* find the proper way of nesting the expression in Flagged:

finally that's more subtle than just using ei_nested, because when flagging with NestByValueBit we want to store the expression by value already, regardless of whether it already had the NestByValueBit set. * rename temporary() ----> nestByValue() * move the old Product.h to disabled/, replace by what was ProductWIP.h * tweak -O and -g flags for tests and examples * reorder the tests -- basic things go first * simplifications, e.g. in many methoeds return derived() and count on implicit casting to the actual return type. * strip some not-really-useful stuff from the heaviest tests
2026-04-10 11:34:33 +08:00 · 2008-05-28 04:38:16 +00:00
parent 559233c73e
commit aebecae510
15 changed files with 394 additions and 399 deletions
--- a/Eigen/src/Core/CwiseUnaryOp.h
+++ b/Eigen/src/Core/CwiseUnaryOp.h
@@ -118,7 +118,7 @@ template<typename Derived>
 inline const CwiseUnaryOp<ei_scalar_opposite_op<typename ei_traits<Derived>::Scalar>,Derived>
 MatrixBase<Derived>::operator-() const
 {
-  return CwiseUnaryOp<ei_scalar_opposite_op<Scalar>, Derived>(derived());
+  return derived();
 }

 /** \returns an expression of the coefficient-wise absolute value of \c *this
@@ -127,7 +127,7 @@ template<typename Derived>
 inline const CwiseUnaryOp<ei_scalar_abs_op<typename ei_traits<Derived>::Scalar>,Derived>
 MatrixBase<Derived>::cwiseAbs() const
 {
-  return CwiseUnaryOp<ei_scalar_abs_op<Scalar>,Derived>(derived());
+  return derived();
 }

 /** \returns an expression of the coefficient-wise squared absolute value of \c *this
@@ -136,7 +136,7 @@ template<typename Derived>
 inline const CwiseUnaryOp<ei_scalar_abs2_op<typename ei_traits<Derived>::Scalar>,Derived>
 MatrixBase<Derived>::cwiseAbs2() const
 {
-  return CwiseUnaryOp<ei_scalar_abs2_op<Scalar>,Derived>(derived());
+  return derived();
 }

 /** \returns an expression of the complex conjugate of *this.
@@ -146,7 +146,7 @@ template<typename Derived>
 inline const CwiseUnaryOp<ei_scalar_conjugate_op<typename ei_traits<Derived>::Scalar>, Derived>
 MatrixBase<Derived>::conjugate() const
 {
-  return CwiseUnaryOp<ei_scalar_conjugate_op<Scalar>, Derived>(derived());
+  return derived();
 }

 /** \returns an expression of *this with the \a Scalar type casted to
@@ -161,7 +161,7 @@ template<typename NewType>
 inline const CwiseUnaryOp<ei_scalar_cast_op<typename ei_traits<Derived>::Scalar, NewType>, Derived>
 MatrixBase<Derived>::cast() const
 {
-  return CwiseUnaryOp<ei_scalar_cast_op<Scalar, NewType>, Derived>(derived());
+  return derived();
 }

 /** \relates MatrixBase */
@@ -201,7 +201,7 @@ template<typename Derived>
 inline const CwiseUnaryOp<ei_scalar_sqrt_op<typename ei_traits<Derived>::Scalar>, Derived>
 MatrixBase<Derived>::cwiseSqrt() const
 {
-  return CwiseUnaryOp<ei_scalar_sqrt_op<Scalar>, Derived>(derived());
+  return derived();
 }

 /** \returns an expression of the coefficient-wise exponential of *this. */
@@ -209,7 +209,7 @@ template<typename Derived>
 inline const CwiseUnaryOp<ei_scalar_exp_op<typename ei_traits<Derived>::Scalar>, Derived>
 MatrixBase<Derived>::cwiseExp() const
 {
-  return CwiseUnaryOp<ei_scalar_exp_op<Scalar>, Derived>(derived());
+  return derived();
 }

 /** \returns an expression of the coefficient-wise logarithm of *this. */
@@ -217,7 +217,7 @@ template<typename Derived>
 inline const CwiseUnaryOp<ei_scalar_log_op<typename ei_traits<Derived>::Scalar>, Derived>
 MatrixBase<Derived>::cwiseLog() const
 {
-  return CwiseUnaryOp<ei_scalar_log_op<Scalar>, Derived>(derived());
+  return derived();
 }

 /** \returns an expression of the coefficient-wise cosine of *this. */
@@ -225,7 +225,7 @@ template<typename Derived>
 inline const CwiseUnaryOp<ei_scalar_cos_op<typename ei_traits<Derived>::Scalar>, Derived>
 MatrixBase<Derived>::cwiseCos() const
 {
-  return CwiseUnaryOp<ei_scalar_cos_op<Scalar>, Derived>(derived());
+  return derived();
 }

 /** \returns an expression of the coefficient-wise sine of *this. */
@@ -233,10 +233,10 @@ template<typename Derived>
 inline const CwiseUnaryOp<ei_scalar_sin_op<typename ei_traits<Derived>::Scalar>, Derived>
 MatrixBase<Derived>::cwiseSin() const
 {
-  return CwiseUnaryOp<ei_scalar_sin_op<Scalar>, Derived>(derived());
+  return derived();
 }

-/** \relates MatrixBase */
+/** \returns an expression of the coefficient-wise power of *this to the given exponent. */
 template<typename Derived>
 inline const CwiseUnaryOp<ei_scalar_pow_op<typename ei_traits<Derived>::Scalar>, Derived>
 MatrixBase<Derived>::cwisePow(const Scalar& exponent) const
--- a/Eigen/src/Core/DiagonalMatrix.h
+++ b/Eigen/src/Core/DiagonalMatrix.h
@@ -95,7 +95,7 @@ template<typename Derived>
 inline const DiagonalMatrix<Derived>
 MatrixBase<Derived>::asDiagonal() const
 {
-  return DiagonalMatrix<Derived>(derived());
+  return derived();
 }

 /** \returns true if *this is approximately equal to a diagonal matrix,
--- a/Eigen/src/Core/Flagged.h
+++ b/Eigen/src/Core/Flagged.h
@@ -33,7 +33,7 @@
  * \param Added the flags added to the expression
  * \param Removed the flags removed from the expression (has priority over Added).
  *
-  * This class represents an expression whose flags have been modified
+  * This class represents an expression whose flags have been modified.
  * It is the return type of MatrixBase::flagged()
  * and most of the time this is the only way it is used.
  *
@@ -94,7 +94,11 @@ template<typename ExpressionType, unsigned int Added, unsigned int Removed> clas
    }

  protected:
-    const ExpressionType m_matrix;
+    const typename ei_meta_if<
+      Added & ~Removed & NestByValueBit,
+      ExpressionType,
+      typename ExpressionType::Nested
+    >::ret m_matrix;
 };

 /** \returns an expression of *this with added flags
@@ -121,7 +125,7 @@ MatrixBase<Derived>::lazy() const
  */
 template<typename Derived>
 inline const Flagged<Derived, NestByValueBit, 0>
-MatrixBase<Derived>::temporary() const
+MatrixBase<Derived>::nestByValue() const
 {
  return derived();
 }
--- a/Eigen/src/Core/MatrixBase.h
+++ b/Eigen/src/Core/MatrixBase.h
@@ -452,7 +452,7 @@ template<typename Derived> class MatrixBase
    template<unsigned int Added>
    const Flagged<Derived, Added, 0> marked() const;
    const Flagged<Derived, 0, EvalBeforeNestingBit | EvalBeforeAssigningBit> lazy() const;
-    const Flagged<Derived, NestByValueBit, 0> temporary() const;
+    const Flagged<Derived, NestByValueBit, 0> nestByValue() const;

    /** \returns number of elements to skip to pass from one row (resp. column) to another
      * for a row-major (resp. column-major) matrix.
--- a/Eigen/src/Core/Product.h
+++ b/Eigen/src/Core/Product.h
@@ -41,7 +41,7 @@ template<int Size, typename Lhs, typename Rhs>
 struct ei_product_unroller<0, Size, Lhs, Rhs>
 {
  inline static void run(int row, int col, const Lhs& lhs, const Rhs& rhs,
-                               typename Lhs::Scalar &res)
+                  typename Lhs::Scalar &res)
  {
    res = lhs.coeff(row, 0) * rhs.coeff(0, col);
  }
@@ -60,12 +60,6 @@ struct ei_product_unroller<Index, 0, Lhs, Rhs>
  inline static void run(int, int, const Lhs&, const Rhs&, typename Lhs::Scalar&) {}
 };

-template<typename Lhs, typename Rhs>
-struct ei_product_unroller<0, Dynamic, Lhs, Rhs>
-{
-  static void run(int, int, const Lhs&, const Rhs&, typename Lhs::Scalar&) {}
-};
-
 template<bool RowMajor, int Index, int Size, typename Lhs, typename Rhs, typename PacketScalar>
 struct ei_packet_product_unroller;

@@ -119,12 +113,6 @@ struct ei_packet_product_unroller<false, Index, Dynamic, Lhs, Rhs, PacketScalar>
  inline static void run(int, int, const Lhs&, const Rhs&, PacketScalar&) {}
 };

-template<typename Lhs, typename Rhs, typename PacketScalar>
-struct ei_packet_product_unroller<false, 0, Dynamic, Lhs, Rhs, PacketScalar>
-{
-  static void run(int, int, const Lhs&, const Rhs&, PacketScalar&) {}
-};
-
 template<typename Product, bool RowMajor = true> struct ProductPacketCoeffImpl {
  inline static typename Product::PacketScalar execute(const Product& product, int row, int col)
  { return product._packetCoeffRowMajor(row,col); }
@@ -153,18 +141,74 @@ template<typename Lhs, typename Rhs> struct ei_product_eval_mode
 {
  enum{ value =  Lhs::MaxRowsAtCompileTime >= EIGEN_CACHEFRIENDLY_PRODUCT_THRESHOLD
              && Rhs::MaxColsAtCompileTime >= EIGEN_CACHEFRIENDLY_PRODUCT_THRESHOLD
-              && (!( (Lhs::Flags&RowMajorBit) && ((Rhs::Flags&RowMajorBit) ^ RowMajorBit)))
+              && Lhs::MaxColsAtCompileTime >= EIGEN_CACHEFRIENDLY_PRODUCT_THRESHOLD
              ? CacheFriendlyProduct : NormalProduct };
 };

+template<typename T> class ei_product_eval_to_column_major
+{
+    typedef typename ei_traits<T>::Scalar _Scalar;
+    enum {_MaxRows = ei_traits<T>::MaxRowsAtCompileTime,
+          _MaxCols = ei_traits<T>::MaxColsAtCompileTime,
+          _Flags = ei_traits<T>::Flags
+    };
+
+  public:
+    typedef Matrix<_Scalar,
+                  ei_traits<T>::RowsAtCompileTime,
+                  ei_traits<T>::ColsAtCompileTime,
+                  ei_corrected_matrix_flags<_Scalar, ei_size_at_compile_time<_MaxRows,_MaxCols>::ret, _Flags>::ret & ~RowMajorBit,
+                  ei_traits<T>::MaxRowsAtCompileTime,
+                  ei_traits<T>::MaxColsAtCompileTime> type;
+};
+
+template<typename T, int n=1> struct ei_product_nested_rhs
+{
+  typedef typename ei_meta_if<
+    (ei_traits<T>::Flags & NestByValueBit) && (!(ei_traits<T>::Flags & RowMajorBit)) && (int(ei_traits<T>::Flags) & DirectAccessBit),
+    T,
+    typename ei_meta_if<
+        ((ei_traits<T>::Flags & EvalBeforeNestingBit)
+      || (ei_traits<T>::Flags & RowMajorBit)
+      || (!(ei_traits<T>::Flags & DirectAccessBit))
+      || (n+1) * (NumTraits<typename ei_traits<T>::Scalar>::ReadCost) < (n-1) * T::CoeffReadCost),
+      typename ei_product_eval_to_column_major<T>::type,
+      const T&
+    >::ret
+  >::ret type;
+};
+
+template<typename T, int n=1> struct ei_product_nested_lhs
+{
+  typedef typename ei_meta_if<
+    ei_traits<T>::Flags & NestByValueBit && (int(ei_traits<T>::Flags) & DirectAccessBit),
+    T,
+    typename ei_meta_if<
+         int(ei_traits<T>::Flags) & EvalBeforeNestingBit
+      || (!(int(ei_traits<T>::Flags) & DirectAccessBit))
+      || (n+1) * int(NumTraits<typename ei_traits<T>::Scalar>::ReadCost) < (n-1) * int(T::CoeffReadCost),
+      typename ei_eval<T>::type,
+      const T&
+    >::ret
+  >::ret type;
+};
+
 template<typename Lhs, typename Rhs, int EvalMode>
 struct ei_traits<Product<Lhs, Rhs, EvalMode> >
 {
  typedef typename Lhs::Scalar Scalar;
-  typedef typename ei_nested<Lhs,Rhs::ColsAtCompileTime>::type LhsNested;
-  typedef typename ei_nested<Rhs,Lhs::RowsAtCompileTime>::type RhsNested;
-  typedef typename ei_unref<LhsNested>::type _LhsNested;
-  typedef typename ei_unref<RhsNested>::type _RhsNested;
+  // the cache friendly product evals lhs once only
+  // FIXME what to do if we chose to dynamically call the normal product from the cache friendly one for small matrices ?
+  typedef typename ei_meta_if<EvalMode==CacheFriendlyProduct,
+      typename ei_product_nested_lhs<Lhs,1>::type,
+      typename ei_nested<Lhs,Rhs::ColsAtCompileTime>::type>::ret LhsNested;
+
+  // NOTE that rhs must be ColumnMajor, so we might need a special nested type calculation
+  typedef typename ei_meta_if<EvalMode==CacheFriendlyProduct,
+      typename ei_product_nested_rhs<Rhs,Lhs::RowsAtCompileTime>::type,
+      typename ei_nested<Rhs,Lhs::RowsAtCompileTime>::type>::ret RhsNested;
+  typedef typename ei_unconst<typename ei_unref<LhsNested>::type>::type _LhsNested;
+  typedef typename ei_unconst<typename ei_unref<RhsNested>::type>::type _RhsNested;
  enum {
    LhsCoeffReadCost = _LhsNested::CoeffReadCost,
    RhsCoeffReadCost = _RhsNested::CoeffReadCost,
@@ -174,6 +218,8 @@ struct ei_traits<Product<Lhs, Rhs, EvalMode> >
    ColsAtCompileTime = Rhs::ColsAtCompileTime,
    MaxRowsAtCompileTime = Lhs::MaxRowsAtCompileTime,
    MaxColsAtCompileTime = Rhs::MaxColsAtCompileTime,
+    // the vectorization flags are only used by the normal product,
+    // the other one is always vectorized !
    _RhsVectorizable = (RhsFlags & RowMajorBit) && (RhsFlags & VectorizableBit) && (ColsAtCompileTime % ei_packet_traits<Scalar>::size == 0),
    _LhsVectorizable = (!(LhsFlags & RowMajorBit)) && (LhsFlags & VectorizableBit) && (RowsAtCompileTime % ei_packet_traits<Scalar>::size == 0),
    _Vectorizable = (_LhsVectorizable || _RhsVectorizable) ? 1 : 0,
@@ -207,6 +253,10 @@ template<typename Lhs, typename Rhs, int EvalMode> class Product : ei_no_assignm
    typedef typename ei_traits<Product>::_LhsNested _LhsNested;
    typedef typename ei_traits<Product>::_RhsNested _RhsNested;

+    enum {
+      PacketSize = ei_packet_traits<Scalar>::size
+    };
+
    inline Product(const Lhs& lhs, const Rhs& rhs)
      : m_lhs(lhs), m_rhs(rhs)
    {
@@ -214,12 +264,12 @@ template<typename Lhs, typename Rhs, int EvalMode> class Product : ei_no_assignm
    }

    /** \internal */
-    template<typename DestDerived, int AlignedMode>
-    void _cacheOptimalEval(DestDerived& res, ei_meta_false) const;
-    #ifdef EIGEN_VECTORIZE
-    template<typename DestDerived, int AlignedMode>
-    void _cacheOptimalEval(DestDerived& res, ei_meta_true) const;
-    #endif
+    template<typename DestDerived>
+    void _cacheFriendlyEval(DestDerived& res) const;
+
+    /** \internal */
+    template<typename DestDerived>
+    void _cacheFriendlyEvalAndAdd(DestDerived& res) const;

  private:

@@ -252,7 +302,7 @@ template<typename Lhs, typename Rhs, int EvalMode> class Product : ei_no_assignm
      if(Lhs::ColsAtCompileTime <= EIGEN_UNROLLING_LIMIT)
      {
        PacketScalar res;
-        ei_packet_product_unroller<Flags&RowMajorBit, Lhs::ColsAtCompileTime-1,
+        ei_packet_product_unroller<Flags&RowMajorBit ? true : false, Lhs::ColsAtCompileTime-1,
                            Lhs::ColsAtCompileTime <= EIGEN_UNROLLING_LIMIT
                              ? Lhs::ColsAtCompileTime : Dynamic,
                            _LhsNested, _RhsNested, PacketScalar>
@@ -279,16 +329,10 @@ template<typename Lhs, typename Rhs, int EvalMode> class Product : ei_no_assignm
      for(int i = 1; i < m_lhs.cols(); i++)
        res =  ei_pmadd(m_lhs.template packetCoeff<Aligned>(row, i), ei_pset1(m_rhs.coeff(i, col)), res);
      return res;
-//       const PacketScalar tmp[4];
-//       ei_punpack(m_rhs.packetCoeff(0,col), tmp);
-//
-//       return
-//         ei_pmadd(m_lhs.packetCoeff(row, 0), tmp[0],
-//           ei_pmadd(m_lhs.packetCoeff(row, 1), tmp[1],
-//             ei_pmadd(m_lhs.packetCoeff(row, 2), tmp[2]
-//               ei_pmul(m_lhs.packetCoeff(row, 3), tmp[3]))));
    }

+    template<typename Lhs_, typename Rhs_, int EvalMode_, typename DestDerived_, bool DirectAccess_>
+    friend struct ei_cache_friendly_selector;

  protected:
    const LhsNested m_lhs;
@@ -296,9 +340,6 @@ template<typename Lhs, typename Rhs, int EvalMode> class Product : ei_no_assignm
 };

 /** \returns the matrix product of \c *this and \a other.
-  *
-  * \note This function causes an immediate evaluation. If you want to perform a matrix product
-  * without immediate evaluation, call .lazy() on one of the matrices before taking the product.
  *
  * \sa lazy(), operator*=(const MatrixBase&)
  */
@@ -322,168 +363,107 @@ MatrixBase<Derived>::operator*=(const MatrixBase<OtherDerived> &other)
  return *this = *this * other;
 }

+/** \internal */
+template<typename Derived>
+template<typename Lhs,typename Rhs>
+inline Derived&
+MatrixBase<Derived>::operator+=(const Flagged<Product<Lhs,Rhs,CacheFriendlyProduct>, 0, EvalBeforeNestingBit | EvalBeforeAssigningBit>& other)
+{
+  other._expression()._cacheFriendlyEvalAndAdd(const_cast_derived());
+  return derived();
+}
+
 template<typename Derived>
 template<typename Lhs, typename Rhs>
 inline Derived& MatrixBase<Derived>::lazyAssign(const Product<Lhs,Rhs,CacheFriendlyProduct>& product)
 {
-  product.template _cacheOptimalEval<Derived, Aligned>(derived(),
-    #ifdef EIGEN_VECTORIZE
-    typename ei_meta_if<Flags & VectorizableBit, ei_meta_true, ei_meta_false>::ret()
-    #else
-    ei_meta_false()
-    #endif
-    );
+  product._cacheFriendlyEval(derived());
  return derived();
 }

-template<typename Lhs, typename Rhs, int EvalMode>
-template<typename DestDerived, int AlignedMode>
-void Product<Lhs,Rhs,EvalMode>::_cacheOptimalEval(DestDerived& res, ei_meta_false) const
+template<typename Lhs, typename Rhs, int EvalMode, typename DestDerived, bool DirectAccess>
+struct ei_cache_friendly_selector
 {
-  res.setZero();
-  const int cols4 = m_lhs.cols() & 0xfffffffC;
-  if (Lhs::Flags&RowMajorBit)
+  typedef Product<Lhs,Rhs,EvalMode> Prod;
+  typedef typename Prod::_LhsNested _LhsNested;
+  typedef typename Prod::_RhsNested _RhsNested;
+  typedef typename Prod::Scalar Scalar;
+  static inline void eval(const Prod& product, DestDerived& res)
  {
-//     std::cout << "opt rhs\n";
-    int j=0;
-    for(; j<cols4; j+=4)
+    if ( product._rows()>=EIGEN_CACHEFRIENDLY_PRODUCT_THRESHOLD
+      && product._cols()>=EIGEN_CACHEFRIENDLY_PRODUCT_THRESHOLD
+      && product.m_lhs.cols()>=EIGEN_CACHEFRIENDLY_PRODUCT_THRESHOLD
+    )
    {
-      for(int k=0; k<this->rows(); ++k)
-      {
-        const Scalar tmp0 = m_lhs.coeff(k,j  );
-        const Scalar tmp1 = m_lhs.coeff(k,j+1);
-        const Scalar tmp2 = m_lhs.coeff(k,j+2);
-        const Scalar tmp3 = m_lhs.coeff(k,j+3);
-        for (int i=0; i<this->cols(); ++i)
-          res.coeffRef(k,i) += tmp0 * m_rhs.coeff(j+0,i) + tmp1 * m_rhs.coeff(j+1,i)
-                             + tmp2 * m_rhs.coeff(j+2,i) + tmp3 * m_rhs.coeff(j+3,i);
-      }
+      res.setZero();
+      ei_cache_friendly_product<Scalar>(
+        product._rows(), product._cols(), product.m_lhs.cols(),
+        _LhsNested::Flags&RowMajorBit, &(product.m_lhs.const_cast_derived().coeffRef(0,0)), product.m_lhs.stride(),
+        _RhsNested::Flags&RowMajorBit, &(product.m_rhs.const_cast_derived().coeffRef(0,0)), product.m_rhs.stride(),
+        Prod::Flags&RowMajorBit, &(res.coeffRef(0,0)), res.stride()
+      );
    }
-    for(; j<m_lhs.cols(); ++j)
+    else
    {
-      for(int k=0; k<this->rows(); ++k)
-      {
-        const Scalar tmp = m_rhs.coeff(k,j);
-        for (int i=0; i<this->cols(); ++i)
-          res.coeffRef(k,i) += tmp * m_lhs.coeff(j,i);
-      }
+      res = Product<_LhsNested,_RhsNested,NormalProduct>(product.m_lhs, product.m_rhs).lazy();
    }
  }
-  else
+
+  static inline void eval_and_add(const Prod& product, DestDerived& res)
  {
-//     std::cout << "opt lhs\n";
-    int j = 0;
-    for(; j<cols4; j+=4)
+    if ( product._rows()>=EIGEN_CACHEFRIENDLY_PRODUCT_THRESHOLD
+      && product._cols()>=EIGEN_CACHEFRIENDLY_PRODUCT_THRESHOLD
+      && product.m_lhs.cols()>=EIGEN_CACHEFRIENDLY_PRODUCT_THRESHOLD
+    )
    {
-      for(int k=0; k<this->cols(); ++k)
-      {
-        const Scalar tmp0 = m_rhs.coeff(j  ,k);
-        const Scalar tmp1 = m_rhs.coeff(j+1,k);
-        const Scalar tmp2 = m_rhs.coeff(j+2,k);
-        const Scalar tmp3 = m_rhs.coeff(j+3,k);
-        for (int i=0; i<this->rows(); ++i)
-          res.coeffRef(i,k) += tmp0 * m_lhs.coeff(i,j+0) + tmp1 * m_lhs.coeff(i,j+1)
-                             + tmp2 * m_lhs.coeff(i,j+2) + tmp3 * m_lhs.coeff(i,j+3);
-      }
+      ei_cache_friendly_product<Scalar>(
+        product._rows(), product._cols(), product.m_lhs.cols(),
+        _LhsNested::Flags&RowMajorBit, &(product.m_lhs.const_cast_derived().coeffRef(0,0)), product.m_lhs.stride(),
+        _RhsNested::Flags&RowMajorBit, &(product.m_rhs.const_cast_derived().coeffRef(0,0)), product.m_rhs.stride(),
+        Prod::Flags&RowMajorBit, &(res.coeffRef(0,0)), res.stride()
+      );
    }
-    for(; j<m_lhs.cols(); ++j)
+    else
    {
-      for(int k=0; k<this->cols(); ++k)
-      {
-        const Scalar tmp = m_rhs.coeff(j,k);
-        for (int i=0; i<this->rows(); ++i)
-          res.coeffRef(i,k) += tmp * m_lhs.coeff(i,j);
-      }
+      res += Product<_LhsNested,_RhsNested,NormalProduct>(product.m_lhs, product.m_rhs).lazy();
    }
  }
+};
+
+template<typename Lhs, typename Rhs, int EvalMode, typename DestDerived>
+struct ei_cache_friendly_selector<Lhs,Rhs,EvalMode,DestDerived,false>
+{
+  typedef Product<Lhs,Rhs,EvalMode> Prod;
+  typedef typename Prod::_LhsNested _LhsNested;
+  typedef typename Prod::_RhsNested _RhsNested;
+  typedef typename Prod::Scalar Scalar;
+  static inline void eval(const Prod& product, DestDerived& res)
+  {
+    res = Product<_LhsNested,_RhsNested,NormalProduct>(product.m_lhs, product.m_rhs).lazy();
+  }
+
+  static inline void eval_and_add(const Prod& product, DestDerived& res)
+  {
+    res += Product<_LhsNested,_RhsNested,NormalProduct>(product.m_lhs, product.m_rhs).lazy();
+  }
+};
+
+template<typename Lhs, typename Rhs, int EvalMode>
+template<typename DestDerived>
+inline void Product<Lhs,Rhs,EvalMode>::_cacheFriendlyEval(DestDerived& res) const
+{
+  ei_cache_friendly_selector<Lhs,Rhs,EvalMode,DestDerived,
+                             _LhsNested::Flags&_RhsNested::Flags&DirectAccessBit>
+    ::eval(*this, res);
 }

-#ifdef EIGEN_VECTORIZE
 template<typename Lhs, typename Rhs, int EvalMode>
-template<typename DestDerived, int AlignedMode>
-void Product<Lhs,Rhs,EvalMode>::_cacheOptimalEval(DestDerived& res, ei_meta_true) const
+template<typename DestDerived>
+inline void Product<Lhs,Rhs,EvalMode>::_cacheFriendlyEvalAndAdd(DestDerived& res) const
 {
-
-  if (((Lhs::Flags&RowMajorBit) && (_cols() % ei_packet_traits<Scalar>::size != 0))
-    || (_rows() % ei_packet_traits<Scalar>::size != 0))
-  {
-    return _cacheOptimalEval<DestDerived, AlignedMode>(res, ei_meta_false());
-  }
-
-  res.setZero();
-  const int cols4 = m_lhs.cols() & 0xfffffffC;
-  if (Lhs::Flags&RowMajorBit)
-  {
-//     std::cout << "packet rhs\n";
-    int j=0;
-    for(; j<cols4; j+=4)
-    {
-      for(int k=0; k<this->rows(); k++)
-      {
-        const typename ei_packet_traits<Scalar>::type tmp0 = ei_pset1(m_lhs.coeff(k,j+0));
-        const typename ei_packet_traits<Scalar>::type tmp1 = ei_pset1(m_lhs.coeff(k,j+1));
-        const typename ei_packet_traits<Scalar>::type tmp2 = ei_pset1(m_lhs.coeff(k,j+2));
-        const typename ei_packet_traits<Scalar>::type tmp3 = ei_pset1(m_lhs.coeff(k,j+3));
-        for (int i=0; i<this->cols(); i+=ei_packet_traits<Scalar>::size)
-        {
-          res.template writePacketCoeff<AlignedMode>(k,i,
-            ei_pmadd(tmp0, m_rhs.template packetCoeff<AlignedMode>(j+0,i),
-              ei_pmadd(tmp1, m_rhs.template packetCoeff<AlignedMode>(j+1,i),
-                ei_pmadd(tmp2, m_rhs.template packetCoeff<AlignedMode>(j+2,i),
-                  ei_pmadd(tmp3, m_rhs.template packetCoeff<AlignedMode>(j+3,i),
-                    res.template packetCoeff<AlignedMode>(k,i)))))
-          );
-        }
-      }
-    }
-    for(; j<m_lhs.cols(); ++j)
-    {
-      for(int k=0; k<this->rows(); k++)
-      {
-        const typename ei_packet_traits<Scalar>::type tmp = ei_pset1(m_lhs.coeff(k,j));
-        for (int i=0; i<this->cols(); i+=ei_packet_traits<Scalar>::size)
-          res.template writePacketCoeff<AlignedMode>(k,i,
-            ei_pmadd(tmp, m_rhs.template packetCoeff<AlignedMode>(j,i), res.template packetCoeff<AlignedMode>(k,i)));
-      }
-    }
-  }
-  else
-  {
-//     std::cout << "packet lhs\n";
-    int k=0;
-    for(; k<cols4; k+=4)
-    {
-      for(int j=0; j<this->cols(); j+=1)
-      {
-        const typename ei_packet_traits<Scalar>::type tmp0 = ei_pset1(m_rhs.coeff(k+0,j));
-        const typename ei_packet_traits<Scalar>::type tmp1 = ei_pset1(m_rhs.coeff(k+1,j));
-        const typename ei_packet_traits<Scalar>::type tmp2 = ei_pset1(m_rhs.coeff(k+2,j));
-        const typename ei_packet_traits<Scalar>::type tmp3 = ei_pset1(m_rhs.coeff(k+3,j));
-
-        for (int i=0; i<this->rows(); i+=ei_packet_traits<Scalar>::size)
-        {
-          res.template writePacketCoeff<AlignedMode>(i,j,
-            ei_pmadd(tmp0, m_lhs.template packetCoeff<AlignedMode>(i,k),
-              ei_pmadd(tmp1, m_lhs.template packetCoeff<AlignedMode>(i,k+1),
-                ei_pmadd(tmp2, m_lhs.template packetCoeff<AlignedMode>(i,k+2),
-                  ei_pmadd(tmp3, m_lhs.template packetCoeff<AlignedMode>(i,k+3),
-                    res.template packetCoeff<AlignedMode>(i,j)))))
-          );
-        }
-      }
-    }
-    for(; k<m_lhs.cols(); ++k)
-    {
-      for(int j=0; j<this->cols(); j++)
-      {
-        const typename ei_packet_traits<Scalar>::type tmp = ei_pset1(m_rhs.coeff(k,j));
-        for (int i=0; i<this->rows(); i+=ei_packet_traits<Scalar>::size)
-          res.template writePacketCoeff<AlignedMode>(k,j,
-            ei_pmadd(tmp, m_lhs.template packetCoeff<AlignedMode>(i,k), res.template packetCoeff<AlignedMode>(i,j)));
-      }
-    }
-  }
+  ei_cache_friendly_selector<Lhs,Rhs,EvalMode,DestDerived,
+                             _LhsNested::Flags&_RhsNested::Flags&DirectAccessBit>
+    ::eval_and_add(*this, res);
 }
-#endif // EIGEN_VECTORIZE

 #endif // EIGEN_PRODUCT_H
--- a/Eigen/src/Core/ProductWIP.h
+++ b/Eigen/src/Core/ProductWIP.h
@@ -1,471 +0,0 @@
-// This file is part of Eigen, a lightweight C++ template library
-// for linear algebra. Eigen itself is part of the KDE project.
-//
-// Copyright (C) 2006-2008 Benoit Jacob <jacob@math.jussieu.fr>
-// Copyright (C) 2008 Gael Guennebaud <g.gael@free.fr>
-//
-// Eigen is free software; you can redistribute it and/or
-// modify it under the terms of the GNU Lesser General Public
-// License as published by the Free Software Foundation; either
-// version 3 of the License, or (at your option) any later version.
-//
-// Alternatively, you can redistribute it and/or
-// modify it under the terms of the GNU General Public License as
-// published by the Free Software Foundation; either version 2 of
-// the License, or (at your option) any later version.
-//
-// Eigen is distributed in the hope that it will be useful, but WITHOUT ANY
-// WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS
-// FOR A PARTICULAR PURPOSE. See the GNU Lesser General Public License or the
-// GNU General Public License for more details.
-//
-// You should have received a copy of the GNU Lesser General Public
-// License and a copy of the GNU General Public License along with
-// Eigen. If not, see <http://www.gnu.org/licenses/>.
-
-#ifndef EIGEN_PRODUCT_H
-#define EIGEN_PRODUCT_H
-
-#include "CacheFriendlyProduct.h"
-
-template<int Index, int Size, typename Lhs, typename Rhs>
-struct ei_product_unroller
-{
-  inline static void run(int row, int col, const Lhs& lhs, const Rhs& rhs,
-                               typename Lhs::Scalar &res)
-  {
-    ei_product_unroller<Index-1, Size, Lhs, Rhs>::run(row, col, lhs, rhs, res);
-    res += lhs.coeff(row, Index) * rhs.coeff(Index, col);
-  }
-};
-
-template<int Size, typename Lhs, typename Rhs>
-struct ei_product_unroller<0, Size, Lhs, Rhs>
-{
-  inline static void run(int row, int col, const Lhs& lhs, const Rhs& rhs,
-                  typename Lhs::Scalar &res)
-  {
-    res = lhs.coeff(row, 0) * rhs.coeff(0, col);
-  }
-};
-
-template<int Index, typename Lhs, typename Rhs>
-struct ei_product_unroller<Index, Dynamic, Lhs, Rhs>
-{
-  inline static void run(int, int, const Lhs&, const Rhs&, typename Lhs::Scalar&) {}
-};
-
-// prevent buggy user code from causing an infinite recursion
-template<int Index, typename Lhs, typename Rhs>
-struct ei_product_unroller<Index, 0, Lhs, Rhs>
-{
-  inline static void run(int, int, const Lhs&, const Rhs&, typename Lhs::Scalar&) {}
-};
-
-template<bool RowMajor, int Index, int Size, typename Lhs, typename Rhs, typename PacketScalar>
-struct ei_packet_product_unroller;
-
-template<int Index, int Size, typename Lhs, typename Rhs, typename PacketScalar>
-struct ei_packet_product_unroller<true, Index, Size, Lhs, Rhs, PacketScalar>
-{
-  inline static void run(int row, int col, const Lhs& lhs, const Rhs& rhs, PacketScalar &res)
-  {
-    ei_packet_product_unroller<true, Index-1, Size, Lhs, Rhs, PacketScalar>::run(row, col, lhs, rhs, res);
-    res =  ei_pmadd(ei_pset1(lhs.coeff(row, Index)), rhs.template packetCoeff<Aligned>(Index, col), res);
-  }
-};
-
-template<int Index, int Size, typename Lhs, typename Rhs, typename PacketScalar>
-struct ei_packet_product_unroller<false, Index, Size, Lhs, Rhs, PacketScalar>
-{
-  inline static void run(int row, int col, const Lhs& lhs, const Rhs& rhs, PacketScalar &res)
-  {
-    ei_packet_product_unroller<false, Index-1, Size, Lhs, Rhs, PacketScalar>::run(row, col, lhs, rhs, res);
-    res =  ei_pmadd(lhs.template packetCoeff<Aligned>(row, Index), ei_pset1(rhs.coeff(Index, col)), res);
-  }
-};
-
-template<int Size, typename Lhs, typename Rhs, typename PacketScalar>
-struct ei_packet_product_unroller<true, 0, Size, Lhs, Rhs, PacketScalar>
-{
-  inline static void run(int row, int col, const Lhs& lhs, const Rhs& rhs, PacketScalar &res)
-  {
-    res = ei_pmul(ei_pset1(lhs.coeff(row, 0)),rhs.template packetCoeff<Aligned>(0, col));
-  }
-};
-
-template<int Size, typename Lhs, typename Rhs, typename PacketScalar>
-struct ei_packet_product_unroller<false, 0, Size, Lhs, Rhs, PacketScalar>
-{
-  inline static void run(int row, int col, const Lhs& lhs, const Rhs& rhs, PacketScalar &res)
-  {
-    res = ei_pmul(lhs.template packetCoeff<Aligned>(row, 0), ei_pset1(rhs.coeff(0, col)));
-  }
-};
-
-template<bool RowMajor, int Index, typename Lhs, typename Rhs, typename PacketScalar>
-struct ei_packet_product_unroller<RowMajor, Index, Dynamic, Lhs, Rhs, PacketScalar>
-{
-  inline static void run(int, int, const Lhs&, const Rhs&, PacketScalar&) {}
-};
-
-template<int Index, typename Lhs, typename Rhs, typename PacketScalar>
-struct ei_packet_product_unroller<false, Index, Dynamic, Lhs, Rhs, PacketScalar>
-{
-  inline static void run(int, int, const Lhs&, const Rhs&, PacketScalar&) {}
-};
-
-template<typename Product, bool RowMajor = true> struct ProductPacketCoeffImpl {
-  inline static typename Product::PacketScalar execute(const Product& product, int row, int col)
-  { return product._packetCoeffRowMajor(row,col); }
-};
-
-template<typename Product> struct ProductPacketCoeffImpl<Product, false> {
-  inline static typename Product::PacketScalar execute(const Product& product, int row, int col)
-  { return product._packetCoeffColumnMajor(row,col); }
-};
-
-/** \class Product
-  *
-  * \brief Expression of the product of two matrices
-  *
-  * \param Lhs the type of the left-hand side
-  * \param Rhs the type of the right-hand side
-  * \param EvalMode internal use only
-  *
-  * This class represents an expression of the product of two matrices.
-  * It is the return type of the operator* between matrices, and most of the time
-  * this is the only way it is used.
-  *
-  * \sa class Sum, class Difference
-  */
-template<typename Lhs, typename Rhs> struct ei_product_eval_mode
-{
-  enum{ value =  Lhs::MaxRowsAtCompileTime >= EIGEN_CACHEFRIENDLY_PRODUCT_THRESHOLD
-              && Rhs::MaxColsAtCompileTime >= EIGEN_CACHEFRIENDLY_PRODUCT_THRESHOLD
-              && Lhs::MaxColsAtCompileTime >= EIGEN_CACHEFRIENDLY_PRODUCT_THRESHOLD
-              ? CacheFriendlyProduct : NormalProduct };
-};
-
-template<typename T> class ei_product_eval_to_column_major
-{
-    typedef typename ei_traits<T>::Scalar _Scalar;
-    enum {_MaxRows = ei_traits<T>::MaxRowsAtCompileTime,
-          _MaxCols = ei_traits<T>::MaxColsAtCompileTime,
-          _Flags = ei_traits<T>::Flags
-    };
-
-  public:
-    typedef Matrix<_Scalar,
-                  ei_traits<T>::RowsAtCompileTime,
-                  ei_traits<T>::ColsAtCompileTime,
-                  ei_corrected_matrix_flags<_Scalar, ei_size_at_compile_time<_MaxRows,_MaxCols>::ret, _Flags>::ret & ~RowMajorBit,
-                  ei_traits<T>::MaxRowsAtCompileTime,
-                  ei_traits<T>::MaxColsAtCompileTime> type;
-};
-
-template<typename T, int n=1> struct ei_product_nested_rhs
-{
-  typedef typename ei_meta_if<
-    (ei_traits<T>::Flags & NestByValueBit) && (!(ei_traits<T>::Flags & RowMajorBit)) && (int(ei_traits<T>::Flags) & DirectAccessBit),
-    T,
-    typename ei_meta_if<
-        ((ei_traits<T>::Flags & EvalBeforeNestingBit)
-      || (ei_traits<T>::Flags & RowMajorBit)
-      || (!(ei_traits<T>::Flags & DirectAccessBit))
-      || (n+1) * (NumTraits<typename ei_traits<T>::Scalar>::ReadCost) < (n-1) * T::CoeffReadCost),
-      typename ei_product_eval_to_column_major<T>::type,
-      const T&
-    >::ret
-  >::ret type;
-};
-
-template<typename T, int n=1> struct ei_product_nested_lhs
-{
-  typedef typename ei_meta_if<
-    ei_traits<T>::Flags & NestByValueBit && (int(ei_traits<T>::Flags) & DirectAccessBit),
-    T,
-    typename ei_meta_if<
-         int(ei_traits<T>::Flags) & EvalBeforeNestingBit
-      || (!(int(ei_traits<T>::Flags) & DirectAccessBit))
-      || (n+1) * int(NumTraits<typename ei_traits<T>::Scalar>::ReadCost) < (n-1) * int(T::CoeffReadCost),
-      typename ei_eval<T>::type,
-      const T&
-    >::ret
-  >::ret type;
-};
-
-template<typename Lhs, typename Rhs, int EvalMode>
-struct ei_traits<Product<Lhs, Rhs, EvalMode> >
-{
-  typedef typename Lhs::Scalar Scalar;
-  // the cache friendly product evals lhs once only
-  // FIXME what to do if we chose to dynamically call the normal product from the cache friendly one for small matrices ?
-  typedef typename ei_meta_if<EvalMode==CacheFriendlyProduct,
-      typename ei_product_nested_lhs<Lhs,1>::type,
-      typename ei_nested<Lhs,Rhs::ColsAtCompileTime>::type>::ret LhsNested;
-
-  // NOTE that rhs must be ColumnMajor, so we might need a special nested type calculation
-  typedef typename ei_meta_if<EvalMode==CacheFriendlyProduct,
-      typename ei_product_nested_rhs<Rhs,Lhs::RowsAtCompileTime>::type,
-      typename ei_nested<Rhs,Lhs::RowsAtCompileTime>::type>::ret RhsNested;
-  typedef typename ei_unconst<typename ei_unref<LhsNested>::type>::type _LhsNested;
-  typedef typename ei_unconst<typename ei_unref<RhsNested>::type>::type _RhsNested;
-  enum {
-    LhsCoeffReadCost = _LhsNested::CoeffReadCost,
-    RhsCoeffReadCost = _RhsNested::CoeffReadCost,
-    LhsFlags = _LhsNested::Flags,
-    RhsFlags = _RhsNested::Flags,
-    RowsAtCompileTime = Lhs::RowsAtCompileTime,
-    ColsAtCompileTime = Rhs::ColsAtCompileTime,
-    MaxRowsAtCompileTime = Lhs::MaxRowsAtCompileTime,
-    MaxColsAtCompileTime = Rhs::MaxColsAtCompileTime,
-    // the vectorization flags are only used by the normal product,
-    // the other one is always vectorized !
-    _RhsVectorizable = (RhsFlags & RowMajorBit) && (RhsFlags & VectorizableBit) && (ColsAtCompileTime % ei_packet_traits<Scalar>::size == 0),
-    _LhsVectorizable = (!(LhsFlags & RowMajorBit)) && (LhsFlags & VectorizableBit) && (RowsAtCompileTime % ei_packet_traits<Scalar>::size == 0),
-    _Vectorizable = (_LhsVectorizable || _RhsVectorizable) ? 1 : 0,
-    _RowMajor = (RhsFlags & RowMajorBit)
-              && (EvalMode==(int)CacheFriendlyProduct ? (int)LhsFlags & RowMajorBit : (!_LhsVectorizable)),
-    _LostBits = HereditaryBits & ~(
-                (_RowMajor ? 0 : RowMajorBit)
-              | ((RowsAtCompileTime == Dynamic || ColsAtCompileTime == Dynamic) ? 0 : LargeBit)),
-    Flags = ((unsigned int)(LhsFlags | RhsFlags) & _LostBits)
-          | EvalBeforeAssigningBit
-          | EvalBeforeNestingBit
-          | (_Vectorizable ? VectorizableBit : 0),
-    CoeffReadCost
-      = Lhs::ColsAtCompileTime == Dynamic
-      ? Dynamic
-      : Lhs::ColsAtCompileTime
-        * (NumTraits<Scalar>::MulCost + LhsCoeffReadCost + RhsCoeffReadCost)
-        + (Lhs::ColsAtCompileTime - 1) * NumTraits<Scalar>::AddCost
-  };
-};
-
-template<typename Lhs, typename Rhs, int EvalMode> class Product : ei_no_assignment_operator,
-  public MatrixBase<Product<Lhs, Rhs, EvalMode> >
-{
-  public:
-
-    EIGEN_GENERIC_PUBLIC_INTERFACE(Product)
-    friend class ProductPacketCoeffImpl<Product,Flags&RowMajorBit>;
-    typedef typename ei_traits<Product>::LhsNested LhsNested;
-    typedef typename ei_traits<Product>::RhsNested RhsNested;
-    typedef typename ei_traits<Product>::_LhsNested _LhsNested;
-    typedef typename ei_traits<Product>::_RhsNested _RhsNested;
-
-    enum {
-      PacketSize = ei_packet_traits<Scalar>::size
-    };
-
-    inline Product(const Lhs& lhs, const Rhs& rhs)
-      : m_lhs(lhs), m_rhs(rhs)
-    {
-      ei_assert(lhs.cols() == rhs.rows());
-    }
-
-    /** \internal */
-    template<typename DestDerived>
-    void _cacheFriendlyEval(DestDerived& res) const;
-
-    /** \internal */
-    template<typename DestDerived>
-    void _cacheFriendlyEvalAndAdd(DestDerived& res) const;
-
-  private:
-
-    inline int _rows() const { return m_lhs.rows(); }
-    inline int _cols() const { return m_rhs.cols(); }
-
-    const Scalar _coeff(int row, int col) const
-    {
-      Scalar res;
-      const bool unroll = CoeffReadCost <= EIGEN_UNROLLING_LIMIT;
-      if(unroll)
-      {
-        ei_product_unroller<Lhs::ColsAtCompileTime-1,
-                            unroll ? Lhs::ColsAtCompileTime : Dynamic,
-                            _LhsNested, _RhsNested>
-          ::run(row, col, m_lhs, m_rhs, res);
-      }
-      else
-      {
-        res = m_lhs.coeff(row, 0) * m_rhs.coeff(0, col);
-        for(int i = 1; i < m_lhs.cols(); i++)
-          res += m_lhs.coeff(row, i) * m_rhs.coeff(i, col);
-      }
-      return res;
-    }
-
-    template<int LoadMode>
-    const PacketScalar _packetCoeff(int row, int col) const
-    {
-      if(Lhs::ColsAtCompileTime <= EIGEN_UNROLLING_LIMIT)
-      {
-        PacketScalar res;
-        ei_packet_product_unroller<Flags&RowMajorBit ? true : false, Lhs::ColsAtCompileTime-1,
-                            Lhs::ColsAtCompileTime <= EIGEN_UNROLLING_LIMIT
-                              ? Lhs::ColsAtCompileTime : Dynamic,
-                            _LhsNested, _RhsNested, PacketScalar>
-          ::run(row, col, m_lhs, m_rhs, res);
-        return res;
-      }
-      else
-        return ProductPacketCoeffImpl<Product,Flags&RowMajorBit>::execute(*this, row, col);
-    }
-
-    const PacketScalar _packetCoeffRowMajor(int row, int col) const
-    {
-      PacketScalar res;
-      res = ei_pmul(ei_pset1(m_lhs.coeff(row, 0)),m_rhs.template packetCoeff<Aligned>(0, col));
-      for(int i = 1; i < m_lhs.cols(); i++)
-        res =  ei_pmadd(ei_pset1(m_lhs.coeff(row, i)), m_rhs.template packetCoeff<Aligned>(i, col), res);
-      return res;
-    }
-
-    const PacketScalar _packetCoeffColumnMajor(int row, int col) const
-    {
-      PacketScalar res;
-      res = ei_pmul(m_lhs.template packetCoeff<Aligned>(row, 0), ei_pset1(m_rhs.coeff(0, col)));
-      for(int i = 1; i < m_lhs.cols(); i++)
-        res =  ei_pmadd(m_lhs.template packetCoeff<Aligned>(row, i), ei_pset1(m_rhs.coeff(i, col)), res);
-      return res;
-    }
-
-    template<typename Lhs_, typename Rhs_, int EvalMode_, typename DestDerived_, bool DirectAccess_>
-    friend struct ei_cache_friendly_selector;
-
-  protected:
-    const LhsNested m_lhs;
-    const RhsNested m_rhs;
-};
-
-/** \returns the matrix product of \c *this and \a other.
-  *
-  * \sa lazy(), operator*=(const MatrixBase&)
-  */
-template<typename Derived>
-template<typename OtherDerived>
-inline const Product<Derived,OtherDerived>
-MatrixBase<Derived>::operator*(const MatrixBase<OtherDerived> &other) const
-{
-  return Product<Derived,OtherDerived>(derived(), other.derived());
-}
-
-/** replaces \c *this by \c *this * \a other.
-  *
-  * \returns a reference to \c *this
-  */
-template<typename Derived>
-template<typename OtherDerived>
-inline Derived &
-MatrixBase<Derived>::operator*=(const MatrixBase<OtherDerived> &other)
-{
-  return *this = *this * other;
-}
-
-/** \internal */
-template<typename Derived>
-template<typename Lhs,typename Rhs>
-inline Derived&
-MatrixBase<Derived>::operator+=(const Flagged<Product<Lhs,Rhs,CacheFriendlyProduct>, 0, EvalBeforeNestingBit | EvalBeforeAssigningBit>& other)
-{
-  other._expression()._cacheFriendlyEvalAndAdd(const_cast_derived());
-  return derived();
-}
-
-template<typename Derived>
-template<typename Lhs, typename Rhs>
-inline Derived& MatrixBase<Derived>::lazyAssign(const Product<Lhs,Rhs,CacheFriendlyProduct>& product)
-{
-  product._cacheFriendlyEval(derived());
-  return derived();
-}
-
-template<typename Lhs, typename Rhs, int EvalMode, typename DestDerived, bool DirectAccess>
-struct ei_cache_friendly_selector
-{
-  typedef Product<Lhs,Rhs,EvalMode> Prod;
-  typedef typename Prod::_LhsNested _LhsNested;
-  typedef typename Prod::_RhsNested _RhsNested;
-  typedef typename Prod::Scalar Scalar;
-  static inline void eval(const Prod& product, DestDerived& res)
-  {
-    if ( product._rows()>=EIGEN_CACHEFRIENDLY_PRODUCT_THRESHOLD
-      && product._cols()>=EIGEN_CACHEFRIENDLY_PRODUCT_THRESHOLD
-      && product.m_lhs.cols()>=EIGEN_CACHEFRIENDLY_PRODUCT_THRESHOLD
-    )
-    {
-      res.setZero();
-      ei_cache_friendly_product<Scalar>(
-        product._rows(), product._cols(), product.m_lhs.cols(),
-        _LhsNested::Flags&RowMajorBit, &(product.m_lhs.const_cast_derived().coeffRef(0,0)), product.m_lhs.stride(),
-        _RhsNested::Flags&RowMajorBit, &(product.m_rhs.const_cast_derived().coeffRef(0,0)), product.m_rhs.stride(),
-        Prod::Flags&RowMajorBit, &(res.coeffRef(0,0)), res.stride()
-      );
-    }
-    else
-    {
-      res = Product<_LhsNested,_RhsNested,NormalProduct>(product.m_lhs, product.m_rhs).lazy();
-    }
-  }
-
-  static inline void eval_and_add(const Prod& product, DestDerived& res)
-  {
-    if ( product._rows()>=EIGEN_CACHEFRIENDLY_PRODUCT_THRESHOLD
-      && product._cols()>=EIGEN_CACHEFRIENDLY_PRODUCT_THRESHOLD
-      && product.m_lhs.cols()>=EIGEN_CACHEFRIENDLY_PRODUCT_THRESHOLD
-    )
-    {
-      ei_cache_friendly_product<Scalar>(
-        product._rows(), product._cols(), product.m_lhs.cols(),
-        _LhsNested::Flags&RowMajorBit, &(product.m_lhs.const_cast_derived().coeffRef(0,0)), product.m_lhs.stride(),
-        _RhsNested::Flags&RowMajorBit, &(product.m_rhs.const_cast_derived().coeffRef(0,0)), product.m_rhs.stride(),
-        Prod::Flags&RowMajorBit, &(res.coeffRef(0,0)), res.stride()
-      );
-    }
-    else
-    {
-      res += Product<_LhsNested,_RhsNested,NormalProduct>(product.m_lhs, product.m_rhs).lazy();
-    }
-  }
-};
-
-template<typename Lhs, typename Rhs, int EvalMode, typename DestDerived>
-struct ei_cache_friendly_selector<Lhs,Rhs,EvalMode,DestDerived,false>
-{
-  typedef Product<Lhs,Rhs,EvalMode> Prod;
-  typedef typename Prod::_LhsNested _LhsNested;
-  typedef typename Prod::_RhsNested _RhsNested;
-  typedef typename Prod::Scalar Scalar;
-  static inline void eval(const Prod& product, DestDerived& res)
-  {
-    res = Product<_LhsNested,_RhsNested,NormalProduct>(product.m_lhs, product.m_rhs).lazy();
-  }
-
-  static inline void eval_and_add(const Prod& product, DestDerived& res)
-  {
-    res += Product<_LhsNested,_RhsNested,NormalProduct>(product.m_lhs, product.m_rhs).lazy();
-  }
-};
-
-template<typename Lhs, typename Rhs, int EvalMode>
-template<typename DestDerived>
-inline void Product<Lhs,Rhs,EvalMode>::_cacheFriendlyEval(DestDerived& res) const
-{
-  ei_cache_friendly_selector<Lhs,Rhs,EvalMode,DestDerived,
-                             _LhsNested::Flags&_RhsNested::Flags&DirectAccessBit>
-    ::eval(*this, res);
-}
-
-template<typename Lhs, typename Rhs, int EvalMode>
-template<typename DestDerived>
-inline void Product<Lhs,Rhs,EvalMode>::_cacheFriendlyEvalAndAdd(DestDerived& res) const
-{
-  ei_cache_friendly_selector<Lhs,Rhs,EvalMode,DestDerived,
-                             _LhsNested::Flags&_RhsNested::Flags&DirectAccessBit>
-    ::eval_and_add(*this, res);
-}
-
-#endif // EIGEN_PRODUCT_H
--- a/Eigen/src/Core/Transpose.h
+++ b/Eigen/src/Core/Transpose.h
@@ -107,7 +107,7 @@ template<typename Derived>
 inline Transpose<Derived>
 MatrixBase<Derived>::transpose()
 {
-  return Transpose<Derived>(derived());
+  return derived();
 }

 /** This is the const version of transpose(). \sa adjoint() */
@@ -115,7 +115,7 @@ template<typename Derived>
 inline const Transpose<Derived>
 MatrixBase<Derived>::transpose() const
 {
-  return Transpose<Derived>(derived());
+  return derived();
 }

 /** \returns an expression of the adjoint (i.e. conjugate transpose) of *this.
@@ -130,7 +130,7 @@ inline const Transpose<
               , NestByValueBit, 0> >
 MatrixBase<Derived>::adjoint() const
 {
-  return conjugate().temporary().transpose();
+  return conjugate().nestByValue();
 }

 #endif // EIGEN_TRANSPOSE_H
--- a/Eigen/src/Core/util/Constants.h
+++ b/Eigen/src/Core/util/Constants.h
@@ -53,16 +53,16 @@ const unsigned int HereditaryBits = RowMajorBit
                                  | EvalBeforeAssigningBit
                                  | LargeBit;

-// Possible values for the PartType parameter of part() and the ExtractType parameter of extract()
+// Possible values for the Mode parameter of part() and of extract()
 const unsigned int Upper = UpperTriangularBit;
 const unsigned int StrictlyUpper = UpperTriangularBit | ZeroDiagBit;
 const unsigned int Lower = LowerTriangularBit;
 const unsigned int StrictlyLower = LowerTriangularBit | ZeroDiagBit;

-// additional possible values for the PartType parameter of part()
+// additional possible values for the Mode parameter of part()
 const unsigned int SelfAdjoint = SelfAdjointBit;

-// additional possible values for the ExtractType parameter of extract()
+// additional possible values for the Mode parameter of extract()
 const unsigned int UnitUpper = UpperTriangularBit | UnitDiagBit;
 const unsigned int UnitLower = LowerTriangularBit | UnitDiagBit;