bump to 3.3.4

(grafted from 9fbdf02059
)

CMakeLists.txt

@@ -380,7 +380,7 @@ else()
       )
 endif()
 set(CMAKEPACKAGE_INSTALL_DIR
-    "${CMAKE_INSTALL_LIBDIR}/cmake/eigen3"
+    "${CMAKE_INSTALL_DATADIR}/eigen3/cmake"
     CACHE PATH "The directory relative to CMAKE_PREFIX_PATH where Eigen3Config.cmake is installed"
     )
 set(PKGCONFIG_INSTALL_DIR
@@ -507,18 +507,89 @@ set ( EIGEN_VERSION_MINOR  ${EIGEN_MAJOR_VERSION} )
 set ( EIGEN_VERSION_PATCH  ${EIGEN_MINOR_VERSION} )
 set ( EIGEN_DEFINITIONS "")
 set ( EIGEN_INCLUDE_DIR "${CMAKE_INSTALL_PREFIX}/${INCLUDE_INSTALL_DIR}" )
-set ( EIGEN_INCLUDE_DIRS ${EIGEN_INCLUDE_DIR} )
 set ( EIGEN_ROOT_DIR ${CMAKE_INSTALL_PREFIX} )
 
-configure_file ( ${CMAKE_CURRENT_SOURCE_DIR}/cmake/Eigen3Config.cmake.in
-                 ${CMAKE_CURRENT_BINARY_DIR}/Eigen3Config.cmake
-                 @ONLY ESCAPE_QUOTES
-               )
+# Interface libraries require at least CMake 3.0
+if (NOT CMAKE_VERSION VERSION_LESS 3.0)
+  include (CMakePackageConfigHelpers)
+
+  # Imported target support
+  add_library (eigen INTERFACE)
+
+  target_compile_definitions (eigen INTERFACE ${EIGEN_DEFINITIONS})
+  target_include_directories (eigen INTERFACE
+    $<BUILD_INTERFACE:${CMAKE_CURRENT_SOURCE_DIR}>
+    $<INSTALL_INTERFACE:${INCLUDE_INSTALL_DIR}>
+  )
+
+  # Export as title case Eigen
+  set_target_properties (eigen PROPERTIES EXPORT_NAME Eigen)
+
+  install (TARGETS eigen EXPORT Eigen3Targets)
+
+  configure_package_config_file (
+    ${CMAKE_CURRENT_SOURCE_DIR}/cmake/Eigen3Config.cmake.in
+    ${CMAKE_CURRENT_BINARY_DIR}/Eigen3Config.cmake
+    PATH_VARS EIGEN_INCLUDE_DIR EIGEN_ROOT_DIR
+    INSTALL_DESTINATION ${CMAKEPACKAGE_INSTALL_DIR}
+    NO_CHECK_REQUIRED_COMPONENTS_MACRO # Eigen does not provide components
+  )
+  # Remove CMAKE_SIZEOF_VOID_P from Eigen3ConfigVersion.cmake since Eigen does
+  # not depend on architecture specific settings or libraries. More
+  # specifically, an Eigen3Config.cmake generated from a 64 bit target can be
+  # used for 32 bit targets as well (and vice versa).
+  set (_Eigen3_CMAKE_SIZEOF_VOID_P ${CMAKE_SIZEOF_VOID_P})
+  unset (CMAKE_SIZEOF_VOID_P)
+  write_basic_package_version_file (Eigen3ConfigVersion.cmake
+                                    VERSION ${EIGEN_VERSION_NUMBER}
+                                    COMPATIBILITY SameMajorVersion)
+  set (CMAKE_SIZEOF_VOID_P ${_Eigen3_CMAKE_SIZEOF_VOID_P})
+
+  # The Eigen target will be located in the Eigen3 namespace. Other CMake
+  # targets can refer to it using Eigen3::Eigen.
+  export (TARGETS eigen NAMESPACE Eigen3:: FILE Eigen3Targets.cmake)
+  # Export Eigen3 package to CMake registry such that it can be easily found by
+  # CMake even if it has not been installed to a standard directory.
+  export (PACKAGE Eigen3)
+
+  install (EXPORT Eigen3Targets NAMESPACE Eigen3:: DESTINATION ${CMAKEPACKAGE_INSTALL_DIR})
+
+else (NOT CMAKE_VERSION VERSION_LESS 3.0)
+  # Fallback to legacy Eigen3Config.cmake without the imported target
+  
+  # If CMakePackageConfigHelpers module is available (CMake >= 2.8.8)
+  # create a relocatable Config file, otherwise leave the hardcoded paths       
+  include(CMakePackageConfigHelpers OPTIONAL RESULT_VARIABLE CPCH_PATH)
+  
+  if(CPCH_PATH)
+    configure_package_config_file (
+      ${CMAKE_CURRENT_SOURCE_DIR}/cmake/Eigen3ConfigLegacy.cmake.in
+      ${CMAKE_CURRENT_BINARY_DIR}/Eigen3Config.cmake
+      PATH_VARS EIGEN_INCLUDE_DIR EIGEN_ROOT_DIR
+      INSTALL_DESTINATION ${CMAKEPACKAGE_INSTALL_DIR}
+      NO_CHECK_REQUIRED_COMPONENTS_MACRO # Eigen does not provide components
+    )
+  else() 
+    # The PACKAGE_* variables are defined by the configure_package_config_file
+    # but without it we define them manually to the hardcoded paths
+    set(PACKAGE_INIT "")
+    set(PACKAGE_EIGEN_INCLUDE_DIR ${EIGEN_INCLUDE_DIR})
+    set(PACKAGE_EIGEN_ROOT_DIR ${EIGEN_ROOT_DIR})
+    configure_file ( ${CMAKE_CURRENT_SOURCE_DIR}/cmake/Eigen3ConfigLegacy.cmake.in
+                     ${CMAKE_CURRENT_BINARY_DIR}/Eigen3Config.cmake
+                     @ONLY ESCAPE_QUOTES )
+  endif()
+
+  write_basic_package_version_file( Eigen3ConfigVersion.cmake
+                                    VERSION ${EIGEN_VERSION_NUMBER}
+                                    COMPATIBILITY SameMajorVersion )
+
+endif (NOT CMAKE_VERSION VERSION_LESS 3.0)
 
 install ( FILES ${CMAKE_CURRENT_SOURCE_DIR}/cmake/UseEigen3.cmake
                 ${CMAKE_CURRENT_BINARY_DIR}/Eigen3Config.cmake
-          DESTINATION ${CMAKEPACKAGE_INSTALL_DIR}
-        )
+                ${CMAKE_CURRENT_BINARY_DIR}/Eigen3ConfigVersion.cmake
+          DESTINATION ${CMAKEPACKAGE_INSTALL_DIR} )
 
 # Add uninstall target
 add_custom_target ( uninstall

CTestConfig.cmake

@@ -4,14 +4,10 @@
 ## # The following are required to uses Dart and the Cdash dashboard
 ##   ENABLE_TESTING()
 ##   INCLUDE(CTest)
-set(CTEST_PROJECT_NAME "Eigen")
+set(CTEST_PROJECT_NAME "Eigen3.3")
 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=Eigen")
+set(CTEST_DROP_LOCATION "/CDash/submit.php?project=Eigen3.3")
 set(CTEST_DROP_SITE_CDASH TRUE)
-set(CTEST_PROJECT_SUBPROJECTS
-Official
-Unsupported
-)

Eigen/Core

@@ -321,12 +321,16 @@ inline static const char *SimdInstructionSetsInUse(void) {
 #error Eigen2-support is only available up to version 3.2. Please go to "http://eigen.tuxfamily.org/index.php?title=Eigen2" for further information
 #endif
 
+namespace Eigen {
+
 // we use size_t frequently and we'll never remember to prepend it with std:: everytime just to
 // ensure QNX/QCC support
 using std::size_t;
 // gcc 4.6.0 wants std:: for ptrdiff_t
 using std::ptrdiff_t;
 
+}
+
 /** \defgroup Core_Module Core module
   * This is the main module of Eigen providing dense matrix and vector support
   * (both fixed and dynamic size) with all the features corresponding to a BLAS library
@@ -405,6 +409,7 @@ using std::ptrdiff_t;
 // on CUDA devices
 #include "src/Core/arch/CUDA/Complex.h"
 
+#include "src/Core/IO.h"
 #include "src/Core/DenseCoeffsBase.h"
 #include "src/Core/DenseBase.h"
 #include "src/Core/MatrixBase.h"
@@ -452,7 +457,6 @@ using std::ptrdiff_t;
 #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/GeneralProduct.h"

Eigen/QtAlignedMalloc

@@ -14,7 +14,7 @@
 
 #include "src/Core/util/DisableStupidWarnings.h"
 
-void *qMalloc(size_t size)
+void *qMalloc(std::size_t size)
 {
   return Eigen::internal::aligned_malloc(size);
 }
@@ -24,7 +24,7 @@ void qFree(void *ptr)
   Eigen::internal::aligned_free(ptr);
 }
 
-void *qRealloc(void *ptr, size_t size)
+void *qRealloc(void *ptr, std::size_t size)
 {
   void* newPtr = Eigen::internal::aligned_malloc(size);
   memcpy(newPtr, ptr, size);

Eigen/Sparse

@@ -25,7 +25,9 @@
 
 #include "SparseCore"
 #include "OrderingMethods"
+#ifndef EIGEN_MPL2_ONLY
 #include "SparseCholesky"
+#endif
 #include "SparseLU"
 #include "SparseQR"
 #include "IterativeLinearSolvers"

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 EIGEN_COMP_MSVC && EIGEN_OS_WIN64 && (EIGEN_MAX_STATIC_ALIGN_BYTES<=16) /* MSVC auto aligns up to 16 bytes in 64 bit builds */
 
 #define EIGEN_DEFINE_STL_DEQUE_SPECIALIZATION(...)
 

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 EIGEN_COMP_MSVC && EIGEN_OS_WIN64 && (EIGEN_MAX_STATIC_ALIGN_BYTES<=16) /* MSVC auto aligns up to 16 bytes in 64 bit builds */
 
 #define EIGEN_DEFINE_STL_LIST_SPECIALIZATION(...)
 

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 EIGEN_COMP_MSVC && EIGEN_OS_WIN64 && (EIGEN_MAX_STATIC_ALIGN_BYTES<=16) /* MSVC auto aligns up to 16 bytes in 64 bit builds */
 
 #define EIGEN_DEFINE_STL_VECTOR_SPECIALIZATION(...)
 

Eigen/src/Core/Array.h

@@ -231,10 +231,16 @@ class Array
             : Base(other)
     { }
 
+  private:
+    struct PrivateType {};
+  public:
+
     /** \sa MatrixBase::operator=(const EigenBase<OtherDerived>&) */
     template<typename OtherDerived>
     EIGEN_DEVICE_FUNC
-    EIGEN_STRONG_INLINE Array(const EigenBase<OtherDerived> &other)
+    EIGEN_STRONG_INLINE Array(const EigenBase<OtherDerived> &other,
+                              typename internal::enable_if<internal::is_convertible<typename OtherDerived::Scalar,Scalar>::value,
+                                                           PrivateType>::type = PrivateType())
       : Base(other.derived())
     { }
 

Eigen/src/Core/ArrayBase.h

@@ -175,7 +175,7 @@ template<typename Derived> class ArrayBase
   */
 template<typename Derived>
 template<typename OtherDerived>
-EIGEN_STRONG_INLINE Derived &
+EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE Derived &
 ArrayBase<Derived>::operator-=(const ArrayBase<OtherDerived> &other)
 {
   call_assignment(derived(), other.derived(), internal::sub_assign_op<Scalar,typename OtherDerived::Scalar>());
@@ -188,7 +188,7 @@ ArrayBase<Derived>::operator-=(const ArrayBase<OtherDerived> &other)
   */
 template<typename Derived>
 template<typename OtherDerived>
-EIGEN_STRONG_INLINE Derived &
+EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE Derived &
 ArrayBase<Derived>::operator+=(const ArrayBase<OtherDerived>& other)
 {
   call_assignment(derived(), other.derived(), internal::add_assign_op<Scalar,typename OtherDerived::Scalar>());
@@ -201,7 +201,7 @@ ArrayBase<Derived>::operator+=(const ArrayBase<OtherDerived>& other)
   */
 template<typename Derived>
 template<typename OtherDerived>
-EIGEN_STRONG_INLINE Derived &
+EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE Derived &
 ArrayBase<Derived>::operator*=(const ArrayBase<OtherDerived>& other)
 {
   call_assignment(derived(), other.derived(), internal::mul_assign_op<Scalar,typename OtherDerived::Scalar>());
@@ -214,7 +214,7 @@ ArrayBase<Derived>::operator*=(const ArrayBase<OtherDerived>& other)
   */
 template<typename Derived>
 template<typename OtherDerived>
-EIGEN_STRONG_INLINE Derived &
+EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE Derived &
 ArrayBase<Derived>::operator/=(const ArrayBase<OtherDerived>& other)
 {
   call_assignment(derived(), other.derived(), internal::div_assign_op<Scalar,typename OtherDerived::Scalar>());

Eigen/src/Core/ArrayWrapper.h

@@ -32,7 +32,8 @@ struct traits<ArrayWrapper<ExpressionType> >
   // Let's remove NestByRefBit
   enum {
     Flags0 = traits<typename remove_all<typename ExpressionType::Nested>::type >::Flags,
-    Flags = Flags0 & ~NestByRefBit
+    LvalueBitFlag = is_lvalue<ExpressionType>::value ? LvalueBit : 0,
+    Flags = (Flags0 & ~(NestByRefBit | LvalueBit)) | LvalueBitFlag
   };
 };
 }
@@ -129,7 +130,8 @@ struct traits<MatrixWrapper<ExpressionType> >
   // Let's remove NestByRefBit
   enum {
     Flags0 = traits<typename remove_all<typename ExpressionType::Nested>::type >::Flags,
-    Flags = Flags0 & ~NestByRefBit
+    LvalueBitFlag = is_lvalue<ExpressionType>::value ? LvalueBit : 0,
+    Flags = (Flags0 & ~(NestByRefBit | LvalueBit)) | LvalueBitFlag
   };
 };
 }

Eigen/src/Core/AssignEvaluator.h

@@ -407,7 +407,7 @@ struct dense_assignment_loop<Kernel, LinearVectorizedTraversal, NoUnrolling>
                                                             : int(Kernel::AssignmentTraits::DstAlignment),
       srcAlignment = Kernel::AssignmentTraits::JointAlignment
     };
-    const Index alignedStart = dstIsAligned ? 0 : internal::first_aligned<requestedAlignment>(&kernel.dstEvaluator().coeffRef(0), size);
+    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);
@@ -515,7 +515,7 @@ struct dense_assignment_loop<Kernel, LinearTraversal, CompleteUnrolling>
 template<typename Kernel>
 struct dense_assignment_loop<Kernel, SliceVectorizedTraversal, NoUnrolling>
 {
-  EIGEN_DEVICE_FUNC static inline void run(Kernel &kernel)
+  EIGEN_DEVICE_FUNC static EIGEN_STRONG_INLINE void run(Kernel &kernel)
   {
     typedef typename Kernel::Scalar Scalar;
     typedef typename Kernel::PacketType PacketType;
@@ -527,7 +527,7 @@ struct dense_assignment_loop<Kernel, SliceVectorizedTraversal, NoUnrolling>
       dstAlignment = alignable ? int(requestedAlignment)
                                : int(Kernel::AssignmentTraits::DstAlignment)
     };
-    const Scalar *dst_ptr = &kernel.dstEvaluator().coeffRef(0,0);
+    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
@@ -563,7 +563,7 @@ struct dense_assignment_loop<Kernel, SliceVectorizedTraversal, NoUnrolling>
 template<typename Kernel>
 struct dense_assignment_loop<Kernel, SliceVectorizedTraversal, InnerUnrolling>
 {
-  EIGEN_DEVICE_FUNC static inline void run(Kernel &kernel)
+  EIGEN_DEVICE_FUNC static EIGEN_STRONG_INLINE void run(Kernel &kernel)
   {
     typedef typename Kernel::DstEvaluatorType::XprType DstXprType;
     typedef typename Kernel::PacketType PacketType;
@@ -683,6 +683,11 @@ public:
       : int(DstEvaluatorType::Flags)&RowMajorBit ? inner
       : outer;
   }
+
+  EIGEN_DEVICE_FUNC const Scalar* dstDataPtr() const
+  {
+    return m_dstExpr.data();
+  }
   
 protected:
   DstEvaluatorType& m_dst;
@@ -696,6 +701,26 @@ protected:
 * Part 5 : Entry point for dense rectangular assignment
 ***************************************************************************/
 
+template<typename DstXprType,typename SrcXprType, typename Functor>
+EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
+void resize_if_allowed(DstXprType &dst, const SrcXprType& src, const Functor &/*func*/)
+{
+  EIGEN_ONLY_USED_FOR_DEBUG(dst);
+  EIGEN_ONLY_USED_FOR_DEBUG(src);
+  eigen_assert(dst.rows() == src.rows() && dst.cols() == src.cols());
+}
+
+template<typename DstXprType,typename SrcXprType, typename T1, typename T2>
+EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
+void resize_if_allowed(DstXprType &dst, const SrcXprType& src, const internal::assign_op<T1,T2> &/*func*/)
+{
+  Index dstRows = src.rows();
+  Index dstCols = src.cols();
+  if(((dst.rows()!=dstRows) || (dst.cols()!=dstCols)))
+    dst.resize(dstRows, dstCols);
+  eigen_assert(dst.rows() == dstRows && dst.cols() == dstCols);
+}
+
 template<typename DstXprType, typename SrcXprType, typename Functor>
 EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE void call_dense_assignment_loop(DstXprType& dst, const SrcXprType& src, const Functor &func)
 {
@@ -706,10 +731,7 @@ EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE void call_dense_assignment_loop(DstXprType
 
   // 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);
+  resize_if_allowed(dst, src, func);
 
   DstEvaluatorType dstEvaluator(dst);
     
@@ -876,6 +898,34 @@ struct Assignment<DstXprType, SrcXprType, Functor, EigenBase2EigenBase, Weak>
     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

Eigen/src/Core/CoreEvaluators.h

@@ -1556,9 +1556,7 @@ struct evaluator<Diagonal<ArgType, DiagIndex> >
   { }
  
   typedef typename XprType::Scalar Scalar;
-  // FIXME having to check whether ArgType is sparse here i not very nice.
-  typedef typename internal::conditional<!internal::is_same<typename ArgType::StorageKind,Sparse>::value,
-                                         typename XprType::CoeffReturnType,Scalar>::type CoeffReturnType;
+  typedef typename XprType::CoeffReturnType CoeffReturnType;
 
   EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
   CoeffReturnType coeff(Index row, Index) const

Eigen/src/Core/CwiseBinaryOp.h

@@ -46,7 +46,7 @@ struct traits<CwiseBinaryOp<BinaryOp, Lhs, Rhs> >
   typedef typename remove_reference<LhsNested>::type _LhsNested;
   typedef typename remove_reference<RhsNested>::type _RhsNested;
   enum {
-    Flags = _LhsNested::Flags & RowMajorBit
+    Flags = cwise_promote_storage_order<typename traits<Lhs>::StorageKind,typename traits<Rhs>::StorageKind,_LhsNested::Flags & RowMajorBit,_RhsNested::Flags & RowMajorBit>::value
   };
 };
 } // end namespace internal
@@ -84,6 +84,7 @@ class CwiseBinaryOp :
 {
   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;
 

Eigen/src/Core/CwiseNullaryOp.h

@@ -105,7 +105,7 @@ 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_DEVICE_FUNC EIGEN_STRONG_INLINE const CwiseNullaryOp<CustomNullaryOp, typename DenseBase<Derived>::PlainObject>
 DenseBase<Derived>::NullaryExpr(Index rows, Index cols, const CustomNullaryOp& func)
 {
   return CwiseNullaryOp<CustomNullaryOp, PlainObject>(rows, cols, func);
@@ -150,7 +150,7 @@ 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_DEVICE_FUNC EIGEN_STRONG_INLINE const CwiseNullaryOp<CustomNullaryOp, typename DenseBase<Derived>::PlainObject>
 DenseBase<Derived>::NullaryExpr(const CustomNullaryOp& func)
 {
   return CwiseNullaryOp<CustomNullaryOp, PlainObject>(RowsAtCompileTime, ColsAtCompileTime, func);
@@ -192,7 +192,7 @@ DenseBase<Derived>::Constant(Index rows, Index cols, const Scalar& value)
   * \sa class CwiseNullaryOp
   */
 template<typename Derived>
-EIGEN_STRONG_INLINE const typename DenseBase<Derived>::ConstantReturnType
+EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE const typename DenseBase<Derived>::ConstantReturnType
 DenseBase<Derived>::Constant(Index size, const Scalar& value)
 {
   return DenseBase<Derived>::NullaryExpr(size, internal::scalar_constant_op<Scalar>(value));
@@ -208,7 +208,7 @@ DenseBase<Derived>::Constant(Index size, const Scalar& value)
   * \sa class CwiseNullaryOp
   */
 template<typename Derived>
-EIGEN_STRONG_INLINE const typename DenseBase<Derived>::ConstantReturnType
+EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE const typename DenseBase<Derived>::ConstantReturnType
 DenseBase<Derived>::Constant(const Scalar& value)
 {
   EIGEN_STATIC_ASSERT_FIXED_SIZE(Derived)
@@ -220,7 +220,7 @@ DenseBase<Derived>::Constant(const Scalar& value)
   * \sa LinSpaced(Index,Scalar,Scalar), setLinSpaced(Index,const Scalar&,const Scalar&)
   */
 template<typename Derived>
-EIGEN_STRONG_INLINE const typename DenseBase<Derived>::RandomAccessLinSpacedReturnType
+EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE const typename DenseBase<Derived>::RandomAccessLinSpacedReturnType
 DenseBase<Derived>::LinSpaced(Sequential_t, Index size, const Scalar& low, const Scalar& high)
 {
   EIGEN_STATIC_ASSERT_VECTOR_ONLY(Derived)
@@ -232,7 +232,7 @@ DenseBase<Derived>::LinSpaced(Sequential_t, Index size, const Scalar& low, const
   * \sa LinSpaced(Scalar,Scalar)
   */
 template<typename Derived>
-EIGEN_STRONG_INLINE const typename DenseBase<Derived>::RandomAccessLinSpacedReturnType
+EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE const typename DenseBase<Derived>::RandomAccessLinSpacedReturnType
 DenseBase<Derived>::LinSpaced(Sequential_t, const Scalar& low, const Scalar& high)
 {
   EIGEN_STATIC_ASSERT_VECTOR_ONLY(Derived)
@@ -264,7 +264,7 @@ DenseBase<Derived>::LinSpaced(Sequential_t, const Scalar& low, const Scalar& hig
   * \sa setLinSpaced(Index,const Scalar&,const Scalar&), CwiseNullaryOp
   */
 template<typename Derived>
-EIGEN_STRONG_INLINE const typename DenseBase<Derived>::RandomAccessLinSpacedReturnType
+EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE const typename DenseBase<Derived>::RandomAccessLinSpacedReturnType
 DenseBase<Derived>::LinSpaced(Index size, const Scalar& low, const Scalar& high)
 {
   EIGEN_STATIC_ASSERT_VECTOR_ONLY(Derived)
@@ -276,7 +276,7 @@ DenseBase<Derived>::LinSpaced(Index size, const Scalar& low, const Scalar& high)
   * Special version for fixed size types which does not require the size parameter.
   */
 template<typename Derived>
-EIGEN_STRONG_INLINE const typename DenseBase<Derived>::RandomAccessLinSpacedReturnType
+EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE const typename DenseBase<Derived>::RandomAccessLinSpacedReturnType
 DenseBase<Derived>::LinSpaced(const Scalar& low, const Scalar& high)
 {
   EIGEN_STATIC_ASSERT_VECTOR_ONLY(Derived)
@@ -286,7 +286,7 @@ DenseBase<Derived>::LinSpaced(const Scalar& low, const Scalar& high)
 
 /** \returns true if all coefficients in this matrix are approximately equal to \a val, to within precision \a prec */
 template<typename Derived>
-bool DenseBase<Derived>::isApproxToConstant
+EIGEN_DEVICE_FUNC bool DenseBase<Derived>::isApproxToConstant
 (const Scalar& val, const RealScalar& prec) const
 {
   typename internal::nested_eval<Derived,1>::type self(derived());
@@ -301,7 +301,7 @@ bool DenseBase<Derived>::isApproxToConstant
   *
   * \returns true if all coefficients in this matrix are approximately equal to \a value, to within precision \a prec */
 template<typename Derived>
-bool DenseBase<Derived>::isConstant
+EIGEN_DEVICE_FUNC bool DenseBase<Derived>::isConstant
 (const Scalar& val, const RealScalar& prec) const
 {
   return isApproxToConstant(val, prec);
@@ -312,7 +312,7 @@ bool DenseBase<Derived>::isConstant
   * \sa setConstant(), Constant(), class CwiseNullaryOp
   */
 template<typename Derived>
-EIGEN_STRONG_INLINE void DenseBase<Derived>::fill(const Scalar& val)
+EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE void DenseBase<Derived>::fill(const Scalar& val)
 {
   setConstant(val);
 }
@@ -322,7 +322,7 @@ EIGEN_STRONG_INLINE void DenseBase<Derived>::fill(const Scalar& val)
   * \sa fill(), setConstant(Index,const Scalar&), setConstant(Index,Index,const Scalar&), setZero(), setOnes(), Constant(), class CwiseNullaryOp, setZero(), setOnes()
   */
 template<typename Derived>
-EIGEN_STRONG_INLINE Derived& DenseBase<Derived>::setConstant(const Scalar& val)
+EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE Derived& DenseBase<Derived>::setConstant(const Scalar& val)
 {
   return derived() = Constant(rows(), cols(), val);
 }
@@ -337,7 +337,7 @@ EIGEN_STRONG_INLINE Derived& DenseBase<Derived>::setConstant(const Scalar& val)
   * \sa MatrixBase::setConstant(const Scalar&), setConstant(Index,Index,const Scalar&), class CwiseNullaryOp, MatrixBase::Constant(const Scalar&)
   */
 template<typename Derived>
-EIGEN_STRONG_INLINE Derived&
+EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE Derived&
 PlainObjectBase<Derived>::setConstant(Index size, const Scalar& val)
 {
   resize(size);
@@ -356,7 +356,7 @@ PlainObjectBase<Derived>::setConstant(Index size, const Scalar& val)
   * \sa MatrixBase::setConstant(const Scalar&), setConstant(Index,const Scalar&), class CwiseNullaryOp, MatrixBase::Constant(const Scalar&)
   */
 template<typename Derived>
-EIGEN_STRONG_INLINE Derived&
+EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE Derived&
 PlainObjectBase<Derived>::setConstant(Index rows, Index cols, const Scalar& val)
 {
   resize(rows, cols);
@@ -380,7 +380,7 @@ PlainObjectBase<Derived>::setConstant(Index rows, Index cols, const Scalar& val)
   * \sa LinSpaced(Index,const Scalar&,const Scalar&), CwiseNullaryOp
   */
 template<typename Derived>
-EIGEN_STRONG_INLINE Derived& DenseBase<Derived>::setLinSpaced(Index newSize, const Scalar& low, const Scalar& high)
+EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE Derived& DenseBase<Derived>::setLinSpaced(Index newSize, const Scalar& low, const Scalar& high)
 {
   EIGEN_STATIC_ASSERT_VECTOR_ONLY(Derived)
   return derived() = Derived::NullaryExpr(newSize, internal::linspaced_op<Scalar,PacketScalar>(low,high,newSize));
@@ -400,7 +400,7 @@ EIGEN_STRONG_INLINE Derived& DenseBase<Derived>::setLinSpaced(Index newSize, con
   * \sa LinSpaced(Index,const Scalar&,const Scalar&), setLinSpaced(Index, const Scalar&, const Scalar&), CwiseNullaryOp
   */
 template<typename Derived>
-EIGEN_STRONG_INLINE Derived& DenseBase<Derived>::setLinSpaced(const Scalar& low, const Scalar& high)
+EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE Derived& DenseBase<Derived>::setLinSpaced(const Scalar& low, const Scalar& high)
 {
   EIGEN_STATIC_ASSERT_VECTOR_ONLY(Derived)
   return setLinSpaced(size(), low, high);
@@ -423,7 +423,7 @@ EIGEN_STRONG_INLINE Derived& DenseBase<Derived>::setLinSpaced(const Scalar& low,
   * \sa Zero(), Zero(Index)
   */
 template<typename Derived>
-EIGEN_STRONG_INLINE const typename DenseBase<Derived>::ConstantReturnType
+EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE const typename DenseBase<Derived>::ConstantReturnType
 DenseBase<Derived>::Zero(Index rows, Index cols)
 {
   return Constant(rows, cols, Scalar(0));
@@ -446,7 +446,7 @@ DenseBase<Derived>::Zero(Index rows, Index cols)
   * \sa Zero(), Zero(Index,Index)
   */
 template<typename Derived>
-EIGEN_STRONG_INLINE const typename DenseBase<Derived>::ConstantReturnType
+EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE const typename DenseBase<Derived>::ConstantReturnType
 DenseBase<Derived>::Zero(Index size)
 {
   return Constant(size, Scalar(0));
@@ -463,7 +463,7 @@ DenseBase<Derived>::Zero(Index size)
   * \sa Zero(Index), Zero(Index,Index)
   */
 template<typename Derived>
-EIGEN_STRONG_INLINE const typename DenseBase<Derived>::ConstantReturnType
+EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE const typename DenseBase<Derived>::ConstantReturnType
 DenseBase<Derived>::Zero()
 {
   return Constant(Scalar(0));
@@ -478,7 +478,7 @@ DenseBase<Derived>::Zero()
   * \sa class CwiseNullaryOp, Zero()
   */
 template<typename Derived>
-bool DenseBase<Derived>::isZero(const RealScalar& prec) const
+EIGEN_DEVICE_FUNC bool DenseBase<Derived>::isZero(const RealScalar& prec) const
 {
   typename internal::nested_eval<Derived,1>::type self(derived());
   for(Index j = 0; j < cols(); ++j)
@@ -496,7 +496,7 @@ bool DenseBase<Derived>::isZero(const RealScalar& prec) const
   * \sa class CwiseNullaryOp, Zero()
   */
 template<typename Derived>
-EIGEN_STRONG_INLINE Derived& DenseBase<Derived>::setZero()
+EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE Derived& DenseBase<Derived>::setZero()
 {
   return setConstant(Scalar(0));
 }
@@ -511,7 +511,7 @@ EIGEN_STRONG_INLINE Derived& DenseBase<Derived>::setZero()
   * \sa DenseBase::setZero(), setZero(Index,Index), class CwiseNullaryOp, DenseBase::Zero()
   */
 template<typename Derived>
-EIGEN_STRONG_INLINE Derived&
+EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE Derived&
 PlainObjectBase<Derived>::setZero(Index newSize)
 {
   resize(newSize);
@@ -529,7 +529,7 @@ PlainObjectBase<Derived>::setZero(Index newSize)
   * \sa DenseBase::setZero(), setZero(Index), class CwiseNullaryOp, DenseBase::Zero()
   */
 template<typename Derived>
-EIGEN_STRONG_INLINE Derived&
+EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE Derived&
 PlainObjectBase<Derived>::setZero(Index rows, Index cols)
 {
   resize(rows, cols);
@@ -553,7 +553,7 @@ PlainObjectBase<Derived>::setZero(Index rows, Index cols)
   * \sa Ones(), Ones(Index), isOnes(), class Ones
   */
 template<typename Derived>
-EIGEN_STRONG_INLINE const typename DenseBase<Derived>::ConstantReturnType
+EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE const typename DenseBase<Derived>::ConstantReturnType
 DenseBase<Derived>::Ones(Index rows, Index cols)
 {
   return Constant(rows, cols, Scalar(1));
@@ -576,7 +576,7 @@ DenseBase<Derived>::Ones(Index rows, Index cols)
   * \sa Ones(), Ones(Index,Index), isOnes(), class Ones
   */
 template<typename Derived>
-EIGEN_STRONG_INLINE const typename DenseBase<Derived>::ConstantReturnType
+EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE const typename DenseBase<Derived>::ConstantReturnType
 DenseBase<Derived>::Ones(Index newSize)
 {
   return Constant(newSize, Scalar(1));
@@ -593,7 +593,7 @@ DenseBase<Derived>::Ones(Index newSize)
   * \sa Ones(Index), Ones(Index,Index), isOnes(), class Ones
   */
 template<typename Derived>
-EIGEN_STRONG_INLINE const typename DenseBase<Derived>::ConstantReturnType
+EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE const typename DenseBase<Derived>::ConstantReturnType
 DenseBase<Derived>::Ones()
 {
   return Constant(Scalar(1));
@@ -608,7 +608,7 @@ DenseBase<Derived>::Ones()
   * \sa class CwiseNullaryOp, Ones()
   */
 template<typename Derived>
-bool DenseBase<Derived>::isOnes
+EIGEN_DEVICE_FUNC bool DenseBase<Derived>::isOnes
 (const RealScalar& prec) const
 {
   return isApproxToConstant(Scalar(1), prec);
@@ -622,7 +622,7 @@ bool DenseBase<Derived>::isOnes
   * \sa class CwiseNullaryOp, Ones()
   */
 template<typename Derived>
-EIGEN_STRONG_INLINE Derived& DenseBase<Derived>::setOnes()
+EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE Derived& DenseBase<Derived>::setOnes()
 {
   return setConstant(Scalar(1));
 }
@@ -637,7 +637,7 @@ EIGEN_STRONG_INLINE Derived& DenseBase<Derived>::setOnes()
   * \sa MatrixBase::setOnes(), setOnes(Index,Index), class CwiseNullaryOp, MatrixBase::Ones()
   */
 template<typename Derived>
-EIGEN_STRONG_INLINE Derived&
+EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE Derived&
 PlainObjectBase<Derived>::setOnes(Index newSize)
 {
   resize(newSize);
@@ -655,7 +655,7 @@ PlainObjectBase<Derived>::setOnes(Index newSize)
   * \sa MatrixBase::setOnes(), setOnes(Index), class CwiseNullaryOp, MatrixBase::Ones()
   */
 template<typename Derived>
-EIGEN_STRONG_INLINE Derived&
+EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE Derived&
 PlainObjectBase<Derived>::setOnes(Index rows, Index cols)
 {
   resize(rows, cols);
@@ -679,7 +679,7 @@ PlainObjectBase<Derived>::setOnes(Index rows, Index cols)
   * \sa Identity(), setIdentity(), isIdentity()
   */
 template<typename Derived>
-EIGEN_STRONG_INLINE const typename MatrixBase<Derived>::IdentityReturnType
+EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE const typename MatrixBase<Derived>::IdentityReturnType
 MatrixBase<Derived>::Identity(Index rows, Index cols)
 {
   return DenseBase<Derived>::NullaryExpr(rows, cols, internal::scalar_identity_op<Scalar>());
@@ -696,7 +696,7 @@ MatrixBase<Derived>::Identity(Index rows, Index cols)
   * \sa Identity(Index,Index), setIdentity(), isIdentity()
   */
 template<typename Derived>
-EIGEN_STRONG_INLINE const typename MatrixBase<Derived>::IdentityReturnType
+EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE const typename MatrixBase<Derived>::IdentityReturnType
 MatrixBase<Derived>::Identity()
 {
   EIGEN_STATIC_ASSERT_FIXED_SIZE(Derived)
@@ -771,7 +771,7 @@ struct setIdentity_impl<Derived, true>
   * \sa class CwiseNullaryOp, Identity(), Identity(Index,Index), isIdentity()
   */
 template<typename Derived>
-EIGEN_STRONG_INLINE Derived& MatrixBase<Derived>::setIdentity()
+EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE Derived& MatrixBase<Derived>::setIdentity()
 {
   return internal::setIdentity_impl<Derived>::run(derived());
 }
@@ -787,7 +787,7 @@ 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_DEVICE_FUNC EIGEN_STRONG_INLINE Derived& MatrixBase<Derived>::setIdentity(Index rows, Index cols)
 {
   derived().resize(rows, cols);
   return setIdentity();
@@ -800,7 +800,7 @@ EIGEN_STRONG_INLINE Derived& MatrixBase<Derived>::setIdentity(Index rows, Index
   * \sa MatrixBase::Unit(Index), MatrixBase::UnitX(), MatrixBase::UnitY(), MatrixBase::UnitZ(), MatrixBase::UnitW()
   */
 template<typename Derived>
-EIGEN_STRONG_INLINE const typename MatrixBase<Derived>::BasisReturnType MatrixBase<Derived>::Unit(Index newSize, Index i)
+EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE const typename MatrixBase<Derived>::BasisReturnType MatrixBase<Derived>::Unit(Index newSize, Index i)
 {
   EIGEN_STATIC_ASSERT_VECTOR_ONLY(Derived)
   return BasisReturnType(SquareMatrixType::Identity(newSize,newSize), i);
@@ -815,7 +815,7 @@ EIGEN_STRONG_INLINE const typename MatrixBase<Derived>::BasisReturnType MatrixBa
   * \sa MatrixBase::Unit(Index,Index), MatrixBase::UnitX(), MatrixBase::UnitY(), MatrixBase::UnitZ(), MatrixBase::UnitW()
   */
 template<typename Derived>
-EIGEN_STRONG_INLINE const typename MatrixBase<Derived>::BasisReturnType MatrixBase<Derived>::Unit(Index i)
+EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE const typename MatrixBase<Derived>::BasisReturnType MatrixBase<Derived>::Unit(Index i)
 {
   EIGEN_STATIC_ASSERT_VECTOR_ONLY(Derived)
   return BasisReturnType(SquareMatrixType::Identity(),i);
@@ -828,7 +828,7 @@ EIGEN_STRONG_INLINE const typename MatrixBase<Derived>::BasisReturnType MatrixBa
   * \sa MatrixBase::Unit(Index,Index), MatrixBase::Unit(Index), MatrixBase::UnitY(), MatrixBase::UnitZ(), MatrixBase::UnitW()
   */
 template<typename Derived>
-EIGEN_STRONG_INLINE const typename MatrixBase<Derived>::BasisReturnType MatrixBase<Derived>::UnitX()
+EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE const typename MatrixBase<Derived>::BasisReturnType MatrixBase<Derived>::UnitX()
 { return Derived::Unit(0); }
 
 /** \returns an expression of the Y axis unit vector (0,1{,0}^*)
@@ -838,7 +838,7 @@ EIGEN_STRONG_INLINE const typename MatrixBase<Derived>::BasisReturnType MatrixBa
   * \sa MatrixBase::Unit(Index,Index), MatrixBase::Unit(Index), MatrixBase::UnitY(), MatrixBase::UnitZ(), MatrixBase::UnitW()
   */
 template<typename Derived>
-EIGEN_STRONG_INLINE const typename MatrixBase<Derived>::BasisReturnType MatrixBase<Derived>::UnitY()
+EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE const typename MatrixBase<Derived>::BasisReturnType MatrixBase<Derived>::UnitY()
 { return Derived::Unit(1); }
 
 /** \returns an expression of the Z axis unit vector (0,0,1{,0}^*)
@@ -848,7 +848,7 @@ EIGEN_STRONG_INLINE const typename MatrixBase<Derived>::BasisReturnType MatrixBa
   * \sa MatrixBase::Unit(Index,Index), MatrixBase::Unit(Index), MatrixBase::UnitY(), MatrixBase::UnitZ(), MatrixBase::UnitW()
   */
 template<typename Derived>
-EIGEN_STRONG_INLINE const typename MatrixBase<Derived>::BasisReturnType MatrixBase<Derived>::UnitZ()
+EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE const typename MatrixBase<Derived>::BasisReturnType MatrixBase<Derived>::UnitZ()
 { return Derived::Unit(2); }
 
 /** \returns an expression of the W axis unit vector (0,0,0,1)
@@ -858,7 +858,7 @@ EIGEN_STRONG_INLINE const typename MatrixBase<Derived>::BasisReturnType MatrixBa
   * \sa MatrixBase::Unit(Index,Index), MatrixBase::Unit(Index), MatrixBase::UnitY(), MatrixBase::UnitZ(), MatrixBase::UnitW()
   */
 template<typename Derived>
-EIGEN_STRONG_INLINE const typename MatrixBase<Derived>::BasisReturnType MatrixBase<Derived>::UnitW()
+EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE const typename MatrixBase<Derived>::BasisReturnType MatrixBase<Derived>::UnitW()
 { return Derived::Unit(3); }
 
 } // end namespace Eigen

Eigen/src/Core/DenseBase.h

@@ -296,7 +296,7 @@ template<typename Derived> class DenseBase
     EIGEN_DEVICE_FUNC
     Derived& operator=(const ReturnByValue<OtherDerived>& func);
 
-    /** \ínternal
+    /** \internal
       * Copies \a other into *this without evaluating other. \returns a reference to *this.
       * \deprecated */
     template<typename OtherDerived>
@@ -463,7 +463,17 @@ template<typename Derived> class DenseBase
     EIGEN_DEVICE_FUNC
     void visit(Visitor& func) const;
 
-    inline const WithFormat<Derived> format(const IOFormat& fmt) const;
+    /** \returns a WithFormat proxy object allowing to print a matrix the with given
+      * format \a fmt.
+      *
+      * See class IOFormat for some examples.
+      *
+      * \sa class IOFormat, class WithFormat
+      */
+    inline const WithFormat<Derived> format(const IOFormat& fmt) const
+    {
+      return WithFormat<Derived>(derived(), fmt);
+    }
 
     /** \returns the unique coefficient of a 1x1 expression */
     EIGEN_DEVICE_FUNC
@@ -474,9 +484,9 @@ template<typename Derived> class DenseBase
       return derived().coeff(0,0);
     }
 
-    bool all() const;
-    bool any() const;
-    Index count() const;
+    EIGEN_DEVICE_FUNC bool all() const;
+    EIGEN_DEVICE_FUNC bool any() const;
+    EIGEN_DEVICE_FUNC Index count() const;
 
     typedef VectorwiseOp<Derived, Horizontal> RowwiseReturnType;
     typedef const VectorwiseOp<const Derived, Horizontal> ConstRowwiseReturnType;

Eigen/src/Core/DenseStorage.h

@@ -13,9 +13,9 @@
 #define EIGEN_MATRIXSTORAGE_H
 
 #ifdef EIGEN_DENSE_STORAGE_CTOR_PLUGIN
-  #define EIGEN_INTERNAL_DENSE_STORAGE_CTOR_PLUGIN EIGEN_DENSE_STORAGE_CTOR_PLUGIN;
+  #define EIGEN_INTERNAL_DENSE_STORAGE_CTOR_PLUGIN(X) X; EIGEN_DENSE_STORAGE_CTOR_PLUGIN;
 #else
-  #define EIGEN_INTERNAL_DENSE_STORAGE_CTOR_PLUGIN
+  #define EIGEN_INTERNAL_DENSE_STORAGE_CTOR_PLUGIN(X)
 #endif
 
 namespace Eigen {
@@ -184,12 +184,16 @@ 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 DenseStorage() {
+      EIGEN_INTERNAL_DENSE_STORAGE_CTOR_PLUGIN(Index size = Size)
+    }
     EIGEN_DEVICE_FUNC
     explicit DenseStorage(internal::constructor_without_unaligned_array_assert)
       : m_data(internal::constructor_without_unaligned_array_assert()) {}
     EIGEN_DEVICE_FUNC 
-    DenseStorage(const DenseStorage& other) : m_data(other.m_data) {}
+    DenseStorage(const DenseStorage& other) : m_data(other.m_data) {
+      EIGEN_INTERNAL_DENSE_STORAGE_CTOR_PLUGIN(Index size = Size)
+    }
     EIGEN_DEVICE_FUNC 
     DenseStorage& operator=(const DenseStorage& other)
     { 
@@ -197,7 +201,7 @@ template<typename T, int Size, int _Rows, int _Cols, int _Options> class DenseSt
       return *this; 
     }
     EIGEN_DEVICE_FUNC DenseStorage(Index size, Index rows, Index cols) {
-      EIGEN_INTERNAL_DENSE_STORAGE_CTOR_PLUGIN
+      EIGEN_INTERNAL_DENSE_STORAGE_CTOR_PLUGIN({})
       eigen_internal_assert(size==rows*cols && rows==_Rows && cols==_Cols);
       EIGEN_UNUSED_VARIABLE(size);
       EIGEN_UNUSED_VARIABLE(rows);
@@ -343,7 +347,7 @@ template<typename T, int _Options> class DenseStorage<T, Dynamic, Dynamic, Dynam
     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_DENSE_STORAGE_CTOR_PLUGIN({})
       eigen_internal_assert(size==rows*cols && rows>=0 && cols >=0);
     }
     EIGEN_DEVICE_FUNC DenseStorage(const DenseStorage& other)
@@ -351,6 +355,7 @@ template<typename T, int _Options> class DenseStorage<T, Dynamic, Dynamic, Dynam
       , m_rows(other.m_rows)
       , m_cols(other.m_cols)
     {
+      EIGEN_INTERNAL_DENSE_STORAGE_CTOR_PLUGIN(Index size = m_rows*m_cols)
       internal::smart_copy(other.m_data, other.m_data+other.m_rows*other.m_cols, m_data);
     }
     EIGEN_DEVICE_FUNC DenseStorage& operator=(const DenseStorage& other)
@@ -403,7 +408,7 @@ template<typename T, int _Options> class DenseStorage<T, Dynamic, Dynamic, Dynam
           m_data = internal::conditional_aligned_new_auto<T,(_Options&DontAlign)==0>(size);
         else
           m_data = 0;
-        EIGEN_INTERNAL_DENSE_STORAGE_CTOR_PLUGIN
+        EIGEN_INTERNAL_DENSE_STORAGE_CTOR_PLUGIN({})
       }
       m_rows = rows;
       m_cols = cols;
@@ -422,7 +427,7 @@ template<typename T, int _Rows, int _Options> class DenseStorage<T, Dynamic, _Ro
     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_DENSE_STORAGE_CTOR_PLUGIN({})
       eigen_internal_assert(size==rows*cols && rows==_Rows && cols >=0);
       EIGEN_UNUSED_VARIABLE(rows);
     }
@@ -430,6 +435,7 @@ template<typename T, int _Rows, int _Options> class DenseStorage<T, Dynamic, _Ro
       : m_data(internal::conditional_aligned_new_auto<T,(_Options&DontAlign)==0>(_Rows*other.m_cols))
       , m_cols(other.m_cols)
     {
+      EIGEN_INTERNAL_DENSE_STORAGE_CTOR_PLUGIN(Index size = m_cols*_Rows)
       internal::smart_copy(other.m_data, other.m_data+_Rows*m_cols, m_data);
     }
     EIGEN_DEVICE_FUNC DenseStorage& operator=(const DenseStorage& other)
@@ -477,7 +483,7 @@ template<typename T, int _Rows, int _Options> class DenseStorage<T, Dynamic, _Ro
           m_data = internal::conditional_aligned_new_auto<T,(_Options&DontAlign)==0>(size);
         else
           m_data = 0;
-        EIGEN_INTERNAL_DENSE_STORAGE_CTOR_PLUGIN
+        EIGEN_INTERNAL_DENSE_STORAGE_CTOR_PLUGIN({})
       }
       m_cols = cols;
     }
@@ -495,7 +501,7 @@ template<typename T, int _Cols, int _Options> class DenseStorage<T, Dynamic, Dyn
     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_DENSE_STORAGE_CTOR_PLUGIN({})
       eigen_internal_assert(size==rows*cols && rows>=0 && cols == _Cols);
       EIGEN_UNUSED_VARIABLE(cols);
     }
@@ -503,6 +509,7 @@ template<typename T, int _Cols, int _Options> class DenseStorage<T, Dynamic, Dyn
       : m_data(internal::conditional_aligned_new_auto<T,(_Options&DontAlign)==0>(other.m_rows*_Cols))
       , m_rows(other.m_rows)
     {
+      EIGEN_INTERNAL_DENSE_STORAGE_CTOR_PLUGIN(Index size = m_rows*_Cols)
       internal::smart_copy(other.m_data, other.m_data+other.m_rows*_Cols, m_data);
     }
     EIGEN_DEVICE_FUNC DenseStorage& operator=(const DenseStorage& other)
@@ -550,7 +557,7 @@ template<typename T, int _Cols, int _Options> class DenseStorage<T, Dynamic, Dyn
           m_data = internal::conditional_aligned_new_auto<T,(_Options&DontAlign)==0>(size);
         else
           m_data = 0;
-        EIGEN_INTERNAL_DENSE_STORAGE_CTOR_PLUGIN
+        EIGEN_INTERNAL_DENSE_STORAGE_CTOR_PLUGIN({})
       }
       m_rows = rows;
     }

Eigen/src/Core/Diagonal.h

@@ -21,7 +21,7 @@ namespace Eigen {
   * \param MatrixType the type of the object in which we are taking a sub/main/super diagonal
   * \param DiagIndex the index of the sub/super diagonal. The default is 0 and it means the main diagonal.
   *              A positive value means a superdiagonal, a negative value means a subdiagonal.
-  *              You can also use Dynamic so the index can be set at runtime.
+  *              You can also use DynamicIndex so the index can be set at runtime.
   *
   * The matrix is not required to be square.
   *

Eigen/src/Core/Dot.h

@@ -51,7 +51,8 @@ struct dot_nocheck<T, U, true>
 
 } // end namespace internal
 
-/** \returns the dot product of *this with other.
+/** \fn MatrixBase::dot
+  * \returns the dot product of *this with other.
   *
   * \only_for_vectors
   *
@@ -70,9 +71,11 @@ 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);

Eigen/src/Core/EigenBase.h

@@ -14,6 +14,7 @@
 namespace Eigen {
 
 /** \class EigenBase
+  * \ingroup Core_Module
   * 
   * Common base class for all classes T such that MatrixBase has an operator=(T) and a constructor MatrixBase(T).
   *
@@ -128,6 +129,7 @@ template<typename Derived> struct EigenBase
   */
 template<typename Derived>
 template<typename OtherDerived>
+EIGEN_DEVICE_FUNC
 Derived& DenseBase<Derived>::operator=(const EigenBase<OtherDerived> &other)
 {
   call_assignment(derived(), other.derived());
@@ -136,6 +138,7 @@ Derived& DenseBase<Derived>::operator=(const EigenBase<OtherDerived> &other)
 
 template<typename Derived>
 template<typename OtherDerived>
+EIGEN_DEVICE_FUNC
 Derived& DenseBase<Derived>::operator+=(const EigenBase<OtherDerived> &other)
 {
   call_assignment(derived(), other.derived(), internal::add_assign_op<Scalar,typename OtherDerived::Scalar>());
@@ -144,6 +147,7 @@ Derived& DenseBase<Derived>::operator+=(const EigenBase<OtherDerived> &other)
 
 template<typename Derived>
 template<typename OtherDerived>
+EIGEN_DEVICE_FUNC
 Derived& DenseBase<Derived>::operator-=(const EigenBase<OtherDerived> &other)
 {
   call_assignment(derived(), other.derived(), internal::sub_assign_op<Scalar,typename OtherDerived::Scalar>());

Eigen/src/Core/GeneralProduct.h

@@ -25,7 +25,8 @@ 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 >= EIGEN_CACHEFRIENDLY_PRODUCT_THRESHOLD ||
+                    (Size==Dynamic && MaxSize>=EIGEN_CACHEFRIENDLY_PRODUCT_THRESHOLD),
          value = is_large  ? Large
                : Size == 1 ? 1
                            : Small
@@ -223,50 +224,65 @@ template<> struct gemv_dense_selector<OnTheRight,ColMajor,true>
       // on, the other hand it is good for the cache to pack the vector anyways...
       EvalToDestAtCompileTime = (ActualDest::InnerStrideAtCompileTime==1),
       ComplexByReal = (NumTraits<LhsScalar>::IsComplex) && (!NumTraits<RhsScalar>::IsComplex),
-      MightCannotUseDest = (ActualDest::InnerStrideAtCompileTime!=1) || ComplexByReal
+      MightCannotUseDest = (!EvalToDestAtCompileTime) || ComplexByReal
     };
 
-    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;
-
-    RhsScalar compatibleAlpha = get_factor<ResScalar,RhsScalar>::run(actualAlpha);
-
-    ei_declare_aligned_stack_constructed_variable(ResScalar,actualDestPtr,dest.size(),
-                                                  evalToDest ? dest.data() : static_dest.data());
-
-    if(!evalToDest)
-    {
-      #ifdef EIGEN_DENSE_STORAGE_CTOR_PLUGIN
-      Index size = dest.size();
-      EIGEN_DENSE_STORAGE_CTOR_PLUGIN
-      #endif
-      if(!alphaIsCompatible)
-      {
-        MappedDest(actualDestPtr, dest.size()).setZero();
-        compatibleAlpha = RhsScalar(1);
-      }
-      else
-        MappedDest(actualDestPtr, dest.size()) = dest;
-    }
-
     typedef const_blas_data_mapper<LhsScalar,Index,ColMajor> LhsMapper;
     typedef const_blas_data_mapper<RhsScalar,Index,RowMajor> RhsMapper;
-    general_matrix_vector_product
-        <Index,LhsScalar,LhsMapper,ColMajor,LhsBlasTraits::NeedToConjugate,RhsScalar,RhsMapper,RhsBlasTraits::NeedToConjugate>::run(
-        actualLhs.rows(), actualLhs.cols(),
-        LhsMapper(actualLhs.data(), actualLhs.outerStride()),
-        RhsMapper(actualRhs.data(), actualRhs.innerStride()),
-        actualDestPtr, 1,
-        compatibleAlpha);
+    RhsScalar compatibleAlpha = get_factor<ResScalar,RhsScalar>::run(actualAlpha);
 
-    if (!evalToDest)
+    if(!MightCannotUseDest)
     {
-      if(!alphaIsCompatible)
-        dest += actualAlpha * MappedDest(actualDestPtr, dest.size());
-      else
-        dest = MappedDest(actualDestPtr, dest.size());
+      // shortcut if we are sure to be able to use dest directly,
+      // this ease the compiler to generate cleaner and more optimzized code for most common cases
+      general_matrix_vector_product
+          <Index,LhsScalar,LhsMapper,ColMajor,LhsBlasTraits::NeedToConjugate,RhsScalar,RhsMapper,RhsBlasTraits::NeedToConjugate>::run(
+          actualLhs.rows(), actualLhs.cols(),
+          LhsMapper(actualLhs.data(), actualLhs.outerStride()),
+          RhsMapper(actualRhs.data(), actualRhs.innerStride()),
+          dest.data(), 1,
+          compatibleAlpha);
+    }
+    else
+    {
+      gemv_static_vector_if<ResScalar,ActualDest::SizeAtCompileTime,ActualDest::MaxSizeAtCompileTime,MightCannotUseDest> static_dest;
+
+      const bool alphaIsCompatible = (!ComplexByReal) || (numext::imag(actualAlpha)==RealScalar(0));
+      const bool evalToDest = EvalToDestAtCompileTime && alphaIsCompatible;
+
+      ei_declare_aligned_stack_constructed_variable(ResScalar,actualDestPtr,dest.size(),
+                                                    evalToDest ? dest.data() : static_dest.data());
+
+      if(!evalToDest)
+      {
+        #ifdef EIGEN_DENSE_STORAGE_CTOR_PLUGIN
+        Index size = dest.size();
+        EIGEN_DENSE_STORAGE_CTOR_PLUGIN
+        #endif
+        if(!alphaIsCompatible)
+        {
+          MappedDest(actualDestPtr, dest.size()).setZero();
+          compatibleAlpha = RhsScalar(1);
+        }
+        else
+          MappedDest(actualDestPtr, dest.size()) = dest;
+      }
+
+      general_matrix_vector_product
+          <Index,LhsScalar,LhsMapper,ColMajor,LhsBlasTraits::NeedToConjugate,RhsScalar,RhsMapper,RhsBlasTraits::NeedToConjugate>::run(
+          actualLhs.rows(), actualLhs.cols(),
+          LhsMapper(actualLhs.data(), actualLhs.outerStride()),
+          RhsMapper(actualRhs.data(), actualRhs.innerStride()),
+          actualDestPtr, 1,
+          compatibleAlpha);
+
+      if (!evalToDest)
+      {
+        if(!alphaIsCompatible)
+          dest.matrix() += actualAlpha * MappedDest(actualDestPtr, dest.size());
+        else
+          dest = MappedDest(actualDestPtr, dest.size());
+      }
     }
   }
 };
@@ -329,6 +345,7 @@ template<> struct gemv_dense_selector<OnTheRight,ColMajor,false>
   template<typename Lhs, typename Rhs, typename Dest>
   static void run(const Lhs &lhs, const Rhs &rhs, Dest& dest, const typename Dest::Scalar& alpha)
   {
+    EIGEN_STATIC_ASSERT((!nested_eval<Lhs,1>::Evaluate),EIGEN_INTERNAL_COMPILATION_ERROR_OR_YOU_MADE_A_PROGRAMMING_MISTAKE);
     // TODO if rhs is large enough it might be beneficial to make sure that dest is sequentially stored in memory, otherwise use a temp
     typename nested_eval<Rhs,1>::type actual_rhs(rhs);
     const Index size = rhs.rows();
@@ -342,6 +359,7 @@ template<> struct gemv_dense_selector<OnTheRight,RowMajor,false>
   template<typename Lhs, typename Rhs, typename Dest>
   static void run(const Lhs &lhs, const Rhs &rhs, Dest& dest, const typename Dest::Scalar& alpha)
   {
+    EIGEN_STATIC_ASSERT((!nested_eval<Lhs,1>::Evaluate),EIGEN_INTERNAL_COMPILATION_ERROR_OR_YOU_MADE_A_PROGRAMMING_MISTAKE);
     typename nested_eval<Rhs,Lhs::RowsAtCompileTime>::type actual_rhs(rhs);
     const Index rows = dest.rows();
     for(Index i=0; i<rows; ++i)

Eigen/src/Core/GenericPacketMath.h

@@ -230,7 +230,7 @@ pload1(const typename unpacket_traits<Packet>::type  *a) { return pset1<Packet>(
   * 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> EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE Packet
 ploaddup(const typename unpacket_traits<Packet>::type* from) { return *from; }
 
 /** \internal \returns a packet with elements of \a *from quadrupled.
@@ -278,7 +278,7 @@ inline void pbroadcast2(const typename unpacket_traits<Packet>::type *a,
 }
 
 /** \internal \brief Returns a packet with coefficients (a,a+1,...,a+packet_size-1). */
-template<typename Packet> inline Packet
+template<typename Packet> EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE Packet
 plset(const typename unpacket_traits<Packet>::type& a) { return a; }
 
 /** \internal copy the packet \a from to \a *to, \a to must be 16 bytes aligned */
@@ -482,7 +482,7 @@ EIGEN_DEVICE_FUNC EIGEN_ALWAYS_INLINE void pstoret(Scalar* to, const Packet& fro
   * by the current computation.
   */
 template<typename Packet, int LoadMode>
-inline Packet ploadt_ro(const typename unpacket_traits<Packet>::type* from)
+EIGEN_DEVICE_FUNC EIGEN_ALWAYS_INLINE Packet ploadt_ro(const typename unpacket_traits<Packet>::type* from)
 {
   return ploadt<Packet, LoadMode>(from);
 }

Eigen/src/Core/IO.h

@@ -105,24 +105,10 @@ class WithFormat
     }
 
   protected:
-    const typename ExpressionType::Nested m_matrix;
+    typename ExpressionType::Nested m_matrix;
     IOFormat m_format;
 };
 
-/** \returns a WithFormat proxy object allowing to print a matrix the with given
-  * format \a fmt.
-  *
-  * See class IOFormat for some examples.
-  *
-  * \sa class IOFormat, class WithFormat
-  */
-template<typename Derived>
-inline const WithFormat<Derived>
-DenseBase<Derived>::format(const IOFormat& fmt) const
-{
-  return WithFormat<Derived>(derived(), fmt);
-}
-
 namespace internal {
 
 // NOTE: This helper is kept for backward compatibility with previous code specializing

Eigen/src/Core/Inverse.h

@@ -45,6 +45,7 @@ class Inverse : public InverseImpl<XprType,typename internal::traits<XprType>::S
 public:
   typedef typename XprType::StorageIndex StorageIndex;
   typedef typename XprType::PlainObject                       PlainObject;
+  typedef typename XprType::Scalar                            Scalar;
   typedef typename internal::ref_selector<XprType>::type      XprTypeNested;
   typedef typename internal::remove_all<XprTypeNested>::type  XprTypeNestedCleaned;
   typedef typename internal::ref_selector<Inverse>::type Nested;

Eigen/src/Core/MathFunctions.h

@@ -1061,11 +1061,24 @@ double log(const double &x) { return ::log(x); }
 
 template<typename T>
 EIGEN_DEVICE_FUNC EIGEN_ALWAYS_INLINE
-typename NumTraits<T>::Real abs(const T &x) {
+typename internal::enable_if<NumTraits<T>::IsSigned || NumTraits<T>::IsComplex,typename NumTraits<T>::Real>::type
+abs(const T &x) {
   EIGEN_USING_STD_MATH(abs);
   return abs(x);
 }
 
+template<typename T>
+EIGEN_DEVICE_FUNC EIGEN_ALWAYS_INLINE
+typename internal::enable_if<!(NumTraits<T>::IsSigned || NumTraits<T>::IsComplex),typename NumTraits<T>::Real>::type
+abs(const T &x) {
+  return x;
+}
+
+#if defined(__SYCL_DEVICE_ONLY__)
+EIGEN_ALWAYS_INLINE float   abs(float x) { return cl::sycl::fabs(x); }
+EIGEN_ALWAYS_INLINE double  abs(double x) { return cl::sycl::fabs(x); }
+#endif // defined(__SYCL_DEVICE_ONLY__)
+
 #ifdef __CUDACC__
 template<> EIGEN_DEVICE_FUNC EIGEN_ALWAYS_INLINE
 float abs(const float &x) { return ::fabsf(x); }

Eigen/src/Core/MatrixBase.h

@@ -294,7 +294,7 @@ template<typename Derived> class MatrixBase
       *          fuzzy comparison such as isApprox()
       * \sa isApprox(), operator!= */
     template<typename OtherDerived>
-    inline bool operator==(const MatrixBase<OtherDerived>& other) const
+    EIGEN_DEVICE_FUNC inline bool operator==(const MatrixBase<OtherDerived>& other) const
     { return cwiseEqual(other).all(); }
 
     /** \returns true if at least one pair of coefficients of \c *this and \a other are not exactly equal to each other.
@@ -302,7 +302,7 @@ template<typename Derived> class MatrixBase
       *          fuzzy comparison such as isApprox()
       * \sa isApprox(), operator== */
     template<typename OtherDerived>
-    inline bool operator!=(const MatrixBase<OtherDerived>& other) const
+    EIGEN_DEVICE_FUNC inline bool operator!=(const MatrixBase<OtherDerived>& other) const
     { return cwiseNotEqual(other).any(); }
 
     NoAlias<Derived,Eigen::MatrixBase > noalias();

Eigen/src/Core/NumTraits.h

@@ -215,6 +215,8 @@ struct NumTraits<Array<Scalar, Rows, Cols, Options, MaxRows, MaxCols> >
   static inline RealScalar epsilon() { return NumTraits<RealScalar>::epsilon(); }
   EIGEN_DEVICE_FUNC
   static inline RealScalar dummy_precision() { return NumTraits<RealScalar>::dummy_precision(); }
+
+  static inline int digits10() { return NumTraits<Scalar>::digits10(); }
 };
 
 template<> struct NumTraits<std::string>

Eigen/src/Core/PlainObjectBase.h

@@ -41,7 +41,7 @@ template<> struct check_rows_cols_for_overflow<Dynamic> {
   {
     // http://hg.mozilla.org/mozilla-central/file/6c8a909977d3/xpcom/ds/CheckedInt.h#l242
     // we assume Index is signed
-    Index max_index = (size_t(1) << (8 * sizeof(Index) - 1)) - 1; // assume Index is signed
+    Index max_index = (std::size_t(1) << (8 * sizeof(Index) - 1)) - 1; // assume Index is signed
     bool error = (rows == 0 || cols == 0) ? false
                : (rows > max_index / cols);
     if (error)
@@ -58,6 +58,28 @@ template<typename MatrixTypeA, typename MatrixTypeB, bool SwapPointers> struct m
 
 } // end namespace internal
 
+#ifdef EIGEN_PARSED_BY_DOXYGEN
+namespace doxygen {
+
+// This is a workaround to doxygen not being able to understand the inheritance logic
+// when it is hidden by the dense_xpr_base helper struct.
+// Moreover, doxygen fails to include members that are not documented in the declaration body of
+// MatrixBase if we inherits MatrixBase<Matrix<_Scalar, _Rows, _Cols, _Options, _MaxRows, _MaxCols> >,
+// this is why we simply inherits MatrixBase, though this does not make sense.
+
+/** This class is just a workaround for Doxygen and it does not not actually exist. */
+template<typename Derived> struct dense_xpr_base_dispatcher;
+/** 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> >
+    : public MatrixBase {};
+/** 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> >
+    : public ArrayBase {};
+
+} // namespace doxygen
+
 /** \class PlainObjectBase
   * \ingroup Core_Module
   * \brief %Dense storage base class for matrices and arrays.
@@ -65,26 +87,10 @@ template<typename MatrixTypeA, typename MatrixTypeB, bool SwapPointers> struct m
   * 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.
   *
+  * \tparam Derived is the derived type, e.g., a Matrix or Array
+  *
   * \sa \ref TopicClassHierarchy
   */
-#ifdef EIGEN_PARSED_BY_DOXYGEN
-namespace doxygen {
-
-// this is a workaround to doxygen not being able to understand the inheritance logic
-// when it is hidden by the dense_xpr_base helper struct.
-/** This class is just a workaround for Doxygen and it does not not actually exist. */
-template<typename Derived> struct dense_xpr_base_dispatcher;
-/** 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> >
-    : 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> >
-    : public ArrayBase<Array<_Scalar, _Rows, _Cols, _Options, _MaxRows, _MaxCols> > {};
-
-} // namespace doxygen
-
 template<typename Derived>
 class PlainObjectBase : public doxygen::dense_xpr_base_dispatcher<Derived>
 #else
@@ -554,7 +560,8 @@ class PlainObjectBase : public internal::dense_xpr_base<Derived>::type
 
   public:
 
-    /** \copydoc DenseBase::operator=(const EigenBase<OtherDerived>&)
+    /** \brief Copies the generic expression \a other into *this.
+      * \copydetails DenseBase::operator=(const EigenBase<OtherDerived> &other)
       */
     template<typename OtherDerived>
     EIGEN_DEVICE_FUNC 
@@ -763,6 +770,7 @@ class PlainObjectBase : public internal::dense_xpr_base<Derived>::type
     {
       // 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);
@@ -804,6 +812,13 @@ class PlainObjectBase : public internal::dense_xpr_base<Derived>::type
       this->_set_noalias(other);
     }
 
+    // Initialize an arbitrary matrix from an object convertible to the Derived type.
+    template<typename T>
+    EIGEN_DEVICE_FUNC
+    EIGEN_STRONG_INLINE void _init1(const Derived& other){
+      this->_set_noalias(other);
+    }
+
     // Initialize an arbitrary matrix from a generic Eigen expression
     template<typename T, typename OtherDerived>
     EIGEN_DEVICE_FUNC
@@ -826,7 +841,7 @@ class PlainObjectBase : public internal::dense_xpr_base<Derived>::type
       this->derived() = r;
     }
     
-    // For fixed -size arrays:
+    // For fixed-size Array<Scalar,...>
     template<typename T>
     EIGEN_DEVICE_FUNC
     EIGEN_STRONG_INLINE void _init1(const Scalar& val0,
@@ -838,6 +853,7 @@ class PlainObjectBase : public internal::dense_xpr_base<Derived>::type
       Base::setConstant(val0);
     }
     
+    // For fixed-size Array<Index,...>
     template<typename T>
     EIGEN_DEVICE_FUNC
     EIGEN_STRONG_INLINE void _init1(const Index& val0,

Eigen/src/Core/ProductEvaluators.h

@@ -158,10 +158,7 @@ struct Assignment<DstXprType, Product<Lhs,Rhs,Options>, internal::add_assign_op<
   static EIGEN_STRONG_INLINE
   void run(DstXprType &dst, const SrcXprType &src, const internal::add_assign_op<Scalar,Scalar> &)
   {
-    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());
     // FIXME shall we handle nested_eval here?
     generic_product_impl<Lhs, Rhs>::addTo(dst, src.lhs(), src.rhs());
   }
@@ -176,10 +173,7 @@ struct Assignment<DstXprType, Product<Lhs,Rhs,Options>, internal::sub_assign_op<
   static EIGEN_STRONG_INLINE
   void run(DstXprType &dst, const SrcXprType &src, const internal::sub_assign_op<Scalar,Scalar> &)
   {
-    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());
     // FIXME shall we handle nested_eval here?
     generic_product_impl<Lhs, Rhs>::subTo(dst, src.lhs(), src.rhs());
   }
@@ -213,6 +207,12 @@ struct evaluator_assume_aliasing<CwiseBinaryOp<internal::scalar_sum_op<typename
   static const bool value = true;
 };
 
+template<typename OtherXpr, typename Lhs, typename Rhs>
+struct evaluator_assume_aliasing<CwiseBinaryOp<internal::scalar_difference_op<typename OtherXpr::Scalar,typename Product<Lhs,Rhs,DefaultProduct>::Scalar>, const OtherXpr,
+                                               const Product<Lhs,Rhs,DefaultProduct> >, DenseShape > {
+  static const bool value = true;
+};
+
 template<typename DstXprType, typename OtherXpr, typename ProductType, typename Func1, typename Func2>
 struct assignment_from_xpr_op_product
 {
@@ -366,17 +366,21 @@ template<typename Lhs, typename Rhs>
 struct generic_product_impl<Lhs,Rhs,DenseShape,DenseShape,GemvProduct>
   : generic_product_impl_base<Lhs,Rhs,generic_product_impl<Lhs,Rhs,DenseShape,DenseShape,GemvProduct> >
 {
+  typedef typename nested_eval<Lhs,1>::type LhsNested;
+  typedef typename nested_eval<Rhs,1>::type RhsNested;
   typedef typename Product<Lhs,Rhs>::Scalar Scalar;
   enum { Side = Lhs::IsVectorAtCompileTime ? OnTheLeft : OnTheRight };
-  typedef typename internal::conditional<int(Side)==OnTheRight,Lhs,Rhs>::type MatrixType;
+  typedef typename internal::remove_all<typename internal::conditional<int(Side)==OnTheRight,LhsNested,RhsNested>::type>::type MatrixType;
 
   template<typename Dest>
   static EIGEN_STRONG_INLINE void scaleAndAddTo(Dest& dst, const Lhs& lhs, const Rhs& rhs, const Scalar& alpha)
   {
+    LhsNested actual_lhs(lhs);
+    RhsNested actual_rhs(rhs);
     internal::gemv_dense_selector<Side,
                             (int(MatrixType::Flags)&RowMajorBit) ? RowMajor : ColMajor,
                             bool(internal::blas_traits<MatrixType>::HasUsableDirectAccess)
-                           >::run(lhs, rhs, dst, alpha);
+                           >::run(actual_lhs, actual_rhs, dst, alpha);
   }
 };
 

Eigen/src/Core/SelfAdjointView.h

@@ -45,7 +45,7 @@ struct traits<SelfAdjointView<MatrixType, UpLo> > : traits<MatrixType>
 };
 }
 
-// FIXME could also be called SelfAdjointWrapper to be consistent with DiagonalWrapper ??
+
 template<typename _MatrixType, unsigned int UpLo> class SelfAdjointView
   : public TriangularBase<SelfAdjointView<_MatrixType, UpLo> >
 {
@@ -60,10 +60,12 @@ template<typename _MatrixType, unsigned int UpLo> class SelfAdjointView
     /** \brief The type of coefficients in this matrix */
     typedef typename internal::traits<SelfAdjointView>::Scalar Scalar; 
     typedef typename MatrixType::StorageIndex StorageIndex;
+    typedef typename internal::remove_all<typename MatrixType::ConjugateReturnType>::type MatrixConjugateReturnType;
 
     enum {
       Mode = internal::traits<SelfAdjointView>::Mode,
-      Flags = internal::traits<SelfAdjointView>::Flags
+      Flags = internal::traits<SelfAdjointView>::Flags,
+      TransposeMode = ((Mode & Upper) ? Lower : 0) | ((Mode & Lower) ? Upper : 0)
     };
     typedef typename MatrixType::PlainObject PlainObject;
 
@@ -187,6 +189,36 @@ template<typename _MatrixType, unsigned int UpLo> class SelfAdjointView
                                    TriangularView<typename MatrixType::AdjointReturnType,TriMode> >::type(tmp2);
     }
 
+    typedef SelfAdjointView<const MatrixConjugateReturnType,Mode> ConjugateReturnType;
+    /** \sa MatrixBase::conjugate() const */
+    EIGEN_DEVICE_FUNC
+    inline const ConjugateReturnType conjugate() const
+    { return ConjugateReturnType(m_matrix.conjugate()); }
+
+    typedef SelfAdjointView<const typename MatrixType::AdjointReturnType,TransposeMode> AdjointReturnType;
+    /** \sa MatrixBase::adjoint() const */
+    EIGEN_DEVICE_FUNC
+    inline const AdjointReturnType adjoint() const
+    { return AdjointReturnType(m_matrix.adjoint()); }
+
+    typedef SelfAdjointView<typename MatrixType::TransposeReturnType,TransposeMode> TransposeReturnType;
+     /** \sa MatrixBase::transpose() */
+    EIGEN_DEVICE_FUNC
+    inline TransposeReturnType transpose()
+    {
+      EIGEN_STATIC_ASSERT_LVALUE(MatrixType)
+      typename MatrixType::TransposeReturnType tmp(m_matrix);
+      return TransposeReturnType(tmp);
+    }
+
+    typedef SelfAdjointView<const typename MatrixType::ConstTransposeReturnType,TransposeMode> ConstTransposeReturnType;
+    /** \sa MatrixBase::transpose() const */
+    EIGEN_DEVICE_FUNC
+    inline const ConstTransposeReturnType transpose() const
+    {
+      return ConstTransposeReturnType(m_matrix.transpose());
+    }
+
     /** \returns a const expression of the main diagonal of the matrix \c *this
       *
       * This method simply returns the diagonal of the nested expression, thus by-passing the SelfAdjointView decorator.
@@ -287,6 +319,7 @@ public:
 * Implementation of MatrixBase methods
 ***************************************************************************/
 
+/** This is the const version of MatrixBase::selfadjointView() */
 template<typename Derived>
 template<unsigned int UpLo>
 typename MatrixBase<Derived>::template ConstSelfAdjointViewReturnType<UpLo>::Type
@@ -295,6 +328,15 @@ MatrixBase<Derived>::selfadjointView() const
   return typename ConstSelfAdjointViewReturnType<UpLo>::Type(derived());
 }
 
+/** \returns an expression of a symmetric/self-adjoint view extracted from the upper or lower triangular part of the current matrix
+  *
+  * The parameter \a UpLo can be either \c #Upper or \c #Lower
+  *
+  * Example: \include MatrixBase_selfadjointView.cpp
+  * Output: \verbinclude MatrixBase_selfadjointView.out
+  *
+  * \sa class SelfAdjointView
+  */
 template<typename Derived>
 template<unsigned int UpLo>
 typename MatrixBase<Derived>::template SelfAdjointViewReturnType<UpLo>::Type

Eigen/src/Core/SelfCwiseBinaryOp.h

@@ -15,7 +15,7 @@ namespace Eigen {
 // TODO generalize the scalar type of 'other'
 
 template<typename Derived>
-EIGEN_STRONG_INLINE Derived& DenseBase<Derived>::operator*=(const Scalar& other)
+EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE Derived& DenseBase<Derived>::operator*=(const Scalar& other)
 {
   typedef typename Derived::PlainObject PlainObject;
   internal::call_assignment(this->derived(), PlainObject::Constant(rows(),cols(),other), internal::mul_assign_op<Scalar,Scalar>());
@@ -23,7 +23,7 @@ EIGEN_STRONG_INLINE Derived& DenseBase<Derived>::operator*=(const Scalar& other)
 }
 
 template<typename Derived>
-EIGEN_STRONG_INLINE Derived& ArrayBase<Derived>::operator+=(const Scalar& other)
+EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE Derived& ArrayBase<Derived>::operator+=(const Scalar& other)
 {
   typedef typename Derived::PlainObject PlainObject;
   internal::call_assignment(this->derived(), PlainObject::Constant(rows(),cols(),other), internal::add_assign_op<Scalar,Scalar>());
@@ -31,7 +31,7 @@ EIGEN_STRONG_INLINE Derived& ArrayBase<Derived>::operator+=(const Scalar& other)
 }
 
 template<typename Derived>
-EIGEN_STRONG_INLINE Derived& ArrayBase<Derived>::operator-=(const Scalar& other)
+EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE Derived& ArrayBase<Derived>::operator-=(const Scalar& other)
 {
   typedef typename Derived::PlainObject PlainObject;
   internal::call_assignment(this->derived(), PlainObject::Constant(rows(),cols(),other), internal::sub_assign_op<Scalar,Scalar>());
@@ -39,7 +39,7 @@ EIGEN_STRONG_INLINE Derived& ArrayBase<Derived>::operator-=(const Scalar& other)
 }
 
 template<typename Derived>
-EIGEN_STRONG_INLINE Derived& DenseBase<Derived>::operator/=(const Scalar& other)
+EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE Derived& DenseBase<Derived>::operator/=(const Scalar& other)
 {
   typedef typename Derived::PlainObject PlainObject;
   internal::call_assignment(this->derived(), PlainObject::Constant(rows(),cols(),other), internal::div_assign_op<Scalar,Scalar>());

Eigen/src/Core/Solve.h

@@ -34,12 +34,12 @@ template<typename Decomposition, typename RhsType,typename StorageKind> struct s
 template<typename Decomposition, typename RhsType>
 struct solve_traits<Decomposition,RhsType,Dense>
 {
-  typedef Matrix<typename RhsType::Scalar,
+  typedef typename make_proper_matrix_type<typename RhsType::Scalar,
                  Decomposition::ColsAtCompileTime,
                  RhsType::ColsAtCompileTime,
                  RhsType::PlainObject::Options,
                  Decomposition::MaxColsAtCompileTime,
-                 RhsType::MaxColsAtCompileTime> PlainObject;  
+                 RhsType::MaxColsAtCompileTime>::type PlainObject;
 };
 
 template<typename Decomposition, typename RhsType>

Eigen/src/Core/SolveTriangular.h

@@ -161,6 +161,7 @@ struct triangular_solver_selector<Lhs,Rhs,OnTheRight,Mode,CompleteUnrolling,1> {
 * TriangularView methods
 ***************************************************************************/
 
+#ifndef EIGEN_PARSED_BY_DOXYGEN
 template<typename MatrixType, unsigned int Mode>
 template<int Side, typename OtherDerived>
 void TriangularViewImpl<MatrixType,Mode,Dense>::solveInPlace(const MatrixBase<OtherDerived>& _other) const
@@ -188,6 +189,7 @@ TriangularViewImpl<Derived,Mode,Dense>::solve(const MatrixBase<Other>& other) co
 {
   return internal::triangular_solve_retval<Side,TriangularViewType,Other>(derived(), other.derived());
 }
+#endif
 
 namespace internal {
 

Eigen/src/Core/StableNorm.h

@@ -170,7 +170,8 @@ MatrixBase<Derived>::stableNorm() const
   enum {
     CanAlign = (   (int(DerivedCopyClean::Flags)&DirectAccessBit)
                 || (int(internal::evaluator<DerivedCopyClean>::Alignment)>0) // FIXME Alignment)>0 might not be enough
-               ) && (blockSize*sizeof(Scalar)*2<EIGEN_STACK_ALLOCATION_LIMIT) // ifwe cannot allocate on the stack, then let's not bother about this optimization
+               ) && (blockSize*sizeof(Scalar)*2<EIGEN_STACK_ALLOCATION_LIMIT)
+                 && (EIGEN_MAX_STATIC_ALIGN_BYTES>0) // if we cannot allocate on the stack, then let's not bother about this optimization
   };
   typedef typename internal::conditional<CanAlign, Ref<const Matrix<Scalar,Dynamic,1,0,blockSize,1>, internal::evaluator<DerivedCopyClean>::Alignment>,
                                                    typename DerivedCopyClean::ConstSegmentReturnType>::type SegmentWrapper;

Eigen/src/Core/TriangularMatrix.h

@@ -470,6 +470,8 @@ template<typename _MatrixType, unsigned int _Mode> class TriangularViewImpl<_Mat
       * \a Side==OnTheLeft (the default), or the right-inverse-multiply  \a other * inverse(\c *this) if
       * \a Side==OnTheRight.
       *
+      * Note that the template parameter \c Side can be ommitted, in which case \c Side==OnTheLeft
+      *
       * The matrix \c *this must be triangular and invertible (i.e., all the coefficients of the
       * diagonal must be non zero). It works as a forward (resp. backward) substitution if \c *this
       * is an upper (resp. lower) triangular matrix.
@@ -495,6 +497,8 @@ template<typename _MatrixType, unsigned int _Mode> class TriangularViewImpl<_Mat
       * \warning The parameter is only marked 'const' to make the C++ compiler accept a temporary expression here.
       * This function will const_cast it, so constness isn't honored here.
       *
+      * Note that the template parameter \c Side can be ommitted, in which case \c Side==OnTheLeft
+      *
       * See TriangularView:solve() for the details.
       */
     template<int Side, typename OtherDerived>
@@ -539,13 +543,14 @@ template<typename _MatrixType, unsigned int _Mode> class TriangularViewImpl<_Mat
 
     template<typename ProductType>
     EIGEN_DEVICE_FUNC
-    EIGEN_STRONG_INLINE TriangularViewType& _assignProduct(const ProductType& prod, const Scalar& alpha);
+    EIGEN_STRONG_INLINE TriangularViewType& _assignProduct(const ProductType& prod, const Scalar& alpha, bool beta);
 };
 
 /***************************************************************************
 * Implementation of triangular evaluation/assignment
 ***************************************************************************/
 
+#ifndef EIGEN_PARSED_BY_DOXYGEN
 // FIXME should we keep that possibility
 template<typename MatrixType, unsigned int Mode>
 template<typename OtherDerived>
@@ -583,6 +588,7 @@ void TriangularViewImpl<MatrixType, Mode, Dense>::lazyAssign(const TriangularBas
   eigen_assert(Mode == int(OtherDerived::Mode));
   internal::call_assignment_no_alias(derived(), other.derived());
 }
+#endif
 
 /***************************************************************************
 * Implementation of TriangularBase methods
@@ -944,8 +950,7 @@ struct Assignment<DstXprType, Product<Lhs,Rhs,DefaultProduct>, internal::assign_
     if((dst.rows()!=dstRows) || (dst.cols()!=dstCols))
       dst.resize(dstRows, dstCols);
 
-    dst.setZero();
-    dst._assignProduct(src, 1);
+    dst._assignProduct(src, 1, 0);
   }
 };
 
@@ -956,7 +961,7 @@ struct Assignment<DstXprType, Product<Lhs,Rhs,DefaultProduct>, internal::add_ass
   typedef Product<Lhs,Rhs,DefaultProduct> SrcXprType;
   static void run(DstXprType &dst, const SrcXprType &src, const internal::add_assign_op<Scalar,typename SrcXprType::Scalar> &)
   {
-    dst._assignProduct(src, 1);
+    dst._assignProduct(src, 1, 1);
   }
 };
 
@@ -967,7 +972,7 @@ struct Assignment<DstXprType, Product<Lhs,Rhs,DefaultProduct>, internal::sub_ass
   typedef Product<Lhs,Rhs,DefaultProduct> SrcXprType;
   static void run(DstXprType &dst, const SrcXprType &src, const internal::sub_assign_op<Scalar,typename SrcXprType::Scalar> &)
   {
-    dst._assignProduct(src, -1);
+    dst._assignProduct(src, -1, 1);
   }
 };
 

Eigen/src/Core/Visitor.h

@@ -194,7 +194,8 @@ struct functor_traits<max_coeff_visitor<Scalar> > {
 
 } // end namespace internal
 
-/** \returns the minimum of all coefficients of *this and puts in *row and *col its location.
+/** \fn DenseBase<Derived>::minCoeff(IndexType* rowId, IndexType* colId) const
+  * \returns the minimum of all coefficients of *this and puts in *row and *col its location.
   * \warning the result is undefined if \c *this contains NaN.
   *
   * \sa DenseBase::minCoeff(Index*), DenseBase::maxCoeff(Index*,Index*), DenseBase::visit(), DenseBase::minCoeff()
@@ -230,7 +231,8 @@ DenseBase<Derived>::minCoeff(IndexType* index) const
   return minVisitor.res;
 }
 
-/** \returns the maximum of all coefficients of *this and puts in *row and *col its location.
+/** \fn DenseBase<Derived>::maxCoeff(IndexType* rowId, IndexType* colId) const
+  * \returns the maximum of all coefficients of *this and puts in *row and *col its location.
   * \warning the result is undefined if \c *this contains NaN. 
   *
   * \sa DenseBase::minCoeff(IndexType*,IndexType*), DenseBase::visit(), DenseBase::maxCoeff()

Eigen/src/Core/arch/AVX/PacketMath.h

@@ -395,14 +395,11 @@ template<> EIGEN_STRONG_INLINE Packet4d preduxp<Packet4d>(const Packet4d* vecs)
 
 template<> EIGEN_STRONG_INLINE float predux<Packet8f>(const Packet8f& a)
 {
-  Packet8f tmp0 = _mm256_hadd_ps(a,_mm256_permute2f128_ps(a,a,1));
-  tmp0 = _mm256_hadd_ps(tmp0,tmp0);
-  return pfirst(_mm256_hadd_ps(tmp0, tmp0));
+  return predux(Packet4f(_mm_add_ps(_mm256_castps256_ps128(a),_mm256_extractf128_ps(a,1))));
 }
 template<> EIGEN_STRONG_INLINE double predux<Packet4d>(const Packet4d& a)
 {
-  Packet4d tmp0 = _mm256_hadd_pd(a,_mm256_permute2f128_pd(a,a,1));
-  return pfirst(_mm256_hadd_pd(tmp0,tmp0));
+  return predux(Packet2d(_mm_add_pd(_mm256_castpd256_pd128(a),_mm256_extractf128_pd(a,1))));
 }
 
 template<> EIGEN_STRONG_INLINE Packet4f predux_downto4<Packet8f>(const Packet8f& a)

Eigen/src/Core/arch/AltiVec/Complex.h

@@ -15,14 +15,14 @@ namespace Eigen {
 
 namespace internal {
 
-static Packet4ui  p4ui_CONJ_XOR = vec_mergeh((Packet4ui)p4i_ZERO, (Packet4ui)p4f_ZERO_);//{ 0x00000000, 0x80000000, 0x00000000, 0x80000000 };
+static Packet4ui  p4ui_CONJ_XOR = vec_mergeh((Packet4ui)p4i_ZERO, (Packet4ui)p4f_MZERO);//{ 0x00000000, 0x80000000, 0x00000000, 0x80000000 };
 #ifdef __VSX__
 #if defined(_BIG_ENDIAN)
-static Packet2ul  p2ul_CONJ_XOR1 = (Packet2ul) vec_sld((Packet4ui) p2d_ZERO_, (Packet4ui) p2l_ZERO, 8);//{ 0x8000000000000000, 0x0000000000000000 };
-static Packet2ul  p2ul_CONJ_XOR2 = (Packet2ul) vec_sld((Packet4ui) p2l_ZERO,  (Packet4ui) p2d_ZERO_, 8);//{ 0x8000000000000000, 0x0000000000000000 };
+static Packet2ul  p2ul_CONJ_XOR1 = (Packet2ul) vec_sld((Packet4ui) p2d_MZERO, (Packet4ui) p2l_ZERO, 8);//{ 0x8000000000000000, 0x0000000000000000 };
+static Packet2ul  p2ul_CONJ_XOR2 = (Packet2ul) vec_sld((Packet4ui) p2l_ZERO,  (Packet4ui) p2d_MZERO, 8);//{ 0x8000000000000000, 0x0000000000000000 };
 #else
-static Packet2ul  p2ul_CONJ_XOR1 = (Packet2ul) vec_sld((Packet4ui) p2l_ZERO,  (Packet4ui) p2d_ZERO_, 8);//{ 0x8000000000000000, 0x0000000000000000 };
-static Packet2ul  p2ul_CONJ_XOR2 = (Packet2ul) vec_sld((Packet4ui) p2d_ZERO_, (Packet4ui) p2l_ZERO, 8);//{ 0x8000000000000000, 0x0000000000000000 };
+static Packet2ul  p2ul_CONJ_XOR1 = (Packet2ul) vec_sld((Packet4ui) p2l_ZERO,  (Packet4ui) p2d_MZERO, 8);//{ 0x8000000000000000, 0x0000000000000000 };
+static Packet2ul  p2ul_CONJ_XOR2 = (Packet2ul) vec_sld((Packet4ui) p2d_MZERO, (Packet4ui) p2l_ZERO, 8);//{ 0x8000000000000000, 0x0000000000000000 };
 #endif
 #endif
 
@@ -65,7 +65,7 @@ template<> struct unpacket_traits<Packet2cf> { typedef std::complex<float> type;
 template<> EIGEN_STRONG_INLINE Packet2cf pset1<Packet2cf>(const std::complex<float>&  from)
 {
   Packet2cf res;
-  if((ptrdiff_t(&from) % 16) == 0)
+  if((std::ptrdiff_t(&from) % 16) == 0)
     res.v = pload<Packet4f>((const float *)&from);
   else
     res.v = ploadu<Packet4f>((const float *)&from);

Eigen/src/Core/arch/AltiVec/MathFunctions.h

@@ -84,8 +84,10 @@ static _EIGEN_DECLARE_CONST_Packet2d(cephes_exp_q3, 2.00000000000000000009e0);
 static _EIGEN_DECLARE_CONST_Packet2d(cephes_exp_C1, 0.693145751953125);
 static _EIGEN_DECLARE_CONST_Packet2d(cephes_exp_C2, 1.42860682030941723212e-6);
 
+#ifdef __POWER8_VECTOR__
 static Packet2l p2l_1023 = { 1023, 1023 };
 static Packet2ul p2ul_52 = { 52, 52 };
+#endif
 
 #endif
 

Eigen/src/Core/arch/AltiVec/PacketMath.h

@@ -72,7 +72,7 @@ static _EIGEN_DECLARE_CONST_FAST_Packet4i(ZERO, 0); //{ 0, 0, 0, 0,}
 static _EIGEN_DECLARE_CONST_FAST_Packet4i(ONE,1); //{ 1, 1, 1, 1}
 static _EIGEN_DECLARE_CONST_FAST_Packet4i(MINUS16,-16); //{ -16, -16, -16, -16}
 static _EIGEN_DECLARE_CONST_FAST_Packet4i(MINUS1,-1); //{ -1, -1, -1, -1}
-static Packet4f p4f_ZERO_ = (Packet4f) vec_sl((Packet4ui)p4i_MINUS1, (Packet4ui)p4i_MINUS1); //{ 0x80000000, 0x80000000, 0x80000000, 0x80000000}
+static Packet4f p4f_MZERO = (Packet4f) vec_sl((Packet4ui)p4i_MINUS1, (Packet4ui)p4i_MINUS1); //{ 0x80000000, 0x80000000, 0x80000000, 0x80000000}
 #ifndef __VSX__
 static Packet4f p4f_ONE = vec_ctf(p4i_ONE, 0); //{ 1.0, 1.0, 1.0, 1.0}
 #endif
@@ -90,7 +90,7 @@ static Packet16uc p16uc_DUPLICATE32_HI = { 0,1,2,3, 0,1,2,3, 4,5,6,7, 4,5,6,7 };
 #define _EIGEN_MASK_ALIGNMENT	0xfffffff0
 #endif
 
-#define _EIGEN_ALIGNED_PTR(x)	((ptrdiff_t)(x) & _EIGEN_MASK_ALIGNMENT)
+#define _EIGEN_ALIGNED_PTR(x)	((std::ptrdiff_t)(x) & _EIGEN_MASK_ALIGNMENT)
 
 // Handle endianness properly while loading constants
 // Define global static constants:
@@ -358,7 +358,7 @@ template<> EIGEN_STRONG_INLINE Packet4i pnegate(const Packet4i& a) { return p4i_
 template<> EIGEN_STRONG_INLINE Packet4f pconj(const Packet4f& a) { return a; }
 template<> EIGEN_STRONG_INLINE Packet4i pconj(const Packet4i& a) { return a; }
 
-template<> EIGEN_STRONG_INLINE Packet4f pmul<Packet4f>(const Packet4f& a, const Packet4f& b) { return vec_madd(a,b, p4f_ZERO); }
+template<> EIGEN_STRONG_INLINE Packet4f pmul<Packet4f>(const Packet4f& a, const Packet4f& b) { return vec_madd(a,b, p4f_MZERO); }
 template<> EIGEN_STRONG_INLINE Packet4i pmul<Packet4i>(const Packet4i& a, const Packet4i& b) { return a * b; }
 
 template<> EIGEN_STRONG_INLINE Packet4f pdiv<Packet4f>(const Packet4f& a, const Packet4f& b)
@@ -373,7 +373,7 @@ template<> EIGEN_STRONG_INLINE Packet4f pdiv<Packet4f>(const Packet4f& a, const
   t   = vec_nmsub(y_0, b, p4f_ONE);
   y_1 = vec_madd(y_0, t, y_0);
 
-  return vec_madd(a, y_1, p4f_ZERO);
+  return vec_madd(a, y_1, p4f_MZERO);
 #else
   return vec_div(a, b);
 #endif
@@ -450,15 +450,15 @@ template<> EIGEN_STRONG_INLINE Packet4f ploadu<Packet4f>(const float* from)
 template<> EIGEN_STRONG_INLINE Packet4f ploaddup<Packet4f>(const float*   from)
 {
   Packet4f p;
-  if((ptrdiff_t(from) % 16) == 0)  p = pload<Packet4f>(from);
-  else                             p = ploadu<Packet4f>(from);
+  if((std::ptrdiff_t(from) % 16) == 0)  p = pload<Packet4f>(from);
+  else                                  p = ploadu<Packet4f>(from);
   return vec_perm(p, p, p16uc_DUPLICATE32_HI);
 }
 template<> EIGEN_STRONG_INLINE Packet4i ploaddup<Packet4i>(const int*     from)
 {
   Packet4i p;
-  if((ptrdiff_t(from) % 16) == 0)  p = pload<Packet4i>(from);
-  else                             p = ploadu<Packet4i>(from);
+  if((std::ptrdiff_t(from) % 16) == 0)  p = pload<Packet4i>(from);
+  else                                  p = ploadu<Packet4i>(from);
   return vec_perm(p, p, p16uc_DUPLICATE32_HI);
 }
 
@@ -766,7 +766,7 @@ static Packet2l  p2l_ONE  = { 1, 1 };
 static Packet2l  p2l_ZERO = reinterpret_cast<Packet2l>(p4i_ZERO);
 static Packet2d  p2d_ONE  = { 1.0, 1.0 }; 
 static Packet2d  p2d_ZERO = reinterpret_cast<Packet2d>(p4f_ZERO);
-static Packet2d  p2d_ZERO_ = { -0.0, -0.0 };
+static Packet2d  p2d_MZERO = { -0.0, -0.0 };
 
 #ifdef _BIG_ENDIAN
 static Packet2d p2d_COUNTDOWN = reinterpret_cast<Packet2d>(vec_sld(reinterpret_cast<Packet4f>(p2d_ZERO), reinterpret_cast<Packet4f>(p2d_ONE), 8));
@@ -904,7 +904,7 @@ template<> EIGEN_STRONG_INLINE Packet2d pnegate(const Packet2d& a) { return p2d_
 
 template<> EIGEN_STRONG_INLINE Packet2d pconj(const Packet2d& a) { return a; }
 
-template<> EIGEN_STRONG_INLINE Packet2d pmul<Packet2d>(const Packet2d& a, const Packet2d& b) { return vec_madd(a,b,p2d_ZERO); }
+template<> EIGEN_STRONG_INLINE Packet2d pmul<Packet2d>(const Packet2d& a, const Packet2d& b) { return vec_madd(a,b,p2d_MZERO); }
 template<> EIGEN_STRONG_INLINE Packet2d pdiv<Packet2d>(const Packet2d& a, const Packet2d& b) { return vec_div(a,b); }
 
 // for some weird raisons, it has to be overloaded for packet of integers
@@ -935,8 +935,8 @@ template<> EIGEN_STRONG_INLINE Packet2d ploadu<Packet2d>(const double* from)
 template<> EIGEN_STRONG_INLINE Packet2d ploaddup<Packet2d>(const double*   from)
 {
   Packet2d p;
-  if((ptrdiff_t(from) % 16) == 0)  p = pload<Packet2d>(from);
-  else                             p = ploadu<Packet2d>(from);
+  if((std::ptrdiff_t(from) % 16) == 0)  p = pload<Packet2d>(from);
+  else                                  p = ploadu<Packet2d>(from);
   return vec_splat_dbl<0>(p);
 }
 

Eigen/src/Core/arch/CUDA/Half.h

@@ -13,7 +13,7 @@
 // Redistribution and use in source and binary forms, with or without
 // modification, are permitted.
 // THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
-// “AS IS” AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+// "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
 // LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
 // A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
 // HOLDER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
@@ -474,9 +474,59 @@ template<> struct is_arithmetic<half> { enum { value = true }; };
 
 } // end namespace internal
 
+}  // end namespace Eigen
+
+namespace std {
+template<>
+struct numeric_limits<Eigen::half> {
+  static const bool is_specialized = true;
+  static const bool is_signed = true;
+  static const bool is_integer = false;
+  static const bool is_exact = false;
+  static const bool has_infinity = true;
+  static const bool has_quiet_NaN = true;
+  static const bool has_signaling_NaN = true;
+  static const float_denorm_style has_denorm = denorm_present;
+  static const bool has_denorm_loss = false;
+  static const std::float_round_style round_style = std::round_to_nearest;
+  static const bool is_iec559 = false;
+  static const bool is_bounded = false;
+  static const bool is_modulo = false;
+  static const int digits = 11;
+  static const int digits10 = 2;
+  //static const int max_digits10 = ;
+  static const int radix = 2;
+  static const int min_exponent = -13;
+  static const int min_exponent10 = -4;
+  static const int max_exponent = 16;
+  static const int max_exponent10 = 4;
+  static const bool traps = true;
+  static const bool tinyness_before = false;
+
+  static Eigen::half (min)() { return Eigen::half_impl::raw_uint16_to_half(0x400); }
+  static Eigen::half lowest() { return Eigen::half_impl::raw_uint16_to_half(0xfbff); }
+  static Eigen::half (max)() { return Eigen::half_impl::raw_uint16_to_half(0x7bff); }
+  static Eigen::half epsilon() { return Eigen::half_impl::raw_uint16_to_half(0x0800); }
+  static Eigen::half round_error() { return Eigen::half(0.5); }
+  static Eigen::half infinity() { return Eigen::half_impl::raw_uint16_to_half(0x7c00); }
+  static Eigen::half quiet_NaN() { return Eigen::half_impl::raw_uint16_to_half(0x7e00); }
+  static Eigen::half signaling_NaN() { return Eigen::half_impl::raw_uint16_to_half(0x7e00); }
+  static Eigen::half denorm_min() { return Eigen::half_impl::raw_uint16_to_half(0x1); }
+};
+}
+
+namespace Eigen {
+
 template<> struct NumTraits<Eigen::half>
     : GenericNumTraits<Eigen::half>
 {
+  enum {
+    IsSigned = true,
+    IsInteger = false,
+    IsComplex = false,
+    RequireInitialization = false
+  };
+
   EIGEN_DEVICE_FUNC static EIGEN_STRONG_INLINE Eigen::half epsilon() {
     return half_impl::raw_uint16_to_half(0x0800);
   }

Eigen/src/Core/arch/CUDA/PacketMath.h

@@ -291,7 +291,7 @@ template<> EIGEN_DEVICE_FUNC inline double2 pabs<double2>(const double2& a) {
 
 EIGEN_DEVICE_FUNC inline void
 ptranspose(PacketBlock<float4,4>& kernel) {
-  double tmp = kernel.packet[0].y;
+  float tmp = kernel.packet[0].y;
   kernel.packet[0].y = kernel.packet[1].x;
   kernel.packet[1].x = tmp;
 

Eigen/src/Core/arch/NEON/PacketMath.h

@@ -28,11 +28,13 @@ namespace internal {
 #define EIGEN_HAS_SINGLE_INSTRUCTION_CJMADD
 #endif
 
-// FIXME NEON has 16 quad registers, but since the current register allocator
-// is so bad, it is much better to reduce it to 8
 #ifndef EIGEN_ARCH_DEFAULT_NUMBER_OF_REGISTERS
+#if EIGEN_ARCH_ARM64
+#define EIGEN_ARCH_DEFAULT_NUMBER_OF_REGISTERS 32
+#else
 #define EIGEN_ARCH_DEFAULT_NUMBER_OF_REGISTERS 16 
 #endif
+#endif
 
 typedef float32x2_t Packet2f;
 typedef float32x4_t Packet4f;
@@ -44,19 +46,22 @@ typedef uint32x4_t  Packet4ui;
   const Packet4f p4f_##NAME = pset1<Packet4f>(X)
 
 #define _EIGEN_DECLARE_CONST_Packet4f_FROM_INT(NAME,X) \
-  const Packet4f p4f_##NAME = vreinterpretq_f32_u32(pset1<int>(X))
+  const Packet4f p4f_##NAME = vreinterpretq_f32_u32(pset1<int32_t>(X))
 
 #define _EIGEN_DECLARE_CONST_Packet4i(NAME,X) \
   const Packet4i p4i_##NAME = pset1<Packet4i>(X)
 
-// arm64 does have the pld instruction. If available, let's trust the __builtin_prefetch built-in function
-// which available on LLVM and GCC (at least)
-#if EIGEN_HAS_BUILTIN(__builtin_prefetch) || EIGEN_COMP_GNUC
+#if EIGEN_ARCH_ARM64
+  // __builtin_prefetch tends to do nothing on ARM64 compilers because the
+  // prefetch instructions there are too detailed for __builtin_prefetch to map
+  // meaningfully to them.
+  #define EIGEN_ARM_PREFETCH(ADDR)  __asm__ __volatile__("prfm pldl1keep, [%[addr]]\n" ::[addr] "r"(ADDR) : );
+#elif EIGEN_HAS_BUILTIN(__builtin_prefetch) || EIGEN_COMP_GNUC
   #define EIGEN_ARM_PREFETCH(ADDR) __builtin_prefetch(ADDR);
 #elif defined __pld
   #define EIGEN_ARM_PREFETCH(ADDR) __pld(ADDR)
-#elif !EIGEN_ARCH_ARM64
-  #define EIGEN_ARM_PREFETCH(ADDR) __asm__ __volatile__ ( "   pld [%[addr]]\n" :: [addr] "r" (ADDR) : "cc" );
+#elif EIGEN_ARCH_ARM32
+  #define EIGEN_ARM_PREFETCH(ADDR) __asm__ __volatile__ ("pld [%[addr]]\n" :: [addr] "r" (ADDR) : );
 #else
   // by default no explicit prefetching
   #define EIGEN_ARM_PREFETCH(ADDR)
@@ -81,7 +86,7 @@ template<> struct packet_traits<float>  : default_packet_traits
     HasSqrt = 0
   };
 };
-template<> struct packet_traits<int>    : default_packet_traits
+template<> struct packet_traits<int32_t>    : default_packet_traits
 {
   typedef Packet4i type;
   typedef Packet4i half; // Packet2i intrinsics not implemented yet
@@ -103,19 +108,19 @@ EIGEN_STRONG_INLINE void        vst1q_f32(float* to, float32x4_t from) { ::vst1q
 EIGEN_STRONG_INLINE void        vst1_f32 (float* to, float32x2_t from) { ::vst1_f32 ((float32_t*)to,from); }
 #endif
 
-template<> struct unpacket_traits<Packet4f> { typedef float  type; enum {size=4, alignment=Aligned16}; typedef Packet4f half; };
-template<> struct unpacket_traits<Packet4i> { typedef int    type; enum {size=4, alignment=Aligned16}; typedef Packet4i half; };
+template<> struct unpacket_traits<Packet4f> { typedef float   type; enum {size=4, alignment=Aligned16}; typedef Packet4f half; };
+template<> struct unpacket_traits<Packet4i> { typedef int32_t type; enum {size=4, alignment=Aligned16}; typedef Packet4i half; };
 
 template<> EIGEN_STRONG_INLINE Packet4f pset1<Packet4f>(const float&  from) { return vdupq_n_f32(from); }
-template<> EIGEN_STRONG_INLINE Packet4i pset1<Packet4i>(const int&    from)   { return vdupq_n_s32(from); }
+template<> EIGEN_STRONG_INLINE Packet4i pset1<Packet4i>(const int32_t&    from)   { return vdupq_n_s32(from); }
 
 template<> EIGEN_STRONG_INLINE Packet4f plset<Packet4f>(const float& a)
 {
-  const float32_t f[] = {0, 1, 2, 3};
+  const float f[] = {0, 1, 2, 3};
   Packet4f countdown = vld1q_f32(f);
   return vaddq_f32(pset1<Packet4f>(a), countdown);
 }
-template<> EIGEN_STRONG_INLINE Packet4i plset<Packet4i>(const int& a)
+template<> EIGEN_STRONG_INLINE Packet4i plset<Packet4i>(const int32_t& a)
 {
   const int32_t i[] = {0, 1, 2, 3};
   Packet4i countdown = vld1q_s32(i);
@@ -238,20 +243,20 @@ template<> EIGEN_STRONG_INLINE Packet4f pandnot<Packet4f>(const Packet4f& a, con
 }
 template<> EIGEN_STRONG_INLINE Packet4i pandnot<Packet4i>(const Packet4i& a, const Packet4i& b) { return vbicq_s32(a,b); }
 
-template<> EIGEN_STRONG_INLINE Packet4f pload<Packet4f>(const float* from) { EIGEN_DEBUG_ALIGNED_LOAD return vld1q_f32(from); }
-template<> EIGEN_STRONG_INLINE Packet4i pload<Packet4i>(const int*   from) { EIGEN_DEBUG_ALIGNED_LOAD return vld1q_s32(from); }
+template<> EIGEN_STRONG_INLINE Packet4f pload<Packet4f>(const float*    from) { EIGEN_DEBUG_ALIGNED_LOAD return vld1q_f32(from); }
+template<> EIGEN_STRONG_INLINE Packet4i pload<Packet4i>(const int32_t*  from) { EIGEN_DEBUG_ALIGNED_LOAD return vld1q_s32(from); }
 
-template<> EIGEN_STRONG_INLINE Packet4f ploadu<Packet4f>(const float* from) { EIGEN_DEBUG_UNALIGNED_LOAD return vld1q_f32(from); }
-template<> EIGEN_STRONG_INLINE Packet4i ploadu<Packet4i>(const int* from)   { EIGEN_DEBUG_UNALIGNED_LOAD return vld1q_s32(from); }
+template<> EIGEN_STRONG_INLINE Packet4f ploadu<Packet4f>(const float*   from) { EIGEN_DEBUG_UNALIGNED_LOAD return vld1q_f32(from); }
+template<> EIGEN_STRONG_INLINE Packet4i ploadu<Packet4i>(const int32_t* from) { EIGEN_DEBUG_UNALIGNED_LOAD return vld1q_s32(from); }
 
-template<> EIGEN_STRONG_INLINE Packet4f ploaddup<Packet4f>(const float*   from)
+template<> EIGEN_STRONG_INLINE Packet4f ploaddup<Packet4f>(const float* from)
 {
   float32x2_t lo, hi;
   lo = vld1_dup_f32(from);
   hi = vld1_dup_f32(from+1);
   return vcombine_f32(lo, hi);
 }
-template<> EIGEN_STRONG_INLINE Packet4i ploaddup<Packet4i>(const int*     from)
+template<> EIGEN_STRONG_INLINE Packet4i ploaddup<Packet4i>(const int32_t* from)
 {
   int32x2_t lo, hi;
   lo = vld1_dup_s32(from);
@@ -259,11 +264,11 @@ template<> EIGEN_STRONG_INLINE Packet4i ploaddup<Packet4i>(const int*     from)
   return vcombine_s32(lo, hi);
 }
 
-template<> EIGEN_STRONG_INLINE void pstore<float>(float*   to, const Packet4f& from) { EIGEN_DEBUG_ALIGNED_STORE vst1q_f32(to, from); }
-template<> EIGEN_STRONG_INLINE void pstore<int>(int*       to, const Packet4i& from) { EIGEN_DEBUG_ALIGNED_STORE vst1q_s32(to, from); }
+template<> EIGEN_STRONG_INLINE void pstore<float>  (float*    to, const Packet4f& from) { EIGEN_DEBUG_ALIGNED_STORE vst1q_f32(to, from); }
+template<> EIGEN_STRONG_INLINE void pstore<int32_t>(int32_t*  to, const Packet4i& from) { EIGEN_DEBUG_ALIGNED_STORE vst1q_s32(to, from); }
 
-template<> EIGEN_STRONG_INLINE void pstoreu<float>(float*  to, const Packet4f& from) { EIGEN_DEBUG_UNALIGNED_STORE vst1q_f32(to, from); }
-template<> EIGEN_STRONG_INLINE void pstoreu<int>(int*      to, const Packet4i& from) { EIGEN_DEBUG_UNALIGNED_STORE vst1q_s32(to, from); }
+template<> EIGEN_STRONG_INLINE void pstoreu<float>  (float*   to, const Packet4f& from) { EIGEN_DEBUG_UNALIGNED_STORE vst1q_f32(to, from); }
+template<> EIGEN_STRONG_INLINE void pstoreu<int32_t>(int32_t* to, const Packet4i& from) { EIGEN_DEBUG_UNALIGNED_STORE vst1q_s32(to, from); }
 
 template<> EIGEN_DEVICE_FUNC inline Packet4f pgather<float, Packet4f>(const float* from, Index stride)
 {
@@ -274,7 +279,7 @@ template<> EIGEN_DEVICE_FUNC inline Packet4f pgather<float, Packet4f>(const floa
   res = vsetq_lane_f32(from[3*stride], res, 3);
   return res;
 }
-template<> EIGEN_DEVICE_FUNC inline Packet4i pgather<int, Packet4i>(const int* from, Index stride)
+template<> EIGEN_DEVICE_FUNC inline Packet4i pgather<int32_t, Packet4i>(const int32_t* from, Index stride)
 {
   Packet4i res = pset1<Packet4i>(0);
   res = vsetq_lane_s32(from[0*stride], res, 0);
@@ -291,7 +296,7 @@ template<> EIGEN_DEVICE_FUNC inline void pscatter<float, Packet4f>(float* to, co
   to[stride*2] = vgetq_lane_f32(from, 2);
   to[stride*3] = vgetq_lane_f32(from, 3);
 }
-template<> EIGEN_DEVICE_FUNC inline void pscatter<int, Packet4i>(int* to, const Packet4i& from, Index stride)
+template<> EIGEN_DEVICE_FUNC inline void pscatter<int32_t, Packet4i>(int32_t* to, const Packet4i& from, Index stride)
 {
   to[stride*0] = vgetq_lane_s32(from, 0);
   to[stride*1] = vgetq_lane_s32(from, 1);
@@ -299,12 +304,12 @@ template<> EIGEN_DEVICE_FUNC inline void pscatter<int, Packet4i>(int* to, const
   to[stride*3] = vgetq_lane_s32(from, 3);
 }
 
-template<> EIGEN_STRONG_INLINE void prefetch<float>(const float* addr) { EIGEN_ARM_PREFETCH(addr); }
-template<> EIGEN_STRONG_INLINE void prefetch<int>(const int*     addr) { EIGEN_ARM_PREFETCH(addr); }
+template<> EIGEN_STRONG_INLINE void prefetch<float>  (const float*    addr) { EIGEN_ARM_PREFETCH(addr); }
+template<> EIGEN_STRONG_INLINE void prefetch<int32_t>(const int32_t*  addr) { EIGEN_ARM_PREFETCH(addr); }
 
 // FIXME only store the 2 first elements ?
-template<> EIGEN_STRONG_INLINE float  pfirst<Packet4f>(const Packet4f& a) { float EIGEN_ALIGN16 x[4]; vst1q_f32(x, a); return x[0]; }
-template<> EIGEN_STRONG_INLINE int    pfirst<Packet4i>(const Packet4i& a) { int   EIGEN_ALIGN16 x[4]; vst1q_s32(x, a); return x[0]; }
+template<> EIGEN_STRONG_INLINE float   pfirst<Packet4f>(const Packet4f& a) { float   EIGEN_ALIGN16 x[4]; vst1q_f32(x, a); return x[0]; }
+template<> EIGEN_STRONG_INLINE int32_t pfirst<Packet4i>(const Packet4i& a) { int32_t EIGEN_ALIGN16 x[4]; vst1q_s32(x, a); return x[0]; }
 
 template<> EIGEN_STRONG_INLINE Packet4f preverse(const Packet4f& a) {
   float32x2_t a_lo, a_hi;
@@ -359,7 +364,7 @@ template<> EIGEN_STRONG_INLINE Packet4f preduxp<Packet4f>(const Packet4f* vecs)
   return sum;
 }
 
-template<> EIGEN_STRONG_INLINE int predux<Packet4i>(const Packet4i& a)
+template<> EIGEN_STRONG_INLINE int32_t predux<Packet4i>(const Packet4i& a)
 {
   int32x2_t a_lo, a_hi, sum;
 
@@ -406,7 +411,7 @@ template<> EIGEN_STRONG_INLINE float predux_mul<Packet4f>(const Packet4f& a)
 
   return vget_lane_f32(prod, 0);
 }
-template<> EIGEN_STRONG_INLINE int predux_mul<Packet4i>(const Packet4i& a)
+template<> EIGEN_STRONG_INLINE int32_t predux_mul<Packet4i>(const Packet4i& a)
 {
   int32x2_t a_lo, a_hi, prod;
 
@@ -434,7 +439,7 @@ template<> EIGEN_STRONG_INLINE float predux_min<Packet4f>(const Packet4f& a)
   return vget_lane_f32(min, 0);
 }
 
-template<> EIGEN_STRONG_INLINE int predux_min<Packet4i>(const Packet4i& a)
+template<> EIGEN_STRONG_INLINE int32_t predux_min<Packet4i>(const Packet4i& a)
 {
   int32x2_t a_lo, a_hi, min;
 
@@ -459,7 +464,7 @@ template<> EIGEN_STRONG_INLINE float predux_max<Packet4f>(const Packet4f& a)
   return vget_lane_f32(max, 0);
 }
 
-template<> EIGEN_STRONG_INLINE int predux_max<Packet4i>(const Packet4i& a)
+template<> EIGEN_STRONG_INLINE int32_t predux_max<Packet4i>(const Packet4i& a)
 {
   int32x2_t a_lo, a_hi, max;
 

Eigen/src/Core/arch/SSE/PacketMath.h

@@ -28,7 +28,7 @@ namespace internal {
 #endif
 #endif
 
-#if (defined EIGEN_VECTORIZE_AVX) && EIGEN_COMP_GNUC_STRICT && (__GXX_ABI_VERSION < 1004)
+#if (defined EIGEN_VECTORIZE_AVX) && (EIGEN_COMP_GNUC_STRICT || EIGEN_COMP_MINGW) && (__GXX_ABI_VERSION < 1004)
 // With GCC's default ABI version, a __m128 or __m256 are the same types and therefore we cannot
 // have overloads for both types without linking error.
 // One solution is to increase ABI version using -fabi-version=4 (or greater).
@@ -504,30 +504,13 @@ template<> EIGEN_STRONG_INLINE Packet4f preduxp<Packet4f>(const Packet4f* vecs)
 {
   return _mm_hadd_ps(_mm_hadd_ps(vecs[0], vecs[1]),_mm_hadd_ps(vecs[2], vecs[3]));
 }
+
 template<> EIGEN_STRONG_INLINE Packet2d preduxp<Packet2d>(const Packet2d* vecs)
 {
   return _mm_hadd_pd(vecs[0], vecs[1]);
 }
 
-template<> EIGEN_STRONG_INLINE float predux<Packet4f>(const Packet4f& a)
-{
-  Packet4f tmp0 = _mm_hadd_ps(a,a);
-  return pfirst<Packet4f>(_mm_hadd_ps(tmp0, tmp0));
-}
-
-template<> EIGEN_STRONG_INLINE double predux<Packet2d>(const Packet2d& a) { return pfirst<Packet2d>(_mm_hadd_pd(a, a)); }
 #else
-// SSE2 versions
-template<> EIGEN_STRONG_INLINE float predux<Packet4f>(const Packet4f& a)
-{
-  Packet4f tmp = _mm_add_ps(a, _mm_movehl_ps(a,a));
-  return pfirst<Packet4f>(_mm_add_ss(tmp, _mm_shuffle_ps(tmp,tmp, 1)));
-}
-template<> EIGEN_STRONG_INLINE double predux<Packet2d>(const Packet2d& a)
-{
-  return pfirst<Packet2d>(_mm_add_sd(a, _mm_unpackhi_pd(a,a)));
-}
-
 template<> EIGEN_STRONG_INLINE Packet4f preduxp<Packet4f>(const Packet4f* vecs)
 {
   Packet4f tmp0, tmp1, tmp2;
@@ -548,6 +531,29 @@ template<> EIGEN_STRONG_INLINE Packet2d preduxp<Packet2d>(const Packet2d* vecs)
 }
 #endif  // SSE3
 
+template<> EIGEN_STRONG_INLINE float predux<Packet4f>(const Packet4f& a)
+{
+  // Disable SSE3 _mm_hadd_pd that is extremely slow on all existing Intel's architectures
+  // (from Nehalem to Haswell)
+// #ifdef EIGEN_VECTORIZE_SSE3
+//   Packet4f tmp = _mm_add_ps(a, vec4f_swizzle1(a,2,3,2,3));
+//   return pfirst<Packet4f>(_mm_hadd_ps(tmp, tmp));
+// #else
+  Packet4f tmp = _mm_add_ps(a, _mm_movehl_ps(a,a));
+  return pfirst<Packet4f>(_mm_add_ss(tmp, _mm_shuffle_ps(tmp,tmp, 1)));
+// #endif
+}
+
+template<> EIGEN_STRONG_INLINE double predux<Packet2d>(const Packet2d& a)
+{
+  // Disable SSE3 _mm_hadd_pd that is extremely slow on all existing Intel's architectures
+  // (from Nehalem to Haswell)
+// #ifdef EIGEN_VECTORIZE_SSE3
+//   return pfirst<Packet2d>(_mm_hadd_pd(a, a));
+// #else
+  return pfirst<Packet2d>(_mm_add_sd(a, _mm_unpackhi_pd(a,a)));
+// #endif
+}
 
 #ifdef EIGEN_VECTORIZE_SSSE3
 template<> EIGEN_STRONG_INLINE Packet4i preduxp<Packet4i>(const Packet4i* vecs)

Eigen/src/Core/arch/ZVector/Complex.h

@@ -2,6 +2,7 @@
 // for linear algebra.
 //
 // Copyright (C) 2010 Gael Guennebaud <gael.guennebaud@inria.fr>
+// Copyright (C) 2016 Konstantinos Margaritis <markos@freevec.org>
 //
 // 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
@@ -24,13 +25,48 @@ struct Packet1cd
   Packet2d v;
 };
 
+struct Packet2cf
+{
+  EIGEN_STRONG_INLINE Packet2cf() {}
+  EIGEN_STRONG_INLINE explicit Packet2cf(const Packet4f& a) : v(a) {}
+  union {
+    Packet4f v;
+    Packet1cd cd[2];
+  };
+};
+
+template<> struct packet_traits<std::complex<float> >  : default_packet_traits
+{
+  typedef Packet2cf type;
+  typedef Packet2cf half;
+  enum {
+    Vectorizable = 1,
+    AlignedOnScalar = 1,
+    size = 2,
+    HasHalfPacket = 0,
+
+    HasAdd    = 1,
+    HasSub    = 1,
+    HasMul    = 1,
+    HasDiv    = 1,
+    HasNegate = 1,
+    HasAbs    = 0,
+    HasAbs2   = 0,
+    HasMin    = 0,
+    HasMax    = 0,
+    HasBlend  = 1,
+    HasSetLinear = 0
+  };
+};
+
+
 template<> struct packet_traits<std::complex<double> >  : default_packet_traits
 {
   typedef Packet1cd type;
   typedef Packet1cd half;
   enum {
     Vectorizable = 1,
-    AlignedOnScalar = 0,
+    AlignedOnScalar = 1,
     size = 1,
     HasHalfPacket = 0,
 
@@ -47,20 +83,68 @@ template<> struct packet_traits<std::complex<double> >  : default_packet_traits
   };
 };
 
+template<> struct unpacket_traits<Packet2cf> { typedef std::complex<float>  type; enum {size=2, alignment=Aligned16}; typedef Packet2cf half; };
 template<> struct unpacket_traits<Packet1cd> { typedef std::complex<double> type; enum {size=1, alignment=Aligned16}; typedef Packet1cd half; };
 
+/* Forward declaration */
+EIGEN_STRONG_INLINE void ptranspose(PacketBlock<Packet2cf,2>& kernel);
+
+template<> EIGEN_STRONG_INLINE Packet2cf pload <Packet2cf>(const std::complex<float>* from)  { EIGEN_DEBUG_ALIGNED_LOAD return Packet2cf(pload<Packet4f>((const float*)from)); }
 template<> EIGEN_STRONG_INLINE Packet1cd pload <Packet1cd>(const std::complex<double>* from) { EIGEN_DEBUG_ALIGNED_LOAD return Packet1cd(pload<Packet2d>((const double*)from)); }
+template<> EIGEN_STRONG_INLINE Packet2cf ploadu<Packet2cf>(const std::complex<float>* from)  { EIGEN_DEBUG_UNALIGNED_LOAD return Packet2cf(ploadu<Packet4f>((const float*)from)); }
 template<> EIGEN_STRONG_INLINE Packet1cd ploadu<Packet1cd>(const std::complex<double>* from) { EIGEN_DEBUG_UNALIGNED_LOAD return Packet1cd(ploadu<Packet2d>((const double*)from)); }
+template<> EIGEN_STRONG_INLINE void pstore <std::complex<float> >(std::complex<float> *     to, const Packet2cf& from) { EIGEN_DEBUG_ALIGNED_STORE pstore((float*)to, from.v); }
 template<> EIGEN_STRONG_INLINE void pstore <std::complex<double> >(std::complex<double> *   to, const Packet1cd& from) { EIGEN_DEBUG_ALIGNED_STORE pstore((double*)to, from.v); }
+template<> EIGEN_STRONG_INLINE void pstoreu<std::complex<float> >(std::complex<float> *     to, const Packet2cf& from) { EIGEN_DEBUG_UNALIGNED_STORE pstoreu((float*)to, from.v); }
 template<> EIGEN_STRONG_INLINE void pstoreu<std::complex<double> >(std::complex<double> *   to, const Packet1cd& from) { EIGEN_DEBUG_UNALIGNED_STORE pstoreu((double*)to, from.v); }
 
 template<> EIGEN_STRONG_INLINE Packet1cd pset1<Packet1cd>(const std::complex<double>&  from)
 { /* here we really have to use unaligned loads :( */ return ploadu<Packet1cd>(&from); }
 
+template<> EIGEN_STRONG_INLINE Packet2cf pset1<Packet2cf>(const std::complex<float>&  from)
+{
+  Packet2cf res;
+  res.cd[0] = Packet1cd(vec_ld2f((const float *)&from));
+  res.cd[1] = res.cd[0];
+  return res;
+}
+template<> EIGEN_DEVICE_FUNC inline Packet2cf pgather<std::complex<float>, Packet2cf>(const std::complex<float>* from, Index stride)
+{
+  std::complex<float> EIGEN_ALIGN16 af[2];
+  af[0] = from[0*stride];
+  af[1] = from[1*stride];
+  return pload<Packet2cf>(af);
+}
+template<> EIGEN_DEVICE_FUNC inline Packet1cd pgather<std::complex<double>, Packet1cd>(const std::complex<double>* from, Index stride EIGEN_UNUSED)
+{
+  return pload<Packet1cd>(from);
+}
+template<> EIGEN_DEVICE_FUNC inline void pscatter<std::complex<float>, Packet2cf>(std::complex<float>* to, const Packet2cf& from, Index stride)
+{
+  std::complex<float> EIGEN_ALIGN16 af[2];
+  pstore<std::complex<float> >((std::complex<float> *) af, from);
+  to[0*stride] = af[0];
+  to[1*stride] = af[1];
+}
+template<> EIGEN_DEVICE_FUNC inline void pscatter<std::complex<double>, Packet1cd>(std::complex<double>* to, const Packet1cd& from, Index stride EIGEN_UNUSED)
+{
+  pstore<std::complex<double> >(to, from);
+}
+
+template<> EIGEN_STRONG_INLINE Packet2cf padd<Packet2cf>(const Packet2cf& a, const Packet2cf& b) { return Packet2cf(padd<Packet4f>(a.v, b.v)); }
 template<> EIGEN_STRONG_INLINE Packet1cd padd<Packet1cd>(const Packet1cd& a, const Packet1cd& b) { return Packet1cd(a.v + b.v); }
+template<> EIGEN_STRONG_INLINE Packet2cf psub<Packet2cf>(const Packet2cf& a, const Packet2cf& b) { return Packet2cf(psub<Packet4f>(a.v, b.v)); }
 template<> EIGEN_STRONG_INLINE Packet1cd psub<Packet1cd>(const Packet1cd& a, const Packet1cd& b) { return Packet1cd(a.v - b.v); }
 template<> EIGEN_STRONG_INLINE Packet1cd pnegate(const Packet1cd& a) { return Packet1cd(pnegate(Packet2d(a.v))); }
+template<> EIGEN_STRONG_INLINE Packet2cf pnegate(const Packet2cf& a) { return Packet2cf(pnegate(Packet4f(a.v))); }
 template<> EIGEN_STRONG_INLINE Packet1cd pconj(const Packet1cd& a) { return Packet1cd((Packet2d)vec_xor((Packet2d)a.v, (Packet2d)p2ul_CONJ_XOR2)); }
+template<> EIGEN_STRONG_INLINE Packet2cf pconj(const Packet2cf& a)
+{
+  Packet2cf res;
+  res.v.v4f[0] = pconj(Packet1cd(reinterpret_cast<Packet2d>(a.v.v4f[0]))).v;
+  res.v.v4f[1] = pconj(Packet1cd(reinterpret_cast<Packet2d>(a.v.v4f[1]))).v;
+  return res;
+}
 
 template<> EIGEN_STRONG_INLINE Packet1cd pmul<Packet1cd>(const Packet1cd& a, const Packet1cd& b)
 {
@@ -79,43 +163,90 @@ template<> EIGEN_STRONG_INLINE Packet1cd pmul<Packet1cd>(const Packet1cd& a, con
 
   return Packet1cd(v1 + v2);
 }
-
-template<> EIGEN_STRONG_INLINE Packet1cd pand   <Packet1cd>(const Packet1cd& a, const Packet1cd& b) { return Packet1cd(vec_and(a.v,b.v)); }
-template<> EIGEN_STRONG_INLINE Packet1cd por    <Packet1cd>(const Packet1cd& a, const Packet1cd& b) { return Packet1cd(vec_or(a.v,b.v)); }
-template<> EIGEN_STRONG_INLINE Packet1cd pxor   <Packet1cd>(const Packet1cd& a, const Packet1cd& b) { return Packet1cd(vec_xor(a.v,b.v)); }
-template<> EIGEN_STRONG_INLINE Packet1cd pandnot<Packet1cd>(const Packet1cd& a, const Packet1cd& b) { return Packet1cd(vec_and(a.v, vec_nor(b.v,b.v))); }
-
-template<> EIGEN_STRONG_INLINE Packet1cd ploaddup<Packet1cd>(const std::complex<double>*     from)
+template<> EIGEN_STRONG_INLINE Packet2cf pmul<Packet2cf>(const Packet2cf& a, const Packet2cf& b)
 {
-  return pset1<Packet1cd>(*from);
+  Packet2cf res;
+  res.v.v4f[0] = pmul(Packet1cd(reinterpret_cast<Packet2d>(a.v.v4f[0])), Packet1cd(reinterpret_cast<Packet2d>(b.v.v4f[0]))).v;
+  res.v.v4f[1] = pmul(Packet1cd(reinterpret_cast<Packet2d>(a.v.v4f[1])), Packet1cd(reinterpret_cast<Packet2d>(b.v.v4f[1]))).v;
+  return res;
 }
 
+template<> EIGEN_STRONG_INLINE Packet1cd pand   <Packet1cd>(const Packet1cd& a, const Packet1cd& b) { return Packet1cd(vec_and(a.v,b.v)); }
+template<> EIGEN_STRONG_INLINE Packet2cf pand   <Packet2cf>(const Packet2cf& a, const Packet2cf& b) { return Packet2cf(pand<Packet4f>(a.v,b.v)); }
+template<> EIGEN_STRONG_INLINE Packet1cd por    <Packet1cd>(const Packet1cd& a, const Packet1cd& b) { return Packet1cd(vec_or(a.v,b.v)); }
+template<> EIGEN_STRONG_INLINE Packet2cf por    <Packet2cf>(const Packet2cf& a, const Packet2cf& b) { return Packet2cf(por<Packet4f>(a.v,b.v)); }
+template<> EIGEN_STRONG_INLINE Packet1cd pxor   <Packet1cd>(const Packet1cd& a, const Packet1cd& b) { return Packet1cd(vec_xor(a.v,b.v)); }
+template<> EIGEN_STRONG_INLINE Packet2cf pxor   <Packet2cf>(const Packet2cf& a, const Packet2cf& b) { return Packet2cf(pxor<Packet4f>(a.v,b.v)); }
+template<> EIGEN_STRONG_INLINE Packet1cd pandnot<Packet1cd>(const Packet1cd& a, const Packet1cd& b) { return Packet1cd(vec_and(a.v, vec_nor(b.v,b.v))); }
+template<> EIGEN_STRONG_INLINE Packet2cf pandnot<Packet2cf>(const Packet2cf& a, const Packet2cf& b) { return Packet2cf(pandnot<Packet4f>(a.v,b.v)); }
+
+template<> EIGEN_STRONG_INLINE Packet1cd ploaddup<Packet1cd>(const std::complex<double>*     from) {  return pset1<Packet1cd>(*from); }
+template<> EIGEN_STRONG_INLINE Packet2cf ploaddup<Packet2cf>(const std::complex<float>*      from) {  return pset1<Packet2cf>(*from); }
+
+template<> EIGEN_STRONG_INLINE void prefetch<std::complex<float> >(const std::complex<float> *     addr) { EIGEN_ZVECTOR_PREFETCH(addr); }
 template<> EIGEN_STRONG_INLINE void prefetch<std::complex<double> >(const std::complex<double> *   addr) { EIGEN_ZVECTOR_PREFETCH(addr); }
 
 template<> EIGEN_STRONG_INLINE std::complex<double>  pfirst<Packet1cd>(const Packet1cd& a)
 {
-  std::complex<double> EIGEN_ALIGN16 res[2];
-  pstore<std::complex<double> >(res, a);
+  std::complex<double> EIGEN_ALIGN16 res;
+  pstore<std::complex<double> >(&res, a);
+
+  return res;
+}
+template<> EIGEN_STRONG_INLINE std::complex<float>  pfirst<Packet2cf>(const Packet2cf& a)
+{
+  std::complex<float> EIGEN_ALIGN16 res[2];
+  pstore<std::complex<float> >(res, a);
 
   return res[0];
 }
 
 template<> EIGEN_STRONG_INLINE Packet1cd preverse(const Packet1cd& a) { return a; }
+template<> EIGEN_STRONG_INLINE Packet2cf preverse(const Packet2cf& a)
+{
+  Packet2cf res;
+  res.cd[0] = a.cd[1];
+  res.cd[1] = a.cd[0];
+  return res;
+}
 
 template<> EIGEN_STRONG_INLINE std::complex<double> predux<Packet1cd>(const Packet1cd& a)
 {
   return pfirst(a);
 }
+template<> EIGEN_STRONG_INLINE std::complex<float> predux<Packet2cf>(const Packet2cf& a)
+{
+  std::complex<float> res;
+  Packet1cd b = padd<Packet1cd>(a.cd[0], a.cd[1]);
+  vec_st2f(b.v, (float*)&res);
+  return res;
+}
 
 template<> EIGEN_STRONG_INLINE Packet1cd preduxp<Packet1cd>(const Packet1cd* vecs)
 {
   return vecs[0];
 }
+template<> EIGEN_STRONG_INLINE Packet2cf preduxp<Packet2cf>(const Packet2cf* vecs)
+{
+  PacketBlock<Packet2cf,2> transpose;
+  transpose.packet[0] = vecs[0];
+  transpose.packet[1] = vecs[1];
+  ptranspose(transpose);
+
+  return padd<Packet2cf>(transpose.packet[0], transpose.packet[1]);
+} 
 
 template<> EIGEN_STRONG_INLINE std::complex<double> predux_mul<Packet1cd>(const Packet1cd& a)
 {
   return pfirst(a);
 }
+template<> EIGEN_STRONG_INLINE std::complex<float> predux_mul<Packet2cf>(const Packet2cf& a)
+{
+  std::complex<float> res;
+  Packet1cd b = pmul<Packet1cd>(a.cd[0], a.cd[1]);
+  vec_st2f(b.v, (float*)&res);
+  return res;
+}
 
 template<int Offset>
 struct palign_impl<Offset,Packet1cd>
@@ -127,6 +258,18 @@ struct palign_impl<Offset,Packet1cd>
   }
 };
 
+template<int Offset>
+struct palign_impl<Offset,Packet2cf>
+{
+  static EIGEN_STRONG_INLINE void run(Packet2cf& first, const Packet2cf& second)
+  {
+    if (Offset == 1) {
+      first.cd[0] = first.cd[1];
+      first.cd[1] = second.cd[0];
+    }
+  }
+};
+
 template<> struct conj_helper<Packet1cd, Packet1cd, false,true>
 {
   EIGEN_STRONG_INLINE Packet1cd pmadd(const Packet1cd& x, const Packet1cd& y, const Packet1cd& c) const
@@ -160,6 +303,39 @@ template<> struct conj_helper<Packet1cd, Packet1cd, true,true>
   }
 };
 
+template<> struct conj_helper<Packet2cf, Packet2cf, false,true>
+{
+  EIGEN_STRONG_INLINE Packet2cf pmadd(const Packet2cf& x, const Packet2cf& y, const Packet2cf& c) const
+  { return padd(pmul(x,y),c); }
+
+  EIGEN_STRONG_INLINE Packet2cf pmul(const Packet2cf& a, const Packet2cf& b) const
+  {
+    return internal::pmul(a, pconj(b));
+  }
+};
+
+template<> struct conj_helper<Packet2cf, Packet2cf, true,false>
+{
+  EIGEN_STRONG_INLINE Packet2cf pmadd(const Packet2cf& x, const Packet2cf& y, const Packet2cf& c) const
+  { return padd(pmul(x,y),c); }
+
+  EIGEN_STRONG_INLINE Packet2cf pmul(const Packet2cf& a, const Packet2cf& b) const
+  {
+    return internal::pmul(pconj(a), b);
+  }
+};
+
+template<> struct conj_helper<Packet2cf, Packet2cf, true,true>
+{
+  EIGEN_STRONG_INLINE Packet2cf pmadd(const Packet2cf& x, const Packet2cf& y, const Packet2cf& c) const
+  { return padd(pmul(x,y),c); }
+
+  EIGEN_STRONG_INLINE Packet2cf pmul(const Packet2cf& a, const Packet2cf& b) const
+  {
+    return pconj(internal::pmul(a, b));
+  }
+};
+
 template<> EIGEN_STRONG_INLINE Packet1cd pdiv<Packet1cd>(const Packet1cd& a, const Packet1cd& b)
 {
   // TODO optimize it for AltiVec
@@ -168,17 +344,49 @@ template<> EIGEN_STRONG_INLINE Packet1cd pdiv<Packet1cd>(const Packet1cd& a, con
   return Packet1cd(pdiv(res.v, s + vec_perm(s, s, p16uc_REVERSE64)));
 }
 
+template<> EIGEN_STRONG_INLINE Packet2cf pdiv<Packet2cf>(const Packet2cf& a, const Packet2cf& b)
+{
+  // TODO optimize it for AltiVec
+  Packet2cf res;
+  res.cd[0] = pdiv<Packet1cd>(a.cd[0], b.cd[0]);
+  res.cd[1] = pdiv<Packet1cd>(a.cd[1], b.cd[1]);
+  return res;
+}
+
 EIGEN_STRONG_INLINE Packet1cd pcplxflip/*<Packet1cd>*/(const Packet1cd& x)
 {
   return Packet1cd(preverse(Packet2d(x.v)));
 }
 
+EIGEN_STRONG_INLINE Packet2cf pcplxflip/*<Packet2cf>*/(const Packet2cf& x)
+{
+  Packet2cf res;
+  res.cd[0] = pcplxflip(x.cd[0]);
+  res.cd[1] = pcplxflip(x.cd[1]);
+  return res;
+}
+
 EIGEN_STRONG_INLINE void ptranspose(PacketBlock<Packet1cd,2>& kernel)
 {
   Packet2d tmp = vec_perm(kernel.packet[0].v, kernel.packet[1].v, p16uc_TRANSPOSE64_HI);
   kernel.packet[1].v = vec_perm(kernel.packet[0].v, kernel.packet[1].v, p16uc_TRANSPOSE64_LO);
   kernel.packet[0].v = tmp;
 }
+
+EIGEN_STRONG_INLINE void ptranspose(PacketBlock<Packet2cf,2>& kernel)
+{
+  Packet1cd tmp = kernel.packet[0].cd[1];
+  kernel.packet[0].cd[1] = kernel.packet[1].cd[0];
+  kernel.packet[1].cd[0] = tmp;
+}
+
+template<> EIGEN_STRONG_INLINE Packet2cf pblend(const Selector<2>& ifPacket, const Packet2cf& thenPacket, const Packet2cf& elsePacket) {
+  Packet2cf result;
+  const Selector<4> ifPacket4 = { ifPacket.select[0], ifPacket.select[0], ifPacket.select[1], ifPacket.select[1] };
+  result.v = pblend<Packet4f>(ifPacket4, thenPacket.v, elsePacket.v);
+  return result;
+}
+
 } // end namespace internal
 
 } // end namespace Eigen

Eigen/src/Core/arch/ZVector/MathFunctions.h

@@ -3,6 +3,7 @@
 //
 // Copyright (C) 2007 Julien Pommier
 // Copyright (C) 2009 Gael Guennebaud <gael.guennebaud@inria.fr>
+// Copyright (C) 2016 Konstantinos Margaritis <markos@freevec.org>
 //
 // 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
@@ -19,32 +20,32 @@ namespace Eigen {
 
 namespace internal {
 
+static _EIGEN_DECLARE_CONST_Packet2d(1 , 1.0);
+static _EIGEN_DECLARE_CONST_Packet2d(2 , 2.0);
+static _EIGEN_DECLARE_CONST_Packet2d(half, 0.5);
+
+static _EIGEN_DECLARE_CONST_Packet2d(exp_hi,  709.437);
+static _EIGEN_DECLARE_CONST_Packet2d(exp_lo, -709.436139303);
+
+static _EIGEN_DECLARE_CONST_Packet2d(cephes_LOG2EF, 1.4426950408889634073599);
+
+static _EIGEN_DECLARE_CONST_Packet2d(cephes_exp_p0, 1.26177193074810590878e-4);
+static _EIGEN_DECLARE_CONST_Packet2d(cephes_exp_p1, 3.02994407707441961300e-2);
+static _EIGEN_DECLARE_CONST_Packet2d(cephes_exp_p2, 9.99999999999999999910e-1);
+
+static _EIGEN_DECLARE_CONST_Packet2d(cephes_exp_q0, 3.00198505138664455042e-6);
+static _EIGEN_DECLARE_CONST_Packet2d(cephes_exp_q1, 2.52448340349684104192e-3);
+static _EIGEN_DECLARE_CONST_Packet2d(cephes_exp_q2, 2.27265548208155028766e-1);
+static _EIGEN_DECLARE_CONST_Packet2d(cephes_exp_q3, 2.00000000000000000009e0);
+
+static _EIGEN_DECLARE_CONST_Packet2d(cephes_exp_C1, 0.693145751953125);
+static _EIGEN_DECLARE_CONST_Packet2d(cephes_exp_C2, 1.42860682030941723212e-6);
+
 template<> EIGEN_DEFINE_FUNCTION_ALLOWING_MULTIPLE_DEFINITIONS EIGEN_UNUSED
 Packet2d pexp<Packet2d>(const Packet2d& _x)
 {
   Packet2d x = _x;
 
-  _EIGEN_DECLARE_CONST_Packet2d(1 , 1.0);
-  _EIGEN_DECLARE_CONST_Packet2d(2 , 2.0);
-  _EIGEN_DECLARE_CONST_Packet2d(half, 0.5);
-
-  _EIGEN_DECLARE_CONST_Packet2d(exp_hi,  709.437);
-  _EIGEN_DECLARE_CONST_Packet2d(exp_lo, -709.436139303);
-
-  _EIGEN_DECLARE_CONST_Packet2d(cephes_LOG2EF, 1.4426950408889634073599);
-
-  _EIGEN_DECLARE_CONST_Packet2d(cephes_exp_p0, 1.26177193074810590878e-4);
-  _EIGEN_DECLARE_CONST_Packet2d(cephes_exp_p1, 3.02994407707441961300e-2);
-  _EIGEN_DECLARE_CONST_Packet2d(cephes_exp_p2, 9.99999999999999999910e-1);
-
-  _EIGEN_DECLARE_CONST_Packet2d(cephes_exp_q0, 3.00198505138664455042e-6);
-  _EIGEN_DECLARE_CONST_Packet2d(cephes_exp_q1, 2.52448340349684104192e-3);
-  _EIGEN_DECLARE_CONST_Packet2d(cephes_exp_q2, 2.27265548208155028766e-1);
-  _EIGEN_DECLARE_CONST_Packet2d(cephes_exp_q3, 2.00000000000000000009e0);
-
-  _EIGEN_DECLARE_CONST_Packet2d(cephes_exp_C1, 0.693145751953125);
-  _EIGEN_DECLARE_CONST_Packet2d(cephes_exp_C2, 1.42860682030941723212e-6);
-
   Packet2d tmp, fx;
   Packet2l emm0;
 
@@ -91,18 +92,44 @@ Packet2d pexp<Packet2d>(const Packet2d& _x)
                  isnumber_mask);
 }
 
+template<> EIGEN_DEFINE_FUNCTION_ALLOWING_MULTIPLE_DEFINITIONS EIGEN_UNUSED
+Packet4f pexp<Packet4f>(const Packet4f& x)
+{
+  Packet4f res;
+  res.v4f[0] = pexp<Packet2d>(x.v4f[0]);
+  res.v4f[1] = pexp<Packet2d>(x.v4f[1]);
+  return res;
+}
+
 template<> EIGEN_DEFINE_FUNCTION_ALLOWING_MULTIPLE_DEFINITIONS EIGEN_UNUSED
 Packet2d psqrt<Packet2d>(const Packet2d& x)
 {
   return  __builtin_s390_vfsqdb(x);
 }
 
+template<> EIGEN_DEFINE_FUNCTION_ALLOWING_MULTIPLE_DEFINITIONS EIGEN_UNUSED
+Packet4f psqrt<Packet4f>(const Packet4f& x)
+{
+  Packet4f res;
+  res.v4f[0] = psqrt<Packet2d>(x.v4f[0]);
+  res.v4f[1] = psqrt<Packet2d>(x.v4f[1]);
+  return res;
+}
+
 template<> EIGEN_DEFINE_FUNCTION_ALLOWING_MULTIPLE_DEFINITIONS EIGEN_UNUSED
 Packet2d prsqrt<Packet2d>(const Packet2d& x) {
   // Unfortunately we can't use the much faster mm_rqsrt_pd since it only provides an approximation.
   return pset1<Packet2d>(1.0) / psqrt<Packet2d>(x);
 }
 
+template<> EIGEN_DEFINE_FUNCTION_ALLOWING_MULTIPLE_DEFINITIONS EIGEN_UNUSED
+Packet4f prsqrt<Packet4f>(const Packet4f& x) {
+  Packet4f res;
+  res.v4f[0] = prsqrt<Packet2d>(x.v4f[0]);
+  res.v4f[1] = prsqrt<Packet2d>(x.v4f[1]);
+  return res;
+}
+
 }  // end namespace internal
 
 }  // end namespace Eigen

Eigen/src/Core/arch/ZVector/PacketMath.h

@@ -28,9 +28,8 @@ namespace internal {
 #define EIGEN_HAS_SINGLE_INSTRUCTION_CJMADD
 #endif
 
-// NOTE Altivec has 32 registers, but Eigen only accepts a value of 8 or 16
 #ifndef EIGEN_ARCH_DEFAULT_NUMBER_OF_REGISTERS
-#define EIGEN_ARCH_DEFAULT_NUMBER_OF_REGISTERS  32
+#define EIGEN_ARCH_DEFAULT_NUMBER_OF_REGISTERS  16
 #endif
 
 typedef __vector int                 Packet4i;
@@ -42,6 +41,10 @@ typedef __vector double              Packet2d;
 typedef __vector unsigned long long  Packet2ul;
 typedef __vector long long           Packet2l;
 
+typedef struct {
+	Packet2d  v4f[2];
+} Packet4f;
+
 typedef union {
   int32_t   i[4];
   uint32_t ui[4];
@@ -88,6 +91,7 @@ static Packet2d p2d_ONE = { 1.0, 1.0 };
 static Packet2d p2d_ZERO_ = { -0.0, -0.0 };
 
 static Packet4i p4i_COUNTDOWN = { 0, 1, 2, 3 };
+static Packet4f p4f_COUNTDOWN = { 0.0, 1.0, 2.0, 3.0 };
 static Packet2d p2d_COUNTDOWN = reinterpret_cast<Packet2d>(vec_sld(reinterpret_cast<Packet16uc>(p2d_ZERO), reinterpret_cast<Packet16uc>(p2d_ONE), 8));
 
 static Packet16uc p16uc_PSET64_HI = { 0,1,2,3, 4,5,6,7, 0,1,2,3, 4,5,6,7 };
@@ -96,7 +100,7 @@ static Packet16uc p16uc_DUPLICATE32_HI = { 0,1,2,3, 0,1,2,3, 4,5,6,7, 4,5,6,7 };
 // Mask alignment
 #define _EIGEN_MASK_ALIGNMENT	0xfffffffffffffff0
 
-#define _EIGEN_ALIGNED_PTR(x)	((ptrdiff_t)(x) & _EIGEN_MASK_ALIGNMENT)
+#define _EIGEN_ALIGNED_PTR(x)	((std::ptrdiff_t)(x) & _EIGEN_MASK_ALIGNMENT)
 
 // Handle endianness properly while loading constants
 // Define global static constants:
@@ -132,13 +136,11 @@ template<> struct packet_traits<int>    : default_packet_traits
   typedef Packet4i type;
   typedef Packet4i half;
   enum {
-    // FIXME check the Has*
     Vectorizable = 1,
     AlignedOnScalar = 1,
     size = 4,
     HasHalfPacket = 0,
 
-    // FIXME check the Has*
     HasAdd  = 1,
     HasSub  = 1,
     HasMul  = 1,
@@ -147,6 +149,37 @@ template<> struct packet_traits<int>    : default_packet_traits
   };
 };
 
+template<> struct packet_traits<float> : default_packet_traits
+{
+  typedef Packet4f type;
+  typedef Packet4f half;
+  enum {
+    Vectorizable = 1,
+    AlignedOnScalar = 1,
+    size=4,
+    HasHalfPacket = 0,
+
+    HasAdd  = 1,
+    HasSub  = 1,
+    HasMul  = 1,
+    HasDiv  = 1,
+    HasMin  = 1,
+    HasMax  = 1,
+    HasAbs  = 1,
+    HasSin  = 0,
+    HasCos  = 0,
+    HasLog  = 0,
+    HasExp  = 1,
+    HasSqrt = 1,
+    HasRsqrt = 1,
+    HasRound = 1,
+    HasFloor = 1,
+    HasCeil = 1,
+    HasNegate = 1,
+    HasBlend = 1
+  };
+};
+
 template<> struct packet_traits<double> : default_packet_traits
 {
   typedef Packet2d type;
@@ -157,7 +190,6 @@ template<> struct packet_traits<double> : default_packet_traits
     size=2,
     HasHalfPacket = 1,
 
-    // FIXME check the Has*
     HasAdd  = 1,
     HasSub  = 1,
     HasMul  = 1,
@@ -180,8 +212,12 @@ template<> struct packet_traits<double> : default_packet_traits
 };
 
 template<> struct unpacket_traits<Packet4i> { typedef int    type; enum {size=4, alignment=Aligned16}; typedef Packet4i half; };
+template<> struct unpacket_traits<Packet4f> { typedef float  type; enum {size=4, alignment=Aligned16}; typedef Packet4f half; };
 template<> struct unpacket_traits<Packet2d> { typedef double type; enum {size=2, alignment=Aligned16}; typedef Packet2d half; };
 
+/* Forward declaration */
+EIGEN_DEVICE_FUNC inline void ptranspose(PacketBlock<Packet4f,4>& kernel);
+ 
 inline std::ostream & operator <<(std::ostream & s, const Packet4i & v)
 {
   Packet vt;
@@ -222,6 +258,32 @@ inline std::ostream & operator <<(std::ostream & s, const Packet2d & v)
   return s;
 }
 
+/* Helper function to simulate a vec_splat_packet4f
+ */
+template<int element> EIGEN_STRONG_INLINE Packet4f vec_splat_packet4f(const Packet4f&   from)
+{
+  Packet4f splat;
+  switch (element) {
+  case 0:
+    splat.v4f[0] = vec_splat(from.v4f[0], 0);
+    splat.v4f[1] = splat.v4f[0];
+    break;
+  case 1:
+    splat.v4f[0] = vec_splat(from.v4f[0], 1);
+    splat.v4f[1] = splat.v4f[0];
+    break;
+  case 2:
+    splat.v4f[0] = vec_splat(from.v4f[1], 0);
+    splat.v4f[1] = splat.v4f[0];
+    break;
+  case 3:
+    splat.v4f[0] = vec_splat(from.v4f[1], 1);
+    splat.v4f[1] = splat.v4f[0];
+    break;
+  }
+  return splat;
+}
+
 template<int Offset>
 struct palign_impl<Offset,Packet4i>
 {
@@ -238,6 +300,31 @@ struct palign_impl<Offset,Packet4i>
   }
 };
 
+/* This is a tricky one, we have to translate float alignment to vector elements of sizeof double
+ */
+template<int Offset>
+struct palign_impl<Offset,Packet4f>
+{
+  static EIGEN_STRONG_INLINE void run(Packet4f& first, const Packet4f& second)
+  {
+    switch (Offset % 4) {
+    case 1:
+      first.v4f[0] = vec_sld(first.v4f[0], first.v4f[1], 8);
+      first.v4f[1] = vec_sld(first.v4f[1], second.v4f[0], 8);
+      break;
+    case 2:
+      first.v4f[0] = first.v4f[1];
+      first.v4f[1] = second.v4f[0];
+      break;
+    case 3:
+      first.v4f[0] = vec_sld(first.v4f[1],  second.v4f[0], 8);
+      first.v4f[1] = vec_sld(second.v4f[0], second.v4f[1], 8);
+      break;
+    }
+  }
+};
+
+
 template<int Offset>
 struct palign_impl<Offset,Packet2d>
 {
@@ -257,6 +344,16 @@ template<> EIGEN_STRONG_INLINE Packet4i pload<Packet4i>(const int*     from)
   return vfrom->v4i;
 }
 
+template<> EIGEN_STRONG_INLINE Packet4f pload<Packet4f>(const float*   from)
+{
+  // FIXME: No intrinsic yet
+  EIGEN_DEBUG_ALIGNED_LOAD
+  Packet4f vfrom;
+  vfrom.v4f[0] = vec_ld2f(&from[0]);
+  vfrom.v4f[1] = vec_ld2f(&from[2]);
+  return vfrom;
+}
+
 template<> EIGEN_STRONG_INLINE Packet2d pload<Packet2d>(const double* from)
 {
   // FIXME: No intrinsic yet
@@ -275,6 +372,15 @@ template<> EIGEN_STRONG_INLINE void pstore<int>(int*       to, const Packet4i& f
   vto->v4i = from;
 }
 
+template<> EIGEN_STRONG_INLINE void pstore<float>(float*   to, const Packet4f& from)
+{
+  // FIXME: No intrinsic yet
+  EIGEN_DEBUG_ALIGNED_STORE
+  vec_st2f(from.v4f[0], &to[0]);
+  vec_st2f(from.v4f[1], &to[2]);
+}
+
+
 template<> EIGEN_STRONG_INLINE void pstore<double>(double*   to, const Packet2d& from)
 {
   // FIXME: No intrinsic yet
@@ -288,10 +394,16 @@ template<> EIGEN_STRONG_INLINE Packet4i pset1<Packet4i>(const int&    from)
 {
   return vec_splats(from);
 }
-
 template<> EIGEN_STRONG_INLINE Packet2d pset1<Packet2d>(const double& from) {
   return vec_splats(from);
 }
+template<> EIGEN_STRONG_INLINE Packet4f pset1<Packet4f>(const float&    from)
+{
+  Packet4f to;
+  to.v4f[0] = pset1<Packet2d>(static_cast<const double&>(from));
+  to.v4f[1] = to.v4f[0];
+  return to;
+}
 
 template<> EIGEN_STRONG_INLINE void
 pbroadcast4<Packet4i>(const int *a,
@@ -304,6 +416,17 @@ pbroadcast4<Packet4i>(const int *a,
   a3 = vec_splat(a3, 3);
 }
 
+template<> EIGEN_STRONG_INLINE void
+pbroadcast4<Packet4f>(const float *a,
+                      Packet4f& a0, Packet4f& a1, Packet4f& a2, Packet4f& a3)
+{
+  a3 = pload<Packet4f>(a);
+  a0 = vec_splat_packet4f<0>(a3);
+  a1 = vec_splat_packet4f<1>(a3);
+  a2 = vec_splat_packet4f<2>(a3);
+  a3 = vec_splat_packet4f<3>(a3);
+}
+
 template<> EIGEN_STRONG_INLINE void
 pbroadcast4<Packet2d>(const double *a,
                       Packet2d& a0, Packet2d& a1, Packet2d& a2, Packet2d& a3)
@@ -326,6 +449,16 @@ template<> EIGEN_DEVICE_FUNC inline Packet4i pgather<int, Packet4i>(const int* f
  return pload<Packet4i>(ai);
 }
 
+template<> EIGEN_DEVICE_FUNC inline Packet4f pgather<float, Packet4f>(const float* from, Index stride)
+{
+  float EIGEN_ALIGN16 ai[4];
+  ai[0] = from[0*stride];
+  ai[1] = from[1*stride];
+  ai[2] = from[2*stride];
+  ai[3] = from[3*stride];
+ return pload<Packet4f>(ai);
+}
+
 template<> EIGEN_DEVICE_FUNC inline Packet2d pgather<double, Packet2d>(const double* from, Index stride)
 {
   double EIGEN_ALIGN16 af[2];
@@ -344,6 +477,16 @@ template<> EIGEN_DEVICE_FUNC inline void pscatter<int, Packet4i>(int* to, const
   to[3*stride] = ai[3];
 }
 
+template<> EIGEN_DEVICE_FUNC inline void pscatter<float, Packet4f>(float* to, const Packet4f& from, Index stride)
+{
+  float EIGEN_ALIGN16 ai[4];
+  pstore<float>((float *)ai, from);
+  to[0*stride] = ai[0];
+  to[1*stride] = ai[1];
+  to[2*stride] = ai[2];
+  to[3*stride] = ai[3];
+}
+
 template<> EIGEN_DEVICE_FUNC inline void pscatter<double, Packet2d>(double* to, const Packet2d& from, Index stride)
 {
   double EIGEN_ALIGN16 af[2];
@@ -353,52 +496,160 @@ template<> EIGEN_DEVICE_FUNC inline void pscatter<double, Packet2d>(double* to,
 }
 
 template<> EIGEN_STRONG_INLINE Packet4i padd<Packet4i>(const Packet4i& a, const Packet4i& b) { return (a + b); }
+template<> EIGEN_STRONG_INLINE Packet4f padd<Packet4f>(const Packet4f& a, const Packet4f& b)
+{
+  Packet4f c;
+  c.v4f[0] = a.v4f[0] + b.v4f[0];
+  c.v4f[1] = a.v4f[1] + b.v4f[1];
+  return c;
+}
 template<> EIGEN_STRONG_INLINE Packet2d padd<Packet2d>(const Packet2d& a, const Packet2d& b) { return (a + b); }
 
 template<> EIGEN_STRONG_INLINE Packet4i psub<Packet4i>(const Packet4i& a, const Packet4i& b) { return (a - b); }
+template<> EIGEN_STRONG_INLINE Packet4f psub<Packet4f>(const Packet4f& a, const Packet4f& b)
+{
+  Packet4f c;
+  c.v4f[0] = a.v4f[0] - b.v4f[0];
+  c.v4f[1] = a.v4f[1] - b.v4f[1];
+  return c;
+}
 template<> EIGEN_STRONG_INLINE Packet2d psub<Packet2d>(const Packet2d& a, const Packet2d& b) { return (a - b); }
 
 template<> EIGEN_STRONG_INLINE Packet4i pmul<Packet4i>(const Packet4i& a, const Packet4i& b) { return (a * b); }
+template<> EIGEN_STRONG_INLINE Packet4f pmul<Packet4f>(const Packet4f& a, const Packet4f& b)
+{
+  Packet4f c;
+  c.v4f[0] = a.v4f[0] * b.v4f[0];
+  c.v4f[1] = a.v4f[1] * b.v4f[1];
+  return c;
+}
 template<> EIGEN_STRONG_INLINE Packet2d pmul<Packet2d>(const Packet2d& a, const Packet2d& b) { return (a * b); }
 
 template<> EIGEN_STRONG_INLINE Packet4i pdiv<Packet4i>(const Packet4i& a, const Packet4i& b) { return (a / b); }
+template<> EIGEN_STRONG_INLINE Packet4f pdiv<Packet4f>(const Packet4f& a, const Packet4f& b)
+{
+  Packet4f c;
+  c.v4f[0] = a.v4f[0] / b.v4f[0];
+  c.v4f[1] = a.v4f[1] / b.v4f[1];
+  return c;
+}
 template<> EIGEN_STRONG_INLINE Packet2d pdiv<Packet2d>(const Packet2d& a, const Packet2d& b) { return (a / b); }
 
 template<> EIGEN_STRONG_INLINE Packet4i pnegate(const Packet4i& a) { return (-a); }
+template<> EIGEN_STRONG_INLINE Packet4f pnegate(const Packet4f& a)
+{
+  Packet4f c;
+  c.v4f[0] = -a.v4f[0];
+  c.v4f[1] = -a.v4f[1];
+  return c;
+}
 template<> EIGEN_STRONG_INLINE Packet2d pnegate(const Packet2d& a) { return (-a); }
 
 template<> EIGEN_STRONG_INLINE Packet4i pconj(const Packet4i& a) { return a; }
+template<> EIGEN_STRONG_INLINE Packet4f pconj(const Packet4f& a) { return a; }
 template<> EIGEN_STRONG_INLINE Packet2d pconj(const Packet2d& a) { return a; }
 
 template<> EIGEN_STRONG_INLINE Packet4i pmadd(const Packet4i& a, const Packet4i& b, const Packet4i& c) { return padd<Packet4i>(pmul<Packet4i>(a, b), c); }
+template<> EIGEN_STRONG_INLINE Packet4f pmadd(const Packet4f& a, const Packet4f& b, const Packet4f& c)
+{
+  Packet4f res;
+  res.v4f[0] = vec_madd(a.v4f[0], b.v4f[0], c.v4f[0]);
+  res.v4f[1] = vec_madd(a.v4f[1], b.v4f[1], c.v4f[1]);
+  return res;
+}
 template<> EIGEN_STRONG_INLINE Packet2d pmadd(const Packet2d& a, const Packet2d& b, const Packet2d& c) { return vec_madd(a, b, c); }
 
 template<> EIGEN_STRONG_INLINE Packet4i plset<Packet4i>(const int& a)    { return padd<Packet4i>(pset1<Packet4i>(a), p4i_COUNTDOWN); }
+template<> EIGEN_STRONG_INLINE Packet4f plset<Packet4f>(const float& a)  { return padd<Packet4f>(pset1<Packet4f>(a), p4f_COUNTDOWN); }
 template<> EIGEN_STRONG_INLINE Packet2d plset<Packet2d>(const double& a) { return padd<Packet2d>(pset1<Packet2d>(a), p2d_COUNTDOWN); }
 
 template<> EIGEN_STRONG_INLINE Packet4i pmin<Packet4i>(const Packet4i& a, const Packet4i& b) { return vec_min(a, b); }
 template<> EIGEN_STRONG_INLINE Packet2d pmin<Packet2d>(const Packet2d& a, const Packet2d& b) { return vec_min(a, b); }
+template<> EIGEN_STRONG_INLINE Packet4f pmin<Packet4f>(const Packet4f& a, const Packet4f& b)
+{
+  Packet4f res;
+  res.v4f[0] = pmin(a.v4f[0], b.v4f[0]);
+  res.v4f[1] = pmin(a.v4f[1], b.v4f[1]);
+  return res;
+}
 
 template<> EIGEN_STRONG_INLINE Packet4i pmax<Packet4i>(const Packet4i& a, const Packet4i& b) { return vec_max(a, b); }
 template<> EIGEN_STRONG_INLINE Packet2d pmax<Packet2d>(const Packet2d& a, const Packet2d& b) { return vec_max(a, b); }
+template<> EIGEN_STRONG_INLINE Packet4f pmax<Packet4f>(const Packet4f& a, const Packet4f& b)
+{
+  Packet4f res;
+  res.v4f[0] = pmax(a.v4f[0], b.v4f[0]);
+  res.v4f[1] = pmax(a.v4f[1], b.v4f[1]);
+  return res;
+}
 
 template<> EIGEN_STRONG_INLINE Packet4i pand<Packet4i>(const Packet4i& a, const Packet4i& b) { return vec_and(a, b); }
 template<> EIGEN_STRONG_INLINE Packet2d pand<Packet2d>(const Packet2d& a, const Packet2d& b) { return vec_and(a, b); }
+template<> EIGEN_STRONG_INLINE Packet4f pand<Packet4f>(const Packet4f& a, const Packet4f& b)
+{
+  Packet4f res;
+  res.v4f[0] = pand(a.v4f[0], b.v4f[0]);
+  res.v4f[1] = pand(a.v4f[1], b.v4f[1]);
+  return res;
+}
 
 template<> EIGEN_STRONG_INLINE Packet4i por<Packet4i>(const Packet4i& a, const Packet4i& b) { return vec_or(a, b); }
 template<> EIGEN_STRONG_INLINE Packet2d por<Packet2d>(const Packet2d& a, const Packet2d& b) { return vec_or(a, b); }
+template<> EIGEN_STRONG_INLINE Packet4f por<Packet4f>(const Packet4f& a, const Packet4f& b)
+{
+  Packet4f res;
+  res.v4f[0] = pand(a.v4f[0], b.v4f[0]);
+  res.v4f[1] = pand(a.v4f[1], b.v4f[1]);
+  return res;
+}
 
 template<> EIGEN_STRONG_INLINE Packet4i pxor<Packet4i>(const Packet4i& a, const Packet4i& b) { return vec_xor(a, b); }
 template<> EIGEN_STRONG_INLINE Packet2d pxor<Packet2d>(const Packet2d& a, const Packet2d& b) { return vec_xor(a, b); }
+template<> EIGEN_STRONG_INLINE Packet4f pxor<Packet4f>(const Packet4f& a, const Packet4f& b)
+{
+  Packet4f res;
+  res.v4f[0] = pand(a.v4f[0], b.v4f[0]);
+  res.v4f[1] = pand(a.v4f[1], b.v4f[1]);
+  return res;
+}
 
 template<> EIGEN_STRONG_INLINE Packet4i pandnot<Packet4i>(const Packet4i& a, const Packet4i& b) { return pand<Packet4i>(a, vec_nor(b, b)); }
 template<> EIGEN_STRONG_INLINE Packet2d pandnot<Packet2d>(const Packet2d& a, const Packet2d& b) { return vec_and(a, vec_nor(b, b)); }
+template<> EIGEN_STRONG_INLINE Packet4f pandnot<Packet4f>(const Packet4f& a, const Packet4f& b)
+{
+  Packet4f res;
+  res.v4f[0] = pandnot(a.v4f[0], b.v4f[0]);
+  res.v4f[1] = pandnot(a.v4f[1], b.v4f[1]);
+  return res;
+}
 
+template<> EIGEN_STRONG_INLINE Packet4f pround<Packet4f>(const Packet4f& a)
+{
+  Packet4f res;
+  res.v4f[0] = vec_round(a.v4f[0]);
+  res.v4f[1] = vec_round(a.v4f[1]);
+  return res;
+}
 template<> EIGEN_STRONG_INLINE Packet2d pround<Packet2d>(const Packet2d& a) { return vec_round(a); }
+template<> EIGEN_STRONG_INLINE Packet4f pceil<Packet4f>(const  Packet4f& a)
+{
+  Packet4f res;
+  res.v4f[0] = vec_ceil(a.v4f[0]);
+  res.v4f[1] = vec_ceil(a.v4f[1]);
+  return res;
+}
 template<> EIGEN_STRONG_INLINE Packet2d pceil<Packet2d>(const  Packet2d& a) { return vec_ceil(a); }
+template<> EIGEN_STRONG_INLINE Packet4f pfloor<Packet4f>(const Packet4f& a)
+{
+  Packet4f res;
+  res.v4f[0] = vec_floor(a.v4f[0]);
+  res.v4f[1] = vec_floor(a.v4f[1]);
+  return res;
+}
 template<> EIGEN_STRONG_INLINE Packet2d pfloor<Packet2d>(const Packet2d& a) { return vec_floor(a); }
 
 template<> EIGEN_STRONG_INLINE Packet4i ploadu<Packet4i>(const int*       from) { return pload<Packet4i>(from); }
+template<> EIGEN_STRONG_INLINE Packet4f ploadu<Packet4f>(const float*     from) { return pload<Packet4f>(from); }
 template<> EIGEN_STRONG_INLINE Packet2d ploadu<Packet2d>(const double*    from) { return pload<Packet2d>(from); }
 
 
@@ -408,6 +659,14 @@ template<> EIGEN_STRONG_INLINE Packet4i ploaddup<Packet4i>(const int*     from)
   return vec_perm(p, p, p16uc_DUPLICATE32_HI);
 }
 
+template<> EIGEN_STRONG_INLINE Packet4f ploaddup<Packet4f>(const float*    from)
+{
+  Packet4f p = pload<Packet4f>(from);
+  p.v4f[1] = vec_splat(p.v4f[0], 1);
+  p.v4f[0] = vec_splat(p.v4f[0], 0);
+  return p;
+}
+
 template<> EIGEN_STRONG_INLINE Packet2d ploaddup<Packet2d>(const double*   from)
 {
   Packet2d p = pload<Packet2d>(from);
@@ -415,12 +674,15 @@ template<> EIGEN_STRONG_INLINE Packet2d ploaddup<Packet2d>(const double*   from)
 }
 
 template<> EIGEN_STRONG_INLINE void pstoreu<int>(int*        to, const Packet4i& from) { pstore<int>(to, from); }
+template<> EIGEN_STRONG_INLINE void pstoreu<float>(float*    to, const Packet4f& from) { pstore<float>(to, from); }
 template<> EIGEN_STRONG_INLINE void pstoreu<double>(double*  to, const Packet2d& from) { pstore<double>(to, from); }
 
 template<> EIGEN_STRONG_INLINE void prefetch<int>(const int*       addr) { EIGEN_ZVECTOR_PREFETCH(addr); }
+template<> EIGEN_STRONG_INLINE void prefetch<float>(const float*   addr) { EIGEN_ZVECTOR_PREFETCH(addr); }
 template<> EIGEN_STRONG_INLINE void prefetch<double>(const double* addr) { EIGEN_ZVECTOR_PREFETCH(addr); }
 
 template<> EIGEN_STRONG_INLINE int    pfirst<Packet4i>(const Packet4i& a) { int    EIGEN_ALIGN16 x[4]; pstore(x, a); return x[0]; }
+template<> EIGEN_STRONG_INLINE float  pfirst<Packet4f>(const Packet4f& a) { float  EIGEN_ALIGN16 x[2]; vec_st2f(a.v4f[0], &x[0]); return x[0]; }
 template<> EIGEN_STRONG_INLINE double pfirst<Packet2d>(const Packet2d& a) { double EIGEN_ALIGN16 x[2]; pstore(x, a); return x[0]; }
 
 template<> EIGEN_STRONG_INLINE Packet4i preverse(const Packet4i& a)
@@ -433,8 +695,23 @@ template<> EIGEN_STRONG_INLINE Packet2d preverse(const Packet2d& a)
   return reinterpret_cast<Packet2d>(vec_perm(reinterpret_cast<Packet16uc>(a), reinterpret_cast<Packet16uc>(a), p16uc_REVERSE64));
 }
 
-template<> EIGEN_STRONG_INLINE Packet4i pabs(const Packet4i& a) { return vec_abs(a); }
-template<> EIGEN_STRONG_INLINE Packet2d pabs(const Packet2d& a) { return vec_abs(a); }
+template<> EIGEN_STRONG_INLINE Packet4f preverse(const Packet4f& a)
+{
+  Packet4f rev;
+  rev.v4f[0] = preverse<Packet2d>(a.v4f[1]);
+  rev.v4f[1] = preverse<Packet2d>(a.v4f[0]);
+  return rev;
+}
+
+template<> EIGEN_STRONG_INLINE Packet4i pabs<Packet4i>(const Packet4i& a) { return vec_abs(a); }
+template<> EIGEN_STRONG_INLINE Packet2d pabs<Packet2d>(const Packet2d& a) { return vec_abs(a); }
+template<> EIGEN_STRONG_INLINE Packet4f pabs<Packet4f>(const Packet4f& a)
+{
+  Packet4f res;
+  res.v4f[0] = pabs(a.v4f[0]);
+  res.v4f[1] = pabs(a.v4f[1]);
+  return res;
+}
 
 template<> EIGEN_STRONG_INLINE int predux<Packet4i>(const Packet4i& a)
 {
@@ -453,6 +730,13 @@ template<> EIGEN_STRONG_INLINE double predux<Packet2d>(const Packet2d& a)
   sum = padd<Packet2d>(a, b);
   return pfirst(sum);
 }
+template<> EIGEN_STRONG_INLINE float predux<Packet4f>(const Packet4f& a)
+{
+  Packet2d sum;
+  sum = padd<Packet2d>(a.v4f[0], a.v4f[1]);
+  double first = predux<Packet2d>(sum);
+  return static_cast<float>(first);
+}
 
 template<> EIGEN_STRONG_INLINE Packet4i preduxp<Packet4i>(const Packet4i* vecs)
 {
@@ -493,6 +777,21 @@ template<> EIGEN_STRONG_INLINE Packet2d preduxp<Packet2d>(const Packet2d* vecs)
   return sum;
 }
 
+template<> EIGEN_STRONG_INLINE Packet4f preduxp<Packet4f>(const Packet4f* vecs)
+{
+  PacketBlock<Packet4f,4> transpose;
+  transpose.packet[0] = vecs[0];
+  transpose.packet[1] = vecs[1];
+  transpose.packet[2] = vecs[2];
+  transpose.packet[3] = vecs[3];
+  ptranspose(transpose);
+
+  Packet4f sum = padd(transpose.packet[0], transpose.packet[1]);
+  sum = padd(sum, transpose.packet[2]);
+  sum = padd(sum, transpose.packet[3]);
+  return sum;
+}
+
 // Other reduction functions:
 // mul
 template<> EIGEN_STRONG_INLINE int predux_mul<Packet4i>(const Packet4i& a)
@@ -507,6 +806,12 @@ template<> EIGEN_STRONG_INLINE double predux_mul<Packet2d>(const Packet2d& a)
   return pfirst(pmul(a, reinterpret_cast<Packet2d>(vec_sld(reinterpret_cast<Packet4i>(a), reinterpret_cast<Packet4i>(a), 8))));
 }
 
+template<> EIGEN_STRONG_INLINE float predux_mul<Packet4f>(const Packet4f& a)
+{
+  // Return predux_mul<Packet2d> of the subvectors product
+  return static_cast<float>(pfirst(predux_mul(pmul(a.v4f[0], a.v4f[1]))));
+}
+
 // min
 template<> EIGEN_STRONG_INLINE int predux_min<Packet4i>(const Packet4i& a)
 {
@@ -521,6 +826,14 @@ template<> EIGEN_STRONG_INLINE double predux_min<Packet2d>(const Packet2d& a)
   return pfirst(pmin<Packet2d>(a, reinterpret_cast<Packet2d>(vec_sld(reinterpret_cast<Packet4i>(a), reinterpret_cast<Packet4i>(a), 8))));
 }
 
+template<> EIGEN_STRONG_INLINE float predux_min<Packet4f>(const Packet4f& a)
+{
+  Packet2d b, res;
+  b   = pmin<Packet2d>(a.v4f[0], a.v4f[1]);
+  res = pmin<Packet2d>(b, reinterpret_cast<Packet2d>(vec_sld(reinterpret_cast<Packet4i>(b), reinterpret_cast<Packet4i>(b), 8)));
+  return static_cast<float>(pfirst(res));
+}
+
 // max
 template<> EIGEN_STRONG_INLINE int predux_max<Packet4i>(const Packet4i& a)
 {
@@ -536,6 +849,14 @@ template<> EIGEN_STRONG_INLINE double predux_max<Packet2d>(const Packet2d& a)
   return pfirst(pmax<Packet2d>(a, reinterpret_cast<Packet2d>(vec_sld(reinterpret_cast<Packet4i>(a), reinterpret_cast<Packet4i>(a), 8))));
 }
 
+template<> EIGEN_STRONG_INLINE float predux_max<Packet4f>(const Packet4f& a)
+{
+  Packet2d b, res;
+  b   = pmax<Packet2d>(a.v4f[0], a.v4f[1]);
+  res = pmax<Packet2d>(b, reinterpret_cast<Packet2d>(vec_sld(reinterpret_cast<Packet4i>(b), reinterpret_cast<Packet4i>(b), 8)));
+  return static_cast<float>(pfirst(res));
+}
+
 EIGEN_DEVICE_FUNC inline void
 ptranspose(PacketBlock<Packet4i,4>& kernel) {
   Packet4i t0 = vec_mergeh(kernel.packet[0], kernel.packet[2]);
@@ -556,12 +877,61 @@ ptranspose(PacketBlock<Packet2d,2>& kernel) {
   kernel.packet[1] = t1;
 }
 
+/* Split the Packet4f PacketBlock into 4 Packet2d PacketBlocks and transpose each one
+ */
+EIGEN_DEVICE_FUNC inline void
+ptranspose(PacketBlock<Packet4f,4>& kernel) {
+  PacketBlock<Packet2d,2> t0,t1,t2,t3;
+  // copy top-left 2x2 Packet2d block
+  t0.packet[0] = kernel.packet[0].v4f[0];
+  t0.packet[1] = kernel.packet[1].v4f[0];
+
+  // copy top-right 2x2 Packet2d block
+  t1.packet[0] = kernel.packet[0].v4f[1];
+  t1.packet[1] = kernel.packet[1].v4f[1];
+
+  // copy bottom-left 2x2 Packet2d block
+  t2.packet[0] = kernel.packet[2].v4f[0];
+  t2.packet[1] = kernel.packet[3].v4f[0];
+
+  // copy bottom-right 2x2 Packet2d block
+  t3.packet[0] = kernel.packet[2].v4f[1];
+  t3.packet[1] = kernel.packet[3].v4f[1];
+
+  // Transpose all 2x2 blocks
+  ptranspose(t0);
+  ptranspose(t1);
+  ptranspose(t2);
+  ptranspose(t3);
+
+  // Copy back transposed blocks, but exchange t1 and t2 due to transposition
+  kernel.packet[0].v4f[0] = t0.packet[0];
+  kernel.packet[0].v4f[1] = t2.packet[0];
+  kernel.packet[1].v4f[0] = t0.packet[1];
+  kernel.packet[1].v4f[1] = t2.packet[1];
+  kernel.packet[2].v4f[0] = t1.packet[0];
+  kernel.packet[2].v4f[1] = t3.packet[0];
+  kernel.packet[3].v4f[0] = t1.packet[1];
+  kernel.packet[3].v4f[1] = t3.packet[1];
+}
+
 template<> EIGEN_STRONG_INLINE Packet4i pblend(const Selector<4>& ifPacket, const Packet4i& thenPacket, const Packet4i& elsePacket) {
   Packet4ui select = { ifPacket.select[0], ifPacket.select[1], ifPacket.select[2], ifPacket.select[3] };
   Packet4ui mask = vec_cmpeq(select, reinterpret_cast<Packet4ui>(p4i_ONE));
   return vec_sel(elsePacket, thenPacket, mask);
 }
 
+template<> EIGEN_STRONG_INLINE Packet4f pblend(const Selector<4>& ifPacket, const Packet4f& thenPacket, const Packet4f& elsePacket) {
+  Packet2ul select_hi = { ifPacket.select[0], ifPacket.select[1] };
+  Packet2ul select_lo = { ifPacket.select[2], ifPacket.select[3] };
+  Packet2ul mask_hi = vec_cmpeq(select_hi, reinterpret_cast<Packet2ul>(p2l_ONE));
+  Packet2ul mask_lo = vec_cmpeq(select_lo, reinterpret_cast<Packet2ul>(p2l_ONE));
+  Packet4f result;
+  result.v4f[0] = vec_sel(elsePacket.v4f[0], thenPacket.v4f[0], mask_hi);
+  result.v4f[1] = vec_sel(elsePacket.v4f[1], thenPacket.v4f[1], mask_lo);
+  return result;
+}
+
 template<> EIGEN_STRONG_INLINE Packet2d pblend(const Selector<2>& ifPacket, const Packet2d& thenPacket, const Packet2d& elsePacket) {
   Packet2ul select = { ifPacket.select[0], ifPacket.select[1] };
   Packet2ul mask = vec_cmpeq(select, reinterpret_cast<Packet2ul>(p2l_ONE));

Eigen/src/Core/functors/AssignmentFunctors.h

@@ -28,7 +28,7 @@ template<typename DstScalar,typename SrcScalar> struct assign_op {
   { internal::pstoret<DstScalar,Packet,Alignment>(a,b); }
 };
 
-// Empty overload for void type (used by PermutationMatrix
+// Empty overload for void type (used by PermutationMatrix)
 template<typename DstScalar> struct assign_op<DstScalar,void> {};
 
 template<typename DstScalar,typename SrcScalar>

Eigen/src/Core/functors/NullaryFunctors.h

@@ -44,16 +44,16 @@ struct linspaced_op_impl<Scalar,Packet,/*IsInteger*/false>
 {
   linspaced_op_impl(const Scalar& low, const Scalar& high, Index num_steps) :
     m_low(low), m_high(high), m_size1(num_steps==1 ? 1 : num_steps-1), m_step(num_steps==1 ? Scalar() : (high-low)/Scalar(num_steps-1)),
-    m_interPacket(plset<Packet>(0)),
-    m_flip(std::abs(high)<std::abs(low))
+    m_flip(numext::abs(high)<numext::abs(low))
   {}
 
   template<typename IndexType>
   EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE const Scalar operator() (IndexType i) const {
+    typedef typename NumTraits<Scalar>::Real RealScalar;
     if(m_flip)
-      return (i==0)? m_low : (m_high - (m_size1-i)*m_step);
+      return (i==0)? m_low : (m_high - RealScalar(m_size1-i)*m_step);
     else
-      return (i==m_size1)? m_high : (m_low + i*m_step);
+      return (i==m_size1)? m_high : (m_low + RealScalar(i)*m_step);
   }
 
   template<typename IndexType>
@@ -63,7 +63,7 @@ struct linspaced_op_impl<Scalar,Packet,/*IsInteger*/false>
     // [low, ..., low] + ( [step, ..., step] * ( [i, ..., i] + [0, ..., size] ) )
     if(m_flip)
     {
-      Packet pi = padd(pset1<Packet>(Scalar(i-m_size1)),m_interPacket);
+      Packet pi = plset<Packet>(Scalar(i-m_size1));
       Packet res = padd(pset1<Packet>(m_high), pmul(pset1<Packet>(m_step), pi));
       if(i==0)
         res = pinsertfirst(res, m_low);
@@ -71,7 +71,7 @@ struct linspaced_op_impl<Scalar,Packet,/*IsInteger*/false>
     }
     else
     {
-      Packet pi = padd(pset1<Packet>(Scalar(i)),m_interPacket);
+      Packet pi = plset<Packet>(Scalar(i));
       Packet res = padd(pset1<Packet>(m_low), pmul(pset1<Packet>(m_step), pi));
       if(i==m_size1-unpacket_traits<Packet>::size+1)
         res = pinsertlast(res, m_high);
@@ -83,7 +83,6 @@ struct linspaced_op_impl<Scalar,Packet,/*IsInteger*/false>
   const Scalar m_high;
   const Index m_size1;
   const Scalar m_step;
-  const Packet m_interPacket;
   const bool m_flip;
 };
 
@@ -93,8 +92,8 @@ struct linspaced_op_impl<Scalar,Packet,/*IsInteger*/true>
   linspaced_op_impl(const Scalar& low, const Scalar& high, Index num_steps) :
     m_low(low),
     m_multiplier((high-low)/convert_index<Scalar>(num_steps<=1 ? 1 : num_steps-1)),
-    m_divisor(convert_index<Scalar>(num_steps+high-low)/(high-low+1)),
-    m_use_divisor((high-low+1)<num_steps)
+    m_divisor(convert_index<Scalar>((high>=low?num_steps:-num_steps)+(high-low))/((numext::abs(high-low)+1)==0?1:(numext::abs(high-low)+1))),
+    m_use_divisor(num_steps>1 && (numext::abs(high-low)+1)<num_steps)
   {}
 
   template<typename IndexType>
@@ -173,6 +172,13 @@ template<typename Scalar, typename PacketType,typename IndexType>
 struct has_unary_operator<linspaced_op<Scalar,PacketType>,IndexType> { enum { value = 1}; };
 template<typename Scalar, typename PacketType,typename IndexType>
 struct has_binary_operator<linspaced_op<Scalar,PacketType>,IndexType> { enum { value = 0}; };
+
+template<typename Scalar,typename IndexType>
+struct has_nullary_operator<scalar_random_op<Scalar>,IndexType> { enum { value = 1}; };
+template<typename Scalar,typename IndexType>
+struct has_unary_operator<scalar_random_op<Scalar>,IndexType> { enum { value = 0}; };
+template<typename Scalar,typename IndexType>
+struct has_binary_operator<scalar_random_op<Scalar>,IndexType> { enum { value = 0}; };
 #endif
 
 } // end namespace internal

Eigen/src/Core/functors/StlFunctors.h

@@ -72,7 +72,7 @@ template<typename T>
 struct functor_traits<std::not_equal_to<T> >
 { enum { Cost = 1, PacketAccess = false }; };
 
-#if(__cplusplus < 201103L)
+#if (__cplusplus < 201103L) && (EIGEN_COMP_MSVC <= 1900)
 // std::binder* are deprecated since c++11 and will be removed in c++17
 template<typename T>
 struct functor_traits<std::binder2nd<T> >

Eigen/src/Core/products/GeneralBlockPanelKernel.h

@@ -972,7 +972,7 @@ void gebp_kernel<LhsScalar,RhsScalar,Index,DataMapper,mr,nr,ConjugateLhs,Conjuga
               EIGEN_ASM_COMMENT("begin step of gebp micro kernel 3pX4"); \
               EIGEN_ASM_COMMENT("Note: these asm comments work around bug 935!"); \
               internal::prefetch(blA+(3*K+16)*LhsProgress); \
-              if (EIGEN_ARCH_ARM) internal::prefetch(blB+(4*K+16)*RhsProgress); /* Bug 953 */ \
+              if (EIGEN_ARCH_ARM) { internal::prefetch(blB+(4*K+16)*RhsProgress); } /* Bug 953 */ \
               traits.loadLhs(&blA[(0+3*K)*LhsProgress], A0);  \
               traits.loadLhs(&blA[(1+3*K)*LhsProgress], A1);  \
               traits.loadLhs(&blA[(2+3*K)*LhsProgress], A2);  \

Eigen/src/Core/products/GeneralMatrixMatrix.h

@@ -83,8 +83,8 @@ static void run(Index rows, Index cols, Index depth,
   if(info)
   {
     // this is the parallel version!
-    Index tid = omp_get_thread_num();
-    Index threads = omp_get_num_threads();
+    int tid = omp_get_thread_num();
+    int threads = omp_get_num_threads();
 
     LhsScalar* blockA = blocking.blockA();
     eigen_internal_assert(blockA!=0);
@@ -116,9 +116,9 @@ static void run(Index rows, Index cols, Index depth,
       info[tid].sync = k;
 
       // Computes C_i += A' * B' per A'_i
-      for(Index shift=0; shift<threads; ++shift)
+      for(int shift=0; shift<threads; ++shift)
       {
-        Index i = (tid+shift)%threads;
+        int i = (tid+shift)%threads;
 
         // At this point we have to make sure that A'_i has been updated by the thread i,
         // we use testAndSetOrdered to mimic a volatile access.

Eigen/src/Core/products/GeneralMatrixMatrixTriangular.h

@@ -148,7 +148,7 @@ struct tribb_kernel
     ResMapper res(_res, resStride);
     gebp_kernel<LhsScalar, RhsScalar, Index, ResMapper, mr, nr, ConjLhs, ConjRhs> gebp_kernel;
 
-    Matrix<ResScalar,BlockSize,BlockSize,ColMajor> buffer;
+    Matrix<ResScalar,BlockSize,BlockSize,ColMajor> buffer((internal::constructor_without_unaligned_array_assert()));
 
     // let's process the block per panel of actual_mc x BlockSize,
     // again, each is split into three parts, etc.
@@ -199,7 +199,7 @@ struct general_product_to_triangular_selector;
 template<typename MatrixType, typename ProductType, int UpLo>
 struct general_product_to_triangular_selector<MatrixType,ProductType,UpLo,true>
 {
-  static void run(MatrixType& mat, const ProductType& prod, const typename MatrixType::Scalar& alpha)
+  static void run(MatrixType& mat, const ProductType& prod, const typename MatrixType::Scalar& alpha, bool beta)
   {
     typedef typename MatrixType::Scalar Scalar;
     
@@ -217,6 +217,9 @@ struct general_product_to_triangular_selector<MatrixType,ProductType,UpLo,true>
 
     Scalar actualAlpha = alpha * LhsBlasTraits::extractScalarFactor(prod.lhs().derived()) * RhsBlasTraits::extractScalarFactor(prod.rhs().derived());
 
+    if(!beta)
+      mat.template triangularView<UpLo>().setZero();
+
     enum {
       StorageOrder = (internal::traits<MatrixType>::Flags&RowMajorBit) ? RowMajor : ColMajor,
       UseLhsDirectly = _ActualLhs::InnerStrideAtCompileTime==1,
@@ -244,7 +247,7 @@ struct general_product_to_triangular_selector<MatrixType,ProductType,UpLo,true>
 template<typename MatrixType, typename ProductType, int UpLo>
 struct general_product_to_triangular_selector<MatrixType,ProductType,UpLo,false>
 {
-  static void run(MatrixType& mat, const ProductType& prod, const typename MatrixType::Scalar& alpha)
+  static void run(MatrixType& mat, const ProductType& prod, const typename MatrixType::Scalar& alpha, bool beta)
   {
     typedef typename internal::remove_all<typename ProductType::LhsNested>::type Lhs;
     typedef internal::blas_traits<Lhs> LhsBlasTraits;
@@ -260,13 +263,19 @@ struct general_product_to_triangular_selector<MatrixType,ProductType,UpLo,false>
 
     typename ProductType::Scalar actualAlpha = alpha * LhsBlasTraits::extractScalarFactor(prod.lhs().derived()) * RhsBlasTraits::extractScalarFactor(prod.rhs().derived());
 
+    if(!beta)
+      mat.template triangularView<UpLo>().setZero();
+
     enum {
       IsRowMajor = (internal::traits<MatrixType>::Flags&RowMajorBit) ? 1 : 0,
       LhsIsRowMajor = _ActualLhs::Flags&RowMajorBit ? 1 : 0,
-      RhsIsRowMajor = _ActualRhs::Flags&RowMajorBit ? 1 : 0
+      RhsIsRowMajor = _ActualRhs::Flags&RowMajorBit ? 1 : 0,
+      SkipDiag = (UpLo&(UnitDiag|ZeroDiag))!=0
     };
 
     Index size = mat.cols();
+    if(SkipDiag)
+      size--;
     Index depth = actualLhs.cols();
 
     typedef internal::gemm_blocking_space<IsRowMajor ? RowMajor : ColMajor,typename Lhs::Scalar,typename Rhs::Scalar,
@@ -277,20 +286,22 @@ struct general_product_to_triangular_selector<MatrixType,ProductType,UpLo,false>
     internal::general_matrix_matrix_triangular_product<Index,
       typename Lhs::Scalar, LhsIsRowMajor ? RowMajor : ColMajor, LhsBlasTraits::NeedToConjugate,
       typename Rhs::Scalar, RhsIsRowMajor ? RowMajor : ColMajor, RhsBlasTraits::NeedToConjugate,
-      IsRowMajor ? RowMajor : ColMajor, UpLo>
+      IsRowMajor ? RowMajor : ColMajor, UpLo&(Lower|Upper)>
       ::run(size, depth,
-            &actualLhs.coeffRef(0,0), actualLhs.outerStride(), &actualRhs.coeffRef(0,0), actualRhs.outerStride(),
-            mat.data(), mat.outerStride(), actualAlpha, blocking);
+            &actualLhs.coeffRef(SkipDiag&&(UpLo&Lower)==Lower ? 1 : 0,0), actualLhs.outerStride(),
+            &actualRhs.coeffRef(0,SkipDiag&&(UpLo&Upper)==Upper ? 1 : 0), actualRhs.outerStride(),
+            mat.data() + (SkipDiag ? (bool(IsRowMajor) != ((UpLo&Lower)==Lower) ? 1 : mat.outerStride() ) : 0), mat.outerStride(), actualAlpha, blocking);
   }
 };
 
 template<typename MatrixType, unsigned int UpLo>
 template<typename ProductType>
-TriangularView<MatrixType,UpLo>& TriangularViewImpl<MatrixType,UpLo,Dense>::_assignProduct(const ProductType& prod, const Scalar& alpha)
+TriangularView<MatrixType,UpLo>& TriangularViewImpl<MatrixType,UpLo,Dense>::_assignProduct(const ProductType& prod, const Scalar& alpha, bool beta)
 {
+  EIGEN_STATIC_ASSERT((UpLo&UnitDiag)==0, WRITING_TO_TRIANGULAR_PART_WITH_UNIT_DIAGONAL_IS_NOT_SUPPORTED);
   eigen_assert(derived().nestedExpression().rows() == prod.rows() && derived().cols() == prod.cols());
   
-  general_product_to_triangular_selector<MatrixType, ProductType, UpLo, internal::traits<ProductType>::InnerSize==1>::run(derived().nestedExpression().const_cast_derived(), prod, alpha);
+  general_product_to_triangular_selector<MatrixType, ProductType, UpLo, internal::traits<ProductType>::InnerSize==1>::run(derived().nestedExpression().const_cast_derived(), prod, alpha, beta);
   
   return derived();
 }

Eigen/src/Core/products/GeneralMatrixMatrixTriangular_BLAS.h

@@ -33,7 +33,7 @@
 #ifndef EIGEN_GENERAL_MATRIX_MATRIX_TRIANGULAR_BLAS_H
 #define EIGEN_GENERAL_MATRIX_MATRIX_TRIANGULAR_BLAS_H
 
-namespace Eigen { 
+namespace Eigen {
 
 namespace internal {
 
@@ -52,7 +52,7 @@ struct general_matrix_matrix_triangular_product<Index,Scalar,LhsStorageOrder,Con
   static EIGEN_STRONG_INLINE void run(Index size, Index depth,const Scalar* lhs, Index lhsStride, \
                           const Scalar* rhs, Index rhsStride, Scalar* res, Index resStride, Scalar alpha, level3_blocking<Scalar, Scalar>& blocking) \
   { \
-    if (lhs==rhs) { \
+    if ( lhs==rhs && ((UpLo&(Lower|Upper)==UpLo)) ) { \
       general_matrix_matrix_rankupdate<Index,Scalar,LhsStorageOrder,ConjugateLhs,ColMajor,UpLo> \
       ::run(size,depth,lhs,lhsStride,rhs,rhsStride,res,resStride,alpha,blocking); \
     } else { \
@@ -86,8 +86,8 @@ struct general_matrix_matrix_rankupdate<Index,EIGTYPE,AStorageOrder,ConjugateA,C
   /* typedef Matrix<EIGTYPE, Dynamic, Dynamic, RhsStorageOrder> MatrixRhs;*/ \
 \
    BlasIndex lda=convert_index<BlasIndex>(lhsStride), ldc=convert_index<BlasIndex>(resStride), n=convert_index<BlasIndex>(size), k=convert_index<BlasIndex>(depth); \
-   char uplo=(IsLower) ? 'L' : 'U', trans=(AStorageOrder==RowMajor) ? 'T':'N'; \
-   EIGTYPE beta; \
+   char uplo=((IsLower) ? 'L' : 'U'), trans=((AStorageOrder==RowMajor) ? 'T':'N'); \
+   EIGTYPE beta(1); \
    BLASFUNC(&uplo, &trans, &n, &k, &numext::real_ref(alpha), lhs, &lda, &numext::real_ref(beta), res, &ldc); \
   } \
 };
@@ -107,7 +107,7 @@ struct general_matrix_matrix_rankupdate<Index,EIGTYPE,AStorageOrder,ConjugateA,C
    typedef Matrix<EIGTYPE, Dynamic, Dynamic, AStorageOrder> MatrixType; \
 \
    BlasIndex lda=convert_index<BlasIndex>(lhsStride), ldc=convert_index<BlasIndex>(resStride), n=convert_index<BlasIndex>(size), k=convert_index<BlasIndex>(depth); \
-   char uplo=(IsLower) ? 'L' : 'U', trans=(AStorageOrder==RowMajor) ? 'C':'N'; \
+   char uplo=((IsLower) ? 'L' : 'U'), trans=((AStorageOrder==RowMajor) ? 'C':'N'); \
    RTYPE alpha_, beta_; \
    const EIGTYPE* a_ptr; \
 \

Eigen/src/Core/products/Parallelizer.h

@@ -75,7 +75,7 @@ template<typename Index> struct GemmParallelInfo
 {
   GemmParallelInfo() : sync(-1), users(0), lhs_start(0), lhs_length(0) {}
 
-  int volatile sync;
+  Index volatile sync;
   int volatile users;
 
   Index lhs_start;
@@ -104,13 +104,14 @@ void parallelize_gemm(const Functor& func, Index rows, Index cols, Index depth,
   // - the sizes are large enough
 
   // compute the maximal number of threads from the size of the product:
-  // FIXME this has to be fine tuned
+  // This first heuristic takes into account that the product kernel is fully optimized when working with nr columns at once.
   Index size = transpose ? rows : cols;
-  Index pb_max_threads = std::max<Index>(1,size / 32);
+  Index pb_max_threads = std::max<Index>(1,size / Functor::Traits::nr);
+
   // compute the maximal number of threads from the total amount of work:
   double work = static_cast<double>(rows) * static_cast<double>(cols) *
       static_cast<double>(depth);
-  double kMinTaskSize = 50000;  // Heuristic.
+  double kMinTaskSize = 50000;  // FIXME improve this heuristic.
   pb_max_threads = std::max<Index>(1, std::min<Index>(pb_max_threads, work / kMinTaskSize));
 
   // compute the number of threads we are going to use

Eigen/src/Core/products/SelfadjointMatrixVector.h

@@ -83,10 +83,10 @@ EIGEN_DONT_INLINE void selfadjoint_matrix_vector_product<Scalar,Index,StorageOrd
     Scalar t3(0);
     Packet ptmp3 = pset1<Packet>(t3);
 
-    size_t starti = FirstTriangular ? 0 : j+2;
-    size_t endi   = FirstTriangular ? j : size;
-    size_t alignedStart = (starti) + internal::first_default_aligned(&res[starti], endi-starti);
-    size_t alignedEnd = alignedStart + ((endi-alignedStart)/(PacketSize))*(PacketSize);
+    Index starti = FirstTriangular ? 0 : j+2;
+    Index endi   = FirstTriangular ? j : size;
+    Index alignedStart = (starti) + internal::first_default_aligned(&res[starti], endi-starti);
+    Index alignedEnd = alignedStart + ((endi-alignedStart)/(PacketSize))*(PacketSize);
 
     res[j]   += cjd.pmul(numext::real(A0[j]), t0);
     res[j+1] += cjd.pmul(numext::real(A1[j+1]), t1);
@@ -101,7 +101,7 @@ EIGEN_DONT_INLINE void selfadjoint_matrix_vector_product<Scalar,Index,StorageOrd
       t2 += cj1.pmul(A0[j+1], rhs[j+1]);
     }
 
-    for (size_t i=starti; i<alignedStart; ++i)
+    for (Index i=starti; i<alignedStart; ++i)
     {
       res[i] += cj0.pmul(A0[i], t0) + cj0.pmul(A1[i],t1);
       t2 += cj1.pmul(A0[i], rhs[i]);
@@ -113,7 +113,7 @@ EIGEN_DONT_INLINE void selfadjoint_matrix_vector_product<Scalar,Index,StorageOrd
     const Scalar* EIGEN_RESTRICT a1It  = A1  + alignedStart;
     const Scalar* EIGEN_RESTRICT rhsIt = rhs + alignedStart;
           Scalar* EIGEN_RESTRICT resIt = res + alignedStart;
-    for (size_t i=alignedStart; i<alignedEnd; i+=PacketSize)
+    for (Index i=alignedStart; i<alignedEnd; i+=PacketSize)
     {
       Packet A0i = ploadu<Packet>(a0It);  a0It  += PacketSize;
       Packet A1i = ploadu<Packet>(a1It);  a1It  += PacketSize;
@@ -125,7 +125,7 @@ EIGEN_DONT_INLINE void selfadjoint_matrix_vector_product<Scalar,Index,StorageOrd
       ptmp3 = pcj1.pmadd(A1i,  Bi, ptmp3);
       pstore(resIt,Xi); resIt += PacketSize;
     }
-    for (size_t i=alignedEnd; i<endi; i++)
+    for (Index i=alignedEnd; i<endi; i++)
     {
       res[i] += cj0.pmul(A0[i], t0) + cj0.pmul(A1[i],t1);
       t2 += cj1.pmul(A0[i], rhs[i]);

Eigen/src/Core/products/TriangularMatrixMatrix.h

@@ -137,7 +137,7 @@ EIGEN_DONT_INLINE void product_triangular_matrix_matrix<Scalar,Index,Mode,true,
     ei_declare_aligned_stack_constructed_variable(Scalar, blockA, sizeA, blocking.blockA());
     ei_declare_aligned_stack_constructed_variable(Scalar, blockB, sizeB, blocking.blockB());
 
-    Matrix<Scalar,SmallPanelWidth,SmallPanelWidth,LhsStorageOrder> triangularBuffer;
+    Matrix<Scalar,SmallPanelWidth,SmallPanelWidth,LhsStorageOrder> triangularBuffer((internal::constructor_without_unaligned_array_assert()));
     triangularBuffer.setZero();
     if((Mode&ZeroDiag)==ZeroDiag)
       triangularBuffer.diagonal().setZero();
@@ -284,7 +284,7 @@ EIGEN_DONT_INLINE void product_triangular_matrix_matrix<Scalar,Index,Mode,false,
     ei_declare_aligned_stack_constructed_variable(Scalar, blockA, sizeA, blocking.blockA());
     ei_declare_aligned_stack_constructed_variable(Scalar, blockB, sizeB, blocking.blockB());
 
-    Matrix<Scalar,SmallPanelWidth,SmallPanelWidth,RhsStorageOrder> triangularBuffer;
+    Matrix<Scalar,SmallPanelWidth,SmallPanelWidth,RhsStorageOrder> triangularBuffer((internal::constructor_without_unaligned_array_assert()));
     triangularBuffer.setZero();
     if((Mode&ZeroDiag)==ZeroDiag)
       triangularBuffer.diagonal().setZero();

Eigen/src/Core/products/TriangularSolverMatrix.h

@@ -183,7 +183,7 @@ EIGEN_DONT_INLINE void triangular_solve_matrix<Scalar,Index,OnTheLeft,Mode,Conju
     }
   }
 
-/* Optimized triangular solver with multiple left hand sides and the trinagular matrix on the right
+/* Optimized triangular solver with multiple left hand sides and the triangular matrix on the right
  */
 template <typename Scalar, typename Index, int Mode, bool Conjugate, int TriStorageOrder>
 struct triangular_solve_matrix<Scalar,Index,OnTheRight,Mode,Conjugate,TriStorageOrder,ColMajor>
@@ -202,6 +202,7 @@ EIGEN_DONT_INLINE void triangular_solve_matrix<Scalar,Index,OnTheRight,Mode,Conj
     level3_blocking<Scalar,Scalar>& blocking)
   {
     Index rows = otherSize;
+    typedef typename NumTraits<Scalar>::Real RealScalar;
 
     typedef blas_data_mapper<Scalar, Index, ColMajor> LhsMapper;
     typedef const_blas_data_mapper<Scalar, Index, TriStorageOrder> RhsMapper;
@@ -306,9 +307,9 @@ EIGEN_DONT_INLINE void triangular_solve_matrix<Scalar,Index,OnTheRight,Mode,Conj
               }
               if((Mode & UnitDiag)==0)
               {
-                Scalar b = conj(rhs(j,j));
+                Scalar inv_rjj = RealScalar(1)/conj(rhs(j,j));
                 for (Index i=0; i<actual_mc; ++i)
-                  r[i] /= b;
+                  r[i] *= inv_rjj;
               }
             }
 

Eigen/src/Core/util/Macros.h

@@ -13,7 +13,7 @@
 
 #define EIGEN_WORLD_VERSION 3
 #define EIGEN_MAJOR_VERSION 3
-#define EIGEN_MINOR_VERSION 0
+#define EIGEN_MINOR_VERSION 4
 
 #define EIGEN_VERSION_AT_LEAST(x,y,z) (EIGEN_WORLD_VERSION>x || (EIGEN_WORLD_VERSION>=x && \
                                       (EIGEN_MAJOR_VERSION>y || (EIGEN_MAJOR_VERSION>=y && \
@@ -80,8 +80,8 @@
 //  2015   14      1900
 //  "15"   15      1900
 
-/// \internal EIGEN_COMP_MSVC_STRICT set to 1 if the compiler is really Microsoft Visual C++ and not ,e.g., ICC
-#if EIGEN_COMP_MSVC && !(EIGEN_COMP_ICC)
+/// \internal EIGEN_COMP_MSVC_STRICT set to 1 if the compiler is really Microsoft Visual C++ and not ,e.g., ICC or clang-cl
+#if EIGEN_COMP_MSVC && !(EIGEN_COMP_ICC || EIGEN_COMP_LLVM || EIGEN_COMP_CLANG)
   #define EIGEN_COMP_MSVC_STRICT _MSC_VER
 #else
   #define EIGEN_COMP_MSVC_STRICT 0
@@ -356,7 +356,7 @@
 #define EIGEN_MAX_CPP_VER 99
 #endif
 
-#if EIGEN_MAX_CPP_VER>=11 && defined(__cplusplus) && (__cplusplus >= 201103L)
+#if EIGEN_MAX_CPP_VER>=11 && (defined(__cplusplus) && (__cplusplus >= 201103L) || EIGEN_COMP_MSVC >= 1900)
 #define EIGEN_HAS_CXX11 1
 #else
 #define EIGEN_HAS_CXX11 0
@@ -497,10 +497,11 @@
 // attribute to maximize inlining. This should only be used when really necessary: in particular,
 // it uses __attribute__((always_inline)) on GCC, which most of the time is useless and can severely harm compile times.
 // FIXME with the always_inline attribute,
-// gcc 3.4.x reports the following compilation error:
+// gcc 3.4.x and 4.1 reports the following compilation error:
 //   Eval.h:91: sorry, unimplemented: inlining failed in call to 'const Eigen::Eval<Derived> Eigen::MatrixBase<Scalar, Derived>::eval() const'
 //    : function body not available
-#if EIGEN_GNUC_AT_LEAST(4,0)
+//   See also bug 1367
+#if EIGEN_GNUC_AT_LEAST(4,2)
 #define EIGEN_ALWAYS_INLINE __attribute__((always_inline)) inline
 #else
 #define EIGEN_ALWAYS_INLINE EIGEN_STRONG_INLINE
@@ -811,7 +812,7 @@ namespace Eigen {
 // just an empty macro !
 #define EIGEN_EMPTY
 
-#if EIGEN_COMP_MSVC_STRICT && (EIGEN_COMP_MSVC < 1900 ||  __CUDACC_VER__) // for older MSVC versions, as well as 1900 && CUDA 8, using the base operator is sufficient (cf Bugs 1000, 1324)
+#if EIGEN_COMP_MSVC_STRICT && (EIGEN_COMP_MSVC < 1900 ||  defined(__CUDACC_VER__)) // for older MSVC versions, as well as 1900 && CUDA 8, using the base operator is sufficient (cf Bugs 1000, 1324)
   #define EIGEN_INHERIT_ASSIGNMENT_EQUAL_OPERATOR(Derived) \
     using Base::operator =;
 #elif EIGEN_COMP_CLANG // workaround clang bug (see http://forum.kde.org/viewtopic.php?f=74&t=102653)

Eigen/src/Core/util/Memory.h

@@ -150,7 +150,7 @@ EIGEN_DEVICE_FUNC inline void check_that_malloc_is_allowed()
 /** \internal Allocates \a size bytes. The returned pointer is guaranteed to have 16 or 32 bytes alignment depending on the requirements.
   * On allocation error, the returned pointer is null, and std::bad_alloc is thrown.
   */
-EIGEN_DEVICE_FUNC inline void* aligned_malloc(size_t size)
+EIGEN_DEVICE_FUNC inline void* aligned_malloc(std::size_t size)
 {
   check_that_malloc_is_allowed();
 
@@ -185,7 +185,7 @@ EIGEN_DEVICE_FUNC inline void aligned_free(void *ptr)
   * \brief Reallocates an aligned block of memory.
   * \throws std::bad_alloc on allocation failure
   */
-inline void* aligned_realloc(void *ptr, size_t new_size, size_t old_size)
+inline void* aligned_realloc(void *ptr, std::size_t new_size, std::size_t old_size)
 {
   EIGEN_UNUSED_VARIABLE(old_size);
 
@@ -209,12 +209,12 @@ inline void* aligned_realloc(void *ptr, size_t new_size, size_t old_size)
 /** \internal Allocates \a size bytes. If Align is true, then the returned ptr is 16-byte-aligned.
   * On allocation error, the returned pointer is null, and a std::bad_alloc is thrown.
   */
-template<bool Align> EIGEN_DEVICE_FUNC inline void* conditional_aligned_malloc(size_t size)
+template<bool Align> EIGEN_DEVICE_FUNC inline void* conditional_aligned_malloc(std::size_t size)
 {
   return aligned_malloc(size);
 }
 
-template<> EIGEN_DEVICE_FUNC inline void* conditional_aligned_malloc<false>(size_t size)
+template<> EIGEN_DEVICE_FUNC inline void* conditional_aligned_malloc<false>(std::size_t size)
 {
   check_that_malloc_is_allowed();
 
@@ -235,12 +235,12 @@ template<> EIGEN_DEVICE_FUNC inline void conditional_aligned_free<false>(void *p
   std::free(ptr);
 }
 
-template<bool Align> inline void* conditional_aligned_realloc(void* ptr, size_t new_size, size_t old_size)
+template<bool Align> inline void* conditional_aligned_realloc(void* ptr, std::size_t new_size, std::size_t old_size)
 {
   return aligned_realloc(ptr, new_size, old_size);
 }
 
-template<> inline void* conditional_aligned_realloc<false>(void* ptr, size_t new_size, size_t)
+template<> inline void* conditional_aligned_realloc<false>(void* ptr, std::size_t new_size, std::size_t)
 {
   return std::realloc(ptr, new_size);
 }
@@ -252,7 +252,7 @@ template<> inline void* conditional_aligned_realloc<false>(void* ptr, size_t new
 /** \internal Destructs the elements of an array.
   * The \a size parameters tells on how many objects to call the destructor of T.
   */
-template<typename T> EIGEN_DEVICE_FUNC inline void destruct_elements_of_array(T *ptr, size_t size)
+template<typename T> EIGEN_DEVICE_FUNC inline void destruct_elements_of_array(T *ptr, std::size_t size)
 {
   // always destruct an array starting from the end.
   if(ptr)
@@ -262,9 +262,9 @@ template<typename T> EIGEN_DEVICE_FUNC inline void destruct_elements_of_array(T
 /** \internal Constructs the elements of an array.
   * The \a size parameter tells on how many objects to call the constructor of T.
   */
-template<typename T> EIGEN_DEVICE_FUNC inline T* construct_elements_of_array(T *ptr, size_t size)
+template<typename T> EIGEN_DEVICE_FUNC inline T* construct_elements_of_array(T *ptr, std::size_t size)
 {
-  size_t i;
+  std::size_t i;
   EIGEN_TRY
   {
       for (i = 0; i < size; ++i) ::new (ptr + i) T;
@@ -283,9 +283,9 @@ template<typename T> EIGEN_DEVICE_FUNC inline T* construct_elements_of_array(T *
 *****************************************************************************/
 
 template<typename T>
-EIGEN_DEVICE_FUNC EIGEN_ALWAYS_INLINE void check_size_for_overflow(size_t size)
+EIGEN_DEVICE_FUNC EIGEN_ALWAYS_INLINE void check_size_for_overflow(std::size_t size)
 {
-  if(size > size_t(-1) / sizeof(T))
+  if(size > std::size_t(-1) / sizeof(T))
     throw_std_bad_alloc();
 }
 
@@ -293,7 +293,7 @@ EIGEN_DEVICE_FUNC EIGEN_ALWAYS_INLINE void check_size_for_overflow(size_t size)
   * On allocation error, the returned pointer is undefined, but a std::bad_alloc is thrown.
   * The default constructor of T is called.
   */
-template<typename T> EIGEN_DEVICE_FUNC inline T* aligned_new(size_t size)
+template<typename T> EIGEN_DEVICE_FUNC inline T* aligned_new(std::size_t size)
 {
   check_size_for_overflow<T>(size);
   T *result = reinterpret_cast<T*>(aligned_malloc(sizeof(T)*size));
@@ -309,7 +309,7 @@ template<typename T> EIGEN_DEVICE_FUNC inline T* aligned_new(size_t size)
   return result;
 }
 
-template<typename T, bool Align> EIGEN_DEVICE_FUNC inline T* conditional_aligned_new(size_t size)
+template<typename T, bool Align> EIGEN_DEVICE_FUNC inline T* conditional_aligned_new(std::size_t size)
 {
   check_size_for_overflow<T>(size);
   T *result = reinterpret_cast<T*>(conditional_aligned_malloc<Align>(sizeof(T)*size));
@@ -328,7 +328,7 @@ template<typename T, bool Align> EIGEN_DEVICE_FUNC inline T* conditional_aligned
 /** \internal Deletes objects constructed with aligned_new
   * The \a size parameters tells on how many objects to call the destructor of T.
   */
-template<typename T> EIGEN_DEVICE_FUNC inline void aligned_delete(T *ptr, size_t size)
+template<typename T> EIGEN_DEVICE_FUNC inline void aligned_delete(T *ptr, std::size_t size)
 {
   destruct_elements_of_array<T>(ptr, size);
   aligned_free(ptr);
@@ -337,13 +337,13 @@ template<typename T> EIGEN_DEVICE_FUNC inline void aligned_delete(T *ptr, size_t
 /** \internal Deletes objects constructed with conditional_aligned_new
   * The \a size parameters tells on how many objects to call the destructor of T.
   */
-template<typename T, bool Align> EIGEN_DEVICE_FUNC inline void conditional_aligned_delete(T *ptr, size_t size)
+template<typename T, bool Align> EIGEN_DEVICE_FUNC inline void conditional_aligned_delete(T *ptr, std::size_t size)
 {
   destruct_elements_of_array<T>(ptr, size);
   conditional_aligned_free<Align>(ptr);
 }
 
-template<typename T, bool Align> EIGEN_DEVICE_FUNC inline T* conditional_aligned_realloc_new(T* pts, size_t new_size, size_t old_size)
+template<typename T, bool Align> EIGEN_DEVICE_FUNC inline T* conditional_aligned_realloc_new(T* pts, std::size_t new_size, std::size_t old_size)
 {
   check_size_for_overflow<T>(new_size);
   check_size_for_overflow<T>(old_size);
@@ -366,7 +366,7 @@ template<typename T, bool Align> EIGEN_DEVICE_FUNC inline T* conditional_aligned
 }
 
 
-template<typename T, bool Align> EIGEN_DEVICE_FUNC inline T* conditional_aligned_new_auto(size_t size)
+template<typename T, bool Align> EIGEN_DEVICE_FUNC inline T* conditional_aligned_new_auto(std::size_t size)
 {
   if(size==0)
     return 0; // short-cut. Also fixes Bug 884
@@ -387,7 +387,7 @@ template<typename T, bool Align> EIGEN_DEVICE_FUNC inline T* conditional_aligned
   return result;
 }
 
-template<typename T, bool Align> inline T* conditional_aligned_realloc_new_auto(T* pts, size_t new_size, size_t old_size)
+template<typename T, bool Align> inline T* conditional_aligned_realloc_new_auto(T* pts, std::size_t new_size, std::size_t old_size)
 {
   check_size_for_overflow<T>(new_size);
   check_size_for_overflow<T>(old_size);
@@ -409,7 +409,7 @@ template<typename T, bool Align> inline T* conditional_aligned_realloc_new_auto(
   return result;
 }
 
-template<typename T, bool Align> EIGEN_DEVICE_FUNC inline void conditional_aligned_delete_auto(T *ptr, size_t size)
+template<typename T, bool Align> EIGEN_DEVICE_FUNC inline void conditional_aligned_delete_auto(T *ptr, std::size_t size)
 {
   if(NumTraits<T>::RequireInitialization)
     destruct_elements_of_array<T>(ptr, size);
@@ -523,7 +523,7 @@ template<typename T> struct smart_memmove_helper<T,true> {
 template<typename T> struct smart_memmove_helper<T,false> {
   static inline void run(const T* start, const T* end, T* target)
   { 
-    if (uintptr_t(target) < uintptr_t(start))
+    if (UIntPtr(target) < UIntPtr(start))
     {
       std::copy(start, end, target);
     }
@@ -561,7 +561,7 @@ template<typename T> class aligned_stack_memory_handler : noncopyable
      * In this case, the buffer elements will also be destructed when this handler will be destructed.
      * Finally, if \a dealloc is true, then the pointer \a ptr is freed.
      **/
-    aligned_stack_memory_handler(T* ptr, size_t size, bool dealloc)
+    aligned_stack_memory_handler(T* ptr, std::size_t size, bool dealloc)
       : m_ptr(ptr), m_size(size), m_deallocate(dealloc)
     {
       if(NumTraits<T>::RequireInitialization && m_ptr)
@@ -576,7 +576,7 @@ template<typename T> class aligned_stack_memory_handler : noncopyable
     }
   protected:
     T* m_ptr;
-    size_t m_size;
+    std::size_t m_size;
     bool m_deallocate;
 };
 
@@ -655,15 +655,15 @@ template<typename T> void swap(scoped_array<T> &a,scoped_array<T> &b)
 
 #if EIGEN_MAX_ALIGN_BYTES!=0
   #define EIGEN_MAKE_ALIGNED_OPERATOR_NEW_NOTHROW(NeedsToAlign) \
-      void* operator new(size_t size, const std::nothrow_t&) EIGEN_NO_THROW { \
+      void* operator new(std::size_t size, const std::nothrow_t&) EIGEN_NO_THROW { \
         EIGEN_TRY { return Eigen::internal::conditional_aligned_malloc<NeedsToAlign>(size); } \
         EIGEN_CATCH (...) { return 0; } \
       }
   #define EIGEN_MAKE_ALIGNED_OPERATOR_NEW_IF(NeedsToAlign) \
-      void *operator new(size_t size) { \
+      void *operator new(std::size_t size) { \
         return Eigen::internal::conditional_aligned_malloc<NeedsToAlign>(size); \
       } \
-      void *operator new[](size_t size) { \
+      void *operator new[](std::size_t size) { \
         return Eigen::internal::conditional_aligned_malloc<NeedsToAlign>(size); \
       } \
       void operator delete(void * ptr) EIGEN_NO_THROW { Eigen::internal::conditional_aligned_free<NeedsToAlign>(ptr); } \
@@ -673,8 +673,8 @@ template<typename T> void swap(scoped_array<T> &a,scoped_array<T> &b)
       /* in-place new and delete. since (at least afaik) there is no actual   */ \
       /* memory allocated we can safely let the default implementation handle */ \
       /* this particular case. */ \
-      static void *operator new(size_t size, void *ptr) { return ::operator new(size,ptr); } \
-      static void *operator new[](size_t size, void* ptr) { return ::operator new[](size,ptr); } \
+      static void *operator new(std::size_t size, void *ptr) { return ::operator new(size,ptr); } \
+      static void *operator new[](std::size_t size, void* ptr) { return ::operator new[](size,ptr); } \
       void operator delete(void * memory, void *ptr) EIGEN_NO_THROW { return ::operator delete(memory,ptr); } \
       void operator delete[](void * memory, void *ptr) EIGEN_NO_THROW { return ::operator delete[](memory,ptr); } \
       /* nothrow-new (returns zero instead of std::bad_alloc) */ \
@@ -713,7 +713,7 @@ template<class T>
 class aligned_allocator : public std::allocator<T>
 {
 public:
-  typedef size_t          size_type;
+  typedef std::size_t     size_type;
   typedef std::ptrdiff_t  difference_type;
   typedef T*              pointer;
   typedef const T*        const_pointer;

Eigen/src/Core/util/Meta.h

@@ -381,12 +381,12 @@ struct has_ReturnType
   enum { value = sizeof(testFunctor<T>(0)) == sizeof(meta_yes) };
 };
 
-template<typename T> const T& return_ref();
+template<typename T> const T* return_ptr();
 
 template <typename T, typename IndexType=Index>
 struct has_nullary_operator
 {
-  template <typename C> static meta_yes testFunctor(C const *,typename enable_if<(sizeof(return_ref<C>().operator()())>0)>::type * = 0);
+  template <typename C> static meta_yes testFunctor(C const *,typename enable_if<(sizeof(return_ptr<C>()->operator()())>0)>::type * = 0);
   static meta_no testFunctor(...);
 
   enum { value = sizeof(testFunctor(static_cast<T*>(0))) == sizeof(meta_yes) };
@@ -395,7 +395,7 @@ struct has_nullary_operator
 template <typename T, typename IndexType=Index>
 struct has_unary_operator
 {
-  template <typename C> static meta_yes testFunctor(C const *,typename enable_if<(sizeof(return_ref<C>().operator()(IndexType(0)))>0)>::type * = 0);
+  template <typename C> static meta_yes testFunctor(C const *,typename enable_if<(sizeof(return_ptr<C>()->operator()(IndexType(0)))>0)>::type * = 0);
   static meta_no testFunctor(...);
 
   enum { value = sizeof(testFunctor(static_cast<T*>(0))) == sizeof(meta_yes) };
@@ -404,7 +404,7 @@ struct has_unary_operator
 template <typename T, typename IndexType=Index>
 struct has_binary_operator
 {
-  template <typename C> static meta_yes testFunctor(C const *,typename enable_if<(sizeof(return_ref<C>().operator()(IndexType(0),IndexType(0)))>0)>::type * = 0);
+  template <typename C> static meta_yes testFunctor(C const *,typename enable_if<(sizeof(return_ptr<C>()->operator()(IndexType(0),IndexType(0)))>0)>::type * = 0);
   static meta_no testFunctor(...);
 
   enum { value = sizeof(testFunctor(static_cast<T*>(0))) == sizeof(meta_yes) };

Eigen/src/Core/util/XprHelper.h

@@ -445,15 +445,11 @@ template<typename T, int n, typename PlainObject = typename plain_object_eval<T>
                                                   //      Another solution could be to count the number of temps?
     NAsInteger = n == Dynamic ? HugeCost : n,
     CostEval   = (NAsInteger+1) * ScalarReadCost + CoeffReadCost,
-    CostNoEval = NAsInteger * CoeffReadCost
+    CostNoEval = NAsInteger * CoeffReadCost,
+    Evaluate = (int(evaluator<T>::Flags) & EvalBeforeNestingBit) || (int(CostEval) < int(CostNoEval))
   };
 
-  typedef typename conditional<
-        ( (int(evaluator<T>::Flags) & EvalBeforeNestingBit) ||
-          (int(CostEval) < int(CostNoEval)) ),
-        PlainObject,
-        typename ref_selector<T>::type
-  >::type type;
+  typedef typename conditional<Evaluate, PlainObject, typename ref_selector<T>::type>::type type;
 };
 
 template<typename T>
@@ -536,6 +532,15 @@ template <typename B, typename Functor>                   struct cwise_promote_s
 template <typename Functor>                               struct cwise_promote_storage_type<Sparse,Dense,Functor>                             { typedef Sparse ret; };
 template <typename Functor>                               struct cwise_promote_storage_type<Dense,Sparse,Functor>                             { typedef Sparse ret; };
 
+template <typename LhsKind, typename RhsKind, int LhsOrder, int RhsOrder> struct cwise_promote_storage_order {
+  enum { value = LhsOrder };
+};
+
+template <typename LhsKind, int LhsOrder, int RhsOrder>   struct cwise_promote_storage_order<LhsKind,Sparse,LhsOrder,RhsOrder>                { enum { value = RhsOrder }; };
+template <typename RhsKind, int LhsOrder, int RhsOrder>   struct cwise_promote_storage_order<Sparse,RhsKind,LhsOrder,RhsOrder>                { enum { value = LhsOrder }; };
+template <int Order>                                      struct cwise_promote_storage_order<Sparse,Sparse,Order,Order>                       { enum { value = Order }; };
+
+
 /** \internal Specify the "storage kind" of multiplying an expression of kind A with kind B.
   * The template parameter ProductTag permits to specialize the resulting storage kind wrt to
   * some compile-time properties of the product: GemmProduct, GemvProduct, OuterProduct, InnerProduct.
@@ -633,7 +638,7 @@ struct plain_constant_type
 template<typename ExpressionType>
 struct is_lvalue
 {
-  enum { value = !bool(is_const<ExpressionType>::value) &&
+  enum { value = (!bool(is_const<ExpressionType>::value)) &&
                  bool(traits<ExpressionType>::Flags & LvalueBit) };
 };
 

Eigen/src/Eigenvalues/ComplexEigenSolver.h

@@ -250,7 +250,7 @@ template<typename _MatrixType> class ComplexEigenSolver
     EigenvectorType m_matX;
 
   private:
-    void doComputeEigenvectors(const RealScalar& matrixnorm);
+    void doComputeEigenvectors(RealScalar matrixnorm);
     void sortEigenvalues(bool computeEigenvectors);
 };
 
@@ -284,10 +284,12 @@ ComplexEigenSolver<MatrixType>::compute(const EigenBase<InputType>& matrix, bool
 
 
 template<typename MatrixType>
-void ComplexEigenSolver<MatrixType>::doComputeEigenvectors(const RealScalar& matrixnorm)
+void ComplexEigenSolver<MatrixType>::doComputeEigenvectors(RealScalar matrixnorm)
 {
   const Index n = m_eivalues.size();
 
+  matrixnorm = numext::maxi(matrixnorm,(std::numeric_limits<RealScalar>::min)());
+
   // Compute X such that T = X D X^(-1), where D is the diagonal of T.
   // The matrix X is unit triangular.
   m_matX = EigenvectorType::Zero(n, n);

Eigen/src/Eigenvalues/RealSchur.h

@@ -248,12 +248,24 @@ template<typename MatrixType>
 template<typename InputType>
 RealSchur<MatrixType>& RealSchur<MatrixType>::compute(const EigenBase<InputType>& matrix, bool computeU)
 {
+  const Scalar considerAsZero = (std::numeric_limits<Scalar>::min)();
+
   eigen_assert(matrix.cols() == matrix.rows());
   Index maxIters = m_maxIters;
   if (maxIters == -1)
     maxIters = m_maxIterationsPerRow * matrix.rows();
 
   Scalar scale = matrix.derived().cwiseAbs().maxCoeff();
+  if(scale<considerAsZero)
+  {
+    m_matT.setZero(matrix.rows(),matrix.cols());
+    if(computeU)
+      m_matU.setIdentity(matrix.rows(),matrix.cols());
+    m_info = Success;
+    m_isInitialized = true;
+    m_matUisUptodate = computeU;
+    return *this;
+  }
 
   // Step 1. Reduce to Hessenberg form
   m_hess.compute(matrix.derived()/scale);

Eigen/src/Eigenvalues/SelfAdjointEigenSolver.h

@@ -414,7 +414,8 @@ SelfAdjointEigenSolver<MatrixType>& SelfAdjointEigenSolver<MatrixType>
 
   if(n==1)
   {
-    m_eivalues.coeffRef(0,0) = numext::real(matrix.diagonal()[0]);
+    m_eivec = matrix;
+    m_eivalues.coeffRef(0,0) = numext::real(m_eivec.coeff(0,0));
     if(computeEigenvectors)
       m_eivec.setOnes(n,n);
     m_info = Success;

Eigen/src/Geometry/Hyperplane.h

@@ -217,7 +217,10 @@ public:
   EIGEN_DEVICE_FUNC inline Hyperplane& transform(const MatrixBase<XprType>& mat, TransformTraits traits = Affine)
   {
     if (traits==Affine)
+    {
       normal() = mat.inverse().transpose() * normal();
+      m_coeffs /= normal().norm();
+    }
     else if (traits==Isometry)
       normal() = mat * normal();
     else

Eigen/src/Geometry/Quaternion.h

@@ -423,7 +423,7 @@ typedef Map<Quaternion<double>, Aligned>  QuaternionMapAlignedd;
 // Generic Quaternion * Quaternion product
 // This product can be specialized for a given architecture via the Arch template argument.
 namespace internal {
-template<int Arch, class Derived1, class Derived2, typename Scalar, int _Options> struct quat_product
+template<int Arch, class Derived1, class Derived2, typename Scalar> struct quat_product
 {
   EIGEN_DEVICE_FUNC static EIGEN_STRONG_INLINE Quaternion<Scalar> run(const QuaternionBase<Derived1>& a, const QuaternionBase<Derived2>& b){
     return Quaternion<Scalar>
@@ -446,8 +446,7 @@ QuaternionBase<Derived>::operator* (const QuaternionBase<OtherDerived>& other) c
   EIGEN_STATIC_ASSERT((internal::is_same<typename Derived::Scalar, typename OtherDerived::Scalar>::value),
    YOU_MIXED_DIFFERENT_NUMERIC_TYPES__YOU_NEED_TO_USE_THE_CAST_METHOD_OF_MATRIXBASE_TO_CAST_NUMERIC_TYPES_EXPLICITLY)
   return internal::quat_product<Architecture::Target, Derived, OtherDerived,
-                         typename internal::traits<Derived>::Scalar,
-                         EIGEN_PLAIN_ENUM_MIN(internal::traits<Derived>::Alignment, internal::traits<OtherDerived>::Alignment)>::run(*this, other);
+                         typename internal::traits<Derived>::Scalar>::run(*this, other);
 }
 
 /** \sa operator*(Quaternion) */
@@ -672,7 +671,7 @@ EIGEN_DEVICE_FUNC inline Quaternion<typename internal::traits<Derived>::Scalar>
 
 // Generic conjugate of a Quaternion
 namespace internal {
-template<int Arch, class Derived, typename Scalar, int _Options> struct quat_conj
+template<int Arch, class Derived, typename Scalar> struct quat_conj
 {
   EIGEN_DEVICE_FUNC static EIGEN_STRONG_INLINE Quaternion<Scalar> run(const QuaternionBase<Derived>& q){
     return Quaternion<Scalar>(q.w(),-q.x(),-q.y(),-q.z());
@@ -691,8 +690,7 @@ EIGEN_DEVICE_FUNC inline Quaternion<typename internal::traits<Derived>::Scalar>
 QuaternionBase<Derived>::conjugate() const
 {
   return internal::quat_conj<Architecture::Target, Derived,
-                         typename internal::traits<Derived>::Scalar,
-                         internal::traits<Derived>::Alignment>::run(*this);
+                         typename internal::traits<Derived>::Scalar>::run(*this);
                          
 }
 

Eigen/src/Geometry/arch/Geometry_SSE.h

@@ -16,17 +16,23 @@ namespace Eigen {
 namespace internal {
 
 template<class Derived, class OtherDerived>
-struct quat_product<Architecture::SSE, Derived, OtherDerived, float, Aligned16>
+struct quat_product<Architecture::SSE, Derived, OtherDerived, float>
 {
+  enum {
+    AAlignment = traits<Derived>::Alignment,
+    BAlignment = traits<OtherDerived>::Alignment,
+    ResAlignment = traits<Quaternion<float> >::Alignment
+  };
   static inline Quaternion<float> run(const QuaternionBase<Derived>& _a, const QuaternionBase<OtherDerived>& _b)
   {
     Quaternion<float> res;
     const __m128 mask = _mm_setr_ps(0.f,0.f,0.f,-0.f);
-    __m128 a = _a.coeffs().template packet<Aligned16>(0);
-    __m128 b = _b.coeffs().template packet<Aligned16>(0);
+    __m128 a = _a.coeffs().template packet<AAlignment>(0);
+    __m128 b = _b.coeffs().template packet<BAlignment>(0);
     __m128 s1 = _mm_mul_ps(vec4f_swizzle1(a,1,2,0,2),vec4f_swizzle1(b,2,0,1,2));
     __m128 s2 = _mm_mul_ps(vec4f_swizzle1(a,3,3,3,1),vec4f_swizzle1(b,0,1,2,1));
-    pstore(&res.x(),
+    pstoret<float,Packet4f,ResAlignment>(
+              &res.x(),
               _mm_add_ps(_mm_sub_ps(_mm_mul_ps(a,vec4f_swizzle1(b,3,3,3,3)),
                                     _mm_mul_ps(vec4f_swizzle1(a,2,0,1,0),
                                                vec4f_swizzle1(b,1,2,0,0))),
@@ -36,14 +42,17 @@ struct quat_product<Architecture::SSE, Derived, OtherDerived, float, Aligned16>
   }
 };
 
-template<class Derived, int Alignment>
-struct quat_conj<Architecture::SSE, Derived, float, Alignment>
+template<class Derived>
+struct quat_conj<Architecture::SSE, Derived, float>
 {
+  enum {
+    ResAlignment = traits<Quaternion<float> >::Alignment
+  };
   static inline Quaternion<float> run(const QuaternionBase<Derived>& q)
   {
     Quaternion<float> res;
     const __m128 mask = _mm_setr_ps(-0.f,-0.f,-0.f,0.f);
-    pstore(&res.x(), _mm_xor_ps(mask, q.coeffs().template packet<Alignment>(0)));
+    pstoret<float,Packet4f,ResAlignment>(&res.x(), _mm_xor_ps(mask, q.coeffs().template packet<traits<Derived>::Alignment>(0)));
     return res;
   }
 };
@@ -52,6 +61,9 @@ struct quat_conj<Architecture::SSE, Derived, float, Alignment>
 template<typename VectorLhs,typename VectorRhs>
 struct cross3_impl<Architecture::SSE,VectorLhs,VectorRhs,float,true>
 {
+  enum {
+    ResAlignment = traits<typename plain_matrix_type<VectorLhs>::type>::Alignment
+  };
   static inline typename plain_matrix_type<VectorLhs>::type
   run(const VectorLhs& lhs, const VectorRhs& rhs)
   {
@@ -60,7 +72,7 @@ struct cross3_impl<Architecture::SSE,VectorLhs,VectorRhs,float,true>
     __m128 mul1=_mm_mul_ps(vec4f_swizzle1(a,1,2,0,3),vec4f_swizzle1(b,2,0,1,3));
     __m128 mul2=_mm_mul_ps(vec4f_swizzle1(a,2,0,1,3),vec4f_swizzle1(b,1,2,0,3));
     typename plain_matrix_type<VectorLhs>::type res;
-    pstore(&res.x(),_mm_sub_ps(mul1,mul2));
+    pstoret<float,Packet4f,ResAlignment>(&res.x(),_mm_sub_ps(mul1,mul2));
     return res;
   }
 };
@@ -68,9 +80,14 @@ struct cross3_impl<Architecture::SSE,VectorLhs,VectorRhs,float,true>
 
 
 
-template<class Derived, class OtherDerived, int Alignment>
-struct quat_product<Architecture::SSE, Derived, OtherDerived, double, Alignment>
+template<class Derived, class OtherDerived>
+struct quat_product<Architecture::SSE, Derived, OtherDerived, double>
 {
+  enum {
+    BAlignment = traits<OtherDerived>::Alignment,
+    ResAlignment = traits<Quaternion<double> >::Alignment
+  };
+
   static inline Quaternion<double> run(const QuaternionBase<Derived>& _a, const QuaternionBase<OtherDerived>& _b)
   {
   const Packet2d mask = _mm_castsi128_pd(_mm_set_epi32(0x0,0x0,0x80000000,0x0));
@@ -78,8 +95,8 @@ struct quat_product<Architecture::SSE, Derived, OtherDerived, double, Alignment>
   Quaternion<double> res;
 
   const double* a = _a.coeffs().data();
-  Packet2d b_xy = _b.coeffs().template packet<Alignment>(0);
-  Packet2d b_zw = _b.coeffs().template packet<Alignment>(2);
+  Packet2d b_xy = _b.coeffs().template packet<BAlignment>(0);
+  Packet2d b_zw = _b.coeffs().template packet<BAlignment>(2);
   Packet2d a_xx = pset1<Packet2d>(a[0]);
   Packet2d a_yy = pset1<Packet2d>(a[1]);
   Packet2d a_zz = pset1<Packet2d>(a[2]);
@@ -97,9 +114,9 @@ struct quat_product<Architecture::SSE, Derived, OtherDerived, double, Alignment>
   t2 = psub(pmul(a_zz, b_xy), pmul(a_xx, b_zw));
 #ifdef EIGEN_VECTORIZE_SSE3
   EIGEN_UNUSED_VARIABLE(mask)
-  pstore(&res.x(), _mm_addsub_pd(t1, preverse(t2)));
+  pstoret<double,Packet2d,ResAlignment>(&res.x(), _mm_addsub_pd(t1, preverse(t2)));
 #else
-  pstore(&res.x(), padd(t1, pxor(mask,preverse(t2))));
+  pstoret<double,Packet2d,ResAlignment>(&res.x(), padd(t1, pxor(mask,preverse(t2))));
 #endif
   
   /*
@@ -111,25 +128,28 @@ struct quat_product<Architecture::SSE, Derived, OtherDerived, double, Alignment>
   t2 = padd(pmul(a_zz, b_zw), pmul(a_xx, b_xy));
 #ifdef EIGEN_VECTORIZE_SSE3
   EIGEN_UNUSED_VARIABLE(mask)
-  pstore(&res.z(), preverse(_mm_addsub_pd(preverse(t1), t2)));
+  pstoret<double,Packet2d,ResAlignment>(&res.z(), preverse(_mm_addsub_pd(preverse(t1), t2)));
 #else
-  pstore(&res.z(), psub(t1, pxor(mask,preverse(t2))));
+  pstoret<double,Packet2d,ResAlignment>(&res.z(), psub(t1, pxor(mask,preverse(t2))));
 #endif
 
   return res;
 }
 };
 
-template<class Derived, int Alignment>
-struct quat_conj<Architecture::SSE, Derived, double, Alignment>
+template<class Derived>
+struct quat_conj<Architecture::SSE, Derived, double>
 {
+  enum {
+    ResAlignment = traits<Quaternion<double> >::Alignment
+  };
   static inline Quaternion<double> run(const QuaternionBase<Derived>& q)
   {
     Quaternion<double> res;
     const __m128d mask0 = _mm_setr_pd(-0.,-0.);
     const __m128d mask2 = _mm_setr_pd(-0.,0.);
-    pstore(&res.x(), _mm_xor_pd(mask0, q.coeffs().template packet<Alignment>(0)));
-    pstore(&res.z(), _mm_xor_pd(mask2, q.coeffs().template packet<Alignment>(2)));
+    pstoret<double,Packet2d,ResAlignment>(&res.x(), _mm_xor_pd(mask0, q.coeffs().template packet<traits<Derived>::Alignment>(0)));
+    pstoret<double,Packet2d,ResAlignment>(&res.z(), _mm_xor_pd(mask2, q.coeffs().template packet<traits<Derived>::Alignment>(2)));
     return res;
   }
 };

Eigen/src/Householder/BlockHouseholder.h

@@ -87,7 +87,8 @@ void apply_block_householder_on_the_left(MatrixType& mat, const VectorsType& vec
   const TriangularView<const VectorsType, UnitLower> V(vectors);
 
   // A -= V T V^* A
-  Matrix<typename MatrixType::Scalar,VectorsType::ColsAtCompileTime,MatrixType::ColsAtCompileTime,0,
+  Matrix<typename MatrixType::Scalar,VectorsType::ColsAtCompileTime,MatrixType::ColsAtCompileTime,
+         (VectorsType::MaxColsAtCompileTime==1 && MatrixType::MaxColsAtCompileTime!=1)?RowMajor:ColMajor,
          VectorsType::MaxColsAtCompileTime,MatrixType::MaxColsAtCompileTime> tmp = V.adjoint() * mat;
   // FIXME add .noalias() once the triangular product can work inplace
   if(forward) tmp = T.template triangularView<Upper>()           * tmp;

Eigen/src/IterativeLinearSolvers/BasicPreconditioners.h

@@ -152,13 +152,28 @@ class LeastSquareDiagonalPreconditioner : public DiagonalPreconditioner<_Scalar>
     {
       // Compute the inverse squared-norm of each column of mat
       m_invdiag.resize(mat.cols());
-      for(Index j=0; j<mat.outerSize(); ++j)
+      if(MatType::IsRowMajor)
       {
-        RealScalar sum = mat.innerVector(j).squaredNorm();
-        if(sum>0)
-          m_invdiag(j) = RealScalar(1)/sum;
-        else
-          m_invdiag(j) = RealScalar(1);
+        m_invdiag.setZero();
+        for(Index j=0; j<mat.outerSize(); ++j)
+        {
+          for(typename MatType::InnerIterator it(mat,j); it; ++it)
+            m_invdiag(it.index()) += numext::abs2(it.value());
+        }
+        for(Index j=0; j<mat.cols(); ++j)
+          if(numext::real(m_invdiag(j))>RealScalar(0))
+            m_invdiag(j) = RealScalar(1)/numext::real(m_invdiag(j));
+      }
+      else
+      {
+        for(Index j=0; j<mat.outerSize(); ++j)
+        {
+          RealScalar sum = mat.innerVector(j).squaredNorm();
+          if(sum>RealScalar(0))
+            m_invdiag(j) = RealScalar(1)/sum;
+          else
+            m_invdiag(j) = RealScalar(1);
+        }
       }
       Base::m_isInitialized = true;
       return *this;

Eigen/src/Jacobi/Jacobi.h

@@ -302,8 +302,12 @@ template<typename VectorX, typename VectorY, typename OtherScalar>
 void /*EIGEN_DONT_INLINE*/ apply_rotation_in_the_plane(DenseBase<VectorX>& xpr_x, DenseBase<VectorY>& xpr_y, const JacobiRotation<OtherScalar>& j)
 {
   typedef typename VectorX::Scalar Scalar;
-  enum { PacketSize = packet_traits<Scalar>::size };
+  enum {
+    PacketSize = packet_traits<Scalar>::size,
+    OtherPacketSize = packet_traits<OtherScalar>::size
+  };
   typedef typename packet_traits<Scalar>::type Packet;
+  typedef typename packet_traits<OtherScalar>::type OtherPacket;
   eigen_assert(xpr_x.size() == xpr_y.size());
   Index size = xpr_x.size();
   Index incrx = xpr_x.derived().innerStride();
@@ -321,6 +325,7 @@ void /*EIGEN_DONT_INLINE*/ apply_rotation_in_the_plane(DenseBase<VectorX>& xpr_x
 
   if(VectorX::SizeAtCompileTime == Dynamic &&
     (VectorX::Flags & VectorY::Flags & PacketAccessBit) &&
+    (PacketSize == OtherPacketSize) &&
     ((incrx==1 && incry==1) || PacketSize == 1))
   {
     // both vectors are sequentially stored in memory => vectorization
@@ -329,9 +334,10 @@ void /*EIGEN_DONT_INLINE*/ apply_rotation_in_the_plane(DenseBase<VectorX>& xpr_x
     Index alignedStart = internal::first_default_aligned(y, size);
     Index alignedEnd = alignedStart + ((size-alignedStart)/PacketSize)*PacketSize;
 
-    const Packet pc = pset1<Packet>(c);
-    const Packet ps = pset1<Packet>(s);
-    conj_helper<Packet,Packet,NumTraits<Scalar>::IsComplex,false> pcj;
+    const OtherPacket pc = pset1<OtherPacket>(c);
+    const OtherPacket ps = pset1<OtherPacket>(s);
+    conj_helper<OtherPacket,Packet,NumTraits<OtherScalar>::IsComplex,false> pcj;
+    conj_helper<OtherPacket,Packet,false,false> pm;
 
     for(Index i=0; i<alignedStart; ++i)
     {
@@ -350,8 +356,8 @@ void /*EIGEN_DONT_INLINE*/ apply_rotation_in_the_plane(DenseBase<VectorX>& xpr_x
       {
         Packet xi = pload<Packet>(px);
         Packet yi = pload<Packet>(py);
-        pstore(px, padd(pmul(pc,xi),pcj.pmul(ps,yi)));
-        pstore(py, psub(pcj.pmul(pc,yi),pmul(ps,xi)));
+        pstore(px, padd(pm.pmul(pc,xi),pcj.pmul(ps,yi)));
+        pstore(py, psub(pcj.pmul(pc,yi),pm.pmul(ps,xi)));
         px += PacketSize;
         py += PacketSize;
       }
@@ -365,10 +371,10 @@ void /*EIGEN_DONT_INLINE*/ apply_rotation_in_the_plane(DenseBase<VectorX>& xpr_x
         Packet xi1  = ploadu<Packet>(px+PacketSize);
         Packet yi   = pload <Packet>(py);
         Packet yi1  = pload <Packet>(py+PacketSize);
-        pstoreu(px, padd(pmul(pc,xi),pcj.pmul(ps,yi)));
-        pstoreu(px+PacketSize, padd(pmul(pc,xi1),pcj.pmul(ps,yi1)));
-        pstore (py, psub(pcj.pmul(pc,yi),pmul(ps,xi)));
-        pstore (py+PacketSize, psub(pcj.pmul(pc,yi1),pmul(ps,xi1)));
+        pstoreu(px, padd(pm.pmul(pc,xi),pcj.pmul(ps,yi)));
+        pstoreu(px+PacketSize, padd(pm.pmul(pc,xi1),pcj.pmul(ps,yi1)));
+        pstore (py, psub(pcj.pmul(pc,yi),pm.pmul(ps,xi)));
+        pstore (py+PacketSize, psub(pcj.pmul(pc,yi1),pm.pmul(ps,xi1)));
         px += Peeling*PacketSize;
         py += Peeling*PacketSize;
       }
@@ -376,8 +382,8 @@ void /*EIGEN_DONT_INLINE*/ apply_rotation_in_the_plane(DenseBase<VectorX>& xpr_x
       {
         Packet xi = ploadu<Packet>(x+peelingEnd);
         Packet yi = pload <Packet>(y+peelingEnd);
-        pstoreu(x+peelingEnd, padd(pmul(pc,xi),pcj.pmul(ps,yi)));
-        pstore (y+peelingEnd, psub(pcj.pmul(pc,yi),pmul(ps,xi)));
+        pstoreu(x+peelingEnd, padd(pm.pmul(pc,xi),pcj.pmul(ps,yi)));
+        pstore (y+peelingEnd, psub(pcj.pmul(pc,yi),pm.pmul(ps,xi)));
       }
     }
 
@@ -393,19 +399,21 @@ void /*EIGEN_DONT_INLINE*/ apply_rotation_in_the_plane(DenseBase<VectorX>& xpr_x
   /*** fixed-size vectorized path ***/
   else if(VectorX::SizeAtCompileTime != Dynamic &&
           (VectorX::Flags & VectorY::Flags & PacketAccessBit) &&
+          (PacketSize == OtherPacketSize) &&
           (EIGEN_PLAIN_ENUM_MIN(evaluator<VectorX>::Alignment, evaluator<VectorY>::Alignment)>0)) // FIXME should be compared to the required alignment
   {
-    const Packet pc = pset1<Packet>(c);
-    const Packet ps = pset1<Packet>(s);
-    conj_helper<Packet,Packet,NumTraits<Scalar>::IsComplex,false> pcj;
+    const OtherPacket pc = pset1<OtherPacket>(c);
+    const OtherPacket ps = pset1<OtherPacket>(s);
+    conj_helper<OtherPacket,Packet,NumTraits<OtherPacket>::IsComplex,false> pcj;
+    conj_helper<OtherPacket,Packet,false,false> pm;
     Scalar* EIGEN_RESTRICT px = x;
     Scalar* EIGEN_RESTRICT py = y;
     for(Index i=0; i<size; i+=PacketSize)
     {
       Packet xi = pload<Packet>(px);
       Packet yi = pload<Packet>(py);
-      pstore(px, padd(pmul(pc,xi),pcj.pmul(ps,yi)));
-      pstore(py, psub(pcj.pmul(pc,yi),pmul(ps,xi)));
+      pstore(px, padd(pm.pmul(pc,xi),pcj.pmul(ps,yi)));
+      pstore(py, psub(pcj.pmul(pc,yi),pm.pmul(ps,xi)));
       px += PacketSize;
       py += PacketSize;
     }

Eigen/src/OrderingMethods/Eigen_Colamd.h

@@ -1004,7 +1004,7 @@ static IndexType find_ordering /* return the number of garbage collections */
     COLAMD_ASSERT (head [min_score] >= COLAMD_EMPTY) ;
 
     /* get pivot column from head of minimum degree list */
-    while (head [min_score] == COLAMD_EMPTY && min_score < n_col)
+    while (min_score < n_col && head [min_score] == COLAMD_EMPTY)
     {
       min_score++ ;
     }

Eigen/src/QR/ColPivHouseholderQR.h

@@ -506,8 +506,8 @@ void ColPivHouseholderQR<MatrixType>::computeInPlace()
     m_colNormsUpdated.coeffRef(k) = m_colNormsDirect.coeffRef(k);
   }
 
-  RealScalar threshold_helper =  numext::abs2(m_colNormsUpdated.maxCoeff() * NumTraits<Scalar>::epsilon()) / RealScalar(rows);
-  RealScalar norm_downdate_threshold = numext::sqrt(NumTraits<Scalar>::epsilon());
+  RealScalar threshold_helper =  numext::abs2<RealScalar>(m_colNormsUpdated.maxCoeff() * NumTraits<RealScalar>::epsilon()) / RealScalar(rows);
+  RealScalar norm_downdate_threshold = numext::sqrt(NumTraits<RealScalar>::epsilon());
 
   m_nonzero_pivots = size; // the generic case is that in which all pivots are nonzero (invertible case)
   m_maxpivot = RealScalar(0);
@@ -553,12 +553,12 @@ void ColPivHouseholderQR<MatrixType>::computeInPlace()
       // http://www.netlib.org/lapack/lawnspdf/lawn176.pdf
       // and used in LAPACK routines xGEQPF and xGEQP3.
       // See lines 278-297 in http://www.netlib.org/lapack/explore-html/dc/df4/sgeqpf_8f_source.html
-      if (m_colNormsUpdated.coeffRef(j) != 0) {
+      if (m_colNormsUpdated.coeffRef(j) != RealScalar(0)) {
         RealScalar temp = abs(m_qr.coeffRef(k, j)) / m_colNormsUpdated.coeffRef(j);
         temp = (RealScalar(1) + temp) * (RealScalar(1) - temp);
-        temp = temp < 0 ? 0 : temp;
-        RealScalar temp2 = temp * numext::abs2(m_colNormsUpdated.coeffRef(j) /
-                                               m_colNormsDirect.coeffRef(j));
+        temp = temp <  RealScalar(0) ? RealScalar(0) : temp;
+        RealScalar temp2 = temp * numext::abs2<RealScalar>(m_colNormsUpdated.coeffRef(j) /
+                                                           m_colNormsDirect.coeffRef(j));
         if (temp2 <= norm_downdate_threshold) {
           // The updated norm has become too inaccurate so re-compute the column
           // norm directly.

Eigen/src/QR/CompleteOrthogonalDecomposition.h

@@ -138,7 +138,7 @@ class CompleteOrthogonalDecomposition {
    * problem \f[\mathrm{minimize} \|A X - B\|, \f] where \b A is the matrix of
    * which \c *this is the complete orthogonal decomposition.
    *
-   * \param B the right-hand sides of the problem to solve.
+   * \param b the right-hand sides of the problem to solve.
    *
    * \returns a solution.
    *

Eigen/src/SVD/BDCSVD.h

@@ -77,6 +77,7 @@ public:
   typedef _MatrixType MatrixType;
   typedef typename MatrixType::Scalar Scalar;
   typedef typename NumTraits<typename MatrixType::Scalar>::Real RealScalar;
+  typedef typename NumTraits<RealScalar>::Literal Literal;
   enum {
     RowsAtCompileTime = MatrixType::RowsAtCompileTime, 
     ColsAtCompileTime = MatrixType::ColsAtCompileTime, 
@@ -259,7 +260,7 @@ BDCSVD<MatrixType>& BDCSVD<MatrixType>::compute(const MatrixType& matrix, unsign
   
   //**** step 0 - Copy the input matrix and apply scaling to reduce over/under-flows
   RealScalar scale = matrix.cwiseAbs().maxCoeff();
-  if(scale==RealScalar(0)) scale = RealScalar(1);
+  if(scale==Literal(0)) scale = Literal(1);
   MatrixX copy;
   if (m_isTranspose) copy = matrix.adjoint()/scale;
   else               copy = matrix/scale;
@@ -351,13 +352,13 @@ void BDCSVD<MatrixType>::structured_update(Block<MatrixXr,Dynamic,Dynamic> A, co
     Index k1=0, k2=0;
     for(Index j=0; j<n; ++j)
     {
-      if( (A.col(j).head(n1).array()!=0).any() )
+      if( (A.col(j).head(n1).array()!=Literal(0)).any() )
       {
         A1.col(k1) = A.col(j).head(n1);
         B1.row(k1) = B.row(j);
         ++k1;
       }
-      if( (A.col(j).tail(n2).array()!=0).any() )
+      if( (A.col(j).tail(n2).array()!=Literal(0)).any() )
       {
         A2.col(k2) = A.col(j).tail(n2);
         B2.row(k2) = B.row(j);
@@ -449,11 +450,11 @@ void BDCSVD<MatrixType>::divide (Index firstCol, Index lastCol, Index firstRowW,
     l = m_naiveU.row(1).segment(firstCol, k);
     f = m_naiveU.row(0).segment(firstCol + k + 1, n - k - 1);
   }
-  if (m_compV) m_naiveV(firstRowW+k, firstColW) = 1;
+  if (m_compV) m_naiveV(firstRowW+k, firstColW) = Literal(1);
   if (r0<considerZero)
   {
-    c0 = 1;
-    s0 = 0;
+    c0 = Literal(1);
+    s0 = Literal(0);
   }
   else
   {
@@ -574,7 +575,7 @@ void BDCSVD<MatrixType>::computeSVDofM(Index firstCol, Index n, MatrixXr& U, Vec
   ArrayRef col0 = m_computed.col(firstCol).segment(firstCol, n);
   m_workspace.head(n) =  m_computed.block(firstCol, firstCol, n, n).diagonal();
   ArrayRef diag = m_workspace.head(n);
-  diag(0) = 0;
+  diag(0) = Literal(0);
 
   // Allocate space for singular values and vectors
   singVals.resize(n);
@@ -590,7 +591,7 @@ void BDCSVD<MatrixType>::computeSVDofM(Index firstCol, Index n, MatrixXr& U, Vec
   // but others are interleaved and we must ignore them at this stage.
   // To this end, let's compute a permutation skipping them:
   Index actual_n = n;
-  while(actual_n>1 && diag(actual_n-1)==0) --actual_n;
+  while(actual_n>1 && diag(actual_n-1)==Literal(0)) --actual_n;
   Index m = 0; // size of the deflated problem
   for(Index k=0;k<actual_n;++k)
     if(abs(col0(k))>considerZero)
@@ -691,7 +692,7 @@ template <typename MatrixType>
 typename BDCSVD<MatrixType>::RealScalar BDCSVD<MatrixType>::secularEq(RealScalar mu, const ArrayRef& col0, const ArrayRef& diag, const IndicesRef &perm, const ArrayRef& diagShifted, RealScalar shift)
 {
   Index m = perm.size();
-  RealScalar res = 1;
+  RealScalar res = Literal(1);
   for(Index i=0; i<m; ++i)
   {
     Index j = perm(i);
@@ -710,16 +711,16 @@ void BDCSVD<MatrixType>::computeSingVals(const ArrayRef& col0, const ArrayRef& d
 
   Index n = col0.size();
   Index actual_n = n;
-  while(actual_n>1 && col0(actual_n-1)==0) --actual_n;
+  while(actual_n>1 && col0(actual_n-1)==Literal(0)) --actual_n;
 
   for (Index k = 0; k < n; ++k)
   {
-    if (col0(k) == 0 || actual_n==1)
+    if (col0(k) == Literal(0) || actual_n==1)
     {
       // if col0(k) == 0, then entry is deflated, so singular value is on diagonal
       // if actual_n==1, then the deflated problem is already diagonalized
       singVals(k) = k==0 ? col0(0) : diag(k);
-      mus(k) = 0;
+      mus(k) = Literal(0);
       shifts(k) = k==0 ? col0(0) : diag(k);
       continue;
     } 
@@ -733,13 +734,13 @@ void BDCSVD<MatrixType>::computeSingVals(const ArrayRef& col0, const ArrayRef& d
     {
       // Skip deflated singular values
       Index l = k+1;
-      while(col0(l)==0) { ++l; eigen_internal_assert(l<actual_n); }
+      while(col0(l)==Literal(0)) { ++l; eigen_internal_assert(l<actual_n); }
       right = diag(l);
     }
 
     // first decide whether it's closer to the left end or the right end
-    RealScalar mid = left + (right-left) / 2;
-    RealScalar fMid = secularEq(mid, col0, diag, perm, diag, 0);
+    RealScalar mid = left + (right-left) / Literal(2);
+    RealScalar fMid = secularEq(mid, col0, diag, perm, diag, Literal(0));
 #ifdef EIGEN_BDCSVD_DEBUG_VERBOSE
     std::cout << right-left << "\n";
     std::cout << "fMid = " << fMid << " " << secularEq(mid-left, col0, diag, perm, diag-left, left) << " " << secularEq(mid-right, col0, diag, perm, diag-right, right)   << "\n";
@@ -755,7 +756,7 @@ void BDCSVD<MatrixType>::computeSingVals(const ArrayRef& col0, const ArrayRef& d
               << " "       << secularEq(0.8*(left+right), col0, diag, perm, diag, 0)
               << " "       << secularEq(0.9*(left+right), col0, diag, perm, diag, 0) << "\n";
 #endif
-    RealScalar shift = (k == actual_n-1 || fMid > 0) ? left : right;
+    RealScalar shift = (k == actual_n-1 || fMid > Literal(0)) ? left : right;
     
     // measure everything relative to shift
     Map<ArrayXr> diagShifted(m_workspace.data()+4*n, n);
@@ -785,13 +786,13 @@ void BDCSVD<MatrixType>::computeSingVals(const ArrayRef& col0, const ArrayRef& d
 
     // rational interpolation: fit a function of the form a / mu + b through the two previous
     // iterates and use its zero to compute the next iterate
-    bool useBisection = fPrev*fCur>0;
-    while (fCur!=0 && abs(muCur - muPrev) > 8 * NumTraits<RealScalar>::epsilon() * numext::maxi<RealScalar>(abs(muCur), abs(muPrev)) && abs(fCur - fPrev)>NumTraits<RealScalar>::epsilon() && !useBisection)
+    bool useBisection = fPrev*fCur>Literal(0);
+    while (fCur!=Literal(0) && abs(muCur - muPrev) > Literal(8) * NumTraits<RealScalar>::epsilon() * numext::maxi<RealScalar>(abs(muCur), abs(muPrev)) && abs(fCur - fPrev)>NumTraits<RealScalar>::epsilon() && !useBisection)
     {
       ++m_numIters;
 
       // Find a and b such that the function f(mu) = a / mu + b matches the current and previous samples.
-      RealScalar a = (fCur - fPrev) / (1/muCur - 1/muPrev);
+      RealScalar a = (fCur - fPrev) / (Literal(1)/muCur - Literal(1)/muPrev);
       RealScalar b = fCur - a / muCur;
       // And find mu such that f(mu)==0:
       RealScalar muZero = -a/b;
@@ -803,8 +804,8 @@ void BDCSVD<MatrixType>::computeSingVals(const ArrayRef& col0, const ArrayRef& d
       fCur = fZero;
       
       
-      if (shift == left  && (muCur < 0 || muCur > right - left)) useBisection = true;
-      if (shift == right && (muCur < -(right - left) || muCur > 0)) useBisection = true;
+      if (shift == left  && (muCur < Literal(0) || muCur > right - left)) useBisection = true;
+      if (shift == right && (muCur < -(right - left) || muCur > Literal(0))) useBisection = true;
       if (abs(fCur)>abs(fPrev)) useBisection = true;
     }
 
@@ -841,13 +842,13 @@ void BDCSVD<MatrixType>::computeSingVals(const ArrayRef& col0, const ArrayRef& d
         std::cout << k << " : " <<  fLeft << " * " << fRight << " == " << fLeft * fRight << "  ;  " << left << " - " << right << " -> " <<  leftShifted << " " << rightShifted << "   shift=" << shift << "\n";
       }
 #endif
-      eigen_internal_assert(fLeft * fRight < 0);
+      eigen_internal_assert(fLeft * fRight < Literal(0));
       
-      while (rightShifted - leftShifted > 2 * NumTraits<RealScalar>::epsilon() * numext::maxi<RealScalar>(abs(leftShifted), abs(rightShifted)))
+      while (rightShifted - leftShifted > Literal(2) * NumTraits<RealScalar>::epsilon() * numext::maxi<RealScalar>(abs(leftShifted), abs(rightShifted)))
       {
-        RealScalar midShifted = (leftShifted + rightShifted) / 2;
+        RealScalar midShifted = (leftShifted + rightShifted) / Literal(2);
         fMid = secularEq(midShifted, col0, diag, perm, diagShifted, shift);
-        if (fLeft * fMid < 0)
+        if (fLeft * fMid < Literal(0))
         {
           rightShifted = midShifted;
         }
@@ -858,7 +859,7 @@ void BDCSVD<MatrixType>::computeSingVals(const ArrayRef& col0, const ArrayRef& d
         }
       }
 
-      muCur = (leftShifted + rightShifted) / 2;
+      muCur = (leftShifted + rightShifted) / Literal(2);
     }
       
     singVals[k] = shift + muCur;
@@ -892,8 +893,8 @@ void BDCSVD<MatrixType>::perturbCol0
   // The offset permits to skip deflated entries while computing zhat
   for (Index k = 0; k < n; ++k)
   {
-    if (col0(k) == 0) // deflated
-      zhat(k) = 0;
+    if (col0(k) == Literal(0)) // deflated
+      zhat(k) = Literal(0);
     else
     {
       // see equation (3.6)
@@ -918,7 +919,7 @@ void BDCSVD<MatrixType>::perturbCol0
       std::cout << "zhat(" << k << ") =  sqrt( " << prod << ")  ;  " << (singVals(last) + dk) << " * " << mus(last) + shifts(last) << " - " << dk << "\n";
 #endif
       RealScalar tmp = sqrt(prod);
-      zhat(k) = col0(k) > 0 ? tmp : -tmp;
+      zhat(k) = col0(k) > Literal(0) ? tmp : -tmp;
     }
   }
 }
@@ -934,7 +935,7 @@ void BDCSVD<MatrixType>::computeSingVecs
   
   for (Index k = 0; k < n; ++k)
   {
-    if (zhat(k) == 0)
+    if (zhat(k) == Literal(0))
     {
       U.col(k) = VectorType::Unit(n+1, k);
       if (m_compV) V.col(k) = VectorType::Unit(n, k);
@@ -947,7 +948,7 @@ void BDCSVD<MatrixType>::computeSingVecs
         Index i = perm(l);
         U(i,k) = zhat(i)/(((diag(i) - shifts(k)) - mus(k)) )/( (diag(i) + singVals[k]));
       }
-      U(n,k) = 0;      
+      U(n,k) = Literal(0);
       U.col(k).normalize();
     
       if (m_compV)
@@ -958,7 +959,7 @@ void BDCSVD<MatrixType>::computeSingVecs
           Index i = perm(l);
           V(i,k) = diag(i) * zhat(i) / (((diag(i) - shifts(k)) - mus(k)) )/( (diag(i) + singVals[k]));
         }
-        V(0,k) = -1;
+        V(0,k) = Literal(-1);
         V.col(k).normalize();
       }
     }
@@ -980,14 +981,14 @@ void BDCSVD<MatrixType>::deflation43(Index firstCol, Index shift, Index i, Index
   RealScalar c = m_computed(start, start);
   RealScalar s = m_computed(start+i, start);
   RealScalar r = sqrt(numext::abs2(c) + numext::abs2(s));
-  if (r == 0)
+  if (r == Literal(0))
   {
-    m_computed(start+i, start+i) = 0;
+    m_computed(start+i, start+i) = Literal(0);
     return;
   }
   m_computed(start,start) = r;  
-  m_computed(start+i, start) = 0;
-  m_computed(start+i, start+i) = 0;
+  m_computed(start+i, start) = Literal(0);
+  m_computed(start+i, start+i) = Literal(0);
   
   JacobiRotation<RealScalar> J(c/r,-s/r);
   if (m_compU)  m_naiveU.middleRows(firstCol, size+1).applyOnTheRight(firstCol, firstCol+i, J);
@@ -1020,7 +1021,7 @@ void BDCSVD<MatrixType>::deflation44(Index firstColu , Index firstColm, Index fi
     << m_computed(firstColm + i+1, firstColm+i+1) << " "
     << m_computed(firstColm + i+2, firstColm+i+2) << "\n";
 #endif
-  if (r==0)
+  if (r==Literal(0))
   {
     m_computed(firstColm + i, firstColm + i) = m_computed(firstColm + j, firstColm + j);
     return;
@@ -1029,7 +1030,7 @@ void BDCSVD<MatrixType>::deflation44(Index firstColu , Index firstColm, Index fi
   s/=r;
   m_computed(firstColm + i, firstColm) = r;  
   m_computed(firstColm + j, firstColm + j) = m_computed(firstColm + i, firstColm + i);
-  m_computed(firstColm + j, firstColm) = 0;
+  m_computed(firstColm + j, firstColm) = Literal(0);
 
   JacobiRotation<RealScalar> J(c,-s);
   if (m_compU)  m_naiveU.middleRows(firstColu, size+1).applyOnTheRight(firstColu + i, firstColu + j, J);
@@ -1053,7 +1054,7 @@ void BDCSVD<MatrixType>::deflation(Index firstCol, Index lastCol, Index k, Index
   const RealScalar considerZero = (std::numeric_limits<RealScalar>::min)();
   RealScalar maxDiag = diag.tail((std::max)(Index(1),length-1)).cwiseAbs().maxCoeff();
   RealScalar epsilon_strict = numext::maxi<RealScalar>(considerZero,NumTraits<RealScalar>::epsilon() * maxDiag);
-  RealScalar epsilon_coarse = 8 * NumTraits<RealScalar>::epsilon() * numext::maxi<RealScalar>(col0.cwiseAbs().maxCoeff(), maxDiag);
+  RealScalar epsilon_coarse = Literal(8) * NumTraits<RealScalar>::epsilon() * numext::maxi<RealScalar>(col0.cwiseAbs().maxCoeff(), maxDiag);
   
 #ifdef EIGEN_BDCSVD_SANITY_CHECKS
   assert(m_naiveU.allFinite());
@@ -1081,7 +1082,7 @@ void BDCSVD<MatrixType>::deflation(Index firstCol, Index lastCol, Index k, Index
 #ifdef  EIGEN_BDCSVD_DEBUG_VERBOSE
       std::cout << "deflation 4.2, set z(" << i << ") to zero because " << abs(col0(i)) << " < " << epsilon_strict << "  (diag(" << i << ")=" << diag(i) << ")\n";
 #endif
-      col0(i) = 0;
+      col0(i) = Literal(0);
     }
 
   //condition 4.3

Eigen/src/SVD/JacobiSVD.h

@@ -112,9 +112,11 @@ public:
     ColsAtCompileTime = MatrixType::ColsAtCompileTime,
     MaxRowsAtCompileTime = MatrixType::MaxRowsAtCompileTime,
     MaxColsAtCompileTime = MatrixType::MaxColsAtCompileTime,
-    Options = MatrixType::Options
+    TrOptions = RowsAtCompileTime==1 ? (MatrixType::Options & ~(RowMajor))
+              : ColsAtCompileTime==1 ? (MatrixType::Options |   RowMajor)
+              : MatrixType::Options
   };
-  typedef Matrix<Scalar, ColsAtCompileTime, RowsAtCompileTime, Options, MaxColsAtCompileTime, MaxRowsAtCompileTime>
+  typedef Matrix<Scalar, ColsAtCompileTime, RowsAtCompileTime, TrOptions, MaxColsAtCompileTime, MaxRowsAtCompileTime>
           TransposeTypeWithSameStorageOrder;
 
   void allocate(const JacobiSVD<MatrixType, FullPivHouseholderQRPreconditioner>& svd)
@@ -200,10 +202,12 @@ public:
     ColsAtCompileTime = MatrixType::ColsAtCompileTime,
     MaxRowsAtCompileTime = MatrixType::MaxRowsAtCompileTime,
     MaxColsAtCompileTime = MatrixType::MaxColsAtCompileTime,
-    Options = MatrixType::Options
+    TrOptions = RowsAtCompileTime==1 ? (MatrixType::Options & ~(RowMajor))
+              : ColsAtCompileTime==1 ? (MatrixType::Options |   RowMajor)
+              : MatrixType::Options
   };
 
-  typedef Matrix<Scalar, ColsAtCompileTime, RowsAtCompileTime, Options, MaxColsAtCompileTime, MaxRowsAtCompileTime>
+  typedef Matrix<Scalar, ColsAtCompileTime, RowsAtCompileTime, TrOptions, MaxColsAtCompileTime, MaxRowsAtCompileTime>
           TransposeTypeWithSameStorageOrder;
 
   void allocate(const JacobiSVD<MatrixType, ColPivHouseholderQRPreconditioner>& svd)
@@ -412,7 +416,7 @@ struct svd_precondition_2x2_block_to_be_real<MatrixType, QRPreconditioner, true>
     }
 
     // update largest diagonal entry
-    maxDiagEntry = numext::maxi(maxDiagEntry,numext::maxi(abs(work_matrix.coeff(p,p)), abs(work_matrix.coeff(q,q))));
+    maxDiagEntry = numext::maxi<RealScalar>(maxDiagEntry,numext::maxi<RealScalar>(abs(work_matrix.coeff(p,p)), abs(work_matrix.coeff(q,q))));
     // and check whether the 2x2 block is already diagonal
     RealScalar threshold = numext::maxi<RealScalar>(considerAsZero, precision * maxDiagEntry);
     return abs(work_matrix.coeff(p,q))>threshold || abs(work_matrix.coeff(q,p)) > threshold;
@@ -725,7 +729,7 @@ JacobiSVD<MatrixType, QRPreconditioner>::compute(const MatrixType& matrix, unsig
             if(computeV()) m_matrixV.applyOnTheRight(p,q,j_right);
 
             // keep track of the largest diagonal coefficient
-            maxDiagEntry = numext::maxi<RealScalar>(maxDiagEntry,numext::maxi(abs(m_workMatrix.coeff(p,p)), abs(m_workMatrix.coeff(q,q))));
+            maxDiagEntry = numext::maxi<RealScalar>(maxDiagEntry,numext::maxi<RealScalar>(abs(m_workMatrix.coeff(p,p)), abs(m_workMatrix.coeff(q,q))));
           }
         }
       }

Eigen/src/SVD/UpperBidiagonalization.h

@@ -159,6 +159,8 @@ void upperbidiagonalization_blocked_helper(MatrixType& A,
                                                       traits<MatrixType>::Flags & RowMajorBit> > Y)
 {
   typedef typename MatrixType::Scalar Scalar;
+  typedef typename MatrixType::RealScalar RealScalar;
+  typedef typename NumTraits<RealScalar>::Literal Literal;
   enum { StorageOrder = traits<MatrixType>::Flags & RowMajorBit };
   typedef InnerStride<int(StorageOrder) == int(ColMajor) ? 1 : Dynamic> ColInnerStride;
   typedef InnerStride<int(StorageOrder) == int(ColMajor) ? Dynamic : 1> RowInnerStride;
@@ -263,7 +265,7 @@ void upperbidiagonalization_blocked_helper(MatrixType& A,
     SubMatType A10( A.block(bs,0, brows-bs,bs) );
     SubMatType A01( A.block(0,bs, bs,bcols-bs) );
     Scalar tmp = A01(bs-1,0);
-    A01(bs-1,0) = 1;
+    A01(bs-1,0) = Literal(1);
     A11.noalias() -= A10 * Y.topLeftCorner(bcols,bs).bottomRows(bcols-bs).adjoint();
     A11.noalias() -= X.topLeftCorner(brows,bs).bottomRows(brows-bs) * A01;
     A01(bs-1,0) = tmp;

Eigen/src/SparseCore/AmbiVector.h

@@ -336,7 +336,7 @@ class AmbiVector<_Scalar,_StorageIndex>::Iterator
       {
         do {
           ++m_cachedIndex;
-        } while (m_cachedIndex<m_vector.m_end && abs(m_vector.m_buffer[m_cachedIndex])<m_epsilon);
+        } while (m_cachedIndex<m_vector.m_end && abs(m_vector.m_buffer[m_cachedIndex])<=m_epsilon);
         if (m_cachedIndex<m_vector.m_end)
           m_cachedValue = m_vector.m_buffer[m_cachedIndex];
         else
@@ -347,7 +347,7 @@ class AmbiVector<_Scalar,_StorageIndex>::Iterator
         ListEl* EIGEN_RESTRICT llElements = reinterpret_cast<ListEl*>(m_vector.m_buffer);
         do {
           m_currentEl = llElements[m_currentEl].next;
-        } while (m_currentEl>=0 && abs(llElements[m_currentEl].value)<m_epsilon);
+        } while (m_currentEl>=0 && abs(llElements[m_currentEl].value)<=m_epsilon);
         if (m_currentEl<0)
         {
           m_cachedIndex = -1;
@@ -363,9 +363,9 @@ class AmbiVector<_Scalar,_StorageIndex>::Iterator
 
   protected:
     const AmbiVector& m_vector; // the target vector
-    StorageIndex m_currentEl;            // the current element in sparse/linked-list mode
+    StorageIndex m_currentEl;   // the current element in sparse/linked-list mode
     RealScalar m_epsilon;       // epsilon used to prune zero coefficients
-    StorageIndex m_cachedIndex;          // current coordinate
+    StorageIndex m_cachedIndex; // current coordinate
     Scalar m_cachedValue;       // current value
     bool m_isDense;             // mode of the vector
 };

Eigen/src/SparseCore/SparseAssign.h

@@ -143,10 +143,7 @@ struct Assignment<DstXprType, SrcXprType, Functor, Sparse2Dense>
       dst.setZero();
     
     internal::evaluator<SrcXprType> srcEval(src);
-    Index dstRows = src.rows();
-    Index dstCols = src.cols();
-    if((dst.rows()!=dstRows) || (dst.cols()!=dstCols))
-      dst.resize(dstRows, dstCols);
+    resize_if_allowed(dst, src, func);
     internal::evaluator<DstXprType> dstEval(dst);
     
     const Index outerEvaluationSize = (internal::evaluator<SrcXprType>::Flags&RowMajorBit) ? src.rows() : src.cols();

Eigen/src/SparseCore/SparseCompressedBase.h

@@ -224,11 +224,11 @@ class SparseCompressedBase<Derived>::ReverseInnerIterator
       }
       else
       {
-        m_start.value() = mat.outerIndexPtr()[outer];
+        m_start = mat.outerIndexPtr()[outer];
         if(mat.isCompressed())
           m_id = mat.outerIndexPtr()[outer+1];
         else
-          m_id = m_start.value() + mat.innerNonZeroPtr()[outer];
+          m_id = m_start + mat.innerNonZeroPtr()[outer];
       }
     }
 
@@ -254,14 +254,15 @@ class SparseCompressedBase<Derived>::ReverseInnerIterator
     inline Index row() const { return IsRowMajor ? m_outer.value() : index(); }
     inline Index col() const { return IsRowMajor ? index() : m_outer.value(); }
 
-    inline operator bool() const { return (m_id > m_start.value()); }
+    inline operator bool() const { return (m_id > m_start); }
 
   protected:
     const Scalar* m_values;
     const StorageIndex* m_indices;
-    const internal::variable_if_dynamic<Index,Derived::IsVectorAtCompileTime?0:Dynamic> m_outer;
+    typedef internal::variable_if_dynamic<Index,Derived::IsVectorAtCompileTime?0:Dynamic> OuterType;
+    const OuterType m_outer;
+    Index m_start;
     Index m_id;
-    const internal::variable_if_dynamic<Index,Derived::IsVectorAtCompileTime?0:Dynamic> m_start;
 };
 
 namespace internal {
@@ -272,18 +273,17 @@ struct evaluator<SparseCompressedBase<Derived> >
 {
   typedef typename Derived::Scalar Scalar;
   typedef typename Derived::InnerIterator InnerIterator;
-  typedef typename Derived::ReverseInnerIterator ReverseInnerIterator;
   
   enum {
     CoeffReadCost = NumTraits<Scalar>::ReadCost,
     Flags = Derived::Flags
   };
   
-  evaluator() : m_matrix(0)
+  evaluator() : m_matrix(0), m_zero(0)
   {
     EIGEN_INTERNAL_CHECK_COST_VALUE(CoeffReadCost);
   }
-  explicit evaluator(const Derived &mat) : m_matrix(&mat)
+  explicit evaluator(const Derived &mat) : m_matrix(&mat), m_zero(0)
   {
     EIGEN_INTERNAL_CHECK_COST_VALUE(CoeffReadCost);
   }
@@ -296,26 +296,42 @@ struct evaluator<SparseCompressedBase<Derived> >
   operator const Derived&() const { return *m_matrix; }
   
   typedef typename DenseCoeffsBase<Derived,ReadOnlyAccessors>::CoeffReturnType CoeffReturnType;
-  Scalar coeff(Index row, Index col) const
-  { return m_matrix->coeff(row,col); }
-  
+  const Scalar& coeff(Index row, Index col) const
+  {
+    Index p = find(row,col);
+
+    if(p==Dynamic)
+      return m_zero;
+    else
+      return m_matrix->const_cast_derived().valuePtr()[p];
+  }
+
   Scalar& coeffRef(Index row, Index col)
+  {
+    Index p = find(row,col);
+    eigen_assert(p!=Dynamic && "written coefficient does not exist");
+    return m_matrix->const_cast_derived().valuePtr()[p];
+  }
+
+protected:
+
+  Index find(Index row, Index col) const
   {
     eigen_internal_assert(row>=0 && row<m_matrix->rows() && col>=0 && col<m_matrix->cols());
-      
+
     const Index outer = Derived::IsRowMajor ? row : col;
     const Index inner = Derived::IsRowMajor ? col : row;
 
     Index start = m_matrix->outerIndexPtr()[outer];
     Index end = m_matrix->isCompressed() ? m_matrix->outerIndexPtr()[outer+1] : m_matrix->outerIndexPtr()[outer] + m_matrix->innerNonZeroPtr()[outer];
-    eigen_assert(end>start && "you are using a non finalized sparse matrix or written coefficient does not exist");
-    const Index p =   std::lower_bound(m_matrix->innerIndexPtr()+start, m_matrix->innerIndexPtr()+end,inner)
-                    - m_matrix->innerIndexPtr();
-    eigen_assert((p<end) && (m_matrix->innerIndexPtr()[p]==inner) && "written coefficient does not exist");
-    return m_matrix->const_cast_derived().valuePtr()[p];
+    eigen_assert(end>=start && "you are using a non finalized sparse matrix or written coefficient does not exist");
+    const Index p = std::lower_bound(m_matrix->innerIndexPtr()+start, m_matrix->innerIndexPtr()+end,inner) - m_matrix->innerIndexPtr();
+
+    return ((p<end) && (m_matrix->innerIndexPtr()[p]==inner)) ? p : Dynamic;
   }
 
   const Derived *m_matrix;
+  const Scalar m_zero;
 };
 
 }

Eigen/src/SparseCore/SparseCwiseBinaryOp.h

@@ -45,7 +45,7 @@ class CwiseBinaryOpImpl<BinaryOp, Lhs, Rhs, Sparse>
       EIGEN_STATIC_ASSERT((
                 (!internal::is_same<typename internal::traits<Lhs>::StorageKind,
                                     typename internal::traits<Rhs>::StorageKind>::value)
-            ||  ((Lhs::Flags&RowMajorBit) == (Rhs::Flags&RowMajorBit))),
+            ||  ((internal::evaluator<Lhs>::Flags&RowMajorBit) == (internal::evaluator<Rhs>::Flags&RowMajorBit))),
             THE_STORAGE_ORDER_OF_BOTH_SIDES_MUST_MATCH);
     }
 };
@@ -68,7 +68,6 @@ protected:
   typedef typename XprType::StorageIndex StorageIndex;
 public:
 
-  class ReverseInnerIterator;
   class InnerIterator
   {
   public:
@@ -111,6 +110,7 @@ public:
     EIGEN_STRONG_INLINE Scalar value() const { return m_value; }
 
     EIGEN_STRONG_INLINE StorageIndex index() const { return m_id; }
+    EIGEN_STRONG_INLINE Index outer() const { return m_lhsIter.outer(); }
     EIGEN_STRONG_INLINE Index row() const { return Lhs::IsRowMajor ? m_lhsIter.row() : index(); }
     EIGEN_STRONG_INLINE Index col() const { return Lhs::IsRowMajor ? index() : m_lhsIter.col(); }
 
@@ -161,7 +161,6 @@ protected:
   typedef typename XprType::StorageIndex StorageIndex;
 public:
 
-  class ReverseInnerIterator;
   class InnerIterator
   {
     enum { IsRowMajor = (int(Rhs::Flags)&RowMajorBit)==RowMajorBit };
@@ -195,6 +194,7 @@ public:
     EIGEN_STRONG_INLINE Scalar value() const { eigen_internal_assert(m_id<m_innerSize); return m_value; }
 
     EIGEN_STRONG_INLINE StorageIndex index() const { return m_id; }
+    EIGEN_STRONG_INLINE Index outer() const { return m_rhsIter.outer(); }
     EIGEN_STRONG_INLINE Index row() const { return IsRowMajor ? m_rhsIter.outer() : m_id; }
     EIGEN_STRONG_INLINE Index col() const { return IsRowMajor ? m_id : m_rhsIter.outer(); }
 
@@ -249,7 +249,6 @@ protected:
   typedef typename XprType::StorageIndex StorageIndex;
 public:
 
-  class ReverseInnerIterator;
   class InnerIterator
   {
     enum { IsRowMajor = (int(Lhs::Flags)&RowMajorBit)==RowMajorBit };
@@ -283,6 +282,7 @@ public:
     EIGEN_STRONG_INLINE Scalar value() const { eigen_internal_assert(m_id<m_innerSize); return m_value; }
 
     EIGEN_STRONG_INLINE StorageIndex index() const { return m_id; }
+    EIGEN_STRONG_INLINE Index outer() const { return m_lhsIter.outer(); }
     EIGEN_STRONG_INLINE Index row() const { return IsRowMajor ? m_lhsIter.outer() : m_id; }
     EIGEN_STRONG_INLINE Index col() const { return IsRowMajor ? m_id : m_lhsIter.outer(); }
 
@@ -325,26 +325,98 @@ protected:
   const XprType &m_expr;
 };
 
+template<typename T,
+         typename LhsKind   = typename evaluator_traits<typename T::Lhs>::Kind,
+         typename RhsKind   = typename evaluator_traits<typename T::Rhs>::Kind,
+         typename LhsScalar = typename traits<typename T::Lhs>::Scalar,
+         typename RhsScalar = typename traits<typename T::Rhs>::Scalar> struct sparse_conjunction_evaluator;
+
 // "sparse .* sparse"
 template<typename T1, typename T2, typename Lhs, typename Rhs>
 struct binary_evaluator<CwiseBinaryOp<scalar_product_op<T1,T2>, Lhs, Rhs>, IteratorBased, IteratorBased>
-  : evaluator_base<CwiseBinaryOp<scalar_product_op<T1,T2>, Lhs, Rhs> >
+  : sparse_conjunction_evaluator<CwiseBinaryOp<scalar_product_op<T1,T2>, Lhs, Rhs> >
+{
+  typedef CwiseBinaryOp<scalar_product_op<T1,T2>, Lhs, Rhs> XprType;
+  typedef sparse_conjunction_evaluator<XprType> Base;
+  explicit binary_evaluator(const XprType& xpr) : Base(xpr) {}
+};
+// "dense .* sparse"
+template<typename T1, typename T2, typename Lhs, typename Rhs>
+struct binary_evaluator<CwiseBinaryOp<scalar_product_op<T1,T2>, Lhs, Rhs>, IndexBased, IteratorBased>
+  : sparse_conjunction_evaluator<CwiseBinaryOp<scalar_product_op<T1,T2>, Lhs, Rhs> >
+{
+  typedef CwiseBinaryOp<scalar_product_op<T1,T2>, Lhs, Rhs> XprType;
+  typedef sparse_conjunction_evaluator<XprType> Base;
+  explicit binary_evaluator(const XprType& xpr) : Base(xpr) {}
+};
+// "sparse .* dense"
+template<typename T1, typename T2, typename Lhs, typename Rhs>
+struct binary_evaluator<CwiseBinaryOp<scalar_product_op<T1,T2>, Lhs, Rhs>, IteratorBased, IndexBased>
+  : sparse_conjunction_evaluator<CwiseBinaryOp<scalar_product_op<T1,T2>, Lhs, Rhs> >
+{
+  typedef CwiseBinaryOp<scalar_product_op<T1,T2>, Lhs, Rhs> XprType;
+  typedef sparse_conjunction_evaluator<XprType> Base;
+  explicit binary_evaluator(const XprType& xpr) : Base(xpr) {}
+};
+
+// "sparse ./ dense"
+template<typename T1, typename T2, typename Lhs, typename Rhs>
+struct binary_evaluator<CwiseBinaryOp<scalar_quotient_op<T1,T2>, Lhs, Rhs>, IteratorBased, IndexBased>
+  : sparse_conjunction_evaluator<CwiseBinaryOp<scalar_quotient_op<T1,T2>, Lhs, Rhs> >
+{
+  typedef CwiseBinaryOp<scalar_quotient_op<T1,T2>, Lhs, Rhs> XprType;
+  typedef sparse_conjunction_evaluator<XprType> Base;
+  explicit binary_evaluator(const XprType& xpr) : Base(xpr) {}
+};
+
+// "sparse && sparse"
+template<typename Lhs, typename Rhs>
+struct binary_evaluator<CwiseBinaryOp<scalar_boolean_and_op, Lhs, Rhs>, IteratorBased, IteratorBased>
+  : sparse_conjunction_evaluator<CwiseBinaryOp<scalar_boolean_and_op, Lhs, Rhs> >
+{
+  typedef CwiseBinaryOp<scalar_boolean_and_op, Lhs, Rhs> XprType;
+  typedef sparse_conjunction_evaluator<XprType> Base;
+  explicit binary_evaluator(const XprType& xpr) : Base(xpr) {}
+};
+// "dense && sparse"
+template<typename Lhs, typename Rhs>
+struct binary_evaluator<CwiseBinaryOp<scalar_boolean_and_op, Lhs, Rhs>, IndexBased, IteratorBased>
+  : sparse_conjunction_evaluator<CwiseBinaryOp<scalar_boolean_and_op, Lhs, Rhs> >
+{
+  typedef CwiseBinaryOp<scalar_boolean_and_op, Lhs, Rhs> XprType;
+  typedef sparse_conjunction_evaluator<XprType> Base;
+  explicit binary_evaluator(const XprType& xpr) : Base(xpr) {}
+};
+// "sparse && dense"
+template<typename Lhs, typename Rhs>
+struct binary_evaluator<CwiseBinaryOp<scalar_boolean_and_op, Lhs, Rhs>, IteratorBased, IndexBased>
+  : sparse_conjunction_evaluator<CwiseBinaryOp<scalar_boolean_and_op, Lhs, Rhs> >
+{
+  typedef CwiseBinaryOp<scalar_boolean_and_op, Lhs, Rhs> XprType;
+  typedef sparse_conjunction_evaluator<XprType> Base;
+  explicit binary_evaluator(const XprType& xpr) : Base(xpr) {}
+};
+
+// "sparse ^ sparse"
+template<typename XprType>
+struct sparse_conjunction_evaluator<XprType, IteratorBased, IteratorBased>
+  : evaluator_base<XprType>
 {
 protected:
-  typedef scalar_product_op<T1,T2> BinaryOp;
-  typedef typename evaluator<Lhs>::InnerIterator  LhsIterator;
-  typedef typename evaluator<Rhs>::InnerIterator  RhsIterator;
-  typedef CwiseBinaryOp<BinaryOp, Lhs, Rhs> XprType;
+  typedef typename XprType::Functor BinaryOp;
+  typedef typename XprType::Lhs LhsArg;
+  typedef typename XprType::Rhs RhsArg;
+  typedef typename evaluator<LhsArg>::InnerIterator  LhsIterator;
+  typedef typename evaluator<RhsArg>::InnerIterator  RhsIterator;
   typedef typename XprType::StorageIndex StorageIndex;
   typedef typename traits<XprType>::Scalar Scalar;
 public:
 
-  class ReverseInnerIterator;
   class InnerIterator
   {
   public:
     
-    EIGEN_STRONG_INLINE InnerIterator(const binary_evaluator& aEval, Index outer)
+    EIGEN_STRONG_INLINE InnerIterator(const sparse_conjunction_evaluator& aEval, Index outer)
       : m_lhsIter(aEval.m_lhsImpl,outer), m_rhsIter(aEval.m_rhsImpl,outer), m_functor(aEval.m_functor)
     {
       while (m_lhsIter && m_rhsIter && (m_lhsIter.index() != m_rhsIter.index()))
@@ -373,6 +445,7 @@ public:
     EIGEN_STRONG_INLINE Scalar value() const { return m_functor(m_lhsIter.value(), m_rhsIter.value()); }
 
     EIGEN_STRONG_INLINE StorageIndex index() const { return m_lhsIter.index(); }
+    EIGEN_STRONG_INLINE Index outer() const { return m_lhsIter.outer(); }
     EIGEN_STRONG_INLINE Index row() const { return m_lhsIter.row(); }
     EIGEN_STRONG_INLINE Index col() const { return m_lhsIter.col(); }
 
@@ -386,11 +459,11 @@ public:
   
   
   enum {
-    CoeffReadCost = evaluator<Lhs>::CoeffReadCost + evaluator<Rhs>::CoeffReadCost + functor_traits<BinaryOp>::Cost,
+    CoeffReadCost = evaluator<LhsArg>::CoeffReadCost + evaluator<RhsArg>::CoeffReadCost + functor_traits<BinaryOp>::Cost,
     Flags = XprType::Flags
   };
   
-  explicit binary_evaluator(const XprType& xpr)
+  explicit sparse_conjunction_evaluator(const XprType& xpr)
     : m_functor(xpr.functor()),
       m_lhsImpl(xpr.lhs()), 
       m_rhsImpl(xpr.rhs())  
@@ -405,32 +478,32 @@ public:
 
 protected:
   const BinaryOp m_functor;
-  evaluator<Lhs> m_lhsImpl;
-  evaluator<Rhs> m_rhsImpl;
+  evaluator<LhsArg> m_lhsImpl;
+  evaluator<RhsArg> m_rhsImpl;
 };
 
-// "dense .* sparse"
-template<typename T1, typename T2, typename Lhs, typename Rhs>
-struct binary_evaluator<CwiseBinaryOp<scalar_product_op<T1,T2>, Lhs, Rhs>, IndexBased, IteratorBased>
-  : evaluator_base<CwiseBinaryOp<scalar_product_op<T1,T2>, Lhs, Rhs> >
+// "dense ^ sparse"
+template<typename XprType>
+struct sparse_conjunction_evaluator<XprType, IndexBased, IteratorBased>
+  : evaluator_base<XprType>
 {
 protected:
-  typedef scalar_product_op<T1,T2> BinaryOp;
-  typedef evaluator<Lhs>  LhsEvaluator;
-  typedef typename evaluator<Rhs>::InnerIterator  RhsIterator;
-  typedef CwiseBinaryOp<BinaryOp, Lhs, Rhs> XprType;
+  typedef typename XprType::Functor BinaryOp;
+  typedef typename XprType::Lhs LhsArg;
+  typedef typename XprType::Rhs RhsArg;
+  typedef evaluator<LhsArg> LhsEvaluator;
+  typedef typename evaluator<RhsArg>::InnerIterator  RhsIterator;
   typedef typename XprType::StorageIndex StorageIndex;
   typedef typename traits<XprType>::Scalar Scalar;
 public:
 
-  class ReverseInnerIterator;
   class InnerIterator
   {
-    enum { IsRowMajor = (int(Rhs::Flags)&RowMajorBit)==RowMajorBit };
+    enum { IsRowMajor = (int(RhsArg::Flags)&RowMajorBit)==RowMajorBit };
 
   public:
     
-    EIGEN_STRONG_INLINE InnerIterator(const binary_evaluator& aEval, Index outer)
+    EIGEN_STRONG_INLINE InnerIterator(const sparse_conjunction_evaluator& aEval, Index outer)
       : m_lhsEval(aEval.m_lhsImpl), m_rhsIter(aEval.m_rhsImpl,outer), m_functor(aEval.m_functor), m_outer(outer)
     {}
 
@@ -444,6 +517,7 @@ public:
     { return m_functor(m_lhsEval.coeff(IsRowMajor?m_outer:m_rhsIter.index(),IsRowMajor?m_rhsIter.index():m_outer), m_rhsIter.value()); }
 
     EIGEN_STRONG_INLINE StorageIndex index() const { return m_rhsIter.index(); }
+    EIGEN_STRONG_INLINE Index outer() const { return m_rhsIter.outer(); }
     EIGEN_STRONG_INLINE Index row() const { return m_rhsIter.row(); }
     EIGEN_STRONG_INLINE Index col() const { return m_rhsIter.col(); }
 
@@ -458,12 +532,12 @@ public:
   
   
   enum {
-    CoeffReadCost = evaluator<Lhs>::CoeffReadCost + evaluator<Rhs>::CoeffReadCost + functor_traits<BinaryOp>::Cost,
+    CoeffReadCost = evaluator<LhsArg>::CoeffReadCost + evaluator<RhsArg>::CoeffReadCost + functor_traits<BinaryOp>::Cost,
     // Expose storage order of the sparse expression
-    Flags = (XprType::Flags & ~RowMajorBit) | (int(Rhs::Flags)&RowMajorBit)
+    Flags = (XprType::Flags & ~RowMajorBit) | (int(RhsArg::Flags)&RowMajorBit)
   };
   
-  explicit binary_evaluator(const XprType& xpr)
+  explicit sparse_conjunction_evaluator(const XprType& xpr)
     : m_functor(xpr.functor()),
       m_lhsImpl(xpr.lhs()), 
       m_rhsImpl(xpr.rhs())  
@@ -478,32 +552,32 @@ public:
 
 protected:
   const BinaryOp m_functor;
-  evaluator<Lhs> m_lhsImpl;
-  evaluator<Rhs> m_rhsImpl;
+  evaluator<LhsArg> m_lhsImpl;
+  evaluator<RhsArg> m_rhsImpl;
 };
 
-// "sparse .* dense"
-template<typename T1, typename T2, typename Lhs, typename Rhs>
-struct binary_evaluator<CwiseBinaryOp<scalar_product_op<T1,T2>, Lhs, Rhs>, IteratorBased, IndexBased>
-  : evaluator_base<CwiseBinaryOp<scalar_product_op<T1,T2>, Lhs, Rhs> >
+// "sparse ^ dense"
+template<typename XprType>
+struct sparse_conjunction_evaluator<XprType, IteratorBased, IndexBased>
+  : evaluator_base<XprType>
 {
 protected:
-  typedef scalar_product_op<T1,T2> BinaryOp;
-  typedef typename evaluator<Lhs>::InnerIterator  LhsIterator;
-  typedef evaluator<Rhs>  RhsEvaluator;
-  typedef CwiseBinaryOp<BinaryOp, Lhs, Rhs> XprType;
+  typedef typename XprType::Functor BinaryOp;
+  typedef typename XprType::Lhs LhsArg;
+  typedef typename XprType::Rhs RhsArg;
+  typedef typename evaluator<LhsArg>::InnerIterator LhsIterator;
+  typedef evaluator<RhsArg> RhsEvaluator;
   typedef typename XprType::StorageIndex StorageIndex;
   typedef typename traits<XprType>::Scalar Scalar;
 public:
 
-  class ReverseInnerIterator;
   class InnerIterator
   {
-    enum { IsRowMajor = (int(Lhs::Flags)&RowMajorBit)==RowMajorBit };
+    enum { IsRowMajor = (int(LhsArg::Flags)&RowMajorBit)==RowMajorBit };
 
   public:
     
-    EIGEN_STRONG_INLINE InnerIterator(const binary_evaluator& aEval, Index outer)
+    EIGEN_STRONG_INLINE InnerIterator(const sparse_conjunction_evaluator& aEval, Index outer)
       : m_lhsIter(aEval.m_lhsImpl,outer), m_rhsEval(aEval.m_rhsImpl), m_functor(aEval.m_functor), m_outer(outer)
     {}
 
@@ -518,6 +592,7 @@ public:
                        m_rhsEval.coeff(IsRowMajor?m_outer:m_lhsIter.index(),IsRowMajor?m_lhsIter.index():m_outer)); }
 
     EIGEN_STRONG_INLINE StorageIndex index() const { return m_lhsIter.index(); }
+    EIGEN_STRONG_INLINE Index outer() const { return m_lhsIter.outer(); }
     EIGEN_STRONG_INLINE Index row() const { return m_lhsIter.row(); }
     EIGEN_STRONG_INLINE Index col() const { return m_lhsIter.col(); }
 
@@ -525,19 +600,19 @@ public:
     
   protected:
     LhsIterator m_lhsIter;
-    const evaluator<Rhs> &m_rhsEval;
+    const evaluator<RhsArg> &m_rhsEval;
     const BinaryOp& m_functor;
     const Index m_outer;
   };
   
   
   enum {
-    CoeffReadCost = evaluator<Lhs>::CoeffReadCost + evaluator<Rhs>::CoeffReadCost + functor_traits<BinaryOp>::Cost,
+    CoeffReadCost = evaluator<LhsArg>::CoeffReadCost + evaluator<RhsArg>::CoeffReadCost + functor_traits<BinaryOp>::Cost,
     // Expose storage order of the sparse expression
-    Flags = (XprType::Flags & ~RowMajorBit) | (int(Lhs::Flags)&RowMajorBit)
+    Flags = (XprType::Flags & ~RowMajorBit) | (int(LhsArg::Flags)&RowMajorBit)
   };
   
-  explicit binary_evaluator(const XprType& xpr)
+  explicit sparse_conjunction_evaluator(const XprType& xpr)
     : m_functor(xpr.functor()),
       m_lhsImpl(xpr.lhs()), 
       m_rhsImpl(xpr.rhs())  
@@ -552,8 +627,8 @@ public:
 
 protected:
   const BinaryOp m_functor;
-  evaluator<Lhs> m_lhsImpl;
-  evaluator<Rhs> m_rhsImpl;
+  evaluator<LhsArg> m_lhsImpl;
+  evaluator<RhsArg> m_rhsImpl;
 };
 
 }
@@ -562,6 +637,22 @@ protected:
 * Implementation of SparseMatrixBase and SparseCwise functions/operators
 ***************************************************************************/
 
+template<typename Derived>
+template<typename OtherDerived>
+Derived& SparseMatrixBase<Derived>::operator+=(const EigenBase<OtherDerived> &other)
+{
+  call_assignment(derived(), other.derived(), internal::add_assign_op<Scalar,typename OtherDerived::Scalar>());
+  return derived();
+}
+
+template<typename Derived>
+template<typename OtherDerived>
+Derived& SparseMatrixBase<Derived>::operator-=(const EigenBase<OtherDerived> &other)
+{
+  call_assignment(derived(), other.derived(), internal::assign_op<Scalar,typename OtherDerived::Scalar>());
+  return derived();
+}
+
 template<typename Derived>
 template<typename OtherDerived>
 EIGEN_STRONG_INLINE Derived &

Eigen/src/SparseCore/SparseCwiseUnaryOp.h

@@ -22,7 +22,6 @@ struct unary_evaluator<CwiseUnaryOp<UnaryOp,ArgType>, IteratorBased>
     typedef CwiseUnaryOp<UnaryOp, ArgType> XprType;
 
     class InnerIterator;
-    class ReverseInnerIterator;
     
     enum {
       CoeffReadCost = evaluator<ArgType>::CoeffReadCost + functor_traits<UnaryOp>::Cost,
@@ -41,7 +40,6 @@ struct unary_evaluator<CwiseUnaryOp<UnaryOp,ArgType>, IteratorBased>
 
   protected:
     typedef typename evaluator<ArgType>::InnerIterator        EvalIterator;
-//     typedef typename evaluator<ArgType>::ReverseInnerIterator EvalReverseIterator;
     
     const UnaryOp m_functor;
     evaluator<ArgType> m_argImpl;
@@ -70,33 +68,6 @@ class unary_evaluator<CwiseUnaryOp<UnaryOp,ArgType>, IteratorBased>::InnerIterat
     Scalar& valueRef();
 };
 
-// template<typename UnaryOp, typename ArgType>
-// class unary_evaluator<CwiseUnaryOp<UnaryOp,ArgType>, IteratorBased>::ReverseInnerIterator
-//     : public unary_evaluator<CwiseUnaryOp<UnaryOp,ArgType>, IteratorBased>::EvalReverseIterator
-// {
-//     typedef typename XprType::Scalar Scalar;
-//     typedef typename unary_evaluator<CwiseUnaryOp<UnaryOp,ArgType>, IteratorBased>::EvalReverseIterator Base;
-//   public:
-// 
-//     EIGEN_STRONG_INLINE ReverseInnerIterator(const XprType& unaryOp, typename XprType::Index outer)
-//       : Base(unaryOp.derived().nestedExpression(),outer), m_functor(unaryOp.derived().functor())
-//     {}
-// 
-//     EIGEN_STRONG_INLINE ReverseInnerIterator& operator--()
-//     { Base::operator--(); return *this; }
-// 
-//     EIGEN_STRONG_INLINE Scalar value() const { return m_functor(Base::value()); }
-// 
-//   protected:
-//     const UnaryOp m_functor;
-//   private:
-//     Scalar& valueRef();
-// };
-
-
-
-
-
 template<typename ViewOp, typename ArgType>
 struct unary_evaluator<CwiseUnaryView<ViewOp,ArgType>, IteratorBased>
   : public evaluator_base<CwiseUnaryView<ViewOp,ArgType> >
@@ -105,7 +76,6 @@ struct unary_evaluator<CwiseUnaryView<ViewOp,ArgType>, IteratorBased>
     typedef CwiseUnaryView<ViewOp, ArgType> XprType;
 
     class InnerIterator;
-    class ReverseInnerIterator;
     
     enum {
       CoeffReadCost = evaluator<ArgType>::CoeffReadCost + functor_traits<ViewOp>::Cost,
@@ -120,7 +90,6 @@ struct unary_evaluator<CwiseUnaryView<ViewOp,ArgType>, IteratorBased>
 
   protected:
     typedef typename evaluator<ArgType>::InnerIterator        EvalIterator;
-//     typedef typename evaluator<ArgType>::ReverseInnerIterator EvalReverseIterator;
     
     const ViewOp m_functor;
     evaluator<ArgType> m_argImpl;
@@ -148,37 +117,16 @@ class unary_evaluator<CwiseUnaryView<ViewOp,ArgType>, IteratorBased>::InnerItera
     const ViewOp m_functor;
 };
 
-// template<typename ViewOp, typename ArgType>
-// class unary_evaluator<CwiseUnaryView<ViewOp,ArgType>, IteratorBased>::ReverseInnerIterator
-//     : public unary_evaluator<CwiseUnaryView<ViewOp,ArgType>, IteratorBased>::EvalReverseIterator
-// {
-//     typedef typename XprType::Scalar Scalar;
-//     typedef typename unary_evaluator<CwiseUnaryView<ViewOp,ArgType>, IteratorBased>::EvalReverseIterator Base;
-//   public:
-// 
-//     EIGEN_STRONG_INLINE ReverseInnerIterator(const XprType& unaryOp, typename XprType::Index outer)
-//       : Base(unaryOp.derived().nestedExpression(),outer), m_functor(unaryOp.derived().functor())
-//     {}
-// 
-//     EIGEN_STRONG_INLINE ReverseInnerIterator& operator--()
-//     { Base::operator--(); return *this; }
-// 
-//     EIGEN_STRONG_INLINE Scalar value() const { return m_functor(Base::value()); }
-//     EIGEN_STRONG_INLINE Scalar& valueRef() { return m_functor(Base::valueRef()); }
-// 
-//   protected:
-//     const ViewOp m_functor;
-// };
-
-
 } // end namespace internal
 
 template<typename Derived>
 EIGEN_STRONG_INLINE Derived&
 SparseMatrixBase<Derived>::operator*=(const Scalar& other)
 {
+  typedef typename internal::evaluator<Derived>::InnerIterator EvalIterator;
+  internal::evaluator<Derived> thisEval(derived());
   for (Index j=0; j<outerSize(); ++j)
-    for (typename Derived::InnerIterator i(derived(),j); i; ++i)
+    for (EvalIterator i(thisEval,j); i; ++i)
       i.valueRef() *= other;
   return derived();
 }
@@ -187,8 +135,10 @@ template<typename Derived>
 EIGEN_STRONG_INLINE Derived&
 SparseMatrixBase<Derived>::operator/=(const Scalar& other)
 {
+  typedef typename internal::evaluator<Derived>::InnerIterator EvalIterator;
+  internal::evaluator<Derived> thisEval(derived());
   for (Index j=0; j<outerSize(); ++j)
-    for (typename Derived::InnerIterator i(derived(),j); i; ++i)
+    for (EvalIterator i(thisEval,j); i; ++i)
       i.valueRef() /= other;
   return derived();
 }

Eigen/src/SparseCore/SparseDiagonalProduct.h

@@ -80,6 +80,8 @@ public:
   sparse_diagonal_product_evaluator(const SparseXprType &sparseXpr, const DiagonalCoeffType &diagCoeff)
     : m_sparseXprImpl(sparseXpr), m_diagCoeffImpl(diagCoeff)
   {}
+
+  Index nonZerosEstimate() const { return m_sparseXprImpl.nonZerosEstimate(); }
     
 protected:
   evaluator<SparseXprType> m_sparseXprImpl;
@@ -121,6 +123,8 @@ struct sparse_diagonal_product_evaluator<SparseXprType, DiagCoeffType, SDP_AsCwi
   sparse_diagonal_product_evaluator(const SparseXprType &sparseXpr, const DiagCoeffType &diagCoeff)
     : m_sparseXprEval(sparseXpr), m_diagCoeffNested(diagCoeff)
   {}
+
+  Index nonZerosEstimate() const { return m_sparseXprEval.nonZerosEstimate(); }
     
 protected:
   evaluator<SparseXprType> m_sparseXprEval;

Eigen/src/SparseCore/SparseMatrix.h

@@ -32,18 +32,22 @@ namespace Eigen {
   * \tparam _Scalar the scalar type, i.e. the type of the coefficients
   * \tparam _Options Union of bit flags controlling the storage scheme. Currently the only possibility
   *                 is ColMajor or RowMajor. The default is 0 which means column-major.
-  * \tparam _Index the type of the indices. It has to be a \b signed type (e.g., short, int, std::ptrdiff_t). Default is \c int.
+  * \tparam _StorageIndex the type of the indices. It has to be a \b signed type (e.g., short, int, std::ptrdiff_t). Default is \c int.
+  *
+  * \warning In %Eigen 3.2, the undocumented type \c SparseMatrix::Index was improperly defined as the storage index type (e.g., int),
+  *          whereas it is now (starting from %Eigen 3.3) deprecated and always defined as Eigen::Index.
+  *          Codes making use of \c SparseMatrix::Index, might thus likely have to be changed to use \c SparseMatrix::StorageIndex instead.
   *
   * 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_SPARSEMATRIX_PLUGIN.
   */
 
 namespace internal {
-template<typename _Scalar, int _Options, typename _Index>
-struct traits<SparseMatrix<_Scalar, _Options, _Index> >
+template<typename _Scalar, int _Options, typename _StorageIndex>
+struct traits<SparseMatrix<_Scalar, _Options, _StorageIndex> >
 {
   typedef _Scalar Scalar;
-  typedef _Index StorageIndex;
+  typedef _StorageIndex StorageIndex;
   typedef Sparse StorageKind;
   typedef MatrixXpr XprKind;
   enum {
@@ -56,16 +60,16 @@ struct traits<SparseMatrix<_Scalar, _Options, _Index> >
   };
 };
 
-template<typename _Scalar, int _Options, typename _Index, int DiagIndex>
-struct traits<Diagonal<SparseMatrix<_Scalar, _Options, _Index>, DiagIndex> >
+template<typename _Scalar, int _Options, typename _StorageIndex, int DiagIndex>
+struct traits<Diagonal<SparseMatrix<_Scalar, _Options, _StorageIndex>, DiagIndex> >
 {
-  typedef SparseMatrix<_Scalar, _Options, _Index> MatrixType;
+  typedef SparseMatrix<_Scalar, _Options, _StorageIndex> MatrixType;
   typedef typename ref_selector<MatrixType>::type MatrixTypeNested;
   typedef typename remove_reference<MatrixTypeNested>::type _MatrixTypeNested;
 
   typedef _Scalar Scalar;
   typedef Dense StorageKind;
-  typedef _Index StorageIndex;
+  typedef _StorageIndex StorageIndex;
   typedef MatrixXpr XprKind;
 
   enum {
@@ -77,9 +81,9 @@ struct traits<Diagonal<SparseMatrix<_Scalar, _Options, _Index>, DiagIndex> >
   };
 };
 
-template<typename _Scalar, int _Options, typename _Index, int DiagIndex>
-struct traits<Diagonal<const SparseMatrix<_Scalar, _Options, _Index>, DiagIndex> >
- : public traits<Diagonal<SparseMatrix<_Scalar, _Options, _Index>, DiagIndex> >
+template<typename _Scalar, int _Options, typename _StorageIndex, int DiagIndex>
+struct traits<Diagonal<const SparseMatrix<_Scalar, _Options, _StorageIndex>, DiagIndex> >
+ : public traits<Diagonal<SparseMatrix<_Scalar, _Options, _StorageIndex>, DiagIndex> >
 {
   enum {
     Flags = 0
@@ -88,13 +92,13 @@ struct traits<Diagonal<const SparseMatrix<_Scalar, _Options, _Index>, DiagIndex>
 
 } // end namespace internal
 
-template<typename _Scalar, int _Options, typename _Index>
+template<typename _Scalar, int _Options, typename _StorageIndex>
 class SparseMatrix
-  : public SparseCompressedBase<SparseMatrix<_Scalar, _Options, _Index> >
+  : public SparseCompressedBase<SparseMatrix<_Scalar, _Options, _StorageIndex> >
 {
     typedef SparseCompressedBase<SparseMatrix> Base;
     using Base::convert_index;
-    friend class SparseVector<_Scalar,0,_Index>;
+    friend class SparseVector<_Scalar,0,_StorageIndex>;
   public:
     using Base::isCompressed;
     using Base::nonZeros;
@@ -786,30 +790,38 @@ class SparseMatrix
       EIGEN_DBG_SPARSE(
         s << "Nonzero entries:\n";
         if(m.isCompressed())
+        {
           for (Index i=0; i<m.nonZeros(); ++i)
             s << "(" << m.m_data.value(i) << "," << m.m_data.index(i) << ") ";
+        }
         else
+        {
           for (Index i=0; i<m.outerSize(); ++i)
           {
             Index p = m.m_outerIndex[i];
             Index pe = m.m_outerIndex[i]+m.m_innerNonZeros[i];
             Index k=p;
-            for (; k<pe; ++k)
+            for (; k<pe; ++k) {
               s << "(" << m.m_data.value(k) << "," << m.m_data.index(k) << ") ";
-            for (; k<m.m_outerIndex[i+1]; ++k)
+            }
+            for (; k<m.m_outerIndex[i+1]; ++k) {
               s << "(_,_) ";
+            }
           }
+        }
         s << std::endl;
         s << std::endl;
         s << "Outer pointers:\n";
-        for (Index i=0; i<m.outerSize(); ++i)
+        for (Index i=0; i<m.outerSize(); ++i) {
           s << m.m_outerIndex[i] << " ";
+        }
         s << " $" << std::endl;
         if(!m.isCompressed())
         {
           s << "Inner non zeros:\n";
-          for (Index i=0; i<m.outerSize(); ++i)
+          for (Index i=0; i<m.outerSize(); ++i) {
             s << m.m_innerNonZeros[i] << " ";
+          }
           s << " $" << std::endl;
         }
         s << std::endl;
@@ -976,11 +988,11 @@ void set_from_triplets(const InputIterator& begin, const InputIterator& end, Spa
   * an abstract iterator over a complex data-structure that would be expensive to evaluate. The triplets should rather
   * be explicitely stored into a std::vector for instance.
   */
-template<typename Scalar, int _Options, typename _Index>
+template<typename Scalar, int _Options, typename _StorageIndex>
 template<typename InputIterators>
-void SparseMatrix<Scalar,_Options,_Index>::setFromTriplets(const InputIterators& begin, const InputIterators& end)
+void SparseMatrix<Scalar,_Options,_StorageIndex>::setFromTriplets(const InputIterators& begin, const InputIterators& end)
 {
-  internal::set_from_triplets<InputIterators, SparseMatrix<Scalar,_Options,_Index> >(begin, end, *this, internal::scalar_sum_op<Scalar,Scalar>());
+  internal::set_from_triplets<InputIterators, SparseMatrix<Scalar,_Options,_StorageIndex> >(begin, end, *this, internal::scalar_sum_op<Scalar,Scalar>());
 }
 
 /** The same as setFromTriplets but when duplicates are met the functor \a dup_func is applied:
@@ -992,17 +1004,17 @@ void SparseMatrix<Scalar,_Options,_Index>::setFromTriplets(const InputIterators&
   * mat.setFromTriplets(triplets.begin(), triplets.end(), [] (const Scalar&,const Scalar &b) { return b; });
   * \endcode
   */
-template<typename Scalar, int _Options, typename _Index>
+template<typename Scalar, int _Options, typename _StorageIndex>
 template<typename InputIterators,typename DupFunctor>
-void SparseMatrix<Scalar,_Options,_Index>::setFromTriplets(const InputIterators& begin, const InputIterators& end, DupFunctor dup_func)
+void SparseMatrix<Scalar,_Options,_StorageIndex>::setFromTriplets(const InputIterators& begin, const InputIterators& end, DupFunctor dup_func)
 {
-  internal::set_from_triplets<InputIterators, SparseMatrix<Scalar,_Options,_Index>, DupFunctor>(begin, end, *this, dup_func);
+  internal::set_from_triplets<InputIterators, SparseMatrix<Scalar,_Options,_StorageIndex>, DupFunctor>(begin, end, *this, dup_func);
 }
 
 /** \internal */
-template<typename Scalar, int _Options, typename _Index>
+template<typename Scalar, int _Options, typename _StorageIndex>
 template<typename DupFunctor>
-void SparseMatrix<Scalar,_Options,_Index>::collapseDuplicates(DupFunctor dup_func)
+void SparseMatrix<Scalar,_Options,_StorageIndex>::collapseDuplicates(DupFunctor dup_func)
 {
   eigen_assert(!isCompressed());
   // TODO, in practice we should be able to use m_innerNonZeros for that task
@@ -1040,9 +1052,9 @@ void SparseMatrix<Scalar,_Options,_Index>::collapseDuplicates(DupFunctor dup_fun
   m_data.resize(m_outerIndex[m_outerSize]);
 }
 
-template<typename Scalar, int _Options, typename _Index>
+template<typename Scalar, int _Options, typename _StorageIndex>
 template<typename OtherDerived>
-EIGEN_DONT_INLINE SparseMatrix<Scalar,_Options,_Index>& SparseMatrix<Scalar,_Options,_Index>::operator=(const SparseMatrixBase<OtherDerived>& other)
+EIGEN_DONT_INLINE SparseMatrix<Scalar,_Options,_StorageIndex>& SparseMatrix<Scalar,_Options,_StorageIndex>::operator=(const SparseMatrixBase<OtherDerived>& other)
 {
   EIGEN_STATIC_ASSERT((internal::is_same<Scalar, typename OtherDerived::Scalar>::value),
         YOU_MIXED_DIFFERENT_NUMERIC_TYPES__YOU_NEED_TO_USE_THE_CAST_METHOD_OF_MATRIXBASE_TO_CAST_NUMERIC_TYPES_EXPLICITLY)
@@ -1113,8 +1125,8 @@ EIGEN_DONT_INLINE SparseMatrix<Scalar,_Options,_Index>& SparseMatrix<Scalar,_Opt
   }
 }
 
-template<typename _Scalar, int _Options, typename _Index>
-typename SparseMatrix<_Scalar,_Options,_Index>::Scalar& SparseMatrix<_Scalar,_Options,_Index>::insert(Index row, Index col)
+template<typename _Scalar, int _Options, typename _StorageIndex>
+typename SparseMatrix<_Scalar,_Options,_StorageIndex>::Scalar& SparseMatrix<_Scalar,_Options,_StorageIndex>::insert(Index row, Index col)
 {
   eigen_assert(row>=0 && row<rows() && col>=0 && col<cols());
   
@@ -1233,8 +1245,8 @@ typename SparseMatrix<_Scalar,_Options,_Index>::Scalar& SparseMatrix<_Scalar,_Op
   return insertUncompressed(row,col);
 }
     
-template<typename _Scalar, int _Options, typename _Index>
-EIGEN_DONT_INLINE typename SparseMatrix<_Scalar,_Options,_Index>::Scalar& SparseMatrix<_Scalar,_Options,_Index>::insertUncompressed(Index row, Index col)
+template<typename _Scalar, int _Options, typename _StorageIndex>
+EIGEN_DONT_INLINE typename SparseMatrix<_Scalar,_Options,_StorageIndex>::Scalar& SparseMatrix<_Scalar,_Options,_StorageIndex>::insertUncompressed(Index row, Index col)
 {
   eigen_assert(!isCompressed());
 
@@ -1265,8 +1277,8 @@ EIGEN_DONT_INLINE typename SparseMatrix<_Scalar,_Options,_Index>::Scalar& Sparse
   return (m_data.value(p) = 0);
 }
 
-template<typename _Scalar, int _Options, typename _Index>
-EIGEN_DONT_INLINE typename SparseMatrix<_Scalar,_Options,_Index>::Scalar& SparseMatrix<_Scalar,_Options,_Index>::insertCompressed(Index row, Index col)
+template<typename _Scalar, int _Options, typename _StorageIndex>
+EIGEN_DONT_INLINE typename SparseMatrix<_Scalar,_Options,_StorageIndex>::Scalar& SparseMatrix<_Scalar,_Options,_StorageIndex>::insertCompressed(Index row, Index col)
 {
   eigen_assert(isCompressed());
 
@@ -1289,11 +1301,11 @@ EIGEN_DONT_INLINE typename SparseMatrix<_Scalar,_Options,_Index>::Scalar& Sparse
   // starts with: [ 0 0 0 0 0 1 ...] and we are inserted in, e.g.,
   // the 2nd inner vector...
   bool isLastVec = (!(previousOuter==-1 && m_data.size()!=0))
-                && (size_t(m_outerIndex[outer+1]) == m_data.size());
+                && (std::size_t(m_outerIndex[outer+1]) == m_data.size());
 
-  size_t startId = m_outerIndex[outer];
-  // FIXME let's make sure sizeof(long int) == sizeof(size_t)
-  size_t p = m_outerIndex[outer+1];
+  std::size_t startId = m_outerIndex[outer];
+  // FIXME let's make sure sizeof(long int) == sizeof(std::size_t)
+  std::size_t p = m_outerIndex[outer+1];
   ++m_outerIndex[outer+1];
 
   double reallocRatio = 1;
@@ -1374,12 +1386,12 @@ EIGEN_DONT_INLINE typename SparseMatrix<_Scalar,_Options,_Index>::Scalar& Sparse
 
 namespace internal {
 
-template<typename _Scalar, int _Options, typename _Index>
-struct evaluator<SparseMatrix<_Scalar,_Options,_Index> >
-  : evaluator<SparseCompressedBase<SparseMatrix<_Scalar,_Options,_Index> > >
+template<typename _Scalar, int _Options, typename _StorageIndex>
+struct evaluator<SparseMatrix<_Scalar,_Options,_StorageIndex> >
+  : evaluator<SparseCompressedBase<SparseMatrix<_Scalar,_Options,_StorageIndex> > >
 {
-  typedef evaluator<SparseCompressedBase<SparseMatrix<_Scalar,_Options,_Index> > > Base;
-  typedef SparseMatrix<_Scalar,_Options,_Index> SparseMatrixType;
+  typedef evaluator<SparseCompressedBase<SparseMatrix<_Scalar,_Options,_StorageIndex> > > Base;
+  typedef SparseMatrix<_Scalar,_Options,_StorageIndex> SparseMatrixType;
   evaluator() : Base() {}
   explicit evaluator(const SparseMatrixType &mat) : Base(mat) {}
 };

Eigen/src/SparseCore/SparseMatrixBase.h

@@ -37,7 +37,11 @@ template<typename Derived> class SparseMatrixBase
     
     typedef typename internal::packet_traits<Scalar>::type PacketScalar;
     typedef typename internal::traits<Derived>::StorageKind StorageKind;
+
+    /** The integer type used to \b store indices within a SparseMatrix.
+      * For a \c SparseMatrix<Scalar,Options,IndexType> it an alias of the third template parameter \c IndexType. */
     typedef typename internal::traits<Derived>::StorageIndex StorageIndex;
+
     typedef typename internal::add_const_on_value_type_if_arithmetic<
                          typename internal::packet_traits<Scalar>::type
                      >::type PacketReturnType;
@@ -213,7 +217,7 @@ template<typename Derived> class SparseMatrixBase
 
       if (Flags&RowMajorBit)
       {
-        const Nested nm(m.derived());
+        Nested nm(m.derived());
         internal::evaluator<NestedCleaned> thisEval(nm);
         for (Index row=0; row<nm.outerSize(); ++row)
         {
@@ -232,7 +236,7 @@ template<typename Derived> class SparseMatrixBase
       }
       else
       {
-        const Nested nm(m.derived());
+        Nested nm(m.derived());
         internal::evaluator<NestedCleaned> thisEval(nm);
         if (m.cols() == 1) {
           Index row = 0;
@@ -265,6 +269,11 @@ template<typename Derived> class SparseMatrixBase
     template<typename OtherDerived>
     Derived& operator-=(const DiagonalBase<OtherDerived>& other);
 
+    template<typename OtherDerived>
+    Derived& operator+=(const EigenBase<OtherDerived> &other);
+    template<typename OtherDerived>
+    Derived& operator-=(const EigenBase<OtherDerived> &other);
+
     Derived& operator*=(const Scalar& other);
     Derived& operator/=(const Scalar& other);
 

Eigen/src/SparseCore/SparseRef.h

@@ -185,20 +185,27 @@ class Ref<const SparseMatrix<MatScalar,MatOptions,MatIndex>, Options, StrideType
     EIGEN_SPARSE_PUBLIC_INTERFACE(Ref)
 
     template<typename Derived>
-    inline Ref(const SparseMatrixBase<Derived>& expr)
+    inline Ref(const SparseMatrixBase<Derived>& expr) : m_hasCopy(false)
     {
       construct(expr.derived(), typename Traits::template match<Derived>::type());
     }
 
-    inline Ref(const Ref& other) : Base(other) {
+    inline Ref(const Ref& other) : Base(other), m_hasCopy(false) {
       // copy constructor shall not copy the m_object, to avoid unnecessary malloc and copy
     }
 
     template<typename OtherRef>
-    inline Ref(const RefBase<OtherRef>& other) {
+    inline Ref(const RefBase<OtherRef>& other) : m_hasCopy(false) {
       construct(other.derived(), typename Traits::template match<OtherRef>::type());
     }
 
+    ~Ref() {
+      if(m_hasCopy) {
+        TPlainObjectType* obj = reinterpret_cast<TPlainObjectType*>(m_object_bytes);
+        obj->~TPlainObjectType();
+      }
+    }
+
   protected:
 
     template<typename Expression>
@@ -208,6 +215,7 @@ class Ref<const SparseMatrix<MatScalar,MatOptions,MatIndex>, Options, StrideType
       {
         TPlainObjectType* obj = reinterpret_cast<TPlainObjectType*>(m_object_bytes);
         ::new (obj) TPlainObjectType(expr);
+        m_hasCopy = true;
         Base::construct(*obj);
       }
       else
@@ -221,11 +229,13 @@ class Ref<const SparseMatrix<MatScalar,MatOptions,MatIndex>, Options, StrideType
     {
       TPlainObjectType* obj = reinterpret_cast<TPlainObjectType*>(m_object_bytes);
       ::new (obj) TPlainObjectType(expr);
+      m_hasCopy = true;
       Base::construct(*obj);
     }
 
   protected:
     char m_object_bytes[sizeof(TPlainObjectType)];
+    bool m_hasCopy;
 };
 
 
@@ -293,20 +303,27 @@ class Ref<const SparseVector<MatScalar,MatOptions,MatIndex>, Options, StrideType
     EIGEN_SPARSE_PUBLIC_INTERFACE(Ref)
 
     template<typename Derived>
-    inline Ref(const SparseMatrixBase<Derived>& expr)
+    inline Ref(const SparseMatrixBase<Derived>& expr) : m_hasCopy(false)
     {
       construct(expr.derived(), typename Traits::template match<Derived>::type());
     }
 
-    inline Ref(const Ref& other) : Base(other) {
+    inline Ref(const Ref& other) : Base(other), m_hasCopy(false) {
       // copy constructor shall not copy the m_object, to avoid unnecessary malloc and copy
     }
 
     template<typename OtherRef>
-    inline Ref(const RefBase<OtherRef>& other) {
+    inline Ref(const RefBase<OtherRef>& other) : m_hasCopy(false) {
       construct(other.derived(), typename Traits::template match<OtherRef>::type());
     }
 
+    ~Ref() {
+      if(m_hasCopy) {
+        TPlainObjectType* obj = reinterpret_cast<TPlainObjectType*>(m_object_bytes);
+        obj->~TPlainObjectType();
+      }
+    }
+
   protected:
 
     template<typename Expression>
@@ -320,11 +337,13 @@ class Ref<const SparseVector<MatScalar,MatOptions,MatIndex>, Options, StrideType
     {
       TPlainObjectType* obj = reinterpret_cast<TPlainObjectType*>(m_object_bytes);
       ::new (obj) TPlainObjectType(expr);
+      m_hasCopy = true;
       Base::construct(*obj);
     }
 
   protected:
     char m_object_bytes[sizeof(TPlainObjectType)];
+    bool m_hasCopy;
 };
 
 namespace internal {

Eigen/src/SparseCore/SparseSelfAdjointView.h

@@ -47,6 +47,7 @@ template<typename MatrixType, unsigned int _Mode> class SparseSelfAdjointView
     
     enum {
       Mode = _Mode,
+      TransposeMode = ((Mode & Upper) ? Lower : 0) | ((Mode & Lower) ? Upper : 0),
       RowsAtCompileTime = internal::traits<SparseSelfAdjointView>::RowsAtCompileTime,
       ColsAtCompileTime = internal::traits<SparseSelfAdjointView>::ColsAtCompileTime
     };
@@ -222,14 +223,43 @@ template< typename DstXprType, typename SrcXprType, typename Functor>
 struct Assignment<DstXprType, SrcXprType, Functor, SparseSelfAdjoint2Sparse>
 {
   typedef typename DstXprType::StorageIndex StorageIndex;
+  typedef internal::assign_op<typename DstXprType::Scalar,typename SrcXprType::Scalar> AssignOpType;
+
   template<typename DestScalar,int StorageOrder>
-  static void run(SparseMatrix<DestScalar,StorageOrder,StorageIndex> &dst, const SrcXprType &src, const internal::assign_op<typename DstXprType::Scalar,typename SrcXprType::Scalar> &/*func*/)
+  static void run(SparseMatrix<DestScalar,StorageOrder,StorageIndex> &dst, const SrcXprType &src, const AssignOpType&/*func*/)
   {
     internal::permute_symm_to_fullsymm<SrcXprType::Mode>(src.matrix(), dst);
   }
+
+  // FIXME: the handling of += and -= in sparse matrices should be cleanup so that next two overloads could be reduced to:
+  template<typename DestScalar,int StorageOrder,typename AssignFunc>
+  static void run(SparseMatrix<DestScalar,StorageOrder,StorageIndex> &dst, const SrcXprType &src, const AssignFunc& func)
+  {
+    SparseMatrix<DestScalar,StorageOrder,StorageIndex> tmp(src.rows(),src.cols());
+    run(tmp, src, AssignOpType());
+    call_assignment_no_alias_no_transpose(dst, tmp, func);
+  }
+
+  template<typename DestScalar,int StorageOrder>
+  static void run(SparseMatrix<DestScalar,StorageOrder,StorageIndex> &dst, const SrcXprType &src,
+                  const internal::add_assign_op<typename DstXprType::Scalar,typename SrcXprType::Scalar>& /* func */)
+  {
+    SparseMatrix<DestScalar,StorageOrder,StorageIndex> tmp(src.rows(),src.cols());
+    run(tmp, src, AssignOpType());
+    dst += tmp;
+  }
+
+  template<typename DestScalar,int StorageOrder>
+  static void run(SparseMatrix<DestScalar,StorageOrder,StorageIndex> &dst, const SrcXprType &src,
+                  const internal::sub_assign_op<typename DstXprType::Scalar,typename SrcXprType::Scalar>& /* func */)
+  {
+    SparseMatrix<DestScalar,StorageOrder,StorageIndex> tmp(src.rows(),src.cols());
+    run(tmp, src, AssignOpType());
+    dst -= tmp;
+  }
   
   template<typename DestScalar>
-  static void run(DynamicSparseMatrix<DestScalar,ColMajor,StorageIndex>& dst, const SrcXprType &src, const internal::assign_op<typename DstXprType::Scalar,typename SrcXprType::Scalar> &/*func*/)
+  static void run(DynamicSparseMatrix<DestScalar,ColMajor,StorageIndex>& dst, const SrcXprType &src, const AssignOpType&/*func*/)
   {
     // TODO directly evaluate into dst;
     SparseMatrix<DestScalar,ColMajor,StorageIndex> tmp(dst.rows(),dst.cols());
@@ -339,7 +369,7 @@ struct generic_product_impl<Lhs, RhsView, DenseShape, SparseSelfAdjointShape, Pr
     
     // transpose everything
     Transpose<Dest> dstT(dst);
-    internal::sparse_selfadjoint_time_dense_product<RhsView::Mode>(rhsNested.transpose(), lhsNested.transpose(), dstT, alpha);
+    internal::sparse_selfadjoint_time_dense_product<RhsView::TransposeMode>(rhsNested.transpose(), lhsNested.transpose(), dstT, alpha);
   }
 };
 

Eigen/src/SparseCore/SparseTranspose.h

@@ -56,7 +56,6 @@ struct unary_evaluator<Transpose<ArgType>, IteratorBased>
   : public evaluator_base<Transpose<ArgType> >
 {
     typedef typename evaluator<ArgType>::InnerIterator        EvalIterator;
-    typedef typename evaluator<ArgType>::ReverseInnerIterator EvalReverseIterator;
   public:
     typedef Transpose<ArgType> XprType;
     
@@ -75,17 +74,6 @@ struct unary_evaluator<Transpose<ArgType>, IteratorBased>
       Index col() const { return EvalIterator::row(); }
     };
     
-    class ReverseInnerIterator : public EvalReverseIterator
-    {
-    public:
-      EIGEN_STRONG_INLINE ReverseInnerIterator(const unary_evaluator& unaryOp, Index outer)
-        : EvalReverseIterator(unaryOp.m_argImpl,outer)
-      {}
-      
-      Index row() const { return EvalReverseIterator::col(); }
-      Index col() const { return EvalReverseIterator::row(); }
-    };
-    
     enum {
       CoeffReadCost = evaluator<ArgType>::CoeffReadCost,
       Flags = XprType::Flags

Eigen/src/SparseCore/SparseTriangularView.h

@@ -55,7 +55,10 @@ template<typename MatrixType, unsigned int Mode> class TriangularViewImpl<Matrix
       this->solveInPlace(dst);
     }
 
+    /** Applies the inverse of \c *this to the dense vector or matrix \a other, "in-place" */
     template<typename OtherDerived> void solveInPlace(MatrixBase<OtherDerived>& other) const;
+
+    /** Applies the inverse of \c *this to the sparse vector or matrix \a other, "in-place" */
     template<typename OtherDerived> void solveInPlace(SparseMatrixBase<OtherDerived>& other) const;
   
 };

Eigen/src/SparseCore/SparseView.h

@@ -27,6 +27,20 @@ struct traits<SparseView<MatrixType> > : traits<MatrixType>
 
 } // end namespace internal
 
+/** \ingroup SparseCore_Module
+  * \class SparseView
+  *
+  * \brief Expression of a dense or sparse matrix with zero or too small values removed
+  *
+  * \tparam MatrixType the type of the object of which we are removing the small entries
+  *
+  * This class represents an expression of a given dense or sparse matrix with
+  * entries smaller than \c reference * \c epsilon are removed.
+  * It is the return type of MatrixBase::sparseView() and SparseMatrixBase::pruned()
+  * and most of the time this is the only way it is used.
+  *
+  * \sa MatrixBase::sparseView(), SparseMatrixBase::pruned()
+  */
 template<typename MatrixType>
 class SparseView : public SparseMatrixBase<SparseView<MatrixType> >
 {
@@ -190,6 +204,23 @@ struct unary_evaluator<SparseView<ArgType>, IndexBased>
 
 } // end namespace internal
 
+/** \ingroup SparseCore_Module
+  *
+  * \returns a sparse expression of the dense expression \c *this with values smaller than
+  * \a reference * \a epsilon removed.
+  *
+  * This method is typically used when prototyping to convert a quickly assembled dense Matrix \c D to a SparseMatrix \c S:
+  * \code
+  * MatrixXd D(n,m);
+  * SparseMatrix<double> S;
+  * S = D.sparseView();             // suppress numerical zeros (exact)
+  * S = D.sparseView(reference);
+  * S = D.sparseView(reference,epsilon);
+  * \endcode
+  * where \a reference is a meaningful non zero reference value,
+  * and \a epsilon is a tolerance factor defaulting to NumTraits<Scalar>::dummy_precision().
+  *
+  * \sa SparseMatrixBase::pruned(), class SparseView */
 template<typename Derived>
 const SparseView<Derived> MatrixBase<Derived>::sparseView(const Scalar& reference,
                                                           const typename NumTraits<Scalar>::Real& epsilon) const
@@ -198,7 +229,7 @@ const SparseView<Derived> MatrixBase<Derived>::sparseView(const Scalar& referenc
 }
 
 /** \returns an expression of \c *this with values smaller than
-  * \a reference * \a epsilon are removed.
+  * \a reference * \a epsilon removed.
   *
   * This method is typically used in conjunction with the product of two sparse matrices
   * to automatically prune the smallest values as follows:

Eigen/src/SparseCore/TriangularSolver.h

@@ -171,6 +171,8 @@ struct sparse_solve_triangular_selector<Lhs,Rhs,Mode,Upper,ColMajor>
 
 } // end namespace internal
 
+#ifndef EIGEN_PARSED_BY_DOXYGEN
+
 template<typename ExpressionType,unsigned int Mode>
 template<typename OtherDerived>
 void TriangularViewImpl<ExpressionType,Mode,Sparse>::solveInPlace(MatrixBase<OtherDerived>& other) const
@@ -189,6 +191,7 @@ void TriangularViewImpl<ExpressionType,Mode,Sparse>::solveInPlace(MatrixBase<Oth
   if (copy)
     other = otherCopy;
 }
+#endif
 
 // pure sparse path
 
@@ -286,6 +289,7 @@ struct sparse_solve_triangular_sparse_selector<Lhs,Rhs,Mode,UpLo,ColMajor>
 
 } // end namespace internal
 
+#ifndef EIGEN_PARSED_BY_DOXYGEN
 template<typename ExpressionType,unsigned int Mode>
 template<typename OtherDerived>
 void TriangularViewImpl<ExpressionType,Mode,Sparse>::solveInPlace(SparseMatrixBase<OtherDerived>& other) const
@@ -304,6 +308,7 @@ void TriangularViewImpl<ExpressionType,Mode,Sparse>::solveInPlace(SparseMatrixBa
 //   if (copy)
 //     other = otherCopy;
 }
+#endif
 
 } // end namespace Eigen
 

Eigen/src/SparseLU/SparseLU.h

@@ -748,7 +748,7 @@ struct SparseLUMatrixUReturnType : internal::no_assignment_operator
       else
       {
         Map<const Matrix<Scalar,Dynamic,Dynamic, ColMajor>, 0, OuterStride<> > A( &(m_mapL.valuePtr()[luptr]), nsupc, nsupc, OuterStride<>(lda) );
-        Map< Matrix<Scalar,Dynamic,Dynamic, ColMajor>, 0, OuterStride<> > U (&(X(fsupc,0)), nsupc, nrhs, OuterStride<>(n) );
+        Map< Matrix<Scalar,Dynamic,Dest::ColsAtCompileTime, ColMajor>, 0, OuterStride<> > U (&(X(fsupc,0)), nsupc, nrhs, OuterStride<>(n) );
         U = A.template triangularView<Upper>().solve(U);
       }
 

Eigen/src/SparseLU/SparseLU_SupernodalMatrix.h

@@ -239,7 +239,7 @@ void MappedSuperNodalMatrix<Scalar,Index_>::solveInPlace( MatrixBase<Dest>&X) co
     Index n    = int(X.rows());
     Index nrhs = Index(X.cols());
     const Scalar * Lval = valuePtr();                 // Nonzero values 
-    Matrix<Scalar,Dynamic,Dynamic, ColMajor> work(n, nrhs);     // working vector
+    Matrix<Scalar,Dynamic,Dest::ColsAtCompileTime, ColMajor> work(n, nrhs);     // working vector
     work.setZero();
     for (Index k = 0; k <= nsuper(); k ++)
     {
@@ -271,12 +271,12 @@ void MappedSuperNodalMatrix<Scalar,Index_>::solveInPlace( MatrixBase<Dest>&X) co
         
         // Triangular solve 
         Map<const Matrix<Scalar,Dynamic,Dynamic, ColMajor>, 0, OuterStride<> > A( &(Lval[luptr]), nsupc, nsupc, OuterStride<>(lda) );
-        Map< Matrix<Scalar,Dynamic,Dynamic, ColMajor>, 0, OuterStride<> > U (&(X(fsupc,0)), nsupc, nrhs, OuterStride<>(n) ); 
+        Map< Matrix<Scalar,Dynamic,Dest::ColsAtCompileTime, ColMajor>, 0, OuterStride<> > U (&(X(fsupc,0)), nsupc, nrhs, OuterStride<>(n) );
         U = A.template triangularView<UnitLower>().solve(U); 
         
         // Matrix-vector product 
         new (&A) Map<const Matrix<Scalar,Dynamic,Dynamic, ColMajor>, 0, OuterStride<> > ( &(Lval[luptr+nsupc]), nrow, nsupc, OuterStride<>(lda) );
-        work.block(0, 0, nrow, nrhs) = A * U; 
+        work.topRows(nrow).noalias() = A * U;
         
         //Begin Scatter 
         for (Index j = 0; j < nrhs; j++)

Eigen/src/SparseLU/SparseLU_gemm_kernel.h

@@ -106,22 +106,22 @@ void sparselu_gemm(Index m, Index n, Index d, const Scalar* A, Index lda, const
         
 #define KMADD(c, a, b, tmp) {tmp = b; tmp = pmul(a,tmp); c = padd(c,tmp);}
 #define WORK(I)  \
-                    c0 = pload<Packet>(C0+i+(I)*PacketSize);   \
-                    c1 = pload<Packet>(C1+i+(I)*PacketSize);   \
-                    KMADD(c0, a0, b00, t0)      \
-                    KMADD(c1, a0, b01, t1)      \
-                    a0 = pload<Packet>(A0+i+(I+1)*PacketSize); \
-                    KMADD(c0, a1, b10, t0)      \
-                    KMADD(c1, a1, b11, t1)       \
-                    a1 = pload<Packet>(A1+i+(I+1)*PacketSize); \
-          if(RK==4) KMADD(c0, a2, b20, t0)       \
-          if(RK==4) KMADD(c1, a2, b21, t1)       \
-          if(RK==4) a2 = pload<Packet>(A2+i+(I+1)*PacketSize); \
-          if(RK==4) KMADD(c0, a3, b30, t0)       \
-          if(RK==4) KMADD(c1, a3, b31, t1)       \
-          if(RK==4) a3 = pload<Packet>(A3+i+(I+1)*PacketSize); \
-                    pstore(C0+i+(I)*PacketSize, c0);           \
-                    pstore(C1+i+(I)*PacketSize, c1)
+                     c0 = pload<Packet>(C0+i+(I)*PacketSize);    \
+                     c1 = pload<Packet>(C1+i+(I)*PacketSize);    \
+                     KMADD(c0, a0, b00, t0)                      \
+                     KMADD(c1, a0, b01, t1)                      \
+                     a0 = pload<Packet>(A0+i+(I+1)*PacketSize);  \
+                     KMADD(c0, a1, b10, t0)                      \
+                     KMADD(c1, a1, b11, t1)                      \
+                     a1 = pload<Packet>(A1+i+(I+1)*PacketSize);  \
+          if(RK==4){ KMADD(c0, a2, b20, t0)                     }\
+          if(RK==4){ KMADD(c1, a2, b21, t1)                     }\
+          if(RK==4){ a2 = pload<Packet>(A2+i+(I+1)*PacketSize); }\
+          if(RK==4){ KMADD(c0, a3, b30, t0)                     }\
+          if(RK==4){ KMADD(c1, a3, b31, t1)                     }\
+          if(RK==4){ a3 = pload<Packet>(A3+i+(I+1)*PacketSize); }\
+                     pstore(C0+i+(I)*PacketSize, c0);            \
+                     pstore(C1+i+(I)*PacketSize, c1)
         
         // process rows of A' - C' with aggressive vectorization and peeling 
         for(Index i=0; i<actual_b_end1; i+=PacketSize*8)
@@ -131,14 +131,15 @@ void sparselu_gemm(Index m, Index n, Index d, const Scalar* A, Index lda, const
                     prefetch((A1+i+(5)*PacketSize));
           if(RK==4) prefetch((A2+i+(5)*PacketSize));
           if(RK==4) prefetch((A3+i+(5)*PacketSize));
-                    WORK(0);
-                    WORK(1);
-                    WORK(2);
-                    WORK(3);
-                    WORK(4);
-                    WORK(5);
-                    WORK(6);
-                    WORK(7);
+
+          WORK(0);
+          WORK(1);
+          WORK(2);
+          WORK(3);
+          WORK(4);
+          WORK(5);
+          WORK(6);
+          WORK(7);
         }
         // process the remaining rows with vectorization only
         for(Index i=actual_b_end1; i<actual_b_end2; i+=PacketSize)
@@ -203,16 +204,16 @@ void sparselu_gemm(Index m, Index n, Index d, const Scalar* A, Index lda, const
         }
         
 #define WORK(I) \
-                  c0 = pload<Packet>(C0+i+(I)*PacketSize);   \
-                  KMADD(c0, a0, b00, t0)       \
-                  a0 = pload<Packet>(A0+i+(I+1)*PacketSize); \
-                  KMADD(c0, a1, b10, t0)       \
-                  a1 = pload<Packet>(A1+i+(I+1)*PacketSize); \
-        if(RK==4) KMADD(c0, a2, b20, t0)       \
-        if(RK==4) a2 = pload<Packet>(A2+i+(I+1)*PacketSize); \
-        if(RK==4) KMADD(c0, a3, b30, t0)       \
-        if(RK==4) a3 = pload<Packet>(A3+i+(I+1)*PacketSize); \
-                  pstore(C0+i+(I)*PacketSize, c0);
+                   c0 = pload<Packet>(C0+i+(I)*PacketSize);     \
+                   KMADD(c0, a0, b00, t0)                       \
+                   a0 = pload<Packet>(A0+i+(I+1)*PacketSize);   \
+                   KMADD(c0, a1, b10, t0)                       \
+                   a1 = pload<Packet>(A1+i+(I+1)*PacketSize);   \
+        if(RK==4){ KMADD(c0, a2, b20, t0)                      }\
+        if(RK==4){ a2 = pload<Packet>(A2+i+(I+1)*PacketSize);  }\
+        if(RK==4){ KMADD(c0, a3, b30, t0)                      }\
+        if(RK==4){ a3 = pload<Packet>(A3+i+(I+1)*PacketSize);  }\
+                   pstore(C0+i+(I)*PacketSize, c0);
         
         // agressive vectorization and peeling
         for(Index i=0; i<actual_b_end1; i+=PacketSize*8)

Eigen/src/StlSupport/details.h

@@ -22,13 +22,13 @@ namespace Eigen {
   class aligned_allocator_indirection : public EIGEN_ALIGNED_ALLOCATOR<T>
   {
   public:
-    typedef size_t    size_type;
-    typedef ptrdiff_t difference_type;
-    typedef T*        pointer;
-    typedef const T*  const_pointer;
-    typedef T&        reference;
-    typedef const T&  const_reference;
-    typedef T         value_type;
+    typedef std::size_t     size_type;
+    typedef std::ptrdiff_t  difference_type;
+    typedef T*              pointer;
+    typedef const T*        const_pointer;
+    typedef T&              reference;
+    typedef const T&        const_reference;
+    typedef T               value_type;
 
     template<class U>
     struct rebind

Eigen/src/SuperLUSupport/SuperLUSupport.h

@@ -967,6 +967,7 @@ void SuperILU<MatrixType>::factorize(const MatrixType& a)
   m_factorizationIsOk = true;
 }
 
+#ifndef EIGEN_PARSED_BY_DOXYGEN
 template<typename MatrixType>
 template<typename Rhs,typename Dest>
 void SuperILU<MatrixType>::_solve_impl(const MatrixBase<Rhs> &b, MatrixBase<Dest>& x) const
@@ -1019,6 +1020,8 @@ void SuperILU<MatrixType>::_solve_impl(const MatrixBase<Rhs> &b, MatrixBase<Dest
 }
 #endif
 
+#endif
+
 } // end namespace Eigen
 
 #endif // EIGEN_SUPERLUSUPPORT_H

Eigen/src/UmfPackSupport/UmfPackSupport.h

@@ -10,19 +10,37 @@
 #ifndef EIGEN_UMFPACKSUPPORT_H
 #define EIGEN_UMFPACKSUPPORT_H
 
-namespace Eigen { 
+namespace Eigen {
 
 /* TODO extract L, extract U, compute det, etc... */
 
 // generic double/complex<double> wrapper functions:
 
 
-inline void umfpack_defaults(double control[UMFPACK_CONTROL], double) 
+inline void umfpack_defaults(double control[UMFPACK_CONTROL], double)
 { umfpack_di_defaults(control); }
 
-inline void umfpack_defaults(double control[UMFPACK_CONTROL], std::complex<double>) 
+inline void umfpack_defaults(double control[UMFPACK_CONTROL], std::complex<double>)
 { umfpack_zi_defaults(control); }
 
+inline void umfpack_report_info(double control[UMFPACK_CONTROL], double info[UMFPACK_INFO], double)
+{ umfpack_di_report_info(control, info);}
+
+inline void umfpack_report_info(double control[UMFPACK_CONTROL], double info[UMFPACK_INFO], std::complex<double>)
+{ umfpack_zi_report_info(control, info);}
+
+inline void umfpack_report_status(double control[UMFPACK_CONTROL], int status, double)
+{ umfpack_di_report_status(control, status);}
+
+inline void umfpack_report_status(double control[UMFPACK_CONTROL], int status, std::complex<double>)
+{ umfpack_zi_report_status(control, status);}
+
+inline void umfpack_report_control(double control[UMFPACK_CONTROL], double)
+{ umfpack_di_report_control(control);}
+
+inline void umfpack_report_control(double control[UMFPACK_CONTROL], std::complex<double>)
+{ umfpack_zi_report_control(control);}
+
 inline void umfpack_free_numeric(void **Numeric, double)
 { umfpack_di_free_numeric(Numeric); *Numeric = 0; }
 
@@ -156,6 +174,7 @@ class UmfPackLU : public SparseSolverBase<UmfPackLU<_MatrixType> >
   public:
 
     typedef Array<double, UMFPACK_CONTROL, 1> UmfpackControl;
+    typedef Array<double, UMFPACK_INFO, 1> UmfpackInfo;
 
     UmfPackLU()
       : m_dummy(0,0), mp_matrix(m_dummy)
@@ -215,7 +234,7 @@ class UmfPackLU : public SparseSolverBase<UmfPackLU<_MatrixType> >
       return m_q;
     }
 
-    /** Computes the sparse Cholesky decomposition of \a matrix 
+    /** Computes the sparse Cholesky decomposition of \a matrix
      *  Note that the matrix should be column-major, and in compressed format for best performance.
      *  \sa SparseMatrix::makeCompressed().
      */
@@ -240,7 +259,7 @@ class UmfPackLU : public SparseSolverBase<UmfPackLU<_MatrixType> >
     {
       if(m_symbolic) umfpack_free_symbolic(&m_symbolic,Scalar());
       if(m_numeric)  umfpack_free_numeric(&m_numeric,Scalar());
-      
+
       grab(matrix.derived());
 
       analyzePattern_impl();
@@ -267,7 +286,7 @@ class UmfPackLU : public SparseSolverBase<UmfPackLU<_MatrixType> >
     {
       return m_control;
     }
-    
+
     /** Provides access to the control settings array used by UmfPack.
       *
       * If this array contains NaN's, the default values are used.
@@ -278,7 +297,7 @@ class UmfPackLU : public SparseSolverBase<UmfPackLU<_MatrixType> >
     {
       return m_control;
     }
-    
+
     /** Performs a numeric decomposition of \a matrix
       *
       * The given matrix must has the same sparcity than the matrix on which the pattern anylysis has been performed.
@@ -293,10 +312,38 @@ class UmfPackLU : public SparseSolverBase<UmfPackLU<_MatrixType> >
         umfpack_free_numeric(&m_numeric,Scalar());
 
       grab(matrix.derived());
-      
+
       factorize_impl();
     }
 
+    /** Prints the current UmfPack control settings.
+      *
+      * \sa umfpackControl()
+      */
+    void umfpackReportControl()
+    {
+      umfpack_report_control(m_control.data(), Scalar());
+    }
+
+    /** Prints statistics collected by UmfPack.
+      *
+      * \sa analyzePattern(), compute()
+      */
+    void umfpackReportInfo()
+    {
+      eigen_assert(m_analysisIsOk && "UmfPackLU: you must first call analyzePattern()");
+      umfpack_report_info(m_control.data(), m_umfpackInfo.data(), Scalar());
+    }
+
+    /** Prints the status of the previous factorization operation performed by UmfPack (symbolic or numerical factorization).
+      *
+      * \sa analyzePattern(), compute()
+      */
+    void umfpackReportStatus() {
+      eigen_assert(m_analysisIsOk && "UmfPackLU: you must first call analyzePattern()");
+      umfpack_report_status(m_control.data(), m_fact_errorCode, Scalar());
+    }
+
     /** \internal */
     template<typename BDerived,typename XDerived>
     bool _solve_impl(const MatrixBase<BDerived> &b, MatrixBase<XDerived> &x) const;
@@ -314,41 +361,42 @@ class UmfPackLU : public SparseSolverBase<UmfPackLU<_MatrixType> >
       m_numeric               = 0;
       m_symbolic              = 0;
       m_extractedDataAreDirty = true;
+
+      umfpack_defaults(m_control.data(), Scalar());
     }
-    
+
     void analyzePattern_impl()
     {
-      umfpack_defaults(m_control.data(), Scalar());
-      int errorCode = 0;
-      errorCode = umfpack_symbolic(internal::convert_index<int>(mp_matrix.rows()),
-                                   internal::convert_index<int>(mp_matrix.cols()),
-                                   mp_matrix.outerIndexPtr(), mp_matrix.innerIndexPtr(), mp_matrix.valuePtr(),
-                                   &m_symbolic, m_control.data(), 0);
+      m_fact_errorCode = umfpack_symbolic(internal::convert_index<int>(mp_matrix.rows()),
+                                          internal::convert_index<int>(mp_matrix.cols()),
+                                          mp_matrix.outerIndexPtr(), mp_matrix.innerIndexPtr(), mp_matrix.valuePtr(),
+                                          &m_symbolic, m_control.data(), m_umfpackInfo.data());
 
       m_isInitialized = true;
-      m_info = errorCode ? InvalidInput : Success;
+      m_info = m_fact_errorCode ? InvalidInput : Success;
       m_analysisIsOk = true;
       m_factorizationIsOk = false;
       m_extractedDataAreDirty = true;
     }
-    
+
     void factorize_impl()
     {
+
       m_fact_errorCode = umfpack_numeric(mp_matrix.outerIndexPtr(), mp_matrix.innerIndexPtr(), mp_matrix.valuePtr(),
-                                         m_symbolic, &m_numeric, m_control.data(), 0);
+                                         m_symbolic, &m_numeric, m_control.data(), m_umfpackInfo.data());
 
       m_info = m_fact_errorCode == UMFPACK_OK ? Success : NumericalIssue;
       m_factorizationIsOk = true;
       m_extractedDataAreDirty = true;
     }
-    
+
     template<typename MatrixDerived>
     void grab(const EigenBase<MatrixDerived> &A)
     {
       mp_matrix.~UmfpackMatrixRef();
       ::new (&mp_matrix) UmfpackMatrixRef(A.derived());
     }
-    
+
     void grab(const UmfpackMatrixRef &A)
     {
       if(&(A.derived()) != &mp_matrix)
@@ -357,19 +405,20 @@ class UmfPackLU : public SparseSolverBase<UmfPackLU<_MatrixType> >
         ::new (&mp_matrix) UmfpackMatrixRef(A);
       }
     }
-  
+
     // cached data to reduce reallocation, etc.
     mutable LUMatrixType m_l;
     int m_fact_errorCode;
     UmfpackControl m_control;
-    
+    mutable UmfpackInfo m_umfpackInfo;
+
     mutable LUMatrixType m_u;
     mutable IntColVectorType m_p;
     mutable IntRowVectorType m_q;
 
     UmfpackMatrixType m_dummy;
     UmfpackMatrixRef mp_matrix;
-  
+
     void* m_numeric;
     void* m_symbolic;
 
@@ -377,7 +426,7 @@ class UmfPackLU : public SparseSolverBase<UmfPackLU<_MatrixType> >
     int m_factorizationIsOk;
     int m_analysisIsOk;
     mutable bool m_extractedDataAreDirty;
-    
+
   private:
     UmfPackLU(const UmfPackLU& ) { }
 };
@@ -427,7 +476,7 @@ bool UmfPackLU<MatrixType>::_solve_impl(const MatrixBase<BDerived> &b, MatrixBas
   eigen_assert((BDerived::Flags&RowMajorBit)==0 && "UmfPackLU backend does not support non col-major rhs yet");
   eigen_assert((XDerived::Flags&RowMajorBit)==0 && "UmfPackLU backend does not support non col-major result yet");
   eigen_assert(b.derived().data() != x.derived().data() && " Umfpack does not support inplace solve");
-  
+
   int errorCode;
   Scalar* x_ptr = 0;
   Matrix<Scalar,Dynamic,1> x_tmp;
@@ -442,7 +491,7 @@ bool UmfPackLU<MatrixType>::_solve_impl(const MatrixBase<BDerived> &b, MatrixBas
       x_ptr = &x.col(j).coeffRef(0);
     errorCode = umfpack_solve(UMFPACK_A,
         mp_matrix.outerIndexPtr(), mp_matrix.innerIndexPtr(), mp_matrix.valuePtr(),
-        x_ptr, &b.const_cast_derived().col(j).coeffRef(0), m_numeric, m_control.data(), 0);
+        x_ptr, &b.const_cast_derived().col(j).coeffRef(0), m_numeric, m_control.data(), m_umfpackInfo.data());
     if(x.innerStride()!=1)
       x.col(j) = x_tmp;
     if (errorCode!=0)

Eigen/src/plugins/ArrayCwiseBinaryOps.h

@@ -269,44 +269,6 @@ const CwiseBinaryOp<internal::scalar_difference_op<T,Scalar>,Constant<T>,Derived
   operator/(const T& s,const StorageBaseType& a);
 #endif
 
-/** \returns an expression of the coefficient-wise && operator of *this and \a other
-  *
-  * \warning this operator is for expression of bool only.
-  *
-  * Example: \include Cwise_boolean_and.cpp
-  * Output: \verbinclude Cwise_boolean_and.out
-  *
-  * \sa operator||(), select()
-  */
-template<typename OtherDerived>
-EIGEN_DEVICE_FUNC
-inline const CwiseBinaryOp<internal::scalar_boolean_and_op, const Derived, const OtherDerived>
-operator&&(const EIGEN_CURRENT_STORAGE_BASE_CLASS<OtherDerived> &other) const
-{
-  EIGEN_STATIC_ASSERT((internal::is_same<bool,Scalar>::value && internal::is_same<bool,typename OtherDerived::Scalar>::value),
-                      THIS_METHOD_IS_ONLY_FOR_EXPRESSIONS_OF_BOOL);
-  return CwiseBinaryOp<internal::scalar_boolean_and_op, const Derived, const OtherDerived>(derived(),other.derived());
-}
-
-/** \returns an expression of the coefficient-wise || operator of *this and \a other
-  *
-  * \warning this operator is for expression of bool only.
-  *
-  * Example: \include Cwise_boolean_or.cpp
-  * Output: \verbinclude Cwise_boolean_or.out
-  *
-  * \sa operator&&(), select()
-  */
-template<typename OtherDerived>
-EIGEN_DEVICE_FUNC
-inline const CwiseBinaryOp<internal::scalar_boolean_or_op, const Derived, const OtherDerived>
-operator||(const EIGEN_CURRENT_STORAGE_BASE_CLASS<OtherDerived> &other) const
-{
-  EIGEN_STATIC_ASSERT((internal::is_same<bool,Scalar>::value && internal::is_same<bool,typename OtherDerived::Scalar>::value),
-                      THIS_METHOD_IS_ONLY_FOR_EXPRESSIONS_OF_BOOL);
-  return CwiseBinaryOp<internal::scalar_boolean_or_op, const Derived, const OtherDerived>(derived(),other.derived());
-}
-
 /** \returns an expression of the coefficient-wise ^ operator of *this and \a other
  *
  * \warning this operator is for expression of bool only.

Eigen/src/plugins/BlockMethods.h

@@ -818,7 +818,7 @@ inline typename FixedBlockXpr<NRows,NCols>::Type block(Index startRow, Index sta
   return typename FixedBlockXpr<NRows,NCols>::Type(derived(), startRow, startCol, blockRows, blockCols);
 }
 
-/// This is the const version of block<>(Index, Index, Index, Index). */
+/// This is the const version of block<>(Index, Index, Index, Index).
 template<int NRows, int NCols>
 inline const typename ConstFixedBlockXpr<NRows,NCols>::Type block(Index startRow, Index startCol,
                                                               Index blockRows, Index blockCols) const
@@ -832,15 +832,15 @@ inline const typename ConstFixedBlockXpr<NRows,NCols>::Type block(Index startRow
 /// Output: \verbinclude MatrixBase_col.out
 ///
 EIGEN_DOC_BLOCK_ADDONS_INNER_PANEL_IF(column-major)
-///
-/// \sa row(), class Block */
+/**
+  * \sa row(), class Block */
 EIGEN_DEVICE_FUNC
 inline ColXpr col(Index i)
 {
   return ColXpr(derived(), i);
 }
 
-/// This is the const version of col(). */
+/// This is the const version of col().
 EIGEN_DEVICE_FUNC
 inline ConstColXpr col(Index i) const
 {
@@ -853,8 +853,8 @@ inline ConstColXpr col(Index i) const
 /// Output: \verbinclude MatrixBase_row.out
 ///
 EIGEN_DOC_BLOCK_ADDONS_INNER_PANEL_IF(row-major)
-///
-/// \sa col(), class Block */
+/**
+  * \sa col(), class Block */
 EIGEN_DEVICE_FUNC
 inline RowXpr row(Index i)
 {