* add ei_pdiv intrinsic, make quotient functor vectorizable

* add vdw benchmark from Tim's real-world use case

Eigen/src/Core/DummyPacketMath.h

@@ -38,6 +38,9 @@ template <typename Scalar> inline Scalar ei_psub(const Scalar&  a, const Scalar&
 /** \internal \returns a * b (coeff-wise) */
 template <typename Scalar> inline Scalar ei_pmul(const Scalar&  a, const Scalar&  b) { return a * b; }
 
+/** \internal \returns a / b (coeff-wise) */
+template <typename Scalar> inline Scalar ei_pdiv(const Scalar&  a, const Scalar&  b) { return a / b; }
+
 /** \internal \returns a * b - c (coeff-wise) */
 template <typename Scalar> inline Scalar ei_pmadd(const Scalar&  a, const Scalar&  b, const Scalar&  c)
 { return ei_padd(ei_pmul(a, b),c); }

Eigen/src/Core/Functors.h

@@ -131,12 +131,18 @@ struct ei_functor_traits<ei_scalar_difference_op<Scalar> > {
   * \sa class CwiseBinaryOp, MatrixBase::cwiseQuotient()
   */
 template<typename Scalar> struct ei_scalar_quotient_op EIGEN_EMPTY_STRUCT {
-    inline const Scalar operator() (const Scalar& a, const Scalar& b) const { return a / b; }
+  inline const Scalar operator() (const Scalar& a, const Scalar& b) const { return a / b; }
+  template<typename PacketScalar>
+  inline const PacketScalar packetOp(const PacketScalar& a, const PacketScalar& b) const
+  { return ei_pdiv(a,b); }
 };
 template<typename Scalar>
-struct ei_functor_traits<ei_scalar_quotient_op<Scalar> >
-{ enum { Cost = 2 * NumTraits<Scalar>::MulCost, PacketAccess = false }; };
-
+struct ei_functor_traits<ei_scalar_quotient_op<Scalar> > {
+  enum {
+    Cost = 2 * NumTraits<Scalar>::MulCost,
+    PacketAccess = ei_packet_traits<Scalar>::size>1
+  };
+};
 
 // unary functors:
 
@@ -179,7 +185,7 @@ template<typename Scalar> struct ei_scalar_abs2_op EIGEN_EMPTY_STRUCT {
 };
 template<typename Scalar>
 struct ei_functor_traits<ei_scalar_abs2_op<Scalar> >
-{ enum { Cost = NumTraits<Scalar>::MulCost, PacketAccess = NumTraits<Scalar>::IsComplex==false && int(ei_packet_traits<Scalar>::size)>1 }; };
+{ enum { Cost = NumTraits<Scalar>::MulCost, PacketAccess = int(ei_packet_traits<Scalar>::size)>1 }; };
 
 /** \internal
   * \brief Template functor to compute the conjugate of a complex value
@@ -272,7 +278,7 @@ struct ei_functor_traits<ei_scalar_quotient1_impl<Scalar,false> >
   * \brief Template functor to divide a scalar by a fixed other one
   *
   * This functor is used to implement the quotient of a matrix by
-  * a scalar where the scalar type is not a floating point type.
+  * a scalar where the scalar type is not necessarily a floating point type.
   *
   * \sa class CwiseUnaryOp, MatrixBase::operator/
   */

Eigen/src/Core/Sum.h

@@ -216,7 +216,7 @@ struct ei_sum_impl<Derived, LinearVectorization, NoUnrolling>
       if(alignedSize == size) return res;
     }
     else // too small to vectorize anything.
-         // since this is dynamic-size hence inefficient anyway, don't try to optimize.
+         // since this is dynamic-size hence inefficient anyway for such small sizes, don't try to optimize.
     {
       res = Scalar(0);
     }

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

@@ -59,6 +59,8 @@ inline vector int    ei_pmul(const vector int     a, const vector int     b)
   return vec_add( lowProduct, highProduct );
 }
 
+inline vector float  ei_pdiv(const vector float   a, const vector float   b) { return vec_div(a,b); }
+
 inline vector float ei_pmadd(const vector float   a, const vector float   b, const vector float c) { return vec_madd(a, b, c); }
 
 inline vector float  ei_pmin(const vector float   a, const vector float   b) { return vec_min(a,b); }

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

@@ -55,6 +55,9 @@ template<> inline __m128i ei_pmul(const __m128i& a, const __m128i& b)
         _mm_setr_epi32(0xffffffff,0,0xffffffff,0)), 4));
 }
 
+template<> inline __m128  ei_pdiv(const __m128&  a, const __m128&  b) { return _mm_div_ps(a,b); }
+template<> inline __m128d ei_pdiv(const __m128d& a, const __m128d& b) { return _mm_div_pd(a,b); }
+
 // for some weird raisons, it has to be overloaded for packet integer
 template<> inline __m128i ei_pmadd(const __m128i& a, const __m128i& b, const __m128i& c) { return ei_padd(ei_pmul(a,b), c); }