@smake/eigen 1.0.1 → 1.1.0

package/eigen/Eigen/src/Core/Visitor.h

@@ -10,7 +10,7 @@
 #ifndef EIGEN_VISITOR_H
 #define EIGEN_VISITOR_H
 
-namespace Eigen { 
+namespace Eigen {
 
 namespace internal {
 
@@ -40,6 +40,14 @@ struct visitor_impl<Visitor, Derived, 1>
   }
 };
 
+// This specialization enables visitors on empty matrices at compile-time
+template<typename Visitor, typename Derived>
+struct visitor_impl<Visitor, Derived, 0> {
+  EIGEN_DEVICE_FUNC
+  static inline void run(const Derived &/*mat*/, Visitor& /*visitor*/)
+  {}
+};
+
 template<typename Visitor, typename Derived>
 struct visitor_impl<Visitor, Derived, Dynamic>
 {
@@ -62,22 +70,22 @@ class visitor_evaluator
 public:
   EIGEN_DEVICE_FUNC
   explicit visitor_evaluator(const XprType &xpr) : m_evaluator(xpr), m_xpr(xpr) {}
-  
+
   typedef typename XprType::Scalar Scalar;
   typedef typename XprType::CoeffReturnType CoeffReturnType;
-  
+
   enum {
     RowsAtCompileTime = XprType::RowsAtCompileTime,
     CoeffReadCost = internal::evaluator<XprType>::CoeffReadCost
   };
-  
-  EIGEN_DEVICE_FUNC Index rows() const { return m_xpr.rows(); }
-  EIGEN_DEVICE_FUNC Index cols() const { return m_xpr.cols(); }
-  EIGEN_DEVICE_FUNC Index size() const { return m_xpr.size(); }
+
+  EIGEN_DEVICE_FUNC EIGEN_CONSTEXPR Index rows() const EIGEN_NOEXCEPT { return m_xpr.rows(); }
+  EIGEN_DEVICE_FUNC EIGEN_CONSTEXPR Index cols() const EIGEN_NOEXCEPT { return m_xpr.cols(); }
+  EIGEN_DEVICE_FUNC EIGEN_CONSTEXPR Index size() const EIGEN_NOEXCEPT { return m_xpr.size(); }
 
   EIGEN_DEVICE_FUNC CoeffReturnType coeff(Index row, Index col) const
   { return m_evaluator.coeff(row, col); }
-  
+
 protected:
   internal::evaluator<XprType> m_evaluator;
   const XprType &m_xpr;
@@ -99,6 +107,8 @@ protected:
   * \note compared to one or two \em for \em loops, visitors offer automatic
   * unrolling for small fixed size matrix.
   *
+  * \note if the matrix is empty, then the visitor is left unchanged.
+  *
   * \sa minCoeff(Index*,Index*), maxCoeff(Index*,Index*), DenseBase::redux()
   */
 template<typename Derived>
@@ -106,12 +116,15 @@ template<typename Visitor>
 EIGEN_DEVICE_FUNC
 void DenseBase<Derived>::visit(Visitor& visitor) const
 {
+  if(size()==0)
+    return;
+
   typedef typename internal::visitor_evaluator<Derived> ThisEvaluator;
   ThisEvaluator thisEval(derived());
-  
+
   enum {
     unroll =  SizeAtCompileTime != Dynamic
-           && SizeAtCompileTime * ThisEvaluator::CoeffReadCost + (SizeAtCompileTime-1) * internal::functor_traits<Visitor>::Cost <= EIGEN_UNROLLING_LIMIT
+           && SizeAtCompileTime * int(ThisEvaluator::CoeffReadCost) + (SizeAtCompileTime-1) * int(internal::functor_traits<Visitor>::Cost) <= EIGEN_UNROLLING_LIMIT
   };
   return internal::visitor_impl<Visitor, ThisEvaluator, unroll ? int(SizeAtCompileTime) : Dynamic>::run(thisEval, visitor);
 }
@@ -124,6 +137,9 @@ namespace internal {
 template <typename Derived>
 struct coeff_visitor
 {
+  // default initialization to avoid countless invalid maybe-uninitialized warnings by gcc
+  EIGEN_DEVICE_FUNC
+  coeff_visitor() : row(-1), col(-1), res(0) {}
   typedef typename Derived::Scalar Scalar;
   Index row, col;
   Scalar res;
@@ -141,7 +157,7 @@ struct coeff_visitor
   *
   * \sa DenseBase::minCoeff(Index*, Index*)
   */
-template <typename Derived>
+template <typename Derived, int NaNPropagation>
 struct min_coeff_visitor : coeff_visitor<Derived>
 {
   typedef typename Derived::Scalar Scalar;
@@ -157,8 +173,40 @@ struct min_coeff_visitor : coeff_visitor<Derived>
   }
 };
 
-template<typename Scalar>
-struct functor_traits<min_coeff_visitor<Scalar> > {
+template <typename Derived>
+struct min_coeff_visitor<Derived, PropagateNumbers> : coeff_visitor<Derived>
+{
+  typedef typename Derived::Scalar Scalar;
+  EIGEN_DEVICE_FUNC
+  void operator() (const Scalar& value, Index i, Index j)
+  {
+    if((numext::isnan)(this->res) || (!(numext::isnan)(value) && value < this->res))
+    {
+      this->res = value;
+      this->row = i;
+      this->col = j;
+    }
+  }
+};
+
+template <typename Derived>
+struct min_coeff_visitor<Derived, PropagateNaN> : coeff_visitor<Derived>
+{
+  typedef typename Derived::Scalar Scalar;
+  EIGEN_DEVICE_FUNC
+  void operator() (const Scalar& value, Index i, Index j)
+  {
+    if((numext::isnan)(value) || value < this->res)
+    {
+      this->res = value;
+      this->row = i;
+      this->col = j;
+    }
+  }
+};
+
+template<typename Scalar, int NaNPropagation>
+    struct functor_traits<min_coeff_visitor<Scalar, NaNPropagation> > {
   enum {
     Cost = NumTraits<Scalar>::AddCost
   };
@@ -169,10 +217,10 @@ struct functor_traits<min_coeff_visitor<Scalar> > {
   *
   * \sa DenseBase::maxCoeff(Index*, Index*)
   */
-template <typename Derived>
+template <typename Derived, int NaNPropagation>
 struct max_coeff_visitor : coeff_visitor<Derived>
 {
-  typedef typename Derived::Scalar Scalar; 
+  typedef typename Derived::Scalar Scalar;
   EIGEN_DEVICE_FUNC
   void operator() (const Scalar& value, Index i, Index j)
   {
@@ -185,8 +233,40 @@ struct max_coeff_visitor : coeff_visitor<Derived>
   }
 };
 
-template<typename Scalar>
-struct functor_traits<max_coeff_visitor<Scalar> > {
+template <typename Derived>
+struct max_coeff_visitor<Derived, PropagateNumbers> : coeff_visitor<Derived>
+{
+  typedef typename Derived::Scalar Scalar;
+  EIGEN_DEVICE_FUNC
+  void operator() (const Scalar& value, Index i, Index j)
+  {
+    if((numext::isnan)(this->res) || (!(numext::isnan)(value) && value > this->res))
+    {
+      this->res = value;
+      this->row = i;
+      this->col = j;
+    }
+  }
+};
+
+template <typename Derived>
+struct max_coeff_visitor<Derived, PropagateNaN> : coeff_visitor<Derived>
+{
+  typedef typename Derived::Scalar Scalar;
+  EIGEN_DEVICE_FUNC
+  void operator() (const Scalar& value, Index i, Index j)
+  {
+    if((numext::isnan)(value) || value > this->res)
+    {
+      this->res = value;
+      this->row = i;
+      this->col = j;
+    }
+  }
+};
+
+template<typename Scalar, int NaNPropagation>
+struct functor_traits<max_coeff_visitor<Scalar, NaNPropagation> > {
   enum {
     Cost = NumTraits<Scalar>::AddCost
   };
@@ -196,17 +276,24 @@ struct functor_traits<max_coeff_visitor<Scalar> > {
 
 /** \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.
+  *
+  * In case \c *this contains NaN, NaNPropagation determines the behavior:
+  *   NaNPropagation == PropagateFast : undefined
+  *   NaNPropagation == PropagateNaN : result is NaN
+  *   NaNPropagation == PropagateNumbers : result is maximum of elements that are not NaN
+  * \warning the matrix must be not empty, otherwise an assertion is triggered.
   *
   * \sa DenseBase::minCoeff(Index*), DenseBase::maxCoeff(Index*,Index*), DenseBase::visit(), DenseBase::minCoeff()
   */
 template<typename Derived>
-template<typename IndexType>
+template<int NaNPropagation, typename IndexType>
 EIGEN_DEVICE_FUNC
 typename internal::traits<Derived>::Scalar
 DenseBase<Derived>::minCoeff(IndexType* rowId, IndexType* colId) const
 {
-  internal::min_coeff_visitor<Derived> minVisitor;
+  eigen_assert(this->rows()>0 && this->cols()>0 && "you are using an empty matrix");
+
+  internal::min_coeff_visitor<Derived, NaNPropagation> minVisitor;
   this->visit(minVisitor);
   *rowId = minVisitor.row;
   if (colId) *colId = minVisitor.col;
@@ -214,18 +301,25 @@ DenseBase<Derived>::minCoeff(IndexType* rowId, IndexType* colId) const
 }
 
 /** \returns the minimum of all coefficients of *this and puts in *index its location.
-  * \warning the result is undefined if \c *this contains NaN. 
+  *
+  * In case \c *this contains NaN, NaNPropagation determines the behavior:
+  *   NaNPropagation == PropagateFast : undefined
+  *   NaNPropagation == PropagateNaN : result is NaN
+  *   NaNPropagation == PropagateNumbers : result is maximum of elements that are not NaN
+  * \warning the matrix must be not empty, otherwise an assertion is triggered.
   *
   * \sa DenseBase::minCoeff(IndexType*,IndexType*), DenseBase::maxCoeff(IndexType*,IndexType*), DenseBase::visit(), DenseBase::minCoeff()
   */
 template<typename Derived>
-template<typename IndexType>
+template<int NaNPropagation, typename IndexType>
 EIGEN_DEVICE_FUNC
 typename internal::traits<Derived>::Scalar
 DenseBase<Derived>::minCoeff(IndexType* index) const
 {
+  eigen_assert(this->rows()>0 && this->cols()>0 && "you are using an empty matrix");
+
   EIGEN_STATIC_ASSERT_VECTOR_ONLY(Derived)
-  internal::min_coeff_visitor<Derived> minVisitor;
+      internal::min_coeff_visitor<Derived, NaNPropagation> minVisitor;
   this->visit(minVisitor);
   *index = IndexType((RowsAtCompileTime==1) ? minVisitor.col : minVisitor.row);
   return minVisitor.res;
@@ -233,17 +327,24 @@ DenseBase<Derived>::minCoeff(IndexType* index) const
 
 /** \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. 
+  *
+  * In case \c *this contains NaN, NaNPropagation determines the behavior:
+  *   NaNPropagation == PropagateFast : undefined
+  *   NaNPropagation == PropagateNaN : result is NaN
+  *   NaNPropagation == PropagateNumbers : result is maximum of elements that are not NaN
+  * \warning the matrix must be not empty, otherwise an assertion is triggered.
   *
   * \sa DenseBase::minCoeff(IndexType*,IndexType*), DenseBase::visit(), DenseBase::maxCoeff()
   */
 template<typename Derived>
-template<typename IndexType>
+template<int NaNPropagation, typename IndexType>
 EIGEN_DEVICE_FUNC
 typename internal::traits<Derived>::Scalar
 DenseBase<Derived>::maxCoeff(IndexType* rowPtr, IndexType* colPtr) const
 {
-  internal::max_coeff_visitor<Derived> maxVisitor;
+  eigen_assert(this->rows()>0 && this->cols()>0 && "you are using an empty matrix");
+
+  internal::max_coeff_visitor<Derived, NaNPropagation> maxVisitor;
   this->visit(maxVisitor);
   *rowPtr = maxVisitor.row;
   if (colPtr) *colPtr = maxVisitor.col;
@@ -251,18 +352,25 @@ DenseBase<Derived>::maxCoeff(IndexType* rowPtr, IndexType* colPtr) const
 }
 
 /** \returns the maximum of all coefficients of *this and puts in *index its location.
-  * \warning the result is undefined if \c *this contains NaN.
+  *
+  * In case \c *this contains NaN, NaNPropagation determines the behavior:
+  *   NaNPropagation == PropagateFast : undefined
+  *   NaNPropagation == PropagateNaN : result is NaN
+  *   NaNPropagation == PropagateNumbers : result is maximum of elements that are not NaN
+  * \warning the matrix must be not empty, otherwise an assertion is triggered.
   *
   * \sa DenseBase::maxCoeff(IndexType*,IndexType*), DenseBase::minCoeff(IndexType*,IndexType*), DenseBase::visitor(), DenseBase::maxCoeff()
   */
 template<typename Derived>
-template<typename IndexType>
+template<int NaNPropagation, typename IndexType>
 EIGEN_DEVICE_FUNC
 typename internal::traits<Derived>::Scalar
 DenseBase<Derived>::maxCoeff(IndexType* index) const
 {
+  eigen_assert(this->rows()>0 && this->cols()>0 && "you are using an empty matrix");
+
   EIGEN_STATIC_ASSERT_VECTOR_ONLY(Derived)
-  internal::max_coeff_visitor<Derived> maxVisitor;
+      internal::max_coeff_visitor<Derived, NaNPropagation> maxVisitor;
   this->visit(maxVisitor);
   *index = (RowsAtCompileTime==1) ? maxVisitor.col : maxVisitor.row;
   return maxVisitor.res;

package/eigen/Eigen/src/Core/arch/AVX/Complex.h

@@ -22,6 +22,7 @@ struct Packet4cf
   __m256  v;
 };
 
+#ifndef EIGEN_VECTORIZE_AVX512
 template<> struct packet_traits<std::complex<float> >  : default_packet_traits
 {
   typedef Packet4cf type;
@@ -37,6 +38,7 @@ template<> struct packet_traits<std::complex<float> >  : default_packet_traits
     HasMul    = 1,
     HasDiv    = 1,
     HasNegate = 1,
+    HasSqrt   = 1,
     HasAbs    = 0,
     HasAbs2   = 0,
     HasMin    = 0,
@@ -44,8 +46,20 @@ template<> struct packet_traits<std::complex<float> >  : default_packet_traits
     HasSetLinear = 0
   };
 };
+#endif
 
-template<> struct unpacket_traits<Packet4cf> { typedef std::complex<float> type; enum {size=4, alignment=Aligned32}; typedef Packet2cf half; };
+template<> struct unpacket_traits<Packet4cf> {
+  typedef std::complex<float> type;
+  typedef Packet2cf half;
+  typedef Packet8f as_real;
+  enum {
+    size=4,
+    alignment=Aligned32,
+    vectorizable=true,
+    masked_load_available=false,
+    masked_store_available=false
+  };
+};
 
 template<> EIGEN_STRONG_INLINE Packet4cf padd<Packet4cf>(const Packet4cf& a, const Packet4cf& b) { return Packet4cf(_mm256_add_ps(a.v,b.v)); }
 template<> EIGEN_STRONG_INLINE Packet4cf psub<Packet4cf>(const Packet4cf& a, const Packet4cf& b) { return Packet4cf(_mm256_sub_ps(a.v,b.v)); }
@@ -67,10 +81,17 @@ template<> EIGEN_STRONG_INLINE Packet4cf pmul<Packet4cf>(const Packet4cf& a, con
   return Packet4cf(result);
 }
 
+template <>
+EIGEN_STRONG_INLINE Packet4cf pcmp_eq(const Packet4cf& a, const Packet4cf& b) {
+  __m256 eq = _mm256_cmp_ps(a.v, b.v, _CMP_EQ_OQ);
+  return Packet4cf(_mm256_and_ps(eq, _mm256_permute_ps(eq, 0xb1)));
+}
+
+template<> EIGEN_STRONG_INLINE Packet4cf ptrue<Packet4cf>(const Packet4cf& a) { return Packet4cf(ptrue(Packet8f(a.v))); }
 template<> EIGEN_STRONG_INLINE Packet4cf pand   <Packet4cf>(const Packet4cf& a, const Packet4cf& b) { return Packet4cf(_mm256_and_ps(a.v,b.v)); }
 template<> EIGEN_STRONG_INLINE Packet4cf por    <Packet4cf>(const Packet4cf& a, const Packet4cf& b) { return Packet4cf(_mm256_or_ps(a.v,b.v)); }
 template<> EIGEN_STRONG_INLINE Packet4cf pxor   <Packet4cf>(const Packet4cf& a, const Packet4cf& b) { return Packet4cf(_mm256_xor_ps(a.v,b.v)); }
-template<> EIGEN_STRONG_INLINE Packet4cf pandnot<Packet4cf>(const Packet4cf& a, const Packet4cf& b) { return Packet4cf(_mm256_andnot_ps(a.v,b.v)); }
+template<> EIGEN_STRONG_INLINE Packet4cf pandnot<Packet4cf>(const Packet4cf& a, const Packet4cf& b) { return Packet4cf(_mm256_andnot_ps(b.v,a.v)); }
 
 template<> EIGEN_STRONG_INLINE Packet4cf pload <Packet4cf>(const std::complex<float>* from) { EIGEN_DEBUG_ALIGNED_LOAD return Packet4cf(pload<Packet8f>(&numext::real_ref(*from))); }
 template<> EIGEN_STRONG_INLINE Packet4cf ploadu<Packet4cf>(const std::complex<float>* from) { EIGEN_DEBUG_UNALIGNED_LOAD return Packet4cf(ploadu<Packet8f>(&numext::real_ref(*from))); }
@@ -140,70 +161,12 @@ template<> EIGEN_STRONG_INLINE std::complex<float> predux<Packet4cf>(const Packe
                      Packet2cf(_mm256_extractf128_ps(a.v,1))));
 }
 
-template<> EIGEN_STRONG_INLINE Packet4cf preduxp<Packet4cf>(const Packet4cf* vecs)
-{
-  Packet8f t0 = _mm256_shuffle_ps(vecs[0].v, vecs[0].v, _MM_SHUFFLE(3, 1, 2 ,0));
-  Packet8f t1 = _mm256_shuffle_ps(vecs[1].v, vecs[1].v, _MM_SHUFFLE(3, 1, 2 ,0));
-  t0 = _mm256_hadd_ps(t0,t1);
-  Packet8f t2 = _mm256_shuffle_ps(vecs[2].v, vecs[2].v, _MM_SHUFFLE(3, 1, 2 ,0));
-  Packet8f t3 = _mm256_shuffle_ps(vecs[3].v, vecs[3].v, _MM_SHUFFLE(3, 1, 2 ,0));
-  t2 = _mm256_hadd_ps(t2,t3);
-  
-  t1 = _mm256_permute2f128_ps(t0,t2, 0 + (2<<4));
-  t3 = _mm256_permute2f128_ps(t0,t2, 1 + (3<<4));
-
-  return Packet4cf(_mm256_add_ps(t1,t3));
-}
-
 template<> EIGEN_STRONG_INLINE std::complex<float> predux_mul<Packet4cf>(const Packet4cf& a)
 {
   return predux_mul(pmul(Packet2cf(_mm256_extractf128_ps(a.v, 0)),
                          Packet2cf(_mm256_extractf128_ps(a.v, 1))));
 }
 
-template<int Offset>
-struct palign_impl<Offset,Packet4cf>
-{
-  static EIGEN_STRONG_INLINE void run(Packet4cf& first, const Packet4cf& second)
-  {
-    if (Offset==0) return;
-    palign_impl<Offset*2,Packet8f>::run(first.v, second.v);
-  }
-};
-
-template<> struct conj_helper<Packet4cf, Packet4cf, false,true>
-{
-  EIGEN_STRONG_INLINE Packet4cf pmadd(const Packet4cf& x, const Packet4cf& y, const Packet4cf& c) const
-  { return padd(pmul(x,y),c); }
-
-  EIGEN_STRONG_INLINE Packet4cf pmul(const Packet4cf& a, const Packet4cf& b) const
-  {
-    return internal::pmul(a, pconj(b));
-  }
-};
-
-template<> struct conj_helper<Packet4cf, Packet4cf, true,false>
-{
-  EIGEN_STRONG_INLINE Packet4cf pmadd(const Packet4cf& x, const Packet4cf& y, const Packet4cf& c) const
-  { return padd(pmul(x,y),c); }
-
-  EIGEN_STRONG_INLINE Packet4cf pmul(const Packet4cf& a, const Packet4cf& b) const
-  {
-    return internal::pmul(pconj(a), b);
-  }
-};
-
-template<> struct conj_helper<Packet4cf, Packet4cf, true,true>
-{
-  EIGEN_STRONG_INLINE Packet4cf pmadd(const Packet4cf& x, const Packet4cf& y, const Packet4cf& c) const
-  { return padd(pmul(x,y),c); }
-
-  EIGEN_STRONG_INLINE Packet4cf pmul(const Packet4cf& a, const Packet4cf& b) const
-  {
-    return pconj(internal::pmul(a, b));
-  }
-};
-
 EIGEN_MAKE_CONJ_HELPER_CPLX_REAL(Packet4cf,Packet8f)
 
 template<> EIGEN_STRONG_INLINE Packet4cf pdiv<Packet4cf>(const Packet4cf& a, const Packet4cf& b)
@@ -228,6 +191,7 @@ struct Packet2cd
   __m256d  v;
 };
 
+#ifndef EIGEN_VECTORIZE_AVX512
 template<> struct packet_traits<std::complex<double> >  : default_packet_traits
 {
   typedef Packet2cd type;
@@ -243,6 +207,7 @@ template<> struct packet_traits<std::complex<double> >  : default_packet_traits
     HasMul    = 1,
     HasDiv    = 1,
     HasNegate = 1,
+    HasSqrt   = 1,
     HasAbs    = 0,
     HasAbs2   = 0,
     HasMin    = 0,
@@ -250,8 +215,20 @@ template<> struct packet_traits<std::complex<double> >  : default_packet_traits
     HasSetLinear = 0
   };
 };
+#endif
 
-template<> struct unpacket_traits<Packet2cd> { typedef std::complex<double> type; enum {size=2, alignment=Aligned32}; typedef Packet1cd half; };
+template<> struct unpacket_traits<Packet2cd> {
+  typedef std::complex<double> type;
+  typedef Packet1cd half;
+  typedef Packet4d as_real;
+  enum {
+    size=2,
+    alignment=Aligned32,
+    vectorizable=true,
+    masked_load_available=false,
+    masked_store_available=false
+  };
+};
 
 template<> EIGEN_STRONG_INLINE Packet2cd padd<Packet2cd>(const Packet2cd& a, const Packet2cd& b) { return Packet2cd(_mm256_add_pd(a.v,b.v)); }
 template<> EIGEN_STRONG_INLINE Packet2cd psub<Packet2cd>(const Packet2cd& a, const Packet2cd& b) { return Packet2cd(_mm256_sub_pd(a.v,b.v)); }
@@ -272,10 +249,17 @@ template<> EIGEN_STRONG_INLINE Packet2cd pmul<Packet2cd>(const Packet2cd& a, con
   return Packet2cd(_mm256_addsub_pd(even, odd));
 }
 
+template <>
+EIGEN_STRONG_INLINE Packet2cd pcmp_eq(const Packet2cd& a, const Packet2cd& b) {
+  __m256d eq = _mm256_cmp_pd(a.v, b.v, _CMP_EQ_OQ);
+  return Packet2cd(pand(eq, _mm256_permute_pd(eq, 0x5)));
+}
+
+template<> EIGEN_STRONG_INLINE Packet2cd ptrue<Packet2cd>(const Packet2cd& a) { return Packet2cd(ptrue(Packet4d(a.v))); }
 template<> EIGEN_STRONG_INLINE Packet2cd pand   <Packet2cd>(const Packet2cd& a, const Packet2cd& b) { return Packet2cd(_mm256_and_pd(a.v,b.v)); }
 template<> EIGEN_STRONG_INLINE Packet2cd por    <Packet2cd>(const Packet2cd& a, const Packet2cd& b) { return Packet2cd(_mm256_or_pd(a.v,b.v)); }
 template<> EIGEN_STRONG_INLINE Packet2cd pxor   <Packet2cd>(const Packet2cd& a, const Packet2cd& b) { return Packet2cd(_mm256_xor_pd(a.v,b.v)); }
-template<> EIGEN_STRONG_INLINE Packet2cd pandnot<Packet2cd>(const Packet2cd& a, const Packet2cd& b) { return Packet2cd(_mm256_andnot_pd(a.v,b.v)); }
+template<> EIGEN_STRONG_INLINE Packet2cd pandnot<Packet2cd>(const Packet2cd& a, const Packet2cd& b) { return Packet2cd(_mm256_andnot_pd(b.v,a.v)); }
 
 template<> EIGEN_STRONG_INLINE Packet2cd pload <Packet2cd>(const std::complex<double>* from)
 { EIGEN_DEBUG_ALIGNED_LOAD return Packet2cd(pload<Packet4d>((const double*)from)); }
@@ -327,63 +311,12 @@ template<> EIGEN_STRONG_INLINE std::complex<double> predux<Packet2cd>(const Pack
                      Packet1cd(_mm256_extractf128_pd(a.v,1))));
 }
 
-template<> EIGEN_STRONG_INLINE Packet2cd preduxp<Packet2cd>(const Packet2cd* vecs)
-{
-  Packet4d t0 = _mm256_permute2f128_pd(vecs[0].v,vecs[1].v, 0 + (2<<4));
-  Packet4d t1 = _mm256_permute2f128_pd(vecs[0].v,vecs[1].v, 1 + (3<<4));
-
-  return Packet2cd(_mm256_add_pd(t0,t1));
-}
-
 template<> EIGEN_STRONG_INLINE std::complex<double> predux_mul<Packet2cd>(const Packet2cd& a)
 {
   return predux(pmul(Packet1cd(_mm256_extractf128_pd(a.v,0)),
                      Packet1cd(_mm256_extractf128_pd(a.v,1))));
 }
 
-template<int Offset>
-struct palign_impl<Offset,Packet2cd>
-{
-  static EIGEN_STRONG_INLINE void run(Packet2cd& first, const Packet2cd& second)
-  {
-    if (Offset==0) return;
-    palign_impl<Offset*2,Packet4d>::run(first.v, second.v);
-  }
-};
-
-template<> struct conj_helper<Packet2cd, Packet2cd, false,true>
-{
-  EIGEN_STRONG_INLINE Packet2cd pmadd(const Packet2cd& x, const Packet2cd& y, const Packet2cd& c) const
-  { return padd(pmul(x,y),c); }
-
-  EIGEN_STRONG_INLINE Packet2cd pmul(const Packet2cd& a, const Packet2cd& b) const
-  {
-    return internal::pmul(a, pconj(b));
-  }
-};
-
-template<> struct conj_helper<Packet2cd, Packet2cd, true,false>
-{
-  EIGEN_STRONG_INLINE Packet2cd pmadd(const Packet2cd& x, const Packet2cd& y, const Packet2cd& c) const
-  { return padd(pmul(x,y),c); }
-
-  EIGEN_STRONG_INLINE Packet2cd pmul(const Packet2cd& a, const Packet2cd& b) const
-  {
-    return internal::pmul(pconj(a), b);
-  }
-};
-
-template<> struct conj_helper<Packet2cd, Packet2cd, true,true>
-{
-  EIGEN_STRONG_INLINE Packet2cd pmadd(const Packet2cd& x, const Packet2cd& y, const Packet2cd& c) const
-  { return padd(pmul(x,y),c); }
-
-  EIGEN_STRONG_INLINE Packet2cd pmul(const Packet2cd& a, const Packet2cd& b) const
-  {
-    return pconj(internal::pmul(a, b));
-  }
-};
-
 EIGEN_MAKE_CONJ_HELPER_CPLX_REAL(Packet2cd,Packet4d)
 
 template<> EIGEN_STRONG_INLINE Packet2cd pdiv<Packet2cd>(const Packet2cd& a, const Packet2cd& b)
@@ -424,24 +357,12 @@ ptranspose(PacketBlock<Packet2cd,2>& kernel) {
  kernel.packet[0].v = tmp;
 }
 
-template<> EIGEN_STRONG_INLINE Packet4cf pinsertfirst(const Packet4cf& a, std::complex<float> b)
-{
-  return Packet4cf(_mm256_blend_ps(a.v,pset1<Packet4cf>(b).v,1|2));
-}
-
-template<> EIGEN_STRONG_INLINE Packet2cd pinsertfirst(const Packet2cd& a, std::complex<double> b)
-{
-  return Packet2cd(_mm256_blend_pd(a.v,pset1<Packet2cd>(b).v,1|2));
+template<> EIGEN_STRONG_INLINE Packet2cd psqrt<Packet2cd>(const Packet2cd& a) {
+  return psqrt_complex<Packet2cd>(a);
 }
 
-template<> EIGEN_STRONG_INLINE Packet4cf pinsertlast(const Packet4cf& a, std::complex<float> b)
-{
-  return Packet4cf(_mm256_blend_ps(a.v,pset1<Packet4cf>(b).v,(1<<7)|(1<<6)));
-}
-
-template<> EIGEN_STRONG_INLINE Packet2cd pinsertlast(const Packet2cd& a, std::complex<double> b)
-{
-  return Packet2cd(_mm256_blend_pd(a.v,pset1<Packet2cd>(b).v,(1<<3)|(1<<2)));
+template<> EIGEN_STRONG_INLINE Packet4cf psqrt<Packet4cf>(const Packet4cf& a) {
+  return psqrt_complex<Packet4cf>(a);
 }
 
 } // end namespace internal