complete int/double specialized traits for ZVector

1 files changed, 93 insertions, 419 deletions

diff --git a/Eigen/src/Core/arch/ZVector/PacketMath.h b/Eigen/src/Core/arch/ZVector/PacketMath.h
index c786aeec0..3586f87af 100755
--- a/Eigen/src/Core/arch/ZVector/PacketMath.h
+++ b/Eigen/src/Core/arch/ZVector/PacketMath.h
@@ -10,6 +10,8 @@
 #ifndef EIGEN_PACKET_MATH_ZVECTOR_H
 #define EIGEN_PACKET_MATH_ZVECTOR_H
 
+#include <stdint.h>
+
 namespace Eigen {
 
 namespace internal {
@@ -31,7 +33,6 @@ namespace internal {
 #define EIGEN_ARCH_DEFAULT_NUMBER_OF_REGISTERS  32
 #endif
 
-typedef __vector float               Packet4f;
 typedef __vector int                 Packet4i;
 typedef __vector unsigned int        Packet4ui;
 typedef __vector __bool int          Packet4bi;
@@ -42,20 +43,21 @@ typedef __vector unsigned long long  Packet2ul;
 typedef __vector long long           Packet2l;
 
 typedef union {
-  float f[4];
-  double d[2];
-  int i[4];
-  Packet4f v4f;
-  Packet4i v4i;
-  Packet2d v2d;
+  int32_t   i[4];
+  uint32_t ui[4];
+  int64_t   l[2];
+  uint64_t ul[2];
+  double    d[2];
+  Packet4i  v4i;
+  Packet4ui v4ui;
+  Packet2l  v2l;
+  Packet2ul v2ul;
+  Packet2d  v2d;
 } Packet;
 
 // We don't want to write the same code all the time, but we need to reuse the constants
 // and it doesn't really work to declare them global, so we define macros instead
 
-#define _EIGEN_DECLARE_CONST_FAST_Packet4f(NAME,X) \
-  Packet4f p4f_##NAME = reinterpret_cast<Packet4f>(vec_splat_s32(X))
-
 #define _EIGEN_DECLARE_CONST_FAST_Packet4i(NAME,X) \
   Packet4i p4i_##NAME = reinterpret_cast<Packet4i>(vec_splat_s32(X))
 
@@ -65,9 +67,6 @@ typedef union {
 #define _EIGEN_DECLARE_CONST_FAST_Packet2l(NAME,X) \
   Packet2l p2l_##NAME = reinterpret_cast<Packet2l>(vec_splat_s64(X))
 
-#define _EIGEN_DECLARE_CONST_Packet4f(NAME,X) \
-  Packet4f p4f_##NAME = pset1<Packet4f>(X)
-
 #define _EIGEN_DECLARE_CONST_Packet4i(NAME,X) \
   Packet4i p4i_##NAME = pset1<Packet4i>(X)
 
@@ -77,18 +76,14 @@ typedef union {
 #define _EIGEN_DECLARE_CONST_Packet2l(NAME,X) \
   Packet2l p2l_##NAME = pset1<Packet2l>(X)
 
-#define _EIGEN_DECLARE_CONST_Packet4f_FROM_INT(NAME,X) \
-  const Packet4f p4f_##NAME = reinterpret_cast<Packet4f>(pset1<Packet4i>(X))
-
 // These constants are endian-agnostic
-static _EIGEN_DECLARE_CONST_FAST_Packet4f(ZERO, 0); //{ 0.0, 0.0, 0.0, 0.0}
-static _EIGEN_DECLARE_CONST_FAST_Packet4i(ZERO, 0); //{ 0, 0, 0, 0,}
+//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_Packet2d(ZERO, 0);
 static _EIGEN_DECLARE_CONST_FAST_Packet2l(ZERO, 0);
+static _EIGEN_DECLARE_CONST_FAST_Packet2l(ONE, 1);
 
-static Packet4f p4f_ONE = { 1.0, 1.0, 1.0, 1.0 }; 
 static Packet2d p2d_ONE = { 1.0, 1.0 }; 
 static Packet2d p2d_ZERO_ = { -0.0, -0.0 };
 
@@ -98,7 +93,6 @@ 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_COUNTDOWN = { 0.0, 1.0, 2.0, 3.0 };
 static Packet4i p4i_COUNTDOWN = { 0, 1, 2, 3 };
 static Packet2d p2d_COUNTDOWN = reinterpret_cast<Packet2d>(vec_sld(reinterpret_cast<Packet16uc>(p2d_ZERO), reinterpret_cast<Packet16uc>(p2d_ONE), 8));
 
@@ -112,23 +106,23 @@ static Packet16uc p16uc_DUPLICATE32_HI = { 0,1,2,3, 0,1,2,3, 4,5,6,7, 4,5,6,7 };
 
 // Handle endianness properly while loading constants
 // Define global static constants:
-/*
-static Packet16uc p16uc_FORWARD = vec_lvsl(0, (float*)0);*/
+
+static Packet16uc p16uc_FORWARD =   { 0,1,2,3, 4,5,6,7, 8,9,10,11, 12,13,14,15 };
 static Packet16uc p16uc_REVERSE32 = { 12,13,14,15, 8,9,10,11, 4,5,6,7, 0,1,2,3 };
 static Packet16uc p16uc_REVERSE64 = { 8,9,10,11, 12,13,14,15, 0,1,2,3, 4,5,6,7 };
-/*
+
 static Packet16uc p16uc_PSET32_WODD   = vec_sld((Packet16uc) vec_splat((Packet4ui)p16uc_FORWARD, 0), (Packet16uc) vec_splat((Packet4ui)p16uc_FORWARD, 2), 8);//{ 0,1,2,3, 0,1,2,3, 8,9,10,11, 8,9,10,11 };
 static Packet16uc p16uc_PSET32_WEVEN  = vec_sld(p16uc_DUPLICATE32_HI, (Packet16uc) vec_splat((Packet4ui)p16uc_FORWARD, 3), 8);//{ 4,5,6,7, 4,5,6,7, 12,13,14,15, 12,13,14,15 };
-static Packet16uc p16uc_HALF64_0_16 = vec_sld((Packet16uc)p4i_ZERO, vec_splat((Packet16uc) vec_abs(p4i_MINUS16), 3), 8);      //{ 0,0,0,0, 0,0,0,0, 16,16,16,16, 16,16,16,16};
+/*static Packet16uc p16uc_HALF64_0_16 = vec_sld((Packet16uc)p4i_ZERO, vec_splat((Packet16uc) vec_abs(p4i_MINUS16), 3), 8);      //{ 0,0,0,0, 0,0,0,0, 16,16,16,16, 16,16,16,16};
 
-static Packet16uc p16uc_PSET64_HI = (Packet16uc) vec_mergeh((Packet4ui)p16uc_PSET32_WODD, (Packet4ui)p16uc_PSET32_WEVEN);     //{ 0,1,2,3, 4,5,6,7, 0,1,2,3, 4,5,6,7 };
+static Packet16uc p16uc_PSET64_HI = (Packet16uc) vec_mergeh((Packet4ui)p16uc_PSET32_WODD, (Packet4ui)p16uc_PSET32_WEVEN);     //{ 0,1,2,3, 4,5,6,7, 0,1,2,3, 4,5,6,7 };*/
 static Packet16uc p16uc_PSET64_LO = (Packet16uc) vec_mergel((Packet4ui)p16uc_PSET32_WODD, (Packet4ui)p16uc_PSET32_WEVEN);     //{ 8,9,10,11, 12,13,14,15, 8,9,10,11, 12,13,14,15 };
-static Packet16uc p16uc_TRANSPOSE64_HI = vec_add(p16uc_PSET64_HI, p16uc_HALF64_0_16);                                         //{ 0,1,2,3, 4,5,6,7, 16,17,18,19, 20,21,22,23};
+/*static Packet16uc p16uc_TRANSPOSE64_HI = vec_add(p16uc_PSET64_HI, p16uc_HALF64_0_16);                                         //{ 0,1,2,3, 4,5,6,7, 16,17,18,19, 20,21,22,23};
 static Packet16uc p16uc_TRANSPOSE64_LO = vec_add(p16uc_PSET64_LO, p16uc_HALF64_0_16);                                         //{ 8,9,10,11, 12,13,14,15, 24,25,26,27, 28,29,30,31};*/
 static Packet16uc p16uc_TRANSPOSE64_HI = { 0,1,2,3, 4,5,6,7, 16,17,18,19, 20,21,22,23};
 static Packet16uc p16uc_TRANSPOSE64_LO = { 8,9,10,11, 12,13,14,15, 24,25,26,27, 28,29,30,31};
 
-static Packet16uc p16uc_COMPLEX32_REV = vec_sld(p16uc_REVERSE32, p16uc_REVERSE32, 8);                                         //{ 4,5,6,7, 0,1,2,3, 12,13,14,15, 8,9,10,11 };
+//static Packet16uc p16uc_COMPLEX32_REV = vec_sld(p16uc_REVERSE32, p16uc_REVERSE32, 8);                                         //{ 4,5,6,7, 0,1,2,3, 12,13,14,15, 8,9,10,11 };
 
 //static Packet16uc p16uc_COMPLEX32_REV2 = vec_sld(p16uc_FORWARD, p16uc_FORWARD, 8);                                            //{ 8,9,10,11, 12,13,14,15, 0,1,2,3, 4,5,6,7 };
 
@@ -139,29 +133,6 @@ static Packet16uc p16uc_COMPLEX32_REV = vec_sld(p16uc_REVERSE32, p16uc_REVERSE32
   #define EIGEN_ZVECTOR_PREFETCH(ADDR) asm( "   pfd [%[addr]]\n" :: [addr] "r" (ADDR) : "cc" );
 #endif
 
-template<> struct packet_traits<float>  : default_packet_traits
-{
-  typedef Packet4f type;
-  typedef Packet4f half;
-  enum {
-    Vectorizable = 1,
-    AlignedOnScalar = 1,
-    size = 4,
-    HasHalfPacket = 0,
-
-    // FIXME check the Has*
-    HasAdd  = 1,
-    HasSub  = 1,
-    HasMul  = 1,
-    HasDiv  = 1,
-    HasSin  = 0,
-    HasCos  = 0,
-    HasLog  = 1,
-    HasExp  = 1,
-    HasSqrt = 0
-  };
-};
-
 template<> struct packet_traits<int>    : default_packet_traits
 {
   typedef Packet4i type;
@@ -178,11 +149,7 @@ template<> struct packet_traits<int>    : default_packet_traits
     HasSub  = 1,
     HasMul  = 1,
     HasDiv  = 1,
-    HasSin  = 0,
-    HasCos  = 0,
-    HasLog  = 1,
-    HasExp  = 1,
-    HasSqrt = 0
+    HasBlend = 1
   };
 };
 
@@ -203,89 +170,61 @@ template<> struct packet_traits<double> : default_packet_traits
     HasDiv  = 1,
     HasSin  = 0,
     HasCos  = 0,
-    HasLog  = 1,
+    HasLog  = 0,
     HasExp  = 1,
-    HasSqrt = 1
+    HasSqrt = 1,
+    HasRsqrt = 1,
+    HasBlend = 1,
+    HasRound = 1,
+    HasFloor = 1,
+    HasCeil = 1
   };
 };
 
-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<Packet2d> { typedef double type; enum {size=2, alignment=Aligned16}; typedef Packet2d half; };
 
-inline std::ostream & operator <<(std::ostream & s, const Packet16uc & v)
+inline std::ostream & operator <<(std::ostream & s, const Packet4i & v)
 {
-  union {
-    Packet16uc   v;
-    unsigned char n[16];
-  } vt;
-  vt.v = v;
-  for (int i=0; i< 16; i++)
-    s << (int)vt.n[i] << ", ";
+  Packet vt;
+  vt.v4i = v;
+  s << vt.i[0] << ", " << vt.i[1] << ", " << vt.i[2] << ", " << vt.i[3];
   return s;
 }
 
-inline std::ostream & operator <<(std::ostream & s, const Packet4f & v)
+inline std::ostream & operator <<(std::ostream & s, const Packet4ui & v)
 {
-  union {
-    Packet4f   v;
-    float n[4];
-  } vt;
-  vt.v = v;
-  s << vt.n[0] << ", " << vt.n[1] << ", " << vt.n[2] << ", " << vt.n[3];
+  Packet vt;
+  vt.v4ui = v;
+  s << vt.ui[0] << ", " << vt.ui[1] << ", " << vt.ui[2] << ", " << vt.ui[3];
   return s;
 }
 
-inline std::ostream & operator <<(std::ostream & s, const Packet4i & v)
+inline std::ostream & operator <<(std::ostream & s, const Packet2l & v)
 {
-  union {
-    Packet4i   v;
-    int n[4];
-  } vt;
-  vt.v = v;
-  s << vt.n[0] << ", " << vt.n[1] << ", " << vt.n[2] << ", " << vt.n[3];
+  Packet vt;
+  vt.v2l = v;
+  s << vt.l[0] << ", " << vt.l[1];
   return s;
 }
 
-inline std::ostream & operator <<(std::ostream & s, const Packet4ui & v)
+inline std::ostream & operator <<(std::ostream & s, const Packet2ul & v)
 {
-  union {
-    Packet4ui   v;
-    unsigned int n[4];
-  } vt;
-  vt.v = v;
-  s << vt.n[0] << ", " << vt.n[1] << ", " << vt.n[2] << ", " << vt.n[3];
+  Packet vt;
+  vt.v2ul = v;
+  s << vt.ul[0] << ", " << vt.ul[1] ;
   return s;
 }
 
 inline std::ostream & operator <<(std::ostream & s, const Packet2d & v)
 {
-  union {
-    Packet2d   v;
-    double n[2];
-  } vt;
-  vt.v = v;
-  s << vt.n[0] << ", " << vt.n[1];
+  Packet vt;
+  vt.v2d = v;
+  s << vt.d[0] << ", " << vt.d[1];
   return s;
 }
 
 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 = reinterpret_cast<Packet4f>(vec_sld(reinterpret_cast<Packet4i>(first), reinterpret_cast<Packet4i>(second), 4)); break;
-    case 2:
-      first = reinterpret_cast<Packet4f>(vec_sld(reinterpret_cast<Packet4i>(first), reinterpret_cast<Packet4i>(second), 8)); break;
-    case 3:
-      first = reinterpret_cast<Packet4f>(vec_sld(reinterpret_cast<Packet4i>(first), reinterpret_cast<Packet4i>(second), 12)); break;
-    }
-  }
-};
-
-template<int Offset>
 struct palign_impl<Offset,Packet4i>
 {
   static EIGEN_STRONG_INLINE void run(Packet4i& first, const Packet4i& second)
@@ -311,16 +250,6 @@ struct palign_impl<Offset,Packet2d>
   }
 };
 
-// Need to define them first or we get specialization after instantiation errors
-template<> EIGEN_STRONG_INLINE Packet4f pload<Packet4f>(const float* from)
-{
-  // FIXME: No intrinsic yet
-  EIGEN_DEBUG_ALIGNED_LOAD
-  Packet *vfrom;
-  vfrom = (Packet *) from;
-  return vfrom->v4f;
-}
-
 template<> EIGEN_STRONG_INLINE Packet4i pload<Packet4i>(const int*     from)
 {
   // FIXME: No intrinsic yet
@@ -338,15 +267,6 @@ template<> EIGEN_STRONG_INLINE Packet2d pload<Packet2d>(const double* from)
   return vfrom->v2d;
 }
 
-template<> EIGEN_STRONG_INLINE void pstore<float>(float*   to, const Packet4f& from)
-{
-  // FIXME: No intrinsic yet
-  EIGEN_DEBUG_ALIGNED_STORE
-  Packet *vto;
-  vto = (Packet *) to;
-  vto->v4f = from;
-}
-
 template<> EIGEN_STRONG_INLINE void pstore<int>(int*       to, const Packet4i& from)
 {
   // FIXME: No intrinsic yet
@@ -365,15 +285,6 @@ template<> EIGEN_STRONG_INLINE void pstore<double>(double*   to, const Packet2d&
   vto->v2d = from;
 }
 
-template<> EIGEN_STRONG_INLINE Packet4f pset1<Packet4f>(const float&  from)
-{
-  // FIXME: Check if proper intrinsic exists
-  Packet res;
-  res.f[0] = from;
-  res.v4f = reinterpret_cast<Packet4f>(vec_splats(res.i[0]));
-  return res.v4f;
-}
-
 template<> EIGEN_STRONG_INLINE Packet4i pset1<Packet4i>(const int&    from)
 {
   return vec_splats(from);
@@ -386,17 +297,6 @@ template<> EIGEN_STRONG_INLINE Packet2d pset1<Packet2d>(const double&  from) {
 }
 
 template<> EIGEN_STRONG_INLINE void
-pbroadcast4<Packet4f>(const float *a,
-                      Packet4f& a0, Packet4f& a1, Packet4f& a2, Packet4f& a3)
-{
-  a3 = pload<Packet4f>(a);
-  a0 = reinterpret_cast<Packet4f>(vec_splat(reinterpret_cast<Packet4i>(a3), 0));
-  a1 = reinterpret_cast<Packet4f>(vec_splat(reinterpret_cast<Packet4i>(a3), 1));
-  a2 = reinterpret_cast<Packet4f>(vec_splat(reinterpret_cast<Packet4i>(a3), 2));
-  a3 = reinterpret_cast<Packet4f>(vec_splat(reinterpret_cast<Packet4i>(a3), 3));
-}
-
-template<> EIGEN_STRONG_INLINE void
 pbroadcast4<Packet4i>(const int *a,
                       Packet4i& a0, Packet4i& a1, Packet4i& a2, Packet4i& a3)
 {
@@ -419,16 +319,6 @@ pbroadcast4<Packet2d>(const double *a,
   a3 = vec_splat(a3, 1);
 }
 
-template<> EIGEN_DEVICE_FUNC inline Packet4f pgather<float, Packet4f>(const float* from, Index stride)
-{
-  float EIGEN_ALIGN16 af[4];
-  af[0] = from[0*stride];
-  af[1] = from[1*stride];
-  af[2] = from[2*stride];
-  af[3] = from[3*stride];
- return pload<Packet4f>(af);
-}
-
 template<> EIGEN_DEVICE_FUNC inline Packet4i pgather<int, Packet4i>(const int* from, Index stride)
 {
   int EIGEN_ALIGN16 ai[4];
@@ -447,16 +337,6 @@ template<> EIGEN_DEVICE_FUNC inline Packet2d pgather<double, Packet2d>(const dou
  return pload<Packet2d>(af);
 }
 
-template<> EIGEN_DEVICE_FUNC inline void pscatter<float, Packet4f>(float* to, const Packet4f& from, Index stride)
-{
-  float EIGEN_ALIGN16 af[4];
-  pstore<float>(af, from);
-  to[0*stride] = af[0];
-  to[1*stride] = af[1];
-  to[2*stride] = af[2];
-  to[3*stride] = af[3];
-}
-
 template<> EIGEN_DEVICE_FUNC inline void pscatter<int, Packet4i>(int* to, const Packet4i& from, Index stride)
 {
   int EIGEN_ALIGN16 ai[4];
@@ -475,159 +355,52 @@ template<> EIGEN_DEVICE_FUNC inline void pscatter<double, Packet2d>(double* to,
   to[1*stride] = af[1];
 }
 
-/*
-template<> EIGEN_STRONG_INLINE Packet4f psub<Packet4f>(const Packet4f& a, const Packet4f& b) { return vec_sub(a,b); }
-template<> EIGEN_STRONG_INLINE Packet4i psub<Packet4i>(const Packet4i& a, const Packet4i& b) { return vec_sub(a,b); }
-template<> EIGEN_STRONG_INLINE Packet2d psub<Packet2d>(const Packet2d& a, const Packet2d& b) { return vec_sub(a,b); }
-
-template<> EIGEN_STRONG_INLINE Packet4f pnegate(const Packet4f& a) { return psub<Packet4f>(p4f_ZERO, a); }
-template<> EIGEN_STRONG_INLINE Packet4i pnegate(const Packet4i& a) { return psub<Packet4i>(p4i_ZERO, a); }
-template<> EIGEN_STRONG_INLINE Packet2d pnegate(const Packet2d& a) { return psub<Packet2d>(p2d_ZERO, a); }
-*/
-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 Packet2d pconj(const Packet2d& a) { return a; }
-
-template<> EIGEN_STRONG_INLINE Packet4f pmadd(const Packet4f& a, const Packet4f& b, const Packet4f& c)
-{
-  return a;
-}
-
-template<> EIGEN_STRONG_INLINE Packet4i pmadd(const Packet4i& a, const Packet4i& b, const Packet4i& c)
-{
-  return reinterpret_cast<Packet4i>(__builtin_s390_vmalf(reinterpret_cast<Packet4ui>(a), reinterpret_cast<Packet4ui>(b), reinterpret_cast<Packet4ui>(c)));
-}
+template<> EIGEN_STRONG_INLINE Packet4i padd<Packet4i>(const Packet4i& a, const Packet4i& b) { return (a + b); }
+template<> EIGEN_STRONG_INLINE Packet2d padd<Packet2d>(const Packet2d& a, const Packet2d& b) { return (a + b); }
 
-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 psub<Packet4i>(const Packet4i& a, const Packet4i& b) { return (a - b); }
+template<> EIGEN_STRONG_INLINE Packet2d psub<Packet2d>(const Packet2d& a, const Packet2d& b) { return (a - b); }
 
-template<> EIGEN_STRONG_INLINE Packet4f pmul<Packet4f>(const Packet4f& a, const Packet4f& b)
-{
-  return pmadd<Packet4f>(a,b,p4f_ZERO);
-}
-
-template<> EIGEN_STRONG_INLINE Packet4i pmul<Packet4i>(const Packet4i& a, const Packet4i& b)
-{
-  return pmadd<Packet4i>(a,b,p4i_ZERO);
-}
-
-template<> EIGEN_STRONG_INLINE Packet2d pmul<Packet2d>(const Packet2d& a, const Packet2d& b)
-{
-  return pmadd<Packet2d>(a,b,p2d_ZERO);
-}
-
-/*template<> EIGEN_STRONG_INLINE Packet4f pdiv<Packet4f>(const Packet4f& a, const Packet4f& b)
-{
-#ifndef __VSX__  // VSX actually provides a div instruction
-  Packet4f t, y_0, y_1;
+template<> EIGEN_STRONG_INLINE Packet4i pmul<Packet4i>(const Packet4i& a, const Packet4i& b) { return (a * b); }
+template<> EIGEN_STRONG_INLINE Packet2d pmul<Packet2d>(const Packet2d& a, const Packet2d& b) { return (a * b); }
 
-  // Altivec does not offer a divide instruction, we have to do a reciprocal approximation
-  y_0 = vec_re(b);
+template<> EIGEN_STRONG_INLINE Packet4i pdiv<Packet4i>(const Packet4i& a, const Packet4i& b) { return (a / b); }
+template<> EIGEN_STRONG_INLINE Packet2d pdiv<Packet2d>(const Packet2d& a, const Packet2d& b) { return (a / b); }
 
-  // Do one Newton-Raphson iteration to get the needed accuracy
-  t   = vec_nmsub(y_0, b, p4f_ONE);
-  y_1 = vec_madd(y_0, t, y_0);
+template<> EIGEN_STRONG_INLINE Packet4i pnegate(const Packet4i& a) { return (-a); }
+template<> EIGEN_STRONG_INLINE Packet2d pnegate(const Packet2d& a) { return (-a); }
 
-  return vec_madd(a, y_1, p4f_ZERO);
-#else
-  return vec_div(a, b);
-#endif
-}
-
-template<> EIGEN_STRONG_INLINE Packet4i pdiv<Packet4i>(const Packet4i& a, const Packet4i& b)
-{ eigen_assert(false && "packet integer division are not supported by AltiVec");
-  return pset1<Packet4i>(0);
-}
-template<> EIGEN_STRONG_INLINE Packet2d pdiv<Packet2d>(const Packet2d& a, const Packet2d& b) { return vec_div(a,b); }
-*/
+template<> EIGEN_STRONG_INLINE Packet4i pconj(const Packet4i& a) { return a; }
+template<> EIGEN_STRONG_INLINE Packet2d pconj(const Packet2d& a) { return a; }
 
-template<> EIGEN_STRONG_INLINE Packet4f padd<Packet4f>(const Packet4f& a, const Packet4f& b) { return pmadd<Packet4f>(a, p4f_ONE, b); }
-template<> EIGEN_STRONG_INLINE Packet4i padd<Packet4i>(const Packet4i& a, const Packet4i& b) { return pmadd<Packet4i>(a, p4i_ONE, b); }
-template<> EIGEN_STRONG_INLINE Packet2d padd<Packet2d>(const Packet2d& a, const Packet2d& b) { return pmadd<Packet2d>(a, p2d_ONE, b); }
+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 Packet2d pmadd(const Packet2d& a, const Packet2d& b, const Packet2d& c) { return vec_madd(a, b, c); }
 
-template<> EIGEN_STRONG_INLINE Packet4f plset<Packet4f>(const float& a) { return padd<Packet4f>(pset1<Packet4f>(a), p4f_COUNTDOWN); }
 template<> EIGEN_STRONG_INLINE Packet4i plset<Packet4i>(const int& a)   { return padd<Packet4i>(pset1<Packet4i>(a), p4i_COUNTDOWN); }
 template<> EIGEN_STRONG_INLINE Packet2d plset<Packet2d>(const double& a) { return padd<Packet2d>(pset1<Packet2d>(a), p2d_COUNTDOWN); }
 
-
-template<> EIGEN_STRONG_INLINE Packet4f pmin<Packet4f>(const Packet4f& a, const Packet4f& b) { return a; /*vec_min(a, b);*/ }
 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 pmax<Packet4f>(const Packet4f& a, const Packet4f& b) { return a; /*vec_max(a, b);*/ }
 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 pand<Packet4f>(const Packet4f& a, const Packet4f& b)
-{
-  return reinterpret_cast<Packet4f>(vec_and(reinterpret_cast<Packet4i>(a), reinterpret_cast<Packet4i>(b)));
-}
-
-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 por<Packet4f>(const Packet4f& a, const Packet4f& b)
-{
-  return reinterpret_cast<Packet4f>(vec_or(reinterpret_cast<Packet4i>(a), reinterpret_cast<Packet4i>(b)));
-}
-
-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 pxor<Packet4f>(const Packet4f& a, const Packet4f& b)
-{
-  return reinterpret_cast<Packet4f>(vec_xor(reinterpret_cast<Packet4i>(a), reinterpret_cast<Packet4i>(b)));
-}
-
-template<> EIGEN_STRONG_INLINE Packet4i pxor<Packet4i>(const Packet4i& a, const Packet4i& b)
-{
-  return vec_xor(a, b);
-}
+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 Packet2d pandnot<Packet2d>(const Packet2d& a, const Packet2d& b)
-{
-  return vec_and(a, vec_nor(b, b));
-}
+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 Packet2d pxor<Packet2d>(const Packet2d& a, const Packet2d& b)
-{
-  return vec_xor(a, b);
-}
+template<> EIGEN_STRONG_INLINE Packet4i pxor<Packet4i>(const Packet4i& a, const Packet4i& b)    { return vec_xor(a, 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)
-{
-  return pand<Packet4f>(a, reinterpret_cast<Packet4f>(vec_nor(reinterpret_cast<Packet4i>(b), reinterpret_cast<Packet4i>(b))));
-}
+template<> EIGEN_STRONG_INLINE Packet2d pxor<Packet2d>(const Packet2d& a, const Packet2d& b)    { return vec_xor(a, b); }
+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 Packet4i pandnot<Packet4i>(const Packet4i& a, const Packet4i& b)
-{
-  return pand<Packet4i>(a, vec_nor(b, b));
-}
-
-template<> EIGEN_STRONG_INLINE Packet4f ploadu<Packet4f>(const float* from)
-{
-  EIGEN_DEBUG_UNALIGNED_LOAD
-  Packet *vfrom;
-  vfrom = (Packet *) from;
-  return vfrom->v4f;
-}
+template<> EIGEN_STRONG_INLINE Packet2d pround<Packet2d>(const Packet2d& a) { return vec_round(a); }
+template<> EIGEN_STRONG_INLINE Packet2d pceil<Packet2d>(const Packet2d& a) { return vec_ceil(a); }
+template<> EIGEN_STRONG_INLINE Packet2d pfloor<Packet2d>(const Packet2d& a) { return vec_floor(a); }
 
 template<> EIGEN_STRONG_INLINE Packet4i ploadu<Packet4i>(const int*     from)
 {
@@ -645,14 +418,6 @@ template<> EIGEN_STRONG_INLINE Packet2d ploadu<Packet2d>(const double* from)
   return vfrom->v2d;
 }
 
-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);
-  return (Packet4f) vec_perm((Packet16uc)(p), (Packet16uc)(p), p16uc_DUPLICATE32_HI);
-}
-
 template<> EIGEN_STRONG_INLINE Packet4i ploaddup<Packet4i>(const int*     from)
 {
   Packet4i p;
@@ -669,14 +434,6 @@ template<> EIGEN_STRONG_INLINE Packet2d ploaddup<Packet2d>(const double*   from)
   return vec_perm(p, p, p16uc_PSET64_HI);
 }
 
-template<> EIGEN_STRONG_INLINE void pstoreu<float>(float*   to, const Packet4f& from)
-{
-  EIGEN_DEBUG_UNALIGNED_STORE
-  Packet *vto;
-  vto = (Packet *) to;
-  vto->v4f = from;
-}
-
 template<> EIGEN_STRONG_INLINE void pstoreu<int>(int*       to, const Packet4i& from)
 {
   EIGEN_DEBUG_UNALIGNED_STORE
@@ -693,30 +450,12 @@ template<> EIGEN_STRONG_INLINE void pstoreu<double>(double*  to, const Packet2d&
   vto->v2d = from;
 }
 
-template<> EIGEN_STRONG_INLINE void prefetch<float>(const float* addr)
-{
-  EIGEN_ZVECTOR_PREFETCH(addr);
-}
-
-template<> EIGEN_STRONG_INLINE void prefetch<int>(const int*     addr)
-{
-  EIGEN_ZVECTOR_PREFETCH(addr);
-}
+template<> EIGEN_STRONG_INLINE void prefetch<int>(const int*     addr)   { EIGEN_ZVECTOR_PREFETCH(addr); }
+template<> EIGEN_STRONG_INLINE void prefetch<double>(const double* addr) { EIGEN_ZVECTOR_PREFETCH(addr); }
 
-template<> EIGEN_STRONG_INLINE void prefetch<double>(const double* addr)
-{
-  EIGEN_ZVECTOR_PREFETCH(addr);
-}
-
-template<> EIGEN_STRONG_INLINE float  pfirst<Packet4f>(const Packet4f& a) { float EIGEN_ALIGN16 x[4]; pstore(x, a); return x[0]; }
 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 double pfirst<Packet2d>(const Packet2d& a) { double EIGEN_ALIGN16 x[2]; pstore(x, a); return x[0]; }
 
-template<> EIGEN_STRONG_INLINE Packet4f preverse(const Packet4f& a)
-{
-  return reinterpret_cast<Packet4f>(vec_perm(reinterpret_cast<Packet16uc>(a), reinterpret_cast<Packet16uc>(a), p16uc_REVERSE32));
-}
-
 template<> EIGEN_STRONG_INLINE Packet4i preverse(const Packet4i& a)
 {
   return reinterpret_cast<Packet4i>(vec_perm(reinterpret_cast<Packet16uc>(a), reinterpret_cast<Packet16uc>(a), p16uc_REVERSE32));
@@ -727,20 +466,9 @@ 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 Packet4f pabs(const Packet4f& a) { return a; /*vec_abs(a);*/ }
 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 float predux<Packet4f>(const Packet4f& a)
-{
-  Packet4f b, sum;
-  b   = reinterpret_cast<Packet4f>(vec_sld(reinterpret_cast<Packet4i>(a), reinterpret_cast<Packet4i>(a), 8));
-  sum = padd<Packet4f>(a, b);
-  b   = reinterpret_cast<Packet4f>(vec_sld(reinterpret_cast<Packet4i>(sum), reinterpret_cast<Packet4i>(sum), 4));
-  sum = padd<Packet4f>(sum, b);
-  return pfirst(sum);
-}
-
 template<> EIGEN_STRONG_INLINE int predux<Packet4i>(const Packet4i& a)
 {
   Packet4i b, sum;
@@ -760,34 +488,6 @@ template<> EIGEN_STRONG_INLINE double predux<Packet2d>(const Packet2d& a)
 }
 
 
-template<> EIGEN_STRONG_INLINE Packet4f preduxp<Packet4f>(const Packet4f* vecs)
-{
-  Packet4f v[4], sum[4];
-
-  // It's easier and faster to transpose then add as columns
-  // Check: http://www.freevec.org/function/matrix_4x4_transpose_floats for explanation
-  // Do the transpose, first set of moves
-  v[0] = reinterpret_cast<Packet4f>(vec_mergeh(reinterpret_cast<Packet4i>(vecs[0]), reinterpret_cast<Packet4i>(vecs[2])));
-  v[1] = reinterpret_cast<Packet4f>(vec_mergeh(reinterpret_cast<Packet4i>(vecs[0]), reinterpret_cast<Packet4i>(vecs[2])));
-  v[2] = reinterpret_cast<Packet4f>(vec_mergeh(reinterpret_cast<Packet4i>(vecs[1]), reinterpret_cast<Packet4i>(vecs[3])));
-  v[3] = reinterpret_cast<Packet4f>(vec_mergeh(reinterpret_cast<Packet4i>(vecs[1]), reinterpret_cast<Packet4i>(vecs[3])));
-  // Get the resulting vectors
-  sum[0] = reinterpret_cast<Packet4f>(vec_mergeh(reinterpret_cast<Packet4i>(v[0]), reinterpret_cast<Packet4i>(v[2])));
-  sum[1] = reinterpret_cast<Packet4f>(vec_mergeh(reinterpret_cast<Packet4i>(v[0]), reinterpret_cast<Packet4i>(v[2])));
-  sum[2] = reinterpret_cast<Packet4f>(vec_mergeh(reinterpret_cast<Packet4i>(v[1]), reinterpret_cast<Packet4i>(v[3])));
-  sum[3] = reinterpret_cast<Packet4f>(vec_mergeh(reinterpret_cast<Packet4i>(v[1]), reinterpret_cast<Packet4i>(v[3])));
-
-  // Now do the summation:
-  // Lines 0+1
-  sum[0] = padd<Packet4f>(sum[0], sum[1]);
-  // Lines 2+3
-  sum[1] = padd<Packet4f>(sum[2], sum[3]);
-  // Add the results
-  sum[0] = padd<Packet4f>(sum[0], sum[1]);
-
-  return sum[0];
-}
-
 template<> EIGEN_STRONG_INLINE Packet4i preduxp<Packet4i>(const Packet4i* vecs)
 {
   Packet4i v[4], sum[4];
@@ -829,13 +529,6 @@ template<> EIGEN_STRONG_INLINE Packet2d preduxp<Packet2d>(const Packet2d* vecs)
 
 // Other reduction functions:
 // mul
-template<> EIGEN_STRONG_INLINE float predux_mul<Packet4f>(const Packet4f& a)
-{
-  Packet4f prod;
-  prod = pmul(a, reinterpret_cast<Packet4f>(vec_sld(reinterpret_cast<Packet4i>(a), reinterpret_cast<Packet4i>(a), 8)));
-  return pfirst(pmul(prod, reinterpret_cast<Packet4f>(vec_sld(reinterpret_cast<Packet4i>(prod), reinterpret_cast<Packet4i>(prod), 4))));
-}
-
 template<> EIGEN_STRONG_INLINE int predux_mul<Packet4i>(const Packet4i& a)
 {
   EIGEN_ALIGN16 int aux[4];
@@ -849,14 +542,6 @@ template<> EIGEN_STRONG_INLINE double predux_mul<Packet2d>(const Packet2d& a)
 }
 
 // min
-template<> EIGEN_STRONG_INLINE float predux_min<Packet4f>(const Packet4f& a)
-{
-  Packet4f b, res;
-  b = pmin<Packet4f>(a, reinterpret_cast<Packet4f>(vec_sld(reinterpret_cast<Packet4i>(a), reinterpret_cast<Packet4i>(a), 8)));
-  res = pmin<Packet4f>(b, reinterpret_cast<Packet4f>(vec_sld(reinterpret_cast<Packet4i>(b), reinterpret_cast<Packet4i>(b), 4)));
-  return pfirst(res);
-}
-
 template<> EIGEN_STRONG_INLINE int predux_min<Packet4i>(const Packet4i& a)
 {
   Packet4i b, res;
@@ -871,14 +556,6 @@ template<> EIGEN_STRONG_INLINE double predux_min<Packet2d>(const Packet2d& a)
 }
 
 // max
-template<> EIGEN_STRONG_INLINE float predux_max<Packet4f>(const Packet4f& a)
-{
-  Packet4f b, res;
-  b = pmax<Packet4f>(a, reinterpret_cast<Packet4f>(vec_sld(reinterpret_cast<Packet4i>(a), reinterpret_cast<Packet4i>(a), 8)));
-  res = pmax<Packet4f>(b, reinterpret_cast<Packet4f>(vec_sld(reinterpret_cast<Packet4i>(b), reinterpret_cast<Packet4i>(b), 4)));
-  return pfirst(res);
-}
-
 template<> EIGEN_STRONG_INLINE int predux_max<Packet4i>(const Packet4i& a)
 {
   Packet4i b, res;
@@ -894,25 +571,11 @@ template<> EIGEN_STRONG_INLINE double predux_max<Packet2d>(const Packet2d& a)
 }
 
 EIGEN_DEVICE_FUNC inline void
-ptranspose(PacketBlock<Packet4f,4>& kernel) {
-  Packet4f t0, t1, t2, t3;
-  t0 = reinterpret_cast<Packet4f>(vec_mergeh(reinterpret_cast<Packet4i>(kernel.packet[0]), reinterpret_cast<Packet4i>(kernel.packet[2])));
-  t1 = reinterpret_cast<Packet4f>(vec_mergel(reinterpret_cast<Packet4i>(kernel.packet[0]), reinterpret_cast<Packet4i>(kernel.packet[2])));
-  t2 = reinterpret_cast<Packet4f>(vec_mergeh(reinterpret_cast<Packet4i>(kernel.packet[1]), reinterpret_cast<Packet4i>(kernel.packet[3])));
-  t3 = reinterpret_cast<Packet4f>(vec_mergel(reinterpret_cast<Packet4i>(kernel.packet[1]), reinterpret_cast<Packet4i>(kernel.packet[3])));
-  kernel.packet[0] = reinterpret_cast<Packet4f>(vec_mergeh(reinterpret_cast<Packet4i>(t0), reinterpret_cast<Packet4i>(t2)));
-  kernel.packet[1] = reinterpret_cast<Packet4f>(vec_mergel(reinterpret_cast<Packet4i>(t0), reinterpret_cast<Packet4i>(t2)));
-  kernel.packet[2] = reinterpret_cast<Packet4f>(vec_mergeh(reinterpret_cast<Packet4i>(t1), reinterpret_cast<Packet4i>(t3)));
-  kernel.packet[3] = reinterpret_cast<Packet4f>(vec_mergel(reinterpret_cast<Packet4i>(t1), reinterpret_cast<Packet4i>(t3)));
-}
-
-EIGEN_DEVICE_FUNC inline void
 ptranspose(PacketBlock<Packet4i,4>& kernel) {
-  Packet4i t0, t1, t2, t3;
-  t0 = vec_mergeh(kernel.packet[0], kernel.packet[2]);
-  t1 = vec_mergel(kernel.packet[0], kernel.packet[2]);
-  t2 = vec_mergeh(kernel.packet[1], kernel.packet[3]);
-  t3 = vec_mergel(kernel.packet[1], kernel.packet[3]);
+  Packet4i t0 = vec_mergeh(kernel.packet[0], kernel.packet[2]);
+  Packet4i t1 = vec_mergel(kernel.packet[0], kernel.packet[2]);
+  Packet4i t2 = vec_mergeh(kernel.packet[1], kernel.packet[3]);
+  Packet4i t3 = vec_mergel(kernel.packet[1], kernel.packet[3]);
   kernel.packet[0] = vec_mergeh(t0, t2);
   kernel.packet[1] = vec_mergel(t0, t2);
   kernel.packet[2] = vec_mergeh(t1, t3);
@@ -921,13 +584,24 @@ ptranspose(PacketBlock<Packet4i,4>& kernel) {
 
 EIGEN_DEVICE_FUNC inline void
 ptranspose(PacketBlock<Packet2d,2>& kernel) {
-  Packet2d t0, t1;
-  t0 = vec_perm(kernel.packet[0], kernel.packet[1], p16uc_TRANSPOSE64_HI);
-  t1 = vec_perm(kernel.packet[0], kernel.packet[1], p16uc_TRANSPOSE64_LO);
+  Packet2d t0 = vec_perm(kernel.packet[0], kernel.packet[1], p16uc_TRANSPOSE64_HI);
+  Packet2d t1 = vec_perm(kernel.packet[0], kernel.packet[1], p16uc_TRANSPOSE64_LO);
   kernel.packet[0] = t0;
   kernel.packet[1] = t1;
 }
 
+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 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));
+  return vec_sel(elsePacket, thenPacket, mask);
+}
+
 } // end namespace internal
 
 } // end namespace Eigen