bump to 3.2

(transplanted from 660b905e12
)

.gitlab-ci.yml

@@ -1,23 +0,0 @@
-# This file is part of Eigen, a lightweight C++ template library
-# for linear algebra.
-#
-# Copyright (C) 2020 Arm Ltd. and Contributors
-#
-# 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/.
-
-stages:
-  - buildsmoketests
-  - smoketests
-  - build
-  - test
-
-variables:
-  BUILDDIR: builddir
-  EIGEN_CI_CMAKE_GENEATOR: "Ninja"
-
-include:
-  - "/ci/smoketests.gitlab-ci.yml"
-  - "/ci/build.gitlab-ci.yml"
-  - "/ci/test.gitlab-ci.yml"

.gitlab/issue_templates/Bug Report.md

@@ -1,69 +0,0 @@
-<!--
-Please read this!
-
-Before opening a new issue, make sure to search for keywords in the issues
-filtered by "bug::confirmed" or "bug::unconfirmed" and "bugzilla" label:
-
-- https://gitlab.com/libeigen/eigen/-/issues?scope=all&utf8=%E2%9C%93&state=opened&label_name[]=bug%3A%3Aconfirmed
-- https://gitlab.com/libeigen/eigen/-/issues?scope=all&utf8=%E2%9C%93&state=opened&label_name[]=bug%3A%3Aunconfirmed
-- https://gitlab.com/libeigen/eigen/-/issues?scope=all&utf8=%E2%9C%93&state=opened&label_name[]=bugzilla
-
-and verify the issue you're about to submit isn't a duplicate. -->
-
-### Summary
-<!-- Summarize the bug encountered concisely. -->
-
-### Environment
-<!-- Please provide your development environment here -->
-- **Operating System** : Windows/Linux
-- **Architecture** : x64/Arm64/PowerPC ...
-- **Eigen Version** : 3.3.9
-- **Compiler Version** : Gcc7.0
-- **Compile Flags** : -O3 -march=native
-- **Vector Extension** : SSE/AVX/NEON ...
-
-### Minimal Example
-<!-- If possible, please create a minimal example here that exhibits the problematic behavior.
-You can also link to [godbolt](https://godbolt.org). But please note that you need to click 
-the "Share" button in the top right-hand corner of the godbolt page where you reproduce the sample 
-code to get the share link instead of in your browser address bar. 
-
-You can read [the guidelines on stackoverflow](https://stackoverflow.com/help/minimal-reproducible-example)
-on how to create a good minimal example. -->
-
-```cpp
-//show your code here
-```
-
-### Steps to reproduce
-<!-- Describe how one can reproduce the issue - this is very important. Please use an ordered list. -->
-
-1. first step
-2. second step
-3. ... 
-
-### What is the current *bug* behavior?
-<!-- Describe what actually happens. -->
-
-### What is the expected *correct* behavior?
-<!-- Describe what you should see instead. -->
-
-### Relevant logs
-<!-- Add relevant code snippets or program output within blocks marked by " ``` " -->
-
-<!-- OPTIONAL: remove this section if you are not reporting a compilation warning issue.-->
-### Warning Messages
-<!-- Show us the warning messages you got! -->
-
-<!-- OPTIONAL: remove this section if you are not reporting a performance issue. -->
-### Benchmark scripts and results
-<!-- Please share any benchmark scripts - either standalone, or using [Google Benchmark](https://github.com/google/benchmark). -->
-
-### Anything else that might help
-<!-- It will be better to provide us more information to help narrow down the cause. 
-Including but not limited to the following: 
-- lines of code that might help us diagnose the problem. 
-- potential ways to address the issue.
-- last known working/first broken version (release number or commit hash). --> 
-
-- [ ] Have a plan to fix this issue.

.gitlab/issue_templates/Feature Request.md

@@ -1,7 +0,0 @@
-### Describe the feature you would like to be implemented.
-
-### Would such a feature be useful for other users? Why?
-
-### Any hints on how to implement the requested feature?
-
-### Additional resources

.gitlab/merge_request_templates/Merge Request Template.md

@@ -1,26 +0,0 @@
-<!-- 
-Thanks for contributing a merge request! Please name and fully describe your MR as you would for a commit message.
-If the MR fixes an issue, please include "Fixes #issue" in the commit message and the MR description.
-
-In addition, we recommend that first-time contributors read our [contribution guidelines](https://eigen.tuxfamily.org/index.php?title=Contributing_to_Eigen) and [git page](https://eigen.tuxfamily.org/index.php?title=Git), which will help you submit a more standardized MR.
-
-Before submitting the MR, you also need to complete the following checks:
-- Make one PR per feature/bugfix (don't mix multiple changes into one PR). Avoid committing unrelated changes.
-- Rebase before committing
-- For code changes, run the test suite (at least the tests that are likely affected by the change).
-  See our [test guidelines](https://eigen.tuxfamily.org/index.php?title=Tests).
-- If possible, add a test (both for bug-fixes as well as new features)
-- Make sure new features are documented
-
-Note that we are a team of volunteers; we appreciate your patience during the review process.
-
-Again, thanks for contributing! -->
-
-### Reference issue
-<!-- You can link to a specific issue using the gitlab syntax #<issue number>  -->
-
-### What does this implement/fix?
-<!--Please explain your changes.-->
-
-### Additional information
-<!--Any additional information you think is important.-->

.hgeol

@@ -1,9 +1,6 @@
 [patterns]
-*.sh = LF
-*.MINPACK = CRLF
 scripts/*.in = LF
 debug/msvc/*.dat = CRLF
-debug/msvc/*.natvis = CRLF
 unsupported/test/mpreal/*.* = CRLF
 ** = native
 

.hgignore

@@ -1,3 +1,4 @@
+syntax: glob
 qrc_*cxx
 *.orig
 *.pyc
@@ -12,7 +13,7 @@ core
 core.*
 *.bak
 *~
-*build*
+build*
 *.moc.*
 *.moc
 ui_*
@@ -27,12 +28,5 @@ activity.png
 *.rej
 log
 patch
-*.patch
 a
 a.*
-lapack/testing
-lapack/reference
-.*project
-.settings
-Makefile
-!ci/build.gitlab-ci.yml

CMakeLists.txt

@@ -1,7 +1,6 @@
-# cmake_minimum_require must be the first command of the file
-cmake_minimum_required(VERSION 3.5.0)
+project(Eigen)
 
-project(Eigen3)
+cmake_minimum_required(VERSION 2.8.2)
 
 # guard against in-source builds
 
@@ -9,21 +8,22 @@ if(${CMAKE_SOURCE_DIR} STREQUAL ${CMAKE_BINARY_DIR})
   message(FATAL_ERROR "In-source builds not allowed. Please make a new directory (called a build directory) and run CMake from there. You may need to remove CMakeCache.txt. ")
 endif()
 
-
-# Alias Eigen_*_DIR to Eigen3_*_DIR:
-
-set(Eigen_SOURCE_DIR ${Eigen3_SOURCE_DIR})
-set(Eigen_BINARY_DIR ${Eigen3_BINARY_DIR})
-
 # guard against bad build-type strings
 
 if (NOT CMAKE_BUILD_TYPE)
   set(CMAKE_BUILD_TYPE "Release")
 endif()
 
+string(TOLOWER "${CMAKE_BUILD_TYPE}" cmake_build_type_tolower)
+if(    NOT cmake_build_type_tolower STREQUAL "debug"
+   AND NOT cmake_build_type_tolower STREQUAL "release"
+   AND NOT cmake_build_type_tolower STREQUAL "relwithdebinfo")
+  message(FATAL_ERROR "Unknown build type \"${CMAKE_BUILD_TYPE}\". Allowed values are Debug, Release, RelWithDebInfo (case-insensitive).")
+endif()
+
 
 #############################################################################
-# retrieve version information                                               #
+# retrieve version infomation                                               #
 #############################################################################
 
 # automatically parse the version number
@@ -36,58 +36,28 @@ string(REGEX MATCH "define[ \t]+EIGEN_MINOR_VERSION[ \t]+([0-9]+)" _eigen_minor_
 set(EIGEN_MINOR_VERSION "${CMAKE_MATCH_1}")
 set(EIGEN_VERSION_NUMBER ${EIGEN_WORLD_VERSION}.${EIGEN_MAJOR_VERSION}.${EIGEN_MINOR_VERSION})
 
-# if we are not in a git clone
-if(IS_DIRECTORY ${CMAKE_SOURCE_DIR}/.git)
-  # if the git program is absent or this will leave the EIGEN_GIT_REVNUM string empty,
-  # but won't stop CMake.
-  execute_process(COMMAND git ls-remote --refs -q ${CMAKE_SOURCE_DIR} HEAD OUTPUT_VARIABLE EIGEN_GIT_OUTPUT)
-endif()
+# if the mercurial program is absent, this will leave the EIGEN_HG_CHANGESET string empty,
+# but won't stop CMake.
+execute_process(COMMAND hg tip -R ${CMAKE_SOURCE_DIR} OUTPUT_VARIABLE EIGEN_HGTIP_OUTPUT)
+execute_process(COMMAND hg branch -R ${CMAKE_SOURCE_DIR} OUTPUT_VARIABLE EIGEN_BRANCH_OUTPUT)
 
-# extract the git rev number from the git output...
-if(EIGEN_GIT_OUTPUT)
-string(REGEX MATCH "^([0-9;a-f]+).*" EIGEN_GIT_CHANGESET_MATCH "${EIGEN_GIT_OUTPUT}")
-set(EIGEN_GIT_REVNUM "${CMAKE_MATCH_1}")
-endif()
+# if this is the default (aka development) branch, extract the mercurial changeset number from the hg tip output...
+if(EIGEN_BRANCH_OUTPUT MATCHES "default")
+string(REGEX MATCH "^changeset: *[0-9]*:([0-9;a-f]+).*" EIGEN_HG_CHANGESET_MATCH "${EIGEN_HGTIP_OUTPUT}")
+set(EIGEN_HG_CHANGESET "${CMAKE_MATCH_1}")
+endif(EIGEN_BRANCH_OUTPUT MATCHES "default")
 #...and show it next to the version number
-if(EIGEN_GIT_REVNUM)
-  set(EIGEN_VERSION "${EIGEN_VERSION_NUMBER} (git rev ${EIGEN_GIT_REVNUM})")
-else()
+if(EIGEN_HG_CHANGESET)
+  set(EIGEN_VERSION "${EIGEN_VERSION_NUMBER} (mercurial changeset ${EIGEN_HG_CHANGESET})")
+else(EIGEN_HG_CHANGESET)
   set(EIGEN_VERSION "${EIGEN_VERSION_NUMBER}")
-endif()
+endif(EIGEN_HG_CHANGESET)
+
 
 include(CheckCXXCompilerFlag)
-include(GNUInstallDirs)
-include(CMakeDependentOption)
 
 set(CMAKE_MODULE_PATH ${PROJECT_SOURCE_DIR}/cmake)
 
-
-option(EIGEN_TEST_CXX11 "Enable testing with C++11 and C++11 features (e.g. Tensor module)." OFF)
-
-
-macro(ei_add_cxx_compiler_flag FLAG)
-  string(REGEX REPLACE "-" "" SFLAG1 ${FLAG})
-  string(REGEX REPLACE "\\+" "p" SFLAG ${SFLAG1})
-  check_cxx_compiler_flag(${FLAG} COMPILER_SUPPORT_${SFLAG})
-  if(COMPILER_SUPPORT_${SFLAG})
-    set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} ${FLAG}")
-  endif()
-endmacro()
-
-check_cxx_compiler_flag("-std=c++11" EIGEN_COMPILER_SUPPORT_CPP11)
-
-if(EIGEN_TEST_CXX11)
-  set(CMAKE_CXX_STANDARD 11)
-  set(CMAKE_CXX_EXTENSIONS OFF)
-  if(EIGEN_COMPILER_SUPPORT_CPP11)
-    set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -std=c++11")
-  endif()
-else()
-  #set(CMAKE_CXX_STANDARD 03)
-  #set(CMAKE_CXX_EXTENSIONS OFF)
-  ei_add_cxx_compiler_flag("-std=c++03")
-endif()
-
 #############################################################################
 # find how to link to the standard libraries                                #
 #############################################################################
@@ -122,13 +92,11 @@ else()
 endif()
 
 option(EIGEN_BUILD_BTL "Build benchmark suite" OFF)
-
-# Disable pkgconfig only for native Windows builds
-if(NOT WIN32 OR NOT CMAKE_HOST_SYSTEM_NAME MATCHES Windows)
+if(NOT WIN32)
   option(EIGEN_BUILD_PKGCONFIG "Build pkg-config .pc file for Eigen" ON)
-endif()
+endif(NOT WIN32)
 
-set(CMAKE_INCLUDE_CURRENT_DIR OFF)
+set(CMAKE_INCLUDE_CURRENT_DIR ON)
 
 option(EIGEN_SPLIT_LARGE_TESTS "Split large tests into smaller executables" ON)
 
@@ -139,20 +107,33 @@ endif()
 
 set(EIGEN_TEST_MAX_SIZE "320" CACHE STRING "Maximal matrix/vector size, default is 320")
 
+macro(ei_add_cxx_compiler_flag FLAG)
+  string(REGEX REPLACE "-" "" SFLAG ${FLAG})
+  check_cxx_compiler_flag(${FLAG} COMPILER_SUPPORT_${SFLAG})
+  if(COMPILER_SUPPORT_${SFLAG})
+    set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} ${FLAG}")
+  endif()
+endmacro(ei_add_cxx_compiler_flag)
+
 if(NOT MSVC)
   # We assume that other compilers are partly compatible with GNUCC
-
-  # clang outputs some warnings for unknown flags that are not caught by check_cxx_compiler_flag
+  
+  set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -fexceptions")
+  set(CMAKE_CXX_FLAGS_DEBUG "-g3")
+  set(CMAKE_CXX_FLAGS_RELEASE "-g0 -O2")
+  
+  # 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")
@@ -161,42 +142,28 @@ if(NOT MSVC)
   ei_add_cxx_compiler_flag("-Wpointer-arith")
   ei_add_cxx_compiler_flag("-Wwrite-strings")
   ei_add_cxx_compiler_flag("-Wformat-security")
-  ei_add_cxx_compiler_flag("-Wshorten-64-to-32")
-  ei_add_cxx_compiler_flag("-Wlogical-op")
-  ei_add_cxx_compiler_flag("-Wenum-conversion")
-  ei_add_cxx_compiler_flag("-Wc++11-extensions")
-  ei_add_cxx_compiler_flag("-Wdouble-promotion")
-#  ei_add_cxx_compiler_flag("-Wconversion")
-
-  ei_add_cxx_compiler_flag("-Wshadow")
-
+  
   ei_add_cxx_compiler_flag("-Wno-psabi")
   ei_add_cxx_compiler_flag("-Wno-variadic-macros")
   ei_add_cxx_compiler_flag("-Wno-long-long")
-
+  
   ei_add_cxx_compiler_flag("-fno-check-new")
   ei_add_cxx_compiler_flag("-fno-common")
   ei_add_cxx_compiler_flag("-fstrict-aliasing")
   ei_add_cxx_compiler_flag("-wd981")                    # disable ICC's "operands are evaluated in unspecified order" remark
-  ei_add_cxx_compiler_flag("-wd2304")                   # disable ICC's "warning #2304: non-explicit constructor with single argument may cause implicit type conversion" produced by -Wnon-virtual-dtor
-
-
+  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()
-
-  if(ANDROID_NDK)
-    ei_add_cxx_compiler_flag("-pie")
-    ei_add_cxx_compiler_flag("-fPIE")
-  endif()
-
+  
   set(CMAKE_REQUIRED_FLAGS "")
 
   option(EIGEN_TEST_SSE2 "Enable/Disable SSE2 in tests/examples" OFF)
@@ -229,95 +196,18 @@ if(NOT MSVC)
     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_AVX2 "Enable/Disable AVX2 in tests/examples" OFF)
-  if(EIGEN_TEST_AVX2)
-    set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -mavx2 -mfma")
-    message(STATUS "Enabling AVX2 in tests/examples")
-  endif()
-
-  option(EIGEN_TEST_AVX512 "Enable/Disable AVX512 in tests/examples" OFF)
-  if(EIGEN_TEST_AVX512)
-    set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -mavx512f -mfma")
-    if (NOT "${CMAKE_CXX_COMPILER_ID}" STREQUAL "Clang")
-      set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -fabi-version=6")
-    endif()
-    message(STATUS "Enabling AVX512 in tests/examples")
-  endif()
-
-  option(EIGEN_TEST_AVX512DQ "Enable/Disable AVX512DQ in tests/examples" OFF)
-  if(EIGEN_TEST_AVX512DQ)
-    set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -mavx512dq")
-    if (NOT "${CMAKE_CXX_COMPILER_ID}" STREQUAL "Clang")
-      set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -fabi-version=6")
-    endif()
-    message(STATUS "Enabling AVX512DQ in tests/examples")
-  endif()
-
-  option(EIGEN_TEST_F16C "Enable/Disable F16C in tests/examples" OFF)
-  if(EIGEN_TEST_F16C)
-    set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -mf16c")
-    message(STATUS "Enabling F16C in tests/examples")
-  endif()
-
   option(EIGEN_TEST_ALTIVEC "Enable/Disable AltiVec in tests/examples" OFF)
   if(EIGEN_TEST_ALTIVEC)
     set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -maltivec -mabi=altivec")
     message(STATUS "Enabling AltiVec in tests/examples")
   endif()
 
-  option(EIGEN_TEST_VSX "Enable/Disable VSX in tests/examples" OFF)
-  if(EIGEN_TEST_VSX)
-    set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -m64 -mvsx")
-    message(STATUS "Enabling VSX in tests/examples")
-  endif()
-
-  option(EIGEN_TEST_MSA "Enable/Disable MSA in tests/examples" OFF)
-  if(EIGEN_TEST_MSA)
-    set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -mmsa")
-    message(STATUS "Enabling MSA in tests/examples")
-  endif()
-
   option(EIGEN_TEST_NEON "Enable/Disable Neon in tests/examples" OFF)
   if(EIGEN_TEST_NEON)
-    if(EIGEN_TEST_FMA)
-      set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -mfpu=neon-vfpv4")
-    else()
-      set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -mfpu=neon")
-    endif()
-    set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -mfloat-abi=hard")
+    set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -mfpu=neon -mcpu=cortex-a"8)
     message(STATUS "Enabling NEON in tests/examples")
   endif()
 
-  option(EIGEN_TEST_NEON64 "Enable/Disable Neon in tests/examples" OFF)
-  if(EIGEN_TEST_NEON64)
-    set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS}")
-    message(STATUS "Enabling NEON in tests/examples")
-  endif()
-
-  option(EIGEN_TEST_Z13 "Enable/Disable S390X(zEC13) ZVECTOR in tests/examples" OFF)
-  if(EIGEN_TEST_Z13)
-    set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -march=z13 -mzvector")
-    message(STATUS "Enabling S390X(zEC13) ZVECTOR in tests/examples")
-  endif()
-
-  option(EIGEN_TEST_Z14 "Enable/Disable S390X(zEC14) ZVECTOR in tests/examples" OFF)
-  if(EIGEN_TEST_Z14)
-    set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -march=z14 -mzvector")
-    message(STATUS "Enabling S390X(zEC13) ZVECTOR in tests/examples")
-  endif()
-
   check_cxx_compiler_flag("-fopenmp" COMPILER_SUPPORT_OPENMP)
   if(COMPILER_SUPPORT_OPENMP)
     option(EIGEN_TEST_OPENMP "Enable/Disable OpenMP in tests/examples" OFF)
@@ -327,7 +217,7 @@ if(NOT MSVC)
     endif()
   endif()
 
-else()
+else(NOT MSVC)
 
   # C4127 - conditional expression is constant
   # C4714 - marked as __forceinline not inlined (I failed to deactivate it selectively)
@@ -335,7 +225,7 @@ else()
   #         because we are oftentimes returning objects that have a destructor or may
   #         throw exceptions - in particular in the unit tests we are throwing extra many
   #         exceptions to cover indexing errors.
-  # C4505 - unreferenced local function has been removed (impossible to deactivate selectively)
+  # C4505 - unreferenced local function has been removed (impossible to deactive selectively)
   set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} /EHsc /wd4127 /wd4505 /wd4714")
 
   # replace all /Wx by /W4
@@ -355,23 +245,10 @@ else()
     if(NOT CMAKE_CL_64)
       # arch is not supported on 64 bit systems, SSE is enabled automatically.
       set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} /arch:SSE2")
-    endif()
+    endif(NOT CMAKE_CL_64)
     message(STATUS "Enabling SSE2 in tests/examples")
-  endif()
-
-  option(EIGEN_TEST_AVX "Enable/Disable AVX in tests/examples" OFF)
-  if(EIGEN_TEST_AVX)
-    set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} /arch:AVX")
-    message(STATUS "Enabling AVX in tests/examples")
-  endif()
-
-  option(EIGEN_TEST_FMA "Enable/Disable FMA/AVX2 in tests/examples" OFF)
-  if(EIGEN_TEST_FMA AND NOT EIGEN_TEST_NEON)
-    set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} /arch:AVX2")
-    message(STATUS "Enabling FMA/AVX2 in tests/examples")
-  endif()
-
-endif()
+  endif(EIGEN_TEST_SSE2)
+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)
@@ -407,44 +284,28 @@ if(EIGEN_TEST_NO_EXPLICIT_ALIGNMENT)
   message(STATUS "Disabling alignment in tests/examples")
 endif()
 
-option(EIGEN_TEST_NO_EXCEPTIONS "Disables C++ exceptions" OFF)
-if(EIGEN_TEST_NO_EXCEPTIONS)
-  ei_add_cxx_compiler_flag("-fno-exceptions")
-  message(STATUS "Disabling exceptions in tests/examples")
-endif()
+option(EIGEN_TEST_C++0x "Enables all C++0x features." OFF)
 
-set(EIGEN_CUDA_COMPUTE_ARCH 30 CACHE STRING "The CUDA compute architecture level to target when compiling CUDA code")
+include_directories(${CMAKE_CURRENT_SOURCE_DIR} ${CMAKE_CURRENT_BINARY_DIR})
 
-include_directories(${CMAKE_CURRENT_SOURCE_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)")
 
-# Backward compatibility support for EIGEN_INCLUDE_INSTALL_DIR
+# 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.
 if(EIGEN_INCLUDE_INSTALL_DIR)
-  message(WARNING "EIGEN_INCLUDE_INSTALL_DIR is deprecated. Use INCLUDE_INSTALL_DIR instead.")
-endif()
-
-if(EIGEN_INCLUDE_INSTALL_DIR AND NOT INCLUDE_INSTALL_DIR)
-  set(INCLUDE_INSTALL_DIR ${EIGEN_INCLUDE_INSTALL_DIR}
-      CACHE STRING "The directory relative to CMAKE_PREFIX_PATH where Eigen header files are installed")
+  set(INCLUDE_INSTALL_DIR
+    ${EIGEN_INCLUDE_INSTALL_DIR}
+    CACHE INTERNAL
+    "The directory where we install the header files (internal)"
+  )
 else()
   set(INCLUDE_INSTALL_DIR
-      "${CMAKE_INSTALL_INCLUDEDIR}/eigen3"
-      CACHE STRING "The directory relative to CMAKE_PREFIX_PATH where Eigen header files are installed"
-      )
+    "${CMAKE_INSTALL_PREFIX}/include/eigen3"
+    CACHE INTERNAL
+    "The directory where we install the header files (internal)"
+  )
 endif()
-set(CMAKEPACKAGE_INSTALL_DIR
-    "${CMAKE_INSTALL_DATADIR}/eigen3/cmake"
-    CACHE STRING "The directory relative to CMAKE_PREFIX_PATH where Eigen3Config.cmake is installed"
-    )
-set(PKGCONFIG_INSTALL_DIR
-    "${CMAKE_INSTALL_DATADIR}/pkgconfig"
-    CACHE STRING "The directory relative to CMAKE_PREFIX_PATH where eigen3.pc is installed"
-    )
-
-foreach(var INCLUDE_INSTALL_DIR CMAKEPACKAGE_INSTALL_DIR PKGCONFIG_INSTALL_DIR)
-  if(IS_ABSOLUTE "${${var}}")
-    message(FATAL_ERROR "${var} must be relative to CMAKE_PREFIX_PATH. Got: ${${var}}")
-  endif()
-endforeach()
 
 # similar to set_target_properties but append the property instead of overwriting it
 macro(ei_add_target_property target prop value)
@@ -453,9 +314,9 @@ macro(ei_add_target_property target prop value)
   # if the property wasn't previously set, ${previous} is now "previous-NOTFOUND" which cmake allows catching with plain if()
   if(NOT previous)
     set(previous "")
-  endif()
+  endif(NOT previous)
   set_target_properties(${target} PROPERTIES ${prop} "${previous} ${value}")
-endmacro()
+endmacro(ei_add_target_property)
 
 install(FILES
   signature_of_eigen3_matrix_library
@@ -463,32 +324,38 @@ install(FILES
   )
 
 if(EIGEN_BUILD_PKGCONFIG)
-    configure_file(eigen3.pc.in eigen3.pc @ONLY)
+    SET(path_separator ":")
+    STRING(REPLACE ${path_separator} ";" pkg_config_libdir_search "$ENV{PKG_CONFIG_LIBDIR}")
+    message(STATUS "searching for 'pkgconfig' directory in PKG_CONFIG_LIBDIR ( $ENV{PKG_CONFIG_LIBDIR} ), ${CMAKE_INSTALL_PREFIX}/share, and ${CMAKE_INSTALL_PREFIX}/lib")
+    FIND_PATH(pkg_config_libdir pkgconfig ${pkg_config_libdir_search} ${CMAKE_INSTALL_PREFIX}/share ${CMAKE_INSTALL_PREFIX}/lib ${pkg_config_libdir_search})
+    if(pkg_config_libdir)
+        SET(pkg_config_install_dir ${pkg_config_libdir})
+        message(STATUS "found ${pkg_config_libdir}/pkgconfig" )
+    else(pkg_config_libdir)
+        SET(pkg_config_install_dir ${CMAKE_INSTALL_PREFIX}/share)
+        message(STATUS "pkgconfig not found; installing in ${pkg_config_install_dir}" )
+    endif(pkg_config_libdir)
+
+    configure_file(eigen3.pc.in eigen3.pc)
     install(FILES ${CMAKE_CURRENT_BINARY_DIR}/eigen3.pc
-        DESTINATION ${PKGCONFIG_INSTALL_DIR}
+        DESTINATION ${pkg_config_install_dir}/pkgconfig
         )
-endif()
+endif(EIGEN_BUILD_PKGCONFIG)
 
-install(DIRECTORY Eigen DESTINATION ${INCLUDE_INSTALL_DIR} COMPONENT Devel)
+add_subdirectory(Eigen)
+
+add_subdirectory(doc EXCLUDE_FROM_ALL)
+
+include(EigenConfigureTesting)
+
+# fixme, not sure this line is still needed:
+enable_testing() # must be called from the root CMakeLists, see man page
 
 
-option(EIGEN_BUILD_DOC "Enable creation of Eigen documentation" ON)
-if(EIGEN_BUILD_DOC)
-  add_subdirectory(doc EXCLUDE_FROM_ALL)
-endif()
-
-
-option(BUILD_TESTING "Enable creation of Eigen tests." ON)
-if(BUILD_TESTING)
-  include(EigenConfigureTesting)
-
-  if(EIGEN_LEAVE_TEST_IN_ALL_TARGET)
-    add_subdirectory(test) # can't do EXCLUDE_FROM_ALL here, breaks CTest
-  else()
-    add_subdirectory(test EXCLUDE_FROM_ALL)
-  endif()
-
-  add_subdirectory(failtest)
+if(EIGEN_LEAVE_TEST_IN_ALL_TARGET)
+  add_subdirectory(test) # can't do EXCLUDE_FROM_ALL here, breaks CTest
+else()
+  add_subdirectory(test EXCLUDE_FROM_ALL)
 endif()
 
 if(EIGEN_LEAVE_TEST_IN_ALL_TARGET)
@@ -499,36 +366,6 @@ else()
   add_subdirectory(lapack EXCLUDE_FROM_ALL)
 endif()
 
-# add SYCL
-option(EIGEN_TEST_SYCL "Add Sycl support." OFF)
-option(EIGEN_SYCL_TRISYCL "Use the triSYCL Sycl implementation (ComputeCPP by default)." OFF)
-if(EIGEN_TEST_SYCL)
-  set (CMAKE_MODULE_PATH "${CMAKE_ROOT}/Modules" "cmake/Modules/" "${CMAKE_MODULE_PATH}")
-  find_package(Threads REQUIRED)
-  if(EIGEN_SYCL_TRISYCL)
-    message(STATUS "Using triSYCL")
-    include(FindTriSYCL)
-  else()
-    message(STATUS "Using ComputeCPP SYCL")
-    include(FindComputeCpp)
-    set(COMPUTECPP_DRIVER_DEFAULT_VALUE OFF)
-    if (NOT MSVC)
-      set(COMPUTECPP_DRIVER_DEFAULT_VALUE ON)
-    endif()
-    option(COMPUTECPP_USE_COMPILER_DRIVER
-      "Use ComputeCpp driver instead of a 2 steps compilation"
-      ${COMPUTECPP_DRIVER_DEFAULT_VALUE}
-    )
-  endif(EIGEN_SYCL_TRISYCL)
-  option(EIGEN_DONT_VECTORIZE_SYCL "Don't use vectorisation in the SYCL tests." OFF)
-  if(EIGEN_DONT_VECTORIZE_SYCL)
-    message(STATUS "Disabling SYCL vectorization in tests/examples")
-    # When disabling SYCL vectorization, also disable Eigen default vectorization
-    add_definitions(-DEIGEN_DONT_VECTORIZE=1)
-    add_definitions(-DEIGEN_DONT_VECTORIZE_SYCL=1)
-  endif()
-endif()
-
 add_subdirectory(unsupported)
 
 add_subdirectory(demos EXCLUDE_FROM_ALL)
@@ -539,111 +376,45 @@ add_subdirectory(scripts EXCLUDE_FROM_ALL)
 # TODO: consider also replacing EIGEN_BUILD_BTL by a custom target "make btl"?
 if(EIGEN_BUILD_BTL)
   add_subdirectory(bench/btl EXCLUDE_FROM_ALL)
-endif()
+endif(EIGEN_BUILD_BTL)
 
 if(NOT WIN32)
   add_subdirectory(bench/spbench EXCLUDE_FROM_ALL)
-endif()
+endif(NOT WIN32)
 
 configure_file(scripts/cdashtesting.cmake.in cdashtesting.cmake @ONLY)
 
-if(BUILD_TESTING)
-  ei_testing_print_summary()
-endif()
+ei_testing_print_summary()
 
 message(STATUS "")
 message(STATUS "Configured Eigen ${EIGEN_VERSION_NUMBER}")
 message(STATUS "")
 
+option(EIGEN_FAILTEST "Enable failtests." OFF)
+if(EIGEN_FAILTEST)
+  add_subdirectory(failtest)
+endif()
+
 string(TOLOWER "${CMAKE_GENERATOR}" cmake_generator_tolower)
 if(cmake_generator_tolower MATCHES "makefile")
-  message(STATUS "Available targets (use: make TARGET):")
+  message(STATUS "Some things you can do now:")
+  message(STATUS "--------------+--------------------------------------------------------------")
+  message(STATUS "Command       |   Description")
+  message(STATUS "--------------+--------------------------------------------------------------")
+  message(STATUS "make install  | Install to ${CMAKE_INSTALL_PREFIX}. To change that:")
+  message(STATUS "              |     cmake . -DCMAKE_INSTALL_PREFIX=yourpath")
+  message(STATUS "              |   Eigen headers will then be installed to:")
+  message(STATUS "              |     ${INCLUDE_INSTALL_DIR}")
+  message(STATUS "              |   To install Eigen headers to a separate location, do:")
+  message(STATUS "              |     cmake . -DEIGEN_INCLUDE_INSTALL_DIR=yourpath")
+  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 "--------------+--------------------------------------------------------------")
 else()
-  message(STATUS "Available targets (use: cmake --build . --target TARGET):")
+  message(STATUS "To build/run the unit tests, read this page:")
+  message(STATUS "  http://eigen.tuxfamily.org/index.php?title=Tests")
 endif()
-message(STATUS "---------+--------------------------------------------------------------")
-message(STATUS "Target   |   Description")
-message(STATUS "---------+--------------------------------------------------------------")
-message(STATUS "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 "doc      | Generate the API documentation, requires Doxygen & LaTeX")
-if(BUILD_TESTING)
-  message(STATUS "check    | Build and run the unit-tests. Read this page:")
-  message(STATUS "         |   http://eigen.tuxfamily.org/index.php?title=Tests")
-endif()
-message(STATUS "blas     | Build BLAS library (not the same thing as Eigen)")
-message(STATUS "uninstall| Remove files installed by the install target")
-message(STATUS "---------+--------------------------------------------------------------")
+
 message(STATUS "")
-
-
-set ( EIGEN_VERSION_STRING ${EIGEN_VERSION_NUMBER} )
-set ( EIGEN_VERSION_MAJOR  ${EIGEN_WORLD_VERSION} )
-set ( EIGEN_VERSION_MINOR  ${EIGEN_MAJOR_VERSION} )
-set ( EIGEN_VERSION_PATCH  ${EIGEN_MINOR_VERSION} )
-set ( EIGEN_DEFINITIONS "")
-set ( EIGEN_INCLUDE_DIR "${CMAKE_INSTALL_PREFIX}/${INCLUDE_INSTALL_DIR}" )
-set ( EIGEN_ROOT_DIR ${CMAKE_INSTALL_PREFIX} )
-
-include (CMakePackageConfigHelpers)
-
-# Imported target support
-add_library (eigen INTERFACE)
-add_library (Eigen3::Eigen ALIAS eigen)
-target_compile_definitions (eigen INTERFACE ${EIGEN_DEFINITIONS})
-target_include_directories (eigen INTERFACE
-  $<BUILD_INTERFACE:${CMAKE_CURRENT_SOURCE_DIR}>
-  $<INSTALL_INTERFACE:${INCLUDE_INSTALL_DIR}>
-)
-
-# Export as title case Eigen
-set_target_properties (eigen PROPERTIES EXPORT_NAME Eigen)
-
-install (TARGETS eigen EXPORT Eigen3Targets)
-
-configure_package_config_file (
-  ${CMAKE_CURRENT_SOURCE_DIR}/cmake/Eigen3Config.cmake.in
-  ${CMAKE_CURRENT_BINARY_DIR}/Eigen3Config.cmake
-  PATH_VARS EIGEN_INCLUDE_DIR EIGEN_ROOT_DIR
-  INSTALL_DESTINATION ${CMAKEPACKAGE_INSTALL_DIR}
-  NO_CHECK_REQUIRED_COMPONENTS_MACRO # Eigen does not provide components
-)
-# Remove CMAKE_SIZEOF_VOID_P from Eigen3ConfigVersion.cmake since Eigen does
-# not depend on architecture specific settings or libraries. More
-# specifically, an Eigen3Config.cmake generated from a 64 bit target can be
-# used for 32 bit targets as well (and vice versa).
-set (_Eigen3_CMAKE_SIZEOF_VOID_P ${CMAKE_SIZEOF_VOID_P})
-unset (CMAKE_SIZEOF_VOID_P)
-write_basic_package_version_file (Eigen3ConfigVersion.cmake
-                                  VERSION ${EIGEN_VERSION_NUMBER}
-                                  COMPATIBILITY SameMajorVersion)
-set (CMAKE_SIZEOF_VOID_P ${_Eigen3_CMAKE_SIZEOF_VOID_P})
-
-# The Eigen target will be located in the Eigen3 namespace. Other CMake
-# targets can refer to it using Eigen3::Eigen.
-export (TARGETS eigen NAMESPACE Eigen3:: FILE Eigen3Targets.cmake)
-# Export Eigen3 package to CMake registry such that it can be easily found by
-# CMake even if it has not been installed to a standard directory.
-export (PACKAGE Eigen3)
-
-install (EXPORT Eigen3Targets NAMESPACE Eigen3:: DESTINATION ${CMAKEPACKAGE_INSTALL_DIR})
-
-install ( FILES ${CMAKE_CURRENT_SOURCE_DIR}/cmake/UseEigen3.cmake
-                ${CMAKE_CURRENT_BINARY_DIR}/Eigen3Config.cmake
-                ${CMAKE_CURRENT_BINARY_DIR}/Eigen3ConfigVersion.cmake
-          DESTINATION ${CMAKEPACKAGE_INSTALL_DIR} )
-
-# Add uninstall target
-add_custom_target ( uninstall
-    COMMAND ${CMAKE_COMMAND} -P ${CMAKE_CURRENT_SOURCE_DIR}/cmake/EigenUninstall.cmake)
-
-if (EIGEN_SPLIT_TESTSUITE)
-  ei_split_testsuite("${EIGEN_SPLIT_TESTSUITE}")
-endif()

COPYING.APACHE

@@ -1,203 +0,0 @@
-/*
-                                 Apache License
-                           Version 2.0, January 2004
-                        http://www.apache.org/licenses/
-
-   TERMS AND CONDITIONS FOR USE, REPRODUCTION, AND DISTRIBUTION
-
-   1. Definitions.
-
-      "License" shall mean the terms and conditions for use, reproduction,
-      and distribution as defined by Sections 1 through 9 of this document.
-
-      "Licensor" shall mean the copyright owner or entity authorized by
-      the copyright owner that is granting the License.
-
-      "Legal Entity" shall mean the union of the acting entity and all
-      other entities that control, are controlled by, or are under common
-      control with that entity. For the purposes of this definition,
-      "control" means (i) the power, direct or indirect, to cause the
-      direction or management of such entity, whether by contract or
-      otherwise, or (ii) ownership of fifty percent (50%) or more of the
-      outstanding shares, or (iii) beneficial ownership of such entity.
-
-      "You" (or "Your") shall mean an individual or Legal Entity
-      exercising permissions granted by this License.
-
-      "Source" form shall mean the preferred form for making modifications,
-      including but not limited to software source code, documentation
-      source, and configuration files.
-
-      "Object" form shall mean any form resulting from mechanical
-      transformation or translation of a Source form, including but
-      not limited to compiled object code, generated documentation,
-      and conversions to other media types.
-
-      "Work" shall mean the work of authorship, whether in Source or
-      Object form, made available under the License, as indicated by a
-      copyright notice that is included in or attached to the work
-      (an example is provided in the Appendix below).
-
-      "Derivative Works" shall mean any work, whether in Source or Object
-      form, that is based on (or derived from) the Work and for which the
-      editorial revisions, annotations, elaborations, or other modifications
-      represent, as a whole, an original work of authorship. For the purposes
-      of this License, Derivative Works shall not include works that remain
-      separable from, or merely link (or bind by name) to the interfaces of,
-      the Work and Derivative Works thereof.
-
-      "Contribution" shall mean any work of authorship, including
-      the original version of the Work and any modifications or additions
-      to that Work or Derivative Works thereof, that is intentionally
-      submitted to Licensor for inclusion in the Work by the copyright owner
-      or by an individual or Legal Entity authorized to submit on behalf of
-      the copyright owner. For the purposes of this definition, "submitted"
-      means any form of electronic, verbal, or written communication sent
-      to the Licensor or its representatives, including but not limited to
-      communication on electronic mailing lists, source code control systems,
-      and issue tracking systems that are managed by, or on behalf of, the
-      Licensor for the purpose of discussing and improving the Work, but
-      excluding communication that is conspicuously marked or otherwise
-      designated in writing by the copyright owner as "Not a Contribution."
-
-      "Contributor" shall mean Licensor and any individual or Legal Entity
-      on behalf of whom a Contribution has been received by Licensor and
-      subsequently incorporated within the Work.
-
-   2. Grant of Copyright License. Subject to the terms and conditions of
-      this License, each Contributor hereby grants to You a perpetual,
-      worldwide, non-exclusive, no-charge, royalty-free, irrevocable
-      copyright license to reproduce, prepare Derivative Works of,
-      publicly display, publicly perform, sublicense, and distribute the
-      Work and such Derivative Works in Source or Object form.
-
-   3. Grant of Patent License. Subject to the terms and conditions of
-      this License, each Contributor hereby grants to You a perpetual,
-      worldwide, non-exclusive, no-charge, royalty-free, irrevocable
-      (except as stated in this section) patent license to make, have made,
-      use, offer to sell, sell, import, and otherwise transfer the Work,
-      where such license applies only to those patent claims licensable
-      by such Contributor that are necessarily infringed by their
-      Contribution(s) alone or by combination of their Contribution(s)
-      with the Work to which such Contribution(s) was submitted. If You
-      institute patent litigation against any entity (including a
-      cross-claim or counterclaim in a lawsuit) alleging that the Work
-      or a Contribution incorporated within the Work constitutes direct
-      or contributory patent infringement, then any patent licenses
-      granted to You under this License for that Work shall terminate
-      as of the date such litigation is filed.
-
-   4. Redistribution. You may reproduce and distribute copies of the
-      Work or Derivative Works thereof in any medium, with or without
-      modifications, and in Source or Object form, provided that You
-      meet the following conditions:
-
-      (a) You must give any other recipients of the Work or
-          Derivative Works a copy of this License; and
-
-      (b) You must cause any modified files to carry prominent notices
-          stating that You changed the files; and
-
-      (c) You must retain, in the Source form of any Derivative Works
-          that You distribute, all copyright, patent, trademark, and
-          attribution notices from the Source form of the Work,
-          excluding those notices that do not pertain to any part of
-          the Derivative Works; and
-
-      (d) If the Work includes a "NOTICE" text file as part of its
-          distribution, then any Derivative Works that You distribute must
-          include a readable copy of the attribution notices contained
-          within such NOTICE file, excluding those notices that do not
-          pertain to any part of the Derivative Works, in at least one
-          of the following places: within a NOTICE text file distributed
-          as part of the Derivative Works; within the Source form or
-          documentation, if provided along with the Derivative Works; or,
-          within a display generated by the Derivative Works, if and
-          wherever such third-party notices normally appear. The contents
-          of the NOTICE file are for informational purposes only and
-          do not modify the License. You may add Your own attribution
-          notices within Derivative Works that You distribute, alongside
-          or as an addendum to the NOTICE text from the Work, provided
-          that such additional attribution notices cannot be construed
-          as modifying the License.
-
-      You may add Your own copyright statement to Your modifications and
-      may provide additional or different license terms and conditions
-      for use, reproduction, or distribution of Your modifications, or
-      for any such Derivative Works as a whole, provided Your use,
-      reproduction, and distribution of the Work otherwise complies with
-      the conditions stated in this License.
-
-   5. Submission of Contributions. Unless You explicitly state otherwise,
-      any Contribution intentionally submitted for inclusion in the Work
-      by You to the Licensor shall be under the terms and conditions of
-      this License, without any additional terms or conditions.
-      Notwithstanding the above, nothing herein shall supersede or modify
-      the terms of any separate license agreement you may have executed
-      with Licensor regarding such Contributions.
-
-   6. Trademarks. This License does not grant permission to use the trade
-      names, trademarks, service marks, or product names of the Licensor,
-      except as required for reasonable and customary use in describing the
-      origin of the Work and reproducing the content of the NOTICE file.
-
-   7. Disclaimer of Warranty. Unless required by applicable law or
-      agreed to in writing, Licensor provides the Work (and each
-      Contributor provides its Contributions) on an "AS IS" BASIS,
-      WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or
-      implied, including, without limitation, any warranties or conditions
-      of TITLE, NON-INFRINGEMENT, MERCHANTABILITY, or FITNESS FOR A
-      PARTICULAR PURPOSE. You are solely responsible for determining the
-      appropriateness of using or redistributing the Work and assume any
-      risks associated with Your exercise of permissions under this License.
-
-   8. Limitation of Liability. In no event and under no legal theory,
-      whether in tort (including negligence), contract, or otherwise,
-      unless required by applicable law (such as deliberate and grossly
-      negligent acts) or agreed to in writing, shall any Contributor be
-      liable to You for damages, including any direct, indirect, special,
-      incidental, or consequential damages of any character arising as a
-      result of this License or out of the use or inability to use the
-      Work (including but not limited to damages for loss of goodwill,
-      work stoppage, computer failure or malfunction, or any and all
-      other commercial damages or losses), even if such Contributor
-      has been advised of the possibility of such damages.
-
-   9. Accepting Warranty or Additional Liability. While redistributing
-      the Work or Derivative Works thereof, You may choose to offer,
-      and charge a fee for, acceptance of support, warranty, indemnity,
-      or other liability obligations and/or rights consistent with this
-      License. However, in accepting such obligations, You may act only
-      on Your own behalf and on Your sole responsibility, not on behalf
-      of any other Contributor, and only if You agree to indemnify,
-      defend, and hold each Contributor harmless for any liability
-      incurred by, or claims asserted against, such Contributor by reason
-      of your accepting any such warranty or additional liability.
-
-   END OF TERMS AND CONDITIONS
-
-   APPENDIX: How to apply the Apache License to your work.
-
-      To apply the Apache License to your work, attach the following
-      boilerplate notice, with the fields enclosed by brackets "[]"
-      replaced with your own identifying information. (Don't include
-      the brackets!)  The text should be enclosed in the appropriate
-      comment syntax for the file format. We also recommend that a
-      file or class name and description of purpose be included on the
-      same "printed page" as the copyright notice for easier
-      identification within third-party archives.
-
-   Copyright [yyyy] [name of copyright owner]
-
-   Licensed under the Apache License, Version 2.0 (the "License");
-   you may not use this file except in compliance with the License.
-   You may obtain a copy of the License at
-
-       http://www.apache.org/licenses/LICENSE-2.0
-
-   Unless required by applicable law or agreed to in writing, software
-   distributed under the License is distributed on an "AS IS" BASIS,
-   WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
-   See the License for the specific language governing permissions and
-   limitations under the License.
-*/

COPYING.BSD

@@ -23,4 +23,4 @@
  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.MINPACK

@@ -1,51 +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.
+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.
+

CTestConfig.cmake

@@ -2,16 +2,16 @@
 ## Then modify the CMakeLists.txt file in the root directory of your
 ## project to incorporate the testing dashboard.
 ## # The following are required to uses Dart and the Cdash dashboard
-##   enable_testing()
-##   include(CTest)
+##   ENABLE_TESTING()
+##   INCLUDE(CTest)
 set(CTEST_PROJECT_NAME "Eigen")
 set(CTEST_NIGHTLY_START_TIME "00:00:00 UTC")
 
 set(CTEST_DROP_METHOD "http")
-set(CTEST_DROP_SITE "my.cdash.org")
-set(CTEST_DROP_LOCATION "/submit.php?project=Eigen")
+set(CTEST_DROP_SITE "manao.inria.fr")
+set(CTEST_DROP_LOCATION "/CDash/submit.php?project=Eigen")
 set(CTEST_DROP_SITE_CDASH TRUE)
-#set(CTEST_PROJECT_SUBPROJECTS
-#Official
-#Unsupported
-#)
+set(CTEST_PROJECT_SUBPROJECTS
+Official
+Unsupported
+)

CTestCustom.cmake.in

@@ -1,4 +1,3 @@
 
 set(CTEST_CUSTOM_MAXIMUM_NUMBER_OF_WARNINGS "2000")
 set(CTEST_CUSTOM_MAXIMUM_NUMBER_OF_ERRORS   "2000")
-list(APPEND CTEST_CUSTOM_ERROR_EXCEPTION    @EIGEN_CTEST_ERROR_EXCEPTION@)

Eigen/Array

@@ -0,0 +1,11 @@
+#ifndef EIGEN_ARRAY_MODULE_H
+#define EIGEN_ARRAY_MODULE_H
+
+// include Core first to handle Eigen2 support macros
+#include "Core"
+
+#ifndef EIGEN2_SUPPORT
+  #error The Eigen/Array header does no longer exist in Eigen3. All that functionality has moved to Eigen/Core.
+#endif
+
+#endif // EIGEN_ARRAY_MODULE_H

Eigen/CMakeLists.txt

@@ -0,0 +1,19 @@
+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 ${Eigen_directory_files})
+  if(NOT f MATCHES "\\.txt" AND NOT f MATCHES "${ESCAPED_CMAKE_CURRENT_SOURCE_DIR}/[.].+" AND NOT f MATCHES "${ESCAPED_CMAKE_CURRENT_SOURCE_DIR}/src")
+    list(APPEND Eigen_directory_files_to_install ${f})
+  endif()
+endforeach(f ${Eigen_directory_files})
+
+install(FILES
+  ${Eigen_directory_files_to_install}
+  DESTINATION ${INCLUDE_INSTALL_DIR}/Eigen COMPONENT Devel
+  )
+
+add_subdirectory(src)

Eigen/Cholesky

@@ -1,15 +1,7 @@
-// This file is part of Eigen, a lightweight C++ template library
-// for linear algebra.
-//
-// This Source Code Form is subject to the terms of the Mozilla
-// Public License v. 2.0. If a copy of the MPL was not distributed
-// with this file, You can obtain one at http://mozilla.org/MPL/2.0/.
-
 #ifndef EIGEN_CHOLESKY_MODULE_H
 #define EIGEN_CHOLESKY_MODULE_H
 
 #include "Core"
-#include "Jacobi"
 
 #include "src/Core/util/DisableStupidWarnings.h"
 
@@ -18,28 +10,23 @@
   *
   *
   * This module provides two variants of the Cholesky decomposition for selfadjoint (hermitian) matrices.
-  * Those decompositions are also accessible via the following methods:
-  *  - MatrixBase::llt()
+  * Those decompositions are accessible via the following MatrixBase 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
-#ifdef EIGEN_USE_MKL
-#include "mkl_lapacke.h"
-#else
-#include "src/misc/lapacke.h"
-#endif
-#include "src/Cholesky/LLT_LAPACKE.h"
+#include "src/Cholesky/LLT_MKL.h"
 #endif
 
 #include "src/Core/util/ReenableStupidWarnings.h"
 
 #endif // EIGEN_CHOLESKY_MODULE_H
+/* vim: set filetype=cpp et sw=2 ts=2 ai: */

Eigen/CholmodSupport

@@ -1,10 +1,3 @@
-// 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
 
@@ -19,7 +12,7 @@ extern "C" {
 /** \ingroup Support_modules
   * \defgroup CholmodSupport_Module CholmodSupport module
   *
-  * This module provides an interface to the Cholmod library which is part of the <a href="http://www.suitesparse.com">suitesparse</a> package.
+  * 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).
@@ -40,8 +33,12 @@ extern "C" {
   *
   */
 
+#include "src/misc/Solve.h"
+#include "src/misc/SparseSolve.h"
+
 #include "src/CholmodSupport/CholmodSupport.h"
 
+
 #include "src/Core/util/ReenableStupidWarnings.h"
 
 #endif // EIGEN_CHOLMODSUPPORT_MODULE_H

Eigen/Core

@@ -11,55 +11,123 @@
 #ifndef EIGEN_CORE_H
 #define EIGEN_CORE_H
 
-// first thing Eigen does: stop the compiler from reporting useless warnings.
+// first thing Eigen does: stop the compiler from committing suicide
 #include "src/Core/util/DisableStupidWarnings.h"
 
 // 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 compiler/OS/arch detections and define most defaults.
+// it's where we do all the alignment settings (platform detection and honoring the user's will if he
+// defined e.g. EIGEN_DONT_ALIGN) so it needs to be done before we do anything with vectorization.
 #include "src/Core/util/Macros.h"
 
-// This detects SSE/AVX/NEON/etc. and configure alignment settings
-#include "src/Core/util/ConfigureVectorization.h"
-
-// We need cuda_runtime.h/hip_runtime.h to ensure that
-// the EIGEN_USING_STD macro works properly on the device side
-#if defined(EIGEN_CUDACC)
-  #include <cuda_runtime.h>
-#elif defined(EIGEN_HIPCC)
-  #include <hip/hip_runtime.h>
-#endif
-
-
-#ifdef EIGEN_EXCEPTIONS
-  #include <new>
-#endif
-
 // 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) && EIGEN_GNUC_AT_MOST(5,5)
+#if defined(__MINGW32__) && EIGEN_GNUC_AT_LEAST(4,6)
   #pragma GCC optimize ("-fno-ipa-cp-clone")
 #endif
 
-// Prevent ICC from specializing std::complex operators that silently fail
-// on device. This allows us to use our own device-compatible specializations
-// instead.
-#if defined(EIGEN_COMP_ICC) && defined(EIGEN_GPU_COMPILE_PHASE) \
-    && !defined(_OVERRIDE_COMPLEX_SPECIALIZATION_)
-#define _OVERRIDE_COMPLEX_SPECIALIZATION_ 1
-#endif
 #include <complex>
 
 // this include file manages BLAS and MKL related macros
 // and inclusion of their respective header files
 #include "src/Core/util/MKL_support.h"
 
-
-#if defined(EIGEN_HAS_CUDA_FP16) || defined(EIGEN_HAS_HIP_FP16)
-  #define EIGEN_HAS_GPU_FP16
+// 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
+  #ifndef EIGEN_DONT_VECTORIZE
+    #define EIGEN_DONT_VECTORIZE
+  #endif
 #endif
 
-#if defined(EIGEN_HAS_CUDA_BF16) || defined(EIGEN_HAS_HIP_BF16)
-  #define EIGEN_HAS_GPU_BF16
+#ifdef _MSC_VER
+  #include <malloc.h> // for _aligned_malloc -- need it regardless of whether vectorization is enabled
+  #if (_MSC_VER >= 1500) // 2008 or later
+    // Remember that usage of defined() in a #define is undefined by the standard.
+    // a user reported that in 64-bit mode, MSVC doesn't care to define _M_IX86_FP.
+    #if (defined(_M_IX86_FP) && (_M_IX86_FP >= 2)) || defined(_M_X64)
+      #define EIGEN_SSE2_ON_MSVC_2008_OR_LATER
+    #endif
+  #endif
+#else
+  // Remember that usage of defined() in a #define is undefined by the standard
+  #if (defined __SSE2__) && ( (!defined __GNUC__) || (defined __INTEL_COMPILER) || 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_ON_NON_MSVC_BUT_NOT_OLD_GCC) || defined(EIGEN_SSE2_ON_MSVC_2008_OR_LATER)
+
+    // Defines symbols for compile-time detection of which instructions are
+    // used.
+    // EIGEN_VECTORIZE_YY is defined if and only if the instruction set YY is used
+    #define EIGEN_VECTORIZE
+    #define EIGEN_VECTORIZE_SSE
+    #define EIGEN_VECTORIZE_SSE2
+
+    // Detect sse3/ssse3/sse4:
+    // gcc and icc defines __SSE3__, ...
+    // there is no way to know about this on msvc. You can define EIGEN_VECTORIZE_SSE* if you
+    // want to force the use of those instructions with msvc.
+    #ifdef __SSE3__
+      #define EIGEN_VECTORIZE_SSE3
+    #endif
+    #ifdef __SSSE3__
+      #define EIGEN_VECTORIZE_SSSE3
+    #endif
+    #ifdef __SSE4_1__
+      #define EIGEN_VECTORIZE_SSE4_1
+    #endif
+    #ifdef __SSE4_2__
+      #define EIGEN_VECTORIZE_SSE4_2
+    #endif
+
+    // include files
+
+    // This extern "C" works around a MINGW-w64 compilation issue
+    // https://sourceforge.net/tracker/index.php?func=detail&aid=3018394&group_id=202880&atid=983354
+    // In essence, intrin.h is included by windows.h and also declares intrinsics (just as emmintrin.h etc. below do).
+    // However, intrin.h uses an extern "C" declaration, and g++ thus complains of duplicate declarations
+    // with conflicting linkage.  The linkage for intrinsics doesn't matter, but at that stage the compiler doesn't know;
+    // so, to avoid compile errors when windows.h is included after Eigen/Core, ensure intrinsics are extern "C" here too.
+    // notice that since these are C headers, the extern "C" is theoretically needed anyways.
+    extern "C" {
+      // In theory we should only include immintrin.h and not the other *mmintrin.h header files directly.
+      // Doing so triggers some issues with ICC. However old gcc versions seems to not have this file, thus:
+      #ifdef __INTEL_COMPILER
+        #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
+      #endif
+    } // end extern "C"
+  #elif defined __ALTIVEC__
+    #define EIGEN_VECTORIZE
+    #define EIGEN_VECTORIZE_ALTIVEC
+    #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  __ARM_NEON__
+    #define EIGEN_VECTORIZE
+    #define EIGEN_VECTORIZE_NEON
+    #include <arm_neon.h>
+  #endif
 #endif
 
 #if (defined _OPENMP) && (!defined EIGEN_DONT_PARALLELIZE)
@@ -71,7 +139,7 @@
 #endif
 
 // MSVC for windows mobile does not have the errno.h file
-#if !(EIGEN_COMP_MSVC && EIGEN_OS_WINCE) && !EIGEN_COMP_ARM
+#if !(defined(_MSC_VER) && defined(_WIN32_WCE)) && !defined(__ARMCC_VERSION)
 #define EIGEN_HAS_ERRNO
 #endif
 
@@ -83,10 +151,7 @@
 #include <cmath>
 #include <cassert>
 #include <functional>
-#include <sstream>
-#ifndef EIGEN_NO_IO
-  #include <iosfwd>
-#endif
+#include <iosfwd>
 #include <cstring>
 #include <string>
 #include <limits>
@@ -94,60 +159,85 @@
 // for min/max:
 #include <algorithm>
 
-#if EIGEN_HAS_CXX11
-#include <array>
-#endif
-
-// for std::is_nothrow_move_assignable
-#ifdef EIGEN_INCLUDE_TYPE_TRAITS
-#include <type_traits>
-#endif
-
 // for outputting debug info
 #ifdef EIGEN_DEBUG_ASSIGN
 #include <iostream>
 #endif
 
 // required for __cpuid, needs to be included after cmath
-#if EIGEN_COMP_MSVC && EIGEN_ARCH_i386_OR_x86_64 && !EIGEN_OS_WINCE
+#if defined(_MSC_VER) && (defined(_M_IX86)||defined(_M_X64))
   #include <intrin.h>
 #endif
 
-#if defined(EIGEN_USE_SYCL)
-  #undef min
-  #undef max
-  #undef isnan
-  #undef isinf
-  #undef isfinite
-  #include <CL/sycl.hpp>
-  #include <map>
-  #include <memory>
-  #include <utility>
-  #include <thread>
-  #ifndef EIGEN_SYCL_LOCAL_THREAD_DIM0
-  #define EIGEN_SYCL_LOCAL_THREAD_DIM0 16
-  #endif
-  #ifndef EIGEN_SYCL_LOCAL_THREAD_DIM1
-  #define EIGEN_SYCL_LOCAL_THREAD_DIM1 16
-  #endif
+#if defined(_CPPUNWIND) || defined(__EXCEPTIONS)
+  #define EIGEN_EXCEPTIONS
 #endif
 
-
-#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
+#ifdef EIGEN_EXCEPTIONS
+  #include <new>
 #endif
 
+/** \brief Namespace containing all symbols from the %Eigen library. */
 namespace Eigen {
 
-// we use size_t frequently and we'll never remember to prepend it with std:: every time just to
+inline static const char *SimdInstructionSetsInUse(void) {
+#if defined(EIGEN_VECTORIZE_SSE4_2)
+  return "SSE, SSE2, SSE3, SSSE3, SSE4.1, SSE4.2";
+#elif defined(EIGEN_VECTORIZE_SSE4_1)
+  return "SSE, SSE2, SSE3, SSSE3, SSE4.1";
+#elif defined(EIGEN_VECTORIZE_SSSE3)
+  return "SSE, SSE2, SSE3, SSSE3";
+#elif defined(EIGEN_VECTORIZE_SSE3)
+  return "SSE, SSE2, SSE3";
+#elif defined(EIGEN_VECTORIZE_SSE2)
+  return "SSE, SSE2";
+#elif defined(EIGEN_VECTORIZE_ALTIVEC)
+  return "AltiVec";
+#elif defined(EIGEN_VECTORIZE_NEON)
+  return "ARM NEON";
+#else
+  return "None";
+#endif
+}
+
+} // end namespace Eigen
+
+#define STAGE10_FULL_EIGEN2_API             10
+#define STAGE20_RESOLVE_API_CONFLICTS       20
+#define STAGE30_FULL_EIGEN3_API             30
+#define STAGE40_FULL_EIGEN3_STRICTNESS      40
+#define STAGE99_NO_EIGEN2_SUPPORT           99
+
+#if   defined EIGEN2_SUPPORT_STAGE40_FULL_EIGEN3_STRICTNESS
+  #define EIGEN2_SUPPORT
+  #define EIGEN2_SUPPORT_STAGE STAGE40_FULL_EIGEN3_STRICTNESS
+#elif defined EIGEN2_SUPPORT_STAGE30_FULL_EIGEN3_API
+  #define EIGEN2_SUPPORT
+  #define EIGEN2_SUPPORT_STAGE STAGE30_FULL_EIGEN3_API
+#elif defined EIGEN2_SUPPORT_STAGE20_RESOLVE_API_CONFLICTS
+  #define EIGEN2_SUPPORT
+  #define EIGEN2_SUPPORT_STAGE STAGE20_RESOLVE_API_CONFLICTS
+#elif defined EIGEN2_SUPPORT_STAGE10_FULL_EIGEN2_API
+  #define EIGEN2_SUPPORT
+  #define EIGEN2_SUPPORT_STAGE STAGE10_FULL_EIGEN2_API
+#elif defined EIGEN2_SUPPORT
+  // default to stage 3, that's what it's always meant
+  #define EIGEN2_SUPPORT_STAGE30_FULL_EIGEN3_API
+  #define EIGEN2_SUPPORT_STAGE STAGE30_FULL_EIGEN3_API
+#else
+  #define EIGEN2_SUPPORT_STAGE STAGE99_NO_EIGEN2_SUPPORT
+#endif
+
+#ifdef EIGEN2_SUPPORT
+#undef minor
+#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
+// gcc 4.6.0 wants std:: for ptrdiff_t 
 using std::ptrdiff_t;
 
-}
-
 /** \defgroup Core_Module Core module
   * This is the main module of Eigen providing dense matrix and vector support
   * (both fixed and dynamic size) with all the features corresponding to a BLAS library
@@ -159,142 +249,50 @@ using std::ptrdiff_t;
   */
 
 #include "src/Core/util/Constants.h"
-#include "src/Core/util/Meta.h"
 #include "src/Core/util/ForwardDeclarations.h"
+#include "src/Core/util/Meta.h"
 #include "src/Core/util/StaticAssert.h"
 #include "src/Core/util/XprHelper.h"
 #include "src/Core/util/Memory.h"
-#include "src/Core/util/IntegralConstant.h"
-#include "src/Core/util/SymbolicIndex.h"
 
 #include "src/Core/NumTraits.h"
 #include "src/Core/MathFunctions.h"
 #include "src/Core/GenericPacketMath.h"
-#include "src/Core/MathFunctionsImpl.h"
-#include "src/Core/arch/Default/ConjHelper.h"
-// Generic half float support
-#include "src/Core/arch/Default/Half.h"
-#include "src/Core/arch/Default/BFloat16.h"
-#include "src/Core/arch/Default/TypeCasting.h"
-#include "src/Core/arch/Default/GenericPacketMathFunctionsFwd.h"
 
-#if defined EIGEN_VECTORIZE_AVX512
+#if defined EIGEN_VECTORIZE_SSE
   #include "src/Core/arch/SSE/PacketMath.h"
-  #include "src/Core/arch/SSE/TypeCasting.h"
-  #include "src/Core/arch/SSE/Complex.h"
-  #include "src/Core/arch/AVX/PacketMath.h"
-  #include "src/Core/arch/AVX/TypeCasting.h"
-  #include "src/Core/arch/AVX/Complex.h"
-  #include "src/Core/arch/AVX512/PacketMath.h"
-  #include "src/Core/arch/AVX512/TypeCasting.h"
-  #include "src/Core/arch/AVX512/Complex.h"
-  #include "src/Core/arch/SSE/MathFunctions.h"
-  #include "src/Core/arch/AVX/MathFunctions.h"
-  #include "src/Core/arch/AVX512/MathFunctions.h"
-#elif defined EIGEN_VECTORIZE_AVX
-  // Use AVX for floats and doubles, SSE for integers
-  #include "src/Core/arch/SSE/PacketMath.h"
-  #include "src/Core/arch/SSE/TypeCasting.h"
-  #include "src/Core/arch/SSE/Complex.h"
-  #include "src/Core/arch/AVX/PacketMath.h"
-  #include "src/Core/arch/AVX/TypeCasting.h"
-  #include "src/Core/arch/AVX/Complex.h"
-  #include "src/Core/arch/SSE/MathFunctions.h"
-  #include "src/Core/arch/AVX/MathFunctions.h"
-#elif defined EIGEN_VECTORIZE_SSE
-  #include "src/Core/arch/SSE/PacketMath.h"
-  #include "src/Core/arch/SSE/TypeCasting.h"
   #include "src/Core/arch/SSE/MathFunctions.h"
   #include "src/Core/arch/SSE/Complex.h"
-#elif defined(EIGEN_VECTORIZE_ALTIVEC) || defined(EIGEN_VECTORIZE_VSX)
+#elif defined EIGEN_VECTORIZE_ALTIVEC
   #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/TypeCasting.h"
-  #include "src/Core/arch/NEON/MathFunctions.h"
   #include "src/Core/arch/NEON/Complex.h"
-#elif defined EIGEN_VECTORIZE_SVE
-  #include "src/Core/arch/SVE/PacketMath.h"
-  #include "src/Core/arch/SVE/TypeCasting.h"
-  #include "src/Core/arch/SVE/MathFunctions.h"
-#elif defined EIGEN_VECTORIZE_ZVECTOR
-  #include "src/Core/arch/ZVector/PacketMath.h"
-  #include "src/Core/arch/ZVector/MathFunctions.h"
-  #include "src/Core/arch/ZVector/Complex.h"
-#elif defined EIGEN_VECTORIZE_MSA
-  #include "src/Core/arch/MSA/PacketMath.h"
-  #include "src/Core/arch/MSA/MathFunctions.h"
-  #include "src/Core/arch/MSA/Complex.h"
-#endif
-
-#if defined EIGEN_VECTORIZE_GPU
-  #include "src/Core/arch/GPU/PacketMath.h"
-  #include "src/Core/arch/GPU/MathFunctions.h"
-  #include "src/Core/arch/GPU/TypeCasting.h"
-#endif
-
-#if defined(EIGEN_USE_SYCL)
-  #include "src/Core/arch/SYCL/SyclMemoryModel.h"
-  #include "src/Core/arch/SYCL/InteropHeaders.h"
-#if !defined(EIGEN_DONT_VECTORIZE_SYCL)
-  #include "src/Core/arch/SYCL/PacketMath.h"
-  #include "src/Core/arch/SYCL/MathFunctions.h"
-  #include "src/Core/arch/SYCL/TypeCasting.h"
-#endif
 #endif
 
 #include "src/Core/arch/Default/Settings.h"
-// This file provides generic implementations valid for scalar as well
-#include "src/Core/arch/Default/GenericPacketMathFunctions.h"
 
-#include "src/Core/functors/TernaryFunctors.h"
-#include "src/Core/functors/BinaryFunctors.h"
-#include "src/Core/functors/UnaryFunctors.h"
-#include "src/Core/functors/NullaryFunctors.h"
-#include "src/Core/functors/StlFunctors.h"
-#include "src/Core/functors/AssignmentFunctors.h"
-
-// Specialized functors to enable the processing of complex numbers
-// on CUDA devices
-#ifdef EIGEN_CUDACC
-#include "src/Core/arch/CUDA/Complex.h"
-#endif
-
-#include "src/Core/util/IndexedViewHelper.h"
-#include "src/Core/util/ReshapedHelper.h"
-#include "src/Core/ArithmeticSequence.h"
-#ifndef EIGEN_NO_IO
-  #include "src/Core/IO.h"
-#endif
+#include "src/Core/Functors.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/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/PlainObjectBase.h"
 #include "src/Core/Matrix.h"
 #include "src/Core/Array.h"
-#include "src/Core/CwiseTernaryOp.h"
 #include "src/Core/CwiseBinaryOp.h"
 #include "src/Core/CwiseUnaryOp.h"
 #include "src/Core/CwiseNullaryOp.h"
@@ -302,34 +300,32 @@ using std::ptrdiff_t;
 #include "src/Core/SelfCwiseBinaryOp.h"
 #include "src/Core/Dot.h"
 #include "src/Core/StableNorm.h"
-#include "src/Core/Stride.h"
 #include "src/Core/MapBase.h"
+#include "src/Core/Stride.h"
 #include "src/Core/Map.h"
-#include "src/Core/Ref.h"
 #include "src/Core/Block.h"
 #include "src/Core/VectorBlock.h"
-#include "src/Core/IndexedView.h"
-#include "src/Core/Reshaped.h"
+#include "src/Core/Ref.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/GeneralProduct.h"
-#include "src/Core/Solve.h"
-#include "src/Core/Inverse.h"
-#include "src/Core/SolverBase.h"
-#include "src/Core/PermutationMatrix.h"
-#include "src/Core/Transpositions.h"
 #include "src/Core/TriangularMatrix.h"
 #include "src/Core/SelfAdjointView.h"
 #include "src/Core/products/GeneralBlockPanelKernel.h"
 #include "src/Core/products/Parallelizer.h"
-#include "src/Core/ProductEvaluators.h"
+#include "src/Core/products/CoeffBasedProduct.h"
 #include "src/Core/products/GeneralMatrixVector.h"
 #include "src/Core/products/GeneralMatrixMatrix.h"
 #include "src/Core/SolveTriangular.h"
@@ -344,33 +340,25 @@ using std::ptrdiff_t;
 #include "src/Core/products/TriangularSolverVector.h"
 #include "src/Core/BandMatrix.h"
 #include "src/Core/CoreIterators.h"
-#include "src/Core/ConditionEstimator.h"
-
-#if defined(EIGEN_VECTORIZE_ALTIVEC) || defined(EIGEN_VECTORIZE_VSX)
-  #include "src/Core/arch/AltiVec/MatrixProduct.h"
-#elif defined EIGEN_VECTORIZE_NEON
-  #include "src/Core/arch/NEON/GeneralBlockPanelKernel.h"
-#endif
 
 #include "src/Core/BooleanRedux.h"
 #include "src/Core/Select.h"
 #include "src/Core/VectorwiseOp.h"
-#include "src/Core/PartialReduxEvaluator.h"
 #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/StlIterators.h"
 
 #ifdef EIGEN_USE_BLAS
-#include "src/Core/products/GeneralMatrixMatrix_BLAS.h"
-#include "src/Core/products/GeneralMatrixVector_BLAS.h"
-#include "src/Core/products/GeneralMatrixMatrixTriangular_BLAS.h"
-#include "src/Core/products/SelfadjointMatrixMatrix_BLAS.h"
-#include "src/Core/products/SelfadjointMatrixVector_BLAS.h"
-#include "src/Core/products/TriangularMatrixMatrix_BLAS.h"
-#include "src/Core/products/TriangularMatrixVector_BLAS.h"
-#include "src/Core/products/TriangularSolverMatrix_BLAS.h"
+#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
@@ -381,4 +369,8 @@ using std::ptrdiff_t;
 
 #include "src/Core/util/ReenableStupidWarnings.h"
 
+#ifdef EIGEN2_SUPPORT
+#include "Eigen2Support"
+#endif
+
 #endif // EIGEN_CORE_H

Eigen/Eigen

@@ -1,2 +1,2 @@
 #include "Dense"
-#include "Sparse"
+//#include "Sparse"

Eigen/Eigen2Support

@@ -0,0 +1,82 @@
+// This file is part of Eigen, a lightweight C++ template library
+// for linear algebra.
+//
+// Copyright (C) 2009 Gael Guennebaud <gael.guennebaud@inria.fr>
+//
+// This Source Code Form is subject to the terms of the Mozilla
+// Public License v. 2.0. If a copy of the MPL was not distributed
+// with this file, You can obtain one at http://mozilla.org/MPL/2.0/.
+
+#ifndef 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/DisableStupidWarnings.h"
+
+/** \ingroup Support_modules
+  * \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/Macros.h"
+#include "src/Eigen2Support/Memory.h"
+#include "src/Eigen2Support/Meta.h"
+#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"
+#include "src/Eigen2Support/MathFunctions.h"
+
+
+#include "src/Core/util/ReenableStupidWarnings.h"
+
+// Eigen2 used to include iostream
+#include<iostream>
+
+#define EIGEN_USING_MATRIX_TYPEDEFS_FOR_TYPE_AND_SIZE(TypeSuffix, SizeSuffix) \
+using Eigen::Matrix##SizeSuffix##TypeSuffix; \
+using Eigen::Vector##SizeSuffix##TypeSuffix; \
+using Eigen::RowVector##SizeSuffix##TypeSuffix;
+
+#define EIGEN_USING_MATRIX_TYPEDEFS_FOR_TYPE(TypeSuffix) \
+EIGEN_USING_MATRIX_TYPEDEFS_FOR_TYPE_AND_SIZE(TypeSuffix, 2) \
+EIGEN_USING_MATRIX_TYPEDEFS_FOR_TYPE_AND_SIZE(TypeSuffix, 3) \
+EIGEN_USING_MATRIX_TYPEDEFS_FOR_TYPE_AND_SIZE(TypeSuffix, 4) \
+EIGEN_USING_MATRIX_TYPEDEFS_FOR_TYPE_AND_SIZE(TypeSuffix, X) \
+
+#define EIGEN_USING_MATRIX_TYPEDEFS \
+EIGEN_USING_MATRIX_TYPEDEFS_FOR_TYPE(i) \
+EIGEN_USING_MATRIX_TYPEDEFS_FOR_TYPE(f) \
+EIGEN_USING_MATRIX_TYPEDEFS_FOR_TYPE(d) \
+EIGEN_USING_MATRIX_TYPEDEFS_FOR_TYPE(cf) \
+EIGEN_USING_MATRIX_TYPEDEFS_FOR_TYPE(cd)
+
+#define USING_PART_OF_NAMESPACE_EIGEN \
+EIGEN_USING_MATRIX_TYPEDEFS \
+using Eigen::Matrix; \
+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

@@ -1,23 +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_EIGENVALUES_MODULE_H
 #define EIGEN_EIGENVALUES_MODULE_H
 
 #include "Core"
 
+#include "src/Core/util/DisableStupidWarnings.h"
+
 #include "Cholesky"
 #include "Jacobi"
 #include "Householder"
 #include "LU"
 #include "Geometry"
 
-#include "src/Core/util/DisableStupidWarnings.h"
-
 /** \defgroup Eigenvalues_Module Eigenvalues module
   *
   *
@@ -32,7 +25,6 @@
   * \endcode
   */
 
-#include "src/misc/RealSvd2x2.h"
 #include "src/Eigenvalues/Tridiagonalization.h"
 #include "src/Eigenvalues/RealSchur.h"
 #include "src/Eigenvalues/EigenSolver.h"
@@ -45,16 +37,12 @@
 #include "src/Eigenvalues/GeneralizedEigenSolver.h"
 #include "src/Eigenvalues/MatrixBaseEigenvalues.h"
 #ifdef EIGEN_USE_LAPACKE
-#ifdef EIGEN_USE_MKL
-#include "mkl_lapacke.h"
-#else
-#include "src/misc/lapacke.h"
-#endif
-#include "src/Eigenvalues/RealSchur_LAPACKE.h"
-#include "src/Eigenvalues/ComplexSchur_LAPACKE.h"
-#include "src/Eigenvalues/SelfAdjointEigenSolver_LAPACKE.h"
+#include "src/Eigenvalues/RealSchur_MKL.h"
+#include "src/Eigenvalues/ComplexSchur_MKL.h"
+#include "src/Eigenvalues/SelfAdjointEigenSolver_MKL.h"
 #endif
 
 #include "src/Core/util/ReenableStupidWarnings.h"
 
 #endif // EIGEN_EIGENVALUES_MODULE_H
+/* vim: set filetype=cpp et sw=2 ts=2 ai: */

Eigen/Geometry

@@ -1,32 +1,29 @@
-// 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/DisableStupidWarnings.h"
+
 #include "SVD"
 #include "LU"
 #include <limits>
 
-#include "src/Core/util/DisableStupidWarnings.h"
+#ifndef M_PI
+#define M_PI 3.14159265358979323846
+#endif
 
 /** \defgroup Geometry_Module Geometry module
+  *
+  *
   *
   * This module provides support for:
   *  - fixed-size homogeneous transformations
   *  - translation, scaling, 2D and 3D rotations
-  *  - \link Quaternion quaternions \endlink
-  *  - cross products (\ref MatrixBase::cross, \ref MatrixBase::cross3)
-  *  - orthognal vector generation (\ref MatrixBase::unitOrthogonal)
-  *  - some linear components: \link ParametrizedLine parametrized-lines \endlink and \link Hyperplane hyperplanes \endlink
-  *  - \link AlignedBox axis aligned bounding boxes \endlink
-  *  - \link umeyama least-square transformation fitting \endlink
+  *  - quaternions
+  *  - \ref MatrixBase::cross() "cross product"
+  *  - \ref MatrixBase::unitOrthogonal() "orthognal vector generation"
+  *  - some linear components: parametrized-lines and hyperplanes
   *
   * \code
   * #include <Eigen/Geometry>
@@ -36,24 +33,31 @@
 #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/Transform.h"
-#include "src/Geometry/Translation.h"
-#include "src/Geometry/Scaling.h"
-#include "src/Geometry/Hyperplane.h"
-#include "src/Geometry/ParametrizedLine.h"
-#include "src/Geometry/AlignedBox.h"
-#include "src/Geometry/Umeyama.h"
+#if EIGEN2_SUPPORT_STAGE > STAGE20_RESOLVE_API_CONFLICTS
+  #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/Transform.h"
+  #include "src/Geometry/Translation.h"
+  #include "src/Geometry/Scaling.h"
+  #include "src/Geometry/Hyperplane.h"
+  #include "src/Geometry/ParametrizedLine.h"
+  #include "src/Geometry/AlignedBox.h"
+  #include "src/Geometry/Umeyama.h"
 
-// Use the SSE optimized version whenever possible.
-#if (defined EIGEN_VECTORIZE_SSE) || (defined EIGEN_VECTORIZE_NEON)
-#include "src/Geometry/arch/Geometry_SIMD.h"
+  #if defined EIGEN_VECTORIZE_SSE
+    #include "src/Geometry/arch/Geometry_SSE.h"
+  #endif
+#endif
+
+#ifdef EIGEN2_SUPPORT
+#include "src/Eigen2Support/Geometry/All.h"
 #endif
 
 #include "src/Core/util/ReenableStupidWarnings.h"
 
 #endif // EIGEN_GEOMETRY_MODULE_H
+/* vim: set filetype=cpp et sw=2 ts=2 ai: */
+

Eigen/Householder

@@ -1,10 +1,3 @@
-// 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
 
@@ -27,3 +20,4 @@
 #include "src/Core/util/ReenableStupidWarnings.h"
 
 #endif // EIGEN_HOUSEHOLDER_MODULE_H
+/* vim: set filetype=cpp et sw=2 ts=2 ai: */

Eigen/IterativeLinearSolvers

@@ -1,10 +1,3 @@
-// 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
 
@@ -19,29 +12,28 @@
   * 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
+  *  - DiagonalPreconditioner - also called JAcobi preconditioner, work very well on diagonal dominant matrices.
+  *  - IncompleteILUT - 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
+  * \code
+  * #include <Eigen/IterativeLinearSolvers>
+  * \endcode
   */
 
-#include "src/IterativeLinearSolvers/SolveWithGuess.h"
+#include "src/misc/Solve.h"
+#include "src/misc/SparseSolve.h"
+
 #include "src/IterativeLinearSolvers/IterativeSolverBase.h"
 #include "src/IterativeLinearSolvers/BasicPreconditioners.h"
 #include "src/IterativeLinearSolvers/ConjugateGradient.h"
-#include "src/IterativeLinearSolvers/LeastSquareConjugateGradient.h"
 #include "src/IterativeLinearSolvers/BiCGSTAB.h"
 #include "src/IterativeLinearSolvers/IncompleteLUT.h"
-#include "src/IterativeLinearSolvers/IncompleteCholesky.h"
 
 #include "src/Core/util/ReenableStupidWarnings.h"
 

Eigen/Jacobi

@@ -1,10 +1,3 @@
-// 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
 
@@ -29,4 +22,5 @@
 #include "src/Core/util/ReenableStupidWarnings.h"
 
 #endif // EIGEN_JACOBI_MODULE_H
+/* vim: set filetype=cpp et sw=2 ts=2 ai: */
 

Eigen/KLUSupport

@@ -1,41 +0,0 @@
-// 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_KLUSUPPORT_MODULE_H
-#define EIGEN_KLUSUPPORT_MODULE_H
-
-#include <Eigen/SparseCore>
-
-#include <Eigen/src/Core/util/DisableStupidWarnings.h>
-
-extern "C" {
-#include <btf.h>
-#include <klu.h>
-   }
-
-/** \ingroup Support_modules
-  * \defgroup KLUSupport_Module KLUSupport module
-  *
-  * This module provides an interface to the KLU library which is part of the <a href="http://www.suitesparse.com">suitesparse</a> package.
-  * It provides the following factorization class:
-  * - class KLU: a sparse LU factorization, well-suited for circuit simulation.
-  *
-  * \code
-  * #include <Eigen/KLUSupport>
-  * \endcode
-  *
-  * In order to use this module, the klu and btf headers must be accessible from the include paths, and your binary must be linked to the klu library and its dependencies.
-  * The dependencies depend on how umfpack has been compiled.
-  * For a cmake based project, you can use our FindKLU.cmake module to help you in this task.
-  *
-  */
-
-#include "src/KLUSupport/KLUSupport.h"
-
-#include <Eigen/src/Core/util/ReenableStupidWarnings.h>
-
-#endif // EIGEN_KLUSUPPORT_MODULE_H

Eigen/LU

@@ -1,10 +1,3 @@
-// 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
 
@@ -23,27 +16,26 @@
   * \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
-#ifdef EIGEN_USE_MKL
-#include "mkl_lapacke.h"
-#else
-#include "src/misc/lapacke.h"
-#endif
-#include "src/LU/PartialPivLU_LAPACKE.h"
+#include "src/LU/PartialPivLU_MKL.h"
 #endif
 #include "src/LU/Determinant.h"
-#include "src/LU/InverseImpl.h"
+#include "src/LU/Inverse.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) || defined EIGEN_VECTORIZE_NEON
-  #include "src/LU/arch/InverseSize4.h"
+#if defined EIGEN_VECTORIZE_SSE
+  #include "src/LU/arch/Inverse_SSE.h"
+#endif
+
+#ifdef EIGEN2_SUPPORT
+  #include "src/Eigen2Support/LU.h"
 #endif
 
 #include "src/Core/util/ReenableStupidWarnings.h"
 
 #endif // EIGEN_LU_MODULE_H
+/* vim: set filetype=cpp et sw=2 ts=2 ai: */

Eigen/LeastSquares

@@ -0,0 +1,32 @@
+#ifndef EIGEN_REGRESSION_MODULE_H
+#define EIGEN_REGRESSION_MODULE_H
+
+#ifndef EIGEN2_SUPPORT
+#error LeastSquares is only available in Eigen2 support mode (define EIGEN2_SUPPORT)
+#endif
+
+// exclude from normal eigen3-only documentation
+#ifdef EIGEN2_SUPPORT
+
+#include "Core"
+
+#include "src/Core/util/DisableStupidWarnings.h"
+
+#include "Eigenvalues"
+#include "Geometry"
+
+/** \defgroup LeastSquares_Module LeastSquares module
+  * This module provides linear regression and related features.
+  *
+  * \code
+  * #include <Eigen/LeastSquares>
+  * \endcode
+  */
+
+#include "src/Eigen2Support/LeastSquares.h"
+
+#include "src/Core/util/ReenableStupidWarnings.h"
+
+#endif // EIGEN2_SUPPORT
+
+#endif // EIGEN_REGRESSION_MODULE_H

Eigen/MetisSupport

@@ -1,10 +1,3 @@
-// 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
 

Eigen/OrderingMethods

@@ -1,10 +1,3 @@
-// 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
 
@@ -63,7 +56,10 @@
   * \endcode
   */
 
+#ifndef EIGEN_MPL2_ONLY
 #include "src/OrderingMethods/Amd.h"
+#endif
+
 #include "src/OrderingMethods/Ordering.h"
 #include "src/Core/util/ReenableStupidWarnings.h"
 

Eigen/PaStiXSupport

@@ -1,10 +1,3 @@
-// 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
 
@@ -12,6 +5,7 @@
 
 #include "src/Core/util/DisableStupidWarnings.h"
 
+#include <complex.h>
 extern "C" {
 #include <pastix_nompi.h>
 #include <pastix.h>
@@ -36,14 +30,17 @@ extern "C" {
   * \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.
-  * This wrapper resuires PaStiX version 5.x compiled without MPI support.
   * 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/misc/Solve.h"
+#include "src/misc/SparseSolve.h"
+
 #include "src/PaStiXSupport/PaStiXSupport.h"
 
+
 #include "src/Core/util/ReenableStupidWarnings.h"
 
 #endif // EIGEN_PASTIXSUPPORT_MODULE_H

Eigen/PardisoSupport

@@ -1,10 +1,3 @@
-// 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
 
@@ -14,6 +7,8 @@
 
 #include <mkl_pardiso.h>
 
+#include <unsupported/Eigen/SparseExtra>
+
 /** \ingroup Support_modules
   * \defgroup PardisoSupport_Module PardisoSupport module
   *

Eigen/QR

@@ -1,50 +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/DisableStupidWarnings.h"
+
 #include "Cholesky"
 #include "Jacobi"
 #include "Householder"
 
-#include "src/Core/util/DisableStupidWarnings.h"
-
 /** \defgroup QR_Module QR module
   *
   *
   *
   * This module provides various QR decompositions
   * This module also provides some MatrixBase methods, including:
-  *  - MatrixBase::householderQr()
-  *  - MatrixBase::colPivHouseholderQr()
-  *  - MatrixBase::fullPivHouseholderQr()
+  *  - MatrixBase::qr(),
   *
   * \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"
-#include "src/QR/CompleteOrthogonalDecomposition.h"
 #ifdef EIGEN_USE_LAPACKE
-#ifdef EIGEN_USE_MKL
-#include "mkl_lapacke.h"
-#else
-#include "src/misc/lapacke.h"
+#include "src/QR/HouseholderQR_MKL.h"
+#include "src/QR/ColPivHouseholderQR_MKL.h"
 #endif
-#include "src/QR/HouseholderQR_LAPACKE.h"
-#include "src/QR/ColPivHouseholderQR_LAPACKE.h"
+
+#ifdef EIGEN2_SUPPORT
+#include "src/Eigen2Support/QR.h"
 #endif
 
 #include "src/Core/util/ReenableStupidWarnings.h"
 
+#ifdef EIGEN2_SUPPORT
+#include "Eigenvalues"
+#endif
+
 #endif // EIGEN_QR_MODULE_H
+/* vim: set filetype=cpp et sw=2 ts=2 ai: */

Eigen/QtAlignedMalloc

@@ -1,9 +1,3 @@
-// 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
@@ -14,7 +8,7 @@
 
 #include "src/Core/util/DisableStupidWarnings.h"
 
-void *qMalloc(std::size_t size)
+void *qMalloc(size_t size)
 {
   return Eigen::internal::aligned_malloc(size);
 }
@@ -24,10 +18,10 @@ void qFree(void *ptr)
   Eigen::internal::aligned_free(ptr);
 }
 
-void *qRealloc(void *ptr, std::size_t size)
+void *qRealloc(void *ptr, size_t size)
 {
   void* newPtr = Eigen::internal::aligned_malloc(size);
-  std::memcpy(newPtr, ptr, size);
+  memcpy(newPtr, ptr, size);
   Eigen::internal::aligned_free(ptr);
   return newPtr;
 }
@@ -37,3 +31,4 @@ void *qRealloc(void *ptr, std::size_t size)
 #endif
 
 #endif // EIGEN_QTMALLOC_MODULE_H
+/* vim: set filetype=cpp et sw=2 ts=2 ai: */

Eigen/SPQRSupport

@@ -1,10 +1,3 @@
-// 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
 
@@ -17,7 +10,7 @@
 /** \ingroup Support_modules
   * \defgroup SPQRSupport_Module SuiteSparseQR module
   * 
-  * This module provides an interface to the SPQR library, which is part of the <a href="http://www.suitesparse.com">suitesparse</a> package.
+  * 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>
@@ -28,6 +21,8 @@
   *
   */
 
+#include "src/misc/Solve.h"
+#include "src/misc/SparseSolve.h"
 #include "src/CholmodSupport/CholmodSupport.h"
 #include "src/SPQRSupport/SuiteSparseQRSupport.h"
 

Eigen/SVD

@@ -1,10 +1,3 @@
-// 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
 
@@ -19,32 +12,26 @@
   *
   *
   * 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:
+  * This decomposition is accessible via the following MatrixBase method:
   *  - MatrixBase::jacobiSvd()
-  *  - MatrixBase::bdcSvd()
   *
   * \code
   * #include <Eigen/SVD>
   * \endcode
   */
 
-#include "src/misc/RealSvd2x2.h"
-#include "src/SVD/UpperBidiagonalization.h"
-#include "src/SVD/SVDBase.h"
+#include "src/misc/Solve.h"
 #include "src/SVD/JacobiSVD.h"
-#include "src/SVD/BDCSVD.h"
 #if defined(EIGEN_USE_LAPACKE) && !defined(EIGEN_USE_LAPACKE_STRICT)
-#ifdef EIGEN_USE_MKL
-#include "mkl_lapacke.h"
-#else
-#include "src/misc/lapacke.h"
+#include "src/SVD/JacobiSVD_MKL.h"
 #endif
-#include "src/SVD/JacobiSVD_LAPACKE.h"
+#include "src/SVD/UpperBidiagonalization.h"
+
+#ifdef EIGEN2_SUPPORT
+#include "src/Eigen2Support/SVD.h"
 #endif
 
 #include "src/Core/util/ReenableStupidWarnings.h"
 
 #endif // EIGEN_SVD_MODULE_H
+/* vim: set filetype=cpp et sw=2 ts=2 ai: */

Eigen/Sparse

@@ -1,10 +1,3 @@
-// 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
 
@@ -18,9 +11,9 @@
   * - \ref SparseQR_Module
   * - \ref IterativeLinearSolvers_Module
   *
-    \code
-    #include <Eigen/Sparse>
-    \endcode
+  * \code
+  * #include <Eigen/Sparse>
+  * \endcode
   */
 
 #include "SparseCore"

Eigen/SparseCholesky

@@ -30,8 +30,18 @@
   * \endcode
   */
 
+#ifdef EIGEN_MPL2_ONLY
+#error The SparseCholesky module has nothing to offer in MPL2 only mode
+#endif
+
+#include "src/misc/Solve.h"
+#include "src/misc/SparseSolve.h"
 #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

@@ -1,10 +1,3 @@
-// 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
 
@@ -21,7 +14,7 @@
 /** 
   * \defgroup SparseCore_Module SparseCore module
   *
-  * This module provides a sparse matrix representation, and basic associated matrix manipulations
+  * This module provides a sparse matrix representation, and basic associatd matrix manipulations
   * and operations.
   *
   * See the \ref TutorialSparse "Sparse tutorial"
@@ -33,35 +26,37 @@
   * This module depends on: Core.
   */
 
+namespace Eigen {
+
+/** The type used to identify a general sparse storage. */
+struct Sparse {};
+
+}
+
 #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/SparseBlock.h"
+#include "src/SparseCore/SparseTranspose.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/SparsePermutation.h"
 #include "src/SparseCore/SparseRedux.h"
-#include "src/SparseCore/SparseView.h"
-#include "src/SparseCore/SparseDiagonalProduct.h"
+#include "src/SparseCore/SparseFuzzy.h"
 #include "src/SparseCore/ConservativeSparseSparseProduct.h"
 #include "src/SparseCore/SparseSparseProductWithPruning.h"
 #include "src/SparseCore/SparseProduct.h"
 #include "src/SparseCore/SparseDenseProduct.h"
-#include "src/SparseCore/SparseSelfAdjointView.h"
+#include "src/SparseCore/SparseDiagonalProduct.h"
 #include "src/SparseCore/SparseTriangularView.h"
+#include "src/SparseCore/SparseSelfAdjointView.h"
 #include "src/SparseCore/TriangularSolver.h"
-#include "src/SparseCore/SparsePermutation.h"
-#include "src/SparseCore/SparseFuzzy.h"
-#include "src/SparseCore/SparseSolverBase.h"
+#include "src/SparseCore/SparseView.h"
 
 #include "src/Core/util/ReenableStupidWarnings.h"
 

Eigen/SparseLU

@@ -20,11 +20,12 @@
   * Please, see the documentation of the SparseLU class for more details.
   */
 
+#include "src/misc/Solve.h"
+#include "src/misc/SparseSolve.h"
+
 // Ordering interface
 #include "OrderingMethods"
 
-#include "src/Core/util/DisableStupidWarnings.h"
-
 #include "src/SparseLU/SparseLU_gemm_kernel.h"
 
 #include "src/SparseLU/SparseLU_Structs.h"
@@ -45,6 +46,4 @@
 #include "src/SparseLU/SparseLU_Utils.h"
 #include "src/SparseLU/SparseLU.h"
 
-#include "src/Core/util/ReenableStupidWarnings.h"
-
 #endif // EIGEN_SPARSELU_MODULE_H

Eigen/SparseQR

@@ -1,10 +1,3 @@
-// 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
 
@@ -28,6 +21,10 @@
   * 
   */
 
+#include "src/misc/Solve.h"
+#include "src/misc/SparseSolve.h"
+
+#include "OrderingMethods"
 #include "src/SparseCore/SparseColEtree.h"
 #include "src/SparseQR/SparseQR.h"
 

Eigen/StdDeque

@@ -14,7 +14,7 @@
 #include "Core"
 #include <deque>
 
-#if EIGEN_COMP_MSVC && EIGEN_OS_WIN64 && (EIGEN_MAX_STATIC_ALIGN_BYTES<=16) /* MSVC auto aligns up to 16 bytes in 64 bit builds */
+#if (defined(_MSC_VER) && defined(_WIN64)) /* MSVC auto aligns in 64 bit builds */
 
 #define EIGEN_DEFINE_STL_DEQUE_SPECIALIZATION(...)
 

Eigen/StdList

@@ -13,7 +13,7 @@
 #include "Core"
 #include <list>
 
-#if EIGEN_COMP_MSVC && EIGEN_OS_WIN64 && (EIGEN_MAX_STATIC_ALIGN_BYTES<=16) /* MSVC auto aligns up to 16 bytes in 64 bit builds */
+#if (defined(_MSC_VER) && defined(_WIN64)) /* MSVC auto aligns in 64 bit builds */    
 
 #define EIGEN_DEFINE_STL_LIST_SPECIALIZATION(...)
 

Eigen/StdVector

@@ -14,7 +14,7 @@
 #include "Core"
 #include <vector>
 
-#if EIGEN_COMP_MSVC && EIGEN_OS_WIN64 && (EIGEN_MAX_STATIC_ALIGN_BYTES<=16) /* MSVC auto aligns up to 16 bytes in 64 bit builds */
+#if (defined(_MSC_VER) && defined(_WIN64)) /* MSVC auto aligns in 64 bit builds */
 
 #define EIGEN_DEFINE_STL_VECTOR_SPECIALIZATION(...)
 

Eigen/SuperLUSupport

@@ -1,10 +1,3 @@
-// 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
 
@@ -43,8 +36,6 @@ namespace Eigen { struct SluMatrix; }
   * - class SuperLU: a supernodal sequential LU factorization.
   * - class SuperILU: a supernodal sequential incomplete LU factorization (to be used as a preconditioner for iterative methods).
   *
-  * \warning This wrapper requires at least versions 4.0 of SuperLU. The 3.x versions are not supported.
-  *
   * \warning When including this module, you have to use SUPERLU_EMPTY instead of EMPTY which is no longer defined because it is too polluting.
   *
   * \code
@@ -57,8 +48,12 @@ namespace Eigen { struct SluMatrix; }
   *
   */
 
+#include "src/misc/Solve.h"
+#include "src/misc/SparseSolve.h"
+
 #include "src/SuperLUSupport/SuperLUSupport.h"
 
+
 #include "src/Core/util/ReenableStupidWarnings.h"
 
 #endif // EIGEN_SUPERLUSUPPORT_MODULE_H

Eigen/UmfPackSupport

@@ -1,10 +1,3 @@
-// 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
 
@@ -19,7 +12,7 @@ extern "C" {
 /** \ingroup Support_modules
   * \defgroup UmfPackSupport_Module UmfPackSupport module
   *
-  * This module provides an interface to the UmfPack library which is part of the <a href="http://www.suitesparse.com">suitesparse</a> package.
+  * 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.
   *
@@ -33,6 +26,9 @@ extern "C" {
   *
   */
 
+#include "src/misc/Solve.h"
+#include "src/misc/SparseSolve.h"
+
 #include "src/UmfPackSupport/UmfPackSupport.h"
 
 #include "src/Core/util/ReenableStupidWarnings.h"

Eigen/src/CMakeLists.txt

@@ -0,0 +1,7 @@
+file(GLOB Eigen_src_subdirectories "*")
+escape_string_as_regex(ESCAPED_CMAKE_CURRENT_SOURCE_DIR "${CMAKE_CURRENT_SOURCE_DIR}")
+foreach(f ${Eigen_src_subdirectories})
+  if(NOT f MATCHES "\\.txt" AND NOT f MATCHES "${ESCAPED_CMAKE_CURRENT_SOURCE_DIR}/[.].+" )
+    add_subdirectory(${f})
+  endif()
+endforeach()

Eigen/src/Cholesky/CMakeLists.txt

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

Eigen/src/Cholesky/LDLT.h

@@ -13,22 +13,10 @@
 #ifndef EIGEN_LDLT_H
 #define EIGEN_LDLT_H
 
-namespace Eigen {
+namespace Eigen { 
 
 namespace internal {
-  template<typename _MatrixType, int _UpLo> struct traits<LDLT<_MatrixType, _UpLo> >
-   : traits<_MatrixType>
-  {
-    typedef MatrixXpr XprKind;
-    typedef SolverStorage StorageKind;
-    typedef int StorageIndex;
-    enum { Flags = 0 };
-  };
-
-  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 };
+template<typename MatrixType, int UpLo> struct LDLT_Traits;
 }
 
 /** \ingroup Cholesky_Module
@@ -37,40 +25,39 @@ namespace internal {
   *
   * \brief Robust Cholesky decomposition of a matrix with pivoting
   *
-  * \tparam _MatrixType the type of the matrix of which to compute the LDL^T Cholesky decomposition
-  * \tparam _UpLo the triangular part that will be used for the decompositon: Lower (default) or Upper.
+  * \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
   * is lower triangular with a unit diagonal and D is a diagonal matrix.
   *
-  * The decomposition uses pivoting to ensure stability, so that D will have
+  * The decomposition uses pivoting to ensure stability, so that L will have
   * zeros in the bottom right rank(A) - n submatrix. Avoiding the square root
   * on D also stabilizes the computation.
   *
   * 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.
   *
-  * This class supports the \link InplaceDecomposition inplace decomposition \endlink mechanism.
-  *
-  * \sa MatrixBase::ldlt(), SelfAdjointView::ldlt(), class LLT
+  * \sa MatrixBase::ldlt(), class LLT
   */
 template<typename _MatrixType, int _UpLo> class LDLT
-        : public SolverBase<LDLT<_MatrixType, _UpLo> >
 {
   public:
     typedef _MatrixType MatrixType;
-    typedef SolverBase<LDLT> Base;
-    friend class SolverBase<LDLT>;
-
-    EIGEN_GENERIC_PUBLIC_INTERFACE(LDLT)
     enum {
+      RowsAtCompileTime = MatrixType::RowsAtCompileTime,
+      ColsAtCompileTime = MatrixType::ColsAtCompileTime,
+      Options = MatrixType::Options & ~RowMajorBit, // these are the options for the TmpMatrixType, we need a ColMajor matrix here!
       MaxRowsAtCompileTime = MatrixType::MaxRowsAtCompileTime,
       MaxColsAtCompileTime = MatrixType::MaxColsAtCompileTime,
       UpLo = _UpLo
     };
-    typedef Matrix<Scalar, RowsAtCompileTime, 1, 0, MaxRowsAtCompileTime, 1> TmpMatrixType;
+    typedef typename MatrixType::Scalar Scalar;
+    typedef typename NumTraits<typename MatrixType::Scalar>::Real RealScalar;
+    typedef typename MatrixType::Index Index;
+    typedef Matrix<Scalar, RowsAtCompileTime, 1, Options, MaxRowsAtCompileTime, 1> TmpMatrixType;
 
     typedef Transpositions<RowsAtCompileTime, MaxRowsAtCompileTime> TranspositionType;
     typedef PermutationMatrix<RowsAtCompileTime, MaxRowsAtCompileTime> PermutationType;
@@ -82,12 +69,7 @@ template<typename _MatrixType, int _UpLo> class LDLT
       * The default constructor is useful in cases in which the user intends to
       * perform decompositions via LDLT::compute(const MatrixType&).
       */
-    LDLT()
-      : m_matrix(),
-        m_transpositions(),
-        m_sign(internal::ZeroSign),
-        m_isInitialized(false)
-    {}
+    LDLT() : m_matrix(), m_transpositions(), m_isInitialized(false) {}
 
     /** \brief Default Constructor with memory preallocation
       *
@@ -95,46 +77,25 @@ template<typename _MatrixType, int _UpLo> class LDLT
       * according to the specified problem \a size.
       * \sa LDLT()
       */
-    explicit LDLT(Index size)
+    LDLT(Index size)
       : m_matrix(size, size),
         m_transpositions(size),
         m_temporary(size),
-        m_sign(internal::ZeroSign),
         m_isInitialized(false)
     {}
 
     /** \brief Constructor with decomposition
       *
       * This calculates the decomposition for the input \a matrix.
-      *
       * \sa LDLT(Index size)
       */
-    template<typename InputType>
-    explicit LDLT(const EigenBase<InputType>& matrix)
+    LDLT(const MatrixType& matrix)
       : m_matrix(matrix.rows(), matrix.cols()),
         m_transpositions(matrix.rows()),
         m_temporary(matrix.rows()),
-        m_sign(internal::ZeroSign),
         m_isInitialized(false)
     {
-      compute(matrix.derived());
-    }
-
-    /** \brief Constructs a LDLT factorization from a given matrix
-      *
-      * This overloaded constructor is provided for \link InplaceDecomposition inplace decomposition \endlink when \c MatrixType is a Eigen::Ref.
-      *
-      * \sa LDLT(const EigenBase&)
-      */
-    template<typename InputType>
-    explicit LDLT(EigenBase<InputType>& matrix)
-      : m_matrix(matrix.derived()),
-        m_transpositions(matrix.rows()),
-        m_temporary(matrix.rows()),
-        m_sign(internal::ZeroSign),
-        m_isInitialized(false)
-    {
-      compute(matrix.derived());
+      compute(matrix);
     }
 
     /** Clear any existing decomposition
@@ -178,17 +139,23 @@ template<typename _MatrixType, int _UpLo> class LDLT
     inline bool isPositive() const
     {
       eigen_assert(m_isInitialized && "LDLT is not initialized.");
-      return m_sign == internal::PositiveSemiDef || m_sign == internal::ZeroSign;
+      return m_sign == 1;
     }
+    
+    #ifdef EIGEN2_SUPPORT
+    inline bool isPositiveDefinite() const
+    {
+      return isPositive();
+    }
+    #endif
 
     /** \returns true if the matrix is negative (semidefinite) */
     inline bool isNegative(void) const
     {
       eigen_assert(m_isInitialized && "LDLT is not initialized.");
-      return m_sign == internal::NegativeSemiDef || m_sign == internal::ZeroSign;
+      return m_sign == -1;
     }
 
-    #ifdef EIGEN_PARSED_BY_DOXYGEN
     /** \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> .
@@ -196,33 +163,37 @@ template<typename _MatrixType, int _UpLo> class LDLT
       * \note_about_checking_solutions
       *
       * More precisely, this method solves \f$ A x = b \f$ using the decomposition \f$ A = P^T L D L^* P \f$
-      * by solving the systems \f$ P^T y_1 = b \f$, \f$ L y_2 = y_1 \f$, \f$ D y_3 = y_2 \f$,
+      * by solving the systems \f$ 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$ if \f$ A \f$ is singular.
+      * computes the least-square solution of \f$ A x = b \f$ is \f$ A \f$ is singular.
       *
-      * \sa MatrixBase::ldlt(), SelfAdjointView::ldlt()
+      * \sa MatrixBase::ldlt()
       */
     template<typename Rhs>
-    inline const Solve<LDLT, Rhs>
-    solve(const MatrixBase<Rhs>& b) const;
+    inline const internal::solve_retval<LDLT, Rhs>
+    solve(const MatrixBase<Rhs>& b) const
+    {
+      eigen_assert(m_isInitialized && "LDLT is not initialized.");
+      eigen_assert(m_matrix.rows()==b.rows()
+                && "LDLT::solve(): invalid number of rows of the right hand side matrix b");
+      return internal::solve_retval<LDLT, Rhs>(*this, b.derived());
+    }
+
+    #ifdef EIGEN2_SUPPORT
+    template<typename OtherDerived, typename ResultType>
+    bool solve(const MatrixBase<OtherDerived>& b, ResultType *result) const
+    {
+      *result = this->solve(b);
+      return true;
+    }
     #endif
 
     template<typename Derived>
     bool solveInPlace(MatrixBase<Derived> &bAndX) const;
 
-    template<typename InputType>
-    LDLT& compute(const EigenBase<InputType>& matrix);
-
-    /** \returns an estimate of the reciprocal condition number of the matrix of
-     *  which \c *this is the LDLT decomposition.
-     */
-    RealScalar rcond() const
-    {
-      eigen_assert(m_isInitialized && "LDLT is not initialized.");
-      return internal::rcond_estimate_helper(m_l1_norm, *this);
-    }
+    LDLT& compute(const MatrixType& matrix);
 
     template <typename Derived>
     LDLT& rankUpdate(const MatrixBase<Derived>& w, const RealScalar& alpha=1);
@@ -239,42 +210,22 @@ template<typename _MatrixType, int _UpLo> class LDLT
 
     MatrixType reconstructedMatrix() const;
 
-    /** \returns the adjoint of \c *this, that is, a const reference to the decomposition itself as the underlying matrix is self-adjoint.
-      *
-      * This method is provided for compatibility with other matrix decompositions, thus enabling generic code such as:
-      * \code x = decomposition.adjoint().solve(b) \endcode
-      */
-    const LDLT& adjoint() const { return *this; };
-
-    EIGEN_DEVICE_FUNC inline EIGEN_CONSTEXPR Index rows() const EIGEN_NOEXCEPT { return m_matrix.rows(); }
-    EIGEN_DEVICE_FUNC inline EIGEN_CONSTEXPR Index cols() const EIGEN_NOEXCEPT { return m_matrix.cols(); }
+    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 successful,
-      *          \c NumericalIssue if the factorization failed because of a zero pivot.
+      * \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 m_info;
+      return Success;
     }
 
-    #ifndef EIGEN_PARSED_BY_DOXYGEN
-    template<typename RhsType, typename DstType>
-    void _solve_impl(const RhsType &rhs, DstType &dst) const;
-
-    template<bool Conjugate, typename RhsType, typename DstType>
-    void _solve_impl_transposed(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.
       * The strict upper part is used during the decomposition, the strict lower
@@ -282,12 +233,10 @@ template<typename _MatrixType, int _UpLo> class LDLT
       * is not stored), and the diagonal entries correspond to D.
       */
     MatrixType m_matrix;
-    RealScalar m_l1_norm;
     TranspositionType m_transpositions;
     TmpMatrixType m_temporary;
-    internal::SignMatrix m_sign;
+    int m_sign;
     bool m_isInitialized;
-    ComputationInfo m_info;
 };
 
 namespace internal {
@@ -297,35 +246,49 @@ 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, SignMatrix& sign)
+  static bool unblocked(MatrixType& mat, TranspositionType& transpositions, Workspace& temp, int* sign=0)
   {
     using std::abs;
     typedef typename MatrixType::Scalar Scalar;
     typedef typename MatrixType::RealScalar RealScalar;
-    typedef typename TranspositionType::StorageIndex IndexType;
+    typedef typename MatrixType::Index Index;
     eigen_assert(mat.rows()==mat.cols());
     const Index size = mat.rows();
-    bool found_zero_pivot = false;
-    bool ret = true;
 
     if (size <= 1)
     {
       transpositions.setIdentity();
-      if(size==0) sign = ZeroSign;
-      else if (numext::real(mat.coeff(0,0)) > static_cast<RealScalar>(0) ) sign = PositiveSemiDef;
-      else if (numext::real(mat.coeff(0,0)) < static_cast<RealScalar>(0)) sign = NegativeSemiDef;
-      else sign = ZeroSign;
+      if(sign)
+        *sign = numext::real(mat.coeff(0,0))>0 ? 1:-1;
       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;
-      mat.diagonal().tail(size-k).cwiseAbs().maxCoeff(&index_of_biggest_in_corner);
+      biggest_in_corner = mat.diagonal().tail(size-k).cwiseAbs().maxCoeff(&index_of_biggest_in_corner);
       index_of_biggest_in_corner += k;
 
-      transpositions.coeffRef(k) = IndexType(index_of_biggest_in_corner);
+      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 = abs(NumTraits<Scalar>::epsilon() * biggest_in_corner);
+      }
+
+      // 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;
+        if(sign) *sign = 0;
+        break;
+      }
+
+      transpositions.coeffRef(k) = index_of_biggest_in_corner;
       if(k != index_of_biggest_in_corner)
       {
         // apply the transposition while taking care to consider only
@@ -334,7 +297,7 @@ template<> struct 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(Index i=k+1;i<index_of_biggest_in_corner;++i)
+        for(int i=k+1;i<index_of_biggest_in_corner;++i)
         {
           Scalar tmp = mat.coeffRef(i,k);
           mat.coeffRef(i,k) = numext::conj(mat.coeffRef(index_of_biggest_in_corner,i));
@@ -355,51 +318,26 @@ template<> struct ldlt_inplace<Lower>
 
       if(k>0)
       {
-        temp.head(k) = mat.diagonal().real().head(k).asDiagonal() * A10.adjoint();
+        temp.head(k) = mat.diagonal().head(k).asDiagonal() * A10.adjoint();
         mat.coeffRef(k,k) -= (A10 * temp.head(k)).value();
         if(rs>0)
           A21.noalias() -= A20 * temp.head(k);
       }
-
-      // In some previous versions of Eigen (e.g., 3.2.1), the scaling was omitted if the pivot
-      // was smaller than the cutoff value. However, since LDLT is not rank-revealing
-      // we should only make sure that we do not introduce INF or NaN values.
-      // Remark that LAPACK also uses 0 as the cutoff value.
-      RealScalar realAkk = numext::real(mat.coeffRef(k,k));
-      bool pivot_is_valid = (abs(realAkk) > RealScalar(0));
-
-      if(k==0 && !pivot_is_valid)
+      if((rs>0) && (abs(mat.coeffRef(k,k)) > cutoff))
+        A21 /= mat.coeffRef(k,k);
+      
+      if(sign)
       {
-        // The entire diagonal is zero, there is nothing more to do
-        // except filling the transpositions, and checking whether the matrix is zero.
-        sign = ZeroSign;
-        for(Index j = 0; j<size; ++j)
-        {
-          transpositions.coeffRef(j) = IndexType(j);
-          ret = ret && (mat.col(j).tail(size-j-1).array()==Scalar(0)).all();
-        }
-        return ret;
-      }
-
-      if((rs>0) && pivot_is_valid)
-        A21 /= realAkk;
-      else if(rs>0)
-        ret = ret && (A21.array()==Scalar(0)).all();
-
-      if(found_zero_pivot && pivot_is_valid) ret = false; // factorization failed
-      else if(!pivot_is_valid) found_zero_pivot = true;
-
-      if (sign == PositiveSemiDef) {
-        if (realAkk < static_cast<RealScalar>(0)) sign = Indefinite;
-      } else if (sign == NegativeSemiDef) {
-        if (realAkk > static_cast<RealScalar>(0)) sign = Indefinite;
-      } else if (sign == ZeroSign) {
-        if (realAkk > static_cast<RealScalar>(0)) sign = PositiveSemiDef;
-        else if (realAkk < static_cast<RealScalar>(0)) sign = NegativeSemiDef;
+        // LDLT is not guaranteed to work for indefinite matrices, but let's try to get the sign right
+        int newSign = numext::real(mat.diagonal().coeff(index_of_biggest_in_corner)) > 0;
+        if(k == 0)
+          *sign = newSign;
+        else if(*sign != newSign)
+          *sign = 0;
       }
     }
 
-    return ret;
+    return true;
   }
 
   // Reference for the algorithm: Davis and Hager, "Multiple Rank
@@ -415,6 +353,7 @@ template<> struct ldlt_inplace<Lower>
     using numext::isfinite;
     typedef typename MatrixType::Scalar Scalar;
     typedef typename MatrixType::RealScalar RealScalar;
+    typedef typename MatrixType::Index Index;
 
     const Index size = mat.rows();
     eigen_assert(mat.cols() == size && w.size()==size);
@@ -460,7 +399,7 @@ template<> struct ldlt_inplace<Lower>
 template<> struct ldlt_inplace<Upper>
 {
   template<typename MatrixType, typename TranspositionType, typename Workspace>
-  static EIGEN_STRONG_INLINE bool unblocked(MatrixType& mat, TranspositionType& transpositions, Workspace& temp, SignMatrix& sign)
+  static EIGEN_STRONG_INLINE bool unblocked(MatrixType& mat, TranspositionType& transpositions, Workspace& temp, int* sign=0)
   {
     Transpose<MatrixType> matt(mat);
     return ldlt_inplace<Lower>::unblocked(matt, transpositions, temp, sign);
@@ -478,16 +417,16 @@ template<typename MatrixType> struct LDLT_Traits<MatrixType,Lower>
 {
   typedef const TriangularView<const MatrixType, UnitLower> MatrixL;
   typedef const TriangularView<const typename MatrixType::AdjointReturnType, UnitUpper> MatrixU;
-  static inline MatrixL getL(const MatrixType& m) { return MatrixL(m); }
-  static inline MatrixU getU(const MatrixType& m) { return MatrixU(m.adjoint()); }
+  static inline MatrixL getL(const MatrixType& m) { return m; }
+  static inline MatrixU getU(const MatrixType& m) { return m.adjoint(); }
 };
 
 template<typename MatrixType> struct LDLT_Traits<MatrixType,Upper>
 {
   typedef const TriangularView<const typename MatrixType::AdjointReturnType, UnitLower> MatrixL;
   typedef const TriangularView<const MatrixType, UnitUpper> MatrixU;
-  static inline MatrixL getL(const MatrixType& m) { return MatrixL(m.adjoint()); }
-  static inline MatrixU getU(const MatrixType& m) { return MatrixU(m); }
+  static inline MatrixL getL(const MatrixType& m) { return m.adjoint(); }
+  static inline MatrixU getU(const MatrixType& m) { return m; }
 };
 
 } // end namespace internal
@@ -495,35 +434,18 @@ template<typename MatrixType> struct LDLT_Traits<MatrixType,Upper>
 /** Compute / recompute the LDLT decomposition A = L D L^* = U^* D U of \a matrix
   */
 template<typename MatrixType, int _UpLo>
-template<typename InputType>
-LDLT<MatrixType,_UpLo>& LDLT<MatrixType,_UpLo>::compute(const EigenBase<InputType>& a)
+LDLT<MatrixType,_UpLo>& LDLT<MatrixType,_UpLo>::compute(const MatrixType& a)
 {
-  check_template_parameters();
-
   eigen_assert(a.rows()==a.cols());
   const Index size = a.rows();
 
-  m_matrix = a.derived();
-
-  // Compute matrix L1 norm = max abs column sum.
-  m_l1_norm = RealScalar(0);
-  // TODO move this code to SelfAdjointView
-  for (Index col = 0; col < size; ++col) {
-    RealScalar abs_col_sum;
-    if (_UpLo == Lower)
-      abs_col_sum = m_matrix.col(col).tail(size - col).template lpNorm<1>() + m_matrix.row(col).head(col).template lpNorm<1>();
-    else
-      abs_col_sum = m_matrix.col(col).head(col).template lpNorm<1>() + m_matrix.row(col).tail(size - col).template lpNorm<1>();
-    if (abs_col_sum > m_l1_norm)
-      m_l1_norm = abs_col_sum;
-  }
+  m_matrix = a;
 
   m_transpositions.resize(size);
   m_isInitialized = false;
   m_temporary.resize(size);
-  m_sign = internal::ZeroSign;
 
-  m_info = internal::ldlt_inplace<UpLo>::unblocked(m_matrix, m_transpositions, m_temporary, m_sign) ? Success : NumericalIssue;
+  internal::ldlt_inplace<UpLo>::unblocked(m_matrix, m_transpositions, m_temporary, &m_sign);
 
   m_isInitialized = true;
   return *this;
@@ -536,23 +458,22 @@ LDLT<MatrixType,_UpLo>& LDLT<MatrixType,_UpLo>::compute(const EigenBase<InputTyp
   */
 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)
+LDLT<MatrixType,_UpLo>& LDLT<MatrixType,_UpLo>::rankUpdate(const MatrixBase<Derived>& w, const typename NumTraits<typename MatrixType::Scalar>::Real& sigma)
 {
-  typedef typename TranspositionType::StorageIndex IndexType;
   const Index size = w.rows();
   if (m_isInitialized)
   {
     eigen_assert(m_matrix.rows()==size);
   }
   else
-  {
+  {    
     m_matrix.resize(size,size);
     m_matrix.setZero();
     m_transpositions.resize(size);
     for (Index i = 0; i < size; i++)
-      m_transpositions.coeffRef(i) = IndexType(i);
+      m_transpositions.coeffRef(i) = i;
     m_temporary.resize(size);
-    m_sign = sigma>=0 ? internal::PositiveSemiDef : internal::NegativeSemiDef;
+    m_sign = sigma>=0 ? 1 : -1;
     m_isInitialized = true;
   }
 
@@ -561,55 +482,48 @@ LDLT<MatrixType,_UpLo>& LDLT<MatrixType,_UpLo>::rankUpdate(const MatrixBase<Deri
   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
+namespace internal {
+template<typename _MatrixType, int _UpLo, typename Rhs>
+struct solve_retval<LDLT<_MatrixType,_UpLo>, Rhs>
+  : solve_retval_base<LDLT<_MatrixType,_UpLo>, Rhs>
 {
-  _solve_impl_transposed<true>(rhs, dst);
-}
+  typedef LDLT<_MatrixType,_UpLo> LDLTType;
+  EIGEN_MAKE_SOLVE_HELPERS(LDLTType,Rhs)
 
-template<typename _MatrixType,int _UpLo>
-template<bool Conjugate, typename RhsType, typename DstType>
-void LDLT<_MatrixType,_UpLo>::_solve_impl_transposed(const RhsType &rhs, DstType &dst) const
-{
-  // dst = P b
-  dst = m_transpositions * rhs;
-
-  // dst = L^-1 (P b)
-  // dst = L^-*T (P b)
-  matrixL().template conjugateIf<!Conjugate>().solveInPlace(dst);
-
-  // dst = D^-* (L^-1 P b)
-  // dst = D^-1 (L^-*T P b)
-  // more precisely, use pseudo-inverse of D (see bug 241)
-  using std::abs;
-  const typename Diagonal<const MatrixType>::RealReturnType vecD(vectorD());
-  // In some previous versions, tolerance was set to the max of 1/highest (or rather numeric_limits::min())
-  // and the maximal diagonal entry * epsilon as motivated by LAPACK's xGELSS:
-  // RealScalar tolerance = numext::maxi(vecD.array().abs().maxCoeff() * NumTraits<RealScalar>::epsilon(),RealScalar(1) / NumTraits<RealScalar>::highest());
-  // However, LDLT is not rank revealing, and so adjusting the tolerance wrt to the highest
-  // diagonal element is not well justified and leads to numerical issues in some cases.
-  // Moreover, Lapack's xSYTRS routines use 0 for the tolerance.
-  // Using numeric_limits::min() gives us more robustness to denormals.
-  RealScalar tolerance = (std::numeric_limits<RealScalar>::min)();
-  for (Index i = 0; i < vecD.size(); ++i)
+  template<typename Dest> void evalTo(Dest& dst) const
   {
-    if(abs(vecD(i)) > tolerance)
-      dst.row(i) /= vecD(i);
-    else
-      dst.row(i).setZero();
+    eigen_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)
+    // more precisely, use pseudo-inverse of D (see bug 241)
+    using std::abs;
+    using std::max;
+    typedef typename LDLTType::MatrixType MatrixType;
+    typedef typename LDLTType::Scalar Scalar;
+    typedef typename LDLTType::RealScalar RealScalar;
+    const Diagonal<const MatrixType> vectorD = dec().vectorD();
+    RealScalar tolerance = (max)(vectorD.array().abs().maxCoeff() * NumTraits<Scalar>::epsilon(),
+				 RealScalar(1) / NumTraits<RealScalar>::highest()); // motivated by LAPACK's xGELSS
+    for (Index i = 0; i < vectorD.size(); ++i) {
+      if(abs(vectorD(i)) > tolerance)
+	dst.row(i) /= vectorD(i);
+      else
+	dst.row(i).setZero();
+    }
+
+    // dst = 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;
   }
-
-  // dst = L^-* (D^-* L^-1 P b)
-  // dst = L^-T (D^-1 L^-*T P b)
-  matrixL().transpose().template conjugateIf<Conjugate>().solveInPlace(dst);
-
-  // dst = P^T (L^-* D^-* L^-1 P b) = A^-1 b
-  // dst = P^-T (L^-T D^-1 L^-*T P b) = A^-1 b
-  dst = m_transpositions.transpose() * dst;
+};
 }
-#endif
 
 /** \internal use x = ldlt_object.solve(x);
   *
@@ -652,7 +566,7 @@ MatrixType LDLT<MatrixType,_UpLo>::reconstructedMatrix() const
   // L^* P
   res = matrixU() * res;
   // D(L^*P)
-  res = vectorD().real().asDiagonal() * res;
+  res = vectorD().asDiagonal() * res;
   // L(DL^*P)
   res = matrixL() * res;
   // P^T (LDL^*P)
@@ -663,7 +577,6 @@ MatrixType LDLT<MatrixType,_UpLo>::reconstructedMatrix() const
 
 /** \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>
@@ -674,7 +587,6 @@ 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>

Eigen/src/Cholesky/LLT.h

@@ -10,19 +10,9 @@
 #ifndef EIGEN_LLT_H
 #define EIGEN_LLT_H
 
-namespace Eigen {
+namespace Eigen { 
 
 namespace internal{
-
-template<typename _MatrixType, int _UpLo> struct traits<LLT<_MatrixType, _UpLo> >
- : traits<_MatrixType>
-{
-  typedef MatrixXpr XprKind;
-  typedef SolverStorage StorageKind;
-  typedef int StorageIndex;
-  enum { Flags = 0 };
-};
-
 template<typename MatrixType, int UpLo> struct LLT_Traits;
 }
 
@@ -32,9 +22,9 @@ template<typename MatrixType, int UpLo> struct LLT_Traits;
   *
   * \brief Standard Cholesky decomposition (LL^T) of a matrix and associated features
   *
-  * \tparam _MatrixType the type of the matrix of which we are computing the LL^T Cholesky decomposition
-  * \tparam _UpLo the triangular part that will be used for the decompositon: Lower (default) or Upper.
-  *               The other triangular part won't be read.
+  * \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.
@@ -50,31 +40,26 @@ template<typename MatrixType, int UpLo> struct LLT_Traits;
   *
   * Example: \include LLT_example.cpp
   * Output: \verbinclude LLT_example.out
-  *
-  * \b Performance: for best performance, it is recommended to use a column-major storage format
-  * with the Lower triangular part (the default), or, equivalently, a row-major storage format
-  * with the Upper triangular part. Otherwise, you might get a 20% slowdown for the full factorization
-  * step, and rank-updates can be up to 3 times slower.
-  *
-  * This class supports the \link InplaceDecomposition inplace decomposition \endlink mechanism.
-  *
-  * Note that during the decomposition, only the lower (or upper, as defined by _UpLo) triangular part of A is considered.
-  * Therefore, the strict lower part does not have to store correct values.
-  *
-  * \sa MatrixBase::llt(), SelfAdjointView::llt(), class LDLT
+  *    
+  * \sa MatrixBase::llt(), class LDLT
+  */
+ /* HEY THIS DOX IS DISABLED BECAUSE THERE's A BUG EITHER HERE OR IN LDLT ABOUT THAT (OR BOTH)
+  * Note that during the decomposition, only the upper triangular part of A is considered. Therefore,
+  * the strict lower part does not have to store correct values.
   */
 template<typename _MatrixType, int _UpLo> class LLT
-        : public SolverBase<LLT<_MatrixType, _UpLo> >
 {
   public:
     typedef _MatrixType MatrixType;
-    typedef SolverBase<LLT> Base;
-    friend class SolverBase<LLT>;
-
-    EIGEN_GENERIC_PUBLIC_INTERFACE(LLT)
     enum {
+      RowsAtCompileTime = MatrixType::RowsAtCompileTime,
+      ColsAtCompileTime = MatrixType::ColsAtCompileTime,
+      Options = MatrixType::Options,
       MaxColsAtCompileTime = MatrixType::MaxColsAtCompileTime
     };
+    typedef typename MatrixType::Scalar Scalar;
+    typedef typename NumTraits<typename MatrixType::Scalar>::Real RealScalar;
+    typedef typename MatrixType::Index Index;
 
     enum {
       PacketSize = internal::packet_traits<Scalar>::size,
@@ -98,30 +83,14 @@ template<typename _MatrixType, int _UpLo> class LLT
       * according to the specified problem \a size.
       * \sa LLT()
       */
-    explicit LLT(Index size) : m_matrix(size, size),
+    LLT(Index size) : m_matrix(size, size),
                     m_isInitialized(false) {}
 
-    template<typename InputType>
-    explicit LLT(const EigenBase<InputType>& matrix)
+    LLT(const MatrixType& matrix)
       : m_matrix(matrix.rows(), matrix.cols()),
         m_isInitialized(false)
     {
-      compute(matrix.derived());
-    }
-
-    /** \brief Constructs a LLT factorization from a given matrix
-      *
-      * This overloaded constructor is provided for \link InplaceDecomposition inplace decomposition \endlink when
-      * \c MatrixType is a Eigen::Ref.
-      *
-      * \sa LLT(const EigenBase&)
-      */
-    template<typename InputType>
-    explicit LLT(EigenBase<InputType>& matrix)
-      : m_matrix(matrix.derived()),
-        m_isInitialized(false)
-    {
-      compute(matrix.derived());
+      compute(matrix);
     }
 
     /** \returns a view of the upper triangular matrix U */
@@ -138,7 +107,6 @@ template<typename _MatrixType, int _UpLo> class LLT
       return Traits::getL(m_matrix);
     }
 
-    #ifdef EIGEN_PARSED_BY_DOXYGEN
     /** \returns the solution x of \f$ A x = b \f$ using the current decomposition of A.
       *
       * Since this LLT class assumes anyway that the matrix A is invertible, the solution
@@ -147,28 +115,33 @@ template<typename _MatrixType, int _UpLo> class LLT
       * Example: \include LLT_solve.cpp
       * Output: \verbinclude LLT_solve.out
       *
-      * \sa solveInPlace(), MatrixBase::llt(), SelfAdjointView::llt()
+      * \sa solveInPlace(), MatrixBase::llt()
       */
     template<typename Rhs>
-    inline const Solve<LLT, Rhs>
-    solve(const MatrixBase<Rhs>& b) const;
+    inline const internal::solve_retval<LLT, Rhs>
+    solve(const MatrixBase<Rhs>& b) const
+    {
+      eigen_assert(m_isInitialized && "LLT is not initialized.");
+      eigen_assert(m_matrix.rows()==b.rows()
+                && "LLT::solve(): invalid number of rows of the right hand side matrix b");
+      return internal::solve_retval<LLT, Rhs>(*this, b.derived());
+    }
+
+    #ifdef EIGEN2_SUPPORT
+    template<typename OtherDerived, typename ResultType>
+    bool solve(const MatrixBase<OtherDerived>& b, ResultType *result) const
+    {
+      *result = this->solve(b);
+      return true;
+    }
+    
+    bool isPositiveDefinite() const { return true; }
     #endif
 
     template<typename Derived>
-    void solveInPlace(const MatrixBase<Derived> &bAndX) const;
+    void solveInPlace(MatrixBase<Derived> &bAndX) const;
 
-    template<typename InputType>
-    LLT& compute(const EigenBase<InputType>& matrix);
-
-    /** \returns an estimate of the reciprocal condition number of the matrix of
-      *  which \c *this is the Cholesky decomposition.
-      */
-    RealScalar rcond() const
-    {
-      eigen_assert(m_isInitialized && "LLT is not initialized.");
-      eigen_assert(m_info == Success && "LLT failed because matrix appears to be negative");
-      return internal::rcond_estimate_helper(m_l1_norm, *this);
-    }
+    LLT& compute(const MatrixType& matrix);
 
     /** \returns the LLT decomposition matrix
       *
@@ -185,8 +158,8 @@ template<typename _MatrixType, int _UpLo> class LLT
 
     /** \brief Reports whether previous computation was successful.
       *
-      * \returns \c Success if computation was successful,
-      *          \c NumericalIssue if the matrix.appears not to be positive definite.
+      * \returns \c Success if computation was succesful,
+      *          \c NumericalIssue if the matrix.appears to be negative.
       */
     ComputationInfo info() const
     {
@@ -194,40 +167,18 @@ template<typename _MatrixType, int _UpLo> class LLT
       return m_info;
     }
 
-    /** \returns the adjoint of \c *this, that is, a const reference to the decomposition itself as the underlying matrix is self-adjoint.
-      *
-      * This method is provided for compatibility with other matrix decompositions, thus enabling generic code such as:
-      * \code x = decomposition.adjoint().solve(b) \endcode
-      */
-    const LLT& adjoint() const EIGEN_NOEXCEPT { return *this; };
-
-    inline EIGEN_CONSTEXPR Index rows() const EIGEN_NOEXCEPT { return m_matrix.rows(); }
-    inline EIGEN_CONSTEXPR Index cols() const EIGEN_NOEXCEPT { return m_matrix.cols(); }
+    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>
-    void _solve_impl(const RhsType &rhs, DstType &dst) const;
-
-    template<bool Conjugate, typename RhsType, typename DstType>
-    void _solve_impl_transposed(const RhsType &rhs, DstType &dst) const;
-    #endif
+    LLT rankUpdate(const VectorType& vec, const RealScalar& sigma = 1);
 
   protected:
-
-    static void check_template_parameters()
-    {
-      EIGEN_STATIC_ASSERT_NON_INTEGER(Scalar);
-    }
-
     /** \internal
       * Used to compute and store L
       * The strict upper part is not used and even not initialized.
       */
     MatrixType m_matrix;
-    RealScalar m_l1_norm;
     bool m_isInitialized;
     ComputationInfo m_info;
 };
@@ -237,11 +188,12 @@ namespace internal {
 template<typename Scalar, int UpLo> struct llt_inplace;
 
 template<typename MatrixType, typename VectorType>
-static Index llt_rank_update_lower(MatrixType& mat, const VectorType& vec, const typename MatrixType::RealScalar& sigma)
+static typename MatrixType::Index llt_rank_update_lower(MatrixType& mat, const VectorType& vec, const typename MatrixType::RealScalar& sigma)
 {
   using std::sqrt;
   typedef typename MatrixType::Scalar Scalar;
   typedef typename MatrixType::RealScalar RealScalar;
+  typedef typename MatrixType::Index Index;
   typedef typename MatrixType::ColXpr ColXpr;
   typedef typename internal::remove_all<ColXpr>::type ColXprCleaned;
   typedef typename ColXprCleaned::SegmentReturnType ColXprSegment;
@@ -310,10 +262,11 @@ template<typename Scalar> struct llt_inplace<Scalar, Lower>
 {
   typedef typename NumTraits<Scalar>::Real RealScalar;
   template<typename MatrixType>
-  static Index unblocked(MatrixType& mat)
+  static typename MatrixType::Index unblocked(MatrixType& mat)
   {
     using std::sqrt;
-
+    typedef typename MatrixType::Index Index;
+    
     eigen_assert(mat.rows()==mat.cols());
     const Index size = mat.rows();
     for(Index k = 0; k < size; ++k)
@@ -330,14 +283,15 @@ template<typename Scalar> struct llt_inplace<Scalar, Lower>
         return k;
       mat.coeffRef(k,k) = x = sqrt(x);
       if (k>0 && rs>0) A21.noalias() -= A20 * A10.adjoint();
-      if (rs>0) A21 /= x;
+      if (rs>0) A21 *= RealScalar(1)/x;
     }
     return -1;
   }
 
   template<typename MatrixType>
-  static Index blocked(MatrixType& m)
+  static typename MatrixType::Index blocked(MatrixType& m)
   {
+    typedef typename MatrixType::Index Index;
     eigen_assert(m.rows()==m.cols());
     Index size = m.rows();
     if(size<32)
@@ -362,36 +316,36 @@ template<typename Scalar> struct llt_inplace<Scalar, Lower>
       Index ret;
       if((ret=unblocked(A11))>=0) return k+ret;
       if(rs>0) A11.adjoint().template triangularView<Upper>().template solveInPlace<OnTheRight>(A21);
-      if(rs>0) A22.template selfadjointView<Lower>().rankUpdate(A21,typename NumTraits<RealScalar>::Literal(-1)); // bottleneck
+      if(rs>0) A22.template selfadjointView<Lower>().rankUpdate(A21,-1); // bottleneck
     }
     return -1;
   }
 
   template<typename MatrixType, typename VectorType>
-  static Index rankUpdate(MatrixType& mat, const VectorType& vec, const RealScalar& sigma)
+  static typename MatrixType::Index rankUpdate(MatrixType& mat, const VectorType& vec, const RealScalar& sigma)
   {
     return Eigen::internal::llt_rank_update_lower(mat, vec, sigma);
   }
 };
-
+  
 template<typename Scalar> struct llt_inplace<Scalar, Upper>
 {
   typedef typename NumTraits<Scalar>::Real RealScalar;
 
   template<typename MatrixType>
-  static EIGEN_STRONG_INLINE Index unblocked(MatrixType& mat)
+  static EIGEN_STRONG_INLINE typename MatrixType::Index unblocked(MatrixType& mat)
   {
     Transpose<MatrixType> matt(mat);
     return llt_inplace<Scalar, Lower>::unblocked(matt);
   }
   template<typename MatrixType>
-  static EIGEN_STRONG_INLINE Index blocked(MatrixType& mat)
+  static EIGEN_STRONG_INLINE typename MatrixType::Index blocked(MatrixType& mat)
   {
     Transpose<MatrixType> matt(mat);
     return llt_inplace<Scalar, Lower>::blocked(matt);
   }
   template<typename MatrixType, typename VectorType>
-  static Index rankUpdate(MatrixType& mat, const VectorType& vec, const RealScalar& sigma)
+  static typename MatrixType::Index rankUpdate(MatrixType& mat, const VectorType& vec, const RealScalar& sigma)
   {
     Transpose<MatrixType> matt(mat);
     return llt_inplace<Scalar, Lower>::rankUpdate(matt, vec.conjugate(), sigma);
@@ -402,8 +356,8 @@ template<typename MatrixType> struct LLT_Traits<MatrixType,Lower>
 {
   typedef const TriangularView<const MatrixType, Lower> MatrixL;
   typedef const TriangularView<const typename MatrixType::AdjointReturnType, Upper> MatrixU;
-  static inline MatrixL getL(const MatrixType& m) { return MatrixL(m); }
-  static inline MatrixU getU(const MatrixType& m) { return MatrixU(m.adjoint()); }
+  static inline MatrixL getL(const MatrixType& m) { return m; }
+  static inline MatrixU getU(const MatrixType& m) { return m.adjoint(); }
   static bool inplace_decomposition(MatrixType& m)
   { return llt_inplace<typename MatrixType::Scalar, Lower>::blocked(m)==-1; }
 };
@@ -412,8 +366,8 @@ template<typename MatrixType> struct LLT_Traits<MatrixType,Upper>
 {
   typedef const TriangularView<const typename MatrixType::AdjointReturnType, Lower> MatrixL;
   typedef const TriangularView<const MatrixType, Upper> MatrixU;
-  static inline MatrixL getL(const MatrixType& m) { return MatrixL(m.adjoint()); }
-  static inline MatrixU getU(const MatrixType& m) { return MatrixU(m); }
+  static inline MatrixL getL(const MatrixType& m) { return m.adjoint(); }
+  static inline MatrixU getU(const MatrixType& m) { return m; }
   static bool inplace_decomposition(MatrixType& m)
   { return llt_inplace<typename MatrixType::Scalar, Upper>::blocked(m)==-1; }
 };
@@ -428,29 +382,12 @@ template<typename MatrixType> struct LLT_Traits<MatrixType,Upper>
   * Output: \verbinclude TutorialLinAlgComputeTwice.out
   */
 template<typename MatrixType, int _UpLo>
-template<typename InputType>
-LLT<MatrixType,_UpLo>& LLT<MatrixType,_UpLo>::compute(const EigenBase<InputType>& a)
+LLT<MatrixType,_UpLo>& LLT<MatrixType,_UpLo>::compute(const MatrixType& a)
 {
-  check_template_parameters();
-
   eigen_assert(a.rows()==a.cols());
   const Index size = a.rows();
   m_matrix.resize(size, size);
-  if (!internal::is_same_dense(m_matrix, a.derived()))
-    m_matrix = a.derived();
-
-  // Compute matrix L1 norm = max abs column sum.
-  m_l1_norm = RealScalar(0);
-  // TODO move this code to SelfAdjointView
-  for (Index col = 0; col < size; ++col) {
-    RealScalar abs_col_sum;
-    if (_UpLo == Lower)
-      abs_col_sum = m_matrix.col(col).tail(size - col).template lpNorm<1>() + m_matrix.row(col).head(col).template lpNorm<1>();
-    else
-      abs_col_sum = m_matrix.col(col).head(col).template lpNorm<1>() + m_matrix.row(col).tail(size - col).template lpNorm<1>();
-    if (abs_col_sum > m_l1_norm)
-      m_l1_norm = abs_col_sum;
-  }
+  m_matrix = a;
 
   m_isInitialized = true;
   bool ok = Traits::inplace_decomposition(m_matrix);
@@ -466,7 +403,7 @@ LLT<MatrixType,_UpLo>& LLT<MatrixType,_UpLo>::compute(const EigenBase<InputType>
   */
 template<typename _MatrixType, int _UpLo>
 template<typename VectorType>
-LLT<_MatrixType,_UpLo> & LLT<_MatrixType,_UpLo>::rankUpdate(const VectorType& v, const RealScalar& sigma)
+LLT<_MatrixType,_UpLo> LLT<_MatrixType,_UpLo>::rankUpdate(const VectorType& v, const RealScalar& sigma)
 {
   EIGEN_STATIC_ASSERT_VECTOR_ONLY(VectorType);
   eigen_assert(v.size()==m_matrix.cols());
@@ -478,42 +415,39 @@ LLT<_MatrixType,_UpLo> & LLT<_MatrixType,_UpLo>::rankUpdate(const VectorType& v,
 
   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
+    
+namespace internal {
+template<typename _MatrixType, int UpLo, typename Rhs>
+struct solve_retval<LLT<_MatrixType, UpLo>, Rhs>
+  : solve_retval_base<LLT<_MatrixType, UpLo>, Rhs>
 {
-  _solve_impl_transposed<true>(rhs, dst);
-}
+  typedef LLT<_MatrixType,UpLo> LLTType;
+  EIGEN_MAKE_SOLVE_HELPERS(LLTType,Rhs)
 
-template<typename _MatrixType,int _UpLo>
-template<bool Conjugate, typename RhsType, typename DstType>
-void LLT<_MatrixType,_UpLo>::_solve_impl_transposed(const RhsType &rhs, DstType &dst) const
-{
-    dst = rhs;
-
-    matrixL().template conjugateIf<!Conjugate>().solveInPlace(dst);
-    matrixU().template conjugateIf<!Conjugate>().solveInPlace(dst);
+  template<typename Dest> void evalTo(Dest& dst) const
+  {
+    dst = rhs();
+    dec().solveInPlace(dst);
+  }
+};
 }
-#endif
 
 /** \internal use x = llt_object.solve(x);
-  *
+  * 
   * This is the \em in-place version of solve().
   *
   * \param bAndX represents both the right-hand side matrix b and result x.
   *
-  * This version avoids a copy when the right hand side matrix b is not needed anymore.
+  * \returns true always! If you need to check for existence of solutions, use another decomposition like LU, QR, or SVD.
   *
-  * \warning The parameter is only marked 'const' to make the C++ compiler accept a temporary expression here.
-  * This function will const_cast it, so constness isn't honored here.
+  * This version avoids a copy when the right hand side matrix b is not
+  * needed anymore.
   *
   * \sa LLT::solve(), MatrixBase::llt()
   */
 template<typename MatrixType, int _UpLo>
 template<typename Derived>
-void LLT<MatrixType,_UpLo>::solveInPlace(const MatrixBase<Derived> &bAndX) const
+void LLT<MatrixType,_UpLo>::solveInPlace(MatrixBase<Derived> &bAndX) const
 {
   eigen_assert(m_isInitialized && "LLT is not initialized.");
   eigen_assert(m_matrix.rows()==bAndX.rows());
@@ -533,7 +467,6 @@ MatrixType LLT<MatrixType,_UpLo>::reconstructedMatrix() const
 
 /** \cholesky_module
   * \returns the LLT decomposition of \c *this
-  * \sa SelfAdjointView::llt()
   */
 template<typename Derived>
 inline const LLT<typename MatrixBase<Derived>::PlainObject>
@@ -544,7 +477,6 @@ 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>

Eigen/src/Cholesky/LLT_MKL.h

@@ -25,75 +25,78 @@
  SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 
  ********************************************************************************
- *   Content : Eigen bindings to LAPACKe
+ *   Content : Eigen bindings to Intel(R) MKL
  *     LLt decomposition based on LAPACKE_?potrf function.
  ********************************************************************************
 */
 
-#ifndef EIGEN_LLT_LAPACKE_H
-#define EIGEN_LLT_LAPACKE_H
+#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 lapacke_llt;
+template<typename Scalar> struct mkl_llt;
 
-#define EIGEN_LAPACKE_LLT(EIGTYPE, BLASTYPE, LAPACKE_PREFIX) \
-template<> struct lapacke_llt<EIGTYPE> \
+#define EIGEN_MKL_LLT(EIGTYPE, MKLTYPE, MKLPREFIX) \
+template<> struct mkl_llt<EIGTYPE> \
 { \
   template<typename MatrixType> \
-  static inline Index potrf(MatrixType& m, char uplo) \
+  static inline typename MatrixType::Index potrf(MatrixType& m, char uplo) \
   { \
     lapack_int matrix_order; \
     lapack_int size, lda, info, StorageOrder; \
     EIGTYPE* a; \
     eigen_assert(m.rows()==m.cols()); \
     /* Set up parameters for ?potrf */ \
-    size = convert_index<lapack_int>(m.rows()); \
+    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 = convert_index<lapack_int>(m.outerStride()); \
+    lda = m.outerStride(); \
 \
-    info = LAPACKE_##LAPACKE_PREFIX##potrf( matrix_order, uplo, size, (BLASTYPE*)a, lda ); \
-    info = (info==0) ? -1 : info>0 ? info-1 : size; \
+    info = LAPACKE_##MKLPREFIX##potrf( matrix_order, uplo, size, (MKLTYPE*)a, lda ); \
+    info = (info==0) ? Success : NumericalIssue; \
     return info; \
   } \
 }; \
 template<> struct llt_inplace<EIGTYPE, Lower> \
 { \
   template<typename MatrixType> \
-  static Index blocked(MatrixType& m) \
+  static typename MatrixType::Index blocked(MatrixType& m) \
   { \
-    return lapacke_llt<EIGTYPE>::potrf(m, 'L'); \
+    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) \
+  static typename MatrixType::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) \
+  static typename MatrixType::Index blocked(MatrixType& m) \
   { \
-    return lapacke_llt<EIGTYPE>::potrf(m, 'U'); \
+    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) \
+  static typename MatrixType::Index rankUpdate(MatrixType& mat, const VectorType& vec, const typename MatrixType::RealScalar& sigma) \
   { \
     Transpose<MatrixType> matt(mat); \
     return llt_inplace<EIGTYPE, Lower>::rankUpdate(matt, vec.conjugate(), sigma); \
   } \
 };
 
-EIGEN_LAPACKE_LLT(double, double, d)
-EIGEN_LAPACKE_LLT(float, float, s)
-EIGEN_LAPACKE_LLT(dcomplex, lapack_complex_double, z)
-EIGEN_LAPACKE_LLT(scomplex, lapack_complex_float, c)
+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_LAPACKE_H
+#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

@@ -10,66 +10,58 @@
 #ifndef EIGEN_CHOLMODSUPPORT_H
 #define EIGEN_CHOLMODSUPPORT_H
 
-namespace Eigen {
+namespace Eigen { 
 
 namespace internal {
 
-template<typename Scalar> struct cholmod_configure_matrix;
-
-template<> struct cholmod_configure_matrix<double> {
-  template<typename CholmodType>
-  static void run(CholmodType& mat) {
+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;
   }
-};
-
-template<> struct cholmod_configure_matrix<std::complex<double> > {
-  template<typename CholmodType>
-  static void run(CholmodType& mat) {
+  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;
   }
-};
-
-// Other scalar types are not yet supported by Cholmod
-// template<> struct cholmod_configure_matrix<float> {
-//   template<typename CholmodType>
-//   static void run(CholmodType& mat) {
-//     mat.xtype = CHOLMOD_REAL;
-//     mat.dtype = CHOLMOD_SINGLE;
-//   }
-// };
-//
-// template<> struct cholmod_configure_matrix<std::complex<float> > {
-//   template<typename CholmodType>
-//   static void run(CholmodType& mat) {
-//     mat.xtype = CHOLMOD_COMPLEX;
-//     mat.dtype = CHOLMOD_SINGLE;
-//   }
-// };
+  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(Ref<SparseMatrix<_Scalar,_Options,_StorageIndex> > mat)
+template<typename _Scalar, int _Options, typename _Index>
+cholmod_sparse viewAsCholmod(SparseMatrix<_Scalar,_Options,_Index>& mat)
 {
   cholmod_sparse res;
   res.nzmax   = mat.nonZeros();
-  res.nrow    = mat.rows();
+  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
   {
@@ -79,12 +71,12 @@ cholmod_sparse viewAsCholmod(Ref<SparseMatrix<_Scalar,_Options,_StorageIndex> >
 
   res.dtype   = 0;
   res.stype   = -1;
-
-  if (internal::is_same<_StorageIndex,int>::value)
+  
+  if (internal::is_same<_Index,int>::value)
   {
     res.itype = CHOLMOD_INT;
   }
-  else if (internal::is_same<_StorageIndex,SuiteSparse_long>::value)
+  else if (internal::is_same<_Index,UF_long>::value)
   {
     res.itype = CHOLMOD_LONG;
   }
@@ -94,39 +86,29 @@ cholmod_sparse viewAsCholmod(Ref<SparseMatrix<_Scalar,_Options,_StorageIndex> >
   }
 
   // setup res.xtype
-  internal::cholmod_configure_matrix<_Scalar>::run(res);
-
+  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(Ref<SparseMatrix<_Scalar,_Options,_Index> >(mat.const_cast_derived()));
-  return res;
-}
-
-template<typename _Scalar, int _Options, typename _Index>
-const cholmod_sparse viewAsCholmod(const SparseVector<_Scalar,_Options,_Index>& mat)
-{
-  cholmod_sparse res = viewAsCholmod(Ref<SparseMatrix<_Scalar,_Options,_Index> >(mat.const_cast_derived()));
+  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 viewAsCholmod(const SparseSelfAdjointView<SparseMatrix<_Scalar,_Options,_Index>, UpLo>& mat)
 {
-  cholmod_sparse res = viewAsCholmod(Ref<SparseMatrix<_Scalar,_Options,_Index> >(mat.matrix().const_cast_derived()));
-
+  cholmod_sparse res = viewAsCholmod(mat.matrix().const_cast_derived());
+  
   if(UpLo==Upper) res.stype =  1;
   if(UpLo==Lower) res.stype = -1;
-  // swap stype for rowmajor matrices (only works for real matrices)
-  EIGEN_STATIC_ASSERT((_Options & RowMajorBit) == 0 || NumTraits<_Scalar>::IsComplex == 0, THIS_METHOD_IS_ONLY_FOR_COLUMN_MAJOR_MATRICES);
-  if(_Options & RowMajorBit) res.stype *=-1;
 
   return res;
 }
@@ -147,59 +129,21 @@ cholmod_dense viewAsCholmod(MatrixBase<Derived>& mat)
   res.x      = (void*)(mat.derived().data());
   res.z      = 0;
 
-  internal::cholmod_configure_matrix<Scalar>::run(res);
+  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)
+template<typename Scalar, int Flags, typename Index>
+MappedSparseMatrix<Scalar,Flags,Index> 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) );
+  return MappedSparseMatrix<Scalar,Flags,Index>
+         (cm.nrow, cm.ncol, static_cast<Index*>(cm.p)[cm.ncol],
+          static_cast<Index*>(cm.p), static_cast<Index*>(cm.i),static_cast<Scalar*>(cm.x) );
 }
 
-namespace internal {
-
-// template specializations for int and long that call the correct cholmod method
-
-#define EIGEN_CHOLMOD_SPECIALIZE0(ret, name) \
-    template<typename _StorageIndex> inline ret cm_ ## name       (cholmod_common &Common) { return cholmod_ ## name   (&Common); } \
-    template<>                       inline ret cm_ ## name<SuiteSparse_long> (cholmod_common &Common) { return cholmod_l_ ## name (&Common); }
-
-#define EIGEN_CHOLMOD_SPECIALIZE1(ret, name, t1, a1) \
-    template<typename _StorageIndex> inline ret cm_ ## name       (t1& a1, cholmod_common &Common) { return cholmod_ ## name   (&a1, &Common); } \
-    template<>                       inline ret cm_ ## name<SuiteSparse_long> (t1& a1, cholmod_common &Common) { return cholmod_l_ ## name (&a1, &Common); }
-
-EIGEN_CHOLMOD_SPECIALIZE0(int, start)
-EIGEN_CHOLMOD_SPECIALIZE0(int, finish)
-
-EIGEN_CHOLMOD_SPECIALIZE1(int, free_factor, cholmod_factor*, L)
-EIGEN_CHOLMOD_SPECIALIZE1(int, free_dense,  cholmod_dense*,  X)
-EIGEN_CHOLMOD_SPECIALIZE1(int, free_sparse, cholmod_sparse*, A)
-
-EIGEN_CHOLMOD_SPECIALIZE1(cholmod_factor*, analyze, cholmod_sparse, A)
-
-template<typename _StorageIndex> inline cholmod_dense*  cm_solve         (int sys, cholmod_factor& L, cholmod_dense&  B, cholmod_common &Common) { return cholmod_solve     (sys, &L, &B, &Common); }
-template<>                       inline cholmod_dense*  cm_solve<SuiteSparse_long>   (int sys, cholmod_factor& L, cholmod_dense&  B, cholmod_common &Common) { return cholmod_l_solve   (sys, &L, &B, &Common); }
-
-template<typename _StorageIndex> inline cholmod_sparse* cm_spsolve       (int sys, cholmod_factor& L, cholmod_sparse& B, cholmod_common &Common) { return cholmod_spsolve   (sys, &L, &B, &Common); }
-template<>                       inline cholmod_sparse* cm_spsolve<SuiteSparse_long> (int sys, cholmod_factor& L, cholmod_sparse& B, cholmod_common &Common) { return cholmod_l_spsolve (sys, &L, &B, &Common); }
-
-template<typename _StorageIndex>
-inline int  cm_factorize_p       (cholmod_sparse*  A, double beta[2], _StorageIndex* fset, std::size_t fsize, cholmod_factor* L, cholmod_common &Common) { return cholmod_factorize_p   (A, beta, fset, fsize, L, &Common); }
-template<>
-inline int  cm_factorize_p<SuiteSparse_long> (cholmod_sparse*  A, double beta[2], SuiteSparse_long* fset,          std::size_t fsize, cholmod_factor* L, cholmod_common &Common) { return cholmod_l_factorize_p (A, beta, fset, fsize, L, &Common); }
-
-#undef EIGEN_CHOLMOD_SPECIALIZE0
-#undef EIGEN_CHOLMOD_SPECIALIZE1
-
-}  // namespace internal
-
-
 enum CholmodMode {
   CholmodAuto, CholmodSimplicialLLt, CholmodSupernodalLLt, CholmodLDLt
 };
@@ -211,56 +155,48 @@ enum CholmodMode {
   * \sa class CholmodSupernodalLLT, class CholmodSimplicialLDLT, class CholmodSimplicialLLT
   */
 template<typename _MatrixType, int _UpLo, typename Derived>
-class CholmodBase : public SparseSolverBase<Derived>
+class CholmodBase : internal::noncopyable
 {
-  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
-    };
+    typedef typename MatrixType::Index Index;
 
   public:
 
     CholmodBase()
-      : m_cholmodFactor(0), m_info(Success), m_factorizationIsOk(false), m_analysisIsOk(false)
+      : m_cholmodFactor(0), m_info(Success), m_isInitialized(false)
     {
-      EIGEN_STATIC_ASSERT((internal::is_same<double,RealScalar>::value), CHOLMOD_SUPPORTS_DOUBLE_PRECISION_ONLY);
-      m_shiftOffset[0] = m_shiftOffset[1] = 0.0;
-      internal::cm_start<StorageIndex>(m_cholmod);
+      cholmod_start(&m_cholmod);
     }
 
-    explicit CholmodBase(const MatrixType& matrix)
-      : m_cholmodFactor(0), m_info(Success), m_factorizationIsOk(false), m_analysisIsOk(false)
+    CholmodBase(const MatrixType& matrix)
+      : m_cholmodFactor(0), m_info(Success), m_isInitialized(false)
     {
-      EIGEN_STATIC_ASSERT((internal::is_same<double,RealScalar>::value), CHOLMOD_SUPPORTS_DOUBLE_PRECISION_ONLY);
-      m_shiftOffset[0] = m_shiftOffset[1] = 0.0;
-      internal::cm_start<StorageIndex>(m_cholmod);
+      m_shiftOffset[0] = m_shiftOffset[1] = RealScalar(0.0);
+      cholmod_start(&m_cholmod);
       compute(matrix);
     }
 
     ~CholmodBase()
     {
       if(m_cholmodFactor)
-        internal::cm_free_factor<StorageIndex>(m_cholmodFactor, m_cholmod);
-      internal::cm_finish<StorageIndex>(m_cholmod);
+        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); }
-
+    
+    inline Index cols() const { return m_cholmodFactor->n; }
+    inline Index rows() const { return m_cholmodFactor->n; }
+    
+    Derived& derived() { return *static_cast<Derived*>(this); }
+    const Derived& derived() const { return *static_cast<const Derived*>(this); }
+    
     /** \brief Reports whether previous computation was successful.
       *
-      * \returns \c Success if computation was successful,
+      * \returns \c Success if computation was succesful,
       *          \c NumericalIssue if the matrix.appears to be negative.
       */
     ComputationInfo info() const
@@ -276,32 +212,60 @@ class CholmodBase : public SparseSolverBase<Derived>
       factorize(matrix);
       return derived();
     }
-
-    /** Performs a symbolic decomposition on the sparsity pattern of \a matrix.
+    
+    /** \returns the solution x of \f$ A x = b \f$ using the current decomposition of A.
+      *
+      * \sa compute()
+      */
+    template<typename Rhs>
+    inline const internal::solve_retval<CholmodBase, Rhs>
+    solve(const MatrixBase<Rhs>& b) const
+    {
+      eigen_assert(m_isInitialized && "LLT is not initialized.");
+      eigen_assert(rows()==b.rows()
+                && "CholmodDecomposition::solve(): invalid number of rows of the right hand side matrix b");
+      return internal::solve_retval<CholmodBase, Rhs>(*this, b.derived());
+    }
+    
+    /** \returns the solution x of \f$ A x = b \f$ using the current decomposition of A.
+      *
+      * \sa compute()
+      */
+    template<typename Rhs>
+    inline const internal::sparse_solve_retval<CholmodBase, Rhs>
+    solve(const SparseMatrixBase<Rhs>& b) const
+    {
+      eigen_assert(m_isInitialized && "LLT is not initialized.");
+      eigen_assert(rows()==b.rows()
+                && "CholmodDecomposition::solve(): invalid number of rows of the right hand side matrix b");
+      return internal::sparse_solve_retval<CholmodBase, Rhs>(*this, b.derived());
+    }
+    
+    /** Performs a symbolic decomposition on the sparcity 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)
       {
-        internal::cm_free_factor<StorageIndex>(m_cholmodFactor, m_cholmod);
+        cholmod_free_factor(&m_cholmodFactor, &m_cholmod);
         m_cholmodFactor = 0;
       }
       cholmod_sparse A = viewAsCholmod(matrix.template selfadjointView<UpLo>());
-      m_cholmodFactor = internal::cm_analyze<StorageIndex>(A, m_cholmod);
-
+      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.
+      * The given matrix must has the same sparcity than the matrix on which the symbolic decomposition has been performed.
       *
       * \sa analyzePattern()
       */
@@ -309,46 +273,43 @@ class CholmodBase : public SparseSolverBase<Derived>
     {
       eigen_assert(m_analysisIsOk && "You must first call analyzePattern()");
       cholmod_sparse A = viewAsCholmod(matrix.template selfadjointView<UpLo>());
-      internal::cm_factorize_p<StorageIndex>(&A, m_shiftOffset, 0, 0, m_cholmodFactor, m_cholmod);
-
+      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
+    void _solve(const MatrixBase<Rhs> &b, MatrixBase<Dest> &dest) const
     {
       eigen_assert(m_factorizationIsOk && "The decomposition is not in a valid state for solving, you must first call either compute() or symbolic()/numeric()");
       const Index size = m_cholmodFactor->n;
       EIGEN_UNUSED_VARIABLE(size);
       eigen_assert(size==b.rows());
 
-      // Cholmod needs column-major storage without inner-stride, which corresponds to the default behavior of Ref.
-      Ref<const Matrix<typename Rhs::Scalar,Dynamic,Dynamic,ColMajor> > b_ref(b.derived());
-
+      // note: cd stands for Cholmod Dense
+      Rhs& b_ref(b.const_cast_derived());
       cholmod_dense b_cd = viewAsCholmod(b_ref);
-      cholmod_dense* x_cd = internal::cm_solve<StorageIndex>(CHOLMOD_A, *m_cholmodFactor, b_cd, m_cholmod);
+      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.)
-      // NOTE Actually, the copy can be avoided by calling cholmod_solve2 instead of cholmod_solve
+      // TODO optimize this copy by swapping when possible (be carreful with alignment, etc.)
       dest = Matrix<Scalar,Dest::RowsAtCompileTime,Dest::ColsAtCompileTime>::Map(reinterpret_cast<Scalar*>(x_cd->x),b.rows(),b.cols());
-      internal::cm_free_dense<StorageIndex>(x_cd, m_cholmod);
+      cholmod_free_dense(&x_cd, &m_cholmod);
     }
-
+    
     /** \internal */
-    template<typename RhsDerived, typename DestDerived>
-    void _solve_impl(const SparseMatrixBase<RhsDerived> &b, SparseMatrixBase<DestDerived> &dest) const
+    template<typename RhsScalar, int RhsOptions, typename RhsIndex, typename DestScalar, int DestOptions, typename DestIndex>
+    void _solve(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;
@@ -356,22 +317,19 @@ class CholmodBase : public SparseSolverBase<Derived>
       eigen_assert(size==b.rows());
 
       // note: cs stands for Cholmod Sparse
-      Ref<SparseMatrix<typename RhsDerived::Scalar,ColMajor,typename RhsDerived::StorageIndex> > b_ref(b.const_cast_derived());
-      cholmod_sparse b_cs = viewAsCholmod(b_ref);
-      cholmod_sparse* x_cs = internal::cm_spsolve<StorageIndex>(CHOLMOD_A, *m_cholmodFactor, b_cs, m_cholmod);
+      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.)
-      // NOTE cholmod_spsolve in fact just calls the dense solver for blocks of 4 columns at a time (similar to Eigen's sparse solver)
-      dest.derived() = viewAsEigen<typename DestDerived::Scalar,ColMajor,typename DestDerived::StorageIndex>(*x_cs);
-      internal::cm_free_sparse<StorageIndex>(x_cs, m_cholmod);
+      // TODO optimize this copy by swapping when possible (be carreful 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
@@ -383,70 +341,20 @@ class CholmodBase : public SparseSolverBase<Derived>
       */
     Derived& setShift(const RealScalar& offset)
     {
-      m_shiftOffset[0] = double(offset);
+      m_shiftOffset[0] = offset;
       return derived();
     }
-
-    /** \returns the determinant of the underlying matrix from the current factorization */
-    Scalar determinant() const
-    {
-      using std::exp;
-      return exp(logDeterminant());
-    }
-
-    /** \returns the log determinant of the underlying matrix from the current factorization */
-    Scalar logDeterminant() const
-    {
-      using std::log;
-      using numext::real;
-      eigen_assert(m_factorizationIsOk && "The decomposition is not in a valid state for solving, you must first call either compute() or symbolic()/numeric()");
-
-      RealScalar logDet = 0;
-      Scalar *x = static_cast<Scalar*>(m_cholmodFactor->x);
-      if (m_cholmodFactor->is_super)
-      {
-        // Supernodal factorization stored as a packed list of dense column-major blocs,
-        // as described by the following structure:
-
-        // super[k] == index of the first column of the j-th super node
-        StorageIndex *super = static_cast<StorageIndex*>(m_cholmodFactor->super);
-        // pi[k] == offset to the description of row indices
-        StorageIndex *pi = static_cast<StorageIndex*>(m_cholmodFactor->pi);
-        // px[k] == offset to the respective dense block
-        StorageIndex *px = static_cast<StorageIndex*>(m_cholmodFactor->px);
-
-        Index nb_super_nodes = m_cholmodFactor->nsuper;
-        for (Index k=0; k < nb_super_nodes; ++k)
-        {
-          StorageIndex ncols = super[k + 1] - super[k];
-          StorageIndex nrows = pi[k + 1] - pi[k];
-
-          Map<const Array<Scalar,1,Dynamic>, 0, InnerStride<> > sk(x + px[k], ncols, InnerStride<>(nrows+1));
-          logDet += sk.real().log().sum();
-        }
-      }
-      else
-      {
-        // Simplicial factorization stored as standard CSC matrix.
-        StorageIndex *p = static_cast<StorageIndex*>(m_cholmodFactor->p);
-        Index size = m_cholmodFactor->n;
-        for (Index k=0; k<size; ++k)
-          logDet += log(real( x[p[k]] ));
-      }
-      if (m_cholmodFactor->is_ll)
-        logDet *= 2.0;
-      return logDet;
-    };
-
+    
     template<typename Stream>
     void dumpMemory(Stream& /*s*/)
     {}
-
+    
   protected:
     mutable cholmod_common m_cholmod;
     cholmod_factor* m_cholmodFactor;
-    double m_shiftOffset[2];
+    RealScalar m_shiftOffset[2];
     mutable ComputationInfo m_info;
+    bool m_isInitialized;
     int m_factorizationIsOk;
     int m_analysisIsOk;
 };
@@ -457,38 +365,34 @@ class CholmodBase : public SparseSolverBase<Derived>
   *
   * 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
+  * This simplicial variant is equivalent to Eigen's built-in SimplicialLLT class. Thefore, it has little practical interest.
+  * The sparse matrix A must be selfajoint and positive definite. The vectors or matrices
   * X and B can be either dense or sparse.
   *
   * \tparam _MatrixType the type of the sparse matrix A, it must be a SparseMatrix<>
   * \tparam _UpLo the triangular part that will be used for the computations. It can be Lower
   *               or Upper. Default is Lower.
   *
-  * \implsparsesolverconcept
-  *
   * This class supports all kind of SparseMatrix<>: row or column major; upper, lower, or both; compressed or non compressed.
   *
-  * \warning Only double precision real and complex scalar types are supported by Cholmod.
-  *
-  * \sa \ref TutorialSparseSolverConcept, class CholmodSupernodalLLT, class SimplicialLLT
+  * \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);
+      compute(matrix);
     }
 
     ~CholmodSimplicialLLT() {}
@@ -508,38 +412,34 @@ class CholmodSimplicialLLT : public CholmodBase<_MatrixType, _UpLo, CholmodSimpl
   *
   * 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
+  * This simplicial variant is equivalent to Eigen's built-in SimplicialLDLT class. Thefore, it has little practical interest.
+  * The sparse matrix A must be selfajoint and positive definite. The vectors or matrices
   * X and B can be either dense or sparse.
   *
   * \tparam _MatrixType the type of the sparse matrix A, it must be a SparseMatrix<>
   * \tparam _UpLo the triangular part that will be used for the computations. It can be Lower
   *               or Upper. Default is Lower.
   *
-  * \implsparsesolverconcept
-  *
   * This class supports all kind of SparseMatrix<>: row or column major; upper, lower, or both; compressed or non compressed.
   *
-  * \warning Only double precision real and complex scalar types are supported by Cholmod.
-  *
-  * \sa \ref TutorialSparseSolverConcept, class CholmodSupernodalLLT, class SimplicialLDLT
+  * \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);
+      compute(matrix);
     }
 
     ~CholmodSimplicialLDLT() {}
@@ -558,37 +458,33 @@ class CholmodSimplicialLDLT : public CholmodBase<_MatrixType, _UpLo, CholmodSimp
   * 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
+  * The sparse matrix A must be selfajoint and positive definite. The vectors or matrices
   * X and B can be either dense or sparse.
   *
   * \tparam _MatrixType the type of the sparse matrix A, it must be a SparseMatrix<>
   * \tparam _UpLo the triangular part that will be used for the computations. It can be Lower
   *               or Upper. Default is Lower.
   *
-  * \implsparsesolverconcept
-  *
   * This class supports all kind of SparseMatrix<>: row or column major; upper, lower, or both; compressed or non compressed.
   *
-  * \warning Only double precision real and complex scalar types are supported by Cholmod.
-  *
-  * \sa \ref TutorialSparseSolverConcept
+  * \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);
+      compute(matrix);
     }
 
     ~CholmodSupernodalLLT() {}
@@ -605,7 +501,7 @@ class CholmodSupernodalLLT : public CholmodBase<_MatrixType, _UpLo, CholmodSuper
   * \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
+  * using the Cholmod library. The sparse matrix A must be selfajoint 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.
@@ -616,34 +512,30 @@ class CholmodSupernodalLLT : public CholmodBase<_MatrixType, _UpLo, CholmodSuper
   * \tparam _UpLo the triangular part that will be used for the computations. It can be Lower
   *               or Upper. Default is Lower.
   *
-  * \implsparsesolverconcept
-  *
   * This class supports all kind of SparseMatrix<>: row or column major; upper, lower, or both; compressed or non compressed.
   *
-  * \warning Only double precision real and complex scalar types are supported by Cholmod.
-  *
-  * \sa \ref TutorialSparseSolverConcept
+  * \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);
+      compute(matrix);
     }
 
     ~CholmodDecomposition() {}
-
+    
     void setMode(CholmodMode mode)
     {
       switch(mode)
@@ -677,6 +569,36 @@ class CholmodDecomposition : public CholmodBase<_MatrixType, _UpLo, CholmodDecom
     }
 };
 
+namespace internal {
+  
+template<typename _MatrixType, int _UpLo, typename Derived, typename Rhs>
+struct solve_retval<CholmodBase<_MatrixType,_UpLo,Derived>, Rhs>
+  : solve_retval_base<CholmodBase<_MatrixType,_UpLo,Derived>, Rhs>
+{
+  typedef CholmodBase<_MatrixType,_UpLo,Derived> Dec;
+  EIGEN_MAKE_SOLVE_HELPERS(Dec,Rhs)
+
+  template<typename Dest> void evalTo(Dest& dst) const
+  {
+    dec()._solve(rhs(),dst);
+  }
+};
+
+template<typename _MatrixType, int _UpLo, typename Derived, typename Rhs>
+struct sparse_solve_retval<CholmodBase<_MatrixType,_UpLo,Derived>, Rhs>
+  : sparse_solve_retval_base<CholmodBase<_MatrixType,_UpLo,Derived>, Rhs>
+{
+  typedef CholmodBase<_MatrixType,_UpLo,Derived> Dec;
+  EIGEN_MAKE_SPARSE_SOLVE_HELPERS(Dec,Rhs)
+
+  template<typename Dest> void evalTo(Dest& dst) const
+  {
+    dec()._solve(rhs(),dst);
+  }
+};
+
+} // end namespace internal
+
 } // end namespace Eigen
 
 #endif // EIGEN_CHOLMODSUPPORT_H

Eigen/src/Core/ArithmeticSequence.h

@@ -1,413 +0,0 @@
-// This file is part of Eigen, a lightweight C++ template library
-// for linear algebra.
-//
-// Copyright (C) 2017 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_ARITHMETIC_SEQUENCE_H
-#define EIGEN_ARITHMETIC_SEQUENCE_H
-
-namespace Eigen {
-
-namespace internal {
-
-#if (!EIGEN_HAS_CXX11) || !((!EIGEN_COMP_GNUC) || EIGEN_COMP_GNUC>=48)
-template<typename T> struct aseq_negate {};
-
-template<> struct aseq_negate<Index> {
-  typedef Index type;
-};
-
-template<int N> struct aseq_negate<FixedInt<N> > {
-  typedef FixedInt<-N> type;
-};
-
-// Compilation error in the following case:
-template<> struct aseq_negate<FixedInt<DynamicIndex> > {};
-
-template<typename FirstType,typename SizeType,typename IncrType,
-         bool FirstIsSymbolic=symbolic::is_symbolic<FirstType>::value,
-         bool SizeIsSymbolic =symbolic::is_symbolic<SizeType>::value>
-struct aseq_reverse_first_type {
-  typedef Index type;
-};
-
-template<typename FirstType,typename SizeType,typename IncrType>
-struct aseq_reverse_first_type<FirstType,SizeType,IncrType,true,true> {
-  typedef symbolic::AddExpr<FirstType,
-                            symbolic::ProductExpr<symbolic::AddExpr<SizeType,symbolic::ValueExpr<FixedInt<-1> > >,
-                                                  symbolic::ValueExpr<IncrType> >
-                           > type;
-};
-
-template<typename SizeType,typename IncrType,typename EnableIf = void>
-struct aseq_reverse_first_type_aux {
-  typedef Index type;
-};
-
-template<typename SizeType,typename IncrType>
-struct aseq_reverse_first_type_aux<SizeType,IncrType,typename internal::enable_if<bool((SizeType::value+IncrType::value)|0x1)>::type> {
-  typedef FixedInt<(SizeType::value-1)*IncrType::value> type;
-};
-
-template<typename FirstType,typename SizeType,typename IncrType>
-struct aseq_reverse_first_type<FirstType,SizeType,IncrType,true,false> {
-  typedef typename aseq_reverse_first_type_aux<SizeType,IncrType>::type Aux;
-  typedef symbolic::AddExpr<FirstType,symbolic::ValueExpr<Aux> > type;
-};
-
-template<typename FirstType,typename SizeType,typename IncrType>
-struct aseq_reverse_first_type<FirstType,SizeType,IncrType,false,true> {
-  typedef symbolic::AddExpr<symbolic::ProductExpr<symbolic::AddExpr<SizeType,symbolic::ValueExpr<FixedInt<-1> > >,
-                                                  symbolic::ValueExpr<IncrType> >,
-                            symbolic::ValueExpr<> > type;
-};
-#endif
-
-// Helper to cleanup the type of the increment:
-template<typename T> struct cleanup_seq_incr {
-  typedef typename cleanup_index_type<T,DynamicIndex>::type type;
-};
-
-}
-
-//--------------------------------------------------------------------------------
-// seq(first,last,incr) and seqN(first,size,incr)
-//--------------------------------------------------------------------------------
-
-template<typename FirstType=Index,typename SizeType=Index,typename IncrType=internal::FixedInt<1> >
-class ArithmeticSequence;
-
-template<typename FirstType,typename SizeType,typename IncrType>
-ArithmeticSequence<typename internal::cleanup_index_type<FirstType>::type,
-                   typename internal::cleanup_index_type<SizeType>::type,
-                   typename internal::cleanup_seq_incr<IncrType>::type >
-seqN(FirstType first, SizeType size, IncrType incr);
-
-/** \class ArithmeticSequence
-  * \ingroup Core_Module
-  *
-  * This class represents an arithmetic progression \f$ a_0, a_1, a_2, ..., a_{n-1}\f$ defined by
-  * its \em first value \f$ a_0 \f$, its \em size (aka length) \em n, and the \em increment (aka stride)
-  * that is equal to \f$ a_{i+1}-a_{i}\f$ for any \em i.
-  *
-  * It is internally used as the return type of the Eigen::seq and Eigen::seqN functions, and as the input arguments
-  * of DenseBase::operator()(const RowIndices&, const ColIndices&), and most of the time this is the
-  * only way it is used.
-  *
-  * \tparam FirstType type of the first element, usually an Index,
-  *                   but internally it can be a symbolic expression
-  * \tparam SizeType type representing the size of the sequence, usually an Index
-  *                  or a compile time integral constant. Internally, it can also be a symbolic expression
-  * \tparam IncrType type of the increment, can be a runtime Index, or a compile time integral constant (default is compile-time 1)
-  *
-  * \sa Eigen::seq, Eigen::seqN, DenseBase::operator()(const RowIndices&, const ColIndices&), class IndexedView
-  */
-template<typename FirstType,typename SizeType,typename IncrType>
-class ArithmeticSequence
-{
-public:
-  ArithmeticSequence(FirstType first, SizeType size) : m_first(first), m_size(size) {}
-  ArithmeticSequence(FirstType first, SizeType size, IncrType incr) : m_first(first), m_size(size), m_incr(incr) {}
-
-  enum {
-    SizeAtCompileTime = internal::get_fixed_value<SizeType>::value,
-    IncrAtCompileTime = internal::get_fixed_value<IncrType,DynamicIndex>::value
-  };
-
-  /** \returns the size, i.e., number of elements, of the sequence */
-  Index size()  const { return m_size; }
-
-  /** \returns the first element \f$ a_0 \f$ in the sequence */
-  Index first()  const { return m_first; }
-
-  /** \returns the value \f$ a_i \f$ at index \a i in the sequence. */
-  Index operator[](Index i) const { return m_first + i * m_incr; }
-
-  const FirstType& firstObject() const { return m_first; }
-  const SizeType&  sizeObject()  const { return m_size; }
-  const IncrType&  incrObject()  const { return m_incr; }
-
-protected:
-  FirstType m_first;
-  SizeType  m_size;
-  IncrType  m_incr;
-
-public:
-
-#if EIGEN_HAS_CXX11 && ((!EIGEN_COMP_GNUC) || EIGEN_COMP_GNUC>=48)
-  auto reverse() const -> decltype(Eigen::seqN(m_first+(m_size+fix<-1>())*m_incr,m_size,-m_incr)) {
-    return seqN(m_first+(m_size+fix<-1>())*m_incr,m_size,-m_incr);
-  }
-#else
-protected:
-  typedef typename internal::aseq_negate<IncrType>::type ReverseIncrType;
-  typedef typename internal::aseq_reverse_first_type<FirstType,SizeType,IncrType>::type ReverseFirstType;
-public:
-  ArithmeticSequence<ReverseFirstType,SizeType,ReverseIncrType>
-  reverse() const {
-    return seqN(m_first+(m_size+fix<-1>())*m_incr,m_size,-m_incr);
-  }
-#endif
-};
-
-/** \returns an ArithmeticSequence starting at \a first, of length \a size, and increment \a incr
-  *
-  * \sa seqN(FirstType,SizeType), seq(FirstType,LastType,IncrType) */
-template<typename FirstType,typename SizeType,typename IncrType>
-ArithmeticSequence<typename internal::cleanup_index_type<FirstType>::type,typename internal::cleanup_index_type<SizeType>::type,typename internal::cleanup_seq_incr<IncrType>::type >
-seqN(FirstType first, SizeType size, IncrType incr)  {
-  return ArithmeticSequence<typename internal::cleanup_index_type<FirstType>::type,typename internal::cleanup_index_type<SizeType>::type,typename internal::cleanup_seq_incr<IncrType>::type>(first,size,incr);
-}
-
-/** \returns an ArithmeticSequence starting at \a first, of length \a size, and unit increment
-  *
-  * \sa seqN(FirstType,SizeType,IncrType), seq(FirstType,LastType) */
-template<typename FirstType,typename SizeType>
-ArithmeticSequence<typename internal::cleanup_index_type<FirstType>::type,typename internal::cleanup_index_type<SizeType>::type >
-seqN(FirstType first, SizeType size)  {
-  return ArithmeticSequence<typename internal::cleanup_index_type<FirstType>::type,typename internal::cleanup_index_type<SizeType>::type>(first,size);
-}
-
-#ifdef EIGEN_PARSED_BY_DOXYGEN
-
-/** \returns an ArithmeticSequence starting at \a f, up (or down) to \a l, and with positive (or negative) increment \a incr
-  *
-  * It is essentially an alias to:
-  * \code
-  * seqN(f, (l-f+incr)/incr, incr);
-  * \endcode
-  *
-  * \sa seqN(FirstType,SizeType,IncrType), seq(FirstType,LastType)
-  */
-template<typename FirstType,typename LastType, typename IncrType>
-auto seq(FirstType f, LastType l, IncrType incr);
-
-/** \returns an ArithmeticSequence starting at \a f, up (or down) to \a l, and unit increment
-  *
-  * It is essentially an alias to:
-  * \code
-  * seqN(f,l-f+1);
-  * \endcode
-  *
-  * \sa seqN(FirstType,SizeType), seq(FirstType,LastType,IncrType)
-  */
-template<typename FirstType,typename LastType>
-auto seq(FirstType f, LastType l);
-
-#else // EIGEN_PARSED_BY_DOXYGEN
-
-#if EIGEN_HAS_CXX11
-template<typename FirstType,typename LastType>
-auto seq(FirstType f, LastType l) -> decltype(seqN(typename internal::cleanup_index_type<FirstType>::type(f),
-                                                   (  typename internal::cleanup_index_type<LastType>::type(l)
-                                                    - typename internal::cleanup_index_type<FirstType>::type(f)+fix<1>())))
-{
-  return seqN(typename internal::cleanup_index_type<FirstType>::type(f),
-              (typename internal::cleanup_index_type<LastType>::type(l)
-               -typename internal::cleanup_index_type<FirstType>::type(f)+fix<1>()));
-}
-
-template<typename FirstType,typename LastType, typename IncrType>
-auto seq(FirstType f, LastType l, IncrType incr)
-  -> decltype(seqN(typename internal::cleanup_index_type<FirstType>::type(f),
-                   (   typename internal::cleanup_index_type<LastType>::type(l)
-                     - typename internal::cleanup_index_type<FirstType>::type(f)+typename internal::cleanup_seq_incr<IncrType>::type(incr)
-                   ) / typename internal::cleanup_seq_incr<IncrType>::type(incr),
-                   typename internal::cleanup_seq_incr<IncrType>::type(incr)))
-{
-  typedef typename internal::cleanup_seq_incr<IncrType>::type CleanedIncrType;
-  return seqN(typename internal::cleanup_index_type<FirstType>::type(f),
-              ( typename internal::cleanup_index_type<LastType>::type(l)
-               -typename internal::cleanup_index_type<FirstType>::type(f)+CleanedIncrType(incr)) / CleanedIncrType(incr),
-              CleanedIncrType(incr));
-}
-
-#else // EIGEN_HAS_CXX11
-
-template<typename FirstType,typename LastType>
-typename internal::enable_if<!(symbolic::is_symbolic<FirstType>::value || symbolic::is_symbolic<LastType>::value),
-                             ArithmeticSequence<typename internal::cleanup_index_type<FirstType>::type,Index> >::type
-seq(FirstType f, LastType l)
-{
-  return seqN(typename internal::cleanup_index_type<FirstType>::type(f),
-              Index((typename internal::cleanup_index_type<LastType>::type(l)-typename internal::cleanup_index_type<FirstType>::type(f)+fix<1>())));
-}
-
-template<typename FirstTypeDerived,typename LastType>
-typename internal::enable_if<!symbolic::is_symbolic<LastType>::value,
-    ArithmeticSequence<FirstTypeDerived, symbolic::AddExpr<symbolic::AddExpr<symbolic::NegateExpr<FirstTypeDerived>,symbolic::ValueExpr<> >,
-                                                            symbolic::ValueExpr<internal::FixedInt<1> > > > >::type
-seq(const symbolic::BaseExpr<FirstTypeDerived> &f, LastType l)
-{
-  return seqN(f.derived(),(typename internal::cleanup_index_type<LastType>::type(l)-f.derived()+fix<1>()));
-}
-
-template<typename FirstType,typename LastTypeDerived>
-typename internal::enable_if<!symbolic::is_symbolic<FirstType>::value,
-    ArithmeticSequence<typename internal::cleanup_index_type<FirstType>::type,
-                        symbolic::AddExpr<symbolic::AddExpr<LastTypeDerived,symbolic::ValueExpr<> >,
-                                          symbolic::ValueExpr<internal::FixedInt<1> > > > >::type
-seq(FirstType f, const symbolic::BaseExpr<LastTypeDerived> &l)
-{
-  return seqN(typename internal::cleanup_index_type<FirstType>::type(f),(l.derived()-typename internal::cleanup_index_type<FirstType>::type(f)+fix<1>()));
-}
-
-template<typename FirstTypeDerived,typename LastTypeDerived>
-ArithmeticSequence<FirstTypeDerived,
-                    symbolic::AddExpr<symbolic::AddExpr<LastTypeDerived,symbolic::NegateExpr<FirstTypeDerived> >,symbolic::ValueExpr<internal::FixedInt<1> > > >
-seq(const symbolic::BaseExpr<FirstTypeDerived> &f, const symbolic::BaseExpr<LastTypeDerived> &l)
-{
-  return seqN(f.derived(),(l.derived()-f.derived()+fix<1>()));
-}
-
-
-template<typename FirstType,typename LastType, typename IncrType>
-typename internal::enable_if<!(symbolic::is_symbolic<FirstType>::value || symbolic::is_symbolic<LastType>::value),
-    ArithmeticSequence<typename internal::cleanup_index_type<FirstType>::type,Index,typename internal::cleanup_seq_incr<IncrType>::type> >::type
-seq(FirstType f, LastType l, IncrType incr)
-{
-  typedef typename internal::cleanup_seq_incr<IncrType>::type CleanedIncrType;
-  return seqN(typename internal::cleanup_index_type<FirstType>::type(f),
-              Index((typename internal::cleanup_index_type<LastType>::type(l)-typename internal::cleanup_index_type<FirstType>::type(f)+CleanedIncrType(incr))/CleanedIncrType(incr)), incr);
-}
-
-template<typename FirstTypeDerived,typename LastType, typename IncrType>
-typename internal::enable_if<!symbolic::is_symbolic<LastType>::value,
-    ArithmeticSequence<FirstTypeDerived,
-                        symbolic::QuotientExpr<symbolic::AddExpr<symbolic::AddExpr<symbolic::NegateExpr<FirstTypeDerived>,
-                                                                                   symbolic::ValueExpr<> >,
-                                                                 symbolic::ValueExpr<typename internal::cleanup_seq_incr<IncrType>::type> >,
-                                              symbolic::ValueExpr<typename internal::cleanup_seq_incr<IncrType>::type> >,
-                        typename internal::cleanup_seq_incr<IncrType>::type> >::type
-seq(const symbolic::BaseExpr<FirstTypeDerived> &f, LastType l, IncrType incr)
-{
-  typedef typename internal::cleanup_seq_incr<IncrType>::type CleanedIncrType;
-  return seqN(f.derived(),(typename internal::cleanup_index_type<LastType>::type(l)-f.derived()+CleanedIncrType(incr))/CleanedIncrType(incr), incr);
-}
-
-template<typename FirstType,typename LastTypeDerived, typename IncrType>
-typename internal::enable_if<!symbolic::is_symbolic<FirstType>::value,
-    ArithmeticSequence<typename internal::cleanup_index_type<FirstType>::type,
-                        symbolic::QuotientExpr<symbolic::AddExpr<symbolic::AddExpr<LastTypeDerived,symbolic::ValueExpr<> >,
-                                                                 symbolic::ValueExpr<typename internal::cleanup_seq_incr<IncrType>::type> >,
-                                               symbolic::ValueExpr<typename internal::cleanup_seq_incr<IncrType>::type> >,
-                        typename internal::cleanup_seq_incr<IncrType>::type> >::type
-seq(FirstType f, const symbolic::BaseExpr<LastTypeDerived> &l, IncrType incr)
-{
-  typedef typename internal::cleanup_seq_incr<IncrType>::type CleanedIncrType;
-  return seqN(typename internal::cleanup_index_type<FirstType>::type(f),
-              (l.derived()-typename internal::cleanup_index_type<FirstType>::type(f)+CleanedIncrType(incr))/CleanedIncrType(incr), incr);
-}
-
-template<typename FirstTypeDerived,typename LastTypeDerived, typename IncrType>
-ArithmeticSequence<FirstTypeDerived,
-                    symbolic::QuotientExpr<symbolic::AddExpr<symbolic::AddExpr<LastTypeDerived,
-                                                                               symbolic::NegateExpr<FirstTypeDerived> >,
-                                                             symbolic::ValueExpr<typename internal::cleanup_seq_incr<IncrType>::type> >,
-                                          symbolic::ValueExpr<typename internal::cleanup_seq_incr<IncrType>::type> >,
-                    typename internal::cleanup_seq_incr<IncrType>::type>
-seq(const symbolic::BaseExpr<FirstTypeDerived> &f, const symbolic::BaseExpr<LastTypeDerived> &l, IncrType incr)
-{
-  typedef typename internal::cleanup_seq_incr<IncrType>::type CleanedIncrType;
-  return seqN(f.derived(),(l.derived()-f.derived()+CleanedIncrType(incr))/CleanedIncrType(incr), incr);
-}
-#endif // EIGEN_HAS_CXX11
-
-#endif // EIGEN_PARSED_BY_DOXYGEN
-
-
-#if EIGEN_HAS_CXX11 || defined(EIGEN_PARSED_BY_DOXYGEN)
-/** \cpp11
-  * \returns a symbolic ArithmeticSequence representing the last \a size elements with increment \a incr.
-  *
-  * It is a shortcut for: \code seqN(last-(size-fix<1>)*incr, size, incr) \endcode
-  * 
-  * \sa lastN(SizeType), seqN(FirstType,SizeType), seq(FirstType,LastType,IncrType) */
-template<typename SizeType,typename IncrType>
-auto lastN(SizeType size, IncrType incr)
--> decltype(seqN(Eigen::last-(size-fix<1>())*incr, size, incr))
-{
-  return seqN(Eigen::last-(size-fix<1>())*incr, size, incr);
-}
-
-/** \cpp11
-  * \returns a symbolic ArithmeticSequence representing the last \a size elements with a unit increment.
-  *
-  *  It is a shortcut for: \code seq(last+fix<1>-size, last) \endcode
-  * 
-  * \sa lastN(SizeType,IncrType, seqN(FirstType,SizeType), seq(FirstType,LastType) */
-template<typename SizeType>
-auto lastN(SizeType size)
--> decltype(seqN(Eigen::last+fix<1>()-size, size))
-{
-  return seqN(Eigen::last+fix<1>()-size, size);
-}
-#endif
-
-namespace internal {
-
-// Convert a symbolic span into a usable one (i.e., remove last/end "keywords")
-template<typename T>
-struct make_size_type {
-  typedef typename internal::conditional<symbolic::is_symbolic<T>::value, Index, T>::type type;
-};
-
-template<typename FirstType,typename SizeType,typename IncrType,int XprSize>
-struct IndexedViewCompatibleType<ArithmeticSequence<FirstType,SizeType,IncrType>, XprSize> {
-  typedef ArithmeticSequence<Index,typename make_size_type<SizeType>::type,IncrType> type;
-};
-
-template<typename FirstType,typename SizeType,typename IncrType>
-ArithmeticSequence<Index,typename make_size_type<SizeType>::type,IncrType>
-makeIndexedViewCompatible(const ArithmeticSequence<FirstType,SizeType,IncrType>& ids, Index size,SpecializedType) {
-  return ArithmeticSequence<Index,typename make_size_type<SizeType>::type,IncrType>(
-            eval_expr_given_size(ids.firstObject(),size),eval_expr_given_size(ids.sizeObject(),size),ids.incrObject());
-}
-
-template<typename FirstType,typename SizeType,typename IncrType>
-struct get_compile_time_incr<ArithmeticSequence<FirstType,SizeType,IncrType> > {
-  enum { value = get_fixed_value<IncrType,DynamicIndex>::value };
-};
-
-} // end namespace internal
-
-/** \namespace Eigen::indexing
-  * \ingroup Core_Module
-  * 
-  * The sole purpose of this namespace is to be able to import all functions
-  * and symbols that are expected to be used within operator() for indexing
-  * and slicing. If you already imported the whole Eigen namespace:
-  * \code using namespace Eigen; \endcode
-  * then you are already all set. Otherwise, if you don't want/cannot import
-  * the whole Eigen namespace, the following line:
-  * \code using namespace Eigen::indexing; \endcode
-  * is equivalent to:
-  * \code
-  using Eigen::all;
-  using Eigen::seq;
-  using Eigen::seqN;
-  using Eigen::lastN; // c++11 only
-  using Eigen::last;
-  using Eigen::lastp1;
-  using Eigen::fix;
-  \endcode
-  */
-namespace indexing {
-  using Eigen::all;
-  using Eigen::seq;
-  using Eigen::seqN;
-  #if EIGEN_HAS_CXX11
-  using Eigen::lastN;
-  #endif
-  using Eigen::last;
-  using Eigen::lastp1;
-  using Eigen::fix;
-}
-
-} // end namespace Eigen
-
-#endif // EIGEN_ARITHMETIC_SEQUENCE_H

Eigen/src/Core/Array.h

@@ -12,16 +12,7 @@
 
 namespace Eigen {
 
-namespace internal {
-template<typename _Scalar, int _Rows, int _Cols, int _Options, int _MaxRows, int _MaxCols>
-struct traits<Array<_Scalar, _Rows, _Cols, _Options, _MaxRows, _MaxCols> > : traits<Matrix<_Scalar, _Rows, _Cols, _Options, _MaxRows, _MaxCols> >
-{
-  typedef ArrayXpr XprKind;
-  typedef ArrayBase<Array<_Scalar, _Rows, _Cols, _Options, _MaxRows, _MaxCols> > XprBase;
-};
-}
-
-/** \class Array
+/** \class Array 
   * \ingroup Core_Module
   *
   * \brief General-purpose arrays with easy API for coefficient-wise operations
@@ -33,14 +24,20 @@ struct traits<Array<_Scalar, _Rows, _Cols, _Options, _MaxRows, _MaxCols> > : tra
   * API for the %Matrix class provides easy access to linear-algebra
   * operations.
   *
-  * See documentation of class Matrix for detailed information on the template parameters
-  * storage layout.
-  *
   * This class can be extended with the help of the plugin mechanism described on the page
-  * \ref TopicCustomizing_Plugins by defining the preprocessor symbol \c EIGEN_ARRAY_PLUGIN.
+  * \ref TopicCustomizingEigen by defining the preprocessor symbol \c EIGEN_ARRAY_PLUGIN.
   *
-  * \sa \blank \ref TutorialArrayClass, \ref TopicClassHierarchy
+  * \sa \ref TutorialArrayClass, \ref TopicClassHierarchy
   */
+namespace internal {
+template<typename _Scalar, int _Rows, int _Cols, int _Options, int _MaxRows, int _MaxCols>
+struct traits<Array<_Scalar, _Rows, _Cols, _Options, _MaxRows, _MaxCols> > : traits<Matrix<_Scalar, _Rows, _Cols, _Options, _MaxRows, _MaxCols> >
+{
+  typedef ArrayXpr XprKind;
+  typedef ArrayBase<Array<_Scalar, _Rows, _Cols, _Options, _MaxRows, _MaxCols> > XprBase;
+};
+}
+
 template<typename _Scalar, int _Rows, int _Cols, int _Options, int _MaxRows, int _MaxCols>
 class Array
   : public PlainObjectBase<Array<_Scalar, _Rows, _Cols, _Options, _MaxRows, _MaxCols> >
@@ -72,27 +69,11 @@ 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),
@@ -103,8 +84,7 @@ class Array
       * remain row-vectors and vectors remain vectors.
       */
     template<typename OtherDerived>
-    EIGEN_DEVICE_FUNC
-    EIGEN_STRONG_INLINE Array& operator=(const DenseBase<OtherDerived>& other)
+    EIGEN_STRONG_INLINE Array& operator=(const ArrayBase<OtherDerived>& other)
     {
       return Base::_set(other);
     }
@@ -112,7 +92,6 @@ 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);
@@ -128,7 +107,6 @@ class Array
       *
       * \sa resize(Index,Index)
       */
-    EIGEN_DEVICE_FUNC
     EIGEN_STRONG_INLINE Array() : Base()
     {
       Base::_check_template_params();
@@ -138,7 +116,6 @@ class Array
 #ifndef EIGEN_PARSED_BY_DOXYGEN
     // FIXME is it still needed ??
     /** \internal */
-    EIGEN_DEVICE_FUNC
     Array(internal::constructor_without_unaligned_array_assert)
       : Base(internal::constructor_without_unaligned_array_assert())
     {
@@ -147,106 +124,41 @@ class Array
     }
 #endif
 
-#if EIGEN_HAS_RVALUE_REFERENCES
-    EIGEN_DEVICE_FUNC
-    Array(Array&& other) EIGEN_NOEXCEPT_IF(std::is_nothrow_move_constructible<Scalar>::value)
-      : Base(std::move(other))
+    /** 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 Array(Index dim)
+      : Base(dim, RowsAtCompileTime == 1 ? 1 : dim, ColsAtCompileTime == 1 ? 1 : dim)
     {
       Base::_check_template_params();
+      EIGEN_STATIC_ASSERT_VECTOR_ONLY(Array)
+      eigen_assert(dim >= 0);
+      eigen_assert(SizeAtCompileTime == Dynamic || SizeAtCompileTime == dim);
+      EIGEN_INITIALIZE_COEFFS_IF_THAT_OPTION_IS_ENABLED
     }
-    EIGEN_DEVICE_FUNC
-    Array& operator=(Array&& other) EIGEN_NOEXCEPT_IF(std::is_nothrow_move_assignable<Scalar>::value)
-    {
-      Base::operator=(std::move(other));
-      return *this;
-    }
-#endif
-
-    #if EIGEN_HAS_CXX11
-    /** \copydoc PlainObjectBase(const Scalar& a0, const Scalar& a1, const Scalar& a2, const Scalar& a3, const ArgTypes&... args)
-     *
-     * Example: \include Array_variadic_ctor_cxx11.cpp
-     * Output: \verbinclude Array_variadic_ctor_cxx11.out
-     *
-     * \sa Array(const std::initializer_list<std::initializer_list<Scalar>>&)
-     * \sa Array(const Scalar&), Array(const Scalar&,const Scalar&)
-     */
-    template <typename... ArgTypes>
-    EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
-    Array(const Scalar& a0, const Scalar& a1, const Scalar& a2, const Scalar& a3, const ArgTypes&... args)
-      : Base(a0, a1, a2, a3, args...) {}
-
-    /** \brief Constructs an array and initializes it from the coefficients given as initializer-lists grouped by row. \cpp11
-      *
-      * In the general case, the constructor takes a list of rows, each row being represented as a list of coefficients:
-      *
-      * Example: \include Array_initializer_list_23_cxx11.cpp
-      * Output: \verbinclude Array_initializer_list_23_cxx11.out
-      *
-      * Each of the inner initializer lists must contain the exact same number of elements, otherwise an assertion is triggered.
-      *
-      * In the case of a compile-time column 1D array, implicit transposition from a single row is allowed.
-      * Therefore <code> Array<int,Dynamic,1>{{1,2,3,4,5}}</code> is legal and the more verbose syntax
-      * <code>Array<int,Dynamic,1>{{1},{2},{3},{4},{5}}</code> can be avoided:
-      *
-      * Example: \include Array_initializer_list_vector_cxx11.cpp
-      * Output: \verbinclude Array_initializer_list_vector_cxx11.out
-      *
-      * In the case of fixed-sized arrays, the initializer list sizes must exactly match the array sizes,
-      * and implicit transposition is allowed for compile-time 1D arrays only.
-      *
-      * \sa  Array(const Scalar& a0, const Scalar& a1, const Scalar& a2, const Scalar& a3, const ArgTypes&... args)
-      */
-    EIGEN_DEVICE_FUNC
-    EIGEN_STRONG_INLINE Array(const std::initializer_list<std::initializer_list<Scalar>>& list) : Base(list) {}
-    #endif // end EIGEN_HAS_CXX11
 
     #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
+    /** constructs an uninitialized matrix with \a rows rows and \a cols columns.
       *
-      * Note that this is only useful for dynamic-size vectors. For fixed-size vectors,
-      * it is redundant to pass the dimension here, so it makes more sense to use the default
-      * constructor Array() instead.
-      */
-    EIGEN_DEVICE_FUNC
-    EIGEN_STRONG_INLINE explicit Array(Index dim);
-    /** constructs an initialized 1x1 Array with the given coefficient
-      * \sa const Scalar& a0, const Scalar& a1, const Scalar& a2, const Scalar& a3, const ArgTypes&... args */
-    Array(const Scalar& value);
-    /** constructs an uninitialized array with \a rows rows and \a cols columns.
-      *
-      * This is useful for dynamic-size arrays. For fixed-size arrays,
+      * 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
-      * Array() instead. */
+      * Matrix() instead. */
     Array(Index rows, Index cols);
-    /** constructs an initialized 2D vector with given coefficients
-      * \sa Array(const Scalar& a0, const Scalar& a1, const Scalar& a2, const Scalar& a3, const ArgTypes&... args) */
+    /** constructs an initialized 2D vector with given coefficients */
     Array(const Scalar& val0, const Scalar& val1);
-    #endif  // end EIGEN_PARSED_BY_DOXYGEN
+    #endif
 
-    /** constructs an initialized 3D vector with given coefficients
-      * \sa Array(const Scalar& a0, const Scalar& a1, const Scalar& a2, const Scalar& a3, const ArgTypes&... args)
-      */
-    EIGEN_DEVICE_FUNC
+    /** constructs an initialized 3D vector with given coefficients */
     EIGEN_STRONG_INLINE Array(const Scalar& val0, const Scalar& val1, const Scalar& val2)
     {
       Base::_check_template_params();
@@ -255,10 +167,7 @@ class Array
       m_storage.data()[1] = val1;
       m_storage.data()[2] = val2;
     }
-    /** constructs an initialized 4D vector with given coefficients
-      * \sa Array(const Scalar& a0, const Scalar& a1, const Scalar& a2, const Scalar& a3, const ArgTypes&... args)
-      */
-    EIGEN_DEVICE_FUNC
+    /** constructs an initialized 4D vector with given coefficients */
     EIGEN_STRONG_INLINE Array(const Scalar& val0, const Scalar& val1, const Scalar& val2, const Scalar& val3)
     {
       Base::_check_template_params();
@@ -269,29 +178,51 @@ class Array
       m_storage.data()[3] = val3;
     }
 
-    /** Copy constructor */
-    EIGEN_DEVICE_FUNC
-    EIGEN_STRONG_INLINE Array(const Array& other)
-            : Base(other)
-    { }
+    explicit Array(const Scalar *data);
 
-  private:
-    struct PrivateType {};
-  public:
+    /** 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_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);
+    }
 
     /** \sa MatrixBase::operator=(const EigenBase<OtherDerived>&) */
     template<typename OtherDerived>
-    EIGEN_DEVICE_FUNC
-    EIGEN_STRONG_INLINE Array(const EigenBase<OtherDerived> &other,
-                              typename internal::enable_if<internal::is_convertible<typename OtherDerived::Scalar,Scalar>::value,
-                                                           PrivateType>::type = PrivateType())
-      : Base(other.derived())
-    { }
+    EIGEN_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;
+    }
 
-    EIGEN_DEVICE_FUNC EIGEN_CONSTEXPR
-    inline Index innerStride() const EIGEN_NOEXCEPT{ return 1; }
-    EIGEN_DEVICE_FUNC EIGEN_CONSTEXPR
-    inline Index outerStride() const EIGEN_NOEXCEPT { return this->innerSize(); }
+    /** 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> const & other)
+    { this->_swap(other.derived()); }
+
+    inline Index innerStride() const { return 1; }
+    inline Index outerStride() const { return this->innerSize(); }
 
     #ifdef EIGEN_ARRAY_PLUGIN
     #include EIGEN_ARRAY_PLUGIN
@@ -306,7 +237,7 @@ class Array
 /** \defgroup arraytypedefs Global array typedefs
   * \ingroup Core_Module
   *
-  * %Eigen defines several typedef shortcuts for most common 1D and 2D array types.
+  * Eigen defines several typedef shortcuts for most common 1D and 2D array types.
   *
   * The general patterns are the following:
   *
@@ -319,12 +250,6 @@ class Array
   * There are also \c ArraySizeType which are self-explanatory. For example, \c Array4cf is
   * a fixed-size 1D array of 4 complex floats.
   *
-  * With \cpp11, template alias are also defined for common sizes.
-  * They follow the same pattern as above except that the scalar type suffix is replaced by a
-  * template parameter, i.e.:
-  *   - `ArrayRowsCols<Type>` where `Rows` and `Cols` can be \c 2,\c 3,\c 4, or \c X for fixed or dynamic size.
-  *   - `ArraySize<Type>` where `Size` can be \c 2,\c 3,\c 4 or \c X for fixed or dynamic size 1D arrays.
-  *
   * \sa class Array
   */
 
@@ -357,42 +282,8 @@ EIGEN_MAKE_ARRAY_TYPEDEFS_ALL_SIZES(std::complex<double>, cd)
 
 #undef EIGEN_MAKE_ARRAY_TYPEDEFS_ALL_SIZES
 #undef EIGEN_MAKE_ARRAY_TYPEDEFS
-#undef EIGEN_MAKE_ARRAY_FIXED_TYPEDEFS
 
-#if EIGEN_HAS_CXX11
-
-#define EIGEN_MAKE_ARRAY_TYPEDEFS(Size, SizeSuffix)               \
-/** \ingroup arraytypedefs */                                     \
-/** \brief \cpp11 */                                              \
-template <typename Type>                                          \
-using Array##SizeSuffix##SizeSuffix = Array<Type, Size, Size>;    \
-/** \ingroup arraytypedefs */                                     \
-/** \brief \cpp11 */                                              \
-template <typename Type>                                          \
-using Array##SizeSuffix = Array<Type, Size, 1>;
-
-#define EIGEN_MAKE_ARRAY_FIXED_TYPEDEFS(Size)                     \
-/** \ingroup arraytypedefs */                                     \
-/** \brief \cpp11 */                                              \
-template <typename Type>                                          \
-using Array##Size##X = Array<Type, Size, Dynamic>;                \
-/** \ingroup arraytypedefs */                                     \
-/** \brief \cpp11 */                                              \
-template <typename Type>                                          \
-using Array##X##Size = Array<Type, Dynamic, Size>;
-
-EIGEN_MAKE_ARRAY_TYPEDEFS(2, 2)
-EIGEN_MAKE_ARRAY_TYPEDEFS(3, 3)
-EIGEN_MAKE_ARRAY_TYPEDEFS(4, 4)
-EIGEN_MAKE_ARRAY_TYPEDEFS(Dynamic, X)
-EIGEN_MAKE_ARRAY_FIXED_TYPEDEFS(2)
-EIGEN_MAKE_ARRAY_FIXED_TYPEDEFS(3)
-EIGEN_MAKE_ARRAY_FIXED_TYPEDEFS(4)
-
-#undef EIGEN_MAKE_ARRAY_TYPEDEFS
-#undef EIGEN_MAKE_ARRAY_FIXED_TYPEDEFS
-
-#endif // EIGEN_HAS_CXX11
+#undef EIGEN_MAKE_ARRAY_TYPEDEFS_LARGE
 
 #define EIGEN_USING_ARRAY_TYPEDEFS_FOR_TYPE_AND_SIZE(TypeSuffix, SizeSuffix) \
 using Eigen::Matrix##SizeSuffix##TypeSuffix; \

Eigen/src/Core/ArrayBase.h

@@ -32,7 +32,7 @@ template<typename ExpressionType> class MatrixWrapper;
   * \tparam Derived is the derived type, e.g., an array or an expression type.
   *
   * This class can be extended with the help of the plugin mechanism described on the page
-  * \ref TopicCustomizing_Plugins by defining the preprocessor symbol \c EIGEN_ARRAYBASE_PLUGIN.
+  * \ref TopicCustomizingEigen by defining the preprocessor symbol \c EIGEN_ARRAYBASE_PLUGIN.
   *
   * \sa class MatrixBase, \ref TopicClassHierarchy
   */
@@ -46,7 +46,11 @@ template<typename Derived> class ArrayBase
 
     typedef ArrayBase Eigen_BaseClassForSpecializationOfGlobalMathFuncImpl;
 
+    using internal::special_scalar_op_base<Derived,typename internal::traits<Derived>::Scalar,
+                typename NumTraits<typename internal::traits<Derived>::Scalar>::Real>::operator*;
+
     typedef typename internal::traits<Derived>::StorageKind StorageKind;
+    typedef typename internal::traits<Derived>::Index Index;
     typedef typename internal::traits<Derived>::Scalar Scalar;
     typedef typename internal::packet_traits<Scalar>::type PacketScalar;
     typedef typename NumTraits<Scalar>::Real RealScalar;
@@ -60,7 +64,8 @@ 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;
@@ -69,7 +74,6 @@ template<typename Derived> class ArrayBase
     using Base::coeff;
     using Base::coeffRef;
     using Base::lazyAssign;
-    using Base::operator-;
     using Base::operator=;
     using Base::operator+=;
     using Base::operator-=;
@@ -81,14 +85,26 @@ template<typename Derived> class ArrayBase
 #endif // not EIGEN_PARSED_BY_DOXYGEN
 
 #ifndef EIGEN_PARSED_BY_DOXYGEN
-    typedef typename Base::PlainObject PlainObject;
+    /** \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 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
+          > PlainObject;
+
 
     /** \internal Represents a matrix with all coefficients equal to one another*/
-    typedef CwiseNullaryOp<internal::scalar_constant_op<Scalar>,PlainObject> ConstantReturnType;
+    typedef CwiseNullaryOp<internal::scalar_constant_op<Scalar>,Derived> ConstantReturnType;
 #endif // not EIGEN_PARSED_BY_DOXYGEN
 
 #define EIGEN_CURRENT_STORAGE_BASE_CLASS Eigen::ArrayBase
-#define EIGEN_DOC_UNARY_ADDONS(X,Y)
+#   include "../plugins/CommonCwiseUnaryOps.h"
 #   include "../plugins/MatrixCwiseUnaryOps.h"
 #   include "../plugins/ArrayCwiseUnaryOps.h"
 #   include "../plugins/CommonCwiseBinaryOps.h"
@@ -98,63 +114,45 @@ template<typename Derived> class ArrayBase
 #     include EIGEN_ARRAYBASE_PLUGIN
 #   endif
 #undef EIGEN_CURRENT_STORAGE_BASE_CLASS
-#undef EIGEN_DOC_UNARY_ADDONS
 
     /** Special case of the template operator=, in order to prevent the compiler
       * from generating a default operator= (issue hit with g++ 4.1)
       */
-    EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
     Derived& operator=(const ArrayBase& other)
     {
-      internal::call_assignment(derived(), other.derived());
-      return derived();
+      return internal::assign_selector<Derived,Derived>::run(derived(), other.derived());
     }
-    
-    /** Set all the entries to \a value.
-      * \sa DenseBase::setConstant(), DenseBase::fill() */
-    EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
-    Derived& operator=(const Scalar &value)
-    { Base::setConstant(value); return derived(); }
 
-    EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
-    Derived& operator+=(const Scalar& scalar);
-    EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
-    Derived& operator-=(const Scalar& scalar);
+    Derived& operator+=(const Scalar& scalar)
+    { return *this = derived() + scalar; }
+    Derived& operator-=(const Scalar& scalar)
+    { return *this = derived() - scalar; }
 
     template<typename OtherDerived>
-    EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
     Derived& operator+=(const ArrayBase<OtherDerived>& other);
     template<typename OtherDerived>
-    EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
     Derived& operator-=(const ArrayBase<OtherDerived>& other);
 
     template<typename OtherDerived>
-    EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
     Derived& operator*=(const ArrayBase<OtherDerived>& other);
 
     template<typename OtherDerived>
-    EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
     Derived& operator/=(const ArrayBase<OtherDerived>& other);
 
   public:
-    EIGEN_DEVICE_FUNC
     ArrayBase<Derived>& array() { return *this; }
-    EIGEN_DEVICE_FUNC
     const ArrayBase<Derived>& array() const { return *this; }
 
     /** \returns an \link Eigen::MatrixBase Matrix \endlink expression of this array
       * \sa MatrixBase::array() */
-    EIGEN_DEVICE_FUNC
-    MatrixWrapper<Derived> matrix() { return MatrixWrapper<Derived>(derived()); }
-    EIGEN_DEVICE_FUNC
-    const MatrixWrapper<const Derived> matrix() const { return MatrixWrapper<const Derived>(derived()); }
+    MatrixWrapper<Derived> matrix() { return derived(); }
+    const MatrixWrapper<const Derived> matrix() const { return derived(); }
 
 //     template<typename Dest>
 //     inline void evalTo(Dest& dst) const { dst = matrix(); }
 
   protected:
-    EIGEN_DEFAULT_COPY_CONSTRUCTOR(ArrayBase)
-    EIGEN_DEFAULT_EMPTY_CONSTRUCTOR_AND_DESTRUCTOR(ArrayBase)
+    ArrayBase() : Base() {}
 
   private:
     explicit ArrayBase(Index);
@@ -175,10 +173,11 @@ template<typename Derived> class ArrayBase
   */
 template<typename Derived>
 template<typename OtherDerived>
-EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE Derived &
+EIGEN_STRONG_INLINE Derived &
 ArrayBase<Derived>::operator-=(const ArrayBase<OtherDerived> &other)
 {
-  call_assignment(derived(), other.derived(), internal::sub_assign_op<Scalar,typename OtherDerived::Scalar>());
+  SelfCwiseBinaryOp<internal::scalar_difference_op<Scalar>, Derived, OtherDerived> tmp(derived());
+  tmp = other.derived();
   return derived();
 }
 
@@ -188,10 +187,11 @@ ArrayBase<Derived>::operator-=(const ArrayBase<OtherDerived> &other)
   */
 template<typename Derived>
 template<typename OtherDerived>
-EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE Derived &
+EIGEN_STRONG_INLINE Derived &
 ArrayBase<Derived>::operator+=(const ArrayBase<OtherDerived>& other)
 {
-  call_assignment(derived(), other.derived(), internal::add_assign_op<Scalar,typename OtherDerived::Scalar>());
+  SelfCwiseBinaryOp<internal::scalar_sum_op<Scalar>, Derived, OtherDerived> tmp(derived());
+  tmp = other.derived();
   return derived();
 }
 
@@ -201,10 +201,11 @@ ArrayBase<Derived>::operator+=(const ArrayBase<OtherDerived>& other)
   */
 template<typename Derived>
 template<typename OtherDerived>
-EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE Derived &
+EIGEN_STRONG_INLINE Derived &
 ArrayBase<Derived>::operator*=(const ArrayBase<OtherDerived>& other)
 {
-  call_assignment(derived(), other.derived(), internal::mul_assign_op<Scalar,typename OtherDerived::Scalar>());
+  SelfCwiseBinaryOp<internal::scalar_product_op<Scalar>, Derived, OtherDerived> tmp(derived());
+  tmp = other.derived();
   return derived();
 }
 
@@ -214,10 +215,11 @@ ArrayBase<Derived>::operator*=(const ArrayBase<OtherDerived>& other)
   */
 template<typename Derived>
 template<typename OtherDerived>
-EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE Derived &
+EIGEN_STRONG_INLINE Derived &
 ArrayBase<Derived>::operator/=(const ArrayBase<OtherDerived>& other)
 {
-  call_assignment(derived(), other.derived(), internal::div_assign_op<Scalar,typename OtherDerived::Scalar>());
+  SelfCwiseBinaryOp<internal::scalar_quotient_op<Scalar>, Derived, OtherDerived> tmp(derived());
+  tmp = other.derived();
   return derived();
 }
 

Eigen/src/Core/ArrayWrapper.h

@@ -10,7 +10,7 @@
 #ifndef EIGEN_ARRAYWRAPPER_H
 #define EIGEN_ARRAYWRAPPER_H
 
-namespace Eigen {
+namespace Eigen { 
 
 /** \class ArrayWrapper
   * \ingroup Core_Module
@@ -29,12 +29,6 @@ struct traits<ArrayWrapper<ExpressionType> >
   : public traits<typename remove_all<typename ExpressionType::Nested>::type >
 {
   typedef ArrayXpr XprKind;
-  // Let's remove NestByRefBit
-  enum {
-    Flags0 = traits<typename remove_all<typename ExpressionType::Nested>::type >::Flags,
-    LvalueBitFlag = is_lvalue<ExpressionType>::value ? LvalueBit : 0,
-    Flags = (Flags0 & ~(NestByRefBit | LvalueBit)) | LvalueBitFlag
-  };
 };
 }
 
@@ -45,7 +39,6 @@ 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 internal::conditional<
                        internal::is_lvalue<ExpressionType>::value,
@@ -53,58 +46,87 @@ class ArrayWrapper : public ArrayBase<ArrayWrapper<ExpressionType> >
                        const Scalar
                      >::type ScalarWithConstIfNotLvalue;
 
-    typedef typename internal::ref_selector<ExpressionType>::non_const_type NestedExpressionType;
+    typedef typename internal::nested<ExpressionType>::type NestedExpressionType;
 
-    using Base::coeffRef;
+    inline ArrayWrapper(ExpressionType& matrix) : m_expression(matrix) {}
 
-    EIGEN_DEVICE_FUNC
-    explicit EIGEN_STRONG_INLINE ArrayWrapper(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 EIGEN_CONSTEXPR
-    inline Index rows() const EIGEN_NOEXCEPT { return m_expression.rows(); }
-    EIGEN_DEVICE_FUNC EIGEN_CONSTEXPR
-    inline Index cols() const EIGEN_NOEXCEPT { return m_expression.cols(); }
-    EIGEN_DEVICE_FUNC EIGEN_CONSTEXPR
-    inline Index outerStride() const EIGEN_NOEXCEPT { return m_expression.outerStride(); }
-    EIGEN_DEVICE_FUNC EIGEN_CONSTEXPR
-    inline Index innerStride() const EIGEN_NOEXCEPT { return m_expression.innerStride(); }
-
-    EIGEN_DEVICE_FUNC
-    inline ScalarWithConstIfNotLvalue* data() { return m_expression.data(); }
-    EIGEN_DEVICE_FUNC
+    inline ScalarWithConstIfNotLvalue* data() { return m_expression.const_cast_derived().data(); }
     inline const Scalar* data() const { return m_expression.data(); }
 
-    EIGEN_DEVICE_FUNC
-    inline const Scalar& coeffRef(Index rowId, Index colId) const
+    inline CoeffReturnType coeff(Index rowId, Index colId) const
     {
-      return m_expression.coeffRef(rowId, colId);
+      return m_expression.coeff(rowId, colId);
+    }
+
+    inline Scalar& coeffRef(Index rowId, Index colId)
+    {
+      return m_expression.const_cast_derived().coeffRef(rowId, colId);
+    }
+
+    inline const Scalar& coeffRef(Index rowId, Index colId) const
+    {
+      return m_expression.const_cast_derived().coeffRef(rowId, colId);
+    }
+
+    inline CoeffReturnType coeff(Index index) const
+    {
+      return m_expression.coeff(index);
+    }
+
+    inline Scalar& coeffRef(Index index)
+    {
+      return m_expression.const_cast_derived().coeffRef(index);
     }
 
-    EIGEN_DEVICE_FUNC
     inline const Scalar& coeffRef(Index index) const
     {
-      return m_expression.coeffRef(index);
+      return m_expression.const_cast_derived().coeffRef(index);
+    }
+
+    template<int LoadMode>
+    inline const PacketScalar packet(Index rowId, Index colId) const
+    {
+      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>
+    inline const PacketScalar packet(Index index) const
+    {
+      return m_expression.template packet<LoadMode>(index);
+    }
+
+    template<int LoadMode>
+    inline void writePacket(Index index, const PacketScalar& val)
+    {
+      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; }
 
-    EIGEN_DEVICE_FUNC
-    const typename internal::remove_all<NestedExpressionType>::type&
-    nestedExpression() const
+    const typename internal::remove_all<NestedExpressionType>::type& 
+    nestedExpression() const 
     {
       return m_expression;
     }
 
     /** Forwards the resizing request to the nested expression
       * \sa DenseBase::resize(Index)  */
-    EIGEN_DEVICE_FUNC
-    void resize(Index newSize) { m_expression.resize(newSize); }
+    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.resize(rows,cols); }
+    void resize(Index nbRows, Index nbCols) { m_expression.const_cast_derived().resize(nbRows,nbCols); }
 
   protected:
     NestedExpressionType m_expression;
@@ -127,12 +149,6 @@ struct traits<MatrixWrapper<ExpressionType> >
  : public traits<typename remove_all<typename ExpressionType::Nested>::type >
 {
   typedef MatrixXpr XprKind;
-  // Let's remove NestByRefBit
-  enum {
-    Flags0 = traits<typename remove_all<typename ExpressionType::Nested>::type >::Flags,
-    LvalueBitFlag = is_lvalue<ExpressionType>::value ? LvalueBit : 0,
-    Flags = (Flags0 & ~(NestByRefBit | LvalueBit)) | LvalueBitFlag
-  };
 };
 }
 
@@ -143,7 +159,6 @@ 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 internal::conditional<
                        internal::is_lvalue<ExpressionType>::value,
@@ -151,54 +166,84 @@ class MatrixWrapper : public MatrixBase<MatrixWrapper<ExpressionType> >
                        const Scalar
                      >::type ScalarWithConstIfNotLvalue;
 
-    typedef typename internal::ref_selector<ExpressionType>::non_const_type NestedExpressionType;
+    typedef typename internal::nested<ExpressionType>::type NestedExpressionType;
 
-    using Base::coeffRef;
+    inline MatrixWrapper(ExpressionType& a_matrix) : m_expression(a_matrix) {}
 
-    EIGEN_DEVICE_FUNC
-    explicit inline MatrixWrapper(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 EIGEN_CONSTEXPR
-    inline Index rows() const EIGEN_NOEXCEPT { return m_expression.rows(); }
-    EIGEN_DEVICE_FUNC EIGEN_CONSTEXPR
-    inline Index cols() const EIGEN_NOEXCEPT { return m_expression.cols(); }
-    EIGEN_DEVICE_FUNC EIGEN_CONSTEXPR
-    inline Index outerStride() const EIGEN_NOEXCEPT { return m_expression.outerStride(); }
-    EIGEN_DEVICE_FUNC EIGEN_CONSTEXPR
-    inline Index innerStride() const EIGEN_NOEXCEPT { return m_expression.innerStride(); }
-
-    EIGEN_DEVICE_FUNC
-    inline ScalarWithConstIfNotLvalue* data() { return m_expression.data(); }
-    EIGEN_DEVICE_FUNC
+    inline ScalarWithConstIfNotLvalue* data() { return m_expression.const_cast_derived().data(); }
     inline const Scalar* data() const { return m_expression.data(); }
 
-    EIGEN_DEVICE_FUNC
+    inline CoeffReturnType coeff(Index rowId, Index colId) const
+    {
+      return m_expression.coeff(rowId, colId);
+    }
+
+    inline Scalar& coeffRef(Index rowId, Index colId)
+    {
+      return m_expression.const_cast_derived().coeffRef(rowId, colId);
+    }
+
     inline const Scalar& coeffRef(Index rowId, Index colId) const
     {
       return m_expression.derived().coeffRef(rowId, colId);
     }
 
-    EIGEN_DEVICE_FUNC
-    inline const Scalar& coeffRef(Index index) const
+    inline CoeffReturnType coeff(Index index) const
     {
-      return m_expression.coeffRef(index);
+      return m_expression.coeff(index);
     }
 
-    EIGEN_DEVICE_FUNC
-    const typename internal::remove_all<NestedExpressionType>::type&
-    nestedExpression() const
+    inline Scalar& coeffRef(Index index)
+    {
+      return m_expression.const_cast_derived().coeffRef(index);
+    }
+
+    inline const Scalar& coeffRef(Index index) const
+    {
+      return m_expression.const_cast_derived().coeffRef(index);
+    }
+
+    template<int LoadMode>
+    inline const PacketScalar packet(Index rowId, Index colId) const
+    {
+      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>
+    inline const PacketScalar packet(Index index) const
+    {
+      return m_expression.template packet<LoadMode>(index);
+    }
+
+    template<int LoadMode>
+    inline void writePacket(Index index, const PacketScalar& val)
+    {
+      m_expression.const_cast_derived().template writePacket<LoadMode>(index, val);
+    }
+
+    const typename internal::remove_all<NestedExpressionType>::type& 
+    nestedExpression() const 
     {
       return m_expression;
     }
 
     /** Forwards the resizing request to the nested expression
       * \sa DenseBase::resize(Index)  */
-    EIGEN_DEVICE_FUNC
-    void resize(Index newSize) { m_expression.resize(newSize); }
+    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.resize(rows,cols); }
+    void resize(Index nbRows, Index nbCols) { m_expression.const_cast_derived().resize(nbRows,nbCols); }
 
   protected:
     NestedExpressionType m_expression;

Eigen/src/Core/Assign.h

@@ -14,9 +14,474 @@
 
 namespace Eigen {
 
+namespace internal {
+
+/***************************************************************************
+* Part 1 : the logic deciding a strategy for traversal and unrolling       *
+***************************************************************************/
+
+template <typename Derived, typename OtherDerived>
+struct assign_traits
+{
+public:
+  enum {
+    DstIsAligned = Derived::Flags & AlignedBit,
+    DstHasDirectAccess = Derived::Flags & DirectAccessBit,
+    SrcIsAligned = OtherDerived::Flags & AlignedBit,
+    JointAlignment = bool(DstIsAligned) && bool(SrcIsAligned) ? Aligned : Unaligned
+  };
+
+private:
+  enum {
+    InnerSize = int(Derived::IsVectorAtCompileTime) ? int(Derived::SizeAtCompileTime)
+              : int(Derived::Flags)&RowMajorBit ? int(Derived::ColsAtCompileTime)
+              : int(Derived::RowsAtCompileTime),
+    InnerMaxSize = int(Derived::IsVectorAtCompileTime) ? int(Derived::MaxSizeAtCompileTime)
+              : int(Derived::Flags)&RowMajorBit ? int(Derived::MaxColsAtCompileTime)
+              : int(Derived::MaxRowsAtCompileTime),
+    MaxSizeAtCompileTime = Derived::SizeAtCompileTime,
+    PacketSize = packet_traits<typename Derived::Scalar>::size
+  };
+
+  enum {
+    StorageOrdersAgree = (int(Derived::IsRowMajor) == int(OtherDerived::IsRowMajor)),
+    MightVectorize = StorageOrdersAgree
+                  && (int(Derived::Flags) & int(OtherDerived::Flags) & ActualPacketAccessBit),
+    MayInnerVectorize  = MightVectorize && int(InnerSize)!=Dynamic && int(InnerSize)%int(PacketSize)==0
+                       && int(DstIsAligned) && int(SrcIsAligned),
+    MayLinearize = StorageOrdersAgree && (int(Derived::Flags) & int(OtherDerived::Flags) & LinearAccessBit),
+    MayLinearVectorize = MightVectorize && MayLinearize && DstHasDirectAccess
+                       && (DstIsAligned || MaxSizeAtCompileTime == Dynamic),
+      /* If the destination isn't aligned, we have to do runtime checks and we don't unroll,
+         so it's only good for large enough sizes. */
+    MaySliceVectorize  = MightVectorize && DstHasDirectAccess
+                       && (int(InnerMaxSize)==Dynamic || int(InnerMaxSize)>=3*PacketSize)
+      /* slice vectorization can be slow, so we only want it if the slices are big, which is
+         indicated by InnerMaxSize rather than InnerSize, think of the case of a dynamic block
+         in a fixed-size matrix */
+  };
+
+public:
+  enum {
+    Traversal = int(MayInnerVectorize)  ? int(InnerVectorizedTraversal)
+              : int(MayLinearVectorize) ? int(LinearVectorizedTraversal)
+              : int(MaySliceVectorize)  ? int(SliceVectorizedTraversal)
+              : int(MayLinearize)       ? int(LinearTraversal)
+                                        : int(DefaultTraversal),
+    Vectorized = int(Traversal) == InnerVectorizedTraversal
+              || int(Traversal) == LinearVectorizedTraversal
+              || int(Traversal) == SliceVectorizedTraversal
+  };
+
+private:
+  enum {
+    UnrollingLimit      = EIGEN_UNROLLING_LIMIT * (Vectorized ? int(PacketSize) : 1),
+    MayUnrollCompletely = int(Derived::SizeAtCompileTime) != Dynamic
+                       && int(OtherDerived::CoeffReadCost) != Dynamic
+                       && int(Derived::SizeAtCompileTime) * int(OtherDerived::CoeffReadCost) <= int(UnrollingLimit),
+    MayUnrollInner      = int(InnerSize) != Dynamic
+                       && int(OtherDerived::CoeffReadCost) != Dynamic
+                       && int(InnerSize) * int(OtherDerived::CoeffReadCost) <= int(UnrollingLimit)
+  };
+
+public:
+  enum {
+    Unrolling = (int(Traversal) == int(InnerVectorizedTraversal) || int(Traversal) == int(DefaultTraversal))
+                ? (
+                    int(MayUnrollCompletely) ? int(CompleteUnrolling)
+                  : int(MayUnrollInner)      ? int(InnerUnrolling)
+                                             : int(NoUnrolling)
+                  )
+              : int(Traversal) == int(LinearVectorizedTraversal)
+                ? ( bool(MayUnrollCompletely) && bool(DstIsAligned) ? int(CompleteUnrolling) : int(NoUnrolling) )
+              : int(Traversal) == int(LinearTraversal)
+                ? ( bool(MayUnrollCompletely) ? int(CompleteUnrolling) : int(NoUnrolling) )
+              : int(NoUnrolling)
+  };
+
+#ifdef EIGEN_DEBUG_ASSIGN
+  static void debug()
+  {
+    EIGEN_DEBUG_VAR(DstIsAligned)
+    EIGEN_DEBUG_VAR(SrcIsAligned)
+    EIGEN_DEBUG_VAR(JointAlignment)
+    EIGEN_DEBUG_VAR(InnerSize)
+    EIGEN_DEBUG_VAR(InnerMaxSize)
+    EIGEN_DEBUG_VAR(PacketSize)
+    EIGEN_DEBUG_VAR(StorageOrdersAgree)
+    EIGEN_DEBUG_VAR(MightVectorize)
+    EIGEN_DEBUG_VAR(MayLinearize)
+    EIGEN_DEBUG_VAR(MayInnerVectorize)
+    EIGEN_DEBUG_VAR(MayLinearVectorize)
+    EIGEN_DEBUG_VAR(MaySliceVectorize)
+    EIGEN_DEBUG_VAR(Traversal)
+    EIGEN_DEBUG_VAR(UnrollingLimit)
+    EIGEN_DEBUG_VAR(MayUnrollCompletely)
+    EIGEN_DEBUG_VAR(MayUnrollInner)
+    EIGEN_DEBUG_VAR(Unrolling)
+  }
+#endif
+};
+
+/***************************************************************************
+* Part 2 : meta-unrollers
+***************************************************************************/
+
+/************************
+*** Default traversal ***
+************************/
+
+template<typename Derived1, typename Derived2, int Index, int Stop>
+struct assign_DefaultTraversal_CompleteUnrolling
+{
+  enum {
+    outer = Index / Derived1::InnerSizeAtCompileTime,
+    inner = Index % Derived1::InnerSizeAtCompileTime
+  };
+
+  static EIGEN_STRONG_INLINE void run(Derived1 &dst, const Derived2 &src)
+  {
+    dst.copyCoeffByOuterInner(outer, inner, src);
+    assign_DefaultTraversal_CompleteUnrolling<Derived1, Derived2, Index+1, Stop>::run(dst, src);
+  }
+};
+
+template<typename Derived1, typename Derived2, int Stop>
+struct assign_DefaultTraversal_CompleteUnrolling<Derived1, Derived2, Stop, Stop>
+{
+  static EIGEN_STRONG_INLINE void run(Derived1 &, const Derived2 &) {}
+};
+
+template<typename Derived1, typename Derived2, int Index, int Stop>
+struct assign_DefaultTraversal_InnerUnrolling
+{
+  static EIGEN_STRONG_INLINE void run(Derived1 &dst, const Derived2 &src, typename Derived1::Index outer)
+  {
+    dst.copyCoeffByOuterInner(outer, Index, src);
+    assign_DefaultTraversal_InnerUnrolling<Derived1, Derived2, Index+1, Stop>::run(dst, src, outer);
+  }
+};
+
+template<typename Derived1, typename Derived2, int Stop>
+struct assign_DefaultTraversal_InnerUnrolling<Derived1, Derived2, Stop, Stop>
+{
+  static EIGEN_STRONG_INLINE void run(Derived1 &, const Derived2 &, typename Derived1::Index) {}
+};
+
+/***********************
+*** Linear traversal ***
+***********************/
+
+template<typename Derived1, typename Derived2, int Index, int Stop>
+struct assign_LinearTraversal_CompleteUnrolling
+{
+  static EIGEN_STRONG_INLINE void run(Derived1 &dst, const Derived2 &src)
+  {
+    dst.copyCoeff(Index, src);
+    assign_LinearTraversal_CompleteUnrolling<Derived1, Derived2, Index+1, Stop>::run(dst, src);
+  }
+};
+
+template<typename Derived1, typename Derived2, int Stop>
+struct assign_LinearTraversal_CompleteUnrolling<Derived1, Derived2, Stop, Stop>
+{
+  static EIGEN_STRONG_INLINE void run(Derived1 &, const Derived2 &) {}
+};
+
+/**************************
+*** Inner vectorization ***
+**************************/
+
+template<typename Derived1, typename Derived2, int Index, int Stop>
+struct assign_innervec_CompleteUnrolling
+{
+  enum {
+    outer = Index / Derived1::InnerSizeAtCompileTime,
+    inner = Index % Derived1::InnerSizeAtCompileTime,
+    JointAlignment = assign_traits<Derived1,Derived2>::JointAlignment
+  };
+
+  static EIGEN_STRONG_INLINE void run(Derived1 &dst, const Derived2 &src)
+  {
+    dst.template copyPacketByOuterInner<Derived2, Aligned, JointAlignment>(outer, inner, src);
+    assign_innervec_CompleteUnrolling<Derived1, Derived2,
+      Index+packet_traits<typename Derived1::Scalar>::size, Stop>::run(dst, src);
+  }
+};
+
+template<typename Derived1, typename Derived2, int Stop>
+struct assign_innervec_CompleteUnrolling<Derived1, Derived2, Stop, Stop>
+{
+  static EIGEN_STRONG_INLINE void run(Derived1 &, const Derived2 &) {}
+};
+
+template<typename Derived1, typename Derived2, int Index, int Stop>
+struct assign_innervec_InnerUnrolling
+{
+  static EIGEN_STRONG_INLINE void run(Derived1 &dst, const Derived2 &src, typename Derived1::Index outer)
+  {
+    dst.template copyPacketByOuterInner<Derived2, Aligned, Aligned>(outer, Index, src);
+    assign_innervec_InnerUnrolling<Derived1, Derived2,
+      Index+packet_traits<typename Derived1::Scalar>::size, Stop>::run(dst, src, outer);
+  }
+};
+
+template<typename Derived1, typename Derived2, int Stop>
+struct assign_innervec_InnerUnrolling<Derived1, Derived2, Stop, Stop>
+{
+  static EIGEN_STRONG_INLINE void run(Derived1 &, const Derived2 &, typename Derived1::Index) {}
+};
+
+/***************************************************************************
+* Part 3 : implementation of all cases
+***************************************************************************/
+
+template<typename Derived1, typename Derived2,
+         int Traversal = assign_traits<Derived1, Derived2>::Traversal,
+         int Unrolling = assign_traits<Derived1, Derived2>::Unrolling,
+         int Version = Specialized>
+struct assign_impl;
+
+/************************
+*** Default traversal ***
+************************/
+
+template<typename Derived1, typename Derived2, int Unrolling, int Version>
+struct assign_impl<Derived1, Derived2, InvalidTraversal, Unrolling, Version>
+{
+  static inline void run(Derived1 &, const Derived2 &) { }
+};
+
+template<typename Derived1, typename Derived2, int Version>
+struct assign_impl<Derived1, Derived2, DefaultTraversal, NoUnrolling, Version>
+{
+  typedef typename Derived1::Index Index;
+  static inline 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, int Version>
+struct assign_impl<Derived1, Derived2, DefaultTraversal, CompleteUnrolling, Version>
+{
+  static EIGEN_STRONG_INLINE void run(Derived1 &dst, const Derived2 &src)
+  {
+    assign_DefaultTraversal_CompleteUnrolling<Derived1, Derived2, 0, Derived1::SizeAtCompileTime>
+      ::run(dst, src);
+  }
+};
+
+template<typename Derived1, typename Derived2, int Version>
+struct assign_impl<Derived1, Derived2, DefaultTraversal, InnerUnrolling, Version>
+{
+  typedef typename Derived1::Index Index;
+  static EIGEN_STRONG_INLINE void run(Derived1 &dst, const Derived2 &src)
+  {
+    const Index outerSize = dst.outerSize();
+    for(Index outer = 0; outer < outerSize; ++outer)
+      assign_DefaultTraversal_InnerUnrolling<Derived1, Derived2, 0, Derived1::InnerSizeAtCompileTime>
+        ::run(dst, src, outer);
+  }
+};
+
+/***********************
+*** Linear traversal ***
+***********************/
+
+template<typename Derived1, typename Derived2, int Version>
+struct assign_impl<Derived1, Derived2, LinearTraversal, NoUnrolling, Version>
+{
+  typedef typename Derived1::Index Index;
+  static inline 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, int Version>
+struct assign_impl<Derived1, Derived2, LinearTraversal, CompleteUnrolling, Version>
+{
+  static EIGEN_STRONG_INLINE void run(Derived1 &dst, const Derived2 &src)
+  {
+    assign_LinearTraversal_CompleteUnrolling<Derived1, Derived2, 0, Derived1::SizeAtCompileTime>
+      ::run(dst, src);
+  }
+};
+
+/**************************
+*** Inner vectorization ***
+**************************/
+
+template<typename Derived1, typename Derived2, int Version>
+struct assign_impl<Derived1, Derived2, InnerVectorizedTraversal, NoUnrolling, Version>
+{
+  typedef typename Derived1::Index Index;
+  static inline void run(Derived1 &dst, const Derived2 &src)
+  {
+    const Index innerSize = dst.innerSize();
+    const Index outerSize = dst.outerSize();
+    const Index packetSize = 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, int Version>
+struct assign_impl<Derived1, Derived2, InnerVectorizedTraversal, CompleteUnrolling, Version>
+{
+  static EIGEN_STRONG_INLINE void run(Derived1 &dst, const Derived2 &src)
+  {
+    assign_innervec_CompleteUnrolling<Derived1, Derived2, 0, Derived1::SizeAtCompileTime>
+      ::run(dst, src);
+  }
+};
+
+template<typename Derived1, typename Derived2, int Version>
+struct assign_impl<Derived1, Derived2, InnerVectorizedTraversal, InnerUnrolling, Version>
+{
+  typedef typename Derived1::Index Index;
+  static EIGEN_STRONG_INLINE void run(Derived1 &dst, const Derived2 &src)
+  {
+    const Index outerSize = dst.outerSize();
+    for(Index outer = 0; outer < outerSize; ++outer)
+      assign_innervec_InnerUnrolling<Derived1, Derived2, 0, Derived1::InnerSizeAtCompileTime>
+        ::run(dst, src, outer);
+  }
+};
+
+/***************************
+*** Linear vectorization ***
+***************************/
+
+template <bool IsAligned = false>
+struct 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 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, int Version>
+struct assign_impl<Derived1, Derived2, LinearVectorizedTraversal, NoUnrolling, Version>
+{
+  typedef typename Derived1::Index Index;
+  static EIGEN_STRONG_INLINE void run(Derived1 &dst, const Derived2 &src)
+  {
+    const Index size = dst.size();
+    typedef packet_traits<typename Derived1::Scalar> PacketTraits;
+    enum {
+      packetSize = PacketTraits::size,
+      dstAlignment = PacketTraits::AlignedOnScalar ? Aligned : int(assign_traits<Derived1,Derived2>::DstIsAligned) ,
+      srcAlignment = assign_traits<Derived1,Derived2>::JointAlignment
+    };
+    const Index alignedStart = assign_traits<Derived1,Derived2>::DstIsAligned ? 0
+                             : internal::first_aligned(&dst.coeffRef(0), size);
+    const Index alignedEnd = alignedStart + ((size-alignedStart)/packetSize)*packetSize;
+
+    unaligned_assign_impl<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);
+    }
+
+    unaligned_assign_impl<>::run(src,dst,alignedEnd,size);
+  }
+};
+
+template<typename Derived1, typename Derived2, int Version>
+struct assign_impl<Derived1, Derived2, LinearVectorizedTraversal, CompleteUnrolling, Version>
+{
+  typedef typename Derived1::Index Index;
+  static EIGEN_STRONG_INLINE void run(Derived1 &dst, const Derived2 &src)
+  {
+    enum { size = Derived1::SizeAtCompileTime,
+           packetSize = packet_traits<typename Derived1::Scalar>::size,
+           alignedSize = (size/packetSize)*packetSize };
+
+    assign_innervec_CompleteUnrolling<Derived1, Derived2, 0, alignedSize>::run(dst, src);
+    assign_DefaultTraversal_CompleteUnrolling<Derived1, Derived2, alignedSize, size>::run(dst, src);
+  }
+};
+
+/**************************
+*** Slice vectorization ***
+***************************/
+
+template<typename Derived1, typename Derived2, int Version>
+struct assign_impl<Derived1, Derived2, SliceVectorizedTraversal, NoUnrolling, Version>
+{
+  typedef typename Derived1::Index Index;
+  static inline void run(Derived1 &dst, const Derived2 &src)
+  {
+    typedef packet_traits<typename Derived1::Scalar> PacketTraits;
+    enum {
+      packetSize = PacketTraits::size,
+      alignable = PacketTraits::AlignedOnScalar,
+      dstAlignment = alignable ? Aligned : int(assign_traits<Derived1,Derived2>::DstIsAligned) ,
+      srcAlignment = 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) || assign_traits<Derived1,Derived2>::DstIsAligned) ? 0
+                       : internal::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, dstAlignment, 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);
+    }
+  }
+};
+
+} // end namespace internal
+
+/***************************************************************************
+* Part 4 : implementation of DenseBase methods
+***************************************************************************/
+
 template<typename Derived>
 template<typename OtherDerived>
-EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE Derived& DenseBase<Derived>
+EIGEN_STRONG_INLINE Derived& DenseBase<Derived>
   ::lazyAssign(const DenseBase<OtherDerived>& other)
 {
   enum{
@@ -27,62 +492,90 @@ EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE Derived& DenseBase<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)
 
+#ifdef EIGEN_DEBUG_ASSIGN
+  internal::assign_traits<Derived, OtherDerived>::debug();
+#endif
   eigen_assert(rows() == other.rows() && cols() == other.cols());
-  internal::call_assignment_no_alias(derived(),other.derived());
-  
+  internal::assign_impl<Derived, OtherDerived, int(SameType) ? int(internal::assign_traits<Derived, OtherDerived>::Traversal)
+                                                       : int(InvalidTraversal)>::run(derived(),other.derived());
+#ifndef EIGEN_NO_DEBUG
+  checkTransposeAliasing(other.derived());
+#endif
   return derived();
 }
 
+namespace internal {
+
+template<typename Derived, typename OtherDerived,
+         bool EvalBeforeAssigning = (int(internal::traits<OtherDerived>::Flags) & EvalBeforeAssigningBit) != 0,
+         bool NeedToTranspose = ((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 assign_selector;
+
+template<typename Derived, typename OtherDerived>
+struct assign_selector<Derived,OtherDerived,false,false> {
+  static EIGEN_STRONG_INLINE Derived& run(Derived& dst, const OtherDerived& other) { return dst.lazyAssign(other.derived()); }
+  template<typename ActualDerived, typename ActualOtherDerived>
+  static EIGEN_STRONG_INLINE Derived& evalTo(ActualDerived& dst, const ActualOtherDerived& other) { other.evalTo(dst); return dst; }
+};
+template<typename Derived, typename OtherDerived>
+struct assign_selector<Derived,OtherDerived,true,false> {
+  static EIGEN_STRONG_INLINE Derived& run(Derived& dst, const OtherDerived& other) { return dst.lazyAssign(other.eval()); }
+};
+template<typename Derived, typename OtherDerived>
+struct assign_selector<Derived,OtherDerived,false,true> {
+  static EIGEN_STRONG_INLINE Derived& run(Derived& dst, const OtherDerived& other) { return dst.lazyAssign(other.transpose()); }
+  template<typename ActualDerived, typename ActualOtherDerived>
+  static EIGEN_STRONG_INLINE Derived& evalTo(ActualDerived& dst, const ActualOtherDerived& other) { Transpose<ActualDerived> dstTrans(dst); other.evalTo(dstTrans); return dst; }
+};
+template<typename Derived, typename OtherDerived>
+struct assign_selector<Derived,OtherDerived,true,true> {
+  static EIGEN_STRONG_INLINE Derived& run(Derived& dst, const OtherDerived& other) { return dst.lazyAssign(other.transpose().eval()); }
+};
+
+} // end namespace internal
+
 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();
+  return internal::assign_selector<Derived,OtherDerived>::run(derived(), other.derived());
 }
 
 template<typename Derived>
-EIGEN_DEVICE_FUNC
 EIGEN_STRONG_INLINE Derived& DenseBase<Derived>::operator=(const DenseBase& other)
 {
-  internal::call_assignment(derived(), other.derived());
-  return derived();
+  return internal::assign_selector<Derived,Derived>::run(derived(), other.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();
+  return internal::assign_selector<Derived,Derived>::run(derived(), other.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();
+  return internal::assign_selector<Derived,OtherDerived>::run(derived(), other.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();
+  return internal::assign_selector<Derived,OtherDerived,false>::evalTo(derived(), other.derived());
 }
 
 template<typename Derived>
 template<typename OtherDerived>
-EIGEN_DEVICE_FUNC
 EIGEN_STRONG_INLINE Derived& MatrixBase<Derived>::operator=(const ReturnByValue<OtherDerived>& other)
 {
-  other.derived().evalTo(derived());
-  return derived();
+  return internal::assign_selector<Derived,OtherDerived,false>::evalTo(derived(), other.derived());
 }
 
 } // end namespace Eigen

Eigen/src/Core/Assign_MKL.h

@@ -1,7 +1,6 @@
 /*
  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:
 
@@ -38,13 +37,17 @@ namespace Eigen {
 
 namespace internal {
 
-template<typename Dst, typename Src>
+template<typename Op> struct vml_call
+{ enum { IsSupported = 0 }; };
+
+template<typename Dst, typename Src, typename UnaryOp>
 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)
@@ -54,122 +57,165 @@ class vml_assign_traits
                     : int(Dst::MaxRowsAtCompileTime),
       MaxSizeAtCompileTime = Dst::SizeAtCompileTime,
 
-      MightEnableVml = StorageOrdersAgree && DstHasDirectAccess && SrcHasDirectAccess && Src::InnerStrideAtCompileTime==1 && Dst::InnerStrideAtCompileTime==1,
+      MightEnableVml =  vml_call<UnaryOp>::IsSupported && 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
+      LargeEnough = VmlSize==Dynamic || VmlSize>=EIGEN_MKL_VML_THRESHOLD,
+      MayEnableVml = MightEnableVml && LargeEnough,
+      MayLinearize = MayEnableVml && MightLinearize
     };
   public:
     enum {
-      EnableVml = MightEnableVml && LargeEnough,
-      Traversal = MightLinearize ? LinearTraversal : DefaultTraversal
+      Traversal = MayLinearize ? LinearVectorizedTraversal
+                : MayEnableVml ? InnerVectorizedTraversal
+                : DefaultTraversal
     };
 };
 
-#define EIGEN_PP_EXPAND(ARG) ARG
+template<typename Derived1, typename Derived2, typename UnaryOp, int Traversal, int Unrolling,
+         int VmlTraversal = vml_assign_traits<Derived1, Derived2, UnaryOp>::Traversal >
+struct vml_assign_impl
+  : assign_impl<Derived1, Eigen::CwiseUnaryOp<UnaryOp, Derived2>,Traversal,Unrolling,BuiltIn>
+{
+};
+
+template<typename Derived1, typename Derived2, typename UnaryOp, int Traversal, int Unrolling>
+struct vml_assign_impl<Derived1, Derived2, UnaryOp, Traversal, Unrolling, InnerVectorizedTraversal>
+{
+  typedef typename Derived1::Scalar Scalar;
+  typedef typename Derived1::Index Index;
+  static inline void run(Derived1& dst, const CwiseUnaryOp<UnaryOp, Derived2>& src)
+  {
+    // in case we want to (or have to) skip VML at runtime we can call:
+    // assign_impl<Derived1,Eigen::CwiseUnaryOp<UnaryOp, Derived2>,Traversal,Unrolling,BuiltIn>::run(dst,src);
+    const Index innerSize = dst.innerSize();
+    const Index outerSize = dst.outerSize();
+    for(Index outer = 0; outer < outerSize; ++outer) {
+      const Scalar *src_ptr = src.IsRowMajor ?  &(src.nestedExpression().coeffRef(outer,0)) :
+                                                &(src.nestedExpression().coeffRef(0, outer));
+      Scalar *dst_ptr = dst.IsRowMajor ? &(dst.coeffRef(outer,0)) : &(dst.coeffRef(0, outer));
+      vml_call<UnaryOp>::run(src.functor(), innerSize, src_ptr, dst_ptr );
+    }
+  }
+};
+
+template<typename Derived1, typename Derived2, typename UnaryOp, int Traversal, int Unrolling>
+struct vml_assign_impl<Derived1, Derived2, UnaryOp, Traversal, Unrolling, LinearVectorizedTraversal>
+{
+  static inline void run(Derived1& dst, const CwiseUnaryOp<UnaryOp, Derived2>& src)
+  {
+    // in case we want to (or have to) skip VML at runtime we can call:
+    // assign_impl<Derived1,Eigen::CwiseUnaryOp<UnaryOp, Derived2>,Traversal,Unrolling,BuiltIn>::run(dst,src);
+    vml_call<UnaryOp>::run(src.functor(), dst.size(), src.nestedExpression().data(), dst.data() );
+  }
+};
+
+// Macroses
+
+#define EIGEN_MKL_VML_SPECIALIZE_ASSIGN(TRAVERSAL,UNROLLING) \
+  template<typename Derived1, typename Derived2, typename UnaryOp> \
+  struct assign_impl<Derived1, Eigen::CwiseUnaryOp<UnaryOp, Derived2>, TRAVERSAL, UNROLLING, Specialized>  {  \
+    static inline void run(Derived1 &dst, const Eigen::CwiseUnaryOp<UnaryOp, Derived2> &src) { \
+      vml_assign_impl<Derived1,Derived2,UnaryOp,TRAVERSAL,UNROLLING>::run(dst, src); \
+    } \
+  };
+
+EIGEN_MKL_VML_SPECIALIZE_ASSIGN(DefaultTraversal,NoUnrolling)
+EIGEN_MKL_VML_SPECIALIZE_ASSIGN(DefaultTraversal,CompleteUnrolling)
+EIGEN_MKL_VML_SPECIALIZE_ASSIGN(DefaultTraversal,InnerUnrolling)
+EIGEN_MKL_VML_SPECIALIZE_ASSIGN(LinearTraversal,NoUnrolling)
+EIGEN_MKL_VML_SPECIALIZE_ASSIGN(LinearTraversal,CompleteUnrolling)
+EIGEN_MKL_VML_SPECIALIZE_ASSIGN(InnerVectorizedTraversal,NoUnrolling)
+EIGEN_MKL_VML_SPECIALIZE_ASSIGN(InnerVectorizedTraversal,CompleteUnrolling)
+EIGEN_MKL_VML_SPECIALIZE_ASSIGN(InnerVectorizedTraversal,InnerUnrolling)
+EIGEN_MKL_VML_SPECIALIZE_ASSIGN(LinearVectorizedTraversal,CompleteUnrolling)
+EIGEN_MKL_VML_SPECIALIZE_ASSIGN(LinearVectorizedTraversal,NoUnrolling)
+EIGEN_MKL_VML_SPECIALIZE_ASSIGN(SliceVectorizedTraversal,NoUnrolling)
+
+
 #if !defined (EIGEN_FAST_MATH) || (EIGEN_FAST_MATH != 1)
-#define EIGEN_VMLMODE_EXPAND_xLA , VML_HA
+#define  EIGEN_MKL_VML_MODE VML_HA
 #else
-#define EIGEN_VMLMODE_EXPAND_xLA , VML_LA
+#define  EIGEN_MKL_VML_MODE VML_LA
 #endif
 
-#define EIGEN_VMLMODE_EXPAND_x_
-
-#define EIGEN_VMLMODE_PREFIX_xLA vm
-#define EIGEN_VMLMODE_PREFIX_x_  v
-#define EIGEN_VMLMODE_PREFIX(VMLMODE) EIGEN_CAT(EIGEN_VMLMODE_PREFIX_x,VMLMODE)
-
-#define EIGEN_MKL_VML_DECLARE_UNARY_CALL(EIGENOP, VMLOP, EIGENTYPE, VMLTYPE, VMLMODE)                                           \
-  template< typename DstXprType, typename SrcXprNested>                                                                         \
-  struct Assignment<DstXprType, CwiseUnaryOp<scalar_##EIGENOP##_op<EIGENTYPE>, SrcXprNested>, assign_op<EIGENTYPE,EIGENTYPE>,   \
-                   Dense2Dense, typename enable_if<vml_assign_traits<DstXprType,SrcXprNested>::EnableVml>::type> {              \
-    typedef CwiseUnaryOp<scalar_##EIGENOP##_op<EIGENTYPE>, SrcXprNested> SrcXprType;                                            \
-    static void run(DstXprType &dst, const SrcXprType &src, const assign_op<EIGENTYPE,EIGENTYPE> &func) {                       \
-      resize_if_allowed(dst, src, func);                                                                                        \
-      eigen_assert(dst.rows() == src.rows() && dst.cols() == src.cols());                                                       \
-      if(vml_assign_traits<DstXprType,SrcXprNested>::Traversal==LinearTraversal) {                                              \
-        VMLOP(dst.size(), (const VMLTYPE*)src.nestedExpression().data(),                                                        \
-              (VMLTYPE*)dst.data() EIGEN_PP_EXPAND(EIGEN_VMLMODE_EXPAND_x##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_x##VMLMODE));                                             \
-        }                                                                                                                       \
-      }                                                                                                                         \
-    }                                                                                                                           \
-  };                                                                                                                            \
-
-
-#define EIGEN_MKL_VML_DECLARE_UNARY_CALLS_REAL(EIGENOP, VMLOP, VMLMODE)                                                         \
-  EIGEN_MKL_VML_DECLARE_UNARY_CALL(EIGENOP, EIGEN_CAT(EIGEN_VMLMODE_PREFIX(VMLMODE),s##VMLOP), float, float, VMLMODE)           \
-  EIGEN_MKL_VML_DECLARE_UNARY_CALL(EIGENOP, EIGEN_CAT(EIGEN_VMLMODE_PREFIX(VMLMODE),d##VMLOP), double, double, VMLMODE)
-
-#define EIGEN_MKL_VML_DECLARE_UNARY_CALLS_CPLX(EIGENOP, VMLOP, VMLMODE)                                                         \
-  EIGEN_MKL_VML_DECLARE_UNARY_CALL(EIGENOP, EIGEN_CAT(EIGEN_VMLMODE_PREFIX(VMLMODE),c##VMLOP), scomplex, MKL_Complex8, VMLMODE) \
-  EIGEN_MKL_VML_DECLARE_UNARY_CALL(EIGENOP, EIGEN_CAT(EIGEN_VMLMODE_PREFIX(VMLMODE),z##VMLOP), dcomplex, MKL_Complex16, VMLMODE)
-  
-#define EIGEN_MKL_VML_DECLARE_UNARY_CALLS(EIGENOP, VMLOP, VMLMODE)                                                              \
-  EIGEN_MKL_VML_DECLARE_UNARY_CALLS_REAL(EIGENOP, VMLOP, VMLMODE)                                                               \
-  EIGEN_MKL_VML_DECLARE_UNARY_CALLS_CPLX(EIGENOP, VMLOP, VMLMODE)
-
-  
-EIGEN_MKL_VML_DECLARE_UNARY_CALLS(sin,   Sin,   LA)
-EIGEN_MKL_VML_DECLARE_UNARY_CALLS(asin,  Asin,  LA)
-EIGEN_MKL_VML_DECLARE_UNARY_CALLS(sinh,  Sinh,  LA)
-EIGEN_MKL_VML_DECLARE_UNARY_CALLS(cos,   Cos,   LA)
-EIGEN_MKL_VML_DECLARE_UNARY_CALLS(acos,  Acos,  LA)
-EIGEN_MKL_VML_DECLARE_UNARY_CALLS(cosh,  Cosh,  LA)
-EIGEN_MKL_VML_DECLARE_UNARY_CALLS(tan,   Tan,   LA)
-EIGEN_MKL_VML_DECLARE_UNARY_CALLS(atan,  Atan,  LA)
-EIGEN_MKL_VML_DECLARE_UNARY_CALLS(tanh,  Tanh,  LA)
-// EIGEN_MKL_VML_DECLARE_UNARY_CALLS(abs,   Abs,    _)
-EIGEN_MKL_VML_DECLARE_UNARY_CALLS(exp,   Exp,   LA)
-EIGEN_MKL_VML_DECLARE_UNARY_CALLS(log,   Ln,    LA)
-EIGEN_MKL_VML_DECLARE_UNARY_CALLS(log10, Log10, LA)
-EIGEN_MKL_VML_DECLARE_UNARY_CALLS(sqrt,  Sqrt,  _)
-
-EIGEN_MKL_VML_DECLARE_UNARY_CALLS_REAL(square, Sqr,   _)
-EIGEN_MKL_VML_DECLARE_UNARY_CALLS_CPLX(arg, Arg,      _)
-EIGEN_MKL_VML_DECLARE_UNARY_CALLS_REAL(round, Round,  _)
-EIGEN_MKL_VML_DECLARE_UNARY_CALLS_REAL(floor, Floor,  _)
-EIGEN_MKL_VML_DECLARE_UNARY_CALLS_REAL(ceil,  Ceil,   _)
-
-#define EIGEN_MKL_VML_DECLARE_POW_CALL(EIGENOP, VMLOP, EIGENTYPE, VMLTYPE, VMLMODE)                                           \
-  template< typename DstXprType, typename SrcXprNested, typename Plain>                                                       \
-  struct Assignment<DstXprType, CwiseBinaryOp<scalar_##EIGENOP##_op<EIGENTYPE,EIGENTYPE>, SrcXprNested,                       \
-                    const CwiseNullaryOp<internal::scalar_constant_op<EIGENTYPE>,Plain> >, assign_op<EIGENTYPE,EIGENTYPE>,    \
-                   Dense2Dense, typename enable_if<vml_assign_traits<DstXprType,SrcXprNested>::EnableVml>::type> {            \
-    typedef CwiseBinaryOp<scalar_##EIGENOP##_op<EIGENTYPE,EIGENTYPE>, SrcXprNested,                                           \
-                    const CwiseNullaryOp<internal::scalar_constant_op<EIGENTYPE>,Plain> > SrcXprType;                         \
-    static void run(DstXprType &dst, const SrcXprType &src, const assign_op<EIGENTYPE,EIGENTYPE> &func) {                     \
-      resize_if_allowed(dst, src, func);                                                                                      \
-      eigen_assert(dst.rows() == src.rows() && dst.cols() == src.cols());                                                     \
-      VMLTYPE exponent = reinterpret_cast<const VMLTYPE&>(src.rhs().functor().m_other);                                       \
-      if(vml_assign_traits<DstXprType,SrcXprNested>::Traversal==LinearTraversal)                                              \
-      {                                                                                                                       \
-        VMLOP( dst.size(), (const VMLTYPE*)src.lhs().data(), exponent,                                                        \
-              (VMLTYPE*)dst.data() EIGEN_PP_EXPAND(EIGEN_VMLMODE_EXPAND_x##VMLMODE) );                                         \
-      } else {                                                                                                                \
-        const Index outerSize = dst.outerSize();                                                                              \
-        for(Index outer = 0; outer < outerSize; ++outer) {                                                                    \
-          const EIGENTYPE *src_ptr = src.IsRowMajor ? &(src.lhs().coeffRef(outer,0)) :                                        \
-                                                      &(src.lhs().coeffRef(0, outer));                                        \
-          EIGENTYPE *dst_ptr = dst.IsRowMajor ? &(dst.coeffRef(outer,0)) : &(dst.coeffRef(0, outer));                         \
-          VMLOP( dst.innerSize(), (const VMLTYPE*)src_ptr, exponent,                                                          \
-                 (VMLTYPE*)dst_ptr EIGEN_PP_EXPAND(EIGEN_VMLMODE_EXPAND_x##VMLMODE));                                          \
-        }                                                                                                                     \
-      }                                                                                                                       \
-    }                                                                                                                         \
+#define EIGEN_MKL_VML_DECLARE_UNARY_CALL(EIGENOP, VMLOP, EIGENTYPE, VMLTYPE)     \
+  template<> struct vml_call< scalar_##EIGENOP##_op<EIGENTYPE> > {               \
+    enum { IsSupported = 1 };                                                    \
+    static inline void run( const scalar_##EIGENOP##_op<EIGENTYPE>& /*func*/,        \
+                            int size, const EIGENTYPE* src, EIGENTYPE* dst) {    \
+      VMLOP(size, (const VMLTYPE*)src, (VMLTYPE*)dst);                           \
+    }                                                                            \
   };
-  
-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)
+
+#define EIGEN_MKL_VML_DECLARE_UNARY_CALL_LA(EIGENOP, VMLOP, EIGENTYPE, VMLTYPE)  \
+  template<> struct vml_call< scalar_##EIGENOP##_op<EIGENTYPE> > {               \
+    enum { IsSupported = 1 };                                                    \
+    static inline void run( const scalar_##EIGENOP##_op<EIGENTYPE>& /*func*/,        \
+                            int size, const EIGENTYPE* src, EIGENTYPE* dst) {    \
+      MKL_INT64 vmlMode = EIGEN_MKL_VML_MODE;                                    \
+      VMLOP(size, (const VMLTYPE*)src, (VMLTYPE*)dst, vmlMode);                  \
+    }                                                                            \
+  };
+
+#define EIGEN_MKL_VML_DECLARE_POW_CALL(EIGENOP, VMLOP, EIGENTYPE, VMLTYPE)       \
+  template<> struct vml_call< scalar_##EIGENOP##_op<EIGENTYPE> > {               \
+    enum { IsSupported = 1 };                                                    \
+    static inline void run( const scalar_##EIGENOP##_op<EIGENTYPE>& func,        \
+                          int size, const EIGENTYPE* src, EIGENTYPE* dst) {      \
+      EIGENTYPE exponent = func.m_exponent;                                      \
+      MKL_INT64 vmlMode = EIGEN_MKL_VML_MODE;                                    \
+      VMLOP(&size, (const VMLTYPE*)src, (const VMLTYPE*)&exponent,               \
+                        (VMLTYPE*)dst, &vmlMode);                                \
+    }                                                                            \
+  };
+
+#define EIGEN_MKL_VML_DECLARE_UNARY_CALLS_REAL(EIGENOP, VMLOP)                   \
+  EIGEN_MKL_VML_DECLARE_UNARY_CALL(EIGENOP, vs##VMLOP, float, float)             \
+  EIGEN_MKL_VML_DECLARE_UNARY_CALL(EIGENOP, vd##VMLOP, double, double)
+
+#define EIGEN_MKL_VML_DECLARE_UNARY_CALLS_COMPLEX(EIGENOP, VMLOP)                \
+  EIGEN_MKL_VML_DECLARE_UNARY_CALL(EIGENOP, vc##VMLOP, scomplex, MKL_Complex8)   \
+  EIGEN_MKL_VML_DECLARE_UNARY_CALL(EIGENOP, vz##VMLOP, dcomplex, MKL_Complex16)
+
+#define EIGEN_MKL_VML_DECLARE_UNARY_CALLS(EIGENOP, VMLOP)                        \
+  EIGEN_MKL_VML_DECLARE_UNARY_CALLS_REAL(EIGENOP, VMLOP)                         \
+  EIGEN_MKL_VML_DECLARE_UNARY_CALLS_COMPLEX(EIGENOP, VMLOP)
+
+
+#define EIGEN_MKL_VML_DECLARE_UNARY_CALLS_REAL_LA(EIGENOP, VMLOP)                \
+  EIGEN_MKL_VML_DECLARE_UNARY_CALL_LA(EIGENOP, vms##VMLOP, float, float)         \
+  EIGEN_MKL_VML_DECLARE_UNARY_CALL_LA(EIGENOP, vmd##VMLOP, double, double)
+
+#define EIGEN_MKL_VML_DECLARE_UNARY_CALLS_COMPLEX_LA(EIGENOP, VMLOP)             \
+  EIGEN_MKL_VML_DECLARE_UNARY_CALL_LA(EIGENOP, vmc##VMLOP, scomplex, MKL_Complex8)  \
+  EIGEN_MKL_VML_DECLARE_UNARY_CALL_LA(EIGENOP, vmz##VMLOP, dcomplex, MKL_Complex16)
+
+#define EIGEN_MKL_VML_DECLARE_UNARY_CALLS_LA(EIGENOP, VMLOP)                     \
+  EIGEN_MKL_VML_DECLARE_UNARY_CALLS_REAL_LA(EIGENOP, VMLOP)                      \
+  EIGEN_MKL_VML_DECLARE_UNARY_CALLS_COMPLEX_LA(EIGENOP, VMLOP)
+
+
+EIGEN_MKL_VML_DECLARE_UNARY_CALLS_LA(sin,  Sin)
+EIGEN_MKL_VML_DECLARE_UNARY_CALLS_LA(asin, Asin)
+EIGEN_MKL_VML_DECLARE_UNARY_CALLS_LA(cos,  Cos)
+EIGEN_MKL_VML_DECLARE_UNARY_CALLS_LA(acos, Acos)
+EIGEN_MKL_VML_DECLARE_UNARY_CALLS_LA(tan,  Tan)
+//EIGEN_MKL_VML_DECLARE_UNARY_CALLS(abs,  Abs)
+EIGEN_MKL_VML_DECLARE_UNARY_CALLS_LA(exp,  Exp)
+EIGEN_MKL_VML_DECLARE_UNARY_CALLS_LA(log,  Ln)
+EIGEN_MKL_VML_DECLARE_UNARY_CALLS_LA(sqrt, Sqrt)
+
+EIGEN_MKL_VML_DECLARE_UNARY_CALLS_REAL(square, Sqr)
+
+// The vm*powx functions are not avaibale in the windows version of MKL.
+#ifndef _WIN32
+EIGEN_MKL_VML_DECLARE_POW_CALL(pow, vmspowx_, float, float)
+EIGEN_MKL_VML_DECLARE_POW_CALL(pow, vmdpowx_, double, double)
+EIGEN_MKL_VML_DECLARE_POW_CALL(pow, vmcpowx_, scomplex, MKL_Complex8)
+EIGEN_MKL_VML_DECLARE_POW_CALL(pow, vmzpowx_, dcomplex, MKL_Complex16)
+#endif
 
 } // end namespace internal
 

Eigen/src/Core/BandMatrix.h

@@ -10,7 +10,7 @@
 #ifndef EIGEN_BANDMATRIX_H
 #define EIGEN_BANDMATRIX_H
 
-namespace Eigen {
+namespace Eigen { 
 
 namespace internal {
 
@@ -32,7 +32,7 @@ class BandMatrixBase : public EigenBase<Derived>
     };
     typedef typename internal::traits<Derived>::Scalar Scalar;
     typedef Matrix<Scalar,RowsAtCompileTime,ColsAtCompileTime> DenseMatrixType;
-    typedef typename DenseMatrixType::StorageIndex StorageIndex;
+    typedef typename DenseMatrixType::Index Index;
     typedef typename internal::traits<Derived>::CoefficientsType CoefficientsType;
     typedef EigenBase<Derived> Base;
 
@@ -45,7 +45,7 @@ class BandMatrixBase : public EigenBase<Derived>
     };
 
   public:
-
+    
     using Base::derived;
     using Base::rows;
     using Base::cols;
@@ -55,10 +55,10 @@ class BandMatrixBase : public EigenBase<Derived>
 
     /** \returns the number of sub diagonals */
     inline Index subs() const { return derived().subs(); }
-
+    
     /** \returns an expression of the underlying coefficient matrix */
     inline const CoefficientsType& coeffs() const { return derived().coeffs(); }
-
+    
     /** \returns an expression of the underlying coefficient matrix */
     inline CoefficientsType& coeffs() { return derived().coeffs(); }
 
@@ -130,7 +130,7 @@ class BandMatrixBase : public EigenBase<Derived>
       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());
@@ -161,15 +161,15 @@ class BandMatrixBase : public EigenBase<Derived>
   *
   * \brief Represents a rectangular matrix with a banded storage
   *
-  * \tparam _Scalar Numeric type, i.e. float, double, int
-  * \tparam _Rows Number of rows, or \b Dynamic
-  * \tparam _Cols Number of columns, or \b Dynamic
-  * \tparam _Supers Number of super diagonal
-  * \tparam _Subs Number of sub diagonal
-  * \tparam _Options A combination of either \b #RowMajor or \b #ColMajor, and of \b #SelfAdjoint
-  *                  The former controls \ref TopicStorageOrders "storage order", and defaults to
-  *                  column-major. The latter controls whether the matrix represents a selfadjoint
-  *                  matrix in which case either Supers of Subs have to be null.
+  * \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
   */
@@ -179,7 +179,7 @@ struct traits<BandMatrix<_Scalar,_Rows,_Cols,_Supers,_Subs,_Options> >
 {
   typedef _Scalar Scalar;
   typedef Dense StorageKind;
-  typedef Eigen::Index StorageIndex;
+  typedef DenseIndex Index;
   enum {
     CoeffReadCost = NumTraits<Scalar>::ReadCost,
     RowsAtCompileTime = _Rows,
@@ -201,26 +201,26 @@ class BandMatrix : public BandMatrixBase<BandMatrix<_Scalar,Rows,Cols,Supers,Sub
   public:
 
     typedef typename internal::traits<BandMatrix>::Scalar Scalar;
-    typedef typename internal::traits<BandMatrix>::StorageIndex StorageIndex;
+    typedef typename internal::traits<BandMatrix>::Index Index;
     typedef typename internal::traits<BandMatrix>::CoefficientsType CoefficientsType;
 
-    explicit inline BandMatrix(Index rows=Rows, Index cols=Cols, Index supers=Supers, Index subs=Subs)
+    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 EIGEN_CONSTEXPR Index rows() const { return m_rows.value(); }
+    inline Index rows() const { return m_rows.value(); }
 
     /** \returns the number of rows */
-    inline EIGEN_CONSTEXPR Index cols() const { return m_coeffs.cols(); }
+    inline Index cols() const { return m_coeffs.cols(); }
 
     /** \returns the number of super diagonals */
-    inline EIGEN_CONSTEXPR Index supers() const { return m_supers.value(); }
+    inline Index supers() const { return m_supers.value(); }
 
     /** \returns the number of sub diagonals */
-    inline EIGEN_CONSTEXPR Index subs() const { return m_subs.value(); }
+    inline Index subs() const { return m_subs.value(); }
 
     inline const CoefficientsType& coeffs() const { return m_coeffs; }
     inline CoefficientsType& coeffs() { return m_coeffs; }
@@ -241,7 +241,7 @@ struct traits<BandMatrixWrapper<_CoefficientsType,_Rows,_Cols,_Supers,_Subs,_Opt
 {
   typedef typename _CoefficientsType::Scalar Scalar;
   typedef typename _CoefficientsType::StorageKind StorageKind;
-  typedef typename _CoefficientsType::StorageIndex StorageIndex;
+  typedef typename _CoefficientsType::Index Index;
   enum {
     CoeffReadCost = internal::traits<_CoefficientsType>::CoeffReadCost,
     RowsAtCompileTime = _Rows,
@@ -264,9 +264,9 @@ class BandMatrixWrapper : public BandMatrixBase<BandMatrixWrapper<_CoefficientsT
 
     typedef typename internal::traits<BandMatrixWrapper>::Scalar Scalar;
     typedef typename internal::traits<BandMatrixWrapper>::CoefficientsType CoefficientsType;
-    typedef typename internal::traits<BandMatrixWrapper>::StorageIndex StorageIndex;
+    typedef typename internal::traits<BandMatrixWrapper>::Index Index;
 
-    explicit inline BandMatrixWrapper(const CoefficientsType& coeffs, Index rows=_Rows, Index cols=_Cols, Index supers=_Supers, Index subs=_Subs)
+    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)
     {
@@ -275,16 +275,16 @@ class BandMatrixWrapper : public BandMatrixBase<BandMatrixWrapper<_CoefficientsT
     }
 
     /** \returns the number of columns */
-    inline EIGEN_CONSTEXPR Index rows() const { return m_rows.value(); }
+    inline Index rows() const { return m_rows.value(); }
 
     /** \returns the number of rows */
-    inline EIGEN_CONSTEXPR Index cols() const { return m_coeffs.cols(); }
+    inline Index cols() const { return m_coeffs.cols(); }
 
     /** \returns the number of super diagonals */
-    inline EIGEN_CONSTEXPR Index supers() const { return m_supers.value(); }
+    inline Index supers() const { return m_supers.value(); }
 
     /** \returns the number of sub diagonals */
-    inline EIGEN_CONSTEXPR Index subs() const { return m_subs.value(); }
+    inline Index subs() const { return m_subs.value(); }
 
     inline const CoefficientsType& coeffs() const { return m_coeffs; }
 
@@ -302,9 +302,9 @@ class BandMatrixWrapper : public BandMatrixBase<BandMatrixWrapper<_CoefficientsT
   *
   * \brief Represents a tridiagonal matrix with a compact banded storage
   *
-  * \tparam Scalar Numeric type, i.e. float, double, int
-  * \tparam Size Number of rows and cols, or \b Dynamic
-  * \tparam Options Can be 0 or \b SelfAdjoint
+  * \param _Scalar Numeric type, i.e. float, double, int
+  * \param Size Number of rows and cols, or \b Dynamic
+  * \param _Options Can be 0 or \b SelfAdjoint
   *
   * \sa class BandMatrix
   */
@@ -312,9 +312,9 @@ template<typename Scalar, int Size, int Options>
 class TridiagonalMatrix : public BandMatrix<Scalar,Size,Size,Options&SelfAdjoint?0:1,1,Options|RowMajor>
 {
     typedef BandMatrix<Scalar,Size,Size,Options&SelfAdjoint?0:1,1,Options|RowMajor> Base;
-    typedef typename Base::StorageIndex StorageIndex;
+    typedef typename Base::Index Index;
   public:
-    explicit TridiagonalMatrix(Index size = Size) : Base(size,size,Options&SelfAdjoint?0:1,1) {}
+    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>(); }
@@ -327,25 +327,6 @@ 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

Eigen/src/Core/Block.h

@@ -11,72 +11,16 @@
 #ifndef EIGEN_BLOCK_H
 #define EIGEN_BLOCK_H
 
-namespace Eigen {
-
-namespace internal {
-template<typename XprType, int BlockRows, int BlockCols, bool InnerPanel>
-struct traits<Block<XprType, BlockRows, BlockCols, InnerPanel> > : traits<XprType>
-{
-  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 = 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(traits<XprType>::MaxRowsAtCompileTime),
-    MaxColsAtCompileTime = BlockCols==0 ? 0
-                         : ColsAtCompileTime != Dynamic ? int(ColsAtCompileTime)
-                         : 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(inner_stride_at_compile_time<XprType>::ret)
-                             : int(outer_stride_at_compile_time<XprType>::ret),
-    OuterStrideAtCompileTime = HasSameStorageOrderAsXprType
-                             ? 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=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;
+namespace Eigen { 
 
 /** \class Block
   * \ingroup Core_Module
   *
   * \brief Expression of a fixed-size or dynamic-size block
   *
-  * \tparam XprType the type of the expression in which we are taking a block
-  * \tparam BlockRows the number of rows of the block we are taking at compile time (optional)
-  * \tparam BlockCols the number of columns of the block we are taking at compile time (optional)
-  * \tparam 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.
+  * \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)
   *
   * 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
@@ -100,6 +44,61 @@ template<typename XprType, int BlockRows, int BlockCols, bool InnerPanel, typena
   *
   * \sa DenseBase::block(Index,Index,Index,Index), DenseBase::block(Index,Index), class VectorBlock
   */
+
+namespace internal {
+template<typename XprType, int BlockRows, int BlockCols, bool InnerPanel>
+struct traits<Block<XprType, BlockRows, BlockCols, InnerPanel> > : traits<XprType>
+{
+  typedef typename traits<XprType>::Scalar Scalar;
+  typedef typename traits<XprType>::StorageKind StorageKind;
+  typedef typename traits<XprType>::XprKind XprKind;
+  typedef typename nested<XprType>::type XprTypeNested;
+  typedef typename remove_reference<XprTypeNested>::type _XprTypeNested;
+  enum{
+    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(traits<XprType>::MaxRowsAtCompileTime),
+    MaxColsAtCompileTime = BlockCols==0 ? 0
+                         : ColsAtCompileTime != Dynamic ? int(ColsAtCompileTime)
+                         : 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(inner_stride_at_compile_time<XprType>::ret)
+                             : int(outer_stride_at_compile_time<XprType>::ret),
+    OuterStrideAtCompileTime = HasSameStorageOrderAsXprType
+                             ? int(outer_stride_at_compile_time<XprType>::ret)
+                             : int(inner_stride_at_compile_time<XprType>::ret),
+    MaskPacketAccessBit = (InnerSize == Dynamic || (InnerSize % packet_traits<Scalar>::size) == 0)
+                       && (InnerStrideAtCompileTime == 1)
+                        ? PacketAccessBit : 0,
+    MaskAlignedBit = (InnerPanel && (OuterStrideAtCompileTime!=Dynamic) && (((OuterStrideAtCompileTime * int(sizeof(Scalar))) % 16) == 0)) ? AlignedBit : 0,
+    FlagsLinearAccessBit = (RowsAtCompileTime == 1 || ColsAtCompileTime == 1) ? LinearAccessBit : 0,
+    FlagsLvalueBit = is_lvalue<XprType>::value ? LvalueBit : 0,
+    FlagsRowMajorBit = IsRowMajor ? RowMajorBit : 0,
+    Flags0 = traits<XprType>::Flags & ( (HereditaryBits & ~RowMajorBit) |
+                                        DirectAccessBit |
+                                        MaskPacketAccessBit |
+                                        MaskAlignedBit),
+    Flags = Flags0 | FlagsLinearAccessBit | FlagsLvalueBit | FlagsRowMajorBit
+  };
+};
+
+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>
 {
@@ -109,13 +108,10 @@ template<typename XprType, int BlockRows, int BlockCols, bool InnerPanel> class
     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 EIGEN_STRONG_INLINE
-    Block(XprType& xpr, Index i) : Impl(xpr,i)
+    inline Block(XprType& xpr, Index i) : Impl(xpr,i)
     {
       eigen_assert( (i>=0) && (
           ((BlockRows==1) && (BlockCols==XprType::ColsAtCompileTime) && i<xpr.rows())
@@ -124,30 +120,28 @@ template<typename XprType, int BlockRows, int BlockCols, bool InnerPanel> class
 
     /** Fixed-size constructor
       */
-    EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
-    Block(XprType& xpr, Index startRow, Index startCol)
-      : Impl(xpr, startRow, startCol)
+    inline Block(XprType& xpr, Index a_startRow, Index a_startCol)
+      : Impl(xpr, a_startRow, a_startCol)
     {
       EIGEN_STATIC_ASSERT(RowsAtCompileTime!=Dynamic && ColsAtCompileTime!=Dynamic,THIS_METHOD_IS_ONLY_FOR_FIXED_SIZE)
-      eigen_assert(startRow >= 0 && BlockRows >= 0 && startRow + BlockRows <= xpr.rows()
-             && startCol >= 0 && BlockCols >= 0 && startCol + BlockCols <= xpr.cols());
+      eigen_assert(a_startRow >= 0 && BlockRows >= 1 && a_startRow + BlockRows <= xpr.rows()
+             && a_startCol >= 0 && BlockCols >= 1 && a_startCol + BlockCols <= xpr.cols());
     }
 
     /** Dynamic-size constructor
       */
-    EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
-    Block(XprType& xpr,
-          Index startRow, Index startCol,
+    inline Block(XprType& xpr,
+          Index a_startRow, Index a_startCol,
           Index blockRows, Index blockCols)
-      : Impl(xpr, startRow, startCol, blockRows, blockCols)
+      : Impl(xpr, a_startRow, a_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);
+      eigen_assert(a_startRow >= 0 && blockRows >= 0 && a_startRow  <= xpr.rows() - blockRows
+          && a_startCol >= 0 && blockCols >= 0 && a_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>
@@ -155,15 +149,14 @@ 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;
+    typedef typename XprType::Index Index;
   public:
     typedef Impl Base;
     EIGEN_INHERIT_ASSIGNMENT_OPERATORS(BlockImpl)
-    EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE BlockImpl(XprType& xpr, Index i) : Impl(xpr,i) {}
-    EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE BlockImpl(XprType& xpr, Index startRow, Index startCol) : Impl(xpr, startRow, startCol) {}
-    EIGEN_DEVICE_FUNC
-    EIGEN_STRONG_INLINE BlockImpl(XprType& xpr, Index startRow, Index startCol, Index blockRows, Index blockCols)
-      : Impl(xpr, startRow, startCol, blockRows, blockCols) {}
+    inline BlockImpl(XprType& xpr, Index i) : Impl(xpr,i) {}
+    inline BlockImpl(XprType& xpr, Index a_startRow, Index a_startCol) : Impl(xpr, a_startRow, a_startCol) {}
+    inline BlockImpl(XprType& xpr, Index a_startRow, Index a_startCol, Index blockRows, Index blockCols)
+      : Impl(xpr, a_startRow, a_startCol, blockRows, blockCols) {}
 };
 
 namespace internal {
@@ -173,18 +166,16 @@ template<typename XprType, int BlockRows, int BlockCols, bool InnerPanel, bool H
   : public internal::dense_xpr_base<Block<XprType, BlockRows, BlockCols, InnerPanel> >::type
 {
     typedef Block<XprType, BlockRows, BlockCols, InnerPanel> BlockType;
-    typedef typename internal::ref_selector<XprType>::non_const_type XprTypeNested;
   public:
 
     typedef typename internal::dense_xpr_base<BlockType>::type Base;
     EIGEN_DENSE_PUBLIC_INTERFACE(BlockType)
     EIGEN_INHERIT_ASSIGNMENT_OPERATORS(BlockImpl_dense)
 
-    // class InnerIterator; // FIXME apparently never used
+    class InnerIterator;
 
     /** Column or Row constructor
       */
-    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,
@@ -199,76 +190,75 @@ template<typename XprType, int BlockRows, int BlockCols, bool InnerPanel, bool H
 
     /** Fixed-size constructor
       */
-    EIGEN_DEVICE_FUNC
-    inline BlockImpl_dense(XprType& xpr, Index startRow, Index startCol)
-      : m_xpr(xpr), m_startRow(startRow), m_startCol(startCol),
+    inline BlockImpl_dense(XprType& xpr, Index a_startRow, Index a_startCol)
+      : m_xpr(xpr), m_startRow(a_startRow), m_startCol(a_startCol),
                     m_blockRows(BlockRows), m_blockCols(BlockCols)
     {}
 
     /** Dynamic-size constructor
       */
-    EIGEN_DEVICE_FUNC
     inline BlockImpl_dense(XprType& xpr,
-          Index startRow, Index startCol,
+          Index a_startRow, Index a_startCol,
           Index blockRows, Index blockCols)
-      : m_xpr(xpr), m_startRow(startRow), m_startCol(startCol),
+      : m_xpr(xpr), m_startRow(a_startRow), m_startCol(a_startCol),
                     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(); }
+    inline Index rows() const { return m_blockRows.value(); }
+    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.coeffRef(rowId + m_startRow.value(), colId + m_startCol.value());
+      return m_xpr.const_cast_derived()
+               .coeffRef(rowId + m_startRow.value(), colId + m_startCol.value());
     }
 
-    EIGEN_DEVICE_FUNC
     inline const Scalar& coeffRef(Index rowId, Index colId) const
     {
-      return m_xpr.derived().coeffRef(rowId + m_startRow.value(), colId + 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.coeffRef(m_startRow.value() + (RowsAtCompileTime == 1 ? 0 : index),
-                            m_startCol.value() + (RowsAtCompileTime == 1 ? index : 0));
+      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.coeffRef(m_startRow.value() + (RowsAtCompileTime == 1 ? 0 : index),
-                            m_startCol.value() + (RowsAtCompileTime == 1 ? index : 0));
+      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.coeff(m_startRow.value() + (RowsAtCompileTime == 1 ? 0 : index),
-                         m_startCol.value() + (RowsAtCompileTime == 1 ? index : 0));
+      return m_xpr
+             .coeff(m_startRow.value() + (RowsAtCompileTime == 1 ? 0 : index),
+                    m_startCol.value() + (RowsAtCompileTime == 1 ? index : 0));
     }
 
     template<int LoadMode>
     inline PacketScalar packet(Index rowId, Index colId) const
     {
-      return m_xpr.template packet<Unaligned>(rowId + m_startRow.value(), colId + m_startCol.value());
+      return m_xpr.template packet<Unaligned>
+              (rowId + m_startRow.value(), colId + m_startCol.value());
     }
 
     template<int LoadMode>
     inline void writePacket(Index rowId, Index colId, const PacketScalar& val)
     {
-      m_xpr.template writePacket<Unaligned>(rowId + m_startRow.value(), colId + m_startCol.value(), val);
+      m_xpr.const_cast_derived().template writePacket<Unaligned>
+              (rowId + m_startRow.value(), colId + m_startCol.value(), val);
     }
 
     template<int LoadMode>
@@ -282,46 +272,40 @@ template<typename XprType, int BlockRows, int BlockCols, bool InnerPanel, bool H
     template<int LoadMode>
     inline void writePacket(Index index, const PacketScalar& val)
     {
-      m_xpr.template writePacket<Unaligned>
+      m_xpr.const_cast_derived().template writePacket<Unaligned>
          (m_startRow.value() + (RowsAtCompileTime == 1 ? 0 : index),
           m_startCol.value() + (RowsAtCompileTime == 1 ? index : 0), val);
     }
 
     #ifdef EIGEN_PARSED_BY_DOXYGEN
     /** \sa MapBase::data() */
-    EIGEN_DEVICE_FUNC inline const Scalar* data() const;
-    EIGEN_DEVICE_FUNC inline Index innerStride() const;
-    EIGEN_DEVICE_FUNC inline Index outerStride() const;
+    inline const Scalar* data() const;
+    inline Index innerStride() const;
+    inline Index outerStride() const;
     #endif
 
-    EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
-    const typename internal::remove_all<XprTypeNested>::type& nestedExpression() const
-    {
-      return m_xpr;
+    const typename internal::remove_all<typename XprType::Nested>::type& nestedExpression() const 
+    { 
+      return m_xpr; 
     }
-
-    EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
-    XprType& nestedExpression() { return m_xpr; }
-
-    EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE EIGEN_CONSTEXPR
-    StorageIndex startRow() const EIGEN_NOEXCEPT
-    {
-      return m_startRow.value();
+      
+    Index startRow() const 
+    { 
+      return m_startRow.value(); 
     }
-
-    EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE EIGEN_CONSTEXPR
-    StorageIndex startCol() const EIGEN_NOEXCEPT
-    {
-      return m_startCol.value();
+      
+    Index startCol() const 
+    { 
+      return m_startCol.value(); 
     }
 
   protected:
 
-    XprTypeNested m_xpr;
-    const internal::variable_if_dynamic<StorageIndex, (XprType::RowsAtCompileTime == 1 && BlockRows==1) ? 0 : Dynamic> m_startRow;
-    const internal::variable_if_dynamic<StorageIndex, (XprType::ColsAtCompileTime == 1 && BlockCols==1) ? 0 : Dynamic> m_startCol;
-    const internal::variable_if_dynamic<StorageIndex, RowsAtCompileTime> m_blockRows;
-    const internal::variable_if_dynamic<StorageIndex, ColsAtCompileTime> m_blockCols;
+    const typename XprType::Nested m_xpr;
+    const internal::variable_if_dynamic<Index, XprType::RowsAtCompileTime == 1 ? 0 : Dynamic> m_startRow;
+    const internal::variable_if_dynamic<Index, XprType::ColsAtCompileTime == 1 ? 0 : Dynamic> m_startCol;
+    const internal::variable_if_dynamic<Index, RowsAtCompileTime> m_blockRows;
+    const internal::variable_if_dynamic<Index, ColsAtCompileTime> m_blockCols;
 };
 
 /** \internal Internal implementation of dense Blocks in the direct access case.*/
@@ -330,10 +314,6 @@ class BlockImpl_dense<XprType,BlockRows,BlockCols, InnerPanel,true>
   : public MapBase<Block<XprType, BlockRows, BlockCols, InnerPanel> >
 {
     typedef Block<XprType, BlockRows, BlockCols, InnerPanel> BlockType;
-    typedef typename internal::ref_selector<XprType>::non_const_type XprTypeNested;
-    enum {
-      XprTypeIsRowMajor = (int(traits<XprType>::Flags)&RowMajorBit) != 0
-    };
   public:
 
     typedef MapBase<BlockType> Base;
@@ -342,53 +322,43 @@ class BlockImpl_dense<XprType,BlockRows,BlockCols, InnerPanel,true>
 
     /** Column or Row constructor
       */
-    EIGEN_DEVICE_FUNC EIGEN_STRONG_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()),
+    inline BlockImpl_dense(XprType& xpr, Index i)
+      : Base(internal::const_cast_ptr(&xpr.coeffRef(
+              (BlockRows==1) && (BlockCols==XprType::ColsAtCompileTime) ? i : 0,
+              (BlockRows==XprType::RowsAtCompileTime) && (BlockCols==1) ? i : 0)),
              BlockRows==1 ? 1 : xpr.rows(),
              BlockCols==1 ? 1 : xpr.cols()),
-        m_xpr(xpr),
-        m_startRow( (BlockRows==1) && (BlockCols==XprType::ColsAtCompileTime) ? i : 0),
-        m_startCol( (BlockRows==XprType::RowsAtCompileTime) && (BlockCols==1) ? i : 0)
+        m_xpr(xpr)
     {
       init();
     }
 
     /** Fixed-size constructor
       */
-    EIGEN_DEVICE_FUNC EIGEN_STRONG_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), m_startRow(startRow), m_startCol(startCol)
+    inline BlockImpl_dense(XprType& xpr, Index startRow, Index startCol)
+      : Base(internal::const_cast_ptr(&xpr.coeffRef(startRow,startCol))), m_xpr(xpr)
     {
       init();
     }
 
     /** Dynamic-size constructor
       */
-    EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
-    BlockImpl_dense(XprType& xpr,
+    inline BlockImpl_dense(XprType& xpr,
           Index startRow, Index startCol,
           Index blockRows, Index blockCols)
-      : Base(xpr.data()+xpr.innerStride()*(XprTypeIsRowMajor?startCol:startRow) + xpr.outerStride()*(XprTypeIsRowMajor?startRow:startCol), blockRows, blockCols),
-        m_xpr(xpr), m_startRow(startRow), m_startCol(startCol)
+      : Base(internal::const_cast_ptr(&xpr.coeffRef(startRow,startCol)), blockRows, blockCols),
+        m_xpr(xpr)
     {
       init();
     }
 
-    EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
-    const typename internal::remove_all<XprTypeNested>::type& nestedExpression() const EIGEN_NOEXCEPT
-    {
-      return m_xpr;
+    const typename internal::remove_all<typename XprType::Nested>::type& nestedExpression() const 
+    { 
+      return m_xpr; 
     }
-
-    EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
-    XprType& nestedExpression() { return m_xpr; }
-
+      
     /** \sa MapBase::innerStride() */
-    EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE EIGEN_CONSTEXPR
-    Index innerStride() const EIGEN_NOEXCEPT
+    inline Index innerStride() const
     {
       return internal::traits<BlockType>::HasSameStorageOrderAsXprType
              ? m_xpr.innerStride()
@@ -396,20 +366,11 @@ class BlockImpl_dense<XprType,BlockRows,BlockCols, InnerPanel,true>
     }
 
     /** \sa MapBase::outerStride() */
-    EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE EIGEN_CONSTEXPR
-    Index outerStride() const EIGEN_NOEXCEPT
+    inline Index outerStride() const
     {
-      return internal::traits<BlockType>::HasSameStorageOrderAsXprType
-                    ? m_xpr.outerStride()
-                    : m_xpr.innerStride();
+      return m_outerStride;
     }
 
-    EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE EIGEN_CONSTEXPR
-    StorageIndex startRow() const EIGEN_NOEXCEPT { return m_startRow.value(); }
-
-    EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE EIGEN_CONSTEXPR
-    StorageIndex startCol() const EIGEN_NOEXCEPT { return m_startCol.value(); }
-
   #ifndef __SUNPRO_CC
   // FIXME sunstudio is not friendly with the above friend...
   // META-FIXME there is no 'friend' keyword around here. Is this obsolete?
@@ -418,8 +379,7 @@ class BlockImpl_dense<XprType,BlockRows,BlockCols, InnerPanel,true>
 
     #ifndef EIGEN_PARSED_BY_DOXYGEN
     /** \internal used by allowAligned() */
-    EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
-    BlockImpl_dense(XprType& xpr, const Scalar* data, Index blockRows, Index blockCols)
+    inline BlockImpl_dense(XprType& xpr, const Scalar* data, Index blockRows, Index blockCols)
       : Base(data, blockRows, blockCols), m_xpr(xpr)
     {
       init();
@@ -427,7 +387,6 @@ class BlockImpl_dense<XprType,BlockRows,BlockCols, InnerPanel,true>
     #endif
 
   protected:
-    EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
     void init()
     {
       m_outerStride = internal::traits<BlockType>::HasSameStorageOrderAsXprType
@@ -435,9 +394,7 @@ class BlockImpl_dense<XprType,BlockRows,BlockCols, InnerPanel,true>
                     : m_xpr.innerStride();
     }
 
-    XprTypeNested m_xpr;
-    const internal::variable_if_dynamic<StorageIndex, (XprType::RowsAtCompileTime == 1 && BlockRows==1) ? 0 : Dynamic> m_startRow;
-    const internal::variable_if_dynamic<StorageIndex, (XprType::ColsAtCompileTime == 1 && BlockCols==1) ? 0 : Dynamic> m_startCol;
+    typename XprType::Nested m_xpr;
     Index m_outerStride;
 };
 

Eigen/src/Core/BooleanRedux.h

@@ -14,56 +14,56 @@ namespace Eigen {
 
 namespace internal {
 
-template<typename Derived, int UnrollCount, int Rows>
+template<typename Derived, int UnrollCount>
 struct all_unroller
 {
   enum {
-    col = (UnrollCount-1) / Rows,
-    row = (UnrollCount-1) % Rows
+    col = (UnrollCount-1) / Derived::RowsAtCompileTime,
+    row = (UnrollCount-1) % Derived::RowsAtCompileTime
   };
 
-  EIGEN_DEVICE_FUNC static inline bool run(const Derived &mat)
+  static inline bool run(const Derived &mat)
   {
-    return all_unroller<Derived, UnrollCount-1, Rows>::run(mat) && mat.coeff(row, col);
+    return all_unroller<Derived, UnrollCount-1>::run(mat) && mat.coeff(row, col);
   }
 };
 
-template<typename Derived, int Rows>
-struct all_unroller<Derived, 0, Rows>
+template<typename Derived>
+struct all_unroller<Derived, 1>
 {
-  EIGEN_DEVICE_FUNC static inline bool run(const Derived &/*mat*/) { return true; }
+  static inline bool run(const Derived &mat) { return mat.coeff(0, 0); }
 };
 
-template<typename Derived, int Rows>
-struct all_unroller<Derived, Dynamic, Rows>
+template<typename Derived>
+struct all_unroller<Derived, Dynamic>
 {
-  EIGEN_DEVICE_FUNC static inline bool run(const Derived &) { return false; }
+  static inline bool run(const Derived &) { return false; }
 };
 
-template<typename Derived, int UnrollCount, int Rows>
+template<typename Derived, int UnrollCount>
 struct any_unroller
 {
   enum {
-    col = (UnrollCount-1) / Rows,
-    row = (UnrollCount-1) % Rows
+    col = (UnrollCount-1) / Derived::RowsAtCompileTime,
+    row = (UnrollCount-1) % Derived::RowsAtCompileTime
   };
-  
-  EIGEN_DEVICE_FUNC static inline bool run(const Derived &mat)
+
+  static inline bool run(const Derived &mat)
   {
-    return any_unroller<Derived, UnrollCount-1, Rows>::run(mat) || mat.coeff(row, col);
+    return any_unroller<Derived, UnrollCount-1>::run(mat) || mat.coeff(row, col);
   }
 };
 
-template<typename Derived, int Rows>
-struct any_unroller<Derived, 0, Rows>
+template<typename Derived>
+struct any_unroller<Derived, 1>
 {
-  EIGEN_DEVICE_FUNC static inline bool run(const Derived & /*mat*/) { return false; }
+  static inline bool run(const Derived &mat) { return mat.coeff(0, 0); }
 };
 
-template<typename Derived, int Rows>
-struct any_unroller<Derived, Dynamic, Rows>
+template<typename Derived>
+struct any_unroller<Derived, Dynamic>
 {
-  EIGEN_DEVICE_FUNC static inline bool run(const Derived &) { return false; }
+  static inline bool run(const Derived &) { return false; }
 };
 
 } // end namespace internal
@@ -76,21 +76,21 @@ struct any_unroller<Derived, Dynamic, Rows>
   * \sa any(), Cwise::operator<()
   */
 template<typename Derived>
-EIGEN_DEVICE_FUNC inline bool DenseBase<Derived>::all() const
+inline bool DenseBase<Derived>::all() const
 {
-  typedef internal::evaluator<Derived> Evaluator;
   enum {
     unroll = SizeAtCompileTime != Dynamic
-          && SizeAtCompileTime * (Evaluator::CoeffReadCost + NumTraits<Scalar>::AddCost) <= EIGEN_UNROLLING_LIMIT
+          && CoeffReadCost != Dynamic
+          && NumTraits<Scalar>::AddCost != Dynamic
+          && SizeAtCompileTime * (CoeffReadCost + NumTraits<Scalar>::AddCost) <= EIGEN_UNROLLING_LIMIT
   };
-  Evaluator evaluator(derived());
   if(unroll)
-    return internal::all_unroller<Evaluator, unroll ? int(SizeAtCompileTime) : Dynamic, internal::traits<Derived>::RowsAtCompileTime>::run(evaluator);
+    return internal::all_unroller<Derived, unroll ? int(SizeAtCompileTime) : Dynamic>::run(derived());
   else
   {
     for(Index j = 0; j < cols(); ++j)
       for(Index i = 0; i < rows(); ++i)
-        if (!evaluator.coeff(i, j)) return false;
+        if (!coeff(i, j)) return false;
     return true;
   }
 }
@@ -100,21 +100,21 @@ EIGEN_DEVICE_FUNC inline bool DenseBase<Derived>::all() const
   * \sa all()
   */
 template<typename Derived>
-EIGEN_DEVICE_FUNC inline bool DenseBase<Derived>::any() const
+inline bool DenseBase<Derived>::any() const
 {
-  typedef internal::evaluator<Derived> Evaluator;
   enum {
     unroll = SizeAtCompileTime != Dynamic
-          && SizeAtCompileTime * (Evaluator::CoeffReadCost + NumTraits<Scalar>::AddCost) <= EIGEN_UNROLLING_LIMIT
+          && CoeffReadCost != Dynamic
+          && NumTraits<Scalar>::AddCost != Dynamic
+          && SizeAtCompileTime * (CoeffReadCost + NumTraits<Scalar>::AddCost) <= EIGEN_UNROLLING_LIMIT
   };
-  Evaluator evaluator(derived());
   if(unroll)
-    return internal::any_unroller<Evaluator, unroll ? int(SizeAtCompileTime) : Dynamic, internal::traits<Derived>::RowsAtCompileTime>::run(evaluator);
+    return internal::any_unroller<Derived, unroll ? int(SizeAtCompileTime) : Dynamic>::run(derived());
   else
   {
     for(Index j = 0; j < cols(); ++j)
       for(Index i = 0; i < rows(); ++i)
-        if (evaluator.coeff(i, j)) return true;
+        if (coeff(i, j)) return true;
     return false;
   }
 }
@@ -124,7 +124,7 @@ EIGEN_DEVICE_FUNC inline bool DenseBase<Derived>::any() const
   * \sa all(), any()
   */
 template<typename Derived>
-EIGEN_DEVICE_FUNC inline Eigen::Index DenseBase<Derived>::count() const
+inline typename DenseBase<Derived>::Index DenseBase<Derived>::count() const
 {
   return derived().template cast<bool>().template cast<Index>().sum();
 }
@@ -136,11 +136,7 @@ EIGEN_DEVICE_FUNC inline Eigen::Index DenseBase<Derived>::count() const
 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.
@@ -150,11 +146,7 @@ inline bool DenseBase<Derived>::hasNaN() const
 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

Eigen/src/Core/CMakeLists.txt

@@ -0,0 +1,10 @@
+FILE(GLOB Eigen_Core_SRCS "*.h")
+
+INSTALL(FILES
+  ${Eigen_Core_SRCS}
+  DESTINATION ${INCLUDE_INSTALL_DIR}/Eigen/src/Core COMPONENT Devel
+  )
+
+ADD_SUBDIRECTORY(products)
+ADD_SUBDIRECTORY(util)
+ADD_SUBDIRECTORY(arch)

Eigen/src/Core/CommaInitializer.h

@@ -22,46 +22,28 @@ namespace Eigen {
   * the return type of MatrixBase::operator<<, and most of the time this is the only
   * way it is used.
   *
-  * \sa \blank \ref MatrixBaseCommaInitRef "MatrixBase::operator<<", CommaInitializer::finished()
+  * \sa \ref MatrixBaseCommaInitRef "MatrixBase::operator<<", CommaInitializer::finished()
   */
 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)
   {
-    eigen_assert(m_xpr.rows() > 0 && m_xpr.cols() > 0
-      && "Cannot comma-initialize a 0x0 matrix (operator<<)");
     m_xpr.coeffRef(0,0) = s;
   }
 
   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())
   {
-    eigen_assert(m_xpr.rows() >= other.rows() && m_xpr.cols() >= other.cols()
-      && "Cannot comma-initialize a 0x0 matrix (operator<<)");
     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())
@@ -81,10 +63,11 @@ struct CommaInitializer
 
   /* inserts a matrix expression in the target matrix */
   template<typename OtherDerived>
-  EIGEN_DEVICE_FUNC
   CommaInitializer& operator,(const DenseBase<OtherDerived>& other)
   {
-    if (m_col==m_xpr.cols() && (other.cols()!=0 || other.rows()!=m_currentBlockRows))
+    if(other.cols()==0 || other.rows()==0)
+      return *this;
+    if (m_col==m_xpr.cols())
     {
       m_row+=m_currentBlockRows;
       m_col = 0;
@@ -92,22 +75,24 @@ struct CommaInitializer
       eigen_assert(m_row+m_currentBlockRows<=m_xpr.rows()
         && "Too many rows passed to comma initializer (operator<<)");
     }
-    eigen_assert((m_col + other.cols() <= m_xpr.cols())
+    eigen_assert(m_col<m_xpr.cols()
       && "Too many coefficients passed to comma initializer (operator<<)");
     eigen_assert(m_currentBlockRows==other.rows());
-    m_xpr.template block<OtherDerived::RowsAtCompileTime, OtherDerived::ColsAtCompileTime>
-                    (m_row, m_col, other.rows(), other.cols()) = other;
+    if (OtherDerived::SizeAtCompileTime != Dynamic)
+      m_xpr.template block<OtherDerived::RowsAtCompileTime != Dynamic ? OtherDerived::RowsAtCompileTime : 1,
+                              OtherDerived::ColsAtCompileTime != Dynamic ? OtherDerived::ColsAtCompileTime : 1>
+                    (m_row, m_col) = other;
+    else
+      m_xpr.block(m_row, m_col, other.rows(), other.cols()) = other;
     m_col += other.cols();
     return *this;
   }
 
-  EIGEN_DEVICE_FUNC
   inline ~CommaInitializer()
-#if defined VERIFY_RAISES_ASSERT && (!defined EIGEN_NO_ASSERTION_CHECKING) && defined EIGEN_EXCEPTIONS
-  EIGEN_EXCEPTION_SPEC(Eigen::eigen_assert_exception)
-#endif
   {
-    finished();
+    eigen_assert((m_row+m_currentBlockRows) == m_xpr.rows()
+         && m_col == m_xpr.cols()
+         && "Too few coefficients passed to comma initializer (operator<<)");
   }
 
   /** \returns the built matrix once all its coefficients have been set.
@@ -117,15 +102,9 @@ struct CommaInitializer
     * quaternion.fromRotationMatrix((Matrix3f() << axis0, axis1, axis2).finished());
     * \endcode
     */
-  EIGEN_DEVICE_FUNC
-  inline XprType& finished() {
-      eigen_assert(((m_row+m_currentBlockRows) == m_xpr.rows() || m_xpr.cols() == 0)
-           && m_col == m_xpr.cols()
-           && "Too few coefficients passed to comma initializer (operator<<)");
-      return m_xpr;
-  }
+  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
@@ -145,7 +124,7 @@ struct CommaInitializer
   * \sa CommaInitializer::finished(), class CommaInitializer
   */
 template<typename Derived>
-EIGEN_DEVICE_FUNC inline CommaInitializer<Derived> DenseBase<Derived>::operator<< (const Scalar& s)
+inline CommaInitializer<Derived> DenseBase<Derived>::operator<< (const Scalar& s)
 {
   return CommaInitializer<Derived>(*static_cast<Derived*>(this), s);
 }
@@ -153,7 +132,7 @@ EIGEN_DEVICE_FUNC inline CommaInitializer<Derived> DenseBase<Derived>::operator<
 /** \sa operator<<(const Scalar&) */
 template<typename Derived>
 template<typename OtherDerived>
-EIGEN_DEVICE_FUNC inline CommaInitializer<Derived>
+inline CommaInitializer<Derived>
 DenseBase<Derived>::operator<<(const DenseBase<OtherDerived>& other)
 {
   return CommaInitializer<Derived>(*static_cast<Derived *>(this), other);

Eigen/src/Core/ConditionEstimator.h

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

Eigen/src/Core/CoreIterators.h

@@ -1,7 +1,7 @@
 // This file is part of Eigen, a lightweight C++ template library
 // for linear algebra.
 //
-// Copyright (C) 2008-2014 Gael Guennebaud <gael.guennebaud@inria.fr>
+// Copyright (C) 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
@@ -15,118 +15,47 @@ 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
+/** \ingroup SparseCore_Module
+  * \class InnerIterator
+  * \brief An InnerIterator allows to loop over the element of a sparse (or dense) matrix or expression
+  *
+  * todo
   */
-template<typename XprType>
-class InnerIterator
+
+// generic version for dense matrix and expressions
+template<typename Derived> class DenseBase<Derived>::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; }
-  EIGEN_STRONG_INLINE InnerIterator& operator+=(Index i) { m_iter.operator+=(i); return *this; }
-  EIGEN_STRONG_INLINE InnerIterator operator+(Index i) 
-  { InnerIterator result(*this); result+=i; return result; }
-    
+  protected:
+    typedef typename Derived::Scalar Scalar;
+    typedef typename Derived::Index Index;
 
-  /// \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);
+    enum { IsRowMajor = (Derived::Flags&RowMajorBit)==RowMajorBit };
+  public:
+    EIGEN_STRONG_INLINE InnerIterator(const Derived& expr, Index outer)
+      : m_expression(expr), m_inner(0), m_outer(outer), m_end(expr.innerSize())
+    {}
+
+    EIGEN_STRONG_INLINE Scalar value() const
+    {
+      return (IsRowMajor) ? m_expression.coeff(m_outer, m_inner)
+                          : m_expression.coeff(m_inner, m_outer);
+    }
+
+    EIGEN_STRONG_INLINE InnerIterator& 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 Derived& m_expression;
+    Index m_inner;
+    const Index m_outer;
+    const Index m_end;
 };
 
-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

@@ -1,7 +1,7 @@
 // This file is part of Eigen, a lightweight C++ template library
 // for linear algebra.
 //
-// Copyright (C) 2008-2014 Gael Guennebaud <gael.guennebaud@inria.fr>
+// Copyright (C) 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
@@ -13,6 +13,26 @@
 
 namespace Eigen {
 
+/** \class CwiseBinaryOp
+  * \ingroup Core_Module
+  *
+  * \brief Generic expression where a coefficient-wise binary operator is applied to two expressions
+  *
+  * \param BinaryOp template functor implementing the operator
+  * \param Lhs the type of the left-hand side
+  * \param Rhs the type of the right-hand side
+  *
+  * This class represents an expression  where a coefficient-wise binary operator is applied to two expressions.
+  * It is the return type of binary operators, by which we mean only those binary operators where
+  * both the left-hand side and the right-hand side are Eigen expressions.
+  * For example, the return type of matrix1+matrix2 is a CwiseBinaryOp.
+  *
+  * Most of the time, this is the only way that it is used, so you typically don't have to name
+  * CwiseBinaryOp types explicitly.
+  *
+  * \sa MatrixBase::binaryExpr(const MatrixBase<OtherDerived> &,const CustomBinaryOp &) const, class CwiseUnaryOp, class CwiseNullaryOp
+  */
+
 namespace internal {
 template<typename BinaryOp, typename Lhs, typename Rhs>
 struct traits<CwiseBinaryOp<BinaryOp, Lhs, Rhs> >
@@ -32,82 +52,77 @@ struct traits<CwiseBinaryOp<BinaryOp, Lhs, Rhs> >
   // we still want to handle the case when the result type is different.
   typedef typename result_of<
                      BinaryOp(
-                       const typename Lhs::Scalar&,
-                       const typename Rhs::Scalar&
+                       typename Lhs::Scalar,
+                       typename Rhs::Scalar
                      )
                    >::type Scalar;
-  typedef typename cwise_promote_storage_type<typename traits<Lhs>::StorageKind,
-                                              typename traits<Rhs>::StorageKind,
-                                              BinaryOp>::ret StorageKind;
-  typedef typename promote_index_type<typename traits<Lhs>::StorageIndex,
-                                      typename traits<Rhs>::StorageIndex>::type StorageIndex;
+  typedef typename promote_storage_type<typename traits<Lhs>::StorageKind,
+                                           typename traits<Rhs>::StorageKind>::ret StorageKind;
+  typedef typename promote_index_type<typename traits<Lhs>::Index,
+                                         typename traits<Rhs>::Index>::type Index;
   typedef typename Lhs::Nested LhsNested;
   typedef typename Rhs::Nested RhsNested;
   typedef typename remove_reference<LhsNested>::type _LhsNested;
   typedef typename remove_reference<RhsNested>::type _RhsNested;
   enum {
-    Flags = cwise_promote_storage_order<typename traits<Lhs>::StorageKind,typename traits<Rhs>::StorageKind,_LhsNested::Flags & RowMajorBit,_RhsNested::Flags & RowMajorBit>::value
+    LhsCoeffReadCost = _LhsNested::CoeffReadCost,
+    RhsCoeffReadCost = _RhsNested::CoeffReadCost,
+    LhsFlags = _LhsNested::Flags,
+    RhsFlags = _RhsNested::Flags,
+    SameType = is_same<typename _LhsNested::Scalar,typename _RhsNested::Scalar>::value,
+    StorageOrdersAgree = (int(Lhs::Flags)&RowMajorBit)==(int(Rhs::Flags)&RowMajorBit),
+    Flags0 = (int(LhsFlags) | int(RhsFlags)) & (
+        HereditaryBits
+      | (int(LhsFlags) & int(RhsFlags) &
+           ( AlignedBit
+           | (StorageOrdersAgree ? LinearAccessBit : 0)
+           | (functor_traits<BinaryOp>::PacketAccess && StorageOrdersAgree && SameType ? PacketAccessBit : 0)
+           )
+        )
+     ),
+    Flags = (Flags0 & ~RowMajorBit) | (LhsFlags & RowMajorBit),
+    CoeffReadCost = LhsCoeffReadCost + RhsCoeffReadCost + functor_traits<BinaryOp>::Cost
   };
 };
 } // end namespace internal
 
+// 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((internal::functor_is_product_like<BINOP>::ret \
+                        ? int(internal::scalar_product_traits<LHS, RHS>::Defined) \
+                        : int(internal::is_same<LHS, RHS>::value)), \
+    YOU_MIXED_DIFFERENT_NUMERIC_TYPES__YOU_NEED_TO_USE_THE_CAST_METHOD_OF_MATRIXBASE_TO_CAST_NUMERIC_TYPES_EXPLICITLY)
+
 template<typename BinaryOp, typename Lhs, typename Rhs, typename StorageKind>
 class CwiseBinaryOpImpl;
 
-/** \class CwiseBinaryOp
-  * \ingroup Core_Module
-  *
-  * \brief Generic expression where a coefficient-wise binary operator is applied to two expressions
-  *
-  * \tparam BinaryOp template functor implementing the operator
-  * \tparam LhsType the type of the left-hand side
-  * \tparam RhsType the type of the right-hand side
-  *
-  * This class represents an expression  where a coefficient-wise binary operator is applied to two expressions.
-  * It is the return type of binary operators, by which we mean only those binary operators where
-  * both the left-hand side and the right-hand side are Eigen expressions.
-  * For example, the return type of matrix1+matrix2 is a CwiseBinaryOp.
-  *
-  * Most of the time, this is the only way that it is used, so you typically don't have to name
-  * CwiseBinaryOp types explicitly.
-  *
-  * \sa MatrixBase::binaryExpr(const MatrixBase<OtherDerived> &,const CustomBinaryOp &) const, class CwiseUnaryOp, class CwiseNullaryOp
-  */
-template<typename BinaryOp, typename LhsType, typename RhsType>
-class CwiseBinaryOp :
+template<typename BinaryOp, typename Lhs, typename Rhs>
+class CwiseBinaryOp : internal::no_assignment_operator,
   public CwiseBinaryOpImpl<
-          BinaryOp, LhsType, RhsType,
-          typename internal::cwise_promote_storage_type<typename internal::traits<LhsType>::StorageKind,
-                                                        typename internal::traits<RhsType>::StorageKind,
-                                                        BinaryOp>::ret>,
-  internal::no_assignment_operator
+          BinaryOp, Lhs, Rhs,
+          typename internal::promote_storage_type<typename internal::traits<Lhs>::StorageKind,
+                                           typename internal::traits<Rhs>::StorageKind>::ret>
 {
   public:
 
-    typedef typename internal::remove_all<BinaryOp>::type Functor;
-    typedef typename internal::remove_all<LhsType>::type Lhs;
-    typedef typename internal::remove_all<RhsType>::type Rhs;
-
     typedef typename CwiseBinaryOpImpl<
-        BinaryOp, LhsType, RhsType,
-        typename internal::cwise_promote_storage_type<typename internal::traits<LhsType>::StorageKind,
-                                                      typename internal::traits<Rhs>::StorageKind,
-                                                      BinaryOp>::ret>::Base Base;
+        BinaryOp, Lhs, Rhs,
+        typename internal::promote_storage_type<typename internal::traits<Lhs>::StorageKind,
+                                         typename internal::traits<Rhs>::StorageKind>::ret>::Base Base;
     EIGEN_GENERIC_PUBLIC_INTERFACE(CwiseBinaryOp)
 
-    typedef typename internal::ref_selector<LhsType>::type LhsNested;
-    typedef typename internal::ref_selector<RhsType>::type RhsNested;
+    typedef typename internal::nested<Lhs>::type LhsNested;
+    typedef typename internal::nested<Rhs>::type RhsNested;
     typedef typename internal::remove_reference<LhsNested>::type _LhsNested;
     typedef typename internal::remove_reference<RhsNested>::type _RhsNested;
 
-#if EIGEN_COMP_MSVC && EIGEN_HAS_CXX11
-    //Required for Visual Studio or the Copy constructor will probably not get inlined!
-    EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
-    CwiseBinaryOp(const CwiseBinaryOp<BinaryOp,LhsType,RhsType>&) = default;
-#endif
-
-    EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
-    CwiseBinaryOp(const Lhs& aLhs, const Rhs& aRhs, const BinaryOp& func = BinaryOp())
+    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);
@@ -116,25 +131,26 @@ class CwiseBinaryOp :
       eigen_assert(aLhs.rows() == aRhs.rows() && aLhs.cols() == aRhs.cols());
     }
 
-    EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE EIGEN_CONSTEXPR
-    Index rows() const EIGEN_NOEXCEPT {
+    EIGEN_STRONG_INLINE Index rows() const {
       // return the fixed size type if available to enable compile time optimizations
-      return internal::traits<typename internal::remove_all<LhsNested>::type>::RowsAtCompileTime==Dynamic ? m_rhs.rows() : m_lhs.rows();
+      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 EIGEN_CONSTEXPR
-    Index cols() const EIGEN_NOEXCEPT {
+    EIGEN_STRONG_INLINE Index cols() const {
       // return the fixed size type if available to enable compile time optimizations
-      return internal::traits<typename internal::remove_all<LhsNested>::type>::ColsAtCompileTime==Dynamic ? m_rhs.cols() : m_lhs.cols();
+      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 EIGEN_STRONG_INLINE
     const _LhsNested& lhs() const { return m_lhs; }
     /** \returns the right hand side nested expression */
-    EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
     const _RhsNested& rhs() const { return m_rhs; }
     /** \returns the functor representing the binary operation */
-    EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
     const BinaryOp& functor() const { return m_functor; }
 
   protected:
@@ -143,13 +159,41 @@ class CwiseBinaryOp :
     const BinaryOp m_functor;
 };
 
-// Generic API dispatcher
-template<typename BinaryOp, typename Lhs, typename Rhs, typename StorageKind>
-class CwiseBinaryOpImpl
-  : public internal::generic_xpr_base<CwiseBinaryOp<BinaryOp, Lhs, Rhs> >::type
+template<typename BinaryOp, typename Lhs, typename Rhs>
+class CwiseBinaryOpImpl<BinaryOp, Lhs, Rhs, Dense>
+  : public internal::dense_xpr_base<CwiseBinaryOp<BinaryOp, Lhs, Rhs> >::type
 {
-public:
-  typedef typename internal::generic_xpr_base<CwiseBinaryOp<BinaryOp, Lhs, Rhs> >::type Base;
+    typedef CwiseBinaryOp<BinaryOp, Lhs, Rhs> Derived;
+  public:
+
+    typedef typename internal::dense_xpr_base<CwiseBinaryOp<BinaryOp, Lhs, Rhs> >::type Base;
+    EIGEN_DENSE_PUBLIC_INTERFACE( Derived )
+
+    EIGEN_STRONG_INLINE const Scalar coeff(Index rowId, Index colId) const
+    {
+      return derived().functor()(derived().lhs().coeff(rowId, colId),
+                                 derived().rhs().coeff(rowId, colId));
+    }
+
+    template<int LoadMode>
+    EIGEN_STRONG_INLINE PacketScalar packet(Index rowId, Index colId) const
+    {
+      return derived().functor().packetOp(derived().lhs().template packet<LoadMode>(rowId, colId),
+                                          derived().rhs().template packet<LoadMode>(rowId, colId));
+    }
+
+    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));
+    }
 };
 
 /** replaces \c *this by \c *this - \a other.
@@ -158,10 +202,11 @@ public:
   */
 template<typename Derived>
 template<typename OtherDerived>
-EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE Derived &
+EIGEN_STRONG_INLINE Derived &
 MatrixBase<Derived>::operator-=(const MatrixBase<OtherDerived> &other)
 {
-  call_assignment(derived(), other.derived(), internal::sub_assign_op<Scalar,typename OtherDerived::Scalar>());
+  SelfCwiseBinaryOp<internal::scalar_difference_op<Scalar>, Derived, OtherDerived> tmp(derived());
+  tmp = other.derived();
   return derived();
 }
 
@@ -171,10 +216,11 @@ MatrixBase<Derived>::operator-=(const MatrixBase<OtherDerived> &other)
   */
 template<typename Derived>
 template<typename OtherDerived>
-EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE Derived &
+EIGEN_STRONG_INLINE Derived &
 MatrixBase<Derived>::operator+=(const MatrixBase<OtherDerived>& other)
 {
-  call_assignment(derived(), other.derived(), internal::add_assign_op<Scalar,typename OtherDerived::Scalar>());
+  SelfCwiseBinaryOp<internal::scalar_sum_op<Scalar>, Derived, OtherDerived> tmp(derived());
+  tmp = other.derived();
   return derived();
 }
 

Eigen/src/Core/CwiseNullaryOp.h

@@ -12,24 +12,13 @@
 
 namespace Eigen {
 
-namespace internal {
-template<typename NullaryOp, typename PlainObjectType>
-struct traits<CwiseNullaryOp<NullaryOp, PlainObjectType> > : traits<PlainObjectType>
-{
-  enum {
-    Flags = traits<PlainObjectType>::Flags & RowMajorBit
-  };
-};
-
-} // namespace internal
-
 /** \class CwiseNullaryOp
   * \ingroup Core_Module
   *
   * \brief Generic expression of a matrix where all coefficients are defined by a functor
   *
-  * \tparam NullaryOp template functor implementing the operator
-  * \tparam PlainObjectType the underlying plain matrix/array type
+  * \param NullaryOp template functor implementing the operator
+  * \param PlainObjectType the underlying plain matrix/array type
   *
   * This class represents an expression of a generic nullary operator.
   * It is the return type of the Ones(), Zero(), Constant(), Identity() and Random() methods,
@@ -38,49 +27,68 @@ struct traits<CwiseNullaryOp<NullaryOp, PlainObjectType> > : traits<PlainObjectT
   * However, if you want to write a function returning such an expression, you
   * will need to use this class.
   *
-  * The functor NullaryOp must expose one of the following method:
-    <table class="manual">
-    <tr            ><td>\c operator()() </td><td>if the procedural generation does not depend on the coefficient entries (e.g., random numbers)</td></tr>
-    <tr class="alt"><td>\c operator()(Index i)</td><td>if the procedural generation makes sense for vectors only and that it depends on the coefficient index \c i (e.g., linspace) </td></tr>
-    <tr            ><td>\c operator()(Index i,Index j)</td><td>if the procedural generation depends on the matrix coordinates \c i, \c j (e.g., to generate a checkerboard with 0 and 1)</td></tr>
-    </table>
-  * It is also possible to expose the last two operators if the generation makes sense for matrices but can be optimized for vectors.
-  *
-  * See DenseBase::NullaryExpr(Index,const CustomNullaryOp&) for an example binding
-  * C++11 random number generators.
-  *
-  * A nullary expression can also be used to implement custom sophisticated matrix manipulations
-  * that cannot be covered by the existing set of natively supported matrix manipulations.
-  * See this \ref TopicCustomizing_NullaryExpr "page" for some examples and additional explanations
-  * on the behavior of CwiseNullaryOp.
-  *
-  * \sa class CwiseUnaryOp, class CwiseBinaryOp, DenseBase::NullaryExpr
+  * \sa class CwiseUnaryOp, class CwiseBinaryOp, DenseBase::NullaryExpr()
   */
+
+namespace internal {
 template<typename NullaryOp, typename PlainObjectType>
-class CwiseNullaryOp : public internal::dense_xpr_base< CwiseNullaryOp<NullaryOp, PlainObjectType> >::type, internal::no_assignment_operator
+struct traits<CwiseNullaryOp<NullaryOp, PlainObjectType> > : traits<PlainObjectType>
+{
+  enum {
+    Flags = (traits<PlainObjectType>::Flags
+      & (  HereditaryBits
+         | (functor_has_linear_access<NullaryOp>::ret ? LinearAccessBit : 0)
+         | (functor_traits<NullaryOp>::PacketAccess ? PacketAccessBit : 0)))
+      | (functor_traits<NullaryOp>::IsRepeatable ? 0 : EvalBeforeNestingBit),
+    CoeffReadCost = functor_traits<NullaryOp>::Cost
+  };
+};
+}
+
+template<typename NullaryOp, typename PlainObjectType>
+class CwiseNullaryOp : internal::no_assignment_operator,
+  public internal::dense_xpr_base< CwiseNullaryOp<NullaryOp, PlainObjectType> >::type
 {
   public:
 
     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)
+    CwiseNullaryOp(Index nbRows, Index nbCols, const NullaryOp& func = NullaryOp())
+      : m_rows(nbRows), m_cols(nbCols), m_functor(func)
     {
-      eigen_assert(rows >= 0
-            && (RowsAtCompileTime == Dynamic || RowsAtCompileTime == rows)
-            &&  cols >= 0
-            && (ColsAtCompileTime == Dynamic || ColsAtCompileTime == cols));
+      eigen_assert(nbRows >= 0
+            && (RowsAtCompileTime == Dynamic || RowsAtCompileTime == nbRows)
+            &&  nbCols >= 0
+            && (ColsAtCompileTime == Dynamic || ColsAtCompileTime == nbCols));
     }
 
-    EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE EIGEN_CONSTEXPR
-    Index rows() const { return m_rows.value(); }
-    EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE EIGEN_CONSTEXPR
-    Index cols() const { return m_cols.value(); }
+    EIGEN_STRONG_INLINE Index rows() const { return m_rows.value(); }
+    EIGEN_STRONG_INLINE Index cols() const { return m_cols.value(); }
+
+    EIGEN_STRONG_INLINE const Scalar coeff(Index rowId, Index colId) const
+    {
+      return m_functor(rowId, colId);
+    }
+
+    template<int LoadMode>
+    EIGEN_STRONG_INLINE PacketScalar packet(Index rowId, Index colId) const
+    {
+      return m_functor.packetOp(rowId, colId);
+    }
+
+    EIGEN_STRONG_INLINE const Scalar coeff(Index index) const
+    {
+      return m_functor(index);
+    }
+
+    template<int LoadMode>
+    EIGEN_STRONG_INLINE PacketScalar packet(Index index) const
+    {
+      return m_functor.packetOp(index);
+    }
 
     /** \returns the functor representing the nullary operation */
-    EIGEN_DEVICE_FUNC
     const NullaryOp& functor() const { return m_functor; }
 
   protected:
@@ -105,15 +113,10 @@ class CwiseNullaryOp : public internal::dense_xpr_base< CwiseNullaryOp<NullaryOp
   */
 template<typename Derived>
 template<typename CustomNullaryOp>
-EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
-#ifndef EIGEN_PARSED_BY_DOXYGEN
-const CwiseNullaryOp<CustomNullaryOp,typename DenseBase<Derived>::PlainObject>
-#else
-const CwiseNullaryOp<CustomNullaryOp,PlainObject>
-#endif
+EIGEN_STRONG_INLINE const CwiseNullaryOp<CustomNullaryOp, Derived>
 DenseBase<Derived>::NullaryExpr(Index rows, Index cols, const CustomNullaryOp& func)
 {
-  return CwiseNullaryOp<CustomNullaryOp, PlainObject>(rows, cols, func);
+  return CwiseNullaryOp<CustomNullaryOp, Derived>(rows, cols, func);
 }
 
 /** \returns an expression of a matrix defined by a custom functor \a func
@@ -129,24 +132,16 @@ 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_DEVICE_FUNC EIGEN_STRONG_INLINE
-#ifndef EIGEN_PARSED_BY_DOXYGEN
-const CwiseNullaryOp<CustomNullaryOp, typename DenseBase<Derived>::PlainObject>
-#else
-const CwiseNullaryOp<CustomNullaryOp, PlainObject>
-#endif
+EIGEN_STRONG_INLINE const CwiseNullaryOp<CustomNullaryOp, Derived>
 DenseBase<Derived>::NullaryExpr(Index size, const CustomNullaryOp& func)
 {
   EIGEN_STATIC_ASSERT_VECTOR_ONLY(Derived)
-  if(RowsAtCompileTime == 1) return CwiseNullaryOp<CustomNullaryOp, PlainObject>(1, size, func);
-  else return CwiseNullaryOp<CustomNullaryOp, PlainObject>(size, 1, func);
+  if(RowsAtCompileTime == 1) return CwiseNullaryOp<CustomNullaryOp, Derived>(1, size, func);
+  else return CwiseNullaryOp<CustomNullaryOp, Derived>(size, 1, func);
 }
 
 /** \returns an expression of a matrix defined by a custom functor \a func
@@ -160,24 +155,19 @@ DenseBase<Derived>::NullaryExpr(Index size, const CustomNullaryOp& func)
   */
 template<typename Derived>
 template<typename CustomNullaryOp>
-EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
-#ifndef EIGEN_PARSED_BY_DOXYGEN
-const CwiseNullaryOp<CustomNullaryOp, typename DenseBase<Derived>::PlainObject>
-#else
-const CwiseNullaryOp<CustomNullaryOp, PlainObject>
-#endif
+EIGEN_STRONG_INLINE const CwiseNullaryOp<CustomNullaryOp, Derived>
 DenseBase<Derived>::NullaryExpr(const CustomNullaryOp& func)
 {
-  return CwiseNullaryOp<CustomNullaryOp, PlainObject>(RowsAtCompileTime, ColsAtCompileTime, func);
+  return CwiseNullaryOp<CustomNullaryOp, Derived>(RowsAtCompileTime, ColsAtCompileTime, func);
 }
 
 /** \returns an expression of a constant matrix of value \a value
   *
-  * The parameters \a rows and \a cols are the number of rows and of columns of
+  * The parameters \a nbRows and \a nbCols are the number of rows and of columns of
   * the returned matrix. Must be compatible with this DenseBase type.
   *
   * This variant is meant to be used for dynamic-size matrix types. For fixed-size types,
-  * it is redundant to pass \a rows and \a cols as arguments, so Zero() should be used
+  * it is redundant to pass \a nbRows and \a nbCols as arguments, so Zero() should be used
   * instead.
   *
   * The template parameter \a CustomNullaryOp is the type of the functor.
@@ -185,10 +175,10 @@ DenseBase<Derived>::NullaryExpr(const CustomNullaryOp& func)
   * \sa class CwiseNullaryOp
   */
 template<typename Derived>
-EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE const typename DenseBase<Derived>::ConstantReturnType
-DenseBase<Derived>::Constant(Index rows, Index cols, const Scalar& value)
+EIGEN_STRONG_INLINE const typename DenseBase<Derived>::ConstantReturnType
+DenseBase<Derived>::Constant(Index nbRows, Index nbCols, const Scalar& value)
 {
-  return DenseBase<Derived>::NullaryExpr(rows, cols, internal::scalar_constant_op<Scalar>(value));
+  return DenseBase<Derived>::NullaryExpr(nbRows, nbCols, internal::scalar_constant_op<Scalar>(value));
 }
 
 /** \returns an expression of a constant matrix of value \a value
@@ -207,7 +197,7 @@ DenseBase<Derived>::Constant(Index rows, Index cols, const Scalar& value)
   * \sa class CwiseNullaryOp
   */
 template<typename Derived>
-EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE const typename DenseBase<Derived>::ConstantReturnType
+EIGEN_STRONG_INLINE const typename DenseBase<Derived>::ConstantReturnType
 DenseBase<Derived>::Constant(Index size, const Scalar& value)
 {
   return DenseBase<Derived>::NullaryExpr(size, internal::scalar_constant_op<Scalar>(value));
@@ -223,45 +213,53 @@ DenseBase<Derived>::Constant(Index size, const Scalar& value)
   * \sa class CwiseNullaryOp
   */
 template<typename Derived>
-EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE const typename DenseBase<Derived>::ConstantReturnType
+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, internal::scalar_constant_op<Scalar>(value));
 }
 
-/** \deprecated because of accuracy loss. In Eigen 3.3, it is an alias for LinSpaced(Index,const Scalar&,const Scalar&)
+/**
+  * \brief Sets a linearly space vector.
+  *
+  * The function generates 'size' equally spaced values in the closed interval [low,high].
+  * This particular version of LinSpaced() uses sequential access, i.e. vector access is
+  * 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_deprecated.cpp
-  * Output: \verbinclude DenseBase_LinSpaced_seq_deprecated.out
+  * Example: \include DenseBase_LinSpaced_seq.cpp
+  * Output: \verbinclude DenseBase_LinSpaced_seq.out
   *
-  * \sa LinSpaced(Index,const Scalar&, const Scalar&), setLinSpaced(Index,const Scalar&,const Scalar&)
+  * \sa setLinSpaced(Index,const Scalar&,const Scalar&), LinSpaced(Index,Scalar,Scalar), CwiseNullaryOp
   */
 template<typename Derived>
-EIGEN_DEPRECATED EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE const typename DenseBase<Derived>::RandomAccessLinSpacedReturnType
+EIGEN_STRONG_INLINE const typename DenseBase<Derived>::SequentialLinSpacedReturnType
 DenseBase<Derived>::LinSpaced(Sequential_t, Index size, const Scalar& low, const Scalar& high)
 {
   EIGEN_STATIC_ASSERT_VECTOR_ONLY(Derived)
-  return DenseBase<Derived>::NullaryExpr(size, internal::linspaced_op<Scalar>(low,high,size));
+  return DenseBase<Derived>::NullaryExpr(size, internal::linspaced_op<Scalar,false>(low,high,size));
 }
 
-/** \deprecated because of accuracy loss. In Eigen 3.3, it is an alias for LinSpaced(const Scalar&,const Scalar&)
-  *
-  * \sa LinSpaced(const Scalar&, const Scalar&)
+/**
+  * \copydoc DenseBase::LinSpaced(Sequential_t, Index, const Scalar&, const Scalar&)
+  * Special version for fixed size types which does not require the size parameter.
   */
 template<typename Derived>
-EIGEN_DEPRECATED EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE const typename DenseBase<Derived>::RandomAccessLinSpacedReturnType
+EIGEN_STRONG_INLINE const typename DenseBase<Derived>::SequentialLinSpacedReturnType
 DenseBase<Derived>::LinSpaced(Sequential_t, const Scalar& low, const Scalar& high)
 {
   EIGEN_STATIC_ASSERT_VECTOR_ONLY(Derived)
   EIGEN_STATIC_ASSERT_FIXED_SIZE(Derived)
-  return DenseBase<Derived>::NullaryExpr(Derived::SizeAtCompileTime, internal::linspaced_op<Scalar>(low,high,Derived::SizeAtCompileTime));
+  return DenseBase<Derived>::NullaryExpr(Derived::SizeAtCompileTime, internal::linspaced_op<Scalar,false>(low,high,Derived::SizeAtCompileTime));
 }
 
 /**
-  * \brief Sets a linearly spaced vector.
+  * \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.
@@ -271,24 +269,14 @@ DenseBase<Derived>::LinSpaced(Sequential_t, const Scalar& low, const Scalar& hig
   * Example: \include DenseBase_LinSpaced.cpp
   * Output: \verbinclude DenseBase_LinSpaced.out
   *
-  * For integer scalar types, an even spacing is possible if and only if the length of the range,
-  * i.e., \c high-low is a scalar multiple of \c size-1, or if \c size is a scalar multiple of the
-  * number of values \c high-low+1 (meaning each value can be repeated the same number of time).
-  * If one of these two considions is not satisfied, then \c high is lowered to the largest value
-  * satisfying one of this constraint.
-  * Here are some examples:
-  *
-  * Example: \include DenseBase_LinSpacedInt.cpp
-  * Output: \verbinclude DenseBase_LinSpacedInt.out
-  *
-  * \sa setLinSpaced(Index,const Scalar&,const Scalar&), CwiseNullaryOp
+  * \sa setLinSpaced(Index,const Scalar&,const Scalar&), LinSpaced(Sequential_t,Index,const Scalar&,const Scalar&,Index), CwiseNullaryOp
   */
 template<typename Derived>
-EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE const typename DenseBase<Derived>::RandomAccessLinSpacedReturnType
+EIGEN_STRONG_INLINE const typename DenseBase<Derived>::RandomAccessLinSpacedReturnType
 DenseBase<Derived>::LinSpaced(Index size, const Scalar& low, const Scalar& high)
 {
   EIGEN_STATIC_ASSERT_VECTOR_ONLY(Derived)
-  return DenseBase<Derived>::NullaryExpr(size, internal::linspaced_op<Scalar>(low,high,size));
+  return DenseBase<Derived>::NullaryExpr(size, internal::linspaced_op<Scalar,true>(low,high,size));
 }
 
 /**
@@ -296,23 +284,22 @@ DenseBase<Derived>::LinSpaced(Index size, const Scalar& low, const Scalar& high)
   * Special version for fixed size types which does not require the size parameter.
   */
 template<typename Derived>
-EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE const typename DenseBase<Derived>::RandomAccessLinSpacedReturnType
+EIGEN_STRONG_INLINE const typename DenseBase<Derived>::RandomAccessLinSpacedReturnType
 DenseBase<Derived>::LinSpaced(const Scalar& low, const Scalar& high)
 {
   EIGEN_STATIC_ASSERT_VECTOR_ONLY(Derived)
   EIGEN_STATIC_ASSERT_FIXED_SIZE(Derived)
-  return DenseBase<Derived>::NullaryExpr(Derived::SizeAtCompileTime, internal::linspaced_op<Scalar>(low,high,Derived::SizeAtCompileTime));
+  return DenseBase<Derived>::NullaryExpr(Derived::SizeAtCompileTime, internal::linspaced_op<Scalar,true>(low,high,Derived::SizeAtCompileTime));
 }
 
 /** \returns true if all coefficients in this matrix are approximately equal to \a val, to within precision \a prec */
 template<typename Derived>
-EIGEN_DEVICE_FUNC bool DenseBase<Derived>::isApproxToConstant
+bool DenseBase<Derived>::isApproxToConstant
 (const Scalar& val, const RealScalar& prec) const
 {
-  typename internal::nested_eval<Derived,1>::type self(derived());
   for(Index j = 0; j < cols(); ++j)
     for(Index i = 0; i < rows(); ++i)
-      if(!internal::isApprox(self.coeff(i, j), val, prec))
+      if(!internal::isApprox(this->coeff(i, j), val, prec))
         return false;
   return true;
 }
@@ -321,7 +308,7 @@ EIGEN_DEVICE_FUNC 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>
-EIGEN_DEVICE_FUNC bool DenseBase<Derived>::isConstant
+bool DenseBase<Derived>::isConstant
 (const Scalar& val, const RealScalar& prec) const
 {
   return isApproxToConstant(val, prec);
@@ -332,22 +319,22 @@ EIGEN_DEVICE_FUNC bool DenseBase<Derived>::isConstant
   * \sa setConstant(), Constant(), class CwiseNullaryOp
   */
 template<typename Derived>
-EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE void DenseBase<Derived>::fill(const Scalar& val)
+EIGEN_STRONG_INLINE void DenseBase<Derived>::fill(const Scalar& val)
 {
   setConstant(val);
 }
 
-/** Sets all coefficients in this expression to value \a val.
+/** Sets all coefficients in this expression to \a value.
   *
   * \sa fill(), setConstant(Index,const Scalar&), setConstant(Index,Index,const Scalar&), setZero(), setOnes(), Constant(), class CwiseNullaryOp, setZero(), setOnes()
   */
 template<typename Derived>
-EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE Derived& DenseBase<Derived>::setConstant(const Scalar& val)
+EIGEN_STRONG_INLINE Derived& DenseBase<Derived>::setConstant(const Scalar& val)
 {
   return derived() = Constant(rows(), cols(), val);
 }
 
-/** Resizes to the given \a size, and sets all coefficients in this expression to the given value \a val.
+/** Resizes to the given \a size, and sets all coefficients in this expression to the given \a value.
   *
   * \only_for_vectors
   *
@@ -357,17 +344,17 @@ EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE Derived& DenseBase<Derived>::setConstant(c
   * \sa MatrixBase::setConstant(const Scalar&), setConstant(Index,Index,const Scalar&), class CwiseNullaryOp, MatrixBase::Constant(const Scalar&)
   */
 template<typename Derived>
-EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE Derived&
+EIGEN_STRONG_INLINE Derived&
 PlainObjectBase<Derived>::setConstant(Index size, const Scalar& val)
 {
   resize(size);
   return setConstant(val);
 }
 
-/** Resizes to the given size, and sets all coefficients in this expression to the given value \a 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 nbRows the new number of rows
+  * \param nbCols the new number of columns
   * \param val the value to which all coefficients are set
   *
   * Example: \include Matrix_setConstant_int_int.cpp
@@ -376,42 +363,15 @@ PlainObjectBase<Derived>::setConstant(Index size, const Scalar& val)
   * \sa MatrixBase::setConstant(const Scalar&), setConstant(Index,const Scalar&), class CwiseNullaryOp, MatrixBase::Constant(const Scalar&)
   */
 template<typename Derived>
-EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE Derived&
-PlainObjectBase<Derived>::setConstant(Index rows, Index cols, const Scalar& val)
+EIGEN_STRONG_INLINE Derived&
+PlainObjectBase<Derived>::setConstant(Index nbRows, Index nbCols, const Scalar& val)
 {
-  resize(rows, cols);
+  resize(nbRows, nbCols);
   return setConstant(val);
 }
 
-/** Resizes to the given size, changing only the number of columns, and sets all
-  * coefficients in this expression to the given value \a val. For the parameter
-  * of type NoChange_t, just pass the special value \c NoChange.
-  *
-  * \sa MatrixBase::setConstant(const Scalar&), setConstant(Index,const Scalar&), class CwiseNullaryOp, MatrixBase::Constant(const Scalar&)
-  */
-template<typename Derived>
-EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE Derived&
-PlainObjectBase<Derived>::setConstant(NoChange_t, Index cols, const Scalar& val)
-{
-  return setConstant(rows(), cols, val);
-}
-
-/** Resizes to the given size, changing only the number of rows, and sets all
-  * coefficients in this expression to the given value \a val. For the parameter
-  * of type NoChange_t, just pass the special value \c NoChange.
-  *
-  * \sa MatrixBase::setConstant(const Scalar&), setConstant(Index,const Scalar&), class CwiseNullaryOp, MatrixBase::Constant(const Scalar&)
-  */
-template<typename Derived>
-EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE Derived&
-PlainObjectBase<Derived>::setConstant(Index rows, NoChange_t, const Scalar& val)
-{
-  return setConstant(rows, cols(), val);
-}
-
-
 /**
-  * \brief Sets a linearly spaced vector.
+  * \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.
@@ -421,33 +381,27 @@ PlainObjectBase<Derived>::setConstant(Index rows, NoChange_t, const Scalar& val)
   * Example: \include DenseBase_setLinSpaced.cpp
   * Output: \verbinclude DenseBase_setLinSpaced.out
   *
-  * For integer scalar types, do not miss the explanations on the definition
-  * of \link LinSpaced(Index,const Scalar&,const Scalar&) even spacing \endlink.
-  *
-  * \sa LinSpaced(Index,const Scalar&,const Scalar&), CwiseNullaryOp
+  * \sa CwiseNullaryOp
   */
 template<typename Derived>
-EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE Derived& DenseBase<Derived>::setLinSpaced(Index newSize, 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(newSize, internal::linspaced_op<Scalar>(low,high,newSize));
+  return derived() = Derived::NullaryExpr(newSize, internal::linspaced_op<Scalar,false>(low,high,newSize));
 }
 
 /**
-  * \brief Sets a linearly spaced vector.
+  * \brief Sets a linearly space vector.
   *
-  * The function fills \c *this with equally spaced values in the closed interval [low,high].
+  * 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
   *
-  * For integer scalar types, do not miss the explanations on the definition
-  * of \link LinSpaced(Index,const Scalar&,const Scalar&) even spacing \endlink.
-  *
-  * \sa LinSpaced(Index,const Scalar&,const Scalar&), setLinSpaced(Index, const Scalar&, const Scalar&), CwiseNullaryOp
+  * \sa setLinSpaced(Index, const Scalar&, const Scalar&), CwiseNullaryOp
   */
 template<typename Derived>
-EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE Derived& DenseBase<Derived>::setLinSpaced(const Scalar& low, const Scalar& high)
+EIGEN_STRONG_INLINE Derived& DenseBase<Derived>::setLinSpaced(const Scalar& low, const Scalar& high)
 {
   EIGEN_STATIC_ASSERT_VECTOR_ONLY(Derived)
   return setLinSpaced(size(), low, high);
@@ -470,10 +424,10 @@ EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE Derived& DenseBase<Derived>::setLinSpaced(
   * \sa Zero(), Zero(Index)
   */
 template<typename Derived>
-EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE const typename DenseBase<Derived>::ConstantReturnType
-DenseBase<Derived>::Zero(Index rows, Index cols)
+EIGEN_STRONG_INLINE const typename DenseBase<Derived>::ConstantReturnType
+DenseBase<Derived>::Zero(Index nbRows, Index nbCols)
 {
-  return Constant(rows, cols, Scalar(0));
+  return Constant(nbRows, nbCols, Scalar(0));
 }
 
 /** \returns an expression of a zero vector.
@@ -493,7 +447,7 @@ DenseBase<Derived>::Zero(Index rows, Index cols)
   * \sa Zero(), Zero(Index,Index)
   */
 template<typename Derived>
-EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE const typename DenseBase<Derived>::ConstantReturnType
+EIGEN_STRONG_INLINE const typename DenseBase<Derived>::ConstantReturnType
 DenseBase<Derived>::Zero(Index size)
 {
   return Constant(size, Scalar(0));
@@ -510,7 +464,7 @@ DenseBase<Derived>::Zero(Index size)
   * \sa Zero(Index), Zero(Index,Index)
   */
 template<typename Derived>
-EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE const typename DenseBase<Derived>::ConstantReturnType
+EIGEN_STRONG_INLINE const typename DenseBase<Derived>::ConstantReturnType
 DenseBase<Derived>::Zero()
 {
   return Constant(Scalar(0));
@@ -525,12 +479,11 @@ DenseBase<Derived>::Zero()
   * \sa class CwiseNullaryOp, Zero()
   */
 template<typename Derived>
-EIGEN_DEVICE_FUNC bool DenseBase<Derived>::isZero(const 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(!internal::isMuchSmallerThan(self.coeff(i, j), static_cast<Scalar>(1), prec))
+      if(!internal::isMuchSmallerThan(this->coeff(i, j), static_cast<Scalar>(1), prec))
         return false;
   return true;
 }
@@ -543,7 +496,7 @@ EIGEN_DEVICE_FUNC bool DenseBase<Derived>::isZero(const RealScalar& prec) const
   * \sa class CwiseNullaryOp, Zero()
   */
 template<typename Derived>
-EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE Derived& DenseBase<Derived>::setZero()
+EIGEN_STRONG_INLINE Derived& DenseBase<Derived>::setZero()
 {
   return setConstant(Scalar(0));
 }
@@ -558,7 +511,7 @@ EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE Derived& DenseBase<Derived>::setZero()
   * \sa DenseBase::setZero(), setZero(Index,Index), class CwiseNullaryOp, DenseBase::Zero()
   */
 template<typename Derived>
-EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE Derived&
+EIGEN_STRONG_INLINE Derived&
 PlainObjectBase<Derived>::setZero(Index newSize)
 {
   resize(newSize);
@@ -567,8 +520,8 @@ PlainObjectBase<Derived>::setZero(Index newSize)
 
 /** Resizes to the given size, and sets all coefficients in this expression to zero.
   *
-  * \param rows the new number of rows
-  * \param cols the new number of columns
+  * \param nbRows the new number of rows
+  * \param nbCols the new number of columns
   *
   * Example: \include Matrix_setZero_int_int.cpp
   * Output: \verbinclude Matrix_setZero_int_int.out
@@ -576,44 +529,18 @@ PlainObjectBase<Derived>::setZero(Index newSize)
   * \sa DenseBase::setZero(), setZero(Index), class CwiseNullaryOp, DenseBase::Zero()
   */
 template<typename Derived>
-EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE Derived&
-PlainObjectBase<Derived>::setZero(Index rows, Index cols)
+EIGEN_STRONG_INLINE Derived&
+PlainObjectBase<Derived>::setZero(Index nbRows, Index nbCols)
 {
-  resize(rows, cols);
+  resize(nbRows, nbCols);
   return setConstant(Scalar(0));
 }
 
-/** Resizes to the given size, changing only the number of columns, and sets all
-  * coefficients in this expression to zero. For the parameter of type NoChange_t,
-  * just pass the special value \c NoChange.
-  *
-  * \sa DenseBase::setZero(), setZero(Index), setZero(Index, Index), setZero(Index, NoChange_t), class CwiseNullaryOp, DenseBase::Zero()
-  */
-template<typename Derived>
-EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE Derived&
-PlainObjectBase<Derived>::setZero(NoChange_t, Index cols)
-{
-  return setZero(rows(), cols);
-}
-
-/** Resizes to the given size, changing only the number of rows, and sets all
-  * coefficients in this expression to zero. For the parameter of type NoChange_t,
-  * just pass the special value \c NoChange.
-  *
-  * \sa DenseBase::setZero(), setZero(Index), setZero(Index, Index), setZero(NoChange_t, Index), class CwiseNullaryOp, DenseBase::Zero()
-  */
-template<typename Derived>
-EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE Derived&
-PlainObjectBase<Derived>::setZero(Index rows, NoChange_t)
-{
-  return setZero(rows, cols());
-}
-
 // ones:
 
 /** \returns an expression of a matrix where all coefficients equal one.
   *
-  * The parameters \a rows and \a cols are the number of rows and of columns of
+  * The parameters \a nbRows and \a nbCols are the number of rows and of columns of
   * the returned matrix. Must be compatible with this MatrixBase type.
   *
   * This variant is meant to be used for dynamic-size matrix types. For fixed-size types,
@@ -626,10 +553,10 @@ PlainObjectBase<Derived>::setZero(Index rows, NoChange_t)
   * \sa Ones(), Ones(Index), isOnes(), class Ones
   */
 template<typename Derived>
-EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE const typename DenseBase<Derived>::ConstantReturnType
-DenseBase<Derived>::Ones(Index rows, Index cols)
+EIGEN_STRONG_INLINE const typename DenseBase<Derived>::ConstantReturnType
+DenseBase<Derived>::Ones(Index nbRows, Index nbCols)
 {
-  return Constant(rows, cols, Scalar(1));
+  return Constant(nbRows, nbCols, Scalar(1));
 }
 
 /** \returns an expression of a vector where all coefficients equal one.
@@ -649,7 +576,7 @@ DenseBase<Derived>::Ones(Index rows, Index cols)
   * \sa Ones(), Ones(Index,Index), isOnes(), class Ones
   */
 template<typename Derived>
-EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE const typename DenseBase<Derived>::ConstantReturnType
+EIGEN_STRONG_INLINE const typename DenseBase<Derived>::ConstantReturnType
 DenseBase<Derived>::Ones(Index newSize)
 {
   return Constant(newSize, Scalar(1));
@@ -666,7 +593,7 @@ DenseBase<Derived>::Ones(Index newSize)
   * \sa Ones(Index), Ones(Index,Index), isOnes(), class Ones
   */
 template<typename Derived>
-EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE const typename DenseBase<Derived>::ConstantReturnType
+EIGEN_STRONG_INLINE const typename DenseBase<Derived>::ConstantReturnType
 DenseBase<Derived>::Ones()
 {
   return Constant(Scalar(1));
@@ -681,7 +608,7 @@ DenseBase<Derived>::Ones()
   * \sa class CwiseNullaryOp, Ones()
   */
 template<typename Derived>
-EIGEN_DEVICE_FUNC bool DenseBase<Derived>::isOnes
+bool DenseBase<Derived>::isOnes
 (const RealScalar& prec) const
 {
   return isApproxToConstant(Scalar(1), prec);
@@ -695,7 +622,7 @@ EIGEN_DEVICE_FUNC bool DenseBase<Derived>::isOnes
   * \sa class CwiseNullaryOp, Ones()
   */
 template<typename Derived>
-EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE Derived& DenseBase<Derived>::setOnes()
+EIGEN_STRONG_INLINE Derived& DenseBase<Derived>::setOnes()
 {
   return setConstant(Scalar(1));
 }
@@ -710,7 +637,7 @@ EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE Derived& DenseBase<Derived>::setOnes()
   * \sa MatrixBase::setOnes(), setOnes(Index,Index), class CwiseNullaryOp, MatrixBase::Ones()
   */
 template<typename Derived>
-EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE Derived&
+EIGEN_STRONG_INLINE Derived&
 PlainObjectBase<Derived>::setOnes(Index newSize)
 {
   resize(newSize);
@@ -719,8 +646,8 @@ PlainObjectBase<Derived>::setOnes(Index newSize)
 
 /** Resizes to the given size, and sets all coefficients in this expression to one.
   *
-  * \param rows the new number of rows
-  * \param cols the new number of columns
+  * \param nbRows the new number of rows
+  * \param nbCols the new number of columns
   *
   * Example: \include Matrix_setOnes_int_int.cpp
   * Output: \verbinclude Matrix_setOnes_int_int.out
@@ -728,44 +655,18 @@ PlainObjectBase<Derived>::setOnes(Index newSize)
   * \sa MatrixBase::setOnes(), setOnes(Index), class CwiseNullaryOp, MatrixBase::Ones()
   */
 template<typename Derived>
-EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE Derived&
-PlainObjectBase<Derived>::setOnes(Index rows, Index cols)
+EIGEN_STRONG_INLINE Derived&
+PlainObjectBase<Derived>::setOnes(Index nbRows, Index nbCols)
 {
-  resize(rows, cols);
+  resize(nbRows, nbCols);
   return setConstant(Scalar(1));
 }
 
-/** Resizes to the given size, changing only the number of rows, and sets all
-  * coefficients in this expression to one. For the parameter of type NoChange_t,
-  * just pass the special value \c NoChange.
-  *
- * \sa MatrixBase::setOnes(), setOnes(Index), setOnes(Index, Index), setOnes(NoChange_t, Index), class CwiseNullaryOp, MatrixBase::Ones()
-  */
-template<typename Derived>
-EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE Derived&
-PlainObjectBase<Derived>::setOnes(Index rows, NoChange_t)
-{
-  return setOnes(rows, cols());
-}
-
-/** Resizes to the given size, changing only the number of columns, and sets all
-  * coefficients in this expression to one. For the parameter of type NoChange_t,
-  * just pass the special value \c NoChange.
-  *
- * \sa MatrixBase::setOnes(), setOnes(Index), setOnes(Index, Index), setOnes(Index, NoChange_t) class CwiseNullaryOp, MatrixBase::Ones()
-  */
-template<typename Derived>
-EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE Derived&
-PlainObjectBase<Derived>::setOnes(NoChange_t, Index cols)
-{
-  return setOnes(rows(), cols);
-}
-
 // Identity:
 
 /** \returns an expression of the identity matrix (not necessarily square).
   *
-  * The parameters \a rows and \a cols are the number of rows and of columns of
+  * The parameters \a nbRows and \a nbCols are the number of rows and of columns of
   * the returned matrix. Must be compatible with this MatrixBase type.
   *
   * This variant is meant to be used for dynamic-size matrix types. For fixed-size types,
@@ -778,10 +679,10 @@ PlainObjectBase<Derived>::setOnes(NoChange_t, Index cols)
   * \sa Identity(), setIdentity(), isIdentity()
   */
 template<typename Derived>
-EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE const typename MatrixBase<Derived>::IdentityReturnType
-MatrixBase<Derived>::Identity(Index rows, Index cols)
+EIGEN_STRONG_INLINE const typename MatrixBase<Derived>::IdentityReturnType
+MatrixBase<Derived>::Identity(Index nbRows, Index nbCols)
 {
-  return DenseBase<Derived>::NullaryExpr(rows, cols, internal::scalar_identity_op<Scalar>());
+  return DenseBase<Derived>::NullaryExpr(nbRows, nbCols, internal::scalar_identity_op<Scalar>());
 }
 
 /** \returns an expression of the identity matrix (not necessarily square).
@@ -795,7 +696,7 @@ MatrixBase<Derived>::Identity(Index rows, Index cols)
   * \sa Identity(Index,Index), setIdentity(), isIdentity()
   */
 template<typename Derived>
-EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE const typename MatrixBase<Derived>::IdentityReturnType
+EIGEN_STRONG_INLINE const typename MatrixBase<Derived>::IdentityReturnType
 MatrixBase<Derived>::Identity()
 {
   EIGEN_STATIC_ASSERT_FIXED_SIZE(Derived)
@@ -815,19 +716,18 @@ template<typename Derived>
 bool MatrixBase<Derived>::isIdentity
 (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(!internal::isApprox(self.coeff(i, j), static_cast<Scalar>(1), prec))
+        if(!internal::isApprox(this->coeff(i, j), static_cast<Scalar>(1), prec))
           return false;
       }
       else
       {
-        if(!internal::isMuchSmallerThan(self.coeff(i, j), static_cast<RealScalar>(1), prec))
+        if(!internal::isMuchSmallerThan(this->coeff(i, j), static_cast<RealScalar>(1), prec))
           return false;
       }
     }
@@ -840,7 +740,6 @@ namespace internal {
 template<typename Derived, bool Big = (Derived::SizeAtCompileTime>=16)>
 struct setIdentity_impl
 {
-  EIGEN_DEVICE_FUNC
   static EIGEN_STRONG_INLINE Derived& run(Derived& m)
   {
     return m = Derived::Identity(m.rows(), m.cols());
@@ -850,11 +749,11 @@ struct setIdentity_impl
 template<typename Derived>
 struct setIdentity_impl<Derived, true>
 {
-  EIGEN_DEVICE_FUNC
+  typedef typename Derived::Index Index;
   static EIGEN_STRONG_INLINE Derived& run(Derived& m)
   {
     m.setZero();
-    const Index size = numext::mini(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;
   }
@@ -870,15 +769,15 @@ struct setIdentity_impl<Derived, true>
   * \sa class CwiseNullaryOp, Identity(), Identity(Index,Index), isIdentity()
   */
 template<typename Derived>
-EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE Derived& MatrixBase<Derived>::setIdentity()
+EIGEN_STRONG_INLINE Derived& MatrixBase<Derived>::setIdentity()
 {
   return internal::setIdentity_impl<Derived>::run(derived());
 }
 
 /** \brief Resizes to the given size, and writes the identity expression (not necessarily square) into *this.
   *
-  * \param rows the new number of rows
-  * \param cols the new number of columns
+  * \param nbRows the new number of rows
+  * \param nbCols the new number of columns
   *
   * Example: \include Matrix_setIdentity_int_int.cpp
   * Output: \verbinclude Matrix_setIdentity_int_int.out
@@ -886,9 +785,9 @@ EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE Derived& MatrixBase<Derived>::setIdentity(
   * \sa MatrixBase::setIdentity(), class CwiseNullaryOp, MatrixBase::Identity()
   */
 template<typename Derived>
-EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE Derived& MatrixBase<Derived>::setIdentity(Index rows, Index cols)
+EIGEN_STRONG_INLINE Derived& MatrixBase<Derived>::setIdentity(Index nbRows, Index nbCols)
 {
-  derived().resize(rows, cols);
+  derived().resize(nbRows, nbCols);
   return setIdentity();
 }
 
@@ -899,7 +798,7 @@ EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE Derived& MatrixBase<Derived>::setIdentity(
   * \sa MatrixBase::Unit(Index), MatrixBase::UnitX(), MatrixBase::UnitY(), MatrixBase::UnitZ(), MatrixBase::UnitW()
   */
 template<typename Derived>
-EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE const typename MatrixBase<Derived>::BasisReturnType MatrixBase<Derived>::Unit(Index newSize, 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(newSize,newSize), i);
@@ -914,7 +813,7 @@ EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE const typename MatrixBase<Derived>::BasisR
   * \sa MatrixBase::Unit(Index,Index), MatrixBase::UnitX(), MatrixBase::UnitY(), MatrixBase::UnitZ(), MatrixBase::UnitW()
   */
 template<typename Derived>
-EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE const typename MatrixBase<Derived>::BasisReturnType MatrixBase<Derived>::Unit(Index i)
+EIGEN_STRONG_INLINE const typename MatrixBase<Derived>::BasisReturnType MatrixBase<Derived>::Unit(Index i)
 {
   EIGEN_STATIC_ASSERT_VECTOR_ONLY(Derived)
   return BasisReturnType(SquareMatrixType::Identity(),i);
@@ -927,7 +826,7 @@ EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE const typename MatrixBase<Derived>::BasisR
   * \sa MatrixBase::Unit(Index,Index), MatrixBase::Unit(Index), MatrixBase::UnitY(), MatrixBase::UnitZ(), MatrixBase::UnitW()
   */
 template<typename Derived>
-EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE const typename MatrixBase<Derived>::BasisReturnType MatrixBase<Derived>::UnitX()
+EIGEN_STRONG_INLINE const typename MatrixBase<Derived>::BasisReturnType MatrixBase<Derived>::UnitX()
 { return Derived::Unit(0); }
 
 /** \returns an expression of the Y axis unit vector (0,1{,0}^*)
@@ -937,7 +836,7 @@ EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE const typename MatrixBase<Derived>::BasisR
   * \sa MatrixBase::Unit(Index,Index), MatrixBase::Unit(Index), MatrixBase::UnitY(), MatrixBase::UnitZ(), MatrixBase::UnitW()
   */
 template<typename Derived>
-EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE const typename MatrixBase<Derived>::BasisReturnType MatrixBase<Derived>::UnitY()
+EIGEN_STRONG_INLINE const typename MatrixBase<Derived>::BasisReturnType MatrixBase<Derived>::UnitY()
 { return Derived::Unit(1); }
 
 /** \returns an expression of the Z axis unit vector (0,0,1{,0}^*)
@@ -947,7 +846,7 @@ EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE const typename MatrixBase<Derived>::BasisR
   * \sa MatrixBase::Unit(Index,Index), MatrixBase::Unit(Index), MatrixBase::UnitY(), MatrixBase::UnitZ(), MatrixBase::UnitW()
   */
 template<typename Derived>
-EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE const typename MatrixBase<Derived>::BasisReturnType MatrixBase<Derived>::UnitZ()
+EIGEN_STRONG_INLINE const typename MatrixBase<Derived>::BasisReturnType MatrixBase<Derived>::UnitZ()
 { return Derived::Unit(2); }
 
 /** \returns an expression of the W axis unit vector (0,0,0,1)
@@ -957,45 +856,9 @@ EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE const typename MatrixBase<Derived>::BasisR
   * \sa MatrixBase::Unit(Index,Index), MatrixBase::Unit(Index), MatrixBase::UnitY(), MatrixBase::UnitZ(), MatrixBase::UnitW()
   */
 template<typename Derived>
-EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE const typename MatrixBase<Derived>::BasisReturnType MatrixBase<Derived>::UnitW()
+EIGEN_STRONG_INLINE const typename MatrixBase<Derived>::BasisReturnType MatrixBase<Derived>::UnitW()
 { return Derived::Unit(3); }
 
-/** \brief Set the coefficients of \c *this to the i-th unit (basis) vector
-  *
-  * \param i index of the unique coefficient to be set to 1
-  *
-  * \only_for_vectors
-  *
-  * \sa MatrixBase::setIdentity(), class CwiseNullaryOp, MatrixBase::Unit(Index,Index)
-  */
-template<typename Derived>
-EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE Derived& MatrixBase<Derived>::setUnit(Index i)
-{
-  EIGEN_STATIC_ASSERT_VECTOR_ONLY(Derived);
-  eigen_assert(i<size());
-  derived().setZero();
-  derived().coeffRef(i) = Scalar(1);
-  return derived();
-}
-
-/** \brief Resizes to the given \a newSize, and writes the i-th unit (basis) vector into *this.
-  *
-  * \param newSize the new size of the vector
-  * \param i index of the unique coefficient to be set to 1
-  *
-  * \only_for_vectors
-  *
-  * \sa MatrixBase::setIdentity(), class CwiseNullaryOp, MatrixBase::Unit(Index,Index)
-  */
-template<typename Derived>
-EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE Derived& MatrixBase<Derived>::setUnit(Index newSize, Index i)
-{
-  EIGEN_STATIC_ASSERT_VECTOR_ONLY(Derived);
-  eigen_assert(i<newSize);
-  derived().resize(newSize);
-  return setUnit(i);
-}
-
 } // end namespace Eigen
 
 #endif // EIGEN_CWISE_NULLARY_OP_H

Eigen/src/Core/CwiseTernaryOp.h

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

Eigen/src/Core/CwiseUnaryOp.h

@@ -1,7 +1,7 @@
 // This file is part of Eigen, a lightweight C++ template library
 // for linear algebra.
 //
-// Copyright (C) 2008-2014 Gael Guennebaud <gael.guennebaud@inria.fr>
+// Copyright (C) 2008-2010 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
@@ -11,34 +11,15 @@
 #ifndef EIGEN_CWISE_UNARY_OP_H
 #define EIGEN_CWISE_UNARY_OP_H
 
-namespace Eigen {
-
-namespace internal {
-template<typename UnaryOp, typename XprType>
-struct traits<CwiseUnaryOp<UnaryOp, XprType> >
- : traits<XprType>
-{
-  typedef typename result_of<
-                     UnaryOp(const typename XprType::Scalar&)
-                   >::type Scalar;
-  typedef typename XprType::Nested XprTypeNested;
-  typedef typename remove_reference<XprTypeNested>::type _XprTypeNested;
-  enum {
-    Flags = _XprTypeNested::Flags & RowMajorBit
-  };
-};
-}
-
-template<typename UnaryOp, typename XprType, typename StorageKind>
-class CwiseUnaryOpImpl;
+namespace Eigen { 
 
 /** \class CwiseUnaryOp
   * \ingroup Core_Module
   *
   * \brief Generic expression where a coefficient-wise unary operator is applied to an expression
   *
-  * \tparam UnaryOp template functor implementing the operator
-  * \tparam XprType the type of the expression to which we are applying the unary operator
+  * \param UnaryOp template functor implementing the operator
+  * \param XprType the type of the expression to which we are applying the unary operator
   *
   * This class represents an expression where a unary operator is applied to an expression.
   * It is the return type of all operations taking exactly 1 input expression, regardless of the
@@ -51,51 +32,93 @@ class CwiseUnaryOpImpl;
   *
   * \sa MatrixBase::unaryExpr(const CustomUnaryOp &) const, class CwiseBinaryOp, class CwiseNullaryOp
   */
+
+namespace internal {
 template<typename UnaryOp, typename XprType>
-class CwiseUnaryOp : public CwiseUnaryOpImpl<UnaryOp, XprType, typename internal::traits<XprType>::StorageKind>, internal::no_assignment_operator
+struct traits<CwiseUnaryOp<UnaryOp, XprType> >
+ : traits<XprType>
+{
+  typedef typename result_of<
+                     UnaryOp(typename XprType::Scalar)
+                   >::type Scalar;
+  typedef typename XprType::Nested XprTypeNested;
+  typedef typename remove_reference<XprTypeNested>::type _XprTypeNested;
+  enum {
+    Flags = _XprTypeNested::Flags & (
+      HereditaryBits | LinearAccessBit | AlignedBit
+      | (functor_traits<UnaryOp>::PacketAccess ? PacketAccessBit : 0)),
+    CoeffReadCost = _XprTypeNested::CoeffReadCost + functor_traits<UnaryOp>::Cost
+  };
+};
+}
+
+template<typename UnaryOp, typename XprType, typename StorageKind>
+class CwiseUnaryOpImpl;
+
+template<typename UnaryOp, typename XprType>
+class CwiseUnaryOp : internal::no_assignment_operator,
+  public CwiseUnaryOpImpl<UnaryOp, XprType, typename internal::traits<XprType>::StorageKind>
 {
   public:
 
     typedef typename CwiseUnaryOpImpl<UnaryOp, XprType,typename internal::traits<XprType>::StorageKind>::Base Base;
     EIGEN_GENERIC_PUBLIC_INTERFACE(CwiseUnaryOp)
-    typedef typename internal::ref_selector<XprType>::type XprTypeNested;
-    typedef typename internal::remove_all<XprType>::type NestedExpression;
 
-    EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
-    explicit CwiseUnaryOp(const XprType& xpr, const UnaryOp& func = UnaryOp())
+    inline CwiseUnaryOp(const XprType& xpr, const UnaryOp& func = UnaryOp())
       : m_xpr(xpr), m_functor(func) {}
 
-    EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE EIGEN_CONSTEXPR
-    Index rows() const EIGEN_NOEXCEPT { return m_xpr.rows(); }
-    EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE EIGEN_CONSTEXPR
-    Index cols() const EIGEN_NOEXCEPT { return m_xpr.cols(); }
+    EIGEN_STRONG_INLINE Index rows() const { return m_xpr.rows(); }
+    EIGEN_STRONG_INLINE Index cols() const { return m_xpr.cols(); }
 
     /** \returns the functor representing the unary operation */
-    EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
     const UnaryOp& functor() const { return m_functor; }
 
     /** \returns the nested expression */
-    EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
-    const typename internal::remove_all<XprTypeNested>::type&
+    const typename internal::remove_all<typename XprType::Nested>::type&
     nestedExpression() const { return m_xpr; }
 
     /** \returns the nested expression */
-    EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
-    typename internal::remove_all<XprTypeNested>::type&
-    nestedExpression() { return m_xpr; }
+    typename internal::remove_all<typename XprType::Nested>::type&
+    nestedExpression() { return m_xpr.const_cast_derived(); }
 
   protected:
-    XprTypeNested m_xpr;
+    typename XprType::Nested m_xpr;
     const UnaryOp m_functor;
 };
 
-// Generic API dispatcher
-template<typename UnaryOp, typename XprType, typename StorageKind>
-class CwiseUnaryOpImpl
-  : public internal::generic_xpr_base<CwiseUnaryOp<UnaryOp, XprType> >::type
+// 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 internal::dense_xpr_base<CwiseUnaryOp<UnaryOp, XprType> >::type
 {
-public:
-  typedef typename internal::generic_xpr_base<CwiseUnaryOp<UnaryOp, XprType> >::type Base;
+  public:
+
+    typedef CwiseUnaryOp<UnaryOp, XprType> Derived;
+    typedef typename internal::dense_xpr_base<CwiseUnaryOp<UnaryOp, XprType> >::type Base;
+    EIGEN_DENSE_PUBLIC_INTERFACE(Derived)
+
+    EIGEN_STRONG_INLINE const Scalar coeff(Index rowId, Index colId) const
+    {
+      return derived().functor()(derived().nestedExpression().coeff(rowId, colId));
+    }
+
+    template<int LoadMode>
+    EIGEN_STRONG_INLINE PacketScalar packet(Index rowId, Index colId) const
+    {
+      return derived().functor().packetOp(derived().nestedExpression().template packet<LoadMode>(rowId, colId));
+    }
+
+    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));
+    }
 };
 
 } // end namespace Eigen

Eigen/src/Core/CwiseUnaryView.h

@@ -12,19 +12,33 @@
 
 namespace Eigen {
 
+/** \class CwiseUnaryView
+  * \ingroup Core_Module
+  *
+  * \brief Generic lvalue expression of a coefficient-wise unary operator of a matrix or a vector
+  *
+  * \param ViewOp template functor implementing the view
+  * \param MatrixType the type of the matrix we are applying the unary operator
+  *
+  * This class represents a lvalue expression of a generic unary view operator of a matrix or a vector.
+  * It is the return type of real() and imag(), and most of the time this is the only way it is used.
+  *
+  * \sa MatrixBase::unaryViewExpr(const CustomUnaryOp &) const, class CwiseUnaryOp
+  */
+
 namespace internal {
 template<typename ViewOp, typename MatrixType>
 struct traits<CwiseUnaryView<ViewOp, MatrixType> >
  : traits<MatrixType>
 {
   typedef typename result_of<
-                     ViewOp(const typename traits<MatrixType>::Scalar&)
+                     ViewOp(typename traits<MatrixType>::Scalar)
                    >::type Scalar;
   typedef typename MatrixType::Nested MatrixTypeNested;
   typedef typename remove_all<MatrixTypeNested>::type _MatrixTypeNested;
   enum {
-    FlagsLvalueBit = is_lvalue<MatrixType>::value ? LvalueBit : 0,
-    Flags = traits<_MatrixTypeNested>::Flags & (RowMajorBit | FlagsLvalueBit | DirectAccessBit), // FIXME DirectAccessBit should not be handled by expressions
+    Flags = (traits<_MatrixTypeNested>::Flags & (HereditaryBits | LvalueBit | LinearAccessBit | DirectAccessBit)),
+    CoeffReadCost = traits<_MatrixTypeNested>::CoeffReadCost + functor_traits<ViewOp>::Cost,
     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
@@ -41,19 +55,6 @@ struct traits<CwiseUnaryView<ViewOp, MatrixType> >
 template<typename ViewOp, typename MatrixType, typename StorageKind>
 class CwiseUnaryViewImpl;
 
-/** \class CwiseUnaryView
-  * \ingroup Core_Module
-  *
-  * \brief Generic lvalue expression of a coefficient-wise unary operator of a matrix or a vector
-  *
-  * \tparam ViewOp template functor implementing the view
-  * \tparam MatrixType the type of the matrix we are applying the unary operator
-  *
-  * This class represents a lvalue expression of a generic unary view operator of a matrix or a vector.
-  * It is the return type of real() and imag(), and most of the time this is the only way it is used.
-  *
-  * \sa MatrixBase::unaryViewExpr(const CustomUnaryOp &) const, class CwiseUnaryOp
-  */
 template<typename ViewOp, typename MatrixType>
 class CwiseUnaryView : public CwiseUnaryViewImpl<ViewOp, MatrixType, typename internal::traits<MatrixType>::StorageKind>
 {
@@ -61,44 +62,32 @@ class CwiseUnaryView : public CwiseUnaryViewImpl<ViewOp, MatrixType, typename in
 
     typedef typename CwiseUnaryViewImpl<ViewOp, MatrixType,typename internal::traits<MatrixType>::StorageKind>::Base Base;
     EIGEN_GENERIC_PUBLIC_INTERFACE(CwiseUnaryView)
-    typedef typename internal::ref_selector<MatrixType>::non_const_type MatrixTypeNested;
-    typedef typename internal::remove_all<MatrixType>::type NestedExpression;
 
-    explicit EIGEN_DEVICE_FUNC inline CwiseUnaryView(MatrixType& mat, const ViewOp& func = ViewOp())
+    inline CwiseUnaryView(const MatrixType& mat, const ViewOp& func = ViewOp())
       : m_matrix(mat), m_functor(func) {}
 
     EIGEN_INHERIT_ASSIGNMENT_OPERATORS(CwiseUnaryView)
 
-    EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE EIGEN_CONSTEXPR
-    Index rows() const EIGEN_NOEXCEPT { return m_matrix.rows(); }
-    EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE EIGEN_CONSTEXPR
-    Index cols() const EIGEN_NOEXCEPT { return m_matrix.cols(); }
+    EIGEN_STRONG_INLINE Index rows() const { return m_matrix.rows(); }
+    EIGEN_STRONG_INLINE Index cols() const { return m_matrix.cols(); }
 
     /** \returns the functor representing unary operation */
-    EIGEN_DEVICE_FUNC const ViewOp& functor() const { return m_functor; }
+    const ViewOp& functor() const { return m_functor; }
 
     /** \returns the nested expression */
-    EIGEN_DEVICE_FUNC const typename internal::remove_all<MatrixTypeNested>::type&
+    const typename internal::remove_all<typename MatrixType::Nested>::type&
     nestedExpression() const { return m_matrix; }
 
     /** \returns the nested expression */
-    EIGEN_DEVICE_FUNC typename internal::remove_reference<MatrixTypeNested>::type&
-    nestedExpression() { return m_matrix; }
+    typename internal::remove_all<typename MatrixType::Nested>::type&
+    nestedExpression() { return m_matrix.const_cast_derived(); }
 
   protected:
-    MatrixTypeNested m_matrix;
+    // FIXME changed from MatrixType::Nested because of a weird compilation error with sun CC
+    typename internal::nested<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 internal::dense_xpr_base< CwiseUnaryView<ViewOp, MatrixType> >::type
@@ -110,21 +99,39 @@ class CwiseUnaryViewImpl<ViewOp,MatrixType,Dense>
 
     EIGEN_DENSE_PUBLIC_INTERFACE(Derived)
     EIGEN_INHERIT_ASSIGNMENT_OPERATORS(CwiseUnaryViewImpl)
+    
+    inline Scalar* data() { return &coeffRef(0); }
+    inline const Scalar* data() const { return &coeff(0); }
 
-    EIGEN_DEVICE_FUNC inline Scalar* data() { return &(this->coeffRef(0)); }
-    EIGEN_DEVICE_FUNC inline const Scalar* data() const { return &(this->coeff(0)); }
-
-    EIGEN_DEVICE_FUNC EIGEN_CONSTEXPR inline Index innerStride() const
+    inline Index innerStride() const
     {
       return derived().nestedExpression().innerStride() * sizeof(typename internal::traits<MatrixType>::Scalar) / sizeof(Scalar);
     }
 
-    EIGEN_DEVICE_FUNC EIGEN_CONSTEXPR inline Index outerStride() const
+    inline Index outerStride() const
     {
       return derived().nestedExpression().outerStride() * sizeof(typename internal::traits<MatrixType>::Scalar) / sizeof(Scalar);
     }
-  protected:
-    EIGEN_DEFAULT_EMPTY_CONSTRUCTOR_AND_DESTRUCTOR(CwiseUnaryViewImpl)
+
+    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));
+    }
 };
 
 } // end namespace Eigen

Eigen/src/Core/DenseBase.h

@@ -14,15 +14,15 @@
 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)
+  EIGEN_STATIC_ASSERT(NumTraits<DenseIndex>::IsSigned,THE_INDEX_TYPE_MUST_BE_A_SIGNED_TYPE); 
 }
 
 } // end namespace internal
-
+  
 /** \class DenseBase
   * \ingroup Core_Module
   *
@@ -34,45 +34,37 @@ static inline void check_DenseIndex_is_signed() {
   * \tparam Derived is the derived type, e.g., a matrix type or an expression.
   *
   * This class can be extended with the help of the plugin mechanism described on the page
-  * \ref TopicCustomizing_Plugins by defining the preprocessor symbol \c EIGEN_DENSEBASE_PLUGIN.
+  * \ref TopicCustomizingEigen by defining the preprocessor symbol \c EIGEN_DENSEBASE_PLUGIN.
   *
-  * \sa \blank \ref TopicClassHierarchy
+  * \sa \ref TopicClassHierarchy
   */
 template<typename Derived> class DenseBase
 #ifndef EIGEN_PARSED_BY_DOXYGEN
-  : public DenseCoeffsBase<Derived, internal::accessors_level<Derived>::value>
+  : public internal::special_scalar_op_base<Derived,typename internal::traits<Derived>::Scalar,
+                                     typename NumTraits<typename internal::traits<Derived>::Scalar>::Real>
 #else
-  : public DenseCoeffsBase<Derived,DirectWriteAccessors>
+  : public DenseCoeffsBase<Derived>
 #endif // not EIGEN_PARSED_BY_DOXYGEN
 {
   public:
+    using internal::special_scalar_op_base<Derived,typename internal::traits<Derived>::Scalar,
+                typename NumTraits<typename internal::traits<Derived>::Scalar>::Real>::operator*;
 
-    /** Inner iterator type to iterate over the coefficients of a row or column.
-      * \sa class InnerIterator
-      */
-    typedef Eigen::InnerIterator<Derived> InnerIterator;
+    class InnerIterator;
 
     typedef typename internal::traits<Derived>::StorageKind StorageKind;
 
-    /**
-      * \brief The type used to store indices
-      * \details This typedef is relevant for types that store multiple indices such as
-      *          PermutationMatrix or Transpositions, otherwise it defaults to Eigen::Index
-      * \sa \blank \ref TopicPreprocessorDirectives, Eigen::Index, SparseMatrixBase.
-     */
-    typedef typename internal::traits<Derived>::StorageIndex StorageIndex;
+    /** \brief The type of indices 
+      * \details To change this, \c \#define the preprocessor symbol \c EIGEN_DEFAULT_DENSE_INDEX_TYPE.
+      * \sa \ref TopicPreprocessorDirectives.
+      */
+    typedef typename internal::traits<Derived>::Index Index; 
 
-    /** 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 internal::packet_traits<Scalar>::type PacketScalar;
     typedef typename NumTraits<Scalar>::Real RealScalar;
-    typedef DenseCoeffsBase<Derived, internal::accessors_level<Derived>::value> Base;
 
+    typedef DenseCoeffsBase<Derived> Base;
     using Base::derived;
     using Base::const_cast_derived;
     using Base::rows;
@@ -82,6 +74,16 @@ 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;
@@ -150,18 +152,13 @@ template<typename Derived> class DenseBase
           * \sa SizeAtCompileTime, MaxRowsAtCompileTime, MaxColsAtCompileTime
           */
 
-      IsVectorAtCompileTime = internal::traits<Derived>::RowsAtCompileTime == 1
-                           || internal::traits<Derived>::ColsAtCompileTime == 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). */
 
-      NumDimensions = int(MaxSizeAtCompileTime) == 1 ? 0 : bool(IsVectorAtCompileTime) ? 1 : 2,
-        /**< This value is equal to Tensor::NumDimensions, i.e. 0 for scalars, 1 for vectors,
-         * and 2 for matrices.
-         */
-
       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".
@@ -172,54 +169,30 @@ template<typename Derived> class DenseBase
       InnerSizeAtCompileTime = int(IsVectorAtCompileTime) ? int(SizeAtCompileTime)
                              : int(IsRowMajor) ? int(ColsAtCompileTime) : int(RowsAtCompileTime),
 
+      CoeffReadCost = internal::traits<Derived>::CoeffReadCost,
+        /**< This is a rough measure of how expensive it is to read one coefficient from
+          * this expression.
+          */
+
       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;
+    enum { ThisConstantIsPrivateInPlainObjectBase };
 
     /** \returns the number of nonzero coefficients which is in practice the number
       * of stored coefficients. */
-    EIGEN_DEVICE_FUNC EIGEN_CONSTEXPR
     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 \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 EIGEN_CONSTEXPR
     Index outerSize() const
     {
       return IsVectorAtCompileTime ? 1
@@ -229,9 +202,8 @@ 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 \ref TopicStorageOrders "storage order", i.e., the number of rows for a
+      * 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 EIGEN_CONSTEXPR
     Index innerSize() const
     {
       return IsVectorAtCompileTime ? this->size()
@@ -242,7 +214,6 @@ 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.
       */
-    EIGEN_DEVICE_FUNC
     void resize(Index newSize)
     {
       EIGEN_ONLY_USED_FOR_DEBUG(newSize);
@@ -253,22 +224,22 @@ 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.
       */
-    EIGEN_DEVICE_FUNC
-    void resize(Index rows, Index cols)
+    void resize(Index nbRows, Index nbCols)
     {
-      EIGEN_ONLY_USED_FOR_DEBUG(rows);
-      EIGEN_ONLY_USED_FOR_DEBUG(cols);
-      eigen_assert(rows == this->rows() && cols == this->cols()
+      EIGEN_ONLY_USED_FOR_DEBUG(nbRows);
+      EIGEN_ONLY_USED_FOR_DEBUG(nbCols);
+      eigen_assert(nbRows == this->rows() && nbCols == this->cols()
                 && "DenseBase::resize() does not actually allow to resize.");
     }
 
 #ifndef EIGEN_PARSED_BY_DOXYGEN
+
     /** \internal Represents a matrix with all coefficients equal to one another*/
-    typedef CwiseNullaryOp<internal::scalar_constant_op<Scalar>,PlainObject> ConstantReturnType;
-    /** \internal \deprecated Represents a vector with linearly spaced coefficients that allows sequential access only. */
-    EIGEN_DEPRECATED typedef CwiseNullaryOp<internal::linspaced_op<Scalar>,PlainObject> SequentialLinSpacedReturnType;
+    typedef CwiseNullaryOp<internal::scalar_constant_op<Scalar>,Derived> ConstantReturnType;
+    /** \internal Represents a vector with linearly spaced coefficients that allows sequential access only. */
+    typedef CwiseNullaryOp<internal::linspaced_op<Scalar,false>,Derived> SequentialLinSpacedReturnType;
     /** \internal Represents a vector with linearly spaced coefficients that allows random access. */
-    typedef CwiseNullaryOp<internal::linspaced_op<Scalar>,PlainObject> RandomAccessLinSpacedReturnType;
+    typedef CwiseNullaryOp<internal::linspaced_op<Scalar,true>,Derived> RandomAccessLinSpacedReturnType;
     /** \internal the return type of MatrixBase::eigenvalues() */
     typedef Matrix<typename NumTraits<typename internal::traits<Derived>::Scalar>::Real, internal::traits<Derived>::ColsAtCompileTime, 1> EigenvaluesReturnType;
 
@@ -276,134 +247,118 @@ template<typename Derived> class DenseBase
 
     /** Copies \a other into *this. \returns a reference to *this. */
     template<typename OtherDerived>
-    EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
     Derived& operator=(const DenseBase<OtherDerived>& other);
 
     /** Special case of the template operator=, in order to prevent the compiler
       * from generating a default operator= (issue hit with g++ 4.1)
       */
-    EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
     Derived& operator=(const DenseBase& other);
 
     template<typename OtherDerived>
-    EIGEN_DEVICE_FUNC
     Derived& operator=(const EigenBase<OtherDerived> &other);
 
     template<typename OtherDerived>
-    EIGEN_DEVICE_FUNC
     Derived& operator+=(const EigenBase<OtherDerived> &other);
 
     template<typename OtherDerived>
-    EIGEN_DEVICE_FUNC
     Derived& operator-=(const EigenBase<OtherDerived> &other);
 
     template<typename OtherDerived>
-    EIGEN_DEVICE_FUNC
     Derived& operator=(const ReturnByValue<OtherDerived>& func);
 
-    /** \internal
-      * Copies \a other into *this without evaluating other. \returns a reference to *this. */
+#ifndef EIGEN_PARSED_BY_DOXYGEN
+    /** Copies \a other into *this without evaluating other. \returns a reference to *this. */
     template<typename OtherDerived>
-    /** \deprecated */
-    EIGEN_DEPRECATED EIGEN_DEVICE_FUNC
     Derived& lazyAssign(const DenseBase<OtherDerived>& other);
+#endif // not EIGEN_PARSED_BY_DOXYGEN
 
-    EIGEN_DEVICE_FUNC
     CommaInitializer<Derived> operator<< (const Scalar& s);
 
     template<unsigned int Added,unsigned int Removed>
-    /** \deprecated it now returns \c *this */
-    EIGEN_DEPRECATED
-    const Derived& flagged() const
-    { return derived(); }
+    const Flagged<Derived, Added, Removed> flagged() const;
 
     template<typename OtherDerived>
-    EIGEN_DEVICE_FUNC
     CommaInitializer<Derived> operator<< (const DenseBase<OtherDerived>& other);
 
-    typedef Transpose<Derived> TransposeReturnType;
-    EIGEN_DEVICE_FUNC
-    TransposeReturnType transpose();
-    typedef typename internal::add_const<Transpose<const Derived> >::type ConstTransposeReturnType;
-    EIGEN_DEVICE_FUNC
+    Eigen::Transpose<Derived> transpose();
+	typedef typename internal::add_const<Transpose<const Derived> >::type ConstTransposeReturnType;
     ConstTransposeReturnType transpose() const;
-    EIGEN_DEVICE_FUNC
     void transposeInPlace();
+#ifndef EIGEN_NO_DEBUG
+  protected:
+    template<typename OtherDerived>
+    void checkTransposeAliasing(const OtherDerived& other) const;
+  public:
+#endif
 
-    EIGEN_DEVICE_FUNC static const ConstantReturnType
+
+    static const ConstantReturnType
     Constant(Index rows, Index cols, const Scalar& value);
-    EIGEN_DEVICE_FUNC static const ConstantReturnType
+    static const ConstantReturnType
     Constant(Index size, const Scalar& value);
-    EIGEN_DEVICE_FUNC static const ConstantReturnType
+    static const ConstantReturnType
     Constant(const Scalar& value);
 
-    EIGEN_DEPRECATED EIGEN_DEVICE_FUNC static const RandomAccessLinSpacedReturnType
+    static const SequentialLinSpacedReturnType
     LinSpaced(Sequential_t, Index size, const Scalar& low, const Scalar& high);
-    EIGEN_DEPRECATED EIGEN_DEVICE_FUNC static const RandomAccessLinSpacedReturnType
-    LinSpaced(Sequential_t, const Scalar& low, const Scalar& high);
-
-    EIGEN_DEVICE_FUNC static const RandomAccessLinSpacedReturnType
+    static const RandomAccessLinSpacedReturnType
     LinSpaced(Index size, const Scalar& low, const Scalar& high);
-    EIGEN_DEVICE_FUNC static const RandomAccessLinSpacedReturnType
+    static const SequentialLinSpacedReturnType
+    LinSpaced(Sequential_t, const Scalar& low, const Scalar& high);
+    static const RandomAccessLinSpacedReturnType
     LinSpaced(const Scalar& low, const Scalar& high);
 
-    template<typename CustomNullaryOp> EIGEN_DEVICE_FUNC
-    static const CwiseNullaryOp<CustomNullaryOp, PlainObject>
+    template<typename CustomNullaryOp>
+    static const CwiseNullaryOp<CustomNullaryOp, Derived>
     NullaryExpr(Index rows, Index cols, const CustomNullaryOp& func);
-    template<typename CustomNullaryOp> EIGEN_DEVICE_FUNC
-    static const CwiseNullaryOp<CustomNullaryOp, PlainObject>
+    template<typename CustomNullaryOp>
+    static const CwiseNullaryOp<CustomNullaryOp, Derived>
     NullaryExpr(Index size, const CustomNullaryOp& func);
-    template<typename CustomNullaryOp> EIGEN_DEVICE_FUNC
-    static const CwiseNullaryOp<CustomNullaryOp, PlainObject>
+    template<typename CustomNullaryOp>
+    static const CwiseNullaryOp<CustomNullaryOp, Derived>
     NullaryExpr(const CustomNullaryOp& func);
 
-    EIGEN_DEVICE_FUNC static const ConstantReturnType Zero(Index rows, Index cols);
-    EIGEN_DEVICE_FUNC static const ConstantReturnType Zero(Index size);
-    EIGEN_DEVICE_FUNC static const ConstantReturnType Zero();
-    EIGEN_DEVICE_FUNC static const ConstantReturnType Ones(Index rows, Index cols);
-    EIGEN_DEVICE_FUNC static const ConstantReturnType Ones(Index size);
-    EIGEN_DEVICE_FUNC static const ConstantReturnType Ones();
+    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 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();
+    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();
 
-    template<typename OtherDerived> EIGEN_DEVICE_FUNC
+    template<typename OtherDerived>
     bool isApprox(const DenseBase<OtherDerived>& other,
                   const RealScalar& prec = NumTraits<Scalar>::dummy_precision()) const;
-    EIGEN_DEVICE_FUNC
     bool isMuchSmallerThan(const RealScalar& other,
                            const RealScalar& prec = NumTraits<Scalar>::dummy_precision()) const;
-    template<typename OtherDerived> EIGEN_DEVICE_FUNC
+    template<typename OtherDerived>
     bool isMuchSmallerThan(const DenseBase<OtherDerived>& other,
                            const RealScalar& prec = NumTraits<Scalar>::dummy_precision()) const;
 
-    EIGEN_DEVICE_FUNC bool isApproxToConstant(const Scalar& value, const RealScalar& prec = NumTraits<Scalar>::dummy_precision()) const;
-    EIGEN_DEVICE_FUNC bool isConstant(const Scalar& value, const RealScalar& prec = NumTraits<Scalar>::dummy_precision()) const;
-    EIGEN_DEVICE_FUNC bool isZero(const RealScalar& prec = NumTraits<Scalar>::dummy_precision()) const;
-    EIGEN_DEVICE_FUNC bool isOnes(const RealScalar& prec = NumTraits<Scalar>::dummy_precision()) const;
-
+    bool isApproxToConstant(const Scalar& value, const RealScalar& prec = NumTraits<Scalar>::dummy_precision()) const;
+    bool isConstant(const Scalar& value, const RealScalar& prec = NumTraits<Scalar>::dummy_precision()) const;
+    bool isZero(const RealScalar& prec = NumTraits<Scalar>::dummy_precision()) const;
+    bool isOnes(const RealScalar& prec = NumTraits<Scalar>::dummy_precision()) const;
+    
     inline bool hasNaN() const;
     inline bool allFinite() const;
 
-    EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
-    Derived& operator*=(const Scalar& other);
-    EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
-    Derived& operator/=(const Scalar& other);
+    inline Derived& operator*=(const Scalar& other);
+    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 careful with eval() and the auto C++ keyword, as detailed in this \link TopicPitfalls_auto_keyword page \endlink.
       */
-    EIGEN_DEVICE_FUNC
     EIGEN_STRONG_INLINE EvalReturnType eval() const
     {
       // Even though MSVC does not honor strong inlining when the return type
@@ -416,113 +371,56 @@ template<typename Derived> class DenseBase
       *
       */
     template<typename OtherDerived>
-    EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
-    void swap(const DenseBase<OtherDerived>& other)
+    void swap(const DenseBase<OtherDerived>& other,
+              int = OtherDerived::ThisConstantIsPrivateInPlainObjectBase)
     {
-      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>());
+      SwapWrapper<Derived>(derived()).lazyAssign(other.derived());
     }
 
     /** swaps *this with the matrix or array \a other.
       *
       */
     template<typename OtherDerived>
-    EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
     void swap(PlainObjectBase<OtherDerived>& other)
     {
-      eigen_assert(rows()==other.rows() && cols()==other.cols());
-      call_assignment(derived(), other.derived(), internal::swap_assign_op<Scalar>());
+      SwapWrapper<Derived>(derived()).lazyAssign(other.derived());
     }
 
-    EIGEN_DEVICE_FUNC inline const NestByValue<Derived> nestByValue() const;
-    EIGEN_DEVICE_FUNC inline const ForceAlignedAccess<Derived> forceAlignedAccess() const;
-    EIGEN_DEVICE_FUNC inline ForceAlignedAccess<Derived> forceAlignedAccess();
-    template<bool Enable> EIGEN_DEVICE_FUNC
-    inline const typename internal::conditional<Enable,ForceAlignedAccess<Derived>,Derived&>::type forceAlignedAccessIf() const;
-    template<bool Enable> EIGEN_DEVICE_FUNC
-    inline typename internal::conditional<Enable,ForceAlignedAccess<Derived>,Derived&>::type forceAlignedAccessIf();
 
-    EIGEN_DEVICE_FUNC Scalar sum() const;
-    EIGEN_DEVICE_FUNC Scalar mean() const;
-    EIGEN_DEVICE_FUNC Scalar trace() const;
+    inline const NestByValue<Derived> nestByValue() const;
+    inline const ForceAlignedAccess<Derived> forceAlignedAccess() const;
+    inline ForceAlignedAccess<Derived> forceAlignedAccess();
+    template<bool Enable> inline const typename internal::conditional<Enable,ForceAlignedAccess<Derived>,Derived&>::type forceAlignedAccessIf() const;
+    template<bool Enable> inline typename internal::conditional<Enable,ForceAlignedAccess<Derived>,Derived&>::type forceAlignedAccessIf();
 
-    EIGEN_DEVICE_FUNC Scalar prod() const;
+    Scalar sum() const;
+    Scalar mean() const;
+    Scalar trace() const;
 
-    template<int NaNPropagation>
-    EIGEN_DEVICE_FUNC typename internal::traits<Derived>::Scalar minCoeff() const;
-    template<int NaNPropagation>
-    EIGEN_DEVICE_FUNC typename internal::traits<Derived>::Scalar maxCoeff() const;
+    Scalar prod() const;
 
+    typename internal::traits<Derived>::Scalar minCoeff() const;
+    typename internal::traits<Derived>::Scalar maxCoeff() const;
 
-    // By default, the fastest version with undefined NaN propagation semantics is
-    // used.
-    // TODO(rmlarsen): Replace with default template argument when we move to
-    // c++11 or beyond.
-    EIGEN_DEVICE_FUNC inline typename internal::traits<Derived>::Scalar minCoeff() const {
-      return minCoeff<PropagateFast>();
-    }
-    EIGEN_DEVICE_FUNC inline typename internal::traits<Derived>::Scalar maxCoeff() const {
-      return maxCoeff<PropagateFast>();
-    }
-
-    template<int NaNPropagation, typename IndexType>
-    EIGEN_DEVICE_FUNC
+    template<typename IndexType>
     typename internal::traits<Derived>::Scalar minCoeff(IndexType* row, IndexType* col) const;
-    template<int NaNPropagation, typename IndexType>
-    EIGEN_DEVICE_FUNC
+    template<typename IndexType>
     typename internal::traits<Derived>::Scalar maxCoeff(IndexType* row, IndexType* col) const;
-    template<int NaNPropagation, typename IndexType>
-    EIGEN_DEVICE_FUNC
+    template<typename IndexType>
     typename internal::traits<Derived>::Scalar minCoeff(IndexType* index) const;
-    template<int NaNPropagation, typename IndexType>
-    EIGEN_DEVICE_FUNC
+    template<typename IndexType>
     typename internal::traits<Derived>::Scalar maxCoeff(IndexType* index) const;
 
-    // TODO(rmlarsen): Replace these methods with a default template argument.
-    template<typename IndexType>
-    EIGEN_DEVICE_FUNC inline
-    typename internal::traits<Derived>::Scalar minCoeff(IndexType* row, IndexType* col) const {
-      return minCoeff<PropagateFast>(row, col);
-    }
-    template<typename IndexType>
-    EIGEN_DEVICE_FUNC inline
-    typename internal::traits<Derived>::Scalar maxCoeff(IndexType* row, IndexType* col) const {
-      return maxCoeff<PropagateFast>(row, col);
-    }
-    template<typename IndexType>
-     EIGEN_DEVICE_FUNC inline
-    typename internal::traits<Derived>::Scalar minCoeff(IndexType* index) const {
-      return minCoeff<PropagateFast>(index);
-    }
-    template<typename IndexType>
-    EIGEN_DEVICE_FUNC inline
-    typename internal::traits<Derived>::Scalar maxCoeff(IndexType* index) const {
-      return maxCoeff<PropagateFast>(index);
-    }
-  
     template<typename BinaryOp>
-    EIGEN_DEVICE_FUNC
-    Scalar redux(const BinaryOp& func) const;
+    typename internal::result_of<BinaryOp(typename internal::traits<Derived>::Scalar)>::type
+    redux(const BinaryOp& func) const;
 
     template<typename Visitor>
-    EIGEN_DEVICE_FUNC
     void visit(Visitor& func) const;
 
-    /** \returns a WithFormat proxy object allowing to print a matrix the with given
-      * format \a fmt.
-      *
-      * See class IOFormat for some examples.
-      *
-      * \sa class IOFormat, class WithFormat
-      */
-    inline const WithFormat<Derived> format(const IOFormat& fmt) const
-    {
-      return WithFormat<Derived>(derived(), fmt);
-    }
+    inline const WithFormat<Derived> format(const IOFormat& fmt) const;
 
     /** \returns the unique coefficient of a 1x1 expression */
-    EIGEN_DEVICE_FUNC
     CoeffReturnType value() const
     {
       EIGEN_STATIC_ASSERT_SIZE_1x1(Derived)
@@ -530,158 +428,80 @@ template<typename Derived> class DenseBase
       return derived().coeff(0,0);
     }
 
-    EIGEN_DEVICE_FUNC bool all() const;
-    EIGEN_DEVICE_FUNC bool any() const;
-    EIGEN_DEVICE_FUNC Index count() const;
+    bool all(void) const;
+    bool any(void) const;
+    Index count() const;
 
     typedef VectorwiseOp<Derived, Horizontal> RowwiseReturnType;
     typedef const VectorwiseOp<const Derived, Horizontal> ConstRowwiseReturnType;
     typedef VectorwiseOp<Derived, Vertical> ColwiseReturnType;
     typedef const VectorwiseOp<const Derived, Vertical> ConstColwiseReturnType;
 
-    /** \returns a VectorwiseOp wrapper of *this for broadcasting and partial reductions
-    *
-    * 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();
+    ConstRowwiseReturnType rowwise() const;
+    RowwiseReturnType rowwise();
+    ConstColwiseReturnType colwise() const;
+    ColwiseReturnType colwise();
 
-    /** \returns a VectorwiseOp wrapper of *this broadcasting and partial reductions
-    *
-    * 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();
+    static const CwiseNullaryOp<internal::scalar_random_op<Scalar>,Derived> Random(Index rows, Index cols);
+    static const CwiseNullaryOp<internal::scalar_random_op<Scalar>,Derived> Random(Index size);
+    static const CwiseNullaryOp<internal::scalar_random_op<Scalar>,Derived> Random();
 
     template<typename ThenDerived,typename ElseDerived>
-    inline EIGEN_DEVICE_FUNC const Select<Derived,ThenDerived,ElseDerived>
+    const Select<Derived,ThenDerived,ElseDerived>
     select(const DenseBase<ThenDerived>& thenMatrix,
            const DenseBase<ElseDerived>& elseMatrix) const;
 
     template<typename ThenDerived>
-    inline EIGEN_DEVICE_FUNC const Select<Derived,ThenDerived, typename ThenDerived::ConstantReturnType>
+    inline const Select<Derived,ThenDerived, typename ThenDerived::ConstantReturnType>
     select(const DenseBase<ThenDerived>& thenMatrix, const typename ThenDerived::Scalar& elseScalar) const;
 
     template<typename ElseDerived>
-    inline EIGEN_DEVICE_FUNC const Select<Derived, typename ElseDerived::ConstantReturnType, ElseDerived >
+    inline const Select<Derived, typename ElseDerived::ConstantReturnType, ElseDerived >
     select(const typename ElseDerived::Scalar& thenScalar, const DenseBase<ElseDerived>& elseMatrix) const;
 
     template<int p> RealScalar lpNorm() const;
 
     template<int RowFactor, int ColFactor>
-    EIGEN_DEVICE_FUNC
     const Replicate<Derived,RowFactor,ColFactor> replicate() const;
-    /**
-    * \return an expression of the replication of \c *this
-    *
-    * Example: \include MatrixBase_replicate_int_int.cpp
-    * Output: \verbinclude MatrixBase_replicate_int_int.out
-    *
-    * \sa VectorwiseOp::replicate(), DenseBase::replicate<int,int>(), class Replicate
-    */
-    //Code moved here due to a CUDA compiler bug
-    EIGEN_DEVICE_FUNC
-    const Replicate<Derived, Dynamic, Dynamic> replicate(Index rowFactor, Index colFactor) const
-    {
-      return Replicate<Derived, Dynamic, Dynamic>(derived(), rowFactor, colFactor);
-    }
+    const Replicate<Derived,Dynamic,Dynamic> replicate(Index rowFacor,Index colFactor) const;
 
     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();
-
-    #ifdef EIGEN_PARSED_BY_DOXYGEN
-    /** STL-like <a href="https://en.cppreference.com/w/cpp/named_req/RandomAccessIterator">RandomAccessIterator</a>
-      * iterator type as returned by the begin() and end() methods.
-      */
-    typedef random_access_iterator_type iterator;
-    /** This is the const version of iterator (aka read-only) */
-    typedef random_access_iterator_type const_iterator;
-    #else
-    typedef typename internal::conditional< (Flags&DirectAccessBit)==DirectAccessBit,
-                                            internal::pointer_based_stl_iterator<Derived>,
-                                            internal::generic_randaccess_stl_iterator<Derived>
-                                          >::type iterator_type;
-
-    typedef typename internal::conditional< (Flags&DirectAccessBit)==DirectAccessBit,
-                                            internal::pointer_based_stl_iterator<const Derived>,
-                                            internal::generic_randaccess_stl_iterator<const Derived>
-                                          >::type const_iterator_type;
-
-    // Stl-style iterators are supported only for vectors.
-
-    typedef typename internal::conditional< IsVectorAtCompileTime,
-                                            iterator_type,
-                                            void
-                                          >::type iterator;
-
-    typedef typename internal::conditional< IsVectorAtCompileTime,
-                                            const_iterator_type,
-                                            void
-                                          >::type const_iterator;
-    #endif
-
-    inline iterator begin();
-    inline const_iterator begin() const;
-    inline const_iterator cbegin() const;
-    inline iterator end();
-    inline const_iterator end() const;
-    inline const_iterator cend() const;
+    ReverseReturnType reverse();
+    ConstReverseReturnType reverse() const;
+    void reverseInPlace();
 
 #define EIGEN_CURRENT_STORAGE_BASE_CLASS Eigen::DenseBase
-#define EIGEN_DOC_BLOCK_ADDONS_NOT_INNER_PANEL
-#define EIGEN_DOC_BLOCK_ADDONS_INNER_PANEL_IF(COND)
-#define EIGEN_DOC_UNARY_ADDONS(X,Y)
-#   include "../plugins/CommonCwiseUnaryOps.h"
 #   include "../plugins/BlockMethods.h"
-#   include "../plugins/IndexedViewMethods.h"
-#   include "../plugins/ReshapedMethods.h"
 #   ifdef EIGEN_DENSEBASE_PLUGIN
 #     include EIGEN_DENSEBASE_PLUGIN
 #   endif
 #undef EIGEN_CURRENT_STORAGE_BASE_CLASS
-#undef EIGEN_DOC_BLOCK_ADDONS_NOT_INNER_PANEL
-#undef EIGEN_DOC_BLOCK_ADDONS_INNER_PANEL_IF
-#undef EIGEN_DOC_UNARY_ADDONS
+
+#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>
-    EIGEN_DEVICE_FUNC
-    inline void evalTo(Dest& ) const
+    template<typename Dest> inline void evalTo(Dest& ) const
     {
       EIGEN_STATIC_ASSERT((internal::is_same<Dest,void>::value),THE_EVAL_EVALTO_FUNCTION_SHOULD_NEVER_BE_CALLED_FOR_DENSE_OBJECTS);
     }
 
   protected:
-    EIGEN_DEFAULT_COPY_CONSTRUCTOR(DenseBase)
     /** Default constructor. Do nothing. */
-    EIGEN_DEVICE_FUNC DenseBase()
+    DenseBase()
     {
       /* Just checks for self-consistency of the flags.
-       * Only do it when debugging Eigen, as this borders on paranoia and could slow compilation down
+       * Only do it when debugging Eigen, as this borders on paranoiac and could slow compilation down
        */
 #ifdef EIGEN_INTERNAL_DEBUGGING
       EIGEN_STATIC_ASSERT((EIGEN_IMPLIES(MaxRowsAtCompileTime==1 && MaxColsAtCompileTime!=1, int(IsRowMajor))
@@ -691,9 +511,9 @@ template<typename Derived> class DenseBase
     }
 
   private:
-    EIGEN_DEVICE_FUNC explicit DenseBase(int);
-    EIGEN_DEVICE_FUNC DenseBase(int,int);
-    template<typename OtherDerived> EIGEN_DEVICE_FUNC explicit DenseBase(const DenseBase<OtherDerived>&);
+    explicit DenseBase(int);
+    DenseBase(int,int);
+    template<typename OtherDerived> explicit DenseBase(const DenseBase<OtherDerived>&);
 };
 
 } // end namespace Eigen

Eigen/src/Core/DenseCoeffsBase.h

@@ -22,12 +22,11 @@ template<typename T> struct add_const_on_value_type_if_arithmetic
 /** \brief Base class providing read-only coefficient access to matrices and arrays.
   * \ingroup Core_Module
   * \tparam Derived Type of the derived class
-  *
-  * \note #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.
-  *
+  * 
   * \sa DenseCoeffsBase<Derived, WriteAccessors>, DenseCoeffsBase<Derived, DirectAccessors>,
   *     \ref TopicClassHierarchy
   */
@@ -36,6 +35,7 @@ class DenseCoeffsBase<Derived,ReadOnlyAccessors> : public EigenBase<Derived>
 {
   public:
     typedef typename internal::traits<Derived>::StorageKind StorageKind;
+    typedef typename internal::traits<Derived>::Index Index;
     typedef typename internal::traits<Derived>::Scalar Scalar;
     typedef typename internal::packet_traits<Scalar>::type PacketScalar;
 
@@ -61,7 +61,6 @@ 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
@@ -70,7 +69,6 @@ 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
@@ -93,15 +91,13 @@ 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
     {
       eigen_internal_assert(row >= 0 && row < rows()
-                         && col >= 0 && col < cols());
-      return internal::evaluator<Derived>(derived()).coeff(row,col);
+                        && col >= 0 && col < cols());
+      return derived().coeff(row, col);
     }
 
-    EIGEN_DEVICE_FUNC
     EIGEN_STRONG_INLINE CoeffReturnType coeffByOuterInner(Index outer, Index inner) const
     {
       return coeff(rowIndexByOuterInner(outer, inner),
@@ -112,12 +108,11 @@ 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
     {
       eigen_assert(row >= 0 && row < rows()
           && col >= 0 && col < cols());
-      return coeff(row, col);
+      return derived().coeff(row, col);
     }
 
     /** Short version: don't use this function, use
@@ -135,14 +130,11 @@ 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
     {
-      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);
+      return derived().coeff(index);
     }
 
 
@@ -154,14 +146,15 @@ class DenseCoeffsBase<Derived,ReadOnlyAccessors> : public EigenBase<Derived>
       * z() const, w() const
       */
 
-    EIGEN_DEVICE_FUNC
     EIGEN_STRONG_INLINE CoeffReturnType
     operator[](Index index) const
     {
+      #ifndef EIGEN2_SUPPORT
       EIGEN_STATIC_ASSERT(Derived::IsVectorAtCompileTime,
                           THE_BRACKET_OPERATOR_IS_ONLY_FOR_VECTORS__USE_THE_PARENTHESIS_OPERATOR_INSTEAD)
+      #endif
       eigen_assert(index >= 0 && index < size());
-      return coeff(index);
+      return derived().coeff(index);
     }
 
     /** \returns the coefficient at given index.
@@ -174,49 +167,32 @@ class DenseCoeffsBase<Derived,ReadOnlyAccessors> : public EigenBase<Derived>
       * z() const, w() const
       */
 
-    EIGEN_DEVICE_FUNC
     EIGEN_STRONG_INLINE CoeffReturnType
     operator()(Index index) const
     {
       eigen_assert(index >= 0 && index < size());
-      return coeff(index);
+      return derived().coeff(index);
     }
 
     /** equivalent to operator[](0).  */
 
-    EIGEN_DEVICE_FUNC
     EIGEN_STRONG_INLINE CoeffReturnType
     x() const { return (*this)[0]; }
 
     /** equivalent to operator[](1).  */
 
-    EIGEN_DEVICE_FUNC
     EIGEN_STRONG_INLINE CoeffReturnType
-    y() const
-    {
-      EIGEN_STATIC_ASSERT(Derived::SizeAtCompileTime==-1 || Derived::SizeAtCompileTime>=2, OUT_OF_RANGE_ACCESS);
-      return (*this)[1];
-    }
+    y() const { return (*this)[1]; }
 
     /** equivalent to operator[](2).  */
 
-    EIGEN_DEVICE_FUNC
     EIGEN_STRONG_INLINE CoeffReturnType
-    z() const
-    {
-      EIGEN_STATIC_ASSERT(Derived::SizeAtCompileTime==-1 || Derived::SizeAtCompileTime>=3, OUT_OF_RANGE_ACCESS);
-      return (*this)[2];
-    }
+    z() const { return (*this)[2]; }
 
     /** equivalent to operator[](3).  */
 
-    EIGEN_DEVICE_FUNC
     EIGEN_STRONG_INLINE CoeffReturnType
-    w() const
-    {
-      EIGEN_STATIC_ASSERT(Derived::SizeAtCompileTime==-1 || Derived::SizeAtCompileTime>=4, OUT_OF_RANGE_ACCESS);
-      return (*this)[3];
-    }
+    w() const { return (*this)[3]; }
 
     /** \internal
       * \returns the packet of coefficients starting at the given row and column. It is your responsibility
@@ -231,9 +207,9 @@ class DenseCoeffsBase<Derived,ReadOnlyAccessors> : public EigenBase<Derived>
     template<int LoadMode>
     EIGEN_STRONG_INLINE PacketReturnType packet(Index row, Index col) const
     {
-      typedef typename internal::packet_traits<Scalar>::type DefaultPacketType;
-      eigen_internal_assert(row >= 0 && row < rows() && col >= 0 && col < cols());
-      return internal::evaluator<Derived>(derived()).template packet<LoadMode,DefaultPacketType>(row,col);
+      eigen_internal_assert(row >= 0 && row < rows()
+                      && col >= 0 && col < cols());
+      return derived().template packet<LoadMode>(row,col);
     }
 
 
@@ -258,11 +234,8 @@ class DenseCoeffsBase<Derived,ReadOnlyAccessors> : public EigenBase<Derived>
     template<int LoadMode>
     EIGEN_STRONG_INLINE PacketReturnType packet(Index index) const
     {
-      EIGEN_STATIC_ASSERT(internal::evaluator<Derived>::Flags & LinearAccessBit,
-                          THIS_COEFFICIENT_ACCESSOR_TAKING_ONE_ACCESS_IS_ONLY_FOR_EXPRESSIONS_ALLOWING_LINEAR_ACCESS)
-      typedef typename internal::packet_traits<Scalar>::type DefaultPacketType;
       eigen_internal_assert(index >= 0 && index < size());
-      return internal::evaluator<Derived>(derived()).template packet<LoadMode,DefaultPacketType>(index);
+      return derived().template packet<LoadMode>(index);
     }
 
   protected:
@@ -289,13 +262,12 @@ 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
-  *
-  * \note #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
   * defines the const variant for reading specific entries.
-  *
+  * 
   * \sa DenseCoeffsBase<Derived, DirectAccessors>, \ref TopicClassHierarchy
   */
 template<typename Derived>
@@ -306,6 +278,7 @@ class DenseCoeffsBase<Derived, WriteAccessors> : public DenseCoeffsBase<Derived,
     typedef DenseCoeffsBase<Derived, ReadOnlyAccessors> Base;
 
     typedef typename internal::traits<Derived>::StorageKind StorageKind;
+    typedef typename internal::traits<Derived>::Index Index;
     typedef typename internal::traits<Derived>::Scalar Scalar;
     typedef typename internal::packet_traits<Scalar>::type PacketScalar;
     typedef typename NumTraits<Scalar>::Real RealScalar;
@@ -338,15 +311,13 @@ 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)
     {
       eigen_internal_assert(row >= 0 && row < rows()
-                         && col >= 0 && col < cols());
-      return internal::evaluator<Derived>(derived()).coeffRef(row,col);
+                        && col >= 0 && col < cols());
+      return derived().coeffRef(row, col);
     }
 
-    EIGEN_DEVICE_FUNC
     EIGEN_STRONG_INLINE Scalar&
     coeffRefByOuterInner(Index outer, Index inner)
     {
@@ -359,13 +330,12 @@ class DenseCoeffsBase<Derived, WriteAccessors> : public DenseCoeffsBase<Derived,
       * \sa operator[](Index)
       */
 
-    EIGEN_DEVICE_FUNC
     EIGEN_STRONG_INLINE Scalar&
     operator()(Index row, Index col)
     {
       eigen_assert(row >= 0 && row < rows()
           && col >= 0 && col < cols());
-      return coeffRef(row, col);
+      return derived().coeffRef(row, col);
     }
 
 
@@ -384,14 +354,11 @@ 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)
     {
-      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);
+      return derived().coeffRef(index);
     }
 
     /** \returns a reference to the coefficient at given index.
@@ -401,14 +368,15 @@ 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)
     {
+      #ifndef EIGEN2_SUPPORT
       EIGEN_STATIC_ASSERT(Derived::IsVectorAtCompileTime,
                           THE_BRACKET_OPERATOR_IS_ONLY_FOR_VECTORS__USE_THE_PARENTHESIS_OPERATOR_INSTEAD)
+      #endif
       eigen_assert(index >= 0 && index < size());
-      return coeffRef(index);
+      return derived().coeffRef(index);
     }
 
     /** \returns a reference to the coefficient at given index.
@@ -420,62 +388,179 @@ 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_assert(index >= 0 && index < size());
-      return coeffRef(index);
+      return derived().coeffRef(index);
     }
 
     /** equivalent to operator[](0).  */
 
-    EIGEN_DEVICE_FUNC
     EIGEN_STRONG_INLINE Scalar&
     x() { return (*this)[0]; }
 
     /** equivalent to operator[](1).  */
 
-    EIGEN_DEVICE_FUNC
     EIGEN_STRONG_INLINE Scalar&
-    y()
-    {
-      EIGEN_STATIC_ASSERT(Derived::SizeAtCompileTime==-1 || Derived::SizeAtCompileTime>=2, OUT_OF_RANGE_ACCESS);
-      return (*this)[1];
-    }
+    y() { return (*this)[1]; }
 
     /** equivalent to operator[](2).  */
 
-    EIGEN_DEVICE_FUNC
     EIGEN_STRONG_INLINE Scalar&
-    z()
-    {
-      EIGEN_STATIC_ASSERT(Derived::SizeAtCompileTime==-1 || Derived::SizeAtCompileTime>=3, OUT_OF_RANGE_ACCESS);
-      return (*this)[2];
-    }
+    z() { return (*this)[2]; }
 
     /** equivalent to operator[](3).  */
 
-    EIGEN_DEVICE_FUNC
     EIGEN_STRONG_INLINE Scalar&
-    w()
+    w() { return (*this)[3]; }
+
+    /** \internal
+      * 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 internal::packet_traits<Scalar>::type& val)
     {
-      EIGEN_STATIC_ASSERT(Derived::SizeAtCompileTime==-1 || Derived::SizeAtCompileTime>=4, OUT_OF_RANGE_ACCESS);
-      return (*this)[3];
+      eigen_internal_assert(row >= 0 && row < rows()
+                        && col >= 0 && col < cols());
+      derived().template writePacket<StoreMode>(row,col,val);
     }
+
+
+    /** \internal */
+    template<int StoreMode>
+    EIGEN_STRONG_INLINE void writePacketByOuterInner
+    (Index outer, Index inner, const typename internal::packet_traits<Scalar>::type& val)
+    {
+      writePacket<StoreMode>(rowIndexByOuterInner(outer, inner),
+                            colIndexByOuterInner(outer, inner),
+                            val);
+    }
+
+    /** \internal
+      * 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 internal::packet_traits<Scalar>::type& val)
+    {
+      eigen_internal_assert(index >= 0 && index < size());
+      derived().template writePacket<StoreMode>(index,val);
+    }
+
+#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)
+    {
+      eigen_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)
+    {
+      eigen_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)
+    {
+      eigen_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)
+    {
+      eigen_internal_assert(index >= 0 && index < size());
+      derived().template writePacket<StoreMode>(index,
+        other.derived().template packet<LoadMode>(index));
+    }
+
+    /** \internal */
+    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 read-only coefficient access to matrices and arrays.
   * \ingroup Core_Module
   * \tparam Derived Type of the derived class
-  *
-  * \note #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, ReadOnlyAccessors> which defines functions to access entries read-only using
   * \c operator() .
   *
-  * \sa \blank \ref TopicClassHierarchy
+  * \sa \ref TopicClassHierarchy
   */
 template<typename Derived>
 class DenseCoeffsBase<Derived, DirectAccessors> : public DenseCoeffsBase<Derived, ReadOnlyAccessors>
@@ -483,6 +568,7 @@ class DenseCoeffsBase<Derived, DirectAccessors> : public DenseCoeffsBase<Derived
   public:
 
     typedef DenseCoeffsBase<Derived, ReadOnlyAccessors> Base;
+    typedef typename internal::traits<Derived>::Index Index;
     typedef typename internal::traits<Derived>::Scalar Scalar;
     typedef typename NumTraits<Scalar>::Real RealScalar;
 
@@ -495,7 +581,6 @@ class DenseCoeffsBase<Derived, DirectAccessors> : public DenseCoeffsBase<Derived
       *
       * \sa outerStride(), rowStride(), colStride()
       */
-    EIGEN_DEVICE_FUNC EIGEN_CONSTEXPR
     inline Index innerStride() const
     {
       return derived().innerStride();
@@ -506,14 +591,13 @@ class DenseCoeffsBase<Derived, DirectAccessors> : public DenseCoeffsBase<Derived
       *
       * \sa innerStride(), rowStride(), colStride()
       */
-    EIGEN_DEVICE_FUNC EIGEN_CONSTEXPR
     inline Index outerStride() const
     {
       return derived().outerStride();
     }
 
     // FIXME shall we remove it ?
-    EIGEN_CONSTEXPR inline Index stride() const
+    inline Index stride() const
     {
       return Derived::IsVectorAtCompileTime ? innerStride() : outerStride();
     }
@@ -522,7 +606,6 @@ class DenseCoeffsBase<Derived, DirectAccessors> : public DenseCoeffsBase<Derived
       *
       * \sa innerStride(), outerStride(), colStride()
       */
-    EIGEN_DEVICE_FUNC EIGEN_CONSTEXPR
     inline Index rowStride() const
     {
       return Derived::IsRowMajor ? outerStride() : innerStride();
@@ -532,7 +615,6 @@ class DenseCoeffsBase<Derived, DirectAccessors> : public DenseCoeffsBase<Derived
       *
       * \sa innerStride(), outerStride(), rowStride()
       */
-    EIGEN_DEVICE_FUNC EIGEN_CONSTEXPR
     inline Index colStride() const
     {
       return Derived::IsRowMajor ? innerStride() : outerStride();
@@ -542,14 +624,13 @@ class DenseCoeffsBase<Derived, DirectAccessors> : public DenseCoeffsBase<Derived
 /** \brief Base class providing direct read/write coefficient access to matrices and arrays.
   * \ingroup Core_Module
   * \tparam Derived Type of the derived class
-  *
-  * \note #DirectWriteAccessors Constant indicating direct access
+  * \tparam #DirectWriteAccessors Constant indicating direct access
   *
   * This class defines functions to work with strides which can be used to access entries directly. This class
   * inherits DenseCoeffsBase<Derived, WriteAccessors> which defines functions to access entries read/write using
   * \c operator().
   *
-  * \sa \blank \ref TopicClassHierarchy
+  * \sa \ref TopicClassHierarchy
   */
 template<typename Derived>
 class DenseCoeffsBase<Derived, DirectWriteAccessors>
@@ -558,6 +639,7 @@ class DenseCoeffsBase<Derived, DirectWriteAccessors>
   public:
 
     typedef DenseCoeffsBase<Derived, WriteAccessors> Base;
+    typedef typename internal::traits<Derived>::Index Index;
     typedef typename internal::traits<Derived>::Scalar Scalar;
     typedef typename NumTraits<Scalar>::Real RealScalar;
 
@@ -570,8 +652,7 @@ class DenseCoeffsBase<Derived, DirectWriteAccessors>
       *
       * \sa outerStride(), rowStride(), colStride()
       */
-    EIGEN_DEVICE_FUNC EIGEN_CONSTEXPR
-    inline Index innerStride() const EIGEN_NOEXCEPT
+    inline Index innerStride() const
     {
       return derived().innerStride();
     }
@@ -581,14 +662,13 @@ class DenseCoeffsBase<Derived, DirectWriteAccessors>
       *
       * \sa innerStride(), rowStride(), colStride()
       */
-    EIGEN_DEVICE_FUNC EIGEN_CONSTEXPR
-    inline Index outerStride() const EIGEN_NOEXCEPT
+    inline Index outerStride() const
     {
       return derived().outerStride();
     }
 
     // FIXME shall we remove it ?
-    EIGEN_CONSTEXPR inline Index stride() const EIGEN_NOEXCEPT
+    inline Index stride() const
     {
       return Derived::IsVectorAtCompileTime ? innerStride() : outerStride();
     }
@@ -597,8 +677,7 @@ class DenseCoeffsBase<Derived, DirectWriteAccessors>
       *
       * \sa innerStride(), outerStride(), colStride()
       */
-    EIGEN_DEVICE_FUNC EIGEN_CONSTEXPR
-    inline Index rowStride() const EIGEN_NOEXCEPT
+    inline Index rowStride() const
     {
       return Derived::IsRowMajor ? outerStride() : innerStride();
     }
@@ -607,8 +686,7 @@ class DenseCoeffsBase<Derived, DirectWriteAccessors>
       *
       * \sa innerStride(), outerStride(), rowStride()
       */
-    EIGEN_DEVICE_FUNC EIGEN_CONSTEXPR
-    inline Index colStride() const EIGEN_NOEXCEPT
+    inline Index colStride() const
     {
       return Derived::IsRowMajor ? innerStride() : outerStride();
     }
@@ -616,42 +694,33 @@ class DenseCoeffsBase<Derived, DirectWriteAccessors>
 
 namespace internal {
 
-template<int Alignment, typename Derived, bool JustReturnZero>
+template<typename Derived, bool JustReturnZero>
 struct first_aligned_impl
 {
-  static EIGEN_CONSTEXPR inline Index run(const Derived&) EIGEN_NOEXCEPT
+  static inline typename Derived::Index run(const Derived&)
   { return 0; }
 };
 
-template<int Alignment, typename Derived>
-struct first_aligned_impl<Alignment, Derived, false>
+template<typename Derived>
+struct first_aligned_impl<Derived, false>
 {
-  static inline Index run(const Derived& m)
+  static inline typename Derived::Index run(const Derived& m)
   {
-    return internal::first_aligned<Alignment>(m.data(), m.size());
+    return internal::first_aligned(&m.const_cast_derived().coeffRef(0,0), m.size());
   }
 };
 
-/** \internal \returns the index of the first element of the array stored by \a m that is properly aligned with respect to \a Alignment for vectorization.
-  *
-  * \tparam Alignment requested alignment in Bytes.
+/** \internal \returns the index of the first element of the array that is well aligned for vectorization.
   *
   * There is also the variant first_aligned(const Scalar*, Integer) defined in Memory.h. See it for more
   * documentation.
   */
-template<int Alignment, typename Derived>
-static inline Index first_aligned(const DenseBase<Derived>& m)
-{
-  enum { ReturnZero = (int(evaluator<Derived>::Alignment) >= Alignment) || !(Derived::Flags & DirectAccessBit) };
-  return first_aligned_impl<Alignment, Derived, ReturnZero>::run(m.derived());
-}
-
 template<typename Derived>
-static inline Index first_default_aligned(const DenseBase<Derived>& m)
+static inline typename Derived::Index first_aligned(const Derived& m)
 {
-  typedef typename Derived::Scalar Scalar;
-  typedef typename packet_traits<Scalar>::type DefaultPacketType;
-  return internal::first_aligned<int(unpacket_traits<DefaultPacketType>::alignment),Derived>(m);
+  return first_aligned_impl
+          <Derived, (Derived::Flags & AlignedBit) || !(Derived::Flags & DirectAccessBit)>
+          ::run(m);
 }
 
 template<typename Derived, bool HasDirectAccess = has_direct_access<Derived>::ret>

Eigen/src/Core/DenseStorage.h

@@ -3,7 +3,7 @@
 //
 // 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>
+// Copyright (C) 2010 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
@@ -13,9 +13,9 @@
 #define EIGEN_MATRIXSTORAGE_H
 
 #ifdef EIGEN_DENSE_STORAGE_CTOR_PLUGIN
-  #define EIGEN_INTERNAL_DENSE_STORAGE_CTOR_PLUGIN(X) X; 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(X)
+  #define EIGEN_INTERNAL_DENSE_STORAGE_CTOR_PLUGIN
 #endif
 
 namespace Eigen {
@@ -24,143 +24,73 @@ 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 >
+                        : (((Size*sizeof(T))%16)==0) ? 16
+                        : 0 >
 struct plain_array
 {
   T array[Size];
 
-  EIGEN_DEVICE_FUNC
-  plain_array()
-  {
-    check_static_allocation_size<T,Size>();
+  plain_array() 
+  { 
+    EIGEN_STATIC_ASSERT(Size * sizeof(T) <= 128 * 128 * 8, OBJECT_ALLOCATED_ON_STACK_IS_TOO_BIG);
   }
 
-  EIGEN_DEVICE_FUNC
-  plain_array(constructor_without_unaligned_array_assert)
-  {
-    check_static_allocation_size<T,Size>();
+  plain_array(constructor_without_unaligned_array_assert) 
+  { 
+    EIGEN_STATIC_ASSERT(Size * sizeof(T) <= 128 * 128 * 8, OBJECT_ALLOCATED_ON_STACK_IS_TOO_BIG);
   }
 };
 
 #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 alignment test based on the fact the array is declared to be aligned.
+#elif EIGEN_GNUC_AT_LEAST(4,7) 
+  // GCC 4.7 is too aggressive in its optimizations and remove the alignement test based on the fact the array is declared to be aligned.
   // See this bug report: http://gcc.gnu.org/bugzilla/show_bug.cgi?id=53900
   // Hiding the origin of the array pointer behind a function argument seems to do the trick even if the function is inlined:
   template<typename PtrType>
   EIGEN_ALWAYS_INLINE PtrType eigen_unaligned_array_assert_workaround_gcc47(PtrType array) { return array; }
   #define EIGEN_MAKE_UNALIGNED_ARRAY_ASSERT(sizemask) \
-    eigen_assert((internal::UIntPtr(eigen_unaligned_array_assert_workaround_gcc47(array)) & (sizemask)) == 0 \
+    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((internal::UIntPtr(array) & (sizemask)) == 0 \
+    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_USER_ALIGN16 T array[Size];
 
-  EIGEN_DEVICE_FUNC
-  plain_array()
-  {
-    EIGEN_MAKE_UNALIGNED_ARRAY_ASSERT(15);
-    check_static_allocation_size<T,Size>();
+  plain_array() 
+  { 
+    EIGEN_MAKE_UNALIGNED_ARRAY_ASSERT(0xf);
+    EIGEN_STATIC_ASSERT(Size * sizeof(T) <= 128 * 128 * 8, OBJECT_ALLOCATED_ON_STACK_IS_TOO_BIG);
   }
 
-  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>();
+  plain_array(constructor_without_unaligned_array_assert) 
+  { 
+    EIGEN_STATIC_ASSERT(Size * sizeof(T) <= 128 * 128 * 8, OBJECT_ALLOCATED_ON_STACK_IS_TOO_BIG);
   }
 };
 
 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) {}
+  EIGEN_USER_ALIGN16 T array[1];
+  plain_array() {}
+  plain_array(constructor_without_unaligned_array_assert) {}
 };
 
 } // end namespace internal
@@ -184,68 +114,33 @@ template<typename T, int Size, int _Rows, int _Cols, int _Options> class DenseSt
 {
     internal::plain_array<T,Size,_Options> m_data;
   public:
-    EIGEN_DEVICE_FUNC DenseStorage() {
-      EIGEN_INTERNAL_DENSE_STORAGE_CTOR_PLUGIN(Index size = Size)
-    }
-    EIGEN_DEVICE_FUNC
-    explicit DenseStorage(internal::constructor_without_unaligned_array_assert)
+    inline DenseStorage() {}
+    inline DenseStorage(internal::constructor_without_unaligned_array_assert)
       : m_data(internal::constructor_without_unaligned_array_assert()) {}
-    EIGEN_DEVICE_FUNC
-    DenseStorage(const DenseStorage& other) : m_data(other.m_data) {
-      EIGEN_INTERNAL_DENSE_STORAGE_CTOR_PLUGIN(Index size = Size)
-    }
-    EIGEN_DEVICE_FUNC
-    DenseStorage& operator=(const DenseStorage& other)
-    {
-      if (this != &other) m_data = other.m_data;
-      return *this;
-    }
-#if EIGEN_HAS_RVALUE_REFERENCES
-    EIGEN_DEVICE_FUNC DenseStorage(DenseStorage&& other) EIGEN_NOEXCEPT
-      : m_data(std::move(other.m_data))
-    {
-    }
-    EIGEN_DEVICE_FUNC DenseStorage& operator=(DenseStorage&& other) EIGEN_NOEXCEPT
-    {
-      if (this != &other)
-        m_data = std::move(other.m_data);
-      return *this;
-    }
-#endif
-    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) {
-      numext::swap(m_data, other.m_data);
-    }
-    EIGEN_DEVICE_FUNC static EIGEN_CONSTEXPR Index rows(void) EIGEN_NOEXCEPT {return _Rows;}
-    EIGEN_DEVICE_FUNC static EIGEN_CONSTEXPR Index cols(void) EIGEN_NOEXCEPT {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; }
+    inline DenseStorage(DenseIndex,DenseIndex,DenseIndex) {}
+    inline void swap(DenseStorage& other) { std::swap(m_data,other.m_data); }
+    static inline DenseIndex rows(void) {return _Rows;}
+    static inline 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 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 EIGEN_CONSTEXPR Index rows(void) EIGEN_NOEXCEPT {return _Rows;}
-    EIGEN_DEVICE_FUNC static EIGEN_CONSTEXPR Index cols(void) EIGEN_NOEXCEPT {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; }
+    inline DenseStorage() {}
+    inline DenseStorage(internal::constructor_without_unaligned_array_assert) {}
+    inline DenseStorage(DenseIndex,DenseIndex,DenseIndex) {}
+    inline void swap(DenseStorage& ) {}
+    static inline DenseIndex rows(void) {return _Rows;}
+    static inline 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; }
 };
 
 // more specializations for null matrices; these are necessary to resolve ambiguities
@@ -262,339 +157,173 @@ template<typename T, int _Cols, int _Options> class DenseStorage<T, 0, Dynamic,
 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;
+    DenseIndex m_rows;
+    DenseIndex m_cols;
   public:
-    EIGEN_DEVICE_FUNC DenseStorage() : m_rows(0), m_cols(0) {}
-    EIGEN_DEVICE_FUNC explicit DenseStorage(internal::constructor_without_unaligned_array_assert)
+    inline DenseStorage() : m_rows(0), m_cols(0) {}
+    inline 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)
-    {
-      numext::swap(m_data,other.m_data);
-      numext::swap(m_rows,other.m_rows);
-      numext::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; }
+    inline DenseStorage(DenseIndex, DenseIndex nbRows, DenseIndex nbCols) : m_rows(nbRows), m_cols(nbCols) {}
+    inline 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); }
+    inline DenseIndex rows() const {return m_rows;}
+    inline DenseIndex cols() const {return m_cols;}
+    inline void conservativeResize(DenseIndex, DenseIndex nbRows, DenseIndex nbCols) { m_rows = nbRows; m_cols = nbCols; }
+    inline void resize(DenseIndex, DenseIndex nbRows, DenseIndex nbCols) { m_rows = nbRows; m_cols = nbCols; }
+    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 DenseStorage<T, Size, Dynamic, _Cols, _Options>
 {
     internal::plain_array<T,Size,_Options> m_data;
-    Index m_rows;
+    DenseIndex m_rows;
   public:
-    EIGEN_DEVICE_FUNC DenseStorage() : m_rows(0) {}
-    EIGEN_DEVICE_FUNC explicit DenseStorage(internal::constructor_without_unaligned_array_assert)
+    inline DenseStorage() : m_rows(0) {}
+    inline 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)
-    {
-      numext::swap(m_data,other.m_data);
-      numext::swap(m_rows,other.m_rows);
-    }
-    EIGEN_DEVICE_FUNC Index rows(void) const EIGEN_NOEXCEPT {return m_rows;}
-    EIGEN_DEVICE_FUNC EIGEN_CONSTEXPR Index cols(void) const EIGEN_NOEXCEPT {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; }
+    inline DenseStorage(DenseIndex, DenseIndex nbRows, DenseIndex) : m_rows(nbRows) {}
+    inline void swap(DenseStorage& 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 nbRows, DenseIndex) { m_rows = nbRows; }
+    inline void resize(DenseIndex, DenseIndex nbRows, DenseIndex) { m_rows = nbRows; }
+    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 DenseStorage<T, Size, _Rows, Dynamic, _Options>
 {
     internal::plain_array<T,Size,_Options> m_data;
-    Index m_cols;
+    DenseIndex m_cols;
   public:
-    EIGEN_DEVICE_FUNC DenseStorage() : m_cols(0) {}
-    EIGEN_DEVICE_FUNC explicit DenseStorage(internal::constructor_without_unaligned_array_assert)
+    inline DenseStorage() : m_cols(0) {}
+    inline 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) {
-      numext::swap(m_data,other.m_data);
-      numext::swap(m_cols,other.m_cols);
-    }
-    EIGEN_DEVICE_FUNC EIGEN_CONSTEXPR Index rows(void) const EIGEN_NOEXCEPT {return _Rows;}
-    EIGEN_DEVICE_FUNC Index cols(void) const EIGEN_NOEXCEPT {return m_cols;}
-    EIGEN_DEVICE_FUNC void conservativeResize(Index, Index, Index cols) { m_cols = cols; }
-    EIGEN_DEVICE_FUNC 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; }
+    inline DenseStorage(DenseIndex, DenseIndex, DenseIndex nbCols) : m_cols(nbCols) {}
+    inline void swap(DenseStorage& 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 nbCols) { m_cols = nbCols; }
+    inline void resize(DenseIndex, DenseIndex, DenseIndex nbCols) { m_cols = nbCols; }
+    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 DenseStorage<T, Dynamic, Dynamic, Dynamic, _Options>
 {
     T *m_data;
-    Index m_rows;
-    Index m_cols;
+    DenseIndex m_rows;
+    DenseIndex 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)
+    inline DenseStorage() : m_data(0), m_rows(0), m_cols(0) {}
+    inline DenseStorage(internal::constructor_without_unaligned_array_assert)
        : m_data(0), m_rows(0), m_cols(0) {}
-    EIGEN_DEVICE_FUNC DenseStorage(Index size, Index rows, Index cols)
-      : m_data(internal::conditional_aligned_new_auto<T,(_Options&DontAlign)==0>(size)), m_rows(rows), m_cols(cols)
-    {
-      EIGEN_INTERNAL_DENSE_STORAGE_CTOR_PLUGIN({})
-      eigen_internal_assert(size==rows*cols && rows>=0 && cols >=0);
-    }
-    EIGEN_DEVICE_FUNC DenseStorage(const DenseStorage& other)
-      : m_data(internal::conditional_aligned_new_auto<T,(_Options&DontAlign)==0>(other.m_rows*other.m_cols))
-      , m_rows(other.m_rows)
-      , m_cols(other.m_cols)
-    {
-      EIGEN_INTERNAL_DENSE_STORAGE_CTOR_PLUGIN(Index size = m_rows*m_cols)
-      internal::smart_copy(other.m_data, other.m_data+other.m_rows*other.m_cols, m_data);
-    }
-    EIGEN_DEVICE_FUNC DenseStorage& operator=(const DenseStorage& other)
-    {
-      if (this != &other)
-      {
-        DenseStorage tmp(other);
-        this->swap(tmp);
-      }
-      return *this;
-    }
-#if EIGEN_HAS_RVALUE_REFERENCES
-    EIGEN_DEVICE_FUNC
-    DenseStorage(DenseStorage&& other) EIGEN_NOEXCEPT
-      : m_data(std::move(other.m_data))
-      , m_rows(std::move(other.m_rows))
-      , m_cols(std::move(other.m_cols))
-    {
-      other.m_data = nullptr;
-      other.m_rows = 0;
-      other.m_cols = 0;
-    }
-    EIGEN_DEVICE_FUNC
-    DenseStorage& operator=(DenseStorage&& other) EIGEN_NOEXCEPT
-    {
-      numext::swap(m_data, other.m_data);
-      numext::swap(m_rows, other.m_rows);
-      numext::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)
-    {
-      numext::swap(m_data,other.m_data);
-      numext::swap(m_rows,other.m_rows);
-      numext::swap(m_cols,other.m_cols);
-    }
-    EIGEN_DEVICE_FUNC Index rows(void) const EIGEN_NOEXCEPT {return m_rows;}
-    EIGEN_DEVICE_FUNC Index cols(void) const EIGEN_NOEXCEPT {return m_cols;}
-    void conservativeResize(Index size, Index rows, Index cols)
+    inline DenseStorage(DenseIndex size, DenseIndex nbRows, DenseIndex nbCols)
+      : m_data(internal::conditional_aligned_new_auto<T,(_Options&DontAlign)==0>(size)), m_rows(nbRows), m_cols(nbCols)
+    { EIGEN_INTERNAL_DENSE_STORAGE_CTOR_PLUGIN }
+    inline ~DenseStorage() { internal::conditional_aligned_delete_auto<T,(_Options&DontAlign)==0>(m_data, m_rows*m_cols); }
+    inline 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); }
+    inline DenseIndex rows(void) const {return m_rows;}
+    inline DenseIndex cols(void) const {return m_cols;}
+    inline void conservativeResize(DenseIndex size, DenseIndex nbRows, DenseIndex nbCols)
     {
       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;
+      m_rows = nbRows;
+      m_cols = nbCols;
     }
-    EIGEN_DEVICE_FUNC void resize(Index size, Index rows, Index cols)
+    void resize(DenseIndex size, DenseIndex nbRows, DenseIndex nbCols)
     {
       if(size != m_rows*m_cols)
       {
         internal::conditional_aligned_delete_auto<T,(_Options&DontAlign)==0>(m_data, m_rows*m_cols);
-        if (size>0) // >0 and not simply !=0 to let the compiler knows that size cannot be negative
+        if (size)
           m_data = internal::conditional_aligned_new_auto<T,(_Options&DontAlign)==0>(size);
         else
           m_data = 0;
-        EIGEN_INTERNAL_DENSE_STORAGE_CTOR_PLUGIN({})
+        EIGEN_INTERNAL_DENSE_STORAGE_CTOR_PLUGIN
       }
-      m_rows = rows;
-      m_cols = cols;
+      m_rows = nbRows;
+      m_cols = nbCols;
     }
-    EIGEN_DEVICE_FUNC const T *data() const { return m_data; }
-    EIGEN_DEVICE_FUNC T *data() { return m_data; }
+    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 DenseStorage<T, Dynamic, _Rows, Dynamic, _Options>
 {
     T *m_data;
-    Index m_cols;
+    DenseIndex m_cols;
   public:
-    EIGEN_DEVICE_FUNC DenseStorage() : m_data(0), m_cols(0) {}
-    explicit DenseStorage(internal::constructor_without_unaligned_array_assert) : m_data(0), m_cols(0) {}
-    EIGEN_DEVICE_FUNC DenseStorage(Index size, Index rows, Index cols) : m_data(internal::conditional_aligned_new_auto<T,(_Options&DontAlign)==0>(size)), m_cols(cols)
-    {
-      EIGEN_INTERNAL_DENSE_STORAGE_CTOR_PLUGIN({})
-      eigen_internal_assert(size==rows*cols && rows==_Rows && cols >=0);
-      EIGEN_UNUSED_VARIABLE(rows);
-    }
-    EIGEN_DEVICE_FUNC DenseStorage(const DenseStorage& other)
-      : m_data(internal::conditional_aligned_new_auto<T,(_Options&DontAlign)==0>(_Rows*other.m_cols))
-      , m_cols(other.m_cols)
-    {
-      EIGEN_INTERNAL_DENSE_STORAGE_CTOR_PLUGIN(Index size = m_cols*_Rows)
-      internal::smart_copy(other.m_data, other.m_data+_Rows*m_cols, m_data);
-    }
-    EIGEN_DEVICE_FUNC DenseStorage& operator=(const DenseStorage& other)
-    {
-      if (this != &other)
-      {
-        DenseStorage tmp(other);
-        this->swap(tmp);
-      }
-      return *this;
-    }
-#if EIGEN_HAS_RVALUE_REFERENCES
-    EIGEN_DEVICE_FUNC
-    DenseStorage(DenseStorage&& other) EIGEN_NOEXCEPT
-      : m_data(std::move(other.m_data))
-      , m_cols(std::move(other.m_cols))
-    {
-      other.m_data = nullptr;
-      other.m_cols = 0;
-    }
-    EIGEN_DEVICE_FUNC
-    DenseStorage& operator=(DenseStorage&& other) EIGEN_NOEXCEPT
-    {
-      numext::swap(m_data, other.m_data);
-      numext::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) {
-      numext::swap(m_data,other.m_data);
-      numext::swap(m_cols,other.m_cols);
-    }
-    EIGEN_DEVICE_FUNC static EIGEN_CONSTEXPR Index rows(void) EIGEN_NOEXCEPT {return _Rows;}
-    EIGEN_DEVICE_FUNC Index cols(void) const EIGEN_NOEXCEPT {return m_cols;}
-    EIGEN_DEVICE_FUNC void conservativeResize(Index size, Index, Index cols)
+    inline DenseStorage() : m_data(0), m_cols(0) {}
+    inline DenseStorage(internal::constructor_without_unaligned_array_assert) : m_data(0), m_cols(0) {}
+    inline DenseStorage(DenseIndex size, DenseIndex, DenseIndex nbCols) : m_data(internal::conditional_aligned_new_auto<T,(_Options&DontAlign)==0>(size)), m_cols(nbCols)
+    { EIGEN_INTERNAL_DENSE_STORAGE_CTOR_PLUGIN }
+    inline ~DenseStorage() { internal::conditional_aligned_delete_auto<T,(_Options&DontAlign)==0>(m_data, _Rows*m_cols); }
+    inline void swap(DenseStorage& other) { std::swap(m_data,other.m_data); std::swap(m_cols,other.m_cols); }
+    static inline DenseIndex rows(void) {return _Rows;}
+    inline DenseIndex cols(void) const {return m_cols;}
+    inline void conservativeResize(DenseIndex size, DenseIndex, DenseIndex nbCols)
     {
       m_data = internal::conditional_aligned_realloc_new_auto<T,(_Options&DontAlign)==0>(m_data, size, _Rows*m_cols);
-      m_cols = cols;
+      m_cols = nbCols;
     }
-    EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE void resize(Index size, Index, Index cols)
+    EIGEN_STRONG_INLINE void resize(DenseIndex size, DenseIndex, DenseIndex nbCols)
     {
       if(size != _Rows*m_cols)
       {
         internal::conditional_aligned_delete_auto<T,(_Options&DontAlign)==0>(m_data, _Rows*m_cols);
-        if (size>0) // >0 and not simply !=0 to let the compiler knows that size cannot be negative
+        if (size)
           m_data = internal::conditional_aligned_new_auto<T,(_Options&DontAlign)==0>(size);
         else
           m_data = 0;
-        EIGEN_INTERNAL_DENSE_STORAGE_CTOR_PLUGIN({})
+        EIGEN_INTERNAL_DENSE_STORAGE_CTOR_PLUGIN
       }
-      m_cols = cols;
+      m_cols = nbCols;
     }
-    EIGEN_DEVICE_FUNC const T *data() const { return m_data; }
-    EIGEN_DEVICE_FUNC T *data() { return m_data; }
+    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 DenseStorage<T, Dynamic, Dynamic, _Cols, _Options>
 {
     T *m_data;
-    Index m_rows;
+    DenseIndex m_rows;
   public:
-    EIGEN_DEVICE_FUNC DenseStorage() : m_data(0), m_rows(0) {}
-    explicit DenseStorage(internal::constructor_without_unaligned_array_assert) : m_data(0), m_rows(0) {}
-    EIGEN_DEVICE_FUNC DenseStorage(Index size, Index rows, Index cols) : m_data(internal::conditional_aligned_new_auto<T,(_Options&DontAlign)==0>(size)), m_rows(rows)
-    {
-      EIGEN_INTERNAL_DENSE_STORAGE_CTOR_PLUGIN({})
-      eigen_internal_assert(size==rows*cols && rows>=0 && cols == _Cols);
-      EIGEN_UNUSED_VARIABLE(cols);
-    }
-    EIGEN_DEVICE_FUNC DenseStorage(const DenseStorage& other)
-      : m_data(internal::conditional_aligned_new_auto<T,(_Options&DontAlign)==0>(other.m_rows*_Cols))
-      , m_rows(other.m_rows)
-    {
-      EIGEN_INTERNAL_DENSE_STORAGE_CTOR_PLUGIN(Index size = m_rows*_Cols)
-      internal::smart_copy(other.m_data, other.m_data+other.m_rows*_Cols, m_data);
-    }
-    EIGEN_DEVICE_FUNC DenseStorage& operator=(const DenseStorage& other)
-    {
-      if (this != &other)
-      {
-        DenseStorage tmp(other);
-        this->swap(tmp);
-      }
-      return *this;
-    }
-#if EIGEN_HAS_RVALUE_REFERENCES
-    EIGEN_DEVICE_FUNC
-    DenseStorage(DenseStorage&& other) EIGEN_NOEXCEPT
-      : m_data(std::move(other.m_data))
-      , m_rows(std::move(other.m_rows))
-    {
-      other.m_data = nullptr;
-      other.m_rows = 0;
-    }
-    EIGEN_DEVICE_FUNC
-    DenseStorage& operator=(DenseStorage&& other) EIGEN_NOEXCEPT
-    {
-      numext::swap(m_data, other.m_data);
-      numext::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) {
-      numext::swap(m_data,other.m_data);
-      numext::swap(m_rows,other.m_rows);
-    }
-    EIGEN_DEVICE_FUNC Index rows(void) const EIGEN_NOEXCEPT {return m_rows;}
-    EIGEN_DEVICE_FUNC static EIGEN_CONSTEXPR Index cols(void) {return _Cols;}
-    void conservativeResize(Index size, Index rows, Index)
+    inline DenseStorage() : m_data(0), m_rows(0) {}
+    inline DenseStorage(internal::constructor_without_unaligned_array_assert) : m_data(0), m_rows(0) {}
+    inline DenseStorage(DenseIndex size, DenseIndex nbRows, DenseIndex) : m_data(internal::conditional_aligned_new_auto<T,(_Options&DontAlign)==0>(size)), m_rows(nbRows)
+    { EIGEN_INTERNAL_DENSE_STORAGE_CTOR_PLUGIN }
+    inline ~DenseStorage() { internal::conditional_aligned_delete_auto<T,(_Options&DontAlign)==0>(m_data, _Cols*m_rows); }
+    inline void swap(DenseStorage& other) { std::swap(m_data,other.m_data); std::swap(m_rows,other.m_rows); }
+    inline DenseIndex rows(void) const {return m_rows;}
+    static inline DenseIndex cols(void) {return _Cols;}
+    inline void conservativeResize(DenseIndex size, DenseIndex nbRows, DenseIndex)
     {
       m_data = internal::conditional_aligned_realloc_new_auto<T,(_Options&DontAlign)==0>(m_data, size, m_rows*_Cols);
-      m_rows = rows;
+      m_rows = nbRows;
     }
-    EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE void resize(Index size, Index rows, Index)
+    EIGEN_STRONG_INLINE void resize(DenseIndex size, DenseIndex nbRows, DenseIndex)
     {
       if(size != m_rows*_Cols)
       {
         internal::conditional_aligned_delete_auto<T,(_Options&DontAlign)==0>(m_data, _Cols*m_rows);
-        if (size>0) // >0 and not simply !=0 to let the compiler knows that size cannot be negative
+        if (size)
           m_data = internal::conditional_aligned_new_auto<T,(_Options&DontAlign)==0>(size);
         else
           m_data = 0;
-        EIGEN_INTERNAL_DENSE_STORAGE_CTOR_PLUGIN({})
+        EIGEN_INTERNAL_DENSE_STORAGE_CTOR_PLUGIN
       }
-      m_rows = rows;
+      m_rows = nbRows;
     }
-    EIGEN_DEVICE_FUNC const T *data() const { return m_data; }
-    EIGEN_DEVICE_FUNC T *data() { return m_data; }
+    inline const T *data() const { return m_data; }
+    inline T *data() { return m_data; }
 };
 
 } // end namespace Eigen

Eigen/src/Core/Diagonal.h

@@ -11,7 +11,7 @@
 #ifndef EIGEN_DIAGONAL_H
 #define EIGEN_DIAGONAL_H
 
-namespace Eigen {
+namespace Eigen { 
 
 /** \class Diagonal
   * \ingroup Core_Module
@@ -21,7 +21,7 @@ namespace Eigen {
   * \param MatrixType the type of the object in which we are taking a sub/main/super diagonal
   * \param DiagIndex the index of the sub/super diagonal. The default is 0 and it means the main diagonal.
   *              A positive value means a superdiagonal, a negative value means a subdiagonal.
-  *              You can also use DynamicIndex so the index can be set at runtime.
+  *              You can also use Dynamic so the index can be set at runtime.
   *
   * The matrix is not required to be square.
   *
@@ -37,7 +37,7 @@ template<typename MatrixType, int DiagIndex>
 struct traits<Diagonal<MatrixType,DiagIndex> >
  : traits<MatrixType>
 {
-  typedef typename ref_selector<MatrixType>::type MatrixTypeNested;
+  typedef typename nested<MatrixType>::type MatrixTypeNested;
   typedef typename remove_reference<MatrixTypeNested>::type _MatrixTypeNested;
   typedef typename MatrixType::StorageKind StorageKind;
   enum {
@@ -52,7 +52,8 @@ struct traits<Diagonal<MatrixType,DiagIndex> >
                                                  MatrixType::MaxColsAtCompileTime - EIGEN_PLAIN_ENUM_MAX( DiagIndex, 0))),
     MaxColsAtCompileTime = 1,
     MaskLvalueBit = is_lvalue<MatrixType>::value ? LvalueBit : 0,
-    Flags = (unsigned int)_MatrixTypeNested::Flags & (RowMajorBit | MaskLvalueBit | DirectAccessBit) & ~RowMajorBit, // FIXME DirectAccessBit should not be handled by expressions
+    Flags = (unsigned int)_MatrixTypeNested::Flags & (HereditaryBits | LinearAccessBit | MaskLvalueBit | DirectAccessBit) & ~RowMajorBit,
+    CoeffReadCost = _MatrixTypeNested::CoeffReadCost,
     MatrixTypeOuterStride = outer_stride_at_compile_time<MatrixType>::ret,
     InnerStrideAtCompileTime = MatrixTypeOuterStride == Dynamic ? Dynamic : MatrixTypeOuterStride+1,
     OuterStrideAtCompileTime = 0
@@ -69,31 +70,24 @@ template<typename MatrixType, int _DiagIndex> class Diagonal
     typedef typename internal::dense_xpr_base<Diagonal>::type Base;
     EIGEN_DENSE_PUBLIC_INTERFACE(Diagonal)
 
-    EIGEN_DEVICE_FUNC
-    explicit inline Diagonal(MatrixType& matrix, Index a_index = DiagIndex) : m_matrix(matrix), m_index(a_index)
-    {
-      eigen_assert( a_index <= m_matrix.cols() && -a_index <= m_matrix.rows() );
-    }
+    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<Index>)(m_matrix.cols(),m_matrix.rows()+m_index.value()) : (std::min<Index>)(m_matrix.rows(),m_matrix.cols()-m_index.value()); }
+
+    inline Index cols() const { return 1; }
+
+    inline Index innerStride() const
     {
-      return m_index.value()<0 ? numext::mini<Index>(m_matrix.cols(),m_matrix.rows()+m_index.value())
-                               : numext::mini<Index>(m_matrix.rows(),m_matrix.cols()-m_index.value());
-    }
-
-    EIGEN_DEVICE_FUNC EIGEN_CONSTEXPR
-    inline Index cols() const EIGEN_NOEXCEPT { return 1; }
-
-    EIGEN_DEVICE_FUNC EIGEN_CONSTEXPR
-    inline Index innerStride() const EIGEN_NOEXCEPT {
       return m_matrix.outerStride() + 1;
     }
 
-    EIGEN_DEVICE_FUNC EIGEN_CONSTEXPR
-    inline Index outerStride() const EIGEN_NOEXCEPT { return 0; }
+    inline Index outerStride() const
+    {
+      return 0;
+    }
 
     typedef typename internal::conditional<
                        internal::is_lvalue<MatrixType>::value,
@@ -101,75 +95,62 @@ template<typename MatrixType, int _DiagIndex> class Diagonal
                        const Scalar
                      >::type ScalarWithConstIfNotLvalue;
 
-    EIGEN_DEVICE_FUNC
-    inline ScalarWithConstIfNotLvalue* data() { return &(m_matrix.coeffRef(rowOffset(), colOffset())); }
-    EIGEN_DEVICE_FUNC
-    inline const Scalar* data() const { return &(m_matrix.coeffRef(rowOffset(), colOffset())); }
+    inline ScalarWithConstIfNotLvalue* data() { return &(m_matrix.const_cast_derived().coeffRef(rowOffset(), colOffset())); }
+    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.coeffRef(row+rowOffset(), row+colOffset());
+      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.coeffRef(row+rowOffset(), row+colOffset());
+      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());
     }
 
-    EIGEN_DEVICE_FUNC
     inline Scalar& coeffRef(Index idx)
     {
       EIGEN_STATIC_ASSERT_LVALUE(MatrixType)
-      return m_matrix.coeffRef(idx+rowOffset(), idx+colOffset());
+      return m_matrix.const_cast_derived().coeffRef(idx+rowOffset(), idx+colOffset());
     }
 
-    EIGEN_DEVICE_FUNC
     inline const Scalar& coeffRef(Index idx) const
     {
-      return m_matrix.coeffRef(idx+rowOffset(), idx+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
+    const typename internal::remove_all<typename MatrixType::Nested>::type& 
+    nestedExpression() const 
     {
       return m_matrix;
     }
 
-    EIGEN_DEVICE_FUNC
-    inline Index index() const
+    int index() const
     {
       return m_index.value();
     }
 
   protected:
-    typename internal::ref_selector<MatrixType>::non_const_type m_matrix;
+    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 EIGEN_CONSTEXPR
-    Index absDiagIndex() const EIGEN_NOEXCEPT { return m_index.value()>0 ? m_index.value() : -m_index.value(); }
-    EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE EIGEN_CONSTEXPR
-    Index rowOffset() const EIGEN_NOEXCEPT { return m_index.value()>0 ? 0 : -m_index.value(); }
-    EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE EIGEN_CONSTEXPR
-    Index colOffset() const EIGEN_NOEXCEPT { return m_index.value()>0 ? m_index.value() : 0; }
-    // trigger a compile-time error if someone try to call packet
+    EIGEN_STRONG_INLINE Index absDiagIndex() const { return m_index.value()>0 ? m_index.value() : -m_index.value(); }
+    EIGEN_STRONG_INLINE Index rowOffset() const { return m_index.value()>0 ? 0 : -m_index.value(); }
+    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
     template<int LoadMode> typename MatrixType::PacketReturnType packet(Index) const;
     template<int LoadMode> typename MatrixType::PacketReturnType packet(Index,Index) const;
 };
@@ -183,15 +164,15 @@ template<typename MatrixType, int _DiagIndex> class Diagonal
   *
   * \sa class Diagonal */
 template<typename Derived>
-EIGEN_DEVICE_FUNC inline typename MatrixBase<Derived>::DiagonalReturnType
+inline typename MatrixBase<Derived>::DiagonalReturnType
 MatrixBase<Derived>::diagonal()
 {
-  return DiagonalReturnType(derived());
+  return derived();
 }
 
 /** This is the const version of diagonal(). */
 template<typename Derived>
-EIGEN_DEVICE_FUNC inline typename MatrixBase<Derived>::ConstDiagonalReturnType
+inline typename MatrixBase<Derived>::ConstDiagonalReturnType
 MatrixBase<Derived>::diagonal() const
 {
   return ConstDiagonalReturnType(derived());
@@ -209,18 +190,18 @@ MatrixBase<Derived>::diagonal() const
   *
   * \sa MatrixBase::diagonal(), class Diagonal */
 template<typename Derived>
-EIGEN_DEVICE_FUNC inline typename MatrixBase<Derived>::DiagonalDynamicIndexReturnType
+inline typename MatrixBase<Derived>::template DiagonalIndexReturnType<DynamicIndex>::Type
 MatrixBase<Derived>::diagonal(Index index)
 {
-  return DiagonalDynamicIndexReturnType(derived(), index);
+  return typename DiagonalIndexReturnType<DynamicIndex>::Type(derived(), index);
 }
 
 /** This is the const version of diagonal(Index). */
 template<typename Derived>
-EIGEN_DEVICE_FUNC inline typename MatrixBase<Derived>::ConstDiagonalDynamicIndexReturnType
+inline typename MatrixBase<Derived>::template ConstDiagonalIndexReturnType<DynamicIndex>::Type
 MatrixBase<Derived>::diagonal(Index index) const
 {
-  return ConstDiagonalDynamicIndexReturnType(derived(), index);
+  return typename ConstDiagonalIndexReturnType<DynamicIndex>::Type(derived(), index);
 }
 
 /** \returns an expression of the \a DiagIndex-th sub or super diagonal of the matrix \c *this
@@ -235,22 +216,20 @@ MatrixBase<Derived>::diagonal(Index index) const
   *
   * \sa MatrixBase::diagonal(), class Diagonal */
 template<typename Derived>
-template<int Index_>
-EIGEN_DEVICE_FUNC
-inline typename MatrixBase<Derived>::template DiagonalIndexReturnType<Index_>::Type
+template<int Index>
+inline typename MatrixBase<Derived>::template DiagonalIndexReturnType<Index>::Type
 MatrixBase<Derived>::diagonal()
 {
-  return typename DiagonalIndexReturnType<Index_>::Type(derived());
+  return derived();
 }
 
 /** This is the const version of diagonal<int>(). */
 template<typename Derived>
-template<int Index_>
-EIGEN_DEVICE_FUNC
-inline typename MatrixBase<Derived>::template ConstDiagonalIndexReturnType<Index_>::Type
+template<int Index>
+inline typename MatrixBase<Derived>::template ConstDiagonalIndexReturnType<Index>::Type
 MatrixBase<Derived>::diagonal() const
 {
-  return typename ConstDiagonalIndexReturnType<Index_>::Type(derived());
+  return derived();
 }
 
 } // end namespace Eigen

Eigen/src/Core/DiagonalMatrix.h

@@ -22,7 +22,7 @@ class DiagonalBase : public EigenBase<Derived>
     typedef typename DiagonalVectorType::Scalar Scalar;
     typedef typename DiagonalVectorType::RealScalar RealScalar;
     typedef typename internal::traits<Derived>::StorageKind StorageKind;
-    typedef typename internal::traits<Derived>::StorageIndex StorageIndex;
+    typedef typename internal::traits<Derived>::Index Index;
 
     enum {
       RowsAtCompileTime = DiagonalVectorType::SizeAtCompileTime,
@@ -30,85 +30,79 @@ class DiagonalBase : public EigenBase<Derived>
       MaxRowsAtCompileTime = DiagonalVectorType::MaxSizeAtCompileTime,
       MaxColsAtCompileTime = DiagonalVectorType::MaxSizeAtCompileTime,
       IsVectorAtCompileTime = 0,
-      Flags = NoPreferredStorageOrderBit
+      Flags = 0
     };
 
     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(); }
 
+    /** \returns the diagonal matrix product of \c *this by the matrix \a matrix.
+      */
     template<typename MatrixDerived>
-    EIGEN_DEVICE_FUNC
-    const Product<Derived,MatrixDerived,LazyProduct>
+    const DiagonalProduct<MatrixDerived, Derived, OnTheLeft>
     operator*(const MatrixBase<MatrixDerived> &matrix) const
     {
-      return Product<Derived, MatrixDerived, LazyProduct>(derived(),matrix.derived());
+      return DiagonalProduct<MatrixDerived, Derived, OnTheLeft>(matrix.derived(), derived());
     }
 
-    typedef DiagonalWrapper<const CwiseUnaryOp<internal::scalar_inverse_op<Scalar>, const DiagonalVectorType> > InverseReturnType;
-    EIGEN_DEVICE_FUNC
-    inline const InverseReturnType
+    inline const DiagonalWrapper<const CwiseUnaryOp<internal::scalar_inverse_op<Scalar>, const DiagonalVectorType> >
     inverse() const
     {
-      return InverseReturnType(diagonal().cwiseInverse());
+      return diagonal().cwiseInverse();
     }
     
-    EIGEN_DEVICE_FUNC
-    inline const DiagonalWrapper<const EIGEN_EXPR_BINARYOP_SCALAR_RETURN_TYPE(DiagonalVectorType,Scalar,product) >
+    inline const DiagonalWrapper<const CwiseUnaryOp<internal::scalar_multiple_op<Scalar>, const DiagonalVectorType> >
     operator*(const Scalar& scalar) const
     {
-      return DiagonalWrapper<const EIGEN_EXPR_BINARYOP_SCALAR_RETURN_TYPE(DiagonalVectorType,Scalar,product) >(diagonal() * scalar);
+      return diagonal() * scalar;
     }
-    EIGEN_DEVICE_FUNC
-    friend inline const DiagonalWrapper<const EIGEN_SCALAR_BINARYOP_EXPR_RETURN_TYPE(Scalar,DiagonalVectorType,product) >
+    friend inline const DiagonalWrapper<const CwiseUnaryOp<internal::scalar_multiple_op<Scalar>, const DiagonalVectorType> >
     operator*(const Scalar& scalar, const DiagonalBase& other)
     {
-      return DiagonalWrapper<const EIGEN_SCALAR_BINARYOP_EXPR_RETURN_TYPE(Scalar,DiagonalVectorType,product) >(scalar * other.diagonal());
+      return other.diagonal() * scalar;
     }
-
+    
+    #ifdef EIGEN2_SUPPORT
     template<typename OtherDerived>
-    EIGEN_DEVICE_FUNC
-    #ifdef EIGEN_PARSED_BY_DOXYGEN
-    inline unspecified_expression_type
-    #else
-    inline const DiagonalWrapper<const EIGEN_CWISE_BINARY_RETURN_TYPE(DiagonalVectorType,typename OtherDerived::DiagonalVectorType,sum) >
-    #endif
-    operator+(const DiagonalBase<OtherDerived>& other) const
+    bool isApprox(const DiagonalBase<OtherDerived>& other, typename NumTraits<Scalar>::Real precision = NumTraits<Scalar>::dummy_precision()) const
     {
-      return (diagonal() + other.diagonal()).asDiagonal();
+      return diagonal().isApprox(other.diagonal(), precision);
     }
-
     template<typename OtherDerived>
-    EIGEN_DEVICE_FUNC
-    #ifdef EIGEN_PARSED_BY_DOXYGEN
-    inline unspecified_expression_type
-    #else
-    inline const DiagonalWrapper<const EIGEN_CWISE_BINARY_RETURN_TYPE(DiagonalVectorType,typename OtherDerived::DiagonalVectorType,difference) >
-    #endif
-    operator-(const DiagonalBase<OtherDerived>& other) const
+    bool isApprox(const MatrixBase<OtherDerived>& other, typename NumTraits<Scalar>::Real precision = NumTraits<Scalar>::dummy_precision()) const
     {
-      return (diagonal() - other.diagonal()).asDiagonal();
+      return toDenseMatrix().isApprox(other, precision);
     }
+    #endif
 };
 
+template<typename Derived>
+template<typename DenseDerived>
+void DiagonalBase<Derived>::evalTo(MatrixBase<DenseDerived> &other) const
+{
+  other.setZero();
+  other.diagonal() = diagonal();
+}
 #endif
 
 /** \class DiagonalMatrix
@@ -130,9 +124,10 @@ struct traits<DiagonalMatrix<_Scalar,SizeAtCompileTime,MaxSizeAtCompileTime> >
  : traits<Matrix<_Scalar,SizeAtCompileTime,SizeAtCompileTime,0,MaxSizeAtCompileTime,MaxSizeAtCompileTime> >
 {
   typedef Matrix<_Scalar,SizeAtCompileTime,1,0,MaxSizeAtCompileTime,1> DiagonalVectorType;
-  typedef DiagonalShape StorageKind;
+  typedef Dense StorageKind;
+  typedef DenseIndex Index;
   enum {
-    Flags = LvalueBit | NoPreferredStorageOrderBit
+    Flags = LvalueBit
   };
 };
 }
@@ -146,7 +141,7 @@ class DiagonalMatrix
     typedef const DiagonalMatrix& Nested;
     typedef _Scalar Scalar;
     typedef typename internal::traits<DiagonalMatrix>::StorageKind StorageKind;
-    typedef typename internal::traits<DiagonalMatrix>::StorageIndex StorageIndex;
+    typedef typename internal::traits<DiagonalMatrix>::Index Index;
     #endif
 
   protected:
@@ -156,55 +151,24 @@ 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  */
-    EIGEN_DEVICE_FUNC
-    explicit inline DiagonalMatrix(Index dim) : m_diagonal(dim) {}
+    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) {}
 
-    #if EIGEN_HAS_CXX11
-    /** \brief Construct a diagonal matrix with fixed size from an arbitrary number of coefficients. \cpp11
-      * 
-      * There exists C++98 anologue constructors for fixed-size diagonal matrices having 2 or 3 coefficients.
-      * 
-      * \warning To construct a diagonal matrix of fixed size, the number of values passed to this 
-      * constructor must match the fixed dimension of \c *this.
-      * 
-      * \sa DiagonalMatrix(const Scalar&, const Scalar&)
-      * \sa DiagonalMatrix(const Scalar&, const Scalar&, const Scalar&)
-      */
-    template <typename... ArgTypes>
-    EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
-    DiagonalMatrix(const Scalar& a0, const Scalar& a1, const Scalar& a2, const ArgTypes&... args)
-      : m_diagonal(a0, a1, a2, args...) {}
-
-    /** \brief Constructs a DiagonalMatrix and initializes it by elements given by an initializer list of initializer
-      * lists \cpp11
-      */
-    EIGEN_DEVICE_FUNC
-    explicit EIGEN_STRONG_INLINE DiagonalMatrix(const std::initializer_list<std::initializer_list<Scalar>>& list)
-      : m_diagonal(list) {}
-    #endif  // EIGEN_HAS_CXX11
-
     /** Copy constructor. */
     template<typename OtherDerived>
-    EIGEN_DEVICE_FUNC
     inline DiagonalMatrix(const DiagonalBase<OtherDerived>& other) : m_diagonal(other.diagonal()) {}
 
     #ifndef EIGEN_PARSED_BY_DOXYGEN
@@ -214,13 +178,11 @@ 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();
@@ -231,7 +193,6 @@ 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();
@@ -240,19 +201,14 @@ 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); }
 };
 
@@ -276,15 +232,14 @@ struct traits<DiagonalWrapper<_DiagonalVectorType> >
 {
   typedef _DiagonalVectorType DiagonalVectorType;
   typedef typename DiagonalVectorType::Scalar Scalar;
-  typedef typename DiagonalVectorType::StorageIndex StorageIndex;
-  typedef DiagonalShape StorageKind;
-  typedef typename traits<DiagonalVectorType>::XprKind XprKind;
+  typedef typename DiagonalVectorType::Index Index;
+  typedef typename DiagonalVectorType::StorageKind StorageKind;
   enum {
     RowsAtCompileTime = DiagonalVectorType::SizeAtCompileTime,
     ColsAtCompileTime = DiagonalVectorType::SizeAtCompileTime,
-    MaxRowsAtCompileTime = DiagonalVectorType::MaxSizeAtCompileTime,
-    MaxColsAtCompileTime = DiagonalVectorType::MaxSizeAtCompileTime,
-    Flags =  (traits<DiagonalVectorType>::Flags & LvalueBit) | NoPreferredStorageOrderBit
+    MaxRowsAtCompileTime = DiagonalVectorType::SizeAtCompileTime,
+    MaxColsAtCompileTime = DiagonalVectorType::SizeAtCompileTime,
+    Flags =  traits<DiagonalVectorType>::Flags & LvalueBit
   };
 };
 }
@@ -300,11 +255,9 @@ class DiagonalWrapper
     #endif
 
     /** Constructor from expression of diagonal coefficients to wrap. */
-    EIGEN_DEVICE_FUNC
-    explicit inline DiagonalWrapper(DiagonalVectorType& a_diagonal) : m_diagonal(a_diagonal) {}
+    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:
@@ -321,10 +274,10 @@ class DiagonalWrapper
   * \sa class DiagonalWrapper, class DiagonalMatrix, diagonal(), isDiagonal()
   **/
 template<typename Derived>
-EIGEN_DEVICE_FUNC inline const DiagonalWrapper<const Derived>
+inline const DiagonalWrapper<const Derived>
 MatrixBase<Derived>::asDiagonal() const
 {
-  return DiagonalWrapper<const Derived>(derived());
+  return derived();
 }
 
 /** \returns true if *this is approximately equal to a diagonal matrix,
@@ -338,11 +291,12 @@ MatrixBase<Derived>::asDiagonal() const
 template<typename Derived>
 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 = numext::abs(coeff(j,j));
+    RealScalar absOnDiagonal = abs(coeff(j,j));
     if(absOnDiagonal > maxAbsOnDiagonal) maxAbsOnDiagonal = absOnDiagonal;
   }
   for(Index j = 0; j < cols(); ++j)
@@ -354,38 +308,6 @@ bool MatrixBase<Derived>::isDiagonal(const RealScalar& prec) const
   return true;
 }
 
-namespace internal {
-
-template<> struct storage_kind_to_shape<DiagonalShape> { typedef DiagonalShape Shape; };
-
-struct Diagonal2Dense {};
-
-template<> struct AssignmentKind<DenseShape,DiagonalShape> { typedef Diagonal2Dense Kind; };
-
-// Diagonal matrix to Dense assignment
-template< typename DstXprType, typename SrcXprType, typename Functor>
-struct Assignment<DstXprType, SrcXprType, Functor, Diagonal2Dense>
-{
-  static void run(DstXprType &dst, const SrcXprType &src, const internal::assign_op<typename DstXprType::Scalar,typename SrcXprType::Scalar> &/*func*/)
-  {
-    Index dstRows = src.rows();
-    Index dstCols = src.cols();
-    if((dst.rows()!=dstRows) || (dst.cols()!=dstCols))
-      dst.resize(dstRows, dstCols);
-    
-    dst.setZero();
-    dst.diagonal() = src.diagonal();
-  }
-  
-  static void run(DstXprType &dst, const SrcXprType &src, const internal::add_assign_op<typename DstXprType::Scalar,typename SrcXprType::Scalar> &/*func*/)
-  { dst.diagonal() += src.diagonal(); }
-  
-  static void run(DstXprType &dst, const SrcXprType &src, const internal::sub_assign_op<typename DstXprType::Scalar,typename SrcXprType::Scalar> &/*func*/)
-  { dst.diagonal() -= src.diagonal(); }
-};
-
-} // namespace internal
-
 } // end namespace Eigen
 
 #endif // EIGEN_DIAGONALMATRIX_H

Eigen/src/Core/DiagonalProduct.h

@@ -13,14 +13,116 @@
 
 namespace Eigen { 
 
+namespace internal {
+template<typename MatrixType, typename DiagonalType, int ProductOrder>
+struct traits<DiagonalProduct<MatrixType, DiagonalType, ProductOrder> >
+ : traits<MatrixType>
+{
+  typedef typename 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,
+    _ScalarAccessOnDiag =  !((int(_StorageOrder) == ColMajor && int(ProductOrder) == OnTheLeft)
+                          ||(int(_StorageOrder) == RowMajor && int(ProductOrder) == OnTheRight)),
+    _SameTypes = is_same<typename MatrixType::Scalar, typename DiagonalType::Scalar>::value,
+    // 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::DiagonalVectorType::Flags)&PacketAccessBit))),
+    _Vectorizable = bool(int(MatrixType::Flags)&PacketAccessBit) && _SameTypes && (_ScalarAccessOnDiag || (bool(int(DiagonalType::DiagonalVectorType::Flags)&PacketAccessBit))),
+    _LinearAccessMask = (RowsAtCompileTime==1 || ColsAtCompileTime==1) ? LinearAccessBit : 0,
+
+    Flags = ((HereditaryBits|_LinearAccessMask) & (unsigned int)(MatrixType::Flags)) | (_Vectorizable ? PacketAccessBit : 0) | AlignedBit,//(int(MatrixType::Flags)&int(DiagonalType::DiagonalVectorType::Flags)&AlignedBit),
+    CoeffReadCost = NumTraits<Scalar>::MulCost + MatrixType::CoeffReadCost + DiagonalType::DiagonalVectorType::CoeffReadCost
+  };
+};
+}
+
+template<typename MatrixType, typename DiagonalType, int ProductOrder>
+class DiagonalProduct : internal::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)
+    {
+      eigen_assert(diagonal.diagonal().size() == (ProductOrder == OnTheLeft ? matrix.rows() : matrix.cols()));
+    }
+
+    EIGEN_STRONG_INLINE Index rows() const { return m_matrix.rows(); }
+    EIGEN_STRONG_INLINE Index cols() const { return m_matrix.cols(); }
+
+    EIGEN_STRONG_INLINE const Scalar coeff(Index row, Index col) const
+    {
+      return m_diagonal.diagonal().coeff(ProductOrder == OnTheLeft ? row : col) * m_matrix.coeff(row, col);
+    }
+    
+    EIGEN_STRONG_INLINE const Scalar coeff(Index idx) const
+    {
+      enum {
+        StorageOrder = int(MatrixType::Flags) & RowMajorBit ? RowMajor : ColMajor
+      };
+      return coeff(int(StorageOrder)==ColMajor?idx:0,int(StorageOrder)==ColMajor?0:idx);
+    }
+
+    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 internal::conditional<
+        ((int(StorageOrder) == RowMajor && int(ProductOrder) == OnTheLeft)
+       ||(int(StorageOrder) == ColMajor && int(ProductOrder) == OnTheRight)), internal::true_type, internal::false_type>::type());
+    }
+    
+    template<int LoadMode>
+    EIGEN_STRONG_INLINE PacketScalar packet(Index idx) const
+    {
+      enum {
+        StorageOrder = int(MatrixType::Flags) & RowMajorBit ? RowMajor : ColMajor
+      };
+      return packet<LoadMode>(int(StorageOrder)==ColMajor?idx:0,int(StorageOrder)==ColMajor?0:idx);
+    }
+
+  protected:
+    template<int LoadMode>
+    EIGEN_STRONG_INLINE PacketScalar packet_impl(Index row, Index col, Index id, internal::true_type) const
+    {
+      return internal::pmul(m_matrix.template packet<LoadMode>(row, col),
+                     internal::pset1<PacketScalar>(m_diagonal.diagonal().coeff(id)));
+    }
+
+    template<int LoadMode>
+    EIGEN_STRONG_INLINE PacketScalar packet_impl(Index row, Index col, Index id, internal::false_type) const
+    {
+      enum {
+        InnerSize = (MatrixType::Flags & RowMajorBit) ? MatrixType::ColsAtCompileTime : MatrixType::RowsAtCompileTime,
+        DiagonalVectorPacketLoadMode = (LoadMode == Aligned && (((InnerSize%16) == 0) || (int(DiagonalType::DiagonalVectorType::Flags)&AlignedBit)==AlignedBit) ? Aligned : Unaligned)
+      };
+      return internal::pmul(m_matrix.template packet<LoadMode>(row, col),
+                     m_diagonal.diagonal().template packet<DiagonalVectorPacketLoadMode>(id));
+    }
+
+    typename MatrixType::Nested m_matrix;
+    typename DiagonalType::Nested m_diagonal;
+};
+
 /** \returns the diagonal matrix product of \c *this by the diagonal matrix \a diagonal.
   */
 template<typename Derived>
 template<typename DiagonalDerived>
-EIGEN_DEVICE_FUNC inline const Product<Derived, DiagonalDerived, LazyProduct>
+inline const DiagonalProduct<Derived, DiagonalDerived, OnTheRight>
 MatrixBase<Derived>::operator*(const DiagonalBase<DiagonalDerived> &a_diagonal) const
 {
-  return Product<Derived, DiagonalDerived, LazyProduct>(derived(),a_diagonal.derived());
+  return DiagonalProduct<Derived, DiagonalDerived, OnTheRight>(derived(), a_diagonal.derived());
 }
 
 } // end namespace Eigen

Eigen/src/Core/Dot.h

@@ -28,33 +28,26 @@ template<typename T, typename U,
 >
 struct dot_nocheck
 {
-  typedef scalar_conj_product_op<typename traits<T>::Scalar,typename traits<U>::Scalar> conj_prod;
-  typedef typename conj_prod::result_type ResScalar;
-  EIGEN_DEVICE_FUNC
-  EIGEN_STRONG_INLINE
-  static ResScalar run(const MatrixBase<T>& a, const MatrixBase<U>& b)
+  typedef typename scalar_product_traits<typename traits<T>::Scalar,typename traits<U>::Scalar>::ReturnType ResScalar;
+  static inline ResScalar run(const MatrixBase<T>& a, const MatrixBase<U>& b)
   {
-    return a.template binaryExpr<conj_prod>(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 dot_nocheck<T, U, true>
 {
-  typedef scalar_conj_product_op<typename traits<T>::Scalar,typename traits<U>::Scalar> conj_prod;
-  typedef typename conj_prod::result_type ResScalar;
-  EIGEN_DEVICE_FUNC
-  EIGEN_STRONG_INLINE
-  static ResScalar run(const MatrixBase<T>& a, const MatrixBase<U>& b)
+  typedef typename scalar_product_traits<typename traits<T>::Scalar,typename traits<U>::Scalar>::ReturnType ResScalar;
+  static inline ResScalar run(const MatrixBase<T>& a, const MatrixBase<U>& b)
   {
-    return a.transpose().template binaryExpr<conj_prod>(b).sum();
+    return a.transpose().template binaryExpr<scalar_conj_product_op<typename traits<T>::Scalar,typename traits<U>::Scalar> >(b).sum();
   }
 };
 
 } // end namespace internal
 
-/** \fn MatrixBase::dot
-  * \returns the dot product of *this with other.
+/** \returns the dot product of *this with other.
   *
   * \only_for_vectors
   *
@@ -66,34 +59,58 @@ struct dot_nocheck<T, U, true>
   */
 template<typename Derived>
 template<typename OtherDerived>
-EIGEN_DEVICE_FUNC
-EIGEN_STRONG_INLINE
-typename ScalarBinaryOpTraits<typename internal::traits<Derived>::Scalar,typename internal::traits<OtherDerived>::Scalar>::ReturnType
+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)
-#if !(defined(EIGEN_NO_STATIC_ASSERT) && defined(EIGEN_NO_DEBUG))
   typedef internal::scalar_conj_product_op<Scalar,typename OtherDerived::Scalar> func;
   EIGEN_CHECK_BINARY_COMPATIBILIY(func,Scalar,typename OtherDerived::Scalar);
-#endif
-  
+
   eigen_assert(size() == other.size());
 
   return internal::dot_nocheck<Derived,OtherDerived>::run(*this, other);
 }
 
+#ifdef EIGEN2_SUPPORT
+/** \returns the dot product of *this with other, with the Eigen2 convention that the dot product is linear in the first variable
+  * (conjugating the second variable). Of course this only makes a difference in the complex case.
+  *
+  * This method is only available in EIGEN2_SUPPORT mode.
+  *
+  * \only_for_vectors
+  *
+  * \sa dot()
+  */
+template<typename Derived>
+template<typename OtherDerived>
+typename internal::traits<Derived>::Scalar
+MatrixBase<Derived>::eigen2_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((internal::is_same<Scalar, typename OtherDerived::Scalar>::value),
+    YOU_MIXED_DIFFERENT_NUMERIC_TYPES__YOU_NEED_TO_USE_THE_CAST_METHOD_OF_MATRIXBASE_TO_CAST_NUMERIC_TYPES_EXPLICITLY)
+
+  eigen_assert(size() == other.size());
+
+  return internal::dot_nocheck<OtherDerived,Derived>::run(other,*this);
+}
+#endif
+
+
 //---------- implementation of L2 norm and related functions ----------
 
 /** \returns, for vectors, the squared \em l2 norm of \c *this, and for matrices the Frobenius norm.
   * In both cases, it consists in the sum of the square of all the matrix entries.
   * For vectors, this is also equals to the dot product of \c *this with itself.
   *
-  * \sa dot(), norm(), lpNorm()
+  * \sa dot(), norm()
   */
 template<typename Derived>
-EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE typename NumTraits<typename internal::traits<Derived>::Scalar>::Real MatrixBase<Derived>::squaredNorm() const
+EIGEN_STRONG_INLINE typename NumTraits<typename internal::traits<Derived>::Scalar>::Real MatrixBase<Derived>::squaredNorm() const
 {
   return numext::real((*this).cwiseAbs2().sum());
 }
@@ -102,98 +119,41 @@ EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE typename NumTraits<typename internal::trai
   * 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 lpNorm(), dot(), squaredNorm()
+  * \sa dot(), squaredNorm()
   */
 template<typename Derived>
-EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE typename NumTraits<typename internal::traits<Derived>::Scalar>::Real MatrixBase<Derived>::norm() const
+inline typename NumTraits<typename internal::traits<Derived>::Scalar>::Real MatrixBase<Derived>::norm() const
 {
-  return numext::sqrt(squaredNorm());
+  using std::sqrt;
+  return sqrt(squaredNorm());
 }
 
-/** \returns an expression of the quotient of \c *this by its own norm.
-  *
-  * \warning If the input vector is too small (i.e., this->norm()==0),
-  *          then this function returns a copy of the input.
+/** \returns an expression of the quotient of *this by its own norm.
   *
   * \only_for_vectors
   *
   * \sa norm(), normalize()
   */
 template<typename Derived>
-EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE const typename MatrixBase<Derived>::PlainObject
+inline const typename MatrixBase<Derived>::PlainObject
 MatrixBase<Derived>::normalized() const
 {
-  typedef typename internal::nested_eval<Derived,2>::type _Nested;
+  typedef typename internal::nested<Derived>::type Nested;
+  typedef typename internal::remove_reference<Nested>::type _Nested;
   _Nested n(derived());
-  RealScalar z = n.squaredNorm();
-  // NOTE: after extensive benchmarking, this conditional does not impact performance, at least on recent x86 CPU
-  if(z>RealScalar(0))
-    return n / numext::sqrt(z);
-  else
-    return n;
+  return n / n.norm();
 }
 
 /** Normalizes the vector, i.e. divides it by its own norm.
   *
   * \only_for_vectors
   *
-  * \warning If the input vector is too small (i.e., this->norm()==0), then \c *this is left unchanged.
-  *
   * \sa norm(), normalized()
   */
 template<typename Derived>
-EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE void MatrixBase<Derived>::normalize()
+inline void MatrixBase<Derived>::normalize()
 {
-  RealScalar z = squaredNorm();
-  // NOTE: after extensive benchmarking, this conditional does not impact performance, at least on recent x86 CPU
-  if(z>RealScalar(0))
-    derived() /= numext::sqrt(z);
-}
-
-/** \returns an expression of the quotient of \c *this by its own norm while avoiding underflow and overflow.
-  *
-  * \only_for_vectors
-  *
-  * This method is analogue to the normalized() method, but it reduces the risk of
-  * underflow and overflow when computing the norm.
-  *
-  * \warning If the input vector is too small (i.e., this->norm()==0),
-  *          then this function returns a copy of the input.
-  *
-  * \sa stableNorm(), stableNormalize(), normalized()
-  */
-template<typename Derived>
-EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE const typename MatrixBase<Derived>::PlainObject
-MatrixBase<Derived>::stableNormalized() const
-{
-  typedef typename internal::nested_eval<Derived,3>::type _Nested;
-  _Nested n(derived());
-  RealScalar w = n.cwiseAbs().maxCoeff();
-  RealScalar z = (n/w).squaredNorm();
-  if(z>RealScalar(0))
-    return n / (numext::sqrt(z)*w);
-  else
-    return n;
-}
-
-/** Normalizes the vector while avoid underflow and overflow
-  *
-  * \only_for_vectors
-  *
-  * This method is analogue to the normalize() method, but it reduces the risk of
-  * underflow and overflow when computing the norm.
-  *
-  * \warning If the input vector is too small (i.e., this->norm()==0), then \c *this is left unchanged.
-  *
-  * \sa stableNorm(), stableNormalized(), normalize()
-  */
-template<typename Derived>
-EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE void MatrixBase<Derived>::stableNormalize()
-{
-  RealScalar w = cwiseAbs().maxCoeff();
-  RealScalar z = (derived()/w).squaredNorm();
-  if(z>RealScalar(0))
-    derived() /= numext::sqrt(z)*w;
+  *this /= norm();
 }
 
 //---------- implementation of other norms ----------
@@ -204,10 +164,9 @@ template<typename Derived, int p>
 struct lpNorm_selector
 {
   typedef typename NumTraits<typename traits<Derived>::Scalar>::Real RealScalar;
-  EIGEN_DEVICE_FUNC
   static inline RealScalar run(const MatrixBase<Derived>& m)
   {
-    EIGEN_USING_STD(pow)
+    using std::pow;
     return pow(m.cwiseAbs().array().pow(p).sum(), RealScalar(1)/p);
   }
 };
@@ -215,7 +174,6 @@ struct lpNorm_selector
 template<typename Derived>
 struct lpNorm_selector<Derived, 1>
 {
-  EIGEN_DEVICE_FUNC
   static inline typename NumTraits<typename traits<Derived>::Scalar>::Real run(const MatrixBase<Derived>& m)
   {
     return m.cwiseAbs().sum();
@@ -225,7 +183,6 @@ struct lpNorm_selector<Derived, 1>
 template<typename Derived>
 struct lpNorm_selector<Derived, 2>
 {
-  EIGEN_DEVICE_FUNC
   static inline typename NumTraits<typename traits<Derived>::Scalar>::Real run(const MatrixBase<Derived>& m)
   {
     return m.norm();
@@ -235,35 +192,23 @@ struct lpNorm_selector<Derived, 2>
 template<typename Derived>
 struct lpNorm_selector<Derived, Infinity>
 {
-  typedef typename NumTraits<typename traits<Derived>::Scalar>::Real RealScalar;
-  EIGEN_DEVICE_FUNC
-  static inline RealScalar run(const MatrixBase<Derived>& m)
+  static inline typename NumTraits<typename traits<Derived>::Scalar>::Real run(const MatrixBase<Derived>& m)
   {
-    if(Derived::SizeAtCompileTime==0 || (Derived::SizeAtCompileTime==Dynamic && m.size()==0))
-      return RealScalar(0);
     return m.cwiseAbs().maxCoeff();
   }
 };
 
 } // end namespace internal
 
-/** \returns the \b coefficient-wise \f$ \ell^p \f$ norm of \c *this, that is, returns the p-th root of the sum of the p-th powers of the absolute values
-  *          of the coefficients of \c *this. If \a p is the special value \a Eigen::Infinity, this function returns the \f$ \ell^\infty \f$
-  *          norm, that is the maximum of the absolute values of the coefficients of \c *this.
-  *
-  * In all cases, if \c *this is empty, then the value 0 is returned.
-  *
-  * \note For matrices, this function does not compute the <a href="https://en.wikipedia.org/wiki/Operator_norm">operator-norm</a>. That is, if \c *this is a matrix, then its coefficients are interpreted as a 1D vector. Nonetheless, you can easily compute the 1-norm and \f$\infty\f$-norm matrix operator norms using \link TutorialReductionsVisitorsBroadcastingReductionsNorm partial reductions \endlink.
+/** \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^\infty \f$
+  *          norm, that is the maximum of the absolute values of the coefficients of *this.
   *
   * \sa norm()
   */
 template<typename Derived>
 template<int p>
-#ifndef EIGEN_PARSED_BY_DOXYGEN
-EIGEN_DEVICE_FUNC inline typename NumTraits<typename internal::traits<Derived>::Scalar>::Real
-#else
-EIGEN_DEVICE_FUNC MatrixBase<Derived>::RealScalar
-#endif
+inline typename NumTraits<typename internal::traits<Derived>::Scalar>::Real
 MatrixBase<Derived>::lpNorm() const
 {
   return internal::lpNorm_selector<Derived, p>::run(*this);
@@ -282,8 +227,8 @@ template<typename OtherDerived>
 bool MatrixBase<Derived>::isOrthogonal
 (const MatrixBase<OtherDerived>& other, const RealScalar& prec) const
 {
-  typename internal::nested_eval<Derived,2>::type nested(derived());
-  typename internal::nested_eval<OtherDerived,2>::type otherNested(other.derived());
+  typename internal::nested<Derived,2>::type nested(derived());
+  typename internal::nested<OtherDerived,2>::type otherNested(other.derived());
   return numext::abs2(nested.dot(otherNested)) <= prec * prec * nested.squaredNorm() * otherNested.squaredNorm();
 }
 
@@ -301,13 +246,13 @@ bool MatrixBase<Derived>::isOrthogonal
 template<typename Derived>
 bool MatrixBase<Derived>::isUnitary(const RealScalar& prec) const
 {
-  typename internal::nested_eval<Derived,1>::type self(derived());
+  typename Derived::Nested nested(derived());
   for(Index i = 0; i < cols(); ++i)
   {
-    if(!internal::isApprox(self.col(i).squaredNorm(), static_cast<RealScalar>(1), prec))
+    if(!internal::isApprox(nested.col(i).squaredNorm(), static_cast<RealScalar>(1), prec))
       return false;
     for(Index j = 0; j < i; ++j)
-      if(!internal::isMuchSmallerThan(self.col(i).dot(self.col(j)), static_cast<Scalar>(1), prec))
+      if(!internal::isMuchSmallerThan(nested.col(i).dot(nested.col(j)), static_cast<Scalar>(1), prec))
         return false;
   }
   return true;

Eigen/src/Core/EigenBase.h

@@ -13,10 +13,7 @@
 
 namespace Eigen {
 
-/** \class EigenBase
-  * \ingroup Core_Module
-  *
-  * Common base class for all classes T such that MatrixBase has an operator=(T) and a constructor MatrixBase(T).
+/** 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.
   *
@@ -24,58 +21,39 @@ namespace Eigen {
   *
   * Notice that this class is trivial, it is only used to disambiguate overloaded functions.
   *
-  * \sa \blank \ref TopicClassHierarchy
+  * \sa \ref TopicClassHierarchy
   */
 template<typename Derived> struct EigenBase
 {
 //   typedef typename internal::plain_matrix_type<Derived>::type PlainObject;
 
-  /** \brief The interface type of indices
-    * \details To change this, \c \#define the preprocessor symbol \c EIGEN_DEFAULT_DENSE_INDEX_TYPE.
-    * \sa StorageIndex, \ref TopicPreprocessorDirectives.
-    * DEPRECATED: Since Eigen 3.3, its usage is deprecated. Use Eigen::Index instead.
-    * Deprecation is not marked with a doxygen comment because there are too many existing usages to add the deprecation attribute.
-    */
-  typedef Eigen::Index Index;
-
-  // FIXME is it needed?
   typedef typename internal::traits<Derived>::StorageKind StorageKind;
+  typedef typename internal::traits<Derived>::Index Index;
 
   /** \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 EIGEN_CONSTEXPR
-  inline Index rows() const EIGEN_NOEXCEPT { return derived().rows(); }
+  inline Index rows() const { return derived().rows(); }
   /** \returns the number of columns. \sa rows(), ColsAtCompileTime*/
-  EIGEN_DEVICE_FUNC EIGEN_CONSTEXPR
-  inline Index cols() const EIGEN_NOEXCEPT { return derived().cols(); }
+  inline Index cols() const { return derived().cols(); }
   /** \returns the number of coefficients, which is rows()*cols().
     * \sa rows(), cols(), SizeAtCompileTime. */
-  EIGEN_DEVICE_FUNC EIGEN_CONSTEXPR
-  inline Index size() const EIGEN_NOEXCEPT { return rows() * cols(); }
+  inline Index size() const { return rows() * cols(); }
 
   /** \internal Don't use it, but do the equivalent: \code dst = *this; \endcode */
-  template<typename Dest>
-  EIGEN_DEVICE_FUNC
-  inline void evalTo(Dest& dst) const
+  template<typename Dest> inline void evalTo(Dest& dst) const
   { derived().evalTo(dst); }
 
   /** \internal Don't use it, but do the equivalent: \code dst += *this; \endcode */
-  template<typename Dest>
-  EIGEN_DEVICE_FUNC
-  inline void addTo(Dest& dst) const
+  template<typename Dest> inline void addTo(Dest& dst) const
   {
     // This is the default implementation,
     // derived class can reimplement it in a more optimized way.
@@ -85,9 +63,7 @@ template<typename Derived> struct EigenBase
   }
 
   /** \internal Don't use it, but do the equivalent: \code dst -= *this; \endcode */
-  template<typename Dest>
-  EIGEN_DEVICE_FUNC
-  inline void subTo(Dest& dst) const
+  template<typename Dest> inline void subTo(Dest& dst) const
   {
     // This is the default implementation,
     // derived class can reimplement it in a more optimized way.
@@ -97,8 +73,7 @@ template<typename Derived> struct EigenBase
   }
 
   /** \internal Don't use it, but do the equivalent: \code dst.applyOnTheRight(*this); \endcode */
-  template<typename Dest>
-  EIGEN_DEVICE_FUNC inline void applyThisOnTheRight(Dest& dst) const
+  template<typename Dest> inline void applyThisOnTheRight(Dest& dst) const
   {
     // This is the default implementation,
     // derived class can reimplement it in a more optimized way.
@@ -106,8 +81,7 @@ template<typename Derived> struct EigenBase
   }
 
   /** \internal Don't use it, but do the equivalent: \code dst.applyOnTheLeft(*this); \endcode */
-  template<typename Dest>
-  EIGEN_DEVICE_FUNC inline void applyThisOnTheLeft(Dest& dst) const
+  template<typename Dest> inline void applyThisOnTheLeft(Dest& dst) const
   {
     // This is the default implementation,
     // derived class can reimplement it in a more optimized way.
@@ -130,31 +104,58 @@ template<typename Derived> struct EigenBase
   */
 template<typename Derived>
 template<typename OtherDerived>
-EIGEN_DEVICE_FUNC
 Derived& DenseBase<Derived>::operator=(const EigenBase<OtherDerived> &other)
 {
-  call_assignment(derived(), other.derived());
+  other.derived().evalTo(derived());
   return derived();
 }
 
 template<typename Derived>
 template<typename OtherDerived>
-EIGEN_DEVICE_FUNC
 Derived& DenseBase<Derived>::operator+=(const EigenBase<OtherDerived> &other)
 {
-  call_assignment(derived(), other.derived(), internal::add_assign_op<Scalar,typename OtherDerived::Scalar>());
+  other.derived().addTo(derived());
   return derived();
 }
 
 template<typename Derived>
 template<typename OtherDerived>
-EIGEN_DEVICE_FUNC
 Derived& DenseBase<Derived>::operator-=(const EigenBase<OtherDerived> &other)
 {
-  call_assignment(derived(), other.derived(), internal::sub_assign_op<Scalar,typename OtherDerived::Scalar>());
+  other.derived().subTo(derived());
   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

@@ -0,0 +1,140 @@
+// This file is part of Eigen, a lightweight C++ template library
+// for linear algebra.
+//
+// Copyright (C) 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_FLAGGED_H
+#define EIGEN_FLAGGED_H
+
+namespace Eigen { 
+
+/** \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()
+  */
+
+namespace internal {
+template<typename ExpressionType, unsigned int Added, unsigned int Removed>
+struct traits<Flagged<ExpressionType, Added, Removed> > : 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 internal::conditional<internal::must_nest_by_value<ExpressionType>::ret,
+        ExpressionType, const ExpressionType&>::type 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 CoeffReturnType coeff(Index row, Index col) const
+    {
+      return m_matrix.coeff(row, col);
+    }
+
+    inline CoeffReturnType coeff(Index index) const
+    {
+      return m_matrix.coeff(index);
+    }
+    
+    inline const Scalar& coeffRef(Index row, Index col) const
+    {
+      return m_matrix.const_cast_derived().coeffRef(row, col);
+    }
+
+    inline const Scalar& coeffRef(Index index) const
+    {
+      return m_matrix.const_cast_derived().coeffRef(index);
+    }
+
+    inline Scalar& coeffRef(Index row, Index col)
+    {
+      return m_matrix.const_cast_derived().coeffRef(row, col);
+    }
+
+    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();
+}
+
+} // end namespace Eigen
+
+#endif // EIGEN_FLAGGED_H

Eigen/src/Core/ForceAlignedAccess.h

@@ -39,33 +39,29 @@ template<typename ExpressionType> class ForceAlignedAccess
     typedef typename internal::dense_xpr_base<ForceAlignedAccess>::type Base;
     EIGEN_DENSE_PUBLIC_INTERFACE(ForceAlignedAccess)
 
-    EIGEN_DEVICE_FUNC explicit inline ForceAlignedAccess(const ExpressionType& matrix) : m_expression(matrix) {}
+    inline ForceAlignedAccess(const ExpressionType& matrix) : m_expression(matrix) {}
 
-    EIGEN_DEVICE_FUNC EIGEN_CONSTEXPR
-    inline Index rows() const EIGEN_NOEXCEPT { return m_expression.rows(); }
-    EIGEN_DEVICE_FUNC EIGEN_CONSTEXPR
-    inline Index cols() const EIGEN_NOEXCEPT { return m_expression.cols(); }
-    EIGEN_DEVICE_FUNC EIGEN_CONSTEXPR
-    inline Index outerStride() const EIGEN_NOEXCEPT { return m_expression.outerStride(); }
-    EIGEN_DEVICE_FUNC EIGEN_CONSTEXPR
-    inline Index innerStride() const EIGEN_NOEXCEPT { return m_expression.innerStride(); }
+    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 const CoeffReturnType coeff(Index row, Index col) const
+    inline const CoeffReturnType coeff(Index row, Index col) const
     {
       return m_expression.coeff(row, col);
     }
 
-    EIGEN_DEVICE_FUNC inline Scalar& coeffRef(Index row, Index col)
+    inline Scalar& coeffRef(Index row, Index col)
     {
       return m_expression.const_cast_derived().coeffRef(row, col);
     }
 
-    EIGEN_DEVICE_FUNC inline const CoeffReturnType coeff(Index index) const
+    inline const CoeffReturnType coeff(Index index) const
     {
       return m_expression.coeff(index);
     }
 
-    EIGEN_DEVICE_FUNC inline Scalar& coeffRef(Index index)
+    inline Scalar& coeffRef(Index index)
     {
       return m_expression.const_cast_derived().coeffRef(index);
     }
@@ -94,7 +90,7 @@ template<typename ExpressionType> class ForceAlignedAccess
       m_expression.const_cast_derived().template writePacket<Aligned>(index, x);
     }
 
-    EIGEN_DEVICE_FUNC operator const ExpressionType&() const { return m_expression; }
+    operator const ExpressionType&() const { return m_expression; }
 
   protected:
     const ExpressionType& m_expression;
@@ -131,7 +127,7 @@ template<bool Enable>
 inline typename internal::add_const_on_value_type<typename internal::conditional<Enable,ForceAlignedAccess<Derived>,Derived&>::type>::type
 MatrixBase<Derived>::forceAlignedAccessIf() const
 {
-  return derived();  // FIXME This should not work but apparently is never used
+  return derived();
 }
 
 /** \returns an expression of *this with forced aligned access if \a Enable is true.
@@ -142,7 +138,7 @@ template<bool Enable>
 inline typename internal::conditional<Enable,ForceAlignedAccess<Derived>,Derived&>::type
 MatrixBase<Derived>::forceAlignedAccessIf()
 {
-  return derived();  // FIXME This should not work but apparently is never used
+  return derived();
 }
 
 } // end namespace Eigen

Eigen/src/Core/Functors.h

@@ -0,0 +1,985 @@
+// 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_FUNCTORS_H
+#define EIGEN_FUNCTORS_H
+
+namespace Eigen {
+
+namespace internal {
+
+// 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 scalar_sum_op {
+  EIGEN_EMPTY_STRUCT_CTOR(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 internal::padd(a,b); }
+  template<typename Packet>
+  EIGEN_STRONG_INLINE const Scalar predux(const Packet& a) const
+  { return internal::predux(a); }
+};
+template<typename Scalar>
+struct functor_traits<scalar_sum_op<Scalar> > {
+  enum {
+    Cost = NumTraits<Scalar>::AddCost,
+    PacketAccess = 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 scalar_product_op {
+  enum {
+    // TODO vectorize mixed product
+    Vectorizable = is_same<LhsScalar,RhsScalar>::value && packet_traits<LhsScalar>::HasMul && packet_traits<RhsScalar>::HasMul
+  };
+  typedef typename scalar_product_traits<LhsScalar,RhsScalar>::ReturnType result_type;
+  EIGEN_EMPTY_STRUCT_CTOR(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 internal::pmul(a,b); }
+  template<typename Packet>
+  EIGEN_STRONG_INLINE const result_type predux(const Packet& a) const
+  { return internal::predux_mul(a); }
+};
+template<typename LhsScalar,typename RhsScalar>
+struct functor_traits<scalar_product_op<LhsScalar,RhsScalar> > {
+  enum {
+    Cost = (NumTraits<LhsScalar>::MulCost + NumTraits<RhsScalar>::MulCost)/2, // rough estimate!
+    PacketAccess = scalar_product_op<LhsScalar,RhsScalar>::Vectorizable
+  };
+};
+
+/** \internal
+  * \brief Template functor to compute the conjugate product of two scalars
+  *
+  * This is a short cut for conj(x) * y which is needed for optimization purpose; in Eigen2 support mode, this becomes x * conj(y)
+  */
+template<typename LhsScalar,typename RhsScalar> struct scalar_conj_product_op {
+
+  enum {
+    Conj = NumTraits<LhsScalar>::IsComplex
+  };
+  
+  typedef typename scalar_product_traits<LhsScalar,RhsScalar>::ReturnType result_type;
+  
+  EIGEN_EMPTY_STRUCT_CTOR(scalar_conj_product_op)
+  EIGEN_STRONG_INLINE const result_type operator() (const LhsScalar& a, const RhsScalar& b) const
+  { return conj_helper<LhsScalar,RhsScalar,Conj,false>().pmul(a,b); }
+  
+  template<typename Packet>
+  EIGEN_STRONG_INLINE const Packet packetOp(const Packet& a, const Packet& b) const
+  { return conj_helper<Packet,Packet,Conj,false>().pmul(a,b); }
+};
+template<typename LhsScalar,typename RhsScalar>
+struct functor_traits<scalar_conj_product_op<LhsScalar,RhsScalar> > {
+  enum {
+    Cost = NumTraits<LhsScalar>::MulCost,
+    PacketAccess = internal::is_same<LhsScalar, RhsScalar>::value && packet_traits<LhsScalar>::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 scalar_min_op {
+  EIGEN_EMPTY_STRUCT_CTOR(scalar_min_op)
+  EIGEN_STRONG_INLINE const Scalar operator() (const Scalar& a, const Scalar& b) const { using std::min; return (min)(a, b); }
+  template<typename Packet>
+  EIGEN_STRONG_INLINE const Packet packetOp(const Packet& a, const Packet& b) const
+  { return internal::pmin(a,b); }
+  template<typename Packet>
+  EIGEN_STRONG_INLINE const Scalar predux(const Packet& a) const
+  { return internal::predux_min(a); }
+};
+template<typename Scalar>
+struct functor_traits<scalar_min_op<Scalar> > {
+  enum {
+    Cost = NumTraits<Scalar>::AddCost,
+    PacketAccess = 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 scalar_max_op {
+  EIGEN_EMPTY_STRUCT_CTOR(scalar_max_op)
+  EIGEN_STRONG_INLINE const Scalar operator() (const Scalar& a, const Scalar& b) const { using std::max; return (max)(a, b); }
+  template<typename Packet>
+  EIGEN_STRONG_INLINE const Packet packetOp(const Packet& a, const Packet& b) const
+  { return internal::pmax(a,b); }
+  template<typename Packet>
+  EIGEN_STRONG_INLINE const Scalar predux(const Packet& a) const
+  { return internal::predux_max(a); }
+};
+template<typename Scalar>
+struct functor_traits<scalar_max_op<Scalar> > {
+  enum {
+    Cost = NumTraits<Scalar>::AddCost,
+    PacketAccess = packet_traits<Scalar>::HasMax
+  };
+};
+
+/** \internal
+  * \brief Template functor to compute the hypot of two scalars
+  *
+  * \sa MatrixBase::stableNorm(), class Redux
+  */
+template<typename Scalar> struct scalar_hypot_op {
+  EIGEN_EMPTY_STRUCT_CTOR(scalar_hypot_op)
+//   typedef typename NumTraits<Scalar>::Real result_type;
+  EIGEN_STRONG_INLINE const Scalar operator() (const Scalar& _x, const Scalar& _y) const
+  {
+    using std::max;
+    using std::min;
+    using std::sqrt;
+    Scalar p = (max)(_x, _y);
+    Scalar q = (min)(_x, _y);
+    Scalar qp = q/p;
+    return p * sqrt(Scalar(1) + qp*qp);
+  }
+};
+template<typename Scalar>
+struct functor_traits<scalar_hypot_op<Scalar> > {
+  enum { Cost = 5 * NumTraits<Scalar>::MulCost, PacketAccess=0 };
+};
+
+/** \internal
+  * \brief Template functor to compute the pow of two scalars
+  */
+template<typename Scalar, typename OtherScalar> struct scalar_binary_pow_op {
+  EIGEN_EMPTY_STRUCT_CTOR(scalar_binary_pow_op)
+  inline Scalar operator() (const Scalar& a, const OtherScalar& b) const { return numext::pow(a, b); }
+};
+template<typename Scalar, typename OtherScalar>
+struct functor_traits<scalar_binary_pow_op<Scalar,OtherScalar> > {
+  enum { Cost = 5 * NumTraits<Scalar>::MulCost, PacketAccess = false };
+};
+
+// other binary functors:
+
+/** \internal
+  * \brief Template functor to compute the difference of two scalars
+  *
+  * \sa class CwiseBinaryOp, MatrixBase::operator-
+  */
+template<typename Scalar> struct scalar_difference_op {
+  EIGEN_EMPTY_STRUCT_CTOR(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 internal::psub(a,b); }
+};
+template<typename Scalar>
+struct functor_traits<scalar_difference_op<Scalar> > {
+  enum {
+    Cost = NumTraits<Scalar>::AddCost,
+    PacketAccess = packet_traits<Scalar>::HasSub
+  };
+};
+
+/** \internal
+  * \brief Template functor to compute the quotient of two scalars
+  *
+  * \sa class CwiseBinaryOp, Cwise::operator/()
+  */
+template<typename LhsScalar,typename RhsScalar> struct scalar_quotient_op {
+  enum {
+    // TODO vectorize mixed product
+    Vectorizable = is_same<LhsScalar,RhsScalar>::value && packet_traits<LhsScalar>::HasDiv && packet_traits<RhsScalar>::HasDiv
+  };
+  typedef typename scalar_product_traits<LhsScalar,RhsScalar>::ReturnType result_type;
+  EIGEN_EMPTY_STRUCT_CTOR(scalar_quotient_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 internal::pdiv(a,b); }
+};
+template<typename LhsScalar,typename RhsScalar>
+struct functor_traits<scalar_quotient_op<LhsScalar,RhsScalar> > {
+  enum {
+    Cost = (NumTraits<LhsScalar>::MulCost + NumTraits<RhsScalar>::MulCost), // rough estimate!
+    PacketAccess = scalar_quotient_op<LhsScalar,RhsScalar>::Vectorizable
+  };
+};
+
+
+
+/** \internal
+  * \brief Template functor to compute the and of two booleans
+  *
+  * \sa class CwiseBinaryOp, ArrayBase::operator&&
+  */
+struct scalar_boolean_and_op {
+  EIGEN_EMPTY_STRUCT_CTOR(scalar_boolean_and_op)
+  EIGEN_STRONG_INLINE bool operator() (const bool& a, const bool& b) const { return a && b; }
+};
+template<> struct functor_traits<scalar_boolean_and_op> {
+  enum {
+    Cost = NumTraits<bool>::AddCost,
+    PacketAccess = false
+  };
+};
+
+/** \internal
+  * \brief Template functor to compute the or of two booleans
+  *
+  * \sa class CwiseBinaryOp, ArrayBase::operator||
+  */
+struct scalar_boolean_or_op {
+  EIGEN_EMPTY_STRUCT_CTOR(scalar_boolean_or_op)
+  EIGEN_STRONG_INLINE bool operator() (const bool& a, const bool& b) const { return a || b; }
+};
+template<> struct functor_traits<scalar_boolean_or_op> {
+  enum {
+    Cost = NumTraits<bool>::AddCost,
+    PacketAccess = false
+  };
+};
+
+// unary functors:
+
+/** \internal
+  * \brief Template functor to compute the opposite of a scalar
+  *
+  * \sa class CwiseUnaryOp, MatrixBase::operator-
+  */
+template<typename Scalar> struct scalar_opposite_op {
+  EIGEN_EMPTY_STRUCT_CTOR(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 internal::pnegate(a); }
+};
+template<typename Scalar>
+struct functor_traits<scalar_opposite_op<Scalar> >
+{ enum {
+    Cost = NumTraits<Scalar>::AddCost,
+    PacketAccess = 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 scalar_abs_op {
+  EIGEN_EMPTY_STRUCT_CTOR(scalar_abs_op)
+  typedef typename NumTraits<Scalar>::Real result_type;
+  EIGEN_STRONG_INLINE const result_type operator() (const Scalar& a) const { using std::abs; return abs(a); }
+  template<typename Packet>
+  EIGEN_STRONG_INLINE const Packet packetOp(const Packet& a) const
+  { return internal::pabs(a); }
+};
+template<typename Scalar>
+struct functor_traits<scalar_abs_op<Scalar> >
+{
+  enum {
+    Cost = NumTraits<Scalar>::AddCost,
+    PacketAccess = 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 scalar_abs2_op {
+  EIGEN_EMPTY_STRUCT_CTOR(scalar_abs2_op)
+  typedef typename NumTraits<Scalar>::Real result_type;
+  EIGEN_STRONG_INLINE const result_type operator() (const Scalar& a) const { return numext::abs2(a); }
+  template<typename Packet>
+  EIGEN_STRONG_INLINE const Packet packetOp(const Packet& a) const
+  { return internal::pmul(a,a); }
+};
+template<typename Scalar>
+struct functor_traits<scalar_abs2_op<Scalar> >
+{ enum { Cost = NumTraits<Scalar>::MulCost, PacketAccess = 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 scalar_conjugate_op {
+  EIGEN_EMPTY_STRUCT_CTOR(scalar_conjugate_op)
+  EIGEN_STRONG_INLINE const Scalar operator() (const Scalar& a) const { using numext::conj; return conj(a); }
+  template<typename Packet>
+  EIGEN_STRONG_INLINE const Packet packetOp(const Packet& a) const { return internal::pconj(a); }
+};
+template<typename Scalar>
+struct functor_traits<scalar_conjugate_op<Scalar> >
+{
+  enum {
+    Cost = NumTraits<Scalar>::IsComplex ? NumTraits<Scalar>::AddCost : 0,
+    PacketAccess = 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 scalar_cast_op {
+  EIGEN_EMPTY_STRUCT_CTOR(scalar_cast_op)
+  typedef NewType result_type;
+  EIGEN_STRONG_INLINE const NewType operator() (const Scalar& a) const { return cast<Scalar, NewType>(a); }
+};
+template<typename Scalar, typename NewType>
+struct functor_traits<scalar_cast_op<Scalar,NewType> >
+{ enum { Cost = is_same<Scalar, NewType>::value ? 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 scalar_real_op {
+  EIGEN_EMPTY_STRUCT_CTOR(scalar_real_op)
+  typedef typename NumTraits<Scalar>::Real result_type;
+  EIGEN_STRONG_INLINE result_type operator() (const Scalar& a) const { return numext::real(a); }
+};
+template<typename Scalar>
+struct functor_traits<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 scalar_imag_op {
+  EIGEN_EMPTY_STRUCT_CTOR(scalar_imag_op)
+  typedef typename NumTraits<Scalar>::Real result_type;
+  EIGEN_STRONG_INLINE result_type operator() (const Scalar& a) const { return numext::imag(a); }
+};
+template<typename Scalar>
+struct functor_traits<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 scalar_real_ref_op {
+  EIGEN_EMPTY_STRUCT_CTOR(scalar_real_ref_op)
+  typedef typename NumTraits<Scalar>::Real result_type;
+  EIGEN_STRONG_INLINE result_type& operator() (const Scalar& a) const { return numext::real_ref(*const_cast<Scalar*>(&a)); }
+};
+template<typename Scalar>
+struct functor_traits<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 scalar_imag_ref_op {
+  EIGEN_EMPTY_STRUCT_CTOR(scalar_imag_ref_op)
+  typedef typename NumTraits<Scalar>::Real result_type;
+  EIGEN_STRONG_INLINE result_type& operator() (const Scalar& a) const { return numext::imag_ref(*const_cast<Scalar*>(&a)); }
+};
+template<typename Scalar>
+struct functor_traits<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 scalar_exp_op {
+  EIGEN_EMPTY_STRUCT_CTOR(scalar_exp_op)
+  inline const Scalar operator() (const Scalar& a) const { using std::exp; return exp(a); }
+  typedef typename packet_traits<Scalar>::type Packet;
+  inline Packet packetOp(const Packet& a) const { return internal::pexp(a); }
+};
+template<typename Scalar>
+struct functor_traits<scalar_exp_op<Scalar> >
+{ enum { Cost = 5 * NumTraits<Scalar>::MulCost, PacketAccess = packet_traits<Scalar>::HasExp }; };
+
+/** \internal
+  *
+  * \brief Template functor to compute the logarithm of a scalar
+  *
+  * \sa class CwiseUnaryOp, Cwise::log()
+  */
+template<typename Scalar> struct scalar_log_op {
+  EIGEN_EMPTY_STRUCT_CTOR(scalar_log_op)
+  inline const Scalar operator() (const Scalar& a) const { using std::log; return log(a); }
+  typedef typename packet_traits<Scalar>::type Packet;
+  inline Packet packetOp(const Packet& a) const { return internal::plog(a); }
+};
+template<typename Scalar>
+struct functor_traits<scalar_log_op<Scalar> >
+{ enum { Cost = 5 * NumTraits<Scalar>::MulCost, PacketAccess = 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 pset1() in packetOp *is* an optimization ?
+ * indeed it seems better to declare m_other as a Packet and do the 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 pset1() outside 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 scalar_multiple_op {
+  typedef typename packet_traits<Scalar>::type Packet;
+  // FIXME default copy constructors seems bugged with std::complex<>
+  EIGEN_STRONG_INLINE scalar_multiple_op(const scalar_multiple_op& other) : m_other(other.m_other) { }
+  EIGEN_STRONG_INLINE 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 internal::pmul(a, pset1<Packet>(m_other)); }
+  typename add_const_on_value_type<typename NumTraits<Scalar>::Nested>::type m_other;
+};
+template<typename Scalar>
+struct functor_traits<scalar_multiple_op<Scalar> >
+{ enum { Cost = NumTraits<Scalar>::MulCost, PacketAccess = packet_traits<Scalar>::HasMul }; };
+
+template<typename Scalar1, typename Scalar2>
+struct scalar_multiple2_op {
+  typedef typename scalar_product_traits<Scalar1,Scalar2>::ReturnType result_type;
+  EIGEN_STRONG_INLINE scalar_multiple2_op(const scalar_multiple2_op& other) : m_other(other.m_other) { }
+  EIGEN_STRONG_INLINE scalar_multiple2_op(const Scalar2& other) : m_other(other) { }
+  EIGEN_STRONG_INLINE result_type operator() (const Scalar1& a) const { return a * m_other; }
+  typename add_const_on_value_type<typename NumTraits<Scalar2>::Nested>::type m_other;
+};
+template<typename Scalar1,typename Scalar2>
+struct functor_traits<scalar_multiple2_op<Scalar1,Scalar2> >
+{ enum { Cost = NumTraits<Scalar1>::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 scalar_quotient1_op {
+  typedef typename packet_traits<Scalar>::type Packet;
+  // FIXME default copy constructors seems bugged with std::complex<>
+  EIGEN_STRONG_INLINE scalar_quotient1_op(const scalar_quotient1_op& other) : m_other(other.m_other) { }
+  EIGEN_STRONG_INLINE scalar_quotient1_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 internal::pdiv(a, pset1<Packet>(m_other)); }
+  typename add_const_on_value_type<typename NumTraits<Scalar>::Nested>::type m_other;
+};
+template<typename Scalar>
+struct functor_traits<scalar_quotient1_op<Scalar> >
+{ enum { Cost = 2 * NumTraits<Scalar>::MulCost, PacketAccess = packet_traits<Scalar>::HasDiv }; };
+
+// nullary functors
+
+template<typename Scalar>
+struct scalar_constant_op {
+  typedef typename packet_traits<Scalar>::type Packet;
+  EIGEN_STRONG_INLINE scalar_constant_op(const scalar_constant_op& other) : m_other(other.m_other) { }
+  EIGEN_STRONG_INLINE 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 internal::pset1<Packet>(m_other); }
+  const Scalar m_other;
+};
+template<typename Scalar>
+struct functor_traits<scalar_constant_op<Scalar> >
+// FIXME replace this packet test by a safe one
+{ enum { Cost = 1, PacketAccess = packet_traits<Scalar>::Vectorizable, IsRepeatable = true }; };
+
+template<typename Scalar> struct scalar_identity_op {
+  EIGEN_EMPTY_STRUCT_CTOR(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 functor_traits<scalar_identity_op<Scalar> >
+{ enum { Cost = NumTraits<Scalar>::AddCost, PacketAccess = false, IsRepeatable = true }; };
+
+template <typename Scalar, bool RandomAccess> struct linspaced_op_impl;
+
+// linear access for packet ops:
+// 1) initialization
+//   base = [low, ..., low] + ([step, ..., step] * [-size, ..., 0])
+// 2) each step (where size is 1 for coeff access or PacketSize for packet access)
+//   base += [size*step, ..., size*step]
+//
+// TODO: Perhaps it's better to initialize lazily (so not in the constructor but in packetOp)
+//       in order to avoid the padd() in operator() ?
+template <typename Scalar>
+struct linspaced_op_impl<Scalar,false>
+{
+  typedef typename packet_traits<Scalar>::type Packet;
+
+  linspaced_op_impl(const Scalar& low, const Scalar& step) :
+  m_low(low), m_step(step),
+  m_packetStep(pset1<Packet>(packet_traits<Scalar>::size*step)),
+  m_base(padd(pset1<Packet>(low), pmul(pset1<Packet>(step),plset<Scalar>(-packet_traits<Scalar>::size)))) {}
+
+  template<typename Index>
+  EIGEN_STRONG_INLINE const Scalar operator() (Index i) const 
+  { 
+    m_base = padd(m_base, pset1<Packet>(m_step));
+    return m_low+Scalar(i)*m_step; 
+  }
+
+  template<typename Index>
+  EIGEN_STRONG_INLINE const Packet packetOp(Index) const { return m_base = 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 linspaced_op_impl<Scalar,true>
+{
+  typedef typename packet_traits<Scalar>::type Packet;
+
+  linspaced_op_impl(const Scalar& low, const Scalar& step) :
+  m_low(low), m_step(step),
+  m_lowPacket(pset1<Packet>(m_low)), m_stepPacket(pset1<Packet>(m_step)), m_interPacket(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 internal::padd(m_lowPacket, pmul(m_stepPacket, padd(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 linspaced_op;
+template <typename Scalar, bool RandomAccess> struct functor_traits< linspaced_op<Scalar,RandomAccess> >
+{ enum { Cost = 1, PacketAccess = packet_traits<Scalar>::HasSetLinear, IsRepeatable = true }; };
+template <typename Scalar, bool RandomAccess> struct linspaced_op
+{
+  typedef typename packet_traits<Scalar>::type Packet;
+  linspaced_op(const Scalar& low, const Scalar& high, DenseIndex num_steps) : impl((num_steps==1 ? high : low), (num_steps==1 ? Scalar() : (high-low)/(num_steps-1))) {}
+
+  template<typename Index>
+  EIGEN_STRONG_INLINE const Scalar operator() (Index i) const { return impl(i); }
+
+  // We need this function when assigning e.g. a RowVectorXd to a MatrixXd since
+  // there row==0 and col is used for the actual iteration.
+  template<typename Index>
+  EIGEN_STRONG_INLINE const Scalar operator() (Index row, Index col) const 
+  {
+    eigen_assert(col==0 || row==0);
+    return impl(col + row);
+  }
+
+  template<typename Index>
+  EIGEN_STRONG_INLINE const Packet packetOp(Index i) const { return impl.packetOp(i); }
+
+  // We need this function when assigning e.g. a RowVectorXd to a MatrixXd since
+  // there row==0 and col is used for the actual iteration.
+  template<typename Index>
+  EIGEN_STRONG_INLINE const Packet packetOp(Index row, Index col) const
+  {
+    eigen_assert(col==0 || row==0);
+    return impl.packetOp(col + row);
+  }
+
+  // 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 linspaced_op_impl<Scalar,RandomAccess> impl;
+};
+
+// all functors allow linear access, except scalar_identity_op. So we fix here a quick meta
+// to indicate whether a functor allows linear access, just always answering 'yes' except for
+// scalar_identity_op.
+// FIXME move this to functor_traits adding a functor_default
+template<typename Functor> struct functor_has_linear_access { enum { ret = 1 }; };
+template<typename Scalar> struct functor_has_linear_access<scalar_identity_op<Scalar> > { enum { ret = 0 }; };
+
+// In Eigen, any binary op (Product, CwiseBinaryOp) require the Lhs and Rhs to have the same scalar type, except for multiplication
+// where the mixing of different types is handled by scalar_product_traits
+// In particular, real * complex<real> is allowed.
+// FIXME move this to functor_traits adding a functor_default
+template<typename Functor> struct functor_is_product_like { enum { ret = 0 }; };
+template<typename LhsScalar,typename RhsScalar> struct functor_is_product_like<scalar_product_op<LhsScalar,RhsScalar> > { enum { ret = 1 }; };
+template<typename LhsScalar,typename RhsScalar> struct functor_is_product_like<scalar_conj_product_op<LhsScalar,RhsScalar> > { enum { ret = 1 }; };
+template<typename LhsScalar,typename RhsScalar> struct functor_is_product_like<scalar_quotient_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 pset1() in packetOp() is an optimization check scalar_multiple_op */
+template<typename Scalar>
+struct scalar_add_op {
+  typedef typename packet_traits<Scalar>::type Packet;
+  // FIXME default copy constructors seems bugged with std::complex<>
+  inline scalar_add_op(const scalar_add_op& other) : m_other(other.m_other) { }
+  inline 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 internal::padd(a, pset1<Packet>(m_other)); }
+  const Scalar m_other;
+};
+template<typename Scalar>
+struct functor_traits<scalar_add_op<Scalar> >
+{ enum { Cost = NumTraits<Scalar>::AddCost, PacketAccess = 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 scalar_sqrt_op {
+  EIGEN_EMPTY_STRUCT_CTOR(scalar_sqrt_op)
+  inline const Scalar operator() (const Scalar& a) const { using std::sqrt; return sqrt(a); }
+  typedef typename packet_traits<Scalar>::type Packet;
+  inline Packet packetOp(const Packet& a) const { return internal::psqrt(a); }
+};
+template<typename Scalar>
+struct functor_traits<scalar_sqrt_op<Scalar> >
+{ enum {
+    Cost = 5 * NumTraits<Scalar>::MulCost,
+    PacketAccess = packet_traits<Scalar>::HasSqrt
+  };
+};
+
+/** \internal
+  * \brief Template functor to compute the cosine of a scalar
+  * \sa class CwiseUnaryOp, ArrayBase::cos()
+  */
+template<typename Scalar> struct scalar_cos_op {
+  EIGEN_EMPTY_STRUCT_CTOR(scalar_cos_op)
+  inline Scalar operator() (const Scalar& a) const { using std::cos; return cos(a); }
+  typedef typename packet_traits<Scalar>::type Packet;
+  inline Packet packetOp(const Packet& a) const { return internal::pcos(a); }
+};
+template<typename Scalar>
+struct functor_traits<scalar_cos_op<Scalar> >
+{
+  enum {
+    Cost = 5 * NumTraits<Scalar>::MulCost,
+    PacketAccess = packet_traits<Scalar>::HasCos
+  };
+};
+
+/** \internal
+  * \brief Template functor to compute the sine of a scalar
+  * \sa class CwiseUnaryOp, ArrayBase::sin()
+  */
+template<typename Scalar> struct scalar_sin_op {
+  EIGEN_EMPTY_STRUCT_CTOR(scalar_sin_op)
+  inline const Scalar operator() (const Scalar& a) const { using std::sin; return sin(a); }
+  typedef typename packet_traits<Scalar>::type Packet;
+  inline Packet packetOp(const Packet& a) const { return internal::psin(a); }
+};
+template<typename Scalar>
+struct functor_traits<scalar_sin_op<Scalar> >
+{
+  enum {
+    Cost = 5 * NumTraits<Scalar>::MulCost,
+    PacketAccess = packet_traits<Scalar>::HasSin
+  };
+};
+
+
+/** \internal
+  * \brief Template functor to compute the tan of a scalar
+  * \sa class CwiseUnaryOp, ArrayBase::tan()
+  */
+template<typename Scalar> struct scalar_tan_op {
+  EIGEN_EMPTY_STRUCT_CTOR(scalar_tan_op)
+  inline const Scalar operator() (const Scalar& a) const { using std::tan; return tan(a); }
+  typedef typename packet_traits<Scalar>::type Packet;
+  inline Packet packetOp(const Packet& a) const { return internal::ptan(a); }
+};
+template<typename Scalar>
+struct functor_traits<scalar_tan_op<Scalar> >
+{
+  enum {
+    Cost = 5 * NumTraits<Scalar>::MulCost,
+    PacketAccess = packet_traits<Scalar>::HasTan
+  };
+};
+
+/** \internal
+  * \brief Template functor to compute the arc cosine of a scalar
+  * \sa class CwiseUnaryOp, ArrayBase::acos()
+  */
+template<typename Scalar> struct scalar_acos_op {
+  EIGEN_EMPTY_STRUCT_CTOR(scalar_acos_op)
+  inline const Scalar operator() (const Scalar& a) const { using std::acos; return acos(a); }
+  typedef typename packet_traits<Scalar>::type Packet;
+  inline Packet packetOp(const Packet& a) const { return internal::pacos(a); }
+};
+template<typename Scalar>
+struct functor_traits<scalar_acos_op<Scalar> >
+{
+  enum {
+    Cost = 5 * NumTraits<Scalar>::MulCost,
+    PacketAccess = packet_traits<Scalar>::HasACos
+  };
+};
+
+/** \internal
+  * \brief Template functor to compute the arc sine of a scalar
+  * \sa class CwiseUnaryOp, ArrayBase::asin()
+  */
+template<typename Scalar> struct scalar_asin_op {
+  EIGEN_EMPTY_STRUCT_CTOR(scalar_asin_op)
+  inline const Scalar operator() (const Scalar& a) const { using std::asin; return asin(a); }
+  typedef typename packet_traits<Scalar>::type Packet;
+  inline Packet packetOp(const Packet& a) const { return internal::pasin(a); }
+};
+template<typename Scalar>
+struct functor_traits<scalar_asin_op<Scalar> >
+{
+  enum {
+    Cost = 5 * NumTraits<Scalar>::MulCost,
+    PacketAccess = packet_traits<Scalar>::HasASin
+  };
+};
+
+/** \internal
+  * \brief Template functor to raise a scalar to a power
+  * \sa class CwiseUnaryOp, Cwise::pow
+  */
+template<typename Scalar>
+struct scalar_pow_op {
+  // FIXME default copy constructors seems bugged with std::complex<>
+  inline scalar_pow_op(const scalar_pow_op& other) : m_exponent(other.m_exponent) { }
+  inline scalar_pow_op(const Scalar& exponent) : m_exponent(exponent) {}
+  inline Scalar operator() (const Scalar& a) const { return numext::pow(a, m_exponent); }
+  const Scalar m_exponent;
+};
+template<typename Scalar>
+struct functor_traits<scalar_pow_op<Scalar> >
+{ enum { Cost = 5 * NumTraits<Scalar>::MulCost, PacketAccess = false }; };
+
+/** \internal
+  * \brief Template functor to compute the quotient between a scalar and array entries.
+  * \sa class CwiseUnaryOp, Cwise::inverse()
+  */
+template<typename Scalar>
+struct scalar_inverse_mult_op {
+  scalar_inverse_mult_op(const Scalar& other) : m_other(other) {}
+  inline Scalar operator() (const Scalar& a) const { return m_other / a; }
+  template<typename Packet>
+  inline const Packet packetOp(const Packet& a) const
+  { return internal::pdiv(pset1<Packet>(m_other),a); }
+  Scalar m_other;
+};
+
+/** \internal
+  * \brief Template functor to compute the inverse of a scalar
+  * \sa class CwiseUnaryOp, Cwise::inverse()
+  */
+template<typename Scalar>
+struct scalar_inverse_op {
+  EIGEN_EMPTY_STRUCT_CTOR(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 internal::pdiv(pset1<Packet>(Scalar(1)),a); }
+};
+template<typename Scalar>
+struct functor_traits<scalar_inverse_op<Scalar> >
+{ enum { Cost = NumTraits<Scalar>::MulCost, PacketAccess = packet_traits<Scalar>::HasDiv }; };
+
+/** \internal
+  * \brief Template functor to compute the square of a scalar
+  * \sa class CwiseUnaryOp, Cwise::square()
+  */
+template<typename Scalar>
+struct scalar_square_op {
+  EIGEN_EMPTY_STRUCT_CTOR(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 internal::pmul(a,a); }
+};
+template<typename Scalar>
+struct functor_traits<scalar_square_op<Scalar> >
+{ enum { Cost = NumTraits<Scalar>::MulCost, PacketAccess = packet_traits<Scalar>::HasMul }; };
+
+/** \internal
+  * \brief Template functor to compute the cube of a scalar
+  * \sa class CwiseUnaryOp, Cwise::cube()
+  */
+template<typename Scalar>
+struct scalar_cube_op {
+  EIGEN_EMPTY_STRUCT_CTOR(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 internal::pmul(a,pmul(a,a)); }
+};
+template<typename Scalar>
+struct functor_traits<scalar_cube_op<Scalar> >
+{ enum { Cost = 2*NumTraits<Scalar>::MulCost, PacketAccess = packet_traits<Scalar>::HasMul }; };
+
+// default functor traits for STL functors:
+
+template<typename T>
+struct functor_traits<std::multiplies<T> >
+{ enum { Cost = NumTraits<T>::MulCost, PacketAccess = false }; };
+
+template<typename T>
+struct functor_traits<std::divides<T> >
+{ enum { Cost = NumTraits<T>::MulCost, PacketAccess = false }; };
+
+template<typename T>
+struct functor_traits<std::plus<T> >
+{ enum { Cost = NumTraits<T>::AddCost, PacketAccess = false }; };
+
+template<typename T>
+struct functor_traits<std::minus<T> >
+{ enum { Cost = NumTraits<T>::AddCost, PacketAccess = false }; };
+
+template<typename T>
+struct functor_traits<std::negate<T> >
+{ enum { Cost = NumTraits<T>::AddCost, PacketAccess = false }; };
+
+template<typename T>
+struct functor_traits<std::logical_or<T> >
+{ enum { Cost = 1, PacketAccess = false }; };
+
+template<typename T>
+struct functor_traits<std::logical_and<T> >
+{ enum { Cost = 1, PacketAccess = false }; };
+
+template<typename T>
+struct functor_traits<std::logical_not<T> >
+{ enum { Cost = 1, PacketAccess = false }; };
+
+template<typename T>
+struct functor_traits<std::greater<T> >
+{ enum { Cost = 1, PacketAccess = false }; };
+
+template<typename T>
+struct functor_traits<std::less<T> >
+{ enum { Cost = 1, PacketAccess = false }; };
+
+template<typename T>
+struct functor_traits<std::greater_equal<T> >
+{ enum { Cost = 1, PacketAccess = false }; };
+
+template<typename T>
+struct functor_traits<std::less_equal<T> >
+{ enum { Cost = 1, PacketAccess = false }; };
+
+template<typename T>
+struct functor_traits<std::equal_to<T> >
+{ enum { Cost = 1, PacketAccess = false }; };
+
+template<typename T>
+struct functor_traits<std::not_equal_to<T> >
+{ enum { Cost = 1, PacketAccess = false }; };
+
+template<typename T>
+struct functor_traits<std::binder2nd<T> >
+{ enum { Cost = functor_traits<T>::Cost, PacketAccess = false }; };
+
+template<typename T>
+struct functor_traits<std::binder1st<T> >
+{ enum { Cost = functor_traits<T>::Cost, PacketAccess = false }; };
+
+template<typename T>
+struct functor_traits<std::unary_negate<T> >
+{ enum { Cost = 1 + functor_traits<T>::Cost, PacketAccess = false }; };
+
+template<typename T>
+struct functor_traits<std::binary_negate<T> >
+{ enum { Cost = 1 + functor_traits<T>::Cost, PacketAccess = false }; };
+
+#ifdef EIGEN_STDEXT_SUPPORT
+
+template<typename T0,typename T1>
+struct functor_traits<std::project1st<T0,T1> >
+{ enum { Cost = 0, PacketAccess = false }; };
+
+template<typename T0,typename T1>
+struct functor_traits<std::project2nd<T0,T1> >
+{ enum { Cost = 0, PacketAccess = false }; };
+
+template<typename T0,typename T1>
+struct functor_traits<std::select2nd<std::pair<T0,T1> > >
+{ enum { Cost = 0, PacketAccess = false }; };
+
+template<typename T0,typename T1>
+struct functor_traits<std::select1st<std::pair<T0,T1> > >
+{ enum { Cost = 0, PacketAccess = false }; };
+
+template<typename T0,typename T1>
+struct functor_traits<std::unary_compose<T0,T1> >
+{ enum { Cost = functor_traits<T0>::Cost + functor_traits<T1>::Cost, PacketAccess = false }; };
+
+template<typename T0,typename T1,typename T2>
+struct functor_traits<std::binary_compose<T0,T1,T2> >
+{ enum { Cost = functor_traits<T0>::Cost + functor_traits<T1>::Cost + functor_traits<T2>::Cost, PacketAccess = false }; };
+
+#endif // EIGEN_STDEXT_SUPPORT
+
+// allow to add new functors and specializations of functor_traits from outside Eigen.
+// this macro is really needed because functor_traits must be specialized after it is declared but before it is used...
+#ifdef EIGEN_FUNCTORS_PLUGIN
+#include EIGEN_FUNCTORS_PLUGIN
+#endif
+
+} // end namespace internal
+
+} // end namespace Eigen
+
+#endif // EIGEN_FUNCTORS_H

Eigen/src/Core/Fuzzy.h

@@ -19,19 +19,18 @@ 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());
+    using std::min;
+    typename internal::nested<Derived,2>::type nested(x);
+    typename internal::nested<OtherDerived,2>::type otherNested(y);
+    return (nested - otherNested).cwiseAbs2().sum() <= prec * prec * (min)(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();
@@ -41,7 +40,6 @@ struct isApprox_selector<Derived, OtherDerived, true>
 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();
@@ -51,7 +49,6 @@ struct isMuchSmallerThan_object_selector
 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();
@@ -61,7 +58,6 @@ struct isMuchSmallerThan_object_selector<Derived, OtherDerived, true>
 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);
@@ -71,7 +67,6 @@ struct isMuchSmallerThan_scalar_selector
 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();
@@ -100,7 +95,7 @@ struct isMuchSmallerThan_scalar_selector<Derived, true>
   */
 template<typename Derived>
 template<typename OtherDerived>
-EIGEN_DEVICE_FUNC bool DenseBase<Derived>::isApprox(
+bool DenseBase<Derived>::isApprox(
   const DenseBase<OtherDerived>& other,
   const RealScalar& prec
 ) const
@@ -122,7 +117,7 @@ EIGEN_DEVICE_FUNC bool DenseBase<Derived>::isApprox(
   * \sa isApprox(), isMuchSmallerThan(const DenseBase<OtherDerived>&, RealScalar) const
   */
 template<typename Derived>
-EIGEN_DEVICE_FUNC bool DenseBase<Derived>::isMuchSmallerThan(
+bool DenseBase<Derived>::isMuchSmallerThan(
   const typename NumTraits<Scalar>::Real& other,
   const RealScalar& prec
 ) const
@@ -142,7 +137,7 @@ EIGEN_DEVICE_FUNC bool DenseBase<Derived>::isMuchSmallerThan(
   */
 template<typename Derived>
 template<typename OtherDerived>
-EIGEN_DEVICE_FUNC bool DenseBase<Derived>::isMuchSmallerThan(
+bool DenseBase<Derived>::isMuchSmallerThan(
   const DenseBase<OtherDerived>& other,
   const RealScalar& prec
 ) const

Eigen/src/Core/GeneralProduct.h

@@ -11,40 +11,46 @@
 #ifndef EIGEN_GENERAL_PRODUCT_H
 #define EIGEN_GENERAL_PRODUCT_H
 
-namespace Eigen {
+namespace Eigen { 
+
+/** \class GeneralProduct
+  * \ingroup Core_Module
+  *
+  * \brief Expression of the product of two general matrices or vectors
+  *
+  * \param LhsNested the type used to store the left-hand side
+  * \param RhsNested the type used to store the right-hand side
+  * \param ProductMode the type of the product
+  *
+  * This class represents an expression of the product of two general matrices.
+  * We call a general matrix, a dense matrix with full storage. For instance,
+  * This excludes triangular, selfadjoint, and sparse matrices.
+  * It is the return type of the operator* between general matrices. Its template
+  * arguments are determined automatically by ProductReturnType. Therefore,
+  * GeneralProduct should never be used direclty. To determine the result type of a
+  * function which involves a matrix product, use ProductReturnType::Type.
+  *
+  * \sa ProductReturnType, MatrixBase::operator*(const MatrixBase<OtherDerived>&)
+  */
+template<typename Lhs, typename Rhs, int ProductType = internal::product_type<Lhs,Rhs>::value>
+class GeneralProduct;
 
 enum {
   Large = 2,
   Small = 3
 };
 
-// Define the threshold value to fallback from the generic matrix-matrix product
-// implementation (heavy) to the lightweight coeff-based product one.
-// See generic_product_impl<Lhs,Rhs,DenseShape,DenseShape,GemmProduct>
-// in products/GeneralMatrixMatrix.h for more details.
-// TODO This threshold should also be used in the compile-time selector below.
-#ifndef EIGEN_GEMM_TO_COEFFBASED_THRESHOLD
-// This default value has been obtained on a Haswell architecture.
-#define EIGEN_GEMM_TO_COEFFBASED_THRESHOLD 20
-#endif
-
 namespace internal {
 
 template<int Rows, int Cols, int Depth> struct product_type_selector;
 
 template<int Size, int MaxSize> struct product_size_category
 {
-  enum {
-    #ifndef EIGEN_GPU_COMPILE_PHASE
-    is_large = MaxSize == Dynamic ||
-               Size >= EIGEN_CACHEFRIENDLY_PRODUCT_THRESHOLD ||
-               (Size==Dynamic && MaxSize>=EIGEN_CACHEFRIENDLY_PRODUCT_THRESHOLD),
-    #else
-    is_large = 0,
-    #endif
-    value = is_large  ? Large
-          : Size == 1 ? 1
-                      : Small
+  enum { is_large = MaxSize == Dynamic ||
+                    Size >= EIGEN_CACHEFRIENDLY_PRODUCT_THRESHOLD,
+         value = is_large  ? Large
+               : Size == 1 ? 1
+                           : Small
   };
 };
 
@@ -53,14 +59,15 @@ 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)
+    MaxRows  = _Lhs::MaxRowsAtCompileTime,
+    Rows  = _Lhs::RowsAtCompileTime,
+    MaxCols  = _Rhs::MaxColsAtCompileTime,
+    Cols  = _Rhs::ColsAtCompileTime,
+    MaxDepth = EIGEN_SIZE_MIN_PREFER_FIXED(_Lhs::MaxColsAtCompileTime,
+                                           _Rhs::MaxRowsAtCompileTime),
+    Depth = EIGEN_SIZE_MIN_PREFER_FIXED(_Lhs::ColsAtCompileTime,
+                                        _Rhs::RowsAtCompileTime),
+    LargeThreshold = EIGEN_CACHEFRIENDLY_PRODUCT_THRESHOLD
   };
 
   // the splitting into different lines of code here, introducing the _select enums and the typedef below,
@@ -75,8 +82,7 @@ private:
 
 public:
   enum {
-    value = selector::ret,
-    ret = selector::ret
+    value = selector::ret
   };
 #ifdef EIGEN_DEBUG_PRODUCT
   static void debug()
@@ -92,13 +98,12 @@ public:
 #endif
 };
 
+
 /* The following allows to select the kind of product at compile time
  * based on the three dimensions of the product.
  * This is a compile time mapping from {1,Small,Large}^3 -> {product types} */
 // FIXME I'm not sure the current mapping is the ideal one.
 template<int M, int N>  struct product_type_selector<M,N,1>              { enum { ret = OuterProduct }; };
-template<int M>         struct product_type_selector<M, 1, 1>            { enum { ret = LazyCoeffBasedProductMode }; };
-template<int N>         struct product_type_selector<1, N, 1>            { enum { ret = LazyCoeffBasedProductMode }; };
 template<int Depth>     struct product_type_selector<1,    1,    Depth>  { enum { ret = InnerProduct }; };
 template<>              struct product_type_selector<1,    1,    1>      { enum { ret = InnerProduct }; };
 template<>              struct product_type_selector<Small,1,    Small>  { enum { ret = CoeffBasedProductMode }; };
@@ -117,12 +122,60 @@ template<>              struct product_type_selector<Small,Small,Large>  { enum
 template<>              struct product_type_selector<Large,Small,Large>  { enum { ret = GemmProduct }; };
 template<>              struct product_type_selector<Small,Large,Large>  { enum { ret = GemmProduct }; };
 template<>              struct product_type_selector<Large,Large,Large>  { enum { ret = GemmProduct }; };
-template<>              struct product_type_selector<Large,Small,Small>  { enum { ret = CoeffBasedProductMode }; };
-template<>              struct product_type_selector<Small,Large,Small>  { enum { ret = CoeffBasedProductMode }; };
+template<>              struct product_type_selector<Large,Small,Small>  { enum { ret = GemmProduct }; };
+template<>              struct product_type_selector<Small,Large,Small>  { enum { ret = GemmProduct }; };
 template<>              struct product_type_selector<Large,Large,Small>  { enum { ret = GemmProduct }; };
 
 } // end namespace internal
 
+/** \class ProductReturnType
+  * \ingroup Core_Module
+  *
+  * \brief Helper class to get the correct and optimized returned type of operator*
+  *
+  * \param Lhs the type of the left-hand side
+  * \param Rhs the type of the right-hand side
+  * \param ProductMode the type of the product (determined automatically by internal::product_mode)
+  *
+  * This class defines the typename Type representing the optimized product expression
+  * between two matrix expressions. In practice, using ProductReturnType<Lhs,Rhs>::Type
+  * is the recommended way to define the result type of a function returning an expression
+  * which involve a matrix product. The class Product should never be
+  * used directly.
+  *
+  * \sa class Product, MatrixBase::operator*(const MatrixBase<OtherDerived>&)
+  */
+template<typename Lhs, typename Rhs, int ProductType>
+struct ProductReturnType
+{
+  // TODO use the nested type to reduce instanciations ????
+//   typedef typename internal::nested<Lhs,Rhs::ColsAtCompileTime>::type LhsNested;
+//   typedef typename internal::nested<Rhs,Lhs::RowsAtCompileTime>::type RhsNested;
+
+  typedef GeneralProduct<Lhs/*Nested*/, Rhs/*Nested*/, ProductType> Type;
+};
+
+template<typename Lhs, typename Rhs>
+struct ProductReturnType<Lhs,Rhs,CoeffBasedProductMode>
+{
+  typedef typename internal::nested<Lhs, Rhs::ColsAtCompileTime, typename internal::plain_matrix_type<Lhs>::type >::type LhsNested;
+  typedef typename internal::nested<Rhs, Lhs::RowsAtCompileTime, typename internal::plain_matrix_type<Rhs>::type >::type RhsNested;
+  typedef CoeffBasedProduct<LhsNested, RhsNested, EvalBeforeAssigningBit | EvalBeforeNestingBit> Type;
+};
+
+template<typename Lhs, typename Rhs>
+struct ProductReturnType<Lhs,Rhs,LazyCoeffBasedProductMode>
+{
+  typedef typename internal::nested<Lhs, Rhs::ColsAtCompileTime, typename internal::plain_matrix_type<Lhs>::type >::type LhsNested;
+  typedef typename internal::nested<Rhs, Lhs::RowsAtCompileTime, typename internal::plain_matrix_type<Rhs>::type >::type RhsNested;
+  typedef CoeffBasedProduct<LhsNested, RhsNested, NestByRefBit> Type;
+};
+
+// this is a workaround for sun CC
+template<typename Lhs, typename Rhs>
+struct LazyProductReturnType : public ProductReturnType<Lhs,Rhs,LazyCoeffBasedProductMode>
+{};
+
 /***********************************************************************
 *  Implementation of Inner Vector Vector Product
 ***********************************************************************/
@@ -134,10 +187,119 @@ template<>              struct product_type_selector<Large,Large,Small>  { enum
 // product ends up to a row-vector times col-vector product... To tackle this use
 // case, we could have a specialization for Block<MatrixType,1,1> with: operator=(Scalar x);
 
+namespace internal {
+
+template<typename Lhs, typename Rhs>
+struct traits<GeneralProduct<Lhs,Rhs,InnerProduct> >
+ : traits<Matrix<typename scalar_product_traits<typename Lhs::Scalar, typename Rhs::Scalar>::ReturnType,1,1> >
+{};
+
+}
+
+template<typename Lhs, typename Rhs>
+class GeneralProduct<Lhs, Rhs, InnerProduct>
+  : internal::no_assignment_operator,
+    public Matrix<typename internal::scalar_product_traits<typename Lhs::Scalar, typename Rhs::Scalar>::ReturnType,1,1>
+{
+    typedef Matrix<typename internal::scalar_product_traits<typename Lhs::Scalar, typename Rhs::Scalar>::ReturnType,1,1> Base;
+  public:
+    GeneralProduct(const Lhs& lhs, const Rhs& rhs)
+    {
+      EIGEN_STATIC_ASSERT((internal::is_same<typename Lhs::RealScalar, typename Rhs::RealScalar>::value),
+        YOU_MIXED_DIFFERENT_NUMERIC_TYPES__YOU_NEED_TO_USE_THE_CAST_METHOD_OF_MATRIXBASE_TO_CAST_NUMERIC_TYPES_EXPLICITLY)
+
+      Base::coeffRef(0,0) = (lhs.transpose().cwiseProduct(rhs)).sum();
+    }
+
+    /** Convertion to scalar */
+    operator const typename Base::Scalar() const {
+      return Base::coeff(0,0);
+    }
+};
+
 /***********************************************************************
 *  Implementation of Outer Vector Vector Product
 ***********************************************************************/
 
+namespace internal {
+
+// Column major
+template<typename ProductType, typename Dest, typename Func>
+EIGEN_DONT_INLINE void outer_product_selector_run(const ProductType& prod, Dest& dest, const Func& func, const false_type&)
+{
+  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)
+    func(dest.col(j), prod.rhs().coeff(j) * prod.lhs());
+}
+
+// Row major
+template<typename ProductType, typename Dest, typename Func>
+EIGEN_DONT_INLINE void outer_product_selector_run(const ProductType& prod, Dest& dest, const Func& func, const true_type&) {
+  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)
+    func(dest.row(i), prod.lhs().coeff(i) * prod.rhs());
+}
+
+template<typename Lhs, typename Rhs>
+struct traits<GeneralProduct<Lhs,Rhs,OuterProduct> >
+ : traits<ProductBase<GeneralProduct<Lhs,Rhs,OuterProduct>, Lhs, Rhs> >
+{};
+
+}
+
+template<typename Lhs, typename Rhs>
+class GeneralProduct<Lhs, Rhs, OuterProduct>
+  : public ProductBase<GeneralProduct<Lhs,Rhs,OuterProduct>, Lhs, Rhs>
+{
+    template<typename T> struct IsRowMajor : internal::conditional<(int(T::Flags)&RowMajorBit), internal::true_type, internal::false_type>::type {};
+    
+  public:
+    EIGEN_PRODUCT_PUBLIC_INTERFACE(GeneralProduct)
+
+    GeneralProduct(const Lhs& lhs, const Rhs& rhs) : Base(lhs,rhs)
+    {
+      EIGEN_STATIC_ASSERT((internal::is_same<typename Lhs::RealScalar, typename Rhs::RealScalar>::value),
+        YOU_MIXED_DIFFERENT_NUMERIC_TYPES__YOU_NEED_TO_USE_THE_CAST_METHOD_OF_MATRIXBASE_TO_CAST_NUMERIC_TYPES_EXPLICITLY)
+    }
+    
+    struct set  { template<typename Dst, typename Src> void operator()(const Dst& dst, const Src& src) const { dst.const_cast_derived()  = src; } };
+    struct add  { template<typename Dst, typename Src> void operator()(const Dst& dst, const Src& src) const { dst.const_cast_derived() += src; } };
+    struct sub  { template<typename Dst, typename Src> void operator()(const Dst& dst, const Src& src) const { dst.const_cast_derived() -= src; } };
+    struct adds {
+      Scalar m_scale;
+      adds(const Scalar& s) : m_scale(s) {}
+      template<typename Dst, typename Src> void operator()(const Dst& dst, const Src& src) const {
+        dst.const_cast_derived() += m_scale * src;
+      }
+    };
+    
+    template<typename Dest>
+    inline void evalTo(Dest& dest) const {
+      internal::outer_product_selector_run(*this, dest, set(), IsRowMajor<Dest>());
+    }
+    
+    template<typename Dest>
+    inline void addTo(Dest& dest) const {
+      internal::outer_product_selector_run(*this, dest, add(), IsRowMajor<Dest>());
+    }
+
+    template<typename Dest>
+    inline void subTo(Dest& dest) const {
+      internal::outer_product_selector_run(*this, dest, sub(), IsRowMajor<Dest>());
+    }
+
+    template<typename Dest> void scaleAndAddTo(Dest& dest, const Scalar& alpha) const
+    {
+      internal::outer_product_selector_run(*this, dest, adds(alpha), IsRowMajor<Dest>());
+    }
+};
+
 /***********************************************************************
 *  Implementation of General Matrix Vector Product
 ***********************************************************************/
@@ -151,183 +313,197 @@ template<>              struct product_type_selector<Large,Large,Small>  { enum
  */
 namespace internal {
 
+template<typename Lhs, typename Rhs>
+struct traits<GeneralProduct<Lhs,Rhs,GemvProduct> >
+ : traits<ProductBase<GeneralProduct<Lhs,Rhs,GemvProduct>, Lhs, Rhs> >
+{};
+
 template<int Side, int StorageOrder, bool BlasCompatible>
-struct gemv_dense_selector;
+struct gemv_selector;
 
 } // end namespace internal
 
+template<typename Lhs, typename Rhs>
+class GeneralProduct<Lhs, Rhs, GemvProduct>
+  : public ProductBase<GeneralProduct<Lhs,Rhs,GemvProduct>, Lhs, Rhs>
+{
+  public:
+    EIGEN_PRODUCT_PUBLIC_INTERFACE(GeneralProduct)
+
+    typedef typename Lhs::Scalar LhsScalar;
+    typedef typename Rhs::Scalar RhsScalar;
+
+    GeneralProduct(const Lhs& a_lhs, const Rhs& a_rhs) : Base(a_lhs,a_rhs)
+    {
+//       EIGEN_STATIC_ASSERT((internal::is_same<typename Lhs::Scalar, typename Rhs::Scalar>::value),
+//         YOU_MIXED_DIFFERENT_NUMERIC_TYPES__YOU_NEED_TO_USE_THE_CAST_METHOD_OF_MATRIXBASE_TO_CAST_NUMERIC_TYPES_EXPLICITLY)
+    }
+
+    enum { Side = Lhs::IsVectorAtCompileTime ? OnTheLeft : OnTheRight };
+    typedef typename internal::conditional<int(Side)==OnTheRight,_LhsNested,_RhsNested>::type MatrixType;
+
+    template<typename Dest> void scaleAndAddTo(Dest& dst, const Scalar& alpha) const
+    {
+      eigen_assert(m_lhs.rows() == dst.rows() && m_rhs.cols() == dst.cols());
+      internal::gemv_selector<Side,(int(MatrixType::Flags)&RowMajorBit) ? RowMajor : ColMajor,
+                       bool(internal::blas_traits<MatrixType>::HasUsableDirectAccess)>::run(*this, dst, alpha);
+    }
+};
+
 namespace internal {
 
+// The vector is on the left => transposition
+template<int StorageOrder, bool BlasCompatible>
+struct gemv_selector<OnTheLeft,StorageOrder,BlasCompatible>
+{
+  template<typename ProductType, typename Dest>
+  static void run(const ProductType& prod, Dest& dest, const typename ProductType::Scalar& alpha)
+  {
+    Transpose<Dest> destT(dest);
+    enum { OtherStorageOrder = StorageOrder == RowMajor ? ColMajor : RowMajor };
+    gemv_selector<OnTheRight,OtherStorageOrder,BlasCompatible>
+      ::run(GeneralProduct<Transpose<const typename ProductType::_RhsNested>,Transpose<const typename ProductType::_LhsNested>, GemvProduct>
+        (prod.rhs().transpose(), prod.lhs().transpose()), destT, alpha);
+  }
+};
+
 template<typename Scalar,int Size,int MaxSize,bool Cond> struct gemv_static_vector_if;
 
 template<typename Scalar,int Size,int MaxSize>
 struct gemv_static_vector_if<Scalar,Size,MaxSize,false>
 {
-  EIGEN_STRONG_INLINE EIGEN_DEVICE_FUNC Scalar* data() { eigen_internal_assert(false && "should never be called"); return 0; }
+  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 EIGEN_DEVICE_FUNC Scalar* data() { return 0; }
+  EIGEN_STRONG_INLINE Scalar* data() { return 0; }
 };
 
 template<typename Scalar,int Size,int MaxSize>
 struct gemv_static_vector_if<Scalar,Size,MaxSize,true>
 {
-  enum {
-    ForceAlignment  = internal::packet_traits<Scalar>::Vectorizable,
-    PacketSize      = internal::packet_traits<Scalar>::size
-  };
-  #if EIGEN_MAX_STATIC_ALIGN_BYTES!=0
-  internal::plain_array<Scalar,EIGEN_SIZE_MIN_PREFER_FIXED(Size,MaxSize),0,EIGEN_PLAIN_ENUM_MIN(AlignedMax,PacketSize)> m_data;
+  #if EIGEN_ALIGN_STATICALLY
+  internal::plain_array<Scalar,EIGEN_SIZE_MIN_PREFER_FIXED(Size,MaxSize),0> m_data;
   EIGEN_STRONG_INLINE Scalar* data() { return m_data.array; }
   #else
   // Some architectures cannot align on the stack,
   // => let's manually enforce alignment by allocating more data and return the address of the first aligned element.
-  internal::plain_array<Scalar,EIGEN_SIZE_MIN_PREFER_FIXED(Size,MaxSize)+(ForceAlignment?EIGEN_MAX_ALIGN_BYTES:0),0> m_data;
+  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*>((internal::UIntPtr(m_data.array) & ~(std::size_t(EIGEN_MAX_ALIGN_BYTES-1))) + EIGEN_MAX_ALIGN_BYTES)
+            ? reinterpret_cast<Scalar*>((reinterpret_cast<size_t>(m_data.array) & ~(size_t(15))) + 16)
             : m_data.array;
   }
   #endif
 };
 
-// The vector is on the left => transposition
-template<int StorageOrder, bool BlasCompatible>
-struct gemv_dense_selector<OnTheLeft,StorageOrder,BlasCompatible>
+template<> struct gemv_selector<OnTheRight,ColMajor,true>
 {
-  template<typename Lhs, typename Rhs, typename Dest>
-  static void run(const Lhs &lhs, const Rhs &rhs, Dest& dest, const typename Dest::Scalar& alpha)
+  template<typename ProductType, typename Dest>
+  static inline void run(const ProductType& prod, Dest& dest, const typename ProductType::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);
-  }
-};
+    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;
 
-template<> struct gemv_dense_selector<OnTheRight,ColMajor,true>
-{
-  template<typename Lhs, typename Rhs, typename Dest>
-  static inline void run(const Lhs &lhs, const Rhs &rhs, Dest& dest, const typename Dest::Scalar& alpha)
-  {
-    typedef typename Lhs::Scalar   LhsScalar;
-    typedef typename Rhs::Scalar   RhsScalar;
-    typedef typename Dest::Scalar  ResScalar;
-    typedef typename Dest::RealScalar  RealScalar;
-    
-    typedef internal::blas_traits<Lhs> LhsBlasTraits;
-    typedef typename LhsBlasTraits::DirectLinearAccessType ActualLhsType;
-    typedef internal::blas_traits<Rhs> RhsBlasTraits;
-    typedef typename RhsBlasTraits::DirectLinearAccessType ActualRhsType;
-  
-    typedef Map<Matrix<ResScalar,Dynamic,1>, EIGEN_PLAIN_ENUM_MIN(AlignedMax,internal::packet_traits<ResScalar>::size)> MappedDest;
+    ActualLhsType actualLhs = LhsBlasTraits::extract(prod.lhs());
+    ActualRhsType actualRhs = RhsBlasTraits::extract(prod.rhs());
 
-    ActualLhsType actualLhs = LhsBlasTraits::extract(lhs);
-    ActualRhsType actualRhs = RhsBlasTraits::extract(rhs);
-
-    ResScalar actualAlpha = combine_scalar_factors(alpha, lhs, rhs);
-
-    // make sure Dest is a compile-time vector type (bug 1166)
-    typedef typename conditional<Dest::IsVectorAtCompileTime, Dest, typename Dest::ColXpr>::type ActualDest;
+    ResScalar actualAlpha = alpha * LhsBlasTraits::extractScalarFactor(prod.lhs())
+                                  * RhsBlasTraits::extractScalarFactor(prod.rhs());
 
     enum {
       // FIXME find a way to allow an inner stride on the result if packet_traits<Scalar>::size==1
       // on, the other hand it is good for the cache to pack the vector anyways...
-      EvalToDestAtCompileTime = (ActualDest::InnerStrideAtCompileTime==1),
+      EvalToDestAtCompileTime = Dest::InnerStrideAtCompileTime==1,
       ComplexByReal = (NumTraits<LhsScalar>::IsComplex) && (!NumTraits<RhsScalar>::IsComplex),
-      MightCannotUseDest = ((!EvalToDestAtCompileTime) || ComplexByReal) && (ActualDest::MaxSizeAtCompileTime!=0)
+      MightCannotUseDest = (Dest::InnerStrideAtCompileTime!=1) || ComplexByReal
     };
 
-    typedef const_blas_data_mapper<LhsScalar,Index,ColMajor> LhsMapper;
-    typedef const_blas_data_mapper<RhsScalar,Index,RowMajor> RhsMapper;
+    gemv_static_vector_if<ResScalar,Dest::SizeAtCompileTime,Dest::MaxSizeAtCompileTime,MightCannotUseDest> static_dest;
+
+    bool alphaIsCompatible = (!ComplexByReal) || (numext::imag(actualAlpha)==RealScalar(0));
+    bool evalToDest = EvalToDestAtCompileTime && alphaIsCompatible;
+    
     RhsScalar compatibleAlpha = get_factor<ResScalar,RhsScalar>::run(actualAlpha);
 
-    if(!MightCannotUseDest)
+    ei_declare_aligned_stack_constructed_variable(ResScalar,actualDestPtr,dest.size(),
+                                                  evalToDest ? dest.data() : static_dest.data());
+    
+    if(!evalToDest)
     {
-      // shortcut if we are sure to be able to use dest directly,
-      // this ease the compiler to generate cleaner and more optimzized code for most common cases
-      general_matrix_vector_product
-          <Index,LhsScalar,LhsMapper,ColMajor,LhsBlasTraits::NeedToConjugate,RhsScalar,RhsMapper,RhsBlasTraits::NeedToConjugate>::run(
-          actualLhs.rows(), actualLhs.cols(),
-          LhsMapper(actualLhs.data(), actualLhs.outerStride()),
-          RhsMapper(actualRhs.data(), actualRhs.innerStride()),
-          dest.data(), 1,
-          compatibleAlpha);
+      #ifdef EIGEN_DENSE_STORAGE_CTOR_PLUGIN
+      int size = dest.size();
+      EIGEN_DENSE_STORAGE_CTOR_PLUGIN
+      #endif
+      if(!alphaIsCompatible)
+      {
+        MappedDest(actualDestPtr, dest.size()).setZero();
+        compatibleAlpha = RhsScalar(1);
+      }
+      else
+        MappedDest(actualDestPtr, dest.size()) = dest;
     }
-    else
+
+    general_matrix_vector_product
+      <Index,LhsScalar,ColMajor,LhsBlasTraits::NeedToConjugate,RhsScalar,RhsBlasTraits::NeedToConjugate>::run(
+        actualLhs.rows(), actualLhs.cols(),
+        actualLhs.data(), actualLhs.outerStride(),
+        actualRhs.data(), actualRhs.innerStride(),
+        actualDestPtr, 1,
+        compatibleAlpha);
+
+    if (!evalToDest)
     {
-      gemv_static_vector_if<ResScalar,ActualDest::SizeAtCompileTime,ActualDest::MaxSizeAtCompileTime,MightCannotUseDest> static_dest;
-
-      const bool alphaIsCompatible = (!ComplexByReal) || (numext::imag(actualAlpha)==RealScalar(0));
-      const bool evalToDest = EvalToDestAtCompileTime && alphaIsCompatible;
-
-      ei_declare_aligned_stack_constructed_variable(ResScalar,actualDestPtr,dest.size(),
-                                                    evalToDest ? dest.data() : static_dest.data());
-
-      if(!evalToDest)
-      {
-        #ifdef EIGEN_DENSE_STORAGE_CTOR_PLUGIN
-        Index size = dest.size();
-        EIGEN_DENSE_STORAGE_CTOR_PLUGIN
-        #endif
-        if(!alphaIsCompatible)
-        {
-          MappedDest(actualDestPtr, dest.size()).setZero();
-          compatibleAlpha = RhsScalar(1);
-        }
-        else
-          MappedDest(actualDestPtr, dest.size()) = dest;
-      }
-
-      general_matrix_vector_product
-          <Index,LhsScalar,LhsMapper,ColMajor,LhsBlasTraits::NeedToConjugate,RhsScalar,RhsMapper,RhsBlasTraits::NeedToConjugate>::run(
-          actualLhs.rows(), actualLhs.cols(),
-          LhsMapper(actualLhs.data(), actualLhs.outerStride()),
-          RhsMapper(actualRhs.data(), actualRhs.innerStride()),
-          actualDestPtr, 1,
-          compatibleAlpha);
-
-      if (!evalToDest)
-      {
-        if(!alphaIsCompatible)
-          dest.matrix() += actualAlpha * MappedDest(actualDestPtr, dest.size());
-        else
-          dest = MappedDest(actualDestPtr, dest.size());
-      }
+      if(!alphaIsCompatible)
+        dest += actualAlpha * MappedDest(actualDestPtr, dest.size());
+      else
+        dest = MappedDest(actualDestPtr, dest.size());
     }
   }
 };
 
-template<> struct gemv_dense_selector<OnTheRight,RowMajor,true>
+template<> struct gemv_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)
+  template<typename ProductType, typename Dest>
+  static void run(const ProductType& prod, Dest& dest, const typename ProductType::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;
+    typedef typename ProductType::LhsScalar LhsScalar;
+    typedef typename ProductType::RhsScalar RhsScalar;
+    typedef typename ProductType::Scalar    ResScalar;
+    typedef typename ProductType::Index Index;
+    typedef typename ProductType::ActualLhsType ActualLhsType;
+    typedef typename ProductType::ActualRhsType ActualRhsType;
+    typedef typename ProductType::_ActualRhsType _ActualRhsType;
+    typedef typename ProductType::LhsBlasTraits LhsBlasTraits;
+    typedef typename ProductType::RhsBlasTraits RhsBlasTraits;
 
-    typename add_const<ActualLhsType>::type actualLhs = LhsBlasTraits::extract(lhs);
-    typename add_const<ActualRhsType>::type actualRhs = RhsBlasTraits::extract(rhs);
+    typename add_const<ActualLhsType>::type actualLhs = LhsBlasTraits::extract(prod.lhs());
+    typename add_const<ActualRhsType>::type actualRhs = RhsBlasTraits::extract(prod.rhs());
 
-    ResScalar actualAlpha = combine_scalar_factors(alpha, lhs, rhs);
+    ResScalar actualAlpha = alpha * LhsBlasTraits::extractScalarFactor(prod.lhs())
+                                  * RhsBlasTraits::extractScalarFactor(prod.rhs());
 
     enum {
       // FIXME find a way to allow an inner stride on the result if packet_traits<Scalar>::size==1
       // on, the other hand it is good for the cache to pack the vector anyways...
-      DirectlyUseRhs = ActualRhsTypeCleaned::InnerStrideAtCompileTime==1 || ActualRhsTypeCleaned::MaxSizeAtCompileTime==0
+      DirectlyUseRhs = _ActualRhsType::InnerStrideAtCompileTime==1
     };
 
-    gemv_static_vector_if<RhsScalar,ActualRhsTypeCleaned::SizeAtCompileTime,ActualRhsTypeCleaned::MaxSizeAtCompileTime,!DirectlyUseRhs> static_rhs;
+    gemv_static_vector_if<RhsScalar,_ActualRhsType::SizeAtCompileTime,_ActualRhsType::MaxSizeAtCompileTime,!DirectlyUseRhs> static_rhs;
 
     ei_declare_aligned_stack_constructed_variable(RhsScalar,actualRhsPtr,actualRhs.size(),
         DirectlyUseRhs ? const_cast<RhsScalar*>(actualRhs.data()) : static_rhs.data());
@@ -335,48 +511,45 @@ template<> struct gemv_dense_selector<OnTheRight,RowMajor,true>
     if(!DirectlyUseRhs)
     {
       #ifdef EIGEN_DENSE_STORAGE_CTOR_PLUGIN
-      Index size = actualRhs.size();
+      int size = actualRhs.size();
       EIGEN_DENSE_STORAGE_CTOR_PLUGIN
       #endif
-      Map<typename ActualRhsTypeCleaned::PlainObject>(actualRhsPtr, actualRhs.size()) = actualRhs;
+      Map<typename _ActualRhsType::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(
+      <Index,LhsScalar,RowMajor,LhsBlasTraits::NeedToConjugate,RhsScalar,RhsBlasTraits::NeedToConjugate>::run(
         actualLhs.rows(), actualLhs.cols(),
-        LhsMapper(actualLhs.data(), actualLhs.outerStride()),
-        RhsMapper(actualRhsPtr, 1),
-        dest.data(), dest.col(0).innerStride(), //NOTE  if dest is not a vector at compile-time, then dest.innerStride() might be wrong. (bug 1166)
+        actualLhs.data(), actualLhs.outerStride(),
+        actualRhsPtr, 1,
+        dest.data(), dest.innerStride(),
         actualAlpha);
   }
 };
 
-template<> struct gemv_dense_selector<OnTheRight,ColMajor,false>
+template<> struct gemv_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)
+  template<typename ProductType, typename Dest>
+  static void run(const ProductType& prod, Dest& dest, const typename ProductType::Scalar& alpha)
   {
-    EIGEN_STATIC_ASSERT((!nested_eval<Lhs,1>::Evaluate),EIGEN_INTERNAL_COMPILATION_ERROR_OR_YOU_MADE_A_PROGRAMMING_MISTAKE);
-    // TODO if rhs is large enough it might be beneficial to make sure that dest is sequentially stored in memory, otherwise use a temp
-    typename nested_eval<Rhs,1>::type actual_rhs(rhs);
-    const Index size = rhs.rows();
+    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*actual_rhs.coeff(k)) * lhs.col(k);
+      dest += (alpha*prod.rhs().coeff(k)) * prod.lhs().col(k);
   }
 };
 
-template<> struct gemv_dense_selector<OnTheRight,RowMajor,false>
+template<> struct gemv_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)
+  template<typename ProductType, typename Dest>
+  static void run(const ProductType& prod, Dest& dest, const typename ProductType::Scalar& alpha)
   {
-    EIGEN_STATIC_ASSERT((!nested_eval<Lhs,1>::Evaluate),EIGEN_INTERNAL_COMPILATION_ERROR_OR_YOU_MADE_A_PROGRAMMING_MISTAKE);
-    typename nested_eval<Rhs,Lhs::RowsAtCompileTime>::type actual_rhs(rhs);
-    const Index rows = dest.rows();
+    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 * (lhs.row(i).cwiseProduct(actual_rhs.transpose())).sum();
+      dest.coeffRef(i) += alpha * (prod.lhs().row(i).cwiseProduct(prod.rhs().transpose())).sum();
   }
 };
 
@@ -394,8 +567,7 @@ template<> struct gemv_dense_selector<OnTheRight,RowMajor,false>
   */
 template<typename Derived>
 template<typename OtherDerived>
-EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
-const Product<Derived, 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
@@ -420,8 +592,7 @@ MatrixBase<Derived>::operator*(const MatrixBase<OtherDerived> &other) const
 #ifdef EIGEN_DEBUG_PRODUCT
   internal::product_type<Derived,OtherDerived>::debug();
 #endif
-
-  return Product<Derived, OtherDerived>(derived(), other.derived());
+  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.
@@ -437,8 +608,7 @@ MatrixBase<Derived>::operator*(const MatrixBase<OtherDerived> &other) const
   */
 template<typename Derived>
 template<typename OtherDerived>
-EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
-const Product<Derived,OtherDerived,LazyProduct>
+const typename LazyProductReturnType<Derived,OtherDerived>::Type
 MatrixBase<Derived>::lazyProduct(const MatrixBase<OtherDerived> &other) const
 {
   enum {
@@ -457,7 +627,7 @@ MatrixBase<Derived>::lazyProduct(const MatrixBase<OtherDerived> &other) const
     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());
+  return typename LazyProductReturnType<Derived,OtherDerived>::Type(derived(), other.derived());
 }
 
 } // end namespace Eigen

Eigen/src/Core/GlobalFunctions.h

@@ -1,7 +1,7 @@
 // This file is part of Eigen, a lightweight C++ template library
 // for linear algebra.
 //
-// Copyright (C) 2010-2016 Gael Guennebaud <gael.guennebaud@inria.fr>
+// Copyright (C) 2010-2012 Gael Guennebaud <gael.guennebaud@inria.fr>
 // Copyright (C) 2010 Benoit Jacob <jacob.benoit.1@gmail.com>
 //
 // This Source Code Form is subject to the terms of the Mozilla
@@ -11,30 +11,13 @@
 #ifndef EIGEN_GLOBAL_FUNCTIONS_H
 #define EIGEN_GLOBAL_FUNCTIONS_H
 
-#ifdef EIGEN_PARSED_BY_DOXYGEN
-
-#define EIGEN_ARRAY_DECLARE_GLOBAL_UNARY(NAME,FUNCTOR,DOC_OP,DOC_DETAILS) \
-  /** \returns an expression of the coefficient-wise DOC_OP of \a x
-
-    DOC_DETAILS
-
-    \sa <a href="group__CoeffwiseMathFunctions.html#cwisetable_##NAME">Math functions</a>, class CwiseUnaryOp
-    */ \
+#define EIGEN_ARRAY_DECLARE_GLOBAL_UNARY(NAME,FUNCTOR) \
   template<typename Derived> \
   inline const Eigen::CwiseUnaryOp<Eigen::internal::FUNCTOR<typename Derived::Scalar>, const Derived> \
-  NAME(const Eigen::ArrayBase<Derived>& x);
-
-#else
-
-#define EIGEN_ARRAY_DECLARE_GLOBAL_UNARY(NAME,FUNCTOR,DOC_OP,DOC_DETAILS) \
-  template<typename Derived> \
-  inline const Eigen::CwiseUnaryOp<Eigen::internal::FUNCTOR<typename Derived::Scalar>, const Derived> \
-  (NAME)(const Eigen::ArrayBase<Derived>& x) { \
-    return Eigen::CwiseUnaryOp<Eigen::internal::FUNCTOR<typename Derived::Scalar>, const Derived>(x.derived()); \
+  NAME(const Eigen::ArrayBase<Derived>& x) { \
+    return x.derived(); \
   }
 
-#endif // EIGEN_PARSED_BY_DOXYGEN
-
 #define EIGEN_ARRAY_DECLARE_GLOBAL_EIGEN_UNARY(NAME,FUNCTOR) \
   \
   template<typename Derived> \
@@ -47,139 +30,54 @@
   { \
     static inline typename NAME##_retval<ArrayBase<Derived> >::type run(const Eigen::ArrayBase<Derived>& x) \
     { \
-      return typename NAME##_retval<ArrayBase<Derived> >::type(x.derived()); \
+      return x.derived(); \
     } \
   };
 
+
 namespace Eigen
 {
-  EIGEN_ARRAY_DECLARE_GLOBAL_UNARY(real,scalar_real_op,real part,\sa ArrayBase::real)
-  EIGEN_ARRAY_DECLARE_GLOBAL_UNARY(imag,scalar_imag_op,imaginary part,\sa ArrayBase::imag)
-  EIGEN_ARRAY_DECLARE_GLOBAL_UNARY(conj,scalar_conjugate_op,complex conjugate,\sa ArrayBase::conjugate)
-  EIGEN_ARRAY_DECLARE_GLOBAL_UNARY(inverse,scalar_inverse_op,inverse,\sa ArrayBase::inverse)
-  EIGEN_ARRAY_DECLARE_GLOBAL_UNARY(sin,scalar_sin_op,sine,\sa ArrayBase::sin)
-  EIGEN_ARRAY_DECLARE_GLOBAL_UNARY(cos,scalar_cos_op,cosine,\sa ArrayBase::cos)
-  EIGEN_ARRAY_DECLARE_GLOBAL_UNARY(tan,scalar_tan_op,tangent,\sa ArrayBase::tan)
-  EIGEN_ARRAY_DECLARE_GLOBAL_UNARY(atan,scalar_atan_op,arc-tangent,\sa ArrayBase::atan)
-  EIGEN_ARRAY_DECLARE_GLOBAL_UNARY(asin,scalar_asin_op,arc-sine,\sa ArrayBase::asin)
-  EIGEN_ARRAY_DECLARE_GLOBAL_UNARY(acos,scalar_acos_op,arc-consine,\sa ArrayBase::acos)
-  EIGEN_ARRAY_DECLARE_GLOBAL_UNARY(sinh,scalar_sinh_op,hyperbolic sine,\sa ArrayBase::sinh)
-  EIGEN_ARRAY_DECLARE_GLOBAL_UNARY(cosh,scalar_cosh_op,hyperbolic cosine,\sa ArrayBase::cosh)
-  EIGEN_ARRAY_DECLARE_GLOBAL_UNARY(tanh,scalar_tanh_op,hyperbolic tangent,\sa ArrayBase::tanh)
-#if EIGEN_HAS_CXX11_MATH
-  EIGEN_ARRAY_DECLARE_GLOBAL_UNARY(asinh,scalar_asinh_op,inverse hyperbolic sine,\sa ArrayBase::asinh)
-  EIGEN_ARRAY_DECLARE_GLOBAL_UNARY(acosh,scalar_acosh_op,inverse hyperbolic cosine,\sa ArrayBase::acosh)
-  EIGEN_ARRAY_DECLARE_GLOBAL_UNARY(atanh,scalar_atanh_op,inverse hyperbolic tangent,\sa ArrayBase::atanh)
-#endif
-  EIGEN_ARRAY_DECLARE_GLOBAL_UNARY(logistic,scalar_logistic_op,logistic function,\sa ArrayBase::logistic)
-  EIGEN_ARRAY_DECLARE_GLOBAL_UNARY(lgamma,scalar_lgamma_op,natural logarithm of the gamma function,\sa ArrayBase::lgamma)
-  EIGEN_ARRAY_DECLARE_GLOBAL_UNARY(digamma,scalar_digamma_op,derivative of lgamma,\sa ArrayBase::digamma)
-  EIGEN_ARRAY_DECLARE_GLOBAL_UNARY(erf,scalar_erf_op,error function,\sa ArrayBase::erf)
-  EIGEN_ARRAY_DECLARE_GLOBAL_UNARY(erfc,scalar_erfc_op,complement error function,\sa ArrayBase::erfc)
-  EIGEN_ARRAY_DECLARE_GLOBAL_UNARY(ndtri,scalar_ndtri_op,inverse normal distribution function,\sa ArrayBase::ndtri)
-  EIGEN_ARRAY_DECLARE_GLOBAL_UNARY(exp,scalar_exp_op,exponential,\sa ArrayBase::exp)
-  EIGEN_ARRAY_DECLARE_GLOBAL_UNARY(expm1,scalar_expm1_op,exponential of a value minus 1,\sa ArrayBase::expm1)
-  EIGEN_ARRAY_DECLARE_GLOBAL_UNARY(log,scalar_log_op,natural logarithm,\sa Eigen::log10 DOXCOMMA ArrayBase::log)
-  EIGEN_ARRAY_DECLARE_GLOBAL_UNARY(log1p,scalar_log1p_op,natural logarithm of 1 plus the value,\sa ArrayBase::log1p)
-  EIGEN_ARRAY_DECLARE_GLOBAL_UNARY(log10,scalar_log10_op,base 10 logarithm,\sa Eigen::log DOXCOMMA ArrayBase::log10)
-  EIGEN_ARRAY_DECLARE_GLOBAL_UNARY(log2,scalar_log2_op,base 2 logarithm,\sa Eigen::log DOXCOMMA ArrayBase::log2)
-  EIGEN_ARRAY_DECLARE_GLOBAL_UNARY(abs,scalar_abs_op,absolute value,\sa ArrayBase::abs DOXCOMMA MatrixBase::cwiseAbs)
-  EIGEN_ARRAY_DECLARE_GLOBAL_UNARY(abs2,scalar_abs2_op,squared absolute value,\sa ArrayBase::abs2 DOXCOMMA MatrixBase::cwiseAbs2)
-  EIGEN_ARRAY_DECLARE_GLOBAL_UNARY(arg,scalar_arg_op,complex argument,\sa ArrayBase::arg DOXCOMMA MatrixBase::cwiseArg)
-  EIGEN_ARRAY_DECLARE_GLOBAL_UNARY(sqrt,scalar_sqrt_op,square root,\sa ArrayBase::sqrt DOXCOMMA MatrixBase::cwiseSqrt)
-  EIGEN_ARRAY_DECLARE_GLOBAL_UNARY(rsqrt,scalar_rsqrt_op,reciprocal square root,\sa ArrayBase::rsqrt)
-  EIGEN_ARRAY_DECLARE_GLOBAL_UNARY(square,scalar_square_op,square (power 2),\sa Eigen::abs2 DOXCOMMA Eigen::pow DOXCOMMA ArrayBase::square)
-  EIGEN_ARRAY_DECLARE_GLOBAL_UNARY(cube,scalar_cube_op,cube (power 3),\sa Eigen::pow DOXCOMMA ArrayBase::cube)
-  EIGEN_ARRAY_DECLARE_GLOBAL_UNARY(rint,scalar_rint_op,nearest integer,\sa Eigen::floor DOXCOMMA Eigen::ceil DOXCOMMA ArrayBase::round)
-  EIGEN_ARRAY_DECLARE_GLOBAL_UNARY(round,scalar_round_op,nearest integer,\sa Eigen::floor DOXCOMMA Eigen::ceil DOXCOMMA ArrayBase::round)
-  EIGEN_ARRAY_DECLARE_GLOBAL_UNARY(floor,scalar_floor_op,nearest integer not greater than the giben value,\sa Eigen::ceil DOXCOMMA ArrayBase::floor)
-  EIGEN_ARRAY_DECLARE_GLOBAL_UNARY(ceil,scalar_ceil_op,nearest integer not less than the giben value,\sa Eigen::floor DOXCOMMA ArrayBase::ceil)
-  EIGEN_ARRAY_DECLARE_GLOBAL_UNARY(isnan,scalar_isnan_op,not-a-number test,\sa Eigen::isinf DOXCOMMA Eigen::isfinite DOXCOMMA ArrayBase::isnan)
-  EIGEN_ARRAY_DECLARE_GLOBAL_UNARY(isinf,scalar_isinf_op,infinite value test,\sa Eigen::isnan DOXCOMMA Eigen::isfinite DOXCOMMA ArrayBase::isinf)
-  EIGEN_ARRAY_DECLARE_GLOBAL_UNARY(isfinite,scalar_isfinite_op,finite value test,\sa Eigen::isinf DOXCOMMA Eigen::isnan DOXCOMMA ArrayBase::isfinite)
-  EIGEN_ARRAY_DECLARE_GLOBAL_UNARY(sign,scalar_sign_op,sign (or 0),\sa ArrayBase::sign)
-
-  /** \returns an expression of the coefficient-wise power of \a x to the given constant \a exponent.
-    *
-    * \tparam ScalarExponent is the scalar type of \a exponent. It must be compatible with the scalar type of the given expression (\c Derived::Scalar).
-    *
-    * \sa ArrayBase::pow()
-    *
-    * \relates ArrayBase
-    */
-#ifdef EIGEN_PARSED_BY_DOXYGEN
-  template<typename Derived,typename ScalarExponent>
-  inline const CwiseBinaryOp<internal::scalar_pow_op<Derived::Scalar,ScalarExponent>,Derived,Constant<ScalarExponent> >
-  pow(const Eigen::ArrayBase<Derived>& x, const ScalarExponent& exponent);
-#else
-  template <typename Derived,typename ScalarExponent>
-  EIGEN_DEVICE_FUNC inline
-  EIGEN_MSVC10_WORKAROUND_BINARYOP_RETURN_TYPE(
-    const EIGEN_EXPR_BINARYOP_SCALAR_RETURN_TYPE(Derived,typename internal::promote_scalar_arg<typename Derived::Scalar
-                                                 EIGEN_COMMA ScalarExponent EIGEN_COMMA
-                                                 EIGEN_SCALAR_BINARY_SUPPORTED(pow,typename Derived::Scalar,ScalarExponent)>::type,pow))
-  pow(const Eigen::ArrayBase<Derived>& x, const ScalarExponent& exponent)
-  {
-    typedef typename internal::promote_scalar_arg<typename Derived::Scalar,ScalarExponent,
-                                                  EIGEN_SCALAR_BINARY_SUPPORTED(pow,typename Derived::Scalar,ScalarExponent)>::type PromotedExponent;
-    return EIGEN_EXPR_BINARYOP_SCALAR_RETURN_TYPE(Derived,PromotedExponent,pow)(x.derived(),
-           typename internal::plain_constant_type<Derived,PromotedExponent>::type(x.derived().rows(), x.derived().cols(), internal::scalar_constant_op<PromotedExponent>(exponent)));
+  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(sin,scalar_sin_op)
+  EIGEN_ARRAY_DECLARE_GLOBAL_UNARY(cos,scalar_cos_op)
+  EIGEN_ARRAY_DECLARE_GLOBAL_UNARY(asin,scalar_asin_op)
+  EIGEN_ARRAY_DECLARE_GLOBAL_UNARY(acos,scalar_acos_op)
+  EIGEN_ARRAY_DECLARE_GLOBAL_UNARY(tan,scalar_tan_op)
+  EIGEN_ARRAY_DECLARE_GLOBAL_UNARY(exp,scalar_exp_op)
+  EIGEN_ARRAY_DECLARE_GLOBAL_UNARY(log,scalar_log_op)
+  EIGEN_ARRAY_DECLARE_GLOBAL_UNARY(abs,scalar_abs_op)
+  EIGEN_ARRAY_DECLARE_GLOBAL_UNARY(sqrt,scalar_sqrt_op)
+  
+  template<typename Derived>
+  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);
   }
-#endif
 
-  /** \returns an expression of the coefficient-wise power of \a x to the given array of \a exponents.
-    *
-    * This function computes the coefficient-wise power.
-    *
-    * Example: \include Cwise_array_power_array.cpp
-    * Output: \verbinclude Cwise_array_power_array.out
-    *
-    * \sa ArrayBase::pow()
-    *
-    * \relates ArrayBase
-    */
-  template<typename Derived,typename ExponentDerived>
-  inline const Eigen::CwiseBinaryOp<Eigen::internal::scalar_pow_op<typename Derived::Scalar, typename ExponentDerived::Scalar>, const Derived, const ExponentDerived>
-  pow(const Eigen::ArrayBase<Derived>& x, const Eigen::ArrayBase<ExponentDerived>& exponents)
+  template<typename Derived>
+  inline const Eigen::CwiseBinaryOp<Eigen::internal::scalar_binary_pow_op<typename Derived::Scalar, typename Derived::Scalar>, const Derived, const Derived>
+  pow(const Eigen::ArrayBase<Derived>& x, const Eigen::ArrayBase<Derived>& exponents) 
   {
-    return Eigen::CwiseBinaryOp<Eigen::internal::scalar_pow_op<typename Derived::Scalar, typename ExponentDerived::Scalar>, const Derived, const ExponentDerived>(
+    return Eigen::CwiseBinaryOp<Eigen::internal::scalar_binary_pow_op<typename Derived::Scalar, typename Derived::Scalar>, const Derived, const Derived>(
       x.derived(),
       exponents.derived()
     );
   }
-
-  /** \returns an expression of the coefficient-wise power of the scalar \a x to the given array of \a exponents.
-    *
-    * This function computes the coefficient-wise power between a scalar and an array of exponents.
-    *
-    * \tparam Scalar is the scalar type of \a x. It must be compatible with the scalar type of the given array expression (\c Derived::Scalar).
-    *
-    * Example: \include Cwise_scalar_power_array.cpp
-    * Output: \verbinclude Cwise_scalar_power_array.out
-    *
-    * \sa ArrayBase::pow()
-    *
-    * \relates ArrayBase
-    */
-#ifdef EIGEN_PARSED_BY_DOXYGEN
-  template<typename Scalar,typename Derived>
-  inline const CwiseBinaryOp<internal::scalar_pow_op<Scalar,Derived::Scalar>,Constant<Scalar>,Derived>
-  pow(const Scalar& x,const Eigen::ArrayBase<Derived>& x);
-#else
-  template <typename Scalar, typename Derived>
-  EIGEN_DEVICE_FUNC inline
-  EIGEN_MSVC10_WORKAROUND_BINARYOP_RETURN_TYPE(
-    const EIGEN_SCALAR_BINARYOP_EXPR_RETURN_TYPE(typename internal::promote_scalar_arg<typename Derived::Scalar
-                                                 EIGEN_COMMA Scalar EIGEN_COMMA
-                                                 EIGEN_SCALAR_BINARY_SUPPORTED(pow,Scalar,typename Derived::Scalar)>::type,Derived,pow))
-  pow(const Scalar& x, const Eigen::ArrayBase<Derived>& exponents) {
-    typedef typename internal::promote_scalar_arg<typename Derived::Scalar,Scalar,
-                                                  EIGEN_SCALAR_BINARY_SUPPORTED(pow,Scalar,typename Derived::Scalar)>::type PromotedScalar;
-    return EIGEN_SCALAR_BINARYOP_EXPR_RETURN_TYPE(PromotedScalar,Derived,pow)(
-           typename internal::plain_constant_type<Derived,PromotedScalar>::type(exponents.derived().rows(), exponents.derived().cols(), internal::scalar_constant_op<PromotedScalar>(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)  
+    );
   }
-#endif
-
 
   namespace internal
   {

Eigen/src/Core/IO.h

@@ -41,7 +41,6 @@ std::ostream & print_matrix(std::ostream & s, const Derived& _m, const IOFormat&
   *  - \b rowSuffix string printed at the end of each row
   *  - \b matPrefix string printed at the beginning of the matrix
   *  - \b matSuffix string printed at the end of the matrix
-  *  - \b fill character printed to fill the empty space in aligned columns
   *
   * Example: \include IOFormat.cpp
   * Output: \verbinclude IOFormat.out
@@ -50,18 +49,14 @@ std::ostream & print_matrix(std::ostream & s, const Derived& _m, const IOFormat&
   */
 struct IOFormat
 {
-  /** Default constructor, see class IOFormat for the meaning of the parameters */
+  /** Default contructor, 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="", const char _fill=' ')
+    const std::string& _matPrefix="", const std::string& _matSuffix="")
   : matPrefix(_matPrefix), matSuffix(_matSuffix), rowPrefix(_rowPrefix), rowSuffix(_rowSuffix), rowSeparator(_rowSeparator),
-    rowSpacer(""), coeffSeparator(_coeffSeparator), fill(_fill), precision(_precision), flags(_flags)
+    rowSpacer(""), coeffSeparator(_coeffSeparator), precision(_precision), flags(_flags)
   {
-    // 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')
     {
@@ -72,7 +67,6 @@ struct IOFormat
   std::string matPrefix, matSuffix;
   std::string rowPrefix, rowSuffix, rowSeparator, rowSpacer;
   std::string coeffSeparator;
-  char fill;
   int precision;
   int flags;
 };
@@ -82,7 +76,7 @@ struct IOFormat
   *
   * \brief Pseudo expression providing matrix output with given format
   *
-  * \tparam ExpressionType the type of the object on which IO stream operations are performed
+  * \param ExpressionType the type of the object on which IO stream operations are performed
   *
   * This class represents an expression with stream operators controlled by a given IOFormat.
   * It is the return type of DenseBase::format()
@@ -107,32 +101,57 @@ class WithFormat
     }
 
   protected:
-    typename ExpressionType::Nested m_matrix;
+    const typename ExpressionType::Nested m_matrix;
     IOFormat m_format;
 };
 
+/** \returns a WithFormat proxy object allowing to print a matrix the with given
+  * format \a fmt.
+  *
+  * See class IOFormat for some examples.
+  *
+  * \sa class IOFormat, class WithFormat
+  */
+template<typename Derived>
+inline const WithFormat<Derived>
+DenseBase<Derived>::format(const IOFormat& fmt) const
+{
+  return WithFormat<Derived>(derived(), fmt);
+}
+
 namespace internal {
 
-// NOTE: This helper is kept for backward compatibility with previous code specializing
-//       this internal::significant_decimals_impl structure. In the future we should directly
-//       call digits10() which has been introduced in July 2016 in 3.3.
+template<typename Scalar, bool IsInteger>
+struct significant_decimals_default_impl
+{
+  typedef typename NumTraits<Scalar>::Real RealScalar;
+  static inline int run()
+  {
+    using std::ceil;
+    using std::log;
+    return cast<RealScalar,int>(ceil(-log(NumTraits<RealScalar>::epsilon())/log(RealScalar(10))));
+  }
+};
+
 template<typename Scalar>
-struct significant_decimals_impl
+struct significant_decimals_default_impl<Scalar, true>
 {
   static inline int run()
   {
-    return NumTraits<Scalar>::digits10();
+    return 0;
   }
 };
 
+template<typename Scalar>
+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 & print_matrix(std::ostream & s, const Derived& _m, const IOFormat& fmt)
 {
-  using internal::is_same;
-  using internal::conditional;
-
   if(_m.size() == 0)
   {
     s << fmt.matPrefix << fmt.matSuffix;
@@ -141,22 +160,7 @@ std::ostream & print_matrix(std::ostream & s, const Derived& _m, const IOFormat&
   
   typename Derived::Nested m = _m;
   typedef typename Derived::Scalar Scalar;
-  typedef typename
-      conditional<
-          is_same<Scalar, char>::value ||
-            is_same<Scalar, unsigned char>::value ||
-            is_same<Scalar, numext::int8_t>::value ||
-            is_same<Scalar, numext::uint8_t>::value,
-          int,
-          typename conditional<
-              is_same<Scalar, std::complex<char> >::value ||
-                is_same<Scalar, std::complex<unsigned char> >::value ||
-                is_same<Scalar, std::complex<numext::int8_t> >::value ||
-                is_same<Scalar, std::complex<numext::uint8_t> >::value,
-              std::complex<int>,
-              const Scalar&
-            >::type
-        >::type PrintType;
+  typedef typename Derived::Index Index;
 
   Index width = 0;
 
@@ -181,43 +185,34 @@ std::ostream & print_matrix(std::ostream & s, const Derived& _m, const IOFormat&
     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 = 0; j < m.cols(); ++j)
+    for(Index j = 1; j < m.cols(); ++j)
       for(Index i = 0; i < m.rows(); ++i)
       {
         std::stringstream sstr;
-        sstr.copyfmt(s);
-        sstr << static_cast<PrintType>(m.coeff(i,j));
+        if(explicit_precision) sstr.precision(explicit_precision);
+        sstr << m.coeff(i,j);
         width = std::max<Index>(width, Index(sstr.str().length()));
       }
   }
-  std::streamsize old_width = s.width();
-  char old_fill_character = s.fill();
+  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)
   {
     if (i)
       s << fmt.rowSpacer;
     s << fmt.rowPrefix;
-    if(width) {
-      s.fill(fmt.fill);
-      s.width(width);
-    }
-    s << static_cast<PrintType>(m.coeff(i, 0));
+    if(width) s.width(width);
+    s << m.coeff(i, 0);
     for(Index j = 1; j < m.cols(); ++j)
     {
       s << fmt.coeffSeparator;
-      if(width) {
-        s.fill(fmt.fill);
-        s.width(width);
-      }
-      s << static_cast<PrintType>(m.coeff(i, j));
+      if (width) s.width(width);
+      s << m.coeff(i, j);
     }
     s << fmt.rowSuffix;
     if( i < m.rows() - 1)
@@ -225,10 +220,6 @@ std::ostream & print_matrix(std::ostream & s, const Derived& _m, const IOFormat&
   }
   s << fmt.matSuffix;
   if(explicit_precision) s.precision(old_precision);
-  if(width) {
-    s.fill(old_fill_character);
-    s.width(old_width);
-  }
   return s;
 }
 

Eigen/src/Core/IndexedView.h

@@ -1,237 +0,0 @@
-// This file is part of Eigen, a lightweight C++ template library
-// for linear algebra.
-//
-// Copyright (C) 2017 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_INDEXED_VIEW_H
-#define EIGEN_INDEXED_VIEW_H
-
-namespace Eigen {
-
-namespace internal {
-
-template<typename XprType, typename RowIndices, typename ColIndices>
-struct traits<IndexedView<XprType, RowIndices, ColIndices> >
- : traits<XprType>
-{
-  enum {
-    RowsAtCompileTime = int(array_size<RowIndices>::value),
-    ColsAtCompileTime = int(array_size<ColIndices>::value),
-    MaxRowsAtCompileTime = RowsAtCompileTime != Dynamic ? int(RowsAtCompileTime) : Dynamic,
-    MaxColsAtCompileTime = ColsAtCompileTime != Dynamic ? int(ColsAtCompileTime) : Dynamic,
-
-    XprTypeIsRowMajor = (int(traits<XprType>::Flags)&RowMajorBit) != 0,
-    IsRowMajor = (MaxRowsAtCompileTime==1&&MaxColsAtCompileTime!=1) ? 1
-               : (MaxColsAtCompileTime==1&&MaxRowsAtCompileTime!=1) ? 0
-               : XprTypeIsRowMajor,
-
-    RowIncr = int(get_compile_time_incr<RowIndices>::value),
-    ColIncr = int(get_compile_time_incr<ColIndices>::value),
-    InnerIncr = IsRowMajor ? ColIncr : RowIncr,
-    OuterIncr = IsRowMajor ? RowIncr : ColIncr,
-
-    HasSameStorageOrderAsXprType = (IsRowMajor == XprTypeIsRowMajor),
-    XprInnerStride = HasSameStorageOrderAsXprType ? int(inner_stride_at_compile_time<XprType>::ret) : int(outer_stride_at_compile_time<XprType>::ret),
-    XprOuterstride = HasSameStorageOrderAsXprType ? int(outer_stride_at_compile_time<XprType>::ret) : int(inner_stride_at_compile_time<XprType>::ret),
-
-    InnerSize = XprTypeIsRowMajor ? ColsAtCompileTime : RowsAtCompileTime,
-    IsBlockAlike = InnerIncr==1 && OuterIncr==1,
-    IsInnerPannel = HasSameStorageOrderAsXprType && is_same<AllRange<InnerSize>,typename conditional<XprTypeIsRowMajor,ColIndices,RowIndices>::type>::value,
-
-    InnerStrideAtCompileTime = InnerIncr<0 || InnerIncr==DynamicIndex || XprInnerStride==Dynamic ? Dynamic : XprInnerStride * InnerIncr,
-    OuterStrideAtCompileTime = OuterIncr<0 || OuterIncr==DynamicIndex || XprOuterstride==Dynamic ? Dynamic : XprOuterstride * OuterIncr,
-
-    ReturnAsScalar = is_same<RowIndices,SingleRange>::value && is_same<ColIndices,SingleRange>::value,
-    ReturnAsBlock = (!ReturnAsScalar) && IsBlockAlike,
-    ReturnAsIndexedView = (!ReturnAsScalar) && (!ReturnAsBlock),
-
-    // FIXME we deal with compile-time strides if and only if we have DirectAccessBit flag,
-    // but this is too strict regarding negative strides...
-    DirectAccessMask = (int(InnerIncr)!=UndefinedIncr && int(OuterIncr)!=UndefinedIncr && InnerIncr>=0 && OuterIncr>=0) ? DirectAccessBit : 0,
-    FlagsRowMajorBit = IsRowMajor ? RowMajorBit : 0,
-    FlagsLvalueBit = is_lvalue<XprType>::value ? LvalueBit : 0,
-    FlagsLinearAccessBit = (RowsAtCompileTime == 1 || ColsAtCompileTime == 1) ? LinearAccessBit : 0,
-    Flags = (traits<XprType>::Flags & (HereditaryBits | DirectAccessMask )) | FlagsLvalueBit | FlagsRowMajorBit | FlagsLinearAccessBit
-  };
-
-  typedef Block<XprType,RowsAtCompileTime,ColsAtCompileTime,IsInnerPannel> BlockType;
-};
-
-}
-
-template<typename XprType, typename RowIndices, typename ColIndices, typename StorageKind>
-class IndexedViewImpl;
-
-
-/** \class IndexedView
-  * \ingroup Core_Module
-  *
-  * \brief Expression of a non-sequential sub-matrix defined by arbitrary sequences of row and column indices
-  *
-  * \tparam XprType the type of the expression in which we are taking the intersections of sub-rows and sub-columns
-  * \tparam RowIndices the type of the object defining the sequence of row indices
-  * \tparam ColIndices the type of the object defining the sequence of column indices
-  *
-  * This class represents an expression of a sub-matrix (or sub-vector) defined as the intersection
-  * of sub-sets of rows and columns, that are themself defined by generic sequences of row indices \f$ \{r_0,r_1,..r_{m-1}\} \f$
-  * and column indices \f$ \{c_0,c_1,..c_{n-1} \}\f$. Let \f$ A \f$  be the nested matrix, then the resulting matrix \f$ B \f$ has \c m
-  * rows and \c n columns, and its entries are given by: \f$ B(i,j) = A(r_i,c_j) \f$.
-  *
-  * The \c RowIndices and \c ColIndices types must be compatible with the following API:
-  * \code
-  * <integral type> operator[](Index) const;
-  * Index size() const;
-  * \endcode
-  *
-  * Typical supported types thus include:
-  *  - std::vector<int>
-  *  - std::valarray<int>
-  *  - std::array<int>
-  *  - Plain C arrays: int[N]
-  *  - Eigen::ArrayXi
-  *  - decltype(ArrayXi::LinSpaced(...))
-  *  - Any view/expressions of the previous types
-  *  - Eigen::ArithmeticSequence
-  *  - Eigen::internal::AllRange      (helper for Eigen::all)
-  *  - Eigen::internal::SingleRange  (helper for single index)
-  *  - etc.
-  *
-  * In typical usages of %Eigen, this class should never be used directly. It is the return type of
-  * DenseBase::operator()(const RowIndices&, const ColIndices&).
-  *
-  * \sa class Block
-  */
-template<typename XprType, typename RowIndices, typename ColIndices>
-class IndexedView : public IndexedViewImpl<XprType, RowIndices, ColIndices, typename internal::traits<XprType>::StorageKind>
-{
-public:
-  typedef typename IndexedViewImpl<XprType, RowIndices, ColIndices, typename internal::traits<XprType>::StorageKind>::Base Base;
-  EIGEN_GENERIC_PUBLIC_INTERFACE(IndexedView)
-  EIGEN_INHERIT_ASSIGNMENT_OPERATORS(IndexedView)
-
-  typedef typename internal::ref_selector<XprType>::non_const_type MatrixTypeNested;
-  typedef typename internal::remove_all<XprType>::type NestedExpression;
-
-  template<typename T0, typename T1>
-  IndexedView(XprType& xpr, const T0& rowIndices, const T1& colIndices)
-    : m_xpr(xpr), m_rowIndices(rowIndices), m_colIndices(colIndices)
-  {}
-
-  /** \returns number of rows */
-  Index rows() const { return internal::size(m_rowIndices); }
-
-  /** \returns number of columns */
-  Index cols() const { return internal::size(m_colIndices); }
-
-  /** \returns the nested expression */
-  const typename internal::remove_all<XprType>::type&
-  nestedExpression() const { return m_xpr; }
-
-  /** \returns the nested expression */
-  typename internal::remove_reference<XprType>::type&
-  nestedExpression() { return m_xpr; }
-
-  /** \returns a const reference to the object storing/generating the row indices */
-  const RowIndices& rowIndices() const { return m_rowIndices; }
-
-  /** \returns a const reference to the object storing/generating the column indices */
-  const ColIndices& colIndices() const { return m_colIndices; }
-
-protected:
-  MatrixTypeNested m_xpr;
-  RowIndices m_rowIndices;
-  ColIndices m_colIndices;
-};
-
-
-// Generic API dispatcher
-template<typename XprType, typename RowIndices, typename ColIndices, typename StorageKind>
-class IndexedViewImpl
-  : public internal::generic_xpr_base<IndexedView<XprType, RowIndices, ColIndices> >::type
-{
-public:
-  typedef typename internal::generic_xpr_base<IndexedView<XprType, RowIndices, ColIndices> >::type Base;
-};
-
-namespace internal {
-
-
-template<typename ArgType, typename RowIndices, typename ColIndices>
-struct unary_evaluator<IndexedView<ArgType, RowIndices, ColIndices>, IndexBased>
-  : evaluator_base<IndexedView<ArgType, RowIndices, ColIndices> >
-{
-  typedef IndexedView<ArgType, RowIndices, ColIndices> XprType;
-
-  enum {
-    CoeffReadCost = evaluator<ArgType>::CoeffReadCost /* TODO + cost of row/col index */,
-
-    FlagsLinearAccessBit = (traits<XprType>::RowsAtCompileTime == 1 || traits<XprType>::ColsAtCompileTime == 1) ? LinearAccessBit : 0,
-
-    FlagsRowMajorBit = traits<XprType>::FlagsRowMajorBit, 
-
-    Flags = (evaluator<ArgType>::Flags & (HereditaryBits & ~RowMajorBit /*| LinearAccessBit | DirectAccessBit*/)) | FlagsLinearAccessBit | FlagsRowMajorBit,
-
-    Alignment = 0
-  };
-
-  EIGEN_DEVICE_FUNC explicit unary_evaluator(const XprType& xpr) : m_argImpl(xpr.nestedExpression()), m_xpr(xpr)
-  {
-    EIGEN_INTERNAL_CHECK_COST_VALUE(CoeffReadCost);
-  }
-
-  typedef typename XprType::Scalar Scalar;
-  typedef typename XprType::CoeffReturnType CoeffReturnType;
-
-  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
-  CoeffReturnType coeff(Index row, Index col) const
-  {
-    return m_argImpl.coeff(m_xpr.rowIndices()[row], m_xpr.colIndices()[col]);
-  }
-
-  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
-  Scalar& coeffRef(Index row, Index col)
-  {
-    return m_argImpl.coeffRef(m_xpr.rowIndices()[row], m_xpr.colIndices()[col]);
-  }
-
-  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
-  Scalar& coeffRef(Index index)
-  {
-    EIGEN_STATIC_ASSERT_LVALUE(XprType)
-    Index row = XprType::RowsAtCompileTime == 1 ? 0 : index;
-    Index col = XprType::RowsAtCompileTime == 1 ? index : 0;
-    return m_argImpl.coeffRef( m_xpr.rowIndices()[row], m_xpr.colIndices()[col]);
-  }
-
-  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
-  const Scalar& coeffRef(Index index) const
-  {
-    Index row = XprType::RowsAtCompileTime == 1 ? 0 : index;
-    Index col = XprType::RowsAtCompileTime == 1 ? index : 0;
-    return m_argImpl.coeffRef( m_xpr.rowIndices()[row], m_xpr.colIndices()[col]);
-  }
-
-  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
-  const CoeffReturnType coeff(Index index) const
-  {
-    Index row = XprType::RowsAtCompileTime == 1 ? 0 : index;
-    Index col = XprType::RowsAtCompileTime == 1 ? index : 0;
-    return m_argImpl.coeff( m_xpr.rowIndices()[row], m_xpr.colIndices()[col]);
-  }
-
-protected:
-
-  evaluator<ArgType> m_argImpl;
-  const XprType& m_xpr;
-
-};
-
-} // end namespace internal
-
-} // end namespace Eigen
-
-#endif // EIGEN_INDEXED_VIEW_H

Eigen/src/Core/Inverse.h

@@ -1,117 +0,0 @@
-// This file is part of Eigen, a lightweight C++ template library
-// for linear algebra.
-//
-// Copyright (C) 2014-2019 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::Scalar                            Scalar;
-  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 EIGEN_DEVICE_FUNC Inverse(const XprType &xpr)
-    : m_xpr(xpr)
-  {}
-
-  EIGEN_DEVICE_FUNC EIGEN_CONSTEXPR  Index rows() const EIGEN_NOEXCEPT { return m_xpr.cols(); }
-  EIGEN_DEVICE_FUNC EIGEN_CONSTEXPR  Index cols() const EIGEN_NOEXCEPT { return m_xpr.rows(); }
-
-  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

@@ -11,35 +11,7 @@
 #ifndef EIGEN_MAP_H
 #define EIGEN_MAP_H
 
-namespace Eigen {
-
-namespace internal {
-template<typename PlainObjectType, int MapOptions, typename StrideType>
-struct traits<Map<PlainObjectType, MapOptions, StrideType> >
-  : public traits<PlainObjectType>
-{
-  typedef traits<PlainObjectType> TraitsBase;
-  enum {
-    PlainObjectTypeInnerSize = ((traits<PlainObjectType>::Flags&RowMajorBit)==RowMajorBit)
-                             ? PlainObjectType::ColsAtCompileTime
-                             : PlainObjectType::RowsAtCompileTime,
-
-    InnerStrideAtCompileTime = StrideType::InnerStrideAtCompileTime == 0
-                             ? int(PlainObjectType::InnerStrideAtCompileTime)
-                             : int(StrideType::InnerStrideAtCompileTime),
-    OuterStrideAtCompileTime = StrideType::OuterStrideAtCompileTime == 0
-                             ? (InnerStrideAtCompileTime==Dynamic || PlainObjectTypeInnerSize==Dynamic
-                                ? Dynamic
-                                : int(InnerStrideAtCompileTime) * int(PlainObjectTypeInnerSize))
-                             : int(StrideType::OuterStrideAtCompileTime),
-    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 have Options
-};
-}
+namespace Eigen { 
 
 /** \class Map
   * \ingroup Core_Module
@@ -47,7 +19,7 @@ private:
   * \brief A matrix or vector expression mapping an existing array of data.
   *
   * \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.
+  * \tparam MapOptions specifies whether the pointer is \c #Aligned, or \c #Unaligned.
   *                The default is \c #Unaligned.
   * \tparam StrideType optionally specifies strides. By default, Map assumes the memory layout
   *                   of an ordinary, contiguous array. This can be overridden by specifying strides.
@@ -91,6 +63,44 @@ private:
   *
   * \sa PlainObjectBase::Map(), \ref TopicStorageOrders
   */
+
+namespace internal {
+template<typename PlainObjectType, int MapOptions, typename StrideType>
+struct traits<Map<PlainObjectType, MapOptions, StrideType> >
+  : public traits<PlainObjectType>
+{
+  typedef traits<PlainObjectType> TraitsBase;
+  typedef typename PlainObjectType::Index Index;
+  typedef typename PlainObjectType::Scalar Scalar;
+  enum {
+    InnerStrideAtCompileTime = StrideType::InnerStrideAtCompileTime == 0
+                             ? int(PlainObjectType::InnerStrideAtCompileTime)
+                             : int(StrideType::InnerStrideAtCompileTime),
+    OuterStrideAtCompileTime = StrideType::OuterStrideAtCompileTime == 0
+                             ? int(PlainObjectType::OuterStrideAtCompileTime)
+                             : int(StrideType::OuterStrideAtCompileTime),
+    HasNoInnerStride = InnerStrideAtCompileTime == 1,
+    HasNoOuterStride = StrideType::OuterStrideAtCompileTime == 0,
+    HasNoStride = HasNoInnerStride && HasNoOuterStride,
+    IsAligned = bool(EIGEN_ALIGN) && ((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 = TraitsBase::Flags & (~NestByRefBit),
+    Flags1 = IsAligned ? (int(Flags0) | AlignedBit) : (int(Flags0) & ~AlignedBit),
+    Flags2 = (bool(HasNoStride) || bool(PlainObjectType::IsVectorAtCompileTime))
+           ? int(Flags1) : int(Flags1 & ~LinearAccessBit),
+    Flags3 = is_lvalue<PlainObjectType>::value ? int(Flags2) : (int(Flags2) & ~LvalueBit),
+    Flags = KeepsPacketAccess ? int(Flags3) : (int(Flags3) & ~PacketAccessBit)
+  };
+private:
+  enum { Options }; // Expressions don't have Options
+};
+}
+
 template<typename PlainObjectType, int MapOptions, typename StrideType> class Map
   : public MapBase<Map<PlainObjectType, MapOptions, StrideType> >
 {
@@ -100,34 +110,34 @@ template<typename PlainObjectType, int MapOptions, typename StrideType> class Ma
     EIGEN_DENSE_PUBLIC_INTERFACE(Map)
 
     typedef typename Base::PointerType PointerType;
+#if EIGEN2_SUPPORT_STAGE <= STAGE30_FULL_EIGEN3_API
+    typedef const Scalar* PointerArgType;
+    inline PointerType cast_to_pointer_type(PointerArgType ptr) { return const_cast<PointerType>(ptr); }
+#else
     typedef PointerType PointerArgType;
-    EIGEN_DEVICE_FUNC
     inline PointerType cast_to_pointer_type(PointerArgType ptr) { return ptr; }
+#endif
 
-    EIGEN_DEVICE_FUNC EIGEN_CONSTEXPR
     inline Index innerStride() const
     {
       return StrideType::InnerStrideAtCompileTime != 0 ? m_stride.inner() : 1;
     }
 
-    EIGEN_DEVICE_FUNC EIGEN_CONSTEXPR
     inline Index outerStride() const
     {
       return StrideType::OuterStrideAtCompileTime != 0 ? m_stride.outer()
-           : internal::traits<Map>::OuterStrideAtCompileTime != Dynamic ? Index(internal::traits<Map>::OuterStrideAtCompileTime)
-           : IsVectorAtCompileTime ? (this->size() * innerStride())
-           : int(Flags)&RowMajorBit ? (this->cols() * innerStride())
-           : (this->rows() * innerStride());
+           : IsVectorAtCompileTime ? this->size()
+           : int(Flags)&RowMajorBit ? this->cols()
+           : this->rows();
     }
 
     /** Constructor in the fixed-size case.
       *
       * \param dataPtr pointer to the array to map
-      * \param stride optional Stride object, passing the strides.
+      * \param a_stride optional Stride object, passing the strides.
       */
-    EIGEN_DEVICE_FUNC
-    explicit inline Map(PointerArgType dataPtr, const StrideType& stride = StrideType())
-      : Base(cast_to_pointer_type(dataPtr)), m_stride(stride)
+    inline Map(PointerArgType dataPtr, const StrideType& a_stride = StrideType())
+      : Base(cast_to_pointer_type(dataPtr)), m_stride(a_stride)
     {
       PlainObjectType::Base::_check_template_params();
     }
@@ -135,12 +145,11 @@ template<typename PlainObjectType, int MapOptions, typename StrideType> class Ma
     /** Constructor in the dynamic-size vector case.
       *
       * \param dataPtr pointer to the array to map
-      * \param size the size of the vector expression
-      * \param stride optional Stride object, passing the strides.
+      * \param a_size the size of the vector expression
+      * \param a_stride optional Stride object, passing the strides.
       */
-    EIGEN_DEVICE_FUNC
-    inline Map(PointerArgType dataPtr, Index size, const StrideType& stride = StrideType())
-      : Base(cast_to_pointer_type(dataPtr), size), m_stride(stride)
+    inline Map(PointerArgType dataPtr, Index a_size, const StrideType& a_stride = StrideType())
+      : Base(cast_to_pointer_type(dataPtr), a_size), m_stride(a_stride)
     {
       PlainObjectType::Base::_check_template_params();
     }
@@ -148,13 +157,12 @@ template<typename PlainObjectType, int MapOptions, typename StrideType> class Ma
     /** Constructor in the dynamic-size matrix case.
       *
       * \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.
+      * \param nbRows the number of rows of the matrix expression
+      * \param nbCols the number of columns of the matrix expression
+      * \param a_stride optional Stride object, passing the strides.
       */
-    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)
+    inline Map(PointerArgType dataPtr, Index nbRows, Index nbCols, const StrideType& a_stride = StrideType())
+      : Base(cast_to_pointer_type(dataPtr), nbRows, nbCols), m_stride(a_stride)
     {
       PlainObjectType::Base::_check_template_params();
     }
@@ -165,6 +173,19 @@ template<typename PlainObjectType, int MapOptions, typename StrideType> class Ma
     StrideType m_stride;
 };
 
+template<typename _Scalar, int _Rows, int _Cols, int _Options, int _MaxRows, int _MaxCols>
+inline Array<_Scalar, _Rows, _Cols, _Options, _MaxRows, _MaxCols>
+  ::Array(const Scalar *data)
+{
+  this->_set_noalias(Eigen::Map<const Array>(data));
+}
+
+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)
+{
+  this->_set_noalias(Eigen::Map<const Matrix>(data));
+}
 
 } // end namespace Eigen
 

Eigen/src/Core/MapBase.h

@@ -12,25 +12,15 @@
 #define EIGEN_MAPBASE_H
 
 #define EIGEN_STATIC_ASSERT_INDEX_BASED_ACCESS(Derived) \
-      EIGEN_STATIC_ASSERT((int(internal::evaluator<Derived>::Flags) & LinearAccessBit) || Derived::IsVectorAtCompileTime, \
+      EIGEN_STATIC_ASSERT((int(internal::traits<Derived>::Flags) & LinearAccessBit) || Derived::IsVectorAtCompileTime, \
                           YOU_ARE_TRYING_TO_USE_AN_INDEX_BASED_ACCESSOR_ON_AN_EXPRESSION_THAT_DOES_NOT_SUPPORT_THAT)
 
-namespace Eigen {
+namespace Eigen { 
 
-/** \ingroup Core_Module
+/** \class MapBase
+  * \ingroup Core_Module
   *
-  * \brief Base class for dense Map and Block expression with direct access
-  *
-  * This base class provides the const low-level accessors (e.g. coeff, coeffRef) of dense
-  * Map and Block objects with direct access.
-  * Typical users do not have to directly deal with this class.
-  *
-  * This class can be extended by through the macro plugin \c EIGEN_MAPBASE_PLUGIN.
-  * See \link TopicCustomizing_Plugins customizing Eigen \endlink for details.
-  *
-  * The \c Derived class has to provide the following two methods describing the memory layout:
-  *  \code Index innerStride() const; \endcode
-  *  \code Index outerStride() const; \endcode
+  * \brief Base class for Map and Block expression with direct access
   *
   * \sa class Map, class Block
   */
@@ -43,11 +33,11 @@ template<typename Derived> class MapBase<Derived, ReadOnlyAccessors>
     enum {
       RowsAtCompileTime = internal::traits<Derived>::RowsAtCompileTime,
       ColsAtCompileTime = internal::traits<Derived>::ColsAtCompileTime,
-      InnerStrideAtCompileTime = internal::traits<Derived>::InnerStrideAtCompileTime,
       SizeAtCompileTime = Base::SizeAtCompileTime
     };
 
     typedef typename internal::traits<Derived>::StorageKind StorageKind;
+    typedef typename internal::traits<Derived>::Index Index;
     typedef typename internal::traits<Derived>::Scalar Scalar;
     typedef typename internal::packet_traits<Scalar>::type PacketScalar;
     typedef typename NumTraits<Scalar>::Real RealScalar;
@@ -86,12 +76,8 @@ template<typename Derived> class MapBase<Derived, ReadOnlyAccessors>
 
     typedef typename Base::CoeffReturnType CoeffReturnType;
 
-    /** \copydoc DenseBase::rows() */
-    EIGEN_DEVICE_FUNC EIGEN_CONSTEXPR
-    inline Index rows() const EIGEN_NOEXCEPT { return m_rows.value(); }
-    /** \copydoc DenseBase::cols() */
-    EIGEN_DEVICE_FUNC EIGEN_CONSTEXPR
-    inline Index cols() const EIGEN_NOEXCEPT { return m_cols.value(); }
+    inline Index rows() const { return m_rows.value(); }
+    inline Index cols() const { return m_cols.value(); }
 
     /** Returns a pointer to the first coefficient of the matrix or vector.
       *
@@ -99,39 +85,30 @@ template<typename Derived> class MapBase<Derived, ReadOnlyAccessors>
       *
       * \sa innerStride(), outerStride()
       */
-    EIGEN_DEVICE_FUNC inline const Scalar* data() const { return m_data; }
+    inline const Scalar* data() const { return m_data; }
 
-    /** \copydoc PlainObjectBase::coeff(Index,Index) const */
-    EIGEN_DEVICE_FUNC
     inline const Scalar& coeff(Index rowId, Index colId) const
     {
       return m_data[colId * colStride() + rowId * rowStride()];
     }
 
-    /** \copydoc PlainObjectBase::coeff(Index) const */
-    EIGEN_DEVICE_FUNC
     inline const Scalar& coeff(Index index) const
     {
       EIGEN_STATIC_ASSERT_INDEX_BASED_ACCESS(Derived)
       return m_data[index * innerStride()];
     }
 
-    /** \copydoc PlainObjectBase::coeffRef(Index,Index) const */
-    EIGEN_DEVICE_FUNC
     inline const Scalar& coeffRef(Index rowId, Index colId) const
     {
       return this->m_data[colId * colStride() + rowId * rowStride()];
     }
 
-    /** \copydoc PlainObjectBase::coeffRef(Index) const */
-    EIGEN_DEVICE_FUNC
     inline const Scalar& coeffRef(Index index) const
     {
       EIGEN_STATIC_ASSERT_INDEX_BASED_ACCESS(Derived)
       return this->m_data[index * innerStride()];
     }
 
-    /** \internal */
     template<int LoadMode>
     inline PacketScalar packet(Index rowId, Index colId) const
     {
@@ -139,7 +116,6 @@ template<typename Derived> class MapBase<Derived, ReadOnlyAccessors>
                (m_data + (colId * colStride() + rowId * rowStride()));
     }
 
-    /** \internal */
     template<int LoadMode>
     inline PacketScalar packet(Index index) const
     {
@@ -147,16 +123,12 @@ template<typename Derived> class MapBase<Derived, ReadOnlyAccessors>
       return internal::ploadt<PacketScalar, LoadMode>(m_data + index * innerStride());
     }
 
-    /** \internal Constructor for fixed size matrices or vectors */
-    EIGEN_DEVICE_FUNC
-    explicit inline MapBase(PointerType dataPtr) : m_data(dataPtr), m_rows(RowsAtCompileTime), m_cols(ColsAtCompileTime)
+    inline MapBase(PointerType dataPtr) : m_data(dataPtr), m_rows(RowsAtCompileTime), m_cols(ColsAtCompileTime)
     {
       EIGEN_STATIC_ASSERT_FIXED_SIZE(Derived)
-      checkSanity<Derived>();
+      checkSanity();
     }
 
-    /** \internal Constructor for dynamically sized vectors */
-    EIGEN_DEVICE_FUNC
     inline MapBase(PointerType dataPtr, Index vecSize)
             : m_data(dataPtr),
               m_rows(RowsAtCompileTime == Dynamic ? vecSize : Index(RowsAtCompileTime)),
@@ -165,72 +137,44 @@ template<typename Derived> class MapBase<Derived, ReadOnlyAccessors>
       EIGEN_STATIC_ASSERT_VECTOR_ONLY(Derived)
       eigen_assert(vecSize >= 0);
       eigen_assert(dataPtr == 0 || SizeAtCompileTime == Dynamic || SizeAtCompileTime == vecSize);
-      checkSanity<Derived>();
+      checkSanity();
     }
 
-    /** \internal Constructor for dynamically sized matrices */
-    EIGEN_DEVICE_FUNC
-    inline MapBase(PointerType dataPtr, Index rows, Index cols)
-            : m_data(dataPtr), m_rows(rows), m_cols(cols)
+    inline MapBase(PointerType dataPtr, Index nbRows, Index nbCols)
+            : m_data(dataPtr), m_rows(nbRows), m_cols(nbCols)
     {
       eigen_assert( (dataPtr == 0)
-              || (   rows >= 0 && (RowsAtCompileTime == Dynamic || RowsAtCompileTime == rows)
-                  && cols >= 0 && (ColsAtCompileTime == Dynamic || ColsAtCompileTime == cols)));
-      checkSanity<Derived>();
+              || (   nbRows >= 0 && (RowsAtCompileTime == Dynamic || RowsAtCompileTime == nbRows)
+                  && nbCols >= 0 && (ColsAtCompileTime == Dynamic || ColsAtCompileTime == nbCols)));
+      checkSanity();
     }
 
-    #ifdef EIGEN_MAPBASE_PLUGIN
-    #include EIGEN_MAPBASE_PLUGIN
-    #endif
-
   protected:
-    EIGEN_DEFAULT_COPY_CONSTRUCTOR(MapBase)
-    EIGEN_DEFAULT_EMPTY_CONSTRUCTOR_AND_DESTRUCTOR(MapBase)
 
-    template<typename T>
-    EIGEN_DEVICE_FUNC
-    void checkSanity(typename internal::enable_if<(internal::traits<T>::Alignment>0),void*>::type = 0) const
+    void checkSanity() const
     {
-#if EIGEN_MAX_ALIGN_BYTES>0
-      // innerStride() is not set yet when this function is called, so we optimistically assume the lowest plausible value:
-      const Index minInnerStride = InnerStrideAtCompileTime == Dynamic ? 1 : Index(InnerStrideAtCompileTime);
-      EIGEN_ONLY_USED_FOR_DEBUG(minInnerStride);
-      eigen_assert((   ((internal::UIntPtr(m_data) % internal::traits<Derived>::Alignment) == 0)
-                    || (cols() * rows() * minInnerStride * sizeof(Scalar)) < internal::traits<Derived>::Alignment ) && "data is not aligned");
-#endif
+      EIGEN_STATIC_ASSERT(EIGEN_IMPLIES(internal::traits<Derived>::Flags&PacketAccessBit,
+                                        internal::inner_stride_at_compile_time<Derived>::ret==1),
+                          PACKET_ACCESS_REQUIRES_TO_HAVE_INNER_STRIDE_FIXED_TO_1);
+      eigen_assert(EIGEN_IMPLIES(internal::traits<Derived>::Flags&AlignedBit, (size_t(m_data) % 16) == 0)
+                   && "data is not aligned");
     }
 
-    template<typename T>
-    EIGEN_DEVICE_FUNC
-    void checkSanity(typename internal::enable_if<internal::traits<T>::Alignment==0,void*>::type = 0) const
-    {}
-
     PointerType m_data;
     const internal::variable_if_dynamic<Index, RowsAtCompileTime> m_rows;
     const internal::variable_if_dynamic<Index, ColsAtCompileTime> m_cols;
 };
 
-/** \ingroup Core_Module
-  *
-  * \brief Base class for non-const dense Map and Block expression with direct access
-  *
-  * This base class provides the non-const low-level accessors (e.g. coeff and coeffRef) of
-  * dense Map and Block objects with direct access.
-  * It inherits MapBase<Derived, ReadOnlyAccessors> which defines the const variant for reading specific entries.
-  *
-  * \sa class Map, class Block
-  */
 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::Index Index;
     typedef typename Base::PointerType PointerType;
 
     using Base::derived;
@@ -251,18 +195,14 @@ template<typename Derived> class MapBase<Derived, WriteAccessors>
                     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)
@@ -284,27 +224,19 @@ template<typename Derived> class MapBase<Derived, WriteAccessors>
                 (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) {}
+    explicit inline MapBase(PointerType dataPtr) : Base(dataPtr) {}
+    inline MapBase(PointerType dataPtr, Index vecSize) : Base(dataPtr, vecSize) {}
+    inline MapBase(PointerType dataPtr, Index nbRows, Index nbCols) : Base(dataPtr, nbRows, nbCols) {}
 
-    EIGEN_DEVICE_FUNC
     Derived& operator=(const MapBase& other)
     {
-      ReadOnlyMapBase::Base::operator=(other);
+      Base::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=;
-  protected:
-    EIGEN_DEFAULT_COPY_CONSTRUCTOR(MapBase)
-    EIGEN_DEFAULT_EMPTY_CONSTRUCTOR_AND_DESTRUCTOR(MapBase)
+    using Base::Base::operator=;
 };
 
-#undef EIGEN_STATIC_ASSERT_INDEX_BASED_ACCESS
-
 } // end namespace Eigen
 
 #endif // EIGEN_MAPBASE_H

Eigen/src/Core/MathFunctionsImpl.h

@@ -1,191 +0,0 @@
-// This file is part of Eigen, a lightweight C++ template library
-// for linear algebra.
-//
-// Copyright (C) 2014 Pedro Gonnet (pedro.gonnet@gmail.com)
-// Copyright (C) 2016 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_MATHFUNCTIONSIMPL_H
-#define EIGEN_MATHFUNCTIONSIMPL_H
-
-namespace Eigen {
-
-namespace internal {
-
-/** \internal \returns the hyperbolic tan of \a a (coeff-wise)
-    Doesn't do anything fancy, just a 13/6-degree rational interpolant which
-    is accurate up to a couple of ulps in the (approximate) range [-8, 8],
-    outside of which tanh(x) = +/-1 in single precision. The input is clamped
-    to the range [-c, c]. The value c is chosen as the smallest value where
-    the approximation evaluates to exactly 1. In the reange [-0.0004, 0.0004]
-    the approxmation tanh(x) ~= x is used for better accuracy as x tends to zero.
-
-    This implementation works on both scalars and packets.
-*/
-template<typename T>
-T generic_fast_tanh_float(const T& a_x)
-{
-  // Clamp the inputs to the range [-c, c]
-#ifdef EIGEN_VECTORIZE_FMA
-  const T plus_clamp = pset1<T>(7.99881172180175781f);
-  const T minus_clamp = pset1<T>(-7.99881172180175781f);
-#else
-  const T plus_clamp = pset1<T>(7.90531110763549805f);
-  const T minus_clamp = pset1<T>(-7.90531110763549805f);
-#endif
-  const T tiny = pset1<T>(0.0004f);
-  const T x = pmax(pmin(a_x, plus_clamp), minus_clamp);
-  const T tiny_mask = pcmp_lt(pabs(a_x), tiny);
-  // The monomial coefficients of the numerator polynomial (odd).
-  const T alpha_1 = pset1<T>(4.89352455891786e-03f);
-  const T alpha_3 = pset1<T>(6.37261928875436e-04f);
-  const T alpha_5 = pset1<T>(1.48572235717979e-05f);
-  const T alpha_7 = pset1<T>(5.12229709037114e-08f);
-  const T alpha_9 = pset1<T>(-8.60467152213735e-11f);
-  const T alpha_11 = pset1<T>(2.00018790482477e-13f);
-  const T alpha_13 = pset1<T>(-2.76076847742355e-16f);
-
-  // The monomial coefficients of the denominator polynomial (even).
-  const T beta_0 = pset1<T>(4.89352518554385e-03f);
-  const T beta_2 = pset1<T>(2.26843463243900e-03f);
-  const T beta_4 = pset1<T>(1.18534705686654e-04f);
-  const T beta_6 = pset1<T>(1.19825839466702e-06f);
-
-  // Since the polynomials are odd/even, we need x^2.
-  const T x2 = pmul(x, x);
-
-  // Evaluate the numerator polynomial p.
-  T p = pmadd(x2, alpha_13, alpha_11);
-  p = pmadd(x2, p, alpha_9);
-  p = pmadd(x2, p, alpha_7);
-  p = pmadd(x2, p, alpha_5);
-  p = pmadd(x2, p, alpha_3);
-  p = pmadd(x2, p, alpha_1);
-  p = pmul(x, p);
-
-  // Evaluate the denominator polynomial q.
-  T q = pmadd(x2, beta_6, beta_4);
-  q = pmadd(x2, q, beta_2);
-  q = pmadd(x2, q, beta_0);
-
-  // Divide the numerator by the denominator.
-  return pselect(tiny_mask, x, pdiv(p, q));
-}
-
-template<typename RealScalar>
-EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
-RealScalar positive_real_hypot(const RealScalar& x, const RealScalar& y)
-{
-  // IEEE IEC 6059 special cases.
-  if ((numext::isinf)(x) || (numext::isinf)(y))
-    return NumTraits<RealScalar>::infinity();
-  if ((numext::isnan)(x) || (numext::isnan)(y))
-    return NumTraits<RealScalar>::quiet_NaN();
-    
-  EIGEN_USING_STD(sqrt);
-  RealScalar p, qp;
-  p = numext::maxi(x,y);
-  if(p==RealScalar(0)) return RealScalar(0);
-  qp = numext::mini(y,x) / p;
-  return p * sqrt(RealScalar(1) + qp*qp);
-}
-
-template<typename Scalar>
-struct hypot_impl
-{
-  typedef typename NumTraits<Scalar>::Real RealScalar;
-  static EIGEN_DEVICE_FUNC
-  inline RealScalar run(const Scalar& x, const Scalar& y)
-  {
-    EIGEN_USING_STD(abs);
-    return positive_real_hypot<RealScalar>(abs(x), abs(y));
-  }
-};
-
-// Generic complex sqrt implementation that correctly handles corner cases
-// according to https://en.cppreference.com/w/cpp/numeric/complex/sqrt
-template<typename T>
-EIGEN_DEVICE_FUNC std::complex<T> complex_sqrt(const std::complex<T>& z) {
-  // Computes the principal sqrt of the input.
-  //
-  // For a complex square root of the number x + i*y. We want to find real
-  // numbers u and v such that
-  //    (u + i*v)^2 = x + i*y  <=>
-  //    u^2 - v^2 + i*2*u*v = x + i*v.
-  // By equating the real and imaginary parts we get:
-  //    u^2 - v^2 = x
-  //    2*u*v = y.
-  //
-  // For x >= 0, this has the numerically stable solution
-  //    u = sqrt(0.5 * (x + sqrt(x^2 + y^2)))
-  //    v = y / (2 * u)
-  // and for x < 0,
-  //    v = sign(y) * sqrt(0.5 * (-x + sqrt(x^2 + y^2)))
-  //    u = y / (2 * v)
-  //
-  // Letting w = sqrt(0.5 * (|x| + |z|)),
-  //   if x == 0: u = w, v = sign(y) * w
-  //   if x > 0:  u = w, v = y / (2 * w)
-  //   if x < 0:  u = |y| / (2 * w), v = sign(y) * w
-
-  const T x = numext::real(z);
-  const T y = numext::imag(z);
-  const T zero = T(0);
-  const T w = numext::sqrt(T(0.5) * (numext::abs(x) + numext::hypot(x, y)));
-
-  return
-    (numext::isinf)(y) ? std::complex<T>(NumTraits<T>::infinity(), y)
-      : x == zero ? std::complex<T>(w, y < zero ? -w : w)
-      : x > zero ? std::complex<T>(w, y / (2 * w))
-      : std::complex<T>(numext::abs(y) / (2 * w), y < zero ? -w : w );
-}
-
-// Generic complex rsqrt implementation.
-template<typename T>
-EIGEN_DEVICE_FUNC std::complex<T> complex_rsqrt(const std::complex<T>& z) {
-  // Computes the principal reciprocal sqrt of the input.
-  //
-  // For a complex reciprocal square root of the number z = x + i*y. We want to
-  // find real numbers u and v such that
-  //    (u + i*v)^2 = 1 / (x + i*y)  <=>
-  //    u^2 - v^2 + i*2*u*v = x/|z|^2 - i*v/|z|^2.
-  // By equating the real and imaginary parts we get:
-  //    u^2 - v^2 = x/|z|^2
-  //    2*u*v = y/|z|^2.
-  //
-  // For x >= 0, this has the numerically stable solution
-  //    u = sqrt(0.5 * (x + |z|)) / |z|
-  //    v = -y / (2 * u * |z|)
-  // and for x < 0,
-  //    v = -sign(y) * sqrt(0.5 * (-x + |z|)) / |z|
-  //    u = -y / (2 * v * |z|)
-  //
-  // Letting w = sqrt(0.5 * (|x| + |z|)),
-  //   if x == 0: u = w / |z|, v = -sign(y) * w / |z|
-  //   if x > 0:  u = w / |z|, v = -y / (2 * w * |z|)
-  //   if x < 0:  u = |y| / (2 * w * |z|), v = -sign(y) * w / |z|
-
-  const T x = numext::real(z);
-  const T y = numext::imag(z);
-  const T zero = T(0);
-
-  const T abs_z = numext::hypot(x, y);
-  const T w = numext::sqrt(T(0.5) * (numext::abs(x) + abs_z));
-  const T woz = w / abs_z;
-  // Corner cases consistent with 1/sqrt(z) on gcc/clang.
-  return
-    abs_z == zero ? std::complex<T>(NumTraits<T>::infinity(), NumTraits<T>::quiet_NaN())
-      : ((numext::isinf)(x) || (numext::isinf)(y)) ? std::complex<T>(zero, zero)
-      : x == zero ? std::complex<T>(woz, y < zero ? woz : -woz)
-      : x > zero ? std::complex<T>(woz, -y / (2 * w * abs_z))
-      : std::complex<T>(numext::abs(y) / (2 * w * abs_z), y < zero ? woz : -woz );
-}
-
-} // end namespace internal
-
-} // end namespace Eigen
-
-#endif // EIGEN_MATHFUNCTIONSIMPL_H

Eigen/src/Core/Matrix.h

@@ -13,45 +13,6 @@
 
 namespace Eigen {
 
-namespace internal {
-template<typename _Scalar, int _Rows, int _Cols, int _Options, int _MaxRows, int _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 && (EIGEN_UNALIGNED_VECTORIZE || (actual_alignment>=required_alignment))) ? PacketAccessBit : 0
-    };
-
-public:
-  typedef _Scalar Scalar;
-  typedef Dense StorageKind;
-  typedef Eigen::Index StorageIndex;
-  typedef MatrixXpr XprKind;
-  enum {
-    RowsAtCompileTime = _Rows,
-    ColsAtCompileTime = _Cols,
-    MaxRowsAtCompileTime = _MaxRows,
-    MaxColsAtCompileTime = _MaxCols,
-    Flags = compute_matrix_flags<_Scalar, _Rows, _Cols, _Options, _MaxRows, _MaxCols>::ret,
-    Options = _Options,
-    InnerStrideAtCompileTime = 1,
-    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
-  };
-};
-}
-
 /** \class Matrix
   * \ingroup Core_Module
   *
@@ -63,13 +24,13 @@ public:
   * The %Matrix class encompasses \em both fixed-size and dynamic-size objects (\ref fixedsize "note").
   *
   * The first three template parameters are required:
-  * \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 _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 _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 A combination of either \b #RowMajor or \b #ColMajor, and of either
+  * \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 \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.
@@ -106,7 +67,7 @@ public:
   * \endcode
   *
   * This class can be extended with the help of the plugin mechanism described on the page
-  * \ref TopicCustomizing_Plugins by defining the preprocessor symbol \c EIGEN_MATRIX_PLUGIN.
+  * \ref TopicCustomizingEigen by defining the preprocessor symbol \c EIGEN_MATRIX_PLUGIN.
   *
   * <i><b>Some notes:</b></i>
   *
@@ -136,44 +97,32 @@ public:
   * are the dimensions of the original matrix, while _Rows and _Cols are Dynamic.</dd>
   * </dl>
   *
-  * <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
+  * \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 traits<Matrix<_Scalar, _Rows, _Cols, _Options, _MaxRows, _MaxCols> >
+{
+  typedef _Scalar Scalar;
+  typedef Dense StorageKind;
+  typedef DenseIndex Index;
+  typedef MatrixXpr XprKind;
+  enum {
+    RowsAtCompileTime = _Rows,
+    ColsAtCompileTime = _Cols,
+    MaxRowsAtCompileTime = _MaxRows,
+    MaxColsAtCompileTime = _MaxCols,
+    Flags = compute_matrix_flags<_Scalar, _Rows, _Cols, _Options, _MaxRows, _MaxCols>::ret,
+    CoeffReadCost = NumTraits<Scalar>::ReadCost,
+    Options = _Options,
+    InnerStrideAtCompileTime = 1,
+    OuterStrideAtCompileTime = (Options&RowMajor) ? ColsAtCompileTime : RowsAtCompileTime
+  };
+};
+}
+
 template<typename _Scalar, int _Rows, int _Cols, int _Options, int _MaxRows, int _MaxCols>
 class Matrix
   : public PlainObjectBase<Matrix<_Scalar, _Rows, _Cols, _Options, _MaxRows, _MaxCols> >
@@ -202,7 +151,6 @@ class Matrix
       *
       * \callgraph
       */
-    EIGEN_DEVICE_FUNC
     EIGEN_STRONG_INLINE Matrix& operator=(const Matrix& other)
     {
       return Base::_set(other);
@@ -219,8 +167,7 @@ class Matrix
       * remain row-vectors and vectors remain vectors.
       */
     template<typename OtherDerived>
-    EIGEN_DEVICE_FUNC
-    EIGEN_STRONG_INLINE Matrix& operator=(const DenseBase<OtherDerived>& other)
+    EIGEN_STRONG_INLINE Matrix& operator=(const MatrixBase<OtherDerived>& other)
     {
       return Base::_set(other);
     }
@@ -232,14 +179,12 @@ 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);
@@ -255,137 +200,52 @@ class Matrix
       *
       * \sa resize(Index,Index)
       */
-    EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
-    Matrix() : Base()
+    EIGEN_STRONG_INLINE Matrix() : Base()
     {
       Base::_check_template_params();
       EIGEN_INITIALIZE_COEFFS_IF_THAT_OPTION_IS_ENABLED
     }
 
     // FIXME is it still needed
-    EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
-    explicit Matrix(internal::constructor_without_unaligned_array_assert)
+    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 }
 
-#if EIGEN_HAS_RVALUE_REFERENCES
-    EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
-    Matrix(Matrix&& other) EIGEN_NOEXCEPT_IF(std::is_nothrow_move_constructible<Scalar>::value)
-      : Base(std::move(other))
-    {
-      Base::_check_template_params();
-    }
-    EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
-    Matrix& operator=(Matrix&& other) EIGEN_NOEXCEPT_IF(std::is_nothrow_move_assignable<Scalar>::value)
-    {
-      Base::operator=(std::move(other));
-      return *this;
-    }
-#endif
-
-#if EIGEN_HAS_CXX11
-    /** \copydoc PlainObjectBase(const Scalar&, const Scalar&, const Scalar&,  const Scalar&, const ArgTypes&... args)
-     *
-     * Example: \include Matrix_variadic_ctor_cxx11.cpp
-     * Output: \verbinclude Matrix_variadic_ctor_cxx11.out
-     *
-     * \sa Matrix(const std::initializer_list<std::initializer_list<Scalar>>&)
-     */
-    template <typename... ArgTypes>
-    EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
-    Matrix(const Scalar& a0, const Scalar& a1, const Scalar& a2,  const Scalar& a3, const ArgTypes&... args)
-      : Base(a0, a1, a2, a3, args...) {}
-
-    /** \brief Constructs a Matrix and initializes it from the coefficients given as initializer-lists grouped by row. \cpp11
+    /** \brief Constructs a vector or row-vector with given dimension. \only_for_vectors
       *
-      * In the general case, the constructor takes a list of rows, each row being represented as a list of coefficients:
-      *
-      * Example: \include Matrix_initializer_list_23_cxx11.cpp
-      * Output: \verbinclude Matrix_initializer_list_23_cxx11.out
-      *
-      * Each of the inner initializer lists must contain the exact same number of elements, otherwise an assertion is triggered.
-      *
-      * In the case of a compile-time column vector, implicit transposition from a single row is allowed.
-      * Therefore <code>VectorXd{{1,2,3,4,5}}</code> is legal and the more verbose syntax
-      * <code>RowVectorXd{{1},{2},{3},{4},{5}}</code> can be avoided:
-      *
-      * Example: \include Matrix_initializer_list_vector_cxx11.cpp
-      * Output: \verbinclude Matrix_initializer_list_vector_cxx11.out
-      *
-      * In the case of fixed-sized matrices, the initializer list sizes must exactly match the matrix sizes,
-      * and implicit transposition is allowed for compile-time vectors only.
-      *
-      * \sa Matrix(const Scalar& a0, const Scalar& a1, const Scalar& a2,  const Scalar& a3, const ArgTypes&... args)
+      * 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_DEVICE_FUNC
-    explicit EIGEN_STRONG_INLINE Matrix(const std::initializer_list<std::initializer_list<Scalar>>& list) : Base(list) {}
-#endif // end EIGEN_HAS_CXX11
-
-#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)
+    EIGEN_STRONG_INLINE explicit Matrix(Index dim)
+      : Base(dim, RowsAtCompileTime == 1 ? 1 : dim, ColsAtCompileTime == 1 ? 1 : dim)
     {
       Base::_check_template_params();
-      Base::template _init1<T>(x);
+      EIGEN_STATIC_ASSERT_VECTOR_ONLY(Matrix)
+      eigen_assert(dim >= 0);
+      eigen_assert(SizeAtCompileTime == Dynamic || SizeAtCompileTime == dim);
+      EIGEN_INITIALIZE_COEFFS_IF_THAT_OPTION_IS_ENABLED
     }
 
+    #ifndef EIGEN_PARSED_BY_DOXYGEN
     template<typename T0, typename T1>
-    EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
-    Matrix(const T0& x, const T1& y)
+    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
-      * \sa Matrix(const Scalar&, const Scalar&, const Scalar&,  const Scalar&, const ArgTypes&...) */
-    Matrix(const Scalar& x);
+    #else
     /** \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.
-      *
-      * \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() instead. */
     Matrix(Index rows, Index cols);
-
-    /** \brief Constructs an initialized 2D vector with given coefficients
-      * \sa Matrix(const Scalar&, const Scalar&, const Scalar&,  const Scalar&, const ArgTypes&...) */
+    /** \brief Constructs an initialized 2D vector with given coefficients */
     Matrix(const Scalar& x, const Scalar& y);
-    #endif  // end EIGEN_PARSED_BY_DOXYGEN
+    #endif
 
-    /** \brief Constructs an initialized 3D vector with given coefficients
-      * \sa Matrix(const Scalar&, const Scalar&, const Scalar&,  const Scalar&, const ArgTypes&...)
-      */
-    EIGEN_DEVICE_FUNC
+    /** \brief Constructs an initialized 3D vector with given coefficients */
     EIGEN_STRONG_INLINE Matrix(const Scalar& x, const Scalar& y, const Scalar& z)
     {
       Base::_check_template_params();
@@ -394,10 +254,7 @@ class Matrix
       m_storage.data()[1] = y;
       m_storage.data()[2] = z;
     }
-    /** \brief Constructs an initialized 4D vector with given coefficients
-      * \sa Matrix(const Scalar&, const Scalar&, const Scalar&,  const Scalar&, const ArgTypes&...)
-      */
-    EIGEN_DEVICE_FUNC
+    /** \brief Constructs an initialized 4D vector with given coefficients */
     EIGEN_STRONG_INLINE Matrix(const Scalar& x, const Scalar& y, const Scalar& z, const Scalar& w)
     {
       Base::_check_template_params();
@@ -408,35 +265,76 @@ 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())
+    {
+      // This test resides here, to bring the error messages closer to the user. Normally, these checks
+      // are performed deeply within the library, thus causing long and scary error traces.
+      EIGEN_STATIC_ASSERT((internal::is_same<Scalar, typename OtherDerived::Scalar>::value),
+        YOU_MIXED_DIFFERENT_NUMERIC_TYPES__YOU_NEED_TO_USE_THE_CAST_METHOD_OF_MATRIXBASE_TO_CAST_NUMERIC_TYPES_EXPLICITLY)
+
+      Base::_check_template_params();
+      Base::_set_noalias(other);
+    }
     /** \brief Copy constructor */
-    EIGEN_DEVICE_FUNC
-    EIGEN_STRONG_INLINE Matrix(const Matrix& other) : Base(other)
-    { }
+    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);
+    }
 
     /** \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())
-    { }
+      : 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;
+    }
 
-    EIGEN_DEVICE_FUNC EIGEN_CONSTEXPR
-    inline Index innerStride() const EIGEN_NOEXCEPT { return 1; }
-    EIGEN_DEVICE_FUNC EIGEN_CONSTEXPR
-    inline Index outerStride() const EIGEN_NOEXCEPT { return this->innerSize(); }
+    /** \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> const & other)
+    { this->_swap(other.derived()); }
+
+    inline Index innerStride() const { return 1; }
+    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);
 
+    #ifdef EIGEN2_SUPPORT
+    template<typename OtherDerived>
+    explicit Matrix(const eigen2_RotationBase<OtherDerived,ColsAtCompileTime>& r);
+    template<typename OtherDerived>
+    Matrix& operator=(const eigen2_RotationBase<OtherDerived,ColsAtCompileTime>& r);
+    #endif
+
     // allow to extend Matrix outside Eigen
     #ifdef EIGEN_MATRIX_PLUGIN
     #include EIGEN_MATRIX_PLUGIN
@@ -453,7 +351,7 @@ class Matrix
   *
   * \ingroup Core_Module
   *
-  * %Eigen defines several typedef shortcuts for most common matrix and vector types.
+  * Eigen defines several typedef shortcuts for most common matrix and vector types.
   *
   * The general patterns are the following:
   *
@@ -466,15 +364,6 @@ class Matrix
   * There are also \c VectorSizeType and \c RowVectorSizeType which are self-explanatory. For example, \c Vector4cf is
   * a fixed-size vector of 4 complex floats.
   *
-  * With \cpp11, template alias are also defined for common sizes.
-  * They follow the same pattern as above except that the scalar type suffix is replaced by a
-  * template parameter, i.e.:
-  *   - `MatrixSize<Type>` where `Size` can be \c 2,\c 3,\c 4 for fixed size square matrices or \c X for dynamic size.
-  *   - `MatrixXSize<Type>` and `MatrixSizeX<Type>` where `Size` can be \c 2,\c 3,\c 4 for hybrid dynamic/fixed matrices.
-  *   - `VectorSize<Type>` and `RowVectorSize<Type>` for column and row vectors.
-  *
-  * With \cpp11, you can also use fully generic column and row vector types: `Vector<Type,Size>` and `RowVector<Type,Size>`.
-  *
   * \sa class Matrix
   */
 
@@ -511,55 +400,6 @@ EIGEN_MAKE_TYPEDEFS_ALL_SIZES(std::complex<double>, cd)
 #undef EIGEN_MAKE_TYPEDEFS
 #undef EIGEN_MAKE_FIXED_TYPEDEFS
 
-#if EIGEN_HAS_CXX11
-
-#define EIGEN_MAKE_TYPEDEFS(Size, SizeSuffix)                     \
-/** \ingroup matrixtypedefs */                                    \
-/** \brief \cpp11 */                                              \
-template <typename Type>                                          \
-using Matrix##SizeSuffix = Matrix<Type, Size, Size>;              \
-/** \ingroup matrixtypedefs */                                    \
-/** \brief \cpp11 */                                              \
-template <typename Type>                                          \
-using Vector##SizeSuffix = Matrix<Type, Size, 1>;                 \
-/** \ingroup matrixtypedefs */                                    \
-/** \brief \cpp11 */                                              \
-template <typename Type>                                          \
-using RowVector##SizeSuffix = Matrix<Type, 1, Size>;
-
-#define EIGEN_MAKE_FIXED_TYPEDEFS(Size)                           \
-/** \ingroup matrixtypedefs */                                    \
-/** \brief \cpp11 */                                              \
-template <typename Type>                                          \
-using Matrix##Size##X = Matrix<Type, Size, Dynamic>;              \
-/** \ingroup matrixtypedefs */                                    \
-/** \brief \cpp11 */                                              \
-template <typename Type>                                          \
-using Matrix##X##Size = Matrix<Type, Dynamic, Size>;
-
-EIGEN_MAKE_TYPEDEFS(2, 2)
-EIGEN_MAKE_TYPEDEFS(3, 3)
-EIGEN_MAKE_TYPEDEFS(4, 4)
-EIGEN_MAKE_TYPEDEFS(Dynamic, X)
-EIGEN_MAKE_FIXED_TYPEDEFS(2)
-EIGEN_MAKE_FIXED_TYPEDEFS(3)
-EIGEN_MAKE_FIXED_TYPEDEFS(4)
-
-/** \ingroup matrixtypedefs
-  * \brief \cpp11 */
-template <typename Type, int Size>
-using Vector = Matrix<Type, Size, 1>;
-
-/** \ingroup matrixtypedefs
-  * \brief \cpp11 */
-template <typename Type, int Size>
-using RowVector = Matrix<Type, 1, Size>;
-
-#undef EIGEN_MAKE_TYPEDEFS
-#undef EIGEN_MAKE_FIXED_TYPEDEFS
-
-#endif // EIGEN_HAS_CXX11
-
 } // end namespace Eigen
 
 #endif // EIGEN_MATRIX_H

Eigen/src/Core/MatrixBase.h

@@ -41,9 +41,9 @@ namespace Eigen {
   * \endcode
   *
   * This class can be extended with the help of the plugin mechanism described on the page
-  * \ref TopicCustomizing_Plugins by defining the preprocessor symbol \c EIGEN_MATRIXBASE_PLUGIN.
+  * \ref TopicCustomizingEigen by defining the preprocessor symbol \c EIGEN_MATRIXBASE_PLUGIN.
   *
-  * \sa \blank \ref TopicClassHierarchy
+  * \sa \ref TopicClassHierarchy
   */
 template<typename Derived> class MatrixBase
   : public DenseBase<Derived>
@@ -52,7 +52,7 @@ template<typename Derived> class MatrixBase
 #ifndef EIGEN_PARSED_BY_DOXYGEN
     typedef MatrixBase StorageBaseType;
     typedef typename internal::traits<Derived>::StorageKind StorageKind;
-    typedef typename internal::traits<Derived>::StorageIndex StorageIndex;
+    typedef typename internal::traits<Derived>::Index Index;
     typedef typename internal::traits<Derived>::Scalar Scalar;
     typedef typename internal::packet_traits<Scalar>::type PacketScalar;
     typedef typename NumTraits<Scalar>::Real RealScalar;
@@ -66,6 +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;
@@ -76,7 +77,6 @@ template<typename Derived> class MatrixBase
     using Base::coeffRef;
     using Base::lazyAssign;
     using Base::eval;
-    using Base::operator-;
     using Base::operator+=;
     using Base::operator-=;
     using Base::operator*=;
@@ -98,14 +98,25 @@ template<typename Derived> class MatrixBase
 
     /** \returns the size of the main diagonal, which is min(rows(),cols()).
       * \sa rows(), cols(), SizeAtCompileTime. */
-    EIGEN_DEVICE_FUNC
-    inline Index diagonalSize() const { return (numext::mini)(rows(),cols()); }
+    inline Index diagonalSize() const { return (std::min)(rows(),cols()); }
 
-    typedef typename Base::PlainObject PlainObject;
+    /** \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 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
+          > PlainObject;
 
 #ifndef EIGEN_PARSED_BY_DOXYGEN
     /** \internal Represents a matrix with all coefficients equal to one another*/
-    typedef CwiseNullaryOp<internal::scalar_constant_op<Scalar>,PlainObject> ConstantReturnType;
+    typedef CwiseNullaryOp<internal::scalar_constant_op<Scalar>,Derived> ConstantReturnType;
     /** \internal the return type of MatrixBase::adjoint() */
     typedef typename internal::conditional<NumTraits<Scalar>::IsComplex,
                         CwiseUnaryOp<internal::scalar_conjugate_op<Scalar>, ConstTransposeReturnType>,
@@ -114,7 +125,7 @@ template<typename Derived> class MatrixBase
     /** \internal Return type of eigenvalues() */
     typedef Matrix<std::complex<RealScalar>, internal::traits<Derived>::ColsAtCompileTime, 1, ColMajor> EigenvaluesReturnType;
     /** \internal the return type of identity */
-    typedef CwiseNullaryOp<internal::scalar_identity_op<Scalar>,PlainObject> IdentityReturnType;
+    typedef CwiseNullaryOp<internal::scalar_identity_op<Scalar>,Derived> IdentityReturnType;
     /** \internal the return type of unit vectors */
     typedef Block<const CwiseNullaryOp<internal::scalar_identity_op<Scalar>, SquareMatrixType>,
                   internal::traits<Derived>::RowsAtCompileTime,
@@ -122,7 +133,7 @@ template<typename Derived> class MatrixBase
 #endif // not EIGEN_PARSED_BY_DOXYGEN
 
 #define EIGEN_CURRENT_STORAGE_BASE_CLASS Eigen::MatrixBase
-#define EIGEN_DOC_UNARY_ADDONS(X,Y)
+#   include "../plugins/CommonCwiseUnaryOps.h"
 #   include "../plugins/CommonCwiseBinaryOps.h"
 #   include "../plugins/MatrixCwiseUnaryOps.h"
 #   include "../plugins/MatrixCwiseBinaryOps.h"
@@ -130,44 +141,43 @@ template<typename Derived> class MatrixBase
 #     include EIGEN_MATRIXBASE_PLUGIN
 #   endif
 #undef EIGEN_CURRENT_STORAGE_BASE_CLASS
-#undef EIGEN_DOC_UNARY_ADDONS
 
     /** Special case of the template operator=, in order to prevent the compiler
       * from generating a default operator= (issue hit with g++ 4.1)
       */
-    EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
     Derived& operator=(const MatrixBase& other);
 
     // We cannot inherit here via Base::operator= since it is causing
     // trouble with MSVC.
 
     template <typename OtherDerived>
-    EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
     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);
+
+    template<typename MatrixPower, typename Lhs, typename Rhs>
+    Derived& lazyAssign(const MatrixPowerProduct<MatrixPower, Lhs,Rhs>& other);
+#endif // not EIGEN_PARSED_BY_DOXYGEN
+
     template<typename OtherDerived>
-    EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
     Derived& operator+=(const MatrixBase<OtherDerived>& other);
     template<typename OtherDerived>
-    EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
     Derived& operator-=(const MatrixBase<OtherDerived>& other);
 
     template<typename OtherDerived>
-    EIGEN_DEVICE_FUNC
-    const Product<Derived,OtherDerived>
+    const typename ProductReturnType<Derived,OtherDerived>::Type
     operator*(const MatrixBase<OtherDerived> &other) const;
 
     template<typename OtherDerived>
-    EIGEN_DEVICE_FUNC
-    const Product<Derived,OtherDerived,LazyProduct>
+    const typename LazyProductReturnType<Derived,OtherDerived>::Type
     lazyProduct(const MatrixBase<OtherDerived> &other) const;
 
     template<typename OtherDerived>
@@ -180,96 +190,89 @@ template<typename Derived> class MatrixBase
     void applyOnTheRight(const EigenBase<OtherDerived>& other);
 
     template<typename DiagonalDerived>
-    EIGEN_DEVICE_FUNC
-    const Product<Derived, DiagonalDerived, LazyProduct>
+    const DiagonalProduct<Derived, DiagonalDerived, OnTheRight>
     operator*(const DiagonalBase<DiagonalDerived> &diagonal) const;
 
     template<typename OtherDerived>
-    EIGEN_DEVICE_FUNC
-    typename ScalarBinaryOpTraits<typename internal::traits<Derived>::Scalar,typename internal::traits<OtherDerived>::Scalar>::ReturnType
+    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;
+    #ifdef EIGEN2_SUPPORT
+      template<typename OtherDerived>
+      Scalar eigen2_dot(const MatrixBase<OtherDerived>& other) const;
+    #endif
+
+    RealScalar squaredNorm() const;
+    RealScalar norm() const;
     RealScalar stableNorm() const;
     RealScalar blueNorm() const;
     RealScalar hypotNorm() const;
-    EIGEN_DEVICE_FUNC const PlainObject normalized() const;
-    EIGEN_DEVICE_FUNC const PlainObject stableNormalized() const;
-    EIGEN_DEVICE_FUNC void normalize();
-    EIGEN_DEVICE_FUNC void stableNormalize();
+    const PlainObject normalized() const;
+    void normalize();
 
-    EIGEN_DEVICE_FUNC const AdjointReturnType adjoint() const;
-    EIGEN_DEVICE_FUNC void adjointInPlace();
+    const AdjointReturnType adjoint() const;
+    void adjointInPlace();
 
     typedef Diagonal<Derived> DiagonalReturnType;
-    EIGEN_DEVICE_FUNC
     DiagonalReturnType diagonal();
-
-    typedef typename internal::add_const<Diagonal<const Derived> >::type ConstDiagonalReturnType;
-    EIGEN_DEVICE_FUNC
+	typedef typename internal::add_const<Diagonal<const Derived> >::type ConstDiagonalReturnType;
     ConstDiagonalReturnType diagonal() const;
 
     template<int Index> struct DiagonalIndexReturnType { typedef Diagonal<Derived,Index> Type; };
     template<int Index> struct ConstDiagonalIndexReturnType { typedef const Diagonal<const Derived,Index> Type; };
 
-    template<int Index>
-    EIGEN_DEVICE_FUNC
-    typename DiagonalIndexReturnType<Index>::Type diagonal();
+    template<int Index> typename DiagonalIndexReturnType<Index>::Type diagonal();
+    template<int Index> typename ConstDiagonalIndexReturnType<Index>::Type diagonal() const;
 
-    template<int Index>
-    EIGEN_DEVICE_FUNC
-    typename ConstDiagonalIndexReturnType<Index>::Type diagonal() const;
+    // Note: The "MatrixBase::" prefixes are added to help MSVC9 to match these declarations with the later implementations.
+    // On the other hand they confuse MSVC8...
+    #if (defined _MSC_VER) && (_MSC_VER >= 1500) // 2008 or later
+    typename MatrixBase::template DiagonalIndexReturnType<DynamicIndex>::Type diagonal(Index index);
+    typename MatrixBase::template ConstDiagonalIndexReturnType<DynamicIndex>::Type diagonal(Index index) const;
+    #else
+    typename DiagonalIndexReturnType<DynamicIndex>::Type diagonal(Index index);
+    typename ConstDiagonalIndexReturnType<DynamicIndex>::Type diagonal(Index index) const;
+    #endif
 
-    typedef Diagonal<Derived,DynamicIndex> DiagonalDynamicIndexReturnType;
-    typedef typename internal::add_const<Diagonal<const Derived,DynamicIndex> >::type ConstDiagonalDynamicIndexReturnType;
-
-    EIGEN_DEVICE_FUNC
-    DiagonalDynamicIndexReturnType diagonal(Index index);
-    EIGEN_DEVICE_FUNC
-    ConstDiagonalDynamicIndexReturnType diagonal(Index index) const;
+    #ifdef EIGEN2_SUPPORT
+    template<unsigned int Mode> typename internal::eigen2_part_return_type<Derived, Mode>::type part();
+    template<unsigned int Mode> const typename internal::eigen2_part_return_type<Derived, Mode>::type part() const;
+    
+    // huuuge hack. make Eigen2's matrix.part<Diagonal>() work in eigen3. Problem: Diagonal is now a class template instead
+    // of an integer constant. Solution: overload the part() method template wrt template parameters list.
+    template<template<typename T, int N> class U>
+    const DiagonalWrapper<ConstDiagonalReturnType> part() const
+    { return diagonal().asDiagonal(); }
+    #endif // EIGEN2_SUPPORT
 
     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 Mode> typename TriangularViewReturnType<Mode>::Type triangularView();
+    template<unsigned int Mode> 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;
+    template<unsigned int UpLo> typename SelfAdjointViewReturnType<UpLo>::Type selfadjointView();
+    template<unsigned int UpLo> typename ConstSelfAdjointViewReturnType<UpLo>::Type selfadjointView() const;
 
     const SparseView<Derived> sparseView(const Scalar& m_reference = Scalar(0),
                                          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();
+    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();
 
-    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);
-    EIGEN_DEVICE_FUNC Derived& setUnit(Index i);
-    EIGEN_DEVICE_FUNC Derived& setUnit(Index newSize, Index i);
 
     bool isIdentity(const RealScalar& prec = NumTraits<Scalar>::dummy_precision()) const;
     bool isDiagonal(const RealScalar& prec = NumTraits<Scalar>::dummy_precision()) const;
@@ -287,7 +290,7 @@ template<typename Derived> class MatrixBase
       *          fuzzy comparison such as isApprox()
       * \sa isApprox(), operator!= */
     template<typename OtherDerived>
-    EIGEN_DEVICE_FUNC inline bool operator==(const MatrixBase<OtherDerived>& other) const
+    inline bool operator==(const MatrixBase<OtherDerived>& other) const
     { return cwiseEqual(other).all(); }
 
     /** \returns true if at least one pair of coefficients of \c *this and \a other are not exactly equal to each other.
@@ -295,142 +298,143 @@ template<typename Derived> class MatrixBase
       *          fuzzy comparison such as isApprox()
       * \sa isApprox(), operator== */
     template<typename OtherDerived>
-    EIGEN_DEVICE_FUNC inline bool operator!=(const MatrixBase<OtherDerived>& other) const
+    inline bool operator!=(const MatrixBase<OtherDerived>& other) const
     { return cwiseNotEqual(other).any(); }
 
-    NoAlias<Derived,Eigen::MatrixBase > EIGEN_DEVICE_FUNC noalias();
+    NoAlias<Derived,Eigen::MatrixBase > noalias();
 
-    // 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(); }
+    inline const ForceAlignedAccess<Derived> forceAlignedAccess() const;
+    inline ForceAlignedAccess<Derived> forceAlignedAccess();
+    template<bool Enable> inline typename internal::add_const_on_value_type<typename internal::conditional<Enable,ForceAlignedAccess<Derived>,Derived&>::type>::type forceAlignedAccessIf() const;
+    template<bool Enable> inline typename internal::conditional<Enable,ForceAlignedAccess<Derived>,Derived&>::type forceAlignedAccessIf();
 
-    EIGEN_DEVICE_FUNC Scalar trace() const;
+    Scalar trace() const;
 
-    template<int p> EIGEN_DEVICE_FUNC RealScalar lpNorm() const;
+/////////// Array module ///////////
 
-    EIGEN_DEVICE_FUNC MatrixBase<Derived>& matrix() { return *this; }
-    EIGEN_DEVICE_FUNC const MatrixBase<Derived>& matrix() const { return *this; }
+    template<int p> RealScalar lpNorm() const;
+
+    MatrixBase<Derived>& matrix() { return *this; }
+    const MatrixBase<Derived>& matrix() const { return *this; }
 
     /** \returns an \link Eigen::ArrayBase Array \endlink expression of this matrix
       * \sa ArrayBase::matrix() */
-    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()); }
+    ArrayWrapper<Derived> array() { return derived(); }
+    const ArrayWrapper<const Derived> array() const { return derived(); }
 
 /////////// LU module ///////////
 
-    inline const FullPivLU<PlainObject> fullPivLu() const;
-    inline const PartialPivLU<PlainObject> partialPivLu() const;
+    const FullPivLU<PlainObject> fullPivLu() const;
+    const PartialPivLU<PlainObject> partialPivLu() const;
 
-    inline const PartialPivLU<PlainObject> lu() const;
+    #if EIGEN2_SUPPORT_STAGE < STAGE20_RESOLVE_API_CONFLICTS
+    const LU<PlainObject> lu() const;
+    #endif
 
-    EIGEN_DEVICE_FUNC
-    inline const Inverse<Derived> inverse() const;
+    #ifdef EIGEN2_SUPPORT
+    const LU<PlainObject> eigen2_lu() const;
+    #endif
 
+    #if EIGEN2_SUPPORT_STAGE > STAGE20_RESOLVE_API_CONFLICTS
+    const PartialPivLU<PlainObject> lu() const;
+    #endif
+    
+    #ifdef EIGEN2_SUPPORT
     template<typename ResultType>
-    inline void computeInverseAndDetWithCheck(
+    void computeInverse(MatrixBase<ResultType> *result) const {
+      *result = this->inverse();
+    }
+    #endif
+
+    const internal::inverse_impl<Derived> inverse() const;
+    template<typename ResultType>
+    void computeInverseAndDetWithCheck(
       ResultType& inverse,
       typename ResultType::Scalar& determinant,
       bool& invertible,
       const RealScalar& absDeterminantThreshold = NumTraits<Scalar>::dummy_precision()
     ) const;
-
     template<typename ResultType>
-    inline void computeInverseWithCheck(
+    void computeInverseWithCheck(
       ResultType& inverse,
       bool& invertible,
       const RealScalar& absDeterminantThreshold = NumTraits<Scalar>::dummy_precision()
     ) const;
-
-    EIGEN_DEVICE_FUNC
     Scalar determinant() const;
 
 /////////// Cholesky module ///////////
 
-    inline const LLT<PlainObject>  llt() const;
-    inline const LDLT<PlainObject> ldlt() const;
+    const LLT<PlainObject>  llt() const;
+    const LDLT<PlainObject> ldlt() const;
 
 /////////// QR module ///////////
 
-    inline const HouseholderQR<PlainObject> householderQr() const;
-    inline const ColPivHouseholderQR<PlainObject> colPivHouseholderQr() const;
-    inline const FullPivHouseholderQR<PlainObject> fullPivHouseholderQr() const;
-    inline const CompleteOrthogonalDecomposition<PlainObject> completeOrthogonalDecomposition() const;
+    const HouseholderQR<PlainObject> householderQr() const;
+    const ColPivHouseholderQR<PlainObject> colPivHouseholderQr() const;
+    const FullPivHouseholderQR<PlainObject> fullPivHouseholderQr() const;
+    
+    #ifdef EIGEN2_SUPPORT
+    const QR<PlainObject> qr() const;
+    #endif
 
-/////////// Eigenvalues module ///////////
-
-    inline EigenvaluesReturnType eigenvalues() const;
-    inline RealScalar operatorNorm() const;
+    EigenvaluesReturnType eigenvalues() const;
+    RealScalar operatorNorm() const;
 
 /////////// SVD module ///////////
 
-    inline JacobiSVD<PlainObject> jacobiSvd(unsigned int computationOptions = 0) const;
-    inline BDCSVD<PlainObject>    bdcSvd(unsigned int computationOptions = 0) const;
+    JacobiSVD<PlainObject> jacobiSvd(unsigned int computationOptions = 0) const;
+
+    #ifdef EIGEN2_SUPPORT
+    SVD<PlainObject> svd() const;
+    #endif
 
 /////////// 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 ScalarBinaryOpTraits<typename internal::traits<Derived>::Scalar,typename internal::traits<OtherDerived>::Scalar>::ReturnType Scalar;
+      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>
-    EIGEN_DEVICE_FUNC
-#ifndef EIGEN_PARSED_BY_DOXYGEN
-    inline typename cross_product_return_type<OtherDerived>::type
-#else
-    inline PlainObject
-#endif
+    typename cross_product_return_type<OtherDerived>::type
     cross(const MatrixBase<OtherDerived>& other) const;
-
     template<typename OtherDerived>
-    EIGEN_DEVICE_FUNC
-    inline PlainObject cross3(const MatrixBase<OtherDerived>& other) const;
-
-    EIGEN_DEVICE_FUNC
-    inline PlainObject unitOrthogonal(void) const;
-
-    EIGEN_DEVICE_FUNC
-    inline Matrix<Scalar,3,1> eulerAngles(Index a0, Index a1, Index a2) const;
-
+    PlainObject cross3(const MatrixBase<OtherDerived>& other) const;
+    PlainObject unitOrthogonal(void) const;
+    Matrix<Scalar,3,1> eulerAngles(Index a0, Index a1, Index a2) const;
+    
+    #if EIGEN2_SUPPORT_STAGE > STAGE20_RESOLVE_API_CONFLICTS
+    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 };
+    enum { HomogeneousReturnTypeDirection = ColsAtCompileTime==1?Vertical:Horizontal };
     typedef Homogeneous<Derived, HomogeneousReturnTypeDirection> HomogeneousReturnType;
-    EIGEN_DEVICE_FUNC
-    inline HomogeneousReturnType homogeneous() const;
-
+    HomogeneousReturnType homogeneous() const;
+    #endif
+    
     enum {
       SizeMinusOne = SizeAtCompileTime==Dynamic ? Dynamic : SizeAtCompileTime-1
     };
     typedef Block<const Derived,
                   internal::traits<Derived>::ColsAtCompileTime==1 ? SizeMinusOne : 1,
                   internal::traits<Derived>::ColsAtCompileTime==1 ? 1 : SizeMinusOne> ConstStartMinusOne;
-    typedef EIGEN_EXPR_BINARYOP_SCALAR_RETURN_TYPE(ConstStartMinusOne,Scalar,quotient) HNormalizedReturnType;
-    EIGEN_DEVICE_FUNC
-    inline const HNormalizedReturnType hnormalized() const;
+    typedef CwiseUnaryOp<internal::scalar_quotient1_op<typename internal::traits<Derived>::Scalar>,
+                const ConstStartMinusOne > HNormalizedReturnType;
+
+    const HNormalizedReturnType hnormalized() const;
 
 ////////// Householder module ///////////
 
-    EIGEN_DEVICE_FUNC
     void makeHouseholderInPlace(Scalar& tau, RealScalar& beta);
     template<typename EssentialPart>
-    EIGEN_DEVICE_FUNC
     void makeHouseholder(EssentialPart& essential,
                          Scalar& tau, RealScalar& beta) const;
     template<typename EssentialPart>
-    EIGEN_DEVICE_FUNC
     void applyHouseholderOnTheLeft(const EssentialPart& essential,
                                    const Scalar& tau,
                                    Scalar* workspace);
     template<typename EssentialPart>
-    EIGEN_DEVICE_FUNC
     void applyHouseholderOnTheRight(const EssentialPart& essential,
                                     const Scalar& tau,
                                     Scalar* workspace);
@@ -438,56 +442,65 @@ template<typename Derived> class MatrixBase
 ///////// Jacobi module /////////
 
     template<typename OtherScalar>
-    EIGEN_DEVICE_FUNC
     void applyOnTheLeft(Index p, Index q, const JacobiRotation<OtherScalar>& j);
     template<typename OtherScalar>
-    EIGEN_DEVICE_FUNC
     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 internal::stem_function<Scalar>::type StemFunction;
-#define EIGEN_MATRIX_FUNCTION(ReturnType, Name, Description) \
-    /** \returns an expression of the matrix Description of \c *this. \brief This function requires the <a href="unsupported/group__MatrixFunctions__Module.html"> unsupported MatrixFunctions module</a>. To compute the coefficient-wise Description use ArrayBase::##Name . */ \
-    const ReturnType<Derived> Name() const;
-#define EIGEN_MATRIX_FUNCTION_1(ReturnType, Name, Description, Argument) \
-    /** \returns an expression of the matrix Description of \c *this. \brief This function requires the <a href="unsupported/group__MatrixFunctions__Module.html"> unsupported MatrixFunctions module</a>. To compute the coefficient-wise Description use ArrayBase::##Name . */ \
-    const ReturnType<Derived> Name(Argument) const;
-
-    EIGEN_MATRIX_FUNCTION(MatrixExponentialReturnValue, exp, exponential)
-    /** \brief Helper function for the <a href="unsupported/group__MatrixFunctions__Module.html"> unsupported MatrixFunctions module</a>.*/
+    const MatrixExponentialReturnValue<Derived> exp() const;
     const MatrixFunctionReturnValue<Derived> matrixFunction(StemFunction f) const;
-    EIGEN_MATRIX_FUNCTION(MatrixFunctionReturnValue, cosh, hyperbolic cosine)
-    EIGEN_MATRIX_FUNCTION(MatrixFunctionReturnValue, sinh, hyperbolic sine)
-#if EIGEN_HAS_CXX11_MATH
-    EIGEN_MATRIX_FUNCTION(MatrixFunctionReturnValue, atanh, inverse hyperbolic cosine)
-    EIGEN_MATRIX_FUNCTION(MatrixFunctionReturnValue, acosh, inverse hyperbolic cosine)
-    EIGEN_MATRIX_FUNCTION(MatrixFunctionReturnValue, asinh, inverse hyperbolic sine)
+    const MatrixFunctionReturnValue<Derived> cosh() const;
+    const MatrixFunctionReturnValue<Derived> sinh() const;
+    const MatrixFunctionReturnValue<Derived> cos() const;
+    const MatrixFunctionReturnValue<Derived> sin() const;
+    const MatrixSquareRootReturnValue<Derived> sqrt() const;
+    const MatrixLogarithmReturnValue<Derived> log() const;
+    const MatrixPowerReturnValue<Derived> pow(const RealScalar& p) 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<const Derived> start(Index size) const;
+    VectorBlock<Derived> end(Index size);
+    const VectorBlock<const Derived> end(Index size) const;
+    template<int Size> VectorBlock<Derived,Size> start();
+    template<int Size> const VectorBlock<const Derived,Size> start() const;
+    template<int Size> VectorBlock<Derived,Size> end();
+    template<int Size> const VectorBlock<const Derived,Size> end() const;
+
+    Minor<Derived> minor(Index row, Index col);
+    const Minor<Derived> minor(Index row, Index col) const;
 #endif
-    EIGEN_MATRIX_FUNCTION(MatrixFunctionReturnValue, cos, cosine)
-    EIGEN_MATRIX_FUNCTION(MatrixFunctionReturnValue, sin, sine)
-    EIGEN_MATRIX_FUNCTION(MatrixSquareRootReturnValue, sqrt, square root)
-    EIGEN_MATRIX_FUNCTION(MatrixLogarithmReturnValue, log, logarithm)
-    EIGEN_MATRIX_FUNCTION_1(MatrixPowerReturnValue,        pow, power to \c p, const RealScalar& p)
-    EIGEN_MATRIX_FUNCTION_1(MatrixComplexPowerReturnValue, pow, power to \c p, const std::complex<RealScalar>& p)
 
   protected:
-    EIGEN_DEFAULT_COPY_CONSTRUCTOR(MatrixBase)
-    EIGEN_DEFAULT_EMPTY_CONSTRUCTOR_AND_DESTRUCTOR(MatrixBase)
+    MatrixBase() : Base() {}
 
   private:
-    EIGEN_DEVICE_FUNC explicit MatrixBase(int);
-    EIGEN_DEVICE_FUNC MatrixBase(int,int);
-    template<typename OtherDerived> EIGEN_DEVICE_FUNC explicit MatrixBase(const MatrixBase<OtherDerived>&);
+    explicit MatrixBase(int);
+    MatrixBase(int,int);
+    template<typename OtherDerived> explicit MatrixBase(const MatrixBase<OtherDerived>&);
   protected:
     // mixing arrays and matrices is not legal
     template<typename OtherDerived> Derived& operator+=(const ArrayBase<OtherDerived>& )
@@ -497,51 +510,6 @@ template<typename Derived> class MatrixBase
     {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/NestByValue.h

@@ -13,28 +13,25 @@
 
 namespace Eigen {
 
-namespace internal {
-template<typename ExpressionType>
-struct traits<NestByValue<ExpressionType> > : public traits<ExpressionType>
-{
-  enum {
-    Flags = traits<ExpressionType>::Flags & ~NestByRefBit
-  };
-};
-}
-
 /** \class NestByValue
   * \ingroup Core_Module
   *
   * \brief Expression which must be nested by value
   *
-  * \tparam ExpressionType the type of the object of which we are requiring nesting-by-value
+  * \param ExpressionType the type of the object of which we are requiring nesting-by-value
   *
   * This class is the return type of MatrixBase::nestByValue()
   * and most of the time this is the only way it is used.
   *
   * \sa MatrixBase::nestByValue()
   */
+
+namespace internal {
+template<typename ExpressionType>
+struct traits<NestByValue<ExpressionType> > : public traits<ExpressionType>
+{};
+}
+
 template<typename ExpressionType> class NestByValue
   : public internal::dense_xpr_base< NestByValue<ExpressionType> >::type
 {
@@ -43,14 +40,58 @@ template<typename ExpressionType> class NestByValue
     typedef typename internal::dense_xpr_base<NestByValue>::type Base;
     EIGEN_DENSE_PUBLIC_INTERFACE(NestByValue)
 
-    EIGEN_DEVICE_FUNC explicit inline NestByValue(const ExpressionType& matrix) : m_expression(matrix) {}
+    inline NestByValue(const ExpressionType& matrix) : m_expression(matrix) {}
 
-    EIGEN_DEVICE_FUNC EIGEN_CONSTEXPR inline Index rows() const EIGEN_NOEXCEPT { return m_expression.rows(); }
-    EIGEN_DEVICE_FUNC EIGEN_CONSTEXPR inline Index cols() const EIGEN_NOEXCEPT { return m_expression.cols(); }
+    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 operator const ExpressionType&() const { return m_expression; }
+    inline const CoeffReturnType coeff(Index row, Index col) const
+    {
+      return m_expression.coeff(row, col);
+    }
 
-    EIGEN_DEVICE_FUNC const ExpressionType& nestedExpression() const { return m_expression; }
+    inline Scalar& coeffRef(Index row, Index col)
+    {
+      return m_expression.const_cast_derived().coeffRef(row, col);
+    }
+
+    inline const CoeffReturnType coeff(Index index) const
+    {
+      return m_expression.coeff(index);
+    }
+
+    inline Scalar& coeffRef(Index index)
+    {
+      return m_expression.const_cast_derived().coeffRef(index);
+    }
+
+    template<int LoadMode>
+    inline const PacketScalar packet(Index row, Index col) const
+    {
+      return m_expression.template packet<LoadMode>(row, col);
+    }
+
+    template<int LoadMode>
+    inline void writePacket(Index row, Index col, const PacketScalar& x)
+    {
+      m_expression.const_cast_derived().template writePacket<LoadMode>(row, col, x);
+    }
+
+    template<int LoadMode>
+    inline const PacketScalar packet(Index index) const
+    {
+      return m_expression.template packet<LoadMode>(index);
+    }
+
+    template<int LoadMode>
+    inline void writePacket(Index index, const PacketScalar& x)
+    {
+      m_expression.const_cast_derived().template writePacket<LoadMode>(index, x);
+    }
+
+    operator const ExpressionType&() const { return m_expression; }
 
   protected:
     const ExpressionType m_expression;
@@ -59,27 +100,12 @@ template<typename ExpressionType> class NestByValue
 /** \returns an expression of the temporary version of *this.
   */
 template<typename Derived>
-EIGEN_DEVICE_FUNC inline const NestByValue<Derived>
+inline const NestByValue<Derived>
 DenseBase<Derived>::nestByValue() const
 {
   return NestByValue<Derived>(derived());
 }
 
-namespace internal {
-
-// Evaluator of Solve -> eval into a temporary
-template<typename ArgType>
-struct evaluator<NestByValue<ArgType> >
-  : public evaluator<ArgType>
-{
-  typedef evaluator<ArgType> Base;
-
-  EIGEN_DEVICE_FUNC explicit evaluator(const NestByValue<ArgType>& xpr)
-    : Base(xpr.nestedExpression())
-  {}
-};
-}
-
 } // end namespace Eigen
 
 #endif // EIGEN_NESTBYVALUE_H

Eigen/src/Core/NoAlias.h

@@ -17,7 +17,7 @@ namespace Eigen {
   *
   * \brief Pseudo expression providing an operator = assuming no aliasing
   *
-  * \tparam ExpressionType the type of the object on which to do the lazy assignment
+  * \param ExpressionType the type of the object on which to do the lazy assignment
   *
   * This class represents an expression with special assignment operators
   * assuming no aliasing between the target expression and the source expression.
@@ -30,37 +30,62 @@ namespace Eigen {
 template<typename ExpressionType, template <typename> class StorageBase>
 class NoAlias
 {
-  public:
     typedef typename ExpressionType::Scalar Scalar;
-    
-    EIGEN_DEVICE_FUNC
-    explicit NoAlias(ExpressionType& expression) : m_expression(expression) {}
-    
+  public:
+    NoAlias(ExpressionType& expression) : m_expression(expression) {}
+
+    /** Behaves like MatrixBase::lazyAssign(other)
+      * \sa MatrixBase::lazyAssign() */
     template<typename OtherDerived>
-    EIGEN_DEVICE_FUNC
     EIGEN_STRONG_INLINE ExpressionType& operator=(const StorageBase<OtherDerived>& other)
-    {
-      call_assignment_no_alias(m_expression, other.derived(), internal::assign_op<Scalar,typename OtherDerived::Scalar>());
-      return m_expression;
-    }
-    
+    { return internal::assign_selector<ExpressionType,OtherDerived,false>::run(m_expression,other.derived()); }
+
+    /** \sa MatrixBase::operator+= */
     template<typename OtherDerived>
-    EIGEN_DEVICE_FUNC
     EIGEN_STRONG_INLINE ExpressionType& operator+=(const StorageBase<OtherDerived>& other)
     {
-      call_assignment_no_alias(m_expression, other.derived(), internal::add_assign_op<Scalar,typename OtherDerived::Scalar>());
-      return m_expression;
-    }
-    
-    template<typename OtherDerived>
-    EIGEN_DEVICE_FUNC
-    EIGEN_STRONG_INLINE ExpressionType& operator-=(const StorageBase<OtherDerived>& other)
-    {
-      call_assignment_no_alias(m_expression, other.derived(), internal::sub_assign_op<Scalar,typename OtherDerived::Scalar>());
+      typedef SelfCwiseBinaryOp<internal::scalar_sum_op<Scalar>, ExpressionType, OtherDerived> SelfAdder;
+      SelfAdder tmp(m_expression);
+      typedef typename internal::nested<OtherDerived>::type OtherDerivedNested;
+      typedef typename internal::remove_all<OtherDerivedNested>::type _OtherDerivedNested;
+      internal::assign_selector<SelfAdder,_OtherDerivedNested,false>::run(tmp,OtherDerivedNested(other.derived()));
       return m_expression;
     }
 
-    EIGEN_DEVICE_FUNC
+    /** \sa MatrixBase::operator-= */
+    template<typename OtherDerived>
+    EIGEN_STRONG_INLINE ExpressionType& operator-=(const StorageBase<OtherDerived>& other)
+    {
+      typedef SelfCwiseBinaryOp<internal::scalar_difference_op<Scalar>, ExpressionType, OtherDerived> SelfAdder;
+      SelfAdder tmp(m_expression);
+      typedef typename internal::nested<OtherDerived>::type OtherDerivedNested;
+      typedef typename internal::remove_all<OtherDerivedNested>::type _OtherDerivedNested;
+      internal::assign_selector<SelfAdder,_OtherDerivedNested,false>::run(tmp,OtherDerivedNested(other.derived()));
+      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()); }
+    
+    template<typename OtherDerived>
+    ExpressionType& operator=(const ReturnByValue<OtherDerived>& func)
+    { return m_expression = func; }
+#endif
+
     ExpressionType& expression() const
     {
       return m_expression;
@@ -75,10 +100,10 @@ class NoAlias
   *
   * More precisely, noalias() allows to bypass the EvalBeforeAssignBit flag.
   * Currently, even though several expressions may alias, only product
-  * expressions have this flag. Therefore, noalias() is only useful when
+  * expressions have this flag. Therefore, noalias() is only usefull when
   * the source expression contains a matrix product.
   *
-  * Here are some examples where noalias is useful:
+  * Here are some examples where noalias is usefull:
   * \code
   * D.noalias()  = A * B;
   * D.noalias() += A.transpose() * B;
@@ -99,9 +124,9 @@ class NoAlias
   * \sa class NoAlias
   */
 template<typename Derived>
-NoAlias<Derived,MatrixBase> EIGEN_DEVICE_FUNC MatrixBase<Derived>::noalias()
+NoAlias<Derived,MatrixBase> MatrixBase<Derived>::noalias()
 {
-  return NoAlias<Derived, Eigen::MatrixBase >(derived());
+  return derived();
 }
 
 } // end namespace Eigen

Eigen/src/Core/NumTraits.h

@@ -12,112 +12,24 @@
 
 namespace Eigen {
 
-namespace internal {
-
-// default implementation of digits10(), based on numeric_limits if specialized,
-// 0 for integer types, and log10(epsilon()) otherwise.
-template< typename T,
-          bool use_numeric_limits = std::numeric_limits<T>::is_specialized,
-          bool is_integer = NumTraits<T>::IsInteger>
-struct default_digits10_impl
-{
-  EIGEN_DEVICE_FUNC EIGEN_CONSTEXPR
-  static int run() { return std::numeric_limits<T>::digits10; }
-};
-
-template<typename T>
-struct default_digits10_impl<T,false,false> // Floating point
-{
-  EIGEN_DEVICE_FUNC EIGEN_CONSTEXPR
-  static int run() {
-    using std::log10;
-    using std::ceil;
-    typedef typename NumTraits<T>::Real Real;
-    return int(ceil(-log10(NumTraits<Real>::epsilon())));
-  }
-};
-
-template<typename T>
-struct default_digits10_impl<T,false,true> // Integer
-{
-  EIGEN_DEVICE_FUNC EIGEN_CONSTEXPR
-  static int run() { return 0; }
-};
-
-
-// default implementation of digits(), based on numeric_limits if specialized,
-// 0 for integer types, and log2(epsilon()) otherwise.
-template< typename T,
-          bool use_numeric_limits = std::numeric_limits<T>::is_specialized,
-          bool is_integer = NumTraits<T>::IsInteger>
-struct default_digits_impl
-{
-  EIGEN_DEVICE_FUNC EIGEN_CONSTEXPR
-  static int run() { return std::numeric_limits<T>::digits; }
-};
-
-template<typename T>
-struct default_digits_impl<T,false,false> // Floating point
-{
-  EIGEN_DEVICE_FUNC EIGEN_CONSTEXPR
-  static int run() {
-    using std::log;
-    using std::ceil;
-    typedef typename NumTraits<T>::Real Real;
-    return int(ceil(-log(NumTraits<Real>::epsilon())/log(static_cast<Real>(2))));
-  }
-};
-
-template<typename T>
-struct default_digits_impl<T,false,true> // Integer
-{
-  EIGEN_DEVICE_FUNC EIGEN_CONSTEXPR
-  static int run() { return 0; }
-};
-
-} // end namespace internal
-
-namespace numext {
-/** \internal bit-wise cast without changing the underlying bit representation. */
-
-// TODO: Replace by std::bit_cast (available in C++20)
-template <typename Tgt, typename Src>
-EIGEN_STRONG_INLINE EIGEN_DEVICE_FUNC Tgt bit_cast(const Src& src) {
-#if EIGEN_HAS_TYPE_TRAITS
-  // The behaviour of memcpy is not specified for non-trivially copyable types
-  EIGEN_STATIC_ASSERT(std::is_trivially_copyable<Src>::value, THIS_TYPE_IS_NOT_SUPPORTED);
-  EIGEN_STATIC_ASSERT(std::is_trivially_copyable<Tgt>::value && std::is_default_constructible<Tgt>::value,
-                      THIS_TYPE_IS_NOT_SUPPORTED);
-#endif
-
-  EIGEN_STATIC_ASSERT(sizeof(Src) == sizeof(Tgt), THIS_TYPE_IS_NOT_SUPPORTED);
-  Tgt tgt;
-  EIGEN_USING_STD(memcpy)
-  memcpy(&tgt, &src, sizeof(Tgt));
-  return tgt;
-}
-}  // namespace numext
-
 /** \class NumTraits
   * \ingroup Core_Module
   *
   * \brief Holds information about the various numeric (i.e. scalar) types allowed by Eigen.
   *
-  * \tparam T the numeric type at hand
+  * \param T the numeric type at hand
   *
   * This class stores enums, typedefs and static methods giving information about a numeric type.
   *
   * The provided data consists of:
-  * \li A typedef \c Real, giving the "real part" type of \a T. If \a T is already real,
-  *     then \c Real is just a typedef to \a T. If \a T is \c std::complex<U> then \c Real
+  * \li A typedef \a Real, giving the "real part" type of \a T. If \a T is already real,
+  *     then \a Real is just a typedef to \a T. If \a T is \c std::complex<U> then \a Real
   *     is a typedef to \a U.
-  * \li A typedef \c NonInteger, giving the type that should be used for operations producing non-integral values,
+  * \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 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 \c Literal giving the type to use for numeric literals such as "2" or "0.5". For instance, for \c std::complex<U>, Literal is defined as \c U.
-  *     Of course, this type must be fully compatible with \a T. In doubt, just use \a T here.
   * \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
   *     this means, just use \a T here.
   * \li An enum value \a IsComplex. It is equal to 1 if \a T is a \c std::complex
@@ -126,27 +38,14 @@ EIGEN_STRONG_INLINE EIGEN_DEVICE_FUNC Tgt bit_cast(const Src& src) {
   *     and to \c 0 otherwise.
   * \li Enum values ReadCost, AddCost and MulCost representing a rough estimate of the number of CPU cycles needed
   *     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. If you don't know what this means, just use \c Eigen::HugeCost.
+  *     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 <a href="http://en.cppreference.com/w/cpp/types/numeric_limits/epsilon">std::numeric_limits::epsilon()</a>,
-  *     it returns a \a Real instead of a \a T.
+  * \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.
   * \li highest() and lowest() functions returning the highest and lowest possible values respectively.
-  * \li digits() function returning the number of radix digits (non-sign digits for integers, mantissa for floating-point). This is
-  *     the analogue of <a href="http://en.cppreference.com/w/cpp/types/numeric_limits/digits">std::numeric_limits<T>::digits</a>
-  *     which is used as the default implementation if specialized.
-  * \li digits10() function returning the number of decimal digits that can be represented without change. This is
-  *     the analogue of <a href="http://en.cppreference.com/w/cpp/types/numeric_limits/digits10">std::numeric_limits<T>::digits10</a>
-  *     which is used as the default implementation if specialized.
-  * \li min_exponent() and max_exponent() functions returning the highest and lowest possible values, respectively,
-  *     such that the radix raised to the power exponent-1 is a normalized floating-point number.  These are equivalent to
-  *     <a href="http://en.cppreference.com/w/cpp/types/numeric_limits/min_exponent">std::numeric_limits<T>::min_exponent</a>/
-  *     <a href="http://en.cppreference.com/w/cpp/types/numeric_limits/max_exponent">std::numeric_limits<T>::max_exponent</a>.
-  * \li infinity() function returning a representation of positive infinity, if available.
-  * \li quiet_NaN function returning a non-signaling "not-a-number", if available.
   */
 
 template<typename T> struct GenericNumTraits
@@ -168,65 +67,22 @@ template<typename T> struct GenericNumTraits
                      T
                    >::type NonInteger;
   typedef T Nested;
-  typedef T Literal;
 
-  EIGEN_DEVICE_FUNC EIGEN_CONSTEXPR
-  static inline Real epsilon()
-  {
-    return numext::numeric_limits<T>::epsilon();
-  }
-
-  EIGEN_DEVICE_FUNC EIGEN_CONSTEXPR
-  static inline int digits10()
-  {
-    return internal::default_digits10_impl<T>::run();
-  }
-
-  EIGEN_DEVICE_FUNC EIGEN_CONSTEXPR
-  static inline int digits()
-  {
-    return internal::default_digits_impl<T>::run();
-  }
-
-  EIGEN_DEVICE_FUNC EIGEN_CONSTEXPR
-  static inline int min_exponent()
-  {
-    return numext::numeric_limits<T>::min_exponent;
-  }
-
-  EIGEN_DEVICE_FUNC EIGEN_CONSTEXPR
-  static inline int max_exponent()
-  {
-    return numext::numeric_limits<T>::max_exponent;
-  }
-
-  EIGEN_DEVICE_FUNC EIGEN_CONSTEXPR
+  static inline Real epsilon() { return std::numeric_limits<T>::epsilon(); }
   static inline Real dummy_precision()
   {
     // make sure to override this for floating-point types
     return Real(0);
   }
-
-  EIGEN_DEVICE_FUNC EIGEN_CONSTEXPR
-  static inline T highest() {
-    return (numext::numeric_limits<T>::max)();
-  }
-
-  EIGEN_DEVICE_FUNC EIGEN_CONSTEXPR
-  static inline T lowest()  {
-    return IsInteger ? (numext::numeric_limits<T>::min)()
-                     : static_cast<T>(-(numext::numeric_limits<T>::max)());
-  }
-
-  EIGEN_DEVICE_FUNC EIGEN_CONSTEXPR
-  static inline T infinity() {
-    return numext::numeric_limits<T>::infinity();
-  }
-
-  EIGEN_DEVICE_FUNC EIGEN_CONSTEXPR
-  static inline T quiet_NaN() {
-    return numext::numeric_limits<T>::quiet_NaN();
-  }
+  static inline T highest() { return (std::numeric_limits<T>::max)(); }
+  static inline T lowest()  { return IsInteger ? (std::numeric_limits<T>::min)() : (-(std::numeric_limits<T>::max)()); }
+  
+#ifdef EIGEN2_SUPPORT
+  enum {
+    HasFloatingPoint = !IsInteger
+  };
+  typedef NonInteger FloatingPoint;
+#endif
 };
 
 template<typename T> struct NumTraits : GenericNumTraits<T>
@@ -235,20 +91,17 @@ template<typename T> struct NumTraits : GenericNumTraits<T>
 template<> struct NumTraits<float>
   : GenericNumTraits<float>
 {
-  EIGEN_DEVICE_FUNC EIGEN_CONSTEXPR
   static inline float dummy_precision() { return 1e-5f; }
 };
 
 template<> struct NumTraits<double> : GenericNumTraits<double>
 {
-  EIGEN_DEVICE_FUNC EIGEN_CONSTEXPR
   static inline double dummy_precision() { return 1e-12; }
 };
 
 template<> struct NumTraits<long double>
   : GenericNumTraits<long double>
 {
-  EIGEN_CONSTEXPR
   static inline long double dummy_precision() { return 1e-15l; }
 };
 
@@ -256,7 +109,6 @@ template<typename _Real> struct NumTraits<std::complex<_Real> >
   : GenericNumTraits<std::complex<_Real> >
 {
   typedef _Real Real;
-  typedef typename NumTraits<_Real>::Literal Literal;
   enum {
     IsComplex = 1,
     RequireInitialization = NumTraits<_Real>::RequireInitialization,
@@ -265,12 +117,8 @@ template<typename _Real> struct NumTraits<std::complex<_Real> >
     MulCost = 4 * NumTraits<Real>::MulCost + 2 * NumTraits<Real>::AddCost
   };
 
-  EIGEN_DEVICE_FUNC EIGEN_CONSTEXPR
   static inline Real epsilon() { return NumTraits<Real>::epsilon(); }
-  EIGEN_DEVICE_FUNC EIGEN_CONSTEXPR
   static inline Real dummy_precision() { return NumTraits<Real>::dummy_precision(); }
-  EIGEN_DEVICE_FUNC EIGEN_CONSTEXPR
-  static inline int digits10() { return NumTraits<Real>::digits10(); }
 };
 
 template<typename Scalar, int Rows, int Cols, int Options, int MaxRows, int MaxCols>
@@ -282,54 +130,21 @@ struct NumTraits<Array<Scalar, Rows, Cols, Options, MaxRows, MaxCols> >
   typedef typename NumTraits<Scalar>::NonInteger NonIntegerScalar;
   typedef Array<NonIntegerScalar, Rows, Cols, Options, MaxRows, MaxCols> NonInteger;
   typedef ArrayType & Nested;
-  typedef typename NumTraits<Scalar>::Literal Literal;
-
+  
   enum {
     IsComplex = NumTraits<Scalar>::IsComplex,
     IsInteger = NumTraits<Scalar>::IsInteger,
     IsSigned  = NumTraits<Scalar>::IsSigned,
     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
+    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
   };
-
-  EIGEN_DEVICE_FUNC EIGEN_CONSTEXPR
+  
   static inline RealScalar epsilon() { return NumTraits<RealScalar>::epsilon(); }
-  EIGEN_DEVICE_FUNC EIGEN_CONSTEXPR
   static inline RealScalar dummy_precision() { return NumTraits<RealScalar>::dummy_precision(); }
-
-  EIGEN_CONSTEXPR
-  static inline int digits10() { return NumTraits<Scalar>::digits10(); }
 };
 
-template<> struct NumTraits<std::string>
-  : GenericNumTraits<std::string>
-{
-  enum {
-    RequireInitialization = 1,
-    ReadCost = HugeCost,
-    AddCost  = HugeCost,
-    MulCost  = HugeCost
-  };
-
-  EIGEN_CONSTEXPR
-  static inline int digits10() { return 0; }
-
-private:
-  static inline std::string epsilon();
-  static inline std::string dummy_precision();
-  static inline std::string lowest();
-  static inline std::string highest();
-  static inline std::string infinity();
-  static inline std::string quiet_NaN();
-};
-
-// Empty specialization for void to allow template specialization based on NumTraits<T>::Real with T==void and SFINAE.
-template<> struct NumTraits<void> {};
-
-template<> struct NumTraits<bool> : GenericNumTraits<bool> {};
-
 } // end namespace Eigen
 
 #endif // EIGEN_NUMTRAITS_H

Eigen/src/Core/PartialReduxEvaluator.h

@@ -1,232 +0,0 @@
-// This file is part of Eigen, a lightweight C++ template library
-// for linear algebra.
-//
-// Copyright (C) 2011-2018 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_PARTIALREDUX_H
-#define EIGEN_PARTIALREDUX_H
-
-namespace Eigen { 
-
-namespace internal {
-
-
-/***************************************************************************
-*
-* This file provides evaluators for partial reductions.
-* There are two modes:
-*
-*  - scalar path: simply calls the respective function on the column or row.
-*    -> nothing special here, all the tricky part is handled by the return
-*       types of VectorwiseOp's members. They embed the functor calling the
-*       respective DenseBase's member function.
-*
-*  - vectorized path: implements a packet-wise reductions followed by
-*    some (optional) processing of the outcome, e.g., division by n for mean.
-*
-* For the vectorized path let's observe that the packet-size and outer-unrolling
-* are both decided by the assignement logic. So all we have to do is to decide
-* on the inner unrolling.
-*
-* For the unrolling, we can reuse "internal::redux_vec_unroller" from Redux.h,
-* but be need to be careful to specify correct increment.
-*
-***************************************************************************/
-
-
-/* logic deciding a strategy for unrolling of vectorized paths */
-template<typename Func, typename Evaluator>
-struct packetwise_redux_traits
-{
-  enum {
-    OuterSize = int(Evaluator::IsRowMajor) ? Evaluator::RowsAtCompileTime : Evaluator::ColsAtCompileTime,
-    Cost = OuterSize == Dynamic ? HugeCost
-         : OuterSize * Evaluator::CoeffReadCost + (OuterSize-1) * functor_traits<Func>::Cost,
-    Unrolling = Cost <= EIGEN_UNROLLING_LIMIT ? CompleteUnrolling : NoUnrolling
-  };
-
-};
-
-/* Value to be returned when size==0 , by default let's return 0 */
-template<typename PacketType,typename Func>
-EIGEN_DEVICE_FUNC
-PacketType packetwise_redux_empty_value(const Func& ) { return pset1<PacketType>(0); }
-
-/* For products the default is 1 */
-template<typename PacketType,typename Scalar>
-EIGEN_DEVICE_FUNC
-PacketType packetwise_redux_empty_value(const scalar_product_op<Scalar,Scalar>& ) { return pset1<PacketType>(1); }
-
-/* Perform the actual reduction */
-template<typename Func, typename Evaluator,
-         int Unrolling = packetwise_redux_traits<Func, Evaluator>::Unrolling
->
-struct packetwise_redux_impl;
-
-/* Perform the actual reduction with unrolling */
-template<typename Func, typename Evaluator>
-struct packetwise_redux_impl<Func, Evaluator, CompleteUnrolling>
-{
-  typedef redux_novec_unroller<Func,Evaluator, 0, Evaluator::SizeAtCompileTime> Base;
-  typedef typename Evaluator::Scalar Scalar;
-
-  template<typename PacketType>
-  EIGEN_DEVICE_FUNC static EIGEN_STRONG_INLINE
-  PacketType run(const Evaluator &eval, const Func& func, Index /*size*/)
-  {
-    return redux_vec_unroller<Func, Evaluator, 0, packetwise_redux_traits<Func, Evaluator>::OuterSize>::template run<PacketType>(eval,func);
-  }
-};
-
-/* Add a specialization of redux_vec_unroller for size==0 at compiletime.
- * This specialization is not required for general reductions, which is
- * why it is defined here.
- */
-template<typename Func, typename Evaluator, int Start>
-struct redux_vec_unroller<Func, Evaluator, Start, 0>
-{
-  template<typename PacketType>
-  EIGEN_DEVICE_FUNC
-  static EIGEN_STRONG_INLINE PacketType run(const Evaluator &, const Func& f)
-  {
-    return packetwise_redux_empty_value<PacketType>(f);
-  }
-};
-
-/* Perform the actual reduction for dynamic sizes */
-template<typename Func, typename Evaluator>
-struct packetwise_redux_impl<Func, Evaluator, NoUnrolling>
-{
-  typedef typename Evaluator::Scalar Scalar;
-  typedef typename redux_traits<Func, Evaluator>::PacketType PacketScalar;
-
-  template<typename PacketType>
-  EIGEN_DEVICE_FUNC
-  static PacketType run(const Evaluator &eval, const Func& func, Index size)
-  {
-    if(size==0)
-      return packetwise_redux_empty_value<PacketType>(func);
-    
-    const Index size4 = (size-1)&(~3);
-    PacketType p = eval.template packetByOuterInner<Unaligned,PacketType>(0,0);
-    Index i = 1;
-    // This loop is optimized for instruction pipelining:
-    // - each iteration generates two independent instructions
-    // - thanks to branch prediction and out-of-order execution we have independent instructions across loops
-    for(; i<size4; i+=4)
-      p = func.packetOp(p,
-            func.packetOp(
-              func.packetOp(eval.template packetByOuterInner<Unaligned,PacketType>(i+0,0),eval.template packetByOuterInner<Unaligned,PacketType>(i+1,0)),
-              func.packetOp(eval.template packetByOuterInner<Unaligned,PacketType>(i+2,0),eval.template packetByOuterInner<Unaligned,PacketType>(i+3,0))));
-    for(; i<size; ++i)
-      p = func.packetOp(p, eval.template packetByOuterInner<Unaligned,PacketType>(i,0));
-    return p;
-  }
-};
-
-template< typename ArgType, typename MemberOp, int Direction>
-struct evaluator<PartialReduxExpr<ArgType, MemberOp, Direction> >
-  : evaluator_base<PartialReduxExpr<ArgType, MemberOp, Direction> >
-{
-  typedef PartialReduxExpr<ArgType, MemberOp, Direction> XprType;
-  typedef typename internal::nested_eval<ArgType,1>::type ArgTypeNested;
-  typedef typename internal::add_const_on_value_type<ArgTypeNested>::type ConstArgTypeNested;
-  typedef typename internal::remove_all<ArgTypeNested>::type ArgTypeNestedCleaned;
-  typedef typename ArgType::Scalar InputScalar;
-  typedef typename XprType::Scalar Scalar;
-  enum {
-    TraversalSize = Direction==int(Vertical) ? int(ArgType::RowsAtCompileTime) :  int(ArgType::ColsAtCompileTime)
-  };
-  typedef typename MemberOp::template Cost<int(TraversalSize)> CostOpType;
-  enum {
-    CoeffReadCost = TraversalSize==Dynamic ? HugeCost
-                  : TraversalSize==0 ? 1
-                  : TraversalSize * evaluator<ArgType>::CoeffReadCost + int(CostOpType::value),
-    
-    _ArgFlags = evaluator<ArgType>::Flags,
-
-    _Vectorizable =  bool(int(_ArgFlags)&PacketAccessBit)
-                  && bool(MemberOp::Vectorizable)
-                  && (Direction==int(Vertical) ? bool(_ArgFlags&RowMajorBit) : (_ArgFlags&RowMajorBit)==0)
-                  && (TraversalSize!=0),
-                  
-    Flags = (traits<XprType>::Flags&RowMajorBit)
-          | (evaluator<ArgType>::Flags&(HereditaryBits&(~RowMajorBit)))
-          | (_Vectorizable ? PacketAccessBit : 0)
-          | LinearAccessBit,
-    
-    Alignment = 0 // FIXME this will need to be improved once PartialReduxExpr is vectorized
-  };
-
-  EIGEN_DEVICE_FUNC explicit evaluator(const XprType xpr)
-    : m_arg(xpr.nestedExpression()), m_functor(xpr.functor())
-  {
-    EIGEN_INTERNAL_CHECK_COST_VALUE(TraversalSize==Dynamic ? HugeCost : (TraversalSize==0 ? 1 : int(CostOpType::value)));
-    EIGEN_INTERNAL_CHECK_COST_VALUE(CoeffReadCost);
-  }
-
-  typedef typename XprType::CoeffReturnType CoeffReturnType;
-
-  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
-  const Scalar coeff(Index i, Index j) const
-  {
-    return coeff(Direction==Vertical ? j : i);
-  }
-
-  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
-  const Scalar coeff(Index index) const
-  {
-    return m_functor(m_arg.template subVector<DirectionType(Direction)>(index));
-  }
-
-  template<int LoadMode,typename PacketType>
-  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
-  PacketType packet(Index i, Index j) const
-  {
-    return packet<LoadMode,PacketType>(Direction==Vertical ? j : i);
-  }
-  
-  template<int LoadMode,typename PacketType>
-  EIGEN_STRONG_INLINE EIGEN_DEVICE_FUNC
-  PacketType packet(Index idx) const
-  {
-    enum { PacketSize = internal::unpacket_traits<PacketType>::size };
-    typedef Block<const ArgTypeNestedCleaned,
-                  Direction==Vertical ? int(ArgType::RowsAtCompileTime) : int(PacketSize),
-                  Direction==Vertical ? int(PacketSize) : int(ArgType::ColsAtCompileTime),
-                  true /* InnerPanel */> PanelType;
-    
-    PanelType panel(m_arg,
-                    Direction==Vertical ? 0 : idx,
-                    Direction==Vertical ? idx : 0,
-                    Direction==Vertical ? m_arg.rows() : Index(PacketSize),
-                    Direction==Vertical ? Index(PacketSize) : m_arg.cols());
-
-    // FIXME
-    // See bug 1612, currently if PacketSize==1 (i.e. complex<double> with 128bits registers) then the storage-order of panel get reversed
-    // and methods like packetByOuterInner do not make sense anymore in this context.
-    // So let's just by pass "vectorization" in this case:
-    if(PacketSize==1)
-      return internal::pset1<PacketType>(coeff(idx));
-    
-    typedef typename internal::redux_evaluator<PanelType> PanelEvaluator;
-    PanelEvaluator panel_eval(panel);
-    typedef typename MemberOp::BinaryOp BinaryOp;
-    PacketType p = internal::packetwise_redux_impl<BinaryOp,PanelEvaluator>::template run<PacketType>(panel_eval,m_functor.binaryFunc(),m_arg.outerSize());
-    return p;
-  }
-
-protected:
-  ConstArgTypeNested m_arg;
-  const MemberOp m_functor;
-};
-
-} // end namespace internal
-
-} // end namespace Eigen
-
-#endif // EIGEN_PARTIALREDUX_H