Bump to 3.3.8-rc1

Bug corrupts results on Maxwell and earlier GPU architectures.

(cherry picked from commit dd6de618c3)

.gitignore

@@ -1,10 +1,10 @@
-syntax: glob
 qrc_*cxx
 *.orig
 *.pyc
 *.diff
 diff
 *.save
+save
 *.old
 *.gmo
 *.qm
@@ -12,7 +12,7 @@ core
 core.*
 *.bak
 *~
-build
+*build*
 *.moc.*
 *.moc
 ui_*
@@ -22,3 +22,16 @@ tags
 activity.png
 *.out
 *.php*
+*.log
+*.orig
+*.rej
+log
+patch
+*.patch
+a
+a.*
+lapack/testing
+lapack/reference
+.*project
+.settings
+Makefile

.hgeol

@@ -0,0 +1,11 @@
+[patterns]
+*.sh = LF
+*.MINPACK = CRLF
+scripts/*.in = LF
+debug/msvc/*.dat = CRLF
+debug/msvc/*.natvis = CRLF
+unsupported/test/mpreal/*.* = CRLF
+** = native
+
+[repository]
+native = LF

.krazy

@@ -1,3 +0,0 @@
-SKIP /disabled/
-SKIP /bench/
-SKIP /build/

CMakeLists.txt

@@ -1,105 +1,616 @@
-project(Eigen)
-cmake_minimum_required(VERSION 2.6.2)
+project(Eigen3)
 
-set(EIGEN_VERSION_NUMBER "2.0.8")
-set(EIGEN_VERSION "${EIGEN_VERSION_NUMBER}")
+cmake_minimum_required(VERSION 2.8.5)
+
+# guard against in-source builds
+
+if(${CMAKE_SOURCE_DIR} STREQUAL ${CMAKE_BINARY_DIR})
+  message(FATAL_ERROR "In-source builds not allowed. Please make a new directory (called a build directory) and run CMake from there. You may need to remove CMakeCache.txt. ")
+endif()
+
+# Alias Eigen_*_DIR to Eigen3_*_DIR:
+
+set(Eigen_SOURCE_DIR ${Eigen3_SOURCE_DIR})
+set(Eigen_BINARY_DIR ${Eigen3_BINARY_DIR})
+
+# guard against bad build-type strings
+
+if (NOT CMAKE_BUILD_TYPE)
+  set(CMAKE_BUILD_TYPE "Release")
+endif()
+
+string(TOLOWER "${CMAKE_BUILD_TYPE}" cmake_build_type_tolower)
+if(    NOT cmake_build_type_tolower STREQUAL "debug"
+   AND NOT cmake_build_type_tolower STREQUAL "release"
+   AND NOT cmake_build_type_tolower STREQUAL "relwithdebinfo")
+  message(FATAL_ERROR "Unknown build type \"${CMAKE_BUILD_TYPE}\". Allowed values are Debug, Release, RelWithDebInfo (case-insensitive).")
+endif()
+
+
+#############################################################################
+# retrieve version infomation                                               #
+#############################################################################
+
+# automatically parse the version number
+file(READ "${PROJECT_SOURCE_DIR}/Eigen/src/Core/util/Macros.h" _eigen_version_header)
+string(REGEX MATCH "define[ \t]+EIGEN_WORLD_VERSION[ \t]+([0-9]+)" _eigen_world_version_match "${_eigen_version_header}")
+set(EIGEN_WORLD_VERSION "${CMAKE_MATCH_1}")
+string(REGEX MATCH "define[ \t]+EIGEN_MAJOR_VERSION[ \t]+([0-9]+)" _eigen_major_version_match "${_eigen_version_header}")
+set(EIGEN_MAJOR_VERSION "${CMAKE_MATCH_1}")
+string(REGEX MATCH "define[ \t]+EIGEN_MINOR_VERSION[ \t]+([0-9]+)" _eigen_minor_version_match "${_eigen_version_header}")
+set(EIGEN_MINOR_VERSION "${CMAKE_MATCH_1}")
+set(EIGEN_VERSION_NUMBER ${EIGEN_WORLD_VERSION}.${EIGEN_MAJOR_VERSION}.${EIGEN_MINOR_VERSION})
+
+# if we are not in a mercurial clone
+if(IS_DIRECTORY ${CMAKE_SOURCE_DIR}/.hg)
+  # if the mercurial program is absent or this will leave the EIGEN_HG_CHANGESET string empty,
+  # but won't stop CMake.
+  execute_process(COMMAND hg tip -R ${CMAKE_SOURCE_DIR} OUTPUT_VARIABLE EIGEN_HGTIP_OUTPUT)
+  execute_process(COMMAND hg branch -R ${CMAKE_SOURCE_DIR} OUTPUT_VARIABLE EIGEN_BRANCH_OUTPUT)
+endif()
+
+# if this is the default (aka development) branch, extract the mercurial changeset number from the hg tip output...
+if(EIGEN_BRANCH_OUTPUT MATCHES "default")
+string(REGEX MATCH "^changeset: *[0-9]*:([0-9;a-f]+).*" EIGEN_HG_CHANGESET_MATCH "${EIGEN_HGTIP_OUTPUT}")
+set(EIGEN_HG_CHANGESET "${CMAKE_MATCH_1}")
+endif(EIGEN_BRANCH_OUTPUT MATCHES "default")
+#...and show it next to the version number
+if(EIGEN_HG_CHANGESET)
+  set(EIGEN_VERSION "${EIGEN_VERSION_NUMBER} (mercurial changeset ${EIGEN_HG_CHANGESET})")
+else(EIGEN_HG_CHANGESET)
+  set(EIGEN_VERSION "${EIGEN_VERSION_NUMBER}")
+endif(EIGEN_HG_CHANGESET)
+
+
+include(CheckCXXCompilerFlag)
+include(GNUInstallDirs)
 
 set(CMAKE_MODULE_PATH ${PROJECT_SOURCE_DIR}/cmake)
 
-option(EIGEN_BUILD_TESTS "Build tests" OFF)
-option(EIGEN_BUILD_DEMOS "Build demos" OFF)
-if(NOT WIN32)
-  option(EIGEN_BUILD_LIB "Build the binary shared library" OFF)
-endif(NOT WIN32)
-option(EIGEN_BUILD_BTL "Build benchmark suite" OFF)
-if(NOT WIN32)
-  option(EIGEN_BUILD_PKGCONFIG "Build pkg-config .pc file for Eigen" ON)
-endif(NOT WIN32)
 
-if(EIGEN_BUILD_LIB)
-  option(EIGEN_TEST_LIB "Build the unit tests using the library (disable -pedantic)" OFF)
-endif(EIGEN_BUILD_LIB)
+option(EIGEN_TEST_CXX11 "Enable testing with C++11 and C++11 features (e.g. Tensor module)." OFF)
+
+
+macro(ei_add_cxx_compiler_flag FLAG)
+  string(REGEX REPLACE "-" "" SFLAG1 ${FLAG})
+  string(REGEX REPLACE "\\+" "p" SFLAG ${SFLAG1})
+  check_cxx_compiler_flag(${FLAG} COMPILER_SUPPORT_${SFLAG})
+  if(COMPILER_SUPPORT_${SFLAG})
+    set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} ${FLAG}")
+  endif()
+endmacro(ei_add_cxx_compiler_flag)
+
+check_cxx_compiler_flag("-std=c++11" EIGEN_COMPILER_SUPPORT_CPP11)
+
+if(EIGEN_TEST_CXX11)
+  set(CMAKE_CXX_STANDARD 11)
+  set(CMAKE_CXX_EXTENSIONS OFF)
+  if(EIGEN_COMPILER_SUPPORT_CPP11)
+    set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -std=c++11")
+  endif()
+else()
+  #set(CMAKE_CXX_STANDARD 03)
+  #set(CMAKE_CXX_EXTENSIONS OFF)
+  ei_add_cxx_compiler_flag("-std=c++03")
+endif()
+
+#############################################################################
+# find how to link to the standard libraries                                #
+#############################################################################
+
+find_package(StandardMathLibrary)
+
+
+set(EIGEN_TEST_CUSTOM_LINKER_FLAGS  "" CACHE STRING "Additional linker flags when linking unit tests.")
+set(EIGEN_TEST_CUSTOM_CXX_FLAGS     "" CACHE STRING "Additional compiler flags when compiling unit tests.")
+
+set(EIGEN_STANDARD_LIBRARIES_TO_LINK_TO "")
+
+if(NOT STANDARD_MATH_LIBRARY_FOUND)
+
+  message(FATAL_ERROR
+    "Can't link to the standard math library. Please report to the Eigen developers, telling them about your platform.")
+
+else()
+
+  if(EIGEN_STANDARD_LIBRARIES_TO_LINK_TO)
+    set(EIGEN_STANDARD_LIBRARIES_TO_LINK_TO "${EIGEN_STANDARD_LIBRARIES_TO_LINK_TO} ${STANDARD_MATH_LIBRARY}")
+  else()
+    set(EIGEN_STANDARD_LIBRARIES_TO_LINK_TO "${STANDARD_MATH_LIBRARY}")
+  endif()
+
+endif()
+
+if(EIGEN_STANDARD_LIBRARIES_TO_LINK_TO)
+  message(STATUS "Standard libraries to link to explicitly: ${EIGEN_STANDARD_LIBRARIES_TO_LINK_TO}")
+else()
+  message(STATUS "Standard libraries to link to explicitly: none")
+endif()
+
+option(EIGEN_BUILD_BTL "Build benchmark suite" OFF)
+
+# Disable pkgconfig only for native Windows builds
+if(NOT WIN32 OR NOT CMAKE_HOST_SYSTEM_NAME MATCHES Windows)
+  option(EIGEN_BUILD_PKGCONFIG "Build pkg-config .pc file for Eigen" ON)
+endif()
 
 set(CMAKE_INCLUDE_CURRENT_DIR ON)
 
-if(CMAKE_COMPILER_IS_GNUCXX)
-  if(CMAKE_SYSTEM_NAME MATCHES Linux)
-    include(CheckCXXCompilerFlag)
-    set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -Wnon-virtual-dtor -Wno-long-long -ansi -Wundef -Wcast-align -Wchar-subscripts -Wall -W -Wpointer-arith -Wwrite-strings -Wformat-security -fno-exceptions -fno-check-new -fno-common -fstrict-aliasing")
-    check_cxx_compiler_flag("-Wextra" has_wextra)
-    if(has_wextra)
-      set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -Wextra")
-    endif()
-    if(NOT EIGEN_TEST_LIB)
-      set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -pedantic")
-    endif(NOT EIGEN_TEST_LIB)
+option(EIGEN_SPLIT_LARGE_TESTS "Split large tests into smaller executables" ON)
 
-    option(EIGEN_TEST_SSE2 "Enable/Disable SSE2 in tests/examples" OFF)
-    if(EIGEN_TEST_SSE2)
-      set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -msse2")
-      message("Enabling SSE2 in tests/examples")
-    endif(EIGEN_TEST_SSE2)
+option(EIGEN_DEFAULT_TO_ROW_MAJOR "Use row-major as default matrix storage order" OFF)
+if(EIGEN_DEFAULT_TO_ROW_MAJOR)
+  add_definitions("-DEIGEN_DEFAULT_TO_ROW_MAJOR")
+endif()
 
-    option(EIGEN_TEST_SSE3 "Enable/Disable SSE3 in tests/examples" OFF)
-    if(EIGEN_TEST_SSE3)
-      set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -msse3")
-      message("Enabling SSE3 in tests/examples")
-    endif(EIGEN_TEST_SSE3)
+set(EIGEN_TEST_MAX_SIZE "320" CACHE STRING "Maximal matrix/vector size, default is 320")
 
-    option(EIGEN_TEST_SSSE3 "Enable/Disable SSSE3 in tests/examples" OFF)
-    if(EIGEN_TEST_SSSE3)
-      set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -mssse3")
-      message("Enabling SSSE3 in tests/examples")
-    endif(EIGEN_TEST_SSSE3)
+if(NOT MSVC)
+  # We assume that other compilers are partly compatible with GNUCC
 
-    option(EIGEN_TEST_ALTIVEC "Enable/Disable altivec in tests/examples" OFF)
-    if(EIGEN_TEST_ALTIVEC)
-      set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -maltivec -mabi=altivec")
-      message("Enabling AltiVec in tests/examples")
-    endif(EIGEN_TEST_ALTIVEC)
+  # clang outputs some warnings for unknown flags that are not caught by check_cxx_compiler_flag
+  # adding -Werror turns such warnings into errors
+  check_cxx_compiler_flag("-Werror" COMPILER_SUPPORT_WERROR)
+  if(COMPILER_SUPPORT_WERROR)
+    set(CMAKE_REQUIRED_FLAGS "-Werror")
+  endif()
+  ei_add_cxx_compiler_flag("-pedantic")
+  ei_add_cxx_compiler_flag("-Wall")
+  ei_add_cxx_compiler_flag("-Wextra")
+  #ei_add_cxx_compiler_flag("-Weverything")              # clang
+  
+  ei_add_cxx_compiler_flag("-Wundef")
+  ei_add_cxx_compiler_flag("-Wcast-align")
+  ei_add_cxx_compiler_flag("-Wchar-subscripts")
+  ei_add_cxx_compiler_flag("-Wnon-virtual-dtor")
+  ei_add_cxx_compiler_flag("-Wunused-local-typedefs")
+  ei_add_cxx_compiler_flag("-Wpointer-arith")
+  ei_add_cxx_compiler_flag("-Wwrite-strings")
+  ei_add_cxx_compiler_flag("-Wformat-security")
+  ei_add_cxx_compiler_flag("-Wshorten-64-to-32")
+  ei_add_cxx_compiler_flag("-Wlogical-op")
+  ei_add_cxx_compiler_flag("-Wenum-conversion")
+  ei_add_cxx_compiler_flag("-Wc++11-extensions")
+  ei_add_cxx_compiler_flag("-Wdouble-promotion")
+#  ei_add_cxx_compiler_flag("-Wconversion")
+  
+  # -Wshadow is insanely too strict with gcc, hopefully it will become usable with gcc 6
+  # if(NOT CMAKE_COMPILER_IS_GNUCXX OR (CMAKE_CXX_COMPILER_VERSION VERSION_GREATER "5.0.0"))
+  if(NOT CMAKE_COMPILER_IS_GNUCXX)
+    ei_add_cxx_compiler_flag("-Wshadow")
+  endif()
+  
+  ei_add_cxx_compiler_flag("-Wno-psabi")
+  ei_add_cxx_compiler_flag("-Wno-variadic-macros")
+  ei_add_cxx_compiler_flag("-Wno-long-long")
+  
+  ei_add_cxx_compiler_flag("-fno-check-new")
+  ei_add_cxx_compiler_flag("-fno-common")
+  ei_add_cxx_compiler_flag("-fstrict-aliasing")
+  ei_add_cxx_compiler_flag("-wd981")                    # disable ICC's "operands are evaluated in unspecified order" remark
+  ei_add_cxx_compiler_flag("-wd2304")                   # disable ICC's "warning #2304: non-explicit constructor with single argument may cause implicit type conversion" produced by -Wnon-virtual-dtor
+  
+  
+  # The -ansi flag must be added last, otherwise it is also used as a linker flag by check_cxx_compiler_flag making it fails
+  # Moreover we should not set both -strict-ansi and -ansi
+  check_cxx_compiler_flag("-strict-ansi" COMPILER_SUPPORT_STRICTANSI)
+  ei_add_cxx_compiler_flag("-Qunused-arguments")        # disable clang warning: argument unused during compilation: '-ansi'
+  
+  if(COMPILER_SUPPORT_STRICTANSI)
+    set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -strict-ansi")
+  else()
+    ei_add_cxx_compiler_flag("-ansi")
+  endif()
 
-  endif(CMAKE_SYSTEM_NAME MATCHES Linux)
-endif(CMAKE_COMPILER_IS_GNUCXX)
-
-if(MSVC)
-  set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} ")
+  if(ANDROID_NDK)
+    ei_add_cxx_compiler_flag("-pie")
+    ei_add_cxx_compiler_flag("-fPIE")
+  endif()
+  
+  set(CMAKE_REQUIRED_FLAGS "")
 
   option(EIGEN_TEST_SSE2 "Enable/Disable SSE2 in tests/examples" OFF)
   if(EIGEN_TEST_SSE2)
-    set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} /arch:SSE2")
-    message("Enabling SSE2 in tests/examples")
+    set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -msse2")
+    message(STATUS "Enabling SSE2 in tests/examples")
+  endif()
+
+  option(EIGEN_TEST_SSE3 "Enable/Disable SSE3 in tests/examples" OFF)
+  if(EIGEN_TEST_SSE3)
+    set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -msse3")
+    message(STATUS "Enabling SSE3 in tests/examples")
+  endif()
+
+  option(EIGEN_TEST_SSSE3 "Enable/Disable SSSE3 in tests/examples" OFF)
+  if(EIGEN_TEST_SSSE3)
+    set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -mssse3")
+    message(STATUS "Enabling SSSE3 in tests/examples")
+  endif()
+
+  option(EIGEN_TEST_SSE4_1 "Enable/Disable SSE4.1 in tests/examples" OFF)
+  if(EIGEN_TEST_SSE4_1)
+    set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -msse4.1")
+    message(STATUS "Enabling SSE4.1 in tests/examples")
+  endif()
+
+  option(EIGEN_TEST_SSE4_2 "Enable/Disable SSE4.2 in tests/examples" OFF)
+  if(EIGEN_TEST_SSE4_2)
+    set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -msse4.2")
+    message(STATUS "Enabling SSE4.2 in tests/examples")
+  endif()
+
+  option(EIGEN_TEST_AVX "Enable/Disable AVX in tests/examples" OFF)
+  if(EIGEN_TEST_AVX)
+    set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -mavx")
+    message(STATUS "Enabling AVX in tests/examples")
+  endif()
+
+  option(EIGEN_TEST_FMA "Enable/Disable FMA in tests/examples" OFF)
+  if(EIGEN_TEST_FMA AND NOT EIGEN_TEST_NEON)
+    set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -mfma")
+    message(STATUS "Enabling FMA in tests/examples")
+  endif()
+
+  option(EIGEN_TEST_AVX512 "Enable/Disable AVX512 in tests/examples" OFF)
+  if(EIGEN_TEST_AVX512)
+    set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -mavx512f -fabi-version=6 -DEIGEN_ENABLE_AVX512")
+    message(STATUS "Enabling AVX512 in tests/examples")
+  endif()
+
+  option(EIGEN_TEST_F16C "Enable/Disable F16C in tests/examples" OFF)
+  if(EIGEN_TEST_F16C)
+    set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -mf16c")
+    message(STATUS "Enabling F16C in tests/examples")
+  endif()
+
+  option(EIGEN_TEST_ALTIVEC "Enable/Disable AltiVec in tests/examples" OFF)
+  if(EIGEN_TEST_ALTIVEC)
+    set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -maltivec -mabi=altivec")
+    message(STATUS "Enabling AltiVec in tests/examples")
+  endif()
+
+  option(EIGEN_TEST_VSX "Enable/Disable VSX in tests/examples" OFF)
+  if(EIGEN_TEST_VSX)
+    set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -m64 -mvsx")
+    message(STATUS "Enabling VSX in tests/examples")
+  endif()
+
+  option(EIGEN_TEST_NEON "Enable/Disable Neon in tests/examples" OFF)
+  if(EIGEN_TEST_NEON)
+    if(EIGEN_TEST_FMA)
+      set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -mfpu=neon-vfpv4")
+    else()
+      set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -mfpu=neon")
+    endif()
+    set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -mfloat-abi=hard")
+    message(STATUS "Enabling NEON in tests/examples")
+  endif()
+
+  option(EIGEN_TEST_NEON64 "Enable/Disable Neon in tests/examples" OFF)
+  if(EIGEN_TEST_NEON64)
+    set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS}")
+    message(STATUS "Enabling NEON in tests/examples")
+  endif()
+
+  option(EIGEN_TEST_ZVECTOR "Enable/Disable S390X(zEC13) ZVECTOR in tests/examples" OFF)
+  if(EIGEN_TEST_ZVECTOR)
+    set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -march=z13 -mzvector")
+    message(STATUS "Enabling S390X(zEC13) ZVECTOR in tests/examples")
+  endif()
+
+  check_cxx_compiler_flag("-fopenmp" COMPILER_SUPPORT_OPENMP)
+  if(COMPILER_SUPPORT_OPENMP)
+    option(EIGEN_TEST_OPENMP "Enable/Disable OpenMP in tests/examples" OFF)
+    if(EIGEN_TEST_OPENMP)
+      set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -fopenmp")
+      message(STATUS "Enabling OpenMP in tests/examples")
+    endif()
+  endif()
+
+else(NOT MSVC)
+
+  # C4127 - conditional expression is constant
+  # C4714 - marked as __forceinline not inlined (I failed to deactivate it selectively)
+  #         We can disable this warning in the unit tests since it is clear that it occurs
+  #         because we are oftentimes returning objects that have a destructor or may
+  #         throw exceptions - in particular in the unit tests we are throwing extra many
+  #         exceptions to cover indexing errors.
+  # C4505 - unreferenced local function has been removed (impossible to deactive selectively)
+  set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} /EHsc /wd4127 /wd4505 /wd4714")
+
+  # replace all /Wx by /W4
+  string(REGEX REPLACE "/W[0-9]" "/W4" CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS}")
+
+  check_cxx_compiler_flag("/openmp" COMPILER_SUPPORT_OPENMP)
+  if(COMPILER_SUPPORT_OPENMP)
+    option(EIGEN_TEST_OPENMP "Enable/Disable OpenMP in tests/examples" OFF)
+    if(EIGEN_TEST_OPENMP)
+      set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} /openmp")
+      message(STATUS "Enabling OpenMP in tests/examples")
+    endif()
+  endif()
+
+  option(EIGEN_TEST_SSE2 "Enable/Disable SSE2 in tests/examples" OFF)
+  if(EIGEN_TEST_SSE2)
+    if(NOT CMAKE_CL_64)
+      # arch is not supported on 64 bit systems, SSE is enabled automatically.
+      set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} /arch:SSE2")
+    endif(NOT CMAKE_CL_64)
+    message(STATUS "Enabling SSE2 in tests/examples")
   endif(EIGEN_TEST_SSE2)
-endif(MSVC)
+endif(NOT MSVC)
 
 option(EIGEN_TEST_NO_EXPLICIT_VECTORIZATION "Disable explicit vectorization in tests/examples" OFF)
+option(EIGEN_TEST_X87 "Force using X87 instructions. Implies no vectorization." OFF)
+option(EIGEN_TEST_32BIT "Force generating 32bit code." OFF)
+
+if(EIGEN_TEST_X87)
+  set(EIGEN_TEST_NO_EXPLICIT_VECTORIZATION ON)
+  if(CMAKE_COMPILER_IS_GNUCXX)
+    set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -mfpmath=387")
+    message(STATUS "Forcing use of x87 instructions in tests/examples")
+  else()
+    message(STATUS "EIGEN_TEST_X87 ignored on your compiler")
+  endif()
+endif()
+
+if(EIGEN_TEST_32BIT)
+  if(CMAKE_COMPILER_IS_GNUCXX)
+    set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -m32")
+    message(STATUS "Forcing generation of 32-bit code in tests/examples")
+  else()
+    message(STATUS "EIGEN_TEST_32BIT ignored on your compiler")
+  endif()
+endif()
+
 if(EIGEN_TEST_NO_EXPLICIT_VECTORIZATION)
   add_definitions(-DEIGEN_DONT_VECTORIZE=1)
-  message("Disabling vectorization in tests/examples")
-endif(EIGEN_TEST_NO_EXPLICIT_VECTORIZATION)
+  message(STATUS "Disabling vectorization in tests/examples")
+endif()
+
+option(EIGEN_TEST_NO_EXPLICIT_ALIGNMENT "Disable explicit alignment (hence vectorization) in tests/examples" OFF)
+if(EIGEN_TEST_NO_EXPLICIT_ALIGNMENT)
+  add_definitions(-DEIGEN_DONT_ALIGN=1)
+  message(STATUS "Disabling alignment in tests/examples")
+endif()
+
+option(EIGEN_TEST_NO_EXCEPTIONS "Disables C++ exceptions" OFF)
+if(EIGEN_TEST_NO_EXCEPTIONS)
+  ei_add_cxx_compiler_flag("-fno-exceptions")
+  message(STATUS "Disabling exceptions in tests/examples")
+endif()
+
+set(EIGEN_CUDA_COMPUTE_ARCH 30 CACHE STRING "The CUDA compute architecture level to target when compiling CUDA code")
 
 include_directories(${CMAKE_CURRENT_SOURCE_DIR} ${CMAKE_CURRENT_BINARY_DIR})
 
+# Backward compatibility support for EIGEN_INCLUDE_INSTALL_DIR
+if(EIGEN_INCLUDE_INSTALL_DIR)
+  message(WARNING "EIGEN_INCLUDE_INSTALL_DIR is deprecated. Use INCLUDE_INSTALL_DIR instead.")
+endif()
+
+if(EIGEN_INCLUDE_INSTALL_DIR AND NOT INCLUDE_INSTALL_DIR)
+  set(INCLUDE_INSTALL_DIR ${EIGEN_INCLUDE_INSTALL_DIR}
+      CACHE PATH "The directory relative to CMAKE_PREFIX_PATH where Eigen header files are installed")
+else()
+  set(INCLUDE_INSTALL_DIR
+      "${CMAKE_INSTALL_INCLUDEDIR}/eigen3"
+      CACHE PATH "The directory relative to CMAKE_PREFIX_PATH where Eigen header files are installed"
+      )
+endif()
+set(CMAKEPACKAGE_INSTALL_DIR
+    "${CMAKE_INSTALL_DATADIR}/eigen3/cmake"
+    CACHE PATH "The directory relative to CMAKE_PREFIX_PATH where Eigen3Config.cmake is installed"
+    )
+set(PKGCONFIG_INSTALL_DIR
+    "${CMAKE_INSTALL_DATADIR}/pkgconfig"
+    CACHE PATH "The directory relative to CMAKE_PREFIX_PATH where eigen3.pc is installed"
+    )
+
+
+# similar to set_target_properties but append the property instead of overwriting it
+macro(ei_add_target_property target prop value)
+
+  get_target_property(previous ${target} ${prop})
+  # if the property wasn't previously set, ${previous} is now "previous-NOTFOUND" which cmake allows catching with plain if()
+  if(NOT previous)
+    set(previous "")
+  endif(NOT previous)
+  set_target_properties(${target} PROPERTIES ${prop} "${previous} ${value}")
+endmacro(ei_add_target_property)
+
+install(FILES
+  signature_of_eigen3_matrix_library
+  DESTINATION ${INCLUDE_INSTALL_DIR} COMPONENT Devel
+  )
+
 if(EIGEN_BUILD_PKGCONFIG)
-    configure_file(eigen2.pc.in eigen2.pc)
-    install(FILES ${CMAKE_INSTALL_PREFIX}/eigen2.pc
-        DESTINATION lib/pkgconfig
+    configure_file(eigen3.pc.in eigen3.pc @ONLY)
+    install(FILES ${CMAKE_CURRENT_BINARY_DIR}/eigen3.pc
+        DESTINATION ${PKGCONFIG_INSTALL_DIR}
         )
-endif(EIGEN_BUILD_PKGCONFIG)
+endif()
 
 add_subdirectory(Eigen)
+
+add_subdirectory(doc EXCLUDE_FROM_ALL)
+
+option(BUILD_TESTING "Enable creation of Eigen tests." ON)
+if(BUILD_TESTING)
+  include(EigenConfigureTesting)
+
+  if(EIGEN_LEAVE_TEST_IN_ALL_TARGET)
+    add_subdirectory(test) # can't do EXCLUDE_FROM_ALL here, breaks CTest
+  else()
+    add_subdirectory(test EXCLUDE_FROM_ALL)
+  endif()
+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()
+
+# add SYCL
+option(EIGEN_TEST_SYCL "Add Sycl support." OFF)
+if(EIGEN_TEST_SYCL)
+  set (CMAKE_MODULE_PATH "${CMAKE_ROOT}/Modules" "cmake/Modules/" "${CMAKE_MODULE_PATH}")
+  include(FindComputeCpp)
+endif()
+
 add_subdirectory(unsupported)
 
-if(EIGEN_BUILD_TESTS)
-  include(CTest)
-  add_subdirectory(test)
-endif(EIGEN_BUILD_TESTS)
+add_subdirectory(demos EXCLUDE_FROM_ALL)
 
-add_subdirectory(doc)
-
-if(EIGEN_BUILD_DEMOS)
-  add_subdirectory(demos)
-endif(EIGEN_BUILD_DEMOS)
+# must be after test and unsupported, for configuring buildtests.in
+add_subdirectory(scripts EXCLUDE_FROM_ALL)
 
+# TODO: consider also replacing EIGEN_BUILD_BTL by a custom target "make btl"?
 if(EIGEN_BUILD_BTL)
-  add_subdirectory(bench/btl)
+  add_subdirectory(bench/btl EXCLUDE_FROM_ALL)
 endif(EIGEN_BUILD_BTL)
+
+if(NOT WIN32)
+  add_subdirectory(bench/spbench EXCLUDE_FROM_ALL)
+endif(NOT WIN32)
+
+configure_file(scripts/cdashtesting.cmake.in cdashtesting.cmake @ONLY)
+
+if(BUILD_TESTING)
+  ei_testing_print_summary()
+endif()
+
+message(STATUS "")
+message(STATUS "Configured Eigen ${EIGEN_VERSION_NUMBER}")
+message(STATUS "")
+
+option(EIGEN_FAILTEST "Enable failtests." OFF)
+if(EIGEN_FAILTEST)
+  add_subdirectory(failtest)
+endif()
+
+string(TOLOWER "${CMAKE_GENERATOR}" cmake_generator_tolower)
+if(cmake_generator_tolower MATCHES "makefile")
+  message(STATUS "Some things you can do now:")
+  message(STATUS "--------------+--------------------------------------------------------------")
+  message(STATUS "Command       |   Description")
+  message(STATUS "--------------+--------------------------------------------------------------")
+  message(STATUS "make install  | Install Eigen. Headers will be installed to:")
+  message(STATUS "              |     <CMAKE_INSTALL_PREFIX>/<INCLUDE_INSTALL_DIR>")
+  message(STATUS "              |   Using the following values:")
+  message(STATUS "              |     CMAKE_INSTALL_PREFIX: ${CMAKE_INSTALL_PREFIX}")
+  message(STATUS "              |     INCLUDE_INSTALL_DIR:  ${INCLUDE_INSTALL_DIR}")
+  message(STATUS "              |   Change the install location of Eigen headers using:")
+  message(STATUS "              |     cmake . -DCMAKE_INSTALL_PREFIX=yourprefix")
+  message(STATUS "              |   Or:")
+  message(STATUS "              |     cmake . -DINCLUDE_INSTALL_DIR=yourdir")
+  message(STATUS "make doc      | Generate the API documentation, requires Doxygen & LaTeX")
+  message(STATUS "make check    | Build and run the unit-tests. Read this page:")
+  message(STATUS "              |   http://eigen.tuxfamily.org/index.php?title=Tests")
+  message(STATUS "make blas     | Build BLAS library (not the same thing as Eigen)")
+  message(STATUS "make uninstall| Removes files installed by make install")
+  message(STATUS "--------------+--------------------------------------------------------------")
+else()
+  message(STATUS "To build/run the unit tests, read this page:")
+  message(STATUS "  http://eigen.tuxfamily.org/index.php?title=Tests")
+endif()
+
+message(STATUS "")
+
+
+set ( EIGEN_VERSION_STRING ${EIGEN_VERSION_NUMBER} )
+set ( EIGEN_VERSION_MAJOR  ${EIGEN_WORLD_VERSION} )
+set ( EIGEN_VERSION_MINOR  ${EIGEN_MAJOR_VERSION} )
+set ( EIGEN_VERSION_PATCH  ${EIGEN_MINOR_VERSION} )
+set ( EIGEN_DEFINITIONS "")
+set ( EIGEN_INCLUDE_DIR "${CMAKE_INSTALL_PREFIX}/${INCLUDE_INSTALL_DIR}" )
+set ( EIGEN_ROOT_DIR ${CMAKE_INSTALL_PREFIX} )
+
+# Interface libraries require at least CMake 3.0
+if (NOT CMAKE_VERSION VERSION_LESS 3.0)
+  include (CMakePackageConfigHelpers)
+
+  # Imported target support
+  add_library (eigen INTERFACE)
+
+  target_compile_definitions (eigen INTERFACE ${EIGEN_DEFINITIONS})
+  target_include_directories (eigen INTERFACE
+    $<BUILD_INTERFACE:${CMAKE_CURRENT_SOURCE_DIR}>
+    $<INSTALL_INTERFACE:${INCLUDE_INSTALL_DIR}>
+  )
+
+  # Export as title case Eigen
+  set_target_properties (eigen PROPERTIES EXPORT_NAME Eigen)
+
+  install (TARGETS eigen EXPORT Eigen3Targets)
+
+  configure_package_config_file (
+    ${CMAKE_CURRENT_SOURCE_DIR}/cmake/Eigen3Config.cmake.in
+    ${CMAKE_CURRENT_BINARY_DIR}/Eigen3Config.cmake
+    PATH_VARS EIGEN_INCLUDE_DIR EIGEN_ROOT_DIR
+    INSTALL_DESTINATION ${CMAKEPACKAGE_INSTALL_DIR}
+    NO_CHECK_REQUIRED_COMPONENTS_MACRO # Eigen does not provide components
+  )
+  # Remove CMAKE_SIZEOF_VOID_P from Eigen3ConfigVersion.cmake since Eigen does
+  # not depend on architecture specific settings or libraries. More
+  # specifically, an Eigen3Config.cmake generated from a 64 bit target can be
+  # used for 32 bit targets as well (and vice versa).
+  set (_Eigen3_CMAKE_SIZEOF_VOID_P ${CMAKE_SIZEOF_VOID_P})
+  unset (CMAKE_SIZEOF_VOID_P)
+  write_basic_package_version_file (Eigen3ConfigVersion.cmake
+                                    VERSION ${EIGEN_VERSION_NUMBER}
+                                    COMPATIBILITY SameMajorVersion)
+  set (CMAKE_SIZEOF_VOID_P ${_Eigen3_CMAKE_SIZEOF_VOID_P})
+
+  # The Eigen target will be located in the Eigen3 namespace. Other CMake
+  # targets can refer to it using Eigen3::Eigen.
+  export (TARGETS eigen NAMESPACE Eigen3:: FILE Eigen3Targets.cmake)
+  # Export Eigen3 package to CMake registry such that it can be easily found by
+  # CMake even if it has not been installed to a standard directory.
+  export (PACKAGE Eigen3)
+
+  install (EXPORT Eigen3Targets NAMESPACE Eigen3:: DESTINATION ${CMAKEPACKAGE_INSTALL_DIR})
+
+else (NOT CMAKE_VERSION VERSION_LESS 3.0)
+  # Fallback to legacy Eigen3Config.cmake without the imported target
+  
+  # If CMakePackageConfigHelpers module is available (CMake >= 2.8.8)
+  # create a relocatable Config file, otherwise leave the hardcoded paths       
+  include(CMakePackageConfigHelpers OPTIONAL RESULT_VARIABLE CPCH_PATH)
+  
+  if(CPCH_PATH)
+    configure_package_config_file (
+      ${CMAKE_CURRENT_SOURCE_DIR}/cmake/Eigen3ConfigLegacy.cmake.in
+      ${CMAKE_CURRENT_BINARY_DIR}/Eigen3Config.cmake
+      PATH_VARS EIGEN_INCLUDE_DIR EIGEN_ROOT_DIR
+      INSTALL_DESTINATION ${CMAKEPACKAGE_INSTALL_DIR}
+      NO_CHECK_REQUIRED_COMPONENTS_MACRO # Eigen does not provide components
+    )
+  else() 
+    # The PACKAGE_* variables are defined by the configure_package_config_file
+    # but without it we define them manually to the hardcoded paths
+    set(PACKAGE_INIT "")
+    set(PACKAGE_EIGEN_INCLUDE_DIR ${EIGEN_INCLUDE_DIR})
+    set(PACKAGE_EIGEN_ROOT_DIR ${EIGEN_ROOT_DIR})
+    configure_file ( ${CMAKE_CURRENT_SOURCE_DIR}/cmake/Eigen3ConfigLegacy.cmake.in
+                     ${CMAKE_CURRENT_BINARY_DIR}/Eigen3Config.cmake
+                     @ONLY ESCAPE_QUOTES )
+  endif()
+
+  write_basic_package_version_file( Eigen3ConfigVersion.cmake
+                                    VERSION ${EIGEN_VERSION_NUMBER}
+                                    COMPATIBILITY SameMajorVersion )
+
+endif (NOT CMAKE_VERSION VERSION_LESS 3.0)
+
+install ( FILES ${CMAKE_CURRENT_SOURCE_DIR}/cmake/UseEigen3.cmake
+                ${CMAKE_CURRENT_BINARY_DIR}/Eigen3Config.cmake
+                ${CMAKE_CURRENT_BINARY_DIR}/Eigen3ConfigVersion.cmake
+          DESTINATION ${CMAKEPACKAGE_INSTALL_DIR} )
+
+# Add uninstall target
+add_custom_target ( uninstall
+    COMMAND ${CMAKE_COMMAND} -P ${CMAKE_CURRENT_SOURCE_DIR}/cmake/EigenUninstall.cmake)

COPYING.BSD

@@ -0,0 +1,26 @@
+/*
+ Copyright (c) 2011, Intel Corporation. All rights reserved.
+
+ Redistribution and use in source and binary forms, with or without modification,
+ are permitted provided that the following conditions are met:
+
+ * Redistributions of source code must retain the above copyright notice, this
+   list of conditions and the following disclaimer.
+ * Redistributions in binary form must reproduce the above copyright notice,
+   this list of conditions and the following disclaimer in the documentation
+   and/or other materials provided with the distribution.
+ * Neither the name of Intel Corporation nor the names of its contributors may
+   be used to endorse or promote products derived from this software without
+   specific prior written permission.
+
+ THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" AND
+ ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED
+ WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE
+ DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE LIABLE FOR
+ ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES
+ (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES;
+ LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON
+ ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+ (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS
+ SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+*/

COPYING.LESSER

@@ -1,165 +0,0 @@
-		   GNU LESSER GENERAL PUBLIC LICENSE
-                       Version 3, 29 June 2007
-
- Copyright (C) 2007 Free Software Foundation, Inc. <http://fsf.org/>
- Everyone is permitted to copy and distribute verbatim copies
- of this license document, but changing it is not allowed.
-
-
-  This version of the GNU Lesser General Public License incorporates
-the terms and conditions of version 3 of the GNU General Public
-License, supplemented by the additional permissions listed below.
-
-  0. Additional Definitions. 
-
-  As used herein, "this License" refers to version 3 of the GNU Lesser
-General Public License, and the "GNU GPL" refers to version 3 of the GNU
-General Public License.
-
-  "The Library" refers to a covered work governed by this License,
-other than an Application or a Combined Work as defined below.
-
-  An "Application" is any work that makes use of an interface provided
-by the Library, but which is not otherwise based on the Library.
-Defining a subclass of a class defined by the Library is deemed a mode
-of using an interface provided by the Library.
-
-  A "Combined Work" is a work produced by combining or linking an
-Application with the Library.  The particular version of the Library
-with which the Combined Work was made is also called the "Linked
-Version".
-
-  The "Minimal Corresponding Source" for a Combined Work means the
-Corresponding Source for the Combined Work, excluding any source code
-for portions of the Combined Work that, considered in isolation, are
-based on the Application, and not on the Linked Version.
-
-  The "Corresponding Application Code" for a Combined Work means the
-object code and/or source code for the Application, including any data
-and utility programs needed for reproducing the Combined Work from the
-Application, but excluding the System Libraries of the Combined Work.
-
-  1. Exception to Section 3 of the GNU GPL.
-
-  You may convey a covered work under sections 3 and 4 of this License
-without being bound by section 3 of the GNU GPL.
-
-  2. Conveying Modified Versions.
-
-  If you modify a copy of the Library, and, in your modifications, a
-facility refers to a function or data to be supplied by an Application
-that uses the facility (other than as an argument passed when the
-facility is invoked), then you may convey a copy of the modified
-version:
-
-   a) under this License, provided that you make a good faith effort to
-   ensure that, in the event an Application does not supply the
-   function or data, the facility still operates, and performs
-   whatever part of its purpose remains meaningful, or
-
-   b) under the GNU GPL, with none of the additional permissions of
-   this License applicable to that copy.
-
-  3. Object Code Incorporating Material from Library Header Files.
-
-  The object code form of an Application may incorporate material from
-a header file that is part of the Library.  You may convey such object
-code under terms of your choice, provided that, if the incorporated
-material is not limited to numerical parameters, data structure
-layouts and accessors, or small macros, inline functions and templates
-(ten or fewer lines in length), you do both of the following:
-
-   a) Give prominent notice with each copy of the object code that the
-   Library is used in it and that the Library and its use are
-   covered by this License.
-
-   b) Accompany the object code with a copy of the GNU GPL and this license
-   document.
-
-  4. Combined Works.
-
-  You may convey a Combined Work under terms of your choice that,
-taken together, effectively do not restrict modification of the
-portions of the Library contained in the Combined Work and reverse
-engineering for debugging such modifications, if you also do each of
-the following:
-
-   a) Give prominent notice with each copy of the Combined Work that
-   the Library is used in it and that the Library and its use are
-   covered by this License.
-
-   b) Accompany the Combined Work with a copy of the GNU GPL and this license
-   document.
-
-   c) For a Combined Work that displays copyright notices during
-   execution, include the copyright notice for the Library among
-   these notices, as well as a reference directing the user to the
-   copies of the GNU GPL and this license document.
-
-   d) Do one of the following:
-
-       0) Convey the Minimal Corresponding Source under the terms of this
-       License, and the Corresponding Application Code in a form
-       suitable for, and under terms that permit, the user to
-       recombine or relink the Application with a modified version of
-       the Linked Version to produce a modified Combined Work, in the
-       manner specified by section 6 of the GNU GPL for conveying
-       Corresponding Source.
-
-       1) Use a suitable shared library mechanism for linking with the
-       Library.  A suitable mechanism is one that (a) uses at run time
-       a copy of the Library already present on the user's computer
-       system, and (b) will operate properly with a modified version
-       of the Library that is interface-compatible with the Linked
-       Version. 
-
-   e) Provide Installation Information, but only if you would otherwise
-   be required to provide such information under section 6 of the
-   GNU GPL, and only to the extent that such information is
-   necessary to install and execute a modified version of the
-   Combined Work produced by recombining or relinking the
-   Application with a modified version of the Linked Version. (If
-   you use option 4d0, the Installation Information must accompany
-   the Minimal Corresponding Source and Corresponding Application
-   Code. If you use option 4d1, you must provide the Installation
-   Information in the manner specified by section 6 of the GNU GPL
-   for conveying Corresponding Source.)
-
-  5. Combined Libraries.
-
-  You may place library facilities that are a work based on the
-Library side by side in a single library together with other library
-facilities that are not Applications and are not covered by this
-License, and convey such a combined library under terms of your
-choice, if you do both of the following:
-
-   a) Accompany the combined library with a copy of the same work based
-   on the Library, uncombined with any other library facilities,
-   conveyed under the terms of this License.
-
-   b) Give prominent notice with the combined library that part of it
-   is a work based on the Library, and explaining where to find the
-   accompanying uncombined form of the same work.
-
-  6. Revised Versions of the GNU Lesser General Public License.
-
-  The Free Software Foundation may publish revised and/or new versions
-of the GNU Lesser General Public License from time to time. Such new
-versions will be similar in spirit to the present version, but may
-differ in detail to address new problems or concerns.
-
-  Each version is given a distinguishing version number. If the
-Library as you received it specifies that a certain numbered version
-of the GNU Lesser General Public License "or any later version"
-applies to it, you have the option of following the terms and
-conditions either of that published version or of any later version
-published by the Free Software Foundation. If the Library as you
-received it does not specify a version number of the GNU Lesser
-General Public License, you may choose any version of the GNU Lesser
-General Public License ever published by the Free Software Foundation.
-
-  If the Library as you received it specifies that a proxy can decide
-whether future versions of the GNU Lesser General Public License shall
-apply, that proxy's public statement of acceptance of any version is
-permanent authorization for you to choose that version for the
-Library.

COPYING.LGPL

@@ -0,0 +1,502 @@
+                  GNU LESSER GENERAL PUBLIC LICENSE
+                       Version 2.1, February 1999
+
+ Copyright (C) 1991, 1999 Free Software Foundation, Inc.
+ 51 Franklin Street, Fifth Floor, Boston, MA  02110-1301  USA
+ Everyone is permitted to copy and distribute verbatim copies
+ of this license document, but changing it is not allowed.
+
+[This is the first released version of the Lesser GPL.  It also counts
+ as the successor of the GNU Library Public License, version 2, hence
+ the version number 2.1.]
+
+                            Preamble
+
+  The licenses for most software are designed to take away your
+freedom to share and change it.  By contrast, the GNU General Public
+Licenses are intended to guarantee your freedom to share and change
+free software--to make sure the software is free for all its users.
+
+  This license, the Lesser General Public License, applies to some
+specially designated software packages--typically libraries--of the
+Free Software Foundation and other authors who decide to use it.  You
+can use it too, but we suggest you first think carefully about whether
+this license or the ordinary General Public License is the better
+strategy to use in any particular case, based on the explanations below.
+
+  When we speak of free software, we are referring to freedom of use,
+not price.  Our General Public Licenses are designed to make sure that
+you have the freedom to distribute copies of free software (and charge
+for this service if you wish); that you receive source code or can get
+it if you want it; that you can change the software and use pieces of
+it in new free programs; and that you are informed that you can do
+these things.
+
+  To protect your rights, we need to make restrictions that forbid
+distributors to deny you these rights or to ask you to surrender these
+rights.  These restrictions translate to certain responsibilities for
+you if you distribute copies of the library or if you modify it.
+
+  For example, if you distribute copies of the library, whether gratis
+or for a fee, you must give the recipients all the rights that we gave
+you.  You must make sure that they, too, receive or can get the source
+code.  If you link other code with the library, you must provide
+complete object files to the recipients, so that they can relink them
+with the library after making changes to the library and recompiling
+it.  And you must show them these terms so they know their rights.
+
+  We protect your rights with a two-step method: (1) we copyright the
+library, and (2) we offer you this license, which gives you legal
+permission to copy, distribute and/or modify the library.
+
+  To protect each distributor, we want to make it very clear that
+there is no warranty for the free library.  Also, if the library is
+modified by someone else and passed on, the recipients should know
+that what they have is not the original version, so that the original
+author's reputation will not be affected by problems that might be
+introduced by others.
+
+  Finally, software patents pose a constant threat to the existence of
+any free program.  We wish to make sure that a company cannot
+effectively restrict the users of a free program by obtaining a
+restrictive license from a patent holder.  Therefore, we insist that
+any patent license obtained for a version of the library must be
+consistent with the full freedom of use specified in this license.
+
+  Most GNU software, including some libraries, is covered by the
+ordinary GNU General Public License.  This license, the GNU Lesser
+General Public License, applies to certain designated libraries, and
+is quite different from the ordinary General Public License.  We use
+this license for certain libraries in order to permit linking those
+libraries into non-free programs.
+
+  When a program is linked with a library, whether statically or using
+a shared library, the combination of the two is legally speaking a
+combined work, a derivative of the original library.  The ordinary
+General Public License therefore permits such linking only if the
+entire combination fits its criteria of freedom.  The Lesser General
+Public License permits more lax criteria for linking other code with
+the library.
+
+  We call this license the "Lesser" General Public License because it
+does Less to protect the user's freedom than the ordinary General
+Public License.  It also provides other free software developers Less
+of an advantage over competing non-free programs.  These disadvantages
+are the reason we use the ordinary General Public License for many
+libraries.  However, the Lesser license provides advantages in certain
+special circumstances.
+
+  For example, on rare occasions, there may be a special need to
+encourage the widest possible use of a certain library, so that it becomes
+a de-facto standard.  To achieve this, non-free programs must be
+allowed to use the library.  A more frequent case is that a free
+library does the same job as widely used non-free libraries.  In this
+case, there is little to gain by limiting the free library to free
+software only, so we use the Lesser General Public License.
+
+  In other cases, permission to use a particular library in non-free
+programs enables a greater number of people to use a large body of
+free software.  For example, permission to use the GNU C Library in
+non-free programs enables many more people to use the whole GNU
+operating system, as well as its variant, the GNU/Linux operating
+system.
+
+  Although the Lesser General Public License is Less protective of the
+users' freedom, it does ensure that the user of a program that is
+linked with the Library has the freedom and the wherewithal to run
+that program using a modified version of the Library.
+
+  The precise terms and conditions for copying, distribution and
+modification follow.  Pay close attention to the difference between a
+"work based on the library" and a "work that uses the library".  The
+former contains code derived from the library, whereas the latter must
+be combined with the library in order to run.
+
+                  GNU LESSER GENERAL PUBLIC LICENSE
+   TERMS AND CONDITIONS FOR COPYING, DISTRIBUTION AND MODIFICATION
+
+  0. This License Agreement applies to any software library or other
+program which contains a notice placed by the copyright holder or
+other authorized party saying it may be distributed under the terms of
+this Lesser General Public License (also called "this License").
+Each licensee is addressed as "you".
+
+  A "library" means a collection of software functions and/or data
+prepared so as to be conveniently linked with application programs
+(which use some of those functions and data) to form executables.
+
+  The "Library", below, refers to any such software library or work
+which has been distributed under these terms.  A "work based on the
+Library" means either the Library or any derivative work under
+copyright law: that is to say, a work containing the Library or a
+portion of it, either verbatim or with modifications and/or translated
+straightforwardly into another language.  (Hereinafter, translation is
+included without limitation in the term "modification".)
+
+  "Source code" for a work means the preferred form of the work for
+making modifications to it.  For a library, complete source code means
+all the source code for all modules it contains, plus any associated
+interface definition files, plus the scripts used to control compilation
+and installation of the library.
+
+  Activities other than copying, distribution and modification are not
+covered by this License; they are outside its scope.  The act of
+running a program using the Library is not restricted, and output from
+such a program is covered only if its contents constitute a work based
+on the Library (independent of the use of the Library in a tool for
+writing it).  Whether that is true depends on what the Library does
+and what the program that uses the Library does.
+
+  1. You may copy and distribute verbatim copies of the Library's
+complete source code as you receive it, in any medium, provided that
+you conspicuously and appropriately publish on each copy an
+appropriate copyright notice and disclaimer of warranty; keep intact
+all the notices that refer to this License and to the absence of any
+warranty; and distribute a copy of this License along with the
+Library.
+
+  You may charge a fee for the physical act of transferring a copy,
+and you may at your option offer warranty protection in exchange for a
+fee.
+
+  2. You may modify your copy or copies of the Library or any portion
+of it, thus forming a work based on the Library, and copy and
+distribute such modifications or work under the terms of Section 1
+above, provided that you also meet all of these conditions:
+
+    a) The modified work must itself be a software library.
+
+    b) You must cause the files modified to carry prominent notices
+    stating that you changed the files and the date of any change.
+
+    c) You must cause the whole of the work to be licensed at no
+    charge to all third parties under the terms of this License.
+
+    d) If a facility in the modified Library refers to a function or a
+    table of data to be supplied by an application program that uses
+    the facility, other than as an argument passed when the facility
+    is invoked, then you must make a good faith effort to ensure that,
+    in the event an application does not supply such function or
+    table, the facility still operates, and performs whatever part of
+    its purpose remains meaningful.
+
+    (For example, a function in a library to compute square roots has
+    a purpose that is entirely well-defined independent of the
+    application.  Therefore, Subsection 2d requires that any
+    application-supplied function or table used by this function must
+    be optional: if the application does not supply it, the square
+    root function must still compute square roots.)
+
+These requirements apply to the modified work as a whole.  If
+identifiable sections of that work are not derived from the Library,
+and can be reasonably considered independent and separate works in
+themselves, then this License, and its terms, do not apply to those
+sections when you distribute them as separate works.  But when you
+distribute the same sections as part of a whole which is a work based
+on the Library, the distribution of the whole must be on the terms of
+this License, whose permissions for other licensees extend to the
+entire whole, and thus to each and every part regardless of who wrote
+it.
+
+Thus, it is not the intent of this section to claim rights or contest
+your rights to work written entirely by you; rather, the intent is to
+exercise the right to control the distribution of derivative or
+collective works based on the Library.
+
+In addition, mere aggregation of another work not based on the Library
+with the Library (or with a work based on the Library) on a volume of
+a storage or distribution medium does not bring the other work under
+the scope of this License.
+
+  3. You may opt to apply the terms of the ordinary GNU General Public
+License instead of this License to a given copy of the Library.  To do
+this, you must alter all the notices that refer to this License, so
+that they refer to the ordinary GNU General Public License, version 2,
+instead of to this License.  (If a newer version than version 2 of the
+ordinary GNU General Public License has appeared, then you can specify
+that version instead if you wish.)  Do not make any other change in
+these notices.
+
+  Once this change is made in a given copy, it is irreversible for
+that copy, so the ordinary GNU General Public License applies to all
+subsequent copies and derivative works made from that copy.
+
+  This option is useful when you wish to copy part of the code of
+the Library into a program that is not a library.
+
+  4. You may copy and distribute the Library (or a portion or
+derivative of it, under Section 2) in object code or executable form
+under the terms of Sections 1 and 2 above provided that you accompany
+it with the complete corresponding machine-readable source code, which
+must be distributed under the terms of Sections 1 and 2 above on a
+medium customarily used for software interchange.
+
+  If distribution of object code is made by offering access to copy
+from a designated place, then offering equivalent access to copy the
+source code from the same place satisfies the requirement to
+distribute the source code, even though third parties are not
+compelled to copy the source along with the object code.
+
+  5. A program that contains no derivative of any portion of the
+Library, but is designed to work with the Library by being compiled or
+linked with it, is called a "work that uses the Library".  Such a
+work, in isolation, is not a derivative work of the Library, and
+therefore falls outside the scope of this License.
+
+  However, linking a "work that uses the Library" with the Library
+creates an executable that is a derivative of the Library (because it
+contains portions of the Library), rather than a "work that uses the
+library".  The executable is therefore covered by this License.
+Section 6 states terms for distribution of such executables.
+
+  When a "work that uses the Library" uses material from a header file
+that is part of the Library, the object code for the work may be a
+derivative work of the Library even though the source code is not.
+Whether this is true is especially significant if the work can be
+linked without the Library, or if the work is itself a library.  The
+threshold for this to be true is not precisely defined by law.
+
+  If such an object file uses only numerical parameters, data
+structure layouts and accessors, and small macros and small inline
+functions (ten lines or less in length), then the use of the object
+file is unrestricted, regardless of whether it is legally a derivative
+work.  (Executables containing this object code plus portions of the
+Library will still fall under Section 6.)
+
+  Otherwise, if the work is a derivative of the Library, you may
+distribute the object code for the work under the terms of Section 6.
+Any executables containing that work also fall under Section 6,
+whether or not they are linked directly with the Library itself.
+
+  6. As an exception to the Sections above, you may also combine or
+link a "work that uses the Library" with the Library to produce a
+work containing portions of the Library, and distribute that work
+under terms of your choice, provided that the terms permit
+modification of the work for the customer's own use and reverse
+engineering for debugging such modifications.
+
+  You must give prominent notice with each copy of the work that the
+Library is used in it and that the Library and its use are covered by
+this License.  You must supply a copy of this License.  If the work
+during execution displays copyright notices, you must include the
+copyright notice for the Library among them, as well as a reference
+directing the user to the copy of this License.  Also, you must do one
+of these things:
+
+    a) Accompany the work with the complete corresponding
+    machine-readable source code for the Library including whatever
+    changes were used in the work (which must be distributed under
+    Sections 1 and 2 above); and, if the work is an executable linked
+    with the Library, with the complete machine-readable "work that
+    uses the Library", as object code and/or source code, so that the
+    user can modify the Library and then relink to produce a modified
+    executable containing the modified Library.  (It is understood
+    that the user who changes the contents of definitions files in the
+    Library will not necessarily be able to recompile the application
+    to use the modified definitions.)
+
+    b) Use a suitable shared library mechanism for linking with the
+    Library.  A suitable mechanism is one that (1) uses at run time a
+    copy of the library already present on the user's computer system,
+    rather than copying library functions into the executable, and (2)
+    will operate properly with a modified version of the library, if
+    the user installs one, as long as the modified version is
+    interface-compatible with the version that the work was made with.
+
+    c) Accompany the work with a written offer, valid for at
+    least three years, to give the same user the materials
+    specified in Subsection 6a, above, for a charge no more
+    than the cost of performing this distribution.
+
+    d) If distribution of the work is made by offering access to copy
+    from a designated place, offer equivalent access to copy the above
+    specified materials from the same place.
+
+    e) Verify that the user has already received a copy of these
+    materials or that you have already sent this user a copy.
+
+  For an executable, the required form of the "work that uses the
+Library" must include any data and utility programs needed for
+reproducing the executable from it.  However, as a special exception,
+the materials to be distributed need not include anything that is
+normally distributed (in either source or binary form) with the major
+components (compiler, kernel, and so on) of the operating system on
+which the executable runs, unless that component itself accompanies
+the executable.
+
+  It may happen that this requirement contradicts the license
+restrictions of other proprietary libraries that do not normally
+accompany the operating system.  Such a contradiction means you cannot
+use both them and the Library together in an executable that you
+distribute.
+
+  7. You may place library facilities that are a work based on the
+Library side-by-side in a single library together with other library
+facilities not covered by this License, and distribute such a combined
+library, provided that the separate distribution of the work based on
+the Library and of the other library facilities is otherwise
+permitted, and provided that you do these two things:
+
+    a) Accompany the combined library with a copy of the same work
+    based on the Library, uncombined with any other library
+    facilities.  This must be distributed under the terms of the
+    Sections above.
+
+    b) Give prominent notice with the combined library of the fact
+    that part of it is a work based on the Library, and explaining
+    where to find the accompanying uncombined form of the same work.
+
+  8. You may not copy, modify, sublicense, link with, or distribute
+the Library except as expressly provided under this License.  Any
+attempt otherwise to copy, modify, sublicense, link with, or
+distribute the Library is void, and will automatically terminate your
+rights under this License.  However, parties who have received copies,
+or rights, from you under this License will not have their licenses
+terminated so long as such parties remain in full compliance.
+
+  9. You are not required to accept this License, since you have not
+signed it.  However, nothing else grants you permission to modify or
+distribute the Library or its derivative works.  These actions are
+prohibited by law if you do not accept this License.  Therefore, by
+modifying or distributing the Library (or any work based on the
+Library), you indicate your acceptance of this License to do so, and
+all its terms and conditions for copying, distributing or modifying
+the Library or works based on it.
+
+  10. Each time you redistribute the Library (or any work based on the
+Library), the recipient automatically receives a license from the
+original licensor to copy, distribute, link with or modify the Library
+subject to these terms and conditions.  You may not impose any further
+restrictions on the recipients' exercise of the rights granted herein.
+You are not responsible for enforcing compliance by third parties with
+this License.
+
+  11. If, as a consequence of a court judgment or allegation of patent
+infringement or for any other reason (not limited to patent issues),
+conditions are imposed on you (whether by court order, agreement or
+otherwise) that contradict the conditions of this License, they do not
+excuse you from the conditions of this License.  If you cannot
+distribute so as to satisfy simultaneously your obligations under this
+License and any other pertinent obligations, then as a consequence you
+may not distribute the Library at all.  For example, if a patent
+license would not permit royalty-free redistribution of the Library by
+all those who receive copies directly or indirectly through you, then
+the only way you could satisfy both it and this License would be to
+refrain entirely from distribution of the Library.
+
+If any portion of this section is held invalid or unenforceable under any
+particular circumstance, the balance of the section is intended to apply,
+and the section as a whole is intended to apply in other circumstances.
+
+It is not the purpose of this section to induce you to infringe any
+patents or other property right claims or to contest validity of any
+such claims; this section has the sole purpose of protecting the
+integrity of the free software distribution system which is
+implemented by public license practices.  Many people have made
+generous contributions to the wide range of software distributed
+through that system in reliance on consistent application of that
+system; it is up to the author/donor to decide if he or she is willing
+to distribute software through any other system and a licensee cannot
+impose that choice.
+
+This section is intended to make thoroughly clear what is believed to
+be a consequence of the rest of this License.
+
+  12. If the distribution and/or use of the Library is restricted in
+certain countries either by patents or by copyrighted interfaces, the
+original copyright holder who places the Library under this License may add
+an explicit geographical distribution limitation excluding those countries,
+so that distribution is permitted only in or among countries not thus
+excluded.  In such case, this License incorporates the limitation as if
+written in the body of this License.
+
+  13. The Free Software Foundation may publish revised and/or new
+versions of the Lesser General Public License from time to time.
+Such new versions will be similar in spirit to the present version,
+but may differ in detail to address new problems or concerns.
+
+Each version is given a distinguishing version number.  If the Library
+specifies a version number of this License which applies to it and
+"any later version", you have the option of following the terms and
+conditions either of that version or of any later version published by
+the Free Software Foundation.  If the Library does not specify a
+license version number, you may choose any version ever published by
+the Free Software Foundation.
+
+  14. If you wish to incorporate parts of the Library into other free
+programs whose distribution conditions are incompatible with these,
+write to the author to ask for permission.  For software which is
+copyrighted by the Free Software Foundation, write to the Free
+Software Foundation; we sometimes make exceptions for this.  Our
+decision will be guided by the two goals of preserving the free status
+of all derivatives of our free software and of promoting the sharing
+and reuse of software generally.
+
+                            NO WARRANTY
+
+  15. BECAUSE THE LIBRARY IS LICENSED FREE OF CHARGE, THERE IS NO
+WARRANTY FOR THE LIBRARY, TO THE EXTENT PERMITTED BY APPLICABLE LAW.
+EXCEPT WHEN OTHERWISE STATED IN WRITING THE COPYRIGHT HOLDERS AND/OR
+OTHER PARTIES PROVIDE THE LIBRARY "AS IS" WITHOUT WARRANTY OF ANY
+KIND, EITHER EXPRESSED OR IMPLIED, INCLUDING, BUT NOT LIMITED TO, THE
+IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
+PURPOSE.  THE ENTIRE RISK AS TO THE QUALITY AND PERFORMANCE OF THE
+LIBRARY IS WITH YOU.  SHOULD THE LIBRARY PROVE DEFECTIVE, YOU ASSUME
+THE COST OF ALL NECESSARY SERVICING, REPAIR OR CORRECTION.
+
+  16. IN NO EVENT UNLESS REQUIRED BY APPLICABLE LAW OR AGREED TO IN
+WRITING WILL ANY COPYRIGHT HOLDER, OR ANY OTHER PARTY WHO MAY MODIFY
+AND/OR REDISTRIBUTE THE LIBRARY AS PERMITTED ABOVE, BE LIABLE TO YOU
+FOR DAMAGES, INCLUDING ANY GENERAL, SPECIAL, INCIDENTAL OR
+CONSEQUENTIAL DAMAGES ARISING OUT OF THE USE OR INABILITY TO USE THE
+LIBRARY (INCLUDING BUT NOT LIMITED TO LOSS OF DATA OR DATA BEING
+RENDERED INACCURATE OR LOSSES SUSTAINED BY YOU OR THIRD PARTIES OR A
+FAILURE OF THE LIBRARY TO OPERATE WITH ANY OTHER SOFTWARE), EVEN IF
+SUCH HOLDER OR OTHER PARTY HAS BEEN ADVISED OF THE POSSIBILITY OF SUCH
+DAMAGES.
+
+                     END OF TERMS AND CONDITIONS
+
+           How to Apply These Terms to Your New Libraries
+
+  If you develop a new library, and you want it to be of the greatest
+possible use to the public, we recommend making it free software that
+everyone can redistribute and change.  You can do so by permitting
+redistribution under these terms (or, alternatively, under the terms of the
+ordinary General Public License).
+
+  To apply these terms, attach the following notices to the library.  It is
+safest to attach them to the start of each source file to most effectively
+convey the exclusion of warranty; and each file should have at least the
+"copyright" line and a pointer to where the full notice is found.
+
+    <one line to give the library's name and a brief idea of what it does.>
+    Copyright (C) <year>  <name of author>
+
+    This library 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 2.1 of the License, or (at your option) any later version.
+
+    This library 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 for more details.
+
+    You should have received a copy of the GNU Lesser General Public
+    License along with this library; if not, write to the Free Software
+    Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA  02110-1301  USA
+
+Also add information on how to contact you by electronic and paper mail.
+
+You should also get your employer (if you work as a programmer) or your
+school, if any, to sign a "copyright disclaimer" for the library, if
+necessary.  Here is a sample; alter the names:
+
+  Yoyodyne, Inc., hereby disclaims all copyright interest in the
+  library `Frob' (a library for tweaking knobs) written by James Random Hacker.
+
+  <signature of Ty Coon>, 1 April 1990
+  Ty Coon, President of Vice
+
+That's all there is to it!

COPYING.MINPACK

@@ -0,0 +1,52 @@
+Minpack Copyright Notice (1999) University of Chicago.  All rights reserved
+
+Redistribution and use in source and binary forms, with or
+without modification, are permitted provided that the
+following conditions are met:
+
+1. Redistributions of source code must retain the above
+copyright notice, this list of conditions and the following
+disclaimer.
+
+2. Redistributions in binary form must reproduce the above
+copyright notice, this list of conditions and the following
+disclaimer in the documentation and/or other materials
+provided with the distribution.
+
+3. The end-user documentation included with the
+redistribution, if any, must include the following
+acknowledgment:
+
+   "This product includes software developed by the
+   University of Chicago, as Operator of Argonne National
+   Laboratory.
+
+Alternately, this acknowledgment may appear in the software
+itself, if and wherever such third-party acknowledgments
+normally appear.
+
+4. WARRANTY DISCLAIMER. THE SOFTWARE IS SUPPLIED "AS IS"
+WITHOUT WARRANTY OF ANY KIND. THE COPYRIGHT HOLDER, THE
+UNITED STATES, THE UNITED STATES DEPARTMENT OF ENERGY, AND
+THEIR EMPLOYEES: (1) DISCLAIM ANY WARRANTIES, EXPRESS OR
+IMPLIED, INCLUDING BUT NOT LIMITED TO ANY IMPLIED WARRANTIES
+OF MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE, TITLE
+OR NON-INFRINGEMENT, (2) DO NOT ASSUME ANY LEGAL LIABILITY
+OR RESPONSIBILITY FOR THE ACCURACY, COMPLETENESS, OR
+USEFULNESS OF THE SOFTWARE, (3) DO NOT REPRESENT THAT USE OF
+THE SOFTWARE WOULD NOT INFRINGE PRIVATELY OWNED RIGHTS, (4)
+DO NOT WARRANT THAT THE SOFTWARE WILL FUNCTION
+UNINTERRUPTED, THAT IT IS ERROR-FREE OR THAT ANY ERRORS WILL
+BE CORRECTED.
+
+5. LIMITATION OF LIABILITY. IN NO EVENT WILL THE COPYRIGHT
+HOLDER, THE UNITED STATES, THE UNITED STATES DEPARTMENT OF
+ENERGY, OR THEIR EMPLOYEES: BE LIABLE FOR ANY INDIRECT,
+INCIDENTAL, CONSEQUENTIAL, SPECIAL OR PUNITIVE DAMAGES OF
+ANY KIND OR NATURE, INCLUDING BUT NOT LIMITED TO LOSS OF
+PROFITS OR LOSS OF DATA, FOR ANY REASON WHATSOEVER, WHETHER
+SUCH LIABILITY IS ASSERTED ON THE BASIS OF CONTRACT, TORT
+(INCLUDING NEGLIGENCE OR STRICT LIABILITY), OR OTHERWISE,
+EVEN IF ANY OF SAID PARTIES HAS BEEN WARNED OF THE
+POSSIBILITY OF SUCH LOSS OR DAMAGES.
+

COPYING.MPL2

@@ -0,0 +1,373 @@
+Mozilla Public License Version 2.0
+==================================
+
+1. Definitions
+--------------
+
+1.1. "Contributor"
+    means each individual or legal entity that creates, contributes to
+    the creation of, or owns Covered Software.
+
+1.2. "Contributor Version"
+    means the combination of the Contributions of others (if any) used
+    by a Contributor and that particular Contributor's Contribution.
+
+1.3. "Contribution"
+    means Covered Software of a particular Contributor.
+
+1.4. "Covered Software"
+    means Source Code Form to which the initial Contributor has attached
+    the notice in Exhibit A, the Executable Form of such Source Code
+    Form, and Modifications of such Source Code Form, in each case
+    including portions thereof.
+
+1.5. "Incompatible With Secondary Licenses"
+    means
+
+    (a) that the initial Contributor has attached the notice described
+        in Exhibit B to the Covered Software; or
+
+    (b) that the Covered Software was made available under the terms of
+        version 1.1 or earlier of the License, but not also under the
+        terms of a Secondary License.
+
+1.6. "Executable Form"
+    means any form of the work other than Source Code Form.
+
+1.7. "Larger Work"
+    means a work that combines Covered Software with other material, in 
+    a separate file or files, that is not Covered Software.
+
+1.8. "License"
+    means this document.
+
+1.9. "Licensable"
+    means having the right to grant, to the maximum extent possible,
+    whether at the time of the initial grant or subsequently, any and
+    all of the rights conveyed by this License.
+
+1.10. "Modifications"
+    means any of the following:
+
+    (a) any file in Source Code Form that results from an addition to,
+        deletion from, or modification of the contents of Covered
+        Software; or
+
+    (b) any new file in Source Code Form that contains any Covered
+        Software.
+
+1.11. "Patent Claims" of a Contributor
+    means any patent claim(s), including without limitation, method,
+    process, and apparatus claims, in any patent Licensable by such
+    Contributor that would be infringed, but for the grant of the
+    License, by the making, using, selling, offering for sale, having
+    made, import, or transfer of either its Contributions or its
+    Contributor Version.
+
+1.12. "Secondary License"
+    means either the GNU General Public License, Version 2.0, the GNU
+    Lesser General Public License, Version 2.1, the GNU Affero General
+    Public License, Version 3.0, or any later versions of those
+    licenses.
+
+1.13. "Source Code Form"
+    means the form of the work preferred for making modifications.
+
+1.14. "You" (or "Your")
+    means an individual or a legal entity exercising rights under this
+    License. For legal entities, "You" includes any entity that
+    controls, is controlled by, or is under common control with You. For
+    purposes of this definition, "control" means (a) the power, direct
+    or indirect, to cause the direction or management of such entity,
+    whether by contract or otherwise, or (b) ownership of more than
+    fifty percent (50%) of the outstanding shares or beneficial
+    ownership of such entity.
+
+2. License Grants and Conditions
+--------------------------------
+
+2.1. Grants
+
+Each Contributor hereby grants You a world-wide, royalty-free,
+non-exclusive license:
+
+(a) under intellectual property rights (other than patent or trademark)
+    Licensable by such Contributor to use, reproduce, make available,
+    modify, display, perform, distribute, and otherwise exploit its
+    Contributions, either on an unmodified basis, with Modifications, or
+    as part of a Larger Work; and
+
+(b) under Patent Claims of such Contributor to make, use, sell, offer
+    for sale, have made, import, and otherwise transfer either its
+    Contributions or its Contributor Version.
+
+2.2. Effective Date
+
+The licenses granted in Section 2.1 with respect to any Contribution
+become effective for each Contribution on the date the Contributor first
+distributes such Contribution.
+
+2.3. Limitations on Grant Scope
+
+The licenses granted in this Section 2 are the only rights granted under
+this License. No additional rights or licenses will be implied from the
+distribution or licensing of Covered Software under this License.
+Notwithstanding Section 2.1(b) above, no patent license is granted by a
+Contributor:
+
+(a) for any code that a Contributor has removed from Covered Software;
+    or
+
+(b) for infringements caused by: (i) Your and any other third party's
+    modifications of Covered Software, or (ii) the combination of its
+    Contributions with other software (except as part of its Contributor
+    Version); or
+
+(c) under Patent Claims infringed by Covered Software in the absence of
+    its Contributions.
+
+This License does not grant any rights in the trademarks, service marks,
+or logos of any Contributor (except as may be necessary to comply with
+the notice requirements in Section 3.4).
+
+2.4. Subsequent Licenses
+
+No Contributor makes additional grants as a result of Your choice to
+distribute the Covered Software under a subsequent version of this
+License (see Section 10.2) or under the terms of a Secondary License (if
+permitted under the terms of Section 3.3).
+
+2.5. Representation
+
+Each Contributor represents that the Contributor believes its
+Contributions are its original creation(s) or it has sufficient rights
+to grant the rights to its Contributions conveyed by this License.
+
+2.6. Fair Use
+
+This License is not intended to limit any rights You have under
+applicable copyright doctrines of fair use, fair dealing, or other
+equivalents.
+
+2.7. Conditions
+
+Sections 3.1, 3.2, 3.3, and 3.4 are conditions of the licenses granted
+in Section 2.1.
+
+3. Responsibilities
+-------------------
+
+3.1. Distribution of Source Form
+
+All distribution of Covered Software in Source Code Form, including any
+Modifications that You create or to which You contribute, must be under
+the terms of this License. You must inform recipients that the Source
+Code Form of the Covered Software is governed by the terms of this
+License, and how they can obtain a copy of this License. You may not
+attempt to alter or restrict the recipients' rights in the Source Code
+Form.
+
+3.2. Distribution of Executable Form
+
+If You distribute Covered Software in Executable Form then:
+
+(a) such Covered Software must also be made available in Source Code
+    Form, as described in Section 3.1, and You must inform recipients of
+    the Executable Form how they can obtain a copy of such Source Code
+    Form by reasonable means in a timely manner, at a charge no more
+    than the cost of distribution to the recipient; and
+
+(b) You may distribute such Executable Form under the terms of this
+    License, or sublicense it under different terms, provided that the
+    license for the Executable Form does not attempt to limit or alter
+    the recipients' rights in the Source Code Form under this License.
+
+3.3. Distribution of a Larger Work
+
+You may create and distribute a Larger Work under terms of Your choice,
+provided that You also comply with the requirements of this License for
+the Covered Software. If the Larger Work is a combination of Covered
+Software with a work governed by one or more Secondary Licenses, and the
+Covered Software is not Incompatible With Secondary Licenses, this
+License permits You to additionally distribute such Covered Software
+under the terms of such Secondary License(s), so that the recipient of
+the Larger Work may, at their option, further distribute the Covered
+Software under the terms of either this License or such Secondary
+License(s).
+
+3.4. Notices
+
+You may not remove or alter the substance of any license notices
+(including copyright notices, patent notices, disclaimers of warranty,
+or limitations of liability) contained within the Source Code Form of
+the Covered Software, except that You may alter any license notices to
+the extent required to remedy known factual inaccuracies.
+
+3.5. Application of Additional Terms
+
+You may choose to offer, and to charge a fee for, warranty, support,
+indemnity or liability obligations to one or more recipients of Covered
+Software. However, You may do so only on Your own behalf, and not on
+behalf of any Contributor. You must make it absolutely clear that any
+such warranty, support, indemnity, or liability obligation is offered by
+You alone, and You hereby agree to indemnify every Contributor for any
+liability incurred by such Contributor as a result of warranty, support,
+indemnity or liability terms You offer. You may include additional
+disclaimers of warranty and limitations of liability specific to any
+jurisdiction.
+
+4. Inability to Comply Due to Statute or Regulation
+---------------------------------------------------
+
+If it is impossible for You to comply with any of the terms of this
+License with respect to some or all of the Covered Software due to
+statute, judicial order, or regulation then You must: (a) comply with
+the terms of this License to the maximum extent possible; and (b)
+describe the limitations and the code they affect. Such description must
+be placed in a text file included with all distributions of the Covered
+Software under this License. Except to the extent prohibited by statute
+or regulation, such description must be sufficiently detailed for a
+recipient of ordinary skill to be able to understand it.
+
+5. Termination
+--------------
+
+5.1. The rights granted under this License will terminate automatically
+if You fail to comply with any of its terms. However, if You become
+compliant, then the rights granted under this License from a particular
+Contributor are reinstated (a) provisionally, unless and until such
+Contributor explicitly and finally terminates Your grants, and (b) on an
+ongoing basis, if such Contributor fails to notify You of the
+non-compliance by some reasonable means prior to 60 days after You have
+come back into compliance. Moreover, Your grants from a particular
+Contributor are reinstated on an ongoing basis if such Contributor
+notifies You of the non-compliance by some reasonable means, this is the
+first time You have received notice of non-compliance with this License
+from such Contributor, and You become compliant prior to 30 days after
+Your receipt of the notice.
+
+5.2. If You initiate litigation against any entity by asserting a patent
+infringement claim (excluding declaratory judgment actions,
+counter-claims, and cross-claims) alleging that a Contributor Version
+directly or indirectly infringes any patent, then the rights granted to
+You by any and all Contributors for the Covered Software under Section
+2.1 of this License shall terminate.
+
+5.3. In the event of termination under Sections 5.1 or 5.2 above, all
+end user license agreements (excluding distributors and resellers) which
+have been validly granted by You or Your distributors under this License
+prior to termination shall survive termination.
+
+************************************************************************
+*                                                                      *
+*  6. Disclaimer of Warranty                                           *
+*  -------------------------                                           *
+*                                                                      *
+*  Covered Software is provided under this License on an "as is"       *
+*  basis, without warranty of any kind, either expressed, implied, or  *
+*  statutory, including, without limitation, warranties that the       *
+*  Covered Software is free of defects, merchantable, fit for a        *
+*  particular purpose or non-infringing. The entire risk as to the     *
+*  quality and performance of the Covered Software is with You.        *
+*  Should any Covered Software prove defective in any respect, You     *
+*  (not any Contributor) assume the cost of any necessary servicing,   *
+*  repair, or correction. This disclaimer of warranty constitutes an   *
+*  essential part of this License. No use of any Covered Software is   *
+*  authorized under this License except under this disclaimer.         *
+*                                                                      *
+************************************************************************
+
+************************************************************************
+*                                                                      *
+*  7. Limitation of Liability                                          *
+*  --------------------------                                          *
+*                                                                      *
+*  Under no circumstances and under no legal theory, whether tort      *
+*  (including negligence), contract, or otherwise, shall any           *
+*  Contributor, or anyone who distributes Covered Software as          *
+*  permitted above, be liable to You for any direct, indirect,         *
+*  special, incidental, or consequential damages of any character      *
+*  including, without limitation, damages for lost profits, loss of    *
+*  goodwill, work stoppage, computer failure or malfunction, or any    *
+*  and all other commercial damages or losses, even if such party      *
+*  shall have been informed of the possibility of such damages. This   *
+*  limitation of liability shall not apply to liability for death or   *
+*  personal injury resulting from such party's negligence to the       *
+*  extent applicable law prohibits such limitation. Some               *
+*  jurisdictions do not allow the exclusion or limitation of           *
+*  incidental or consequential damages, so this exclusion and          *
+*  limitation may not apply to You.                                    *
+*                                                                      *
+************************************************************************
+
+8. Litigation
+-------------
+
+Any litigation relating to this License may be brought only in the
+courts of a jurisdiction where the defendant maintains its principal
+place of business and such litigation shall be governed by laws of that
+jurisdiction, without reference to its conflict-of-law provisions.
+Nothing in this Section shall prevent a party's ability to bring
+cross-claims or counter-claims.
+
+9. Miscellaneous
+----------------
+
+This License represents the complete agreement concerning the subject
+matter hereof. If any provision of this License is held to be
+unenforceable, such provision shall be reformed only to the extent
+necessary to make it enforceable. Any law or regulation which provides
+that the language of a contract shall be construed against the drafter
+shall not be used to construe this License against a Contributor.
+
+10. Versions of the License
+---------------------------
+
+10.1. New Versions
+
+Mozilla Foundation is the license steward. Except as provided in Section
+10.3, no one other than the license steward has the right to modify or
+publish new versions of this License. Each version will be given a
+distinguishing version number.
+
+10.2. Effect of New Versions
+
+You may distribute the Covered Software under the terms of the version
+of the License under which You originally received the Covered Software,
+or under the terms of any subsequent version published by the license
+steward.
+
+10.3. Modified Versions
+
+If you create software not governed by this License, and you want to
+create a new license for such software, you may create and use a
+modified version of this License if you rename the license and remove
+any references to the name of the license steward (except to note that
+such modified license differs from this License).
+
+10.4. Distributing Source Code Form that is Incompatible With Secondary
+Licenses
+
+If You choose to distribute Source Code Form that is Incompatible With
+Secondary Licenses under the terms of this version of the License, the
+notice described in Exhibit B of this License must be attached.
+
+Exhibit A - Source Code Form License Notice
+-------------------------------------------
+
+  This Source Code Form is subject to the terms of the Mozilla Public
+  License, v. 2.0. If a copy of the MPL was not distributed with this
+  file, You can obtain one at http://mozilla.org/MPL/2.0/.
+
+If it is not possible or desirable to put the notice in a particular
+file, then You may include the notice in a location (such as a LICENSE
+file in a relevant directory) where a recipient would be likely to look
+for such a notice.
+
+You may add additional accurate notices of copyright ownership.
+
+Exhibit B - "Incompatible With Secondary Licenses" Notice
+---------------------------------------------------------
+
+  This Source Code Form is "Incompatible With Secondary Licenses", as
+  defined by the Mozilla Public License, v. 2.0.

COPYING.README

@@ -0,0 +1,18 @@
+Eigen is primarily MPL2 licensed. See COPYING.MPL2 and these links:
+  http://www.mozilla.org/MPL/2.0/
+  http://www.mozilla.org/MPL/2.0/FAQ.html
+
+Some files contain third-party code under BSD or LGPL licenses, whence the other
+COPYING.* files here.
+
+All the LGPL code is either LGPL 2.1-only, or LGPL 2.1-or-later.
+For this reason, the COPYING.LGPL file contains the LGPL 2.1 text.
+
+If you want to guarantee that the Eigen code that you are #including is licensed
+under the MPL2 and possibly more permissive licenses (like BSD), #define this
+preprocessor symbol:
+  EIGEN_MPL2_ONLY
+For example, with most compilers, you could add this to your project CXXFLAGS:
+  -DEIGEN_MPL2_ONLY
+This will cause a compilation error to be generated if you #include any code that is
+LGPL licensed.

CTestConfig.cmake

@@ -4,10 +4,10 @@
 ## # The following are required to uses Dart and the Cdash dashboard
 ##   ENABLE_TESTING()
 ##   INCLUDE(CTest)
-set(CTEST_PROJECT_NAME "Eigen 2.0")
-set(CTEST_NIGHTLY_START_TIME "06:00:00 UTC")
+set(CTEST_PROJECT_NAME "Eigen")
+set(CTEST_NIGHTLY_START_TIME "00:00:00 UTC")
 
 set(CTEST_DROP_METHOD "http")
-set(CTEST_DROP_SITE "eigen.tuxfamily.org")
-set(CTEST_DROP_LOCATION "/CDash/submit.php?project=Eigen+2.0")
+set(CTEST_DROP_SITE "my.cdash.org")
+set(CTEST_DROP_LOCATION "/submit.php?project=Eigen")
 set(CTEST_DROP_SITE_CDASH TRUE)

CTestCustom.cmake.in

@@ -0,0 +1,4 @@
+
+set(CTEST_CUSTOM_MAXIMUM_NUMBER_OF_WARNINGS "2000")
+set(CTEST_CUSTOM_MAXIMUM_NUMBER_OF_ERRORS   "2000")
+list(APPEND CTEST_CUSTOM_ERROR_EXCEPTION    @EIGEN_CTEST_ERROR_EXCEPTION@)

Doxyfile

@@ -1,281 +0,0 @@
-# Doxyfile 1.5.3
-
-#---------------------------------------------------------------------------
-# Project related configuration options
-#---------------------------------------------------------------------------
-DOXYFILE_ENCODING      = UTF-8
-PROJECT_NAME           = Eigen
-PROJECT_NUMBER         = 2.0
-OUTPUT_DIRECTORY       = ./
-CREATE_SUBDIRS         = NO
-OUTPUT_LANGUAGE        = English
-BRIEF_MEMBER_DESC      = YES
-REPEAT_BRIEF           = YES
-ABBREVIATE_BRIEF       = "The $name class" \
-                         "The $name widget" \
-                         "The $name file" \
-                         is \
-                         provides \
-                         specifies \
-                         contains \
-                         represents \
-                         a \
-                         an \
-                         the
-ALWAYS_DETAILED_SEC    = NO
-INLINE_INHERITED_MEMB  = NO
-FULL_PATH_NAMES        = NO
-STRIP_FROM_PATH        = 
-STRIP_FROM_INC_PATH    = 
-SHORT_NAMES            = NO
-JAVADOC_AUTOBRIEF      = NO
-QT_AUTOBRIEF           = NO
-MULTILINE_CPP_IS_BRIEF = NO
-DETAILS_AT_TOP         = NO
-INHERIT_DOCS           = YES
-SEPARATE_MEMBER_PAGES  = NO
-TAB_SIZE               = 8
-ALIASES                = 
-OPTIMIZE_OUTPUT_FOR_C  = NO
-OPTIMIZE_OUTPUT_JAVA   = NO
-BUILTIN_STL_SUPPORT    = NO
-CPP_CLI_SUPPORT        = NO
-DISTRIBUTE_GROUP_DOC   = NO
-SUBGROUPING            = YES
-#---------------------------------------------------------------------------
-# Build related configuration options
-#---------------------------------------------------------------------------
-EXTRACT_ALL            = NO
-EXTRACT_PRIVATE        = NO
-EXTRACT_STATIC         = NO
-EXTRACT_LOCAL_CLASSES  = NO
-EXTRACT_LOCAL_METHODS  = NO
-EXTRACT_ANON_NSPACES   = NO
-HIDE_UNDOC_MEMBERS     = YES
-HIDE_UNDOC_CLASSES     = YES
-HIDE_FRIEND_COMPOUNDS  = YES
-HIDE_IN_BODY_DOCS      = NO
-INTERNAL_DOCS          = NO
-CASE_SENSE_NAMES       = YES
-HIDE_SCOPE_NAMES       = YES
-SHOW_INCLUDE_FILES     = YES
-INLINE_INFO            = YES
-SORT_MEMBER_DOCS       = YES
-SORT_BRIEF_DOCS        = NO
-SORT_BY_SCOPE_NAME     = NO
-GENERATE_TODOLIST      = YES
-GENERATE_TESTLIST      = YES
-GENERATE_BUGLIST       = YES
-GENERATE_DEPRECATEDLIST= YES
-ENABLED_SECTIONS       = 
-MAX_INITIALIZER_LINES  = 30
-SHOW_USED_FILES        = YES
-SHOW_DIRECTORIES       = NO
-FILE_VERSION_FILTER    = 
-#---------------------------------------------------------------------------
-# configuration options related to warning and progress messages
-#---------------------------------------------------------------------------
-QUIET                  = NO
-WARNINGS               = YES
-WARN_IF_UNDOCUMENTED   = YES
-WARN_IF_DOC_ERROR      = YES
-WARN_NO_PARAMDOC       = NO
-WARN_FORMAT            = "$file:$line: $text"
-WARN_LOGFILE           = 
-#---------------------------------------------------------------------------
-# configuration options related to the input files
-#---------------------------------------------------------------------------
-INPUT                  = ./
-INPUT_ENCODING         = UTF-8
-FILE_PATTERNS          = *.c \
-                         *.cc \
-                         *.cxx \
-                         *.cpp \
-                         *.c++ \
-                         *.d \
-                         *.java \
-                         *.ii \
-                         *.ixx \
-                         *.ipp \
-                         *.i++ \
-                         *.inl \
-                         *.h \
-                         *.hh \
-                         *.hxx \
-                         *.hpp \
-                         *.h++ \
-                         *.idl \
-                         *.odl \
-                         *.cs \
-                         *.php \
-                         *.php3 \
-                         *.inc \
-                         *.m \
-                         *.mm \
-                         *.dox \
-                         *.py \
-                         *.C \
-                         *.CC \
-                         *.C++ \
-                         *.II \
-                         *.I++ \
-                         *.H \
-                         *.HH \
-                         *.H++ \
-                         *.CS \
-                         *.PHP \
-                         *.PHP3 \
-                         *.M \
-                         *.MM \
-                         *.PY
-RECURSIVE              = NO
-EXCLUDE                = 
-EXCLUDE_SYMLINKS       = NO
-EXCLUDE_PATTERNS       = 
-EXCLUDE_SYMBOLS        = 
-EXAMPLE_PATH           = doc/examples/
-EXAMPLE_PATTERNS       = *
-EXAMPLE_RECURSIVE      = NO
-IMAGE_PATH             = 
-INPUT_FILTER           = 
-FILTER_PATTERNS        = 
-FILTER_SOURCE_FILES    = NO
-#---------------------------------------------------------------------------
-# configuration options related to source browsing
-#---------------------------------------------------------------------------
-SOURCE_BROWSER         = NO
-INLINE_SOURCES         = NO
-STRIP_CODE_COMMENTS    = YES
-REFERENCED_BY_RELATION = YES
-REFERENCES_RELATION    = YES
-REFERENCES_LINK_SOURCE = YES
-USE_HTAGS              = NO
-VERBATIM_HEADERS       = YES
-#---------------------------------------------------------------------------
-# configuration options related to the alphabetical class index
-#---------------------------------------------------------------------------
-ALPHABETICAL_INDEX     = NO
-COLS_IN_ALPHA_INDEX    = 5
-IGNORE_PREFIX          = 
-#---------------------------------------------------------------------------
-# configuration options related to the HTML output
-#---------------------------------------------------------------------------
-GENERATE_HTML          = YES
-HTML_OUTPUT            = html
-HTML_FILE_EXTENSION    = .html
-HTML_HEADER            = 
-HTML_FOOTER            = 
-HTML_STYLESHEET        = 
-HTML_ALIGN_MEMBERS     = YES
-GENERATE_HTMLHELP      = NO
-HTML_DYNAMIC_SECTIONS  = NO
-CHM_FILE               = 
-HHC_LOCATION           = 
-GENERATE_CHI           = NO
-BINARY_TOC             = NO
-TOC_EXPAND             = NO
-DISABLE_INDEX          = NO
-ENUM_VALUES_PER_LINE   = 4
-GENERATE_TREEVIEW      = NO
-TREEVIEW_WIDTH         = 250
-#---------------------------------------------------------------------------
-# configuration options related to the LaTeX output
-#---------------------------------------------------------------------------
-GENERATE_LATEX         = YES
-LATEX_OUTPUT           = latex
-LATEX_CMD_NAME         = latex
-MAKEINDEX_CMD_NAME     = makeindex
-COMPACT_LATEX          = NO
-PAPER_TYPE             = a4wide
-EXTRA_PACKAGES         = 
-LATEX_HEADER           = 
-PDF_HYPERLINKS         = NO
-USE_PDFLATEX           = NO
-LATEX_BATCHMODE        = NO
-LATEX_HIDE_INDICES     = NO
-#---------------------------------------------------------------------------
-# configuration options related to the RTF output
-#---------------------------------------------------------------------------
-GENERATE_RTF           = NO
-RTF_OUTPUT             = rtf
-COMPACT_RTF            = NO
-RTF_HYPERLINKS         = NO
-RTF_STYLESHEET_FILE    = 
-RTF_EXTENSIONS_FILE    = 
-#---------------------------------------------------------------------------
-# configuration options related to the man page output
-#---------------------------------------------------------------------------
-GENERATE_MAN           = NO
-MAN_OUTPUT             = man
-MAN_EXTENSION          = .3
-MAN_LINKS              = NO
-#---------------------------------------------------------------------------
-# configuration options related to the XML output
-#---------------------------------------------------------------------------
-GENERATE_XML           = NO
-XML_OUTPUT             = xml
-XML_SCHEMA             = 
-XML_DTD                = 
-XML_PROGRAMLISTING     = YES
-#---------------------------------------------------------------------------
-# configuration options for the AutoGen Definitions output
-#---------------------------------------------------------------------------
-GENERATE_AUTOGEN_DEF   = NO
-#---------------------------------------------------------------------------
-# configuration options related to the Perl module output
-#---------------------------------------------------------------------------
-GENERATE_PERLMOD       = NO
-PERLMOD_LATEX          = NO
-PERLMOD_PRETTY         = YES
-PERLMOD_MAKEVAR_PREFIX = 
-#---------------------------------------------------------------------------
-# Configuration options related to the preprocessor   
-#---------------------------------------------------------------------------
-ENABLE_PREPROCESSING   = YES
-MACRO_EXPANSION        = NO
-EXPAND_ONLY_PREDEF     = NO
-SEARCH_INCLUDES        = YES
-INCLUDE_PATH           = 
-INCLUDE_FILE_PATTERNS  = 
-PREDEFINED             = 
-EXPAND_AS_DEFINED      = 
-SKIP_FUNCTION_MACROS   = YES
-#---------------------------------------------------------------------------
-# Configuration::additions related to external references   
-#---------------------------------------------------------------------------
-TAGFILES               = 
-GENERATE_TAGFILE       = 
-ALLEXTERNALS           = NO
-EXTERNAL_GROUPS        = YES
-PERL_PATH              = /usr/bin/perl
-#---------------------------------------------------------------------------
-# Configuration options related to the dot tool   
-#---------------------------------------------------------------------------
-CLASS_DIAGRAMS         = YES
-MSCGEN_PATH            = 
-HIDE_UNDOC_RELATIONS   = YES
-HAVE_DOT               = YES
-CLASS_GRAPH            = YES
-COLLABORATION_GRAPH    = YES
-GROUP_GRAPHS           = YES
-UML_LOOK               = NO
-TEMPLATE_RELATIONS     = NO
-INCLUDE_GRAPH          = YES
-INCLUDED_BY_GRAPH      = YES
-CALL_GRAPH             = NO
-CALLER_GRAPH           = NO
-GRAPHICAL_HIERARCHY    = YES
-DIRECTORY_GRAPH        = YES
-DOT_IMAGE_FORMAT       = png
-DOT_PATH               = 
-DOTFILE_DIRS           = 
-DOT_GRAPH_MAX_NODES    = 50
-MAX_DOT_GRAPH_DEPTH    = 1000
-DOT_TRANSPARENT        = NO
-DOT_MULTI_TARGETS      = NO
-GENERATE_LEGEND        = YES
-DOT_CLEANUP            = YES
-#---------------------------------------------------------------------------
-# Configuration::additions related to the search engine   
-#---------------------------------------------------------------------------
-SEARCHENGINE           = NO

Eigen/Array

@@ -1,39 +0,0 @@
-#ifndef EIGEN_ARRAY_MODULE_H
-#define EIGEN_ARRAY_MODULE_H
-
-#include "Core"
-
-#include "src/Core/util/DisableMSVCWarnings.h"
-
-namespace Eigen {
-
-/** \defgroup Array_Module Array module
-  * This module provides several handy features to manipulate matrices as simple array of values.
-  * In addition to listed classes, it defines various methods of the Cwise interface
-  * (accessible from MatrixBase::cwise()), including:
-  *  - matrix-scalar sum,
-  *  - coeff-wise comparison operators,
-  *  - sin, cos, sqrt, pow, exp, log, square, cube, inverse (reciprocal).
-  *
-  * This module also provides various MatrixBase methods, including:
-  *  - \ref MatrixBase::all() "all", \ref MatrixBase::any() "any",
-  *  - \ref MatrixBase::Random() "random matrix initialization"
-  *
-  * \code
-  * #include <Eigen/Array>
-  * \endcode
-  */
-
-#include "src/Array/CwiseOperators.h"
-#include "src/Array/Functors.h"
-#include "src/Array/BooleanRedux.h"
-#include "src/Array/Select.h"
-#include "src/Array/PartialRedux.h"
-#include "src/Array/Random.h"
-#include "src/Array/Norms.h"
-
-} // namespace Eigen
-
-#include "src/Core/util/EnableMSVCWarnings.h"
-
-#endif // EIGEN_ARRAY_MODULE_H

Eigen/CMakeLists.txt

@@ -1,34 +1,19 @@
-set(Eigen_HEADERS Core LU Cholesky QR Geometry Sparse Array SVD LeastSquares QtAlignedMalloc StdVector)
+include(RegexUtils)
+test_escape_string_as_regex()
 
-if(EIGEN_BUILD_LIB)
-    set(Eigen_SRCS
-      src/Core/CoreInstantiations.cpp
-      src/Cholesky/CholeskyInstantiations.cpp
-      src/QR/QrInstantiations.cpp
-    )
+file(GLOB Eigen_directory_files "*")
 
-    add_library(Eigen2 SHARED ${Eigen_SRCS})
+escape_string_as_regex(ESCAPED_CMAKE_CURRENT_SOURCE_DIR "${CMAKE_CURRENT_SOURCE_DIR}")
 
-    install(TARGETS Eigen2
-            RUNTIME DESTINATION bin
-            LIBRARY DESTINATION lib
-            ARCHIVE DESTINATION lib)
-endif(EIGEN_BUILD_LIB)
-
-if(CMAKE_COMPILER_IS_GNUCXX)
-    set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -g1 -O2")
-    set(CMAKE_CXX_FLAGS_RELWITHDEBINFO "${CMAKE_CXX_FLAGS_RELWITHDEBINFO} -g1 -O2")
-endif(CMAKE_COMPILER_IS_GNUCXX)
-
-set(INCLUDE_INSTALL_DIR
-    "${CMAKE_INSTALL_PREFIX}/include/eigen2"
-    CACHE PATH
-    "The directory where we install the header files"
-    FORCE)
+foreach(f ${Eigen_directory_files})
+  if(NOT f MATCHES "\\.txt" AND NOT f MATCHES "${ESCAPED_CMAKE_CURRENT_SOURCE_DIR}/[.].+" AND NOT f MATCHES "${ESCAPED_CMAKE_CURRENT_SOURCE_DIR}/src")
+    list(APPEND Eigen_directory_files_to_install ${f})
+  endif()
+endforeach(f ${Eigen_directory_files})
 
 install(FILES
-  ${Eigen_HEADERS}
-  DESTINATION ${INCLUDE_INSTALL_DIR}/Eigen
+  ${Eigen_directory_files_to_install}
+  DESTINATION ${INCLUDE_INSTALL_DIR}/Eigen COMPONENT Devel
   )
 
-add_subdirectory(src)
+install(DIRECTORY src DESTINATION ${INCLUDE_INSTALL_DIR}/Eigen COMPONENT Devel FILES_MATCHING PATTERN "*.h")

Eigen/Cholesky

@@ -1,65 +1,46 @@
+// This file is part of Eigen, a lightweight C++ template library
+// for linear algebra.
+//
+// This Source Code Form is subject to the terms of the Mozilla
+// Public License v. 2.0. If a copy of the MPL was not distributed
+// with this file, You can obtain one at http://mozilla.org/MPL/2.0/.
+
 #ifndef EIGEN_CHOLESKY_MODULE_H
 #define EIGEN_CHOLESKY_MODULE_H
 
 #include "Core"
+#include "Jacobi"
 
-#include "src/Core/util/DisableMSVCWarnings.h"
-
-// Note that EIGEN_HIDE_HEAVY_CODE has to be defined per module
-#if (defined EIGEN_EXTERN_INSTANTIATIONS) && (EIGEN_EXTERN_INSTANTIATIONS>=2)
-  #ifndef EIGEN_HIDE_HEAVY_CODE
-  #define EIGEN_HIDE_HEAVY_CODE
-  #endif
-#elif defined EIGEN_HIDE_HEAVY_CODE
-  #undef EIGEN_HIDE_HEAVY_CODE
-#endif
-
-namespace Eigen {
+#include "src/Core/util/DisableStupidWarnings.h"
 
 /** \defgroup Cholesky_Module Cholesky module
   *
-  * \nonstableyet
+  *
   *
   * This module provides two variants of the Cholesky decomposition for selfadjoint (hermitian) matrices.
-  * Those decompositions are accessible via the following MatrixBase methods:
-  *  - MatrixBase::llt(),
+  * Those decompositions are also accessible via the following methods:
+  *  - MatrixBase::llt()
   *  - MatrixBase::ldlt()
+  *  - SelfAdjointView::llt()
+  *  - SelfAdjointView::ldlt()
   *
   * \code
   * #include <Eigen/Cholesky>
   * \endcode
   */
 
-#include "src/Array/CwiseOperators.h"
-#include "src/Array/Functors.h"
 #include "src/Cholesky/LLT.h"
 #include "src/Cholesky/LDLT.h"
-
-} // namespace Eigen
-
-#define EIGEN_CHOLESKY_MODULE_INSTANTIATE_TYPE(MATRIXTYPE,PREFIX) \
-  PREFIX template class LLT<MATRIXTYPE>; \
-  PREFIX template class LDLT<MATRIXTYPE>
-
-#define EIGEN_CHOLESKY_MODULE_INSTANTIATE(PREFIX) \
-  EIGEN_CHOLESKY_MODULE_INSTANTIATE_TYPE(Matrix2f,PREFIX); \
-  EIGEN_CHOLESKY_MODULE_INSTANTIATE_TYPE(Matrix2d,PREFIX); \
-  EIGEN_CHOLESKY_MODULE_INSTANTIATE_TYPE(Matrix3f,PREFIX); \
-  EIGEN_CHOLESKY_MODULE_INSTANTIATE_TYPE(Matrix3d,PREFIX); \
-  EIGEN_CHOLESKY_MODULE_INSTANTIATE_TYPE(Matrix4f,PREFIX); \
-  EIGEN_CHOLESKY_MODULE_INSTANTIATE_TYPE(Matrix4d,PREFIX); \
-  EIGEN_CHOLESKY_MODULE_INSTANTIATE_TYPE(MatrixXf,PREFIX); \
-  EIGEN_CHOLESKY_MODULE_INSTANTIATE_TYPE(MatrixXd,PREFIX); \
-  EIGEN_CHOLESKY_MODULE_INSTANTIATE_TYPE(MatrixXcf,PREFIX); \
-  EIGEN_CHOLESKY_MODULE_INSTANTIATE_TYPE(MatrixXcd,PREFIX)
-
-#ifdef EIGEN_EXTERN_INSTANTIATIONS
-
-namespace Eigen {
-  EIGEN_CHOLESKY_MODULE_INSTANTIATE(extern);
-} // namespace Eigen
+#ifdef EIGEN_USE_LAPACKE
+#ifdef EIGEN_USE_MKL
+#include "mkl_lapacke.h"
+#else
+#include "src/misc/lapacke.h"
+#endif
+#include "src/Cholesky/LLT_LAPACKE.h"
 #endif
 
-#include "src/Core/util/EnableMSVCWarnings.h"
+#include "src/Core/util/ReenableStupidWarnings.h"
 
 #endif // EIGEN_CHOLESKY_MODULE_H
+/* vim: set filetype=cpp et sw=2 ts=2 ai: */

Eigen/CholmodSupport

@@ -0,0 +1,48 @@
+// This file is part of Eigen, a lightweight C++ template library
+// for linear algebra.
+//
+// This Source Code Form is subject to the terms of the Mozilla
+// Public License v. 2.0. If a copy of the MPL was not distributed
+// with this file, You can obtain one at http://mozilla.org/MPL/2.0/.
+
+#ifndef EIGEN_CHOLMODSUPPORT_MODULE_H
+#define EIGEN_CHOLMODSUPPORT_MODULE_H
+
+#include "SparseCore"
+
+#include "src/Core/util/DisableStupidWarnings.h"
+
+extern "C" {
+  #include <cholmod.h>
+}
+
+/** \ingroup Support_modules
+  * \defgroup CholmodSupport_Module CholmodSupport module
+  *
+  * This module provides an interface to the Cholmod library which is part of the <a href="http://www.suitesparse.com">suitesparse</a> package.
+  * It provides the two following main factorization classes:
+  * - class CholmodSupernodalLLT: a supernodal LLT Cholesky factorization.
+  * - class CholmodDecomposiiton: a general L(D)LT Cholesky factorization with automatic or explicit runtime selection of the underlying factorization method (supernodal or simplicial).
+  *
+  * For the sake of completeness, this module also propose the two following classes:
+  * - class CholmodSimplicialLLT
+  * - class CholmodSimplicialLDLT
+  * Note that these classes does not bring any particular advantage compared to the built-in
+  * SimplicialLLT and SimplicialLDLT factorization classes.
+  *
+  * \code
+  * #include <Eigen/CholmodSupport>
+  * \endcode
+  *
+  * In order to use this module, the cholmod headers must be accessible from the include paths, and your binary must be linked to the cholmod library and its dependencies.
+  * The dependencies depend on how cholmod has been compiled.
+  * For a cmake based project, you can use our FindCholmod.cmake module to help you in this task.
+  *
+  */
+
+#include "src/CholmodSupport/CholmodSupport.h"
+
+#include "src/Core/util/ReenableStupidWarnings.h"
+
+#endif // EIGEN_CHOLMODSUPPORT_MODULE_H
+

Eigen/Core

@@ -1,43 +1,225 @@
+// This file is part of Eigen, a lightweight C++ template library
+// for linear algebra.
+//
+// Copyright (C) 2008 Gael Guennebaud <gael.guennebaud@inria.fr>
+// Copyright (C) 2007-2011 Benoit Jacob <jacob.benoit.1@gmail.com>
+//
+// This Source Code Form is subject to the terms of the Mozilla
+// Public License v. 2.0. If a copy of the MPL was not distributed
+// with this file, You can obtain one at http://mozilla.org/MPL/2.0/.
+
 #ifndef EIGEN_CORE_H
 #define EIGEN_CORE_H
 
-// first thing Eigen does: prevent MSVC from committing suicide
-#include "src/Core/util/DisableMSVCWarnings.h"
+// first thing Eigen does: stop the compiler from committing suicide
+#include "src/Core/util/DisableStupidWarnings.h"
 
-#ifdef _MSC_VER
-  #include <malloc.h> // for _aligned_malloc -- need it regardless of whether vectorization is enabled
-  #if (_MSC_VER >= 1500) // 2008 or later
-    // Remember that usage of defined() in a #define is undefined by the standard.
-    // a user reported that in 64-bit mode, MSVC doesn't care to define _M_IX86_FP.
-    #if (defined(_M_IX86_FP) && (_M_IX86_FP >= 2)) || defined(_M_X64)
-      #define EIGEN_SSE2_ON_MSVC_2008_OR_LATER
-    #endif   
+#if defined(__CUDACC__) && !defined(EIGEN_NO_CUDA)
+  #define EIGEN_CUDACC __CUDACC__
+#endif
+
+#if defined(__CUDA_ARCH__) && !defined(EIGEN_NO_CUDA)
+  #define EIGEN_CUDA_ARCH __CUDA_ARCH__
+#endif
+
+#if defined(__CUDACC_VER_MAJOR__) && (__CUDACC_VER_MAJOR__ >= 9)
+#define EIGEN_CUDACC_VER  ((__CUDACC_VER_MAJOR__ * 10000) + (__CUDACC_VER_MINOR__ * 100))
+#elif defined(__CUDACC_VER__)
+#define EIGEN_CUDACC_VER __CUDACC_VER__
+#else
+#define EIGEN_CUDACC_VER 0
+#endif
+
+// Handle NVCC/CUDA/SYCL
+#if defined(__CUDACC__) || defined(__SYCL_DEVICE_ONLY__)
+  // Do not try asserts on CUDA and SYCL!
+  #ifndef EIGEN_NO_DEBUG
+  #define EIGEN_NO_DEBUG
+  #endif
+
+  #ifdef EIGEN_INTERNAL_DEBUGGING
+  #undef EIGEN_INTERNAL_DEBUGGING
+  #endif
+
+  #ifdef EIGEN_EXCEPTIONS
+  #undef EIGEN_EXCEPTIONS
+  #endif
+
+  // All functions callable from CUDA code must be qualified with __device__
+  #ifdef __CUDACC__
+    // Do not try to vectorize on CUDA and SYCL!
+    #ifndef EIGEN_DONT_VECTORIZE
+    #define EIGEN_DONT_VECTORIZE
+    #endif
+
+    #define EIGEN_DEVICE_FUNC __host__ __device__
+    // We need cuda_runtime.h to ensure that that EIGEN_USING_STD_MATH macro
+    // works properly on the device side
+    #include <cuda_runtime.h>
+  #else
+    #define EIGEN_DEVICE_FUNC
+  #endif
+
+#else
+  #define EIGEN_DEVICE_FUNC
+
+#endif
+
+// When compiling CUDA device code with NVCC, pull in math functions from the
+// global namespace.  In host mode, and when device doee with clang, use the
+// std versions.
+#if defined(__CUDA_ARCH__) && defined(__NVCC__)
+  #define EIGEN_USING_STD_MATH(FUNC) using ::FUNC;
+#else
+  #define EIGEN_USING_STD_MATH(FUNC) using std::FUNC;
+#endif
+
+#if (defined(_CPPUNWIND) || defined(__EXCEPTIONS)) && !defined(__CUDA_ARCH__) && !defined(EIGEN_EXCEPTIONS) && !defined(EIGEN_USE_SYCL)
+  #define EIGEN_EXCEPTIONS
+#endif
+
+#ifdef EIGEN_EXCEPTIONS
+  #include <new>
+#endif
+
+// then include this file where all our macros are defined. It's really important to do it first because
+// it's where we do all the alignment settings (platform detection and honoring the user's will if he
+// defined e.g. EIGEN_DONT_ALIGN) so it needs to be done before we do anything with vectorization.
+#include "src/Core/util/Macros.h"
+
+// Disable the ipa-cp-clone optimization flag with MinGW 6.x or newer (enabled by default with -O3)
+// See http://eigen.tuxfamily.org/bz/show_bug.cgi?id=556 for details.
+#if EIGEN_COMP_MINGW && EIGEN_GNUC_AT_LEAST(4,6)
+  #pragma GCC optimize ("-fno-ipa-cp-clone")
+#endif
+
+#include <complex>
+
+// this include file manages BLAS and MKL related macros
+// and inclusion of their respective header files
+#include "src/Core/util/MKL_support.h"
+
+// if alignment is disabled, then disable vectorization. Note: EIGEN_MAX_ALIGN_BYTES is the proper check, it takes into
+// account both the user's will (EIGEN_MAX_ALIGN_BYTES,EIGEN_DONT_ALIGN) and our own platform checks
+#if EIGEN_MAX_ALIGN_BYTES==0
+  #ifndef EIGEN_DONT_VECTORIZE
+    #define EIGEN_DONT_VECTORIZE
   #endif
 #endif
 
-#ifdef __GNUC__
-  #define EIGEN_GNUC_AT_LEAST(x,y) ((__GNUC__>=x && __GNUC_MINOR__>=y) || __GNUC__>x)
+#if EIGEN_COMP_MSVC
+  #include <malloc.h> // for _aligned_malloc -- need it regardless of whether vectorization is enabled
+  #if (EIGEN_COMP_MSVC >= 1500) // 2008 or later
+    // Remember that usage of defined() in a #define is undefined by the standard.
+    // a user reported that in 64-bit mode, MSVC doesn't care to define _M_IX86_FP.
+    #if (defined(_M_IX86_FP) && (_M_IX86_FP >= 2)) || EIGEN_ARCH_x86_64
+      #define EIGEN_SSE2_ON_MSVC_2008_OR_LATER
+    #endif
+  #endif
 #else
-  #define EIGEN_GNUC_AT_LEAST(x,y) 0
-#endif
-
-// Remember that usage of defined() in a #define is undefined by the standard
-#if (defined __SSE2__) && ( (!defined __GNUC__) || EIGEN_GNUC_AT_LEAST(4,2) )
-  #define EIGEN_SSE2_BUT_NOT_OLD_GCC
+  // Remember that usage of defined() in a #define is undefined by the standard
+  #if (defined __SSE2__) && ( (!EIGEN_COMP_GNUC) || EIGEN_COMP_ICC || EIGEN_GNUC_AT_LEAST(4,2) )
+    #define EIGEN_SSE2_ON_NON_MSVC_BUT_NOT_OLD_GCC
+  #endif
 #endif
 
 #ifndef EIGEN_DONT_VECTORIZE
-  #if defined (EIGEN_SSE2_BUT_NOT_OLD_GCC) || defined(EIGEN_SSE2_ON_MSVC_2008_OR_LATER)
+
+  #if defined (EIGEN_SSE2_ON_NON_MSVC_BUT_NOT_OLD_GCC) || defined(EIGEN_SSE2_ON_MSVC_2008_OR_LATER)
+
+    // Defines symbols for compile-time detection of which instructions are
+    // used.
+    // EIGEN_VECTORIZE_YY is defined if and only if the instruction set YY is used
     #define EIGEN_VECTORIZE
     #define EIGEN_VECTORIZE_SSE
-    #include <emmintrin.h>
-    #include <xmmintrin.h>
+    #define EIGEN_VECTORIZE_SSE2
+
+    // Detect sse3/ssse3/sse4:
+    // gcc and icc defines __SSE3__, ...
+    // there is no way to know about this on msvc. You can define EIGEN_VECTORIZE_SSE* if you
+    // want to force the use of those instructions with msvc.
     #ifdef __SSE3__
-      #include <pmmintrin.h>
+      #define EIGEN_VECTORIZE_SSE3
     #endif
     #ifdef __SSSE3__
-      #include <tmmintrin.h>
+      #define EIGEN_VECTORIZE_SSSE3
     #endif
+    #ifdef __SSE4_1__
+      #define EIGEN_VECTORIZE_SSE4_1
+    #endif
+    #ifdef __SSE4_2__
+      #define EIGEN_VECTORIZE_SSE4_2
+    #endif
+    #ifdef __AVX__
+      #define EIGEN_VECTORIZE_AVX
+      #define EIGEN_VECTORIZE_SSE3
+      #define EIGEN_VECTORIZE_SSSE3
+      #define EIGEN_VECTORIZE_SSE4_1
+      #define EIGEN_VECTORIZE_SSE4_2
+    #endif
+    #ifdef __AVX2__
+      #define EIGEN_VECTORIZE_AVX2
+    #endif
+    #ifdef __FMA__
+      #define EIGEN_VECTORIZE_FMA
+    #endif
+    #if defined(__AVX512F__) && defined(EIGEN_ENABLE_AVX512)
+      #define EIGEN_VECTORIZE_AVX512
+      #define EIGEN_VECTORIZE_AVX2
+      #define EIGEN_VECTORIZE_AVX
+      #define EIGEN_VECTORIZE_FMA
+      #ifdef __AVX512DQ__
+        #define EIGEN_VECTORIZE_AVX512DQ
+      #endif
+      #ifdef __AVX512ER__
+        #define EIGEN_VECTORIZE_AVX512ER
+      #endif
+    #endif
+
+    // include files
+
+    // This extern "C" works around a MINGW-w64 compilation issue
+    // https://sourceforge.net/tracker/index.php?func=detail&aid=3018394&group_id=202880&atid=983354
+    // In essence, intrin.h is included by windows.h and also declares intrinsics (just as emmintrin.h etc. below do).
+    // However, intrin.h uses an extern "C" declaration, and g++ thus complains of duplicate declarations
+    // with conflicting linkage.  The linkage for intrinsics doesn't matter, but at that stage the compiler doesn't know;
+    // so, to avoid compile errors when windows.h is included after Eigen/Core, ensure intrinsics are extern "C" here too.
+    // notice that since these are C headers, the extern "C" is theoretically needed anyways.
+    extern "C" {
+      // In theory we should only include immintrin.h and not the other *mmintrin.h header files directly.
+      // Doing so triggers some issues with ICC. However old gcc versions seems to not have this file, thus:
+      #if EIGEN_COMP_ICC >= 1110
+        #include <immintrin.h>
+      #else
+        #include <mmintrin.h>
+        #include <emmintrin.h>
+        #include <xmmintrin.h>
+        #ifdef  EIGEN_VECTORIZE_SSE3
+        #include <pmmintrin.h>
+        #endif
+        #ifdef EIGEN_VECTORIZE_SSSE3
+        #include <tmmintrin.h>
+        #endif
+        #ifdef EIGEN_VECTORIZE_SSE4_1
+        #include <smmintrin.h>
+        #endif
+        #ifdef EIGEN_VECTORIZE_SSE4_2
+        #include <nmmintrin.h>
+        #endif
+        #if defined(EIGEN_VECTORIZE_AVX) || defined(EIGEN_VECTORIZE_AVX512)
+        #include <immintrin.h>
+        #endif
+      #endif
+    } // end extern "C"
+  #elif defined __VSX__
+    #define EIGEN_VECTORIZE
+    #define EIGEN_VECTORIZE_VSX
+    #include <altivec.h>
+    // We need to #undef all these ugly tokens defined in <altivec.h>
+    // => use __vector instead of vector
+    #undef bool
+    #undef vector
+    #undef pixel
   #elif defined __ALTIVEC__
     #define EIGEN_VECTORIZE
     #define EIGEN_VECTORIZE_ALTIVEC
@@ -47,36 +229,130 @@
     #undef bool
     #undef vector
     #undef pixel
+  #elif (defined  __ARM_NEON) || (defined __ARM_NEON__)
+    #define EIGEN_VECTORIZE
+    #define EIGEN_VECTORIZE_NEON
+    #include <arm_neon.h>
+  #elif (defined __s390x__ && defined __VEC__)
+    #define EIGEN_VECTORIZE
+    #define EIGEN_VECTORIZE_ZVECTOR
+    #include <vecintrin.h>
   #endif
 #endif
 
+#if defined(__F16C__) && !defined(EIGEN_COMP_CLANG)
+  // We can use the optimized fp16 to float and float to fp16 conversion routines
+  #define EIGEN_HAS_FP16_C
+#endif
+
+#if defined __CUDACC__
+  #define EIGEN_VECTORIZE_CUDA
+  #include <vector_types.h>
+  #if EIGEN_CUDACC_VER >= 70500
+    #define EIGEN_HAS_CUDA_FP16
+  #endif
+#endif
+
+#if defined EIGEN_HAS_CUDA_FP16
+  #include <host_defines.h>
+  #include <cuda_fp16.h>
+#endif
+
+#if (defined _OPENMP) && (!defined EIGEN_DONT_PARALLELIZE)
+  #define EIGEN_HAS_OPENMP
+#endif
+
+#ifdef EIGEN_HAS_OPENMP
+#include <omp.h>
+#endif
+
+// MSVC for windows mobile does not have the errno.h file
+#if !(EIGEN_COMP_MSVC && EIGEN_OS_WINCE) && !EIGEN_COMP_ARM
+#define EIGEN_HAS_ERRNO
+#endif
+
+#ifdef EIGEN_HAS_ERRNO
+#include <cerrno>
+#endif
+#include <cstddef>
 #include <cstdlib>
 #include <cmath>
-#include <complex>
 #include <cassert>
 #include <functional>
-#include <iostream>
+#include <sstream>
+#ifndef EIGEN_NO_IO
+  #include <iosfwd>
+#endif
 #include <cstring>
 #include <string>
 #include <limits>
+#include <climits> // for CHAR_BIT
+// for min/max:
+#include <algorithm>
 
-#if (defined(_CPPUNWIND) || defined(__EXCEPTIONS)) && !defined(EIGEN_NO_EXCEPTIONS)
-  #define EIGEN_EXCEPTIONS
+// for std::is_nothrow_move_assignable
+#ifdef EIGEN_INCLUDE_TYPE_TRAITS
+#include <type_traits>
 #endif
 
-#ifdef EIGEN_EXCEPTIONS
-  #include <new>
+// for outputting debug info
+#ifdef EIGEN_DEBUG_ASSIGN
+#include <iostream>
 #endif
 
-// this needs to be done after all possible windows C header includes and before any Eigen source includes
-// (system C++ includes are supposed to be able to deal with this already):
-// windows.h defines min and max macros which would make Eigen fail to compile.
-#if defined(min) || defined(max)
-#error The preprocessor symbols 'min' or 'max' are defined. If you are compiling on Windows, do #define NOMINMAX to prevent windows.h from defining these symbols.
+// required for __cpuid, needs to be included after cmath
+#if EIGEN_COMP_MSVC && EIGEN_ARCH_i386_OR_x86_64 && !EIGEN_OS_WINCE
+  #include <intrin.h>
+#endif
+
+/** \brief Namespace containing all symbols from the %Eigen library. */
+namespace Eigen {
+
+inline static const char *SimdInstructionSetsInUse(void) {
+#if defined(EIGEN_VECTORIZE_AVX512)
+  return "AVX512, FMA, AVX2, AVX, SSE, SSE2, SSE3, SSSE3, SSE4.1, SSE4.2";
+#elif defined(EIGEN_VECTORIZE_AVX)
+  return "AVX SSE, SSE2, SSE3, SSSE3, SSE4.1, SSE4.2";
+#elif defined(EIGEN_VECTORIZE_SSE4_2)
+  return "SSE, SSE2, SSE3, SSSE3, SSE4.1, SSE4.2";
+#elif defined(EIGEN_VECTORIZE_SSE4_1)
+  return "SSE, SSE2, SSE3, SSSE3, SSE4.1";
+#elif defined(EIGEN_VECTORIZE_SSSE3)
+  return "SSE, SSE2, SSE3, SSSE3";
+#elif defined(EIGEN_VECTORIZE_SSE3)
+  return "SSE, SSE2, SSE3";
+#elif defined(EIGEN_VECTORIZE_SSE2)
+  return "SSE, SSE2";
+#elif defined(EIGEN_VECTORIZE_ALTIVEC)
+  return "AltiVec";
+#elif defined(EIGEN_VECTORIZE_VSX)
+  return "VSX";
+#elif defined(EIGEN_VECTORIZE_NEON)
+  return "ARM NEON";
+#elif defined(EIGEN_VECTORIZE_ZVECTOR)
+  return "S390X ZVECTOR";
+#else
+  return "None";
+#endif
+}
+
+} // end namespace Eigen
+
+#if defined EIGEN2_SUPPORT_STAGE40_FULL_EIGEN3_STRICTNESS || defined EIGEN2_SUPPORT_STAGE30_FULL_EIGEN3_API || defined EIGEN2_SUPPORT_STAGE20_RESOLVE_API_CONFLICTS || defined EIGEN2_SUPPORT_STAGE10_FULL_EIGEN2_API || defined EIGEN2_SUPPORT
+// This will generate an error message:
+#error Eigen2-support is only available up to version 3.2. Please go to "http://eigen.tuxfamily.org/index.php?title=Eigen2" for further information
 #endif
 
 namespace Eigen {
 
+// we use size_t frequently and we'll never remember to prepend it with std:: everytime just to
+// ensure QNX/QCC support
+using std::size_t;
+// gcc 4.6.0 wants std:: for ptrdiff_t
+using std::ptrdiff_t;
+
+}
+
 /** \defgroup Core_Module Core module
   * This is the main module of Eigen providing dense matrix and vector support
   * (both fixed and dynamic size) with all the features corresponding to a BLAS library
@@ -87,68 +363,180 @@ namespace Eigen {
   * \endcode
   */
 
-#include "src/Core/util/Macros.h"
 #include "src/Core/util/Constants.h"
-#include "src/Core/util/ForwardDeclarations.h"
 #include "src/Core/util/Meta.h"
-#include "src/Core/util/XprHelper.h"
+#include "src/Core/util/ForwardDeclarations.h"
 #include "src/Core/util/StaticAssert.h"
+#include "src/Core/util/XprHelper.h"
 #include "src/Core/util/Memory.h"
 
 #include "src/Core/NumTraits.h"
 #include "src/Core/MathFunctions.h"
 #include "src/Core/GenericPacketMath.h"
+#include "src/Core/MathFunctionsImpl.h"
+#include "src/Core/arch/Default/ConjHelper.h"
 
-#if defined EIGEN_VECTORIZE_SSE
+#if defined EIGEN_VECTORIZE_AVX512
   #include "src/Core/arch/SSE/PacketMath.h"
-#elif defined EIGEN_VECTORIZE_ALTIVEC
+  #include "src/Core/arch/SSE/MathFunctions.h"
+  #include "src/Core/arch/AVX/PacketMath.h"
+  #include "src/Core/arch/AVX/MathFunctions.h"
+  #include "src/Core/arch/AVX512/PacketMath.h"
+  #include "src/Core/arch/AVX512/MathFunctions.h"
+#elif defined EIGEN_VECTORIZE_AVX
+  // Use AVX for floats and doubles, SSE for integers
+  #include "src/Core/arch/SSE/PacketMath.h"
+  #include "src/Core/arch/SSE/Complex.h"
+  #include "src/Core/arch/SSE/MathFunctions.h"
+  #include "src/Core/arch/AVX/PacketMath.h"
+  #include "src/Core/arch/AVX/MathFunctions.h"
+  #include "src/Core/arch/AVX/Complex.h"
+  #include "src/Core/arch/AVX/TypeCasting.h"
+  #include "src/Core/arch/SSE/TypeCasting.h"
+#elif defined EIGEN_VECTORIZE_SSE
+  #include "src/Core/arch/SSE/PacketMath.h"
+  #include "src/Core/arch/SSE/MathFunctions.h"
+  #include "src/Core/arch/SSE/Complex.h"
+  #include "src/Core/arch/SSE/TypeCasting.h"
+#elif defined(EIGEN_VECTORIZE_ALTIVEC) || defined(EIGEN_VECTORIZE_VSX)
   #include "src/Core/arch/AltiVec/PacketMath.h"
+  #include "src/Core/arch/AltiVec/MathFunctions.h"
+  #include "src/Core/arch/AltiVec/Complex.h"
+#elif defined EIGEN_VECTORIZE_NEON
+  #include "src/Core/arch/NEON/PacketMath.h"
+  #include "src/Core/arch/NEON/MathFunctions.h"
+  #include "src/Core/arch/NEON/Complex.h"
+#elif defined EIGEN_VECTORIZE_ZVECTOR
+  #include "src/Core/arch/ZVector/PacketMath.h"
+  #include "src/Core/arch/ZVector/MathFunctions.h"
+  #include "src/Core/arch/ZVector/Complex.h"
 #endif
 
-#ifndef EIGEN_CACHEFRIENDLY_PRODUCT_THRESHOLD
-#define EIGEN_CACHEFRIENDLY_PRODUCT_THRESHOLD 16
+// Half float support
+#include "src/Core/arch/CUDA/Half.h"
+#include "src/Core/arch/CUDA/PacketMathHalf.h"
+#include "src/Core/arch/CUDA/TypeCasting.h"
+
+#if defined EIGEN_VECTORIZE_CUDA
+  #include "src/Core/arch/CUDA/PacketMath.h"
+  #include "src/Core/arch/CUDA/MathFunctions.h"
 #endif
 
-#include "src/Core/Functors.h"
+#include "src/Core/arch/Default/Settings.h"
+
+#include "src/Core/functors/TernaryFunctors.h"
+#include "src/Core/functors/BinaryFunctors.h"
+#include "src/Core/functors/UnaryFunctors.h"
+#include "src/Core/functors/NullaryFunctors.h"
+#include "src/Core/functors/StlFunctors.h"
+#include "src/Core/functors/AssignmentFunctors.h"
+
+// Specialized functors to enable the processing of complex numbers
+// on CUDA devices
+#include "src/Core/arch/CUDA/Complex.h"
+
+#include "src/Core/IO.h"
+#include "src/Core/DenseCoeffsBase.h"
+#include "src/Core/DenseBase.h"
 #include "src/Core/MatrixBase.h"
-#include "src/Core/Coeffs.h"
+#include "src/Core/EigenBase.h"
+
+#include "src/Core/Product.h"
+#include "src/Core/CoreEvaluators.h"
+#include "src/Core/AssignEvaluator.h"
 
 #ifndef EIGEN_PARSED_BY_DOXYGEN // work around Doxygen bug triggered by Assign.h r814874
                                 // at least confirmed with Doxygen 1.5.5 and 1.5.6
   #include "src/Core/Assign.h"
 #endif
 
-#include "src/Core/MatrixStorage.h"
+#include "src/Core/ArrayBase.h"
+#include "src/Core/util/BlasUtil.h"
+#include "src/Core/DenseStorage.h"
 #include "src/Core/NestByValue.h"
-#include "src/Core/Flagged.h"
+
+// #include "src/Core/ForceAlignedAccess.h"
+
+#include "src/Core/ReturnByValue.h"
+#include "src/Core/NoAlias.h"
+#include "src/Core/PlainObjectBase.h"
 #include "src/Core/Matrix.h"
-#include "src/Core/Cwise.h"
+#include "src/Core/Array.h"
+#include "src/Core/CwiseTernaryOp.h"
 #include "src/Core/CwiseBinaryOp.h"
 #include "src/Core/CwiseUnaryOp.h"
 #include "src/Core/CwiseNullaryOp.h"
+#include "src/Core/CwiseUnaryView.h"
+#include "src/Core/SelfCwiseBinaryOp.h"
 #include "src/Core/Dot.h"
-#include "src/Core/Product.h"
-#include "src/Core/DiagonalProduct.h"
-#include "src/Core/SolveTriangular.h"
+#include "src/Core/StableNorm.h"
+#include "src/Core/Stride.h"
 #include "src/Core/MapBase.h"
 #include "src/Core/Map.h"
+#include "src/Core/Ref.h"
 #include "src/Core/Block.h"
-#include "src/Core/Minor.h"
+#include "src/Core/VectorBlock.h"
 #include "src/Core/Transpose.h"
 #include "src/Core/DiagonalMatrix.h"
-#include "src/Core/DiagonalCoeffs.h"
-#include "src/Core/Sum.h"
+#include "src/Core/Diagonal.h"
+#include "src/Core/DiagonalProduct.h"
 #include "src/Core/Redux.h"
 #include "src/Core/Visitor.h"
 #include "src/Core/Fuzzy.h"
-#include "src/Core/IO.h"
 #include "src/Core/Swap.h"
 #include "src/Core/CommaInitializer.h"
-#include "src/Core/Part.h"
-#include "src/Core/CacheFriendlyProduct.h"
+#include "src/Core/GeneralProduct.h"
+#include "src/Core/Solve.h"
+#include "src/Core/Inverse.h"
+#include "src/Core/SolverBase.h"
+#include "src/Core/PermutationMatrix.h"
+#include "src/Core/Transpositions.h"
+#include "src/Core/TriangularMatrix.h"
+#include "src/Core/SelfAdjointView.h"
+#include "src/Core/products/GeneralBlockPanelKernel.h"
+#include "src/Core/products/Parallelizer.h"
+#include "src/Core/ProductEvaluators.h"
+#include "src/Core/products/GeneralMatrixVector.h"
+#include "src/Core/products/GeneralMatrixMatrix.h"
+#include "src/Core/SolveTriangular.h"
+#include "src/Core/products/GeneralMatrixMatrixTriangular.h"
+#include "src/Core/products/SelfadjointMatrixVector.h"
+#include "src/Core/products/SelfadjointMatrixMatrix.h"
+#include "src/Core/products/SelfadjointProduct.h"
+#include "src/Core/products/SelfadjointRank2Update.h"
+#include "src/Core/products/TriangularMatrixVector.h"
+#include "src/Core/products/TriangularMatrixMatrix.h"
+#include "src/Core/products/TriangularSolverMatrix.h"
+#include "src/Core/products/TriangularSolverVector.h"
+#include "src/Core/BandMatrix.h"
+#include "src/Core/CoreIterators.h"
+#include "src/Core/ConditionEstimator.h"
 
-} // namespace Eigen
+#include "src/Core/BooleanRedux.h"
+#include "src/Core/Select.h"
+#include "src/Core/VectorwiseOp.h"
+#include "src/Core/Random.h"
+#include "src/Core/Replicate.h"
+#include "src/Core/Reverse.h"
+#include "src/Core/ArrayWrapper.h"
 
-#include "src/Core/util/EnableMSVCWarnings.h"
+#ifdef EIGEN_USE_BLAS
+#include "src/Core/products/GeneralMatrixMatrix_BLAS.h"
+#include "src/Core/products/GeneralMatrixVector_BLAS.h"
+#include "src/Core/products/GeneralMatrixMatrixTriangular_BLAS.h"
+#include "src/Core/products/SelfadjointMatrixMatrix_BLAS.h"
+#include "src/Core/products/SelfadjointMatrixVector_BLAS.h"
+#include "src/Core/products/TriangularMatrixMatrix_BLAS.h"
+#include "src/Core/products/TriangularMatrixVector_BLAS.h"
+#include "src/Core/products/TriangularSolverMatrix_BLAS.h"
+#endif // EIGEN_USE_BLAS
+
+#ifdef EIGEN_USE_MKL_VML
+#include "src/Core/Assign_MKL.h"
+#endif
+
+#include "src/Core/GlobalFunctions.h"
+
+#include "src/Core/util/ReenableStupidWarnings.h"
 
 #endif // EIGEN_CORE_H

Eigen/Dense

@@ -1,8 +1,7 @@
 #include "Core"
-#include "Array"
 #include "LU"
 #include "Cholesky"
 #include "QR"
 #include "SVD"
 #include "Geometry"
-#include "LeastSquares"
+#include "Eigenvalues"

Eigen/Eigenvalues

@@ -0,0 +1,61 @@
+// This file is part of Eigen, a lightweight C++ template library
+// for linear algebra.
+//
+// This Source Code Form is subject to the terms of the Mozilla
+// Public License v. 2.0. If a copy of the MPL was not distributed
+// with this file, You can obtain one at http://mozilla.org/MPL/2.0/.
+
+#ifndef EIGEN_EIGENVALUES_MODULE_H
+#define EIGEN_EIGENVALUES_MODULE_H
+
+#include "Core"
+
+#include "Cholesky"
+#include "Jacobi"
+#include "Householder"
+#include "LU"
+#include "Geometry"
+
+#include "src/Core/util/DisableStupidWarnings.h"
+
+/** \defgroup Eigenvalues_Module Eigenvalues module
+  *
+  *
+  *
+  * This module mainly provides various eigenvalue solvers.
+  * This module also provides some MatrixBase methods, including:
+  *  - MatrixBase::eigenvalues(),
+  *  - MatrixBase::operatorNorm()
+  *
+  * \code
+  * #include <Eigen/Eigenvalues>
+  * \endcode
+  */
+
+#include "src/misc/RealSvd2x2.h"
+#include "src/Eigenvalues/Tridiagonalization.h"
+#include "src/Eigenvalues/RealSchur.h"
+#include "src/Eigenvalues/EigenSolver.h"
+#include "src/Eigenvalues/SelfAdjointEigenSolver.h"
+#include "src/Eigenvalues/GeneralizedSelfAdjointEigenSolver.h"
+#include "src/Eigenvalues/HessenbergDecomposition.h"
+#include "src/Eigenvalues/ComplexSchur.h"
+#include "src/Eigenvalues/ComplexEigenSolver.h"
+#include "src/Eigenvalues/RealQZ.h"
+#include "src/Eigenvalues/GeneralizedEigenSolver.h"
+#include "src/Eigenvalues/MatrixBaseEigenvalues.h"
+#ifdef EIGEN_USE_LAPACKE
+#ifdef EIGEN_USE_MKL
+#include "mkl_lapacke.h"
+#else
+#include "src/misc/lapacke.h"
+#endif
+#include "src/Eigenvalues/RealSchur_LAPACKE.h"
+#include "src/Eigenvalues/ComplexSchur_LAPACKE.h"
+#include "src/Eigenvalues/SelfAdjointEigenSolver_LAPACKE.h"
+#endif
+
+#include "src/Core/util/ReenableStupidWarnings.h"
+
+#endif // EIGEN_EIGENVALUES_MODULE_H
+/* vim: set filetype=cpp et sw=2 ts=2 ai: */

Eigen/Geometry

@@ -1,30 +1,32 @@
+// This file is part of Eigen, a lightweight C++ template library
+// for linear algebra.
+//
+// This Source Code Form is subject to the terms of the Mozilla
+// Public License v. 2.0. If a copy of the MPL was not distributed
+// with this file, You can obtain one at http://mozilla.org/MPL/2.0/.
+
 #ifndef EIGEN_GEOMETRY_MODULE_H
 #define EIGEN_GEOMETRY_MODULE_H
 
 #include "Core"
 
-#include "src/Core/util/DisableMSVCWarnings.h"
-
-#include "Array"
+#include "SVD"
+#include "LU"
 #include <limits>
 
-#ifndef M_PI
-#define M_PI 3.14159265358979323846
-#endif
-
-namespace Eigen {
+#include "src/Core/util/DisableStupidWarnings.h"
 
 /** \defgroup Geometry_Module Geometry module
-  *
-  * \nonstableyet
   *
   * This module provides support for:
   *  - fixed-size homogeneous transformations
   *  - translation, scaling, 2D and 3D rotations
-  *  - quaternions
-  *  - \ref MatrixBase::cross() "cross product"
-  *  - \ref MatrixBase::unitOrthogonal() "orthognal vector generation"
-  *  - some linear components: parametrized-lines and hyperplanes
+  *  - \link Quaternion quaternions \endlink
+  *  - cross products (\ref MatrixBase::cross, \ref MatrixBase::cross3)
+  *  - orthognal vector generation (\ref MatrixBase::unitOrthogonal)
+  *  - some linear components: \link ParametrizedLine parametrized-lines \endlink and \link Hyperplane hyperplanes \endlink
+  *  - \link AlignedBox axis aligned bounding boxes \endlink
+  *  - \link umeyama least-square transformation fitting \endlink
   *
   * \code
   * #include <Eigen/Geometry>
@@ -32,20 +34,29 @@ namespace Eigen {
   */
 
 #include "src/Geometry/OrthoMethods.h"
+#include "src/Geometry/EulerAngles.h"
+
+#include "src/Geometry/Homogeneous.h"
 #include "src/Geometry/RotationBase.h"
 #include "src/Geometry/Rotation2D.h"
 #include "src/Geometry/Quaternion.h"
 #include "src/Geometry/AngleAxis.h"
-#include "src/Geometry/EulerAngles.h"
 #include "src/Geometry/Transform.h"
 #include "src/Geometry/Translation.h"
 #include "src/Geometry/Scaling.h"
 #include "src/Geometry/Hyperplane.h"
 #include "src/Geometry/ParametrizedLine.h"
 #include "src/Geometry/AlignedBox.h"
+#include "src/Geometry/Umeyama.h"
 
-} // namespace Eigen
+// Use the SSE optimized version whenever possible. At the moment the
+// SSE version doesn't compile when AVX is enabled
+#if defined EIGEN_VECTORIZE_SSE && !defined EIGEN_VECTORIZE_AVX
+#include "src/Geometry/arch/Geometry_SSE.h"
+#endif
 
-#include "src/Core/util/EnableMSVCWarnings.h"
+#include "src/Core/util/ReenableStupidWarnings.h"
 
 #endif // EIGEN_GEOMETRY_MODULE_H
+/* vim: set filetype=cpp et sw=2 ts=2 ai: */
+

Eigen/Householder

@@ -0,0 +1,30 @@
+// This file is part of Eigen, a lightweight C++ template library
+// for linear algebra.
+//
+// This Source Code Form is subject to the terms of the Mozilla
+// Public License v. 2.0. If a copy of the MPL was not distributed
+// with this file, You can obtain one at http://mozilla.org/MPL/2.0/.
+
+#ifndef EIGEN_HOUSEHOLDER_MODULE_H
+#define EIGEN_HOUSEHOLDER_MODULE_H
+
+#include "Core"
+
+#include "src/Core/util/DisableStupidWarnings.h"
+
+/** \defgroup Householder_Module Householder module
+  * This module provides Householder transformations.
+  *
+  * \code
+  * #include <Eigen/Householder>
+  * \endcode
+  */
+
+#include "src/Householder/Householder.h"
+#include "src/Householder/HouseholderSequence.h"
+#include "src/Householder/BlockHouseholder.h"
+
+#include "src/Core/util/ReenableStupidWarnings.h"
+
+#endif // EIGEN_HOUSEHOLDER_MODULE_H
+/* vim: set filetype=cpp et sw=2 ts=2 ai: */

Eigen/IterativeLinearSolvers

@@ -0,0 +1,48 @@
+// This file is part of Eigen, a lightweight C++ template library
+// for linear algebra.
+//
+// This Source Code Form is subject to the terms of the Mozilla
+// Public License v. 2.0. If a copy of the MPL was not distributed
+// with this file, You can obtain one at http://mozilla.org/MPL/2.0/.
+
+#ifndef EIGEN_ITERATIVELINEARSOLVERS_MODULE_H
+#define EIGEN_ITERATIVELINEARSOLVERS_MODULE_H
+
+#include "SparseCore"
+#include "OrderingMethods"
+
+#include "src/Core/util/DisableStupidWarnings.h"
+
+/** 
+  * \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,
+  *  - LeastSquaresConjugateGradient for rectangular least-square problems,
+  *  - BiCGSTAB for general square matrices.
+  *
+  * These iterative solvers are associated with some preconditioners:
+  *  - IdentityPreconditioner - not really useful
+  *  - DiagonalPreconditioner - also called Jacobi preconditioner, work very well on diagonal dominant matrices.
+  *  - IncompleteLUT - incomplete LU factorization with dual thresholding
+  *
+  * Such problems can also be solved using the direct sparse decomposition modules: SparseCholesky, CholmodSupport, UmfPackSupport, SuperLUSupport.
+  *
+    \code
+    #include <Eigen/IterativeLinearSolvers>
+    \endcode
+  */
+
+#include "src/IterativeLinearSolvers/SolveWithGuess.h"
+#include "src/IterativeLinearSolvers/IterativeSolverBase.h"
+#include "src/IterativeLinearSolvers/BasicPreconditioners.h"
+#include "src/IterativeLinearSolvers/ConjugateGradient.h"
+#include "src/IterativeLinearSolvers/LeastSquareConjugateGradient.h"
+#include "src/IterativeLinearSolvers/BiCGSTAB.h"
+#include "src/IterativeLinearSolvers/IncompleteLUT.h"
+#include "src/IterativeLinearSolvers/IncompleteCholesky.h"
+
+#include "src/Core/util/ReenableStupidWarnings.h"
+
+#endif // EIGEN_ITERATIVELINEARSOLVERS_MODULE_H

Eigen/Jacobi

@@ -0,0 +1,33 @@
+// This file is part of Eigen, a lightweight C++ template library
+// for linear algebra.
+//
+// This Source Code Form is subject to the terms of the Mozilla
+// Public License v. 2.0. If a copy of the MPL was not distributed
+// with this file, You can obtain one at http://mozilla.org/MPL/2.0/.
+
+#ifndef EIGEN_JACOBI_MODULE_H
+#define EIGEN_JACOBI_MODULE_H
+
+#include "Core"
+
+#include "src/Core/util/DisableStupidWarnings.h"
+
+/** \defgroup Jacobi_Module Jacobi module
+  * This module provides Jacobi and Givens rotations.
+  *
+  * \code
+  * #include <Eigen/Jacobi>
+  * \endcode
+  *
+  * In addition to listed classes, it defines the two following MatrixBase methods to apply a Jacobi or Givens rotation:
+  *  - MatrixBase::applyOnTheLeft()
+  *  - MatrixBase::applyOnTheRight().
+  */
+
+#include "src/Jacobi/Jacobi.h"
+
+#include "src/Core/util/ReenableStupidWarnings.h"
+
+#endif // EIGEN_JACOBI_MODULE_H
+/* vim: set filetype=cpp et sw=2 ts=2 ai: */
+

Eigen/LU

@@ -1,11 +1,16 @@
+// This file is part of Eigen, a lightweight C++ template library
+// for linear algebra.
+//
+// This Source Code Form is subject to the terms of the Mozilla
+// Public License v. 2.0. If a copy of the MPL was not distributed
+// with this file, You can obtain one at http://mozilla.org/MPL/2.0/.
+
 #ifndef EIGEN_LU_MODULE_H
 #define EIGEN_LU_MODULE_H
 
 #include "Core"
 
-#include "src/Core/util/DisableMSVCWarnings.h"
-
-namespace Eigen {
+#include "src/Core/util/DisableStupidWarnings.h"
 
 /** \defgroup LU_Module LU module
   * This module includes %LU decomposition and related notions such as matrix inversion and determinant.
@@ -18,12 +23,28 @@ namespace Eigen {
   * \endcode
   */
 
-#include "src/LU/LU.h"
+#include "src/misc/Kernel.h"
+#include "src/misc/Image.h"
+#include "src/LU/FullPivLU.h"
+#include "src/LU/PartialPivLU.h"
+#ifdef EIGEN_USE_LAPACKE
+#ifdef EIGEN_USE_MKL
+#include "mkl_lapacke.h"
+#else
+#include "src/misc/lapacke.h"
+#endif
+#include "src/LU/PartialPivLU_LAPACKE.h"
+#endif
 #include "src/LU/Determinant.h"
-#include "src/LU/Inverse.h"
+#include "src/LU/InverseImpl.h"
 
-} // namespace Eigen
+// Use the SSE optimized version whenever possible. At the moment the
+// SSE version doesn't compile when AVX is enabled
+#if defined EIGEN_VECTORIZE_SSE && !defined EIGEN_VECTORIZE_AVX
+  #include "src/LU/arch/Inverse_SSE.h"
+#endif
 
-#include "src/Core/util/EnableMSVCWarnings.h"
+#include "src/Core/util/ReenableStupidWarnings.h"
 
 #endif // EIGEN_LU_MODULE_H
+/* vim: set filetype=cpp et sw=2 ts=2 ai: */

Eigen/LeastSquares

@@ -1,27 +0,0 @@
-#ifndef EIGEN_REGRESSION_MODULE_H
-#define EIGEN_REGRESSION_MODULE_H
-
-#include "Core"
-
-#include "src/Core/util/DisableMSVCWarnings.h"
-
-#include "QR"
-#include "Geometry"
-
-namespace Eigen {
-
-/** \defgroup LeastSquares_Module LeastSquares module
-  * This module provides linear regression and related features.
-  *
-  * \code
-  * #include <Eigen/LeastSquares>
-  * \endcode
-  */
-
-#include "src/LeastSquares/LeastSquares.h"
-
-} // namespace Eigen
-
-#include "src/Core/util/EnableMSVCWarnings.h"
-
-#endif // EIGEN_REGRESSION_MODULE_H

Eigen/MetisSupport

@@ -0,0 +1,35 @@
+// This file is part of Eigen, a lightweight C++ template library
+// for linear algebra.
+//
+// This Source Code Form is subject to the terms of the Mozilla
+// Public License v. 2.0. If a copy of the MPL was not distributed
+// with this file, You can obtain one at http://mozilla.org/MPL/2.0/.
+
+#ifndef EIGEN_METISSUPPORT_MODULE_H
+#define EIGEN_METISSUPPORT_MODULE_H
+
+#include "SparseCore"
+
+#include "src/Core/util/DisableStupidWarnings.h"
+
+extern "C" {
+#include <metis.h>
+}
+
+
+/** \ingroup Support_modules
+  * \defgroup MetisSupport_Module MetisSupport module
+  *
+  * \code
+  * #include <Eigen/MetisSupport>
+  * \endcode
+  * This module defines an interface to the METIS reordering package (http://glaros.dtc.umn.edu/gkhome/views/metis). 
+  * It can be used just as any other built-in method as explained in \link OrderingMethods_Module here. \endlink
+  */
+
+
+#include "src/MetisSupport/MetisSupport.h"
+
+#include "src/Core/util/ReenableStupidWarnings.h"
+
+#endif // EIGEN_METISSUPPORT_MODULE_H

Eigen/NewStdVector

@@ -1,168 +0,0 @@
-// This file is part of Eigen, a lightweight C++ template library
-// for linear algebra.
-//
-// Copyright (C) 2009 Gael Guennebaud <g.gael@free.fr>
-// Copyright (C) 2009 Hauke Heibel <hauke.heibel@googlemail.com>
-//
-// 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_STDVECTOR_MODULE_H
-#define EIGEN_STDVECTOR_MODULE_H
-
-#include "Core"
-#include <vector>
-
-namespace Eigen {
-
-// This one is needed to prevent reimplementing the whole std::vector.
-template <class T>
-class aligned_allocator_indirection : public aligned_allocator<T>
-{
-public:
-  typedef size_t    size_type;
-  typedef ptrdiff_t difference_type;
-  typedef T*        pointer;
-  typedef const T*  const_pointer;
-  typedef T&        reference;
-  typedef const T&  const_reference;
-  typedef T         value_type;
-
-  template<class U>
-  struct rebind
-  {
-    typedef aligned_allocator_indirection<U> other;
-  };
-
-  aligned_allocator_indirection() throw() {}
-  aligned_allocator_indirection(const aligned_allocator_indirection& ) throw() : aligned_allocator<T>() {}
-  aligned_allocator_indirection(const aligned_allocator<T>& ) throw() {}
-  template<class U>
-  aligned_allocator_indirection(const aligned_allocator_indirection<U>& ) throw() {}
-  template<class U>
-  aligned_allocator_indirection(const aligned_allocator<U>& ) throw() {}
-  ~aligned_allocator_indirection() throw() {}
-};
-
-#ifdef _MSC_VER
-
-  // sometimes, MSVC detects, at compile time, that the argument x
-  // in std::vector::resize(size_t s,T x) won't be aligned and generate an error
-  // even if this function is never called. Whence this little wrapper.
-  #define EIGEN_WORKAROUND_MSVC_STD_VECTOR(T) Eigen::ei_workaround_msvc_std_vector<T>
-  template<typename T> struct ei_workaround_msvc_std_vector : public T
-  {
-    inline ei_workaround_msvc_std_vector() : T() {}
-    inline ei_workaround_msvc_std_vector(const T& other) : T(other) {}
-    inline operator T& () { return *static_cast<T*>(this); }
-    inline operator const T& () const { return *static_cast<const T*>(this); }
-    template<typename OtherT>
-    inline T& operator=(const OtherT& other)
-    { T::operator=(other); return *this; }
-    inline ei_workaround_msvc_std_vector& operator=(const ei_workaround_msvc_std_vector& other)
-    { T::operator=(other); return *this; }
-  };
-
-#else
-
-  #define EIGEN_WORKAROUND_MSVC_STD_VECTOR(T) T
-
-#endif
-
-}
-
-namespace std {
-
-#define EIGEN_STD_VECTOR_SPECIALIZATION_BODY \
-  public:  \
-    typedef T value_type; \
-    typedef typename vector_base::allocator_type allocator_type; \
-    typedef typename vector_base::size_type size_type;  \
-    typedef typename vector_base::iterator iterator;  \
-    typedef typename vector_base::const_iterator const_iterator;  \
-    explicit vector(const allocator_type& a = allocator_type()) : vector_base(a) {}  \
-    template<typename InputIterator> \
-    vector(InputIterator first, InputIterator last, const allocator_type& a = allocator_type()) \
-    : vector_base(first, last, a) {} \
-    vector(const vector& c) : vector_base(c) {}  \
-    explicit vector(size_type num, const value_type& val = value_type()) : vector_base(num, val) {} \
-    vector(iterator start, iterator end) : vector_base(start, end) {}  \
-    vector& operator=(const vector& x) {  \
-      vector_base::operator=(x);  \
-      return *this;  \
-    }
-
-template<typename T>
-class vector<T,Eigen::aligned_allocator<T> >
-  : public vector<EIGEN_WORKAROUND_MSVC_STD_VECTOR(T),
-                  Eigen::aligned_allocator_indirection<EIGEN_WORKAROUND_MSVC_STD_VECTOR(T)> >
-{
-  typedef vector<EIGEN_WORKAROUND_MSVC_STD_VECTOR(T),
-                 Eigen::aligned_allocator_indirection<EIGEN_WORKAROUND_MSVC_STD_VECTOR(T)> > vector_base;
-  EIGEN_STD_VECTOR_SPECIALIZATION_BODY
-
-  void resize(size_type new_size)
-  { resize(new_size, T()); }
-
-#if defined(_VECTOR_)
-  // workaround MSVC std::vector implementation
-  void resize(size_type new_size, const value_type& x)
-  {
-    if (vector_base::size() < new_size)
-      vector_base::_Insert_n(vector_base::end(), new_size - vector_base::size(), x);
-    else if (new_size < vector_base::size())
-      vector_base::erase(vector_base::begin() + new_size, vector_base::end());
-  }
-  void push_back(const value_type& x)
-  { vector_base::push_back(x); } 
-  using vector_base::insert;  
-  iterator insert(const_iterator position, const value_type& x)
-  { return vector_base::insert(position,x); }
-  void insert(const_iterator position, size_type new_size, const value_type& x)
-  { vector_base::insert(position, new_size, x); }
-#elif defined(_GLIBCXX_VECTOR) && EIGEN_GNUC_AT_LEAST(4,2)
-  // workaround GCC std::vector implementation
-  void resize(size_type new_size, const value_type& x)
-  {
-    if (new_size < vector_base::size())
-      vector_base::_M_erase_at_end(this->_M_impl._M_start + new_size);
-    else
-      vector_base::insert(vector_base::end(), new_size - vector_base::size(), x);
-  }
-#elif defined(_GLIBCXX_VECTOR) && (!EIGEN_GNUC_AT_LEAST(4,1))
-  // Note that before gcc-4.1 we already have: std::vector::resize(size_type,const T&),
-  // no no need to workaround !
-  using vector_base::resize;
-#else
-  // either GCC 4.1 or non-GCC
-  // default implementation which should always work.
-  void resize(size_type new_size, const value_type& x)
-  {
-    if (new_size < vector_base::size())
-      vector_base::erase(vector_base::begin() + new_size, vector_base::end());
-    else if (new_size > vector_base::size())
-      vector_base::insert(vector_base::end(), new_size - vector_base::size(), x);
-  }
-#endif
-
-};
-
-}
-
-#endif // EIGEN_STDVECTOR_MODULE_H

Eigen/OrderingMethods

@@ -0,0 +1,73 @@
+// This file is part of Eigen, a lightweight C++ template library
+// for linear algebra.
+//
+// This Source Code Form is subject to the terms of the Mozilla
+// Public License v. 2.0. If a copy of the MPL was not distributed
+// with this file, You can obtain one at http://mozilla.org/MPL/2.0/.
+
+#ifndef EIGEN_ORDERINGMETHODS_MODULE_H
+#define EIGEN_ORDERINGMETHODS_MODULE_H
+
+#include "SparseCore"
+
+#include "src/Core/util/DisableStupidWarnings.h"
+
+/** 
+  * \defgroup OrderingMethods_Module OrderingMethods module
+  *
+  * This module is currently for internal use only
+  * 
+  * It defines various built-in and external ordering methods for sparse matrices. 
+  * They are typically used to reduce the number of elements during 
+  * the sparse matrix decomposition (LLT, LU, QR).
+  * Precisely, in a preprocessing step, a permutation matrix P is computed using 
+  * those ordering methods and applied to the columns of the matrix. 
+  * Using for instance the sparse Cholesky decomposition, it is expected that 
+  * the nonzeros elements in LLT(A*P) will be much smaller than that in LLT(A).
+  * 
+  * 
+  * Usage : 
+  * \code
+  * #include <Eigen/OrderingMethods>
+  * \endcode
+  * 
+  * A simple usage is as a template parameter in the sparse decomposition classes : 
+  * 
+  * \code 
+  * SparseLU<MatrixType, COLAMDOrdering<int> > solver;
+  * \endcode 
+  * 
+  * \code 
+  * SparseQR<MatrixType, COLAMDOrdering<int> > solver;
+  * \endcode
+  * 
+  * It is possible as well to call directly a particular ordering method for your own purpose, 
+  * \code 
+  * AMDOrdering<int> ordering;
+  * PermutationMatrix<Dynamic, Dynamic, int> perm;
+  * SparseMatrix<double> A; 
+  * //Fill the matrix ...
+  * 
+  * ordering(A, perm); // Call AMD
+  * \endcode
+  * 
+  * \note Some of these methods (like AMD or METIS), need the sparsity pattern 
+  * of the input matrix to be symmetric. When the matrix is structurally unsymmetric, 
+  * Eigen computes internally the pattern of \f$A^T*A\f$ before calling the method.
+  * If your matrix is already symmetric (at leat in structure), you can avoid that
+  * by calling the method with a SelfAdjointView type.
+  * 
+  * \code
+  *  // Call the ordering on the pattern of the lower triangular matrix A
+  * ordering(A.selfadjointView<Lower>(), perm);
+  * \endcode
+  */
+
+#ifndef EIGEN_MPL2_ONLY
+#include "src/OrderingMethods/Amd.h"
+#endif
+
+#include "src/OrderingMethods/Ordering.h"
+#include "src/Core/util/ReenableStupidWarnings.h"
+
+#endif // EIGEN_ORDERINGMETHODS_MODULE_H

Eigen/PaStiXSupport

@@ -0,0 +1,48 @@
+// This file is part of Eigen, a lightweight C++ template library
+// for linear algebra.
+//
+// This Source Code Form is subject to the terms of the Mozilla
+// Public License v. 2.0. If a copy of the MPL was not distributed
+// with this file, You can obtain one at http://mozilla.org/MPL/2.0/.
+
+#ifndef EIGEN_PASTIXSUPPORT_MODULE_H
+#define EIGEN_PASTIXSUPPORT_MODULE_H
+
+#include "SparseCore"
+
+#include "src/Core/util/DisableStupidWarnings.h"
+
+extern "C" {
+#include <pastix_nompi.h>
+#include <pastix.h>
+}
+
+#ifdef complex
+#undef complex
+#endif
+
+/** \ingroup Support_modules
+  * \defgroup PaStiXSupport_Module PaStiXSupport module
+  * 
+  * This module provides an interface to the <a href="http://pastix.gforge.inria.fr/">PaSTiX</a> library.
+  * PaSTiX is a general \b supernodal, \b parallel and \b opensource sparse solver.
+  * It provides the two following main factorization classes:
+  * - class PastixLLT : a supernodal, parallel LLt Cholesky factorization.
+  * - class PastixLDLT: a supernodal, parallel LDLt Cholesky factorization.
+  * - class PastixLU : a supernodal, parallel LU factorization (optimized for a symmetric pattern).
+  * 
+  * \code
+  * #include <Eigen/PaStiXSupport>
+  * \endcode
+  *
+  * In order to use this module, the PaSTiX headers must be accessible from the include paths, and your binary must be linked to the PaSTiX library and its dependencies.
+  * The dependencies depend on how PaSTiX has been compiled.
+  * For a cmake based project, you can use our FindPaSTiX.cmake module to help you in this task.
+  *
+  */
+
+#include "src/PaStiXSupport/PaStiXSupport.h"
+
+#include "src/Core/util/ReenableStupidWarnings.h"
+
+#endif // EIGEN_PASTIXSUPPORT_MODULE_H

Eigen/PardisoSupport

@@ -0,0 +1,35 @@
+// This file is part of Eigen, a lightweight C++ template library
+// for linear algebra.
+//
+// This Source Code Form is subject to the terms of the Mozilla
+// Public License v. 2.0. If a copy of the MPL was not distributed
+// with this file, You can obtain one at http://mozilla.org/MPL/2.0/.
+
+#ifndef EIGEN_PARDISOSUPPORT_MODULE_H
+#define EIGEN_PARDISOSUPPORT_MODULE_H
+
+#include "SparseCore"
+
+#include "src/Core/util/DisableStupidWarnings.h"
+
+#include <mkl_pardiso.h>
+
+/** \ingroup Support_modules
+  * \defgroup PardisoSupport_Module PardisoSupport module
+  *
+  * This module brings support for the Intel(R) MKL PARDISO direct sparse solvers.
+  *
+  * \code
+  * #include <Eigen/PardisoSupport>
+  * \endcode
+  *
+  * In order to use this module, the MKL headers must be accessible from the include paths, and your binary must be linked to the MKL library and its dependencies.
+  * See this \ref TopicUsingIntelMKL "page" for more information on MKL-Eigen integration.
+  * 
+  */
+
+#include "src/PardisoSupport/PardisoSupport.h"
+
+#include "src/Core/util/ReenableStupidWarnings.h"
+
+#endif // EIGEN_PARDISOSUPPORT_MODULE_H

Eigen/QR

@@ -1,73 +1,51 @@
+// This file is part of Eigen, a lightweight C++ template library
+// for linear algebra.
+//
+// This Source Code Form is subject to the terms of the Mozilla
+// Public License v. 2.0. If a copy of the MPL was not distributed
+// with this file, You can obtain one at http://mozilla.org/MPL/2.0/.
+
 #ifndef EIGEN_QR_MODULE_H
 #define EIGEN_QR_MODULE_H
 
 #include "Core"
 
-#include "src/Core/util/DisableMSVCWarnings.h"
-
 #include "Cholesky"
+#include "Jacobi"
+#include "Householder"
 
-// Note that EIGEN_HIDE_HEAVY_CODE has to be defined per module
-#if (defined EIGEN_EXTERN_INSTANTIATIONS) && (EIGEN_EXTERN_INSTANTIATIONS>=2)
-  #ifndef EIGEN_HIDE_HEAVY_CODE
-  #define EIGEN_HIDE_HEAVY_CODE
-  #endif
-#elif defined EIGEN_HIDE_HEAVY_CODE
-  #undef EIGEN_HIDE_HEAVY_CODE
-#endif
-
-namespace Eigen {
+#include "src/Core/util/DisableStupidWarnings.h"
 
 /** \defgroup QR_Module QR module
   *
-  * \nonstableyet
   *
-  * This module mainly provides QR decomposition and an eigen value solver.
+  *
+  * This module provides various QR decompositions
   * This module also provides some MatrixBase methods, including:
-  *  - MatrixBase::qr(),
-  *  - MatrixBase::eigenvalues(),
-  *  - MatrixBase::operatorNorm()
+  *  - MatrixBase::householderQr()
+  *  - MatrixBase::colPivHouseholderQr()
+  *  - MatrixBase::fullPivHouseholderQr()
   *
   * \code
   * #include <Eigen/QR>
   * \endcode
   */
 
-#include "src/QR/QR.h"
-#include "src/QR/Tridiagonalization.h"
-#include "src/QR/EigenSolver.h"
-#include "src/QR/SelfAdjointEigenSolver.h"
-#include "src/QR/HessenbergDecomposition.h"
+#include "src/QR/HouseholderQR.h"
+#include "src/QR/FullPivHouseholderQR.h"
+#include "src/QR/ColPivHouseholderQR.h"
+#include "src/QR/CompleteOrthogonalDecomposition.h"
+#ifdef EIGEN_USE_LAPACKE
+#ifdef EIGEN_USE_MKL
+#include "mkl_lapacke.h"
+#else
+#include "src/misc/lapacke.h"
+#endif
+#include "src/QR/HouseholderQR_LAPACKE.h"
+#include "src/QR/ColPivHouseholderQR_LAPACKE.h"
+#endif
 
-// declare all classes for a given matrix type
-#define EIGEN_QR_MODULE_INSTANTIATE_TYPE(MATRIXTYPE,PREFIX) \
-  PREFIX template class QR<MATRIXTYPE>; \
-  PREFIX template class Tridiagonalization<MATRIXTYPE>; \
-  PREFIX template class HessenbergDecomposition<MATRIXTYPE>; \
-  PREFIX template class SelfAdjointEigenSolver<MATRIXTYPE>
-
-// removed because it does not support complex yet
-//  PREFIX template class EigenSolver<MATRIXTYPE>
-
-// declare all class for all types
-#define EIGEN_QR_MODULE_INSTANTIATE(PREFIX) \
-  EIGEN_QR_MODULE_INSTANTIATE_TYPE(Matrix2f,PREFIX); \
-  EIGEN_QR_MODULE_INSTANTIATE_TYPE(Matrix2d,PREFIX); \
-  EIGEN_QR_MODULE_INSTANTIATE_TYPE(Matrix3f,PREFIX); \
-  EIGEN_QR_MODULE_INSTANTIATE_TYPE(Matrix3d,PREFIX); \
-  EIGEN_QR_MODULE_INSTANTIATE_TYPE(Matrix4f,PREFIX); \
-  EIGEN_QR_MODULE_INSTANTIATE_TYPE(Matrix4d,PREFIX); \
-  EIGEN_QR_MODULE_INSTANTIATE_TYPE(MatrixXf,PREFIX); \
-  EIGEN_QR_MODULE_INSTANTIATE_TYPE(MatrixXd,PREFIX); \
-  EIGEN_QR_MODULE_INSTANTIATE_TYPE(MatrixXcf,PREFIX); \
-  EIGEN_QR_MODULE_INSTANTIATE_TYPE(MatrixXcd,PREFIX)
-
-#ifdef EIGEN_EXTERN_INSTANTIATIONS
-  EIGEN_QR_MODULE_INSTANTIATE(extern);
-#endif // EIGEN_EXTERN_INSTANTIATIONS
-
-} // namespace Eigen
-
-#include "src/Core/util/EnableMSVCWarnings.h"
+#include "src/Core/util/ReenableStupidWarnings.h"
 
 #endif // EIGEN_QR_MODULE_H
+/* vim: set filetype=cpp et sw=2 ts=2 ai: */

Eigen/QtAlignedMalloc

@@ -1,49 +1,40 @@
+// This file is part of Eigen, a lightweight C++ template library
+// for linear algebra.
+//
+// This Source Code Form is subject to the terms of the Mozilla
+// Public License v. 2.0. If a copy of the MPL was not distributed
+// with this file, You can obtain one at http://mozilla.org/MPL/2.0/.
+
 #ifndef EIGEN_QTMALLOC_MODULE_H
 #define EIGEN_QTMALLOC_MODULE_H
 
-#if (!EIGEN_MALLOC_ALREADY_ALIGNED)
-
-#ifdef QVECTOR_H
-#error You must include <Eigen/QtAlignedMalloc> before <QtCore/QVector>.
-#endif                                           
-
-#ifdef Q_DECL_IMPORT
-  #define Q_DECL_IMPORT_ORIG Q_DECL_IMPORT
-  #undef Q_DECL_IMPORT
-  #define Q_DECL_IMPORT
-#else
-  #define Q_DECL_IMPORT
-#endif
-
 #include "Core"
 
-#include <QtCore/QVector>
+#if (!EIGEN_MALLOC_ALREADY_ALIGNED)
 
-inline void *qMalloc(size_t size)
+#include "src/Core/util/DisableStupidWarnings.h"
+
+void *qMalloc(std::size_t size)
 {
-  return Eigen::ei_aligned_malloc(size);
+  return Eigen::internal::aligned_malloc(size);
 }
 
-inline void qFree(void *ptr)
+void qFree(void *ptr)
 {
-  Eigen::ei_aligned_free(ptr);
+  Eigen::internal::aligned_free(ptr);
 }
 
-inline void *qRealloc(void *ptr, size_t size)
+void *qRealloc(void *ptr, std::size_t size)
 {
-  void* newPtr = Eigen::ei_aligned_malloc(size);
-  memcpy(newPtr, ptr, size);
-  Eigen::ei_aligned_free(ptr);
+  void* newPtr = Eigen::internal::aligned_malloc(size);
+  std::memcpy(newPtr, ptr, size);
+  Eigen::internal::aligned_free(ptr);
   return newPtr;
 }
 
-#endif
+#include "src/Core/util/ReenableStupidWarnings.h"
 
-#ifdef Q_DECL_IMPORT_ORIG
-  #define Q_DECL_IMPORT Q_DECL_IMPORT_ORIG
-  #undef Q_DECL_IMPORT_ORIG
-#else
-  #undef Q_DECL_IMPORT
 #endif
 
 #endif // EIGEN_QTMALLOC_MODULE_H
+/* vim: set filetype=cpp et sw=2 ts=2 ai: */

Eigen/SPQRSupport

@@ -0,0 +1,34 @@
+// This file is part of Eigen, a lightweight C++ template library
+// for linear algebra.
+//
+// This Source Code Form is subject to the terms of the Mozilla
+// Public License v. 2.0. If a copy of the MPL was not distributed
+// with this file, You can obtain one at http://mozilla.org/MPL/2.0/.
+
+#ifndef EIGEN_SPQRSUPPORT_MODULE_H
+#define EIGEN_SPQRSUPPORT_MODULE_H
+
+#include "SparseCore"
+
+#include "src/Core/util/DisableStupidWarnings.h"
+
+#include "SuiteSparseQR.hpp"
+
+/** \ingroup Support_modules
+  * \defgroup SPQRSupport_Module SuiteSparseQR module
+  * 
+  * This module provides an interface to the SPQR library, which is part of the <a href="http://www.suitesparse.com">suitesparse</a> package.
+  *
+  * \code
+  * #include <Eigen/SPQRSupport>
+  * \endcode
+  *
+  * In order to use this module, the SPQR headers must be accessible from the include paths, and your binary must be linked to the SPQR library and its dependencies (Cholmod, AMD, COLAMD,...).
+  * For a cmake based project, you can use our FindSPQR.cmake and FindCholmod.Cmake modules
+  *
+  */
+
+#include "src/CholmodSupport/CholmodSupport.h"
+#include "src/SPQRSupport/SuiteSparseQRSupport.h"
+
+#endif

Eigen/SVD

@@ -1,29 +1,51 @@
+// This file is part of Eigen, a lightweight C++ template library
+// for linear algebra.
+//
+// This Source Code Form is subject to the terms of the Mozilla
+// Public License v. 2.0. If a copy of the MPL was not distributed
+// with this file, You can obtain one at http://mozilla.org/MPL/2.0/.
+
 #ifndef EIGEN_SVD_MODULE_H
 #define EIGEN_SVD_MODULE_H
 
-#include "Core"
+#include "QR"
+#include "Householder"
+#include "Jacobi"
 
-#include "src/Core/util/DisableMSVCWarnings.h"
-
-namespace Eigen {
+#include "src/Core/util/DisableStupidWarnings.h"
 
 /** \defgroup SVD_Module SVD module
   *
-  * \nonstableyet
   *
-  * This module provides SVD decomposition for (currently) real matrices.
-  * This decomposition is accessible via the following MatrixBase method:
-  *  - MatrixBase::svd()
+  *
+  * This module provides SVD decomposition for matrices (both real and complex).
+  * Two decomposition algorithms are provided:
+  *  - JacobiSVD implementing two-sided Jacobi iterations is numerically very accurate, fast for small matrices, but very slow for larger ones.
+  *  - BDCSVD implementing a recursive divide & conquer strategy on top of an upper-bidiagonalization which remains fast for large problems.
+  * These decompositions are accessible via the respective classes and following MatrixBase methods:
+  *  - MatrixBase::jacobiSvd()
+  *  - MatrixBase::bdcSvd()
   *
   * \code
   * #include <Eigen/SVD>
   * \endcode
   */
 
-#include "src/SVD/SVD.h"
+#include "src/misc/RealSvd2x2.h"
+#include "src/SVD/UpperBidiagonalization.h"
+#include "src/SVD/SVDBase.h"
+#include "src/SVD/JacobiSVD.h"
+#include "src/SVD/BDCSVD.h"
+#if defined(EIGEN_USE_LAPACKE) && !defined(EIGEN_USE_LAPACKE_STRICT)
+#ifdef EIGEN_USE_MKL
+#include "mkl_lapacke.h"
+#else
+#include "src/misc/lapacke.h"
+#endif
+#include "src/SVD/JacobiSVD_LAPACKE.h"
+#endif
 
-} // namespace Eigen
-
-#include "src/Core/util/EnableMSVCWarnings.h"
+#include "src/Core/util/ReenableStupidWarnings.h"
 
 #endif // EIGEN_SVD_MODULE_H
+/* vim: set filetype=cpp et sw=2 ts=2 ai: */

Eigen/Sparse

@@ -1,132 +1,36 @@
+// This file is part of Eigen, a lightweight C++ template library
+// for linear algebra.
+//
+// This Source Code Form is subject to the terms of the Mozilla
+// Public License v. 2.0. If a copy of the MPL was not distributed
+// with this file, You can obtain one at http://mozilla.org/MPL/2.0/.
+
 #ifndef EIGEN_SPARSE_MODULE_H
 #define EIGEN_SPARSE_MODULE_H
 
-#include "Core"
-
-#include "src/Core/util/DisableMSVCWarnings.h"
-
-#include <vector>
-#include <map>
-#include <cstdlib>
-#include <cstring>
-#include <algorithm>
-
-#ifdef EIGEN_GOOGLEHASH_SUPPORT
-  #include <google/dense_hash_map>
-#endif
-
-#ifdef EIGEN_CHOLMOD_SUPPORT
-  extern "C" {
-    #include "cholmod.h"
-  }
-#endif
-
-#ifdef EIGEN_TAUCS_SUPPORT
-  // taucs.h declares a lot of mess
-  #define isnan
-  #define finite
-  #define isinf
-  extern "C" {
-    #include "taucs.h"
-  }
-  #undef isnan
-  #undef finite
-  #undef isinf
-
-  #ifdef min
-    #undef min
-  #endif
-  #ifdef max
-    #undef max
-  #endif
-  #ifdef complex
-    #undef complex
-  #endif
-#endif
-
-#ifdef EIGEN_SUPERLU_SUPPORT
-  typedef int int_t;
-  #include "superlu/slu_Cnames.h"
-  #include "superlu/supermatrix.h"
-  #include "superlu/slu_util.h"
-
-  namespace SuperLU_S {
-  #include "superlu/slu_sdefs.h"
-  }
-  namespace SuperLU_D {
-  #include "superlu/slu_ddefs.h"
-  }
-  namespace SuperLU_C {
-  #include "superlu/slu_cdefs.h"
-  }
-  namespace SuperLU_Z {
-  #include "superlu/slu_zdefs.h"
-  }
-  namespace Eigen { struct SluMatrix; }
-#endif
-
-#ifdef EIGEN_UMFPACK_SUPPORT
-  #include "umfpack.h"
-#endif
-
-namespace Eigen {
-
-/** \defgroup Sparse_Module Sparse module
+/** \defgroup Sparse_Module Sparse meta-module
   *
-  * \nonstableyet
+  * Meta-module including all related modules:
+  * - \ref SparseCore_Module
+  * - \ref OrderingMethods_Module
+  * - \ref SparseCholesky_Module
+  * - \ref SparseLU_Module
+  * - \ref SparseQR_Module
+  * - \ref IterativeLinearSolvers_Module
   *
-  * See the \ref TutorialSparse "Sparse tutorial"
-  *
-  * \code
-  * #include <Eigen/QR>
-  * \endcode
+    \code
+    #include <Eigen/Sparse>
+    \endcode
   */
 
-#include "src/Sparse/SparseUtil.h"
-#include "src/Sparse/SparseMatrixBase.h"
-#include "src/Sparse/CompressedStorage.h"
-#include "src/Sparse/AmbiVector.h"
-#include "src/Sparse/RandomSetter.h"
-#include "src/Sparse/SparseBlock.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/SparseTranspose.h"
-#include "src/Sparse/SparseCwise.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/SparseFlagged.h"
-#include "src/Sparse/SparseProduct.h"
-#include "src/Sparse/SparseDiagonalProduct.h"
-#include "src/Sparse/TriangularSolver.h"
-#include "src/Sparse/SparseLLT.h"
-#include "src/Sparse/SparseLDLT.h"
-#include "src/Sparse/SparseLU.h"
-
-#ifdef EIGEN_CHOLMOD_SUPPORT
-# include "src/Sparse/CholmodSupport.h"
+#include "SparseCore"
+#include "OrderingMethods"
+#ifndef EIGEN_MPL2_ONLY
+#include "SparseCholesky"
 #endif
-
-#ifdef EIGEN_TAUCS_SUPPORT
-# include "src/Sparse/TaucsSupport.h"
-#endif
-
-#ifdef EIGEN_SUPERLU_SUPPORT
-# include "src/Sparse/SuperLUSupport.h"
-#endif
-
-#ifdef EIGEN_UMFPACK_SUPPORT
-# include "src/Sparse/UmfPackSupport.h"
-#endif
-
-} // namespace Eigen
-
-#include "src/Core/util/EnableMSVCWarnings.h"
+#include "SparseLU"
+#include "SparseQR"
+#include "IterativeLinearSolvers"
 
 #endif // EIGEN_SPARSE_MODULE_H
+

Eigen/SparseCholesky

@@ -0,0 +1,45 @@
+// This file is part of Eigen, a lightweight C++ template library
+// for linear algebra.
+//
+// Copyright (C) 2008-2013 Gael Guennebaud <gael.guennebaud@inria.fr>
+//
+// This Source Code Form is subject to the terms of the Mozilla
+// Public License v. 2.0. If a copy of the MPL was not distributed
+// with this file, You can obtain one at http://mozilla.org/MPL/2.0/.
+
+#ifndef EIGEN_SPARSECHOLESKY_MODULE_H
+#define EIGEN_SPARSECHOLESKY_MODULE_H
+
+#include "SparseCore"
+#include "OrderingMethods"
+
+#include "src/Core/util/DisableStupidWarnings.h"
+
+/** 
+  * \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
+  */
+
+#ifdef EIGEN_MPL2_ONLY
+#error The SparseCholesky module has nothing to offer in MPL2 only mode
+#endif
+
+#include "src/SparseCholesky/SimplicialCholesky.h"
+
+#ifndef EIGEN_MPL2_ONLY
+#include "src/SparseCholesky/SimplicialCholesky_impl.h"
+#endif
+
+#include "src/Core/util/ReenableStupidWarnings.h"
+
+#endif // EIGEN_SPARSECHOLESKY_MODULE_H

Eigen/SparseCore

@@ -0,0 +1,69 @@
+// This file is part of Eigen, a lightweight C++ template library
+// for linear algebra.
+//
+// This Source Code Form is subject to the terms of the Mozilla
+// Public License v. 2.0. If a copy of the MPL was not distributed
+// with this file, You can obtain one at http://mozilla.org/MPL/2.0/.
+
+#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>
+
+/** 
+  * \defgroup SparseCore_Module SparseCore module
+  *
+  * This module provides a sparse matrix representation, and basic associated matrix manipulations
+  * and operations.
+  *
+  * See the \ref TutorialSparse "Sparse tutorial"
+  *
+  * \code
+  * #include <Eigen/SparseCore>
+  * \endcode
+  *
+  * This module depends on: Core.
+  */
+
+#include "src/SparseCore/SparseUtil.h"
+#include "src/SparseCore/SparseMatrixBase.h"
+#include "src/SparseCore/SparseAssign.h"
+#include "src/SparseCore/CompressedStorage.h"
+#include "src/SparseCore/AmbiVector.h"
+#include "src/SparseCore/SparseCompressedBase.h"
+#include "src/SparseCore/SparseMatrix.h"
+#include "src/SparseCore/SparseMap.h"
+#include "src/SparseCore/MappedSparseMatrix.h"
+#include "src/SparseCore/SparseVector.h"
+#include "src/SparseCore/SparseRef.h"
+#include "src/SparseCore/SparseCwiseUnaryOp.h"
+#include "src/SparseCore/SparseCwiseBinaryOp.h"
+#include "src/SparseCore/SparseTranspose.h"
+#include "src/SparseCore/SparseBlock.h"
+#include "src/SparseCore/SparseDot.h"
+#include "src/SparseCore/SparseRedux.h"
+#include "src/SparseCore/SparseView.h"
+#include "src/SparseCore/SparseDiagonalProduct.h"
+#include "src/SparseCore/ConservativeSparseSparseProduct.h"
+#include "src/SparseCore/SparseSparseProductWithPruning.h"
+#include "src/SparseCore/SparseProduct.h"
+#include "src/SparseCore/SparseDenseProduct.h"
+#include "src/SparseCore/SparseSelfAdjointView.h"
+#include "src/SparseCore/SparseTriangularView.h"
+#include "src/SparseCore/TriangularSolver.h"
+#include "src/SparseCore/SparsePermutation.h"
+#include "src/SparseCore/SparseFuzzy.h"
+#include "src/SparseCore/SparseSolverBase.h"
+
+#include "src/Core/util/ReenableStupidWarnings.h"
+
+#endif // EIGEN_SPARSECORE_MODULE_H
+

Eigen/SparseLU

@@ -0,0 +1,46 @@
+// This file is part of Eigen, a lightweight C++ template library
+// for linear algebra.
+//
+// Copyright (C) 2012 Désiré Nuentsa-Wakam <desire.nuentsa_wakam@inria.fr>
+// Copyright (C) 2012 Gael Guennebaud <gael.guennebaud@inria.fr>
+//
+// This Source Code Form is subject to the terms of the Mozilla
+// Public License v. 2.0. If a copy of the MPL was not distributed
+// with this file, You can obtain one at http://mozilla.org/MPL/2.0/.
+
+#ifndef EIGEN_SPARSELU_MODULE_H
+#define EIGEN_SPARSELU_MODULE_H
+
+#include "SparseCore"
+
+/** 
+  * \defgroup SparseLU_Module SparseLU module
+  * This module defines a supernodal factorization of general sparse matrices.
+  * The code is fully optimized for supernode-panel updates with specialized kernels.
+  * Please, see the documentation of the SparseLU class for more details.
+  */
+
+// Ordering interface
+#include "OrderingMethods"
+
+#include "src/SparseLU/SparseLU_gemm_kernel.h"
+
+#include "src/SparseLU/SparseLU_Structs.h"
+#include "src/SparseLU/SparseLU_SupernodalMatrix.h"
+#include "src/SparseLU/SparseLUImpl.h"
+#include "src/SparseCore/SparseColEtree.h"
+#include "src/SparseLU/SparseLU_Memory.h"
+#include "src/SparseLU/SparseLU_heap_relax_snode.h"
+#include "src/SparseLU/SparseLU_relax_snode.h"
+#include "src/SparseLU/SparseLU_pivotL.h"
+#include "src/SparseLU/SparseLU_panel_dfs.h"
+#include "src/SparseLU/SparseLU_kernel_bmod.h"
+#include "src/SparseLU/SparseLU_panel_bmod.h"
+#include "src/SparseLU/SparseLU_column_dfs.h"
+#include "src/SparseLU/SparseLU_column_bmod.h"
+#include "src/SparseLU/SparseLU_copy_to_ucol.h"
+#include "src/SparseLU/SparseLU_pruneL.h"
+#include "src/SparseLU/SparseLU_Utils.h"
+#include "src/SparseLU/SparseLU.h"
+
+#endif // EIGEN_SPARSELU_MODULE_H

Eigen/SparseQR

@@ -0,0 +1,36 @@
+// This file is part of Eigen, a lightweight C++ template library
+// for linear algebra.
+//
+// This Source Code Form is subject to the terms of the Mozilla
+// Public License v. 2.0. If a copy of the MPL was not distributed
+// with this file, You can obtain one at http://mozilla.org/MPL/2.0/.
+
+#ifndef EIGEN_SPARSEQR_MODULE_H
+#define EIGEN_SPARSEQR_MODULE_H
+
+#include "SparseCore"
+#include "OrderingMethods"
+#include "src/Core/util/DisableStupidWarnings.h"
+
+/** \defgroup SparseQR_Module SparseQR module
+  * \brief Provides QR decomposition for sparse matrices
+  * 
+  * This module provides a simplicial version of the left-looking Sparse QR decomposition. 
+  * The columns of the input matrix should be reordered to limit the fill-in during the 
+  * decomposition. Built-in methods (COLAMD, AMD) or external  methods (METIS) can be used to this end.
+  * See the \link OrderingMethods_Module OrderingMethods\endlink module for the list 
+  * of built-in and external ordering methods.
+  * 
+  * \code
+  * #include <Eigen/SparseQR>
+  * \endcode
+  * 
+  * 
+  */
+
+#include "src/SparseCore/SparseColEtree.h"
+#include "src/SparseQR/SparseQR.h"
+
+#include "src/Core/util/ReenableStupidWarnings.h"
+
+#endif

Eigen/StdDeque

@@ -0,0 +1,27 @@
+// This file is part of Eigen, a lightweight C++ template library
+// for linear algebra.
+//
+// Copyright (C) 2009 Gael Guennebaud <gael.guennebaud@inria.fr>
+// Copyright (C) 2009 Hauke Heibel <hauke.heibel@googlemail.com>
+//
+// This Source Code Form is subject to the terms of the Mozilla
+// Public License v. 2.0. If a copy of the MPL was not distributed
+// with this file, You can obtain one at http://mozilla.org/MPL/2.0/.
+
+#ifndef EIGEN_STDDEQUE_MODULE_H
+#define EIGEN_STDDEQUE_MODULE_H
+
+#include "Core"
+#include <deque>
+
+#if EIGEN_COMP_MSVC && EIGEN_OS_WIN64 && (EIGEN_MAX_STATIC_ALIGN_BYTES<=16) /* MSVC auto aligns up to 16 bytes in 64 bit builds */
+
+#define EIGEN_DEFINE_STL_DEQUE_SPECIALIZATION(...)
+
+#else
+
+#include "src/StlSupport/StdDeque.h"
+
+#endif
+
+#endif // EIGEN_STDDEQUE_MODULE_H

Eigen/StdList

@@ -0,0 +1,26 @@
+// This file is part of Eigen, a lightweight C++ template library
+// for linear algebra.
+//
+// Copyright (C) 2009 Hauke Heibel <hauke.heibel@googlemail.com>
+//
+// This Source Code Form is subject to the terms of the Mozilla
+// Public License v. 2.0. If a copy of the MPL was not distributed
+// with this file, You can obtain one at http://mozilla.org/MPL/2.0/.
+
+#ifndef EIGEN_STDLIST_MODULE_H
+#define EIGEN_STDLIST_MODULE_H
+
+#include "Core"
+#include <list>
+
+#if EIGEN_COMP_MSVC && EIGEN_OS_WIN64 && (EIGEN_MAX_STATIC_ALIGN_BYTES<=16) /* MSVC auto aligns up to 16 bytes in 64 bit builds */
+
+#define EIGEN_DEFINE_STL_LIST_SPECIALIZATION(...)
+
+#else
+
+#include "src/StlSupport/StdList.h"
+
+#endif
+
+#endif // EIGEN_STDLIST_MODULE_H

Eigen/StdVector

@@ -1,147 +1,27 @@
-#ifdef EIGEN_USE_NEW_STDVECTOR
-#include "NewStdVector"
-#else
+// This file is part of Eigen, a lightweight C++ template library
+// for linear algebra.
+//
+// Copyright (C) 2009 Gael Guennebaud <gael.guennebaud@inria.fr>
+// Copyright (C) 2009 Hauke Heibel <hauke.heibel@googlemail.com>
+//
+// This Source Code Form is subject to the terms of the Mozilla
+// Public License v. 2.0. If a copy of the MPL was not distributed
+// with this file, You can obtain one at http://mozilla.org/MPL/2.0/.
 
 #ifndef EIGEN_STDVECTOR_MODULE_H
 #define EIGEN_STDVECTOR_MODULE_H
 
-#if defined(_GLIBCXX_VECTOR) || defined(_VECTOR_)
-#error you must include <Eigen/StdVector> before <vector>. Also note that <Eigen/Sparse> includes <vector>, so it must be included after <Eigen/StdVector> too.
-#endif                                                    
+#include "Core"
+#include <vector>
 
-#ifndef EIGEN_GNUC_AT_LEAST
-#ifdef __GNUC__
-  #define EIGEN_GNUC_AT_LEAST(x,y) ((__GNUC__>=x && __GNUC_MINOR__>=y) || __GNUC__>x)
-#else
-  #define EIGEN_GNUC_AT_LEAST(x,y) 0
-#endif
-#endif
+#if EIGEN_COMP_MSVC && EIGEN_OS_WIN64 && (EIGEN_MAX_STATIC_ALIGN_BYTES<=16) /* MSVC auto aligns up to 16 bytes in 64 bit builds */
 
-#define vector std_vector
-#include <vector>        
-#undef vector
-
-namespace Eigen {
-
-template<typename T> class aligned_allocator;
-
-// meta programming to determine if a class has a given member
-struct ei_does_not_have_aligned_operator_new_marker_sizeof {int a[1];};
-struct ei_has_aligned_operator_new_marker_sizeof {int a[2];};
-
-template<typename ClassType>
-struct ei_has_aligned_operator_new {
-    template<typename T>
-    static ei_has_aligned_operator_new_marker_sizeof
-    test(T const *, typename T::ei_operator_new_marker_type const * = 0);
-    static ei_does_not_have_aligned_operator_new_marker_sizeof
-    test(...);
-
-    // note that the following indirection is needed for gcc-3.3
-    enum {ret =  sizeof(test(static_cast<ClassType*>(0))) 
-              == sizeof(ei_has_aligned_operator_new_marker_sizeof) };
-};
-
-#ifdef _MSC_VER
-  
-  // sometimes, MSVC detects, at compile time, that the argument x
-  // in std::vector::resize(size_t s,T x) won't be aligned and generate an error
-  // even if this function is never called. Whence this little wrapper.
-  #define _EIGEN_WORKAROUND_MSVC_STD_VECTOR(T) Eigen::ei_workaround_msvc_std_vector<T>
-  template<typename T> struct ei_workaround_msvc_std_vector : public T
-  {
-    inline ei_workaround_msvc_std_vector() : T() {}
-    inline ei_workaround_msvc_std_vector(const T& other) : T(other) {}
-    inline operator T& () { return *static_cast<T*>(this); }
-    inline operator const T& () const { return *static_cast<const T*>(this); }
-    template<typename OtherT>
-    inline T& operator=(const OtherT& other)
-    { T::operator=(other); return *this; }
-    inline ei_workaround_msvc_std_vector& operator=(const ei_workaround_msvc_std_vector& other)
-    { T::operator=(other); return *this; }
-  };
+#define EIGEN_DEFINE_STL_VECTOR_SPECIALIZATION(...)
 
 #else
 
-  #define _EIGEN_WORKAROUND_MSVC_STD_VECTOR(T) T
+#include "src/StlSupport/StdVector.h"
 
 #endif
 
-}
-
-namespace std {
-
-#define EIGEN_STD_VECTOR_SPECIALIZATION_BODY \
-  public:  \
-    typedef T value_type; \
-    typedef typename vector_base::allocator_type allocator_type; \
-    typedef typename vector_base::size_type size_type;  \
-    typedef typename vector_base::iterator iterator;  \
-    explicit vector(const allocator_type& __a = allocator_type()) : vector_base(__a) {}  \
-    vector(const vector& c) : vector_base(c) {}  \
-    vector(size_type num, const value_type& val = value_type()) : vector_base(num, val) {} \
-    vector(iterator start, iterator end) : vector_base(start, end) {}  \
-    vector& operator=(const vector& __x) {  \
-      vector_base::operator=(__x);  \
-      return *this;  \
-    }
-
-template<typename T,
-         typename AllocT = std::allocator<T>,
-         bool HasAlignedNew = Eigen::ei_has_aligned_operator_new<T>::ret>
-class vector : public std::std_vector<T,AllocT>
-{
-  typedef std_vector<T, AllocT> vector_base;
-  EIGEN_STD_VECTOR_SPECIALIZATION_BODY
-};
-
-template<typename T,typename DummyAlloc>
-class vector<T,DummyAlloc,true>
-  : public std::std_vector<_EIGEN_WORKAROUND_MSVC_STD_VECTOR(T),
-                           Eigen::aligned_allocator<_EIGEN_WORKAROUND_MSVC_STD_VECTOR(T)> >          
-{
-  typedef std_vector<_EIGEN_WORKAROUND_MSVC_STD_VECTOR(T),
-                     Eigen::aligned_allocator<_EIGEN_WORKAROUND_MSVC_STD_VECTOR(T)> > vector_base;
-  EIGEN_STD_VECTOR_SPECIALIZATION_BODY
-
-  void resize(size_type __new_size)
-  { resize(__new_size, T()); }     
-
-  #if defined(_VECTOR_)
-  // workaround MSVC std::vector implementation
-  void resize(size_type __new_size, const value_type& __x)                 
-  {                                                              
-    if (vector_base::size() < __new_size)                                 
-      vector_base::_Insert_n(vector_base::end(), __new_size - vector_base::size(), __x);
-    else if (__new_size < vector_base::size())
-      vector_base::erase(vector_base::begin() + __new_size, vector_base::end());
-  }
-  #elif defined(_GLIBCXX_VECTOR) && EIGEN_GNUC_AT_LEAST(4,2)
-  // workaround GCC std::vector implementation
-  void resize(size_type __new_size, const value_type& __x)
-  {                                              
-    if (__new_size < vector_base::size())               
-      vector_base::_M_erase_at_end(this->_M_impl._M_start + __new_size);
-    else                       
-      vector_base::insert(vector_base::end(), __new_size - vector_base::size(), __x); 
-  }                                                              
-  #elif defined(_GLIBCXX_VECTOR) && EIGEN_GNUC_AT_LEAST(4,1)
-  void resize(size_type __new_size, const value_type& __x)
-  {
-    if (__new_size < vector_base::size())
-      vector_base::erase(vector_base::begin() + __new_size, vector_base::end());
-    else
-      vector_base::insert(vector_base::end(), __new_size - vector_base::size(), __x);
-  }
-  #else
-  // Before gcc-4.1 we already have: std::vector::resize(size_type,const T&),
-  // so no need for a workaround !
-  using vector_base::resize;
-  #endif  
-};
-
-}
-
 #endif // EIGEN_STDVECTOR_MODULE_H
-
-#endif // EIGEN_USE_NEW_STDVECTOR

Eigen/SuperLUSupport

@@ -0,0 +1,64 @@
+// This file is part of Eigen, a lightweight C++ template library
+// for linear algebra.
+//
+// This Source Code Form is subject to the terms of the Mozilla
+// Public License v. 2.0. If a copy of the MPL was not distributed
+// with this file, You can obtain one at http://mozilla.org/MPL/2.0/.
+
+#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; }
+
+/** \ingroup Support_modules
+  * \defgroup SuperLUSupport_Module SuperLUSupport module
+  *
+  * This module provides an interface to the <a href="http://crd-legacy.lbl.gov/~xiaoye/SuperLU/">SuperLU</a> library.
+  * It provides the following factorization class:
+  * - class SuperLU: a supernodal sequential LU factorization.
+  * - class SuperILU: a supernodal sequential incomplete LU factorization (to be used as a preconditioner for iterative methods).
+  *
+  * \warning This wrapper requires at least versions 4.0 of SuperLU. The 3.x versions are not supported.
+  *
+  * \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
+  *
+  * In order to use this module, the superlu headers must be accessible from the include paths, and your binary must be linked to the superlu library and its dependencies.
+  * The dependencies depend on how superlu has been compiled.
+  * For a cmake based project, you can use our FindSuperLU.cmake module to help you in this task.
+  *
+  */
+
+#include "src/SuperLUSupport/SuperLUSupport.h"
+
+#include "src/Core/util/ReenableStupidWarnings.h"
+
+#endif // EIGEN_SUPERLUSUPPORT_MODULE_H

Eigen/UmfPackSupport

@@ -0,0 +1,40 @@
+// This file is part of Eigen, a lightweight C++ template library
+// for linear algebra.
+//
+// This Source Code Form is subject to the terms of the Mozilla
+// Public License v. 2.0. If a copy of the MPL was not distributed
+// with this file, You can obtain one at http://mozilla.org/MPL/2.0/.
+
+#ifndef EIGEN_UMFPACKSUPPORT_MODULE_H
+#define EIGEN_UMFPACKSUPPORT_MODULE_H
+
+#include "SparseCore"
+
+#include "src/Core/util/DisableStupidWarnings.h"
+
+extern "C" {
+#include <umfpack.h>
+}
+
+/** \ingroup Support_modules
+  * \defgroup UmfPackSupport_Module UmfPackSupport module
+  *
+  * This module provides an interface to the UmfPack library which is part of the <a href="http://www.suitesparse.com">suitesparse</a> package.
+  * It provides the following factorization class:
+  * - class UmfPackLU: a multifrontal sequential LU factorization.
+  *
+  * \code
+  * #include <Eigen/UmfPackSupport>
+  * \endcode
+  *
+  * In order to use this module, the umfpack headers must be accessible from the include paths, and your binary must be linked to the umfpack library and its dependencies.
+  * The dependencies depend on how umfpack has been compiled.
+  * For a cmake based project, you can use our FindUmfPack.cmake module to help you in this task.
+  *
+  */
+
+#include "src/UmfPackSupport/UmfPackSupport.h"
+
+#include "src/Core/util/ReenableStupidWarnings.h"
+
+#endif // EIGEN_UMFPACKSUPPORT_MODULE_H

Eigen/src/Array/BooleanRedux.h

@@ -1,145 +0,0 @@
-// This file is part of Eigen, a lightweight C++ template library
-// for linear algebra. Eigen itself is part of the KDE project.
-//
-// Copyright (C) 2008 Gael Guennebaud <g.gael@free.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_ALLANDANY_H
-#define EIGEN_ALLANDANY_H
-
-template<typename Derived, int UnrollCount>
-struct ei_all_unroller
-{
-  enum {
-    col = (UnrollCount-1) / Derived::RowsAtCompileTime,
-    row = (UnrollCount-1) % Derived::RowsAtCompileTime
-  };
-
-  inline static bool run(const Derived &mat)
-  {
-    return ei_all_unroller<Derived, UnrollCount-1>::run(mat) && mat.coeff(row, col);
-  }
-};
-
-template<typename Derived>
-struct ei_all_unroller<Derived, 1>
-{
-  inline static bool run(const Derived &mat) { return mat.coeff(0, 0); }
-};
-
-template<typename Derived>
-struct ei_all_unroller<Derived, Dynamic>
-{
-  inline static bool run(const Derived &) { return false; }
-};
-
-template<typename Derived, int UnrollCount>
-struct ei_any_unroller
-{
-  enum {
-    col = (UnrollCount-1) / Derived::RowsAtCompileTime,
-    row = (UnrollCount-1) % Derived::RowsAtCompileTime
-  };
-
-  inline static bool run(const Derived &mat)
-  {
-    return ei_any_unroller<Derived, UnrollCount-1>::run(mat) || mat.coeff(row, col);
-  }
-};
-
-template<typename Derived>
-struct ei_any_unroller<Derived, 1>
-{
-  inline static bool run(const Derived &mat) { return mat.coeff(0, 0); }
-};
-
-template<typename Derived>
-struct ei_any_unroller<Derived, Dynamic>
-{
-  inline static bool run(const Derived &) { return false; }
-};
-
-/** \array_module
-  * 
-  * \returns true if all coefficients are true
-  *
-  * \addexample CwiseAll \label How to check whether a point is inside a box (using operator< and all())
-  *
-  * Example: \include MatrixBase_all.cpp
-  * Output: \verbinclude MatrixBase_all.out
-  *
-  * \sa MatrixBase::any(), Cwise::operator<()
-  */
-template<typename Derived>
-inline bool MatrixBase<Derived>::all() const
-{
-  const bool unroll = SizeAtCompileTime * (CoeffReadCost + NumTraits<Scalar>::AddCost)
-                      <= EIGEN_UNROLLING_LIMIT;
-  if(unroll)
-    return ei_all_unroller<Derived,
-                           unroll ? int(SizeAtCompileTime) : Dynamic
-     >::run(derived());
-  else
-  {
-    for(int j = 0; j < cols(); ++j)
-      for(int i = 0; i < rows(); ++i)
-        if (!coeff(i, j)) return false;
-    return true;
-  }
-}
-
-/** \array_module
-  * 
-  * \returns true if at least one coefficient is true
-  *
-  * \sa MatrixBase::all()
-  */
-template<typename Derived>
-inline bool MatrixBase<Derived>::any() const
-{
-  const bool unroll = SizeAtCompileTime * (CoeffReadCost + NumTraits<Scalar>::AddCost)
-                      <= EIGEN_UNROLLING_LIMIT;
-  if(unroll)
-    return ei_any_unroller<Derived,
-                           unroll ? int(SizeAtCompileTime) : Dynamic
-           >::run(derived());
-  else
-  {
-    for(int j = 0; j < cols(); ++j)
-      for(int i = 0; i < rows(); ++i)
-        if (coeff(i, j)) return true;
-    return false;
-  }
-}
-
-/** \array_module
-  * 
-  * \returns the number of coefficients which evaluate to true
-  *
-  * \sa MatrixBase::all(), MatrixBase::any()
-  */
-template<typename Derived>
-inline int MatrixBase<Derived>::count() const
-{
-  return this->cast<bool>().cast<int>().sum();
-}
-
-#endif // EIGEN_ALLANDANY_H

Eigen/src/Array/CMakeLists.txt

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

Eigen/src/Array/CwiseOperators.h

@@ -1,453 +0,0 @@
-// This file is part of Eigen, a lightweight C++ template library
-// for linear algebra. Eigen itself is part of the KDE project.
-//
-// Copyright (C) 2008 Gael Guennebaud <g.gael@free.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_ARRAY_CWISE_OPERATORS_H
-#define EIGEN_ARRAY_CWISE_OPERATORS_H
-
-// -- unary operators --
-
-/** \array_module
-  * 
-  * \returns an expression of the coefficient-wise square root of *this.
-  *
-  * Example: \include Cwise_sqrt.cpp
-  * Output: \verbinclude Cwise_sqrt.out
-  *
-  * \sa pow(), square()
-  */
-template<typename ExpressionType>
-inline const EIGEN_CWISE_UNOP_RETURN_TYPE(ei_scalar_sqrt_op)
-Cwise<ExpressionType>::sqrt() const
-{
-  return _expression();
-}
-
-/** \array_module
-  * 
-  * \returns an expression of the coefficient-wise exponential of *this.
-  *
-  * Example: \include Cwise_exp.cpp
-  * Output: \verbinclude Cwise_exp.out
-  *
-  * \sa pow(), log(), sin(), cos()
-  */
-template<typename ExpressionType>
-inline const EIGEN_CWISE_UNOP_RETURN_TYPE(ei_scalar_exp_op)
-Cwise<ExpressionType>::exp() const
-{
-  return _expression();
-}
-
-/** \array_module
-  * 
-  * \returns an expression of the coefficient-wise logarithm of *this.
-  *
-  * Example: \include Cwise_log.cpp
-  * Output: \verbinclude Cwise_log.out
-  *
-  * \sa exp()
-  */
-template<typename ExpressionType>
-inline const EIGEN_CWISE_UNOP_RETURN_TYPE(ei_scalar_log_op)
-Cwise<ExpressionType>::log() const
-{
-  return _expression();
-}
-
-/** \array_module
-  * 
-  * \returns an expression of the coefficient-wise cosine of *this.
-  *
-  * Example: \include Cwise_cos.cpp
-  * Output: \verbinclude Cwise_cos.out
-  *
-  * \sa sin(), exp()
-  */
-template<typename ExpressionType>
-inline const EIGEN_CWISE_UNOP_RETURN_TYPE(ei_scalar_cos_op)
-Cwise<ExpressionType>::cos() const
-{
-  return _expression();
-}
-
-
-/** \array_module
-  * 
-  * \returns an expression of the coefficient-wise sine of *this.
-  *
-  * Example: \include Cwise_sin.cpp
-  * Output: \verbinclude Cwise_sin.out
-  *
-  * \sa cos(), exp()
-  */
-template<typename ExpressionType>
-inline const EIGEN_CWISE_UNOP_RETURN_TYPE(ei_scalar_sin_op)
-Cwise<ExpressionType>::sin() const
-{
-  return _expression();
-}
-
-
-/** \array_module
-  * 
-  * \returns an expression of the coefficient-wise power of *this to the given exponent.
-  *
-  * Example: \include Cwise_pow.cpp
-  * Output: \verbinclude Cwise_pow.out
-  *
-  * \sa exp(), log()
-  */
-template<typename ExpressionType>
-inline const EIGEN_CWISE_UNOP_RETURN_TYPE(ei_scalar_pow_op)
-Cwise<ExpressionType>::pow(const Scalar& exponent) const
-{
-  return EIGEN_CWISE_UNOP_RETURN_TYPE(ei_scalar_pow_op)(_expression(), ei_scalar_pow_op<Scalar>(exponent));
-}
-
-
-/** \array_module
-  * 
-  * \returns an expression of the coefficient-wise inverse of *this.
-  *
-  * Example: \include Cwise_inverse.cpp
-  * Output: \verbinclude Cwise_inverse.out
-  *
-  * \sa operator/(), operator*()
-  */
-template<typename ExpressionType>
-inline const EIGEN_CWISE_UNOP_RETURN_TYPE(ei_scalar_inverse_op)
-Cwise<ExpressionType>::inverse() const
-{
-  return _expression();
-}
-
-/** \array_module
-  *
-  * \returns an expression of the coefficient-wise square of *this.
-  *
-  * Example: \include Cwise_square.cpp
-  * Output: \verbinclude Cwise_square.out
-  *
-  * \sa operator/(), operator*(), abs2()
-  */
-template<typename ExpressionType>
-inline const EIGEN_CWISE_UNOP_RETURN_TYPE(ei_scalar_square_op)
-Cwise<ExpressionType>::square() const
-{
-  return _expression();
-}
-
-/** \array_module
-  *
-  * \returns an expression of the coefficient-wise cube of *this.
-  *
-  * Example: \include Cwise_cube.cpp
-  * Output: \verbinclude Cwise_cube.out
-  *
-  * \sa square(), pow()
-  */
-template<typename ExpressionType>
-inline const EIGEN_CWISE_UNOP_RETURN_TYPE(ei_scalar_cube_op)
-Cwise<ExpressionType>::cube() const
-{
-  return _expression();
-}
-
-
-// -- binary operators --
-
-/** \array_module
-  * 
-  * \returns an expression of the coefficient-wise \< operator of *this and \a other
-  *
-  * Example: \include Cwise_less.cpp
-  * Output: \verbinclude Cwise_less.out
-  *
-  * \sa MatrixBase::all(), MatrixBase::any(), operator>(), operator<=()
-  */
-template<typename ExpressionType>
-template<typename OtherDerived>
-inline const EIGEN_CWISE_BINOP_RETURN_TYPE(std::less)
-Cwise<ExpressionType>::operator<(const MatrixBase<OtherDerived> &other) const
-{
-  return EIGEN_CWISE_BINOP_RETURN_TYPE(std::less)(_expression(), other.derived());
-}
-
-/** \array_module
-  * 
-  * \returns an expression of the coefficient-wise \<= operator of *this and \a other
-  *
-  * Example: \include Cwise_less_equal.cpp
-  * Output: \verbinclude Cwise_less_equal.out
-  *
-  * \sa MatrixBase::all(), MatrixBase::any(), operator>=(), operator<()
-  */
-template<typename ExpressionType>
-template<typename OtherDerived>
-inline const EIGEN_CWISE_BINOP_RETURN_TYPE(std::less_equal)
-Cwise<ExpressionType>::operator<=(const MatrixBase<OtherDerived> &other) const
-{
-  return EIGEN_CWISE_BINOP_RETURN_TYPE(std::less_equal)(_expression(), other.derived());
-}
-
-/** \array_module
-  * 
-  * \returns an expression of the coefficient-wise \> operator of *this and \a other
-  *
-  * Example: \include Cwise_greater.cpp
-  * Output: \verbinclude Cwise_greater.out
-  *
-  * \sa MatrixBase::all(), MatrixBase::any(), operator>=(), operator<()
-  */
-template<typename ExpressionType>
-template<typename OtherDerived>
-inline const EIGEN_CWISE_BINOP_RETURN_TYPE(std::greater)
-Cwise<ExpressionType>::operator>(const MatrixBase<OtherDerived> &other) const
-{
-  return EIGEN_CWISE_BINOP_RETURN_TYPE(std::greater)(_expression(), other.derived());
-}
-
-/** \array_module
-  * 
-  * \returns an expression of the coefficient-wise \>= operator of *this and \a other
-  *
-  * Example: \include Cwise_greater_equal.cpp
-  * Output: \verbinclude Cwise_greater_equal.out
-  *
-  * \sa MatrixBase::all(), MatrixBase::any(), operator>(), operator<=()
-  */
-template<typename ExpressionType>
-template<typename OtherDerived>
-inline const EIGEN_CWISE_BINOP_RETURN_TYPE(std::greater_equal)
-Cwise<ExpressionType>::operator>=(const MatrixBase<OtherDerived> &other) const
-{
-  return EIGEN_CWISE_BINOP_RETURN_TYPE(std::greater_equal)(_expression(), other.derived());
-}
-
-/** \array_module
-  * 
-  * \returns an expression of the coefficient-wise == operator of *this and \a other
-  *
-  * \warning this performs an exact comparison, which is generally a bad idea with floating-point types.
-  * In order to check for equality between two vectors or matrices with floating-point coefficients, it is
-  * generally a far better idea to use a fuzzy comparison as provided by MatrixBase::isApprox() and
-  * MatrixBase::isMuchSmallerThan().
-  *
-  * Example: \include Cwise_equal_equal.cpp
-  * Output: \verbinclude Cwise_equal_equal.out
-  *
-  * \sa MatrixBase::all(), MatrixBase::any(), MatrixBase::isApprox(), MatrixBase::isMuchSmallerThan()
-  */
-template<typename ExpressionType>
-template<typename OtherDerived>
-inline const EIGEN_CWISE_BINOP_RETURN_TYPE(std::equal_to)
-Cwise<ExpressionType>::operator==(const MatrixBase<OtherDerived> &other) const
-{
-  return EIGEN_CWISE_BINOP_RETURN_TYPE(std::equal_to)(_expression(), other.derived());
-}
-
-/** \array_module
-  * 
-  * \returns an expression of the coefficient-wise != operator of *this and \a other
-  *
-  * \warning this performs an exact comparison, which is generally a bad idea with floating-point types.
-  * In order to check for equality between two vectors or matrices with floating-point coefficients, it is
-  * generally a far better idea to use a fuzzy comparison as provided by MatrixBase::isApprox() and
-  * MatrixBase::isMuchSmallerThan().
-  *
-  * Example: \include Cwise_not_equal.cpp
-  * Output: \verbinclude Cwise_not_equal.out
-  *
-  * \sa MatrixBase::all(), MatrixBase::any(), MatrixBase::isApprox(), MatrixBase::isMuchSmallerThan()
-  */
-template<typename ExpressionType>
-template<typename OtherDerived>
-inline const EIGEN_CWISE_BINOP_RETURN_TYPE(std::not_equal_to)
-Cwise<ExpressionType>::operator!=(const MatrixBase<OtherDerived> &other) const
-{
-  return EIGEN_CWISE_BINOP_RETURN_TYPE(std::not_equal_to)(_expression(), other.derived());
-}
-
-// comparisons to scalar value
-
-/** \array_module
-  * 
-  * \returns an expression of the coefficient-wise \< operator of *this and a scalar \a s
-  *
-  * \sa operator<(const MatrixBase<OtherDerived> &) const
-  */
-template<typename ExpressionType>
-inline const EIGEN_CWISE_COMP_TO_SCALAR_RETURN_TYPE(std::less)
-Cwise<ExpressionType>::operator<(Scalar s) const
-{
-  return EIGEN_CWISE_COMP_TO_SCALAR_RETURN_TYPE(std::less)(_expression(),
-            typename ExpressionType::ConstantReturnType(_expression().rows(), _expression().cols(), s));
-}
-
-/** \array_module
-  * 
-  * \returns an expression of the coefficient-wise \<= operator of *this and a scalar \a s
-  *
-  * \sa operator<=(const MatrixBase<OtherDerived> &) const
-  */
-template<typename ExpressionType>
-inline const EIGEN_CWISE_COMP_TO_SCALAR_RETURN_TYPE(std::less_equal)
-Cwise<ExpressionType>::operator<=(Scalar s) const
-{
-  return EIGEN_CWISE_COMP_TO_SCALAR_RETURN_TYPE(std::less_equal)(_expression(),
-            typename ExpressionType::ConstantReturnType(_expression().rows(), _expression().cols(), s));
-}
-
-/** \array_module
-  * 
-  * \returns an expression of the coefficient-wise \> operator of *this and a scalar \a s
-  *
-  * \sa operator>(const MatrixBase<OtherDerived> &) const
-  */
-template<typename ExpressionType>
-inline const EIGEN_CWISE_COMP_TO_SCALAR_RETURN_TYPE(std::greater)
-Cwise<ExpressionType>::operator>(Scalar s) const
-{
-  return EIGEN_CWISE_COMP_TO_SCALAR_RETURN_TYPE(std::greater)(_expression(),
-            typename ExpressionType::ConstantReturnType(_expression().rows(), _expression().cols(), s));
-}
-
-/** \array_module
-  * 
-  * \returns an expression of the coefficient-wise \>= operator of *this and a scalar \a s
-  *
-  * \sa operator>=(const MatrixBase<OtherDerived> &) const
-  */
-template<typename ExpressionType>
-inline const EIGEN_CWISE_COMP_TO_SCALAR_RETURN_TYPE(std::greater_equal)
-Cwise<ExpressionType>::operator>=(Scalar s) const
-{
-  return EIGEN_CWISE_COMP_TO_SCALAR_RETURN_TYPE(std::greater_equal)(_expression(),
-            typename ExpressionType::ConstantReturnType(_expression().rows(), _expression().cols(), s));
-}
-
-/** \array_module
-  * 
-  * \returns an expression of the coefficient-wise == operator of *this and a scalar \a s
-  *
-  * \warning this performs an exact comparison, which is generally a bad idea with floating-point types.
-  * In order to check for equality between two vectors or matrices with floating-point coefficients, it is
-  * generally a far better idea to use a fuzzy comparison as provided by MatrixBase::isApprox() and
-  * MatrixBase::isMuchSmallerThan().
-  *
-  * \sa operator==(const MatrixBase<OtherDerived> &) const
-  */
-template<typename ExpressionType>
-inline const EIGEN_CWISE_COMP_TO_SCALAR_RETURN_TYPE(std::equal_to)
-Cwise<ExpressionType>::operator==(Scalar s) const
-{
-  return EIGEN_CWISE_COMP_TO_SCALAR_RETURN_TYPE(std::equal_to)(_expression(),
-            typename ExpressionType::ConstantReturnType(_expression().rows(), _expression().cols(), s));
-}
-
-/** \array_module
-  * 
-  * \returns an expression of the coefficient-wise != operator of *this and a scalar \a s
-  *
-  * \warning this performs an exact comparison, which is generally a bad idea with floating-point types.
-  * In order to check for equality between two vectors or matrices with floating-point coefficients, it is
-  * generally a far better idea to use a fuzzy comparison as provided by MatrixBase::isApprox() and
-  * MatrixBase::isMuchSmallerThan().
-  *
-  * \sa operator!=(const MatrixBase<OtherDerived> &) const
-  */
-template<typename ExpressionType>
-inline const EIGEN_CWISE_COMP_TO_SCALAR_RETURN_TYPE(std::not_equal_to)
-Cwise<ExpressionType>::operator!=(Scalar s) const
-{
-  return EIGEN_CWISE_COMP_TO_SCALAR_RETURN_TYPE(std::not_equal_to)(_expression(),
-            typename ExpressionType::ConstantReturnType(_expression().rows(), _expression().cols(), s));
-}
-
-// scalar addition
-
-/** \array_module
-  *
-  * \returns an expression of \c *this with each coeff incremented by the constant \a scalar
-  *
-  * Example: \include Cwise_plus.cpp
-  * Output: \verbinclude Cwise_plus.out
-  *
-  * \sa operator+=(), operator-()
-  */
-template<typename ExpressionType>
-inline const typename Cwise<ExpressionType>::ScalarAddReturnType
-Cwise<ExpressionType>::operator+(const Scalar& scalar) const
-{
-  return typename Cwise<ExpressionType>::ScalarAddReturnType(m_matrix, ei_scalar_add_op<Scalar>(scalar));
-}
-
-/** \array_module
-  *
-  * Adds the given \a scalar to each coeff of this expression.
-  *
-  * Example: \include Cwise_plus_equal.cpp
-  * Output: \verbinclude Cwise_plus_equal.out
-  *
-  * \sa operator+(), operator-=()
-  */
-template<typename ExpressionType>
-inline ExpressionType& Cwise<ExpressionType>::operator+=(const Scalar& scalar)
-{
-  return m_matrix.const_cast_derived() = *this + scalar;
-}
-
-/** \array_module
-  *
-  * \returns an expression of \c *this with each coeff decremented by the constant \a scalar
-  *
-  * Example: \include Cwise_minus.cpp
-  * Output: \verbinclude Cwise_minus.out
-  *
-  * \sa operator+(), operator-=()
-  */
-template<typename ExpressionType>
-inline const typename Cwise<ExpressionType>::ScalarAddReturnType
-Cwise<ExpressionType>::operator-(const Scalar& scalar) const
-{
-  return *this + (-scalar);
-}
-
-/** \array_module
-  *
-  * Substracts the given \a scalar from each coeff of this expression.
-  *
-  * Example: \include Cwise_minus_equal.cpp
-  * Output: \verbinclude Cwise_minus_equal.out
-  *
-  * \sa operator+=(), operator-()
-  */
-
-template<typename ExpressionType>
-inline ExpressionType& Cwise<ExpressionType>::operator-=(const Scalar& scalar)
-{
-  return m_matrix.const_cast_derived() = *this - scalar;
-}
-
-#endif // EIGEN_ARRAY_CWISE_OPERATORS_H

Eigen/src/Array/Functors.h

@@ -1,309 +0,0 @@
-// This file is part of Eigen, a lightweight C++ template library
-// for linear algebra. Eigen itself is part of the KDE project.
-//
-// Copyright (C) 2008 Gael Guennebaud <g.gael@free.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_ARRAY_FUNCTORS_H
-#define EIGEN_ARRAY_FUNCTORS_H
-
-/** \internal
-  * \array_module
-  *
-  * \brief Template functor to add a scalar to a fixed other one
-  *
-  * \sa class CwiseUnaryOp, Array::operator+
-  */
-/* If you wonder why doing the ei_pset1() in packetOp() is an optimization check ei_scalar_multiple_op */
-template<typename Scalar>
-struct ei_scalar_add_op {
-  typedef typename ei_packet_traits<Scalar>::type PacketScalar;
-  // FIXME default copy constructors seems bugged with std::complex<>
-  inline ei_scalar_add_op(const ei_scalar_add_op& other) : m_other(other.m_other) { }
-  inline ei_scalar_add_op(const Scalar& other) : m_other(other) { }
-  inline Scalar operator() (const Scalar& a) const { return a + m_other; }
-  inline const PacketScalar packetOp(const PacketScalar& a) const
-  { return ei_padd(a, ei_pset1(m_other)); }
-  const Scalar m_other;
-private:
-  ei_scalar_add_op& operator=(const ei_scalar_add_op&);
-};
-template<typename Scalar>
-struct ei_functor_traits<ei_scalar_add_op<Scalar> >
-{ enum { Cost = NumTraits<Scalar>::AddCost, PacketAccess = ei_packet_traits<Scalar>::size>1 }; };
-
-/** \internal
-  *
-  * \array_module
-  *
-  * \brief Template functor to compute the square root of a scalar
-  *
-  * \sa class CwiseUnaryOp, Cwise::sqrt()
-  */
-template<typename Scalar> struct ei_scalar_sqrt_op EIGEN_EMPTY_STRUCT {
-  inline const Scalar operator() (const Scalar& a) const { return ei_sqrt(a); }
-};
-template<typename Scalar>
-struct ei_functor_traits<ei_scalar_sqrt_op<Scalar> >
-{ enum { Cost = 5 * NumTraits<Scalar>::MulCost, PacketAccess = false }; };
-
-/** \internal
-  *
-  * \array_module
-  *
-  * \brief Template functor to compute the exponential of a scalar
-  *
-  * \sa class CwiseUnaryOp, Cwise::exp()
-  */
-template<typename Scalar> struct ei_scalar_exp_op EIGEN_EMPTY_STRUCT {
-  inline const Scalar operator() (const Scalar& a) const { return ei_exp(a); }
-};
-template<typename Scalar>
-struct ei_functor_traits<ei_scalar_exp_op<Scalar> >
-{ enum { Cost = 5 * NumTraits<Scalar>::MulCost, PacketAccess = false }; };
-
-/** \internal
-  *
-  * \array_module
-  *
-  * \brief Template functor to compute the logarithm of a scalar
-  *
-  * \sa class CwiseUnaryOp, Cwise::log()
-  */
-template<typename Scalar> struct ei_scalar_log_op EIGEN_EMPTY_STRUCT {
-  inline const Scalar operator() (const Scalar& a) const { return ei_log(a); }
-};
-template<typename Scalar>
-struct ei_functor_traits<ei_scalar_log_op<Scalar> >
-{ enum { Cost = 5 * NumTraits<Scalar>::MulCost, PacketAccess = false }; };
-
-/** \internal
-  *
-  * \array_module
-  *
-  * \brief Template functor to compute the cosine of a scalar
-  *
-  * \sa class CwiseUnaryOp, Cwise::cos()
-  */
-template<typename Scalar> struct ei_scalar_cos_op EIGEN_EMPTY_STRUCT {
-  inline const Scalar operator() (const Scalar& a) const { return ei_cos(a); }
-};
-template<typename Scalar>
-struct ei_functor_traits<ei_scalar_cos_op<Scalar> >
-{ enum { Cost = 5 * NumTraits<Scalar>::MulCost, PacketAccess = false }; };
-
-/** \internal
-  *
-  * \array_module
-  *
-  * \brief Template functor to compute the sine of a scalar
-  *
-  * \sa class CwiseUnaryOp, Cwise::sin()
-  */
-template<typename Scalar> struct ei_scalar_sin_op EIGEN_EMPTY_STRUCT {
-  inline const Scalar operator() (const Scalar& a) const { return ei_sin(a); }
-};
-template<typename Scalar>
-struct ei_functor_traits<ei_scalar_sin_op<Scalar> >
-{ enum { Cost = 5 * NumTraits<Scalar>::MulCost, PacketAccess = false }; };
-
-/** \internal
-  *
-  * \array_module
-  *
-  * \brief Template functor to raise a scalar to a power
-  *
-  * \sa class CwiseUnaryOp, Cwise::pow
-  */
-template<typename Scalar>
-struct ei_scalar_pow_op {
-  // FIXME default copy constructors seems bugged with std::complex<>
-  inline ei_scalar_pow_op(const ei_scalar_pow_op& other) : m_exponent(other.m_exponent) { }
-  inline ei_scalar_pow_op(const Scalar& exponent) : m_exponent(exponent) {}
-  inline Scalar operator() (const Scalar& a) const { return ei_pow(a, m_exponent); }
-  const Scalar m_exponent;
-private:
-  ei_scalar_pow_op& operator=(const ei_scalar_pow_op&);
-};
-template<typename Scalar>
-struct ei_functor_traits<ei_scalar_pow_op<Scalar> >
-{ enum { Cost = 5 * NumTraits<Scalar>::MulCost, PacketAccess = false }; };
-
-/** \internal
-  *
-  * \array_module
-  *
-  * \brief Template functor to compute the inverse of a scalar
-  *
-  * \sa class CwiseUnaryOp, Cwise::inverse()
-  */
-template<typename Scalar>
-struct ei_scalar_inverse_op {
-  inline Scalar operator() (const Scalar& a) const { return Scalar(1)/a; }
-  template<typename PacketScalar>
-  inline const PacketScalar packetOp(const PacketScalar& a) const
-  { return ei_pdiv(ei_pset1(Scalar(1)),a); }
-};
-template<typename Scalar>
-struct ei_functor_traits<ei_scalar_inverse_op<Scalar> >
-{ enum { Cost = NumTraits<Scalar>::MulCost, PacketAccess = int(ei_packet_traits<Scalar>::size)>1 }; };
-
-/** \internal
-  *
-  * \array_module
-  *
-  * \brief Template functor to compute the square of a scalar
-  *
-  * \sa class CwiseUnaryOp, Cwise::square()
-  */
-template<typename Scalar>
-struct ei_scalar_square_op {
-  inline Scalar operator() (const Scalar& a) const { return a*a; }
-  template<typename PacketScalar>
-  inline const PacketScalar packetOp(const PacketScalar& a) const
-  { return ei_pmul(a,a); }
-};
-template<typename Scalar>
-struct ei_functor_traits<ei_scalar_square_op<Scalar> >
-{ enum { Cost = NumTraits<Scalar>::MulCost, PacketAccess = int(ei_packet_traits<Scalar>::size)>1 }; };
-
-/** \internal
-  *
-  * \array_module
-  *
-  * \brief Template functor to compute the cube of a scalar
-  *
-  * \sa class CwiseUnaryOp, Cwise::cube()
-  */
-template<typename Scalar>
-struct ei_scalar_cube_op {
-  inline Scalar operator() (const Scalar& a) const { return a*a*a; }
-  template<typename PacketScalar>
-  inline const PacketScalar packetOp(const PacketScalar& a) const
-  { return ei_pmul(a,ei_pmul(a,a)); }
-};
-template<typename Scalar>
-struct ei_functor_traits<ei_scalar_cube_op<Scalar> >
-{ enum { Cost = 2*NumTraits<Scalar>::MulCost, PacketAccess = int(ei_packet_traits<Scalar>::size)>1 }; };
-
-// default ei_functor_traits for STL functors:
-
-template<typename T>
-struct ei_functor_traits<std::multiplies<T> >
-{ enum { Cost = NumTraits<T>::MulCost, PacketAccess = false }; };
-
-template<typename T>
-struct ei_functor_traits<std::divides<T> >
-{ enum { Cost = NumTraits<T>::MulCost, PacketAccess = false }; };
-
-template<typename T>
-struct ei_functor_traits<std::plus<T> >
-{ enum { Cost = NumTraits<T>::AddCost, PacketAccess = false }; };
-
-template<typename T>
-struct ei_functor_traits<std::minus<T> >
-{ enum { Cost = NumTraits<T>::AddCost, PacketAccess = false }; };
-
-template<typename T>
-struct ei_functor_traits<std::negate<T> >
-{ enum { Cost = NumTraits<T>::AddCost, PacketAccess = false }; };
-
-template<typename T>
-struct ei_functor_traits<std::logical_or<T> >
-{ enum { Cost = 1, PacketAccess = false }; };
-
-template<typename T>
-struct ei_functor_traits<std::logical_and<T> >
-{ enum { Cost = 1, PacketAccess = false }; };
-
-template<typename T>
-struct ei_functor_traits<std::logical_not<T> >
-{ enum { Cost = 1, PacketAccess = false }; };
-
-template<typename T>
-struct ei_functor_traits<std::greater<T> >
-{ enum { Cost = 1, PacketAccess = false }; };
-
-template<typename T>
-struct ei_functor_traits<std::less<T> >
-{ enum { Cost = 1, PacketAccess = false }; };
-
-template<typename T>
-struct ei_functor_traits<std::greater_equal<T> >
-{ enum { Cost = 1, PacketAccess = false }; };
-
-template<typename T>
-struct ei_functor_traits<std::less_equal<T> >
-{ enum { Cost = 1, PacketAccess = false }; };
-
-template<typename T>
-struct ei_functor_traits<std::equal_to<T> >
-{ enum { Cost = 1, PacketAccess = false }; };
-
-template<typename T>
-struct ei_functor_traits<std::not_equal_to<T> >
-{ enum { Cost = 1, PacketAccess = false }; };
-
-template<typename T>
-struct ei_functor_traits<std::binder2nd<T> >
-{ enum { Cost = ei_functor_traits<T>::Cost, PacketAccess = false }; };
-
-template<typename T>
-struct ei_functor_traits<std::binder1st<T> >
-{ enum { Cost = ei_functor_traits<T>::Cost, PacketAccess = false }; };
-
-template<typename T>
-struct ei_functor_traits<std::unary_negate<T> >
-{ enum { Cost = 1 + ei_functor_traits<T>::Cost, PacketAccess = false }; };
-
-template<typename T>
-struct ei_functor_traits<std::binary_negate<T> >
-{ enum { Cost = 1 + ei_functor_traits<T>::Cost, PacketAccess = false }; };
-
-#ifdef EIGEN_STDEXT_SUPPORT
-
-template<typename T0,typename T1>
-struct ei_functor_traits<std::project1st<T0,T1> >
-{ enum { Cost = 0, PacketAccess = false }; };
-
-template<typename T0,typename T1>
-struct ei_functor_traits<std::project2nd<T0,T1> >
-{ enum { Cost = 0, PacketAccess = false }; };
-
-template<typename T0,typename T1>
-struct ei_functor_traits<std::select2nd<std::pair<T0,T1> > >
-{ enum { Cost = 0, PacketAccess = false }; };
-
-template<typename T0,typename T1>
-struct ei_functor_traits<std::select1st<std::pair<T0,T1> > >
-{ enum { Cost = 0, PacketAccess = false }; };
-
-template<typename T0,typename T1>
-struct ei_functor_traits<std::unary_compose<T0,T1> >
-{ enum { Cost = ei_functor_traits<T0>::Cost + ei_functor_traits<T1>::Cost, PacketAccess = false }; };
-
-template<typename T0,typename T1,typename T2>
-struct ei_functor_traits<std::binary_compose<T0,T1,T2> >
-{ enum { Cost = ei_functor_traits<T0>::Cost + ei_functor_traits<T1>::Cost + ei_functor_traits<T2>::Cost, PacketAccess = false }; };
-
-#endif // EIGEN_STDEXT_SUPPORT
-
-#endif // EIGEN_ARRAY_FUNCTORS_H

Eigen/src/Array/Norms.h

@@ -1,80 +0,0 @@
-// This file is part of Eigen, a lightweight C++ template library
-// for linear algebra. Eigen itself is part of the KDE project.
-//
-// Copyright (C) 2008 Benoit Jacob <jacob.benoit.1@gmail.com>
-//
-// 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_ARRAY_NORMS_H
-#define EIGEN_ARRAY_NORMS_H
-
-template<typename Derived, int p>
-struct ei_lpNorm_selector
-{
-  typedef typename NumTraits<typename ei_traits<Derived>::Scalar>::Real RealScalar;
-  inline static RealScalar run(const MatrixBase<Derived>& m)
-  {
-    return ei_pow(m.cwise().abs().cwise().pow(p).sum(), RealScalar(1)/p);
-  }
-};
-
-template<typename Derived>
-struct ei_lpNorm_selector<Derived, 1>
-{
-  inline static typename NumTraits<typename ei_traits<Derived>::Scalar>::Real run(const MatrixBase<Derived>& m)
-  {
-    return m.cwise().abs().sum();
-  }
-};
-
-template<typename Derived>
-struct ei_lpNorm_selector<Derived, 2>
-{
-  inline static typename NumTraits<typename ei_traits<Derived>::Scalar>::Real run(const MatrixBase<Derived>& m)
-  {
-    return m.norm();
-  }
-};
-
-template<typename Derived>
-struct ei_lpNorm_selector<Derived, Infinity>
-{
-  inline static typename NumTraits<typename ei_traits<Derived>::Scalar>::Real run(const MatrixBase<Derived>& m)
-  {
-    return m.cwise().abs().maxCoeff();
-  }
-};
-
-/** \array_module
-  * 
-  * \returns the \f$ \ell^p \f$ norm of *this, that is, returns the p-th root of the sum of the p-th powers of the absolute values
-  *          of the coefficients of *this. If \a p is the special value \a Eigen::Infinity, this function returns the \f$ \ell^p\infty \f$
-  *          norm, that is the maximum of the absolute values of the coefficients of *this.
-  *
-  * \sa norm()
-  */
-template<typename Derived>
-template<int p>
-inline typename NumTraits<typename ei_traits<Derived>::Scalar>::Real MatrixBase<Derived>::lpNorm() const
-{
-  return ei_lpNorm_selector<Derived, p>::run(*this);
-}
-
-#endif // EIGEN_ARRAY_NORMS_H

Eigen/src/Array/PartialRedux.h

@@ -1,347 +0,0 @@
-// This file is part of Eigen, a lightweight C++ template library
-// for linear algebra. Eigen itself is part of the KDE project.
-//
-// Copyright (C) 2008 Gael Guennebaud <g.gael@free.fr>
-// Copyright (C) 2006-2008 Benoit Jacob <jacob.benoit.1@gmail.com>
-//
-// 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_PARTIAL_REDUX_H
-#define EIGEN_PARTIAL_REDUX_H
-
-/** \array_module \ingroup Array
-  *
-  * \class PartialReduxExpr
-  *
-  * \brief Generic expression of a partially reduxed matrix
-  *
-  * \param MatrixType the type of the matrix we are applying the redux operation
-  * \param MemberOp type of the member functor
-  * \param Direction indicates the direction of the redux (Vertical or Horizontal)
-  *
-  * This class represents an expression of a partial redux operator of a matrix.
-  * It is the return type of PartialRedux functions,
-  * and most of the time this is the only way it is used.
-  *
-  * \sa class PartialRedux
-  */
-
-template< typename MatrixType, typename MemberOp, int Direction>
-class PartialReduxExpr;
-
-template<typename MatrixType, typename MemberOp, int Direction>
-struct ei_traits<PartialReduxExpr<MatrixType, MemberOp, Direction> >
-{
-  typedef typename MemberOp::result_type Scalar;
-  typedef typename MatrixType::Scalar InputScalar;
-  typedef typename ei_nested<MatrixType>::type MatrixTypeNested;
-  typedef typename ei_cleantype<MatrixTypeNested>::type _MatrixTypeNested;
-  enum {
-    RowsAtCompileTime = Direction==Vertical   ? 1 : MatrixType::RowsAtCompileTime,
-    ColsAtCompileTime = Direction==Horizontal ? 1 : MatrixType::ColsAtCompileTime,
-    MaxRowsAtCompileTime = Direction==Vertical   ? 1 : MatrixType::MaxRowsAtCompileTime,
-    MaxColsAtCompileTime = Direction==Horizontal ? 1 : MatrixType::MaxColsAtCompileTime,
-    Flags = (unsigned int)_MatrixTypeNested::Flags & HereditaryBits,
-    TraversalSize = Direction==Vertical ? RowsAtCompileTime : ColsAtCompileTime
-  };
-  #if EIGEN_GNUC_AT_LEAST(3,4)
-  typedef typename MemberOp::template Cost<InputScalar,int(TraversalSize)> CostOpType;
-  #else
-  typedef typename MemberOp::template Cost<InputScalar,TraversalSize> CostOpType;
-  #endif
-  enum {
-    CoeffReadCost = TraversalSize * ei_traits<_MatrixTypeNested>::CoeffReadCost + int(CostOpType::value)
-  };
-};
-
-template< typename MatrixType, typename MemberOp, int Direction>
-class PartialReduxExpr : ei_no_assignment_operator,
-  public MatrixBase<PartialReduxExpr<MatrixType, MemberOp, Direction> >
-{
-  public:
-
-    EIGEN_GENERIC_PUBLIC_INTERFACE(PartialReduxExpr)
-    typedef typename ei_traits<PartialReduxExpr>::MatrixTypeNested MatrixTypeNested;
-    typedef typename ei_traits<PartialReduxExpr>::_MatrixTypeNested _MatrixTypeNested;
-
-    PartialReduxExpr(const MatrixType& mat, const MemberOp& func = MemberOp())
-      : m_matrix(mat), m_functor(func) {}
-
-    int rows() const { return (Direction==Vertical   ? 1 : m_matrix.rows()); }
-    int cols() const { return (Direction==Horizontal ? 1 : m_matrix.cols()); }
-
-    const Scalar coeff(int i, int j) const
-    {
-      if (Direction==Vertical)
-        return m_functor(m_matrix.col(j));
-      else
-        return m_functor(m_matrix.row(i));
-    }
-
-  protected:
-    const MatrixTypeNested m_matrix;
-    const MemberOp m_functor;
-};
-
-#define EIGEN_MEMBER_FUNCTOR(MEMBER,COST)                           \
-  template <typename ResultType>                                    \
-  struct ei_member_##MEMBER EIGEN_EMPTY_STRUCT {                    \
-    typedef ResultType result_type;                                 \
-    template<typename Scalar, int Size> struct Cost                 \
-    { enum { value = COST }; };                                     \
-    template<typename Derived>                                      \
-    inline ResultType operator()(const MatrixBase<Derived>& mat) const     \
-    { return mat.MEMBER(); } \
-  }
-
-EIGEN_MEMBER_FUNCTOR(squaredNorm, Size * NumTraits<Scalar>::MulCost + (Size-1)*NumTraits<Scalar>::AddCost);
-EIGEN_MEMBER_FUNCTOR(norm, (Size+5) * NumTraits<Scalar>::MulCost + (Size-1)*NumTraits<Scalar>::AddCost);
-EIGEN_MEMBER_FUNCTOR(sum, (Size-1)*NumTraits<Scalar>::AddCost);
-EIGEN_MEMBER_FUNCTOR(minCoeff, (Size-1)*NumTraits<Scalar>::AddCost);
-EIGEN_MEMBER_FUNCTOR(maxCoeff, (Size-1)*NumTraits<Scalar>::AddCost);
-EIGEN_MEMBER_FUNCTOR(all, (Size-1)*NumTraits<Scalar>::AddCost);
-EIGEN_MEMBER_FUNCTOR(any, (Size-1)*NumTraits<Scalar>::AddCost);
-EIGEN_MEMBER_FUNCTOR(count, (Size-1)*NumTraits<Scalar>::AddCost);
-
-/** \internal */
-template <typename BinaryOp, typename Scalar>
-struct ei_member_redux {
-  typedef typename ei_result_of<
-                     BinaryOp(Scalar)
-                   >::type  result_type;
-  template<typename _Scalar, int Size> struct Cost
-  { enum { value = (Size-1) * ei_functor_traits<BinaryOp>::Cost }; };
-  ei_member_redux(const BinaryOp func) : m_functor(func) {}
-  template<typename Derived>
-  inline result_type operator()(const MatrixBase<Derived>& mat) const
-  { return mat.redux(m_functor); }
-  const BinaryOp m_functor;
-private:
-  ei_member_redux& operator=(const ei_member_redux&);
-};
-
-/** \array_module \ingroup Array
-  *
-  * \class PartialRedux
-  *
-  * \brief Pseudo expression providing partial reduction operations
-  *
-  * \param ExpressionType the type of the object on which to do partial reductions
-  * \param Direction indicates the direction of the redux (Vertical or Horizontal)
-  *
-  * This class represents a pseudo expression with partial reduction features.
-  * It is the return type of MatrixBase::colwise() and MatrixBase::rowwise()
-  * and most of the time this is the only way it is used.
-  *
-  * Example: \include MatrixBase_colwise.cpp
-  * Output: \verbinclude MatrixBase_colwise.out
-  *
-  * \sa MatrixBase::colwise(), MatrixBase::rowwise(), class PartialReduxExpr
-  */
-template<typename ExpressionType, int Direction> class PartialRedux
-{
-  public:
-
-    typedef typename ei_traits<ExpressionType>::Scalar Scalar;
-    typedef typename ei_meta_if<ei_must_nest_by_value<ExpressionType>::ret,
-        ExpressionType, const ExpressionType&>::ret ExpressionTypeNested;
-
-    template<template<typename _Scalar> class Functor> struct ReturnType
-    {
-      typedef PartialReduxExpr<ExpressionType,
-                               Functor<typename ei_traits<ExpressionType>::Scalar>,
-                               Direction
-                              > Type;
-    };
-
-    template<typename BinaryOp> struct ReduxReturnType
-    {
-      typedef PartialReduxExpr<ExpressionType,
-                               ei_member_redux<BinaryOp,typename ei_traits<ExpressionType>::Scalar>,
-                               Direction
-                              > Type;
-    };
-
-    typedef typename ExpressionType::PlainMatrixType CrossReturnType;
-    
-    inline PartialRedux(const ExpressionType& matrix) : m_matrix(matrix) {}
-
-    /** \internal */
-    inline const ExpressionType& _expression() const { return m_matrix; }
-
-    template<typename BinaryOp>
-    const typename ReduxReturnType<BinaryOp>::Type
-    redux(const BinaryOp& func = BinaryOp()) const;
-
-    /** \returns a row (or column) vector expression of the smallest coefficient
-      * of each column (or row) of the referenced expression.
-      *
-      * Example: \include PartialRedux_minCoeff.cpp
-      * Output: \verbinclude PartialRedux_minCoeff.out
-      *
-      * \sa MatrixBase::minCoeff() */
-    const typename ReturnType<ei_member_minCoeff>::Type minCoeff() const
-    { return _expression(); }
-
-    /** \returns a row (or column) vector expression of the largest coefficient
-      * of each column (or row) of the referenced expression.
-      *
-      * Example: \include PartialRedux_maxCoeff.cpp
-      * Output: \verbinclude PartialRedux_maxCoeff.out
-      *
-      * \sa MatrixBase::maxCoeff() */
-    const typename ReturnType<ei_member_maxCoeff>::Type maxCoeff() const
-    { return _expression(); }
-
-    /** \returns a row (or column) vector expression of the squared norm
-      * of each column (or row) of the referenced expression.
-      *
-      * Example: \include PartialRedux_squaredNorm.cpp
-      * Output: \verbinclude PartialRedux_squaredNorm.out
-      *
-      * \sa MatrixBase::squaredNorm() */
-    const typename ReturnType<ei_member_squaredNorm>::Type squaredNorm() const
-    { return _expression(); }
-
-    /** \returns a row (or column) vector expression of the norm
-      * of each column (or row) of the referenced expression.
-      *
-      * Example: \include PartialRedux_norm.cpp
-      * Output: \verbinclude PartialRedux_norm.out
-      *
-      * \sa MatrixBase::norm() */
-    const typename ReturnType<ei_member_norm>::Type norm() const
-    { return _expression(); }
-
-    /** \returns a row (or column) vector expression of the sum
-      * of each column (or row) of the referenced expression.
-      *
-      * Example: \include PartialRedux_sum.cpp
-      * Output: \verbinclude PartialRedux_sum.out
-      *
-      * \sa MatrixBase::sum() */
-    const typename ReturnType<ei_member_sum>::Type sum() const
-    { return _expression(); }
-
-    /** \returns a row (or column) vector expression representing
-      * whether \b all coefficients of each respective column (or row) are \c true.
-      *
-      * \sa MatrixBase::all() */
-    const typename ReturnType<ei_member_all>::Type all() const
-    { return _expression(); }
-
-    /** \returns a row (or column) vector expression representing
-      * whether \b at \b least one coefficient of each respective column (or row) is \c true.
-      *
-      * \sa MatrixBase::any() */
-    const typename ReturnType<ei_member_any>::Type any() const
-    { return _expression(); }
-    
-    /** \returns a row (or column) vector expression representing
-      * the number of \c true coefficients of each respective column (or row).
-      *
-      * Example: \include PartialRedux_count.cpp
-      * Output: \verbinclude PartialRedux_count.out
-      *
-      * \sa MatrixBase::count() */
-    const PartialReduxExpr<ExpressionType, ei_member_count<int>, Direction> count() const
-    { return _expression(); }
-
-    /** \returns a 3x3 matrix expression of the cross product
-      * of each column or row of the referenced expression with the \a other vector.
-      *
-      * \geometry_module
-      *
-      * \sa MatrixBase::cross() */
-    template<typename OtherDerived>
-    const CrossReturnType cross(const MatrixBase<OtherDerived>& other) const
-    {
-      EIGEN_STATIC_ASSERT_MATRIX_SPECIFIC_SIZE(CrossReturnType,3,3)
-      EIGEN_STATIC_ASSERT_VECTOR_SPECIFIC_SIZE(OtherDerived,3)
-      EIGEN_STATIC_ASSERT((ei_is_same_type<Scalar, typename OtherDerived::Scalar>::ret),
-        YOU_MIXED_DIFFERENT_NUMERIC_TYPES__YOU_NEED_TO_USE_THE_CAST_METHOD_OF_MATRIXBASE_TO_CAST_NUMERIC_TYPES_EXPLICITLY)
-
-      if(Direction==Vertical)
-        return (CrossReturnType()
-                                 << _expression().col(0).cross(other),
-                                    _expression().col(1).cross(other),
-                                    _expression().col(2).cross(other)).finished();
-      else
-        return (CrossReturnType() 
-                                 << _expression().row(0).cross(other),
-                                    _expression().row(1).cross(other),
-                                    _expression().row(2).cross(other)).finished();
-    }
-
-  protected:
-    ExpressionTypeNested m_matrix;
-
-  private:
-    PartialRedux& operator=(const PartialRedux&);
-};
-
-/** \array_module
-  *
-  * \returns a PartialRedux wrapper of *this providing additional partial reduction operations
-  *
-  * Example: \include MatrixBase_colwise.cpp
-  * Output: \verbinclude MatrixBase_colwise.out
-  *
-  * \sa rowwise(), class PartialRedux
-  */
-template<typename Derived>
-inline const PartialRedux<Derived,Vertical>
-MatrixBase<Derived>::colwise() const
-{
-  return derived();
-}
-
-/** \array_module
-  *
-  * \returns a PartialRedux wrapper of *this providing additional partial reduction operations
-  *
-  * Example: \include MatrixBase_rowwise.cpp
-  * Output: \verbinclude MatrixBase_rowwise.out
-  *
-  * \sa colwise(), class PartialRedux
-  */
-template<typename Derived>
-inline const PartialRedux<Derived,Horizontal>
-MatrixBase<Derived>::rowwise() const
-{
-  return derived();
-}
-
-/** \returns a row or column vector expression of \c *this reduxed by \a func
-  *
-  * The template parameter \a BinaryOp is the type of the functor
-  * of the custom redux operator. Note that func must be an associative operator.
-  *
-  * \sa class PartialRedux, MatrixBase::colwise(), MatrixBase::rowwise()
-  */
-template<typename ExpressionType, int Direction>
-template<typename BinaryOp>
-const typename PartialRedux<ExpressionType,Direction>::template ReduxReturnType<BinaryOp>::Type
-PartialRedux<ExpressionType,Direction>::redux(const BinaryOp& func) const
-{
-  return typename ReduxReturnType<BinaryOp>::Type(_expression(), func);
-}
-
-#endif // EIGEN_PARTIAL_REDUX_H

Eigen/src/Array/Random.h

@@ -1,156 +0,0 @@
-// This file is part of Eigen, a lightweight C++ template library
-// for linear algebra. Eigen itself is part of the KDE project.
-//
-// Copyright (C) 2008 Gael Guennebaud <g.gael@free.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_RANDOM_H
-#define EIGEN_RANDOM_H
-
-template<typename Scalar> struct ei_scalar_random_op EIGEN_EMPTY_STRUCT {
-  inline ei_scalar_random_op(void) {}
-  inline const Scalar operator() (int, int) const { return ei_random<Scalar>(); }
-};
-template<typename Scalar>
-struct ei_functor_traits<ei_scalar_random_op<Scalar> >
-{ enum { Cost = 5 * NumTraits<Scalar>::MulCost, PacketAccess = false, IsRepeatable = false }; };
-
-/** \array_module
-  * 
-  * \returns a random matrix (not an expression, the matrix is immediately evaluated).
-  *
-  * The parameters \a rows and \a cols are the number of rows and of columns of
-  * the returned matrix. Must be compatible with this MatrixBase type.
-  *
-  * This variant is meant to be used for dynamic-size matrix types. For fixed-size types,
-  * it is redundant to pass \a rows and \a cols as arguments, so ei_random() should be used
-  * instead.
-  *
-  * \addexample RandomExample \label How to create a matrix with random coefficients
-  *
-  * Example: \include MatrixBase_random_int_int.cpp
-  * Output: \verbinclude MatrixBase_random_int_int.out
-  *
-  * \sa MatrixBase::setRandom(), MatrixBase::Random(int), MatrixBase::Random()
-  */
-template<typename Derived>
-inline const CwiseNullaryOp<ei_scalar_random_op<typename ei_traits<Derived>::Scalar>, Derived>
-MatrixBase<Derived>::Random(int rows, int cols)
-{
-  return NullaryExpr(rows, cols, ei_scalar_random_op<Scalar>());
-}
-
-/** \array_module
-  * 
-  * \returns a random vector (not an expression, the vector is immediately evaluated).
-  *
-  * The parameter \a size is the size of the returned vector.
-  * Must be compatible with this MatrixBase type.
-  *
-  * \only_for_vectors
-  *
-  * This variant is meant to be used for dynamic-size vector types. For fixed-size types,
-  * it is redundant to pass \a size as argument, so ei_random() should be used
-  * instead.
-  *
-  * Example: \include MatrixBase_random_int.cpp
-  * Output: \verbinclude MatrixBase_random_int.out
-  *
-  * \sa MatrixBase::setRandom(), MatrixBase::Random(int,int), MatrixBase::Random()
-  */
-template<typename Derived>
-inline const CwiseNullaryOp<ei_scalar_random_op<typename ei_traits<Derived>::Scalar>, Derived>
-MatrixBase<Derived>::Random(int size)
-{
-  return NullaryExpr(size, ei_scalar_random_op<Scalar>());
-}
-
-/** \array_module
-  * 
-  * \returns a fixed-size random matrix or vector
-  * (not an expression, the matrix is immediately evaluated).
-  *
-  * This variant is only for fixed-size MatrixBase types. For dynamic-size types, you
-  * need to use the variants taking size arguments.
-  *
-  * Example: \include MatrixBase_random.cpp
-  * Output: \verbinclude MatrixBase_random.out
-  *
-  * \sa MatrixBase::setRandom(), MatrixBase::Random(int,int), MatrixBase::Random(int)
-  */
-template<typename Derived>
-inline const CwiseNullaryOp<ei_scalar_random_op<typename ei_traits<Derived>::Scalar>, Derived>
-MatrixBase<Derived>::Random()
-{
-  return NullaryExpr(RowsAtCompileTime, ColsAtCompileTime, ei_scalar_random_op<Scalar>());
-}
-
-/** \array_module
-  * 
-  * Sets all coefficients in this expression to random values.
-  *
-  * Example: \include MatrixBase_setRandom.cpp
-  * Output: \verbinclude MatrixBase_setRandom.out
-  *
-  * \sa class CwiseNullaryOp, setRandom(int), setRandom(int,int)
-  */
-template<typename Derived>
-inline Derived& MatrixBase<Derived>::setRandom()
-{
-  return *this = Random(rows(), cols());
-}
-
-/** Resizes to the given \a size, and sets all coefficients in this expression to random values.
-  *
-  * \only_for_vectors
-  *
-  * Example: \include Matrix_setRandom_int.cpp
-  * Output: \verbinclude Matrix_setRandom_int.out
-  *
-  * \sa MatrixBase::setRandom(), setRandom(int,int), class CwiseNullaryOp, MatrixBase::Random()
-  */
-template<typename _Scalar, int _Rows, int _Cols, int _Options, int _MaxRows, int _MaxCols>
-EIGEN_STRONG_INLINE Matrix<_Scalar, _Rows, _Cols, _Options, _MaxRows, _MaxCols>&
-Matrix<_Scalar, _Rows, _Cols, _Options, _MaxRows, _MaxCols>::setRandom(int size)
-{
-  resize(size);
-  return setRandom();
-}
-
-/** Resizes to the given size, and sets all coefficients in this expression to random values.
-  *
-  * \param rows the new number of rows
-  * \param cols the new number of columns
-  *
-  * Example: \include Matrix_setRandom_int_int.cpp
-  * Output: \verbinclude Matrix_setRandom_int_int.out
-  *
-  * \sa MatrixBase::setRandom(), setRandom(int), class CwiseNullaryOp, MatrixBase::Random()
-  */
-template<typename _Scalar, int _Rows, int _Cols, int _Options, int _MaxRows, int _MaxCols>
-EIGEN_STRONG_INLINE Matrix<_Scalar, _Rows, _Cols, _Options, _MaxRows, _MaxCols>&
-Matrix<_Scalar, _Rows, _Cols, _Options, _MaxRows, _MaxCols>::setRandom(int rows, int cols)
-{
-  resize(rows, cols);
-  return setRandom();
-}
-
-#endif // EIGEN_RANDOM_H

Eigen/src/Array/Select.h

@@ -1,159 +0,0 @@
-// This file is part of Eigen, a lightweight C++ template library
-// for linear algebra. Eigen itself is part of the KDE project.
-//
-// Copyright (C) 2008 Gael Guennebaud <g.gael@free.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_SELECT_H
-#define EIGEN_SELECT_H
-
-/** \array_module \ingroup Array
-  *
-  * \class Select
-  *
-  * \brief Expression of a coefficient wise version of the C++ ternary operator ?:
-  *
-  * \param ConditionMatrixType the type of the \em condition expression which must be a boolean matrix
-  * \param ThenMatrixType the type of the \em then expression
-  * \param ElseMatrixType the type of the \em else expression
-  *
-  * This class represents an expression of a coefficient wise version of the C++ ternary operator ?:.
-  * It is the return type of MatrixBase::select() and most of the time this is the only way it is used.
-  *
-  * \sa MatrixBase::select(const MatrixBase<ThenDerived>&, const MatrixBase<ElseDerived>&) const
-  */
-
-template<typename ConditionMatrixType, typename ThenMatrixType, typename ElseMatrixType>
-struct ei_traits<Select<ConditionMatrixType, ThenMatrixType, ElseMatrixType> >
-{
-  typedef typename ei_traits<ThenMatrixType>::Scalar Scalar;
-  typedef typename ConditionMatrixType::Nested ConditionMatrixNested;
-  typedef typename ThenMatrixType::Nested ThenMatrixNested;
-  typedef typename ElseMatrixType::Nested ElseMatrixNested;
-  enum {
-    RowsAtCompileTime = ConditionMatrixType::RowsAtCompileTime,
-    ColsAtCompileTime = ConditionMatrixType::ColsAtCompileTime,
-    MaxRowsAtCompileTime = ConditionMatrixType::MaxRowsAtCompileTime,
-    MaxColsAtCompileTime = ConditionMatrixType::MaxColsAtCompileTime,
-    Flags = (unsigned int)ThenMatrixType::Flags & ElseMatrixType::Flags & HereditaryBits,
-	CoeffReadCost = ei_traits<typename ei_cleantype<ConditionMatrixNested>::type>::CoeffReadCost
-	+ EIGEN_ENUM_MAX(ei_traits<typename ei_cleantype<ThenMatrixNested>::type>::CoeffReadCost,
-	                 ei_traits<typename ei_cleantype<ElseMatrixNested>::type>::CoeffReadCost)
-  };
-};
-
-template<typename ConditionMatrixType, typename ThenMatrixType, typename ElseMatrixType>
-class Select : ei_no_assignment_operator,
-  public MatrixBase<Select<ConditionMatrixType, ThenMatrixType, ElseMatrixType> >
-{
-  public:
-
-    EIGEN_GENERIC_PUBLIC_INTERFACE(Select)
-
-    Select(const ConditionMatrixType& conditionMatrix,
-           const ThenMatrixType& thenMatrix,
-           const ElseMatrixType& elseMatrix)
-      : m_condition(conditionMatrix), m_then(thenMatrix), m_else(elseMatrix)
-    {
-      ei_assert(m_condition.rows() == m_then.rows() && m_condition.rows() == m_else.rows());
-      ei_assert(m_condition.cols() == m_then.cols() && m_condition.cols() == m_else.cols());
-    }
-
-    int rows() const { return m_condition.rows(); }
-    int cols() const { return m_condition.cols(); }
-
-    const Scalar coeff(int i, int j) const
-    {
-      if (m_condition.coeff(i,j))
-        return m_then.coeff(i,j);
-      else
-        return m_else.coeff(i,j);
-    }
-    
-    const Scalar coeff(int i) const
-    {
-      if (m_condition.coeff(i))
-        return m_then.coeff(i);
-      else
-        return m_else.coeff(i);
-    }
-
-  protected:
-    const typename ConditionMatrixType::Nested m_condition;
-    const typename ThenMatrixType::Nested m_then;
-    const typename ElseMatrixType::Nested m_else;
-};
-
-
-/** \array_module
-  *
-  * \returns a matrix where each coefficient (i,j) is equal to \a thenMatrix(i,j)
-  * if \c *this(i,j), and \a elseMatrix(i,j) otherwise.
-  *
-  * Example: \include MatrixBase_select.cpp
-  * Output: \verbinclude MatrixBase_select.out
-  *
-  * \sa class Select
-  */
-template<typename Derived>
-template<typename ThenDerived,typename ElseDerived>
-inline const Select<Derived,ThenDerived,ElseDerived>
-MatrixBase<Derived>::select(const MatrixBase<ThenDerived>& thenMatrix,
-                            const MatrixBase<ElseDerived>& elseMatrix) const
-{
-  return Select<Derived,ThenDerived,ElseDerived>(derived(), thenMatrix.derived(), elseMatrix.derived());
-}
-
-/** \array_module
-  *
-  * Version of MatrixBase::select(const MatrixBase&, const MatrixBase&) with
-  * the \em else expression being a scalar value.
-  *
-  * \sa MatrixBase::select(const MatrixBase<ThenDerived>&, const MatrixBase<ElseDerived>&) const, class Select
-  */
-template<typename Derived>
-template<typename ThenDerived>
-inline const Select<Derived,ThenDerived, NestByValue<typename ThenDerived::ConstantReturnType> >
-MatrixBase<Derived>::select(const MatrixBase<ThenDerived>& thenMatrix,
-                            typename ThenDerived::Scalar elseScalar) const
-{
-  return Select<Derived,ThenDerived,NestByValue<typename ThenDerived::ConstantReturnType> >(
-    derived(), thenMatrix.derived(), ThenDerived::Constant(rows(),cols(),elseScalar));
-}
-
-/** \array_module
-  *
-  * Version of MatrixBase::select(const MatrixBase&, const MatrixBase&) with
-  * the \em then expression being a scalar value.
-  *
-  * \sa MatrixBase::select(const MatrixBase<ThenDerived>&, const MatrixBase<ElseDerived>&) const, class Select
-  */
-template<typename Derived>
-template<typename ElseDerived>
-inline const Select<Derived, NestByValue<typename ElseDerived::ConstantReturnType>, ElseDerived >
-MatrixBase<Derived>::select(typename ElseDerived::Scalar thenScalar,
-                            const MatrixBase<ElseDerived>& elseMatrix) const
-{
-  return Select<Derived,NestByValue<typename ElseDerived::ConstantReturnType>,ElseDerived>(
-    derived(), ElseDerived::Constant(rows(),cols(),thenScalar), elseMatrix.derived());
-}
-
-#endif // EIGEN_SELECT_H

Eigen/src/CMakeLists.txt

@@ -1,9 +0,0 @@
-ADD_SUBDIRECTORY(Core)
-ADD_SUBDIRECTORY(LU)
-ADD_SUBDIRECTORY(QR)
-ADD_SUBDIRECTORY(SVD)
-ADD_SUBDIRECTORY(Cholesky)
-ADD_SUBDIRECTORY(Array)
-ADD_SUBDIRECTORY(Geometry)
-ADD_SUBDIRECTORY(LeastSquares)
-ADD_SUBDIRECTORY(Sparse)

Eigen/src/Cholesky/CMakeLists.txt

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

Eigen/src/Cholesky/CholeskyInstantiations.cpp

@@ -1,35 +0,0 @@
-// This file is part of Eigen, a lightweight C++ template library
-// for linear algebra. Eigen itself is part of the KDE project.
-//
-// Copyright (C) 2008 Gael Guennebaud <g.gael@free.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_EXTERN_INSTANTIATIONS
-#define EIGEN_EXTERN_INSTANTIATIONS
-#endif
-#include "../../Core"
-#undef EIGEN_EXTERN_INSTANTIATIONS
-
-#include "../../Cholesky"
-
-namespace Eigen {
-  EIGEN_CHOLESKY_MODULE_INSTANTIATE();
-}

Eigen/src/Cholesky/LDLT.h

@@ -1,198 +1,673 @@
 // This file is part of Eigen, a lightweight C++ template library
-// for linear algebra. Eigen itself is part of the KDE project.
+// for linear algebra.
 //
-// Copyright (C) 2008 Gael Guennebaud <g.gael@free.fr>
+// Copyright (C) 2008-2011 Gael Guennebaud <gael.guennebaud@inria.fr>
+// Copyright (C) 2009 Keir Mierle <mierle@gmail.com>
+// Copyright (C) 2009 Benoit Jacob <jacob.benoit.1@gmail.com>
+// Copyright (C) 2011 Timothy E. Holy <tim.holy@gmail.com >
 //
-// 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/>.
+// This Source Code Form is subject to the terms of the Mozilla
+// Public License v. 2.0. If a copy of the MPL was not distributed
+// with this file, You can obtain one at http://mozilla.org/MPL/2.0/.
 
 #ifndef EIGEN_LDLT_H
 #define EIGEN_LDLT_H
 
-/** \ingroup cholesky_Module
+namespace Eigen {
+
+namespace internal {
+  template<typename MatrixType, int UpLo> struct LDLT_Traits;
+
+  // PositiveSemiDef means positive semi-definite and non-zero; same for NegativeSemiDef
+  enum SignMatrix { PositiveSemiDef, NegativeSemiDef, ZeroSign, Indefinite };
+}
+
+/** \ingroup Cholesky_Module
   *
   * \class LDLT
   *
-  * \brief Robust Cholesky decomposition of a matrix and associated features
+  * \brief Robust Cholesky decomposition of a matrix with pivoting
   *
-  * \param MatrixType the type of the matrix of which we are computing the LDL^T Cholesky decomposition
+  * \tparam _MatrixType the type of the matrix of which to compute the LDL^T Cholesky decomposition
+  * \tparam _UpLo the triangular part that will be used for the decompositon: Lower (default) or Upper.
+  *             The other triangular part won't be read.
   *
-  * This class performs a Cholesky decomposition without square root of a symmetric, positive definite
-  * matrix A such that A = L D L^* = U^* D U, where L is lower triangular with a unit diagonal
-  * and D is a diagonal matrix.
+  * Perform a robust Cholesky decomposition of a positive semidefinite or negative semidefinite
+  * matrix \f$ A \f$ such that \f$ A =  P^TLDL^*P \f$, where P is a permutation matrix, L
+  * is lower triangular with a unit diagonal and D is a diagonal matrix.
   *
-  * Compared to a standard Cholesky decomposition, avoiding the square roots allows for faster and more
-  * stable computation.
+  * The decomposition uses pivoting to ensure stability, so that L will have
+  * zeros in the bottom right rank(A) - n submatrix. Avoiding the square root
+  * on D also stabilizes the computation.
   *
-  * Note that during the decomposition, only the upper triangular part of A is considered. Therefore,
-  * the strict lower part does not have to store correct values.
+  * Remember that Cholesky decompositions are not rank-revealing. Also, do not use a Cholesky
+  * decomposition to determine whether a system of equations has a solution.
   *
-  * \sa MatrixBase::ldlt(), class LLT
+  * This class supports the \link InplaceDecomposition inplace decomposition \endlink mechanism.
+  * 
+  * \sa MatrixBase::ldlt(), SelfAdjointView::ldlt(), class LLT
   */
-template<typename MatrixType> class LDLT
+template<typename _MatrixType, int _UpLo> class LDLT
 {
   public:
-
+    typedef _MatrixType MatrixType;
+    enum {
+      RowsAtCompileTime = MatrixType::RowsAtCompileTime,
+      ColsAtCompileTime = MatrixType::ColsAtCompileTime,
+      MaxRowsAtCompileTime = MatrixType::MaxRowsAtCompileTime,
+      MaxColsAtCompileTime = MatrixType::MaxColsAtCompileTime,
+      UpLo = _UpLo
+    };
     typedef typename MatrixType::Scalar Scalar;
     typedef typename NumTraits<typename MatrixType::Scalar>::Real RealScalar;
-    typedef Matrix<Scalar, MatrixType::ColsAtCompileTime, 1> VectorType;
+    typedef Eigen::Index Index; ///< \deprecated since Eigen 3.3
+    typedef typename MatrixType::StorageIndex StorageIndex;
+    typedef Matrix<Scalar, RowsAtCompileTime, 1, 0, MaxRowsAtCompileTime, 1> TmpMatrixType;
 
-    LDLT(const MatrixType& matrix)
-      : m_matrix(matrix.rows(), matrix.cols())
+    typedef Transpositions<RowsAtCompileTime, MaxRowsAtCompileTime> TranspositionType;
+    typedef PermutationMatrix<RowsAtCompileTime, MaxRowsAtCompileTime> PermutationType;
+
+    typedef internal::LDLT_Traits<MatrixType,UpLo> Traits;
+
+    /** \brief Default Constructor.
+      *
+      * The default constructor is useful in cases in which the user intends to
+      * perform decompositions via LDLT::compute(const MatrixType&).
+      */
+    LDLT()
+      : m_matrix(),
+        m_transpositions(),
+        m_sign(internal::ZeroSign),
+        m_isInitialized(false)
+    {}
+
+    /** \brief Default Constructor with memory preallocation
+      *
+      * Like the default constructor but with preallocation of the internal data
+      * according to the specified problem \a size.
+      * \sa LDLT()
+      */
+    explicit LDLT(Index size)
+      : m_matrix(size, size),
+        m_transpositions(size),
+        m_temporary(size),
+        m_sign(internal::ZeroSign),
+        m_isInitialized(false)
+    {}
+
+    /** \brief Constructor with decomposition
+      *
+      * This calculates the decomposition for the input \a matrix.
+      *
+      * \sa LDLT(Index size)
+      */
+    template<typename InputType>
+    explicit LDLT(const EigenBase<InputType>& matrix)
+      : m_matrix(matrix.rows(), matrix.cols()),
+        m_transpositions(matrix.rows()),
+        m_temporary(matrix.rows()),
+        m_sign(internal::ZeroSign),
+        m_isInitialized(false)
     {
-      compute(matrix);
+      compute(matrix.derived());
     }
 
-    /** \returns the lower triangular matrix L */
-    inline Part<MatrixType, UnitLowerTriangular> matrixL(void) const { return m_matrix; }
+    /** \brief Constructs a LDLT factorization from a given matrix
+      *
+      * This overloaded constructor is provided for \link InplaceDecomposition inplace decomposition \endlink when \c MatrixType is a Eigen::Ref.
+      *
+      * \sa LDLT(const EigenBase&)
+      */
+    template<typename InputType>
+    explicit LDLT(EigenBase<InputType>& matrix)
+      : m_matrix(matrix.derived()),
+        m_transpositions(matrix.rows()),
+        m_temporary(matrix.rows()),
+        m_sign(internal::ZeroSign),
+        m_isInitialized(false)
+    {
+      compute(matrix.derived());
+    }
+
+    /** Clear any existing decomposition
+     * \sa rankUpdate(w,sigma)
+     */
+    void setZero()
+    {
+      m_isInitialized = false;
+    }
+
+    /** \returns a view of the upper triangular matrix U */
+    inline typename Traits::MatrixU matrixU() const
+    {
+      eigen_assert(m_isInitialized && "LDLT is not initialized.");
+      return Traits::getU(m_matrix);
+    }
+
+    /** \returns a view of the lower triangular matrix L */
+    inline typename Traits::MatrixL matrixL() const
+    {
+      eigen_assert(m_isInitialized && "LDLT is not initialized.");
+      return Traits::getL(m_matrix);
+    }
+
+    /** \returns the permutation matrix P as a transposition sequence.
+      */
+    inline const TranspositionType& transpositionsP() const
+    {
+      eigen_assert(m_isInitialized && "LDLT is not initialized.");
+      return m_transpositions;
+    }
 
     /** \returns the coefficients of the diagonal matrix D */
-    inline DiagonalCoeffs<MatrixType> vectorD(void) const { return m_matrix.diagonal(); }
+    inline Diagonal<const MatrixType> vectorD() const
+    {
+      eigen_assert(m_isInitialized && "LDLT is not initialized.");
+      return m_matrix.diagonal();
+    }
 
-    /** \returns true if the matrix is positive definite */
-    inline bool isPositiveDefinite(void) const { return m_isPositiveDefinite; }
+    /** \returns true if the matrix is positive (semidefinite) */
+    inline bool isPositive() const
+    {
+      eigen_assert(m_isInitialized && "LDLT is not initialized.");
+      return m_sign == internal::PositiveSemiDef || m_sign == internal::ZeroSign;
+    }
 
-    template<typename RhsDerived, typename ResultType>
-    bool solve(const MatrixBase<RhsDerived> &b, ResultType *result) const;
+    /** \returns true if the matrix is negative (semidefinite) */
+    inline bool isNegative(void) const
+    {
+      eigen_assert(m_isInitialized && "LDLT is not initialized.");
+      return m_sign == internal::NegativeSemiDef || m_sign == internal::ZeroSign;
+    }
+
+    /** \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(), SelfAdjointView::ldlt()
+      */
+    template<typename Rhs>
+    inline const Solve<LDLT, Rhs>
+    solve(const MatrixBase<Rhs>& b) const
+    {
+      eigen_assert(m_isInitialized && "LDLT is not initialized.");
+      eigen_assert(m_matrix.rows()==b.rows()
+                && "LDLT::solve(): invalid number of rows of the right hand side matrix b");
+      return Solve<LDLT, Rhs>(*this, b.derived());
+    }
 
     template<typename Derived>
     bool solveInPlace(MatrixBase<Derived> &bAndX) const;
 
-    void compute(const MatrixType& matrix);
+    template<typename InputType>
+    LDLT& compute(const EigenBase<InputType>& matrix);
+
+    /** \returns an estimate of the reciprocal condition number of the matrix of
+     *  which \c *this is the LDLT decomposition.
+     */
+    RealScalar rcond() const
+    {
+      eigen_assert(m_isInitialized && "LDLT is not initialized.");
+      return internal::rcond_estimate_helper(m_l1_norm, *this);
+    }
+
+    template <typename Derived>
+    LDLT& rankUpdate(const MatrixBase<Derived>& w, const RealScalar& alpha=1);
+
+    /** \returns the internal LDLT decomposition matrix
+      *
+      * TODO: document the storage layout
+      */
+    inline const MatrixType& matrixLDLT() const
+    {
+      eigen_assert(m_isInitialized && "LDLT is not initialized.");
+      return m_matrix;
+    }
+
+    MatrixType reconstructedMatrix() const;
+
+    /** \returns the adjoint of \c *this, that is, a const reference to the decomposition itself as the underlying matrix is self-adjoint.
+      *
+      * This method is provided for compatibility with other matrix decompositions, thus enabling generic code such as:
+      * \code x = decomposition.adjoint().solve(b) \endcode
+      */
+    const LDLT& adjoint() const { return *this; };
+
+    inline Index rows() const { return m_matrix.rows(); }
+    inline Index cols() const { return m_matrix.cols(); }
+
+    /** \brief Reports whether previous computation was successful.
+      *
+      * \returns \c Success if computation was succesful,
+      *          \c NumericalIssue if the factorization failed because of a zero pivot.
+      */
+    ComputationInfo info() const
+    {
+      eigen_assert(m_isInitialized && "LDLT is not initialized.");
+      return m_info;
+    }
+
+    #ifndef EIGEN_PARSED_BY_DOXYGEN
+    template<typename RhsType, typename DstType>
+    EIGEN_DEVICE_FUNC
+    void _solve_impl(const RhsType &rhs, DstType &dst) const;
+    #endif
 
   protected:
+
+    static void check_template_parameters()
+    {
+      EIGEN_STATIC_ASSERT_NON_INTEGER(Scalar);
+    }
+
     /** \internal
-      * Used to compute and store the cholesky decomposition A = L D L^* = U^* D U.
+      * Used to compute and store the Cholesky decomposition A = L D L^* = U^* D U.
       * The strict upper part is used during the decomposition, the strict lower
       * part correspond to the coefficients of L (its diagonal is equal to 1 and
       * is not stored), and the diagonal entries correspond to D.
       */
     MatrixType m_matrix;
-
-    bool m_isPositiveDefinite;
+    RealScalar m_l1_norm;
+    TranspositionType m_transpositions;
+    TmpMatrixType m_temporary;
+    internal::SignMatrix m_sign;
+    bool m_isInitialized;
+    ComputationInfo m_info;
 };
 
-/** Compute / recompute the LLT decomposition A = L D L^* = U^* D U of \a matrix
-  */
-template<typename MatrixType>
-void LDLT<MatrixType>::compute(const MatrixType& a)
+namespace internal {
+
+template<int UpLo> struct ldlt_inplace;
+
+template<> struct ldlt_inplace<Lower>
 {
-  assert(a.rows()==a.cols());
-  const int size = a.rows();
-  m_matrix.resize(size, size);
-  m_isPositiveDefinite = true;
-  const RealScalar eps = ei_sqrt(precision<Scalar>());
-
-  if (size<=1)
+  template<typename MatrixType, typename TranspositionType, typename Workspace>
+  static bool unblocked(MatrixType& mat, TranspositionType& transpositions, Workspace& temp, SignMatrix& sign)
   {
-    m_matrix = a;
-    return;
-  }
+    using std::abs;
+    typedef typename MatrixType::Scalar Scalar;
+    typedef typename MatrixType::RealScalar RealScalar;
+    typedef typename TranspositionType::StorageIndex IndexType;
+    eigen_assert(mat.rows()==mat.cols());
+    const Index size = mat.rows();
+    bool found_zero_pivot = false;
+    bool ret = true;
 
-  // Let's preallocate a temporay vector to evaluate the matrix-vector product into it.
-  // Unlike the standard LLT decomposition, here we cannot evaluate it to the destination
-  // matrix because it a sub-row which is not compatible suitable for efficient packet evaluation.
-  // (at least if we assume the matrix is col-major)
-  Matrix<Scalar,MatrixType::RowsAtCompileTime,1> _temporary(size);
-
-  // Note that, in this algorithm the rows of the strict upper part of m_matrix is used to store
-  // column vector, thus the strange .conjugate() and .transpose()...
-
-  m_matrix.row(0) = a.row(0).conjugate();
-  m_matrix.col(0).end(size-1) = m_matrix.row(0).end(size-1) / m_matrix.coeff(0,0);
-  for (int j = 1; j < size; ++j)
-  {
-    RealScalar tmp = ei_real(a.coeff(j,j) - (m_matrix.row(j).start(j) * m_matrix.col(j).start(j).conjugate()).coeff(0,0));
-    m_matrix.coeffRef(j,j) = tmp;
-
-    if (tmp < eps)
+    if (size <= 1)
     {
-      m_isPositiveDefinite = false;
-      return;
+      transpositions.setIdentity();
+      if(size==0) sign = ZeroSign;
+      else if (numext::real(mat.coeff(0,0)) > static_cast<RealScalar>(0) ) sign = PositiveSemiDef;
+      else if (numext::real(mat.coeff(0,0)) < static_cast<RealScalar>(0)) sign = NegativeSemiDef;
+      else sign = ZeroSign;
+      return true;
     }
 
-    int endSize = size-j-1;
-    if (endSize>0)
+    for (Index k = 0; k < size; ++k)
     {
-      _temporary.end(endSize) = ( m_matrix.block(j+1,0, endSize, j)
-                                  * m_matrix.col(j).start(j).conjugate() ).lazy();
+      // Find largest diagonal element
+      Index index_of_biggest_in_corner;
+      mat.diagonal().tail(size-k).cwiseAbs().maxCoeff(&index_of_biggest_in_corner);
+      index_of_biggest_in_corner += k;
 
-      m_matrix.row(j).end(endSize) = a.row(j).end(endSize).conjugate()
-                                   - _temporary.end(endSize).transpose();
+      transpositions.coeffRef(k) = IndexType(index_of_biggest_in_corner);
+      if(k != index_of_biggest_in_corner)
+      {
+        // apply the transposition while taking care to consider only
+        // the lower triangular part
+        Index s = size-index_of_biggest_in_corner-1; // trailing size after the biggest element
+        mat.row(k).head(k).swap(mat.row(index_of_biggest_in_corner).head(k));
+        mat.col(k).tail(s).swap(mat.col(index_of_biggest_in_corner).tail(s));
+        std::swap(mat.coeffRef(k,k),mat.coeffRef(index_of_biggest_in_corner,index_of_biggest_in_corner));
+        for(Index i=k+1;i<index_of_biggest_in_corner;++i)
+        {
+          Scalar tmp = mat.coeffRef(i,k);
+          mat.coeffRef(i,k) = numext::conj(mat.coeffRef(index_of_biggest_in_corner,i));
+          mat.coeffRef(index_of_biggest_in_corner,i) = numext::conj(tmp);
+        }
+        if(NumTraits<Scalar>::IsComplex)
+          mat.coeffRef(index_of_biggest_in_corner,k) = numext::conj(mat.coeff(index_of_biggest_in_corner,k));
+      }
 
-      m_matrix.col(j).end(endSize) = m_matrix.row(j).end(endSize) / tmp;
+      // partition the matrix:
+      //       A00 |  -  |  -
+      // lu  = A10 | A11 |  -
+      //       A20 | A21 | A22
+      Index rs = size - k - 1;
+      Block<MatrixType,Dynamic,1> A21(mat,k+1,k,rs,1);
+      Block<MatrixType,1,Dynamic> A10(mat,k,0,1,k);
+      Block<MatrixType,Dynamic,Dynamic> A20(mat,k+1,0,rs,k);
+
+      if(k>0)
+      {
+        temp.head(k) = mat.diagonal().real().head(k).asDiagonal() * A10.adjoint();
+        mat.coeffRef(k,k) -= (A10 * temp.head(k)).value();
+        if(rs>0)
+          A21.noalias() -= A20 * temp.head(k);
+      }
+
+      // In some previous versions of Eigen (e.g., 3.2.1), the scaling was omitted if the pivot
+      // was smaller than the cutoff value. However, since LDLT is not rank-revealing
+      // we should only make sure that we do not introduce INF or NaN values.
+      // Remark that LAPACK also uses 0 as the cutoff value.
+      RealScalar realAkk = numext::real(mat.coeffRef(k,k));
+      bool pivot_is_valid = (abs(realAkk) > RealScalar(0));
+
+      if(k==0 && !pivot_is_valid)
+      {
+        // The entire diagonal is zero, there is nothing more to do
+        // except filling the transpositions, and checking whether the matrix is zero.
+        sign = ZeroSign;
+        for(Index j = 0; j<size; ++j)
+        {
+          transpositions.coeffRef(j) = IndexType(j);
+          ret = ret && (mat.col(j).tail(size-j-1).array()==Scalar(0)).all();
+        }
+        return ret;
+      }
+
+      if((rs>0) && pivot_is_valid)
+        A21 /= realAkk;
+      else if(rs>0)
+        ret = ret && (A21.array()==Scalar(0)).all();
+
+      if(found_zero_pivot && pivot_is_valid) ret = false; // factorization failed
+      else if(!pivot_is_valid) found_zero_pivot = true;
+
+      if (sign == PositiveSemiDef) {
+        if (realAkk < static_cast<RealScalar>(0)) sign = Indefinite;
+      } else if (sign == NegativeSemiDef) {
+        if (realAkk > static_cast<RealScalar>(0)) sign = Indefinite;
+      } else if (sign == ZeroSign) {
+        if (realAkk > static_cast<RealScalar>(0)) sign = PositiveSemiDef;
+        else if (realAkk < static_cast<RealScalar>(0)) sign = NegativeSemiDef;
+      }
     }
+
+    return ret;
   }
-}
 
-/** Computes the solution x of \f$ A x = b \f$ using the current decomposition of A.
-  * The result is stored in \a result
-  *
-  * \returns true in case of success, false otherwise.
-  *
-  * In other words, it computes \f$ b = A^{-1} b \f$ with
-  * \f$ {L^{*}}^{-1} D^{-1} L^{-1} b \f$ from right to left.
-  *
-  * \sa LDLT::solveInPlace(), MatrixBase::ldlt()
-  */
-template<typename MatrixType>
-template<typename RhsDerived, typename ResultType>
-bool LDLT<MatrixType>
-::solve(const MatrixBase<RhsDerived> &b, ResultType *result) const
+  // Reference for the algorithm: Davis and Hager, "Multiple Rank
+  // Modifications of a Sparse Cholesky Factorization" (Algorithm 1)
+  // Trivial rearrangements of their computations (Timothy E. Holy)
+  // allow their algorithm to work for rank-1 updates even if the
+  // original matrix is not of full rank.
+  // Here only rank-1 updates are implemented, to reduce the
+  // requirement for intermediate storage and improve accuracy
+  template<typename MatrixType, typename WDerived>
+  static bool updateInPlace(MatrixType& mat, MatrixBase<WDerived>& w, const typename MatrixType::RealScalar& sigma=1)
+  {
+    using numext::isfinite;
+    typedef typename MatrixType::Scalar Scalar;
+    typedef typename MatrixType::RealScalar RealScalar;
+
+    const Index size = mat.rows();
+    eigen_assert(mat.cols() == size && w.size()==size);
+
+    RealScalar alpha = 1;
+
+    // Apply the update
+    for (Index j = 0; j < size; j++)
+    {
+      // Check for termination due to an original decomposition of low-rank
+      if (!(isfinite)(alpha))
+        break;
+
+      // Update the diagonal terms
+      RealScalar dj = numext::real(mat.coeff(j,j));
+      Scalar wj = w.coeff(j);
+      RealScalar swj2 = sigma*numext::abs2(wj);
+      RealScalar gamma = dj*alpha + swj2;
+
+      mat.coeffRef(j,j) += swj2/alpha;
+      alpha += swj2/dj;
+
+
+      // Update the terms of L
+      Index rs = size-j-1;
+      w.tail(rs) -= wj * mat.col(j).tail(rs);
+      if(gamma != 0)
+        mat.col(j).tail(rs) += (sigma*numext::conj(wj)/gamma)*w.tail(rs);
+    }
+    return true;
+  }
+
+  template<typename MatrixType, typename TranspositionType, typename Workspace, typename WType>
+  static bool update(MatrixType& mat, const TranspositionType& transpositions, Workspace& tmp, const WType& w, const typename MatrixType::RealScalar& sigma=1)
+  {
+    // Apply the permutation to the input w
+    tmp = transpositions * w;
+
+    return ldlt_inplace<Lower>::updateInPlace(mat,tmp,sigma);
+  }
+};
+
+template<> struct ldlt_inplace<Upper>
 {
-  const int size = m_matrix.rows();
-  ei_assert(size==b.rows() && "LLT::solve(): invalid number of rows of the right hand side matrix b");
-  *result = b;
-  return solveInPlace(*result);
+  template<typename MatrixType, typename TranspositionType, typename Workspace>
+  static EIGEN_STRONG_INLINE bool unblocked(MatrixType& mat, TranspositionType& transpositions, Workspace& temp, SignMatrix& sign)
+  {
+    Transpose<MatrixType> matt(mat);
+    return ldlt_inplace<Lower>::unblocked(matt, transpositions, temp, sign);
+  }
+
+  template<typename MatrixType, typename TranspositionType, typename Workspace, typename WType>
+  static EIGEN_STRONG_INLINE bool update(MatrixType& mat, TranspositionType& transpositions, Workspace& tmp, WType& w, const typename MatrixType::RealScalar& sigma=1)
+  {
+    Transpose<MatrixType> matt(mat);
+    return ldlt_inplace<Lower>::update(matt, transpositions, tmp, w.conjugate(), sigma);
+  }
+};
+
+template<typename MatrixType> struct LDLT_Traits<MatrixType,Lower>
+{
+  typedef const TriangularView<const MatrixType, UnitLower> MatrixL;
+  typedef const TriangularView<const typename MatrixType::AdjointReturnType, UnitUpper> MatrixU;
+  static inline MatrixL getL(const MatrixType& m) { return MatrixL(m); }
+  static inline MatrixU getU(const MatrixType& m) { return MatrixU(m.adjoint()); }
+};
+
+template<typename MatrixType> struct LDLT_Traits<MatrixType,Upper>
+{
+  typedef const TriangularView<const typename MatrixType::AdjointReturnType, UnitLower> MatrixL;
+  typedef const TriangularView<const MatrixType, UnitUpper> MatrixU;
+  static inline MatrixL getL(const MatrixType& m) { return MatrixL(m.adjoint()); }
+  static inline MatrixU getU(const MatrixType& m) { return MatrixU(m); }
+};
+
+} // end namespace internal
+
+/** Compute / recompute the LDLT decomposition A = L D L^* = U^* D U of \a matrix
+  */
+template<typename MatrixType, int _UpLo>
+template<typename InputType>
+LDLT<MatrixType,_UpLo>& LDLT<MatrixType,_UpLo>::compute(const EigenBase<InputType>& a)
+{
+  check_template_parameters();
+
+  eigen_assert(a.rows()==a.cols());
+  const Index size = a.rows();
+
+  m_matrix = a.derived();
+
+  // Compute matrix L1 norm = max abs column sum.
+  m_l1_norm = RealScalar(0);
+  // TODO move this code to SelfAdjointView
+  for (Index col = 0; col < size; ++col) {
+    RealScalar abs_col_sum;
+    if (_UpLo == Lower)
+      abs_col_sum = m_matrix.col(col).tail(size - col).template lpNorm<1>() + m_matrix.row(col).head(col).template lpNorm<1>();
+    else
+      abs_col_sum = m_matrix.col(col).head(col).template lpNorm<1>() + m_matrix.row(col).tail(size - col).template lpNorm<1>();
+    if (abs_col_sum > m_l1_norm)
+      m_l1_norm = abs_col_sum;
+  }
+
+  m_transpositions.resize(size);
+  m_isInitialized = false;
+  m_temporary.resize(size);
+  m_sign = internal::ZeroSign;
+
+  m_info = internal::ldlt_inplace<UpLo>::unblocked(m_matrix, m_transpositions, m_temporary, m_sign) ? Success : NumericalIssue;
+
+  m_isInitialized = true;
+  return *this;
 }
 
-/** This is the \em in-place version of solve().
+/** Update the LDLT decomposition:  given A = L D L^T, efficiently compute the decomposition of A + sigma w w^T.
+ * \param w a vector to be incorporated into the decomposition.
+ * \param sigma a scalar, +1 for updates and -1 for "downdates," which correspond to removing previously-added column vectors. Optional; default value is +1.
+ * \sa setZero()
+  */
+template<typename MatrixType, int _UpLo>
+template<typename Derived>
+LDLT<MatrixType,_UpLo>& LDLT<MatrixType,_UpLo>::rankUpdate(const MatrixBase<Derived>& w, const typename LDLT<MatrixType,_UpLo>::RealScalar& sigma)
+{
+  typedef typename TranspositionType::StorageIndex IndexType;
+  const Index size = w.rows();
+  if (m_isInitialized)
+  {
+    eigen_assert(m_matrix.rows()==size);
+  }
+  else
+  {
+    m_matrix.resize(size,size);
+    m_matrix.setZero();
+    m_transpositions.resize(size);
+    for (Index i = 0; i < size; i++)
+      m_transpositions.coeffRef(i) = IndexType(i);
+    m_temporary.resize(size);
+    m_sign = sigma>=0 ? internal::PositiveSemiDef : internal::NegativeSemiDef;
+    m_isInitialized = true;
+  }
+
+  internal::ldlt_inplace<UpLo>::update(m_matrix, m_transpositions, m_temporary, w, sigma);
+
+  return *this;
+}
+
+#ifndef EIGEN_PARSED_BY_DOXYGEN
+template<typename _MatrixType, int _UpLo>
+template<typename RhsType, typename DstType>
+void LDLT<_MatrixType,_UpLo>::_solve_impl(const RhsType &rhs, DstType &dst) const
+{
+  eigen_assert(rhs.rows() == rows());
+  // dst = P b
+  dst = m_transpositions * rhs;
+
+  // dst = L^-1 (P b)
+  matrixL().solveInPlace(dst);
+
+  // dst = D^-1 (L^-1 P b)
+  // more precisely, use pseudo-inverse of D (see bug 241)
+  using std::abs;
+  const typename Diagonal<const MatrixType>::RealReturnType vecD(vectorD());
+  // In some previous versions, tolerance was set to the max of 1/highest (or rather numeric_limits::min())
+  // and the maximal diagonal entry * epsilon as motivated by LAPACK's xGELSS:
+  // RealScalar tolerance = numext::maxi(vecD.array().abs().maxCoeff() * NumTraits<RealScalar>::epsilon(),RealScalar(1) / NumTraits<RealScalar>::highest());
+  // However, LDLT is not rank revealing, and so adjusting the tolerance wrt to the highest
+  // diagonal element is not well justified and leads to numerical issues in some cases.
+  // Moreover, Lapack's xSYTRS routines use 0 for the tolerance.
+  // Using numeric_limits::min() gives us more robustness to denormals.
+  RealScalar tolerance = (std::numeric_limits<RealScalar>::min)();
+
+  for (Index i = 0; i < vecD.size(); ++i)
+  {
+    if(abs(vecD(i)) > tolerance)
+      dst.row(i) /= vecD(i);
+    else
+      dst.row(i).setZero();
+  }
+
+  // dst = L^-T (D^-1 L^-1 P b)
+  matrixU().solveInPlace(dst);
+
+  // dst = P^-1 (L^-T D^-1 L^-1 P b) = A^-1 b
+  dst = m_transpositions.transpose() * dst;
+}
+#endif
+
+/** \internal use x = ldlt_object.solve(x);
+  *
+  * This is the \em in-place version of solve().
   *
   * \param bAndX represents both the right-hand side matrix b and result x.
   *
+  * \returns true always! If you need to check for existence of solutions, use another decomposition like LU, QR, or SVD.
+  *
   * This version avoids a copy when the right hand side matrix b is not
   * needed anymore.
   *
   * \sa LDLT::solve(), MatrixBase::ldlt()
   */
-template<typename MatrixType>
+template<typename MatrixType,int _UpLo>
 template<typename Derived>
-bool LDLT<MatrixType>::solveInPlace(MatrixBase<Derived> &bAndX) const
+bool LDLT<MatrixType,_UpLo>::solveInPlace(MatrixBase<Derived> &bAndX) const
 {
-  const int size = m_matrix.rows();
-  ei_assert(size==bAndX.rows());
-  if (!m_isPositiveDefinite)
-    return false;
-  matrixL().solveTriangularInPlace(bAndX);
-  bAndX = (m_matrix.cwise().inverse().template part<Diagonal>() * bAndX).lazy();
-  m_matrix.adjoint().template part<UnitUpperTriangular>().solveTriangularInPlace(bAndX);
+  eigen_assert(m_isInitialized && "LDLT is not initialized.");
+  eigen_assert(m_matrix.rows() == bAndX.rows());
+
+  bAndX = this->solve(bAndX);
+
   return true;
 }
 
-/** \cholesky_module
-  * \returns the Cholesky decomposition without square root of \c *this
-  */
-template<typename Derived>
-inline const LDLT<typename MatrixBase<Derived>::PlainMatrixType>
-MatrixBase<Derived>::ldlt() const
+/** \returns the matrix represented by the decomposition,
+ * i.e., it returns the product: P^T L D L^* P.
+ * This function is provided for debug purpose. */
+template<typename MatrixType, int _UpLo>
+MatrixType LDLT<MatrixType,_UpLo>::reconstructedMatrix() const
 {
-  return derived();
+  eigen_assert(m_isInitialized && "LDLT is not initialized.");
+  const Index size = m_matrix.rows();
+  MatrixType res(size,size);
+
+  // P
+  res.setIdentity();
+  res = transpositionsP() * res;
+  // L^* P
+  res = matrixU() * res;
+  // D(L^*P)
+  res = vectorD().real().asDiagonal() * res;
+  // L(DL^*P)
+  res = matrixL() * res;
+  // P^T (LDL^*P)
+  res = transpositionsP().transpose() * res;
+
+  return res;
 }
 
+/** \cholesky_module
+  * \returns the Cholesky decomposition with full pivoting without square root of \c *this
+  * \sa MatrixBase::ldlt()
+  */
+template<typename MatrixType, unsigned int UpLo>
+inline const LDLT<typename SelfAdjointView<MatrixType, UpLo>::PlainObject, UpLo>
+SelfAdjointView<MatrixType, UpLo>::ldlt() const
+{
+  return LDLT<PlainObject,UpLo>(m_matrix);
+}
+
+/** \cholesky_module
+  * \returns the Cholesky decomposition with full pivoting without square root of \c *this
+  * \sa SelfAdjointView::ldlt()
+  */
+template<typename Derived>
+inline const LDLT<typename MatrixBase<Derived>::PlainObject>
+MatrixBase<Derived>::ldlt() const
+{
+  return LDLT<PlainObject>(derived());
+}
+
+} // end namespace Eigen
+
 #endif // EIGEN_LDLT_H

Eigen/src/Cholesky/LLT.h

@@ -1,37 +1,30 @@
 // This file is part of Eigen, a lightweight C++ template library
 // for linear algebra.
 //
-// Copyright (C) 2008 Gael Guennebaud <g.gael@free.fr>
+// 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
-// 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/>.
+// This Source Code Form is subject to the terms of the Mozilla
+// Public License v. 2.0. If a copy of the MPL was not distributed
+// with this file, You can obtain one at http://mozilla.org/MPL/2.0/.
 
 #ifndef EIGEN_LLT_H
 #define EIGEN_LLT_H
 
-/** \ingroup cholesky_Module
+namespace Eigen {
+
+namespace internal{
+template<typename MatrixType, int UpLo> struct LLT_Traits;
+}
+
+/** \ingroup Cholesky_Module
   *
   * \class LLT
   *
   * \brief Standard Cholesky decomposition (LL^T) of a matrix and associated features
   *
-  * \param MatrixType the type of the matrix of which we are computing the LL^T Cholesky decomposition
+  * \tparam _MatrixType the type of the matrix of which we are computing the LL^T Cholesky decomposition
+  * \tparam _UpLo the triangular part that will be used for the decompositon: Lower (default) or Upper.
+  *               The other triangular part won't be read.
   *
   * This class performs a LL^T Cholesky decomposition of a symmetric, positive definite
   * matrix A such that A = LL^* = U^*U, where L is lower triangular.
@@ -45,175 +38,505 @@
   * use LDLT instead for the semidefinite case. Also, do not use a Cholesky decomposition to determine whether a system of equations
   * has a solution.
   *
-  * \sa MatrixBase::llt(), class LDLT
+  * Example: \include LLT_example.cpp
+  * Output: \verbinclude LLT_example.out
+  *
+  * \b Performance: for best performance, it is recommended to use a column-major storage format
+  * with the Lower triangular part (the default), or, equivalently, a row-major storage format
+  * with the Upper triangular part. Otherwise, you might get a 20% slowdown for the full factorization
+  * step, and rank-updates can be up to 3 times slower.
+  *
+  * This class supports the \link InplaceDecomposition inplace decomposition \endlink mechanism.
+  *
+  * Note that during the decomposition, only the lower (or upper, as defined by _UpLo) triangular part of A is considered.
+  * Therefore, the strict lower part does not have to store correct values.
+  *
+  * \sa MatrixBase::llt(), SelfAdjointView::llt(), class LDLT
   */
- /* HEY THIS DOX IS DISABLED BECAUSE THERE's A BUG EITHER HERE OR IN LDLT ABOUT THAT (OR BOTH)
-  * Note that during the decomposition, only the upper triangular part of A is considered. Therefore,
-  * the strict lower part does not have to store correct values.
-  */
-template<typename MatrixType> class LLT
+template<typename _MatrixType, int _UpLo> class LLT
 {
-  private:
+  public:
+    typedef _MatrixType MatrixType;
+    enum {
+      RowsAtCompileTime = MatrixType::RowsAtCompileTime,
+      ColsAtCompileTime = MatrixType::ColsAtCompileTime,
+      MaxColsAtCompileTime = MatrixType::MaxColsAtCompileTime
+    };
     typedef typename MatrixType::Scalar Scalar;
     typedef typename NumTraits<typename MatrixType::Scalar>::Real RealScalar;
-    typedef Matrix<Scalar, MatrixType::ColsAtCompileTime, 1> VectorType;
+    typedef Eigen::Index Index; ///< \deprecated since Eigen 3.3
+    typedef typename MatrixType::StorageIndex StorageIndex;
 
     enum {
-      PacketSize = ei_packet_traits<Scalar>::size,
-      AlignmentMask = int(PacketSize)-1
+      PacketSize = internal::packet_traits<Scalar>::size,
+      AlignmentMask = int(PacketSize)-1,
+      UpLo = _UpLo
     };
 
-  public:
+    typedef internal::LLT_Traits<MatrixType,UpLo> Traits;
 
-    /** 
-    * \brief Default Constructor.
-    *
-    * The default constructor is useful in cases in which the user intends to
-    * perform decompositions via LLT::compute(const MatrixType&).
-    */
+    /**
+      * \brief Default Constructor.
+      *
+      * The default constructor is useful in cases in which the user intends to
+      * perform decompositions via LLT::compute(const MatrixType&).
+      */
     LLT() : m_matrix(), m_isInitialized(false) {}
 
-    LLT(const MatrixType& matrix)
+    /** \brief Default Constructor with memory preallocation
+      *
+      * Like the default constructor but with preallocation of the internal data
+      * according to the specified problem \a size.
+      * \sa LLT()
+      */
+    explicit LLT(Index size) : m_matrix(size, size),
+                    m_isInitialized(false) {}
+
+    template<typename InputType>
+    explicit LLT(const EigenBase<InputType>& matrix)
       : m_matrix(matrix.rows(), matrix.cols()),
         m_isInitialized(false)
     {
-      compute(matrix);
+      compute(matrix.derived());
     }
 
-    /** \returns the lower triangular matrix L */
-    inline Part<MatrixType, LowerTriangular> matrixL(void) const 
-    { 
-      ei_assert(m_isInitialized && "LLT is not initialized.");
-      return m_matrix; 
+    /** \brief Constructs a LDLT factorization from a given matrix
+      *
+      * This overloaded constructor is provided for \link InplaceDecomposition inplace decomposition \endlink when
+      * \c MatrixType is a Eigen::Ref.
+      *
+      * \sa LLT(const EigenBase&)
+      */
+    template<typename InputType>
+    explicit LLT(EigenBase<InputType>& matrix)
+      : m_matrix(matrix.derived()),
+        m_isInitialized(false)
+    {
+      compute(matrix.derived());
     }
-    
-    /** \deprecated */
-    inline bool isPositiveDefinite(void) const { return m_isInitialized && m_isPositiveDefinite; }
 
-    template<typename RhsDerived, typename ResultType>
-    bool solve(const MatrixBase<RhsDerived> &b, ResultType *result) const;
+    /** \returns a view of the upper triangular matrix U */
+    inline typename Traits::MatrixU matrixU() const
+    {
+      eigen_assert(m_isInitialized && "LLT is not initialized.");
+      return Traits::getU(m_matrix);
+    }
+
+    /** \returns a view of the lower triangular matrix L */
+    inline typename Traits::MatrixL matrixL() const
+    {
+      eigen_assert(m_isInitialized && "LLT is not initialized.");
+      return Traits::getL(m_matrix);
+    }
+
+    /** \returns the solution x of \f$ A x = b \f$ using the current decomposition of A.
+      *
+      * Since this LLT class assumes anyway that the matrix A is invertible, the solution
+      * theoretically exists and is unique regardless of b.
+      *
+      * Example: \include LLT_solve.cpp
+      * Output: \verbinclude LLT_solve.out
+      *
+      * \sa solveInPlace(), MatrixBase::llt(), SelfAdjointView::llt()
+      */
+    template<typename Rhs>
+    inline const Solve<LLT, Rhs>
+    solve(const MatrixBase<Rhs>& b) const
+    {
+      eigen_assert(m_isInitialized && "LLT is not initialized.");
+      eigen_assert(m_matrix.rows()==b.rows()
+                && "LLT::solve(): invalid number of rows of the right hand side matrix b");
+      return Solve<LLT, Rhs>(*this, b.derived());
+    }
 
     template<typename Derived>
-    bool solveInPlace(MatrixBase<Derived> &bAndX) const;
+    void solveInPlace(const MatrixBase<Derived> &bAndX) const;
 
-    void compute(const MatrixType& matrix);
+    template<typename InputType>
+    LLT& compute(const EigenBase<InputType>& matrix);
+
+    /** \returns an estimate of the reciprocal condition number of the matrix of
+      *  which \c *this is the Cholesky decomposition.
+      */
+    RealScalar rcond() const
+    {
+      eigen_assert(m_isInitialized && "LLT is not initialized.");
+      eigen_assert(m_info == Success && "LLT failed because matrix appears to be negative");
+      return internal::rcond_estimate_helper(m_l1_norm, *this);
+    }
+
+    /** \returns the LLT decomposition matrix
+      *
+      * TODO: document the storage layout
+      */
+    inline const MatrixType& matrixLLT() const
+    {
+      eigen_assert(m_isInitialized && "LLT is not initialized.");
+      return m_matrix;
+    }
+
+    MatrixType reconstructedMatrix() const;
+
+
+    /** \brief Reports whether previous computation was successful.
+      *
+      * \returns \c Success if computation was succesful,
+      *          \c NumericalIssue if the matrix.appears not to be positive definite.
+      */
+    ComputationInfo info() const
+    {
+      eigen_assert(m_isInitialized && "LLT is not initialized.");
+      return m_info;
+    }
+
+    /** \returns the adjoint of \c *this, that is, a const reference to the decomposition itself as the underlying matrix is self-adjoint.
+      *
+      * This method is provided for compatibility with other matrix decompositions, thus enabling generic code such as:
+      * \code x = decomposition.adjoint().solve(b) \endcode
+      */
+    const LLT& adjoint() const { return *this; };
+
+    inline Index rows() const { return m_matrix.rows(); }
+    inline Index cols() const { return m_matrix.cols(); }
+
+    template<typename VectorType>
+    LLT rankUpdate(const VectorType& vec, const RealScalar& sigma = 1);
+
+    #ifndef EIGEN_PARSED_BY_DOXYGEN
+    template<typename RhsType, typename DstType>
+    EIGEN_DEVICE_FUNC
+    void _solve_impl(const RhsType &rhs, DstType &dst) const;
+    #endif
 
   protected:
+
+    static void check_template_parameters()
+    {
+      EIGEN_STATIC_ASSERT_NON_INTEGER(Scalar);
+    }
+
     /** \internal
       * Used to compute and store L
       * The strict upper part is not used and even not initialized.
       */
     MatrixType m_matrix;
+    RealScalar m_l1_norm;
     bool m_isInitialized;
-    bool m_isPositiveDefinite;
+    ComputationInfo m_info;
 };
 
-/** Computes / recomputes the Cholesky decomposition A = LL^* = U^*U of \a matrix
-  */
-template<typename MatrixType>
-void LLT<MatrixType>::compute(const MatrixType& a)
+namespace internal {
+
+template<typename Scalar, int UpLo> struct llt_inplace;
+
+template<typename MatrixType, typename VectorType>
+static Index llt_rank_update_lower(MatrixType& mat, const VectorType& vec, const typename MatrixType::RealScalar& sigma)
 {
-  assert(a.rows()==a.cols());
-  m_isPositiveDefinite = true;
-  const int size = a.rows();
-  m_matrix.resize(size, size);
-  // The biggest overall is the point of reference to which further diagonals
-  // are compared; if any diagonal is negligible compared
-  // to the largest overall, the algorithm bails.  This cutoff is suggested
-  // in "Analysis of the Cholesky Decomposition of a Semi-definite Matrix" by
-  // Nicholas J. Higham. Also see "Accuracy and Stability of Numerical
-  // Algorithms" page 217, also by Higham.
-  const RealScalar cutoff = machine_epsilon<Scalar>() * size * a.diagonal().cwise().abs().maxCoeff();
-  RealScalar x;
-  x = ei_real(a.coeff(0,0));
-  m_matrix.coeffRef(0,0) = ei_sqrt(x);
-  if(size==1)
+  using std::sqrt;
+  typedef typename MatrixType::Scalar Scalar;
+  typedef typename MatrixType::RealScalar RealScalar;
+  typedef typename MatrixType::ColXpr ColXpr;
+  typedef typename internal::remove_all<ColXpr>::type ColXprCleaned;
+  typedef typename ColXprCleaned::SegmentReturnType ColXprSegment;
+  typedef Matrix<Scalar,Dynamic,1> TempVectorType;
+  typedef typename TempVectorType::SegmentReturnType TempVecSegment;
+
+  Index n = mat.cols();
+  eigen_assert(mat.rows()==n && vec.size()==n);
+
+  TempVectorType temp;
+
+  if(sigma>0)
   {
-    m_isInitialized = true;
-    return;
-  }
-  m_matrix.col(0).end(size-1) = a.row(0).end(size-1).adjoint() / ei_real(m_matrix.coeff(0,0));
-  for (int j = 1; j < size; ++j)
-  {
-    x = ei_real(a.coeff(j,j)) - m_matrix.row(j).start(j).squaredNorm();
-    if (x < cutoff)
+    // This version is based on Givens rotations.
+    // It is faster than the other one below, but only works for updates,
+    // i.e., for sigma > 0
+    temp = sqrt(sigma) * vec;
+
+    for(Index i=0; i<n; ++i)
     {
-      m_isPositiveDefinite = false;
-      continue;
+      JacobiRotation<Scalar> g;
+      g.makeGivens(mat(i,i), -temp(i), &mat(i,i));
+
+      Index 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);
+      }
     }
+  }
+  else
+  {
+    temp = vec;
+    RealScalar beta = 1;
+    for(Index j=0; j<n; ++j)
+    {
+      RealScalar Ljj = numext::real(mat.coeff(j,j));
+      RealScalar dj = numext::abs2(Ljj);
+      Scalar wj = temp.coeff(j);
+      RealScalar swj2 = sigma*numext::abs2(wj);
+      RealScalar gamma = dj*beta + swj2;
 
-    m_matrix.coeffRef(j,j) = x = ei_sqrt(x);
+      RealScalar x = dj + swj2/beta;
+      if (x<=RealScalar(0))
+        return j;
+      RealScalar nLjj = sqrt(x);
+      mat.coeffRef(j,j) = nLjj;
+      beta += swj2/dj;
 
-    int endSize = size-j-1;
-    if (endSize>0) {
-      // Note that when all matrix columns have good alignment, then the following
-      // product is guaranteed to be optimal with respect to alignment.
-      m_matrix.col(j).end(endSize) =
-        (m_matrix.block(j+1, 0, endSize, j) * m_matrix.row(j).start(j).adjoint()).lazy();
-
-      // FIXME could use a.col instead of a.row
-      m_matrix.col(j).end(endSize) = (a.row(j).end(endSize).adjoint()
-        - m_matrix.col(j).end(endSize) ) / x;
+      // Update the terms of L
+      Index rs = n-j-1;
+      if(rs)
+      {
+        temp.tail(rs) -= (wj/Ljj) * mat.col(j).tail(rs);
+        if(gamma != 0)
+          mat.col(j).tail(rs) = (nLjj/Ljj) * mat.col(j).tail(rs) + (nLjj * sigma*numext::conj(wj)/gamma)*temp.tail(rs);
+      }
     }
   }
+  return -1;
+}
+
+template<typename Scalar> struct llt_inplace<Scalar, Lower>
+{
+  typedef typename NumTraits<Scalar>::Real RealScalar;
+  template<typename MatrixType>
+  static Index unblocked(MatrixType& mat)
+  {
+    using std::sqrt;
+
+    eigen_assert(mat.rows()==mat.cols());
+    const Index size = mat.rows();
+    for(Index k = 0; k < size; ++k)
+    {
+      Index rs = size-k-1; // remaining size
+
+      Block<MatrixType,Dynamic,1> A21(mat,k+1,k,rs,1);
+      Block<MatrixType,1,Dynamic> A10(mat,k,0,1,k);
+      Block<MatrixType,Dynamic,Dynamic> A20(mat,k+1,0,rs,k);
+
+      RealScalar x = numext::real(mat.coeff(k,k));
+      if (k>0) x -= A10.squaredNorm();
+      if (x<=RealScalar(0))
+        return k;
+      mat.coeffRef(k,k) = x = sqrt(x);
+      if (k>0 && rs>0) A21.noalias() -= A20 * A10.adjoint();
+      if (rs>0) A21 /= x;
+    }
+    return -1;
+  }
+
+  template<typename MatrixType>
+  static Index blocked(MatrixType& m)
+  {
+    eigen_assert(m.rows()==m.cols());
+    Index size = m.rows();
+    if(size<32)
+      return unblocked(m);
+
+    Index blockSize = size/8;
+    blockSize = (blockSize/16)*16;
+    blockSize = (std::min)((std::max)(blockSize,Index(8)), Index(128));
+
+    for (Index k=0; k<size; k+=blockSize)
+    {
+      // partition the matrix:
+      //       A00 |  -  |  -
+      // lu  = A10 | A11 |  -
+      //       A20 | A21 | A22
+      Index bs = (std::min)(blockSize, size-k);
+      Index rs = size - k - bs;
+      Block<MatrixType,Dynamic,Dynamic> A11(m,k,   k,   bs,bs);
+      Block<MatrixType,Dynamic,Dynamic> A21(m,k+bs,k,   rs,bs);
+      Block<MatrixType,Dynamic,Dynamic> A22(m,k+bs,k+bs,rs,rs);
+
+      Index ret;
+      if((ret=unblocked(A11))>=0) return k+ret;
+      if(rs>0) A11.adjoint().template triangularView<Upper>().template solveInPlace<OnTheRight>(A21);
+      if(rs>0) A22.template selfadjointView<Lower>().rankUpdate(A21,typename NumTraits<RealScalar>::Literal(-1)); // bottleneck
+    }
+    return -1;
+  }
+
+  template<typename MatrixType, typename VectorType>
+  static Index rankUpdate(MatrixType& mat, const VectorType& vec, const RealScalar& sigma)
+  {
+    return Eigen::internal::llt_rank_update_lower(mat, vec, sigma);
+  }
+};
+
+template<typename Scalar> struct llt_inplace<Scalar, Upper>
+{
+  typedef typename NumTraits<Scalar>::Real RealScalar;
+
+  template<typename MatrixType>
+  static EIGEN_STRONG_INLINE Index unblocked(MatrixType& mat)
+  {
+    Transpose<MatrixType> matt(mat);
+    return llt_inplace<Scalar, Lower>::unblocked(matt);
+  }
+  template<typename MatrixType>
+  static EIGEN_STRONG_INLINE Index blocked(MatrixType& mat)
+  {
+    Transpose<MatrixType> matt(mat);
+    return llt_inplace<Scalar, Lower>::blocked(matt);
+  }
+  template<typename MatrixType, typename VectorType>
+  static Index rankUpdate(MatrixType& mat, const VectorType& vec, const RealScalar& sigma)
+  {
+    Transpose<MatrixType> matt(mat);
+    return llt_inplace<Scalar, Lower>::rankUpdate(matt, vec.conjugate(), sigma);
+  }
+};
+
+template<typename MatrixType> struct LLT_Traits<MatrixType,Lower>
+{
+  typedef const TriangularView<const MatrixType, Lower> MatrixL;
+  typedef const TriangularView<const typename MatrixType::AdjointReturnType, Upper> MatrixU;
+  static inline MatrixL getL(const MatrixType& m) { return MatrixL(m); }
+  static inline MatrixU getU(const MatrixType& m) { return MatrixU(m.adjoint()); }
+  static bool inplace_decomposition(MatrixType& m)
+  { return llt_inplace<typename MatrixType::Scalar, Lower>::blocked(m)==-1; }
+};
+
+template<typename MatrixType> struct LLT_Traits<MatrixType,Upper>
+{
+  typedef const TriangularView<const typename MatrixType::AdjointReturnType, Lower> MatrixL;
+  typedef const TriangularView<const MatrixType, Upper> MatrixU;
+  static inline MatrixL getL(const MatrixType& m) { return MatrixL(m.adjoint()); }
+  static inline MatrixU getU(const MatrixType& m) { return MatrixU(m); }
+  static bool inplace_decomposition(MatrixType& m)
+  { return llt_inplace<typename MatrixType::Scalar, Upper>::blocked(m)==-1; }
+};
+
+} // end namespace internal
+
+/** 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
+  */
+template<typename MatrixType, int _UpLo>
+template<typename InputType>
+LLT<MatrixType,_UpLo>& LLT<MatrixType,_UpLo>::compute(const EigenBase<InputType>& a)
+{
+  check_template_parameters();
+
+  eigen_assert(a.rows()==a.cols());
+  const Index size = a.rows();
+  m_matrix.resize(size, size);
+  if (!internal::is_same_dense(m_matrix, a.derived()))
+    m_matrix = a.derived();
+
+  // Compute matrix L1 norm = max abs column sum.
+  m_l1_norm = RealScalar(0);
+  // TODO move this code to SelfAdjointView
+  for (Index col = 0; col < size; ++col) {
+    RealScalar abs_col_sum;
+    if (_UpLo == Lower)
+      abs_col_sum = m_matrix.col(col).tail(size - col).template lpNorm<1>() + m_matrix.row(col).head(col).template lpNorm<1>();
+    else
+      abs_col_sum = m_matrix.col(col).head(col).template lpNorm<1>() + m_matrix.row(col).tail(size - col).template lpNorm<1>();
+    if (abs_col_sum > m_l1_norm)
+      m_l1_norm = abs_col_sum;
+  }
 
   m_isInitialized = true;
+  bool ok = Traits::inplace_decomposition(m_matrix);
+  m_info = ok ? Success : NumericalIssue;
+
+  return *this;
 }
 
-/** Computes the solution x of \f$ A x = b \f$ using the current decomposition of A.
-  * The result is stored in \a result
-  *
-  * \returns true always! If you need to check for existence of solutions, use another decomposition like LU, QR, or SVD.
-  *
-  * In other words, it computes \f$ b = A^{-1} b \f$ with
-  * \f$ {L^{*}}^{-1} L^{-1} b \f$ from right to left.
-  *
-  * Example: \include LLT_solve.cpp
-  * Output: \verbinclude LLT_solve.out
-  *
-  * \sa LLT::solveInPlace(), MatrixBase::llt()
+/** Performs a rank one update (or dowdate) of the current decomposition.
+  * If A = LL^* before the rank one update,
+  * then after it we have LL^* = A + sigma * v v^* where \a v must be a vector
+  * of same dimension.
   */
-template<typename MatrixType>
-template<typename RhsDerived, typename ResultType>
-bool LLT<MatrixType>::solve(const MatrixBase<RhsDerived> &b, ResultType *result) const
+template<typename _MatrixType, int _UpLo>
+template<typename VectorType>
+LLT<_MatrixType,_UpLo> LLT<_MatrixType,_UpLo>::rankUpdate(const VectorType& v, const RealScalar& sigma)
 {
-  ei_assert(m_isInitialized && "LLT is not initialized.");
-  const int size = m_matrix.rows();
-  ei_assert(size==b.rows() && "LLT::solve(): invalid number of rows of the right hand side matrix b");
-  return solveInPlace((*result) = b);
+  EIGEN_STATIC_ASSERT_VECTOR_ONLY(VectorType);
+  eigen_assert(v.size()==m_matrix.cols());
+  eigen_assert(m_isInitialized);
+  if(internal::llt_inplace<typename MatrixType::Scalar, UpLo>::rankUpdate(m_matrix,v,sigma)>=0)
+    m_info = NumericalIssue;
+  else
+    m_info = Success;
+
+  return *this;
 }
 
-/** This is the \em in-place version of solve().
+#ifndef EIGEN_PARSED_BY_DOXYGEN
+template<typename _MatrixType,int _UpLo>
+template<typename RhsType, typename DstType>
+void LLT<_MatrixType,_UpLo>::_solve_impl(const RhsType &rhs, DstType &dst) const
+{
+  dst = rhs;
+  solveInPlace(dst);
+}
+#endif
+
+/** \internal use x = llt_object.solve(x);
+  *
+  * This is the \em in-place version of solve().
   *
   * \param bAndX represents both the right-hand side matrix b and result x.
   *
-  * \returns true always! If you need to check for existence of solutions, use another decomposition like LU, QR, or SVD.
+  * This version avoids a copy when the right hand side matrix b is not needed anymore.
   *
-  * This version avoids a copy when the right hand side matrix b is not
-  * needed anymore.
+  * \warning The parameter is only marked 'const' to make the C++ compiler accept a temporary expression here.
+  * This function will const_cast it, so constness isn't honored here.
   *
   * \sa LLT::solve(), MatrixBase::llt()
   */
-template<typename MatrixType>
+template<typename MatrixType, int _UpLo>
 template<typename Derived>
-bool LLT<MatrixType>::solveInPlace(MatrixBase<Derived> &bAndX) const
+void LLT<MatrixType,_UpLo>::solveInPlace(const MatrixBase<Derived> &bAndX) const
 {
-  ei_assert(m_isInitialized && "LLT is not initialized.");
-  const int size = m_matrix.rows();
-  ei_assert(size==bAndX.rows());
-  matrixL().solveTriangularInPlace(bAndX);
-  m_matrix.adjoint().template part<UpperTriangular>().solveTriangularInPlace(bAndX);
-  return true;
+  eigen_assert(m_isInitialized && "LLT is not initialized.");
+  eigen_assert(m_matrix.rows()==bAndX.rows());
+  matrixL().solveInPlace(bAndX);
+  matrixU().solveInPlace(bAndX);
+}
+
+/** \returns the matrix represented by the decomposition,
+ * i.e., it returns the product: L L^*.
+ * This function is provided for debug purpose. */
+template<typename MatrixType, int _UpLo>
+MatrixType LLT<MatrixType,_UpLo>::reconstructedMatrix() const
+{
+  eigen_assert(m_isInitialized && "LLT is not initialized.");
+  return matrixL() * matrixL().adjoint().toDenseMatrix();
 }
 
 /** \cholesky_module
   * \returns the LLT decomposition of \c *this
+  * \sa SelfAdjointView::llt()
   */
 template<typename Derived>
-inline const LLT<typename MatrixBase<Derived>::PlainMatrixType>
+inline const LLT<typename MatrixBase<Derived>::PlainObject>
 MatrixBase<Derived>::llt() const
 {
-  return LLT<PlainMatrixType>(derived());
+  return LLT<PlainObject>(derived());
 }
 
+/** \cholesky_module
+  * \returns the LLT decomposition of \c *this
+  * \sa SelfAdjointView::llt()
+  */
+template<typename MatrixType, unsigned int UpLo>
+inline const LLT<typename SelfAdjointView<MatrixType, UpLo>::PlainObject, UpLo>
+SelfAdjointView<MatrixType, UpLo>::llt() const
+{
+  return LLT<PlainObject,UpLo>(m_matrix);
+}
+
+} // end namespace Eigen
+
 #endif // EIGEN_LLT_H

Eigen/src/Cholesky/LLT_LAPACKE.h

@@ -0,0 +1,99 @@
+/*
+ Copyright (c) 2011, Intel Corporation. All rights reserved.
+
+ Redistribution and use in source and binary forms, with or without modification,
+ are permitted provided that the following conditions are met:
+
+ * Redistributions of source code must retain the above copyright notice, this
+   list of conditions and the following disclaimer.
+ * Redistributions in binary form must reproduce the above copyright notice,
+   this list of conditions and the following disclaimer in the documentation
+   and/or other materials provided with the distribution.
+ * Neither the name of Intel Corporation nor the names of its contributors may
+   be used to endorse or promote products derived from this software without
+   specific prior written permission.
+
+ THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" AND
+ ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED
+ WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE
+ DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE LIABLE FOR
+ ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES
+ (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES;
+ LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON
+ ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+ (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS
+ SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+
+ ********************************************************************************
+ *   Content : Eigen bindings to LAPACKe
+ *     LLt decomposition based on LAPACKE_?potrf function.
+ ********************************************************************************
+*/
+
+#ifndef EIGEN_LLT_LAPACKE_H
+#define EIGEN_LLT_LAPACKE_H
+
+namespace Eigen { 
+
+namespace internal {
+
+template<typename Scalar> struct lapacke_llt;
+
+#define EIGEN_LAPACKE_LLT(EIGTYPE, BLASTYPE, LAPACKE_PREFIX) \
+template<> struct lapacke_llt<EIGTYPE> \
+{ \
+  template<typename MatrixType> \
+  static inline Index potrf(MatrixType& m, char uplo) \
+  { \
+    lapack_int matrix_order; \
+    lapack_int size, lda, info, StorageOrder; \
+    EIGTYPE* a; \
+    eigen_assert(m.rows()==m.cols()); \
+    /* Set up parameters for ?potrf */ \
+    size = convert_index<lapack_int>(m.rows()); \
+    StorageOrder = MatrixType::Flags&RowMajorBit?RowMajor:ColMajor; \
+    matrix_order = StorageOrder==RowMajor ? LAPACK_ROW_MAJOR : LAPACK_COL_MAJOR; \
+    a = &(m.coeffRef(0,0)); \
+    lda = convert_index<lapack_int>(m.outerStride()); \
+\
+    info = LAPACKE_##LAPACKE_PREFIX##potrf( matrix_order, uplo, size, (BLASTYPE*)a, lda ); \
+    info = (info==0) ? -1 : info>0 ? info-1 : size; \
+    return info; \
+  } \
+}; \
+template<> struct llt_inplace<EIGTYPE, Lower> \
+{ \
+  template<typename MatrixType> \
+  static Index blocked(MatrixType& m) \
+  { \
+    return lapacke_llt<EIGTYPE>::potrf(m, 'L'); \
+  } \
+  template<typename MatrixType, typename VectorType> \
+  static Index rankUpdate(MatrixType& mat, const VectorType& vec, const typename MatrixType::RealScalar& sigma) \
+  { return Eigen::internal::llt_rank_update_lower(mat, vec, sigma); } \
+}; \
+template<> struct llt_inplace<EIGTYPE, Upper> \
+{ \
+  template<typename MatrixType> \
+  static Index blocked(MatrixType& m) \
+  { \
+    return lapacke_llt<EIGTYPE>::potrf(m, 'U'); \
+  } \
+  template<typename MatrixType, typename VectorType> \
+  static Index rankUpdate(MatrixType& mat, const VectorType& vec, const typename MatrixType::RealScalar& sigma) \
+  { \
+    Transpose<MatrixType> matt(mat); \
+    return llt_inplace<EIGTYPE, Lower>::rankUpdate(matt, vec.conjugate(), sigma); \
+  } \
+};
+
+EIGEN_LAPACKE_LLT(double, double, d)
+EIGEN_LAPACKE_LLT(float, float, s)
+EIGEN_LAPACKE_LLT(dcomplex, lapack_complex_double, z)
+EIGEN_LAPACKE_LLT(scomplex, lapack_complex_float, c)
+
+} // end namespace internal
+
+} // end namespace Eigen
+
+#endif // EIGEN_LLT_LAPACKE_H

Eigen/src/CholmodSupport/CholmodSupport.h

@@ -0,0 +1,639 @@
+// This file is part of Eigen, a lightweight C++ template library
+// for linear algebra.
+//
+// Copyright (C) 2008-2010 Gael Guennebaud <gael.guennebaud@inria.fr>
+//
+// This Source Code Form is subject to the terms of the Mozilla
+// Public License v. 2.0. If a copy of the MPL was not distributed
+// with this file, You can obtain one at http://mozilla.org/MPL/2.0/.
+
+#ifndef EIGEN_CHOLMODSUPPORT_H
+#define EIGEN_CHOLMODSUPPORT_H
+
+namespace Eigen { 
+
+namespace internal {
+
+template<typename Scalar> struct cholmod_configure_matrix;
+
+template<> struct cholmod_configure_matrix<double> {
+  template<typename CholmodType>
+  static void run(CholmodType& mat) {
+    mat.xtype = CHOLMOD_REAL;
+    mat.dtype = CHOLMOD_DOUBLE;
+  }
+};
+
+template<> struct cholmod_configure_matrix<std::complex<double> > {
+  template<typename CholmodType>
+  static void run(CholmodType& mat) {
+    mat.xtype = CHOLMOD_COMPLEX;
+    mat.dtype = CHOLMOD_DOUBLE;
+  }
+};
+
+// Other scalar types are not yet suppotred by Cholmod
+// template<> struct cholmod_configure_matrix<float> {
+//   template<typename CholmodType>
+//   static void run(CholmodType& mat) {
+//     mat.xtype = CHOLMOD_REAL;
+//     mat.dtype = CHOLMOD_SINGLE;
+//   }
+// };
+//
+// template<> struct cholmod_configure_matrix<std::complex<float> > {
+//   template<typename CholmodType>
+//   static void run(CholmodType& mat) {
+//     mat.xtype = CHOLMOD_COMPLEX;
+//     mat.dtype = CHOLMOD_SINGLE;
+//   }
+// };
+
+} // namespace internal
+
+/** Wraps the Eigen sparse matrix \a mat into a Cholmod sparse matrix object.
+  * Note that the data are shared.
+  */
+template<typename _Scalar, int _Options, typename _StorageIndex>
+cholmod_sparse viewAsCholmod(Ref<SparseMatrix<_Scalar,_Options,_StorageIndex> > mat)
+{
+  cholmod_sparse res;
+  res.nzmax   = mat.nonZeros();
+  res.nrow    = mat.rows();
+  res.ncol    = mat.cols();
+  res.p       = mat.outerIndexPtr();
+  res.i       = mat.innerIndexPtr();
+  res.x       = mat.valuePtr();
+  res.z       = 0;
+  res.sorted  = 1;
+  if(mat.isCompressed())
+  {
+    res.packed  = 1;
+    res.nz = 0;
+  }
+  else
+  {
+    res.packed  = 0;
+    res.nz = mat.innerNonZeroPtr();
+  }
+
+  res.dtype   = 0;
+  res.stype   = -1;
+  
+  if (internal::is_same<_StorageIndex,int>::value)
+  {
+    res.itype = CHOLMOD_INT;
+  }
+  else if (internal::is_same<_StorageIndex,long>::value)
+  {
+    res.itype = CHOLMOD_LONG;
+  }
+  else
+  {
+    eigen_assert(false && "Index type not supported yet");
+  }
+
+  // setup res.xtype
+  internal::cholmod_configure_matrix<_Scalar>::run(res);
+  
+  res.stype = 0;
+  
+  return res;
+}
+
+template<typename _Scalar, int _Options, typename _Index>
+const cholmod_sparse viewAsCholmod(const SparseMatrix<_Scalar,_Options,_Index>& mat)
+{
+  cholmod_sparse res = viewAsCholmod(Ref<SparseMatrix<_Scalar,_Options,_Index> >(mat.const_cast_derived()));
+  return res;
+}
+
+template<typename _Scalar, int _Options, typename _Index>
+const cholmod_sparse viewAsCholmod(const SparseVector<_Scalar,_Options,_Index>& mat)
+{
+  cholmod_sparse res = viewAsCholmod(Ref<SparseMatrix<_Scalar,_Options,_Index> >(mat.const_cast_derived()));
+  return res;
+}
+
+/** Returns a view of the Eigen sparse matrix \a mat as Cholmod sparse matrix.
+  * The data are not copied but shared. */
+template<typename _Scalar, int _Options, typename _Index, unsigned int UpLo>
+cholmod_sparse viewAsCholmod(const SparseSelfAdjointView<const SparseMatrix<_Scalar,_Options,_Index>, UpLo>& mat)
+{
+  cholmod_sparse res = viewAsCholmod(Ref<SparseMatrix<_Scalar,_Options,_Index> >(mat.matrix().const_cast_derived()));
+  
+  if(UpLo==Upper) res.stype =  1;
+  if(UpLo==Lower) res.stype = -1;
+
+  return res;
+}
+
+/** Returns a view of the Eigen \b dense matrix \a mat as Cholmod dense matrix.
+  * The data are not copied but shared. */
+template<typename Derived>
+cholmod_dense viewAsCholmod(MatrixBase<Derived>& mat)
+{
+  EIGEN_STATIC_ASSERT((internal::traits<Derived>::Flags&RowMajorBit)==0,THIS_METHOD_IS_ONLY_FOR_COLUMN_MAJOR_MATRICES);
+  typedef typename Derived::Scalar Scalar;
+
+  cholmod_dense res;
+  res.nrow   = mat.rows();
+  res.ncol   = mat.cols();
+  res.nzmax  = res.nrow * res.ncol;
+  res.d      = Derived::IsVectorAtCompileTime ? mat.derived().size() : mat.derived().outerStride();
+  res.x      = (void*)(mat.derived().data());
+  res.z      = 0;
+
+  internal::cholmod_configure_matrix<Scalar>::run(res);
+
+  return res;
+}
+
+/** Returns a view of the Cholmod sparse matrix \a cm as an Eigen sparse matrix.
+  * The data are not copied but shared. */
+template<typename Scalar, int Flags, typename StorageIndex>
+MappedSparseMatrix<Scalar,Flags,StorageIndex> viewAsEigen(cholmod_sparse& cm)
+{
+  return MappedSparseMatrix<Scalar,Flags,StorageIndex>
+         (cm.nrow, cm.ncol, static_cast<StorageIndex*>(cm.p)[cm.ncol],
+          static_cast<StorageIndex*>(cm.p), static_cast<StorageIndex*>(cm.i),static_cast<Scalar*>(cm.x) );
+}
+
+enum CholmodMode {
+  CholmodAuto, CholmodSimplicialLLt, CholmodSupernodalLLt, CholmodLDLt
+};
+
+
+/** \ingroup CholmodSupport_Module
+  * \class CholmodBase
+  * \brief The base class for the direct Cholesky factorization of Cholmod
+  * \sa class CholmodSupernodalLLT, class CholmodSimplicialLDLT, class CholmodSimplicialLLT
+  */
+template<typename _MatrixType, int _UpLo, typename Derived>
+class CholmodBase : public SparseSolverBase<Derived>
+{
+  protected:
+    typedef SparseSolverBase<Derived> Base;
+    using Base::derived;
+    using Base::m_isInitialized;
+  public:
+    typedef _MatrixType MatrixType;
+    enum { UpLo = _UpLo };
+    typedef typename MatrixType::Scalar Scalar;
+    typedef typename MatrixType::RealScalar RealScalar;
+    typedef MatrixType CholMatrixType;
+    typedef typename MatrixType::StorageIndex StorageIndex;
+    enum {
+      ColsAtCompileTime = MatrixType::ColsAtCompileTime,
+      MaxColsAtCompileTime = MatrixType::MaxColsAtCompileTime
+    };
+
+  public:
+
+    CholmodBase()
+      : m_cholmodFactor(0), m_info(Success), m_factorizationIsOk(false), m_analysisIsOk(false)
+    {
+      EIGEN_STATIC_ASSERT((internal::is_same<double,RealScalar>::value), CHOLMOD_SUPPORTS_DOUBLE_PRECISION_ONLY);
+      m_shiftOffset[0] = m_shiftOffset[1] = 0.0;
+      cholmod_start(&m_cholmod);
+    }
+
+    explicit CholmodBase(const MatrixType& matrix)
+      : m_cholmodFactor(0), m_info(Success), m_factorizationIsOk(false), m_analysisIsOk(false)
+    {
+      EIGEN_STATIC_ASSERT((internal::is_same<double,RealScalar>::value), CHOLMOD_SUPPORTS_DOUBLE_PRECISION_ONLY);
+      m_shiftOffset[0] = m_shiftOffset[1] = 0.0;
+      cholmod_start(&m_cholmod);
+      compute(matrix);
+    }
+
+    ~CholmodBase()
+    {
+      if(m_cholmodFactor)
+        cholmod_free_factor(&m_cholmodFactor, &m_cholmod);
+      cholmod_finish(&m_cholmod);
+    }
+    
+    inline StorageIndex cols() const { return internal::convert_index<StorageIndex, Index>(m_cholmodFactor->n); }
+    inline StorageIndex rows() const { return internal::convert_index<StorageIndex, Index>(m_cholmodFactor->n); }
+    
+    /** \brief Reports whether previous computation was successful.
+      *
+      * \returns \c Success if computation was succesful,
+      *          \c NumericalIssue if the matrix.appears to be negative.
+      */
+    ComputationInfo info() const
+    {
+      eigen_assert(m_isInitialized && "Decomposition is not initialized.");
+      return m_info;
+    }
+
+    /** Computes the sparse Cholesky decomposition of \a matrix */
+    Derived& compute(const MatrixType& matrix)
+    {
+      analyzePattern(matrix);
+      factorize(matrix);
+      return derived();
+    }
+    
+    /** Performs a symbolic decomposition on the sparsity pattern of \a matrix.
+      *
+      * This function is particularly useful when solving for several problems having the same structure.
+      * 
+      * \sa factorize()
+      */
+    void analyzePattern(const MatrixType& matrix)
+    {
+      if(m_cholmodFactor)
+      {
+        cholmod_free_factor(&m_cholmodFactor, &m_cholmod);
+        m_cholmodFactor = 0;
+      }
+      cholmod_sparse A = viewAsCholmod(matrix.template selfadjointView<UpLo>());
+      m_cholmodFactor = cholmod_analyze(&A, &m_cholmod);
+      
+      this->m_isInitialized = true;
+      this->m_info = Success;
+      m_analysisIsOk = true;
+      m_factorizationIsOk = false;
+    }
+    
+    /** Performs a numeric decomposition of \a matrix
+      *
+      * The given matrix must have the same sparsity pattern as the matrix on which the symbolic decomposition has been performed.
+      *
+      * \sa analyzePattern()
+      */
+    void factorize(const MatrixType& matrix)
+    {
+      eigen_assert(m_analysisIsOk && "You must first call analyzePattern()");
+      cholmod_sparse A = viewAsCholmod(matrix.template selfadjointView<UpLo>());
+      cholmod_factorize_p(&A, m_shiftOffset, 0, 0, m_cholmodFactor, &m_cholmod);
+
+      // If the factorization failed, minor is the column at which it did. On success minor == n.
+      this->m_info = (m_cholmodFactor->minor == m_cholmodFactor->n ? Success : NumericalIssue);
+      m_factorizationIsOk = true;
+    }
+    
+    /** Returns a reference to the Cholmod's configuration structure to get a full control over the performed operations.
+     *  See the Cholmod user guide for details. */
+    cholmod_common& cholmod() { return m_cholmod; }
+    
+    #ifndef EIGEN_PARSED_BY_DOXYGEN
+    /** \internal */
+    template<typename Rhs,typename Dest>
+    void _solve_impl(const MatrixBase<Rhs> &b, MatrixBase<Dest> &dest) const
+    {
+      eigen_assert(m_factorizationIsOk && "The decomposition is not in a valid state for solving, you must first call either compute() or symbolic()/numeric()");
+      const Index size = m_cholmodFactor->n;
+      EIGEN_UNUSED_VARIABLE(size);
+      eigen_assert(size==b.rows());
+      
+      // Cholmod needs column-major stoarge without inner-stride, which corresponds to the default behavior of Ref.
+      Ref<const Matrix<typename Rhs::Scalar,Dynamic,Dynamic,ColMajor> > b_ref(b.derived());
+
+      cholmod_dense b_cd = viewAsCholmod(b_ref);
+      cholmod_dense* x_cd = cholmod_solve(CHOLMOD_A, m_cholmodFactor, &b_cd, &m_cholmod);
+      if(!x_cd)
+      {
+        this->m_info = NumericalIssue;
+        return;
+      }
+      // TODO optimize this copy by swapping when possible (be careful with alignment, etc.)
+      dest = Matrix<Scalar,Dest::RowsAtCompileTime,Dest::ColsAtCompileTime>::Map(reinterpret_cast<Scalar*>(x_cd->x),b.rows(),b.cols());
+      cholmod_free_dense(&x_cd, &m_cholmod);
+    }
+    
+    /** \internal */
+    template<typename RhsDerived, typename DestDerived>
+    void _solve_impl(const SparseMatrixBase<RhsDerived> &b, SparseMatrixBase<DestDerived> &dest) const
+    {
+      eigen_assert(m_factorizationIsOk && "The decomposition is not in a valid state for solving, you must first call either compute() or symbolic()/numeric()");
+      const Index size = m_cholmodFactor->n;
+      EIGEN_UNUSED_VARIABLE(size);
+      eigen_assert(size==b.rows());
+
+      // note: cs stands for Cholmod Sparse
+      Ref<SparseMatrix<typename RhsDerived::Scalar,ColMajor,typename RhsDerived::StorageIndex> > b_ref(b.const_cast_derived());
+      cholmod_sparse b_cs = viewAsCholmod(b_ref);
+      cholmod_sparse* x_cs = cholmod_spsolve(CHOLMOD_A, m_cholmodFactor, &b_cs, &m_cholmod);
+      if(!x_cs)
+      {
+        this->m_info = NumericalIssue;
+        return;
+      }
+      // TODO optimize this copy by swapping when possible (be careful with alignment, etc.)
+      dest.derived() = viewAsEigen<typename DestDerived::Scalar,ColMajor,typename DestDerived::StorageIndex>(*x_cs);
+      cholmod_free_sparse(&x_cs, &m_cholmod);
+    }
+    #endif // EIGEN_PARSED_BY_DOXYGEN
+    
+    
+    /** Sets the shift parameter that will be used to adjust the diagonal coefficients during the numerical factorization.
+      *
+      * During the numerical factorization, an offset term is added to the diagonal coefficients:\n
+      * \c d_ii = \a offset + \c d_ii
+      *
+      * The default is \a offset=0.
+      *
+      * \returns a reference to \c *this.
+      */
+    Derived& setShift(const RealScalar& offset)
+    {
+      m_shiftOffset[0] = double(offset);
+      return derived();
+    }
+    
+    /** \returns the determinant of the underlying matrix from the current factorization */
+    Scalar determinant() const
+    {
+      using std::exp;
+      return exp(logDeterminant());
+    }
+
+    /** \returns the log determinant of the underlying matrix from the current factorization */
+    Scalar logDeterminant() const
+    {
+      using std::log;
+      using numext::real;
+      eigen_assert(m_factorizationIsOk && "The decomposition is not in a valid state for solving, you must first call either compute() or symbolic()/numeric()");
+
+      RealScalar logDet = 0;
+      Scalar *x = static_cast<Scalar*>(m_cholmodFactor->x);
+      if (m_cholmodFactor->is_super)
+      {
+        // Supernodal factorization stored as a packed list of dense column-major blocs,
+        // as described by the following structure:
+
+        // super[k] == index of the first column of the j-th super node
+        StorageIndex *super = static_cast<StorageIndex*>(m_cholmodFactor->super);
+        // pi[k] == offset to the description of row indices
+        StorageIndex *pi = static_cast<StorageIndex*>(m_cholmodFactor->pi);
+        // px[k] == offset to the respective dense block
+        StorageIndex *px = static_cast<StorageIndex*>(m_cholmodFactor->px);
+
+        Index nb_super_nodes = m_cholmodFactor->nsuper;
+        for (Index k=0; k < nb_super_nodes; ++k)
+        {
+          StorageIndex ncols = super[k + 1] - super[k];
+          StorageIndex nrows = pi[k + 1] - pi[k];
+
+          Map<const Array<Scalar,1,Dynamic>, 0, InnerStride<> > sk(x + px[k], ncols, InnerStride<>(nrows+1));
+          logDet += sk.real().log().sum();
+        }
+      }
+      else
+      {
+        // Simplicial factorization stored as standard CSC matrix.
+        StorageIndex *p = static_cast<StorageIndex*>(m_cholmodFactor->p);
+        Index size = m_cholmodFactor->n;
+        for (Index k=0; k<size; ++k)
+          logDet += log(real( x[p[k]] ));
+      }
+      if (m_cholmodFactor->is_ll)
+        logDet *= 2.0;
+      return logDet;
+    };
+
+    template<typename Stream>
+    void dumpMemory(Stream& /*s*/)
+    {}
+    
+  protected:
+    mutable cholmod_common m_cholmod;
+    cholmod_factor* m_cholmodFactor;
+    double m_shiftOffset[2];
+    mutable ComputationInfo m_info;
+    int m_factorizationIsOk;
+    int m_analysisIsOk;
+};
+
+/** \ingroup CholmodSupport_Module
+  * \class CholmodSimplicialLLT
+  * \brief A simplicial direct Cholesky (LLT) factorization and solver based on Cholmod
+  *
+  * This class allows to solve for A.X = B sparse linear problems via a simplicial LL^T Cholesky factorization
+  * using the Cholmod library.
+  * This simplicial variant is equivalent to Eigen's built-in SimplicialLLT class. Therefore, it has little practical interest.
+  * The sparse matrix A must be selfadjoint and positive definite. The vectors or matrices
+  * X and B can be either dense or sparse.
+  *
+  * \tparam _MatrixType the type of the sparse matrix A, it must be a SparseMatrix<>
+  * \tparam _UpLo the triangular part that will be used for the computations. It can be Lower
+  *               or Upper. Default is Lower.
+  *
+  * \implsparsesolverconcept
+  *
+  * This class supports all kind of SparseMatrix<>: row or column major; upper, lower, or both; compressed or non compressed.
+  *
+  * \warning Only double precision real and complex scalar types are supported by Cholmod.
+  *
+  * \sa \ref TutorialSparseSolverConcept, class CholmodSupernodalLLT, class SimplicialLLT
+  */
+template<typename _MatrixType, int _UpLo = Lower>
+class CholmodSimplicialLLT : public CholmodBase<_MatrixType, _UpLo, CholmodSimplicialLLT<_MatrixType, _UpLo> >
+{
+    typedef CholmodBase<_MatrixType, _UpLo, CholmodSimplicialLLT> Base;
+    using Base::m_cholmod;
+    
+  public:
+    
+    typedef _MatrixType MatrixType;
+    
+    CholmodSimplicialLLT() : Base() { init(); }
+
+    CholmodSimplicialLLT(const MatrixType& matrix) : Base()
+    {
+      init();
+      this->compute(matrix);
+    }
+
+    ~CholmodSimplicialLLT() {}
+  protected:
+    void init()
+    {
+      m_cholmod.final_asis = 0;
+      m_cholmod.supernodal = CHOLMOD_SIMPLICIAL;
+      m_cholmod.final_ll = 1;
+    }
+};
+
+
+/** \ingroup CholmodSupport_Module
+  * \class CholmodSimplicialLDLT
+  * \brief A simplicial direct Cholesky (LDLT) factorization and solver based on Cholmod
+  *
+  * This class allows to solve for A.X = B sparse linear problems via a simplicial LDL^T Cholesky factorization
+  * using the Cholmod library.
+  * This simplicial variant is equivalent to Eigen's built-in SimplicialLDLT class. Therefore, it has little practical interest.
+  * The sparse matrix A must be selfadjoint and positive definite. The vectors or matrices
+  * X and B can be either dense or sparse.
+  *
+  * \tparam _MatrixType the type of the sparse matrix A, it must be a SparseMatrix<>
+  * \tparam _UpLo the triangular part that will be used for the computations. It can be Lower
+  *               or Upper. Default is Lower.
+  *
+  * \implsparsesolverconcept
+  *
+  * This class supports all kind of SparseMatrix<>: row or column major; upper, lower, or both; compressed or non compressed.
+  *
+  * \warning Only double precision real and complex scalar types are supported by Cholmod.
+  *
+  * \sa \ref TutorialSparseSolverConcept, class CholmodSupernodalLLT, class SimplicialLDLT
+  */
+template<typename _MatrixType, int _UpLo = Lower>
+class CholmodSimplicialLDLT : public CholmodBase<_MatrixType, _UpLo, CholmodSimplicialLDLT<_MatrixType, _UpLo> >
+{
+    typedef CholmodBase<_MatrixType, _UpLo, CholmodSimplicialLDLT> Base;
+    using Base::m_cholmod;
+    
+  public:
+    
+    typedef _MatrixType MatrixType;
+    
+    CholmodSimplicialLDLT() : Base() { init(); }
+
+    CholmodSimplicialLDLT(const MatrixType& matrix) : Base()
+    {
+      init();
+      this->compute(matrix);
+    }
+
+    ~CholmodSimplicialLDLT() {}
+  protected:
+    void init()
+    {
+      m_cholmod.final_asis = 1;
+      m_cholmod.supernodal = CHOLMOD_SIMPLICIAL;
+    }
+};
+
+/** \ingroup CholmodSupport_Module
+  * \class CholmodSupernodalLLT
+  * \brief A supernodal Cholesky (LLT) factorization and solver based on Cholmod
+  *
+  * This class allows to solve for A.X = B sparse linear problems via a supernodal LL^T Cholesky factorization
+  * using the Cholmod library.
+  * This supernodal variant performs best on dense enough problems, e.g., 3D FEM, or very high order 2D FEM.
+  * The sparse matrix A must be selfadjoint and positive definite. The vectors or matrices
+  * X and B can be either dense or sparse.
+  *
+  * \tparam _MatrixType the type of the sparse matrix A, it must be a SparseMatrix<>
+  * \tparam _UpLo the triangular part that will be used for the computations. It can be Lower
+  *               or Upper. Default is Lower.
+  *
+  * \implsparsesolverconcept
+  *
+  * This class supports all kind of SparseMatrix<>: row or column major; upper, lower, or both; compressed or non compressed.
+  *
+  * \warning Only double precision real and complex scalar types are supported by Cholmod.
+  *
+  * \sa \ref TutorialSparseSolverConcept
+  */
+template<typename _MatrixType, int _UpLo = Lower>
+class CholmodSupernodalLLT : public CholmodBase<_MatrixType, _UpLo, CholmodSupernodalLLT<_MatrixType, _UpLo> >
+{
+    typedef CholmodBase<_MatrixType, _UpLo, CholmodSupernodalLLT> Base;
+    using Base::m_cholmod;
+    
+  public:
+    
+    typedef _MatrixType MatrixType;
+    
+    CholmodSupernodalLLT() : Base() { init(); }
+
+    CholmodSupernodalLLT(const MatrixType& matrix) : Base()
+    {
+      init();
+      this->compute(matrix);
+    }
+
+    ~CholmodSupernodalLLT() {}
+  protected:
+    void init()
+    {
+      m_cholmod.final_asis = 1;
+      m_cholmod.supernodal = CHOLMOD_SUPERNODAL;
+    }
+};
+
+/** \ingroup CholmodSupport_Module
+  * \class CholmodDecomposition
+  * \brief A general Cholesky factorization and solver based on Cholmod
+  *
+  * This class allows to solve for A.X = B sparse linear problems via a LL^T or LDL^T Cholesky factorization
+  * using the Cholmod library. The sparse matrix A must be selfadjoint and positive definite. The vectors or matrices
+  * X and B can be either dense or sparse.
+  *
+  * This variant permits to change the underlying Cholesky method at runtime.
+  * On the other hand, it does not provide access to the result of the factorization.
+  * The default is to let Cholmod automatically choose between a simplicial and supernodal factorization.
+  *
+  * \tparam _MatrixType the type of the sparse matrix A, it must be a SparseMatrix<>
+  * \tparam _UpLo the triangular part that will be used for the computations. It can be Lower
+  *               or Upper. Default is Lower.
+  *
+  * \implsparsesolverconcept
+  *
+  * This class supports all kind of SparseMatrix<>: row or column major; upper, lower, or both; compressed or non compressed.
+  *
+  * \warning Only double precision real and complex scalar types are supported by Cholmod.
+  *
+  * \sa \ref TutorialSparseSolverConcept
+  */
+template<typename _MatrixType, int _UpLo = Lower>
+class CholmodDecomposition : public CholmodBase<_MatrixType, _UpLo, CholmodDecomposition<_MatrixType, _UpLo> >
+{
+    typedef CholmodBase<_MatrixType, _UpLo, CholmodDecomposition> Base;
+    using Base::m_cholmod;
+    
+  public:
+    
+    typedef _MatrixType MatrixType;
+    
+    CholmodDecomposition() : Base() { init(); }
+
+    CholmodDecomposition(const MatrixType& matrix) : Base()
+    {
+      init();
+      this->compute(matrix);
+    }
+
+    ~CholmodDecomposition() {}
+    
+    void setMode(CholmodMode mode)
+    {
+      switch(mode)
+      {
+        case CholmodAuto:
+          m_cholmod.final_asis = 1;
+          m_cholmod.supernodal = CHOLMOD_AUTO;
+          break;
+        case CholmodSimplicialLLt:
+          m_cholmod.final_asis = 0;
+          m_cholmod.supernodal = CHOLMOD_SIMPLICIAL;
+          m_cholmod.final_ll = 1;
+          break;
+        case CholmodSupernodalLLt:
+          m_cholmod.final_asis = 1;
+          m_cholmod.supernodal = CHOLMOD_SUPERNODAL;
+          break;
+        case CholmodLDLt:
+          m_cholmod.final_asis = 1;
+          m_cholmod.supernodal = CHOLMOD_SIMPLICIAL;
+          break;
+        default:
+          break;
+      }
+    }
+  protected:
+    void init()
+    {
+      m_cholmod.final_asis = 1;
+      m_cholmod.supernodal = CHOLMOD_AUTO;
+    }
+};
+
+} // end namespace Eigen
+
+#endif // EIGEN_CHOLMODSUPPORT_H

Eigen/src/Core/Array.h

@@ -0,0 +1,329 @@
+// This file is part of Eigen, a lightweight C++ template library
+// for linear algebra.
+//
+// Copyright (C) 2009 Gael Guennebaud <gael.guennebaud@inria.fr>
+//
+// This Source Code Form is subject to the terms of the Mozilla
+// Public License v. 2.0. If a copy of the MPL was not distributed
+// with this file, You can obtain one at http://mozilla.org/MPL/2.0/.
+
+#ifndef EIGEN_ARRAY_H
+#define EIGEN_ARRAY_H
+
+namespace Eigen {
+
+namespace internal {
+template<typename _Scalar, int _Rows, int _Cols, int _Options, int _MaxRows, int _MaxCols>
+struct traits<Array<_Scalar, _Rows, _Cols, _Options, _MaxRows, _MaxCols> > : traits<Matrix<_Scalar, _Rows, _Cols, _Options, _MaxRows, _MaxCols> >
+{
+  typedef ArrayXpr XprKind;
+  typedef ArrayBase<Array<_Scalar, _Rows, _Cols, _Options, _MaxRows, _MaxCols> > XprBase;
+};
+}
+
+/** \class Array
+  * \ingroup Core_Module
+  *
+  * \brief General-purpose arrays with easy API for coefficient-wise operations
+  *
+  * The %Array class is very similar to the Matrix class. It provides
+  * general-purpose one- and two-dimensional arrays. The difference between the
+  * %Array and the %Matrix class is primarily in the API: the API for the
+  * %Array class provides easy access to coefficient-wise operations, while the
+  * API for the %Matrix class provides easy access to linear-algebra
+  * operations.
+  *
+  * See documentation of class Matrix for detailed information on the template parameters
+  * storage layout.
+  *
+  * This class can be extended with the help of the plugin mechanism described on the page
+  * \ref TopicCustomizing_Plugins by defining the preprocessor symbol \c EIGEN_ARRAY_PLUGIN.
+  *
+  * \sa \blank \ref TutorialArrayClass, \ref TopicClassHierarchy
+  */
+template<typename _Scalar, int _Rows, int _Cols, int _Options, int _MaxRows, int _MaxCols>
+class Array
+  : public PlainObjectBase<Array<_Scalar, _Rows, _Cols, _Options, _MaxRows, _MaxCols> >
+{
+  public:
+
+    typedef PlainObjectBase<Array> Base;
+    EIGEN_DENSE_PUBLIC_INTERFACE(Array)
+
+    enum { Options = _Options };
+    typedef typename Base::PlainObject PlainObject;
+
+  protected:
+    template <typename Derived, typename OtherDerived, bool IsVector>
+    friend struct internal::conservative_resize_like_impl;
+
+    using Base::m_storage;
+
+  public:
+
+    using Base::base;
+    using Base::coeff;
+    using Base::coeffRef;
+
+    /**
+      * The usage of
+      *   using Base::operator=;
+      * fails on MSVC. Since the code below is working with GCC and MSVC, we skipped
+      * the usage of 'using'. This should be done only for operator=.
+      */
+    template<typename OtherDerived>
+    EIGEN_DEVICE_FUNC
+    EIGEN_STRONG_INLINE Array& operator=(const EigenBase<OtherDerived> &other)
+    {
+      return Base::operator=(other);
+    }
+
+    /** Set all the entries to \a value.
+      * \sa DenseBase::setConstant(), DenseBase::fill()
+      */
+    /* This overload is needed because the usage of
+      *   using Base::operator=;
+      * fails on MSVC. Since the code below is working with GCC and MSVC, we skipped
+      * the usage of 'using'. This should be done only for operator=.
+      */
+    EIGEN_DEVICE_FUNC
+    EIGEN_STRONG_INLINE Array& operator=(const Scalar &value)
+    {
+      Base::setConstant(value);
+      return *this;
+    }
+
+    /** Copies the value of the expression \a other into \c *this with automatic resizing.
+      *
+      * *this might be resized to match the dimensions of \a other. If *this was a null matrix (not already initialized),
+      * it will be initialized.
+      *
+      * Note that copying a row-vector into a vector (and conversely) is allowed.
+      * The resizing, if any, is then done in the appropriate way so that row-vectors
+      * remain row-vectors and vectors remain vectors.
+      */
+    template<typename OtherDerived>
+    EIGEN_DEVICE_FUNC
+    EIGEN_STRONG_INLINE Array& operator=(const DenseBase<OtherDerived>& other)
+    {
+      return Base::_set(other);
+    }
+
+    /** This is a special case of the templated operator=. Its purpose is to
+      * prevent a default operator= from hiding the templated operator=.
+      */
+    EIGEN_DEVICE_FUNC
+    EIGEN_STRONG_INLINE Array& operator=(const Array& other)
+    {
+      return Base::_set(other);
+    }
+    
+    /** Default constructor.
+      *
+      * For fixed-size matrices, does nothing.
+      *
+      * For dynamic-size matrices, creates an empty matrix of size 0. Does not allocate any array. Such a matrix
+      * is called a null matrix. This constructor is the unique way to create null matrices: resizing
+      * a matrix to 0 is not supported.
+      *
+      * \sa resize(Index,Index)
+      */
+    EIGEN_DEVICE_FUNC
+    EIGEN_STRONG_INLINE Array() : Base()
+    {
+      Base::_check_template_params();
+      EIGEN_INITIALIZE_COEFFS_IF_THAT_OPTION_IS_ENABLED
+    }
+
+#ifndef EIGEN_PARSED_BY_DOXYGEN
+    // FIXME is it still needed ??
+    /** \internal */
+    EIGEN_DEVICE_FUNC
+    Array(internal::constructor_without_unaligned_array_assert)
+      : Base(internal::constructor_without_unaligned_array_assert())
+    {
+      Base::_check_template_params();
+      EIGEN_INITIALIZE_COEFFS_IF_THAT_OPTION_IS_ENABLED
+    }
+#endif
+
+#if EIGEN_HAS_RVALUE_REFERENCES
+    EIGEN_DEVICE_FUNC
+    Array(Array&& other) EIGEN_NOEXCEPT_IF(std::is_nothrow_move_constructible<Scalar>::value)
+      : Base(std::move(other))
+    {
+      Base::_check_template_params();
+    }
+    EIGEN_DEVICE_FUNC
+    Array& operator=(Array&& other) EIGEN_NOEXCEPT_IF(std::is_nothrow_move_assignable<Scalar>::value)
+    {
+      other.swap(*this);
+      return *this;
+    }
+#endif
+
+    #ifndef EIGEN_PARSED_BY_DOXYGEN
+    template<typename T>
+    EIGEN_DEVICE_FUNC
+    EIGEN_STRONG_INLINE explicit Array(const T& x)
+    {
+      Base::_check_template_params();
+      Base::template _init1<T>(x);
+    }
+
+    template<typename T0, typename T1>
+    EIGEN_DEVICE_FUNC
+    EIGEN_STRONG_INLINE Array(const T0& val0, const T1& val1)
+    {
+      Base::_check_template_params();
+      this->template _init2<T0,T1>(val0, val1);
+    }
+    #else
+    /** \brief Constructs a fixed-sized array initialized with coefficients starting at \a data */
+    EIGEN_DEVICE_FUNC explicit Array(const Scalar *data);
+    /** Constructs a vector or row-vector with given dimension. \only_for_vectors
+      *
+      * Note that this is only useful for dynamic-size vectors. For fixed-size vectors,
+      * it is redundant to pass the dimension here, so it makes more sense to use the default
+      * constructor Array() instead.
+      */
+    EIGEN_DEVICE_FUNC
+    EIGEN_STRONG_INLINE explicit Array(Index dim);
+    /** constructs an initialized 1x1 Array with the given coefficient */
+    Array(const Scalar& value);
+    /** constructs an uninitialized array with \a rows rows and \a cols columns.
+      *
+      * This is useful for dynamic-size arrays. For fixed-size arrays,
+      * it is redundant to pass these parameters, so one should use the default constructor
+      * Array() instead. */
+    Array(Index rows, Index cols);
+    /** constructs an initialized 2D vector with given coefficients */
+    Array(const Scalar& val0, const Scalar& val1);
+    #endif
+
+    /** constructs an initialized 3D vector with given coefficients */
+    EIGEN_DEVICE_FUNC
+    EIGEN_STRONG_INLINE Array(const Scalar& val0, const Scalar& val1, const Scalar& val2)
+    {
+      Base::_check_template_params();
+      EIGEN_STATIC_ASSERT_VECTOR_SPECIFIC_SIZE(Array, 3)
+      m_storage.data()[0] = val0;
+      m_storage.data()[1] = val1;
+      m_storage.data()[2] = val2;
+    }
+    /** constructs an initialized 4D vector with given coefficients */
+    EIGEN_DEVICE_FUNC
+    EIGEN_STRONG_INLINE Array(const Scalar& val0, const Scalar& val1, const Scalar& val2, const Scalar& val3)
+    {
+      Base::_check_template_params();
+      EIGEN_STATIC_ASSERT_VECTOR_SPECIFIC_SIZE(Array, 4)
+      m_storage.data()[0] = val0;
+      m_storage.data()[1] = val1;
+      m_storage.data()[2] = val2;
+      m_storage.data()[3] = val3;
+    }
+
+    /** Copy constructor */
+    EIGEN_DEVICE_FUNC
+    EIGEN_STRONG_INLINE Array(const Array& other)
+            : Base(other)
+    { }
+
+  private:
+    struct PrivateType {};
+  public:
+
+    /** \sa MatrixBase::operator=(const EigenBase<OtherDerived>&) */
+    template<typename OtherDerived>
+    EIGEN_DEVICE_FUNC
+    EIGEN_STRONG_INLINE Array(const EigenBase<OtherDerived> &other,
+                              typename internal::enable_if<internal::is_convertible<typename OtherDerived::Scalar,Scalar>::value,
+                                                           PrivateType>::type = PrivateType())
+      : Base(other.derived())
+    { }
+
+    EIGEN_DEVICE_FUNC inline Index innerStride() const { return 1; }
+    EIGEN_DEVICE_FUNC inline Index outerStride() const { return this->innerSize(); }
+
+    #ifdef EIGEN_ARRAY_PLUGIN
+    #include EIGEN_ARRAY_PLUGIN
+    #endif
+
+  private:
+
+    template<typename MatrixType, typename OtherDerived, bool SwapPointers>
+    friend struct internal::matrix_swap_impl;
+};
+
+/** \defgroup arraytypedefs Global array typedefs
+  * \ingroup Core_Module
+  *
+  * Eigen defines several typedef shortcuts for most common 1D and 2D array types.
+  *
+  * The general patterns are the following:
+  *
+  * \c ArrayRowsColsType where \c Rows and \c Cols can be \c 2,\c 3,\c 4 for fixed size square matrices or \c X for dynamic size,
+  * and where \c Type can be \c i for integer, \c f for float, \c d for double, \c cf for complex float, \c cd
+  * for complex double.
+  *
+  * For example, \c Array33d is a fixed-size 3x3 array type of doubles, and \c ArrayXXf is a dynamic-size matrix of floats.
+  *
+  * There are also \c ArraySizeType which are self-explanatory. For example, \c Array4cf is
+  * a fixed-size 1D array of 4 complex floats.
+  *
+  * \sa class Array
+  */
+
+#define EIGEN_MAKE_ARRAY_TYPEDEFS(Type, TypeSuffix, Size, SizeSuffix)   \
+/** \ingroup arraytypedefs */                                    \
+typedef Array<Type, Size, Size> Array##SizeSuffix##SizeSuffix##TypeSuffix;  \
+/** \ingroup arraytypedefs */                                    \
+typedef Array<Type, Size, 1>    Array##SizeSuffix##TypeSuffix;
+
+#define EIGEN_MAKE_ARRAY_FIXED_TYPEDEFS(Type, TypeSuffix, Size)         \
+/** \ingroup arraytypedefs */                                    \
+typedef Array<Type, Size, Dynamic> Array##Size##X##TypeSuffix;  \
+/** \ingroup arraytypedefs */                                    \
+typedef Array<Type, Dynamic, Size> Array##X##Size##TypeSuffix;
+
+#define EIGEN_MAKE_ARRAY_TYPEDEFS_ALL_SIZES(Type, TypeSuffix) \
+EIGEN_MAKE_ARRAY_TYPEDEFS(Type, TypeSuffix, 2, 2) \
+EIGEN_MAKE_ARRAY_TYPEDEFS(Type, TypeSuffix, 3, 3) \
+EIGEN_MAKE_ARRAY_TYPEDEFS(Type, TypeSuffix, 4, 4) \
+EIGEN_MAKE_ARRAY_TYPEDEFS(Type, TypeSuffix, Dynamic, X) \
+EIGEN_MAKE_ARRAY_FIXED_TYPEDEFS(Type, TypeSuffix, 2) \
+EIGEN_MAKE_ARRAY_FIXED_TYPEDEFS(Type, TypeSuffix, 3) \
+EIGEN_MAKE_ARRAY_FIXED_TYPEDEFS(Type, TypeSuffix, 4)
+
+EIGEN_MAKE_ARRAY_TYPEDEFS_ALL_SIZES(int,                  i)
+EIGEN_MAKE_ARRAY_TYPEDEFS_ALL_SIZES(float,                f)
+EIGEN_MAKE_ARRAY_TYPEDEFS_ALL_SIZES(double,               d)
+EIGEN_MAKE_ARRAY_TYPEDEFS_ALL_SIZES(std::complex<float>,  cf)
+EIGEN_MAKE_ARRAY_TYPEDEFS_ALL_SIZES(std::complex<double>, cd)
+
+#undef EIGEN_MAKE_ARRAY_TYPEDEFS_ALL_SIZES
+#undef EIGEN_MAKE_ARRAY_TYPEDEFS
+
+#undef EIGEN_MAKE_ARRAY_TYPEDEFS_LARGE
+
+#define EIGEN_USING_ARRAY_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_ARRAY_TYPEDEFS_FOR_TYPE(TypeSuffix) \
+EIGEN_USING_ARRAY_TYPEDEFS_FOR_TYPE_AND_SIZE(TypeSuffix, 2) \
+EIGEN_USING_ARRAY_TYPEDEFS_FOR_TYPE_AND_SIZE(TypeSuffix, 3) \
+EIGEN_USING_ARRAY_TYPEDEFS_FOR_TYPE_AND_SIZE(TypeSuffix, 4) \
+EIGEN_USING_ARRAY_TYPEDEFS_FOR_TYPE_AND_SIZE(TypeSuffix, X) \
+
+#define EIGEN_USING_ARRAY_TYPEDEFS \
+EIGEN_USING_ARRAY_TYPEDEFS_FOR_TYPE(i) \
+EIGEN_USING_ARRAY_TYPEDEFS_FOR_TYPE(f) \
+EIGEN_USING_ARRAY_TYPEDEFS_FOR_TYPE(d) \
+EIGEN_USING_ARRAY_TYPEDEFS_FOR_TYPE(cf) \
+EIGEN_USING_ARRAY_TYPEDEFS_FOR_TYPE(cd)
+
+} // end namespace Eigen
+
+#endif // EIGEN_ARRAY_H

Eigen/src/Core/ArrayBase.h

@@ -0,0 +1,226 @@
+// This file is part of Eigen, a lightweight C++ template library
+// for linear algebra.
+//
+// Copyright (C) 2009 Gael Guennebaud <gael.guennebaud@inria.fr>
+//
+// This Source Code Form is subject to the terms of the Mozilla
+// Public License v. 2.0. If a copy of the MPL was not distributed
+// with this file, You can obtain one at http://mozilla.org/MPL/2.0/.
+
+#ifndef EIGEN_ARRAYBASE_H
+#define EIGEN_ARRAYBASE_H
+
+namespace Eigen { 
+
+template<typename ExpressionType> class MatrixWrapper;
+
+/** \class ArrayBase
+  * \ingroup Core_Module
+  *
+  * \brief Base class for all 1D and 2D array, and related expressions
+  *
+  * An array is similar to a dense vector or matrix. While matrices are mathematical
+  * objects with well defined linear algebra operators, an array is just a collection
+  * of scalar values arranged in a one or two dimensionnal fashion. As the main consequence,
+  * all operations applied to an array are performed coefficient wise. Furthermore,
+  * arrays support scalar math functions of the c++ standard library (e.g., std::sin(x)), and convenient
+  * constructors allowing to easily write generic code working for both scalar values
+  * and arrays.
+  *
+  * This class is the base that is inherited by all array expression types.
+  *
+  * \tparam Derived is the derived type, e.g., an array or an expression type.
+  *
+  * This class can be extended with the help of the plugin mechanism described on the page
+  * \ref TopicCustomizing_Plugins by defining the preprocessor symbol \c EIGEN_ARRAYBASE_PLUGIN.
+  *
+  * \sa class MatrixBase, \ref TopicClassHierarchy
+  */
+template<typename Derived> class ArrayBase
+  : public DenseBase<Derived>
+{
+  public:
+#ifndef EIGEN_PARSED_BY_DOXYGEN
+    /** The base class for a given storage type. */
+    typedef ArrayBase StorageBaseType;
+
+    typedef ArrayBase Eigen_BaseClassForSpecializationOfGlobalMathFuncImpl;
+
+    typedef typename internal::traits<Derived>::StorageKind StorageKind;
+    typedef typename internal::traits<Derived>::Scalar Scalar;
+    typedef typename internal::packet_traits<Scalar>::type PacketScalar;
+    typedef typename NumTraits<Scalar>::Real RealScalar;
+
+    typedef DenseBase<Derived> Base;
+    using Base::RowsAtCompileTime;
+    using Base::ColsAtCompileTime;
+    using Base::SizeAtCompileTime;
+    using Base::MaxRowsAtCompileTime;
+    using Base::MaxColsAtCompileTime;
+    using Base::MaxSizeAtCompileTime;
+    using Base::IsVectorAtCompileTime;
+    using Base::Flags;
+    
+    using Base::derived;
+    using Base::const_cast_derived;
+    using Base::rows;
+    using Base::cols;
+    using Base::size;
+    using Base::coeff;
+    using Base::coeffRef;
+    using Base::lazyAssign;
+    using Base::operator=;
+    using Base::operator+=;
+    using Base::operator-=;
+    using Base::operator*=;
+    using Base::operator/=;
+
+    typedef typename Base::CoeffReturnType CoeffReturnType;
+
+#endif // not EIGEN_PARSED_BY_DOXYGEN
+
+#ifndef EIGEN_PARSED_BY_DOXYGEN
+    typedef typename Base::PlainObject PlainObject;
+
+    /** \internal Represents a matrix with all coefficients equal to one another*/
+    typedef CwiseNullaryOp<internal::scalar_constant_op<Scalar>,PlainObject> ConstantReturnType;
+#endif // not EIGEN_PARSED_BY_DOXYGEN
+
+#define EIGEN_CURRENT_STORAGE_BASE_CLASS Eigen::ArrayBase
+#define EIGEN_DOC_UNARY_ADDONS(X,Y)
+#   include "../plugins/CommonCwiseUnaryOps.h"
+#   include "../plugins/MatrixCwiseUnaryOps.h"
+#   include "../plugins/ArrayCwiseUnaryOps.h"
+#   include "../plugins/CommonCwiseBinaryOps.h"
+#   include "../plugins/MatrixCwiseBinaryOps.h"
+#   include "../plugins/ArrayCwiseBinaryOps.h"
+#   ifdef EIGEN_ARRAYBASE_PLUGIN
+#     include EIGEN_ARRAYBASE_PLUGIN
+#   endif
+#undef EIGEN_CURRENT_STORAGE_BASE_CLASS
+#undef EIGEN_DOC_UNARY_ADDONS
+
+    /** Special case of the template operator=, in order to prevent the compiler
+      * from generating a default operator= (issue hit with g++ 4.1)
+      */
+    EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
+    Derived& operator=(const ArrayBase& other)
+    {
+      internal::call_assignment(derived(), other.derived());
+      return derived();
+    }
+    
+    /** Set all the entries to \a value.
+      * \sa DenseBase::setConstant(), DenseBase::fill() */
+    EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
+    Derived& operator=(const Scalar &value)
+    { Base::setConstant(value); return derived(); }
+
+    EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
+    Derived& operator+=(const Scalar& scalar);
+    EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
+    Derived& operator-=(const Scalar& scalar);
+
+    template<typename OtherDerived>
+    EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
+    Derived& operator+=(const ArrayBase<OtherDerived>& other);
+    template<typename OtherDerived>
+    EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
+    Derived& operator-=(const ArrayBase<OtherDerived>& other);
+
+    template<typename OtherDerived>
+    EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
+    Derived& operator*=(const ArrayBase<OtherDerived>& other);
+
+    template<typename OtherDerived>
+    EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
+    Derived& operator/=(const ArrayBase<OtherDerived>& other);
+
+  public:
+    EIGEN_DEVICE_FUNC
+    ArrayBase<Derived>& array() { return *this; }
+    EIGEN_DEVICE_FUNC
+    const ArrayBase<Derived>& array() const { return *this; }
+
+    /** \returns an \link Eigen::MatrixBase Matrix \endlink expression of this array
+      * \sa MatrixBase::array() */
+    EIGEN_DEVICE_FUNC
+    MatrixWrapper<Derived> matrix() { return MatrixWrapper<Derived>(derived()); }
+    EIGEN_DEVICE_FUNC
+    const MatrixWrapper<const Derived> matrix() const { return MatrixWrapper<const Derived>(derived()); }
+
+//     template<typename Dest>
+//     inline void evalTo(Dest& dst) const { dst = matrix(); }
+
+  protected:
+    EIGEN_DEFAULT_COPY_CONSTRUCTOR(ArrayBase)
+    EIGEN_DEFAULT_EMPTY_CONSTRUCTOR_AND_DESTRUCTOR(ArrayBase)
+
+  private:
+    explicit ArrayBase(Index);
+    ArrayBase(Index,Index);
+    template<typename OtherDerived> explicit ArrayBase(const ArrayBase<OtherDerived>&);
+  protected:
+    // mixing arrays and matrices is not legal
+    template<typename OtherDerived> Derived& operator+=(const MatrixBase<OtherDerived>& )
+    {EIGEN_STATIC_ASSERT(std::ptrdiff_t(sizeof(typename OtherDerived::Scalar))==-1,YOU_CANNOT_MIX_ARRAYS_AND_MATRICES); return *this;}
+    // mixing arrays and matrices is not legal
+    template<typename OtherDerived> Derived& operator-=(const MatrixBase<OtherDerived>& )
+    {EIGEN_STATIC_ASSERT(std::ptrdiff_t(sizeof(typename OtherDerived::Scalar))==-1,YOU_CANNOT_MIX_ARRAYS_AND_MATRICES); return *this;}
+};
+
+/** replaces \c *this by \c *this - \a other.
+  *
+  * \returns a reference to \c *this
+  */
+template<typename Derived>
+template<typename OtherDerived>
+EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE Derived &
+ArrayBase<Derived>::operator-=(const ArrayBase<OtherDerived> &other)
+{
+  call_assignment(derived(), other.derived(), internal::sub_assign_op<Scalar,typename OtherDerived::Scalar>());
+  return derived();
+}
+
+/** replaces \c *this by \c *this + \a other.
+  *
+  * \returns a reference to \c *this
+  */
+template<typename Derived>
+template<typename OtherDerived>
+EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE Derived &
+ArrayBase<Derived>::operator+=(const ArrayBase<OtherDerived>& other)
+{
+  call_assignment(derived(), other.derived(), internal::add_assign_op<Scalar,typename OtherDerived::Scalar>());
+  return derived();
+}
+
+/** replaces \c *this by \c *this * \a other coefficient wise.
+  *
+  * \returns a reference to \c *this
+  */
+template<typename Derived>
+template<typename OtherDerived>
+EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE Derived &
+ArrayBase<Derived>::operator*=(const ArrayBase<OtherDerived>& other)
+{
+  call_assignment(derived(), other.derived(), internal::mul_assign_op<Scalar,typename OtherDerived::Scalar>());
+  return derived();
+}
+
+/** replaces \c *this by \c *this / \a other coefficient wise.
+  *
+  * \returns a reference to \c *this
+  */
+template<typename Derived>
+template<typename OtherDerived>
+EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE Derived &
+ArrayBase<Derived>::operator/=(const ArrayBase<OtherDerived>& other)
+{
+  call_assignment(derived(), other.derived(), internal::div_assign_op<Scalar,typename OtherDerived::Scalar>());
+  return derived();
+}
+
+} // end namespace Eigen
+
+#endif // EIGEN_ARRAYBASE_H

Eigen/src/Core/ArrayWrapper.h

@@ -0,0 +1,209 @@
+// This file is part of Eigen, a lightweight C++ template library
+// for linear algebra.
+//
+// Copyright (C) 2009-2010 Gael Guennebaud <gael.guennebaud@inria.fr>
+//
+// This Source Code Form is subject to the terms of the Mozilla
+// Public License v. 2.0. If a copy of the MPL was not distributed
+// with this file, You can obtain one at http://mozilla.org/MPL/2.0/.
+
+#ifndef EIGEN_ARRAYWRAPPER_H
+#define EIGEN_ARRAYWRAPPER_H
+
+namespace Eigen { 
+
+/** \class ArrayWrapper
+  * \ingroup Core_Module
+  *
+  * \brief Expression of a mathematical vector or matrix as an array object
+  *
+  * This class is the return type of MatrixBase::array(), and most of the time
+  * this is the only way it is use.
+  *
+  * \sa MatrixBase::array(), class MatrixWrapper
+  */
+
+namespace internal {
+template<typename ExpressionType>
+struct traits<ArrayWrapper<ExpressionType> >
+  : public traits<typename remove_all<typename ExpressionType::Nested>::type >
+{
+  typedef ArrayXpr XprKind;
+  // Let's remove NestByRefBit
+  enum {
+    Flags0 = traits<typename remove_all<typename ExpressionType::Nested>::type >::Flags,
+    LvalueBitFlag = is_lvalue<ExpressionType>::value ? LvalueBit : 0,
+    Flags = (Flags0 & ~(NestByRefBit | LvalueBit)) | LvalueBitFlag
+  };
+};
+}
+
+template<typename ExpressionType>
+class ArrayWrapper : public ArrayBase<ArrayWrapper<ExpressionType> >
+{
+  public:
+    typedef ArrayBase<ArrayWrapper> Base;
+    EIGEN_DENSE_PUBLIC_INTERFACE(ArrayWrapper)
+    EIGEN_INHERIT_ASSIGNMENT_OPERATORS(ArrayWrapper)
+    typedef typename internal::remove_all<ExpressionType>::type NestedExpression;
+
+    typedef typename internal::conditional<
+                       internal::is_lvalue<ExpressionType>::value,
+                       Scalar,
+                       const Scalar
+                     >::type ScalarWithConstIfNotLvalue;
+
+    typedef typename internal::ref_selector<ExpressionType>::non_const_type NestedExpressionType;
+
+    using Base::coeffRef;
+
+    EIGEN_DEVICE_FUNC
+    explicit EIGEN_STRONG_INLINE ArrayWrapper(ExpressionType& matrix) : m_expression(matrix) {}
+
+    EIGEN_DEVICE_FUNC
+    inline Index rows() const { return m_expression.rows(); }
+    EIGEN_DEVICE_FUNC
+    inline Index cols() const { return m_expression.cols(); }
+    EIGEN_DEVICE_FUNC
+    inline Index outerStride() const { return m_expression.outerStride(); }
+    EIGEN_DEVICE_FUNC
+    inline Index innerStride() const { return m_expression.innerStride(); }
+
+    EIGEN_DEVICE_FUNC
+    inline ScalarWithConstIfNotLvalue* data() { return m_expression.data(); }
+    EIGEN_DEVICE_FUNC
+    inline const Scalar* data() const { return m_expression.data(); }
+
+    EIGEN_DEVICE_FUNC
+    inline const Scalar& coeffRef(Index rowId, Index colId) const
+    {
+      return m_expression.coeffRef(rowId, colId);
+    }
+
+    EIGEN_DEVICE_FUNC
+    inline const Scalar& coeffRef(Index index) const
+    {
+      return m_expression.coeffRef(index);
+    }
+
+    template<typename Dest>
+    EIGEN_DEVICE_FUNC
+    inline void evalTo(Dest& dst) const { dst = m_expression; }
+
+    const typename internal::remove_all<NestedExpressionType>::type& 
+    EIGEN_DEVICE_FUNC
+    nestedExpression() const 
+    {
+      return m_expression;
+    }
+
+    /** Forwards the resizing request to the nested expression
+      * \sa DenseBase::resize(Index)  */
+    EIGEN_DEVICE_FUNC
+    void resize(Index newSize) { m_expression.resize(newSize); }
+    /** Forwards the resizing request to the nested expression
+      * \sa DenseBase::resize(Index,Index)*/
+    EIGEN_DEVICE_FUNC
+    void resize(Index rows, Index cols) { m_expression.resize(rows,cols); }
+
+  protected:
+    NestedExpressionType m_expression;
+};
+
+/** \class MatrixWrapper
+  * \ingroup Core_Module
+  *
+  * \brief Expression of an array as a mathematical vector or matrix
+  *
+  * This class is the return type of ArrayBase::matrix(), and most of the time
+  * this is the only way it is use.
+  *
+  * \sa MatrixBase::matrix(), class ArrayWrapper
+  */
+
+namespace internal {
+template<typename ExpressionType>
+struct traits<MatrixWrapper<ExpressionType> >
+ : public traits<typename remove_all<typename ExpressionType::Nested>::type >
+{
+  typedef MatrixXpr XprKind;
+  // Let's remove NestByRefBit
+  enum {
+    Flags0 = traits<typename remove_all<typename ExpressionType::Nested>::type >::Flags,
+    LvalueBitFlag = is_lvalue<ExpressionType>::value ? LvalueBit : 0,
+    Flags = (Flags0 & ~(NestByRefBit | LvalueBit)) | LvalueBitFlag
+  };
+};
+}
+
+template<typename ExpressionType>
+class MatrixWrapper : public MatrixBase<MatrixWrapper<ExpressionType> >
+{
+  public:
+    typedef MatrixBase<MatrixWrapper<ExpressionType> > Base;
+    EIGEN_DENSE_PUBLIC_INTERFACE(MatrixWrapper)
+    EIGEN_INHERIT_ASSIGNMENT_OPERATORS(MatrixWrapper)
+    typedef typename internal::remove_all<ExpressionType>::type NestedExpression;
+
+    typedef typename internal::conditional<
+                       internal::is_lvalue<ExpressionType>::value,
+                       Scalar,
+                       const Scalar
+                     >::type ScalarWithConstIfNotLvalue;
+
+    typedef typename internal::ref_selector<ExpressionType>::non_const_type NestedExpressionType;
+
+    using Base::coeffRef;
+
+    EIGEN_DEVICE_FUNC
+    explicit inline MatrixWrapper(ExpressionType& matrix) : m_expression(matrix) {}
+
+    EIGEN_DEVICE_FUNC
+    inline Index rows() const { return m_expression.rows(); }
+    EIGEN_DEVICE_FUNC
+    inline Index cols() const { return m_expression.cols(); }
+    EIGEN_DEVICE_FUNC
+    inline Index outerStride() const { return m_expression.outerStride(); }
+    EIGEN_DEVICE_FUNC
+    inline Index innerStride() const { return m_expression.innerStride(); }
+
+    EIGEN_DEVICE_FUNC
+    inline ScalarWithConstIfNotLvalue* data() { return m_expression.data(); }
+    EIGEN_DEVICE_FUNC
+    inline const Scalar* data() const { return m_expression.data(); }
+
+    EIGEN_DEVICE_FUNC
+    inline const Scalar& coeffRef(Index rowId, Index colId) const
+    {
+      return m_expression.derived().coeffRef(rowId, colId);
+    }
+
+    EIGEN_DEVICE_FUNC
+    inline const Scalar& coeffRef(Index index) const
+    {
+      return m_expression.coeffRef(index);
+    }
+
+    EIGEN_DEVICE_FUNC
+    const typename internal::remove_all<NestedExpressionType>::type& 
+    nestedExpression() const 
+    {
+      return m_expression;
+    }
+
+    /** Forwards the resizing request to the nested expression
+      * \sa DenseBase::resize(Index)  */
+    EIGEN_DEVICE_FUNC
+    void resize(Index newSize) { m_expression.resize(newSize); }
+    /** Forwards the resizing request to the nested expression
+      * \sa DenseBase::resize(Index,Index)*/
+    EIGEN_DEVICE_FUNC
+    void resize(Index rows, Index cols) { m_expression.resize(rows,cols); }
+
+  protected:
+    NestedExpressionType m_expression;
+};
+
+} // end namespace Eigen
+
+#endif // EIGEN_ARRAYWRAPPER_H

Eigen/src/Core/Assign.h

@@ -1,445 +1,90 @@
 // This file is part of Eigen, a lightweight C++ template library
-// for linear algebra. Eigen itself is part of the KDE project.
+// for linear algebra.
 //
 // Copyright (C) 2007 Michael Olbrich <michael.olbrich@gmx.net>
-// Copyright (C) 2006-2008 Benoit Jacob <jacob.benoit.1@gmail.com>
-// Copyright (C) 2008 Gael Guennebaud <g.gael@free.fr>
+// Copyright (C) 2006-2010 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
-// 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/>.
+// This Source Code Form is subject to the terms of the Mozilla
+// Public License v. 2.0. If a copy of the MPL was not distributed
+// with this file, You can obtain one at http://mozilla.org/MPL/2.0/.
 
 #ifndef EIGEN_ASSIGN_H
 #define EIGEN_ASSIGN_H
 
-/***************************************************************************
-* Part 1 : the logic deciding a strategy for vectorization and unrolling
-***************************************************************************/
-
-template <typename Derived, typename OtherDerived>
-struct ei_assign_traits
-{
-public:
-  enum {
-    DstIsAligned = Derived::Flags & AlignedBit,
-    SrcIsAligned = OtherDerived::Flags & AlignedBit,
-    SrcAlignment = DstIsAligned && SrcIsAligned ? Aligned : Unaligned
-  };
-
-private:
-  enum {
-    InnerSize = int(Derived::Flags)&RowMajorBit
-              ? Derived::ColsAtCompileTime
-              : Derived::RowsAtCompileTime,
-    InnerMaxSize = int(Derived::Flags)&RowMajorBit
-              ? Derived::MaxColsAtCompileTime
-              : Derived::MaxRowsAtCompileTime,
-    PacketSize = ei_packet_traits<typename Derived::Scalar>::size
-  };
-
-  enum {
-    MightVectorize = (int(Derived::Flags) & int(OtherDerived::Flags) & ActualPacketAccessBit)
-                  && ((int(Derived::Flags)&RowMajorBit)==(int(OtherDerived::Flags)&RowMajorBit)),
-    MayInnerVectorize  = MightVectorize && int(InnerSize)!=Dynamic && int(InnerSize)%int(PacketSize)==0
-                       && int(DstIsAligned) && int(SrcIsAligned),
-    MayLinearVectorize = MightVectorize && (int(Derived::Flags) & int(OtherDerived::Flags) & LinearAccessBit),
-    MaySliceVectorize  = MightVectorize && 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 {
-    Vectorization = int(MayInnerVectorize)  ? int(InnerVectorization)
-                  : int(MayLinearVectorize) ? int(LinearVectorization)
-                  : int(MaySliceVectorize)  ? int(SliceVectorization)
-                                            : int(NoVectorization)
-  };
-
-private:
-  enum {
-    UnrollingLimit      = EIGEN_UNROLLING_LIMIT * (int(Vectorization) == int(NoVectorization) ? 1 : int(PacketSize)),
-    MayUnrollCompletely = int(Derived::SizeAtCompileTime) * int(OtherDerived::CoeffReadCost) <= int(UnrollingLimit),
-    MayUnrollInner      = int(InnerSize * OtherDerived::CoeffReadCost) <= int(UnrollingLimit)
-  };
-
-public:
-  enum {
-    Unrolling = (int(Vectorization) == int(InnerVectorization) || int(Vectorization) == int(NoVectorization))
-              ? (
-                   int(MayUnrollCompletely) ? int(CompleteUnrolling)
-                 : int(MayUnrollInner)      ? int(InnerUnrolling)
-                                            : int(NoUnrolling)
-                )
-              : int(Vectorization) == int(LinearVectorization)
-              ? ( int(MayUnrollCompletely) && int(DstIsAligned) ? int(CompleteUnrolling) : int(NoUnrolling) )
-              : int(NoUnrolling)
-  };
-};
-
-/***************************************************************************
-* Part 2 : meta-unrollers
-***************************************************************************/
-
-/***********************
-*** No vectorization ***
-***********************/
-
-template<typename Derived1, typename Derived2, int Index, int Stop>
-struct ei_assign_novec_CompleteUnrolling
-{
-  enum {
-    row = int(Derived1::Flags)&RowMajorBit
-        ? Index / int(Derived1::ColsAtCompileTime)
-        : Index % Derived1::RowsAtCompileTime,
-    col = int(Derived1::Flags)&RowMajorBit
-        ? Index % int(Derived1::ColsAtCompileTime)
-        : Index / Derived1::RowsAtCompileTime
-  };
-
-  EIGEN_STRONG_INLINE static void run(Derived1 &dst, const Derived2 &src)
-  {
-    dst.copyCoeff(row, col, src);
-    ei_assign_novec_CompleteUnrolling<Derived1, Derived2, Index+1, Stop>::run(dst, src);
-  }
-};
-
-template<typename Derived1, typename Derived2, int Stop>
-struct ei_assign_novec_CompleteUnrolling<Derived1, Derived2, Stop, Stop>
-{
-  EIGEN_STRONG_INLINE static void run(Derived1 &, const Derived2 &) {}
-};
-
-template<typename Derived1, typename Derived2, int Index, int Stop>
-struct ei_assign_novec_InnerUnrolling
-{
-  EIGEN_STRONG_INLINE static void run(Derived1 &dst, const Derived2 &src, int row_or_col)
-  {
-    const bool rowMajor = int(Derived1::Flags)&RowMajorBit;
-    const int row = rowMajor ? row_or_col : Index;
-    const int col = rowMajor ? Index : row_or_col;
-    dst.copyCoeff(row, col, src);
-    ei_assign_novec_InnerUnrolling<Derived1, Derived2, Index+1, Stop>::run(dst, src, row_or_col);
-  }
-};
-
-template<typename Derived1, typename Derived2, int Stop>
-struct ei_assign_novec_InnerUnrolling<Derived1, Derived2, Stop, Stop>
-{
-  EIGEN_STRONG_INLINE static void run(Derived1 &, const Derived2 &, int) {}
-};
-
-/**************************
-*** Inner vectorization ***
-**************************/
-
-template<typename Derived1, typename Derived2, int Index, int Stop>
-struct ei_assign_innervec_CompleteUnrolling
-{
-  enum {
-    row = int(Derived1::Flags)&RowMajorBit
-        ? Index / int(Derived1::ColsAtCompileTime)
-        : Index % Derived1::RowsAtCompileTime,
-    col = int(Derived1::Flags)&RowMajorBit
-        ? Index % int(Derived1::ColsAtCompileTime)
-        : Index / Derived1::RowsAtCompileTime,
-    SrcAlignment = ei_assign_traits<Derived1,Derived2>::SrcAlignment
-  };
-
-  EIGEN_STRONG_INLINE static void run(Derived1 &dst, const Derived2 &src)
-  {
-    dst.template copyPacket<Derived2, Aligned, SrcAlignment>(row, col, src);
-    ei_assign_innervec_CompleteUnrolling<Derived1, Derived2,
-      Index+ei_packet_traits<typename Derived1::Scalar>::size, Stop>::run(dst, src);
-  }
-};
-
-template<typename Derived1, typename Derived2, int Stop>
-struct ei_assign_innervec_CompleteUnrolling<Derived1, Derived2, Stop, Stop>
-{
-  EIGEN_STRONG_INLINE static void run(Derived1 &, const Derived2 &) {}
-};
-
-template<typename Derived1, typename Derived2, int Index, int Stop>
-struct ei_assign_innervec_InnerUnrolling
-{
-  EIGEN_STRONG_INLINE static void run(Derived1 &dst, const Derived2 &src, int row_or_col)
-  {
-    const int row = int(Derived1::Flags)&RowMajorBit ? row_or_col : Index;
-    const int col = int(Derived1::Flags)&RowMajorBit ? Index : row_or_col;
-    dst.template copyPacket<Derived2, Aligned, Aligned>(row, col, src);
-    ei_assign_innervec_InnerUnrolling<Derived1, Derived2,
-      Index+ei_packet_traits<typename Derived1::Scalar>::size, Stop>::run(dst, src, row_or_col);
-  }
-};
-
-template<typename Derived1, typename Derived2, int Stop>
-struct ei_assign_innervec_InnerUnrolling<Derived1, Derived2, Stop, Stop>
-{
-  EIGEN_STRONG_INLINE static void run(Derived1 &, const Derived2 &, int) {}
-};
-
-/***************************************************************************
-* Part 3 : implementation of all cases
-***************************************************************************/
-
-template<typename Derived1, typename Derived2,
-         int Vectorization = ei_assign_traits<Derived1, Derived2>::Vectorization,
-         int Unrolling = ei_assign_traits<Derived1, Derived2>::Unrolling>
-struct ei_assign_impl;
-
-/***********************
-*** No vectorization ***
-***********************/
-
-template<typename Derived1, typename Derived2>
-struct ei_assign_impl<Derived1, Derived2, NoVectorization, NoUnrolling>
-{
-  inline static void run(Derived1 &dst, const Derived2 &src)
-  {
-    const int innerSize = dst.innerSize();
-    const int outerSize = dst.outerSize();
-    for(int j = 0; j < outerSize; ++j)
-      for(int i = 0; i < innerSize; ++i)
-      {
-        if(int(Derived1::Flags)&RowMajorBit)
-          dst.copyCoeff(j, i, src);
-        else
-          dst.copyCoeff(i, j, src);
-      }
-  }
-};
-
-template<typename Derived1, typename Derived2>
-struct ei_assign_impl<Derived1, Derived2, NoVectorization, CompleteUnrolling>
-{
-  EIGEN_STRONG_INLINE static void run(Derived1 &dst, const Derived2 &src)
-  {
-    ei_assign_novec_CompleteUnrolling<Derived1, Derived2, 0, Derived1::SizeAtCompileTime>
-      ::run(dst, src);
-  }
-};
-
-template<typename Derived1, typename Derived2>
-struct ei_assign_impl<Derived1, Derived2, NoVectorization, InnerUnrolling>
-{
-  EIGEN_STRONG_INLINE static void run(Derived1 &dst, const Derived2 &src)
-  {
-    const bool rowMajor = int(Derived1::Flags)&RowMajorBit;
-    const int innerSize = rowMajor ? Derived1::ColsAtCompileTime : Derived1::RowsAtCompileTime;
-    const int outerSize = dst.outerSize();
-    for(int j = 0; j < outerSize; ++j)
-      ei_assign_novec_InnerUnrolling<Derived1, Derived2, 0, innerSize>
-        ::run(dst, src, j);
-  }
-};
-
-/**************************
-*** Inner vectorization ***
-**************************/
-
-template<typename Derived1, typename Derived2>
-struct ei_assign_impl<Derived1, Derived2, InnerVectorization, NoUnrolling>
-{
-  inline static void run(Derived1 &dst, const Derived2 &src)
-  {
-    const int innerSize = dst.innerSize();
-    const int outerSize = dst.outerSize();
-    const int packetSize = ei_packet_traits<typename Derived1::Scalar>::size;
-    for(int j = 0; j < outerSize; ++j)
-      for(int i = 0; i < innerSize; i+=packetSize)
-      {
-        if(int(Derived1::Flags)&RowMajorBit)
-          dst.template copyPacket<Derived2, Aligned, Aligned>(j, i, src);
-        else
-          dst.template copyPacket<Derived2, Aligned, Aligned>(i, j, src);
-      }
-  }
-};
-
-template<typename Derived1, typename Derived2>
-struct ei_assign_impl<Derived1, Derived2, InnerVectorization, CompleteUnrolling>
-{
-  EIGEN_STRONG_INLINE static void run(Derived1 &dst, const Derived2 &src)
-  {
-    ei_assign_innervec_CompleteUnrolling<Derived1, Derived2, 0, Derived1::SizeAtCompileTime>
-      ::run(dst, src);
-  }
-};
-
-template<typename Derived1, typename Derived2>
-struct ei_assign_impl<Derived1, Derived2, InnerVectorization, InnerUnrolling>
-{
-  EIGEN_STRONG_INLINE static void run(Derived1 &dst, const Derived2 &src)
-  {
-    const bool rowMajor = int(Derived1::Flags)&RowMajorBit;
-    const int innerSize = rowMajor ? Derived1::ColsAtCompileTime : Derived1::RowsAtCompileTime;
-    const int outerSize = dst.outerSize();
-    for(int j = 0; j < outerSize; ++j)
-      ei_assign_innervec_InnerUnrolling<Derived1, Derived2, 0, innerSize>
-        ::run(dst, src, j);
-  }
-};
-
-/***************************
-*** Linear vectorization ***
-***************************/
-
-template<typename Derived1, typename Derived2>
-struct ei_assign_impl<Derived1, Derived2, LinearVectorization, NoUnrolling>
-{
-  inline static void run(Derived1 &dst, const Derived2 &src)
-  {
-    const int size = dst.size();
-    const int packetSize = ei_packet_traits<typename Derived1::Scalar>::size;
-    const int alignedStart = ei_assign_traits<Derived1,Derived2>::DstIsAligned ? 0
-                           : ei_alignmentOffset(&dst.coeffRef(0), size);
-    const int alignedEnd = alignedStart + ((size-alignedStart)/packetSize)*packetSize;
-
-    for(int index = 0; index < alignedStart; ++index)
-      dst.copyCoeff(index, src);
-
-    for(int index = alignedStart; index < alignedEnd; index += packetSize)
-    {
-      dst.template copyPacket<Derived2, Aligned, ei_assign_traits<Derived1,Derived2>::SrcAlignment>(index, src);
-    }
-
-    for(int index = alignedEnd; index < size; ++index)
-      dst.copyCoeff(index, src);
-  }
-};
-
-template<typename Derived1, typename Derived2>
-struct ei_assign_impl<Derived1, Derived2, LinearVectorization, CompleteUnrolling>
-{
-  EIGEN_STRONG_INLINE static void run(Derived1 &dst, const Derived2 &src)
-  {
-    const int size = Derived1::SizeAtCompileTime;
-    const int packetSize = ei_packet_traits<typename Derived1::Scalar>::size;
-    const int alignedSize = (size/packetSize)*packetSize;
-
-    ei_assign_innervec_CompleteUnrolling<Derived1, Derived2, 0, alignedSize>::run(dst, src);
-    ei_assign_novec_CompleteUnrolling<Derived1, Derived2, alignedSize, size>::run(dst, src);
-  }
-};
-
-/**************************
-*** Slice vectorization ***
-***************************/
-
-template<typename Derived1, typename Derived2>
-struct ei_assign_impl<Derived1, Derived2, SliceVectorization, NoUnrolling>
-{
-  inline static void run(Derived1 &dst, const Derived2 &src)
-  {
-    const int packetSize = ei_packet_traits<typename Derived1::Scalar>::size;
-    const int packetAlignedMask = packetSize - 1;
-    const int innerSize = dst.innerSize();
-    const int outerSize = dst.outerSize();
-    const int alignedStep = (packetSize - dst.stride() % packetSize) & packetAlignedMask;
-    int alignedStart = ei_assign_traits<Derived1,Derived2>::DstIsAligned ? 0
-                     : ei_alignmentOffset(&dst.coeffRef(0,0), innerSize);
-
-    for(int i = 0; i < outerSize; ++i)
-    {
-      const int alignedEnd = alignedStart + ((innerSize-alignedStart) & ~packetAlignedMask);
-
-      // do the non-vectorizable part of the assignment
-      for (int index = 0; index<alignedStart ; ++index)
-      {
-        if(Derived1::Flags&RowMajorBit)
-          dst.copyCoeff(i, index, src);
-        else
-          dst.copyCoeff(index, i, src);
-      }
-
-      // do the vectorizable part of the assignment
-      for (int index = alignedStart; index<alignedEnd; index+=packetSize)
-      {
-        if(Derived1::Flags&RowMajorBit)
-          dst.template copyPacket<Derived2, Aligned, Unaligned>(i, index, src);
-        else
-          dst.template copyPacket<Derived2, Aligned, Unaligned>(index, i, src);
-      }
-
-      // do the non-vectorizable part of the assignment
-      for (int index = alignedEnd; index<innerSize ; ++index)
-      {
-        if(Derived1::Flags&RowMajorBit)
-          dst.copyCoeff(i, index, src);
-        else
-          dst.copyCoeff(index, i, src);
-      }
-
-      alignedStart = std::min<int>((alignedStart+alignedStep)%packetSize, innerSize);
-    }
-  }
-};
-
-/***************************************************************************
-* Part 4 : implementation of MatrixBase methods
-***************************************************************************/
+namespace Eigen {
 
 template<typename Derived>
 template<typename OtherDerived>
-EIGEN_STRONG_INLINE Derived& MatrixBase<Derived>
-  ::lazyAssign(const MatrixBase<OtherDerived>& other)
+EIGEN_STRONG_INLINE Derived& DenseBase<Derived>
+  ::lazyAssign(const DenseBase<OtherDerived>& other)
 {
+  enum{
+    SameType = internal::is_same<typename Derived::Scalar,typename OtherDerived::Scalar>::value
+  };
+
+  EIGEN_STATIC_ASSERT_LVALUE(Derived)
   EIGEN_STATIC_ASSERT_SAME_MATRIX_SIZE(Derived,OtherDerived)
-  EIGEN_STATIC_ASSERT((ei_is_same_type<typename Derived::Scalar, typename OtherDerived::Scalar>::ret),
-    YOU_MIXED_DIFFERENT_NUMERIC_TYPES__YOU_NEED_TO_USE_THE_CAST_METHOD_OF_MATRIXBASE_TO_CAST_NUMERIC_TYPES_EXPLICITLY)
-  ei_assert(rows() == other.rows() && cols() == other.cols());
-  ei_assign_impl<Derived, OtherDerived>::run(derived(),other.derived());
+  EIGEN_STATIC_ASSERT(SameType,YOU_MIXED_DIFFERENT_NUMERIC_TYPES__YOU_NEED_TO_USE_THE_CAST_METHOD_OF_MATRIXBASE_TO_CAST_NUMERIC_TYPES_EXPLICITLY)
+
+  eigen_assert(rows() == other.rows() && cols() == other.cols());
+  internal::call_assignment_no_alias(derived(),other.derived());
+  
   return derived();
 }
 
-template<typename Derived, typename OtherDerived,
-         bool EvalBeforeAssigning = (int(OtherDerived::Flags) & EvalBeforeAssigningBit) != 0,
-         bool NeedToTranspose = Derived::IsVectorAtCompileTime
-                && OtherDerived::IsVectorAtCompileTime
-                && int(Derived::RowsAtCompileTime) == int(OtherDerived::ColsAtCompileTime)
-                && int(Derived::ColsAtCompileTime) == int(OtherDerived::RowsAtCompileTime)
-                && int(Derived::SizeAtCompileTime) != 1>
-struct ei_assign_selector;
+template<typename Derived>
+template<typename OtherDerived>
+EIGEN_DEVICE_FUNC
+EIGEN_STRONG_INLINE Derived& DenseBase<Derived>::operator=(const DenseBase<OtherDerived>& other)
+{
+  internal::call_assignment(derived(), other.derived());
+  return derived();
+}
 
-template<typename Derived, typename OtherDerived>
-struct ei_assign_selector<Derived,OtherDerived,false,false> {
-  EIGEN_STRONG_INLINE static Derived& run(Derived& dst, const OtherDerived& other) { return dst.lazyAssign(other.derived()); }
-};
-template<typename Derived, typename OtherDerived>
-struct ei_assign_selector<Derived,OtherDerived,true,false> {
-  EIGEN_STRONG_INLINE static Derived& run(Derived& dst, const OtherDerived& other) { return dst.lazyAssign(other.eval()); }
-};
-template<typename Derived, typename OtherDerived>
-struct ei_assign_selector<Derived,OtherDerived,false,true> {
-  EIGEN_STRONG_INLINE static Derived& run(Derived& dst, const OtherDerived& other) { return dst.lazyAssign(other.transpose()); }
-};
-template<typename Derived, typename OtherDerived>
-struct ei_assign_selector<Derived,OtherDerived,true,true> {
-  EIGEN_STRONG_INLINE static Derived& run(Derived& dst, const OtherDerived& other) { return dst.lazyAssign(other.transpose().eval()); }
-};
+template<typename Derived>
+EIGEN_DEVICE_FUNC
+EIGEN_STRONG_INLINE Derived& DenseBase<Derived>::operator=(const DenseBase& other)
+{
+  internal::call_assignment(derived(), other.derived());
+  return derived();
+}
+
+template<typename Derived>
+EIGEN_DEVICE_FUNC
+EIGEN_STRONG_INLINE Derived& MatrixBase<Derived>::operator=(const MatrixBase& other)
+{
+  internal::call_assignment(derived(), other.derived());
+  return derived();
+}
+
+template<typename Derived>
+template <typename OtherDerived>
+EIGEN_DEVICE_FUNC
+EIGEN_STRONG_INLINE Derived& MatrixBase<Derived>::operator=(const DenseBase<OtherDerived>& other)
+{
+  internal::call_assignment(derived(), other.derived());
+  return derived();
+}
+
+template<typename Derived>
+template <typename OtherDerived>
+EIGEN_DEVICE_FUNC
+EIGEN_STRONG_INLINE Derived& MatrixBase<Derived>::operator=(const EigenBase<OtherDerived>& other)
+{
+  internal::call_assignment(derived(), other.derived());
+  return derived();
+}
 
 template<typename Derived>
 template<typename OtherDerived>
-EIGEN_STRONG_INLINE Derived& MatrixBase<Derived>
-  ::operator=(const MatrixBase<OtherDerived>& other)
+EIGEN_DEVICE_FUNC
+EIGEN_STRONG_INLINE Derived& MatrixBase<Derived>::operator=(const ReturnByValue<OtherDerived>& other)
 {
-  return ei_assign_selector<Derived,OtherDerived>::run(derived(), other.derived());
+  other.derived().evalTo(derived());
+  return derived();
 }
 
+} // end namespace Eigen
+
 #endif // EIGEN_ASSIGN_H

Eigen/src/Core/AssignEvaluator.h

@@ -0,0 +1,935 @@
+// 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) 2011-2014 Gael Guennebaud <gael.guennebaud@inria.fr>
+// Copyright (C) 2011-2012 Jitse Niesen <jitse@maths.leeds.ac.uk>
+//
+// This Source Code Form is subject to the terms of the Mozilla
+// Public License v. 2.0. If a copy of the MPL was not distributed
+// with this file, You can obtain one at http://mozilla.org/MPL/2.0/.
+
+#ifndef EIGEN_ASSIGN_EVALUATOR_H
+#define EIGEN_ASSIGN_EVALUATOR_H
+
+namespace Eigen {
+
+// 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
+
+template <typename DstEvaluator, typename SrcEvaluator, typename AssignFunc>
+struct copy_using_evaluator_traits
+{
+  typedef typename DstEvaluator::XprType Dst;
+  typedef typename Dst::Scalar DstScalar;
+  
+  enum {
+    DstFlags = DstEvaluator::Flags,
+    SrcFlags = SrcEvaluator::Flags
+  };
+  
+public:
+  enum {
+    DstAlignment = DstEvaluator::Alignment,
+    SrcAlignment = SrcEvaluator::Alignment,
+    DstHasDirectAccess = (DstFlags & DirectAccessBit) == DirectAccessBit,
+    JointAlignment = EIGEN_PLAIN_ENUM_MIN(DstAlignment,SrcAlignment)
+  };
+
+private:
+  enum {
+    InnerSize = int(Dst::IsVectorAtCompileTime) ? int(Dst::SizeAtCompileTime)
+              : int(DstFlags)&RowMajorBit ? int(Dst::ColsAtCompileTime)
+              : int(Dst::RowsAtCompileTime),
+    InnerMaxSize = int(Dst::IsVectorAtCompileTime) ? int(Dst::MaxSizeAtCompileTime)
+              : int(DstFlags)&RowMajorBit ? int(Dst::MaxColsAtCompileTime)
+              : int(Dst::MaxRowsAtCompileTime),
+    OuterStride = int(outer_stride_at_compile_time<Dst>::ret),
+    MaxSizeAtCompileTime = Dst::SizeAtCompileTime
+  };
+
+  // TODO distinguish between linear traversal and inner-traversals
+  typedef typename find_best_packet<DstScalar,Dst::SizeAtCompileTime>::type LinearPacketType;
+  typedef typename find_best_packet<DstScalar,InnerSize>::type InnerPacketType;
+
+  enum {
+    LinearPacketSize = unpacket_traits<LinearPacketType>::size,
+    InnerPacketSize = unpacket_traits<InnerPacketType>::size
+  };
+
+public:
+  enum {
+    LinearRequiredAlignment = unpacket_traits<LinearPacketType>::alignment,
+    InnerRequiredAlignment = unpacket_traits<InnerPacketType>::alignment
+  };
+
+private:
+  enum {
+    DstIsRowMajor = DstFlags&RowMajorBit,
+    SrcIsRowMajor = SrcFlags&RowMajorBit,
+    StorageOrdersAgree = (int(DstIsRowMajor) == int(SrcIsRowMajor)),
+    MightVectorize = bool(StorageOrdersAgree)
+                  && (int(DstFlags) & int(SrcFlags) & ActualPacketAccessBit)
+                  && bool(functor_traits<AssignFunc>::PacketAccess),
+    MayInnerVectorize  = MightVectorize
+                       && int(InnerSize)!=Dynamic && int(InnerSize)%int(InnerPacketSize)==0
+                       && int(OuterStride)!=Dynamic && int(OuterStride)%int(InnerPacketSize)==0
+                       && (EIGEN_UNALIGNED_VECTORIZE  || int(JointAlignment)>=int(InnerRequiredAlignment)),
+    MayLinearize = bool(StorageOrdersAgree) && (int(DstFlags) & int(SrcFlags) & LinearAccessBit),
+    MayLinearVectorize = bool(MightVectorize) && bool(MayLinearize) && bool(DstHasDirectAccess)
+                       && (EIGEN_UNALIGNED_VECTORIZE || (int(DstAlignment)>=int(LinearRequiredAlignment)) || 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  = bool(MightVectorize) && bool(DstHasDirectAccess)
+                       && (int(InnerMaxSize)==Dynamic || int(InnerMaxSize)>=(EIGEN_UNALIGNED_VECTORIZE?InnerPacketSize:(3*InnerPacketSize)))
+      /* 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
+         However, with EIGEN_UNALIGNED_VECTORIZE and unrolling, slice vectorization is still worth it */
+  };
+
+public:
+  enum {
+    Traversal = int(MayLinearVectorize) && (LinearPacketSize>InnerPacketSize) ? int(LinearVectorizedTraversal)
+              : 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
+  };
+
+  typedef typename conditional<int(Traversal)==LinearVectorizedTraversal, LinearPacketType, InnerPacketType>::type PacketType;
+
+private:
+  enum {
+    ActualPacketSize    = int(Traversal)==LinearVectorizedTraversal ? LinearPacketSize
+                        : Vectorized ? InnerPacketSize
+                        : 1,
+    UnrollingLimit      = EIGEN_UNROLLING_LIMIT * ActualPacketSize,
+    MayUnrollCompletely = int(Dst::SizeAtCompileTime) != Dynamic
+                       && int(Dst::SizeAtCompileTime) * (int(DstEvaluator::CoeffReadCost)+int(SrcEvaluator::CoeffReadCost)) <= int(UnrollingLimit),
+    MayUnrollInner      = int(InnerSize) != Dynamic
+                       && int(InnerSize) * (int(DstEvaluator::CoeffReadCost)+int(SrcEvaluator::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) && ( EIGEN_UNALIGNED_VECTORIZE || (int(DstAlignment)>=int(LinearRequiredAlignment)))
+                          ? int(CompleteUnrolling)
+                          : int(NoUnrolling) )
+              : int(Traversal) == int(LinearTraversal)
+                ? ( bool(MayUnrollCompletely) ? int(CompleteUnrolling) 
+                                              : int(NoUnrolling) )
+#if EIGEN_UNALIGNED_VECTORIZE
+              : int(Traversal) == int(SliceVectorizedTraversal)
+                ? ( bool(MayUnrollInner) ? int(InnerUnrolling)
+                                         : int(NoUnrolling) )
+#endif
+              : int(NoUnrolling)
+  };
+
+#ifdef EIGEN_DEBUG_ASSIGN
+  static void debug()
+  {
+    std::cerr << "DstXpr: " << typeid(typename DstEvaluator::XprType).name() << std::endl;
+    std::cerr << "SrcXpr: " << typeid(typename SrcEvaluator::XprType).name() << std::endl;
+    std::cerr.setf(std::ios::hex, std::ios::basefield);
+    std::cerr << "DstFlags" << " = " << DstFlags << " (" << demangle_flags(DstFlags) << " )" << std::endl;
+    std::cerr << "SrcFlags" << " = " << SrcFlags << " (" << demangle_flags(SrcFlags) << " )" << std::endl;
+    std::cerr.unsetf(std::ios::hex);
+    EIGEN_DEBUG_VAR(DstAlignment)
+    EIGEN_DEBUG_VAR(SrcAlignment)
+    EIGEN_DEBUG_VAR(LinearRequiredAlignment)
+    EIGEN_DEBUG_VAR(InnerRequiredAlignment)
+    EIGEN_DEBUG_VAR(JointAlignment)
+    EIGEN_DEBUG_VAR(InnerSize)
+    EIGEN_DEBUG_VAR(InnerMaxSize)
+    EIGEN_DEBUG_VAR(LinearPacketSize)
+    EIGEN_DEBUG_VAR(InnerPacketSize)
+    EIGEN_DEBUG_VAR(ActualPacketSize)
+    EIGEN_DEBUG_VAR(StorageOrdersAgree)
+    EIGEN_DEBUG_VAR(MightVectorize)
+    EIGEN_DEBUG_VAR(MayLinearize)
+    EIGEN_DEBUG_VAR(MayInnerVectorize)
+    EIGEN_DEBUG_VAR(MayLinearVectorize)
+    EIGEN_DEBUG_VAR(MaySliceVectorize)
+    std::cerr << "Traversal" << " = " << Traversal << " (" << demangle_traversal(Traversal) << ")" << std::endl;
+    EIGEN_DEBUG_VAR(SrcEvaluator::CoeffReadCost)
+    EIGEN_DEBUG_VAR(UnrollingLimit)
+    EIGEN_DEBUG_VAR(MayUnrollCompletely)
+    EIGEN_DEBUG_VAR(MayUnrollInner)
+    std::cerr << "Unrolling" << " = " << Unrolling << " (" << demangle_unrolling(Unrolling) << ")" << std::endl;
+    std::cerr << std::endl;
+  }
+#endif
+};
+
+/***************************************************************************
+* Part 2 : meta-unrollers
+***************************************************************************/
+
+/************************
+*** Default traversal ***
+************************/
+
+template<typename Kernel, int Index, int Stop>
+struct copy_using_evaluator_DefaultTraversal_CompleteUnrolling
+{
+  // FIXME: this is not very clean, perhaps this information should be provided by the kernel?
+  typedef typename Kernel::DstEvaluatorType DstEvaluatorType;
+  typedef typename DstEvaluatorType::XprType DstXprType;
+  
+  enum {
+    outer = Index / DstXprType::InnerSizeAtCompileTime,
+    inner = Index % DstXprType::InnerSizeAtCompileTime
+  };
+
+  EIGEN_DEVICE_FUNC static EIGEN_STRONG_INLINE void run(Kernel &kernel)
+  {
+    kernel.assignCoeffByOuterInner(outer, inner);
+    copy_using_evaluator_DefaultTraversal_CompleteUnrolling<Kernel, Index+1, Stop>::run(kernel);
+  }
+};
+
+template<typename Kernel, int Stop>
+struct copy_using_evaluator_DefaultTraversal_CompleteUnrolling<Kernel, Stop, Stop>
+{
+  EIGEN_DEVICE_FUNC static EIGEN_STRONG_INLINE void run(Kernel&) { }
+};
+
+template<typename Kernel, int Index_, int Stop>
+struct copy_using_evaluator_DefaultTraversal_InnerUnrolling
+{
+  EIGEN_DEVICE_FUNC static EIGEN_STRONG_INLINE void run(Kernel &kernel, Index outer)
+  {
+    kernel.assignCoeffByOuterInner(outer, Index_);
+    copy_using_evaluator_DefaultTraversal_InnerUnrolling<Kernel, Index_+1, Stop>::run(kernel, outer);
+  }
+};
+
+template<typename Kernel, int Stop>
+struct copy_using_evaluator_DefaultTraversal_InnerUnrolling<Kernel, Stop, Stop>
+{
+  EIGEN_DEVICE_FUNC static EIGEN_STRONG_INLINE void run(Kernel&, Index) { }
+};
+
+/***********************
+*** Linear traversal ***
+***********************/
+
+template<typename Kernel, int Index, int Stop>
+struct copy_using_evaluator_LinearTraversal_CompleteUnrolling
+{
+  EIGEN_DEVICE_FUNC static EIGEN_STRONG_INLINE void run(Kernel& kernel)
+  {
+    kernel.assignCoeff(Index);
+    copy_using_evaluator_LinearTraversal_CompleteUnrolling<Kernel, Index+1, Stop>::run(kernel);
+  }
+};
+
+template<typename Kernel, int Stop>
+struct copy_using_evaluator_LinearTraversal_CompleteUnrolling<Kernel, Stop, Stop>
+{
+  EIGEN_DEVICE_FUNC static EIGEN_STRONG_INLINE void run(Kernel&) { }
+};
+
+/**************************
+*** Inner vectorization ***
+**************************/
+
+template<typename Kernel, int Index, int Stop>
+struct copy_using_evaluator_innervec_CompleteUnrolling
+{
+  // FIXME: this is not very clean, perhaps this information should be provided by the kernel?
+  typedef typename Kernel::DstEvaluatorType DstEvaluatorType;
+  typedef typename DstEvaluatorType::XprType DstXprType;
+  typedef typename Kernel::PacketType PacketType;
+  
+  enum {
+    outer = Index / DstXprType::InnerSizeAtCompileTime,
+    inner = Index % DstXprType::InnerSizeAtCompileTime,
+    SrcAlignment = Kernel::AssignmentTraits::SrcAlignment,
+    DstAlignment = Kernel::AssignmentTraits::DstAlignment
+  };
+
+  EIGEN_DEVICE_FUNC static EIGEN_STRONG_INLINE void run(Kernel &kernel)
+  {
+    kernel.template assignPacketByOuterInner<DstAlignment, SrcAlignment, PacketType>(outer, inner);
+    enum { NextIndex = Index + unpacket_traits<PacketType>::size };
+    copy_using_evaluator_innervec_CompleteUnrolling<Kernel, NextIndex, Stop>::run(kernel);
+  }
+};
+
+template<typename Kernel, int Stop>
+struct copy_using_evaluator_innervec_CompleteUnrolling<Kernel, Stop, Stop>
+{
+  EIGEN_DEVICE_FUNC static EIGEN_STRONG_INLINE void run(Kernel&) { }
+};
+
+template<typename Kernel, int Index_, int Stop, int SrcAlignment, int DstAlignment>
+struct copy_using_evaluator_innervec_InnerUnrolling
+{
+  typedef typename Kernel::PacketType PacketType;
+  EIGEN_DEVICE_FUNC static EIGEN_STRONG_INLINE void run(Kernel &kernel, Index outer)
+  {
+    kernel.template assignPacketByOuterInner<DstAlignment, SrcAlignment, PacketType>(outer, Index_);
+    enum { NextIndex = Index_ + unpacket_traits<PacketType>::size };
+    copy_using_evaluator_innervec_InnerUnrolling<Kernel, NextIndex, Stop, SrcAlignment, DstAlignment>::run(kernel, outer);
+  }
+};
+
+template<typename Kernel, int Stop, int SrcAlignment, int DstAlignment>
+struct copy_using_evaluator_innervec_InnerUnrolling<Kernel, Stop, Stop, SrcAlignment, DstAlignment>
+{
+  EIGEN_DEVICE_FUNC static EIGEN_STRONG_INLINE void run(Kernel &, Index) { }
+};
+
+/***************************************************************************
+* Part 3 : implementation of all cases
+***************************************************************************/
+
+// dense_assignment_loop is based on assign_impl
+
+template<typename Kernel,
+         int Traversal = Kernel::AssignmentTraits::Traversal,
+         int Unrolling = Kernel::AssignmentTraits::Unrolling>
+struct dense_assignment_loop;
+
+/************************
+*** Default traversal ***
+************************/
+
+template<typename Kernel>
+struct dense_assignment_loop<Kernel, DefaultTraversal, NoUnrolling>
+{
+  EIGEN_DEVICE_FUNC static void EIGEN_STRONG_INLINE run(Kernel &kernel)
+  {
+    for(Index outer = 0; outer < kernel.outerSize(); ++outer) {
+      for(Index inner = 0; inner < kernel.innerSize(); ++inner) {
+        kernel.assignCoeffByOuterInner(outer, inner);
+      }
+    }
+  }
+};
+
+template<typename Kernel>
+struct dense_assignment_loop<Kernel, DefaultTraversal, CompleteUnrolling>
+{
+  EIGEN_DEVICE_FUNC static EIGEN_STRONG_INLINE void run(Kernel &kernel)
+  {
+    typedef typename Kernel::DstEvaluatorType::XprType DstXprType;
+    copy_using_evaluator_DefaultTraversal_CompleteUnrolling<Kernel, 0, DstXprType::SizeAtCompileTime>::run(kernel);
+  }
+};
+
+template<typename Kernel>
+struct dense_assignment_loop<Kernel, DefaultTraversal, InnerUnrolling>
+{
+  EIGEN_DEVICE_FUNC static EIGEN_STRONG_INLINE void run(Kernel &kernel)
+  {
+    typedef typename Kernel::DstEvaluatorType::XprType DstXprType;
+
+    const Index outerSize = kernel.outerSize();
+    for(Index outer = 0; outer < outerSize; ++outer)
+      copy_using_evaluator_DefaultTraversal_InnerUnrolling<Kernel, 0, DstXprType::InnerSizeAtCompileTime>::run(kernel, outer);
+  }
+};
+
+/***************************
+*** Linear vectorization ***
+***************************/
+
+
+// The goal of unaligned_dense_assignment_loop is simply to factorize the handling
+// of the non vectorizable beginning and ending parts
+
+template <bool IsAligned = false>
+struct unaligned_dense_assignment_loop
+{
+  // if IsAligned = true, then do nothing
+  template <typename Kernel>
+  EIGEN_DEVICE_FUNC static EIGEN_STRONG_INLINE void run(Kernel&, Index, Index) {}
+};
+
+template <>
+struct unaligned_dense_assignment_loop<false>
+{
+  // MSVC must not inline this functions. If it does, it fails to optimize the
+  // packet access path.
+  // FIXME check which version exhibits this issue
+#if EIGEN_COMP_MSVC
+  template <typename Kernel>
+  static EIGEN_DONT_INLINE void run(Kernel &kernel,
+                                    Index start,
+                                    Index end)
+#else
+  template <typename Kernel>
+  EIGEN_DEVICE_FUNC static EIGEN_STRONG_INLINE void run(Kernel &kernel,
+                                      Index start,
+                                      Index end)
+#endif
+  {
+    for (Index index = start; index < end; ++index)
+      kernel.assignCoeff(index);
+  }
+};
+
+template<typename Kernel>
+struct dense_assignment_loop<Kernel, LinearVectorizedTraversal, NoUnrolling>
+{
+  EIGEN_DEVICE_FUNC static EIGEN_STRONG_INLINE void run(Kernel &kernel)
+  {
+    const Index size = kernel.size();
+    typedef typename Kernel::Scalar Scalar;
+    typedef typename Kernel::PacketType PacketType;
+    enum {
+      requestedAlignment = Kernel::AssignmentTraits::LinearRequiredAlignment,
+      packetSize = unpacket_traits<PacketType>::size,
+      dstIsAligned = int(Kernel::AssignmentTraits::DstAlignment)>=int(requestedAlignment),
+      dstAlignment = packet_traits<Scalar>::AlignedOnScalar ? int(requestedAlignment)
+                                                            : int(Kernel::AssignmentTraits::DstAlignment),
+      srcAlignment = Kernel::AssignmentTraits::JointAlignment
+    };
+    const Index alignedStart = dstIsAligned ? 0 : internal::first_aligned<requestedAlignment>(kernel.dstDataPtr(), size);
+    const Index alignedEnd = alignedStart + ((size-alignedStart)/packetSize)*packetSize;
+
+    unaligned_dense_assignment_loop<dstIsAligned!=0>::run(kernel, 0, alignedStart);
+
+    for(Index index = alignedStart; index < alignedEnd; index += packetSize)
+      kernel.template assignPacket<dstAlignment, srcAlignment, PacketType>(index);
+
+    unaligned_dense_assignment_loop<>::run(kernel, alignedEnd, size);
+  }
+};
+
+template<typename Kernel>
+struct dense_assignment_loop<Kernel, LinearVectorizedTraversal, CompleteUnrolling>
+{
+  EIGEN_DEVICE_FUNC static EIGEN_STRONG_INLINE void run(Kernel &kernel)
+  {
+    typedef typename Kernel::DstEvaluatorType::XprType DstXprType;
+    typedef typename Kernel::PacketType PacketType;
+    
+    enum { size = DstXprType::SizeAtCompileTime,
+           packetSize =unpacket_traits<PacketType>::size,
+           alignedSize = (size/packetSize)*packetSize };
+
+    copy_using_evaluator_innervec_CompleteUnrolling<Kernel, 0, alignedSize>::run(kernel);
+    copy_using_evaluator_DefaultTraversal_CompleteUnrolling<Kernel, alignedSize, size>::run(kernel);
+  }
+};
+
+/**************************
+*** Inner vectorization ***
+**************************/
+
+template<typename Kernel>
+struct dense_assignment_loop<Kernel, InnerVectorizedTraversal, NoUnrolling>
+{
+  typedef typename Kernel::PacketType PacketType;
+  enum {
+    SrcAlignment = Kernel::AssignmentTraits::SrcAlignment,
+    DstAlignment = Kernel::AssignmentTraits::DstAlignment
+  };
+  EIGEN_DEVICE_FUNC static EIGEN_STRONG_INLINE void run(Kernel &kernel)
+  {
+    const Index innerSize = kernel.innerSize();
+    const Index outerSize = kernel.outerSize();
+    const Index packetSize = unpacket_traits<PacketType>::size;
+    for(Index outer = 0; outer < outerSize; ++outer)
+      for(Index inner = 0; inner < innerSize; inner+=packetSize)
+        kernel.template assignPacketByOuterInner<DstAlignment, SrcAlignment, PacketType>(outer, inner);
+  }
+};
+
+template<typename Kernel>
+struct dense_assignment_loop<Kernel, InnerVectorizedTraversal, CompleteUnrolling>
+{
+  EIGEN_DEVICE_FUNC static EIGEN_STRONG_INLINE void run(Kernel &kernel)
+  {
+    typedef typename Kernel::DstEvaluatorType::XprType DstXprType;
+    copy_using_evaluator_innervec_CompleteUnrolling<Kernel, 0, DstXprType::SizeAtCompileTime>::run(kernel);
+  }
+};
+
+template<typename Kernel>
+struct dense_assignment_loop<Kernel, InnerVectorizedTraversal, InnerUnrolling>
+{
+  EIGEN_DEVICE_FUNC static EIGEN_STRONG_INLINE void run(Kernel &kernel)
+  {
+    typedef typename Kernel::DstEvaluatorType::XprType DstXprType;
+    typedef typename Kernel::AssignmentTraits Traits;
+    const Index outerSize = kernel.outerSize();
+    for(Index outer = 0; outer < outerSize; ++outer)
+      copy_using_evaluator_innervec_InnerUnrolling<Kernel, 0, DstXprType::InnerSizeAtCompileTime,
+                                                   Traits::SrcAlignment, Traits::DstAlignment>::run(kernel, outer);
+  }
+};
+
+/***********************
+*** Linear traversal ***
+***********************/
+
+template<typename Kernel>
+struct dense_assignment_loop<Kernel, LinearTraversal, NoUnrolling>
+{
+  EIGEN_DEVICE_FUNC static EIGEN_STRONG_INLINE void run(Kernel &kernel)
+  {
+    const Index size = kernel.size();
+    for(Index i = 0; i < size; ++i)
+      kernel.assignCoeff(i);
+  }
+};
+
+template<typename Kernel>
+struct dense_assignment_loop<Kernel, LinearTraversal, CompleteUnrolling>
+{
+  EIGEN_DEVICE_FUNC static EIGEN_STRONG_INLINE void run(Kernel &kernel)
+  {
+    typedef typename Kernel::DstEvaluatorType::XprType DstXprType;
+    copy_using_evaluator_LinearTraversal_CompleteUnrolling<Kernel, 0, DstXprType::SizeAtCompileTime>::run(kernel);
+  }
+};
+
+/**************************
+*** Slice vectorization ***
+***************************/
+
+template<typename Kernel>
+struct dense_assignment_loop<Kernel, SliceVectorizedTraversal, NoUnrolling>
+{
+  EIGEN_DEVICE_FUNC static EIGEN_STRONG_INLINE void run(Kernel &kernel)
+  {
+    typedef typename Kernel::Scalar Scalar;
+    typedef typename Kernel::PacketType PacketType;
+    enum {
+      packetSize = unpacket_traits<PacketType>::size,
+      requestedAlignment = int(Kernel::AssignmentTraits::InnerRequiredAlignment),
+      alignable = packet_traits<Scalar>::AlignedOnScalar || int(Kernel::AssignmentTraits::DstAlignment)>=sizeof(Scalar),
+      dstIsAligned = int(Kernel::AssignmentTraits::DstAlignment)>=int(requestedAlignment),
+      dstAlignment = alignable ? int(requestedAlignment)
+                               : int(Kernel::AssignmentTraits::DstAlignment)
+    };
+    const Scalar *dst_ptr = kernel.dstDataPtr();
+    if((!bool(dstIsAligned)) && (UIntPtr(dst_ptr) % sizeof(Scalar))>0)
+    {
+      // the pointer is not aligend-on scalar, so alignment is not possible
+      return dense_assignment_loop<Kernel,DefaultTraversal,NoUnrolling>::run(kernel);
+    }
+    const Index packetAlignedMask = packetSize - 1;
+    const Index innerSize = kernel.innerSize();
+    const Index outerSize = kernel.outerSize();
+    const Index alignedStep = alignable ? (packetSize - kernel.outerStride() % packetSize) & packetAlignedMask : 0;
+    Index alignedStart = ((!alignable) || bool(dstIsAligned)) ? 0 : internal::first_aligned<requestedAlignment>(dst_ptr, 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)
+        kernel.assignCoeffByOuterInner(outer, inner);
+
+      // do the vectorizable part of the assignment
+      for(Index inner = alignedStart; inner<alignedEnd; inner+=packetSize)
+        kernel.template assignPacketByOuterInner<dstAlignment, Unaligned, PacketType>(outer, inner);
+
+      // do the non-vectorizable part of the assignment
+      for(Index inner = alignedEnd; inner<innerSize ; ++inner)
+        kernel.assignCoeffByOuterInner(outer, inner);
+
+      alignedStart = numext::mini((alignedStart+alignedStep)%packetSize, innerSize);
+    }
+  }
+};
+
+#if EIGEN_UNALIGNED_VECTORIZE
+template<typename Kernel>
+struct dense_assignment_loop<Kernel, SliceVectorizedTraversal, InnerUnrolling>
+{
+  EIGEN_DEVICE_FUNC static EIGEN_STRONG_INLINE void run(Kernel &kernel)
+  {
+    typedef typename Kernel::DstEvaluatorType::XprType DstXprType;
+    typedef typename Kernel::PacketType PacketType;
+
+    enum { size = DstXprType::InnerSizeAtCompileTime,
+           packetSize =unpacket_traits<PacketType>::size,
+           vectorizableSize = (size/packetSize)*packetSize };
+
+    for(Index outer = 0; outer < kernel.outerSize(); ++outer)
+    {
+      copy_using_evaluator_innervec_InnerUnrolling<Kernel, 0, vectorizableSize, 0, 0>::run(kernel, outer);
+      copy_using_evaluator_DefaultTraversal_InnerUnrolling<Kernel, vectorizableSize, size>::run(kernel, outer);
+    }
+  }
+};
+#endif
+
+
+/***************************************************************************
+* Part 4 : Generic dense assignment kernel
+***************************************************************************/
+
+// This class generalize the assignment of a coefficient (or packet) from one dense evaluator
+// to another dense writable evaluator.
+// It is parametrized by the two evaluators, and the actual assignment functor.
+// This abstraction level permits to keep the evaluation loops as simple and as generic as possible.
+// One can customize the assignment using this generic dense_assignment_kernel with different
+// functors, or by completely overloading it, by-passing a functor.
+template<typename DstEvaluatorTypeT, typename SrcEvaluatorTypeT, typename Functor, int Version = Specialized>
+class generic_dense_assignment_kernel
+{
+protected:
+  typedef typename DstEvaluatorTypeT::XprType DstXprType;
+  typedef typename SrcEvaluatorTypeT::XprType SrcXprType;
+public:
+  
+  typedef DstEvaluatorTypeT DstEvaluatorType;
+  typedef SrcEvaluatorTypeT SrcEvaluatorType;
+  typedef typename DstEvaluatorType::Scalar Scalar;
+  typedef copy_using_evaluator_traits<DstEvaluatorTypeT, SrcEvaluatorTypeT, Functor> AssignmentTraits;
+  typedef typename AssignmentTraits::PacketType PacketType;
+  
+  
+  EIGEN_DEVICE_FUNC generic_dense_assignment_kernel(DstEvaluatorType &dst, const SrcEvaluatorType &src, const Functor &func, DstXprType& dstExpr)
+    : m_dst(dst), m_src(src), m_functor(func), m_dstExpr(dstExpr)
+  {
+    #ifdef EIGEN_DEBUG_ASSIGN
+    AssignmentTraits::debug();
+    #endif
+  }
+  
+  EIGEN_DEVICE_FUNC Index size() const        { return m_dstExpr.size(); }
+  EIGEN_DEVICE_FUNC Index innerSize() const   { return m_dstExpr.innerSize(); }
+  EIGEN_DEVICE_FUNC Index outerSize() const   { return m_dstExpr.outerSize(); }
+  EIGEN_DEVICE_FUNC Index rows() const        { return m_dstExpr.rows(); }
+  EIGEN_DEVICE_FUNC Index cols() const        { return m_dstExpr.cols(); }
+  EIGEN_DEVICE_FUNC Index outerStride() const { return m_dstExpr.outerStride(); }
+  
+  EIGEN_DEVICE_FUNC DstEvaluatorType& dstEvaluator() { return m_dst; }
+  EIGEN_DEVICE_FUNC const SrcEvaluatorType& srcEvaluator() const { return m_src; }
+  
+  /// Assign src(row,col) to dst(row,col) through the assignment functor.
+  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE void assignCoeff(Index row, Index col)
+  {
+    m_functor.assignCoeff(m_dst.coeffRef(row,col), m_src.coeff(row,col));
+  }
+  
+  /// \sa assignCoeff(Index,Index)
+  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE void assignCoeff(Index index)
+  {
+    m_functor.assignCoeff(m_dst.coeffRef(index), m_src.coeff(index));
+  }
+  
+  /// \sa assignCoeff(Index,Index)
+  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE void assignCoeffByOuterInner(Index outer, Index inner)
+  {
+    Index row = rowIndexByOuterInner(outer, inner); 
+    Index col = colIndexByOuterInner(outer, inner); 
+    assignCoeff(row, col);
+  }
+  
+  
+  template<int StoreMode, int LoadMode, typename PacketType>
+  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE void assignPacket(Index row, Index col)
+  {
+    m_functor.template assignPacket<StoreMode>(&m_dst.coeffRef(row,col), m_src.template packet<LoadMode,PacketType>(row,col));
+  }
+  
+  template<int StoreMode, int LoadMode, typename PacketType>
+  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE void assignPacket(Index index)
+  {
+    m_functor.template assignPacket<StoreMode>(&m_dst.coeffRef(index), m_src.template packet<LoadMode,PacketType>(index));
+  }
+  
+  template<int StoreMode, int LoadMode, typename PacketType>
+  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE void assignPacketByOuterInner(Index outer, Index inner)
+  {
+    Index row = rowIndexByOuterInner(outer, inner); 
+    Index col = colIndexByOuterInner(outer, inner);
+    assignPacket<StoreMode,LoadMode,PacketType>(row, col);
+  }
+  
+  EIGEN_DEVICE_FUNC static EIGEN_STRONG_INLINE Index rowIndexByOuterInner(Index outer, Index inner)
+  {
+    typedef typename DstEvaluatorType::ExpressionTraits Traits;
+    return int(Traits::RowsAtCompileTime) == 1 ? 0
+      : int(Traits::ColsAtCompileTime) == 1 ? inner
+      : int(DstEvaluatorType::Flags)&RowMajorBit ? outer
+      : inner;
+  }
+
+  EIGEN_DEVICE_FUNC static EIGEN_STRONG_INLINE Index colIndexByOuterInner(Index outer, Index inner)
+  {
+    typedef typename DstEvaluatorType::ExpressionTraits Traits;
+    return int(Traits::ColsAtCompileTime) == 1 ? 0
+      : int(Traits::RowsAtCompileTime) == 1 ? inner
+      : int(DstEvaluatorType::Flags)&RowMajorBit ? inner
+      : outer;
+  }
+
+  EIGEN_DEVICE_FUNC const Scalar* dstDataPtr() const
+  {
+    return m_dstExpr.data();
+  }
+  
+protected:
+  DstEvaluatorType& m_dst;
+  const SrcEvaluatorType& m_src;
+  const Functor &m_functor;
+  // TODO find a way to avoid the needs of the original expression
+  DstXprType& m_dstExpr;
+};
+
+/***************************************************************************
+* Part 5 : Entry point for dense rectangular assignment
+***************************************************************************/
+
+template<typename DstXprType,typename SrcXprType, typename Functor>
+EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
+void resize_if_allowed(DstXprType &dst, const SrcXprType& src, const Functor &/*func*/)
+{
+  EIGEN_ONLY_USED_FOR_DEBUG(dst);
+  EIGEN_ONLY_USED_FOR_DEBUG(src);
+  eigen_assert(dst.rows() == src.rows() && dst.cols() == src.cols());
+}
+
+template<typename DstXprType,typename SrcXprType, typename T1, typename T2>
+EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
+void resize_if_allowed(DstXprType &dst, const SrcXprType& src, const internal::assign_op<T1,T2> &/*func*/)
+{
+  Index dstRows = src.rows();
+  Index dstCols = src.cols();
+  if(((dst.rows()!=dstRows) || (dst.cols()!=dstCols)))
+    dst.resize(dstRows, dstCols);
+  eigen_assert(dst.rows() == dstRows && dst.cols() == dstCols);
+}
+
+template<typename DstXprType, typename SrcXprType, typename Functor>
+EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE void call_dense_assignment_loop(DstXprType& dst, const SrcXprType& src, const Functor &func)
+{
+  typedef evaluator<DstXprType> DstEvaluatorType;
+  typedef evaluator<SrcXprType> SrcEvaluatorType;
+
+  SrcEvaluatorType srcEvaluator(src);
+
+  // NOTE To properly handle A = (A*A.transpose())/s with A rectangular,
+  // we need to resize the destination after the source evaluator has been created.
+  resize_if_allowed(dst, src, func);
+
+  DstEvaluatorType dstEvaluator(dst);
+    
+  typedef generic_dense_assignment_kernel<DstEvaluatorType,SrcEvaluatorType,Functor> Kernel;
+  Kernel kernel(dstEvaluator, srcEvaluator, func, dst.const_cast_derived());
+
+  dense_assignment_loop<Kernel>::run(kernel);
+}
+
+template<typename DstXprType, typename SrcXprType>
+EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE void call_dense_assignment_loop(DstXprType& dst, const SrcXprType& src)
+{
+  call_dense_assignment_loop(dst, src, internal::assign_op<typename DstXprType::Scalar,typename SrcXprType::Scalar>());
+}
+
+/***************************************************************************
+* Part 6 : Generic assignment
+***************************************************************************/
+
+// Based on the respective shapes of the destination and source,
+// the class AssignmentKind determine the kind of assignment mechanism.
+// AssignmentKind must define a Kind typedef.
+template<typename DstShape, typename SrcShape> struct AssignmentKind;
+
+// Assignement kind defined in this file:
+struct Dense2Dense {};
+struct EigenBase2EigenBase {};
+
+template<typename,typename> struct AssignmentKind { typedef EigenBase2EigenBase Kind; };
+template<> struct AssignmentKind<DenseShape,DenseShape> { typedef Dense2Dense Kind; };
+    
+// This is the main assignment class
+template< typename DstXprType, typename SrcXprType, typename Functor,
+          typename Kind = typename AssignmentKind< typename evaluator_traits<DstXprType>::Shape , typename evaluator_traits<SrcXprType>::Shape >::Kind,
+          typename EnableIf = void>
+struct Assignment;
+
+
+// The only purpose of this call_assignment() function is to deal with noalias() / "assume-aliasing" and automatic transposition.
+// Indeed, I (Gael) think that this concept of "assume-aliasing" was a mistake, and it makes thing quite complicated.
+// So this intermediate function removes everything related to "assume-aliasing" such that Assignment
+// does not has to bother about these annoying details.
+
+template<typename Dst, typename Src>
+EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
+void call_assignment(Dst& dst, const Src& src)
+{
+  call_assignment(dst, src, internal::assign_op<typename Dst::Scalar,typename Src::Scalar>());
+}
+template<typename Dst, typename Src>
+EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
+void call_assignment(const Dst& dst, const Src& src)
+{
+  call_assignment(dst, src, internal::assign_op<typename Dst::Scalar,typename Src::Scalar>());
+}
+                     
+// Deal with "assume-aliasing"
+template<typename Dst, typename Src, typename Func>
+EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
+void call_assignment(Dst& dst, const Src& src, const Func& func, typename enable_if< evaluator_assume_aliasing<Src>::value, void*>::type = 0)
+{
+  typename plain_matrix_type<Src>::type tmp(src);
+  call_assignment_no_alias(dst, tmp, func);
+}
+
+template<typename Dst, typename Src, typename Func>
+EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
+void call_assignment(Dst& dst, const Src& src, const Func& func, typename enable_if<!evaluator_assume_aliasing<Src>::value, void*>::type = 0)
+{
+  call_assignment_no_alias(dst, src, func);
+}
+
+// by-pass "assume-aliasing"
+// When there is no aliasing, we require that 'dst' has been properly resized
+template<typename Dst, template <typename> class StorageBase, typename Src, typename Func>
+EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
+void call_assignment(NoAlias<Dst,StorageBase>& dst, const Src& src, const Func& func)
+{
+  call_assignment_no_alias(dst.expression(), src, func);
+}
+
+
+template<typename Dst, typename Src, typename Func>
+EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
+void call_assignment_no_alias(Dst& dst, const Src& src, const Func& func)
+{
+  enum {
+    NeedToTranspose = (    (int(Dst::RowsAtCompileTime) == 1 && int(Src::ColsAtCompileTime) == 1)
+                        || (int(Dst::ColsAtCompileTime) == 1 && int(Src::RowsAtCompileTime) == 1)
+                      ) && int(Dst::SizeAtCompileTime) != 1
+  };
+
+  typedef typename internal::conditional<NeedToTranspose, Transpose<Dst>, Dst>::type ActualDstTypeCleaned;
+  typedef typename internal::conditional<NeedToTranspose, Transpose<Dst>, Dst&>::type ActualDstType;
+  ActualDstType actualDst(dst);
+  
+  // TODO check whether this is the right place to perform these checks:
+  EIGEN_STATIC_ASSERT_LVALUE(Dst)
+  EIGEN_STATIC_ASSERT_SAME_MATRIX_SIZE(ActualDstTypeCleaned,Src)
+  EIGEN_CHECK_BINARY_COMPATIBILIY(Func,typename ActualDstTypeCleaned::Scalar,typename Src::Scalar);
+  
+  Assignment<ActualDstTypeCleaned,Src,Func>::run(actualDst, src, func);
+}
+template<typename Dst, typename Src>
+EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
+void call_assignment_no_alias(Dst& dst, const Src& src)
+{
+  call_assignment_no_alias(dst, src, internal::assign_op<typename Dst::Scalar,typename Src::Scalar>());
+}
+
+template<typename Dst, typename Src, typename Func>
+EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
+void call_assignment_no_alias_no_transpose(Dst& dst, const Src& src, const Func& func)
+{
+  // TODO check whether this is the right place to perform these checks:
+  EIGEN_STATIC_ASSERT_LVALUE(Dst)
+  EIGEN_STATIC_ASSERT_SAME_MATRIX_SIZE(Dst,Src)
+  EIGEN_CHECK_BINARY_COMPATIBILIY(Func,typename Dst::Scalar,typename Src::Scalar);
+
+  Assignment<Dst,Src,Func>::run(dst, src, func);
+}
+template<typename Dst, typename Src>
+EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
+void call_assignment_no_alias_no_transpose(Dst& dst, const Src& src)
+{
+  call_assignment_no_alias_no_transpose(dst, src, internal::assign_op<typename Dst::Scalar,typename Src::Scalar>());
+}
+
+// forward declaration
+template<typename Dst, typename Src> void check_for_aliasing(const Dst &dst, const Src &src);
+
+// Generic Dense to Dense assignment
+// Note that the last template argument "Weak" is needed to make it possible to perform
+// both partial specialization+SFINAE without ambiguous specialization
+template< typename DstXprType, typename SrcXprType, typename Functor, typename Weak>
+struct Assignment<DstXprType, SrcXprType, Functor, Dense2Dense, Weak>
+{
+  EIGEN_DEVICE_FUNC
+  static EIGEN_STRONG_INLINE void run(DstXprType &dst, const SrcXprType &src, const Functor &func)
+  {
+#ifndef EIGEN_NO_DEBUG
+    internal::check_for_aliasing(dst, src);
+#endif
+    
+    call_dense_assignment_loop(dst, src, func);
+  }
+};
+
+// Generic assignment through evalTo.
+// TODO: not sure we have to keep that one, but it helps porting current code to new evaluator mechanism.
+// Note that the last template argument "Weak" is needed to make it possible to perform
+// both partial specialization+SFINAE without ambiguous specialization
+template< typename DstXprType, typename SrcXprType, typename Functor, typename Weak>
+struct Assignment<DstXprType, SrcXprType, Functor, EigenBase2EigenBase, Weak>
+{
+  EIGEN_DEVICE_FUNC
+  static EIGEN_STRONG_INLINE void run(DstXprType &dst, const SrcXprType &src, const internal::assign_op<typename DstXprType::Scalar,typename SrcXprType::Scalar> &/*func*/)
+  {
+    Index dstRows = src.rows();
+    Index dstCols = src.cols();
+    if((dst.rows()!=dstRows) || (dst.cols()!=dstCols))
+      dst.resize(dstRows, dstCols);
+
+    eigen_assert(dst.rows() == src.rows() && dst.cols() == src.cols());
+    src.evalTo(dst);
+  }
+
+  // NOTE The following two functions are templated to avoid their instanciation if not needed
+  //      This is needed because some expressions supports evalTo only and/or have 'void' as scalar type.
+  template<typename SrcScalarType>
+  EIGEN_DEVICE_FUNC
+  static EIGEN_STRONG_INLINE void run(DstXprType &dst, const SrcXprType &src, const internal::add_assign_op<typename DstXprType::Scalar,SrcScalarType> &/*func*/)
+  {
+    Index dstRows = src.rows();
+    Index dstCols = src.cols();
+    if((dst.rows()!=dstRows) || (dst.cols()!=dstCols))
+      dst.resize(dstRows, dstCols);
+
+    eigen_assert(dst.rows() == src.rows() && dst.cols() == src.cols());
+    src.addTo(dst);
+  }
+
+  template<typename SrcScalarType>
+  EIGEN_DEVICE_FUNC
+  static EIGEN_STRONG_INLINE void run(DstXprType &dst, const SrcXprType &src, const internal::sub_assign_op<typename DstXprType::Scalar,SrcScalarType> &/*func*/)
+  {
+    Index dstRows = src.rows();
+    Index dstCols = src.cols();
+    if((dst.rows()!=dstRows) || (dst.cols()!=dstCols))
+      dst.resize(dstRows, dstCols);
+
+    eigen_assert(dst.rows() == src.rows() && dst.cols() == src.cols());
+    src.subTo(dst);
+  }
+};
+
+} // namespace internal
+
+} // end namespace Eigen
+
+#endif // EIGEN_ASSIGN_EVALUATOR_H

Eigen/src/Core/Assign_MKL.h

@@ -0,0 +1,178 @@
+/*
+ Copyright (c) 2011, Intel Corporation. All rights reserved.
+ Copyright (C) 2015 Gael Guennebaud <gael.guennebaud@inria.fr>
+ 
+ Redistribution and use in source and binary forms, with or without modification,
+ are permitted provided that the following conditions are met:
+
+ * Redistributions of source code must retain the above copyright notice, this
+   list of conditions and the following disclaimer.
+ * Redistributions in binary form must reproduce the above copyright notice,
+   this list of conditions and the following disclaimer in the documentation
+   and/or other materials provided with the distribution.
+ * Neither the name of Intel Corporation nor the names of its contributors may
+   be used to endorse or promote products derived from this software without
+   specific prior written permission.
+
+ THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" AND
+ ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED
+ WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE
+ DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE LIABLE FOR
+ ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES
+ (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES;
+ LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON
+ ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+ (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS
+ SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+
+ ********************************************************************************
+ *   Content : Eigen bindings to Intel(R) MKL
+ *   MKL VML support for coefficient-wise unary Eigen expressions like a=b.sin()
+ ********************************************************************************
+*/
+
+#ifndef EIGEN_ASSIGN_VML_H
+#define EIGEN_ASSIGN_VML_H
+
+namespace Eigen { 
+
+namespace internal {
+
+template<typename Dst, typename Src>
+class vml_assign_traits
+{
+  private:
+    enum {
+      DstHasDirectAccess = Dst::Flags & DirectAccessBit,
+      SrcHasDirectAccess = Src::Flags & DirectAccessBit,
+      StorageOrdersAgree = (int(Dst::IsRowMajor) == int(Src::IsRowMajor)),
+      InnerSize = int(Dst::IsVectorAtCompileTime) ? int(Dst::SizeAtCompileTime)
+                : int(Dst::Flags)&RowMajorBit ? int(Dst::ColsAtCompileTime)
+                : int(Dst::RowsAtCompileTime),
+      InnerMaxSize  = int(Dst::IsVectorAtCompileTime) ? int(Dst::MaxSizeAtCompileTime)
+                    : int(Dst::Flags)&RowMajorBit ? int(Dst::MaxColsAtCompileTime)
+                    : int(Dst::MaxRowsAtCompileTime),
+      MaxSizeAtCompileTime = Dst::SizeAtCompileTime,
+
+      MightEnableVml = StorageOrdersAgree && DstHasDirectAccess && SrcHasDirectAccess && Src::InnerStrideAtCompileTime==1 && Dst::InnerStrideAtCompileTime==1,
+      MightLinearize = MightEnableVml && (int(Dst::Flags) & int(Src::Flags) & LinearAccessBit),
+      VmlSize = MightLinearize ? MaxSizeAtCompileTime : InnerMaxSize,
+      LargeEnough = VmlSize==Dynamic || VmlSize>=EIGEN_MKL_VML_THRESHOLD
+    };
+  public:
+    enum {
+      EnableVml = MightEnableVml && LargeEnough,
+      Traversal = MightLinearize ? LinearTraversal : DefaultTraversal
+    };
+};
+
+#define EIGEN_PP_EXPAND(ARG) ARG
+#if !defined (EIGEN_FAST_MATH) || (EIGEN_FAST_MATH != 1)
+#define EIGEN_VMLMODE_EXPAND_LA , VML_HA
+#else
+#define EIGEN_VMLMODE_EXPAND_LA , VML_LA
+#endif
+
+#define EIGEN_VMLMODE_EXPAND__ 
+
+#define EIGEN_VMLMODE_PREFIX_LA vm
+#define EIGEN_VMLMODE_PREFIX__  v
+#define EIGEN_VMLMODE_PREFIX(VMLMODE) EIGEN_CAT(EIGEN_VMLMODE_PREFIX_,VMLMODE)
+
+#define EIGEN_MKL_VML_DECLARE_UNARY_CALL(EIGENOP, VMLOP, EIGENTYPE, VMLTYPE, VMLMODE)                                           \
+  template< typename DstXprType, typename SrcXprNested>                                                                         \
+  struct Assignment<DstXprType, CwiseUnaryOp<scalar_##EIGENOP##_op<EIGENTYPE>, SrcXprNested>, assign_op<EIGENTYPE,EIGENTYPE>,   \
+                   Dense2Dense, typename enable_if<vml_assign_traits<DstXprType,SrcXprNested>::EnableVml>::type> {              \
+    typedef CwiseUnaryOp<scalar_##EIGENOP##_op<EIGENTYPE>, SrcXprNested> SrcXprType;                                            \
+    static void run(DstXprType &dst, const SrcXprType &src, const assign_op<EIGENTYPE,EIGENTYPE> &func) {                       \
+      resize_if_allowed(dst, src, func);                                                                                        \
+      eigen_assert(dst.rows() == src.rows() && dst.cols() == src.cols());                                                       \
+      if(vml_assign_traits<DstXprType,SrcXprNested>::Traversal==LinearTraversal) {                                              \
+        VMLOP(dst.size(), (const VMLTYPE*)src.nestedExpression().data(),                                                        \
+              (VMLTYPE*)dst.data() EIGEN_PP_EXPAND(EIGEN_VMLMODE_EXPAND_##VMLMODE) );                                           \
+      } else {                                                                                                                  \
+        const Index outerSize = dst.outerSize();                                                                                \
+        for(Index outer = 0; outer < outerSize; ++outer) {                                                                      \
+          const EIGENTYPE *src_ptr = src.IsRowMajor ? &(src.nestedExpression().coeffRef(outer,0)) :                             \
+                                                      &(src.nestedExpression().coeffRef(0, outer));                             \
+          EIGENTYPE *dst_ptr = dst.IsRowMajor ? &(dst.coeffRef(outer,0)) : &(dst.coeffRef(0, outer));                           \
+          VMLOP( dst.innerSize(), (const VMLTYPE*)src_ptr,                                                                      \
+                (VMLTYPE*)dst_ptr EIGEN_PP_EXPAND(EIGEN_VMLMODE_EXPAND_##VMLMODE));                                             \
+        }                                                                                                                       \
+      }                                                                                                                         \
+    }                                                                                                                           \
+  };                                                                                                                            \
+
+
+#define EIGEN_MKL_VML_DECLARE_UNARY_CALLS_REAL(EIGENOP, VMLOP, VMLMODE)                                                         \
+  EIGEN_MKL_VML_DECLARE_UNARY_CALL(EIGENOP, EIGEN_CAT(EIGEN_VMLMODE_PREFIX(VMLMODE),s##VMLOP), float, float, VMLMODE)           \
+  EIGEN_MKL_VML_DECLARE_UNARY_CALL(EIGENOP, EIGEN_CAT(EIGEN_VMLMODE_PREFIX(VMLMODE),d##VMLOP), double, double, VMLMODE)
+
+#define EIGEN_MKL_VML_DECLARE_UNARY_CALLS_CPLX(EIGENOP, VMLOP, VMLMODE)                                                         \
+  EIGEN_MKL_VML_DECLARE_UNARY_CALL(EIGENOP, EIGEN_CAT(EIGEN_VMLMODE_PREFIX(VMLMODE),c##VMLOP), scomplex, MKL_Complex8, VMLMODE) \
+  EIGEN_MKL_VML_DECLARE_UNARY_CALL(EIGENOP, EIGEN_CAT(EIGEN_VMLMODE_PREFIX(VMLMODE),z##VMLOP), dcomplex, MKL_Complex16, VMLMODE)
+  
+#define EIGEN_MKL_VML_DECLARE_UNARY_CALLS(EIGENOP, VMLOP, VMLMODE)                                                              \
+  EIGEN_MKL_VML_DECLARE_UNARY_CALLS_REAL(EIGENOP, VMLOP, VMLMODE)                                                               \
+  EIGEN_MKL_VML_DECLARE_UNARY_CALLS_CPLX(EIGENOP, VMLOP, VMLMODE)
+
+  
+EIGEN_MKL_VML_DECLARE_UNARY_CALLS(sin,   Sin,   LA)
+EIGEN_MKL_VML_DECLARE_UNARY_CALLS(asin,  Asin,  LA)
+EIGEN_MKL_VML_DECLARE_UNARY_CALLS(sinh,  Sinh,  LA)
+EIGEN_MKL_VML_DECLARE_UNARY_CALLS(cos,   Cos,   LA)
+EIGEN_MKL_VML_DECLARE_UNARY_CALLS(acos,  Acos,  LA)
+EIGEN_MKL_VML_DECLARE_UNARY_CALLS(cosh,  Cosh,  LA)
+EIGEN_MKL_VML_DECLARE_UNARY_CALLS(tan,   Tan,   LA)
+EIGEN_MKL_VML_DECLARE_UNARY_CALLS(atan,  Atan,  LA)
+EIGEN_MKL_VML_DECLARE_UNARY_CALLS(tanh,  Tanh,  LA)
+// EIGEN_MKL_VML_DECLARE_UNARY_CALLS(abs,   Abs,    _)
+EIGEN_MKL_VML_DECLARE_UNARY_CALLS(exp,   Exp,   LA)
+EIGEN_MKL_VML_DECLARE_UNARY_CALLS(log,   Ln,    LA)
+EIGEN_MKL_VML_DECLARE_UNARY_CALLS(log10, Log10, LA)
+EIGEN_MKL_VML_DECLARE_UNARY_CALLS(sqrt,  Sqrt,  _)
+
+EIGEN_MKL_VML_DECLARE_UNARY_CALLS_REAL(square, Sqr,   _)
+EIGEN_MKL_VML_DECLARE_UNARY_CALLS_CPLX(arg, Arg,      _)
+EIGEN_MKL_VML_DECLARE_UNARY_CALLS_REAL(round, Round,  _)
+EIGEN_MKL_VML_DECLARE_UNARY_CALLS_REAL(floor, Floor,  _)
+EIGEN_MKL_VML_DECLARE_UNARY_CALLS_REAL(ceil,  Ceil,   _)
+
+#define EIGEN_MKL_VML_DECLARE_POW_CALL(EIGENOP, VMLOP, EIGENTYPE, VMLTYPE, VMLMODE)                                           \
+  template< typename DstXprType, typename SrcXprNested, typename Plain>                                                       \
+  struct Assignment<DstXprType, CwiseBinaryOp<scalar_##EIGENOP##_op<EIGENTYPE,EIGENTYPE>, SrcXprNested,                       \
+                    const CwiseNullaryOp<internal::scalar_constant_op<EIGENTYPE>,Plain> >, assign_op<EIGENTYPE,EIGENTYPE>,    \
+                   Dense2Dense, typename enable_if<vml_assign_traits<DstXprType,SrcXprNested>::EnableVml>::type> {            \
+    typedef CwiseBinaryOp<scalar_##EIGENOP##_op<EIGENTYPE,EIGENTYPE>, SrcXprNested,                                           \
+                    const CwiseNullaryOp<internal::scalar_constant_op<EIGENTYPE>,Plain> > SrcXprType;                         \
+    static void run(DstXprType &dst, const SrcXprType &src, const assign_op<EIGENTYPE,EIGENTYPE> &func) {                     \
+      resize_if_allowed(dst, src, func);                                                                                      \
+      eigen_assert(dst.rows() == src.rows() && dst.cols() == src.cols());                                                     \
+      VMLTYPE exponent = reinterpret_cast<const VMLTYPE&>(src.rhs().functor().m_other);                                       \
+      if(vml_assign_traits<DstXprType,SrcXprNested>::Traversal==LinearTraversal)                                              \
+      {                                                                                                                       \
+        VMLOP( dst.size(), (const VMLTYPE*)src.lhs().data(), exponent,                                                        \
+              (VMLTYPE*)dst.data() EIGEN_PP_EXPAND(EIGEN_VMLMODE_EXPAND_##VMLMODE) );                                         \
+      } else {                                                                                                                \
+        const Index outerSize = dst.outerSize();                                                                              \
+        for(Index outer = 0; outer < outerSize; ++outer) {                                                                    \
+          const EIGENTYPE *src_ptr = src.IsRowMajor ? &(src.lhs().coeffRef(outer,0)) :                                        \
+                                                      &(src.lhs().coeffRef(0, outer));                                        \
+          EIGENTYPE *dst_ptr = dst.IsRowMajor ? &(dst.coeffRef(outer,0)) : &(dst.coeffRef(0, outer));                         \
+          VMLOP( dst.innerSize(), (const VMLTYPE*)src_ptr, exponent,                                                          \
+                 (VMLTYPE*)dst_ptr EIGEN_PP_EXPAND(EIGEN_VMLMODE_EXPAND_##VMLMODE));                                          \
+        }                                                                                                                     \
+      }                                                                                                                       \
+    }                                                                                                                         \
+  };
+  
+EIGEN_MKL_VML_DECLARE_POW_CALL(pow, vmsPowx, float,    float,         LA)
+EIGEN_MKL_VML_DECLARE_POW_CALL(pow, vmdPowx, double,   double,        LA)
+EIGEN_MKL_VML_DECLARE_POW_CALL(pow, vmcPowx, scomplex, MKL_Complex8,  LA)
+EIGEN_MKL_VML_DECLARE_POW_CALL(pow, vmzPowx, dcomplex, MKL_Complex16, LA)
+
+} // end namespace internal
+
+} // end namespace Eigen
+
+#endif // EIGEN_ASSIGN_VML_H

Eigen/src/Core/BandMatrix.h

@@ -0,0 +1,353 @@
+// This file is part of Eigen, a lightweight C++ template library
+// for linear algebra.
+//
+// Copyright (C) 2009 Gael Guennebaud <gael.guennebaud@inria.fr>
+//
+// This Source Code Form is subject to the terms of the Mozilla
+// Public License v. 2.0. If a copy of the MPL was not distributed
+// with this file, You can obtain one at http://mozilla.org/MPL/2.0/.
+
+#ifndef EIGEN_BANDMATRIX_H
+#define EIGEN_BANDMATRIX_H
+
+namespace Eigen { 
+
+namespace internal {
+
+template<typename Derived>
+class BandMatrixBase : public EigenBase<Derived>
+{
+  public:
+
+    enum {
+      Flags = internal::traits<Derived>::Flags,
+      CoeffReadCost = internal::traits<Derived>::CoeffReadCost,
+      RowsAtCompileTime = internal::traits<Derived>::RowsAtCompileTime,
+      ColsAtCompileTime = internal::traits<Derived>::ColsAtCompileTime,
+      MaxRowsAtCompileTime = internal::traits<Derived>::MaxRowsAtCompileTime,
+      MaxColsAtCompileTime = internal::traits<Derived>::MaxColsAtCompileTime,
+      Supers = internal::traits<Derived>::Supers,
+      Subs   = internal::traits<Derived>::Subs,
+      Options = internal::traits<Derived>::Options
+    };
+    typedef typename internal::traits<Derived>::Scalar Scalar;
+    typedef Matrix<Scalar,RowsAtCompileTime,ColsAtCompileTime> DenseMatrixType;
+    typedef typename DenseMatrixType::StorageIndex StorageIndex;
+    typedef typename internal::traits<Derived>::CoefficientsType CoefficientsType;
+    typedef EigenBase<Derived> Base;
+
+  protected:
+    enum {
+      DataRowsAtCompileTime = ((Supers!=Dynamic) && (Subs!=Dynamic))
+                            ? 1 + Supers + Subs
+                            : Dynamic,
+      SizeAtCompileTime = EIGEN_SIZE_MIN_PREFER_DYNAMIC(RowsAtCompileTime,ColsAtCompileTime)
+    };
+
+  public:
+    
+    using Base::derived;
+    using Base::rows;
+    using Base::cols;
+
+    /** \returns the number of super diagonals */
+    inline Index supers() const { return derived().supers(); }
+
+    /** \returns the number of sub diagonals */
+    inline Index subs() const { return derived().subs(); }
+    
+    /** \returns an expression of the underlying coefficient matrix */
+    inline const CoefficientsType& coeffs() const { return derived().coeffs(); }
+    
+    /** \returns an expression of the underlying coefficient matrix */
+    inline CoefficientsType& coeffs() { return derived().coeffs(); }
+
+    /** \returns a vector expression of the \a i -th column,
+      * only the meaningful part is returned.
+      * \warning the internal storage must be column major. */
+    inline Block<CoefficientsType,Dynamic,1> col(Index i)
+    {
+      EIGEN_STATIC_ASSERT((Options&RowMajor)==0,THIS_METHOD_IS_ONLY_FOR_COLUMN_MAJOR_MATRICES);
+      Index start = 0;
+      Index len = coeffs().rows();
+      if (i<=supers())
+      {
+        start = supers()-i;
+        len = (std::min)(rows(),std::max<Index>(0,coeffs().rows() - (supers()-i)));
+      }
+      else if (i>=rows()-subs())
+        len = std::max<Index>(0,coeffs().rows() - (i + 1 - rows() + subs()));
+      return Block<CoefficientsType,Dynamic,1>(coeffs(), start, i, len, 1);
+    }
+
+    /** \returns a vector expression of the main diagonal */
+    inline Block<CoefficientsType,1,SizeAtCompileTime> diagonal()
+    { return Block<CoefficientsType,1,SizeAtCompileTime>(coeffs(),supers(),0,1,(std::min)(rows(),cols())); }
+
+    /** \returns a vector expression of the main diagonal (const version) */
+    inline const Block<const CoefficientsType,1,SizeAtCompileTime> diagonal() const
+    { return Block<const CoefficientsType,1,SizeAtCompileTime>(coeffs(),supers(),0,1,(std::min)(rows(),cols())); }
+
+    template<int Index> struct DiagonalIntReturnType {
+      enum {
+        ReturnOpposite = (Options&SelfAdjoint) && (((Index)>0 && Supers==0) || ((Index)<0 && Subs==0)),
+        Conjugate = ReturnOpposite && NumTraits<Scalar>::IsComplex,
+        ActualIndex = ReturnOpposite ? -Index : Index,
+        DiagonalSize = (RowsAtCompileTime==Dynamic || ColsAtCompileTime==Dynamic)
+                     ? Dynamic
+                     : (ActualIndex<0
+                     ? EIGEN_SIZE_MIN_PREFER_DYNAMIC(ColsAtCompileTime, RowsAtCompileTime + ActualIndex)
+                     : EIGEN_SIZE_MIN_PREFER_DYNAMIC(RowsAtCompileTime, ColsAtCompileTime - ActualIndex))
+      };
+      typedef Block<CoefficientsType,1, DiagonalSize> BuildType;
+      typedef typename internal::conditional<Conjugate,
+                 CwiseUnaryOp<internal::scalar_conjugate_op<Scalar>,BuildType >,
+                 BuildType>::type Type;
+    };
+
+    /** \returns a vector expression of the \a N -th sub or super diagonal */
+    template<int N> inline typename DiagonalIntReturnType<N>::Type diagonal()
+    {
+      return typename DiagonalIntReturnType<N>::BuildType(coeffs(), supers()-N, (std::max)(0,N), 1, diagonalLength(N));
+    }
+
+    /** \returns a vector expression of the \a N -th sub or super diagonal */
+    template<int N> inline const typename DiagonalIntReturnType<N>::Type diagonal() const
+    {
+      return typename DiagonalIntReturnType<N>::BuildType(coeffs(), supers()-N, (std::max)(0,N), 1, diagonalLength(N));
+    }
+
+    /** \returns a vector expression of the \a i -th sub or super diagonal */
+    inline Block<CoefficientsType,1,Dynamic> diagonal(Index i)
+    {
+      eigen_assert((i<0 && -i<=subs()) || (i>=0 && i<=supers()));
+      return Block<CoefficientsType,1,Dynamic>(coeffs(), supers()-i, std::max<Index>(0,i), 1, diagonalLength(i));
+    }
+
+    /** \returns a vector expression of the \a i -th sub or super diagonal */
+    inline const Block<const CoefficientsType,1,Dynamic> diagonal(Index i) const
+    {
+      eigen_assert((i<0 && -i<=subs()) || (i>=0 && i<=supers()));
+      return Block<const CoefficientsType,1,Dynamic>(coeffs(), supers()-i, std::max<Index>(0,i), 1, diagonalLength(i));
+    }
+    
+    template<typename Dest> inline void evalTo(Dest& dst) const
+    {
+      dst.resize(rows(),cols());
+      dst.setZero();
+      dst.diagonal() = diagonal();
+      for (Index i=1; i<=supers();++i)
+        dst.diagonal(i) = diagonal(i);
+      for (Index i=1; i<=subs();++i)
+        dst.diagonal(-i) = diagonal(-i);
+    }
+
+    DenseMatrixType toDenseMatrix() const
+    {
+      DenseMatrixType res(rows(),cols());
+      evalTo(res);
+      return res;
+    }
+
+  protected:
+
+    inline Index diagonalLength(Index i) const
+    { return i<0 ? (std::min)(cols(),rows()+i) : (std::min)(rows(),cols()-i); }
+};
+
+/**
+  * \class BandMatrix
+  * \ingroup Core_Module
+  *
+  * \brief Represents a rectangular matrix with a banded storage
+  *
+  * \tparam _Scalar Numeric type, i.e. float, double, int
+  * \tparam _Rows Number of rows, or \b Dynamic
+  * \tparam _Cols Number of columns, or \b Dynamic
+  * \tparam _Supers Number of super diagonal
+  * \tparam _Subs Number of sub diagonal
+  * \tparam _Options A combination of either \b #RowMajor or \b #ColMajor, and of \b #SelfAdjoint
+  *                  The former controls \ref TopicStorageOrders "storage order", and defaults to
+  *                  column-major. The latter controls whether the matrix represents a selfadjoint
+  *                  matrix in which case either Supers of Subs have to be null.
+  *
+  * \sa class TridiagonalMatrix
+  */
+
+template<typename _Scalar, int _Rows, int _Cols, int _Supers, int _Subs, int _Options>
+struct traits<BandMatrix<_Scalar,_Rows,_Cols,_Supers,_Subs,_Options> >
+{
+  typedef _Scalar Scalar;
+  typedef Dense StorageKind;
+  typedef Eigen::Index StorageIndex;
+  enum {
+    CoeffReadCost = NumTraits<Scalar>::ReadCost,
+    RowsAtCompileTime = _Rows,
+    ColsAtCompileTime = _Cols,
+    MaxRowsAtCompileTime = _Rows,
+    MaxColsAtCompileTime = _Cols,
+    Flags = LvalueBit,
+    Supers = _Supers,
+    Subs = _Subs,
+    Options = _Options,
+    DataRowsAtCompileTime = ((Supers!=Dynamic) && (Subs!=Dynamic)) ? 1 + Supers + Subs : Dynamic
+  };
+  typedef Matrix<Scalar,DataRowsAtCompileTime,ColsAtCompileTime,Options&RowMajor?RowMajor:ColMajor> CoefficientsType;
+};
+
+template<typename _Scalar, int Rows, int Cols, int Supers, int Subs, int Options>
+class BandMatrix : public BandMatrixBase<BandMatrix<_Scalar,Rows,Cols,Supers,Subs,Options> >
+{
+  public:
+
+    typedef typename internal::traits<BandMatrix>::Scalar Scalar;
+    typedef typename internal::traits<BandMatrix>::StorageIndex StorageIndex;
+    typedef typename internal::traits<BandMatrix>::CoefficientsType CoefficientsType;
+
+    explicit inline BandMatrix(Index rows=Rows, Index cols=Cols, Index supers=Supers, Index subs=Subs)
+      : m_coeffs(1+supers+subs,cols),
+        m_rows(rows), m_supers(supers), m_subs(subs)
+    {
+    }
+
+    /** \returns the number of columns */
+    inline Index rows() const { return m_rows.value(); }
+
+    /** \returns the number of rows */
+    inline Index cols() const { return m_coeffs.cols(); }
+
+    /** \returns the number of super diagonals */
+    inline Index supers() const { return m_supers.value(); }
+
+    /** \returns the number of sub diagonals */
+    inline Index subs() const { return m_subs.value(); }
+
+    inline const CoefficientsType& coeffs() const { return m_coeffs; }
+    inline CoefficientsType& coeffs() { return m_coeffs; }
+
+  protected:
+
+    CoefficientsType m_coeffs;
+    internal::variable_if_dynamic<Index, Rows>   m_rows;
+    internal::variable_if_dynamic<Index, Supers> m_supers;
+    internal::variable_if_dynamic<Index, Subs>   m_subs;
+};
+
+template<typename _CoefficientsType,int _Rows, int _Cols, int _Supers, int _Subs,int _Options>
+class BandMatrixWrapper;
+
+template<typename _CoefficientsType,int _Rows, int _Cols, int _Supers, int _Subs,int _Options>
+struct traits<BandMatrixWrapper<_CoefficientsType,_Rows,_Cols,_Supers,_Subs,_Options> >
+{
+  typedef typename _CoefficientsType::Scalar Scalar;
+  typedef typename _CoefficientsType::StorageKind StorageKind;
+  typedef typename _CoefficientsType::StorageIndex StorageIndex;
+  enum {
+    CoeffReadCost = internal::traits<_CoefficientsType>::CoeffReadCost,
+    RowsAtCompileTime = _Rows,
+    ColsAtCompileTime = _Cols,
+    MaxRowsAtCompileTime = _Rows,
+    MaxColsAtCompileTime = _Cols,
+    Flags = LvalueBit,
+    Supers = _Supers,
+    Subs = _Subs,
+    Options = _Options,
+    DataRowsAtCompileTime = ((Supers!=Dynamic) && (Subs!=Dynamic)) ? 1 + Supers + Subs : Dynamic
+  };
+  typedef _CoefficientsType CoefficientsType;
+};
+
+template<typename _CoefficientsType,int _Rows, int _Cols, int _Supers, int _Subs,int _Options>
+class BandMatrixWrapper : public BandMatrixBase<BandMatrixWrapper<_CoefficientsType,_Rows,_Cols,_Supers,_Subs,_Options> >
+{
+  public:
+
+    typedef typename internal::traits<BandMatrixWrapper>::Scalar Scalar;
+    typedef typename internal::traits<BandMatrixWrapper>::CoefficientsType CoefficientsType;
+    typedef typename internal::traits<BandMatrixWrapper>::StorageIndex StorageIndex;
+
+    explicit inline BandMatrixWrapper(const CoefficientsType& coeffs, Index rows=_Rows, Index cols=_Cols, Index supers=_Supers, Index subs=_Subs)
+      : m_coeffs(coeffs),
+        m_rows(rows), m_supers(supers), m_subs(subs)
+    {
+      EIGEN_UNUSED_VARIABLE(cols);
+      //internal::assert(coeffs.cols()==cols() && (supers()+subs()+1)==coeffs.rows());
+    }
+
+    /** \returns the number of columns */
+    inline Index rows() const { return m_rows.value(); }
+
+    /** \returns the number of rows */
+    inline Index cols() const { return m_coeffs.cols(); }
+
+    /** \returns the number of super diagonals */
+    inline Index supers() const { return m_supers.value(); }
+
+    /** \returns the number of sub diagonals */
+    inline Index subs() const { return m_subs.value(); }
+
+    inline const CoefficientsType& coeffs() const { return m_coeffs; }
+
+  protected:
+
+    const CoefficientsType& m_coeffs;
+    internal::variable_if_dynamic<Index, _Rows>   m_rows;
+    internal::variable_if_dynamic<Index, _Supers> m_supers;
+    internal::variable_if_dynamic<Index, _Subs>   m_subs;
+};
+
+/**
+  * \class TridiagonalMatrix
+  * \ingroup Core_Module
+  *
+  * \brief Represents a tridiagonal matrix with a compact banded storage
+  *
+  * \tparam Scalar Numeric type, i.e. float, double, int
+  * \tparam Size Number of rows and cols, or \b Dynamic
+  * \tparam Options Can be 0 or \b SelfAdjoint
+  *
+  * \sa class BandMatrix
+  */
+template<typename Scalar, int Size, int Options>
+class TridiagonalMatrix : public BandMatrix<Scalar,Size,Size,Options&SelfAdjoint?0:1,1,Options|RowMajor>
+{
+    typedef BandMatrix<Scalar,Size,Size,Options&SelfAdjoint?0:1,1,Options|RowMajor> Base;
+    typedef typename Base::StorageIndex StorageIndex;
+  public:
+    explicit TridiagonalMatrix(Index size = Size) : Base(size,size,Options&SelfAdjoint?0:1,1) {}
+
+    inline typename Base::template DiagonalIntReturnType<1>::Type super()
+    { return Base::template diagonal<1>(); }
+    inline const typename Base::template DiagonalIntReturnType<1>::Type super() const
+    { return Base::template diagonal<1>(); }
+    inline typename Base::template DiagonalIntReturnType<-1>::Type sub()
+    { return Base::template diagonal<-1>(); }
+    inline const typename Base::template DiagonalIntReturnType<-1>::Type sub() const
+    { return Base::template diagonal<-1>(); }
+  protected:
+};
+
+
+struct BandShape {};
+
+template<typename _Scalar, int _Rows, int _Cols, int _Supers, int _Subs, int _Options>
+struct evaluator_traits<BandMatrix<_Scalar,_Rows,_Cols,_Supers,_Subs,_Options> >
+  : public evaluator_traits_base<BandMatrix<_Scalar,_Rows,_Cols,_Supers,_Subs,_Options> >
+{
+  typedef BandShape Shape;
+};
+
+template<typename _CoefficientsType,int _Rows, int _Cols, int _Supers, int _Subs,int _Options>
+struct evaluator_traits<BandMatrixWrapper<_CoefficientsType,_Rows,_Cols,_Supers,_Subs,_Options> >
+  : public evaluator_traits_base<BandMatrixWrapper<_CoefficientsType,_Rows,_Cols,_Supers,_Subs,_Options> >
+{
+  typedef BandShape Shape;
+};
+
+template<> struct AssignmentKind<DenseShape,BandShape> { typedef EigenBase2EigenBase Kind; };
+
+} // end namespace internal
+
+} // end namespace Eigen
+
+#endif // EIGEN_BANDMATRIX_H

Eigen/src/Core/BooleanRedux.h

@@ -0,0 +1,164 @@
+// This file is part of Eigen, a lightweight C++ template library
+// for linear algebra.
+//
+// Copyright (C) 2008 Gael Guennebaud <gael.guennebaud@inria.fr>
+//
+// This Source Code Form is subject to the terms of the Mozilla
+// Public License v. 2.0. If a copy of the MPL was not distributed
+// with this file, You can obtain one at http://mozilla.org/MPL/2.0/.
+
+#ifndef EIGEN_ALLANDANY_H
+#define EIGEN_ALLANDANY_H
+
+namespace Eigen { 
+
+namespace internal {
+
+template<typename Derived, int UnrollCount>
+struct all_unroller
+{
+  typedef typename Derived::ExpressionTraits Traits;
+  enum {
+    col = (UnrollCount-1) / Traits::RowsAtCompileTime,
+    row = (UnrollCount-1) % Traits::RowsAtCompileTime
+  };
+
+  static inline bool run(const Derived &mat)
+  {
+    return all_unroller<Derived, UnrollCount-1>::run(mat) && mat.coeff(row, col);
+  }
+};
+
+template<typename Derived>
+struct all_unroller<Derived, 0>
+{
+  static inline bool run(const Derived &/*mat*/) { return true; }
+};
+
+template<typename Derived>
+struct all_unroller<Derived, Dynamic>
+{
+  static inline bool run(const Derived &) { return false; }
+};
+
+template<typename Derived, int UnrollCount>
+struct any_unroller
+{
+  typedef typename Derived::ExpressionTraits Traits;
+  enum {
+    col = (UnrollCount-1) / Traits::RowsAtCompileTime,
+    row = (UnrollCount-1) % Traits::RowsAtCompileTime
+  };
+  
+  static inline bool run(const Derived &mat)
+  {
+    return any_unroller<Derived, UnrollCount-1>::run(mat) || mat.coeff(row, col);
+  }
+};
+
+template<typename Derived>
+struct any_unroller<Derived, 0>
+{
+  static inline bool run(const Derived & /*mat*/) { return false; }
+};
+
+template<typename Derived>
+struct any_unroller<Derived, Dynamic>
+{
+  static inline bool run(const Derived &) { return false; }
+};
+
+} // end namespace internal
+
+/** \returns true if all coefficients are true
+  *
+  * Example: \include MatrixBase_all.cpp
+  * Output: \verbinclude MatrixBase_all.out
+  *
+  * \sa any(), Cwise::operator<()
+  */
+template<typename Derived>
+inline bool DenseBase<Derived>::all() const
+{
+  typedef internal::evaluator<Derived> Evaluator;
+  enum {
+    unroll = SizeAtCompileTime != Dynamic
+          && SizeAtCompileTime * (Evaluator::CoeffReadCost + NumTraits<Scalar>::AddCost) <= EIGEN_UNROLLING_LIMIT
+  };
+  Evaluator evaluator(derived());
+  if(unroll)
+    return internal::all_unroller<Evaluator, unroll ? int(SizeAtCompileTime) : Dynamic>::run(evaluator);
+  else
+  {
+    for(Index j = 0; j < cols(); ++j)
+      for(Index i = 0; i < rows(); ++i)
+        if (!evaluator.coeff(i, j)) return false;
+    return true;
+  }
+}
+
+/** \returns true if at least one coefficient is true
+  *
+  * \sa all()
+  */
+template<typename Derived>
+inline bool DenseBase<Derived>::any() const
+{
+  typedef internal::evaluator<Derived> Evaluator;
+  enum {
+    unroll = SizeAtCompileTime != Dynamic
+          && SizeAtCompileTime * (Evaluator::CoeffReadCost + NumTraits<Scalar>::AddCost) <= EIGEN_UNROLLING_LIMIT
+  };
+  Evaluator evaluator(derived());
+  if(unroll)
+    return internal::any_unroller<Evaluator, unroll ? int(SizeAtCompileTime) : Dynamic>::run(evaluator);
+  else
+  {
+    for(Index j = 0; j < cols(); ++j)
+      for(Index i = 0; i < rows(); ++i)
+        if (evaluator.coeff(i, j)) return true;
+    return false;
+  }
+}
+
+/** \returns the number of coefficients which evaluate to true
+  *
+  * \sa all(), any()
+  */
+template<typename Derived>
+inline Eigen::Index DenseBase<Derived>::count() const
+{
+  return derived().template cast<bool>().template cast<Index>().sum();
+}
+
+/** \returns true is \c *this contains at least one Not A Number (NaN).
+  *
+  * \sa allFinite()
+  */
+template<typename Derived>
+inline bool DenseBase<Derived>::hasNaN() const
+{
+#if EIGEN_COMP_MSVC || (defined __FAST_MATH__)
+  return derived().array().isNaN().any();
+#else
+  return !((derived().array()==derived().array()).all());
+#endif
+}
+
+/** \returns true if \c *this contains only finite numbers, i.e., no NaN and no +/-INF values.
+  *
+  * \sa hasNaN()
+  */
+template<typename Derived>
+inline bool DenseBase<Derived>::allFinite() const
+{
+#if EIGEN_COMP_MSVC || (defined __FAST_MATH__)
+  return derived().array().isFinite().all();
+#else
+  return !((derived()-derived()).hasNaN());
+#endif
+}
+    
+} // end namespace Eigen
+
+#endif // EIGEN_ALLANDANY_H

Eigen/src/Core/CMakeLists.txt

@@ -1,9 +0,0 @@
-FILE(GLOB Eigen_Core_SRCS "*.h")
-
-INSTALL(FILES
-  ${Eigen_Core_SRCS}
-  DESTINATION ${INCLUDE_INSTALL_DIR}/Eigen/src/Core
-  )
-
-ADD_SUBDIRECTORY(util)
-ADD_SUBDIRECTORY(arch)

Eigen/src/Core/CacheFriendlyProduct.h

@@ -1,753 +0,0 @@
-// This file is part of Eigen, a lightweight C++ template library
-// for linear algebra. Eigen itself is part of the KDE project.
-//
-// Copyright (C) 2008 Gael Guennebaud <g.gael@free.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_CACHE_FRIENDLY_PRODUCT_H
-#define EIGEN_CACHE_FRIENDLY_PRODUCT_H
-
-template <int L2MemorySize,typename Scalar>
-struct ei_L2_block_traits {
-  enum {width = 8 * ei_meta_sqrt<L2MemorySize/(64*sizeof(Scalar))>::ret };
-};
-
-#ifndef EIGEN_EXTERN_INSTANTIATIONS
-
-template<typename Scalar>
-static void ei_cache_friendly_product(
-  int _rows, int _cols, int depth,
-  bool _lhsRowMajor, const Scalar* _lhs, int _lhsStride,
-  bool _rhsRowMajor, const Scalar* _rhs, int _rhsStride,
-  bool resRowMajor, Scalar* res, int resStride)
-{
-  const Scalar* EIGEN_RESTRICT lhs;
-  const Scalar* EIGEN_RESTRICT rhs;
-  int lhsStride, rhsStride, rows, cols;
-  bool lhsRowMajor;
-
-  if (resRowMajor)
-  {
-    lhs = _rhs;
-    rhs = _lhs;
-    lhsStride = _rhsStride;
-    rhsStride = _lhsStride;
-    cols = _rows;
-    rows = _cols;
-    lhsRowMajor = !_rhsRowMajor;
-    ei_assert(_lhsRowMajor);
-  }
-  else
-  {
-    lhs = _lhs;
-    rhs = _rhs;
-    lhsStride = _lhsStride;
-    rhsStride = _rhsStride;
-    rows = _rows;
-    cols = _cols;
-    lhsRowMajor = _lhsRowMajor;
-    ei_assert(!_rhsRowMajor);
-  }
-
-  typedef typename ei_packet_traits<Scalar>::type PacketType;
-
-  enum {
-    PacketSize = sizeof(PacketType)/sizeof(Scalar),
-    #if (defined __i386__)
-    // i386 architecture provides only 8 xmm registers,
-    // so let's reduce the max number of rows processed at once.
-    MaxBlockRows = 4,
-    MaxBlockRows_ClampingMask = 0xFFFFFC,
-    #else
-    MaxBlockRows = 8,
-    MaxBlockRows_ClampingMask = 0xFFFFF8,
-    #endif
-    // maximal size of the blocks fitted in L2 cache
-    MaxL2BlockSize = ei_L2_block_traits<EIGEN_TUNE_FOR_CPU_CACHE_SIZE,Scalar>::width
-  };
-
-  const bool resIsAligned = (PacketSize==1) || (((resStride%PacketSize) == 0) && (size_t(res)%16==0));
-
-  const int remainingSize = depth % PacketSize;
-  const int size = depth - remainingSize; // third dimension of the product clamped to packet boundaries
-  const int l2BlockRows = MaxL2BlockSize > rows ? rows : MaxL2BlockSize;
-  const int l2BlockCols = MaxL2BlockSize > cols ? cols : MaxL2BlockSize;
-  const int l2BlockSize = MaxL2BlockSize > size ? size : MaxL2BlockSize;
-  const int l2BlockSizeAligned = (1 + std::max(l2BlockSize,l2BlockCols)/PacketSize)*PacketSize;
-  const bool needRhsCopy = (PacketSize>1) && ((rhsStride%PacketSize!=0) || (size_t(rhs)%16!=0));
-  Scalar* EIGEN_RESTRICT block = 0;
-  const int allocBlockSize = l2BlockRows*size;
-  block = ei_aligned_stack_new(Scalar, allocBlockSize);
-  Scalar* EIGEN_RESTRICT rhsCopy
-    = ei_aligned_stack_new(Scalar, l2BlockSizeAligned*l2BlockSizeAligned);
-
-  // loops on each L2 cache friendly blocks of the result
-  for(int l2i=0; l2i<rows; l2i+=l2BlockRows)
-  {
-    const int l2blockRowEnd = std::min(l2i+l2BlockRows, rows);
-    const int l2blockRowEndBW = l2blockRowEnd & MaxBlockRows_ClampingMask;    // end of the rows aligned to bw
-    const int l2blockRemainingRows = l2blockRowEnd - l2blockRowEndBW;         // number of remaining rows
-    //const int l2blockRowEndBWPlusOne = l2blockRowEndBW + (l2blockRemainingRows?0:MaxBlockRows);
-
-    // build a cache friendly blocky matrix
-    int count = 0;
-
-    // copy l2blocksize rows of m_lhs to blocks of ps x bw
-    for(int l2k=0; l2k<size; l2k+=l2BlockSize)
-    {
-      const int l2blockSizeEnd = std::min(l2k+l2BlockSize, size);
-
-      for (int i = l2i; i<l2blockRowEndBW/*PlusOne*/; i+=MaxBlockRows)
-      {
-        // TODO merge the "if l2blockRemainingRows" using something like:
-        // const int blockRows = std::min(i+MaxBlockRows, rows) - i;
-
-        for (int k=l2k; k<l2blockSizeEnd; k+=PacketSize)
-        {
-          // TODO write these loops using meta unrolling
-          // negligible for large matrices but useful for small ones
-          if (lhsRowMajor)
-          {
-            for (int w=0; w<MaxBlockRows; ++w)
-              for (int s=0; s<PacketSize; ++s)
-                block[count++] = lhs[(i+w)*lhsStride + (k+s)];
-          }
-          else
-          {
-            for (int w=0; w<MaxBlockRows; ++w)
-              for (int s=0; s<PacketSize; ++s)
-                block[count++] = lhs[(i+w) + (k+s)*lhsStride];
-          }
-        }
-      }
-      if (l2blockRemainingRows>0)
-      {
-        for (int k=l2k; k<l2blockSizeEnd; k+=PacketSize)
-        {
-          if (lhsRowMajor)
-          {
-            for (int w=0; w<l2blockRemainingRows; ++w)
-              for (int s=0; s<PacketSize; ++s)
-                block[count++] = lhs[(l2blockRowEndBW+w)*lhsStride + (k+s)];
-          }
-          else
-          {
-            for (int w=0; w<l2blockRemainingRows; ++w)
-              for (int s=0; s<PacketSize; ++s)
-                block[count++] = lhs[(l2blockRowEndBW+w) + (k+s)*lhsStride];
-          }
-        }
-      }
-    }
-
-    for(int l2j=0; l2j<cols; l2j+=l2BlockCols)
-    {
-      int l2blockColEnd = std::min(l2j+l2BlockCols, cols);
-
-      for(int l2k=0; l2k<size; l2k+=l2BlockSize)
-      {
-        // acumulate bw rows of lhs time a single column of rhs to a bw x 1 block of res
-        int l2blockSizeEnd = std::min(l2k+l2BlockSize, size);
-
-        // if not aligned, copy the rhs block
-        if (needRhsCopy)
-          for(int l1j=l2j; l1j<l2blockColEnd; l1j+=1)
-          {
-            ei_internal_assert(l2BlockSizeAligned*(l1j-l2j)+(l2blockSizeEnd-l2k) < l2BlockSizeAligned*l2BlockSizeAligned);
-            memcpy(rhsCopy+l2BlockSizeAligned*(l1j-l2j),&(rhs[l1j*rhsStride+l2k]),(l2blockSizeEnd-l2k)*sizeof(Scalar));
-          }
-
-        // for each bw x 1 result's block
-        for(int l1i=l2i; l1i<l2blockRowEndBW; l1i+=MaxBlockRows)
-        {
-          int offsetblock = l2k * (l2blockRowEnd-l2i) + (l1i-l2i)*(l2blockSizeEnd-l2k) - l2k*MaxBlockRows;
-          const Scalar* EIGEN_RESTRICT localB = &block[offsetblock];
-
-          for(int l1j=l2j; l1j<l2blockColEnd; l1j+=1)
-          {
-            const Scalar* EIGEN_RESTRICT rhsColumn;
-            if (needRhsCopy)
-              rhsColumn = &(rhsCopy[l2BlockSizeAligned*(l1j-l2j)-l2k]);
-            else
-              rhsColumn = &(rhs[l1j*rhsStride]);
-
-            PacketType dst[MaxBlockRows];
-            dst[3] = dst[2] = dst[1] = dst[0] = ei_pset1(Scalar(0.));
-            if (MaxBlockRows==8)
-              dst[7] = dst[6] = dst[5] = dst[4] = dst[0];
-
-            PacketType tmp;
-
-            for(int k=l2k; k<l2blockSizeEnd; k+=PacketSize)
-            {
-              tmp = ei_ploadu(&rhsColumn[k]);
-              PacketType A0, A1, A2, A3, A4, A5;
-              A0 = ei_pload(localB + k*MaxBlockRows);
-              A1 = ei_pload(localB + k*MaxBlockRows+1*PacketSize);
-              A2 = ei_pload(localB + k*MaxBlockRows+2*PacketSize);
-              A3 = ei_pload(localB + k*MaxBlockRows+3*PacketSize);
-              if (MaxBlockRows==8) A4 = ei_pload(localB + k*MaxBlockRows+4*PacketSize);
-              if (MaxBlockRows==8) A5 = ei_pload(localB + k*MaxBlockRows+5*PacketSize);
-              dst[0] = ei_pmadd(tmp, A0, dst[0]);
-              if (MaxBlockRows==8) A0 = ei_pload(localB + k*MaxBlockRows+6*PacketSize);
-              dst[1] = ei_pmadd(tmp, A1, dst[1]);
-              if (MaxBlockRows==8) A1 = ei_pload(localB + k*MaxBlockRows+7*PacketSize);
-              dst[2] = ei_pmadd(tmp, A2, dst[2]);
-              dst[3] = ei_pmadd(tmp, A3, dst[3]);
-              if (MaxBlockRows==8)
-              {
-                dst[4] = ei_pmadd(tmp, A4, dst[4]);
-                dst[5] = ei_pmadd(tmp, A5, dst[5]);
-                dst[6] = ei_pmadd(tmp, A0, dst[6]);
-                dst[7] = ei_pmadd(tmp, A1, dst[7]);
-              }
-            }
-
-            Scalar* EIGEN_RESTRICT localRes = &(res[l1i + l1j*resStride]);
-
-            if (PacketSize>1 && resIsAligned)
-            {
-              // the result is aligned: let's do packet reduction
-              ei_pstore(&(localRes[0]), ei_padd(ei_pload(&(localRes[0])), ei_preduxp(&dst[0])));
-              if (PacketSize==2)
-                ei_pstore(&(localRes[2]), ei_padd(ei_pload(&(localRes[2])), ei_preduxp(&(dst[2]))));
-              if (MaxBlockRows==8)
-              {
-                ei_pstore(&(localRes[4]), ei_padd(ei_pload(&(localRes[4])), ei_preduxp(&(dst[4]))));
-                if (PacketSize==2)
-                  ei_pstore(&(localRes[6]), ei_padd(ei_pload(&(localRes[6])), ei_preduxp(&(dst[6]))));
-              }
-            }
-            else
-            {
-              // not aligned => per coeff packet reduction
-              localRes[0] += ei_predux(dst[0]);
-              localRes[1] += ei_predux(dst[1]);
-              localRes[2] += ei_predux(dst[2]);
-              localRes[3] += ei_predux(dst[3]);
-              if (MaxBlockRows==8)
-              {
-                localRes[4] += ei_predux(dst[4]);
-                localRes[5] += ei_predux(dst[5]);
-                localRes[6] += ei_predux(dst[6]);
-                localRes[7] += ei_predux(dst[7]);
-              }
-            }
-          }
-        }
-        if (l2blockRemainingRows>0)
-        {
-          int offsetblock = l2k * (l2blockRowEnd-l2i) + (l2blockRowEndBW-l2i)*(l2blockSizeEnd-l2k) - l2k*l2blockRemainingRows;
-          const Scalar* localB = &block[offsetblock];
-
-          for(int l1j=l2j; l1j<l2blockColEnd; l1j+=1)
-          {
-            const Scalar* EIGEN_RESTRICT rhsColumn;
-            if (needRhsCopy)
-              rhsColumn = &(rhsCopy[l2BlockSizeAligned*(l1j-l2j)-l2k]);
-            else
-              rhsColumn = &(rhs[l1j*rhsStride]);
-
-            PacketType dst[MaxBlockRows];
-            dst[3] = dst[2] = dst[1] = dst[0] = ei_pset1(Scalar(0.));
-            if (MaxBlockRows==8)
-              dst[7] = dst[6] = dst[5] = dst[4] = dst[0];
-
-            // let's declare a few other temporary registers
-            PacketType tmp;
-
-            for(int k=l2k; k<l2blockSizeEnd; k+=PacketSize)
-            {
-              tmp = ei_pload(&rhsColumn[k]);
-
-                                           dst[0] = ei_pmadd(tmp, ei_pload(&(localB[k*l2blockRemainingRows             ])), dst[0]);
-              if (l2blockRemainingRows>=2) dst[1] = ei_pmadd(tmp, ei_pload(&(localB[k*l2blockRemainingRows+  PacketSize])), dst[1]);
-              if (l2blockRemainingRows>=3) dst[2] = ei_pmadd(tmp, ei_pload(&(localB[k*l2blockRemainingRows+2*PacketSize])), dst[2]);
-              if (l2blockRemainingRows>=4) dst[3] = ei_pmadd(tmp, ei_pload(&(localB[k*l2blockRemainingRows+3*PacketSize])), dst[3]);
-              if (MaxBlockRows==8)
-              {
-                if (l2blockRemainingRows>=5) dst[4] = ei_pmadd(tmp, ei_pload(&(localB[k*l2blockRemainingRows+4*PacketSize])), dst[4]);
-                if (l2blockRemainingRows>=6) dst[5] = ei_pmadd(tmp, ei_pload(&(localB[k*l2blockRemainingRows+5*PacketSize])), dst[5]);
-                if (l2blockRemainingRows>=7) dst[6] = ei_pmadd(tmp, ei_pload(&(localB[k*l2blockRemainingRows+6*PacketSize])), dst[6]);
-                if (l2blockRemainingRows>=8) dst[7] = ei_pmadd(tmp, ei_pload(&(localB[k*l2blockRemainingRows+7*PacketSize])), dst[7]);
-              }
-            }
-
-            Scalar* EIGEN_RESTRICT localRes = &(res[l2blockRowEndBW + l1j*resStride]);
-
-            // process the remaining rows once at a time
-                                         localRes[0] += ei_predux(dst[0]);
-            if (l2blockRemainingRows>=2) localRes[1] += ei_predux(dst[1]);
-            if (l2blockRemainingRows>=3) localRes[2] += ei_predux(dst[2]);
-            if (l2blockRemainingRows>=4) localRes[3] += ei_predux(dst[3]);
-            if (MaxBlockRows==8)
-            {
-              if (l2blockRemainingRows>=5) localRes[4] += ei_predux(dst[4]);
-              if (l2blockRemainingRows>=6) localRes[5] += ei_predux(dst[5]);
-              if (l2blockRemainingRows>=7) localRes[6] += ei_predux(dst[6]);
-              if (l2blockRemainingRows>=8) localRes[7] += ei_predux(dst[7]);
-            }
-
-          }
-        }
-      }
-    }
-  }
-  if (PacketSize>1 && remainingSize)
-  {
-    if (lhsRowMajor)
-    {
-      for (int j=0; j<cols; ++j)
-        for (int i=0; i<rows; ++i)
-        {
-          Scalar tmp = lhs[i*lhsStride+size] * rhs[j*rhsStride+size];
-          // FIXME this loop get vectorized by the compiler !
-          for (int k=1; k<remainingSize; ++k)
-            tmp += lhs[i*lhsStride+size+k] * rhs[j*rhsStride+size+k];
-          res[i+j*resStride] += tmp;
-        }
-    }
-    else
-    {
-      for (int j=0; j<cols; ++j)
-        for (int i=0; i<rows; ++i)
-        {
-          Scalar tmp = lhs[i+size*lhsStride] * rhs[j*rhsStride+size];
-          for (int k=1; k<remainingSize; ++k)
-            tmp += lhs[i+(size+k)*lhsStride] * rhs[j*rhsStride+size+k];
-          res[i+j*resStride] += tmp;
-        }
-    }
-  }
-
-  ei_aligned_stack_delete(Scalar, block, allocBlockSize);
-  ei_aligned_stack_delete(Scalar, rhsCopy, l2BlockSizeAligned*l2BlockSizeAligned);
-}
-
-#endif // EIGEN_EXTERN_INSTANTIATIONS
-
-/* Optimized col-major matrix * vector product:
- * This algorithm processes 4 columns at onces that allows to both reduce
- * the number of load/stores of the result by a factor 4 and to reduce
- * the instruction dependency. Moreover, we know that all bands have the
- * same alignment pattern.
- * TODO: since rhs gets evaluated only once, no need to evaluate it
- */
-template<typename Scalar, typename RhsType>
-static EIGEN_DONT_INLINE void ei_cache_friendly_product_colmajor_times_vector(
-  int size,
-  const Scalar* lhs, int lhsStride,
-  const RhsType& rhs,
-  Scalar* res)
-{
-  #ifdef _EIGEN_ACCUMULATE_PACKETS
-  #error _EIGEN_ACCUMULATE_PACKETS has already been defined
-  #endif
-  #define _EIGEN_ACCUMULATE_PACKETS(A0,A13,A2) \
-    ei_pstore(&res[j], \
-      ei_padd(ei_pload(&res[j]), \
-        ei_padd( \
-          ei_padd(ei_pmul(ptmp0,EIGEN_CAT(ei_ploa , A0)(&lhs0[j])), \
-                  ei_pmul(ptmp1,EIGEN_CAT(ei_ploa , A13)(&lhs1[j]))), \
-          ei_padd(ei_pmul(ptmp2,EIGEN_CAT(ei_ploa , A2)(&lhs2[j])), \
-                  ei_pmul(ptmp3,EIGEN_CAT(ei_ploa , A13)(&lhs3[j]))) )))
-
-  typedef typename ei_packet_traits<Scalar>::type Packet;
-  const int PacketSize = sizeof(Packet)/sizeof(Scalar);
-
-  enum { AllAligned = 0, EvenAligned, FirstAligned, NoneAligned };
-  const int columnsAtOnce = 4;
-  const int peels = 2;
-  const int PacketAlignedMask = PacketSize-1;
-  const int PeelAlignedMask = PacketSize*peels-1;
-
-  // How many coeffs of the result do we have to skip to be aligned.
-  // Here we assume data are at least aligned on the base scalar type that is mandatory anyway.
-  const int alignedStart = ei_alignmentOffset(res,size);
-  const int alignedSize = PacketSize>1 ? alignedStart + ((size-alignedStart) & ~PacketAlignedMask) : 0;
-  const int peeledSize  = peels>1 ? alignedStart + ((alignedSize-alignedStart) & ~PeelAlignedMask) : alignedStart;
-
-  const int alignmentStep = PacketSize>1 ? (PacketSize - lhsStride % PacketSize) & PacketAlignedMask : 0;
-  int alignmentPattern = alignmentStep==0 ? AllAligned
-                       : alignmentStep==(PacketSize/2) ? EvenAligned
-                       : FirstAligned;
-
-  // we cannot assume the first element is aligned because of sub-matrices
-  const int lhsAlignmentOffset = ei_alignmentOffset(lhs,size);
-
-  // find how many columns do we have to skip to be aligned with the result (if possible)
-  int skipColumns = 0;
-  if (PacketSize>1)
-  {
-    ei_internal_assert(size_t(lhs+lhsAlignmentOffset)%sizeof(Packet)==0 || size<PacketSize);
-
-    while (skipColumns<PacketSize &&
-           alignedStart != ((lhsAlignmentOffset + alignmentStep*skipColumns)%PacketSize))
-      ++skipColumns;
-    if (skipColumns==PacketSize)
-    {
-      // nothing can be aligned, no need to skip any column
-      alignmentPattern = NoneAligned;
-      skipColumns = 0;
-    }
-    else
-    {
-      skipColumns = std::min(skipColumns,rhs.size());
-      // note that the skiped columns are processed later.
-    }
-
-    ei_internal_assert((alignmentPattern==NoneAligned) || (size_t(lhs+alignedStart+lhsStride*skipColumns)%sizeof(Packet))==0);
-  }
-
-  int offset1 = (FirstAligned && alignmentStep==1?3:1);
-  int offset3 = (FirstAligned && alignmentStep==1?1:3);
-
-  int columnBound = ((rhs.size()-skipColumns)/columnsAtOnce)*columnsAtOnce + skipColumns;
-  for (int i=skipColumns; i<columnBound; i+=columnsAtOnce)
-  {
-    Packet ptmp0 = ei_pset1(rhs[i]),   ptmp1 = ei_pset1(rhs[i+offset1]),
-           ptmp2 = ei_pset1(rhs[i+2]), ptmp3 = ei_pset1(rhs[i+offset3]);
-
-    // this helps a lot generating better binary code
-    const Scalar *lhs0 = lhs + i*lhsStride, *lhs1 = lhs + (i+offset1)*lhsStride,
-                 *lhs2 = lhs + (i+2)*lhsStride, *lhs3 = lhs + (i+offset3)*lhsStride;
-
-    if (PacketSize>1)
-    {
-      /* explicit vectorization */
-      // process initial unaligned coeffs
-      for (int j=0; j<alignedStart; ++j)
-        res[j] += ei_pfirst(ptmp0)*lhs0[j] + ei_pfirst(ptmp1)*lhs1[j] + ei_pfirst(ptmp2)*lhs2[j] + ei_pfirst(ptmp3)*lhs3[j];
-
-      if (alignedSize>alignedStart)
-      {
-        switch(alignmentPattern)
-        {
-          case AllAligned:
-            for (int j = alignedStart; j<alignedSize; j+=PacketSize)
-              _EIGEN_ACCUMULATE_PACKETS(d,d,d);
-            break;
-          case EvenAligned:
-            for (int j = alignedStart; j<alignedSize; j+=PacketSize)
-              _EIGEN_ACCUMULATE_PACKETS(d,du,d);
-            break;
-          case FirstAligned:
-            if(peels>1)
-            {
-              Packet A00, A01, A02, A03, A10, A11, A12, A13;
-
-              A01 = ei_pload(&lhs1[alignedStart-1]);
-              A02 = ei_pload(&lhs2[alignedStart-2]);
-              A03 = ei_pload(&lhs3[alignedStart-3]);
-
-              for (int j = alignedStart; j<peeledSize; j+=peels*PacketSize)
-              {
-                A11 = ei_pload(&lhs1[j-1+PacketSize]);  ei_palign<1>(A01,A11);
-                A12 = ei_pload(&lhs2[j-2+PacketSize]);  ei_palign<2>(A02,A12);
-                A13 = ei_pload(&lhs3[j-3+PacketSize]);  ei_palign<3>(A03,A13);
-
-                A00 = ei_pload (&lhs0[j]);
-                A10 = ei_pload (&lhs0[j+PacketSize]);
-                A00 = ei_pmadd(ptmp0, A00, ei_pload(&res[j]));
-                A10 = ei_pmadd(ptmp0, A10, ei_pload(&res[j+PacketSize]));
-
-                A00 = ei_pmadd(ptmp1, A01, A00);
-                A01 = ei_pload(&lhs1[j-1+2*PacketSize]);  ei_palign<1>(A11,A01);
-                A00 = ei_pmadd(ptmp2, A02, A00);
-                A02 = ei_pload(&lhs2[j-2+2*PacketSize]);  ei_palign<2>(A12,A02);
-                A00 = ei_pmadd(ptmp3, A03, A00);
-                ei_pstore(&res[j],A00);
-                A03 = ei_pload(&lhs3[j-3+2*PacketSize]);  ei_palign<3>(A13,A03);
-                A10 = ei_pmadd(ptmp1, A11, A10);
-                A10 = ei_pmadd(ptmp2, A12, A10);
-                A10 = ei_pmadd(ptmp3, A13, A10);
-                ei_pstore(&res[j+PacketSize],A10);
-              }
-            }
-            for (int j = peeledSize; j<alignedSize; j+=PacketSize)
-              _EIGEN_ACCUMULATE_PACKETS(d,du,du);
-            break;
-          default:
-            for (int j = alignedStart; j<alignedSize; j+=PacketSize)
-              _EIGEN_ACCUMULATE_PACKETS(du,du,du);
-            break;
-        }
-      }
-    } // end explicit vectorization
-
-    /* process remaining coeffs (or all if there is no explicit vectorization) */
-    for (int j=alignedSize; j<size; ++j)
-      res[j] += ei_pfirst(ptmp0)*lhs0[j] + ei_pfirst(ptmp1)*lhs1[j] + ei_pfirst(ptmp2)*lhs2[j] + ei_pfirst(ptmp3)*lhs3[j];
-  }
-
-  // process remaining first and last columns (at most columnsAtOnce-1)
-  int end = rhs.size();
-  int start = columnBound;
-  do
-  {
-    for (int i=start; i<end; ++i)
-    {
-      Packet ptmp0 = ei_pset1(rhs[i]);
-      const Scalar* lhs0 = lhs + i*lhsStride;
-
-      if (PacketSize>1)
-      {
-        /* explicit vectorization */
-        // process first unaligned result's coeffs
-        for (int j=0; j<alignedStart; ++j)
-          res[j] += ei_pfirst(ptmp0) * lhs0[j];
-
-        // process aligned result's coeffs
-        if ((size_t(lhs0+alignedStart)%sizeof(Packet))==0)
-          for (int j = alignedStart;j<alignedSize;j+=PacketSize)
-            ei_pstore(&res[j], ei_pmadd(ptmp0,ei_pload(&lhs0[j]),ei_pload(&res[j])));
-        else
-          for (int j = alignedStart;j<alignedSize;j+=PacketSize)
-            ei_pstore(&res[j], ei_pmadd(ptmp0,ei_ploadu(&lhs0[j]),ei_pload(&res[j])));
-      }
-
-      // process remaining scalars (or all if no explicit vectorization)
-      for (int j=alignedSize; j<size; ++j)
-        res[j] += ei_pfirst(ptmp0) * lhs0[j];
-    }
-    if (skipColumns)
-    {
-      start = 0;
-      end = skipColumns;
-      skipColumns = 0;
-    }
-    else
-      break;
-  } while(PacketSize>1);
-  #undef _EIGEN_ACCUMULATE_PACKETS
-}
-
-// TODO add peeling to mask unaligned load/stores
-template<typename Scalar, typename ResType>
-static EIGEN_DONT_INLINE void ei_cache_friendly_product_rowmajor_times_vector(
-  const Scalar* lhs, int lhsStride,
-  const Scalar* rhs, int rhsSize,
-  ResType& res)
-{
-  #ifdef _EIGEN_ACCUMULATE_PACKETS
-  #error _EIGEN_ACCUMULATE_PACKETS has already been defined
-  #endif
-
-  #define _EIGEN_ACCUMULATE_PACKETS(A0,A13,A2) {\
-    Packet b = ei_pload(&rhs[j]); \
-    ptmp0 = ei_pmadd(b, EIGEN_CAT(ei_ploa,A0) (&lhs0[j]), ptmp0); \
-    ptmp1 = ei_pmadd(b, EIGEN_CAT(ei_ploa,A13)(&lhs1[j]), ptmp1); \
-    ptmp2 = ei_pmadd(b, EIGEN_CAT(ei_ploa,A2) (&lhs2[j]), ptmp2); \
-    ptmp3 = ei_pmadd(b, EIGEN_CAT(ei_ploa,A13)(&lhs3[j]), ptmp3); }
-
-  typedef typename ei_packet_traits<Scalar>::type Packet;
-  const int PacketSize = sizeof(Packet)/sizeof(Scalar);
-
-  enum { AllAligned=0, EvenAligned=1, FirstAligned=2, NoneAligned=3 };
-  const int rowsAtOnce = 4;
-  const int peels = 2;
-  const int PacketAlignedMask = PacketSize-1;
-  const int PeelAlignedMask = PacketSize*peels-1;
-  const int size = rhsSize;
-
-  // How many coeffs of the result do we have to skip to be aligned.
-  // Here we assume data are at least aligned on the base scalar type that is mandatory anyway.
-  const int alignedStart = ei_alignmentOffset(rhs, size);
-  const int alignedSize = PacketSize>1 ? alignedStart + ((size-alignedStart) & ~PacketAlignedMask) : 0;
-  const int peeledSize  = peels>1 ? alignedStart + ((alignedSize-alignedStart) & ~PeelAlignedMask) : alignedStart;
-
-  const int alignmentStep = PacketSize>1 ? (PacketSize - lhsStride % PacketSize) & PacketAlignedMask : 0;
-  int alignmentPattern = alignmentStep==0 ? AllAligned
-                       : alignmentStep==(PacketSize/2) ? EvenAligned
-                       : FirstAligned;
-
-  // we cannot assume the first element is aligned because of sub-matrices
-  const int lhsAlignmentOffset = ei_alignmentOffset(lhs,size);
-
-  // find how many rows do we have to skip to be aligned with rhs (if possible)
-  int skipRows = 0;
-  if (PacketSize>1)
-  {
-    ei_internal_assert(size_t(lhs+lhsAlignmentOffset)%sizeof(Packet)==0  || size<PacketSize);
-
-    while (skipRows<PacketSize &&
-           alignedStart != ((lhsAlignmentOffset + alignmentStep*skipRows)%PacketSize))
-      ++skipRows;
-    if (skipRows==PacketSize)
-    {
-      // nothing can be aligned, no need to skip any column
-      alignmentPattern = NoneAligned;
-      skipRows = 0;
-    }
-    else
-    {
-      skipRows = std::min(skipRows,res.size());
-      // note that the skiped columns are processed later.
-    }
-    ei_internal_assert((alignmentPattern==NoneAligned) || PacketSize==1
-      || (size_t(lhs+alignedStart+lhsStride*skipRows)%sizeof(Packet))==0);
-  }
-
-  int offset1 = (FirstAligned && alignmentStep==1?3:1);
-  int offset3 = (FirstAligned && alignmentStep==1?1:3);
-
-  int rowBound = ((res.size()-skipRows)/rowsAtOnce)*rowsAtOnce + skipRows;
-  for (int i=skipRows; i<rowBound; i+=rowsAtOnce)
-  {
-    Scalar tmp0 = Scalar(0), tmp1 = Scalar(0), tmp2 = Scalar(0), tmp3 = Scalar(0);
-
-    // this helps the compiler generating good binary code
-    const Scalar *lhs0 = lhs + i*lhsStride,     *lhs1 = lhs + (i+offset1)*lhsStride,
-                 *lhs2 = lhs + (i+2)*lhsStride, *lhs3 = lhs + (i+offset3)*lhsStride;
-
-    if (PacketSize>1)
-    {
-      /* explicit vectorization */
-      Packet ptmp0 = ei_pset1(Scalar(0)), ptmp1 = ei_pset1(Scalar(0)), ptmp2 = ei_pset1(Scalar(0)), ptmp3 = ei_pset1(Scalar(0));
-
-      // process initial unaligned coeffs
-      // FIXME this loop get vectorized by the compiler !
-      for (int j=0; j<alignedStart; ++j)
-      {
-        Scalar b = rhs[j];
-        tmp0 += b*lhs0[j]; tmp1 += b*lhs1[j]; tmp2 += b*lhs2[j]; tmp3 += b*lhs3[j];
-      }
-
-      if (alignedSize>alignedStart)
-      {
-        switch(alignmentPattern)
-        {
-          case AllAligned:
-            for (int j = alignedStart; j<alignedSize; j+=PacketSize)
-              _EIGEN_ACCUMULATE_PACKETS(d,d,d);
-            break;
-          case EvenAligned:
-            for (int j = alignedStart; j<alignedSize; j+=PacketSize)
-              _EIGEN_ACCUMULATE_PACKETS(d,du,d);
-            break;
-          case FirstAligned:
-            if (peels>1)
-            {
-              /* Here we proccess 4 rows with with two peeled iterations to hide
-               * tghe overhead of unaligned loads. Moreover unaligned loads are handled
-               * using special shift/move operations between the two aligned packets
-               * overlaping the desired unaligned packet. This is *much* more efficient
-               * than basic unaligned loads.
-               */
-              Packet A01, A02, A03, b, A11, A12, A13;
-              A01 = ei_pload(&lhs1[alignedStart-1]);
-              A02 = ei_pload(&lhs2[alignedStart-2]);
-              A03 = ei_pload(&lhs3[alignedStart-3]);
-
-              for (int j = alignedStart; j<peeledSize; j+=peels*PacketSize)
-              {
-                b = ei_pload(&rhs[j]);
-                A11 = ei_pload(&lhs1[j-1+PacketSize]);  ei_palign<1>(A01,A11);
-                A12 = ei_pload(&lhs2[j-2+PacketSize]);  ei_palign<2>(A02,A12);
-                A13 = ei_pload(&lhs3[j-3+PacketSize]);  ei_palign<3>(A03,A13);
-
-                ptmp0 = ei_pmadd(b, ei_pload (&lhs0[j]), ptmp0);
-                ptmp1 = ei_pmadd(b, A01, ptmp1);
-                A01 = ei_pload(&lhs1[j-1+2*PacketSize]);  ei_palign<1>(A11,A01);
-                ptmp2 = ei_pmadd(b, A02, ptmp2);
-                A02 = ei_pload(&lhs2[j-2+2*PacketSize]);  ei_palign<2>(A12,A02);
-                ptmp3 = ei_pmadd(b, A03, ptmp3);
-                A03 = ei_pload(&lhs3[j-3+2*PacketSize]);  ei_palign<3>(A13,A03);
-
-                b = ei_pload(&rhs[j+PacketSize]);
-                ptmp0 = ei_pmadd(b, ei_pload (&lhs0[j+PacketSize]), ptmp0);
-                ptmp1 = ei_pmadd(b, A11, ptmp1);
-                ptmp2 = ei_pmadd(b, A12, ptmp2);
-                ptmp3 = ei_pmadd(b, A13, ptmp3);
-              }
-            }
-            for (int j = peeledSize; j<alignedSize; j+=PacketSize)
-              _EIGEN_ACCUMULATE_PACKETS(d,du,du);
-            break;
-          default:
-            for (int j = alignedStart; j<alignedSize; j+=PacketSize)
-              _EIGEN_ACCUMULATE_PACKETS(du,du,du);
-            break;
-        }
-        tmp0 += ei_predux(ptmp0);
-        tmp1 += ei_predux(ptmp1);
-        tmp2 += ei_predux(ptmp2);
-        tmp3 += ei_predux(ptmp3);
-      }
-    } // end explicit vectorization
-
-    // process remaining coeffs (or all if no explicit vectorization)
-    // FIXME this loop get vectorized by the compiler !
-    for (int j=alignedSize; j<size; ++j)
-    {
-      Scalar b = rhs[j];
-      tmp0 += b*lhs0[j]; tmp1 += b*lhs1[j]; tmp2 += b*lhs2[j]; tmp3 += b*lhs3[j];
-    }
-    res[i] += tmp0; res[i+offset1] += tmp1; res[i+2] += tmp2; res[i+offset3] += tmp3;
-  }
-
-  // process remaining first and last rows (at most columnsAtOnce-1)
-  int end = res.size();
-  int start = rowBound;
-  do
-  {
-    for (int i=start; i<end; ++i)
-    {
-      Scalar tmp0 = Scalar(0);
-      Packet ptmp0 = ei_pset1(tmp0);
-      const Scalar* lhs0 = lhs + i*lhsStride;
-      // process first unaligned result's coeffs
-      // FIXME this loop get vectorized by the compiler !
-      for (int j=0; j<alignedStart; ++j)
-        tmp0 += rhs[j] * lhs0[j];
-
-      if (alignedSize>alignedStart)
-      {
-        // process aligned rhs coeffs
-        if ((size_t(lhs0+alignedStart)%sizeof(Packet))==0)
-          for (int j = alignedStart;j<alignedSize;j+=PacketSize)
-            ptmp0 = ei_pmadd(ei_pload(&rhs[j]), ei_pload(&lhs0[j]), ptmp0);
-        else
-          for (int j = alignedStart;j<alignedSize;j+=PacketSize)
-            ptmp0 = ei_pmadd(ei_pload(&rhs[j]), ei_ploadu(&lhs0[j]), ptmp0);
-        tmp0 += ei_predux(ptmp0);
-      }
-
-      // process remaining scalars
-      // FIXME this loop get vectorized by the compiler !
-      for (int j=alignedSize; j<size; ++j)
-        tmp0 += rhs[j] * lhs0[j];
-      res[i] += tmp0;
-    }
-    if (skipRows)
-    {
-      start = 0;
-      end = skipRows;
-      skipRows = 0;
-    }
-    else
-      break;
-  } while(PacketSize>1);
-
-  #undef _EIGEN_ACCUMULATE_PACKETS
-}
-
-#endif // EIGEN_CACHE_FRIENDLY_PRODUCT_H

Eigen/src/Core/Coeffs.h

@@ -1,384 +0,0 @@
-// This file is part of Eigen, a lightweight C++ template library
-// for linear algebra. Eigen itself is part of the KDE project.
-//
-// Copyright (C) 2006-2008 Benoit Jacob <jacob.benoit.1@gmail.com>
-//
-// 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_COEFFS_H
-#define EIGEN_COEFFS_H
-
-/** Short version: don't use this function, use
-  * \link operator()(int,int) const \endlink instead.
-  *
-  * Long version: this function is similar to
-  * \link operator()(int,int) const \endlink, but without the assertion.
-  * Use this for limiting the performance cost of debugging code when doing
-  * repeated coefficient access. Only use this when it is guaranteed that the
-  * parameters \a row and \a col are in range.
-  *
-  * If EIGEN_INTERNAL_DEBUGGING is defined, an assertion will be made, making this
-  * function equivalent to \link operator()(int,int) const \endlink.
-  *
-  * \sa operator()(int,int) const, coeffRef(int,int), coeff(int) const
-  */
-template<typename Derived>
-EIGEN_STRONG_INLINE const typename ei_traits<Derived>::Scalar MatrixBase<Derived>
-  ::coeff(int row, int col) const
-{
-  ei_internal_assert(row >= 0 && row < rows()
-                     && col >= 0 && col < cols());
-  return derived().coeff(row, col);
-}
-
-/** \returns the coefficient at given the given row and column.
-  *
-  * \sa operator()(int,int), operator[](int) const
-  */
-template<typename Derived>
-EIGEN_STRONG_INLINE const typename ei_traits<Derived>::Scalar MatrixBase<Derived>
-  ::operator()(int row, int col) const
-{
-  ei_assert(row >= 0 && row < rows()
-      && col >= 0 && col < cols());
-  return derived().coeff(row, col);
-}
-
-/** Short version: don't use this function, use
-  * \link operator()(int,int) \endlink instead.
-  *
-  * Long version: this function is similar to
-  * \link operator()(int,int) \endlink, but without the assertion.
-  * Use this for limiting the performance cost of debugging code when doing
-  * repeated coefficient access. Only use this when it is guaranteed that the
-  * parameters \a row and \a col are in range.
-  *
-  * If EIGEN_INTERNAL_DEBUGGING is defined, an assertion will be made, making this
-  * function equivalent to \link operator()(int,int) \endlink.
-  *
-  * \sa operator()(int,int), coeff(int, int) const, coeffRef(int)
-  */
-template<typename Derived>
-EIGEN_STRONG_INLINE typename ei_traits<Derived>::Scalar& MatrixBase<Derived>
-  ::coeffRef(int row, int col)
-{
-  ei_internal_assert(row >= 0 && row < rows()
-                     && col >= 0 && col < cols());
-  return derived().coeffRef(row, col);
-}
-
-/** \returns a reference to the coefficient at given the given row and column.
-  *
-  * \sa operator()(int,int) const, operator[](int)
-  */
-template<typename Derived>
-EIGEN_STRONG_INLINE typename ei_traits<Derived>::Scalar& MatrixBase<Derived>
-  ::operator()(int row, int col)
-{
-  ei_assert(row >= 0 && row < rows()
-      && col >= 0 && col < cols());
-  return derived().coeffRef(row, col);
-}
-
-/** Short version: don't use this function, use
-  * \link operator[](int) const \endlink instead.
-  *
-  * Long version: this function is similar to
-  * \link operator[](int) const \endlink, but without the assertion.
-  * Use this for limiting the performance cost of debugging code when doing
-  * repeated coefficient access. Only use this when it is guaranteed that the
-  * parameter \a index is in range.
-  *
-  * If EIGEN_INTERNAL_DEBUGGING is defined, an assertion will be made, making this
-  * function equivalent to \link operator[](int) const \endlink.
-  *
-  * \sa operator[](int) const, coeffRef(int), coeff(int,int) const
-  */
-template<typename Derived>
-EIGEN_STRONG_INLINE const typename ei_traits<Derived>::Scalar MatrixBase<Derived>
-  ::coeff(int index) const
-{
-  ei_internal_assert(index >= 0 && index < size());
-  return derived().coeff(index);
-}
-
-/** \returns the coefficient at given index.
-  *
-  * This method is allowed only for vector expressions, and for matrix expressions having the LinearAccessBit.
-  *
-  * \sa operator[](int), operator()(int,int) const, x() const, y() const,
-  * z() const, w() const
-  */
-template<typename Derived>
-EIGEN_STRONG_INLINE const typename ei_traits<Derived>::Scalar MatrixBase<Derived>
-  ::operator[](int index) const
-{
-  ei_assert(index >= 0 && index < size());
-  return derived().coeff(index);
-}
-
-/** \returns the coefficient at given index.
-  *
-  * This is synonymous to operator[](int) const.
-  *
-  * This method is allowed only for vector expressions, and for matrix expressions having the LinearAccessBit.
-  *
-  * \sa operator[](int), operator()(int,int) const, x() const, y() const,
-  * z() const, w() const
-  */
-template<typename Derived>
-EIGEN_STRONG_INLINE const typename ei_traits<Derived>::Scalar MatrixBase<Derived>
-  ::operator()(int index) const
-{
-  ei_assert(index >= 0 && index < size());
-  return derived().coeff(index);
-}
-
-/** Short version: don't use this function, use
-  * \link operator[](int) \endlink instead.
-  *
-  * Long version: this function is similar to
-  * \link operator[](int) \endlink, but without the assertion.
-  * Use this for limiting the performance cost of debugging code when doing
-  * repeated coefficient access. Only use this when it is guaranteed that the
-  * parameters \a row and \a col are in range.
-  *
-  * If EIGEN_INTERNAL_DEBUGGING is defined, an assertion will be made, making this
-  * function equivalent to \link operator[](int) \endlink.
-  *
-  * \sa operator[](int), coeff(int) const, coeffRef(int,int)
-  */
-template<typename Derived>
-EIGEN_STRONG_INLINE typename ei_traits<Derived>::Scalar& MatrixBase<Derived>
-  ::coeffRef(int index)
-{
-  ei_internal_assert(index >= 0 && index < size());
-  return derived().coeffRef(index);
-}
-
-/** \returns a reference to the coefficient at given index.
-  *
-  * This method is allowed only for vector expressions, and for matrix expressions having the LinearAccessBit.
-  *
-  * \sa operator[](int) const, operator()(int,int), x(), y(), z(), w()
-  */
-template<typename Derived>
-EIGEN_STRONG_INLINE typename ei_traits<Derived>::Scalar& MatrixBase<Derived>
-  ::operator[](int index)
-{
-  ei_assert(index >= 0 && index < size());
-  return derived().coeffRef(index);
-}
-
-/** \returns a reference to the coefficient at given index.
-  *
-  * This is synonymous to operator[](int).
-  *
-  * This method is allowed only for vector expressions, and for matrix expressions having the LinearAccessBit.
-  *
-  * \sa operator[](int) const, operator()(int,int), x(), y(), z(), w()
-  */
-template<typename Derived>
-EIGEN_STRONG_INLINE typename ei_traits<Derived>::Scalar& MatrixBase<Derived>
-  ::operator()(int index)
-{
-  ei_assert(index >= 0 && index < size());
-  return derived().coeffRef(index);
-}
-
-/** equivalent to operator[](0).  */
-template<typename Derived>
-EIGEN_STRONG_INLINE const typename ei_traits<Derived>::Scalar MatrixBase<Derived>
-  ::x() const { return (*this)[0]; }
-
-/** equivalent to operator[](1).  */
-template<typename Derived>
-EIGEN_STRONG_INLINE const typename ei_traits<Derived>::Scalar MatrixBase<Derived>
-  ::y() const { return (*this)[1]; }
-
-/** equivalent to operator[](2).  */
-template<typename Derived>
-EIGEN_STRONG_INLINE const typename ei_traits<Derived>::Scalar MatrixBase<Derived>
-  ::z() const { return (*this)[2]; }
-
-/** equivalent to operator[](3).  */
-template<typename Derived>
-EIGEN_STRONG_INLINE const typename ei_traits<Derived>::Scalar MatrixBase<Derived>
-  ::w() const { return (*this)[3]; }
-
-/** equivalent to operator[](0).  */
-template<typename Derived>
-EIGEN_STRONG_INLINE typename ei_traits<Derived>::Scalar& MatrixBase<Derived>
-  ::x() { return (*this)[0]; }
-
-/** equivalent to operator[](1).  */
-template<typename Derived>
-EIGEN_STRONG_INLINE typename ei_traits<Derived>::Scalar& MatrixBase<Derived>
-  ::y() { return (*this)[1]; }
-
-/** equivalent to operator[](2).  */
-template<typename Derived>
-EIGEN_STRONG_INLINE typename ei_traits<Derived>::Scalar& MatrixBase<Derived>
-  ::z() { return (*this)[2]; }
-
-/** equivalent to operator[](3).  */
-template<typename Derived>
-EIGEN_STRONG_INLINE typename ei_traits<Derived>::Scalar& MatrixBase<Derived>
-  ::w() { return (*this)[3]; }
-
-/** \returns the packet of coefficients starting at the given row and column. It is your responsibility
-  * to ensure that a packet really starts there. This method is only available on expressions having the
-  * PacketAccessBit.
-  *
-  * The \a LoadMode parameter may have the value \a Aligned or \a Unaligned. Its effect is to select
-  * the appropriate vectorization instruction. Aligned access is faster, but is only possible for packets
-  * starting at an address which is a multiple of the packet size.
-  */
-template<typename Derived>
-template<int LoadMode>
-EIGEN_STRONG_INLINE typename ei_packet_traits<typename ei_traits<Derived>::Scalar>::type
-MatrixBase<Derived>::packet(int row, int col) const
-{
-  ei_internal_assert(row >= 0 && row < rows()
-                     && col >= 0 && col < cols());
-  return derived().template packet<LoadMode>(row,col);
-}
-
-/** Stores the given packet of coefficients, at the given row and column of this expression. It is your responsibility
-  * to ensure that a packet really starts there. This method is only available on expressions having the
-  * PacketAccessBit.
-  *
-  * The \a LoadMode parameter may have the value \a Aligned or \a Unaligned. Its effect is to select
-  * the appropriate vectorization instruction. Aligned access is faster, but is only possible for packets
-  * starting at an address which is a multiple of the packet size.
-  */
-template<typename Derived>
-template<int StoreMode>
-EIGEN_STRONG_INLINE void MatrixBase<Derived>::writePacket
-(int row, int col, const typename ei_packet_traits<typename ei_traits<Derived>::Scalar>::type& x)
-{
-  ei_internal_assert(row >= 0 && row < rows()
-                     && col >= 0 && col < cols());
-  derived().template writePacket<StoreMode>(row,col,x);
-}
-
-/** \returns the packet of coefficients starting at the given index. It is your responsibility
-  * to ensure that a packet really starts there. This method is only available on expressions having the
-  * PacketAccessBit and the LinearAccessBit.
-  *
-  * The \a LoadMode parameter may have the value \a Aligned or \a Unaligned. Its effect is to select
-  * the appropriate vectorization instruction. Aligned access is faster, but is only possible for packets
-  * starting at an address which is a multiple of the packet size.
-  */
-template<typename Derived>
-template<int LoadMode>
-EIGEN_STRONG_INLINE typename ei_packet_traits<typename ei_traits<Derived>::Scalar>::type
-MatrixBase<Derived>::packet(int index) const
-{
-  ei_internal_assert(index >= 0 && index < size());
-  return derived().template packet<LoadMode>(index);
-}
-
-/** Stores the given packet of coefficients, at the given index in this expression. It is your responsibility
-  * to ensure that a packet really starts there. This method is only available on expressions having the
-  * PacketAccessBit and the LinearAccessBit.
-  *
-  * The \a LoadMode parameter may have the value \a Aligned or \a Unaligned. Its effect is to select
-  * the appropriate vectorization instruction. Aligned access is faster, but is only possible for packets
-  * starting at an address which is a multiple of the packet size.
-  */
-template<typename Derived>
-template<int StoreMode>
-EIGEN_STRONG_INLINE void MatrixBase<Derived>::writePacket
-(int index, const typename ei_packet_traits<typename ei_traits<Derived>::Scalar>::type& x)
-{
-  ei_internal_assert(index >= 0 && index < size());
-  derived().template writePacket<StoreMode>(index,x);
-}
-
-#ifndef EIGEN_PARSED_BY_DOXYGEN
-
-/** \internal Copies the coefficient at position (row,col) of other into *this.
-  *
-  * This method is overridden in SwapWrapper, allowing swap() assignments to share 99% of their code
-  * with usual assignments.
-  *
-  * Outside of this internal usage, this method has probably no usefulness. It is hidden in the public API dox.
-  */
-template<typename Derived>
-template<typename OtherDerived>
-EIGEN_STRONG_INLINE void MatrixBase<Derived>::copyCoeff(int row, int col, const MatrixBase<OtherDerived>& other)
-{
-  ei_internal_assert(row >= 0 && row < rows()
-                     && col >= 0 && col < cols());
-  derived().coeffRef(row, col) = other.derived().coeff(row, col);
-}
-
-/** \internal Copies the coefficient at the given index of other into *this.
-  *
-  * This method is overridden in SwapWrapper, allowing swap() assignments to share 99% of their code
-  * with usual assignments.
-  *
-  * Outside of this internal usage, this method has probably no usefulness. It is hidden in the public API dox.
-  */
-template<typename Derived>
-template<typename OtherDerived>
-EIGEN_STRONG_INLINE void MatrixBase<Derived>::copyCoeff(int index, const MatrixBase<OtherDerived>& other)
-{
-  ei_internal_assert(index >= 0 && index < size());
-  derived().coeffRef(index) = other.derived().coeff(index);
-}
-
-/** \internal Copies the packet at position (row,col) of other into *this.
-  *
-  * This method is overridden in SwapWrapper, allowing swap() assignments to share 99% of their code
-  * with usual assignments.
-  *
-  * Outside of this internal usage, this method has probably no usefulness. It is hidden in the public API dox.
-  */
-template<typename Derived>
-template<typename OtherDerived, int StoreMode, int LoadMode>
-EIGEN_STRONG_INLINE void MatrixBase<Derived>::copyPacket(int row, int col, const MatrixBase<OtherDerived>& other)
-{
-  ei_internal_assert(row >= 0 && row < rows()
-                     && col >= 0 && col < cols());
-  derived().template writePacket<StoreMode>(row, col,
-    other.derived().template packet<LoadMode>(row, col));
-}
-
-/** \internal Copies the packet at the given index of other into *this.
-  *
-  * This method is overridden in SwapWrapper, allowing swap() assignments to share 99% of their code
-  * with usual assignments.
-  *
-  * Outside of this internal usage, this method has probably no usefulness. It is hidden in the public API dox.
-  */
-template<typename Derived>
-template<typename OtherDerived, int StoreMode, int LoadMode>
-EIGEN_STRONG_INLINE void MatrixBase<Derived>::copyPacket(int index, const MatrixBase<OtherDerived>& other)
-{
-  ei_internal_assert(index >= 0 && index < size());
-  derived().template writePacket<StoreMode>(index,
-    other.derived().template packet<LoadMode>(index));
-}
-
-#endif
-
-#endif // EIGEN_COEFFS_H

Eigen/src/Core/CommaInitializer.h

@@ -1,32 +1,20 @@
 // This file is part of Eigen, a lightweight C++ template library
-// for linear algebra. Eigen itself is part of the KDE project.
+// for linear algebra.
 //
-// Copyright (C) 2008 Gael Guennebaud <g.gael@free.fr>
+// Copyright (C) 2008 Gael Guennebaud <gael.guennebaud@inria.fr>
 // Copyright (C) 2006-2008 Benoit Jacob <jacob.benoit.1@gmail.com>
 //
-// 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/>.
+// This Source Code Form is subject to the terms of the Mozilla
+// Public License v. 2.0. If a copy of the MPL was not distributed
+// with this file, You can obtain one at http://mozilla.org/MPL/2.0/.
 
 #ifndef EIGEN_COMMAINITIALIZER_H
 #define EIGEN_COMMAINITIALIZER_H
 
+namespace Eigen { 
+
 /** \class CommaInitializer
+  * \ingroup Core_Module
   *
   * \brief Helper class used by the comma initializer operator
   *
@@ -34,73 +22,88 @@
   * the return type of MatrixBase::operator<<, and most of the time this is the only
   * way it is used.
   *
-  * \sa \ref MatrixBaseCommaInitRef "MatrixBase::operator<<", CommaInitializer::finished()
+  * \sa \blank \ref MatrixBaseCommaInitRef "MatrixBase::operator<<", CommaInitializer::finished()
   */
-template<typename MatrixType>
+template<typename XprType>
 struct CommaInitializer
 {
-  typedef typename ei_traits<MatrixType>::Scalar Scalar;
-  inline CommaInitializer(MatrixType& mat, const Scalar& s)
-    : m_matrix(mat), m_row(0), m_col(1), m_currentBlockRows(1)
+  typedef typename XprType::Scalar Scalar;
+
+  EIGEN_DEVICE_FUNC
+  inline CommaInitializer(XprType& xpr, const Scalar& s)
+    : m_xpr(xpr), m_row(0), m_col(1), m_currentBlockRows(1)
   {
-    m_matrix.coeffRef(0,0) = s;
+    m_xpr.coeffRef(0,0) = s;
   }
 
   template<typename OtherDerived>
-  inline CommaInitializer(MatrixType& mat, const MatrixBase<OtherDerived>& other)
-    : m_matrix(mat), m_row(0), m_col(other.cols()), m_currentBlockRows(other.rows())
+  EIGEN_DEVICE_FUNC
+  inline CommaInitializer(XprType& xpr, const DenseBase<OtherDerived>& other)
+    : m_xpr(xpr), m_row(0), m_col(other.cols()), m_currentBlockRows(other.rows())
   {
-    m_matrix.block(0, 0, other.rows(), other.cols()) = other;
+    m_xpr.block(0, 0, other.rows(), other.cols()) = other;
+  }
+
+  /* Copy/Move constructor which transfers ownership. This is crucial in 
+   * absence of return value optimization to avoid assertions during destruction. */
+  // FIXME in C++11 mode this could be replaced by a proper RValue constructor
+  EIGEN_DEVICE_FUNC
+  inline CommaInitializer(const CommaInitializer& o)
+  : m_xpr(o.m_xpr), m_row(o.m_row), m_col(o.m_col), m_currentBlockRows(o.m_currentBlockRows) {
+    // Mark original object as finished. In absence of R-value references we need to const_cast:
+    const_cast<CommaInitializer&>(o).m_row = m_xpr.rows();
+    const_cast<CommaInitializer&>(o).m_col = m_xpr.cols();
+    const_cast<CommaInitializer&>(o).m_currentBlockRows = 0;
   }
 
   /* inserts a scalar value in the target matrix */
+  EIGEN_DEVICE_FUNC
   CommaInitializer& operator,(const Scalar& s)
   {
-    if (m_col==m_matrix.cols())
+    if (m_col==m_xpr.cols())
     {
       m_row+=m_currentBlockRows;
       m_col = 0;
       m_currentBlockRows = 1;
-      ei_assert(m_row<m_matrix.rows()
+      eigen_assert(m_row<m_xpr.rows()
         && "Too many rows passed to comma initializer (operator<<)");
     }
-    ei_assert(m_col<m_matrix.cols()
+    eigen_assert(m_col<m_xpr.cols()
       && "Too many coefficients passed to comma initializer (operator<<)");
-    ei_assert(m_currentBlockRows==1);
-    m_matrix.coeffRef(m_row, m_col++) = s;
+    eigen_assert(m_currentBlockRows==1);
+    m_xpr.coeffRef(m_row, m_col++) = s;
     return *this;
   }
 
   /* inserts a matrix expression in the target matrix */
   template<typename OtherDerived>
-  CommaInitializer& operator,(const MatrixBase<OtherDerived>& other)
+  EIGEN_DEVICE_FUNC
+  CommaInitializer& operator,(const DenseBase<OtherDerived>& other)
   {
-    if (m_col==m_matrix.cols())
+    if (m_col==m_xpr.cols() && (other.cols()!=0 || other.rows()!=m_currentBlockRows))
     {
       m_row+=m_currentBlockRows;
       m_col = 0;
       m_currentBlockRows = other.rows();
-      ei_assert(m_row+m_currentBlockRows<=m_matrix.rows()
+      eigen_assert(m_row+m_currentBlockRows<=m_xpr.rows()
         && "Too many rows passed to comma initializer (operator<<)");
     }
-    ei_assert(m_col<m_matrix.cols()
+    eigen_assert((m_col + other.cols() <= m_xpr.cols())
       && "Too many coefficients passed to comma initializer (operator<<)");
-    ei_assert(m_currentBlockRows==other.rows());
-    if (OtherDerived::SizeAtCompileTime != Dynamic)
-      m_matrix.template block<OtherDerived::RowsAtCompileTime != Dynamic ? OtherDerived::RowsAtCompileTime : 1,
-                              OtherDerived::ColsAtCompileTime != Dynamic ? OtherDerived::ColsAtCompileTime : 1>
-                    (m_row, m_col) = other;
-    else
-      m_matrix.block(m_row, m_col, other.rows(), other.cols()) = other;
+    eigen_assert(m_currentBlockRows==other.rows());
+    m_xpr.template block<OtherDerived::RowsAtCompileTime, OtherDerived::ColsAtCompileTime>
+                    (m_row, m_col, other.rows(), other.cols()) = other;
     m_col += other.cols();
     return *this;
   }
 
+  EIGEN_DEVICE_FUNC
   inline ~CommaInitializer()
+#if defined VERIFY_RAISES_ASSERT && (!defined EIGEN_NO_ASSERTION_CHECKING) && defined EIGEN_EXCEPTIONS
+  EIGEN_EXCEPTION_SPEC(Eigen::eigen_assert_exception)
+#endif
   {
-    ei_assert((m_row+m_currentBlockRows) == m_matrix.rows()
-         && m_col == m_matrix.cols()
-         && "Too few coefficients passed to comma initializer (operator<<)");
+      finished();
   }
 
   /** \returns the built matrix once all its coefficients have been set.
@@ -110,15 +113,18 @@ struct CommaInitializer
     * quaternion.fromRotationMatrix((Matrix3f() << axis0, axis1, axis2).finished());
     * \endcode
     */
-  inline MatrixType& finished() { return m_matrix; }
+  EIGEN_DEVICE_FUNC
+  inline XprType& finished() {
+      eigen_assert(((m_row+m_currentBlockRows) == m_xpr.rows() || m_xpr.cols() == 0)
+           && m_col == m_xpr.cols()
+           && "Too few coefficients passed to comma initializer (operator<<)");
+      return m_xpr;
+  }
 
-  MatrixType& m_matrix;   // target matrix
-  int m_row;              // current row id
-  int m_col;              // current col id
-  int m_currentBlockRows; // current block height
-
-private:
-  CommaInitializer& operator=(const CommaInitializer&);
+  XprType& m_xpr;           // target expression
+  Index m_row;              // current row id
+  Index m_col;              // current col id
+  Index m_currentBlockRows; // current block height
 };
 
 /** \anchor MatrixBaseCommaInitRef
@@ -127,15 +133,15 @@ private:
   * The coefficients must be provided in a row major order and exactly match
   * the size of the matrix. Otherwise an assertion is raised.
   *
-  * \addexample CommaInit \label How to easily set all the coefficients of a matrix
-  *
   * Example: \include MatrixBase_set.cpp
   * Output: \verbinclude MatrixBase_set.out
+  * 
+  * \note According the c++ standard, the argument expressions of this comma initializer are evaluated in arbitrary order.
   *
   * \sa CommaInitializer::finished(), class CommaInitializer
   */
 template<typename Derived>
-inline CommaInitializer<Derived> MatrixBase<Derived>::operator<< (const Scalar& s)
+inline CommaInitializer<Derived> DenseBase<Derived>::operator<< (const Scalar& s)
 {
   return CommaInitializer<Derived>(*static_cast<Derived*>(this), s);
 }
@@ -144,9 +150,11 @@ inline CommaInitializer<Derived> MatrixBase<Derived>::operator<< (const Scalar&
 template<typename Derived>
 template<typename OtherDerived>
 inline CommaInitializer<Derived>
-MatrixBase<Derived>::operator<<(const MatrixBase<OtherDerived>& other)
+DenseBase<Derived>::operator<<(const DenseBase<OtherDerived>& other)
 {
   return CommaInitializer<Derived>(*static_cast<Derived *>(this), other);
 }
 
+} // end namespace Eigen
+
 #endif // EIGEN_COMMAINITIALIZER_H

Eigen/src/Core/ConditionEstimator.h

@@ -0,0 +1,175 @@
+// This file is part of Eigen, a lightweight C++ template library
+// for linear algebra.
+//
+// Copyright (C) 2016 Rasmus Munk Larsen (rmlarsen@google.com)
+//
+// This Source Code Form is subject to the terms of the Mozilla
+// Public License v. 2.0. If a copy of the MPL was not distributed
+// with this file, You can obtain one at http://mozilla.org/MPL/2.0/.
+
+#ifndef EIGEN_CONDITIONESTIMATOR_H
+#define EIGEN_CONDITIONESTIMATOR_H
+
+namespace Eigen {
+
+namespace internal {
+
+template <typename Vector, typename RealVector, bool IsComplex>
+struct rcond_compute_sign {
+  static inline Vector run(const Vector& v) {
+    const RealVector v_abs = v.cwiseAbs();
+    return (v_abs.array() == static_cast<typename Vector::RealScalar>(0))
+            .select(Vector::Ones(v.size()), v.cwiseQuotient(v_abs));
+  }
+};
+
+// Partial specialization to avoid elementwise division for real vectors.
+template <typename Vector>
+struct rcond_compute_sign<Vector, Vector, false> {
+  static inline Vector run(const Vector& v) {
+    return (v.array() < static_cast<typename Vector::RealScalar>(0))
+           .select(-Vector::Ones(v.size()), Vector::Ones(v.size()));
+  }
+};
+
+/**
+  * \returns an estimate of ||inv(matrix)||_1 given a decomposition of
+  * \a matrix that implements .solve() and .adjoint().solve() methods.
+  *
+  * This function implements Algorithms 4.1 and 5.1 from
+  *   http://www.maths.manchester.ac.uk/~higham/narep/narep135.pdf
+  * which also forms the basis for the condition number estimators in
+  * LAPACK. Since at most 10 calls to the solve method of dec are
+  * performed, the total cost is O(dims^2), as opposed to O(dims^3)
+  * needed to compute the inverse matrix explicitly.
+  *
+  * The most common usage is in estimating the condition number
+  * ||matrix||_1 * ||inv(matrix)||_1. The first term ||matrix||_1 can be
+  * computed directly in O(n^2) operations.
+  *
+  * Supports the following decompositions: FullPivLU, PartialPivLU, LDLT, and
+  * LLT.
+  *
+  * \sa FullPivLU, PartialPivLU, LDLT, LLT.
+  */
+template <typename Decomposition>
+typename Decomposition::RealScalar rcond_invmatrix_L1_norm_estimate(const Decomposition& dec)
+{
+  typedef typename Decomposition::MatrixType MatrixType;
+  typedef typename Decomposition::Scalar Scalar;
+  typedef typename Decomposition::RealScalar RealScalar;
+  typedef typename internal::plain_col_type<MatrixType>::type Vector;
+  typedef typename internal::plain_col_type<MatrixType, RealScalar>::type RealVector;
+  const bool is_complex = (NumTraits<Scalar>::IsComplex != 0);
+
+  eigen_assert(dec.rows() == dec.cols());
+  const Index n = dec.rows();
+  if (n == 0)
+    return 0;
+
+  // Disable Index to float conversion warning
+#ifdef __INTEL_COMPILER
+  #pragma warning push
+  #pragma warning ( disable : 2259 )
+#endif
+  Vector v = dec.solve(Vector::Ones(n) / Scalar(n));
+#ifdef __INTEL_COMPILER
+  #pragma warning pop
+#endif
+
+  // lower_bound is a lower bound on
+  //   ||inv(matrix)||_1  = sup_v ||inv(matrix) v||_1 / ||v||_1
+  // and is the objective maximized by the ("super-") gradient ascent
+  // algorithm below.
+  RealScalar lower_bound = v.template lpNorm<1>();
+  if (n == 1)
+    return lower_bound;
+
+  // Gradient ascent algorithm follows: We know that the optimum is achieved at
+  // one of the simplices v = e_i, so in each iteration we follow a
+  // super-gradient to move towards the optimal one.
+  RealScalar old_lower_bound = lower_bound;
+  Vector sign_vector(n);
+  Vector old_sign_vector;
+  Index v_max_abs_index = -1;
+  Index old_v_max_abs_index = v_max_abs_index;
+  for (int k = 0; k < 4; ++k)
+  {
+    sign_vector = internal::rcond_compute_sign<Vector, RealVector, is_complex>::run(v);
+    if (k > 0 && !is_complex && sign_vector == old_sign_vector) {
+      // Break if the solution stagnated.
+      break;
+    }
+    // v_max_abs_index = argmax |real( inv(matrix)^T * sign_vector )|
+    v = dec.adjoint().solve(sign_vector);
+    v.real().cwiseAbs().maxCoeff(&v_max_abs_index);
+    if (v_max_abs_index == old_v_max_abs_index) {
+      // Break if the solution stagnated.
+      break;
+    }
+    // Move to the new simplex e_j, where j = v_max_abs_index.
+    v = dec.solve(Vector::Unit(n, v_max_abs_index));  // v = inv(matrix) * e_j.
+    lower_bound = v.template lpNorm<1>();
+    if (lower_bound <= old_lower_bound) {
+      // Break if the gradient step did not increase the lower_bound.
+      break;
+    }
+    if (!is_complex) {
+      old_sign_vector = sign_vector;
+    }
+    old_v_max_abs_index = v_max_abs_index;
+    old_lower_bound = lower_bound;
+  }
+  // The following calculates an independent estimate of ||matrix||_1 by
+  // multiplying matrix by a vector with entries of slowly increasing
+  // magnitude and alternating sign:
+  //   v_i = (-1)^{i} (1 + (i / (dim-1))), i = 0,...,dim-1.
+  // This improvement to Hager's algorithm above is due to Higham. It was
+  // added to make the algorithm more robust in certain corner cases where
+  // large elements in the matrix might otherwise escape detection due to
+  // exact cancellation (especially when op and op_adjoint correspond to a
+  // sequence of backsubstitutions and permutations), which could cause
+  // Hager's algorithm to vastly underestimate ||matrix||_1.
+  Scalar alternating_sign(RealScalar(1));
+  for (Index i = 0; i < n; ++i) {
+    // The static_cast is needed when Scalar is a complex and RealScalar implements expression templates
+    v[i] = alternating_sign * static_cast<RealScalar>(RealScalar(1) + (RealScalar(i) / (RealScalar(n - 1))));
+    alternating_sign = -alternating_sign;
+  }
+  v = dec.solve(v);
+  const RealScalar alternate_lower_bound = (2 * v.template lpNorm<1>()) / (3 * RealScalar(n));
+  return numext::maxi(lower_bound, alternate_lower_bound);
+}
+
+/** \brief Reciprocal condition number estimator.
+  *
+  * Computing a decomposition of a dense matrix takes O(n^3) operations, while
+  * this method estimates the condition number quickly and reliably in O(n^2)
+  * operations.
+  *
+  * \returns an estimate of the reciprocal condition number
+  * (1 / (||matrix||_1 * ||inv(matrix)||_1)) of matrix, given ||matrix||_1 and
+  * its decomposition. Supports the following decompositions: FullPivLU,
+  * PartialPivLU, LDLT, and LLT.
+  *
+  * \sa FullPivLU, PartialPivLU, LDLT, LLT.
+  */
+template <typename Decomposition>
+typename Decomposition::RealScalar
+rcond_estimate_helper(typename Decomposition::RealScalar matrix_norm, const Decomposition& dec)
+{
+  typedef typename Decomposition::RealScalar RealScalar;
+  eigen_assert(dec.rows() == dec.cols());
+  if (dec.rows() == 0)              return NumTraits<RealScalar>::infinity();
+  if (matrix_norm == RealScalar(0)) return RealScalar(0);
+  if (dec.rows() == 1)              return RealScalar(1);
+  const RealScalar inverse_matrix_norm = rcond_invmatrix_L1_norm_estimate(dec);
+  return (inverse_matrix_norm == RealScalar(0) ? RealScalar(0)
+                                               : (RealScalar(1) / inverse_matrix_norm) / matrix_norm);
+}
+
+}  // namespace internal
+
+}  // namespace Eigen
+
+#endif

Eigen/src/Core/CoreInstantiations.cpp

@@ -1,47 +0,0 @@
-// This file is part of Eigen, a lightweight C++ template library
-// for linear algebra. Eigen itself is part of the KDE project.
-//
-// Copyright (C) 2008 Gael Guennebaud <g.gael@free.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/>.
-
-#ifdef EIGEN_EXTERN_INSTANTIATIONS
-#undef EIGEN_EXTERN_INSTANTIATIONS
-#endif
-
-#include "../../Core"
-
-namespace Eigen
-{
-
-#define EIGEN_INSTANTIATE_PRODUCT(TYPE) \
-template static void ei_cache_friendly_product<TYPE>( \
-  int _rows, int _cols, int depth, \
-  bool _lhsRowMajor, const TYPE* _lhs, int _lhsStride, \
-  bool _rhsRowMajor, const TYPE* _rhs, int _rhsStride, \
-  bool resRowMajor, TYPE* res, int resStride)
-
-EIGEN_INSTANTIATE_PRODUCT(float);
-EIGEN_INSTANTIATE_PRODUCT(double);
-EIGEN_INSTANTIATE_PRODUCT(int);
-EIGEN_INSTANTIATE_PRODUCT(std::complex<float>);
-EIGEN_INSTANTIATE_PRODUCT(std::complex<double>);
-
-}

Eigen/src/Core/CoreIterators.h

@@ -0,0 +1,127 @@
+// This file is part of Eigen, a lightweight C++ template library
+// for linear algebra.
+//
+// Copyright (C) 2008-2014 Gael Guennebaud <gael.guennebaud@inria.fr>
+//
+// This Source Code Form is subject to the terms of the Mozilla
+// Public License v. 2.0. If a copy of the MPL was not distributed
+// with this file, You can obtain one at http://mozilla.org/MPL/2.0/.
+
+#ifndef EIGEN_COREITERATORS_H
+#define EIGEN_COREITERATORS_H
+
+namespace Eigen { 
+
+/* This file contains the respective InnerIterator definition of the expressions defined in Eigen/Core
+ */
+
+namespace internal {
+
+template<typename XprType, typename EvaluatorKind>
+class inner_iterator_selector;
+
+}
+
+/** \class InnerIterator
+  * \brief An InnerIterator allows to loop over the element of any matrix expression.
+  * 
+  * \warning To be used with care because an evaluator is constructed every time an InnerIterator iterator is constructed.
+  * 
+  * TODO: add a usage example
+  */
+template<typename XprType>
+class InnerIterator
+{
+protected:
+  typedef internal::inner_iterator_selector<XprType, typename internal::evaluator_traits<XprType>::Kind> IteratorType;
+  typedef internal::evaluator<XprType> EvaluatorType;
+  typedef typename internal::traits<XprType>::Scalar Scalar;
+public:
+  /** Construct an iterator over the \a outerId -th row or column of \a xpr */
+  InnerIterator(const XprType &xpr, const Index &outerId)
+    : m_eval(xpr), m_iter(m_eval, outerId, xpr.innerSize())
+  {}
+  
+  /// \returns the value of the current coefficient.
+  EIGEN_STRONG_INLINE Scalar value() const          { return m_iter.value(); }
+  /** Increment the iterator \c *this to the next non-zero coefficient.
+    * Explicit zeros are not skipped over. To skip explicit zeros, see class SparseView
+    */
+  EIGEN_STRONG_INLINE InnerIterator& operator++()   { m_iter.operator++(); return *this; }
+  /// \returns the column or row index of the current coefficient.
+  EIGEN_STRONG_INLINE Index index() const           { return m_iter.index(); }
+  /// \returns the row index of the current coefficient.
+  EIGEN_STRONG_INLINE Index row() const             { return m_iter.row(); }
+  /// \returns the column index of the current coefficient.
+  EIGEN_STRONG_INLINE Index col() const             { return m_iter.col(); }
+  /// \returns \c true if the iterator \c *this still references a valid coefficient.
+  EIGEN_STRONG_INLINE operator bool() const         { return m_iter; }
+  
+protected:
+  EvaluatorType m_eval;
+  IteratorType m_iter;
+private:
+  // If you get here, then you're not using the right InnerIterator type, e.g.:
+  //   SparseMatrix<double,RowMajor> A;
+  //   SparseMatrix<double>::InnerIterator it(A,0);
+  template<typename T> InnerIterator(const EigenBase<T>&,Index outer);
+};
+
+namespace internal {
+
+// Generic inner iterator implementation for dense objects
+template<typename XprType>
+class inner_iterator_selector<XprType, IndexBased>
+{
+protected:
+  typedef evaluator<XprType> EvaluatorType;
+  typedef typename traits<XprType>::Scalar Scalar;
+  enum { IsRowMajor = (XprType::Flags&RowMajorBit)==RowMajorBit };
+  
+public:
+  EIGEN_STRONG_INLINE inner_iterator_selector(const EvaluatorType &eval, const Index &outerId, const Index &innerSize)
+    : m_eval(eval), m_inner(0), m_outer(outerId), m_end(innerSize)
+  {}
+
+  EIGEN_STRONG_INLINE Scalar value() const
+  {
+    return (IsRowMajor) ? m_eval.coeff(m_outer, m_inner)
+                        : m_eval.coeff(m_inner, m_outer);
+  }
+
+  EIGEN_STRONG_INLINE inner_iterator_selector& operator++() { m_inner++; return *this; }
+
+  EIGEN_STRONG_INLINE Index index() const { return m_inner; }
+  inline Index row() const { return IsRowMajor ? m_outer : index(); }
+  inline Index col() const { return IsRowMajor ? index() : m_outer; }
+
+  EIGEN_STRONG_INLINE operator bool() const { return m_inner < m_end && m_inner>=0; }
+
+protected:
+  const EvaluatorType& m_eval;
+  Index m_inner;
+  const Index m_outer;
+  const Index m_end;
+};
+
+// For iterator-based evaluator, inner-iterator is already implemented as
+// evaluator<>::InnerIterator
+template<typename XprType>
+class inner_iterator_selector<XprType, IteratorBased>
+ : public evaluator<XprType>::InnerIterator
+{
+protected:
+  typedef typename evaluator<XprType>::InnerIterator Base;
+  typedef evaluator<XprType> EvaluatorType;
+  
+public:
+  EIGEN_STRONG_INLINE inner_iterator_selector(const EvaluatorType &eval, const Index &outerId, const Index &/*innerSize*/)
+    : Base(eval, outerId)
+  {}  
+};
+
+} // end namespace internal
+
+} // end namespace Eigen
+
+#endif // EIGEN_COREITERATORS_H

Eigen/src/Core/Cwise.h

@@ -1,214 +0,0 @@
-// This file is part of Eigen, a lightweight C++ template library
-// for linear algebra. Eigen itself is part of the KDE project.
-//
-// Copyright (C) 2008 Gael Guennebaud <g.gael@free.fr>
-// Copyright (C) 2008 Benoit Jacob <jacob.benoit.1@gmail.com>
-//
-// 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_CWISE_H
-#define EIGEN_CWISE_H
-
-/** \internal
-  * convenient macro to defined the return type of a cwise binary operation */
-#define EIGEN_CWISE_BINOP_RETURN_TYPE(OP) \
-    CwiseBinaryOp<OP<typename ei_traits<ExpressionType>::Scalar>, ExpressionType, OtherDerived>
-
-#define EIGEN_CWISE_PRODUCT_RETURN_TYPE \
-    CwiseBinaryOp< \
-      ei_scalar_product_op< \
-        typename ei_scalar_product_traits< \
-          typename ei_traits<ExpressionType>::Scalar, \
-          typename ei_traits<OtherDerived>::Scalar \
-        >::ReturnType \
-      >, \
-      ExpressionType, \
-      OtherDerived \
-    >
-
-/** \internal
-  * convenient macro to defined the return type of a cwise unary operation */
-#define EIGEN_CWISE_UNOP_RETURN_TYPE(OP) \
-    CwiseUnaryOp<OP<typename ei_traits<ExpressionType>::Scalar>, ExpressionType>
-
-/** \internal
-  * convenient macro to defined the return type of a cwise comparison to a scalar */
-#define EIGEN_CWISE_COMP_TO_SCALAR_RETURN_TYPE(OP) \
-    CwiseBinaryOp<OP<typename ei_traits<ExpressionType>::Scalar>, ExpressionType, \
-        NestByValue<typename ExpressionType::ConstantReturnType> >
-
-/** \class Cwise
-  *
-  * \brief Pseudo expression providing additional coefficient-wise operations
-  *
-  * \param ExpressionType the type of the object on which to do coefficient-wise operations
-  *
-  * This class represents an expression with additional coefficient-wise features.
-  * It is the return type of MatrixBase::cwise()
-  * and most of the time this is the only way it is used.
-  *
-  * Note that some methods are defined in the \ref Array module.
-  *
-  * Example: \include MatrixBase_cwise_const.cpp
-  * Output: \verbinclude MatrixBase_cwise_const.out
-  *
-  * \sa MatrixBase::cwise() const, MatrixBase::cwise()
-  */
-template<typename ExpressionType> class Cwise
-{
-  public:
-
-    typedef typename ei_traits<ExpressionType>::Scalar Scalar;
-    typedef typename ei_meta_if<ei_must_nest_by_value<ExpressionType>::ret,
-        ExpressionType, const ExpressionType&>::ret ExpressionTypeNested;
-    typedef CwiseUnaryOp<ei_scalar_add_op<Scalar>, ExpressionType> ScalarAddReturnType;
-
-    inline Cwise(const ExpressionType& matrix) : m_matrix(matrix) {}
-
-    /** \internal */
-    inline const ExpressionType& _expression() const { return m_matrix; }
-
-    template<typename OtherDerived>
-    const EIGEN_CWISE_PRODUCT_RETURN_TYPE
-    operator*(const MatrixBase<OtherDerived> &other) const;
-
-    template<typename OtherDerived>
-    const EIGEN_CWISE_BINOP_RETURN_TYPE(ei_scalar_quotient_op)
-    operator/(const MatrixBase<OtherDerived> &other) const;
-
-    template<typename OtherDerived>
-    const EIGEN_CWISE_BINOP_RETURN_TYPE(ei_scalar_min_op)
-    min(const MatrixBase<OtherDerived> &other) const;
-
-    template<typename OtherDerived>
-    const EIGEN_CWISE_BINOP_RETURN_TYPE(ei_scalar_max_op)
-    max(const MatrixBase<OtherDerived> &other) const;
-
-    const EIGEN_CWISE_UNOP_RETURN_TYPE(ei_scalar_abs_op)      abs() const;
-    const EIGEN_CWISE_UNOP_RETURN_TYPE(ei_scalar_abs2_op)     abs2() const;
-    const EIGEN_CWISE_UNOP_RETURN_TYPE(ei_scalar_square_op)   square() const;
-    const EIGEN_CWISE_UNOP_RETURN_TYPE(ei_scalar_cube_op)     cube() const;
-    const EIGEN_CWISE_UNOP_RETURN_TYPE(ei_scalar_inverse_op)  inverse() const;
-    const EIGEN_CWISE_UNOP_RETURN_TYPE(ei_scalar_sqrt_op)     sqrt() const;
-    const EIGEN_CWISE_UNOP_RETURN_TYPE(ei_scalar_exp_op)      exp() const;
-    const EIGEN_CWISE_UNOP_RETURN_TYPE(ei_scalar_log_op)      log() const;
-    const EIGEN_CWISE_UNOP_RETURN_TYPE(ei_scalar_cos_op)      cos() const;
-    const EIGEN_CWISE_UNOP_RETURN_TYPE(ei_scalar_sin_op)      sin() const;
-    const EIGEN_CWISE_UNOP_RETURN_TYPE(ei_scalar_pow_op)      pow(const Scalar& exponent) const;
-
-    const ScalarAddReturnType
-    operator+(const Scalar& scalar) const;
-
-    /** \relates Cwise */
-    friend const ScalarAddReturnType
-    operator+(const Scalar& scalar, const Cwise& mat)
-    { return mat + scalar; }
-
-    ExpressionType& operator+=(const Scalar& scalar);
-
-    const ScalarAddReturnType
-    operator-(const Scalar& scalar) const;
-
-    ExpressionType& operator-=(const Scalar& scalar);
-
-    template<typename OtherDerived>
-    inline ExpressionType& operator*=(const MatrixBase<OtherDerived> &other);
-
-    template<typename OtherDerived>
-    inline ExpressionType& operator/=(const MatrixBase<OtherDerived> &other);
-
-    template<typename OtherDerived> const EIGEN_CWISE_BINOP_RETURN_TYPE(std::less)
-    operator<(const MatrixBase<OtherDerived>& other) const;
-
-    template<typename OtherDerived> const EIGEN_CWISE_BINOP_RETURN_TYPE(std::less_equal)
-    operator<=(const MatrixBase<OtherDerived>& other) const;
-
-    template<typename OtherDerived> const EIGEN_CWISE_BINOP_RETURN_TYPE(std::greater)
-    operator>(const MatrixBase<OtherDerived>& other) const;
-
-    template<typename OtherDerived> const EIGEN_CWISE_BINOP_RETURN_TYPE(std::greater_equal)
-    operator>=(const MatrixBase<OtherDerived>& other) const;
-
-    template<typename OtherDerived> const EIGEN_CWISE_BINOP_RETURN_TYPE(std::equal_to)
-    operator==(const MatrixBase<OtherDerived>& other) const;
-
-    template<typename OtherDerived> const EIGEN_CWISE_BINOP_RETURN_TYPE(std::not_equal_to)
-    operator!=(const MatrixBase<OtherDerived>& other) const;
-
-    // comparisons to a scalar value
-    const EIGEN_CWISE_COMP_TO_SCALAR_RETURN_TYPE(std::less)
-    operator<(Scalar s) const;
-
-    const EIGEN_CWISE_COMP_TO_SCALAR_RETURN_TYPE(std::less_equal)
-    operator<=(Scalar s) const;
-
-    const EIGEN_CWISE_COMP_TO_SCALAR_RETURN_TYPE(std::greater)
-    operator>(Scalar s) const;
-
-    const EIGEN_CWISE_COMP_TO_SCALAR_RETURN_TYPE(std::greater_equal)
-    operator>=(Scalar s) const;
-
-    const EIGEN_CWISE_COMP_TO_SCALAR_RETURN_TYPE(std::equal_to)
-    operator==(Scalar s) const;
-
-    const EIGEN_CWISE_COMP_TO_SCALAR_RETURN_TYPE(std::not_equal_to)
-    operator!=(Scalar s) const;
-
-    // allow to extend Cwise outside Eigen
-    #ifdef EIGEN_CWISE_PLUGIN
-    #include EIGEN_CWISE_PLUGIN
-    #endif
-
-  protected:
-    ExpressionTypeNested m_matrix;
-
-  private:
-    Cwise& operator=(const Cwise&);
-};
-
-/** \returns a Cwise wrapper of *this providing additional coefficient-wise operations
-  *
-  * Example: \include MatrixBase_cwise_const.cpp
-  * Output: \verbinclude MatrixBase_cwise_const.out
-  *
-  * \sa class Cwise, cwise()
-  */
-template<typename Derived>
-inline const Cwise<Derived>
-MatrixBase<Derived>::cwise() const
-{
-  return derived();
-}
-
-/** \returns a Cwise wrapper of *this providing additional coefficient-wise operations
-  *
-  * Example: \include MatrixBase_cwise.cpp
-  * Output: \verbinclude MatrixBase_cwise.out
-  *
-  * \sa class Cwise, cwise() const
-  */
-template<typename Derived>
-inline Cwise<Derived>
-MatrixBase<Derived>::cwise()
-{
-  return derived();
-}
-
-#endif // EIGEN_CWISE_H

Eigen/src/Core/CwiseBinaryOp.h

@@ -1,156 +1,156 @@
 // This file is part of Eigen, a lightweight C++ template library
-// for linear algebra. Eigen itself is part of the KDE project.
+// for linear algebra.
 //
-// Copyright (C) 2008 Gael Guennebaud <g.gael@free.fr>
+// Copyright (C) 2008-2014 Gael Guennebaud <gael.guennebaud@inria.fr>
 // Copyright (C) 2006-2008 Benoit Jacob <jacob.benoit.1@gmail.com>
 //
-// 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/>.
+// This Source Code Form is subject to the terms of the Mozilla
+// Public License v. 2.0. If a copy of the MPL was not distributed
+// with this file, You can obtain one at http://mozilla.org/MPL/2.0/.
 
 #ifndef EIGEN_CWISE_BINARY_OP_H
 #define EIGEN_CWISE_BINARY_OP_H
 
+namespace Eigen {
+
+namespace internal {
+template<typename BinaryOp, typename Lhs, typename Rhs>
+struct traits<CwiseBinaryOp<BinaryOp, Lhs, Rhs> >
+{
+  // we must not inherit from traits<Lhs> since it has
+  // the potential to cause problems with MSVC
+  typedef typename remove_all<Lhs>::type Ancestor;
+  typedef typename traits<Ancestor>::XprKind XprKind;
+  enum {
+    RowsAtCompileTime = traits<Ancestor>::RowsAtCompileTime,
+    ColsAtCompileTime = traits<Ancestor>::ColsAtCompileTime,
+    MaxRowsAtCompileTime = traits<Ancestor>::MaxRowsAtCompileTime,
+    MaxColsAtCompileTime = traits<Ancestor>::MaxColsAtCompileTime
+  };
+
+  // even though we require Lhs and Rhs to have the same scalar type (see CwiseBinaryOp constructor),
+  // we still want to handle the case when the result type is different.
+  typedef typename result_of<
+                     BinaryOp(
+                       const typename Lhs::Scalar&,
+                       const typename Rhs::Scalar&
+                     )
+                   >::type Scalar;
+  typedef typename cwise_promote_storage_type<typename traits<Lhs>::StorageKind,
+                                              typename traits<Rhs>::StorageKind,
+                                              BinaryOp>::ret StorageKind;
+  typedef typename promote_index_type<typename traits<Lhs>::StorageIndex,
+                                      typename traits<Rhs>::StorageIndex>::type StorageIndex;
+  typedef typename Lhs::Nested LhsNested;
+  typedef typename Rhs::Nested RhsNested;
+  typedef typename remove_reference<LhsNested>::type _LhsNested;
+  typedef typename remove_reference<RhsNested>::type _RhsNested;
+  enum {
+    Flags = cwise_promote_storage_order<typename traits<Lhs>::StorageKind,typename traits<Rhs>::StorageKind,_LhsNested::Flags & RowMajorBit,_RhsNested::Flags & RowMajorBit>::value
+  };
+};
+} // end namespace internal
+
+template<typename BinaryOp, typename Lhs, typename Rhs, typename StorageKind>
+class CwiseBinaryOpImpl;
+
 /** \class CwiseBinaryOp
+  * \ingroup Core_Module
   *
-  * \brief Generic expression of a coefficient-wise operator between two matrices or vectors
+  * \brief Generic expression where a coefficient-wise binary operator is applied to two expressions
   *
-  * \param BinaryOp template functor implementing the operator
-  * \param Lhs the type of the left-hand side
-  * \param Rhs the type of the right-hand side
+  * \tparam BinaryOp template functor implementing the operator
+  * \tparam LhsType the type of the left-hand side
+  * \tparam RhsType the type of the right-hand side
   *
-  * This class represents an expression of a generic binary operator of two matrices or vectors.
-  * It is the return type of the operator+, operator-, and the Cwise methods, and most
-  * of the time this is the only way it is used.
+  * This class represents an expression  where a coefficient-wise binary operator is applied to two expressions.
+  * It is the return type of binary operators, by which we mean only those binary operators where
+  * both the left-hand side and the right-hand side are Eigen expressions.
+  * For example, the return type of matrix1+matrix2 is a CwiseBinaryOp.
   *
-  * However, if you want to write a function returning such an expression, you
-  * will need to use this class.
+  * Most of the time, this is the only way that it is used, so you typically don't have to name
+  * CwiseBinaryOp types explicitly.
   *
   * \sa MatrixBase::binaryExpr(const MatrixBase<OtherDerived> &,const CustomBinaryOp &) const, class CwiseUnaryOp, class CwiseNullaryOp
   */
-template<typename BinaryOp, typename Lhs, typename Rhs>
-struct ei_traits<CwiseBinaryOp<BinaryOp, Lhs, Rhs> >
-{
-  // even though we require Lhs and Rhs to have the same scalar type (see CwiseBinaryOp constructor),
-  // we still want to handle the case when the result type is different.
-  typedef typename ei_result_of<
-                     BinaryOp(
-                       typename Lhs::Scalar,
-                       typename Rhs::Scalar
-                     )
-                   >::type Scalar;
-  typedef typename Lhs::Nested LhsNested;
-  typedef typename Rhs::Nested RhsNested;
-  typedef typename ei_unref<LhsNested>::type _LhsNested;
-  typedef typename ei_unref<RhsNested>::type _RhsNested;
-  enum {
-    LhsCoeffReadCost = _LhsNested::CoeffReadCost,
-    RhsCoeffReadCost = _RhsNested::CoeffReadCost,
-    LhsFlags = _LhsNested::Flags,
-    RhsFlags = _RhsNested::Flags,
-    RowsAtCompileTime = Lhs::RowsAtCompileTime,
-    ColsAtCompileTime = Lhs::ColsAtCompileTime,
-    MaxRowsAtCompileTime = Lhs::MaxRowsAtCompileTime,
-    MaxColsAtCompileTime = Lhs::MaxColsAtCompileTime,
-    Flags = (int(LhsFlags) | int(RhsFlags)) & (
-        HereditaryBits
-      | (int(LhsFlags) & int(RhsFlags) & (LinearAccessBit | AlignedBit))
-      | (ei_functor_traits<BinaryOp>::PacketAccess && ((int(LhsFlags) & RowMajorBit)==(int(RhsFlags) & RowMajorBit))
-        ? (int(LhsFlags) & int(RhsFlags) & PacketAccessBit) : 0)),
-    CoeffReadCost = LhsCoeffReadCost + RhsCoeffReadCost + ei_functor_traits<BinaryOp>::Cost
-  };
-};
-
-template<typename BinaryOp, typename Lhs, typename Rhs>
-class CwiseBinaryOp : ei_no_assignment_operator,
-  public MatrixBase<CwiseBinaryOp<BinaryOp, Lhs, Rhs> >
+template<typename BinaryOp, typename LhsType, typename RhsType>
+class CwiseBinaryOp : 
+  public CwiseBinaryOpImpl<
+          BinaryOp, LhsType, RhsType,
+          typename internal::cwise_promote_storage_type<typename internal::traits<LhsType>::StorageKind,
+                                                        typename internal::traits<RhsType>::StorageKind,
+                                                        BinaryOp>::ret>,
+  internal::no_assignment_operator
 {
   public:
+    
+    typedef typename internal::remove_all<BinaryOp>::type Functor;
+    typedef typename internal::remove_all<LhsType>::type Lhs;
+    typedef typename internal::remove_all<RhsType>::type Rhs;
 
+    typedef typename CwiseBinaryOpImpl<
+        BinaryOp, LhsType, RhsType,
+        typename internal::cwise_promote_storage_type<typename internal::traits<LhsType>::StorageKind,
+                                                      typename internal::traits<Rhs>::StorageKind,
+                                                      BinaryOp>::ret>::Base Base;
     EIGEN_GENERIC_PUBLIC_INTERFACE(CwiseBinaryOp)
-    typedef typename ei_traits<CwiseBinaryOp>::LhsNested LhsNested;
-    typedef typename ei_traits<CwiseBinaryOp>::RhsNested RhsNested;
 
-    EIGEN_STRONG_INLINE CwiseBinaryOp(const Lhs& lhs, const Rhs& rhs, const BinaryOp& func = BinaryOp())
-      : m_lhs(lhs), m_rhs(rhs), m_functor(func)
+    typedef typename internal::ref_selector<LhsType>::type LhsNested;
+    typedef typename internal::ref_selector<RhsType>::type RhsNested;
+    typedef typename internal::remove_reference<LhsNested>::type _LhsNested;
+    typedef typename internal::remove_reference<RhsNested>::type _RhsNested;
+
+    EIGEN_DEVICE_FUNC
+    EIGEN_STRONG_INLINE CwiseBinaryOp(const Lhs& aLhs, const Rhs& aRhs, const BinaryOp& func = BinaryOp())
+      : m_lhs(aLhs), m_rhs(aRhs), m_functor(func)
     {
-      // we require Lhs and Rhs to have the same scalar type. Currently there is no example of a binary functor
-      // that would take two operands of different types. If there were such an example, then this check should be
-      // moved to the BinaryOp functors, on a per-case basis. This would however require a change in the BinaryOp functors, as
-      // currently they take only one typename Scalar template parameter.
-      // It is tempting to always allow mixing different types but remember that this is often impossible in the vectorized paths.
-      // So allowing mixing different types gives very unexpected errors when enabling vectorization, when the user tries to
-      // add together a float matrix and a double matrix.
-      EIGEN_STATIC_ASSERT((ei_functor_allows_mixing_real_and_complex<BinaryOp>::ret
-                           ? int(ei_is_same_type<typename Lhs::RealScalar, typename Rhs::RealScalar>::ret)
-                           : int(ei_is_same_type<typename Lhs::Scalar, typename Rhs::Scalar>::ret)),
-        YOU_MIXED_DIFFERENT_NUMERIC_TYPES__YOU_NEED_TO_USE_THE_CAST_METHOD_OF_MATRIXBASE_TO_CAST_NUMERIC_TYPES_EXPLICITLY)
+      EIGEN_CHECK_BINARY_COMPATIBILIY(BinaryOp,typename Lhs::Scalar,typename Rhs::Scalar);
       // require the sizes to match
       EIGEN_STATIC_ASSERT_SAME_MATRIX_SIZE(Lhs, Rhs)
-      ei_assert(lhs.rows() == rhs.rows() && lhs.cols() == rhs.cols());
+      eigen_assert(aLhs.rows() == aRhs.rows() && aLhs.cols() == aRhs.cols());
     }
 
-    EIGEN_STRONG_INLINE int rows() const { return m_lhs.rows(); }
-    EIGEN_STRONG_INLINE int cols() const { return m_lhs.cols(); }
-
-    EIGEN_STRONG_INLINE const Scalar coeff(int row, int col) const
-    {
-      return m_functor(m_lhs.coeff(row, col), m_rhs.coeff(row, col));
+    EIGEN_DEVICE_FUNC
+    EIGEN_STRONG_INLINE Index rows() const {
+      // return the fixed size type if available to enable compile time optimizations
+      if (internal::traits<typename internal::remove_all<LhsNested>::type>::RowsAtCompileTime==Dynamic)
+        return m_rhs.rows();
+      else
+        return m_lhs.rows();
+    }
+    EIGEN_DEVICE_FUNC
+    EIGEN_STRONG_INLINE Index cols() const {
+      // return the fixed size type if available to enable compile time optimizations
+      if (internal::traits<typename internal::remove_all<LhsNested>::type>::ColsAtCompileTime==Dynamic)
+        return m_rhs.cols();
+      else
+        return m_lhs.cols();
     }
 
-    template<int LoadMode>
-    EIGEN_STRONG_INLINE PacketScalar packet(int row, int col) const
-    {
-      return m_functor.packetOp(m_lhs.template packet<LoadMode>(row, col), m_rhs.template packet<LoadMode>(row, col));
-    }
-
-    EIGEN_STRONG_INLINE const Scalar coeff(int index) const
-    {
-      return m_functor(m_lhs.coeff(index), m_rhs.coeff(index));
-    }
-
-    template<int LoadMode>
-    EIGEN_STRONG_INLINE PacketScalar packet(int index) const
-    {
-      return m_functor.packetOp(m_lhs.template packet<LoadMode>(index), m_rhs.template packet<LoadMode>(index));
-    }
+    /** \returns the left hand side nested expression */
+    EIGEN_DEVICE_FUNC
+    const _LhsNested& lhs() const { return m_lhs; }
+    /** \returns the right hand side nested expression */
+    EIGEN_DEVICE_FUNC
+    const _RhsNested& rhs() const { return m_rhs; }
+    /** \returns the functor representing the binary operation */
+    EIGEN_DEVICE_FUNC
+    const BinaryOp& functor() const { return m_functor; }
 
   protected:
-    const LhsNested m_lhs;
-    const RhsNested m_rhs;
+    LhsNested m_lhs;
+    RhsNested m_rhs;
     const BinaryOp m_functor;
 };
 
-/**\returns an expression of the difference of \c *this and \a other
-  *
-  * \note If you want to substract a given scalar from all coefficients, see Cwise::operator-().
-  *
-  * \sa class CwiseBinaryOp, MatrixBase::operator-=(), Cwise::operator-()
-  */
-template<typename Derived>
-template<typename OtherDerived>
-EIGEN_STRONG_INLINE const CwiseBinaryOp<ei_scalar_difference_op<typename ei_traits<Derived>::Scalar>,
-                                 Derived, OtherDerived>
-MatrixBase<Derived>::operator-(const MatrixBase<OtherDerived> &other) const
+// Generic API dispatcher
+template<typename BinaryOp, typename Lhs, typename Rhs, typename StorageKind>
+class CwiseBinaryOpImpl
+  : public internal::generic_xpr_base<CwiseBinaryOp<BinaryOp, Lhs, Rhs> >::type
 {
-  return CwiseBinaryOp<ei_scalar_difference_op<Scalar>,
-                       Derived, OtherDerived>(derived(), other.derived());
-}
+public:
+  typedef typename internal::generic_xpr_base<CwiseBinaryOp<BinaryOp, Lhs, Rhs> >::type Base;
+};
 
 /** replaces \c *this by \c *this - \a other.
   *
@@ -161,23 +161,8 @@ template<typename OtherDerived>
 EIGEN_STRONG_INLINE Derived &
 MatrixBase<Derived>::operator-=(const MatrixBase<OtherDerived> &other)
 {
-  return *this = *this - other;
-}
-
-/** \relates MatrixBase
-  *
-  * \returns an expression of the sum of \c *this and \a other
-  *
-  * \note If you want to add a given scalar to all coefficients, see Cwise::operator+().
-  *
-  * \sa class CwiseBinaryOp, MatrixBase::operator+=(), Cwise::operator+()
-  */
-template<typename Derived>
-template<typename OtherDerived>
-EIGEN_STRONG_INLINE const CwiseBinaryOp<ei_scalar_sum_op<typename ei_traits<Derived>::Scalar>, Derived, OtherDerived>
-MatrixBase<Derived>::operator+(const MatrixBase<OtherDerived> &other) const
-{
-  return CwiseBinaryOp<ei_scalar_sum_op<Scalar>, Derived, OtherDerived>(derived(), other.derived());
+  call_assignment(derived(), other.derived(), internal::sub_assign_op<Scalar,typename OtherDerived::Scalar>());
+  return derived();
 }
 
 /** replaces \c *this by \c *this + \a other.
@@ -189,116 +174,11 @@ template<typename OtherDerived>
 EIGEN_STRONG_INLINE Derived &
 MatrixBase<Derived>::operator+=(const MatrixBase<OtherDerived>& other)
 {
-  return *this = *this + other;
+  call_assignment(derived(), other.derived(), internal::add_assign_op<Scalar,typename OtherDerived::Scalar>());
+  return derived();
 }
 
-/** \returns an expression of the Schur product (coefficient wise product) of *this and \a other
-  *
-  * Example: \include Cwise_product.cpp
-  * Output: \verbinclude Cwise_product.out
-  *
-  * \sa class CwiseBinaryOp, operator/(), square()
-  */
-template<typename ExpressionType>
-template<typename OtherDerived>
-EIGEN_STRONG_INLINE const EIGEN_CWISE_PRODUCT_RETURN_TYPE
-Cwise<ExpressionType>::operator*(const MatrixBase<OtherDerived> &other) const
-{
-  return EIGEN_CWISE_PRODUCT_RETURN_TYPE(_expression(), other.derived());
-}
-
-/** \returns an expression of the coefficient-wise quotient of *this and \a other
-  *
-  * Example: \include Cwise_quotient.cpp
-  * Output: \verbinclude Cwise_quotient.out
-  *
-  * \sa class CwiseBinaryOp, operator*(), inverse()
-  */
-template<typename ExpressionType>
-template<typename OtherDerived>
-EIGEN_STRONG_INLINE const EIGEN_CWISE_BINOP_RETURN_TYPE(ei_scalar_quotient_op)
-Cwise<ExpressionType>::operator/(const MatrixBase<OtherDerived> &other) const
-{
-  return EIGEN_CWISE_BINOP_RETURN_TYPE(ei_scalar_quotient_op)(_expression(), other.derived());
-}
-
-/** Replaces this expression by its coefficient-wise product with \a other.
-  *
-  * Example: \include Cwise_times_equal.cpp
-  * Output: \verbinclude Cwise_times_equal.out
-  *
-  * \sa operator*(), operator/=()
-  */
-template<typename ExpressionType>
-template<typename OtherDerived>
-inline ExpressionType& Cwise<ExpressionType>::operator*=(const MatrixBase<OtherDerived> &other)
-{
-  return m_matrix.const_cast_derived() = *this * other;
-}
-
-/** Replaces this expression by its coefficient-wise quotient by \a other.
-  *
-  * Example: \include Cwise_slash_equal.cpp
-  * Output: \verbinclude Cwise_slash_equal.out
-  *
-  * \sa operator/(), operator*=()
-  */
-template<typename ExpressionType>
-template<typename OtherDerived>
-inline ExpressionType& Cwise<ExpressionType>::operator/=(const MatrixBase<OtherDerived> &other)
-{
-  return m_matrix.const_cast_derived() = *this / other;
-}
-
-/** \returns an expression of the coefficient-wise min of *this and \a other
-  *
-  * Example: \include Cwise_min.cpp
-  * Output: \verbinclude Cwise_min.out
-  *
-  * \sa class CwiseBinaryOp
-  */
-template<typename ExpressionType>
-template<typename OtherDerived>
-EIGEN_STRONG_INLINE const EIGEN_CWISE_BINOP_RETURN_TYPE(ei_scalar_min_op)
-Cwise<ExpressionType>::min(const MatrixBase<OtherDerived> &other) const
-{
-  return EIGEN_CWISE_BINOP_RETURN_TYPE(ei_scalar_min_op)(_expression(), other.derived());
-}
-
-/** \returns an expression of the coefficient-wise max of *this and \a other
-  *
-  * Example: \include Cwise_max.cpp
-  * Output: \verbinclude Cwise_max.out
-  *
-  * \sa class CwiseBinaryOp
-  */
-template<typename ExpressionType>
-template<typename OtherDerived>
-EIGEN_STRONG_INLINE const EIGEN_CWISE_BINOP_RETURN_TYPE(ei_scalar_max_op)
-Cwise<ExpressionType>::max(const MatrixBase<OtherDerived> &other) const
-{
-  return EIGEN_CWISE_BINOP_RETURN_TYPE(ei_scalar_max_op)(_expression(), other.derived());
-}
-
-/** \returns an expression of a custom coefficient-wise operator \a func of *this and \a other
-  *
-  * The template parameter \a CustomBinaryOp is the type of the functor
-  * of the custom operator (see class CwiseBinaryOp for an example)
-  *
-  * \addexample CustomCwiseBinaryFunctors \label How to use custom coeff wise binary functors
-  *
-  * Here is an example illustrating the use of custom functors:
-  * \include class_CwiseBinaryOp.cpp
-  * Output: \verbinclude class_CwiseBinaryOp.out
-  *
-  * \sa class CwiseBinaryOp, MatrixBase::operator+, MatrixBase::operator-, Cwise::operator*, Cwise::operator/
-  */
-template<typename Derived>
-template<typename CustomBinaryOp, typename OtherDerived>
-EIGEN_STRONG_INLINE const CwiseBinaryOp<CustomBinaryOp, Derived, OtherDerived>
-MatrixBase<Derived>::binaryExpr(const MatrixBase<OtherDerived> &other, const CustomBinaryOp& func) const
-{
-  return CwiseBinaryOp<CustomBinaryOp, Derived, OtherDerived>(derived(), other.derived(), func);
-}
+} // end namespace Eigen
 
 #endif // EIGEN_CWISE_BINARY_OP_H
+

Eigen/src/Core/CwiseNullaryOp.h

@@ -1,106 +1,91 @@
 // This file is part of Eigen, a lightweight C++ template library
-// for linear algebra. Eigen itself is part of the KDE project.
+// for linear algebra.
 //
-// Copyright (C) 2008 Gael Guennebaud <g.gael@free.fr>
+// 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/>.
+// This Source Code Form is subject to the terms of the Mozilla
+// Public License v. 2.0. If a copy of the MPL was not distributed
+// with this file, You can obtain one at http://mozilla.org/MPL/2.0/.
 
 #ifndef EIGEN_CWISE_NULLARY_OP_H
 #define EIGEN_CWISE_NULLARY_OP_H
 
+namespace Eigen {
+
+namespace internal {
+template<typename NullaryOp, typename PlainObjectType>
+struct traits<CwiseNullaryOp<NullaryOp, PlainObjectType> > : traits<PlainObjectType>
+{
+  enum {
+    Flags = traits<PlainObjectType>::Flags & RowMajorBit
+  };
+};
+
+} // namespace internal
+
 /** \class CwiseNullaryOp
+  * \ingroup Core_Module
   *
   * \brief Generic expression of a matrix where all coefficients are defined by a functor
   *
-  * \param NullaryOp template functor implementing the operator
+  * \tparam NullaryOp template functor implementing the operator
+  * \tparam PlainObjectType the underlying plain matrix/array type
   *
   * This class represents an expression of a generic nullary operator.
-  * It is the return type of the Ones(), Zero(), Constant(), Identity() and Random() functions,
+  * It is the return type of the Ones(), Zero(), Constant(), Identity() and Random() methods,
   * and most of the time this is the only way it is used.
   *
   * However, if you want to write a function returning such an expression, you
   * will need to use this class.
   *
-  * \sa class CwiseUnaryOp, class CwiseBinaryOp, MatrixBase::NullaryExpr()
+  * The functor NullaryOp must expose one of the following method:
+    <table class="manual">
+    <tr            ><td>\c operator()() </td><td>if the procedural generation does not depend on the coefficient entries (e.g., random numbers)</td></tr>
+    <tr class="alt"><td>\c operator()(Index i)</td><td>if the procedural generation makes sense for vectors only and that it depends on the coefficient index \c i (e.g., linspace) </td></tr>
+    <tr            ><td>\c operator()(Index i,Index j)</td><td>if the procedural generation depends on the matrix coordinates \c i, \c j (e.g., to generate a checkerboard with 0 and 1)</td></tr>
+    </table>
+  * It is also possible to expose the last two operators if the generation makes sense for matrices but can be optimized for vectors.
+  *
+  * See DenseBase::NullaryExpr(Index,const CustomNullaryOp&) for an example binding
+  * C++11 random number generators.
+  *
+  * A nullary expression can also be used to implement custom sophisticated matrix manipulations
+  * that cannot be covered by the existing set of natively supported matrix manipulations.
+  * See this \ref TopicCustomizing_NullaryExpr "page" for some examples and additional explanations
+  * on the behavior of CwiseNullaryOp.
+  *
+  * \sa class CwiseUnaryOp, class CwiseBinaryOp, DenseBase::NullaryExpr
   */
-template<typename NullaryOp, typename MatrixType>
-struct ei_traits<CwiseNullaryOp<NullaryOp, MatrixType> > : ei_traits<MatrixType>
-{
-  enum {
-    Flags = (ei_traits<MatrixType>::Flags
-      & (  HereditaryBits
-         | (ei_functor_has_linear_access<NullaryOp>::ret ? LinearAccessBit : 0)
-         | (ei_functor_traits<NullaryOp>::PacketAccess ? PacketAccessBit : 0)))
-      | (ei_functor_traits<NullaryOp>::IsRepeatable ? 0 : EvalBeforeNestingBit),
-    CoeffReadCost = ei_functor_traits<NullaryOp>::Cost
-  };
-};
-
-template<typename NullaryOp, typename MatrixType>
-class CwiseNullaryOp : ei_no_assignment_operator,
-  public MatrixBase<CwiseNullaryOp<NullaryOp, MatrixType> >
+template<typename NullaryOp, typename PlainObjectType>
+class CwiseNullaryOp : public internal::dense_xpr_base< CwiseNullaryOp<NullaryOp, PlainObjectType> >::type, internal::no_assignment_operator
 {
   public:
 
-    EIGEN_GENERIC_PUBLIC_INTERFACE(CwiseNullaryOp)
+    typedef typename internal::dense_xpr_base<CwiseNullaryOp>::type Base;
+    EIGEN_DENSE_PUBLIC_INTERFACE(CwiseNullaryOp)
 
-    CwiseNullaryOp(int rows, int cols, const NullaryOp& func = NullaryOp())
+    EIGEN_DEVICE_FUNC
+    CwiseNullaryOp(Index rows, Index cols, const NullaryOp& func = NullaryOp())
       : m_rows(rows), m_cols(cols), m_functor(func)
     {
-      ei_assert(rows > 0
-          && (RowsAtCompileTime == Dynamic || RowsAtCompileTime == rows)
-          && cols > 0
-          && (ColsAtCompileTime == Dynamic || ColsAtCompileTime == cols));
+      eigen_assert(rows >= 0
+            && (RowsAtCompileTime == Dynamic || RowsAtCompileTime == rows)
+            &&  cols >= 0
+            && (ColsAtCompileTime == Dynamic || ColsAtCompileTime == cols));
     }
 
-    EIGEN_STRONG_INLINE int rows() const { return m_rows.value(); }
-    EIGEN_STRONG_INLINE int cols() const { return m_cols.value(); }
+    EIGEN_DEVICE_FUNC
+    EIGEN_STRONG_INLINE Index rows() const { return m_rows.value(); }
+    EIGEN_DEVICE_FUNC
+    EIGEN_STRONG_INLINE Index cols() const { return m_cols.value(); }
 
-    EIGEN_STRONG_INLINE const Scalar coeff(int rows, int cols) const
-    {
-      return m_functor(rows, cols);
-    }
-
-    template<int LoadMode>
-    EIGEN_STRONG_INLINE PacketScalar packet(int, int) const
-    {
-      return m_functor.packetOp();
-    }
-
-    EIGEN_STRONG_INLINE const Scalar coeff(int index) const
-    {
-      if(RowsAtCompileTime == 1)
-        return m_functor(0, index);
-      else
-        return m_functor(index, 0);
-    }
-
-    template<int LoadMode>
-    EIGEN_STRONG_INLINE PacketScalar packet(int) const
-    {
-      return m_functor.packetOp();
-    }
+    /** \returns the functor representing the nullary operation */
+    EIGEN_DEVICE_FUNC
+    const NullaryOp& functor() const { return m_functor; }
 
   protected:
-    const ei_int_if_dynamic<RowsAtCompileTime> m_rows;
-    const ei_int_if_dynamic<ColsAtCompileTime> m_cols;
+    const internal::variable_if_dynamic<Index, RowsAtCompileTime> m_rows;
+    const internal::variable_if_dynamic<Index, ColsAtCompileTime> m_cols;
     const NullaryOp m_functor;
 };
 
@@ -120,10 +105,10 @@ class CwiseNullaryOp : ei_no_assignment_operator,
   */
 template<typename Derived>
 template<typename CustomNullaryOp>
-EIGEN_STRONG_INLINE const CwiseNullaryOp<CustomNullaryOp, Derived>
-MatrixBase<Derived>::NullaryExpr(int rows, int cols, const CustomNullaryOp& func)
+EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE const CwiseNullaryOp<CustomNullaryOp, typename DenseBase<Derived>::PlainObject>
+DenseBase<Derived>::NullaryExpr(Index rows, Index cols, const CustomNullaryOp& func)
 {
-  return CwiseNullaryOp<CustomNullaryOp, Derived>(rows, cols, func);
+  return CwiseNullaryOp<CustomNullaryOp, PlainObject>(rows, cols, func);
 }
 
 /** \returns an expression of a matrix defined by a custom functor \a func
@@ -139,22 +124,24 @@ MatrixBase<Derived>::NullaryExpr(int rows, int cols, const CustomNullaryOp& func
   *
   * The template parameter \a CustomNullaryOp is the type of the functor.
   *
+  * Here is an example with C++11 random generators: \include random_cpp11.cpp
+  * Output: \verbinclude random_cpp11.out
+  * 
   * \sa class CwiseNullaryOp
   */
 template<typename Derived>
 template<typename CustomNullaryOp>
-EIGEN_STRONG_INLINE const CwiseNullaryOp<CustomNullaryOp, Derived>
-MatrixBase<Derived>::NullaryExpr(int size, const CustomNullaryOp& func)
+EIGEN_STRONG_INLINE const CwiseNullaryOp<CustomNullaryOp, typename DenseBase<Derived>::PlainObject>
+DenseBase<Derived>::NullaryExpr(Index size, const CustomNullaryOp& func)
 {
   EIGEN_STATIC_ASSERT_VECTOR_ONLY(Derived)
-  ei_assert(IsVectorAtCompileTime);
-  if(RowsAtCompileTime == 1) return CwiseNullaryOp<CustomNullaryOp, Derived>(1, size, func);
-  else return CwiseNullaryOp<CustomNullaryOp, Derived>(size, 1, func);
+  if(RowsAtCompileTime == 1) return CwiseNullaryOp<CustomNullaryOp, PlainObject>(1, size, func);
+  else return CwiseNullaryOp<CustomNullaryOp, PlainObject>(size, 1, func);
 }
 
 /** \returns an expression of a matrix defined by a custom functor \a func
   *
-  * This variant is only for fixed-size MatrixBase types. For dynamic-size types, you
+  * This variant is only for fixed-size DenseBase types. For dynamic-size types, you
   * need to use the variants taking size arguments.
   *
   * The template parameter \a CustomNullaryOp is the type of the functor.
@@ -163,16 +150,16 @@ MatrixBase<Derived>::NullaryExpr(int size, const CustomNullaryOp& func)
   */
 template<typename Derived>
 template<typename CustomNullaryOp>
-EIGEN_STRONG_INLINE const CwiseNullaryOp<CustomNullaryOp, Derived>
-MatrixBase<Derived>::NullaryExpr(const CustomNullaryOp& func)
+EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE const CwiseNullaryOp<CustomNullaryOp, typename DenseBase<Derived>::PlainObject>
+DenseBase<Derived>::NullaryExpr(const CustomNullaryOp& func)
 {
-  return CwiseNullaryOp<CustomNullaryOp, Derived>(RowsAtCompileTime, ColsAtCompileTime, func);
+  return CwiseNullaryOp<CustomNullaryOp, PlainObject>(RowsAtCompileTime, ColsAtCompileTime, func);
 }
 
 /** \returns an expression of a constant matrix of value \a value
   *
   * The parameters \a rows and \a cols are the number of rows and of columns of
-  * the returned matrix. Must be compatible with this MatrixBase type.
+  * the returned matrix. Must be compatible with this DenseBase type.
   *
   * This variant is meant to be used for dynamic-size matrix types. For fixed-size types,
   * it is redundant to pass \a rows and \a cols as arguments, so Zero() should be used
@@ -183,16 +170,16 @@ MatrixBase<Derived>::NullaryExpr(const CustomNullaryOp& func)
   * \sa class CwiseNullaryOp
   */
 template<typename Derived>
-EIGEN_STRONG_INLINE const typename MatrixBase<Derived>::ConstantReturnType
-MatrixBase<Derived>::Constant(int rows, int cols, const Scalar& value)
+EIGEN_STRONG_INLINE const typename DenseBase<Derived>::ConstantReturnType
+DenseBase<Derived>::Constant(Index rows, Index cols, const Scalar& value)
 {
-  return NullaryExpr(rows, cols, ei_scalar_constant_op<Scalar>(value));
+  return DenseBase<Derived>::NullaryExpr(rows, cols, internal::scalar_constant_op<Scalar>(value));
 }
 
 /** \returns an expression of a constant matrix of value \a value
   *
   * The parameter \a size is the size of the returned vector.
-  * Must be compatible with this MatrixBase type.
+  * Must be compatible with this DenseBase type.
   *
   * \only_for_vectors
   *
@@ -205,15 +192,15 @@ MatrixBase<Derived>::Constant(int rows, int cols, const Scalar& value)
   * \sa class CwiseNullaryOp
   */
 template<typename Derived>
-EIGEN_STRONG_INLINE const typename MatrixBase<Derived>::ConstantReturnType
-MatrixBase<Derived>::Constant(int size, const Scalar& value)
+EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE const typename DenseBase<Derived>::ConstantReturnType
+DenseBase<Derived>::Constant(Index size, const Scalar& value)
 {
-  return NullaryExpr(size, ei_scalar_constant_op<Scalar>(value));
+  return DenseBase<Derived>::NullaryExpr(size, internal::scalar_constant_op<Scalar>(value));
 }
 
 /** \returns an expression of a constant matrix of value \a value
   *
-  * This variant is only for fixed-size MatrixBase types. For dynamic-size types, you
+  * This variant is only for fixed-size DenseBase types. For dynamic-size types, you
   * need to use the variants taking size arguments.
   *
   * The template parameter \a CustomNullaryOp is the type of the functor.
@@ -221,21 +208,91 @@ MatrixBase<Derived>::Constant(int size, const Scalar& value)
   * \sa class CwiseNullaryOp
   */
 template<typename Derived>
-EIGEN_STRONG_INLINE const typename MatrixBase<Derived>::ConstantReturnType
-MatrixBase<Derived>::Constant(const Scalar& value)
+EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE const typename DenseBase<Derived>::ConstantReturnType
+DenseBase<Derived>::Constant(const Scalar& value)
 {
   EIGEN_STATIC_ASSERT_FIXED_SIZE(Derived)
-  return NullaryExpr(RowsAtCompileTime, ColsAtCompileTime, ei_scalar_constant_op<Scalar>(value));
+  return DenseBase<Derived>::NullaryExpr(RowsAtCompileTime, ColsAtCompileTime, internal::scalar_constant_op<Scalar>(value));
 }
 
-/** \returns true if all coefficients in this matrix are approximately equal to \a value, to within precision \a prec */
+/** \deprecated because of accuracy loss. In Eigen 3.3, it is an alias for LinSpaced(Index,const Scalar&,const Scalar&)
+  *
+  * \sa LinSpaced(Index,Scalar,Scalar), setLinSpaced(Index,const Scalar&,const Scalar&)
+  */
 template<typename Derived>
-bool MatrixBase<Derived>::isApproxToConstant
-(const Scalar& value, RealScalar prec) const
+EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE const typename DenseBase<Derived>::RandomAccessLinSpacedReturnType
+DenseBase<Derived>::LinSpaced(Sequential_t, Index size, const Scalar& low, const Scalar& high)
 {
-  for(int j = 0; j < cols(); ++j)
-    for(int i = 0; i < rows(); ++i)
-      if(!ei_isApprox(coeff(i, j), value, prec))
+  EIGEN_STATIC_ASSERT_VECTOR_ONLY(Derived)
+  return DenseBase<Derived>::NullaryExpr(size, internal::linspaced_op<Scalar,PacketScalar>(low,high,size));
+}
+
+/** \deprecated because of accuracy loss. In Eigen 3.3, it is an alias for LinSpaced(const Scalar&,const Scalar&)
+  *
+  * \sa LinSpaced(Scalar,Scalar)
+  */
+template<typename Derived>
+EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE const typename DenseBase<Derived>::RandomAccessLinSpacedReturnType
+DenseBase<Derived>::LinSpaced(Sequential_t, const Scalar& low, const Scalar& high)
+{
+  EIGEN_STATIC_ASSERT_VECTOR_ONLY(Derived)
+  EIGEN_STATIC_ASSERT_FIXED_SIZE(Derived)
+  return DenseBase<Derived>::NullaryExpr(Derived::SizeAtCompileTime, internal::linspaced_op<Scalar,PacketScalar>(low,high,Derived::SizeAtCompileTime));
+}
+
+/**
+  * \brief Sets a linearly spaced vector.
+  *
+  * The function generates 'size' equally spaced values in the closed interval [low,high].
+  * When size is set to 1, a vector of length 1 containing 'high' is returned.
+  *
+  * \only_for_vectors
+  *
+  * Example: \include DenseBase_LinSpaced.cpp
+  * Output: \verbinclude DenseBase_LinSpaced.out
+  *
+  * For integer scalar types, an even spacing is possible if and only if the length of the range,
+  * i.e., \c high-low is a scalar multiple of \c size-1, or if \c size is a scalar multiple of the
+  * number of values \c high-low+1 (meaning each value can be repeated the same number of time).
+  * If one of these two considions is not satisfied, then \c high is lowered to the largest value
+  * satisfying one of this constraint.
+  * Here are some examples:
+  *
+  * Example: \include DenseBase_LinSpacedInt.cpp
+  * Output: \verbinclude DenseBase_LinSpacedInt.out
+  *
+  * \sa setLinSpaced(Index,const Scalar&,const Scalar&), CwiseNullaryOp
+  */
+template<typename Derived>
+EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE const typename DenseBase<Derived>::RandomAccessLinSpacedReturnType
+DenseBase<Derived>::LinSpaced(Index size, const Scalar& low, const Scalar& high)
+{
+  EIGEN_STATIC_ASSERT_VECTOR_ONLY(Derived)
+  return DenseBase<Derived>::NullaryExpr(size, internal::linspaced_op<Scalar,PacketScalar>(low,high,size));
+}
+
+/**
+  * \copydoc DenseBase::LinSpaced(Index, const Scalar&, const Scalar&)
+  * Special version for fixed size types which does not require the size parameter.
+  */
+template<typename Derived>
+EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE const typename DenseBase<Derived>::RandomAccessLinSpacedReturnType
+DenseBase<Derived>::LinSpaced(const Scalar& low, const Scalar& high)
+{
+  EIGEN_STATIC_ASSERT_VECTOR_ONLY(Derived)
+  EIGEN_STATIC_ASSERT_FIXED_SIZE(Derived)
+  return DenseBase<Derived>::NullaryExpr(Derived::SizeAtCompileTime, internal::linspaced_op<Scalar,PacketScalar>(low,high,Derived::SizeAtCompileTime));
+}
+
+/** \returns true if all coefficients in this matrix are approximately equal to \a val, to within precision \a prec */
+template<typename Derived>
+EIGEN_DEVICE_FUNC bool DenseBase<Derived>::isApproxToConstant
+(const Scalar& val, const RealScalar& prec) const
+{
+  typename internal::nested_eval<Derived,1>::type self(derived());
+  for(Index j = 0; j < cols(); ++j)
+    for(Index i = 0; i < rows(); ++i)
+      if(!internal::isApprox(self.coeff(i, j), val, prec))
         return false;
   return true;
 }
@@ -244,67 +301,110 @@ bool MatrixBase<Derived>::isApproxToConstant
   *
   * \returns true if all coefficients in this matrix are approximately equal to \a value, to within precision \a prec */
 template<typename Derived>
-bool MatrixBase<Derived>::isConstant
-(const Scalar& value, RealScalar prec) const
+EIGEN_DEVICE_FUNC bool DenseBase<Derived>::isConstant
+(const Scalar& val, const RealScalar& prec) const
 {
-  return isApproxToConstant(value, prec);
+  return isApproxToConstant(val, prec);
 }
 
-/** Alias for setConstant(): sets all coefficients in this expression to \a value.
+/** Alias for setConstant(): sets all coefficients in this expression to \a val.
   *
   * \sa setConstant(), Constant(), class CwiseNullaryOp
   */
 template<typename Derived>
-EIGEN_STRONG_INLINE void MatrixBase<Derived>::fill(const Scalar& value)
+EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE void DenseBase<Derived>::fill(const Scalar& val)
 {
-  setConstant(value);
+  setConstant(val);
 }
 
-/** Sets all coefficients in this expression to \a value.
+/** Sets all coefficients in this expression to value \a val.
   *
-  * \sa fill(), setConstant(int,const Scalar&), setConstant(int,int,const Scalar&), setZero(), setOnes(), Constant(), class CwiseNullaryOp, setZero(), setOnes()
+  * \sa fill(), setConstant(Index,const Scalar&), setConstant(Index,Index,const Scalar&), setZero(), setOnes(), Constant(), class CwiseNullaryOp, setZero(), setOnes()
   */
 template<typename Derived>
-EIGEN_STRONG_INLINE Derived& MatrixBase<Derived>::setConstant(const Scalar& value)
+EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE Derived& DenseBase<Derived>::setConstant(const Scalar& val)
 {
-  return derived() = Constant(rows(), cols(), value);
+  return derived() = Constant(rows(), cols(), val);
 }
 
-/** Resizes to the given \a size, and sets all coefficients in this expression to the given \a value.
+/** Resizes to the given \a size, and sets all coefficients in this expression to the given value \a val.
   *
   * \only_for_vectors
   *
-  * Example: \include Matrix_set_int.cpp
+  * Example: \include Matrix_setConstant_int.cpp
   * Output: \verbinclude Matrix_setConstant_int.out
   *
-  * \sa MatrixBase::setConstant(const Scalar&), setConstant(int,int,const Scalar&), class CwiseNullaryOp, MatrixBase::Constant(const Scalar&)
+  * \sa MatrixBase::setConstant(const Scalar&), setConstant(Index,Index,const Scalar&), class CwiseNullaryOp, MatrixBase::Constant(const Scalar&)
   */
-template<typename _Scalar, int _Rows, int _Cols, int _Options, int _MaxRows, int _MaxCols>
-EIGEN_STRONG_INLINE Matrix<_Scalar, _Rows, _Cols, _Options, _MaxRows, _MaxCols>&
-Matrix<_Scalar, _Rows, _Cols, _Options, _MaxRows, _MaxCols>::setConstant(int size, const Scalar& value)
+template<typename Derived>
+EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE Derived&
+PlainObjectBase<Derived>::setConstant(Index size, const Scalar& val)
 {
   resize(size);
-  return setConstant(value);
+  return setConstant(val);
 }
 
-/** Resizes to the given size, and sets all coefficients in this expression to the given \a value.
+/** Resizes to the given size, and sets all coefficients in this expression to the given value \a val.
   *
   * \param rows the new number of rows
   * \param cols the new number of columns
+  * \param val the value to which all coefficients are set
   *
   * Example: \include Matrix_setConstant_int_int.cpp
   * Output: \verbinclude Matrix_setConstant_int_int.out
   *
-  * \sa MatrixBase::setConstant(const Scalar&), setConstant(int,const Scalar&), class CwiseNullaryOp, MatrixBase::Constant(const Scalar&)
+  * \sa MatrixBase::setConstant(const Scalar&), setConstant(Index,const Scalar&), class CwiseNullaryOp, MatrixBase::Constant(const Scalar&)
   */
-template<typename _Scalar, int _Rows, int _Cols, int _Options, int _MaxRows, int _MaxCols>
-EIGEN_STRONG_INLINE Matrix<_Scalar, _Rows, _Cols, _Options, _MaxRows, _MaxCols>&
-Matrix<_Scalar, _Rows, _Cols, _Options, _MaxRows, _MaxCols>::setConstant(int rows, int cols, const Scalar& value)
+template<typename Derived>
+EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE Derived&
+PlainObjectBase<Derived>::setConstant(Index rows, Index cols, const Scalar& val)
 {
   resize(rows, cols);
-  return setConstant(value);
+  return setConstant(val);
 }
 
+/**
+  * \brief Sets a linearly spaced vector.
+  *
+  * The function generates 'size' equally spaced values in the closed interval [low,high].
+  * When size is set to 1, a vector of length 1 containing 'high' is returned.
+  *
+  * \only_for_vectors
+  *
+  * Example: \include DenseBase_setLinSpaced.cpp
+  * Output: \verbinclude DenseBase_setLinSpaced.out
+  *
+  * For integer scalar types, do not miss the explanations on the definition
+  * of \link LinSpaced(Index,const Scalar&,const Scalar&) even spacing \endlink.
+  *
+  * \sa LinSpaced(Index,const Scalar&,const Scalar&), CwiseNullaryOp
+  */
+template<typename Derived>
+EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE Derived& DenseBase<Derived>::setLinSpaced(Index newSize, const Scalar& low, const Scalar& high)
+{
+  EIGEN_STATIC_ASSERT_VECTOR_ONLY(Derived)
+  return derived() = Derived::NullaryExpr(newSize, internal::linspaced_op<Scalar,PacketScalar>(low,high,newSize));
+}
+
+/**
+  * \brief Sets a linearly spaced vector.
+  *
+  * The function fills \c *this with equally spaced values in the closed interval [low,high].
+  * When size is set to 1, a vector of length 1 containing 'high' is returned.
+  *
+  * \only_for_vectors
+  *
+  * For integer scalar types, do not miss the explanations on the definition
+  * of \link LinSpaced(Index,const Scalar&,const Scalar&) even spacing \endlink.
+  *
+  * \sa LinSpaced(Index,const Scalar&,const Scalar&), setLinSpaced(Index, const Scalar&, const Scalar&), CwiseNullaryOp
+  */
+template<typename Derived>
+EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE Derived& DenseBase<Derived>::setLinSpaced(const Scalar& low, const Scalar& high)
+{
+  EIGEN_STATIC_ASSERT_VECTOR_ONLY(Derived)
+  return setLinSpaced(size(), low, high);
+}
 
 // zero:
 
@@ -317,16 +417,14 @@ Matrix<_Scalar, _Rows, _Cols, _Options, _MaxRows, _MaxCols>::setConstant(int row
   * it is redundant to pass \a rows and \a cols as arguments, so Zero() should be used
   * instead.
   *
-  * \addexample Zero \label How to take get a zero matrix
-  *
   * Example: \include MatrixBase_zero_int_int.cpp
   * Output: \verbinclude MatrixBase_zero_int_int.out
   *
-  * \sa Zero(), Zero(int)
+  * \sa Zero(), Zero(Index)
   */
 template<typename Derived>
-EIGEN_STRONG_INLINE const typename MatrixBase<Derived>::ConstantReturnType
-MatrixBase<Derived>::Zero(int rows, int cols)
+EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE const typename DenseBase<Derived>::ConstantReturnType
+DenseBase<Derived>::Zero(Index rows, Index cols)
 {
   return Constant(rows, cols, Scalar(0));
 }
@@ -345,11 +443,11 @@ MatrixBase<Derived>::Zero(int rows, int cols)
   * Example: \include MatrixBase_zero_int.cpp
   * Output: \verbinclude MatrixBase_zero_int.out
   *
-  * \sa Zero(), Zero(int,int)
+  * \sa Zero(), Zero(Index,Index)
   */
 template<typename Derived>
-EIGEN_STRONG_INLINE const typename MatrixBase<Derived>::ConstantReturnType
-MatrixBase<Derived>::Zero(int size)
+EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE const typename DenseBase<Derived>::ConstantReturnType
+DenseBase<Derived>::Zero(Index size)
 {
   return Constant(size, Scalar(0));
 }
@@ -362,11 +460,11 @@ MatrixBase<Derived>::Zero(int size)
   * Example: \include MatrixBase_zero.cpp
   * Output: \verbinclude MatrixBase_zero.out
   *
-  * \sa Zero(int), Zero(int,int)
+  * \sa Zero(Index), Zero(Index,Index)
   */
 template<typename Derived>
-EIGEN_STRONG_INLINE const typename MatrixBase<Derived>::ConstantReturnType
-MatrixBase<Derived>::Zero()
+EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE const typename DenseBase<Derived>::ConstantReturnType
+DenseBase<Derived>::Zero()
 {
   return Constant(Scalar(0));
 }
@@ -380,11 +478,12 @@ MatrixBase<Derived>::Zero()
   * \sa class CwiseNullaryOp, Zero()
   */
 template<typename Derived>
-bool MatrixBase<Derived>::isZero(RealScalar prec) const
+EIGEN_DEVICE_FUNC bool DenseBase<Derived>::isZero(const RealScalar& prec) const
 {
-  for(int j = 0; j < cols(); ++j)
-    for(int i = 0; i < rows(); ++i)
-      if(!ei_isMuchSmallerThan(coeff(i, j), static_cast<Scalar>(1), prec))
+  typename internal::nested_eval<Derived,1>::type self(derived());
+  for(Index j = 0; j < cols(); ++j)
+    for(Index i = 0; i < rows(); ++i)
+      if(!internal::isMuchSmallerThan(self.coeff(i, j), static_cast<Scalar>(1), prec))
         return false;
   return true;
 }
@@ -397,7 +496,7 @@ bool MatrixBase<Derived>::isZero(RealScalar prec) const
   * \sa class CwiseNullaryOp, Zero()
   */
 template<typename Derived>
-EIGEN_STRONG_INLINE Derived& MatrixBase<Derived>::setZero()
+EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE Derived& DenseBase<Derived>::setZero()
 {
   return setConstant(Scalar(0));
 }
@@ -409,13 +508,13 @@ EIGEN_STRONG_INLINE Derived& MatrixBase<Derived>::setZero()
   * Example: \include Matrix_setZero_int.cpp
   * Output: \verbinclude Matrix_setZero_int.out
   *
-  * \sa MatrixBase::setZero(), setZero(int,int), class CwiseNullaryOp, MatrixBase::Zero()
+  * \sa DenseBase::setZero(), setZero(Index,Index), class CwiseNullaryOp, DenseBase::Zero()
   */
-template<typename _Scalar, int _Rows, int _Cols, int _Options, int _MaxRows, int _MaxCols>
-EIGEN_STRONG_INLINE Matrix<_Scalar, _Rows, _Cols, _Options, _MaxRows, _MaxCols>&
-Matrix<_Scalar, _Rows, _Cols, _Options, _MaxRows, _MaxCols>::setZero(int size)
+template<typename Derived>
+EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE Derived&
+PlainObjectBase<Derived>::setZero(Index newSize)
 {
-  resize(size);
+  resize(newSize);
   return setConstant(Scalar(0));
 }
 
@@ -427,11 +526,11 @@ Matrix<_Scalar, _Rows, _Cols, _Options, _MaxRows, _MaxCols>::setZero(int size)
   * Example: \include Matrix_setZero_int_int.cpp
   * Output: \verbinclude Matrix_setZero_int_int.out
   *
-  * \sa MatrixBase::setZero(), setZero(int), class CwiseNullaryOp, MatrixBase::Zero()
+  * \sa DenseBase::setZero(), setZero(Index), class CwiseNullaryOp, DenseBase::Zero()
   */
-template<typename _Scalar, int _Rows, int _Cols, int _Options, int _MaxRows, int _MaxCols>
-EIGEN_STRONG_INLINE Matrix<_Scalar, _Rows, _Cols, _Options, _MaxRows, _MaxCols>&
-Matrix<_Scalar, _Rows, _Cols, _Options, _MaxRows, _MaxCols>::setZero(int rows, int cols)
+template<typename Derived>
+EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE Derived&
+PlainObjectBase<Derived>::setZero(Index rows, Index cols)
 {
   resize(rows, cols);
   return setConstant(Scalar(0));
@@ -448,23 +547,21 @@ Matrix<_Scalar, _Rows, _Cols, _Options, _MaxRows, _MaxCols>::setZero(int rows, i
   * it is redundant to pass \a rows and \a cols as arguments, so Ones() should be used
   * instead.
   *
-  * \addexample One \label How to get a matrix with all coefficients equal one
-  *
   * Example: \include MatrixBase_ones_int_int.cpp
   * Output: \verbinclude MatrixBase_ones_int_int.out
   *
-  * \sa Ones(), Ones(int), isOnes(), class Ones
+  * \sa Ones(), Ones(Index), isOnes(), class Ones
   */
 template<typename Derived>
-EIGEN_STRONG_INLINE const typename MatrixBase<Derived>::ConstantReturnType
-MatrixBase<Derived>::Ones(int rows, int cols)
+EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE const typename DenseBase<Derived>::ConstantReturnType
+DenseBase<Derived>::Ones(Index rows, Index cols)
 {
   return Constant(rows, cols, Scalar(1));
 }
 
 /** \returns an expression of a vector where all coefficients equal one.
   *
-  * The parameter \a size is the size of the returned vector.
+  * The parameter \a newSize is the size of the returned vector.
   * Must be compatible with this MatrixBase type.
   *
   * \only_for_vectors
@@ -476,13 +573,13 @@ MatrixBase<Derived>::Ones(int rows, int cols)
   * Example: \include MatrixBase_ones_int.cpp
   * Output: \verbinclude MatrixBase_ones_int.out
   *
-  * \sa Ones(), Ones(int,int), isOnes(), class Ones
+  * \sa Ones(), Ones(Index,Index), isOnes(), class Ones
   */
 template<typename Derived>
-EIGEN_STRONG_INLINE const typename MatrixBase<Derived>::ConstantReturnType
-MatrixBase<Derived>::Ones(int size)
+EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE const typename DenseBase<Derived>::ConstantReturnType
+DenseBase<Derived>::Ones(Index newSize)
 {
-  return Constant(size, Scalar(1));
+  return Constant(newSize, Scalar(1));
 }
 
 /** \returns an expression of a fixed-size matrix or vector where all coefficients equal one.
@@ -493,11 +590,11 @@ MatrixBase<Derived>::Ones(int size)
   * Example: \include MatrixBase_ones.cpp
   * Output: \verbinclude MatrixBase_ones.out
   *
-  * \sa Ones(int), Ones(int,int), isOnes(), class Ones
+  * \sa Ones(Index), Ones(Index,Index), isOnes(), class Ones
   */
 template<typename Derived>
-EIGEN_STRONG_INLINE const typename MatrixBase<Derived>::ConstantReturnType
-MatrixBase<Derived>::Ones()
+EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE const typename DenseBase<Derived>::ConstantReturnType
+DenseBase<Derived>::Ones()
 {
   return Constant(Scalar(1));
 }
@@ -511,8 +608,8 @@ MatrixBase<Derived>::Ones()
   * \sa class CwiseNullaryOp, Ones()
   */
 template<typename Derived>
-bool MatrixBase<Derived>::isOnes
-(RealScalar prec) const
+EIGEN_DEVICE_FUNC bool DenseBase<Derived>::isOnes
+(const RealScalar& prec) const
 {
   return isApproxToConstant(Scalar(1), prec);
 }
@@ -525,25 +622,25 @@ bool MatrixBase<Derived>::isOnes
   * \sa class CwiseNullaryOp, Ones()
   */
 template<typename Derived>
-EIGEN_STRONG_INLINE Derived& MatrixBase<Derived>::setOnes()
+EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE Derived& DenseBase<Derived>::setOnes()
 {
   return setConstant(Scalar(1));
 }
 
-/** Resizes to the given \a size, and sets all coefficients in this expression to one.
+/** Resizes to the given \a newSize, and sets all coefficients in this expression to one.
   *
   * \only_for_vectors
   *
   * Example: \include Matrix_setOnes_int.cpp
   * Output: \verbinclude Matrix_setOnes_int.out
   *
-  * \sa MatrixBase::setOnes(), setOnes(int,int), class CwiseNullaryOp, MatrixBase::Ones()
+  * \sa MatrixBase::setOnes(), setOnes(Index,Index), class CwiseNullaryOp, MatrixBase::Ones()
   */
-template<typename _Scalar, int _Rows, int _Cols, int _Options, int _MaxRows, int _MaxCols>
-EIGEN_STRONG_INLINE Matrix<_Scalar, _Rows, _Cols, _Options, _MaxRows, _MaxCols>&
-Matrix<_Scalar, _Rows, _Cols, _Options, _MaxRows, _MaxCols>::setOnes(int size)
+template<typename Derived>
+EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE Derived&
+PlainObjectBase<Derived>::setOnes(Index newSize)
 {
-  resize(size);
+  resize(newSize);
   return setConstant(Scalar(1));
 }
 
@@ -555,11 +652,11 @@ Matrix<_Scalar, _Rows, _Cols, _Options, _MaxRows, _MaxCols>::setOnes(int size)
   * Example: \include Matrix_setOnes_int_int.cpp
   * Output: \verbinclude Matrix_setOnes_int_int.out
   *
-  * \sa MatrixBase::setOnes(), setOnes(int), class CwiseNullaryOp, MatrixBase::Ones()
+  * \sa MatrixBase::setOnes(), setOnes(Index), class CwiseNullaryOp, MatrixBase::Ones()
   */
-template<typename _Scalar, int _Rows, int _Cols, int _Options, int _MaxRows, int _MaxCols>
-EIGEN_STRONG_INLINE Matrix<_Scalar, _Rows, _Cols, _Options, _MaxRows, _MaxCols>&
-Matrix<_Scalar, _Rows, _Cols, _Options, _MaxRows, _MaxCols>::setOnes(int rows, int cols)
+template<typename Derived>
+EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE Derived&
+PlainObjectBase<Derived>::setOnes(Index rows, Index cols)
 {
   resize(rows, cols);
   return setConstant(Scalar(1));
@@ -576,18 +673,16 @@ Matrix<_Scalar, _Rows, _Cols, _Options, _MaxRows, _MaxCols>::setOnes(int rows, i
   * it is redundant to pass \a rows and \a cols as arguments, so Identity() should be used
   * instead.
   *
-  * \addexample Identity \label How to get an identity matrix
-  *
   * Example: \include MatrixBase_identity_int_int.cpp
   * Output: \verbinclude MatrixBase_identity_int_int.out
   *
   * \sa Identity(), setIdentity(), isIdentity()
   */
 template<typename Derived>
-EIGEN_STRONG_INLINE const typename MatrixBase<Derived>::IdentityReturnType
-MatrixBase<Derived>::Identity(int rows, int cols)
+EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE const typename MatrixBase<Derived>::IdentityReturnType
+MatrixBase<Derived>::Identity(Index rows, Index cols)
 {
-  return NullaryExpr(rows, cols, ei_scalar_identity_op<Scalar>());
+  return DenseBase<Derived>::NullaryExpr(rows, cols, internal::scalar_identity_op<Scalar>());
 }
 
 /** \returns an expression of the identity matrix (not necessarily square).
@@ -598,14 +693,14 @@ MatrixBase<Derived>::Identity(int rows, int cols)
   * Example: \include MatrixBase_identity.cpp
   * Output: \verbinclude MatrixBase_identity.out
   *
-  * \sa Identity(int,int), setIdentity(), isIdentity()
+  * \sa Identity(Index,Index), setIdentity(), isIdentity()
   */
 template<typename Derived>
-EIGEN_STRONG_INLINE const typename MatrixBase<Derived>::IdentityReturnType
+EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE const typename MatrixBase<Derived>::IdentityReturnType
 MatrixBase<Derived>::Identity()
 {
   EIGEN_STATIC_ASSERT_FIXED_SIZE(Derived)
-  return NullaryExpr(RowsAtCompileTime, ColsAtCompileTime, ei_scalar_identity_op<Scalar>());
+  return MatrixBase<Derived>::NullaryExpr(RowsAtCompileTime, ColsAtCompileTime, internal::scalar_identity_op<Scalar>());
 }
 
 /** \returns true if *this is approximately equal to the identity matrix
@@ -615,24 +710,25 @@ MatrixBase<Derived>::Identity()
   * Example: \include MatrixBase_isIdentity.cpp
   * Output: \verbinclude MatrixBase_isIdentity.out
   *
-  * \sa class CwiseNullaryOp, Identity(), Identity(int,int), setIdentity()
+  * \sa class CwiseNullaryOp, Identity(), Identity(Index,Index), setIdentity()
   */
 template<typename Derived>
 bool MatrixBase<Derived>::isIdentity
-(RealScalar prec) const
+(const RealScalar& prec) const
 {
-  for(int j = 0; j < cols(); ++j)
+  typename internal::nested_eval<Derived,1>::type self(derived());
+  for(Index j = 0; j < cols(); ++j)
   {
-    for(int i = 0; i < rows(); ++i)
+    for(Index i = 0; i < rows(); ++i)
     {
       if(i == j)
       {
-        if(!ei_isApprox(coeff(i, j), static_cast<Scalar>(1), prec))
+        if(!internal::isApprox(self.coeff(i, j), static_cast<Scalar>(1), prec))
           return false;
       }
       else
       {
-        if(!ei_isMuchSmallerThan(coeff(i, j), static_cast<RealScalar>(1), prec))
+        if(!internal::isMuchSmallerThan(self.coeff(i, j), static_cast<RealScalar>(1), prec))
           return false;
       }
     }
@@ -640,9 +736,12 @@ bool MatrixBase<Derived>::isIdentity
   return true;
 }
 
+namespace internal {
+
 template<typename Derived, bool Big = (Derived::SizeAtCompileTime>=16)>
-struct ei_setIdentity_impl
+struct setIdentity_impl
 {
+  EIGEN_DEVICE_FUNC
   static EIGEN_STRONG_INLINE Derived& run(Derived& m)
   {
     return m = Derived::Identity(m.rows(), m.cols());
@@ -650,31 +749,34 @@ struct ei_setIdentity_impl
 };
 
 template<typename Derived>
-struct ei_setIdentity_impl<Derived, true>
+struct setIdentity_impl<Derived, true>
 {
+  EIGEN_DEVICE_FUNC
   static EIGEN_STRONG_INLINE Derived& run(Derived& m)
   {
     m.setZero();
-    const int size = std::min(m.rows(), m.cols());
-    for(int i = 0; i < size; ++i) m.coeffRef(i,i) = typename Derived::Scalar(1);
+    const Index size = numext::mini(m.rows(), m.cols());
+    for(Index i = 0; i < size; ++i) m.coeffRef(i,i) = typename Derived::Scalar(1);
     return m;
   }
 };
 
+} // end namespace internal
+
 /** Writes the identity expression (not necessarily square) into *this.
   *
   * Example: \include MatrixBase_setIdentity.cpp
   * Output: \verbinclude MatrixBase_setIdentity.out
   *
-  * \sa class CwiseNullaryOp, Identity(), Identity(int,int), isIdentity()
+  * \sa class CwiseNullaryOp, Identity(), Identity(Index,Index), isIdentity()
   */
 template<typename Derived>
-EIGEN_STRONG_INLINE Derived& MatrixBase<Derived>::setIdentity()
+EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE Derived& MatrixBase<Derived>::setIdentity()
 {
-  return ei_setIdentity_impl<Derived>::run(derived());
+  return internal::setIdentity_impl<Derived>::run(derived());
 }
 
-/** Resizes to the given size, and writes the identity expression (not necessarily square) into *this.
+/** \brief Resizes to the given size, and writes the identity expression (not necessarily square) into *this.
   *
   * \param rows the new number of rows
   * \param cols the new number of columns
@@ -684,11 +786,10 @@ EIGEN_STRONG_INLINE Derived& MatrixBase<Derived>::setIdentity()
   *
   * \sa MatrixBase::setIdentity(), class CwiseNullaryOp, MatrixBase::Identity()
   */
-template<typename _Scalar, int _Rows, int _Cols, int _Options, int _MaxRows, int _MaxCols>
-EIGEN_STRONG_INLINE Matrix<_Scalar, _Rows, _Cols, _Options, _MaxRows, _MaxCols>&
-Matrix<_Scalar, _Rows, _Cols, _Options, _MaxRows, _MaxCols>::setIdentity(int rows, int cols)
+template<typename Derived>
+EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE Derived& MatrixBase<Derived>::setIdentity(Index rows, Index cols)
 {
-  resize(rows, cols);
+  derived().resize(rows, cols);
   return setIdentity();
 }
 
@@ -696,13 +797,13 @@ Matrix<_Scalar, _Rows, _Cols, _Options, _MaxRows, _MaxCols>::setIdentity(int row
   *
   * \only_for_vectors
   *
-  * \sa MatrixBase::Unit(int), MatrixBase::UnitX(), MatrixBase::UnitY(), MatrixBase::UnitZ(), MatrixBase::UnitW()
+  * \sa MatrixBase::Unit(Index), MatrixBase::UnitX(), MatrixBase::UnitY(), MatrixBase::UnitZ(), MatrixBase::UnitW()
   */
 template<typename Derived>
-EIGEN_STRONG_INLINE const typename MatrixBase<Derived>::BasisReturnType MatrixBase<Derived>::Unit(int size, int i)
+EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE const typename MatrixBase<Derived>::BasisReturnType MatrixBase<Derived>::Unit(Index newSize, Index i)
 {
   EIGEN_STATIC_ASSERT_VECTOR_ONLY(Derived)
-  return BasisReturnType(SquareMatrixType::Identity(size,size), i);
+  return BasisReturnType(SquareMatrixType::Identity(newSize,newSize), i);
 }
 
 /** \returns an expression of the i-th unit (basis) vector.
@@ -711,10 +812,10 @@ EIGEN_STRONG_INLINE const typename MatrixBase<Derived>::BasisReturnType MatrixBa
   *
   * This variant is for fixed-size vector only.
   *
-  * \sa MatrixBase::Unit(int,int), MatrixBase::UnitX(), MatrixBase::UnitY(), MatrixBase::UnitZ(), MatrixBase::UnitW()
+  * \sa MatrixBase::Unit(Index,Index), MatrixBase::UnitX(), MatrixBase::UnitY(), MatrixBase::UnitZ(), MatrixBase::UnitW()
   */
 template<typename Derived>
-EIGEN_STRONG_INLINE const typename MatrixBase<Derived>::BasisReturnType MatrixBase<Derived>::Unit(int i)
+EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE const typename MatrixBase<Derived>::BasisReturnType MatrixBase<Derived>::Unit(Index i)
 {
   EIGEN_STATIC_ASSERT_VECTOR_ONLY(Derived)
   return BasisReturnType(SquareMatrixType::Identity(),i);
@@ -724,40 +825,42 @@ EIGEN_STRONG_INLINE const typename MatrixBase<Derived>::BasisReturnType MatrixBa
   *
   * \only_for_vectors
   *
-  * \sa MatrixBase::Unit(int,int), MatrixBase::Unit(int), MatrixBase::UnitY(), MatrixBase::UnitZ(), MatrixBase::UnitW()
+  * \sa MatrixBase::Unit(Index,Index), MatrixBase::Unit(Index), MatrixBase::UnitY(), MatrixBase::UnitZ(), MatrixBase::UnitW()
   */
 template<typename Derived>
-EIGEN_STRONG_INLINE const typename MatrixBase<Derived>::BasisReturnType MatrixBase<Derived>::UnitX()
+EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE const typename MatrixBase<Derived>::BasisReturnType MatrixBase<Derived>::UnitX()
 { return Derived::Unit(0); }
 
 /** \returns an expression of the Y axis unit vector (0,1{,0}^*)
   *
   * \only_for_vectors
   *
-  * \sa MatrixBase::Unit(int,int), MatrixBase::Unit(int), MatrixBase::UnitY(), MatrixBase::UnitZ(), MatrixBase::UnitW()
+  * \sa MatrixBase::Unit(Index,Index), MatrixBase::Unit(Index), MatrixBase::UnitY(), MatrixBase::UnitZ(), MatrixBase::UnitW()
   */
 template<typename Derived>
-EIGEN_STRONG_INLINE const typename MatrixBase<Derived>::BasisReturnType MatrixBase<Derived>::UnitY()
+EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE const typename MatrixBase<Derived>::BasisReturnType MatrixBase<Derived>::UnitY()
 { return Derived::Unit(1); }
 
 /** \returns an expression of the Z axis unit vector (0,0,1{,0}^*)
   *
   * \only_for_vectors
   *
-  * \sa MatrixBase::Unit(int,int), MatrixBase::Unit(int), MatrixBase::UnitY(), MatrixBase::UnitZ(), MatrixBase::UnitW()
+  * \sa MatrixBase::Unit(Index,Index), MatrixBase::Unit(Index), MatrixBase::UnitY(), MatrixBase::UnitZ(), MatrixBase::UnitW()
   */
 template<typename Derived>
-EIGEN_STRONG_INLINE const typename MatrixBase<Derived>::BasisReturnType MatrixBase<Derived>::UnitZ()
+EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE const typename MatrixBase<Derived>::BasisReturnType MatrixBase<Derived>::UnitZ()
 { return Derived::Unit(2); }
 
 /** \returns an expression of the W axis unit vector (0,0,0,1)
   *
   * \only_for_vectors
   *
-  * \sa MatrixBase::Unit(int,int), MatrixBase::Unit(int), MatrixBase::UnitY(), MatrixBase::UnitZ(), MatrixBase::UnitW()
+  * \sa MatrixBase::Unit(Index,Index), MatrixBase::Unit(Index), MatrixBase::UnitY(), MatrixBase::UnitZ(), MatrixBase::UnitW()
   */
 template<typename Derived>
-EIGEN_STRONG_INLINE const typename MatrixBase<Derived>::BasisReturnType MatrixBase<Derived>::UnitW()
+EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE const typename MatrixBase<Derived>::BasisReturnType MatrixBase<Derived>::UnitW()
 { return Derived::Unit(3); }
 
+} // end namespace Eigen
+
 #endif // EIGEN_CWISE_NULLARY_OP_H

Eigen/src/Core/CwiseTernaryOp.h

@@ -0,0 +1,197 @@
+// This file is part of Eigen, a lightweight C++ template library
+// for linear algebra.
+//
+// Copyright (C) 2008-2014 Gael Guennebaud <gael.guennebaud@inria.fr>
+// Copyright (C) 2006-2008 Benoit Jacob <jacob.benoit.1@gmail.com>
+// Copyright (C) 2016 Eugene Brevdo <ebrevdo@gmail.com>
+//
+// This Source Code Form is subject to the terms of the Mozilla
+// Public License v. 2.0. If a copy of the MPL was not distributed
+// with this file, You can obtain one at http://mozilla.org/MPL/2.0/.
+
+#ifndef EIGEN_CWISE_TERNARY_OP_H
+#define EIGEN_CWISE_TERNARY_OP_H
+
+namespace Eigen {
+
+namespace internal {
+template <typename TernaryOp, typename Arg1, typename Arg2, typename Arg3>
+struct traits<CwiseTernaryOp<TernaryOp, Arg1, Arg2, Arg3> > {
+  // we must not inherit from traits<Arg1> since it has
+  // the potential to cause problems with MSVC
+  typedef typename remove_all<Arg1>::type Ancestor;
+  typedef typename traits<Ancestor>::XprKind XprKind;
+  enum {
+    RowsAtCompileTime = traits<Ancestor>::RowsAtCompileTime,
+    ColsAtCompileTime = traits<Ancestor>::ColsAtCompileTime,
+    MaxRowsAtCompileTime = traits<Ancestor>::MaxRowsAtCompileTime,
+    MaxColsAtCompileTime = traits<Ancestor>::MaxColsAtCompileTime
+  };
+
+  // even though we require Arg1, Arg2, and Arg3 to have the same scalar type
+  // (see CwiseTernaryOp constructor),
+  // we still want to handle the case when the result type is different.
+  typedef typename result_of<TernaryOp(
+      const typename Arg1::Scalar&, const typename Arg2::Scalar&,
+      const typename Arg3::Scalar&)>::type Scalar;
+
+  typedef typename internal::traits<Arg1>::StorageKind StorageKind;
+  typedef typename internal::traits<Arg1>::StorageIndex StorageIndex;
+
+  typedef typename Arg1::Nested Arg1Nested;
+  typedef typename Arg2::Nested Arg2Nested;
+  typedef typename Arg3::Nested Arg3Nested;
+  typedef typename remove_reference<Arg1Nested>::type _Arg1Nested;
+  typedef typename remove_reference<Arg2Nested>::type _Arg2Nested;
+  typedef typename remove_reference<Arg3Nested>::type _Arg3Nested;
+  enum { Flags = _Arg1Nested::Flags & RowMajorBit };
+};
+}  // end namespace internal
+
+template <typename TernaryOp, typename Arg1, typename Arg2, typename Arg3,
+          typename StorageKind>
+class CwiseTernaryOpImpl;
+
+/** \class CwiseTernaryOp
+  * \ingroup Core_Module
+  *
+  * \brief Generic expression where a coefficient-wise ternary operator is
+ * applied to two expressions
+  *
+  * \tparam TernaryOp template functor implementing the operator
+  * \tparam Arg1Type the type of the first argument
+  * \tparam Arg2Type the type of the second argument
+  * \tparam Arg3Type the type of the third argument
+  *
+  * This class represents an expression where a coefficient-wise ternary
+ * operator is applied to three expressions.
+  * It is the return type of ternary operators, by which we mean only those
+ * ternary operators where
+  * all three arguments are Eigen expressions.
+  * For example, the return type of betainc(matrix1, matrix2, matrix3) is a
+ * CwiseTernaryOp.
+  *
+  * Most of the time, this is the only way that it is used, so you typically
+ * don't have to name
+  * CwiseTernaryOp types explicitly.
+  *
+  * \sa MatrixBase::ternaryExpr(const MatrixBase<Argument2> &, const
+ * MatrixBase<Argument3> &, const CustomTernaryOp &) const, class CwiseBinaryOp,
+ * class CwiseUnaryOp, class CwiseNullaryOp
+  */
+template <typename TernaryOp, typename Arg1Type, typename Arg2Type,
+          typename Arg3Type>
+class CwiseTernaryOp : public CwiseTernaryOpImpl<
+                           TernaryOp, Arg1Type, Arg2Type, Arg3Type,
+                           typename internal::traits<Arg1Type>::StorageKind>,
+                       internal::no_assignment_operator
+{
+ public:
+  typedef typename internal::remove_all<Arg1Type>::type Arg1;
+  typedef typename internal::remove_all<Arg2Type>::type Arg2;
+  typedef typename internal::remove_all<Arg3Type>::type Arg3;
+
+  typedef typename CwiseTernaryOpImpl<
+      TernaryOp, Arg1Type, Arg2Type, Arg3Type,
+      typename internal::traits<Arg1Type>::StorageKind>::Base Base;
+  EIGEN_GENERIC_PUBLIC_INTERFACE(CwiseTernaryOp)
+
+  typedef typename internal::ref_selector<Arg1Type>::type Arg1Nested;
+  typedef typename internal::ref_selector<Arg2Type>::type Arg2Nested;
+  typedef typename internal::ref_selector<Arg3Type>::type Arg3Nested;
+  typedef typename internal::remove_reference<Arg1Nested>::type _Arg1Nested;
+  typedef typename internal::remove_reference<Arg2Nested>::type _Arg2Nested;
+  typedef typename internal::remove_reference<Arg3Nested>::type _Arg3Nested;
+
+  EIGEN_DEVICE_FUNC
+  EIGEN_STRONG_INLINE CwiseTernaryOp(const Arg1& a1, const Arg2& a2,
+                                     const Arg3& a3,
+                                     const TernaryOp& func = TernaryOp())
+      : m_arg1(a1), m_arg2(a2), m_arg3(a3), m_functor(func) {
+    // require the sizes to match
+    EIGEN_STATIC_ASSERT_SAME_MATRIX_SIZE(Arg1, Arg2)
+    EIGEN_STATIC_ASSERT_SAME_MATRIX_SIZE(Arg1, Arg3)
+
+    // The index types should match
+    EIGEN_STATIC_ASSERT((internal::is_same<
+                         typename internal::traits<Arg1Type>::StorageKind,
+                         typename internal::traits<Arg2Type>::StorageKind>::value),
+                        STORAGE_KIND_MUST_MATCH)
+    EIGEN_STATIC_ASSERT((internal::is_same<
+                         typename internal::traits<Arg1Type>::StorageKind,
+                         typename internal::traits<Arg3Type>::StorageKind>::value),
+                        STORAGE_KIND_MUST_MATCH)
+
+    eigen_assert(a1.rows() == a2.rows() && a1.cols() == a2.cols() &&
+                 a1.rows() == a3.rows() && a1.cols() == a3.cols());
+  }
+
+  EIGEN_DEVICE_FUNC
+  EIGEN_STRONG_INLINE Index rows() const {
+    // return the fixed size type if available to enable compile time
+    // optimizations
+    if (internal::traits<typename internal::remove_all<Arg1Nested>::type>::
+                RowsAtCompileTime == Dynamic &&
+        internal::traits<typename internal::remove_all<Arg2Nested>::type>::
+                RowsAtCompileTime == Dynamic)
+      return m_arg3.rows();
+    else if (internal::traits<typename internal::remove_all<Arg1Nested>::type>::
+                     RowsAtCompileTime == Dynamic &&
+             internal::traits<typename internal::remove_all<Arg3Nested>::type>::
+                     RowsAtCompileTime == Dynamic)
+      return m_arg2.rows();
+    else
+      return m_arg1.rows();
+  }
+  EIGEN_DEVICE_FUNC
+  EIGEN_STRONG_INLINE Index cols() const {
+    // return the fixed size type if available to enable compile time
+    // optimizations
+    if (internal::traits<typename internal::remove_all<Arg1Nested>::type>::
+                ColsAtCompileTime == Dynamic &&
+        internal::traits<typename internal::remove_all<Arg2Nested>::type>::
+                ColsAtCompileTime == Dynamic)
+      return m_arg3.cols();
+    else if (internal::traits<typename internal::remove_all<Arg1Nested>::type>::
+                     ColsAtCompileTime == Dynamic &&
+             internal::traits<typename internal::remove_all<Arg3Nested>::type>::
+                     ColsAtCompileTime == Dynamic)
+      return m_arg2.cols();
+    else
+      return m_arg1.cols();
+  }
+
+  /** \returns the first argument nested expression */
+  EIGEN_DEVICE_FUNC
+  const _Arg1Nested& arg1() const { return m_arg1; }
+  /** \returns the first argument nested expression */
+  EIGEN_DEVICE_FUNC
+  const _Arg2Nested& arg2() const { return m_arg2; }
+  /** \returns the third argument nested expression */
+  EIGEN_DEVICE_FUNC
+  const _Arg3Nested& arg3() const { return m_arg3; }
+  /** \returns the functor representing the ternary operation */
+  EIGEN_DEVICE_FUNC
+  const TernaryOp& functor() const { return m_functor; }
+
+ protected:
+  Arg1Nested m_arg1;
+  Arg2Nested m_arg2;
+  Arg3Nested m_arg3;
+  const TernaryOp m_functor;
+};
+
+// Generic API dispatcher
+template <typename TernaryOp, typename Arg1, typename Arg2, typename Arg3,
+          typename StorageKind>
+class CwiseTernaryOpImpl
+    : public internal::generic_xpr_base<
+          CwiseTernaryOp<TernaryOp, Arg1, Arg2, Arg3> >::type {
+ public:
+  typedef typename internal::generic_xpr_base<
+      CwiseTernaryOp<TernaryOp, Arg1, Arg2, Arg3> >::type Base;
+};
+
+}  // end namespace Eigen
+
+#endif  // EIGEN_CWISE_TERNARY_OP_H

Eigen/src/Core/CwiseUnaryOp.h

@@ -1,229 +1,103 @@
 // This file is part of Eigen, a lightweight C++ template library
-// for linear algebra. Eigen itself is part of the KDE project.
+// for linear algebra.
 //
-// Copyright (C) 2008 Gael Guennebaud <g.gael@free.fr>
+// Copyright (C) 2008-2014 Gael Guennebaud <gael.guennebaud@inria.fr>
 // Copyright (C) 2006-2008 Benoit Jacob <jacob.benoit.1@gmail.com>
 //
-// 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/>.
+// This Source Code Form is subject to the terms of the Mozilla
+// Public License v. 2.0. If a copy of the MPL was not distributed
+// with this file, You can obtain one at http://mozilla.org/MPL/2.0/.
 
 #ifndef EIGEN_CWISE_UNARY_OP_H
 #define EIGEN_CWISE_UNARY_OP_H
 
+namespace Eigen { 
+
+namespace internal {
+template<typename UnaryOp, typename XprType>
+struct traits<CwiseUnaryOp<UnaryOp, XprType> >
+ : traits<XprType>
+{
+  typedef typename result_of<
+                     UnaryOp(const typename XprType::Scalar&)
+                   >::type Scalar;
+  typedef typename XprType::Nested XprTypeNested;
+  typedef typename remove_reference<XprTypeNested>::type _XprTypeNested;
+  enum {
+    Flags = _XprTypeNested::Flags & RowMajorBit 
+  };
+};
+}
+
+template<typename UnaryOp, typename XprType, typename StorageKind>
+class CwiseUnaryOpImpl;
+
 /** \class CwiseUnaryOp
+  * \ingroup Core_Module
   *
-  * \brief Generic expression of a coefficient-wise unary operator of a matrix or a vector
+  * \brief Generic expression where a coefficient-wise unary operator is applied to an expression
   *
-  * \param UnaryOp template functor implementing the operator
-  * \param MatrixType the type of the matrix we are applying the unary operator
+  * \tparam UnaryOp template functor implementing the operator
+  * \tparam XprType the type of the expression to which we are applying the unary operator
   *
-  * This class represents an expression of a generic unary operator of a matrix or a vector.
-  * It is the return type of the unary operator-, of a matrix or a vector, and most
-  * of the time this is the only way it is used.
+  * This class represents an expression where a unary operator is applied to an expression.
+  * It is the return type of all operations taking exactly 1 input expression, regardless of the
+  * presence of other inputs such as scalars. For example, the operator* in the expression 3*matrix
+  * is considered unary, because only the right-hand side is an expression, and its
+  * return type is a specialization of CwiseUnaryOp.
+  *
+  * Most of the time, this is the only way that it is used, so you typically don't have to name
+  * CwiseUnaryOp types explicitly.
   *
   * \sa MatrixBase::unaryExpr(const CustomUnaryOp &) const, class CwiseBinaryOp, class CwiseNullaryOp
   */
-template<typename UnaryOp, typename MatrixType>
-struct ei_traits<CwiseUnaryOp<UnaryOp, MatrixType> >
- : ei_traits<MatrixType>
-{
-  typedef typename ei_result_of<
-                     UnaryOp(typename MatrixType::Scalar)
-                   >::type Scalar;
-  typedef typename MatrixType::Nested MatrixTypeNested;
-  typedef typename ei_unref<MatrixTypeNested>::type _MatrixTypeNested;
-  enum {
-    Flags = (_MatrixTypeNested::Flags & (
-      HereditaryBits | LinearAccessBit | AlignedBit
-      | (ei_functor_traits<UnaryOp>::PacketAccess ? PacketAccessBit : 0))),
-    CoeffReadCost = _MatrixTypeNested::CoeffReadCost + ei_functor_traits<UnaryOp>::Cost
-  };
-};
-
-template<typename UnaryOp, typename MatrixType>
-class CwiseUnaryOp : ei_no_assignment_operator,
-  public MatrixBase<CwiseUnaryOp<UnaryOp, MatrixType> >
+template<typename UnaryOp, typename XprType>
+class CwiseUnaryOp : public CwiseUnaryOpImpl<UnaryOp, XprType, typename internal::traits<XprType>::StorageKind>, internal::no_assignment_operator
 {
   public:
 
+    typedef typename CwiseUnaryOpImpl<UnaryOp, XprType,typename internal::traits<XprType>::StorageKind>::Base Base;
     EIGEN_GENERIC_PUBLIC_INTERFACE(CwiseUnaryOp)
+    typedef typename internal::ref_selector<XprType>::type XprTypeNested;
+    typedef typename internal::remove_all<XprType>::type NestedExpression;
 
-    inline CwiseUnaryOp(const MatrixType& mat, const UnaryOp& func = UnaryOp())
-      : m_matrix(mat), m_functor(func) {}
+    EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
+    explicit CwiseUnaryOp(const XprType& xpr, const UnaryOp& func = UnaryOp())
+      : m_xpr(xpr), m_functor(func) {}
 
-    EIGEN_STRONG_INLINE int rows() const { return m_matrix.rows(); }
-    EIGEN_STRONG_INLINE int cols() const { return m_matrix.cols(); }
+    EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
+    Index rows() const { return m_xpr.rows(); }
+    EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
+    Index cols() const { return m_xpr.cols(); }
 
-    EIGEN_STRONG_INLINE const Scalar coeff(int row, int col) const
-    {
-      return m_functor(m_matrix.coeff(row, col));
-    }
+    /** \returns the functor representing the unary operation */
+    EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
+    const UnaryOp& functor() const { return m_functor; }
 
-    template<int LoadMode>
-    EIGEN_STRONG_INLINE PacketScalar packet(int row, int col) const
-    {
-      return m_functor.packetOp(m_matrix.template packet<LoadMode>(row, col));
-    }
+    /** \returns the nested expression */
+    EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
+    const typename internal::remove_all<XprTypeNested>::type&
+    nestedExpression() const { return m_xpr; }
 
-    EIGEN_STRONG_INLINE const Scalar coeff(int index) const
-    {
-      return m_functor(m_matrix.coeff(index));
-    }
-
-    template<int LoadMode>
-    EIGEN_STRONG_INLINE PacketScalar packet(int index) const
-    {
-      return m_functor.packetOp(m_matrix.template packet<LoadMode>(index));
-    }
+    /** \returns the nested expression */
+    EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
+    typename internal::remove_all<XprTypeNested>::type&
+    nestedExpression() { return m_xpr; }
 
   protected:
-    const typename MatrixType::Nested m_matrix;
+    XprTypeNested m_xpr;
     const UnaryOp m_functor;
 };
 
-/** \returns an expression of a custom coefficient-wise unary operator \a func of *this
-  *
-  * The template parameter \a CustomUnaryOp is the type of the functor
-  * of the custom unary operator.
-  *
-  * \addexample CustomCwiseUnaryFunctors \label How to use custom coeff wise unary functors
-  *
-  * Example:
-  * \include class_CwiseUnaryOp.cpp
-  * Output: \verbinclude class_CwiseUnaryOp.out
-  *
-  * \sa class CwiseUnaryOp, class CwiseBinarOp, MatrixBase::operator-, Cwise::abs
-  */
-template<typename Derived>
-template<typename CustomUnaryOp>
-EIGEN_STRONG_INLINE const CwiseUnaryOp<CustomUnaryOp, Derived>
-MatrixBase<Derived>::unaryExpr(const CustomUnaryOp& func) const
+// Generic API dispatcher
+template<typename UnaryOp, typename XprType, typename StorageKind>
+class CwiseUnaryOpImpl
+  : public internal::generic_xpr_base<CwiseUnaryOp<UnaryOp, XprType> >::type
 {
-  return CwiseUnaryOp<CustomUnaryOp, Derived>(derived(), func);
-}
+public:
+  typedef typename internal::generic_xpr_base<CwiseUnaryOp<UnaryOp, XprType> >::type Base;
+};
 
-/** \returns an expression of the opposite of \c *this
-  */
-template<typename Derived>
-EIGEN_STRONG_INLINE const CwiseUnaryOp<ei_scalar_opposite_op<typename ei_traits<Derived>::Scalar>,Derived>
-MatrixBase<Derived>::operator-() const
-{
-  return derived();
-}
-
-/** \returns an expression of the coefficient-wise absolute value of \c *this
-  *
-  * Example: \include Cwise_abs.cpp
-  * Output: \verbinclude Cwise_abs.out
-  *
-  * \sa abs2()
-  */
-template<typename ExpressionType>
-EIGEN_STRONG_INLINE const EIGEN_CWISE_UNOP_RETURN_TYPE(ei_scalar_abs_op)
-Cwise<ExpressionType>::abs() const
-{
-  return _expression();
-}
-
-/** \returns an expression of the coefficient-wise squared absolute value of \c *this
-  *
-  * Example: \include Cwise_abs2.cpp
-  * Output: \verbinclude Cwise_abs2.out
-  *
-  * \sa abs(), square()
-  */
-template<typename ExpressionType>
-EIGEN_STRONG_INLINE const EIGEN_CWISE_UNOP_RETURN_TYPE(ei_scalar_abs2_op)
-Cwise<ExpressionType>::abs2() const
-{
-  return _expression();
-}
-
-/** \returns an expression of the complex conjugate of \c *this.
-  *
-  * \sa adjoint() */
-template<typename Derived>
-EIGEN_STRONG_INLINE typename MatrixBase<Derived>::ConjugateReturnType
-MatrixBase<Derived>::conjugate() const
-{
-  return ConjugateReturnType(derived());
-}
-
-/** \returns an expression of the real part of \c *this.
-  *
-  * \sa imag() */
-template<typename Derived>
-EIGEN_STRONG_INLINE const typename MatrixBase<Derived>::RealReturnType
-MatrixBase<Derived>::real() const { return derived(); }
-
-/** \returns an expression of the imaginary part of \c *this.
-  *
-  * \sa real() */
-template<typename Derived>
-EIGEN_STRONG_INLINE const typename MatrixBase<Derived>::ImagReturnType
-MatrixBase<Derived>::imag() const { return derived(); }
-
-/** \returns an expression of *this with the \a Scalar type casted to
-  * \a NewScalar.
-  *
-  * The template parameter \a NewScalar is the type we are casting the scalars to.
-  *
-  * \sa class CwiseUnaryOp
-  */
-template<typename Derived>
-template<typename NewType>
-EIGEN_STRONG_INLINE const CwiseUnaryOp<ei_scalar_cast_op<typename ei_traits<Derived>::Scalar, NewType>, Derived>
-MatrixBase<Derived>::cast() const
-{
-  return derived();
-}
-
-/** \relates MatrixBase */
-template<typename Derived>
-EIGEN_STRONG_INLINE const typename MatrixBase<Derived>::ScalarMultipleReturnType
-MatrixBase<Derived>::operator*(const Scalar& scalar) const
-{
-  return CwiseUnaryOp<ei_scalar_multiple_op<Scalar>, Derived>
-    (derived(), ei_scalar_multiple_op<Scalar>(scalar));
-}
-
-/** \relates MatrixBase */
-template<typename Derived>
-EIGEN_STRONG_INLINE const CwiseUnaryOp<ei_scalar_quotient1_op<typename ei_traits<Derived>::Scalar>, Derived>
-MatrixBase<Derived>::operator/(const Scalar& scalar) const
-{
-  return CwiseUnaryOp<ei_scalar_quotient1_op<Scalar>, Derived>
-    (derived(), ei_scalar_quotient1_op<Scalar>(scalar));
-}
-
-template<typename Derived>
-EIGEN_STRONG_INLINE Derived&
-MatrixBase<Derived>::operator*=(const Scalar& other)
-{
-  return *this = *this * other;
-}
-
-template<typename Derived>
-EIGEN_STRONG_INLINE Derived&
-MatrixBase<Derived>::operator/=(const Scalar& other)
-{
-  return *this = *this / other;
-}
+} // end namespace Eigen
 
 #endif // EIGEN_CWISE_UNARY_OP_H

Eigen/src/Core/CwiseUnaryView.h

@@ -0,0 +1,130 @@
+// This file is part of Eigen, a lightweight C++ template library
+// for linear algebra.
+//
+// Copyright (C) 2009-2010 Gael Guennebaud <gael.guennebaud@inria.fr>
+//
+// This Source Code Form is subject to the terms of the Mozilla
+// Public License v. 2.0. If a copy of the MPL was not distributed
+// with this file, You can obtain one at http://mozilla.org/MPL/2.0/.
+
+#ifndef EIGEN_CWISE_UNARY_VIEW_H
+#define EIGEN_CWISE_UNARY_VIEW_H
+
+namespace Eigen {
+
+namespace internal {
+template<typename ViewOp, typename MatrixType>
+struct traits<CwiseUnaryView<ViewOp, MatrixType> >
+ : traits<MatrixType>
+{
+  typedef typename result_of<
+                     ViewOp(const typename traits<MatrixType>::Scalar&)
+                   >::type Scalar;
+  typedef typename MatrixType::Nested MatrixTypeNested;
+  typedef typename remove_all<MatrixTypeNested>::type _MatrixTypeNested;
+  enum {
+    FlagsLvalueBit = is_lvalue<MatrixType>::value ? LvalueBit : 0,
+    Flags = traits<_MatrixTypeNested>::Flags & (RowMajorBit | FlagsLvalueBit | DirectAccessBit), // FIXME DirectAccessBit should not be handled by expressions
+    MatrixTypeInnerStride =  inner_stride_at_compile_time<MatrixType>::ret,
+    // need to cast the sizeof's from size_t to int explicitly, otherwise:
+    // "error: no integral type can represent all of the enumerator values
+    InnerStrideAtCompileTime = MatrixTypeInnerStride == Dynamic
+                             ? int(Dynamic)
+                             : int(MatrixTypeInnerStride) * int(sizeof(typename traits<MatrixType>::Scalar) / sizeof(Scalar)),
+    OuterStrideAtCompileTime = outer_stride_at_compile_time<MatrixType>::ret == Dynamic
+                             ? int(Dynamic)
+                             : outer_stride_at_compile_time<MatrixType>::ret * int(sizeof(typename traits<MatrixType>::Scalar) / sizeof(Scalar))
+  };
+};
+}
+
+template<typename ViewOp, typename MatrixType, typename StorageKind>
+class CwiseUnaryViewImpl;
+
+/** \class CwiseUnaryView
+  * \ingroup Core_Module
+  *
+  * \brief Generic lvalue expression of a coefficient-wise unary operator of a matrix or a vector
+  *
+  * \tparam ViewOp template functor implementing the view
+  * \tparam MatrixType the type of the matrix we are applying the unary operator
+  *
+  * This class represents a lvalue expression of a generic unary view operator of a matrix or a vector.
+  * It is the return type of real() and imag(), and most of the time this is the only way it is used.
+  *
+  * \sa MatrixBase::unaryViewExpr(const CustomUnaryOp &) const, class CwiseUnaryOp
+  */
+template<typename ViewOp, typename MatrixType>
+class CwiseUnaryView : public CwiseUnaryViewImpl<ViewOp, MatrixType, typename internal::traits<MatrixType>::StorageKind>
+{
+  public:
+
+    typedef typename CwiseUnaryViewImpl<ViewOp, MatrixType,typename internal::traits<MatrixType>::StorageKind>::Base Base;
+    EIGEN_GENERIC_PUBLIC_INTERFACE(CwiseUnaryView)
+    typedef typename internal::ref_selector<MatrixType>::non_const_type MatrixTypeNested;
+    typedef typename internal::remove_all<MatrixType>::type NestedExpression;
+
+    explicit inline CwiseUnaryView(MatrixType& mat, const ViewOp& func = ViewOp())
+      : m_matrix(mat), m_functor(func) {}
+
+    EIGEN_INHERIT_ASSIGNMENT_OPERATORS(CwiseUnaryView)
+
+    EIGEN_STRONG_INLINE Index rows() const { return m_matrix.rows(); }
+    EIGEN_STRONG_INLINE Index cols() const { return m_matrix.cols(); }
+
+    /** \returns the functor representing unary operation */
+    const ViewOp& functor() const { return m_functor; }
+
+    /** \returns the nested expression */
+    const typename internal::remove_all<MatrixTypeNested>::type&
+    nestedExpression() const { return m_matrix; }
+
+    /** \returns the nested expression */
+    typename internal::remove_reference<MatrixTypeNested>::type&
+    nestedExpression() { return m_matrix.const_cast_derived(); }
+
+  protected:
+    MatrixTypeNested m_matrix;
+    ViewOp m_functor;
+};
+
+// Generic API dispatcher
+template<typename ViewOp, typename XprType, typename StorageKind>
+class CwiseUnaryViewImpl
+  : public internal::generic_xpr_base<CwiseUnaryView<ViewOp, XprType> >::type
+{
+public:
+  typedef typename internal::generic_xpr_base<CwiseUnaryView<ViewOp, XprType> >::type Base;
+};
+
+template<typename ViewOp, typename MatrixType>
+class CwiseUnaryViewImpl<ViewOp,MatrixType,Dense>
+  : public internal::dense_xpr_base< CwiseUnaryView<ViewOp, MatrixType> >::type
+{
+  public:
+
+    typedef CwiseUnaryView<ViewOp, MatrixType> Derived;
+    typedef typename internal::dense_xpr_base< CwiseUnaryView<ViewOp, MatrixType> >::type Base;
+
+    EIGEN_DENSE_PUBLIC_INTERFACE(Derived)
+    EIGEN_INHERIT_ASSIGNMENT_OPERATORS(CwiseUnaryViewImpl)
+    
+    EIGEN_DEVICE_FUNC inline Scalar* data() { return &(this->coeffRef(0)); }
+    EIGEN_DEVICE_FUNC inline const Scalar* data() const { return &(this->coeff(0)); }
+
+    EIGEN_DEVICE_FUNC inline Index innerStride() const
+    {
+      return derived().nestedExpression().innerStride() * sizeof(typename internal::traits<MatrixType>::Scalar) / sizeof(Scalar);
+    }
+
+    EIGEN_DEVICE_FUNC inline Index outerStride() const
+    {
+      return derived().nestedExpression().outerStride() * sizeof(typename internal::traits<MatrixType>::Scalar) / sizeof(Scalar);
+    }
+  protected:
+    EIGEN_DEFAULT_EMPTY_CONSTRUCTOR_AND_DESTRUCTOR(CwiseUnaryViewImpl)
+};
+
+} // end namespace Eigen
+
+#endif // EIGEN_CWISE_UNARY_VIEW_H

Eigen/src/Core/DenseBase.h

@@ -0,0 +1,612 @@
+// This file is part of Eigen, a lightweight C++ template library
+// for linear algebra.
+//
+// Copyright (C) 2007-2010 Benoit Jacob <jacob.benoit.1@gmail.com>
+// Copyright (C) 2008-2010 Gael Guennebaud <gael.guennebaud@inria.fr>
+//
+// This Source Code Form is subject to the terms of the Mozilla
+// Public License v. 2.0. If a copy of the MPL was not distributed
+// with this file, You can obtain one at http://mozilla.org/MPL/2.0/.
+
+#ifndef EIGEN_DENSEBASE_H
+#define EIGEN_DENSEBASE_H
+
+namespace Eigen {
+
+namespace internal {
+  
+// The index type defined by EIGEN_DEFAULT_DENSE_INDEX_TYPE must be a signed type.
+// This dummy function simply aims at checking that at compile time.
+static inline void check_DenseIndex_is_signed() {
+  EIGEN_STATIC_ASSERT(NumTraits<DenseIndex>::IsSigned,THE_INDEX_TYPE_MUST_BE_A_SIGNED_TYPE); 
+}
+
+} // end namespace internal
+  
+/** \class DenseBase
+  * \ingroup Core_Module
+  *
+  * \brief Base class for all dense matrices, vectors, and arrays
+  *
+  * This class is the base that is inherited by all dense objects (matrix, vector, arrays,
+  * and related expression types). The common Eigen API for dense objects is contained in this class.
+  *
+  * \tparam Derived is the derived type, e.g., a matrix type or an expression.
+  *
+  * This class can be extended with the help of the plugin mechanism described on the page
+  * \ref TopicCustomizing_Plugins by defining the preprocessor symbol \c EIGEN_DENSEBASE_PLUGIN.
+  *
+  * \sa \blank \ref TopicClassHierarchy
+  */
+template<typename Derived> class DenseBase
+#ifndef EIGEN_PARSED_BY_DOXYGEN
+  : public DenseCoeffsBase<Derived, internal::accessors_level<Derived>::value>
+#else
+  : public DenseCoeffsBase<Derived,DirectWriteAccessors>
+#endif // not EIGEN_PARSED_BY_DOXYGEN
+{
+  public:
+
+    /** Inner iterator type to iterate over the coefficients of a row or column.
+      * \sa class InnerIterator
+      */
+    typedef Eigen::InnerIterator<Derived> InnerIterator;
+
+    typedef typename internal::traits<Derived>::StorageKind StorageKind;
+
+    /**
+      * \brief The type used to store indices
+      * \details This typedef is relevant for types that store multiple indices such as
+      *          PermutationMatrix or Transpositions, otherwise it defaults to Eigen::Index
+      * \sa \blank \ref TopicPreprocessorDirectives, Eigen::Index, SparseMatrixBase.
+     */
+    typedef typename internal::traits<Derived>::StorageIndex StorageIndex;
+
+    /** The numeric type of the expression' coefficients, e.g. float, double, int or std::complex<float>, etc. */
+    typedef typename internal::traits<Derived>::Scalar Scalar;
+    
+    /** The numeric type of the expression' coefficients, e.g. float, double, int or std::complex<float>, etc.
+      *
+      * It is an alias for the Scalar type */
+    typedef Scalar value_type;
+    
+    typedef typename NumTraits<Scalar>::Real RealScalar;
+    typedef DenseCoeffsBase<Derived, internal::accessors_level<Derived>::value> Base;
+
+    using Base::derived;
+    using Base::const_cast_derived;
+    using Base::rows;
+    using Base::cols;
+    using Base::size;
+    using Base::rowIndexByOuterInner;
+    using Base::colIndexByOuterInner;
+    using Base::coeff;
+    using Base::coeffByOuterInner;
+    using Base::operator();
+    using Base::operator[];
+    using Base::x;
+    using Base::y;
+    using Base::z;
+    using Base::w;
+    using Base::stride;
+    using Base::innerStride;
+    using Base::outerStride;
+    using Base::rowStride;
+    using Base::colStride;
+    typedef typename Base::CoeffReturnType CoeffReturnType;
+
+    enum {
+
+      RowsAtCompileTime = internal::traits<Derived>::RowsAtCompileTime,
+        /**< The number of rows at compile-time. This is just a copy of the value provided
+          * by the \a Derived type. If a value is not known at compile-time,
+          * it is set to the \a Dynamic constant.
+          * \sa MatrixBase::rows(), MatrixBase::cols(), ColsAtCompileTime, SizeAtCompileTime */
+
+      ColsAtCompileTime = internal::traits<Derived>::ColsAtCompileTime,
+        /**< The number of columns at compile-time. This is just a copy of the value provided
+          * by the \a Derived type. If a value is not known at compile-time,
+          * it is set to the \a Dynamic constant.
+          * \sa MatrixBase::rows(), MatrixBase::cols(), RowsAtCompileTime, SizeAtCompileTime */
+
+
+      SizeAtCompileTime = (internal::size_at_compile_time<internal::traits<Derived>::RowsAtCompileTime,
+                                                   internal::traits<Derived>::ColsAtCompileTime>::ret),
+        /**< This is equal to the number of coefficients, i.e. the number of
+          * rows times the number of columns, or to \a Dynamic if this is not
+          * known at compile-time. \sa RowsAtCompileTime, ColsAtCompileTime */
+
+      MaxRowsAtCompileTime = internal::traits<Derived>::MaxRowsAtCompileTime,
+        /**< This value is equal to the maximum possible number of rows that this expression
+          * might have. If this expression might have an arbitrarily high number of rows,
+          * this value is set to \a Dynamic.
+          *
+          * This value is useful to know when evaluating an expression, in order to determine
+          * whether it is possible to avoid doing a dynamic memory allocation.
+          *
+          * \sa RowsAtCompileTime, MaxColsAtCompileTime, MaxSizeAtCompileTime
+          */
+
+      MaxColsAtCompileTime = internal::traits<Derived>::MaxColsAtCompileTime,
+        /**< This value is equal to the maximum possible number of columns that this expression
+          * might have. If this expression might have an arbitrarily high number of columns,
+          * this value is set to \a Dynamic.
+          *
+          * This value is useful to know when evaluating an expression, in order to determine
+          * whether it is possible to avoid doing a dynamic memory allocation.
+          *
+          * \sa ColsAtCompileTime, MaxRowsAtCompileTime, MaxSizeAtCompileTime
+          */
+
+      MaxSizeAtCompileTime = (internal::size_at_compile_time<internal::traits<Derived>::MaxRowsAtCompileTime,
+                                                      internal::traits<Derived>::MaxColsAtCompileTime>::ret),
+        /**< This value is equal to the maximum possible number of coefficients that this expression
+          * might have. If this expression might have an arbitrarily high number of coefficients,
+          * this value is set to \a Dynamic.
+          *
+          * This value is useful to know when evaluating an expression, in order to determine
+          * whether it is possible to avoid doing a dynamic memory allocation.
+          *
+          * \sa SizeAtCompileTime, MaxRowsAtCompileTime, MaxColsAtCompileTime
+          */
+
+      IsVectorAtCompileTime = internal::traits<Derived>::MaxRowsAtCompileTime == 1
+                           || internal::traits<Derived>::MaxColsAtCompileTime == 1,
+        /**< This is set to true if either the number of rows or the number of
+          * columns is known at compile-time to be equal to 1. Indeed, in that case,
+          * we are dealing with a column-vector (if there is only one column) or with
+          * a row-vector (if there is only one row). */
+
+      Flags = internal::traits<Derived>::Flags,
+        /**< This stores expression \ref flags flags which may or may not be inherited by new expressions
+          * constructed from this one. See the \ref flags "list of flags".
+          */
+
+      IsRowMajor = int(Flags) & RowMajorBit, /**< True if this expression has row-major storage order. */
+
+      InnerSizeAtCompileTime = int(IsVectorAtCompileTime) ? int(SizeAtCompileTime)
+                             : int(IsRowMajor) ? int(ColsAtCompileTime) : int(RowsAtCompileTime),
+
+      InnerStrideAtCompileTime = internal::inner_stride_at_compile_time<Derived>::ret,
+      OuterStrideAtCompileTime = internal::outer_stride_at_compile_time<Derived>::ret
+    };
+    
+    typedef typename internal::find_best_packet<Scalar,SizeAtCompileTime>::type PacketScalar;
+
+    enum { IsPlainObjectBase = 0 };
+    
+    /** The plain matrix type corresponding to this expression.
+      * \sa PlainObject */
+    typedef Matrix<typename internal::traits<Derived>::Scalar,
+                internal::traits<Derived>::RowsAtCompileTime,
+                internal::traits<Derived>::ColsAtCompileTime,
+                AutoAlign | (internal::traits<Derived>::Flags&RowMajorBit ? RowMajor : ColMajor),
+                internal::traits<Derived>::MaxRowsAtCompileTime,
+                internal::traits<Derived>::MaxColsAtCompileTime
+          > PlainMatrix;
+    
+    /** The plain array type corresponding to this expression.
+      * \sa PlainObject */
+    typedef Array<typename internal::traits<Derived>::Scalar,
+                internal::traits<Derived>::RowsAtCompileTime,
+                internal::traits<Derived>::ColsAtCompileTime,
+                AutoAlign | (internal::traits<Derived>::Flags&RowMajorBit ? RowMajor : ColMajor),
+                internal::traits<Derived>::MaxRowsAtCompileTime,
+                internal::traits<Derived>::MaxColsAtCompileTime
+          > PlainArray;
+
+    /** \brief The plain matrix or array type corresponding to this expression.
+      *
+      * This is not necessarily exactly the return type of eval(). In the case of plain matrices,
+      * the return type of eval() is a const reference to a matrix, not a matrix! It is however guaranteed
+      * that the return type of eval() is either PlainObject or const PlainObject&.
+      */
+    typedef typename internal::conditional<internal::is_same<typename internal::traits<Derived>::XprKind,MatrixXpr >::value,
+                                 PlainMatrix, PlainArray>::type PlainObject;
+
+    /** \returns the number of nonzero coefficients which is in practice the number
+      * of stored coefficients. */
+    EIGEN_DEVICE_FUNC
+    inline Index nonZeros() const { return size(); }
+
+    /** \returns the outer size.
+      *
+      * \note For a vector, this returns just 1. For a matrix (non-vector), this is the major dimension
+      * with respect to the \ref TopicStorageOrders "storage order", i.e., the number of columns for a
+      * column-major matrix, and the number of rows for a row-major matrix. */
+    EIGEN_DEVICE_FUNC
+    Index outerSize() const
+    {
+      return IsVectorAtCompileTime ? 1
+           : int(IsRowMajor) ? this->rows() : this->cols();
+    }
+
+    /** \returns the inner size.
+      *
+      * \note For a vector, this is just the size. For a matrix (non-vector), this is the minor dimension
+      * with respect to the \ref TopicStorageOrders "storage order", i.e., the number of rows for a 
+      * column-major matrix, and the number of columns for a row-major matrix. */
+    EIGEN_DEVICE_FUNC
+    Index innerSize() const
+    {
+      return IsVectorAtCompileTime ? this->size()
+           : int(IsRowMajor) ? this->cols() : this->rows();
+    }
+
+    /** Only plain matrices/arrays, not expressions, may be resized; therefore the only useful resize methods are
+      * Matrix::resize() and Array::resize(). The present method only asserts that the new size equals the old size, and does
+      * nothing else.
+      */
+    EIGEN_DEVICE_FUNC
+    void resize(Index newSize)
+    {
+      EIGEN_ONLY_USED_FOR_DEBUG(newSize);
+      eigen_assert(newSize == this->size()
+                && "DenseBase::resize() does not actually allow to resize.");
+    }
+    /** Only plain matrices/arrays, not expressions, may be resized; therefore the only useful resize methods are
+      * Matrix::resize() and Array::resize(). The present method only asserts that the new size equals the old size, and does
+      * nothing else.
+      */
+    EIGEN_DEVICE_FUNC
+    void resize(Index rows, Index cols)
+    {
+      EIGEN_ONLY_USED_FOR_DEBUG(rows);
+      EIGEN_ONLY_USED_FOR_DEBUG(cols);
+      eigen_assert(rows == this->rows() && cols == this->cols()
+                && "DenseBase::resize() does not actually allow to resize.");
+    }
+
+#ifndef EIGEN_PARSED_BY_DOXYGEN
+    /** \internal Represents a matrix with all coefficients equal to one another*/
+    typedef CwiseNullaryOp<internal::scalar_constant_op<Scalar>,PlainObject> ConstantReturnType;
+    /** \internal \deprecated Represents a vector with linearly spaced coefficients that allows sequential access only. */
+    typedef CwiseNullaryOp<internal::linspaced_op<Scalar,PacketScalar>,PlainObject> SequentialLinSpacedReturnType;
+    /** \internal Represents a vector with linearly spaced coefficients that allows random access. */
+    typedef CwiseNullaryOp<internal::linspaced_op<Scalar,PacketScalar>,PlainObject> RandomAccessLinSpacedReturnType;
+    /** \internal the return type of MatrixBase::eigenvalues() */
+    typedef Matrix<typename NumTraits<typename internal::traits<Derived>::Scalar>::Real, internal::traits<Derived>::ColsAtCompileTime, 1> EigenvaluesReturnType;
+
+#endif // not EIGEN_PARSED_BY_DOXYGEN
+
+    /** Copies \a other into *this. \returns a reference to *this. */
+    template<typename OtherDerived>
+    EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
+    Derived& operator=(const DenseBase<OtherDerived>& other);
+
+    /** Special case of the template operator=, in order to prevent the compiler
+      * from generating a default operator= (issue hit with g++ 4.1)
+      */
+    EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
+    Derived& operator=(const DenseBase& other);
+
+    template<typename OtherDerived>
+    EIGEN_DEVICE_FUNC
+    Derived& operator=(const EigenBase<OtherDerived> &other);
+
+    template<typename OtherDerived>
+    EIGEN_DEVICE_FUNC
+    Derived& operator+=(const EigenBase<OtherDerived> &other);
+
+    template<typename OtherDerived>
+    EIGEN_DEVICE_FUNC
+    Derived& operator-=(const EigenBase<OtherDerived> &other);
+
+    template<typename OtherDerived>
+    EIGEN_DEVICE_FUNC
+    Derived& operator=(const ReturnByValue<OtherDerived>& func);
+
+    /** \internal
+      * Copies \a other into *this without evaluating other. \returns a reference to *this.
+      * \deprecated */
+    template<typename OtherDerived>
+    EIGEN_DEVICE_FUNC
+    Derived& lazyAssign(const DenseBase<OtherDerived>& other);
+
+    EIGEN_DEVICE_FUNC
+    CommaInitializer<Derived> operator<< (const Scalar& s);
+
+    /** \deprecated it now returns \c *this */
+    template<unsigned int Added,unsigned int Removed>
+    EIGEN_DEPRECATED
+    const Derived& flagged() const
+    { return derived(); }
+
+    template<typename OtherDerived>
+    EIGEN_DEVICE_FUNC
+    CommaInitializer<Derived> operator<< (const DenseBase<OtherDerived>& other);
+
+    typedef Transpose<Derived> TransposeReturnType;
+    EIGEN_DEVICE_FUNC
+    TransposeReturnType transpose();
+    typedef typename internal::add_const<Transpose<const Derived> >::type ConstTransposeReturnType;
+    EIGEN_DEVICE_FUNC
+    ConstTransposeReturnType transpose() const;
+    EIGEN_DEVICE_FUNC
+    void transposeInPlace();
+
+    EIGEN_DEVICE_FUNC static const ConstantReturnType
+    Constant(Index rows, Index cols, const Scalar& value);
+    EIGEN_DEVICE_FUNC static const ConstantReturnType
+    Constant(Index size, const Scalar& value);
+    EIGEN_DEVICE_FUNC static const ConstantReturnType
+    Constant(const Scalar& value);
+
+    EIGEN_DEVICE_FUNC static const SequentialLinSpacedReturnType
+    LinSpaced(Sequential_t, Index size, const Scalar& low, const Scalar& high);
+    EIGEN_DEVICE_FUNC static const RandomAccessLinSpacedReturnType
+    LinSpaced(Index size, const Scalar& low, const Scalar& high);
+    EIGEN_DEVICE_FUNC static const SequentialLinSpacedReturnType
+    LinSpaced(Sequential_t, const Scalar& low, const Scalar& high);
+    EIGEN_DEVICE_FUNC static const RandomAccessLinSpacedReturnType
+    LinSpaced(const Scalar& low, const Scalar& high);
+
+    template<typename CustomNullaryOp> EIGEN_DEVICE_FUNC
+    static const CwiseNullaryOp<CustomNullaryOp, PlainObject>
+    NullaryExpr(Index rows, Index cols, const CustomNullaryOp& func);
+    template<typename CustomNullaryOp> EIGEN_DEVICE_FUNC
+    static const CwiseNullaryOp<CustomNullaryOp, PlainObject>
+    NullaryExpr(Index size, const CustomNullaryOp& func);
+    template<typename CustomNullaryOp> EIGEN_DEVICE_FUNC
+    static const CwiseNullaryOp<CustomNullaryOp, PlainObject>
+    NullaryExpr(const CustomNullaryOp& func);
+
+    EIGEN_DEVICE_FUNC static const ConstantReturnType Zero(Index rows, Index cols);
+    EIGEN_DEVICE_FUNC static const ConstantReturnType Zero(Index size);
+    EIGEN_DEVICE_FUNC static const ConstantReturnType Zero();
+    EIGEN_DEVICE_FUNC static const ConstantReturnType Ones(Index rows, Index cols);
+    EIGEN_DEVICE_FUNC static const ConstantReturnType Ones(Index size);
+    EIGEN_DEVICE_FUNC static const ConstantReturnType Ones();
+
+    EIGEN_DEVICE_FUNC void fill(const Scalar& value);
+    EIGEN_DEVICE_FUNC Derived& setConstant(const Scalar& value);
+    EIGEN_DEVICE_FUNC Derived& setLinSpaced(Index size, const Scalar& low, const Scalar& high);
+    EIGEN_DEVICE_FUNC Derived& setLinSpaced(const Scalar& low, const Scalar& high);
+    EIGEN_DEVICE_FUNC Derived& setZero();
+    EIGEN_DEVICE_FUNC Derived& setOnes();
+    EIGEN_DEVICE_FUNC Derived& setRandom();
+
+    template<typename OtherDerived> EIGEN_DEVICE_FUNC
+    bool isApprox(const DenseBase<OtherDerived>& other,
+                  const RealScalar& prec = NumTraits<Scalar>::dummy_precision()) const;
+    EIGEN_DEVICE_FUNC 
+    bool isMuchSmallerThan(const RealScalar& other,
+                           const RealScalar& prec = NumTraits<Scalar>::dummy_precision()) const;
+    template<typename OtherDerived> EIGEN_DEVICE_FUNC
+    bool isMuchSmallerThan(const DenseBase<OtherDerived>& other,
+                           const RealScalar& prec = NumTraits<Scalar>::dummy_precision()) const;
+
+    EIGEN_DEVICE_FUNC bool isApproxToConstant(const Scalar& value, const RealScalar& prec = NumTraits<Scalar>::dummy_precision()) const;
+    EIGEN_DEVICE_FUNC bool isConstant(const Scalar& value, const RealScalar& prec = NumTraits<Scalar>::dummy_precision()) const;
+    EIGEN_DEVICE_FUNC bool isZero(const RealScalar& prec = NumTraits<Scalar>::dummy_precision()) const;
+    EIGEN_DEVICE_FUNC bool isOnes(const RealScalar& prec = NumTraits<Scalar>::dummy_precision()) const;
+    
+    inline bool hasNaN() const;
+    inline bool allFinite() const;
+
+    EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
+    Derived& operator*=(const Scalar& other);
+    EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
+    Derived& operator/=(const Scalar& other);
+
+    typedef typename internal::add_const_on_value_type<typename internal::eval<Derived>::type>::type EvalReturnType;
+    /** \returns the matrix or vector obtained by evaluating this expression.
+      *
+      * Notice that in the case of a plain matrix or vector (not an expression) this function just returns
+      * a const reference, in order to avoid a useless copy.
+      * 
+      * \warning Be carefull with eval() and the auto C++ keyword, as detailed in this \link TopicPitfalls_auto_keyword page \endlink.
+      */
+    EIGEN_DEVICE_FUNC
+    EIGEN_STRONG_INLINE EvalReturnType eval() const
+    {
+      // Even though MSVC does not honor strong inlining when the return type
+      // is a dynamic matrix, we desperately need strong inlining for fixed
+      // size types on MSVC.
+      return typename internal::eval<Derived>::type(derived());
+    }
+    
+    /** swaps *this with the expression \a other.
+      *
+      */
+    template<typename OtherDerived>
+    EIGEN_DEVICE_FUNC
+    void swap(const DenseBase<OtherDerived>& other)
+    {
+      EIGEN_STATIC_ASSERT(!OtherDerived::IsPlainObjectBase,THIS_EXPRESSION_IS_NOT_A_LVALUE__IT_IS_READ_ONLY);
+      eigen_assert(rows()==other.rows() && cols()==other.cols());
+      call_assignment(derived(), other.const_cast_derived(), internal::swap_assign_op<Scalar>());
+    }
+
+    /** swaps *this with the matrix or array \a other.
+      *
+      */
+    template<typename OtherDerived>
+    EIGEN_DEVICE_FUNC
+    void swap(PlainObjectBase<OtherDerived>& other)
+    {
+      eigen_assert(rows()==other.rows() && cols()==other.cols());
+      call_assignment(derived(), other.derived(), internal::swap_assign_op<Scalar>());
+    }
+
+    EIGEN_DEVICE_FUNC inline const NestByValue<Derived> nestByValue() const;
+    EIGEN_DEVICE_FUNC inline const ForceAlignedAccess<Derived> forceAlignedAccess() const;
+    EIGEN_DEVICE_FUNC inline ForceAlignedAccess<Derived> forceAlignedAccess();
+    template<bool Enable> EIGEN_DEVICE_FUNC
+    inline const typename internal::conditional<Enable,ForceAlignedAccess<Derived>,Derived&>::type forceAlignedAccessIf() const;
+    template<bool Enable> EIGEN_DEVICE_FUNC
+    inline typename internal::conditional<Enable,ForceAlignedAccess<Derived>,Derived&>::type forceAlignedAccessIf();
+
+    EIGEN_DEVICE_FUNC Scalar sum() const;
+    EIGEN_DEVICE_FUNC Scalar mean() const;
+    EIGEN_DEVICE_FUNC Scalar trace() const;
+
+    EIGEN_DEVICE_FUNC Scalar prod() const;
+
+    EIGEN_DEVICE_FUNC typename internal::traits<Derived>::Scalar minCoeff() const;
+    EIGEN_DEVICE_FUNC typename internal::traits<Derived>::Scalar maxCoeff() const;
+
+    template<typename IndexType> EIGEN_DEVICE_FUNC
+    typename internal::traits<Derived>::Scalar minCoeff(IndexType* row, IndexType* col) const;
+    template<typename IndexType> EIGEN_DEVICE_FUNC
+    typename internal::traits<Derived>::Scalar maxCoeff(IndexType* row, IndexType* col) const;
+    template<typename IndexType> EIGEN_DEVICE_FUNC
+    typename internal::traits<Derived>::Scalar minCoeff(IndexType* index) const;
+    template<typename IndexType> EIGEN_DEVICE_FUNC
+    typename internal::traits<Derived>::Scalar maxCoeff(IndexType* index) const;
+
+    template<typename BinaryOp>
+    EIGEN_DEVICE_FUNC
+    Scalar redux(const BinaryOp& func) const;
+
+    template<typename Visitor>
+    EIGEN_DEVICE_FUNC
+    void visit(Visitor& func) const;
+
+    /** \returns a WithFormat proxy object allowing to print a matrix the with given
+      * format \a fmt.
+      *
+      * See class IOFormat for some examples.
+      *
+      * \sa class IOFormat, class WithFormat
+      */
+    inline const WithFormat<Derived> format(const IOFormat& fmt) const
+    {
+      return WithFormat<Derived>(derived(), fmt);
+    }
+
+    /** \returns the unique coefficient of a 1x1 expression */
+    EIGEN_DEVICE_FUNC
+    CoeffReturnType value() const
+    {
+      EIGEN_STATIC_ASSERT_SIZE_1x1(Derived)
+      eigen_assert(this->rows() == 1 && this->cols() == 1);
+      return derived().coeff(0,0);
+    }
+
+    EIGEN_DEVICE_FUNC bool all() const;
+    EIGEN_DEVICE_FUNC bool any() const;
+    EIGEN_DEVICE_FUNC Index count() const;
+
+    typedef VectorwiseOp<Derived, Horizontal> RowwiseReturnType;
+    typedef const VectorwiseOp<const Derived, Horizontal> ConstRowwiseReturnType;
+    typedef VectorwiseOp<Derived, Vertical> ColwiseReturnType;
+    typedef const VectorwiseOp<const Derived, Vertical> ConstColwiseReturnType;
+
+    /** \returns a VectorwiseOp wrapper of *this providing additional partial reduction operations
+    *
+    * Example: \include MatrixBase_rowwise.cpp
+    * Output: \verbinclude MatrixBase_rowwise.out
+    *
+    * \sa colwise(), class VectorwiseOp, \ref TutorialReductionsVisitorsBroadcasting
+    */
+    //Code moved here due to a CUDA compiler bug
+    EIGEN_DEVICE_FUNC inline ConstRowwiseReturnType rowwise() const {
+      return ConstRowwiseReturnType(derived());
+    }
+    EIGEN_DEVICE_FUNC RowwiseReturnType rowwise();
+
+    /** \returns a VectorwiseOp wrapper of *this providing additional partial reduction operations
+    *
+    * Example: \include MatrixBase_colwise.cpp
+    * Output: \verbinclude MatrixBase_colwise.out
+    *
+    * \sa rowwise(), class VectorwiseOp, \ref TutorialReductionsVisitorsBroadcasting
+    */
+    EIGEN_DEVICE_FUNC inline ConstColwiseReturnType colwise() const {
+      return ConstColwiseReturnType(derived());
+    }
+    EIGEN_DEVICE_FUNC ColwiseReturnType colwise();
+
+    typedef CwiseNullaryOp<internal::scalar_random_op<Scalar>,PlainObject> RandomReturnType;
+    static const RandomReturnType Random(Index rows, Index cols);
+    static const RandomReturnType Random(Index size);
+    static const RandomReturnType Random();
+
+    template<typename ThenDerived,typename ElseDerived>
+    const Select<Derived,ThenDerived,ElseDerived>
+    select(const DenseBase<ThenDerived>& thenMatrix,
+           const DenseBase<ElseDerived>& elseMatrix) const;
+
+    template<typename ThenDerived>
+    inline const Select<Derived,ThenDerived, typename ThenDerived::ConstantReturnType>
+    select(const DenseBase<ThenDerived>& thenMatrix, const typename ThenDerived::Scalar& elseScalar) const;
+
+    template<typename ElseDerived>
+    inline const Select<Derived, typename ElseDerived::ConstantReturnType, ElseDerived >
+    select(const typename ElseDerived::Scalar& thenScalar, const DenseBase<ElseDerived>& elseMatrix) const;
+
+    template<int p> RealScalar lpNorm() const;
+
+    template<int RowFactor, int ColFactor>
+    EIGEN_DEVICE_FUNC
+    const Replicate<Derived,RowFactor,ColFactor> replicate() const;
+    /**
+    * \return an expression of the replication of \c *this
+    *
+    * Example: \include MatrixBase_replicate_int_int.cpp
+    * Output: \verbinclude MatrixBase_replicate_int_int.out
+    *
+    * \sa VectorwiseOp::replicate(), DenseBase::replicate<int,int>(), class Replicate
+    */
+    //Code moved here due to a CUDA compiler bug
+    EIGEN_DEVICE_FUNC
+    const Replicate<Derived, Dynamic, Dynamic> replicate(Index rowFactor, Index colFactor) const
+    {
+      return Replicate<Derived, Dynamic, Dynamic>(derived(), rowFactor, colFactor);
+    }
+
+    typedef Reverse<Derived, BothDirections> ReverseReturnType;
+    typedef const Reverse<const Derived, BothDirections> ConstReverseReturnType;
+    EIGEN_DEVICE_FUNC ReverseReturnType reverse();
+    /** This is the const version of reverse(). */
+    //Code moved here due to a CUDA compiler bug
+    EIGEN_DEVICE_FUNC ConstReverseReturnType reverse() const
+    {
+      return ConstReverseReturnType(derived());
+    }
+    EIGEN_DEVICE_FUNC void reverseInPlace();
+
+#define EIGEN_CURRENT_STORAGE_BASE_CLASS Eigen::DenseBase
+#define EIGEN_DOC_BLOCK_ADDONS_NOT_INNER_PANEL
+#define EIGEN_DOC_BLOCK_ADDONS_INNER_PANEL_IF(COND)
+#   include "../plugins/BlockMethods.h"
+#   ifdef EIGEN_DENSEBASE_PLUGIN
+#     include EIGEN_DENSEBASE_PLUGIN
+#   endif
+#undef EIGEN_CURRENT_STORAGE_BASE_CLASS
+#undef EIGEN_DOC_BLOCK_ADDONS_NOT_INNER_PANEL
+#undef EIGEN_DOC_BLOCK_ADDONS_INNER_PANEL_IF
+
+    // disable the use of evalTo for dense objects with a nice compilation error
+    template<typename Dest>
+    EIGEN_DEVICE_FUNC
+    inline void evalTo(Dest& ) const
+    {
+      EIGEN_STATIC_ASSERT((internal::is_same<Dest,void>::value),THE_EVAL_EVALTO_FUNCTION_SHOULD_NEVER_BE_CALLED_FOR_DENSE_OBJECTS);
+    }
+
+  protected:
+    EIGEN_DEFAULT_COPY_CONSTRUCTOR(DenseBase)
+    /** Default constructor. Do nothing. */
+    EIGEN_DEVICE_FUNC DenseBase()
+    {
+      /* Just checks for self-consistency of the flags.
+       * Only do it when debugging Eigen, as this borders on paranoia and could slow compilation down
+       */
+#ifdef EIGEN_INTERNAL_DEBUGGING
+      EIGEN_STATIC_ASSERT((EIGEN_IMPLIES(MaxRowsAtCompileTime==1 && MaxColsAtCompileTime!=1, int(IsRowMajor))
+                        && EIGEN_IMPLIES(MaxColsAtCompileTime==1 && MaxRowsAtCompileTime!=1, int(!IsRowMajor))),
+                          INVALID_STORAGE_ORDER_FOR_THIS_VECTOR_EXPRESSION)
+#endif
+    }
+
+  private:
+    EIGEN_DEVICE_FUNC explicit DenseBase(int);
+    EIGEN_DEVICE_FUNC DenseBase(int,int);
+    template<typename OtherDerived> EIGEN_DEVICE_FUNC explicit DenseBase(const DenseBase<OtherDerived>&);
+};
+
+} // end namespace Eigen
+
+#endif // EIGEN_DENSEBASE_H

Eigen/src/Core/DenseCoeffsBase.h

@@ -0,0 +1,681 @@
+// This file is part of Eigen, a lightweight C++ template library
+// for linear algebra.
+//
+// Copyright (C) 2006-2010 Benoit Jacob <jacob.benoit.1@gmail.com>
+//
+// This Source Code Form is subject to the terms of the Mozilla
+// Public License v. 2.0. If a copy of the MPL was not distributed
+// with this file, You can obtain one at http://mozilla.org/MPL/2.0/.
+
+#ifndef EIGEN_DENSECOEFFSBASE_H
+#define EIGEN_DENSECOEFFSBASE_H
+
+namespace Eigen {
+
+namespace internal {
+template<typename T> struct add_const_on_value_type_if_arithmetic
+{
+  typedef typename conditional<is_arithmetic<T>::value, T, typename add_const_on_value_type<T>::type>::type type;
+};
+}
+
+/** \brief Base class providing read-only coefficient access to matrices and arrays.
+  * \ingroup Core_Module
+  * \tparam Derived Type of the derived class
+  * \tparam #ReadOnlyAccessors Constant indicating read-only access
+  *
+  * This class defines the \c operator() \c const function and friends, which can be used to read specific
+  * entries of a matrix or array.
+  * 
+  * \sa DenseCoeffsBase<Derived, WriteAccessors>, DenseCoeffsBase<Derived, DirectAccessors>,
+  *     \ref TopicClassHierarchy
+  */
+template<typename Derived>
+class DenseCoeffsBase<Derived,ReadOnlyAccessors> : public EigenBase<Derived>
+{
+  public:
+    typedef typename internal::traits<Derived>::StorageKind StorageKind;
+    typedef typename internal::traits<Derived>::Scalar Scalar;
+    typedef typename internal::packet_traits<Scalar>::type PacketScalar;
+
+    // Explanation for this CoeffReturnType typedef.
+    // - This is the return type of the coeff() method.
+    // - The LvalueBit means exactly that we can offer a coeffRef() method, which means exactly that we can get references
+    // to coeffs, which means exactly that we can have coeff() return a const reference (as opposed to returning a value).
+    // - The is_artihmetic check is required since "const int", "const double", etc. will cause warnings on some systems
+    // while the declaration of "const T", where T is a non arithmetic type does not. Always returning "const Scalar&" is
+    // not possible, since the underlying expressions might not offer a valid address the reference could be referring to.
+    typedef typename internal::conditional<bool(internal::traits<Derived>::Flags&LvalueBit),
+                         const Scalar&,
+                         typename internal::conditional<internal::is_arithmetic<Scalar>::value, Scalar, const Scalar>::type
+                     >::type CoeffReturnType;
+
+    typedef typename internal::add_const_on_value_type_if_arithmetic<
+                         typename internal::packet_traits<Scalar>::type
+                     >::type PacketReturnType;
+
+    typedef EigenBase<Derived> Base;
+    using Base::rows;
+    using Base::cols;
+    using Base::size;
+    using Base::derived;
+
+    EIGEN_DEVICE_FUNC
+    EIGEN_STRONG_INLINE Index rowIndexByOuterInner(Index outer, Index inner) const
+    {
+      return int(Derived::RowsAtCompileTime) == 1 ? 0
+          : int(Derived::ColsAtCompileTime) == 1 ? inner
+          : int(Derived::Flags)&RowMajorBit ? outer
+          : inner;
+    }
+
+    EIGEN_DEVICE_FUNC
+    EIGEN_STRONG_INLINE Index colIndexByOuterInner(Index outer, Index inner) const
+    {
+      return int(Derived::ColsAtCompileTime) == 1 ? 0
+          : int(Derived::RowsAtCompileTime) == 1 ? inner
+          : int(Derived::Flags)&RowMajorBit ? inner
+          : outer;
+    }
+
+    /** Short version: don't use this function, use
+      * \link operator()(Index,Index) const \endlink instead.
+      *
+      * Long version: this function is similar to
+      * \link operator()(Index,Index) const \endlink, but without the assertion.
+      * Use this for limiting the performance cost of debugging code when doing
+      * repeated coefficient access. Only use this when it is guaranteed that the
+      * parameters \a row and \a col are in range.
+      *
+      * If EIGEN_INTERNAL_DEBUGGING is defined, an assertion will be made, making this
+      * function equivalent to \link operator()(Index,Index) const \endlink.
+      *
+      * \sa operator()(Index,Index) const, coeffRef(Index,Index), coeff(Index) const
+      */
+    EIGEN_DEVICE_FUNC
+    EIGEN_STRONG_INLINE CoeffReturnType coeff(Index row, Index col) const
+    {
+      eigen_internal_assert(row >= 0 && row < rows()
+                         && col >= 0 && col < cols());
+      return internal::evaluator<Derived>(derived()).coeff(row,col);
+    }
+
+    EIGEN_DEVICE_FUNC
+    EIGEN_STRONG_INLINE CoeffReturnType coeffByOuterInner(Index outer, Index inner) const
+    {
+      return coeff(rowIndexByOuterInner(outer, inner),
+                   colIndexByOuterInner(outer, inner));
+    }
+
+    /** \returns the coefficient at given the given row and column.
+      *
+      * \sa operator()(Index,Index), operator[](Index)
+      */
+    EIGEN_DEVICE_FUNC
+    EIGEN_STRONG_INLINE CoeffReturnType operator()(Index row, Index col) const
+    {
+      eigen_assert(row >= 0 && row < rows()
+          && col >= 0 && col < cols());
+      return coeff(row, col);
+    }
+
+    /** Short version: don't use this function, use
+      * \link operator[](Index) const \endlink instead.
+      *
+      * Long version: this function is similar to
+      * \link operator[](Index) const \endlink, but without the assertion.
+      * Use this for limiting the performance cost of debugging code when doing
+      * repeated coefficient access. Only use this when it is guaranteed that the
+      * parameter \a index is in range.
+      *
+      * If EIGEN_INTERNAL_DEBUGGING is defined, an assertion will be made, making this
+      * function equivalent to \link operator[](Index) const \endlink.
+      *
+      * \sa operator[](Index) const, coeffRef(Index), coeff(Index,Index) const
+      */
+
+    EIGEN_DEVICE_FUNC
+    EIGEN_STRONG_INLINE CoeffReturnType
+    coeff(Index index) const
+    {
+      EIGEN_STATIC_ASSERT(internal::evaluator<Derived>::Flags & LinearAccessBit,
+                          THIS_COEFFICIENT_ACCESSOR_TAKING_ONE_ACCESS_IS_ONLY_FOR_EXPRESSIONS_ALLOWING_LINEAR_ACCESS)
+      eigen_internal_assert(index >= 0 && index < size());
+      return internal::evaluator<Derived>(derived()).coeff(index);
+    }
+
+
+    /** \returns the coefficient at given index.
+      *
+      * This method is allowed only for vector expressions, and for matrix expressions having the LinearAccessBit.
+      *
+      * \sa operator[](Index), operator()(Index,Index) const, x() const, y() const,
+      * z() const, w() const
+      */
+
+    EIGEN_DEVICE_FUNC
+    EIGEN_STRONG_INLINE CoeffReturnType
+    operator[](Index index) const
+    {
+      EIGEN_STATIC_ASSERT(Derived::IsVectorAtCompileTime,
+                          THE_BRACKET_OPERATOR_IS_ONLY_FOR_VECTORS__USE_THE_PARENTHESIS_OPERATOR_INSTEAD)
+      eigen_assert(index >= 0 && index < size());
+      return coeff(index);
+    }
+
+    /** \returns the coefficient at given index.
+      *
+      * This is synonymous to operator[](Index) const.
+      *
+      * This method is allowed only for vector expressions, and for matrix expressions having the LinearAccessBit.
+      *
+      * \sa operator[](Index), operator()(Index,Index) const, x() const, y() const,
+      * z() const, w() const
+      */
+
+    EIGEN_DEVICE_FUNC
+    EIGEN_STRONG_INLINE CoeffReturnType
+    operator()(Index index) const
+    {
+      eigen_assert(index >= 0 && index < size());
+      return coeff(index);
+    }
+
+    /** equivalent to operator[](0).  */
+
+    EIGEN_DEVICE_FUNC
+    EIGEN_STRONG_INLINE CoeffReturnType
+    x() const { return (*this)[0]; }
+
+    /** equivalent to operator[](1).  */
+
+    EIGEN_DEVICE_FUNC
+    EIGEN_STRONG_INLINE CoeffReturnType
+    y() const
+    {
+      EIGEN_STATIC_ASSERT(Derived::SizeAtCompileTime==-1 || Derived::SizeAtCompileTime>=2, OUT_OF_RANGE_ACCESS);
+      return (*this)[1];
+    }
+
+    /** equivalent to operator[](2).  */
+
+    EIGEN_DEVICE_FUNC
+    EIGEN_STRONG_INLINE CoeffReturnType
+    z() const
+    {
+      EIGEN_STATIC_ASSERT(Derived::SizeAtCompileTime==-1 || Derived::SizeAtCompileTime>=3, OUT_OF_RANGE_ACCESS);
+      return (*this)[2];
+    }
+
+    /** equivalent to operator[](3).  */
+
+    EIGEN_DEVICE_FUNC
+    EIGEN_STRONG_INLINE CoeffReturnType
+    w() const
+    {
+      EIGEN_STATIC_ASSERT(Derived::SizeAtCompileTime==-1 || Derived::SizeAtCompileTime>=4, OUT_OF_RANGE_ACCESS);
+      return (*this)[3];
+    }
+
+    /** \internal
+      * \returns the packet of coefficients starting at the given row and column. It is your responsibility
+      * to ensure that a packet really starts there. This method is only available on expressions having the
+      * PacketAccessBit.
+      *
+      * The \a LoadMode parameter may have the value \a #Aligned or \a #Unaligned. Its effect is to select
+      * the appropriate vectorization instruction. Aligned access is faster, but is only possible for packets
+      * starting at an address which is a multiple of the packet size.
+      */
+
+    template<int LoadMode>
+    EIGEN_STRONG_INLINE PacketReturnType packet(Index row, Index col) const
+    {
+      typedef typename internal::packet_traits<Scalar>::type DefaultPacketType;
+      eigen_internal_assert(row >= 0 && row < rows() && col >= 0 && col < cols());
+      return internal::evaluator<Derived>(derived()).template packet<LoadMode,DefaultPacketType>(row,col);
+    }
+
+
+    /** \internal */
+    template<int LoadMode>
+    EIGEN_STRONG_INLINE PacketReturnType packetByOuterInner(Index outer, Index inner) const
+    {
+      return packet<LoadMode>(rowIndexByOuterInner(outer, inner),
+                              colIndexByOuterInner(outer, inner));
+    }
+
+    /** \internal
+      * \returns the packet of coefficients starting at the given index. It is your responsibility
+      * to ensure that a packet really starts there. This method is only available on expressions having the
+      * PacketAccessBit and the LinearAccessBit.
+      *
+      * The \a LoadMode parameter may have the value \a #Aligned or \a #Unaligned. Its effect is to select
+      * the appropriate vectorization instruction. Aligned access is faster, but is only possible for packets
+      * starting at an address which is a multiple of the packet size.
+      */
+
+    template<int LoadMode>
+    EIGEN_STRONG_INLINE PacketReturnType packet(Index index) const
+    {
+      EIGEN_STATIC_ASSERT(internal::evaluator<Derived>::Flags & LinearAccessBit,
+                          THIS_COEFFICIENT_ACCESSOR_TAKING_ONE_ACCESS_IS_ONLY_FOR_EXPRESSIONS_ALLOWING_LINEAR_ACCESS)
+      typedef typename internal::packet_traits<Scalar>::type DefaultPacketType;
+      eigen_internal_assert(index >= 0 && index < size());
+      return internal::evaluator<Derived>(derived()).template packet<LoadMode,DefaultPacketType>(index);
+    }
+
+  protected:
+    // explanation: DenseBase is doing "using ..." on the methods from DenseCoeffsBase.
+    // But some methods are only available in the DirectAccess case.
+    // So we add dummy methods here with these names, so that "using... " doesn't fail.
+    // It's not private so that the child class DenseBase can access them, and it's not public
+    // either since it's an implementation detail, so has to be protected.
+    void coeffRef();
+    void coeffRefByOuterInner();
+    void writePacket();
+    void writePacketByOuterInner();
+    void copyCoeff();
+    void copyCoeffByOuterInner();
+    void copyPacket();
+    void copyPacketByOuterInner();
+    void stride();
+    void innerStride();
+    void outerStride();
+    void rowStride();
+    void colStride();
+};
+
+/** \brief Base class providing read/write coefficient access to matrices and arrays.
+  * \ingroup Core_Module
+  * \tparam Derived Type of the derived class
+  * \tparam #WriteAccessors Constant indicating read/write access
+  *
+  * This class defines the non-const \c operator() function and friends, which can be used to write specific
+  * entries of a matrix or array. This class inherits DenseCoeffsBase<Derived, ReadOnlyAccessors> which
+  * defines the const variant for reading specific entries.
+  * 
+  * \sa DenseCoeffsBase<Derived, DirectAccessors>, \ref TopicClassHierarchy
+  */
+template<typename Derived>
+class DenseCoeffsBase<Derived, WriteAccessors> : public DenseCoeffsBase<Derived, ReadOnlyAccessors>
+{
+  public:
+
+    typedef DenseCoeffsBase<Derived, ReadOnlyAccessors> Base;
+
+    typedef typename internal::traits<Derived>::StorageKind StorageKind;
+    typedef typename internal::traits<Derived>::Scalar Scalar;
+    typedef typename internal::packet_traits<Scalar>::type PacketScalar;
+    typedef typename NumTraits<Scalar>::Real RealScalar;
+
+    using Base::coeff;
+    using Base::rows;
+    using Base::cols;
+    using Base::size;
+    using Base::derived;
+    using Base::rowIndexByOuterInner;
+    using Base::colIndexByOuterInner;
+    using Base::operator[];
+    using Base::operator();
+    using Base::x;
+    using Base::y;
+    using Base::z;
+    using Base::w;
+
+    /** Short version: don't use this function, use
+      * \link operator()(Index,Index) \endlink instead.
+      *
+      * Long version: this function is similar to
+      * \link operator()(Index,Index) \endlink, but without the assertion.
+      * Use this for limiting the performance cost of debugging code when doing
+      * repeated coefficient access. Only use this when it is guaranteed that the
+      * parameters \a row and \a col are in range.
+      *
+      * If EIGEN_INTERNAL_DEBUGGING is defined, an assertion will be made, making this
+      * function equivalent to \link operator()(Index,Index) \endlink.
+      *
+      * \sa operator()(Index,Index), coeff(Index, Index) const, coeffRef(Index)
+      */
+    EIGEN_DEVICE_FUNC
+    EIGEN_STRONG_INLINE Scalar& coeffRef(Index row, Index col)
+    {
+      eigen_internal_assert(row >= 0 && row < rows()
+                         && col >= 0 && col < cols());
+      return internal::evaluator<Derived>(derived()).coeffRef(row,col);
+    }
+
+    EIGEN_DEVICE_FUNC
+    EIGEN_STRONG_INLINE Scalar&
+    coeffRefByOuterInner(Index outer, Index inner)
+    {
+      return coeffRef(rowIndexByOuterInner(outer, inner),
+                      colIndexByOuterInner(outer, inner));
+    }
+
+    /** \returns a reference to the coefficient at given the given row and column.
+      *
+      * \sa operator[](Index)
+      */
+
+    EIGEN_DEVICE_FUNC
+    EIGEN_STRONG_INLINE Scalar&
+    operator()(Index row, Index col)
+    {
+      eigen_assert(row >= 0 && row < rows()
+          && col >= 0 && col < cols());
+      return coeffRef(row, col);
+    }
+
+
+    /** Short version: don't use this function, use
+      * \link operator[](Index) \endlink instead.
+      *
+      * Long version: this function is similar to
+      * \link operator[](Index) \endlink, but without the assertion.
+      * Use this for limiting the performance cost of debugging code when doing
+      * repeated coefficient access. Only use this when it is guaranteed that the
+      * parameters \a row and \a col are in range.
+      *
+      * If EIGEN_INTERNAL_DEBUGGING is defined, an assertion will be made, making this
+      * function equivalent to \link operator[](Index) \endlink.
+      *
+      * \sa operator[](Index), coeff(Index) const, coeffRef(Index,Index)
+      */
+
+    EIGEN_DEVICE_FUNC
+    EIGEN_STRONG_INLINE Scalar&
+    coeffRef(Index index)
+    {
+      EIGEN_STATIC_ASSERT(internal::evaluator<Derived>::Flags & LinearAccessBit,
+                          THIS_COEFFICIENT_ACCESSOR_TAKING_ONE_ACCESS_IS_ONLY_FOR_EXPRESSIONS_ALLOWING_LINEAR_ACCESS)
+      eigen_internal_assert(index >= 0 && index < size());
+      return internal::evaluator<Derived>(derived()).coeffRef(index);
+    }
+
+    /** \returns a reference to the coefficient at given index.
+      *
+      * This method is allowed only for vector expressions, and for matrix expressions having the LinearAccessBit.
+      *
+      * \sa operator[](Index) const, operator()(Index,Index), x(), y(), z(), w()
+      */
+
+    EIGEN_DEVICE_FUNC
+    EIGEN_STRONG_INLINE Scalar&
+    operator[](Index index)
+    {
+      EIGEN_STATIC_ASSERT(Derived::IsVectorAtCompileTime,
+                          THE_BRACKET_OPERATOR_IS_ONLY_FOR_VECTORS__USE_THE_PARENTHESIS_OPERATOR_INSTEAD)
+      eigen_assert(index >= 0 && index < size());
+      return coeffRef(index);
+    }
+
+    /** \returns a reference to the coefficient at given index.
+      *
+      * This is synonymous to operator[](Index).
+      *
+      * This method is allowed only for vector expressions, and for matrix expressions having the LinearAccessBit.
+      *
+      * \sa operator[](Index) const, operator()(Index,Index), x(), y(), z(), w()
+      */
+
+    EIGEN_DEVICE_FUNC
+    EIGEN_STRONG_INLINE Scalar&
+    operator()(Index index)
+    {
+      eigen_assert(index >= 0 && index < size());
+      return coeffRef(index);
+    }
+
+    /** equivalent to operator[](0).  */
+
+    EIGEN_DEVICE_FUNC
+    EIGEN_STRONG_INLINE Scalar&
+    x() { return (*this)[0]; }
+
+    /** equivalent to operator[](1).  */
+
+    EIGEN_DEVICE_FUNC
+    EIGEN_STRONG_INLINE Scalar&
+    y()
+    {
+      EIGEN_STATIC_ASSERT(Derived::SizeAtCompileTime==-1 || Derived::SizeAtCompileTime>=2, OUT_OF_RANGE_ACCESS);
+      return (*this)[1];
+    }
+
+    /** equivalent to operator[](2).  */
+
+    EIGEN_DEVICE_FUNC
+    EIGEN_STRONG_INLINE Scalar&
+    z()
+    {
+      EIGEN_STATIC_ASSERT(Derived::SizeAtCompileTime==-1 || Derived::SizeAtCompileTime>=3, OUT_OF_RANGE_ACCESS);
+      return (*this)[2];
+    }
+
+    /** equivalent to operator[](3).  */
+
+    EIGEN_DEVICE_FUNC
+    EIGEN_STRONG_INLINE Scalar&
+    w()
+    {
+      EIGEN_STATIC_ASSERT(Derived::SizeAtCompileTime==-1 || Derived::SizeAtCompileTime>=4, OUT_OF_RANGE_ACCESS);
+      return (*this)[3];
+    }
+};
+
+/** \brief Base class providing direct read-only coefficient access to matrices and arrays.
+  * \ingroup Core_Module
+  * \tparam Derived Type of the derived class
+  * \tparam #DirectAccessors Constant indicating direct access
+  *
+  * This class defines functions to work with strides which can be used to access entries directly. This class
+  * inherits DenseCoeffsBase<Derived, ReadOnlyAccessors> which defines functions to access entries read-only using
+  * \c operator() .
+  *
+  * \sa \blank \ref TopicClassHierarchy
+  */
+template<typename Derived>
+class DenseCoeffsBase<Derived, DirectAccessors> : public DenseCoeffsBase<Derived, ReadOnlyAccessors>
+{
+  public:
+
+    typedef DenseCoeffsBase<Derived, ReadOnlyAccessors> Base;
+    typedef typename internal::traits<Derived>::Scalar Scalar;
+    typedef typename NumTraits<Scalar>::Real RealScalar;
+
+    using Base::rows;
+    using Base::cols;
+    using Base::size;
+    using Base::derived;
+
+    /** \returns the pointer increment between two consecutive elements within a slice in the inner direction.
+      *
+      * \sa outerStride(), rowStride(), colStride()
+      */
+    EIGEN_DEVICE_FUNC
+    inline Index innerStride() const
+    {
+      return derived().innerStride();
+    }
+
+    /** \returns the pointer increment between two consecutive inner slices (for example, between two consecutive columns
+      *          in a column-major matrix).
+      *
+      * \sa innerStride(), rowStride(), colStride()
+      */
+    EIGEN_DEVICE_FUNC
+    inline Index outerStride() const
+    {
+      return derived().outerStride();
+    }
+
+    // FIXME shall we remove it ?
+    inline Index stride() const
+    {
+      return Derived::IsVectorAtCompileTime ? innerStride() : outerStride();
+    }
+
+    /** \returns the pointer increment between two consecutive rows.
+      *
+      * \sa innerStride(), outerStride(), colStride()
+      */
+    EIGEN_DEVICE_FUNC
+    inline Index rowStride() const
+    {
+      return Derived::IsRowMajor ? outerStride() : innerStride();
+    }
+
+    /** \returns the pointer increment between two consecutive columns.
+      *
+      * \sa innerStride(), outerStride(), rowStride()
+      */
+    EIGEN_DEVICE_FUNC
+    inline Index colStride() const
+    {
+      return Derived::IsRowMajor ? innerStride() : outerStride();
+    }
+};
+
+/** \brief Base class providing direct read/write coefficient access to matrices and arrays.
+  * \ingroup Core_Module
+  * \tparam Derived Type of the derived class
+  * \tparam #DirectWriteAccessors Constant indicating direct access
+  *
+  * This class defines functions to work with strides which can be used to access entries directly. This class
+  * inherits DenseCoeffsBase<Derived, WriteAccessors> which defines functions to access entries read/write using
+  * \c operator().
+  *
+  * \sa \blank \ref TopicClassHierarchy
+  */
+template<typename Derived>
+class DenseCoeffsBase<Derived, DirectWriteAccessors>
+  : public DenseCoeffsBase<Derived, WriteAccessors>
+{
+  public:
+
+    typedef DenseCoeffsBase<Derived, WriteAccessors> Base;
+    typedef typename internal::traits<Derived>::Scalar Scalar;
+    typedef typename NumTraits<Scalar>::Real RealScalar;
+
+    using Base::rows;
+    using Base::cols;
+    using Base::size;
+    using Base::derived;
+
+    /** \returns the pointer increment between two consecutive elements within a slice in the inner direction.
+      *
+      * \sa outerStride(), rowStride(), colStride()
+      */
+    EIGEN_DEVICE_FUNC
+    inline Index innerStride() const
+    {
+      return derived().innerStride();
+    }
+
+    /** \returns the pointer increment between two consecutive inner slices (for example, between two consecutive columns
+      *          in a column-major matrix).
+      *
+      * \sa innerStride(), rowStride(), colStride()
+      */
+    EIGEN_DEVICE_FUNC
+    inline Index outerStride() const
+    {
+      return derived().outerStride();
+    }
+
+    // FIXME shall we remove it ?
+    inline Index stride() const
+    {
+      return Derived::IsVectorAtCompileTime ? innerStride() : outerStride();
+    }
+
+    /** \returns the pointer increment between two consecutive rows.
+      *
+      * \sa innerStride(), outerStride(), colStride()
+      */
+    EIGEN_DEVICE_FUNC
+    inline Index rowStride() const
+    {
+      return Derived::IsRowMajor ? outerStride() : innerStride();
+    }
+
+    /** \returns the pointer increment between two consecutive columns.
+      *
+      * \sa innerStride(), outerStride(), rowStride()
+      */
+    EIGEN_DEVICE_FUNC
+    inline Index colStride() const
+    {
+      return Derived::IsRowMajor ? innerStride() : outerStride();
+    }
+};
+
+namespace internal {
+
+template<int Alignment, typename Derived, bool JustReturnZero>
+struct first_aligned_impl
+{
+  static inline Index run(const Derived&)
+  { return 0; }
+};
+
+template<int Alignment, typename Derived>
+struct first_aligned_impl<Alignment, Derived, false>
+{
+  static inline Index run(const Derived& m)
+  {
+    return internal::first_aligned<Alignment>(m.data(), m.size());
+  }
+};
+
+/** \internal \returns the index of the first element of the array stored by \a m that is properly aligned with respect to \a Alignment for vectorization.
+  *
+  * \tparam Alignment requested alignment in Bytes.
+  *
+  * There is also the variant first_aligned(const Scalar*, Integer) defined in Memory.h. See it for more
+  * documentation.
+  */
+template<int Alignment, typename Derived>
+static inline Index first_aligned(const DenseBase<Derived>& m)
+{
+  enum { ReturnZero = (int(evaluator<Derived>::Alignment) >= Alignment) || !(Derived::Flags & DirectAccessBit) };
+  return first_aligned_impl<Alignment, Derived, ReturnZero>::run(m.derived());
+}
+
+template<typename Derived>
+static inline Index first_default_aligned(const DenseBase<Derived>& m)
+{
+  typedef typename Derived::Scalar Scalar;
+  typedef typename packet_traits<Scalar>::type DefaultPacketType;
+  return internal::first_aligned<int(unpacket_traits<DefaultPacketType>::alignment),Derived>(m);
+}
+
+template<typename Derived, bool HasDirectAccess = has_direct_access<Derived>::ret>
+struct inner_stride_at_compile_time
+{
+  enum { ret = traits<Derived>::InnerStrideAtCompileTime };
+};
+
+template<typename Derived>
+struct inner_stride_at_compile_time<Derived, false>
+{
+  enum { ret = 0 };
+};
+
+template<typename Derived, bool HasDirectAccess = has_direct_access<Derived>::ret>
+struct outer_stride_at_compile_time
+{
+  enum { ret = traits<Derived>::OuterStrideAtCompileTime };
+};
+
+template<typename Derived>
+struct outer_stride_at_compile_time<Derived, false>
+{
+  enum { ret = 0 };
+};
+
+} // end namespace internal
+
+} // end namespace Eigen
+
+#endif // EIGEN_DENSECOEFFSBASE_H

Eigen/src/Core/DenseStorage.h

@@ -0,0 +1,570 @@
+// This file is part of Eigen, a lightweight C++ template library
+// for linear algebra.
+//
+// Copyright (C) 2008 Gael Guennebaud <gael.guennebaud@inria.fr>
+// Copyright (C) 2006-2009 Benoit Jacob <jacob.benoit.1@gmail.com>
+// Copyright (C) 2010-2013 Hauke Heibel <hauke.heibel@gmail.com>
+//
+// This Source Code Form is subject to the terms of the Mozilla
+// Public License v. 2.0. If a copy of the MPL was not distributed
+// with this file, You can obtain one at http://mozilla.org/MPL/2.0/.
+
+#ifndef EIGEN_MATRIXSTORAGE_H
+#define EIGEN_MATRIXSTORAGE_H
+
+#ifdef EIGEN_DENSE_STORAGE_CTOR_PLUGIN
+  #define EIGEN_INTERNAL_DENSE_STORAGE_CTOR_PLUGIN(X) X; EIGEN_DENSE_STORAGE_CTOR_PLUGIN;
+#else
+  #define EIGEN_INTERNAL_DENSE_STORAGE_CTOR_PLUGIN(X)
+#endif
+
+namespace Eigen {
+
+namespace internal {
+
+struct constructor_without_unaligned_array_assert {};
+
+template<typename T, int Size>
+EIGEN_DEVICE_FUNC
+void check_static_allocation_size()
+{
+  // if EIGEN_STACK_ALLOCATION_LIMIT is defined to 0, then no limit
+  #if EIGEN_STACK_ALLOCATION_LIMIT
+  EIGEN_STATIC_ASSERT(Size * sizeof(T) <= EIGEN_STACK_ALLOCATION_LIMIT, OBJECT_ALLOCATED_ON_STACK_IS_TOO_BIG);
+  #endif
+}
+
+/** \internal
+  * Static array. If the MatrixOrArrayOptions require auto-alignment, the array will be automatically aligned:
+  * to 16 bytes boundary if the total size is a multiple of 16 bytes.
+  */
+template <typename T, int Size, int MatrixOrArrayOptions,
+          int Alignment = (MatrixOrArrayOptions&DontAlign) ? 0
+                        : compute_default_alignment<T,Size>::value >
+struct plain_array
+{
+  T array[Size];
+
+  EIGEN_DEVICE_FUNC
+  plain_array()
+  { 
+    check_static_allocation_size<T,Size>();
+  }
+
+  EIGEN_DEVICE_FUNC
+  plain_array(constructor_without_unaligned_array_assert)
+  { 
+    check_static_allocation_size<T,Size>();
+  }
+};
+
+#if defined(EIGEN_DISABLE_UNALIGNED_ARRAY_ASSERT)
+  #define EIGEN_MAKE_UNALIGNED_ARRAY_ASSERT(sizemask)
+#elif EIGEN_GNUC_AT_LEAST(4,7) 
+  // GCC 4.7 is too aggressive in its optimizations and remove the alignement test based on the fact the array is declared to be aligned.
+  // See this bug report: http://gcc.gnu.org/bugzilla/show_bug.cgi?id=53900
+  // Hiding the origin of the array pointer behind a function argument seems to do the trick even if the function is inlined:
+  template<typename PtrType>
+  EIGEN_ALWAYS_INLINE PtrType eigen_unaligned_array_assert_workaround_gcc47(PtrType array) { return array; }
+  #define EIGEN_MAKE_UNALIGNED_ARRAY_ASSERT(sizemask) \
+    eigen_assert((internal::UIntPtr(eigen_unaligned_array_assert_workaround_gcc47(array)) & (sizemask)) == 0 \
+              && "this assertion is explained here: " \
+              "http://eigen.tuxfamily.org/dox-devel/group__TopicUnalignedArrayAssert.html" \
+              " **** READ THIS WEB PAGE !!! ****");
+#else
+  #define EIGEN_MAKE_UNALIGNED_ARRAY_ASSERT(sizemask) \
+    eigen_assert((internal::UIntPtr(array) & (sizemask)) == 0 \
+              && "this assertion is explained here: " \
+              "http://eigen.tuxfamily.org/dox-devel/group__TopicUnalignedArrayAssert.html" \
+              " **** READ THIS WEB PAGE !!! ****");
+#endif
+
+template <typename T, int Size, int MatrixOrArrayOptions>
+struct plain_array<T, Size, MatrixOrArrayOptions, 8>
+{
+  EIGEN_ALIGN_TO_BOUNDARY(8) T array[Size];
+
+  EIGEN_DEVICE_FUNC
+  plain_array() 
+  {
+    EIGEN_MAKE_UNALIGNED_ARRAY_ASSERT(7);
+    check_static_allocation_size<T,Size>();
+  }
+
+  EIGEN_DEVICE_FUNC
+  plain_array(constructor_without_unaligned_array_assert) 
+  { 
+    check_static_allocation_size<T,Size>();
+  }
+};
+
+template <typename T, int Size, int MatrixOrArrayOptions>
+struct plain_array<T, Size, MatrixOrArrayOptions, 16>
+{
+  EIGEN_ALIGN_TO_BOUNDARY(16) T array[Size];
+
+  EIGEN_DEVICE_FUNC
+  plain_array() 
+  { 
+    EIGEN_MAKE_UNALIGNED_ARRAY_ASSERT(15);
+    check_static_allocation_size<T,Size>();
+  }
+
+  EIGEN_DEVICE_FUNC
+  plain_array(constructor_without_unaligned_array_assert) 
+  { 
+    check_static_allocation_size<T,Size>();
+  }
+};
+
+template <typename T, int Size, int MatrixOrArrayOptions>
+struct plain_array<T, Size, MatrixOrArrayOptions, 32>
+{
+  EIGEN_ALIGN_TO_BOUNDARY(32) T array[Size];
+
+  EIGEN_DEVICE_FUNC
+  plain_array() 
+  {
+    EIGEN_MAKE_UNALIGNED_ARRAY_ASSERT(31);
+    check_static_allocation_size<T,Size>();
+  }
+
+  EIGEN_DEVICE_FUNC
+  plain_array(constructor_without_unaligned_array_assert) 
+  { 
+    check_static_allocation_size<T,Size>();
+  }
+};
+
+template <typename T, int Size, int MatrixOrArrayOptions>
+struct plain_array<T, Size, MatrixOrArrayOptions, 64>
+{
+  EIGEN_ALIGN_TO_BOUNDARY(64) T array[Size];
+
+  EIGEN_DEVICE_FUNC
+  plain_array() 
+  { 
+    EIGEN_MAKE_UNALIGNED_ARRAY_ASSERT(63);
+    check_static_allocation_size<T,Size>();
+  }
+
+  EIGEN_DEVICE_FUNC
+  plain_array(constructor_without_unaligned_array_assert) 
+  { 
+    check_static_allocation_size<T,Size>();
+  }
+};
+
+template <typename T, int MatrixOrArrayOptions, int Alignment>
+struct plain_array<T, 0, MatrixOrArrayOptions, Alignment>
+{
+  T array[1];
+  EIGEN_DEVICE_FUNC plain_array() {}
+  EIGEN_DEVICE_FUNC plain_array(constructor_without_unaligned_array_assert) {}
+};
+
+} // end namespace internal
+
+/** \internal
+  *
+  * \class DenseStorage
+  * \ingroup Core_Module
+  *
+  * \brief Stores the data of a matrix
+  *
+  * This class stores the data of fixed-size, dynamic-size or mixed matrices
+  * in a way as compact as possible.
+  *
+  * \sa Matrix
+  */
+template<typename T, int Size, int _Rows, int _Cols, int _Options> class DenseStorage;
+
+// purely fixed-size matrix
+template<typename T, int Size, int _Rows, int _Cols, int _Options> class DenseStorage
+{
+    internal::plain_array<T,Size,_Options> m_data;
+  public:
+    EIGEN_DEVICE_FUNC DenseStorage() {
+      EIGEN_INTERNAL_DENSE_STORAGE_CTOR_PLUGIN(Index size = Size)
+    }
+    EIGEN_DEVICE_FUNC
+    explicit DenseStorage(internal::constructor_without_unaligned_array_assert)
+      : m_data(internal::constructor_without_unaligned_array_assert()) {}
+    EIGEN_DEVICE_FUNC 
+    DenseStorage(const DenseStorage& other) : m_data(other.m_data) {
+      EIGEN_INTERNAL_DENSE_STORAGE_CTOR_PLUGIN(Index size = Size)
+    }
+    EIGEN_DEVICE_FUNC 
+    DenseStorage& operator=(const DenseStorage& other)
+    { 
+      if (this != &other) m_data = other.m_data;
+      return *this; 
+    }
+    EIGEN_DEVICE_FUNC DenseStorage(Index size, Index rows, Index cols) {
+      EIGEN_INTERNAL_DENSE_STORAGE_CTOR_PLUGIN({})
+      eigen_internal_assert(size==rows*cols && rows==_Rows && cols==_Cols);
+      EIGEN_UNUSED_VARIABLE(size);
+      EIGEN_UNUSED_VARIABLE(rows);
+      EIGEN_UNUSED_VARIABLE(cols);
+    }
+    EIGEN_DEVICE_FUNC void swap(DenseStorage& other) { std::swap(m_data,other.m_data); }
+    EIGEN_DEVICE_FUNC static Index rows(void) {return _Rows;}
+    EIGEN_DEVICE_FUNC static Index cols(void) {return _Cols;}
+    EIGEN_DEVICE_FUNC void conservativeResize(Index,Index,Index) {}
+    EIGEN_DEVICE_FUNC void resize(Index,Index,Index) {}
+    EIGEN_DEVICE_FUNC const T *data() const { return m_data.array; }
+    EIGEN_DEVICE_FUNC T *data() { return m_data.array; }
+};
+
+// null matrix
+template<typename T, int _Rows, int _Cols, int _Options> class DenseStorage<T, 0, _Rows, _Cols, _Options>
+{
+  public:
+    EIGEN_DEVICE_FUNC DenseStorage() {}
+    EIGEN_DEVICE_FUNC explicit DenseStorage(internal::constructor_without_unaligned_array_assert) {}
+    EIGEN_DEVICE_FUNC DenseStorage(const DenseStorage&) {}
+    EIGEN_DEVICE_FUNC DenseStorage& operator=(const DenseStorage&) { return *this; }
+    EIGEN_DEVICE_FUNC DenseStorage(Index,Index,Index) {}
+    EIGEN_DEVICE_FUNC void swap(DenseStorage& ) {}
+    EIGEN_DEVICE_FUNC static Index rows(void) {return _Rows;}
+    EIGEN_DEVICE_FUNC static Index cols(void) {return _Cols;}
+    EIGEN_DEVICE_FUNC void conservativeResize(Index,Index,Index) {}
+    EIGEN_DEVICE_FUNC void resize(Index,Index,Index) {}
+    EIGEN_DEVICE_FUNC const T *data() const { return 0; }
+    EIGEN_DEVICE_FUNC 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>
+{
+    internal::plain_array<T,Size,_Options> m_data;
+    Index m_rows;
+    Index m_cols;
+  public:
+    EIGEN_DEVICE_FUNC DenseStorage() : m_rows(0), m_cols(0) {}
+    EIGEN_DEVICE_FUNC explicit DenseStorage(internal::constructor_without_unaligned_array_assert)
+      : m_data(internal::constructor_without_unaligned_array_assert()), m_rows(0), m_cols(0) {}
+    EIGEN_DEVICE_FUNC DenseStorage(const DenseStorage& other) : m_data(other.m_data), m_rows(other.m_rows), m_cols(other.m_cols) {}
+    EIGEN_DEVICE_FUNC DenseStorage& operator=(const DenseStorage& other) 
+    { 
+      if (this != &other)
+      {
+        m_data = other.m_data;
+        m_rows = other.m_rows;
+        m_cols = other.m_cols;
+      }
+      return *this; 
+    }
+    EIGEN_DEVICE_FUNC DenseStorage(Index, Index rows, Index cols) : m_rows(rows), m_cols(cols) {}
+    EIGEN_DEVICE_FUNC void swap(DenseStorage& other)
+    { std::swap(m_data,other.m_data); std::swap(m_rows,other.m_rows); std::swap(m_cols,other.m_cols); }
+    EIGEN_DEVICE_FUNC Index rows() const {return m_rows;}
+    EIGEN_DEVICE_FUNC Index cols() const {return m_cols;}
+    EIGEN_DEVICE_FUNC void conservativeResize(Index, Index rows, Index cols) { m_rows = rows; m_cols = cols; }
+    EIGEN_DEVICE_FUNC void resize(Index, Index rows, Index cols) { m_rows = rows; m_cols = cols; }
+    EIGEN_DEVICE_FUNC const T *data() const { return m_data.array; }
+    EIGEN_DEVICE_FUNC T *data() { return m_data.array; }
+};
+
+// dynamic-size matrix with fixed-size storage and fixed width
+template<typename T, int Size, int _Cols, int _Options> class DenseStorage<T, Size, Dynamic, _Cols, _Options>
+{
+    internal::plain_array<T,Size,_Options> m_data;
+    Index m_rows;
+  public:
+    EIGEN_DEVICE_FUNC DenseStorage() : m_rows(0) {}
+    EIGEN_DEVICE_FUNC explicit DenseStorage(internal::constructor_without_unaligned_array_assert)
+      : m_data(internal::constructor_without_unaligned_array_assert()), m_rows(0) {}
+    EIGEN_DEVICE_FUNC DenseStorage(const DenseStorage& other) : m_data(other.m_data), m_rows(other.m_rows) {}
+    EIGEN_DEVICE_FUNC DenseStorage& operator=(const DenseStorage& other) 
+    {
+      if (this != &other)
+      {
+        m_data = other.m_data;
+        m_rows = other.m_rows;
+      }
+      return *this; 
+    }
+    EIGEN_DEVICE_FUNC DenseStorage(Index, Index rows, Index) : m_rows(rows) {}
+    EIGEN_DEVICE_FUNC void swap(DenseStorage& other) { std::swap(m_data,other.m_data); std::swap(m_rows,other.m_rows); }
+    EIGEN_DEVICE_FUNC Index rows(void) const {return m_rows;}
+    EIGEN_DEVICE_FUNC Index cols(void) const {return _Cols;}
+    EIGEN_DEVICE_FUNC void conservativeResize(Index, Index rows, Index) { m_rows = rows; }
+    EIGEN_DEVICE_FUNC void resize(Index, Index rows, Index) { m_rows = rows; }
+    EIGEN_DEVICE_FUNC const T *data() const { return m_data.array; }
+    EIGEN_DEVICE_FUNC T *data() { return m_data.array; }
+};
+
+// dynamic-size matrix with fixed-size storage and fixed height
+template<typename T, int Size, int _Rows, int _Options> class DenseStorage<T, Size, _Rows, Dynamic, _Options>
+{
+    internal::plain_array<T,Size,_Options> m_data;
+    Index m_cols;
+  public:
+    EIGEN_DEVICE_FUNC DenseStorage() : m_cols(0) {}
+    EIGEN_DEVICE_FUNC explicit DenseStorage(internal::constructor_without_unaligned_array_assert)
+      : m_data(internal::constructor_without_unaligned_array_assert()), m_cols(0) {}
+    EIGEN_DEVICE_FUNC DenseStorage(const DenseStorage& other) : m_data(other.m_data), m_cols(other.m_cols) {}
+    EIGEN_DEVICE_FUNC DenseStorage& operator=(const DenseStorage& other)
+    {
+      if (this != &other)
+      {
+        m_data = other.m_data;
+        m_cols = other.m_cols;
+      }
+      return *this;
+    }
+    EIGEN_DEVICE_FUNC DenseStorage(Index, Index, Index cols) : m_cols(cols) {}
+    EIGEN_DEVICE_FUNC void swap(DenseStorage& other) { std::swap(m_data,other.m_data); std::swap(m_cols,other.m_cols); }
+    EIGEN_DEVICE_FUNC Index rows(void) const {return _Rows;}
+    EIGEN_DEVICE_FUNC Index cols(void) const {return m_cols;}
+    void conservativeResize(Index, Index, Index cols) { m_cols = cols; }
+    void resize(Index, Index, Index cols) { m_cols = cols; }
+    EIGEN_DEVICE_FUNC const T *data() const { return m_data.array; }
+    EIGEN_DEVICE_FUNC T *data() { return m_data.array; }
+};
+
+// purely dynamic matrix.
+template<typename T, int _Options> class DenseStorage<T, Dynamic, Dynamic, Dynamic, _Options>
+{
+    T *m_data;
+    Index m_rows;
+    Index m_cols;
+  public:
+    EIGEN_DEVICE_FUNC DenseStorage() : m_data(0), m_rows(0), m_cols(0) {}
+    EIGEN_DEVICE_FUNC explicit DenseStorage(internal::constructor_without_unaligned_array_assert)
+       : m_data(0), m_rows(0), m_cols(0) {}
+    EIGEN_DEVICE_FUNC DenseStorage(Index size, Index rows, Index cols)
+      : m_data(internal::conditional_aligned_new_auto<T,(_Options&DontAlign)==0>(size)), m_rows(rows), m_cols(cols)
+    {
+      EIGEN_INTERNAL_DENSE_STORAGE_CTOR_PLUGIN({})
+      eigen_internal_assert(size==rows*cols && rows>=0 && cols >=0);
+    }
+    EIGEN_DEVICE_FUNC DenseStorage(const DenseStorage& other)
+      : m_data(internal::conditional_aligned_new_auto<T,(_Options&DontAlign)==0>(other.m_rows*other.m_cols))
+      , m_rows(other.m_rows)
+      , m_cols(other.m_cols)
+    {
+      EIGEN_INTERNAL_DENSE_STORAGE_CTOR_PLUGIN(Index size = m_rows*m_cols)
+      internal::smart_copy(other.m_data, other.m_data+other.m_rows*other.m_cols, m_data);
+    }
+    EIGEN_DEVICE_FUNC DenseStorage& operator=(const DenseStorage& other)
+    {
+      if (this != &other)
+      {
+        DenseStorage tmp(other);
+        this->swap(tmp);
+      }
+      return *this;
+    }
+#if EIGEN_HAS_RVALUE_REFERENCES
+    EIGEN_DEVICE_FUNC
+    DenseStorage(DenseStorage&& other) EIGEN_NOEXCEPT
+      : m_data(std::move(other.m_data))
+      , m_rows(std::move(other.m_rows))
+      , m_cols(std::move(other.m_cols))
+    {
+      other.m_data = nullptr;
+      other.m_rows = 0;
+      other.m_cols = 0;
+    }
+    EIGEN_DEVICE_FUNC
+    DenseStorage& operator=(DenseStorage&& other) EIGEN_NOEXCEPT
+    {
+      using std::swap;
+      swap(m_data, other.m_data);
+      swap(m_rows, other.m_rows);
+      swap(m_cols, other.m_cols);
+      return *this;
+    }
+#endif
+    EIGEN_DEVICE_FUNC ~DenseStorage() { internal::conditional_aligned_delete_auto<T,(_Options&DontAlign)==0>(m_data, m_rows*m_cols); }
+    EIGEN_DEVICE_FUNC void swap(DenseStorage& other)
+    { std::swap(m_data,other.m_data); std::swap(m_rows,other.m_rows); std::swap(m_cols,other.m_cols); }
+    EIGEN_DEVICE_FUNC Index rows(void) const {return m_rows;}
+    EIGEN_DEVICE_FUNC Index cols(void) const {return m_cols;}
+    void conservativeResize(Index size, Index rows, Index cols)
+    {
+      m_data = internal::conditional_aligned_realloc_new_auto<T,(_Options&DontAlign)==0>(m_data, size, m_rows*m_cols);
+      m_rows = rows;
+      m_cols = cols;
+    }
+    EIGEN_DEVICE_FUNC void resize(Index size, Index rows, Index cols)
+    {
+      if(size != m_rows*m_cols)
+      {
+        internal::conditional_aligned_delete_auto<T,(_Options&DontAlign)==0>(m_data, m_rows*m_cols);
+        if (size>0) // >0 and not simply !=0 to let the compiler knows that size cannot be negative
+          m_data = internal::conditional_aligned_new_auto<T,(_Options&DontAlign)==0>(size);
+        else
+          m_data = 0;
+        EIGEN_INTERNAL_DENSE_STORAGE_CTOR_PLUGIN({})
+      }
+      m_rows = rows;
+      m_cols = cols;
+    }
+    EIGEN_DEVICE_FUNC const T *data() const { return m_data; }
+    EIGEN_DEVICE_FUNC T *data() { return m_data; }
+};
+
+// matrix with dynamic width and fixed height (so that matrix has dynamic size).
+template<typename T, int _Rows, int _Options> class DenseStorage<T, Dynamic, _Rows, Dynamic, _Options>
+{
+    T *m_data;
+    Index m_cols;
+  public:
+    EIGEN_DEVICE_FUNC DenseStorage() : m_data(0), m_cols(0) {}
+    explicit DenseStorage(internal::constructor_without_unaligned_array_assert) : m_data(0), m_cols(0) {}
+    EIGEN_DEVICE_FUNC DenseStorage(Index size, Index rows, Index cols) : m_data(internal::conditional_aligned_new_auto<T,(_Options&DontAlign)==0>(size)), m_cols(cols)
+    {
+      EIGEN_INTERNAL_DENSE_STORAGE_CTOR_PLUGIN({})
+      eigen_internal_assert(size==rows*cols && rows==_Rows && cols >=0);
+      EIGEN_UNUSED_VARIABLE(rows);
+    }
+    EIGEN_DEVICE_FUNC DenseStorage(const DenseStorage& other)
+      : m_data(internal::conditional_aligned_new_auto<T,(_Options&DontAlign)==0>(_Rows*other.m_cols))
+      , m_cols(other.m_cols)
+    {
+      EIGEN_INTERNAL_DENSE_STORAGE_CTOR_PLUGIN(Index size = m_cols*_Rows)
+      internal::smart_copy(other.m_data, other.m_data+_Rows*m_cols, m_data);
+    }
+    EIGEN_DEVICE_FUNC DenseStorage& operator=(const DenseStorage& other)
+    {
+      if (this != &other)
+      {
+        DenseStorage tmp(other);
+        this->swap(tmp);
+      }
+      return *this;
+    }    
+#if EIGEN_HAS_RVALUE_REFERENCES
+    EIGEN_DEVICE_FUNC
+    DenseStorage(DenseStorage&& other) EIGEN_NOEXCEPT
+      : m_data(std::move(other.m_data))
+      , m_cols(std::move(other.m_cols))
+    {
+      other.m_data = nullptr;
+      other.m_cols = 0;
+    }
+    EIGEN_DEVICE_FUNC
+    DenseStorage& operator=(DenseStorage&& other) EIGEN_NOEXCEPT
+    {
+      using std::swap;
+      swap(m_data, other.m_data);
+      swap(m_cols, other.m_cols);
+      return *this;
+    }
+#endif
+    EIGEN_DEVICE_FUNC ~DenseStorage() { internal::conditional_aligned_delete_auto<T,(_Options&DontAlign)==0>(m_data, _Rows*m_cols); }
+    EIGEN_DEVICE_FUNC void swap(DenseStorage& other) { std::swap(m_data,other.m_data); std::swap(m_cols,other.m_cols); }
+    EIGEN_DEVICE_FUNC static Index rows(void) {return _Rows;}
+    EIGEN_DEVICE_FUNC Index cols(void) const {return m_cols;}
+    EIGEN_DEVICE_FUNC void conservativeResize(Index size, Index, Index cols)
+    {
+      m_data = internal::conditional_aligned_realloc_new_auto<T,(_Options&DontAlign)==0>(m_data, size, _Rows*m_cols);
+      m_cols = cols;
+    }
+    EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE void resize(Index size, Index, Index cols)
+    {
+      if(size != _Rows*m_cols)
+      {
+        internal::conditional_aligned_delete_auto<T,(_Options&DontAlign)==0>(m_data, _Rows*m_cols);
+        if (size>0) // >0 and not simply !=0 to let the compiler knows that size cannot be negative
+          m_data = internal::conditional_aligned_new_auto<T,(_Options&DontAlign)==0>(size);
+        else
+          m_data = 0;
+        EIGEN_INTERNAL_DENSE_STORAGE_CTOR_PLUGIN({})
+      }
+      m_cols = cols;
+    }
+    EIGEN_DEVICE_FUNC const T *data() const { return m_data; }
+    EIGEN_DEVICE_FUNC T *data() { return m_data; }
+};
+
+// matrix with dynamic height and fixed width (so that matrix has dynamic size).
+template<typename T, int _Cols, int _Options> class DenseStorage<T, Dynamic, Dynamic, _Cols, _Options>
+{
+    T *m_data;
+    Index m_rows;
+  public:
+    EIGEN_DEVICE_FUNC DenseStorage() : m_data(0), m_rows(0) {}
+    explicit DenseStorage(internal::constructor_without_unaligned_array_assert) : m_data(0), m_rows(0) {}
+    EIGEN_DEVICE_FUNC DenseStorage(Index size, Index rows, Index cols) : m_data(internal::conditional_aligned_new_auto<T,(_Options&DontAlign)==0>(size)), m_rows(rows)
+    {
+      EIGEN_INTERNAL_DENSE_STORAGE_CTOR_PLUGIN({})
+      eigen_internal_assert(size==rows*cols && rows>=0 && cols == _Cols);
+      EIGEN_UNUSED_VARIABLE(cols);
+    }
+    EIGEN_DEVICE_FUNC DenseStorage(const DenseStorage& other)
+      : m_data(internal::conditional_aligned_new_auto<T,(_Options&DontAlign)==0>(other.m_rows*_Cols))
+      , m_rows(other.m_rows)
+    {
+      EIGEN_INTERNAL_DENSE_STORAGE_CTOR_PLUGIN(Index size = m_rows*_Cols)
+      internal::smart_copy(other.m_data, other.m_data+other.m_rows*_Cols, m_data);
+    }
+    EIGEN_DEVICE_FUNC DenseStorage& operator=(const DenseStorage& other)
+    {
+      if (this != &other)
+      {
+        DenseStorage tmp(other);
+        this->swap(tmp);
+      }
+      return *this;
+    }    
+#if EIGEN_HAS_RVALUE_REFERENCES
+    EIGEN_DEVICE_FUNC
+    DenseStorage(DenseStorage&& other) EIGEN_NOEXCEPT
+      : m_data(std::move(other.m_data))
+      , m_rows(std::move(other.m_rows))
+    {
+      other.m_data = nullptr;
+      other.m_rows = 0;
+    }
+    EIGEN_DEVICE_FUNC
+    DenseStorage& operator=(DenseStorage&& other) EIGEN_NOEXCEPT
+    {
+      using std::swap;
+      swap(m_data, other.m_data);
+      swap(m_rows, other.m_rows);
+      return *this;
+    }
+#endif
+    EIGEN_DEVICE_FUNC ~DenseStorage() { internal::conditional_aligned_delete_auto<T,(_Options&DontAlign)==0>(m_data, _Cols*m_rows); }
+    EIGEN_DEVICE_FUNC void swap(DenseStorage& other) { std::swap(m_data,other.m_data); std::swap(m_rows,other.m_rows); }
+    EIGEN_DEVICE_FUNC Index rows(void) const {return m_rows;}
+    EIGEN_DEVICE_FUNC static Index cols(void) {return _Cols;}
+    void conservativeResize(Index size, Index rows, Index)
+    {
+      m_data = internal::conditional_aligned_realloc_new_auto<T,(_Options&DontAlign)==0>(m_data, size, m_rows*_Cols);
+      m_rows = rows;
+    }
+    EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE void resize(Index size, Index rows, Index)
+    {
+      if(size != m_rows*_Cols)
+      {
+        internal::conditional_aligned_delete_auto<T,(_Options&DontAlign)==0>(m_data, _Cols*m_rows);
+        if (size>0) // >0 and not simply !=0 to let the compiler knows that size cannot be negative
+          m_data = internal::conditional_aligned_new_auto<T,(_Options&DontAlign)==0>(size);
+        else
+          m_data = 0;
+        EIGEN_INTERNAL_DENSE_STORAGE_CTOR_PLUGIN({})
+      }
+      m_rows = rows;
+    }
+    EIGEN_DEVICE_FUNC const T *data() const { return m_data; }
+    EIGEN_DEVICE_FUNC T *data() { return m_data; }
+};
+
+} // end namespace Eigen
+
+#endif // EIGEN_MATRIX_H

Eigen/src/Core/Diagonal.h

@@ -0,0 +1,260 @@
+// This file is part of Eigen, a lightweight C++ template library
+// for linear algebra.
+//
+// Copyright (C) 2007-2009 Benoit Jacob <jacob.benoit.1@gmail.com>
+// Copyright (C) 2009-2010 Gael Guennebaud <gael.guennebaud@inria.fr>
+//
+// This Source Code Form is subject to the terms of the Mozilla
+// Public License v. 2.0. If a copy of the MPL was not distributed
+// with this file, You can obtain one at http://mozilla.org/MPL/2.0/.
+
+#ifndef EIGEN_DIAGONAL_H
+#define EIGEN_DIAGONAL_H
+
+namespace Eigen { 
+
+/** \class Diagonal
+  * \ingroup Core_Module
+  *
+  * \brief Expression of a diagonal/subdiagonal/superdiagonal in a matrix
+  *
+  * \param MatrixType the type of the object in which we are taking a sub/main/super diagonal
+  * \param DiagIndex the index of the sub/super diagonal. The default is 0 and it means the main diagonal.
+  *              A positive value means a superdiagonal, a negative value means a subdiagonal.
+  *              You can also use DynamicIndex so the index can be set at runtime.
+  *
+  * The matrix is not required to be square.
+  *
+  * This class represents an expression of the main diagonal, or any sub/super diagonal
+  * of a square matrix. It is the return type of MatrixBase::diagonal() and MatrixBase::diagonal(Index) and most of the
+  * time this is the only way it is used.
+  *
+  * \sa MatrixBase::diagonal(), MatrixBase::diagonal(Index)
+  */
+
+namespace internal {
+template<typename MatrixType, int DiagIndex>
+struct traits<Diagonal<MatrixType,DiagIndex> >
+ : traits<MatrixType>
+{
+  typedef typename ref_selector<MatrixType>::type MatrixTypeNested;
+  typedef typename remove_reference<MatrixTypeNested>::type _MatrixTypeNested;
+  typedef typename MatrixType::StorageKind StorageKind;
+  enum {
+    RowsAtCompileTime = (int(DiagIndex) == DynamicIndex || int(MatrixType::SizeAtCompileTime) == Dynamic) ? Dynamic
+                      : (EIGEN_PLAIN_ENUM_MIN(MatrixType::RowsAtCompileTime - EIGEN_PLAIN_ENUM_MAX(-DiagIndex, 0),
+                                              MatrixType::ColsAtCompileTime - EIGEN_PLAIN_ENUM_MAX( DiagIndex, 0))),
+    ColsAtCompileTime = 1,
+    MaxRowsAtCompileTime = int(MatrixType::MaxSizeAtCompileTime) == Dynamic ? Dynamic
+                         : DiagIndex == DynamicIndex ? EIGEN_SIZE_MIN_PREFER_FIXED(MatrixType::MaxRowsAtCompileTime,
+                                                                              MatrixType::MaxColsAtCompileTime)
+                         : (EIGEN_PLAIN_ENUM_MIN(MatrixType::MaxRowsAtCompileTime - EIGEN_PLAIN_ENUM_MAX(-DiagIndex, 0),
+                                                 MatrixType::MaxColsAtCompileTime - EIGEN_PLAIN_ENUM_MAX( DiagIndex, 0))),
+    MaxColsAtCompileTime = 1,
+    MaskLvalueBit = is_lvalue<MatrixType>::value ? LvalueBit : 0,
+    Flags = (unsigned int)_MatrixTypeNested::Flags & (RowMajorBit | MaskLvalueBit | DirectAccessBit) & ~RowMajorBit, // FIXME DirectAccessBit should not be handled by expressions
+    MatrixTypeOuterStride = outer_stride_at_compile_time<MatrixType>::ret,
+    InnerStrideAtCompileTime = MatrixTypeOuterStride == Dynamic ? Dynamic : MatrixTypeOuterStride+1,
+    OuterStrideAtCompileTime = 0
+  };
+};
+}
+
+template<typename MatrixType, int _DiagIndex> class Diagonal
+   : public internal::dense_xpr_base< Diagonal<MatrixType,_DiagIndex> >::type
+{
+  public:
+
+    enum { DiagIndex = _DiagIndex };
+    typedef typename internal::dense_xpr_base<Diagonal>::type Base;
+    EIGEN_DENSE_PUBLIC_INTERFACE(Diagonal)
+
+    EIGEN_DEVICE_FUNC
+    explicit inline Diagonal(MatrixType& matrix, Index a_index = DiagIndex) : m_matrix(matrix), m_index(a_index)
+    {
+      eigen_assert( a_index <= m_matrix.cols() && -a_index <= m_matrix.rows() );
+    }
+
+    EIGEN_INHERIT_ASSIGNMENT_OPERATORS(Diagonal)
+
+    EIGEN_DEVICE_FUNC
+    inline Index rows() const
+    {
+      return m_index.value()<0 ? numext::mini<Index>(m_matrix.cols(),m_matrix.rows()+m_index.value())
+                               : numext::mini<Index>(m_matrix.rows(),m_matrix.cols()-m_index.value());
+    }
+
+    EIGEN_DEVICE_FUNC
+    inline Index cols() const { return 1; }
+
+    EIGEN_DEVICE_FUNC
+    inline Index innerStride() const
+    {
+      return m_matrix.outerStride() + 1;
+    }
+
+    EIGEN_DEVICE_FUNC
+    inline Index outerStride() const
+    {
+      return 0;
+    }
+
+    typedef typename internal::conditional<
+                       internal::is_lvalue<MatrixType>::value,
+                       Scalar,
+                       const Scalar
+                     >::type ScalarWithConstIfNotLvalue;
+
+    EIGEN_DEVICE_FUNC
+    inline ScalarWithConstIfNotLvalue* data() { return &(m_matrix.coeffRef(rowOffset(), colOffset())); }
+    EIGEN_DEVICE_FUNC
+    inline const Scalar* data() const { return &(m_matrix.coeffRef(rowOffset(), colOffset())); }
+
+    EIGEN_DEVICE_FUNC
+    inline Scalar& coeffRef(Index row, Index)
+    {
+      EIGEN_STATIC_ASSERT_LVALUE(MatrixType)
+      return m_matrix.coeffRef(row+rowOffset(), row+colOffset());
+    }
+
+    EIGEN_DEVICE_FUNC
+    inline const Scalar& coeffRef(Index row, Index) const
+    {
+      return m_matrix.coeffRef(row+rowOffset(), row+colOffset());
+    }
+
+    EIGEN_DEVICE_FUNC
+    inline CoeffReturnType coeff(Index row, Index) const
+    {
+      return m_matrix.coeff(row+rowOffset(), row+colOffset());
+    }
+
+    EIGEN_DEVICE_FUNC
+    inline Scalar& coeffRef(Index idx)
+    {
+      EIGEN_STATIC_ASSERT_LVALUE(MatrixType)
+      return m_matrix.coeffRef(idx+rowOffset(), idx+colOffset());
+    }
+
+    EIGEN_DEVICE_FUNC
+    inline const Scalar& coeffRef(Index idx) const
+    {
+      return m_matrix.coeffRef(idx+rowOffset(), idx+colOffset());
+    }
+
+    EIGEN_DEVICE_FUNC
+    inline CoeffReturnType coeff(Index idx) const
+    {
+      return m_matrix.coeff(idx+rowOffset(), idx+colOffset());
+    }
+
+    EIGEN_DEVICE_FUNC
+    inline const typename internal::remove_all<typename MatrixType::Nested>::type& 
+    nestedExpression() const 
+    {
+      return m_matrix;
+    }
+
+    EIGEN_DEVICE_FUNC
+    inline Index index() const
+    {
+      return m_index.value();
+    }
+
+  protected:
+    typename internal::ref_selector<MatrixType>::non_const_type m_matrix;
+    const internal::variable_if_dynamicindex<Index, DiagIndex> m_index;
+
+  private:
+    // some compilers may fail to optimize std::max etc in case of compile-time constants...
+    EIGEN_DEVICE_FUNC
+    EIGEN_STRONG_INLINE Index absDiagIndex() const { return m_index.value()>0 ? m_index.value() : -m_index.value(); }
+    EIGEN_DEVICE_FUNC
+    EIGEN_STRONG_INLINE Index rowOffset() const { return m_index.value()>0 ? 0 : -m_index.value(); }
+    EIGEN_DEVICE_FUNC
+    EIGEN_STRONG_INLINE Index colOffset() const { return m_index.value()>0 ? m_index.value() : 0; }
+    // trigger a compile-time error if someone try to call packet
+    template<int LoadMode> typename MatrixType::PacketReturnType packet(Index) const;
+    template<int LoadMode> typename MatrixType::PacketReturnType packet(Index,Index) const;
+};
+
+/** \returns an expression of the main diagonal of the matrix \c *this
+  *
+  * \c *this is not required to be square.
+  *
+  * Example: \include MatrixBase_diagonal.cpp
+  * Output: \verbinclude MatrixBase_diagonal.out
+  *
+  * \sa class Diagonal */
+template<typename Derived>
+inline typename MatrixBase<Derived>::DiagonalReturnType
+MatrixBase<Derived>::diagonal()
+{
+  return DiagonalReturnType(derived());
+}
+
+/** This is the const version of diagonal(). */
+template<typename Derived>
+inline typename MatrixBase<Derived>::ConstDiagonalReturnType
+MatrixBase<Derived>::diagonal() const
+{
+  return ConstDiagonalReturnType(derived());
+}
+
+/** \returns an expression of the \a DiagIndex-th sub or super diagonal of the matrix \c *this
+  *
+  * \c *this is not required to be square.
+  *
+  * The template parameter \a DiagIndex represent a super diagonal if \a DiagIndex > 0
+  * and a sub diagonal otherwise. \a DiagIndex == 0 is equivalent to the main diagonal.
+  *
+  * Example: \include MatrixBase_diagonal_int.cpp
+  * Output: \verbinclude MatrixBase_diagonal_int.out
+  *
+  * \sa MatrixBase::diagonal(), class Diagonal */
+template<typename Derived>
+inline typename MatrixBase<Derived>::DiagonalDynamicIndexReturnType
+MatrixBase<Derived>::diagonal(Index index)
+{
+  return DiagonalDynamicIndexReturnType(derived(), index);
+}
+
+/** This is the const version of diagonal(Index). */
+template<typename Derived>
+inline typename MatrixBase<Derived>::ConstDiagonalDynamicIndexReturnType
+MatrixBase<Derived>::diagonal(Index index) const
+{
+  return ConstDiagonalDynamicIndexReturnType(derived(), index);
+}
+
+/** \returns an expression of the \a DiagIndex-th sub or super diagonal of the matrix \c *this
+  *
+  * \c *this is not required to be square.
+  *
+  * The template parameter \a DiagIndex represent a super diagonal if \a DiagIndex > 0
+  * and a sub diagonal otherwise. \a DiagIndex == 0 is equivalent to the main diagonal.
+  *
+  * Example: \include MatrixBase_diagonal_template_int.cpp
+  * Output: \verbinclude MatrixBase_diagonal_template_int.out
+  *
+  * \sa MatrixBase::diagonal(), class Diagonal */
+template<typename Derived>
+template<int Index_>
+inline typename MatrixBase<Derived>::template DiagonalIndexReturnType<Index_>::Type
+MatrixBase<Derived>::diagonal()
+{
+  return typename DiagonalIndexReturnType<Index_>::Type(derived());
+}
+
+/** This is the const version of diagonal<int>(). */
+template<typename Derived>
+template<int Index_>
+inline typename MatrixBase<Derived>::template ConstDiagonalIndexReturnType<Index_>::Type
+MatrixBase<Derived>::diagonal() const
+{
+  return typename ConstDiagonalIndexReturnType<Index_>::Type(derived());
+}
+
+} // end namespace Eigen
+
+#endif // EIGEN_DIAGONAL_H

Eigen/src/Core/DiagonalCoeffs.h

@@ -1,124 +0,0 @@
-// This file is part of Eigen, a lightweight C++ template library
-// for linear algebra. Eigen itself is part of the KDE project.
-//
-// Copyright (C) 2006-2008 Benoit Jacob <jacob.benoit.1@gmail.com>
-//
-// 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_DIAGONALCOEFFS_H
-#define EIGEN_DIAGONALCOEFFS_H
-
-/** \class DiagonalCoeffs
-  *
-  * \brief Expression of the main diagonal of a matrix
-  *
-  * \param MatrixType the type of the object in which we are taking the main diagonal
-  *
-  * The matrix is not required to be square.
-  *
-  * This class represents an expression of the main diagonal of a square matrix.
-  * It is the return type of MatrixBase::diagonal() and most of the time this is
-  * the only way it is used.
-  *
-  * \sa MatrixBase::diagonal()
-  */
-template<typename MatrixType>
-struct ei_traits<DiagonalCoeffs<MatrixType> >
-{
-  typedef typename MatrixType::Scalar Scalar;
-  typedef typename ei_nested<MatrixType>::type MatrixTypeNested;
-  typedef typename ei_unref<MatrixTypeNested>::type _MatrixTypeNested;
-  enum {
-    RowsAtCompileTime = int(MatrixType::SizeAtCompileTime) == Dynamic ? Dynamic
-                      : EIGEN_ENUM_MIN(MatrixType::RowsAtCompileTime,
-                                       MatrixType::ColsAtCompileTime),
-    ColsAtCompileTime = 1,
-    MaxRowsAtCompileTime = int(MatrixType::MaxSizeAtCompileTime) == Dynamic ? Dynamic
-                            : EIGEN_ENUM_MIN(MatrixType::MaxRowsAtCompileTime,
-                                             MatrixType::MaxColsAtCompileTime),
-    MaxColsAtCompileTime = 1,
-    Flags = (unsigned int)_MatrixTypeNested::Flags & (HereditaryBits | LinearAccessBit),
-    CoeffReadCost = _MatrixTypeNested::CoeffReadCost
-  };
-};
-
-template<typename MatrixType> class DiagonalCoeffs
-   : public MatrixBase<DiagonalCoeffs<MatrixType> >
-{
-  public:
-
-    EIGEN_GENERIC_PUBLIC_INTERFACE(DiagonalCoeffs)
-
-    inline DiagonalCoeffs(const MatrixType& matrix) : m_matrix(matrix) {}
-
-    EIGEN_INHERIT_ASSIGNMENT_OPERATORS(DiagonalCoeffs)
-
-    inline int rows() const { return std::min(m_matrix.rows(), m_matrix.cols()); }
-    inline int cols() const { return 1; }
-
-    inline Scalar& coeffRef(int row, int)
-    {
-      return m_matrix.const_cast_derived().coeffRef(row, row);
-    }
-
-    inline const Scalar coeff(int row, int) const
-    {
-      return m_matrix.coeff(row, row);
-    }
-
-    inline Scalar& coeffRef(int index)
-    {
-      return m_matrix.const_cast_derived().coeffRef(index, index);
-    }
-
-    inline const Scalar coeff(int index) const
-    {
-      return m_matrix.coeff(index, index);
-    }
-
-  protected:
-
-    const typename MatrixType::Nested m_matrix;
-};
-
-/** \returns an expression of the main diagonal of the matrix \c *this
-  *
-  * \c *this is not required to be square.
-  *
-  * Example: \include MatrixBase_diagonal.cpp
-  * Output: \verbinclude MatrixBase_diagonal.out
-  *
-  * \sa class DiagonalCoeffs */
-template<typename Derived>
-inline DiagonalCoeffs<Derived>
-MatrixBase<Derived>::diagonal()
-{
-  return DiagonalCoeffs<Derived>(derived());
-}
-
-/** This is the const version of diagonal(). */
-template<typename Derived>
-inline const DiagonalCoeffs<Derived>
-MatrixBase<Derived>::diagonal() const
-{
-  return DiagonalCoeffs<Derived>(derived());
-}
-
-#endif // EIGEN_DIAGONALCOEFFS_H

Eigen/src/Core/DiagonalMatrix.h

@@ -1,119 +1,285 @@
 // This file is part of Eigen, a lightweight C++ template library
-// for linear algebra. Eigen itself is part of the KDE project.
+// for linear algebra.
 //
-// Copyright (C) 2006-2008 Benoit Jacob <jacob.benoit.1@gmail.com>
+// Copyright (C) 2009 Gael Guennebaud <gael.guennebaud@inria.fr>
+// Copyright (C) 2007-2009 Benoit Jacob <jacob.benoit.1@gmail.com>
 //
-// 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/>.
+// This Source Code Form is subject to the terms of the Mozilla
+// Public License v. 2.0. If a copy of the MPL was not distributed
+// with this file, You can obtain one at http://mozilla.org/MPL/2.0/.
 
 #ifndef EIGEN_DIAGONALMATRIX_H
 #define EIGEN_DIAGONALMATRIX_H
 
+namespace Eigen { 
+
+#ifndef EIGEN_PARSED_BY_DOXYGEN
+template<typename Derived>
+class DiagonalBase : public EigenBase<Derived>
+{
+  public:
+    typedef typename internal::traits<Derived>::DiagonalVectorType DiagonalVectorType;
+    typedef typename DiagonalVectorType::Scalar Scalar;
+    typedef typename DiagonalVectorType::RealScalar RealScalar;
+    typedef typename internal::traits<Derived>::StorageKind StorageKind;
+    typedef typename internal::traits<Derived>::StorageIndex StorageIndex;
+
+    enum {
+      RowsAtCompileTime = DiagonalVectorType::SizeAtCompileTime,
+      ColsAtCompileTime = DiagonalVectorType::SizeAtCompileTime,
+      MaxRowsAtCompileTime = DiagonalVectorType::MaxSizeAtCompileTime,
+      MaxColsAtCompileTime = DiagonalVectorType::MaxSizeAtCompileTime,
+      IsVectorAtCompileTime = 0,
+      Flags = NoPreferredStorageOrderBit
+    };
+
+    typedef Matrix<Scalar, RowsAtCompileTime, ColsAtCompileTime, 0, MaxRowsAtCompileTime, MaxColsAtCompileTime> DenseMatrixType;
+    typedef DenseMatrixType DenseType;
+    typedef DiagonalMatrix<Scalar,DiagonalVectorType::SizeAtCompileTime,DiagonalVectorType::MaxSizeAtCompileTime> PlainObject;
+
+    EIGEN_DEVICE_FUNC
+    inline const Derived& derived() const { return *static_cast<const Derived*>(this); }
+    EIGEN_DEVICE_FUNC
+    inline Derived& derived() { return *static_cast<Derived*>(this); }
+
+    EIGEN_DEVICE_FUNC
+    DenseMatrixType toDenseMatrix() const { return derived(); }
+    
+    EIGEN_DEVICE_FUNC
+    inline const DiagonalVectorType& diagonal() const { return derived().diagonal(); }
+    EIGEN_DEVICE_FUNC
+    inline DiagonalVectorType& diagonal() { return derived().diagonal(); }
+
+    EIGEN_DEVICE_FUNC
+    inline Index rows() const { return diagonal().size(); }
+    EIGEN_DEVICE_FUNC
+    inline Index cols() const { return diagonal().size(); }
+
+    template<typename MatrixDerived>
+    EIGEN_DEVICE_FUNC
+    const Product<Derived,MatrixDerived,LazyProduct>
+    operator*(const MatrixBase<MatrixDerived> &matrix) const
+    {
+      return Product<Derived, MatrixDerived, LazyProduct>(derived(),matrix.derived());
+    }
+
+    typedef DiagonalWrapper<const CwiseUnaryOp<internal::scalar_inverse_op<Scalar>, const DiagonalVectorType> > InverseReturnType;
+    EIGEN_DEVICE_FUNC
+    inline const InverseReturnType
+    inverse() const
+    {
+      return InverseReturnType(diagonal().cwiseInverse());
+    }
+    
+    EIGEN_DEVICE_FUNC
+    inline const DiagonalWrapper<const EIGEN_EXPR_BINARYOP_SCALAR_RETURN_TYPE(DiagonalVectorType,Scalar,product) >
+    operator*(const Scalar& scalar) const
+    {
+      return DiagonalWrapper<const EIGEN_EXPR_BINARYOP_SCALAR_RETURN_TYPE(DiagonalVectorType,Scalar,product) >(diagonal() * scalar);
+    }
+    EIGEN_DEVICE_FUNC
+    friend inline const DiagonalWrapper<const EIGEN_SCALAR_BINARYOP_EXPR_RETURN_TYPE(Scalar,DiagonalVectorType,product) >
+    operator*(const Scalar& scalar, const DiagonalBase& other)
+    {
+      return DiagonalWrapper<const EIGEN_SCALAR_BINARYOP_EXPR_RETURN_TYPE(Scalar,DiagonalVectorType,product) >(scalar * other.diagonal());
+    }
+};
+
+#endif
+
 /** \class DiagonalMatrix
-  * \nonstableyet 
+  * \ingroup Core_Module
+  *
+  * \brief Represents a diagonal matrix with its storage
+  *
+  * \param _Scalar the type of coefficients
+  * \param SizeAtCompileTime the dimension of the matrix, or Dynamic
+  * \param MaxSizeAtCompileTime the dimension of the matrix, or Dynamic. This parameter is optional and defaults
+  *        to SizeAtCompileTime. Most of the time, you do not need to specify it.
+  *
+  * \sa class DiagonalWrapper
+  */
+
+namespace internal {
+template<typename _Scalar, int SizeAtCompileTime, int MaxSizeAtCompileTime>
+struct traits<DiagonalMatrix<_Scalar,SizeAtCompileTime,MaxSizeAtCompileTime> >
+ : traits<Matrix<_Scalar,SizeAtCompileTime,SizeAtCompileTime,0,MaxSizeAtCompileTime,MaxSizeAtCompileTime> >
+{
+  typedef Matrix<_Scalar,SizeAtCompileTime,1,0,MaxSizeAtCompileTime,1> DiagonalVectorType;
+  typedef DiagonalShape StorageKind;
+  enum {
+    Flags = LvalueBit | NoPreferredStorageOrderBit
+  };
+};
+}
+template<typename _Scalar, int SizeAtCompileTime, int MaxSizeAtCompileTime>
+class DiagonalMatrix
+  : public DiagonalBase<DiagonalMatrix<_Scalar,SizeAtCompileTime,MaxSizeAtCompileTime> >
+{
+  public:
+    #ifndef EIGEN_PARSED_BY_DOXYGEN
+    typedef typename internal::traits<DiagonalMatrix>::DiagonalVectorType DiagonalVectorType;
+    typedef const DiagonalMatrix& Nested;
+    typedef _Scalar Scalar;
+    typedef typename internal::traits<DiagonalMatrix>::StorageKind StorageKind;
+    typedef typename internal::traits<DiagonalMatrix>::StorageIndex StorageIndex;
+    #endif
+
+  protected:
+
+    DiagonalVectorType m_diagonal;
+
+  public:
+
+    /** const version of diagonal(). */
+    EIGEN_DEVICE_FUNC
+    inline const DiagonalVectorType& diagonal() const { return m_diagonal; }
+    /** \returns a reference to the stored vector of diagonal coefficients. */
+    EIGEN_DEVICE_FUNC
+    inline DiagonalVectorType& diagonal() { return m_diagonal; }
+
+    /** Default constructor without initialization */
+    EIGEN_DEVICE_FUNC
+    inline DiagonalMatrix() {}
+
+    /** Constructs a diagonal matrix with given dimension  */
+    EIGEN_DEVICE_FUNC
+    explicit inline DiagonalMatrix(Index dim) : m_diagonal(dim) {}
+
+    /** 2D constructor. */
+    EIGEN_DEVICE_FUNC
+    inline DiagonalMatrix(const Scalar& x, const Scalar& y) : m_diagonal(x,y) {}
+
+    /** 3D constructor. */
+    EIGEN_DEVICE_FUNC
+    inline DiagonalMatrix(const Scalar& x, const Scalar& y, const Scalar& z) : m_diagonal(x,y,z) {}
+
+    /** Copy constructor. */
+    template<typename OtherDerived>
+    EIGEN_DEVICE_FUNC
+    inline DiagonalMatrix(const DiagonalBase<OtherDerived>& other) : m_diagonal(other.diagonal()) {}
+
+    #ifndef EIGEN_PARSED_BY_DOXYGEN
+    /** copy constructor. prevent a default copy constructor from hiding the other templated constructor */
+    inline DiagonalMatrix(const DiagonalMatrix& other) : m_diagonal(other.diagonal()) {}
+    #endif
+
+    /** generic constructor from expression of the diagonal coefficients */
+    template<typename OtherDerived>
+    EIGEN_DEVICE_FUNC
+    explicit inline DiagonalMatrix(const MatrixBase<OtherDerived>& other) : m_diagonal(other)
+    {}
+
+    /** Copy operator. */
+    template<typename OtherDerived>
+    EIGEN_DEVICE_FUNC
+    DiagonalMatrix& operator=(const DiagonalBase<OtherDerived>& other)
+    {
+      m_diagonal = other.diagonal();
+      return *this;
+    }
+
+    #ifndef EIGEN_PARSED_BY_DOXYGEN
+    /** This is a special case of the templated operator=. Its purpose is to
+      * prevent a default operator= from hiding the templated operator=.
+      */
+    EIGEN_DEVICE_FUNC
+    DiagonalMatrix& operator=(const DiagonalMatrix& other)
+    {
+      m_diagonal = other.diagonal();
+      return *this;
+    }
+    #endif
+
+    /** Resizes to given size. */
+    EIGEN_DEVICE_FUNC
+    inline void resize(Index size) { m_diagonal.resize(size); }
+    /** Sets all coefficients to zero. */
+    EIGEN_DEVICE_FUNC
+    inline void setZero() { m_diagonal.setZero(); }
+    /** Resizes and sets all coefficients to zero. */
+    EIGEN_DEVICE_FUNC
+    inline void setZero(Index size) { m_diagonal.setZero(size); }
+    /** Sets this matrix to be the identity matrix of the current size. */
+    EIGEN_DEVICE_FUNC
+    inline void setIdentity() { m_diagonal.setOnes(); }
+    /** Sets this matrix to be the identity matrix of the given size. */
+    EIGEN_DEVICE_FUNC
+    inline void setIdentity(Index size) { m_diagonal.setOnes(size); }
+};
+
+/** \class DiagonalWrapper
+  * \ingroup Core_Module
   *
   * \brief Expression of a diagonal matrix
   *
-  * \param CoeffsVectorType the type of the vector of diagonal coefficients
+  * \param _DiagonalVectorType the type of the vector of diagonal coefficients
   *
-  * This class is an expression of a diagonal matrix with given vector of diagonal
-  * coefficients. It is the return
-  * type of MatrixBase::diagonal(const OtherDerived&) and most of the time this is
-  * the only way it is used.
+  * This class is an expression of a diagonal matrix, but not storing its own vector of diagonal coefficients,
+  * instead wrapping an existing vector expression. It is the return type of MatrixBase::asDiagonal()
+  * and most of the time this is the only way that it is used.
   *
-  * \sa MatrixBase::diagonal(const OtherDerived&)
+  * \sa class DiagonalMatrix, class DiagonalBase, MatrixBase::asDiagonal()
   */
-template<typename CoeffsVectorType>
-struct ei_traits<DiagonalMatrix<CoeffsVectorType> >
+
+namespace internal {
+template<typename _DiagonalVectorType>
+struct traits<DiagonalWrapper<_DiagonalVectorType> >
 {
-  typedef typename CoeffsVectorType::Scalar Scalar;
-  typedef typename ei_nested<CoeffsVectorType>::type CoeffsVectorTypeNested;
-  typedef typename ei_unref<CoeffsVectorTypeNested>::type _CoeffsVectorTypeNested;
+  typedef _DiagonalVectorType DiagonalVectorType;
+  typedef typename DiagonalVectorType::Scalar Scalar;
+  typedef typename DiagonalVectorType::StorageIndex StorageIndex;
+  typedef DiagonalShape StorageKind;
+  typedef typename traits<DiagonalVectorType>::XprKind XprKind;
   enum {
-    RowsAtCompileTime = CoeffsVectorType::SizeAtCompileTime,
-    ColsAtCompileTime = CoeffsVectorType::SizeAtCompileTime,
-    MaxRowsAtCompileTime = CoeffsVectorType::MaxSizeAtCompileTime,
-    MaxColsAtCompileTime = CoeffsVectorType::MaxSizeAtCompileTime,
-    Flags = (_CoeffsVectorTypeNested::Flags & HereditaryBits) | Diagonal,
-    CoeffReadCost = _CoeffsVectorTypeNested::CoeffReadCost
+    RowsAtCompileTime = DiagonalVectorType::SizeAtCompileTime,
+    ColsAtCompileTime = DiagonalVectorType::SizeAtCompileTime,
+    MaxRowsAtCompileTime = DiagonalVectorType::MaxSizeAtCompileTime,
+    MaxColsAtCompileTime = DiagonalVectorType::MaxSizeAtCompileTime,
+    Flags =  (traits<DiagonalVectorType>::Flags & LvalueBit) | NoPreferredStorageOrderBit
   };
 };
+}
 
-template<typename CoeffsVectorType>
-class DiagonalMatrix : ei_no_assignment_operator,
-   public MatrixBase<DiagonalMatrix<CoeffsVectorType> >
+template<typename _DiagonalVectorType>
+class DiagonalWrapper
+  : public DiagonalBase<DiagonalWrapper<_DiagonalVectorType> >, internal::no_assignment_operator
 {
   public:
+    #ifndef EIGEN_PARSED_BY_DOXYGEN
+    typedef _DiagonalVectorType DiagonalVectorType;
+    typedef DiagonalWrapper Nested;
+    #endif
 
-    EIGEN_GENERIC_PUBLIC_INTERFACE(DiagonalMatrix)
-    typedef CoeffsVectorType _CoeffsVectorType;
+    /** Constructor from expression of diagonal coefficients to wrap. */
+    EIGEN_DEVICE_FUNC
+    explicit inline DiagonalWrapper(DiagonalVectorType& a_diagonal) : m_diagonal(a_diagonal) {}
 
-    // needed to evaluate a DiagonalMatrix<Xpr> to a DiagonalMatrix<NestByValue<Vector> >
-    template<typename OtherCoeffsVectorType>
-    inline DiagonalMatrix(const DiagonalMatrix<OtherCoeffsVectorType>& other) : m_coeffs(other.diagonal())
-    {
-      EIGEN_STATIC_ASSERT_VECTOR_ONLY(CoeffsVectorType);
-      EIGEN_STATIC_ASSERT_VECTOR_ONLY(OtherCoeffsVectorType);
-      ei_assert(m_coeffs.size() > 0);
-    }
-
-    inline DiagonalMatrix(const CoeffsVectorType& coeffs) : m_coeffs(coeffs)
-    {
-      EIGEN_STATIC_ASSERT_VECTOR_ONLY(CoeffsVectorType);
-      ei_assert(coeffs.size() > 0);
-    }
-
-    inline int rows() const { return m_coeffs.size(); }
-    inline int cols() const { return m_coeffs.size(); }
-
-    inline const Scalar coeff(int row, int col) const
-    {
-      return row == col ? m_coeffs.coeff(row) : static_cast<Scalar>(0);
-    }
-
-    inline const CoeffsVectorType& diagonal() const { return m_coeffs; }
+    /** \returns a const reference to the wrapped expression of diagonal coefficients. */
+    EIGEN_DEVICE_FUNC
+    const DiagonalVectorType& diagonal() const { return m_diagonal; }
 
   protected:
-    const typename CoeffsVectorType::Nested m_coeffs;
+    typename DiagonalVectorType::Nested m_diagonal;
 };
 
-/** \nonstableyet 
-  * \returns an expression of a diagonal matrix with *this as vector of diagonal coefficients
+/** \returns a pseudo-expression of a diagonal matrix with *this as vector of diagonal coefficients
   *
   * \only_for_vectors
   *
-  * \addexample AsDiagonalExample \label How to build a diagonal matrix from a vector
-  *
   * Example: \include MatrixBase_asDiagonal.cpp
   * Output: \verbinclude MatrixBase_asDiagonal.out
   *
-  * \sa class DiagonalMatrix, isDiagonal()
+  * \sa class DiagonalWrapper, class DiagonalMatrix, diagonal(), isDiagonal()
   **/
 template<typename Derived>
-inline const DiagonalMatrix<Derived>
+inline const DiagonalWrapper<const Derived>
 MatrixBase<Derived>::asDiagonal() const
 {
-  return derived();
+  return DiagonalWrapper<const Derived>(derived());
 }
 
-/** \nonstableyet 
-  * \returns true if *this is approximately equal to a diagonal matrix,
+/** \returns true if *this is approximately equal to a diagonal matrix,
   *          within the precision given by \a prec.
   *
   * Example: \include MatrixBase_isDiagonal.cpp
@@ -122,23 +288,56 @@ MatrixBase<Derived>::asDiagonal() const
   * \sa asDiagonal()
   */
 template<typename Derived>
-bool MatrixBase<Derived>::isDiagonal
-(RealScalar prec) const
+bool MatrixBase<Derived>::isDiagonal(const RealScalar& prec) const
 {
   if(cols() != rows()) return false;
   RealScalar maxAbsOnDiagonal = static_cast<RealScalar>(-1);
-  for(int j = 0; j < cols(); ++j)
+  for(Index j = 0; j < cols(); ++j)
   {
-    RealScalar absOnDiagonal = ei_abs(coeff(j,j));
+    RealScalar absOnDiagonal = numext::abs(coeff(j,j));
     if(absOnDiagonal > maxAbsOnDiagonal) maxAbsOnDiagonal = absOnDiagonal;
   }
-  for(int j = 0; j < cols(); ++j)
-    for(int i = 0; i < j; ++i)
+  for(Index j = 0; j < cols(); ++j)
+    for(Index i = 0; i < j; ++i)
     {
-      if(!ei_isMuchSmallerThan(coeff(i, j), maxAbsOnDiagonal, prec)) return false;
-      if(!ei_isMuchSmallerThan(coeff(j, i), maxAbsOnDiagonal, prec)) return false;
+      if(!internal::isMuchSmallerThan(coeff(i, j), maxAbsOnDiagonal, prec)) return false;
+      if(!internal::isMuchSmallerThan(coeff(j, i), maxAbsOnDiagonal, prec)) return false;
     }
   return true;
 }
 
+namespace internal {
+
+template<> struct storage_kind_to_shape<DiagonalShape> { typedef DiagonalShape Shape; };
+
+struct Diagonal2Dense {};
+
+template<> struct AssignmentKind<DenseShape,DiagonalShape> { typedef Diagonal2Dense Kind; };
+
+// Diagonal matrix to Dense assignment
+template< typename DstXprType, typename SrcXprType, typename Functor>
+struct Assignment<DstXprType, SrcXprType, Functor, Diagonal2Dense>
+{
+  static void run(DstXprType &dst, const SrcXprType &src, const internal::assign_op<typename DstXprType::Scalar,typename SrcXprType::Scalar> &/*func*/)
+  {
+    Index dstRows = src.rows();
+    Index dstCols = src.cols();
+    if((dst.rows()!=dstRows) || (dst.cols()!=dstCols))
+      dst.resize(dstRows, dstCols);
+    
+    dst.setZero();
+    dst.diagonal() = src.diagonal();
+  }
+  
+  static void run(DstXprType &dst, const SrcXprType &src, const internal::add_assign_op<typename DstXprType::Scalar,typename SrcXprType::Scalar> &/*func*/)
+  { dst.diagonal() += src.diagonal(); }
+  
+  static void run(DstXprType &dst, const SrcXprType &src, const internal::sub_assign_op<typename DstXprType::Scalar,typename SrcXprType::Scalar> &/*func*/)
+  { dst.diagonal() -= src.diagonal(); }
+};
+
+} // namespace internal
+
+} // end namespace Eigen
+
 #endif // EIGEN_DIAGONALMATRIX_H

Eigen/src/Core/DiagonalProduct.h

@@ -1,130 +1,28 @@
 // This file is part of Eigen, a lightweight C++ template library
-// for linear algebra. Eigen itself is part of the KDE project.
+// for linear algebra.
 //
-// Copyright (C) 2008 Gael Guennebaud <g.gael@free.fr>
-// Copyright (C) 2006-2008 Benoit Jacob <jacob.benoit.1@gmail.com>
+// Copyright (C) 2008 Gael Guennebaud <gael.guennebaud@inria.fr>
+// Copyright (C) 2007-2009 Benoit Jacob <jacob.benoit.1@gmail.com>
 //
-// 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/>.
+// This Source Code Form is subject to the terms of the Mozilla
+// Public License v. 2.0. If a copy of the MPL was not distributed
+// with this file, You can obtain one at http://mozilla.org/MPL/2.0/.
 
 #ifndef EIGEN_DIAGONALPRODUCT_H
 #define EIGEN_DIAGONALPRODUCT_H
 
-/** \internal Specialization of ei_nested for DiagonalMatrix.
- *  Unlike ei_nested, if the argument is a DiagonalMatrix and if it must be evaluated,
- *  then it evaluated to a DiagonalMatrix having its own argument evaluated.
- */
-template<typename T, int N> struct ei_nested_diagonal : ei_nested<T,N> {};
-template<typename T, int N> struct ei_nested_diagonal<DiagonalMatrix<T>,N >
- : ei_nested<DiagonalMatrix<T>, N, DiagonalMatrix<NestByValue<typename ei_plain_matrix_type<T>::type> > >
-{};
+namespace Eigen { 
 
-// specialization of ProductReturnType
-template<typename Lhs, typename Rhs>
-struct ProductReturnType<Lhs,Rhs,DiagonalProduct>
+/** \returns the diagonal matrix product of \c *this by the diagonal matrix \a diagonal.
+  */
+template<typename Derived>
+template<typename DiagonalDerived>
+inline const Product<Derived, DiagonalDerived, LazyProduct>
+MatrixBase<Derived>::operator*(const DiagonalBase<DiagonalDerived> &a_diagonal) const
 {
-  typedef typename ei_nested_diagonal<Lhs,Rhs::ColsAtCompileTime>::type LhsNested;
-  typedef typename ei_nested_diagonal<Rhs,Lhs::RowsAtCompileTime>::type RhsNested;
+  return Product<Derived, DiagonalDerived, LazyProduct>(derived(),a_diagonal.derived());
+}
 
-  typedef Product<LhsNested, RhsNested, DiagonalProduct> Type;
-};
-
-template<typename LhsNested, typename RhsNested>
-struct ei_traits<Product<LhsNested, RhsNested, DiagonalProduct> >
-{
-  // clean the nested types:
-  typedef typename ei_cleantype<LhsNested>::type _LhsNested;
-  typedef typename ei_cleantype<RhsNested>::type _RhsNested;
-  typedef typename _LhsNested::Scalar Scalar;
-
-  enum {
-    LhsFlags = _LhsNested::Flags,
-    RhsFlags = _RhsNested::Flags,
-    RowsAtCompileTime = _LhsNested::RowsAtCompileTime,
-    ColsAtCompileTime = _RhsNested::ColsAtCompileTime,
-    MaxRowsAtCompileTime = _LhsNested::MaxRowsAtCompileTime,
-    MaxColsAtCompileTime = _RhsNested::MaxColsAtCompileTime,
-
-    LhsIsDiagonal = (_LhsNested::Flags&Diagonal)==Diagonal,
-    RhsIsDiagonal = (_RhsNested::Flags&Diagonal)==Diagonal,
-
-    CanVectorizeRhs =  (!RhsIsDiagonal) && (RhsFlags & RowMajorBit) && (RhsFlags & PacketAccessBit)
-                     && (ColsAtCompileTime % ei_packet_traits<Scalar>::size == 0),
-
-    CanVectorizeLhs =  (!LhsIsDiagonal) && (!(LhsFlags & RowMajorBit)) && (LhsFlags & PacketAccessBit)
-                     && (RowsAtCompileTime % ei_packet_traits<Scalar>::size == 0),
-
-    RemovedBits = ~((RhsFlags & RowMajorBit) && (!CanVectorizeLhs) ? 0 : RowMajorBit),
-
-    Flags = ((unsigned int)(LhsFlags | RhsFlags) & HereditaryBits & RemovedBits)
-          | (((CanVectorizeLhs&&RhsIsDiagonal) || (CanVectorizeRhs&&LhsIsDiagonal)) ? PacketAccessBit : 0),
-
-    CoeffReadCost = NumTraits<Scalar>::MulCost + _LhsNested::CoeffReadCost + _RhsNested::CoeffReadCost
-  };
-};
-
-template<typename LhsNested, typename RhsNested> class Product<LhsNested, RhsNested, DiagonalProduct> : ei_no_assignment_operator,
-  public MatrixBase<Product<LhsNested, RhsNested, DiagonalProduct> >
-{
-    typedef typename ei_traits<Product>::_LhsNested _LhsNested;
-    typedef typename ei_traits<Product>::_RhsNested _RhsNested;
-
-    enum {
-      RhsIsDiagonal = (_RhsNested::Flags&Diagonal)==Diagonal
-    };
-
-  public:
-
-    EIGEN_GENERIC_PUBLIC_INTERFACE(Product)
-
-    template<typename Lhs, typename Rhs>
-    inline Product(const Lhs& lhs, const Rhs& rhs)
-      : m_lhs(lhs), m_rhs(rhs)
-    {
-      ei_assert(lhs.cols() == rhs.rows());
-    }
-
-    inline int rows() const { return m_lhs.rows(); }
-    inline int cols() const { return m_rhs.cols(); }
-
-    const Scalar coeff(int row, int col) const
-    {
-      const int unique = RhsIsDiagonal ? col : row;
-      return m_lhs.coeff(row, unique) * m_rhs.coeff(unique, col);
-    }
-
-    template<int LoadMode>
-    const PacketScalar packet(int row, int col) const
-    {
-      if (RhsIsDiagonal)
-      {
-        return ei_pmul(m_lhs.template packet<LoadMode>(row, col), ei_pset1(m_rhs.coeff(col, col)));
-      }
-      else
-      {
-        return ei_pmul(ei_pset1(m_lhs.coeff(row, row)), m_rhs.template packet<LoadMode>(row, col));
-      }
-    }
-
-  protected:
-    const LhsNested m_lhs;
-    const RhsNested m_rhs;
-};
+} // end namespace Eigen
 
 #endif // EIGEN_DIAGONALPRODUCT_H

Eigen/src/Core/Dot.h

@@ -1,322 +1,276 @@
 // This file is part of Eigen, a lightweight C++ template library
-// for linear algebra. Eigen itself is part of the KDE project.
+// for linear algebra.
 //
-// Copyright (C) 2006-2008 Benoit Jacob <jacob.benoit.1@gmail.com>
+// Copyright (C) 2006-2008, 2010 Benoit Jacob <jacob.benoit.1@gmail.com>
 //
-// 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/>.
+// This Source Code Form is subject to the terms of the Mozilla
+// Public License v. 2.0. If a copy of the MPL was not distributed
+// with this file, You can obtain one at http://mozilla.org/MPL/2.0/.
 
 #ifndef EIGEN_DOT_H
 #define EIGEN_DOT_H
 
-/***************************************************************************
-* Part 1 : the logic deciding a strategy for vectorization and unrolling
-***************************************************************************/
+namespace Eigen { 
 
-template<typename Derived1, typename Derived2>
-struct ei_dot_traits
-{
-public:
-  enum {
-    Vectorization = (int(Derived1::Flags)&int(Derived2::Flags)&ActualPacketAccessBit)
-                 && (int(Derived1::Flags)&int(Derived2::Flags)&LinearAccessBit)
-                  ? LinearVectorization
-                  : NoVectorization
-  };
+namespace internal {
 
-private:
-  typedef typename Derived1::Scalar Scalar;
-  enum {
-    PacketSize = ei_packet_traits<Scalar>::size,
-    Cost = Derived1::SizeAtCompileTime * (Derived1::CoeffReadCost + Derived2::CoeffReadCost + NumTraits<Scalar>::MulCost)
-           + (Derived1::SizeAtCompileTime-1) * NumTraits<Scalar>::AddCost,
-    UnrollingLimit = EIGEN_UNROLLING_LIMIT * (int(Vectorization) == int(NoVectorization) ? 1 : int(PacketSize))
-  };
-
-public:
-  enum {
-    Unrolling = Cost <= UnrollingLimit
-              ? CompleteUnrolling
-              : NoUnrolling
-  };
-};
-
-/***************************************************************************
-* Part 2 : unrollers
-***************************************************************************/
-
-/*** no vectorization ***/
-
-template<typename Derived1, typename Derived2, int Start, int Length>
-struct ei_dot_novec_unroller
-{
-  enum {
-    HalfLength = Length/2
-  };
-
-  typedef typename Derived1::Scalar Scalar;
-
-  inline static Scalar run(const Derived1& v1, const Derived2& v2)
-  {
-    return ei_dot_novec_unroller<Derived1, Derived2, Start, HalfLength>::run(v1, v2)
-         + ei_dot_novec_unroller<Derived1, Derived2, Start+HalfLength, Length-HalfLength>::run(v1, v2);
-  }
-};
-
-template<typename Derived1, typename Derived2, int Start>
-struct ei_dot_novec_unroller<Derived1, Derived2, Start, 1>
-{
-  typedef typename Derived1::Scalar Scalar;
-
-  inline static Scalar run(const Derived1& v1, const Derived2& v2)
-  {
-    return v1.coeff(Start) * ei_conj(v2.coeff(Start));
-  }
-};
-
-/*** vectorization ***/
-
-template<typename Derived1, typename Derived2, int Index, int Stop,
-         bool LastPacket = (Stop-Index == ei_packet_traits<typename Derived1::Scalar>::size)>
-struct ei_dot_vec_unroller
-{
-  typedef typename Derived1::Scalar Scalar;
-  typedef typename ei_packet_traits<Scalar>::type PacketScalar;
-
-  enum {
-    row1 = Derived1::RowsAtCompileTime == 1 ? 0 : Index,
-    col1 = Derived1::RowsAtCompileTime == 1 ? Index : 0,
-    row2 = Derived2::RowsAtCompileTime == 1 ? 0 : Index,
-    col2 = Derived2::RowsAtCompileTime == 1 ? Index : 0
-  };
-
-  inline static PacketScalar run(const Derived1& v1, const Derived2& v2)
-  {
-    return ei_pmadd(
-      v1.template packet<Aligned>(row1, col1),
-      v2.template packet<Aligned>(row2, col2),
-      ei_dot_vec_unroller<Derived1, Derived2, Index+ei_packet_traits<Scalar>::size, Stop>::run(v1, v2)
-    );
-  }
-};
-
-template<typename Derived1, typename Derived2, int Index, int Stop>
-struct ei_dot_vec_unroller<Derived1, Derived2, Index, Stop, true>
-{
-  enum {
-    row1 = Derived1::RowsAtCompileTime == 1 ? 0 : Index,
-    col1 = Derived1::RowsAtCompileTime == 1 ? Index : 0,
-    row2 = Derived2::RowsAtCompileTime == 1 ? 0 : Index,
-    col2 = Derived2::RowsAtCompileTime == 1 ? Index : 0,
-    alignment1 = (Derived1::Flags & AlignedBit) ? Aligned : Unaligned,
-    alignment2 = (Derived2::Flags & AlignedBit) ? Aligned : Unaligned
-  };
-
-  typedef typename Derived1::Scalar Scalar;
-  typedef typename ei_packet_traits<Scalar>::type PacketScalar;
-
-  inline static PacketScalar run(const Derived1& v1, const Derived2& v2)
-  {
-    return ei_pmul(v1.template packet<alignment1>(row1, col1), v2.template packet<alignment2>(row2, col2));
-  }
-};
-
-/***************************************************************************
-* Part 3 : implementation of all cases
-***************************************************************************/
-
-template<typename Derived1, typename Derived2,
-         int Vectorization = ei_dot_traits<Derived1, Derived2>::Vectorization,
-         int Unrolling = ei_dot_traits<Derived1, Derived2>::Unrolling
+// helper function for dot(). The problem is that if we put that in the body of dot(), then upon calling dot
+// with mismatched types, the compiler emits errors about failing to instantiate cwiseProduct BEFORE
+// looking at the static assertions. Thus this is a trick to get better compile errors.
+template<typename T, typename U,
+// the NeedToTranspose condition here is taken straight from Assign.h
+         bool NeedToTranspose = T::IsVectorAtCompileTime
+                && U::IsVectorAtCompileTime
+                && ((int(T::RowsAtCompileTime) == 1 && int(U::ColsAtCompileTime) == 1)
+                      |  // FIXME | instead of || to please GCC 4.4.0 stupid warning "suggest parentheses around &&".
+                         // revert to || as soon as not needed anymore.
+                    (int(T::ColsAtCompileTime) == 1 && int(U::RowsAtCompileTime) == 1))
 >
-struct ei_dot_impl;
-
-template<typename Derived1, typename Derived2>
-struct ei_dot_impl<Derived1, Derived2, NoVectorization, NoUnrolling>
+struct dot_nocheck
 {
-  typedef typename Derived1::Scalar Scalar;
-  static Scalar run(const Derived1& v1, const Derived2& v2)
+  typedef scalar_conj_product_op<typename traits<T>::Scalar,typename traits<U>::Scalar> conj_prod;
+  typedef typename conj_prod::result_type ResScalar;
+  EIGEN_DEVICE_FUNC
+  EIGEN_STRONG_INLINE
+  static ResScalar run(const MatrixBase<T>& a, const MatrixBase<U>& b)
   {
-    ei_assert(v1.size()>0 && "you are using a non initialized vector");
-    Scalar res;
-    res = v1.coeff(0) * ei_conj(v2.coeff(0));
-    for(int i = 1; i < v1.size(); ++i)
-      res += v1.coeff(i) * ei_conj(v2.coeff(i));
-    return res;
+    return a.template binaryExpr<conj_prod>(b).sum();
   }
 };
 
-template<typename Derived1, typename Derived2>
-struct ei_dot_impl<Derived1, Derived2, NoVectorization, CompleteUnrolling>
-  : public ei_dot_novec_unroller<Derived1, Derived2, 0, Derived1::SizeAtCompileTime>
-{};
-
-template<typename Derived1, typename Derived2>
-struct ei_dot_impl<Derived1, Derived2, LinearVectorization, NoUnrolling>
+template<typename T, typename U>
+struct dot_nocheck<T, U, true>
 {
-  typedef typename Derived1::Scalar Scalar;
-  typedef typename ei_packet_traits<Scalar>::type PacketScalar;
-
-  static Scalar run(const Derived1& v1, const Derived2& v2)
+  typedef scalar_conj_product_op<typename traits<T>::Scalar,typename traits<U>::Scalar> conj_prod;
+  typedef typename conj_prod::result_type ResScalar;
+  EIGEN_DEVICE_FUNC
+  EIGEN_STRONG_INLINE
+  static ResScalar run(const MatrixBase<T>& a, const MatrixBase<U>& b)
   {
-    const int size = v1.size();
-    const int packetSize = ei_packet_traits<Scalar>::size;
-    const int alignedSize = (size/packetSize)*packetSize;
-    enum {
-      alignment1 = (Derived1::Flags & AlignedBit) ? Aligned : Unaligned,
-      alignment2 = (Derived2::Flags & AlignedBit) ? Aligned : Unaligned
-    };
-    Scalar res;
-
-    // do the vectorizable part of the sum
-    if(size >= packetSize)
-    {
-      PacketScalar packet_res = ei_pmul(
-                                  v1.template packet<alignment1>(0),
-                                  v2.template packet<alignment2>(0)
-                                );
-      for(int index = packetSize; index<alignedSize; index += packetSize)
-      {
-        packet_res = ei_pmadd(
-                       v1.template packet<alignment1>(index),
-                       v2.template packet<alignment2>(index),
-                       packet_res
-                     );
-      }
-      res = ei_predux(packet_res);
-
-      // now we must do the rest without vectorization.
-      if(alignedSize == size) return res;
-    }
-    else // too small to vectorize anything.
-         // since this is dynamic-size hence inefficient anyway for such small sizes, don't try to optimize.
-    {
-      res = Scalar(0);
-    }
-
-    // do the remainder of the vector
-    for(int index = alignedSize; index < size; ++index)
-    {
-      res += v1.coeff(index) * v2.coeff(index);
-    }
-
-    return res;
+    return a.transpose().template binaryExpr<conj_prod>(b).sum();
   }
 };
 
-template<typename Derived1, typename Derived2>
-struct ei_dot_impl<Derived1, Derived2, LinearVectorization, CompleteUnrolling>
-{
-  typedef typename Derived1::Scalar Scalar;
-  typedef typename ei_packet_traits<Scalar>::type PacketScalar;
-  enum {
-    PacketSize = ei_packet_traits<Scalar>::size,
-    Size = Derived1::SizeAtCompileTime,
-    VectorizationSize = (Size / PacketSize) * PacketSize
-  };
-  static Scalar run(const Derived1& v1, const Derived2& v2)
-  {
-    Scalar res = ei_predux(ei_dot_vec_unroller<Derived1, Derived2, 0, VectorizationSize>::run(v1, v2));
-    if (VectorizationSize != Size)
-      res += ei_dot_novec_unroller<Derived1, Derived2, VectorizationSize, Size-VectorizationSize>::run(v1, v2);
-    return res;
-  }
-};
+} // end namespace internal
 
-/***************************************************************************
-* Part 4 : implementation of MatrixBase methods
-***************************************************************************/
-
-/** \returns the dot product of *this with other.
+/** \fn MatrixBase::dot
+  * \returns the dot product of *this with other.
   *
   * \only_for_vectors
   *
   * \note If the scalar type is complex numbers, then this function returns the hermitian
-  * (sesquilinear) dot product, linear in the first variable and conjugate-linear in the
+  * (sesquilinear) dot product, conjugate-linear in the first variable and linear in the
   * second variable.
   *
   * \sa squaredNorm(), norm()
   */
 template<typename Derived>
 template<typename OtherDerived>
-typename ei_traits<Derived>::Scalar
+EIGEN_DEVICE_FUNC
+EIGEN_STRONG_INLINE
+typename ScalarBinaryOpTraits<typename internal::traits<Derived>::Scalar,typename internal::traits<OtherDerived>::Scalar>::ReturnType
 MatrixBase<Derived>::dot(const MatrixBase<OtherDerived>& other) const
 {
   EIGEN_STATIC_ASSERT_VECTOR_ONLY(Derived)
   EIGEN_STATIC_ASSERT_VECTOR_ONLY(OtherDerived)
   EIGEN_STATIC_ASSERT_SAME_VECTOR_SIZE(Derived,OtherDerived)
-  EIGEN_STATIC_ASSERT((ei_is_same_type<Scalar, typename OtherDerived::Scalar>::ret),
-    YOU_MIXED_DIFFERENT_NUMERIC_TYPES__YOU_NEED_TO_USE_THE_CAST_METHOD_OF_MATRIXBASE_TO_CAST_NUMERIC_TYPES_EXPLICITLY)
+#if !(defined(EIGEN_NO_STATIC_ASSERT) && defined(EIGEN_NO_DEBUG))
+  typedef internal::scalar_conj_product_op<Scalar,typename OtherDerived::Scalar> func;
+  EIGEN_CHECK_BINARY_COMPATIBILIY(func,Scalar,typename OtherDerived::Scalar);
+#endif
+  
+  eigen_assert(size() == other.size());
 
-  ei_assert(size() == other.size());
-
-  return ei_dot_impl<Derived, OtherDerived>::run(derived(), other.derived());
+  return internal::dot_nocheck<Derived,OtherDerived>::run(*this, other);
 }
 
-/** \returns the squared \em l2 norm of *this, i.e., for vectors, the dot product of *this with itself.
+//---------- implementation of L2 norm and related functions ----------
+
+/** \returns, for vectors, the squared \em l2 norm of \c *this, and for matrices the Frobenius norm.
+  * In both cases, it consists in the sum of the square of all the matrix entries.
+  * For vectors, this is also equals to the dot product of \c *this with itself.
   *
-  * \sa dot(), norm()
+  * \sa dot(), norm(), lpNorm()
   */
 template<typename Derived>
-inline typename NumTraits<typename ei_traits<Derived>::Scalar>::Real MatrixBase<Derived>::squaredNorm() const
+EIGEN_STRONG_INLINE typename NumTraits<typename internal::traits<Derived>::Scalar>::Real MatrixBase<Derived>::squaredNorm() const
 {
-  return ei_real((*this).cwise().abs2().sum());
+  return numext::real((*this).cwiseAbs2().sum());
 }
 
-/** \returns the \em l2 norm of *this, i.e., for vectors, the square root of the dot product of *this with itself.
+/** \returns, for vectors, the \em l2 norm of \c *this, and for matrices the Frobenius norm.
+  * In both cases, it consists in the square root of the sum of the square of all the matrix entries.
+  * For vectors, this is also equals to the square root of the dot product of \c *this with itself.
   *
-  * \sa dot(), squaredNorm()
+  * \sa lpNorm(), dot(), squaredNorm()
   */
 template<typename Derived>
-inline typename NumTraits<typename ei_traits<Derived>::Scalar>::Real MatrixBase<Derived>::norm() const
+EIGEN_STRONG_INLINE typename NumTraits<typename internal::traits<Derived>::Scalar>::Real MatrixBase<Derived>::norm() const
 {
-  return ei_sqrt(squaredNorm());
+  return numext::sqrt(squaredNorm());
 }
 
-/** \returns an expression of the quotient of *this by its own norm.
+/** \returns an expression of the quotient of \c *this by its own norm.
+  *
+  * \warning If the input vector is too small (i.e., this->norm()==0),
+  *          then this function returns a copy of the input.
   *
   * \only_for_vectors
   *
   * \sa norm(), normalize()
   */
 template<typename Derived>
-inline const typename MatrixBase<Derived>::PlainMatrixType
+EIGEN_STRONG_INLINE const typename MatrixBase<Derived>::PlainObject
 MatrixBase<Derived>::normalized() const
 {
-  typedef typename ei_nested<Derived>::type Nested;
-  typedef typename ei_unref<Nested>::type _Nested;
+  typedef typename internal::nested_eval<Derived,2>::type _Nested;
   _Nested n(derived());
-  return n / n.norm();
+  RealScalar z = n.squaredNorm();
+  // NOTE: after extensive benchmarking, this conditional does not impact performance, at least on recent x86 CPU
+  if(z>RealScalar(0))
+    return n / numext::sqrt(z);
+  else
+    return n;
 }
 
 /** Normalizes the vector, i.e. divides it by its own norm.
   *
   * \only_for_vectors
   *
+  * \warning If the input vector is too small (i.e., this->norm()==0), then \c *this is left unchanged.
+  *
   * \sa norm(), normalized()
   */
 template<typename Derived>
-inline void MatrixBase<Derived>::normalize()
+EIGEN_STRONG_INLINE void MatrixBase<Derived>::normalize()
 {
-  *this /= norm();
+  RealScalar z = squaredNorm();
+  // NOTE: after extensive benchmarking, this conditional does not impact performance, at least on recent x86 CPU
+  if(z>RealScalar(0))
+    derived() /= numext::sqrt(z);
 }
 
+/** \returns an expression of the quotient of \c *this by its own norm while avoiding underflow and overflow.
+  *
+  * \only_for_vectors
+  *
+  * This method is analogue to the normalized() method, but it reduces the risk of
+  * underflow and overflow when computing the norm.
+  *
+  * \warning If the input vector is too small (i.e., this->norm()==0),
+  *          then this function returns a copy of the input.
+  *
+  * \sa stableNorm(), stableNormalize(), normalized()
+  */
+template<typename Derived>
+EIGEN_STRONG_INLINE const typename MatrixBase<Derived>::PlainObject
+MatrixBase<Derived>::stableNormalized() const
+{
+  typedef typename internal::nested_eval<Derived,3>::type _Nested;
+  _Nested n(derived());
+  RealScalar w = n.cwiseAbs().maxCoeff();
+  RealScalar z = (n/w).squaredNorm();
+  if(z>RealScalar(0))
+    return n / (numext::sqrt(z)*w);
+  else
+    return n;
+}
+
+/** Normalizes the vector while avoid underflow and overflow
+  *
+  * \only_for_vectors
+  *
+  * This method is analogue to the normalize() method, but it reduces the risk of
+  * underflow and overflow when computing the norm.
+  *
+  * \warning If the input vector is too small (i.e., this->norm()==0), then \c *this is left unchanged.
+  *
+  * \sa stableNorm(), stableNormalized(), normalize()
+  */
+template<typename Derived>
+EIGEN_STRONG_INLINE void MatrixBase<Derived>::stableNormalize()
+{
+  RealScalar w = cwiseAbs().maxCoeff();
+  RealScalar z = (derived()/w).squaredNorm();
+  if(z>RealScalar(0))
+    derived() /= numext::sqrt(z)*w;
+}
+
+//---------- implementation of other norms ----------
+
+namespace internal {
+
+template<typename Derived, int p>
+struct lpNorm_selector
+{
+  typedef typename NumTraits<typename traits<Derived>::Scalar>::Real RealScalar;
+  EIGEN_DEVICE_FUNC
+  static inline RealScalar run(const MatrixBase<Derived>& m)
+  {
+    EIGEN_USING_STD_MATH(pow)
+    return pow(m.cwiseAbs().array().pow(p).sum(), RealScalar(1)/p);
+  }
+};
+
+template<typename Derived>
+struct lpNorm_selector<Derived, 1>
+{
+  EIGEN_DEVICE_FUNC
+  static inline typename NumTraits<typename traits<Derived>::Scalar>::Real run(const MatrixBase<Derived>& m)
+  {
+    return m.cwiseAbs().sum();
+  }
+};
+
+template<typename Derived>
+struct lpNorm_selector<Derived, 2>
+{
+  EIGEN_DEVICE_FUNC
+  static inline typename NumTraits<typename traits<Derived>::Scalar>::Real run(const MatrixBase<Derived>& m)
+  {
+    return m.norm();
+  }
+};
+
+template<typename Derived>
+struct lpNorm_selector<Derived, Infinity>
+{
+  typedef typename NumTraits<typename traits<Derived>::Scalar>::Real RealScalar;
+  EIGEN_DEVICE_FUNC
+  static inline RealScalar run(const MatrixBase<Derived>& m)
+  {
+    if(Derived::SizeAtCompileTime==0 || (Derived::SizeAtCompileTime==Dynamic && m.size()==0))
+      return RealScalar(0);
+    return m.cwiseAbs().maxCoeff();
+  }
+};
+
+} // end namespace internal
+
+/** \returns the \b coefficient-wise \f$ \ell^p \f$ norm of \c *this, that is, returns the p-th root of the sum of the p-th powers of the absolute values
+  *          of the coefficients of \c *this. If \a p is the special value \a Eigen::Infinity, this function returns the \f$ \ell^\infty \f$
+  *          norm, that is the maximum of the absolute values of the coefficients of \c *this.
+  *
+  * In all cases, if \c *this is empty, then the value 0 is returned.
+  *
+  * \note For matrices, this function does not compute the <a href="https://en.wikipedia.org/wiki/Operator_norm">operator-norm</a>. That is, if \c *this is a matrix, then its coefficients are interpreted as a 1D vector. Nonetheless, you can easily compute the 1-norm and \f$\infty\f$-norm matrix operator norms using \link TutorialReductionsVisitorsBroadcastingReductionsNorm partial reductions \endlink.
+  *
+  * \sa norm()
+  */
+template<typename Derived>
+template<int p>
+#ifndef EIGEN_PARSED_BY_DOXYGEN
+inline typename NumTraits<typename internal::traits<Derived>::Scalar>::Real
+#else
+MatrixBase<Derived>::RealScalar
+#endif
+MatrixBase<Derived>::lpNorm() const
+{
+  return internal::lpNorm_selector<Derived, p>::run(*this);
+}
+
+//---------- implementation of isOrthogonal / isUnitary ----------
+
 /** \returns true if *this is approximately orthogonal to \a other,
   *          within the precision given by \a prec.
   *
@@ -326,11 +280,11 @@ inline void MatrixBase<Derived>::normalize()
 template<typename Derived>
 template<typename OtherDerived>
 bool MatrixBase<Derived>::isOrthogonal
-(const MatrixBase<OtherDerived>& other, RealScalar prec) const
+(const MatrixBase<OtherDerived>& other, const RealScalar& prec) const
 {
-  typename ei_nested<Derived,2>::type nested(derived());
-  typename ei_nested<OtherDerived,2>::type otherNested(other.derived());
-  return ei_abs2(nested.dot(otherNested)) <= prec * prec * nested.squaredNorm() * otherNested.squaredNorm();
+  typename internal::nested_eval<Derived,2>::type nested(derived());
+  typename internal::nested_eval<OtherDerived,2>::type otherNested(other.derived());
+  return numext::abs2(nested.dot(otherNested)) <= prec * prec * nested.squaredNorm() * otherNested.squaredNorm();
 }
 
 /** \returns true if *this is approximately an unitary matrix,
@@ -345,17 +299,20 @@ bool MatrixBase<Derived>::isOrthogonal
   * Output: \verbinclude MatrixBase_isUnitary.out
   */
 template<typename Derived>
-bool MatrixBase<Derived>::isUnitary(RealScalar prec) const
+bool MatrixBase<Derived>::isUnitary(const RealScalar& prec) const
 {
-  typename Derived::Nested nested(derived());
-  for(int i = 0; i < cols(); ++i)
+  typename internal::nested_eval<Derived,1>::type self(derived());
+  for(Index i = 0; i < cols(); ++i)
   {
-    if(!ei_isApprox(nested.col(i).squaredNorm(), static_cast<Scalar>(1), prec))
+    if(!internal::isApprox(self.col(i).squaredNorm(), static_cast<RealScalar>(1), prec))
       return false;
-    for(int j = 0; j < i; ++j)
-      if(!ei_isMuchSmallerThan(nested.col(i).dot(nested.col(j)), static_cast<Scalar>(1), prec))
+    for(Index j = 0; j < i; ++j)
+      if(!internal::isMuchSmallerThan(self.col(i).dot(self.col(j)), static_cast<Scalar>(1), prec))
         return false;
   }
   return true;
 }
+
+} // end namespace Eigen
+
 #endif // EIGEN_DOT_H

Eigen/src/Core/EigenBase.h

@@ -0,0 +1,159 @@
+// This file is part of Eigen, a lightweight C++ template library
+// for linear algebra.
+//
+// Copyright (C) 2009 Benoit Jacob <jacob.benoit.1@gmail.com>
+// Copyright (C) 2009 Gael Guennebaud <gael.guennebaud@inria.fr>
+//
+// This Source Code Form is subject to the terms of the Mozilla
+// Public License v. 2.0. If a copy of the MPL was not distributed
+// with this file, You can obtain one at http://mozilla.org/MPL/2.0/.
+
+#ifndef EIGEN_EIGENBASE_H
+#define EIGEN_EIGENBASE_H
+
+namespace Eigen {
+
+/** \class EigenBase
+  * \ingroup Core_Module
+  * 
+  * Common base class for all classes T such that MatrixBase has an operator=(T) and a constructor MatrixBase(T).
+  *
+  * In other words, an EigenBase object is an object that can be copied into a MatrixBase.
+  *
+  * Besides MatrixBase-derived classes, this also includes special matrix classes such as diagonal matrices, etc.
+  *
+  * Notice that this class is trivial, it is only used to disambiguate overloaded functions.
+  *
+  * \sa \blank \ref TopicClassHierarchy
+  */
+template<typename Derived> struct EigenBase
+{
+//   typedef typename internal::plain_matrix_type<Derived>::type PlainObject;
+  
+  /** \brief The interface type of indices
+    * \details To change this, \c \#define the preprocessor symbol \c EIGEN_DEFAULT_DENSE_INDEX_TYPE.
+    * \deprecated Since Eigen 3.3, its usage is deprecated. Use Eigen::Index instead.
+    * \sa StorageIndex, \ref TopicPreprocessorDirectives.
+    */
+  typedef Eigen::Index Index;
+
+  // FIXME is it needed?
+  typedef typename internal::traits<Derived>::StorageKind StorageKind;
+
+  /** \returns a reference to the derived object */
+  EIGEN_DEVICE_FUNC
+  Derived& derived() { return *static_cast<Derived*>(this); }
+  /** \returns a const reference to the derived object */
+  EIGEN_DEVICE_FUNC
+  const Derived& derived() const { return *static_cast<const Derived*>(this); }
+
+  EIGEN_DEVICE_FUNC
+  inline Derived& const_cast_derived() const
+  { return *static_cast<Derived*>(const_cast<EigenBase*>(this)); }
+  EIGEN_DEVICE_FUNC
+  inline const Derived& const_derived() const
+  { return *static_cast<const Derived*>(this); }
+
+  /** \returns the number of rows. \sa cols(), RowsAtCompileTime */
+  EIGEN_DEVICE_FUNC
+  inline Index rows() const { return derived().rows(); }
+  /** \returns the number of columns. \sa rows(), ColsAtCompileTime*/
+  EIGEN_DEVICE_FUNC
+  inline Index cols() const { return derived().cols(); }
+  /** \returns the number of coefficients, which is rows()*cols().
+    * \sa rows(), cols(), SizeAtCompileTime. */
+  EIGEN_DEVICE_FUNC
+  inline Index size() const { return rows() * cols(); }
+
+  /** \internal Don't use it, but do the equivalent: \code dst = *this; \endcode */
+  template<typename Dest>
+  EIGEN_DEVICE_FUNC
+  inline void evalTo(Dest& dst) const
+  { derived().evalTo(dst); }
+
+  /** \internal Don't use it, but do the equivalent: \code dst += *this; \endcode */
+  template<typename Dest>
+  EIGEN_DEVICE_FUNC
+  inline void addTo(Dest& dst) const
+  {
+    // This is the default implementation,
+    // derived class can reimplement it in a more optimized way.
+    typename Dest::PlainObject res(rows(),cols());
+    evalTo(res);
+    dst += res;
+  }
+
+  /** \internal Don't use it, but do the equivalent: \code dst -= *this; \endcode */
+  template<typename Dest>
+  EIGEN_DEVICE_FUNC
+  inline void subTo(Dest& dst) const
+  {
+    // This is the default implementation,
+    // derived class can reimplement it in a more optimized way.
+    typename Dest::PlainObject res(rows(),cols());
+    evalTo(res);
+    dst -= res;
+  }
+
+  /** \internal Don't use it, but do the equivalent: \code dst.applyOnTheRight(*this); \endcode */
+  template<typename Dest>
+  EIGEN_DEVICE_FUNC inline void applyThisOnTheRight(Dest& dst) const
+  {
+    // This is the default implementation,
+    // derived class can reimplement it in a more optimized way.
+    dst = dst * this->derived();
+  }
+
+  /** \internal Don't use it, but do the equivalent: \code dst.applyOnTheLeft(*this); \endcode */
+  template<typename Dest>
+  EIGEN_DEVICE_FUNC inline void applyThisOnTheLeft(Dest& dst) const
+  {
+    // This is the default implementation,
+    // derived class can reimplement it in a more optimized way.
+    dst = this->derived() * dst;
+  }
+
+};
+
+/***************************************************************************
+* Implementation of matrix base methods
+***************************************************************************/
+
+/** \brief Copies the generic expression \a other into *this.
+  *
+  * \details The expression must provide a (templated) evalTo(Derived& dst) const
+  * function which does the actual job. In practice, this allows any user to write
+  * its own special matrix without having to modify MatrixBase
+  *
+  * \returns a reference to *this.
+  */
+template<typename Derived>
+template<typename OtherDerived>
+EIGEN_DEVICE_FUNC
+Derived& DenseBase<Derived>::operator=(const EigenBase<OtherDerived> &other)
+{
+  call_assignment(derived(), other.derived());
+  return derived();
+}
+
+template<typename Derived>
+template<typename OtherDerived>
+EIGEN_DEVICE_FUNC
+Derived& DenseBase<Derived>::operator+=(const EigenBase<OtherDerived> &other)
+{
+  call_assignment(derived(), other.derived(), internal::add_assign_op<Scalar,typename OtherDerived::Scalar>());
+  return derived();
+}
+
+template<typename Derived>
+template<typename OtherDerived>
+EIGEN_DEVICE_FUNC
+Derived& DenseBase<Derived>::operator-=(const EigenBase<OtherDerived> &other)
+{
+  call_assignment(derived(), other.derived(), internal::sub_assign_op<Scalar,typename OtherDerived::Scalar>());
+  return derived();
+}
+
+} // end namespace Eigen
+
+#endif // EIGEN_EIGENBASE_H

Eigen/src/Core/Flagged.h

@@ -1,149 +0,0 @@
-// This file is part of Eigen, a lightweight C++ template library
-// for linear algebra. Eigen itself is part of the KDE project.
-//
-// Copyright (C) 2008 Benoit Jacob <jacob.benoit.1@gmail.com>
-//
-// 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_FLAGGED_H
-#define EIGEN_FLAGGED_H
-
-/** \class Flagged
-  *
-  * \brief Expression with modified flags
-  *
-  * \param ExpressionType the type of the object of which we are modifying the flags
-  * \param Added the flags added to the expression
-  * \param Removed the flags removed from the expression (has priority over Added).
-  *
-  * This class represents an expression whose flags have been modified.
-  * It is the return type of MatrixBase::flagged()
-  * and most of the time this is the only way it is used.
-  *
-  * \sa MatrixBase::flagged()
-  */
-template<typename ExpressionType, unsigned int Added, unsigned int Removed>
-struct ei_traits<Flagged<ExpressionType, Added, Removed> > : ei_traits<ExpressionType>
-{
-  enum { Flags = (ExpressionType::Flags | Added) & ~Removed };
-};
-
-template<typename ExpressionType, unsigned int Added, unsigned int Removed> class Flagged
-  : public MatrixBase<Flagged<ExpressionType, Added, Removed> >
-{
-  public:
-
-    EIGEN_GENERIC_PUBLIC_INTERFACE(Flagged)
-    typedef typename ei_meta_if<ei_must_nest_by_value<ExpressionType>::ret,
-        ExpressionType, const ExpressionType&>::ret ExpressionTypeNested;
-    typedef typename ExpressionType::InnerIterator InnerIterator;
-
-    inline Flagged(const ExpressionType& matrix) : m_matrix(matrix) {}
-
-    inline int rows() const { return m_matrix.rows(); }
-    inline int cols() const { return m_matrix.cols(); }
-    inline int stride() const { return m_matrix.stride(); }
-
-    inline const Scalar coeff(int row, int col) const
-    {
-      return m_matrix.coeff(row, col);
-    }
-
-    inline Scalar& coeffRef(int row, int col)
-    {
-      return m_matrix.const_cast_derived().coeffRef(row, col);
-    }
-
-    inline const Scalar coeff(int index) const
-    {
-      return m_matrix.coeff(index);
-    }
-
-    inline Scalar& coeffRef(int index)
-    {
-      return m_matrix.const_cast_derived().coeffRef(index);
-    }
-
-    template<int LoadMode>
-    inline const PacketScalar packet(int row, int col) const
-    {
-      return m_matrix.template packet<LoadMode>(row, col);
-    }
-
-    template<int LoadMode>
-    inline void writePacket(int row, int col, const PacketScalar& x)
-    {
-      m_matrix.const_cast_derived().template writePacket<LoadMode>(row, col, x);
-    }
-
-    template<int LoadMode>
-    inline const PacketScalar packet(int index) const
-    {
-      return m_matrix.template packet<LoadMode>(index);
-    }
-
-    template<int LoadMode>
-    inline void writePacket(int index, const PacketScalar& x)
-    {
-      m_matrix.const_cast_derived().template writePacket<LoadMode>(index, x);
-    }
-
-    const ExpressionType& _expression() const { return m_matrix; }
-
-  protected:
-    ExpressionTypeNested m_matrix;
-
-private:
-  Flagged& operator=(const Flagged&);
-};
-
-/** \returns an expression of *this with added flags
-  *
-  * \addexample MarkExample \label How to mark a triangular matrix as triangular
-  *
-  * Example: \include MatrixBase_marked.cpp
-  * Output: \verbinclude MatrixBase_marked.out
-  *
-  * \sa class Flagged, extract(), part()
-  */
-template<typename Derived>
-template<unsigned int Added>
-inline const Flagged<Derived, Added, 0>
-MatrixBase<Derived>::marked() const
-{
-  return derived();
-}
-
-/** \returns an expression of *this with the following flags removed:
-  * EvalBeforeNestingBit and EvalBeforeAssigningBit.
-  *
-  * Example: \include MatrixBase_lazy.cpp
-  * Output: \verbinclude MatrixBase_lazy.out
-  *
-  * \sa class Flagged, marked()
-  */
-template<typename Derived>
-inline const Flagged<Derived, 0, EvalBeforeNestingBit | EvalBeforeAssigningBit>
-MatrixBase<Derived>::lazy() const
-{
-  return derived();
-}
-
-#endif // EIGEN_FLAGGED_H

Eigen/src/Core/ForceAlignedAccess.h

@@ -0,0 +1,146 @@
+// This file is part of Eigen, a lightweight C++ template library
+// for linear algebra.
+//
+// Copyright (C) 2009-2010 Gael Guennebaud <gael.guennebaud@inria.fr>
+//
+// This Source Code Form is subject to the terms of the Mozilla
+// Public License v. 2.0. If a copy of the MPL was not distributed
+// with this file, You can obtain one at http://mozilla.org/MPL/2.0/.
+
+#ifndef EIGEN_FORCEALIGNEDACCESS_H
+#define EIGEN_FORCEALIGNEDACCESS_H
+
+namespace Eigen {
+
+/** \class ForceAlignedAccess
+  * \ingroup Core_Module
+  *
+  * \brief Enforce aligned packet loads and stores regardless of what is requested
+  *
+  * \param ExpressionType the type of the object of which we are forcing aligned packet access
+  *
+  * This class is the return type of MatrixBase::forceAlignedAccess()
+  * and most of the time this is the only way it is used.
+  *
+  * \sa MatrixBase::forceAlignedAccess()
+  */
+
+namespace internal {
+template<typename ExpressionType>
+struct traits<ForceAlignedAccess<ExpressionType> > : public traits<ExpressionType>
+{};
+}
+
+template<typename ExpressionType> class ForceAlignedAccess
+  : public internal::dense_xpr_base< ForceAlignedAccess<ExpressionType> >::type
+{
+  public:
+
+    typedef typename internal::dense_xpr_base<ForceAlignedAccess>::type Base;
+    EIGEN_DENSE_PUBLIC_INTERFACE(ForceAlignedAccess)
+
+    EIGEN_DEVICE_FUNC explicit inline ForceAlignedAccess(const ExpressionType& matrix) : m_expression(matrix) {}
+
+    EIGEN_DEVICE_FUNC inline Index rows() const { return m_expression.rows(); }
+    EIGEN_DEVICE_FUNC inline Index cols() const { return m_expression.cols(); }
+    EIGEN_DEVICE_FUNC inline Index outerStride() const { return m_expression.outerStride(); }
+    EIGEN_DEVICE_FUNC inline Index innerStride() const { return m_expression.innerStride(); }
+
+    EIGEN_DEVICE_FUNC inline const CoeffReturnType coeff(Index row, Index col) const
+    {
+      return m_expression.coeff(row, col);
+    }
+
+    EIGEN_DEVICE_FUNC inline Scalar& coeffRef(Index row, Index col)
+    {
+      return m_expression.const_cast_derived().coeffRef(row, col);
+    }
+
+    EIGEN_DEVICE_FUNC inline const CoeffReturnType coeff(Index index) const
+    {
+      return m_expression.coeff(index);
+    }
+
+    EIGEN_DEVICE_FUNC inline Scalar& coeffRef(Index index)
+    {
+      return m_expression.const_cast_derived().coeffRef(index);
+    }
+
+    template<int LoadMode>
+    inline const PacketScalar packet(Index row, Index col) const
+    {
+      return m_expression.template packet<Aligned>(row, col);
+    }
+
+    template<int LoadMode>
+    inline void writePacket(Index row, Index col, const PacketScalar& x)
+    {
+      m_expression.const_cast_derived().template writePacket<Aligned>(row, col, x);
+    }
+
+    template<int LoadMode>
+    inline const PacketScalar packet(Index index) const
+    {
+      return m_expression.template packet<Aligned>(index);
+    }
+
+    template<int LoadMode>
+    inline void writePacket(Index index, const PacketScalar& x)
+    {
+      m_expression.const_cast_derived().template writePacket<Aligned>(index, x);
+    }
+
+    EIGEN_DEVICE_FUNC operator const ExpressionType&() const { return m_expression; }
+
+  protected:
+    const ExpressionType& m_expression;
+
+  private:
+    ForceAlignedAccess& operator=(const ForceAlignedAccess&);
+};
+
+/** \returns an expression of *this with forced aligned access
+  * \sa forceAlignedAccessIf(),class ForceAlignedAccess
+  */
+template<typename Derived>
+inline const ForceAlignedAccess<Derived>
+MatrixBase<Derived>::forceAlignedAccess() const
+{
+  return ForceAlignedAccess<Derived>(derived());
+}
+
+/** \returns an expression of *this with forced aligned access
+  * \sa forceAlignedAccessIf(), class ForceAlignedAccess
+  */
+template<typename Derived>
+inline ForceAlignedAccess<Derived>
+MatrixBase<Derived>::forceAlignedAccess()
+{
+  return ForceAlignedAccess<Derived>(derived());
+}
+
+/** \returns an expression of *this with forced aligned access if \a Enable is true.
+  * \sa forceAlignedAccess(), class ForceAlignedAccess
+  */
+template<typename Derived>
+template<bool Enable>
+inline typename internal::add_const_on_value_type<typename internal::conditional<Enable,ForceAlignedAccess<Derived>,Derived&>::type>::type
+MatrixBase<Derived>::forceAlignedAccessIf() const
+{
+  return derived();  // FIXME This should not work but apparently is never used
+}
+
+/** \returns an expression of *this with forced aligned access if \a Enable is true.
+  * \sa forceAlignedAccess(), class ForceAlignedAccess
+  */
+template<typename Derived>
+template<bool Enable>
+inline typename internal::conditional<Enable,ForceAlignedAccess<Derived>,Derived&>::type
+MatrixBase<Derived>::forceAlignedAccessIf()
+{
+  return derived();  // FIXME This should not work but apparently is never used
+}
+
+} // end namespace Eigen
+
+#endif // EIGEN_FORCEALIGNEDACCESS_H

Eigen/src/Core/Functors.h

@@ -1,378 +0,0 @@
-// This file is part of Eigen, a lightweight C++ template library
-// for linear algebra. Eigen itself is part of the KDE project.
-//
-// Copyright (C) 2008 Gael Guennebaud <g.gael@free.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_FUNCTORS_H
-#define EIGEN_FUNCTORS_H
-
-// associative functors:
-
-/** \internal
-  * \brief Template functor to compute the sum of two scalars
-  *
-  * \sa class CwiseBinaryOp, MatrixBase::operator+, class PartialRedux, MatrixBase::sum()
-  */
-template<typename Scalar> struct ei_scalar_sum_op EIGEN_EMPTY_STRUCT {
-  EIGEN_STRONG_INLINE const Scalar operator() (const Scalar& a, const Scalar& b) const { return a + b; }
-  template<typename PacketScalar>
-  EIGEN_STRONG_INLINE const PacketScalar packetOp(const PacketScalar& a, const PacketScalar& b) const
-  { return ei_padd(a,b); }
-};
-template<typename Scalar>
-struct ei_functor_traits<ei_scalar_sum_op<Scalar> > {
-  enum {
-    Cost = NumTraits<Scalar>::AddCost,
-    PacketAccess = ei_packet_traits<Scalar>::size>1
-  };
-};
-
-/** \internal
-  * \brief Template functor to compute the product of two scalars
-  *
-  * \sa class CwiseBinaryOp, Cwise::operator*(), class PartialRedux, MatrixBase::redux()
-  */
-template<typename Scalar> struct ei_scalar_product_op EIGEN_EMPTY_STRUCT {
-  EIGEN_STRONG_INLINE const Scalar operator() (const Scalar& a, const Scalar& b) const { return a * b; }
-  template<typename PacketScalar>
-  EIGEN_STRONG_INLINE const PacketScalar packetOp(const PacketScalar& a, const PacketScalar& b) const
-  { return ei_pmul(a,b); }
-};
-template<typename Scalar>
-struct ei_functor_traits<ei_scalar_product_op<Scalar> > {
-  enum {
-    Cost = NumTraits<Scalar>::MulCost,
-    PacketAccess = ei_packet_traits<Scalar>::size>1
-  };
-};
-
-/** \internal
-  * \brief Template functor to compute the min of two scalars
-  *
-  * \sa class CwiseBinaryOp, MatrixBase::cwiseMin, class PartialRedux, MatrixBase::minCoeff()
-  */
-template<typename Scalar> struct ei_scalar_min_op EIGEN_EMPTY_STRUCT {
-  EIGEN_STRONG_INLINE const Scalar operator() (const Scalar& a, const Scalar& b) const { return std::min(a, b); }
-  template<typename PacketScalar>
-  EIGEN_STRONG_INLINE const PacketScalar packetOp(const PacketScalar& a, const PacketScalar& b) const
-  { return ei_pmin(a,b); }
-};
-template<typename Scalar>
-struct ei_functor_traits<ei_scalar_min_op<Scalar> > {
-  enum {
-    Cost = NumTraits<Scalar>::AddCost,
-    PacketAccess = ei_packet_traits<Scalar>::size>1
-  };
-};
-
-/** \internal
-  * \brief Template functor to compute the max of two scalars
-  *
-  * \sa class CwiseBinaryOp, MatrixBase::cwiseMax, class PartialRedux, MatrixBase::maxCoeff()
-  */
-template<typename Scalar> struct ei_scalar_max_op EIGEN_EMPTY_STRUCT {
-  EIGEN_STRONG_INLINE const Scalar operator() (const Scalar& a, const Scalar& b) const { return std::max(a, b); }
-  template<typename PacketScalar>
-  EIGEN_STRONG_INLINE const PacketScalar packetOp(const PacketScalar& a, const PacketScalar& b) const
-  { return ei_pmax(a,b); }
-};
-template<typename Scalar>
-struct ei_functor_traits<ei_scalar_max_op<Scalar> > {
-  enum {
-    Cost = NumTraits<Scalar>::AddCost,
-    PacketAccess = ei_packet_traits<Scalar>::size>1
-  };
-};
-
-
-// other binary functors:
-
-/** \internal
-  * \brief Template functor to compute the difference of two scalars
-  *
-  * \sa class CwiseBinaryOp, MatrixBase::operator-
-  */
-template<typename Scalar> struct ei_scalar_difference_op EIGEN_EMPTY_STRUCT {
-  EIGEN_STRONG_INLINE const Scalar operator() (const Scalar& a, const Scalar& b) const { return a - b; }
-  template<typename PacketScalar>
-  EIGEN_STRONG_INLINE const PacketScalar packetOp(const PacketScalar& a, const PacketScalar& b) const
-  { return ei_psub(a,b); }
-};
-template<typename Scalar>
-struct ei_functor_traits<ei_scalar_difference_op<Scalar> > {
-  enum {
-    Cost = NumTraits<Scalar>::AddCost,
-    PacketAccess = ei_packet_traits<Scalar>::size>1
-  };
-};
-
-/** \internal
-  * \brief Template functor to compute the quotient of two scalars
-  *
-  * \sa class CwiseBinaryOp, Cwise::operator/()
-  */
-template<typename Scalar> struct ei_scalar_quotient_op EIGEN_EMPTY_STRUCT {
-  EIGEN_STRONG_INLINE const Scalar operator() (const Scalar& a, const Scalar& b) const { return a / b; }
-  template<typename PacketScalar>
-  EIGEN_STRONG_INLINE const PacketScalar packetOp(const PacketScalar& a, const PacketScalar& b) const
-  { return ei_pdiv(a,b); }
-};
-template<typename Scalar>
-struct ei_functor_traits<ei_scalar_quotient_op<Scalar> > {
-  enum {
-    Cost = 2 * NumTraits<Scalar>::MulCost,
-    PacketAccess = ei_packet_traits<Scalar>::size>1
-                  #if (defined EIGEN_VECTORIZE_SSE)
-                  && NumTraits<Scalar>::HasFloatingPoint
-                  #endif
-  };
-};
-
-// unary functors:
-
-/** \internal
-  * \brief Template functor to compute the opposite of a scalar
-  *
-  * \sa class CwiseUnaryOp, MatrixBase::operator-
-  */
-template<typename Scalar> struct ei_scalar_opposite_op EIGEN_EMPTY_STRUCT {
-  EIGEN_STRONG_INLINE const Scalar operator() (const Scalar& a) const { return -a; }
-};
-template<typename Scalar>
-struct ei_functor_traits<ei_scalar_opposite_op<Scalar> >
-{ enum { Cost = NumTraits<Scalar>::AddCost, PacketAccess = false }; };
-
-/** \internal
-  * \brief Template functor to compute the absolute value of a scalar
-  *
-  * \sa class CwiseUnaryOp, Cwise::abs
-  */
-template<typename Scalar> struct ei_scalar_abs_op EIGEN_EMPTY_STRUCT {
-  typedef typename NumTraits<Scalar>::Real result_type;
-  EIGEN_STRONG_INLINE const result_type operator() (const Scalar& a) const { return ei_abs(a); }
-};
-template<typename Scalar>
-struct ei_functor_traits<ei_scalar_abs_op<Scalar> >
-{
-  enum {
-    Cost = NumTraits<Scalar>::AddCost,
-    PacketAccess = false // this could actually be vectorized with SSSE3.
-  };
-};
-
-/** \internal
-  * \brief Template functor to compute the squared absolute value of a scalar
-  *
-  * \sa class CwiseUnaryOp, Cwise::abs2
-  */
-template<typename Scalar> struct ei_scalar_abs2_op EIGEN_EMPTY_STRUCT {
-  typedef typename NumTraits<Scalar>::Real result_type;
-  EIGEN_STRONG_INLINE const result_type operator() (const Scalar& a) const { return ei_abs2(a); }
-  template<typename PacketScalar>
-  EIGEN_STRONG_INLINE const PacketScalar packetOp(const PacketScalar& a) const
-  { return ei_pmul(a,a); }
-};
-template<typename Scalar>
-struct ei_functor_traits<ei_scalar_abs2_op<Scalar> >
-{ enum { Cost = NumTraits<Scalar>::MulCost, PacketAccess = int(ei_packet_traits<Scalar>::size)>1 }; };
-
-/** \internal
-  * \brief Template functor to compute the conjugate of a complex value
-  *
-  * \sa class CwiseUnaryOp, MatrixBase::conjugate()
-  */
-template<typename Scalar> struct ei_scalar_conjugate_op EIGEN_EMPTY_STRUCT {
-  EIGEN_STRONG_INLINE const Scalar operator() (const Scalar& a) const { return ei_conj(a); }
-  template<typename PacketScalar>
-  EIGEN_STRONG_INLINE const PacketScalar packetOp(const PacketScalar& a) const { return a; }
-};
-template<typename Scalar>
-struct ei_functor_traits<ei_scalar_conjugate_op<Scalar> >
-{
-  enum {
-    Cost = NumTraits<Scalar>::IsComplex ? NumTraits<Scalar>::AddCost : 0,
-    PacketAccess = int(ei_packet_traits<Scalar>::size)>1
-  };
-};
-
-/** \internal
-  * \brief Template functor to cast a scalar to another type
-  *
-  * \sa class CwiseUnaryOp, MatrixBase::cast()
-  */
-template<typename Scalar, typename NewType>
-struct ei_scalar_cast_op EIGEN_EMPTY_STRUCT {
-  typedef NewType result_type;
-  EIGEN_STRONG_INLINE const NewType operator() (const Scalar& a) const { return static_cast<NewType>(a); }
-};
-template<typename Scalar, typename NewType>
-struct ei_functor_traits<ei_scalar_cast_op<Scalar,NewType> >
-{ enum { Cost = ei_is_same_type<Scalar, NewType>::ret ? 0 : NumTraits<NewType>::AddCost, PacketAccess = false }; };
-
-/** \internal
-  * \brief Template functor to extract the real part of a complex
-  *
-  * \sa class CwiseUnaryOp, MatrixBase::real()
-  */
-template<typename Scalar>
-struct ei_scalar_real_op EIGEN_EMPTY_STRUCT {
-  typedef typename NumTraits<Scalar>::Real result_type;
-  EIGEN_STRONG_INLINE result_type operator() (const Scalar& a) const { return ei_real(a); }
-};
-template<typename Scalar>
-struct ei_functor_traits<ei_scalar_real_op<Scalar> >
-{ enum { Cost = 0, PacketAccess = false }; };
-
-/** \internal
-  * \brief Template functor to extract the imaginary part of a complex
-  *
-  * \sa class CwiseUnaryOp, MatrixBase::imag()
-  */
-template<typename Scalar>
-struct ei_scalar_imag_op EIGEN_EMPTY_STRUCT {
-  typedef typename NumTraits<Scalar>::Real result_type;
-  EIGEN_STRONG_INLINE result_type operator() (const Scalar& a) const { return ei_imag(a); }
-};
-template<typename Scalar>
-struct ei_functor_traits<ei_scalar_imag_op<Scalar> >
-{ enum { Cost = 0, PacketAccess = false }; };
-
-/** \internal
-  * \brief Template functor to multiply a scalar by a fixed other one
-  *
-  * \sa class CwiseUnaryOp, MatrixBase::operator*, MatrixBase::operator/
-  */
-/* NOTE why doing the ei_pset1() in packetOp *is* an optimization ?
- * indeed it seems better to declare m_other as a PacketScalar and do the ei_pset1() once
- * in the constructor. However, in practice:
- *  - GCC does not like m_other as a PacketScalar and generate a load every time it needs it
- *  - one the other hand GCC is able to moves the ei_pset1() away the loop :)
- *  - simpler code ;)
- * (ICC and gcc 4.4 seems to perform well in both cases, the issue is visible with y = a*x + b*y)
- */
-template<typename Scalar>
-struct ei_scalar_multiple_op {
-  typedef typename ei_packet_traits<Scalar>::type PacketScalar;
-  // FIXME default copy constructors seems bugged with std::complex<>
-  EIGEN_STRONG_INLINE ei_scalar_multiple_op(const ei_scalar_multiple_op& other) : m_other(other.m_other) { }
-  EIGEN_STRONG_INLINE ei_scalar_multiple_op(const Scalar& other) : m_other(other) { }
-  EIGEN_STRONG_INLINE Scalar operator() (const Scalar& a) const { return a * m_other; }
-  EIGEN_STRONG_INLINE const PacketScalar packetOp(const PacketScalar& a) const
-  { return ei_pmul(a, ei_pset1(m_other)); }
-  const Scalar m_other;
-private:
-  ei_scalar_multiple_op& operator=(const ei_scalar_multiple_op&);
-};
-template<typename Scalar>
-struct ei_functor_traits<ei_scalar_multiple_op<Scalar> >
-{ enum { Cost = NumTraits<Scalar>::MulCost, PacketAccess = ei_packet_traits<Scalar>::size>1 }; };
-
-template<typename Scalar, bool HasFloatingPoint>
-struct ei_scalar_quotient1_impl {
-  typedef typename ei_packet_traits<Scalar>::type PacketScalar;
-  // FIXME default copy constructors seems bugged with std::complex<>
-  EIGEN_STRONG_INLINE ei_scalar_quotient1_impl(const ei_scalar_quotient1_impl& other) : m_other(other.m_other) { }
-  EIGEN_STRONG_INLINE ei_scalar_quotient1_impl(const Scalar& other) : m_other(static_cast<Scalar>(1) / other) {}
-  EIGEN_STRONG_INLINE Scalar operator() (const Scalar& a) const { return a * m_other; }
-  EIGEN_STRONG_INLINE const PacketScalar packetOp(const PacketScalar& a) const
-  { return ei_pmul(a, ei_pset1(m_other)); }
-  const Scalar m_other;
-private:
-  ei_scalar_quotient1_impl& operator=(const ei_scalar_quotient1_impl&);
-};
-template<typename Scalar>
-struct ei_functor_traits<ei_scalar_quotient1_impl<Scalar,true> >
-{ enum { Cost = NumTraits<Scalar>::MulCost, PacketAccess = ei_packet_traits<Scalar>::size>1 }; };
-
-template<typename Scalar>
-struct ei_scalar_quotient1_impl<Scalar,false> {
-  // FIXME default copy constructors seems bugged with std::complex<>
-  EIGEN_STRONG_INLINE ei_scalar_quotient1_impl(const ei_scalar_quotient1_impl& other) : m_other(other.m_other) { }
-  EIGEN_STRONG_INLINE ei_scalar_quotient1_impl(const Scalar& other) : m_other(other) {}
-  EIGEN_STRONG_INLINE Scalar operator() (const Scalar& a) const { return a / m_other; }
-  const Scalar m_other;
-private:
-  ei_scalar_quotient1_impl& operator=(const ei_scalar_quotient1_impl&);
-};
-template<typename Scalar>
-struct ei_functor_traits<ei_scalar_quotient1_impl<Scalar,false> >
-{ enum { Cost = 2 * NumTraits<Scalar>::MulCost, PacketAccess = false }; };
-
-/** \internal
-  * \brief Template functor to divide a scalar by a fixed other one
-  *
-  * This functor is used to implement the quotient of a matrix by
-  * a scalar where the scalar type is not necessarily a floating point type.
-  *
-  * \sa class CwiseUnaryOp, MatrixBase::operator/
-  */
-template<typename Scalar>
-struct ei_scalar_quotient1_op : ei_scalar_quotient1_impl<Scalar, NumTraits<Scalar>::HasFloatingPoint > {
-  EIGEN_STRONG_INLINE ei_scalar_quotient1_op(const Scalar& other)
-    : ei_scalar_quotient1_impl<Scalar, NumTraits<Scalar>::HasFloatingPoint >(other) {}
-private:
-  ei_scalar_quotient1_op& operator=(const ei_scalar_quotient1_op&);
-};
-
-// nullary functors
-
-template<typename Scalar>
-struct ei_scalar_constant_op {
-  typedef typename ei_packet_traits<Scalar>::type PacketScalar;
-  EIGEN_STRONG_INLINE ei_scalar_constant_op(const ei_scalar_constant_op& other) : m_other(other.m_other) { }
-  EIGEN_STRONG_INLINE ei_scalar_constant_op(const Scalar& other) : m_other(other) { }
-  EIGEN_STRONG_INLINE const Scalar operator() (int, int = 0) const { return m_other; }
-  EIGEN_STRONG_INLINE const PacketScalar packetOp() const { return ei_pset1(m_other); }
-  const Scalar m_other;
-private:
-  ei_scalar_constant_op& operator=(const ei_scalar_constant_op&);
-};
-template<typename Scalar>
-struct ei_functor_traits<ei_scalar_constant_op<Scalar> >
-{ enum { Cost = 1, PacketAccess = ei_packet_traits<Scalar>::size>1, IsRepeatable = true }; };
-
-template<typename Scalar> struct ei_scalar_identity_op EIGEN_EMPTY_STRUCT {
-  EIGEN_STRONG_INLINE ei_scalar_identity_op(void) {}
-  EIGEN_STRONG_INLINE const Scalar operator() (int row, int col) const { return row==col ? Scalar(1) : Scalar(0); }
-};
-template<typename Scalar>
-struct ei_functor_traits<ei_scalar_identity_op<Scalar> >
-{ enum { Cost = NumTraits<Scalar>::AddCost, PacketAccess = false, IsRepeatable = true }; };
-
-// allow to add new functors and specializations of ei_functor_traits from outside Eigen.
-// this macro is really needed because ei_functor_traits must be specialized after it is declared but before it is used...
-#ifdef EIGEN_FUNCTORS_PLUGIN
-#include EIGEN_FUNCTORS_PLUGIN
-#endif
-
-// all functors allow linear access, except ei_scalar_identity_op. So we fix here a quick meta
-// to indicate whether a functor allows linear access, just always answering 'yes' except for
-// ei_scalar_identity_op.
-template<typename Functor> struct ei_functor_has_linear_access { enum { ret = 1 }; };
-template<typename Scalar> struct ei_functor_has_linear_access<ei_scalar_identity_op<Scalar> > { enum { ret = 0 }; };
-
-// in CwiseBinaryOp, we require the Lhs and Rhs to have the same scalar type, except for multiplication
-// where we only require them to have the same _real_ scalar type so one may multiply, say, float by complex<float>.
-template<typename Functor> struct ei_functor_allows_mixing_real_and_complex { enum { ret = 0 }; };
-template<typename Scalar> struct ei_functor_allows_mixing_real_and_complex<ei_scalar_product_op<Scalar> > { enum { ret = 1 }; };
-
-#endif // EIGEN_FUNCTORS_H

Eigen/src/Core/Fuzzy.h

@@ -1,32 +1,85 @@
 // This file is part of Eigen, a lightweight C++ template library
-// for linear algebra. Eigen itself is part of the KDE project.
+// for linear algebra.
 //
 // Copyright (C) 2006-2008 Benoit Jacob <jacob.benoit.1@gmail.com>
-// Copyright (C) 2008 Gael Guennebaud <g.gael@free.fr>
+// 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
-// 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/>.
+// This Source Code Form is subject to the terms of the Mozilla
+// Public License v. 2.0. If a copy of the MPL was not distributed
+// with this file, You can obtain one at http://mozilla.org/MPL/2.0/.
 
 #ifndef EIGEN_FUZZY_H
 #define EIGEN_FUZZY_H
 
-#ifndef EIGEN_LEGACY_COMPARES
+namespace Eigen { 
+
+namespace internal
+{
+
+template<typename Derived, typename OtherDerived, bool is_integer = NumTraits<typename Derived::Scalar>::IsInteger>
+struct isApprox_selector
+{
+  EIGEN_DEVICE_FUNC
+  static bool run(const Derived& x, const OtherDerived& y, const typename Derived::RealScalar& prec)
+  {
+    typename internal::nested_eval<Derived,2>::type nested(x);
+    typename internal::nested_eval<OtherDerived,2>::type otherNested(y);
+    return (nested - otherNested).cwiseAbs2().sum() <= prec * prec * numext::mini(nested.cwiseAbs2().sum(), otherNested.cwiseAbs2().sum());
+  }
+};
+
+template<typename Derived, typename OtherDerived>
+struct isApprox_selector<Derived, OtherDerived, true>
+{
+  EIGEN_DEVICE_FUNC
+  static bool run(const Derived& x, const OtherDerived& y, const typename Derived::RealScalar&)
+  {
+    return x.matrix() == y.matrix();
+  }
+};
+
+template<typename Derived, typename OtherDerived, bool is_integer = NumTraits<typename Derived::Scalar>::IsInteger>
+struct isMuchSmallerThan_object_selector
+{
+  EIGEN_DEVICE_FUNC
+  static bool run(const Derived& x, const OtherDerived& y, const typename Derived::RealScalar& prec)
+  {
+    return x.cwiseAbs2().sum() <= numext::abs2(prec) * y.cwiseAbs2().sum();
+  }
+};
+
+template<typename Derived, typename OtherDerived>
+struct isMuchSmallerThan_object_selector<Derived, OtherDerived, true>
+{
+  EIGEN_DEVICE_FUNC
+  static bool run(const Derived& x, const OtherDerived&, const typename Derived::RealScalar&)
+  {
+    return x.matrix() == Derived::Zero(x.rows(), x.cols()).matrix();
+  }
+};
+
+template<typename Derived, bool is_integer = NumTraits<typename Derived::Scalar>::IsInteger>
+struct isMuchSmallerThan_scalar_selector
+{
+  EIGEN_DEVICE_FUNC
+  static bool run(const Derived& x, const typename Derived::RealScalar& y, const typename Derived::RealScalar& prec)
+  {
+    return x.cwiseAbs2().sum() <= numext::abs2(prec * y);
+  }
+};
+
+template<typename Derived>
+struct isMuchSmallerThan_scalar_selector<Derived, true>
+{
+  EIGEN_DEVICE_FUNC
+  static bool run(const Derived& x, const typename Derived::RealScalar&, const typename Derived::RealScalar&)
+  {
+    return x.matrix() == Derived::Zero(x.rows(), x.cols()).matrix();
+  }
+};
+
+} // end namespace internal
+
 
 /** \returns \c true if \c *this is approximately equal to \a other, within the precision
   * determined by \a prec.
@@ -40,21 +93,19 @@
   * \note Because of the multiplicativeness of this comparison, one can't use this function
   * to check whether \c *this is approximately equal to the zero matrix or vector.
   * Indeed, \c isApprox(zero) returns false unless \c *this itself is exactly the zero matrix
-  * or vector. If you want to test whether \c *this is zero, use ei_isMuchSmallerThan(const
+  * or vector. If you want to test whether \c *this is zero, use internal::isMuchSmallerThan(const
   * RealScalar&, RealScalar) instead.
   *
-  * \sa ei_isMuchSmallerThan(const RealScalar&, RealScalar) const
+  * \sa internal::isMuchSmallerThan(const RealScalar&, RealScalar) const
   */
 template<typename Derived>
 template<typename OtherDerived>
-bool MatrixBase<Derived>::isApprox(
-  const MatrixBase<OtherDerived>& other,
-  typename NumTraits<Scalar>::Real prec
+bool DenseBase<Derived>::isApprox(
+  const DenseBase<OtherDerived>& other,
+  const RealScalar& prec
 ) const
 {
-  const typename ei_nested<Derived,2>::type nested(derived());
-  const typename ei_nested<OtherDerived,2>::type otherNested(other.derived());
-  return (nested - otherNested).cwise().abs2().sum() <= prec * prec * std::min(nested.cwise().abs2().sum(), otherNested.cwise().abs2().sum());
+  return internal::isApprox_selector<Derived, OtherDerived>::run(derived(), other.derived(), prec);
 }
 
 /** \returns \c true if the norm of \c *this is much smaller than \a other,
@@ -68,15 +119,15 @@ bool MatrixBase<Derived>::isApprox(
   * the value of the reference scalar \a other should come from the Hilbert-Schmidt norm
   * of a reference matrix of same dimensions.
   *
-  * \sa isApprox(), isMuchSmallerThan(const MatrixBase<OtherDerived>&, RealScalar) const
+  * \sa isApprox(), isMuchSmallerThan(const DenseBase<OtherDerived>&, RealScalar) const
   */
 template<typename Derived>
-bool MatrixBase<Derived>::isMuchSmallerThan(
+bool DenseBase<Derived>::isMuchSmallerThan(
   const typename NumTraits<Scalar>::Real& other,
-  typename NumTraits<Scalar>::Real prec
+  const RealScalar& prec
 ) const
 {
-  return cwise().abs2().sum() <= prec * prec * other * other;
+  return internal::isMuchSmallerThan_scalar_selector<Derived>::run(derived(), other, prec);
 }
 
 /** \returns \c true if the norm of \c *this is much smaller than the norm of \a other,
@@ -91,144 +142,14 @@ bool MatrixBase<Derived>::isMuchSmallerThan(
   */
 template<typename Derived>
 template<typename OtherDerived>
-bool MatrixBase<Derived>::isMuchSmallerThan(
-  const MatrixBase<OtherDerived>& other,
-  typename NumTraits<Scalar>::Real prec
+bool DenseBase<Derived>::isMuchSmallerThan(
+  const DenseBase<OtherDerived>& other,
+  const RealScalar& prec
 ) const
 {
-  return this->cwise().abs2().sum() <= prec * prec * other.cwise().abs2().sum();
+  return internal::isMuchSmallerThan_object_selector<Derived, OtherDerived>::run(derived(), other.derived(), prec);
 }
 
-#else
-
-template<typename Derived, typename OtherDerived=Derived, bool IsVector=Derived::IsVectorAtCompileTime>
-struct ei_fuzzy_selector;
-
-/** \returns \c true if \c *this is approximately equal to \a other, within the precision
-  * determined by \a prec.
-  *
-  * \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]
-  * For matrices, the comparison is done on all columns.
-  *
-  * \note Because of the multiplicativeness of this comparison, one can't use this function
-  * to check whether \c *this is approximately equal to the zero matrix or vector.
-  * Indeed, \c isApprox(zero) returns false unless \c *this itself is exactly the zero matrix
-  * or vector. If you want to test whether \c *this is zero, use ei_isMuchSmallerThan(const
-  * RealScalar&, RealScalar) instead.
-  *
-  * \sa ei_isMuchSmallerThan(const RealScalar&, RealScalar) const
-  */
-template<typename Derived>
-template<typename OtherDerived>
-bool MatrixBase<Derived>::isApprox(
-  const MatrixBase<OtherDerived>& other,
-  typename NumTraits<Scalar>::Real prec
-) const
-{
-  return ei_fuzzy_selector<Derived,OtherDerived>::isApprox(derived(), other.derived(), prec);
-}
-
-/** \returns \c true if the norm of \c *this is much smaller than \a other,
-  * within the precision determined by \a prec.
-  *
-  * \note The fuzzy compares are done multiplicatively. A vector \f$ v \f$ is
-  * considered to be much smaller than \f$ x \f$ within precision \f$ p \f$ if
-  * \f[ \Vert v \Vert \leqslant p\,\vert x\vert. \f]
-  * For matrices, the comparison is done on all columns.
-  *
-  * \sa isApprox(), isMuchSmallerThan(const MatrixBase<OtherDerived>&, RealScalar) const
-  */
-template<typename Derived>
-bool MatrixBase<Derived>::isMuchSmallerThan(
-  const typename NumTraits<Scalar>::Real& other,
-  typename NumTraits<Scalar>::Real prec
-) const
-{
-  return ei_fuzzy_selector<Derived>::isMuchSmallerThan(derived(), other, prec);
-}
-
-/** \returns \c true if the norm of \c *this is much smaller than the norm of \a other,
-  * within the precision determined by \a prec.
-  *
-  * \note The fuzzy compares are done multiplicatively. A vector \f$ v \f$ is
-  * considered to be much smaller than a vector \f$ w \f$ within precision \f$ p \f$ if
-  * \f[ \Vert v \Vert \leqslant p\,\Vert w\Vert. \f]
-  * For matrices, the comparison is done on all columns.
-  *
-  * \sa isApprox(), isMuchSmallerThan(const RealScalar&, RealScalar) const
-  */
-template<typename Derived>
-template<typename OtherDerived>
-bool MatrixBase<Derived>::isMuchSmallerThan(
-  const MatrixBase<OtherDerived>& other,
-  typename NumTraits<Scalar>::Real prec
-) const
-{
-  return ei_fuzzy_selector<Derived,OtherDerived>::isMuchSmallerThan(derived(), other.derived(), prec);
-}
-
-
-template<typename Derived, typename OtherDerived>
-struct ei_fuzzy_selector<Derived,OtherDerived,true>
-{
-  typedef typename Derived::RealScalar RealScalar;
-  static bool isApprox(const Derived& self, const OtherDerived& other, RealScalar prec)
-  {
-    EIGEN_STATIC_ASSERT_SAME_VECTOR_SIZE(Derived,OtherDerived)
-    ei_assert(self.size() == other.size());
-    return((self - other).squaredNorm() <= std::min(self.squaredNorm(), other.squaredNorm()) * prec * prec);
-  }
-  static bool isMuchSmallerThan(const Derived& self, const RealScalar& other, RealScalar prec)
-  {
-    return(self.squaredNorm() <= ei_abs2(other * prec));
-  }
-  static bool isMuchSmallerThan(const Derived& self, const OtherDerived& other, RealScalar prec)
-  {
-    EIGEN_STATIC_ASSERT_SAME_VECTOR_SIZE(Derived,OtherDerived)
-    ei_assert(self.size() == other.size());
-    return(self.squaredNorm() <= other.squaredNorm() * prec * prec);
-  }
-};
-
-template<typename Derived, typename OtherDerived>
-struct ei_fuzzy_selector<Derived,OtherDerived,false>
-{
-  typedef typename Derived::RealScalar RealScalar;
-  static bool isApprox(const Derived& self, const OtherDerived& other, RealScalar prec)
-  {
-    EIGEN_STATIC_ASSERT_SAME_MATRIX_SIZE(Derived,OtherDerived)
-    ei_assert(self.rows() == other.rows() && self.cols() == other.cols());
-    typename Derived::Nested nested(self);
-    typename OtherDerived::Nested otherNested(other);
-    for(int i = 0; i < self.cols(); ++i)
-      if((nested.col(i) - otherNested.col(i)).squaredNorm()
-          > std::min(nested.col(i).squaredNorm(), otherNested.col(i).squaredNorm()) * prec * prec)
-        return false;
-    return true;
-  }
-  static bool isMuchSmallerThan(const Derived& self, const RealScalar& other, RealScalar prec)
-  {
-    typename Derived::Nested nested(self);
-    for(int i = 0; i < self.cols(); ++i)
-      if(nested.col(i).squaredNorm() > ei_abs2(other * prec))
-        return false;
-    return true;
-  }
-  static bool isMuchSmallerThan(const Derived& self, const OtherDerived& other, RealScalar prec)
-  {
-    EIGEN_STATIC_ASSERT_SAME_MATRIX_SIZE(Derived,OtherDerived)
-    ei_assert(self.rows() == other.rows() && self.cols() == other.cols());
-    typename Derived::Nested nested(self);
-    typename OtherDerived::Nested otherNested(other);
-    for(int i = 0; i < self.cols(); ++i)
-      if(nested.col(i).squaredNorm() > otherNested.col(i).squaredNorm() * prec * prec)
-        return false;
-    return true;
-  }
-};
-
-#endif
+} // end namespace Eigen
 
 #endif // EIGEN_FUZZY_H

Eigen/src/Core/GeneralProduct.h

@@ -0,0 +1,455 @@
+// 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>
+//
+// This Source Code Form is subject to the terms of the Mozilla
+// Public License v. 2.0. If a copy of the MPL was not distributed
+// with this file, You can obtain one at http://mozilla.org/MPL/2.0/.
+
+#ifndef EIGEN_GENERAL_PRODUCT_H
+#define EIGEN_GENERAL_PRODUCT_H
+
+namespace Eigen {
+
+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 {
+    #ifndef EIGEN_CUDA_ARCH
+    is_large = MaxSize == Dynamic ||
+               Size >= EIGEN_CACHEFRIENDLY_PRODUCT_THRESHOLD ||
+               (Size==Dynamic && MaxSize>=EIGEN_CACHEFRIENDLY_PRODUCT_THRESHOLD),
+    #else
+    is_large = 0,
+    #endif
+    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 = traits<_Lhs>::MaxRowsAtCompileTime,
+    Rows    = traits<_Lhs>::RowsAtCompileTime,
+    MaxCols = traits<_Rhs>::MaxColsAtCompileTime,
+    Cols    = traits<_Rhs>::ColsAtCompileTime,
+    MaxDepth = EIGEN_SIZE_MIN_PREFER_FIXED(traits<_Lhs>::MaxColsAtCompileTime,
+                                           traits<_Rhs>::MaxRowsAtCompileTime),
+    Depth = EIGEN_SIZE_MIN_PREFER_FIXED(traits<_Lhs>::ColsAtCompileTime,
+                                        traits<_Rhs>::RowsAtCompileTime)
+  };
+
+  // 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,
+    ret = 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 M>         struct product_type_selector<M, 1, 1>            { enum { ret = LazyCoeffBasedProductMode }; };
+template<int N>         struct product_type_selector<1, N, 1>            { enum { ret = LazyCoeffBasedProductMode }; };
+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 = CoeffBasedProductMode }; };
+template<>              struct product_type_selector<Small,Large,Small>  { enum { ret = CoeffBasedProductMode }; };
+template<>              struct product_type_selector<Large,Large,Small>  { enum { ret = GemmProduct }; };
+
+} // end namespace internal
+
+/***********************************************************************
+*  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);
+
+/***********************************************************************
+*  Implementation of Outer Vector Vector Product
+***********************************************************************/
+
+/***********************************************************************
+*  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<int Side, int StorageOrder, bool BlasCompatible>
+struct gemv_dense_selector;
+
+} // end namespace internal
+
+namespace internal {
+
+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>
+{
+  enum {
+    ForceAlignment  = internal::packet_traits<Scalar>::Vectorizable,
+    PacketSize      = internal::packet_traits<Scalar>::size
+  };
+  #if EIGEN_MAX_STATIC_ALIGN_BYTES!=0
+  internal::plain_array<Scalar,EIGEN_SIZE_MIN_PREFER_FIXED(Size,MaxSize),0,EIGEN_PLAIN_ENUM_MIN(AlignedMax,PacketSize)> 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.
+  internal::plain_array<Scalar,EIGEN_SIZE_MIN_PREFER_FIXED(Size,MaxSize)+(ForceAlignment?EIGEN_MAX_ALIGN_BYTES:0),0> m_data;
+  EIGEN_STRONG_INLINE Scalar* data() {
+    return ForceAlignment
+            ? reinterpret_cast<Scalar*>((internal::UIntPtr(m_data.array) & ~(std::size_t(EIGEN_MAX_ALIGN_BYTES-1))) + EIGEN_MAX_ALIGN_BYTES)
+            : m_data.array;
+  }
+  #endif
+};
+
+// The vector is on the left => transposition
+template<int StorageOrder, bool BlasCompatible>
+struct gemv_dense_selector<OnTheLeft,StorageOrder,BlasCompatible>
+{
+  template<typename Lhs, typename Rhs, typename Dest>
+  static void run(const Lhs &lhs, const Rhs &rhs, Dest& dest, const typename Dest::Scalar& alpha)
+  {
+    Transpose<Dest> destT(dest);
+    enum { OtherStorageOrder = StorageOrder == RowMajor ? ColMajor : RowMajor };
+    gemv_dense_selector<OnTheRight,OtherStorageOrder,BlasCompatible>
+      ::run(rhs.transpose(), lhs.transpose(), destT, alpha);
+  }
+};
+
+template<> struct gemv_dense_selector<OnTheRight,ColMajor,true>
+{
+  template<typename Lhs, typename Rhs, typename Dest>
+  static inline void run(const Lhs &lhs, const Rhs &rhs, Dest& dest, const typename Dest::Scalar& alpha)
+  {
+    typedef typename Lhs::Scalar   LhsScalar;
+    typedef typename Rhs::Scalar   RhsScalar;
+    typedef typename Dest::Scalar  ResScalar;
+    typedef typename Dest::RealScalar  RealScalar;
+    
+    typedef internal::blas_traits<Lhs> LhsBlasTraits;
+    typedef typename LhsBlasTraits::DirectLinearAccessType ActualLhsType;
+    typedef internal::blas_traits<Rhs> RhsBlasTraits;
+    typedef typename RhsBlasTraits::DirectLinearAccessType ActualRhsType;
+  
+    typedef Map<Matrix<ResScalar,Dynamic,1>, EIGEN_PLAIN_ENUM_MIN(AlignedMax,internal::packet_traits<ResScalar>::size)> MappedDest;
+
+    ActualLhsType actualLhs = LhsBlasTraits::extract(lhs);
+    ActualRhsType actualRhs = RhsBlasTraits::extract(rhs);
+
+    ResScalar actualAlpha = alpha * LhsBlasTraits::extractScalarFactor(lhs)
+                                  * RhsBlasTraits::extractScalarFactor(rhs);
+
+    // make sure Dest is a compile-time vector type (bug 1166)
+    typedef typename conditional<Dest::IsVectorAtCompileTime, Dest, typename Dest::ColXpr>::type ActualDest;
+
+    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 = (ActualDest::InnerStrideAtCompileTime==1),
+      ComplexByReal = (NumTraits<LhsScalar>::IsComplex) && (!NumTraits<RhsScalar>::IsComplex),
+      MightCannotUseDest = (!EvalToDestAtCompileTime) || ComplexByReal
+    };
+
+    typedef const_blas_data_mapper<LhsScalar,Index,ColMajor> LhsMapper;
+    typedef const_blas_data_mapper<RhsScalar,Index,RowMajor> RhsMapper;
+    RhsScalar compatibleAlpha = get_factor<ResScalar,RhsScalar>::run(actualAlpha);
+
+    if(!MightCannotUseDest)
+    {
+      // shortcut if we are sure to be able to use dest directly,
+      // this ease the compiler to generate cleaner and more optimzized code for most common cases
+      general_matrix_vector_product
+          <Index,LhsScalar,LhsMapper,ColMajor,LhsBlasTraits::NeedToConjugate,RhsScalar,RhsMapper,RhsBlasTraits::NeedToConjugate>::run(
+          actualLhs.rows(), actualLhs.cols(),
+          LhsMapper(actualLhs.data(), actualLhs.outerStride()),
+          RhsMapper(actualRhs.data(), actualRhs.innerStride()),
+          dest.data(), 1,
+          compatibleAlpha);
+    }
+    else
+    {
+      gemv_static_vector_if<ResScalar,ActualDest::SizeAtCompileTime,ActualDest::MaxSizeAtCompileTime,MightCannotUseDest> static_dest;
+
+      const bool alphaIsCompatible = (!ComplexByReal) || (numext::imag(actualAlpha)==RealScalar(0));
+      const bool evalToDest = EvalToDestAtCompileTime && alphaIsCompatible;
+
+      ei_declare_aligned_stack_constructed_variable(ResScalar,actualDestPtr,dest.size(),
+                                                    evalToDest ? dest.data() : static_dest.data());
+
+      if(!evalToDest)
+      {
+        #ifdef EIGEN_DENSE_STORAGE_CTOR_PLUGIN
+        Index 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,LhsMapper,ColMajor,LhsBlasTraits::NeedToConjugate,RhsScalar,RhsMapper,RhsBlasTraits::NeedToConjugate>::run(
+          actualLhs.rows(), actualLhs.cols(),
+          LhsMapper(actualLhs.data(), actualLhs.outerStride()),
+          RhsMapper(actualRhs.data(), actualRhs.innerStride()),
+          actualDestPtr, 1,
+          compatibleAlpha);
+
+      if (!evalToDest)
+      {
+        if(!alphaIsCompatible)
+          dest.matrix() += actualAlpha * MappedDest(actualDestPtr, dest.size());
+        else
+          dest = MappedDest(actualDestPtr, dest.size());
+      }
+    }
+  }
+};
+
+template<> struct gemv_dense_selector<OnTheRight,RowMajor,true>
+{
+  template<typename Lhs, typename Rhs, typename Dest>
+  static void run(const Lhs &lhs, const Rhs &rhs, Dest& dest, const typename Dest::Scalar& alpha)
+  {
+    typedef typename Lhs::Scalar   LhsScalar;
+    typedef typename Rhs::Scalar   RhsScalar;
+    typedef typename Dest::Scalar  ResScalar;
+    
+    typedef internal::blas_traits<Lhs> LhsBlasTraits;
+    typedef typename LhsBlasTraits::DirectLinearAccessType ActualLhsType;
+    typedef internal::blas_traits<Rhs> RhsBlasTraits;
+    typedef typename RhsBlasTraits::DirectLinearAccessType ActualRhsType;
+    typedef typename internal::remove_all<ActualRhsType>::type ActualRhsTypeCleaned;
+
+    typename add_const<ActualLhsType>::type actualLhs = LhsBlasTraits::extract(lhs);
+    typename add_const<ActualRhsType>::type actualRhs = RhsBlasTraits::extract(rhs);
+
+    ResScalar actualAlpha = alpha * LhsBlasTraits::extractScalarFactor(lhs)
+                                  * RhsBlasTraits::extractScalarFactor(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 = ActualRhsTypeCleaned::InnerStrideAtCompileTime==1
+    };
+
+    gemv_static_vector_if<RhsScalar,ActualRhsTypeCleaned::SizeAtCompileTime,ActualRhsTypeCleaned::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
+      Index size = actualRhs.size();
+      EIGEN_DENSE_STORAGE_CTOR_PLUGIN
+      #endif
+      Map<typename ActualRhsTypeCleaned::PlainObject>(actualRhsPtr, actualRhs.size()) = actualRhs;
+    }
+
+    typedef const_blas_data_mapper<LhsScalar,Index,RowMajor> LhsMapper;
+    typedef const_blas_data_mapper<RhsScalar,Index,ColMajor> RhsMapper;
+    general_matrix_vector_product
+        <Index,LhsScalar,LhsMapper,RowMajor,LhsBlasTraits::NeedToConjugate,RhsScalar,RhsMapper,RhsBlasTraits::NeedToConjugate>::run(
+        actualLhs.rows(), actualLhs.cols(),
+        LhsMapper(actualLhs.data(), actualLhs.outerStride()),
+        RhsMapper(actualRhsPtr, 1),
+        dest.data(), dest.col(0).innerStride(), //NOTE  if dest is not a vector at compile-time, then dest.innerStride() might be wrong. (bug 1166)
+        actualAlpha);
+  }
+};
+
+template<> struct gemv_dense_selector<OnTheRight,ColMajor,false>
+{
+  template<typename Lhs, typename Rhs, typename Dest>
+  static void run(const Lhs &lhs, const Rhs &rhs, Dest& dest, const typename Dest::Scalar& alpha)
+  {
+    EIGEN_STATIC_ASSERT((!nested_eval<Lhs,1>::Evaluate),EIGEN_INTERNAL_COMPILATION_ERROR_OR_YOU_MADE_A_PROGRAMMING_MISTAKE);
+    // TODO if rhs is large enough it might be beneficial to make sure that dest is sequentially stored in memory, otherwise use a temp
+    typename nested_eval<Rhs,1>::type actual_rhs(rhs);
+    const Index size = rhs.rows();
+    for(Index k=0; k<size; ++k)
+      dest += (alpha*actual_rhs.coeff(k)) * lhs.col(k);
+  }
+};
+
+template<> struct gemv_dense_selector<OnTheRight,RowMajor,false>
+{
+  template<typename Lhs, typename Rhs, typename Dest>
+  static void run(const Lhs &lhs, const Rhs &rhs, Dest& dest, const typename Dest::Scalar& alpha)
+  {
+    EIGEN_STATIC_ASSERT((!nested_eval<Lhs,1>::Evaluate),EIGEN_INTERNAL_COMPILATION_ERROR_OR_YOU_MADE_A_PROGRAMMING_MISTAKE);
+    typename nested_eval<Rhs,Lhs::RowsAtCompileTime>::type actual_rhs(rhs);
+    const Index rows = dest.rows();
+    for(Index i=0; i<rows; ++i)
+      dest.coeffRef(i) += alpha * (lhs.row(i).cwiseProduct(actual_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 Product<Derived, OtherDerived>
+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 Product<Derived, OtherDerived>(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 Product<Derived,OtherDerived,LazyProduct>
+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 Product<Derived,OtherDerived,LazyProduct>(derived(), other.derived());
+}
+
+} // end namespace Eigen
+
+#endif // EIGEN_PRODUCT_H

Eigen/src/Core/GenericPacketMath.h

@@ -1,31 +1,20 @@
 // This file is part of Eigen, a lightweight C++ template library
-// for linear algebra. Eigen itself is part of the KDE project.
+// for linear algebra.
 //
-// Copyright (C) 2008 Gael Guennebaud <g.gael@free.fr>
+// Copyright (C) 2008 Gael Guennebaud <gael.guennebaud@inria.fr>
 // Copyright (C) 2006-2008 Benoit Jacob <jacob.benoit.1@gmail.com>
 //
-// 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/>.
+// This Source Code Form is subject to the terms of the Mozilla
+// Public License v. 2.0. If a copy of the MPL was not distributed
+// with this file, You can obtain one at http://mozilla.org/MPL/2.0/.
 
 #ifndef EIGEN_GENERIC_PACKET_MATH_H
 #define EIGEN_GENERIC_PACKET_MATH_H
 
+namespace Eigen {
+
+namespace internal {
+
 /** \internal
   * \file GenericPacketMath.h
   *
@@ -34,117 +23,568 @@
   * of generic vectorized code.
   */
 
+#ifndef EIGEN_DEBUG_ALIGNED_LOAD
+#define EIGEN_DEBUG_ALIGNED_LOAD
+#endif
+
+#ifndef EIGEN_DEBUG_UNALIGNED_LOAD
+#define EIGEN_DEBUG_UNALIGNED_LOAD
+#endif
+
+#ifndef EIGEN_DEBUG_ALIGNED_STORE
+#define EIGEN_DEBUG_ALIGNED_STORE
+#endif
+
+#ifndef EIGEN_DEBUG_UNALIGNED_STORE
+#define EIGEN_DEBUG_UNALIGNED_STORE
+#endif
+
+struct default_packet_traits
+{
+  enum {
+    HasHalfPacket = 0,
+
+    HasAdd    = 1,
+    HasSub    = 1,
+    HasMul    = 1,
+    HasNegate = 1,
+    HasAbs    = 1,
+    HasArg    = 0,
+    HasAbs2   = 1,
+    HasMin    = 1,
+    HasMax    = 1,
+    HasConj   = 1,
+    HasSetLinear = 1,
+    HasBlend  = 0,
+
+    HasDiv    = 0,
+    HasSqrt   = 0,
+    HasRsqrt  = 0,
+    HasExp    = 0,
+    HasLog    = 0,
+    HasLog1p  = 0,
+    HasLog10  = 0,
+    HasPow    = 0,
+
+    HasSin    = 0,
+    HasCos    = 0,
+    HasTan    = 0,
+    HasASin   = 0,
+    HasACos   = 0,
+    HasATan   = 0,
+    HasSinh   = 0,
+    HasCosh   = 0,
+    HasTanh   = 0,
+    HasLGamma = 0,
+    HasDiGamma = 0,
+    HasZeta = 0,
+    HasPolygamma = 0,
+    HasErf = 0,
+    HasErfc = 0,
+    HasIGamma = 0,
+    HasIGammac = 0,
+    HasBetaInc = 0,
+
+    HasRound  = 0,
+    HasFloor  = 0,
+    HasCeil   = 0,
+
+    HasSign   = 0
+  };
+};
+
+template<typename T> struct packet_traits : default_packet_traits
+{
+  typedef T type;
+  typedef T half;
+  enum {
+    Vectorizable = 0,
+    size = 1,
+    AlignedOnScalar = 0,
+    HasHalfPacket = 0
+  };
+  enum {
+    HasAdd    = 0,
+    HasSub    = 0,
+    HasMul    = 0,
+    HasNegate = 0,
+    HasAbs    = 0,
+    HasAbs2   = 0,
+    HasMin    = 0,
+    HasMax    = 0,
+    HasConj   = 0,
+    HasSetLinear = 0
+  };
+};
+
+template<typename T> struct packet_traits<const T> : packet_traits<T> { };
+
+template <typename Src, typename Tgt> struct type_casting_traits {
+  enum {
+    VectorizedCast = 0,
+    SrcCoeffRatio = 1,
+    TgtCoeffRatio = 1
+  };
+};
+
+
+/** \internal \returns static_cast<TgtType>(a) (coeff-wise) */
+template <typename SrcPacket, typename TgtPacket>
+EIGEN_DEVICE_FUNC inline TgtPacket
+pcast(const SrcPacket& a) {
+  return static_cast<TgtPacket>(a);
+}
+template <typename SrcPacket, typename TgtPacket>
+EIGEN_DEVICE_FUNC inline TgtPacket
+pcast(const SrcPacket& a, const SrcPacket& /*b*/) {
+  return static_cast<TgtPacket>(a);
+}
+
+template <typename SrcPacket, typename TgtPacket>
+EIGEN_DEVICE_FUNC inline TgtPacket
+pcast(const SrcPacket& a, const SrcPacket& /*b*/, const SrcPacket& /*c*/, const SrcPacket& /*d*/) {
+  return static_cast<TgtPacket>(a);
+}
+
 /** \internal \returns a + b (coeff-wise) */
-template<typename Packet> inline Packet
-ei_padd(const Packet& a,
+template<typename Packet> EIGEN_DEVICE_FUNC inline Packet
+padd(const Packet& a,
         const Packet& b) { return a+b; }
 
 /** \internal \returns a - b (coeff-wise) */
-template<typename Packet> inline Packet
-ei_psub(const Packet& a,
+template<typename Packet> EIGEN_DEVICE_FUNC inline Packet
+psub(const Packet& a,
         const Packet& b) { return a-b; }
 
+/** \internal \returns -a (coeff-wise) */
+template<typename Packet> EIGEN_DEVICE_FUNC inline Packet
+pnegate(const Packet& a) { return -a; }
+
+/** \internal \returns conj(a) (coeff-wise) */
+
+template<typename Packet> EIGEN_DEVICE_FUNC inline Packet
+pconj(const Packet& a) { return numext::conj(a); }
+
 /** \internal \returns a * b (coeff-wise) */
-template<typename Packet> inline Packet
-ei_pmul(const Packet& a,
+template<typename Packet> EIGEN_DEVICE_FUNC inline Packet
+pmul(const Packet& a,
         const Packet& b) { return a*b; }
 
 /** \internal \returns a / b (coeff-wise) */
-template<typename Packet> inline Packet
-ei_pdiv(const Packet& a,
+template<typename Packet> EIGEN_DEVICE_FUNC inline Packet
+pdiv(const Packet& a,
         const Packet& b) { return a/b; }
 
 /** \internal \returns the min of \a a and \a b  (coeff-wise) */
-template<typename Packet> inline Packet
-ei_pmin(const Packet& a,
-        const Packet& b) { return std::min(a, b); }
+template<typename Packet> EIGEN_DEVICE_FUNC inline Packet
+pmin(const Packet& a,
+        const Packet& b) { return numext::mini(a, b); }
 
 /** \internal \returns the max of \a a and \a b  (coeff-wise) */
-template<typename Packet> inline Packet
-ei_pmax(const Packet& a,
-        const Packet& b) { return std::max(a, b); }
+template<typename Packet> EIGEN_DEVICE_FUNC inline Packet
+pmax(const Packet& a,
+        const Packet& b) { return numext::maxi(a, b); }
+
+/** \internal \returns the absolute value of \a a */
+template<typename Packet> EIGEN_DEVICE_FUNC inline Packet
+pabs(const Packet& a) { using std::abs; return abs(a); }
+
+/** \internal \returns the phase angle of \a a */
+template<typename Packet> EIGEN_DEVICE_FUNC inline Packet
+parg(const Packet& a) { using numext::arg; return arg(a); }
+
+/** \internal \returns the bitwise and of \a a and \a b */
+template<typename Packet> EIGEN_DEVICE_FUNC inline Packet
+pand(const Packet& a, const Packet& b) { return a & b; }
+
+/** \internal \returns the bitwise or of \a a and \a b */
+template<typename Packet> EIGEN_DEVICE_FUNC inline Packet
+por(const Packet& a, const Packet& b) { return a | b; }
+
+/** \internal \returns the bitwise xor of \a a and \a b */
+template<typename Packet> EIGEN_DEVICE_FUNC inline Packet
+pxor(const Packet& a, const Packet& b) { return a ^ b; }
+
+/** \internal \returns the bitwise andnot of \a a and \a b */
+template<typename Packet> EIGEN_DEVICE_FUNC inline Packet
+pandnot(const Packet& a, const Packet& b) { return a & (!b); }
 
 /** \internal \returns a packet version of \a *from, from must be 16 bytes aligned */
-template<typename Scalar> inline typename ei_packet_traits<Scalar>::type
-ei_pload(const Scalar* from) { return *from; }
+template<typename Packet> EIGEN_DEVICE_FUNC inline Packet
+pload(const typename unpacket_traits<Packet>::type* from) { return *from; }
 
 /** \internal \returns a packet version of \a *from, (un-aligned load) */
-template<typename Scalar> inline typename ei_packet_traits<Scalar>::type
-ei_ploadu(const Scalar* from) { return *from; }
+template<typename Packet> EIGEN_DEVICE_FUNC inline Packet
+ploadu(const typename unpacket_traits<Packet>::type* from) { return *from; }
 
 /** \internal \returns a packet with constant coefficients \a a, e.g.: (a,a,a,a) */
-template<typename Scalar> inline typename ei_packet_traits<Scalar>::type
-ei_pset1(const Scalar& a) { return a; }
+template<typename Packet> EIGEN_DEVICE_FUNC inline Packet
+pset1(const typename unpacket_traits<Packet>::type& a) { return a; }
+
+/** \internal \returns a packet with constant coefficients \a a[0], e.g.: (a[0],a[0],a[0],a[0]) */
+template<typename Packet> EIGEN_DEVICE_FUNC inline Packet
+pload1(const typename unpacket_traits<Packet>::type  *a) { return pset1<Packet>(*a); }
+
+/** \internal \returns a packet with elements of \a *from duplicated.
+  * For instance, for a packet of 8 elements, 4 scalars will be read from \a *from and
+  * duplicated to form: {from[0],from[0],from[1],from[1],from[2],from[2],from[3],from[3]}
+  * Currently, this function is only used for scalar * complex products.
+  */
+template<typename Packet> EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE Packet
+ploaddup(const typename unpacket_traits<Packet>::type* from) { return *from; }
+
+/** \internal \returns a packet with elements of \a *from quadrupled.
+  * For instance, for a packet of 8 elements, 2 scalars will be read from \a *from and
+  * replicated to form: {from[0],from[0],from[0],from[0],from[1],from[1],from[1],from[1]}
+  * Currently, this function is only used in matrix products.
+  * For packet-size smaller or equal to 4, this function is equivalent to pload1 
+  */
+template<typename Packet> EIGEN_DEVICE_FUNC inline Packet
+ploadquad(const typename unpacket_traits<Packet>::type* from)
+{ return pload1<Packet>(from); }
+
+/** \internal equivalent to
+  * \code
+  * a0 = pload1(a+0);
+  * a1 = pload1(a+1);
+  * a2 = pload1(a+2);
+  * a3 = pload1(a+3);
+  * \endcode
+  * \sa pset1, pload1, ploaddup, pbroadcast2
+  */
+template<typename Packet> EIGEN_DEVICE_FUNC
+inline void pbroadcast4(const typename unpacket_traits<Packet>::type *a,
+                        Packet& a0, Packet& a1, Packet& a2, Packet& a3)
+{
+  a0 = pload1<Packet>(a+0);
+  a1 = pload1<Packet>(a+1);
+  a2 = pload1<Packet>(a+2);
+  a3 = pload1<Packet>(a+3);
+}
+
+/** \internal equivalent to
+  * \code
+  * a0 = pload1(a+0);
+  * a1 = pload1(a+1);
+  * \endcode
+  * \sa pset1, pload1, ploaddup, pbroadcast4
+  */
+template<typename Packet> EIGEN_DEVICE_FUNC
+inline void pbroadcast2(const typename unpacket_traits<Packet>::type *a,
+                        Packet& a0, Packet& a1)
+{
+  a0 = pload1<Packet>(a+0);
+  a1 = pload1<Packet>(a+1);
+}
+
+/** \internal \brief Returns a packet with coefficients (a,a+1,...,a+packet_size-1). */
+template<typename Packet> EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE Packet
+plset(const typename unpacket_traits<Packet>::type& a) { return a; }
 
 /** \internal copy the packet \a from to \a *to, \a to must be 16 bytes aligned */
-template<typename Scalar, typename Packet> inline void ei_pstore(Scalar* to, const Packet& from)
+template<typename Scalar, typename Packet> EIGEN_DEVICE_FUNC inline void pstore(Scalar* to, const Packet& from)
 { (*to) = from; }
 
 /** \internal copy the packet \a from to \a *to, (un-aligned store) */
-template<typename Scalar, typename Packet> inline void ei_pstoreu(Scalar* to, const Packet& from)
-{ (*to) = from; }
+template<typename Scalar, typename Packet> EIGEN_DEVICE_FUNC inline void pstoreu(Scalar* to, const Packet& from)
+{  (*to) = from; }
+
+ template<typename Scalar, typename Packet> EIGEN_DEVICE_FUNC inline Packet pgather(const Scalar* from, Index /*stride*/)
+ { return ploadu<Packet>(from); }
+
+ template<typename Scalar, typename Packet> EIGEN_DEVICE_FUNC inline void pscatter(Scalar* to, const Packet& from, Index /*stride*/)
+ { pstore(to, from); }
+
+/** \internal tries to do cache prefetching of \a addr */
+template<typename Scalar> EIGEN_DEVICE_FUNC inline void prefetch(const Scalar* addr)
+{
+#ifdef __CUDA_ARCH__
+#if defined(__LP64__)
+  // 64-bit pointer operand constraint for inlined asm
+  asm(" prefetch.L1 [ %1 ];" : "=l"(addr) : "l"(addr));
+#else
+  // 32-bit pointer operand constraint for inlined asm
+  asm(" prefetch.L1 [ %1 ];" : "=r"(addr) : "r"(addr));
+#endif
+#elif (!EIGEN_COMP_MSVC) && (EIGEN_COMP_GNUC || EIGEN_COMP_CLANG || EIGEN_COMP_ICC)
+  __builtin_prefetch(addr);
+#endif
+}
 
 /** \internal \returns the first element of a packet */
-template<typename Packet> inline typename ei_unpacket_traits<Packet>::type ei_pfirst(const Packet& a)
+template<typename Packet> EIGEN_DEVICE_FUNC inline typename unpacket_traits<Packet>::type pfirst(const Packet& a)
 { return a; }
 
 /** \internal \returns a packet where the element i contains the sum of the packet of \a vec[i] */
-template<typename Packet> inline Packet
-ei_preduxp(const Packet* vecs) { return vecs[0]; }
+template<typename Packet> EIGEN_DEVICE_FUNC inline Packet
+preduxp(const Packet* vecs) { return vecs[0]; }
 
 /** \internal \returns the sum of the elements of \a a*/
-template<typename Packet> inline typename ei_unpacket_traits<Packet>::type ei_predux(const Packet& a)
+template<typename Packet> EIGEN_DEVICE_FUNC inline typename unpacket_traits<Packet>::type predux(const Packet& a)
 { return a; }
 
+/** \internal \returns the sum of the elements of \a a by block of 4 elements.
+  * For a packet {a0, a1, a2, a3, a4, a5, a6, a7}, it returns a half packet {a0+a4, a1+a5, a2+a6, a3+a7}
+  * For packet-size smaller or equal to 4, this boils down to a noop.
+  */
+template<typename Packet> EIGEN_DEVICE_FUNC inline
+typename conditional<(unpacket_traits<Packet>::size%8)==0,typename unpacket_traits<Packet>::half,Packet>::type
+predux_downto4(const Packet& a)
+{ return a; }
+
+/** \internal \returns the product of the elements of \a a*/
+template<typename Packet> EIGEN_DEVICE_FUNC inline typename unpacket_traits<Packet>::type predux_mul(const Packet& a)
+{ return a; }
+
+/** \internal \returns the min of the elements of \a a*/
+template<typename Packet> EIGEN_DEVICE_FUNC inline typename unpacket_traits<Packet>::type predux_min(const Packet& a)
+{ return a; }
+
+/** \internal \returns the max of the elements of \a a*/
+template<typename Packet> EIGEN_DEVICE_FUNC inline typename unpacket_traits<Packet>::type predux_max(const Packet& a)
+{ return a; }
+
+/** \internal \returns the reversed elements of \a a*/
+template<typename Packet> EIGEN_DEVICE_FUNC inline Packet preverse(const Packet& a)
+{ return a; }
+
+/** \internal \returns \a a with real and imaginary part flipped (for complex type only) */
+template<typename Packet> EIGEN_DEVICE_FUNC inline Packet pcplxflip(const Packet& a)
+{
+  return Packet(a.imag(),a.real());
+}
+
+/**************************
+* Special math functions
+***************************/
+
+/** \internal \returns the sine of \a a (coeff-wise) */
+template<typename Packet> EIGEN_DECLARE_FUNCTION_ALLOWING_MULTIPLE_DEFINITIONS
+Packet psin(const Packet& a) { using std::sin; return sin(a); }
+
+/** \internal \returns the cosine of \a a (coeff-wise) */
+template<typename Packet> EIGEN_DECLARE_FUNCTION_ALLOWING_MULTIPLE_DEFINITIONS
+Packet pcos(const Packet& a) { using std::cos; return cos(a); }
+
+/** \internal \returns the tan of \a a (coeff-wise) */
+template<typename Packet> EIGEN_DECLARE_FUNCTION_ALLOWING_MULTIPLE_DEFINITIONS
+Packet ptan(const Packet& a) { using std::tan; return tan(a); }
+
+/** \internal \returns the arc sine of \a a (coeff-wise) */
+template<typename Packet> EIGEN_DECLARE_FUNCTION_ALLOWING_MULTIPLE_DEFINITIONS
+Packet pasin(const Packet& a) { using std::asin; return asin(a); }
+
+/** \internal \returns the arc cosine of \a a (coeff-wise) */
+template<typename Packet> EIGEN_DECLARE_FUNCTION_ALLOWING_MULTIPLE_DEFINITIONS
+Packet pacos(const Packet& a) { using std::acos; return acos(a); }
+
+/** \internal \returns the arc tangent of \a a (coeff-wise) */
+template<typename Packet> EIGEN_DECLARE_FUNCTION_ALLOWING_MULTIPLE_DEFINITIONS
+Packet patan(const Packet& a) { using std::atan; return atan(a); }
+
+/** \internal \returns the hyperbolic sine of \a a (coeff-wise) */
+template<typename Packet> EIGEN_DECLARE_FUNCTION_ALLOWING_MULTIPLE_DEFINITIONS
+Packet psinh(const Packet& a) { using std::sinh; return sinh(a); }
+
+/** \internal \returns the hyperbolic cosine of \a a (coeff-wise) */
+template<typename Packet> EIGEN_DECLARE_FUNCTION_ALLOWING_MULTIPLE_DEFINITIONS
+Packet pcosh(const Packet& a) { using std::cosh; return cosh(a); }
+
+/** \internal \returns the hyperbolic tan of \a a (coeff-wise) */
+template<typename Packet> EIGEN_DECLARE_FUNCTION_ALLOWING_MULTIPLE_DEFINITIONS
+Packet ptanh(const Packet& a) { using std::tanh; return tanh(a); }
+
+/** \internal \returns the exp of \a a (coeff-wise) */
+template<typename Packet> EIGEN_DECLARE_FUNCTION_ALLOWING_MULTIPLE_DEFINITIONS
+Packet pexp(const Packet& a) { using std::exp; return exp(a); }
+
+/** \internal \returns the log of \a a (coeff-wise) */
+template<typename Packet> EIGEN_DECLARE_FUNCTION_ALLOWING_MULTIPLE_DEFINITIONS
+Packet plog(const Packet& a) { using std::log; return log(a); }
+
+/** \internal \returns the log1p of \a a (coeff-wise) */
+template<typename Packet> EIGEN_DECLARE_FUNCTION_ALLOWING_MULTIPLE_DEFINITIONS
+Packet plog1p(const Packet& a) { return numext::log1p(a); }
+
+/** \internal \returns the log10 of \a a (coeff-wise) */
+template<typename Packet> EIGEN_DECLARE_FUNCTION_ALLOWING_MULTIPLE_DEFINITIONS
+Packet plog10(const Packet& a) { using std::log10; return log10(a); }
+
+/** \internal \returns the square-root of \a a (coeff-wise) */
+template<typename Packet> EIGEN_DECLARE_FUNCTION_ALLOWING_MULTIPLE_DEFINITIONS
+Packet psqrt(const Packet& a) { using std::sqrt; return sqrt(a); }
+
+/** \internal \returns the reciprocal square-root of \a a (coeff-wise) */
+template<typename Packet> EIGEN_DECLARE_FUNCTION_ALLOWING_MULTIPLE_DEFINITIONS
+Packet prsqrt(const Packet& a) {
+  return pdiv(pset1<Packet>(1), psqrt(a));
+}
+
+/** \internal \returns the rounded value of \a a (coeff-wise) */
+template<typename Packet> EIGEN_DECLARE_FUNCTION_ALLOWING_MULTIPLE_DEFINITIONS
+Packet pround(const Packet& a) { using numext::round; return round(a); }
+
+/** \internal \returns the floor of \a a (coeff-wise) */
+template<typename Packet> EIGEN_DECLARE_FUNCTION_ALLOWING_MULTIPLE_DEFINITIONS
+Packet pfloor(const Packet& a) { using numext::floor; return floor(a); }
+
+/** \internal \returns the ceil of \a a (coeff-wise) */
+template<typename Packet> EIGEN_DECLARE_FUNCTION_ALLOWING_MULTIPLE_DEFINITIONS
+Packet pceil(const Packet& a) { using numext::ceil; return ceil(a); }
 
 /***************************************************************************
 * The following functions might not have to be overwritten for vectorized types
 ***************************************************************************/
 
+/** \internal copy a packet with constant coeficient \a a (e.g., [a,a,a,a]) to \a *to. \a to must be 16 bytes aligned */
+// NOTE: this function must really be templated on the packet type (think about different packet types for the same scalar type)
+template<typename Packet>
+inline void pstore1(typename unpacket_traits<Packet>::type* to, const typename unpacket_traits<Packet>::type& a)
+{
+  pstore(to, pset1<Packet>(a));
+}
+
 /** \internal \returns a * b + c (coeff-wise) */
-template<typename Packet> inline Packet
-ei_pmadd(const Packet&  a,
+template<typename Packet> EIGEN_DEVICE_FUNC inline Packet
+pmadd(const Packet&  a,
          const Packet&  b,
          const Packet&  c)
-{ return ei_padd(ei_pmul(a, b),c); }
+{ return padd(pmul(a, b),c); }
 
 /** \internal \returns a packet version of \a *from.
-  * \If LoadMode equals Aligned, \a from must be 16 bytes aligned */
-template<typename Scalar, int LoadMode>
-inline typename ei_packet_traits<Scalar>::type ei_ploadt(const Scalar* from)
+  * The pointer \a from must be aligned on a \a Alignment bytes boundary. */
+template<typename Packet, int Alignment>
+EIGEN_DEVICE_FUNC EIGEN_ALWAYS_INLINE Packet ploadt(const typename unpacket_traits<Packet>::type* from)
 {
-  if(LoadMode == Aligned)
-    return ei_pload(from);
+  if(Alignment >= unpacket_traits<Packet>::alignment)
+    return pload<Packet>(from);
   else
-    return ei_ploadu(from);
+    return ploadu<Packet>(from);
 }
 
 /** \internal copy the packet \a from to \a *to.
-  * If StoreMode equals Aligned, \a to must be 16 bytes aligned */
-template<typename Scalar, typename Packet, int LoadMode>
-inline void ei_pstoret(Scalar* to, const Packet& from)
+  * The pointer \a from must be aligned on a \a Alignment bytes boundary. */
+template<typename Scalar, typename Packet, int Alignment>
+EIGEN_DEVICE_FUNC EIGEN_ALWAYS_INLINE void pstoret(Scalar* to, const Packet& from)
 {
-  if(LoadMode == Aligned)
-    ei_pstore(to, from);
+  if(Alignment >= unpacket_traits<Packet>::alignment)
+    pstore(to, from);
   else
-    ei_pstoreu(to, from);
+    pstoreu(to, from);
 }
 
-/** \internal default implementation of ei_palign() allowing partial specialization */
+/** \internal \returns a packet version of \a *from.
+  * Unlike ploadt, ploadt_ro takes advantage of the read-only memory path on the
+  * hardware if available to speedup the loading of data that won't be modified
+  * by the current computation.
+  */
+template<typename Packet, int LoadMode>
+EIGEN_DEVICE_FUNC EIGEN_ALWAYS_INLINE Packet ploadt_ro(const typename unpacket_traits<Packet>::type* from)
+{
+  return ploadt<Packet, LoadMode>(from);
+}
+
+/** \internal default implementation of palign() allowing partial specialization */
 template<int Offset,typename PacketType>
-struct ei_palign_impl
+struct palign_impl
 {
   // by default data are aligned, so there is nothing to be done :)
-  inline static void run(PacketType&, const PacketType&) {}
+  static inline void run(PacketType&, const PacketType&) {}
 };
 
-/** \internal update \a first using the concatenation of the \a Offset last elements
-  * of \a first and packet_size minus \a Offset first elements of \a second */
+/** \internal update \a first using the concatenation of the packet_size minus \a Offset last elements
+  * of \a first and \a Offset first elements of \a second.
+  * 
+  * This function is currently only used to optimize matrix-vector products on unligned matrices.
+  * It takes 2 packets that represent a contiguous memory array, and returns a packet starting
+  * at the position \a Offset. For instance, for packets of 4 elements, we have:
+  *  Input:
+  *  - first = {f0,f1,f2,f3}
+  *  - second = {s0,s1,s2,s3}
+  * Output: 
+  *   - if Offset==0 then {f0,f1,f2,f3}
+  *   - if Offset==1 then {f1,f2,f3,s0}
+  *   - if Offset==2 then {f2,f3,s0,s1}
+  *   - if Offset==3 then {f3,s0,s1,s3}
+  */
 template<int Offset,typename PacketType>
-inline void ei_palign(PacketType& first, const PacketType& second)
+inline void palign(PacketType& first, const PacketType& second)
 {
-  ei_palign_impl<Offset,PacketType>::run(first,second);
+  palign_impl<Offset,PacketType>::run(first,second);
 }
 
-#endif // EIGEN_GENERIC_PACKET_MATH_H
+/***************************************************************************
+* Fast complex products (GCC generates a function call which is very slow)
+***************************************************************************/
 
+// Eigen+CUDA does not support complexes.
+#ifndef __CUDACC__
+
+template<> inline std::complex<float> pmul(const std::complex<float>& a, const std::complex<float>& b)
+{ return std::complex<float>(a.real()*b.real() - a.imag()*b.imag(), a.imag()*b.real() + a.real()*b.imag()); }
+
+template<> inline std::complex<double> pmul(const std::complex<double>& a, const std::complex<double>& b)
+{ return std::complex<double>(a.real()*b.real() - a.imag()*b.imag(), a.imag()*b.real() + a.real()*b.imag()); }
+
+#endif
+
+
+/***************************************************************************
+ * PacketBlock, that is a collection of N packets where the number of words
+ * in the packet is a multiple of N.
+***************************************************************************/
+template <typename Packet,int N=unpacket_traits<Packet>::size> struct PacketBlock {
+  Packet packet[N];
+};
+
+template<typename Packet> EIGEN_DEVICE_FUNC inline void
+ptranspose(PacketBlock<Packet,1>& /*kernel*/) {
+  // Nothing to do in the scalar case, i.e. a 1x1 matrix.
+}
+
+/***************************************************************************
+ * Selector, i.e. vector of N boolean values used to select (i.e. blend)
+ * words from 2 packets.
+***************************************************************************/
+template <size_t N> struct Selector {
+  bool select[N];
+};
+
+template<typename Packet> EIGEN_DEVICE_FUNC inline Packet
+pblend(const Selector<unpacket_traits<Packet>::size>& ifPacket, const Packet& thenPacket, const Packet& elsePacket) {
+  return ifPacket.select[0] ? thenPacket : elsePacket;
+}
+
+/** \internal \returns \a a with the first coefficient replaced by the scalar b */
+template<typename Packet> EIGEN_DEVICE_FUNC inline Packet
+pinsertfirst(const Packet& a, typename unpacket_traits<Packet>::type b)
+{
+  // Default implementation based on pblend.
+  // It must be specialized for higher performance.
+  Selector<unpacket_traits<Packet>::size> mask;
+  mask.select[0] = true;
+  // This for loop should be optimized away by the compiler.
+  for(Index i=1; i<unpacket_traits<Packet>::size; ++i)
+    mask.select[i] = false;
+  return pblend(mask, pset1<Packet>(b), a);
+}
+
+/** \internal \returns \a a with the last coefficient replaced by the scalar b */
+template<typename Packet> EIGEN_DEVICE_FUNC inline Packet
+pinsertlast(const Packet& a, typename unpacket_traits<Packet>::type b)
+{
+  // Default implementation based on pblend.
+  // It must be specialized for higher performance.
+  Selector<unpacket_traits<Packet>::size> mask;
+  // This for loop should be optimized away by the compiler.
+  for(Index i=0; i<unpacket_traits<Packet>::size-1; ++i)
+    mask.select[i] = false;
+  mask.select[unpacket_traits<Packet>::size-1] = true;
+  return pblend(mask, pset1<Packet>(b), a);
+}
+
+} // end namespace internal
+
+} // end namespace Eigen
+
+#endif // EIGEN_GENERIC_PACKET_MATH_H

Eigen/src/Core/GlobalFunctions.h

@@ -0,0 +1,187 @@
+// This file is part of Eigen, a lightweight C++ template library
+// for linear algebra.
+//
+// Copyright (C) 2010-2016 Gael Guennebaud <gael.guennebaud@inria.fr>
+// Copyright (C) 2010 Benoit Jacob <jacob.benoit.1@gmail.com>
+//
+// This Source Code Form is subject to the terms of the Mozilla
+// Public License v. 2.0. If a copy of the MPL was not distributed
+// with this file, You can obtain one at http://mozilla.org/MPL/2.0/.
+
+#ifndef EIGEN_GLOBAL_FUNCTIONS_H
+#define EIGEN_GLOBAL_FUNCTIONS_H
+
+#ifdef EIGEN_PARSED_BY_DOXYGEN
+
+#define EIGEN_ARRAY_DECLARE_GLOBAL_UNARY(NAME,FUNCTOR,DOC_OP,DOC_DETAILS) \
+  /** \returns an expression of the coefficient-wise DOC_OP of \a x
+
+    DOC_DETAILS
+
+    \sa <a href="group__CoeffwiseMathFunctions.html#cwisetable_##NAME">Math functions</a>, class CwiseUnaryOp
+    */ \
+  template<typename Derived> \
+  inline const Eigen::CwiseUnaryOp<Eigen::internal::FUNCTOR<typename Derived::Scalar>, const Derived> \
+  NAME(const Eigen::ArrayBase<Derived>& x);
+
+#else
+
+#define EIGEN_ARRAY_DECLARE_GLOBAL_UNARY(NAME,FUNCTOR,DOC_OP,DOC_DETAILS) \
+  template<typename Derived> \
+  inline const Eigen::CwiseUnaryOp<Eigen::internal::FUNCTOR<typename Derived::Scalar>, const Derived> \
+  (NAME)(const Eigen::ArrayBase<Derived>& x) { \
+    return Eigen::CwiseUnaryOp<Eigen::internal::FUNCTOR<typename Derived::Scalar>, const Derived>(x.derived()); \
+  }
+
+#endif // EIGEN_PARSED_BY_DOXYGEN
+
+#define EIGEN_ARRAY_DECLARE_GLOBAL_EIGEN_UNARY(NAME,FUNCTOR) \
+  \
+  template<typename Derived> \
+  struct NAME##_retval<ArrayBase<Derived> > \
+  { \
+    typedef const Eigen::CwiseUnaryOp<Eigen::internal::FUNCTOR<typename Derived::Scalar>, const Derived> type; \
+  }; \
+  template<typename Derived> \
+  struct NAME##_impl<ArrayBase<Derived> > \
+  { \
+    static inline typename NAME##_retval<ArrayBase<Derived> >::type run(const Eigen::ArrayBase<Derived>& x) \
+    { \
+      return typename NAME##_retval<ArrayBase<Derived> >::type(x.derived()); \
+    } \
+  };
+
+namespace Eigen
+{
+  EIGEN_ARRAY_DECLARE_GLOBAL_UNARY(real,scalar_real_op,real part,\sa ArrayBase::real)
+  EIGEN_ARRAY_DECLARE_GLOBAL_UNARY(imag,scalar_imag_op,imaginary part,\sa ArrayBase::imag)
+  EIGEN_ARRAY_DECLARE_GLOBAL_UNARY(conj,scalar_conjugate_op,complex conjugate,\sa ArrayBase::conjugate)
+  EIGEN_ARRAY_DECLARE_GLOBAL_UNARY(inverse,scalar_inverse_op,inverse,\sa ArrayBase::inverse)
+  EIGEN_ARRAY_DECLARE_GLOBAL_UNARY(sin,scalar_sin_op,sine,\sa ArrayBase::sin)
+  EIGEN_ARRAY_DECLARE_GLOBAL_UNARY(cos,scalar_cos_op,cosine,\sa ArrayBase::cos)
+  EIGEN_ARRAY_DECLARE_GLOBAL_UNARY(tan,scalar_tan_op,tangent,\sa ArrayBase::tan)
+  EIGEN_ARRAY_DECLARE_GLOBAL_UNARY(atan,scalar_atan_op,arc-tangent,\sa ArrayBase::atan)
+  EIGEN_ARRAY_DECLARE_GLOBAL_UNARY(asin,scalar_asin_op,arc-sine,\sa ArrayBase::asin)
+  EIGEN_ARRAY_DECLARE_GLOBAL_UNARY(acos,scalar_acos_op,arc-consine,\sa ArrayBase::acos)
+  EIGEN_ARRAY_DECLARE_GLOBAL_UNARY(sinh,scalar_sinh_op,hyperbolic sine,\sa ArrayBase::sinh)
+  EIGEN_ARRAY_DECLARE_GLOBAL_UNARY(cosh,scalar_cosh_op,hyperbolic cosine,\sa ArrayBase::cosh)
+  EIGEN_ARRAY_DECLARE_GLOBAL_UNARY(tanh,scalar_tanh_op,hyperbolic tangent,\sa ArrayBase::tanh)
+  EIGEN_ARRAY_DECLARE_GLOBAL_UNARY(lgamma,scalar_lgamma_op,natural logarithm of the gamma function,\sa ArrayBase::lgamma)
+  EIGEN_ARRAY_DECLARE_GLOBAL_UNARY(digamma,scalar_digamma_op,derivative of lgamma,\sa ArrayBase::digamma)
+  EIGEN_ARRAY_DECLARE_GLOBAL_UNARY(erf,scalar_erf_op,error function,\sa ArrayBase::erf)
+  EIGEN_ARRAY_DECLARE_GLOBAL_UNARY(erfc,scalar_erfc_op,complement error function,\sa ArrayBase::erfc)
+  EIGEN_ARRAY_DECLARE_GLOBAL_UNARY(exp,scalar_exp_op,exponential,\sa ArrayBase::exp)
+  EIGEN_ARRAY_DECLARE_GLOBAL_UNARY(log,scalar_log_op,natural logarithm,\sa Eigen::log10 DOXCOMMA ArrayBase::log)
+  EIGEN_ARRAY_DECLARE_GLOBAL_UNARY(log1p,scalar_log1p_op,natural logarithm of 1 plus the value,\sa ArrayBase::log1p)
+  EIGEN_ARRAY_DECLARE_GLOBAL_UNARY(log10,scalar_log10_op,base 10 logarithm,\sa Eigen::log DOXCOMMA ArrayBase::log)
+  EIGEN_ARRAY_DECLARE_GLOBAL_UNARY(abs,scalar_abs_op,absolute value,\sa ArrayBase::abs DOXCOMMA MatrixBase::cwiseAbs)
+  EIGEN_ARRAY_DECLARE_GLOBAL_UNARY(abs2,scalar_abs2_op,squared absolute value,\sa ArrayBase::abs2 DOXCOMMA MatrixBase::cwiseAbs2)
+  EIGEN_ARRAY_DECLARE_GLOBAL_UNARY(arg,scalar_arg_op,complex argument,\sa ArrayBase::arg)
+  EIGEN_ARRAY_DECLARE_GLOBAL_UNARY(sqrt,scalar_sqrt_op,square root,\sa ArrayBase::sqrt DOXCOMMA MatrixBase::cwiseSqrt)
+  EIGEN_ARRAY_DECLARE_GLOBAL_UNARY(rsqrt,scalar_rsqrt_op,reciprocal square root,\sa ArrayBase::rsqrt)
+  EIGEN_ARRAY_DECLARE_GLOBAL_UNARY(square,scalar_square_op,square (power 2),\sa Eigen::abs2 DOXCOMMA Eigen::pow DOXCOMMA ArrayBase::square)
+  EIGEN_ARRAY_DECLARE_GLOBAL_UNARY(cube,scalar_cube_op,cube (power 3),\sa Eigen::pow DOXCOMMA ArrayBase::cube)
+  EIGEN_ARRAY_DECLARE_GLOBAL_UNARY(round,scalar_round_op,nearest integer,\sa Eigen::floor DOXCOMMA Eigen::ceil DOXCOMMA ArrayBase::round)
+  EIGEN_ARRAY_DECLARE_GLOBAL_UNARY(floor,scalar_floor_op,nearest integer not greater than the giben value,\sa Eigen::ceil DOXCOMMA ArrayBase::floor)
+  EIGEN_ARRAY_DECLARE_GLOBAL_UNARY(ceil,scalar_ceil_op,nearest integer not less than the giben value,\sa Eigen::floor DOXCOMMA ArrayBase::ceil)
+  EIGEN_ARRAY_DECLARE_GLOBAL_UNARY(isnan,scalar_isnan_op,not-a-number test,\sa Eigen::isinf DOXCOMMA Eigen::isfinite DOXCOMMA ArrayBase::isnan)
+  EIGEN_ARRAY_DECLARE_GLOBAL_UNARY(isinf,scalar_isinf_op,infinite value test,\sa Eigen::isnan DOXCOMMA Eigen::isfinite DOXCOMMA ArrayBase::isinf)
+  EIGEN_ARRAY_DECLARE_GLOBAL_UNARY(isfinite,scalar_isfinite_op,finite value test,\sa Eigen::isinf DOXCOMMA Eigen::isnan DOXCOMMA ArrayBase::isfinite)
+  EIGEN_ARRAY_DECLARE_GLOBAL_UNARY(sign,scalar_sign_op,sign (or 0),\sa ArrayBase::sign)
+  
+  /** \returns an expression of the coefficient-wise power of \a x to the given constant \a exponent.
+    *
+    * \tparam ScalarExponent is the scalar type of \a exponent. It must be compatible with the scalar type of the given expression (\c Derived::Scalar).
+    *
+    * \sa ArrayBase::pow()
+    *
+    * \relates ArrayBase
+    */
+#ifdef EIGEN_PARSED_BY_DOXYGEN
+  template<typename Derived,typename ScalarExponent>
+  inline const CwiseBinaryOp<internal::scalar_pow_op<Derived::Scalar,ScalarExponent>,Derived,Constant<ScalarExponent> >
+  pow(const Eigen::ArrayBase<Derived>& x, const ScalarExponent& exponent);
+#else
+  template<typename Derived,typename ScalarExponent>
+  inline typename internal::enable_if<   !(internal::is_same<typename Derived::Scalar,ScalarExponent>::value) && EIGEN_SCALAR_BINARY_SUPPORTED(pow,typename Derived::Scalar,ScalarExponent),
+          const EIGEN_EXPR_BINARYOP_SCALAR_RETURN_TYPE(Derived,ScalarExponent,pow) >::type
+  pow(const Eigen::ArrayBase<Derived>& x, const ScalarExponent& exponent) {
+    return x.derived().pow(exponent);
+  }
+
+  template<typename Derived>
+  inline const EIGEN_EXPR_BINARYOP_SCALAR_RETURN_TYPE(Derived,typename Derived::Scalar,pow)
+  pow(const Eigen::ArrayBase<Derived>& x, const typename Derived::Scalar& exponent) {
+    return x.derived().pow(exponent);
+  }
+#endif
+
+  /** \returns an expression of the coefficient-wise power of \a x to the given array of \a exponents.
+    *
+    * This function computes the coefficient-wise power.
+    *
+    * Example: \include Cwise_array_power_array.cpp
+    * Output: \verbinclude Cwise_array_power_array.out
+    * 
+    * \sa ArrayBase::pow()
+    *
+    * \relates ArrayBase
+    */
+  template<typename Derived,typename ExponentDerived>
+  inline const Eigen::CwiseBinaryOp<Eigen::internal::scalar_pow_op<typename Derived::Scalar, typename ExponentDerived::Scalar>, const Derived, const ExponentDerived>
+  pow(const Eigen::ArrayBase<Derived>& x, const Eigen::ArrayBase<ExponentDerived>& exponents) 
+  {
+    return Eigen::CwiseBinaryOp<Eigen::internal::scalar_pow_op<typename Derived::Scalar, typename ExponentDerived::Scalar>, const Derived, const ExponentDerived>(
+      x.derived(),
+      exponents.derived()
+    );
+  }
+  
+  /** \returns an expression of the coefficient-wise power of the scalar \a x to the given array of \a exponents.
+    *
+    * This function computes the coefficient-wise power between a scalar and an array of exponents.
+    *
+    * \tparam Scalar is the scalar type of \a x. It must be compatible with the scalar type of the given array expression (\c Derived::Scalar).
+    *
+    * Example: \include Cwise_scalar_power_array.cpp
+    * Output: \verbinclude Cwise_scalar_power_array.out
+    * 
+    * \sa ArrayBase::pow()
+    *
+    * \relates ArrayBase
+    */
+#ifdef EIGEN_PARSED_BY_DOXYGEN
+  template<typename Scalar,typename Derived>
+  inline const CwiseBinaryOp<internal::scalar_pow_op<Scalar,Derived::Scalar>,Constant<Scalar>,Derived>
+  pow(const Scalar& x,const Eigen::ArrayBase<Derived>& x);
+#else
+  template<typename Scalar, typename Derived>
+  inline typename internal::enable_if<   !(internal::is_same<typename Derived::Scalar,Scalar>::value) && EIGEN_SCALAR_BINARY_SUPPORTED(pow,Scalar,typename Derived::Scalar),
+          const EIGEN_SCALAR_BINARYOP_EXPR_RETURN_TYPE(Scalar,Derived,pow) >::type
+  pow(const Scalar& x, const Eigen::ArrayBase<Derived>& exponents)
+  {
+    return EIGEN_SCALAR_BINARYOP_EXPR_RETURN_TYPE(Scalar,Derived,pow)(
+            typename internal::plain_constant_type<Derived,Scalar>::type(exponents.rows(), exponents.cols(), x), exponents.derived() );
+  }
+
+  template<typename Derived>
+  inline const EIGEN_SCALAR_BINARYOP_EXPR_RETURN_TYPE(typename Derived::Scalar,Derived,pow)
+  pow(const typename Derived::Scalar& x, const Eigen::ArrayBase<Derived>& exponents)
+  {
+    return EIGEN_SCALAR_BINARYOP_EXPR_RETURN_TYPE(typename Derived::Scalar,Derived,pow)(
+      typename internal::plain_constant_type<Derived,typename Derived::Scalar>::type(exponents.rows(), exponents.cols(), x), exponents.derived() );
+  }
+#endif
+
+
+  namespace internal
+  {
+    EIGEN_ARRAY_DECLARE_GLOBAL_EIGEN_UNARY(real,scalar_real_op)
+    EIGEN_ARRAY_DECLARE_GLOBAL_EIGEN_UNARY(imag,scalar_imag_op)
+    EIGEN_ARRAY_DECLARE_GLOBAL_EIGEN_UNARY(abs2,scalar_abs2_op)
+  }
+}
+
+// TODO: cleanly disable those functions that are not supported on Array (numext::real_ref, internal::random, internal::isApprox...)
+
+#endif // EIGEN_GLOBAL_FUNCTIONS_H