bump to 3.3-beta1

(grafted from f22036f5f8
)

.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

.hgignore

@@ -23,5 +23,12 @@ tags
 activity.png
 *.out
 *.php*
-eigen_gen_credits.log
-
+*.log
+*.orig
+*.rej
+log
+patch
+a
+a.*
+lapack/testing
+lapack/reference

.krazy

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

CMakeLists.txt

@@ -1,20 +1,33 @@
 project(Eigen)
 
-cmake_minimum_required(VERSION 2.6.2)
-
+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 ")
+  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()
 
+# 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 "${CMAKE_SOURCE_DIR}/Eigen/src/Core/util/Macros.h" _eigen_version_header)
+file(READ "${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}")
@@ -42,6 +55,7 @@ endif(EIGEN_HG_CHANGESET)
 
 
 include(CheckCXXCompilerFlag)
+include(GNUInstallDirs)
 
 set(CMAKE_MODULE_PATH ${PROJECT_SOURCE_DIR}/cmake)
 
@@ -51,6 +65,10 @@ set(CMAKE_MODULE_PATH ${PROJECT_SOURCE_DIR}/cmake)
 
 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)
@@ -81,13 +99,6 @@ endif(NOT WIN32)
 
 set(CMAKE_INCLUDE_CURRENT_DIR ON)
 
-string(TOLOWER "${CMAKE_BUILD_TYPE}" cmake_build_type_tolower)
-if(cmake_build_type_tolower STREQUAL "debug")
-  set(CMAKE_BUILD_TYPE "Debug")
-else()
-  set(CMAKE_BUILD_TYPE "Release")
-endif()
-
 option(EIGEN_SPLIT_LARGE_TESTS "Split large tests into smaller executables" ON)
 
 option(EIGEN_DEFAULT_TO_ROW_MAJOR "Use row-major as default matrix storage order" OFF)
@@ -95,77 +106,163 @@ if(EIGEN_DEFAULT_TO_ROW_MAJOR)
   add_definitions("-DEIGEN_DEFAULT_TO_ROW_MAJOR")
 endif()
 
-if(CMAKE_COMPILER_IS_GNUCXX)
-  set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -Wnon-virtual-dtor -Wno-long-long -ansi -Wundef -Wcast-align -Wchar-subscripts -Wall -W -Wpointer-arith -Wwrite-strings -Wformat-security -fexceptions -fno-check-new -fno-common -fstrict-aliasing")
-  set(CMAKE_CXX_FLAGS_DEBUG "-g3")
-  set(CMAKE_CXX_FLAGS_RELEASE "-g0 -O2")
+set(EIGEN_TEST_MAX_SIZE "320" CACHE STRING "Maximal matrix/vector size, default is 320")
 
-  check_cxx_compiler_flag("-Wno-variadic-macros" COMPILER_SUPPORT_WNOVARIADICMACRO)
-  if(COMPILER_SUPPORT_WNOVARIADICMACRO)
-    set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -Wno-variadic-macros")
+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)
+
+if(NOT MSVC)
+  # We assume that other compilers are partly compatible with GNUCC
+
+  # clang outputs some warnings for unknwon 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("-Wenum-conversion")
+  ei_add_cxx_compiler_flag("-Wc++11-extensions")
+  
+  # -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")                   # disbale 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()
 
-  check_cxx_compiler_flag("-Wextra" COMPILER_SUPPORT_WEXTRA)
-  if(COMPILER_SUPPORT_WEXTRA)
-    set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -Wextra")
+  if(ANDROID_NDK)
+    ei_add_cxx_compiler_flag("-pie")
+    ei_add_cxx_compiler_flag("-fPIE")
   endif()
-
-  set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -pedantic")
+  
+  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} -msse2")
-    message("Enabling SSE2 in tests/examples")
+    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("Enabling SSE3 in tests/examples")
+    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("Enabling SSSE3 in tests/examples")
+    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("Enabling SSE4.1 in tests/examples")
+    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 "${CMAKE_CXX_FLAGS} -msse4.2")
-    message("Enabling SSE4.2 in tests/examples")
+    message(STATUS "Enabling SSE4.2 in tests/examples")
+  endif()
+
+  option(EIGEN_TEST_AVX "Enable/Disable AVX in tests/examples" OFF)
+  if(EIGEN_TEST_AVX)
+    set(CMAKE_CXX_FLAGS "${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_ALTIVEC "Enable/Disable AltiVec in tests/examples" OFF)
   if(EIGEN_TEST_ALTIVEC)
     set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -maltivec -mabi=altivec")
-    message("Enabling AltiVec in tests/examples")
+    message(STATUS "Enabling AltiVec in tests/examples")
+  endif()
+
+  option(EIGEN_TEST_VSX "Enable/Disable VSX in tests/examples" OFF)
+  if(EIGEN_TEST_VSX)
+    set(CMAKE_CXX_FLAGS "${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)
-    set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -mfloat-abi=hard -mfpu=neon -mcpu=cortex-a8")
-    message("Enabling NEON in tests/examples")
+    if(EIGEN_TEST_FMA)
+      set(CMAKE_CXX_FLAGS "${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=softfp")
+    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 "${CMAKE_CXX_FLAGS}")
+    message(STATUS "Enabling NEON in tests/examples")
+  endif()
+
+
+
   check_cxx_compiler_flag("-fopenmp" COMPILER_SUPPORT_OPENMP)
   if(COMPILER_SUPPORT_OPENMP)
     option(EIGEN_TEST_OPENMP "Enable/Disable OpenMP in tests/examples" OFF)
     if(EIGEN_TEST_OPENMP)
       set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -fopenmp")
-      message("Enabling OpenMP in tests/examples")
+      message(STATUS "Enabling OpenMP in tests/examples")
     endif()
   endif()
 
-endif(CMAKE_COMPILER_IS_GNUCXX)
+else(NOT MSVC)
 
-if(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
@@ -183,7 +280,7 @@ if(MSVC)
     option(EIGEN_TEST_OPENMP "Enable/Disable OpenMP in tests/examples" OFF)
     if(EIGEN_TEST_OPENMP)
       set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} /openmp")
-      message("Enabling OpenMP in tests/examples")
+      message(STATUS "Enabling OpenMP in tests/examples")
     endif()
   endif()
 
@@ -193,56 +290,77 @@ if(MSVC)
       # 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("Enabling SSE2 in tests/examples")
+    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 "${CMAKE_CXX_FLAGS} -mfpmath=387")
-    message("Forcing use of x87 instructions in tests/examples")
+    message(STATUS "Forcing use of x87 instructions in tests/examples")
   else()
-    message("EIGEN_TEST_X87 ignored on your compiler")
+    message(STATUS "EIGEN_TEST_X87 ignored on your compiler")
+  endif()
+endif()
+
+if(EIGEN_TEST_32BIT)
+  if(CMAKE_COMPILER_IS_GNUCXX)
+    set(CMAKE_CXX_FLAGS "${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")
+  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("Disabling alignment in tests/examples")
+  message(STATUS "Disabling alignment in tests/examples")
 endif()
 
-option(EIGEN_TEST_C++0x "Enables all C++0x features." OFF)
+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()
+
+option(EIGEN_TEST_CXX11 "Enable testing with C++11 and C++11 features (e.g. Tensor module)." OFF)
 
 include_directories(${CMAKE_CURRENT_SOURCE_DIR} ${CMAKE_CURRENT_BINARY_DIR})
 
-# the user modifiable install path for header files
-set(EIGEN_INCLUDE_INSTALL_DIR ${EIGEN_INCLUDE_INSTALL_DIR} CACHE PATH "The directory where we install the header files (optional)")
-
-# set the internal install path for header files which depends on wether the user modifiable
-# EIGEN_INCLUDE_INSTALL_DIR has been set by the user or not.
+# Backward compatibility support for EIGEN_INCLUDE_INSTALL_DIR
 if(EIGEN_INCLUDE_INSTALL_DIR)
-  set(INCLUDE_INSTALL_DIR
-    ${EIGEN_INCLUDE_INSTALL_DIR}
-    CACHE INTERNAL
-    "The directory where we install the header files (internal)"
-  )
+  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_PREFIX}/include/eigen3"
-    CACHE INTERNAL
-    "The directory where we install the header files (internal)"
-  )
+      "${CMAKE_INSTALL_INCLUDEDIR}/eigen3"
+      CACHE PATH "The directory relative to CMAKE_PREFIX_PATH where Eigen header files are installed"
+      )
 endif()
+set(CMAKEPACKAGE_INSTALL_DIR
+    "${CMAKE_INSTALL_LIBDIR}/cmake/eigen3"
+    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)
@@ -261,9 +379,9 @@ install(FILES
   )
 
 if(EIGEN_BUILD_PKGCONFIG)
-    configure_file(eigen3.pc.in eigen3.pc)
+    configure_file(eigen3.pc.in eigen3.pc @ONLY)
     install(FILES ${CMAKE_CURRENT_BINARY_DIR}/eigen3.pc
-        DESTINATION share/pkgconfig
+        DESTINATION ${PKGCONFIG_INSTALL_DIR}
         )
 endif(EIGEN_BUILD_PKGCONFIG)
 
@@ -271,10 +389,11 @@ add_subdirectory(Eigen)
 
 add_subdirectory(doc EXCLUDE_FROM_ALL)
 
-include(CTest)
+include(EigenConfigureTesting)
+
+# fixme, not sure this line is still needed:
 enable_testing() # must be called from the root CMakeLists, see man page
-include(EigenTesting)
-ei_init_testing()
+
 
 if(EIGEN_LEAVE_TEST_IN_ALL_TARGET)
   add_subdirectory(test) # can't do EXCLUDE_FROM_ALL here, breaks CTest
@@ -282,12 +401,18 @@ else()
   add_subdirectory(test EXCLUDE_FROM_ALL)
 endif()
 
+if(EIGEN_LEAVE_TEST_IN_ALL_TARGET)
+  add_subdirectory(blas)
+  add_subdirectory(lapack)
+else()
+  add_subdirectory(blas EXCLUDE_FROM_ALL)
+  add_subdirectory(lapack EXCLUDE_FROM_ALL)
+endif()
+
 add_subdirectory(unsupported)
 
 add_subdirectory(demos EXCLUDE_FROM_ALL)
 
-add_subdirectory(blas EXCLUDE_FROM_ALL)
-
 # must be after test and unsupported, for configuring buildtests.in
 add_subdirectory(scripts EXCLUDE_FROM_ALL)
 
@@ -296,32 +421,71 @@ if(EIGEN_BUILD_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)
+
 ei_testing_print_summary()
 
-message("")
-message("Configured Eigen ${EIGEN_VERSION_NUMBER}")
-message("")
+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("Some things you can do now:")
-  message("--------------+----------------------------------------------------------------")
-  message("Command       |   Description")
-  message("--------------+----------------------------------------------------------------")
-  message("make install  | Install to ${CMAKE_INSTALL_PREFIX}. To change that:")
-  message("              |     cmake . -DCMAKE_INSTALL_PREFIX=yourpath")
-  message("              |   Eigen headers will then be installed to:")
-  message("              |     ${INCLUDE_INSTALL_DIR}")
-  message("              |   To install Eigen headers to a separate location, do:")
-  message("              |     cmake . -DEIGEN_INCLUDE_INSTALL_DIR=yourpath")
-  message("make doc      | Generate the API documentation, requires Doxygen & LaTeX")
-  message("make check    | Build and run the unit-tests. Read this page:")
-  message("              |   http://eigen.tuxfamily.org/index.php?title=Tests")
-  message("make blas     | Build BLAS library (not the same thing as Eigen)")
-  message("--------------+----------------------------------------------------------------")
+  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("To build/run the unit tests, read this page:")
-  message("  http://eigen.tuxfamily.org/index.php?title=Tests")
+  message(STATUS "To build/run the unit tests, read this page:")
+  message(STATUS "  http://eigen.tuxfamily.org/index.php?title=Tests")
 endif()
 
-message("")
+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_INCLUDE_DIRS ${EIGEN_INCLUDE_DIR} )
+set ( EIGEN_ROOT_DIR ${CMAKE_INSTALL_PREFIX} )
+
+configure_file ( ${CMAKE_CURRENT_SOURCE_DIR}/cmake/Eigen3Config.cmake.in
+                 ${CMAKE_CURRENT_BINARY_DIR}/Eigen3Config.cmake
+                 @ONLY ESCAPE_QUOTES
+               )
+
+install ( FILES ${CMAKE_CURRENT_SOURCE_DIR}/cmake/UseEigen3.cmake
+                ${CMAKE_CURRENT_BINARY_DIR}/Eigen3Config.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.LGPL

@@ -1,165 +1,502 @@
                   GNU LESSER GENERAL PUBLIC LICENSE
-                       Version 3, 29 June 2007
+                       Version 2.1, February 1999
 
- Copyright (C) 2007 Free Software Foundation, Inc. <http://fsf.org/>
+ 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.]
 
-  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.
+                            Preamble
 
-  0. Additional Definitions. 
+  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.
 
-  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.
+  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.
 
-  "The Library" refers to a covered work governed by this License,
-other than an Application or a Combined Work as defined 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.
 
-  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.
+  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.
 
-  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".
+  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.
 
-  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.
+  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.
 
-  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.
+  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.
 
-  1. Exception to Section 3 of the GNU GPL.
+  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.
 
-  You may convey a covered work under sections 3 and 4 of this License
-without being bound by section 3 of the GNU GPL.
+  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.
 
-  2. Conveying Modified Versions.
+  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.
 
-  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:
+  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.
 
-   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
+  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.
 
-   b) under the GNU GPL, with none of the additional permissions of
-   this License applicable to that copy.
+  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.
 
-  3. Object Code Incorporating Material from Library Header Files.
+  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
 
-  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:
+  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) 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.
+  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.
 
-   b) Accompany the object code with a copy of the GNU GPL and this license
-   document.
+  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".)
 
-  4. Combined Works.
+  "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.
 
-  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:
+  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.
 
-   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
+  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

@@ -5,13 +5,13 @@
 ##   ENABLE_TESTING()
 ##   INCLUDE(CTest)
 set(CTEST_PROJECT_NAME "Eigen")
-set(CTEST_NIGHTLY_START_TIME "06:00:00 UTC")
+set(CTEST_NIGHTLY_START_TIME "00:00:00 UTC")
 
 set(CTEST_DROP_METHOD "http")
-set(CTEST_DROP_SITE "eigen.tuxfamily.org")
+set(CTEST_DROP_SITE "manao.inria.fr")
 set(CTEST_DROP_LOCATION "/CDash/submit.php?project=Eigen")
 set(CTEST_DROP_SITE_CDASH TRUE)
-
-## A tribute to Dynamic!
-set(CTEST_CUSTOM_MAXIMUM_NUMBER_OF_WARNINGS "33331")
-set(CTEST_CUSTOM_MAXIMUM_NUMBER_OF_ERRORS "33331")
+set(CTEST_PROJECT_SUBPROJECTS
+Official
+Unsupported
+)

CTestCustom.cmake.in

@@ -0,0 +1,3 @@
+
+set(CTEST_CUSTOM_MAXIMUM_NUMBER_OF_WARNINGS "2000")
+set(CTEST_CUSTOM_MAXIMUM_NUMBER_OF_ERRORS   "2000")

Eigen/Array

@@ -1,14 +0,0 @@
-#ifndef EIGEN_ARRAY_MODULE_H
-#define EIGEN_ARRAY_MODULE_H
-
-#ifdef _MSC_VER
-#pragma message("The inclusion of Eigen/Array is deprecated. \
-The array module is available as soon as Eigen/Core is included.")
-#elif __GNUC__
-#warning "The inclusion of Eigen/Array is deprecated. \
-The array module is available as soon as Eigen/Core is included."
-#endif
-
-#include "Core"
-
-#endif // EIGEN_ARRAY_MODULE_H

Eigen/CMakeLists.txt

@@ -1,6 +1,12 @@
+include(RegexUtils)
+test_escape_string_as_regex()
+
 file(GLOB Eigen_directory_files "*")
+
+escape_string_as_regex(ESCAPED_CMAKE_CURRENT_SOURCE_DIR "${CMAKE_CURRENT_SOURCE_DIR}")
+
 foreach(f ${Eigen_directory_files})
-  if(NOT f MATCHES ".txt" AND NOT f MATCHES "${CMAKE_CURRENT_SOURCE_DIR}/src")
+  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})

Eigen/Cholesky

@@ -1,33 +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_CHOLESKY_MODULE_H
 #define EIGEN_CHOLESKY_MODULE_H
 
 #include "Core"
 
-#include "src/Core/util/DisableMSVCWarnings.h"
-
-namespace Eigen {
+#include "src/Core/util/DisableStupidWarnings.h"
 
 /** \defgroup Cholesky_Module Cholesky module
   *
   *
   *
   * 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/misc/Solve.h"
 #include "src/Cholesky/LLT.h"
 #include "src/Cholesky/LDLT.h"
+#ifdef EIGEN_USE_LAPACKE
+#include "src/Cholesky/LLT_MKL.h"
+#endif
 
-} // namespace Eigen
-
-#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.cise.ufl.edu/research/sparse/SuiteSparse/">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

@@ -2,73 +2,104 @@
 // for linear algebra.
 //
 // Copyright (C) 2008 Gael Guennebaud <gael.guennebaud@inria.fr>
-// Copyright (C) 2007-2010 Benoit Jacob <jacob.benoit.1@gmail.com>
+// Copyright (C) 2007-2011 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_CORE_H
 #define EIGEN_CORE_H
 
-#define EIGEN_NO_STATIC_ASSERT
+// first thing Eigen does: stop the compiler from committing suicide
+#include "src/Core/util/DisableStupidWarnings.h"
 
-// first thing Eigen does: prevent MSVC from committing suicide
-#include "src/Core/util/DisableMSVCWarnings.h"
+// Handle NVCC/CUDA
+#ifdef __CUDACC__
+  // Do not try asserts on CUDA!
+  #ifndef EIGEN_NO_DEBUG
+  #define EIGEN_NO_DEBUG
+  #endif
+
+  #ifdef EIGEN_INTERNAL_DEBUGGING
+  #undef EIGEN_INTERNAL_DEBUGGING
+  #endif
+
+  // Do not try to vectorize on CUDA!
+  #ifndef EIGEN_DONT_VECTORIZE
+  #define EIGEN_DONT_VECTORIZE
+  #endif
+
+  #ifdef EIGEN_EXCEPTIONS
+  #undef EIGEN_EXCEPTIONS
+  #endif
+  
+  // All functions callable from CUDA code must be qualified with __device__
+  #define EIGEN_DEVICE_FUNC __host__ __device__
+  
+#else
+  #define EIGEN_DEVICE_FUNC
+  
+#endif
+
+#if defined(__CUDA_ARCH__)
+  #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)
+  #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.
-#ifndef EIGEN_PARSED_BY_DOXYGEN
-  #include "src/Core/util/Macros.h"
-#else
-  namespace Eigen {  // for some reason Doxygen needs this namespace
-    #include "src/Core/util/Macros.h"
-  }
+#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
 
-// if alignment is disabled, then disable vectorization. Note: EIGEN_ALIGN is the proper check, it takes into
-// account both the user's will (EIGEN_DONT_ALIGN) and our own platform checks
-#if !EIGEN_ALIGN
+#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 _MSC_VER
+#if EIGEN_COMP_MSVC
   #include <malloc.h> // for _aligned_malloc -- need it regardless of whether vectorization is enabled
-  #if (_MSC_VER >= 1500) // 2008 or later
+  #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)) || defined(_M_X64)
+    #if (defined(_M_IX86_FP) && (_M_IX86_FP >= 2)) || EIGEN_ARCH_x86_64
       #define EIGEN_SSE2_ON_MSVC_2008_OR_LATER
     #endif
   #endif
-#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
+#else
+  // 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.
@@ -93,30 +124,63 @@
     #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
 
     // include files
-    #if (defined __GNUC__) && (defined __MINGW32__)
-      #include <intrin.h>
-      //including intrin.h works around a MINGW bug http://sourceforge.net/tracker/?func=detail&atid=102435&aid=2962480&group_id=2435
-      //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,
-      //include intrin here.
-    #endif
-    #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
+
+    // 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 <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
+        #ifdef EIGEN_VECTORIZE_AVX
+        #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
@@ -126,13 +190,18 @@
     #undef bool
     #undef vector
     #undef pixel
-  #elif defined  __ARM_NEON__
+  #elif (defined  __ARM_NEON) || (defined __ARM_NEON__)
     #define EIGEN_VECTORIZE
     #define EIGEN_VECTORIZE_NEON
     #include <arm_neon.h>
   #endif
 #endif
 
+#if defined __CUDACC__
+  #define EIGEN_VECTORIZE_CUDA
+  #include <vector_types.h>
+#endif
+
 #if (defined _OPENMP) && (!defined EIGEN_DONT_PARALLELIZE)
   #define EIGEN_HAS_OPENMP
 #endif
@@ -142,57 +211,43 @@
 #endif
 
 // MSVC for windows mobile does not have the errno.h file
-#if !(defined(_MSC_VER) && defined(_WIN32_WCE))
+#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 <iosfwd>
 #include <cstring>
 #include <string>
 #include <limits>
+#include <climits> // for CHAR_BIT
 // for min/max:
 #include <algorithm>
 
 // for outputting debug info
 #ifdef EIGEN_DEBUG_ASSIGN
-#include<iostream>
+#include <iostream>
 #endif
 
 // required for __cpuid, needs to be included after cmath
-#if defined(_MSC_VER) && (defined(_M_IX86)||defined(_M_IX64))
+#if EIGEN_COMP_MSVC && EIGEN_ARCH_i386_OR_x86_64 && !EIGEN_OS_WINCE
   #include <intrin.h>
 #endif
 
-#if (defined(_CPPUNWIND) || defined(__EXCEPTIONS)) && !defined(EIGEN_NO_EXCEPTIONS)
-  #define EIGEN_EXCEPTIONS
-#endif
-
-#ifdef EIGEN_EXCEPTIONS
-  #include <new>
-#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.
-#endif
-
-// defined in bits/termios.h
-#undef B0
-
+/** \brief Namespace containing all symbols from the %Eigen library. */
 namespace Eigen {
 
 inline static const char *SimdInstructionSetsInUse(void) {
-#if defined(EIGEN_VECTORIZE_SSE4_2)
+#if 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";
@@ -204,6 +259,8 @@ inline static const char *SimdInstructionSetsInUse(void) {
   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";
 #else
@@ -211,13 +268,18 @@ inline static const char *SimdInstructionSetsInUse(void) {
 #endif
 }
 
-#ifdef EIGEN2_SUPPORT
-#undef minor
+} // 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
 
 // 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
@@ -230,49 +292,80 @@ using std::size_t;
   */
 
 #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/SpecialFunctions.h"
 #include "src/Core/GenericPacketMath.h"
 
-#if defined EIGEN_VECTORIZE_SSE
+#if 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"
+#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"
-#elif defined EIGEN_VECTORIZE_ALTIVEC
+  #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"
 #endif
 
+#if defined EIGEN_VECTORIZE_CUDA
+  #include "src/Core/arch/CUDA/PacketMath.h"
+  #include "src/Core/arch/CUDA/MathFunctions.h"
+#endif
+
 #include "src/Core/arch/Default/Settings.h"
 
-#include "src/Core/Functors.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"
+
 #include "src/Core/DenseCoeffsBase.h"
 #include "src/Core/DenseBase.h"
 #include "src/Core/MatrixBase.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/ArrayBase.h"
 #include "src/Core/util/BlasUtil.h"
-#include "src/Core/MatrixStorage.h"
+#include "src/Core/DenseStorage.h"
 #include "src/Core/NestByValue.h"
-#include "src/Core/ForceAlignedAccess.h"
+
+// #include "src/Core/ForceAlignedAccess.h"
+
 #include "src/Core/ReturnByValue.h"
 #include "src/Core/NoAlias.h"
-#include "src/Core/DenseStorageBase.h"
+#include "src/Core/PlainObjectBase.h"
 #include "src/Core/Matrix.h"
+#include "src/Core/Array.h"
 #include "src/Core/CwiseBinaryOp.h"
 #include "src/Core/CwiseUnaryOp.h"
 #include "src/Core/CwiseNullaryOp.h"
@@ -280,34 +373,37 @@ using std::size_t;
 #include "src/Core/SelfCwiseBinaryOp.h"
 #include "src/Core/Dot.h"
 #include "src/Core/StableNorm.h"
-#include "src/Core/MapBase.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/VectorBlock.h"
 #include "src/Core/Transpose.h"
 #include "src/Core/DiagonalMatrix.h"
 #include "src/Core/Diagonal.h"
 #include "src/Core/DiagonalProduct.h"
-#include "src/Core/PermutationMatrix.h"
-#include "src/Core/Transpositions.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/Flagged.h"
-#include "src/Core/ProductBase.h"
-#include "src/Core/Product.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/SolveTriangular.h"
-#include "src/Core/products/Parallelizer.h"
-#include "src/Core/products/CoeffBasedProduct.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"
@@ -315,7 +411,9 @@ using std::size_t;
 #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/BooleanRedux.h"
 #include "src/Core/Select.h"
@@ -323,18 +421,25 @@ using std::size_t;
 #include "src/Core/Random.h"
 #include "src/Core/Replicate.h"
 #include "src/Core/Reverse.h"
-#include "src/Core/ArrayBase.h"
 #include "src/Core/ArrayWrapper.h"
-#include "src/Core/Array.h"
 
-} // namespace Eigen
+#ifdef EIGEN_USE_BLAS
+#include "src/Core/products/GeneralMatrixMatrix_MKL.h"
+#include "src/Core/products/GeneralMatrixVector_MKL.h"
+#include "src/Core/products/GeneralMatrixMatrixTriangular_MKL.h"
+#include "src/Core/products/SelfadjointMatrixMatrix_MKL.h"
+#include "src/Core/products/SelfadjointMatrixVector_MKL.h"
+#include "src/Core/products/TriangularMatrixMatrix_MKL.h"
+#include "src/Core/products/TriangularMatrixVector_MKL.h"
+#include "src/Core/products/TriangularSolverMatrix_MKL.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/EnableMSVCWarnings.h"
-
-#ifdef EIGEN2_SUPPORT
-#include "Eigen2Support"
-#endif
+#include "src/Core/util/ReenableStupidWarnings.h"
 
 #endif // EIGEN_CORE_H

Eigen/Eigen2Support

@@ -1,78 +0,0 @@
-// This file is part of Eigen, a lightweight C++ template library
-// for linear algebra.
-//
-// Copyright (C) 2009 Gael Guennebaud <gael.guennebaud@inria.fr>
-//
-// Eigen is free software; you can redistribute it and/or
-// modify it under the terms of the GNU Lesser General Public
-// License as published by the Free Software Foundation; either
-// version 3 of the License, or (at your option) any later version.
-//
-// Alternatively, you can redistribute it and/or
-// modify it under the terms of the GNU General Public License as
-// published by the Free Software Foundation; either version 2 of
-// the License, or (at your option) any later version.
-//
-// Eigen is distributed in the hope that it will be useful, but WITHOUT ANY
-// WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS
-// FOR A PARTICULAR PURPOSE. See the GNU Lesser General Public License or the
-// GNU General Public License for more details.
-//
-// You should have received a copy of the GNU Lesser General Public
-// License and a copy of the GNU General Public License along with
-// Eigen. If not, see <http://www.gnu.org/licenses/>.
-
-#ifndef EIGEN2SUPPORT_H
-#define EIGEN2SUPPORT_H
-
-#if (!defined(EIGEN2_SUPPORT)) || (!defined(EIGEN_CORE_H))
-#error Eigen2 support must be enabled by defining EIGEN2_SUPPORT before including any Eigen header
-#endif
-
-#include "src/Core/util/DisableMSVCWarnings.h"
-
-namespace Eigen {
-
-/** \defgroup Eigen2Support_Module Eigen2 support module
-  * This module provides a couple of deprecated functions improving the compatibility with Eigen2.
-  *
-  * To use it, define EIGEN2_SUPPORT before including any Eigen header
-  * \code
-  * #define EIGEN2_SUPPORT
-  * \endcode
-  *
-  */
-
-#include "src/Eigen2Support/Lazy.h"
-#include "src/Eigen2Support/Cwise.h"
-#include "src/Eigen2Support/CwiseOperators.h"
-#include "src/Eigen2Support/TriangularSolver.h"
-#include "src/Eigen2Support/Block.h"
-#include "src/Eigen2Support/VectorBlock.h"
-#include "src/Eigen2Support/Minor.h"
-
-
-} // namespace Eigen
-
-#include "src/Core/util/EnableMSVCWarnings.h"
-
-// Eigen2 used to include iostream
-#include<iostream>
-
-#define USING_PART_OF_NAMESPACE_EIGEN \
-EIGEN_USING_MATRIX_TYPEDEFS \
-using Eigen::Matrix; \
-using Eigen::MatrixBase; \
-using Eigen::ei_random; \
-using Eigen::ei_real; \
-using Eigen::ei_imag; \
-using Eigen::ei_conj; \
-using Eigen::ei_abs; \
-using Eigen::ei_abs2; \
-using Eigen::ei_sqrt; \
-using Eigen::ei_exp; \
-using Eigen::ei_log; \
-using Eigen::ei_sin; \
-using Eigen::ei_cos;
-
-#endif // EIGEN2SUPPORT_H

Eigen/Eigenvalues

@@ -1,16 +1,22 @@
+// 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 "src/Core/util/DisableMSVCWarnings.h"
+#include "src/Core/util/DisableStupidWarnings.h"
 
 #include "Cholesky"
 #include "Jacobi"
 #include "Householder"
 #include "LU"
-
-namespace Eigen {
+#include "Geometry"
 
 /** \defgroup Eigenvalues_Module Eigenvalues module
   *
@@ -34,11 +40,16 @@ namespace Eigen {
 #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
+#include "src/Eigenvalues/RealSchur_MKL.h"
+#include "src/Eigenvalues/ComplexSchur_MKL.h"
+#include "src/Eigenvalues/SelfAdjointEigenSolver_MKL.h"
+#endif
 
-} // namespace Eigen
-
-#include "src/Core/util/EnableMSVCWarnings.h"
+#include "src/Core/util/ReenableStupidWarnings.h"
 
 #endif // EIGEN_EIGENVALUES_MODULE_H
 /* vim: set filetype=cpp et sw=2 ts=2 ai: */

Eigen/Geometry

@@ -1,20 +1,21 @@
+// 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 "src/Core/util/DisableStupidWarnings.h"
 
 #include "SVD"
 #include "LU"
 #include <limits>
 
-#ifndef M_PI
-#define M_PI 3.14159265358979323846
-#endif
-
-namespace Eigen {
-
 /** \defgroup Geometry_Module Geometry module
   *
   *
@@ -33,12 +34,13 @@ 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"
@@ -47,13 +49,13 @@ namespace Eigen {
 #include "src/Geometry/AlignedBox.h"
 #include "src/Geometry/Umeyama.h"
 
-#if defined EIGEN_VECTORIZE_SSE
-  #include "src/Geometry/arch/Geometry_SSE.h"
+// 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
 
-} // namespace Eigen
-
-#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

@@ -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_HOUSEHOLDER_MODULE_H
 #define EIGEN_HOUSEHOLDER_MODULE_H
 
 #include "Core"
 
-#include "src/Core/util/DisableMSVCWarnings.h"
-
-namespace Eigen {
+#include "src/Core/util/DisableStupidWarnings.h"
 
 /** \defgroup Householder_Module Householder module
   * This module provides Householder transformations.
@@ -19,9 +24,7 @@ namespace Eigen {
 #include "src/Householder/HouseholderSequence.h"
 #include "src/Householder/BlockHouseholder.h"
 
-} // namespace Eigen
-
-#include "src/Core/util/EnableMSVCWarnings.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

@@ -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_JACOBI_MODULE_H
 #define EIGEN_JACOBI_MODULE_H
 
 #include "Core"
 
-#include "src/Core/util/DisableMSVCWarnings.h"
-
-namespace Eigen {
+#include "src/Core/util/DisableStupidWarnings.h"
 
 /** \defgroup Jacobi_Module Jacobi module
   * This module provides Jacobi and Givens rotations.
@@ -21,9 +26,7 @@ namespace Eigen {
 
 #include "src/Jacobi/Jacobi.h"
 
-} // namespace Eigen
-
-#include "src/Core/util/EnableMSVCWarnings.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.
@@ -18,21 +23,23 @@ namespace Eigen {
   * \endcode
   */
 
-#include "src/misc/Solve.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
+#include "src/LU/PartialPivLU_MKL.h"
+#endif
 #include "src/LU/Determinant.h"
-#include "src/LU/Inverse.h"
+#include "src/LU/InverseImpl.h"
 
-#if defined EIGEN_VECTORIZE_SSE
+// 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
 
-} // namespace Eigen
-
-#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/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/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 +1,49 @@
+// 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"
+
+#include <complex.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

@@ -1,41 +1,45 @@
+// 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 "src/Core/util/DisableStupidWarnings.h"
 
 #include "Cholesky"
 #include "Jacobi"
 #include "Householder"
 
-namespace Eigen {
-
 /** \defgroup QR_Module QR module
   *
   *
   *
   * This module provides various QR decompositions
   * This module also provides some MatrixBase methods, including:
-  *  - MatrixBase::qr(),
+  *  - MatrixBase::householderQr()
+  *  - MatrixBase::colPivHouseholderQr()
+  *  - MatrixBase::fullPivHouseholderQr()
   *
   * \code
   * #include <Eigen/QR>
   * \endcode
   */
 
-#include "src/misc/Solve.h"
 #include "src/QR/HouseholderQR.h"
 #include "src/QR/FullPivHouseholderQR.h"
 #include "src/QR/ColPivHouseholderQR.h"
+#ifdef EIGEN_USE_LAPACKE
+#include "src/QR/HouseholderQR_MKL.h"
+#include "src/QR/ColPivHouseholderQR_MKL.h"
+#endif
 
-
-} // namespace Eigen
-
-#include "src/Core/util/EnableMSVCWarnings.h"
-
-// FIXME for compatibility we include Eigenvalues here:
-#include "Eigenvalues"
+#include "src/Core/util/ReenableStupidWarnings.h"
 
 #endif // EIGEN_QR_MODULE_H
 /* vim: set filetype=cpp et sw=2 ts=2 ai: */

Eigen/QtAlignedMalloc

@@ -1,3 +1,9 @@
+// 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
@@ -6,27 +12,27 @@
 
 #if (!EIGEN_MALLOC_ALREADY_ALIGNED)
 
-#include "src/Core/util/DisableMSVCWarnings.h"
+#include "src/Core/util/DisableStupidWarnings.h"
 
 void *qMalloc(size_t size)
 {
-  return Eigen::ei_aligned_malloc(size);
+  return Eigen::internal::aligned_malloc(size);
 }
 
 void qFree(void *ptr)
 {
-  Eigen::ei_aligned_free(ptr);
+  Eigen::internal::aligned_free(ptr);
 }
 
 void *qRealloc(void *ptr, size_t size)
 {
-  void* newPtr = Eigen::ei_aligned_malloc(size);
+  void* newPtr = Eigen::internal::aligned_malloc(size);
   memcpy(newPtr, ptr, size);
-  Eigen::ei_aligned_free(ptr);
+  Eigen::internal::aligned_free(ptr);
   return newPtr;
 }
 
-#include "src/Core/util/EnableMSVCWarnings.h"
+#include "src/Core/util/ReenableStupidWarnings.h"
 
 #endif
 

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.cise.ufl.edu/research/sparse/SuiteSparse/">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

@@ -1,3 +1,10 @@
+// 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
 
@@ -5,30 +12,34 @@
 #include "Householder"
 #include "Jacobi"
 
-#include "src/Core/util/DisableMSVCWarnings.h"
-
-namespace Eigen {
+#include "src/Core/util/DisableStupidWarnings.h"
 
 /** \defgroup SVD_Module SVD module
   *
   *
   *
-  * 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/misc/Solve.h"
-#include "src/SVD/JacobiSVD.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)
+#include "src/SVD/JacobiSVD_MKL.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

@@ -1,61 +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_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>
-
-namespace Eigen {
-
-/** \defgroup Sparse_Module Sparse module
+/** \defgroup Sparse_Module Sparse meta-module
   *
+  * 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/Sparse>
-  * \endcode
+    \code
+    #include <Eigen/Sparse>
+    \endcode
   */
 
-/** The type used to identify a general sparse storage. */
-struct Sparse {};
-
-#include "src/Sparse/SparseUtil.h"
-#include "src/Sparse/SparseMatrixBase.h"
-#include "src/Sparse/CompressedStorage.h"
-#include "src/Sparse/AmbiVector.h"
-#include "src/Sparse/SparseMatrix.h"
-#include "src/Sparse/DynamicSparseMatrix.h"
-#include "src/Sparse/MappedSparseMatrix.h"
-#include "src/Sparse/SparseVector.h"
-#include "src/Sparse/CoreIterators.h"
-#include "src/Sparse/SparseBlock.h"
-#include "src/Sparse/SparseTranspose.h"
-#include "src/Sparse/SparseCwiseUnaryOp.h"
-#include "src/Sparse/SparseCwiseBinaryOp.h"
-#include "src/Sparse/SparseDot.h"
-#include "src/Sparse/SparseAssign.h"
-#include "src/Sparse/SparseRedux.h"
-#include "src/Sparse/SparseFuzzy.h"
-#include "src/Sparse/SparseProduct.h"
-#include "src/Sparse/SparseSparseProduct.h"
-#include "src/Sparse/SparseDenseProduct.h"
-#include "src/Sparse/SparseDiagonalProduct.h"
-#include "src/Sparse/SparseTriangularView.h"
-#include "src/Sparse/SparseSelfAdjointView.h"
-#include "src/Sparse/TriangularSolver.h"
-#include "src/Sparse/SparseView.h"
-
-} // namespace Eigen
-
-#include "src/Core/util/EnableMSVCWarnings.h"
+#include "SparseCore"
+#include "OrderingMethods"
+#include "SparseCholesky"
+#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 +1,37 @@
+// 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 "OrderingMethods"
+#include "src/SparseCore/SparseColEtree.h"
+#include "src/SparseQR/SparseQR.h"
+
+#include "src/Core/util/ReenableStupidWarnings.h"
+
+#endif

Eigen/StdDeque

@@ -4,24 +4,9 @@
 // Copyright (C) 2009 Gael Guennebaud <gael.guennebaud@inria.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/>.
+// 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
@@ -29,7 +14,7 @@
 #include "Core"
 #include <deque>
 
-#if (defined(_MSC_VER) && defined(_WIN64)) /* MSVC auto aligns in 64 bit builds */
+#if EIGEN_COMP_MSVC && EIGEN_OS_WIN64 /* MSVC auto aligns in 64 bit builds */
 
 #define EIGEN_DEFINE_STL_DEQUE_SPECIALIZATION(...)
 

Eigen/StdList

@@ -3,24 +3,9 @@
 //
 // 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/>.
+// 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
@@ -28,7 +13,7 @@
 #include "Core"
 #include <list>
 
-#if (defined(_MSC_VER) && defined(_WIN64)) /* MSVC auto aligns in 64 bit builds */    
+#if EIGEN_COMP_MSVC && EIGEN_OS_WIN64 /* MSVC auto aligns in 64 bit builds */    
 
 #define EIGEN_DEFINE_STL_LIST_SPECIALIZATION(...)
 

Eigen/StdVector

@@ -4,24 +4,9 @@
 // Copyright (C) 2009 Gael Guennebaud <gael.guennebaud@inria.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/>.
+// 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
@@ -29,7 +14,7 @@
 #include "Core"
 #include <vector>
 
-#if (defined(_MSC_VER) && defined(_WIN64)) /* MSVC auto aligns in 64 bit builds */
+#if EIGEN_COMP_MSVC && EIGEN_OS_WIN64 /* MSVC auto aligns in 64 bit builds */
 
 #define EIGEN_DEFINE_STL_VECTOR_SPECIALIZATION(...)
 

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 is only for the 4.x versions of SuperLU. The 3.x and 5.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.cise.ufl.edu/research/sparse/SuiteSparse/">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/CMakeLists.txt

@@ -1,6 +1,7 @@
 file(GLOB Eigen_src_subdirectories "*")
+escape_string_as_regex(ESCAPED_CMAKE_CURRENT_SOURCE_DIR "${CMAKE_CURRENT_SOURCE_DIR}")
 foreach(f ${Eigen_src_subdirectories})
-  if(NOT f MATCHES ".txt")
+  if(NOT f MATCHES "\\.txt" AND NOT f MATCHES "${ESCAPED_CMAKE_CURRENT_SOURCE_DIR}/[.].+" )
     add_subdirectory(${f})
   endif()
 endforeach()

Eigen/src/Cholesky/LDLT.h

@@ -1,41 +1,36 @@
 // This file is part of Eigen, a lightweight C++ template library
 // for linear algebra.
 //
-// Copyright (C) 2008-2010 Gael Guennebaud <gael.guennebaud@inria.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
 
-template<typename MatrixType, int UpLo> struct LDLT_Traits;
+namespace Eigen { 
 
-/** \ingroup cholesky_Module
+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 with pivoting
   *
   * \param MatrixType the type of the matrix of which to compute the LDL^T Cholesky decomposition
+  * \param UpLo the triangular part that will be used for the decompositon: Lower (default) or Upper.
+  *             The other triangular part won't be read.
   *
   * 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
@@ -46,13 +41,9 @@ template<typename MatrixType, int UpLo> struct LDLT_Traits;
   * on D also stabilizes the computation.
   *
   * 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.
+  * decomposition to determine whether a system of equations has a solution.
   *
-  * \sa MatrixBase::ldlt(), class LLT
-  */
- /* THIS PART OF THE DOX IS CURRENTLY DISABLED BECAUSE INACCURATE BECAUSE OF BUG IN THE DECOMPOSITION CODE
-  * 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.
+  * \sa MatrixBase::ldlt(), SelfAdjointView::ldlt(), class LLT
   */
 template<typename _MatrixType, int _UpLo> class LDLT
 {
@@ -68,20 +59,26 @@ template<typename _MatrixType, int _UpLo> class LDLT
     };
     typedef typename MatrixType::Scalar Scalar;
     typedef typename NumTraits<typename MatrixType::Scalar>::Real RealScalar;
-    typedef typename MatrixType::Index Index;
+    typedef Eigen::Index Index; ///< \deprecated since Eigen 3.3
+    typedef typename MatrixType::StorageIndex StorageIndex;
     typedef Matrix<Scalar, RowsAtCompileTime, 1, Options, MaxRowsAtCompileTime, 1> TmpMatrixType;
 
     typedef Transpositions<RowsAtCompileTime, MaxRowsAtCompileTime> TranspositionType;
     typedef PermutationMatrix<RowsAtCompileTime, MaxRowsAtCompileTime> PermutationType;
 
-    typedef LDLT_Traits<MatrixType,UpLo> Traits;
+    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_isInitialized(false) {}
+    LDLT() 
+      : m_matrix(), 
+        m_transpositions(), 
+        m_sign(internal::ZeroSign),
+        m_isInitialized(false) 
+    {}
 
     /** \brief Default Constructor with memory preallocation
       *
@@ -89,33 +86,49 @@ template<typename _MatrixType, int _UpLo> class LDLT
       * according to the specified problem \a size.
       * \sa LDLT()
       */
-    LDLT(Index size)
+    explicit LDLT(Index size)
       : m_matrix(size, size),
         m_transpositions(size),
         m_temporary(size),
+        m_sign(internal::ZeroSign),
         m_isInitialized(false)
     {}
 
-    LDLT(const MatrixType& matrix)
+    /** \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());
+    }
+
+    /** 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
     {
-      ei_assert(m_isInitialized && "LDLT is not initialized.");
+      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
     {
-      ei_assert(m_isInitialized && "LDLT is not initialized.");
+      eigen_assert(m_isInitialized && "LDLT is not initialized.");
       return Traits::getL(m_matrix);
     }
 
@@ -123,51 +136,64 @@ template<typename _MatrixType, int _UpLo> class LDLT
       */
     inline const TranspositionType& transpositionsP() const
     {
-      ei_assert(m_isInitialized && "LDLT is not initialized.");
+      eigen_assert(m_isInitialized && "LDLT is not initialized.");
       return m_transpositions;
     }
 
     /** \returns the coefficients of the diagonal matrix D */
-    inline Diagonal<MatrixType,0> vectorD(void) const
+    inline Diagonal<const MatrixType> vectorD() const
     {
-      ei_assert(m_isInitialized && "LDLT is not initialized.");
+      eigen_assert(m_isInitialized && "LDLT is not initialized.");
       return m_matrix.diagonal();
     }
 
     /** \returns true if the matrix is positive (semidefinite) */
-    inline bool isPositive(void) const
+    inline bool isPositive() const
     {
-      ei_assert(m_isInitialized && "LDLT is not initialized.");
-      return m_sign == 1;
+      eigen_assert(m_isInitialized && "LDLT is not initialized.");
+      return m_sign == internal::PositiveSemiDef || m_sign == internal::ZeroSign;
     }
 
     /** \returns true if the matrix is negative (semidefinite) */
     inline bool isNegative(void) const
     {
-      ei_assert(m_isInitialized && "LDLT is not initialized.");
-      return m_sign == -1;
+      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
       *
-      * \sa solveInPlace(), MatrixBase::ldlt()
+      * 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 ei_solve_retval<LDLT, Rhs>
+    inline const Solve<LDLT, Rhs>
     solve(const MatrixBase<Rhs>& b) const
     {
-      ei_assert(m_isInitialized && "LDLT is not initialized.");
-      ei_assert(m_matrix.rows()==b.rows()
+      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 ei_solve_retval<LDLT, Rhs>(*this, b.derived());
+      return Solve<LDLT, Rhs>(*this, b.derived());
     }
 
     template<typename Derived>
     bool solveInPlace(MatrixBase<Derived> &bAndX) const;
 
-    LDLT& compute(const MatrixType& matrix);
+    template<typename InputType>
+    LDLT& compute(const EigenBase<InputType>& matrix);
+
+    template <typename Derived>
+    LDLT& rankUpdate(const MatrixBase<Derived>& w, const RealScalar& alpha=1);
 
     /** \returns the internal LDLT decomposition matrix
       *
@@ -175,7 +201,7 @@ template<typename _MatrixType, int _UpLo> class LDLT
       */
     inline const MatrixType& matrixLDLT() const
     {
-      ei_assert(m_isInitialized && "LDLT is not initialized.");
+      eigen_assert(m_isInitialized && "LDLT is not initialized.");
       return m_matrix;
     }
 
@@ -184,7 +210,29 @@ template<typename _MatrixType, int _UpLo> class LDLT
     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 matrix.appears to be negative.
+      */
+    ComputationInfo info() const
+    {
+      eigen_assert(m_isInitialized && "LDLT is not initialized.");
+      return Success;
+    }
+    
+    #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.
@@ -195,59 +243,43 @@ template<typename _MatrixType, int _UpLo> class LDLT
     MatrixType m_matrix;
     TranspositionType m_transpositions;
     TmpMatrixType m_temporary;
-    int m_sign;
+    internal::SignMatrix m_sign;
     bool m_isInitialized;
 };
 
-template<int UpLo> struct ei_ldlt_inplace;
+namespace internal {
 
-template<> struct ei_ldlt_inplace<Lower>
+template<int UpLo> struct ldlt_inplace;
+
+template<> struct ldlt_inplace<Lower>
 {
   template<typename MatrixType, typename TranspositionType, typename Workspace>
-  static bool unblocked(MatrixType& mat, TranspositionType& transpositions, Workspace& temp, int* sign=0)
+  static bool unblocked(MatrixType& mat, TranspositionType& transpositions, Workspace& temp, SignMatrix& sign)
   {
+    using std::abs;
     typedef typename MatrixType::Scalar Scalar;
     typedef typename MatrixType::RealScalar RealScalar;
-    typedef typename MatrixType::Index Index;
-    ei_assert(mat.rows()==mat.cols());
+    typedef typename TranspositionType::StorageIndex IndexType;
+    eigen_assert(mat.rows()==mat.cols());
     const Index size = mat.rows();
 
     if (size <= 1)
     {
       transpositions.setIdentity();
-      if(sign)
-        *sign = ei_real(mat.coeff(0,0))>0 ? 1:-1;
+      if (numext::real(mat.coeff(0,0)) > 0) sign = PositiveSemiDef;
+      else if (numext::real(mat.coeff(0,0)) < 0) sign = NegativeSemiDef;
+      else sign = ZeroSign;
       return true;
     }
 
-    RealScalar cutoff = 0, biggest_in_corner;
-
     for (Index k = 0; k < size; ++k)
     {
       // Find largest diagonal element
       Index index_of_biggest_in_corner;
-      biggest_in_corner = mat.diagonal().tail(size-k).cwiseAbs().maxCoeff(&index_of_biggest_in_corner);
+      mat.diagonal().tail(size-k).cwiseAbs().maxCoeff(&index_of_biggest_in_corner);
       index_of_biggest_in_corner += k;
 
-      if(k == 0)
-      {
-        // 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.
-        cutoff = ei_abs(NumTraits<Scalar>::epsilon() * biggest_in_corner);
-
-        if(sign)
-          *sign = ei_real(mat.diagonal().coeff(index_of_biggest_in_corner)) > 0 ? 1 : -1;
-      }
-
-      // Finish early if the matrix is not full rank.
-      if(biggest_in_corner < cutoff)
-      {
-        for(Index i = k; i < size; i++) transpositions.coeffRef(i) = i;
-        break;
-      }
-
-      transpositions.coeffRef(k) = index_of_biggest_in_corner;
+      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
@@ -256,14 +288,14 @@ template<> struct ei_ldlt_inplace<Lower>
         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(int i=k+1;i<index_of_biggest_in_corner;++i)
+        for(Index i=k+1;i<index_of_biggest_in_corner;++i)
         {
           Scalar tmp = mat.coeffRef(i,k);
-          mat.coeffRef(i,k) = ei_conj(mat.coeffRef(index_of_biggest_in_corner,i));
-          mat.coeffRef(index_of_biggest_in_corner,i) = ei_conj(tmp);
+          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) = ei_conj(mat.coeff(index_of_biggest_in_corner,k));
+          mat.coeffRef(index_of_biggest_in_corner,k) = numext::conj(mat.coeff(index_of_biggest_in_corner,k));
       }
 
       // partition the matrix:
@@ -277,91 +309,218 @@ template<> struct ei_ldlt_inplace<Lower>
 
       if(k>0)
       {
-        temp.head(k) = mat.diagonal().head(k).asDiagonal() * A10.adjoint();
+        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);
       }
-      if((rs>0) && (ei_abs(mat.coeffRef(k,k)) > cutoff))
-        A21 /= mat.coeffRef(k,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));
+      if((rs>0) && (abs(realAkk) > RealScalar(0)))
+        A21 /= realAkk;
+
+      if (sign == PositiveSemiDef) {
+        if (realAkk < 0) sign = Indefinite;
+      } else if (sign == NegativeSemiDef) {
+        if (realAkk > 0) sign = Indefinite;
+      } else if (sign == ZeroSign) {
+        if (realAkk > 0) sign = PositiveSemiDef;
+        else if (realAkk < 0) sign = NegativeSemiDef;
+      }
     }
 
     return true;
   }
+
+  // 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 ei_ldlt_inplace<Upper>
+template<> struct ldlt_inplace<Upper>
 {
   template<typename MatrixType, typename TranspositionType, typename Workspace>
-  static EIGEN_STRONG_INLINE bool unblocked(MatrixType& mat, TranspositionType& transpositions, Workspace& temp, int* sign=0)
+  static EIGEN_STRONG_INLINE bool unblocked(MatrixType& mat, TranspositionType& transpositions, Workspace& temp, SignMatrix& sign)
   {
     Transpose<MatrixType> matt(mat);
-    return ei_ldlt_inplace<Lower>::unblocked(matt, transpositions, temp, sign);
+    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 TriangularView<MatrixType, UnitLower> MatrixL;
-  typedef TriangularView<typename MatrixType::AdjointReturnType, UnitUpper> MatrixU;
-  inline static MatrixL getL(const MatrixType& m) { return m; }
-  inline static MatrixU getU(const MatrixType& m) { return m.adjoint(); }
+  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 TriangularView<typename MatrixType::AdjointReturnType, UnitLower> MatrixL;
-  typedef TriangularView<MatrixType, UnitUpper> MatrixU;
-  inline static MatrixL getL(const MatrixType& m) { return m.adjoint(); }
-  inline static MatrixU getU(const MatrixType& m) { return m; }
+  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>
-LDLT<MatrixType,_UpLo>& LDLT<MatrixType,_UpLo>::compute(const MatrixType& a)
+template<typename InputType>
+LDLT<MatrixType,_UpLo>& LDLT<MatrixType,_UpLo>::compute(const EigenBase<InputType>& a)
 {
-  ei_assert(a.rows()==a.cols());
+  check_template_parameters();
+  
+  eigen_assert(a.rows()==a.cols());
   const Index size = a.rows();
 
-  m_matrix = a;
+  m_matrix = a.derived();
 
   m_transpositions.resize(size);
   m_isInitialized = false;
   m_temporary.resize(size);
+  m_sign = internal::ZeroSign;
 
-  ei_ldlt_inplace<UpLo>::unblocked(m_matrix, m_transpositions, m_temporary, &m_sign);
+  internal::ldlt_inplace<UpLo>::unblocked(m_matrix, m_transpositions, m_temporary, m_sign);
 
   m_isInitialized = true;
   return *this;
 }
 
-template<typename _MatrixType, int _UpLo, typename Rhs>
-struct ei_solve_retval<LDLT<_MatrixType,_UpLo>, Rhs>
-  : ei_solve_retval_base<LDLT<_MatrixType,_UpLo>, Rhs>
+/** 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 LDLT<_MatrixType,_UpLo> LDLTType;
-  EIGEN_MAKE_SOLVE_HELPERS(LDLTType,Rhs)
-
-  template<typename Dest> void evalTo(Dest& dst) const
+  typedef typename TranspositionType::StorageIndex IndexType;
+  const Index size = w.rows();
+  if (m_isInitialized)
   {
-    ei_assert(rhs().rows() == dec().matrixLDLT().rows());
-    // dst = P b
-    dst = dec().transpositionsP() * rhs();
-
-    // dst = L^-1 (P b)
-    dec().matrixL().solveInPlace(dst);
-
-    // dst = D^-1 (L^-1 P b)
-    dst = dec().vectorD().asDiagonal().inverse() * dst;
-
-    // dst = L^-T (D^-1 L^-1 P b)
-    dec().matrixU().solveInPlace(dst);
-
-    // dst = P^-1 (L^-T D^-1 L^-1 P b) = A^-1 b
-    dst = dec().transpositionsP().transpose() * dst;
+    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 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.
+  RealScalar tolerance = RealScalar(1) / NumTraits<RealScalar>::highest();
+  
+  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);
   *
@@ -380,9 +539,8 @@ template<typename MatrixType,int _UpLo>
 template<typename Derived>
 bool LDLT<MatrixType,_UpLo>::solveInPlace(MatrixBase<Derived> &bAndX) const
 {
-  ei_assert(m_isInitialized && "LDLT is not initialized.");
-  const Index size = m_matrix.rows();
-  ei_assert(size == bAndX.rows());
+  eigen_assert(m_isInitialized && "LDLT is not initialized.");
+  eigen_assert(m_matrix.rows() == bAndX.rows());
 
   bAndX = this->solve(bAndX);
 
@@ -395,7 +553,7 @@ bool LDLT<MatrixType,_UpLo>::solveInPlace(MatrixBase<Derived> &bAndX) const
 template<typename MatrixType, int _UpLo>
 MatrixType LDLT<MatrixType,_UpLo>::reconstructedMatrix() const
 {
-  ei_assert(m_isInitialized && "LDLT is not initialized.");
+  eigen_assert(m_isInitialized && "LDLT is not initialized.");
   const Index size = m_matrix.rows();
   MatrixType res(size,size);
 
@@ -405,7 +563,7 @@ MatrixType LDLT<MatrixType,_UpLo>::reconstructedMatrix() const
   // L^* P
   res = matrixU() * res;
   // D(L^*P)
-  res = vectorD().asDiagonal() * res;
+  res = vectorD().real().asDiagonal() * res;
   // L(DL^*P)
   res = matrixL() * res;
   // P^T (LDL^*P)
@@ -414,8 +572,10 @@ MatrixType LDLT<MatrixType,_UpLo>::reconstructedMatrix() const
   return res;
 }
 
+#ifndef __CUDACC__
 /** \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>
@@ -426,6 +586,7 @@ SelfAdjointView<MatrixType, UpLo>::ldlt() const
 
 /** \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>
@@ -433,5 +594,8 @@ MatrixBase<Derived>::ldlt() const
 {
   return LDLT<PlainObject>(derived());
 }
+#endif // __CUDACC__
+
+} // end namespace Eigen
 
 #endif // EIGEN_LDLT_H

Eigen/src/Cholesky/LLT.h

@@ -3,37 +3,28 @@
 //
 // 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
 
-template<typename MatrixType, int UpLo> struct LLT_Traits;
+namespace Eigen { 
 
-/** \ingroup cholesky_Module
+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
+  * \param 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.
@@ -47,7 +38,10 @@ template<typename MatrixType, int UpLo> struct LLT_Traits;
   * use LDLT instead for the semidefinite case. Also, do not use a Cholesky decomposition to determine whether a system of equations
   * has a solution.
   *
-  * \sa MatrixBase::llt(), class LDLT
+  * Example: \include LLT_example.cpp
+  * Output: \verbinclude LLT_example.out
+  *    
+  * \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,
@@ -65,15 +59,16 @@ template<typename _MatrixType, int _UpLo> class LLT
     };
     typedef typename MatrixType::Scalar Scalar;
     typedef typename NumTraits<typename MatrixType::Scalar>::Real RealScalar;
-    typedef typename MatrixType::Index Index;
+    typedef Eigen::Index Index; ///< \deprecated since Eigen 3.3
+    typedef typename MatrixType::StorageIndex StorageIndex;
 
     enum {
-      PacketSize = ei_packet_traits<Scalar>::size,
+      PacketSize = internal::packet_traits<Scalar>::size,
       AlignmentMask = int(PacketSize)-1,
       UpLo = _UpLo
     };
 
-    typedef LLT_Traits<MatrixType,UpLo> Traits;
+    typedef internal::LLT_Traits<MatrixType,UpLo> Traits;
 
     /**
       * \brief Default Constructor.
@@ -89,27 +84,28 @@ template<typename _MatrixType, int _UpLo> class LLT
       * according to the specified problem \a size.
       * \sa LLT()
       */
-    LLT(Index size) : m_matrix(size, size),
+    explicit LLT(Index size) : m_matrix(size, size),
                     m_isInitialized(false) {}
 
-    LLT(const MatrixType& matrix)
+    template<typename InputType>
+    explicit LLT(const EigenBase<InputType>& matrix)
       : m_matrix(matrix.rows(), matrix.cols()),
         m_isInitialized(false)
     {
-      compute(matrix);
+      compute(matrix.derived());
     }
 
     /** \returns a view of the upper triangular matrix U */
     inline typename Traits::MatrixU matrixU() const
     {
-      ei_assert(m_isInitialized && "LLT is not initialized.");
+      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
     {
-      ei_assert(m_isInitialized && "LLT is not initialized.");
+      eigen_assert(m_isInitialized && "LLT is not initialized.");
       return Traits::getL(m_matrix);
     }
 
@@ -121,22 +117,23 @@ template<typename _MatrixType, int _UpLo> class LLT
       * Example: \include LLT_solve.cpp
       * Output: \verbinclude LLT_solve.out
       *
-      * \sa solveInPlace(), MatrixBase::llt()
+      * \sa solveInPlace(), MatrixBase::llt(), SelfAdjointView::llt()
       */
     template<typename Rhs>
-    inline const ei_solve_retval<LLT, Rhs>
+    inline const Solve<LLT, Rhs>
     solve(const MatrixBase<Rhs>& b) const
     {
-      ei_assert(m_isInitialized && "LLT is not initialized.");
-      ei_assert(m_matrix.rows()==b.rows()
+      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 ei_solve_retval<LLT, Rhs>(*this, b.derived());
+      return Solve<LLT, Rhs>(*this, b.derived());
     }
 
     template<typename Derived>
     void solveInPlace(MatrixBase<Derived> &bAndX) const;
 
-    LLT& compute(const MatrixType& matrix);
+    template<typename InputType>
+    LLT& compute(const EigenBase<InputType>& matrix);
 
     /** \returns the LLT decomposition matrix
       *
@@ -144,7 +141,7 @@ template<typename _MatrixType, int _UpLo> class LLT
       */
     inline const MatrixType& matrixLLT() const
     {
-      ei_assert(m_isInitialized && "LLT is not initialized.");
+      eigen_assert(m_isInitialized && "LLT is not initialized.");
       return m_matrix;
     }
 
@@ -158,14 +155,29 @@ template<typename _MatrixType, int _UpLo> class LLT
       */
     ComputationInfo info() const
     {
-      ei_assert(m_isInitialized && "LLT is not initialized.");
+      eigen_assert(m_isInitialized && "LLT is not initialized.");
       return m_info;
     }
 
     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.
@@ -175,17 +187,89 @@ template<typename _MatrixType, int _UpLo> class LLT
     ComputationInfo m_info;
 };
 
-template<int UpLo> struct ei_llt_inplace;
+namespace internal {
 
-template<> struct ei_llt_inplace<Lower>
+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)
 {
-  template<typename MatrixType>
-  static bool unblocked(MatrixType& mat)
+  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)
   {
-    typedef typename MatrixType::Scalar Scalar;
-    typedef typename MatrixType::RealScalar RealScalar;
-    typedef typename MatrixType::Index Index;
-    ei_assert(mat.rows()==mat.cols());
+    // 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)
+    {
+      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;
+
+      RealScalar x = dj + swj2/beta;
+      if (x<=RealScalar(0))
+        return j;
+      RealScalar nLjj = sqrt(x);
+      mat.coeffRef(j,j) = nLjj;
+      beta += swj2/dj;
+
+      // 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)
     {
@@ -195,29 +279,28 @@ template<> struct ei_llt_inplace<Lower>
       Block<MatrixType,1,Dynamic> A10(mat,k,0,1,k);
       Block<MatrixType,Dynamic,Dynamic> A20(mat,k+1,0,rs,k);
 
-      RealScalar x = ei_real(mat.coeff(k,k));
-      if (k>0) x -= mat.row(k).head(k).squaredNorm();
+      RealScalar x = numext::real(mat.coeff(k,k));
+      if (k>0) x -= A10.squaredNorm();
       if (x<=RealScalar(0))
-        return false;
-      mat.coeffRef(k,k) = x = ei_sqrt(x);
+        return k;
+      mat.coeffRef(k,k) = x = sqrt(x);
       if (k>0 && rs>0) A21.noalias() -= A20 * A10.adjoint();
-      if (rs>0) A21 *= RealScalar(1)/x;
+      if (rs>0) A21 /= x;
     }
-    return true;
+    return -1;
   }
 
   template<typename MatrixType>
-  static bool blocked(MatrixType& m)
+  static Index blocked(MatrixType& m)
   {
-    typedef typename MatrixType::Index Index;
-    ei_assert(m.rows()==m.cols());
+    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));
+    blockSize = (std::min)((std::max)(blockSize,Index(8)), Index(128));
 
     for (Index k=0; k<size; k+=blockSize)
     {
@@ -225,68 +308,90 @@ template<> struct ei_llt_inplace<Lower>
       //       A00 |  -  |  -
       // lu  = A10 | A11 |  -
       //       A20 | A21 | A22
-      Index bs = std::min(blockSize, size-k);
+      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);
 
-      if(!unblocked(A11)) return false;
+      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,-1); // bottleneck
     }
-    return true;
+    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<> struct ei_llt_inplace<Upper>
+  
+template<typename Scalar> struct llt_inplace<Scalar, Upper>
 {
+  typedef typename NumTraits<Scalar>::Real RealScalar;
+
   template<typename MatrixType>
-  static EIGEN_STRONG_INLINE bool unblocked(MatrixType& mat)
+  static EIGEN_STRONG_INLINE Index unblocked(MatrixType& mat)
   {
     Transpose<MatrixType> matt(mat);
-    return ei_llt_inplace<Lower>::unblocked(matt);
+    return llt_inplace<Scalar, Lower>::unblocked(matt);
   }
   template<typename MatrixType>
-  static EIGEN_STRONG_INLINE bool blocked(MatrixType& mat)
+  static EIGEN_STRONG_INLINE Index blocked(MatrixType& mat)
   {
     Transpose<MatrixType> matt(mat);
-    return ei_llt_inplace<Lower>::blocked(matt);
+    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 TriangularView<MatrixType, Lower> MatrixL;
-  typedef TriangularView<typename MatrixType::AdjointReturnType, Upper> MatrixU;
-  inline static MatrixL getL(const MatrixType& m) { return m; }
-  inline static MatrixU getU(const MatrixType& m) { return m.adjoint(); }
+  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 ei_llt_inplace<Lower>::blocked(m); }
+  { return llt_inplace<typename MatrixType::Scalar, Lower>::blocked(m)==-1; }
 };
 
 template<typename MatrixType> struct LLT_Traits<MatrixType,Upper>
 {
-  typedef TriangularView<typename MatrixType::AdjointReturnType, Lower> MatrixL;
-  typedef TriangularView<MatrixType, Upper> MatrixU;
-  inline static MatrixL getL(const MatrixType& m) { return m.adjoint(); }
-  inline static MatrixU getU(const MatrixType& m) { return m; }
+  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 ei_llt_inplace<Upper>::blocked(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>
-LLT<MatrixType,_UpLo>& LLT<MatrixType,_UpLo>::compute(const MatrixType& a)
+template<typename InputType>
+LLT<MatrixType,_UpLo>& LLT<MatrixType,_UpLo>::compute(const EigenBase<InputType>& a)
 {
-  assert(a.rows()==a.cols());
+  check_template_parameters();
+  
+  eigen_assert(a.rows()==a.cols());
   const Index size = a.rows();
   m_matrix.resize(size, size);
-  m_matrix = a;
+  m_matrix = a.derived();
 
   m_isInitialized = true;
   bool ok = Traits::inplace_decomposition(m_matrix);
@@ -295,19 +400,35 @@ LLT<MatrixType,_UpLo>& LLT<MatrixType,_UpLo>::compute(const MatrixType& a)
   return *this;
 }
 
-template<typename _MatrixType, int UpLo, typename Rhs>
-struct ei_solve_retval<LLT<_MatrixType, UpLo>, Rhs>
-  : ei_solve_retval_base<LLT<_MatrixType, UpLo>, Rhs>
+/** 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, int _UpLo>
+template<typename VectorType>
+LLT<_MatrixType,_UpLo> LLT<_MatrixType,_UpLo>::rankUpdate(const VectorType& v, const RealScalar& sigma)
 {
-  typedef LLT<_MatrixType,UpLo> LLTType;
-  EIGEN_MAKE_SOLVE_HELPERS(LLTType,Rhs)
+  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;
 
-  template<typename Dest> void evalTo(Dest& dst) const
-  {
-    dst = rhs();
-    dec().solveInPlace(dst);
-  }
-};
+  return *this;
+}
+ 
+#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);
   * 
@@ -326,8 +447,8 @@ template<typename MatrixType, int _UpLo>
 template<typename Derived>
 void LLT<MatrixType,_UpLo>::solveInPlace(MatrixBase<Derived> &bAndX) const
 {
-  ei_assert(m_isInitialized && "LLT is not initialized.");
-  ei_assert(m_matrix.rows()==bAndX.rows());
+  eigen_assert(m_isInitialized && "LLT is not initialized.");
+  eigen_assert(m_matrix.rows()==bAndX.rows());
   matrixL().solveInPlace(bAndX);
   matrixU().solveInPlace(bAndX);
 }
@@ -338,12 +459,14 @@ void LLT<MatrixType,_UpLo>::solveInPlace(MatrixBase<Derived> &bAndX) const
 template<typename MatrixType, int _UpLo>
 MatrixType LLT<MatrixType,_UpLo>::reconstructedMatrix() const
 {
-  ei_assert(m_isInitialized && "LLT is not initialized.");
+  eigen_assert(m_isInitialized && "LLT is not initialized.");
   return matrixL() * matrixL().adjoint().toDenseMatrix();
 }
 
+#ifndef __CUDACC__
 /** \cholesky_module
   * \returns the LLT decomposition of \c *this
+  * \sa SelfAdjointView::llt()
   */
 template<typename Derived>
 inline const LLT<typename MatrixBase<Derived>::PlainObject>
@@ -354,6 +477,7 @@ MatrixBase<Derived>::llt() const
 
 /** \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>
@@ -361,5 +485,8 @@ SelfAdjointView<MatrixType, UpLo>::llt() const
 {
   return LLT<PlainObject,UpLo>(m_matrix);
 }
+#endif // __CUDACC__
+  
+} // end namespace Eigen
 
 #endif // EIGEN_LLT_H

Eigen/src/Cholesky/LLT_MKL.h

@@ -0,0 +1,102 @@
+/*
+ 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 Intel(R) MKL
+ *     LLt decomposition based on LAPACKE_?potrf function.
+ ********************************************************************************
+*/
+
+#ifndef EIGEN_LLT_MKL_H
+#define EIGEN_LLT_MKL_H
+
+#include "Eigen/src/Core/util/MKL_support.h"
+#include <iostream>
+
+namespace Eigen { 
+
+namespace internal {
+
+template<typename Scalar> struct mkl_llt;
+
+#define EIGEN_MKL_LLT(EIGTYPE, MKLTYPE, MKLPREFIX) \
+template<> struct mkl_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 = m.rows(); \
+    StorageOrder = MatrixType::Flags&RowMajorBit?RowMajor:ColMajor; \
+    matrix_order = StorageOrder==RowMajor ? LAPACK_ROW_MAJOR : LAPACK_COL_MAJOR; \
+    a = &(m.coeffRef(0,0)); \
+    lda = m.outerStride(); \
+\
+    info = LAPACKE_##MKLPREFIX##potrf( matrix_order, uplo, size, (MKLTYPE*)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 mkl_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 mkl_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_MKL_LLT(double, double, d)
+EIGEN_MKL_LLT(float, float, s)
+EIGEN_MKL_LLT(dcomplex, MKL_Complex16, z)
+EIGEN_MKL_LLT(scomplex, MKL_Complex8, c)
+
+} // end namespace internal
+
+} // end namespace Eigen
+
+#endif // EIGEN_LLT_MKL_H

Eigen/src/CholmodSupport/CMakeLists.txt

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

Eigen/src/CholmodSupport/CholmodSupport.h

@@ -0,0 +1,563 @@
+// 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, typename CholmodType>
+void cholmod_configure_matrix(CholmodType& mat)
+{
+  if (internal::is_same<Scalar,float>::value)
+  {
+    mat.xtype = CHOLMOD_REAL;
+    mat.dtype = CHOLMOD_SINGLE;
+  }
+  else if (internal::is_same<Scalar,double>::value)
+  {
+    mat.xtype = CHOLMOD_REAL;
+    mat.dtype = CHOLMOD_DOUBLE;
+  }
+  else if (internal::is_same<Scalar,std::complex<float> >::value)
+  {
+    mat.xtype = CHOLMOD_COMPLEX;
+    mat.dtype = CHOLMOD_SINGLE;
+  }
+  else if (internal::is_same<Scalar,std::complex<double> >::value)
+  {
+    mat.xtype = CHOLMOD_COMPLEX;
+    mat.dtype = CHOLMOD_DOUBLE;
+  }
+  else
+  {
+    eigen_assert(false && "Scalar type not supported by CHOLMOD");
+  }
+}
+
+} // 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(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,SuiteSparse_long>::value)
+  {
+    res.itype = CHOLMOD_LONG;
+  }
+  else
+  {
+    eigen_assert(false && "Index type not supported yet");
+  }
+
+  // setup res.xtype
+  internal::cholmod_configure_matrix<_Scalar>(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(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(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>(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_shiftOffset[0] = m_shiftOffset[1] = RealScalar(0.0);
+      cholmod_start(&m_cholmod);
+    }
+
+    explicit CholmodBase(const MatrixType& matrix)
+      : m_cholmodFactor(0), m_info(Success)
+    {
+      m_shiftOffset[0] = m_shiftOffset[1] = RealScalar(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());
+
+      // note: cd stands for Cholmod Dense
+      Rhs& b_ref(b.const_cast_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 RhsScalar, int RhsOptions, typename RhsIndex, typename DestScalar, int DestOptions, typename DestIndex>
+    void _solve_impl(const SparseMatrix<RhsScalar,RhsOptions,RhsIndex> &b, SparseMatrix<DestScalar,DestOptions,DestIndex> &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
+      cholmod_sparse b_cs = viewAsCholmod(b);
+      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 = viewAsEigen<DestScalar,DestOptions,DestIndex>(*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] = offset;
+      return derived();
+    }
+    
+    template<typename Stream>
+    void dumpMemory(Stream& /*s*/)
+    {}
+    
+  protected:
+    mutable cholmod_common m_cholmod;
+    cholmod_factor* m_cholmodFactor;
+    RealScalar 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.
+  *
+  * \sa \ref TutorialSparseDirectSolvers, 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.
+  *
+  * \sa \ref TutorialSparseDirectSolvers, 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.
+  *
+  * \sa \ref TutorialSparseDirectSolvers
+  */
+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.
+  *
+  * \sa \ref TutorialSparseDirectSolvers
+  */
+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

@@ -3,28 +3,15 @@
 //
 // Copyright (C) 2009 Gael Guennebaud <gael.guennebaud@inria.fr>
 //
-// Eigen is free software; you can redistribute it and/or
-// modify it under the terms of the GNU Lesser General Public
-// License as published by the Free Software Foundation; either
-// version 3 of the License, or (at your option) any later version.
-//
-// Alternatively, you can redistribute it and/or
-// modify it under the terms of the GNU General Public License as
-// published by the Free Software Foundation; either version 2 of
-// the License, or (at your option) any later version.
-//
-// Eigen is distributed in the hope that it will be useful, but WITHOUT ANY
-// WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS
-// FOR A PARTICULAR PURPOSE. See the GNU Lesser General Public License or the
-// GNU General Public License for more details.
-//
-// You should have received a copy of the GNU Lesser General Public
-// License and a copy of the GNU General Public License along with
-// Eigen. If not, see <http://www.gnu.org/licenses/>.
+// 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 {
+
 /** \class Array 
   * \ingroup Core_Module
   *
@@ -37,22 +24,30 @@
   * 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 TopicCustomizingEigen by defining the preprocessor symbol \c EIGEN_ARRAY_PLUGIN.
+  *
   * \sa \ref TutorialArrayClass, \ref TopicClassHierarchy
   */
+namespace internal {
 template<typename _Scalar, int _Rows, int _Cols, int _Options, int _MaxRows, int _MaxCols>
-struct ei_traits<Array<_Scalar, _Rows, _Cols, _Options, _MaxRows, _MaxCols> > : ei_traits<Matrix<_Scalar, _Rows, _Cols, _Options, _MaxRows, _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;
 };
+}
 
 template<typename _Scalar, int _Rows, int _Cols, int _Options, int _MaxRows, int _MaxCols>
 class Array
-  : public DenseStorageBase<Array<_Scalar, _Rows, _Cols, _Options, _MaxRows, _MaxCols> >
+  : public PlainObjectBase<Array<_Scalar, _Rows, _Cols, _Options, _MaxRows, _MaxCols> >
 {
   public:
 
-    typedef DenseStorageBase<Array> Base;
+    typedef PlainObjectBase<Array> Base;
     EIGEN_DENSE_PUBLIC_INTERFACE(Array)
 
     enum { Options = _Options };
@@ -60,13 +55,11 @@ class Array
 
   protected:
     template <typename Derived, typename OtherDerived, bool IsVector>
-    friend struct ei_conservative_resize_like_impl;
+    friend struct internal::conservative_resize_like_impl;
 
     using Base::m_storage;
+
   public:
-    enum { NeedsToAlign = (!(Options&DontAlign))
-                          && SizeAtCompileTime!=Dynamic && ((static_cast<int>(sizeof(Scalar))*SizeAtCompileTime)%16)==0 };
-    EIGEN_MAKE_ALIGNED_OPERATOR_NEW_IF(NeedsToAlign)
 
     using Base::base;
     using Base::coeff;
@@ -79,11 +72,27 @@ class Array
       * 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),
@@ -94,7 +103,8 @@ class Array
       * remain row-vectors and vectors remain vectors.
       */
     template<typename OtherDerived>
-    EIGEN_STRONG_INLINE Array& operator=(const ArrayBase<OtherDerived>& other)
+    EIGEN_DEVICE_FUNC
+    EIGEN_STRONG_INLINE Array& operator=(const DenseBase<OtherDerived>& other)
     {
       return Base::_set(other);
     }
@@ -102,11 +112,12 @@ class Array
     /** 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.
@@ -117,122 +128,118 @@ class Array
       *
       * \sa resize(Index,Index)
       */
-    EIGEN_STRONG_INLINE explicit Array() : Base()
+    EIGEN_DEVICE_FUNC
+    EIGEN_STRONG_INLINE Array() : Base()
     {
       Base::_check_template_params();
-      EIGEN_INITIALIZE_BY_ZERO_IF_THAT_OPTION_IS_ENABLED
+      EIGEN_INITIALIZE_COEFFS_IF_THAT_OPTION_IS_ENABLED
     }
 
 #ifndef EIGEN_PARSED_BY_DOXYGEN
     // FIXME is it still needed ??
     /** \internal */
-    Array(ei_constructor_without_unaligned_array_assert)
-      : Base(ei_constructor_without_unaligned_array_assert())
+    EIGEN_DEVICE_FUNC
+    Array(internal::constructor_without_unaligned_array_assert)
+      : Base(internal::constructor_without_unaligned_array_assert())
     {
       Base::_check_template_params();
-      EIGEN_INITIALIZE_BY_ZERO_IF_THAT_OPTION_IS_ENABLED
+      EIGEN_INITIALIZE_COEFFS_IF_THAT_OPTION_IS_ENABLED
     }
 #endif
 
+#ifdef EIGEN_HAVE_RVALUE_REFERENCES
+    EIGEN_DEVICE_FUNC
+    Array(Array&& other)
+      : Base(std::move(other))
+    {
+      Base::_check_template_params();
+      if (RowsAtCompileTime!=Dynamic && ColsAtCompileTime!=Dynamic)
+        Base::_set_noalias(other);
+    }
+    EIGEN_DEVICE_FUNC
+    Array& operator=(Array&& other)
+    {
+      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 Matrix() instead.
+      * constructor Array() instead.
       */
-    EIGEN_STRONG_INLINE explicit Array(Index dim)
-      : Base(dim, RowsAtCompileTime == 1 ? 1 : dim, ColsAtCompileTime == 1 ? 1 : dim)
-    {
-      Base::_check_template_params();
-      EIGEN_STATIC_ASSERT_VECTOR_ONLY(Array)
-      ei_assert(dim > 0);
-      ei_assert(SizeAtCompileTime == Dynamic || SizeAtCompileTime == dim);
-      EIGEN_INITIALIZE_BY_ZERO_IF_THAT_OPTION_IS_ENABLED
-    }
-
-    #ifndef EIGEN_PARSED_BY_DOXYGEN
-    template<typename T0, typename T1>
-    EIGEN_STRONG_INLINE Array(const T0& x, const T1& y)
-    {
-      Base::_check_template_params();
-      this->template _init2<T0,T1>(x, y);
-    }
-    #else
-    /** constructs an uninitialized matrix with \a rows rows and \a cols columns.
+    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 matrices. For fixed-size matrices,
+      * 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
-      * Matrix() instead. */
+      * Array() instead. */
     Array(Index rows, Index cols);
     /** constructs an initialized 2D vector with given coefficients */
-    Array(const Scalar& x, const Scalar& y);
+    Array(const Scalar& val0, const Scalar& val1);
     #endif
 
     /** constructs an initialized 3D vector with given coefficients */
-    EIGEN_STRONG_INLINE Array(const Scalar& x, const Scalar& y, const Scalar& z)
+    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] = x;
-      m_storage.data()[1] = y;
-      m_storage.data()[2] = z;
+      m_storage.data()[0] = val0;
+      m_storage.data()[1] = val1;
+      m_storage.data()[2] = val2;
     }
     /** constructs an initialized 4D vector with given coefficients */
-    EIGEN_STRONG_INLINE Array(const Scalar& x, const Scalar& y, const Scalar& z, const Scalar& w)
+    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] = x;
-      m_storage.data()[1] = y;
-      m_storage.data()[2] = z;
-      m_storage.data()[3] = w;
+      m_storage.data()[0] = val0;
+      m_storage.data()[1] = val1;
+      m_storage.data()[2] = val2;
+      m_storage.data()[3] = val3;
     }
 
-    explicit Array(const Scalar *data);
-
-    /** Constructor copying the value of the expression \a other */
-    template<typename OtherDerived>
-    EIGEN_STRONG_INLINE Array(const ArrayBase<OtherDerived>& other)
-             : Base(other.rows() * other.cols(), other.rows(), other.cols())
-    {
-      Base::_check_template_params();
-      Base::_set_noalias(other);
-    }
     /** Copy constructor */
+    EIGEN_DEVICE_FUNC
     EIGEN_STRONG_INLINE Array(const Array& other)
-            : Base(other.rows() * other.cols(), other.rows(), other.cols())
-    {
-      Base::_check_template_params();
-      Base::_set_noalias(other);
-    }
-    /** Copy constructor with in-place evaluation */
-    template<typename OtherDerived>
-    EIGEN_STRONG_INLINE Array(const ReturnByValue<OtherDerived>& other)
-    {
-      Base::_check_template_params();
-      Base::resize(other.rows(), other.cols());
-      other.evalTo(*this);
-    }
+            : Base(other)
+    { }
 
     /** \sa MatrixBase::operator=(const EigenBase<OtherDerived>&) */
     template<typename OtherDerived>
+    EIGEN_DEVICE_FUNC
     EIGEN_STRONG_INLINE Array(const EigenBase<OtherDerived> &other)
-      : Base(other.derived().rows() * other.derived().cols(), other.derived().rows(), other.derived().cols())
-    {
-      Base::_check_template_params();
-      Base::resize(other.rows(), other.cols());
-      *this = other;
-    }
+      : Base(other.derived())
+    { }
 
-    /** Override MatrixBase::swap() since for dynamic-sized matrices of same type it is enough to swap the
-      * data pointers.
-      */
-    template<typename OtherDerived>
-    void swap(ArrayBase<OtherDerived> EIGEN_REF_TO_TEMPORARY other)
-    { this->_swap(other.derived()); }
-
-    inline Index innerStride() const { return 1; }
-    inline Index outerStride() const { return this->innerSize(); }
+    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
@@ -241,7 +248,7 @@ class Array
   private:
 
     template<typename MatrixType, typename OtherDerived, bool SwapPointers>
-    friend struct ei_matrix_swap_impl;
+    friend struct internal::matrix_swap_impl;
 };
 
 /** \defgroup arraytypedefs Global array typedefs
@@ -313,5 +320,6 @@ 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

@@ -3,28 +3,15 @@
 //
 // Copyright (C) 2009 Gael Guennebaud <gael.guennebaud@inria.fr>
 //
-// Eigen is free software; you can redistribute it and/or
-// modify it under the terms of the GNU Lesser General Public
-// License as published by the Free Software Foundation; either
-// version 3 of the License, or (at your option) any later version.
-//
-// Alternatively, you can redistribute it and/or
-// modify it under the terms of the GNU General Public License as
-// published by the Free Software Foundation; either version 2 of
-// the License, or (at your option) any later version.
-//
-// Eigen is distributed in the hope that it will be useful, but WITHOUT ANY
-// WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS
-// FOR A PARTICULAR PURPOSE. See the GNU Lesser General Public License or the
-// GNU General Public License for more details.
-//
-// You should have received a copy of the GNU Lesser General Public
-// License and a copy of the GNU General Public License along with
-// Eigen. If not, see <http://www.gnu.org/licenses/>.
+// 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
@@ -42,7 +29,10 @@ template<typename ExpressionType> class MatrixWrapper;
   *
   * This class is the base that is inherited by all array expression types.
   *
-  * \param Derived is the derived type, e.g., an array or an expression type.
+  * \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 TopicCustomizingEigen by defining the preprocessor symbol \c EIGEN_ARRAYBASE_PLUGIN.
   *
   * \sa class MatrixBase, \ref TopicClassHierarchy
   */
@@ -53,19 +43,17 @@ template<typename Derived> class ArrayBase
 #ifndef EIGEN_PARSED_BY_DOXYGEN
     /** The base class for a given storage type. */
     typedef ArrayBase StorageBaseType;
-    
+
     typedef ArrayBase Eigen_BaseClassForSpecializationOfGlobalMathFuncImpl;
 
-    using ei_special_scalar_op_base<Derived,typename ei_traits<Derived>::Scalar,
-                typename NumTraits<typename ei_traits<Derived>::Scalar>::Real>::operator*;
-
-    typedef typename ei_traits<Derived>::StorageKind StorageKind;
-    typedef typename ei_traits<Derived>::Index Index;
-    typedef typename ei_traits<Derived>::Scalar Scalar;
-    typedef typename ei_packet_traits<Scalar>::type PacketScalar;
+    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::operator*;
+    using Base::operator/;
     using Base::RowsAtCompileTime;
     using Base::ColsAtCompileTime;
     using Base::SizeAtCompileTime;
@@ -74,8 +62,7 @@ template<typename Derived> class ArrayBase
     using Base::MaxSizeAtCompileTime;
     using Base::IsVectorAtCompileTime;
     using Base::Flags;
-    using Base::CoeffReadCost;
-
+    
     using Base::derived;
     using Base::const_cast_derived;
     using Base::rows;
@@ -91,25 +78,14 @@ template<typename Derived> class ArrayBase
     using Base::operator/=;
 
     typedef typename Base::CoeffReturnType CoeffReturnType;
+
 #endif // not EIGEN_PARSED_BY_DOXYGEN
 
 #ifndef EIGEN_PARSED_BY_DOXYGEN
-    /** \internal the plain matrix type corresponding to this expression. Note that 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 Array<typename ei_traits<Derived>::Scalar,
-                ei_traits<Derived>::RowsAtCompileTime,
-                ei_traits<Derived>::ColsAtCompileTime,
-                AutoAlign | (ei_traits<Derived>::Flags&RowMajorBit ? RowMajor : ColMajor),
-                ei_traits<Derived>::MaxRowsAtCompileTime,
-                ei_traits<Derived>::MaxColsAtCompileTime
-          > PlainObject;
-
+    typedef typename Base::PlainObject PlainObject;
 
     /** \internal Represents a matrix with all coefficients equal to one another*/
-    typedef CwiseNullaryOp<ei_scalar_constant_op<Scalar>,Derived> ConstantReturnType;
+    typedef CwiseNullaryOp<internal::scalar_constant_op<Scalar>,PlainObject> ConstantReturnType;
 #endif // not EIGEN_PARSED_BY_DOXYGEN
 
 #define EIGEN_CURRENT_STORAGE_BASE_CLASS Eigen::ArrayBase
@@ -127,40 +103,57 @@ template<typename Derived> class ArrayBase
     /** 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
     Derived& operator=(const ArrayBase& other)
     {
-      return ei_assign_selector<Derived,Derived>::run(derived(), other.derived());
+      internal::call_assignment(derived(), other.derived());
+      return derived();
     }
+    
+    /** Set all the entries to \a value.
+      * \sa DenseBase::setConstant(), DenseBase::fill() */
+    EIGEN_DEVICE_FUNC
+    Derived& operator=(const Scalar &value)
+    { Base::setConstant(value); return derived(); }
 
-    Derived& operator+=(const Scalar& scalar)
-    { return *this = derived() + scalar; }
-    Derived& operator-=(const Scalar& scalar)
-    { return *this = derived() - scalar; }
+    EIGEN_DEVICE_FUNC
+    Derived& operator+=(const Scalar& scalar);
+    EIGEN_DEVICE_FUNC
+    Derived& operator-=(const Scalar& scalar);
 
     template<typename OtherDerived>
+    EIGEN_DEVICE_FUNC
     Derived& operator+=(const ArrayBase<OtherDerived>& other);
     template<typename OtherDerived>
+    EIGEN_DEVICE_FUNC
     Derived& operator-=(const ArrayBase<OtherDerived>& other);
 
     template<typename OtherDerived>
+    EIGEN_DEVICE_FUNC
     Derived& operator*=(const ArrayBase<OtherDerived>& other);
 
     template<typename OtherDerived>
+    EIGEN_DEVICE_FUNC
     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 MatrixBase Matrix \endlink expression of this array
+    /** \returns an \link Eigen::MatrixBase Matrix \endlink expression of this array
       * \sa MatrixBase::array() */
-    MatrixWrapper<Derived> matrix() { return derived(); }
-    const MatrixWrapper<Derived> matrix() const { return derived(); }
+    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_DEVICE_FUNC
     ArrayBase() : Base() {}
 
   private:
@@ -169,11 +162,11 @@ template<typename Derived> class ArrayBase
     template<typename OtherDerived> explicit ArrayBase(const ArrayBase<OtherDerived>&);
   protected:
     // mixing arrays and matrices is not legal
-    template<typename OtherDerived> Derived& operator+=(const MatrixBase<OtherDerived>& mat)
-    {EIGEN_STATIC_ASSERT(sizeof(typename OtherDerived::Scalar)==-1,YOU_CANNOT_MIX_ARRAYS_AND_MATRICES);}
+    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>& mat)
-    {EIGEN_STATIC_ASSERT(sizeof(typename OtherDerived::Scalar)==-1,YOU_CANNOT_MIX_ARRAYS_AND_MATRICES);}
+    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.
@@ -185,8 +178,7 @@ template<typename OtherDerived>
 EIGEN_STRONG_INLINE Derived &
 ArrayBase<Derived>::operator-=(const ArrayBase<OtherDerived> &other)
 {
-  SelfCwiseBinaryOp<ei_scalar_difference_op<Scalar>, Derived, OtherDerived> tmp(derived());
-  tmp = other;
+  call_assignment(derived(), other.derived(), internal::sub_assign_op<Scalar>());
   return derived();
 }
 
@@ -199,8 +191,7 @@ template<typename OtherDerived>
 EIGEN_STRONG_INLINE Derived &
 ArrayBase<Derived>::operator+=(const ArrayBase<OtherDerived>& other)
 {
-  SelfCwiseBinaryOp<ei_scalar_sum_op<Scalar>, Derived, OtherDerived> tmp(derived());
-  tmp = other.derived();
+  call_assignment(derived(), other.derived(), internal::add_assign_op<Scalar>());
   return derived();
 }
 
@@ -213,8 +204,7 @@ template<typename OtherDerived>
 EIGEN_STRONG_INLINE Derived &
 ArrayBase<Derived>::operator*=(const ArrayBase<OtherDerived>& other)
 {
-  SelfCwiseBinaryOp<ei_scalar_product_op<Scalar>, Derived, OtherDerived> tmp(derived());
-  tmp = other.derived();
+  call_assignment(derived(), other.derived(), internal::mul_assign_op<Scalar,typename OtherDerived::Scalar>());
   return derived();
 }
 
@@ -227,9 +217,10 @@ template<typename OtherDerived>
 EIGEN_STRONG_INLINE Derived &
 ArrayBase<Derived>::operator/=(const ArrayBase<OtherDerived>& other)
 {
-  SelfCwiseBinaryOp<ei_scalar_quotient_op<Scalar>, Derived, OtherDerived> tmp(derived());
-  tmp = other.derived();
+  call_assignment(derived(), other.derived(), internal::div_assign_op<Scalar>());
   return derived();
 }
 
+} // end namespace Eigen
+
 #endif // EIGEN_ARRAYBASE_H

Eigen/src/Core/ArrayWrapper.h

@@ -3,28 +3,15 @@
 //
 // Copyright (C) 2009-2010 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_ARRAYWRAPPER_H
 #define EIGEN_ARRAYWRAPPER_H
 
+namespace Eigen { 
+
 /** \class ArrayWrapper
   * \ingroup Core_Module
   *
@@ -35,12 +22,20 @@
   *
   * \sa MatrixBase::array(), class MatrixWrapper
   */
+
+namespace internal {
 template<typename ExpressionType>
-struct ei_traits<ArrayWrapper<ExpressionType> >
-  : public ei_traits<typename ei_cleantype<typename ExpressionType::Nested>::type >
+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,
+    Flags = Flags0 & ~NestByRefBit
+  };
 };
+}
 
 template<typename ExpressionType>
 class ArrayWrapper : public ArrayBase<ArrayWrapper<ExpressionType> >
@@ -49,46 +44,79 @@ class ArrayWrapper : public ArrayBase<ArrayWrapper<ExpressionType> >
     typedef ArrayBase<ArrayWrapper> Base;
     EIGEN_DENSE_PUBLIC_INTERFACE(ArrayWrapper)
     EIGEN_INHERIT_ASSIGNMENT_OPERATORS(ArrayWrapper)
+    typedef typename internal::remove_all<ExpressionType>::type NestedExpression;
 
-    typedef typename ei_nested<ExpressionType>::type NestedExpressionType;
+    typedef typename internal::conditional<
+                       internal::is_lvalue<ExpressionType>::value,
+                       Scalar,
+                       const Scalar
+                     >::type ScalarWithConstIfNotLvalue;
 
-    inline ArrayWrapper(const ExpressionType& matrix) : m_expression(matrix) {}
+    typedef typename internal::ref_selector<ExpressionType>::type NestedExpressionType;
 
+    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(); }
 
-    inline const CoeffReturnType coeff(Index row, Index col) const
+    EIGEN_DEVICE_FUNC
+    inline ScalarWithConstIfNotLvalue* data() { return m_expression.const_cast_derived().data(); }
+    EIGEN_DEVICE_FUNC
+    inline const Scalar* data() const { return m_expression.data(); }
+
+    EIGEN_DEVICE_FUNC
+    inline CoeffReturnType coeff(Index rowId, Index colId) const
     {
-      return m_expression.coeff(row, col);
+      return m_expression.coeff(rowId, colId);
     }
 
-    inline Scalar& coeffRef(Index row, Index col)
+    EIGEN_DEVICE_FUNC
+    inline Scalar& coeffRef(Index rowId, Index colId)
     {
-      return m_expression.const_cast_derived().coeffRef(row, col);
+      return m_expression.const_cast_derived().coeffRef(rowId, colId);
     }
 
-    inline const CoeffReturnType coeff(Index index) const
+    EIGEN_DEVICE_FUNC
+    inline const Scalar& coeffRef(Index rowId, Index colId) const
+    {
+      return m_expression.const_cast_derived().coeffRef(rowId, colId);
+    }
+
+    EIGEN_DEVICE_FUNC
+    inline 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
+    EIGEN_DEVICE_FUNC
+    inline const Scalar& coeffRef(Index index) const
     {
-      return m_expression.template packet<LoadMode>(row, col);
+      return m_expression.const_cast_derived().coeffRef(index);
     }
 
     template<int LoadMode>
-    inline void writePacket(Index row, Index col, const PacketScalar& x)
+    inline const PacketScalar packet(Index rowId, Index colId) const
     {
-      m_expression.const_cast_derived().template writePacket<LoadMode>(row, col, x);
+      return m_expression.template packet<LoadMode>(rowId, colId);
+    }
+
+    template<int LoadMode>
+    inline void writePacket(Index rowId, Index colId, const PacketScalar& val)
+    {
+      m_expression.const_cast_derived().template writePacket<LoadMode>(rowId, colId, val);
     }
 
     template<int LoadMode>
@@ -98,16 +126,33 @@ class ArrayWrapper : public ArrayBase<ArrayWrapper<ExpressionType> >
     }
 
     template<int LoadMode>
-    inline void writePacket(Index index, const PacketScalar& x)
+    inline void writePacket(Index index, const PacketScalar& val)
     {
-      m_expression.const_cast_derived().template writePacket<LoadMode>(index, x);
+      m_expression.const_cast_derived().template writePacket<LoadMode>(index, val);
     }
 
     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.const_cast_derived().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.const_cast_derived().resize(rows,cols); }
+
   protected:
-    const NestedExpressionType m_expression;
+    NestedExpressionType m_expression;
 };
 
 /** \class MatrixWrapper
@@ -121,12 +166,19 @@ class ArrayWrapper : public ArrayBase<ArrayWrapper<ExpressionType> >
   * \sa MatrixBase::matrix(), class ArrayWrapper
   */
 
+namespace internal {
 template<typename ExpressionType>
-struct ei_traits<MatrixWrapper<ExpressionType> >
- : public ei_traits<typename ei_cleantype<typename ExpressionType::Nested>::type >
+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,
+    Flags = Flags0 & ~NestByRefBit
+  };
 };
+}
 
 template<typename ExpressionType>
 class MatrixWrapper : public MatrixBase<MatrixWrapper<ExpressionType> >
@@ -135,46 +187,79 @@ class MatrixWrapper : public MatrixBase<MatrixWrapper<ExpressionType> >
     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 ei_nested<ExpressionType>::type NestedExpressionType;
+    typedef typename internal::conditional<
+                       internal::is_lvalue<ExpressionType>::value,
+                       Scalar,
+                       const Scalar
+                     >::type ScalarWithConstIfNotLvalue;
 
-    inline MatrixWrapper(const ExpressionType& matrix) : m_expression(matrix) {}
+    typedef typename internal::ref_selector<ExpressionType>::type NestedExpressionType;
 
+    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(); }
 
-    inline const CoeffReturnType coeff(Index row, Index col) const
+    EIGEN_DEVICE_FUNC
+    inline ScalarWithConstIfNotLvalue* data() { return m_expression.const_cast_derived().data(); }
+    EIGEN_DEVICE_FUNC
+    inline const Scalar* data() const { return m_expression.data(); }
+
+    EIGEN_DEVICE_FUNC
+    inline CoeffReturnType coeff(Index rowId, Index colId) const
     {
-      return m_expression.coeff(row, col);
+      return m_expression.coeff(rowId, colId);
     }
 
-    inline Scalar& coeffRef(Index row, Index col)
+    EIGEN_DEVICE_FUNC
+    inline Scalar& coeffRef(Index rowId, Index colId)
     {
-      return m_expression.const_cast_derived().coeffRef(row, col);
+      return m_expression.const_cast_derived().coeffRef(rowId, colId);
     }
 
-    inline const CoeffReturnType coeff(Index index) const
+    EIGEN_DEVICE_FUNC
+    inline const Scalar& coeffRef(Index rowId, Index colId) const
+    {
+      return m_expression.derived().coeffRef(rowId, colId);
+    }
+
+    EIGEN_DEVICE_FUNC
+    inline 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
+    EIGEN_DEVICE_FUNC
+    inline const Scalar& coeffRef(Index index) const
     {
-      return m_expression.template packet<LoadMode>(row, col);
+      return m_expression.const_cast_derived().coeffRef(index);
     }
 
     template<int LoadMode>
-    inline void writePacket(Index row, Index col, const PacketScalar& x)
+    inline const PacketScalar packet(Index rowId, Index colId) const
     {
-      m_expression.const_cast_derived().template writePacket<LoadMode>(row, col, x);
+      return m_expression.template packet<LoadMode>(rowId, colId);
+    }
+
+    template<int LoadMode>
+    inline void writePacket(Index rowId, Index colId, const PacketScalar& val)
+    {
+      m_expression.const_cast_derived().template writePacket<LoadMode>(rowId, colId, val);
     }
 
     template<int LoadMode>
@@ -184,13 +269,31 @@ class MatrixWrapper : public MatrixBase<MatrixWrapper<ExpressionType> >
     }
 
     template<int LoadMode>
-    inline void writePacket(Index index, const PacketScalar& x)
+    inline void writePacket(Index index, const PacketScalar& val)
     {
-      m_expression.const_cast_derived().template writePacket<LoadMode>(index, x);
+      m_expression.const_cast_derived().template writePacket<LoadMode>(index, val);
     }
 
+    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.const_cast_derived().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.const_cast_derived().resize(rows,cols); }
+
   protected:
-    const NestedExpressionType m_expression;
+    NestedExpressionType m_expression;
 };
 
+} // end namespace Eigen
+
 #endif // EIGEN_ARRAYWRAPPER_H

Eigen/src/Core/Assign.h

@@ -5,487 +5,14 @@
 // 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 traversal and unrolling       *
-***************************************************************************/
-
-template <typename Derived, typename OtherDerived>
-struct ei_assign_traits
-{
-public:
-  enum {
-    DstIsAligned = Derived::Flags & AlignedBit,
-    DstHasDirectAccess = Derived::Flags & DirectAccessBit,
-    SrcIsAligned = OtherDerived::Flags & AlignedBit,
-    JointAlignment = DstIsAligned && SrcIsAligned ? Aligned : Unaligned
-  };
-
-private:
-  enum {
-    InnerSize = int(Derived::IsVectorAtCompileTime) ? int(Derived::SizeAtCompileTime)
-              : int(Derived::Flags)&RowMajorBit ? int(Derived::ColsAtCompileTime)
-              : int(Derived::RowsAtCompileTime),
-    InnerMaxSize = int(Derived::IsVectorAtCompileTime) ? int(Derived::MaxSizeAtCompileTime)
-              : int(Derived::Flags)&RowMajorBit ? int(Derived::MaxColsAtCompileTime)
-              : int(Derived::MaxRowsAtCompileTime),
-    MaxSizeAtCompileTime = Derived::SizeAtCompileTime,
-    PacketSize = ei_packet_traits<typename Derived::Scalar>::size
-  };
-
-  enum {
-    StorageOrdersAgree = (int(Derived::IsRowMajor) == int(OtherDerived::IsRowMajor)),
-    MightVectorize = StorageOrdersAgree
-                  && (int(Derived::Flags) & int(OtherDerived::Flags) & ActualPacketAccessBit),
-    MayInnerVectorize  = MightVectorize && int(InnerSize)!=Dynamic && int(InnerSize)%int(PacketSize)==0
-                       && int(DstIsAligned) && int(SrcIsAligned),
-    MayLinearize = StorageOrdersAgree && (int(Derived::Flags) & int(OtherDerived::Flags) & LinearAccessBit),
-    MayLinearVectorize = MightVectorize && MayLinearize && DstHasDirectAccess
-                       && (DstIsAligned || MaxSizeAtCompileTime == Dynamic),
-      /* If the destination isn't aligned, we have to do runtime checks and we don't unroll,
-         so it's only good for large enough sizes. */
-    MaySliceVectorize  = MightVectorize && DstHasDirectAccess
-                       && (int(InnerMaxSize)==Dynamic || int(InnerMaxSize)>=3*PacketSize)
-      /* slice vectorization can be slow, so we only want it if the slices are big, which is
-         indicated by InnerMaxSize rather than InnerSize, think of the case of a dynamic block
-         in a fixed-size matrix */
-  };
-
-public:
-  enum {
-    Traversal = int(MayInnerVectorize)  ? int(InnerVectorizedTraversal)
-              : int(MayLinearVectorize) ? int(LinearVectorizedTraversal)
-              : int(MaySliceVectorize)  ? int(SliceVectorizedTraversal)
-              : int(MayLinearize)       ? int(LinearTraversal)
-                                        : int(DefaultTraversal),
-    Vectorized = int(Traversal) == InnerVectorizedTraversal
-              || int(Traversal) == LinearVectorizedTraversal
-              || int(Traversal) == SliceVectorizedTraversal
-  };
-
-private:
-  enum {
-    UnrollingLimit      = EIGEN_UNROLLING_LIMIT * (Vectorized ? int(PacketSize) : 1),
-    MayUnrollCompletely = int(Derived::SizeAtCompileTime) != Dynamic
-                       && int(OtherDerived::CoeffReadCost) != Dynamic
-                       && int(Derived::SizeAtCompileTime) * int(OtherDerived::CoeffReadCost) <= int(UnrollingLimit),
-    MayUnrollInner      = int(InnerSize) != Dynamic
-                       && int(OtherDerived::CoeffReadCost) != Dynamic
-                       && int(InnerSize) * int(OtherDerived::CoeffReadCost) <= int(UnrollingLimit)
-  };
-
-public:
-  enum {
-    Unrolling = (int(Traversal) == int(InnerVectorizedTraversal) || int(Traversal) == int(DefaultTraversal))
-                ? (
-                    int(MayUnrollCompletely) ? int(CompleteUnrolling)
-                  : int(MayUnrollInner)      ? int(InnerUnrolling)
-                                             : int(NoUnrolling)
-                  )
-              : int(Traversal) == int(LinearVectorizedTraversal)
-                ? ( int(MayUnrollCompletely) && int(DstIsAligned) ? int(CompleteUnrolling) : int(NoUnrolling) )
-              : int(Traversal) == int(LinearTraversal)
-                ? ( int(MayUnrollCompletely) ? int(CompleteUnrolling) : int(NoUnrolling) )
-              : int(NoUnrolling)
-  };
-
-#ifdef EIGEN_DEBUG_ASSIGN
-  static void debug()
-  {
-    EIGEN_DEBUG_VAR(DstIsAligned)
-    EIGEN_DEBUG_VAR(SrcIsAligned)
-    EIGEN_DEBUG_VAR(JointAlignment)
-    EIGEN_DEBUG_VAR(InnerSize)
-    EIGEN_DEBUG_VAR(InnerMaxSize)
-    EIGEN_DEBUG_VAR(PacketSize)
-    EIGEN_DEBUG_VAR(StorageOrdersAgree)
-    EIGEN_DEBUG_VAR(MightVectorize)
-    EIGEN_DEBUG_VAR(MayLinearize)
-    EIGEN_DEBUG_VAR(MayInnerVectorize)
-    EIGEN_DEBUG_VAR(MayLinearVectorize)
-    EIGEN_DEBUG_VAR(MaySliceVectorize)
-    EIGEN_DEBUG_VAR(Traversal)
-    EIGEN_DEBUG_VAR(UnrollingLimit)
-    EIGEN_DEBUG_VAR(MayUnrollCompletely)
-    EIGEN_DEBUG_VAR(MayUnrollInner)
-    EIGEN_DEBUG_VAR(Unrolling)
-  }
-#endif
-};
-
-/***************************************************************************
-* Part 2 : meta-unrollers
-***************************************************************************/
-
-/************************
-*** Default traversal ***
-************************/
-
-template<typename Derived1, typename Derived2, int Index, int Stop>
-struct ei_assign_DefaultTraversal_CompleteUnrolling
-{
-  enum {
-    outer = Index / Derived1::InnerSizeAtCompileTime,
-    inner = Index % Derived1::InnerSizeAtCompileTime
-  };
-
-  EIGEN_STRONG_INLINE static void run(Derived1 &dst, const Derived2 &src)
-  {
-    dst.copyCoeffByOuterInner(outer, inner, src);
-    ei_assign_DefaultTraversal_CompleteUnrolling<Derived1, Derived2, Index+1, Stop>::run(dst, src);
-  }
-};
-
-template<typename Derived1, typename Derived2, int Stop>
-struct ei_assign_DefaultTraversal_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_DefaultTraversal_InnerUnrolling
-{
-  EIGEN_STRONG_INLINE static void run(Derived1 &dst, const Derived2 &src, int outer)
-  {
-    dst.copyCoeffByOuterInner(outer, Index, src);
-    ei_assign_DefaultTraversal_InnerUnrolling<Derived1, Derived2, Index+1, Stop>::run(dst, src, outer);
-  }
-};
-
-template<typename Derived1, typename Derived2, int Stop>
-struct ei_assign_DefaultTraversal_InnerUnrolling<Derived1, Derived2, Stop, Stop>
-{
-  EIGEN_STRONG_INLINE static void run(Derived1 &, const Derived2 &, int) {}
-};
-
-/***********************
-*** Linear traversal ***
-***********************/
-
-template<typename Derived1, typename Derived2, int Index, int Stop>
-struct ei_assign_LinearTraversal_CompleteUnrolling
-{
-  EIGEN_STRONG_INLINE static void run(Derived1 &dst, const Derived2 &src)
-  {
-    dst.copyCoeff(Index, src);
-    ei_assign_LinearTraversal_CompleteUnrolling<Derived1, Derived2, Index+1, Stop>::run(dst, src);
-  }
-};
-
-template<typename Derived1, typename Derived2, int Stop>
-struct ei_assign_LinearTraversal_CompleteUnrolling<Derived1, Derived2, Stop, Stop>
-{
-  EIGEN_STRONG_INLINE static void run(Derived1 &, const Derived2 &) {}
-};
-
-/**************************
-*** Inner vectorization ***
-**************************/
-
-template<typename Derived1, typename Derived2, int Index, int Stop>
-struct ei_assign_innervec_CompleteUnrolling
-{
-  enum {
-    outer = Index / Derived1::InnerSizeAtCompileTime,
-    inner = Index % Derived1::InnerSizeAtCompileTime,
-    JointAlignment = ei_assign_traits<Derived1,Derived2>::JointAlignment
-  };
-
-  EIGEN_STRONG_INLINE static void run(Derived1 &dst, const Derived2 &src)
-  {
-    dst.template copyPacketByOuterInner<Derived2, Aligned, JointAlignment>(outer, inner, 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 outer)
-  {
-    dst.template copyPacketByOuterInner<Derived2, Aligned, Aligned>(outer, Index, src);
-    ei_assign_innervec_InnerUnrolling<Derived1, Derived2,
-      Index+ei_packet_traits<typename Derived1::Scalar>::size, Stop>::run(dst, src, outer);
-  }
-};
-
-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 Traversal = ei_assign_traits<Derived1, Derived2>::Traversal,
-         int Unrolling = ei_assign_traits<Derived1, Derived2>::Unrolling>
-struct ei_assign_impl;
-
-/************************
-*** Default traversal ***
-************************/
-
-template<typename Derived1, typename Derived2, int Unrolling>
-struct ei_assign_impl<Derived1, Derived2, InvalidTraversal, Unrolling>
-{
-  inline static void run(Derived1 &, const Derived2 &) { }
-};
-
-template<typename Derived1, typename Derived2>
-struct ei_assign_impl<Derived1, Derived2, DefaultTraversal, NoUnrolling>
-{
-  typedef typename Derived1::Index Index;
-  inline static void run(Derived1 &dst, const Derived2 &src)
-  {
-    const Index innerSize = dst.innerSize();
-    const Index outerSize = dst.outerSize();
-    for(Index outer = 0; outer < outerSize; ++outer)
-      for(Index inner = 0; inner < innerSize; ++inner)
-        dst.copyCoeffByOuterInner(outer, inner, src);
-  }
-};
-
-template<typename Derived1, typename Derived2>
-struct ei_assign_impl<Derived1, Derived2, DefaultTraversal, CompleteUnrolling>
-{
-  EIGEN_STRONG_INLINE static void run(Derived1 &dst, const Derived2 &src)
-  {
-    ei_assign_DefaultTraversal_CompleteUnrolling<Derived1, Derived2, 0, Derived1::SizeAtCompileTime>
-      ::run(dst, src);
-  }
-};
-
-template<typename Derived1, typename Derived2>
-struct ei_assign_impl<Derived1, Derived2, DefaultTraversal, InnerUnrolling>
-{
-  typedef typename Derived1::Index Index;
-  EIGEN_STRONG_INLINE static void run(Derived1 &dst, const Derived2 &src)
-  {
-    const Index outerSize = dst.outerSize();
-    for(Index outer = 0; outer < outerSize; ++outer)
-      ei_assign_DefaultTraversal_InnerUnrolling<Derived1, Derived2, 0, Derived1::InnerSizeAtCompileTime>
-        ::run(dst, src, outer);
-  }
-};
-
-/***********************
-*** Linear traversal ***
-***********************/
-
-template<typename Derived1, typename Derived2>
-struct ei_assign_impl<Derived1, Derived2, LinearTraversal, NoUnrolling>
-{
-  typedef typename Derived1::Index Index;
-  inline static void run(Derived1 &dst, const Derived2 &src)
-  {
-    const Index size = dst.size();
-    for(Index i = 0; i < size; ++i)
-      dst.copyCoeff(i, src);
-  }
-};
-
-template<typename Derived1, typename Derived2>
-struct ei_assign_impl<Derived1, Derived2, LinearTraversal, CompleteUnrolling>
-{
-  EIGEN_STRONG_INLINE static void run(Derived1 &dst, const Derived2 &src)
-  {
-    ei_assign_LinearTraversal_CompleteUnrolling<Derived1, Derived2, 0, Derived1::SizeAtCompileTime>
-      ::run(dst, src);
-  }
-};
-
-/**************************
-*** Inner vectorization ***
-**************************/
-
-template<typename Derived1, typename Derived2>
-struct ei_assign_impl<Derived1, Derived2, InnerVectorizedTraversal, NoUnrolling>
-{
-  typedef typename Derived1::Index Index;
-  inline static void run(Derived1 &dst, const Derived2 &src)
-  {
-    const Index innerSize = dst.innerSize();
-    const Index outerSize = dst.outerSize();
-    const Index packetSize = ei_packet_traits<typename Derived1::Scalar>::size;
-    for(Index outer = 0; outer < outerSize; ++outer)
-      for(Index inner = 0; inner < innerSize; inner+=packetSize)
-        dst.template copyPacketByOuterInner<Derived2, Aligned, Aligned>(outer, inner, src);
-  }
-};
-
-template<typename Derived1, typename Derived2>
-struct ei_assign_impl<Derived1, Derived2, InnerVectorizedTraversal, 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, InnerVectorizedTraversal, InnerUnrolling>
-{
-  typedef typename Derived1::Index Index;
-  EIGEN_STRONG_INLINE static void run(Derived1 &dst, const Derived2 &src)
-  {
-    const Index outerSize = dst.outerSize();
-    for(Index outer = 0; outer < outerSize; ++outer)
-      ei_assign_innervec_InnerUnrolling<Derived1, Derived2, 0, Derived1::InnerSizeAtCompileTime>
-        ::run(dst, src, outer);
-  }
-};
-
-/***************************
-*** Linear vectorization ***
-***************************/
-
-template <bool IsAligned = false>
-struct ei_unaligned_assign_impl
-{
-  template <typename Derived, typename OtherDerived>
-  static EIGEN_STRONG_INLINE void run(const Derived&, OtherDerived&, typename Derived::Index, typename Derived::Index) {}
-};
-
-template <>
-struct ei_unaligned_assign_impl<false>
-{
-  // MSVC must not inline this functions. If it does, it fails to optimize the
-  // packet access path.
-#ifdef _MSC_VER
-  template <typename Derived, typename OtherDerived>
-  static EIGEN_DONT_INLINE void run(const Derived& src, OtherDerived& dst, typename Derived::Index start, typename Derived::Index end)
-#else
-  template <typename Derived, typename OtherDerived>
-  static EIGEN_STRONG_INLINE void run(const Derived& src, OtherDerived& dst, typename Derived::Index start, typename Derived::Index end)
-#endif
-  {
-    for (typename Derived::Index index = start; index < end; ++index)
-      dst.copyCoeff(index, src);
-  }
-};
-
-template<typename Derived1, typename Derived2>
-struct ei_assign_impl<Derived1, Derived2, LinearVectorizedTraversal, NoUnrolling>
-{
-  typedef typename Derived1::Index Index;
-  EIGEN_STRONG_INLINE static void run(Derived1 &dst, const Derived2 &src)
-  {
-    const Index size = dst.size();
-    typedef ei_packet_traits<typename Derived1::Scalar> PacketTraits;
-    enum {
-      packetSize = PacketTraits::size,
-      dstAlignment = PacketTraits::AlignedOnScalar ? Aligned : int(ei_assign_traits<Derived1,Derived2>::DstIsAligned) ,
-      srcAlignment = ei_assign_traits<Derived1,Derived2>::JointAlignment
-    };
-    const Index alignedStart = ei_assign_traits<Derived1,Derived2>::DstIsAligned ? 0
-                             : ei_first_aligned(&dst.coeffRef(0), size);
-    const Index alignedEnd = alignedStart + ((size-alignedStart)/packetSize)*packetSize;
-
-    ei_unaligned_assign_impl<ei_assign_traits<Derived1,Derived2>::DstIsAligned!=0>::run(src,dst,0,alignedStart);
-
-    for(Index index = alignedStart; index < alignedEnd; index += packetSize)
-    {
-      dst.template copyPacket<Derived2, dstAlignment, srcAlignment>(index, src);
-    }
-
-    ei_unaligned_assign_impl<>::run(src,dst,alignedEnd,size);
-  }
-};
-
-template<typename Derived1, typename Derived2>
-struct ei_assign_impl<Derived1, Derived2, LinearVectorizedTraversal, CompleteUnrolling>
-{
-  typedef typename Derived1::Index Index;
-  EIGEN_STRONG_INLINE static void run(Derived1 &dst, const Derived2 &src)
-  {
-    enum { size = Derived1::SizeAtCompileTime,
-           packetSize = ei_packet_traits<typename Derived1::Scalar>::size,
-           alignedSize = (size/packetSize)*packetSize };
-
-    ei_assign_innervec_CompleteUnrolling<Derived1, Derived2, 0, alignedSize>::run(dst, src);
-    ei_assign_DefaultTraversal_CompleteUnrolling<Derived1, Derived2, alignedSize, size>::run(dst, src);
-  }
-};
-
-/**************************
-*** Slice vectorization ***
-***************************/
-
-template<typename Derived1, typename Derived2>
-struct ei_assign_impl<Derived1, Derived2, SliceVectorizedTraversal, NoUnrolling>
-{
-  typedef typename Derived1::Index Index;
-  inline static void run(Derived1 &dst, const Derived2 &src)
-  {
-    typedef ei_packet_traits<typename Derived1::Scalar> PacketTraits;
-    enum {
-      packetSize = PacketTraits::size,
-      alignable = PacketTraits::AlignedOnScalar,
-      dstAlignment = alignable ? Aligned : int(ei_assign_traits<Derived1,Derived2>::DstIsAligned) ,
-      srcAlignment = ei_assign_traits<Derived1,Derived2>::JointAlignment
-    };
-    const Index packetAlignedMask = packetSize - 1;
-    const Index innerSize = dst.innerSize();
-    const Index outerSize = dst.outerSize();
-    const Index alignedStep = alignable ? (packetSize - dst.outerStride() % packetSize) & packetAlignedMask : 0;
-    Index alignedStart = ((!alignable) || ei_assign_traits<Derived1,Derived2>::DstIsAligned) ? 0
-                       : ei_first_aligned(&dst.coeffRef(0,0), innerSize);
-
-    for(Index outer = 0; outer < outerSize; ++outer)
-    {
-      const Index alignedEnd = alignedStart + ((innerSize-alignedStart) & ~packetAlignedMask);
-      // do the non-vectorizable part of the assignment
-      for(Index inner = 0; inner<alignedStart ; ++inner)
-        dst.copyCoeffByOuterInner(outer, inner, src);
-
-      // do the vectorizable part of the assignment
-      for(Index inner = alignedStart; inner<alignedEnd; inner+=packetSize)
-        dst.template copyPacketByOuterInner<Derived2, Aligned, Unaligned>(outer, inner, src);
-
-      // do the non-vectorizable part of the assignment
-      for(Index inner = alignedEnd; inner<innerSize ; ++inner)
-        dst.copyCoeffByOuterInner(outer, inner, src);
-
-      alignedStart = std::min<Index>((alignedStart+alignedStep)%packetSize, innerSize);
-    }
-  }
-};
-
-/***************************************************************************
-* Part 4 : implementation of DenseBase methods
-***************************************************************************/
+namespace Eigen {
 
 template<typename Derived>
 template<typename OtherDerived>
@@ -493,94 +20,71 @@ EIGEN_STRONG_INLINE Derived& DenseBase<Derived>
   ::lazyAssign(const DenseBase<OtherDerived>& other)
 {
   enum{
-    SameType = ei_is_same_type<typename Derived::Scalar,typename OtherDerived::Scalar>::ret
+    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(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());
   
-  
-#ifdef EIGEN_DEBUG_ASSIGN
-  ei_assign_traits<Derived, OtherDerived>::debug();
-#endif
-  ei_assert(rows() == other.rows() && cols() == other.cols());
-  ei_assign_impl<Derived, OtherDerived, int(SameType) ? int(ei_assign_traits<Derived, OtherDerived>::Traversal)
-                                                      : int(InvalidTraversal)>::run(derived(),other.derived());
-#ifndef EIGEN_NO_DEBUG
-  checkTransposeAliasing(other.derived());
-#endif
   return derived();
 }
 
-template<typename Derived, typename OtherDerived,
-         bool EvalBeforeAssigning = (int(OtherDerived::Flags) & EvalBeforeAssigningBit) != 0,
-         bool NeedToTranspose = Derived::IsVectorAtCompileTime
-                && OtherDerived::IsVectorAtCompileTime
-                && ((int(Derived::RowsAtCompileTime) == 1 && int(OtherDerived::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(Derived::ColsAtCompileTime) == 1 && int(OtherDerived::RowsAtCompileTime) == 1))
-                && 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& DenseBase<Derived>::operator=(const DenseBase<OtherDerived>& other)
-{
-  return ei_assign_selector<Derived,OtherDerived>::run(derived(), other.derived());
-}
-
-template<typename Derived>
-EIGEN_STRONG_INLINE Derived& DenseBase<Derived>::operator=(const DenseBase& other)
-{
-  return ei_assign_selector<Derived,Derived>::run(derived(), other.derived());
-}
-
-template<typename Derived>
-EIGEN_STRONG_INLINE Derived& MatrixBase<Derived>::operator=(const MatrixBase& other)
-{
-  return ei_assign_selector<Derived,Derived>::run(derived(), other.derived());
-}
-
-template<typename Derived>
-template <typename OtherDerived>
-EIGEN_STRONG_INLINE Derived& MatrixBase<Derived>::operator=(const DenseBase<OtherDerived>& other)
-{
-  return ei_assign_selector<Derived,OtherDerived>::run(derived(), other.derived());
-}
-
-template<typename Derived>
-template <typename OtherDerived>
-EIGEN_STRONG_INLINE Derived& MatrixBase<Derived>::operator=(const EigenBase<OtherDerived>& other)
+EIGEN_DEVICE_FUNC
+EIGEN_STRONG_INLINE Derived& MatrixBase<Derived>::operator=(const ReturnByValue<OtherDerived>& other)
 {
   other.derived().evalTo(derived());
   return derived();
 }
 
-template<typename Derived>
-template<typename OtherDerived>
-EIGEN_STRONG_INLINE Derived& MatrixBase<Derived>::operator=(const ReturnByValue<OtherDerived>& other)
-{
-  other.evalTo(derived());
-  return derived();
-}
+} // end namespace Eigen
 
 #endif // EIGEN_ASSIGN_H

Eigen/src/Core/AssignEvaluator.h

@@ -0,0 +1,810 @@
+// 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;
+  // TODO distinguish between linear traversal and inner-traversals
+  typedef typename find_best_packet<DstScalar,Dst::SizeAtCompileTime>::type PacketType; 
+  
+  enum {
+    DstFlags = DstEvaluator::Flags,
+    SrcFlags = SrcEvaluator::Flags,
+    RequiredAlignment = unpacket_traits<PacketType>::alignment
+  };
+  
+public:
+  enum {
+    DstAlignment = DstEvaluator::Alignment,
+    SrcAlignment = SrcEvaluator::Alignment,
+    DstHasDirectAccess = DstFlags & 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,
+    PacketSize = unpacket_traits<PacketType>::size
+  };
+
+  enum {
+    DstIsRowMajor = DstFlags&RowMajorBit,
+    SrcIsRowMajor = SrcFlags&RowMajorBit,
+    StorageOrdersAgree = (int(DstIsRowMajor) == int(SrcIsRowMajor)),
+    MightVectorize = StorageOrdersAgree
+                  && (int(DstFlags) & int(SrcFlags) & ActualPacketAccessBit)
+                  && (functor_traits<AssignFunc>::PacketAccess),
+    MayInnerVectorize  = MightVectorize
+                       && int(InnerSize)!=Dynamic && int(InnerSize)%int(PacketSize)==0
+                       && int(OuterStride)!=Dynamic && int(OuterStride)%int(PacketSize)==0
+                       && int(JointAlignment)>=int(RequiredAlignment),
+    MayLinearize = StorageOrdersAgree && (int(DstFlags) & int(SrcFlags) & LinearAccessBit),
+    MayLinearVectorize = MightVectorize && MayLinearize && DstHasDirectAccess
+                       && ((int(DstAlignment)>=int(RequiredAlignment)) || MaxSizeAtCompileTime == Dynamic),
+      /* If the destination isn't aligned, we have to do runtime checks and we don't unroll,
+         so it's only good for large enough sizes. */
+    MaySliceVectorize  = MightVectorize && DstHasDirectAccess
+                       && (int(InnerMaxSize)==Dynamic || int(InnerMaxSize)>=3*PacketSize)
+      /* slice vectorization can be slow, so we only want it if the slices are big, which is
+         indicated by InnerMaxSize rather than InnerSize, think of the case of a dynamic block
+         in a fixed-size matrix */
+  };
+
+public:
+  enum {
+    Traversal = int(MayInnerVectorize)   ? int(InnerVectorizedTraversal)
+              : int(MayLinearVectorize)  ? int(LinearVectorizedTraversal)
+              : int(MaySliceVectorize)   ? int(SliceVectorizedTraversal)
+              : int(MayLinearize)        ? int(LinearTraversal)
+                                         : int(DefaultTraversal),
+    Vectorized = int(Traversal) == InnerVectorizedTraversal
+              || int(Traversal) == LinearVectorizedTraversal
+              || int(Traversal) == SliceVectorizedTraversal
+  };
+
+private:
+  enum {
+    UnrollingLimit      = EIGEN_UNROLLING_LIMIT * (Vectorized ? int(PacketSize) : 1),
+    MayUnrollCompletely = int(Dst::SizeAtCompileTime) != Dynamic
+                       && int(Dst::SizeAtCompileTime) * int(SrcEvaluator::CoeffReadCost) <= int(UnrollingLimit),
+    MayUnrollInner      = int(InnerSize) != Dynamic
+                       && int(InnerSize) * 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) && (int(DstAlignment)>=int(RequiredAlignment)) ? int(CompleteUnrolling)
+                                                                                             : int(NoUnrolling) )
+              : int(Traversal) == int(LinearTraversal)
+                ? ( bool(MayUnrollCompletely) ? int(CompleteUnrolling) 
+                                              : int(NoUnrolling) )
+              : 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(RequiredAlignment)
+    EIGEN_DEBUG_VAR(JointAlignment)
+    EIGEN_DEBUG_VAR(InnerSize)
+    EIGEN_DEBUG_VAR(InnerMaxSize)
+    EIGEN_DEBUG_VAR(PacketSize)
+    EIGEN_DEBUG_VAR(StorageOrdersAgree)
+    EIGEN_DEBUG_VAR(MightVectorize)
+    EIGEN_DEBUG_VAR(MayLinearize)
+    EIGEN_DEBUG_VAR(MayInnerVectorize)
+    EIGEN_DEBUG_VAR(MayLinearVectorize)
+    EIGEN_DEBUG_VAR(MaySliceVectorize)
+    std::cerr << "Traversal" << " = " << Traversal << " (" << demangle_traversal(Traversal) << ")" << std::endl;
+    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,
+    JointAlignment = Kernel::AssignmentTraits::JointAlignment
+  };
+
+  EIGEN_DEVICE_FUNC static EIGEN_STRONG_INLINE void run(Kernel &kernel)
+  {
+    kernel.template assignPacketByOuterInner<Aligned, JointAlignment, 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>
+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<Aligned, Aligned, PacketType>(outer, Index_);
+    enum { NextIndex = Index_ + unpacket_traits<PacketType>::size };
+    copy_using_evaluator_innervec_InnerUnrolling<Kernel, NextIndex, Stop>::run(kernel, outer);
+  }
+};
+
+template<typename Kernel, int Stop>
+struct copy_using_evaluator_innervec_InnerUnrolling<Kernel, Stop, Stop>
+{
+  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::RequiredAlignment,
+      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.dstEvaluator().coeffRef(0), 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;
+    
+    enum { size = DstXprType::SizeAtCompileTime,
+           packetSize = packet_traits<typename Kernel::Scalar>::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;
+  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<Aligned, Aligned, 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;
+    const Index outerSize = kernel.outerSize();
+    for(Index outer = 0; outer < outerSize; ++outer)
+      copy_using_evaluator_innervec_InnerUnrolling<Kernel, 0, DstXprType::InnerSizeAtCompileTime>::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 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::RequiredAlignment),
+      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.dstEvaluator().coeffRef(0,0);
+    if((!bool(dstIsAligned)) && (size_t(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 = std::min<Index>((alignedStart+alignedStep)%packetSize, innerSize);
+    }
+  }
+};
+
+/***************************************************************************
+* 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;
+  }
+  
+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 void call_dense_assignment_loop(const DstXprType& dst, const SrcXprType& src, const Functor &func)
+{
+  eigen_assert(dst.rows() == src.rows() && dst.cols() == src.cols());
+  
+  typedef evaluator<DstXprType> DstEvaluatorType;
+  typedef evaluator<SrcXprType> SrcEvaluatorType;
+
+  DstEvaluatorType dstEvaluator(dst);
+  SrcEvaluatorType srcEvaluator(src);
+    
+  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 void call_dense_assignment_loop(const DstXprType& dst, const SrcXprType& src)
+{
+  call_dense_assignment_loop(dst, src, internal::assign_op<typename DstXprType::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 Scalar = typename DstXprType::Scalar>
+struct Assignment;
+
+
+// The only purpose of this call_assignment() function is to deal with noalias() / AssumeAliasing and automatic transposition.
+// Indeed, I (Gael) think that this concept of AssumeAliasing was a mistake, and it makes thing quite complicated.
+// So this intermediate function removes everything related to AssumeAliasing such that Assignment
+// does not has to bother about these annoying details.
+
+template<typename Dst, typename Src>
+EIGEN_DEVICE_FUNC void call_assignment(Dst& dst, const Src& src)
+{
+  call_assignment(dst, src, internal::assign_op<typename Dst::Scalar>());
+}
+template<typename Dst, typename Src>
+EIGEN_DEVICE_FUNC void call_assignment(const Dst& dst, const Src& src)
+{
+  call_assignment(dst, src, internal::assign_op<typename Dst::Scalar>());
+}
+                     
+// Deal with AssumeAliasing
+template<typename Dst, typename Src, typename Func>
+EIGEN_DEVICE_FUNC void call_assignment(Dst& dst, const Src& src, const Func& func, typename enable_if<evaluator_traits<Src>::AssumeAliasing==1, 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 void call_assignment(Dst& dst, const Src& src, const Func& func, typename enable_if<evaluator_traits<Src>::AssumeAliasing==0, void*>::type = 0)
+{
+  call_assignment_no_alias(dst, src, func);
+}
+
+// by-pass AssumeAliasing
+// 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 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 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
+  };
+
+  Index dstRows = NeedToTranspose ? src.cols() : src.rows();
+  Index dstCols = NeedToTranspose ? src.rows() : src.cols();
+  if((dst.rows()!=dstRows) || (dst.cols()!=dstCols))
+    dst.resize(dstRows, dstCols);
+  
+  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 void call_assignment_no_alias(Dst& dst, const Src& src)
+{
+  call_assignment_no_alias(dst, src, internal::assign_op<typename Dst::Scalar>());
+}
+
+template<typename Dst, typename Src, typename Func>
+EIGEN_DEVICE_FUNC void call_assignment_no_alias_no_transpose(Dst& dst, const Src& src, const Func& func)
+{
+  Index dstRows = src.rows();
+  Index dstCols = src.cols();
+  if((dst.rows()!=dstRows) || (dst.cols()!=dstCols))
+    dst.resize(dstRows, dstCols);
+  
+  // 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)
+  
+  Assignment<Dst,Src,Func>::run(dst, src, func);
+}
+template<typename Dst, typename Src>
+EIGEN_DEVICE_FUNC 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>());
+}
+
+// forward declaration
+template<typename Dst, typename Src> void check_for_aliasing(const Dst &dst, const Src &src);
+
+// Generic Dense to Dense assignment
+template< typename DstXprType, typename SrcXprType, typename Functor, typename Scalar>
+struct Assignment<DstXprType, SrcXprType, Functor, Dense2Dense, Scalar>
+{
+  EIGEN_DEVICE_FUNC static void run(DstXprType &dst, const SrcXprType &src, const Functor &func)
+  {
+    eigen_assert(dst.rows() == src.rows() && dst.cols() == src.cols());
+    
+#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.
+template< typename DstXprType, typename SrcXprType, typename Functor, typename Scalar>
+struct Assignment<DstXprType, SrcXprType, Functor, EigenBase2EigenBase, Scalar>
+{
+  EIGEN_DEVICE_FUNC static void run(DstXprType &dst, const SrcXprType &src, const internal::assign_op<typename DstXprType::Scalar> &/*func*/)
+  {
+    eigen_assert(dst.rows() == src.rows() && dst.cols() == src.cols());
+    src.evalTo(dst);
+  }
+};
+
+} // namespace internal
+
+} // end namespace Eigen
+
+#endif // EIGEN_ASSIGN_EVALUATOR_H

Eigen/src/Core/Assign_MKL.h

@@ -0,0 +1,174 @@
+/*
+ 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>,             \
+                   Dense2Dense, typename enable_if<vml_assign_traits<DstXprType,SrcXprNested>::EnableVml,EIGENTYPE>::type> {    \
+    typedef CwiseUnaryOp<scalar_##EIGENOP##_op<EIGENTYPE>, SrcXprNested> SrcXprType;                                            \
+    static void run(DstXprType &dst, const SrcXprType &src, const assign_op<EIGENTYPE> &/*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>                                                                       \
+  struct Assignment<DstXprType, CwiseUnaryOp<scalar_##EIGENOP##_op<EIGENTYPE>, SrcXprNested>, assign_op<EIGENTYPE>,           \
+                   Dense2Dense, typename enable_if<vml_assign_traits<DstXprType,SrcXprNested>::EnableVml,EIGENTYPE>::type> {  \
+    typedef CwiseUnaryOp<scalar_##EIGENOP##_op<EIGENTYPE>, SrcXprNested> SrcXprType;                                          \
+    static void run(DstXprType &dst, const SrcXprType &src, const assign_op<EIGENTYPE> &/*func*/) {                           \
+      eigen_assert(dst.rows() == src.rows() && dst.cols() == src.cols());                                                     \
+      VMLTYPE exponent = reinterpret_cast<const VMLTYPE&>(src.functor().m_exponent);                                          \
+      if(vml_assign_traits<DstXprType,SrcXprNested>::Traversal==LinearTraversal)                                              \
+      {                                                                                                                       \
+        VMLOP( dst.size(), (const VMLTYPE*)src.nestedExpression().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.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, 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

@@ -3,134 +3,90 @@
 //
 // Copyright (C) 2009 Gael Guennebaud <gael.guennebaud@inria.fr>
 //
-// Eigen is free software; you can redistribute it and/or
-// modify it under the terms of the GNU Lesser General Public
-// License as published by the Free Software Foundation; either
-// version 3 of the License, or (at your option) any later version.
-//
-// Alternatively, you can redistribute it and/or
-// modify it under the terms of the GNU General Public License as
-// published by the Free Software Foundation; either version 2 of
-// the License, or (at your option) any later version.
-//
-// Eigen is distributed in the hope that it will be useful, but WITHOUT ANY
-// WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS
-// FOR A PARTICULAR PURPOSE. See the GNU Lesser General Public License or the
-// GNU General Public License for more details.
-//
-// You should have received a copy of the GNU Lesser General Public
-// License and a copy of the GNU General Public License along with
-// Eigen. If not, see <http://www.gnu.org/licenses/>.
+// 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
 
-/**
-  * \class BandMatrix
-  * \ingroup Core_Module
-  *
-  * \brief Represents a rectangular matrix with a banded storage
-  *
-  * \param _Scalar Numeric type, i.e. float, double, int
-  * \param Rows Number of rows, or \b Dynamic
-  * \param Cols Number of columns, or \b Dynamic
-  * \param Supers Number of super diagonal
-  * \param Subs Number of sub diagonal
-  * \param _Options A combination of either \b RowMajor or \b ColMajor, and of \b SelfAdjoint
-  *                 The former controls storage order, and defaults to column-major. The latter controls
-  *                 whether the matrix represent 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 ei_traits<BandMatrix<_Scalar,Rows,Cols,Supers,Subs,Options> >
-{
-  typedef _Scalar Scalar;
-  typedef Dense StorageKind;
-  typedef DenseIndex Index;
-  enum {
-    CoeffReadCost = NumTraits<Scalar>::ReadCost,
-    RowsAtCompileTime = Rows,
-    ColsAtCompileTime = Cols,
-    MaxRowsAtCompileTime = Rows,
-    MaxColsAtCompileTime = Cols,
-    Flags = LvalueBit
-  };
-};
+namespace Eigen { 
 
-template<typename _Scalar, int Rows, int Cols, int Supers, int Subs, int Options>
-class BandMatrix : public EigenBase<BandMatrix<_Scalar,Rows,Cols,Supers,Subs,Options> >
+namespace internal {
+
+template<typename Derived>
+class BandMatrixBase : public EigenBase<Derived>
 {
   public:
 
     enum {
-      Flags = ei_traits<BandMatrix>::Flags,
-      CoeffReadCost = ei_traits<BandMatrix>::CoeffReadCost,
-      RowsAtCompileTime = ei_traits<BandMatrix>::RowsAtCompileTime,
-      ColsAtCompileTime = ei_traits<BandMatrix>::ColsAtCompileTime,
-      MaxRowsAtCompileTime = ei_traits<BandMatrix>::MaxRowsAtCompileTime,
-      MaxColsAtCompileTime = ei_traits<BandMatrix>::MaxColsAtCompileTime
+      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 ei_traits<BandMatrix>::Scalar Scalar;
+    typedef typename internal::traits<Derived>::Scalar Scalar;
     typedef Matrix<Scalar,RowsAtCompileTime,ColsAtCompileTime> DenseMatrixType;
-    typedef typename DenseMatrixType::Index Index;
+    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(Rows,Cols)
+      SizeAtCompileTime = EIGEN_SIZE_MIN_PREFER_DYNAMIC(RowsAtCompileTime,ColsAtCompileTime)
     };
-    typedef Matrix<Scalar,DataRowsAtCompileTime,ColsAtCompileTime,Options&RowMajor?RowMajor:ColMajor> DataType;
 
   public:
-
-    inline BandMatrix(Index rows=Rows, Index cols=Cols, Index supers=Supers, Index subs=Subs)
-      : m_data(1+supers+subs,cols),
-        m_rows(rows), m_supers(supers), m_subs(subs)
-    {
-        //m_data.setConstant(666);
-    }
-
-    /** \returns the number of columns */
-    inline Index rows() const { return m_rows.value(); }
-
-    /** \returns the number of rows */
-    inline Index cols() const { return m_data.cols(); }
+    
+    using Base::derived;
+    using Base::rows;
+    using Base::cols;
 
     /** \returns the number of super diagonals */
-    inline Index supers() const { return m_supers.value(); }
+    inline Index supers() const { return derived().supers(); }
 
     /** \returns the number of sub diagonals */
-    inline Index subs() const { return m_subs.value(); }
+    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<DataType,Dynamic,1> col(Index i)
+    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 = m_data.rows();
+      Index len = coeffs().rows();
       if (i<=supers())
       {
         start = supers()-i;
-        len = std::min(rows(),std::max<Index>(0,m_data.rows() - (supers()-i)));
+        len = (std::min)(rows(),std::max<Index>(0,coeffs().rows() - (supers()-i)));
       }
       else if (i>=rows()-subs())
-        len = std::max<Index>(0,m_data.rows() - (i + 1 - rows() + subs()));
-      return Block<DataType,Dynamic,1>(m_data, start, i, len, 1);
+        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<DataType,1,SizeAtCompileTime> diagonal()
-    { return Block<DataType,1,SizeAtCompileTime>(m_data,supers(),0,1,std::min(rows(),cols())); }
+    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<DataType,1,SizeAtCompileTime> diagonal() const
-    { return Block<DataType,1,SizeAtCompileTime>(m_data,supers(),0,1,std::min(rows(),cols())); }
+    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 {
@@ -143,38 +99,38 @@ class BandMatrix : public EigenBase<BandMatrix<_Scalar,Rows,Cols,Supers,Subs,Opt
                      ? EIGEN_SIZE_MIN_PREFER_DYNAMIC(ColsAtCompileTime, RowsAtCompileTime + ActualIndex)
                      : EIGEN_SIZE_MIN_PREFER_DYNAMIC(RowsAtCompileTime, ColsAtCompileTime - ActualIndex))
       };
-      typedef Block<DataType,1, DiagonalSize> BuildType;
-      typedef typename ei_meta_if<Conjugate,
-                 CwiseUnaryOp<ei_scalar_conjugate_op<Scalar>,BuildType >,
-                 BuildType>::ret Type;
+      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(m_data, supers()-N, std::max(0,N), 1, diagonalLength(N));
+      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(m_data, supers()-N, std::max(0,N), 1, diagonalLength(N));
+      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<DataType,1,Dynamic> diagonal(Index i)
+    inline Block<CoefficientsType,1,Dynamic> diagonal(Index i)
     {
-      ei_assert((i<0 && -i<=subs()) || (i>=0 && i<=supers()));
-      return Block<DataType,1,Dynamic>(m_data, supers()-i, std::max<Index>(0,i), 1, diagonalLength(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<DataType,1,Dynamic> diagonal(Index i) const
+    inline const Block<const CoefficientsType,1,Dynamic> diagonal(Index i) const
     {
-      ei_assert((i<0 && -i<=subs()) || (i>=0 && i<=supers()));
-      return Block<DataType,1,Dynamic>(m_data, supers()-i, std::max<Index>(0,i), 1, diagonalLength(i));
+      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());
@@ -196,19 +152,155 @@ class BandMatrix : public EigenBase<BandMatrix<_Scalar,Rows,Cols,Supers,Subs,Opt
   protected:
 
     inline Index diagonalLength(Index i) const
-    { return i<0 ? std::min(cols(),rows()+i) : std::min(rows(),cols()-i); }
+    { return i<0 ? (std::min)(cols(),rows()+i) : (std::min)(rows(),cols()-i); }
+};
 
-    DataType m_data;
-    ei_variable_if_dynamic<Index, Rows>   m_rows;
-    ei_variable_if_dynamic<Index, Supers> m_supers;
-    ei_variable_if_dynamic<Index, Subs>   m_subs;
+/**
+  * \class BandMatrix
+  * \ingroup Core_Module
+  *
+  * \brief Represents a rectangular matrix with a banded storage
+  *
+  * \param _Scalar Numeric type, i.e. float, double, int
+  * \param Rows Number of rows, or \b Dynamic
+  * \param Cols Number of columns, or \b Dynamic
+  * \param Supers Number of super diagonal
+  * \param Subs Number of sub diagonal
+  * \param _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
+  * \brief Represents a tridiagonal matrix with a compact banded storage
   *
   * \param _Scalar Numeric type, i.e. float, double, int
   * \param Size Number of rows and cols, or \b Dynamic
@@ -219,10 +311,10 @@ class BandMatrix : public EigenBase<BandMatrix<_Scalar,Rows,Cols,Supers,Subs,Opt
 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,1,Options&SelfAdjoint?0:1,Options|RowMajor> Base;
-    typedef typename Base::Index Index;
+    typedef BandMatrix<Scalar,Size,Size,Options&SelfAdjoint?0:1,1,Options|RowMajor> Base;
+    typedef typename Base::StorageIndex StorageIndex;
   public:
-    TridiagonalMatrix(Index size = Size) : Base(size,size,1,1) {}
+    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>(); }
@@ -235,4 +327,27 @@ class TridiagonalMatrix : public BandMatrix<Scalar,Size,Size,Options&SelfAdjoint
   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/Block.h

@@ -4,28 +4,15 @@
 // Copyright (C) 2008 Gael Guennebaud <gael.guennebaud@inria.fr>
 // Copyright (C) 2006-2010 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_BLOCK_H
 #define EIGEN_BLOCK_H
 
+namespace Eigen { 
+
 /** \class Block
   * \ingroup Core_Module
   *
@@ -34,7 +21,9 @@
   * \param XprType the type of the expression in which we are taking a block
   * \param BlockRows the number of rows of the block we are taking at compile time (optional)
   * \param BlockCols the number of columns of the block we are taking at compile time (optional)
-  * \param _DirectAccessStatus \internal used for partial specialization
+  * \param InnerPanel is true, if the block maps to a set of rows of a row major matrix or
+  *        to set of columns of a column major matrix (optional). The parameter allows to determine
+  *        at compile time whether aligned access is possible on the block expression.
   *
   * This class represents an expression of either a fixed-size or dynamic-size block. It is the return
   * type of DenseBase::block(Index,Index,Index,Index) and DenseBase::block<int,int>(Index,Index) and
@@ -58,61 +47,145 @@
   *
   * \sa DenseBase::block(Index,Index,Index,Index), DenseBase::block(Index,Index), class VectorBlock
   */
-template<typename XprType, int BlockRows, int BlockCols, bool InnerPanel, bool HasDirectAccess>
-struct ei_traits<Block<XprType, BlockRows, BlockCols, InnerPanel, HasDirectAccess> > : ei_traits<XprType>
+
+namespace internal {
+template<typename XprType, int BlockRows, int BlockCols, bool InnerPanel>
+struct traits<Block<XprType, BlockRows, BlockCols, InnerPanel> > : traits<XprType>
 {
-  typedef typename ei_traits<XprType>::Scalar Scalar;
-  typedef typename ei_traits<XprType>::StorageKind StorageKind;
-  typedef typename ei_traits<XprType>::XprKind XprKind;
-  typedef typename ei_nested<XprType>::type XprTypeNested;
-  typedef typename ei_unref<XprTypeNested>::type _XprTypeNested;
+  typedef typename traits<XprType>::Scalar Scalar;
+  typedef typename traits<XprType>::StorageKind StorageKind;
+  typedef typename traits<XprType>::XprKind XprKind;
+  typedef typename ref_selector<XprType>::type XprTypeNested;
+  typedef typename remove_reference<XprTypeNested>::type _XprTypeNested;
   enum{
-    MatrixRows = ei_traits<XprType>::RowsAtCompileTime,
-    MatrixCols = ei_traits<XprType>::ColsAtCompileTime,
+    MatrixRows = traits<XprType>::RowsAtCompileTime,
+    MatrixCols = traits<XprType>::ColsAtCompileTime,
     RowsAtCompileTime = MatrixRows == 0 ? 0 : BlockRows,
     ColsAtCompileTime = MatrixCols == 0 ? 0 : BlockCols,
     MaxRowsAtCompileTime = BlockRows==0 ? 0
                          : RowsAtCompileTime != Dynamic ? int(RowsAtCompileTime)
-                         : int(ei_traits<XprType>::MaxRowsAtCompileTime),
+                         : int(traits<XprType>::MaxRowsAtCompileTime),
     MaxColsAtCompileTime = BlockCols==0 ? 0
                          : ColsAtCompileTime != Dynamic ? int(ColsAtCompileTime)
-                         : int(ei_traits<XprType>::MaxColsAtCompileTime),
-    XprTypeIsRowMajor = (int(ei_traits<XprType>::Flags)&RowMajorBit) != 0,
+                         : int(traits<XprType>::MaxColsAtCompileTime),
+
+    XprTypeIsRowMajor = (int(traits<XprType>::Flags)&RowMajorBit) != 0,
     IsRowMajor = (MaxRowsAtCompileTime==1&&MaxColsAtCompileTime!=1) ? 1
                : (MaxColsAtCompileTime==1&&MaxRowsAtCompileTime!=1) ? 0
                : XprTypeIsRowMajor,
     HasSameStorageOrderAsXprType = (IsRowMajor == XprTypeIsRowMajor),
     InnerSize = IsRowMajor ? int(ColsAtCompileTime) : int(RowsAtCompileTime),
     InnerStrideAtCompileTime = HasSameStorageOrderAsXprType
-                             ? int(ei_inner_stride_at_compile_time<XprType>::ret)
-                             : int(ei_outer_stride_at_compile_time<XprType>::ret),
+                             ? int(inner_stride_at_compile_time<XprType>::ret)
+                             : int(outer_stride_at_compile_time<XprType>::ret),
     OuterStrideAtCompileTime = HasSameStorageOrderAsXprType
-                             ? int(ei_outer_stride_at_compile_time<XprType>::ret)
-                             : int(ei_inner_stride_at_compile_time<XprType>::ret),
-    MaskPacketAccessBit = (InnerSize == Dynamic || (InnerSize % ei_packet_traits<Scalar>::size) == 0)
-                       && (InnerStrideAtCompileTime == 1)
-                        ? PacketAccessBit : 0,
-    MaskAlignedBit = (InnerPanel && (OuterStrideAtCompileTime!=Dynamic) && ((OuterStrideAtCompileTime % ei_packet_traits<Scalar>::size) == 0)) ? AlignedBit : 0,
-    FlagsLinearAccessBit = (RowsAtCompileTime == 1 || ColsAtCompileTime == 1) ? LinearAccessBit : 0,
-    Flags0 = ei_traits<XprType>::Flags & (HereditaryBits | MaskPacketAccessBit | LvalueBit | DirectAccessBit | MaskAlignedBit),
-    Flags1 = Flags0 | FlagsLinearAccessBit,
-    Flags = (Flags1 & ~RowMajorBit) | (IsRowMajor ? RowMajorBit : 0)
+                             ? int(outer_stride_at_compile_time<XprType>::ret)
+                             : int(inner_stride_at_compile_time<XprType>::ret),
+
+    // FIXME, this traits is rather specialized for dense object and it needs to be cleaned further
+    FlagsLvalueBit = is_lvalue<XprType>::value ? LvalueBit : 0,
+    FlagsRowMajorBit = IsRowMajor ? RowMajorBit : 0,
+    Flags = (traits<XprType>::Flags & (DirectAccessBit | (InnerPanel?CompressedAccessBit:0))) | FlagsLvalueBit | FlagsRowMajorBit,
+    // FIXME DirectAccessBit should not be handled by expressions
+    // 
+    // Alignment is needed by MapBase's assertions
+    // We can sefely set it to false here. Internal alignment errors will be detected by an eigen_internal_assert in the respective evaluator
+    Alignment = 0
   };
 };
 
-template<typename XprType, int BlockRows, int BlockCols, bool InnerPanel, bool HasDirectAccess> class Block
-  : public ei_dense_xpr_base<Block<XprType, BlockRows, BlockCols, InnerPanel, HasDirectAccess> >::type
+template<typename XprType, int BlockRows=Dynamic, int BlockCols=Dynamic, bool InnerPanel = false,
+         bool HasDirectAccess = internal::has_direct_access<XprType>::ret> class BlockImpl_dense;
+         
+} // end namespace internal
+
+template<typename XprType, int BlockRows, int BlockCols, bool InnerPanel, typename StorageKind> class BlockImpl;
+
+template<typename XprType, int BlockRows, int BlockCols, bool InnerPanel> class Block
+  : public BlockImpl<XprType, BlockRows, BlockCols, InnerPanel, typename internal::traits<XprType>::StorageKind>
 {
+    typedef BlockImpl<XprType, BlockRows, BlockCols, InnerPanel, typename internal::traits<XprType>::StorageKind> Impl;
+  public:
+    //typedef typename Impl::Base Base;
+    typedef Impl Base;
+    EIGEN_GENERIC_PUBLIC_INTERFACE(Block)
+    EIGEN_INHERIT_ASSIGNMENT_OPERATORS(Block)
+    
+    typedef typename internal::remove_all<XprType>::type NestedExpression;
+  
+    /** Column or Row constructor
+      */
+    EIGEN_DEVICE_FUNC
+    inline Block(XprType& xpr, Index i) : Impl(xpr,i)
+    {
+      eigen_assert( (i>=0) && (
+          ((BlockRows==1) && (BlockCols==XprType::ColsAtCompileTime) && i<xpr.rows())
+        ||((BlockRows==XprType::RowsAtCompileTime) && (BlockCols==1) && i<xpr.cols())));
+    }
+
+    /** Fixed-size constructor
+      */
+    EIGEN_DEVICE_FUNC
+    inline Block(XprType& xpr, Index startRow, Index startCol)
+      : Impl(xpr, startRow, startCol)
+    {
+      EIGEN_STATIC_ASSERT(RowsAtCompileTime!=Dynamic && ColsAtCompileTime!=Dynamic,THIS_METHOD_IS_ONLY_FOR_FIXED_SIZE)
+      eigen_assert(startRow >= 0 && BlockRows >= 1 && startRow + BlockRows <= xpr.rows()
+             && startCol >= 0 && BlockCols >= 1 && startCol + BlockCols <= xpr.cols());
+    }
+
+    /** Dynamic-size constructor
+      */
+    EIGEN_DEVICE_FUNC
+    inline Block(XprType& xpr,
+          Index startRow, Index startCol,
+          Index blockRows, Index blockCols)
+      : Impl(xpr, startRow, startCol, blockRows, blockCols)
+    {
+      eigen_assert((RowsAtCompileTime==Dynamic || RowsAtCompileTime==blockRows)
+          && (ColsAtCompileTime==Dynamic || ColsAtCompileTime==blockCols));
+      eigen_assert(startRow >= 0 && blockRows >= 0 && startRow  <= xpr.rows() - blockRows
+          && startCol >= 0 && blockCols >= 0 && startCol <= xpr.cols() - blockCols);
+    }
+};
+         
+// The generic default implementation for dense block simplu forward to the internal::BlockImpl_dense
+// that must be specialized for direct and non-direct access...
+template<typename XprType, int BlockRows, int BlockCols, bool InnerPanel>
+class BlockImpl<XprType, BlockRows, BlockCols, InnerPanel, Dense>
+  : public internal::BlockImpl_dense<XprType, BlockRows, BlockCols, InnerPanel>
+{
+    typedef internal::BlockImpl_dense<XprType, BlockRows, BlockCols, InnerPanel> Impl;
+    typedef typename XprType::StorageIndex StorageIndex;
+  public:
+    typedef Impl Base;
+    EIGEN_INHERIT_ASSIGNMENT_OPERATORS(BlockImpl)
+    EIGEN_DEVICE_FUNC inline BlockImpl(XprType& xpr, Index i) : Impl(xpr,i) {}
+    EIGEN_DEVICE_FUNC inline BlockImpl(XprType& xpr, Index startRow, Index startCol) : Impl(xpr, startRow, startCol) {}
+    EIGEN_DEVICE_FUNC
+    inline BlockImpl(XprType& xpr, Index startRow, Index startCol, Index blockRows, Index blockCols)
+      : Impl(xpr, startRow, startCol, blockRows, blockCols) {}
+};
+
+namespace internal {
+
+/** \internal Internal implementation of dense Blocks in the general case. */
+template<typename XprType, int BlockRows, int BlockCols, bool InnerPanel, bool HasDirectAccess> class BlockImpl_dense
+  : public internal::dense_xpr_base<Block<XprType, BlockRows, BlockCols, InnerPanel> >::type
+{
+    typedef Block<XprType, BlockRows, BlockCols, InnerPanel> BlockType;
   public:
 
-    typedef typename ei_dense_xpr_base<Block>::type Base;
-    EIGEN_DENSE_PUBLIC_INTERFACE(Block)
+    typedef typename internal::dense_xpr_base<BlockType>::type Base;
+    EIGEN_DENSE_PUBLIC_INTERFACE(BlockType)
+    EIGEN_INHERIT_ASSIGNMENT_OPERATORS(BlockImpl_dense)
 
-    class InnerIterator;
+    // class InnerIterator; // FIXME apparently never used
 
     /** Column or Row constructor
       */
-    inline Block(const XprType& xpr, Index i)
+    EIGEN_DEVICE_FUNC
+    inline BlockImpl_dense(XprType& xpr, Index i)
       : m_xpr(xpr),
         // It is a row if and only if BlockRows==1 and BlockCols==XprType::ColsAtCompileTime,
         // and it is a column if and only if BlockRows==XprType::RowsAtCompileTime and BlockCols==1,
@@ -122,60 +195,68 @@ template<typename XprType, int BlockRows, int BlockCols, bool InnerPanel, bool H
         m_startCol( (BlockRows==XprType::RowsAtCompileTime) && (BlockCols==1) ? i : 0),
         m_blockRows(BlockRows==1 ? 1 : xpr.rows()),
         m_blockCols(BlockCols==1 ? 1 : xpr.cols())
-    {
-      ei_assert( (i>=0) && (
-          ((BlockRows==1) && (BlockCols==XprType::ColsAtCompileTime) && i<xpr.rows())
-        ||((BlockRows==XprType::RowsAtCompileTime) && (BlockCols==1) && i<xpr.cols())));
-    }
+    {}
 
     /** Fixed-size constructor
       */
-    inline Block(const XprType& xpr, Index startRow, Index startCol)
+    EIGEN_DEVICE_FUNC
+    inline BlockImpl_dense(XprType& xpr, Index startRow, Index startCol)
       : m_xpr(xpr), m_startRow(startRow), m_startCol(startCol),
-        m_blockRows(BlockRows), m_blockCols(BlockCols)
-    {
-      EIGEN_STATIC_ASSERT(RowsAtCompileTime!=Dynamic && ColsAtCompileTime!=Dynamic,THIS_METHOD_IS_ONLY_FOR_FIXED_SIZE)
-      ei_assert(startRow >= 0 && BlockRows >= 1 && startRow + BlockRows <= xpr.rows()
-             && startCol >= 0 && BlockCols >= 1 && startCol + BlockCols <= xpr.cols());
-    }
+                    m_blockRows(BlockRows), m_blockCols(BlockCols)
+    {}
 
     /** Dynamic-size constructor
       */
-    inline Block(const XprType& xpr,
+    EIGEN_DEVICE_FUNC
+    inline BlockImpl_dense(XprType& xpr,
           Index startRow, Index startCol,
           Index blockRows, Index blockCols)
       : m_xpr(xpr), m_startRow(startRow), m_startCol(startCol),
-                          m_blockRows(blockRows), m_blockCols(blockCols)
-    {
-      ei_assert((RowsAtCompileTime==Dynamic || RowsAtCompileTime==blockRows)
-          && (ColsAtCompileTime==Dynamic || ColsAtCompileTime==blockCols));
-      ei_assert(startRow >= 0 && blockRows >= 0 && startRow + blockRows <= xpr.rows()
-          && startCol >= 0 && blockCols >= 0 && startCol + blockCols <= xpr.cols());
-    }
-
-    EIGEN_INHERIT_ASSIGNMENT_OPERATORS(Block)
-
-    inline Index rows() const { return m_blockRows.value(); }
-    inline Index cols() const { return m_blockCols.value(); }
-
-    inline Scalar& coeffRef(Index row, Index col)
+                    m_blockRows(blockRows), m_blockCols(blockCols)
+    {}
+
+    EIGEN_DEVICE_FUNC inline Index rows() const { return m_blockRows.value(); }
+    EIGEN_DEVICE_FUNC inline Index cols() const { return m_blockCols.value(); }
+
+    EIGEN_DEVICE_FUNC
+    inline Scalar& coeffRef(Index rowId, Index colId)
     {
+      EIGEN_STATIC_ASSERT_LVALUE(XprType)
       return m_xpr.const_cast_derived()
-               .coeffRef(row + m_startRow.value(), col + m_startCol.value());
+               .coeffRef(rowId + m_startRow.value(), colId + m_startCol.value());
     }
 
-    EIGEN_STRONG_INLINE const CoeffReturnType coeff(Index row, Index col) const
+    EIGEN_DEVICE_FUNC
+    inline const Scalar& coeffRef(Index rowId, Index colId) const
     {
-      return m_xpr.coeff(row + m_startRow.value(), col + m_startCol.value());
+      return m_xpr.derived()
+               .coeffRef(rowId + m_startRow.value(), colId + m_startCol.value());
     }
 
+    EIGEN_DEVICE_FUNC
+    EIGEN_STRONG_INLINE const CoeffReturnType coeff(Index rowId, Index colId) const
+    {
+      return m_xpr.coeff(rowId + m_startRow.value(), colId + m_startCol.value());
+    }
+
+    EIGEN_DEVICE_FUNC
     inline Scalar& coeffRef(Index index)
+    {
+      EIGEN_STATIC_ASSERT_LVALUE(XprType)
+      return m_xpr.const_cast_derived()
+             .coeffRef(m_startRow.value() + (RowsAtCompileTime == 1 ? 0 : index),
+                       m_startCol.value() + (RowsAtCompileTime == 1 ? index : 0));
+    }
+
+    EIGEN_DEVICE_FUNC
+    inline const Scalar& coeffRef(Index index) const
     {
       return m_xpr.const_cast_derived()
              .coeffRef(m_startRow.value() + (RowsAtCompileTime == 1 ? 0 : index),
                        m_startCol.value() + (RowsAtCompileTime == 1 ? index : 0));
     }
 
+    EIGEN_DEVICE_FUNC
     inline const CoeffReturnType coeff(Index index) const
     {
       return m_xpr
@@ -184,17 +265,17 @@ template<typename XprType, int BlockRows, int BlockCols, bool InnerPanel, bool H
     }
 
     template<int LoadMode>
-    inline PacketScalar packet(Index row, Index col) const
+    inline PacketScalar packet(Index rowId, Index colId) const
     {
       return m_xpr.template packet<Unaligned>
-              (row + m_startRow.value(), col + m_startCol.value());
+              (rowId + m_startRow.value(), colId + m_startCol.value());
     }
 
     template<int LoadMode>
-    inline void writePacket(Index row, Index col, const PacketScalar& x)
+    inline void writePacket(Index rowId, Index colId, const PacketScalar& val)
     {
       m_xpr.const_cast_derived().template writePacket<Unaligned>
-              (row + m_startRow.value(), col + m_startCol.value(), x);
+              (rowId + m_startRow.value(), colId + m_startCol.value(), val);
     }
 
     template<int LoadMode>
@@ -206,91 +287,114 @@ template<typename XprType, int BlockRows, int BlockCols, bool InnerPanel, bool H
     }
 
     template<int LoadMode>
-    inline void writePacket(Index index, const PacketScalar& x)
+    inline void writePacket(Index index, const PacketScalar& val)
     {
       m_xpr.const_cast_derived().template writePacket<Unaligned>
          (m_startRow.value() + (RowsAtCompileTime == 1 ? 0 : index),
-          m_startCol.value() + (RowsAtCompileTime == 1 ? index : 0), x);
+          m_startCol.value() + (RowsAtCompileTime == 1 ? index : 0), val);
     }
 
     #ifdef EIGEN_PARSED_BY_DOXYGEN
     /** \sa MapBase::data() */
-    inline const Scalar* data() const;
-    inline Index innerStride() const;
-    inline Index outerStride() const;
+    EIGEN_DEVICE_FUNC inline const Scalar* data() const;
+    EIGEN_DEVICE_FUNC inline Index innerStride() const;
+    EIGEN_DEVICE_FUNC inline Index outerStride() const;
     #endif
 
+    EIGEN_DEVICE_FUNC
+    const typename internal::remove_all<typename XprType::Nested>::type& nestedExpression() const
+    { 
+      return m_xpr; 
+    }
+      
+    EIGEN_DEVICE_FUNC
+    StorageIndex startRow() const
+    { 
+      return m_startRow.value(); 
+    }
+      
+    EIGEN_DEVICE_FUNC
+    StorageIndex startCol() const
+    { 
+      return m_startCol.value(); 
+    }
+
   protected:
 
     const typename XprType::Nested m_xpr;
-    const ei_variable_if_dynamic<Index, XprType::RowsAtCompileTime == 1 ? 0 : Dynamic> m_startRow;
-    const ei_variable_if_dynamic<Index, XprType::ColsAtCompileTime == 1 ? 0 : Dynamic> m_startCol;
-    const ei_variable_if_dynamic<Index, RowsAtCompileTime> m_blockRows;
-    const ei_variable_if_dynamic<Index, ColsAtCompileTime> m_blockCols;
+    const internal::variable_if_dynamic<StorageIndex, XprType::RowsAtCompileTime == 1 ? 0 : Dynamic> m_startRow;
+    const internal::variable_if_dynamic<StorageIndex, XprType::ColsAtCompileTime == 1 ? 0 : Dynamic> m_startCol;
+    const internal::variable_if_dynamic<StorageIndex, RowsAtCompileTime> m_blockRows;
+    const internal::variable_if_dynamic<StorageIndex, ColsAtCompileTime> m_blockCols;
 };
 
-/** \internal */
+/** \internal Internal implementation of dense Blocks in the direct access case.*/
 template<typename XprType, int BlockRows, int BlockCols, bool InnerPanel>
-class Block<XprType,BlockRows,BlockCols, InnerPanel,true>
-  : public MapBase<Block<XprType, BlockRows, BlockCols, InnerPanel, true> >
+class BlockImpl_dense<XprType,BlockRows,BlockCols, InnerPanel,true>
+  : public MapBase<Block<XprType, BlockRows, BlockCols, InnerPanel> >
 {
+    typedef Block<XprType, BlockRows, BlockCols, InnerPanel> BlockType;
+    enum {
+      XprTypeIsRowMajor = (int(traits<XprType>::Flags)&RowMajorBit) != 0
+    };
   public:
 
-    typedef MapBase<Block> Base;
-    EIGEN_DENSE_PUBLIC_INTERFACE(Block)
-
-    EIGEN_INHERIT_ASSIGNMENT_OPERATORS(Block)
+    typedef MapBase<BlockType> Base;
+    EIGEN_DENSE_PUBLIC_INTERFACE(BlockType)
+    EIGEN_INHERIT_ASSIGNMENT_OPERATORS(BlockImpl_dense)
 
     /** Column or Row constructor
       */
-    inline Block(const XprType& xpr, Index i)
-      : Base(&xpr.const_cast_derived().coeffRef(
-              (BlockRows==1) && (BlockCols==XprType::ColsAtCompileTime) ? i : 0,
-              (BlockRows==XprType::RowsAtCompileTime) && (BlockCols==1) ? i : 0),
+    EIGEN_DEVICE_FUNC
+    inline BlockImpl_dense(XprType& xpr, Index i)
+      : Base(xpr.data() + i * (    ((BlockRows==1) && (BlockCols==XprType::ColsAtCompileTime) && (!XprTypeIsRowMajor)) 
+                                || ((BlockRows==XprType::RowsAtCompileTime) && (BlockCols==1) && ( XprTypeIsRowMajor)) ? xpr.innerStride() : xpr.outerStride()),
              BlockRows==1 ? 1 : xpr.rows(),
              BlockCols==1 ? 1 : xpr.cols()),
         m_xpr(xpr)
     {
-      ei_assert( (i>=0) && (
-          ((BlockRows==1) && (BlockCols==XprType::ColsAtCompileTime) && i<xpr.rows())
-        ||((BlockRows==XprType::RowsAtCompileTime) && (BlockCols==1) && i<xpr.cols())));
       init();
     }
 
     /** Fixed-size constructor
       */
-    inline Block(const XprType& xpr, Index startRow, Index startCol)
-      : Base(&xpr.const_cast_derived().coeffRef(startRow,startCol)), m_xpr(xpr)
+    EIGEN_DEVICE_FUNC
+    inline BlockImpl_dense(XprType& xpr, Index startRow, Index startCol)
+      : Base(xpr.data()+xpr.innerStride()*(XprTypeIsRowMajor?startCol:startRow) + xpr.outerStride()*(XprTypeIsRowMajor?startRow:startCol)),
+        m_xpr(xpr)
     {
-      ei_assert(startRow >= 0 && BlockRows >= 1 && startRow + BlockRows <= xpr.rows()
-             && startCol >= 0 && BlockCols >= 1 && startCol + BlockCols <= xpr.cols());
       init();
     }
 
     /** Dynamic-size constructor
       */
-    inline Block(const XprType& xpr,
+    EIGEN_DEVICE_FUNC
+    inline BlockImpl_dense(XprType& xpr,
           Index startRow, Index startCol,
           Index blockRows, Index blockCols)
-      : Base(&xpr.const_cast_derived().coeffRef(startRow,startCol), blockRows, blockCols),
+      : Base(xpr.data()+xpr.innerStride()*(XprTypeIsRowMajor?startCol:startRow) + xpr.outerStride()*(XprTypeIsRowMajor?startRow:startCol), blockRows, blockCols),
         m_xpr(xpr)
     {
-      ei_assert((RowsAtCompileTime==Dynamic || RowsAtCompileTime==blockRows)
-             && (ColsAtCompileTime==Dynamic || ColsAtCompileTime==blockCols));
-      ei_assert(startRow >= 0 && blockRows >= 0 && startRow + blockRows <= xpr.rows()
-             && startCol >= 0 && blockCols >= 0 && startCol + blockCols <= xpr.cols());
       init();
     }
 
+    EIGEN_DEVICE_FUNC
+    const typename internal::remove_all<typename XprType::Nested>::type& nestedExpression() const
+    { 
+      return m_xpr; 
+    }
+      
     /** \sa MapBase::innerStride() */
+    EIGEN_DEVICE_FUNC
     inline Index innerStride() const
     {
-      return ei_traits<Block>::HasSameStorageOrderAsXprType
+      return internal::traits<BlockType>::HasSameStorageOrderAsXprType
              ? m_xpr.innerStride()
              : m_xpr.outerStride();
     }
 
     /** \sa MapBase::outerStride() */
+    EIGEN_DEVICE_FUNC
     inline Index outerStride() const
     {
       return m_outerStride;
@@ -304,7 +408,8 @@ class Block<XprType,BlockRows,BlockCols, InnerPanel,true>
 
     #ifndef EIGEN_PARSED_BY_DOXYGEN
     /** \internal used by allowAligned() */
-    inline Block(const XprType& xpr, const Scalar* data, Index blockRows, Index blockCols)
+    EIGEN_DEVICE_FUNC
+    inline BlockImpl_dense(XprType& xpr, const Scalar* data, Index blockRows, Index blockCols)
       : Base(data, blockRows, blockCols), m_xpr(xpr)
     {
       init();
@@ -312,16 +417,20 @@ class Block<XprType,BlockRows,BlockCols, InnerPanel,true>
     #endif
 
   protected:
+    EIGEN_DEVICE_FUNC
     void init()
     {
-      m_outerStride = ei_traits<Block>::HasSameStorageOrderAsXprType
+      m_outerStride = internal::traits<BlockType>::HasSameStorageOrderAsXprType
                     ? m_xpr.outerStride()
                     : m_xpr.innerStride();
     }
 
-    const typename XprType::Nested m_xpr;
-    int m_outerStride;
+    typename XprType::Nested m_xpr;
+    Index m_outerStride;
 };
 
+} // end namespace internal
+
+} // end namespace Eigen
 
 #endif // EIGEN_BLOCK_H

Eigen/src/Core/BooleanRedux.h

@@ -3,80 +3,73 @@
 //
 // 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_ALLANDANY_H
 #define EIGEN_ALLANDANY_H
 
+namespace Eigen { 
+
+namespace internal {
+
 template<typename Derived, int UnrollCount>
-struct ei_all_unroller
+struct all_unroller
 {
+  typedef typename Derived::ExpressionTraits Traits;
   enum {
-    col = (UnrollCount-1) / Derived::RowsAtCompileTime,
-    row = (UnrollCount-1) % Derived::RowsAtCompileTime
+    col = (UnrollCount-1) / Traits::RowsAtCompileTime,
+    row = (UnrollCount-1) % Traits::RowsAtCompileTime
   };
 
-  inline static bool run(const Derived &mat)
+  static inline bool run(const Derived &mat)
   {
-    return ei_all_unroller<Derived, UnrollCount-1>::run(mat) && mat.coeff(row, col);
+    return all_unroller<Derived, UnrollCount-1>::run(mat) && mat.coeff(row, col);
   }
 };
 
 template<typename Derived>
-struct ei_all_unroller<Derived, 1>
+struct all_unroller<Derived, 0>
 {
-  inline static bool run(const Derived &mat) { return mat.coeff(0, 0); }
+  static inline bool run(const Derived &/*mat*/) { return true; }
 };
 
 template<typename Derived>
-struct ei_all_unroller<Derived, Dynamic>
+struct all_unroller<Derived, Dynamic>
 {
-  inline static bool run(const Derived &) { return false; }
+  static inline bool run(const Derived &) { return false; }
 };
 
 template<typename Derived, int UnrollCount>
-struct ei_any_unroller
+struct any_unroller
 {
+  typedef typename Derived::ExpressionTraits Traits;
   enum {
-    col = (UnrollCount-1) / Derived::RowsAtCompileTime,
-    row = (UnrollCount-1) % Derived::RowsAtCompileTime
+    col = (UnrollCount-1) / Traits::RowsAtCompileTime,
+    row = (UnrollCount-1) % Traits::RowsAtCompileTime
   };
-
-  inline static bool run(const Derived &mat)
+  
+  static inline bool run(const Derived &mat)
   {
-    return ei_any_unroller<Derived, UnrollCount-1>::run(mat) || mat.coeff(row, col);
+    return any_unroller<Derived, UnrollCount-1>::run(mat) || mat.coeff(row, col);
   }
 };
 
 template<typename Derived>
-struct ei_any_unroller<Derived, 1>
+struct any_unroller<Derived, 0>
 {
-  inline static bool run(const Derived &mat) { return mat.coeff(0, 0); }
+  static inline bool run(const Derived & /*mat*/) { return false; }
 };
 
 template<typename Derived>
-struct ei_any_unroller<Derived, Dynamic>
+struct any_unroller<Derived, Dynamic>
 {
-  inline static bool run(const Derived &) { return false; }
+  static inline bool run(const Derived &) { return false; }
 };
 
+} // end namespace internal
+
 /** \returns true if all coefficients are true
   *
   * Example: \include MatrixBase_all.cpp
@@ -87,21 +80,19 @@ struct ei_any_unroller<Derived, Dynamic>
 template<typename Derived>
 inline bool DenseBase<Derived>::all() const
 {
+  typedef internal::evaluator<Derived> Evaluator;
   enum {
     unroll = SizeAtCompileTime != Dynamic
-          && CoeffReadCost != Dynamic
-          && NumTraits<Scalar>::AddCost != Dynamic
-          && SizeAtCompileTime * (CoeffReadCost + NumTraits<Scalar>::AddCost) <= EIGEN_UNROLLING_LIMIT
+          && SizeAtCompileTime * (Evaluator::CoeffReadCost + NumTraits<Scalar>::AddCost) <= EIGEN_UNROLLING_LIMIT
   };
+  Evaluator evaluator(derived());
   if(unroll)
-    return ei_all_unroller<Derived,
-                           unroll ? int(SizeAtCompileTime) : Dynamic
-     >::run(derived());
+    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 (!coeff(i, j)) return false;
+        if (!evaluator.coeff(i, j)) return false;
     return true;
   }
 }
@@ -113,21 +104,19 @@ inline bool DenseBase<Derived>::all() const
 template<typename Derived>
 inline bool DenseBase<Derived>::any() const
 {
+  typedef internal::evaluator<Derived> Evaluator;
   enum {
     unroll = SizeAtCompileTime != Dynamic
-          && CoeffReadCost != Dynamic
-          && NumTraits<Scalar>::AddCost != Dynamic
-          && SizeAtCompileTime * (CoeffReadCost + NumTraits<Scalar>::AddCost) <= EIGEN_UNROLLING_LIMIT
+          && SizeAtCompileTime * (Evaluator::CoeffReadCost + NumTraits<Scalar>::AddCost) <= EIGEN_UNROLLING_LIMIT
   };
+  Evaluator evaluator(derived());
   if(unroll)
-    return ei_any_unroller<Derived,
-                           unroll ? int(SizeAtCompileTime) : Dynamic
-           >::run(derived());
+    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 (coeff(i, j)) return true;
+        if (evaluator.coeff(i, j)) return true;
     return false;
   }
 }
@@ -137,9 +126,39 @@ inline bool DenseBase<Derived>::any() const
   * \sa all(), any()
   */
 template<typename Derived>
-inline typename DenseBase<Derived>::Index DenseBase<Derived>::count() const
+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

@@ -8,3 +8,4 @@ INSTALL(FILES
 ADD_SUBDIRECTORY(products)
 ADD_SUBDIRECTORY(util)
 ADD_SUBDIRECTORY(arch)
+ADD_SUBDIRECTORY(functors)

Eigen/src/Core/CommaInitializer.h

@@ -4,28 +4,15 @@
 // 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
   *
@@ -41,8 +28,8 @@ template<typename XprType>
 struct CommaInitializer
 {
   typedef typename XprType::Scalar Scalar;
-  typedef typename XprType::Index Index;
 
+  EIGEN_DEVICE_FUNC
   inline CommaInitializer(XprType& xpr, const Scalar& s)
     : m_xpr(xpr), m_row(0), m_col(1), m_currentBlockRows(1)
   {
@@ -50,13 +37,27 @@ struct CommaInitializer
   }
 
   template<typename OtherDerived>
+  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_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_xpr.cols())
@@ -64,31 +65,34 @@ struct CommaInitializer
       m_row+=m_currentBlockRows;
       m_col = 0;
       m_currentBlockRows = 1;
-      ei_assert(m_row<m_xpr.rows()
+      eigen_assert(m_row<m_xpr.rows()
         && "Too many rows passed to comma initializer (operator<<)");
     }
-    ei_assert(m_col<m_xpr.cols()
+    eigen_assert(m_col<m_xpr.cols()
       && "Too many coefficients passed to comma initializer (operator<<)");
-    ei_assert(m_currentBlockRows==1);
+    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>
+  EIGEN_DEVICE_FUNC
   CommaInitializer& operator,(const DenseBase<OtherDerived>& other)
   {
+    if(other.cols()==0 || other.rows()==0)
+      return *this;
     if (m_col==m_xpr.cols())
     {
       m_row+=m_currentBlockRows;
       m_col = 0;
       m_currentBlockRows = other.rows();
-      ei_assert(m_row+m_currentBlockRows<=m_xpr.rows()
+      eigen_assert(m_row+m_currentBlockRows<=m_xpr.rows()
         && "Too many rows passed to comma initializer (operator<<)");
     }
-    ei_assert(m_col<m_xpr.cols()
+    eigen_assert(m_col<m_xpr.cols()
       && "Too many coefficients passed to comma initializer (operator<<)");
-    ei_assert(m_currentBlockRows==other.rows());
+    eigen_assert(m_currentBlockRows==other.rows());
     if (OtherDerived::SizeAtCompileTime != Dynamic)
       m_xpr.template block<OtherDerived::RowsAtCompileTime != Dynamic ? OtherDerived::RowsAtCompileTime : 1,
                               OtherDerived::ColsAtCompileTime != Dynamic ? OtherDerived::ColsAtCompileTime : 1>
@@ -99,9 +103,13 @@ struct CommaInitializer
     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_xpr.rows()
+    eigen_assert((m_row+m_currentBlockRows) == m_xpr.rows()
          && m_col == m_xpr.cols()
          && "Too few coefficients passed to comma initializer (operator<<)");
   }
@@ -113,9 +121,10 @@ struct CommaInitializer
     * quaternion.fromRotationMatrix((Matrix3f() << axis0, axis1, axis2).finished());
     * \endcode
     */
+  EIGEN_DEVICE_FUNC
   inline XprType& finished() { return m_xpr; }
 
-  XprType& m_xpr;   // target expression
+  XprType& m_xpr;           // target expression
   Index m_row;              // current row id
   Index m_col;              // current col id
   Index m_currentBlockRows; // current block height
@@ -129,6 +138,8 @@ struct CommaInitializer
   *
   * 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
   */
@@ -147,4 +158,6 @@ 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/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/CwiseBinaryOp.h

@@ -1,31 +1,18 @@
 // This file is part of Eigen, a lightweight C++ template library
 // for linear algebra.
 //
-// Copyright (C) 2008-2009 Gael Guennebaud <gael.guennebaud@inria.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 {
+
 /** \class CwiseBinaryOp
   * \ingroup Core_Module
   *
@@ -45,165 +32,124 @@
   *
   * \sa MatrixBase::binaryExpr(const MatrixBase<OtherDerived> &,const CustomBinaryOp &) const, class CwiseUnaryOp, class CwiseNullaryOp
   */
+
+namespace internal {
 template<typename BinaryOp, typename Lhs, typename Rhs>
-struct ei_traits<CwiseBinaryOp<BinaryOp, Lhs, Rhs> >
+struct traits<CwiseBinaryOp<BinaryOp, Lhs, Rhs> >
 {
-  // we must not inherit from ei_traits<Lhs> since it has
+  // we must not inherit from traits<Lhs> since it has
   // the potential to cause problems with MSVC
-  typedef typename ei_cleantype<Lhs>::type Ancestor;
-  typedef typename ei_traits<Ancestor>::XprKind XprKind;
+  typedef typename remove_all<Lhs>::type Ancestor;
+  typedef typename traits<Ancestor>::XprKind XprKind;
   enum {
-    RowsAtCompileTime = ei_traits<Ancestor>::RowsAtCompileTime,
-    ColsAtCompileTime = ei_traits<Ancestor>::ColsAtCompileTime,
-    MaxRowsAtCompileTime = ei_traits<Ancestor>::MaxRowsAtCompileTime,
-    MaxColsAtCompileTime = ei_traits<Ancestor>::MaxColsAtCompileTime
+    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 ei_result_of<
+  typedef typename result_of<
                      BinaryOp(
                        typename Lhs::Scalar,
                        typename Rhs::Scalar
                      )
                    >::type Scalar;
-  typedef typename ei_promote_storage_type<typename ei_traits<Lhs>::StorageKind,
-                                           typename ei_traits<Rhs>::StorageKind>::ret StorageKind;
-  typedef typename ei_promote_index_type<typename ei_traits<Lhs>::Index,
-                                         typename ei_traits<Rhs>::Index>::type Index;
+  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 ei_unref<LhsNested>::type _LhsNested;
-  typedef typename ei_unref<RhsNested>::type _RhsNested;
+  typedef typename remove_reference<LhsNested>::type _LhsNested;
+  typedef typename remove_reference<RhsNested>::type _RhsNested;
   enum {
-    LhsCoeffReadCost = _LhsNested::CoeffReadCost,
-    RhsCoeffReadCost = _RhsNested::CoeffReadCost,
-    LhsFlags = _LhsNested::Flags,
-    RhsFlags = _RhsNested::Flags,
-    SameType = ei_is_same_type<typename _LhsNested::Scalar,typename _RhsNested::Scalar>::ret,
-    StorageOrdersAgree = (int(Lhs::Flags)&RowMajorBit)==(int(Rhs::Flags)&RowMajorBit),
-    Flags0 = (int(LhsFlags) | int(RhsFlags)) & (
-        HereditaryBits
-      | (int(LhsFlags) & int(RhsFlags) &
-           ( AlignedBit
-           | (StorageOrdersAgree ? LinearAccessBit : 0)
-           | (ei_functor_traits<BinaryOp>::PacketAccess && StorageOrdersAgree && SameType ? PacketAccessBit : 0)
-           )
-        )
-     ),
-    Flags = (Flags0 & ~RowMajorBit) | (LhsFlags & RowMajorBit),
-    CoeffReadCost = LhsCoeffReadCost + RhsCoeffReadCost + ei_functor_traits<BinaryOp>::Cost
+    Flags = _LhsNested::Flags & RowMajorBit
   };
 };
-
-// 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.
-#define EIGEN_CHECK_BINARY_COMPATIBILIY(BINOP,LHS,RHS) \
-  EIGEN_STATIC_ASSERT((ei_functor_allows_mixing_real_and_complex<BINOP>::ret \
-                        ? int(ei_is_same_type<typename NumTraits<LHS>::Real, typename NumTraits<RHS>::Real>::ret) \
-                        : int(ei_is_same_type<LHS, RHS>::ret)), \
-    YOU_MIXED_DIFFERENT_NUMERIC_TYPES__YOU_NEED_TO_USE_THE_CAST_METHOD_OF_MATRIXBASE_TO_CAST_NUMERIC_TYPES_EXPLICITLY)
+} // end namespace internal
 
 template<typename BinaryOp, typename Lhs, typename Rhs, typename StorageKind>
 class CwiseBinaryOpImpl;
 
-template<typename BinaryOp, typename Lhs, typename Rhs>
-class CwiseBinaryOp : ei_no_assignment_operator,
+template<typename BinaryOp, typename LhsType, typename RhsType>
+class CwiseBinaryOp : 
   public CwiseBinaryOpImpl<
-          BinaryOp, Lhs, Rhs,
-          typename ei_promote_storage_type<typename ei_traits<Lhs>::StorageKind,
-                                           typename ei_traits<Rhs>::StorageKind>::ret>
+          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<LhsType>::type Lhs;
+    typedef typename internal::remove_all<RhsType>::type Rhs;
 
     typedef typename CwiseBinaryOpImpl<
-        BinaryOp, Lhs, Rhs,
-        typename ei_promote_storage_type<typename ei_traits<Lhs>::StorageKind,
-                                         typename ei_traits<Rhs>::StorageKind>::ret>::Base Base;
+        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_nested<Lhs>::type LhsNested;
-    typedef typename ei_nested<Rhs>::type RhsNested;
-    typedef typename ei_unref<LhsNested>::type _LhsNested;
-    typedef typename ei_unref<RhsNested>::type _RhsNested;
+    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_STRONG_INLINE CwiseBinaryOp(const Lhs& lhs, const Rhs& rhs, const BinaryOp& func = BinaryOp())
-      : m_lhs(lhs), m_rhs(rhs), m_functor(func)
+    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)
     {
       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_DEVICE_FUNC
     EIGEN_STRONG_INLINE Index rows() const {
       // return the fixed size type if available to enable compile time optimizations
-      if (ei_traits<typename ei_cleantype<LhsNested>::type>::RowsAtCompileTime==Dynamic)
+      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 (ei_traits<typename ei_cleantype<LhsNested>::type>::ColsAtCompileTime==Dynamic)
+      if (internal::traits<typename internal::remove_all<LhsNested>::type>::ColsAtCompileTime==Dynamic)
         return m_rhs.cols();
       else
         return m_lhs.cols();
     }
 
     /** \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;
 };
 
-template<typename BinaryOp, typename Lhs, typename Rhs>
-class CwiseBinaryOpImpl<BinaryOp, Lhs, Rhs, Dense>
-  : public ei_dense_xpr_base<CwiseBinaryOp<BinaryOp, Lhs, Rhs> >::type
+// Generic API dispatcher
+template<typename BinaryOp, typename Lhs, typename Rhs, typename StorageKind>
+class CwiseBinaryOpImpl
+  : public internal::generic_xpr_base<CwiseBinaryOp<BinaryOp, Lhs, Rhs> >::type
 {
-    typedef CwiseBinaryOp<BinaryOp, Lhs, Rhs> Derived;
-  public:
-
-    typedef typename ei_dense_xpr_base<CwiseBinaryOp<BinaryOp, Lhs, Rhs> >::type Base;
-    EIGEN_DENSE_PUBLIC_INTERFACE( Derived )
-
-    EIGEN_STRONG_INLINE const Scalar coeff(Index row, Index col) const
-    {
-      return derived().functor()(derived().lhs().coeff(row, col),
-                                 derived().rhs().coeff(row, col));
-    }
-
-    template<int LoadMode>
-    EIGEN_STRONG_INLINE PacketScalar packet(Index row, Index col) const
-    {
-      return derived().functor().packetOp(derived().lhs().template packet<LoadMode>(row, col),
-                                          derived().rhs().template packet<LoadMode>(row, col));
-    }
-
-    EIGEN_STRONG_INLINE const Scalar coeff(Index index) const
-    {
-      return derived().functor()(derived().lhs().coeff(index),
-                                 derived().rhs().coeff(index));
-    }
-
-    template<int LoadMode>
-    EIGEN_STRONG_INLINE PacketScalar packet(Index index) const
-    {
-      return derived().functor().packetOp(derived().lhs().template packet<LoadMode>(index),
-                                          derived().rhs().template packet<LoadMode>(index));
-    }
+public:
+  typedef typename internal::generic_xpr_base<CwiseBinaryOp<BinaryOp, Lhs, Rhs> >::type Base;
 };
 
 /** replaces \c *this by \c *this - \a other.
@@ -215,8 +161,7 @@ template<typename OtherDerived>
 EIGEN_STRONG_INLINE Derived &
 MatrixBase<Derived>::operator-=(const MatrixBase<OtherDerived> &other)
 {
-  SelfCwiseBinaryOp<ei_scalar_difference_op<Scalar>, Derived, OtherDerived> tmp(derived());
-  tmp = other.derived();
+  call_assignment(derived(), other.derived(), internal::sub_assign_op<Scalar>());
   return derived();
 }
 
@@ -229,9 +174,11 @@ template<typename OtherDerived>
 EIGEN_STRONG_INLINE Derived &
 MatrixBase<Derived>::operator+=(const MatrixBase<OtherDerived>& other)
 {
-  SelfCwiseBinaryOp<ei_scalar_sum_op<Scalar>, Derived, OtherDerived> tmp(derived());
-  tmp = other.derived();
+  call_assignment(derived(), other.derived(), internal::add_assign_op<Scalar>());
   return derived();
 }
 
+} // end namespace Eigen
+
 #endif // EIGEN_CWISE_BINARY_OP_H
+

Eigen/src/Core/CwiseNullaryOp.h

@@ -3,28 +3,15 @@
 //
 // 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 {
+
 /** \class CwiseNullaryOp
   * \ingroup Core_Module
   *
@@ -42,51 +29,53 @@
   *
   * \sa class CwiseUnaryOp, class CwiseBinaryOp, DenseBase::NullaryExpr()
   */
+
+namespace internal {
 template<typename NullaryOp, typename PlainObjectType>
-struct ei_traits<CwiseNullaryOp<NullaryOp, PlainObjectType> > : ei_traits<PlainObjectType>
+struct traits<CwiseNullaryOp<NullaryOp, PlainObjectType> > : traits<PlainObjectType>
 {
   enum {
-    Flags = (ei_traits<PlainObjectType>::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
+    Flags = traits<PlainObjectType>::Flags & RowMajorBit
   };
 };
+}
 
 template<typename NullaryOp, typename PlainObjectType>
-class CwiseNullaryOp : ei_no_assignment_operator,
-  public ei_dense_xpr_base< CwiseNullaryOp<NullaryOp, PlainObjectType> >::type
+class CwiseNullaryOp : public internal::dense_xpr_base< CwiseNullaryOp<NullaryOp, PlainObjectType> >::type, internal::no_assignment_operator
 {
   public:
 
-    typedef typename ei_dense_xpr_base<CwiseNullaryOp>::type Base;
+    typedef typename internal::dense_xpr_base<CwiseNullaryOp>::type Base;
     EIGEN_DENSE_PUBLIC_INTERFACE(CwiseNullaryOp)
 
+    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
+      eigen_assert(rows >= 0
             && (RowsAtCompileTime == Dynamic || RowsAtCompileTime == rows)
             &&  cols >= 0
             && (ColsAtCompileTime == Dynamic || ColsAtCompileTime == cols));
     }
 
+    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(Index rows, Index cols) const
+    EIGEN_DEVICE_FUNC
+    EIGEN_STRONG_INLINE const Scalar coeff(Index rowId, Index colId) const
     {
-      return m_functor(rows, cols);
+      return m_functor(rowId, colId);
     }
 
     template<int LoadMode>
-    EIGEN_STRONG_INLINE PacketScalar packet(Index row, Index col) const
+    EIGEN_STRONG_INLINE PacketScalar packet(Index rowId, Index colId) const
     {
-      return m_functor.packetOp(row, col);
+      return m_functor.packetOp(rowId, colId);
     }
 
+    EIGEN_DEVICE_FUNC
     EIGEN_STRONG_INLINE const Scalar coeff(Index index) const
     {
       return m_functor(index);
@@ -98,9 +87,13 @@ class CwiseNullaryOp : ei_no_assignment_operator,
       return m_functor.packetOp(index);
     }
 
+    /** \returns the functor representing the nullary operation */
+    EIGEN_DEVICE_FUNC
+    const NullaryOp& functor() const { return m_functor; }
+
   protected:
-    const ei_variable_if_dynamic<Index, RowsAtCompileTime> m_rows;
-    const ei_variable_if_dynamic<Index, 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 +113,10 @@ class CwiseNullaryOp : ei_no_assignment_operator,
   */
 template<typename Derived>
 template<typename CustomNullaryOp>
-EIGEN_STRONG_INLINE const CwiseNullaryOp<CustomNullaryOp, Derived>
+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
@@ -139,16 +132,19 @@ DenseBase<Derived>::NullaryExpr(Index rows, Index cols, const CustomNullaryOp& f
   *
   * 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>
+EIGEN_STRONG_INLINE const CwiseNullaryOp<CustomNullaryOp, typename DenseBase<Derived>::PlainObject>
 DenseBase<Derived>::NullaryExpr(Index size, const CustomNullaryOp& func)
 {
   EIGEN_STATIC_ASSERT_VECTOR_ONLY(Derived)
-  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
@@ -162,10 +158,10 @@ DenseBase<Derived>::NullaryExpr(Index size, const CustomNullaryOp& func)
   */
 template<typename Derived>
 template<typename CustomNullaryOp>
-EIGEN_STRONG_INLINE const CwiseNullaryOp<CustomNullaryOp, Derived>
+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
@@ -185,7 +181,7 @@ template<typename Derived>
 EIGEN_STRONG_INLINE const typename DenseBase<Derived>::ConstantReturnType
 DenseBase<Derived>::Constant(Index rows, Index cols, const Scalar& value)
 {
-  return DenseBase<Derived>::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
@@ -207,7 +203,7 @@ template<typename Derived>
 EIGEN_STRONG_INLINE const typename DenseBase<Derived>::ConstantReturnType
 DenseBase<Derived>::Constant(Index size, const Scalar& value)
 {
-  return DenseBase<Derived>::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
@@ -224,7 +220,7 @@ EIGEN_STRONG_INLINE const typename DenseBase<Derived>::ConstantReturnType
 DenseBase<Derived>::Constant(const Scalar& value)
 {
   EIGEN_STATIC_ASSERT_FIXED_SIZE(Derived)
-  return DenseBase<Derived>::NullaryExpr(RowsAtCompileTime, ColsAtCompileTime, ei_scalar_constant_op<Scalar>(value));
+  return DenseBase<Derived>::NullaryExpr(RowsAtCompileTime, ColsAtCompileTime, internal::scalar_constant_op<Scalar>(value));
 }
 
 /**
@@ -235,6 +231,8 @@ DenseBase<Derived>::Constant(const Scalar& value)
   * assumed to be a(0), a(1), ..., a(size). This assumption allows for better vectorization
   * and yields faster code than the random access version.
   *
+  * When size is set to 1, a vector of length 1 containing 'high' is returned.
+  *
   * \only_for_vectors
   *
   * Example: \include DenseBase_LinSpaced_seq.cpp
@@ -247,7 +245,7 @@ EIGEN_STRONG_INLINE const typename DenseBase<Derived>::SequentialLinSpacedReturn
 DenseBase<Derived>::LinSpaced(Sequential_t, Index size, const Scalar& low, const Scalar& high)
 {
   EIGEN_STATIC_ASSERT_VECTOR_ONLY(Derived)
-  return DenseBase<Derived>::NullaryExpr(size, ei_linspaced_op<Scalar,false>(low,high,size));
+  return DenseBase<Derived>::NullaryExpr(size, internal::linspaced_op<Scalar,PacketScalar,false>(low,high,size));
 }
 
 /**
@@ -260,13 +258,14 @@ DenseBase<Derived>::LinSpaced(Sequential_t, const Scalar& low, const Scalar& hig
 {
   EIGEN_STATIC_ASSERT_VECTOR_ONLY(Derived)
   EIGEN_STATIC_ASSERT_FIXED_SIZE(Derived)
-  return DenseBase<Derived>::NullaryExpr(Derived::SizeAtCompileTime, ei_linspaced_op<Scalar,false>(low,high,Derived::SizeAtCompileTime));
+  return DenseBase<Derived>::NullaryExpr(Derived::SizeAtCompileTime, internal::linspaced_op<Scalar,PacketScalar,false>(low,high,Derived::SizeAtCompileTime));
 }
 
 /**
   * \brief Sets a linearly space 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
   *
@@ -280,7 +279,7 @@ EIGEN_STRONG_INLINE const typename DenseBase<Derived>::RandomAccessLinSpacedRetu
 DenseBase<Derived>::LinSpaced(Index size, const Scalar& low, const Scalar& high)
 {
   EIGEN_STATIC_ASSERT_VECTOR_ONLY(Derived)
-  return DenseBase<Derived>::NullaryExpr(size, ei_linspaced_op<Scalar,true>(low,high,size));
+  return DenseBase<Derived>::NullaryExpr(size, internal::linspaced_op<Scalar,PacketScalar,true>(low,high,size));
 }
 
 /**
@@ -293,17 +292,18 @@ 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, ei_linspaced_op<Scalar,true>(low,high,Derived::SizeAtCompileTime));
+  return DenseBase<Derived>::NullaryExpr(Derived::SizeAtCompileTime, internal::linspaced_op<Scalar,PacketScalar,true>(low,high,Derived::SizeAtCompileTime));
 }
 
-/** \returns true if all coefficients in this matrix are approximately equal to \a value, to within precision \a prec */
+/** \returns true if all coefficients in this matrix are approximately equal to \a val, to within precision \a prec */
 template<typename Derived>
 bool DenseBase<Derived>::isApproxToConstant
-(const Scalar& value, RealScalar prec) const
+(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(!ei_isApprox(this->coeff(i, j), value, prec))
+      if(!internal::isApprox(self.coeff(i, j), val, prec))
         return false;
   return true;
 }
@@ -313,19 +313,19 @@ bool DenseBase<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 DenseBase<Derived>::isConstant
-(const Scalar& value, RealScalar prec) const
+(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 DenseBase<Derived>::fill(const Scalar& value)
+EIGEN_STRONG_INLINE void DenseBase<Derived>::fill(const Scalar& val)
 {
-  setConstant(value);
+  setConstant(val);
 }
 
 /** Sets all coefficients in this expression to \a value.
@@ -333,9 +333,9 @@ EIGEN_STRONG_INLINE void DenseBase<Derived>::fill(const Scalar& value)
   * \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& DenseBase<Derived>::setConstant(const Scalar& value)
+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.
@@ -349,17 +349,17 @@ EIGEN_STRONG_INLINE Derived& DenseBase<Derived>::setConstant(const Scalar& value
   */
 template<typename Derived>
 EIGEN_STRONG_INLINE Derived&
-DenseStorageBase<Derived>::setConstant(Index size, const Scalar& value)
+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.
   *
   * \param rows the new number of rows
   * \param cols the new number of columns
-  * \param value the value to which all coefficients are set
+  * \param val the value to which all coefficients are set
   *
   * Example: \include Matrix_setConstant_int_int.cpp
   * Output: \verbinclude Matrix_setConstant_int_int.out
@@ -368,16 +368,17 @@ DenseStorageBase<Derived>::setConstant(Index size, const Scalar& value)
   */
 template<typename Derived>
 EIGEN_STRONG_INLINE Derived&
-DenseStorageBase<Derived>::setConstant(Index rows, Index cols, const Scalar& value)
+PlainObjectBase<Derived>::setConstant(Index rows, Index cols, const Scalar& val)
 {
   resize(rows, cols);
-  return setConstant(value);
+  return setConstant(val);
 }
 
 /**
   * \brief Sets a linearly space 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
   *
@@ -387,10 +388,27 @@ DenseStorageBase<Derived>::setConstant(Index rows, Index cols, const Scalar& val
   * \sa CwiseNullaryOp
   */
 template<typename Derived>
-EIGEN_STRONG_INLINE Derived& DenseBase<Derived>::setLinSpaced(Index size, const Scalar& low, const Scalar& high)
+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(size, ei_linspaced_op<Scalar,false>(low,high,size));
+  return derived() = Derived::NullaryExpr(newSize, internal::linspaced_op<Scalar,PacketScalar,false>(low,high,newSize));
+}
+
+/**
+  * \brief Sets a linearly space vector.
+  *
+  * The function fill *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
+  *
+  * \sa setLinSpaced(Index, const Scalar&, const Scalar&), CwiseNullaryOp
+  */
+template<typename Derived>
+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:
@@ -465,11 +483,12 @@ DenseBase<Derived>::Zero()
   * \sa class CwiseNullaryOp, Zero()
   */
 template<typename Derived>
-bool DenseBase<Derived>::isZero(RealScalar prec) const
+bool DenseBase<Derived>::isZero(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(!ei_isMuchSmallerThan(this->coeff(i, j), static_cast<Scalar>(1), prec))
+      if(!internal::isMuchSmallerThan(self.coeff(i, j), static_cast<Scalar>(1), prec))
         return false;
   return true;
 }
@@ -498,9 +517,9 @@ EIGEN_STRONG_INLINE Derived& DenseBase<Derived>::setZero()
   */
 template<typename Derived>
 EIGEN_STRONG_INLINE Derived&
-DenseStorageBase<Derived>::setZero(Index size)
+PlainObjectBase<Derived>::setZero(Index newSize)
 {
-  resize(size);
+  resize(newSize);
   return setConstant(Scalar(0));
 }
 
@@ -516,7 +535,7 @@ DenseStorageBase<Derived>::setZero(Index size)
   */
 template<typename Derived>
 EIGEN_STRONG_INLINE Derived&
-DenseStorageBase<Derived>::setZero(Index rows, Index cols)
+PlainObjectBase<Derived>::setZero(Index rows, Index cols)
 {
   resize(rows, cols);
   return setConstant(Scalar(0));
@@ -547,7 +566,7 @@ DenseBase<Derived>::Ones(Index rows, Index cols)
 
 /** \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
@@ -563,9 +582,9 @@ DenseBase<Derived>::Ones(Index rows, Index cols)
   */
 template<typename Derived>
 EIGEN_STRONG_INLINE const typename DenseBase<Derived>::ConstantReturnType
-DenseBase<Derived>::Ones(Index size)
+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.
@@ -595,7 +614,7 @@ DenseBase<Derived>::Ones()
   */
 template<typename Derived>
 bool DenseBase<Derived>::isOnes
-(RealScalar prec) const
+(const RealScalar& prec) const
 {
   return isApproxToConstant(Scalar(1), prec);
 }
@@ -613,7 +632,7 @@ 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
   *
@@ -624,9 +643,9 @@ EIGEN_STRONG_INLINE Derived& DenseBase<Derived>::setOnes()
   */
 template<typename Derived>
 EIGEN_STRONG_INLINE Derived&
-DenseStorageBase<Derived>::setOnes(Index size)
+PlainObjectBase<Derived>::setOnes(Index newSize)
 {
-  resize(size);
+  resize(newSize);
   return setConstant(Scalar(1));
 }
 
@@ -642,7 +661,7 @@ DenseStorageBase<Derived>::setOnes(Index size)
   */
 template<typename Derived>
 EIGEN_STRONG_INLINE Derived&
-DenseStorageBase<Derived>::setOnes(Index rows, Index cols)
+PlainObjectBase<Derived>::setOnes(Index rows, Index cols)
 {
   resize(rows, cols);
   return setConstant(Scalar(1));
@@ -668,7 +687,7 @@ template<typename Derived>
 EIGEN_STRONG_INLINE const typename MatrixBase<Derived>::IdentityReturnType
 MatrixBase<Derived>::Identity(Index rows, Index cols)
 {
-  return DenseBase<Derived>::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).
@@ -686,7 +705,7 @@ EIGEN_STRONG_INLINE const typename MatrixBase<Derived>::IdentityReturnType
 MatrixBase<Derived>::Identity()
 {
   EIGEN_STATIC_ASSERT_FIXED_SIZE(Derived)
-  return MatrixBase<Derived>::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
@@ -700,20 +719,21 @@ MatrixBase<Derived>::Identity()
   */
 template<typename Derived>
 bool MatrixBase<Derived>::isIdentity
-(RealScalar prec) const
+(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(i == j)
       {
-        if(!ei_isApprox(this->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(this->coeff(i, j), static_cast<RealScalar>(1), prec))
+        if(!internal::isMuchSmallerThan(self.coeff(i, j), static_cast<RealScalar>(1), prec))
           return false;
       }
     }
@@ -721,9 +741,12 @@ 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());
@@ -731,18 +754,20 @@ struct ei_setIdentity_impl
 };
 
 template<typename Derived>
-struct ei_setIdentity_impl<Derived, true>
+struct setIdentity_impl<Derived, true>
 {
-  typedef typename Derived::Index Index;
+  EIGEN_DEVICE_FUNC
   static EIGEN_STRONG_INLINE Derived& run(Derived& m)
   {
     m.setZero();
-    const Index size = std::min(m.rows(), m.cols());
+    const Index size = (std::min)(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
@@ -753,7 +778,7 @@ struct ei_setIdentity_impl<Derived, true>
 template<typename Derived>
 EIGEN_STRONG_INLINE Derived& MatrixBase<Derived>::setIdentity()
 {
-  return ei_setIdentity_impl<Derived>::run(derived());
+  return internal::setIdentity_impl<Derived>::run(derived());
 }
 
 /** \brief Resizes to the given size, and writes the identity expression (not necessarily square) into *this.
@@ -780,10 +805,10 @@ EIGEN_STRONG_INLINE Derived& MatrixBase<Derived>::setIdentity(Index rows, Index
   * \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(Index size, Index i)
+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.
@@ -841,4 +866,6 @@ template<typename Derived>
 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/CwiseUnaryOp.h

@@ -1,31 +1,18 @@
 // This file is part of Eigen, a lightweight C++ template library
 // for linear algebra.
 //
-// Copyright (C) 2008-2010 Gael Guennebaud <gael.guennebaud@inria.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 { 
+
 /** \class CwiseUnaryOp
   * \ingroup Core_Module
   *
@@ -45,90 +32,72 @@
   *
   * \sa MatrixBase::unaryExpr(const CustomUnaryOp &) const, class CwiseBinaryOp, class CwiseNullaryOp
   */
+
+namespace internal {
 template<typename UnaryOp, typename XprType>
-struct ei_traits<CwiseUnaryOp<UnaryOp, XprType> >
- : ei_traits<XprType>
+struct traits<CwiseUnaryOp<UnaryOp, XprType> >
+ : traits<XprType>
 {
-  typedef typename ei_result_of<
+  typedef typename result_of<
                      UnaryOp(typename XprType::Scalar)
                    >::type Scalar;
   typedef typename XprType::Nested XprTypeNested;
-  typedef typename ei_unref<XprTypeNested>::type _XprTypeNested;
+  typedef typename remove_reference<XprTypeNested>::type _XprTypeNested;
   enum {
-    Flags = _XprTypeNested::Flags & (
-      HereditaryBits | LinearAccessBit | AlignedBit
-      | (ei_functor_traits<UnaryOp>::PacketAccess ? PacketAccessBit : 0)),
-    CoeffReadCost = _XprTypeNested::CoeffReadCost + ei_functor_traits<UnaryOp>::Cost
+    Flags = _XprTypeNested::Flags & RowMajorBit 
   };
 };
+}
 
 template<typename UnaryOp, typename XprType, typename StorageKind>
 class CwiseUnaryOpImpl;
 
 template<typename UnaryOp, typename XprType>
-class CwiseUnaryOp : ei_no_assignment_operator,
-  public CwiseUnaryOpImpl<UnaryOp, XprType, typename ei_traits<XprType>::StorageKind>
+class CwiseUnaryOp : public CwiseUnaryOpImpl<UnaryOp, XprType, typename internal::traits<XprType>::StorageKind>, internal::no_assignment_operator
 {
   public:
 
-    typedef typename CwiseUnaryOpImpl<UnaryOp, XprType,typename ei_traits<XprType>::StorageKind>::Base Base;
+    typedef typename CwiseUnaryOpImpl<UnaryOp, XprType,typename internal::traits<XprType>::StorageKind>::Base Base;
     EIGEN_GENERIC_PUBLIC_INTERFACE(CwiseUnaryOp)
+    typedef typename internal::remove_all<XprType>::type NestedExpression;
 
-    inline CwiseUnaryOp(const XprType& xpr, const UnaryOp& func = UnaryOp())
+    EIGEN_DEVICE_FUNC
+    explicit inline CwiseUnaryOp(const XprType& xpr, const UnaryOp& func = UnaryOp())
       : m_xpr(xpr), m_functor(func) {}
 
+    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(); }
 
     /** \returns the functor representing the unary operation */
+    EIGEN_DEVICE_FUNC
     const UnaryOp& functor() const { return m_functor; }
 
     /** \returns the nested expression */
-    const typename ei_cleantype<typename XprType::Nested>::type&
+    EIGEN_DEVICE_FUNC
+    const typename internal::remove_all<typename XprType::Nested>::type&
     nestedExpression() const { return m_xpr; }
 
     /** \returns the nested expression */
-    typename ei_cleantype<typename XprType::Nested>::type&
+    EIGEN_DEVICE_FUNC
+    typename internal::remove_all<typename XprType::Nested>::type&
     nestedExpression() { return m_xpr.const_cast_derived(); }
 
   protected:
-    const typename XprType::Nested m_xpr;
+    typename XprType::Nested m_xpr;
     const UnaryOp m_functor;
 };
 
-// This is the generic implementation for dense storage.
-// It can be used for any expression types implementing the dense concept.
-template<typename UnaryOp, typename XprType>
-class CwiseUnaryOpImpl<UnaryOp,XprType,Dense>
-  : public ei_dense_xpr_base<CwiseUnaryOp<UnaryOp, XprType> >::type
+// Generic API dispatcher
+template<typename UnaryOp, typename XprType, typename StorageKind>
+class CwiseUnaryOpImpl
+  : public internal::generic_xpr_base<CwiseUnaryOp<UnaryOp, XprType> >::type
 {
-  public:
-
-    typedef CwiseUnaryOp<UnaryOp, XprType> Derived;
-    typedef typename ei_dense_xpr_base<CwiseUnaryOp<UnaryOp, XprType> >::type Base;
-    EIGEN_DENSE_PUBLIC_INTERFACE(Derived)
-
-    EIGEN_STRONG_INLINE const Scalar coeff(Index row, Index col) const
-    {
-      return derived().functor()(derived().nestedExpression().coeff(row, col));
-    }
-
-    template<int LoadMode>
-    EIGEN_STRONG_INLINE PacketScalar packet(Index row, Index col) const
-    {
-      return derived().functor().packetOp(derived().nestedExpression().template packet<LoadMode>(row, col));
-    }
-
-    EIGEN_STRONG_INLINE const Scalar coeff(Index index) const
-    {
-      return derived().functor()(derived().nestedExpression().coeff(index));
-    }
-
-    template<int LoadMode>
-    EIGEN_STRONG_INLINE PacketScalar packet(Index index) const
-    {
-      return derived().functor().packetOp(derived().nestedExpression().template packet<LoadMode>(index));
-    }
+public:
+  typedef typename internal::generic_xpr_base<CwiseUnaryOp<UnaryOp, XprType> >::type Base;
 };
 
+} // end namespace Eigen
+
 #endif // EIGEN_CWISE_UNARY_OP_H

Eigen/src/Core/CwiseUnaryView.h

@@ -3,28 +3,15 @@
 //
 // Copyright (C) 2009-2010 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_UNARY_VIEW_H
 #define EIGEN_CWISE_UNARY_VIEW_H
 
+namespace Eigen {
+
 /** \class CwiseUnaryView
   * \ingroup Core_Module
   *
@@ -38,42 +25,46 @@
   *
   * \sa MatrixBase::unaryViewExpr(const CustomUnaryOp &) const, class CwiseUnaryOp
   */
+
+namespace internal {
 template<typename ViewOp, typename MatrixType>
-struct ei_traits<CwiseUnaryView<ViewOp, MatrixType> >
- : ei_traits<MatrixType>
+struct traits<CwiseUnaryView<ViewOp, MatrixType> >
+ : traits<MatrixType>
 {
-  typedef typename ei_result_of<
-                     ViewOp(typename ei_traits<MatrixType>::Scalar)
+  typedef typename result_of<
+                     ViewOp(typename traits<MatrixType>::Scalar)
                    >::type Scalar;
   typedef typename MatrixType::Nested MatrixTypeNested;
-  typedef typename ei_cleantype<MatrixTypeNested>::type _MatrixTypeNested;
+  typedef typename remove_all<MatrixTypeNested>::type _MatrixTypeNested;
   enum {
-    Flags = (ei_traits<_MatrixTypeNested>::Flags & (HereditaryBits | LvalueBit | LinearAccessBit | DirectAccessBit)),
-    CoeffReadCost = ei_traits<_MatrixTypeNested>::CoeffReadCost + ei_functor_traits<ViewOp>::Cost,
-    MatrixTypeInnerStride =  ei_inner_stride_at_compile_time<MatrixType>::ret,
+    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 ei_traits<MatrixType>::Scalar) / sizeof(Scalar)),
-    OuterStrideAtCompileTime = ei_outer_stride_at_compile_time<MatrixType>::ret
+                             : 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;
 
 template<typename ViewOp, typename MatrixType>
-class CwiseUnaryView : ei_no_assignment_operator,
-  public CwiseUnaryViewImpl<ViewOp, MatrixType, typename ei_traits<MatrixType>::StorageKind>
+class CwiseUnaryView : public CwiseUnaryViewImpl<ViewOp, MatrixType, typename internal::traits<MatrixType>::StorageKind>
 {
   public:
 
-    typedef typename CwiseUnaryViewImpl<ViewOp, MatrixType,typename ei_traits<MatrixType>::StorageKind>::Base Base;
+    typedef typename CwiseUnaryViewImpl<ViewOp, MatrixType,typename internal::traits<MatrixType>::StorageKind>::Base Base;
     EIGEN_GENERIC_PUBLIC_INTERFACE(CwiseUnaryView)
+    typedef typename internal::remove_all<MatrixType>::type NestedExpression;
 
-    inline CwiseUnaryView(const MatrixType& mat, const ViewOp& func = ViewOp())
+    explicit inline CwiseUnaryView(MatrixType& mat, const ViewOp& func = ViewOp())
       : m_matrix(mat), m_functor(func) {}
 
     EIGEN_INHERIT_ASSIGNMENT_OPERATORS(CwiseUnaryView)
@@ -85,61 +76,53 @@ class CwiseUnaryView : ei_no_assignment_operator,
     const ViewOp& functor() const { return m_functor; }
 
     /** \returns the nested expression */
-    const typename ei_cleantype<typename MatrixType::Nested>::type&
+    const typename internal::remove_all<typename MatrixType::Nested>::type&
     nestedExpression() const { return m_matrix; }
 
     /** \returns the nested expression */
-    typename ei_cleantype<typename MatrixType::Nested>::type&
+    typename internal::remove_all<typename MatrixType::Nested>::type&
     nestedExpression() { return m_matrix.const_cast_derived(); }
 
   protected:
-    // FIXME changed from MatrixType::Nested because of a weird compilation error with sun CC
-    const typename ei_nested<MatrixType>::type m_matrix;
+    typename internal::ref_selector<MatrixType>::type 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 ei_dense_xpr_base< CwiseUnaryView<ViewOp, MatrixType> >::type
+  : public internal::dense_xpr_base< CwiseUnaryView<ViewOp, MatrixType> >::type
 {
   public:
 
     typedef CwiseUnaryView<ViewOp, MatrixType> Derived;
-    typedef typename ei_dense_xpr_base< CwiseUnaryView<ViewOp, MatrixType> >::type Base;
+    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)); }
 
-    inline Index innerStride() const
+    EIGEN_DEVICE_FUNC inline Index innerStride() const
     {
-      return derived().nestedExpression().innerStride() * sizeof(typename ei_traits<MatrixType>::Scalar) / sizeof(Scalar);
+      return derived().nestedExpression().innerStride() * sizeof(typename internal::traits<MatrixType>::Scalar) / sizeof(Scalar);
     }
 
-    inline Index outerStride() const
+    EIGEN_DEVICE_FUNC inline Index outerStride() const
     {
-      return derived().nestedExpression().outerStride();
-    }
-
-    EIGEN_STRONG_INLINE CoeffReturnType coeff(Index row, Index col) const
-    {
-      return derived().functor()(derived().nestedExpression().coeff(row, col));
-    }
-
-    EIGEN_STRONG_INLINE CoeffReturnType coeff(Index index) const
-    {
-      return derived().functor()(derived().nestedExpression().coeff(index));
-    }
-
-    EIGEN_STRONG_INLINE Scalar& coeffRef(Index row, Index col)
-    {
-      return derived().functor()(const_cast_derived().nestedExpression().coeffRef(row, col));
-    }
-
-    EIGEN_STRONG_INLINE Scalar& coeffRef(Index index)
-    {
-      return derived().functor()(const_cast_derived().nestedExpression().coeffRef(index));
+      return derived().nestedExpression().outerStride() * sizeof(typename internal::traits<MatrixType>::Scalar) / sizeof(Scalar);
     }
 };
 
-
+} // end namespace Eigen
 
 #endif // EIGEN_CWISE_UNARY_VIEW_H

Eigen/src/Core/DenseBase.h

@@ -4,28 +4,25 @@
 // Copyright (C) 2007-2010 Benoit Jacob <jacob.benoit.1@gmail.com>
 // 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_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
   *
@@ -34,31 +31,52 @@
   * 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.
   *
-  * \param Derived is the derived type, e.g., a matrix type or an expression.
+  * \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 TopicCustomizingEigen by defining the preprocessor symbol \c EIGEN_DENSEBASE_PLUGIN.
   *
   * \sa \ref TopicClassHierarchy
   */
 template<typename Derived> class DenseBase
 #ifndef EIGEN_PARSED_BY_DOXYGEN
-  : public ei_special_scalar_op_base<Derived,typename ei_traits<Derived>::Scalar,
-                                     typename NumTraits<typename ei_traits<Derived>::Scalar>::Real>
+  : public internal::special_scalar_op_base<Derived, typename internal::traits<Derived>::Scalar,
+                                            typename NumTraits<typename internal::traits<Derived>::Scalar>::Real,
+                                            DenseCoeffsBase<Derived> >
 #else
   : public DenseCoeffsBase<Derived>
 #endif // not EIGEN_PARSED_BY_DOXYGEN
 {
   public:
-    using ei_special_scalar_op_base<Derived,typename ei_traits<Derived>::Scalar,
-                typename NumTraits<typename ei_traits<Derived>::Scalar>::Real>::operator*;
 
-    class InnerIterator;
+    /** Inner iterator type to iterate over the coefficients of a row or column.
+      * \sa class InnerIterator
+      */
+    typedef Eigen::InnerIterator<Derived> InnerIterator;
 
-    typedef typename ei_traits<Derived>::StorageKind StorageKind;
-    typedef typename ei_traits<Derived>::Index Index; /**< The type of indices */
-    typedef typename ei_traits<Derived>::Scalar Scalar;
-    typedef typename ei_packet_traits<Scalar>::type PacketScalar;
+    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 \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 internal::special_scalar_op_base<Derived,Scalar,RealScalar, DenseCoeffsBase<Derived> > Base;
 
-    typedef DenseCoeffsBase<Derived> Base;
+    using Base::operator*;
+    using Base::operator/;
     using Base::derived;
     using Base::const_cast_derived;
     using Base::rows;
@@ -68,16 +86,6 @@ template<typename Derived> class DenseBase
     using Base::colIndexByOuterInner;
     using Base::coeff;
     using Base::coeffByOuterInner;
-    using Base::packet;
-    using Base::packetByOuterInner;
-    using Base::writePacket;
-    using Base::writePacketByOuterInner;
-    using Base::coeffRef;
-    using Base::coeffRefByOuterInner;
-    using Base::copyCoeff;
-    using Base::copyCoeffByOuterInner;
-    using Base::copyPacket;
-    using Base::copyPacketByOuterInner;
     using Base::operator();
     using Base::operator[];
     using Base::x;
@@ -93,26 +101,26 @@ template<typename Derived> class DenseBase
 
     enum {
 
-      RowsAtCompileTime = ei_traits<Derived>::RowsAtCompileTime,
+      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 = ei_traits<Derived>::ColsAtCompileTime,
+      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 = (ei_size_at_compile_time<ei_traits<Derived>::RowsAtCompileTime,
-                                                   ei_traits<Derived>::ColsAtCompileTime>::ret),
+      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 = ei_traits<Derived>::MaxRowsAtCompileTime,
+      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.
@@ -123,7 +131,7 @@ template<typename Derived> class DenseBase
           * \sa RowsAtCompileTime, MaxColsAtCompileTime, MaxSizeAtCompileTime
           */
 
-      MaxColsAtCompileTime = ei_traits<Derived>::MaxColsAtCompileTime,
+      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.
@@ -134,8 +142,8 @@ template<typename Derived> class DenseBase
           * \sa ColsAtCompileTime, MaxRowsAtCompileTime, MaxSizeAtCompileTime
           */
 
-      MaxSizeAtCompileTime = (ei_size_at_compile_time<ei_traits<Derived>::MaxRowsAtCompileTime,
-                                                      ei_traits<Derived>::MaxColsAtCompileTime>::ret),
+      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.
@@ -146,45 +154,71 @@ template<typename Derived> class DenseBase
           * \sa SizeAtCompileTime, MaxRowsAtCompileTime, MaxColsAtCompileTime
           */
 
-      IsVectorAtCompileTime = ei_traits<Derived>::MaxRowsAtCompileTime == 1
-                           || ei_traits<Derived>::MaxColsAtCompileTime == 1,
+      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 = ei_traits<Derived>::Flags,
+      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) ? SizeAtCompileTime
-                             : int(IsRowMajor) ? ColsAtCompileTime : RowsAtCompileTime,
+      InnerSizeAtCompileTime = int(IsVectorAtCompileTime) ? int(SizeAtCompileTime)
+                             : int(IsRowMajor) ? int(ColsAtCompileTime) : int(RowsAtCompileTime),
 
-      CoeffReadCost = ei_traits<Derived>::CoeffReadCost,
-        /**< This is a rough measure of how expensive it is to read one coefficient from
-          * this expression.
-          */
-
-      InnerStrideAtCompileTime = ei_inner_stride_at_compile_time<Derived>::ret,
-      OuterStrideAtCompileTime = ei_outer_stride_at_compile_time<Derived>::ret
+      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 true if either the number of rows or the number of columns is equal to 1.
-      * In other words, this function returns
-      * \code rows()==1 || cols()==1 \endcode
-      * \sa rows(), cols(), IsVectorAtCompileTime. */
 
     /** \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 storage order, i.e., the number of columns for a column-major matrix,
-      * and the number of rows for a row-major matrix. */
+      * 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
@@ -194,8 +228,9 @@ template<typename Derived> class DenseBase
     /** \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 storage order, i.e., the number of rows for a column-major matrix,
-      * and the number of columns for a row-major matrix. */
+      * 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()
@@ -206,236 +241,281 @@ template<typename Derived> class DenseBase
       * Matrix::resize() and Array::resize(). The present method only asserts that the new size equals the old size, and does
       * nothing else.
       */
-    void resize(Index size)
+    EIGEN_DEVICE_FUNC
+    void resize(Index newSize)
     {
-      EIGEN_ONLY_USED_FOR_DEBUG(size);
-      ei_assert(size == this->size()
+      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);
-      ei_assert(rows == this->rows() && cols == this->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<ei_scalar_constant_op<Scalar>,Derived> ConstantReturnType;
+    typedef CwiseNullaryOp<internal::scalar_constant_op<Scalar>,PlainObject> ConstantReturnType;
     /** \internal Represents a vector with linearly spaced coefficients that allows sequential access only. */
-    typedef CwiseNullaryOp<ei_linspaced_op<Scalar,false>,Derived> SequentialLinSpacedReturnType;
+    typedef CwiseNullaryOp<internal::linspaced_op<Scalar,PacketScalar,false>,PlainObject> SequentialLinSpacedReturnType;
     /** \internal Represents a vector with linearly spaced coefficients that allows random access. */
-    typedef CwiseNullaryOp<ei_linspaced_op<Scalar,true>,Derived> RandomAccessLinSpacedReturnType;
+    typedef CwiseNullaryOp<internal::linspaced_op<Scalar,PacketScalar,true>,PlainObject> RandomAccessLinSpacedReturnType;
     /** \internal the return type of MatrixBase::eigenvalues() */
-    typedef Matrix<typename NumTraits<typename ei_traits<Derived>::Scalar>::Real, ei_traits<Derived>::ColsAtCompileTime, 1> EigenvaluesReturnType;
+    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
     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
     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);
 
-#ifndef EIGEN_PARSED_BY_DOXYGEN
-    /** Copies \a other into *this without evaluating other. \returns a reference to *this. */
+    /** \ínternal
+      * 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);
-#endif // not EIGEN_PARSED_BY_DOXYGEN
 
+    EIGEN_DEVICE_FUNC
     CommaInitializer<Derived> operator<< (const Scalar& s);
 
+    /** \deprecated it now returns \c *this */
     template<unsigned int Added,unsigned int Removed>
-    const Flagged<Derived, Added, Removed> flagged() const;
+    EIGEN_DEPRECATED
+    const Derived& flagged() const
+    { return derived(); }
 
     template<typename OtherDerived>
+    EIGEN_DEVICE_FUNC
     CommaInitializer<Derived> operator<< (const DenseBase<OtherDerived>& other);
 
-    Eigen::Transpose<Derived> transpose();
-    const Eigen::Transpose<Derived> transpose() const;
+    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();
-#ifndef EIGEN_NO_DEBUG
-  protected:
-    template<typename OtherDerived>
-    void checkTransposeAliasing(const OtherDerived& other) const;
-  public:
-#endif
 
-    VectorBlock<Derived> segment(Index start, Index size);
-    const VectorBlock<Derived> segment(Index start, Index size) const;
-
-    VectorBlock<Derived> head(Index size);
-    const VectorBlock<Derived> head(Index size) const;
-
-    VectorBlock<Derived> tail(Index size);
-    const VectorBlock<Derived> tail(Index size) const;
-
-    template<int Size> VectorBlock<Derived,Size> head(void);
-    template<int Size> const VectorBlock<Derived,Size> head() const;
-
-    template<int Size> VectorBlock<Derived,Size> tail();
-    template<int Size> const VectorBlock<Derived,Size> tail() const;
-
-    template<int Size> VectorBlock<Derived,Size> segment(Index start);
-    template<int Size> const VectorBlock<Derived,Size> segment(Index start) const;
-
-    Diagonal<Derived,0> diagonal();
-    const Diagonal<Derived,0> diagonal() const;
-
-    template<int Index> Diagonal<Derived,Index> diagonal();
-    template<int Index> const Diagonal<Derived,Index> diagonal() const;
-
-    Diagonal<Derived, Dynamic> diagonal(Index index);
-    const Diagonal<Derived, Dynamic> diagonal(Index index) const;
-
-    template<unsigned int Mode> TriangularView<Derived, Mode> part();
-    template<unsigned int Mode> const TriangularView<Derived, Mode> part() const;
-
-    template<unsigned int Mode> TriangularView<Derived, Mode> triangularView();
-    template<unsigned int Mode> const TriangularView<Derived, Mode> triangularView() const;
-
-    template<unsigned int UpLo> SelfAdjointView<Derived, UpLo> selfadjointView();
-    template<unsigned int UpLo> const SelfAdjointView<Derived, UpLo> selfadjointView() const;
-
-    static const ConstantReturnType
+    EIGEN_DEVICE_FUNC static const ConstantReturnType
     Constant(Index rows, Index cols, const Scalar& value);
-    static const ConstantReturnType
+    EIGEN_DEVICE_FUNC static const ConstantReturnType
     Constant(Index size, const Scalar& value);
-    static const ConstantReturnType
+    EIGEN_DEVICE_FUNC static const ConstantReturnType
     Constant(const Scalar& value);
 
-    static const SequentialLinSpacedReturnType
+    EIGEN_DEVICE_FUNC static const SequentialLinSpacedReturnType
     LinSpaced(Sequential_t, Index size, const Scalar& low, const Scalar& high);
-    static const RandomAccessLinSpacedReturnType
+    EIGEN_DEVICE_FUNC static const RandomAccessLinSpacedReturnType
     LinSpaced(Index size, const Scalar& low, const Scalar& high);
-    static const SequentialLinSpacedReturnType
+    EIGEN_DEVICE_FUNC static const SequentialLinSpacedReturnType
     LinSpaced(Sequential_t, const Scalar& low, const Scalar& high);
-    static const RandomAccessLinSpacedReturnType
+    EIGEN_DEVICE_FUNC static const RandomAccessLinSpacedReturnType
     LinSpaced(const Scalar& low, const Scalar& high);
 
-    template<typename CustomNullaryOp>
-    static const CwiseNullaryOp<CustomNullaryOp, Derived>
+    template<typename CustomNullaryOp> EIGEN_DEVICE_FUNC
+    static const CwiseNullaryOp<CustomNullaryOp, PlainObject>
     NullaryExpr(Index rows, Index cols, const CustomNullaryOp& func);
-    template<typename CustomNullaryOp>
-    static const CwiseNullaryOp<CustomNullaryOp, Derived>
+    template<typename CustomNullaryOp> EIGEN_DEVICE_FUNC
+    static const CwiseNullaryOp<CustomNullaryOp, PlainObject>
     NullaryExpr(Index size, const CustomNullaryOp& func);
-    template<typename CustomNullaryOp>
-    static const CwiseNullaryOp<CustomNullaryOp, Derived>
+    template<typename CustomNullaryOp> EIGEN_DEVICE_FUNC
+    static const CwiseNullaryOp<CustomNullaryOp, PlainObject>
     NullaryExpr(const CustomNullaryOp& func);
 
-    static const ConstantReturnType Zero(Index rows, Index cols);
-    static const ConstantReturnType Zero(Index size);
-    static const ConstantReturnType Zero();
-    static const ConstantReturnType Ones(Index rows, Index cols);
-    static const ConstantReturnType Ones(Index size);
-    static const ConstantReturnType Ones();
+    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();
 
-    void fill(const Scalar& value);
-    Derived& setConstant(const Scalar& value);
-    Derived& setLinSpaced(Index size, const Scalar& low, const Scalar& high);
-    Derived& setLinSpaced(const Scalar& low, const Scalar& high);
-    Derived& setZero();
-    Derived& setOnes();
-    Derived& setRandom();
+    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>
+    template<typename OtherDerived> EIGEN_DEVICE_FUNC
     bool isApprox(const DenseBase<OtherDerived>& other,
-                  RealScalar prec = NumTraits<Scalar>::dummy_precision()) const;
+                  const RealScalar& prec = NumTraits<Scalar>::dummy_precision()) const;
+    EIGEN_DEVICE_FUNC 
     bool isMuchSmallerThan(const RealScalar& other,
-                           RealScalar prec = NumTraits<Scalar>::dummy_precision()) const;
-    template<typename OtherDerived>
+                           const RealScalar& prec = NumTraits<Scalar>::dummy_precision()) const;
+    template<typename OtherDerived> EIGEN_DEVICE_FUNC
     bool isMuchSmallerThan(const DenseBase<OtherDerived>& other,
-                           RealScalar prec = NumTraits<Scalar>::dummy_precision()) const;
+                           const RealScalar& prec = NumTraits<Scalar>::dummy_precision()) const;
 
-    bool isApproxToConstant(const Scalar& value, RealScalar prec = NumTraits<Scalar>::dummy_precision()) const;
-    bool isConstant(const Scalar& value, RealScalar prec = NumTraits<Scalar>::dummy_precision()) const;
-    bool isZero(RealScalar prec = NumTraits<Scalar>::dummy_precision()) const;
-    bool isOnes(RealScalar prec = NumTraits<Scalar>::dummy_precision()) const;
+    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
     inline Derived& operator*=(const Scalar& other);
+    EIGEN_DEVICE_FUNC
     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_STRONG_INLINE const typename ei_eval<Derived>::type eval() const
+    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 ei_eval<Derived>::type(derived());
+      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>
-    void swap(DenseBase<OtherDerived> EIGEN_REF_TO_TEMPORARY other);
+    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>());
+    }
 
-    inline const NestByValue<Derived> nestByValue() const;
-    inline const ForceAlignedAccess<Derived> forceAlignedAccess() const;
-    inline ForceAlignedAccess<Derived> forceAlignedAccess();
-    template<bool Enable> inline const typename ei_meta_if<Enable,ForceAlignedAccess<Derived>,Derived&>::ret forceAlignedAccessIf() const;
-    template<bool Enable> inline typename ei_meta_if<Enable,ForceAlignedAccess<Derived>,Derived&>::ret forceAlignedAccessIf();
+    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();
 
-    Scalar sum() const;
-    Scalar mean() const;
-    Scalar trace() const;
+    EIGEN_DEVICE_FUNC Scalar sum() const;
+    EIGEN_DEVICE_FUNC Scalar mean() const;
+    EIGEN_DEVICE_FUNC Scalar trace() const;
 
-    Scalar prod() const;
+    EIGEN_DEVICE_FUNC Scalar prod() const;
 
-    typename ei_traits<Derived>::Scalar minCoeff() const;
-    typename ei_traits<Derived>::Scalar maxCoeff() const;
+    EIGEN_DEVICE_FUNC typename internal::traits<Derived>::Scalar minCoeff() const;
+    EIGEN_DEVICE_FUNC typename internal::traits<Derived>::Scalar maxCoeff() const;
 
-    typename ei_traits<Derived>::Scalar minCoeff(Index* row, Index* col) const;
-    typename ei_traits<Derived>::Scalar maxCoeff(Index* row, Index* col) const;
-
-    typename ei_traits<Derived>::Scalar minCoeff(Index* index) const;
-    typename ei_traits<Derived>::Scalar maxCoeff(Index* index) 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>
-    typename ei_result_of<BinaryOp(typename ei_traits<Derived>::Scalar)>::type
-    redux(const BinaryOp& func) const;
+    EIGEN_DEVICE_FUNC
+    Scalar redux(const BinaryOp& func) const;
 
     template<typename Visitor>
+    EIGEN_DEVICE_FUNC
     void visit(Visitor& func) const;
 
     inline const WithFormat<Derived> format(const IOFormat& fmt) const;
 
-/////////// Array module ///////////
+    /** \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);
+    }
 
-    bool all(void) const;
-    bool any(void) const;
+    bool all() const;
+    bool any() const;
     Index count() const;
 
-    const VectorwiseOp<Derived,Horizontal> rowwise() const;
-    VectorwiseOp<Derived,Horizontal> rowwise();
-    const VectorwiseOp<Derived,Vertical> colwise() const;
-    VectorwiseOp<Derived,Vertical> colwise();
+    typedef VectorwiseOp<Derived, Horizontal> RowwiseReturnType;
+    typedef const VectorwiseOp<const Derived, Horizontal> ConstRowwiseReturnType;
+    typedef VectorwiseOp<Derived, Vertical> ColwiseReturnType;
+    typedef const VectorwiseOp<const Derived, Vertical> ConstColwiseReturnType;
 
-    static const CwiseNullaryOp<ei_scalar_random_op<Scalar>,Derived> Random(Index rows, Index cols);
-    static const CwiseNullaryOp<ei_scalar_random_op<Scalar>,Derived> Random(Index size);
-    static const CwiseNullaryOp<ei_scalar_random_op<Scalar>,Derived> Random();
+    /** \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>
@@ -444,21 +524,42 @@ template<typename Derived> class DenseBase
 
     template<typename ThenDerived>
     inline const Select<Derived,ThenDerived, typename ThenDerived::ConstantReturnType>
-    select(const DenseBase<ThenDerived>& thenMatrix, typename ThenDerived::Scalar elseScalar) const;
+    select(const DenseBase<ThenDerived>& thenMatrix, const typename ThenDerived::Scalar& elseScalar) const;
 
     template<typename ElseDerived>
     inline const Select<Derived, typename ElseDerived::ConstantReturnType, ElseDerived >
-    select(typename ElseDerived::Scalar thenScalar, const DenseBase<ElseDerived>& elseMatrix) const;
+    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;
-    const Replicate<Derived,Dynamic,Dynamic> replicate(Index rowFacor,Index colFactor) 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);
+    }
 
-    Eigen::Reverse<Derived, BothDirections> reverse();
-    const Eigen::Reverse<Derived, BothDirections> reverse() const;
-    void reverseInPlace();
+    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
 #   include "../plugins/BlockMethods.h"
@@ -467,34 +568,23 @@ template<typename Derived> class DenseBase
 #   endif
 #undef EIGEN_CURRENT_STORAGE_BASE_CLASS
 
-#ifdef EIGEN2_SUPPORT
-
-    Block<Derived> corner(CornerType type, Index cRows, Index cCols);
-    const Block<Derived> corner(CornerType type, Index cRows, Index cCols) const;
-    template<int CRows, int CCols>
-    Block<Derived, CRows, CCols> corner(CornerType type);
-    template<int CRows, int CCols>
-    const Block<Derived, CRows, CCols> corner(CornerType type) const;
-
-#endif // EIGEN2_SUPPORT
-
 
     // disable the use of evalTo for dense objects with a nice compilation error
-    template<typename Dest> inline void evalTo(Dest& ) const
+    template<typename Dest>
+    EIGEN_DEVICE_FUNC
+    inline void evalTo(Dest& ) const
     {
-      EIGEN_STATIC_ASSERT((ei_is_same_type<Dest,void>::ret),THE_EVAL_EVALTO_FUNCTION_SHOULD_NEVER_BE_CALLED_FOR_DENSE_OBJECTS);
+      EIGEN_STATIC_ASSERT((internal::is_same<Dest,void>::value),THE_EVAL_EVALTO_FUNCTION_SHOULD_NEVER_BE_CALLED_FOR_DENSE_OBJECTS);
     }
 
   protected:
     /** Default constructor. Do nothing. */
-    DenseBase()
+    EIGEN_DEVICE_FUNC DenseBase()
     {
       /* Just checks for self-consistency of the flags.
        * Only do it when debugging Eigen, as this borders on paranoiac and could slow compilation down
        */
 #ifdef EIGEN_INTERNAL_DEBUGGING
-      EIGEN_STATIC_ASSERT(ei_are_flags_consistent<Flags>::ret,
-                          INVALID_MATRIXBASE_TEMPLATE_PARAMETERS)
       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)
@@ -502,9 +592,11 @@ template<typename Derived> class DenseBase
     }
 
   private:
-    explicit DenseBase(int);
-    DenseBase(int,int);
-    template<typename OtherDerived> explicit DenseBase(const DenseBase<OtherDerived>&);
+    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

@@ -3,32 +3,26 @@
 //
 // Copyright (C) 2006-2010 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_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
+  * \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.
@@ -40,15 +34,25 @@ template<typename Derived>
 class DenseCoeffsBase<Derived,ReadOnlyAccessors> : public EigenBase<Derived>
 {
   public:
-    typedef typename ei_traits<Derived>::StorageKind StorageKind;
-    typedef typename ei_traits<Derived>::Index Index;
-    typedef typename ei_traits<Derived>::Scalar Scalar;
-    typedef typename ei_packet_traits<Scalar>::type PacketScalar;
-    typedef typename ei_meta_if<bool(ei_traits<Derived>::Flags&LvalueBit),
-                                const Scalar&,
-                                typename ei_meta_if<ei_is_arithmetic<Scalar>::ret, Scalar, const Scalar>::ret
-                               >::ret CoeffReturnType;
-    typedef typename ei_makeconst_return_type<typename ei_packet_traits<Scalar>::type>::type PacketReturnType;
+    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;
@@ -56,6 +60,7 @@ class DenseCoeffsBase<Derived,ReadOnlyAccessors> : public EigenBase<Derived>
     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
@@ -64,6 +69,7 @@ class DenseCoeffsBase<Derived,ReadOnlyAccessors> : public EigenBase<Derived>
           : inner;
     }
 
+    EIGEN_DEVICE_FUNC
     EIGEN_STRONG_INLINE Index colIndexByOuterInner(Index outer, Index inner) const
     {
       return int(Derived::ColsAtCompileTime) == 1 ? 0
@@ -86,13 +92,15 @@ class DenseCoeffsBase<Derived,ReadOnlyAccessors> : public EigenBase<Derived>
       *
       * \sa operator()(Index,Index) const, coeffRef(Index,Index), coeff(Index) const
       */
+    EIGEN_DEVICE_FUNC
     EIGEN_STRONG_INLINE CoeffReturnType coeff(Index row, Index col) const
     {
-      ei_internal_assert(row >= 0 && row < rows()
-                        && col >= 0 && col < cols());
-      return derived().coeff(row, col);
+      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),
@@ -103,11 +111,12 @@ class DenseCoeffsBase<Derived,ReadOnlyAccessors> : public EigenBase<Derived>
       *
       * \sa operator()(Index,Index), operator[](Index)
       */
+    EIGEN_DEVICE_FUNC
     EIGEN_STRONG_INLINE CoeffReturnType operator()(Index row, Index col) const
     {
-      ei_assert(row >= 0 && row < rows()
+      eigen_assert(row >= 0 && row < rows()
           && col >= 0 && col < cols());
-      return derived().coeff(row, col);
+      return coeff(row, col);
     }
 
     /** Short version: don't use this function, use
@@ -125,11 +134,14 @@ class DenseCoeffsBase<Derived,ReadOnlyAccessors> : public EigenBase<Derived>
       * \sa operator[](Index) const, coeffRef(Index), coeff(Index,Index) const
       */
 
+    EIGEN_DEVICE_FUNC
     EIGEN_STRONG_INLINE CoeffReturnType
     coeff(Index index) const
     {
-      ei_internal_assert(index >= 0 && index < size());
-      return derived().coeff(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()).coeff(index);
     }
 
 
@@ -141,13 +153,14 @@ class DenseCoeffsBase<Derived,ReadOnlyAccessors> : public EigenBase<Derived>
       * 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)
-      ei_assert(index >= 0 && index < size());
-      return derived().coeff(index);
+      eigen_assert(index >= 0 && index < size());
+      return coeff(index);
     }
 
     /** \returns the coefficient at given index.
@@ -160,38 +173,44 @@ class DenseCoeffsBase<Derived,ReadOnlyAccessors> : public EigenBase<Derived>
       * z() const, w() const
       */
 
+    EIGEN_DEVICE_FUNC
     EIGEN_STRONG_INLINE CoeffReturnType
     operator()(Index index) const
     {
-      ei_assert(index >= 0 && index < size());
-      return derived().coeff(index);
+      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 { return (*this)[1]; }
 
     /** equivalent to operator[](2).  */
 
+    EIGEN_DEVICE_FUNC
     EIGEN_STRONG_INLINE CoeffReturnType
     z() const { return (*this)[2]; }
 
     /** equivalent to operator[](3).  */
 
+    EIGEN_DEVICE_FUNC
     EIGEN_STRONG_INLINE CoeffReturnType
     w() const { return (*this)[3]; }
 
-    /** \returns the packet of coefficients starting at the given row and column. It is your responsibility
+    /** \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 \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.
       */
@@ -199,12 +218,13 @@ class DenseCoeffsBase<Derived,ReadOnlyAccessors> : public EigenBase<Derived>
     template<int LoadMode>
     EIGEN_STRONG_INLINE PacketReturnType packet(Index row, Index col) const
     {
-      ei_internal_assert(row >= 0 && row < rows()
-                      && col >= 0 && col < cols());
-      return derived().template packet<LoadMode>(row,col);
+      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
     {
@@ -212,11 +232,12 @@ class DenseCoeffsBase<Derived,ReadOnlyAccessors> : public EigenBase<Derived>
                               colIndexByOuterInner(outer, inner));
     }
 
-    /** \returns the packet of coefficients starting at the given index. It is your responsibility
+    /** \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 \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.
       */
@@ -224,13 +245,16 @@ class DenseCoeffsBase<Derived,ReadOnlyAccessors> : public EigenBase<Derived>
     template<int LoadMode>
     EIGEN_STRONG_INLINE PacketReturnType packet(Index index) const
     {
-      ei_internal_assert(index >= 0 && index < size());
-      return derived().template packet<LoadMode>(index);
+      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 EnableDirectAccessAPI case.
+    // 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.
@@ -252,7 +276,7 @@ class DenseCoeffsBase<Derived,ReadOnlyAccessors> : public EigenBase<Derived>
 /** \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
+  * \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
@@ -267,10 +291,9 @@ class DenseCoeffsBase<Derived, WriteAccessors> : public DenseCoeffsBase<Derived,
 
     typedef DenseCoeffsBase<Derived, ReadOnlyAccessors> Base;
 
-    typedef typename ei_traits<Derived>::StorageKind StorageKind;
-    typedef typename ei_traits<Derived>::Index Index;
-    typedef typename ei_traits<Derived>::Scalar Scalar;
-    typedef typename ei_packet_traits<Scalar>::type PacketScalar;
+    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;
@@ -301,13 +324,15 @@ class DenseCoeffsBase<Derived, WriteAccessors> : public DenseCoeffsBase<Derived,
       *
       * \sa operator()(Index,Index), coeff(Index, Index) const, coeffRef(Index)
       */
+    EIGEN_DEVICE_FUNC
     EIGEN_STRONG_INLINE Scalar& coeffRef(Index row, Index col)
     {
-      ei_internal_assert(row >= 0 && row < rows()
-                        && col >= 0 && col < cols());
-      return derived().coeffRef(row, 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)
     {
@@ -320,12 +345,13 @@ class DenseCoeffsBase<Derived, WriteAccessors> : public DenseCoeffsBase<Derived,
       * \sa operator[](Index)
       */
 
+    EIGEN_DEVICE_FUNC
     EIGEN_STRONG_INLINE Scalar&
     operator()(Index row, Index col)
     {
-      ei_assert(row >= 0 && row < rows()
+      eigen_assert(row >= 0 && row < rows()
           && col >= 0 && col < cols());
-      return derived().coeffRef(row, col);
+      return coeffRef(row, col);
     }
 
 
@@ -344,11 +370,14 @@ class DenseCoeffsBase<Derived, WriteAccessors> : public DenseCoeffsBase<Derived,
       * \sa operator[](Index), coeff(Index) const, coeffRef(Index,Index)
       */
 
+    EIGEN_DEVICE_FUNC
     EIGEN_STRONG_INLINE Scalar&
     coeffRef(Index index)
     {
-      ei_internal_assert(index >= 0 && index < size());
-      return derived().coeffRef(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.
@@ -358,13 +387,14 @@ class DenseCoeffsBase<Derived, WriteAccessors> : public DenseCoeffsBase<Derived,
       * \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)
-      ei_assert(index >= 0 && index < size());
-      return derived().coeffRef(index);
+      eigen_assert(index >= 0 && index < size());
+      return coeffRef(index);
     }
 
     /** \returns a reference to the coefficient at given index.
@@ -376,185 +406,57 @@ class DenseCoeffsBase<Derived, WriteAccessors> : public DenseCoeffsBase<Derived,
       * \sa operator[](Index) const, operator()(Index,Index), x(), y(), z(), w()
       */
 
+    EIGEN_DEVICE_FUNC
     EIGEN_STRONG_INLINE Scalar&
     operator()(Index index)
     {
-      ei_assert(index >= 0 && index < size());
-      return derived().coeffRef(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() { return (*this)[1]; }
 
     /** equivalent to operator[](2).  */
 
+    EIGEN_DEVICE_FUNC
     EIGEN_STRONG_INLINE Scalar&
     z() { return (*this)[2]; }
 
     /** equivalent to operator[](3).  */
 
+    EIGEN_DEVICE_FUNC
     EIGEN_STRONG_INLINE Scalar&
     w() { return (*this)[3]; }
-
-    /** 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<int StoreMode>
-    EIGEN_STRONG_INLINE void writePacket
-    (Index row, Index col, const typename ei_packet_traits<Scalar>::type& x)
-    {
-      ei_internal_assert(row >= 0 && row < rows()
-                        && col >= 0 && col < cols());
-      derived().template writePacket<StoreMode>(row,col,x);
-    }
-
-
-    template<int StoreMode>
-    EIGEN_STRONG_INLINE void writePacketByOuterInner
-    (Index outer, Index inner, const typename ei_packet_traits<Scalar>::type& x)
-    {
-      writePacket<StoreMode>(rowIndexByOuterInner(outer, inner),
-                            colIndexByOuterInner(outer, inner),
-                            x);
-    }
-
-    /** 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<int StoreMode>
-    EIGEN_STRONG_INLINE void writePacket
-    (Index index, const typename ei_packet_traits<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 OtherDerived>
-    EIGEN_STRONG_INLINE void copyCoeff(Index row, Index col, const DenseBase<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 OtherDerived>
-    EIGEN_STRONG_INLINE void copyCoeff(Index index, const DenseBase<OtherDerived>& other)
-    {
-      ei_internal_assert(index >= 0 && index < size());
-      derived().coeffRef(index) = other.derived().coeff(index);
-    }
-
-
-    template<typename OtherDerived>
-    EIGEN_STRONG_INLINE void copyCoeffByOuterInner(Index outer, Index inner, const DenseBase<OtherDerived>& other)
-    {
-      const Index row = rowIndexByOuterInner(outer,inner);
-      const Index col = colIndexByOuterInner(outer,inner);
-      // derived() is important here: copyCoeff() may be reimplemented in Derived!
-      derived().copyCoeff(row, col, other);
-    }
-
-    /** \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 OtherDerived, int StoreMode, int LoadMode>
-    EIGEN_STRONG_INLINE void copyPacket(Index row, Index col, const DenseBase<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 OtherDerived, int StoreMode, int LoadMode>
-    EIGEN_STRONG_INLINE void copyPacket(Index index, const DenseBase<OtherDerived>& other)
-    {
-      ei_internal_assert(index >= 0 && index < size());
-      derived().template writePacket<StoreMode>(index,
-        other.derived().template packet<LoadMode>(index));
-    }
-
-    template<typename OtherDerived, int StoreMode, int LoadMode>
-    EIGEN_STRONG_INLINE void copyPacketByOuterInner(Index outer, Index inner, const DenseBase<OtherDerived>& other)
-    {
-      const Index row = rowIndexByOuterInner(outer,inner);
-      const Index col = colIndexByOuterInner(outer,inner);
-      // derived() is important here: copyCoeff() may be reimplemented in Derived!
-      derived().template copyPacket< OtherDerived, StoreMode, LoadMode>(row, col, other);
-    }
-#endif
-
 };
 
-/** \brief Base class providing direct coefficient access to matrices and arrays.
+/** \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
+  * \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, WriteAccessors> which defines functions to access entries using 
+  * inherits DenseCoeffsBase<Derived, ReadOnlyAccessors> which defines functions to access entries read-only using
   * \c operator() .
   *
   * \sa \ref TopicClassHierarchy
   */
 template<typename Derived>
-class DenseCoeffsBase<Derived, DirectAccessors> : public DenseCoeffsBase<Derived, WriteAccessors>
+class DenseCoeffsBase<Derived, DirectAccessors> : public DenseCoeffsBase<Derived, ReadOnlyAccessors>
 {
   public:
 
-    typedef DenseCoeffsBase<Derived, WriteAccessors> Base;
-    typedef typename ei_traits<Derived>::Index Index;
-    typedef typename ei_traits<Derived>::Scalar Scalar;
+    typedef DenseCoeffsBase<Derived, ReadOnlyAccessors> Base;
+    typedef typename internal::traits<Derived>::Scalar Scalar;
     typedef typename NumTraits<Scalar>::Real RealScalar;
 
     using Base::rows;
@@ -566,6 +468,7 @@ class DenseCoeffsBase<Derived, DirectAccessors> : public DenseCoeffsBase<Derived
       *
       * \sa outerStride(), rowStride(), colStride()
       */
+    EIGEN_DEVICE_FUNC
     inline Index innerStride() const
     {
       return derived().innerStride();
@@ -576,6 +479,7 @@ class DenseCoeffsBase<Derived, DirectAccessors> : public DenseCoeffsBase<Derived
       *
       * \sa innerStride(), rowStride(), colStride()
       */
+    EIGEN_DEVICE_FUNC
     inline Index outerStride() const
     {
       return derived().outerStride();
@@ -591,6 +495,7 @@ class DenseCoeffsBase<Derived, DirectAccessors> : public DenseCoeffsBase<Derived
       *
       * \sa innerStride(), outerStride(), colStride()
       */
+    EIGEN_DEVICE_FUNC
     inline Index rowStride() const
     {
       return Derived::IsRowMajor ? outerStride() : innerStride();
@@ -600,63 +505,153 @@ class DenseCoeffsBase<Derived, DirectAccessors> : public DenseCoeffsBase<Derived
       *
       * \sa innerStride(), outerStride(), rowStride()
       */
+    EIGEN_DEVICE_FUNC
     inline Index colStride() const
     {
       return Derived::IsRowMajor ? innerStride() : outerStride();
     }
 };
 
-template<typename Derived, bool JustReturnZero>
-struct ei_first_aligned_impl
+/** \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 \ref TopicClassHierarchy
+  */
+template<typename Derived>
+class DenseCoeffsBase<Derived, DirectWriteAccessors>
+  : public DenseCoeffsBase<Derived, WriteAccessors>
 {
-  inline static typename Derived::Index run(const Derived&)
+  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<typename Derived>
-struct ei_first_aligned_impl<Derived, false>
+template<int Alignment, typename Derived>
+struct first_aligned_impl<Alignment, Derived, false>
 {
-  inline static typename Derived::Index run(const Derived& m)
+  static inline Index run(const Derived& m)
   {
-    return ei_first_aligned(&m.const_cast_derived().coeffRef(0,0), m.size());
+    return internal::first_aligned<Alignment>(&m.const_cast_derived().coeffRef(0,0), m.size());
   }
 };
 
-/** \internal \returns the index of the first element of the array that is well aligned for vectorization.
+/** \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.
   *
-  * There is also the variant ei_first_aligned(const Scalar*, Integer) defined in Memory.h. See it for more
+  * \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<typename Derived>
-inline static typename Derived::Index ei_first_aligned(const Derived& m)
+template<int Alignment, typename Derived>
+static inline Index first_aligned(const DenseBase<Derived>& m)
 {
-  return ei_first_aligned_impl
-          <Derived, (Derived::Flags & AlignedBit) || !(Derived::Flags & DirectAccessBit)>
-          ::run(m);
+  enum { ReturnZero = (int(evaluator<Derived>::Alignment) >= Alignment) || !(Derived::Flags & DirectAccessBit) };
+  return first_aligned_impl<Alignment, Derived, ReturnZero>::run(m.derived());
 }
 
-template<typename Derived, bool HasDirectAccess = ei_has_direct_access<Derived>::ret>
-struct ei_inner_stride_at_compile_time
+template<typename Derived>
+static inline Index first_default_aligned(const DenseBase<Derived>& m)
 {
-  enum { ret = ei_traits<Derived>::InnerStrideAtCompileTime };
+  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 ei_inner_stride_at_compile_time<Derived, false>
+struct inner_stride_at_compile_time<Derived, false>
 {
   enum { ret = 0 };
 };
 
-template<typename Derived, bool HasDirectAccess = ei_has_direct_access<Derived>::ret>
-struct ei_outer_stride_at_compile_time
+template<typename Derived, bool HasDirectAccess = has_direct_access<Derived>::ret>
+struct outer_stride_at_compile_time
 {
-  enum { ret = ei_traits<Derived>::OuterStrideAtCompileTime };
+  enum { ret = traits<Derived>::OuterStrideAtCompileTime };
 };
 
 template<typename Derived>
-struct ei_outer_stride_at_compile_time<Derived, false>
+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 +1,563 @@
+// 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 EIGEN_DENSE_STORAGE_CTOR_PLUGIN;
+#else
+  #define EIGEN_INTERNAL_DENSE_STORAGE_CTOR_PLUGIN
+#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((reinterpret_cast<size_t>(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((reinterpret_cast<size_t>(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_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_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)
+    {
+      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;
+    }
+#ifdef EIGEN_HAVE_RVALUE_REFERENCES
+    EIGEN_DEVICE_FUNC
+    DenseStorage(DenseStorage&& other)
+      : 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)
+    {
+      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)
+          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)
+    {
+      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;
+    }    
+#ifdef EIGEN_HAVE_RVALUE_REFERENCES
+    EIGEN_DEVICE_FUNC
+    DenseStorage(DenseStorage&& other)
+      : 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)
+    {
+      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)
+          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)
+    {
+      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;
+    }    
+#ifdef EIGEN_HAVE_RVALUE_REFERENCES
+    EIGEN_DEVICE_FUNC
+    DenseStorage(DenseStorage&& other)
+      : 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)
+    {
+      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)
+          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/DenseStorageBase.h

@@ -1,635 +0,0 @@
-// This file is part of Eigen, a lightweight C++ template library
-// for linear algebra.
-//
-// Copyright (C) 2008-2009 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/>.
-
-#ifndef EIGEN_DENSESTORAGEBASE_H
-#define EIGEN_DENSESTORAGEBASE_H
-
-#ifdef EIGEN_INITIALIZE_MATRICES_BY_ZERO
-# define EIGEN_INITIALIZE_BY_ZERO_IF_THAT_OPTION_IS_ENABLED for(int i=0;i<base().size();++i) coeffRef(i)=Scalar(0);
-#else
-# define EIGEN_INITIALIZE_BY_ZERO_IF_THAT_OPTION_IS_ENABLED
-#endif
-
-template <typename Derived, typename OtherDerived = Derived, bool IsVector = static_cast<bool>(Derived::IsVectorAtCompileTime)> struct ei_conservative_resize_like_impl;
-template<typename MatrixTypeA, typename MatrixTypeB, bool SwapPointers> struct ei_matrix_swap_impl;
-
-/**
-  * \brief %Dense storage base class for matrices and arrays.
-  * \sa \ref TopicClassHierarchy
-  */
-template<typename Derived>
-class DenseStorageBase : public ei_dense_xpr_base<Derived>::type
-{
-  public:
-    enum { Options = ei_traits<Derived>::Options };
-    typedef typename ei_dense_xpr_base<Derived>::type Base;
-
-    typedef typename ei_traits<Derived>::StorageKind StorageKind;
-    typedef typename ei_traits<Derived>::Index Index;
-    typedef typename ei_traits<Derived>::Scalar Scalar;
-    typedef typename ei_packet_traits<Scalar>::type PacketScalar;
-    typedef typename NumTraits<Scalar>::Real RealScalar;
-
-    using Base::RowsAtCompileTime;
-    using Base::ColsAtCompileTime;
-    using Base::SizeAtCompileTime;
-    using Base::MaxRowsAtCompileTime;
-    using Base::MaxColsAtCompileTime;
-    using Base::MaxSizeAtCompileTime;
-    using Base::IsVectorAtCompileTime;
-    using Base::Flags;
-
-    friend  class Eigen::Map<Derived, Unaligned>;
-    typedef class Eigen::Map<Derived, Unaligned>  UnalignedMapType;
-    friend  class Eigen::Map<Derived, Aligned>;
-    typedef class Eigen::Map<Derived, Aligned>    AlignedMapType;
-
-  protected:
-    ei_matrix_storage<Scalar, Base::MaxSizeAtCompileTime, Base::RowsAtCompileTime, Base::ColsAtCompileTime, Options> m_storage;
-
-  public:
-    enum { NeedsToAlign = (!(Options&DontAlign))
-                          && SizeAtCompileTime!=Dynamic && ((static_cast<int>(sizeof(Scalar))*SizeAtCompileTime)%16)==0 };
-    EIGEN_MAKE_ALIGNED_OPERATOR_NEW_IF(NeedsToAlign)
-
-    Base& base() { return *static_cast<Base*>(this); }
-    const Base& base() const { return *static_cast<const Base*>(this); }
-
-    EIGEN_STRONG_INLINE Index rows() const { return m_storage.rows(); }
-    EIGEN_STRONG_INLINE Index cols() const { return m_storage.cols(); }
-
-    EIGEN_STRONG_INLINE const Scalar& coeff(Index row, Index col) const
-    {
-      if(Flags & RowMajorBit)
-        return m_storage.data()[col + row * m_storage.cols()];
-      else // column-major
-        return m_storage.data()[row + col * m_storage.rows()];
-    }
-
-    EIGEN_STRONG_INLINE const Scalar& coeff(Index index) const
-    {
-      return m_storage.data()[index];
-    }
-
-    EIGEN_STRONG_INLINE Scalar& coeffRef(Index row, Index col)
-    {
-      if(Flags & RowMajorBit)
-        return m_storage.data()[col + row * m_storage.cols()];
-      else // column-major
-        return m_storage.data()[row + col * m_storage.rows()];
-    }
-
-    EIGEN_STRONG_INLINE Scalar& coeffRef(Index index)
-    {
-      return m_storage.data()[index];
-    }
-
-    template<int LoadMode>
-    EIGEN_STRONG_INLINE PacketScalar packet(Index row, Index col) const
-    {
-      return ei_ploadt<PacketScalar, LoadMode>
-               (m_storage.data() + (Flags & RowMajorBit
-                                   ? col + row * m_storage.cols()
-                                   : row + col * m_storage.rows()));
-    }
-
-    template<int LoadMode>
-    EIGEN_STRONG_INLINE PacketScalar packet(Index index) const
-    {
-      return ei_ploadt<PacketScalar, LoadMode>(m_storage.data() + index);
-    }
-
-    template<int StoreMode>
-    EIGEN_STRONG_INLINE void writePacket(Index row, Index col, const PacketScalar& x)
-    {
-      ei_pstoret<Scalar, PacketScalar, StoreMode>
-              (m_storage.data() + (Flags & RowMajorBit
-                                   ? col + row * m_storage.cols()
-                                   : row + col * m_storage.rows()), x);
-    }
-
-    template<int StoreMode>
-    EIGEN_STRONG_INLINE void writePacket(Index index, const PacketScalar& x)
-    {
-      ei_pstoret<Scalar, PacketScalar, StoreMode>(m_storage.data() + index, x);
-    }
-
-    /** \returns a const pointer to the data array of this matrix */
-    EIGEN_STRONG_INLINE const Scalar *data() const
-    { return m_storage.data(); }
-
-    /** \returns a pointer to the data array of this matrix */
-    EIGEN_STRONG_INLINE Scalar *data()
-    { return m_storage.data(); }
-
-    /** Resizes \c *this to a \a rows x \a cols matrix.
-      *
-      * This method is intended for dynamic-size matrices, although it is legal to call it on any
-      * matrix as long as fixed dimensions are left unchanged. If you only want to change the number
-      * of rows and/or of columns, you can use resize(NoChange_t, Index), resize(Index, NoChange_t).
-      *
-      * If the current number of coefficients of \c *this exactly matches the
-      * product \a rows * \a cols, then no memory allocation is performed and
-      * the current values are left unchanged. In all other cases, including
-      * shrinking, the data is reallocated and all previous values are lost.
-      *
-      * Example: \include Matrix_resize_int_int.cpp
-      * Output: \verbinclude Matrix_resize_int_int.out
-      *
-      * \sa resize(Index) for vectors, resize(NoChange_t, Index), resize(Index, NoChange_t)
-      */
-    EIGEN_STRONG_INLINE void resize(Index rows, Index cols)
-    {
-      #ifdef EIGEN_INITIALIZE_MATRICES_BY_ZERO
-        Index size = rows*cols;
-        bool size_changed = size != this->size();
-        m_storage.resize(size, rows, cols);
-        if(size_changed) EIGEN_INITIALIZE_BY_ZERO_IF_THAT_OPTION_IS_ENABLED
-      #else
-        m_storage.resize(rows*cols, rows, cols);
-      #endif
-    }
-
-    /** Resizes \c *this to a vector of length \a size
-      *
-      * \only_for_vectors. This method does not work for
-      * partially dynamic matrices when the static dimension is anything other
-      * than 1. For example it will not work with Matrix<double, 2, Dynamic>.
-      *
-      * Example: \include Matrix_resize_int.cpp
-      * Output: \verbinclude Matrix_resize_int.out
-      *
-      * \sa resize(Index,Index), resize(NoChange_t, Index), resize(Index, NoChange_t)
-      */
-    inline void resize(Index size)
-    {
-      EIGEN_STATIC_ASSERT_VECTOR_ONLY(DenseStorageBase)
-      ei_assert(SizeAtCompileTime == Dynamic || SizeAtCompileTime == size);
-      #ifdef EIGEN_INITIALIZE_MATRICES_BY_ZERO
-        bool size_changed = size != this->size();
-      #endif
-      if(RowsAtCompileTime == 1)
-        m_storage.resize(size, 1, size);
-      else
-        m_storage.resize(size, size, 1);
-      #ifdef EIGEN_INITIALIZE_MATRICES_BY_ZERO
-        if(size_changed) EIGEN_INITIALIZE_BY_ZERO_IF_THAT_OPTION_IS_ENABLED
-      #endif
-    }
-
-    /** Resizes the matrix, changing only the number of columns. For the parameter of type NoChange_t, just pass the special value \c NoChange
-      * as in the example below.
-      *
-      * Example: \include Matrix_resize_NoChange_int.cpp
-      * Output: \verbinclude Matrix_resize_NoChange_int.out
-      *
-      * \sa resize(Index,Index)
-      */
-    inline void resize(NoChange_t, Index cols)
-    {
-      resize(rows(), cols);
-    }
-
-    /** Resizes the matrix, changing only the number of rows. For the parameter of type NoChange_t, just pass the special value \c NoChange
-      * as in the example below.
-      *
-      * Example: \include Matrix_resize_int_NoChange.cpp
-      * Output: \verbinclude Matrix_resize_int_NoChange.out
-      *
-      * \sa resize(Index,Index)
-      */
-    inline void resize(Index rows, NoChange_t)
-    {
-      resize(rows, cols());
-    }
-
-    /** Resizes \c *this to have the same dimensions as \a other.
-      * Takes care of doing all the checking that's needed.
-      *
-      * 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_STRONG_INLINE void resizeLike(const EigenBase<OtherDerived>& _other)
-    {
-      const OtherDerived& other = _other.derived();
-      const Index othersize = other.rows()*other.cols();
-      if(RowsAtCompileTime == 1)
-      {
-        ei_assert(other.rows() == 1 || other.cols() == 1);
-        resize(1, othersize);
-      }
-      else if(ColsAtCompileTime == 1)
-      {
-        ei_assert(other.rows() == 1 || other.cols() == 1);
-        resize(othersize, 1);
-      }
-      else resize(other.rows(), other.cols());
-    }
-
-    /** Resizes \c *this to a \a rows x \a cols matrix while leaving old values of \c *this untouched.
-      *
-      * This method is intended for dynamic-size matrices. If you only want to change the number
-      * of rows and/or of columns, you can use conservativeResize(NoChange_t, Index),
-      * conservativeResize(Index, NoChange_t).
-      *
-      * The top-left part of the resized matrix will be the same as the overlapping top-left corner
-      * of \c *this. In case values need to be appended to the matrix they will be uninitialized.
-      */
-    EIGEN_STRONG_INLINE void conservativeResize(Index rows, Index cols)
-    {
-      ei_conservative_resize_like_impl<Derived>::run(*this, rows, cols);
-    }
-
-    EIGEN_STRONG_INLINE void conservativeResize(Index rows, NoChange_t)
-    {
-      // Note: see the comment in conservativeResize(Index,Index)
-      conservativeResize(rows, cols());
-    }
-
-    EIGEN_STRONG_INLINE void conservativeResize(NoChange_t, Index cols)
-    {
-      // Note: see the comment in conservativeResize(Index,Index)
-      conservativeResize(rows(), cols);
-    }
-
-    /** Resizes \c *this to a vector of length \a size while retaining old values of *this.
-      *
-      * \only_for_vectors. This method does not work for
-      * partially dynamic matrices when the static dimension is anything other
-      * than 1. For example it will not work with Matrix<double, 2, Dynamic>.
-      *
-      * When values are appended, they will be uninitialized.
-      */
-    EIGEN_STRONG_INLINE void conservativeResize(Index size)
-    {
-      ei_conservative_resize_like_impl<Derived>::run(*this, size);
-    }
-
-    template<typename OtherDerived>
-    EIGEN_STRONG_INLINE void conservativeResizeLike(const DenseBase<OtherDerived>& other)
-    {
-      ei_conservative_resize_like_impl<Derived,OtherDerived>::run(*this, other);
-    }
-
-    /** This is a special case of the templated operator=. Its purpose is to
-      * prevent a default operator= from hiding the templated operator=.
-      */
-    EIGEN_STRONG_INLINE Derived& operator=(const DenseStorageBase& other)
-    {
-      return _set(other);
-    }
-
-    /** \sa MatrixBase::lazyAssign() */
-    template<typename OtherDerived>
-    EIGEN_STRONG_INLINE Derived& lazyAssign(const DenseBase<OtherDerived>& other)
-    {
-      _resize_to_match(other);
-      return Base::lazyAssign(other.derived());
-    }
-
-    template<typename OtherDerived>
-    EIGEN_STRONG_INLINE Derived& operator=(const ReturnByValue<OtherDerived>& func)
-    {
-      resize(func.rows(), func.cols());
-      return Base::operator=(func);
-    }
-
-    EIGEN_STRONG_INLINE explicit DenseStorageBase() : m_storage()
-    {
-//       _check_template_params();
-//       EIGEN_INITIALIZE_BY_ZERO_IF_THAT_OPTION_IS_ENABLED
-    }
-
-#ifndef EIGEN_PARSED_BY_DOXYGEN
-    // FIXME is it still needed ?
-    /** \internal */
-    DenseStorageBase(ei_constructor_without_unaligned_array_assert)
-      : m_storage(ei_constructor_without_unaligned_array_assert())
-    {
-//       _check_template_params(); EIGEN_INITIALIZE_BY_ZERO_IF_THAT_OPTION_IS_ENABLED
-    }
-#endif
-
-    EIGEN_STRONG_INLINE DenseStorageBase(Index size, Index rows, Index cols)
-      : m_storage(size, rows, cols)
-    {
-//       _check_template_params();
-//       EIGEN_INITIALIZE_BY_ZERO_IF_THAT_OPTION_IS_ENABLED
-    }
-
-    /** \copydoc MatrixBase::operator=(const EigenBase<OtherDerived>&)
-      */
-    template<typename OtherDerived>
-    EIGEN_STRONG_INLINE Derived& operator=(const EigenBase<OtherDerived> &other)
-    {
-      _resize_to_match(other);
-      Base::operator=(other.derived());
-      return this->derived();
-    }
-
-    /** \sa MatrixBase::operator=(const EigenBase<OtherDerived>&) */
-    template<typename OtherDerived>
-    EIGEN_STRONG_INLINE DenseStorageBase(const EigenBase<OtherDerived> &other)
-      : m_storage(other.derived().rows() * other.derived().cols(), other.derived().rows(), other.derived().cols())
-    {
-      _check_template_params();
-      Base::operator=(other.derived());
-    }
-
-    /** \name Map
-      * These are convenience functions returning Map objects. The Map() static functions return unaligned Map objects,
-      * while the AlignedMap() functions return aligned Map objects and thus should be called only with 16-byte-aligned
-      * \a data pointers.
-      *
-      * These methods do not allow to specify strides. If you need to specify strides, you have to
-      * use the Map class directly.
-      *
-      * \see class Map
-      */
-    //@{
-    inline static const UnalignedMapType Map(const Scalar* data)
-    { return UnalignedMapType(data); }
-    inline static UnalignedMapType Map(Scalar* data)
-    { return UnalignedMapType(data); }
-    inline static const UnalignedMapType Map(const Scalar* data, Index size)
-    { return UnalignedMapType(data, size); }
-    inline static UnalignedMapType Map(Scalar* data, Index size)
-    { return UnalignedMapType(data, size); }
-    inline static const UnalignedMapType Map(const Scalar* data, Index rows, Index cols)
-    { return UnalignedMapType(data, rows, cols); }
-    inline static UnalignedMapType Map(Scalar* data, Index rows, Index cols)
-    { return UnalignedMapType(data, rows, cols); }
-
-    inline static const AlignedMapType MapAligned(const Scalar* data)
-    { return AlignedMapType(data); }
-    inline static AlignedMapType MapAligned(Scalar* data)
-    { return AlignedMapType(data); }
-    inline static const AlignedMapType MapAligned(const Scalar* data, Index size)
-    { return AlignedMapType(data, size); }
-    inline static AlignedMapType MapAligned(Scalar* data, Index size)
-    { return AlignedMapType(data, size); }
-    inline static const AlignedMapType MapAligned(const Scalar* data, Index rows, Index cols)
-    { return AlignedMapType(data, rows, cols); }
-    inline static AlignedMapType MapAligned(Scalar* data, Index rows, Index cols)
-    { return AlignedMapType(data, rows, cols); }
-    //@}
-
-    using Base::setConstant;
-    Derived& setConstant(Index size, const Scalar& value);
-    Derived& setConstant(Index rows, Index cols, const Scalar& value);
-
-    using Base::setZero;
-    Derived& setZero(Index size);
-    Derived& setZero(Index rows, Index cols);
-
-    using Base::setOnes;
-    Derived& setOnes(Index size);
-    Derived& setOnes(Index rows, Index cols);
-
-    using Base::setRandom;
-    Derived& setRandom(Index size);
-    Derived& setRandom(Index rows, Index cols);
-
-    #ifdef EIGEN_DENSESTORAGEBASE_PLUGIN
-    #include EIGEN_DENSESTORAGEBASE_PLUGIN
-    #endif
-
-  protected:
-    /** \internal Resizes *this in preparation for assigning \a other to it.
-      * Takes care of doing all the checking that's needed.
-      *
-      * 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_STRONG_INLINE void _resize_to_match(const EigenBase<OtherDerived>& other)
-    {
-      #ifdef EIGEN_NO_AUTOMATIC_RESIZING
-      ei_assert((this->size()==0 || (IsVectorAtCompileTime ? (this->size() == other.size())
-                 : (rows() == other.rows() && cols() == other.cols())))
-        && "Size mismatch. Automatic resizing is disabled because EIGEN_NO_AUTOMATIC_RESIZING is defined");
-      #else
-      resizeLike(other);
-      #endif
-    }
-
-    /**
-      * \brief 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.
-      *
-      * \sa operator=(const MatrixBase<OtherDerived>&), _set_noalias()
-      *
-      * \internal
-      */
-    template<typename OtherDerived>
-    EIGEN_STRONG_INLINE Derived& _set(const DenseBase<OtherDerived>& other)
-    {
-      _set_selector(other.derived(), typename ei_meta_if<static_cast<bool>(int(OtherDerived::Flags) & EvalBeforeAssigningBit), ei_meta_true, ei_meta_false>::ret());
-      return this->derived();
-    }
-
-    template<typename OtherDerived>
-    EIGEN_STRONG_INLINE void _set_selector(const OtherDerived& other, const ei_meta_true&) { _set_noalias(other.eval()); }
-
-    template<typename OtherDerived>
-    EIGEN_STRONG_INLINE void _set_selector(const OtherDerived& other, const ei_meta_false&) { _set_noalias(other); }
-
-    /** \internal Like _set() but additionally makes the assumption that no aliasing effect can happen (which
-      * is the case when creating a new matrix) so one can enforce lazy evaluation.
-      *
-      * \sa operator=(const MatrixBase<OtherDerived>&), _set()
-      */
-    template<typename OtherDerived>
-    EIGEN_STRONG_INLINE Derived& _set_noalias(const DenseBase<OtherDerived>& other)
-    {
-      // I don't think we need this resize call since the lazyAssign will anyways resize
-      // and lazyAssign will be called by the assign selector.
-      //_resize_to_match(other);
-      // the 'false' below means to enforce lazy evaluation. We don't use lazyAssign() because
-      // it wouldn't allow to copy a row-vector into a column-vector.
-      return ei_assign_selector<Derived,OtherDerived,false>::run(this->derived(), other.derived());
-    }
-
-    template<typename T0, typename T1>
-    EIGEN_STRONG_INLINE void _init2(Index rows, Index cols, typename ei_enable_if<Base::SizeAtCompileTime!=2,T0>::type* = 0)
-    {
-      ei_assert(rows >= 0 && (RowsAtCompileTime == Dynamic || RowsAtCompileTime == rows)
-             && cols >= 0 && (ColsAtCompileTime == Dynamic || ColsAtCompileTime == cols));
-      m_storage.resize(rows*cols,rows,cols);
-      EIGEN_INITIALIZE_BY_ZERO_IF_THAT_OPTION_IS_ENABLED
-    }
-    template<typename T0, typename T1>
-    EIGEN_STRONG_INLINE void _init2(const Scalar& x, const Scalar& y, typename ei_enable_if<Base::SizeAtCompileTime==2,T0>::type* = 0)
-    {
-      EIGEN_STATIC_ASSERT_VECTOR_SPECIFIC_SIZE(DenseStorageBase, 2)
-      m_storage.data()[0] = x;
-      m_storage.data()[1] = y;
-    }
-
-    template<typename MatrixTypeA, typename MatrixTypeB, bool SwapPointers>
-    friend struct ei_matrix_swap_impl;
-
-    /** \internal generic implementation of swap for dense storage since for dynamic-sized matrices of same type it is enough to swap the
-      * data pointers.
-      */
-    template<typename OtherDerived>
-    void _swap(DenseBase<OtherDerived> EIGEN_REF_TO_TEMPORARY other)
-    {
-      enum { SwapPointers = ei_is_same_type<Derived, OtherDerived>::ret && Base::SizeAtCompileTime==Dynamic };
-      ei_matrix_swap_impl<Derived, OtherDerived, bool(SwapPointers)>::run(this->derived(), other.const_cast_derived());
-    }
-
-  public:
-#ifndef EIGEN_PARSED_BY_DOXYGEN
-    EIGEN_STRONG_INLINE static void _check_template_params()
-    {
-      EIGEN_STATIC_ASSERT((EIGEN_IMPLIES(MaxRowsAtCompileTime==1 && MaxColsAtCompileTime!=1, (Options&RowMajor)==RowMajor)
-                        && EIGEN_IMPLIES(MaxColsAtCompileTime==1 && MaxRowsAtCompileTime!=1, (Options&RowMajor)==0)
-                        && ((RowsAtCompileTime == Dynamic) || (RowsAtCompileTime >= 0))
-                        && ((ColsAtCompileTime == Dynamic) || (ColsAtCompileTime >= 0))
-                        && ((MaxRowsAtCompileTime == Dynamic) || (MaxRowsAtCompileTime >= 0))
-                        && ((MaxColsAtCompileTime == Dynamic) || (MaxColsAtCompileTime >= 0))
-                        && (MaxRowsAtCompileTime == RowsAtCompileTime || RowsAtCompileTime==Dynamic)
-                        && (MaxColsAtCompileTime == ColsAtCompileTime || ColsAtCompileTime==Dynamic)
-                        && (Options & (DontAlign|RowMajor)) == Options),
-        INVALID_MATRIX_TEMPLATE_PARAMETERS)
-    }
-#endif
-};
-
-template <typename Derived, typename OtherDerived, bool IsVector>
-struct ei_conservative_resize_like_impl
-{
-  typedef typename Derived::Index Index;
-  static void run(DenseBase<Derived>& _this, Index rows, Index cols)
-  {
-    if (_this.rows() == rows && _this.cols() == cols) return;
-    EIGEN_STATIC_ASSERT_DYNAMIC_SIZE(Derived)
-
-    if ( ( Derived::IsRowMajor && _this.cols() == cols) || // row-major and we change only the number of rows
-         (!Derived::IsRowMajor && _this.rows() == rows) )  // column-major and we change only the number of columns
-    {
-      _this.derived().m_storage.conservativeResize(rows*cols,rows,cols);
-    }
-    else
-    {
-      // The storage order does not allow us to use reallocation.
-      typename Derived::PlainObject tmp(rows,cols);
-      const Index common_rows = std::min(rows, _this.rows());
-      const Index common_cols = std::min(cols, _this.cols());
-      tmp.block(0,0,common_rows,common_cols) = _this.block(0,0,common_rows,common_cols);
-      _this.derived().swap(tmp);
-    }
-  }
-
-  static void run(DenseBase<Derived>& _this, const DenseBase<OtherDerived>& other)
-  {
-    if (_this.rows() == other.rows() && _this.cols() == other.cols()) return;
-
-    // Note: Here is space for improvement. Basically, for conservativeResize(Index,Index),
-    // neither RowsAtCompileTime or ColsAtCompileTime must be Dynamic. If only one of the
-    // dimensions is dynamic, one could use either conservativeResize(Index rows, NoChange_t) or
-    // conservativeResize(NoChange_t, Index cols). For these methods new static asserts like
-    // EIGEN_STATIC_ASSERT_DYNAMIC_ROWS and EIGEN_STATIC_ASSERT_DYNAMIC_COLS would be good.
-    EIGEN_STATIC_ASSERT_DYNAMIC_SIZE(Derived)
-    EIGEN_STATIC_ASSERT_DYNAMIC_SIZE(OtherDerived)
-
-    if ( ( Derived::IsRowMajor && _this.cols() == other.cols()) || // row-major and we change only the number of rows
-         (!Derived::IsRowMajor && _this.rows() == other.rows()) )  // column-major and we change only the number of columns
-    {
-      const Index new_rows = other.rows() - _this.rows();
-      const Index new_cols = other.cols() - _this.cols();
-      _this.derived().m_storage.conservativeResize(other.size(),other.rows(),other.cols());
-      if (new_rows>0)
-        _this.bottomRightCorner(new_rows, other.cols()) = other.bottomRows(new_rows);
-      else if (new_cols>0)
-        _this.bottomRightCorner(other.rows(), new_cols) = other.rightCols(new_cols);
-    }
-    else
-    {
-      // The storage order does not allow us to use reallocation.
-      typename Derived::PlainObject tmp(other);
-      const Index common_rows = std::min(tmp.rows(), _this.rows());
-      const Index common_cols = std::min(tmp.cols(), _this.cols());
-      tmp.block(0,0,common_rows,common_cols) = _this.block(0,0,common_rows,common_cols);
-      _this.derived().swap(tmp);
-    }
-  }
-};
-
-template <typename Derived, typename OtherDerived>
-struct ei_conservative_resize_like_impl<Derived,OtherDerived,true>
-{
-  typedef typename Derived::Index Index;
-  static void run(DenseBase<Derived>& _this, Index size)
-  {
-    const Index new_rows = Derived::RowsAtCompileTime==1 ? 1 : size;
-    const Index new_cols = Derived::RowsAtCompileTime==1 ? size : 1;
-    _this.derived().m_storage.conservativeResize(size,new_rows,new_cols);
-  }
-
-  static void run(DenseBase<Derived>& _this, const DenseBase<OtherDerived>& other)
-  {
-    if (_this.rows() == other.rows() && _this.cols() == other.cols()) return;
-
-    const Index num_new_elements = other.size() - _this.size();
-
-    const Index new_rows = Derived::RowsAtCompileTime==1 ? 1 : other.rows();
-    const Index new_cols = Derived::RowsAtCompileTime==1 ? other.cols() : 1;
-    _this.derived().m_storage.conservativeResize(other.size(),new_rows,new_cols);
-
-    if (num_new_elements > 0)
-      _this.tail(num_new_elements) = other.tail(num_new_elements);
-  }
-};
-
-template<typename MatrixTypeA, typename MatrixTypeB, bool SwapPointers>
-struct ei_matrix_swap_impl
-{
-  static inline void run(MatrixTypeA& a, MatrixTypeB& b)
-  {
-    a.base().swap(b);
-  }
-};
-
-template<typename MatrixTypeA, typename MatrixTypeB>
-struct ei_matrix_swap_impl<MatrixTypeA, MatrixTypeB, true>
-{
-  static inline void run(MatrixTypeA& a, MatrixTypeB& b)
-  {
-    static_cast<typename MatrixTypeA::Base&>(a).m_storage.swap(static_cast<typename MatrixTypeB::Base&>(b).m_storage);
-  }
-};
-
-#endif // EIGEN_DENSESTORAGEBASE_H

Eigen/src/Core/Diagonal.h

@@ -2,29 +2,17 @@
 // for linear algebra.
 //
 // Copyright (C) 2007-2009 Benoit Jacob <jacob.benoit.1@gmail.com>
+// Copyright (C) 2009-2010 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_DIAGONAL_H
 #define EIGEN_DIAGONAL_H
 
+namespace Eigen { 
+
 /** \class Diagonal
   * \ingroup Core_Module
   *
@@ -43,90 +31,146 @@
   *
   * \sa MatrixBase::diagonal(), MatrixBase::diagonal(Index)
   */
+
+namespace internal {
 template<typename MatrixType, int DiagIndex>
-struct ei_traits<Diagonal<MatrixType,DiagIndex> >
- : ei_traits<MatrixType>
+struct traits<Diagonal<MatrixType,DiagIndex> >
+ : traits<MatrixType>
 {
-  typedef typename ei_nested<MatrixType>::type MatrixTypeNested;
-  typedef typename ei_unref<MatrixTypeNested>::type _MatrixTypeNested;
+  typedef typename ref_selector<MatrixType>::type MatrixTypeNested;
+  typedef typename remove_reference<MatrixTypeNested>::type _MatrixTypeNested;
   typedef typename MatrixType::StorageKind StorageKind;
   enum {
-    AbsDiagIndex = DiagIndex<0 ? -DiagIndex : DiagIndex, // only used if DiagIndex != Dynamic
-    // FIXME these computations are broken in the case where the matrix is rectangular and DiagIndex!=0
-    RowsAtCompileTime = (int(DiagIndex) == Dynamic || int(MatrixType::SizeAtCompileTime) == Dynamic) ? Dynamic
-                      : (EIGEN_SIZE_MIN_PREFER_DYNAMIC(MatrixType::RowsAtCompileTime,
-                                        MatrixType::ColsAtCompileTime) - AbsDiagIndex),
+    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 == Dynamic ? EIGEN_SIZE_MIN_PREFER_FIXED(MatrixType::MaxRowsAtCompileTime,
-                                                                    MatrixType::MaxColsAtCompileTime)
-                         : (EIGEN_SIZE_MIN_PREFER_FIXED(MatrixType::MaxRowsAtCompileTime, MatrixType::MaxColsAtCompileTime) - AbsDiagIndex),
+                         : 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,
-    Flags = (unsigned int)_MatrixTypeNested::Flags & (HereditaryBits | LinearAccessBit | LvalueBit | DirectAccessBit) & ~RowMajorBit,
-    CoeffReadCost = _MatrixTypeNested::CoeffReadCost,
-    MatrixTypeOuterStride = ei_outer_stride_at_compile_time<MatrixType>::ret,
+    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 ei_dense_xpr_base< Diagonal<MatrixType,DiagIndex> >::type
+template<typename MatrixType, int _DiagIndex> class Diagonal
+   : public internal::dense_xpr_base< Diagonal<MatrixType,_DiagIndex> >::type
 {
   public:
 
-    typedef typename ei_dense_xpr_base<Diagonal>::type Base;
+    enum { DiagIndex = _DiagIndex };
+    typedef typename internal::dense_xpr_base<Diagonal>::type Base;
     EIGEN_DENSE_PUBLIC_INTERFACE(Diagonal)
 
-    inline Diagonal(const MatrixType& matrix, Index index = DiagIndex) : m_matrix(matrix), m_index(index) {}
+    EIGEN_DEVICE_FUNC
+    explicit inline Diagonal(MatrixType& matrix, Index a_index = DiagIndex) : m_matrix(matrix), m_index(a_index) {}
 
     EIGEN_INHERIT_ASSIGNMENT_OPERATORS(Diagonal)
 
+    EIGEN_DEVICE_FUNC
     inline Index rows() const
-    { return m_index.value()<0 ? std::min(m_matrix.cols(),m_matrix.rows()+m_index.value()) : std::min(m_matrix.rows(),m_matrix.cols()-m_index.value()); }
+    {
+      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.const_cast_derived().coeffRef(rowOffset(), colOffset())); }
+    EIGEN_DEVICE_FUNC
+    inline const Scalar* data() const { return &(m_matrix.const_cast_derived().coeffRef(rowOffset(), colOffset())); }
+
+    EIGEN_DEVICE_FUNC
     inline Scalar& coeffRef(Index row, Index)
+    {
+      EIGEN_STATIC_ASSERT_LVALUE(MatrixType)
+      return m_matrix.const_cast_derived().coeffRef(row+rowOffset(), row+colOffset());
+    }
+
+    EIGEN_DEVICE_FUNC
+    inline const Scalar& coeffRef(Index row, Index) const
     {
       return m_matrix.const_cast_derived().coeffRef(row+rowOffset(), row+colOffset());
     }
 
+    EIGEN_DEVICE_FUNC
     inline CoeffReturnType coeff(Index row, Index) const
     {
       return m_matrix.coeff(row+rowOffset(), row+colOffset());
     }
 
-    inline Scalar& coeffRef(Index index)
+    EIGEN_DEVICE_FUNC
+    inline Scalar& coeffRef(Index idx)
     {
-      return m_matrix.const_cast_derived().coeffRef(index+rowOffset(), index+colOffset());
+      EIGEN_STATIC_ASSERT_LVALUE(MatrixType)
+      return m_matrix.const_cast_derived().coeffRef(idx+rowOffset(), idx+colOffset());
     }
 
-    inline CoeffReturnType coeff(Index index) const
+    EIGEN_DEVICE_FUNC
+    inline const Scalar& coeffRef(Index idx) const
     {
-      return m_matrix.coeff(index+rowOffset(), index+colOffset());
+      return m_matrix.const_cast_derived().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:
-    const typename MatrixType::Nested m_matrix;
-    const ei_variable_if_dynamic<Index, DiagIndex> m_index;
+    typename MatrixType::Nested 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; }
-    // triger a compile time error is someone try to call packet
+    // 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;
 };
@@ -140,18 +184,18 @@ template<typename MatrixType, int DiagIndex> class Diagonal
   *
   * \sa class Diagonal */
 template<typename Derived>
-inline Diagonal<Derived, 0>
+inline typename MatrixBase<Derived>::DiagonalReturnType
 MatrixBase<Derived>::diagonal()
 {
-  return Diagonal<Derived, 0>(derived());
+  return DiagonalReturnType(derived());
 }
 
 /** This is the const version of diagonal(). */
 template<typename Derived>
-inline const Diagonal<Derived, 0>
+inline typename MatrixBase<Derived>::ConstDiagonalReturnType
 MatrixBase<Derived>::diagonal() const
 {
-  return Diagonal<Derived, 0>(derived());
+  return ConstDiagonalReturnType(derived());
 }
 
 /** \returns an expression of the \a DiagIndex-th sub or super diagonal of the matrix \c *this
@@ -166,18 +210,18 @@ MatrixBase<Derived>::diagonal() const
   *
   * \sa MatrixBase::diagonal(), class Diagonal */
 template<typename Derived>
-inline Diagonal<Derived, Dynamic>
+inline typename MatrixBase<Derived>::DiagonalDynamicIndexReturnType
 MatrixBase<Derived>::diagonal(Index index)
 {
-  return Diagonal<Derived, Dynamic>(derived(), index);
+  return DiagonalDynamicIndexReturnType(derived(), index);
 }
 
 /** This is the const version of diagonal(Index). */
 template<typename Derived>
-inline const Diagonal<Derived, Dynamic>
+inline typename MatrixBase<Derived>::ConstDiagonalDynamicIndexReturnType
 MatrixBase<Derived>::diagonal(Index index) const
 {
-  return Diagonal<Derived, Dynamic>(derived(), index);
+  return ConstDiagonalDynamicIndexReturnType(derived(), index);
 }
 
 /** \returns an expression of the \a DiagIndex-th sub or super diagonal of the matrix \c *this
@@ -192,20 +236,22 @@ MatrixBase<Derived>::diagonal(Index index) const
   *
   * \sa MatrixBase::diagonal(), class Diagonal */
 template<typename Derived>
-template<int DiagIndex>
-inline Diagonal<Derived,DiagIndex>
+template<int Index_>
+inline typename MatrixBase<Derived>::template DiagonalIndexReturnType<Index_>::Type
 MatrixBase<Derived>::diagonal()
 {
-  return Diagonal<Derived,DiagIndex>(derived());
+  return typename DiagonalIndexReturnType<Index_>::Type(derived());
 }
 
 /** This is the const version of diagonal<int>(). */
 template<typename Derived>
-template<int DiagIndex>
-inline const Diagonal<Derived,DiagIndex>
+template<int Index_>
+inline typename MatrixBase<Derived>::template ConstDiagonalIndexReturnType<Index_>::Type
 MatrixBase<Derived>::diagonal() const
 {
-  return Diagonal<Derived,DiagIndex>(derived());
+  return typename ConstDiagonalIndexReturnType<Index_>::Type(derived());
 }
 
+} // end namespace Eigen
+
 #endif // EIGEN_DIAGONAL_H

Eigen/src/Core/DiagonalMatrix.h

@@ -4,37 +4,25 @@
 // 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 ei_traits<Derived>::DiagonalVectorType DiagonalVectorType;
+    typedef typename internal::traits<Derived>::DiagonalVectorType DiagonalVectorType;
     typedef typename DiagonalVectorType::Scalar Scalar;
-    typedef typename ei_traits<Derived>::StorageKind StorageKind;
-    typedef typename ei_traits<Derived>::Index Index;
+    typedef typename DiagonalVectorType::RealScalar RealScalar;
+    typedef typename internal::traits<Derived>::StorageKind StorageKind;
+    typedef typename internal::traits<Derived>::StorageIndex StorageIndex;
 
     enum {
       RowsAtCompileTime = DiagonalVectorType::SizeAtCompileTime,
@@ -42,48 +30,62 @@ class DiagonalBase : public EigenBase<Derived>
       MaxRowsAtCompileTime = DiagonalVectorType::MaxSizeAtCompileTime,
       MaxColsAtCompileTime = DiagonalVectorType::MaxSizeAtCompileTime,
       IsVectorAtCompileTime = 0,
-      Flags = 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(); }
-    template<typename DenseDerived>
-    void evalTo(MatrixBase<DenseDerived> &other) const;
-    template<typename DenseDerived>
-    void addTo(MatrixBase<DenseDerived> &other) const
-    { other.diagonal() += diagonal(); }
-    template<typename DenseDerived>
-    void subTo(MatrixBase<DenseDerived> &other) const
-    { other.diagonal() -= diagonal(); }
-
+    
+    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>
-    const DiagonalProduct<MatrixDerived, Derived, OnTheLeft>
-    operator*(const MatrixBase<MatrixDerived> &matrix) const;
+    EIGEN_DEVICE_FUNC
+    const Product<Derived,MatrixDerived,LazyProduct>
+    operator*(const MatrixBase<MatrixDerived> &matrix) const
+    {
+      return Product<Derived, MatrixDerived, LazyProduct>(derived(),matrix.derived());
+    }
 
-    inline const DiagonalWrapper<CwiseUnaryOp<ei_scalar_inverse_op<Scalar>, DiagonalVectorType> >
+    typedef DiagonalWrapper<const CwiseUnaryOp<internal::scalar_inverse_op<Scalar>, const DiagonalVectorType> > InverseReturnType;
+    EIGEN_DEVICE_FUNC
+    inline const InverseReturnType
     inverse() const
     {
-      return diagonal().cwiseInverse();
+      return InverseReturnType(diagonal().cwiseInverse());
+    }
+    
+    typedef DiagonalWrapper<const CwiseUnaryOp<internal::scalar_multiple_op<Scalar>, const DiagonalVectorType> > ScalarMultipleReturnType;
+    EIGEN_DEVICE_FUNC
+    inline const ScalarMultipleReturnType
+    operator*(const Scalar& scalar) const
+    {
+      return ScalarMultipleReturnType(diagonal() * scalar);
+    }
+    EIGEN_DEVICE_FUNC
+    friend inline const ScalarMultipleReturnType
+    operator*(const Scalar& scalar, const DiagonalBase& other)
+    {
+      return ScalarMultipleReturnType(other.diagonal() * scalar);
     }
 };
 
-template<typename Derived>
-template<typename DenseDerived>
-void DiagonalBase<Derived>::evalTo(MatrixBase<DenseDerived> &other) const
-{
-  other.setZero();
-  other.diagonal() = diagonal();
-}
 #endif
 
 /** \class DiagonalMatrix
@@ -98,29 +100,30 @@ void DiagonalBase<Derived>::evalTo(MatrixBase<DenseDerived> &other) const
   *
   * \sa class DiagonalWrapper
   */
+
+namespace internal {
 template<typename _Scalar, int SizeAtCompileTime, int MaxSizeAtCompileTime>
-struct ei_traits<DiagonalMatrix<_Scalar,SizeAtCompileTime,MaxSizeAtCompileTime> >
- : ei_traits<Matrix<_Scalar,SizeAtCompileTime,SizeAtCompileTime,0,MaxSizeAtCompileTime,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 Dense StorageKind;
-  typedef DenseIndex Index;
+  typedef DiagonalShape StorageKind;
   enum {
-    Flags = LvalueBit
+    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 ei_traits<DiagonalMatrix>::DiagonalVectorType DiagonalVectorType;
+    typedef typename internal::traits<DiagonalMatrix>::DiagonalVectorType DiagonalVectorType;
     typedef const DiagonalMatrix& Nested;
     typedef _Scalar Scalar;
-    typedef typename ei_traits<DiagonalMatrix>::StorageKind StorageKind;
-    typedef typename ei_traits<DiagonalMatrix>::Index Index;
+    typedef typename internal::traits<DiagonalMatrix>::StorageKind StorageKind;
+    typedef typename internal::traits<DiagonalMatrix>::StorageIndex StorageIndex;
     #endif
 
   protected:
@@ -130,24 +133,31 @@ class DiagonalMatrix
   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  */
-    inline DiagonalMatrix(Index dim) : m_diagonal(dim) {}
+    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
@@ -157,11 +167,13 @@ class DiagonalMatrix
 
     /** 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();
@@ -172,6 +184,7 @@ class DiagonalMatrix
     /** 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();
@@ -180,14 +193,19 @@ class DiagonalMatrix
     #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); }
 };
 
@@ -204,25 +222,29 @@ class DiagonalMatrix
   *
   * \sa class DiagonalMatrix, class DiagonalBase, MatrixBase::asDiagonal()
   */
+
+namespace internal {
 template<typename _DiagonalVectorType>
-struct ei_traits<DiagonalWrapper<_DiagonalVectorType> >
+struct traits<DiagonalWrapper<_DiagonalVectorType> >
 {
   typedef _DiagonalVectorType DiagonalVectorType;
   typedef typename DiagonalVectorType::Scalar Scalar;
-  typedef typename DiagonalVectorType::Index Index;
-  typedef typename DiagonalVectorType::StorageKind StorageKind;
+  typedef typename DiagonalVectorType::StorageIndex StorageIndex;
+  typedef DiagonalShape StorageKind;
+  typedef typename traits<DiagonalVectorType>::XprKind XprKind;
   enum {
     RowsAtCompileTime = DiagonalVectorType::SizeAtCompileTime,
     ColsAtCompileTime = DiagonalVectorType::SizeAtCompileTime,
-    MaxRowsAtCompileTime = DiagonalVectorType::SizeAtCompileTime,
-    MaxColsAtCompileTime = DiagonalVectorType::SizeAtCompileTime,
-    Flags =  ei_traits<DiagonalVectorType>::Flags & LvalueBit
+    MaxRowsAtCompileTime = DiagonalVectorType::MaxSizeAtCompileTime,
+    MaxColsAtCompileTime = DiagonalVectorType::MaxSizeAtCompileTime,
+    Flags =  (traits<DiagonalVectorType>::Flags & LvalueBit) | NoPreferredStorageOrderBit
   };
 };
+}
 
 template<typename _DiagonalVectorType>
 class DiagonalWrapper
-  : public DiagonalBase<DiagonalWrapper<_DiagonalVectorType> >, ei_no_assignment_operator
+  : public DiagonalBase<DiagonalWrapper<_DiagonalVectorType> >, internal::no_assignment_operator
 {
   public:
     #ifndef EIGEN_PARSED_BY_DOXYGEN
@@ -231,13 +253,15 @@ class DiagonalWrapper
     #endif
 
     /** Constructor from expression of diagonal coefficients to wrap. */
-    inline DiagonalWrapper(const DiagonalVectorType& diagonal) : m_diagonal(diagonal) {}
+    EIGEN_DEVICE_FUNC
+    explicit inline DiagonalWrapper(DiagonalVectorType& a_diagonal) : m_diagonal(a_diagonal) {}
 
     /** \returns a const reference to the wrapped expression of diagonal coefficients. */
+    EIGEN_DEVICE_FUNC
     const DiagonalVectorType& diagonal() const { return m_diagonal; }
 
   protected:
-    const typename DiagonalVectorType::Nested m_diagonal;
+    typename DiagonalVectorType::Nested m_diagonal;
 };
 
 /** \returns a pseudo-expression of a diagonal matrix with *this as vector of diagonal coefficients
@@ -250,10 +274,10 @@ class DiagonalWrapper
   * \sa class DiagonalWrapper, class DiagonalMatrix, diagonal(), isDiagonal()
   **/
 template<typename Derived>
-inline const DiagonalWrapper<Derived>
+inline const DiagonalWrapper<const Derived>
 MatrixBase<Derived>::asDiagonal() const
 {
-  return derived();
+  return DiagonalWrapper<const Derived>(derived());
 }
 
 /** \returns true if *this is approximately equal to a diagonal matrix,
@@ -265,23 +289,52 @@ MatrixBase<Derived>::asDiagonal() const
   * \sa asDiagonal()
   */
 template<typename Derived>
-bool MatrixBase<Derived>::isDiagonal
-(RealScalar prec) const
+bool MatrixBase<Derived>::isDiagonal(const RealScalar& prec) const
 {
+  using std::abs;
   if(cols() != rows()) return false;
   RealScalar maxAbsOnDiagonal = static_cast<RealScalar>(-1);
   for(Index j = 0; j < cols(); ++j)
   {
-    RealScalar absOnDiagonal = ei_abs(coeff(j,j));
+    RealScalar absOnDiagonal = abs(coeff(j,j));
     if(absOnDiagonal > maxAbsOnDiagonal) maxAbsOnDiagonal = absOnDiagonal;
   }
   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, typename Scalar>
+struct Assignment<DstXprType, SrcXprType, Functor, Diagonal2Dense, Scalar>
+{
+  static void run(DstXprType &dst, const SrcXprType &src, const internal::assign_op<typename DstXprType::Scalar> &/*func*/)
+  {
+    dst.setZero();
+    dst.diagonal() = src.diagonal();
+  }
+  
+  static void run(DstXprType &dst, const SrcXprType &src, const internal::add_assign_op<typename DstXprType::Scalar> &/*func*/)
+  { dst.diagonal() += src.diagonal(); }
+  
+  static void run(DstXprType &dst, const SrcXprType &src, const internal::sub_assign_op<typename DstXprType::Scalar> &/*func*/)
+  { dst.diagonal() -= src.diagonal(); }
+};
+
+} // namespace internal
+
+} // end namespace Eigen
+
 #endif // EIGEN_DIAGONALMATRIX_H

Eigen/src/Core/DiagonalProduct.h

@@ -4,130 +4,25 @@
 // 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
 
-template<typename MatrixType, typename DiagonalType, int ProductOrder>
-struct ei_traits<DiagonalProduct<MatrixType, DiagonalType, ProductOrder> >
- : ei_traits<MatrixType>
-{
-  typedef typename ei_scalar_product_traits<typename MatrixType::Scalar, typename DiagonalType::Scalar>::ReturnType Scalar;
-  enum {
-    RowsAtCompileTime = MatrixType::RowsAtCompileTime,
-    ColsAtCompileTime = MatrixType::ColsAtCompileTime,
-    MaxRowsAtCompileTime = MatrixType::MaxRowsAtCompileTime,
-    MaxColsAtCompileTime = MatrixType::MaxColsAtCompileTime,
-
-    _StorageOrder = MatrixType::Flags & RowMajorBit ? RowMajor : ColMajor,
-    _PacketOnDiag = !((int(_StorageOrder) == RowMajor && int(ProductOrder) == OnTheLeft)
-                    ||(int(_StorageOrder) == ColMajor && int(ProductOrder) == OnTheRight)),
-    _SameTypes = ei_is_same_type<typename MatrixType::Scalar, typename DiagonalType::Scalar>::ret,
-    // FIXME currently we need same types, but in the future the next rule should be the one
-    //_Vectorizable = bool(int(MatrixType::Flags)&PacketAccessBit) && ((!_PacketOnDiag) || (_SameTypes && bool(int(DiagonalType::Flags)&PacketAccessBit))),
-    _Vectorizable = bool(int(MatrixType::Flags)&PacketAccessBit) && _SameTypes && ((!_PacketOnDiag) || (bool(int(DiagonalType::Flags)&PacketAccessBit))),
-
-    Flags = (HereditaryBits & (unsigned int)(MatrixType::Flags)) | (_Vectorizable ? PacketAccessBit : 0),
-    CoeffReadCost = NumTraits<Scalar>::MulCost + MatrixType::CoeffReadCost + DiagonalType::DiagonalVectorType::CoeffReadCost
-  };
-};
-
-template<typename MatrixType, typename DiagonalType, int ProductOrder>
-class DiagonalProduct : ei_no_assignment_operator,
-                        public MatrixBase<DiagonalProduct<MatrixType, DiagonalType, ProductOrder> >
-{
-  public:
-
-    typedef MatrixBase<DiagonalProduct> Base;
-    EIGEN_DENSE_PUBLIC_INTERFACE(DiagonalProduct)
-
-    inline DiagonalProduct(const MatrixType& matrix, const DiagonalType& diagonal)
-      : m_matrix(matrix), m_diagonal(diagonal)
-    {
-      ei_assert(diagonal.diagonal().size() == (ProductOrder == OnTheLeft ? matrix.rows() : matrix.cols()));
-    }
-
-    inline Index rows() const { return m_matrix.rows(); }
-    inline Index cols() const { return m_matrix.cols(); }
-
-    const Scalar coeff(Index row, Index col) const
-    {
-      return m_diagonal.diagonal().coeff(ProductOrder == OnTheLeft ? row : col) * m_matrix.coeff(row, col);
-    }
-
-    template<int LoadMode>
-    EIGEN_STRONG_INLINE PacketScalar packet(Index row, Index col) const
-    {
-      enum {
-        StorageOrder = Flags & RowMajorBit ? RowMajor : ColMajor
-      };
-      const Index indexInDiagonalVector = ProductOrder == OnTheLeft ? row : col;
-
-      return packet_impl<LoadMode>(row,col,indexInDiagonalVector,typename ei_meta_if<
-        ((int(StorageOrder) == RowMajor && int(ProductOrder) == OnTheLeft)
-       ||(int(StorageOrder) == ColMajor && int(ProductOrder) == OnTheRight)), ei_meta_true, ei_meta_false>::ret());
-    }
-
-  protected:
-    template<int LoadMode>
-    EIGEN_STRONG_INLINE PacketScalar packet_impl(Index row, Index col, Index id, ei_meta_true) const
-    {
-      return ei_pmul(m_matrix.template packet<LoadMode>(row, col),
-                     ei_pset1<PacketScalar>(m_diagonal.diagonal().coeff(id)));
-    }
-
-    template<int LoadMode>
-    EIGEN_STRONG_INLINE PacketScalar packet_impl(Index row, Index col, Index id, ei_meta_false) const
-    {
-      enum {
-        InnerSize = (MatrixType::Flags & RowMajorBit) ? MatrixType::ColsAtCompileTime : MatrixType::RowsAtCompileTime,
-        DiagonalVectorPacketLoadMode = (LoadMode == Aligned && ((InnerSize%16) == 0)) ? Aligned : Unaligned
-      };
-      return ei_pmul(m_matrix.template packet<LoadMode>(row, col),
-                     m_diagonal.diagonal().template packet<DiagonalVectorPacketLoadMode>(id));
-    }
-
-    const typename MatrixType::Nested m_matrix;
-    const typename DiagonalType::Nested m_diagonal;
-};
+namespace Eigen { 
 
 /** \returns the diagonal matrix product of \c *this by the diagonal matrix \a diagonal.
   */
 template<typename Derived>
 template<typename DiagonalDerived>
-inline const DiagonalProduct<Derived, DiagonalDerived, OnTheRight>
-MatrixBase<Derived>::operator*(const DiagonalBase<DiagonalDerived> &diagonal) const
+inline const Product<Derived, DiagonalDerived, LazyProduct>
+MatrixBase<Derived>::operator*(const DiagonalBase<DiagonalDerived> &a_diagonal) const
 {
-  return DiagonalProduct<Derived, DiagonalDerived, OnTheRight>(derived(), diagonal.derived());
-}
-
-/** \returns the diagonal matrix product of \c *this by the matrix \a matrix.
-  */
-template<typename DiagonalDerived>
-template<typename MatrixDerived>
-inline const DiagonalProduct<MatrixDerived, DiagonalDerived, OnTheLeft>
-DiagonalBase<DiagonalDerived>::operator*(const MatrixBase<MatrixDerived> &matrix) const
-{
-  return DiagonalProduct<MatrixDerived, DiagonalDerived, OnTheLeft>(matrix.derived(), derived());
+  return Product<Derived, DiagonalDerived, LazyProduct>(derived(),a_diagonal.derived());
 }
 
+} // end namespace Eigen
 
 #endif // EIGEN_DIAGONALPRODUCT_H

Eigen/src/Core/Dot.h

@@ -3,28 +3,17 @@
 //
 // 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
 
+namespace Eigen { 
+
+namespace internal {
+
 // 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.
@@ -37,23 +26,29 @@ template<typename T, typename U,
                          // revert to || as soon as not needed anymore.
                     (int(T::ColsAtCompileTime) == 1 && int(U::RowsAtCompileTime) == 1))
 >
-struct ei_dot_nocheck
+struct dot_nocheck
 {
-  static inline typename ei_traits<T>::Scalar run(const MatrixBase<T>& a, const MatrixBase<U>& b)
+  typedef typename scalar_product_traits<typename traits<T>::Scalar,typename traits<U>::Scalar>::ReturnType ResScalar;
+  EIGEN_DEVICE_FUNC
+  static inline ResScalar run(const MatrixBase<T>& a, const MatrixBase<U>& b)
   {
-    return a.template binaryExpr<ei_scalar_conj_product_op<typename ei_traits<T>::Scalar> >(b).sum();
+    return a.template binaryExpr<scalar_conj_product_op<typename traits<T>::Scalar,typename traits<U>::Scalar> >(b).sum();
   }
 };
 
 template<typename T, typename U>
-struct ei_dot_nocheck<T, U, true>
+struct dot_nocheck<T, U, true>
 {
-  static inline typename ei_traits<T>::Scalar run(const MatrixBase<T>& a, const MatrixBase<U>& b)
+  typedef typename scalar_product_traits<typename traits<T>::Scalar,typename traits<U>::Scalar>::ReturnType ResScalar;
+  EIGEN_DEVICE_FUNC
+  static inline ResScalar run(const MatrixBase<T>& a, const MatrixBase<U>& b)
   {
-    return a.transpose().template binaryExpr<ei_scalar_conj_product_op<typename ei_traits<T>::Scalar> >(b).sum();
+    return a.transpose().template binaryExpr<scalar_conj_product_op<typename traits<T>::Scalar,typename traits<U>::Scalar> >(b).sum();
   }
 };
 
+} // end namespace internal
+
 /** \returns the dot product of *this with other.
   *
   * \only_for_vectors
@@ -66,40 +61,46 @@ struct ei_dot_nocheck<T, U, true>
   */
 template<typename Derived>
 template<typename OtherDerived>
-typename ei_traits<Derived>::Scalar
+EIGEN_DEVICE_FUNC
+typename internal::scalar_product_traits<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)
+  typedef internal::scalar_conj_product_op<Scalar,typename OtherDerived::Scalar> func;
+  EIGEN_CHECK_BINARY_COMPATIBILIY(func,Scalar,typename OtherDerived::Scalar);
 
-  ei_assert(size() == other.size());
+  eigen_assert(size() == other.size());
 
-  return ei_dot_nocheck<Derived,OtherDerived>::run(*this, other);
+  return internal::dot_nocheck<Derived,OtherDerived>::run(*this, other);
 }
 
 //---------- implementation of L2 norm and related functions ----------
 
-/** \returns the squared \em l2 norm of *this, i.e., for vectors, the dot product of *this with itself.
+/** \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()
   */
 template<typename Derived>
-EIGEN_STRONG_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).cwiseAbs2().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()
   */
 template<typename Derived>
-inline typename NumTraits<typename ei_traits<Derived>::Scalar>::Real MatrixBase<Derived>::norm() const
+inline typename NumTraits<typename internal::traits<Derived>::Scalar>::Real MatrixBase<Derived>::norm() const
 {
-  return ei_sqrt(squaredNorm());
+  EIGEN_USING_STD_MATH(sqrt)
+  return sqrt(squaredNorm());
 }
 
 /** \returns an expression of the quotient of *this by its own norm.
@@ -112,8 +113,7 @@ template<typename Derived>
 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();
 }
@@ -132,55 +132,66 @@ inline void MatrixBase<Derived>::normalize()
 
 //---------- implementation of other norms ----------
 
+namespace internal {
+
 template<typename Derived, int p>
-struct ei_lpNorm_selector
+struct lpNorm_selector
 {
-  typedef typename NumTraits<typename ei_traits<Derived>::Scalar>::Real RealScalar;
-  inline static RealScalar run(const MatrixBase<Derived>& m)
+  typedef typename NumTraits<typename traits<Derived>::Scalar>::Real RealScalar;
+  EIGEN_DEVICE_FUNC
+  static inline RealScalar run(const MatrixBase<Derived>& m)
   {
-    return ei_pow(m.cwiseAbs().array().pow(p).sum(), RealScalar(1)/p);
+    EIGEN_USING_STD_MATH(pow)
+    return pow(m.cwiseAbs().array().pow(p).sum(), RealScalar(1)/p);
   }
 };
 
 template<typename Derived>
-struct ei_lpNorm_selector<Derived, 1>
+struct lpNorm_selector<Derived, 1>
 {
-  inline static typename NumTraits<typename ei_traits<Derived>::Scalar>::Real run(const MatrixBase<Derived>& m)
+  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 ei_lpNorm_selector<Derived, 2>
+struct lpNorm_selector<Derived, 2>
 {
-  inline static typename NumTraits<typename ei_traits<Derived>::Scalar>::Real run(const MatrixBase<Derived>& m)
+  EIGEN_DEVICE_FUNC
+  static inline typename NumTraits<typename traits<Derived>::Scalar>::Real run(const MatrixBase<Derived>& m)
   {
     return m.norm();
   }
 };
 
 template<typename Derived>
-struct ei_lpNorm_selector<Derived, Infinity>
+struct lpNorm_selector<Derived, Infinity>
 {
-  inline static typename NumTraits<typename ei_traits<Derived>::Scalar>::Real run(const MatrixBase<Derived>& m)
+  EIGEN_DEVICE_FUNC
+  static inline typename NumTraits<typename traits<Derived>::Scalar>::Real run(const MatrixBase<Derived>& m)
   {
     return m.cwiseAbs().maxCoeff();
   }
 };
 
-/** \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.
+} // 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.
+  *
+  * \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>
-inline typename NumTraits<typename ei_traits<Derived>::Scalar>::Real
+inline typename NumTraits<typename internal::traits<Derived>::Scalar>::Real
 MatrixBase<Derived>::lpNorm() const
 {
-  return ei_lpNorm_selector<Derived, p>::run(*this);
+  return internal::lpNorm_selector<Derived, p>::run(*this);
 }
 
 //---------- implementation of isOrthogonal / isUnitary ----------
@@ -194,11 +205,11 @@ MatrixBase<Derived>::lpNorm() const
 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,
@@ -213,18 +224,20 @@ 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());
+  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<RealScalar>(1), prec))
+    if(!internal::isApprox(self.col(i).squaredNorm(), static_cast<RealScalar>(1), prec))
       return false;
     for(Index j = 0; j < i; ++j)
-      if(!ei_isMuchSmallerThan(nested.col(i).dot(nested.col(j)), static_cast<Scalar>(1), prec))
+      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

@@ -4,30 +4,18 @@
 // Copyright (C) 2009 Benoit Jacob <jacob.benoit.1@gmail.com>
 // Copyright (C) 2009 Gael Guennebaud <gael.guennebaud@inria.fr>
 //
-// Eigen is free software; you can redistribute it and/or
-// modify it under the terms of the GNU Lesser General Public
-// License as published by the Free Software Foundation; either
-// version 3 of the License, or (at your option) any later version.
-//
-// Alternatively, you can redistribute it and/or
-// modify it under the terms of the GNU General Public License as
-// published by the Free Software Foundation; either version 2 of
-// the License, or (at your option) any later version.
-//
-// Eigen is distributed in the hope that it will be useful, but WITHOUT ANY
-// WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS
-// FOR A PARTICULAR PURPOSE. See the GNU Lesser General Public License or the
-// GNU General Public License for more details.
-//
-// You should have received a copy of the GNU Lesser General Public
-// License and a copy of the GNU General Public License along with
-// Eigen. If not, see <http://www.gnu.org/licenses/>.
+// 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 {
 
-/** Common base class for all classes T such that MatrixBase has an operator=(T) and a constructor MatrixBase(T).
+/** \class EigenBase
+  * 
+  * 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.
   *
@@ -39,33 +27,53 @@
   */
 template<typename Derived> struct EigenBase
 {
-//   typedef typename ei_plain_matrix_type<Derived>::type PlainObject;
+//   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;
 
-  typedef typename ei_traits<Derived>::StorageKind StorageKind;
-  typedef typename ei_traits<Derived>::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> inline void evalTo(Dest& dst) const
+  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> inline void addTo(Dest& dst) const
+  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.
@@ -75,7 +83,9 @@ template<typename Derived> struct EigenBase
   }
 
   /** \internal Don't use it, but do the equivalent: \code dst -= *this; \endcode */
-  template<typename Dest> inline void subTo(Dest& dst) const
+  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.
@@ -85,7 +95,8 @@ template<typename Derived> struct EigenBase
   }
 
   /** \internal Don't use it, but do the equivalent: \code dst.applyOnTheRight(*this); \endcode */
-  template<typename Dest> inline void applyThisOnTheRight(Dest& dst) const
+  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.
@@ -93,7 +104,8 @@ template<typename Derived> struct EigenBase
   }
 
   /** \internal Don't use it, but do the equivalent: \code dst.applyOnTheLeft(*this); \endcode */
-  template<typename Dest> inline void applyThisOnTheLeft(Dest& dst) const
+  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.
@@ -118,7 +130,7 @@ template<typename Derived>
 template<typename OtherDerived>
 Derived& DenseBase<Derived>::operator=(const EigenBase<OtherDerived> &other)
 {
-  other.derived().evalTo(derived());
+  call_assignment(derived(), other.derived());
   return derived();
 }
 
@@ -126,7 +138,7 @@ template<typename Derived>
 template<typename OtherDerived>
 Derived& DenseBase<Derived>::operator+=(const EigenBase<OtherDerived> &other)
 {
-  other.derived().addTo(derived());
+  call_assignment(derived(), other.derived(), internal::add_assign_op<Scalar>());
   return derived();
 }
 
@@ -134,37 +146,10 @@ template<typename Derived>
 template<typename OtherDerived>
 Derived& DenseBase<Derived>::operator-=(const EigenBase<OtherDerived> &other)
 {
-  other.derived().subTo(derived());
+  call_assignment(derived(), other.derived(), internal::sub_assign_op<Scalar>());
   return derived();
 }
 
-/** replaces \c *this by \c *this * \a other.
-  *
-  * \returns a reference to \c *this
-  */
-template<typename Derived>
-template<typename OtherDerived>
-inline Derived&
-MatrixBase<Derived>::operator*=(const EigenBase<OtherDerived> &other)
-{
-  other.derived().applyThisOnTheRight(derived());
-  return derived();
-}
-
-/** replaces \c *this by \c *this * \a other. It is equivalent to MatrixBase::operator*=() */
-template<typename Derived>
-template<typename OtherDerived>
-inline void MatrixBase<Derived>::applyOnTheRight(const EigenBase<OtherDerived> &other)
-{
-  other.derived().applyThisOnTheRight(derived());
-}
-
-/** replaces \c *this by \c *this * \a other. */
-template<typename Derived>
-template<typename OtherDerived>
-inline void MatrixBase<Derived>::applyOnTheLeft(const EigenBase<OtherDerived> &other)
-{
-  other.derived().applyThisOnTheLeft(derived());
-}
+} // end namespace Eigen
 
 #endif // EIGEN_EIGENBASE_H

Eigen/src/Core/Flagged.h

@@ -1,137 +0,0 @@
-// This file is part of Eigen, a lightweight C++ template library
-// for linear algebra.
-//
-// 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
-  * \ingroup Core_Module
-  *
-  * \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:
-
-    typedef MatrixBase<Flagged> Base;
-    EIGEN_DENSE_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 Index rows() const { return m_matrix.rows(); }
-    inline Index cols() const { return m_matrix.cols(); }
-    inline Index outerStride() const { return m_matrix.outerStride(); }
-    inline Index innerStride() const { return m_matrix.innerStride(); }
-
-    inline const Scalar coeff(Index row, Index col) const
-    {
-      return m_matrix.coeff(row, col);
-    }
-
-    inline Scalar& coeffRef(Index row, Index col)
-    {
-      return m_matrix.const_cast_derived().coeffRef(row, col);
-    }
-
-    inline const Scalar coeff(Index index) const
-    {
-      return m_matrix.coeff(index);
-    }
-
-    inline Scalar& coeffRef(Index index)
-    {
-      return m_matrix.const_cast_derived().coeffRef(index);
-    }
-
-    template<int LoadMode>
-    inline const PacketScalar packet(Index row, Index col) const
-    {
-      return m_matrix.template packet<LoadMode>(row, col);
-    }
-
-    template<int LoadMode>
-    inline void writePacket(Index row, Index col, const PacketScalar& x)
-    {
-      m_matrix.const_cast_derived().template writePacket<LoadMode>(row, col, x);
-    }
-
-    template<int LoadMode>
-    inline const PacketScalar packet(Index index) const
-    {
-      return m_matrix.template packet<LoadMode>(index);
-    }
-
-    template<int LoadMode>
-    inline void writePacket(Index index, const PacketScalar& x)
-    {
-      m_matrix.const_cast_derived().template writePacket<LoadMode>(index, x);
-    }
-
-    const ExpressionType& _expression() const { return m_matrix; }
-
-    template<typename OtherDerived>
-    typename ExpressionType::PlainObject solveTriangular(const MatrixBase<OtherDerived>& other) const;
-
-    template<typename OtherDerived>
-    void solveTriangularInPlace(const MatrixBase<OtherDerived>& other) const;
-
-  protected:
-    ExpressionTypeNested m_matrix;
-};
-
-/** \returns an expression of *this with added and removed flags
-  *
-  * This is mostly for internal use.
-  *
-  * \sa class Flagged
-  */
-template<typename Derived>
-template<unsigned int Added,unsigned int Removed>
-inline const Flagged<Derived, Added, Removed>
-DenseBase<Derived>::flagged() const
-{
-  return derived();
-}
-
-#endif // EIGEN_FLAGGED_H

Eigen/src/Core/ForceAlignedAccess.h

@@ -3,28 +3,15 @@
 //
 // Copyright (C) 2009-2010 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_FORCEALIGNEDACCESS_H
 #define EIGEN_FORCEALIGNEDACCESS_H
 
+namespace Eigen {
+
 /** \class ForceAlignedAccess
   * \ingroup Core_Module
   *
@@ -37,41 +24,44 @@
   *
   * \sa MatrixBase::forceAlignedAccess()
   */
+
+namespace internal {
 template<typename ExpressionType>
-struct ei_traits<ForceAlignedAccess<ExpressionType> > : public ei_traits<ExpressionType>
+struct traits<ForceAlignedAccess<ExpressionType> > : public traits<ExpressionType>
 {};
+}
 
 template<typename ExpressionType> class ForceAlignedAccess
-  : public ei_dense_xpr_base< ForceAlignedAccess<ExpressionType> >::type
+  : public internal::dense_xpr_base< ForceAlignedAccess<ExpressionType> >::type
 {
   public:
 
-    typedef typename ei_dense_xpr_base<ForceAlignedAccess>::type Base;
+    typedef typename internal::dense_xpr_base<ForceAlignedAccess>::type Base;
     EIGEN_DENSE_PUBLIC_INTERFACE(ForceAlignedAccess)
 
-    inline ForceAlignedAccess(const ExpressionType& matrix) : m_expression(matrix) {}
+    EIGEN_DEVICE_FUNC explicit inline ForceAlignedAccess(const ExpressionType& matrix) : m_expression(matrix) {}
 
-    inline Index rows() const { return m_expression.rows(); }
-    inline Index cols() const { return m_expression.cols(); }
-    inline Index outerStride() const { return m_expression.outerStride(); }
-    inline Index innerStride() const { return m_expression.innerStride(); }
+    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(); }
 
-    inline const CoeffReturnType coeff(Index row, Index col) const
+    EIGEN_DEVICE_FUNC inline const CoeffReturnType coeff(Index row, Index col) const
     {
       return m_expression.coeff(row, col);
     }
 
-    inline Scalar& coeffRef(Index row, Index col)
+    EIGEN_DEVICE_FUNC inline Scalar& coeffRef(Index row, Index col)
     {
       return m_expression.const_cast_derived().coeffRef(row, col);
     }
 
-    inline const CoeffReturnType coeff(Index index) const
+    EIGEN_DEVICE_FUNC inline const CoeffReturnType coeff(Index index) const
     {
       return m_expression.coeff(index);
     }
 
-    inline Scalar& coeffRef(Index index)
+    EIGEN_DEVICE_FUNC inline Scalar& coeffRef(Index index)
     {
       return m_expression.const_cast_derived().coeffRef(index);
     }
@@ -100,7 +90,7 @@ template<typename ExpressionType> class ForceAlignedAccess
       m_expression.const_cast_derived().template writePacket<Aligned>(index, x);
     }
 
-    operator const ExpressionType&() const { return m_expression; }
+    EIGEN_DEVICE_FUNC operator const ExpressionType&() const { return m_expression; }
 
   protected:
     const ExpressionType& m_expression;
@@ -134,10 +124,10 @@ MatrixBase<Derived>::forceAlignedAccess()
   */
 template<typename Derived>
 template<bool Enable>
-inline typename ei_makeconst<typename ei_meta_if<Enable,ForceAlignedAccess<Derived>,Derived&>::ret>::type
+inline typename internal::add_const_on_value_type<typename internal::conditional<Enable,ForceAlignedAccess<Derived>,Derived&>::type>::type
 MatrixBase<Derived>::forceAlignedAccessIf() const
 {
-  return derived();
+  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.
@@ -145,10 +135,12 @@ MatrixBase<Derived>::forceAlignedAccessIf() const
   */
 template<typename Derived>
 template<bool Enable>
-inline typename ei_meta_if<Enable,ForceAlignedAccess<Derived>,Derived&>::ret
+inline typename internal::conditional<Enable,ForceAlignedAccess<Derived>,Derived&>::type
 MatrixBase<Derived>::forceAlignedAccessIf()
 {
-  return derived();
+  return derived();  // FIXME This should not work but apparently is never used
 }
 
+} // end namespace Eigen
+
 #endif // EIGEN_FORCEALIGNEDACCESS_H

Eigen/src/Core/Functors.h

@@ -1,847 +0,0 @@
-// This file is part of Eigen, a lightweight C++ template library
-// for linear algebra.
-//
-// Copyright (C) 2008-2010 Gael Guennebaud <gael.guennebaud@inria.fr>
-//
-// Eigen is free software; you can redistribute it and/or
-// modify it under the terms of the GNU Lesser General Public
-// License as published by the Free Software Foundation; either
-// version 3 of the License, or (at your option) any later version.
-//
-// Alternatively, you can redistribute it and/or
-// modify it under the terms of the GNU General Public License as
-// published by the Free Software Foundation; either version 2 of
-// the License, or (at your option) any later version.
-//
-// Eigen is distributed in the hope that it will be useful, but WITHOUT ANY
-// WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS
-// FOR A PARTICULAR PURPOSE. See the GNU Lesser General Public License or the
-// GNU General Public License for more details.
-//
-// You should have received a copy of the GNU Lesser General Public
-// License and a copy of the GNU General Public License along with
-// Eigen. If not, see <http://www.gnu.org/licenses/>.
-
-#ifndef EIGEN_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 VectorwiseOp, MatrixBase::sum()
-  */
-template<typename Scalar> struct ei_scalar_sum_op {
-  EIGEN_EMPTY_STRUCT_CTOR(ei_scalar_sum_op)
-  EIGEN_STRONG_INLINE const Scalar operator() (const Scalar& a, const Scalar& b) const { return a + b; }
-  template<typename Packet>
-  EIGEN_STRONG_INLINE const Packet packetOp(const Packet& a, const Packet& b) const
-  { return ei_padd(a,b); }
-  template<typename Packet>
-  EIGEN_STRONG_INLINE const Scalar predux(const Packet& a) const
-  { return ei_predux(a); }
-};
-template<typename Scalar>
-struct ei_functor_traits<ei_scalar_sum_op<Scalar> > {
-  enum {
-    Cost = NumTraits<Scalar>::AddCost,
-    PacketAccess = ei_packet_traits<Scalar>::HasAdd
-  };
-};
-
-/** \internal
-  * \brief Template functor to compute the product of two scalars
-  *
-  * \sa class CwiseBinaryOp, Cwise::operator*(), class VectorwiseOp, MatrixBase::redux()
-  */
-template<typename LhsScalar,typename RhsScalar> struct ei_scalar_product_op {
-  enum {
-    Vectorizable = ei_is_same_type<LhsScalar,RhsScalar>::ret && ei_packet_traits<LhsScalar>::HasMul && ei_packet_traits<RhsScalar>::HasMul
-  };
-  typedef typename ei_scalar_product_traits<LhsScalar,RhsScalar>::ReturnType result_type;
-  EIGEN_EMPTY_STRUCT_CTOR(ei_scalar_product_op)
-  EIGEN_STRONG_INLINE const result_type operator() (const LhsScalar& a, const RhsScalar& b) const { return a * b; }
-  template<typename Packet>
-  EIGEN_STRONG_INLINE const Packet packetOp(const Packet& a, const Packet& b) const
-  { return ei_pmul(a,b); }
-  template<typename Packet>
-  EIGEN_STRONG_INLINE const result_type predux(const Packet& a) const
-  { return ei_predux_mul(a); }
-};
-template<typename LhsScalar,typename RhsScalar>
-struct ei_functor_traits<ei_scalar_product_op<LhsScalar,RhsScalar> > {
-  enum {
-    Cost = (NumTraits<LhsScalar>::MulCost + NumTraits<RhsScalar>::MulCost)/2, // rough estimate!
-    PacketAccess = ei_scalar_product_op<LhsScalar,RhsScalar>::Vectorizable
-  };
-};
-
-/** \internal
-  * \brief Template functor to compute the conjugate product of two scalars
-  *
-  * This is a short cut for ei_conj(x) * y which is needed for optimization purpose
-  */
-template<typename Scalar> struct ei_scalar_conj_product_op {
-  enum { Conj = NumTraits<Scalar>::IsComplex };
-  EIGEN_EMPTY_STRUCT_CTOR(ei_scalar_conj_product_op)
-  EIGEN_STRONG_INLINE const Scalar operator() (const Scalar& a, const Scalar& b) const
-  { return ei_conj_helper<Scalar,Scalar,Conj,false>().pmul(a,b); }
-  template<typename Packet>
-  EIGEN_STRONG_INLINE const Packet packetOp(const Packet& a, const Packet& b) const
-  { return ei_conj_helper<Packet,Packet,Conj,false>().pmul(a,b); }
-};
-template<typename Scalar>
-struct ei_functor_traits<ei_scalar_conj_product_op<Scalar> > {
-  enum {
-    Cost = NumTraits<Scalar>::MulCost,
-    PacketAccess = ei_packet_traits<Scalar>::HasMul
-  };
-};
-
-/** \internal
-  * \brief Template functor to compute the min of two scalars
-  *
-  * \sa class CwiseBinaryOp, MatrixBase::cwiseMin, class VectorwiseOp, MatrixBase::minCoeff()
-  */
-template<typename Scalar> struct ei_scalar_min_op {
-  EIGEN_EMPTY_STRUCT_CTOR(ei_scalar_min_op)
-  EIGEN_STRONG_INLINE const Scalar operator() (const Scalar& a, const Scalar& b) const { return std::min(a, b); }
-  template<typename Packet>
-  EIGEN_STRONG_INLINE const Packet packetOp(const Packet& a, const Packet& b) const
-  { return ei_pmin(a,b); }
-  template<typename Packet>
-  EIGEN_STRONG_INLINE const Scalar predux(const Packet& a) const
-  { return ei_predux_min(a); }
-};
-template<typename Scalar>
-struct ei_functor_traits<ei_scalar_min_op<Scalar> > {
-  enum {
-    Cost = NumTraits<Scalar>::AddCost,
-    PacketAccess = ei_packet_traits<Scalar>::HasMin
-  };
-};
-
-/** \internal
-  * \brief Template functor to compute the max of two scalars
-  *
-  * \sa class CwiseBinaryOp, MatrixBase::cwiseMax, class VectorwiseOp, MatrixBase::maxCoeff()
-  */
-template<typename Scalar> struct ei_scalar_max_op {
-  EIGEN_EMPTY_STRUCT_CTOR(ei_scalar_max_op)
-  EIGEN_STRONG_INLINE const Scalar operator() (const Scalar& a, const Scalar& b) const { return std::max(a, b); }
-  template<typename Packet>
-  EIGEN_STRONG_INLINE const Packet packetOp(const Packet& a, const Packet& b) const
-  { return ei_pmax(a,b); }
-  template<typename Packet>
-  EIGEN_STRONG_INLINE const Scalar predux(const Packet& a) const
-  { return ei_predux_max(a); }
-};
-template<typename Scalar>
-struct ei_functor_traits<ei_scalar_max_op<Scalar> > {
-  enum {
-    Cost = NumTraits<Scalar>::AddCost,
-    PacketAccess = ei_packet_traits<Scalar>::HasMax
-  };
-};
-
-/** \internal
-  * \brief Template functor to compute the hypot of two scalars
-  *
-  * \sa MatrixBase::stableNorm(), class Redux
-  */
-template<typename Scalar> struct ei_scalar_hypot_op {
-  EIGEN_EMPTY_STRUCT_CTOR(ei_scalar_hypot_op)
-//   typedef typename NumTraits<Scalar>::Real result_type;
-  EIGEN_STRONG_INLINE const Scalar operator() (const Scalar& _x, const Scalar& _y) const
-  {
-    Scalar p = std::max(_x, _y);
-    Scalar q = std::min(_x, _y);
-    Scalar qp = q/p;
-    return p * ei_sqrt(Scalar(1) + qp*qp);
-  }
-};
-template<typename Scalar>
-struct ei_functor_traits<ei_scalar_hypot_op<Scalar> > {
-  enum { Cost = 5 * NumTraits<Scalar>::MulCost, PacketAccess=0 };
-};
-
-// 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_CTOR(ei_scalar_difference_op)
-  EIGEN_STRONG_INLINE const Scalar operator() (const Scalar& a, const Scalar& b) const { return a - b; }
-  template<typename Packet>
-  EIGEN_STRONG_INLINE const Packet packetOp(const Packet& a, const Packet& 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>::HasSub
-  };
-};
-
-/** \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_CTOR(ei_scalar_quotient_op)
-  EIGEN_STRONG_INLINE const Scalar operator() (const Scalar& a, const Scalar& b) const { return a / b; }
-  template<typename Packet>
-  EIGEN_STRONG_INLINE const Packet packetOp(const Packet& a, const Packet& 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>::HasDiv
-  };
-};
-
-// 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_CTOR(ei_scalar_opposite_op)
-  EIGEN_STRONG_INLINE const Scalar operator() (const Scalar& a) const { return -a; }
-  template<typename Packet>
-  EIGEN_STRONG_INLINE const Packet packetOp(const Packet& a) const
-  { return ei_pnegate(a); }
-};
-template<typename Scalar>
-struct ei_functor_traits<ei_scalar_opposite_op<Scalar> >
-{ enum {
-    Cost = NumTraits<Scalar>::AddCost,
-    PacketAccess = ei_packet_traits<Scalar>::HasNegate };
-};
-
-/** \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_CTOR(ei_scalar_abs_op)
-  typedef typename NumTraits<Scalar>::Real result_type;
-  EIGEN_STRONG_INLINE const result_type operator() (const Scalar& a) const { return ei_abs(a); }
-  template<typename Packet>
-  EIGEN_STRONG_INLINE const Packet packetOp(const Packet& a) const
-  { return ei_pabs(a); }
-};
-template<typename Scalar>
-struct ei_functor_traits<ei_scalar_abs_op<Scalar> >
-{
-  enum {
-    Cost = NumTraits<Scalar>::AddCost,
-    PacketAccess = ei_packet_traits<Scalar>::HasAbs
-  };
-};
-
-/** \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_CTOR(ei_scalar_abs2_op)
-  typedef typename NumTraits<Scalar>::Real result_type;
-  EIGEN_STRONG_INLINE const result_type operator() (const Scalar& a) const { return ei_abs2(a); }
-  template<typename Packet>
-  EIGEN_STRONG_INLINE const Packet packetOp(const Packet& 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 = ei_packet_traits<Scalar>::HasAbs2 }; };
-
-/** \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_CTOR(ei_scalar_conjugate_op)
-  EIGEN_STRONG_INLINE const Scalar operator() (const Scalar& a) const { return ei_conj(a); }
-  template<typename Packet>
-  EIGEN_STRONG_INLINE const Packet packetOp(const Packet& a) const { return ei_pconj(a); }
-};
-template<typename Scalar>
-struct ei_functor_traits<ei_scalar_conjugate_op<Scalar> >
-{
-  enum {
-    Cost = NumTraits<Scalar>::IsComplex ? NumTraits<Scalar>::AddCost : 0,
-    PacketAccess = ei_packet_traits<Scalar>::HasConj
-  };
-};
-
-/** \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_CTOR(ei_scalar_cast_op)
-  typedef NewType result_type;
-  EIGEN_STRONG_INLINE const NewType operator() (const Scalar& a) const { return ei_cast<Scalar, 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_CTOR(ei_scalar_real_op)
-  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_CTOR(ei_scalar_imag_op)
-  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 extract the real part of a complex as a reference
-  *
-  * \sa class CwiseUnaryOp, MatrixBase::real()
-  */
-template<typename Scalar>
-struct ei_scalar_real_ref_op {
-  EIGEN_EMPTY_STRUCT_CTOR(ei_scalar_real_ref_op)
-  typedef typename NumTraits<Scalar>::Real result_type;
-  EIGEN_STRONG_INLINE result_type& operator() (const Scalar& a) const { return ei_real_ref(*const_cast<Scalar*>(&a)); }
-};
-template<typename Scalar>
-struct ei_functor_traits<ei_scalar_real_ref_op<Scalar> >
-{ enum { Cost = 0, PacketAccess = false }; };
-
-/** \internal
-  * \brief Template functor to extract the imaginary part of a complex as a reference
-  *
-  * \sa class CwiseUnaryOp, MatrixBase::imag()
-  */
-template<typename Scalar>
-struct ei_scalar_imag_ref_op {
-  EIGEN_EMPTY_STRUCT_CTOR(ei_scalar_imag_ref_op)
-  typedef typename NumTraits<Scalar>::Real result_type;
-  EIGEN_STRONG_INLINE result_type& operator() (const Scalar& a) const { return ei_imag_ref(*const_cast<Scalar*>(&a)); }
-};
-template<typename Scalar>
-struct ei_functor_traits<ei_scalar_imag_ref_op<Scalar> >
-{ enum { Cost = 0, PacketAccess = false }; };
-
-/** \internal
-  *
-  * \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_CTOR(ei_scalar_exp_op)
-  inline const Scalar operator() (const Scalar& a) const { return ei_exp(a); }
-  typedef typename ei_packet_traits<Scalar>::type Packet;
-  inline Packet packetOp(const Packet& a) const { return ei_pexp(a); }
-};
-template<typename Scalar>
-struct ei_functor_traits<ei_scalar_exp_op<Scalar> >
-{ enum { Cost = 5 * NumTraits<Scalar>::MulCost, PacketAccess = ei_packet_traits<Scalar>::HasExp }; };
-
-/** \internal
-  *
-  * \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_CTOR(ei_scalar_log_op)
-  inline const Scalar operator() (const Scalar& a) const { return ei_log(a); }
-  typedef typename ei_packet_traits<Scalar>::type Packet;
-  inline Packet packetOp(const Packet& a) const { return ei_plog(a); }
-};
-template<typename Scalar>
-struct ei_functor_traits<ei_scalar_log_op<Scalar> >
-{ enum { Cost = 5 * NumTraits<Scalar>::MulCost, PacketAccess = ei_packet_traits<Scalar>::HasLog }; };
-
-/** \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 Packet and do the ei_pset1() once
- * in the constructor. However, in practice:
- *  - GCC does not like m_other as a Packet and generate a load every time it needs it
- *  - on 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 Packet;
-  // 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 Packet packetOp(const Packet& a) const
-  { return ei_pmul(a, ei_pset1<Packet>(m_other)); }
-  typename ei_makeconst<typename NumTraits<Scalar>::Nested>::type m_other;
-};
-template<typename Scalar>
-struct ei_functor_traits<ei_scalar_multiple_op<Scalar> >
-{ enum { Cost = NumTraits<Scalar>::MulCost, PacketAccess = ei_packet_traits<Scalar>::HasMul }; };
-
-template<typename Scalar1, typename Scalar2>
-struct ei_scalar_multiple2_op {
-  typedef typename ei_scalar_product_traits<Scalar1,Scalar2>::ReturnType result_type;
-  EIGEN_STRONG_INLINE ei_scalar_multiple2_op(const ei_scalar_multiple2_op& other) : m_other(other.m_other) { }
-  EIGEN_STRONG_INLINE ei_scalar_multiple2_op(const Scalar2& other) : m_other(other) { }
-  EIGEN_STRONG_INLINE result_type operator() (const Scalar1& a) const { return a * m_other; }
-  typename ei_makeconst<typename NumTraits<Scalar2>::Nested>::type m_other;
-};
-template<typename Scalar1,typename Scalar2>
-struct ei_functor_traits<ei_scalar_multiple2_op<Scalar1,Scalar2> >
-{ enum { Cost = NumTraits<Scalar1>::MulCost, PacketAccess = false }; };
-
-template<typename Scalar, bool IsInteger>
-struct ei_scalar_quotient1_impl {
-  typedef typename ei_packet_traits<Scalar>::type Packet;
-  // 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 Packet packetOp(const Packet& a) const
-  { return ei_pmul(a, ei_pset1<Packet>(m_other)); }
-  const Scalar m_other;
-};
-template<typename Scalar>
-struct ei_functor_traits<ei_scalar_quotient1_impl<Scalar,false> >
-{ enum { Cost = NumTraits<Scalar>::MulCost, PacketAccess = ei_packet_traits<Scalar>::HasMul }; };
-
-template<typename Scalar>
-struct ei_scalar_quotient1_impl<Scalar,true> {
-  // 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; }
-  typename ei_makeconst<typename NumTraits<Scalar>::Nested>::type m_other;
-};
-template<typename Scalar>
-struct ei_functor_traits<ei_scalar_quotient1_impl<Scalar,true> >
-{ 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>::IsInteger > {
-  EIGEN_STRONG_INLINE ei_scalar_quotient1_op(const Scalar& other)
-    : ei_scalar_quotient1_impl<Scalar, NumTraits<Scalar>::IsInteger >(other) {}
-};
-template<typename Scalar>
-struct ei_functor_traits<ei_scalar_quotient1_op<Scalar> >
-: ei_functor_traits<ei_scalar_quotient1_impl<Scalar, NumTraits<Scalar>::IsInteger> >
-{};
-
-// nullary functors
-
-template<typename Scalar>
-struct ei_scalar_constant_op {
-  typedef typename ei_packet_traits<Scalar>::type Packet;
-  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) { }
-  template<typename Index>
-  EIGEN_STRONG_INLINE const Scalar operator() (Index, Index = 0) const { return m_other; }
-  template<typename Index>
-  EIGEN_STRONG_INLINE const Packet packetOp(Index, Index = 0) const { return ei_pset1<Packet>(m_other); }
-  const Scalar m_other;
-};
-template<typename Scalar>
-struct ei_functor_traits<ei_scalar_constant_op<Scalar> >
-// FIXME replace this packet test by a safe one
-{ enum { Cost = 1, PacketAccess = ei_packet_traits<Scalar>::Vectorizable, IsRepeatable = true }; };
-
-template<typename Scalar> struct ei_scalar_identity_op {
-  EIGEN_EMPTY_STRUCT_CTOR(ei_scalar_identity_op)
-  template<typename Index>
-  EIGEN_STRONG_INLINE const Scalar operator() (Index row, Index 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 }; };
-
-template <typename Scalar, bool RandomAccess> struct ei_linspaced_op_impl;
-
-// linear access for packet ops:
-// 1) initialization
-//   base = [low, ..., low] + ([step, ..., step] * [-size, ..., 0])
-// 2) each step
-//   base += [size*step, ..., size*step]
-template <typename Scalar>
-struct ei_linspaced_op_impl<Scalar,false>
-{
-  typedef typename ei_packet_traits<Scalar>::type Packet;
-
-  ei_linspaced_op_impl(Scalar low, Scalar step) :
-  m_low(low), m_step(step),
-  m_packetStep(ei_pset1<Packet>(ei_packet_traits<Scalar>::size*step)),
-  m_base(ei_padd(ei_pset1<Packet>(low),ei_pmul(ei_pset1<Packet>(step),ei_plset<Scalar>(-ei_packet_traits<Scalar>::size)))) {}
-
-  template<typename Index>
-  EIGEN_STRONG_INLINE const Scalar operator() (Index i) const { return m_low+i*m_step; }
-  template<typename Index>
-  EIGEN_STRONG_INLINE const Packet packetOp(Index) const { return m_base = ei_padd(m_base,m_packetStep); }
-
-  const Scalar m_low;
-  const Scalar m_step;
-  const Packet m_packetStep;
-  mutable Packet m_base;
-};
-
-// random access for packet ops:
-// 1) each step
-//   [low, ..., low] + ( [step, ..., step] * ( [i, ..., i] + [0, ..., size] ) )
-template <typename Scalar>
-struct ei_linspaced_op_impl<Scalar,true>
-{
-  typedef typename ei_packet_traits<Scalar>::type Packet;
-
-  ei_linspaced_op_impl(Scalar low, Scalar step) :
-  m_low(low), m_step(step),
-  m_lowPacket(ei_pset1<Packet>(m_low)), m_stepPacket(ei_pset1<Packet>(m_step)), m_interPacket(ei_plset<Scalar>(0)) {}
-
-  template<typename Index>
-  EIGEN_STRONG_INLINE const Scalar operator() (Index i) const { return m_low+i*m_step; }
-  template<typename Index>
-  EIGEN_STRONG_INLINE const Packet packetOp(Index i) const
-  { return ei_padd(m_lowPacket, ei_pmul(m_stepPacket, ei_padd(ei_pset1<Packet>(i),m_interPacket))); }
-
-  const Scalar m_low;
-  const Scalar m_step;
-  const Packet m_lowPacket;
-  const Packet m_stepPacket;
-  const Packet m_interPacket;
-};
-
-// ----- Linspace functor ----------------------------------------------------------------
-
-// Forward declaration (we default to random access which does not really give
-// us a speed gain when using packet access but it allows to use the functor in
-// nested expressions).
-template <typename Scalar, bool RandomAccess = true> struct ei_linspaced_op;
-template <typename Scalar, bool RandomAccess> struct ei_functor_traits< ei_linspaced_op<Scalar,RandomAccess> >
-{ enum { Cost = 1, PacketAccess = ei_packet_traits<Scalar>::HasSetLinear, IsRepeatable = true }; };
-template <typename Scalar, bool RandomAccess> struct ei_linspaced_op
-{
-  typedef typename ei_packet_traits<Scalar>::type Packet;
-  ei_linspaced_op(Scalar low, Scalar high, int num_steps) : impl(low, (high-low)/(num_steps-1)) {}
-  template<typename Index>
-  EIGEN_STRONG_INLINE const Scalar operator() (Index i, Index = 0) const { return impl(i); }
-  template<typename Index>
-  EIGEN_STRONG_INLINE const Packet packetOp(Index i, Index = 0) const { return impl.packetOp(i); }
-  // This proxy object handles the actual required temporaries, the different
-  // implementations (random vs. sequential access) as well as the
-  // correct piping to size 2/4 packet operations.
-  const ei_linspaced_op_impl<Scalar,RandomAccess> impl;
-};
-
-// 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.
-// FIXME move this to ei_functor_traits adding a ei_functor_default
-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>.
-// FIXME move this to ei_functor_traits adding a ei_functor_default
-template<typename Functor> struct ei_functor_allows_mixing_real_and_complex { enum { ret = 0 }; };
-template<typename LhsScalar,typename RhsScalar> struct ei_functor_allows_mixing_real_and_complex<ei_scalar_product_op<LhsScalar,RhsScalar> > { enum { ret = 1 }; };
-
-
-/** \internal
-  * \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 Packet;
-  // 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 Packet packetOp(const Packet& a) const
-  { return ei_padd(a, ei_pset1<Packet>(m_other)); }
-  const Scalar m_other;
-};
-template<typename Scalar>
-struct ei_functor_traits<ei_scalar_add_op<Scalar> >
-{ enum { Cost = NumTraits<Scalar>::AddCost, PacketAccess = ei_packet_traits<Scalar>::HasAdd }; };
-
-/** \internal
-  * \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_CTOR(ei_scalar_sqrt_op)
-  inline const Scalar operator() (const Scalar& a) const { return ei_sqrt(a); }
-  typedef typename ei_packet_traits<Scalar>::type Packet;
-  inline Packet packetOp(const Packet& a) const { return ei_psqrt(a); }
-};
-template<typename Scalar>
-struct ei_functor_traits<ei_scalar_sqrt_op<Scalar> >
-{ enum {
-    Cost = 5 * NumTraits<Scalar>::MulCost,
-    PacketAccess = ei_packet_traits<Scalar>::HasSqrt
-  };
-};
-
-/** \internal
-  * \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_CTOR(ei_scalar_cos_op)
-  inline Scalar operator() (const Scalar& a) const { return ei_cos(a); }
-  typedef typename ei_packet_traits<Scalar>::type Packet;
-  inline Packet packetOp(const Packet& a) const { return ei_pcos(a); }
-};
-template<typename Scalar>
-struct ei_functor_traits<ei_scalar_cos_op<Scalar> >
-{
-  enum {
-    Cost = 5 * NumTraits<Scalar>::MulCost,
-    PacketAccess = ei_packet_traits<Scalar>::HasCos
-  };
-};
-
-/** \internal
-  * \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_CTOR(ei_scalar_sin_op)
-  inline const Scalar operator() (const Scalar& a) const { return ei_sin(a); }
-  typedef typename ei_packet_traits<Scalar>::type Packet;
-  inline Packet packetOp(const Packet& a) const { return ei_psin(a); }
-};
-template<typename Scalar>
-struct ei_functor_traits<ei_scalar_sin_op<Scalar> >
-{
-  enum {
-    Cost = 5 * NumTraits<Scalar>::MulCost,
-    PacketAccess = ei_packet_traits<Scalar>::HasSin
-  };
-};
-
-/** \internal
-  * \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;
-};
-template<typename Scalar>
-struct ei_functor_traits<ei_scalar_pow_op<Scalar> >
-{ enum { Cost = 5 * NumTraits<Scalar>::MulCost, PacketAccess = false }; };
-
-/** \internal
-  * \brief Template functor to compute the inverse of a scalar
-  * \sa class CwiseUnaryOp, Cwise::inverse()
-  */
-template<typename Scalar>
-struct ei_scalar_inverse_op {
-  EIGEN_EMPTY_STRUCT_CTOR(ei_scalar_inverse_op)
-  inline Scalar operator() (const Scalar& a) const { return Scalar(1)/a; }
-  template<typename Packet>
-  inline const Packet packetOp(const Packet& a) const
-  { return ei_pdiv(ei_pset1<Packet>(Scalar(1)),a); }
-};
-template<typename Scalar>
-struct ei_functor_traits<ei_scalar_inverse_op<Scalar> >
-{ enum { Cost = NumTraits<Scalar>::MulCost, PacketAccess = ei_packet_traits<Scalar>::HasDiv }; };
-
-/** \internal
-  * \brief Template functor to compute the square of a scalar
-  * \sa class CwiseUnaryOp, Cwise::square()
-  */
-template<typename Scalar>
-struct ei_scalar_square_op {
-  EIGEN_EMPTY_STRUCT_CTOR(ei_scalar_square_op)
-  inline Scalar operator() (const Scalar& a) const { return a*a; }
-  template<typename Packet>
-  inline const Packet packetOp(const Packet& 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 = ei_packet_traits<Scalar>::HasMul }; };
-
-/** \internal
-  * \brief Template functor to compute the cube of a scalar
-  * \sa class CwiseUnaryOp, Cwise::cube()
-  */
-template<typename Scalar>
-struct ei_scalar_cube_op {
-  EIGEN_EMPTY_STRUCT_CTOR(ei_scalar_cube_op)
-  inline Scalar operator() (const Scalar& a) const { return a*a*a; }
-  template<typename Packet>
-  inline const Packet packetOp(const Packet& 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 = ei_packet_traits<Scalar>::HasMul }; };
-
-// default 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
-
-// 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
-
-#endif // EIGEN_FUNCTORS_H

Eigen/src/Core/Fuzzy.h

@@ -4,31 +4,82 @@
 // Copyright (C) 2006-2008 Benoit Jacob <jacob.benoit.1@gmail.com>
 // Copyright (C) 2008 Gael Guennebaud <gael.guennebaud@inria.fr>
 //
-// Eigen is free software; you can redistribute it and/or
-// modify it under the terms of the GNU Lesser General Public
-// 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
 
-// TODO support small integer types properly i.e. do exact compare on coeffs --- taking a HS norm is guaranteed to cause integer overflow.
+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
 
-#ifndef EIGEN_LEGACY_COMPARES
 
 /** \returns \c true if \c *this is approximately equal to \a other, within the precision
   * determined by \a prec.
@@ -42,24 +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 DenseBase<Derived>::isApprox(
   const DenseBase<OtherDerived>& other,
-  RealScalar prec
+  const RealScalar& prec
 ) const
 {
-  const typename ei_nested<Derived,2>::type nested(derived());
-  const typename ei_nested<OtherDerived,2>::type otherNested(other.derived());
-//   std::cerr << typeid(Derived).name() << " => " << typeid(typename ei_nested<Derived,2>::type).name() << "\n";
-//   std::cerr << typeid(OtherDerived).name() << " => " << typeid(typename ei_nested<OtherDerived,2>::type).name() << "\n";
-//   return false;
-  return (nested - otherNested).cwiseAbs2().sum() <= prec * prec * std::min(nested.cwiseAbs2().sum(), otherNested.cwiseAbs2().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,
@@ -78,10 +124,10 @@ bool DenseBase<Derived>::isApprox(
 template<typename Derived>
 bool DenseBase<Derived>::isMuchSmallerThan(
   const typename NumTraits<Scalar>::Real& other,
-  RealScalar prec
+  const RealScalar& prec
 ) const
 {
-  return derived().cwiseAbs2().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,
@@ -98,143 +144,12 @@ template<typename Derived>
 template<typename OtherDerived>
 bool DenseBase<Derived>::isMuchSmallerThan(
   const DenseBase<OtherDerived>& other,
-  RealScalar prec
+  const RealScalar& prec
 ) const
 {
-  return derived().cwiseAbs2().sum() <= prec * prec * other.derived().cwiseAbs2().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 DenseBase<Derived>::isApprox(
-  const DenseBase<OtherDerived>& other,
-  RealScalar 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 DenseBase<OtherDerived>&, RealScalar) const
-  */
-template<typename Derived>
-bool DenseBase<Derived>::isMuchSmallerThan(
-  const typename NumTraits<Scalar>::Real& other,
-  RealScalar 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 DenseBase<Derived>::isMuchSmallerThan(
-  const DenseBase<OtherDerived>& other,
-  RealScalar 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;
-  typedef typename Derived::Index Index;
-  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(Index 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(Index 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(Index 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 +1,457 @@
+// 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 { is_large = MaxSize == Dynamic ||
+                    Size >= EIGEN_CACHEFRIENDLY_PRODUCT_THRESHOLD,
+         value = is_large  ? Large
+               : Size == 1 ? 1
+                           : Small
+  };
+};
+
+template<typename Lhs, typename Rhs> struct product_type
+{
+  typedef typename remove_all<Lhs>::type _Lhs;
+  typedef typename remove_all<Rhs>::type _Rhs;
+  enum {
+    MaxRows = 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
+};
+
+// template<typename Lhs, typename Rhs> struct product_tag
+// {
+// private:
+//   
+//   typedef typename remove_all<Lhs>::type _Lhs;
+//   typedef typename remove_all<Rhs>::type _Rhs;
+//   enum {
+//     Rows  = _Lhs::RowsAtCompileTime,
+//     Cols  = _Rhs::ColsAtCompileTime,
+//     Depth = EIGEN_SIZE_MIN_PREFER_FIXED(_Lhs::ColsAtCompileTime, _Rhs::RowsAtCompileTime)
+//   };
+// 
+//   enum {
+//     rows_select = Rows==1 ? int(Rows) : int(Large),
+//     cols_select = Cols==1 ? int(Cols) : int(Large),
+//     depth_select = Depth==1 ? int(Depth) : int(Large)
+//   };
+//   typedef product_type_selector<rows_select, cols_select, depth_select> selector;
+// 
+// public:
+//   enum {
+//     ret = selector::ret
+//   };
+// 
+// };
+
+/* The following allows to select the kind of product at compile time
+ * based on the three dimensions of the product.
+ * This is a compile time mapping from {1,Small,Large}^3 -> {product types} */
+// FIXME I'm not sure the current mapping is the ideal one.
+template<int M, int N>  struct product_type_selector<M,N,1>              { enum { ret = OuterProduct }; };
+template<int Depth>     struct product_type_selector<1,    1,    Depth>  { enum { ret = InnerProduct }; };
+template<>              struct product_type_selector<1,    1,    1>      { enum { ret = InnerProduct }; };
+template<>              struct product_type_selector<Small,1,    Small>  { enum { ret = CoeffBasedProductMode }; };
+template<>              struct product_type_selector<1,    Small,Small>  { enum { ret = CoeffBasedProductMode }; };
+template<>              struct product_type_selector<Small,Small,Small>  { enum { ret = CoeffBasedProductMode }; };
+template<>              struct product_type_selector<Small, Small, 1>    { enum { ret = LazyCoeffBasedProductMode }; };
+template<>              struct product_type_selector<Small, Large, 1>    { enum { ret = LazyCoeffBasedProductMode }; };
+template<>              struct product_type_selector<Large, Small, 1>    { enum { ret = LazyCoeffBasedProductMode }; };
+template<>              struct product_type_selector<1,    Large,Small>  { enum { ret = CoeffBasedProductMode }; };
+template<>              struct product_type_selector<1,    Large,Large>  { enum { ret = GemvProduct }; };
+template<>              struct product_type_selector<1,    Small,Large>  { enum { ret = CoeffBasedProductMode }; };
+template<>              struct product_type_selector<Large,1,    Small>  { enum { ret = CoeffBasedProductMode }; };
+template<>              struct product_type_selector<Large,1,    Large>  { enum { ret = GemvProduct }; };
+template<>              struct product_type_selector<Small,1,    Large>  { enum { ret = CoeffBasedProductMode }; };
+template<>              struct product_type_selector<Small,Small,Large>  { enum { ret = GemmProduct }; };
+template<>              struct product_type_selector<Large,Small,Large>  { enum { ret = GemmProduct }; };
+template<>              struct product_type_selector<Small,Large,Large>  { enum { ret = GemmProduct }; };
+template<>              struct product_type_selector<Large,Large,Large>  { enum { ret = GemmProduct }; };
+template<>              struct product_type_selector<Large,Small,Small>  { enum { ret = GemmProduct }; };
+template<>              struct product_type_selector<Small,Large,Small>  { enum { ret = GemmProduct }; };
+template<>              struct product_type_selector<Large,Large,Small>  { enum { ret = GemmProduct }; };
+
+} // end namespace internal
+
+/***********************************************************************
+*  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>
+{
+  #if EIGEN_MAX_STATIC_ALIGN_BYTES!=0
+  internal::plain_array<Scalar,EIGEN_SIZE_MIN_PREFER_FIXED(Size,MaxSize),0> m_data;
+  EIGEN_STRONG_INLINE Scalar* data() { return m_data.array; }
+  #else
+  // Some architectures cannot align on the stack,
+  // => let's manually enforce alignment by allocating more data and return the address of the first aligned element.
+  enum {
+    ForceAlignment  = internal::packet_traits<Scalar>::Vectorizable,
+    PacketSize      = internal::packet_traits<Scalar>::size
+  };
+  internal::plain_array<Scalar,EIGEN_SIZE_MIN_PREFER_FIXED(Size,MaxSize)+(ForceAlignment?PacketSize:0),0> m_data;
+  EIGEN_STRONG_INLINE Scalar* data() {
+    return ForceAlignment
+            ? reinterpret_cast<Scalar*>((reinterpret_cast<size_t>(m_data.array) & ~(size_t(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>, Aligned> MappedDest;
+
+    ActualLhsType actualLhs = LhsBlasTraits::extract(lhs);
+    ActualRhsType 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...
+      EvalToDestAtCompileTime = Dest::InnerStrideAtCompileTime==1,
+      ComplexByReal = (NumTraits<LhsScalar>::IsComplex) && (!NumTraits<RhsScalar>::IsComplex),
+      MightCannotUseDest = (Dest::InnerStrideAtCompileTime!=1) || ComplexByReal
+    };
+
+    gemv_static_vector_if<ResScalar,Dest::SizeAtCompileTime,Dest::MaxSizeAtCompileTime,MightCannotUseDest> static_dest;
+
+    const bool alphaIsCompatible = (!ComplexByReal) || (numext::imag(actualAlpha)==RealScalar(0));
+    const bool evalToDest = EvalToDestAtCompileTime && alphaIsCompatible;
+
+    RhsScalar compatibleAlpha = get_factor<ResScalar,RhsScalar>::run(actualAlpha);
+
+    ei_declare_aligned_stack_constructed_variable(ResScalar,actualDestPtr,dest.size(),
+                                                  evalToDest ? dest.data() : static_dest.data());
+
+    if(!evalToDest)
+    {
+      #ifdef EIGEN_DENSE_STORAGE_CTOR_PLUGIN
+      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;
+    }
+
+    typedef const_blas_data_mapper<LhsScalar,Index,ColMajor> LhsMapper;
+    typedef const_blas_data_mapper<RhsScalar,Index,RowMajor> RhsMapper;
+    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 += 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.innerStride(),
+        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)
+  {
+    // 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)
+  {
+    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*()
+  */
+#ifndef __CUDACC__
+
+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());
+}
+
+#endif // __CUDACC__
+
+/** \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

@@ -4,28 +4,17 @@
 // 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
   *
@@ -50,24 +39,30 @@
 #define EIGEN_DEBUG_UNALIGNED_STORE
 #endif
 
-struct ei_default_packet_traits
+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,
+    HasLog10    = 0,
     HasPow    = 0,
 
     HasSin    = 0,
@@ -75,17 +70,31 @@ struct ei_default_packet_traits
     HasTan    = 0,
     HasASin   = 0,
     HasACos   = 0,
-    HasATan   = 0
+    HasATan   = 0,
+    HasSinh    = 0,
+    HasCosh    = 0,
+    HasTanh    = 0,
+    HasLGamma = 0,
+    HasErf = 0,
+    HasErfc = 0,
+
+    HasRound  = 0,
+    HasFloor  = 0,
+    HasCeil   = 0,
+
+    HasSign   = 0
   };
 };
 
-template<typename T> struct ei_packet_traits : ei_default_packet_traits
+template<typename T> struct packet_traits : default_packet_traits
 {
   typedef T type;
+  typedef T half;
   enum {
     Vectorizable = 0,
     size = 1,
-    AlignedOnScalar = 0
+    AlignedOnScalar = 0,
+    HasHalfPacket = 0
   };
   enum {
     HasAdd    = 0,
@@ -101,210 +110,468 @@ template<typename T> struct ei_packet_traits : ei_default_packet_traits
   };
 };
 
+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);
+}
+
+
 /** \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> inline Packet
-ei_pnegate(const Packet& a) { return -a; }
+template<typename Packet> EIGEN_DEVICE_FUNC inline Packet
+pnegate(const Packet& a) { return -a; }
 
 /** \internal \returns conj(a) (coeff-wise) */
-template<typename Packet> inline Packet
-ei_pconj(const Packet& a) { return ei_conj(a); }
+
+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> inline Packet
-ei_pabs(const Packet& a) { return ei_abs(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> inline Packet
-ei_pand(const Packet& a, const Packet& b) { return 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> inline Packet
-ei_por(const Packet& a, const Packet& b) { return 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> inline Packet
-ei_pxor(const Packet& a, const Packet& b) { return 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> inline Packet
-ei_pandnot(const Packet& a, const Packet& b) { return 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 Packet> inline Packet
-ei_pload(const typename ei_unpacket_traits<Packet>::type* 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 Packet> inline Packet
-ei_ploadu(const typename ei_unpacket_traits<Packet>::type* from) { return *from; }
-
-/** \internal \returns a packet with elements of \a *from duplicated, e.g.: (from[0],from[0],from[1],from[1]) */
-template<typename Packet> inline Packet
-ei_ploaddup(const typename ei_unpacket_traits<Packet>::type* 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 Packet> inline Packet
-ei_pset1(const typename ei_unpacket_traits<Packet>::type& 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 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 Scalar> inline typename ei_packet_traits<Scalar>::type
-ei_plset(const Scalar& a) { return a; }
+template<typename Packet> 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> inline void ei_prefetch(const Scalar* addr)
+template<typename Scalar> inline void prefetch(const Scalar* addr)
 {
-#if !defined(_MSC_VER)
-__builtin_prefetch(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
+  __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
+predux4(const Packet& a)
 { return a; }
 
 /** \internal \returns the product of the elements of \a a*/
-template<typename Packet> inline typename ei_unpacket_traits<Packet>::type ei_predux_mul(const Packet& 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> inline typename ei_unpacket_traits<Packet>::type ei_predux_min(const Packet& 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> inline typename ei_unpacket_traits<Packet>::type ei_predux_max(const Packet& 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> inline Packet ei_preverse(const Packet& a)
+template<typename Packet> EIGEN_DEVICE_FUNC inline Packet preverse(const Packet& a)
 { return a; }
 
+template<size_t offset, typename Packet>
+struct protate_impl
+{
+  // Empty so attempts to use this unimplemented path will fail to compile.
+  // Only specializations of this template should be used.
+};
+
+/** \internal \returns a packet with the coefficients rotated to the right in little-endian convention,
+  * by the given offset, e.g. for offset == 1:
+  *     (packet[3], packet[2], packet[1], packet[0]) becomes (packet[0], packet[3], packet[2], packet[1])
+  */
+template<size_t offset, typename Packet> EIGEN_DEVICE_FUNC inline Packet protate(const Packet& a)
+{
+  return offset ? protate_impl<offset, Packet>::run(a) : 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)
+{
+  // FIXME: uncomment the following in case we drop the internal imag and real functions.
+//   using std::imag;
+//   using std::real;
+  return Packet(imag(a),real(a));
+}
+
 /**************************
 * Special math functions
 ***************************/
 
-/** \internal \returns the sin of \a a (coeff-wise) */
+/** \internal \returns the sine of \a a (coeff-wise) */
 template<typename Packet> EIGEN_DECLARE_FUNCTION_ALLOWING_MULTIPLE_DEFINITIONS
-Packet ei_psin(const Packet& a) { return ei_sin(a); }
+Packet psin(const Packet& a) { using std::sin; return sin(a); }
 
-/** \internal \returns the cos of \a a (coeff-wise) */
+/** \internal \returns the cosine of \a a (coeff-wise) */
 template<typename Packet> EIGEN_DECLARE_FUNCTION_ALLOWING_MULTIPLE_DEFINITIONS
-Packet ei_pcos(const Packet& a) { return ei_cos(a); }
+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 ei_pexp(const Packet& a) { return ei_exp(a); }
+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 ei_plog(const Packet& a) { return ei_log(a); }
+Packet plog(const Packet& a) { using std::log; return log(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 ei_psqrt(const Packet& a) { return ei_sqrt(a); }
+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); }
+
+/** \internal \returns the ln(|gamma(\a a)|) (coeff-wise) */
+template<typename Packet> EIGEN_DECLARE_FUNCTION_ALLOWING_MULTIPLE_DEFINITIONS
+Packet plgamma(const Packet& a) { using numext::lgamma; return lgamma(a); }
+
+/** \internal \returns the erf(\a a) (coeff-wise) */
+template<typename Packet> EIGEN_DECLARE_FUNCTION_ALLOWING_MULTIPLE_DEFINITIONS
+Packet perf(const Packet& a) { using numext::erf; return erf(a); }
+
+/** \internal \returns the erfc(\a a) (coeff-wise) */
+template<typename Packet> EIGEN_DECLARE_FUNCTION_ALLOWING_MULTIPLE_DEFINITIONS
+Packet perfc(const Packet& a) { using numext::erfc; return erfc(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 Packet, int LoadMode>
-inline Packet ei_ploadt(const typename ei_unpacket_traits<Packet>::type* 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<Packet>(from);
+  if(Alignment >= unpacket_traits<Packet>::alignment)
+    return pload<Packet>(from);
   else
-    return ei_ploadu<Packet>(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>
+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);
 }
 
 /***************************************************************************
 * Fast complex products (GCC generates a function call which is very slow)
 ***************************************************************************/
 
-template<> inline std::complex<float> ei_pmul(const std::complex<float>& a, const std::complex<float>& b)
-{ return std::complex<float>(ei_real(a)*ei_real(b) - ei_imag(a)*ei_imag(b), ei_imag(a)*ei_real(b) + ei_real(a)*ei_imag(b)); }
+// Eigen+CUDA does not support complexes.
+#ifndef __CUDACC__
 
-template<> inline std::complex<double> ei_pmul(const std::complex<double>& a, const std::complex<double>& b)
-{ return std::complex<double>(ei_real(a)*ei_real(b) - ei_imag(a)*ei_imag(b), ei_imag(a)*ei_real(b) + ei_real(a)*ei_imag(b)); }
+template<> inline std::complex<float> pmul(const std::complex<float>& a, const std::complex<float>& b)
+{ return std::complex<float>(real(a)*real(b) - imag(a)*imag(b), imag(a)*real(b) + real(a)*imag(b)); }
+
+template<> inline std::complex<double> pmul(const std::complex<double>& a, const std::complex<double>& b)
+{ return std::complex<double>(real(a)*real(b) - imag(a)*imag(b), imag(a)*real(b) + real(a)*imag(b)); }
+
+#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;
+}
+
+} // end namespace internal
+
+} // end namespace Eigen
 
 #endif // EIGEN_GENERIC_PACKET_MATH_H
-

Eigen/src/Core/GlobalFunctions.h

@@ -1,36 +1,21 @@
 // This file is part of Eigen, a lightweight C++ template library
 // for linear algebra.
 //
-// Copyright (C) 2010 Gael Guennebaud <gael.guennebaud@inria.fr>
+// Copyright (C) 2010-2012 Gael Guennebaud <gael.guennebaud@inria.fr>
 // Copyright (C) 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_GLOBAL_FUNCTIONS_H
 #define EIGEN_GLOBAL_FUNCTIONS_H
 
-#define EIGEN_ARRAY_DECLARE_GLOBAL_STD_UNARY(NAME,FUNCTOR) \
+#define EIGEN_ARRAY_DECLARE_GLOBAL_UNARY(NAME,FUNCTOR) \
   template<typename Derived> \
-  inline const Eigen::CwiseUnaryOp<Eigen::FUNCTOR<typename Derived::Scalar>, Derived> \
-  NAME(const Eigen::ArrayBase<Derived>& x) { \
-    return x.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()); \
   }
 
 #define EIGEN_ARRAY_DECLARE_GLOBAL_EIGEN_UNARY(NAME,FUNCTOR) \
@@ -38,49 +23,119 @@
   template<typename Derived> \
   struct NAME##_retval<ArrayBase<Derived> > \
   { \
-    typedef const Eigen::CwiseUnaryOp<Eigen::FUNCTOR<typename Derived::Scalar>, Derived> type; \
+    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 x.derived(); \
+      return typename NAME##_retval<ArrayBase<Derived> >::type(x.derived()); \
     } \
   };
 
-
-namespace std
+namespace Eigen
 {
-  EIGEN_ARRAY_DECLARE_GLOBAL_STD_UNARY(real,ei_scalar_real_op)
-  EIGEN_ARRAY_DECLARE_GLOBAL_STD_UNARY(imag,ei_scalar_imag_op)
-  EIGEN_ARRAY_DECLARE_GLOBAL_STD_UNARY(sin,ei_scalar_sin_op)
-  EIGEN_ARRAY_DECLARE_GLOBAL_STD_UNARY(cos,ei_scalar_cos_op)
-  EIGEN_ARRAY_DECLARE_GLOBAL_STD_UNARY(exp,ei_scalar_exp_op)
-  EIGEN_ARRAY_DECLARE_GLOBAL_STD_UNARY(log,ei_scalar_log_op)
-  EIGEN_ARRAY_DECLARE_GLOBAL_STD_UNARY(abs,ei_scalar_abs_op)
-  EIGEN_ARRAY_DECLARE_GLOBAL_STD_UNARY(sqrt,ei_scalar_sqrt_op)
-
+  EIGEN_ARRAY_DECLARE_GLOBAL_UNARY(real,scalar_real_op)
+  EIGEN_ARRAY_DECLARE_GLOBAL_UNARY(imag,scalar_imag_op)
+  EIGEN_ARRAY_DECLARE_GLOBAL_UNARY(conj,scalar_conjugate_op)
+  EIGEN_ARRAY_DECLARE_GLOBAL_UNARY(inverse,scalar_inverse_op)
+  EIGEN_ARRAY_DECLARE_GLOBAL_UNARY(sin,scalar_sin_op)
+  EIGEN_ARRAY_DECLARE_GLOBAL_UNARY(cos,scalar_cos_op)
+  EIGEN_ARRAY_DECLARE_GLOBAL_UNARY(tan,scalar_tan_op)
+  EIGEN_ARRAY_DECLARE_GLOBAL_UNARY(atan,scalar_atan_op)
+  EIGEN_ARRAY_DECLARE_GLOBAL_UNARY(asin,scalar_asin_op)
+  EIGEN_ARRAY_DECLARE_GLOBAL_UNARY(acos,scalar_acos_op)
+  EIGEN_ARRAY_DECLARE_GLOBAL_UNARY(sinh,scalar_sinh_op)
+  EIGEN_ARRAY_DECLARE_GLOBAL_UNARY(cosh,scalar_cosh_op)
+  EIGEN_ARRAY_DECLARE_GLOBAL_UNARY(tanh,scalar_tanh_op)
+  EIGEN_ARRAY_DECLARE_GLOBAL_UNARY(lgamma,scalar_lgamma_op)
+  EIGEN_ARRAY_DECLARE_GLOBAL_UNARY(erf,scalar_erf_op)
+  EIGEN_ARRAY_DECLARE_GLOBAL_UNARY(erfc,scalar_erfc_op)
+  EIGEN_ARRAY_DECLARE_GLOBAL_UNARY(exp,scalar_exp_op)
+  EIGEN_ARRAY_DECLARE_GLOBAL_UNARY(log,scalar_log_op)
+  EIGEN_ARRAY_DECLARE_GLOBAL_UNARY(log10,scalar_log10_op)
+  EIGEN_ARRAY_DECLARE_GLOBAL_UNARY(abs,scalar_abs_op)
+  EIGEN_ARRAY_DECLARE_GLOBAL_UNARY(abs2,scalar_abs2_op)
+  EIGEN_ARRAY_DECLARE_GLOBAL_UNARY(arg,scalar_arg_op)
+  EIGEN_ARRAY_DECLARE_GLOBAL_UNARY(sqrt,scalar_sqrt_op)
+  EIGEN_ARRAY_DECLARE_GLOBAL_UNARY(square,scalar_square_op)
+  EIGEN_ARRAY_DECLARE_GLOBAL_UNARY(cube,scalar_cube_op)
+  EIGEN_ARRAY_DECLARE_GLOBAL_UNARY(round,scalar_round_op)
+  EIGEN_ARRAY_DECLARE_GLOBAL_UNARY(floor,scalar_floor_op)
+  EIGEN_ARRAY_DECLARE_GLOBAL_UNARY(ceil,scalar_ceil_op)
+  EIGEN_ARRAY_DECLARE_GLOBAL_UNARY(isnan,scalar_isnan_op)
+  EIGEN_ARRAY_DECLARE_GLOBAL_UNARY(isinf,scalar_isinf_op)
+  EIGEN_ARRAY_DECLARE_GLOBAL_UNARY(isfinite,scalar_isfinite_op)
+  EIGEN_ARRAY_DECLARE_GLOBAL_UNARY(sign,scalar_sign_op)
+  
   template<typename Derived>
-  inline const Eigen::CwiseUnaryOp<Eigen::ei_scalar_pow_op<typename Derived::Scalar>, Derived>
-  pow(const Eigen::ArrayBase<Derived>& x, const typename Derived::Scalar& exponent) { \
-    return x.derived().pow(exponent); \
+  inline const Eigen::CwiseUnaryOp<Eigen::internal::scalar_pow_op<typename Derived::Scalar>, const Derived>
+  pow(const Eigen::ArrayBase<Derived>& x, const typename Derived::Scalar& exponent) {
+    return x.derived().pow(exponent);
+  }
+
+  /** \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()
+    */
+  template<typename Derived,typename ExponentDerived>
+  inline const Eigen::CwiseBinaryOp<Eigen::internal::scalar_binary_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_binary_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.
+    * Beaware that the scalar type of the input scalar \a x and the exponents \a exponents must be the same.
+    *
+    * Example: \include Cwise_scalar_power_array.cpp
+    * Output: \verbinclude Cwise_scalar_power_array.out
+    * 
+    * \sa ArrayBase::pow()
+    */
+  template<typename Derived>
+  inline const Eigen::CwiseBinaryOp<Eigen::internal::scalar_binary_pow_op<typename Derived::Scalar, typename Derived::Scalar>, const typename Derived::ConstantReturnType, const Derived>
+  pow(const typename Derived::Scalar& x, const Eigen::ArrayBase<Derived>& exponents) 
+  {
+    typename Derived::ConstantReturnType constant_x(exponents.rows(), exponents.cols(), x);
+    return Eigen::CwiseBinaryOp<Eigen::internal::scalar_binary_pow_op<typename Derived::Scalar, typename Derived::Scalar>, const typename Derived::ConstantReturnType, const Derived>(
+      constant_x,
+      exponents.derived()
+    );
+  }
+  
+  /**
+  * \brief Component-wise division of a scalar by array elements.
+  **/
+  template <typename Derived>
+  inline const Eigen::CwiseUnaryOp<Eigen::internal::scalar_inverse_mult_op<typename Derived::Scalar>, const Derived>
+    operator/(const typename Derived::Scalar& s, const Eigen::ArrayBase<Derived>& a)
+  {
+    return Eigen::CwiseUnaryOp<Eigen::internal::scalar_inverse_mult_op<typename Derived::Scalar>, const Derived>(
+      a.derived(),
+      Eigen::internal::scalar_inverse_mult_op<typename Derived::Scalar>(s)  
+    );
+  }
+
+  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)
   }
 }
 
-namespace Eigen
-{
-  EIGEN_ARRAY_DECLARE_GLOBAL_EIGEN_UNARY(ei_real,ei_scalar_real_op)
-  EIGEN_ARRAY_DECLARE_GLOBAL_EIGEN_UNARY(ei_imag,ei_scalar_imag_op)
-  EIGEN_ARRAY_DECLARE_GLOBAL_EIGEN_UNARY(ei_sin,ei_scalar_sin_op)
-  EIGEN_ARRAY_DECLARE_GLOBAL_EIGEN_UNARY(ei_cos,ei_scalar_cos_op)
-  EIGEN_ARRAY_DECLARE_GLOBAL_EIGEN_UNARY(ei_exp,ei_scalar_exp_op)
-  EIGEN_ARRAY_DECLARE_GLOBAL_EIGEN_UNARY(ei_log,ei_scalar_log_op)
-  EIGEN_ARRAY_DECLARE_GLOBAL_EIGEN_UNARY(ei_abs,ei_scalar_abs_op)
-  EIGEN_ARRAY_DECLARE_GLOBAL_EIGEN_UNARY(ei_abs2,ei_scalar_abs2_op)
-  EIGEN_ARRAY_DECLARE_GLOBAL_EIGEN_UNARY(ei_sqrt,ei_scalar_sqrt_op)
-}
-
-// TODO: cleanly disable those functions that are not supported on Array (ei_real_ref, ei_random, ei_isApprox...)
+// TODO: cleanly disable those functions that are not supported on Array (numext::real_ref, internal::random, internal::isApprox...)
 
 #endif // EIGEN_GLOBAL_FUNCTIONS_H

Eigen/src/Core/IO.h

@@ -4,32 +4,24 @@
 // Copyright (C) 2006-2008 Benoit Jacob <jacob.benoit.1@gmail.com>
 // Copyright (C) 2008 Gael Guennebaud <gael.guennebaud@inria.fr>
 //
-// Eigen is free software; you can redistribute it and/or
-// modify it under the terms of the GNU Lesser General Public
-// 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_IO_H
 #define EIGEN_IO_H
 
+namespace Eigen { 
+
 enum { DontAlignCols = 1 };
 enum { StreamPrecision = -1,
        FullPrecision = -2 };
 
+namespace internal {
+template<typename Derived>
+std::ostream & print_matrix(std::ostream & s, const Derived& _m, const IOFormat& fmt);
+}
+
 /** \class IOFormat
   * \ingroup Core_Module
   *
@@ -57,15 +49,18 @@ enum { StreamPrecision = -1,
   */
 struct IOFormat
 {
-  /** Default contructor, see class IOFormat for the meaning of the parameters */
+  /** Default constructor, see class IOFormat for the meaning of the parameters */
   IOFormat(int _precision = StreamPrecision, int _flags = 0,
     const std::string& _coeffSeparator = " ",
     const std::string& _rowSeparator = "\n", const std::string& _rowPrefix="", const std::string& _rowSuffix="",
     const std::string& _matPrefix="", const std::string& _matSuffix="")
   : matPrefix(_matPrefix), matSuffix(_matSuffix), rowPrefix(_rowPrefix), rowSuffix(_rowSuffix), rowSeparator(_rowSeparator),
-    coeffSeparator(_coeffSeparator), precision(_precision), flags(_flags)
+    rowSpacer(""), coeffSeparator(_coeffSeparator), precision(_precision), flags(_flags)
   {
-    rowSpacer = "";
+    // TODO check if rowPrefix, rowSuffix or rowSeparator contains a newline
+    // don't add rowSpacer if columns are not to be aligned
+    if((flags & DontAlignCols))
+      return;
     int i = int(matSuffix.length())-1;
     while (i>=0 && matSuffix[i]!='\n')
     {
@@ -106,7 +101,7 @@ class WithFormat
 
     friend std::ostream & operator << (std::ostream & s, const WithFormat& wf)
     {
-      return ei_print_matrix(s, wf.m_matrix.eval(), wf.m_format);
+      return internal::print_matrix(s, wf.m_matrix.eval(), wf.m_format);
     }
 
   protected:
@@ -128,18 +123,22 @@ DenseBase<Derived>::format(const IOFormat& fmt) const
   return WithFormat<Derived>(derived(), fmt);
 }
 
+namespace internal {
+
 template<typename Scalar, bool IsInteger>
-struct ei_significant_decimals_default_impl
+struct significant_decimals_default_impl
 {
   typedef typename NumTraits<Scalar>::Real RealScalar;
   static inline int run()
   {
-    return ei_cast<RealScalar,int>(std::ceil(-ei_log(NumTraits<RealScalar>::epsilon())/ei_log(RealScalar(10))));
+    using std::ceil;
+    using std::log;
+    return cast<RealScalar,int>(ceil(-log(NumTraits<RealScalar>::epsilon())/log(RealScalar(10))));
   }
 };
 
 template<typename Scalar>
-struct ei_significant_decimals_default_impl<Scalar, true>
+struct significant_decimals_default_impl<Scalar, true>
 {
   static inline int run()
   {
@@ -148,14 +147,14 @@ struct ei_significant_decimals_default_impl<Scalar, true>
 };
 
 template<typename Scalar>
-struct ei_significant_decimals_impl
-  : ei_significant_decimals_default_impl<Scalar, NumTraits<Scalar>::IsInteger>
+struct significant_decimals_impl
+  : significant_decimals_default_impl<Scalar, NumTraits<Scalar>::IsInteger>
 {};
 
 /** \internal
   * print the matrix \a _m to the output stream \a s using the output format \a fmt */
 template<typename Derived>
-std::ostream & ei_print_matrix(std::ostream & s, const Derived& _m, const IOFormat& fmt)
+std::ostream & print_matrix(std::ostream & s, const Derived& _m, const IOFormat& fmt)
 {
   if(_m.size() == 0)
   {
@@ -163,9 +162,8 @@ std::ostream & ei_print_matrix(std::ostream & s, const Derived& _m, const IOForm
     return s;
   }
   
-  const typename Derived::Nested m = _m;
+  typename Derived::Nested m = _m;
   typedef typename Derived::Scalar Scalar;
-  typedef typename Derived::Index Index;
 
   Index width = 0;
 
@@ -182,7 +180,7 @@ std::ostream & ei_print_matrix(std::ostream & s, const Derived& _m, const IOForm
     }
     else
     {
-      explicit_precision = ei_significant_decimals_impl<Scalar>::run();
+      explicit_precision = significant_decimals_impl<Scalar>::run();
     }
   }
   else
@@ -190,21 +188,22 @@ std::ostream & ei_print_matrix(std::ostream & s, const Derived& _m, const IOForm
     explicit_precision = fmt.precision;
   }
 
+  std::streamsize old_precision = 0;
+  if(explicit_precision) old_precision = s.precision(explicit_precision);
+
   bool align_cols = !(fmt.flags & DontAlignCols);
   if(align_cols)
   {
     // compute the largest width
-    for(Index j = 1; j < m.cols(); ++j)
+    for(Index j = 0; j < m.cols(); ++j)
       for(Index i = 0; i < m.rows(); ++i)
       {
         std::stringstream sstr;
-        if(explicit_precision) sstr.precision(explicit_precision);
+        sstr.copyfmt(s);
         sstr << m.coeff(i,j);
         width = std::max<Index>(width, Index(sstr.str().length()));
       }
   }
-  std::streamsize old_precision = 0;
-  if(explicit_precision) old_precision = s.precision(explicit_precision);
   s << fmt.matPrefix;
   for(Index i = 0; i < m.rows(); ++i)
   {
@@ -228,6 +227,8 @@ std::ostream & ei_print_matrix(std::ostream & s, const Derived& _m, const IOForm
   return s;
 }
 
+} // end namespace internal
+
 /** \relates DenseBase
   *
   * Outputs the matrix, to the given stream.
@@ -244,7 +245,9 @@ std::ostream & operator <<
 (std::ostream & s,
  const DenseBase<Derived> & m)
 {
-  return ei_print_matrix(s, m.eval(), EIGEN_DEFAULT_IO_FORMAT);
+  return internal::print_matrix(s, m.eval(), EIGEN_DEFAULT_IO_FORMAT);
 }
 
+} // end namespace Eigen
+
 #endif // EIGEN_IO_H

Eigen/src/Core/Inverse.h

@@ -0,0 +1,117 @@
+// This file is part of Eigen, a lightweight C++ template library
+// for linear algebra.
+//
+// Copyright (C) 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_INVERSE_H
+#define EIGEN_INVERSE_H
+
+namespace Eigen { 
+
+template<typename XprType,typename StorageKind> class InverseImpl;
+
+namespace internal {
+
+template<typename XprType>
+struct traits<Inverse<XprType> >
+  : traits<typename XprType::PlainObject>
+{
+  typedef typename XprType::PlainObject PlainObject;
+  typedef traits<PlainObject> BaseTraits;
+  enum {
+    Flags = BaseTraits::Flags & RowMajorBit
+  };
+};
+
+} // end namespace internal
+
+/** \class Inverse
+  *
+  * \brief Expression of the inverse of another expression
+  *
+  * \tparam XprType the type of the expression we are taking the inverse
+  *
+  * This class represents an abstract expression of A.inverse()
+  * and most of the time this is the only way it is used.
+  *
+  */
+template<typename XprType>
+class Inverse : public InverseImpl<XprType,typename internal::traits<XprType>::StorageKind>
+{
+public:
+  typedef typename XprType::StorageIndex StorageIndex;
+  typedef typename XprType::PlainObject                       PlainObject;
+  typedef typename internal::ref_selector<XprType>::type      XprTypeNested;
+  typedef typename internal::remove_all<XprTypeNested>::type  XprTypeNestedCleaned;
+  typedef typename internal::ref_selector<Inverse>::type Nested;
+  typedef typename internal::remove_all<XprType>::type NestedExpression;
+  
+  explicit Inverse(const XprType &xpr)
+    : m_xpr(xpr)
+  {}
+
+  EIGEN_DEVICE_FUNC Index rows() const { return m_xpr.rows(); }
+  EIGEN_DEVICE_FUNC Index cols() const { return m_xpr.cols(); }
+
+  EIGEN_DEVICE_FUNC const XprTypeNestedCleaned& nestedExpression() const { return m_xpr; }
+
+protected:
+  XprTypeNested m_xpr;
+};
+
+// Generic API dispatcher
+template<typename XprType, typename StorageKind>
+class InverseImpl
+  : public internal::generic_xpr_base<Inverse<XprType> >::type
+{
+public:
+  typedef typename internal::generic_xpr_base<Inverse<XprType> >::type Base;
+  typedef typename XprType::Scalar Scalar;
+private:
+
+  Scalar coeff(Index row, Index col) const;
+  Scalar coeff(Index i) const;
+};
+
+namespace internal {
+
+/** \internal
+  * \brief Default evaluator for Inverse expression.
+  * 
+  * This default evaluator for Inverse expression simply evaluate the inverse into a temporary
+  * by a call to internal::call_assignment_no_alias.
+  * Therefore, inverse implementers only have to specialize Assignment<Dst,Inverse<...>, ...> for
+  * there own nested expression.
+  *
+  * \sa class Inverse
+  */
+template<typename ArgType>
+struct unary_evaluator<Inverse<ArgType> >
+  : public evaluator<typename Inverse<ArgType>::PlainObject>
+{
+  typedef Inverse<ArgType> InverseType;
+  typedef typename InverseType::PlainObject PlainObject;
+  typedef evaluator<PlainObject> Base;
+  
+  enum { Flags = Base::Flags | EvalBeforeNestingBit };
+
+  unary_evaluator(const InverseType& inv_xpr)
+    : m_result(inv_xpr.rows(), inv_xpr.cols())
+  {
+    ::new (static_cast<Base*>(this)) Base(m_result);
+    internal::call_assignment_no_alias(m_result, inv_xpr);
+  }
+  
+protected:
+  PlainObject m_result;
+};
+  
+} // end namespace internal
+
+} // end namespace Eigen
+
+#endif // EIGEN_INVERSE_H

Eigen/src/Core/Map.h

@@ -4,37 +4,24 @@
 // Copyright (C) 2007-2010 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_MAP_H
 #define EIGEN_MAP_H
 
+namespace Eigen { 
+
 /** \class Map
   * \ingroup Core_Module
   *
   * \brief A matrix or vector expression mapping an existing array of data.
   *
-  * \param PlainObjectType the equivalent matrix type of the mapped data
-  * \param MapOptions specifies whether the pointer is \c Aligned, or \c Unaligned.
-  *                The default is \c Unaligned.
-  * \param StrideType optionnally specifies strides. By default, Map assumes the memory layout
+  * \tparam PlainObjectType the equivalent matrix type of the mapped data
+  * \tparam MapOptions specifies the pointer alignment in bytes. It can be: \c #Aligned128, , \c #Aligned64, \c #Aligned32, \c #Aligned16, \c #Aligned8 or \c #Unaligned.
+  *                The default is \c #Unaligned.
+  * \tparam StrideType optionally specifies strides. By default, Map assumes the memory layout
   *                   of an ordinary, contiguous array. This can be overridden by specifying strides.
   *                   The type passed here must be a specialization of the Stride template, see examples below.
   *
@@ -44,7 +31,7 @@
   * data is laid out contiguously in memory. You can however override this by explicitly specifying
   * inner and outer strides.
   *
-  * Here's an example of simply mapping a contiguous array as a column-major matrix:
+  * Here's an example of simply mapping a contiguous array as a \ref TopicStorageOrders "column-major" matrix:
   * \include Map_simple.cpp
   * Output: \verbinclude Map_simple.out
   *
@@ -72,16 +59,17 @@
   * Example: \include Map_placement_new.cpp
   * Output: \verbinclude Map_placement_new.out
   *
-  * This class is the return type of Matrix::Map() but can also be used directly.
+  * This class is the return type of PlainObjectBase::Map() but can also be used directly.
   *
-  * \sa Matrix::Map()
+  * \sa PlainObjectBase::Map(), \ref TopicStorageOrders
   */
+
+namespace internal {
 template<typename PlainObjectType, int MapOptions, typename StrideType>
-struct ei_traits<Map<PlainObjectType, MapOptions, StrideType> >
-  : public ei_traits<PlainObjectType>
+struct traits<Map<PlainObjectType, MapOptions, StrideType> >
+  : public traits<PlainObjectType>
 {
-  typedef typename PlainObjectType::Index Index;
-  typedef typename PlainObjectType::Scalar Scalar;
+  typedef traits<PlainObjectType> TraitsBase;
   enum {
     InnerStrideAtCompileTime = StrideType::InnerStrideAtCompileTime == 0
                              ? int(PlainObjectType::InnerStrideAtCompileTime)
@@ -89,24 +77,14 @@ struct ei_traits<Map<PlainObjectType, MapOptions, StrideType> >
     OuterStrideAtCompileTime = StrideType::OuterStrideAtCompileTime == 0
                              ? int(PlainObjectType::OuterStrideAtCompileTime)
                              : int(StrideType::OuterStrideAtCompileTime),
-    HasNoInnerStride = InnerStrideAtCompileTime == 1,
-    HasNoOuterStride = StrideType::OuterStrideAtCompileTime == 0,
-    HasNoStride = HasNoInnerStride && HasNoOuterStride,
-    IsAligned = int(int(MapOptions)&Aligned)==Aligned,
-    IsDynamicSize = PlainObjectType::SizeAtCompileTime==Dynamic,
-    KeepsPacketAccess = bool(HasNoInnerStride)
-                        && ( bool(IsDynamicSize)
-                           || HasNoOuterStride
-                           || ( OuterStrideAtCompileTime!=Dynamic
-                           && ((static_cast<int>(sizeof(Scalar))*OuterStrideAtCompileTime)%16)==0 ) ),
-    Flags0 = ei_traits<PlainObjectType>::Flags,
-    Flags1 = IsAligned ? (int(Flags0) | AlignedBit) : (int(Flags0) & ~AlignedBit),
-    Flags2 = HasNoStride ? int(Flags1) : int(Flags1 & ~LinearAccessBit),
-    Flags = KeepsPacketAccess ? int(Flags2) : (int(Flags2) & ~PacketAccessBit)
+    Alignment = int(MapOptions)&int(AlignedMask),
+    Flags0 = TraitsBase::Flags & (~NestByRefBit),
+    Flags = is_lvalue<PlainObjectType>::value ? int(Flags0) : (int(Flags0) & ~LvalueBit)
   };
 private:
-  enum { Options }; // Expressions don't support Options
+  enum { Options }; // Expressions don't have Options
 };
+}
 
 template<typename PlainObjectType, int MapOptions, typename StrideType> class Map
   : public MapBase<Map<PlainObjectType, MapOptions, StrideType> >
@@ -114,14 +92,20 @@ template<typename PlainObjectType, int MapOptions, typename StrideType> class Ma
   public:
 
     typedef MapBase<Map> Base;
-
     EIGEN_DENSE_PUBLIC_INTERFACE(Map)
 
+    typedef typename Base::PointerType PointerType;
+    typedef PointerType PointerArgType;
+    EIGEN_DEVICE_FUNC
+    inline PointerType cast_to_pointer_type(PointerArgType ptr) { return ptr; }
+
+    EIGEN_DEVICE_FUNC
     inline Index innerStride() const
     {
       return StrideType::InnerStrideAtCompileTime != 0 ? m_stride.inner() : 1;
     }
 
+    EIGEN_DEVICE_FUNC
     inline Index outerStride() const
     {
       return StrideType::OuterStrideAtCompileTime != 0 ? m_stride.outer()
@@ -132,52 +116,50 @@ template<typename PlainObjectType, int MapOptions, typename StrideType> class Ma
 
     /** Constructor in the fixed-size case.
       *
-      * \param data pointer to the array to map
+      * \param dataPtr pointer to the array to map
       * \param stride optional Stride object, passing the strides.
       */
-    inline Map(const Scalar* data, const StrideType& stride = StrideType())
-      : Base(data), m_stride(stride)
+    EIGEN_DEVICE_FUNC
+    explicit inline Map(PointerArgType dataPtr, const StrideType& stride = StrideType())
+      : Base(cast_to_pointer_type(dataPtr)), m_stride(stride)
     {
       PlainObjectType::Base::_check_template_params();
     }
 
     /** Constructor in the dynamic-size vector case.
       *
-      * \param data pointer to the array to map
+      * \param dataPtr pointer to the array to map
       * \param size the size of the vector expression
       * \param stride optional Stride object, passing the strides.
       */
-    inline Map(const Scalar* data, Index size, const StrideType& stride = StrideType())
-      : Base(data, size), m_stride(stride)
+    EIGEN_DEVICE_FUNC
+    inline Map(PointerArgType dataPtr, Index size, const StrideType& stride = StrideType())
+      : Base(cast_to_pointer_type(dataPtr), size), m_stride(stride)
     {
       PlainObjectType::Base::_check_template_params();
     }
 
     /** Constructor in the dynamic-size matrix case.
       *
-      * \param data pointer to the array to map
+      * \param dataPtr pointer to the array to map
       * \param rows the number of rows of the matrix expression
       * \param cols the number of columns of the matrix expression
       * \param stride optional Stride object, passing the strides.
       */
-    inline Map(const Scalar* data, Index rows, Index cols, const StrideType& stride = StrideType())
-      : Base(data, rows, cols), m_stride(stride)
+    EIGEN_DEVICE_FUNC
+    inline Map(PointerArgType dataPtr, Index rows, Index cols, const StrideType& stride = StrideType())
+      : Base(cast_to_pointer_type(dataPtr), rows, cols), m_stride(stride)
     {
       PlainObjectType::Base::_check_template_params();
     }
 
-
     EIGEN_INHERIT_ASSIGNMENT_OPERATORS(Map)
 
   protected:
     StrideType m_stride;
 };
 
-template<typename _Scalar, int _Rows, int _Cols, int _Options, int _MaxRows, int _MaxCols>
-inline Matrix<_Scalar, _Rows, _Cols, _Options, _MaxRows, _MaxCols>
-  ::Matrix(const Scalar *data)
-{
-  _set_noalias(Eigen::Map<Matrix>(data));
-}
+
+} // end namespace Eigen
 
 #endif // EIGEN_MAP_H

Eigen/src/Core/MapBase.h

@@ -4,28 +4,19 @@
 // Copyright (C) 2007-2010 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_MAPBASE_H
 #define EIGEN_MAPBASE_H
 
+#define EIGEN_STATIC_ASSERT_INDEX_BASED_ACCESS(Derived) \
+      EIGEN_STATIC_ASSERT((int(internal::evaluator<Derived>::Flags) & LinearAccessBit) || Derived::IsVectorAtCompileTime, \
+                          YOU_ARE_TRYING_TO_USE_AN_INDEX_BASED_ACCESSOR_ON_AN_EXPRESSION_THAT_DOES_NOT_SUPPORT_THAT)
+
+namespace Eigen { 
+
 /** \class MapBase
   * \ingroup Core_Module
   *
@@ -33,24 +24,27 @@
   *
   * \sa class Map, class Block
   */
-template<typename Derived> class MapBase
-  : public ei_dense_xpr_base<Derived>::type
+template<typename Derived> class MapBase<Derived, ReadOnlyAccessors>
+  : public internal::dense_xpr_base<Derived>::type
 {
   public:
 
-    typedef typename ei_dense_xpr_base<Derived>::type Base;
+    typedef typename internal::dense_xpr_base<Derived>::type Base;
     enum {
-      RowsAtCompileTime = ei_traits<Derived>::RowsAtCompileTime,
-      ColsAtCompileTime = ei_traits<Derived>::ColsAtCompileTime,
+      RowsAtCompileTime = internal::traits<Derived>::RowsAtCompileTime,
+      ColsAtCompileTime = internal::traits<Derived>::ColsAtCompileTime,
       SizeAtCompileTime = Base::SizeAtCompileTime
     };
 
-
-    typedef typename ei_traits<Derived>::StorageKind StorageKind;
-    typedef typename ei_traits<Derived>::Index Index;
-    typedef typename ei_traits<Derived>::Scalar Scalar;
-    typedef typename ei_packet_traits<Scalar>::type PacketScalar;
+    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 typename internal::conditional<
+                         bool(internal::is_lvalue<Derived>::value),
+                         Scalar *,
+                         const Scalar *>::type
+                     PointerType;
 
     using Base::derived;
 //    using Base::RowsAtCompileTime;
@@ -63,10 +57,6 @@ template<typename Derived> class MapBase
     using Base::Flags;
     using Base::IsRowMajor;
 
-    using Base::CoeffReadCost;
-
-//    using Base::derived;
-    using Base::const_cast_derived;
     using Base::rows;
     using Base::cols;
     using Base::size;
@@ -74,22 +64,19 @@ template<typename Derived> class MapBase
     using Base::coeffRef;
     using Base::lazyAssign;
     using Base::eval;
-//    using Base::operator=;
-    using Base::operator+=;
-    using Base::operator-=;
-    using Base::operator*=;
-    using Base::operator/=;
 
     using Base::innerStride;
     using Base::outerStride;
     using Base::rowStride;
     using Base::colStride;
 
+    // bug 217 - compile error on ICC 11.1
+    using Base::operator=;
 
     typedef typename Base::CoeffReturnType CoeffReturnType;
 
-    inline Index rows() const { return m_rows.value(); }
-    inline Index cols() const { return m_cols.value(); }
+    EIGEN_DEVICE_FUNC inline Index rows() const { return m_rows.value(); }
+    EIGEN_DEVICE_FUNC inline Index cols() const { return m_cols.value(); }
 
     /** Returns a pointer to the first coefficient of the matrix or vector.
       *
@@ -97,105 +84,178 @@ template<typename Derived> class MapBase
       *
       * \sa innerStride(), outerStride()
       */
-    inline const Scalar* data() const { return m_data; }
+    EIGEN_DEVICE_FUNC inline const Scalar* data() const { return m_data; }
 
-    inline const Scalar& coeff(Index row, Index col) const
+    EIGEN_DEVICE_FUNC
+    inline const Scalar& coeff(Index rowId, Index colId) const
     {
-      return m_data[col * colStride() + row * rowStride()];
-    }
-
-    inline Scalar& coeffRef(Index row, Index col)
-    {
-      return const_cast<Scalar*>(m_data)[col * colStride() + row * rowStride()];
+      return m_data[colId * colStride() + rowId * rowStride()];
     }
 
+    EIGEN_DEVICE_FUNC
     inline const Scalar& coeff(Index index) const
     {
-      ei_assert(Derived::IsVectorAtCompileTime || (ei_traits<Derived>::Flags & LinearAccessBit));
+      EIGEN_STATIC_ASSERT_INDEX_BASED_ACCESS(Derived)
       return m_data[index * innerStride()];
     }
 
-    inline Scalar& coeffRef(Index index)
+    EIGEN_DEVICE_FUNC
+    inline const Scalar& coeffRef(Index rowId, Index colId) const
     {
-      ei_assert(Derived::IsVectorAtCompileTime || (ei_traits<Derived>::Flags & LinearAccessBit));
-      return const_cast<Scalar*>(m_data)[index * innerStride()];
+      return this->m_data[colId * colStride() + rowId * rowStride()];
+    }
+
+    EIGEN_DEVICE_FUNC
+    inline const Scalar& coeffRef(Index index) const
+    {
+      EIGEN_STATIC_ASSERT_INDEX_BASED_ACCESS(Derived)
+      return this->m_data[index * innerStride()];
     }
 
     template<int LoadMode>
-    inline PacketScalar packet(Index row, Index col) const
+    inline PacketScalar packet(Index rowId, Index colId) const
     {
-      return ei_ploadt<PacketScalar, LoadMode>
-               (m_data + (col * colStride() + row * rowStride()));
+      return internal::ploadt<PacketScalar, LoadMode>
+               (m_data + (colId * colStride() + rowId * rowStride()));
     }
 
     template<int LoadMode>
     inline PacketScalar packet(Index index) const
     {
-      return ei_ploadt<PacketScalar, LoadMode>(m_data + index * innerStride());
+      EIGEN_STATIC_ASSERT_INDEX_BASED_ACCESS(Derived)
+      return internal::ploadt<PacketScalar, LoadMode>(m_data + index * innerStride());
     }
 
-    template<int StoreMode>
-    inline void writePacket(Index row, Index col, const PacketScalar& x)
-    {
-      ei_pstoret<Scalar, PacketScalar, StoreMode>
-               (const_cast<Scalar*>(m_data) + (col * colStride() + row * rowStride()), x);
-    }
-
-    template<int StoreMode>
-    inline void writePacket(Index index, const PacketScalar& x)
-    {
-      ei_pstoret<Scalar, PacketScalar, StoreMode>
-        (const_cast<Scalar*>(m_data) + index * innerStride(), x);
-    }
-
-    inline MapBase(const Scalar* data) : m_data(data), m_rows(RowsAtCompileTime), m_cols(ColsAtCompileTime)
+    EIGEN_DEVICE_FUNC
+    explicit inline MapBase(PointerType dataPtr) : m_data(dataPtr), m_rows(RowsAtCompileTime), m_cols(ColsAtCompileTime)
     {
       EIGEN_STATIC_ASSERT_FIXED_SIZE(Derived)
       checkSanity();
     }
 
-    inline MapBase(const Scalar* data, Index size)
-            : m_data(data),
-              m_rows(RowsAtCompileTime == Dynamic ? size : Index(RowsAtCompileTime)),
-              m_cols(ColsAtCompileTime == Dynamic ? size : Index(ColsAtCompileTime))
+    EIGEN_DEVICE_FUNC
+    inline MapBase(PointerType dataPtr, Index vecSize)
+            : m_data(dataPtr),
+              m_rows(RowsAtCompileTime == Dynamic ? vecSize : Index(RowsAtCompileTime)),
+              m_cols(ColsAtCompileTime == Dynamic ? vecSize : Index(ColsAtCompileTime))
     {
       EIGEN_STATIC_ASSERT_VECTOR_ONLY(Derived)
-      ei_assert(size >= 0);
-      ei_assert(data == 0 || SizeAtCompileTime == Dynamic || SizeAtCompileTime == size);
+      eigen_assert(vecSize >= 0);
+      eigen_assert(dataPtr == 0 || SizeAtCompileTime == Dynamic || SizeAtCompileTime == vecSize);
       checkSanity();
     }
 
-    inline MapBase(const Scalar* data, Index rows, Index cols)
-            : m_data(data), m_rows(rows), m_cols(cols)
+    EIGEN_DEVICE_FUNC
+    inline MapBase(PointerType dataPtr, Index rows, Index cols)
+            : m_data(dataPtr), m_rows(rows), m_cols(cols)
     {
-      ei_assert( (data == 0)
+      eigen_assert( (dataPtr == 0)
               || (   rows >= 0 && (RowsAtCompileTime == Dynamic || RowsAtCompileTime == rows)
                   && cols >= 0 && (ColsAtCompileTime == Dynamic || ColsAtCompileTime == cols)));
       checkSanity();
     }
 
-    Derived& operator=(const MapBase& other)
-    {
-      Base::operator=(other);
-      return derived();
-    }
-
-    using Base::operator=;
+    #ifdef EIGEN_MAPBASE_PLUGIN
+    #include EIGEN_MAPBASE_PLUGIN
+    #endif
 
   protected:
 
+    EIGEN_DEVICE_FUNC
     void checkSanity() const
     {
-      EIGEN_STATIC_ASSERT(EIGEN_IMPLIES(ei_traits<Derived>::Flags&PacketAccessBit,
-                                        ei_inner_stride_at_compile_time<Derived>::ret==1),
-                          PACKET_ACCESS_REQUIRES_TO_HAVE_INNER_STRIDE_FIXED_TO_1);
-      ei_assert(EIGEN_IMPLIES(ei_traits<Derived>::Flags&AlignedBit, (size_t(m_data) % (sizeof(Scalar)*ei_packet_traits<Scalar>::size)) == 0)
-        && "data is not aligned");
+#if EIGEN_MAX_ALIGN_BYTES>0
+      eigen_assert(((size_t(m_data) % EIGEN_PLAIN_ENUM_MAX(1,internal::traits<Derived>::Alignment)) == 0) && "data is not aligned");
+#endif
     }
 
-    const Scalar* EIGEN_RESTRICT m_data;
-    const ei_variable_if_dynamic<Index, RowsAtCompileTime> m_rows;
-    const ei_variable_if_dynamic<Index, ColsAtCompileTime> m_cols;
+    PointerType m_data;
+    const internal::variable_if_dynamic<Index, RowsAtCompileTime> m_rows;
+    const internal::variable_if_dynamic<Index, ColsAtCompileTime> m_cols;
 };
 
+template<typename Derived> class MapBase<Derived, WriteAccessors>
+  : public MapBase<Derived, ReadOnlyAccessors>
+{
+    typedef MapBase<Derived, ReadOnlyAccessors> ReadOnlyMapBase;
+  public:
+
+    typedef MapBase<Derived, ReadOnlyAccessors> Base;
+
+    typedef typename Base::Scalar Scalar;
+    typedef typename Base::PacketScalar PacketScalar;
+    typedef typename Base::StorageIndex StorageIndex;
+    typedef typename Base::PointerType PointerType;
+
+    using Base::derived;
+    using Base::rows;
+    using Base::cols;
+    using Base::size;
+    using Base::coeff;
+    using Base::coeffRef;
+
+    using Base::innerStride;
+    using Base::outerStride;
+    using Base::rowStride;
+    using Base::colStride;
+
+    typedef typename internal::conditional<
+                    internal::is_lvalue<Derived>::value,
+                    Scalar,
+                    const Scalar
+                  >::type ScalarWithConstIfNotLvalue;
+
+    EIGEN_DEVICE_FUNC
+    inline const Scalar* data() const { return this->m_data; }
+    EIGEN_DEVICE_FUNC
+    inline ScalarWithConstIfNotLvalue* data() { return this->m_data; } // no const-cast here so non-const-correct code will give a compile error
+
+    EIGEN_DEVICE_FUNC
+    inline ScalarWithConstIfNotLvalue& coeffRef(Index row, Index col)
+    {
+      return this->m_data[col * colStride() + row * rowStride()];
+    }
+
+    EIGEN_DEVICE_FUNC
+    inline ScalarWithConstIfNotLvalue& coeffRef(Index index)
+    {
+      EIGEN_STATIC_ASSERT_INDEX_BASED_ACCESS(Derived)
+      return this->m_data[index * innerStride()];
+    }
+
+    template<int StoreMode>
+    inline void writePacket(Index row, Index col, const PacketScalar& val)
+    {
+      internal::pstoret<Scalar, PacketScalar, StoreMode>
+               (this->m_data + (col * colStride() + row * rowStride()), val);
+    }
+
+    template<int StoreMode>
+    inline void writePacket(Index index, const PacketScalar& val)
+    {
+      EIGEN_STATIC_ASSERT_INDEX_BASED_ACCESS(Derived)
+      internal::pstoret<Scalar, PacketScalar, StoreMode>
+                (this->m_data + index * innerStride(), val);
+    }
+
+    EIGEN_DEVICE_FUNC explicit inline MapBase(PointerType dataPtr) : Base(dataPtr) {}
+    EIGEN_DEVICE_FUNC inline MapBase(PointerType dataPtr, Index vecSize) : Base(dataPtr, vecSize) {}
+    EIGEN_DEVICE_FUNC inline MapBase(PointerType dataPtr, Index rows, Index cols) : Base(dataPtr, rows, cols) {}
+
+    EIGEN_DEVICE_FUNC
+    Derived& operator=(const MapBase& other)
+    {
+      ReadOnlyMapBase::Base::operator=(other);
+      return derived();
+    }
+
+    // In theory we could simply refer to Base:Base::operator=, but MSVC does not like Base::Base,
+    // see bugs 821 and 920.
+    using ReadOnlyMapBase::Base::operator=;
+};
+
+#undef EIGEN_STATIC_ASSERT_INDEX_BASED_ACCESS
+
+} // end namespace Eigen
+
 #endif // EIGEN_MAPBASE_H

Eigen/src/Core/Matrix.h

@@ -4,28 +4,15 @@
 // Copyright (C) 2006-2010 Benoit Jacob <jacob.benoit.1@gmail.com>
 // Copyright (C) 2008-2009 Gael Guennebaud <gael.guennebaud@inria.fr>
 //
-// Eigen is free software; you can redistribute it and/or
-// modify it under the terms of the GNU Lesser General Public
-// 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_MATRIX_H
 #define EIGEN_MATRIX_H
 
+namespace Eigen {
+
 /** \class Matrix
   * \ingroup Core_Module
   *
@@ -37,15 +24,15 @@
   * The %Matrix class encompasses \em both fixed-size and dynamic-size objects (\ref fixedsize "note").
   *
   * The first three template parameters are required:
-  * \tparam _Scalar \anchor matrix_tparam_scalar Numeric type, e.g. float, double, int or std::complex<float>.
-  *                 User defined sclar types are supported as well (see \ref user_defined_scalars "here").
+  * \tparam _Scalar Numeric type, e.g. float, double, int or std::complex<float>.
+  *                 User defined scalar types are supported as well (see \ref user_defined_scalars "here").
   * \tparam _Rows Number of rows, or \b Dynamic
   * \tparam _Cols Number of columns, or \b Dynamic
   *
   * The remaining template parameters are optional -- in most cases you don't have to worry about them.
-  * \tparam _Options \anchor matrix_tparam_options A combination of either \b RowMajor or \b ColMajor, and of either
-  *                 \b AutoAlign or \b DontAlign.
-  *                 The former controls storage order, and defaults to column-major. The latter controls alignment, which is required
+  * \tparam _Options A combination of either \b #RowMajor or \b #ColMajor, and of either
+  *                 \b #AutoAlign or \b #DontAlign.
+  *                 The former controls \ref TopicStorageOrders "storage order", and defaults to column-major. The latter controls alignment, which is required
   *                 for vectorization. It defaults to aligning matrices except for fixed sizes that aren't a multiple of the packet size.
   * \tparam _MaxRows Maximum number of rows. Defaults to \a _Rows (\ref maxrows "note").
   * \tparam _MaxCols Maximum number of columns. Defaults to \a _Cols (\ref maxrows "note").
@@ -79,6 +66,9 @@
   * m(0, 3) = 3;
   * \endcode
   *
+  * This class can be extended with the help of the plugin mechanism described on the page
+  * \ref TopicCustomizingEigen by defining the preprocessor symbol \c EIGEN_MATRIX_PLUGIN.
+  *
   * <i><b>Some notes:</b></i>
   *
   * <dl>
@@ -107,38 +97,93 @@
   * are the dimensions of the original matrix, while _Rows and _Cols are Dynamic.</dd>
   * </dl>
   *
-  * \see MatrixBase for the majority of the API methods for matrices, \ref TopicClassHierarchy
+  * <i><b>ABI and storage layout</b></i>
+  * 
+  * The table below summarizes the ABI of some possible Matrix instances which is fixed thorough the lifetime of Eigen 3.
+  * <table  class="manual">
+  * <tr><th>Matrix type</th><th>Equivalent C structure</th></tr>
+  * <tr><td>\code Matrix<T,Dynamic,Dynamic> \endcode</td><td>\code
+  * struct {
+  *   T *data;                  // with (size_t(data)%EIGEN_MAX_ALIGN_BYTES)==0
+  *   Eigen::Index rows, cols;
+  *  };
+  * \endcode</td></tr>
+  * <tr class="alt"><td>\code
+  * Matrix<T,Dynamic,1>
+  * Matrix<T,1,Dynamic> \endcode</td><td>\code
+  * struct {
+  *   T *data;                  // with (size_t(data)%EIGEN_MAX_ALIGN_BYTES)==0
+  *   Eigen::Index size;
+  *  };
+  * \endcode</td></tr>
+  * <tr><td>\code Matrix<T,Rows,Cols> \endcode</td><td>\code
+  * struct {
+  *   T data[Rows*Cols];        // with (size_t(data)%A(Rows*Cols*sizeof(T)))==0
+  *  };
+  * \endcode</td></tr>
+  * <tr class="alt"><td>\code Matrix<T,Dynamic,Dynamic,0,MaxRows,MaxCols> \endcode</td><td>\code
+  * struct {
+  *   T data[MaxRows*MaxCols];  // with (size_t(data)%A(MaxRows*MaxCols*sizeof(T)))==0
+  *   Eigen::Index rows, cols;
+  *  };
+  * \endcode</td></tr>
+  * </table>
+  * Note that in this table Rows, Cols, MaxRows and MaxCols are all positive integers. A(S) is defined to the largest possible power-of-two
+  * smaller to EIGEN_MAX_STATIC_ALIGN_BYTES.
+  * 
+  * \see MatrixBase for the majority of the API methods for matrices, \ref TopicClassHierarchy, 
+  * \ref TopicStorageOrders 
   */
+
+namespace internal {
 template<typename _Scalar, int _Rows, int _Cols, int _Options, int _MaxRows, int _MaxCols>
-struct ei_traits<Matrix<_Scalar, _Rows, _Cols, _Options, _MaxRows, _MaxCols> >
+struct traits<Matrix<_Scalar, _Rows, _Cols, _Options, _MaxRows, _MaxCols> >
 {
+private:
+  enum { size = internal::size_at_compile_time<_Rows,_Cols>::ret };
+  typedef typename find_best_packet<_Scalar,size>::type PacketScalar;
+  enum {
+      row_major_bit = _Options&RowMajor ? RowMajorBit : 0,
+      is_dynamic_size_storage = _MaxRows==Dynamic || _MaxCols==Dynamic,
+      max_size = is_dynamic_size_storage ? Dynamic : _MaxRows*_MaxCols,
+      default_alignment = compute_default_alignment<_Scalar,max_size>::value,
+      actual_alignment = ((_Options&DontAlign)==0) ? default_alignment : 0,
+      required_alignment = unpacket_traits<PacketScalar>::alignment,
+      packet_access_bit = packet_traits<_Scalar>::Vectorizable && (actual_alignment>=required_alignment) ? PacketAccessBit : 0
+    };
+    
+public:
   typedef _Scalar Scalar;
   typedef Dense StorageKind;
-  typedef DenseIndex Index;
+  typedef Eigen::Index StorageIndex;
   typedef MatrixXpr XprKind;
   enum {
     RowsAtCompileTime = _Rows,
     ColsAtCompileTime = _Cols,
     MaxRowsAtCompileTime = _MaxRows,
     MaxColsAtCompileTime = _MaxCols,
-    Flags = ei_compute_matrix_flags<_Scalar, _Rows, _Cols, _Options, _MaxRows, _MaxCols>::ret,
-    CoeffReadCost = NumTraits<Scalar>::ReadCost,
+    Flags = compute_matrix_flags<_Scalar, _Rows, _Cols, _Options, _MaxRows, _MaxCols>::ret,
     Options = _Options,
     InnerStrideAtCompileTime = 1,
-    OuterStrideAtCompileTime = (Options&RowMajor) ? ColsAtCompileTime : RowsAtCompileTime
+    OuterStrideAtCompileTime = (Options&RowMajor) ? ColsAtCompileTime : RowsAtCompileTime,
+    
+    // FIXME, the following flag in only used to define NeedsToAlign in PlainObjectBase
+    EvaluatorFlags = LinearAccessBit | DirectAccessBit | packet_access_bit | row_major_bit,
+    Alignment = actual_alignment
   };
 };
+}
 
 template<typename _Scalar, int _Rows, int _Cols, int _Options, int _MaxRows, int _MaxCols>
 class Matrix
-  : public DenseStorageBase<Matrix<_Scalar, _Rows, _Cols, _Options, _MaxRows, _MaxCols> >
+  : public PlainObjectBase<Matrix<_Scalar, _Rows, _Cols, _Options, _MaxRows, _MaxCols> >
 {
   public:
 
     /** \brief Base class typedef.
-      * \sa DenseStorageBase
+      * \sa PlainObjectBase
       */
-    typedef DenseStorageBase<Matrix> Base;
+    typedef PlainObjectBase<Matrix> Base;
 
     enum { Options = _Options };
 
@@ -146,10 +191,6 @@ class Matrix
 
     typedef typename Base::PlainObject PlainObject;
 
-    enum { NeedsToAlign = (!(Options&DontAlign))
-                          && SizeAtCompileTime!=Dynamic && ((static_cast<int>(sizeof(Scalar))*SizeAtCompileTime)%16)==0 };
-    EIGEN_MAKE_ALIGNED_OPERATOR_NEW_IF(NeedsToAlign)
-
     using Base::base;
     using Base::coeffRef;
 
@@ -161,6 +202,7 @@ class Matrix
       *
       * \callgraph
       */
+    EIGEN_DEVICE_FUNC
     EIGEN_STRONG_INLINE Matrix& operator=(const Matrix& other)
     {
       return Base::_set(other);
@@ -177,7 +219,8 @@ class Matrix
       * remain row-vectors and vectors remain vectors.
       */
     template<typename OtherDerived>
-    EIGEN_STRONG_INLINE Matrix& operator=(const MatrixBase<OtherDerived>& other)
+    EIGEN_DEVICE_FUNC
+    EIGEN_STRONG_INLINE Matrix& operator=(const DenseBase<OtherDerived>& other)
     {
       return Base::_set(other);
     }
@@ -189,12 +232,14 @@ class Matrix
       * \copydetails DenseBase::operator=(const EigenBase<OtherDerived> &other)
       */
     template<typename OtherDerived>
+    EIGEN_DEVICE_FUNC
     EIGEN_STRONG_INLINE Matrix& operator=(const EigenBase<OtherDerived> &other)
     {
       return Base::operator=(other);
     }
 
     template<typename OtherDerived>
+    EIGEN_DEVICE_FUNC
     EIGEN_STRONG_INLINE Matrix& operator=(const ReturnByValue<OtherDerived>& func)
     {
       return Base::operator=(func);
@@ -210,52 +255,95 @@ class Matrix
       *
       * \sa resize(Index,Index)
       */
-    EIGEN_STRONG_INLINE explicit Matrix() : Base()
+    EIGEN_DEVICE_FUNC
+    EIGEN_STRONG_INLINE Matrix() : Base()
     {
       Base::_check_template_params();
-      EIGEN_INITIALIZE_BY_ZERO_IF_THAT_OPTION_IS_ENABLED
+      EIGEN_INITIALIZE_COEFFS_IF_THAT_OPTION_IS_ENABLED
     }
 
     // FIXME is it still needed
-    Matrix(ei_constructor_without_unaligned_array_assert)
-      : Base(ei_constructor_without_unaligned_array_assert())
-    { Base::_check_template_params(); EIGEN_INITIALIZE_BY_ZERO_IF_THAT_OPTION_IS_ENABLED }
+    EIGEN_DEVICE_FUNC
+    explicit Matrix(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 }
 
-    /** \brief 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 Matrix() instead.
-      */
-    EIGEN_STRONG_INLINE explicit Matrix(Index dim)
-      : Base(dim, RowsAtCompileTime == 1 ? 1 : dim, ColsAtCompileTime == 1 ? 1 : dim)
+#ifdef EIGEN_HAVE_RVALUE_REFERENCES
+    EIGEN_DEVICE_FUNC
+    Matrix(Matrix&& other)
+      : Base(std::move(other))
     {
       Base::_check_template_params();
-      EIGEN_STATIC_ASSERT_VECTOR_ONLY(Matrix)
-      ei_assert(dim > 0);
-      ei_assert(SizeAtCompileTime == Dynamic || SizeAtCompileTime == dim);
-      EIGEN_INITIALIZE_BY_ZERO_IF_THAT_OPTION_IS_ENABLED
+      if (RowsAtCompileTime!=Dynamic && ColsAtCompileTime!=Dynamic)
+        Base::_set_noalias(other);
     }
+    EIGEN_DEVICE_FUNC
+    Matrix& operator=(Matrix&& other)
+    {
+      other.swap(*this);
+      return *this;
+    }
+#endif
 
     #ifndef EIGEN_PARSED_BY_DOXYGEN
+
+    // This constructor is for both 1x1 matrices and dynamic vectors
+    template<typename T>
+    EIGEN_DEVICE_FUNC
+    EIGEN_STRONG_INLINE explicit Matrix(const T& x)
+    {
+      Base::_check_template_params();
+      Base::template _init1<T>(x);
+    }
+
     template<typename T0, typename T1>
+    EIGEN_DEVICE_FUNC
     EIGEN_STRONG_INLINE Matrix(const T0& x, const T1& y)
     {
       Base::_check_template_params();
       Base::template _init2<T0,T1>(x, y);
     }
     #else
+    /** \brief Constructs a fixed-sized matrix initialized with coefficients starting at \a data */
+    EIGEN_DEVICE_FUNC
+    explicit Matrix(const Scalar *data);
+
+    /** \brief Constructs a vector or row-vector with given dimension. \only_for_vectors
+      *
+      * This is useful for dynamic-size vectors. For fixed-size vectors,
+      * it is redundant to pass these parameters, so one should use the default constructor
+      * Matrix() instead.
+      * 
+      * \warning This constructor is disabled for fixed-size \c 1x1 matrices. For instance,
+      * calling Matrix<double,1,1>(1) will call the initialization constructor: Matrix(const Scalar&).
+      * For fixed-size \c 1x1 matrices it is therefore recommended to use the default
+      * constructor Matrix() instead, especially when using one of the non standard
+      * \c EIGEN_INITIALIZE_MATRICES_BY_{ZERO,\c NAN} macros (see \ref TopicPreprocessorDirectives).
+      */
+    EIGEN_STRONG_INLINE explicit Matrix(Index dim);
+    /** \brief Constructs an initialized 1x1 matrix with the given coefficient */
+    Matrix(const Scalar& x);
     /** \brief Constructs an uninitialized matrix with \a rows rows and \a cols columns.
       *
       * This is useful for dynamic-size matrices. For fixed-size matrices,
       * it is redundant to pass these parameters, so one should use the default constructor
-      * Matrix() instead. */
+      * Matrix() instead.
+      * 
+      * \warning This constructor is disabled for fixed-size \c 1x2 and \c 2x1 vectors. For instance,
+      * calling Matrix2f(2,1) will call the initialization constructor: Matrix(const Scalar& x, const Scalar& y).
+      * For fixed-size \c 1x2 or \c 2x1 vectors it is therefore recommended to use the default
+      * constructor Matrix() instead, especially when using one of the non standard
+      * \c EIGEN_INITIALIZE_MATRICES_BY_{ZERO,\c NAN} macros (see \ref TopicPreprocessorDirectives).
+      */
+    EIGEN_DEVICE_FUNC
     Matrix(Index rows, Index cols);
+    
     /** \brief Constructs an initialized 2D vector with given coefficients */
     Matrix(const Scalar& x, const Scalar& y);
     #endif
 
     /** \brief Constructs an initialized 3D vector with given coefficients */
+    EIGEN_DEVICE_FUNC
     EIGEN_STRONG_INLINE Matrix(const Scalar& x, const Scalar& y, const Scalar& z)
     {
       Base::_check_template_params();
@@ -265,6 +353,7 @@ class Matrix
       m_storage.data()[2] = z;
     }
     /** \brief Constructs an initialized 4D vector with given coefficients */
+    EIGEN_DEVICE_FUNC
     EIGEN_STRONG_INLINE Matrix(const Scalar& x, const Scalar& y, const Scalar& z, const Scalar& w)
     {
       Base::_check_template_params();
@@ -275,62 +364,31 @@ class Matrix
       m_storage.data()[3] = w;
     }
 
-    explicit Matrix(const Scalar *data);
 
-    /** \brief Constructor copying the value of the expression \a other */
-    template<typename OtherDerived>
-    EIGEN_STRONG_INLINE Matrix(const MatrixBase<OtherDerived>& other)
-             : Base(other.rows() * other.cols(), other.rows(), other.cols())
-    {
-      Base::_check_template_params();
-      Base::_set_noalias(other);
-    }
     /** \brief Copy constructor */
-    EIGEN_STRONG_INLINE Matrix(const Matrix& other)
-            : Base(other.rows() * other.cols(), other.rows(), other.cols())
-    {
-      Base::_check_template_params();
-      Base::_set_noalias(other);
-    }
-    /** \brief Copy constructor with in-place evaluation */
-    template<typename OtherDerived>
-    EIGEN_STRONG_INLINE Matrix(const ReturnByValue<OtherDerived>& other)
-    {
-      Base::_check_template_params();
-      Base::resize(other.rows(), other.cols());
-      other.evalTo(*this);
-    }
+    EIGEN_DEVICE_FUNC
+    EIGEN_STRONG_INLINE Matrix(const Matrix& other) : Base(other)
+    { }
 
     /** \brief Copy constructor for generic expressions.
       * \sa MatrixBase::operator=(const EigenBase<OtherDerived>&)
       */
     template<typename OtherDerived>
+    EIGEN_DEVICE_FUNC
     EIGEN_STRONG_INLINE Matrix(const EigenBase<OtherDerived> &other)
-      : Base(other.derived().rows() * other.derived().cols(), other.derived().rows(), other.derived().cols())
-    {
-      Base::_check_template_params();
-      Base::resize(other.rows(), other.cols());
-      // FIXME/CHECK: isn't *this = other.derived() more efficient. it allows to
-      //              go for pure _set() implementations, right?
-      *this = other;
-    }
+      : Base(other.derived())
+    { }
 
-    /** \internal
-      * \brief Override MatrixBase::swap() since for dynamic-sized matrices
-      * of same type it is enough to swap the data pointers.
-      */
-    template<typename OtherDerived>
-    void swap(MatrixBase<OtherDerived> EIGEN_REF_TO_TEMPORARY other)
-    { this->_swap(other.derived()); }
-
-    inline Index innerStride() const { return 1; }
-    inline Index outerStride() const { return this->innerSize(); }
+    EIGEN_DEVICE_FUNC inline Index innerStride() const { return 1; }
+    EIGEN_DEVICE_FUNC inline Index outerStride() const { return this->innerSize(); }
 
     /////////// Geometry module ///////////
 
     template<typename OtherDerived>
+    EIGEN_DEVICE_FUNC
     explicit Matrix(const RotationBase<OtherDerived,ColsAtCompileTime>& r);
     template<typename OtherDerived>
+    EIGEN_DEVICE_FUNC
     Matrix& operator=(const RotationBase<OtherDerived,ColsAtCompileTime>& r);
 
     // allow to extend Matrix outside Eigen
@@ -340,7 +398,7 @@ class Matrix
 
   protected:
     template <typename Derived, typename OtherDerived, bool IsVector>
-    friend struct ei_conservative_resize_like_impl;
+    friend struct internal::conservative_resize_like_impl;
 
     using Base::m_storage;
 };
@@ -396,25 +454,8 @@ EIGEN_MAKE_TYPEDEFS_ALL_SIZES(std::complex<double>, cd)
 
 #undef EIGEN_MAKE_TYPEDEFS_ALL_SIZES
 #undef EIGEN_MAKE_TYPEDEFS
+#undef EIGEN_MAKE_FIXED_TYPEDEFS
 
-#undef EIGEN_MAKE_TYPEDEFS_LARGE
-
-#define EIGEN_USING_MATRIX_TYPEDEFS_FOR_TYPE_AND_SIZE(TypeSuffix, SizeSuffix) \
-using Eigen::Matrix##SizeSuffix##TypeSuffix; \
-using Eigen::Vector##SizeSuffix##TypeSuffix; \
-using Eigen::RowVector##SizeSuffix##TypeSuffix;
-
-#define EIGEN_USING_MATRIX_TYPEDEFS_FOR_TYPE(TypeSuffix) \
-EIGEN_USING_MATRIX_TYPEDEFS_FOR_TYPE_AND_SIZE(TypeSuffix, 2) \
-EIGEN_USING_MATRIX_TYPEDEFS_FOR_TYPE_AND_SIZE(TypeSuffix, 3) \
-EIGEN_USING_MATRIX_TYPEDEFS_FOR_TYPE_AND_SIZE(TypeSuffix, 4) \
-EIGEN_USING_MATRIX_TYPEDEFS_FOR_TYPE_AND_SIZE(TypeSuffix, X) \
-
-#define EIGEN_USING_MATRIX_TYPEDEFS \
-EIGEN_USING_MATRIX_TYPEDEFS_FOR_TYPE(i) \
-EIGEN_USING_MATRIX_TYPEDEFS_FOR_TYPE(f) \
-EIGEN_USING_MATRIX_TYPEDEFS_FOR_TYPE(d) \
-EIGEN_USING_MATRIX_TYPEDEFS_FOR_TYPE(cf) \
-EIGEN_USING_MATRIX_TYPEDEFS_FOR_TYPE(cd)
+} // end namespace Eigen
 
 #endif // EIGEN_MATRIX_H

Eigen/src/Core/MatrixBase.h

@@ -4,28 +4,15 @@
 // Copyright (C) 2006-2009 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_MATRIXBASE_H
 #define EIGEN_MATRIXBASE_H
 
+namespace Eigen {
+
 /** \class MatrixBase
   * \ingroup Core_Module
   *
@@ -38,7 +25,7 @@
   * Note that some methods are defined in other modules such as the \ref LU_Module LU module
   * for all functions related to matrix inversions.
   *
-  * \param Derived is the derived type, e.g. a matrix type, or an expression, etc.
+  * \tparam Derived is the derived type, e.g. a matrix type, or an expression, etc.
   *
   * When writing a function taking Eigen objects as argument, if you want your function
   * to take as argument any matrix, vector, or expression, just let it take a
@@ -53,6 +40,9 @@
     }
   * \endcode
   *
+  * This class can be extended with the help of the plugin mechanism described on the page
+  * \ref TopicCustomizingEigen by defining the preprocessor symbol \c EIGEN_MATRIXBASE_PLUGIN.
+  *
   * \sa \ref TopicClassHierarchy
   */
 template<typename Derived> class MatrixBase
@@ -61,10 +51,10 @@ template<typename Derived> class MatrixBase
   public:
 #ifndef EIGEN_PARSED_BY_DOXYGEN
     typedef MatrixBase StorageBaseType;
-    typedef typename ei_traits<Derived>::StorageKind StorageKind;
-    typedef typename ei_traits<Derived>::Index Index;
-    typedef typename ei_traits<Derived>::Scalar Scalar;
-    typedef typename ei_packet_traits<Scalar>::type PacketScalar;
+    typedef typename internal::traits<Derived>::StorageKind StorageKind;
+    typedef typename internal::traits<Derived>::StorageIndex StorageIndex;
+    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;
@@ -76,8 +66,7 @@ template<typename Derived> class MatrixBase
     using Base::MaxSizeAtCompileTime;
     using Base::IsVectorAtCompileTime;
     using Base::Flags;
-    using Base::CoeffReadCost;
-
+    
     using Base::derived;
     using Base::const_cast_derived;
     using Base::rows;
@@ -91,8 +80,11 @@ template<typename Derived> class MatrixBase
     using Base::operator-=;
     using Base::operator*=;
     using Base::operator/=;
+    using Base::operator*;
+    using Base::operator/;
 
     typedef typename Base::CoeffReturnType CoeffReturnType;
+    typedef typename Base::ConstTransposeReturnType ConstTransposeReturnType;
     typedef typename Base::RowXpr RowXpr;
     typedef typename Base::ColXpr ColXpr;
 #endif // not EIGEN_PARSED_BY_DOXYGEN
@@ -107,38 +99,27 @@ template<typename Derived> class MatrixBase
 
     /** \returns the size of the main diagonal, which is min(rows(),cols()).
       * \sa rows(), cols(), SizeAtCompileTime. */
-    inline Index diagonalSize() const { return std::min(rows(),cols()); }
+    EIGEN_DEVICE_FUNC
+    inline Index diagonalSize() const { return (std::min)(rows(),cols()); }
 
-    /** \brief The plain matrix 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 Matrix<typename ei_traits<Derived>::Scalar,
-                ei_traits<Derived>::RowsAtCompileTime,
-                ei_traits<Derived>::ColsAtCompileTime,
-                AutoAlign | (ei_traits<Derived>::Flags&RowMajorBit ? RowMajor : ColMajor),
-                ei_traits<Derived>::MaxRowsAtCompileTime,
-                ei_traits<Derived>::MaxColsAtCompileTime
-          > PlainObject;
+    typedef typename Base::PlainObject PlainObject;
 
 #ifndef EIGEN_PARSED_BY_DOXYGEN
     /** \internal Represents a matrix with all coefficients equal to one another*/
-    typedef CwiseNullaryOp<ei_scalar_constant_op<Scalar>,Derived> ConstantReturnType;
+    typedef CwiseNullaryOp<internal::scalar_constant_op<Scalar>,PlainObject> ConstantReturnType;
     /** \internal the return type of MatrixBase::adjoint() */
-    typedef typename ei_meta_if<NumTraits<Scalar>::IsComplex,
-                        CwiseUnaryOp<ei_scalar_conjugate_op<Scalar>, Eigen::Transpose<Derived> >,
-                        Transpose<Derived>
-                     >::ret AdjointReturnType;
+    typedef typename internal::conditional<NumTraits<Scalar>::IsComplex,
+                        CwiseUnaryOp<internal::scalar_conjugate_op<Scalar>, ConstTransposeReturnType>,
+                        ConstTransposeReturnType
+                     >::type AdjointReturnType;
     /** \internal Return type of eigenvalues() */
-    typedef Matrix<std::complex<RealScalar>, ei_traits<Derived>::ColsAtCompileTime, 1, ColMajor> EigenvaluesReturnType;
+    typedef Matrix<std::complex<RealScalar>, internal::traits<Derived>::ColsAtCompileTime, 1, ColMajor> EigenvaluesReturnType;
     /** \internal the return type of identity */
-    typedef CwiseNullaryOp<ei_scalar_identity_op<Scalar>,Derived> IdentityReturnType;
+    typedef CwiseNullaryOp<internal::scalar_identity_op<Scalar>,PlainObject> IdentityReturnType;
     /** \internal the return type of unit vectors */
-    typedef Block<CwiseNullaryOp<ei_scalar_identity_op<Scalar>, SquareMatrixType>,
-                  ei_traits<Derived>::RowsAtCompileTime,
-                  ei_traits<Derived>::ColsAtCompileTime> BasisReturnType;
+    typedef Block<const CwiseNullaryOp<internal::scalar_identity_op<Scalar>, SquareMatrixType>,
+                  internal::traits<Derived>::RowsAtCompileTime,
+                  internal::traits<Derived>::ColsAtCompileTime> BasisReturnType;
 #endif // not EIGEN_PARSED_BY_DOXYGEN
 
 #define EIGEN_CURRENT_STORAGE_BASE_CLASS Eigen::MatrixBase
@@ -154,36 +135,48 @@ template<typename Derived> class MatrixBase
     /** 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
     Derived& operator=(const MatrixBase& other);
 
     // We cannot inherit here via Base::operator= since it is causing
     // trouble with MSVC.
 
     template <typename OtherDerived>
+    EIGEN_DEVICE_FUNC
     Derived& operator=(const DenseBase<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>& other);
 
-#ifndef EIGEN_PARSED_BY_DOXYGEN
-    template<typename ProductDerived, typename Lhs, typename Rhs>
-    Derived& lazyAssign(const ProductBase<ProductDerived, Lhs,Rhs>& other);
-#endif // not EIGEN_PARSED_BY_DOXYGEN
-
     template<typename OtherDerived>
+    EIGEN_DEVICE_FUNC
     Derived& operator+=(const MatrixBase<OtherDerived>& other);
     template<typename OtherDerived>
+    EIGEN_DEVICE_FUNC
     Derived& operator-=(const MatrixBase<OtherDerived>& other);
 
+#ifdef __CUDACC__
     template<typename OtherDerived>
-    const typename ProductReturnType<Derived,OtherDerived>::Type
-    operator*(const MatrixBase<OtherDerived> &other) const;
+    EIGEN_DEVICE_FUNC
+    const Product<Derived,OtherDerived,LazyProduct>
+    operator*(const MatrixBase<OtherDerived> &other) const
+    { return this->lazyProduct(other); }
+#else
 
     template<typename OtherDerived>
-    const typename LazyProductReturnType<Derived,OtherDerived>::Type
+    const Product<Derived,OtherDerived>
+    operator*(const MatrixBase<OtherDerived> &other) const;
+
+#endif
+
+    template<typename OtherDerived>
+    EIGEN_DEVICE_FUNC 
+    const Product<Derived,OtherDerived,LazyProduct>
     lazyProduct(const MatrixBase<OtherDerived> &other) const;
 
     template<typename OtherDerived>
@@ -196,66 +189,103 @@ template<typename Derived> class MatrixBase
     void applyOnTheRight(const EigenBase<OtherDerived>& other);
 
     template<typename DiagonalDerived>
-    const DiagonalProduct<Derived, DiagonalDerived, OnTheRight>
+    EIGEN_DEVICE_FUNC
+    const Product<Derived, DiagonalDerived, LazyProduct>
     operator*(const DiagonalBase<DiagonalDerived> &diagonal) const;
 
     template<typename OtherDerived>
-    Scalar dot(const MatrixBase<OtherDerived>& other) const;
-    RealScalar squaredNorm() const;
-    RealScalar norm() const;
+    EIGEN_DEVICE_FUNC
+    typename internal::scalar_product_traits<typename internal::traits<Derived>::Scalar,typename internal::traits<OtherDerived>::Scalar>::ReturnType
+    dot(const MatrixBase<OtherDerived>& other) const;
+
+    EIGEN_DEVICE_FUNC RealScalar squaredNorm() const;
+    EIGEN_DEVICE_FUNC RealScalar norm() const;
     RealScalar stableNorm() const;
     RealScalar blueNorm() const;
     RealScalar hypotNorm() const;
-    const PlainObject normalized() const;
-    void normalize();
+    EIGEN_DEVICE_FUNC const PlainObject normalized() const;
+    EIGEN_DEVICE_FUNC void normalize();
 
-    const AdjointReturnType adjoint() const;
-    void adjointInPlace();
+    EIGEN_DEVICE_FUNC const AdjointReturnType adjoint() const;
+    EIGEN_DEVICE_FUNC void adjointInPlace();
 
-    Diagonal<Derived,0> diagonal();
-    const Diagonal<Derived,0> diagonal() const;
+    typedef Diagonal<Derived> DiagonalReturnType;
+    EIGEN_DEVICE_FUNC
+    DiagonalReturnType diagonal();
+    
+    typedef typename internal::add_const<Diagonal<const Derived> >::type ConstDiagonalReturnType;
+    EIGEN_DEVICE_FUNC
+    ConstDiagonalReturnType diagonal() const;
 
-    template<int Index> Diagonal<Derived,Index> diagonal();
-    template<int Index> const Diagonal<Derived,Index> diagonal() const;
+    template<int Index> struct DiagonalIndexReturnType { typedef Diagonal<Derived,Index> Type; };
+    template<int Index> struct ConstDiagonalIndexReturnType { typedef const Diagonal<const Derived,Index> Type; };
 
-    Diagonal<Derived, Dynamic> diagonal(Index index);
-    const Diagonal<Derived, Dynamic> diagonal(Index index) const;
+    template<int Index> 
+    EIGEN_DEVICE_FUNC
+    typename DiagonalIndexReturnType<Index>::Type diagonal();
 
-    template<unsigned int Mode> TriangularView<Derived, Mode> part();
-    template<unsigned int Mode> const TriangularView<Derived, Mode> part() const;
+    template<int Index>
+    EIGEN_DEVICE_FUNC
+    typename ConstDiagonalIndexReturnType<Index>::Type diagonal() const;
+    
+    typedef Diagonal<Derived,DynamicIndex> DiagonalDynamicIndexReturnType;
+    typedef typename internal::add_const<Diagonal<const Derived,DynamicIndex> >::type ConstDiagonalDynamicIndexReturnType;
 
-    template<unsigned int Mode> TriangularView<Derived, Mode> triangularView();
-    template<unsigned int Mode> const TriangularView<Derived, Mode> triangularView() const;
+    EIGEN_DEVICE_FUNC
+    DiagonalDynamicIndexReturnType diagonal(Index index);
+    EIGEN_DEVICE_FUNC
+    ConstDiagonalDynamicIndexReturnType diagonal(Index index) const;
 
-    template<unsigned int UpLo> SelfAdjointView<Derived, UpLo> selfadjointView();
-    template<unsigned int UpLo> const SelfAdjointView<Derived, UpLo> selfadjointView() const;
+    template<unsigned int Mode> struct TriangularViewReturnType { typedef TriangularView<Derived, Mode> Type; };
+    template<unsigned int Mode> struct ConstTriangularViewReturnType { typedef const TriangularView<const Derived, Mode> Type; };
+
+    template<unsigned int Mode>
+    EIGEN_DEVICE_FUNC
+    typename TriangularViewReturnType<Mode>::Type triangularView();
+    template<unsigned int Mode>
+    EIGEN_DEVICE_FUNC
+    typename ConstTriangularViewReturnType<Mode>::Type triangularView() const;
+
+    template<unsigned int UpLo> struct SelfAdjointViewReturnType { typedef SelfAdjointView<Derived, UpLo> Type; };
+    template<unsigned int UpLo> struct ConstSelfAdjointViewReturnType { typedef const SelfAdjointView<const Derived, UpLo> Type; };
+
+    template<unsigned int UpLo> 
+    EIGEN_DEVICE_FUNC
+    typename SelfAdjointViewReturnType<UpLo>::Type selfadjointView();
+    template<unsigned int UpLo>
+    EIGEN_DEVICE_FUNC
+    typename ConstSelfAdjointViewReturnType<UpLo>::Type selfadjointView() const;
 
     const SparseView<Derived> sparseView(const Scalar& m_reference = Scalar(0),
-                                         typename NumTraits<Scalar>::Real m_epsilon = NumTraits<Scalar>::dummy_precision()) const;
-    static const IdentityReturnType Identity();
-    static const IdentityReturnType Identity(Index rows, Index cols);
-    static const BasisReturnType Unit(Index size, Index i);
-    static const BasisReturnType Unit(Index i);
-    static const BasisReturnType UnitX();
-    static const BasisReturnType UnitY();
-    static const BasisReturnType UnitZ();
-    static const BasisReturnType UnitW();
+                                         const typename NumTraits<Scalar>::Real& m_epsilon = NumTraits<Scalar>::dummy_precision()) const;
+    EIGEN_DEVICE_FUNC static const IdentityReturnType Identity();
+    EIGEN_DEVICE_FUNC static const IdentityReturnType Identity(Index rows, Index cols);
+    EIGEN_DEVICE_FUNC static const BasisReturnType Unit(Index size, Index i);
+    EIGEN_DEVICE_FUNC static const BasisReturnType Unit(Index i);
+    EIGEN_DEVICE_FUNC static const BasisReturnType UnitX();
+    EIGEN_DEVICE_FUNC static const BasisReturnType UnitY();
+    EIGEN_DEVICE_FUNC static const BasisReturnType UnitZ();
+    EIGEN_DEVICE_FUNC static const BasisReturnType UnitW();
 
-    const DiagonalWrapper<Derived> asDiagonal() const;
+    EIGEN_DEVICE_FUNC
+    const DiagonalWrapper<const Derived> asDiagonal() const;
+    const PermutationWrapper<const Derived> asPermutation() const;
 
+    EIGEN_DEVICE_FUNC
     Derived& setIdentity();
+    EIGEN_DEVICE_FUNC
     Derived& setIdentity(Index rows, Index cols);
 
-    bool isIdentity(RealScalar prec = NumTraits<Scalar>::dummy_precision()) const;
-    bool isDiagonal(RealScalar prec = NumTraits<Scalar>::dummy_precision()) const;
+    bool isIdentity(const RealScalar& prec = NumTraits<Scalar>::dummy_precision()) const;
+    bool isDiagonal(const RealScalar& prec = NumTraits<Scalar>::dummy_precision()) const;
 
-    bool isUpperTriangular(RealScalar prec = NumTraits<Scalar>::dummy_precision()) const;
-    bool isLowerTriangular(RealScalar prec = NumTraits<Scalar>::dummy_precision()) const;
+    bool isUpperTriangular(const RealScalar& prec = NumTraits<Scalar>::dummy_precision()) const;
+    bool isLowerTriangular(const RealScalar& prec = NumTraits<Scalar>::dummy_precision()) const;
 
     template<typename OtherDerived>
     bool isOrthogonal(const MatrixBase<OtherDerived>& other,
-                      RealScalar prec = NumTraits<Scalar>::dummy_precision()) const;
-    bool isUnitary(RealScalar prec = NumTraits<Scalar>::dummy_precision()) const;
+                      const RealScalar& prec = NumTraits<Scalar>::dummy_precision()) const;
+    bool isUnitary(const RealScalar& prec = NumTraits<Scalar>::dummy_precision()) const;
 
     /** \returns true if each coefficients of \c *this and \a other are all exactly equal.
       * \warning When using floating point scalar values you probably should rather use a
@@ -275,40 +305,49 @@ template<typename Derived> class MatrixBase
 
     NoAlias<Derived,Eigen::MatrixBase > noalias();
 
-    inline const ForceAlignedAccess<Derived> forceAlignedAccess() const;
-    inline ForceAlignedAccess<Derived> forceAlignedAccess();
-    template<bool Enable> inline typename ei_makeconst<typename ei_meta_if<Enable,ForceAlignedAccess<Derived>,Derived&>::ret>::type forceAlignedAccessIf() const;
-    template<bool Enable> inline typename ei_meta_if<Enable,ForceAlignedAccess<Derived>,Derived&>::ret forceAlignedAccessIf();
+    // TODO forceAlignedAccess is temporarily disabled
+    // Need to find a nicer workaround.
+    inline const Derived& forceAlignedAccess() const { return derived(); }
+    inline Derived& forceAlignedAccess() { return derived(); }
+    template<bool Enable> inline const Derived& forceAlignedAccessIf() const { return derived(); }
+    template<bool Enable> inline Derived& forceAlignedAccessIf() { return derived(); }
 
-    Scalar trace() const;
+    EIGEN_DEVICE_FUNC Scalar trace() const;
 
-/////////// Array module ///////////
+    template<int p> EIGEN_DEVICE_FUNC RealScalar lpNorm() const;
 
-    template<int p> RealScalar lpNorm() const;
+    EIGEN_DEVICE_FUNC MatrixBase<Derived>& matrix() { return *this; }
+    EIGEN_DEVICE_FUNC const MatrixBase<Derived>& matrix() const { return *this; }
 
-    MatrixBase<Derived>& matrix() { return *this; }
-    const MatrixBase<Derived>& matrix() const { return *this; }
-
-    /** \returns an \link ArrayBase Array \endlink expression of this matrix
+    /** \returns an \link Eigen::ArrayBase Array \endlink expression of this matrix
       * \sa ArrayBase::matrix() */
-    ArrayWrapper<Derived> array() { return derived(); }
-    const ArrayWrapper<Derived> array() const { return derived(); }
+    EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE ArrayWrapper<Derived> array() { return ArrayWrapper<Derived>(derived()); }
+    /** \returns a const \link Eigen::ArrayBase Array \endlink expression of this matrix
+      * \sa ArrayBase::matrix() */
+    EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE const ArrayWrapper<const Derived> array() const { return ArrayWrapper<const Derived>(derived()); }
 
 /////////// LU module ///////////
 
-    const FullPivLU<PlainObject> fullPivLu() const;
-    const PartialPivLU<PlainObject> partialPivLu() const;
-    const PartialPivLU<PlainObject> lu() const;
-    const ei_inverse_impl<Derived> inverse() const;
+    EIGEN_DEVICE_FUNC
+    inline const FullPivLU<PlainObject> fullPivLu() const;
+    EIGEN_DEVICE_FUNC
+    inline const PartialPivLU<PlainObject> partialPivLu() const;
+
+    EIGEN_DEVICE_FUNC
+    inline const PartialPivLU<PlainObject> lu() const;
+
+    EIGEN_DEVICE_FUNC
+    inline const Inverse<Derived> inverse() const;
+    
     template<typename ResultType>
-    void computeInverseAndDetWithCheck(
+    inline void computeInverseAndDetWithCheck(
       ResultType& inverse,
       typename ResultType::Scalar& determinant,
       bool& invertible,
       const RealScalar& absDeterminantThreshold = NumTraits<Scalar>::dummy_precision()
     ) const;
     template<typename ResultType>
-    void computeInverseWithCheck(
+    inline void computeInverseWithCheck(
       ResultType& inverse,
       bool& invertible,
       const RealScalar& absDeterminantThreshold = NumTraits<Scalar>::dummy_precision()
@@ -317,45 +356,65 @@ template<typename Derived> class MatrixBase
 
 /////////// Cholesky module ///////////
 
-    const LLT<PlainObject>  llt() const;
-    const LDLT<PlainObject> ldlt() const;
+    inline const LLT<PlainObject>  llt() const;
+    inline const LDLT<PlainObject> ldlt() const;
 
 /////////// QR module ///////////
 
-    const HouseholderQR<PlainObject> householderQr() const;
-    const ColPivHouseholderQR<PlainObject> colPivHouseholderQr() const;
-    const FullPivHouseholderQR<PlainObject> fullPivHouseholderQr() const;
+    inline const HouseholderQR<PlainObject> householderQr() const;
+    inline const ColPivHouseholderQR<PlainObject> colPivHouseholderQr() const;
+    inline const FullPivHouseholderQR<PlainObject> fullPivHouseholderQr() const;
 
-    EigenvaluesReturnType eigenvalues() const;
-    RealScalar operatorNorm() const;
+/////////// Eigenvalues module ///////////
+
+    inline EigenvaluesReturnType eigenvalues() const;
+    inline RealScalar operatorNorm() const;
 
 /////////// SVD module ///////////
 
+    inline JacobiSVD<PlainObject> jacobiSvd(unsigned int computationOptions = 0) const;
+    inline BDCSVD<PlainObject>    bdcSvd(unsigned int computationOptions = 0) const;
+
 /////////// Geometry module ///////////
 
+    #ifndef EIGEN_PARSED_BY_DOXYGEN
+    /// \internal helper struct to form the return type of the cross product
+    template<typename OtherDerived> struct cross_product_return_type {
+      typedef typename internal::scalar_product_traits<typename internal::traits<Derived>::Scalar,typename internal::traits<OtherDerived>::Scalar>::ReturnType Scalar;
+      typedef Matrix<Scalar,MatrixBase::RowsAtCompileTime,MatrixBase::ColsAtCompileTime> type;
+    };
+    #endif // EIGEN_PARSED_BY_DOXYGEN
     template<typename OtherDerived>
-    PlainObject cross(const MatrixBase<OtherDerived>& other) const;
+    EIGEN_DEVICE_FUNC
+    inline typename cross_product_return_type<OtherDerived>::type
+    cross(const MatrixBase<OtherDerived>& other) const;
+    
     template<typename OtherDerived>
-    PlainObject cross3(const MatrixBase<OtherDerived>& other) const;
-    PlainObject unitOrthogonal(void) const;
-    Matrix<Scalar,3,1> eulerAngles(Index a0, Index a1, Index a2) const;
-    ScalarMultipleReturnType operator*(const UniformScaling<Scalar>& s) const;
+    EIGEN_DEVICE_FUNC
+    inline PlainObject cross3(const MatrixBase<OtherDerived>& other) const;
+    
+    EIGEN_DEVICE_FUNC
+    inline PlainObject unitOrthogonal(void) const;
+    
+    inline Matrix<Scalar,3,1> eulerAngles(Index a0, Index a1, Index a2) const;
+    
+    inline ScalarMultipleReturnType operator*(const UniformScaling<Scalar>& s) const;
+    // put this as separate enum value to work around possible GCC 4.3 bug (?)
+    enum { HomogeneousReturnTypeDirection = ColsAtCompileTime==1&&RowsAtCompileTime==1 ? ((internal::traits<Derived>::Flags&RowMajorBit)==RowMajorBit ? Horizontal : Vertical)
+                                          : ColsAtCompileTime==1 ? Vertical : Horizontal };
+    typedef Homogeneous<Derived, HomogeneousReturnTypeDirection> HomogeneousReturnType;
+    inline HomogeneousReturnType homogeneous() const;
+    
     enum {
       SizeMinusOne = SizeAtCompileTime==Dynamic ? Dynamic : SizeAtCompileTime-1
     };
-    typedef Block<Derived,
-                  ei_traits<Derived>::ColsAtCompileTime==1 ? SizeMinusOne : 1,
-                  ei_traits<Derived>::ColsAtCompileTime==1 ? 1 : SizeMinusOne> StartMinusOne;
-    typedef CwiseUnaryOp<ei_scalar_quotient1_op<typename ei_traits<Derived>::Scalar>,
-                StartMinusOne > HNormalizedReturnType;
+    typedef Block<const Derived,
+                  internal::traits<Derived>::ColsAtCompileTime==1 ? SizeMinusOne : 1,
+                  internal::traits<Derived>::ColsAtCompileTime==1 ? 1 : SizeMinusOne> ConstStartMinusOne;
+    typedef CwiseUnaryOp<internal::scalar_quotient1_op<typename internal::traits<Derived>::Scalar>,
+                const ConstStartMinusOne > HNormalizedReturnType;
 
-    HNormalizedReturnType hnormalized() const;
-
-    // put this as separate enum value to work around possible GCC 4.3 bug (?)
-    enum { HomogeneousReturnTypeDirection = ColsAtCompileTime==1?Vertical:Horizontal };
-    typedef Homogeneous<Derived, HomogeneousReturnTypeDirection> HomogeneousReturnType;
-
-    HomogeneousReturnType homogeneous() const;
+    inline const HNormalizedReturnType hnormalized() const;
 
 ////////// Householder module ///////////
 
@@ -375,69 +434,94 @@ template<typename Derived> class MatrixBase
 ///////// Jacobi module /////////
 
     template<typename OtherScalar>
-    void applyOnTheLeft(Index p, Index q, const PlanarRotation<OtherScalar>& j);
+    void applyOnTheLeft(Index p, Index q, const JacobiRotation<OtherScalar>& j);
     template<typename OtherScalar>
-    void applyOnTheRight(Index p, Index q, const PlanarRotation<OtherScalar>& j);
+    void applyOnTheRight(Index p, Index q, const JacobiRotation<OtherScalar>& j);
+
+///////// SparseCore module /////////
+
+    template<typename OtherDerived>
+    EIGEN_STRONG_INLINE const typename SparseMatrixBase<OtherDerived>::template CwiseProductDenseReturnType<Derived>::Type
+    cwiseProduct(const SparseMatrixBase<OtherDerived> &other) const
+    {
+      return other.cwiseProduct(derived());
+    }
 
 ///////// MatrixFunctions module /////////
 
-    typedef typename ei_stem_function<Scalar>::type StemFunction;
+    typedef typename internal::stem_function<Scalar>::type StemFunction;
     const MatrixExponentialReturnValue<Derived> exp() const;
     const MatrixFunctionReturnValue<Derived> matrixFunction(StemFunction f) const;
     const MatrixFunctionReturnValue<Derived> cosh() const;
     const MatrixFunctionReturnValue<Derived> sinh() const;
     const MatrixFunctionReturnValue<Derived> cos() const;
     const MatrixFunctionReturnValue<Derived> sin() const;
-
-#ifdef EIGEN2_SUPPORT
-    template<typename ProductDerived, typename Lhs, typename Rhs>
-    Derived& operator+=(const Flagged<ProductBase<ProductDerived, Lhs,Rhs>, 0,
-                                      EvalBeforeAssigningBit>& other);
-
-    template<typename ProductDerived, typename Lhs, typename Rhs>
-    Derived& operator-=(const Flagged<ProductBase<ProductDerived, Lhs,Rhs>, 0,
-                                      EvalBeforeAssigningBit>& other);
-
-    /** \deprecated because .lazy() is deprecated
-      * Overloaded for cache friendly product evaluation */
-    template<typename OtherDerived>
-    Derived& lazyAssign(const Flagged<OtherDerived, 0, EvalBeforeAssigningBit>& other)
-    { return lazyAssign(other._expression()); }
-
-    template<unsigned int Added>
-    const Flagged<Derived, Added, 0> marked() const;
-    const Flagged<Derived, 0, EvalBeforeAssigningBit> lazy() const;
-
-    inline const Cwise<Derived> cwise() const;
-    inline Cwise<Derived> cwise();
-
-    VectorBlock<Derived> start(Index size);
-    const VectorBlock<Derived> start(Index size) const;
-    VectorBlock<Derived> end(Index size);
-    const VectorBlock<Derived> end(Index size) const;
-    template<int Size> VectorBlock<Derived,Size> start();
-    template<int Size> const VectorBlock<Derived,Size> start() const;
-    template<int Size> VectorBlock<Derived,Size> end();
-    template<int Size> const VectorBlock<Derived,Size> end() const;
-
-    Minor<Derived> minor(Index row, Index col);
-    const Minor<Derived> minor(Index row, Index col) const;
-#endif
+    const MatrixSquareRootReturnValue<Derived> sqrt() const;
+    const MatrixLogarithmReturnValue<Derived> log() const;
+    const MatrixPowerReturnValue<Derived> pow(const RealScalar& p) const;
+    const MatrixComplexPowerReturnValue<Derived> pow(const std::complex<RealScalar>& p) const;
 
   protected:
-    MatrixBase() : Base() {}
+    EIGEN_DEVICE_FUNC MatrixBase() : Base() {}
 
   private:
-    explicit MatrixBase(int);
-    MatrixBase(int,int);
-    template<typename OtherDerived> explicit MatrixBase(const MatrixBase<OtherDerived>&);
+    EIGEN_DEVICE_FUNC explicit MatrixBase(int);
+    EIGEN_DEVICE_FUNC MatrixBase(int,int);
+    template<typename OtherDerived> EIGEN_DEVICE_FUNC explicit MatrixBase(const MatrixBase<OtherDerived>&);
   protected:
     // mixing arrays and matrices is not legal
-    template<typename OtherDerived> Derived& operator+=(const ArrayBase<OtherDerived>& array)
-    {EIGEN_STATIC_ASSERT(sizeof(typename OtherDerived::Scalar)==-1,YOU_CANNOT_MIX_ARRAYS_AND_MATRICES);}
+    template<typename OtherDerived> Derived& operator+=(const ArrayBase<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 ArrayBase<OtherDerived>& array)
-    {EIGEN_STATIC_ASSERT(sizeof(typename OtherDerived::Scalar)==-1,YOU_CANNOT_MIX_ARRAYS_AND_MATRICES);}
+    template<typename OtherDerived> Derived& operator-=(const ArrayBase<OtherDerived>& )
+    {EIGEN_STATIC_ASSERT(std::ptrdiff_t(sizeof(typename OtherDerived::Scalar))==-1,YOU_CANNOT_MIX_ARRAYS_AND_MATRICES); return *this;}
 };
 
+
+/***************************************************************************
+* Implementation of matrix base methods
+***************************************************************************/
+
+/** replaces \c *this by \c *this * \a other.
+  *
+  * \returns a reference to \c *this
+  *
+  * Example: \include MatrixBase_applyOnTheRight.cpp
+  * Output: \verbinclude MatrixBase_applyOnTheRight.out
+  */
+template<typename Derived>
+template<typename OtherDerived>
+inline Derived&
+MatrixBase<Derived>::operator*=(const EigenBase<OtherDerived> &other)
+{
+  other.derived().applyThisOnTheRight(derived());
+  return derived();
+}
+
+/** replaces \c *this by \c *this * \a other. It is equivalent to MatrixBase::operator*=().
+  *
+  * Example: \include MatrixBase_applyOnTheRight.cpp
+  * Output: \verbinclude MatrixBase_applyOnTheRight.out
+  */
+template<typename Derived>
+template<typename OtherDerived>
+inline void MatrixBase<Derived>::applyOnTheRight(const EigenBase<OtherDerived> &other)
+{
+  other.derived().applyThisOnTheRight(derived());
+}
+
+/** replaces \c *this by \a other * \c *this.
+  *
+  * Example: \include MatrixBase_applyOnTheLeft.cpp
+  * Output: \verbinclude MatrixBase_applyOnTheLeft.out
+  */
+template<typename Derived>
+template<typename OtherDerived>
+inline void MatrixBase<Derived>::applyOnTheLeft(const EigenBase<OtherDerived> &other)
+{
+  other.derived().applyThisOnTheLeft(derived());
+}
+
+} // end namespace Eigen
+
 #endif // EIGEN_MATRIXBASE_H

Eigen/src/Core/MatrixStorage.h

@@ -1,300 +0,0 @@
-// 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 Hauke Heibel <hauke.heibel@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_MATRIXSTORAGE_H
-#define EIGEN_MATRIXSTORAGE_H
-
-#ifdef EIGEN_DEBUG_MATRIX_CTOR
-  #define EIGEN_INT_DEBUG_MATRIX_CTOR EIGEN_DEBUG_MATRIX_CTOR;
-#else
-  #define EIGEN_INT_DEBUG_MATRIX_CTOR
-#endif
-
-struct ei_constructor_without_unaligned_array_assert {};
-
-/** \internal
-  * Static array. If the MatrixOptions 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 MatrixOptions,
-          int Alignment = (MatrixOptions&DontAlign) ? 0
-                        : (((Size*sizeof(T))%16)==0) ? 16
-                        : 0 >
-struct ei_matrix_array
-{
-  T array[Size];
-  ei_matrix_array() {}
-  ei_matrix_array(ei_constructor_without_unaligned_array_assert) {}
-};
-
-#ifdef EIGEN_DISABLE_UNALIGNED_ARRAY_ASSERT
-  #define EIGEN_MAKE_UNALIGNED_ARRAY_ASSERT(sizemask)
-#else
-  #define EIGEN_MAKE_UNALIGNED_ARRAY_ASSERT(sizemask) \
-    ei_assert((reinterpret_cast<size_t>(array) & sizemask) == 0 \
-              && "this assertion is explained here: " \
-              "http://eigen.tuxfamily.org/dox/UnalignedArrayAssert.html" \
-              " **** READ THIS WEB PAGE !!! ****");
-#endif
-
-template <typename T, int Size, int MatrixOptions>
-struct ei_matrix_array<T, Size, MatrixOptions, 16>
-{
-  EIGEN_ALIGN16 T array[Size];
-  ei_matrix_array() { EIGEN_MAKE_UNALIGNED_ARRAY_ASSERT(0xf) }
-  ei_matrix_array(ei_constructor_without_unaligned_array_assert) {}
-};
-
-template <typename T, int MatrixOptions, int Alignment>
-struct ei_matrix_array<T, 0, MatrixOptions, Alignment>
-{
-  EIGEN_ALIGN16 T array[1];
-  ei_matrix_array() {}
-  ei_matrix_array(ei_constructor_without_unaligned_array_assert) {}
-};
-
-/** \internal
-  *
-  * \class ei_matrix_storage
-  * \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 ei_matrix_storage;
-
-// purely fixed-size matrix
-template<typename T, int Size, int _Rows, int _Cols, int _Options> class ei_matrix_storage
-{
-    ei_matrix_array<T,Size,_Options> m_data;
-  public:
-    inline explicit ei_matrix_storage() {}
-    inline ei_matrix_storage(ei_constructor_without_unaligned_array_assert)
-      : m_data(ei_constructor_without_unaligned_array_assert()) {}
-    inline ei_matrix_storage(DenseIndex,DenseIndex,DenseIndex) {}
-    inline void swap(ei_matrix_storage& other) { std::swap(m_data,other.m_data); }
-    inline static DenseIndex rows(void) {return _Rows;}
-    inline static DenseIndex cols(void) {return _Cols;}
-    inline void conservativeResize(DenseIndex,DenseIndex,DenseIndex) {}
-    inline void resize(DenseIndex,DenseIndex,DenseIndex) {}
-    inline const T *data() const { return m_data.array; }
-    inline T *data() { return m_data.array; }
-};
-
-// null matrix
-template<typename T, int _Rows, int _Cols, int _Options> class ei_matrix_storage<T, 0, _Rows, _Cols, _Options>
-{
-  public:
-    inline explicit ei_matrix_storage() {}
-    inline ei_matrix_storage(ei_constructor_without_unaligned_array_assert) {}
-    inline ei_matrix_storage(DenseIndex,DenseIndex,DenseIndex) {}
-    inline void swap(ei_matrix_storage& ) {}
-    inline static DenseIndex rows(void) {return _Rows;}
-    inline static DenseIndex cols(void) {return _Cols;}
-    inline void conservativeResize(DenseIndex,DenseIndex,DenseIndex) {}
-    inline void resize(DenseIndex,DenseIndex,DenseIndex) {}
-    inline const T *data() const { return 0; }
-    inline T *data() { return 0; }
-};
-
-// dynamic-size matrix with fixed-size storage
-template<typename T, int Size, int _Options> class ei_matrix_storage<T, Size, Dynamic, Dynamic, _Options>
-{
-    ei_matrix_array<T,Size,_Options> m_data;
-    DenseIndex m_rows;
-    DenseIndex m_cols;
-  public:
-    inline explicit ei_matrix_storage() : m_rows(0), m_cols(0) {}
-    inline ei_matrix_storage(ei_constructor_without_unaligned_array_assert)
-      : m_data(ei_constructor_without_unaligned_array_assert()), m_rows(0), m_cols(0) {}
-    inline ei_matrix_storage(DenseIndex, DenseIndex rows, DenseIndex cols) : m_rows(rows), m_cols(cols) {}
-    inline void swap(ei_matrix_storage& other)
-    { std::swap(m_data,other.m_data); std::swap(m_rows,other.m_rows); std::swap(m_cols,other.m_cols); }
-    inline DenseIndex rows(void) const {return m_rows;}
-    inline DenseIndex cols(void) const {return m_cols;}
-    inline void conservativeResize(DenseIndex, DenseIndex rows, DenseIndex cols) { m_rows = rows; m_cols = cols; }
-    inline void resize(DenseIndex, DenseIndex rows, DenseIndex cols) { m_rows = rows; m_cols = cols; }
-    inline const T *data() const { return m_data.array; }
-    inline 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 ei_matrix_storage<T, Size, Dynamic, _Cols, _Options>
-{
-    ei_matrix_array<T,Size,_Options> m_data;
-    DenseIndex m_rows;
-  public:
-    inline explicit ei_matrix_storage() : m_rows(0) {}
-    inline ei_matrix_storage(ei_constructor_without_unaligned_array_assert)
-      : m_data(ei_constructor_without_unaligned_array_assert()), m_rows(0) {}
-    inline ei_matrix_storage(DenseIndex, DenseIndex rows, DenseIndex) : m_rows(rows) {}
-    inline void swap(ei_matrix_storage& other) { std::swap(m_data,other.m_data); std::swap(m_rows,other.m_rows); }
-    inline DenseIndex rows(void) const {return m_rows;}
-    inline DenseIndex cols(void) const {return _Cols;}
-    inline void conservativeResize(DenseIndex, DenseIndex rows, DenseIndex) { m_rows = rows; }
-    inline void resize(DenseIndex, DenseIndex rows, DenseIndex) { m_rows = rows; }
-    inline const T *data() const { return m_data.array; }
-    inline 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 ei_matrix_storage<T, Size, _Rows, Dynamic, _Options>
-{
-    ei_matrix_array<T,Size,_Options> m_data;
-    DenseIndex m_cols;
-  public:
-    inline explicit ei_matrix_storage() : m_cols(0) {}
-    inline ei_matrix_storage(ei_constructor_without_unaligned_array_assert)
-      : m_data(ei_constructor_without_unaligned_array_assert()), m_cols(0) {}
-    inline ei_matrix_storage(DenseIndex, DenseIndex, DenseIndex cols) : m_cols(cols) {}
-    inline void swap(ei_matrix_storage& other) { std::swap(m_data,other.m_data); std::swap(m_cols,other.m_cols); }
-    inline DenseIndex rows(void) const {return _Rows;}
-    inline DenseIndex cols(void) const {return m_cols;}
-    inline void conservativeResize(DenseIndex, DenseIndex, DenseIndex cols) { m_cols = cols; }
-    inline void resize(DenseIndex, DenseIndex, DenseIndex cols) { m_cols = cols; }
-    inline const T *data() const { return m_data.array; }
-    inline T *data() { return m_data.array; }
-};
-
-// purely dynamic matrix.
-template<typename T, int _Options> class ei_matrix_storage<T, Dynamic, Dynamic, Dynamic, _Options>
-{
-    T *m_data;
-    DenseIndex m_rows;
-    DenseIndex m_cols;
-  public:
-    inline explicit ei_matrix_storage() : m_data(0), m_rows(0), m_cols(0) {}
-    inline ei_matrix_storage(ei_constructor_without_unaligned_array_assert)
-       : m_data(0), m_rows(0), m_cols(0) {}
-    inline ei_matrix_storage(DenseIndex size, DenseIndex rows, DenseIndex cols)
-      : m_data(ei_conditional_aligned_new<T,(_Options&DontAlign)==0>(size)), m_rows(rows), m_cols(cols) 
-    { EIGEN_INT_DEBUG_MATRIX_CTOR }
-    inline ~ei_matrix_storage() { ei_conditional_aligned_delete<T,(_Options&DontAlign)==0>(m_data, m_rows*m_cols); }
-    inline void swap(ei_matrix_storage& other)
-    { std::swap(m_data,other.m_data); std::swap(m_rows,other.m_rows); std::swap(m_cols,other.m_cols); }
-    inline DenseIndex rows(void) const {return m_rows;}
-    inline DenseIndex cols(void) const {return m_cols;}
-    inline void conservativeResize(DenseIndex size, DenseIndex rows, DenseIndex cols)
-    {
-      m_data = ei_conditional_aligned_realloc_new<T,(_Options&DontAlign)==0>(m_data, size, m_rows*m_cols);
-      m_rows = rows;
-      m_cols = cols;
-    }
-    void resize(DenseIndex size, DenseIndex rows, DenseIndex cols)
-    {
-      if(size != m_rows*m_cols)
-      {
-        ei_conditional_aligned_delete<T,(_Options&DontAlign)==0>(m_data, m_rows*m_cols);
-        if (size)
-          m_data = ei_conditional_aligned_new<T,(_Options&DontAlign)==0>(size);
-        else
-          m_data = 0;
-        EIGEN_INT_DEBUG_MATRIX_CTOR
-      }
-      m_rows = rows;
-      m_cols = cols;
-    }
-    inline const T *data() const { return m_data; }
-    inline 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 ei_matrix_storage<T, Dynamic, _Rows, Dynamic, _Options>
-{
-    T *m_data;
-    DenseIndex m_cols;
-  public:
-    inline explicit ei_matrix_storage() : m_data(0), m_cols(0) {}
-    inline ei_matrix_storage(ei_constructor_without_unaligned_array_assert) : m_data(0), m_cols(0) {}
-    inline ei_matrix_storage(DenseIndex size, DenseIndex, DenseIndex cols) : m_data(ei_conditional_aligned_new<T,(_Options&DontAlign)==0>(size)), m_cols(cols)
-    { EIGEN_INT_DEBUG_MATRIX_CTOR }
-    inline ~ei_matrix_storage() { ei_conditional_aligned_delete<T,(_Options&DontAlign)==0>(m_data, _Rows*m_cols); }
-    inline void swap(ei_matrix_storage& other) { std::swap(m_data,other.m_data); std::swap(m_cols,other.m_cols); }
-    inline static DenseIndex rows(void) {return _Rows;}
-    inline DenseIndex cols(void) const {return m_cols;}
-    inline void conservativeResize(DenseIndex size, DenseIndex, DenseIndex cols)
-    {
-      m_data = ei_conditional_aligned_realloc_new<T,(_Options&DontAlign)==0>(m_data, size, _Rows*m_cols);
-      m_cols = cols;
-    }
-    EIGEN_STRONG_INLINE void resize(DenseIndex size, DenseIndex, DenseIndex cols)
-    {
-      if(size != _Rows*m_cols)
-      {
-        ei_conditional_aligned_delete<T,(_Options&DontAlign)==0>(m_data, _Rows*m_cols);
-        if (size)
-          m_data = ei_conditional_aligned_new<T,(_Options&DontAlign)==0>(size);
-        else
-          m_data = 0;
-        EIGEN_INT_DEBUG_MATRIX_CTOR
-      }
-      m_cols = cols;
-    }
-    inline const T *data() const { return m_data; }
-    inline 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 ei_matrix_storage<T, Dynamic, Dynamic, _Cols, _Options>
-{
-    T *m_data;
-    DenseIndex m_rows;
-  public:
-    inline explicit ei_matrix_storage() : m_data(0), m_rows(0) {}
-    inline ei_matrix_storage(ei_constructor_without_unaligned_array_assert) : m_data(0), m_rows(0) {}
-    inline ei_matrix_storage(DenseIndex size, DenseIndex rows, DenseIndex) : m_data(ei_conditional_aligned_new<T,(_Options&DontAlign)==0>(size)), m_rows(rows)
-    { EIGEN_INT_DEBUG_MATRIX_CTOR }
-    inline ~ei_matrix_storage() { ei_conditional_aligned_delete<T,(_Options&DontAlign)==0>(m_data, _Cols*m_rows); }
-    inline void swap(ei_matrix_storage& other) { std::swap(m_data,other.m_data); std::swap(m_rows,other.m_rows); }
-    inline DenseIndex rows(void) const {return m_rows;}
-    inline static DenseIndex cols(void) {return _Cols;}
-    inline void conservativeResize(DenseIndex size, DenseIndex rows, DenseIndex)
-    {
-      m_data = ei_conditional_aligned_realloc_new<T,(_Options&DontAlign)==0>(m_data, size, m_rows*_Cols);
-      m_rows = rows;
-    }
-    EIGEN_STRONG_INLINE void resize(DenseIndex size, DenseIndex rows, DenseIndex)
-    {
-      if(size != m_rows*_Cols)
-      {
-        ei_conditional_aligned_delete<T,(_Options&DontAlign)==0>(m_data, _Cols*m_rows);
-        if (size)
-          m_data = ei_conditional_aligned_new<T,(_Options&DontAlign)==0>(size);
-        else
-          m_data = 0;
-        EIGEN_INT_DEBUG_MATRIX_CTOR
-      }
-      m_rows = rows;
-    }
-    inline const T *data() const { return m_data; }
-    inline T *data() { return m_data; }
-};
-
-#endif // EIGEN_MATRIX_H

Eigen/src/Core/NestByValue.h

@@ -4,28 +4,15 @@
 // 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_NESTBYVALUE_H
 #define EIGEN_NESTBYVALUE_H
 
+namespace Eigen {
+
 /** \class NestByValue
   * \ingroup Core_Module
   *
@@ -38,41 +25,44 @@
   *
   * \sa MatrixBase::nestByValue()
   */
+
+namespace internal {
 template<typename ExpressionType>
-struct ei_traits<NestByValue<ExpressionType> > : public ei_traits<ExpressionType>
+struct traits<NestByValue<ExpressionType> > : public traits<ExpressionType>
 {};
+}
 
 template<typename ExpressionType> class NestByValue
-  : public ei_dense_xpr_base< NestByValue<ExpressionType> >::type
+  : public internal::dense_xpr_base< NestByValue<ExpressionType> >::type
 {
   public:
 
-    typedef typename ei_dense_xpr_base<NestByValue>::type Base;
+    typedef typename internal::dense_xpr_base<NestByValue>::type Base;
     EIGEN_DENSE_PUBLIC_INTERFACE(NestByValue)
 
-    inline NestByValue(const ExpressionType& matrix) : m_expression(matrix) {}
+    EIGEN_DEVICE_FUNC explicit inline NestByValue(const ExpressionType& matrix) : m_expression(matrix) {}
 
-    inline Index rows() const { return m_expression.rows(); }
-    inline Index cols() const { return m_expression.cols(); }
-    inline Index outerStride() const { return m_expression.outerStride(); }
-    inline Index innerStride() const { return m_expression.innerStride(); }
+    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(); }
 
-    inline const CoeffReturnType coeff(Index row, Index col) const
+    EIGEN_DEVICE_FUNC inline const CoeffReturnType coeff(Index row, Index col) const
     {
       return m_expression.coeff(row, col);
     }
 
-    inline Scalar& coeffRef(Index row, Index col)
+    EIGEN_DEVICE_FUNC inline Scalar& coeffRef(Index row, Index col)
     {
       return m_expression.const_cast_derived().coeffRef(row, col);
     }
 
-    inline const CoeffReturnType coeff(Index index) const
+    EIGEN_DEVICE_FUNC inline const CoeffReturnType coeff(Index index) const
     {
       return m_expression.coeff(index);
     }
 
-    inline Scalar& coeffRef(Index index)
+    EIGEN_DEVICE_FUNC inline Scalar& coeffRef(Index index)
     {
       return m_expression.const_cast_derived().coeffRef(index);
     }
@@ -101,7 +91,7 @@ template<typename ExpressionType> class NestByValue
       m_expression.const_cast_derived().template writePacket<LoadMode>(index, x);
     }
 
-    operator const ExpressionType&() const { return m_expression; }
+    EIGEN_DEVICE_FUNC operator const ExpressionType&() const { return m_expression; }
 
   protected:
     const ExpressionType m_expression;
@@ -116,4 +106,6 @@ DenseBase<Derived>::nestByValue() const
   return NestByValue<Derived>(derived());
 }
 
+} // end namespace Eigen
+
 #endif // EIGEN_NESTBYVALUE_H

Eigen/src/Core/NoAlias.h

@@ -3,28 +3,15 @@
 //
 // Copyright (C) 2009 Gael Guennebaud <gael.guennebaud@inria.fr>
 //
-// Eigen is free software; you can redistribute it and/or
-// modify it under the terms of the GNU Lesser General Public
-// License as published by the Free Software Foundation; either
-// version 3 of the License, or (at your option) any later version.
-//
-// Alternatively, you can redistribute it and/or
-// modify it under the terms of the GNU General Public License as
-// published by the Free Software Foundation; either version 2 of
-// the License, or (at your option) any later version.
-//
-// Eigen is distributed in the hope that it will be useful, but WITHOUT ANY
-// WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS
-// FOR A PARTICULAR PURPOSE. See the GNU Lesser General Public License or the
-// GNU General Public License for more details.
-//
-// You should have received a copy of the GNU Lesser General Public
-// License and a copy of the GNU General Public License along with
-// Eigen. If not, see <http://www.gnu.org/licenses/>.
+// 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_NOALIAS_H
 #define EIGEN_NOALIAS_H
 
+namespace Eigen {
+
 /** \class NoAlias
   * \ingroup Core_Module
   *
@@ -43,57 +30,40 @@
 template<typename ExpressionType, template <typename> class StorageBase>
 class NoAlias
 {
-    typedef typename ExpressionType::Scalar Scalar;
   public:
-    NoAlias(ExpressionType& expression) : m_expression(expression) {}
-
-    /** Behaves like MatrixBase::lazyAssign(other)
-      * \sa MatrixBase::lazyAssign() */
+    typedef typename ExpressionType::Scalar Scalar;
+    
+    explicit NoAlias(ExpressionType& expression) : m_expression(expression) {}
+    
     template<typename OtherDerived>
+    EIGEN_DEVICE_FUNC
     EIGEN_STRONG_INLINE ExpressionType& operator=(const StorageBase<OtherDerived>& other)
-    { return ei_assign_selector<ExpressionType,OtherDerived,false>::run(m_expression,other.derived()); }
-
-    /** \sa MatrixBase::operator+= */
+    {
+      call_assignment_no_alias(m_expression, other.derived(), internal::assign_op<Scalar>());
+      return m_expression;
+    }
+    
     template<typename OtherDerived>
+    EIGEN_DEVICE_FUNC
     EIGEN_STRONG_INLINE ExpressionType& operator+=(const StorageBase<OtherDerived>& other)
     {
-      typedef SelfCwiseBinaryOp<ei_scalar_sum_op<Scalar>, ExpressionType, OtherDerived> SelfAdder;
-      SelfAdder tmp(m_expression);
-      typedef typename ei_nested<OtherDerived>::type OtherDerivedNested;
-      typedef typename ei_cleantype<OtherDerivedNested>::type _OtherDerivedNested;
-      ei_assign_selector<SelfAdder,_OtherDerivedNested,false>::run(tmp,OtherDerivedNested(other.derived()));
+      call_assignment_no_alias(m_expression, other.derived(), internal::add_assign_op<Scalar>());
       return m_expression;
     }
-
-    /** \sa MatrixBase::operator-= */
+    
     template<typename OtherDerived>
+    EIGEN_DEVICE_FUNC
     EIGEN_STRONG_INLINE ExpressionType& operator-=(const StorageBase<OtherDerived>& other)
     {
-      typedef SelfCwiseBinaryOp<ei_scalar_difference_op<Scalar>, ExpressionType, OtherDerived> SelfAdder;
-      SelfAdder tmp(m_expression);
-      typedef typename ei_nested<OtherDerived>::type OtherDerivedNested;
-      typedef typename ei_cleantype<OtherDerivedNested>::type _OtherDerivedNested;
-      ei_assign_selector<SelfAdder,_OtherDerivedNested,false>::run(tmp,OtherDerivedNested(other.derived()));
+      call_assignment_no_alias(m_expression, other.derived(), internal::sub_assign_op<Scalar>());
       return m_expression;
     }
 
-#ifndef EIGEN_PARSED_BY_DOXYGEN
-    template<typename ProductDerived, typename Lhs, typename Rhs>
-    EIGEN_STRONG_INLINE ExpressionType& operator+=(const ProductBase<ProductDerived, Lhs,Rhs>& other)
-    { other.derived().addTo(m_expression); return m_expression; }
-
-    template<typename ProductDerived, typename Lhs, typename Rhs>
-    EIGEN_STRONG_INLINE ExpressionType& operator-=(const ProductBase<ProductDerived, Lhs,Rhs>& other)
-    { other.derived().subTo(m_expression); return m_expression; }
-
-    template<typename Lhs, typename Rhs, int NestingFlags>
-    EIGEN_STRONG_INLINE ExpressionType& operator+=(const CoeffBasedProduct<Lhs,Rhs,NestingFlags>& other)
-    { return m_expression.derived() += CoeffBasedProduct<Lhs,Rhs,NestByRefBit>(other.lhs(), other.rhs()); }
-
-    template<typename Lhs, typename Rhs, int NestingFlags>
-    EIGEN_STRONG_INLINE ExpressionType& operator-=(const CoeffBasedProduct<Lhs,Rhs,NestingFlags>& other)
-    { return m_expression.derived() -= CoeffBasedProduct<Lhs,Rhs,NestByRefBit>(other.lhs(), other.rhs()); }
-#endif
+    EIGEN_DEVICE_FUNC
+    ExpressionType& expression() const
+    {
+      return m_expression;
+    }
 
   protected:
     ExpressionType& m_expression;
@@ -130,7 +100,9 @@ class NoAlias
 template<typename Derived>
 NoAlias<Derived,MatrixBase> MatrixBase<Derived>::noalias()
 {
-  return derived();
+  return NoAlias<Derived, Eigen::MatrixBase >(derived());
 }
 
+} // end namespace Eigen
+
 #endif // EIGEN_NOALIAS_H

Eigen/src/Core/NumTraits.h

@@ -3,28 +3,15 @@
 //
 // Copyright (C) 2006-2010 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_NUMTRAITS_H
 #define EIGEN_NUMTRAITS_H
 
+namespace Eigen {
+
 /** \class NumTraits
   * \ingroup Core_Module
   *
@@ -40,7 +27,7 @@
   *     is a typedef to \a U.
   * \li A typedef \a NonInteger, giving the type that should be used for operations producing non-integral values,
   *     such as quotients, square roots, etc. If \a T is a floating-point type, then this typedef just gives
-  *     \a T again. Note however that many Eigen functions such as ei_sqrt simply refuse to
+  *     \a T again. Note however that many Eigen functions such as internal::sqrt simply refuse to
   *     take integers. Outside of a few cases, Eigen doesn't do automatic type promotion. Thus, this typedef is
   *     only intended as a helper for code that needs to explicitly promote types.
   * \li A typedef \a Nested giving the type to use to nest a value inside of the expression tree. If you don't know what
@@ -53,6 +40,8 @@
   *     to by move / add / mul instructions respectively, assuming the data is already stored in CPU registers.
   *     Stay vague here. No need to do architecture-specific stuff.
   * \li An enum value \a IsSigned. It is equal to \c 1 if \a T is a signed type and to 0 if \a T is unsigned.
+  * \li An enum value \a RequireInitialization. It is equal to \c 1 if the constructor of the numeric type \a T must
+  *     be called, and to 0 if it is safe not to call it. Default is 0 if \a T is an arithmetic type, and 1 otherwise.
   * \li An epsilon() function which, unlike std::numeric_limits::epsilon(), returns a \a Real instead of a \a T.
   * \li A dummy_precision() function returning a weak epsilon value. It is mainly used as a default
   *     value by the fuzzy comparison operators.
@@ -65,27 +54,54 @@ template<typename T> struct GenericNumTraits
     IsInteger = std::numeric_limits<T>::is_integer,
     IsSigned = std::numeric_limits<T>::is_signed,
     IsComplex = 0,
+    RequireInitialization = internal::is_arithmetic<T>::value ? 0 : 1,
     ReadCost = 1,
     AddCost = 1,
     MulCost = 1
   };
 
   typedef T Real;
-  typedef typename ei_meta_if<
+  typedef typename internal::conditional<
                      IsInteger,
-                     typename ei_meta_if<sizeof(T)<=2, float, double>::ret,
+                     typename internal::conditional<sizeof(T)<=2, float, double>::type,
                      T
-                   >::ret NonInteger;
+                   >::type NonInteger;
   typedef T Nested;
 
-  inline static Real epsilon() { return std::numeric_limits<T>::epsilon(); }
-  inline static Real dummy_precision()
+  EIGEN_DEVICE_FUNC
+  static inline Real epsilon()
+  {
+    #if defined(__CUDA_ARCH__)
+    return internal::device::numeric_limits<T>::epsilon();
+    #else
+    return std::numeric_limits<T>::epsilon();
+    #endif
+  }
+  EIGEN_DEVICE_FUNC
+  static inline Real dummy_precision()
   {
     // make sure to override this for floating-point types
     return Real(0);
   }
-  inline static T highest() { return std::numeric_limits<T>::max(); }
-  inline static T lowest()  { return IsInteger ? std::numeric_limits<T>::min() : (-std::numeric_limits<T>::max()); }
+
+
+  EIGEN_DEVICE_FUNC
+  static inline T highest() {
+#if defined(__CUDA_ARCH__)
+    return (internal::device::numeric_limits<T>::max)();
+#else
+    return (std::numeric_limits<T>::max)();
+#endif
+  }
+
+  EIGEN_DEVICE_FUNC
+  static inline T lowest()  {
+#if defined(__CUDA_ARCH__)
+    return IsInteger ? (internal::device::numeric_limits<T>::min)() : (-(internal::device::numeric_limits<T>::max)());
+#else
+    return IsInteger ? (std::numeric_limits<T>::min)() : (-(std::numeric_limits<T>::max)());
+#endif
+  }
 };
 
 template<typename T> struct NumTraits : GenericNumTraits<T>
@@ -94,12 +110,14 @@ template<typename T> struct NumTraits : GenericNumTraits<T>
 template<> struct NumTraits<float>
   : GenericNumTraits<float>
 {
-  inline static float dummy_precision() { return 1e-5f; }
+  EIGEN_DEVICE_FUNC
+  static inline float dummy_precision() { return 1e-5f; }
 };
 
 template<> struct NumTraits<double> : GenericNumTraits<double>
 {
-  inline static double dummy_precision() { return 1e-12; }
+  EIGEN_DEVICE_FUNC
+  static inline double dummy_precision() { return 1e-12; }
 };
 
 template<> struct NumTraits<long double>
@@ -114,13 +132,14 @@ template<typename _Real> struct NumTraits<std::complex<_Real> >
   typedef _Real Real;
   enum {
     IsComplex = 1,
+    RequireInitialization = NumTraits<_Real>::RequireInitialization,
     ReadCost = 2 * NumTraits<_Real>::ReadCost,
     AddCost = 2 * NumTraits<Real>::AddCost,
     MulCost = 4 * NumTraits<Real>::MulCost + 2 * NumTraits<Real>::AddCost
   };
 
-  inline static Real epsilon() { return NumTraits<Real>::epsilon(); }
-  inline static Real dummy_precision() { return NumTraits<Real>::dummy_precision(); }
+  static inline Real epsilon() { return NumTraits<Real>::epsilon(); }
+  static inline Real dummy_precision() { return NumTraits<Real>::dummy_precision(); }
 };
 
 template<typename Scalar, int Rows, int Cols, int Options, int MaxRows, int MaxCols>
@@ -137,12 +156,16 @@ struct NumTraits<Array<Scalar, Rows, Cols, Options, MaxRows, MaxCols> >
     IsComplex = NumTraits<Scalar>::IsComplex,
     IsInteger = NumTraits<Scalar>::IsInteger,
     IsSigned  = NumTraits<Scalar>::IsSigned,
-    ReadCost = ArrayType::SizeAtCompileTime==Dynamic ? Dynamic : ArrayType::SizeAtCompileTime * NumTraits<Scalar>::ReadCost,
-    AddCost  = ArrayType::SizeAtCompileTime==Dynamic ? Dynamic : ArrayType::SizeAtCompileTime * NumTraits<Scalar>::AddCost,
-    MulCost  = ArrayType::SizeAtCompileTime==Dynamic ? Dynamic : ArrayType::SizeAtCompileTime * NumTraits<Scalar>::MulCost
+    RequireInitialization = 1,
+    ReadCost = ArrayType::SizeAtCompileTime==Dynamic ? HugeCost : ArrayType::SizeAtCompileTime * NumTraits<Scalar>::ReadCost,
+    AddCost  = ArrayType::SizeAtCompileTime==Dynamic ? HugeCost : ArrayType::SizeAtCompileTime * NumTraits<Scalar>::AddCost,
+    MulCost  = ArrayType::SizeAtCompileTime==Dynamic ? HugeCost : ArrayType::SizeAtCompileTime * NumTraits<Scalar>::MulCost
   };
+  
+  static inline RealScalar epsilon() { return NumTraits<RealScalar>::epsilon(); }
+  static inline RealScalar dummy_precision() { return NumTraits<RealScalar>::dummy_precision(); }
 };
 
-
+} // end namespace Eigen
 
 #endif // EIGEN_NUMTRAITS_H

Eigen/src/Core/PermutationMatrix.h

@@ -2,39 +2,26 @@
 // for linear algebra.
 //
 // Copyright (C) 2009 Benoit Jacob <jacob.benoit.1@gmail.com>
-// Copyright (C) 2009 Gael Guennebaud <gael.guennebaud@inria.fr>
+// Copyright (C) 2009-2015 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_PERMUTATIONMATRIX_H
 #define EIGEN_PERMUTATIONMATRIX_H
 
-/** \class PermutationMatrix
+namespace Eigen { 
+
+/** \class PermutationBase
   * \ingroup Core_Module
   *
-  * \brief Permutation matrix
+  * \brief Base class for permutations
   *
-  * \param SizeAtCompileTime the number of rows/cols, or Dynamic
-  * \param MaxSizeAtCompileTime the maximum number of rows/cols, or Dynamic. This optional parameter defaults to SizeAtCompileTime. Most of the time, you should not have to specify it.
+  * \param Derived the derived class
   *
-  * This class represents a permutation matrix, internally stored as a vector of integers.
+  * This class is the base class for all expressions representing a permutation matrix,
+  * internally stored as a vector of integers.
   * The convention followed here is that if \f$ \sigma \f$ is a permutation, the corresponding permutation matrix
   * \f$ P_\sigma \f$ is such that if \f$ (e_1,\ldots,e_p) \f$ is the canonical basis, we have:
   *  \f[ P_\sigma(e_i) = e_{\sigma(i)}. \f]
@@ -44,51 +31,308 @@
   * Permutation matrices are square and invertible.
   *
   * Notice that in addition to the member functions and operators listed here, there also are non-member
-  * operator* to multiply a PermutationMatrix with any kind of matrix expression (MatrixBase) on either side.
+  * operator* to multiply any kind of permutation object with any kind of matrix expression (MatrixBase)
+  * on either side.
   *
-  * \sa class DiagonalMatrix
+  * \sa class PermutationMatrix, class PermutationWrapper
   */
-template<typename PermutationType, typename MatrixType, int Side, bool Transposed=false> struct ei_permut_matrix_product_retval;
 
-template<int SizeAtCompileTime, int MaxSizeAtCompileTime>
-struct ei_traits<PermutationMatrix<SizeAtCompileTime, MaxSizeAtCompileTime> >
- : ei_traits<Matrix<int,SizeAtCompileTime,SizeAtCompileTime,0,MaxSizeAtCompileTime,MaxSizeAtCompileTime> >
-{};
+namespace internal {
 
-template<int SizeAtCompileTime, int MaxSizeAtCompileTime>
-class PermutationMatrix : public EigenBase<PermutationMatrix<SizeAtCompileTime, MaxSizeAtCompileTime> >
+enum PermPermProduct_t {PermPermProduct};
+
+} // end namespace internal
+
+template<typename Derived>
+class PermutationBase : public EigenBase<Derived>
 {
+    typedef internal::traits<Derived> Traits;
+    typedef EigenBase<Derived> Base;
   public:
 
     #ifndef EIGEN_PARSED_BY_DOXYGEN
-    typedef ei_traits<PermutationMatrix> Traits;
-    typedef Matrix<int,SizeAtCompileTime,SizeAtCompileTime,0,MaxSizeAtCompileTime,MaxSizeAtCompileTime>
-            DenseMatrixType;
+    typedef typename Traits::IndicesType IndicesType;
     enum {
       Flags = Traits::Flags,
-      CoeffReadCost = Traits::CoeffReadCost,
       RowsAtCompileTime = Traits::RowsAtCompileTime,
       ColsAtCompileTime = Traits::ColsAtCompileTime,
       MaxRowsAtCompileTime = Traits::MaxRowsAtCompileTime,
       MaxColsAtCompileTime = Traits::MaxColsAtCompileTime
     };
-    typedef typename Traits::Scalar Scalar;
-    typedef typename Traits::Index Index;
+    typedef typename Traits::StorageIndex StorageIndex;
+    typedef Matrix<StorageIndex,RowsAtCompileTime,ColsAtCompileTime,0,MaxRowsAtCompileTime,MaxColsAtCompileTime>
+            DenseMatrixType;
+    typedef PermutationMatrix<IndicesType::SizeAtCompileTime,IndicesType::MaxSizeAtCompileTime,StorageIndex>
+            PlainPermutationType;
+    typedef PlainPermutationType PlainObject;
+    using Base::derived;
+    typedef Inverse<Derived> InverseReturnType;
+    typedef void Scalar;
     #endif
 
-    typedef Matrix<int, SizeAtCompileTime, 1, 0, MaxSizeAtCompileTime, 1> IndicesType;
+    /** Copies the other permutation into *this */
+    template<typename OtherDerived>
+    Derived& operator=(const PermutationBase<OtherDerived>& other)
+    {
+      indices() = other.indices();
+      return derived();
+    }
+
+    /** Assignment from the Transpositions \a tr */
+    template<typename OtherDerived>
+    Derived& operator=(const TranspositionsBase<OtherDerived>& tr)
+    {
+      setIdentity(tr.size());
+      for(Index k=size()-1; k>=0; --k)
+        applyTranspositionOnTheRight(k,tr.coeff(k));
+      return derived();
+    }
+
+    #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=.
+      */
+    Derived& operator=(const PermutationBase& other)
+    {
+      indices() = other.indices();
+      return derived();
+    }
+    #endif
+
+    /** \returns the number of rows */
+    inline Index rows() const { return Index(indices().size()); }
+
+    /** \returns the number of columns */
+    inline Index cols() const { return Index(indices().size()); }
+
+    /** \returns the size of a side of the respective square matrix, i.e., the number of indices */
+    inline Index size() const { return Index(indices().size()); }
+
+    #ifndef EIGEN_PARSED_BY_DOXYGEN
+    template<typename DenseDerived>
+    void evalTo(MatrixBase<DenseDerived>& other) const
+    {
+      other.setZero();
+      for (Index i=0; i<rows(); ++i)
+        other.coeffRef(indices().coeff(i),i) = typename DenseDerived::Scalar(1);
+    }
+    #endif
+
+    /** \returns a Matrix object initialized from this permutation matrix. Notice that it
+      * is inefficient to return this Matrix object by value. For efficiency, favor using
+      * the Matrix constructor taking EigenBase objects.
+      */
+    DenseMatrixType toDenseMatrix() const
+    {
+      return derived();
+    }
+
+    /** const version of indices(). */
+    const IndicesType& indices() const { return derived().indices(); }
+    /** \returns a reference to the stored array representing the permutation. */
+    IndicesType& indices() { return derived().indices(); }
+
+    /** Resizes to given size.
+      */
+    inline void resize(Index newSize)
+    {
+      indices().resize(newSize);
+    }
+
+    /** Sets *this to be the identity permutation matrix */
+    void setIdentity()
+    {
+      StorageIndex n = StorageIndex(size());
+      for(StorageIndex i = 0; i < n; ++i)
+        indices().coeffRef(i) = i;
+    }
+
+    /** Sets *this to be the identity permutation matrix of given size.
+      */
+    void setIdentity(Index newSize)
+    {
+      resize(newSize);
+      setIdentity();
+    }
+
+    /** Multiplies *this by the transposition \f$(ij)\f$ on the left.
+      *
+      * \returns a reference to *this.
+      *
+      * \warning This is much slower than applyTranspositionOnTheRight(Index,Index):
+      * this has linear complexity and requires a lot of branching.
+      *
+      * \sa applyTranspositionOnTheRight(Index,Index)
+      */
+    Derived& applyTranspositionOnTheLeft(Index i, Index j)
+    {
+      eigen_assert(i>=0 && j>=0 && i<size() && j<size());
+      for(Index k = 0; k < size(); ++k)
+      {
+        if(indices().coeff(k) == i) indices().coeffRef(k) = StorageIndex(j);
+        else if(indices().coeff(k) == j) indices().coeffRef(k) = StorageIndex(i);
+      }
+      return derived();
+    }
+
+    /** Multiplies *this by the transposition \f$(ij)\f$ on the right.
+      *
+      * \returns a reference to *this.
+      *
+      * This is a fast operation, it only consists in swapping two indices.
+      *
+      * \sa applyTranspositionOnTheLeft(Index,Index)
+      */
+    Derived& applyTranspositionOnTheRight(Index i, Index j)
+    {
+      eigen_assert(i>=0 && j>=0 && i<size() && j<size());
+      std::swap(indices().coeffRef(i), indices().coeffRef(j));
+      return derived();
+    }
+
+    /** \returns the inverse permutation matrix.
+      *
+      * \note \note_try_to_help_rvo
+      */
+    inline InverseReturnType inverse() const
+    { return InverseReturnType(derived()); }
+    /** \returns the tranpose permutation matrix.
+      *
+      * \note \note_try_to_help_rvo
+      */
+    inline InverseReturnType transpose() const
+    { return InverseReturnType(derived()); }
+
+    /**** multiplication helpers to hopefully get RVO ****/
+
+  
+#ifndef EIGEN_PARSED_BY_DOXYGEN
+  protected:
+    template<typename OtherDerived>
+    void assignTranspose(const PermutationBase<OtherDerived>& other)
+    {
+      for (Index i=0; i<rows();++i) indices().coeffRef(other.indices().coeff(i)) = i;
+    }
+    template<typename Lhs,typename Rhs>
+    void assignProduct(const Lhs& lhs, const Rhs& rhs)
+    {
+      eigen_assert(lhs.cols() == rhs.rows());
+      for (Index i=0; i<rows();++i) indices().coeffRef(i) = lhs.indices().coeff(rhs.indices().coeff(i));
+    }
+#endif
+
+  public:
+
+    /** \returns the product permutation matrix.
+      *
+      * \note \note_try_to_help_rvo
+      */
+    template<typename Other>
+    inline PlainPermutationType operator*(const PermutationBase<Other>& other) const
+    { return PlainPermutationType(internal::PermPermProduct, derived(), other.derived()); }
+
+    /** \returns the product of a permutation with another inverse permutation.
+      *
+      * \note \note_try_to_help_rvo
+      */
+    template<typename Other>
+    inline PlainPermutationType operator*(const InverseImpl<Other,PermutationStorage>& other) const
+    { return PlainPermutationType(internal::PermPermProduct, *this, other.eval()); }
+
+    /** \returns the product of an inverse permutation with another permutation.
+      *
+      * \note \note_try_to_help_rvo
+      */
+    template<typename Other> friend
+    inline PlainPermutationType operator*(const InverseImpl<Other, PermutationStorage>& other, const PermutationBase& perm)
+    { return PlainPermutationType(internal::PermPermProduct, other.eval(), perm); }
+    
+    /** \returns the determinant of the permutation matrix, which is either 1 or -1 depending on the parity of the permutation.
+      *
+      * This function is O(\c n) procedure allocating a buffer of \c n booleans.
+      */
+    Index determinant() const
+    {
+      Index res = 1;
+      Index n = size();
+      Matrix<bool,RowsAtCompileTime,1,0,MaxRowsAtCompileTime> mask(n);
+      mask.fill(false);
+      Index r = 0;
+      while(r < n)
+      {
+        // search for the next seed
+        while(r<n && mask[r]) r++;
+        if(r>=n)
+          break;
+        // we got one, let's follow it until we are back to the seed
+        Index k0 = r++;
+        mask.coeffRef(k0) = true;
+        for(Index k=indices().coeff(k0); k!=k0; k=indices().coeff(k))
+        {
+          mask.coeffRef(k) = true;
+          res = -res;
+        }
+      }
+      return res;
+    }
+
+  protected:
+
+};
+
+/** \class PermutationMatrix
+  * \ingroup Core_Module
+  *
+  * \brief Permutation matrix
+  *
+  * \param SizeAtCompileTime the number of rows/cols, or Dynamic
+  * \param MaxSizeAtCompileTime the maximum number of rows/cols, or Dynamic. This optional parameter defaults to SizeAtCompileTime. Most of the time, you should not have to specify it.
+  * \param StorageIndex the integer type of the indices
+  *
+  * This class represents a permutation matrix, internally stored as a vector of integers.
+  *
+  * \sa class PermutationBase, class PermutationWrapper, class DiagonalMatrix
+  */
+
+namespace internal {
+template<int SizeAtCompileTime, int MaxSizeAtCompileTime, typename _StorageIndex>
+struct traits<PermutationMatrix<SizeAtCompileTime, MaxSizeAtCompileTime, _StorageIndex> >
+ : traits<Matrix<_StorageIndex,SizeAtCompileTime,SizeAtCompileTime,0,MaxSizeAtCompileTime,MaxSizeAtCompileTime> >
+{
+  typedef PermutationStorage StorageKind;
+  typedef Matrix<_StorageIndex, SizeAtCompileTime, 1, 0, MaxSizeAtCompileTime, 1> IndicesType;
+  typedef _StorageIndex StorageIndex;
+  typedef void Scalar;
+};
+}
+
+template<int SizeAtCompileTime, int MaxSizeAtCompileTime, typename _StorageIndex>
+class PermutationMatrix : public PermutationBase<PermutationMatrix<SizeAtCompileTime, MaxSizeAtCompileTime, _StorageIndex> >
+{
+    typedef PermutationBase<PermutationMatrix> Base;
+    typedef internal::traits<PermutationMatrix> Traits;
+  public:
+
+    typedef const PermutationMatrix& Nested;
+
+    #ifndef EIGEN_PARSED_BY_DOXYGEN
+    typedef typename Traits::IndicesType IndicesType;
+    typedef typename Traits::StorageIndex StorageIndex;
+    #endif
 
     inline PermutationMatrix()
     {}
 
     /** Constructs an uninitialized permutation matrix of given size.
       */
-    inline PermutationMatrix(int size) : m_indices(size)
-    {}
+    explicit inline PermutationMatrix(Index size) : m_indices(size)
+    {
+      eigen_internal_assert(size <= NumTraits<StorageIndex>::highest());
+    }
 
     /** Copy constructor. */
-    template<int OtherSize, int OtherMaxSize>
-    inline PermutationMatrix(const PermutationMatrix<OtherSize, OtherMaxSize>& other)
+    template<typename OtherDerived>
+    inline PermutationMatrix(const PermutationBase<OtherDerived>& other)
       : m_indices(other.indices()) {}
 
     #ifndef EIGEN_PARSED_BY_DOXYGEN
@@ -109,29 +353,26 @@ class PermutationMatrix : public EigenBase<PermutationMatrix<SizeAtCompileTime,
     {}
 
     /** Convert the Transpositions \a tr to a permutation matrix */
-    template<int OtherSize, int OtherMaxSize>
-    explicit PermutationMatrix(const Transpositions<OtherSize,OtherMaxSize>& tr)
+    template<typename Other>
+    explicit PermutationMatrix(const TranspositionsBase<Other>& tr)
       : m_indices(tr.size())
     {
       *this = tr;
     }
 
     /** Copies the other permutation into *this */
-    template<int OtherSize, int OtherMaxSize>
-    PermutationMatrix& operator=(const PermutationMatrix<OtherSize, OtherMaxSize>& other)
+    template<typename Other>
+    PermutationMatrix& operator=(const PermutationBase<Other>& other)
     {
       m_indices = other.indices();
       return *this;
     }
 
     /** Assignment from the Transpositions \a tr */
-    template<int OtherSize, int OtherMaxSize>
-    PermutationMatrix& operator=(const Transpositions<OtherSize,OtherMaxSize>& tr)
+    template<typename Other>
+    PermutationMatrix& operator=(const TranspositionsBase<Other>& tr)
     {
-      setIdentity(tr.size());
-      for(Index k=size()-1; k>=0; --k)
-        applyTranspositionOnTheRight(k,tr.coeff(k));
-      return *this;
+      return Base::operator=(tr.derived());
     }
 
     #ifndef EIGEN_PARSED_BY_DOXYGEN
@@ -145,326 +386,251 @@ class PermutationMatrix : public EigenBase<PermutationMatrix<SizeAtCompileTime,
     }
     #endif
 
-    /** \returns the number of rows */
-    inline Index rows() const { return m_indices.size(); }
-
-    /** \returns the number of columns */
-    inline Index cols() const { return m_indices.size(); }
-
-    /** \returns the size of a side of the respective square matrix, i.e., the number of indices */
-    inline Index size() const { return m_indices.size(); }
-
-    #ifndef EIGEN_PARSED_BY_DOXYGEN
-    template<typename DenseDerived>
-    void evalTo(MatrixBase<DenseDerived>& other) const
-    {
-      other.setZero();
-      for (int i=0; i<rows();++i)
-        other.coeffRef(m_indices.coeff(i),i) = typename DenseDerived::Scalar(1);
-    }
-    #endif
-
-    /** \returns a Matrix object initialized from this permutation matrix. Notice that it
-      * is inefficient to return this Matrix object by value. For efficiency, favor using
-      * the Matrix constructor taking EigenBase objects.
-      */
-    DenseMatrixType toDenseMatrix() const
-    {
-      return *this;
-    }
-
     /** const version of indices(). */
     const IndicesType& indices() const { return m_indices; }
     /** \returns a reference to the stored array representing the permutation. */
     IndicesType& indices() { return m_indices; }
 
-    /** Resizes to given size.
-      */
-    inline void resize(Index size)
-    {
-      m_indices.resize(size);
-    }
-
-    /** Sets *this to be the identity permutation matrix */
-    void setIdentity()
-    {
-      for(Index i = 0; i < m_indices.size(); ++i)
-        m_indices.coeffRef(i) = i;
-    }
-
-    /** Sets *this to be the identity permutation matrix of given size.
-      */
-    void setIdentity(Index size)
-    {
-      resize(size);
-      setIdentity();
-    }
-
-    /** Multiplies *this by the transposition \f$(ij)\f$ on the left.
-      *
-      * \returns a reference to *this.
-      *
-      * \warning This is much slower than applyTranspositionOnTheRight(int,int):
-      * this has linear complexity and requires a lot of branching.
-      *
-      * \sa applyTranspositionOnTheRight(int,int)
-      */
-    PermutationMatrix& applyTranspositionOnTheLeft(Index i, Index j)
-    {
-      ei_assert(i>=0 && j>=0 && i<m_indices.size() && j<m_indices.size());
-      for(Index k = 0; k < m_indices.size(); ++k)
-      {
-        if(m_indices.coeff(k) == i) m_indices.coeffRef(k) = j;
-        else if(m_indices.coeff(k) == j) m_indices.coeffRef(k) = i;
-      }
-      return *this;
-    }
-
-    /** Multiplies *this by the transposition \f$(ij)\f$ on the right.
-      *
-      * \returns a reference to *this.
-      *
-      * This is a fast operation, it only consists in swapping two indices.
-      *
-      * \sa applyTranspositionOnTheLeft(int,int)
-      */
-    PermutationMatrix& applyTranspositionOnTheRight(Index i, Index j)
-    {
-      ei_assert(i>=0 && j>=0 && i<m_indices.size() && j<m_indices.size());
-      std::swap(m_indices.coeffRef(i), m_indices.coeffRef(j));
-      return *this;
-    }
-
-    /** \returns the inverse permutation matrix.
-      *
-      * \note \note_try_to_help_rvo
-      */
-    inline Transpose<PermutationMatrix> inverse() const
-    { return *this; }
-    /** \returns the tranpose permutation matrix.
-      *
-      * \note \note_try_to_help_rvo
-      */
-    inline Transpose<PermutationMatrix> transpose() const
-    { return *this; }
 
     /**** multiplication helpers to hopefully get RVO ****/
 
 #ifndef EIGEN_PARSED_BY_DOXYGEN
-    template<int OtherSize, int OtherMaxSize>
-    PermutationMatrix(const Transpose<PermutationMatrix<OtherSize,OtherMaxSize> >& other)
-      : m_indices(other.nestedPermutation().size())
+    template<typename Other>
+    PermutationMatrix(const InverseImpl<Other,PermutationStorage>& other)
+      : m_indices(other.derived().nestedExpression().size())
     {
-      for (int i=0; i<rows();++i) m_indices.coeffRef(other.nestedPermutation().indices().coeff(i)) = i;
+      eigen_internal_assert(m_indices.size() <= NumTraits<StorageIndex>::highest());
+      StorageIndex end = StorageIndex(m_indices.size());
+      for (StorageIndex i=0; i<end;++i)
+        m_indices.coeffRef(other.derived().nestedExpression().indices().coeff(i)) = i;
     }
-  protected:
-    enum Product_t {Product};
-    PermutationMatrix(Product_t, const PermutationMatrix& lhs, const PermutationMatrix& rhs)
-      : m_indices(lhs.m_indices.size())
+    template<typename Lhs,typename Rhs>
+    PermutationMatrix(internal::PermPermProduct_t, const Lhs& lhs, const Rhs& rhs)
+      : m_indices(lhs.indices().size())
     {
-      ei_assert(lhs.cols() == rhs.rows());
-      for (int i=0; i<rows();++i) m_indices.coeffRef(i) = lhs.m_indices.coeff(rhs.m_indices.coeff(i));
+      Base::assignProduct(lhs,rhs);
     }
 #endif
 
-  public:
-
-    /** \returns the product permutation matrix.
-      *
-      * \note \note_try_to_help_rvo
-      */
-    template<int OtherSize, int OtherMaxSize>
-    inline PermutationMatrix operator*(const PermutationMatrix<OtherSize, OtherMaxSize>& other) const
-    { return PermutationMatrix(Product, *this, other); }
-
-    /** \returns the product of a permutation with another inverse permutation.
-      *
-      * \note \note_try_to_help_rvo
-      */
-    template<int OtherSize, int OtherMaxSize>
-    inline PermutationMatrix operator*(const Transpose<PermutationMatrix<OtherSize,OtherMaxSize> >& other) const
-    { return PermutationMatrix(Product, *this, other.eval()); }
-
-    /** \returns the product of an inverse permutation with another permutation.
-      *
-      * \note \note_try_to_help_rvo
-      */
-    template<int OtherSize, int OtherMaxSize> friend
-    inline PermutationMatrix operator*(const Transpose<PermutationMatrix<OtherSize,OtherMaxSize> >& other, const PermutationMatrix& perm)
-    { return PermutationMatrix(Product, other.eval(), perm); }
-
   protected:
 
     IndicesType m_indices;
 };
 
+
+namespace internal {
+template<int SizeAtCompileTime, int MaxSizeAtCompileTime, typename _StorageIndex, int _PacketAccess>
+struct traits<Map<PermutationMatrix<SizeAtCompileTime, MaxSizeAtCompileTime, _StorageIndex>,_PacketAccess> >
+ : traits<Matrix<_StorageIndex,SizeAtCompileTime,SizeAtCompileTime,0,MaxSizeAtCompileTime,MaxSizeAtCompileTime> >
+{
+  typedef PermutationStorage StorageKind;
+  typedef Map<const Matrix<_StorageIndex, SizeAtCompileTime, 1, 0, MaxSizeAtCompileTime, 1>, _PacketAccess> IndicesType;
+  typedef _StorageIndex StorageIndex;
+  typedef void Scalar;
+};
+}
+
+template<int SizeAtCompileTime, int MaxSizeAtCompileTime, typename _StorageIndex, int _PacketAccess>
+class Map<PermutationMatrix<SizeAtCompileTime, MaxSizeAtCompileTime, _StorageIndex>,_PacketAccess>
+  : public PermutationBase<Map<PermutationMatrix<SizeAtCompileTime, MaxSizeAtCompileTime, _StorageIndex>,_PacketAccess> >
+{
+    typedef PermutationBase<Map> Base;
+    typedef internal::traits<Map> Traits;
+  public:
+
+    #ifndef EIGEN_PARSED_BY_DOXYGEN
+    typedef typename Traits::IndicesType IndicesType;
+    typedef typename IndicesType::Scalar StorageIndex;
+    #endif
+
+    inline Map(const StorageIndex* indicesPtr)
+      : m_indices(indicesPtr)
+    {}
+
+    inline Map(const StorageIndex* indicesPtr, Index size)
+      : m_indices(indicesPtr,size)
+    {}
+
+    /** Copies the other permutation into *this */
+    template<typename Other>
+    Map& operator=(const PermutationBase<Other>& other)
+    { return Base::operator=(other.derived()); }
+
+    /** Assignment from the Transpositions \a tr */
+    template<typename Other>
+    Map& operator=(const TranspositionsBase<Other>& tr)
+    { return Base::operator=(tr.derived()); }
+
+    #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=.
+      */
+    Map& operator=(const Map& other)
+    {
+      m_indices = other.m_indices;
+      return *this;
+    }
+    #endif
+
+    /** const version of indices(). */
+    const IndicesType& indices() const { return m_indices; }
+    /** \returns a reference to the stored array representing the permutation. */
+    IndicesType& indices() { return m_indices; }
+
+  protected:
+
+    IndicesType m_indices;
+};
+
+/** \class PermutationWrapper
+  * \ingroup Core_Module
+  *
+  * \brief Class to view a vector of integers as a permutation matrix
+  *
+  * \param _IndicesType the type of the vector of integer (can be any compatible expression)
+  *
+  * This class allows to view any vector expression of integers as a permutation matrix.
+  *
+  * \sa class PermutationBase, class PermutationMatrix
+  */
+
+template<typename _IndicesType> class TranspositionsWrapper;
+namespace internal {
+template<typename _IndicesType>
+struct traits<PermutationWrapper<_IndicesType> >
+{
+  typedef PermutationStorage StorageKind;
+  typedef void Scalar;
+  typedef typename _IndicesType::Scalar StorageIndex;
+  typedef _IndicesType IndicesType;
+  enum {
+    RowsAtCompileTime = _IndicesType::SizeAtCompileTime,
+    ColsAtCompileTime = _IndicesType::SizeAtCompileTime,
+    MaxRowsAtCompileTime = IndicesType::MaxSizeAtCompileTime,
+    MaxColsAtCompileTime = IndicesType::MaxSizeAtCompileTime,
+    Flags = 0
+  };
+};
+}
+
+template<typename _IndicesType>
+class PermutationWrapper : public PermutationBase<PermutationWrapper<_IndicesType> >
+{
+    typedef PermutationBase<PermutationWrapper> Base;
+    typedef internal::traits<PermutationWrapper> Traits;
+  public:
+
+    #ifndef EIGEN_PARSED_BY_DOXYGEN
+    typedef typename Traits::IndicesType IndicesType;
+    #endif
+
+    inline PermutationWrapper(const IndicesType& indices)
+      : m_indices(indices)
+    {}
+
+    /** const version of indices(). */
+    const typename internal::remove_all<typename IndicesType::Nested>::type&
+    indices() const { return m_indices; }
+
+  protected:
+
+    typename IndicesType::Nested m_indices;
+};
+
+
 /** \returns the matrix with the permutation applied to the columns.
   */
-template<typename Derived, int SizeAtCompileTime, int MaxSizeAtCompileTime>
-inline const ei_permut_matrix_product_retval<PermutationMatrix<SizeAtCompileTime, MaxSizeAtCompileTime>, Derived, OnTheRight>
-operator*(const MatrixBase<Derived>& matrix,
-          const PermutationMatrix<SizeAtCompileTime, MaxSizeAtCompileTime> &permutation)
+template<typename MatrixDerived, typename PermutationDerived>
+EIGEN_DEVICE_FUNC
+const Product<MatrixDerived, PermutationDerived, AliasFreeProduct>
+operator*(const MatrixBase<MatrixDerived> &matrix,
+          const PermutationBase<PermutationDerived>& permutation)
 {
-  return ei_permut_matrix_product_retval
-           <PermutationMatrix<SizeAtCompileTime, MaxSizeAtCompileTime>, Derived, OnTheRight>
-           (permutation, matrix.derived());
+  return Product<MatrixDerived, PermutationDerived, AliasFreeProduct>
+            (matrix.derived(), permutation.derived());
 }
 
 /** \returns the matrix with the permutation applied to the rows.
   */
-template<typename Derived, int SizeAtCompileTime, int MaxSizeAtCompileTime>
-inline const ei_permut_matrix_product_retval
-               <PermutationMatrix<SizeAtCompileTime, MaxSizeAtCompileTime>, Derived, OnTheLeft>
-operator*(const PermutationMatrix<SizeAtCompileTime, MaxSizeAtCompileTime> &permutation,
-          const MatrixBase<Derived>& matrix)
+template<typename PermutationDerived, typename MatrixDerived>
+EIGEN_DEVICE_FUNC
+const Product<PermutationDerived, MatrixDerived, AliasFreeProduct>
+operator*(const PermutationBase<PermutationDerived> &permutation,
+          const MatrixBase<MatrixDerived>& matrix)
 {
-  return ei_permut_matrix_product_retval
-           <PermutationMatrix<SizeAtCompileTime, MaxSizeAtCompileTime>, Derived, OnTheLeft>
-           (permutation, matrix.derived());
+  return Product<PermutationDerived, MatrixDerived, AliasFreeProduct>
+            (permutation.derived(), matrix.derived());
 }
 
-template<typename PermutationType, typename MatrixType, int Side, bool Transposed>
-struct ei_traits<ei_permut_matrix_product_retval<PermutationType, MatrixType, Side, Transposed> >
+
+template<typename PermutationType>
+class InverseImpl<PermutationType, PermutationStorage>
+  : public EigenBase<Inverse<PermutationType> >
 {
-  typedef typename MatrixType::PlainObject ReturnType;
-};
-
-template<typename PermutationType, typename MatrixType, int Side, bool Transposed>
-struct ei_permut_matrix_product_retval
- : public ReturnByValue<ei_permut_matrix_product_retval<PermutationType, MatrixType, Side, Transposed> >
-{
-    typedef typename ei_cleantype<typename MatrixType::Nested>::type MatrixTypeNestedCleaned;
-
-    ei_permut_matrix_product_retval(const PermutationType& perm, const MatrixType& matrix)
-      : m_permutation(perm), m_matrix(matrix)
-    {}
-
-    inline int rows() const { return m_matrix.rows(); }
-    inline int cols() const { return m_matrix.cols(); }
-
-    template<typename Dest> inline void evalTo(Dest& dst) const
-    {
-      const int n = Side==OnTheLeft ? rows() : cols();
-
-      if(ei_is_same_type<MatrixTypeNestedCleaned,Dest>::ret && ei_extract_data(dst) == ei_extract_data(m_matrix))
-      {
-        // apply the permutation inplace
-        Matrix<bool,PermutationType::RowsAtCompileTime,1,0,PermutationType::MaxRowsAtCompileTime> mask(m_permutation.size());
-        mask.fill(false);
-        int r = 0;
-        while(r < m_permutation.size())
-        {
-          // search for the next seed
-          while(r<m_permutation.size() && mask[r]) r++;
-          if(r>=m_permutation.size())
-            break;
-          // we got one, let's follow it until we are back to the seed
-          int k0 = r++;
-          int kPrev = k0;
-          mask.coeffRef(k0) = true;
-          for(int k=m_permutation.indices().coeff(k0); k!=k0; k=m_permutation.indices().coeff(k))
-          {
-                  Block<Dest, Side==OnTheLeft ? 1 : Dest::RowsAtCompileTime, Side==OnTheRight ? 1 : Dest::ColsAtCompileTime>(dst, k)
-            .swap(Block<Dest, Side==OnTheLeft ? 1 : Dest::RowsAtCompileTime, Side==OnTheRight ? 1 : Dest::ColsAtCompileTime>
-                       (dst,((Side==OnTheLeft) ^ Transposed) ? k0 : kPrev));
-
-            mask.coeffRef(k) = true;
-            kPrev = k;
-          }
-        }
-      }
-      else
-      {
-        for(int i = 0; i < n; ++i)
-        {
-          Block<Dest, Side==OnTheLeft ? 1 : Dest::RowsAtCompileTime, Side==OnTheRight ? 1 : Dest::ColsAtCompileTime>
-               (dst, ((Side==OnTheLeft) ^ Transposed) ? m_permutation.indices().coeff(i) : i)
-
-          =
-
-          Block<MatrixTypeNestedCleaned,Side==OnTheLeft ? 1 : MatrixType::RowsAtCompileTime,Side==OnTheRight ? 1 : MatrixType::ColsAtCompileTime>
-               (m_matrix, ((Side==OnTheRight) ^ Transposed) ? m_permutation.indices().coeff(i) : i);
-        }
-      }
-    }
-
+    typedef typename PermutationType::PlainPermutationType PlainPermutationType;
+    typedef internal::traits<PermutationType> PermTraits;
   protected:
-    const PermutationType& m_permutation;
-    const typename MatrixType::Nested m_matrix;
-};
-
-/* Template partial specialization for transposed/inverse permutations */
-
-template<int SizeAtCompileTime, int MaxSizeAtCompileTime>
-struct ei_traits<Transpose<PermutationMatrix<SizeAtCompileTime, MaxSizeAtCompileTime> > >
- : ei_traits<Matrix<int,SizeAtCompileTime,SizeAtCompileTime,0,MaxSizeAtCompileTime,MaxSizeAtCompileTime> >
-{};
-
-template<int SizeAtCompileTime, int MaxSizeAtCompileTime>
-class Transpose<PermutationMatrix<SizeAtCompileTime, MaxSizeAtCompileTime> >
-  : public EigenBase<Transpose<PermutationMatrix<SizeAtCompileTime, MaxSizeAtCompileTime> > >
-{
-    typedef PermutationMatrix<SizeAtCompileTime, MaxSizeAtCompileTime> PermutationType;
-    typedef typename PermutationType::IndicesType IndicesType;
+    InverseImpl() {}
   public:
+    typedef Inverse<PermutationType> InverseType;
+    using EigenBase<Inverse<PermutationType> >::derived;
 
     #ifndef EIGEN_PARSED_BY_DOXYGEN
-    typedef ei_traits<PermutationType> Traits;
-    typedef Matrix<int,SizeAtCompileTime,SizeAtCompileTime,0,MaxSizeAtCompileTime,MaxSizeAtCompileTime>
-            DenseMatrixType;
+    typedef typename PermutationType::DenseMatrixType DenseMatrixType;
     enum {
-      Flags = Traits::Flags,
-      CoeffReadCost = Traits::CoeffReadCost,
-      RowsAtCompileTime = Traits::RowsAtCompileTime,
-      ColsAtCompileTime = Traits::ColsAtCompileTime,
-      MaxRowsAtCompileTime = Traits::MaxRowsAtCompileTime,
-      MaxColsAtCompileTime = Traits::MaxColsAtCompileTime
+      RowsAtCompileTime = PermTraits::RowsAtCompileTime,
+      ColsAtCompileTime = PermTraits::ColsAtCompileTime,
+      MaxRowsAtCompileTime = PermTraits::MaxRowsAtCompileTime,
+      MaxColsAtCompileTime = PermTraits::MaxColsAtCompileTime
     };
-    typedef typename Traits::Scalar Scalar;
     #endif
 
-    Transpose(const PermutationType& p) : m_permutation(p) {}
-
-    inline int rows() const { return m_permutation.rows(); }
-    inline int cols() const { return m_permutation.cols(); }
-
     #ifndef EIGEN_PARSED_BY_DOXYGEN
     template<typename DenseDerived>
     void evalTo(MatrixBase<DenseDerived>& other) const
     {
       other.setZero();
-      for (int i=0; i<rows();++i)
-        other.coeffRef(i, m_permutation.indices().coeff(i)) = typename DenseDerived::Scalar(1);
+      for (Index i=0; i<derived().rows();++i)
+        other.coeffRef(i, derived().nestedExpression().indices().coeff(i)) = typename DenseDerived::Scalar(1);
     }
     #endif
 
     /** \return the equivalent permutation matrix */
-    PermutationType eval() const { return *this; }
+    PlainPermutationType eval() const { return derived(); }
 
-    DenseMatrixType toDenseMatrix() const { return *this; }
+    DenseMatrixType toDenseMatrix() const { return derived(); }
 
     /** \returns the matrix with the inverse permutation applied to the columns.
       */
-    template<typename Derived> friend
-    inline const ei_permut_matrix_product_retval<PermutationType, Derived, OnTheRight, true>
-    operator*(const MatrixBase<Derived>& matrix, const Transpose& trPerm)
+    template<typename OtherDerived> friend
+    const Product<OtherDerived, InverseType, AliasFreeProduct>
+    operator*(const MatrixBase<OtherDerived>& matrix, const InverseType& trPerm)
     {
-      return ei_permut_matrix_product_retval<PermutationType, Derived, OnTheRight, true>(trPerm.m_permutation, matrix.derived());
+      return Product<OtherDerived, InverseType, AliasFreeProduct>(matrix.derived(), trPerm.derived());
     }
 
     /** \returns the matrix with the inverse permutation applied to the rows.
       */
-    template<typename Derived>
-    inline const ei_permut_matrix_product_retval<PermutationType, Derived, OnTheLeft, true>
-    operator*(const MatrixBase<Derived>& matrix) const
+    template<typename OtherDerived>
+    const Product<InverseType, OtherDerived, AliasFreeProduct>
+    operator*(const MatrixBase<OtherDerived>& matrix) const
     {
-      return ei_permut_matrix_product_retval<PermutationType, Derived, OnTheLeft, true>(m_permutation, matrix.derived());
+      return Product<InverseType, OtherDerived, AliasFreeProduct>(derived(), matrix.derived());
     }
-
-    const PermutationType& nestedPermutation() const { return m_permutation; }
-
-  protected:
-    const PermutationType& m_permutation;
 };
 
+template<typename Derived>
+const PermutationWrapper<const Derived> MatrixBase<Derived>::asPermutation() const
+{
+  return derived();
+}
+
+namespace internal {
+
+template<> struct AssignmentKind<DenseShape,PermutationShape> { typedef EigenBase2EigenBase Kind; };
+
+} // end namespace internal
+
+} // end namespace Eigen
+
 #endif // EIGEN_PERMUTATIONMATRIX_H

Eigen/src/Core/PlainObjectBase.h

@@ -0,0 +1,994 @@
+// This file is part of Eigen, a lightweight C++ template library
+// for linear algebra.
+//
+// Copyright (C) 2008-2009 Gael Guennebaud <gael.guennebaud@inria.fr>
+// Copyright (C) 2006-2008 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_DENSESTORAGEBASE_H
+#define EIGEN_DENSESTORAGEBASE_H
+
+#if defined(EIGEN_INITIALIZE_MATRICES_BY_ZERO)
+# define EIGEN_INITIALIZE_COEFFS
+# define EIGEN_INITIALIZE_COEFFS_IF_THAT_OPTION_IS_ENABLED for(int i=0;i<base().size();++i) coeffRef(i)=Scalar(0);
+#elif defined(EIGEN_INITIALIZE_MATRICES_BY_NAN)
+# define EIGEN_INITIALIZE_COEFFS
+# define EIGEN_INITIALIZE_COEFFS_IF_THAT_OPTION_IS_ENABLED for(int i=0;i<base().size();++i) coeffRef(i)=std::numeric_limits<Scalar>::quiet_NaN();
+#else
+# undef EIGEN_INITIALIZE_COEFFS
+# define EIGEN_INITIALIZE_COEFFS_IF_THAT_OPTION_IS_ENABLED
+#endif
+
+namespace Eigen {
+
+namespace internal {
+
+template<int MaxSizeAtCompileTime> struct check_rows_cols_for_overflow {
+  template<typename Index>
+  EIGEN_DEVICE_FUNC
+  static EIGEN_ALWAYS_INLINE void run(Index, Index)
+  {
+  }
+};
+
+template<> struct check_rows_cols_for_overflow<Dynamic> {
+  template<typename Index>
+  EIGEN_DEVICE_FUNC
+  static EIGEN_ALWAYS_INLINE void run(Index rows, Index cols)
+  {
+    // http://hg.mozilla.org/mozilla-central/file/6c8a909977d3/xpcom/ds/CheckedInt.h#l242
+    // we assume Index is signed
+    Index max_index = (size_t(1) << (8 * sizeof(Index) - 1)) - 1; // assume Index is signed
+    bool error = (rows == 0 || cols == 0) ? false
+               : (rows > max_index / cols);
+    if (error)
+      throw_std_bad_alloc();
+  }
+};
+
+template <typename Derived,
+          typename OtherDerived = Derived,
+          bool IsVector = bool(Derived::IsVectorAtCompileTime) && bool(OtherDerived::IsVectorAtCompileTime)>
+struct conservative_resize_like_impl;
+
+template<typename MatrixTypeA, typename MatrixTypeB, bool SwapPointers> struct matrix_swap_impl;
+
+} // end namespace internal
+
+/** \class PlainObjectBase
+  * \brief %Dense storage base class for matrices and arrays.
+  *
+  * This class can be extended with the help of the plugin mechanism described on the page
+  * \ref TopicCustomizingEigen by defining the preprocessor symbol \c EIGEN_PLAINOBJECTBASE_PLUGIN.
+  *
+  * \sa \ref TopicClassHierarchy
+  */
+#ifdef EIGEN_PARSED_BY_DOXYGEN
+namespace internal {
+
+// this is a workaround to doxygen not being able to understand the inheritance logic
+// when it is hidden by the dense_xpr_base helper struct.
+/** This class is just a workaround for Doxygen and it does not not actually exist. */
+template<typename Derived> struct dense_xpr_base_dispatcher_for_doxygen;// : public MatrixBase<Derived> {};
+/** This class is just a workaround for Doxygen and it does not not actually exist. */
+template<typename _Scalar, int _Rows, int _Cols, int _Options, int _MaxRows, int _MaxCols>
+struct dense_xpr_base_dispatcher_for_doxygen<Matrix<_Scalar, _Rows, _Cols, _Options, _MaxRows, _MaxCols> >
+    : public MatrixBase<Matrix<_Scalar, _Rows, _Cols, _Options, _MaxRows, _MaxCols> > {};
+/** This class is just a workaround for Doxygen and it does not not actually exist. */
+template<typename _Scalar, int _Rows, int _Cols, int _Options, int _MaxRows, int _MaxCols>
+struct dense_xpr_base_dispatcher_for_doxygen<Array<_Scalar, _Rows, _Cols, _Options, _MaxRows, _MaxCols> >
+    : public ArrayBase<Array<_Scalar, _Rows, _Cols, _Options, _MaxRows, _MaxCols> > {};
+
+} // namespace internal
+
+template<typename Derived>
+class PlainObjectBase : public internal::dense_xpr_base_dispatcher_for_doxygen<Derived>
+#else
+template<typename Derived>
+class PlainObjectBase : public internal::dense_xpr_base<Derived>::type
+#endif
+{
+  public:
+    enum { Options = internal::traits<Derived>::Options };
+    typedef typename internal::dense_xpr_base<Derived>::type 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;
+    typedef Derived DenseType;
+
+    using Base::RowsAtCompileTime;
+    using Base::ColsAtCompileTime;
+    using Base::SizeAtCompileTime;
+    using Base::MaxRowsAtCompileTime;
+    using Base::MaxColsAtCompileTime;
+    using Base::MaxSizeAtCompileTime;
+    using Base::IsVectorAtCompileTime;
+    using Base::Flags;
+
+    template<typename PlainObjectType, int MapOptions, typename StrideType> friend class Eigen::Map;
+    friend  class Eigen::Map<Derived, Unaligned>;
+    typedef Eigen::Map<Derived, Unaligned>  MapType;
+    friend  class Eigen::Map<const Derived, Unaligned>;
+    typedef const Eigen::Map<const Derived, Unaligned> ConstMapType;
+#if EIGEN_MAX_ALIGN_BYTES>0
+    // for EIGEN_MAX_ALIGN_BYTES==0, AlignedMax==Unaligned, and many compilers generate warnings for friend-ing a class twice.
+    friend  class Eigen::Map<Derived, AlignedMax>;
+    friend  class Eigen::Map<const Derived, AlignedMax>;
+#endif
+    typedef Eigen::Map<Derived, AlignedMax> AlignedMapType;
+    typedef const Eigen::Map<const Derived, AlignedMax> ConstAlignedMapType;
+    template<typename StrideType> struct StridedMapType { typedef Eigen::Map<Derived, Unaligned, StrideType> type; };
+    template<typename StrideType> struct StridedConstMapType { typedef Eigen::Map<const Derived, Unaligned, StrideType> type; };
+    template<typename StrideType> struct StridedAlignedMapType { typedef Eigen::Map<Derived, AlignedMax, StrideType> type; };
+    template<typename StrideType> struct StridedConstAlignedMapType { typedef Eigen::Map<const Derived, AlignedMax, StrideType> type; };
+
+  protected:
+    DenseStorage<Scalar, Base::MaxSizeAtCompileTime, Base::RowsAtCompileTime, Base::ColsAtCompileTime, Options> m_storage;
+
+  public:
+    enum { NeedsToAlign = (SizeAtCompileTime != Dynamic) && (internal::traits<Derived>::Alignment>0) };
+    EIGEN_MAKE_ALIGNED_OPERATOR_NEW_IF(NeedsToAlign)
+
+    EIGEN_DEVICE_FUNC
+    Base& base() { return *static_cast<Base*>(this); }
+    EIGEN_DEVICE_FUNC
+    const Base& base() const { return *static_cast<const Base*>(this); }
+
+    EIGEN_DEVICE_FUNC
+    EIGEN_STRONG_INLINE Index rows() const { return m_storage.rows(); }
+    EIGEN_DEVICE_FUNC
+    EIGEN_STRONG_INLINE Index cols() const { return m_storage.cols(); }
+
+    EIGEN_DEVICE_FUNC
+    EIGEN_STRONG_INLINE const Scalar& coeff(Index rowId, Index colId) const
+    {
+      if(Flags & RowMajorBit)
+        return m_storage.data()[colId + rowId * m_storage.cols()];
+      else // column-major
+        return m_storage.data()[rowId + colId * m_storage.rows()];
+    }
+
+    EIGEN_DEVICE_FUNC
+    EIGEN_STRONG_INLINE const Scalar& coeff(Index index) const
+    {
+      return m_storage.data()[index];
+    }
+
+    EIGEN_DEVICE_FUNC
+    EIGEN_STRONG_INLINE Scalar& coeffRef(Index rowId, Index colId)
+    {
+      if(Flags & RowMajorBit)
+        return m_storage.data()[colId + rowId * m_storage.cols()];
+      else // column-major
+        return m_storage.data()[rowId + colId * m_storage.rows()];
+    }
+
+    EIGEN_DEVICE_FUNC
+    EIGEN_STRONG_INLINE Scalar& coeffRef(Index index)
+    {
+      return m_storage.data()[index];
+    }
+
+    EIGEN_DEVICE_FUNC
+    EIGEN_STRONG_INLINE const Scalar& coeffRef(Index rowId, Index colId) const
+    {
+      if(Flags & RowMajorBit)
+        return m_storage.data()[colId + rowId * m_storage.cols()];
+      else // column-major
+        return m_storage.data()[rowId + colId * m_storage.rows()];
+    }
+
+    EIGEN_DEVICE_FUNC
+    EIGEN_STRONG_INLINE const Scalar& coeffRef(Index index) const
+    {
+      return m_storage.data()[index];
+    }
+
+    /** \internal */
+    template<int LoadMode>
+    EIGEN_STRONG_INLINE PacketScalar packet(Index rowId, Index colId) const
+    {
+      return internal::ploadt<PacketScalar, LoadMode>
+               (m_storage.data() + (Flags & RowMajorBit
+                                   ? colId + rowId * m_storage.cols()
+                                   : rowId + colId * m_storage.rows()));
+    }
+
+    /** \internal */
+    template<int LoadMode>
+    EIGEN_STRONG_INLINE PacketScalar packet(Index index) const
+    {
+      return internal::ploadt<PacketScalar, LoadMode>(m_storage.data() + index);
+    }
+
+    /** \internal */
+    template<int StoreMode>
+    EIGEN_STRONG_INLINE void writePacket(Index rowId, Index colId, const PacketScalar& val)
+    {
+      internal::pstoret<Scalar, PacketScalar, StoreMode>
+              (m_storage.data() + (Flags & RowMajorBit
+                                   ? colId + rowId * m_storage.cols()
+                                   : rowId + colId * m_storage.rows()), val);
+    }
+
+    /** \internal */
+    template<int StoreMode>
+    EIGEN_STRONG_INLINE void writePacket(Index index, const PacketScalar& val)
+    {
+      internal::pstoret<Scalar, PacketScalar, StoreMode>(m_storage.data() + index, val);
+    }
+
+    /** \returns a const pointer to the data array of this matrix */
+    EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE const Scalar *data() const
+    { return m_storage.data(); }
+
+    /** \returns a pointer to the data array of this matrix */
+    EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE Scalar *data()
+    { return m_storage.data(); }
+
+    /** Resizes \c *this to a \a rows x \a cols matrix.
+      *
+      * This method is intended for dynamic-size matrices, although it is legal to call it on any
+      * matrix as long as fixed dimensions are left unchanged. If you only want to change the number
+      * of rows and/or of columns, you can use resize(NoChange_t, Index), resize(Index, NoChange_t).
+      *
+      * If the current number of coefficients of \c *this exactly matches the
+      * product \a rows * \a cols, then no memory allocation is performed and
+      * the current values are left unchanged. In all other cases, including
+      * shrinking, the data is reallocated and all previous values are lost.
+      *
+      * Example: \include Matrix_resize_int_int.cpp
+      * Output: \verbinclude Matrix_resize_int_int.out
+      *
+      * \sa resize(Index) for vectors, resize(NoChange_t, Index), resize(Index, NoChange_t)
+      */
+    EIGEN_DEVICE_FUNC
+    EIGEN_STRONG_INLINE void resize(Index rows, Index cols)
+    {
+      eigen_assert(   EIGEN_IMPLIES(RowsAtCompileTime!=Dynamic,rows==RowsAtCompileTime)
+                   && EIGEN_IMPLIES(ColsAtCompileTime!=Dynamic,cols==ColsAtCompileTime)
+                   && EIGEN_IMPLIES(RowsAtCompileTime==Dynamic && MaxRowsAtCompileTime!=Dynamic,rows<=MaxRowsAtCompileTime)
+                   && EIGEN_IMPLIES(ColsAtCompileTime==Dynamic && MaxColsAtCompileTime!=Dynamic,cols<=MaxColsAtCompileTime)
+                   && rows>=0 && cols>=0 && "Invalid sizes when resizing a matrix or array.");
+      internal::check_rows_cols_for_overflow<MaxSizeAtCompileTime>::run(rows, cols);
+      #ifdef EIGEN_INITIALIZE_COEFFS
+        Index size = rows*cols;
+        bool size_changed = size != this->size();
+        m_storage.resize(size, rows, cols);
+        if(size_changed) EIGEN_INITIALIZE_COEFFS_IF_THAT_OPTION_IS_ENABLED
+      #else
+        m_storage.resize(rows*cols, rows, cols);
+      #endif
+    }
+
+    /** Resizes \c *this to a vector of length \a size
+      *
+      * \only_for_vectors. This method does not work for
+      * partially dynamic matrices when the static dimension is anything other
+      * than 1. For example it will not work with Matrix<double, 2, Dynamic>.
+      *
+      * Example: \include Matrix_resize_int.cpp
+      * Output: \verbinclude Matrix_resize_int.out
+      *
+      * \sa resize(Index,Index), resize(NoChange_t, Index), resize(Index, NoChange_t)
+      */
+    EIGEN_DEVICE_FUNC
+    inline void resize(Index size)
+    {
+      EIGEN_STATIC_ASSERT_VECTOR_ONLY(PlainObjectBase)
+      eigen_assert(((SizeAtCompileTime == Dynamic && (MaxSizeAtCompileTime==Dynamic || size<=MaxSizeAtCompileTime)) || SizeAtCompileTime == size) && size>=0);
+      #ifdef EIGEN_INITIALIZE_COEFFS
+        bool size_changed = size != this->size();
+      #endif
+      if(RowsAtCompileTime == 1)
+        m_storage.resize(size, 1, size);
+      else
+        m_storage.resize(size, size, 1);
+      #ifdef EIGEN_INITIALIZE_COEFFS
+        if(size_changed) EIGEN_INITIALIZE_COEFFS_IF_THAT_OPTION_IS_ENABLED
+      #endif
+    }
+
+    /** Resizes the matrix, changing only the number of columns. For the parameter of type NoChange_t, just pass the special value \c NoChange
+      * as in the example below.
+      *
+      * Example: \include Matrix_resize_NoChange_int.cpp
+      * Output: \verbinclude Matrix_resize_NoChange_int.out
+      *
+      * \sa resize(Index,Index)
+      */
+    EIGEN_DEVICE_FUNC
+    inline void resize(NoChange_t, Index cols)
+    {
+      resize(rows(), cols);
+    }
+
+    /** Resizes the matrix, changing only the number of rows. For the parameter of type NoChange_t, just pass the special value \c NoChange
+      * as in the example below.
+      *
+      * Example: \include Matrix_resize_int_NoChange.cpp
+      * Output: \verbinclude Matrix_resize_int_NoChange.out
+      *
+      * \sa resize(Index,Index)
+      */
+    EIGEN_DEVICE_FUNC
+    inline void resize(Index rows, NoChange_t)
+    {
+      resize(rows, cols());
+    }
+
+    /** Resizes \c *this to have the same dimensions as \a other.
+      * Takes care of doing all the checking that's needed.
+      *
+      * 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 void resizeLike(const EigenBase<OtherDerived>& _other)
+    {
+      const OtherDerived& other = _other.derived();
+      internal::check_rows_cols_for_overflow<MaxSizeAtCompileTime>::run(other.rows(), other.cols());
+      const Index othersize = other.rows()*other.cols();
+      if(RowsAtCompileTime == 1)
+      {
+        eigen_assert(other.rows() == 1 || other.cols() == 1);
+        resize(1, othersize);
+      }
+      else if(ColsAtCompileTime == 1)
+      {
+        eigen_assert(other.rows() == 1 || other.cols() == 1);
+        resize(othersize, 1);
+      }
+      else resize(other.rows(), other.cols());
+    }
+
+    /** Resizes the matrix to \a rows x \a cols while leaving old values untouched.
+      *
+      * The method is intended for matrices of dynamic size. If you only want to change the number
+      * of rows and/or of columns, you can use conservativeResize(NoChange_t, Index) or
+      * conservativeResize(Index, NoChange_t).
+      *
+      * Matrices are resized relative to the top-left element. In case values need to be 
+      * appended to the matrix they will be uninitialized.
+      */
+    EIGEN_DEVICE_FUNC
+    EIGEN_STRONG_INLINE void conservativeResize(Index rows, Index cols)
+    {
+      internal::conservative_resize_like_impl<Derived>::run(*this, rows, cols);
+    }
+
+    /** Resizes the matrix to \a rows x \a cols while leaving old values untouched.
+      *
+      * As opposed to conservativeResize(Index rows, Index cols), this version leaves
+      * the number of columns unchanged.
+      *
+      * In case the matrix is growing, new rows will be uninitialized.
+      */
+    EIGEN_DEVICE_FUNC
+    EIGEN_STRONG_INLINE void conservativeResize(Index rows, NoChange_t)
+    {
+      // Note: see the comment in conservativeResize(Index,Index)
+      conservativeResize(rows, cols());
+    }
+
+    /** Resizes the matrix to \a rows x \a cols while leaving old values untouched.
+      *
+      * As opposed to conservativeResize(Index rows, Index cols), this version leaves
+      * the number of rows unchanged.
+      *
+      * In case the matrix is growing, new columns will be uninitialized.
+      */
+    EIGEN_DEVICE_FUNC
+    EIGEN_STRONG_INLINE void conservativeResize(NoChange_t, Index cols)
+    {
+      // Note: see the comment in conservativeResize(Index,Index)
+      conservativeResize(rows(), cols);
+    }
+
+    /** Resizes the vector to \a size while retaining old values.
+      *
+      * \only_for_vectors. This method does not work for
+      * partially dynamic matrices when the static dimension is anything other
+      * than 1. For example it will not work with Matrix<double, 2, Dynamic>.
+      *
+      * When values are appended, they will be uninitialized.
+      */
+    EIGEN_DEVICE_FUNC
+    EIGEN_STRONG_INLINE void conservativeResize(Index size)
+    {
+      internal::conservative_resize_like_impl<Derived>::run(*this, size);
+    }
+
+    /** Resizes the matrix to \a rows x \a cols of \c other, while leaving old values untouched.
+      *
+      * The method is intended for matrices of dynamic size. If you only want to change the number
+      * of rows and/or of columns, you can use conservativeResize(NoChange_t, Index) or
+      * conservativeResize(Index, NoChange_t).
+      *
+      * Matrices are resized relative to the top-left element. In case values need to be 
+      * appended to the matrix they will copied from \c other.
+      */
+    template<typename OtherDerived>
+    EIGEN_DEVICE_FUNC
+    EIGEN_STRONG_INLINE void conservativeResizeLike(const DenseBase<OtherDerived>& other)
+    {
+      internal::conservative_resize_like_impl<Derived,OtherDerived>::run(*this, 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 Derived& operator=(const PlainObjectBase& other)
+    {
+      return _set(other);
+    }
+
+    /** \sa MatrixBase::lazyAssign() */
+    template<typename OtherDerived>
+    EIGEN_DEVICE_FUNC
+    EIGEN_STRONG_INLINE Derived& lazyAssign(const DenseBase<OtherDerived>& other)
+    {
+      _resize_to_match(other);
+      return Base::lazyAssign(other.derived());
+    }
+
+    template<typename OtherDerived>
+    EIGEN_DEVICE_FUNC
+    EIGEN_STRONG_INLINE Derived& operator=(const ReturnByValue<OtherDerived>& func)
+    {
+      resize(func.rows(), func.cols());
+      return Base::operator=(func);
+    }
+
+    // Prevent user from trying to instantiate PlainObjectBase objects
+    // by making all its constructor protected. See bug 1074.
+  protected:
+
+    EIGEN_DEVICE_FUNC
+    EIGEN_STRONG_INLINE PlainObjectBase() : m_storage()
+    {
+//       _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
+    explicit PlainObjectBase(internal::constructor_without_unaligned_array_assert)
+      : m_storage(internal::constructor_without_unaligned_array_assert())
+    {
+//       _check_template_params(); EIGEN_INITIALIZE_COEFFS_IF_THAT_OPTION_IS_ENABLED
+    }
+#endif
+
+#ifdef EIGEN_HAVE_RVALUE_REFERENCES
+    EIGEN_DEVICE_FUNC
+    PlainObjectBase(PlainObjectBase&& other)
+      : m_storage( std::move(other.m_storage) )
+    {
+    }
+
+    EIGEN_DEVICE_FUNC
+    PlainObjectBase& operator=(PlainObjectBase&& other)
+    {
+      using std::swap;
+      swap(m_storage, other.m_storage);
+      return *this;
+    }
+#endif
+
+    /** Copy constructor */
+    EIGEN_DEVICE_FUNC
+    EIGEN_STRONG_INLINE PlainObjectBase(const PlainObjectBase& other)
+      : Base(), m_storage(other.m_storage) { }
+    EIGEN_DEVICE_FUNC
+    EIGEN_STRONG_INLINE PlainObjectBase(Index size, Index rows, Index cols)
+      : m_storage(size, rows, cols)
+    {
+//       _check_template_params();
+//       EIGEN_INITIALIZE_COEFFS_IF_THAT_OPTION_IS_ENABLED
+    }
+
+    /** \sa PlainObjectBase::operator=(const EigenBase<OtherDerived>&) */
+    template<typename OtherDerived>
+    EIGEN_DEVICE_FUNC
+    EIGEN_STRONG_INLINE PlainObjectBase(const DenseBase<OtherDerived> &other)
+      : m_storage()
+    {
+      _check_template_params();
+      resizeLike(other);
+      _set_noalias(other);
+    }
+
+    /** \sa PlainObjectBase::operator=(const EigenBase<OtherDerived>&) */
+    template<typename OtherDerived>
+    EIGEN_DEVICE_FUNC
+    EIGEN_STRONG_INLINE PlainObjectBase(const EigenBase<OtherDerived> &other)
+      : m_storage()
+    {
+      _check_template_params();
+      resizeLike(other);
+      *this = other.derived();
+    }
+    /** \brief Copy constructor with in-place evaluation */
+    template<typename OtherDerived>
+    EIGEN_DEVICE_FUNC
+    EIGEN_STRONG_INLINE PlainObjectBase(const ReturnByValue<OtherDerived>& other)
+    {
+      _check_template_params();
+      // FIXME this does not automatically transpose vectors if necessary
+      resize(other.rows(), other.cols());
+      other.evalTo(this->derived());
+    }
+
+  public:
+
+    /** \copydoc MatrixBase::operator=(const EigenBase<OtherDerived>&)
+      */
+    template<typename OtherDerived>
+    EIGEN_DEVICE_FUNC 
+    EIGEN_STRONG_INLINE Derived& operator=(const EigenBase<OtherDerived> &other)
+    {
+      _resize_to_match(other);
+      Base::operator=(other.derived());
+      return this->derived();
+    }
+
+    /** \name Map
+      * These are convenience functions returning Map objects. The Map() static functions return unaligned Map objects,
+      * while the AlignedMap() functions return aligned Map objects and thus should be called only with 16-byte-aligned
+      * \a data pointers.
+      *
+      * \see class Map
+      */
+    //@{
+    static inline ConstMapType Map(const Scalar* data)
+    { return ConstMapType(data); }
+    static inline MapType Map(Scalar* data)
+    { return MapType(data); }
+    static inline ConstMapType Map(const Scalar* data, Index size)
+    { return ConstMapType(data, size); }
+    static inline MapType Map(Scalar* data, Index size)
+    { return MapType(data, size); }
+    static inline ConstMapType Map(const Scalar* data, Index rows, Index cols)
+    { return ConstMapType(data, rows, cols); }
+    static inline MapType Map(Scalar* data, Index rows, Index cols)
+    { return MapType(data, rows, cols); }
+
+    static inline ConstAlignedMapType MapAligned(const Scalar* data)
+    { return ConstAlignedMapType(data); }
+    static inline AlignedMapType MapAligned(Scalar* data)
+    { return AlignedMapType(data); }
+    static inline ConstAlignedMapType MapAligned(const Scalar* data, Index size)
+    { return ConstAlignedMapType(data, size); }
+    static inline AlignedMapType MapAligned(Scalar* data, Index size)
+    { return AlignedMapType(data, size); }
+    static inline ConstAlignedMapType MapAligned(const Scalar* data, Index rows, Index cols)
+    { return ConstAlignedMapType(data, rows, cols); }
+    static inline AlignedMapType MapAligned(Scalar* data, Index rows, Index cols)
+    { return AlignedMapType(data, rows, cols); }
+
+    template<int Outer, int Inner>
+    static inline typename StridedConstMapType<Stride<Outer, Inner> >::type Map(const Scalar* data, const Stride<Outer, Inner>& stride)
+    { return typename StridedConstMapType<Stride<Outer, Inner> >::type(data, stride); }
+    template<int Outer, int Inner>
+    static inline typename StridedMapType<Stride<Outer, Inner> >::type Map(Scalar* data, const Stride<Outer, Inner>& stride)
+    { return typename StridedMapType<Stride<Outer, Inner> >::type(data, stride); }
+    template<int Outer, int Inner>
+    static inline typename StridedConstMapType<Stride<Outer, Inner> >::type Map(const Scalar* data, Index size, const Stride<Outer, Inner>& stride)
+    { return typename StridedConstMapType<Stride<Outer, Inner> >::type(data, size, stride); }
+    template<int Outer, int Inner>
+    static inline typename StridedMapType<Stride<Outer, Inner> >::type Map(Scalar* data, Index size, const Stride<Outer, Inner>& stride)
+    { return typename StridedMapType<Stride<Outer, Inner> >::type(data, size, stride); }
+    template<int Outer, int Inner>
+    static inline typename StridedConstMapType<Stride<Outer, Inner> >::type Map(const Scalar* data, Index rows, Index cols, const Stride<Outer, Inner>& stride)
+    { return typename StridedConstMapType<Stride<Outer, Inner> >::type(data, rows, cols, stride); }
+    template<int Outer, int Inner>
+    static inline typename StridedMapType<Stride<Outer, Inner> >::type Map(Scalar* data, Index rows, Index cols, const Stride<Outer, Inner>& stride)
+    { return typename StridedMapType<Stride<Outer, Inner> >::type(data, rows, cols, stride); }
+
+    template<int Outer, int Inner>
+    static inline typename StridedConstAlignedMapType<Stride<Outer, Inner> >::type MapAligned(const Scalar* data, const Stride<Outer, Inner>& stride)
+    { return typename StridedConstAlignedMapType<Stride<Outer, Inner> >::type(data, stride); }
+    template<int Outer, int Inner>
+    static inline typename StridedAlignedMapType<Stride<Outer, Inner> >::type MapAligned(Scalar* data, const Stride<Outer, Inner>& stride)
+    { return typename StridedAlignedMapType<Stride<Outer, Inner> >::type(data, stride); }
+    template<int Outer, int Inner>
+    static inline typename StridedConstAlignedMapType<Stride<Outer, Inner> >::type MapAligned(const Scalar* data, Index size, const Stride<Outer, Inner>& stride)
+    { return typename StridedConstAlignedMapType<Stride<Outer, Inner> >::type(data, size, stride); }
+    template<int Outer, int Inner>
+    static inline typename StridedAlignedMapType<Stride<Outer, Inner> >::type MapAligned(Scalar* data, Index size, const Stride<Outer, Inner>& stride)
+    { return typename StridedAlignedMapType<Stride<Outer, Inner> >::type(data, size, stride); }
+    template<int Outer, int Inner>
+    static inline typename StridedConstAlignedMapType<Stride<Outer, Inner> >::type MapAligned(const Scalar* data, Index rows, Index cols, const Stride<Outer, Inner>& stride)
+    { return typename StridedConstAlignedMapType<Stride<Outer, Inner> >::type(data, rows, cols, stride); }
+    template<int Outer, int Inner>
+    static inline typename StridedAlignedMapType<Stride<Outer, Inner> >::type MapAligned(Scalar* data, Index rows, Index cols, const Stride<Outer, Inner>& stride)
+    { return typename StridedAlignedMapType<Stride<Outer, Inner> >::type(data, rows, cols, stride); }
+    //@}
+
+    using Base::setConstant;
+    EIGEN_DEVICE_FUNC Derived& setConstant(Index size, const Scalar& value);
+    EIGEN_DEVICE_FUNC Derived& setConstant(Index rows, Index cols, const Scalar& value);
+
+    using Base::setZero;
+    EIGEN_DEVICE_FUNC Derived& setZero(Index size);
+    EIGEN_DEVICE_FUNC Derived& setZero(Index rows, Index cols);
+
+    using Base::setOnes;
+    EIGEN_DEVICE_FUNC Derived& setOnes(Index size);
+    EIGEN_DEVICE_FUNC Derived& setOnes(Index rows, Index cols);
+
+    using Base::setRandom;
+    Derived& setRandom(Index size);
+    Derived& setRandom(Index rows, Index cols);
+
+    #ifdef EIGEN_PLAINOBJECTBASE_PLUGIN
+    #include EIGEN_PLAINOBJECTBASE_PLUGIN
+    #endif
+
+  protected:
+    /** \internal Resizes *this in preparation for assigning \a other to it.
+      * Takes care of doing all the checking that's needed.
+      *
+      * 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 void _resize_to_match(const EigenBase<OtherDerived>& other)
+    {
+      #ifdef EIGEN_NO_AUTOMATIC_RESIZING
+      eigen_assert((this->size()==0 || (IsVectorAtCompileTime ? (this->size() == other.size())
+                 : (rows() == other.rows() && cols() == other.cols())))
+        && "Size mismatch. Automatic resizing is disabled because EIGEN_NO_AUTOMATIC_RESIZING is defined");
+      EIGEN_ONLY_USED_FOR_DEBUG(other);
+      #else
+      resizeLike(other);
+      #endif
+    }
+
+    /**
+      * \brief 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.
+      *
+      * \sa operator=(const MatrixBase<OtherDerived>&), _set_noalias()
+      *
+      * \internal
+      */
+    // aliasing is dealt once in internall::call_assignment
+    // so at this stage we have to assume aliasing... and resising has to be done later.
+    template<typename OtherDerived>
+    EIGEN_DEVICE_FUNC 
+    EIGEN_STRONG_INLINE Derived& _set(const DenseBase<OtherDerived>& other)
+    {
+      internal::call_assignment(this->derived(), other.derived());
+      return this->derived();
+    }
+
+    /** \internal Like _set() but additionally makes the assumption that no aliasing effect can happen (which
+      * is the case when creating a new matrix) so one can enforce lazy evaluation.
+      *
+      * \sa operator=(const MatrixBase<OtherDerived>&), _set()
+      */
+    template<typename OtherDerived>
+    EIGEN_DEVICE_FUNC 
+    EIGEN_STRONG_INLINE Derived& _set_noalias(const DenseBase<OtherDerived>& other)
+    {
+      // I don't think we need this resize call since the lazyAssign will anyways resize
+      // and lazyAssign will be called by the assign selector.
+      //_resize_to_match(other);
+      // the 'false' below means to enforce lazy evaluation. We don't use lazyAssign() because
+      // it wouldn't allow to copy a row-vector into a column-vector.
+      internal::call_assignment_no_alias(this->derived(), other.derived(), internal::assign_op<Scalar>());
+      return this->derived();
+    }
+
+    template<typename T0, typename T1>
+    EIGEN_DEVICE_FUNC
+    EIGEN_STRONG_INLINE void _init2(Index rows, Index cols, typename internal::enable_if<Base::SizeAtCompileTime!=2,T0>::type* = 0)
+    {
+      EIGEN_STATIC_ASSERT(bool(NumTraits<T0>::IsInteger) &&
+                          bool(NumTraits<T1>::IsInteger),
+                          FLOATING_POINT_ARGUMENT_PASSED__INTEGER_WAS_EXPECTED)
+      resize(rows,cols);
+    }
+    
+    template<typename T0, typename T1>
+    EIGEN_DEVICE_FUNC 
+    EIGEN_STRONG_INLINE void _init2(const Scalar& val0, const Scalar& val1, typename internal::enable_if<Base::SizeAtCompileTime==2,T0>::type* = 0)
+    {
+      EIGEN_STATIC_ASSERT_VECTOR_SPECIFIC_SIZE(PlainObjectBase, 2)
+      m_storage.data()[0] = val0;
+      m_storage.data()[1] = val1;
+    }
+    
+    template<typename T0, typename T1>
+    EIGEN_DEVICE_FUNC 
+    EIGEN_STRONG_INLINE void _init2(const Index& val0, const Index& val1,
+                                    typename internal::enable_if<    (!internal::is_same<Index,Scalar>::value)
+                                                                  && (internal::is_same<T0,Index>::value)
+                                                                  && (internal::is_same<T1,Index>::value)
+                                                                  && Base::SizeAtCompileTime==2,T1>::type* = 0)
+    {
+      EIGEN_STATIC_ASSERT_VECTOR_SPECIFIC_SIZE(PlainObjectBase, 2)
+      m_storage.data()[0] = Scalar(val0);
+      m_storage.data()[1] = Scalar(val1);
+    }
+
+    // The argument is convertible to the Index type and we either have a non 1x1 Matrix, or a dynamic-sized Array,
+    // then the argument is meant to be the size of the object.
+    template<typename T>
+    EIGEN_DEVICE_FUNC
+    EIGEN_STRONG_INLINE void _init1(Index size, typename internal::enable_if<    (Base::SizeAtCompileTime!=1 || !internal::is_convertible<T, Scalar>::value)
+                                                                              && ((!internal::is_same<typename internal::traits<Derived>::XprKind,ArrayXpr>::value || Base::SizeAtCompileTime==Dynamic)),T>::type* = 0)
+    {
+      // NOTE MSVC 2008 complains if we directly put bool(NumTraits<T>::IsInteger) as the EIGEN_STATIC_ASSERT argument.
+      const bool is_integer = NumTraits<T>::IsInteger;
+      EIGEN_STATIC_ASSERT(is_integer,
+                          FLOATING_POINT_ARGUMENT_PASSED__INTEGER_WAS_EXPECTED)
+      resize(size);
+    }
+    
+    // We have a 1x1 matrix/array => the argument is interpreted as the value of the unique coefficient (case where scalar type can be implicitely converted)
+    template<typename T>
+    EIGEN_DEVICE_FUNC
+    EIGEN_STRONG_INLINE void _init1(const Scalar& val0, typename internal::enable_if<Base::SizeAtCompileTime==1 && internal::is_convertible<T, Scalar>::value,T>::type* = 0)
+    {
+      EIGEN_STATIC_ASSERT_VECTOR_SPECIFIC_SIZE(PlainObjectBase, 1)
+      m_storage.data()[0] = val0;
+    }
+    
+    // We have a 1x1 matrix/array => the argument is interpreted as the value of the unique coefficient (case where scalar type match the index type)
+    template<typename T>
+    EIGEN_DEVICE_FUNC
+    EIGEN_STRONG_INLINE void _init1(const Index& val0,
+                                    typename internal::enable_if<    (!internal::is_same<Index,Scalar>::value)
+                                                                  && (internal::is_same<Index,T>::value)
+                                                                  && Base::SizeAtCompileTime==1
+                                                                  && internal::is_convertible<T, Scalar>::value,T*>::type* = 0)
+    {
+      EIGEN_STATIC_ASSERT_VECTOR_SPECIFIC_SIZE(PlainObjectBase, 1)
+      m_storage.data()[0] = Scalar(val0);
+    }
+
+    // Initialize a fixed size matrix from a pointer to raw data
+    template<typename T>
+    EIGEN_DEVICE_FUNC
+    EIGEN_STRONG_INLINE void _init1(const Scalar* data){
+      this->_set_noalias(ConstMapType(data));
+    }
+
+    // Initialize an arbitrary matrix from a dense expression
+    template<typename T, typename OtherDerived>
+    EIGEN_DEVICE_FUNC
+    EIGEN_STRONG_INLINE void _init1(const DenseBase<OtherDerived>& other){
+      this->_set_noalias(other);
+    }
+
+    // Initialize an arbitrary matrix from a generic Eigen expression
+    template<typename T, typename OtherDerived>
+    EIGEN_DEVICE_FUNC
+    EIGEN_STRONG_INLINE void _init1(const EigenBase<OtherDerived>& other){
+      this->derived() = other;
+    }
+
+    template<typename T, typename OtherDerived>
+    EIGEN_DEVICE_FUNC
+    EIGEN_STRONG_INLINE void _init1(const ReturnByValue<OtherDerived>& other)
+    {
+      resize(other.rows(), other.cols());
+      other.evalTo(this->derived());
+    }
+
+    template<typename T, typename OtherDerived, int ColsAtCompileTime>
+    EIGEN_DEVICE_FUNC
+    EIGEN_STRONG_INLINE void _init1(const RotationBase<OtherDerived,ColsAtCompileTime>& r)
+    {
+      this->derived() = r;
+    }
+    
+    // For fixed -size arrays:
+    template<typename T>
+    EIGEN_DEVICE_FUNC
+    EIGEN_STRONG_INLINE void _init1(const Scalar& val0,
+                                    typename internal::enable_if<    Base::SizeAtCompileTime!=Dynamic
+                                                                  && Base::SizeAtCompileTime!=1
+                                                                  && internal::is_convertible<T, Scalar>::value
+                                                                  && internal::is_same<typename internal::traits<Derived>::XprKind,ArrayXpr>::value,T>::type* = 0)
+    {
+      Base::setConstant(val0);
+    }
+    
+    template<typename T>
+    EIGEN_DEVICE_FUNC
+    EIGEN_STRONG_INLINE void _init1(const Index& val0,
+                                    typename internal::enable_if<    (!internal::is_same<Index,Scalar>::value)
+                                                                  && (internal::is_same<Index,T>::value)
+                                                                  && Base::SizeAtCompileTime!=Dynamic
+                                                                  && Base::SizeAtCompileTime!=1
+                                                                  && internal::is_convertible<T, Scalar>::value
+                                                                  && internal::is_same<typename internal::traits<Derived>::XprKind,ArrayXpr>::value,T*>::type* = 0)
+    {
+      Base::setConstant(val0);
+    }
+    
+    template<typename MatrixTypeA, typename MatrixTypeB, bool SwapPointers>
+    friend struct internal::matrix_swap_impl;
+
+  public:
+    
+#ifndef EIGEN_PARSED_BY_DOXYGEN
+    /** \internal
+      * \brief Override DenseBase::swap() since for dynamic-sized matrices
+      * of same type it is enough to swap the data pointers.
+      */
+    template<typename OtherDerived>
+    EIGEN_DEVICE_FUNC
+    void swap(DenseBase<OtherDerived> & other)
+    {
+      enum { SwapPointers = internal::is_same<Derived, OtherDerived>::value && Base::SizeAtCompileTime==Dynamic };
+      internal::matrix_swap_impl<Derived, OtherDerived, bool(SwapPointers)>::run(this->derived(), other.derived());
+    }
+    
+    /** \internal
+      * \brief const version forwarded to DenseBase::swap
+      */
+    template<typename OtherDerived>
+    EIGEN_DEVICE_FUNC
+    void swap(DenseBase<OtherDerived> const & other)
+    { Base::swap(other.derived()); }
+    
+    EIGEN_DEVICE_FUNC 
+    static EIGEN_STRONG_INLINE void _check_template_params()
+    {
+      EIGEN_STATIC_ASSERT((EIGEN_IMPLIES(MaxRowsAtCompileTime==1 && MaxColsAtCompileTime!=1, (Options&RowMajor)==RowMajor)
+                        && EIGEN_IMPLIES(MaxColsAtCompileTime==1 && MaxRowsAtCompileTime!=1, (Options&RowMajor)==0)
+                        && ((RowsAtCompileTime == Dynamic) || (RowsAtCompileTime >= 0))
+                        && ((ColsAtCompileTime == Dynamic) || (ColsAtCompileTime >= 0))
+                        && ((MaxRowsAtCompileTime == Dynamic) || (MaxRowsAtCompileTime >= 0))
+                        && ((MaxColsAtCompileTime == Dynamic) || (MaxColsAtCompileTime >= 0))
+                        && (MaxRowsAtCompileTime == RowsAtCompileTime || RowsAtCompileTime==Dynamic)
+                        && (MaxColsAtCompileTime == ColsAtCompileTime || ColsAtCompileTime==Dynamic)
+                        && (Options & (DontAlign|RowMajor)) == Options),
+        INVALID_MATRIX_TEMPLATE_PARAMETERS)
+    }
+
+    enum { IsPlainObjectBase = 1 };
+#endif
+};
+
+namespace internal {
+
+template <typename Derived, typename OtherDerived, bool IsVector>
+struct conservative_resize_like_impl
+{
+  static void run(DenseBase<Derived>& _this, Index rows, Index cols)
+  {
+    if (_this.rows() == rows && _this.cols() == cols) return;
+    EIGEN_STATIC_ASSERT_DYNAMIC_SIZE(Derived)
+
+    if ( ( Derived::IsRowMajor && _this.cols() == cols) || // row-major and we change only the number of rows
+         (!Derived::IsRowMajor && _this.rows() == rows) )  // column-major and we change only the number of columns
+    {
+      internal::check_rows_cols_for_overflow<Derived::MaxSizeAtCompileTime>::run(rows, cols);
+      _this.derived().m_storage.conservativeResize(rows*cols,rows,cols);
+    }
+    else
+    {
+      // The storage order does not allow us to use reallocation.
+      typename Derived::PlainObject tmp(rows,cols);
+      const Index common_rows = (std::min)(rows, _this.rows());
+      const Index common_cols = (std::min)(cols, _this.cols());
+      tmp.block(0,0,common_rows,common_cols) = _this.block(0,0,common_rows,common_cols);
+      _this.derived().swap(tmp);
+    }
+  }
+
+  static void run(DenseBase<Derived>& _this, const DenseBase<OtherDerived>& other)
+  {
+    if (_this.rows() == other.rows() && _this.cols() == other.cols()) return;
+
+    // Note: Here is space for improvement. Basically, for conservativeResize(Index,Index),
+    // neither RowsAtCompileTime or ColsAtCompileTime must be Dynamic. If only one of the
+    // dimensions is dynamic, one could use either conservativeResize(Index rows, NoChange_t) or
+    // conservativeResize(NoChange_t, Index cols). For these methods new static asserts like
+    // EIGEN_STATIC_ASSERT_DYNAMIC_ROWS and EIGEN_STATIC_ASSERT_DYNAMIC_COLS would be good.
+    EIGEN_STATIC_ASSERT_DYNAMIC_SIZE(Derived)
+    EIGEN_STATIC_ASSERT_DYNAMIC_SIZE(OtherDerived)
+
+    if ( ( Derived::IsRowMajor && _this.cols() == other.cols()) || // row-major and we change only the number of rows
+         (!Derived::IsRowMajor && _this.rows() == other.rows()) )  // column-major and we change only the number of columns
+    {
+      const Index new_rows = other.rows() - _this.rows();
+      const Index new_cols = other.cols() - _this.cols();
+      _this.derived().m_storage.conservativeResize(other.size(),other.rows(),other.cols());
+      if (new_rows>0)
+        _this.bottomRightCorner(new_rows, other.cols()) = other.bottomRows(new_rows);
+      else if (new_cols>0)
+        _this.bottomRightCorner(other.rows(), new_cols) = other.rightCols(new_cols);
+    }
+    else
+    {
+      // The storage order does not allow us to use reallocation.
+      typename Derived::PlainObject tmp(other);
+      const Index common_rows = (std::min)(tmp.rows(), _this.rows());
+      const Index common_cols = (std::min)(tmp.cols(), _this.cols());
+      tmp.block(0,0,common_rows,common_cols) = _this.block(0,0,common_rows,common_cols);
+      _this.derived().swap(tmp);
+    }
+  }
+};
+
+// Here, the specialization for vectors inherits from the general matrix case
+// to allow calling .conservativeResize(rows,cols) on vectors.
+template <typename Derived, typename OtherDerived>
+struct conservative_resize_like_impl<Derived,OtherDerived,true>
+  : conservative_resize_like_impl<Derived,OtherDerived,false>
+{
+  using conservative_resize_like_impl<Derived,OtherDerived,false>::run;
+  
+  static void run(DenseBase<Derived>& _this, Index size)
+  {
+    const Index new_rows = Derived::RowsAtCompileTime==1 ? 1 : size;
+    const Index new_cols = Derived::RowsAtCompileTime==1 ? size : 1;
+    _this.derived().m_storage.conservativeResize(size,new_rows,new_cols);
+  }
+
+  static void run(DenseBase<Derived>& _this, const DenseBase<OtherDerived>& other)
+  {
+    if (_this.rows() == other.rows() && _this.cols() == other.cols()) return;
+
+    const Index num_new_elements = other.size() - _this.size();
+
+    const Index new_rows = Derived::RowsAtCompileTime==1 ? 1 : other.rows();
+    const Index new_cols = Derived::RowsAtCompileTime==1 ? other.cols() : 1;
+    _this.derived().m_storage.conservativeResize(other.size(),new_rows,new_cols);
+
+    if (num_new_elements > 0)
+      _this.tail(num_new_elements) = other.tail(num_new_elements);
+  }
+};
+
+template<typename MatrixTypeA, typename MatrixTypeB, bool SwapPointers>
+struct matrix_swap_impl
+{
+  EIGEN_DEVICE_FUNC
+  static inline void run(MatrixTypeA& a, MatrixTypeB& b)
+  {
+    a.base().swap(b);
+  }
+};
+
+template<typename MatrixTypeA, typename MatrixTypeB>
+struct matrix_swap_impl<MatrixTypeA, MatrixTypeB, true>
+{
+  EIGEN_DEVICE_FUNC
+  static inline void run(MatrixTypeA& a, MatrixTypeB& b)
+  {
+    static_cast<typename MatrixTypeA::Base&>(a).m_storage.swap(static_cast<typename MatrixTypeB::Base&>(b).m_storage);
+  }
+};
+
+} // end namespace internal
+
+} // end namespace Eigen
+
+#endif // EIGEN_DENSESTORAGEBASE_H

Eigen/src/Core/Product.h

@@ -1,543 +1,222 @@
 // 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 Gael Guennebaud <gael.guennebaud@inria.fr>
+// Copyright (C) 2008-2011 Gael Guennebaud <gael.guennebaud@inria.fr>
 //
-// Eigen is free software; you can redistribute it and/or
-// modify it under the terms of the GNU Lesser General Public
-// License as published by the Free Software Foundation; either
-// version 3 of the License, or (at your option) any later version.
-//
-// Alternatively, you can redistribute it and/or
-// modify it under the terms of the GNU General Public License as
-// published by the Free Software Foundation; either version 2 of
-// the License, or (at your option) any later version.
-//
-// Eigen is distributed in the hope that it will be useful, but WITHOUT ANY
-// WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS
-// FOR A PARTICULAR PURPOSE. See the GNU Lesser General Public License or the
-// GNU General Public License for more details.
-//
-// You should have received a copy of the GNU Lesser General Public
-// License and a copy of the GNU General Public License along with
-// Eigen. If not, see <http://www.gnu.org/licenses/>.
+// 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_PRODUCT_H
 #define EIGEN_PRODUCT_H
 
-/** \class GeneralProduct
+namespace Eigen {
+
+template<typename Lhs, typename Rhs, int Option, typename StorageKind> class ProductImpl;
+
+/** \class Product
   * \ingroup Core_Module
   *
-  * \brief Expression of the product of two general matrices or vectors
+  * \brief Expression of the product of two arbitrary matrices or vectors
   *
-  * \param LhsNested the type used to store the left-hand side
-  * \param RhsNested the type used to store the right-hand side
-  * \param ProductMode the type of the product
+  * \param Lhs the type of the left-hand side expression
+  * \param Rhs the type of the right-hand side expression
   *
-  * This class represents an expression of the product of two general matrices.
-  * We call a general matrix, a dense matrix with full storage. For instance,
-  * This excludes triangular, selfadjoint, and sparse matrices.
-  * It is the return type of the operator* between general matrices. Its template
-  * arguments are determined automatically by ProductReturnType. Therefore,
-  * GeneralProduct should never be used direclty. To determine the result type of a
-  * function which involves a matrix product, use ProductReturnType::Type.
+  * This class represents an expression of the product of two arbitrary matrices.
+  * 
+  * The other template parameters are:
+  * \tparam Option     can be DefaultProduct, AliasFreeProduct, or LazyProduct
   *
-  * \sa ProductReturnType, MatrixBase::operator*(const MatrixBase<OtherDerived>&)
   */
-template<typename Lhs, typename Rhs, int ProductType = ei_product_type<Lhs,Rhs>::value>
-class GeneralProduct;
 
-template<int Rows, int Cols, int Depth> struct ei_product_type_selector;
 
-enum {
-  Large = 2,
-  Small = 3
+namespace internal {
+
+// Determine the scalar of Product<Lhs, Rhs>. This is normally the same as Lhs::Scalar times
+// Rhs::Scalar, but product with permutation matrices inherit the scalar of the other factor.
+template<typename Lhs, typename Rhs, typename LhsShape = typename evaluator_traits<Lhs>::Shape, 
+         typename RhsShape = typename evaluator_traits<Rhs>::Shape >
+struct product_result_scalar
+{
+  typedef typename scalar_product_traits<typename Lhs::Scalar, typename Rhs::Scalar>::ReturnType Scalar;
 };
 
-template<typename Lhs, typename Rhs> struct ei_product_type
+template<typename Lhs, typename Rhs, typename RhsShape>
+struct product_result_scalar<Lhs, Rhs, PermutationShape, RhsShape>
 {
-  typedef typename ei_cleantype<Lhs>::type _Lhs;
-  typedef typename ei_cleantype<Rhs>::type _Rhs;
+  typedef typename Rhs::Scalar Scalar;
+};
+
+template<typename Lhs, typename Rhs, typename LhsShape>
+  struct product_result_scalar<Lhs, Rhs, LhsShape, PermutationShape>
+{
+  typedef typename Lhs::Scalar Scalar;
+};
+
+template<typename Lhs, typename Rhs, typename RhsShape>
+struct product_result_scalar<Lhs, Rhs, TranspositionsShape, RhsShape>
+{
+  typedef typename Rhs::Scalar Scalar;
+};
+
+template<typename Lhs, typename Rhs, typename LhsShape>
+  struct product_result_scalar<Lhs, Rhs, LhsShape, TranspositionsShape>
+{
+  typedef typename Lhs::Scalar Scalar;
+};
+
+template<typename Lhs, typename Rhs, int Option>
+struct traits<Product<Lhs, Rhs, Option> >
+{
+  typedef typename remove_all<Lhs>::type LhsCleaned;
+  typedef typename remove_all<Rhs>::type RhsCleaned;
+  typedef traits<LhsCleaned> LhsTraits;
+  typedef traits<RhsCleaned> RhsTraits;
+  
+  typedef MatrixXpr XprKind;
+  
+  typedef typename product_result_scalar<LhsCleaned,RhsCleaned>::Scalar Scalar;
+  typedef typename product_promote_storage_type<typename LhsTraits::StorageKind,
+                                                typename RhsTraits::StorageKind,
+                                                internal::product_type<Lhs,Rhs>::ret>::ret StorageKind;
+  typedef typename promote_index_type<typename LhsTraits::StorageIndex,
+                                      typename RhsTraits::StorageIndex>::type StorageIndex;
+  
   enum {
-    Rows  = _Lhs::MaxRowsAtCompileTime,
-    Cols  = _Rhs::MaxColsAtCompileTime,
-    Depth = EIGEN_SIZE_MIN_PREFER_FIXED(_Lhs::MaxColsAtCompileTime,_Rhs::MaxRowsAtCompileTime)
-  };
-
-  // 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   = Rows == Dynamic || Rows >=EIGEN_CACHEFRIENDLY_PRODUCT_THRESHOLD ? Large : (Rows==1   ? 1 : Small),
-    cols_select   = Cols == Dynamic || Cols >=EIGEN_CACHEFRIENDLY_PRODUCT_THRESHOLD ? Large : (Cols==1   ? 1 : Small),
-    depth_select  = Depth == Dynamic || Depth>=EIGEN_CACHEFRIENDLY_PRODUCT_THRESHOLD ? Large : (Depth==1  ? 1 : Small)
-  };
-  typedef ei_product_type_selector<rows_select, cols_select, depth_select> product_type_selector;
-
-public:
-  enum {
-    value = product_type_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 ei_product_type_selector<M,N,1>              { enum { ret = OuterProduct }; };
-template<int Depth>     struct ei_product_type_selector<1,    1,    Depth>  { enum { ret = InnerProduct }; };
-template<>              struct ei_product_type_selector<1,    1,    1>      { enum { ret = InnerProduct }; };
-template<>              struct ei_product_type_selector<Small,1,    Small>  { enum { ret = CoeffBasedProductMode }; };
-template<>              struct ei_product_type_selector<1,    Small,Small>  { enum { ret = CoeffBasedProductMode }; };
-template<>              struct ei_product_type_selector<Small,Small,Small>  { enum { ret = CoeffBasedProductMode }; };
-template<>              struct ei_product_type_selector<Small, Small, 1>    { enum { ret = LazyCoeffBasedProductMode }; };
-template<>              struct ei_product_type_selector<Small, Large, 1>    { enum { ret = LazyCoeffBasedProductMode }; };
-template<>              struct ei_product_type_selector<Large, Small, 1>    { enum { ret = LazyCoeffBasedProductMode }; };
-template<>              struct ei_product_type_selector<1,    Large,Small>  { enum { ret = CoeffBasedProductMode }; };
-template<>              struct ei_product_type_selector<1,    Large,Large>  { enum { ret = GemvProduct }; };
-template<>              struct ei_product_type_selector<1,    Small,Large>  { enum { ret = CoeffBasedProductMode }; };
-template<>              struct ei_product_type_selector<Large,1,    Small>  { enum { ret = CoeffBasedProductMode }; };
-template<>              struct ei_product_type_selector<Large,1,    Large>  { enum { ret = GemvProduct }; };
-template<>              struct ei_product_type_selector<Small,1,    Large>  { enum { ret = CoeffBasedProductMode }; };
-template<>              struct ei_product_type_selector<Small,Small,Large>  { enum { ret = GemmProduct }; };
-template<>              struct ei_product_type_selector<Large,Small,Large>  { enum { ret = GemmProduct }; };
-template<>              struct ei_product_type_selector<Small,Large,Large>  { enum { ret = GemmProduct }; };
-template<>              struct ei_product_type_selector<Large,Large,Large>  { enum { ret = GemmProduct }; };
-template<>              struct ei_product_type_selector<Large,Small,Small>  { enum { ret = GemmProduct }; };
-template<>              struct ei_product_type_selector<Small,Large,Small>  { enum { ret = GemmProduct }; };
-template<>              struct ei_product_type_selector<Large,Large,Small>  { enum { ret = GemmProduct }; };
-
-/** \class ProductReturnType
-  * \ingroup Core_Module
-  *
-  * \brief Helper class to get the correct and optimized returned type of operator*
-  *
-  * \param Lhs the type of the left-hand side
-  * \param Rhs the type of the right-hand side
-  * \param ProductMode the type of the product (determined automatically by ei_product_mode)
-  *
-  * This class defines the typename Type representing the optimized product expression
-  * between two matrix expressions. In practice, using ProductReturnType<Lhs,Rhs>::Type
-  * is the recommended way to define the result type of a function returning an expression
-  * which involve a matrix product. The class Product should never be
-  * used directly.
-  *
-  * \sa class Product, MatrixBase::operator*(const MatrixBase<OtherDerived>&)
-  */
-template<typename Lhs, typename Rhs, int ProductType>
-struct ProductReturnType
-{
-  // TODO use the nested type to reduce instanciations ????
-//   typedef typename ei_nested<Lhs,Rhs::ColsAtCompileTime>::type LhsNested;
-//   typedef typename ei_nested<Rhs,Lhs::RowsAtCompileTime>::type RhsNested;
-
-  typedef GeneralProduct<Lhs/*Nested*/, Rhs/*Nested*/, ProductType> Type;
-};
-
-template<typename Lhs, typename Rhs>
-struct ProductReturnType<Lhs,Rhs,CoeffBasedProductMode>
-{
-  typedef typename ei_nested<Lhs, Rhs::ColsAtCompileTime, typename ei_plain_matrix_type<Lhs>::type >::type LhsNested;
-  typedef typename ei_nested<Rhs, Lhs::RowsAtCompileTime, typename ei_plain_matrix_type<Rhs>::type >::type RhsNested;
-  typedef CoeffBasedProduct<LhsNested, RhsNested, EvalBeforeAssigningBit | EvalBeforeNestingBit> Type;
-};
-
-template<typename Lhs, typename Rhs>
-struct ProductReturnType<Lhs,Rhs,LazyCoeffBasedProductMode>
-{
-  typedef typename ei_nested<Lhs, Rhs::ColsAtCompileTime, typename ei_plain_matrix_type<Lhs>::type >::type LhsNested;
-  typedef typename ei_nested<Rhs, Lhs::RowsAtCompileTime, typename ei_plain_matrix_type<Rhs>::type >::type RhsNested;
-  typedef CoeffBasedProduct<LhsNested, RhsNested, NestByRefBit> Type;
-};
-
-// this is a workaround for sun CC
-template<typename Lhs, typename Rhs>
-struct LazyProductReturnType : public ProductReturnType<Lhs,Rhs,LazyCoeffBasedProductMode>
-{};
-
-/***********************************************************************
-*  Implementation of Inner Vector Vector Product
-***********************************************************************/
-
-// FIXME : maybe the "inner product" could return a Scalar
-// instead of a 1x1 matrix ??
-// Pro: more natural for the user
-// Cons: this could be a problem if in a meta unrolled algorithm a matrix-matrix
-// product ends up to a row-vector times col-vector product... To tackle this use
-// case, we could have a specialization for Block<MatrixType,1,1> with: operator=(Scalar x);
-
-template<typename Lhs, typename Rhs>
-struct ei_traits<GeneralProduct<Lhs,Rhs,InnerProduct> >
- : ei_traits<Matrix<typename ei_scalar_product_traits<typename Lhs::Scalar, typename Rhs::Scalar>::ReturnType,1,1> >
-{};
-
-template<typename Lhs, typename Rhs>
-class GeneralProduct<Lhs, Rhs, InnerProduct>
-  : ei_no_assignment_operator,
-    public Matrix<typename ei_scalar_product_traits<typename Lhs::Scalar, typename Rhs::Scalar>::ReturnType,1,1>
-{
-    typedef Matrix<typename ei_scalar_product_traits<typename Lhs::Scalar, typename Rhs::Scalar>::ReturnType,1,1> Base;
-  public:
-    GeneralProduct(const Lhs& lhs, const Rhs& rhs)
-    {
-      EIGEN_STATIC_ASSERT((ei_is_same_type<typename Lhs::RealScalar, typename Rhs::RealScalar>::ret),
-        YOU_MIXED_DIFFERENT_NUMERIC_TYPES__YOU_NEED_TO_USE_THE_CAST_METHOD_OF_MATRIXBASE_TO_CAST_NUMERIC_TYPES_EXPLICITLY)
-
-      Base::coeffRef(0,0) = (lhs.transpose().cwiseProduct(rhs)).sum();
-    }
-
-    typename Base::Scalar value() const { return Base::coeff(0,0); }
-
-    /** Convertion to scalar */
-    operator const typename Base::Scalar() const {
-      return Base::coeff(0,0);
-    }
-};
-
-/***********************************************************************
-*  Implementation of Outer Vector Vector Product
-***********************************************************************/
-template<int StorageOrder> struct ei_outer_product_selector;
-
-template<typename Lhs, typename Rhs>
-struct ei_traits<GeneralProduct<Lhs,Rhs,OuterProduct> >
- : ei_traits<ProductBase<GeneralProduct<Lhs,Rhs,OuterProduct>, Lhs, Rhs> >
-{};
-
-template<typename Lhs, typename Rhs>
-class GeneralProduct<Lhs, Rhs, OuterProduct>
-  : public ProductBase<GeneralProduct<Lhs,Rhs,OuterProduct>, Lhs, Rhs>
-{
-  public:
-    EIGEN_PRODUCT_PUBLIC_INTERFACE(GeneralProduct)
-
-    GeneralProduct(const Lhs& lhs, const Rhs& rhs) : Base(lhs,rhs)
-    {
-      EIGEN_STATIC_ASSERT((ei_is_same_type<typename Lhs::RealScalar, typename Rhs::RealScalar>::ret),
-        YOU_MIXED_DIFFERENT_NUMERIC_TYPES__YOU_NEED_TO_USE_THE_CAST_METHOD_OF_MATRIXBASE_TO_CAST_NUMERIC_TYPES_EXPLICITLY)
-    }
-
-    template<typename Dest> void scaleAndAddTo(Dest& dest, Scalar alpha) const
-    {
-      ei_outer_product_selector<(int(Dest::Flags)&RowMajorBit) ? RowMajor : ColMajor>::run(*this, dest, alpha);
-    }
-};
-
-template<> struct ei_outer_product_selector<ColMajor> {
-  template<typename ProductType, typename Dest>
-  static EIGEN_DONT_INLINE void run(const ProductType& prod, Dest& dest, typename ProductType::Scalar alpha) {
-    typedef typename Dest::Index Index;
-    // FIXME make sure lhs is sequentially stored
-    // FIXME not very good if rhs is real and lhs complex while alpha is real too
-    const Index cols = dest.cols();
-    for (Index j=0; j<cols; ++j)
-      dest.col(j) += (alpha * prod.rhs().coeff(j)) * prod.lhs();
-  }
-};
-
-template<> struct ei_outer_product_selector<RowMajor> {
-  template<typename ProductType, typename Dest>
-  static EIGEN_DONT_INLINE void run(const ProductType& prod, Dest& dest, typename ProductType::Scalar alpha) {
-    typedef typename Dest::Index Index;
-    // FIXME make sure rhs is sequentially stored
-    // FIXME not very good if lhs is real and rhs complex while alpha is real too
-    const Index rows = dest.rows();
-    for (Index i=0; i<rows; ++i)
-      dest.row(i) += (alpha * prod.lhs().coeff(i)) * prod.rhs();
-  }
-};
-
-/***********************************************************************
-*  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.
- */
-template<typename Lhs, typename Rhs>
-struct ei_traits<GeneralProduct<Lhs,Rhs,GemvProduct> >
- : ei_traits<ProductBase<GeneralProduct<Lhs,Rhs,GemvProduct>, Lhs, Rhs> >
-{};
-
-template<int Side, int StorageOrder, bool BlasCompatible>
-struct ei_gemv_selector;
-
-template<typename Lhs, typename Rhs>
-class GeneralProduct<Lhs, Rhs, GemvProduct>
-  : public ProductBase<GeneralProduct<Lhs,Rhs,GemvProduct>, Lhs, Rhs>
-{
-  public:
-    EIGEN_PRODUCT_PUBLIC_INTERFACE(GeneralProduct)
-
-    typedef typename Lhs::Scalar LhsScalar;
-    typedef typename Rhs::Scalar RhsScalar;
-
-    GeneralProduct(const Lhs& lhs, const Rhs& rhs) : Base(lhs,rhs)
-    {
-//       EIGEN_STATIC_ASSERT((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)
-    }
-
-    enum { Side = Lhs::IsVectorAtCompileTime ? OnTheLeft : OnTheRight };
-    typedef typename ei_meta_if<int(Side)==OnTheRight,_LhsNested,_RhsNested>::ret MatrixType;
-
-    template<typename Dest> void scaleAndAddTo(Dest& dst, Scalar alpha) const
-    {
-      ei_assert(m_lhs.rows() == dst.rows() && m_rhs.cols() == dst.cols());
-      ei_gemv_selector<Side,(int(MatrixType::Flags)&RowMajorBit) ? RowMajor : ColMajor,
-                       bool(ei_blas_traits<MatrixType>::HasUsableDirectAccess)>::run(*this, dst, alpha);
-    }
-};
-
-// The vector is on the left => transposition
-template<int StorageOrder, bool BlasCompatible>
-struct ei_gemv_selector<OnTheLeft,StorageOrder,BlasCompatible>
-{
-  template<typename ProductType, typename Dest>
-  static void run(const ProductType& prod, Dest& dest, typename ProductType::Scalar alpha)
-  {
-    Transpose<Dest> destT(dest);
-    enum { OtherStorageOrder = StorageOrder == RowMajor ? ColMajor : RowMajor };
-    ei_gemv_selector<OnTheRight,OtherStorageOrder,BlasCompatible>
-      ::run(GeneralProduct<Transpose<typename ProductType::_RhsNested>,Transpose<typename ProductType::_LhsNested>, GemvProduct>
-        (prod.rhs().transpose(), prod.lhs().transpose()), destT, alpha);
-  }
-};
-
-template<> struct ei_gemv_selector<OnTheRight,ColMajor,true>
-{
-  template<typename ProductType, typename Dest>
-  static void run(const ProductType& prod, Dest& dest, typename ProductType::Scalar alpha)
-  {
-    typedef typename ProductType::Index Index;
-    typedef typename ProductType::LhsScalar   LhsScalar;
-    typedef typename ProductType::RhsScalar   RhsScalar;
-    typedef typename ProductType::Scalar      ResScalar;
-    typedef typename ProductType::RealScalar  RealScalar;
-    typedef typename ProductType::ActualLhsType ActualLhsType;
-    typedef typename ProductType::ActualRhsType ActualRhsType;
-    typedef typename ProductType::LhsBlasTraits LhsBlasTraits;
-    typedef typename ProductType::RhsBlasTraits RhsBlasTraits;
-    typedef Map<Matrix<ResScalar,Dynamic,1>, Aligned> MappedDest;
-
-    ActualLhsType actualLhs = LhsBlasTraits::extract(prod.lhs());
-    ActualRhsType actualRhs = RhsBlasTraits::extract(prod.rhs());
-
-    ResScalar actualAlpha = alpha * LhsBlasTraits::extractScalarFactor(prod.lhs())
-                                  * RhsBlasTraits::extractScalarFactor(prod.rhs());
-
-    enum {
-      // FIXME find a way to allow an inner stride on the result if ei_packet_traits<Scalar>::size==1
-      EvalToDestAtCompileTime = Dest::InnerStrideAtCompileTime==1,
-      ComplexByReal = (NumTraits<LhsScalar>::IsComplex) && (!NumTraits<RhsScalar>::IsComplex)
-    };
-
-    bool alphaIsCompatible = (!ComplexByReal) || (ei_imag(actualAlpha)==RealScalar(0));
-    bool evalToDest = EvalToDestAtCompileTime && alphaIsCompatible;
+    RowsAtCompileTime    = LhsTraits::RowsAtCompileTime,
+    ColsAtCompileTime    = RhsTraits::ColsAtCompileTime,
+    MaxRowsAtCompileTime = LhsTraits::MaxRowsAtCompileTime,
+    MaxColsAtCompileTime = RhsTraits::MaxColsAtCompileTime,
     
-    RhsScalar compatibleAlpha = ei_get_factor<ResScalar,RhsScalar>::run(actualAlpha);
+    // FIXME: only needed by GeneralMatrixMatrixTriangular
+    InnerSize = EIGEN_SIZE_MIN_PREFER_FIXED(LhsTraits::ColsAtCompileTime, RhsTraits::RowsAtCompileTime),
+    
+    // The storage order is somewhat arbitrary here. The correct one will be determined through the evaluator.
+    Flags = (MaxRowsAtCompileTime==1 && MaxColsAtCompileTime!=1) ? RowMajorBit
+          : (MaxColsAtCompileTime==1 && MaxRowsAtCompileTime!=1) ? 0
+          : (   ((LhsTraits::Flags&NoPreferredStorageOrderBit) && (RhsTraits::Flags&RowMajorBit))
+             || ((RhsTraits::Flags&NoPreferredStorageOrderBit) && (LhsTraits::Flags&RowMajorBit)) ) ? RowMajorBit
+          : NoPreferredStorageOrderBit
+  };
+};
 
-    ResScalar* actualDest;
-    if (evalToDest)
+} // end namespace internal
+
+
+template<typename _Lhs, typename _Rhs, int Option>
+class Product : public ProductImpl<_Lhs,_Rhs,Option,
+                                   typename internal::product_promote_storage_type<typename internal::traits<_Lhs>::StorageKind,
+                                                                                   typename internal::traits<_Rhs>::StorageKind,
+                                                                                   internal::product_type<_Lhs,_Rhs>::ret>::ret>
+{
+  public:
+    
+    typedef _Lhs Lhs;
+    typedef _Rhs Rhs;
+    
+    typedef typename ProductImpl<
+        Lhs, Rhs, Option,
+        typename internal::product_promote_storage_type<typename internal::traits<Lhs>::StorageKind,
+                                                        typename internal::traits<Rhs>::StorageKind,
+                                                        internal::product_type<Lhs,Rhs>::ret>::ret>::Base Base;
+    EIGEN_GENERIC_PUBLIC_INTERFACE(Product)
+
+    typedef typename internal::ref_selector<Lhs>::type LhsNested;
+    typedef typename internal::ref_selector<Rhs>::type RhsNested;
+    typedef typename internal::remove_all<LhsNested>::type LhsNestedCleaned;
+    typedef typename internal::remove_all<RhsNested>::type RhsNestedCleaned;
+
+    EIGEN_DEVICE_FUNC Product(const Lhs& lhs, const Rhs& rhs) : m_lhs(lhs), m_rhs(rhs)
     {
-      actualDest = &dest.coeffRef(0);
-    }
-    else
-    {
-      actualDest = ei_aligned_stack_new(ResScalar,dest.size());
-      if(!alphaIsCompatible)
-      {
-        MappedDest(actualDest, dest.size()).setZero();
-        compatibleAlpha = RhsScalar(1);
-      }
-      else
-        MappedDest(actualDest, dest.size()) = dest;
+      eigen_assert(lhs.cols() == rhs.rows()
+        && "invalid matrix product"
+        && "if you wanted a coeff-wise or a dot product use the respective explicit functions");
     }
 
-    ei_general_matrix_vector_product
-      <Index,LhsScalar,ColMajor,LhsBlasTraits::NeedToConjugate,RhsScalar,RhsBlasTraits::NeedToConjugate>::run(
-        actualLhs.rows(), actualLhs.cols(),
-        &actualLhs.const_cast_derived().coeffRef(0,0), actualLhs.outerStride(),
-        actualRhs.data(), actualRhs.innerStride(),
-        actualDest, 1,
-        compatibleAlpha);
+    EIGEN_DEVICE_FUNC inline Index rows() const { return m_lhs.rows(); }
+    EIGEN_DEVICE_FUNC inline Index cols() const { return m_rhs.cols(); }
 
-    if (!evalToDest)
-    {
-      if(!alphaIsCompatible)
-        dest += actualAlpha * MappedDest(actualDest, dest.size());
-      else
-        dest = MappedDest(actualDest, dest.size());
-      ei_aligned_stack_delete(ResScalar, actualDest, dest.size());
-    }
+    EIGEN_DEVICE_FUNC const LhsNestedCleaned& lhs() const { return m_lhs; }
+    EIGEN_DEVICE_FUNC const RhsNestedCleaned& rhs() const { return m_rhs; }
+
+  protected:
+
+    LhsNested m_lhs;
+    RhsNested m_rhs;
+};
+
+namespace internal {
+  
+template<typename Lhs, typename Rhs, int Option, int ProductTag = internal::product_type<Lhs,Rhs>::ret>
+class dense_product_base
+ : public internal::dense_xpr_base<Product<Lhs,Rhs,Option> >::type
+{};
+
+/** Convertion to scalar for inner-products */
+template<typename Lhs, typename Rhs, int Option>
+class dense_product_base<Lhs, Rhs, Option, InnerProduct>
+ : public internal::dense_xpr_base<Product<Lhs,Rhs,Option> >::type
+{
+  typedef Product<Lhs,Rhs,Option> ProductXpr;
+  typedef typename internal::dense_xpr_base<ProductXpr>::type Base;
+public:
+  using Base::derived;
+  typedef typename Base::Scalar Scalar;
+  
+  operator const Scalar() const
+  {
+    return internal::evaluator<ProductXpr>(derived()).coeff(0,0);
   }
 };
 
-template<> struct ei_gemv_selector<OnTheRight,RowMajor,true>
+} // namespace internal
+
+// Generic API dispatcher
+template<typename Lhs, typename Rhs, int Option, typename StorageKind>
+class ProductImpl : public internal::generic_xpr_base<Product<Lhs,Rhs,Option>, MatrixXpr, StorageKind>::type
 {
-  template<typename ProductType, typename Dest>
-  static void run(const ProductType& prod, Dest& dest, typename ProductType::Scalar alpha)
-  {
-    typedef typename ProductType::LhsScalar LhsScalar;
-    typedef typename ProductType::RhsScalar RhsScalar;
-    typedef typename ProductType::Scalar    ResScalar;
-    typedef typename ProductType::Index Index;
-    typedef typename ProductType::ActualLhsType ActualLhsType;
-    typedef typename ProductType::ActualRhsType ActualRhsType;
-    typedef typename ProductType::_ActualRhsType _ActualRhsType;
-    typedef typename ProductType::LhsBlasTraits LhsBlasTraits;
-    typedef typename ProductType::RhsBlasTraits RhsBlasTraits;
-
-    ActualLhsType actualLhs = LhsBlasTraits::extract(prod.lhs());
-    ActualRhsType actualRhs = RhsBlasTraits::extract(prod.rhs());
-
-    ResScalar actualAlpha = alpha * LhsBlasTraits::extractScalarFactor(prod.lhs())
-                                  * RhsBlasTraits::extractScalarFactor(prod.rhs());
+  public:
+    typedef typename internal::generic_xpr_base<Product<Lhs,Rhs,Option>, MatrixXpr, StorageKind>::type Base;
+};
 
+template<typename Lhs, typename Rhs, int Option>
+class ProductImpl<Lhs,Rhs,Option,Dense>
+  : public internal::dense_product_base<Lhs,Rhs,Option>
+{
+    typedef Product<Lhs, Rhs, Option> Derived;
+    
+  public:
+    
+    typedef typename internal::dense_product_base<Lhs, Rhs, Option> Base;
+    EIGEN_DENSE_PUBLIC_INTERFACE(Derived)
+  protected:
     enum {
-      // FIXME I think here we really have to check for ei_packet_traits<Scalar>::size==1
-      // because in this case it is fine to have an inner stride
-      DirectlyUseRhs = ((ei_packet_traits<RhsScalar>::size==1) || (_ActualRhsType::Flags&ActualPacketAccessBit))
-                     && (!(_ActualRhsType::Flags & RowMajorBit))
+      IsOneByOne = (RowsAtCompileTime == 1 || RowsAtCompileTime == Dynamic) && 
+                   (ColsAtCompileTime == 1 || ColsAtCompileTime == Dynamic),
+      EnableCoeff = IsOneByOne || Option==LazyProduct
     };
-
-    RhsScalar* rhs_data;
-    if (DirectlyUseRhs)
-       rhs_data = &actualRhs.const_cast_derived().coeffRef(0);
-    else
+    
+  public:
+  
+    EIGEN_DEVICE_FUNC Scalar coeff(Index row, Index col) const
     {
-      rhs_data = ei_aligned_stack_new(RhsScalar, actualRhs.size());
-      Map<typename _ActualRhsType::PlainObject>(rhs_data, actualRhs.size()) = actualRhs;
+      EIGEN_STATIC_ASSERT(EnableCoeff, THIS_METHOD_IS_ONLY_FOR_INNER_OR_LAZY_PRODUCTS);
+      eigen_assert( (Option==LazyProduct) || (this->rows() == 1 && this->cols() == 1) );
+      
+      return internal::evaluator<Derived>(derived()).coeff(row,col);
     }
 
-    ei_general_matrix_vector_product
-      <Index,LhsScalar,RowMajor,LhsBlasTraits::NeedToConjugate,RhsScalar,RhsBlasTraits::NeedToConjugate>::run(
-        actualLhs.rows(), actualLhs.cols(),
-        &actualLhs.const_cast_derived().coeffRef(0,0), actualLhs.outerStride(),
-        rhs_data, 1,
-        &dest.coeffRef(0,0), dest.innerStride(),
-        actualAlpha);
-
-    if (!DirectlyUseRhs) ei_aligned_stack_delete(RhsScalar, rhs_data, prod.rhs().size());
-  }
+    EIGEN_DEVICE_FUNC Scalar coeff(Index i) const
+    {
+      EIGEN_STATIC_ASSERT(EnableCoeff, THIS_METHOD_IS_ONLY_FOR_INNER_OR_LAZY_PRODUCTS);
+      eigen_assert( (Option==LazyProduct) || (this->rows() == 1 && this->cols() == 1) );
+      
+      return internal::evaluator<Derived>(derived()).coeff(i);
+    }
+    
+  
 };
 
-template<> struct ei_gemv_selector<OnTheRight,ColMajor,false>
-{
-  template<typename ProductType, typename Dest>
-  static void run(const ProductType& prod, Dest& dest, typename ProductType::Scalar alpha)
-  {
-    typedef typename Dest::Index Index;
-    // TODO makes sure dest is sequentially stored in memory, otherwise use a temp
-    const Index size = prod.rhs().rows();
-    for(Index k=0; k<size; ++k)
-      dest += (alpha*prod.rhs().coeff(k)) * prod.lhs().col(k);
-  }
-};
-
-template<> struct ei_gemv_selector<OnTheRight,RowMajor,false>
-{
-  template<typename ProductType, typename Dest>
-  static void run(const ProductType& prod, Dest& dest, typename ProductType::Scalar alpha)
-  {
-    typedef typename Dest::Index Index;
-    // TODO makes sure rhs is sequentially stored in memory, otherwise use a temp
-    const Index rows = prod.rows();
-    for(Index i=0; i<rows; ++i)
-      dest.coeffRef(i) += alpha * (prod.lhs().row(i).cwiseProduct(prod.rhs().transpose())).sum();
-  }
-};
-
-/***************************************************************************
-* Implementation of matrix base methods
-***************************************************************************/
-
-/** \returns the matrix product of \c *this and \a other.
-  *
-  * \note If instead of the matrix product you want the coefficient-wise product, see Cwise::operator*().
-  *
-  * \sa lazyProduct(), operator*=(const MatrixBase&), Cwise::operator*()
-  */
-template<typename Derived>
-template<typename OtherDerived>
-inline const typename ProductReturnType<Derived,OtherDerived>::Type
-MatrixBase<Derived>::operator*(const MatrixBase<OtherDerived> &other) const
-{
-  // A note regarding the function declaration: In MSVC, this function will sometimes
-  // not be inlined since ei_matrix_storage 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
-  ei_product_type<Derived,OtherDerived>::debug();
-#endif
-  return typename ProductReturnType<Derived,OtherDerived>::Type(derived(), other.derived());
-}
-
-/** \returns an expression of the matrix product of \c *this and \a other without implicit evaluation.
-  *
-  * The returned product will behave like any other expressions: the coefficients of the product will be
-  * computed once at a time as requested. This might be useful in some extremely rare cases when only
-  * a small and no coherent fraction of the result's coefficients have to be computed.
-  *
-  * \warning This version of the matrix product can be much much slower. So use it only if you know
-  * what you are doing and that you measured a true speed improvement.
-  *
-  * \sa operator*(const MatrixBase&)
-  */
-template<typename Derived>
-template<typename OtherDerived>
-const typename LazyProductReturnType<Derived,OtherDerived>::Type
-MatrixBase<Derived>::lazyProduct(const MatrixBase<OtherDerived> &other) const
-{
-  enum {
-    ProductIsValid =  Derived::ColsAtCompileTime==Dynamic
-                   || OtherDerived::RowsAtCompileTime==Dynamic
-                   || int(Derived::ColsAtCompileTime)==int(OtherDerived::RowsAtCompileTime),
-    AreVectors = Derived::IsVectorAtCompileTime && OtherDerived::IsVectorAtCompileTime,
-    SameSizes = EIGEN_PREDICATE_SAME_MATRIX_SIZE(Derived,OtherDerived)
-  };
-  // note to the lost user:
-  //    * for a dot product use: v1.dot(v2)
-  //    * for a coeff-wise product use: v1.cwiseProduct(v2)
-  EIGEN_STATIC_ASSERT(ProductIsValid || !(AreVectors && SameSizes),
-    INVALID_VECTOR_VECTOR_PRODUCT__IF_YOU_WANTED_A_DOT_OR_COEFF_WISE_PRODUCT_YOU_MUST_USE_THE_EXPLICIT_FUNCTIONS)
-  EIGEN_STATIC_ASSERT(ProductIsValid || !(SameSizes && !AreVectors),
-    INVALID_MATRIX_PRODUCT__IF_YOU_WANTED_A_COEFF_WISE_PRODUCT_YOU_MUST_USE_THE_EXPLICIT_FUNCTION)
-  EIGEN_STATIC_ASSERT(ProductIsValid || SameSizes, INVALID_MATRIX_PRODUCT)
-
-  return typename LazyProductReturnType<Derived,OtherDerived>::Type(derived(), other.derived());
-}
+} // end namespace Eigen
 
 #endif // EIGEN_PRODUCT_H

Eigen/src/Core/ProductBase.h

@@ -1,253 +0,0 @@
-// This file is part of Eigen, a lightweight C++ template library
-// for linear algebra.
-//
-// Copyright (C) 2009-2010 Gael Guennebaud <gael.guennebaud@inria.fr>
-//
-// Eigen is free software; you can redistribute it and/or
-// modify it under the terms of the GNU Lesser General Public
-// License as published by the Free Software Foundation; either
-// version 3 of the License, or (at your option) any later version.
-//
-// Alternatively, you can redistribute it and/or
-// modify it under the terms of the GNU General Public License as
-// published by the Free Software Foundation; either version 2 of
-// the License, or (at your option) any later version.
-//
-// Eigen is distributed in the hope that it will be useful, but WITHOUT ANY
-// WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS
-// FOR A PARTICULAR PURPOSE. See the GNU Lesser General Public License or the
-// GNU General Public License for more details.
-//
-// You should have received a copy of the GNU Lesser General Public
-// License and a copy of the GNU General Public License along with
-// Eigen. If not, see <http://www.gnu.org/licenses/>.
-
-#ifndef EIGEN_PRODUCTBASE_H
-#define EIGEN_PRODUCTBASE_H
-
-/** \class ProductBase
-  * \ingroup Core_Module
-  *
-  */
-template<typename Derived, typename _Lhs, typename _Rhs>
-struct ei_traits<ProductBase<Derived,_Lhs,_Rhs> >
-{
-  typedef MatrixXpr XprKind;
-  typedef typename ei_cleantype<_Lhs>::type Lhs;
-  typedef typename ei_cleantype<_Rhs>::type Rhs;
-  typedef typename ei_scalar_product_traits<typename Lhs::Scalar, typename Rhs::Scalar>::ReturnType Scalar;
-  typedef typename ei_promote_storage_type<typename ei_traits<Lhs>::StorageKind,
-                                           typename ei_traits<Rhs>::StorageKind>::ret StorageKind;
-  typedef typename ei_promote_index_type<typename ei_traits<Lhs>::Index,
-                                         typename ei_traits<Rhs>::Index>::type Index;
-  enum {
-    RowsAtCompileTime = ei_traits<Lhs>::RowsAtCompileTime,
-    ColsAtCompileTime = ei_traits<Rhs>::ColsAtCompileTime,
-    MaxRowsAtCompileTime = ei_traits<Lhs>::MaxRowsAtCompileTime,
-    MaxColsAtCompileTime = ei_traits<Rhs>::MaxColsAtCompileTime,
-    Flags = (MaxRowsAtCompileTime==1 ? RowMajorBit : 0)
-          | EvalBeforeNestingBit | EvalBeforeAssigningBit | NestByRefBit,
-                  // Note that EvalBeforeNestingBit and NestByRefBit
-                  // are not used in practice because ei_nested is overloaded for products
-    CoeffReadCost = 0 // FIXME why is it needed ?
-  };
-};
-
-#define EIGEN_PRODUCT_PUBLIC_INTERFACE(Derived) \
-  typedef ProductBase<Derived, Lhs, Rhs > Base; \
-  EIGEN_DENSE_PUBLIC_INTERFACE(Derived) \
-  typedef typename Base::LhsNested LhsNested; \
-  typedef typename Base::_LhsNested _LhsNested; \
-  typedef typename Base::LhsBlasTraits LhsBlasTraits; \
-  typedef typename Base::ActualLhsType ActualLhsType; \
-  typedef typename Base::_ActualLhsType _ActualLhsType; \
-  typedef typename Base::RhsNested RhsNested; \
-  typedef typename Base::_RhsNested _RhsNested; \
-  typedef typename Base::RhsBlasTraits RhsBlasTraits; \
-  typedef typename Base::ActualRhsType ActualRhsType; \
-  typedef typename Base::_ActualRhsType _ActualRhsType; \
-  using Base::m_lhs; \
-  using Base::m_rhs;
-
-template<typename Derived, typename Lhs, typename Rhs>
-class ProductBase : public MatrixBase<Derived>
-{
-  public:
-    typedef MatrixBase<Derived> Base;
-    EIGEN_DENSE_PUBLIC_INTERFACE(ProductBase)
-  protected:
-    typedef typename Lhs::Nested LhsNested;
-    typedef typename ei_cleantype<LhsNested>::type _LhsNested;
-    typedef ei_blas_traits<_LhsNested> LhsBlasTraits;
-    typedef typename LhsBlasTraits::DirectLinearAccessType ActualLhsType;
-    typedef typename ei_cleantype<ActualLhsType>::type _ActualLhsType;
-
-    typedef typename Rhs::Nested RhsNested;
-    typedef typename ei_cleantype<RhsNested>::type _RhsNested;
-    typedef ei_blas_traits<_RhsNested> RhsBlasTraits;
-    typedef typename RhsBlasTraits::DirectLinearAccessType ActualRhsType;
-    typedef typename ei_cleantype<ActualRhsType>::type _ActualRhsType;
-
-    // Diagonal of a product: no need to evaluate the arguments because they are going to be evaluated only once
-    typedef CoeffBasedProduct<LhsNested, RhsNested, 0> FullyLazyCoeffBaseProductType;
-
-  public:
-
-    typedef typename Base::PlainObject PlainObject;
-
-    ProductBase(const Lhs& lhs, const Rhs& rhs)
-      : m_lhs(lhs), m_rhs(rhs)
-    {
-      ei_assert(lhs.cols() == rhs.rows()
-        && "invalid matrix product"
-        && "if you wanted a coeff-wise or a dot product use the respective explicit functions");
-    }
-
-    inline Index rows() const { return m_lhs.rows(); }
-    inline Index cols() const { return m_rhs.cols(); }
-
-    template<typename Dest>
-    inline void evalTo(Dest& dst) const { dst.setZero(); scaleAndAddTo(dst,Scalar(1)); }
-
-    template<typename Dest>
-    inline void addTo(Dest& dst) const { scaleAndAddTo(dst,1); }
-
-    template<typename Dest>
-    inline void subTo(Dest& dst) const { scaleAndAddTo(dst,-1); }
-
-    template<typename Dest>
-    inline void scaleAndAddTo(Dest& dst,Scalar alpha) const { derived().scaleAndAddTo(dst,alpha); }
-
-    const _LhsNested& lhs() const { return m_lhs; }
-    const _RhsNested& rhs() const { return m_rhs; }
-
-    // Implicit conversion to the nested type (trigger the evaluation of the product)
-    operator const PlainObject& () const
-    {
-      m_result.resize(m_lhs.rows(), m_rhs.cols());
-      derived().evalTo(m_result);
-      return m_result;
-    }
-
-    const Diagonal<FullyLazyCoeffBaseProductType,0> diagonal() const
-    { return FullyLazyCoeffBaseProductType(m_lhs, m_rhs); }
-
-    template<int Index>
-    const Diagonal<FullyLazyCoeffBaseProductType,Index> diagonal() const
-    { return FullyLazyCoeffBaseProductType(m_lhs, m_rhs); }
-
-    const Diagonal<FullyLazyCoeffBaseProductType,Dynamic> diagonal(Index index) const
-    { return FullyLazyCoeffBaseProductType(m_lhs, m_rhs).diagonal(index); }
-
-  protected:
-
-    const LhsNested m_lhs;
-    const RhsNested m_rhs;
-
-    mutable PlainObject m_result;
-
-  private:
-
-    // discard coeff methods
-    void coeff(Index,Index) const;
-    void coeffRef(Index,Index);
-    void coeff(Index) const;
-    void coeffRef(Index);
-};
-
-// here we need to overload the nested rule for products
-// such that the nested type is a const reference to a plain matrix
-template<typename Lhs, typename Rhs, int Mode, int N, typename PlainObject>
-struct ei_nested<GeneralProduct<Lhs,Rhs,Mode>, N, PlainObject>
-{
-  typedef PlainObject const& type;
-};
-
-template<typename NestedProduct>
-class ScaledProduct;
-
-// Note that these two operator* functions are not defined as member
-// functions of ProductBase, because, otherwise we would have to
-// define all overloads defined in MatrixBase. Furthermore, Using
-// "using Base::operator*" would not work with MSVC.
-//
-// Also note that here we accept any compatible scalar types
-template<typename Derived,typename Lhs,typename Rhs>
-const ScaledProduct<Derived>
-operator*(const ProductBase<Derived,Lhs,Rhs>& prod, typename Derived::Scalar x)
-{ return ScaledProduct<Derived>(prod.derived(), x); }
-
-template<typename Derived,typename Lhs,typename Rhs>
-typename ei_enable_if<!ei_is_same_type<typename Derived::Scalar,typename Derived::RealScalar>::ret,
-                      const ScaledProduct<Derived> >::type
-operator*(const ProductBase<Derived,Lhs,Rhs>& prod, typename Derived::RealScalar x)
-{ return ScaledProduct<Derived>(prod.derived(), x); }
-
-
-template<typename Derived,typename Lhs,typename Rhs>
-const ScaledProduct<Derived>
-operator*(typename Derived::Scalar x,const ProductBase<Derived,Lhs,Rhs>& prod)
-{ return ScaledProduct<Derived>(prod.derived(), x); }
-
-template<typename Derived,typename Lhs,typename Rhs>
-typename ei_enable_if<!ei_is_same_type<typename Derived::Scalar,typename Derived::RealScalar>::ret,
-                      const ScaledProduct<Derived> >::type
-operator*(typename Derived::RealScalar x,const ProductBase<Derived,Lhs,Rhs>& prod)
-{ return ScaledProduct<Derived>(prod.derived(), x); }
-
-
-template<typename NestedProduct>
-struct ei_traits<ScaledProduct<NestedProduct> >
- : ei_traits<ProductBase<ScaledProduct<NestedProduct>,
-                         typename NestedProduct::_LhsNested,
-                         typename NestedProduct::_RhsNested> >
-{
-  typedef typename ei_traits<NestedProduct>::StorageKind StorageKind;
-};
-
-template<typename NestedProduct>
-class ScaledProduct
-  : public ProductBase<ScaledProduct<NestedProduct>,
-                       typename NestedProduct::_LhsNested,
-                       typename NestedProduct::_RhsNested>
-{
-  public:
-    typedef ProductBase<ScaledProduct<NestedProduct>,
-                       typename NestedProduct::_LhsNested,
-                       typename NestedProduct::_RhsNested> Base;
-    typedef typename Base::Scalar Scalar;
-    typedef typename Base::PlainObject PlainObject;
-//     EIGEN_PRODUCT_PUBLIC_INTERFACE(ScaledProduct)
-
-    ScaledProduct(const NestedProduct& prod, Scalar x)
-    : Base(prod.lhs(),prod.rhs()), m_prod(prod), m_alpha(x) {}
-
-    template<typename Dest>
-    inline void evalTo(Dest& dst) const { dst.setZero(); scaleAndAddTo(dst,m_alpha); }
-
-    template<typename Dest>
-    inline void addTo(Dest& dst) const { scaleAndAddTo(dst,m_alpha); }
-
-    template<typename Dest>
-    inline void subTo(Dest& dst) const { scaleAndAddTo(dst,-m_alpha); }
-
-    template<typename Dest>
-    inline void scaleAndAddTo(Dest& dst,Scalar alpha) const { m_prod.derived().scaleAndAddTo(dst,alpha); }
-    
-  protected:
-    const NestedProduct& m_prod;
-    Scalar m_alpha;
-};
-
-/** \internal
-  * Overloaded to perform an efficient C = (A*B).lazy() */
-template<typename Derived>
-template<typename ProductDerived, typename Lhs, typename Rhs>
-Derived& MatrixBase<Derived>::lazyAssign(const ProductBase<ProductDerived, Lhs,Rhs>& other)
-{
-  other.derived().evalTo(derived());
-  return derived();
-}
-
-
-#endif // EIGEN_PRODUCTBASE_H

Eigen/src/Core/Random.h

@@ -3,45 +3,43 @@
 //
 // 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_RANDOM_H
 #define EIGEN_RANDOM_H
 
-template<typename Scalar> struct ei_scalar_random_op {
-  EIGEN_EMPTY_STRUCT_CTOR(ei_scalar_random_op)
+namespace Eigen { 
+
+namespace internal {
+
+template<typename Scalar> struct scalar_random_op {
+  EIGEN_EMPTY_STRUCT_CTOR(scalar_random_op)
   template<typename Index>
-  inline const Scalar operator() (Index, Index = 0) const { return ei_random<Scalar>(); }
+  inline const Scalar operator() (Index, Index = 0) const { return random<Scalar>(); }
 };
+
 template<typename Scalar>
-struct ei_functor_traits<ei_scalar_random_op<Scalar> >
+struct functor_traits<scalar_random_op<Scalar> >
 { enum { Cost = 5 * NumTraits<Scalar>::MulCost, PacketAccess = false, IsRepeatable = false }; };
 
+} // end namespace internal
+
 /** \returns a random matrix expression
   *
+  * Numbers are uniformly spread through their whole definition range for integer types,
+  * and in the [-1:1] range for floating point scalar types.
+  * 
   * 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.
   *
+  * \not_reentrant
+  * 
   * 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 Random() should be used
   * instead.
+  * 
   *
   * Example: \include MatrixBase_random_int_int.cpp
   * Output: \verbinclude MatrixBase_random_int_int.out
@@ -49,22 +47,28 @@ struct ei_functor_traits<ei_scalar_random_op<Scalar> >
   * This expression has the "evaluate before nesting" flag so that it will be evaluated into
   * a temporary matrix whenever it is nested in a larger expression. This prevents unexpected
   * behavior with expressions involving random matrices.
+  * 
+  * See DenseBase::NullaryExpr(Index, const CustomNullaryOp&) for an example using C++11 random generators.
   *
-  * \sa MatrixBase::setRandom(), MatrixBase::Random(Index), MatrixBase::Random()
+  * \sa DenseBase::setRandom(), DenseBase::Random(Index), DenseBase::Random()
   */
 template<typename Derived>
-inline const CwiseNullaryOp<ei_scalar_random_op<typename ei_traits<Derived>::Scalar>, Derived>
+inline const typename DenseBase<Derived>::RandomReturnType
 DenseBase<Derived>::Random(Index rows, Index cols)
 {
-  return NullaryExpr(rows, cols, ei_scalar_random_op<Scalar>());
+  return NullaryExpr(rows, cols, internal::scalar_random_op<Scalar>());
 }
 
 /** \returns a random vector expression
+  *
+  * Numbers are uniformly spread through their whole definition range for integer types,
+  * and in the [-1:1] range for floating point scalar types.
   *
   * The parameter \a size is the size of the returned vector.
   * Must be compatible with this MatrixBase type.
   *
   * \only_for_vectors
+  * \not_reentrant
   *
   * 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 Random() should be used
@@ -77,17 +81,20 @@ DenseBase<Derived>::Random(Index rows, Index cols)
   * a temporary vector whenever it is nested in a larger expression. This prevents unexpected
   * behavior with expressions involving random matrices.
   *
-  * \sa MatrixBase::setRandom(), MatrixBase::Random(Index,Index), MatrixBase::Random()
+  * \sa DenseBase::setRandom(), DenseBase::Random(Index,Index), DenseBase::Random()
   */
 template<typename Derived>
-inline const CwiseNullaryOp<ei_scalar_random_op<typename ei_traits<Derived>::Scalar>, Derived>
+inline const typename DenseBase<Derived>::RandomReturnType
 DenseBase<Derived>::Random(Index size)
 {
-  return NullaryExpr(size, ei_scalar_random_op<Scalar>());
+  return NullaryExpr(size, internal::scalar_random_op<Scalar>());
 }
 
 /** \returns a fixed-size random matrix or vector expression
   *
+  * Numbers are uniformly spread through their whole definition range for integer types,
+  * and in the [-1:1] range for floating point scalar types.
+  * 
   * This variant is only for fixed-size MatrixBase types. For dynamic-size types, you
   * need to use the variants taking size arguments.
   *
@@ -97,18 +104,25 @@ DenseBase<Derived>::Random(Index size)
   * This expression has the "evaluate before nesting" flag so that it will be evaluated into
   * a temporary matrix whenever it is nested in a larger expression. This prevents unexpected
   * behavior with expressions involving random matrices.
+  * 
+  * \not_reentrant
   *
-  * \sa MatrixBase::setRandom(), MatrixBase::Random(Index,Index), MatrixBase::Random(Index)
+  * \sa DenseBase::setRandom(), DenseBase::Random(Index,Index), DenseBase::Random(Index)
   */
 template<typename Derived>
-inline const CwiseNullaryOp<ei_scalar_random_op<typename ei_traits<Derived>::Scalar>, Derived>
+inline const typename DenseBase<Derived>::RandomReturnType
 DenseBase<Derived>::Random()
 {
-  return NullaryExpr(RowsAtCompileTime, ColsAtCompileTime, ei_scalar_random_op<Scalar>());
+  return NullaryExpr(RowsAtCompileTime, ColsAtCompileTime, internal::scalar_random_op<Scalar>());
 }
 
 /** Sets all coefficients in this expression to random values.
   *
+  * Numbers are uniformly spread through their whole definition range for integer types,
+  * and in the [-1:1] range for floating point scalar types.
+  * 
+  * \not_reentrant
+  * 
   * Example: \include MatrixBase_setRandom.cpp
   * Output: \verbinclude MatrixBase_setRandom.out
   *
@@ -120,39 +134,50 @@ inline Derived& DenseBase<Derived>::setRandom()
   return *this = Random(rows(), cols());
 }
 
-/** Resizes to the given \a size, and sets all coefficients in this expression to random values.
+/** Resizes to the given \a newSize, and sets all coefficients in this expression to random values.
   *
+  * Numbers are uniformly spread through their whole definition range for integer types,
+  * and in the [-1:1] range for floating point scalar types.
+  * 
   * \only_for_vectors
+  * \not_reentrant
   *
   * Example: \include Matrix_setRandom_int.cpp
   * Output: \verbinclude Matrix_setRandom_int.out
   *
-  * \sa MatrixBase::setRandom(), setRandom(Index,Index), class CwiseNullaryOp, MatrixBase::Random()
+  * \sa DenseBase::setRandom(), setRandom(Index,Index), class CwiseNullaryOp, DenseBase::Random()
   */
 template<typename Derived>
 EIGEN_STRONG_INLINE Derived&
-DenseStorageBase<Derived>::setRandom(Index size)
+PlainObjectBase<Derived>::setRandom(Index newSize)
 {
-  resize(size);
+  resize(newSize);
   return setRandom();
 }
 
 /** Resizes to the given size, and sets all coefficients in this expression to random values.
   *
+  * Numbers are uniformly spread through their whole definition range for integer types,
+  * and in the [-1:1] range for floating point scalar types.
+  *
+  * \not_reentrant
+  * 
   * \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(Index), class CwiseNullaryOp, MatrixBase::Random()
+  * \sa DenseBase::setRandom(), setRandom(Index), class CwiseNullaryOp, DenseBase::Random()
   */
 template<typename Derived>
 EIGEN_STRONG_INLINE Derived&
-DenseStorageBase<Derived>::setRandom(Index rows, Index cols)
+PlainObjectBase<Derived>::setRandom(Index rows, Index cols)
 {
   resize(rows, cols);
   return setRandom();
 }
 
+} // end namespace Eigen
+
 #endif // EIGEN_RANDOM_H

Eigen/src/Core/Redux.h

@@ -4,28 +4,17 @@
 // 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_REDUX_H
 #define EIGEN_REDUX_H
 
+namespace Eigen { 
+
+namespace internal {
+
 // TODO
 //  * implement other kind of vectorization
 //  * factorize code
@@ -35,11 +24,11 @@
 ***************************************************************************/
 
 template<typename Func, typename Derived>
-struct ei_redux_traits
+struct redux_traits
 {
 public:
   enum {
-    PacketSize = ei_packet_traits<typename Derived::Scalar>::size,
+    PacketSize = packet_traits<typename Derived::Scalar>::size,
     InnerMaxSize = int(Derived::IsRowMajor)
                  ? Derived::MaxColsAtCompileTime
                  : Derived::MaxRowsAtCompileTime
@@ -47,7 +36,7 @@ public:
 
   enum {
     MightVectorize = (int(Derived::Flags)&ActualPacketAccessBit)
-                  && (ei_functor_traits<Func>::PacketAccess),
+                  && (functor_traits<Func>::PacketAccess),
     MayLinearVectorize = MightVectorize && (int(Derived::Flags)&LinearAccessBit),
     MaySliceVectorize  = MightVectorize && int(InnerMaxSize)>=3*PacketSize
   };
@@ -61,21 +50,34 @@ public:
 
 public:
   enum {
-    Cost = (  Derived::SizeAtCompileTime == Dynamic
-           || Derived::CoeffReadCost == Dynamic
-           || (Derived::SizeAtCompileTime!=1 && ei_functor_traits<Func>::Cost == Dynamic)
-           ) ? Dynamic
-           : Derived::SizeAtCompileTime * Derived::CoeffReadCost
-               + (Derived::SizeAtCompileTime-1) * ei_functor_traits<Func>::Cost,
+    Cost = Derived::SizeAtCompileTime == Dynamic ? HugeCost
+         : Derived::SizeAtCompileTime * Derived::CoeffReadCost + (Derived::SizeAtCompileTime-1) * functor_traits<Func>::Cost,
     UnrollingLimit = EIGEN_UNROLLING_LIMIT * (int(Traversal) == int(DefaultTraversal) ? 1 : int(PacketSize))
   };
 
 public:
   enum {
-    Unrolling = Cost != Dynamic && Cost <= UnrollingLimit
-              ? CompleteUnrolling
-              : NoUnrolling
+    Unrolling = Cost <= UnrollingLimit ? CompleteUnrolling : NoUnrolling
   };
+  
+#ifdef EIGEN_DEBUG_ASSIGN
+  static void debug()
+  {
+    std::cerr << "Xpr: " << typeid(typename Derived::XprType).name() << std::endl;
+    std::cerr.setf(std::ios::hex, std::ios::basefield);
+    EIGEN_DEBUG_VAR(Derived::Flags)
+    std::cerr.unsetf(std::ios::hex);
+    EIGEN_DEBUG_VAR(InnerMaxSize)
+    EIGEN_DEBUG_VAR(PacketSize)
+    EIGEN_DEBUG_VAR(MightVectorize)
+    EIGEN_DEBUG_VAR(MayLinearVectorize)
+    EIGEN_DEBUG_VAR(MaySliceVectorize)
+    EIGEN_DEBUG_VAR(Traversal)
+    EIGEN_DEBUG_VAR(UnrollingLimit)
+    EIGEN_DEBUG_VAR(Unrolling)
+    std::cerr << std::endl;
+  }
+#endif
 };
 
 /***************************************************************************
@@ -85,7 +87,7 @@ public:
 /*** no vectorization ***/
 
 template<typename Func, typename Derived, int Start, int Length>
-struct ei_redux_novec_unroller
+struct redux_novec_unroller
 {
   enum {
     HalfLength = Length/2
@@ -93,15 +95,16 @@ struct ei_redux_novec_unroller
 
   typedef typename Derived::Scalar Scalar;
 
-  EIGEN_STRONG_INLINE static Scalar run(const Derived &mat, const Func& func)
+  EIGEN_DEVICE_FUNC
+  static EIGEN_STRONG_INLINE Scalar run(const Derived &mat, const Func& func)
   {
-    return func(ei_redux_novec_unroller<Func, Derived, Start, HalfLength>::run(mat,func),
-                ei_redux_novec_unroller<Func, Derived, Start+HalfLength, Length-HalfLength>::run(mat,func));
+    return func(redux_novec_unroller<Func, Derived, Start, HalfLength>::run(mat,func),
+                redux_novec_unroller<Func, Derived, Start+HalfLength, Length-HalfLength>::run(mat,func));
   }
 };
 
 template<typename Func, typename Derived, int Start>
-struct ei_redux_novec_unroller<Func, Derived, Start, 1>
+struct redux_novec_unroller<Func, Derived, Start, 1>
 {
   enum {
     outer = Start / Derived::InnerSizeAtCompileTime,
@@ -110,7 +113,8 @@ struct ei_redux_novec_unroller<Func, Derived, Start, 1>
 
   typedef typename Derived::Scalar Scalar;
 
-  EIGEN_STRONG_INLINE static Scalar run(const Derived &mat, const Func&)
+  EIGEN_DEVICE_FUNC
+  static EIGEN_STRONG_INLINE Scalar run(const Derived &mat, const Func&)
   {
     return mat.coeffByOuterInner(outer, inner);
   }
@@ -120,49 +124,50 @@ struct ei_redux_novec_unroller<Func, Derived, Start, 1>
 // to prevent false warnings regarding failed inlining though
 // for 0 length run() will never be called at all.
 template<typename Func, typename Derived, int Start>
-struct ei_redux_novec_unroller<Func, Derived, Start, 0>
+struct redux_novec_unroller<Func, Derived, Start, 0>
 {
   typedef typename Derived::Scalar Scalar;
-  EIGEN_STRONG_INLINE static Scalar run(const Derived&, const Func&) { return Scalar(); }
+  EIGEN_DEVICE_FUNC 
+  static EIGEN_STRONG_INLINE Scalar run(const Derived&, const Func&) { return Scalar(); }
 };
 
 /*** vectorization ***/
 
 template<typename Func, typename Derived, int Start, int Length>
-struct ei_redux_vec_unroller
+struct redux_vec_unroller
 {
   enum {
-    PacketSize = ei_packet_traits<typename Derived::Scalar>::size,
+    PacketSize = packet_traits<typename Derived::Scalar>::size,
     HalfLength = Length/2
   };
 
   typedef typename Derived::Scalar Scalar;
-  typedef typename ei_packet_traits<Scalar>::type PacketScalar;
+  typedef typename packet_traits<Scalar>::type PacketScalar;
 
-  EIGEN_STRONG_INLINE static PacketScalar run(const Derived &mat, const Func& func)
+  static EIGEN_STRONG_INLINE PacketScalar run(const Derived &mat, const Func& func)
   {
     return func.packetOp(
-            ei_redux_vec_unroller<Func, Derived, Start, HalfLength>::run(mat,func),
-            ei_redux_vec_unroller<Func, Derived, Start+HalfLength, Length-HalfLength>::run(mat,func) );
+            redux_vec_unroller<Func, Derived, Start, HalfLength>::run(mat,func),
+            redux_vec_unroller<Func, Derived, Start+HalfLength, Length-HalfLength>::run(mat,func) );
   }
 };
 
 template<typename Func, typename Derived, int Start>
-struct ei_redux_vec_unroller<Func, Derived, Start, 1>
+struct redux_vec_unroller<Func, Derived, Start, 1>
 {
   enum {
-    index = Start * ei_packet_traits<typename Derived::Scalar>::size,
+    index = Start * packet_traits<typename Derived::Scalar>::size,
     outer = index / int(Derived::InnerSizeAtCompileTime),
     inner = index % int(Derived::InnerSizeAtCompileTime),
-    alignment = (Derived::Flags & AlignedBit) ? Aligned : Unaligned
+    alignment = Derived::Alignment
   };
 
   typedef typename Derived::Scalar Scalar;
-  typedef typename ei_packet_traits<Scalar>::type PacketScalar;
+  typedef typename packet_traits<Scalar>::type PacketScalar;
 
-  EIGEN_STRONG_INLINE static PacketScalar run(const Derived &mat, const Func&)
+  static EIGEN_STRONG_INLINE PacketScalar run(const Derived &mat, const Func&)
   {
-    return mat.template packetByOuterInner<alignment>(outer, inner);
+    return mat.template packetByOuterInner<alignment,PacketScalar>(outer, inner);
   }
 };
 
@@ -171,19 +176,19 @@ struct ei_redux_vec_unroller<Func, Derived, Start, 1>
 ***************************************************************************/
 
 template<typename Func, typename Derived,
-         int Traversal = ei_redux_traits<Func, Derived>::Traversal,
-         int Unrolling = ei_redux_traits<Func, Derived>::Unrolling
+         int Traversal = redux_traits<Func, Derived>::Traversal,
+         int Unrolling = redux_traits<Func, Derived>::Unrolling
 >
-struct ei_redux_impl;
+struct redux_impl;
 
 template<typename Func, typename Derived>
-struct ei_redux_impl<Func, Derived, DefaultTraversal, NoUnrolling>
+struct redux_impl<Func, Derived, DefaultTraversal, NoUnrolling>
 {
   typedef typename Derived::Scalar Scalar;
-  typedef typename Derived::Index Index;
-  static EIGEN_STRONG_INLINE Scalar run(const Derived& mat, const Func& func)
+  EIGEN_DEVICE_FUNC
+  static EIGEN_STRONG_INLINE Scalar run(const Derived &mat, const Func& func)
   {
-    ei_assert(mat.rows()>0 && mat.cols()>0 && "you are using an empty matrix");
+    eigen_assert(mat.rows()>0 && mat.cols()>0 && "you are using an empty matrix");
     Scalar res;
     res = mat.coeffByOuterInner(0, 0);
     for(Index i = 1; i < mat.innerSize(); ++i)
@@ -196,36 +201,49 @@ struct ei_redux_impl<Func, Derived, DefaultTraversal, NoUnrolling>
 };
 
 template<typename Func, typename Derived>
-struct ei_redux_impl<Func,Derived, DefaultTraversal, CompleteUnrolling>
-  : public ei_redux_novec_unroller<Func,Derived, 0, Derived::SizeAtCompileTime>
+struct redux_impl<Func,Derived, DefaultTraversal, CompleteUnrolling>
+  : public redux_novec_unroller<Func,Derived, 0, Derived::SizeAtCompileTime>
 {};
 
 template<typename Func, typename Derived>
-struct ei_redux_impl<Func, Derived, LinearVectorizedTraversal, NoUnrolling>
+struct redux_impl<Func, Derived, LinearVectorizedTraversal, NoUnrolling>
 {
   typedef typename Derived::Scalar Scalar;
-  typedef typename ei_packet_traits<Scalar>::type PacketScalar;
-  typedef typename Derived::Index Index;
+  typedef typename packet_traits<Scalar>::type PacketScalar;
 
-  static Scalar run(const Derived& mat, const Func& func)
+  static Scalar run(const Derived &mat, const Func& func)
   {
     const Index size = mat.size();
-    ei_assert(size && "you are using an empty matrix");
-    const Index packetSize = ei_packet_traits<Scalar>::size;
-    const Index alignedStart = ei_first_aligned(mat);
+    
+    const Index packetSize = packet_traits<Scalar>::size;
+    const int packetAlignment = unpacket_traits<PacketScalar>::alignment;
     enum {
-      alignment = (Derived::Flags & DirectAccessBit) || (Derived::Flags & AlignedBit)
-                ? Aligned : Unaligned
+      alignment0 = (bool(Derived::Flags & DirectAccessBit) && bool(packet_traits<Scalar>::AlignedOnScalar)) ? int(packetAlignment) : int(Unaligned),
+      alignment = EIGEN_PLAIN_ENUM_MAX(alignment0, Derived::Alignment)
     };
-    const Index alignedSize = ((size-alignedStart)/packetSize)*packetSize;
-    const Index alignedEnd = alignedStart + alignedSize;
+    const Index alignedStart = internal::first_default_aligned(mat.nestedExpression());
+    const Index alignedSize2 = ((size-alignedStart)/(2*packetSize))*(2*packetSize);
+    const Index alignedSize = ((size-alignedStart)/(packetSize))*(packetSize);
+    const Index alignedEnd2 = alignedStart + alignedSize2;
+    const Index alignedEnd  = alignedStart + alignedSize;
     Scalar res;
     if(alignedSize)
     {
-      PacketScalar packet_res = mat.template packet<alignment>(alignedStart);
-      for(Index index = alignedStart + packetSize; index < alignedEnd; index += packetSize)
-        packet_res = func.packetOp(packet_res, mat.template packet<alignment>(index));
-      res = func.predux(packet_res);
+      PacketScalar packet_res0 = mat.template packet<alignment,PacketScalar>(alignedStart);
+      if(alignedSize>packetSize) // we have at least two packets to partly unroll the loop
+      {
+        PacketScalar packet_res1 = mat.template packet<alignment,PacketScalar>(alignedStart+packetSize);
+        for(Index index = alignedStart + 2*packetSize; index < alignedEnd2; index += 2*packetSize)
+        {
+          packet_res0 = func.packetOp(packet_res0, mat.template packet<alignment,PacketScalar>(index));
+          packet_res1 = func.packetOp(packet_res1, mat.template packet<alignment,PacketScalar>(index+packetSize));
+        }
+
+        packet_res0 = func.packetOp(packet_res0,packet_res1);
+        if(alignedEnd>alignedEnd2)
+          packet_res0 = func.packetOp(packet_res0, mat.template packet<alignment,PacketScalar>(alignedEnd2));
+      }
+      res = func.predux(packet_res0);
 
       for(Index index = 0; index < alignedStart; ++index)
         res = func(res,mat.coeff(index));
@@ -245,29 +263,29 @@ struct ei_redux_impl<Func, Derived, LinearVectorizedTraversal, NoUnrolling>
   }
 };
 
-template<typename Func, typename Derived>
-struct ei_redux_impl<Func, Derived, SliceVectorizedTraversal, NoUnrolling>
+// NOTE: for SliceVectorizedTraversal we simply bypass unrolling
+template<typename Func, typename Derived, int Unrolling>
+struct redux_impl<Func, Derived, SliceVectorizedTraversal, Unrolling>
 {
   typedef typename Derived::Scalar Scalar;
-  typedef typename ei_packet_traits<Scalar>::type PacketScalar;
-  typedef typename Derived::Index Index;
+  typedef typename packet_traits<Scalar>::type PacketType;
 
-  static Scalar run(const Derived& mat, const Func& func)
+  EIGEN_DEVICE_FUNC static Scalar run(const Derived &mat, const Func& func)
   {
-    ei_assert(mat.rows()>0 && mat.cols()>0 && "you are using an empty matrix");
+    eigen_assert(mat.rows()>0 && mat.cols()>0 && "you are using an empty matrix");
     const Index innerSize = mat.innerSize();
     const Index outerSize = mat.outerSize();
     enum {
-      packetSize = ei_packet_traits<Scalar>::size
+      packetSize = packet_traits<Scalar>::size
     };
     const Index packetedInnerSize = ((innerSize)/packetSize)*packetSize;
     Scalar res;
     if(packetedInnerSize)
     {
-      PacketScalar packet_res = mat.template packet<Unaligned>(0,0);
+      PacketType packet_res = mat.template packet<Unaligned,PacketType>(0,0);
       for(Index j=0; j<outerSize; ++j)
         for(Index i=(j==0?packetSize:0); i<packetedInnerSize; i+=Index(packetSize))
-          packet_res = func.packetOp(packet_res, mat.template packetByOuterInner<Unaligned>(j,i));
+          packet_res = func.packetOp(packet_res, mat.template packetByOuterInner<Unaligned,PacketType>(j,i));
 
       res = func.predux(packet_res);
       for(Index j=0; j<outerSize; ++j)
@@ -277,7 +295,7 @@ struct ei_redux_impl<Func, Derived, SliceVectorizedTraversal, NoUnrolling>
     else // too small to vectorize anything.
          // since this is dynamic-size hence inefficient anyway for such small sizes, don't try to optimize.
     {
-      res = ei_redux_impl<Func, Derived, DefaultTraversal, NoUnrolling>::run(mat, func);
+      res = redux_impl<Func, Derived, DefaultTraversal, NoUnrolling>::run(mat, func);
     }
 
     return res;
@@ -285,59 +303,137 @@ struct ei_redux_impl<Func, Derived, SliceVectorizedTraversal, NoUnrolling>
 };
 
 template<typename Func, typename Derived>
-struct ei_redux_impl<Func, Derived, LinearVectorizedTraversal, CompleteUnrolling>
+struct redux_impl<Func, Derived, LinearVectorizedTraversal, CompleteUnrolling>
 {
   typedef typename Derived::Scalar Scalar;
-  typedef typename ei_packet_traits<Scalar>::type PacketScalar;
+  typedef typename packet_traits<Scalar>::type PacketScalar;
   enum {
-    PacketSize = ei_packet_traits<Scalar>::size,
+    PacketSize = packet_traits<Scalar>::size,
     Size = Derived::SizeAtCompileTime,
     VectorizedSize = (Size / PacketSize) * PacketSize
   };
-  EIGEN_STRONG_INLINE static Scalar run(const Derived& mat, const Func& func)
+  EIGEN_DEVICE_FUNC static EIGEN_STRONG_INLINE Scalar run(const Derived &mat, const Func& func)
   {
-    ei_assert(mat.rows()>0 && mat.cols()>0 && "you are using an empty matrix");
-    Scalar res = func.predux(ei_redux_vec_unroller<Func, Derived, 0, Size / PacketSize>::run(mat,func));
-    if (VectorizedSize != Size)
-      res = func(res,ei_redux_novec_unroller<Func, Derived, VectorizedSize, Size-VectorizedSize>::run(mat,func));
-    return res;
+    eigen_assert(mat.rows()>0 && mat.cols()>0 && "you are using an empty matrix");
+    if (VectorizedSize > 0) {
+      Scalar res = func.predux(redux_vec_unroller<Func, Derived, 0, Size / PacketSize>::run(mat,func));
+      if (VectorizedSize != Size)
+        res = func(res,redux_novec_unroller<Func, Derived, VectorizedSize, Size-VectorizedSize>::run(mat,func));
+      return res;
+    }
+    else {
+      return redux_novec_unroller<Func, Derived, 0, Size>::run(mat,func);
+    }
   }
 };
 
+// evaluator adaptor
+template<typename _XprType>
+class redux_evaluator
+{
+public:
+  typedef _XprType XprType;
+  EIGEN_DEVICE_FUNC explicit redux_evaluator(const XprType &xpr) : m_evaluator(xpr), m_xpr(xpr) {}
+  
+  typedef typename XprType::Scalar Scalar;
+  typedef typename XprType::CoeffReturnType CoeffReturnType;
+  typedef typename XprType::PacketScalar PacketScalar;
+  typedef typename XprType::PacketReturnType PacketReturnType;
+  
+  enum {
+    MaxRowsAtCompileTime = XprType::MaxRowsAtCompileTime,
+    MaxColsAtCompileTime = XprType::MaxColsAtCompileTime,
+    // TODO we should not remove DirectAccessBit and rather find an elegant way to query the alignment offset at runtime from the evaluator
+    Flags = evaluator<XprType>::Flags & ~DirectAccessBit,
+    IsRowMajor = XprType::IsRowMajor,
+    SizeAtCompileTime = XprType::SizeAtCompileTime,
+    InnerSizeAtCompileTime = XprType::InnerSizeAtCompileTime,
+    CoeffReadCost = evaluator<XprType>::CoeffReadCost,
+    Alignment = evaluator<XprType>::Alignment
+  };
+  
+  EIGEN_DEVICE_FUNC Index rows() const { return m_xpr.rows(); }
+  EIGEN_DEVICE_FUNC Index cols() const { return m_xpr.cols(); }
+  EIGEN_DEVICE_FUNC Index size() const { return m_xpr.size(); }
+  EIGEN_DEVICE_FUNC Index innerSize() const { return m_xpr.innerSize(); }
+  EIGEN_DEVICE_FUNC Index outerSize() const { return m_xpr.outerSize(); }
+
+  EIGEN_DEVICE_FUNC
+  CoeffReturnType coeff(Index row, Index col) const
+  { return m_evaluator.coeff(row, col); }
+
+  EIGEN_DEVICE_FUNC
+  CoeffReturnType coeff(Index index) const
+  { return m_evaluator.coeff(index); }
+
+  template<int LoadMode, typename PacketType>
+  PacketReturnType packet(Index row, Index col) const
+  { return m_evaluator.template packet<LoadMode,PacketType>(row, col); }
+
+  template<int LoadMode, typename PacketType>
+  PacketReturnType packet(Index index) const
+  { return m_evaluator.template packet<LoadMode,PacketType>(index); }
+  
+  EIGEN_DEVICE_FUNC
+  CoeffReturnType coeffByOuterInner(Index outer, Index inner) const
+  { return m_evaluator.coeff(IsRowMajor ? outer : inner, IsRowMajor ? inner : outer); }
+  
+  template<int LoadMode, typename PacketType>
+  PacketReturnType packetByOuterInner(Index outer, Index inner) const
+  { return m_evaluator.template packet<LoadMode,PacketType>(IsRowMajor ? outer : inner, IsRowMajor ? inner : outer); }
+  
+  const XprType & nestedExpression() const { return m_xpr; }
+  
+protected:
+  internal::evaluator<XprType> m_evaluator;
+  const XprType &m_xpr;
+};
+
+} // end namespace internal
+
+/***************************************************************************
+* Part 4 : public API
+***************************************************************************/
+
 
 /** \returns the result of a full redux operation on the whole matrix or vector using \a func
   *
   * The template parameter \a BinaryOp is the type of the functor \a func which must be
-  * an associative operator. Both current STL and TR1 functor styles are handled.
+  * an associative operator. Both current C++98 and C++11 functor styles are handled.
   *
   * \sa DenseBase::sum(), DenseBase::minCoeff(), DenseBase::maxCoeff(), MatrixBase::colwise(), MatrixBase::rowwise()
   */
 template<typename Derived>
 template<typename Func>
-EIGEN_STRONG_INLINE typename ei_result_of<Func(typename ei_traits<Derived>::Scalar)>::type
+typename internal::traits<Derived>::Scalar
 DenseBase<Derived>::redux(const Func& func) const
 {
-  typedef typename ei_cleantype<typename Derived::Nested>::type ThisNested;
-  return ei_redux_impl<Func, ThisNested>
-            ::run(derived(), func);
+  eigen_assert(this->rows()>0 && this->cols()>0 && "you are using an empty matrix");
+
+  typedef typename internal::redux_evaluator<Derived> ThisEvaluator;
+  ThisEvaluator thisEval(derived());
+  
+  return internal::redux_impl<Func, ThisEvaluator>::run(thisEval, func);
 }
 
-/** \returns the minimum of all coefficients of *this
+/** \returns the minimum of all coefficients of \c *this.
+  * \warning the result is undefined if \c *this contains NaN.
   */
 template<typename Derived>
-EIGEN_STRONG_INLINE typename ei_traits<Derived>::Scalar
+EIGEN_STRONG_INLINE typename internal::traits<Derived>::Scalar
 DenseBase<Derived>::minCoeff() const
 {
-  return this->redux(Eigen::ei_scalar_min_op<Scalar>());
+  return derived().redux(Eigen::internal::scalar_min_op<Scalar>());
 }
 
-/** \returns the maximum of all coefficients of *this
+/** \returns the maximum of all coefficients of \c *this.
+  * \warning the result is undefined if \c *this contains NaN.
   */
 template<typename Derived>
-EIGEN_STRONG_INLINE typename ei_traits<Derived>::Scalar
+EIGEN_STRONG_INLINE typename internal::traits<Derived>::Scalar
 DenseBase<Derived>::maxCoeff() const
 {
-  return this->redux(Eigen::ei_scalar_max_op<Scalar>());
+  return derived().redux(Eigen::internal::scalar_max_op<Scalar>());
 }
 
 /** \returns the sum of all coefficients of *this
@@ -345,12 +441,12 @@ DenseBase<Derived>::maxCoeff() const
   * \sa trace(), prod(), mean()
   */
 template<typename Derived>
-EIGEN_STRONG_INLINE typename ei_traits<Derived>::Scalar
+EIGEN_STRONG_INLINE typename internal::traits<Derived>::Scalar
 DenseBase<Derived>::sum() const
 {
   if(SizeAtCompileTime==0 || (SizeAtCompileTime==Dynamic && size()==0))
     return Scalar(0);
-  return this->redux(Eigen::ei_scalar_sum_op<Scalar>());
+  return derived().redux(Eigen::internal::scalar_sum_op<Scalar>());
 }
 
 /** \returns the mean of all coefficients of *this
@@ -358,10 +454,10 @@ DenseBase<Derived>::sum() const
 * \sa trace(), prod(), sum()
 */
 template<typename Derived>
-EIGEN_STRONG_INLINE typename ei_traits<Derived>::Scalar
+EIGEN_STRONG_INLINE typename internal::traits<Derived>::Scalar
 DenseBase<Derived>::mean() const
 {
-  return Scalar(this->redux(Eigen::ei_scalar_sum_op<Scalar>())) / Scalar(this->size());
+  return Scalar(derived().redux(Eigen::internal::scalar_sum_op<Scalar>())) / Scalar(this->size());
 }
 
 /** \returns the product of all coefficients of *this
@@ -372,12 +468,12 @@ DenseBase<Derived>::mean() const
   * \sa sum(), mean(), trace()
   */
 template<typename Derived>
-EIGEN_STRONG_INLINE typename ei_traits<Derived>::Scalar
+EIGEN_STRONG_INLINE typename internal::traits<Derived>::Scalar
 DenseBase<Derived>::prod() const
 {
   if(SizeAtCompileTime==0 || (SizeAtCompileTime==Dynamic && size()==0))
     return Scalar(1);
-  return this->redux(Eigen::ei_scalar_product_op<Scalar>());
+  return derived().redux(Eigen::internal::scalar_product_op<Scalar>());
 }
 
 /** \returns the trace of \c *this, i.e. the sum of the coefficients on the main diagonal.
@@ -387,10 +483,12 @@ DenseBase<Derived>::prod() const
   * \sa diagonal(), sum()
   */
 template<typename Derived>
-EIGEN_STRONG_INLINE typename ei_traits<Derived>::Scalar
+EIGEN_STRONG_INLINE typename internal::traits<Derived>::Scalar
 MatrixBase<Derived>::trace() const
 {
   return derived().diagonal().sum();
 }
 
+} // end namespace Eigen
+
 #endif // EIGEN_REDUX_H

Eigen/src/Core/Ref.h

@@ -0,0 +1,276 @@
+// This file is part of Eigen, a lightweight C++ template library
+// for linear algebra.
+//
+// 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_REF_H
+#define EIGEN_REF_H
+
+namespace Eigen { 
+
+/** \class Ref
+  * \ingroup Core_Module
+  *
+  * \brief A matrix or vector expression mapping an existing expression
+  *
+  * \tparam PlainObjectType the equivalent matrix type of the mapped data
+  * \tparam MapOptions specifies the pointer alignment in bytes. It can be: \c #Aligned128, , \c #Aligned64, \c #Aligned32, \c #Aligned16, \c #Aligned8 or \c #Unaligned.
+  *                The default is \c #Unaligned.
+  * \tparam StrideType optionally specifies strides. By default, Ref implies a contiguous storage along the inner dimension (inner stride==1),
+  *                   but accepts a variable outer stride (leading dimension).
+  *                   This can be overridden by specifying strides.
+  *                   The type passed here must be a specialization of the Stride template, see examples below.
+  *
+  * This class provides a way to write non-template functions taking Eigen objects as parameters while limiting the number of copies.
+  * A Ref<> object can represent either a const expression or a l-value:
+  * \code
+  * // in-out argument:
+  * void foo1(Ref<VectorXf> x);
+  *
+  * // read-only const argument:
+  * void foo2(const Ref<const VectorXf>& x);
+  * \endcode
+  *
+  * In the in-out case, the input argument must satisfy the constraints of the actual Ref<> type, otherwise a compilation issue will be triggered.
+  * By default, a Ref<VectorXf> can reference any dense vector expression of float having a contiguous memory layout.
+  * Likewise, a Ref<MatrixXf> can reference any column-major dense matrix expression of float whose column's elements are contiguously stored with
+  * the possibility to have a constant space in-between each column, i.e. the inner stride must be equal to 1, but the outer stride (or leading dimension)
+  * can be greater than the number of rows.
+  *
+  * In the const case, if the input expression does not match the above requirement, then it is evaluated into a temporary before being passed to the function.
+  * Here are some examples:
+  * \code
+  * MatrixXf A;
+  * VectorXf a;
+  * foo1(a.head());             // OK
+  * foo1(A.col());              // OK
+  * foo1(A.row());              // Compilation error because here innerstride!=1
+  * foo2(A.row());              // Compilation error because A.row() is a 1xN object while foo2 is expecting a Nx1 object
+  * foo2(A.row().transpose());  // The row is copied into a contiguous temporary
+  * foo2(2*a);                  // The expression is evaluated into a temporary
+  * foo2(A.col().segment(2,4)); // No temporary
+  * \endcode
+  *
+  * The range of inputs that can be referenced without temporary can be enlarged using the last two template parameters.
+  * Here is an example accepting an innerstride!=1:
+  * \code
+  * // in-out argument:
+  * void foo3(Ref<VectorXf,0,InnerStride<> > x);
+  * foo3(A.row());              // OK
+  * \endcode
+  * The downside here is that the function foo3 might be significantly slower than foo1 because it won't be able to exploit vectorization, and will involve more
+  * expensive address computations even if the input is contiguously stored in memory. To overcome this issue, one might propose to overload internally calling a
+  * template function, e.g.:
+  * \code
+  * // in the .h:
+  * void foo(const Ref<MatrixXf>& A);
+  * void foo(const Ref<MatrixXf,0,Stride<> >& A);
+  *
+  * // in the .cpp:
+  * template<typename TypeOfA> void foo_impl(const TypeOfA& A) {
+  *     ... // crazy code goes here
+  * }
+  * void foo(const Ref<MatrixXf>& A) { foo_impl(A); }
+  * void foo(const Ref<MatrixXf,0,Stride<> >& A) { foo_impl(A); }
+  * \endcode
+  *
+  *
+  * \sa PlainObjectBase::Map(), \ref TopicStorageOrders
+  */
+
+namespace internal {
+
+template<typename _PlainObjectType, int _Options, typename _StrideType>
+struct traits<Ref<_PlainObjectType, _Options, _StrideType> >
+  : public traits<Map<_PlainObjectType, _Options, _StrideType> >
+{
+  typedef _PlainObjectType PlainObjectType;
+  typedef _StrideType StrideType;
+  enum {
+    Options = _Options,
+    Flags = traits<Map<_PlainObjectType, _Options, _StrideType> >::Flags | NestByRefBit,
+    Alignment = traits<Map<_PlainObjectType, _Options, _StrideType> >::Alignment
+  };
+
+  template<typename Derived> struct match {
+    enum {
+      HasDirectAccess = internal::has_direct_access<Derived>::ret,
+      StorageOrderMatch = PlainObjectType::IsVectorAtCompileTime || Derived::IsVectorAtCompileTime || ((PlainObjectType::Flags&RowMajorBit)==(Derived::Flags&RowMajorBit)),
+      InnerStrideMatch = int(StrideType::InnerStrideAtCompileTime)==int(Dynamic)
+                      || int(StrideType::InnerStrideAtCompileTime)==int(Derived::InnerStrideAtCompileTime)
+                      || (int(StrideType::InnerStrideAtCompileTime)==0 && int(Derived::InnerStrideAtCompileTime)==1),
+      OuterStrideMatch = Derived::IsVectorAtCompileTime
+                      || int(StrideType::OuterStrideAtCompileTime)==int(Dynamic) || int(StrideType::OuterStrideAtCompileTime)==int(Derived::OuterStrideAtCompileTime),
+      AlignmentMatch = (int(traits<PlainObjectType>::Alignment)==int(Unaligned)) || (int(evaluator<Derived>::Alignment) >= int(Alignment)), // FIXME the first condition is not very clear, it should be replaced by the required alignment
+      ScalarTypeMatch = internal::is_same<typename PlainObjectType::Scalar, typename Derived::Scalar>::value,
+      MatchAtCompileTime = HasDirectAccess && StorageOrderMatch && InnerStrideMatch && OuterStrideMatch && AlignmentMatch && ScalarTypeMatch
+    };
+    typedef typename internal::conditional<MatchAtCompileTime,internal::true_type,internal::false_type>::type type;
+  };
+  
+};
+
+template<typename Derived>
+struct traits<RefBase<Derived> > : public traits<Derived> {};
+
+}
+
+template<typename Derived> class RefBase
+ : public MapBase<Derived>
+{
+  typedef typename internal::traits<Derived>::PlainObjectType PlainObjectType;
+  typedef typename internal::traits<Derived>::StrideType StrideType;
+
+public:
+
+  typedef MapBase<Derived> Base;
+  EIGEN_DENSE_PUBLIC_INTERFACE(RefBase)
+
+  EIGEN_DEVICE_FUNC inline Index innerStride() const
+  {
+    return StrideType::InnerStrideAtCompileTime != 0 ? m_stride.inner() : 1;
+  }
+
+  EIGEN_DEVICE_FUNC inline Index outerStride() const
+  {
+    return StrideType::OuterStrideAtCompileTime != 0 ? m_stride.outer()
+         : IsVectorAtCompileTime ? this->size()
+         : int(Flags)&RowMajorBit ? this->cols()
+         : this->rows();
+  }
+
+  EIGEN_DEVICE_FUNC RefBase()
+    : Base(0,RowsAtCompileTime==Dynamic?0:RowsAtCompileTime,ColsAtCompileTime==Dynamic?0:ColsAtCompileTime),
+      // Stride<> does not allow default ctor for Dynamic strides, so let' initialize it with dummy values:
+      m_stride(StrideType::OuterStrideAtCompileTime==Dynamic?0:StrideType::OuterStrideAtCompileTime,
+               StrideType::InnerStrideAtCompileTime==Dynamic?0:StrideType::InnerStrideAtCompileTime)
+  {}
+  
+  EIGEN_INHERIT_ASSIGNMENT_OPERATORS(RefBase)
+
+protected:
+
+  typedef Stride<StrideType::OuterStrideAtCompileTime,StrideType::InnerStrideAtCompileTime> StrideBase;
+
+  template<typename Expression>
+  EIGEN_DEVICE_FUNC void construct(Expression& expr)
+  {
+    if(PlainObjectType::RowsAtCompileTime==1)
+    {
+      eigen_assert(expr.rows()==1 || expr.cols()==1);
+      ::new (static_cast<Base*>(this)) Base(expr.data(), 1, expr.size());
+    }
+    else if(PlainObjectType::ColsAtCompileTime==1)
+    {
+      eigen_assert(expr.rows()==1 || expr.cols()==1);
+      ::new (static_cast<Base*>(this)) Base(expr.data(), expr.size(), 1);
+    }
+    else
+      ::new (static_cast<Base*>(this)) Base(expr.data(), expr.rows(), expr.cols());
+    
+    if(Expression::IsVectorAtCompileTime && (!PlainObjectType::IsVectorAtCompileTime) && ((Expression::Flags&RowMajorBit)!=(PlainObjectType::Flags&RowMajorBit)))
+      ::new (&m_stride) StrideBase(expr.innerStride(), StrideType::InnerStrideAtCompileTime==0?0:1);
+    else
+      ::new (&m_stride) StrideBase(StrideType::OuterStrideAtCompileTime==0?0:expr.outerStride(),
+                                   StrideType::InnerStrideAtCompileTime==0?0:expr.innerStride());    
+  }
+
+  StrideBase m_stride;
+};
+
+
+template<typename PlainObjectType, int Options, typename StrideType> class Ref
+  : public RefBase<Ref<PlainObjectType, Options, StrideType> >
+{
+  private:
+    typedef internal::traits<Ref> Traits;
+    template<typename Derived>
+    EIGEN_DEVICE_FUNC inline Ref(const PlainObjectBase<Derived>& expr,
+                                 typename internal::enable_if<bool(Traits::template match<Derived>::MatchAtCompileTime),Derived>::type* = 0);
+  public:
+
+    typedef RefBase<Ref> Base;
+    EIGEN_DENSE_PUBLIC_INTERFACE(Ref)
+
+
+    #ifndef EIGEN_PARSED_BY_DOXYGEN
+    template<typename Derived>
+    EIGEN_DEVICE_FUNC inline Ref(PlainObjectBase<Derived>& expr,
+                                 typename internal::enable_if<bool(Traits::template match<Derived>::MatchAtCompileTime),Derived>::type* = 0)
+    {
+      EIGEN_STATIC_ASSERT(bool(Traits::template match<Derived>::MatchAtCompileTime), STORAGE_LAYOUT_DOES_NOT_MATCH);
+      Base::construct(expr.derived());
+    }
+    template<typename Derived>
+    EIGEN_DEVICE_FUNC inline Ref(const DenseBase<Derived>& expr,
+                                 typename internal::enable_if<bool(Traits::template match<Derived>::MatchAtCompileTime),Derived>::type* = 0)
+    #else
+    template<typename Derived>
+    inline Ref(DenseBase<Derived>& expr)
+    #endif
+    {
+      EIGEN_STATIC_ASSERT(bool(internal::is_lvalue<Derived>::value), THIS_EXPRESSION_IS_NOT_A_LVALUE__IT_IS_READ_ONLY);
+      EIGEN_STATIC_ASSERT(bool(Traits::template match<Derived>::MatchAtCompileTime), STORAGE_LAYOUT_DOES_NOT_MATCH);
+      EIGEN_STATIC_ASSERT(!Derived::IsPlainObjectBase,THIS_EXPRESSION_IS_NOT_A_LVALUE__IT_IS_READ_ONLY);
+      Base::construct(expr.const_cast_derived());
+    }
+
+    EIGEN_INHERIT_ASSIGNMENT_OPERATORS(Ref)
+
+};
+
+// this is the const ref version
+template<typename TPlainObjectType, int Options, typename StrideType> class Ref<const TPlainObjectType, Options, StrideType>
+  : public RefBase<Ref<const TPlainObjectType, Options, StrideType> >
+{
+    typedef internal::traits<Ref> Traits;
+  public:
+
+    typedef RefBase<Ref> Base;
+    EIGEN_DENSE_PUBLIC_INTERFACE(Ref)
+
+    template<typename Derived>
+    EIGEN_DEVICE_FUNC inline Ref(const DenseBase<Derived>& expr,
+                                 typename internal::enable_if<bool(Traits::template match<Derived>::ScalarTypeMatch),Derived>::type* = 0)
+    {
+//      std::cout << match_helper<Derived>::HasDirectAccess << "," << match_helper<Derived>::OuterStrideMatch << "," << match_helper<Derived>::InnerStrideMatch << "\n";
+//      std::cout << int(StrideType::OuterStrideAtCompileTime) << " - " << int(Derived::OuterStrideAtCompileTime) << "\n";
+//      std::cout << int(StrideType::InnerStrideAtCompileTime) << " - " << int(Derived::InnerStrideAtCompileTime) << "\n";
+      construct(expr.derived(), typename Traits::template match<Derived>::type());
+    }
+
+    EIGEN_DEVICE_FUNC inline Ref(const Ref& other) : Base(other) {
+      // copy constructor shall not copy the m_object, to avoid unnecessary malloc and copy
+    }
+
+    template<typename OtherRef>
+    EIGEN_DEVICE_FUNC inline Ref(const RefBase<OtherRef>& other) {
+      construct(other.derived(), typename Traits::template match<OtherRef>::type());
+    }
+
+  protected:
+
+    template<typename Expression>
+    EIGEN_DEVICE_FUNC void construct(const Expression& expr,internal::true_type)
+    {
+      Base::construct(expr);
+    }
+
+    template<typename Expression>
+    EIGEN_DEVICE_FUNC void construct(const Expression& expr, internal::false_type)
+    {
+      internal::call_assignment_no_alias(m_object,expr,internal::assign_op<Scalar>());
+      Base::construct(m_object);
+    }
+
+  protected:
+    TPlainObjectType m_object;
+};
+
+} // end namespace Eigen
+
+#endif // EIGEN_REF_H

Eigen/src/Core/Replicate.h

@@ -3,28 +3,15 @@
 //
 // Copyright (C) 2009-2010 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_REPLICATE_H
 #define EIGEN_REPLICATE_H
 
+namespace Eigen { 
+
 /**
   * \class Replicate
   * \ingroup Core_Module
@@ -39,15 +26,17 @@
   *
   * \sa DenseBase::replicate()
   */
+
+namespace internal {
 template<typename MatrixType,int RowFactor,int ColFactor>
-struct ei_traits<Replicate<MatrixType,RowFactor,ColFactor> >
- : ei_traits<MatrixType>
+struct traits<Replicate<MatrixType,RowFactor,ColFactor> >
+ : traits<MatrixType>
 {
   typedef typename MatrixType::Scalar Scalar;
-  typedef typename ei_traits<MatrixType>::StorageKind StorageKind;
-  typedef typename ei_traits<MatrixType>::XprKind XprKind;
-  typedef typename ei_nested<MatrixType>::type MatrixTypeNested;
-  typedef typename ei_unref<MatrixTypeNested>::type _MatrixTypeNested;
+  typedef typename traits<MatrixType>::StorageKind StorageKind;
+  typedef typename traits<MatrixType>::XprKind XprKind;
+  typedef typename ref_selector<MatrixType>::type MatrixTypeNested;
+  typedef typename remove_reference<MatrixTypeNested>::type _MatrixTypeNested;
   enum {
     RowsAtCompileTime = RowFactor==Dynamic || int(MatrixType::RowsAtCompileTime)==Dynamic
                       ? Dynamic
@@ -61,69 +50,58 @@ struct ei_traits<Replicate<MatrixType,RowFactor,ColFactor> >
     IsRowMajor = MaxRowsAtCompileTime==1 && MaxColsAtCompileTime!=1 ? 1
                : MaxColsAtCompileTime==1 && MaxRowsAtCompileTime!=1 ? 0
                : (MatrixType::Flags & RowMajorBit) ? 1 : 0,
-    Flags = (_MatrixTypeNested::Flags & HereditaryBits & ~RowMajorBit) | (IsRowMajor ? RowMajorBit : 0),
-    CoeffReadCost = _MatrixTypeNested::CoeffReadCost
+    
+    // FIXME enable DirectAccess with negative strides?
+    Flags = IsRowMajor ? RowMajorBit : 0
   };
 };
+}
 
 template<typename MatrixType,int RowFactor,int ColFactor> class Replicate
-  : public ei_dense_xpr_base< Replicate<MatrixType,RowFactor,ColFactor> >::type
+  : public internal::dense_xpr_base< Replicate<MatrixType,RowFactor,ColFactor> >::type
 {
+    typedef typename internal::traits<Replicate>::MatrixTypeNested MatrixTypeNested;
+    typedef typename internal::traits<Replicate>::_MatrixTypeNested _MatrixTypeNested;
   public:
 
-    typedef typename ei_dense_xpr_base<Replicate>::type Base;
+    typedef typename internal::dense_xpr_base<Replicate>::type Base;
     EIGEN_DENSE_PUBLIC_INTERFACE(Replicate)
+    typedef typename internal::remove_all<MatrixType>::type NestedExpression;
 
     template<typename OriginalMatrixType>
+    EIGEN_DEVICE_FUNC
     inline explicit Replicate(const OriginalMatrixType& matrix)
       : m_matrix(matrix), m_rowFactor(RowFactor), m_colFactor(ColFactor)
     {
-      EIGEN_STATIC_ASSERT((ei_is_same_type<MatrixType,OriginalMatrixType>::ret),
+      EIGEN_STATIC_ASSERT((internal::is_same<typename internal::remove_const<MatrixType>::type,OriginalMatrixType>::value),
                           THE_MATRIX_OR_EXPRESSION_THAT_YOU_PASSED_DOES_NOT_HAVE_THE_EXPECTED_TYPE)
-      ei_assert(RowFactor!=Dynamic && ColFactor!=Dynamic);
+      eigen_assert(RowFactor!=Dynamic && ColFactor!=Dynamic);
     }
 
     template<typename OriginalMatrixType>
-    inline Replicate(const OriginalMatrixType& matrix, int rowFactor, int colFactor)
+    EIGEN_DEVICE_FUNC
+    inline Replicate(const OriginalMatrixType& matrix, Index rowFactor, Index colFactor)
       : m_matrix(matrix), m_rowFactor(rowFactor), m_colFactor(colFactor)
     {
-      EIGEN_STATIC_ASSERT((ei_is_same_type<MatrixType,OriginalMatrixType>::ret),
+      EIGEN_STATIC_ASSERT((internal::is_same<typename internal::remove_const<MatrixType>::type,OriginalMatrixType>::value),
                           THE_MATRIX_OR_EXPRESSION_THAT_YOU_PASSED_DOES_NOT_HAVE_THE_EXPECTED_TYPE)
     }
 
+    EIGEN_DEVICE_FUNC
     inline Index rows() const { return m_matrix.rows() * m_rowFactor.value(); }
+    EIGEN_DEVICE_FUNC
     inline Index cols() const { return m_matrix.cols() * m_colFactor.value(); }
 
-    inline Scalar coeff(Index row, Index col) const
-    {
-      // try to avoid using modulo; this is a pure optimization strategy
-      const Index actual_row  = ei_traits<MatrixType>::RowsAtCompileTime==1 ? 0
-                            : RowFactor==1 ? row
-                            : row%m_matrix.rows();
-      const Index actual_col  = ei_traits<MatrixType>::ColsAtCompileTime==1 ? 0
-                            : ColFactor==1 ? col
-                            : col%m_matrix.cols();
-
-      return m_matrix.coeff(actual_row, actual_col);
+    EIGEN_DEVICE_FUNC
+    const _MatrixTypeNested& nestedExpression() const
+    { 
+      return m_matrix; 
     }
-    template<int LoadMode>
-    inline PacketScalar packet(Index row, Index col) const
-    {
-      const Index actual_row  = ei_traits<MatrixType>::RowsAtCompileTime==1 ? 0
-                            : RowFactor==1 ? row
-                            : row%m_matrix.rows();
-      const Index actual_col  = ei_traits<MatrixType>::ColsAtCompileTime==1 ? 0
-                            : ColFactor==1 ? col
-                            : col%m_matrix.cols();
-
-      return m_matrix.template packet<LoadMode>(actual_row, actual_col);
-    }
-
 
   protected:
-    const typename MatrixType::Nested m_matrix;
-    const ei_variable_if_dynamic<Index, RowFactor> m_rowFactor;
-    const ei_variable_if_dynamic<Index, ColFactor> m_colFactor;
+    MatrixTypeNested m_matrix;
+    const internal::variable_if_dynamic<Index, RowFactor> m_rowFactor;
+    const internal::variable_if_dynamic<Index, ColFactor> m_colFactor;
 };
 
 /**
@@ -136,27 +114,12 @@ template<typename MatrixType,int RowFactor,int ColFactor> class Replicate
   */
 template<typename Derived>
 template<int RowFactor, int ColFactor>
-inline const Replicate<Derived,RowFactor,ColFactor>
+const Replicate<Derived,RowFactor,ColFactor>
 DenseBase<Derived>::replicate() const
 {
   return Replicate<Derived,RowFactor,ColFactor>(derived());
 }
 
-/**
-  * \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
-  */
-template<typename Derived>
-inline const Replicate<Derived,Dynamic,Dynamic>
-DenseBase<Derived>::replicate(Index rowFactor,Index colFactor) const
-{
-  return Replicate<Derived,Dynamic,Dynamic>(derived(),rowFactor,colFactor);
-}
-
 /**
   * \return an expression of the replication of each column (or row) of \c *this
   *
@@ -173,4 +136,6 @@ VectorwiseOp<ExpressionType,Direction>::replicate(Index factor) const
           (_expression(),Direction==Vertical?factor:1,Direction==Horizontal?factor:1);
 }
 
+} // end namespace Eigen
+
 #endif // EIGEN_REPLICATE_H