialized(false) {}
template<typename MatrixType>
DiagonalPreconditioner(const MatrixType& mat) : m_invdiag(mat.cols())
{
compute(mat);
}
Index rows() const { return m_invdiag.size(); }
Index cols() const { return m_invdiag.size(); }
template<typename MatrixType>
DiagonalPreconditioner& compute(const MatrixType& mat)
{
m_invdiag.resize(mat.cols());
for(int j=0; j<mat.outerSize(); ++j)
{
typename MatrixType::InnerIterator it(mat,j);
while(it && it.index()!=j) ++it;
if(it && it.index()==j)
m_invdiag(j) = Scalar(1)/it.value();
else
m_invdiag(j) = 0;
}
m_isInitialized = true;
return *this;
}
template<typename Rhs, typename Dest>
void _solve(const Rhs& b, Dest& x) const
{
x = m_invdiag.array() * b.array() ;
}
template<typename Rhs> inline const internal::solve_retval<DiagonalPreconditioner, Rhs>
solve(const MatrixBase<Rhs>& b) const
{
eigen_assert(m_isInitialized && "DiagonalPreconditioner is not initialized.");
eigen_assert(m_invdiag.size()==b.rows()
&& "DiagonalPreconditioner::solve(): invalid number of rows of the right hand side matrix b");
return internal::solve_retval<DiagonalPreconditioner, Rhs>(*this, b.derived());
}
protected:
Vector m_invdiag;
bool m_isInitialized;
};
namespace internal {
template<typename _MatrixType, typename Rhs>
struct solve_retval<DiagonalPreconditioner<_MatrixType>, Rhs>
: solve_retval_base<DiagonalPreconditioner<_MatrixType>, Rhs>
{
typedef DiagonalPreconditioner<_MatrixType> Dec;
EIGEN_MAKE_SOLVE_HELPERS(Dec,Rhs)
template<typename Dest> void evalTo(Dest& dst) const
{
dec()._solve(rhs(),dst);
}
};
}
/** \ingroup IterativeLinearSolvers_Module
* \brief A naive preconditioner which approximates any matrix as the identity matrix
*
* \sa class DiagonalPreconditioner
*/
class IdentityPreconditioner
{
public:
IdentityPreconditioner() {}
template<typename MatrixType>
IdentityPreconditioner(const MatrixType& ) {}
template<typename MatrixType>
IdentityPreconditioner& compute(const MatrixType& ) { return *this; }
template<typename Rhs>
inline const Rhs& solve(const Rhs& b) const { return b; }
};
#endif // EIGEN_BASIC_PRECONDITIONERS_H

View File

@@ -1,262 +0,0 @@
// This file is part of Eigen, a lightweight C++ template library
// for linear algebra.
//
// Copyright (C) 2011 Gael Guennebaud <gael.guennebaud@inria.fr>
//
// Eigen is free software; you can redistribute it and/or
// modify it under the terms of the GNU Lesser General Public
// License as published by the Free Software Foundation; either
// version 3 of the License, or (at your option) any later version.
//
// Alternatively, you can redistribute it and/or
// modify it under the terms of the GNU General Public License as
// published by the Free Software Foundation; either version 2 of
// the License, or (at your option) any later version.
//
// Eigen is distributed in the hope that it will be useful, but WITHOUT ANY
// WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS
// FOR A PARTICULAR PURPOSE. See the GNU Lesser General Public License or the
// GNU General Public License for more details.
//
// You should have received a copy of the GNU Lesser General Public
// License and a copy of the GNU General Public License along with
// Eigen. If not, see <http://www.gnu.org/licenses/>.
#ifndef EIGEN_BICGSTAB_H
#define EIGEN_BICGSTAB_H
namespace internal {
/** \internal Low-level bi conjugate gradient stabilized algorithm
* \param mat The matrix A
* \param rhs The right hand side vector b
* \param x On input and initial solution, on output the computed solution.
* \param precond A preconditioner being able to efficiently solve for an
* approximation of Ax=b (regardless of b)
* \param iters On input the max number of iteration, on output the number of performed iterations.
* \param tol_error On input the tolerance error, on output an estimation of the relative error.
*/
template<typename MatrixType, typename Rhs, typename Dest, typename Preconditioner>
void bicgstab(const MatrixType& mat, const Rhs& rhs, Dest& x,
const Preconditioner& precond, int& iters,
typename Dest::RealScalar& tol_error)
{
using std::sqrt;
using std::abs;
typedef typename Dest::RealScalar RealScalar;
typedef typename Dest::Scalar Scalar;
typedef Matrix<Scalar,Dynamic,1> VectorType;
RealScalar tol = tol_error;
int maxIters = iters;
int n = mat.cols();
VectorType r = rhs - mat * x;
VectorType r0 = r;
RealScalar r0_sqnorm = r0.squaredNorm();
Scalar rho = 1;
Scalar alpha = 1;
Scalar w = 1;
VectorType v = VectorType::Zero(n), p = VectorType::Zero(n);
VectorType y(n), z(n);
VectorType kt(n), ks(n);
VectorType s(n), t(n);
RealScalar tol2 = tol*tol;
int i = 0;
do
{
Scalar rho_old = rho;
rho = r0.dot(r);
Scalar beta = (rho/rho_old) * (alpha / w);
p = r + beta * (p - w * v);
y = precond.solve(p);
v.noalias() = mat * y;
alpha = rho / r0.dot(v);
s = r - alpha * v;
z = precond.solve(s);
t.noalias() = mat * z;
kt = precond.solve(t);
ks = precond.solve(s);
w = kt.dot(ks) / kt.squaredNorm();
x += alpha * y + w * z;
r = s - w * t;
++i;
} while ( r.squaredNorm()/r0_sqnorm > tol2 && i<maxIters );
tol_error = sqrt(r.squaredNorm()/r0_sqnorm);
//tol_error = sqrt(abs(absNew / absInit));
iters = i;
}
}
template< typename _MatrixType,
typename _Preconditioner = DiagonalPreconditioner<typename _MatrixType::Scalar> >
class BiCGSTAB;
namespace internal {
template< typename _MatrixType, typename _Preconditioner>
struct traits<BiCGSTAB<_MatrixType,_Preconditioner> >
{
typedef _MatrixType MatrixType;
typedef _Preconditioner Preconditioner;
};
}
/** \ingroup IterativeLinearSolvers_Module
* \brief A bi conjugate gradient stabilized solver for sparse square problems
*
* This class allows to solve for A.x = b sparse linear problems using a bi conjugate gradient
* stabilized algorithm. The vectors x and b can be either dense or sparse.
*
* \tparam _MatrixType the type of the sparse matrix A, can be a dense or a sparse matrix.
* \tparam _Preconditioner the type of the preconditioner. Default is DiagonalPreconditioner
*
* The maximal number of iterations and tolerance value can be controlled via the setMaxIterations()
* and setTolerance() methods. The default are 1000 max iterations and NumTraits<Scalar>::epsilon()
* for the tolerance.
*
* This class can be used as the direct solver classes. Here is a typical usage example:
* \code
* int n = 10000;
* VectorXd x(n), b(n);
* SparseMatrix<double> A(n,n);
* // fill A and b
* BiCGSTAB<SparseMatrix<double> > solver;
* solver(A);
* x = solver.solve(b);
* std::cout << "#iterations: " << solver.iterations() << std::endl;
* std::cout << "estimated error: " << solver.error() << std::endl;
* // update b, and solve again
* x = solver.solve(b);
* \endcode
*
* By default the iterations start with x=0 as an initial guess of the solution.
* One can control the start using the solveWithGuess() method. Here is a step by
* step execution example starting with a random guess and printing the evolution
* of the estimated error:
* * \code
* x = VectorXd::Random(n);
* solver.setMaxIterations(1);
* int i = 0;
* do {
* x = solver.solveWithGuess(b,x);
* std::cout << i << " : " << solver.error() << std::endl;
* ++i;
* } while (solver.info()!=Success && i<100);
* \endcode
* Note that such a step by step excution is slightly slower.
*
* \sa class SimplicialCholesky, DiagonalPreconditioner, IdentityPreconditioner
*/
template< typename _MatrixType, typename _Preconditioner>
class BiCGSTAB : public IterativeSolverBase<BiCGSTAB<_MatrixType,_Preconditioner> >
{
typedef IterativeSolverBase<BiCGSTAB> Base;
using Base::mp_matrix;
using Base::m_error;
using Base::m_iterations;
using Base::m_info;
using Base::m_isInitialized;
public:
typedef _MatrixType MatrixType;
typedef typename MatrixType::Scalar Scalar;
typedef typename MatrixType::Index Index;
typedef typename MatrixType::RealScalar RealScalar;
typedef _Preconditioner Preconditioner;
public:
/** Default constructor. */
BiCGSTAB() : Base() {}
/** Initialize the solver with matrix \a A for further \c Ax=b solving.
*
* This constructor is a shortcut for the default constructor followed
* by a call to compute().
*
* \warning this class stores a reference to the matrix A as well as some
* precomputed values that depend on it. Therefore, if \a A is changed
* this class becomes invalid. Call compute() to update it with the new
* matrix A, or modify a copy of A.
*/
BiCGSTAB(const MatrixType& A) : Base(A) {}
~BiCGSTAB() {}
/** \returns the solution x of \f$ A x = b \f$ using the current decomposition of A
* \a x0 as an initial solution.
*
* \sa compute()
*/
template<typename Rhs,typename Guess>
inline const internal::solve_retval_with_guess<BiCGSTAB, Rhs, Guess>
solveWithGuess(const MatrixBase<Rhs>& b, const Guess& x0) const
{
eigen_assert(m_isInitialized && "BiCGSTAB is not initialized.");
eigen_assert(Base::rows()==b.rows()
&& "BiCGSTAB::solve(): invalid number of rows of the right hand side matrix b");
return internal::solve_retval_with_guess
<BiCGSTAB, Rhs, Guess>(*this, b.derived(), x0);
}
/** \internal */
template<typename Rhs,typename Dest>
void _solveWithGuess(const Rhs& b, Dest& x) const
{
for(int j=0; j<b.cols(); ++j)
{
m_iterations = Base::m_maxIterations;
m_error = Base::m_tolerance;
typename Dest::ColXpr xj(x,j);
internal::bicgstab(*mp_matrix, b.col(j), xj, Base::m_preconditioner, m_iterations, m_error);
}
m_isInitialized = true;
m_info = m_error <= Base::m_tolerance ? Success : NoConvergence;
}
/** \internal */
template<typename Rhs,typename Dest>
void _solve(const Rhs& b, Dest& x) const
{
x.setOnes();
_solveWithGuess(b,x);
}
protected:
};
namespace internal {
template<typename _MatrixType, typename _Preconditioner, typename Rhs>
struct solve_retval<BiCGSTAB<_MatrixType, _Preconditioner>, Rhs>
: solve_retval_base<BiCGSTAB<_MatrixType, _Preconditioner>, Rhs>
{
typedef BiCGSTAB<_MatrixType, _Preconditioner> Dec;
EIGEN_MAKE_SOLVE_HELPERS(Dec,Rhs)
template<typename Dest> void evalTo(Dest& dst) const
{
dec()._solve(rhs(),dst);
}
};
}
#endif // EIGEN_BICGSTAB_H

View File

@@ -1,6 +0,0 @@
FILE(GLOB Eigen_IterativeLinearSolvers_SRCS "*.h")
INSTALL(FILES
${Eigen_IterativeLinearSolvers_SRCS}
DESTINATION ${INCLUDE_INSTALL_DIR}/Eigen/src/IterativeLinearSolvers COMPONENT Devel
)

View File

@@ -1,256 +0,0 @@
// This file is part of Eigen, a lightweight C++ template library
// for linear algebra.
//
// Copyright (C) 2011 Gael Guennebaud <gael.guennebaud@inria.fr>
//
// Eigen is free software; you can redistribute it and/or
// modify it under the terms of the GNU Lesser General Public
// License as published by the Free Software Foundation; either
// version 3 of the License, or (at your option) any later version.
//
// Alternatively, you can redistribute it and/or
// modify it under the terms of the GNU General Public License as
// published by the Free Software Foundation; either version 2 of
// the License, or (at your option) any later version.
//
// Eigen is distributed in the hope that it will be useful, but WITHOUT ANY
// WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS
// FOR A PARTICULAR PURPOSE. See the GNU Lesser General Public License or the
// GNU General Public License for more details.
//
// You should have received a copy of the GNU Lesser General Public
// License and a copy of the GNU General Public License along with
// Eigen. If not, see <http://www.gnu.org/licenses/>.
#ifndef EIGEN_CONJUGATE_GRADIENT_H
#define EIGEN_CONJUGATE_GRADIENT_H
namespace internal {
/** \internal Low-level conjugate gradient algorithm
* \param mat The matrix A
* \param rhs The right hand side vector b
* \param x On input and initial solution, on output the computed solution.
* \param precond A preconditioner being able to efficiently solve for an
* approximation of Ax=b (regardless of b)
* \param iters On input the max number of iteration, on output the number of performed iterations.
* \param tol_error On input the tolerance error, on output an estimation of the relative error.
*/
template<typename MatrixType, typename Rhs, typename Dest, typename Preconditioner>
EIGEN_DONT_INLINE
void conjugate_gradient(const MatrixType& mat, const Rhs& rhs, Dest& x,
const Preconditioner& precond, int& iters,
typename Dest::RealScalar& tol_error)
{
using std::sqrt;
using std::abs;
typedef typename Dest::RealScalar RealScalar;
typedef typename Dest::Scalar Scalar;
typedef Matrix<Scalar,Dynamic,1> VectorType;
RealScalar tol = tol_error;
int maxIters = iters;
int n = mat.cols();
VectorType residual = rhs - mat * x; //initial residual
VectorType p(n);
p = precond.solve(residual); //initial search direction
VectorType z(n), tmp(n);
RealScalar absNew = internal::real(residual.dot(p)); // the square of the absolute value of r scaled by invM
RealScalar absInit = absNew; // the initial absolute value
int i = 0;
while ((i < maxIters) && (absNew > tol*tol*absInit))
{
tmp.noalias() = mat * p; // the bottleneck of the algorithm
Scalar alpha = absNew / p.dot(tmp); // the amount we travel on dir
x += alpha * p; // update solution
residual -= alpha * tmp; // update residue
z = precond.solve(residual); // approximately solve for "A z = residual"
RealScalar absOld = absNew;
absNew = internal::real(residual.dot(z)); // update the absolute value of r
RealScalar beta = absNew / absOld; // calculate the Gram-Schmidit value used to create the new search direction
p = z + beta * p; // update search direction
i++;
}
tol_error = sqrt(abs(absNew / absInit));
iters = i;
}
}
template< typename _MatrixType, int _UpLo=Lower,
typename _Preconditioner = DiagonalPreconditioner<typename _MatrixType::Scalar> >
class ConjugateGradient;
namespace internal {
template< typename _MatrixType, int _UpLo, typename _Preconditioner>
struct traits<ConjugateGradient<_MatrixType,_UpLo,_Preconditioner> >
{
typedef _MatrixType MatrixType;
typedef _Preconditioner Preconditioner;
};
}
/** \ingroup IterativeLinearSolvers_Module
* \brief A conjugate gradient solver for sparse self-adjoint problems
*
* This class allows to solve for A.x = b sparse linear problems using a conjugate gradient algorithm.
* The sparse matrix A must be selfadjoint. The vectors x and b can be either dense or sparse.
*
* \tparam _MatrixType the type of the sparse matrix A, can be a dense or a sparse matrix.
* \tparam _UpLo the triangular part that will be used for the computations. It can be Lower
* or Upper. Default is Lower.
* \tparam _Preconditioner the type of the preconditioner. Default is DiagonalPreconditioner
*
* The maximal number of iterations and tolerance value can be controlled via the setMaxIterations()
* and setTolerance() methods. The default are 1000 max iterations and NumTraits<Scalar>::epsilon()
* for the tolerance.
*
* This class can be used as the direct solver classes. Here is a typical usage example:
* \code
* int n = 10000;
* VectorXd x(n), b(n);
* SparseMatrix<double> A(n,n);
* // fill A and b
* ConjugateGradient<SparseMatrix<double> > cg;
* cg.compute(A);
* x = cg.solve(b);
* std::cout << "#iterations: " << cg.iterations() << std::endl;
* std::cout << "estimated error: " << cg.error() << std::endl;
* // update b, and solve again
* x = cg.solve(b);
* \endcode
*
* By default the iterations start with x=0 as an initial guess of the solution.
* One can control the start using the solveWithGuess() method. Here is a step by
* step execution example starting with a random guess and printing the evolution
* of the estimated error:
* * \code
* x = VectorXd::Random(n);
* cg.setMaxIterations(1);
* int i = 0;
* do {
* x = cg.solveWithGuess(b,x);
* std::cout << i << " : " << cg.error() << std::endl;
* ++i;
* } while (cg.info()!=Success && i<100);
* \endcode
* Note that such a step by step excution is slightly slower.
*
* \sa class SimplicialCholesky, DiagonalPreconditioner, IdentityPreconditioner
*/
template< typename _MatrixType, int _UpLo, typename _Preconditioner>
class ConjugateGradient : public IterativeSolverBase<ConjugateGradient<_MatrixType,_UpLo,_Preconditioner> >
{
typedef IterativeSolverBase<ConjugateGradient> Base;
using Base::mp_matrix;
using Base::m_error;
using Base::m_iterations;
using Base::m_info;
using Base::m_isInitialized;
public:
typedef _MatrixType MatrixType;
typedef typename MatrixType::Scalar Scalar;
typedef typename MatrixType::Index Index;
typedef typename MatrixType::RealScalar RealScalar;
typedef _Preconditioner Preconditioner;
enum {
UpLo = _UpLo
};
public:
/** Default constructor. */
ConjugateGradient() : Base() {}
/** Initialize the solver with matrix \a A for further \c Ax=b solving.
*
* This constructor is a shortcut for the default constructor followed
* by a call to compute().
*
* \warning this class stores a reference to the matrix A as well as some
* precomputed values that depend on it. Therefore, if \a A is changed
* this class becomes invalid. Call compute() to update it with the new
* matrix A, or modify a copy of A.
*/
ConjugateGradient(const MatrixType& A) : Base(A) {}
~ConjugateGradient() {}
/** \returns the solution x of \f$ A x = b \f$ using the current decomposition of A
* \a x0 as an initial solution.
*
* \sa compute()
*/
template<typename Rhs,typename Guess>
inline const internal::solve_retval_with_guess<ConjugateGradient, Rhs, Guess>
solveWithGuess(const MatrixBase<Rhs>& b, const Guess& x0) const
{
eigen_assert(m_isInitialized && "ConjugateGradient is not initialized.");
eigen_assert(Base::rows()==b.rows()
&& "ConjugateGradient::solve(): invalid number of rows of the right hand side matrix b");
return internal::solve_retval_with_guess
<ConjugateGradient, Rhs, Guess>(*this, b.derived(), x0);
}
/** \internal */
template<typename Rhs,typename Dest>
void _solveWithGuess(const Rhs& b, Dest& x) const
{
m_iterations = Base::m_maxIterations;
m_error = Base::m_tolerance;
for(int j=0; j<b.cols(); ++j)
{
m_iterations = Base::m_maxIterations;
m_error = Base::m_tolerance;
typename Dest::ColXpr xj(x,j);
internal::conjugate_gradient(mp_matrix->template selfadjointView<UpLo>(), b.col(j), xj,
Base::m_preconditioner, m_iterations, m_error);
}
m_isInitialized = true;
m_info = m_error <= Base::m_tolerance ? Success : NoConvergence;
}
/** \internal */
template<typename Rhs,typename Dest>
void _solve(const Rhs& b, Dest& x) const
{
x.setOnes();
_solveWithGuess(b,x);
}
protected:
};
namespace internal {
template<typename _MatrixType, int _UpLo, typename _Preconditioner, typename Rhs>
struct solve_retval<ConjugateGradient<_MatrixType,_UpLo,_Preconditioner>, Rhs>
: solve_retval_base<ConjugateGradient<_MatrixType,_UpLo,_Preconditioner>, Rhs>
{
typedef ConjugateGradient<_MatrixType,_UpLo,_Preconditioner> Dec;
EIGEN_MAKE_SOLVE_HELPERS(Dec,Rhs)
template<typename Dest> void evalTo(Dest& dst) const
{
dec()._solve(rhs(),dst);
}
};
}
#endif // EIGEN_CONJUGATE_GRADIENT_H

View File

@@ -1,226 +0,0 @@
// This file is part of Eigen, a lightweight C++ template library
// for linear algebra.
//
// Copyright (C) 2011 Gael Guennebaud <gael.guennebaud@inria.fr>
//
// Eigen is free software; you can redistribute it and/or
// modify it under the terms of the GNU Lesser General Public
// License as published by the Free Software Foundation; either
// version 3 of the License, or (at your option) any later version.
//
// Alternatively, you can redistribute it and/or
// modify it under the terms of the GNU General Public License as
// published by the Free Software Foundation; either version 2 of
// the License, or (at your option) any later version.
//
// Eigen is distributed in the hope that it will be useful, but WITHOUT ANY
// WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS
// FOR A PARTICULAR PURPOSE. See the GNU Lesser General Public License or the
// GNU General Public License for more details.
//
// You should have received a copy of the GNU Lesser General Public
// License and a copy of the GNU General Public License along with
// Eigen. If not, see <http://www.gnu.org/licenses/>.
#ifndef EIGEN_ITERATIVE_SOLVER_BASE_H
#define EIGEN_ITERATIVE_SOLVER_BASE_H
/** \ingroup IterativeLinearSolvers_Module
* \brief Base class for linear iterative solvers
*
* \sa class SimplicialCholesky, DiagonalPreconditioner, IdentityPreconditioner
*/
template< typename Derived>
class IterativeSolverBase
{
public:
typedef typename internal::traits<Derived>::MatrixType MatrixType;
typedef typename internal::traits<Derived>::Preconditioner Preconditioner;
typedef typename MatrixType::Scalar Scalar;
typedef typename MatrixType::Index Index;
typedef typename MatrixType::RealScalar RealScalar;
public:
Derived& derived() { return *static_cast<Derived*>(this); }
const Derived& derived() const { return *static_cast<const Derived*>(this); }
/** Default constructor. */
IterativeSolverBase()
: mp_matrix(0)
{
init();
}
/** Initialize the solver with matrix \a A for further \c Ax=b solving.
*
* This constructor is a shortcut for the default constructor followed
* by a call to compute().
*
* \warning this class stores a reference to the matrix A as well as some
* precomputed values that depend on it. Therefore, if \a A is changed
* this class becomes invalid. Call compute() to update it with the new
* matrix A, or modify a copy of A.
*/
IterativeSolverBase(const MatrixType& A)
{
init();
compute(A);
}
~IterativeSolverBase() {}
/** Initializes the iterative solver with the matrix \a A for further solving \c Ax=b problems.
*
* Currently, this function mostly initialized/compute the preconditioner. In the future
* we might, for instance, implement column reodering for faster matrix vector products.
*
* \warning this class stores a reference to the matrix A as well as some
* precomputed values that depend on it. Therefore, if \a A is changed
* this class becomes invalid. Call compute() to update it with the new
* matrix A, or modify a copy of A.
*/
Derived& compute(const MatrixType& A)
{
mp_matrix = &A;
m_preconditioner.compute(A);
m_isInitialized = true;
m_info = Success;
return derived();
}
/** \internal */
Index rows() const { return mp_matrix->rows(); }
/** \internal */
Index cols() const { return mp_matrix->cols(); }
/** \returns the tolerance threshold used by the stopping criteria */
RealScalar tolerance() const { return m_tolerance; }
/** Sets the tolerance threshold used by the stopping criteria */
Derived& setTolerance(RealScalar tolerance)
{
m_tolerance = tolerance;
return derived();
}
/** \returns a read-write reference to the preconditioner for custom configuration. */
Preconditioner& preconditioner() { return m_preconditioner; }
/** \returns a read-only reference to the preconditioner. */
const Preconditioner& preconditioner() const { return m_preconditioner; }
/** \returns the max number of iterations */
int maxIterations() const { return m_maxIterations; }
/** Sets the max number of iterations */
Derived& setMaxIterations(int maxIters)
{
m_maxIterations = maxIters;
return derived();
}
/** \returns the number of iterations performed during the last solve */
int iterations() const
{
eigen_assert(m_isInitialized && "ConjugateGradient is not initialized.");
return m_iterations;
}
/** \returns the tolerance error reached during the last solve */
RealScalar error() const
{
eigen_assert(m_isInitialized && "ConjugateGradient is not initialized.");
return m_error;
}
/** \returns the solution x of \f$ A x = b \f$ using the current decomposition of A.
*
* \sa compute()
*/
template<typename Rhs> inline const internal::solve_retval<Derived, Rhs>
solve(const MatrixBase<Rhs>& b) const
{
eigen_assert(m_isInitialized && "IterativeSolverBase is not initialized.");
eigen_assert(rows()==b.rows()
&& "IterativeSolverBase::solve(): invalid number of rows of the right hand side matrix b");
return internal::solve_retval<Derived, Rhs>(derived(), b.derived());
}
/** \returns the solution x of \f$ A x = b \f$ using the current decomposition of A.
*
* \sa compute()
*/
template<typename Rhs>
inline const internal::sparse_solve_retval<IterativeSolverBase, Rhs>
solve(const SparseMatrixBase<Rhs>& b) const
{
eigen_assert(m_isInitialized && "IterativeSolverBase is not initialized.");
eigen_assert(rows()==b.rows()
&& "IterativeSolverBase::solve(): invalid number of rows of the right hand side matrix b");
return internal::sparse_solve_retval<IterativeSolverBase, Rhs>(*this, b.derived());
}
/** \returns Success if the iterations converged, and NoConvergence otherwise. */
ComputationInfo info() const
{
eigen_assert(m_isInitialized && "IterativeSolverBase is not initialized.");
return m_info;
}
/** \internal */
template<typename Rhs, typename DestScalar, int DestOptions, typename DestIndex>
void _solve_sparse(const Rhs& b, SparseMatrix<DestScalar,DestOptions,DestIndex> &dest) const
{
eigen_assert(rows()==b.rows());
int rhsCols = b.cols();
int size = b.rows();
Eigen::Matrix<DestScalar,Dynamic,1> tb(size);
Eigen::Matrix<DestScalar,Dynamic,1> tx(size);
for(int k=0; k<rhsCols; ++k)
{
tb = b.col(k);
tx = derived().solve(tb);
dest.col(k) = tx.sparseView(0);
}
}
protected:
void init()
{
m_isInitialized = false;
m_maxIterations = 1000;
m_tolerance = NumTraits<Scalar>::epsilon();
}
const MatrixType* mp_matrix;
Preconditioner m_preconditioner;
int m_maxIterations;
RealScalar m_tolerance;
mutable RealScalar m_error;
mutable int m_iterations;
mutable ComputationInfo m_info;
mutable bool m_isInitialized;
};
namespace internal {
template<typename Derived, typename Rhs>
struct sparse_solve_retval<IterativeSolverBase<Derived>, Rhs>
: sparse_solve_retval_base<IterativeSolverBase<Derived>, Rhs>
{
typedef IterativeSolverBase<Derived> Dec;
EIGEN_MAKE_SPARSE_SOLVE_HELPERS(Dec,Rhs)
template<typename Dest> void evalTo(Dest& dst) const
{
dec().derived()._solve_sparse(rhs(),dst);
}
};
}
#endif // EIGEN_ITERATIVE_SOLVER_BASE_H

View File

@@ -443,6 +443,7 @@ FullPivLU<MatrixType>& FullPivLU<MatrixType>::compute(const MatrixType& matrix)
m_nonzero_pivots = size; // the generic case is that in which all pivots are nonzero (invertible case)
m_maxpivot = RealScalar(0);
RealScalar cutoff(0);
for(Index k = 0; k < size; ++k)
{
@@ -457,7 +458,14 @@ FullPivLU<MatrixType>& FullPivLU<MatrixType>::compute(const MatrixType& matrix)
row_of_biggest_in_corner += k; // correct the values! since they were computed in the corner,
col_of_biggest_in_corner += k; // need to add k to them.
if(biggest_in_corner==RealScalar(0))
// when k==0, biggest_in_corner is the biggest coeff absolute value in the original matrix
if(k == 0) cutoff = biggest_in_corner * NumTraits<Scalar>::epsilon();
// if the pivot (hence the corner) is "zero", terminate to avoid generating nan/inf values.
// Notice that using an exact comparison (biggest_in_corner==0) here, as Golub-van Loan do in
// their pseudo-code, results in numerical instability! The cutoff here has been validated
// by running the unit test 'lu' with many repetitions.
if(biggest_in_corner < cutoff)
{
// before exiting, make sure to initialize the still uninitialized transpositions
// in a sane state without destroying what we already have.

View File

@@ -1,6 +0,0 @@
FILE(GLOB Eigen_OrderingMethods_SRCS "*.h")
INSTALL(FILES
${Eigen_OrderingMethods_SRCS}
DESTINATION ${INCLUDE_INSTALL_DIR}/Eigen/src/OrderingMethods COMPONENT Devel
)

View File

@@ -26,18 +26,6 @@
#ifndef EIGEN_FULLPIVOTINGHOUSEHOLDERQR_H
#define EIGEN_FULLPIVOTINGHOUSEHOLDERQR_H
namespace internal {
template<typename MatrixType> struct FullPivHouseholderQRMatrixQReturnType;
template<typename MatrixType>
struct traits<FullPivHouseholderQRMatrixQReturnType<MatrixType> >
{
typedef typename MatrixType::PlainObject ReturnType;
};
}
/** \ingroup QR_Module
*
* \class FullPivHouseholderQR
@@ -74,7 +62,7 @@ template<typename _MatrixType> class FullPivHouseholderQR
typedef typename MatrixType::Scalar Scalar;
typedef typename MatrixType::RealScalar RealScalar;
typedef typename MatrixType::Index Index;
typedef internal::FullPivHouseholderQRMatrixQReturnType<MatrixType> MatrixQReturnType;
typedef Matrix<Scalar, RowsAtCompileTime, RowsAtCompileTime, Options, MaxRowsAtCompileTime, MaxRowsAtCompileTime> MatrixQType;
typedef typename internal::plain_diag_type<MatrixType>::type HCoeffsType;
typedef Matrix<Index, 1, ColsAtCompileTime, RowMajor, 1, MaxColsAtCompileTime> IntRowVectorType;
typedef PermutationMatrix<ColsAtCompileTime, MaxColsAtCompileTime> PermutationType;
@@ -151,9 +139,7 @@ template<typename _MatrixType> class FullPivHouseholderQR
return internal::solve_retval<FullPivHouseholderQR, Rhs>(*this, b.derived());
}
/** \returns Expression object representing the matrix Q
*/
MatrixQReturnType matrixQ(void) const;
MatrixQType matrixQ(void) const;
/** \returns a reference to the matrix where the Householder QR decomposition is stored
*/
@@ -522,73 +508,28 @@ struct solve_retval<FullPivHouseholderQR<_MatrixType>, Rhs>
}
};
/** \ingroup QR_Module
*
* \brief Expression type for return value of FullPivHouseholderQR::matrixQ()
*
* \tparam MatrixType type of underlying dense matrix
*/
template<typename MatrixType> struct FullPivHouseholderQRMatrixQReturnType
: public ReturnByValue<FullPivHouseholderQRMatrixQReturnType<MatrixType> >
{
public:
typedef typename MatrixType::Index Index;
typedef typename internal::plain_col_type<MatrixType, Index>::type IntColVectorType;
typedef typename internal::plain_diag_type<MatrixType>::type HCoeffsType;
typedef Matrix<typename MatrixType::Scalar, 1, MatrixType::RowsAtCompileTime, RowMajor, 1,
MatrixType::MaxRowsAtCompileTime> WorkVectorType;
FullPivHouseholderQRMatrixQReturnType(const MatrixType& qr,
const HCoeffsType& hCoeffs,
const IntColVectorType& rowsTranspositions)
: m_qr(qr),
m_hCoeffs(hCoeffs),
m_rowsTranspositions(rowsTranspositions)
{}
template <typename ResultType>
void evalTo(ResultType& result) const
{
const Index rows = m_qr.rows();
WorkVectorType workspace(rows);
evalTo(result, workspace);
}
template <typename ResultType>
void evalTo(ResultType& result, WorkVectorType& workspace) const
{
// compute the product H'_0 H'_1 ... H'_n-1,
// where H_k is the k-th Householder transformation I - h_k v_k v_k'
// and v_k is the k-th Householder vector [1,m_qr(k+1,k), m_qr(k+2,k), ...]
const Index rows = m_qr.rows();
const Index cols = m_qr.cols();
const Index size = (std::min)(rows, cols);
workspace.resize(rows);
result.setIdentity(rows, rows);
for (Index k = size-1; k >= 0; k--)
{
result.block(k, k, rows-k, rows-k)
.applyHouseholderOnTheLeft(m_qr.col(k).tail(rows-k-1), internal::conj(m_hCoeffs.coeff(k)), &workspace.coeffRef(k));
result.row(k).swap(result.row(m_rowsTranspositions.coeff(k)));
}
}
Index rows() const { return m_qr.rows(); }
Index cols() const { return m_qr.rows(); }
protected:
const typename MatrixType::Nested m_qr;
const typename HCoeffsType::Nested m_hCoeffs;
const typename IntColVectorType::Nested m_rowsTranspositions;
};
} // end namespace internal
/** \returns the matrix Q */
template<typename MatrixType>
inline typename FullPivHouseholderQR<MatrixType>::MatrixQReturnType FullPivHouseholderQR<MatrixType>::matrixQ() const
typename FullPivHouseholderQR<MatrixType>::MatrixQType FullPivHouseholderQR<MatrixType>::matrixQ() const
{
eigen_assert(m_isInitialized && "FullPivHouseholderQR is not initialized.");
return MatrixQReturnType(m_qr, m_hCoeffs, m_rows_transpositions);
// compute the product H'_0 H'_1 ... H'_n-1,
// where H_k is the k-th Householder transformation I - h_k v_k v_k'
// and v_k is the k-th Householder vector [1,m_qr(k+1,k), m_qr(k+2,k), ...]
Index rows = m_qr.rows();
Index cols = m_qr.cols();
Index size = (std::min)(rows,cols);
MatrixQType res = MatrixQType::Identity(rows, rows);
Matrix<Scalar,1,MatrixType::RowsAtCompileTime> temp(rows);
for (Index k = size-1; k >= 0; k--)
{
res.block(k, k, rows-k, rows-k)
.applyHouseholderOnTheLeft(m_qr.col(k).tail(rows-k-1), internal::conj(m_hCoeffs.coeff(k)), &temp.coeffRef(k));
res.row(k).swap(res.row(m_rows_transpositions.coeff(k)));
}
return res;
}
/** \return the full-pivoting Householder QR decomposition of \c *this.

View File

@@ -61,12 +61,9 @@ template<typename MatrixType, int QRPreconditioner, int Case,
> struct qr_preconditioner_impl {};
template<typename MatrixType, int QRPreconditioner, int Case>
class qr_preconditioner_impl<MatrixType, QRPreconditioner, Case, false>
struct qr_preconditioner_impl<MatrixType, QRPreconditioner, Case, false>
{
public:
typedef typename MatrixType::Index Index;
void allocate(const JacobiSVD<MatrixType, QRPreconditioner>&) {}
bool run(JacobiSVD<MatrixType, QRPreconditioner>&, const MatrixType&)
static bool run(JacobiSVD<MatrixType, QRPreconditioner>&, const MatrixType&)
{
return false;
}
@@ -75,279 +72,134 @@ public:
/*** preconditioner using FullPivHouseholderQR ***/
template<typename MatrixType>
class qr_preconditioner_impl<MatrixType, FullPivHouseholderQRPreconditioner, PreconditionIfMoreRowsThanCols, true>
struct qr_preconditioner_impl<MatrixType, FullPivHouseholderQRPreconditioner, PreconditionIfMoreRowsThanCols, true>
{
public:
typedef typename MatrixType::Index Index;
typedef typename MatrixType::Scalar Scalar;
enum
{
RowsAtCompileTime = MatrixType::RowsAtCompileTime,
MaxRowsAtCompileTime = MatrixType::MaxRowsAtCompileTime
};
typedef Matrix<Scalar, 1, RowsAtCompileTime, RowMajor, 1, MaxRowsAtCompileTime> WorkspaceType;
void allocate(const JacobiSVD<MatrixType, FullPivHouseholderQRPreconditioner>& svd)
{
if (svd.rows() != m_qr.rows() || svd.cols() != m_qr.cols())
{
m_qr = FullPivHouseholderQR<MatrixType>(svd.rows(), svd.cols());
}
if (svd.m_computeFullU) m_workspace.resize(svd.rows());
}
bool run(JacobiSVD<MatrixType, FullPivHouseholderQRPreconditioner>& svd, const MatrixType& matrix)
static bool run(JacobiSVD<MatrixType, FullPivHouseholderQRPreconditioner>& svd, const MatrixType& matrix)
{
if(matrix.rows() > matrix.cols())
{
m_qr.compute(matrix);
svd.m_workMatrix = m_qr.matrixQR().block(0,0,matrix.cols(),matrix.cols()).template triangularView<Upper>();
if(svd.m_computeFullU) m_qr.matrixQ().evalTo(svd.m_matrixU, m_workspace);
if(svd.computeV()) svd.m_matrixV = m_qr.colsPermutation();
FullPivHouseholderQR<MatrixType> qr(matrix);
svd.m_workMatrix = qr.matrixQR().block(0,0,matrix.cols(),matrix.cols()).template triangularView<Upper>();
if(svd.m_computeFullU) svd.m_matrixU = qr.matrixQ();
if(svd.computeV()) svd.m_matrixV = qr.colsPermutation();
return true;
}
return false;
}
private:
FullPivHouseholderQR<MatrixType> m_qr;
WorkspaceType m_workspace;
};
template<typename MatrixType>
class qr_preconditioner_impl<MatrixType, FullPivHouseholderQRPreconditioner, PreconditionIfMoreColsThanRows, true>
struct qr_preconditioner_impl<MatrixType, FullPivHouseholderQRPreconditioner, PreconditionIfMoreColsThanRows, true>
{
public:
typedef typename MatrixType::Index Index;
typedef typename MatrixType::Scalar Scalar;
enum
{
RowsAtCompileTime = MatrixType::RowsAtCompileTime,
ColsAtCompileTime = MatrixType::ColsAtCompileTime,
MaxRowsAtCompileTime = MatrixType::MaxRowsAtCompileTime,
MaxColsAtCompileTime = MatrixType::MaxColsAtCompileTime,
Options = MatrixType::Options
};
typedef Matrix<Scalar, ColsAtCompileTime, RowsAtCompileTime, Options, MaxColsAtCompileTime, MaxRowsAtCompileTime>
TransposeTypeWithSameStorageOrder;
void allocate(const JacobiSVD<MatrixType, FullPivHouseholderQRPreconditioner>& svd)
{
if (svd.cols() != m_qr.rows() || svd.rows() != m_qr.cols())
{
m_qr = FullPivHouseholderQR<TransposeTypeWithSameStorageOrder>(svd.cols(), svd.rows());
}
m_adjoint.resize(svd.cols(), svd.rows());
if (svd.m_computeFullV) m_workspace.resize(svd.cols());
}
bool run(JacobiSVD<MatrixType, FullPivHouseholderQRPreconditioner>& svd, const MatrixType& matrix)
static bool run(JacobiSVD<MatrixType, FullPivHouseholderQRPreconditioner>& svd, const MatrixType& matrix)
{
if(matrix.cols() > matrix.rows())
{
m_adjoint = matrix.adjoint();
m_qr.compute(m_adjoint);
svd.m_workMatrix = m_qr.matrixQR().block(0,0,matrix.rows(),matrix.rows()).template triangularView<Upper>().adjoint();
if(svd.m_computeFullV) m_qr.matrixQ().evalTo(svd.m_matrixV, m_workspace);
if(svd.computeU()) svd.m_matrixU = m_qr.colsPermutation();
typedef Matrix<typename MatrixType::Scalar, MatrixType::ColsAtCompileTime, MatrixType::RowsAtCompileTime,
MatrixType::Options, MatrixType::MaxColsAtCompileTime, MatrixType::MaxRowsAtCompileTime>
TransposeTypeWithSameStorageOrder;
FullPivHouseholderQR<TransposeTypeWithSameStorageOrder> qr(matrix.adjoint());
svd.m_workMatrix = qr.matrixQR().block(0,0,matrix.rows(),matrix.rows()).template triangularView<Upper>().adjoint();
if(svd.m_computeFullV) svd.m_matrixV = qr.matrixQ();
if(svd.computeU()) svd.m_matrixU = qr.colsPermutation();
return true;
}
else return false;
}
private:
FullPivHouseholderQR<TransposeTypeWithSameStorageOrder> m_qr;
TransposeTypeWithSameStorageOrder m_adjoint;
typename internal::plain_row_type<MatrixType>::type m_workspace;
};
/*** preconditioner using ColPivHouseholderQR ***/
template<typename MatrixType>
class qr_preconditioner_impl<MatrixType, ColPivHouseholderQRPreconditioner, PreconditionIfMoreRowsThanCols, true>
struct qr_preconditioner_impl<MatrixType, ColPivHouseholderQRPreconditioner, PreconditionIfMoreRowsThanCols, true>
{
public:
typedef typename MatrixType::Index Index;
void allocate(const JacobiSVD<MatrixType, ColPivHouseholderQRPreconditioner>& svd)
{
if (svd.rows() != m_qr.rows() || svd.cols() != m_qr.cols())
{
m_qr = ColPivHouseholderQR<MatrixType>(svd.rows(), svd.cols());
}
if (svd.m_computeFullU) m_workspace.resize(svd.rows());
else if (svd.m_computeThinU) m_workspace.resize(svd.cols());
}
bool run(JacobiSVD<MatrixType, ColPivHouseholderQRPreconditioner>& svd, const MatrixType& matrix)
static bool run(JacobiSVD<MatrixType, ColPivHouseholderQRPreconditioner>& svd, const MatrixType& matrix)
{
if(matrix.rows() > matrix.cols())
{
m_qr.compute(matrix);
svd.m_workMatrix = m_qr.matrixQR().block(0,0,matrix.cols(),matrix.cols()).template triangularView<Upper>();
if(svd.m_computeFullU) m_qr.householderQ().evalTo(svd.m_matrixU, m_workspace);
else if(svd.m_computeThinU)
{
ColPivHouseholderQR<MatrixType> qr(matrix);
svd.m_workMatrix = qr.matrixQR().block(0,0,matrix.cols(),matrix.cols()).template triangularView<Upper>();
if(svd.m_computeFullU) svd.m_matrixU = qr.householderQ();
else if(svd.m_computeThinU) {
svd.m_matrixU.setIdentity(matrix.rows(), matrix.cols());
m_qr.householderQ().applyThisOnTheLeft(svd.m_matrixU, m_workspace);
qr.householderQ().applyThisOnTheLeft(svd.m_matrixU);
}
if(svd.computeV()) svd.m_matrixV = m_qr.colsPermutation();
if(svd.computeV()) svd.m_matrixV = qr.colsPermutation();
return true;
}
return false;
}
private:
ColPivHouseholderQR<MatrixType> m_qr;
typename internal::plain_col_type<MatrixType>::type m_workspace;
};
template<typename MatrixType>
class qr_preconditioner_impl<MatrixType, ColPivHouseholderQRPreconditioner, PreconditionIfMoreColsThanRows, true>
struct qr_preconditioner_impl<MatrixType, ColPivHouseholderQRPreconditioner, PreconditionIfMoreColsThanRows, true>
{
public:
typedef typename MatrixType::Index Index;
typedef typename MatrixType::Scalar Scalar;
enum
{
RowsAtCompileTime = MatrixType::RowsAtCompileTime,
ColsAtCompileTime = MatrixType::ColsAtCompileTime,
MaxRowsAtCompileTime = MatrixType::MaxRowsAtCompileTime,
MaxColsAtCompileTime = MatrixType::MaxColsAtCompileTime,
Options = MatrixType::Options
};
typedef Matrix<Scalar, ColsAtCompileTime, RowsAtCompileTime, Options, MaxColsAtCompileTime, MaxRowsAtCompileTime>
TransposeTypeWithSameStorageOrder;
void allocate(const JacobiSVD<MatrixType, ColPivHouseholderQRPreconditioner>& svd)
{
if (svd.cols() != m_qr.rows() || svd.rows() != m_qr.cols())
{
m_qr = ColPivHouseholderQR<TransposeTypeWithSameStorageOrder>(svd.cols(), svd.rows());
}
if (svd.m_computeFullV) m_workspace.resize(svd.cols());
else if (svd.m_computeThinV) m_workspace.resize(svd.rows());
m_adjoint.resize(svd.cols(), svd.rows());
}
bool run(JacobiSVD<MatrixType, ColPivHouseholderQRPreconditioner>& svd, const MatrixType& matrix)
static bool run(JacobiSVD<MatrixType, ColPivHouseholderQRPreconditioner>& svd, const MatrixType& matrix)
{
if(matrix.cols() > matrix.rows())
{
m_adjoint = matrix.adjoint();
m_qr.compute(m_adjoint);
svd.m_workMatrix = m_qr.matrixQR().block(0,0,matrix.rows(),matrix.rows()).template triangularView<Upper>().adjoint();
if(svd.m_computeFullV) m_qr.householderQ().evalTo(svd.m_matrixV, m_workspace);
else if(svd.m_computeThinV)
{
typedef Matrix<typename MatrixType::Scalar, MatrixType::ColsAtCompileTime, MatrixType::RowsAtCompileTime,
MatrixType::Options, MatrixType::MaxColsAtCompileTime, MatrixType::MaxRowsAtCompileTime>
TransposeTypeWithSameStorageOrder;
ColPivHouseholderQR<TransposeTypeWithSameStorageOrder> qr(matrix.adjoint());
svd.m_workMatrix = qr.matrixQR().block(0,0,matrix.rows(),matrix.rows()).template triangularView<Upper>().adjoint();
if(svd.m_computeFullV) svd.m_matrixV = qr.householderQ();
else if(svd.m_computeThinV) {
svd.m_matrixV.setIdentity(matrix.cols(), matrix.rows());
m_qr.householderQ().applyThisOnTheLeft(svd.m_matrixV, m_workspace);
qr.householderQ().applyThisOnTheLeft(svd.m_matrixV);
}
if(svd.computeU()) svd.m_matrixU = m_qr.colsPermutation();
if(svd.computeU()) svd.m_matrixU = qr.colsPermutation();
return true;
}
else return false;
}
private:
ColPivHouseholderQR<TransposeTypeWithSameStorageOrder> m_qr;
TransposeTypeWithSameStorageOrder m_adjoint;
typename internal::plain_row_type<MatrixType>::type m_workspace;
};
/*** preconditioner using HouseholderQR ***/
template<typename MatrixType>
class qr_preconditioner_impl<MatrixType, HouseholderQRPreconditioner, PreconditionIfMoreRowsThanCols, true>
struct qr_preconditioner_impl<MatrixType, HouseholderQRPreconditioner, PreconditionIfMoreRowsThanCols, true>
{
public:
typedef typename MatrixType::Index Index;
void allocate(const JacobiSVD<MatrixType, HouseholderQRPreconditioner>& svd)
{
if (svd.rows() != m_qr.rows() || svd.cols() != m_qr.cols())
{
m_qr = HouseholderQR<MatrixType>(svd.rows(), svd.cols());
}
if (svd.m_computeFullU) m_workspace.resize(svd.rows());
else if (svd.m_computeThinU) m_workspace.resize(svd.cols());
}
bool run(JacobiSVD<MatrixType, HouseholderQRPreconditioner>& svd, const MatrixType& matrix)
static bool run(JacobiSVD<MatrixType, HouseholderQRPreconditioner>& svd, const MatrixType& matrix)
{
if(matrix.rows() > matrix.cols())
{
m_qr.compute(matrix);
svd.m_workMatrix = m_qr.matrixQR().block(0,0,matrix.cols(),matrix.cols()).template triangularView<Upper>();
if(svd.m_computeFullU) m_qr.householderQ().evalTo(svd.m_matrixU, m_workspace);
else if(svd.m_computeThinU)
{
HouseholderQR<MatrixType> qr(matrix);
svd.m_workMatrix = qr.matrixQR().block(0,0,matrix.cols(),matrix.cols()).template triangularView<Upper>();
if(svd.m_computeFullU) svd.m_matrixU = qr.householderQ();
else if(svd.m_computeThinU) {
svd.m_matrixU.setIdentity(matrix.rows(), matrix.cols());
m_qr.householderQ().applyThisOnTheLeft(svd.m_matrixU, m_workspace);
qr.householderQ().applyThisOnTheLeft(svd.m_matrixU);
}
if(svd.computeV()) svd.m_matrixV.setIdentity(matrix.cols(), matrix.cols());
return true;
}
return false;
}
private:
HouseholderQR<MatrixType> m_qr;
typename internal::plain_col_type<MatrixType>::type m_workspace;
};
template<typename MatrixType>
class qr_preconditioner_impl<MatrixType, HouseholderQRPreconditioner, PreconditionIfMoreColsThanRows, true>
struct qr_preconditioner_impl<MatrixType, HouseholderQRPreconditioner, PreconditionIfMoreColsThanRows, true>
{
public:
typedef typename MatrixType::Index Index;
typedef typename MatrixType::Scalar Scalar;
enum
{
RowsAtCompileTime = MatrixType::RowsAtCompileTime,
ColsAtCompileTime = MatrixType::ColsAtCompileTime,
MaxRowsAtCompileTime = MatrixType::MaxRowsAtCompileTime,
MaxColsAtCompileTime = MatrixType::MaxColsAtCompileTime,
Options = MatrixType::Options
};
typedef Matrix<Scalar, ColsAtCompileTime, RowsAtCompileTime, Options, MaxColsAtCompileTime, MaxRowsAtCompileTime>
TransposeTypeWithSameStorageOrder;
void allocate(const JacobiSVD<MatrixType, HouseholderQRPreconditioner>& svd)
{
if (svd.cols() != m_qr.rows() || svd.rows() != m_qr.cols())
{
m_qr = HouseholderQR<TransposeTypeWithSameStorageOrder>(svd.cols(), svd.rows());
}
if (svd.m_computeFullV) m_workspace.resize(svd.cols());
else if (svd.m_computeThinV) m_workspace.resize(svd.rows());
m_adjoint.resize(svd.cols(), svd.rows());
}
bool run(JacobiSVD<MatrixType, HouseholderQRPreconditioner>& svd, const MatrixType& matrix)
static bool run(JacobiSVD<MatrixType, HouseholderQRPreconditioner>& svd, const MatrixType& matrix)
{
if(matrix.cols() > matrix.rows())
{
m_adjoint = matrix.adjoint();
m_qr.compute(m_adjoint);
svd.m_workMatrix = m_qr.matrixQR().block(0,0,matrix.rows(),matrix.rows()).template triangularView<Upper>().adjoint();
if(svd.m_computeFullV) m_qr.householderQ().evalTo(svd.m_matrixV, m_workspace);
else if(svd.m_computeThinV)
{
typedef Matrix<typename MatrixType::Scalar, MatrixType::ColsAtCompileTime, MatrixType::RowsAtCompileTime,
MatrixType::Options, MatrixType::MaxColsAtCompileTime, MatrixType::MaxRowsAtCompileTime>
TransposeTypeWithSameStorageOrder;
HouseholderQR<TransposeTypeWithSameStorageOrder> qr(matrix.adjoint());
svd.m_workMatrix = qr.matrixQR().block(0,0,matrix.rows(),matrix.rows()).template triangularView<Upper>().adjoint();
if(svd.m_computeFullV) svd.m_matrixV = qr.householderQ();
else if(svd.m_computeThinV) {
svd.m_matrixV.setIdentity(matrix.cols(), matrix.rows());
m_qr.householderQ().applyThisOnTheLeft(svd.m_matrixV, m_workspace);
qr.householderQ().applyThisOnTheLeft(svd.m_matrixV);
}
if(svd.computeU()) svd.m_matrixU.setIdentity(matrix.rows(), matrix.rows());
return true;
}
else return false;
}
private:
HouseholderQR<TransposeTypeWithSameStorageOrder> m_qr;
TransposeTypeWithSameStorageOrder m_adjoint;
typename internal::plain_row_type<MatrixType>::type m_workspace;
};
/*** 2x2 SVD implementation
@@ -464,7 +316,7 @@ void real_2x2_jacobi_svd(const MatrixType& matrix, Index p, Index q,
* Here's an example demonstrating basic usage:
* \include JacobiSVD_basic.cpp
* Output: \verbinclude JacobiSVD_basic.out
*
*
* This JacobiSVD class is a two-sided Jacobi R-SVD decomposition, ensuring optimal reliability and accuracy. The downside is that it's slower than
* bidiagonalizing SVD algorithms for large square matrices; however its complexity is still \f$ O(n^2p) \f$ where \a n is the smaller dimension and
* \a p is the greater dimension, meaning that it is still of the same order of complexity as the faster bidiagonalizing R-SVD algorithms.
@@ -472,7 +324,7 @@ void real_2x2_jacobi_svd(const MatrixType& matrix, Index p, Index q,
*
* If the input matrix has inf or nan coefficients, the result of the computation is undefined, but the computation is guaranteed to
* terminate in finite (and reasonable) time.
*
*
* The possible values for QRPreconditioner are:
* \li ColPivHouseholderQRPreconditioner is the default. In practice it's very safe. It uses column-pivoting QR.
* \li FullPivHouseholderQRPreconditioner, is the safest and slowest. It uses full-pivoting QR.
@@ -642,7 +494,7 @@ template<typename _MatrixType, int QRPreconditioner> class JacobiSVD
* \param b the right-hand-side of the equation to solve.
*
* \note Solving requires both U and V to be computed. Thin U and V are enough, there is no need for full U or V.
*
*
* \note SVD solving is implicitly least-squares. Thus, this method serves both purposes of exact solving and least-squares solving.
* In other words, the returned solution is guaranteed to minimize the Euclidean norm \f$ \Vert A x - b \Vert \f$.
*/
@@ -683,9 +535,6 @@ template<typename _MatrixType, int QRPreconditioner> class JacobiSVD
friend struct internal::svd_precondition_2x2_block_to_be_real;
template<typename __MatrixType, int _QRPreconditioner, int _Case, bool _DoAnything>
friend struct internal::qr_preconditioner_impl;
internal::qr_preconditioner_impl<MatrixType, QRPreconditioner, internal::PreconditionIfMoreColsThanRows> m_qr_precond_morecols;
internal::qr_preconditioner_impl<MatrixType, QRPreconditioner, internal::PreconditionIfMoreRowsThanCols> m_qr_precond_morerows;
};
template<typename MatrixType, int QRPreconditioner>
@@ -729,9 +578,6 @@ void JacobiSVD<MatrixType, QRPreconditioner>::allocate(Index rows, Index cols, u
: m_computeThinV ? m_diagSize
: 0);
m_workMatrix.resize(m_diagSize, m_diagSize);
m_qr_precond_morecols.allocate(*this);
m_qr_precond_morerows.allocate(*this);
}
template<typename MatrixType, int QRPreconditioner>
@@ -744,12 +590,10 @@ JacobiSVD<MatrixType, QRPreconditioner>::compute(const MatrixType& matrix, unsig
// only worsening the precision of U and V as we accumulate more rotations
const RealScalar precision = RealScalar(2) * NumTraits<Scalar>::epsilon();
// limit for very small denormal numbers to be considered zero in order to avoid infinite loops (see bug 286)
const RealScalar considerAsZero = RealScalar(2) * std::numeric_limits<RealScalar>::denorm_min();
/*** step 1. The R-SVD step: we use a QR decomposition to reduce to the case of a square matrix */
if(!m_qr_precond_morecols.run(*this, matrix) && !m_qr_precond_morerows.run(*this, matrix))
if(!internal::qr_preconditioner_impl<MatrixType, QRPreconditioner, internal::PreconditionIfMoreColsThanRows>::run(*this, matrix)
&& !internal::qr_preconditioner_impl<MatrixType, QRPreconditioner, internal::PreconditionIfMoreRowsThanCols>::run(*this, matrix))
{
m_workMatrix = matrix.block(0,0,m_diagSize,m_diagSize);
if(m_computeFullU) m_matrixU.setIdentity(m_rows,m_rows);
@@ -773,11 +617,10 @@ JacobiSVD<MatrixType, QRPreconditioner>::compute(const MatrixType& matrix, unsig
{
// if this 2x2 sub-matrix is not diagonal already...
// notice that this comparison will evaluate to false if any NaN is involved, ensuring that NaN's don't
// keep us iterating forever. Similarly, small denormal numbers are considered zero.
// keep us iterating forever.
using std::max;
RealScalar threshold = (max)(considerAsZero, precision * (max)(internal::abs(m_workMatrix.coeff(p,p)),
internal::abs(m_workMatrix.coeff(q,q))));
if((max)(internal::abs(m_workMatrix.coeff(p,q)),internal::abs(m_workMatrix.coeff(q,p))) > threshold)
if((max)(internal::abs(m_workMatrix.coeff(p,q)),internal::abs(m_workMatrix.coeff(q,p)))
> (max)(internal::abs(m_workMatrix.coeff(p,p)),internal::abs(m_workMatrix.coeff(q,q)))*precision)
{
finished = false;

View File

@@ -25,8 +25,6 @@
#ifndef EIGEN_AMBIVECTOR_H
#define EIGEN_AMBIVECTOR_H
namespace internal {
/** \internal
* Hybrid sparse/dense vector class designed for intensive read-write operations.
*
@@ -301,7 +299,7 @@ class AmbiVector<_Scalar,_Index>::Iterator
* In practice, all coefficients having a magnitude smaller than \a epsilon
* are skipped.
*/
Iterator(const AmbiVector& vec, RealScalar epsilon = 0)
Iterator(const AmbiVector& vec, RealScalar epsilon = RealScalar(0.1)*NumTraits<RealScalar>::dummy_precision())
: m_vector(vec)
{
m_epsilon = epsilon;
@@ -317,7 +315,7 @@ class AmbiVector<_Scalar,_Index>::Iterator
{
ListEl* EIGEN_RESTRICT llElements = reinterpret_cast<ListEl*>(m_vector.m_buffer);
m_currentEl = m_vector.m_llStart;
while (m_currentEl>=0 && internal::abs(llElements[m_currentEl].value)<=m_epsilon)
while (m_currentEl>=0 && internal::abs(llElements[m_currentEl].value)<m_epsilon)
m_currentEl = llElements[m_currentEl].next;
if (m_currentEl<0)
{
@@ -377,6 +375,5 @@ class AmbiVector<_Scalar,_Index>::Iterator
bool m_isDense; // mode of the vector
};
} // namespace internal
#endif // EIGEN_AMBIVECTOR_H

View File

@@ -0,0 +1,6 @@
FILE(GLOB Eigen_Sparse_SRCS "*.h")
INSTALL(FILES
${Eigen_Sparse_SRCS}
DESTINATION ${INCLUDE_INSTALL_DIR}/Eigen/src/Sparse COMPONENT Devel
)

View File

@@ -25,10 +25,7 @@
#ifndef EIGEN_COMPRESSED_STORAGE_H
#define EIGEN_COMPRESSED_STORAGE_H
namespace internal {
/** \internal
* Stores a sparse set of values as a list of values and a list of indices.
/** Stores a sparse set of values as a list of values and a list of indices.
*
*/
template<typename _Scalar,typename _Index>
@@ -221,8 +218,8 @@ class CompressedStorage
Index* newIndices = new Index[size];
size_t copySize = (std::min)(size, m_size);
// copy
internal::smart_copy(m_values, m_values+copySize, newValues);
internal::smart_copy(m_indices, m_indices+copySize, newIndices);
memcpy(newValues, m_values, copySize * sizeof(Scalar));
memcpy(newIndices, m_indices, copySize * sizeof(Index));
// delete old stuff
delete[] m_values;
delete[] m_indices;
@@ -239,6 +236,4 @@ class CompressedStorage
};
} // namespace internal
#endif // EIGEN_COMPRESSED_STORAGE_H

View File

@@ -28,8 +28,7 @@
/* This file contains the respective InnerIterator definition of the expressions defined in Eigen/Core
*/
/** \ingroup SparseCore_Module
* \class InnerIterator
/** \class InnerIterator
* \brief An InnerIterator allows to loop over the element of a sparse (or dense) matrix or expression
*
* todo

View File

@@ -25,9 +25,7 @@
#ifndef EIGEN_DYNAMIC_SPARSEMATRIX_H
#define EIGEN_DYNAMIC_SPARSEMATRIX_H
/** \deprecated use a SparseMatrix in an uncompressed mode
*
* \class DynamicSparseMatrix
/** \class DynamicSparseMatrix
*
* \brief A sparse matrix class designed for matrix assembly purpose
*
@@ -66,7 +64,7 @@ struct traits<DynamicSparseMatrix<_Scalar, _Options, _Index> >
}
template<typename _Scalar, int _Options, typename _Index>
class DynamicSparseMatrix
class DynamicSparseMatrix
: public SparseMatrixBase<DynamicSparseMatrix<_Scalar, _Options, _Index> >
{
public:
@@ -86,7 +84,7 @@ template<typename _Scalar, int _Options, typename _Index>
typedef DynamicSparseMatrix<Scalar,(Flags&~RowMajorBit)|(IsRowMajor?RowMajorBit:0)> TransposedSparseMatrix;
Index m_innerSize;
std::vector<internal::CompressedStorage<Scalar,Index> > m_data;
std::vector<CompressedStorage<Scalar,Index> > m_data;
public:
@@ -96,8 +94,8 @@ template<typename _Scalar, int _Options, typename _Index>
inline Index outerSize() const { return static_cast<Index>(m_data.size()); }
inline Index innerNonZeros(Index j) const { return m_data[j].size(); }
std::vector<internal::CompressedStorage<Scalar,Index> >& _data() { return m_data; }
const std::vector<internal::CompressedStorage<Scalar,Index> >& _data() const { return m_data; }
std::vector<CompressedStorage<Scalar,Index> >& _data() { return m_data; }
const std::vector<CompressedStorage<Scalar,Index> >& _data() const { return m_data; }
/** \returns the coefficient value at given position \a row, \a col
* This operation involes a log(rho*outer_size) binary search.
@@ -121,7 +119,6 @@ template<typename _Scalar, int _Options, typename _Index>
}
class InnerIterator;
class ReverseInnerIterator;
void setZero()
{
@@ -235,23 +232,20 @@ template<typename _Scalar, int _Options, typename _Index>
}
}
/** The class DynamicSparseMatrix is deprectaed */
EIGEN_DEPRECATED inline DynamicSparseMatrix()
inline DynamicSparseMatrix()
: m_innerSize(0), m_data(0)
{
eigen_assert(innerSize()==0 && outerSize()==0);
}
/** The class DynamicSparseMatrix is deprectaed */
EIGEN_DEPRECATED inline DynamicSparseMatrix(Index rows, Index cols)
inline DynamicSparseMatrix(Index rows, Index cols)
: m_innerSize(0)
{
resize(rows, cols);
}
/** The class DynamicSparseMatrix is deprectaed */
template<typename OtherDerived>
EIGEN_DEPRECATED explicit inline DynamicSparseMatrix(const SparseMatrixBase<OtherDerived>& other)
explicit inline DynamicSparseMatrix(const SparseMatrixBase<OtherDerived>& other)
: m_innerSize(0)
{
Base::operator=(other.derived());
@@ -331,12 +325,12 @@ template<typename _Scalar, int _Options, typename _Index>
# ifdef EIGEN_DYNAMICSPARSEMATRIX_PLUGIN
# include EIGEN_DYNAMICSPARSEMATRIX_PLUGIN
# endif
};
};
template<typename Scalar, int _Options, typename _Index>
class DynamicSparseMatrix<Scalar,_Options,_Index>::InnerIterator : public SparseVector<Scalar,_Options,_Index>::InnerIterator
class DynamicSparseMatrix<Scalar,_Options,_Index>::InnerIterator : public SparseVector<Scalar,_Options>::InnerIterator
{
typedef typename SparseVector<Scalar,_Options,_Index>::InnerIterator Base;
typedef typename SparseVector<Scalar,_Options>::InnerIterator Base;
public:
InnerIterator(const DynamicSparseMatrix& mat, Index outer)
: Base(mat.m_data[outer]), m_outer(outer)
@@ -349,20 +343,4 @@ class DynamicSparseMatrix<Scalar,_Options,_Index>::InnerIterator : public Sparse
const Index m_outer;
};
template<typename Scalar, int _Options, typename _Index>
class DynamicSparseMatrix<Scalar,_Options,_Index>::ReverseInnerIterator : public SparseVector<Scalar,_Options,_Index>::ReverseInnerIterator
{
typedef typename SparseVector<Scalar,_Options,_Index>::ReverseInnerIterator Base;
public:
ReverseInnerIterator(const DynamicSparseMatrix& mat, Index outer)
: Base(mat.m_data[outer]), m_outer(outer)
{}
inline Index row() const { return IsRowMajor ? m_outer : Base::index(); }
inline Index col() const { return IsRowMajor ? Base::index() : m_outer; }
protected:
const Index m_outer;
};
#endif // EIGEN_DYNAMIC_SPARSEMATRIX_H

View File

@@ -63,17 +63,18 @@ class MappedSparseMatrix
inline Index cols() const { return IsRowMajor ? m_innerSize : m_outerSize; }
inline Index innerSize() const { return m_innerSize; }
inline Index outerSize() const { return m_outerSize; }
inline Index innerNonZeros(Index j) const { return m_outerIndex[j+1]-m_outerIndex[j]; }
//----------------------------------------
// direct access interface
inline const Scalar* valuePtr() const { return m_values; }
inline Scalar* valuePtr() { return m_values; }
inline const Scalar* _valuePtr() const { return m_values; }
inline Scalar* _valuePtr() { return m_values; }
inline const Index* innerIndexPtr() const { return m_innerIndices; }
inline Index* innerIndexPtr() { return m_innerIndices; }
inline const Index* _innerIndexPtr() const { return m_innerIndices; }
inline Index* _innerIndexPtr() { return m_innerIndices; }
inline const Index* outerIndexPtr() const { return m_outerIndex; }
inline Index* outerIndexPtr() { return m_outerIndex; }
inline const Index* _outerIndexPtr() const { return m_outerIndex; }
inline Index* _outerIndexPtr() { return m_outerIndex; }
//----------------------------------------
inline Scalar coeff(Index row, Index col) const
@@ -111,7 +112,6 @@ class MappedSparseMatrix
}
class InnerIterator;
class ReverseInnerIterator;
/** \returns the number of non zero coefficients */
inline Index nonZeros() const { return m_nnz; }
@@ -132,17 +132,23 @@ class MappedSparseMatrix<Scalar,_Flags,_Index>::InnerIterator
InnerIterator(const MappedSparseMatrix& mat, Index outer)
: m_matrix(mat),
m_outer(outer),
m_id(mat.outerIndexPtr()[outer]),
m_id(mat._outerIndexPtr()[outer]),
m_start(m_id),
m_end(mat.outerIndexPtr()[outer+1])
m_end(mat._outerIndexPtr()[outer+1])
{}
template<unsigned int Added, unsigned int Removed>
InnerIterator(const Flagged<MappedSparseMatrix,Added,Removed>& mat, Index outer)
: m_matrix(mat._expression()), m_id(m_matrix._outerIndexPtr()[outer]),
m_start(m_id), m_end(m_matrix._outerIndexPtr()[outer+1])
{}
inline InnerIterator& operator++() { m_id++; return *this; }
inline Scalar value() const { return m_matrix.valuePtr()[m_id]; }
inline Scalar& valueRef() { return const_cast<Scalar&>(m_matrix.valuePtr()[m_id]); }
inline Scalar value() const { return m_matrix._valuePtr()[m_id]; }
inline Scalar& valueRef() { return const_cast<Scalar&>(m_matrix._valuePtr()[m_id]); }
inline Index index() const { return m_matrix.innerIndexPtr()[m_id]; }
inline Index index() const { return m_matrix._innerIndexPtr()[m_id]; }
inline Index row() const { return IsRowMajor ? m_outer : index(); }
inline Index col() const { return IsRowMajor ? index() : m_outer; }
@@ -156,35 +162,4 @@ class MappedSparseMatrix<Scalar,_Flags,_Index>::InnerIterator
const Index m_end;
};
template<typename Scalar, int _Flags, typename _Index>
class MappedSparseMatrix<Scalar,_Flags,_Index>::ReverseInnerIterator
{
public:
ReverseInnerIterator(const MappedSparseMatrix& mat, Index outer)
: m_matrix(mat),
m_outer(outer),
m_id(mat.outerIndexPtr()[outer+1]),
m_start(mat.outerIndexPtr()[outer]),
m_end(m_id)
{}
inline ReverseInnerIterator& operator--() { m_id--; return *this; }
inline Scalar value() const { return m_matrix.valuePtr()[m_id-1]; }
inline Scalar& valueRef() { return const_cast<Scalar&>(m_matrix.valuePtr()[m_id-1]); }
inline Index index() const { return m_matrix.innerIndexPtr()[m_id-1]; }
inline Index row() const { return IsRowMajor ? m_outer : index(); }
inline Index col() const { return IsRowMajor ? index() : m_outer; }
inline operator bool() const { return (m_id <= m_end) && (m_id>m_start); }
protected:
const MappedSparseMatrix& m_matrix;
const Index m_outer;
Index m_id;
const Index m_start;
const Index m_end;
};
#endif // EIGEN_MAPPED_SPARSEMATRIX_H

View File

@@ -101,6 +101,103 @@ class SparseInnerVectorSet : internal::no_assignment_operator,
const internal::variable_if_dynamic<Index, Size> m_outerSize;
};
/***************************************************************************
* specialisation for DynamicSparseMatrix
***************************************************************************/
template<typename _Scalar, int _Options, int Size>
class SparseInnerVectorSet<DynamicSparseMatrix<_Scalar, _Options>, Size>
: public SparseMatrixBase<SparseInnerVectorSet<DynamicSparseMatrix<_Scalar, _Options>, Size> >
{
typedef DynamicSparseMatrix<_Scalar, _Options> MatrixType;
public:
enum { IsRowMajor = internal::traits<SparseInnerVectorSet>::IsRowMajor };
EIGEN_SPARSE_PUBLIC_INTERFACE(SparseInnerVectorSet)
class InnerIterator: public MatrixType::InnerIterator
{
public:
inline InnerIterator(const SparseInnerVectorSet& xpr, Index outer)
: MatrixType::InnerIterator(xpr.m_matrix, xpr.m_outerStart + outer), m_outer(outer)
{}
inline Index row() const { return IsRowMajor ? m_outer : this->index(); }
inline Index col() const { return IsRowMajor ? this->index() : m_outer; }
protected:
Index m_outer;
};
inline SparseInnerVectorSet(const MatrixType& matrix, Index outerStart, Index outerSize)
: m_matrix(matrix), m_outerStart(outerStart), m_outerSize(outerSize)
{
eigen_assert( (outerStart>=0) && ((outerStart+outerSize)<=matrix.outerSize()) );
}
inline SparseInnerVectorSet(const MatrixType& matrix, Index outer)
: m_matrix(matrix), m_outerStart(outer), m_outerSize(Size)
{
eigen_assert(Size!=Dynamic);
eigen_assert( (outer>=0) && (outer<matrix.outerSize()) );
}
template<typename OtherDerived>
inline SparseInnerVectorSet& operator=(const SparseMatrixBase<OtherDerived>& other)
{
if (IsRowMajor != ((OtherDerived::Flags&RowMajorBit)==RowMajorBit))
{
// need to transpose => perform a block evaluation followed by a big swap
DynamicSparseMatrix<Scalar,IsRowMajor?RowMajorBit:0> aux(other);
*this = aux.markAsRValue();
}
else
{
// evaluate/copy vector per vector
for (Index j=0; j<m_outerSize.value(); ++j)
{
SparseVector<Scalar,IsRowMajor ? RowMajorBit : 0> aux(other.innerVector(j));
m_matrix.const_cast_derived()._data()[m_outerStart+j].swap(aux._data());
}
}
return *this;
}
inline SparseInnerVectorSet& operator=(const SparseInnerVectorSet& other)
{
return operator=<SparseInnerVectorSet>(other);
}
Index nonZeros() const
{
Index count = 0;
for (Index j=0; j<m_outerSize.value(); ++j)
count += m_matrix._data()[m_outerStart+j].size();
return count;
}
const Scalar& lastCoeff() const
{
EIGEN_STATIC_ASSERT_VECTOR_ONLY(SparseInnerVectorSet);
eigen_assert(m_matrix.data()[m_outerStart].size()>0);
return m_matrix.data()[m_outerStart].vale(m_matrix.data()[m_outerStart].size()-1);
}
// template<typename Sparse>
// inline SparseInnerVectorSet& operator=(const SparseMatrixBase<OtherDerived>& other)
// {
// return *this;
// }
EIGEN_STRONG_INLINE Index rows() const { return IsRowMajor ? m_outerSize.value() : m_matrix.rows(); }
EIGEN_STRONG_INLINE Index cols() const { return IsRowMajor ? m_matrix.cols() : m_outerSize.value(); }
protected:
const typename MatrixType::Nested m_matrix;
Index m_outerStart;
const internal::variable_if_dynamic<Index, Size> m_outerSize;
};
/***************************************************************************
* specialisation for SparseMatrix
@@ -108,9 +205,9 @@ class SparseInnerVectorSet : internal::no_assignment_operator,
template<typename _Scalar, int _Options, typename _Index, int Size>
class SparseInnerVectorSet<SparseMatrix<_Scalar, _Options, _Index>, Size>
: public SparseMatrixBase<SparseInnerVectorSet<SparseMatrix<_Scalar, _Options, _Index>, Size> >
: public SparseMatrixBase<SparseInnerVectorSet<SparseMatrix<_Scalar, _Options>, Size> >
{
typedef SparseMatrix<_Scalar, _Options, _Index> MatrixType;
typedef SparseMatrix<_Scalar, _Options> MatrixType;
public:
enum { IsRowMajor = internal::traits<SparseInnerVectorSet>::IsRowMajor };
@@ -146,7 +243,7 @@ class SparseInnerVectorSet<SparseMatrix<_Scalar, _Options, _Index>, Size>
{
typedef typename internal::remove_all<typename MatrixType::Nested>::type _NestedMatrixType;
_NestedMatrixType& matrix = const_cast<_NestedMatrixType&>(m_matrix);;
// This assignement is slow if this vector set is not empty
// This assignement is slow if this vector set not empty
// and/or it is not at the end of the nonzeros of the underlying matrix.
// 1 - eval to a temporary to avoid transposition and/or aliasing issues
@@ -155,9 +252,9 @@ class SparseInnerVectorSet<SparseMatrix<_Scalar, _Options, _Index>, Size>
// 2 - let's check whether there is enough allocated memory
Index nnz = tmp.nonZeros();
Index nnz_previous = nonZeros();
Index free_size = matrix.data().allocatedSize() + nnz_previous;
std::size_t nnz_head = m_outerStart==0 ? 0 : matrix.outerIndexPtr()[m_outerStart];
std::size_t tail = m_matrix.outerIndexPtr()[m_outerStart+m_outerSize.value()];
Index free_size = matrix.data().allocatedSize() - nnz_previous;
std::size_t nnz_head = m_outerStart==0 ? 0 : matrix._outerIndexPtr()[m_outerStart];
std::size_t tail = m_matrix._outerIndexPtr()[m_outerStart+m_outerSize.value()];
std::size_t nnz_tail = matrix.nonZeros() - tail;
if(nnz>free_size)
@@ -201,15 +298,15 @@ class SparseInnerVectorSet<SparseMatrix<_Scalar, _Options, _Index>, Size>
// update outer index pointers
Index p = nnz_head;
for(Index k=0; k<m_outerSize.value(); ++k)
for(Index k=1; k<m_outerSize.value(); ++k)
{
matrix.outerIndexPtr()[m_outerStart+k] = p;
matrix._outerIndexPtr()[m_outerStart+k] = p;
p += tmp.innerVector(k).nonZeros();
}
std::ptrdiff_t offset = nnz - nnz_previous;
for(Index k = m_outerStart + m_outerSize.value(); k<=matrix.outerSize(); ++k)
{
matrix.outerIndexPtr()[k] += offset;
matrix._outerIndexPtr()[k] += offset;
}
return *this;
@@ -220,40 +317,32 @@ class SparseInnerVectorSet<SparseMatrix<_Scalar, _Options, _Index>, Size>
return operator=<SparseInnerVectorSet>(other);
}
inline const Scalar* valuePtr() const
{ return m_matrix.valuePtr() + m_matrix.outerIndexPtr()[m_outerStart]; }
inline Scalar* valuePtr()
{ return m_matrix.const_cast_derived().valuePtr() + m_matrix.outerIndexPtr()[m_outerStart]; }
inline const Scalar* _valuePtr() const
{ return m_matrix._valuePtr() + m_matrix._outerIndexPtr()[m_outerStart]; }
inline Scalar* _valuePtr()
{ return m_matrix.const_cast_derived()._valuePtr() + m_matrix._outerIndexPtr()[m_outerStart]; }
inline const Index* innerIndexPtr() const
{ return m_matrix.innerIndexPtr() + m_matrix.outerIndexPtr()[m_outerStart]; }
inline Index* innerIndexPtr()
{ return m_matrix.const_cast_derived().innerIndexPtr() + m_matrix.outerIndexPtr()[m_outerStart]; }
inline const Index* _innerIndexPtr() const
{ return m_matrix._innerIndexPtr() + m_matrix._outerIndexPtr()[m_outerStart]; }
inline Index* _innerIndexPtr()
{ return m_matrix.const_cast_derived()._innerIndexPtr() + m_matrix._outerIndexPtr()[m_outerStart]; }
inline const Index* outerIndexPtr() const
{ return m_matrix.outerIndexPtr() + m_outerStart; }
inline Index* outerIndexPtr()
{ return m_matrix.const_cast_derived().outerIndexPtr() + m_outerStart; }
inline const Index* _outerIndexPtr() const
{ return m_matrix._outerIndexPtr() + m_outerStart; }
inline Index* _outerIndexPtr()
{ return m_matrix.const_cast_derived()._outerIndexPtr() + m_outerStart; }
Index nonZeros() const
{
if(m_matrix.compressed())
return std::size_t(m_matrix.outerIndexPtr()[m_outerStart+m_outerSize.value()])
- std::size_t(m_matrix.outerIndexPtr()[m_outerStart]);
else if(m_outerSize.value()==0)
return 0;
else
return Map<const Matrix<Index,Size,1> >(m_matrix.innerNonZeroPtr(), m_outerSize.value()).sum();
return std::size_t(m_matrix._outerIndexPtr()[m_outerStart+m_outerSize.value()])
- std::size_t(m_matrix._outerIndexPtr()[m_outerStart]);
}
const Scalar& lastCoeff() const
{
EIGEN_STATIC_ASSERT_VECTOR_ONLY(SparseInnerVectorSet);
eigen_assert(nonZeros()>0);
if(m_matrix.compressed())
return m_matrix.valuePtr()[m_matrix.outerIndexPtr()[m_outerStart+1]-1];
else
return m_matrix.valuePtr()[m_matrix.outerIndexPtr()[m_outerStart]+m_matrix.innerNonZeroPtr()[m_outerStart]-1];
return m_matrix._valuePtr()[m_matrix._outerIndexPtr()[m_outerStart+1]-1];
}
// template<typename Sparse>
@@ -323,9 +412,11 @@ template<typename Derived>
const SparseInnerVectorSet<Derived,1> SparseMatrixBase<Derived>::innerVector(Index outer) const
{ return SparseInnerVectorSet<Derived,1>(derived(), outer); }
//----------
/** \returns the i-th row of the matrix \c *this. For row-major matrix only. */
template<typename Derived>
SparseInnerVectorSet<Derived,Dynamic> SparseMatrixBase<Derived>::middleRows(Index start, Index size)
SparseInnerVectorSet<Derived,Dynamic> SparseMatrixBase<Derived>::subrows(Index start, Index size)
{
EIGEN_STATIC_ASSERT(IsRowMajor,THIS_METHOD_IS_ONLY_FOR_ROW_MAJOR_MATRICES);
return innerVectors(start, size);
@@ -334,7 +425,7 @@ SparseInnerVectorSet<Derived,Dynamic> SparseMatrixBase<Derived>::middleRows(Inde
/** \returns the i-th row of the matrix \c *this. For row-major matrix only.
* (read-only version) */
template<typename Derived>
const SparseInnerVectorSet<Derived,Dynamic> SparseMatrixBase<Derived>::middleRows(Index start, Index size) const
const SparseInnerVectorSet<Derived,Dynamic> SparseMatrixBase<Derived>::subrows(Index start, Index size) const
{
EIGEN_STATIC_ASSERT(IsRowMajor,THIS_METHOD_IS_ONLY_FOR_ROW_MAJOR_MATRICES);
return innerVectors(start, size);
@@ -342,7 +433,7 @@ const SparseInnerVectorSet<Derived,Dynamic> SparseMatrixBase<Derived>::middleRow
/** \returns the i-th column of the matrix \c *this. For column-major matrix only. */
template<typename Derived>
SparseInnerVectorSet<Derived,Dynamic> SparseMatrixBase<Derived>::middleCols(Index start, Index size)
SparseInnerVectorSet<Derived,Dynamic> SparseMatrixBase<Derived>::subcols(Index start, Index size)
{
EIGEN_STATIC_ASSERT(!IsRowMajor,THIS_METHOD_IS_ONLY_FOR_COLUMN_MAJOR_MATRICES);
return innerVectors(start, size);
@@ -351,14 +442,12 @@ SparseInnerVectorSet<Derived,Dynamic> SparseMatrixBase<Derived>::middleCols(Inde
/** \returns the i-th column of the matrix \c *this. For column-major matrix only.
* (read-only version) */
template<typename Derived>
const SparseInnerVectorSet<Derived,Dynamic> SparseMatrixBase<Derived>::middleCols(Index start, Index size) const
const SparseInnerVectorSet<Derived,Dynamic> SparseMatrixBase<Derived>::subcols(Index start, Index size) const
{
EIGEN_STATIC_ASSERT(!IsRowMajor,THIS_METHOD_IS_ONLY_FOR_COLUMN_MAJOR_MATRICES);
return innerVectors(start, size);
}
/** \returns the \a outer -th column (resp. row) of the matrix \c *this if \c *this
* is col-major (resp. row-major).
*/

View File

@@ -298,6 +298,16 @@ class sparse_cwise_binary_op_inner_iterator_selector<scalar_product_op<T>, Lhs,
* Implementation of SparseMatrixBase and SparseCwise functions/operators
***************************************************************************/
// template<typename Derived>
// template<typename OtherDerived>
// EIGEN_STRONG_INLINE const CwiseBinaryOp<internal::scalar_difference_op<typename internal::traits<Derived>::Scalar>,
// Derived, OtherDerived>
// SparseMatrixBase<Derived>::operator-(const SparseMatrixBase<OtherDerived> &other) const
// {
// return CwiseBinaryOp<internal::scalar_difference_op<Scalar>,
// Derived, OtherDerived>(derived(), other.derived());
// }
template<typename Derived>
template<typename OtherDerived>
EIGEN_STRONG_INLINE Derived &
@@ -306,6 +316,14 @@ SparseMatrixBase<Derived>::operator-=(const SparseMatrixBase<OtherDerived> &othe
return *this = derived() - other.derived();
}
// template<typename Derived>
// template<typename OtherDerived>
// EIGEN_STRONG_INLINE const CwiseBinaryOp<internal::scalar_sum_op<typename internal::traits<Derived>::Scalar>, Derived, OtherDerived>
// SparseMatrixBase<Derived>::operator+(const SparseMatrixBase<OtherDerived> &other) const
// {
// return CwiseBinaryOp<internal::scalar_sum_op<Scalar>, Derived, OtherDerived>(derived(), other.derived());
// }
template<typename Derived>
template<typename OtherDerived>
EIGEN_STRONG_INLINE Derived &
@@ -314,6 +332,14 @@ SparseMatrixBase<Derived>::operator+=(const SparseMatrixBase<OtherDerived>& othe
return *this = derived() + other.derived();
}
// template<typename ExpressionType>
// template<typename OtherDerived>
// EIGEN_STRONG_INLINE const EIGEN_SPARSE_CWISE_PRODUCT_RETURN_TYPE
// SparseCwise<ExpressionType>::operator*(const SparseMatrixBase<OtherDerived> &other) const
// {
// return EIGEN_SPARSE_CWISE_PRODUCT_RETURN_TYPE(_expression(), other.derived());
// }
template<typename Derived>
template<typename OtherDerived>
EIGEN_STRONG_INLINE const EIGEN_SPARSE_CWISE_PRODUCT_RETURN_TYPE
@@ -322,4 +348,28 @@ SparseMatrixBase<Derived>::cwiseProduct(const MatrixBase<OtherDerived> &other) c
return EIGEN_SPARSE_CWISE_PRODUCT_RETURN_TYPE(derived(), other.derived());
}
// template<typename ExpressionType>
// template<typename OtherDerived>
// EIGEN_STRONG_INLINE const EIGEN_SPARSE_CWISE_BINOP_RETURN_TYPE(internal::scalar_quotient_op)
// SparseCwise<ExpressionType>::operator/(const SparseMatrixBase<OtherDerived> &other) const
// {
// return EIGEN_SPARSE_CWISE_BINOP_RETURN_TYPE(internal::scalar_quotient_op)(_expression(), other.derived());
// }
//
// template<typename ExpressionType>
// template<typename OtherDerived>
// EIGEN_STRONG_INLINE const EIGEN_SPARSE_CWISE_BINOP_RETURN_TYPE(internal::scalar_quotient_op)
// SparseCwise<ExpressionType>::operator/(const MatrixBase<OtherDerived> &other) const
// {
// return EIGEN_SPARSE_CWISE_BINOP_RETURN_TYPE(internal::scalar_quotient_op)(_expression(), other.derived());
// }
// template<typename ExpressionType>
// template<typename OtherDerived>
// inline ExpressionType& SparseCwise<ExpressionType>::operator*=(const SparseMatrixBase<OtherDerived> &other)
// {
// return m_matrix.const_cast_derived() = _expression() * other.derived();
// }
#endif // EIGEN_SPARSE_CWISE_BINARY_OP_H

View File

@@ -25,6 +25,19 @@
#ifndef EIGEN_SPARSE_CWISE_UNARY_OP_H
#define EIGEN_SPARSE_CWISE_UNARY_OP_H
// template<typename UnaryOp, typename MatrixType>
// struct internal::traits<SparseCwiseUnaryOp<UnaryOp, MatrixType> > : internal::traits<MatrixType>
// {
// typedef typename internal::result_of<
// UnaryOp(typename MatrixType::Scalar)
// >::type Scalar;
// typedef typename MatrixType::Nested MatrixTypeNested;
// typedef typename internal::remove_reference<MatrixTypeNested>::type _MatrixTypeNested;
// enum {
// CoeffReadCost = _MatrixTypeNested::CoeffReadCost + internal::functor_traits<UnaryOp>::Cost
// };
// };
template<typename UnaryOp, typename MatrixType>
class CwiseUnaryOpImpl<UnaryOp,MatrixType,Sparse>
: public SparseMatrixBase<CwiseUnaryOp<UnaryOp, MatrixType> >
@@ -32,61 +45,39 @@ class CwiseUnaryOpImpl<UnaryOp,MatrixType,Sparse>
public:
class InnerIterator;
class ReverseInnerIterator;
// typedef typename internal::remove_reference<LhsNested>::type _LhsNested;
typedef CwiseUnaryOp<UnaryOp, MatrixType> Derived;
EIGEN_SPARSE_PUBLIC_INTERFACE(Derived)
protected:
typedef typename internal::traits<Derived>::_XprTypeNested _MatrixTypeNested;
typedef typename _MatrixTypeNested::InnerIterator MatrixTypeIterator;
typedef typename _MatrixTypeNested::ReverseInnerIterator MatrixTypeReverseIterator;
};
template<typename UnaryOp, typename MatrixType>
class CwiseUnaryOpImpl<UnaryOp,MatrixType,Sparse>::InnerIterator
: public CwiseUnaryOpImpl<UnaryOp,MatrixType,Sparse>::MatrixTypeIterator
{
typedef typename CwiseUnaryOpImpl::Scalar Scalar;
typedef typename CwiseUnaryOpImpl<UnaryOp,MatrixType,Sparse>::MatrixTypeIterator Base;
typedef typename internal::traits<Derived>::_XprTypeNested _MatrixTypeNested;
typedef typename _MatrixTypeNested::InnerIterator MatrixTypeIterator;
typedef typename MatrixType::Index Index;
public:
EIGEN_STRONG_INLINE InnerIterator(const CwiseUnaryOpImpl& unaryOp, Index outer)
: Base(unaryOp.derived().nestedExpression(),outer), m_functor(unaryOp.derived().functor())
: m_iter(unaryOp.derived().nestedExpression(),outer), m_functor(unaryOp.derived().functor())
{}
EIGEN_STRONG_INLINE InnerIterator& operator++()
{ Base::operator++(); return *this; }
{ ++m_iter; return *this; }
EIGEN_STRONG_INLINE Scalar value() const { return m_functor(Base::value()); }
EIGEN_STRONG_INLINE Scalar value() const { return m_functor(m_iter.value()); }
EIGEN_STRONG_INLINE Index index() const { return m_iter.index(); }
EIGEN_STRONG_INLINE Index row() const { return m_iter.row(); }
EIGEN_STRONG_INLINE Index col() const { return m_iter.col(); }
EIGEN_STRONG_INLINE operator bool() const { return m_iter; }
protected:
MatrixTypeIterator m_iter;
const UnaryOp m_functor;
private:
Scalar& valueRef();
};
template<typename UnaryOp, typename MatrixType>
class CwiseUnaryOpImpl<UnaryOp,MatrixType,Sparse>::ReverseInnerIterator
: public CwiseUnaryOpImpl<UnaryOp,MatrixType,Sparse>::MatrixTypeReverseIterator
{
typedef typename CwiseUnaryOpImpl::Scalar Scalar;
typedef typename CwiseUnaryOpImpl<UnaryOp,MatrixType,Sparse>::MatrixTypeReverseIterator Base;
public:
EIGEN_STRONG_INLINE ReverseInnerIterator(const CwiseUnaryOpImpl& unaryOp, Index outer)
: Base(unaryOp.derived().nestedExpression(),outer), m_functor(unaryOp.derived().functor())
{}
EIGEN_STRONG_INLINE ReverseInnerIterator& operator--()
{ Base::operator--(); return *this; }
EIGEN_STRONG_INLINE Scalar value() const { return m_functor(Base::value()); }
protected:
const UnaryOp m_functor;
private:
Scalar& valueRef();
};
template<typename ViewOp, typename MatrixType>
@@ -96,58 +87,39 @@ class CwiseUnaryViewImpl<ViewOp,MatrixType,Sparse>
public:
class InnerIterator;
class ReverseInnerIterator;
// typedef typename internal::remove_reference<LhsNested>::type _LhsNested;
typedef CwiseUnaryView<ViewOp, MatrixType> Derived;
EIGEN_SPARSE_PUBLIC_INTERFACE(Derived)
protected:
typedef typename internal::traits<Derived>::_MatrixTypeNested _MatrixTypeNested;
typedef typename _MatrixTypeNested::InnerIterator MatrixTypeIterator;
typedef typename _MatrixTypeNested::ReverseInnerIterator MatrixTypeReverseIterator;
};
template<typename ViewOp, typename MatrixType>
class CwiseUnaryViewImpl<ViewOp,MatrixType,Sparse>::InnerIterator
: public CwiseUnaryViewImpl<ViewOp,MatrixType,Sparse>::MatrixTypeIterator
{
typedef typename CwiseUnaryViewImpl::Scalar Scalar;
typedef typename CwiseUnaryViewImpl<ViewOp,MatrixType,Sparse>::MatrixTypeIterator Base;
typedef typename internal::traits<Derived>::_MatrixTypeNested _MatrixTypeNested;
typedef typename _MatrixTypeNested::InnerIterator MatrixTypeIterator;
typedef typename MatrixType::Index Index;
public:
EIGEN_STRONG_INLINE InnerIterator(const CwiseUnaryViewImpl& unaryOp, Index outer)
: Base(unaryOp.derived().nestedExpression(),outer), m_functor(unaryOp.derived().functor())
EIGEN_STRONG_INLINE InnerIterator(const CwiseUnaryViewImpl& unaryView, Index outer)
: m_iter(unaryView.derived().nestedExpression(),outer), m_functor(unaryView.derived().functor())
{}
EIGEN_STRONG_INLINE InnerIterator& operator++()
{ Base::operator++(); return *this; }
{ ++m_iter; return *this; }
EIGEN_STRONG_INLINE Scalar value() const { return m_functor(Base::value()); }
EIGEN_STRONG_INLINE Scalar& valueRef() { return m_functor(Base::valueRef()); }
protected:
const ViewOp m_functor;
};
template<typename ViewOp, typename MatrixType>
class CwiseUnaryViewImpl<ViewOp,MatrixType,Sparse>::ReverseInnerIterator
: public CwiseUnaryViewImpl<ViewOp,MatrixType,Sparse>::MatrixTypeReverseIterator
{
typedef typename CwiseUnaryViewImpl::Scalar Scalar;
typedef typename CwiseUnaryViewImpl<ViewOp,MatrixType,Sparse>::MatrixTypeReverseIterator Base;
public:
EIGEN_STRONG_INLINE ReverseInnerIterator(const CwiseUnaryViewImpl& unaryOp, Index outer)
: Base(unaryOp.derived().nestedExpression(),outer), m_functor(unaryOp.derived().functor())
{}
EIGEN_STRONG_INLINE ReverseInnerIterator& operator--()
{ Base::operator--(); return *this; }
EIGEN_STRONG_INLINE Scalar value() const { return m_functor(Base::value()); }
EIGEN_STRONG_INLINE Scalar& valueRef() { return m_functor(Base::valueRef()); }
EIGEN_STRONG_INLINE Scalar value() const { return m_functor(m_iter.value()); }
EIGEN_STRONG_INLINE Scalar& valueRef() { return m_functor(m_iter.valueRef()); }
EIGEN_STRONG_INLINE Index index() const { return m_iter.index(); }
EIGEN_STRONG_INLINE Index row() const { return m_iter.row(); }
EIGEN_STRONG_INLINE Index col() const { return m_iter.col(); }
EIGEN_STRONG_INLINE operator bool() const { return m_iter; }
protected:
MatrixTypeIterator m_iter;
const ViewOp m_functor;
};

View File

@@ -166,25 +166,17 @@ class SparseTimeDenseProduct
typedef typename internal::remove_all<Lhs>::type _Lhs;
typedef typename internal::remove_all<Rhs>::type _Rhs;
typedef typename _Lhs::InnerIterator LhsInnerIterator;
enum {
LhsIsRowMajor = (_Lhs::Flags&RowMajorBit)==RowMajorBit,
RhsIsVector = Rhs::ColsAtCompileTime==1
};
Index j=0;
for(j=0; j<m_lhs.outerSize(); ++j)
enum { LhsIsRowMajor = (_Lhs::Flags&RowMajorBit)==RowMajorBit };
for(Index j=0; j<m_lhs.outerSize(); ++j)
{
typename Rhs::Scalar rhs_j = alpha * m_rhs.coeff(LhsIsRowMajor ? 0 : j,0);
typename Dest::RowXpr dest_j(dest.row(LhsIsRowMajor ? j : 0));
typename Dest::Scalar tmp(0);
for(LhsInnerIterator it(m_lhs,j); it ;++it)
{
if(LhsIsRowMajor && RhsIsVector) tmp += (it.value()) * m_rhs.coeff(it.index());
else if(LhsIsRowMajor) dest_j += (alpha*it.value()) * m_rhs.row(it.index());
else if(RhsIsVector) dest.coeffRef(it.index()) += it.value() * rhs_j;
else dest.row(it.index()) += (alpha*it.value()) * m_rhs.row(j);
if(LhsIsRowMajor) dest_j += (alpha*it.value()) * m_rhs.row(it.index());
else if(Rhs::ColsAtCompileTime==1) dest.coeffRef(it.index()) += it.value() * rhs_j;
else dest.row(it.index()) += (alpha*it.value()) * m_rhs.row(j);
}
if(LhsIsRowMajor && RhsIsVector)
dest.coeffRef(LhsIsRowMajor ? j : 0) = alpha * tmp;
}
}

View File

@@ -0,0 +1,651 @@
// This file is part of Eigen, a lightweight C++ template library
// for linear algebra.
//
// Copyright (C) 2008-2010 Gael Guennebaud <gael.guennebaud@inria.fr>
//
// Eigen is free software; you can redistribute it and/or
// modify it under the terms of the GNU Lesser General Public
// License as published by the Free Software Foundation; either
// version 3 of the License, or (at your option) any later version.
//
// Alternatively, you can redistribute it and/or
// modify it under the terms of the GNU General Public License as
// published by the Free Software Foundation; either version 2 of
// the License, or (at your option) any later version.
//
// Eigen is distributed in the hope that it will be useful, but WITHOUT ANY
// WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS
// FOR A PARTICULAR PURPOSE. See the GNU Lesser General Public License or the
// GNU General Public License for more details.
//
// You should have received a copy of the GNU Lesser General Public
// License and a copy of the GNU General Public License along with
// Eigen. If not, see <http://www.gnu.org/licenses/>.
#ifndef EIGEN_SPARSEMATRIX_H
#define EIGEN_SPARSEMATRIX_H
/** \ingroup Sparse_Module
*
* \class SparseMatrix
*
* \brief The main sparse matrix class
*
* This class implements a sparse matrix using the very common compressed row/column storage
* scheme.
*
* \tparam _Scalar the scalar type, i.e. the type of the coefficients
* \tparam _Options Union of bit flags controlling the storage scheme. Currently the only possibility
* is RowMajor. The default is 0 which means column-major.
* \tparam _Index the type of the indices. Default is \c int.
*
* See http://www.netlib.org/linalg/html_templates/node91.html for details on the storage scheme.
*
* This class can be extended with the help of the plugin mechanism described on the page
* \ref TopicCustomizingEigen by defining the preprocessor symbol \c EIGEN_SPARSEMATRIX_PLUGIN.
*/
namespace internal {
template<typename _Scalar, int _Options, typename _Index>
struct traits<SparseMatrix<_Scalar, _Options, _Index> >
{
typedef _Scalar Scalar;
typedef _Index Index;
typedef Sparse StorageKind;
typedef MatrixXpr XprKind;
enum {
RowsAtCompileTime = Dynamic,
ColsAtCompileTime = Dynamic,
MaxRowsAtCompileTime = Dynamic,
MaxColsAtCompileTime = Dynamic,
Flags = _Options | NestByRefBit | LvalueBit,
CoeffReadCost = NumTraits<Scalar>::ReadCost,
SupportedAccessPatterns = InnerRandomAccessPattern
};
};
} // end namespace internal
template<typename _Scalar, int _Options, typename _Index>
class SparseMatrix
: public SparseMatrixBase<SparseMatrix<_Scalar, _Options, _Index> >
{
public:
EIGEN_SPARSE_PUBLIC_INTERFACE(SparseMatrix)
// using Base::operator=;
EIGEN_SPARSE_INHERIT_ASSIGNMENT_OPERATOR(SparseMatrix, +=)
EIGEN_SPARSE_INHERIT_ASSIGNMENT_OPERATOR(SparseMatrix, -=)
// FIXME: why are these operator already alvailable ???
// EIGEN_SPARSE_INHERIT_SCALAR_ASSIGNMENT_OPERATOR(SparseMatrix, *=)
// EIGEN_SPARSE_INHERIT_SCALAR_ASSIGNMENT_OPERATOR(SparseMatrix, /=)
typedef MappedSparseMatrix<Scalar,Flags> Map;
using Base::IsRowMajor;
typedef CompressedStorage<Scalar,Index> Storage;
enum {
Options = _Options
};
protected:
typedef SparseMatrix<Scalar,(Flags&~RowMajorBit)|(IsRowMajor?RowMajorBit:0)> TransposedSparseMatrix;
Index m_outerSize;
Index m_innerSize;
Index* m_outerIndex;
CompressedStorage<Scalar,Index> m_data;
public:
inline Index rows() const { return IsRowMajor ? m_outerSize : m_innerSize; }
inline Index cols() const { return IsRowMajor ? m_innerSize : m_outerSize; }
inline Index innerSize() const { return m_innerSize; }
inline Index outerSize() const { return m_outerSize; }
inline Index innerNonZeros(Index j) const { return m_outerIndex[j+1]-m_outerIndex[j]; }
inline const Scalar* _valuePtr() const { return &m_data.value(0); }
inline Scalar* _valuePtr() { return &m_data.value(0); }
inline const Index* _innerIndexPtr() const { return &m_data.index(0); }
inline Index* _innerIndexPtr() { return &m_data.index(0); }
inline const Index* _outerIndexPtr() const { return m_outerIndex; }
inline Index* _outerIndexPtr() { return m_outerIndex; }
inline Storage& data() { return m_data; }
inline const Storage& data() const { return m_data; }
inline Scalar coeff(Index row, Index col) const
{
const Index outer = IsRowMajor ? row : col;
const Index inner = IsRowMajor ? col : row;
return m_data.atInRange(m_outerIndex[outer], m_outerIndex[outer+1], inner);
}
inline Scalar& coeffRef(Index row, Index col)
{
const Index outer = IsRowMajor ? row : col;
const Index inner = IsRowMajor ? col : row;
Index start = m_outerIndex[outer];
Index end = m_outerIndex[outer+1];
eigen_assert(end>=start && "you probably called coeffRef on a non finalized matrix");
eigen_assert(end>start && "coeffRef cannot be called on a zero coefficient");
const Index p = m_data.searchLowerIndex(start,end-1,inner);
eigen_assert((p<end) && (m_data.index(p)==inner) && "coeffRef cannot be called on a zero coefficient");
return m_data.value(p);
}
public:
class InnerIterator;
/** Removes all non zeros */
inline void setZero()
{
m_data.clear();
memset(m_outerIndex, 0, (m_outerSize+1)*sizeof(Index));
}
/** \returns the number of non zero coefficients */
inline Index nonZeros() const { return static_cast<Index>(m_data.size()); }
/** Preallocates \a reserveSize non zeros */
inline void reserve(Index reserveSize)
{
m_data.reserve(reserveSize);
}
//--- low level purely coherent filling ---
/** \returns a reference to the non zero coefficient at position \a row, \a col assuming that:
* - the nonzero does not already exist
* - the new coefficient is the last one according to the storage order
*
* Before filling a given inner vector you must call the statVec(Index) function.
*
* After an insertion session, you should call the finalize() function.
*
* \sa insert, insertBackByOuterInner, startVec */
inline Scalar& insertBack(Index row, Index col)
{
return insertBackByOuterInner(IsRowMajor?row:col, IsRowMajor?col:row);
}
/** \sa insertBack, startVec */
inline Scalar& insertBackByOuterInner(Index outer, Index inner)
{
eigen_assert(size_t(m_outerIndex[outer+1]) == m_data.size() && "Invalid ordered insertion (invalid outer index)");
eigen_assert( (m_outerIndex[outer+1]-m_outerIndex[outer]==0 || m_data.index(m_data.size()-1)<inner) && "Invalid ordered insertion (invalid inner index)");
Index p = m_outerIndex[outer+1];
++m_outerIndex[outer+1];
m_data.append(0, inner);
return m_data.value(p);
}
/** \warning use it only if you know what you are doing */
inline Scalar& insertBackByOuterInnerUnordered(Index outer, Index inner)
{
Index p = m_outerIndex[outer+1];
++m_outerIndex[outer+1];
m_data.append(0, inner);
return m_data.value(p);
}
/** \sa insertBack, insertBackByOuterInner */
inline void startVec(Index outer)
{
eigen_assert(m_outerIndex[outer]==int(m_data.size()) && "You must call startVec for each inner vector sequentially");
eigen_assert(m_outerIndex[outer+1]==0 && "You must call startVec for each inner vector sequentially");
m_outerIndex[outer+1] = m_outerIndex[outer];
}
//---
/** \returns a reference to a novel non zero coefficient with coordinates \a row x \a col.
* The non zero coefficient must \b not already exist.
*
* \warning This function can be extremely slow if the non zero coefficients
* are not inserted in a coherent order.
*
* After an insertion session, you should call the finalize() function.
*/
EIGEN_DONT_INLINE Scalar& insert(Index row, Index col)
{
const Index outer = IsRowMajor ? row : col;
const Index inner = IsRowMajor ? col : row;
Index previousOuter = outer;
if (m_outerIndex[outer+1]==0)
{
// we start a new inner vector
while (previousOuter>=0 && m_outerIndex[previousOuter]==0)
{
m_outerIndex[previousOuter] = static_cast<Index>(m_data.size());
--previousOuter;
}
m_outerIndex[outer+1] = m_outerIndex[outer];
}
// here we have to handle the tricky case where the outerIndex array
// starts with: [ 0 0 0 0 0 1 ...] and we are inserting in, e.g.,
// the 2nd inner vector...
bool isLastVec = (!(previousOuter==-1 && m_data.size()!=0))
&& (size_t(m_outerIndex[outer+1]) == m_data.size());
size_t startId = m_outerIndex[outer];
// FIXME let's make sure sizeof(long int) == sizeof(size_t)
size_t p = m_outerIndex[outer+1];
++m_outerIndex[outer+1];
float reallocRatio = 1;
if (m_data.allocatedSize()<=m_data.size())
{
// if there is no preallocated memory, let's reserve a minimum of 32 elements
if (m_data.size()==0)
{
m_data.reserve(32);
}
else
{
// we need to reallocate the data, to reduce multiple reallocations
// we use a smart resize algorithm based on the current filling ratio
// in addition, we use float to avoid integers overflows
float nnzEstimate = float(m_outerIndex[outer])*float(m_outerSize)/float(outer+1);
reallocRatio = (nnzEstimate-float(m_data.size()))/float(m_data.size());
// furthermore we bound the realloc ratio to:
// 1) reduce multiple minor realloc when the matrix is almost filled
// 2) avoid to allocate too much memory when the matrix is almost empty
reallocRatio = (std::min)((std::max)(reallocRatio,1.5f),8.f);
}
}
m_data.resize(m_data.size()+1,reallocRatio);
if (!isLastVec)
{
if (previousOuter==-1)
{
// oops wrong guess.
// let's correct the outer offsets
for (Index k=0; k<=(outer+1); ++k)
m_outerIndex[k] = 0;
Index k=outer+1;
while(m_outerIndex[k]==0)
m_outerIndex[k++] = 1;
while (k<=m_outerSize && m_outerIndex[k]!=0)
m_outerIndex[k++]++;
p = 0;
--k;
k = m_outerIndex[k]-1;
while (k>0)
{
m_data.index(k) = m_data.index(k-1);
m_data.value(k) = m_data.value(k-1);
k--;
}
}
else
{
// we are not inserting into the last inner vec
// update outer indices:
Index j = outer+2;
while (j<=m_outerSize && m_outerIndex[j]!=0)
m_outerIndex[j++]++;
--j;
// shift data of last vecs:
Index k = m_outerIndex[j]-1;
while (k>=Index(p))
{
m_data.index(k) = m_data.index(k-1);
m_data.value(k) = m_data.value(k-1);
k--;
}
}
}
while ( (p > startId) && (m_data.index(p-1) > inner) )
{
m_data.index(p) = m_data.index(p-1);
m_data.value(p) = m_data.value(p-1);
--p;
}
m_data.index(p) = inner;
return (m_data.value(p) = 0);
}
/** Must be called after inserting a set of non zero entries.
*/
inline void finalize()
{
Index size = static_cast<Index>(m_data.size());
Index i = m_outerSize;
// find the last filled column
while (i>=0 && m_outerIndex[i]==0)
--i;
++i;
while (i<=m_outerSize)
{
m_outerIndex[i] = size;
++i;
}
}
/** Suppress all nonzeros which are smaller than \a reference under the tolerence \a epsilon */
void prune(Scalar reference, RealScalar epsilon = NumTraits<RealScalar>::dummy_precision())
{
prune(default_prunning_func(reference,epsilon));
}
/** Suppress all nonzeros which do not satisfy the predicate \a keep.
* The functor type \a KeepFunc must implement the following function:
* \code
* bool operator() (const Index& row, const Index& col, const Scalar& value) const;
* \endcode
* \sa prune(Scalar,RealScalar)
*/
template<typename KeepFunc>
void prune(const KeepFunc& keep = KeepFunc())
{
Index k = 0;
for(Index j=0; j<m_outerSize; ++j)
{
Index previousStart = m_outerIndex[j];
m_outerIndex[j] = k;
Index end = m_outerIndex[j+1];
for(Index i=previousStart; i<end; ++i)
{
if(keep(IsRowMajor?j:m_data.index(i), IsRowMajor?m_data.index(i):j, m_data.value(i)))
{
m_data.value(k) = m_data.value(i);
m_data.index(k) = m_data.index(i);
++k;
}
}
}
m_outerIndex[m_outerSize] = k;
m_data.resize(k,0);
}
/** Resizes the matrix to a \a rows x \a cols matrix and initializes it to zero
* \sa resizeNonZeros(Index), reserve(), setZero()
*/
void resize(Index rows, Index cols)
{
const Index outerSize = IsRowMajor ? rows : cols;
m_innerSize = IsRowMajor ? cols : rows;
m_data.clear();
if (m_outerSize != outerSize || m_outerSize==0)
{
delete[] m_outerIndex;
m_outerIndex = new Index [outerSize+1];
m_outerSize = outerSize;
}
memset(m_outerIndex, 0, (m_outerSize+1)*sizeof(Index));
}
/** Low level API
* Resize the nonzero vector to \a size */
void resizeNonZeros(Index size)
{
m_data.resize(size);
}
/** Default constructor yielding an empty \c 0 \c x \c 0 matrix */
inline SparseMatrix()
: m_outerSize(-1), m_innerSize(0), m_outerIndex(0)
{
resize(0, 0);
}
/** Constructs a \a rows \c x \a cols empty matrix */
inline SparseMatrix(Index rows, Index cols)
: m_outerSize(0), m_innerSize(0), m_outerIndex(0)
{
resize(rows, cols);
}
/** Constructs a sparse matrix from the sparse expression \a other */
template<typename OtherDerived>
inline SparseMatrix(const SparseMatrixBase<OtherDerived>& other)
: m_outerSize(0), m_innerSize(0), m_outerIndex(0)
{
*this = other.derived();
}
/** Copy constructor */
inline SparseMatrix(const SparseMatrix& other)
: Base(), m_outerSize(0), m_innerSize(0), m_outerIndex(0)
{
*this = other.derived();
}
/** Swap the content of two sparse matrices of same type (optimization) */
inline void swap(SparseMatrix& other)
{
//EIGEN_DBG_SPARSE(std::cout << "SparseMatrix:: swap\n");
std::swap(m_outerIndex, other.m_outerIndex);
std::swap(m_innerSize, other.m_innerSize);
std::swap(m_outerSize, other.m_outerSize);
m_data.swap(other.m_data);
}
inline SparseMatrix& operator=(const SparseMatrix& other)
{
// std::cout << "SparseMatrix& operator=(const SparseMatrix& other)\n";
if (other.isRValue())
{
swap(other.const_cast_derived());
}
else
{
resize(other.rows(), other.cols());
memcpy(m_outerIndex, other.m_outerIndex, (m_outerSize+1)*sizeof(Index));
m_data = other.m_data;
}
return *this;
}
#ifndef EIGEN_PARSED_BY_DOXYGEN
template<typename Lhs, typename Rhs>
inline SparseMatrix& operator=(const SparseSparseProduct<Lhs,Rhs>& product)
{ return Base::operator=(product); }
template<typename OtherDerived>
inline SparseMatrix& operator=(const ReturnByValue<OtherDerived>& other)
{ return Base::operator=(other); }
template<typename OtherDerived>
inline SparseMatrix& operator=(const EigenBase<OtherDerived>& other)
{ return Base::operator=(other); }
#endif
template<typename OtherDerived>
EIGEN_DONT_INLINE SparseMatrix& operator=(const SparseMatrixBase<OtherDerived>& other)
{
const bool needToTranspose = (Flags & RowMajorBit) != (OtherDerived::Flags & RowMajorBit);
if (needToTranspose)
{
// two passes algorithm:
// 1 - compute the number of coeffs per dest inner vector
// 2 - do the actual copy/eval
// Since each coeff of the rhs has to be evaluated twice, let's evaluate it if needed
typedef typename internal::nested<OtherDerived,2>::type OtherCopy;
typedef typename internal::remove_all<OtherCopy>::type _OtherCopy;
OtherCopy otherCopy(other.derived());
resize(other.rows(), other.cols());
Eigen::Map<Matrix<Index, Dynamic, 1> > (m_outerIndex,outerSize()).setZero();
// pass 1
// FIXME the above copy could be merged with that pass
for (Index j=0; j<otherCopy.outerSize(); ++j)
for (typename _OtherCopy::InnerIterator it(otherCopy, j); it; ++it)
++m_outerIndex[it.index()];
// prefix sum
Index count = 0;
VectorXi positions(outerSize());
for (Index j=0; j<outerSize(); ++j)
{
Index tmp = m_outerIndex[j];
m_outerIndex[j] = count;
positions[j] = count;
count += tmp;
}
m_outerIndex[outerSize()] = count;
// alloc
m_data.resize(count);
// pass 2
for (Index j=0; j<otherCopy.outerSize(); ++j)
{
for (typename _OtherCopy::InnerIterator it(otherCopy, j); it; ++it)
{
Index pos = positions[it.index()]++;
m_data.index(pos) = j;
m_data.value(pos) = it.value();
}
}
return *this;
}
else
{
// there is no special optimization
return SparseMatrixBase<SparseMatrix>::operator=(other.derived());
}
}
friend std::ostream & operator << (std::ostream & s, const SparseMatrix& m)
{
EIGEN_DBG_SPARSE(
s << "Nonzero entries:\n";
for (Index i=0; i<m.nonZeros(); ++i)
{
s << "(" << m.m_data.value(i) << "," << m.m_data.index(i) << ") ";
}
s << std::endl;
s << std::endl;
s << "Column pointers:\n";
for (Index i=0; i<m.outerSize(); ++i)
{
s << m.m_outerIndex[i] << " ";
}
s << " $" << std::endl;
s << std::endl;
);
s << static_cast<const SparseMatrixBase<SparseMatrix>&>(m);
return s;
}
/** Destructor */
inline ~SparseMatrix()
{
delete[] m_outerIndex;
}
/** Overloaded for performance */
Scalar sum() const;
public:
/** \deprecated use setZero() and reserve()
* Initializes the filling process of \c *this.
* \param reserveSize approximate number of nonzeros
* Note that the matrix \c *this is zero-ed.
*/
EIGEN_DEPRECATED void startFill(Index reserveSize = 1000)
{
setZero();
m_data.reserve(reserveSize);
}
/** \deprecated use insert()
* Like fill() but with random inner coordinates.
*/
EIGEN_DEPRECATED Scalar& fillrand(Index row, Index col)
{
return insert(row,col);
}
/** \deprecated use insert()
*/
EIGEN_DEPRECATED Scalar& fill(Index row, Index col)
{
const Index outer = IsRowMajor ? row : col;
const Index inner = IsRowMajor ? col : row;
if (m_outerIndex[outer+1]==0)
{
// we start a new inner vector
Index i = outer;
while (i>=0 && m_outerIndex[i]==0)
{
m_outerIndex[i] = m_data.size();
--i;
}
m_outerIndex[outer+1] = m_outerIndex[outer];
}
else
{
eigen_assert(m_data.index(m_data.size()-1)<inner && "wrong sorted insertion");
}
// std::cerr << size_t(m_outerIndex[outer+1]) << " == " << m_data.size() << "\n";
assert(size_t(m_outerIndex[outer+1]) == m_data.size());
Index p = m_outerIndex[outer+1];
++m_outerIndex[outer+1];
m_data.append(0, inner);
return m_data.value(p);
}
/** \deprecated use finalize */
EIGEN_DEPRECATED void endFill() { finalize(); }
# ifdef EIGEN_SPARSEMATRIX_PLUGIN
# include EIGEN_SPARSEMATRIX_PLUGIN
# endif
private:
struct default_prunning_func {
default_prunning_func(Scalar ref, RealScalar eps) : reference(ref), epsilon(eps) {}
inline bool operator() (const Index&, const Index&, const Scalar& value) const
{
return !internal::isMuchSmallerThan(value, reference, epsilon);
}
Scalar reference;
RealScalar epsilon;
};
};
template<typename Scalar, int _Options, typename _Index>
class SparseMatrix<Scalar,_Options,_Index>::InnerIterator
{
public:
InnerIterator(const SparseMatrix& mat, Index outer)
: m_values(mat._valuePtr()), m_indices(mat._innerIndexPtr()), m_outer(outer), m_id(mat.m_outerIndex[outer]), m_end(mat.m_outerIndex[outer+1])
{}
inline InnerIterator& operator++() { m_id++; return *this; }
inline const Scalar& value() const { return m_values[m_id]; }
inline Scalar& valueRef() { return const_cast<Scalar&>(m_values[m_id]); }
inline Index index() const { return m_indices[m_id]; }
inline Index outer() const { return m_outer; }
inline Index row() const { return IsRowMajor ? m_outer : index(); }
inline Index col() const { return IsRowMajor ? index() : m_outer; }
inline operator bool() const { return (m_id < m_end); }
protected:
const Scalar* m_values;
const Index* m_indices;
const Index m_outer;
Index m_id;
const Index m_end;
};
#endif // EIGEN_SPARSEMATRIX_H

View File

@@ -1,7 +1,7 @@
// This file is part of Eigen, a lightweight C++ template library
// for linear algebra.
//
// Copyright (C) 2008-2011 Gael Guennebaud <gael.guennebaud@inria.fr>
// Copyright (C) 2008-2009 Gael Guennebaud <gael.guennebaud@inria.fr>
//
// Eigen is free software; you can redistribute it and/or
// modify it under the terms of the GNU Lesser General Public
@@ -25,7 +25,7 @@
#ifndef EIGEN_SPARSEMATRIXBASE_H
#define EIGEN_SPARSEMATRIXBASE_H
/** \ingroup SparseCore_Module
/** \ingroup Sparse_Module
*
* \class SparseMatrixBase
*
@@ -44,9 +44,6 @@ template<typename Derived> class SparseMatrixBase : public EigenBase<Derived>
typedef typename internal::packet_traits<Scalar>::type PacketScalar;
typedef typename internal::traits<Derived>::StorageKind StorageKind;
typedef typename internal::traits<Derived>::Index Index;
typedef typename internal::add_const_on_value_type_if_arithmetic<
typename internal::packet_traits<Scalar>::type
>::type PacketReturnType;
typedef SparseMatrixBase StorageBaseType;
typedef EigenBase<Derived> Base;
@@ -57,6 +54,8 @@ template<typename Derived> class SparseMatrixBase : public EigenBase<Derived>
other.derived().evalTo(derived());
return derived();
}
// using Base::operator=;
enum {
@@ -108,6 +107,15 @@ template<typename Derived> class SparseMatrixBase : public EigenBase<Derived>
#endif
};
/* \internal the return type of MatrixBase::conjugate() */
// typedef typename internal::conditional<NumTraits<Scalar>::IsComplex,
// const SparseCwiseUnaryOp<internal::scalar_conjugate_op<Scalar>, Derived>,
// const Derived&
// >::type ConjugateReturnType;
/* \internal the return type of MatrixBase::real() */
// typedef SparseCwiseUnaryOp<internal::scalar_real_op<Scalar>, Derived> RealReturnType;
/* \internal the return type of MatrixBase::imag() */
// typedef SparseCwiseUnaryOp<internal::scalar_imag_op<Scalar>, Derived> ImagReturnType;
/** \internal the return type of MatrixBase::adjoint() */
typedef typename internal::conditional<NumTraits<Scalar>::IsComplex,
CwiseUnaryOp<internal::scalar_conjugate_op<Scalar>, Eigen::Transpose<const Derived> >,
@@ -154,12 +162,12 @@ template<typename Derived> class SparseMatrixBase : public EigenBase<Derived>
{ return *static_cast<Derived*>(const_cast<SparseMatrixBase*>(this)); }
#endif // not EIGEN_PARSED_BY_DOXYGEN
/** \returns the number of rows. \sa cols() */
/** \returns the number of rows. \sa cols(), RowsAtCompileTime */
inline Index rows() const { return derived().rows(); }
/** \returns the number of columns. \sa rows() */
/** \returns the number of columns. \sa rows(), ColsAtCompileTime*/
inline Index cols() const { return derived().cols(); }
/** \returns the number of coefficients, which is \a rows()*cols().
* \sa rows(), cols(). */
* \sa rows(), cols(), SizeAtCompileTime. */
inline Index size() const { return rows() * cols(); }
/** \returns the number of nonzero coefficients which is in practice the number
* of stored coefficients. */
@@ -180,7 +188,16 @@ template<typename Derived> class SparseMatrixBase : public EigenBase<Derived>
Derived& markAsRValue() { m_isRValue = true; return derived(); }
SparseMatrixBase() : m_isRValue(false) { /* TODO check flags */ }
inline Derived& operator=(const Derived& other)
{
// std::cout << "Derived& operator=(const Derived& other)\n";
// if (other.isRValue())
// derived().swap(other.const_cast_derived());
// else
this->operator=<Derived>(other);
return derived();
}
template<typename OtherDerived>
Derived& operator=(const ReturnByValue<OtherDerived>& other)
@@ -190,55 +207,10 @@ template<typename Derived> class SparseMatrixBase : public EigenBase<Derived>
}
template<typename OtherDerived>
inline Derived& operator=(const SparseMatrixBase<OtherDerived>& other)
{
return assign(other.derived());
}
inline Derived& operator=(const Derived& other)
{
// if (other.isRValue())
// derived().swap(other.const_cast_derived());
// else
return assign(other.derived());
}
protected:
template<typename OtherDerived>
inline Derived& assign(const OtherDerived& other)
{
const bool transpose = (Flags & RowMajorBit) != (OtherDerived::Flags & RowMajorBit);
const Index outerSize = (int(OtherDerived::Flags) & RowMajorBit) ? other.rows() : other.cols();
if ((!transpose) && other.isRValue())
{
// eval without temporary
derived().resize(other.rows(), other.cols());
derived().setZero();
derived().reserve((std::max)(this->rows(),this->cols())*2);
for (Index j=0; j<outerSize; ++j)
{
derived().startVec(j);
for (typename OtherDerived::InnerIterator it(other, j); it; ++it)
{
Scalar v = it.value();
if (v!=Scalar(0))
derived().insertBackByOuterInner(j,it.index()) = v;
}
}
derived().finalize();
}
else
{
assignGeneric(other);
}
return derived();
}
template<typename OtherDerived>
inline void assignGeneric(const OtherDerived& other)
{
// std::cout << "Derived& operator=(const MatrixBase<OtherDerived>& other)\n";
//const bool transpose = (Flags & RowMajorBit) != (OtherDerived::Flags & RowMajorBit);
eigen_assert(( ((internal::traits<Derived>::SupportedAccessPatterns&OuterRandomAccessPattern)==OuterRandomAccessPattern) ||
(!((Flags & RowMajorBit) != (OtherDerived::Flags & RowMajorBit)))) &&
@@ -267,11 +239,46 @@ template<typename Derived> class SparseMatrixBase : public EigenBase<Derived>
derived() = temp.markAsRValue();
}
public:
template<typename OtherDerived>
inline Derived& operator=(const SparseMatrixBase<OtherDerived>& other)
{
// std::cout << typeid(OtherDerived).name() << "\n";
// std::cout << Flags << " " << OtherDerived::Flags << "\n";
const bool transpose = (Flags & RowMajorBit) != (OtherDerived::Flags & RowMajorBit);
// std::cout << "eval transpose = " << transpose << "\n";
const Index outerSize = (int(OtherDerived::Flags) & RowMajorBit) ? other.rows() : other.cols();
if ((!transpose) && other.isRValue())
{
// eval without temporary
derived().resize(other.rows(), other.cols());
derived().setZero();
derived().reserve((std::max)(this->rows(),this->cols())*2);
for (Index j=0; j<outerSize; ++j)
{
derived().startVec(j);
for (typename OtherDerived::InnerIterator it(other.derived(), j); it; ++it)
{
Scalar v = it.value();
if (v!=Scalar(0))
derived().insertBackByOuterInner(j,it.index()) = v;
}
}
derived().finalize();
}
else
{
assignGeneric(other.derived());
}
return derived();
}
template<typename Lhs, typename Rhs>
inline Derived& operator=(const SparseSparseProduct<Lhs,Rhs>& product);
template<typename Lhs, typename Rhs>
inline void _experimentalNewProduct(const Lhs& lhs, const Rhs& rhs);
friend std::ostream & operator << (std::ostream & s, const SparseMatrixBase& m)
{
if (Flags&RowMajorBit)
@@ -314,11 +321,24 @@ template<typename Derived> class SparseMatrixBase : public EigenBase<Derived>
return s;
}
// const SparseCwiseUnaryOp<internal::scalar_opposite_op<typename internal::traits<Derived>::Scalar>,Derived> operator-() const;
// template<typename OtherDerived>
// const CwiseBinaryOp<internal::scalar_sum_op<typename internal::traits<Derived>::Scalar>, Derived, OtherDerived>
// operator+(const SparseMatrixBase<OtherDerived> &other) const;
// template<typename OtherDerived>
// const CwiseBinaryOp<internal::scalar_difference_op<typename internal::traits<Derived>::Scalar>, Derived, OtherDerived>
// operator-(const SparseMatrixBase<OtherDerived> &other) const;
template<typename OtherDerived>
Derived& operator+=(const SparseMatrixBase<OtherDerived>& other);
template<typename OtherDerived>
Derived& operator-=(const SparseMatrixBase<OtherDerived>& other);
// template<typename Lhs,typename Rhs>
// Derived& operator+=(const Flagged<Product<Lhs,Rhs,CacheFriendlyProduct>, 0, EvalBeforeNestingBit | EvalBeforeAssigningBit>& other);
Derived& operator*=(const Scalar& other);
Derived& operator/=(const Scalar& other);
@@ -338,6 +358,16 @@ template<typename Derived> class SparseMatrixBase : public EigenBase<Derived>
EIGEN_STRONG_INLINE const EIGEN_SPARSE_CWISE_PRODUCT_RETURN_TYPE
cwiseProduct(const MatrixBase<OtherDerived> &other) const;
// const SparseCwiseUnaryOp<internal::scalar_multiple_op<typename internal::traits<Derived>::Scalar>, Derived>
// operator*(const Scalar& scalar) const;
// const SparseCwiseUnaryOp<internal::scalar_quotient1_op<typename internal::traits<Derived>::Scalar>, Derived>
// operator/(const Scalar& scalar) const;
// inline friend const SparseCwiseUnaryOp<internal::scalar_multiple_op<typename internal::traits<Derived>::Scalar>, Derived>
// operator*(const Scalar& scalar, const SparseMatrixBase& matrix)
// { return matrix*scalar; }
// sparse * sparse
template<typename OtherDerived>
const typename SparseSparseProductReturnType<Derived,OtherDerived>::Type
@@ -377,6 +407,8 @@ template<typename Derived> class SparseMatrixBase : public EigenBase<Derived>
// deprecated
template<typename OtherDerived>
void solveTriangularInPlace(MatrixBase<OtherDerived>& other) const;
// template<typename OtherDerived>
// void solveTriangularInPlace(SparseMatrixBase<OtherDerived>& other) const;
#endif // EIGEN2_SUPPORT
template<int Mode>
@@ -389,9 +421,12 @@ template<typename Derived> class SparseMatrixBase : public EigenBase<Derived>
template<typename OtherDerived> Scalar dot(const SparseMatrixBase<OtherDerived>& other) const;
RealScalar squaredNorm() const;
RealScalar norm() const;
// const PlainObject normalized() const;
// void normalize();
Transpose<Derived> transpose() { return derived(); }
const Transpose<const Derived> transpose() const { return derived(); }
// void transposeInPlace();
const AdjointReturnType adjoint() const { return transpose(); }
// sub-vector
@@ -407,14 +442,77 @@ template<typename Derived> class SparseMatrixBase : public EigenBase<Derived>
const SparseInnerVectorSet<Derived,Dynamic> subrows(Index start, Index size) const;
SparseInnerVectorSet<Derived,Dynamic> subcols(Index start, Index size);
const SparseInnerVectorSet<Derived,Dynamic> subcols(Index start, Index size) const;
SparseInnerVectorSet<Derived,Dynamic> middleRows(Index start, Index size);
const SparseInnerVectorSet<Derived,Dynamic> middleRows(Index start, Index size) const;
SparseInnerVectorSet<Derived,Dynamic> middleCols(Index start, Index size);
const SparseInnerVectorSet<Derived,Dynamic> middleCols(Index start, Index size) const;
SparseInnerVectorSet<Derived,Dynamic> innerVectors(Index outerStart, Index outerSize);
const SparseInnerVectorSet<Derived,Dynamic> innerVectors(Index outerStart, Index outerSize) const;
// typename BlockReturnType<Derived>::Type block(int startRow, int startCol, int blockRows, int blockCols);
// const typename BlockReturnType<Derived>::Type
// block(int startRow, int startCol, int blockRows, int blockCols) const;
//
// typename BlockReturnType<Derived>::SubVectorType segment(int start, int size);
// const typename BlockReturnType<Derived>::SubVectorType segment(int start, int size) const;
//
// typename BlockReturnType<Derived,Dynamic>::SubVectorType start(int size);
// const typename BlockReturnType<Derived,Dynamic>::SubVectorType start(int size) const;
//
// typename BlockReturnType<Derived,Dynamic>::SubVectorType end(int size);
// const typename BlockReturnType<Derived,Dynamic>::SubVectorType end(int size) const;
//
// template<int BlockRows, int BlockCols>
// typename BlockReturnType<Derived, BlockRows, BlockCols>::Type block(int startRow, int startCol);
// template<int BlockRows, int BlockCols>
// const typename BlockReturnType<Derived, BlockRows, BlockCols>::Type block(int startRow, int startCol) const;
// template<int Size> typename BlockReturnType<Derived,Size>::SubVectorType start(void);
// template<int Size> const typename BlockReturnType<Derived,Size>::SubVectorType start() const;
// template<int Size> typename BlockReturnType<Derived,Size>::SubVectorType end();
// template<int Size> const typename BlockReturnType<Derived,Size>::SubVectorType end() const;
// template<int Size> typename BlockReturnType<Derived,Size>::SubVectorType segment(int start);
// template<int Size> const typename BlockReturnType<Derived,Size>::SubVectorType segment(int start) const;
// Diagonal<Derived> diagonal();
// const Diagonal<Derived> diagonal() const;
// template<unsigned int Mode> Part<Derived, Mode> part();
// template<unsigned int Mode> const Part<Derived, Mode> part() const;
// static const ConstantReturnType Constant(int rows, int cols, const Scalar& value);
// static const ConstantReturnType Constant(int size, const Scalar& value);
// static const ConstantReturnType Constant(const Scalar& value);
// template<typename CustomNullaryOp>
// static const CwiseNullaryOp<CustomNullaryOp, Derived> NullaryExpr(int rows, int cols, const CustomNullaryOp& func);
// template<typename CustomNullaryOp>
// static const CwiseNullaryOp<CustomNullaryOp, Derived> NullaryExpr(int size, const CustomNullaryOp& func);
// template<typename CustomNullaryOp>
// static const CwiseNullaryOp<CustomNullaryOp, Derived> NullaryExpr(const CustomNullaryOp& func);
// static const ConstantReturnType Zero(int rows, int cols);
// static const ConstantReturnType Zero(int size);
// static const ConstantReturnType Zero();
// static const ConstantReturnType Ones(int rows, int cols);
// static const ConstantReturnType Ones(int size);
// static const ConstantReturnType Ones();
// static const IdentityReturnType Identity();
// static const IdentityReturnType Identity(int rows, int cols);
// static const BasisReturnType Unit(int size, int i);
// static const BasisReturnType Unit(int i);
// static const BasisReturnType UnitX();
// static const BasisReturnType UnitY();
// static const BasisReturnType UnitZ();
// static const BasisReturnType UnitW();
// const DiagonalMatrix<Derived> asDiagonal() const;
// Derived& setConstant(const Scalar& value);
// Derived& setZero();
// Derived& setOnes();
// Derived& setRandom();
// Derived& setIdentity();
/** \internal use operator= */
template<typename DenseDerived>
void evalTo(MatrixBase<DenseDerived>& dst) const
@@ -439,6 +537,37 @@ template<typename Derived> class SparseMatrixBase : public EigenBase<Derived>
bool isApprox(const MatrixBase<OtherDerived>& other,
RealScalar prec = NumTraits<Scalar>::dummy_precision()) const
{ return toDense().isApprox(other,prec); }
// bool isMuchSmallerThan(const RealScalar& other,
// RealScalar prec = NumTraits<Scalar>::dummy_precision()) const;
// template<typename OtherDerived>
// bool isMuchSmallerThan(const MatrixBase<OtherDerived>& other,
// RealScalar prec = NumTraits<Scalar>::dummy_precision()) const;
// bool isApproxToConstant(const Scalar& value, RealScalar prec = NumTraits<Scalar>::dummy_precision()) const;
// bool isZero(RealScalar prec = NumTraits<Scalar>::dummy_precision()) const;
// bool isOnes(RealScalar prec = NumTraits<Scalar>::dummy_precision()) const;
// bool isIdentity(RealScalar prec = NumTraits<Scalar>::dummy_precision()) const;
// bool isDiagonal(RealScalar prec = NumTraits<Scalar>::dummy_precision()) const;
// bool isUpper(RealScalar prec = NumTraits<Scalar>::dummy_precision()) const;
// bool isLower(RealScalar prec = NumTraits<Scalar>::dummy_precision()) const;
// template<typename OtherDerived>
// bool isOrthogonal(const MatrixBase<OtherDerived>& other,
// RealScalar prec = NumTraits<Scalar>::dummy_precision()) const;
// bool isUnitary(RealScalar prec = NumTraits<Scalar>::dummy_precision()) const;
// template<typename OtherDerived>
// inline bool operator==(const MatrixBase<OtherDerived>& other) const
// { return (cwise() == other).all(); }
// template<typename OtherDerived>
// inline bool operator!=(const MatrixBase<OtherDerived>& other) const
// { return (cwise() != other).any(); }
// template<typename NewType>
// const SparseCwiseUnaryOp<internal::scalar_cast_op<typename internal::traits<Derived>::Scalar, NewType>, Derived> cast() const;
/** \returns the matrix or vector obtained by evaluating this expression.
*
@@ -448,7 +577,126 @@ template<typename Derived> class SparseMatrixBase : public EigenBase<Derived>
inline const typename internal::eval<Derived>::type eval() const
{ return typename internal::eval<Derived>::type(derived()); }
// template<typename OtherDerived>
// void swap(MatrixBase<OtherDerived> const & other);
// template<unsigned int Added>
// const SparseFlagged<Derived, Added, 0> marked() const;
// const Flagged<Derived, 0, EvalBeforeNestingBit | EvalBeforeAssigningBit> lazy() const;
/** \returns number of elements to skip to pass from one row (resp. column) to another
* for a row-major (resp. column-major) matrix.
* Combined with coeffRef() and the \ref flags flags, it allows a direct access to the data
* of the underlying matrix.
*/
// inline int stride(void) const { return derived().stride(); }
// FIXME
// ConjugateReturnType conjugate() const;
// const RealReturnType real() const;
// const ImagReturnType imag() const;
// template<typename CustomUnaryOp>
// const SparseCwiseUnaryOp<CustomUnaryOp, Derived> unaryExpr(const CustomUnaryOp& func = CustomUnaryOp()) const;
// template<typename CustomBinaryOp, typename OtherDerived>
// const CwiseBinaryOp<CustomBinaryOp, Derived, OtherDerived>
// binaryExpr(const MatrixBase<OtherDerived> &other, const CustomBinaryOp& func = CustomBinaryOp()) const;
Scalar sum() const;
// Scalar trace() const;
// typename internal::traits<Derived>::Scalar minCoeff() const;
// typename internal::traits<Derived>::Scalar maxCoeff() const;
// typename internal::traits<Derived>::Scalar minCoeff(int* row, int* col = 0) const;
// typename internal::traits<Derived>::Scalar maxCoeff(int* row, int* col = 0) const;
// template<typename BinaryOp>
// typename internal::result_of<BinaryOp(typename internal::traits<Derived>::Scalar)>::type
// redux(const BinaryOp& func) const;
// template<typename Visitor>
// void visit(Visitor& func) const;
// const SparseCwise<Derived> cwise() const;
// SparseCwise<Derived> cwise();
// inline const WithFormat<Derived> format(const IOFormat& fmt) const;
/////////// Array module ///////////
/*
bool all(void) const;
bool any(void) const;
const VectorwiseOp<Derived,Horizontal> rowwise() const;
const VectorwiseOp<Derived,Vertical> colwise() const;
static const CwiseNullaryOp<internal::scalar_random_op<Scalar>,Derived> Random(int rows, int cols);
static const CwiseNullaryOp<internal::scalar_random_op<Scalar>,Derived> Random(int size);
static const CwiseNullaryOp<internal::scalar_random_op<Scalar>,Derived> Random();
template<typename ThenDerived,typename ElseDerived>
const Select<Derived,ThenDerived,ElseDerived>
select(const MatrixBase<ThenDerived>& thenMatrix,
const MatrixBase<ElseDerived>& elseMatrix) const;
template<typename ThenDerived>
inline const Select<Derived,ThenDerived, typename ThenDerived::ConstantReturnType>
select(const MatrixBase<ThenDerived>& thenMatrix, typename ThenDerived::Scalar elseScalar) const;
template<typename ElseDerived>
inline const Select<Derived, typename ElseDerived::ConstantReturnType, ElseDerived >
select(typename ElseDerived::Scalar thenScalar, const MatrixBase<ElseDerived>& elseMatrix) const;
template<int p> RealScalar lpNorm() const;
*/
// template<typename OtherDerived>
// Scalar dot(const MatrixBase<OtherDerived>& other) const
// {
// EIGEN_STATIC_ASSERT_VECTOR_ONLY(Derived)
// EIGEN_STATIC_ASSERT_VECTOR_ONLY(OtherDerived)
// EIGEN_STATIC_ASSERT((internal::is_same<Scalar, typename OtherDerived::Scalar>::value),
// YOU_MIXED_DIFFERENT_NUMERIC_TYPES__YOU_NEED_TO_USE_THE_CAST_METHOD_OF_MATRIXBASE_TO_CAST_NUMERIC_TYPES_EXPLICITLY)
//
// eigen_assert(derived().size() == other.size());
// // short version, but the assembly looks more complicated because
// // of the CwiseBinaryOp iterator complexity
// // return res = (derived().cwise() * other.derived().conjugate()).sum();
//
// // optimized, generic version
// typename Derived::InnerIterator i(derived(),0);
// typename OtherDerived::InnerIterator j(other.derived(),0);
// Scalar res = 0;
// while (i && j)
// {
// if (i.index()==j.index())
// {
// // std::cerr << i.value() << " * " << j.value() << "\n";
// res += i.value() * internal::conj(j.value());
// ++i; ++j;
// }
// else if (i.index()<j.index())
// ++i;
// else
// ++j;
// }
// return res;
// }
//
// Scalar sum() const
// {
// Scalar res = 0;
// for (typename Derived::InnerIterator iter(*this,0); iter; ++iter)
// {
// res += iter.value();
// }
// return res;
// }
protected:

View File

@@ -106,42 +106,9 @@ class SparseSparseProduct : internal::no_assignment_operator,
template<typename Lhs, typename Rhs>
EIGEN_STRONG_INLINE SparseSparseProduct(const Lhs& lhs, const Rhs& rhs)
: m_lhs(lhs), m_rhs(rhs), m_tolerance(0), m_conservative(true)
: m_lhs(lhs), m_rhs(rhs)
{
init();
}
template<typename Lhs, typename Rhs>
EIGEN_STRONG_INLINE SparseSparseProduct(const Lhs& lhs, const Rhs& rhs, RealScalar tolerance)
: m_lhs(lhs), m_rhs(rhs), m_tolerance(tolerance), m_conservative(false)
{
init();
}
SparseSparseProduct pruned(Scalar reference = 0, RealScalar epsilon = NumTraits<RealScalar>::dummy_precision()) const
{
return SparseSparseProduct(m_lhs,m_rhs,internal::abs(reference)*epsilon);
}
template<typename Dest>
void evalTo(Dest& result) const
{
if(m_conservative)
internal::conservative_sparse_sparse_product_selector<_LhsNested, _RhsNested, Dest>::run(lhs(),rhs(),result);
else
internal::sparse_sparse_product_with_pruning_selector<_LhsNested, _RhsNested, Dest>::run(lhs(),rhs(),result,m_tolerance);
}
EIGEN_STRONG_INLINE Index rows() const { return m_lhs.rows(); }
EIGEN_STRONG_INLINE Index cols() const { return m_rhs.cols(); }
EIGEN_STRONG_INLINE const _LhsNested& lhs() const { return m_lhs; }
EIGEN_STRONG_INLINE const _RhsNested& rhs() const { return m_rhs; }
protected:
void init()
{
eigen_assert(m_lhs.cols() == m_rhs.rows());
eigen_assert(lhs.cols() == rhs.rows());
enum {
ProductIsValid = _LhsNested::ColsAtCompileTime==Dynamic
@@ -160,38 +127,15 @@ class SparseSparseProduct : internal::no_assignment_operator,
EIGEN_STATIC_ASSERT(ProductIsValid || SameSizes, INVALID_MATRIX_PRODUCT)
}
EIGEN_STRONG_INLINE Index rows() const { return m_lhs.rows(); }
EIGEN_STRONG_INLINE Index cols() const { return m_rhs.cols(); }
EIGEN_STRONG_INLINE const _LhsNested& lhs() const { return m_lhs; }
EIGEN_STRONG_INLINE const _RhsNested& rhs() const { return m_rhs; }
protected:
LhsNested m_lhs;
RhsNested m_rhs;
RealScalar m_tolerance;
bool m_conservative;
};
// sparse = sparse * sparse
template<typename Derived>
template<typename Lhs, typename Rhs>
inline Derived& SparseMatrixBase<Derived>::operator=(const SparseSparseProduct<Lhs,Rhs>& product)
{
product.evalTo(derived());
return derived();
}
/** \returns an expression of the product of two sparse matrices.
* By default a conservative product preserving the symbolic non zeros is performed.
* The automatic pruning of the small values can be achieved by calling the pruned() function
* in which case a totally different product algorithm is employed:
* \code
* C = (A*B).pruned(); // supress numerical zeros (exact)
* C = (A*B).pruned(ref);
* C = (A*B).pruned(ref,epsilon);
* \endcode
* where \c ref is a meaningful non zero reference value.
* */
template<typename Derived>
template<typename OtherDerived>
inline const typename SparseSparseProductReturnType<Derived,OtherDerived>::Type
SparseMatrixBase<Derived>::operator*(const SparseMatrixBase<OtherDerived> &other) const
{
return typename SparseSparseProductReturnType<Derived,OtherDerived>::Type(derived(), other.derived());
}
#endif // EIGEN_SPARSEPRODUCT_H

View File

@@ -25,8 +25,8 @@
#ifndef EIGEN_SPARSE_SELFADJOINTVIEW_H
#define EIGEN_SPARSE_SELFADJOINTVIEW_H
/** \ingroup SparseCore_Module
* \class SparseSelfAdjointView
/** \class SparseSelfAdjointView
*
*
* \brief Pseudo expression to manipulate a triangular sparse matrix as a selfadjoint matrix.
*
@@ -106,6 +106,9 @@ template<typename MatrixType, unsigned int UpLo> class SparseSelfAdjointView
*
* \returns a reference to \c *this
*
* Note that it is faster to set alpha=0 than initializing the matrix to zero
* and then keep the default value alpha=1.
*
* To perform \f$ this = this + \alpha ( u^* u ) \f$ you can simply
* call this function with u.adjoint().
*/
@@ -113,21 +116,21 @@ template<typename MatrixType, unsigned int UpLo> class SparseSelfAdjointView
SparseSelfAdjointView& rankUpdate(const SparseMatrixBase<DerivedU>& u, Scalar alpha = Scalar(1));
/** \internal triggered by sparse_matrix = SparseSelfadjointView; */
template<typename DestScalar,int StorageOrder> void evalTo(SparseMatrix<DestScalar,StorageOrder,Index>& _dest) const
template<typename DestScalar> void evalTo(SparseMatrix<DestScalar>& _dest) const
{
internal::permute_symm_to_fullsymm<UpLo>(m_matrix, _dest);
}
template<typename DestScalar> void evalTo(DynamicSparseMatrix<DestScalar,ColMajor,Index>& _dest) const
template<typename DestScalar> void evalTo(DynamicSparseMatrix<DestScalar>& _dest) const
{
// TODO directly evaluate into _dest;
SparseMatrix<DestScalar,ColMajor,Index> tmp(_dest.rows(),_dest.cols());
SparseMatrix<DestScalar> tmp(_dest.rows(),_dest.cols());
internal::permute_symm_to_fullsymm<UpLo>(m_matrix, tmp);
_dest = tmp;
}
/** \returns an expression of P^-1 H P */
SparseSymmetricPermutationProduct<_MatrixTypeNested,UpLo> twistedBy(const PermutationMatrix<Dynamic,Dynamic,Index>& perm) const
SparseSymmetricPermutationProduct<_MatrixTypeNested,UpLo> twistedBy(const PermutationMatrix<Dynamic>& perm) const
{
return SparseSymmetricPermutationProduct<_MatrixTypeNested,UpLo>(m_matrix, perm);
}
@@ -227,15 +230,12 @@ class SparseSelfAdjointTimeDenseProduct
for (Index j=0; j<m_lhs.outerSize(); ++j)
{
LhsInnerIterator i(m_lhs,j);
if (ProcessSecondHalf)
if (ProcessSecondHalf && i && (i.index()==j))
{
while (i && i.index()<j) ++i;
if(i && i.index()==j)
{
dest.row(j) += i.value() * m_rhs.row(j);
++i;
}
dest.row(j) += i.value() * m_rhs.row(j);
++i;
}
Block<Dest,1,Dest::ColsAtCompileTime> dest_j(dest.row(LhsIsRowMajor ? j : 0));
for(; (ProcessFirstHalf ? i && i.index() < j : i) ; ++i)
{
Index a = LhsIsRowMajor ? j : i.index();
@@ -300,7 +300,7 @@ void permute_symm_to_fullsymm(const MatrixType& mat, SparseMatrix<typename Matri
enum {
StorageOrderMatch = int(Dest::IsRowMajor) == int(MatrixType::IsRowMajor)
};
eigen_assert(perm==0);
Index size = mat.rows();
VectorI count;
count.resize(size);
@@ -326,11 +326,11 @@ void permute_symm_to_fullsymm(const MatrixType& mat, SparseMatrix<typename Matri
// reserve space
dest.reserve(nnz);
dest.outerIndexPtr()[0] = 0;
dest._outerIndexPtr()[0] = 0;
for(Index j=0; j<size; ++j)
dest.outerIndexPtr()[j+1] = dest.outerIndexPtr()[j] + count[j];
dest._outerIndexPtr()[j+1] = dest._outerIndexPtr()[j] + count[j];
for(Index j=0; j<size; ++j)
count[j] = dest.outerIndexPtr()[j];
count[j] = dest._outerIndexPtr()[j];
// copy data
for(Index j = 0; j<size; ++j)
@@ -343,17 +343,17 @@ void permute_symm_to_fullsymm(const MatrixType& mat, SparseMatrix<typename Matri
if(i==j)
{
int k = count[ip]++;
dest.innerIndexPtr()[k] = ip;
dest.valuePtr()[k] = it.value();
dest._innerIndexPtr()[k] = ip;
dest._valuePtr()[k] = it.value();
}
else if((UpLo==Lower && i>j) || (UpLo==Upper && i<j))
{
int k = count[jp]++;
dest.innerIndexPtr()[k] = ip;
dest.valuePtr()[k] = it.value();
dest._innerIndexPtr()[k] = ip;
dest._valuePtr()[k] = it.value();
k = count[ip]++;
dest.innerIndexPtr()[k] = jp;
dest.valuePtr()[k] = internal::conj(it.value());
dest._innerIndexPtr()[k] = jp;
dest._valuePtr()[k] = internal::conj(it.value());
}
}
}
@@ -386,12 +386,12 @@ void permute_symm_to_symm(const MatrixType& mat, SparseMatrix<typename MatrixTyp
count[DstUpLo==Lower ? (std::min)(ip,jp) : (std::max)(ip,jp)]++;
}
}
dest.outerIndexPtr()[0] = 0;
dest._outerIndexPtr()[0] = 0;
for(Index j=0; j<size; ++j)
dest.outerIndexPtr()[j+1] = dest.outerIndexPtr()[j] + count[j];
dest.resizeNonZeros(dest.outerIndexPtr()[size]);
dest._outerIndexPtr()[j+1] = dest._outerIndexPtr()[j] + count[j];
dest.resizeNonZeros(dest._outerIndexPtr()[size]);
for(Index j=0; j<size; ++j)
count[j] = dest.outerIndexPtr()[j];
count[j] = dest._outerIndexPtr()[j];
for(Index j = 0; j<size; ++j)
{
@@ -404,12 +404,12 @@ void permute_symm_to_symm(const MatrixType& mat, SparseMatrix<typename MatrixTyp
Index ip = perm? perm[i] : i;
Index k = count[DstUpLo==Lower ? (std::min)(ip,jp) : (std::max)(ip,jp)]++;
dest.innerIndexPtr()[k] = DstUpLo==Lower ? (std::max)(ip,jp) : (std::min)(ip,jp);
dest._innerIndexPtr()[k] = DstUpLo==Lower ? (std::max)(ip,jp) : (std::min)(ip,jp);
if((DstUpLo==Lower && ip<jp) || (DstUpLo==Upper && ip>jp))
dest.valuePtr()[k] = conj(it.value());
dest._valuePtr()[k] = conj(it.value());
else
dest.valuePtr()[k] = it.value();
dest._valuePtr()[k] = it.value();
}
}
}
@@ -420,12 +420,10 @@ template<typename MatrixType,int UpLo>
class SparseSymmetricPermutationProduct
: public EigenBase<SparseSymmetricPermutationProduct<MatrixType,UpLo> >
{
typedef PermutationMatrix<Dynamic> Perm;
public:
typedef typename MatrixType::Scalar Scalar;
typedef typename MatrixType::Index Index;
protected:
typedef PermutationMatrix<Dynamic,Dynamic,Index> Perm;
public:
typedef Matrix<Index,Dynamic,1> VectorI;
typedef typename MatrixType::Nested MatrixTypeNested;
typedef typename internal::remove_all<MatrixTypeNested>::type _MatrixTypeNested;

View File

@@ -0,0 +1,401 @@
// This file is part of Eigen, a lightweight C++ template library
// for linear algebra.
//
// Copyright (C) 2008-2010 Gael Guennebaud <gael.guennebaud@inria.fr>
//
// Eigen is free software; you can redistribute it and/or
// modify it under the terms of the GNU Lesser General Public
// License as published by the Free Software Foundation; either
// version 3 of the License, or (at your option) any later version.
//
// Alternatively, you can redistribute it and/or
// modify it under the terms of the GNU General Public License as
// published by the Free Software Foundation; either version 2 of
// the License, or (at your option) any later version.
//
// Eigen is distributed in the hope that it will be useful, but WITHOUT ANY
// WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS
// FOR A PARTICULAR PURPOSE. See the GNU Lesser General Public License or the
// GNU General Public License for more details.
//
// You should have received a copy of the GNU Lesser General Public
// License and a copy of the GNU General Public License along with
// Eigen. If not, see <http://www.gnu.org/licenses/>.
#ifndef EIGEN_SPARSESPARSEPRODUCT_H
#define EIGEN_SPARSESPARSEPRODUCT_H
namespace internal {
template<typename Lhs, typename Rhs, typename ResultType>
static void sparse_product_impl2(const Lhs& lhs, const Rhs& rhs, ResultType& res)
{
typedef typename remove_all<Lhs>::type::Scalar Scalar;
typedef typename remove_all<Lhs>::type::Index Index;
// make sure to call innerSize/outerSize since we fake the storage order.
Index rows = lhs.innerSize();
Index cols = rhs.outerSize();
eigen_assert(lhs.outerSize() == rhs.innerSize());
std::vector<bool> mask(rows,false);
Matrix<Scalar,Dynamic,1> values(rows);
Matrix<Index,Dynamic,1> indices(rows);
// estimate the number of non zero entries
float ratioLhs = float(lhs.nonZeros())/(float(lhs.rows())*float(lhs.cols()));
float avgNnzPerRhsColumn = float(rhs.nonZeros())/float(cols);
float ratioRes = (std::min)(ratioLhs * avgNnzPerRhsColumn, 1.f);
// int t200 = rows/(log2(200)*1.39);
// int t = (rows*100)/139;
res.resize(rows, cols);
res.reserve(Index(ratioRes*rows*cols));
// we compute each column of the result, one after the other
for (Index j=0; j<cols; ++j)
{
res.startVec(j);
Index nnz = 0;
for (typename Rhs::InnerIterator rhsIt(rhs, j); rhsIt; ++rhsIt)
{
Scalar y = rhsIt.value();
Index k = rhsIt.index();
for (typename Lhs::InnerIterator lhsIt(lhs, k); lhsIt; ++lhsIt)
{
Index i = lhsIt.index();
Scalar x = lhsIt.value();
if(!mask[i])
{
mask[i] = true;
// values[i] = x * y;
// indices[nnz] = i;
++nnz;
}
else
values[i] += x * y;
}
}
// FIXME reserve nnz non zeros
// FIXME implement fast sort algorithms for very small nnz
// if the result is sparse enough => use a quick sort
// otherwise => loop through the entire vector
// In order to avoid to perform an expensive log2 when the
// result is clearly very sparse we use a linear bound up to 200.
// if((nnz<200 && nnz<t200) || nnz * log2(nnz) < t)
// {
// if(nnz>1) std::sort(indices.data(),indices.data()+nnz);
// for(int k=0; k<nnz; ++k)
// {
// int i = indices[k];
// res.insertBackNoCheck(j,i) = values[i];
// mask[i] = false;
// }
// }
// else
// {
// // dense path
// for(int i=0; i<rows; ++i)
// {
// if(mask[i])
// {
// mask[i] = false;
// res.insertBackNoCheck(j,i) = values[i];
// }
// }
// }
}
res.finalize();
}
// perform a pseudo in-place sparse * sparse product assuming all matrices are col major
template<typename Lhs, typename Rhs, typename ResultType>
static void sparse_product_impl(const Lhs& lhs, const Rhs& rhs, ResultType& res)
{
// return sparse_product_impl2(lhs,rhs,res);
typedef typename remove_all<Lhs>::type::Scalar Scalar;
typedef typename remove_all<Lhs>::type::Index Index;
// make sure to call innerSize/outerSize since we fake the storage order.
Index rows = lhs.innerSize();
Index cols = rhs.outerSize();
//int size = lhs.outerSize();
eigen_assert(lhs.outerSize() == rhs.innerSize());
// allocate a temporary buffer
AmbiVector<Scalar,Index> tempVector(rows);
// estimate the number of non zero entries
float ratioLhs = float(lhs.nonZeros())/(float(lhs.rows())*float(lhs.cols()));
float avgNnzPerRhsColumn = float(rhs.nonZeros())/float(cols);
float ratioRes = (std::min)(ratioLhs * avgNnzPerRhsColumn, 1.f);
// mimics a resizeByInnerOuter:
if(ResultType::IsRowMajor)
res.resize(cols, rows);
else
res.resize(rows, cols);
res.reserve(Index(ratioRes*rows*cols));
for (Index j=0; j<cols; ++j)
{
// let's do a more accurate determination of the nnz ratio for the current column j of res
//float ratioColRes = (std::min)(ratioLhs * rhs.innerNonZeros(j), 1.f);
// FIXME find a nice way to get the number of nonzeros of a sub matrix (here an inner vector)
float ratioColRes = ratioRes;
tempVector.init(ratioColRes);
tempVector.setZero();
for (typename Rhs::InnerIterator rhsIt(rhs, j); rhsIt; ++rhsIt)
{
// FIXME should be written like this: tmp += rhsIt.value() * lhs.col(rhsIt.index())
tempVector.restart();
Scalar x = rhsIt.value();
for (typename Lhs::InnerIterator lhsIt(lhs, rhsIt.index()); lhsIt; ++lhsIt)
{
tempVector.coeffRef(lhsIt.index()) += lhsIt.value() * x;
}
}
res.startVec(j);
for (typename AmbiVector<Scalar,Index>::Iterator it(tempVector); it; ++it)
res.insertBackByOuterInner(j,it.index()) = it.value();
}
res.finalize();
}
template<typename Lhs, typename Rhs, typename ResultType,
int LhsStorageOrder = traits<Lhs>::Flags&RowMajorBit,
int RhsStorageOrder = traits<Rhs>::Flags&RowMajorBit,
int ResStorageOrder = traits<ResultType>::Flags&RowMajorBit>
struct sparse_product_selector;
template<typename Lhs, typename Rhs, typename ResultType>
struct sparse_product_selector<Lhs,Rhs,ResultType,ColMajor,ColMajor,ColMajor>
{
typedef typename traits<typename remove_all<Lhs>::type>::Scalar Scalar;
static void run(const Lhs& lhs, const Rhs& rhs, ResultType& res)
{
// std::cerr << __LINE__ << "\n";
typename remove_all<ResultType>::type _res(res.rows(), res.cols());
sparse_product_impl<Lhs,Rhs,ResultType>(lhs, rhs, _res);
res.swap(_res);
}
};
template<typename Lhs, typename Rhs, typename ResultType>
struct sparse_product_selector<Lhs,Rhs,ResultType,ColMajor,ColMajor,RowMajor>
{
static void run(const Lhs& lhs, const Rhs& rhs, ResultType& res)
{
// std::cerr << __LINE__ << "\n";
// we need a col-major matrix to hold the result
typedef SparseMatrix<typename ResultType::Scalar> SparseTemporaryType;
SparseTemporaryType _res(res.rows(), res.cols());
sparse_product_impl<Lhs,Rhs,SparseTemporaryType>(lhs, rhs, _res);
res = _res;
}
};
template<typename Lhs, typename Rhs, typename ResultType>
struct sparse_product_selector<Lhs,Rhs,ResultType,RowMajor,RowMajor,RowMajor>
{
static void run(const Lhs& lhs, const Rhs& rhs, ResultType& res)
{
// std::cerr << __LINE__ << "\n";
// let's transpose the product to get a column x column product
typename remove_all<ResultType>::type _res(res.rows(), res.cols());
sparse_product_impl<Rhs,Lhs,ResultType>(rhs, lhs, _res);
res.swap(_res);
}
};
template<typename Lhs, typename Rhs, typename ResultType>
struct sparse_product_selector<Lhs,Rhs,ResultType,RowMajor,RowMajor,ColMajor>
{
static void run(const Lhs& lhs, const Rhs& rhs, ResultType& res)
{
// std::cerr << "here...\n";
typedef SparseMatrix<typename ResultType::Scalar,ColMajor> ColMajorMatrix;
ColMajorMatrix colLhs(lhs);
ColMajorMatrix colRhs(rhs);
// std::cerr << "more...\n";
sparse_product_impl<ColMajorMatrix,ColMajorMatrix,ResultType>(colLhs, colRhs, res);
// std::cerr << "OK.\n";
// let's transpose the product to get a column x column product
// typedef SparseMatrix<typename ResultType::Scalar> SparseTemporaryType;
// SparseTemporaryType _res(res.cols(), res.rows());
// sparse_product_impl<Rhs,Lhs,SparseTemporaryType>(rhs, lhs, _res);
// res = _res.transpose();
}
};
// NOTE the 2 others cases (col row *) must never occur since they are caught
// by ProductReturnType which transforms it to (col col *) by evaluating rhs.
} // end namespace internal
// sparse = sparse * sparse
template<typename Derived>
template<typename Lhs, typename Rhs>
inline Derived& SparseMatrixBase<Derived>::operator=(const SparseSparseProduct<Lhs,Rhs>& product)
{
// std::cerr << "there..." << typeid(Lhs).name() << " " << typeid(Lhs).name() << " " << (Derived::Flags&&RowMajorBit) << "\n";
internal::sparse_product_selector<
typename internal::remove_all<Lhs>::type,
typename internal::remove_all<Rhs>::type,
Derived>::run(product.lhs(),product.rhs(),derived());
return derived();
}
namespace internal {
template<typename Lhs, typename Rhs, typename ResultType,
int LhsStorageOrder = traits<Lhs>::Flags&RowMajorBit,
int RhsStorageOrder = traits<Rhs>::Flags&RowMajorBit,
int ResStorageOrder = traits<ResultType>::Flags&RowMajorBit>
struct sparse_product_selector2;
template<typename Lhs, typename Rhs, typename ResultType>
struct sparse_product_selector2<Lhs,Rhs,ResultType,ColMajor,ColMajor,ColMajor>
{
typedef typename traits<typename remove_all<Lhs>::type>::Scalar Scalar;
static void run(const Lhs& lhs, const Rhs& rhs, ResultType& res)
{
sparse_product_impl2<Lhs,Rhs,ResultType>(lhs, rhs, res);
}
};
template<typename Lhs, typename Rhs, typename ResultType>
struct sparse_product_selector2<Lhs,Rhs,ResultType,RowMajor,ColMajor,ColMajor>
{
static void run(const Lhs& lhs, const Rhs& rhs, ResultType& res)
{
// prevent warnings until the code is fixed
EIGEN_UNUSED_VARIABLE(lhs);
EIGEN_UNUSED_VARIABLE(rhs);
EIGEN_UNUSED_VARIABLE(res);
// typedef SparseMatrix<typename ResultType::Scalar,RowMajor> RowMajorMatrix;
// RowMajorMatrix rhsRow = rhs;
// RowMajorMatrix resRow(res.rows(), res.cols());
// sparse_product_impl2<RowMajorMatrix,Lhs,RowMajorMatrix>(rhsRow, lhs, resRow);
// res = resRow;
}
};
template<typename Lhs, typename Rhs, typename ResultType>
struct sparse_product_selector2<Lhs,Rhs,ResultType,ColMajor,RowMajor,ColMajor>
{
static void run(const Lhs& lhs, const Rhs& rhs, ResultType& res)
{
typedef SparseMatrix<typename ResultType::Scalar,RowMajor> RowMajorMatrix;
RowMajorMatrix lhsRow = lhs;
RowMajorMatrix resRow(res.rows(), res.cols());
sparse_product_impl2<Rhs,RowMajorMatrix,RowMajorMatrix>(rhs, lhsRow, resRow);
res = resRow;
}
};
template<typename Lhs, typename Rhs, typename ResultType>
struct sparse_product_selector2<Lhs,Rhs,ResultType,RowMajor,RowMajor,ColMajor>
{
static void run(const Lhs& lhs, const Rhs& rhs, ResultType& res)
{
typedef SparseMatrix<typename ResultType::Scalar,RowMajor> RowMajorMatrix;
RowMajorMatrix resRow(res.rows(), res.cols());
sparse_product_impl2<Rhs,Lhs,RowMajorMatrix>(rhs, lhs, resRow);
res = resRow;
}
};
template<typename Lhs, typename Rhs, typename ResultType>
struct sparse_product_selector2<Lhs,Rhs,ResultType,ColMajor,ColMajor,RowMajor>
{
typedef typename traits<typename remove_all<Lhs>::type>::Scalar Scalar;
static void run(const Lhs& lhs, const Rhs& rhs, ResultType& res)
{
typedef SparseMatrix<typename ResultType::Scalar,ColMajor> ColMajorMatrix;
ColMajorMatrix resCol(res.rows(), res.cols());
sparse_product_impl2<Lhs,Rhs,ColMajorMatrix>(lhs, rhs, resCol);
res = resCol;
}
};
template<typename Lhs, typename Rhs, typename ResultType>
struct sparse_product_selector2<Lhs,Rhs,ResultType,RowMajor,ColMajor,RowMajor>
{
static void run(const Lhs& lhs, const Rhs& rhs, ResultType& res)
{
typedef SparseMatrix<typename ResultType::Scalar,ColMajor> ColMajorMatrix;
ColMajorMatrix lhsCol = lhs;
ColMajorMatrix resCol(res.rows(), res.cols());
sparse_product_impl2<ColMajorMatrix,Rhs,ColMajorMatrix>(lhsCol, rhs, resCol);
res = resCol;
}
};
template<typename Lhs, typename Rhs, typename ResultType>
struct sparse_product_selector2<Lhs,Rhs,ResultType,ColMajor,RowMajor,RowMajor>
{
static void run(const Lhs& lhs, const Rhs& rhs, ResultType& res)
{
typedef SparseMatrix<typename ResultType::Scalar,ColMajor> ColMajorMatrix;
ColMajorMatrix rhsCol = rhs;
ColMajorMatrix resCol(res.rows(), res.cols());
sparse_product_impl2<Lhs,ColMajorMatrix,ColMajorMatrix>(lhs, rhsCol, resCol);
res = resCol;
}
};
template<typename Lhs, typename Rhs, typename ResultType>
struct sparse_product_selector2<Lhs,Rhs,ResultType,RowMajor,RowMajor,RowMajor>
{
static void run(const Lhs& lhs, const Rhs& rhs, ResultType& res)
{
typedef SparseMatrix<typename ResultType::Scalar,ColMajor> ColMajorMatrix;
// ColMajorMatrix lhsTr(lhs);
// ColMajorMatrix rhsTr(rhs);
// ColMajorMatrix aux(res.rows(), res.cols());
// sparse_product_impl2<Rhs,Lhs,ColMajorMatrix>(rhs, lhs, aux);
// // ColMajorMatrix aux2 = aux.transpose();
// res = aux;
typedef SparseMatrix<typename ResultType::Scalar,ColMajor> ColMajorMatrix;
ColMajorMatrix lhsCol(lhs);
ColMajorMatrix rhsCol(rhs);
ColMajorMatrix resCol(res.rows(), res.cols());
sparse_product_impl2<ColMajorMatrix,ColMajorMatrix,ColMajorMatrix>(lhsCol, rhsCol, resCol);
res = resCol;
}
};
} // end namespace internal
template<typename Derived>
template<typename Lhs, typename Rhs>
inline void SparseMatrixBase<Derived>::_experimentalNewProduct(const Lhs& lhs, const Rhs& rhs)
{
//derived().resize(lhs.rows(), rhs.cols());
internal::sparse_product_selector2<
typename internal::remove_all<Lhs>::type,
typename internal::remove_all<Rhs>::type,
Derived>::run(lhs,rhs,derived());
}
// sparse * sparse
template<typename Derived>
template<typename OtherDerived>
inline const typename SparseSparseProductReturnType<Derived,OtherDerived>::Type
SparseMatrixBase<Derived>::operator*(const SparseMatrixBase<OtherDerived> &other) const
{
return typename SparseSparseProductReturnType<Derived,OtherDerived>::Type(derived(), other.derived());
}
#endif // EIGEN_SPARSESPARSEPRODUCT_H

View File

@@ -0,0 +1,100 @@
// This file is part of Eigen, a lightweight C++ template library
// for linear algebra.
//
// Copyright (C) 2009 Gael Guennebaud <gael.guennebaud@inria.fr>
//
// Eigen is free software; you can redistribute it and/or
// modify it under the terms of the GNU Lesser General Public
// License as published by the Free Software Foundation; either
// version 3 of the License, or (at your option) any later version.
//
// Alternatively, you can redistribute it and/or
// modify it under the terms of the GNU General Public License as
// published by the Free Software Foundation; either version 2 of
// the License, or (at your option) any later version.
//
// Eigen is distributed in the hope that it will be useful, but WITHOUT ANY
// WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS
// FOR A PARTICULAR PURPOSE. See the GNU Lesser General Public License or the
// GNU General Public License for more details.
//
// You should have received a copy of the GNU Lesser General Public
// License and a copy of the GNU General Public License along with
// Eigen. If not, see <http://www.gnu.org/licenses/>.
#ifndef EIGEN_SPARSE_TRIANGULARVIEW_H
#define EIGEN_SPARSE_TRIANGULARVIEW_H
namespace internal {
template<typename MatrixType, int Mode>
struct traits<SparseTriangularView<MatrixType,Mode> >
: public traits<MatrixType>
{};
} // namespace internal
template<typename MatrixType, int Mode> class SparseTriangularView
: public SparseMatrixBase<SparseTriangularView<MatrixType,Mode> >
{
enum { SkipFirst = (Mode==Lower && !(MatrixType::Flags&RowMajorBit))
|| (Mode==Upper && (MatrixType::Flags&RowMajorBit)) };
public:
EIGEN_SPARSE_PUBLIC_INTERFACE(SparseTriangularView)
class InnerIterator;
inline Index rows() const { return m_matrix.rows(); }
inline Index cols() const { return m_matrix.cols(); }
typedef typename internal::conditional<internal::must_nest_by_value<MatrixType>::ret,
MatrixType, const MatrixType&>::type MatrixTypeNested;
inline SparseTriangularView(const MatrixType& matrix) : m_matrix(matrix) {}
/** \internal */
inline const MatrixType& nestedExpression() const { return m_matrix; }
template<typename OtherDerived>
typename internal::plain_matrix_type_column_major<OtherDerived>::type
solve(const MatrixBase<OtherDerived>& other) const;
template<typename OtherDerived> void solveInPlace(MatrixBase<OtherDerived>& other) const;
template<typename OtherDerived> void solveInPlace(SparseMatrixBase<OtherDerived>& other) const;
protected:
MatrixTypeNested m_matrix;
};
template<typename MatrixType, int Mode>
class SparseTriangularView<MatrixType,Mode>::InnerIterator : public MatrixType::InnerIterator
{
typedef typename MatrixType::InnerIterator Base;
public:
EIGEN_STRONG_INLINE InnerIterator(const SparseTriangularView& view, Index outer)
: Base(view.nestedExpression(), outer)
{
if(SkipFirst)
while((*this) && this->index()<outer)
++(*this);
}
inline Index row() const { return Base::row(); }
inline Index col() const { return Base::col(); }
EIGEN_STRONG_INLINE operator bool() const
{
return SkipFirst ? Base::operator bool() : (Base::operator bool() && this->index() <= this->outer());
}
};
template<typename Derived>
template<int Mode>
inline const SparseTriangularView<Derived, Mode>
SparseMatrixBase<Derived>::triangularView() const
{
return derived();
}
#endif // EIGEN_SPARSE_TRIANGULARVIEW_H

View File

@@ -25,8 +25,7 @@
#ifndef EIGEN_SPARSEVECTOR_H
#define EIGEN_SPARSEVECTOR_H
/** \ingroup SparseCore_Module
* \class SparseVector
/** \class SparseVector
*
* \brief a sparse vector class
*
@@ -68,6 +67,7 @@ class SparseVector
EIGEN_SPARSE_PUBLIC_INTERFACE(SparseVector)
EIGEN_SPARSE_INHERIT_ASSIGNMENT_OPERATOR(SparseVector, +=)
EIGEN_SPARSE_INHERIT_ASSIGNMENT_OPERATOR(SparseVector, -=)
// EIGEN_SPARSE_INHERIT_ASSIGNMENT_OPERATOR(SparseVector, =)
protected:
public:
@@ -79,11 +79,11 @@ class SparseVector
Options = _Options
};
internal::CompressedStorage<Scalar,Index> m_data;
CompressedStorage<Scalar,Index> m_data;
Index m_size;
internal::CompressedStorage<Scalar,Index>& _data() { return m_data; }
internal::CompressedStorage<Scalar,Index>& _data() const { return m_data; }
CompressedStorage<Scalar,Index>& _data() { return m_data; }
CompressedStorage<Scalar,Index>& _data() const { return m_data; }
public:
@@ -91,12 +91,13 @@ class SparseVector
EIGEN_STRONG_INLINE Index cols() const { return IsColVector ? 1 : m_size; }
EIGEN_STRONG_INLINE Index innerSize() const { return m_size; }
EIGEN_STRONG_INLINE Index outerSize() const { return 1; }
EIGEN_STRONG_INLINE Index innerNonZeros(Index j) const { eigen_assert(j==0); return m_size; }
EIGEN_STRONG_INLINE const Scalar* valuePtr() const { return &m_data.value(0); }
EIGEN_STRONG_INLINE Scalar* valuePtr() { return &m_data.value(0); }
EIGEN_STRONG_INLINE const Scalar* _valuePtr() const { return &m_data.value(0); }
EIGEN_STRONG_INLINE Scalar* _valuePtr() { return &m_data.value(0); }
EIGEN_STRONG_INLINE const Index* innerIndexPtr() const { return &m_data.index(0); }
EIGEN_STRONG_INLINE Index* innerIndexPtr() { return &m_data.index(0); }
EIGEN_STRONG_INLINE const Index* _innerIndexPtr() const { return &m_data.index(0); }
EIGEN_STRONG_INLINE Index* _innerIndexPtr() { return &m_data.index(0); }
inline Scalar coeff(Index row, Index col) const
{
@@ -125,7 +126,6 @@ class SparseVector
public:
class InnerIterator;
class ReverseInnerIterator;
inline void setZero() { m_data.clear(); }
@@ -205,6 +205,13 @@ class SparseVector
inline SparseVector(Index rows, Index cols) : m_size(0) { resize(rows,cols); }
template<typename OtherDerived>
inline SparseVector(const MatrixBase<OtherDerived>& other)
: m_size(0)
{
*this = other.derived();
}
template<typename OtherDerived>
inline SparseVector(const SparseMatrixBase<OtherDerived>& other)
: m_size(0)
@@ -255,6 +262,56 @@ class SparseVector
}
#endif
// const bool needToTranspose = (Flags & RowMajorBit) != (OtherDerived::Flags & RowMajorBit);
// if (needToTranspose)
// {
// // two passes algorithm:
// // 1 - compute the number of coeffs per dest inner vector
// // 2 - do the actual copy/eval
// // Since each coeff of the rhs has to be evaluated twice, let's evauluate it if needed
// typedef typename internal::nested<OtherDerived,2>::type OtherCopy;
// OtherCopy otherCopy(other.derived());
// typedef typename internal::remove_all<OtherCopy>::type _OtherCopy;
//
// resize(other.rows(), other.cols());
// Eigen::Map<VectorXi>(m_outerIndex,outerSize()).setZero();
// // pass 1
// // FIXME the above copy could be merged with that pass
// for (int j=0; j<otherCopy.outerSize(); ++j)
// for (typename _OtherCopy::InnerIterator it(otherCopy, j); it; ++it)
// ++m_outerIndex[it.index()];
//
// // prefix sum
// int count = 0;
// VectorXi positions(outerSize());
// for (int j=0; j<outerSize(); ++j)
// {
// int tmp = m_outerIndex[j];
// m_outerIndex[j] = count;
// positions[j] = count;
// count += tmp;
// }
// m_outerIndex[outerSize()] = count;
// // alloc
// m_data.resize(count);
// // pass 2
// for (int j=0; j<otherCopy.outerSize(); ++j)
// for (typename _OtherCopy::InnerIterator it(otherCopy, j); it; ++it)
// {
// int pos = positions[it.index()]++;
// m_data.index(pos) = j;
// m_data.value(pos) = it.value();
// }
//
// return *this;
// }
// else
// {
// // there is no special optimization
// return SparseMatrixBase<SparseMatrix>::operator=(other.derived());
// }
// }
friend std::ostream & operator << (std::ostream & s, const SparseVector& m)
{
for (Index i=0; i<m.nonZeros(); ++i)
@@ -263,6 +320,28 @@ class SparseVector
return s;
}
// this specialized version does not seems to be faster
// Scalar dot(const SparseVector& other) const
// {
// int i=0, j=0;
// Scalar res = 0;
// asm("#begindot");
// while (i<nonZeros() && j<other.nonZeros())
// {
// if (m_data.index(i)==other.m_data.index(j))
// {
// res += m_data.value(i) * internal::conj(other.m_data.value(j));
// ++i; ++j;
// }
// else if (m_data.index(i)<other.m_data.index(j))
// ++i;
// else
// ++j;
// }
// asm("#enddot");
// return res;
// }
/** Destructor */
inline ~SparseVector() {}
@@ -323,10 +402,15 @@ class SparseVector<Scalar,_Options,_Index>::InnerIterator
eigen_assert(outer==0);
}
InnerIterator(const internal::CompressedStorage<Scalar,Index>& data)
InnerIterator(const CompressedStorage<Scalar,Index>& data)
: m_data(data), m_id(0), m_end(static_cast<Index>(m_data.size()))
{}
template<unsigned int Added, unsigned int Removed>
InnerIterator(const Flagged<SparseVector,Added,Removed>& vec, Index )
: m_data(vec._expression().m_data), m_id(0), m_end(m_data.size())
{}
inline InnerIterator& operator++() { m_id++; return *this; }
inline Scalar value() const { return m_data.value(m_id); }
@@ -339,40 +423,9 @@ class SparseVector<Scalar,_Options,_Index>::InnerIterator
inline operator bool() const { return (m_id < m_end); }
protected:
const internal::CompressedStorage<Scalar,Index>& m_data;
const CompressedStorage<Scalar,Index>& m_data;
Index m_id;
const Index m_end;
};
template<typename Scalar, int _Options, typename _Index>
class SparseVector<Scalar,_Options,_Index>::ReverseInnerIterator
{
public:
ReverseInnerIterator(const SparseVector& vec, Index outer=0)
: m_data(vec.m_data), m_id(static_cast<Index>(m_data.size())), m_start(0)
{
eigen_assert(outer==0);
}
ReverseInnerIterator(const internal::CompressedStorage<Scalar,Index>& data)
: m_data(data), m_id(static_cast<Index>(m_data.size())), m_start(0)
{}
inline ReverseInnerIterator& operator--() { m_id--; return *this; }
inline Scalar value() const { return m_data.value(m_id-1); }
inline Scalar& valueRef() { return const_cast<Scalar&>(m_data.value(m_id-1)); }
inline Index index() const { return m_data.index(m_id-1); }
inline Index row() const { return IsColVector ? index() : 0; }
inline Index col() const { return IsColVector ? 0 : index(); }
inline operator bool() const { return (m_id > m_start); }
protected:
const internal::CompressedStorage<Scalar,Index>& m_data;
Index m_id;
const Index m_start;
};
#endif // EIGEN_SPARSEVECTOR_H

View File

@@ -1,7 +1,7 @@
// This file is part of Eigen, a lightweight C++ template library
// for linear algebra.
//
// Copyright (C) 2011 Gael Guennebaud <gael.guennebaud@inria.fr>
// Copyright (C) 2010 Gael Guennebaud <gael.guennebaud@inria.fr>
// Copyright (C) 2010 Daniel Lowengrub <lowdanie@gmail.com>
//
// Eigen is free software; you can redistribute it and/or
@@ -92,10 +92,10 @@ protected:
private:
void incrementToNonZero()
{
while((bool(*this)) && internal::isMuchSmallerThan(value(), m_view.m_reference, m_view.m_epsilon))
{
IterBase::operator++();
}
while(internal::isMuchSmallerThan(value(), m_view.m_reference, m_view.m_epsilon) && (bool(*this)))
{
IterBase::operator++();
}
}
};

View File

@@ -82,17 +82,8 @@ struct sparse_solve_triangular_selector<Lhs,Rhs,Mode,Upper,RowMajor>
for(int i=lhs.rows()-1 ; i>=0 ; --i)
{
Scalar tmp = other.coeff(i,col);
Scalar l_ii = 0;
typename Lhs::InnerIterator it(lhs, i);
while(it && it.index()<i)
++it;
if(!(Mode & UnitDiag))
{
eigen_assert(it && it.index()==i);
l_ii = it.value();
++it;
}
else if (it && it.index() == i)
if (it && it.index() == i)
++it;
for(; it; ++it)
{
@@ -102,7 +93,11 @@ struct sparse_solve_triangular_selector<Lhs,Rhs,Mode,Upper,RowMajor>
if (Mode & UnitDiag)
other.coeffRef(i,col) = tmp;
else
other.coeffRef(i,col) = tmp/l_ii;
{
typename Lhs::InnerIterator it(lhs, i);
eigen_assert(it && it.index() == i);
other.coeffRef(i,col) = tmp/it.value();
}
}
}
}
@@ -123,11 +118,9 @@ struct sparse_solve_triangular_selector<Lhs,Rhs,Mode,Lower,ColMajor>
if (tmp!=Scalar(0)) // optimization when other is actually sparse
{
typename Lhs::InnerIterator it(lhs, i);
while(it && it.index()<i)
++it;
if(!(Mode & UnitDiag))
{
eigen_assert(it && it.index()==i);
eigen_assert(it.index()==i);
tmp /= it.value();
}
if (it && it.index()==i)
@@ -156,11 +149,9 @@ struct sparse_solve_triangular_selector<Lhs,Rhs,Mode,Upper,ColMajor>
{
if(!(Mode & UnitDiag))
{
typename Lhs::ReverseInnerIterator it(lhs, i);
while(it && it.index()!=i)
--it;
eigen_assert(it && it.index()==i);
other.coeffRef(i,col) /= it.value();
// FIXME lhs.coeff(i,i) might not be always efficient while it must simply be the
// last element of the column !
other.coeffRef(i,col) /= lhs.innerVector(i).lastCoeff();
}
typename Lhs::InnerIterator it(lhs, i);
for(; it && it.index()<i; ++it)

Some files were not shown because too many files have changed in this diff Show More