Add CI to build docs

(grafted from 94e8d8902f
)

.gitignore

@@ -0,0 +1,41 @@
+qrc_*cxx
+*.orig
+*.pyc
+*.diff
+diff
+*.save
+save
+*.old
+*.gmo
+*.qm
+core
+core.*
+*.bak
+*~
+*.build*
+*.moc.*
+*.moc
+ui_*
+CMakeCache.txt
+tags
+.*.swp
+activity.png
+*.out
+*.php*
+*.log
+*.orig
+*.rej
+log
+patch
+*.patch
+a
+a.*
+lapack/testing
+lapack/reference
+.*project
+.settings
+Makefile
+!ci/build.gitlab-ci.yml
+!scripts/buildtests.in
+!Eigen/Core
+!Eigen/src/Core

.gitlab-ci.yml

@@ -0,0 +1,28 @@
+# This file is part of Eigen, a lightweight C++ template library
+# for linear algebra.
+#
+# Copyright (C) 2023, The Eigen Authors
+#
+# 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:
+  - build
+  - deploy
+
+variables:
+  # CMake build directory.
+  EIGEN_CI_BUILDDIR: .build
+  # Specify the CMake build target.
+  EIGEN_CI_BUILD_TARGET: ""
+  # If a test regex is specified, that will be selected.
+  # Otherwise, we will try a label if specified.
+  EIGEN_CI_CTEST_REGEX: ""
+  EIGEN_CI_CTEST_LABEL: ""
+  EIGEN_CI_CTEST_ARGS: ""
+
+include:
+  - "/ci/common.gitlab-ci.yml"
+  - "/ci/build.linux.gitlab-ci.yml"
+  - "/ci/deploy.gitlab-ci.yml"

.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

@@ -30,5 +30,3 @@ log
 patch
 a
 a.*
-lapack/testing
-lapack/reference

CMakeLists.txt

@@ -1,5 +1,4 @@
-project(Eigen3)
-
+project(Eigen)
 cmake_minimum_required(VERSION 2.8.5)
 
 # guard against in-source builds
@@ -8,11 +7,6 @@ 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)
@@ -98,11 +92,9 @@ 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 ON)
 
@@ -116,8 +108,7 @@ 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 "-" "" SFLAG1 ${FLAG})
-  string(REGEX REPLACE "\\+" "p" SFLAG ${SFLAG1})
+  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}")
@@ -126,13 +117,18 @@ 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")
@@ -146,18 +142,6 @@ 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")
-  
-  # -Wshadow is insanely too strict with gcc, hopefully it will become usable with gcc 6
-  # if(NOT CMAKE_COMPILER_IS_GNUCXX OR (CMAKE_CXX_COMPILER_VERSION VERSION_GREATER "5.0.0"))
-  if(NOT CMAKE_COMPILER_IS_GNUCXX)
-    ei_add_cxx_compiler_flag("-Wshadow")
-  endif()
   
   ei_add_cxx_compiler_flag("-Wno-psabi")
   ei_add_cxx_compiler_flag("-Wno-variadic-macros")
@@ -167,8 +151,7 @@ if(NOT MSVC)
   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
@@ -180,11 +163,6 @@ if(NOT MSVC)
   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 "")
 
@@ -218,65 +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_AVX512 "Enable/Disable AVX512 in tests/examples" OFF)
-  if(EIGEN_TEST_AVX512)
-    set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -mavx512f -fabi-version=6 -DEIGEN_ENABLE_AVX512")
-    message(STATUS "Enabling AVX512 in tests/examples")
-  endif()
-
-  option(EIGEN_TEST_F16C "Enable/Disable F16C in tests/examples" OFF)
-  if(EIGEN_TEST_F16C)
-    set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -mf16c")
-    message(STATUS "Enabling F16C in tests/examples")
-  endif()
-
   option(EIGEN_TEST_ALTIVEC "Enable/Disable AltiVec in tests/examples" OFF)
   if(EIGEN_TEST_ALTIVEC)
     set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -maltivec -mabi=altivec")
     message(STATUS "Enabling AltiVec in tests/examples")
   endif()
 
-  option(EIGEN_TEST_VSX "Enable/Disable VSX in tests/examples" OFF)
-  if(EIGEN_TEST_VSX)
-    set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -m64 -mvsx")
-    message(STATUS "Enabling VSX in tests/examples")
-  endif()
-
   option(EIGEN_TEST_NEON "Enable/Disable Neon in tests/examples" OFF)
   if(EIGEN_TEST_NEON)
-    if(EIGEN_TEST_FMA)
-      set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -mfpu=neon-vfpv4")
-    else()
-      set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -mfpu=neon")
-    endif()
-    set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -mfloat-abi=hard")
+    set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -mfpu=neon -mcpu=cortex-a8")
     message(STATUS "Enabling NEON in tests/examples")
   endif()
 
-  option(EIGEN_TEST_NEON64 "Enable/Disable Neon in tests/examples" OFF)
-  if(EIGEN_TEST_NEON64)
-    set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS}")
-    message(STATUS "Enabling NEON in tests/examples")
-  endif()
-
-  option(EIGEN_TEST_ZVECTOR "Enable/Disable S390X(zEC13) ZVECTOR in tests/examples" OFF)
-  if(EIGEN_TEST_ZVECTOR)
-    set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -march=z13 -mzvector")
-    message(STATUS "Enabling S390X(zEC13) ZVECTOR in tests/examples")
-  endif()
-
   check_cxx_compiler_flag("-fopenmp" COMPILER_SUPPORT_OPENMP)
   if(COMPILER_SUPPORT_OPENMP)
     option(EIGEN_TEST_OPENMP "Enable/Disable OpenMP in tests/examples" OFF)
@@ -353,23 +284,11 @@ 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_CXX11 "Enable testing with C++11 and C++11 features (e.g. Tensor module)." OFF)
-
-set(EIGEN_CUDA_COMPUTE_ARCH 30 CACHE STRING "The CUDA compute architecture level to target when compiling CUDA code")
+option(EIGEN_TEST_C++0x "Enables all C++0x features." OFF)
 
 include_directories(${CMAKE_CURRENT_SOURCE_DIR} ${CMAKE_CURRENT_BINARY_DIR})
 
 # Backward compatibility support for EIGEN_INCLUDE_INSTALL_DIR
-if(EIGEN_INCLUDE_INSTALL_DIR)
-  message(WARNING "EIGEN_INCLUDE_INSTALL_DIR is deprecated. Use INCLUDE_INSTALL_DIR instead.")
-endif()
-
 if(EIGEN_INCLUDE_INSTALL_DIR AND NOT INCLUDE_INSTALL_DIR)
   set(INCLUDE_INSTALL_DIR ${EIGEN_INCLUDE_INSTALL_DIR}
       CACHE PATH "The directory relative to CMAKE_PREFIX_PATH where Eigen header files are installed")
@@ -379,8 +298,9 @@ else()
       CACHE PATH "The directory relative to CMAKE_PREFIX_PATH where Eigen header files are installed"
       )
 endif()
+
 set(CMAKEPACKAGE_INSTALL_DIR
-    "${CMAKE_INSTALL_DATADIR}/eigen3/cmake"
+    "${CMAKE_INSTALL_LIBDIR}/cmake/eigen3"
     CACHE PATH "The directory relative to CMAKE_PREFIX_PATH where Eigen3Config.cmake is installed"
     )
 set(PKGCONFIG_INSTALL_DIR
@@ -388,7 +308,6 @@ set(PKGCONFIG_INSTALL_DIR
     CACHE PATH "The directory relative to CMAKE_PREFIX_PATH where eigen3.pc is installed"
     )
 
-
 # similar to set_target_properties but append the property instead of overwriting it
 macro(ei_add_target_property target prop value)
 
@@ -410,7 +329,7 @@ if(EIGEN_BUILD_PKGCONFIG)
     install(FILES ${CMAKE_CURRENT_BINARY_DIR}/eigen3.pc
         DESTINATION ${PKGCONFIG_INSTALL_DIR}
         )
-endif()
+endif(EIGEN_BUILD_PKGCONFIG)
 
 add_subdirectory(Eigen)
 
@@ -436,13 +355,6 @@ else()
   add_subdirectory(lapack EXCLUDE_FROM_ALL)
 endif()
 
-# add SYCL
-option(EIGEN_TEST_SYCL "Add Sycl support." OFF)
-if(EIGEN_TEST_SYCL)
-  set (CMAKE_MODULE_PATH "${CMAKE_ROOT}/Modules" "cmake/Modules/" "${CMAKE_MODULE_PATH}")
-  include(FindComputeCpp)
-endif()
-
 add_subdirectory(unsupported)
 
 add_subdirectory(demos EXCLUDE_FROM_ALL)
@@ -491,7 +403,6 @@ if(cmake_generator_tolower MATCHES "makefile")
   message(STATUS "make check    | Build and run the unit-tests. Read this page:")
   message(STATUS "              |   http://eigen.tuxfamily.org/index.php?title=Tests")
   message(STATUS "make blas     | Build BLAS library (not the same thing as Eigen)")
-  message(STATUS "make uninstall| Removes files installed by make install")
   message(STATUS "--------------+--------------------------------------------------------------")
 else()
   message(STATUS "To build/run the unit tests, read this page:")
@@ -499,98 +410,3 @@ else()
 endif()
 
 message(STATUS "")
-
-
-set ( EIGEN_VERSION_STRING ${EIGEN_VERSION_NUMBER} )
-set ( EIGEN_VERSION_MAJOR  ${EIGEN_WORLD_VERSION} )
-set ( EIGEN_VERSION_MINOR  ${EIGEN_MAJOR_VERSION} )
-set ( EIGEN_VERSION_PATCH  ${EIGEN_MINOR_VERSION} )
-set ( EIGEN_DEFINITIONS "")
-set ( EIGEN_INCLUDE_DIR "${CMAKE_INSTALL_PREFIX}/${INCLUDE_INSTALL_DIR}" )
-set ( EIGEN_ROOT_DIR ${CMAKE_INSTALL_PREFIX} )
-
-# Interface libraries require at least CMake 3.0
-if (NOT CMAKE_VERSION VERSION_LESS 3.0)
-  include (CMakePackageConfigHelpers)
-
-  # Imported target support
-  add_library (eigen INTERFACE)
-
-  target_compile_definitions (eigen INTERFACE ${EIGEN_DEFINITIONS})
-  target_include_directories (eigen INTERFACE
-    $<BUILD_INTERFACE:${CMAKE_CURRENT_SOURCE_DIR}>
-    $<INSTALL_INTERFACE:${INCLUDE_INSTALL_DIR}>
-  )
-
-  # Export as title case Eigen
-  set_target_properties (eigen PROPERTIES EXPORT_NAME Eigen)
-
-  install (TARGETS eigen EXPORT Eigen3Targets)
-
-  configure_package_config_file (
-    ${CMAKE_CURRENT_SOURCE_DIR}/cmake/Eigen3Config.cmake.in
-    ${CMAKE_CURRENT_BINARY_DIR}/Eigen3Config.cmake
-    PATH_VARS EIGEN_INCLUDE_DIR EIGEN_ROOT_DIR
-    INSTALL_DESTINATION ${CMAKEPACKAGE_INSTALL_DIR}
-    NO_CHECK_REQUIRED_COMPONENTS_MACRO # Eigen does not provide components
-  )
-  # Remove CMAKE_SIZEOF_VOID_P from Eigen3ConfigVersion.cmake since Eigen does
-  # not depend on architecture specific settings or libraries. More
-  # specifically, an Eigen3Config.cmake generated from a 64 bit target can be
-  # used for 32 bit targets as well (and vice versa).
-  set (_Eigen3_CMAKE_SIZEOF_VOID_P ${CMAKE_SIZEOF_VOID_P})
-  unset (CMAKE_SIZEOF_VOID_P)
-  write_basic_package_version_file (Eigen3ConfigVersion.cmake
-    VERSION ${EIGEN_VERSION_NUMBER} COMPATIBILITY SameMajorVersion)
-  set (CMAKE_SIZEOF_VOID_P ${_Eigen3_CMAKE_SIZEOF_VOID_P})
-
-  # The Eigen target will be located in the Eigen3 namespace. Other CMake
-  # targets can refer to it using Eigen3::Eigen.
-  export (TARGETS eigen NAMESPACE Eigen3:: FILE Eigen3Targets.cmake)
-  # Export Eigen3 package to CMake registry such that it can be easily found by
-  # CMake even if it has not been installed to a standard directory.
-  export (PACKAGE Eigen3)
-
-  install (EXPORT Eigen3Targets NAMESPACE Eigen3:: DESTINATION
-    ${CMAKEPACKAGE_INSTALL_DIR})
-  install (FILES
-    ${CMAKE_CURRENT_BINARY_DIR}/Eigen3Config.cmake
-    ${CMAKE_CURRENT_BINARY_DIR}/Eigen3ConfigVersion.cmake
-    ${CMAKE_CURRENT_SOURCE_DIR}/cmake/UseEigen3.cmake
-    DESTINATION ${CMAKEPACKAGE_INSTALL_DIR})
-else (NOT CMAKE_VERSION VERSION_LESS 3.0)
-  # Fallback to legacy Eigen3Config.cmake without the imported target
-  
-  # If CMakePackageConfigHelpers module is available (CMake >= 2.8.8)
-  # create a relocatable Config file, otherwise leave the hardcoded paths       
-  include(CMakePackageConfigHelpers OPTIONAL RESULT_VARIABLE CPCH_PATH)
-  
-  if(CPCH_PATH)
-    configure_package_config_file (
-      ${CMAKE_CURRENT_SOURCE_DIR}/cmake/Eigen3ConfigLegacy.cmake.in
-      ${CMAKE_CURRENT_BINARY_DIR}/Eigen3Config.cmake
-      PATH_VARS EIGEN_INCLUDE_DIR EIGEN_ROOT_DIR
-      INSTALL_DESTINATION ${CMAKEPACKAGE_INSTALL_DIR}
-      NO_CHECK_REQUIRED_COMPONENTS_MACRO # Eigen does not provide components
-    )
-  else() 
-    # The PACKAGE_* variables are defined by the configure_package_config_file
-    # but without it we define them manually to the hardcoded paths
-    set(PACKAGE_INIT "")
-    set(PACKAGE_EIGEN_INCLUDE_DIR ${EIGEN_INCLUDE_DIR})
-    set(PACKAGE_EIGEN_ROOT_DIR ${EIGEN_ROOT_DIR})
-    configure_file ( ${CMAKE_CURRENT_SOURCE_DIR}/cmake/Eigen3ConfigLegacy.cmake.in
-                     ${CMAKE_CURRENT_BINARY_DIR}/Eigen3Config.cmake
-                     @ONLY ESCAPE_QUOTES
-                   )
-  endif()
-
-  install ( FILES ${CMAKE_CURRENT_SOURCE_DIR}/cmake/UseEigen3.cmake
-                  ${CMAKE_CURRENT_BINARY_DIR}/Eigen3Config.cmake
-            DESTINATION ${CMAKEPACKAGE_INSTALL_DIR}
-          )
-endif (NOT CMAKE_VERSION VERSION_LESS 3.0)
-
-# Add uninstall target
-add_custom_target ( uninstall
-    COMMAND ${CMAKE_COMMAND} -P ${CMAKE_CURRENT_SOURCE_DIR}/cmake/EigenUninstall.cmake)

CTestConfig.cmake

@@ -4,10 +4,10 @@
 ## # The following are required to uses Dart and the Cdash dashboard
 ##   ENABLE_TESTING()
 ##   INCLUDE(CTest)
-set(CTEST_PROJECT_NAME "Eigen3.3")
+set(CTEST_PROJECT_NAME "Eigen3.2")
 set(CTEST_NIGHTLY_START_TIME "00:00:00 UTC")
 
 set(CTEST_DROP_METHOD "http")
 set(CTEST_DROP_SITE "manao.inria.fr")
-set(CTEST_DROP_LOCATION "/CDash/submit.php?project=Eigen3.3")
+set(CTEST_DROP_LOCATION "/CDash/submit.php?project=Eigen3.2")
 set(CTEST_DROP_SITE_CDASH TRUE)

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

@@ -16,4 +16,4 @@ install(FILES
   DESTINATION ${INCLUDE_INSTALL_DIR}/Eigen COMPONENT Devel
   )
 
-install(DIRECTORY src DESTINATION ${INCLUDE_INSTALL_DIR}/Eigen COMPONENT Devel FILES_MATCHING PATTERN "*.h")
+add_subdirectory(src)

Eigen/Cholesky

@@ -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_CHOLESKY_MODULE_H
 #define EIGEN_CHOLESKY_MODULE_H
 
@@ -17,22 +10,20 @@
   *
   *
   * 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
-#include "src/misc/lapacke.h"
-#include "src/Cholesky/LLT_LAPACKE.h"
+#include "src/Cholesky/LLT_MKL.h"
 #endif
 
 #include "src/Core/util/ReenableStupidWarnings.h"

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
 
@@ -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

@@ -14,58 +14,6 @@
 // first thing Eigen does: stop the compiler from committing suicide
 #include "src/Core/util/DisableStupidWarnings.h"
 
-// Handle NVCC/CUDA/SYCL
-#if defined(__CUDACC__) || defined(__SYCL_DEVICE_ONLY__)
-  // Do not try asserts on CUDA and SYCL!
-  #ifndef EIGEN_NO_DEBUG
-  #define EIGEN_NO_DEBUG
-  #endif
-
-  #ifdef EIGEN_INTERNAL_DEBUGGING
-  #undef EIGEN_INTERNAL_DEBUGGING
-  #endif
-
-  #ifdef EIGEN_EXCEPTIONS
-  #undef EIGEN_EXCEPTIONS
-  #endif
-
-  // All functions callable from CUDA code must be qualified with __device__
-  #ifdef __CUDACC__
-    // Do not try to vectorize on CUDA and SYCL!
-    #ifndef EIGEN_DONT_VECTORIZE
-    #define EIGEN_DONT_VECTORIZE
-    #endif
-
-    #define EIGEN_DEVICE_FUNC __host__ __device__
-    // We need math_functions.hpp to ensure that that EIGEN_USING_STD_MATH macro
-    // works properly on the device side
-    #include <math_functions.hpp>
-  #else
-    #define EIGEN_DEVICE_FUNC
-  #endif
-
-#else
-  #define EIGEN_DEVICE_FUNC
-
-#endif
-
-// When compiling CUDA device code with NVCC, pull in math functions from the
-// global namespace.  In host mode, and when device doee with clang, use the
-// std versions.
-#if defined(__CUDA_ARCH__) && defined(__NVCC__)
-  #define EIGEN_USING_STD_MATH(FUNC) using ::FUNC;
-#else
-  #define EIGEN_USING_STD_MATH(FUNC) using std::FUNC;
-#endif
-
-#if (defined(_CPPUNWIND) || defined(__EXCEPTIONS)) && !defined(__CUDA_ARCH__) && !defined(EIGEN_EXCEPTIONS) && !defined(EIGEN_USE_SYCL)
-  #define EIGEN_EXCEPTIONS
-#endif
-
-#ifdef EIGEN_EXCEPTIONS
-  #include <new>
-#endif
-
 // then include this file where all our macros are defined. It's really important to do it first because
 // it's where we do all the alignment settings (platform detection and honoring the user's will if he
 // defined e.g. EIGEN_DONT_ALIGN) so it needs to be done before we do anything with vectorization.
@@ -73,7 +21,7 @@
 
 // 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)
+#if defined(__MINGW32__) && EIGEN_GNUC_AT_LEAST(4,6)
   #pragma GCC optimize ("-fno-ipa-cp-clone")
 #endif
 
@@ -83,26 +31,26 @@
 // and inclusion of their respective header files
 #include "src/Core/util/MKL_support.h"
 
-// if alignment is disabled, then disable vectorization. Note: EIGEN_MAX_ALIGN_BYTES is the proper check, it takes into
-// account both the user's will (EIGEN_MAX_ALIGN_BYTES,EIGEN_DONT_ALIGN) and our own platform checks
-#if EIGEN_MAX_ALIGN_BYTES==0
+// 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 EIGEN_COMP_MSVC
+#ifdef _MSC_VER
   #include <malloc.h> // for _aligned_malloc -- need it regardless of whether vectorization is enabled
-  #if (EIGEN_COMP_MSVC >= 1500) // 2008 or later
+  #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)) || EIGEN_ARCH_x86_64
+    #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__) && ( (!EIGEN_COMP_GNUC) || EIGEN_COMP_ICC || EIGEN_GNUC_AT_LEAST(4,2) )
+  #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
@@ -134,28 +82,6 @@
     #ifdef __SSE4_2__
       #define EIGEN_VECTORIZE_SSE4_2
     #endif
-    #ifdef __AVX__
-      #define EIGEN_VECTORIZE_AVX
-      #define EIGEN_VECTORIZE_SSE3
-      #define EIGEN_VECTORIZE_SSSE3
-      #define EIGEN_VECTORIZE_SSE4_1
-      #define EIGEN_VECTORIZE_SSE4_2
-    #endif
-    #ifdef __AVX2__
-      #define EIGEN_VECTORIZE_AVX2
-    #endif
-    #ifdef __FMA__
-      #define EIGEN_VECTORIZE_FMA
-    #endif
-    #if defined(__AVX512F__) && defined(EIGEN_ENABLE_AVX512)
-      #define EIGEN_VECTORIZE_AVX512
-      #define EIGEN_VECTORIZE_AVX2
-      #define EIGEN_VECTORIZE_AVX
-      #define EIGEN_VECTORIZE_FMA
-      #ifdef __AVX512DQ__
-        #define EIGEN_VECTORIZE_AVX512DQ
-      #endif
-    #endif
 
     // include files
 
@@ -169,10 +95,9 @@
     extern "C" {
       // In theory we should only include immintrin.h and not the other *mmintrin.h header files directly.
       // Doing so triggers some issues with ICC. However old gcc versions seems to not have this file, thus:
-      #if EIGEN_COMP_ICC >= 1110
+      #if defined(__INTEL_COMPILER) && __INTEL_COMPILER >= 1110
         #include <immintrin.h>
       #else
-        #include <mmintrin.h>
         #include <emmintrin.h>
         #include <xmmintrin.h>
         #ifdef  EIGEN_VECTORIZE_SSE3
@@ -187,20 +112,8 @@
         #ifdef EIGEN_VECTORIZE_SSE4_2
         #include <nmmintrin.h>
         #endif
-        #if defined(EIGEN_VECTORIZE_AVX) || defined(EIGEN_VECTORIZE_AVX512)
-        #include <immintrin.h>
-        #endif
       #endif
     } // end extern "C"
-  #elif defined __VSX__
-    #define EIGEN_VECTORIZE
-    #define EIGEN_VECTORIZE_VSX
-    #include <altivec.h>
-    // We need to #undef all these ugly tokens defined in <altivec.h>
-    // => use __vector instead of vector
-    #undef bool
-    #undef vector
-    #undef pixel
   #elif defined __ALTIVEC__
     #define EIGEN_VECTORIZE
     #define EIGEN_VECTORIZE_ALTIVEC
@@ -210,35 +123,13 @@
     #undef bool
     #undef vector
     #undef pixel
-  #elif (defined  __ARM_NEON) || (defined __ARM_NEON__)
+  #elif defined  __ARM_NEON
     #define EIGEN_VECTORIZE
     #define EIGEN_VECTORIZE_NEON
     #include <arm_neon.h>
-  #elif (defined __s390x__ && defined __VEC__)
-    #define EIGEN_VECTORIZE
-    #define EIGEN_VECTORIZE_ZVECTOR
-    #include <vecintrin.h>
   #endif
 #endif
 
-#if defined(__F16C__) && !defined(EIGEN_COMP_CLANG)
-  // We can use the optimized fp16 to float and float to fp16 conversion routines
-  #define EIGEN_HAS_FP16_C
-#endif
-
-#if defined __CUDACC__
-  #define EIGEN_VECTORIZE_CUDA
-  #include <vector_types.h>
-  #if defined __CUDACC_VER__ && __CUDACC_VER__ >= 70500
-    #define EIGEN_HAS_CUDA_FP16
-  #endif
-#endif
-
-#if defined EIGEN_HAS_CUDA_FP16
-  #include <host_defines.h>
-  #include <cuda_fp16.h>
-#endif
-
 #if (defined _OPENMP) && (!defined EIGEN_DONT_PARALLELIZE)
   #define EIGEN_HAS_OPENMP
 #endif
@@ -248,7 +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
 
@@ -268,30 +159,29 @@
 // for min/max:
 #include <algorithm>
 
-// 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)) && (!defined(_WIN32_WCE))
   #include <intrin.h>
 #endif
 
+#if defined(_CPPUNWIND) || defined(__EXCEPTIONS)
+  #define EIGEN_EXCEPTIONS
+#endif
+
+#ifdef EIGEN_EXCEPTIONS
+  #include <new>
+#endif
+
 /** \brief Namespace containing all symbols from the %Eigen library. */
 namespace Eigen {
 
 inline static const char *SimdInstructionSetsInUse(void) {
-#if defined(EIGEN_VECTORIZE_AVX512)
-  return "AVX512, FMA, AVX2, AVX, SSE, SSE2, SSE3, SSSE3, SSE4.1, SSE4.2";
-#elif defined(EIGEN_VECTORIZE_AVX)
-  return "AVX SSE, SSE2, SSE3, SSSE3, SSE4.1, SSE4.2";
-#elif defined(EIGEN_VECTORIZE_SSE4_2)
+#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";
@@ -303,12 +193,8 @@ inline static const char *SimdInstructionSetsInUse(void) {
   return "SSE, SSE2";
 #elif defined(EIGEN_VECTORIZE_ALTIVEC)
   return "AltiVec";
-#elif defined(EIGEN_VECTORIZE_VSX)
-  return "VSX";
 #elif defined(EIGEN_VECTORIZE_NEON)
   return "ARM NEON";
-#elif defined(EIGEN_VECTORIZE_ZVECTOR)
-  return "S390X ZVECTOR";
 #else
   return "None";
 #endif
@@ -316,15 +202,40 @@ inline static const char *SimdInstructionSetsInUse(void) {
 
 } // end namespace Eigen
 
-#if defined EIGEN2_SUPPORT_STAGE40_FULL_EIGEN3_STRICTNESS || defined EIGEN2_SUPPORT_STAGE30_FULL_EIGEN3_API || defined EIGEN2_SUPPORT_STAGE20_RESOLVE_API_CONFLICTS || defined EIGEN2_SUPPORT_STAGE10_FULL_EIGEN2_API || defined EIGEN2_SUPPORT
-// This will generate an error message:
-#error Eigen2-support is only available up to version 3.2. Please go to "http://eigen.tuxfamily.org/index.php?title=Eigen2" for further information
+#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
@@ -338,8 +249,8 @@ 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"
@@ -347,91 +258,41 @@ using std::ptrdiff_t;
 #include "src/Core/NumTraits.h"
 #include "src/Core/MathFunctions.h"
 #include "src/Core/GenericPacketMath.h"
-#include "src/Core/MathFunctionsImpl.h"
 
-#if defined EIGEN_VECTORIZE_AVX512
-  #include "src/Core/arch/SSE/PacketMath.h"
-  #include "src/Core/arch/AVX/PacketMath.h"
-  #include "src/Core/arch/AVX512/PacketMath.h"
-  #include "src/Core/arch/AVX512/MathFunctions.h"
-#elif defined EIGEN_VECTORIZE_AVX
-  // Use AVX for floats and doubles, SSE for integers
-  #include "src/Core/arch/SSE/PacketMath.h"
-  #include "src/Core/arch/SSE/Complex.h"
-  #include "src/Core/arch/SSE/MathFunctions.h"
-  #include "src/Core/arch/AVX/PacketMath.h"
-  #include "src/Core/arch/AVX/MathFunctions.h"
-  #include "src/Core/arch/AVX/Complex.h"
-  #include "src/Core/arch/AVX/TypeCasting.h"
-#elif defined EIGEN_VECTORIZE_SSE
+#if defined EIGEN_VECTORIZE_SSE
   #include "src/Core/arch/SSE/PacketMath.h"
   #include "src/Core/arch/SSE/MathFunctions.h"
   #include "src/Core/arch/SSE/Complex.h"
-  #include "src/Core/arch/SSE/TypeCasting.h"
-#elif defined(EIGEN_VECTORIZE_ALTIVEC) || defined(EIGEN_VECTORIZE_VSX)
+#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/MathFunctions.h"
   #include "src/Core/arch/NEON/Complex.h"
-#elif defined EIGEN_VECTORIZE_ZVECTOR
-  #include "src/Core/arch/ZVector/PacketMath.h"
-  #include "src/Core/arch/ZVector/MathFunctions.h"
-  #include "src/Core/arch/ZVector/Complex.h"
-#endif
-
-// Half float support
-#include "src/Core/arch/CUDA/Half.h"
-#include "src/Core/arch/CUDA/PacketMathHalf.h"
-#include "src/Core/arch/CUDA/TypeCasting.h"
-
-#if defined EIGEN_VECTORIZE_CUDA
-  #include "src/Core/arch/CUDA/PacketMath.h"
-  #include "src/Core/arch/CUDA/MathFunctions.h"
 #endif
 
 #include "src/Core/arch/Default/Settings.h"
 
-#include "src/Core/functors/TernaryFunctors.h"
-#include "src/Core/functors/BinaryFunctors.h"
-#include "src/Core/functors/UnaryFunctors.h"
-#include "src/Core/functors/NullaryFunctors.h"
-#include "src/Core/functors/StlFunctors.h"
-#include "src/Core/functors/AssignmentFunctors.h"
-
-// Specialized functors to enable the processing of complex numbers
-// on CUDA devices
-#include "src/Core/arch/CUDA/Complex.h"
-
+#include "src/Core/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"
@@ -439,33 +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/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"
@@ -480,7 +340,6 @@ 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"
 
 #include "src/Core/BooleanRedux.h"
 #include "src/Core/Select.h"
@@ -488,17 +347,18 @@ using std::ptrdiff_t;
 #include "src/Core/Random.h"
 #include "src/Core/Replicate.h"
 #include "src/Core/Reverse.h"
+#include "src/Core/ArrayBase.h"
 #include "src/Core/ArrayWrapper.h"
 
 #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
@@ -509,4 +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 +1,95 @@
+// 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
+
+#ifndef EIGEN_NO_EIGEN2_DEPRECATED_WARNING
+
+#if defined(__GNUC__) || defined(__INTEL_COMPILER) || defined(__clang__)
+#warning "Eigen2 support is deprecated in Eigen 3.2.x and it will be removed in Eigen 3.3. (Define EIGEN_NO_EIGEN2_DEPRECATED_WARNING to disable this warning)"
+#else
+#pragma message ("Eigen2 support is deprecated in Eigen 3.2.x and it will be removed in Eigen 3.3. (Define EIGEN_NO_EIGEN2_DEPRECATED_WARNING to disable this warning)")
+#endif
+
+#endif // EIGEN_NO_EIGEN2_DEPRECATED_WARNING
+
+#include "src/Core/util/DisableStupidWarnings.h"
+
+/** \ingroup Support_modules
+  * \defgroup Eigen2Support_Module Eigen2 support module
+  *
+  * \warning Eigen2 support is deprecated in Eigen 3.2.x and it will be removed in Eigen 3.3.
+  *
+  * 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

@@ -1,10 +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_EIGENVALUES_MODULE_H
 #define EIGEN_EIGENVALUES_MODULE_H
 
@@ -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"
@@ -45,10 +37,9 @@
 #include "src/Eigenvalues/GeneralizedEigenSolver.h"
 #include "src/Eigenvalues/MatrixBaseEigenvalues.h"
 #ifdef EIGEN_USE_LAPACKE
-#include "src/misc/lapacke.h"
-#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"

Eigen/Geometry

@@ -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_GEOMETRY_MODULE_H
 #define EIGEN_GEOMETRY_MODULE_H
 
@@ -16,17 +9,21 @@
 #include "LU"
 #include <limits>
 
+#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>
@@ -36,23 +33,27 @@
 #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. At the moment the
-// SSE version doesn't compile when AVX is enabled
-#if defined EIGEN_VECTORIZE_SSE && !defined EIGEN_VECTORIZE_AVX
-#include "src/Geometry/arch/Geometry_SSE.h"
+  #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"

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
 

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
 

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
 
@@ -23,23 +16,25 @@
   * \endcode
   */
 
+#include "src/misc/Solve.h"
 #include "src/misc/Kernel.h"
 #include "src/misc/Image.h"
 #include "src/LU/FullPivLU.h"
 #include "src/LU/PartialPivLU.h"
 #ifdef EIGEN_USE_LAPACKE
-#include "src/misc/lapacke.h"
-#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
+#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

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
 

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>
@@ -41,8 +35,12 @@ extern "C" {
   *
   */
 
+#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

@@ -1,10 +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_QR_MODULE_H
 #define EIGEN_QR_MODULE_H
 
@@ -22,26 +15,31 @@
   *
   * 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
-#include "src/misc/lapacke.h"
-#include "src/QR/HouseholderQR_LAPACKE.h"
-#include "src/QR/ColPivHouseholderQR_LAPACKE.h"
+#include "src/QR/HouseholderQR_MKL.h"
+#include "src/QR/ColPivHouseholderQR_MKL.h"
+#endif
+
+#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

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
 
@@ -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
 
@@ -19,26 +12,23 @@
   *
   *
   * 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)
-#include "src/misc/lapacke.h"
-#include "src/SVD/JacobiSVD_LAPACKE.h"
+#include "src/SVD/JacobiSVD_MKL.h"
+#endif
+#include "src/SVD/UpperBidiagonalization.h"
+
+#ifdef EIGEN2_SUPPORT
+#include "src/Eigen2Support/SVD.h"
 #endif
 
 #include "src/Core/util/ReenableStupidWarnings.h"

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

@@ -34,6 +34,8 @@
 #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

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
 
@@ -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

@@ -20,6 +20,9 @@
   * 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"
 

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 +21,9 @@
   * 
   */
 
+#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 /* MSVC auto aligns 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 /* MSVC auto aligns 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 /* MSVC auto aligns 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
 
@@ -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

@@ -13,7 +13,7 @@
 #ifndef EIGEN_LDLT_H
 #define EIGEN_LDLT_H
 
-namespace Eigen {
+namespace Eigen { 
 
 namespace internal {
   template<typename MatrixType, int UpLo> struct LDLT_Traits;
@@ -28,8 +28,8 @@ 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
@@ -43,9 +43,7 @@ namespace internal {
   * 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
 {
@@ -54,15 +52,15 @@ template<typename _MatrixType, int _UpLo> class 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 typename MatrixType::Scalar Scalar;
     typedef typename NumTraits<typename MatrixType::Scalar>::Real RealScalar;
-    typedef Eigen::Index Index; ///< \deprecated since Eigen 3.3
-    typedef typename MatrixType::StorageIndex StorageIndex;
-    typedef Matrix<Scalar, RowsAtCompileTime, 1, 0, MaxRowsAtCompileTime, 1> TmpMatrixType;
+    typedef typename MatrixType::Index Index;
+    typedef Matrix<Scalar, RowsAtCompileTime, 1, Options, MaxRowsAtCompileTime, 1> TmpMatrixType;
 
     typedef Transpositions<RowsAtCompileTime, MaxRowsAtCompileTime> TranspositionType;
     typedef PermutationMatrix<RowsAtCompileTime, MaxRowsAtCompileTime> PermutationType;
@@ -74,11 +72,11 @@ 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(),
+    LDLT() 
+      : m_matrix(), 
+        m_transpositions(), 
         m_sign(internal::ZeroSign),
-        m_isInitialized(false)
+        m_isInitialized(false) 
     {}
 
     /** \brief Default Constructor with memory preallocation
@@ -87,7 +85,7 @@ 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),
@@ -98,35 +96,16 @@ template<typename _MatrixType, int _UpLo> class LDLT
     /** \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
@@ -172,6 +151,13 @@ template<typename _MatrixType, int _UpLo> class LDLT
       eigen_assert(m_isInitialized && "LDLT is not initialized.");
       return m_sign == internal::PositiveSemiDef || m_sign == internal::ZeroSign;
     }
+    
+    #ifdef EIGEN2_SUPPORT
+    inline bool isPositiveDefinite() const
+    {
+      return isPositive();
+    }
+    #endif
 
     /** \returns true if the matrix is negative (semidefinite) */
     inline bool isNegative(void) const
@@ -187,38 +173,37 @@ 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$ is \f$ A \f$ is singular.
       *
-      * \sa MatrixBase::ldlt(), SelfAdjointView::ldlt()
+      * \sa MatrixBase::ldlt()
       */
     template<typename Rhs>
-    inline const Solve<LDLT, Rhs>
+    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 Solve<LDLT, Rhs>(*this, b.derived());
+      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);
@@ -235,13 +220,6 @@ 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; };
-
     inline Index rows() const { return m_matrix.rows(); }
     inline Index cols() const { return m_matrix.cols(); }
 
@@ -253,17 +231,11 @@ template<typename _MatrixType, int _UpLo> class LDLT
     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>
-    EIGEN_DEVICE_FUNC
-    void _solve_impl(const RhsType &rhs, DstType &dst) const;
-    #endif
-
   protected:
-
+    
     static void check_template_parameters()
     {
       EIGEN_STATIC_ASSERT_NON_INTEGER(Scalar);
@@ -276,12 +248,10 @@ 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;
     bool m_isInitialized;
-    ComputationInfo m_info;
 };
 
 namespace internal {
@@ -296,17 +266,15 @@ template<> struct ldlt_inplace<Lower>
     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 (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;
+      if (numext::real(mat.coeff(0,0)) > 0) sign = PositiveSemiDef;
+      else if (numext::real(mat.coeff(0,0)) < 0) sign = NegativeSemiDef;
       else sign = ZeroSign;
       return true;
     }
@@ -318,7 +286,7 @@ template<> struct ldlt_inplace<Lower>
       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);
+      transpositions.coeffRef(k) = index_of_biggest_in_corner;
       if(k != index_of_biggest_in_corner)
       {
         // apply the transposition while taking care to consider only
@@ -327,7 +295,7 @@ 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));
@@ -353,44 +321,26 @@ template<> struct ldlt_inplace<Lower>
         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.
+      // was smaller than the cutoff value. However, soince LDLT is not rank-revealing
+      // we should only make sure we do not introduce INF or NaN values.
+      // LAPACK also uses 0 as the cutoff value.
       RealScalar realAkk = numext::real(mat.coeffRef(k,k));
-      bool pivot_is_valid = (abs(realAkk) > RealScalar(0));
-
-      if(k==0 && !pivot_is_valid)
-      {
-        // The entire diagonal is zero, there is nothing more to do
-        // except filling the transpositions, and checking whether the matrix is zero.
-        sign = ZeroSign;
-        for(Index j = 0; j<size; ++j)
-        {
-          transpositions.coeffRef(j) = IndexType(j);
-          ret = ret && (mat.col(j).tail(size-j-1).array()==Scalar(0)).all();
-        }
-        return ret;
-      }
-
-      if((rs>0) && pivot_is_valid)
+      if((rs>0) && (abs(realAkk) > RealScalar(0)))
         A21 /= realAkk;
 
-      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;
+        if (realAkk < 0) sign = Indefinite;
       } else if (sign == NegativeSemiDef) {
-        if (realAkk > static_cast<RealScalar>(0)) sign = Indefinite;
+        if (realAkk > 0) sign = Indefinite;
       } else if (sign == ZeroSign) {
-        if (realAkk > static_cast<RealScalar>(0)) sign = PositiveSemiDef;
-        else if (realAkk < static_cast<RealScalar>(0)) sign = NegativeSemiDef;
+        if (realAkk > 0) sign = PositiveSemiDef;
+        else if (realAkk < 0) sign = NegativeSemiDef;
       }
     }
 
-    return ret;
+    return true;
   }
 
   // Reference for the algorithm: Davis and Hager, "Multiple Rank
@@ -406,6 +356,7 @@ 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);
@@ -469,16 +420,16 @@ 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
@@ -486,35 +437,21 @@ 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;
@@ -529,19 +466,18 @@ template<typename MatrixType, int _UpLo>
 template<typename Derived>
 LDLT<MatrixType,_UpLo>& LDLT<MatrixType,_UpLo>::rankUpdate(const MatrixBase<Derived>& w, const typename LDLT<MatrixType,_UpLo>::RealScalar& sigma)
 {
-  typedef typename TranspositionType::StorageIndex IndexType;
   const Index size = w.rows();
   if (m_isInitialized)
   {
     eigen_assert(m_matrix.rows()==size);
   }
   else
-  {
+  {    
     m_matrix.resize(size,size);
     m_matrix.setZero();
     m_transpositions.resize(size);
     for (Index i = 0; i < size; i++)
-      m_transpositions.coeffRef(i) = IndexType(i);
+      m_transpositions.coeffRef(i) = i;
     m_temporary.resize(size);
     m_sign = sigma>=0 ? internal::PositiveSemiDef : internal::NegativeSemiDef;
     m_isInitialized = true;
@@ -552,45 +488,53 @@ 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>
 {
-  eigen_assert(rhs.rows() == rows());
-  // dst = P b
-  dst = m_transpositions * rhs;
+  typedef LDLT<_MatrixType,_UpLo> LDLTType;
+  EIGEN_MAKE_SOLVE_HELPERS(LDLTType,Rhs)
 
-  // dst = L^-1 (P b)
-  matrixL().solveInPlace(dst);
-
-  // dst = D^-1 (L^-1 P b)
-  // more precisely, use pseudo-inverse of D (see bug 241)
-  using std::abs;
-  const typename Diagonal<const MatrixType>::RealReturnType vecD(vectorD());
-  // In some previous versions, tolerance was set to the max of 1/highest and the maximal diagonal entry * epsilon
-  // as motivated by LAPACK's xGELSS:
-  // RealScalar tolerance = numext::maxi(vecD.array().abs().maxCoeff() * NumTraits<RealScalar>::epsilon(),RealScalar(1) / NumTraits<RealScalar>::highest());
-  // However, LDLT is not rank revealing, and so adjusting the tolerance wrt to the highest
-  // diagonal element is not well justified and leads to numerical issues in some cases.
-  // Moreover, Lapack's xSYTRS routines use 0 for the tolerance.
-  RealScalar tolerance = RealScalar(1) / NumTraits<RealScalar>::highest();
-
-  for (Index i = 0; i < vecD.size(); ++i)
+  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::RealScalar RealScalar;
+    const typename Diagonal<const MatrixType>::RealReturnType vectorD(dec().vectorD());
+    // In some previous versions, tolerance was set to the max of 1/highest and the maximal diagonal entry * epsilon
+    // as motivated by LAPACK's xGELSS:
+    // RealScalar tolerance = (max)(vectorD.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 to numerical issues in some cases.
+    // Moreover, Lapack's xSYTRS routines use 0 for the tolerance.
+    RealScalar tolerance = RealScalar(1) / NumTraits<RealScalar>::highest();
+    
+    for (Index i = 0; i < 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^-T (D^-1 L^-1 P b)
-  matrixU().solveInPlace(dst);
-
-  // dst = P^-1 (L^-T D^-1 L^-1 P b) = A^-1 b
-  dst = m_transpositions.transpose() * dst;
+};
 }
-#endif
 
 /** \internal use x = ldlt_object.solve(x);
   *
@@ -644,7 +588,6 @@ 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>
@@ -655,7 +598,6 @@ 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

@@ -10,7 +10,7 @@
 #ifndef EIGEN_LLT_H
 #define EIGEN_LLT_H
 
-namespace Eigen {
+namespace Eigen { 
 
 namespace internal{
 template<typename MatrixType, int UpLo> struct LLT_Traits;
@@ -22,8 +22,8 @@ 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.
+  * \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
@@ -40,10 +40,8 @@ template<typename MatrixType, int UpLo> struct LLT_Traits;
   *
   * Example: \include LLT_example.cpp
   * Output: \verbinclude LLT_example.out
-  *
-  * This class supports the \link InplaceDecomposition inplace decomposition \endlink mechanism.
-  *
-  * \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,
@@ -56,12 +54,12 @@ template<typename _MatrixType, int _UpLo> class 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 Eigen::Index Index; ///< \deprecated since Eigen 3.3
-    typedef typename MatrixType::StorageIndex StorageIndex;
+    typedef typename MatrixType::Index Index;
 
     enum {
       PacketSize = internal::packet_traits<Scalar>::size,
@@ -85,30 +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 LDLT factorization from a given matrix
-      *
-      * This overloaded constructor is provided for \link InplaceDecomposition inplace decomposition \endlink when
-      * \c MatrixType is a Eigen::Ref.
-      *
-      * \sa LLT(const EigenBase&)
-      */
-    template<typename InputType>
-    explicit LLT(EigenBase<InputType>& matrix)
-      : m_matrix(matrix.derived()),
-        m_isInitialized(false)
-    {
-      compute(matrix.derived());
+      compute(matrix);
     }
 
     /** \returns a view of the upper triangular matrix U */
@@ -133,33 +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>
+    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 Solve<LLT, Rhs>(*this, b.derived());
+      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(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
       *
@@ -185,38 +167,24 @@ 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 { return *this; };
-
     inline Index rows() const { return m_matrix.rows(); }
     inline Index cols() const { return m_matrix.cols(); }
 
     template<typename VectorType>
     LLT rankUpdate(const VectorType& vec, const RealScalar& sigma = 1);
 
-    #ifndef EIGEN_PARSED_BY_DOXYGEN
-    template<typename RhsType, typename DstType>
-    EIGEN_DEVICE_FUNC
-    void _solve_impl(const RhsType &rhs, DstType &dst) const;
-    #endif
-
   protected:
-
+    
     static void check_template_parameters()
     {
       EIGEN_STATIC_ASSERT_NON_INTEGER(Scalar);
     }
-
+    
     /** \internal
       * Used to compute and store L
       * The strict upper part is not used and even not initialized.
       */
     MatrixType m_matrix;
-    RealScalar m_l1_norm;
     bool m_isInitialized;
     ComputationInfo m_info;
 };
@@ -226,11 +194,12 @@ 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;
@@ -299,10 +268,11 @@ 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)
@@ -325,8 +295,9 @@ template<typename Scalar> struct llt_inplace<Scalar, Lower>
   }
 
   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)
@@ -351,36 +322,36 @@ 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);
@@ -391,8 +362,8 @@ 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; }
 };
@@ -401,8 +372,8 @@ 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; }
 };
@@ -417,28 +388,14 @@ 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);
-  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);
@@ -466,24 +423,33 @@ LLT<_MatrixType,_UpLo> LLT<_MatrixType,_UpLo>::rankUpdate(const VectorType& v, c
 
   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>
 {
-  dst = rhs;
-  solveInPlace(dst);
+  typedef LLT<_MatrixType,UpLo> LLTType;
+  EIGEN_MAKE_SOLVE_HELPERS(LLTType,Rhs)
+
+  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.
+  *
+  * This version avoids a copy when the right hand side matrix b is not
+  * needed anymore.
   *
   * \sa LLT::solve(), MatrixBase::llt()
   */
@@ -509,7 +475,6 @@ 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>
@@ -520,7 +485,6 @@ 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

@@ -25,38 +25,41 @@
  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 = LAPACKE_##MKLPREFIX##potrf( matrix_order, uplo, size, (MKLTYPE*)a, lda ); \
     info = (info==0) ? -1 : info>0 ? info-1 : size; \
     return info; \
   } \
@@ -64,36 +67,36 @@ template<> struct lapacke_llt<EIGTYPE> \
 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

@@ -54,12 +54,12 @@ template<> struct cholmod_configure_matrix<std::complex<double> > {
 /** 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();
@@ -80,11 +80,11 @@ 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,long>::value)
+  else if (internal::is_same<_Index,SuiteSparse_long>::value)
   {
     res.itype = CHOLMOD_LONG;
   }
@@ -104,23 +104,16 @@ cholmod_sparse viewAsCholmod(Ref<SparseMatrix<_Scalar,_Options,_StorageIndex> >
 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;
@@ -151,12 +144,12 @@ cholmod_dense viewAsCholmod(MatrixBase<Derived>& mat)
 
 /** 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) );
 }
 
 enum CholmodMode {
@@ -170,36 +163,28 @@ 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;
       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;
@@ -214,8 +199,11 @@ class CholmodBase : public SparseSolverBase<Derived>
       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.
       *
@@ -236,6 +224,34 @@ class CholmodBase : public SparseSolverBase<Derived>
       return 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::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 sparsity pattern of \a matrix.
       *
       * This function is particularly useful when solving for several problems having the same structure.
@@ -282,22 +298,20 @@ class CholmodBase : public SparseSolverBase<Derived>
     #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 stoarge without inner-stride, which corresponds to the default behavior of Ref.
-      Ref<const Matrix<typename Rhs::Scalar,Dynamic,Dynamic,ColMajor> > b_ref(b.derived());
 
+      // note: cd stands for Cholmod Dense
+      Rhs& b_ref(b.const_cast_derived());
       cholmod_dense b_cd = viewAsCholmod(b_ref);
       cholmod_dense* x_cd = cholmod_solve(CHOLMOD_A, m_cholmodFactor, &b_cd, &m_cholmod);
       if(!x_cd)
       {
         this->m_info = NumericalIssue;
-        return;
       }
       // TODO optimize this copy by swapping when possible (be careful with alignment, etc.)
       dest = Matrix<Scalar,Dest::RowsAtCompileTime,Dest::ColsAtCompileTime>::Map(reinterpret_cast<Scalar*>(x_cd->x),b.rows(),b.cols());
@@ -305,8 +319,8 @@ class CholmodBase : public SparseSolverBase<Derived>
     }
     
     /** \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;
@@ -314,16 +328,14 @@ 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 b_cs = viewAsCholmod(b);
       cholmod_sparse* x_cs = cholmod_spsolve(CHOLMOD_A, m_cholmodFactor, &b_cs, &m_cholmod);
       if(!x_cs)
       {
         this->m_info = NumericalIssue;
-        return;
       }
       // TODO optimize this copy by swapping when possible (be careful with alignment, etc.)
-      dest.derived() = viewAsEigen<typename DestDerived::Scalar,ColMajor,typename DestDerived::StorageIndex>(*x_cs);
+      dest = viewAsEigen<DestScalar,DestOptions,DestIndex>(*x_cs);
       cholmod_free_sparse(&x_cs, &m_cholmod);
     }
     #endif // EIGEN_PARSED_BY_DOXYGEN
@@ -344,57 +356,6 @@ class CholmodBase : public SparseSolverBase<Derived>
       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*/)
     {}
@@ -404,6 +365,7 @@ class CholmodBase : public SparseSolverBase<Derived>
     cholmod_factor* m_cholmodFactor;
     double m_shiftOffset[2];
     mutable ComputationInfo m_info;
+    bool m_isInitialized;
     int m_factorizationIsOk;
     int m_analysisIsOk;
 };
@@ -422,13 +384,11 @@ class CholmodBase : public SparseSolverBase<Derived>
   * \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> >
@@ -445,7 +405,7 @@ class CholmodSimplicialLLT : public CholmodBase<_MatrixType, _UpLo, CholmodSimpl
     CholmodSimplicialLLT(const MatrixType& matrix) : Base()
     {
       init();
-      this->compute(matrix);
+      Base::compute(matrix);
     }
 
     ~CholmodSimplicialLLT() {}
@@ -473,13 +433,11 @@ class CholmodSimplicialLLT : public CholmodBase<_MatrixType, _UpLo, CholmodSimpl
   * \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> >
@@ -496,7 +454,7 @@ class CholmodSimplicialLDLT : public CholmodBase<_MatrixType, _UpLo, CholmodSimp
     CholmodSimplicialLDLT(const MatrixType& matrix) : Base()
     {
       init();
-      this->compute(matrix);
+      Base::compute(matrix);
     }
 
     ~CholmodSimplicialLDLT() {}
@@ -522,13 +480,11 @@ class CholmodSimplicialLDLT : public CholmodBase<_MatrixType, _UpLo, CholmodSimp
   * \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> >
@@ -545,7 +501,7 @@ class CholmodSupernodalLLT : public CholmodBase<_MatrixType, _UpLo, CholmodSuper
     CholmodSupernodalLLT(const MatrixType& matrix) : Base()
     {
       init();
-      this->compute(matrix);
+      Base::compute(matrix);
     }
 
     ~CholmodSupernodalLLT() {}
@@ -573,13 +529,11 @@ 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> >
@@ -596,7 +550,7 @@ class CholmodDecomposition : public CholmodBase<_MatrixType, _UpLo, CholmodDecom
     CholmodDecomposition(const MatrixType& matrix) : Base()
     {
       init();
-      this->compute(matrix);
+      Base::compute(matrix);
     }
 
     ~CholmodDecomposition() {}
@@ -634,6 +588,36 @@ 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/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);
     }
@@ -112,12 +92,11 @@ class Array
     /** This is a special case of the templated operator=. Its purpose is to
       * prevent a default operator= from hiding the templated operator=.
       */
-    EIGEN_DEVICE_FUNC
     EIGEN_STRONG_INLINE Array& operator=(const Array& other)
     {
       return Base::_set(other);
     }
-    
+
     /** Default constructor.
       *
       * For fixed-size matrices, does nothing.
@@ -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())
     {
@@ -147,64 +124,56 @@ class Array
     }
 #endif
 
-#if EIGEN_HAS_RVALUE_REFERENCES
-    EIGEN_DEVICE_FUNC
-    Array(Array&& other) EIGEN_NOEXCEPT_IF(std::is_nothrow_move_constructible<Scalar>::value)
+#ifdef EIGEN_HAVE_RVALUE_REFERENCES
+    Array(Array&& other)
       : Base(std::move(other))
     {
       Base::_check_template_params();
       if (RowsAtCompileTime!=Dynamic && ColsAtCompileTime!=Dynamic)
         Base::_set_noalias(other);
     }
-    EIGEN_DEVICE_FUNC
-    Array& operator=(Array&& other) EIGEN_NOEXCEPT_IF(std::is_nothrow_move_assignable<Scalar>::value)
+    Array& operator=(Array&& other)
     {
       other.swap(*this);
       return *this;
     }
 #endif
 
-    #ifndef EIGEN_PARSED_BY_DOXYGEN
-    template<typename T>
-    EIGEN_DEVICE_FUNC
-    EIGEN_STRONG_INLINE explicit Array(const T& x)
+    /** 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();
-      Base::template _init1<T>(x);
+      EIGEN_STATIC_ASSERT_VECTOR_ONLY(Array)
+      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 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 */
-    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 */
     Array(const Scalar& val0, const Scalar& val1);
     #endif
 
     /** constructs an initialized 3D vector with given coefficients */
-    EIGEN_DEVICE_FUNC
     EIGEN_STRONG_INLINE Array(const Scalar& val0, const Scalar& val1, const Scalar& val2)
     {
       Base::_check_template_params();
@@ -214,7 +183,6 @@ class Array
       m_storage.data()[2] = val2;
     }
     /** constructs an initialized 4D vector with given coefficients */
-    EIGEN_DEVICE_FUNC
     EIGEN_STRONG_INLINE Array(const Scalar& val0, const Scalar& val1, const Scalar& val2, const Scalar& val3)
     {
       Base::_check_template_params();
@@ -225,21 +193,51 @@ class Array
       m_storage.data()[3] = val3;
     }
 
+    explicit Array(const Scalar *data);
+
+    /** Constructor copying the value of the expression \a other */
+    template<typename OtherDerived>
+    EIGEN_STRONG_INLINE Array(const ArrayBase<OtherDerived>& other)
+             : Base(other.rows() * other.cols(), other.rows(), other.cols())
+    {
+      Base::_check_template_params();
+      Base::_set_noalias(other);
+    }
     /** Copy constructor */
-    EIGEN_DEVICE_FUNC
     EIGEN_STRONG_INLINE Array(const Array& other)
-            : Base(other)
-    { }
+            : 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)
-      : Base(other.derived())
-    { }
+      : Base(other.derived().rows() * other.derived().cols(), other.derived().rows(), other.derived().cols())
+    {
+      Base::_check_template_params();
+      Base::_resize_to_match(other);
+      *this = other;
+    }
 
-    EIGEN_DEVICE_FUNC inline Index innerStride() const { return 1; }
-    EIGEN_DEVICE_FUNC inline Index outerStride() const { 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

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
   */
@@ -47,11 +47,13 @@ template<typename Derived> class ArrayBase
     typedef ArrayBase Eigen_BaseClassForSpecializationOfGlobalMathFuncImpl;
 
     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;
 
     typedef DenseBase<Derived> Base;
+    using Base::operator*;
     using Base::RowsAtCompileTime;
     using Base::ColsAtCompileTime;
     using Base::SizeAtCompileTime;
@@ -60,7 +62,8 @@ 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;
@@ -80,14 +83,25 @@ 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"
@@ -98,62 +112,44 @@ 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_DEVICE_FUNC
     ArrayBase() : Base() {}
 
   private:
@@ -178,7 +174,8 @@ template<typename OtherDerived>
 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();
 }
 
@@ -191,7 +188,8 @@ template<typename OtherDerived>
 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();
 }
 
@@ -204,7 +202,8 @@ template<typename OtherDerived>
 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();
 }
 
@@ -217,7 +216,8 @@ template<typename OtherDerived>
 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

@@ -44,7 +44,6 @@ 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,
@@ -52,45 +51,76 @@ 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) {}
-
-    EIGEN_DEVICE_FUNC
     inline Index rows() const { return m_expression.rows(); }
-    EIGEN_DEVICE_FUNC
     inline Index cols() const { return m_expression.cols(); }
-    EIGEN_DEVICE_FUNC
     inline Index outerStride() const { return m_expression.outerStride(); }
-    EIGEN_DEVICE_FUNC
     inline Index innerStride() const { return m_expression.innerStride(); }
 
-    EIGEN_DEVICE_FUNC
-    inline ScalarWithConstIfNotLvalue* data() { return m_expression.data(); }
-    EIGEN_DEVICE_FUNC
+    inline 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; }
 
     const typename internal::remove_all<NestedExpressionType>::type& 
-    EIGEN_DEVICE_FUNC
     nestedExpression() const 
     {
       return m_expression;
@@ -98,12 +128,10 @@ class ArrayWrapper : public ArrayBase<ArrayWrapper<ExpressionType> >
 
     /** 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;
@@ -141,7 +169,6 @@ 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,
@@ -149,40 +176,72 @@ 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) {}
-
-    EIGEN_DEVICE_FUNC
     inline Index rows() const { return m_expression.rows(); }
-    EIGEN_DEVICE_FUNC
     inline Index cols() const { return m_expression.cols(); }
-    EIGEN_DEVICE_FUNC
     inline Index outerStride() const { return m_expression.outerStride(); }
-    EIGEN_DEVICE_FUNC
     inline Index innerStride() const { return m_expression.innerStride(); }
 
-    EIGEN_DEVICE_FUNC
-    inline ScalarWithConstIfNotLvalue* data() { return m_expression.data(); }
-    EIGEN_DEVICE_FUNC
+    inline 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);
+    }
+
+    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);
     }
 
-    EIGEN_DEVICE_FUNC
     const typename internal::remove_all<NestedExpressionType>::type& 
     nestedExpression() const 
     {
@@ -191,12 +250,10 @@ class MatrixWrapper : public MatrixBase<MatrixWrapper<ExpressionType> >
 
     /** 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

@@ -14,6 +14,478 @@
 
 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 typename Derived1::Scalar Scalar;
+    typedef packet_traits<Scalar> PacketTraits;
+    enum {
+      packetSize = PacketTraits::size,
+      alignable = PacketTraits::AlignedOnScalar,
+      dstIsAligned = assign_traits<Derived1,Derived2>::DstIsAligned,
+      dstAlignment = alignable ? Aligned : int(dstIsAligned),
+      srcAlignment = assign_traits<Derived1,Derived2>::JointAlignment
+    };
+    const Scalar *dst_ptr = &dst.coeffRef(0,0);
+    if((!bool(dstIsAligned)) && (size_t(dst_ptr) % sizeof(Scalar))>0)
+    {
+      // the pointer is not aligend-on scalar, so alignment is not possible
+      return assign_impl<Derived1,Derived2,DefaultTraversal,NoUnrolling>::run(dst, src);
+    }
+    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) || bool(dstIsAligned)) ? 0 : internal::first_aligned(dst_ptr, innerSize);
+
+    for(Index outer = 0; outer < outerSize; ++outer)
+    {
+      const Index alignedEnd = alignedStart + ((innerSize-alignedStart) & ~packetAlignedMask);
+      // do the non-vectorizable part of the assignment
+      for(Index inner = 0; inner<alignedStart ; ++inner)
+        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_STRONG_INLINE Derived& DenseBase<Derived>
@@ -27,62 +499,90 @@ 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/AssignEvaluator.h

@@ -1,918 +0,0 @@
-// This file is part of Eigen, a lightweight C++ template library
-// for linear algebra.
-//
-// Copyright (C) 2011 Benoit Jacob <jacob.benoit.1@gmail.com>
-// Copyright (C) 2011-2014 Gael Guennebaud <gael.guennebaud@inria.fr>
-// Copyright (C) 2011-2012 Jitse Niesen <jitse@maths.leeds.ac.uk>
-//
-// This Source Code Form is subject to the terms of the Mozilla
-// Public License v. 2.0. If a copy of the MPL was not distributed
-// with this file, You can obtain one at http://mozilla.org/MPL/2.0/.
-
-#ifndef EIGEN_ASSIGN_EVALUATOR_H
-#define EIGEN_ASSIGN_EVALUATOR_H
-
-namespace Eigen {
-
-// This implementation is based on Assign.h
-
-namespace internal {
-  
-/***************************************************************************
-* Part 1 : the logic deciding a strategy for traversal and unrolling       *
-***************************************************************************/
-
-// copy_using_evaluator_traits is based on assign_traits
-
-template <typename DstEvaluator, typename SrcEvaluator, typename AssignFunc>
-struct copy_using_evaluator_traits
-{
-  typedef typename DstEvaluator::XprType Dst;
-  typedef typename Dst::Scalar DstScalar;
-  
-  enum {
-    DstFlags = DstEvaluator::Flags,
-    SrcFlags = SrcEvaluator::Flags
-  };
-  
-public:
-  enum {
-    DstAlignment = DstEvaluator::Alignment,
-    SrcAlignment = SrcEvaluator::Alignment,
-    DstHasDirectAccess = DstFlags & DirectAccessBit,
-    JointAlignment = EIGEN_PLAIN_ENUM_MIN(DstAlignment,SrcAlignment)
-  };
-
-private:
-  enum {
-    InnerSize = int(Dst::IsVectorAtCompileTime) ? int(Dst::SizeAtCompileTime)
-              : int(DstFlags)&RowMajorBit ? int(Dst::ColsAtCompileTime)
-              : int(Dst::RowsAtCompileTime),
-    InnerMaxSize = int(Dst::IsVectorAtCompileTime) ? int(Dst::MaxSizeAtCompileTime)
-              : int(DstFlags)&RowMajorBit ? int(Dst::MaxColsAtCompileTime)
-              : int(Dst::MaxRowsAtCompileTime),
-    OuterStride = int(outer_stride_at_compile_time<Dst>::ret),
-    MaxSizeAtCompileTime = Dst::SizeAtCompileTime
-  };
-
-  // TODO distinguish between linear traversal and inner-traversals
-  typedef typename find_best_packet<DstScalar,Dst::SizeAtCompileTime>::type LinearPacketType;
-  typedef typename find_best_packet<DstScalar,InnerSize>::type InnerPacketType;
-
-  enum {
-    LinearPacketSize = unpacket_traits<LinearPacketType>::size,
-    InnerPacketSize = unpacket_traits<InnerPacketType>::size
-  };
-
-public:
-  enum {
-    LinearRequiredAlignment = unpacket_traits<LinearPacketType>::alignment,
-    InnerRequiredAlignment = unpacket_traits<InnerPacketType>::alignment
-  };
-
-private:
-  enum {
-    DstIsRowMajor = DstFlags&RowMajorBit,
-    SrcIsRowMajor = SrcFlags&RowMajorBit,
-    StorageOrdersAgree = (int(DstIsRowMajor) == int(SrcIsRowMajor)),
-    MightVectorize = bool(StorageOrdersAgree)
-                  && (int(DstFlags) & int(SrcFlags) & ActualPacketAccessBit)
-                  && bool(functor_traits<AssignFunc>::PacketAccess),
-    MayInnerVectorize  = MightVectorize
-                       && int(InnerSize)!=Dynamic && int(InnerSize)%int(InnerPacketSize)==0
-                       && int(OuterStride)!=Dynamic && int(OuterStride)%int(InnerPacketSize)==0
-                       && (EIGEN_UNALIGNED_VECTORIZE  || int(JointAlignment)>=int(InnerRequiredAlignment)),
-    MayLinearize = bool(StorageOrdersAgree) && (int(DstFlags) & int(SrcFlags) & LinearAccessBit),
-    MayLinearVectorize = bool(MightVectorize) && MayLinearize && DstHasDirectAccess
-                       && (EIGEN_UNALIGNED_VECTORIZE || (int(DstAlignment)>=int(LinearRequiredAlignment)) || MaxSizeAtCompileTime == Dynamic),
-      /* If the destination isn't aligned, we have to do runtime checks and we don't unroll,
-         so it's only good for large enough sizes. */
-    MaySliceVectorize  = bool(MightVectorize) && bool(DstHasDirectAccess)
-                       && (int(InnerMaxSize)==Dynamic || int(InnerMaxSize)>=(EIGEN_UNALIGNED_VECTORIZE?InnerPacketSize:(3*InnerPacketSize)))
-      /* slice vectorization can be slow, so we only want it if the slices are big, which is
-         indicated by InnerMaxSize rather than InnerSize, think of the case of a dynamic block
-         in a fixed-size matrix
-         However, with EIGEN_UNALIGNED_VECTORIZE and unrolling, slice vectorization is still worth it */
-  };
-
-public:
-  enum {
-    Traversal = int(MayLinearVectorize) && (LinearPacketSize>InnerPacketSize) ? int(LinearVectorizedTraversal)
-              : int(MayInnerVectorize)   ? int(InnerVectorizedTraversal)
-              : int(MayLinearVectorize)  ? int(LinearVectorizedTraversal)
-              : int(MaySliceVectorize)   ? int(SliceVectorizedTraversal)
-              : int(MayLinearize)        ? int(LinearTraversal)
-                                         : int(DefaultTraversal),
-    Vectorized = int(Traversal) == InnerVectorizedTraversal
-              || int(Traversal) == LinearVectorizedTraversal
-              || int(Traversal) == SliceVectorizedTraversal
-  };
-
-  typedef typename conditional<int(Traversal)==LinearVectorizedTraversal, LinearPacketType, InnerPacketType>::type PacketType;
-
-private:
-  enum {
-    ActualPacketSize    = int(Traversal)==LinearVectorizedTraversal ? LinearPacketSize
-                        : Vectorized ? InnerPacketSize
-                        : 1,
-    UnrollingLimit      = EIGEN_UNROLLING_LIMIT * ActualPacketSize,
-    MayUnrollCompletely = int(Dst::SizeAtCompileTime) != Dynamic
-                       && int(Dst::SizeAtCompileTime) * (int(DstEvaluator::CoeffReadCost)+int(SrcEvaluator::CoeffReadCost)) <= int(UnrollingLimit),
-    MayUnrollInner      = int(InnerSize) != Dynamic
-                       && int(InnerSize) * (int(DstEvaluator::CoeffReadCost)+int(SrcEvaluator::CoeffReadCost)) <= int(UnrollingLimit)
-  };
-
-public:
-  enum {
-    Unrolling = (int(Traversal) == int(InnerVectorizedTraversal) || int(Traversal) == int(DefaultTraversal))
-                ? (
-                    int(MayUnrollCompletely) ? int(CompleteUnrolling)
-                  : int(MayUnrollInner)      ? int(InnerUnrolling)
-                                             : int(NoUnrolling)
-                  )
-              : int(Traversal) == int(LinearVectorizedTraversal)
-                ? ( bool(MayUnrollCompletely) && ( EIGEN_UNALIGNED_VECTORIZE || (int(DstAlignment)>=int(LinearRequiredAlignment)))
-                          ? int(CompleteUnrolling)
-                          : int(NoUnrolling) )
-              : int(Traversal) == int(LinearTraversal)
-                ? ( bool(MayUnrollCompletely) ? int(CompleteUnrolling) 
-                                              : int(NoUnrolling) )
-#if EIGEN_UNALIGNED_VECTORIZE
-              : int(Traversal) == int(SliceVectorizedTraversal)
-                ? ( bool(MayUnrollInner) ? int(InnerUnrolling)
-                                         : int(NoUnrolling) )
-#endif
-              : int(NoUnrolling)
-  };
-
-#ifdef EIGEN_DEBUG_ASSIGN
-  static void debug()
-  {
-    std::cerr << "DstXpr: " << typeid(typename DstEvaluator::XprType).name() << std::endl;
-    std::cerr << "SrcXpr: " << typeid(typename SrcEvaluator::XprType).name() << std::endl;
-    std::cerr.setf(std::ios::hex, std::ios::basefield);
-    std::cerr << "DstFlags" << " = " << DstFlags << " (" << demangle_flags(DstFlags) << " )" << std::endl;
-    std::cerr << "SrcFlags" << " = " << SrcFlags << " (" << demangle_flags(SrcFlags) << " )" << std::endl;
-    std::cerr.unsetf(std::ios::hex);
-    EIGEN_DEBUG_VAR(DstAlignment)
-    EIGEN_DEBUG_VAR(SrcAlignment)
-    EIGEN_DEBUG_VAR(LinearRequiredAlignment)
-    EIGEN_DEBUG_VAR(InnerRequiredAlignment)
-    EIGEN_DEBUG_VAR(JointAlignment)
-    EIGEN_DEBUG_VAR(InnerSize)
-    EIGEN_DEBUG_VAR(InnerMaxSize)
-    EIGEN_DEBUG_VAR(LinearPacketSize)
-    EIGEN_DEBUG_VAR(InnerPacketSize)
-    EIGEN_DEBUG_VAR(ActualPacketSize)
-    EIGEN_DEBUG_VAR(StorageOrdersAgree)
-    EIGEN_DEBUG_VAR(MightVectorize)
-    EIGEN_DEBUG_VAR(MayLinearize)
-    EIGEN_DEBUG_VAR(MayInnerVectorize)
-    EIGEN_DEBUG_VAR(MayLinearVectorize)
-    EIGEN_DEBUG_VAR(MaySliceVectorize)
-    std::cerr << "Traversal" << " = " << Traversal << " (" << demangle_traversal(Traversal) << ")" << std::endl;
-    EIGEN_DEBUG_VAR(SrcEvaluator::CoeffReadCost)
-    EIGEN_DEBUG_VAR(UnrollingLimit)
-    EIGEN_DEBUG_VAR(MayUnrollCompletely)
-    EIGEN_DEBUG_VAR(MayUnrollInner)
-    std::cerr << "Unrolling" << " = " << Unrolling << " (" << demangle_unrolling(Unrolling) << ")" << std::endl;
-    std::cerr << std::endl;
-  }
-#endif
-};
-
-/***************************************************************************
-* Part 2 : meta-unrollers
-***************************************************************************/
-
-/************************
-*** Default traversal ***
-************************/
-
-template<typename Kernel, int Index, int Stop>
-struct copy_using_evaluator_DefaultTraversal_CompleteUnrolling
-{
-  // FIXME: this is not very clean, perhaps this information should be provided by the kernel?
-  typedef typename Kernel::DstEvaluatorType DstEvaluatorType;
-  typedef typename DstEvaluatorType::XprType DstXprType;
-  
-  enum {
-    outer = Index / DstXprType::InnerSizeAtCompileTime,
-    inner = Index % DstXprType::InnerSizeAtCompileTime
-  };
-
-  EIGEN_DEVICE_FUNC static EIGEN_STRONG_INLINE void run(Kernel &kernel)
-  {
-    kernel.assignCoeffByOuterInner(outer, inner);
-    copy_using_evaluator_DefaultTraversal_CompleteUnrolling<Kernel, Index+1, Stop>::run(kernel);
-  }
-};
-
-template<typename Kernel, int Stop>
-struct copy_using_evaluator_DefaultTraversal_CompleteUnrolling<Kernel, Stop, Stop>
-{
-  EIGEN_DEVICE_FUNC static EIGEN_STRONG_INLINE void run(Kernel&) { }
-};
-
-template<typename Kernel, int Index_, int Stop>
-struct copy_using_evaluator_DefaultTraversal_InnerUnrolling
-{
-  EIGEN_DEVICE_FUNC static EIGEN_STRONG_INLINE void run(Kernel &kernel, Index outer)
-  {
-    kernel.assignCoeffByOuterInner(outer, Index_);
-    copy_using_evaluator_DefaultTraversal_InnerUnrolling<Kernel, Index_+1, Stop>::run(kernel, outer);
-  }
-};
-
-template<typename Kernel, int Stop>
-struct copy_using_evaluator_DefaultTraversal_InnerUnrolling<Kernel, Stop, Stop>
-{
-  EIGEN_DEVICE_FUNC static EIGEN_STRONG_INLINE void run(Kernel&, Index) { }
-};
-
-/***********************
-*** Linear traversal ***
-***********************/
-
-template<typename Kernel, int Index, int Stop>
-struct copy_using_evaluator_LinearTraversal_CompleteUnrolling
-{
-  EIGEN_DEVICE_FUNC static EIGEN_STRONG_INLINE void run(Kernel& kernel)
-  {
-    kernel.assignCoeff(Index);
-    copy_using_evaluator_LinearTraversal_CompleteUnrolling<Kernel, Index+1, Stop>::run(kernel);
-  }
-};
-
-template<typename Kernel, int Stop>
-struct copy_using_evaluator_LinearTraversal_CompleteUnrolling<Kernel, Stop, Stop>
-{
-  EIGEN_DEVICE_FUNC static EIGEN_STRONG_INLINE void run(Kernel&) { }
-};
-
-/**************************
-*** Inner vectorization ***
-**************************/
-
-template<typename Kernel, int Index, int Stop>
-struct copy_using_evaluator_innervec_CompleteUnrolling
-{
-  // FIXME: this is not very clean, perhaps this information should be provided by the kernel?
-  typedef typename Kernel::DstEvaluatorType DstEvaluatorType;
-  typedef typename DstEvaluatorType::XprType DstXprType;
-  typedef typename Kernel::PacketType PacketType;
-  
-  enum {
-    outer = Index / DstXprType::InnerSizeAtCompileTime,
-    inner = Index % DstXprType::InnerSizeAtCompileTime,
-    SrcAlignment = Kernel::AssignmentTraits::SrcAlignment,
-    DstAlignment = Kernel::AssignmentTraits::DstAlignment
-  };
-
-  EIGEN_DEVICE_FUNC static EIGEN_STRONG_INLINE void run(Kernel &kernel)
-  {
-    kernel.template assignPacketByOuterInner<DstAlignment, SrcAlignment, PacketType>(outer, inner);
-    enum { NextIndex = Index + unpacket_traits<PacketType>::size };
-    copy_using_evaluator_innervec_CompleteUnrolling<Kernel, NextIndex, Stop>::run(kernel);
-  }
-};
-
-template<typename Kernel, int Stop>
-struct copy_using_evaluator_innervec_CompleteUnrolling<Kernel, Stop, Stop>
-{
-  EIGEN_DEVICE_FUNC static EIGEN_STRONG_INLINE void run(Kernel&) { }
-};
-
-template<typename Kernel, int Index_, int Stop, int SrcAlignment, int DstAlignment>
-struct copy_using_evaluator_innervec_InnerUnrolling
-{
-  typedef typename Kernel::PacketType PacketType;
-  EIGEN_DEVICE_FUNC static EIGEN_STRONG_INLINE void run(Kernel &kernel, Index outer)
-  {
-    kernel.template assignPacketByOuterInner<DstAlignment, SrcAlignment, PacketType>(outer, Index_);
-    enum { NextIndex = Index_ + unpacket_traits<PacketType>::size };
-    copy_using_evaluator_innervec_InnerUnrolling<Kernel, NextIndex, Stop, SrcAlignment, DstAlignment>::run(kernel, outer);
-  }
-};
-
-template<typename Kernel, int Stop, int SrcAlignment, int DstAlignment>
-struct copy_using_evaluator_innervec_InnerUnrolling<Kernel, Stop, Stop, SrcAlignment, DstAlignment>
-{
-  EIGEN_DEVICE_FUNC static EIGEN_STRONG_INLINE void run(Kernel &, Index) { }
-};
-
-/***************************************************************************
-* Part 3 : implementation of all cases
-***************************************************************************/
-
-// dense_assignment_loop is based on assign_impl
-
-template<typename Kernel,
-         int Traversal = Kernel::AssignmentTraits::Traversal,
-         int Unrolling = Kernel::AssignmentTraits::Unrolling>
-struct dense_assignment_loop;
-
-/************************
-*** Default traversal ***
-************************/
-
-template<typename Kernel>
-struct dense_assignment_loop<Kernel, DefaultTraversal, NoUnrolling>
-{
-  EIGEN_DEVICE_FUNC static void EIGEN_STRONG_INLINE run(Kernel &kernel)
-  {
-    for(Index outer = 0; outer < kernel.outerSize(); ++outer) {
-      for(Index inner = 0; inner < kernel.innerSize(); ++inner) {
-        kernel.assignCoeffByOuterInner(outer, inner);
-      }
-    }
-  }
-};
-
-template<typename Kernel>
-struct dense_assignment_loop<Kernel, DefaultTraversal, CompleteUnrolling>
-{
-  EIGEN_DEVICE_FUNC static EIGEN_STRONG_INLINE void run(Kernel &kernel)
-  {
-    typedef typename Kernel::DstEvaluatorType::XprType DstXprType;
-    copy_using_evaluator_DefaultTraversal_CompleteUnrolling<Kernel, 0, DstXprType::SizeAtCompileTime>::run(kernel);
-  }
-};
-
-template<typename Kernel>
-struct dense_assignment_loop<Kernel, DefaultTraversal, InnerUnrolling>
-{
-  EIGEN_DEVICE_FUNC static EIGEN_STRONG_INLINE void run(Kernel &kernel)
-  {
-    typedef typename Kernel::DstEvaluatorType::XprType DstXprType;
-
-    const Index outerSize = kernel.outerSize();
-    for(Index outer = 0; outer < outerSize; ++outer)
-      copy_using_evaluator_DefaultTraversal_InnerUnrolling<Kernel, 0, DstXprType::InnerSizeAtCompileTime>::run(kernel, outer);
-  }
-};
-
-/***************************
-*** Linear vectorization ***
-***************************/
-
-
-// The goal of unaligned_dense_assignment_loop is simply to factorize the handling
-// of the non vectorizable beginning and ending parts
-
-template <bool IsAligned = false>
-struct unaligned_dense_assignment_loop
-{
-  // if IsAligned = true, then do nothing
-  template <typename Kernel>
-  EIGEN_DEVICE_FUNC static EIGEN_STRONG_INLINE void run(Kernel&, Index, Index) {}
-};
-
-template <>
-struct unaligned_dense_assignment_loop<false>
-{
-  // MSVC must not inline this functions. If it does, it fails to optimize the
-  // packet access path.
-  // FIXME check which version exhibits this issue
-#if EIGEN_COMP_MSVC
-  template <typename Kernel>
-  static EIGEN_DONT_INLINE void run(Kernel &kernel,
-                                    Index start,
-                                    Index end)
-#else
-  template <typename Kernel>
-  EIGEN_DEVICE_FUNC static EIGEN_STRONG_INLINE void run(Kernel &kernel,
-                                      Index start,
-                                      Index end)
-#endif
-  {
-    for (Index index = start; index < end; ++index)
-      kernel.assignCoeff(index);
-  }
-};
-
-template<typename Kernel>
-struct dense_assignment_loop<Kernel, LinearVectorizedTraversal, NoUnrolling>
-{
-  EIGEN_DEVICE_FUNC static EIGEN_STRONG_INLINE void run(Kernel &kernel)
-  {
-    const Index size = kernel.size();
-    typedef typename Kernel::Scalar Scalar;
-    typedef typename Kernel::PacketType PacketType;
-    enum {
-      requestedAlignment = Kernel::AssignmentTraits::LinearRequiredAlignment,
-      packetSize = unpacket_traits<PacketType>::size,
-      dstIsAligned = int(Kernel::AssignmentTraits::DstAlignment)>=int(requestedAlignment),
-      dstAlignment = packet_traits<Scalar>::AlignedOnScalar ? int(requestedAlignment)
-                                                            : int(Kernel::AssignmentTraits::DstAlignment),
-      srcAlignment = Kernel::AssignmentTraits::JointAlignment
-    };
-    const Index alignedStart = dstIsAligned ? 0 : internal::first_aligned<requestedAlignment>(kernel.dstDataPtr(), size);
-    const Index alignedEnd = alignedStart + ((size-alignedStart)/packetSize)*packetSize;
-
-    unaligned_dense_assignment_loop<dstIsAligned!=0>::run(kernel, 0, alignedStart);
-
-    for(Index index = alignedStart; index < alignedEnd; index += packetSize)
-      kernel.template assignPacket<dstAlignment, srcAlignment, PacketType>(index);
-
-    unaligned_dense_assignment_loop<>::run(kernel, alignedEnd, size);
-  }
-};
-
-template<typename Kernel>
-struct dense_assignment_loop<Kernel, LinearVectorizedTraversal, CompleteUnrolling>
-{
-  EIGEN_DEVICE_FUNC static EIGEN_STRONG_INLINE void run(Kernel &kernel)
-  {
-    typedef typename Kernel::DstEvaluatorType::XprType DstXprType;
-    typedef typename Kernel::PacketType PacketType;
-    
-    enum { size = DstXprType::SizeAtCompileTime,
-           packetSize =unpacket_traits<PacketType>::size,
-           alignedSize = (size/packetSize)*packetSize };
-
-    copy_using_evaluator_innervec_CompleteUnrolling<Kernel, 0, alignedSize>::run(kernel);
-    copy_using_evaluator_DefaultTraversal_CompleteUnrolling<Kernel, alignedSize, size>::run(kernel);
-  }
-};
-
-/**************************
-*** Inner vectorization ***
-**************************/
-
-template<typename Kernel>
-struct dense_assignment_loop<Kernel, InnerVectorizedTraversal, NoUnrolling>
-{
-  typedef typename Kernel::PacketType PacketType;
-  enum {
-    SrcAlignment = Kernel::AssignmentTraits::SrcAlignment,
-    DstAlignment = Kernel::AssignmentTraits::DstAlignment
-  };
-  EIGEN_DEVICE_FUNC static EIGEN_STRONG_INLINE void run(Kernel &kernel)
-  {
-    const Index innerSize = kernel.innerSize();
-    const Index outerSize = kernel.outerSize();
-    const Index packetSize = unpacket_traits<PacketType>::size;
-    for(Index outer = 0; outer < outerSize; ++outer)
-      for(Index inner = 0; inner < innerSize; inner+=packetSize)
-        kernel.template assignPacketByOuterInner<DstAlignment, SrcAlignment, PacketType>(outer, inner);
-  }
-};
-
-template<typename Kernel>
-struct dense_assignment_loop<Kernel, InnerVectorizedTraversal, CompleteUnrolling>
-{
-  EIGEN_DEVICE_FUNC static EIGEN_STRONG_INLINE void run(Kernel &kernel)
-  {
-    typedef typename Kernel::DstEvaluatorType::XprType DstXprType;
-    copy_using_evaluator_innervec_CompleteUnrolling<Kernel, 0, DstXprType::SizeAtCompileTime>::run(kernel);
-  }
-};
-
-template<typename Kernel>
-struct dense_assignment_loop<Kernel, InnerVectorizedTraversal, InnerUnrolling>
-{
-  EIGEN_DEVICE_FUNC static EIGEN_STRONG_INLINE void run(Kernel &kernel)
-  {
-    typedef typename Kernel::DstEvaluatorType::XprType DstXprType;
-    typedef typename Kernel::AssignmentTraits Traits;
-    const Index outerSize = kernel.outerSize();
-    for(Index outer = 0; outer < outerSize; ++outer)
-      copy_using_evaluator_innervec_InnerUnrolling<Kernel, 0, DstXprType::InnerSizeAtCompileTime,
-                                                   Traits::SrcAlignment, Traits::DstAlignment>::run(kernel, outer);
-  }
-};
-
-/***********************
-*** Linear traversal ***
-***********************/
-
-template<typename Kernel>
-struct dense_assignment_loop<Kernel, LinearTraversal, NoUnrolling>
-{
-  EIGEN_DEVICE_FUNC static EIGEN_STRONG_INLINE void run(Kernel &kernel)
-  {
-    const Index size = kernel.size();
-    for(Index i = 0; i < size; ++i)
-      kernel.assignCoeff(i);
-  }
-};
-
-template<typename Kernel>
-struct dense_assignment_loop<Kernel, LinearTraversal, CompleteUnrolling>
-{
-  EIGEN_DEVICE_FUNC static EIGEN_STRONG_INLINE void run(Kernel &kernel)
-  {
-    typedef typename Kernel::DstEvaluatorType::XprType DstXprType;
-    copy_using_evaluator_LinearTraversal_CompleteUnrolling<Kernel, 0, DstXprType::SizeAtCompileTime>::run(kernel);
-  }
-};
-
-/**************************
-*** Slice vectorization ***
-***************************/
-
-template<typename Kernel>
-struct dense_assignment_loop<Kernel, SliceVectorizedTraversal, NoUnrolling>
-{
-  EIGEN_DEVICE_FUNC static inline void run(Kernel &kernel)
-  {
-    typedef typename Kernel::Scalar Scalar;
-    typedef typename Kernel::PacketType PacketType;
-    enum {
-      packetSize = unpacket_traits<PacketType>::size,
-      requestedAlignment = int(Kernel::AssignmentTraits::InnerRequiredAlignment),
-      alignable = packet_traits<Scalar>::AlignedOnScalar || int(Kernel::AssignmentTraits::DstAlignment)>=sizeof(Scalar),
-      dstIsAligned = int(Kernel::AssignmentTraits::DstAlignment)>=int(requestedAlignment),
-      dstAlignment = alignable ? int(requestedAlignment)
-                               : int(Kernel::AssignmentTraits::DstAlignment)
-    };
-    const Scalar *dst_ptr = kernel.dstDataPtr();
-    if((!bool(dstIsAligned)) && (UIntPtr(dst_ptr) % sizeof(Scalar))>0)
-    {
-      // the pointer is not aligend-on scalar, so alignment is not possible
-      return dense_assignment_loop<Kernel,DefaultTraversal,NoUnrolling>::run(kernel);
-    }
-    const Index packetAlignedMask = packetSize - 1;
-    const Index innerSize = kernel.innerSize();
-    const Index outerSize = kernel.outerSize();
-    const Index alignedStep = alignable ? (packetSize - kernel.outerStride() % packetSize) & packetAlignedMask : 0;
-    Index alignedStart = ((!alignable) || bool(dstIsAligned)) ? 0 : internal::first_aligned<requestedAlignment>(dst_ptr, innerSize);
-
-    for(Index outer = 0; outer < outerSize; ++outer)
-    {
-      const Index alignedEnd = alignedStart + ((innerSize-alignedStart) & ~packetAlignedMask);
-      // do the non-vectorizable part of the assignment
-      for(Index inner = 0; inner<alignedStart ; ++inner)
-        kernel.assignCoeffByOuterInner(outer, inner);
-
-      // do the vectorizable part of the assignment
-      for(Index inner = alignedStart; inner<alignedEnd; inner+=packetSize)
-        kernel.template assignPacketByOuterInner<dstAlignment, Unaligned, PacketType>(outer, inner);
-
-      // do the non-vectorizable part of the assignment
-      for(Index inner = alignedEnd; inner<innerSize ; ++inner)
-        kernel.assignCoeffByOuterInner(outer, inner);
-
-      alignedStart = numext::mini((alignedStart+alignedStep)%packetSize, innerSize);
-    }
-  }
-};
-
-#if EIGEN_UNALIGNED_VECTORIZE
-template<typename Kernel>
-struct dense_assignment_loop<Kernel, SliceVectorizedTraversal, InnerUnrolling>
-{
-  EIGEN_DEVICE_FUNC static inline void run(Kernel &kernel)
-  {
-    typedef typename Kernel::DstEvaluatorType::XprType DstXprType;
-    typedef typename Kernel::PacketType PacketType;
-
-    enum { size = DstXprType::InnerSizeAtCompileTime,
-           packetSize =unpacket_traits<PacketType>::size,
-           vectorizableSize = (size/packetSize)*packetSize };
-
-    for(Index outer = 0; outer < kernel.outerSize(); ++outer)
-    {
-      copy_using_evaluator_innervec_InnerUnrolling<Kernel, 0, vectorizableSize, 0, 0>::run(kernel, outer);
-      copy_using_evaluator_DefaultTraversal_InnerUnrolling<Kernel, vectorizableSize, size>::run(kernel, outer);
-    }
-  }
-};
-#endif
-
-
-/***************************************************************************
-* Part 4 : Generic dense assignment kernel
-***************************************************************************/
-
-// This class generalize the assignment of a coefficient (or packet) from one dense evaluator
-// to another dense writable evaluator.
-// It is parametrized by the two evaluators, and the actual assignment functor.
-// This abstraction level permits to keep the evaluation loops as simple and as generic as possible.
-// One can customize the assignment using this generic dense_assignment_kernel with different
-// functors, or by completely overloading it, by-passing a functor.
-template<typename DstEvaluatorTypeT, typename SrcEvaluatorTypeT, typename Functor, int Version = Specialized>
-class generic_dense_assignment_kernel
-{
-protected:
-  typedef typename DstEvaluatorTypeT::XprType DstXprType;
-  typedef typename SrcEvaluatorTypeT::XprType SrcXprType;
-public:
-  
-  typedef DstEvaluatorTypeT DstEvaluatorType;
-  typedef SrcEvaluatorTypeT SrcEvaluatorType;
-  typedef typename DstEvaluatorType::Scalar Scalar;
-  typedef copy_using_evaluator_traits<DstEvaluatorTypeT, SrcEvaluatorTypeT, Functor> AssignmentTraits;
-  typedef typename AssignmentTraits::PacketType PacketType;
-  
-  
-  EIGEN_DEVICE_FUNC generic_dense_assignment_kernel(DstEvaluatorType &dst, const SrcEvaluatorType &src, const Functor &func, DstXprType& dstExpr)
-    : m_dst(dst), m_src(src), m_functor(func), m_dstExpr(dstExpr)
-  {
-    #ifdef EIGEN_DEBUG_ASSIGN
-    AssignmentTraits::debug();
-    #endif
-  }
-  
-  EIGEN_DEVICE_FUNC Index size() const        { return m_dstExpr.size(); }
-  EIGEN_DEVICE_FUNC Index innerSize() const   { return m_dstExpr.innerSize(); }
-  EIGEN_DEVICE_FUNC Index outerSize() const   { return m_dstExpr.outerSize(); }
-  EIGEN_DEVICE_FUNC Index rows() const        { return m_dstExpr.rows(); }
-  EIGEN_DEVICE_FUNC Index cols() const        { return m_dstExpr.cols(); }
-  EIGEN_DEVICE_FUNC Index outerStride() const { return m_dstExpr.outerStride(); }
-  
-  EIGEN_DEVICE_FUNC DstEvaluatorType& dstEvaluator() { return m_dst; }
-  EIGEN_DEVICE_FUNC const SrcEvaluatorType& srcEvaluator() const { return m_src; }
-  
-  /// Assign src(row,col) to dst(row,col) through the assignment functor.
-  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE void assignCoeff(Index row, Index col)
-  {
-    m_functor.assignCoeff(m_dst.coeffRef(row,col), m_src.coeff(row,col));
-  }
-  
-  /// \sa assignCoeff(Index,Index)
-  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE void assignCoeff(Index index)
-  {
-    m_functor.assignCoeff(m_dst.coeffRef(index), m_src.coeff(index));
-  }
-  
-  /// \sa assignCoeff(Index,Index)
-  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE void assignCoeffByOuterInner(Index outer, Index inner)
-  {
-    Index row = rowIndexByOuterInner(outer, inner); 
-    Index col = colIndexByOuterInner(outer, inner); 
-    assignCoeff(row, col);
-  }
-  
-  
-  template<int StoreMode, int LoadMode, typename PacketType>
-  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE void assignPacket(Index row, Index col)
-  {
-    m_functor.template assignPacket<StoreMode>(&m_dst.coeffRef(row,col), m_src.template packet<LoadMode,PacketType>(row,col));
-  }
-  
-  template<int StoreMode, int LoadMode, typename PacketType>
-  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE void assignPacket(Index index)
-  {
-    m_functor.template assignPacket<StoreMode>(&m_dst.coeffRef(index), m_src.template packet<LoadMode,PacketType>(index));
-  }
-  
-  template<int StoreMode, int LoadMode, typename PacketType>
-  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE void assignPacketByOuterInner(Index outer, Index inner)
-  {
-    Index row = rowIndexByOuterInner(outer, inner); 
-    Index col = colIndexByOuterInner(outer, inner);
-    assignPacket<StoreMode,LoadMode,PacketType>(row, col);
-  }
-  
-  EIGEN_DEVICE_FUNC static EIGEN_STRONG_INLINE Index rowIndexByOuterInner(Index outer, Index inner)
-  {
-    typedef typename DstEvaluatorType::ExpressionTraits Traits;
-    return int(Traits::RowsAtCompileTime) == 1 ? 0
-      : int(Traits::ColsAtCompileTime) == 1 ? inner
-      : int(DstEvaluatorType::Flags)&RowMajorBit ? outer
-      : inner;
-  }
-
-  EIGEN_DEVICE_FUNC static EIGEN_STRONG_INLINE Index colIndexByOuterInner(Index outer, Index inner)
-  {
-    typedef typename DstEvaluatorType::ExpressionTraits Traits;
-    return int(Traits::ColsAtCompileTime) == 1 ? 0
-      : int(Traits::RowsAtCompileTime) == 1 ? inner
-      : int(DstEvaluatorType::Flags)&RowMajorBit ? inner
-      : outer;
-  }
-
-  EIGEN_DEVICE_FUNC const Scalar* dstDataPtr() const
-  {
-    return m_dstExpr.data();
-  }
-  
-protected:
-  DstEvaluatorType& m_dst;
-  const SrcEvaluatorType& m_src;
-  const Functor &m_functor;
-  // TODO find a way to avoid the needs of the original expression
-  DstXprType& m_dstExpr;
-};
-
-/***************************************************************************
-* Part 5 : Entry point for dense rectangular assignment
-***************************************************************************/
-
-template<typename DstXprType, typename SrcXprType, typename Functor>
-EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE void call_dense_assignment_loop(DstXprType& dst, const SrcXprType& src, const Functor &func)
-{
-  typedef evaluator<DstXprType> DstEvaluatorType;
-  typedef evaluator<SrcXprType> SrcEvaluatorType;
-
-  SrcEvaluatorType srcEvaluator(src);
-
-  // NOTE To properly handle A = (A*A.transpose())/s with A rectangular,
-  // we need to resize the destination after the source evaluator has been created.
-  Index dstRows = src.rows();
-  Index dstCols = src.cols();
-  if((dst.rows()!=dstRows) || (dst.cols()!=dstCols))
-    dst.resize(dstRows, dstCols);
-
-  DstEvaluatorType dstEvaluator(dst);
-    
-  typedef generic_dense_assignment_kernel<DstEvaluatorType,SrcEvaluatorType,Functor> Kernel;
-  Kernel kernel(dstEvaluator, srcEvaluator, func, dst.const_cast_derived());
-
-  dense_assignment_loop<Kernel>::run(kernel);
-}
-
-template<typename DstXprType, typename SrcXprType>
-EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE void call_dense_assignment_loop(DstXprType& dst, const SrcXprType& src)
-{
-  call_dense_assignment_loop(dst, src, internal::assign_op<typename DstXprType::Scalar,typename SrcXprType::Scalar>());
-}
-
-/***************************************************************************
-* Part 6 : Generic assignment
-***************************************************************************/
-
-// Based on the respective shapes of the destination and source,
-// the class AssignmentKind determine the kind of assignment mechanism.
-// AssignmentKind must define a Kind typedef.
-template<typename DstShape, typename SrcShape> struct AssignmentKind;
-
-// Assignement kind defined in this file:
-struct Dense2Dense {};
-struct EigenBase2EigenBase {};
-
-template<typename,typename> struct AssignmentKind { typedef EigenBase2EigenBase Kind; };
-template<> struct AssignmentKind<DenseShape,DenseShape> { typedef Dense2Dense Kind; };
-    
-// This is the main assignment class
-template< typename DstXprType, typename SrcXprType, typename Functor,
-          typename Kind = typename AssignmentKind< typename evaluator_traits<DstXprType>::Shape , typename evaluator_traits<SrcXprType>::Shape >::Kind,
-          typename EnableIf = void>
-struct Assignment;
-
-
-// The only purpose of this call_assignment() function is to deal with noalias() / "assume-aliasing" and automatic transposition.
-// Indeed, I (Gael) think that this concept of "assume-aliasing" was a mistake, and it makes thing quite complicated.
-// So this intermediate function removes everything related to "assume-aliasing" such that Assignment
-// does not has to bother about these annoying details.
-
-template<typename Dst, typename Src>
-EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
-void call_assignment(Dst& dst, const Src& src)
-{
-  call_assignment(dst, src, internal::assign_op<typename Dst::Scalar,typename Src::Scalar>());
-}
-template<typename Dst, typename Src>
-EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
-void call_assignment(const Dst& dst, const Src& src)
-{
-  call_assignment(dst, src, internal::assign_op<typename Dst::Scalar,typename Src::Scalar>());
-}
-                     
-// Deal with "assume-aliasing"
-template<typename Dst, typename Src, typename Func>
-EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
-void call_assignment(Dst& dst, const Src& src, const Func& func, typename enable_if< evaluator_assume_aliasing<Src>::value, void*>::type = 0)
-{
-  typename plain_matrix_type<Src>::type tmp(src);
-  call_assignment_no_alias(dst, tmp, func);
-}
-
-template<typename Dst, typename Src, typename Func>
-EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
-void call_assignment(Dst& dst, const Src& src, const Func& func, typename enable_if<!evaluator_assume_aliasing<Src>::value, void*>::type = 0)
-{
-  call_assignment_no_alias(dst, src, func);
-}
-
-// by-pass "assume-aliasing"
-// When there is no aliasing, we require that 'dst' has been properly resized
-template<typename Dst, template <typename> class StorageBase, typename Src, typename Func>
-EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
-void call_assignment(NoAlias<Dst,StorageBase>& dst, const Src& src, const Func& func)
-{
-  call_assignment_no_alias(dst.expression(), src, func);
-}
-
-
-template<typename Dst, typename Src, typename Func>
-EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
-void call_assignment_no_alias(Dst& dst, const Src& src, const Func& func)
-{
-  enum {
-    NeedToTranspose = (    (int(Dst::RowsAtCompileTime) == 1 && int(Src::ColsAtCompileTime) == 1)
-                        || (int(Dst::ColsAtCompileTime) == 1 && int(Src::RowsAtCompileTime) == 1)
-                      ) && int(Dst::SizeAtCompileTime) != 1
-  };
-
-  typedef typename internal::conditional<NeedToTranspose, Transpose<Dst>, Dst>::type ActualDstTypeCleaned;
-  typedef typename internal::conditional<NeedToTranspose, Transpose<Dst>, Dst&>::type ActualDstType;
-  ActualDstType actualDst(dst);
-  
-  // TODO check whether this is the right place to perform these checks:
-  EIGEN_STATIC_ASSERT_LVALUE(Dst)
-  EIGEN_STATIC_ASSERT_SAME_MATRIX_SIZE(ActualDstTypeCleaned,Src)
-  EIGEN_CHECK_BINARY_COMPATIBILIY(Func,typename ActualDstTypeCleaned::Scalar,typename Src::Scalar);
-  
-  Assignment<ActualDstTypeCleaned,Src,Func>::run(actualDst, src, func);
-}
-template<typename Dst, typename Src>
-EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
-void call_assignment_no_alias(Dst& dst, const Src& src)
-{
-  call_assignment_no_alias(dst, src, internal::assign_op<typename Dst::Scalar,typename Src::Scalar>());
-}
-
-template<typename Dst, typename Src, typename Func>
-EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
-void call_assignment_no_alias_no_transpose(Dst& dst, const Src& src, const Func& func)
-{
-  // TODO check whether this is the right place to perform these checks:
-  EIGEN_STATIC_ASSERT_LVALUE(Dst)
-  EIGEN_STATIC_ASSERT_SAME_MATRIX_SIZE(Dst,Src)
-  EIGEN_CHECK_BINARY_COMPATIBILIY(Func,typename Dst::Scalar,typename Src::Scalar);
-
-  Assignment<Dst,Src,Func>::run(dst, src, func);
-}
-template<typename Dst, typename Src>
-EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
-void call_assignment_no_alias_no_transpose(Dst& dst, const Src& src)
-{
-  call_assignment_no_alias_no_transpose(dst, src, internal::assign_op<typename Dst::Scalar,typename Src::Scalar>());
-}
-
-// forward declaration
-template<typename Dst, typename Src> void check_for_aliasing(const Dst &dst, const Src &src);
-
-// Generic Dense to Dense assignment
-// Note that the last template argument "Weak" is needed to make it possible to perform
-// both partial specialization+SFINAE without ambiguous specialization
-template< typename DstXprType, typename SrcXprType, typename Functor, typename Weak>
-struct Assignment<DstXprType, SrcXprType, Functor, Dense2Dense, Weak>
-{
-  EIGEN_DEVICE_FUNC
-  static EIGEN_STRONG_INLINE void run(DstXprType &dst, const SrcXprType &src, const Functor &func)
-  {
-#ifndef EIGEN_NO_DEBUG
-    internal::check_for_aliasing(dst, src);
-#endif
-    
-    call_dense_assignment_loop(dst, src, func);
-  }
-};
-
-// Generic assignment through evalTo.
-// TODO: not sure we have to keep that one, but it helps porting current code to new evaluator mechanism.
-// Note that the last template argument "Weak" is needed to make it possible to perform
-// both partial specialization+SFINAE without ambiguous specialization
-template< typename DstXprType, typename SrcXprType, typename Functor, typename Weak>
-struct Assignment<DstXprType, SrcXprType, Functor, EigenBase2EigenBase, Weak>
-{
-  EIGEN_DEVICE_FUNC
-  static EIGEN_STRONG_INLINE void run(DstXprType &dst, const SrcXprType &src, const internal::assign_op<typename DstXprType::Scalar,typename SrcXprType::Scalar> &/*func*/)
-  {
-    Index dstRows = src.rows();
-    Index dstCols = src.cols();
-    if((dst.rows()!=dstRows) || (dst.cols()!=dstCols))
-      dst.resize(dstRows, dstCols);
-
-    eigen_assert(dst.rows() == src.rows() && dst.cols() == src.cols());
-    src.evalTo(dst);
-  }
-
-  // NOTE The following two functions are templated to avoid their instanciation if not needed
-  //      This is needed because some expressions supports evalTo only and/or have 'void' as scalar type.
-  template<typename SrcScalarType>
-  EIGEN_DEVICE_FUNC
-  static EIGEN_STRONG_INLINE void run(DstXprType &dst, const SrcXprType &src, const internal::add_assign_op<typename DstXprType::Scalar,SrcScalarType> &/*func*/)
-  {
-    Index dstRows = src.rows();
-    Index dstCols = src.cols();
-    if((dst.rows()!=dstRows) || (dst.cols()!=dstCols))
-      dst.resize(dstRows, dstCols);
-
-    eigen_assert(dst.rows() == src.rows() && dst.cols() == src.cols());
-    src.addTo(dst);
-  }
-
-  template<typename SrcScalarType>
-  EIGEN_DEVICE_FUNC
-  static EIGEN_STRONG_INLINE void run(DstXprType &dst, const SrcXprType &src, const internal::sub_assign_op<typename DstXprType::Scalar,SrcScalarType> &/*func*/)
-  {
-    Index dstRows = src.rows();
-    Index dstCols = src.cols();
-    if((dst.rows()!=dstRows) || (dst.cols()!=dstCols))
-      dst.resize(dstRows, dstCols);
-
-    eigen_assert(dst.rows() == src.rows() && dst.cols() == src.cols());
-    src.subTo(dst);
-  }
-};
-
-} // namespace internal
-
-} // end namespace Eigen
-
-#endif // EIGEN_ASSIGN_EVALUATOR_H

Eigen/src/Core/Assign_MKL.h

@@ -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)
@@ -54,120 +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_LA , VML_HA
+#define  EIGEN_MKL_VML_MODE VML_HA
 #else
-#define EIGEN_VMLMODE_EXPAND_LA , VML_LA
+#define  EIGEN_MKL_VML_MODE VML_LA
 #endif
 
-#define EIGEN_VMLMODE_EXPAND__ 
-
-#define EIGEN_VMLMODE_PREFIX_LA vm
-#define EIGEN_VMLMODE_PREFIX__  v
-#define EIGEN_VMLMODE_PREFIX(VMLMODE) EIGEN_CAT(EIGEN_VMLMODE_PREFIX_,VMLMODE)
-
-#define EIGEN_MKL_VML_DECLARE_UNARY_CALL(EIGENOP, VMLOP, EIGENTYPE, VMLTYPE, VMLMODE)                                           \
-  template< typename DstXprType, typename SrcXprNested>                                                                         \
-  struct Assignment<DstXprType, CwiseUnaryOp<scalar_##EIGENOP##_op<EIGENTYPE>, SrcXprNested>, assign_op<EIGENTYPE,EIGENTYPE>,   \
-                   Dense2Dense, typename enable_if<vml_assign_traits<DstXprType,SrcXprNested>::EnableVml>::type> {              \
-    typedef CwiseUnaryOp<scalar_##EIGENOP##_op<EIGENTYPE>, SrcXprNested> SrcXprType;                                            \
-    static void run(DstXprType &dst, const SrcXprType &src, const assign_op<EIGENTYPE,EIGENTYPE> &/*func*/) {                   \
-      eigen_assert(dst.rows() == src.rows() && dst.cols() == src.cols());                                                       \
-      if(vml_assign_traits<DstXprType,SrcXprNested>::Traversal==LinearTraversal) {                                              \
-        VMLOP(dst.size(), (const VMLTYPE*)src.nestedExpression().data(),                                                        \
-              (VMLTYPE*)dst.data() EIGEN_PP_EXPAND(EIGEN_VMLMODE_EXPAND_##VMLMODE) );                                           \
-      } else {                                                                                                                  \
-        const Index outerSize = dst.outerSize();                                                                                \
-        for(Index outer = 0; outer < outerSize; ++outer) {                                                                      \
-          const EIGENTYPE *src_ptr = src.IsRowMajor ? &(src.nestedExpression().coeffRef(outer,0)) :                             \
-                                                      &(src.nestedExpression().coeffRef(0, outer));                             \
-          EIGENTYPE *dst_ptr = dst.IsRowMajor ? &(dst.coeffRef(outer,0)) : &(dst.coeffRef(0, outer));                           \
-          VMLOP( dst.innerSize(), (const VMLTYPE*)src_ptr,                                                                      \
-                (VMLTYPE*)dst_ptr EIGEN_PP_EXPAND(EIGEN_VMLMODE_EXPAND_##VMLMODE));                                             \
-        }                                                                                                                       \
-      }                                                                                                                         \
-    }                                                                                                                           \
-  };                                                                                                                            \
-
-
-#define EIGEN_MKL_VML_DECLARE_UNARY_CALLS_REAL(EIGENOP, VMLOP, VMLMODE)                                                         \
-  EIGEN_MKL_VML_DECLARE_UNARY_CALL(EIGENOP, EIGEN_CAT(EIGEN_VMLMODE_PREFIX(VMLMODE),s##VMLOP), float, float, VMLMODE)           \
-  EIGEN_MKL_VML_DECLARE_UNARY_CALL(EIGENOP, EIGEN_CAT(EIGEN_VMLMODE_PREFIX(VMLMODE),d##VMLOP), double, double, VMLMODE)
-
-#define EIGEN_MKL_VML_DECLARE_UNARY_CALLS_CPLX(EIGENOP, VMLOP, VMLMODE)                                                         \
-  EIGEN_MKL_VML_DECLARE_UNARY_CALL(EIGENOP, EIGEN_CAT(EIGEN_VMLMODE_PREFIX(VMLMODE),c##VMLOP), scomplex, MKL_Complex8, VMLMODE) \
-  EIGEN_MKL_VML_DECLARE_UNARY_CALL(EIGENOP, EIGEN_CAT(EIGEN_VMLMODE_PREFIX(VMLMODE),z##VMLOP), dcomplex, MKL_Complex16, VMLMODE)
-  
-#define EIGEN_MKL_VML_DECLARE_UNARY_CALLS(EIGENOP, VMLOP, VMLMODE)                                                              \
-  EIGEN_MKL_VML_DECLARE_UNARY_CALLS_REAL(EIGENOP, VMLOP, VMLMODE)                                                               \
-  EIGEN_MKL_VML_DECLARE_UNARY_CALLS_CPLX(EIGENOP, VMLOP, VMLMODE)
-
-  
-EIGEN_MKL_VML_DECLARE_UNARY_CALLS(sin,   Sin,   LA)
-EIGEN_MKL_VML_DECLARE_UNARY_CALLS(asin,  Asin,  LA)
-EIGEN_MKL_VML_DECLARE_UNARY_CALLS(sinh,  Sinh,  LA)
-EIGEN_MKL_VML_DECLARE_UNARY_CALLS(cos,   Cos,   LA)
-EIGEN_MKL_VML_DECLARE_UNARY_CALLS(acos,  Acos,  LA)
-EIGEN_MKL_VML_DECLARE_UNARY_CALLS(cosh,  Cosh,  LA)
-EIGEN_MKL_VML_DECLARE_UNARY_CALLS(tan,   Tan,   LA)
-EIGEN_MKL_VML_DECLARE_UNARY_CALLS(atan,  Atan,  LA)
-EIGEN_MKL_VML_DECLARE_UNARY_CALLS(tanh,  Tanh,  LA)
-// EIGEN_MKL_VML_DECLARE_UNARY_CALLS(abs,   Abs,    _)
-EIGEN_MKL_VML_DECLARE_UNARY_CALLS(exp,   Exp,   LA)
-EIGEN_MKL_VML_DECLARE_UNARY_CALLS(log,   Ln,    LA)
-EIGEN_MKL_VML_DECLARE_UNARY_CALLS(log10, Log10, LA)
-EIGEN_MKL_VML_DECLARE_UNARY_CALLS(sqrt,  Sqrt,  _)
-
-EIGEN_MKL_VML_DECLARE_UNARY_CALLS_REAL(square, Sqr,   _)
-EIGEN_MKL_VML_DECLARE_UNARY_CALLS_CPLX(arg, Arg,      _)
-EIGEN_MKL_VML_DECLARE_UNARY_CALLS_REAL(round, Round,  _)
-EIGEN_MKL_VML_DECLARE_UNARY_CALLS_REAL(floor, Floor,  _)
-EIGEN_MKL_VML_DECLARE_UNARY_CALLS_REAL(ceil,  Ceil,   _)
-
-#define EIGEN_MKL_VML_DECLARE_POW_CALL(EIGENOP, VMLOP, EIGENTYPE, VMLTYPE, VMLMODE)                                           \
-  template< typename DstXprType, typename SrcXprNested, typename Plain>                                                       \
-  struct Assignment<DstXprType, CwiseBinaryOp<scalar_##EIGENOP##_op<EIGENTYPE,EIGENTYPE>, SrcXprNested,                       \
-                    const CwiseNullaryOp<internal::scalar_constant_op<EIGENTYPE>,Plain> >, assign_op<EIGENTYPE,EIGENTYPE>,    \
-                   Dense2Dense, typename enable_if<vml_assign_traits<DstXprType,SrcXprNested>::EnableVml>::type> {            \
-    typedef CwiseBinaryOp<scalar_##EIGENOP##_op<EIGENTYPE,EIGENTYPE>, SrcXprNested,                                           \
-                    const CwiseNullaryOp<internal::scalar_constant_op<EIGENTYPE>,Plain> > SrcXprType;                         \
-    static void run(DstXprType &dst, const SrcXprType &src, const assign_op<EIGENTYPE,EIGENTYPE> &/*func*/) {                 \
-      eigen_assert(dst.rows() == src.rows() && dst.cols() == src.cols());                                                     \
-      VMLTYPE exponent = reinterpret_cast<const VMLTYPE&>(src.rhs().functor().m_other);                                       \
-      if(vml_assign_traits<DstXprType,SrcXprNested>::Traversal==LinearTraversal)                                              \
-      {                                                                                                                       \
-        VMLOP( dst.size(), (const VMLTYPE*)src.lhs().data(), exponent,                                                        \
-              (VMLTYPE*)dst.data() EIGEN_PP_EXPAND(EIGEN_VMLMODE_EXPAND_##VMLMODE) );                                         \
-      } else {                                                                                                                \
-        const Index outerSize = dst.outerSize();                                                                              \
-        for(Index outer = 0; outer < outerSize; ++outer) {                                                                    \
-          const EIGENTYPE *src_ptr = src.IsRowMajor ? &(src.lhs().coeffRef(outer,0)) :                                        \
-                                                      &(src.lhs().coeffRef(0, outer));                                        \
-          EIGENTYPE *dst_ptr = dst.IsRowMajor ? &(dst.coeffRef(outer,0)) : &(dst.coeffRef(0, outer));                         \
-          VMLOP( dst.innerSize(), (const VMLTYPE*)src_ptr, exponent,                                                          \
-                 (VMLTYPE*)dst_ptr EIGEN_PP_EXPAND(EIGEN_VMLMODE_EXPAND_##VMLMODE));                                          \
-        }                                                                                                                     \
-      }                                                                                                                       \
-    }                                                                                                                         \
+#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

@@ -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;
 
@@ -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,
@@ -201,10 +201,10 @@ 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)
     {
@@ -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)
     {
@@ -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

@@ -13,70 +13,14 @@
 
 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;
-
 /** \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 || (InnerPanel && (traits<XprType>::Flags&LinearAccessBit))) ? 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>
 {
@@ -109,12 +108,9 @@ 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
     inline Block(XprType& xpr, Index i) : Impl(xpr,i)
     {
       eigen_assert( (i>=0) && (
@@ -124,27 +120,25 @@ template<typename XprType, int BlockRows, int BlockCols, bool InnerPanel> class
 
     /** Fixed-size constructor
       */
-    EIGEN_DEVICE_FUNC
-    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
     inline Block(XprType& xpr,
-          Index startRow, Index startCol,
+          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);
     }
 };
          
@@ -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 inline BlockImpl(XprType& xpr, Index i) : Impl(xpr,i) {}
-    EIGEN_DEVICE_FUNC inline BlockImpl(XprType& xpr, Index startRow, Index startCol) : Impl(xpr, startRow, startCol) {}
-    EIGEN_DEVICE_FUNC
-    inline BlockImpl(XprType& xpr, Index startRow, Index startCol, Index blockRows, Index blockCols)
-      : Impl(xpr, startRow, startCol, blockRows, blockCols) {}
+    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
-    const typename internal::remove_all<XprTypeNested>::type& nestedExpression() const
+    const typename internal::remove_all<typename XprType::Nested>::type& nestedExpression() const 
     { 
       return m_xpr; 
     }
-
-    EIGEN_DEVICE_FUNC
-    XprType& nestedExpression() { return m_xpr; }
       
-    EIGEN_DEVICE_FUNC
-    StorageIndex startRow() const
+    Index startRow() const 
     { 
       return m_startRow.value(); 
     }
       
-    EIGEN_DEVICE_FUNC
-    StorageIndex startCol() const
+    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;
@@ -342,52 +322,42 @@ class BlockImpl_dense<XprType,BlockRows,BlockCols, InnerPanel,true>
 
     /** Column or Row constructor
       */
-    EIGEN_DEVICE_FUNC
     inline BlockImpl_dense(XprType& xpr, Index i)
-      : Base(xpr.data() + i * (    ((BlockRows==1) && (BlockCols==XprType::ColsAtCompileTime) && (!XprTypeIsRowMajor)) 
-                                || ((BlockRows==XprType::RowsAtCompileTime) && (BlockCols==1) && ( XprTypeIsRowMajor)) ? xpr.innerStride() : xpr.outerStride()),
+      : 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
     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)
+      : Base(internal::const_cast_ptr(&xpr.coeffRef(startRow,startCol))), m_xpr(xpr)
     {
       init();
     }
 
     /** Dynamic-size constructor
       */
-    EIGEN_DEVICE_FUNC
     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
-    const typename internal::remove_all<XprTypeNested>::type& nestedExpression() const
+    const typename internal::remove_all<typename XprType::Nested>::type& nestedExpression() const 
     { 
       return m_xpr; 
     }
-
-    EIGEN_DEVICE_FUNC
-    XprType& nestedExpression() { return m_xpr; }
       
     /** \sa MapBase::innerStride() */
-    EIGEN_DEVICE_FUNC
     inline Index innerStride() const
     {
       return internal::traits<BlockType>::HasSameStorageOrderAsXprType
@@ -396,24 +366,11 @@ class BlockImpl_dense<XprType,BlockRows,BlockCols, InnerPanel,true>
     }
 
     /** \sa MapBase::outerStride() */
-    EIGEN_DEVICE_FUNC
     inline Index outerStride() const
     {
       return m_outerStride;
     }
 
-    EIGEN_DEVICE_FUNC
-    StorageIndex startRow() const
-    {
-      return m_startRow.value();
-    }
-
-    EIGEN_DEVICE_FUNC
-    StorageIndex startCol() const
-    {
-      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?
@@ -422,7 +379,6 @@ class BlockImpl_dense<XprType,BlockRows,BlockCols, InnerPanel,true>
 
     #ifndef EIGEN_PARSED_BY_DOXYGEN
     /** \internal used by allowAligned() */
-    EIGEN_DEVICE_FUNC
     inline BlockImpl_dense(XprType& xpr, const Scalar* data, Index blockRows, Index blockCols)
       : Base(data, blockRows, blockCols), m_xpr(xpr)
     {
@@ -431,7 +387,6 @@ class BlockImpl_dense<XprType,BlockRows,BlockCols, InnerPanel,true>
     #endif
 
   protected:
-    EIGEN_DEVICE_FUNC
     void init()
     {
       m_outerStride = internal::traits<BlockType>::HasSameStorageOrderAsXprType
@@ -439,9 +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

@@ -17,10 +17,9 @@ namespace internal {
 template<typename Derived, int UnrollCount>
 struct all_unroller
 {
-  typedef typename Derived::ExpressionTraits Traits;
   enum {
-    col = (UnrollCount-1) / Traits::RowsAtCompileTime,
-    row = (UnrollCount-1) % Traits::RowsAtCompileTime
+    col = (UnrollCount-1) / Derived::RowsAtCompileTime,
+    row = (UnrollCount-1) % Derived::RowsAtCompileTime
   };
 
   static inline bool run(const Derived &mat)
@@ -44,12 +43,11 @@ struct all_unroller<Derived, Dynamic>
 template<typename Derived, int UnrollCount>
 struct any_unroller
 {
-  typedef typename Derived::ExpressionTraits Traits;
   enum {
-    col = (UnrollCount-1) / Traits::RowsAtCompileTime,
-    row = (UnrollCount-1) % Traits::RowsAtCompileTime
+    col = (UnrollCount-1) / Derived::RowsAtCompileTime,
+    row = (UnrollCount-1) % Derived::RowsAtCompileTime
   };
-  
+
   static inline bool run(const Derived &mat)
   {
     return any_unroller<Derived, UnrollCount-1>::run(mat) || mat.coeff(row, col);
@@ -80,19 +78,19 @@ struct any_unroller<Derived, Dynamic>
 template<typename Derived>
 inline bool DenseBase<Derived>::all() const
 {
-  typedef internal::evaluator<Derived> Evaluator;
   enum {
     unroll = SizeAtCompileTime != Dynamic
-          && SizeAtCompileTime * (Evaluator::CoeffReadCost + NumTraits<Scalar>::AddCost) <= EIGEN_UNROLLING_LIMIT
+          && 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>::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;
   }
 }
@@ -104,19 +102,19 @@ inline bool DenseBase<Derived>::all() const
 template<typename Derived>
 inline bool DenseBase<Derived>::any() const
 {
-  typedef internal::evaluator<Derived> Evaluator;
   enum {
     unroll = SizeAtCompileTime != Dynamic
-          && 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>::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;
   }
 }
@@ -126,7 +124,7 @@ inline bool DenseBase<Derived>::any() const
   * \sa all(), any()
   */
 template<typename Derived>
-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();
 }
@@ -138,11 +136,7 @@ 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.
@@ -152,11 +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

@@ -22,14 +22,14 @@ 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)
   {
@@ -37,7 +37,6 @@ struct CommaInitializer
   }
 
   template<typename OtherDerived>
-  EIGEN_DEVICE_FUNC
   inline CommaInitializer(XprType& xpr, const DenseBase<OtherDerived>& other)
     : m_xpr(xpr), m_row(0), m_col(other.cols()), m_currentBlockRows(other.rows())
   {
@@ -47,7 +46,6 @@ struct CommaInitializer
   /* 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:
@@ -57,7 +55,6 @@ struct CommaInitializer
   }
 
   /* inserts a scalar value in the target matrix */
-  EIGEN_DEVICE_FUNC
   CommaInitializer& operator,(const Scalar& s)
   {
     if (m_col==m_xpr.cols())
@@ -77,7 +74,6 @@ 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))
@@ -97,11 +93,7 @@ struct CommaInitializer
     return *this;
   }
 
-  EIGEN_DEVICE_FUNC
   inline ~CommaInitializer()
-#if defined VERIFY_RAISES_ASSERT && (!defined EIGEN_NO_ASSERTION_CHECKING) && defined EIGEN_EXCEPTIONS
-  EIGEN_EXCEPTION_SPEC(Eigen::eigen_assert_exception)
-#endif
   {
       finished();
   }
@@ -113,7 +105,6 @@ 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()
@@ -121,7 +112,7 @@ struct CommaInitializer
       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

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 RealScalar(1);
-  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
@@ -15,113 +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; }
-  /// \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);
+  protected:
+    typedef typename Derived::Scalar Scalar;
+    typedef typename Derived::Index Index;
+
+    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> >
@@ -32,75 +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 = _LhsNested::Flags & RowMajorBit
+    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),
+    Cost0 = EIGEN_ADD_COST(LhsCoeffReadCost,RhsCoeffReadCost),
+    CoeffReadCost = EIGEN_ADD_COST(Cost0,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;
 
-    EIGEN_DEVICE_FUNC
     EIGEN_STRONG_INLINE CwiseBinaryOp(const Lhs& aLhs, const Rhs& aRhs, const BinaryOp& func = BinaryOp())
       : m_lhs(aLhs), m_rhs(aRhs), m_functor(func)
     {
@@ -110,7 +132,6 @@ class CwiseBinaryOp :
       eigen_assert(aLhs.rows() == aRhs.rows() && aLhs.cols() == aRhs.cols());
     }
 
-    EIGEN_DEVICE_FUNC
     EIGEN_STRONG_INLINE Index rows() const {
       // return the fixed size type if available to enable compile time optimizations
       if (internal::traits<typename internal::remove_all<LhsNested>::type>::RowsAtCompileTime==Dynamic)
@@ -118,7 +139,6 @@ class CwiseBinaryOp :
       else
         return m_lhs.rows();
     }
-    EIGEN_DEVICE_FUNC
     EIGEN_STRONG_INLINE Index cols() const {
       // return the fixed size type if available to enable compile time optimizations
       if (internal::traits<typename internal::remove_all<LhsNested>::type>::ColsAtCompileTime==Dynamic)
@@ -128,13 +148,10 @@ class CwiseBinaryOp :
     }
 
     /** \returns the left hand side nested expression */
-    EIGEN_DEVICE_FUNC
     const _LhsNested& lhs() const { return m_lhs; }
     /** \returns the right hand side nested expression */
-    EIGEN_DEVICE_FUNC
     const _RhsNested& rhs() const { return m_rhs; }
     /** \returns the functor representing the binary operation */
-    EIGEN_DEVICE_FUNC
     const BinaryOp& functor() const { return m_functor; }
 
   protected:
@@ -143,13 +160,41 @@ 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.
@@ -161,7 +206,8 @@ template<typename OtherDerived>
 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();
 }
 
@@ -174,11 +220,11 @@ template<typename OtherDerived>
 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();
 }
 
 } // end namespace Eigen
 
 #endif // EIGEN_CWISE_BINARY_OP_H
-

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 Index rows() const { return m_rows.value(); }
-    EIGEN_DEVICE_FUNC
     EIGEN_STRONG_INLINE Index cols() const { return m_cols.value(); }
 
+    EIGEN_STRONG_INLINE const Scalar coeff(Index 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:
@@ -105,10 +113,10 @@ class CwiseNullaryOp : public internal::dense_xpr_base< CwiseNullaryOp<NullaryOp
   */
 template<typename Derived>
 template<typename CustomNullaryOp>
-EIGEN_STRONG_INLINE const CwiseNullaryOp<CustomNullaryOp, typename DenseBase<Derived>::PlainObject>
+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
@@ -124,19 +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_STRONG_INLINE const CwiseNullaryOp<CustomNullaryOp, typename DenseBase<Derived>::PlainObject>
+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
@@ -150,19 +155,19 @@ DenseBase<Derived>::NullaryExpr(Index size, const CustomNullaryOp& func)
   */
 template<typename Derived>
 template<typename CustomNullaryOp>
-EIGEN_STRONG_INLINE const CwiseNullaryOp<CustomNullaryOp, typename DenseBase<Derived>::PlainObject>
+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.
@@ -171,9 +176,9 @@ DenseBase<Derived>::NullaryExpr(const CustomNullaryOp& func)
   */
 template<typename Derived>
 EIGEN_STRONG_INLINE const typename DenseBase<Derived>::ConstantReturnType
-DenseBase<Derived>::Constant(Index rows, Index cols, const Scalar& value)
+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
@@ -215,33 +220,46 @@ DenseBase<Derived>::Constant(const Scalar& value)
   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.
   *
-  * \sa LinSpaced(Index,Scalar,Scalar), setLinSpaced(Index,const Scalar&,const Scalar&)
+  * 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.cpp
+  * Output: \verbinclude DenseBase_LinSpaced_seq.out
+  *
+  * \sa setLinSpaced(Index,const Scalar&,const Scalar&), LinSpaced(Index,Scalar,Scalar), CwiseNullaryOp
   */
 template<typename Derived>
-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,PacketScalar>(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(Scalar,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_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,PacketScalar>(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.
@@ -251,24 +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_STRONG_INLINE const typename DenseBase<Derived>::RandomAccessLinSpacedReturnType
 DenseBase<Derived>::LinSpaced(Index size, const Scalar& low, const Scalar& high)
 {
   EIGEN_STATIC_ASSERT_VECTOR_ONLY(Derived)
-  return DenseBase<Derived>::NullaryExpr(size, internal::linspaced_op<Scalar,PacketScalar>(low,high,size));
+  return DenseBase<Derived>::NullaryExpr(size, internal::linspaced_op<Scalar,true>(low,high,size));
 }
 
 /**
@@ -281,7 +289,7 @@ DenseBase<Derived>::LinSpaced(const Scalar& low, const Scalar& high)
 {
   EIGEN_STATIC_ASSERT_VECTOR_ONLY(Derived)
   EIGEN_STATIC_ASSERT_FIXED_SIZE(Derived)
-  return DenseBase<Derived>::NullaryExpr(Derived::SizeAtCompileTime, internal::linspaced_op<Scalar,PacketScalar>(low,high,Derived::SizeAtCompileTime));
+  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 */
@@ -289,10 +297,9 @@ template<typename Derived>
 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;
 }
@@ -317,7 +324,7 @@ 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()
   */
@@ -327,7 +334,7 @@ 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
   *
@@ -344,10 +351,10 @@ PlainObjectBase<Derived>::setConstant(Index size, const Scalar& val)
   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
@@ -357,14 +364,14 @@ PlainObjectBase<Derived>::setConstant(Index size, const Scalar& val)
   */
 template<typename Derived>
 EIGEN_STRONG_INLINE Derived&
-PlainObjectBase<Derived>::setConstant(Index rows, Index cols, const Scalar& val)
+PlainObjectBase<Derived>::setConstant(Index nbRows, Index nbCols, const Scalar& val)
 {
-  resize(rows, cols);
+  resize(nbRows, nbCols);
   return setConstant(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.
@@ -374,30 +381,24 @@ PlainObjectBase<Derived>::setConstant(Index rows, Index cols, 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_STRONG_INLINE Derived& DenseBase<Derived>::setLinSpaced(Index newSize, const Scalar& low, const Scalar& high)
 {
   EIGEN_STATIC_ASSERT_VECTOR_ONLY(Derived)
-  return derived() = Derived::NullaryExpr(newSize, internal::linspaced_op<Scalar,PacketScalar>(low,high,newSize));
+  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_STRONG_INLINE Derived& DenseBase<Derived>::setLinSpaced(const Scalar& low, const Scalar& high)
@@ -424,9 +425,9 @@ EIGEN_STRONG_INLINE Derived& DenseBase<Derived>::setLinSpaced(const Scalar& low,
   */
 template<typename Derived>
 EIGEN_STRONG_INLINE const typename DenseBase<Derived>::ConstantReturnType
-DenseBase<Derived>::Zero(Index rows, Index cols)
+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.
@@ -480,10 +481,9 @@ DenseBase<Derived>::Zero()
 template<typename Derived>
 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;
 }
@@ -520,8 +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
@@ -530,9 +530,9 @@ PlainObjectBase<Derived>::setZero(Index newSize)
   */
 template<typename Derived>
 EIGEN_STRONG_INLINE Derived&
-PlainObjectBase<Derived>::setZero(Index rows, Index cols)
+PlainObjectBase<Derived>::setZero(Index nbRows, Index nbCols)
 {
-  resize(rows, cols);
+  resize(nbRows, nbCols);
   return setConstant(Scalar(0));
 }
 
@@ -540,7 +540,7 @@ PlainObjectBase<Derived>::setZero(Index rows, Index cols)
 
 /** \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,
@@ -554,9 +554,9 @@ PlainObjectBase<Derived>::setZero(Index rows, Index cols)
   */
 template<typename Derived>
 EIGEN_STRONG_INLINE const typename DenseBase<Derived>::ConstantReturnType
-DenseBase<Derived>::Ones(Index rows, Index cols)
+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.
@@ -646,8 +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
@@ -656,9 +656,9 @@ PlainObjectBase<Derived>::setOnes(Index newSize)
   */
 template<typename Derived>
 EIGEN_STRONG_INLINE Derived&
-PlainObjectBase<Derived>::setOnes(Index rows, Index cols)
+PlainObjectBase<Derived>::setOnes(Index nbRows, Index nbCols)
 {
-  resize(rows, cols);
+  resize(nbRows, nbCols);
   return setConstant(Scalar(1));
 }
 
@@ -666,7 +666,7 @@ PlainObjectBase<Derived>::setOnes(Index rows, Index cols)
 
 /** \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,
@@ -680,9 +680,9 @@ PlainObjectBase<Derived>::setOnes(Index rows, Index cols)
   */
 template<typename Derived>
 EIGEN_STRONG_INLINE const typename MatrixBase<Derived>::IdentityReturnType
-MatrixBase<Derived>::Identity(Index rows, Index cols)
+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).
@@ -716,19 +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;
       }
     }
@@ -741,7 +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());
@@ -751,11 +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;
   }
@@ -778,8 +776,8 @@ EIGEN_STRONG_INLINE Derived& MatrixBase<Derived>::setIdentity()
 
 /** \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
@@ -787,9 +785,9 @@ EIGEN_STRONG_INLINE Derived& MatrixBase<Derived>::setIdentity()
   * \sa MatrixBase::setIdentity(), class CwiseNullaryOp, MatrixBase::Identity()
   */
 template<typename Derived>
-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();
 }
 

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
@@ -13,32 +13,13 @@
 
 namespace Eigen { 
 
-namespace internal {
-template<typename UnaryOp, typename XprType>
-struct traits<CwiseUnaryOp<UnaryOp, XprType> >
- : traits<XprType>
-{
-  typedef typename result_of<
-                     UnaryOp(const typename XprType::Scalar&)
-                   >::type Scalar;
-  typedef typename XprType::Nested XprTypeNested;
-  typedef typename remove_reference<XprTypeNested>::type _XprTypeNested;
-  enum {
-    Flags = _XprTypeNested::Flags & RowMajorBit 
-  };
-};
-}
-
-template<typename UnaryOp, typename XprType, typename StorageKind>
-class CwiseUnaryOpImpl;
-
 /** \class CwiseUnaryOp
   * \ingroup Core_Module
   *
   * \brief Generic expression 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 = EIGEN_ADD_COST(_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
-    Index rows() const { return m_xpr.rows(); }
-    EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
-    Index cols() const { 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 = EIGEN_ADD_COST(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>
 {
@@ -61,10 +62,8 @@ 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 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)
@@ -76,27 +75,19 @@ class CwiseUnaryView : public CwiseUnaryViewImpl<ViewOp, MatrixType, typename in
     const ViewOp& functor() const { return m_functor; }
 
     /** \returns the nested expression */
-    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 */
-    typename internal::remove_reference<MatrixTypeNested>::type&
+    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
@@ -109,18 +100,38 @@ class CwiseUnaryViewImpl<ViewOp,MatrixType,Dense>
     EIGEN_DENSE_PUBLIC_INTERFACE(Derived)
     EIGEN_INHERIT_ASSIGNMENT_OPERATORS(CwiseUnaryViewImpl)
     
-    EIGEN_DEVICE_FUNC inline Scalar* data() { return &(this->coeffRef(0)); }
-    EIGEN_DEVICE_FUNC inline const Scalar* data() const { return &(this->coeff(0)); }
+    inline Scalar* data() { return &coeffRef(0); }
+    inline const Scalar* data() const { return &coeff(0); }
 
-    EIGEN_DEVICE_FUNC inline Index innerStride() const
+    inline Index innerStride() const
     {
       return derived().nestedExpression().innerStride() * sizeof(typename internal::traits<MatrixType>::Scalar) / sizeof(Scalar);
     }
 
-    EIGEN_DEVICE_FUNC inline Index outerStride() const
+    inline Index outerStride() const
     {
       return derived().nestedExpression().outerStride() * sizeof(typename internal::traits<MatrixType>::Scalar) / sizeof(Scalar);
     }
+
+    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

@@ -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>
+  : public internal::special_scalar_op_base<Derived, typename internal::traits<Derived>::Scalar,
+                                            typename NumTraits<typename internal::traits<Derived>::Scalar>::Real,
+                                            DenseCoeffsBase<Derived> >
 #else
-  : public DenseCoeffsBase<Derived,DirectWriteAccessors>
+  : public DenseCoeffsBase<Derived>
 #endif // not EIGEN_PARSED_BY_DOXYGEN
 {
   public:
 
-    /** Inner iterator type to iterate over the coefficients of a row or column.
-      * \sa class InnerIterator
-      */
-    typedef Eigen::InnerIterator<Derived> InnerIterator;
+    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> Base;
+    typedef internal::special_scalar_op_base<Derived,Scalar,RealScalar, DenseCoeffsBase<Derived> > Base;
 
+    using Base::operator*;
     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;
@@ -167,46 +169,19 @@ 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
     inline Index nonZeros() const { return size(); }
 
     /** \returns the outer size.
@@ -214,7 +189,6 @@ template<typename Derived> class DenseBase
       * \note For a vector, this returns just 1. For a matrix (non-vector), this is the major dimension
       * with respect to the \ref TopicStorageOrders "storage order", i.e., the number of columns for a
       * column-major matrix, and the number of rows for a row-major matrix. */
-    EIGEN_DEVICE_FUNC
     Index outerSize() const
     {
       return IsVectorAtCompileTime ? 1
@@ -226,7 +200,6 @@ template<typename Derived> class DenseBase
       * \note For a vector, this is just the size. For a matrix (non-vector), this is the minor dimension
       * with respect to the \ref TopicStorageOrders "storage order", i.e., the number of rows for a 
       * column-major matrix, and the number of columns for a row-major matrix. */
-    EIGEN_DEVICE_FUNC
     Index innerSize() const
     {
       return IsVectorAtCompileTime ? this->size()
@@ -237,7 +210,6 @@ 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);
@@ -248,22 +220,22 @@ 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. */
-    typedef CwiseNullaryOp<internal::linspaced_op<Scalar,PacketScalar>,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,PacketScalar>,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;
 
@@ -271,133 +243,120 @@ 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);
 
-    /** \ínternal
-      * Copies \a other into *this without evaluating other. \returns a reference to *this.
-      * \deprecated */
+    /** \internal Copies \a other into *this without evaluating other. \returns a reference to *this. */
     template<typename OtherDerived>
-    EIGEN_DEVICE_FUNC
     Derived& lazyAssign(const DenseBase<OtherDerived>& other);
 
-    EIGEN_DEVICE_FUNC
+    /** \internal Evaluates \a other into *this. \returns a reference to *this. */
+    template<typename OtherDerived>
+    Derived& lazyAssign(const ReturnByValue<OtherDerived>& other);
+
     CommaInitializer<Derived> operator<< (const Scalar& s);
 
-    /** \deprecated it now returns \c *this */
     template<unsigned int Added,unsigned int Removed>
-    EIGEN_DEPRECATED
-    const Derived& flagged() const
-    { return derived(); }
+    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_DEVICE_FUNC static const SequentialLinSpacedReturnType
+    static const SequentialLinSpacedReturnType
     LinSpaced(Sequential_t, Index size, 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 SequentialLinSpacedReturnType
+    static const SequentialLinSpacedReturnType
     LinSpaced(Sequential_t, const Scalar& low, const Scalar& high);
-    EIGEN_DEVICE_FUNC static const RandomAccessLinSpacedReturnType
+    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 carefull with eval() and the auto C++ keyword, as detailed in this \link TopicPitfalls_auto_keyword page \endlink.
       */
-    EIGEN_DEVICE_FUNC
     EIGEN_STRONG_INLINE EvalReturnType eval() const
     {
       // Even though MSVC does not honor strong inlining when the return type
@@ -405,68 +364,61 @@ template<typename Derived> class DenseBase
       // size types on MSVC.
       return typename internal::eval<Derived>::type(derived());
     }
-    
+
     /** swaps *this with the expression \a other.
       *
       */
     template<typename OtherDerived>
-    EIGEN_DEVICE_FUNC
-    void swap(const DenseBase<OtherDerived>& other)
+    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
     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;
 
-    EIGEN_DEVICE_FUNC typename internal::traits<Derived>::Scalar minCoeff() const;
-    EIGEN_DEVICE_FUNC typename internal::traits<Derived>::Scalar maxCoeff() const;
+    Scalar prod() const;
 
-    template<typename IndexType> EIGEN_DEVICE_FUNC
+    typename internal::traits<Derived>::Scalar minCoeff() const;
+    typename internal::traits<Derived>::Scalar maxCoeff() const;
+
+    template<typename IndexType>
     typename internal::traits<Derived>::Scalar minCoeff(IndexType* row, IndexType* col) const;
-    template<typename IndexType> EIGEN_DEVICE_FUNC
+    template<typename IndexType>
     typename internal::traits<Derived>::Scalar maxCoeff(IndexType* row, IndexType* col) const;
-    template<typename IndexType> EIGEN_DEVICE_FUNC
+    template<typename IndexType>
     typename internal::traits<Derived>::Scalar minCoeff(IndexType* index) const;
-    template<typename IndexType> EIGEN_DEVICE_FUNC
+    template<typename IndexType>
     typename internal::traits<Derived>::Scalar maxCoeff(IndexType* index) const;
 
     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;
 
     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)
@@ -474,8 +426,8 @@ template<typename Derived> class DenseBase
       return derived().coeff(0,0);
     }
 
-    bool all() const;
-    bool any() const;
+    bool all(void) const;
+    bool any(void) const;
     Index count() const;
 
     typedef VectorwiseOp<Derived, Horizontal> RowwiseReturnType;
@@ -483,35 +435,14 @@ template<typename Derived> class DenseBase
     typedef VectorwiseOp<Derived, Vertical> ColwiseReturnType;
     typedef const VectorwiseOp<const Derived, Vertical> ConstColwiseReturnType;
 
-    /** \returns a VectorwiseOp wrapper of *this providing additional partial reduction operations
-    *
-    * Example: \include MatrixBase_rowwise.cpp
-    * Output: \verbinclude MatrixBase_rowwise.out
-    *
-    * \sa colwise(), class VectorwiseOp, \ref TutorialReductionsVisitorsBroadcasting
-    */
-    //Code moved here due to a CUDA compiler bug
-    EIGEN_DEVICE_FUNC inline ConstRowwiseReturnType rowwise() const {
-      return ConstRowwiseReturnType(derived());
-    }
-    EIGEN_DEVICE_FUNC RowwiseReturnType rowwise();
+    ConstRowwiseReturnType rowwise() const;
+    RowwiseReturnType rowwise();
+    ConstColwiseReturnType colwise() const;
+    ColwiseReturnType colwise();
 
-    /** \returns a VectorwiseOp wrapper of *this providing additional partial reduction operations
-    *
-    * Example: \include MatrixBase_colwise.cpp
-    * Output: \verbinclude MatrixBase_colwise.out
-    *
-    * \sa rowwise(), class VectorwiseOp, \ref TutorialReductionsVisitorsBroadcasting
-    */
-    EIGEN_DEVICE_FUNC inline ConstColwiseReturnType colwise() const {
-      return ConstColwiseReturnType(derived());
-    }
-    EIGEN_DEVICE_FUNC ColwiseReturnType colwise();
-
-    typedef CwiseNullaryOp<internal::scalar_random_op<Scalar>,PlainObject> RandomReturnType;
-    static const RandomReturnType Random(Index rows, Index cols);
-    static const RandomReturnType Random(Index size);
-    static const RandomReturnType Random();
+    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>
     const Select<Derived,ThenDerived,ElseDerived>
@@ -529,56 +460,45 @@ template<typename Derived> class DenseBase
     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);
-    }
+    inline const Replicate<Derived,RowFactor,ColFactor> replicate() const;
+    
+    typedef Replicate<Derived,Dynamic,Dynamic> ReplicateReturnType;
+    inline const ReplicateReturnType 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();
+    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)
 #   include "../plugins/BlockMethods.h"
 #   ifdef EIGEN_DENSEBASE_PLUGIN
 #     include EIGEN_DENSEBASE_PLUGIN
 #   endif
 #undef EIGEN_CURRENT_STORAGE_BASE_CLASS
-#undef EIGEN_DOC_BLOCK_ADDONS_NOT_INNER_PANEL
-#undef EIGEN_DOC_BLOCK_ADDONS_INNER_PANEL_IF
+
+#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:
     /** 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 paranoiac and could slow compilation down
@@ -591,9 +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

@@ -35,6 +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;
 
@@ -60,7 +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
@@ -69,7 +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
@@ -92,15 +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),
@@ -111,12 +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
@@ -134,14 +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);
     }
 
 
@@ -153,14 +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.
@@ -173,49 +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
@@ -230,9 +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);
     }
 
 
@@ -257,11 +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:
@@ -304,6 +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;
@@ -336,15 +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)
     {
@@ -357,13 +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);
     }
 
 
@@ -382,14 +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.
@@ -399,14 +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.
@@ -418,49 +388,167 @@ 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.
@@ -472,7 +560,7 @@ class DenseCoeffsBase<Derived, WriteAccessors> : public DenseCoeffsBase<Derived,
   * 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>
@@ -480,6 +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;
 
@@ -492,7 +581,6 @@ class DenseCoeffsBase<Derived, DirectAccessors> : public DenseCoeffsBase<Derived
       *
       * \sa outerStride(), rowStride(), colStride()
       */
-    EIGEN_DEVICE_FUNC
     inline Index innerStride() const
     {
       return derived().innerStride();
@@ -503,7 +591,6 @@ class DenseCoeffsBase<Derived, DirectAccessors> : public DenseCoeffsBase<Derived
       *
       * \sa innerStride(), rowStride(), colStride()
       */
-    EIGEN_DEVICE_FUNC
     inline Index outerStride() const
     {
       return derived().outerStride();
@@ -519,7 +606,6 @@ class DenseCoeffsBase<Derived, DirectAccessors> : public DenseCoeffsBase<Derived
       *
       * \sa innerStride(), outerStride(), colStride()
       */
-    EIGEN_DEVICE_FUNC
     inline Index rowStride() const
     {
       return Derived::IsRowMajor ? outerStride() : innerStride();
@@ -529,7 +615,6 @@ class DenseCoeffsBase<Derived, DirectAccessors> : public DenseCoeffsBase<Derived
       *
       * \sa innerStride(), outerStride(), rowStride()
       */
-    EIGEN_DEVICE_FUNC
     inline Index colStride() const
     {
       return Derived::IsRowMajor ? innerStride() : outerStride();
@@ -545,7 +630,7 @@ class DenseCoeffsBase<Derived, DirectAccessors> : public DenseCoeffsBase<Derived
   * 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>
@@ -554,6 +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;
 
@@ -566,7 +652,6 @@ class DenseCoeffsBase<Derived, DirectWriteAccessors>
       *
       * \sa outerStride(), rowStride(), colStride()
       */
-    EIGEN_DEVICE_FUNC
     inline Index innerStride() const
     {
       return derived().innerStride();
@@ -577,7 +662,6 @@ class DenseCoeffsBase<Derived, DirectWriteAccessors>
       *
       * \sa innerStride(), rowStride(), colStride()
       */
-    EIGEN_DEVICE_FUNC
     inline Index outerStride() const
     {
       return derived().outerStride();
@@ -593,7 +677,6 @@ class DenseCoeffsBase<Derived, DirectWriteAccessors>
       *
       * \sa innerStride(), outerStride(), colStride()
       */
-    EIGEN_DEVICE_FUNC
     inline Index rowStride() const
     {
       return Derived::IsRowMajor ? outerStride() : innerStride();
@@ -603,7 +686,6 @@ class DenseCoeffsBase<Derived, DirectWriteAccessors>
       *
       * \sa innerStride(), outerStride(), rowStride()
       */
-    EIGEN_DEVICE_FUNC
     inline Index colStride() const
     {
       return Derived::IsRowMajor ? innerStride() : outerStride();
@@ -612,42 +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 inline Index run(const Derived&)
+  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
@@ -24,9 +24,7 @@ namespace internal {
 
 struct constructor_without_unaligned_array_assert {};
 
-template<typename T, int Size>
-EIGEN_DEVICE_FUNC
-void check_static_allocation_size()
+template<typename T, int Size> void check_static_allocation_size()
 {
   // if EIGEN_STACK_ALLOCATION_LIMIT is defined to 0, then no limit
   #if EIGEN_STACK_ALLOCATION_LIMIT
@@ -40,19 +38,18 @@ void check_static_allocation_size()
   */
 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()
+  plain_array() 
   { 
     check_static_allocation_size<T,Size>();
   }
 
-  EIGEN_DEVICE_FUNC
-  plain_array(constructor_without_unaligned_array_assert)
+  plain_array(constructor_without_unaligned_array_assert) 
   { 
     check_static_allocation_size<T,Size>();
   }
@@ -67,88 +64,29 @@ struct plain_array
   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);
+    EIGEN_MAKE_UNALIGNED_ARRAY_ASSERT(0xf);
     check_static_allocation_size<T,Size>();
   }
 
-  EIGEN_DEVICE_FUNC
-  plain_array(constructor_without_unaligned_array_assert) 
-  { 
-    check_static_allocation_size<T,Size>();
-  }
-};
-
-template <typename T, int Size, int MatrixOrArrayOptions>
-struct plain_array<T, Size, MatrixOrArrayOptions, 32>
-{
-  EIGEN_ALIGN_TO_BOUNDARY(32) T array[Size];
-
-  EIGEN_DEVICE_FUNC
-  plain_array() 
-  {
-    EIGEN_MAKE_UNALIGNED_ARRAY_ASSERT(31);
-    check_static_allocation_size<T,Size>();
-  }
-
-  EIGEN_DEVICE_FUNC
-  plain_array(constructor_without_unaligned_array_assert) 
-  { 
-    check_static_allocation_size<T,Size>();
-  }
-};
-
-template <typename T, int Size, int MatrixOrArrayOptions>
-struct plain_array<T, Size, MatrixOrArrayOptions, 64>
-{
-  EIGEN_ALIGN_TO_BOUNDARY(64) T array[Size];
-
-  EIGEN_DEVICE_FUNC
-  plain_array() 
-  { 
-    EIGEN_MAKE_UNALIGNED_ARRAY_ASSERT(63);
-    check_static_allocation_size<T,Size>();
-  }
-
-  EIGEN_DEVICE_FUNC
   plain_array(constructor_without_unaligned_array_assert) 
   { 
     check_static_allocation_size<T,Size>();
@@ -158,9 +96,9 @@ struct plain_array<T, Size, MatrixOrArrayOptions, 64>
 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
@@ -184,50 +122,41 @@ 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_DEVICE_FUNC
-    explicit DenseStorage(internal::constructor_without_unaligned_array_assert)
+    DenseStorage() {}
+    DenseStorage(internal::constructor_without_unaligned_array_assert)
       : m_data(internal::constructor_without_unaligned_array_assert()) {}
-    EIGEN_DEVICE_FUNC 
     DenseStorage(const DenseStorage& other) : m_data(other.m_data) {}
-    EIGEN_DEVICE_FUNC 
     DenseStorage& operator=(const DenseStorage& other)
-    { 
+    {
       if (this != &other) m_data = other.m_data;
-      return *this; 
+      return *this;
     }
-    EIGEN_DEVICE_FUNC DenseStorage(Index size, Index rows, Index cols) {
-      EIGEN_INTERNAL_DENSE_STORAGE_CTOR_PLUGIN
-      eigen_internal_assert(size==rows*cols && rows==_Rows && cols==_Cols);
-      EIGEN_UNUSED_VARIABLE(size);
-      EIGEN_UNUSED_VARIABLE(rows);
-      EIGEN_UNUSED_VARIABLE(cols);
-    }
-    EIGEN_DEVICE_FUNC void swap(DenseStorage& other) { std::swap(m_data,other.m_data); }
-    EIGEN_DEVICE_FUNC static Index rows(void) {return _Rows;}
-    EIGEN_DEVICE_FUNC static Index cols(void) {return _Cols;}
-    EIGEN_DEVICE_FUNC void conservativeResize(Index,Index,Index) {}
-    EIGEN_DEVICE_FUNC void resize(Index,Index,Index) {}
-    EIGEN_DEVICE_FUNC const T *data() const { return m_data.array; }
-    EIGEN_DEVICE_FUNC T *data() { return m_data.array; }
+    DenseStorage(DenseIndex,DenseIndex,DenseIndex) {}
+    void swap(DenseStorage& other) { std::swap(m_data,other.m_data); }
+    static DenseIndex rows(void) {return _Rows;}
+    static DenseIndex cols(void) {return _Cols;}
+    void conservativeResize(DenseIndex,DenseIndex,DenseIndex) {}
+    void resize(DenseIndex,DenseIndex,DenseIndex) {}
+    const T *data() const { return m_data.array; }
+    T *data() { return m_data.array; }
 };
 
 // null matrix
 template<typename T, int _Rows, int _Cols, int _Options> class DenseStorage<T, 0, _Rows, _Cols, _Options>
 {
   public:
-    EIGEN_DEVICE_FUNC DenseStorage() {}
-    EIGEN_DEVICE_FUNC explicit DenseStorage(internal::constructor_without_unaligned_array_assert) {}
-    EIGEN_DEVICE_FUNC DenseStorage(const DenseStorage&) {}
-    EIGEN_DEVICE_FUNC DenseStorage& operator=(const DenseStorage&) { return *this; }
-    EIGEN_DEVICE_FUNC DenseStorage(Index,Index,Index) {}
-    EIGEN_DEVICE_FUNC void swap(DenseStorage& ) {}
-    EIGEN_DEVICE_FUNC static Index rows(void) {return _Rows;}
-    EIGEN_DEVICE_FUNC static Index cols(void) {return _Cols;}
-    EIGEN_DEVICE_FUNC void conservativeResize(Index,Index,Index) {}
-    EIGEN_DEVICE_FUNC void resize(Index,Index,Index) {}
-    EIGEN_DEVICE_FUNC const T *data() const { return 0; }
-    EIGEN_DEVICE_FUNC T *data() { return 0; }
+    DenseStorage() {}
+    DenseStorage(internal::constructor_without_unaligned_array_assert) {}
+    DenseStorage(const DenseStorage&) {}
+    DenseStorage& operator=(const DenseStorage&) { return *this; }
+    DenseStorage(DenseIndex,DenseIndex,DenseIndex) {}
+    void swap(DenseStorage& ) {}
+    static DenseIndex rows(void) {return _Rows;}
+    static DenseIndex cols(void) {return _Cols;}
+    void conservativeResize(DenseIndex,DenseIndex,DenseIndex) {}
+    void resize(DenseIndex,DenseIndex,DenseIndex) {}
+    const T *data() const { return 0; }
+    T *data() { return 0; }
 };
 
 // more specializations for null matrices; these are necessary to resolve ambiguities
@@ -244,74 +173,74 @@ 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)
+    DenseStorage() : m_rows(0), m_cols(0) {}
+    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) 
-    { 
+    DenseStorage(const DenseStorage& other) : m_data(other.m_data), m_rows(other.m_rows), m_cols(other.m_cols) {}
+    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; 
+      return *this;
     }
-    EIGEN_DEVICE_FUNC DenseStorage(Index, Index rows, Index cols) : m_rows(rows), m_cols(cols) {}
-    EIGEN_DEVICE_FUNC void swap(DenseStorage& other)
+    DenseStorage(DenseIndex, DenseIndex nbRows, DenseIndex nbCols) : m_rows(nbRows), m_cols(nbCols) {}
+    void swap(DenseStorage& other)
     { std::swap(m_data,other.m_data); std::swap(m_rows,other.m_rows); std::swap(m_cols,other.m_cols); }
-    EIGEN_DEVICE_FUNC Index rows() const {return m_rows;}
-    EIGEN_DEVICE_FUNC Index cols() const {return m_cols;}
-    EIGEN_DEVICE_FUNC void conservativeResize(Index, Index rows, Index cols) { m_rows = rows; m_cols = cols; }
-    EIGEN_DEVICE_FUNC void resize(Index, Index rows, Index cols) { m_rows = rows; m_cols = cols; }
-    EIGEN_DEVICE_FUNC const T *data() const { return m_data.array; }
-    EIGEN_DEVICE_FUNC T *data() { return m_data.array; }
+    DenseIndex rows() const {return m_rows;}
+    DenseIndex cols() const {return m_cols;}
+    void conservativeResize(DenseIndex, DenseIndex nbRows, DenseIndex nbCols) { m_rows = nbRows; m_cols = nbCols; }
+    void resize(DenseIndex, DenseIndex nbRows, DenseIndex nbCols) { m_rows = nbRows; m_cols = nbCols; }
+    const T *data() const { return m_data.array; }
+    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)
+    DenseStorage() : m_rows(0) {}
+    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) 
+    DenseStorage(const DenseStorage& other) : m_data(other.m_data), m_rows(other.m_rows) {}
+    DenseStorage& operator=(const DenseStorage& other)
     {
       if (this != &other)
       {
         m_data = other.m_data;
         m_rows = other.m_rows;
       }
-      return *this; 
+      return *this;
     }
-    EIGEN_DEVICE_FUNC DenseStorage(Index, Index rows, Index) : m_rows(rows) {}
-    EIGEN_DEVICE_FUNC void swap(DenseStorage& other) { std::swap(m_data,other.m_data); std::swap(m_rows,other.m_rows); }
-    EIGEN_DEVICE_FUNC Index rows(void) const {return m_rows;}
-    EIGEN_DEVICE_FUNC Index cols(void) const {return _Cols;}
-    EIGEN_DEVICE_FUNC void conservativeResize(Index, Index rows, Index) { m_rows = rows; }
-    EIGEN_DEVICE_FUNC void resize(Index, Index rows, Index) { m_rows = rows; }
-    EIGEN_DEVICE_FUNC const T *data() const { return m_data.array; }
-    EIGEN_DEVICE_FUNC T *data() { return m_data.array; }
+    DenseStorage(DenseIndex, DenseIndex nbRows, DenseIndex) : m_rows(nbRows) {}
+    void swap(DenseStorage& other) { std::swap(m_data,other.m_data); std::swap(m_rows,other.m_rows); }
+    DenseIndex rows(void) const {return m_rows;}
+    DenseIndex cols(void) const {return _Cols;}
+    void conservativeResize(DenseIndex, DenseIndex nbRows, DenseIndex) { m_rows = nbRows; }
+    void resize(DenseIndex, DenseIndex nbRows, DenseIndex) { m_rows = nbRows; }
+    const T *data() const { return m_data.array; }
+    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)
+    DenseStorage() : m_cols(0) {}
+    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)
+    DenseStorage(const DenseStorage& other) : m_data(other.m_data), m_cols(other.m_cols) {}
+    DenseStorage& operator=(const DenseStorage& other)
     {
       if (this != &other)
       {
@@ -320,61 +249,38 @@ template<typename T, int Size, int _Rows, int _Options> class DenseStorage<T, Si
       }
       return *this;
     }
-    EIGEN_DEVICE_FUNC DenseStorage(Index, Index, Index cols) : m_cols(cols) {}
-    EIGEN_DEVICE_FUNC void swap(DenseStorage& other) { std::swap(m_data,other.m_data); std::swap(m_cols,other.m_cols); }
-    EIGEN_DEVICE_FUNC Index rows(void) const {return _Rows;}
-    EIGEN_DEVICE_FUNC Index cols(void) const {return m_cols;}
-    void conservativeResize(Index, Index, Index cols) { m_cols = cols; }
-    void resize(Index, Index, Index cols) { m_cols = cols; }
-    EIGEN_DEVICE_FUNC const T *data() const { return m_data.array; }
-    EIGEN_DEVICE_FUNC T *data() { return m_data.array; }
+    DenseStorage(DenseIndex, DenseIndex, DenseIndex nbCols) : m_cols(nbCols) {}
+    void swap(DenseStorage& other) { std::swap(m_data,other.m_data); std::swap(m_cols,other.m_cols); }
+    DenseIndex rows(void) const {return _Rows;}
+    DenseIndex cols(void) const {return m_cols;}
+    void conservativeResize(DenseIndex, DenseIndex, DenseIndex nbCols) { m_cols = nbCols; }
+    void resize(DenseIndex, DenseIndex, DenseIndex nbCols) { m_cols = nbCols; }
+    const T *data() const { return m_data.array; }
+    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)
+    DenseStorage() : m_data(0), m_rows(0), m_cols(0) {}
+    DenseStorage(internal::constructor_without_unaligned_array_assert)
        : m_data(0), m_rows(0), m_cols(0) {}
-    EIGEN_DEVICE_FUNC DenseStorage(Index size, Index rows, Index cols)
-      : m_data(internal::conditional_aligned_new_auto<T,(_Options&DontAlign)==0>(size)), m_rows(rows), m_cols(cols)
-    {
-      EIGEN_INTERNAL_DENSE_STORAGE_CTOR_PLUGIN
-      eigen_internal_assert(size==rows*cols && rows>=0 && cols >=0);
-    }
-    EIGEN_DEVICE_FUNC DenseStorage(const DenseStorage& other)
-      : m_data(internal::conditional_aligned_new_auto<T,(_Options&DontAlign)==0>(other.m_rows*other.m_cols))
-      , m_rows(other.m_rows)
-      , m_cols(other.m_cols)
-    {
-      internal::smart_copy(other.m_data, other.m_data+other.m_rows*other.m_cols, m_data);
-    }
-    EIGEN_DEVICE_FUNC DenseStorage& operator=(const DenseStorage& other)
-    {
-      if (this != &other)
-      {
-        DenseStorage tmp(other);
-        this->swap(tmp);
-      }
-      return *this;
-    }
-#if EIGEN_HAS_RVALUE_REFERENCES
-    EIGEN_DEVICE_FUNC
-    DenseStorage(DenseStorage&& other) EIGEN_NOEXCEPT
+    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 }
+#ifdef EIGEN_HAVE_RVALUE_REFERENCES
+    DenseStorage(DenseStorage&& other)
       : m_data(std::move(other.m_data))
       , m_rows(std::move(other.m_rows))
       , m_cols(std::move(other.m_cols))
     {
       other.m_data = nullptr;
-      other.m_rows = 0;
-      other.m_cols = 0;
     }
-    EIGEN_DEVICE_FUNC
-    DenseStorage& operator=(DenseStorage&& other) EIGEN_NOEXCEPT
+    DenseStorage& operator=(DenseStorage&& other)
     {
       using std::swap;
       swap(m_data, other.m_data);
@@ -383,18 +289,18 @@ template<typename T, int _Options> class DenseStorage<T, Dynamic, Dynamic, Dynam
       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)
+    ~DenseStorage() { internal::conditional_aligned_delete_auto<T,(_Options&DontAlign)==0>(m_data, m_rows*m_cols); }
+    void swap(DenseStorage& other)
     { std::swap(m_data,other.m_data); std::swap(m_rows,other.m_rows); std::swap(m_cols,other.m_cols); }
-    EIGEN_DEVICE_FUNC Index rows(void) const {return m_rows;}
-    EIGEN_DEVICE_FUNC Index cols(void) const {return m_cols;}
-    void conservativeResize(Index size, Index rows, Index cols)
+    DenseIndex rows(void) const {return m_rows;}
+    DenseIndex cols(void) const {return m_cols;}
+    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)
       {
@@ -405,53 +311,34 @@ template<typename T, int _Options> class DenseStorage<T, Dynamic, Dynamic, Dynam
           m_data = 0;
         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; }
+    const T *data() const { return m_data; }
+    T *data() { return m_data; }
+  private:
+    DenseStorage(const DenseStorage&);
+    DenseStorage& operator=(const DenseStorage&);
 };
 
 // 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)
-    {
-      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
+    DenseStorage() : m_data(0), m_cols(0) {}
+    DenseStorage(internal::constructor_without_unaligned_array_assert) : m_data(0), m_cols(0) {}
+    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 }
+#ifdef EIGEN_HAVE_RVALUE_REFERENCES
+    DenseStorage(DenseStorage&& other)
       : m_data(std::move(other.m_data))
       , m_cols(std::move(other.m_cols))
     {
       other.m_data = nullptr;
-      other.m_cols = 0;
     }
-    EIGEN_DEVICE_FUNC
-    DenseStorage& operator=(DenseStorage&& other) EIGEN_NOEXCEPT
+    DenseStorage& operator=(DenseStorage&& other)
     {
       using std::swap;
       swap(m_data, other.m_data);
@@ -459,16 +346,16 @@ template<typename T, int _Rows, int _Options> class DenseStorage<T, Dynamic, _Ro
       return *this;
     }
 #endif
-    EIGEN_DEVICE_FUNC ~DenseStorage() { internal::conditional_aligned_delete_auto<T,(_Options&DontAlign)==0>(m_data, _Rows*m_cols); }
-    EIGEN_DEVICE_FUNC void swap(DenseStorage& other) { std::swap(m_data,other.m_data); std::swap(m_cols,other.m_cols); }
-    EIGEN_DEVICE_FUNC static Index rows(void) {return _Rows;}
-    EIGEN_DEVICE_FUNC Index cols(void) const {return m_cols;}
-    EIGEN_DEVICE_FUNC void conservativeResize(Index size, Index, Index cols)
+    ~DenseStorage() { internal::conditional_aligned_delete_auto<T,(_Options&DontAlign)==0>(m_data, _Rows*m_cols); }
+    void swap(DenseStorage& other) { std::swap(m_data,other.m_data); std::swap(m_cols,other.m_cols); }
+    static DenseIndex rows(void) {return _Rows;}
+    DenseIndex cols(void) const {return m_cols;}
+    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)
       {
@@ -479,52 +366,33 @@ template<typename T, int _Rows, int _Options> class DenseStorage<T, Dynamic, _Ro
           m_data = 0;
         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; }
+    const T *data() const { return m_data; }
+    T *data() { return m_data; }
+  private:
+    DenseStorage(const DenseStorage&);
+    DenseStorage& operator=(const DenseStorage&);
 };
 
 // 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)
-    {
-      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
+    DenseStorage() : m_data(0), m_rows(0) {}
+    DenseStorage(internal::constructor_without_unaligned_array_assert) : m_data(0), m_rows(0) {}
+    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 }
+#ifdef EIGEN_HAVE_RVALUE_REFERENCES
+    DenseStorage(DenseStorage&& other)
       : m_data(std::move(other.m_data))
       , m_rows(std::move(other.m_rows))
     {
       other.m_data = nullptr;
-      other.m_rows = 0;
     }
-    EIGEN_DEVICE_FUNC
-    DenseStorage& operator=(DenseStorage&& other) EIGEN_NOEXCEPT
+    DenseStorage& operator=(DenseStorage&& other)
     {
       using std::swap;
       swap(m_data, other.m_data);
@@ -532,16 +400,16 @@ template<typename T, int _Cols, int _Options> class DenseStorage<T, Dynamic, Dyn
       return *this;
     }
 #endif
-    EIGEN_DEVICE_FUNC ~DenseStorage() { internal::conditional_aligned_delete_auto<T,(_Options&DontAlign)==0>(m_data, _Cols*m_rows); }
-    EIGEN_DEVICE_FUNC void swap(DenseStorage& other) { std::swap(m_data,other.m_data); std::swap(m_rows,other.m_rows); }
-    EIGEN_DEVICE_FUNC Index rows(void) const {return m_rows;}
-    EIGEN_DEVICE_FUNC static Index cols(void) {return _Cols;}
-    void conservativeResize(Index size, Index rows, Index)
+    ~DenseStorage() { internal::conditional_aligned_delete_auto<T,(_Options&DontAlign)==0>(m_data, _Cols*m_rows); }
+    void swap(DenseStorage& other) { std::swap(m_data,other.m_data); std::swap(m_rows,other.m_rows); }
+    DenseIndex rows(void) const {return m_rows;}
+    static DenseIndex cols(void) {return _Cols;}
+    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)
       {
@@ -552,10 +420,13 @@ template<typename T, int _Cols, int _Options> class DenseStorage<T, Dynamic, Dyn
           m_data = 0;
         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; }
+    const T *data() const { return m_data; }
+    T *data() { return m_data; }
+  private:
+    DenseStorage(const DenseStorage&);
+    DenseStorage& operator=(const DenseStorage&);
 };
 
 } // end namespace Eigen

Eigen/src/Core/Diagonal.h

@@ -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
@@ -69,28 +70,20 @@ 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) {}
+    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 ? numext::mini<Index>(m_matrix.cols(),m_matrix.rows()+m_index.value())
-                               : numext::mini<Index>(m_matrix.rows(),m_matrix.cols()-m_index.value());
-    }
+    { 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()); }
 
-    EIGEN_DEVICE_FUNC
     inline Index cols() const { return 1; }
 
-    EIGEN_DEVICE_FUNC
     inline Index innerStride() const
     {
       return m_matrix.outerStride() + 1;
     }
 
-    EIGEN_DEVICE_FUNC
     inline Index outerStride() const
     {
       return 0;
@@ -102,75 +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& 
+    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 Index absDiagIndex() const { return m_index.value()>0 ? m_index.value() : -m_index.value(); }
-    EIGEN_DEVICE_FUNC
     EIGEN_STRONG_INLINE Index rowOffset() const { return m_index.value()>0 ? 0 : -m_index.value(); }
-    EIGEN_DEVICE_FUNC
     EIGEN_STRONG_INLINE Index colOffset() const { return m_index.value()>0 ? m_index.value() : 0; }
-    // trigger a compile-time error if someone try to call packet
+    // 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;
 };
@@ -187,7 +167,7 @@ template<typename Derived>
 inline typename MatrixBase<Derived>::DiagonalReturnType
 MatrixBase<Derived>::diagonal()
 {
-  return DiagonalReturnType(derived());
+  return derived();
 }
 
 /** This is the const version of diagonal(). */
@@ -236,20 +216,20 @@ MatrixBase<Derived>::diagonal(Index index) const
   *
   * \sa MatrixBase::diagonal(), class Diagonal */
 template<typename Derived>
-template<int Index_>
-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_>
-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,
@@ -30,61 +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(); }
-    
-    EIGEN_DEVICE_FUNC
+    template<typename DenseDerived>
+    void evalTo(MatrixBase<DenseDerived> &other) const;
+    template<typename DenseDerived>
+    inline void addTo(MatrixBase<DenseDerived> &other) const
+    { other.diagonal() += diagonal(); }
+    template<typename DenseDerived>
+    inline void subTo(MatrixBase<DenseDerived> &other) const
+    { other.diagonal() -= diagonal(); }
+
     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>
+    bool isApprox(const DiagonalBase<OtherDerived>& other, typename NumTraits<Scalar>::Real precision = NumTraits<Scalar>::dummy_precision()) const
+    {
+      return diagonal().isApprox(other.diagonal(), precision);
+    }
+    template<typename OtherDerived>
+    bool isApprox(const MatrixBase<OtherDerived>& other, typename NumTraits<Scalar>::Real precision = NumTraits<Scalar>::dummy_precision()) const
+    {
+      return toDenseMatrix().isApprox(other, precision);
+    }
+    #endif
 };
 
+template<typename Derived>
+template<typename DenseDerived>
+inline void DiagonalBase<Derived>::evalTo(MatrixBase<DenseDerived> &other) const
+{
+  other.setZero();
+  other.diagonal() = diagonal();
+}
 #endif
 
 /** \class DiagonalMatrix
@@ -106,9 +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
   };
 };
 }
@@ -122,7 +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:
@@ -132,31 +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) {}
 
     /** Copy constructor. */
     template<typename OtherDerived>
-    EIGEN_DEVICE_FUNC
     inline DiagonalMatrix(const DiagonalBase<OtherDerived>& other) : m_diagonal(other.diagonal()) {}
 
     #ifndef EIGEN_PARSED_BY_DOXYGEN
@@ -166,13 +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();
@@ -183,7 +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();
@@ -192,19 +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); }
 };
 
@@ -228,15 +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
   };
 };
 }
@@ -252,11 +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:
@@ -276,7 +277,7 @@ template<typename 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,
@@ -290,11 +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)
@@ -306,38 +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 +13,117 @@
 
 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|AlignedBit) & (unsigned int)(MatrixType::Flags)) | (_Vectorizable ? PacketAccessBit : 0),//(int(MatrixType::Flags)&int(DiagonalType::DiagonalVectorType::Flags)&AlignedBit),
+    Cost0 = EIGEN_ADD_COST(NumTraits<Scalar>::MulCost, MatrixType::CoeffReadCost),
+    CoeffReadCost = EIGEN_ADD_COST(Cost0,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>
-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

@@ -28,24 +28,20 @@ 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
+  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
+  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();
   }
 };
 
@@ -63,30 +59,55 @@ struct dot_nocheck<T, U, true>
   */
 template<typename Derived>
 template<typename OtherDerived>
-EIGEN_DEVICE_FUNC
-typename ScalarBinaryOpTraits<typename internal::traits<Derived>::Scalar,typename internal::traits<OtherDerived>::Scalar>::ReturnType
+inline 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_STRONG_INLINE typename NumTraits<typename internal::traits<Derived>::Scalar>::Real MatrixBase<Derived>::squaredNorm() const
@@ -98,18 +119,16 @@ EIGEN_STRONG_INLINE typename NumTraits<typename internal::traits<Derived>::Scala
   * 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>
 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
   *
@@ -119,77 +138,22 @@ template<typename Derived>
 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>
 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>
-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>
-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 ----------
@@ -200,10 +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_MATH(pow)
+    using std::pow;
     return pow(m.cwiseAbs().array().pow(p).sum(), RealScalar(1)/p);
   }
 };
@@ -211,7 +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();
@@ -221,7 +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();
@@ -231,35 +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
 inline typename NumTraits<typename internal::traits<Derived>::Scalar>::Real
-#else
-MatrixBase<Derived>::RealScalar
-#endif
 MatrixBase<Derived>::lpNorm() const
 {
   return internal::lpNorm_selector<Derived, p>::run(*this);
@@ -278,8 +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();
 }
 
@@ -297,13 +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

@@ -13,9 +13,7 @@
 
 namespace Eigen {
 
-/** \class EigenBase
-  * 
-  * 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.
   *
@@ -23,57 +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.
-    * \deprecated Since Eigen 3.3, its usage is deprecated. Use Eigen::Index instead.
-    * \sa StorageIndex, \ref TopicPreprocessorDirectives.
-    */
-  typedef Eigen::Index Index;
 
-  // FIXME is it needed?
   typedef typename internal::traits<Derived>::StorageKind StorageKind;
+  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
   inline Index rows() const { return derived().rows(); }
   /** \returns the number of columns. \sa rows(), ColsAtCompileTime*/
-  EIGEN_DEVICE_FUNC
   inline Index cols() const { return derived().cols(); }
   /** \returns the number of coefficients, which is rows()*cols().
     * \sa rows(), cols(), SizeAtCompileTime. */
-  EIGEN_DEVICE_FUNC
   inline Index size() const { return rows() * cols(); }
 
   /** \internal Don't use it, but do the equivalent: \code dst = *this; \endcode */
-  template<typename Dest>
-  EIGEN_DEVICE_FUNC
-  inline void evalTo(Dest& dst) const
+  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.
@@ -83,9 +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.
@@ -95,8 +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.
@@ -104,8 +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.
@@ -130,7 +106,7 @@ template<typename Derived>
 template<typename OtherDerived>
 Derived& DenseBase<Derived>::operator=(const EigenBase<OtherDerived> &other)
 {
-  call_assignment(derived(), other.derived());
+  other.derived().evalTo(derived());
   return derived();
 }
 
@@ -138,7 +114,7 @@ template<typename Derived>
 template<typename OtherDerived>
 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();
 }
 
@@ -146,7 +122,7 @@ template<typename Derived>
 template<typename OtherDerived>
 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();
 }
 

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

@@ -39,29 +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 inline Index rows() const { return m_expression.rows(); }
-    EIGEN_DEVICE_FUNC inline Index cols() const { return m_expression.cols(); }
-    EIGEN_DEVICE_FUNC inline Index outerStride() const { return m_expression.outerStride(); }
-    EIGEN_DEVICE_FUNC inline Index innerStride() const { return m_expression.innerStride(); }
+    inline 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);
     }
@@ -90,7 +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;
@@ -127,7 +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.
@@ -138,7 +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/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();

Eigen/src/Core/GeneralProduct.h

@@ -11,7 +11,29 @@
 #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,
@@ -25,8 +47,7 @@ template<int Rows, int Cols, int Depth> struct product_type_selector;
 template<int Size, int MaxSize> struct product_size_category
 {
   enum { is_large = MaxSize == Dynamic ||
-                    Size >= EIGEN_CACHEFRIENDLY_PRODUCT_THRESHOLD ||
-                    (Size==Dynamic && MaxSize>=EIGEN_CACHEFRIENDLY_PRODUCT_THRESHOLD),
+                    Size >= EIGEN_CACHEFRIENDLY_PRODUCT_THRESHOLD,
          value = is_large  ? Large
                : Size == 1 ? 1
                            : Small
@@ -38,14 +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,
@@ -60,8 +82,7 @@ private:
 
 public:
   enum {
-    value = selector::ret,
-    ret = selector::ret
+    value = selector::ret
   };
 #ifdef EIGEN_DEBUG_PRODUCT
   static void debug()
@@ -77,13 +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 }; };
@@ -102,12 +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
 ***********************************************************************/
@@ -119,10 +187,114 @@ 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)
+    {
+      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(0,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,0) * 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 is_row_major : 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)
+    {
+    }
+    
+    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(), is_row_major<Dest>());
+    }
+    
+    template<typename Dest>
+    inline void addTo(Dest& dest) const {
+      internal::outer_product_selector_run(*this, dest, add(), is_row_major<Dest>());
+    }
+
+    template<typename Dest>
+    inline void subTo(Dest& dest) const {
+      internal::outer_product_selector_run(*this, dest, sub(), is_row_major<Dest>());
+    }
+
+    template<typename Dest> void scaleAndAddTo(Dest& dest, const Scalar& alpha) const
+    {
+      internal::outer_product_selector_run(*this, dest, adds(alpha), is_row_major<Dest>());
+    }
+};
+
 /***********************************************************************
 *  Implementation of General Matrix Vector Product
 ***********************************************************************/
@@ -136,13 +308,60 @@ 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>
@@ -160,61 +379,46 @@ struct gemv_static_vector_if<Scalar,Size,Dynamic,true>
 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 = alpha * LhsBlasTraits::extractScalarFactor(lhs)
-                                  * RhsBlasTraits::extractScalarFactor(rhs);
+    ResScalar actualAlpha = alpha * LhsBlasTraits::extractScalarFactor(prod.lhs())
+                                  * RhsBlasTraits::extractScalarFactor(prod.rhs());
 
     // make sure Dest is a compile-time vector type (bug 1166)
     typedef typename conditional<Dest::IsVectorAtCompileTime, Dest, typename Dest::ColXpr>::type ActualDest;
@@ -229,18 +433,18 @@ template<> struct gemv_dense_selector<OnTheRight,ColMajor,true>
 
     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;
-
+    bool alphaIsCompatible = (!ComplexByReal) || (numext::imag(actualAlpha)==RealScalar(0));
+    bool evalToDest = EvalToDestAtCompileTime && alphaIsCompatible;
+    
     RhsScalar compatibleAlpha = get_factor<ResScalar,RhsScalar>::run(actualAlpha);
 
     ei_declare_aligned_stack_constructed_variable(ResScalar,actualDestPtr,dest.size(),
                                                   evalToDest ? dest.data() : static_dest.data());
-
+    
     if(!evalToDest)
     {
       #ifdef EIGEN_DENSE_STORAGE_CTOR_PLUGIN
-      Index size = dest.size();
+      int size = dest.size();
       EIGEN_DENSE_STORAGE_CTOR_PLUGIN
       #endif
       if(!alphaIsCompatible)
@@ -252,54 +456,52 @@ template<> struct gemv_dense_selector<OnTheRight,ColMajor,true>
         MappedDest(actualDestPtr, dest.size()) = dest;
     }
 
-    typedef const_blas_data_mapper<LhsScalar,Index,ColMajor> LhsMapper;
-    typedef const_blas_data_mapper<RhsScalar,Index,RowMajor> RhsMapper;
     general_matrix_vector_product
-        <Index,LhsScalar,LhsMapper,ColMajor,LhsBlasTraits::NeedToConjugate,RhsScalar,RhsMapper,RhsBlasTraits::NeedToConjugate>::run(
+      <Index,LhsScalar,ColMajor,LhsBlasTraits::NeedToConjugate,RhsScalar,RhsBlasTraits::NeedToConjugate>::run(
         actualLhs.rows(), actualLhs.cols(),
-        LhsMapper(actualLhs.data(), actualLhs.outerStride()),
-        RhsMapper(actualRhs.data(), actualRhs.innerStride()),
+        actualLhs.data(), actualLhs.outerStride(),
+        actualRhs.data(), actualRhs.innerStride(),
         actualDestPtr, 1,
         compatibleAlpha);
 
     if (!evalToDest)
     {
       if(!alphaIsCompatible)
-        dest.matrix() += actualAlpha * MappedDest(actualDestPtr, dest.size());
+        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 = alpha * LhsBlasTraits::extractScalarFactor(lhs)
-                                  * RhsBlasTraits::extractScalarFactor(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
+      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());
@@ -307,48 +509,45 @@ 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),
+        actualLhs.data(), actualLhs.outerStride(),
+        actualRhsPtr, 1,
         dest.data(), dest.col(0).innerStride(), //NOTE  if dest is not a vector at compile-time, then dest.innerStride() might be wrong. (bug 1166)
         actualAlpha);
   }
 };
 
-template<> struct gemv_dense_selector<OnTheRight,ColMajor,false>
+template<> 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();
   }
 };
 
@@ -364,11 +563,9 @@ template<> struct gemv_dense_selector<OnTheRight,RowMajor,false>
   *
   * \sa lazyProduct(), operator*=(const MatrixBase&), Cwise::operator*()
   */
-#ifndef __CUDACC__
-
 template<typename Derived>
 template<typename OtherDerived>
-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
@@ -393,12 +590,9 @@ 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());
 }
 
-#endif // __CUDACC__
-
 /** \returns an expression of the matrix product of \c *this and \a other without implicit evaluation.
   *
   * The returned product will behave like any other expressions: the coefficients of the product will be
@@ -412,7 +606,7 @@ MatrixBase<Derived>::operator*(const MatrixBase<OtherDerived> &other) const
   */
 template<typename Derived>
 template<typename OtherDerived>
-const Product<Derived,OtherDerived,LazyProduct>
+const typename LazyProductReturnType<Derived,OtherDerived>::Type
 MatrixBase<Derived>::lazyProduct(const MatrixBase<OtherDerived> &other) const
 {
   enum {
@@ -431,7 +625,7 @@ 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/GenericPacketMath.h

@@ -42,28 +42,21 @@ namespace internal {
 struct default_packet_traits
 {
   enum {
-    HasHalfPacket = 0,
-
     HasAdd    = 1,
     HasSub    = 1,
     HasMul    = 1,
     HasNegate = 1,
     HasAbs    = 1,
-    HasArg    = 0,
     HasAbs2   = 1,
     HasMin    = 1,
     HasMax    = 1,
     HasConj   = 1,
     HasSetLinear = 1,
-    HasBlend  = 0,
 
     HasDiv    = 0,
     HasSqrt   = 0,
-    HasRsqrt  = 0,
     HasExp    = 0,
     HasLog    = 0,
-    HasLog1p  = 0,
-    HasLog10  = 0,
     HasPow    = 0,
 
     HasSin    = 0,
@@ -71,37 +64,17 @@ struct default_packet_traits
     HasTan    = 0,
     HasASin   = 0,
     HasACos   = 0,
-    HasATan   = 0,
-    HasSinh   = 0,
-    HasCosh   = 0,
-    HasTanh   = 0,
-    HasLGamma = 0,
-    HasDiGamma = 0,
-    HasZeta = 0,
-    HasPolygamma = 0,
-    HasErf = 0,
-    HasErfc = 0,
-    HasIGamma = 0,
-    HasIGammac = 0,
-    HasBetaInc = 0,
-
-    HasRound  = 0,
-    HasFloor  = 0,
-    HasCeil   = 0,
-
-    HasSign   = 0
+    HasATan   = 0
   };
 };
 
 template<typename T> struct packet_traits : default_packet_traits
 {
   typedef T type;
-  typedef T half;
   enum {
     Vectorizable = 0,
     size = 1,
-    AlignedOnScalar = 0,
-    HasHalfPacket = 0
+    AlignedOnScalar = 0
   };
   enum {
     HasAdd    = 0,
@@ -117,239 +90,135 @@ template<typename T> struct packet_traits : default_packet_traits
   };
 };
 
-template<typename T> struct packet_traits<const T> : packet_traits<T> { };
-
-template <typename Src, typename Tgt> struct type_casting_traits {
-  enum {
-    VectorizedCast = 0,
-    SrcCoeffRatio = 1,
-    TgtCoeffRatio = 1
-  };
-};
-
-
-/** \internal \returns static_cast<TgtType>(a) (coeff-wise) */
-template <typename SrcPacket, typename TgtPacket>
-EIGEN_DEVICE_FUNC inline TgtPacket
-pcast(const SrcPacket& a) {
-  return static_cast<TgtPacket>(a);
-}
-template <typename SrcPacket, typename TgtPacket>
-EIGEN_DEVICE_FUNC inline TgtPacket
-pcast(const SrcPacket& a, const SrcPacket& /*b*/) {
-  return static_cast<TgtPacket>(a);
-}
-
-template <typename SrcPacket, typename TgtPacket>
-EIGEN_DEVICE_FUNC inline TgtPacket
-pcast(const SrcPacket& a, const SrcPacket& /*b*/, const SrcPacket& /*c*/, const SrcPacket& /*d*/) {
-  return static_cast<TgtPacket>(a);
-}
-
 /** \internal \returns a + b (coeff-wise) */
-template<typename Packet> EIGEN_DEVICE_FUNC inline Packet
+template<typename Packet> inline Packet
 padd(const Packet& a,
         const Packet& b) { return a+b; }
 
 /** \internal \returns a - b (coeff-wise) */
-template<typename Packet> EIGEN_DEVICE_FUNC inline Packet
+template<typename Packet> inline Packet
 psub(const Packet& a,
         const Packet& b) { return a-b; }
 
 /** \internal \returns -a (coeff-wise) */
-template<typename Packet> EIGEN_DEVICE_FUNC inline Packet
+template<typename Packet> inline Packet
 pnegate(const Packet& a) { return -a; }
 
 /** \internal \returns conj(a) (coeff-wise) */
-
-template<typename Packet> EIGEN_DEVICE_FUNC inline Packet
+template<typename Packet> inline Packet
 pconj(const Packet& a) { return numext::conj(a); }
 
 /** \internal \returns a * b (coeff-wise) */
-template<typename Packet> EIGEN_DEVICE_FUNC inline Packet
+template<typename Packet> inline Packet
 pmul(const Packet& a,
         const Packet& b) { return a*b; }
 
 /** \internal \returns a / b (coeff-wise) */
-template<typename Packet> EIGEN_DEVICE_FUNC inline Packet
+template<typename Packet> inline Packet
 pdiv(const Packet& a,
         const Packet& b) { return a/b; }
 
 /** \internal \returns the min of \a a and \a b  (coeff-wise) */
-template<typename Packet> EIGEN_DEVICE_FUNC inline Packet
+template<typename Packet> inline Packet
 pmin(const Packet& a,
-        const Packet& b) { return numext::mini(a, b); }
+        const Packet& b) { using std::min; return (min)(a, b); }
 
 /** \internal \returns the max of \a a and \a b  (coeff-wise) */
-template<typename Packet> EIGEN_DEVICE_FUNC inline Packet
+template<typename Packet> inline Packet
 pmax(const Packet& a,
-        const Packet& b) { return numext::maxi(a, b); }
+        const Packet& b) { using std::max; return (max)(a, b); }
 
 /** \internal \returns the absolute value of \a a */
-template<typename Packet> EIGEN_DEVICE_FUNC inline Packet
+template<typename Packet> inline Packet
 pabs(const Packet& a) { using std::abs; return abs(a); }
 
-/** \internal \returns the phase angle of \a a */
-template<typename Packet> EIGEN_DEVICE_FUNC inline Packet
-parg(const Packet& a) { using numext::arg; return arg(a); }
-
 /** \internal \returns the bitwise and of \a a and \a b */
-template<typename Packet> EIGEN_DEVICE_FUNC inline Packet
+template<typename Packet> inline Packet
 pand(const Packet& a, const Packet& b) { return a & b; }
 
 /** \internal \returns the bitwise or of \a a and \a b */
-template<typename Packet> EIGEN_DEVICE_FUNC inline Packet
+template<typename Packet> inline Packet
 por(const Packet& a, const Packet& b) { return a | b; }
 
 /** \internal \returns the bitwise xor of \a a and \a b */
-template<typename Packet> EIGEN_DEVICE_FUNC inline Packet
+template<typename Packet> inline Packet
 pxor(const Packet& a, const Packet& b) { return a ^ b; }
 
 /** \internal \returns the bitwise andnot of \a a and \a b */
-template<typename Packet> EIGEN_DEVICE_FUNC inline Packet
+template<typename Packet> inline Packet
 pandnot(const Packet& a, const Packet& b) { return a & (!b); }
 
 /** \internal \returns a packet version of \a *from, from must be 16 bytes aligned */
-template<typename Packet> EIGEN_DEVICE_FUNC inline Packet
+template<typename Packet> inline Packet
 pload(const typename unpacket_traits<Packet>::type* from) { return *from; }
 
 /** \internal \returns a packet version of \a *from, (un-aligned load) */
-template<typename Packet> EIGEN_DEVICE_FUNC inline Packet
+template<typename Packet> inline Packet
 ploadu(const typename unpacket_traits<Packet>::type* from) { return *from; }
 
-/** \internal \returns a packet with constant coefficients \a a, e.g.: (a,a,a,a) */
-template<typename Packet> EIGEN_DEVICE_FUNC inline Packet
-pset1(const typename unpacket_traits<Packet>::type& a) { return a; }
-
-/** \internal \returns a packet with constant coefficients \a a[0], e.g.: (a[0],a[0],a[0],a[0]) */
-template<typename Packet> EIGEN_DEVICE_FUNC inline Packet
-pload1(const typename unpacket_traits<Packet>::type  *a) { return pset1<Packet>(*a); }
-
 /** \internal \returns a packet with elements of \a *from duplicated.
-  * For instance, for a packet of 8 elements, 4 scalars will be read from \a *from and
-  * duplicated to form: {from[0],from[0],from[1],from[1],from[2],from[2],from[3],from[3]}
+  * For instance, for a packet of 8 elements, 4 scalar will be read from \a *from and
+  * duplicated to form: {from[0],from[0],from[1],from[1],,from[2],from[2],,from[3],from[3]}
   * Currently, this function is only used for scalar * complex products.
-  */
-template<typename Packet> EIGEN_DEVICE_FUNC inline Packet
+ */
+template<typename Packet> inline Packet
 ploaddup(const typename unpacket_traits<Packet>::type* from) { return *from; }
 
-/** \internal \returns a packet with elements of \a *from quadrupled.
-  * For instance, for a packet of 8 elements, 2 scalars will be read from \a *from and
-  * replicated to form: {from[0],from[0],from[0],from[0],from[1],from[1],from[1],from[1]}
-  * Currently, this function is only used in matrix products.
-  * For packet-size smaller or equal to 4, this function is equivalent to pload1 
-  */
-template<typename Packet> EIGEN_DEVICE_FUNC inline Packet
-ploadquad(const typename unpacket_traits<Packet>::type* from)
-{ return pload1<Packet>(from); }
-
-/** \internal equivalent to
-  * \code
-  * a0 = pload1(a+0);
-  * a1 = pload1(a+1);
-  * a2 = pload1(a+2);
-  * a3 = pload1(a+3);
-  * \endcode
-  * \sa pset1, pload1, ploaddup, pbroadcast2
-  */
-template<typename Packet> EIGEN_DEVICE_FUNC
-inline void pbroadcast4(const typename unpacket_traits<Packet>::type *a,
-                        Packet& a0, Packet& a1, Packet& a2, Packet& a3)
-{
-  a0 = pload1<Packet>(a+0);
-  a1 = pload1<Packet>(a+1);
-  a2 = pload1<Packet>(a+2);
-  a3 = pload1<Packet>(a+3);
-}
-
-/** \internal equivalent to
-  * \code
-  * a0 = pload1(a+0);
-  * a1 = pload1(a+1);
-  * \endcode
-  * \sa pset1, pload1, ploaddup, pbroadcast4
-  */
-template<typename Packet> EIGEN_DEVICE_FUNC
-inline void pbroadcast2(const typename unpacket_traits<Packet>::type *a,
-                        Packet& a0, Packet& a1)
-{
-  a0 = pload1<Packet>(a+0);
-  a1 = pload1<Packet>(a+1);
-}
+/** \internal \returns a packet with constant coefficients \a a, e.g.: (a,a,a,a) */
+template<typename Packet> inline Packet
+pset1(const typename unpacket_traits<Packet>::type& a) { return a; }
 
 /** \internal \brief Returns a packet with coefficients (a,a+1,...,a+packet_size-1). */
-template<typename Packet> inline Packet
-plset(const typename unpacket_traits<Packet>::type& a) { return a; }
+template<typename Scalar> inline typename packet_traits<Scalar>::type
+plset(const Scalar& a) { return a; }
 
 /** \internal copy the packet \a from to \a *to, \a to must be 16 bytes aligned */
-template<typename Scalar, typename Packet> EIGEN_DEVICE_FUNC inline void pstore(Scalar* to, const Packet& from)
+template<typename Scalar, typename Packet> inline void pstore(Scalar* to, const Packet& from)
 { (*to) = from; }
 
 /** \internal copy the packet \a from to \a *to, (un-aligned store) */
-template<typename Scalar, typename Packet> EIGEN_DEVICE_FUNC inline void pstoreu(Scalar* to, const Packet& from)
-{  (*to) = from; }
-
- template<typename Scalar, typename Packet> EIGEN_DEVICE_FUNC inline Packet pgather(const Scalar* from, Index /*stride*/)
- { return ploadu<Packet>(from); }
-
- template<typename Scalar, typename Packet> EIGEN_DEVICE_FUNC inline void pscatter(Scalar* to, const Packet& from, Index /*stride*/)
- { pstore(to, from); }
+template<typename Scalar, typename Packet> inline void pstoreu(Scalar* to, const Packet& from)
+{ (*to) = from; }
 
 /** \internal tries to do cache prefetching of \a addr */
-template<typename Scalar> EIGEN_DEVICE_FUNC inline void prefetch(const Scalar* addr)
+template<typename Scalar> inline void prefetch(const Scalar* addr)
 {
-#ifdef __CUDA_ARCH__
-#if defined(__LP64__)
-  // 64-bit pointer operand constraint for inlined asm
-  asm(" prefetch.L1 [ %1 ];" : "=l"(addr) : "l"(addr));
-#else
-  // 32-bit pointer operand constraint for inlined asm
-  asm(" prefetch.L1 [ %1 ];" : "=r"(addr) : "r"(addr));
-#endif
-#elif (!EIGEN_COMP_MSVC) && (EIGEN_COMP_GNUC || EIGEN_COMP_CLANG || EIGEN_COMP_ICC)
+#if (!EIGEN_COMP_MSVC) && (EIGEN_COMP_GNUC || EIGEN_COMP_CLANG || EIGEN_COMP_ICC)
   __builtin_prefetch(addr);
 #endif
 }
 
 /** \internal \returns the first element of a packet */
-template<typename Packet> EIGEN_DEVICE_FUNC inline typename unpacket_traits<Packet>::type pfirst(const Packet& a)
+template<typename Packet> inline typename unpacket_traits<Packet>::type pfirst(const Packet& a)
 { return a; }
 
 /** \internal \returns a packet where the element i contains the sum of the packet of \a vec[i] */
-template<typename Packet> EIGEN_DEVICE_FUNC inline Packet
+template<typename Packet> inline Packet
 preduxp(const Packet* vecs) { return vecs[0]; }
 
 /** \internal \returns the sum of the elements of \a a*/
-template<typename Packet> EIGEN_DEVICE_FUNC inline typename unpacket_traits<Packet>::type predux(const Packet& a)
-{ return a; }
-
-/** \internal \returns the sum of the elements of \a a by block of 4 elements.
-  * For a packet {a0, a1, a2, a3, a4, a5, a6, a7}, it returns a half packet {a0+a4, a1+a5, a2+a6, a3+a7}
-  * For packet-size smaller or equal to 4, this boils down to a noop.
-  */
-template<typename Packet> EIGEN_DEVICE_FUNC inline
-typename conditional<(unpacket_traits<Packet>::size%8)==0,typename unpacket_traits<Packet>::half,Packet>::type
-predux_downto4(const Packet& a)
+template<typename Packet> inline typename unpacket_traits<Packet>::type predux(const Packet& a)
 { return a; }
 
 /** \internal \returns the product of the elements of \a a*/
-template<typename Packet> EIGEN_DEVICE_FUNC inline typename unpacket_traits<Packet>::type predux_mul(const Packet& a)
+template<typename Packet> inline typename unpacket_traits<Packet>::type predux_mul(const Packet& a)
 { return a; }
 
 /** \internal \returns the min of the elements of \a a*/
-template<typename Packet> EIGEN_DEVICE_FUNC inline typename unpacket_traits<Packet>::type predux_min(const Packet& a)
+template<typename Packet> inline typename unpacket_traits<Packet>::type predux_min(const Packet& a)
 { return a; }
 
 /** \internal \returns the max of the elements of \a a*/
-template<typename Packet> EIGEN_DEVICE_FUNC inline typename unpacket_traits<Packet>::type predux_max(const Packet& a)
+template<typename Packet> inline typename unpacket_traits<Packet>::type predux_max(const Packet& a)
 { return a; }
 
 /** \internal \returns the reversed elements of \a a*/
-template<typename Packet> EIGEN_DEVICE_FUNC inline Packet preverse(const Packet& a)
+template<typename Packet> inline Packet preverse(const Packet& a)
 { return a; }
 
+
 /** \internal \returns \a a with real and imaginary part flipped (for complex type only) */
-template<typename Packet> EIGEN_DEVICE_FUNC inline Packet pcplxflip(const Packet& a)
+template<typename Packet> inline Packet pcplxflip(const Packet& a)
 {
   // FIXME: uncomment the following in case we drop the internal imag and real functions.
 //   using std::imag;
@@ -381,22 +250,6 @@ Packet pasin(const Packet& a) { using std::asin; return asin(a); }
 template<typename Packet> EIGEN_DECLARE_FUNCTION_ALLOWING_MULTIPLE_DEFINITIONS
 Packet pacos(const Packet& a) { using std::acos; return acos(a); }
 
-/** \internal \returns the arc tangent of \a a (coeff-wise) */
-template<typename Packet> EIGEN_DECLARE_FUNCTION_ALLOWING_MULTIPLE_DEFINITIONS
-Packet patan(const Packet& a) { using std::atan; return atan(a); }
-
-/** \internal \returns the hyperbolic sine of \a a (coeff-wise) */
-template<typename Packet> EIGEN_DECLARE_FUNCTION_ALLOWING_MULTIPLE_DEFINITIONS
-Packet psinh(const Packet& a) { using std::sinh; return sinh(a); }
-
-/** \internal \returns the hyperbolic cosine of \a a (coeff-wise) */
-template<typename Packet> EIGEN_DECLARE_FUNCTION_ALLOWING_MULTIPLE_DEFINITIONS
-Packet pcosh(const Packet& a) { using std::cosh; return cosh(a); }
-
-/** \internal \returns the hyperbolic tan of \a a (coeff-wise) */
-template<typename Packet> EIGEN_DECLARE_FUNCTION_ALLOWING_MULTIPLE_DEFINITIONS
-Packet ptanh(const Packet& a) { using std::tanh; return tanh(a); }
-
 /** \internal \returns the exp of \a a (coeff-wise) */
 template<typename Packet> EIGEN_DECLARE_FUNCTION_ALLOWING_MULTIPLE_DEFINITIONS
 Packet pexp(const Packet& a) { using std::exp; return exp(a); }
@@ -405,36 +258,10 @@ Packet pexp(const Packet& a) { using std::exp; return exp(a); }
 template<typename Packet> EIGEN_DECLARE_FUNCTION_ALLOWING_MULTIPLE_DEFINITIONS
 Packet plog(const Packet& a) { using std::log; return log(a); }
 
-/** \internal \returns the log1p of \a a (coeff-wise) */
-template<typename Packet> EIGEN_DECLARE_FUNCTION_ALLOWING_MULTIPLE_DEFINITIONS
-Packet plog1p(const Packet& a) { return numext::log1p(a); }
-
-/** \internal \returns the log10 of \a a (coeff-wise) */
-template<typename Packet> EIGEN_DECLARE_FUNCTION_ALLOWING_MULTIPLE_DEFINITIONS
-Packet plog10(const Packet& a) { using std::log10; return log10(a); }
-
 /** \internal \returns the square-root of \a a (coeff-wise) */
 template<typename Packet> EIGEN_DECLARE_FUNCTION_ALLOWING_MULTIPLE_DEFINITIONS
 Packet psqrt(const Packet& a) { using std::sqrt; return sqrt(a); }
 
-/** \internal \returns the reciprocal square-root of \a a (coeff-wise) */
-template<typename Packet> EIGEN_DECLARE_FUNCTION_ALLOWING_MULTIPLE_DEFINITIONS
-Packet prsqrt(const Packet& a) {
-  return pdiv(pset1<Packet>(1), psqrt(a));
-}
-
-/** \internal \returns the rounded value of \a a (coeff-wise) */
-template<typename Packet> EIGEN_DECLARE_FUNCTION_ALLOWING_MULTIPLE_DEFINITIONS
-Packet pround(const Packet& a) { using numext::round; return round(a); }
-
-/** \internal \returns the floor of \a a (coeff-wise) */
-template<typename Packet> EIGEN_DECLARE_FUNCTION_ALLOWING_MULTIPLE_DEFINITIONS
-Packet pfloor(const Packet& a) { using numext::floor; return floor(a); }
-
-/** \internal \returns the ceil of \a a (coeff-wise) */
-template<typename Packet> EIGEN_DECLARE_FUNCTION_ALLOWING_MULTIPLE_DEFINITIONS
-Packet pceil(const Packet& a) { using numext::ceil; return ceil(a); }
-
 /***************************************************************************
 * The following functions might not have to be overwritten for vectorized types
 ***************************************************************************/
@@ -448,45 +275,34 @@ inline void pstore1(typename unpacket_traits<Packet>::type* to, const typename u
 }
 
 /** \internal \returns a * b + c (coeff-wise) */
-template<typename Packet> EIGEN_DEVICE_FUNC inline Packet
+template<typename Packet> inline Packet
 pmadd(const Packet&  a,
          const Packet&  b,
          const Packet&  c)
 { return padd(pmul(a, b),c); }
 
 /** \internal \returns a packet version of \a *from.
-  * The pointer \a from must be aligned on a \a Alignment bytes boundary. */
-template<typename Packet, int Alignment>
-EIGEN_DEVICE_FUNC EIGEN_ALWAYS_INLINE Packet ploadt(const typename unpacket_traits<Packet>::type* from)
+  * If LoadMode equals #Aligned, \a from must be 16 bytes aligned */
+template<typename Packet, int LoadMode>
+inline Packet ploadt(const typename unpacket_traits<Packet>::type* from)
 {
-  if(Alignment >= unpacket_traits<Packet>::alignment)
+  if(LoadMode == Aligned)
     return pload<Packet>(from);
   else
     return ploadu<Packet>(from);
 }
 
 /** \internal copy the packet \a from to \a *to.
-  * The pointer \a from must be aligned on a \a Alignment bytes boundary. */
-template<typename Scalar, typename Packet, int Alignment>
-EIGEN_DEVICE_FUNC EIGEN_ALWAYS_INLINE void pstoret(Scalar* to, const Packet& from)
+  * If StoreMode equals #Aligned, \a to must be 16 bytes aligned */
+template<typename Scalar, typename Packet, int LoadMode>
+inline void pstoret(Scalar* to, const Packet& from)
 {
-  if(Alignment >= unpacket_traits<Packet>::alignment)
+  if(LoadMode == Aligned)
     pstore(to, from);
   else
     pstoreu(to, from);
 }
 
-/** \internal \returns a packet version of \a *from.
-  * Unlike ploadt, ploadt_ro takes advantage of the read-only memory path on the
-  * hardware if available to speedup the loading of data that won't be modified
-  * by the current computation.
-  */
-template<typename Packet, int LoadMode>
-inline Packet ploadt_ro(const typename unpacket_traits<Packet>::type* from)
-{
-  return ploadt<Packet, LoadMode>(from);
-}
-
 /** \internal default implementation of palign() allowing partial specialization */
 template<int Offset,typename PacketType>
 struct palign_impl
@@ -520,74 +336,15 @@ inline void palign(PacketType& first, const PacketType& second)
 * Fast complex products (GCC generates a function call which is very slow)
 ***************************************************************************/
 
-// Eigen+CUDA does not support complexes.
-#ifndef __CUDACC__
-
 template<> inline std::complex<float> pmul(const std::complex<float>& a, const std::complex<float>& b)
 { return std::complex<float>(real(a)*real(b) - imag(a)*imag(b), imag(a)*real(b) + real(a)*imag(b)); }
 
 template<> inline std::complex<double> pmul(const std::complex<double>& a, const std::complex<double>& b)
 { return std::complex<double>(real(a)*real(b) - imag(a)*imag(b), imag(a)*real(b) + real(a)*imag(b)); }
 
-#endif
-
-
-/***************************************************************************
- * PacketBlock, that is a collection of N packets where the number of words
- * in the packet is a multiple of N.
-***************************************************************************/
-template <typename Packet,int N=unpacket_traits<Packet>::size> struct PacketBlock {
-  Packet packet[N];
-};
-
-template<typename Packet> EIGEN_DEVICE_FUNC inline void
-ptranspose(PacketBlock<Packet,1>& /*kernel*/) {
-  // Nothing to do in the scalar case, i.e. a 1x1 matrix.
-}
-
-/***************************************************************************
- * Selector, i.e. vector of N boolean values used to select (i.e. blend)
- * words from 2 packets.
-***************************************************************************/
-template <size_t N> struct Selector {
-  bool select[N];
-};
-
-template<typename Packet> EIGEN_DEVICE_FUNC inline Packet
-pblend(const Selector<unpacket_traits<Packet>::size>& ifPacket, const Packet& thenPacket, const Packet& elsePacket) {
-  return ifPacket.select[0] ? thenPacket : elsePacket;
-}
-
-/** \internal \returns \a a with the first coefficient replaced by the scalar b */
-template<typename Packet> EIGEN_DEVICE_FUNC inline Packet
-pinsertfirst(const Packet& a, typename unpacket_traits<Packet>::type b)
-{
-  // Default implementation based on pblend.
-  // It must be specialized for higher performance.
-  Selector<unpacket_traits<Packet>::size> mask;
-  mask.select[0] = true;
-  // This for loop should be optimized away by the compiler.
-  for(Index i=1; i<unpacket_traits<Packet>::size; ++i)
-    mask.select[i] = false;
-  return pblend(mask, pset1<Packet>(b), a);
-}
-
-/** \internal \returns \a a with the last coefficient replaced by the scalar b */
-template<typename Packet> EIGEN_DEVICE_FUNC inline Packet
-pinsertlast(const Packet& a, typename unpacket_traits<Packet>::type b)
-{
-  // Default implementation based on pblend.
-  // It must be specialized for higher performance.
-  Selector<unpacket_traits<Packet>::size> mask;
-  // This for loop should be optimized away by the compiler.
-  for(Index i=0; i<unpacket_traits<Packet>::size-1; ++i)
-    mask.select[i] = false;
-  mask.select[unpacket_traits<Packet>::size-1] = true;
-  return pblend(mask, pset1<Packet>(b), a);
-}
-
 } // end namespace internal
 
 } // end namespace Eigen
 
 #endif // EIGEN_GENERIC_PACKET_MATH_H
+

Eigen/src/Core/GlobalFunctions.h

@@ -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> \
@@ -47,133 +30,55 @@
   { \
     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)
-  EIGEN_ARRAY_DECLARE_GLOBAL_UNARY(lgamma,scalar_lgamma_op,natural logarithm of the gamma function,\sa ArrayBase::lgamma)
-  EIGEN_ARRAY_DECLARE_GLOBAL_UNARY(digamma,scalar_digamma_op,derivative of lgamma,\sa ArrayBase::digamma)
-  EIGEN_ARRAY_DECLARE_GLOBAL_UNARY(erf,scalar_erf_op,error function,\sa ArrayBase::erf)
-  EIGEN_ARRAY_DECLARE_GLOBAL_UNARY(erfc,scalar_erfc_op,complement error function,\sa ArrayBase::erfc)
-  EIGEN_ARRAY_DECLARE_GLOBAL_UNARY(exp,scalar_exp_op,exponential,\sa ArrayBase::exp)
-  EIGEN_ARRAY_DECLARE_GLOBAL_UNARY(log,scalar_log_op,natural logarithm,\sa Eigen::log10 DOXCOMMA ArrayBase::log)
-  EIGEN_ARRAY_DECLARE_GLOBAL_UNARY(log1p,scalar_log1p_op,natural logarithm of 1 plus the value,\sa ArrayBase::log1p)
-  EIGEN_ARRAY_DECLARE_GLOBAL_UNARY(log10,scalar_log10_op,base 10 logarithm,\sa Eigen::log DOXCOMMA ArrayBase::log)
-  EIGEN_ARRAY_DECLARE_GLOBAL_UNARY(abs,scalar_abs_op,absolute value,\sa ArrayBase::abs DOXCOMMA MatrixBase::cwiseAbs)
-  EIGEN_ARRAY_DECLARE_GLOBAL_UNARY(abs2,scalar_abs2_op,squared absolute value,\sa ArrayBase::abs2 DOXCOMMA MatrixBase::cwiseAbs2)
-  EIGEN_ARRAY_DECLARE_GLOBAL_UNARY(arg,scalar_arg_op,complex argument,\sa ArrayBase::arg)
-  EIGEN_ARRAY_DECLARE_GLOBAL_UNARY(sqrt,scalar_sqrt_op,square root,\sa ArrayBase::sqrt DOXCOMMA MatrixBase::cwiseSqrt)
-  EIGEN_ARRAY_DECLARE_GLOBAL_UNARY(rsqrt,scalar_rsqrt_op,reciprocal square root,\sa ArrayBase::rsqrt)
-  EIGEN_ARRAY_DECLARE_GLOBAL_UNARY(square,scalar_square_op,square (power 2),\sa Eigen::abs2 DOXCOMMA Eigen::pow DOXCOMMA ArrayBase::square)
-  EIGEN_ARRAY_DECLARE_GLOBAL_UNARY(cube,scalar_cube_op,cube (power 3),\sa Eigen::pow DOXCOMMA ArrayBase::cube)
-  EIGEN_ARRAY_DECLARE_GLOBAL_UNARY(round,scalar_round_op,nearest integer,\sa Eigen::floor DOXCOMMA Eigen::ceil DOXCOMMA ArrayBase::round)
-  EIGEN_ARRAY_DECLARE_GLOBAL_UNARY(floor,scalar_floor_op,nearest integer not greater than the giben value,\sa Eigen::ceil DOXCOMMA ArrayBase::floor)
-  EIGEN_ARRAY_DECLARE_GLOBAL_UNARY(ceil,scalar_ceil_op,nearest integer not less than the giben value,\sa Eigen::floor DOXCOMMA ArrayBase::ceil)
-  EIGEN_ARRAY_DECLARE_GLOBAL_UNARY(isnan,scalar_isnan_op,not-a-number test,\sa Eigen::isinf DOXCOMMA Eigen::isfinite DOXCOMMA ArrayBase::isnan)
-  EIGEN_ARRAY_DECLARE_GLOBAL_UNARY(isinf,scalar_isinf_op,infinite value test,\sa Eigen::isnan DOXCOMMA Eigen::isfinite DOXCOMMA ArrayBase::isinf)
-  EIGEN_ARRAY_DECLARE_GLOBAL_UNARY(isfinite,scalar_isfinite_op,finite value test,\sa Eigen::isinf DOXCOMMA Eigen::isnan DOXCOMMA ArrayBase::isfinite)
-  EIGEN_ARRAY_DECLARE_GLOBAL_UNARY(sign,scalar_sign_op,sign (or 0),\sa ArrayBase::sign)
+  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)
   
-  /** \returns an expression of the coefficient-wise power of \a x to the given constant \a exponent.
-    *
-    * \tparam ScalarExponent is the scalar type of \a exponent. It must be compatible with the scalar type of the given expression (\c Derived::Scalar).
-    *
-    * \sa ArrayBase::pow()
-    *
-    * \relates ArrayBase
-    */
-#ifdef EIGEN_PARSED_BY_DOXYGEN
-  template<typename Derived,typename ScalarExponent>
-  inline const CwiseBinaryOp<internal::scalar_pow_op<Derived::Scalar,ScalarExponent>,Derived,Constant<ScalarExponent> >
-  pow(const Eigen::ArrayBase<Derived>& x, const ScalarExponent& exponent);
-#else
-  template<typename Derived,typename ScalarExponent>
-  inline typename internal::enable_if<   !(internal::is_same<typename Derived::Scalar,ScalarExponent>::value) && EIGEN_SCALAR_BINARY_SUPPORTED(pow,typename Derived::Scalar,ScalarExponent),
-          const EIGEN_EXPR_BINARYOP_SCALAR_RETURN_TYPE(Derived,ScalarExponent,pow) >::type
-  pow(const Eigen::ArrayBase<Derived>& x, const ScalarExponent& exponent) {
+  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);
   }
 
   template<typename Derived>
-  inline const EIGEN_EXPR_BINARYOP_SCALAR_RETURN_TYPE(Derived,typename Derived::Scalar,pow)
-  pow(const Eigen::ArrayBase<Derived>& x, const typename Derived::Scalar& exponent) {
-    return x.derived().pow(exponent);
-  }
-#endif
-
-  /** \returns an expression of the coefficient-wise power of \a x to the given array of \a exponents.
-    *
-    * This function computes the coefficient-wise power.
-    *
-    * Example: \include Cwise_array_power_array.cpp
-    * Output: \verbinclude Cwise_array_power_array.out
-    * 
-    * \sa ArrayBase::pow()
-    *
-    * \relates ArrayBase
-    */
-  template<typename Derived,typename ExponentDerived>
-  inline const Eigen::CwiseBinaryOp<Eigen::internal::scalar_pow_op<typename Derived::Scalar, typename ExponentDerived::Scalar>, const Derived, const ExponentDerived>
-  pow(const Eigen::ArrayBase<Derived>& x, const Eigen::ArrayBase<ExponentDerived>& exponents) 
+  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>
-  inline typename internal::enable_if<   !(internal::is_same<typename Derived::Scalar,Scalar>::value) && EIGEN_SCALAR_BINARY_SUPPORTED(pow,Scalar,typename Derived::Scalar),
-          const EIGEN_SCALAR_BINARYOP_EXPR_RETURN_TYPE(Scalar,Derived,pow) >::type
-  pow(const Scalar& x, const Eigen::ArrayBase<Derived>& exponents)
+  /**
+  * \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_SCALAR_BINARYOP_EXPR_RETURN_TYPE(Scalar,Derived,pow)(
-            typename internal::plain_constant_type<Derived,Scalar>::type(exponents.rows(), exponents.cols(), x), exponents.derived() );
+    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)  
+    );
   }
 
-  template<typename Derived>
-  inline const EIGEN_SCALAR_BINARYOP_EXPR_RETURN_TYPE(typename Derived::Scalar,Derived,pow)
-  pow(const typename Derived::Scalar& x, const Eigen::ArrayBase<Derived>& exponents)
-  {
-    return EIGEN_SCALAR_BINARYOP_EXPR_RETURN_TYPE(typename Derived::Scalar,Derived,pow)(
-      typename internal::plain_constant_type<Derived,typename Derived::Scalar>::type(exponents.rows(), exponents.cols(), x), exponents.derived() );
-  }
-#endif
-
-
   namespace internal
   {
     EIGEN_ARRAY_DECLARE_GLOBAL_EIGEN_UNARY(real,scalar_real_op)

Eigen/src/Core/IO.h

@@ -49,7 +49,7 @@ 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="",
@@ -57,10 +57,6 @@ struct IOFormat
   : matPrefix(_matPrefix), matSuffix(_matSuffix), rowPrefix(_rowPrefix), rowSuffix(_rowSuffix), rowSeparator(_rowSeparator),
     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')
     {
@@ -80,7 +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()
@@ -125,18 +121,32 @@ DenseBase<Derived>::format(const IOFormat& fmt) const
 
 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>
@@ -150,6 +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 Derived::Index Index;
 
   Index width = 0;
 

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 Gael Guennebaud <gael.guennebaud@inria.fr>
-//
-// This Source Code Form is subject to the terms of the Mozilla
-// Public License v. 2.0. If a copy of the MPL was not distributed
-// with this file, You can obtain one at http://mozilla.org/MPL/2.0/.
-
-#ifndef EIGEN_INVERSE_H
-#define EIGEN_INVERSE_H
-
-namespace Eigen { 
-
-template<typename XprType,typename StorageKind> class InverseImpl;
-
-namespace internal {
-
-template<typename XprType>
-struct traits<Inverse<XprType> >
-  : traits<typename XprType::PlainObject>
-{
-  typedef typename XprType::PlainObject PlainObject;
-  typedef traits<PlainObject> BaseTraits;
-  enum {
-    Flags = BaseTraits::Flags & RowMajorBit
-  };
-};
-
-} // end namespace internal
-
-/** \class Inverse
-  *
-  * \brief Expression of the inverse of another expression
-  *
-  * \tparam XprType the type of the expression we are taking the inverse
-  *
-  * This class represents an abstract expression of A.inverse()
-  * and most of the time this is the only way it is used.
-  *
-  */
-template<typename XprType>
-class Inverse : public InverseImpl<XprType,typename internal::traits<XprType>::StorageKind>
-{
-public:
-  typedef typename XprType::StorageIndex StorageIndex;
-  typedef typename XprType::PlainObject                       PlainObject;
-  typedef typename internal::ref_selector<XprType>::type      XprTypeNested;
-  typedef typename internal::remove_all<XprTypeNested>::type  XprTypeNestedCleaned;
-  typedef typename internal::ref_selector<Inverse>::type Nested;
-  typedef typename internal::remove_all<XprType>::type NestedExpression;
-  
-  explicit EIGEN_DEVICE_FUNC Inverse(const XprType &xpr)
-    : m_xpr(xpr)
-  {}
-
-  EIGEN_DEVICE_FUNC Index rows() const { return m_xpr.rows(); }
-  EIGEN_DEVICE_FUNC Index cols() const { return m_xpr.cols(); }
-
-  EIGEN_DEVICE_FUNC const XprTypeNestedCleaned& nestedExpression() const { return m_xpr; }
-
-protected:
-  XprTypeNested m_xpr;
-};
-
-// Generic API dispatcher
-template<typename XprType, typename StorageKind>
-class InverseImpl
-  : public internal::generic_xpr_base<Inverse<XprType> >::type
-{
-public:
-  typedef typename internal::generic_xpr_base<Inverse<XprType> >::type Base;
-  typedef typename XprType::Scalar Scalar;
-private:
-
-  Scalar coeff(Index row, Index col) const;
-  Scalar coeff(Index i) const;
-};
-
-namespace internal {
-
-/** \internal
-  * \brief Default evaluator for Inverse expression.
-  * 
-  * This default evaluator for Inverse expression simply evaluate the inverse into a temporary
-  * by a call to internal::call_assignment_no_alias.
-  * Therefore, inverse implementers only have to specialize Assignment<Dst,Inverse<...>, ...> for
-  * there own nested expression.
-  *
-  * \sa class Inverse
-  */
-template<typename ArgType>
-struct unary_evaluator<Inverse<ArgType> >
-  : public evaluator<typename Inverse<ArgType>::PlainObject>
-{
-  typedef Inverse<ArgType> InverseType;
-  typedef typename InverseType::PlainObject PlainObject;
-  typedef evaluator<PlainObject> Base;
-  
-  enum { Flags = Base::Flags | EvalBeforeNestingBit };
-
-  unary_evaluator(const InverseType& inv_xpr)
-    : m_result(inv_xpr.rows(), inv_xpr.cols())
-  {
-    ::new (static_cast<Base*>(this)) Base(m_result);
-    internal::call_assignment_no_alias(m_result, inv_xpr);
-  }
-  
-protected:
-  PlainObject m_result;
-};
-  
-} // end namespace internal
-
-} // end namespace Eigen
-
-#endif // EIGEN_INVERSE_H

Eigen/src/Core/Map.h

@@ -13,35 +13,13 @@
 
 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 {
-    InnerStrideAtCompileTime = StrideType::InnerStrideAtCompileTime == 0
-                             ? int(PlainObjectType::InnerStrideAtCompileTime)
-                             : int(StrideType::InnerStrideAtCompileTime),
-    OuterStrideAtCompileTime = StrideType::OuterStrideAtCompileTime == 0
-                             ? int(PlainObjectType::OuterStrideAtCompileTime)
-                             : 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
-};
-}
-
 /** \class Map
   * \ingroup Core_Module
   *
   * \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.
@@ -85,6 +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> >
 {
@@ -94,17 +110,19 @@ 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
     inline Index innerStride() const
     {
       return StrideType::InnerStrideAtCompileTime != 0 ? m_stride.inner() : 1;
     }
 
-    EIGEN_DEVICE_FUNC
     inline Index outerStride() const
     {
       return StrideType::OuterStrideAtCompileTime != 0 ? m_stride.outer()
@@ -116,11 +134,10 @@ template<typename PlainObjectType, int MapOptions, typename StrideType> class Ma
     /** 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();
     }
@@ -128,12 +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();
     }
@@ -141,13 +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();
     }
@@ -158,6 +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 { 
 
-/** \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
   */
@@ -47,6 +37,7 @@ template<typename Derived> class MapBase<Derived, ReadOnlyAccessors>
     };
 
     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;
@@ -85,10 +76,8 @@ template<typename Derived> class MapBase<Derived, ReadOnlyAccessors>
 
     typedef typename Base::CoeffReturnType CoeffReturnType;
 
-    /** \copydoc DenseBase::rows() */
-    EIGEN_DEVICE_FUNC inline Index rows() const { return m_rows.value(); }
-    /** \copydoc DenseBase::cols() */
-    EIGEN_DEVICE_FUNC inline Index cols() const { 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.
       *
@@ -96,39 +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
     {
@@ -136,7 +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
     {
@@ -144,16 +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)
     {
       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)),
@@ -162,18 +137,16 @@ 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
@@ -182,36 +155,20 @@ template<typename Derived> class MapBase<Derived, ReadOnlyAccessors>
 
   protected:
 
-    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
-      eigen_assert((   ((internal::UIntPtr(m_data) % internal::traits<Derived>::Alignment) == 0)
-                    || (cols() * rows() * innerStride() * 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)
+                   && "input pointer is not aligned on a 16 byte boundary");
     }
 
-    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>
 {
@@ -222,7 +179,7 @@ template<typename Derived> class MapBase<Derived, WriteAccessors>
 
     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;
@@ -243,18 +200,14 @@ 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)
@@ -276,11 +229,10 @@ 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);

Eigen/src/Core/MathFunctionsImpl.h

@@ -1,78 +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 ulp in the range [-9, 9], outside of which
-    the tanh(x) = +/-1.
-
-    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 [-9, 9] since anything outside
-  // this range is +/-1.0f in single-precision.
-  const T plus_9 = pset1<T>(9.f);
-  const T minus_9 = pset1<T>(-9.f);
-  // NOTE GCC prior to 6.3 might improperly optimize this max/min
-  //      step such that if a_x is nan, x will be either 9 or -9,
-  //      and tanh will return 1 or -1 instead of nan.
-  //      This is supposed to be fixed in gcc6.3,
-  //      see: https://gcc.gnu.org/bugzilla/show_bug.cgi?id=72867
-  const T x = pmax(minus_9,pmin(plus_9,a_x));
-  // 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 p.
-  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 pdiv(p, q);
-}
-
-} // 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);
@@ -255,7 +200,6 @@ class Matrix
       *
       * \sa resize(Index,Index)
       */
-    EIGEN_DEVICE_FUNC
     EIGEN_STRONG_INLINE Matrix() : Base()
     {
       Base::_check_template_params();
@@ -263,87 +207,60 @@ class Matrix
     }
 
     // FIXME is it still needed
-    EIGEN_DEVICE_FUNC
-    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
-    Matrix(Matrix&& other) EIGEN_NOEXCEPT_IF(std::is_nothrow_move_constructible<Scalar>::value)
+#ifdef EIGEN_HAVE_RVALUE_REFERENCES
+    Matrix(Matrix&& other)
       : Base(std::move(other))
     {
       Base::_check_template_params();
       if (RowsAtCompileTime!=Dynamic && ColsAtCompileTime!=Dynamic)
         Base::_set_noalias(other);
     }
-    EIGEN_DEVICE_FUNC
-    Matrix& operator=(Matrix&& other) EIGEN_NOEXCEPT_IF(std::is_nothrow_move_assignable<Scalar>::value)
+    Matrix& operator=(Matrix&& other)
     {
       other.swap(*this);
       return *this;
     }
 #endif
 
-    #ifndef EIGEN_PARSED_BY_DOXYGEN
-
-    // This constructor is for both 1x1 matrices and dynamic vectors
-    template<typename T>
-    EIGEN_DEVICE_FUNC
-    EIGEN_STRONG_INLINE explicit Matrix(const T& x)
+    /** \brief Constructs a vector or row-vector with given dimension. \only_for_vectors
+      *
+      * Note that this is only useful for dynamic-size vectors. For fixed-size vectors,
+      * it is redundant to pass the dimension here, so it makes more sense to use the default
+      * constructor Matrix() instead.
+      */
+    EIGEN_STRONG_INLINE explicit Matrix(Index dim)
+      : Base(dim, RowsAtCompileTime == 1 ? 1 : dim, ColsAtCompileTime == 1 ? 1 : dim)
     {
       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)
     {
       Base::_check_template_params();
       Base::template _init2<T0,T1>(x, y);
     }
     #else
-    /** \brief Constructs a fixed-sized matrix initialized with coefficients starting at \a data */
-    EIGEN_DEVICE_FUNC
-    explicit Matrix(const Scalar *data);
-
-    /** \brief Constructs a vector or row-vector with given dimension. \only_for_vectors
-      *
-      * This is useful for dynamic-size vectors. For fixed-size vectors,
-      * it is redundant to pass these parameters, so one should use the default constructor
-      * Matrix() instead.
-      * 
-      * \warning This constructor is disabled for fixed-size \c 1x1 matrices. For instance,
-      * calling Matrix<double,1,1>(1) will call the initialization constructor: Matrix(const Scalar&).
-      * For fixed-size \c 1x1 matrices it is therefore recommended to use the default
-      * constructor Matrix() instead, especially when using one of the non standard
-      * \c EIGEN_INITIALIZE_MATRICES_BY_{ZERO,\c NAN} macros (see \ref TopicPreprocessorDirectives).
-      */
-    EIGEN_STRONG_INLINE explicit Matrix(Index dim);
-    /** \brief Constructs an initialized 1x1 matrix with the given coefficient */
-    Matrix(const Scalar& x);
     /** \brief Constructs an uninitialized matrix with \a rows rows and \a cols columns.
       *
       * This is useful for dynamic-size matrices. For fixed-size matrices,
       * it is redundant to pass these parameters, so one should use the default constructor
-      * Matrix() instead.
-      * 
-      * \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 */
     Matrix(const Scalar& x, const Scalar& y);
     #endif
 
     /** \brief Constructs an initialized 3D vector with given coefficients */
-    EIGEN_DEVICE_FUNC
     EIGEN_STRONG_INLINE Matrix(const Scalar& x, const Scalar& y, const Scalar& z)
     {
       Base::_check_template_params();
@@ -353,7 +270,6 @@ class Matrix
       m_storage.data()[2] = z;
     }
     /** \brief Constructs an initialized 4D vector with given coefficients */
-    EIGEN_DEVICE_FUNC
     EIGEN_STRONG_INLINE Matrix(const Scalar& x, const Scalar& y, const Scalar& z, const Scalar& w)
     {
       Base::_check_template_params();
@@ -364,33 +280,76 @@ 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_to_match(other);
+      // FIXME/CHECK: isn't *this = other.derived() more efficient. it allows to
+      //              go for pure _set() implementations, right?
+      *this = other;
+    }
 
-    EIGEN_DEVICE_FUNC inline Index innerStride() const { return 1; }
-    EIGEN_DEVICE_FUNC inline Index outerStride() const { 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

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;
@@ -97,14 +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>,
@@ -113,7 +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,
@@ -121,7 +133,6 @@ 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"
@@ -130,53 +141,41 @@ 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);
 
+    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);
+
     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);
 
-#ifdef __CUDACC__
     template<typename OtherDerived>
-    EIGEN_DEVICE_FUNC
-    const Product<Derived,OtherDerived,LazyProduct>
-    operator*(const MatrixBase<OtherDerived> &other) const
-    { return this->lazyProduct(other); }
-#else
-
-    template<typename OtherDerived>
-    const Product<Derived,OtherDerived>
+    const typename ProductReturnType<Derived,OtherDerived>::Type
     operator*(const MatrixBase<OtherDerived> &other) const;
 
-#endif
-
     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>
@@ -189,93 +188,84 @@ 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
     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>
-    EIGEN_DEVICE_FUNC
-    typename ConstDiagonalIndexReturnType<Index>::Type diagonal() const;
-
+    template<int Index> typename DiagonalIndexReturnType<Index>::Type diagonal();
+    template<int Index> typename ConstDiagonalIndexReturnType<Index>::Type diagonal() const;
+    
     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);
 
     bool isIdentity(const RealScalar& prec = NumTraits<Scalar>::dummy_precision()) const;
@@ -307,45 +297,59 @@ template<typename Derived> class MatrixBase
 
     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
 
-    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()
@@ -354,70 +358,65 @@ template<typename Derived> class MatrixBase
 
 /////////// 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 ///////////
 
@@ -462,15 +461,49 @@ template<typename Derived> class MatrixBase
     const MatrixSquareRootReturnValue<Derived> sqrt() const;
     const MatrixLogarithmReturnValue<Derived> log() const;
     const MatrixPowerReturnValue<Derived> pow(const RealScalar& p) const;
-    const MatrixComplexPowerReturnValue<Derived> pow(const std::complex<RealScalar>& p) const;
+
+#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
 
   protected:
-    EIGEN_DEVICE_FUNC MatrixBase() : Base() {}
+    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>& )

Eigen/src/Core/NestByValue.h

@@ -13,24 +13,25 @@
 
 namespace Eigen {
 
-namespace internal {
-template<typename ExpressionType>
-struct traits<NestByValue<ExpressionType> > : public traits<ExpressionType>
-{};
-}
-
 /** \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
 {
@@ -39,29 +40,29 @@ 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 inline Index rows() const { return m_expression.rows(); }
-    EIGEN_DEVICE_FUNC inline Index cols() const { return m_expression.cols(); }
-    EIGEN_DEVICE_FUNC inline Index outerStride() const { return m_expression.outerStride(); }
-    EIGEN_DEVICE_FUNC inline Index innerStride() const { return m_expression.innerStride(); }
+    inline 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);
     }
@@ -90,7 +91,7 @@ template<typename ExpressionType> class NestByValue
       m_expression.const_cast_derived().template writePacket<LoadMode>(index, x);
     }
 
-    EIGEN_DEVICE_FUNC operator const ExpressionType&() const { return m_expression; }
+    operator const ExpressionType&() const { return m_expression; }
 
   protected:
     const ExpressionType m_expression;

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.
@@ -30,36 +30,62 @@ namespace Eigen {
 template<typename ExpressionType, template <typename> class StorageBase>
 class NoAlias
 {
-  public:
     typedef typename ExpressionType::Scalar Scalar;
-    
-    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;
@@ -100,7 +126,7 @@ class NoAlias
 template<typename Derived>
 NoAlias<Derived,MatrixBase> MatrixBase<Derived>::noalias()
 {
-  return NoAlias<Derived, Eigen::MatrixBase >(derived());
+  return derived();
 }
 
 } // end namespace Eigen

Eigen/src/Core/NumTraits.h

@@ -12,57 +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
-{
-  static int run() { return std::numeric_limits<T>::digits10; }
-};
-
-template<typename T>
-struct default_digits10_impl<T,false,false> // Floating point
-{
-  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
-{
-  static int run() { return 0; }
-};
-
-} // end namespace internal
-
 /** \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
@@ -75,14 +42,10 @@ struct default_digits10_impl<T,false,true> // Integer
   * \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 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.
   */
 
 template<typename T> struct GenericNumTraits
@@ -104,47 +67,22 @@ template<typename T> struct GenericNumTraits
                      T
                    >::type NonInteger;
   typedef T Nested;
-  typedef T Literal;
 
-  EIGEN_DEVICE_FUNC
-  static inline Real epsilon()
-  {
-    return numext::numeric_limits<T>::epsilon();
-  }
-
-  EIGEN_DEVICE_FUNC
-  static inline int digits10()
-  {
-    return internal::default_digits10_impl<T>::run();
-  }
-
-  EIGEN_DEVICE_FUNC
+  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
-  static inline T highest() {
-    return (numext::numeric_limits<T>::max)();
-  }
-
-  EIGEN_DEVICE_FUNC
-  static inline T lowest()  {
-    return IsInteger ? (numext::numeric_limits<T>::min)() : (-(numext::numeric_limits<T>::max)());
-  }
-
-  EIGEN_DEVICE_FUNC
-  static inline T infinity() {
-    return numext::numeric_limits<T>::infinity();
-  }
-
-  EIGEN_DEVICE_FUNC
-  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>
@@ -153,13 +91,11 @@ template<typename T> struct NumTraits : GenericNumTraits<T>
 template<> struct NumTraits<float>
   : GenericNumTraits<float>
 {
-  EIGEN_DEVICE_FUNC
   static inline float dummy_precision() { return 1e-5f; }
 };
 
 template<> struct NumTraits<double> : GenericNumTraits<double>
 {
-  EIGEN_DEVICE_FUNC
   static inline double dummy_precision() { return 1e-12; }
 };
 
@@ -173,7 +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,
@@ -182,12 +117,8 @@ template<typename _Real> struct NumTraits<std::complex<_Real> >
     MulCost = 4 * NumTraits<Real>::MulCost + 2 * NumTraits<Real>::AddCost
   };
 
-  EIGEN_DEVICE_FUNC
   static inline Real epsilon() { return NumTraits<Real>::epsilon(); }
-  EIGEN_DEVICE_FUNC
   static inline Real dummy_precision() { return NumTraits<Real>::dummy_precision(); }
-  EIGEN_DEVICE_FUNC
-  static inline int digits10() { return NumTraits<Real>::digits10(); }
 };
 
 template<typename Scalar, int Rows, int Cols, int Options, int MaxRows, int MaxCols>
@@ -199,48 +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
+  
   static inline RealScalar epsilon() { return NumTraits<RealScalar>::epsilon(); }
-  EIGEN_DEVICE_FUNC
   static inline RealScalar dummy_precision() { return NumTraits<RealScalar>::dummy_precision(); }
 };
 
-template<> struct NumTraits<std::string>
-  : GenericNumTraits<std::string>
-{
-  enum {
-    RequireInitialization = 1,
-    ReadCost = HugeCost,
-    AddCost  = HugeCost,
-    MulCost  = HugeCost
-  };
-
-  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> {};
-
 } // end namespace Eigen
 
 #endif // EIGEN_NUMTRAITS_H

Eigen/src/Core/PermutationMatrix.h

@@ -2,7 +2,7 @@
 // for linear algebra.
 //
 // Copyright (C) 2009 Benoit Jacob <jacob.benoit.1@gmail.com>
-// Copyright (C) 2009-2015 Gael Guennebaud <gael.guennebaud@inria.fr>
+// Copyright (C) 2009-2011 Gael Guennebaud <gael.guennebaud@inria.fr>
 //
 // This Source Code Form is subject to the terms of the Mozilla
 // Public License v. 2.0. If a copy of the MPL was not distributed
@@ -13,18 +13,14 @@
 
 namespace Eigen { 
 
-namespace internal {
-
-enum PermPermProduct_t {PermPermProduct};
-
-} // end namespace internal
+template<int RowCol,typename IndicesType,typename MatrixType, typename StorageKind> class PermutedImpl;
 
 /** \class PermutationBase
   * \ingroup Core_Module
   *
   * \brief Base class for permutations
   *
-  * \tparam Derived the derived class
+  * \param Derived the derived class
   *
   * This class is the base class for all expressions representing a permutation matrix,
   * internally stored as a vector of integers.
@@ -42,6 +38,17 @@ enum PermPermProduct_t {PermPermProduct};
   *
   * \sa class PermutationMatrix, class PermutationWrapper
   */
+
+namespace internal {
+
+template<typename PermutationType, typename MatrixType, int Side, bool Transposed=false>
+struct permut_matrix_product_retval;
+template<typename PermutationType, typename MatrixType, int Side, bool Transposed=false>
+struct permut_sparsematrix_product_retval;
+enum PermPermProduct_t {PermPermProduct};
+
+} // end namespace internal
+
 template<typename Derived>
 class PermutationBase : public EigenBase<Derived>
 {
@@ -53,20 +60,19 @@ class PermutationBase : public EigenBase<Derived>
     typedef typename Traits::IndicesType IndicesType;
     enum {
       Flags = Traits::Flags,
+      CoeffReadCost = Traits::CoeffReadCost,
       RowsAtCompileTime = Traits::RowsAtCompileTime,
       ColsAtCompileTime = Traits::ColsAtCompileTime,
       MaxRowsAtCompileTime = Traits::MaxRowsAtCompileTime,
       MaxColsAtCompileTime = Traits::MaxColsAtCompileTime
     };
-    typedef typename Traits::StorageIndex StorageIndex;
-    typedef Matrix<StorageIndex,RowsAtCompileTime,ColsAtCompileTime,0,MaxRowsAtCompileTime,MaxColsAtCompileTime>
+    typedef typename Traits::Scalar Scalar;
+    typedef typename Traits::Index Index;
+    typedef Matrix<Scalar,RowsAtCompileTime,ColsAtCompileTime,0,MaxRowsAtCompileTime,MaxColsAtCompileTime>
             DenseMatrixType;
-    typedef PermutationMatrix<IndicesType::SizeAtCompileTime,IndicesType::MaxSizeAtCompileTime,StorageIndex>
+    typedef PermutationMatrix<IndicesType::SizeAtCompileTime,IndicesType::MaxSizeAtCompileTime,Index>
             PlainPermutationType;
-    typedef PlainPermutationType PlainObject;
     using Base::derived;
-    typedef Inverse<Derived> InverseReturnType;
-    typedef void Scalar;
     #endif
 
     /** Copies the other permutation into *this */
@@ -112,7 +118,7 @@ class PermutationBase : public EigenBase<Derived>
     void evalTo(MatrixBase<DenseDerived>& other) const
     {
       other.setZero();
-      for (Index i=0; i<rows(); ++i)
+      for (int i=0; i<rows();++i)
         other.coeffRef(indices().coeff(i),i) = typename DenseDerived::Scalar(1);
     }
     #endif
@@ -141,8 +147,7 @@ class PermutationBase : public EigenBase<Derived>
     /** Sets *this to be the identity permutation matrix */
     void setIdentity()
     {
-      StorageIndex n = StorageIndex(size());
-      for(StorageIndex i = 0; i < n; ++i)
+      for(Index i = 0; i < size(); ++i)
         indices().coeffRef(i) = i;
     }
 
@@ -158,18 +163,18 @@ class PermutationBase : public EigenBase<Derived>
       *
       * \returns a reference to *this.
       *
-      * \warning This is much slower than applyTranspositionOnTheRight(Index,Index):
+      * \warning This is much slower than applyTranspositionOnTheRight(int,int):
       * this has linear complexity and requires a lot of branching.
       *
-      * \sa applyTranspositionOnTheRight(Index,Index)
+      * \sa applyTranspositionOnTheRight(int,int)
       */
     Derived& applyTranspositionOnTheLeft(Index i, Index j)
     {
       eigen_assert(i>=0 && j>=0 && i<size() && j<size());
       for(Index k = 0; k < size(); ++k)
       {
-        if(indices().coeff(k) == i) indices().coeffRef(k) = StorageIndex(j);
-        else if(indices().coeff(k) == j) indices().coeffRef(k) = StorageIndex(i);
+        if(indices().coeff(k) == i) indices().coeffRef(k) = j;
+        else if(indices().coeff(k) == j) indices().coeffRef(k) = i;
       }
       return derived();
     }
@@ -180,7 +185,7 @@ class PermutationBase : public EigenBase<Derived>
       *
       * This is a fast operation, it only consists in swapping two indices.
       *
-      * \sa applyTranspositionOnTheLeft(Index,Index)
+      * \sa applyTranspositionOnTheLeft(int,int)
       */
     Derived& applyTranspositionOnTheRight(Index i, Index j)
     {
@@ -191,16 +196,16 @@ class PermutationBase : public EigenBase<Derived>
 
     /** \returns the inverse permutation matrix.
       *
-      * \note \blank \note_try_to_help_rvo
+      * \note \note_try_to_help_rvo
       */
-    inline InverseReturnType inverse() const
-    { return InverseReturnType(derived()); }
+    inline Transpose<PermutationBase> inverse() const
+    { return derived(); }
     /** \returns the tranpose permutation matrix.
       *
-      * \note \blank \note_try_to_help_rvo
+      * \note \note_try_to_help_rvo
       */
-    inline InverseReturnType transpose() const
-    { return InverseReturnType(derived()); }
+    inline Transpose<PermutationBase> transpose() const
+    { return derived(); }
 
     /**** multiplication helpers to hopefully get RVO ****/
 
@@ -210,13 +215,13 @@ class PermutationBase : public EigenBase<Derived>
     template<typename OtherDerived>
     void assignTranspose(const PermutationBase<OtherDerived>& other)
     {
-      for (Index i=0; i<rows();++i) indices().coeffRef(other.indices().coeff(i)) = i;
+      for (int i=0; i<rows();++i) indices().coeffRef(other.indices().coeff(i)) = i;
     }
     template<typename Lhs,typename Rhs>
     void assignProduct(const Lhs& lhs, const Rhs& rhs)
     {
       eigen_assert(lhs.cols() == rhs.rows());
-      for (Index i=0; i<rows();++i) indices().coeffRef(i) = lhs.indices().coeff(rhs.indices().coeff(i));
+      for (int i=0; i<rows();++i) indices().coeffRef(i) = lhs.indices().coeff(rhs.indices().coeff(i));
     }
 #endif
 
@@ -224,7 +229,7 @@ class PermutationBase : public EigenBase<Derived>
 
     /** \returns the product permutation matrix.
       *
-      * \note \blank \note_try_to_help_rvo
+      * \note \note_try_to_help_rvo
       */
     template<typename Other>
     inline PlainPermutationType operator*(const PermutationBase<Other>& other) const
@@ -232,18 +237,18 @@ class PermutationBase : public EigenBase<Derived>
 
     /** \returns the product of a permutation with another inverse permutation.
       *
-      * \note \blank \note_try_to_help_rvo
+      * \note \note_try_to_help_rvo
       */
     template<typename Other>
-    inline PlainPermutationType operator*(const InverseImpl<Other,PermutationStorage>& other) const
+    inline PlainPermutationType operator*(const Transpose<PermutationBase<Other> >& other) const
     { return PlainPermutationType(internal::PermPermProduct, *this, other.eval()); }
 
     /** \returns the product of an inverse permutation with another permutation.
       *
-      * \note \blank \note_try_to_help_rvo
+      * \note \note_try_to_help_rvo
       */
     template<typename Other> friend
-    inline PlainPermutationType operator*(const InverseImpl<Other, PermutationStorage>& other, const PermutationBase& perm)
+    inline PlainPermutationType operator*(const Transpose<PermutationBase<Other> >& other, const PermutationBase& perm)
     { return PlainPermutationType(internal::PermPermProduct, other.eval(), perm); }
     
     /** \returns the determinant of the permutation matrix, which is either 1 or -1 depending on the parity of the permutation.
@@ -279,43 +284,39 @@ class PermutationBase : public EigenBase<Derived>
 
 };
 
-namespace internal {
-template<int SizeAtCompileTime, int MaxSizeAtCompileTime, typename _StorageIndex>
-struct traits<PermutationMatrix<SizeAtCompileTime, MaxSizeAtCompileTime, _StorageIndex> >
- : traits<Matrix<_StorageIndex,SizeAtCompileTime,SizeAtCompileTime,0,MaxSizeAtCompileTime,MaxSizeAtCompileTime> >
-{
-  typedef PermutationStorage StorageKind;
-  typedef Matrix<_StorageIndex, SizeAtCompileTime, 1, 0, MaxSizeAtCompileTime, 1> IndicesType;
-  typedef _StorageIndex StorageIndex;
-  typedef void Scalar;
-};
-}
-
 /** \class PermutationMatrix
   * \ingroup Core_Module
   *
   * \brief Permutation matrix
   *
-  * \tparam SizeAtCompileTime the number of rows/cols, or Dynamic
-  * \tparam MaxSizeAtCompileTime the maximum number of rows/cols, or Dynamic. This optional parameter defaults to SizeAtCompileTime. Most of the time, you should not have to specify it.
-  * \tparam _StorageIndex the integer type of the indices
+  * \param SizeAtCompileTime the number of rows/cols, or Dynamic
+  * \param MaxSizeAtCompileTime the maximum number of rows/cols, or Dynamic. This optional parameter defaults to SizeAtCompileTime. Most of the time, you should not have to specify it.
+  * \param IndexType the interger type of the indices
   *
   * This class represents a permutation matrix, internally stored as a vector of integers.
   *
   * \sa class PermutationBase, class PermutationWrapper, class DiagonalMatrix
   */
-template<int SizeAtCompileTime, int MaxSizeAtCompileTime, typename _StorageIndex>
-class PermutationMatrix : public PermutationBase<PermutationMatrix<SizeAtCompileTime, MaxSizeAtCompileTime, _StorageIndex> >
+
+namespace internal {
+template<int SizeAtCompileTime, int MaxSizeAtCompileTime, typename IndexType>
+struct traits<PermutationMatrix<SizeAtCompileTime, MaxSizeAtCompileTime, IndexType> >
+ : traits<Matrix<IndexType,SizeAtCompileTime,SizeAtCompileTime,0,MaxSizeAtCompileTime,MaxSizeAtCompileTime> >
+{
+  typedef IndexType Index;
+  typedef Matrix<IndexType, SizeAtCompileTime, 1, 0, MaxSizeAtCompileTime, 1> IndicesType;
+};
+}
+
+template<int SizeAtCompileTime, int MaxSizeAtCompileTime, typename IndexType>
+class PermutationMatrix : public PermutationBase<PermutationMatrix<SizeAtCompileTime, MaxSizeAtCompileTime, IndexType> >
 {
     typedef PermutationBase<PermutationMatrix> Base;
     typedef internal::traits<PermutationMatrix> Traits;
   public:
 
-    typedef const PermutationMatrix& Nested;
-
     #ifndef EIGEN_PARSED_BY_DOXYGEN
     typedef typename Traits::IndicesType IndicesType;
-    typedef typename Traits::StorageIndex StorageIndex;
     #endif
 
     inline PermutationMatrix()
@@ -323,10 +324,8 @@ class PermutationMatrix : public PermutationBase<PermutationMatrix<SizeAtCompile
 
     /** Constructs an uninitialized permutation matrix of given size.
       */
-    explicit inline PermutationMatrix(Index size) : m_indices(size)
-    {
-      eigen_internal_assert(size <= NumTraits<StorageIndex>::highest());
-    }
+    inline PermutationMatrix(int size) : m_indices(size)
+    {}
 
     /** Copy constructor. */
     template<typename OtherDerived>
@@ -347,7 +346,7 @@ class PermutationMatrix : public PermutationBase<PermutationMatrix<SizeAtCompile
       * array's size.
       */
     template<typename Other>
-    explicit inline PermutationMatrix(const MatrixBase<Other>& indices) : m_indices(indices)
+    explicit inline PermutationMatrix(const MatrixBase<Other>& a_indices) : m_indices(a_indices)
     {}
 
     /** Convert the Transpositions \a tr to a permutation matrix */
@@ -394,13 +393,10 @@ class PermutationMatrix : public PermutationBase<PermutationMatrix<SizeAtCompile
 
 #ifndef EIGEN_PARSED_BY_DOXYGEN
     template<typename Other>
-    PermutationMatrix(const InverseImpl<Other,PermutationStorage>& other)
-      : m_indices(other.derived().nestedExpression().size())
+    PermutationMatrix(const Transpose<PermutationBase<Other> >& other)
+      : m_indices(other.nestedPermutation().size())
     {
-      eigen_internal_assert(m_indices.size() <= NumTraits<StorageIndex>::highest());
-      StorageIndex end = StorageIndex(m_indices.size());
-      for (StorageIndex i=0; i<end;++i)
-        m_indices.coeffRef(other.derived().nestedExpression().indices().coeff(i)) = i;
+      for (int i=0; i<m_indices.size();++i) m_indices.coeffRef(other.nestedPermutation().indices().coeff(i)) = i;
     }
     template<typename Lhs,typename Rhs>
     PermutationMatrix(internal::PermPermProduct_t, const Lhs& lhs, const Rhs& rhs)
@@ -417,20 +413,18 @@ class PermutationMatrix : public PermutationBase<PermutationMatrix<SizeAtCompile
 
 
 namespace internal {
-template<int SizeAtCompileTime, int MaxSizeAtCompileTime, typename _StorageIndex, int _PacketAccess>
-struct traits<Map<PermutationMatrix<SizeAtCompileTime, MaxSizeAtCompileTime, _StorageIndex>,_PacketAccess> >
- : traits<Matrix<_StorageIndex,SizeAtCompileTime,SizeAtCompileTime,0,MaxSizeAtCompileTime,MaxSizeAtCompileTime> >
+template<int SizeAtCompileTime, int MaxSizeAtCompileTime, typename IndexType, int _PacketAccess>
+struct traits<Map<PermutationMatrix<SizeAtCompileTime, MaxSizeAtCompileTime, IndexType>,_PacketAccess> >
+ : traits<Matrix<IndexType,SizeAtCompileTime,SizeAtCompileTime,0,MaxSizeAtCompileTime,MaxSizeAtCompileTime> >
 {
-  typedef PermutationStorage StorageKind;
-  typedef Map<const Matrix<_StorageIndex, SizeAtCompileTime, 1, 0, MaxSizeAtCompileTime, 1>, _PacketAccess> IndicesType;
-  typedef _StorageIndex StorageIndex;
-  typedef void Scalar;
+  typedef IndexType Index;
+  typedef Map<const Matrix<IndexType, SizeAtCompileTime, 1, 0, MaxSizeAtCompileTime, 1>, _PacketAccess> IndicesType;
 };
 }
 
-template<int SizeAtCompileTime, int MaxSizeAtCompileTime, typename _StorageIndex, int _PacketAccess>
-class Map<PermutationMatrix<SizeAtCompileTime, MaxSizeAtCompileTime, _StorageIndex>,_PacketAccess>
-  : public PermutationBase<Map<PermutationMatrix<SizeAtCompileTime, MaxSizeAtCompileTime, _StorageIndex>,_PacketAccess> >
+template<int SizeAtCompileTime, int MaxSizeAtCompileTime, typename IndexType, int _PacketAccess>
+class Map<PermutationMatrix<SizeAtCompileTime, MaxSizeAtCompileTime, IndexType>,_PacketAccess>
+  : public PermutationBase<Map<PermutationMatrix<SizeAtCompileTime, MaxSizeAtCompileTime, IndexType>,_PacketAccess> >
 {
     typedef PermutationBase<Map> Base;
     typedef internal::traits<Map> Traits;
@@ -438,14 +432,14 @@ class Map<PermutationMatrix<SizeAtCompileTime, MaxSizeAtCompileTime, _StorageInd
 
     #ifndef EIGEN_PARSED_BY_DOXYGEN
     typedef typename Traits::IndicesType IndicesType;
-    typedef typename IndicesType::Scalar StorageIndex;
+    typedef typename IndicesType::Scalar Index;
     #endif
 
-    inline Map(const StorageIndex* indicesPtr)
+    inline Map(const Index* indicesPtr)
       : m_indices(indicesPtr)
     {}
 
-    inline Map(const StorageIndex* indicesPtr, Index size)
+    inline Map(const Index* indicesPtr, Index size)
       : m_indices(indicesPtr,size)
     {}
 
@@ -480,36 +474,40 @@ class Map<PermutationMatrix<SizeAtCompileTime, MaxSizeAtCompileTime, _StorageInd
     IndicesType m_indices;
 };
 
+/** \class PermutationWrapper
+  * \ingroup Core_Module
+  *
+  * \brief Class to view a vector of integers as a permutation matrix
+  *
+  * \param _IndicesType the type of the vector of integer (can be any compatible expression)
+  *
+  * This class allows to view any vector expression of integers as a permutation matrix.
+  *
+  * \sa class PermutationBase, class PermutationMatrix
+  */
+
+struct PermutationStorage {};
+
 template<typename _IndicesType> class TranspositionsWrapper;
 namespace internal {
 template<typename _IndicesType>
 struct traits<PermutationWrapper<_IndicesType> >
 {
   typedef PermutationStorage StorageKind;
-  typedef void Scalar;
-  typedef typename _IndicesType::Scalar StorageIndex;
+  typedef typename _IndicesType::Scalar Scalar;
+  typedef typename _IndicesType::Scalar Index;
   typedef _IndicesType IndicesType;
   enum {
     RowsAtCompileTime = _IndicesType::SizeAtCompileTime,
     ColsAtCompileTime = _IndicesType::SizeAtCompileTime,
-    MaxRowsAtCompileTime = IndicesType::MaxSizeAtCompileTime,
-    MaxColsAtCompileTime = IndicesType::MaxSizeAtCompileTime,
-    Flags = 0
+    MaxRowsAtCompileTime = IndicesType::MaxRowsAtCompileTime,
+    MaxColsAtCompileTime = IndicesType::MaxColsAtCompileTime,
+    Flags = 0,
+    CoeffReadCost = _IndicesType::CoeffReadCost
   };
 };
 }
 
-/** \class PermutationWrapper
-  * \ingroup Core_Module
-  *
-  * \brief Class to view a vector of integers as a permutation matrix
-  *
-  * \tparam _IndicesType the type of the vector of integer (can be any compatible expression)
-  *
-  * This class allows to view any vector expression of integers as a permutation matrix.
-  *
-  * \sa class PermutationBase, class PermutationMatrix
-  */
 template<typename _IndicesType>
 class PermutationWrapper : public PermutationBase<PermutationWrapper<_IndicesType> >
 {
@@ -521,8 +519,8 @@ class PermutationWrapper : public PermutationBase<PermutationWrapper<_IndicesTyp
     typedef typename Traits::IndicesType IndicesType;
     #endif
 
-    inline PermutationWrapper(const IndicesType& indices)
-      : m_indices(indices)
+    inline PermutationWrapper(const IndicesType& a_indices)
+      : m_indices(a_indices)
     {}
 
     /** const version of indices(). */
@@ -534,86 +532,183 @@ class PermutationWrapper : public PermutationBase<PermutationWrapper<_IndicesTyp
     typename IndicesType::Nested m_indices;
 };
 
-
 /** \returns the matrix with the permutation applied to the columns.
   */
-template<typename MatrixDerived, typename PermutationDerived>
-EIGEN_DEVICE_FUNC
-const Product<MatrixDerived, PermutationDerived, AliasFreeProduct>
-operator*(const MatrixBase<MatrixDerived> &matrix,
-          const PermutationBase<PermutationDerived>& permutation)
+template<typename Derived, typename PermutationDerived>
+inline const internal::permut_matrix_product_retval<PermutationDerived, Derived, OnTheRight>
+operator*(const MatrixBase<Derived>& matrix,
+          const PermutationBase<PermutationDerived> &permutation)
 {
-  return Product<MatrixDerived, PermutationDerived, AliasFreeProduct>
-            (matrix.derived(), permutation.derived());
+  return internal::permut_matrix_product_retval
+           <PermutationDerived, Derived, OnTheRight>
+           (permutation.derived(), matrix.derived());
 }
 
 /** \returns the matrix with the permutation applied to the rows.
   */
-template<typename PermutationDerived, typename MatrixDerived>
-EIGEN_DEVICE_FUNC
-const Product<PermutationDerived, MatrixDerived, AliasFreeProduct>
+template<typename Derived, typename PermutationDerived>
+inline const internal::permut_matrix_product_retval
+               <PermutationDerived, Derived, OnTheLeft>
 operator*(const PermutationBase<PermutationDerived> &permutation,
-          const MatrixBase<MatrixDerived>& matrix)
+          const MatrixBase<Derived>& matrix)
 {
-  return Product<PermutationDerived, MatrixDerived, AliasFreeProduct>
-            (permutation.derived(), matrix.derived());
+  return internal::permut_matrix_product_retval
+           <PermutationDerived, Derived, OnTheLeft>
+           (permutation.derived(), matrix.derived());
 }
 
+namespace internal {
 
-template<typename PermutationType>
-class InverseImpl<PermutationType, PermutationStorage>
-  : public EigenBase<Inverse<PermutationType> >
+template<typename PermutationType, typename MatrixType, int Side, bool Transposed>
+struct traits<permut_matrix_product_retval<PermutationType, MatrixType, Side, Transposed> >
 {
-    typedef typename PermutationType::PlainPermutationType PlainPermutationType;
-    typedef internal::traits<PermutationType> PermTraits;
+  typedef typename MatrixType::PlainObject ReturnType;
+};
+
+template<typename PermutationType, typename MatrixType, int Side, bool Transposed>
+struct permut_matrix_product_retval
+ : public ReturnByValue<permut_matrix_product_retval<PermutationType, MatrixType, Side, Transposed> >
+{
+    typedef typename remove_all<typename MatrixType::Nested>::type MatrixTypeNestedCleaned;
+    typedef typename MatrixType::Index Index;
+
+    permut_matrix_product_retval(const PermutationType& perm, const MatrixType& matrix)
+      : m_permutation(perm), m_matrix(matrix)
+    {}
+
+    inline Index rows() const { return m_matrix.rows(); }
+    inline Index cols() const { return m_matrix.cols(); }
+
+    template<typename Dest> inline void evalTo(Dest& dst) const
+    {
+      const Index n = Side==OnTheLeft ? rows() : cols();
+      // FIXME we need an is_same for expression that is not sensitive to constness. For instance
+      // is_same_xpr<Block<const Matrix>, Block<Matrix> >::value should be true.
+      const typename Dest::Scalar *dst_data = internal::extract_data(dst);
+      if(    is_same<MatrixTypeNestedCleaned,Dest>::value
+          && blas_traits<MatrixTypeNestedCleaned>::HasUsableDirectAccess
+          && blas_traits<Dest>::HasUsableDirectAccess
+          && dst_data!=0 && dst_data == extract_data(m_matrix))
+      {
+        // apply the permutation inplace
+        Matrix<bool,PermutationType::RowsAtCompileTime,1,0,PermutationType::MaxRowsAtCompileTime> mask(m_permutation.size());
+        mask.fill(false);
+        Index r = 0;
+        while(r < m_permutation.size())
+        {
+          // search for the next seed
+          while(r<m_permutation.size() && mask[r]) r++;
+          if(r>=m_permutation.size())
+            break;
+          // we got one, let's follow it until we are back to the seed
+          Index k0 = r++;
+          Index kPrev = k0;
+          mask.coeffRef(k0) = true;
+          for(Index k=m_permutation.indices().coeff(k0); k!=k0; k=m_permutation.indices().coeff(k))
+          {
+                  Block<Dest, Side==OnTheLeft ? 1 : Dest::RowsAtCompileTime, Side==OnTheRight ? 1 : Dest::ColsAtCompileTime>(dst, k)
+            .swap(Block<Dest, Side==OnTheLeft ? 1 : Dest::RowsAtCompileTime, Side==OnTheRight ? 1 : Dest::ColsAtCompileTime>
+                       (dst,((Side==OnTheLeft) ^ Transposed) ? k0 : kPrev));
+
+            mask.coeffRef(k) = true;
+            kPrev = k;
+          }
+        }
+      }
+      else
+      {
+        for(int i = 0; i < n; ++i)
+        {
+          Block<Dest, Side==OnTheLeft ? 1 : Dest::RowsAtCompileTime, Side==OnTheRight ? 1 : Dest::ColsAtCompileTime>
+               (dst, ((Side==OnTheLeft) ^ Transposed) ? m_permutation.indices().coeff(i) : i)
+
+          =
+
+          Block<const MatrixTypeNestedCleaned,Side==OnTheLeft ? 1 : MatrixType::RowsAtCompileTime,Side==OnTheRight ? 1 : MatrixType::ColsAtCompileTime>
+               (m_matrix, ((Side==OnTheRight) ^ Transposed) ? m_permutation.indices().coeff(i) : i);
+        }
+      }
+    }
+
   protected:
-    InverseImpl() {}
+    const PermutationType& m_permutation;
+    typename MatrixType::Nested m_matrix;
+};
+
+/* Template partial specialization for transposed/inverse permutations */
+
+template<typename Derived>
+struct traits<Transpose<PermutationBase<Derived> > >
+ : traits<Derived>
+{};
+
+} // end namespace internal
+
+template<typename Derived>
+class Transpose<PermutationBase<Derived> >
+  : public EigenBase<Transpose<PermutationBase<Derived> > >
+{
+    typedef Derived PermutationType;
+    typedef typename PermutationType::IndicesType IndicesType;
+    typedef typename PermutationType::PlainPermutationType PlainPermutationType;
   public:
-    typedef Inverse<PermutationType> InverseType;
-    using EigenBase<Inverse<PermutationType> >::derived;
 
     #ifndef EIGEN_PARSED_BY_DOXYGEN
-    typedef typename PermutationType::DenseMatrixType DenseMatrixType;
+    typedef internal::traits<PermutationType> Traits;
+    typedef typename Derived::DenseMatrixType DenseMatrixType;
     enum {
-      RowsAtCompileTime = PermTraits::RowsAtCompileTime,
-      ColsAtCompileTime = PermTraits::ColsAtCompileTime,
-      MaxRowsAtCompileTime = PermTraits::MaxRowsAtCompileTime,
-      MaxColsAtCompileTime = PermTraits::MaxColsAtCompileTime
+      Flags = Traits::Flags,
+      CoeffReadCost = Traits::CoeffReadCost,
+      RowsAtCompileTime = Traits::RowsAtCompileTime,
+      ColsAtCompileTime = Traits::ColsAtCompileTime,
+      MaxRowsAtCompileTime = Traits::MaxRowsAtCompileTime,
+      MaxColsAtCompileTime = Traits::MaxColsAtCompileTime
     };
+    typedef typename Traits::Scalar Scalar;
     #endif
 
+    Transpose(const PermutationType& p) : m_permutation(p) {}
+
+    inline int rows() const { return m_permutation.rows(); }
+    inline int cols() const { return m_permutation.cols(); }
+
     #ifndef EIGEN_PARSED_BY_DOXYGEN
     template<typename DenseDerived>
     void evalTo(MatrixBase<DenseDerived>& other) const
     {
       other.setZero();
-      for (Index i=0; i<derived().rows();++i)
-        other.coeffRef(i, derived().nestedExpression().indices().coeff(i)) = typename DenseDerived::Scalar(1);
+      for (int i=0; i<rows();++i)
+        other.coeffRef(i, m_permutation.indices().coeff(i)) = typename DenseDerived::Scalar(1);
     }
     #endif
 
     /** \return the equivalent permutation matrix */
-    PlainPermutationType eval() const { return derived(); }
+    PlainPermutationType eval() const { return *this; }
 
-    DenseMatrixType toDenseMatrix() const { return derived(); }
+    DenseMatrixType toDenseMatrix() const { return *this; }
 
     /** \returns the matrix with the inverse permutation applied to the columns.
       */
     template<typename OtherDerived> friend
-    const Product<OtherDerived, InverseType, AliasFreeProduct>
-    operator*(const MatrixBase<OtherDerived>& matrix, const InverseType& trPerm)
+    inline const internal::permut_matrix_product_retval<PermutationType, OtherDerived, OnTheRight, true>
+    operator*(const MatrixBase<OtherDerived>& matrix, const Transpose& trPerm)
     {
-      return Product<OtherDerived, InverseType, AliasFreeProduct>(matrix.derived(), trPerm.derived());
+      return internal::permut_matrix_product_retval<PermutationType, OtherDerived, OnTheRight, true>(trPerm.m_permutation, matrix.derived());
     }
 
     /** \returns the matrix with the inverse permutation applied to the rows.
       */
     template<typename OtherDerived>
-    const Product<InverseType, OtherDerived, AliasFreeProduct>
+    inline const internal::permut_matrix_product_retval<PermutationType, OtherDerived, OnTheLeft, true>
     operator*(const MatrixBase<OtherDerived>& matrix) const
     {
-      return Product<InverseType, OtherDerived, AliasFreeProduct>(derived(), matrix.derived());
+      return internal::permut_matrix_product_retval<PermutationType, OtherDerived, OnTheLeft, true>(m_permutation, matrix.derived());
     }
+
+    const PermutationType& nestedPermutation() const { return m_permutation; }
+
+  protected:
+    const PermutationType& m_permutation;
 };
 
 template<typename Derived>
@@ -622,12 +717,6 @@ const PermutationWrapper<const Derived> MatrixBase<Derived>::asPermutation() con
   return derived();
 }
 
-namespace internal {
-
-template<> struct AssignmentKind<DenseShape,PermutationShape> { typedef EigenBase2EigenBase Kind; };
-
-} // end namespace internal
-
 } // end namespace Eigen
 
 #endif // EIGEN_PERMUTATIONMATRIX_H

Eigen/src/Core/PlainObjectBase.h

@@ -28,7 +28,6 @@ namespace internal {
 
 template<int MaxSizeAtCompileTime> struct check_rows_cols_for_overflow {
   template<typename Index>
-  EIGEN_DEVICE_FUNC
   static EIGEN_ALWAYS_INLINE void run(Index, Index)
   {
   }
@@ -36,7 +35,6 @@ template<int MaxSizeAtCompileTime> struct check_rows_cols_for_overflow {
 
 template<> struct check_rows_cols_for_overflow<Dynamic> {
   template<typename Index>
-  EIGEN_DEVICE_FUNC
   static EIGEN_ALWAYS_INLINE void run(Index rows, Index cols)
   {
     // http://hg.mozilla.org/mozilla-central/file/6c8a909977d3/xpcom/ds/CheckedInt.h#l242
@@ -59,34 +57,32 @@ template<typename MatrixTypeA, typename MatrixTypeB, bool SwapPointers> struct m
 } // end namespace internal
 
 /** \class PlainObjectBase
-  * \ingroup Core_Module
   * \brief %Dense storage base class for matrices and arrays.
   *
   * This class can be extended with the help of the plugin mechanism described on the page
-  * \ref TopicCustomizing_Plugins by defining the preprocessor symbol \c EIGEN_PLAINOBJECTBASE_PLUGIN.
+  * \ref TopicCustomizingEigen by defining the preprocessor symbol \c EIGEN_PLAINOBJECTBASE_PLUGIN.
   *
   * \sa \ref TopicClassHierarchy
   */
 #ifdef EIGEN_PARSED_BY_DOXYGEN
-namespace doxygen {
+namespace internal {
 
-// this is a workaround to doxygen not being able to understand the inheritance logic
+// this is a warkaround to doxygen not being able to understand the inheritence logic
 // when it is hidden by the dense_xpr_base helper struct.
-/** This class is just a workaround for Doxygen and it does not not actually exist. */
-template<typename Derived> struct dense_xpr_base_dispatcher;
+template<typename Derived> struct dense_xpr_base_dispatcher_for_doxygen;// : public MatrixBase<Derived> {};
 /** This class is just a workaround for Doxygen and it does not not actually exist. */
 template<typename _Scalar, int _Rows, int _Cols, int _Options, int _MaxRows, int _MaxCols>
-struct dense_xpr_base_dispatcher<Matrix<_Scalar, _Rows, _Cols, _Options, _MaxRows, _MaxCols> >
+struct dense_xpr_base_dispatcher_for_doxygen<Matrix<_Scalar, _Rows, _Cols, _Options, _MaxRows, _MaxCols> >
     : public MatrixBase<Matrix<_Scalar, _Rows, _Cols, _Options, _MaxRows, _MaxCols> > {};
 /** This class is just a workaround for Doxygen and it does not not actually exist. */
 template<typename _Scalar, int _Rows, int _Cols, int _Options, int _MaxRows, int _MaxCols>
-struct dense_xpr_base_dispatcher<Array<_Scalar, _Rows, _Cols, _Options, _MaxRows, _MaxCols> >
+struct dense_xpr_base_dispatcher_for_doxygen<Array<_Scalar, _Rows, _Cols, _Options, _MaxRows, _MaxCols> >
     : public ArrayBase<Array<_Scalar, _Rows, _Cols, _Options, _MaxRows, _MaxCols> > {};
 
-} // namespace doxygen
+} // namespace internal
 
 template<typename Derived>
-class PlainObjectBase : public doxygen::dense_xpr_base_dispatcher<Derived>
+class PlainObjectBase : public internal::dense_xpr_base_dispatcher_for_doxygen<Derived>
 #else
 template<typename Derived>
 class PlainObjectBase : public internal::dense_xpr_base<Derived>::type
@@ -97,8 +93,8 @@ class PlainObjectBase : public internal::dense_xpr_base<Derived>::type
     typedef typename internal::dense_xpr_base<Derived>::type 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;
     typedef Derived DenseType;
@@ -117,40 +113,28 @@ class PlainObjectBase : public internal::dense_xpr_base<Derived>::type
     typedef Eigen::Map<Derived, Unaligned>  MapType;
     friend  class Eigen::Map<const Derived, Unaligned>;
     typedef const Eigen::Map<const Derived, Unaligned> ConstMapType;
-#if EIGEN_MAX_ALIGN_BYTES>0
-    // for EIGEN_MAX_ALIGN_BYTES==0, AlignedMax==Unaligned, and many compilers generate warnings for friend-ing a class twice.
-    friend  class Eigen::Map<Derived, AlignedMax>;
-    friend  class Eigen::Map<const Derived, AlignedMax>;
-#endif
-    typedef Eigen::Map<Derived, AlignedMax> AlignedMapType;
-    typedef const Eigen::Map<const Derived, AlignedMax> ConstAlignedMapType;
+    friend  class Eigen::Map<Derived, Aligned>;
+    typedef Eigen::Map<Derived, Aligned> AlignedMapType;
+    friend  class Eigen::Map<const Derived, Aligned>;
+    typedef const Eigen::Map<const Derived, Aligned> ConstAlignedMapType;
     template<typename StrideType> struct StridedMapType { typedef Eigen::Map<Derived, Unaligned, StrideType> type; };
     template<typename StrideType> struct StridedConstMapType { typedef Eigen::Map<const Derived, Unaligned, StrideType> type; };
-    template<typename StrideType> struct StridedAlignedMapType { typedef Eigen::Map<Derived, AlignedMax, StrideType> type; };
-    template<typename StrideType> struct StridedConstAlignedMapType { typedef Eigen::Map<const Derived, AlignedMax, StrideType> type; };
+    template<typename StrideType> struct StridedAlignedMapType { typedef Eigen::Map<Derived, Aligned, StrideType> type; };
+    template<typename StrideType> struct StridedConstAlignedMapType { typedef Eigen::Map<const Derived, Aligned, StrideType> type; };
 
   protected:
     DenseStorage<Scalar, Base::MaxSizeAtCompileTime, Base::RowsAtCompileTime, Base::ColsAtCompileTime, Options> m_storage;
 
   public:
-    enum { NeedsToAlign = (SizeAtCompileTime != Dynamic) && (internal::traits<Derived>::Alignment>0) };
+    enum { NeedsToAlign = SizeAtCompileTime != Dynamic && (internal::traits<Derived>::Flags & AlignedBit) != 0 };
     EIGEN_MAKE_ALIGNED_OPERATOR_NEW_IF(NeedsToAlign)
 
-    EIGEN_DEVICE_FUNC
     Base& base() { return *static_cast<Base*>(this); }
-    EIGEN_DEVICE_FUNC
     const Base& base() const { return *static_cast<const Base*>(this); }
 
-    EIGEN_DEVICE_FUNC
     EIGEN_STRONG_INLINE Index rows() const { return m_storage.rows(); }
-    EIGEN_DEVICE_FUNC
     EIGEN_STRONG_INLINE Index cols() const { return m_storage.cols(); }
 
-    /** This is an overloaded version of DenseCoeffsBase<Derived,ReadOnlyAccessors>::coeff(Index,Index) const
-      * provided to by-pass the creation of an evaluator of the expression, thus saving compilation efforts.
-      *
-      * See DenseCoeffsBase<Derived,ReadOnlyAccessors>::coeff(Index) const for details. */
-    EIGEN_DEVICE_FUNC
     EIGEN_STRONG_INLINE const Scalar& coeff(Index rowId, Index colId) const
     {
       if(Flags & RowMajorBit)
@@ -159,21 +143,11 @@ class PlainObjectBase : public internal::dense_xpr_base<Derived>::type
         return m_storage.data()[rowId + colId * m_storage.rows()];
     }
 
-    /** This is an overloaded version of DenseCoeffsBase<Derived,ReadOnlyAccessors>::coeff(Index) const
-      * provided to by-pass the creation of an evaluator of the expression, thus saving compilation efforts.
-      *
-      * See DenseCoeffsBase<Derived,ReadOnlyAccessors>::coeff(Index) const for details. */
-    EIGEN_DEVICE_FUNC
     EIGEN_STRONG_INLINE const Scalar& coeff(Index index) const
     {
       return m_storage.data()[index];
     }
 
-    /** This is an overloaded version of DenseCoeffsBase<Derived,WriteAccessors>::coeffRef(Index,Index) const
-      * provided to by-pass the creation of an evaluator of the expression, thus saving compilation efforts.
-      *
-      * See DenseCoeffsBase<Derived,WriteAccessors>::coeffRef(Index,Index) const for details. */
-    EIGEN_DEVICE_FUNC
     EIGEN_STRONG_INLINE Scalar& coeffRef(Index rowId, Index colId)
     {
       if(Flags & RowMajorBit)
@@ -182,19 +156,11 @@ class PlainObjectBase : public internal::dense_xpr_base<Derived>::type
         return m_storage.data()[rowId + colId * m_storage.rows()];
     }
 
-    /** This is an overloaded version of DenseCoeffsBase<Derived,WriteAccessors>::coeffRef(Index) const
-      * provided to by-pass the creation of an evaluator of the expression, thus saving compilation efforts.
-      *
-      * See DenseCoeffsBase<Derived,WriteAccessors>::coeffRef(Index) const for details. */
-    EIGEN_DEVICE_FUNC
     EIGEN_STRONG_INLINE Scalar& coeffRef(Index index)
     {
       return m_storage.data()[index];
     }
 
-    /** This is the const version of coeffRef(Index,Index) which is thus synonym of coeff(Index,Index).
-      * It is provided for convenience. */
-    EIGEN_DEVICE_FUNC
     EIGEN_STRONG_INLINE const Scalar& coeffRef(Index rowId, Index colId) const
     {
       if(Flags & RowMajorBit)
@@ -203,9 +169,6 @@ class PlainObjectBase : public internal::dense_xpr_base<Derived>::type
         return m_storage.data()[rowId + colId * m_storage.rows()];
     }
 
-    /** This is the const version of coeffRef(Index) which is thus synonym of coeff(Index).
-      * It is provided for convenience. */
-    EIGEN_DEVICE_FUNC
     EIGEN_STRONG_INLINE const Scalar& coeffRef(Index index) const
     {
       return m_storage.data()[index];
@@ -246,11 +209,11 @@ class PlainObjectBase : public internal::dense_xpr_base<Derived>::type
     }
 
     /** \returns a const pointer to the data array of this matrix */
-    EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE const Scalar *data() const
+    EIGEN_STRONG_INLINE const Scalar *data() const
     { return m_storage.data(); }
 
     /** \returns a pointer to the data array of this matrix */
-    EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE Scalar *data()
+    EIGEN_STRONG_INLINE Scalar *data()
     { return m_storage.data(); }
 
     /** Resizes \c *this to a \a rows x \a cols matrix.
@@ -269,22 +232,22 @@ class PlainObjectBase : public internal::dense_xpr_base<Derived>::type
       *
       * \sa resize(Index) for vectors, resize(NoChange_t, Index), resize(Index, NoChange_t)
       */
-    EIGEN_DEVICE_FUNC
-    EIGEN_STRONG_INLINE void resize(Index rows, Index cols)
+    EIGEN_STRONG_INLINE void resize(Index nbRows, Index nbCols)
     {
-      eigen_assert(   EIGEN_IMPLIES(RowsAtCompileTime!=Dynamic,rows==RowsAtCompileTime)
-                   && EIGEN_IMPLIES(ColsAtCompileTime!=Dynamic,cols==ColsAtCompileTime)
-                   && EIGEN_IMPLIES(RowsAtCompileTime==Dynamic && MaxRowsAtCompileTime!=Dynamic,rows<=MaxRowsAtCompileTime)
-                   && EIGEN_IMPLIES(ColsAtCompileTime==Dynamic && MaxColsAtCompileTime!=Dynamic,cols<=MaxColsAtCompileTime)
-                   && rows>=0 && cols>=0 && "Invalid sizes when resizing a matrix or array.");
-      internal::check_rows_cols_for_overflow<MaxSizeAtCompileTime>::run(rows, cols);
+      eigen_assert(   EIGEN_IMPLIES(RowsAtCompileTime!=Dynamic,nbRows==RowsAtCompileTime)
+                   && EIGEN_IMPLIES(ColsAtCompileTime!=Dynamic,nbCols==ColsAtCompileTime)
+                   && EIGEN_IMPLIES(RowsAtCompileTime==Dynamic && MaxRowsAtCompileTime!=Dynamic,nbRows<=MaxRowsAtCompileTime)
+                   && EIGEN_IMPLIES(ColsAtCompileTime==Dynamic && MaxColsAtCompileTime!=Dynamic,nbCols<=MaxColsAtCompileTime)
+                   && nbRows>=0 && nbCols>=0 && "Invalid sizes when resizing a matrix or array.");
+      internal::check_rows_cols_for_overflow<MaxSizeAtCompileTime>::run(nbRows, nbCols);
       #ifdef EIGEN_INITIALIZE_COEFFS
-        Index size = rows*cols;
+        Index size = nbRows*nbCols;
         bool size_changed = size != this->size();
-        m_storage.resize(size, rows, cols);
+        m_storage.resize(size, nbRows, nbCols);
         if(size_changed) EIGEN_INITIALIZE_COEFFS_IF_THAT_OPTION_IS_ENABLED
       #else
-        m_storage.resize(rows*cols, rows, cols);
+        internal::check_rows_cols_for_overflow<MaxSizeAtCompileTime>::run(nbRows, nbCols);
+        m_storage.resize(nbRows*nbCols, nbRows, nbCols);
       #endif
     }
 
@@ -299,7 +262,6 @@ class PlainObjectBase : public internal::dense_xpr_base<Derived>::type
       *
       * \sa resize(Index,Index), resize(NoChange_t, Index), resize(Index, NoChange_t)
       */
-    EIGEN_DEVICE_FUNC
     inline void resize(Index size)
     {
       EIGEN_STATIC_ASSERT_VECTOR_ONLY(PlainObjectBase)
@@ -324,10 +286,9 @@ class PlainObjectBase : public internal::dense_xpr_base<Derived>::type
       *
       * \sa resize(Index,Index)
       */
-    EIGEN_DEVICE_FUNC
-    inline void resize(NoChange_t, Index cols)
+    inline void resize(NoChange_t, Index nbCols)
     {
-      resize(rows(), cols);
+      resize(rows(), nbCols);
     }
 
     /** Resizes the matrix, changing only the number of rows. For the parameter of type NoChange_t, just pass the special value \c NoChange
@@ -338,10 +299,9 @@ class PlainObjectBase : public internal::dense_xpr_base<Derived>::type
       *
       * \sa resize(Index,Index)
       */
-    EIGEN_DEVICE_FUNC
-    inline void resize(Index rows, NoChange_t)
+    inline void resize(Index nbRows, NoChange_t)
     {
-      resize(rows, cols());
+      resize(nbRows, cols());
     }
 
     /** Resizes \c *this to have the same dimensions as \a other.
@@ -352,12 +312,11 @@ class PlainObjectBase : public internal::dense_xpr_base<Derived>::type
       * remain row-vectors and vectors remain vectors.
       */
     template<typename OtherDerived>
-    EIGEN_DEVICE_FUNC 
     EIGEN_STRONG_INLINE void resizeLike(const EigenBase<OtherDerived>& _other)
     {
       const OtherDerived& other = _other.derived();
-      internal::check_rows_cols_for_overflow<MaxSizeAtCompileTime>::run(other.rows(), other.cols());
-      const Index othersize = other.rows()*other.cols();
+      internal::check_rows_cols_for_overflow<MaxSizeAtCompileTime>::run(Index(other.rows()), Index(other.cols()));
+      const Index othersize = Index(other.rows())*Index(other.cols());
       if(RowsAtCompileTime == 1)
       {
         eigen_assert(other.rows() == 1 || other.cols() == 1);
@@ -380,10 +339,9 @@ class PlainObjectBase : public internal::dense_xpr_base<Derived>::type
       * Matrices are resized relative to the top-left element. In case values need to be 
       * appended to the matrix they will be uninitialized.
       */
-    EIGEN_DEVICE_FUNC
-    EIGEN_STRONG_INLINE void conservativeResize(Index rows, Index cols)
+    EIGEN_STRONG_INLINE void conservativeResize(Index nbRows, Index nbCols)
     {
-      internal::conservative_resize_like_impl<Derived>::run(*this, rows, cols);
+      internal::conservative_resize_like_impl<Derived>::run(*this, nbRows, nbCols);
     }
 
     /** Resizes the matrix to \a rows x \a cols while leaving old values untouched.
@@ -393,11 +351,10 @@ class PlainObjectBase : public internal::dense_xpr_base<Derived>::type
       *
       * In case the matrix is growing, new rows will be uninitialized.
       */
-    EIGEN_DEVICE_FUNC
-    EIGEN_STRONG_INLINE void conservativeResize(Index rows, NoChange_t)
+    EIGEN_STRONG_INLINE void conservativeResize(Index nbRows, NoChange_t)
     {
       // Note: see the comment in conservativeResize(Index,Index)
-      conservativeResize(rows, cols());
+      conservativeResize(nbRows, cols());
     }
 
     /** Resizes the matrix to \a rows x \a cols while leaving old values untouched.
@@ -407,11 +364,10 @@ class PlainObjectBase : public internal::dense_xpr_base<Derived>::type
       *
       * In case the matrix is growing, new columns will be uninitialized.
       */
-    EIGEN_DEVICE_FUNC
-    EIGEN_STRONG_INLINE void conservativeResize(NoChange_t, Index cols)
+    EIGEN_STRONG_INLINE void conservativeResize(NoChange_t, Index nbCols)
     {
       // Note: see the comment in conservativeResize(Index,Index)
-      conservativeResize(rows(), cols);
+      conservativeResize(rows(), nbCols);
     }
 
     /** Resizes the vector to \a size while retaining old values.
@@ -422,7 +378,6 @@ class PlainObjectBase : public internal::dense_xpr_base<Derived>::type
       *
       * When values are appended, they will be uninitialized.
       */
-    EIGEN_DEVICE_FUNC
     EIGEN_STRONG_INLINE void conservativeResize(Index size)
     {
       internal::conservative_resize_like_impl<Derived>::run(*this, size);
@@ -438,7 +393,6 @@ class PlainObjectBase : public internal::dense_xpr_base<Derived>::type
       * appended to the matrix they will copied from \c other.
       */
     template<typename OtherDerived>
-    EIGEN_DEVICE_FUNC
     EIGEN_STRONG_INLINE void conservativeResizeLike(const DenseBase<OtherDerived>& other)
     {
       internal::conservative_resize_like_impl<Derived,OtherDerived>::run(*this, other);
@@ -447,7 +401,6 @@ class PlainObjectBase : public internal::dense_xpr_base<Derived>::type
     /** This is a special case of the templated operator=. Its purpose is to
       * prevent a default operator= from hiding the templated operator=.
       */
-    EIGEN_DEVICE_FUNC
     EIGEN_STRONG_INLINE Derived& operator=(const PlainObjectBase& other)
     {
       return _set(other);
@@ -455,7 +408,6 @@ class PlainObjectBase : public internal::dense_xpr_base<Derived>::type
 
     /** \sa MatrixBase::lazyAssign() */
     template<typename OtherDerived>
-    EIGEN_DEVICE_FUNC
     EIGEN_STRONG_INLINE Derived& lazyAssign(const DenseBase<OtherDerived>& other)
     {
       _resize_to_match(other);
@@ -463,18 +415,12 @@ class PlainObjectBase : public internal::dense_xpr_base<Derived>::type
     }
 
     template<typename OtherDerived>
-    EIGEN_DEVICE_FUNC
     EIGEN_STRONG_INLINE Derived& operator=(const ReturnByValue<OtherDerived>& func)
     {
       resize(func.rows(), func.cols());
       return Base::operator=(func);
     }
 
-    // Prevent user from trying to instantiate PlainObjectBase objects
-    // by making all its constructor protected. See bug 1074.
-  protected:
-
-    EIGEN_DEVICE_FUNC
     EIGEN_STRONG_INLINE PlainObjectBase() : m_storage()
     {
 //       _check_template_params();
@@ -484,23 +430,20 @@ class PlainObjectBase : public internal::dense_xpr_base<Derived>::type
 #ifndef EIGEN_PARSED_BY_DOXYGEN
     // FIXME is it still needed ?
     /** \internal */
-    EIGEN_DEVICE_FUNC
-    explicit PlainObjectBase(internal::constructor_without_unaligned_array_assert)
+    PlainObjectBase(internal::constructor_without_unaligned_array_assert)
       : m_storage(internal::constructor_without_unaligned_array_assert())
     {
 //       _check_template_params(); EIGEN_INITIALIZE_COEFFS_IF_THAT_OPTION_IS_ENABLED
     }
 #endif
 
-#if EIGEN_HAS_RVALUE_REFERENCES
-    EIGEN_DEVICE_FUNC
-    PlainObjectBase(PlainObjectBase&& other) EIGEN_NOEXCEPT
+#ifdef EIGEN_HAVE_RVALUE_REFERENCES
+    PlainObjectBase(PlainObjectBase&& other)
       : m_storage( std::move(other.m_storage) )
     {
     }
 
-    EIGEN_DEVICE_FUNC
-    PlainObjectBase& operator=(PlainObjectBase&& other) EIGEN_NOEXCEPT
+    PlainObjectBase& operator=(PlainObjectBase&& other)
     {
       using std::swap;
       swap(m_storage, other.m_storage);
@@ -509,55 +452,31 @@ class PlainObjectBase : public internal::dense_xpr_base<Derived>::type
 #endif
 
     /** Copy constructor */
-    EIGEN_DEVICE_FUNC
     EIGEN_STRONG_INLINE PlainObjectBase(const PlainObjectBase& other)
-      : Base(), m_storage(other.m_storage) { }
-    EIGEN_DEVICE_FUNC
-    EIGEN_STRONG_INLINE PlainObjectBase(Index size, Index rows, Index cols)
-      : m_storage(size, rows, cols)
+      : m_storage()
+    {
+      _check_template_params();
+      lazyAssign(other);
+    }
+
+    template<typename OtherDerived>
+    EIGEN_STRONG_INLINE PlainObjectBase(const DenseBase<OtherDerived> &other)
+      : m_storage()
+    {
+      _check_template_params();
+      lazyAssign(other);
+    }
+
+    EIGEN_STRONG_INLINE PlainObjectBase(Index a_size, Index nbRows, Index nbCols)
+      : m_storage(a_size, nbRows, nbCols)
     {
 //       _check_template_params();
 //       EIGEN_INITIALIZE_COEFFS_IF_THAT_OPTION_IS_ENABLED
     }
 
-    /** \sa PlainObjectBase::operator=(const EigenBase<OtherDerived>&) */
-    template<typename OtherDerived>
-    EIGEN_DEVICE_FUNC
-    EIGEN_STRONG_INLINE PlainObjectBase(const DenseBase<OtherDerived> &other)
-      : m_storage()
-    {
-      _check_template_params();
-      resizeLike(other);
-      _set_noalias(other);
-    }
-
-    /** \sa PlainObjectBase::operator=(const EigenBase<OtherDerived>&) */
-    template<typename OtherDerived>
-    EIGEN_DEVICE_FUNC
-    EIGEN_STRONG_INLINE PlainObjectBase(const EigenBase<OtherDerived> &other)
-      : m_storage()
-    {
-      _check_template_params();
-      resizeLike(other);
-      *this = other.derived();
-    }
-    /** \brief Copy constructor with in-place evaluation */
-    template<typename OtherDerived>
-    EIGEN_DEVICE_FUNC
-    EIGEN_STRONG_INLINE PlainObjectBase(const ReturnByValue<OtherDerived>& other)
-    {
-      _check_template_params();
-      // FIXME this does not automatically transpose vectors if necessary
-      resize(other.rows(), other.cols());
-      other.evalTo(this->derived());
-    }
-
-  public:
-
-    /** \copydoc DenseBase::operator=(const EigenBase<OtherDerived>&)
+    /** \copydoc MatrixBase::operator=(const EigenBase<OtherDerived>&)
       */
     template<typename OtherDerived>
-    EIGEN_DEVICE_FUNC 
     EIGEN_STRONG_INLINE Derived& operator=(const EigenBase<OtherDerived> &other)
     {
       _resize_to_match(other);
@@ -565,6 +484,16 @@ class PlainObjectBase : public internal::dense_xpr_base<Derived>::type
       return this->derived();
     }
 
+    /** \sa MatrixBase::operator=(const EigenBase<OtherDerived>&) */
+    template<typename OtherDerived>
+    EIGEN_STRONG_INLINE PlainObjectBase(const EigenBase<OtherDerived> &other)
+      : m_storage(Index(other.derived().rows()) * Index(other.derived().cols()), other.derived().rows(), other.derived().cols())
+    {
+      _check_template_params();
+      internal::check_rows_cols_for_overflow<MaxSizeAtCompileTime>::run(other.derived().rows(), other.derived().cols());
+      Base::operator=(other.derived());
+    }
+
     /** \name Map
       * These are convenience functions returning Map objects. The Map() static functions return unaligned Map objects,
       * while the AlignedMap() functions return aligned Map objects and thus should be called only with 16-byte-aligned
@@ -639,16 +568,16 @@ class PlainObjectBase : public internal::dense_xpr_base<Derived>::type
     //@}
 
     using Base::setConstant;
-    EIGEN_DEVICE_FUNC Derived& setConstant(Index size, const Scalar& val);
-    EIGEN_DEVICE_FUNC Derived& setConstant(Index rows, Index cols, const Scalar& val);
+    Derived& setConstant(Index size, const Scalar& value);
+    Derived& setConstant(Index rows, Index cols, const Scalar& value);
 
     using Base::setZero;
-    EIGEN_DEVICE_FUNC Derived& setZero(Index size);
-    EIGEN_DEVICE_FUNC Derived& setZero(Index rows, Index cols);
+    Derived& setZero(Index size);
+    Derived& setZero(Index rows, Index cols);
 
     using Base::setOnes;
-    EIGEN_DEVICE_FUNC Derived& setOnes(Index size);
-    EIGEN_DEVICE_FUNC Derived& setOnes(Index rows, Index cols);
+    Derived& setOnes(Index size);
+    Derived& setOnes(Index rows, Index cols);
 
     using Base::setRandom;
     Derived& setRandom(Index size);
@@ -667,7 +596,6 @@ class PlainObjectBase : public internal::dense_xpr_base<Derived>::type
       * remain row-vectors and vectors remain vectors.
       */
     template<typename OtherDerived>
-    EIGEN_DEVICE_FUNC 
     EIGEN_STRONG_INLINE void _resize_to_match(const EigenBase<OtherDerived>& other)
     {
       #ifdef EIGEN_NO_AUTOMATIC_RESIZING
@@ -675,6 +603,8 @@ class PlainObjectBase : public internal::dense_xpr_base<Derived>::type
                  : (rows() == other.rows() && cols() == other.cols())))
         && "Size mismatch. Automatic resizing is disabled because EIGEN_NO_AUTOMATIC_RESIZING is defined");
       EIGEN_ONLY_USED_FOR_DEBUG(other);
+      if(this->size()==0)
+        resizeLike(other);
       #else
       resizeLike(other);
       #endif
@@ -694,23 +624,25 @@ class PlainObjectBase : public internal::dense_xpr_base<Derived>::type
       *
       * \internal
       */
-    // aliasing is dealt once in internall::call_assignment
-    // so at this stage we have to assume aliasing... and resising has to be done later.
     template<typename OtherDerived>
-    EIGEN_DEVICE_FUNC 
     EIGEN_STRONG_INLINE Derived& _set(const DenseBase<OtherDerived>& other)
     {
-      internal::call_assignment(this->derived(), other.derived());
+      _set_selector(other.derived(), typename internal::conditional<static_cast<bool>(int(OtherDerived::Flags) & EvalBeforeAssigningBit), internal::true_type, internal::false_type>::type());
       return this->derived();
     }
 
+    template<typename OtherDerived>
+    EIGEN_STRONG_INLINE void _set_selector(const OtherDerived& other, const internal::true_type&) { _set_noalias(other.eval()); }
+
+    template<typename OtherDerived>
+    EIGEN_STRONG_INLINE void _set_selector(const OtherDerived& other, const internal::false_type&) { _set_noalias(other); }
+
     /** \internal Like _set() but additionally makes the assumption that no aliasing effect can happen (which
       * is the case when creating a new matrix) so one can enforce lazy evaluation.
       *
       * \sa operator=(const MatrixBase<OtherDerived>&), _set()
       */
     template<typename OtherDerived>
-    EIGEN_DEVICE_FUNC 
     EIGEN_STRONG_INLINE Derived& _set_noalias(const DenseBase<OtherDerived>& other)
     {
       // I don't think we need this resize call since the lazyAssign will anyways resize
@@ -718,167 +650,40 @@ class PlainObjectBase : public internal::dense_xpr_base<Derived>::type
       //_resize_to_match(other);
       // the 'false' below means to enforce lazy evaluation. We don't use lazyAssign() because
       // it wouldn't allow to copy a row-vector into a column-vector.
-      internal::call_assignment_no_alias(this->derived(), other.derived(), internal::assign_op<Scalar,typename OtherDerived::Scalar>());
-      return this->derived();
+      return internal::assign_selector<Derived,OtherDerived,false>::run(this->derived(), other.derived());
     }
 
     template<typename T0, typename T1>
-    EIGEN_DEVICE_FUNC
-    EIGEN_STRONG_INLINE void _init2(Index rows, Index cols, typename internal::enable_if<Base::SizeAtCompileTime!=2,T0>::type* = 0)
+    EIGEN_STRONG_INLINE void _init2(Index nbRows, Index nbCols, typename internal::enable_if<Base::SizeAtCompileTime!=2,T0>::type* = 0)
     {
       EIGEN_STATIC_ASSERT(bool(NumTraits<T0>::IsInteger) &&
                           bool(NumTraits<T1>::IsInteger),
                           FLOATING_POINT_ARGUMENT_PASSED__INTEGER_WAS_EXPECTED)
-      resize(rows,cols);
+      resize(nbRows,nbCols);
     }
-    
     template<typename T0, typename T1>
-    EIGEN_DEVICE_FUNC 
-    EIGEN_STRONG_INLINE void _init2(const T0& val0, const T1& val1, typename internal::enable_if<Base::SizeAtCompileTime==2,T0>::type* = 0)
+    EIGEN_STRONG_INLINE void _init2(const Scalar& val0, const Scalar& val1, typename internal::enable_if<Base::SizeAtCompileTime==2,T0>::type* = 0)
     {
       EIGEN_STATIC_ASSERT_VECTOR_SPECIFIC_SIZE(PlainObjectBase, 2)
-      m_storage.data()[0] = Scalar(val0);
-      m_storage.data()[1] = Scalar(val1);
-    }
-    
-    template<typename T0, typename T1>
-    EIGEN_DEVICE_FUNC 
-    EIGEN_STRONG_INLINE void _init2(const Index& val0, const Index& val1,
-                                    typename internal::enable_if<    (!internal::is_same<Index,Scalar>::value)
-                                                                  && (internal::is_same<T0,Index>::value)
-                                                                  && (internal::is_same<T1,Index>::value)
-                                                                  && Base::SizeAtCompileTime==2,T1>::type* = 0)
-    {
-      EIGEN_STATIC_ASSERT_VECTOR_SPECIFIC_SIZE(PlainObjectBase, 2)
-      m_storage.data()[0] = Scalar(val0);
-      m_storage.data()[1] = Scalar(val1);
-    }
-
-    // The argument is convertible to the Index type and we either have a non 1x1 Matrix, or a dynamic-sized Array,
-    // then the argument is meant to be the size of the object.
-    template<typename T>
-    EIGEN_DEVICE_FUNC
-    EIGEN_STRONG_INLINE void _init1(Index size, typename internal::enable_if<    (Base::SizeAtCompileTime!=1 || !internal::is_convertible<T, Scalar>::value)
-                                                                              && ((!internal::is_same<typename internal::traits<Derived>::XprKind,ArrayXpr>::value || Base::SizeAtCompileTime==Dynamic)),T>::type* = 0)
-    {
-      // NOTE MSVC 2008 complains if we directly put bool(NumTraits<T>::IsInteger) as the EIGEN_STATIC_ASSERT argument.
-      const bool is_integer = NumTraits<T>::IsInteger;
-      EIGEN_UNUSED_VARIABLE(is_integer);
-      EIGEN_STATIC_ASSERT(is_integer,
-                          FLOATING_POINT_ARGUMENT_PASSED__INTEGER_WAS_EXPECTED)
-      resize(size);
-    }
-    
-    // We have a 1x1 matrix/array => the argument is interpreted as the value of the unique coefficient (case where scalar type can be implicitely converted)
-    template<typename T>
-    EIGEN_DEVICE_FUNC
-    EIGEN_STRONG_INLINE void _init1(const Scalar& val0, typename internal::enable_if<Base::SizeAtCompileTime==1 && internal::is_convertible<T, Scalar>::value,T>::type* = 0)
-    {
-      EIGEN_STATIC_ASSERT_VECTOR_SPECIFIC_SIZE(PlainObjectBase, 1)
       m_storage.data()[0] = val0;
-    }
-    
-    // We have a 1x1 matrix/array => the argument is interpreted as the value of the unique coefficient (case where scalar type match the index type)
-    template<typename T>
-    EIGEN_DEVICE_FUNC
-    EIGEN_STRONG_INLINE void _init1(const Index& val0,
-                                    typename internal::enable_if<    (!internal::is_same<Index,Scalar>::value)
-                                                                  && (internal::is_same<Index,T>::value)
-                                                                  && Base::SizeAtCompileTime==1
-                                                                  && internal::is_convertible<T, Scalar>::value,T*>::type* = 0)
-    {
-      EIGEN_STATIC_ASSERT_VECTOR_SPECIFIC_SIZE(PlainObjectBase, 1)
-      m_storage.data()[0] = Scalar(val0);
+      m_storage.data()[1] = val1;
     }
 
-    // Initialize a fixed size matrix from a pointer to raw data
-    template<typename T>
-    EIGEN_DEVICE_FUNC
-    EIGEN_STRONG_INLINE void _init1(const Scalar* data){
-      this->_set_noalias(ConstMapType(data));
-    }
-
-    // Initialize an arbitrary matrix from a dense expression
-    template<typename T, typename OtherDerived>
-    EIGEN_DEVICE_FUNC
-    EIGEN_STRONG_INLINE void _init1(const DenseBase<OtherDerived>& other){
-      this->_set_noalias(other);
-    }
-
-    // Initialize an arbitrary matrix from a generic Eigen expression
-    template<typename T, typename OtherDerived>
-    EIGEN_DEVICE_FUNC
-    EIGEN_STRONG_INLINE void _init1(const EigenBase<OtherDerived>& other){
-      this->derived() = other;
-    }
-
-    template<typename T, typename OtherDerived>
-    EIGEN_DEVICE_FUNC
-    EIGEN_STRONG_INLINE void _init1(const ReturnByValue<OtherDerived>& other)
-    {
-      resize(other.rows(), other.cols());
-      other.evalTo(this->derived());
-    }
-
-    template<typename T, typename OtherDerived, int ColsAtCompileTime>
-    EIGEN_DEVICE_FUNC
-    EIGEN_STRONG_INLINE void _init1(const RotationBase<OtherDerived,ColsAtCompileTime>& r)
-    {
-      this->derived() = r;
-    }
-    
-    // For fixed -size arrays:
-    template<typename T>
-    EIGEN_DEVICE_FUNC
-    EIGEN_STRONG_INLINE void _init1(const Scalar& val0,
-                                    typename internal::enable_if<    Base::SizeAtCompileTime!=Dynamic
-                                                                  && Base::SizeAtCompileTime!=1
-                                                                  && internal::is_convertible<T, Scalar>::value
-                                                                  && internal::is_same<typename internal::traits<Derived>::XprKind,ArrayXpr>::value,T>::type* = 0)
-    {
-      Base::setConstant(val0);
-    }
-    
-    template<typename T>
-    EIGEN_DEVICE_FUNC
-    EIGEN_STRONG_INLINE void _init1(const Index& val0,
-                                    typename internal::enable_if<    (!internal::is_same<Index,Scalar>::value)
-                                                                  && (internal::is_same<Index,T>::value)
-                                                                  && Base::SizeAtCompileTime!=Dynamic
-                                                                  && Base::SizeAtCompileTime!=1
-                                                                  && internal::is_convertible<T, Scalar>::value
-                                                                  && internal::is_same<typename internal::traits<Derived>::XprKind,ArrayXpr>::value,T*>::type* = 0)
-    {
-      Base::setConstant(val0);
-    }
-    
     template<typename MatrixTypeA, typename MatrixTypeB, bool SwapPointers>
     friend struct internal::matrix_swap_impl;
 
-  public:
-    
-#ifndef EIGEN_PARSED_BY_DOXYGEN
-    /** \internal
-      * \brief Override DenseBase::swap() since for dynamic-sized matrices
-      * of same type it is enough to swap the data pointers.
+    /** \internal generic implementation of swap for dense storage since for dynamic-sized matrices of same type it is enough to swap the
+      * data pointers.
       */
     template<typename OtherDerived>
-    EIGEN_DEVICE_FUNC
-    void swap(DenseBase<OtherDerived> & other)
+    void _swap(DenseBase<OtherDerived> const & other)
     {
       enum { SwapPointers = internal::is_same<Derived, OtherDerived>::value && Base::SizeAtCompileTime==Dynamic };
-      internal::matrix_swap_impl<Derived, OtherDerived, bool(SwapPointers)>::run(this->derived(), other.derived());
+      internal::matrix_swap_impl<Derived, OtherDerived, bool(SwapPointers)>::run(this->derived(), other.const_cast_derived());
     }
-    
-    /** \internal
-      * \brief const version forwarded to DenseBase::swap
-      */
-    template<typename OtherDerived>
-    EIGEN_DEVICE_FUNC
-    void swap(DenseBase<OtherDerived> const & other)
-    { Base::swap(other.derived()); }
-    
-    EIGEN_DEVICE_FUNC 
+
+  public:
+#ifndef EIGEN_PARSED_BY_DOXYGEN
     static EIGEN_STRONG_INLINE void _check_template_params()
     {
       EIGEN_STATIC_ASSERT((EIGEN_IMPLIES(MaxRowsAtCompileTime==1 && MaxColsAtCompileTime!=1, (Options&RowMajor)==RowMajor)
@@ -892,9 +697,10 @@ class PlainObjectBase : public internal::dense_xpr_base<Derived>::type
                         && (Options & (DontAlign|RowMajor)) == Options),
         INVALID_MATRIX_TEMPLATE_PARAMETERS)
     }
-
-    enum { IsPlainObjectBase = 1 };
 #endif
+
+private:
+    enum { ThisConstantIsPrivateInPlainObjectBase };
 };
 
 namespace internal {
@@ -902,6 +708,7 @@ namespace internal {
 template <typename Derived, typename OtherDerived, bool IsVector>
 struct conservative_resize_like_impl
 {
+  typedef typename Derived::Index Index;
   static void run(DenseBase<Derived>& _this, Index rows, Index cols)
   {
     if (_this.rows() == rows && _this.cols() == cols) return;
@@ -917,8 +724,8 @@ struct conservative_resize_like_impl
     {
       // The storage order does not allow us to use reallocation.
       typename Derived::PlainObject tmp(rows,cols);
-      const Index common_rows = numext::mini(rows, _this.rows());
-      const Index common_cols = numext::mini(cols, _this.cols());
+      const Index common_rows = (std::min)(rows, _this.rows());
+      const Index common_cols = (std::min)(cols, _this.cols());
       tmp.block(0,0,common_rows,common_cols) = _this.block(0,0,common_rows,common_cols);
       _this.derived().swap(tmp);
     }
@@ -951,8 +758,8 @@ struct conservative_resize_like_impl
     {
       // The storage order does not allow us to use reallocation.
       typename Derived::PlainObject tmp(other);
-      const Index common_rows = numext::mini(tmp.rows(), _this.rows());
-      const Index common_cols = numext::mini(tmp.cols(), _this.cols());
+      const Index common_rows = (std::min)(tmp.rows(), _this.rows());
+      const Index common_cols = (std::min)(tmp.cols(), _this.cols());
       tmp.block(0,0,common_rows,common_cols) = _this.block(0,0,common_rows,common_cols);
       _this.derived().swap(tmp);
     }
@@ -967,6 +774,7 @@ struct conservative_resize_like_impl<Derived,OtherDerived,true>
 {
   using conservative_resize_like_impl<Derived,OtherDerived,false>::run;
   
+  typedef typename Derived::Index Index;
   static void run(DenseBase<Derived>& _this, Index size)
   {
     const Index new_rows = Derived::RowsAtCompileTime==1 ? 1 : size;
@@ -992,7 +800,6 @@ struct conservative_resize_like_impl<Derived,OtherDerived,true>
 template<typename MatrixTypeA, typename MatrixTypeB, bool SwapPointers>
 struct matrix_swap_impl
 {
-  EIGEN_DEVICE_FUNC
   static inline void run(MatrixTypeA& a, MatrixTypeB& b)
   {
     a.base().swap(b);
@@ -1002,7 +809,6 @@ struct matrix_swap_impl
 template<typename MatrixTypeA, typename MatrixTypeB>
 struct matrix_swap_impl<MatrixTypeA, MatrixTypeB, true>
 {
-  EIGEN_DEVICE_FUNC
   static inline void run(MatrixTypeA& a, MatrixTypeB& b)
   {
     static_cast<typename MatrixTypeA::Base&>(a).m_storage.swap(static_cast<typename MatrixTypeB::Base&>(b).m_storage);

Eigen/src/Core/Product.h

@@ -1,186 +0,0 @@
-// This file is part of Eigen, a lightweight C++ template library
-// for linear algebra.
-//
-// Copyright (C) 2008-2011 Gael Guennebaud <gael.guennebaud@inria.fr>
-//
-// This Source Code Form is subject to the terms of the Mozilla
-// Public License v. 2.0. If a copy of the MPL was not distributed
-// with this file, You can obtain one at http://mozilla.org/MPL/2.0/.
-
-#ifndef EIGEN_PRODUCT_H
-#define EIGEN_PRODUCT_H
-
-namespace Eigen {
-
-template<typename Lhs, typename Rhs, int Option, typename StorageKind> class ProductImpl;
-
-namespace internal {
-
-template<typename Lhs, typename Rhs, int Option>
-struct traits<Product<Lhs, Rhs, Option> >
-{
-  typedef typename remove_all<Lhs>::type LhsCleaned;
-  typedef typename remove_all<Rhs>::type RhsCleaned;
-  typedef traits<LhsCleaned> LhsTraits;
-  typedef traits<RhsCleaned> RhsTraits;
-  
-  typedef MatrixXpr XprKind;
-  
-  typedef typename ScalarBinaryOpTraits<typename traits<LhsCleaned>::Scalar, typename traits<RhsCleaned>::Scalar>::ReturnType Scalar;
-  typedef typename product_promote_storage_type<typename LhsTraits::StorageKind,
-                                                typename RhsTraits::StorageKind,
-                                                internal::product_type<Lhs,Rhs>::ret>::ret StorageKind;
-  typedef typename promote_index_type<typename LhsTraits::StorageIndex,
-                                      typename RhsTraits::StorageIndex>::type StorageIndex;
-  
-  enum {
-    RowsAtCompileTime    = LhsTraits::RowsAtCompileTime,
-    ColsAtCompileTime    = RhsTraits::ColsAtCompileTime,
-    MaxRowsAtCompileTime = LhsTraits::MaxRowsAtCompileTime,
-    MaxColsAtCompileTime = RhsTraits::MaxColsAtCompileTime,
-    
-    // FIXME: only needed by GeneralMatrixMatrixTriangular
-    InnerSize = EIGEN_SIZE_MIN_PREFER_FIXED(LhsTraits::ColsAtCompileTime, RhsTraits::RowsAtCompileTime),
-    
-    // The storage order is somewhat arbitrary here. The correct one will be determined through the evaluator.
-    Flags = (MaxRowsAtCompileTime==1 && MaxColsAtCompileTime!=1) ? RowMajorBit
-          : (MaxColsAtCompileTime==1 && MaxRowsAtCompileTime!=1) ? 0
-          : (   ((LhsTraits::Flags&NoPreferredStorageOrderBit) && (RhsTraits::Flags&RowMajorBit))
-             || ((RhsTraits::Flags&NoPreferredStorageOrderBit) && (LhsTraits::Flags&RowMajorBit)) ) ? RowMajorBit
-          : NoPreferredStorageOrderBit
-  };
-};
-
-} // end namespace internal
-
-/** \class Product
-  * \ingroup Core_Module
-  *
-  * \brief Expression of the product of two arbitrary matrices or vectors
-  *
-  * \tparam _Lhs the type of the left-hand side expression
-  * \tparam _Rhs the type of the right-hand side expression
-  *
-  * This class represents an expression of the product of two arbitrary matrices.
-  *
-  * The other template parameters are:
-  * \tparam Option     can be DefaultProduct, AliasFreeProduct, or LazyProduct
-  *
-  */
-template<typename _Lhs, typename _Rhs, int Option>
-class Product : public ProductImpl<_Lhs,_Rhs,Option,
-                                   typename internal::product_promote_storage_type<typename internal::traits<_Lhs>::StorageKind,
-                                                                                   typename internal::traits<_Rhs>::StorageKind,
-                                                                                   internal::product_type<_Lhs,_Rhs>::ret>::ret>
-{
-  public:
-    
-    typedef _Lhs Lhs;
-    typedef _Rhs Rhs;
-    
-    typedef typename ProductImpl<
-        Lhs, Rhs, Option,
-        typename internal::product_promote_storage_type<typename internal::traits<Lhs>::StorageKind,
-                                                        typename internal::traits<Rhs>::StorageKind,
-                                                        internal::product_type<Lhs,Rhs>::ret>::ret>::Base Base;
-    EIGEN_GENERIC_PUBLIC_INTERFACE(Product)
-
-    typedef typename internal::ref_selector<Lhs>::type LhsNested;
-    typedef typename internal::ref_selector<Rhs>::type RhsNested;
-    typedef typename internal::remove_all<LhsNested>::type LhsNestedCleaned;
-    typedef typename internal::remove_all<RhsNested>::type RhsNestedCleaned;
-
-    EIGEN_DEVICE_FUNC Product(const Lhs& lhs, const Rhs& rhs) : m_lhs(lhs), m_rhs(rhs)
-    {
-      eigen_assert(lhs.cols() == rhs.rows()
-        && "invalid matrix product"
-        && "if you wanted a coeff-wise or a dot product use the respective explicit functions");
-    }
-
-    EIGEN_DEVICE_FUNC inline Index rows() const { return m_lhs.rows(); }
-    EIGEN_DEVICE_FUNC inline Index cols() const { return m_rhs.cols(); }
-
-    EIGEN_DEVICE_FUNC const LhsNestedCleaned& lhs() const { return m_lhs; }
-    EIGEN_DEVICE_FUNC const RhsNestedCleaned& rhs() const { return m_rhs; }
-
-  protected:
-
-    LhsNested m_lhs;
-    RhsNested m_rhs;
-};
-
-namespace internal {
-  
-template<typename Lhs, typename Rhs, int Option, int ProductTag = internal::product_type<Lhs,Rhs>::ret>
-class dense_product_base
- : public internal::dense_xpr_base<Product<Lhs,Rhs,Option> >::type
-{};
-
-/** Convertion to scalar for inner-products */
-template<typename Lhs, typename Rhs, int Option>
-class dense_product_base<Lhs, Rhs, Option, InnerProduct>
- : public internal::dense_xpr_base<Product<Lhs,Rhs,Option> >::type
-{
-  typedef Product<Lhs,Rhs,Option> ProductXpr;
-  typedef typename internal::dense_xpr_base<ProductXpr>::type Base;
-public:
-  using Base::derived;
-  typedef typename Base::Scalar Scalar;
-  
-  operator const Scalar() const
-  {
-    return internal::evaluator<ProductXpr>(derived()).coeff(0,0);
-  }
-};
-
-} // namespace internal
-
-// Generic API dispatcher
-template<typename Lhs, typename Rhs, int Option, typename StorageKind>
-class ProductImpl : public internal::generic_xpr_base<Product<Lhs,Rhs,Option>, MatrixXpr, StorageKind>::type
-{
-  public:
-    typedef typename internal::generic_xpr_base<Product<Lhs,Rhs,Option>, MatrixXpr, StorageKind>::type Base;
-};
-
-template<typename Lhs, typename Rhs, int Option>
-class ProductImpl<Lhs,Rhs,Option,Dense>
-  : public internal::dense_product_base<Lhs,Rhs,Option>
-{
-    typedef Product<Lhs, Rhs, Option> Derived;
-    
-  public:
-    
-    typedef typename internal::dense_product_base<Lhs, Rhs, Option> Base;
-    EIGEN_DENSE_PUBLIC_INTERFACE(Derived)
-  protected:
-    enum {
-      IsOneByOne = (RowsAtCompileTime == 1 || RowsAtCompileTime == Dynamic) && 
-                   (ColsAtCompileTime == 1 || ColsAtCompileTime == Dynamic),
-      EnableCoeff = IsOneByOne || Option==LazyProduct
-    };
-    
-  public:
-  
-    EIGEN_DEVICE_FUNC Scalar coeff(Index row, Index col) const
-    {
-      EIGEN_STATIC_ASSERT(EnableCoeff, THIS_METHOD_IS_ONLY_FOR_INNER_OR_LAZY_PRODUCTS);
-      eigen_assert( (Option==LazyProduct) || (this->rows() == 1 && this->cols() == 1) );
-      
-      return internal::evaluator<Derived>(derived()).coeff(row,col);
-    }
-
-    EIGEN_DEVICE_FUNC Scalar coeff(Index i) const
-    {
-      EIGEN_STATIC_ASSERT(EnableCoeff, THIS_METHOD_IS_ONLY_FOR_INNER_OR_LAZY_PRODUCTS);
-      eigen_assert( (Option==LazyProduct) || (this->rows() == 1 && this->cols() == 1) );
-      
-      return internal::evaluator<Derived>(derived()).coeff(i);
-    }
-    
-  
-};
-
-} // end namespace Eigen
-
-#endif // EIGEN_PRODUCT_H

Eigen/src/Core/ProductBase.h

@@ -0,0 +1,290 @@
+// This file is part of Eigen, a lightweight C++ template library
+// for linear algebra.
+//
+// Copyright (C) 2009-2010 Gael Guennebaud <gael.guennebaud@inria.fr>
+//
+// This Source Code Form is subject to the terms of the Mozilla
+// Public License v. 2.0. If a copy of the MPL was not distributed
+// with this file, You can obtain one at http://mozilla.org/MPL/2.0/.
+
+#ifndef EIGEN_PRODUCTBASE_H
+#define EIGEN_PRODUCTBASE_H
+
+namespace Eigen { 
+
+/** \class ProductBase
+  * \ingroup Core_Module
+  *
+  */
+
+namespace internal {
+template<typename Derived, typename _Lhs, typename _Rhs>
+struct traits<ProductBase<Derived,_Lhs,_Rhs> >
+{
+  typedef MatrixXpr XprKind;
+  typedef typename remove_all<_Lhs>::type Lhs;
+  typedef typename remove_all<_Rhs>::type Rhs;
+  typedef typename scalar_product_traits<typename Lhs::Scalar, typename Rhs::Scalar>::ReturnType Scalar;
+  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;
+  enum {
+    RowsAtCompileTime = traits<Lhs>::RowsAtCompileTime,
+    ColsAtCompileTime = traits<Rhs>::ColsAtCompileTime,
+    MaxRowsAtCompileTime = traits<Lhs>::MaxRowsAtCompileTime,
+    MaxColsAtCompileTime = traits<Rhs>::MaxColsAtCompileTime,
+    Flags = (MaxRowsAtCompileTime==1 ? RowMajorBit : 0)
+          | EvalBeforeNestingBit | EvalBeforeAssigningBit | NestByRefBit,
+                  // Note that EvalBeforeNestingBit and NestByRefBit
+                  // are not used in practice because nested is overloaded for products
+    CoeffReadCost = 0 // FIXME why is it needed ?
+  };
+};
+}
+
+#define EIGEN_PRODUCT_PUBLIC_INTERFACE(Derived) \
+  typedef ProductBase<Derived, Lhs, Rhs > Base; \
+  EIGEN_DENSE_PUBLIC_INTERFACE(Derived) \
+  typedef typename Base::LhsNested LhsNested; \
+  typedef typename Base::_LhsNested _LhsNested; \
+  typedef typename Base::LhsBlasTraits LhsBlasTraits; \
+  typedef typename Base::ActualLhsType ActualLhsType; \
+  typedef typename Base::_ActualLhsType _ActualLhsType; \
+  typedef typename Base::RhsNested RhsNested; \
+  typedef typename Base::_RhsNested _RhsNested; \
+  typedef typename Base::RhsBlasTraits RhsBlasTraits; \
+  typedef typename Base::ActualRhsType ActualRhsType; \
+  typedef typename Base::_ActualRhsType _ActualRhsType; \
+  using Base::m_lhs; \
+  using Base::m_rhs;
+
+template<typename Derived, typename Lhs, typename Rhs>
+class ProductBase : public MatrixBase<Derived>
+{
+  public:
+    typedef MatrixBase<Derived> Base;
+    EIGEN_DENSE_PUBLIC_INTERFACE(ProductBase)
+    
+    typedef typename Lhs::Nested LhsNested;
+    typedef typename internal::remove_all<LhsNested>::type _LhsNested;
+    typedef internal::blas_traits<_LhsNested> LhsBlasTraits;
+    typedef typename LhsBlasTraits::DirectLinearAccessType ActualLhsType;
+    typedef typename internal::remove_all<ActualLhsType>::type _ActualLhsType;
+    typedef typename internal::traits<Lhs>::Scalar LhsScalar;
+
+    typedef typename Rhs::Nested RhsNested;
+    typedef typename internal::remove_all<RhsNested>::type _RhsNested;
+    typedef internal::blas_traits<_RhsNested> RhsBlasTraits;
+    typedef typename RhsBlasTraits::DirectLinearAccessType ActualRhsType;
+    typedef typename internal::remove_all<ActualRhsType>::type _ActualRhsType;
+    typedef typename internal::traits<Rhs>::Scalar RhsScalar;
+
+    // Diagonal of a product: no need to evaluate the arguments because they are going to be evaluated only once
+    typedef CoeffBasedProduct<LhsNested, RhsNested, 0> FullyLazyCoeffBaseProductType;
+
+  public:
+
+#ifndef EIGEN_NO_MALLOC
+    typedef typename Base::PlainObject BasePlainObject;
+    typedef Matrix<Scalar,RowsAtCompileTime==1?1:Dynamic,ColsAtCompileTime==1?1:Dynamic,BasePlainObject::Options> DynPlainObject;
+    typedef typename internal::conditional<(BasePlainObject::SizeAtCompileTime==Dynamic) || (BasePlainObject::SizeAtCompileTime*int(sizeof(Scalar)) < int(EIGEN_STACK_ALLOCATION_LIMIT)),
+                                           BasePlainObject, DynPlainObject>::type PlainObject;
+#else
+    typedef typename Base::PlainObject PlainObject;
+#endif
+
+    ProductBase(const Lhs& a_lhs, const Rhs& a_rhs)
+      : m_lhs(a_lhs), m_rhs(a_rhs)
+    {
+      eigen_assert(a_lhs.cols() == a_rhs.rows()
+        && "invalid matrix product"
+        && "if you wanted a coeff-wise or a dot product use the respective explicit functions");
+    }
+
+    inline Index rows() const { return m_lhs.rows(); }
+    inline Index cols() const { return m_rhs.cols(); }
+
+    template<typename Dest>
+    inline void evalTo(Dest& dst) const { dst.setZero(); scaleAndAddTo(dst,Scalar(1)); }
+
+    template<typename Dest>
+    inline void addTo(Dest& dst) const { scaleAndAddTo(dst,Scalar(1)); }
+
+    template<typename Dest>
+    inline void subTo(Dest& dst) const { scaleAndAddTo(dst,Scalar(-1)); }
+
+    template<typename Dest>
+    inline void scaleAndAddTo(Dest& dst, const Scalar& alpha) const { derived().scaleAndAddTo(dst,alpha); }
+
+    const _LhsNested& lhs() const { return m_lhs; }
+    const _RhsNested& rhs() const { return m_rhs; }
+
+    // Implicit conversion to the nested type (trigger the evaluation of the product)
+    operator const PlainObject& () const
+    {
+      m_result.resize(m_lhs.rows(), m_rhs.cols());
+      derived().evalTo(m_result);
+      return m_result;
+    }
+
+    const Diagonal<const FullyLazyCoeffBaseProductType,0> diagonal() const
+    { return FullyLazyCoeffBaseProductType(m_lhs, m_rhs); }
+
+    template<int Index>
+    const Diagonal<FullyLazyCoeffBaseProductType,Index> diagonal() const
+    { return FullyLazyCoeffBaseProductType(m_lhs, m_rhs); }
+
+    const Diagonal<FullyLazyCoeffBaseProductType,Dynamic> diagonal(Index index) const
+    { return FullyLazyCoeffBaseProductType(m_lhs, m_rhs).diagonal(index); }
+
+    // restrict coeff accessors to 1x1 expressions. No need to care about mutators here since this isnt a Lvalue expression
+    typename Base::CoeffReturnType coeff(Index row, Index col) const
+    {
+#ifdef EIGEN2_SUPPORT
+      return lhs().row(row).cwiseProduct(rhs().col(col).transpose()).sum();
+#else
+      EIGEN_STATIC_ASSERT_SIZE_1x1(Derived)
+      eigen_assert(this->rows() == 1 && this->cols() == 1);
+      Matrix<Scalar,1,1> result = *this;
+      return result.coeff(row,col);
+#endif
+    }
+
+    typename Base::CoeffReturnType coeff(Index i) const
+    {
+      EIGEN_STATIC_ASSERT_SIZE_1x1(Derived)
+      eigen_assert(this->rows() == 1 && this->cols() == 1);
+      Matrix<Scalar,1,1> result = *this;
+      return result.coeff(i);
+    }
+
+    const Scalar& coeffRef(Index row, Index col) const
+    {
+      EIGEN_STATIC_ASSERT_SIZE_1x1(Derived)
+      eigen_assert(this->rows() == 1 && this->cols() == 1);
+      return derived().coeffRef(row,col);
+    }
+
+    const Scalar& coeffRef(Index i) const
+    {
+      EIGEN_STATIC_ASSERT_SIZE_1x1(Derived)
+      eigen_assert(this->rows() == 1 && this->cols() == 1);
+      return derived().coeffRef(i);
+    }
+
+  protected:
+
+    LhsNested m_lhs;
+    RhsNested m_rhs;
+
+    mutable PlainObject m_result;
+};
+
+// here we need to overload the nested rule for products
+// such that the nested type is a const reference to a plain matrix
+namespace internal {
+template<typename Lhs, typename Rhs, int Mode, int N, typename PlainObject>
+struct nested<GeneralProduct<Lhs,Rhs,Mode>, N, PlainObject>
+{
+  typedef typename GeneralProduct<Lhs,Rhs,Mode>::PlainObject const& type;
+};
+template<typename Lhs, typename Rhs, int Mode, int N, typename PlainObject>
+struct nested<const GeneralProduct<Lhs,Rhs,Mode>, N, PlainObject>
+{
+  typedef typename GeneralProduct<Lhs,Rhs,Mode>::PlainObject const& type;
+};
+}
+
+template<typename NestedProduct>
+class ScaledProduct;
+
+// Note that these two operator* functions are not defined as member
+// functions of ProductBase, because, otherwise we would have to
+// define all overloads defined in MatrixBase. Furthermore, Using
+// "using Base::operator*" would not work with MSVC.
+//
+// Also note that here we accept any compatible scalar types
+template<typename Derived,typename Lhs,typename Rhs>
+const ScaledProduct<Derived>
+operator*(const ProductBase<Derived,Lhs,Rhs>& prod, const typename Derived::Scalar& x)
+{ return ScaledProduct<Derived>(prod.derived(), x); }
+
+template<typename Derived,typename Lhs,typename Rhs>
+typename internal::enable_if<!internal::is_same<typename Derived::Scalar,typename Derived::RealScalar>::value,
+                      const ScaledProduct<Derived> >::type
+operator*(const ProductBase<Derived,Lhs,Rhs>& prod, const typename Derived::RealScalar& x)
+{ return ScaledProduct<Derived>(prod.derived(), x); }
+
+
+template<typename Derived,typename Lhs,typename Rhs>
+const ScaledProduct<Derived>
+operator*(const typename Derived::Scalar& x,const ProductBase<Derived,Lhs,Rhs>& prod)
+{ return ScaledProduct<Derived>(prod.derived(), x); }
+
+template<typename Derived,typename Lhs,typename Rhs>
+typename internal::enable_if<!internal::is_same<typename Derived::Scalar,typename Derived::RealScalar>::value,
+                      const ScaledProduct<Derived> >::type
+operator*(const typename Derived::RealScalar& x,const ProductBase<Derived,Lhs,Rhs>& prod)
+{ return ScaledProduct<Derived>(prod.derived(), x); }
+
+namespace internal {
+template<typename NestedProduct>
+struct traits<ScaledProduct<NestedProduct> >
+ : traits<ProductBase<ScaledProduct<NestedProduct>,
+                         typename NestedProduct::_LhsNested,
+                         typename NestedProduct::_RhsNested> >
+{
+  typedef typename traits<NestedProduct>::StorageKind StorageKind;
+};
+}
+
+template<typename NestedProduct>
+class ScaledProduct
+  : public ProductBase<ScaledProduct<NestedProduct>,
+                       typename NestedProduct::_LhsNested,
+                       typename NestedProduct::_RhsNested>
+{
+  public:
+    typedef ProductBase<ScaledProduct<NestedProduct>,
+                       typename NestedProduct::_LhsNested,
+                       typename NestedProduct::_RhsNested> Base;
+    typedef typename Base::Scalar Scalar;
+    typedef typename Base::PlainObject PlainObject;
+//     EIGEN_PRODUCT_PUBLIC_INTERFACE(ScaledProduct)
+
+    ScaledProduct(const NestedProduct& prod, const Scalar& x)
+    : Base(prod.lhs(),prod.rhs()), m_prod(prod), m_alpha(x) {}
+
+    template<typename Dest>
+    inline void evalTo(Dest& dst) const { dst.setZero(); scaleAndAddTo(dst, Scalar(1)); }
+
+    template<typename Dest>
+    inline void addTo(Dest& dst) const { scaleAndAddTo(dst, Scalar(1)); }
+
+    template<typename Dest>
+    inline void subTo(Dest& dst) const { scaleAndAddTo(dst, Scalar(-1)); }
+
+    template<typename Dest>
+    inline void scaleAndAddTo(Dest& dst, const Scalar& a_alpha) const { m_prod.derived().scaleAndAddTo(dst,a_alpha * m_alpha); }
+
+    const Scalar& alpha() const { return m_alpha; }
+    
+  protected:
+    const NestedProduct& m_prod;
+    Scalar m_alpha;
+};
+
+/** \internal
+  * Overloaded to perform an efficient C = (A*B).lazy() */
+template<typename Derived>
+template<typename ProductDerived, typename Lhs, typename Rhs>
+Derived& MatrixBase<Derived>::lazyAssign(const ProductBase<ProductDerived, Lhs,Rhs>& other)
+{
+  other.derived().evalTo(derived());
+  return derived();
+}
+
+} // end namespace Eigen
+
+#endif // EIGEN_PRODUCTBASE_H

Eigen/src/Core/Random.h

@@ -16,7 +16,8 @@ namespace internal {
 
 template<typename Scalar> struct scalar_random_op {
   EIGEN_EMPTY_STRUCT_CTOR(scalar_random_op)
-  inline const Scalar operator() () const { return random<Scalar>(); }
+  template<typename Index>
+  inline const Scalar operator() (Index, Index = 0) const { return random<Scalar>(); }
 };
 
 template<typename Scalar>
@@ -27,18 +28,12 @@ struct functor_traits<scalar_random_op<Scalar> >
 
 /** \returns a random matrix expression
   *
-  * Numbers are uniformly spread through their whole definition range for integer types,
-  * and in the [-1:1] range for floating point scalar types.
-  * 
   * The parameters \a rows and \a cols are the number of rows and of columns of
   * the returned matrix. Must be compatible with this MatrixBase type.
   *
-  * \not_reentrant
-  * 
   * This variant is meant to be used for dynamic-size matrix types. For fixed-size types,
   * it is redundant to pass \a rows and \a cols as arguments, so Random() should be used
   * instead.
-  * 
   *
   * Example: \include MatrixBase_random_int_int.cpp
   * Output: \verbinclude MatrixBase_random_int_int.out
@@ -46,28 +41,22 @@ struct functor_traits<scalar_random_op<Scalar> >
   * This expression has the "evaluate before nesting" flag so that it will be evaluated into
   * a temporary matrix whenever it is nested in a larger expression. This prevents unexpected
   * behavior with expressions involving random matrices.
-  * 
-  * See DenseBase::NullaryExpr(Index, const CustomNullaryOp&) for an example using C++11 random generators.
   *
-  * \sa DenseBase::setRandom(), DenseBase::Random(Index), DenseBase::Random()
+  * \sa MatrixBase::setRandom(), MatrixBase::Random(Index), MatrixBase::Random()
   */
 template<typename Derived>
-inline const typename DenseBase<Derived>::RandomReturnType
+inline const CwiseNullaryOp<internal::scalar_random_op<typename internal::traits<Derived>::Scalar>, Derived>
 DenseBase<Derived>::Random(Index rows, Index cols)
 {
   return NullaryExpr(rows, cols, internal::scalar_random_op<Scalar>());
 }
 
 /** \returns a random vector expression
-  *
-  * Numbers are uniformly spread through their whole definition range for integer types,
-  * and in the [-1:1] range for floating point scalar types.
   *
   * The parameter \a size is the size of the returned vector.
   * Must be compatible with this MatrixBase type.
   *
   * \only_for_vectors
-  * \not_reentrant
   *
   * This variant is meant to be used for dynamic-size vector types. For fixed-size types,
   * it is redundant to pass \a size as argument, so Random() should be used
@@ -80,10 +69,10 @@ DenseBase<Derived>::Random(Index rows, Index cols)
   * a temporary vector whenever it is nested in a larger expression. This prevents unexpected
   * behavior with expressions involving random matrices.
   *
-  * \sa DenseBase::setRandom(), DenseBase::Random(Index,Index), DenseBase::Random()
+  * \sa MatrixBase::setRandom(), MatrixBase::Random(Index,Index), MatrixBase::Random()
   */
 template<typename Derived>
-inline const typename DenseBase<Derived>::RandomReturnType
+inline const CwiseNullaryOp<internal::scalar_random_op<typename internal::traits<Derived>::Scalar>, Derived>
 DenseBase<Derived>::Random(Index size)
 {
   return NullaryExpr(size, internal::scalar_random_op<Scalar>());
@@ -91,9 +80,6 @@ DenseBase<Derived>::Random(Index size)
 
 /** \returns a fixed-size random matrix or vector expression
   *
-  * Numbers are uniformly spread through their whole definition range for integer types,
-  * and in the [-1:1] range for floating point scalar types.
-  * 
   * This variant is only for fixed-size MatrixBase types. For dynamic-size types, you
   * need to use the variants taking size arguments.
   *
@@ -103,13 +89,11 @@ DenseBase<Derived>::Random(Index size)
   * This expression has the "evaluate before nesting" flag so that it will be evaluated into
   * a temporary matrix whenever it is nested in a larger expression. This prevents unexpected
   * behavior with expressions involving random matrices.
-  * 
-  * \not_reentrant
   *
-  * \sa DenseBase::setRandom(), DenseBase::Random(Index,Index), DenseBase::Random(Index)
+  * \sa MatrixBase::setRandom(), MatrixBase::Random(Index,Index), MatrixBase::Random(Index)
   */
 template<typename Derived>
-inline const typename DenseBase<Derived>::RandomReturnType
+inline const CwiseNullaryOp<internal::scalar_random_op<typename internal::traits<Derived>::Scalar>, Derived>
 DenseBase<Derived>::Random()
 {
   return NullaryExpr(RowsAtCompileTime, ColsAtCompileTime, internal::scalar_random_op<Scalar>());
@@ -117,11 +101,6 @@ DenseBase<Derived>::Random()
 
 /** Sets all coefficients in this expression to random values.
   *
-  * Numbers are uniformly spread through their whole definition range for integer types,
-  * and in the [-1:1] range for floating point scalar types.
-  * 
-  * \not_reentrant
-  * 
   * Example: \include MatrixBase_setRandom.cpp
   * Output: \verbinclude MatrixBase_setRandom.out
   *
@@ -135,16 +114,12 @@ inline Derived& DenseBase<Derived>::setRandom()
 
 /** Resizes to the given \a newSize, and sets all coefficients in this expression to random values.
   *
-  * Numbers are uniformly spread through their whole definition range for integer types,
-  * and in the [-1:1] range for floating point scalar types.
-  * 
   * \only_for_vectors
-  * \not_reentrant
   *
   * Example: \include Matrix_setRandom_int.cpp
   * Output: \verbinclude Matrix_setRandom_int.out
   *
-  * \sa DenseBase::setRandom(), setRandom(Index,Index), class CwiseNullaryOp, DenseBase::Random()
+  * \sa MatrixBase::setRandom(), setRandom(Index,Index), class CwiseNullaryOp, MatrixBase::Random()
   */
 template<typename Derived>
 EIGEN_STRONG_INLINE Derived&
@@ -156,24 +131,19 @@ PlainObjectBase<Derived>::setRandom(Index newSize)
 
 /** Resizes to the given size, and sets all coefficients in this expression to random values.
   *
-  * Numbers are uniformly spread through their whole definition range for integer types,
-  * and in the [-1:1] range for floating point scalar types.
-  *
-  * \not_reentrant
-  * 
-  * \param rows the new number of rows
-  * \param cols the new number of columns
+  * \param nbRows the new number of rows
+  * \param nbCols the new number of columns
   *
   * Example: \include Matrix_setRandom_int_int.cpp
   * Output: \verbinclude Matrix_setRandom_int_int.out
   *
-  * \sa DenseBase::setRandom(), setRandom(Index), class CwiseNullaryOp, DenseBase::Random()
+  * \sa MatrixBase::setRandom(), setRandom(Index), class CwiseNullaryOp, MatrixBase::Random()
   */
 template<typename Derived>
 EIGEN_STRONG_INLINE Derived&
-PlainObjectBase<Derived>::setRandom(Index rows, Index cols)
+PlainObjectBase<Derived>::setRandom(Index nbRows, Index nbCols)
 {
-  resize(rows, cols);
+  resize(nbRows, nbCols);
   return setRandom();
 }
 

Eigen/src/Core/Redux.h

@@ -27,9 +27,8 @@ template<typename Func, typename Derived>
 struct redux_traits
 {
 public:
-    typedef typename find_best_packet<typename Derived::Scalar,Derived::SizeAtCompileTime>::type PacketType;
   enum {
-    PacketSize = unpacket_traits<PacketType>::size,
+    PacketSize = packet_traits<typename Derived::Scalar>::size,
     InnerMaxSize = int(Derived::IsRowMajor)
                  ? Derived::MaxColsAtCompileTime
                  : Derived::MaxRowsAtCompileTime
@@ -38,8 +37,8 @@ public:
   enum {
     MightVectorize = (int(Derived::Flags)&ActualPacketAccessBit)
                   && (functor_traits<Func>::PacketAccess),
-    MayLinearVectorize = bool(MightVectorize) && (int(Derived::Flags)&LinearAccessBit),
-    MaySliceVectorize  = bool(MightVectorize) && int(InnerMaxSize)>=3*PacketSize
+    MayLinearVectorize = MightVectorize && (int(Derived::Flags)&LinearAccessBit),
+    MaySliceVectorize  = MightVectorize && int(InnerMaxSize)>=3*PacketSize
   };
 
 public:
@@ -51,34 +50,21 @@ public:
 
 public:
   enum {
-    Cost = Derived::SizeAtCompileTime == Dynamic ? HugeCost
-         : Derived::SizeAtCompileTime * Derived::CoeffReadCost + (Derived::SizeAtCompileTime-1) * functor_traits<Func>::Cost,
+    Cost = (  Derived::SizeAtCompileTime == Dynamic
+           || Derived::CoeffReadCost == Dynamic
+           || (Derived::SizeAtCompileTime!=1 && functor_traits<Func>::Cost == Dynamic)
+           ) ? Dynamic
+           : Derived::SizeAtCompileTime * Derived::CoeffReadCost
+               + (Derived::SizeAtCompileTime-1) * functor_traits<Func>::Cost,
     UnrollingLimit = EIGEN_UNROLLING_LIMIT * (int(Traversal) == int(DefaultTraversal) ? 1 : int(PacketSize))
   };
 
 public:
   enum {
-    Unrolling = Cost <= UnrollingLimit ? CompleteUnrolling : NoUnrolling
+    Unrolling = Cost != Dynamic && Cost <= UnrollingLimit
+              ? CompleteUnrolling
+              : NoUnrolling
   };
-  
-#ifdef EIGEN_DEBUG_ASSIGN
-  static void debug()
-  {
-    std::cerr << "Xpr: " << typeid(typename Derived::XprType).name() << std::endl;
-    std::cerr.setf(std::ios::hex, std::ios::basefield);
-    EIGEN_DEBUG_VAR(Derived::Flags)
-    std::cerr.unsetf(std::ios::hex);
-    EIGEN_DEBUG_VAR(InnerMaxSize)
-    EIGEN_DEBUG_VAR(PacketSize)
-    EIGEN_DEBUG_VAR(MightVectorize)
-    EIGEN_DEBUG_VAR(MayLinearVectorize)
-    EIGEN_DEBUG_VAR(MaySliceVectorize)
-    EIGEN_DEBUG_VAR(Traversal)
-    EIGEN_DEBUG_VAR(UnrollingLimit)
-    EIGEN_DEBUG_VAR(Unrolling)
-    std::cerr << std::endl;
-  }
-#endif
 };
 
 /***************************************************************************
@@ -96,7 +82,6 @@ struct redux_novec_unroller
 
   typedef typename Derived::Scalar Scalar;
 
-  EIGEN_DEVICE_FUNC
   static EIGEN_STRONG_INLINE Scalar run(const Derived &mat, const Func& func)
   {
     return func(redux_novec_unroller<Func, Derived, Start, HalfLength>::run(mat,func),
@@ -114,7 +99,6 @@ struct redux_novec_unroller<Func, Derived, Start, 1>
 
   typedef typename Derived::Scalar Scalar;
 
-  EIGEN_DEVICE_FUNC
   static EIGEN_STRONG_INLINE Scalar run(const Derived &mat, const Func&)
   {
     return mat.coeffByOuterInner(outer, inner);
@@ -128,7 +112,6 @@ template<typename Func, typename Derived, int Start>
 struct redux_novec_unroller<Func, Derived, Start, 0>
 {
   typedef typename Derived::Scalar Scalar;
-  EIGEN_DEVICE_FUNC 
   static EIGEN_STRONG_INLINE Scalar run(const Derived&, const Func&) { return Scalar(); }
 };
 
@@ -138,12 +121,12 @@ template<typename Func, typename Derived, int Start, int Length>
 struct redux_vec_unroller
 {
   enum {
-    PacketSize = redux_traits<Func, Derived>::PacketSize,
+    PacketSize = packet_traits<typename Derived::Scalar>::size,
     HalfLength = Length/2
   };
 
   typedef typename Derived::Scalar Scalar;
-  typedef typename redux_traits<Func, Derived>::PacketType PacketScalar;
+  typedef typename packet_traits<Scalar>::type PacketScalar;
 
   static EIGEN_STRONG_INLINE PacketScalar run(const Derived &mat, const Func& func)
   {
@@ -157,18 +140,18 @@ template<typename Func, typename Derived, int Start>
 struct redux_vec_unroller<Func, Derived, Start, 1>
 {
   enum {
-    index = Start * redux_traits<Func, Derived>::PacketSize,
+    index = Start * packet_traits<typename Derived::Scalar>::size,
     outer = index / int(Derived::InnerSizeAtCompileTime),
     inner = index % int(Derived::InnerSizeAtCompileTime),
-    alignment = Derived::Alignment
+    alignment = (Derived::Flags & AlignedBit) ? Aligned : Unaligned
   };
 
   typedef typename Derived::Scalar Scalar;
-  typedef typename redux_traits<Func, Derived>::PacketType PacketScalar;
+  typedef typename packet_traits<Scalar>::type PacketScalar;
 
   static EIGEN_STRONG_INLINE PacketScalar run(const Derived &mat, const Func&)
   {
-    return mat.template packetByOuterInner<alignment,PacketScalar>(outer, inner);
+    return mat.template packetByOuterInner<alignment>(outer, inner);
   }
 };
 
@@ -186,8 +169,8 @@ template<typename Func, typename Derived>
 struct redux_impl<Func, Derived, DefaultTraversal, NoUnrolling>
 {
   typedef typename Derived::Scalar Scalar;
-  EIGEN_DEVICE_FUNC
-  static EIGEN_STRONG_INLINE Scalar run(const Derived &mat, const Func& func)
+  typedef typename Derived::Index Index;
+  static EIGEN_STRONG_INLINE Scalar run(const Derived& mat, const Func& func)
   {
     eigen_assert(mat.rows()>0 && mat.cols()>0 && "you are using an empty matrix");
     Scalar res;
@@ -210,19 +193,19 @@ template<typename Func, typename Derived>
 struct redux_impl<Func, Derived, LinearVectorizedTraversal, NoUnrolling>
 {
   typedef typename Derived::Scalar Scalar;
-  typedef typename redux_traits<Func, Derived>::PacketType PacketScalar;
+  typedef typename packet_traits<Scalar>::type PacketScalar;
+  typedef typename Derived::Index Index;
 
-  static Scalar run(const Derived &mat, const Func& func)
+  static Scalar run(const Derived& mat, const Func& func)
   {
     const Index size = mat.size();
-    
-    const Index packetSize = redux_traits<Func, Derived>::PacketSize;
-    const int packetAlignment = unpacket_traits<PacketScalar>::alignment;
+    eigen_assert(size && "you are using an empty matrix");
+    const Index packetSize = packet_traits<Scalar>::size;
+    const Index alignedStart = internal::first_aligned(mat);
     enum {
-      alignment0 = (bool(Derived::Flags & DirectAccessBit) && bool(packet_traits<Scalar>::AlignedOnScalar)) ? int(packetAlignment) : int(Unaligned),
-      alignment = EIGEN_PLAIN_ENUM_MAX(alignment0, Derived::Alignment)
+      alignment = bool(Derived::Flags & DirectAccessBit) || bool(Derived::Flags & AlignedBit)
+                ? Aligned : Unaligned
     };
-    const Index alignedStart = internal::first_default_aligned(mat.nestedExpression());
     const Index alignedSize2 = ((size-alignedStart)/(2*packetSize))*(2*packetSize);
     const Index alignedSize = ((size-alignedStart)/(packetSize))*(packetSize);
     const Index alignedEnd2 = alignedStart + alignedSize2;
@@ -230,19 +213,19 @@ struct redux_impl<Func, Derived, LinearVectorizedTraversal, NoUnrolling>
     Scalar res;
     if(alignedSize)
     {
-      PacketScalar packet_res0 = mat.template packet<alignment,PacketScalar>(alignedStart);
+      PacketScalar packet_res0 = mat.template packet<alignment>(alignedStart);
       if(alignedSize>packetSize) // we have at least two packets to partly unroll the loop
       {
-        PacketScalar packet_res1 = mat.template packet<alignment,PacketScalar>(alignedStart+packetSize);
+        PacketScalar packet_res1 = mat.template packet<alignment>(alignedStart+packetSize);
         for(Index index = alignedStart + 2*packetSize; index < alignedEnd2; index += 2*packetSize)
         {
-          packet_res0 = func.packetOp(packet_res0, mat.template packet<alignment,PacketScalar>(index));
-          packet_res1 = func.packetOp(packet_res1, mat.template packet<alignment,PacketScalar>(index+packetSize));
+          packet_res0 = func.packetOp(packet_res0, mat.template packet<alignment>(index));
+          packet_res1 = func.packetOp(packet_res1, mat.template packet<alignment>(index+packetSize));
         }
 
         packet_res0 = func.packetOp(packet_res0,packet_res1);
         if(alignedEnd>alignedEnd2)
-          packet_res0 = func.packetOp(packet_res0, mat.template packet<alignment,PacketScalar>(alignedEnd2));
+          packet_res0 = func.packetOp(packet_res0, mat.template packet<alignment>(alignedEnd2));
       }
       res = func.predux(packet_res0);
 
@@ -269,24 +252,25 @@ template<typename Func, typename Derived, int Unrolling>
 struct redux_impl<Func, Derived, SliceVectorizedTraversal, Unrolling>
 {
   typedef typename Derived::Scalar Scalar;
-  typedef typename redux_traits<Func, Derived>::PacketType PacketType;
+  typedef typename packet_traits<Scalar>::type PacketScalar;
+  typedef typename Derived::Index Index;
 
-  EIGEN_DEVICE_FUNC static Scalar run(const Derived &mat, const Func& func)
+  static Scalar run(const Derived& mat, const Func& func)
   {
     eigen_assert(mat.rows()>0 && mat.cols()>0 && "you are using an empty matrix");
     const Index innerSize = mat.innerSize();
     const Index outerSize = mat.outerSize();
     enum {
-      packetSize = redux_traits<Func, Derived>::PacketSize
+      packetSize = packet_traits<Scalar>::size
     };
     const Index packetedInnerSize = ((innerSize)/packetSize)*packetSize;
     Scalar res;
     if(packetedInnerSize)
     {
-      PacketType packet_res = mat.template packet<Unaligned,PacketType>(0,0);
+      PacketScalar packet_res = mat.template packet<Unaligned>(0,0);
       for(Index j=0; j<outerSize; ++j)
         for(Index i=(j==0?packetSize:0); i<packetedInnerSize; i+=Index(packetSize))
-          packet_res = func.packetOp(packet_res, mat.template packetByOuterInner<Unaligned,PacketType>(j,i));
+          packet_res = func.packetOp(packet_res, mat.template packetByOuterInner<Unaligned>(j,i));
 
       res = func.predux(packet_res);
       for(Index j=0; j<outerSize; ++j)
@@ -307,90 +291,22 @@ template<typename Func, typename Derived>
 struct redux_impl<Func, Derived, LinearVectorizedTraversal, CompleteUnrolling>
 {
   typedef typename Derived::Scalar Scalar;
-
-  typedef typename redux_traits<Func, Derived>::PacketType PacketScalar;
+  typedef typename packet_traits<Scalar>::type PacketScalar;
   enum {
-    PacketSize = redux_traits<Func, Derived>::PacketSize,
+    PacketSize = packet_traits<Scalar>::size,
     Size = Derived::SizeAtCompileTime,
     VectorizedSize = (Size / PacketSize) * PacketSize
   };
-  EIGEN_DEVICE_FUNC static EIGEN_STRONG_INLINE Scalar run(const Derived &mat, const Func& func)
+  static EIGEN_STRONG_INLINE Scalar run(const Derived& mat, const Func& func)
   {
     eigen_assert(mat.rows()>0 && mat.cols()>0 && "you are using an empty matrix");
-    if (VectorizedSize > 0) {
-      Scalar res = func.predux(redux_vec_unroller<Func, Derived, 0, Size / PacketSize>::run(mat,func));
-      if (VectorizedSize != Size)
-        res = func(res,redux_novec_unroller<Func, Derived, VectorizedSize, Size-VectorizedSize>::run(mat,func));
-      return res;
-    }
-    else {
-      return redux_novec_unroller<Func, Derived, 0, Size>::run(mat,func);
-    }
+    Scalar res = func.predux(redux_vec_unroller<Func, Derived, 0, Size / PacketSize>::run(mat,func));
+    if (VectorizedSize != Size)
+      res = func(res,redux_novec_unroller<Func, Derived, VectorizedSize, Size-VectorizedSize>::run(mat,func));
+    return res;
   }
 };
 
-// evaluator adaptor
-template<typename _XprType>
-class redux_evaluator
-{
-public:
-  typedef _XprType XprType;
-  EIGEN_DEVICE_FUNC explicit redux_evaluator(const XprType &xpr) : m_evaluator(xpr), m_xpr(xpr) {}
-  
-  typedef typename XprType::Scalar Scalar;
-  typedef typename XprType::CoeffReturnType CoeffReturnType;
-  typedef typename XprType::PacketScalar PacketScalar;
-  typedef typename XprType::PacketReturnType PacketReturnType;
-  
-  enum {
-    MaxRowsAtCompileTime = XprType::MaxRowsAtCompileTime,
-    MaxColsAtCompileTime = XprType::MaxColsAtCompileTime,
-    // TODO we should not remove DirectAccessBit and rather find an elegant way to query the alignment offset at runtime from the evaluator
-    Flags = evaluator<XprType>::Flags & ~DirectAccessBit,
-    IsRowMajor = XprType::IsRowMajor,
-    SizeAtCompileTime = XprType::SizeAtCompileTime,
-    InnerSizeAtCompileTime = XprType::InnerSizeAtCompileTime,
-    CoeffReadCost = evaluator<XprType>::CoeffReadCost,
-    Alignment = evaluator<XprType>::Alignment
-  };
-  
-  EIGEN_DEVICE_FUNC Index rows() const { return m_xpr.rows(); }
-  EIGEN_DEVICE_FUNC Index cols() const { return m_xpr.cols(); }
-  EIGEN_DEVICE_FUNC Index size() const { return m_xpr.size(); }
-  EIGEN_DEVICE_FUNC Index innerSize() const { return m_xpr.innerSize(); }
-  EIGEN_DEVICE_FUNC Index outerSize() const { return m_xpr.outerSize(); }
-
-  EIGEN_DEVICE_FUNC
-  CoeffReturnType coeff(Index row, Index col) const
-  { return m_evaluator.coeff(row, col); }
-
-  EIGEN_DEVICE_FUNC
-  CoeffReturnType coeff(Index index) const
-  { return m_evaluator.coeff(index); }
-
-  template<int LoadMode, typename PacketType>
-  PacketType packet(Index row, Index col) const
-  { return m_evaluator.template packet<LoadMode,PacketType>(row, col); }
-
-  template<int LoadMode, typename PacketType>
-  PacketType packet(Index index) const
-  { return m_evaluator.template packet<LoadMode,PacketType>(index); }
-  
-  EIGEN_DEVICE_FUNC
-  CoeffReturnType coeffByOuterInner(Index outer, Index inner) const
-  { return m_evaluator.coeff(IsRowMajor ? outer : inner, IsRowMajor ? inner : outer); }
-  
-  template<int LoadMode, typename PacketType>
-  PacketType packetByOuterInner(Index outer, Index inner) const
-  { return m_evaluator.template packet<LoadMode,PacketType>(IsRowMajor ? outer : inner, IsRowMajor ? inner : outer); }
-  
-  const XprType & nestedExpression() const { return m_xpr; }
-  
-protected:
-  internal::evaluator<XprType> m_evaluator;
-  const XprType &m_xpr;
-};
-
 } // end namespace internal
 
 /***************************************************************************
@@ -401,21 +317,18 @@ protected:
 /** \returns the result of a full redux operation on the whole matrix or vector using \a func
   *
   * The template parameter \a BinaryOp is the type of the functor \a func which must be
-  * an associative operator. Both current C++98 and C++11 functor styles are handled.
+  * an associative operator. Both current STL and TR1 functor styles are handled.
   *
   * \sa DenseBase::sum(), DenseBase::minCoeff(), DenseBase::maxCoeff(), MatrixBase::colwise(), MatrixBase::rowwise()
   */
 template<typename Derived>
 template<typename Func>
-typename internal::traits<Derived>::Scalar
+EIGEN_STRONG_INLINE typename internal::result_of<Func(typename internal::traits<Derived>::Scalar)>::type
 DenseBase<Derived>::redux(const Func& func) const
 {
-  eigen_assert(this->rows()>0 && this->cols()>0 && "you are using an empty matrix");
-
-  typedef typename internal::redux_evaluator<Derived> ThisEvaluator;
-  ThisEvaluator thisEval(derived());
-  
-  return internal::redux_impl<Func, ThisEvaluator>::run(thisEval, func);
+  typedef typename internal::remove_all<typename Derived::Nested>::type ThisNested;
+  return internal::redux_impl<Func, ThisNested>
+            ::run(derived(), func);
 }
 
 /** \returns the minimum of all coefficients of \c *this.
@@ -425,7 +338,7 @@ template<typename Derived>
 EIGEN_STRONG_INLINE typename internal::traits<Derived>::Scalar
 DenseBase<Derived>::minCoeff() const
 {
-  return derived().redux(Eigen::internal::scalar_min_op<Scalar,Scalar>());
+  return this->redux(Eigen::internal::scalar_min_op<Scalar>());
 }
 
 /** \returns the maximum of all coefficients of \c *this.
@@ -435,12 +348,10 @@ template<typename Derived>
 EIGEN_STRONG_INLINE typename internal::traits<Derived>::Scalar
 DenseBase<Derived>::maxCoeff() const
 {
-  return derived().redux(Eigen::internal::scalar_max_op<Scalar,Scalar>());
+  return this->redux(Eigen::internal::scalar_max_op<Scalar>());
 }
 
-/** \returns the sum of all coefficients of \c *this
-  *
-  * If \c *this is empty, then the value 0 is returned.
+/** \returns the sum of all coefficients of *this
   *
   * \sa trace(), prod(), mean()
   */
@@ -450,7 +361,7 @@ DenseBase<Derived>::sum() const
 {
   if(SizeAtCompileTime==0 || (SizeAtCompileTime==Dynamic && size()==0))
     return Scalar(0);
-  return derived().redux(Eigen::internal::scalar_sum_op<Scalar,Scalar>());
+  return this->redux(Eigen::internal::scalar_sum_op<Scalar>());
 }
 
 /** \returns the mean of all coefficients of *this
@@ -461,14 +372,7 @@ template<typename Derived>
 EIGEN_STRONG_INLINE typename internal::traits<Derived>::Scalar
 DenseBase<Derived>::mean() const
 {
-#ifdef __INTEL_COMPILER
-  #pragma warning push
-  #pragma warning ( disable : 2259 )
-#endif
-  return Scalar(derived().redux(Eigen::internal::scalar_sum_op<Scalar,Scalar>())) / Scalar(this->size());
-#ifdef __INTEL_COMPILER
-  #pragma warning pop
-#endif
+  return Scalar(this->redux(Eigen::internal::scalar_sum_op<Scalar>())) / Scalar(this->size());
 }
 
 /** \returns the product of all coefficients of *this
@@ -484,7 +388,7 @@ DenseBase<Derived>::prod() const
 {
   if(SizeAtCompileTime==0 || (SizeAtCompileTime==Dynamic && size()==0))
     return Scalar(1);
-  return derived().redux(Eigen::internal::scalar_product_op<Scalar>());
+  return this->redux(Eigen::internal::scalar_product_op<Scalar>());
 }
 
 /** \returns the trace of \c *this, i.e. the sum of the coefficients on the main diagonal.

Eigen/src/Core/Ref.h

@@ -12,6 +12,79 @@
 
 namespace Eigen { 
 
+template<typename Derived> class RefBase;
+template<typename PlainObjectType, int Options = 0,
+         typename StrideType = typename internal::conditional<PlainObjectType::IsVectorAtCompileTime,InnerStride<1>,OuterStride<> >::type > class Ref;
+
+/** \class Ref
+  * \ingroup Core_Module
+  *
+  * \brief A matrix or vector expression mapping an existing expressions
+  *
+  * \tparam PlainObjectType the equivalent matrix type of the mapped data
+  * \tparam Options specifies whether the pointer is \c #Aligned, or \c #Unaligned.
+  *                The default is \c #Unaligned.
+  * \tparam StrideType optionally specifies strides. By default, Ref implies a contiguous storage along the inner dimension (inner stride==1),
+  *                   but accept a variable outer stride (leading dimension).
+  *                   This can be overridden by specifying strides.
+  *                   The type passed here must be a specialization of the Stride template, see examples below.
+  *
+  * This class permits to write non template functions taking Eigen's object as parameters while limiting the number of copies.
+  * A Ref<> object can represent either a const expression or a l-value:
+  * \code
+  * // in-out argument:
+  * void foo1(Ref<VectorXf> x);
+  *
+  * // read-only const argument:
+  * void foo2(const Ref<const VectorXf>& x);
+  * \endcode
+  *
+  * In the in-out case, the input argument must satisfies the constraints of the actual Ref<> type, otherwise a compilation issue will be triggered.
+  * By default, a Ref<VectorXf> can reference any dense vector expression of float having a contiguous memory layout.
+  * Likewise, a Ref<MatrixXf> can reference any column major dense matrix expression of float whose column's elements are contiguously stored with
+  * the possibility to have a constant space inbetween each column, i.e.: the inner stride mmust be equal to 1, but the outer-stride (or leading dimension),
+  * can be greater than the number of rows.
+  *
+  * In the const case, if the input expression does not match the above requirement, then it is evaluated into a temporary before being passed to the function.
+  * Here are some examples:
+  * \code
+  * MatrixXf A;
+  * VectorXf a;
+  * foo1(a.head());             // OK
+  * foo1(A.col());              // OK
+  * foo1(A.row());              // compilation error because here innerstride!=1
+  * foo2(A.row());              // The row is copied into a contiguous temporary
+  * foo2(2*a);                  // The expression is evaluated into a temporary
+  * foo2(A.col().segment(2,4)); // No temporary
+  * \endcode
+  *
+  * The range of inputs that can be referenced without temporary can be enlarged using the last two template parameter.
+  * Here is an example accepting an innerstride!=1:
+  * \code
+  * // in-out argument:
+  * void foo3(Ref<VectorXf,0,InnerStride<> > x);
+  * foo3(A.row());              // OK
+  * \endcode
+  * The downside here is that the function foo3 might be significantly slower than foo1 because it won't be able to exploit vectorization, and will involved more
+  * expensive address computations even if the input is contiguously stored in memory. To overcome this issue, one might propose to overloads internally calling a
+  * template function, e.g.:
+  * \code
+  * // in the .h:
+  * void foo(const Ref<MatrixXf>& A);
+  * void foo(const Ref<MatrixXf,0,Stride<> >& A);
+  *
+  * // in the .cpp:
+  * template<typename TypeOfA> void foo_impl(const TypeOfA& A) {
+  *     ... // crazy code goes here
+  * }
+  * void foo(const Ref<MatrixXf>& A) { foo_impl(A); }
+  * void foo(const Ref<MatrixXf,0,Stride<> >& A) { foo_impl(A); }
+  * \endcode
+  *
+  *
+  * \sa PlainObjectBase::Map(), \ref TopicStorageOrders
+  */
+
 namespace internal {
 
 template<typename _PlainObjectType, int _Options, typename _StrideType>
@@ -22,8 +95,7 @@ struct traits<Ref<_PlainObjectType, _Options, _StrideType> >
   typedef _StrideType StrideType;
   enum {
     Options = _Options,
-    Flags = traits<Map<_PlainObjectType, _Options, _StrideType> >::Flags | NestByRefBit,
-    Alignment = traits<Map<_PlainObjectType, _Options, _StrideType> >::Alignment
+    Flags = traits<Map<_PlainObjectType, _Options, _StrideType> >::Flags | NestByRefBit
   };
 
   template<typename Derived> struct match {
@@ -35,13 +107,7 @@ struct traits<Ref<_PlainObjectType, _Options, _StrideType> >
                       || (int(StrideType::InnerStrideAtCompileTime)==0 && int(Derived::InnerStrideAtCompileTime)==1),
       OuterStrideMatch = Derived::IsVectorAtCompileTime
                       || int(StrideType::OuterStrideAtCompileTime)==int(Dynamic) || int(StrideType::OuterStrideAtCompileTime)==int(Derived::OuterStrideAtCompileTime),
-      // NOTE, this indirection of evaluator<Derived>::Alignment is needed
-      // to workaround a very strange bug in MSVC related to the instantiation
-      // of has_*ary_operator in evaluator<CwiseNullaryOp>.
-      // This line is surprisingly very sensitive. For instance, simply adding parenthesis
-      // as "DerivedAlignment = (int(evaluator<Derived>::Alignment))," will make MSVC fail...
-      DerivedAlignment = int(evaluator<Derived>::Alignment),
-      AlignmentMatch = (int(traits<PlainObjectType>::Alignment)==int(Unaligned)) || (DerivedAlignment >= int(Alignment)), // FIXME the first condition is not very clear, it should be replaced by the required alignment
+      AlignmentMatch = (_Options!=Aligned) || ((PlainObjectType::Flags&AlignedBit)==0) || ((traits<Derived>::Flags&AlignedBit)==AlignedBit),
       ScalarTypeMatch = internal::is_same<typename PlainObjectType::Scalar, typename Derived::Scalar>::value,
       MatchAtCompileTime = HasDirectAccess && StorageOrderMatch && InnerStrideMatch && OuterStrideMatch && AlignmentMatch && ScalarTypeMatch
     };
@@ -66,12 +132,12 @@ public:
   typedef MapBase<Derived> Base;
   EIGEN_DENSE_PUBLIC_INTERFACE(RefBase)
 
-  EIGEN_DEVICE_FUNC inline Index innerStride() const
+  inline Index innerStride() const
   {
     return StrideType::InnerStrideAtCompileTime != 0 ? m_stride.inner() : 1;
   }
 
-  EIGEN_DEVICE_FUNC inline Index outerStride() const
+  inline Index outerStride() const
   {
     return StrideType::OuterStrideAtCompileTime != 0 ? m_stride.outer()
          : IsVectorAtCompileTime ? this->size()
@@ -79,7 +145,7 @@ public:
          : this->rows();
   }
 
-  EIGEN_DEVICE_FUNC RefBase()
+  RefBase()
     : Base(0,RowsAtCompileTime==Dynamic?0:RowsAtCompileTime,ColsAtCompileTime==Dynamic?0:ColsAtCompileTime),
       // Stride<> does not allow default ctor for Dynamic strides, so let' initialize it with dummy values:
       m_stride(StrideType::OuterStrideAtCompileTime==Dynamic?0:StrideType::OuterStrideAtCompileTime,
@@ -93,7 +159,7 @@ protected:
   typedef Stride<StrideType::OuterStrideAtCompileTime,StrideType::InnerStrideAtCompileTime> StrideBase;
 
   template<typename Expression>
-  EIGEN_DEVICE_FUNC void construct(Expression& expr)
+  void construct(Expression& expr)
   {
     if(PlainObjectType::RowsAtCompileTime==1)
     {
@@ -118,83 +184,15 @@ protected:
   StrideBase m_stride;
 };
 
-/** \class Ref
-  * \ingroup Core_Module
-  *
-  * \brief A matrix or vector expression mapping an existing expression
-  *
-  * \tparam PlainObjectType the equivalent matrix type of the mapped data
-  * \tparam Options specifies the pointer alignment in bytes. It can be: \c #Aligned128, , \c #Aligned64, \c #Aligned32, \c #Aligned16, \c #Aligned8 or \c #Unaligned.
-  *                 The default is \c #Unaligned.
-  * \tparam StrideType optionally specifies strides. By default, Ref implies a contiguous storage along the inner dimension (inner stride==1),
-  *                   but accepts a variable outer stride (leading dimension).
-  *                   This can be overridden by specifying strides.
-  *                   The type passed here must be a specialization of the Stride template, see examples below.
-  *
-  * This class provides a way to write non-template functions taking Eigen objects as parameters while limiting the number of copies.
-  * A Ref<> object can represent either a const expression or a l-value:
-  * \code
-  * // in-out argument:
-  * void foo1(Ref<VectorXf> x);
-  *
-  * // read-only const argument:
-  * void foo2(const Ref<const VectorXf>& x);
-  * \endcode
-  *
-  * In the in-out case, the input argument must satisfy the constraints of the actual Ref<> type, otherwise a compilation issue will be triggered.
-  * By default, a Ref<VectorXf> can reference any dense vector expression of float having a contiguous memory layout.
-  * Likewise, a Ref<MatrixXf> can reference any column-major dense matrix expression of float whose column's elements are contiguously stored with
-  * the possibility to have a constant space in-between each column, i.e. the inner stride must be equal to 1, but the outer stride (or leading dimension)
-  * can be greater than the number of rows.
-  *
-  * In the const case, if the input expression does not match the above requirement, then it is evaluated into a temporary before being passed to the function.
-  * Here are some examples:
-  * \code
-  * MatrixXf A;
-  * VectorXf a;
-  * foo1(a.head());             // OK
-  * foo1(A.col());              // OK
-  * foo1(A.row());              // Compilation error because here innerstride!=1
-  * foo2(A.row());              // Compilation error because A.row() is a 1xN object while foo2 is expecting a Nx1 object
-  * foo2(A.row().transpose());  // The row is copied into a contiguous temporary
-  * foo2(2*a);                  // The expression is evaluated into a temporary
-  * foo2(A.col().segment(2,4)); // No temporary
-  * \endcode
-  *
-  * The range of inputs that can be referenced without temporary can be enlarged using the last two template parameters.
-  * Here is an example accepting an innerstride!=1:
-  * \code
-  * // in-out argument:
-  * void foo3(Ref<VectorXf,0,InnerStride<> > x);
-  * foo3(A.row());              // OK
-  * \endcode
-  * The downside here is that the function foo3 might be significantly slower than foo1 because it won't be able to exploit vectorization, and will involve more
-  * expensive address computations even if the input is contiguously stored in memory. To overcome this issue, one might propose to overload internally calling a
-  * template function, e.g.:
-  * \code
-  * // in the .h:
-  * void foo(const Ref<MatrixXf>& A);
-  * void foo(const Ref<MatrixXf,0,Stride<> >& A);
-  *
-  * // in the .cpp:
-  * template<typename TypeOfA> void foo_impl(const TypeOfA& A) {
-  *     ... // crazy code goes here
-  * }
-  * void foo(const Ref<MatrixXf>& A) { foo_impl(A); }
-  * void foo(const Ref<MatrixXf,0,Stride<> >& A) { foo_impl(A); }
-  * \endcode
-  *
-  *
-  * \sa PlainObjectBase::Map(), \ref TopicStorageOrders
-  */
+
 template<typename PlainObjectType, int Options, typename StrideType> class Ref
   : public RefBase<Ref<PlainObjectType, Options, StrideType> >
 {
   private:
     typedef internal::traits<Ref> Traits;
     template<typename Derived>
-    EIGEN_DEVICE_FUNC inline Ref(const PlainObjectBase<Derived>& expr,
-                                 typename internal::enable_if<bool(Traits::template match<Derived>::MatchAtCompileTime),Derived>::type* = 0);
+    inline Ref(const PlainObjectBase<Derived>& expr,
+               typename internal::enable_if<bool(Traits::template match<Derived>::MatchAtCompileTime),Derived>::type* = 0);
   public:
 
     typedef RefBase<Ref> Base;
@@ -203,24 +201,23 @@ template<typename PlainObjectType, int Options, typename StrideType> class Ref
 
     #ifndef EIGEN_PARSED_BY_DOXYGEN
     template<typename Derived>
-    EIGEN_DEVICE_FUNC inline Ref(PlainObjectBase<Derived>& expr,
-                                 typename internal::enable_if<bool(Traits::template match<Derived>::MatchAtCompileTime),Derived>::type* = 0)
+    inline Ref(PlainObjectBase<Derived>& expr,
+               typename internal::enable_if<bool(Traits::template match<Derived>::MatchAtCompileTime),Derived>::type* = 0)
     {
-      EIGEN_STATIC_ASSERT(bool(Traits::template match<Derived>::MatchAtCompileTime), STORAGE_LAYOUT_DOES_NOT_MATCH);
+      EIGEN_STATIC_ASSERT(static_cast<bool>(Traits::template match<Derived>::MatchAtCompileTime), STORAGE_LAYOUT_DOES_NOT_MATCH);
       Base::construct(expr.derived());
     }
     template<typename Derived>
-    EIGEN_DEVICE_FUNC inline Ref(const DenseBase<Derived>& expr,
-                                 typename internal::enable_if<bool(Traits::template match<Derived>::MatchAtCompileTime),Derived>::type* = 0)
+    inline Ref(const DenseBase<Derived>& expr,
+               typename internal::enable_if<bool(Traits::template match<Derived>::MatchAtCompileTime),Derived>::type* = 0)
     #else
-    /** Implicit constructor from any dense expression */
     template<typename Derived>
     inline Ref(DenseBase<Derived>& expr)
     #endif
     {
-      EIGEN_STATIC_ASSERT(bool(internal::is_lvalue<Derived>::value), THIS_EXPRESSION_IS_NOT_A_LVALUE__IT_IS_READ_ONLY);
-      EIGEN_STATIC_ASSERT(bool(Traits::template match<Derived>::MatchAtCompileTime), STORAGE_LAYOUT_DOES_NOT_MATCH);
-      EIGEN_STATIC_ASSERT(!Derived::IsPlainObjectBase,THIS_EXPRESSION_IS_NOT_A_LVALUE__IT_IS_READ_ONLY);
+      EIGEN_STATIC_ASSERT(static_cast<bool>(internal::is_lvalue<Derived>::value), THIS_EXPRESSION_IS_NOT_A_LVALUE__IT_IS_READ_ONLY);
+      EIGEN_STATIC_ASSERT(static_cast<bool>(Traits::template match<Derived>::MatchAtCompileTime), STORAGE_LAYOUT_DOES_NOT_MATCH);
+      enum { THIS_EXPRESSION_IS_NOT_A_LVALUE__IT_IS_READ_ONLY = Derived::ThisConstantIsPrivateInPlainObjectBase};
       Base::construct(expr.const_cast_derived());
     }
 
@@ -239,36 +236,36 @@ template<typename TPlainObjectType, int Options, typename StrideType> class Ref<
     EIGEN_DENSE_PUBLIC_INTERFACE(Ref)
 
     template<typename Derived>
-    EIGEN_DEVICE_FUNC inline Ref(const DenseBase<Derived>& expr,
-                                 typename internal::enable_if<bool(Traits::template match<Derived>::ScalarTypeMatch),Derived>::type* = 0)
+    inline Ref(const DenseBase<Derived>& expr,
+               typename internal::enable_if<bool(Traits::template match<Derived>::ScalarTypeMatch),Derived>::type* = 0)
     {
 //      std::cout << match_helper<Derived>::HasDirectAccess << "," << match_helper<Derived>::OuterStrideMatch << "," << match_helper<Derived>::InnerStrideMatch << "\n";
 //      std::cout << int(StrideType::OuterStrideAtCompileTime) << " - " << int(Derived::OuterStrideAtCompileTime) << "\n";
 //      std::cout << int(StrideType::InnerStrideAtCompileTime) << " - " << int(Derived::InnerStrideAtCompileTime) << "\n";
       construct(expr.derived(), typename Traits::template match<Derived>::type());
     }
-
-    EIGEN_DEVICE_FUNC inline Ref(const Ref& other) : Base(other) {
+    
+    inline Ref(const Ref& other) : Base(other) {
       // copy constructor shall not copy the m_object, to avoid unnecessary malloc and copy
     }
 
     template<typename OtherRef>
-    EIGEN_DEVICE_FUNC inline Ref(const RefBase<OtherRef>& other) {
+    inline Ref(const RefBase<OtherRef>& other) {
       construct(other.derived(), typename Traits::template match<OtherRef>::type());
     }
 
   protected:
 
     template<typename Expression>
-    EIGEN_DEVICE_FUNC void construct(const Expression& expr,internal::true_type)
+    void construct(const Expression& expr,internal::true_type)
     {
       Base::construct(expr);
     }
 
     template<typename Expression>
-    EIGEN_DEVICE_FUNC void construct(const Expression& expr, internal::false_type)
+    void construct(const Expression& expr, internal::false_type)
     {
-      internal::call_assignment_no_alias(m_object,expr,internal::assign_op<Scalar,Scalar>());
+      m_object.lazyAssign(expr);
       Base::construct(m_object);
     }
 

Eigen/src/Core/Replicate.h

@@ -12,6 +12,21 @@
 
 namespace Eigen { 
 
+/**
+  * \class Replicate
+  * \ingroup Core_Module
+  *
+  * \brief Expression of the multiple replication of a matrix or vector
+  *
+  * \param MatrixType the type of the object we are replicating
+  *
+  * This class represents an expression of the multiple replication of a matrix or vector.
+  * It is the return type of DenseBase::replicate() and most of the time
+  * this is the only way it is used.
+  *
+  * \sa DenseBase::replicate()
+  */
+
 namespace internal {
 template<typename MatrixType,int RowFactor,int ColFactor>
 struct traits<Replicate<MatrixType,RowFactor,ColFactor> >
@@ -20,7 +35,10 @@ struct traits<Replicate<MatrixType,RowFactor,ColFactor> >
   typedef typename MatrixType::Scalar Scalar;
   typedef typename traits<MatrixType>::StorageKind StorageKind;
   typedef typename traits<MatrixType>::XprKind XprKind;
-  typedef typename ref_selector<MatrixType>::type MatrixTypeNested;
+  enum {
+    Factor = (RowFactor==Dynamic || ColFactor==Dynamic) ? Dynamic : RowFactor*ColFactor
+  };
+  typedef typename nested<MatrixType,Factor>::type MatrixTypeNested;
   typedef typename remove_reference<MatrixTypeNested>::type _MatrixTypeNested;
   enum {
     RowsAtCompileTime = RowFactor==Dynamic || int(MatrixType::RowsAtCompileTime)==Dynamic
@@ -35,29 +53,12 @@ struct traits<Replicate<MatrixType,RowFactor,ColFactor> >
     IsRowMajor = MaxRowsAtCompileTime==1 && MaxColsAtCompileTime!=1 ? 1
                : MaxColsAtCompileTime==1 && MaxRowsAtCompileTime!=1 ? 0
                : (MatrixType::Flags & RowMajorBit) ? 1 : 0,
-    
-    // FIXME enable DirectAccess with negative strides?
-    Flags = IsRowMajor ? RowMajorBit : 0
+    Flags = (_MatrixTypeNested::Flags & HereditaryBits & ~RowMajorBit) | (IsRowMajor ? RowMajorBit : 0),
+    CoeffReadCost = _MatrixTypeNested::CoeffReadCost
   };
 };
 }
 
-/**
-  * \class Replicate
-  * \ingroup Core_Module
-  *
-  * \brief Expression of the multiple replication of a matrix or vector
-  *
-  * \tparam MatrixType the type of the object we are replicating
-  * \tparam RowFactor number of repetitions at compile time along the vertical direction, can be Dynamic.
-  * \tparam ColFactor number of repetitions at compile time along the horizontal direction, can be Dynamic.
-  *
-  * This class represents an expression of the multiple replication of a matrix or vector.
-  * It is the return type of DenseBase::replicate() and most of the time
-  * this is the only way it is used.
-  *
-  * \sa DenseBase::replicate()
-  */
 template<typename MatrixType,int RowFactor,int ColFactor> class Replicate
   : public internal::dense_xpr_base< Replicate<MatrixType,RowFactor,ColFactor> >::type
 {
@@ -67,12 +68,10 @@ template<typename MatrixType,int RowFactor,int ColFactor> class Replicate
 
     typedef typename internal::dense_xpr_base<Replicate>::type Base;
     EIGEN_DENSE_PUBLIC_INTERFACE(Replicate)
-    typedef typename internal::remove_all<MatrixType>::type NestedExpression;
 
     template<typename OriginalMatrixType>
-    EIGEN_DEVICE_FUNC
-    inline explicit Replicate(const OriginalMatrixType& matrix)
-      : m_matrix(matrix), m_rowFactor(RowFactor), m_colFactor(ColFactor)
+    inline explicit Replicate(const OriginalMatrixType& a_matrix)
+      : m_matrix(a_matrix), m_rowFactor(RowFactor), m_colFactor(ColFactor)
     {
       EIGEN_STATIC_ASSERT((internal::is_same<typename internal::remove_const<MatrixType>::type,OriginalMatrixType>::value),
                           THE_MATRIX_OR_EXPRESSION_THAT_YOU_PASSED_DOES_NOT_HAVE_THE_EXPECTED_TYPE)
@@ -80,20 +79,41 @@ template<typename MatrixType,int RowFactor,int ColFactor> class Replicate
     }
 
     template<typename OriginalMatrixType>
-    EIGEN_DEVICE_FUNC
-    inline Replicate(const OriginalMatrixType& matrix, Index rowFactor, Index colFactor)
-      : m_matrix(matrix), m_rowFactor(rowFactor), m_colFactor(colFactor)
+    inline Replicate(const OriginalMatrixType& a_matrix, Index rowFactor, Index colFactor)
+      : m_matrix(a_matrix), m_rowFactor(rowFactor), m_colFactor(colFactor)
     {
       EIGEN_STATIC_ASSERT((internal::is_same<typename internal::remove_const<MatrixType>::type,OriginalMatrixType>::value),
                           THE_MATRIX_OR_EXPRESSION_THAT_YOU_PASSED_DOES_NOT_HAVE_THE_EXPECTED_TYPE)
     }
 
-    EIGEN_DEVICE_FUNC
     inline Index rows() const { return m_matrix.rows() * m_rowFactor.value(); }
-    EIGEN_DEVICE_FUNC
     inline Index cols() const { return m_matrix.cols() * m_colFactor.value(); }
 
-    EIGEN_DEVICE_FUNC
+    inline Scalar coeff(Index rowId, Index colId) const
+    {
+      // try to avoid using modulo; this is a pure optimization strategy
+      const Index actual_row  = internal::traits<MatrixType>::RowsAtCompileTime==1 ? 0
+                            : RowFactor==1 ? rowId
+                            : rowId%m_matrix.rows();
+      const Index actual_col  = internal::traits<MatrixType>::ColsAtCompileTime==1 ? 0
+                            : ColFactor==1 ? colId
+                            : colId%m_matrix.cols();
+
+      return m_matrix.coeff(actual_row, actual_col);
+    }
+    template<int LoadMode>
+    inline PacketScalar packet(Index rowId, Index colId) const
+    {
+      const Index actual_row  = internal::traits<MatrixType>::RowsAtCompileTime==1 ? 0
+                            : RowFactor==1 ? rowId
+                            : rowId%m_matrix.rows();
+      const Index actual_col  = internal::traits<MatrixType>::ColsAtCompileTime==1 ? 0
+                            : ColFactor==1 ? colId
+                            : colId%m_matrix.cols();
+
+      return m_matrix.template packet<LoadMode>(actual_row, actual_col);
+    }
+
     const _MatrixTypeNested& nestedExpression() const
     { 
       return m_matrix; 
@@ -121,6 +141,21 @@ DenseBase<Derived>::replicate() const
   return Replicate<Derived,RowFactor,ColFactor>(derived());
 }
 
+/**
+  * \return an expression of the replication of \c *this
+  *
+  * Example: \include MatrixBase_replicate_int_int.cpp
+  * Output: \verbinclude MatrixBase_replicate_int_int.out
+  *
+  * \sa VectorwiseOp::replicate(), DenseBase::replicate<int,int>(), class Replicate
+  */
+template<typename Derived>
+const typename DenseBase<Derived>::ReplicateReturnType
+DenseBase<Derived>::replicate(Index rowFactor,Index colFactor) const
+{
+  return Replicate<Derived,Dynamic,Dynamic>(derived(),rowFactor,colFactor);
+}
+
 /**
   * \return an expression of the replication of each column (or row) of \c *this
   *

Eigen/src/Core/ReturnByValue.h

@@ -13,6 +13,11 @@
 
 namespace Eigen {
 
+/** \class ReturnByValue
+  * \ingroup Core_Module
+  *
+  */
+
 namespace internal {
 
 template<typename Derived>
@@ -33,22 +38,17 @@ struct traits<ReturnByValue<Derived> >
  * So internal::nested always gives the plain return matrix type.
  *
  * FIXME: I don't understand why we need this specialization: isn't this taken care of by the EvalBeforeNestingBit ??
- * Answer: EvalBeforeNestingBit should be deprecated since we have the evaluators
  */
 template<typename Derived,int n,typename PlainObject>
-struct nested_eval<ReturnByValue<Derived>, n, PlainObject>
+struct nested<ReturnByValue<Derived>, n, PlainObject>
 {
   typedef typename traits<Derived>::ReturnType type;
 };
 
 } // end namespace internal
 
-/** \class ReturnByValue
-  * \ingroup Core_Module
-  *
-  */
 template<typename Derived> class ReturnByValue
-  : public internal::dense_xpr_base< ReturnByValue<Derived> >::type, internal::no_assignment_operator
+  : internal::no_assignment_operator, public internal::dense_xpr_base< ReturnByValue<Derived> >::type
 {
   public:
     typedef typename internal::traits<Derived>::ReturnType ReturnType;
@@ -57,11 +57,10 @@ template<typename Derived> class ReturnByValue
     EIGEN_DENSE_PUBLIC_INTERFACE(ReturnByValue)
 
     template<typename Dest>
-    EIGEN_DEVICE_FUNC
     inline void evalTo(Dest& dst) const
     { static_cast<const Derived*>(this)->evalTo(dst); }
-    EIGEN_DEVICE_FUNC inline Index rows() const { return static_cast<const Derived*>(this)->rows(); }
-    EIGEN_DEVICE_FUNC inline Index cols() const { return static_cast<const Derived*>(this)->cols(); }
+    inline Index rows() const { return static_cast<const Derived*>(this)->rows(); }
+    inline Index cols() const { return static_cast<const Derived*>(this)->cols(); }
 
 #ifndef EIGEN_PARSED_BY_DOXYGEN
 #define Unusable YOU_ARE_TRYING_TO_ACCESS_A_SINGLE_COEFFICIENT_IN_A_SPECIAL_EXPRESSION_WHERE_THAT_IS_NOT_ALLOWED_BECAUSE_THAT_WOULD_BE_INEFFICIENT
@@ -73,7 +72,8 @@ template<typename Derived> class ReturnByValue
     const Unusable& coeff(Index,Index) const { return *reinterpret_cast<const Unusable*>(this); }
     Unusable& coeffRef(Index) { return *reinterpret_cast<Unusable*>(this); }
     Unusable& coeffRef(Index,Index) { return *reinterpret_cast<Unusable*>(this); }
-#undef Unusable
+    template<int LoadMode>  Unusable& packet(Index) const;
+    template<int LoadMode>  Unusable& packet(Index, Index) const;
 #endif
 };
 
@@ -85,32 +85,14 @@ Derived& DenseBase<Derived>::operator=(const ReturnByValue<OtherDerived>& other)
   return derived();
 }
 
-namespace internal {
-
-// Expression is evaluated in a temporary; default implementation of Assignment is bypassed so that
-// when a ReturnByValue expression is assigned, the evaluator is not constructed.
-// TODO: Finalize port to new regime; ReturnByValue should not exist in the expression world
-  
 template<typename Derived>
-struct evaluator<ReturnByValue<Derived> >
-  : public evaluator<typename internal::traits<Derived>::ReturnType>
+template<typename OtherDerived>
+Derived& DenseBase<Derived>::lazyAssign(const ReturnByValue<OtherDerived>& other)
 {
-  typedef ReturnByValue<Derived> XprType;
-  typedef typename internal::traits<Derived>::ReturnType PlainObject;
-  typedef evaluator<PlainObject> Base;
-  
-  EIGEN_DEVICE_FUNC explicit evaluator(const XprType& xpr)
-    : m_result(xpr.rows(), xpr.cols())
-  {
-    ::new (static_cast<Base*>(this)) Base(m_result);
-    xpr.evalTo(m_result);
-  }
+  other.evalTo(derived());
+  return derived();
+}
 
-protected:
-  PlainObject m_result;
-};
-
-} // end namespace internal
 
 } // end namespace Eigen