Remove HasCast and fix packetmath cast tests.

The use of the `packet_traits<>::HasCast` field is currently inconsistent with
`type_casting_traits<>`, and is unused apart from within
`test/packetmath.cpp`. In addition, those packetmath cast tests do not
currently reflect how casts are performed in practice: they ignore the
`SrcCoeffRatio` and `TgtCoeffRatio` fields, assuming a 1:1 ratio.

Here we remove the unsed `HasCast`, and modify the packet cast tests to
better reflect their usage.

4 files changed, 95 insertions, 64 deletions

diff --git a/Eigen/src/Core/GenericPacketMath.h b/Eigen/src/Core/GenericPacketMath.h
index 7f35b3114..cb8848731 100644
--- a/Eigen/src/Core/GenericPacketMath.h
+++ b/Eigen/src/Core/GenericPacketMath.h
@@ -96,7 +96,6 @@ struct default_packet_traits
     HasRint   = 0,
     HasFloor  = 0,
     HasCeil   = 0,
-    HasCast   = 0, 
     HasSign   = 0
   };
 };
@@ -325,7 +324,7 @@ pcmp_eq(const Packet& a, const Packet& b) { return a==b ? ptrue(a) : pzero(a); }
 
 /** \internal \returns a < b or a==NaN or b==NaN as a bit mask */
 template<typename Packet> EIGEN_DEVICE_FUNC inline Packet
-pcmp_lt_or_nan(const Packet& a, const Packet& b) { return a>=b ? pzero(a) : ptrue(a); } 
+pcmp_lt_or_nan(const Packet& a, const Packet& b) { return a>=b ? pzero(a) : ptrue(a); }
 
 /** \internal \returns \a or \b for each field in packet according to \mask */
 template<typename Packet> EIGEN_DEVICE_FUNC inline Packet
@@ -392,7 +391,7 @@ ploaddup(const typename unpacket_traits<Packet>::type* from) { return *from; }
   * For instance, for a packet of 8 elements, 2 scalars will be read from \a *from and
   * replicated to form: {from[0],from[0],from[0],from[0],from[1],from[1],from[1],from[1]}
   * Currently, this function is only used in matrix products.
-  * For packet-size smaller or equal to 4, this function is equivalent to pload1 
+  * For packet-size smaller or equal to 4, this function is equivalent to pload1
   */
 template<typename Packet> EIGEN_DEVICE_FUNC inline Packet
 ploadquad(const typename unpacket_traits<Packet>::type* from)
diff --git a/Eigen/src/Core/arch/NEON/PacketMath.h b/Eigen/src/Core/arch/NEON/PacketMath.h
index 065c8100f..ddd43c377 100644
--- a/Eigen/src/Core/arch/NEON/PacketMath.h
+++ b/Eigen/src/Core/arch/NEON/PacketMath.h
@@ -121,7 +121,6 @@ struct packet_traits<float> : default_packet_traits
     size = 4,
     HasHalfPacket = 1,
 
-    HasCast      = 1,
     HasAdd       = 1,
     HasSub       = 1,
     HasShift     = 1,
@@ -162,7 +161,6 @@ struct packet_traits<int8_t> : default_packet_traits
     size = 16,
     HasHalfPacket = 1,
 
-    HasCast      = 1,
     HasAdd       = 1,
     HasSub       = 1,
     HasShift     = 1,
@@ -192,7 +190,6 @@ struct packet_traits<uint8_t> : default_packet_traits
     size = 16,
     HasHalfPacket = 1,
 
-    HasCast      = 1,
     HasAdd       = 1,
     HasSub       = 1,
     HasShift     = 1,
@@ -224,7 +221,6 @@ struct packet_traits<int16_t> : default_packet_traits
     size = 8,
     HasHalfPacket = 1,
 
-    HasCast      = 1,
     HasAdd       = 1,
     HasSub       = 1,
     HasShift     = 1,
@@ -254,7 +250,6 @@ struct packet_traits<uint16_t> : default_packet_traits
     size = 8,
     HasHalfPacket = 1,
 
-    HasCast      = 1,
     HasAdd       = 1,
     HasSub       = 1,
     HasShift     = 1,
@@ -285,7 +280,6 @@ struct packet_traits<int32_t> : default_packet_traits
     size = 4,
     HasHalfPacket = 1,
 
-    HasCast      = 1,
     HasAdd       = 1,
     HasSub       = 1,
     HasShift     = 1,
@@ -315,7 +309,6 @@ struct packet_traits<uint32_t> : default_packet_traits
     size = 4,
     HasHalfPacket = 1,
 
-    HasCast      = 1,
     HasAdd       = 1,
     HasSub       = 1,
     HasShift     = 1,
@@ -347,7 +340,6 @@ struct packet_traits<int64_t> : default_packet_traits
     size = 2,
     HasHalfPacket = 1,
 
-    HasCast      = 1,
     HasCmp       = 1,
     HasAdd       = 1,
     HasSub       = 1,
@@ -378,7 +370,6 @@ struct packet_traits<uint64_t> : default_packet_traits
     size = 2,
     HasHalfPacket = 1,
 
-    HasCast      = 1,
     HasCmp       = 1,
     HasAdd       = 1,
     HasSub       = 1,
@@ -3185,7 +3176,6 @@ template<> struct packet_traits<double>  : default_packet_traits
     size = 2,
     HasHalfPacket = 0,
 
-    HasCast      = 1,
     HasCmp       = 1,
     HasAdd       = 1,
     HasSub       = 1,
diff --git a/test/packetmath.cpp b/test/packetmath.cpp
index 540cf6a86..64ba28741 100644
--- a/test/packetmath.cpp
+++ b/test/packetmath.cpp
@@ -29,58 +29,113 @@ inline bool REF_SUB(const bool& a, const bool& b) { return a ^ b;}
 template <>
 inline bool REF_MUL(const bool& a, const bool& b) { return a && b;}
 
-template<typename FromScalar, typename FromPacket, typename ToScalar, typename ToPacket, bool CanCast = false>
+// Uses pcast to cast from one array to another.
+template<typename SrcPacket, typename TgtPacket, int SrcCoeffRatio, int TgtCoeffRatio>
+struct pcast_array {
+  static void cast(const typename internal::unpacket_traits<SrcPacket>::type* src, size_t size, typename internal::unpacket_traits<TgtPacket>::type* dst) {
+    static const int TgtPacketSize = internal::unpacket_traits<TgtPacket>::size;
+    for (size_t i=0; i<size; i+=TgtPacketSize) {
+      internal::pstoreu(dst+i, internal::pcast<SrcPacket,TgtPacket>(internal::ploadu<SrcPacket>(src+i)));
+    }
+  }
+};
+
+template<typename SrcPacket,typename TgtPacket>
+struct pcast_array<SrcPacket, TgtPacket, 2, 1>{
+  static void cast(const typename internal::unpacket_traits<SrcPacket>::type* src, size_t size, typename internal::unpacket_traits<TgtPacket>::type* dst) {
+    static const int SrcPacketSize = internal::unpacket_traits<SrcPacket>::size;
+    static const int TgtPacketSize = internal::unpacket_traits<TgtPacket>::size;
+    for (size_t i=0; i<size; i+=TgtPacketSize) {
+      SrcPacket a = internal::ploadu<SrcPacket>(src+i);
+      SrcPacket b = internal::ploadu<SrcPacket>(src+i+SrcPacketSize);
+      internal::pstoreu(dst+i, internal::pcast<SrcPacket,TgtPacket>(a, b));
+    }
+  }
+};
+
+template<typename SrcPacket, typename TgtPacket>
+struct pcast_array<SrcPacket, TgtPacket, 4, 1>{
+  static void cast(const typename internal::unpacket_traits<SrcPacket>::type* src, size_t size, typename internal::unpacket_traits<TgtPacket>::type* dst) {
+    static const int SrcPacketSize = internal::unpacket_traits<SrcPacket>::size;
+    static const int TgtPacketSize = internal::unpacket_traits<TgtPacket>::size;
+    for (size_t i=0; i<size; i+=TgtPacketSize) {
+      SrcPacket a = internal::ploadu<SrcPacket>(src+i);
+      SrcPacket b = internal::ploadu<SrcPacket>(src+i+SrcPacketSize);
+      SrcPacket c = internal::ploadu<SrcPacket>(src+i+2*SrcPacketSize);
+      SrcPacket d = internal::ploadu<SrcPacket>(src+i+3*SrcPacketSize);
+      internal::pstoreu(dst+i, internal::pcast<SrcPacket,TgtPacket>(a, b));
+    }
+  }
+};
+
+template<typename SrcPacket, typename TgtPacket, int SrcCoeffRatio, int TgtCoeffRatio, bool CanCast = false>
 struct test_cast_helper;
 
-template<typename FromScalar, typename FromPacket, typename ToScalar, typename ToPacket>
-struct test_cast_helper<FromScalar, FromPacket, ToScalar, ToPacket, false> {
+template<typename SrcPacket, typename TgtPacket, int SrcCoeffRatio, int TgtCoeffRatio>
+struct test_cast_helper<SrcPacket, TgtPacket, SrcCoeffRatio, TgtCoeffRatio, false> {
   static void run() {}
 };
 
-template<typename FromScalar, typename FromPacket, typename ToScalar, typename ToPacket>
-struct test_cast_helper<FromScalar, FromPacket, ToScalar, ToPacket, true> {
+template<typename SrcPacket, typename TgtPacket, int SrcCoeffRatio, int TgtCoeffRatio>
+struct test_cast_helper<SrcPacket, TgtPacket, SrcCoeffRatio, TgtCoeffRatio, true> {
   static void run() {
-    static const int PacketSize = internal::unpacket_traits<FromPacket>::size;
-    EIGEN_ALIGN_MAX FromScalar data1[PacketSize];
-    EIGEN_ALIGN_MAX ToScalar data2[PacketSize];
-    EIGEN_ALIGN_MAX ToScalar ref[PacketSize];
+    typedef typename internal::unpacket_traits<SrcPacket>::type SrcScalar;
+    typedef typename internal::unpacket_traits<TgtPacket>::type TgtScalar;
+    static const int SrcPacketSize = internal::unpacket_traits<SrcPacket>::size;
+    static const int TgtPacketSize = internal::unpacket_traits<TgtPacket>::size;
+    static const int DataSize = SrcPacketSize*SrcCoeffRatio;
+    VERIFY(DataSize == TgtPacketSize*TgtCoeffRatio && "Packet sizes and cast ratios are mismatched.");
+
+    EIGEN_ALIGN_MAX SrcScalar data1[DataSize];
+    EIGEN_ALIGN_MAX TgtScalar data2[DataSize];
+    EIGEN_ALIGN_MAX TgtScalar ref[DataSize];
 
     // Construct a packet of scalars that will not overflow when casting
-    for (int i=0; i<PacketSize; ++i) {
-      const FromScalar from_scalar = Array<FromScalar,1,1>::Random().value();
-      const ToScalar to_scalar = Array<ToScalar,1,1>::Random().value();
-      const FromScalar c = sizeof(ToScalar) > sizeof(FromScalar) ? static_cast<FromScalar>(to_scalar) : from_scalar;
-      data1[i] = (NumTraits<FromScalar>::IsSigned && !NumTraits<ToScalar>::IsSigned) ? numext::abs(c) : c;
+    for (int i=0; i<DataSize; ++i) {
+      const SrcScalar a = Array<SrcScalar,1,1>::Random().value();
+      const TgtScalar b = Array<TgtScalar,1,1>::Random().value();
+      const SrcScalar c = sizeof(TgtScalar) > sizeof(SrcScalar) ? static_cast<SrcScalar>(b) : a;
+      data1[i] = (NumTraits<SrcScalar>::IsSigned && !NumTraits<TgtScalar>::IsSigned) ? numext::abs(c) : c;
     }
 
-    for (int i=0; i<PacketSize; ++i)
-      ref[i] = static_cast<const ToScalar>(data1[i]);
-    internal::pstore(data2, internal::pcast<FromPacket, ToPacket>(internal::pload<FromPacket>(data1)));
+    for (int i=0; i<DataSize; ++i)
+      ref[i] = static_cast<const TgtScalar>(data1[i]);
 
-    VERIFY(test::areApprox(ref, data2, PacketSize) && "internal::pcast<>");
+    pcast_array<SrcPacket, TgtPacket, SrcCoeffRatio, TgtCoeffRatio>::cast(data1, DataSize, data2);
+
+    VERIFY(test::areApprox(ref, data2, DataSize) && "internal::pcast<>");
   }
 };
 
-template<typename FromPacket, typename ToScalar>
+template<typename SrcScalar, typename TgtScalar>
 void test_cast() {
-  typedef typename internal::unpacket_traits<FromPacket>::type FromScalar;
-  typedef typename internal::packet_traits<FromScalar> FromPacketTraits;
-  typedef typename internal::packet_traits<ToScalar>::type Full;
-  typedef typename internal::unpacket_traits<Full>::half Half;
-  typedef typename internal::unpacket_traits<typename internal::unpacket_traits<Full>::half>::half Quarter;
-
-  static const int PacketSize = internal::unpacket_traits<FromPacket>::size;
-  static const bool CanCast =
-      FromPacketTraits::HasCast &&
-      (PacketSize == internal::unpacket_traits<Full>::size ||
-      PacketSize == internal::unpacket_traits<Half>::size ||
-      PacketSize == internal::unpacket_traits<Quarter>::size);
-
-  typedef typename internal::conditional<internal::unpacket_traits<Quarter>::size == PacketSize, Quarter,
-      typename internal::conditional<internal::unpacket_traits<Half>::size == PacketSize, Half, Full>::type>::type
-      ToPacket;
-
-  test_cast_helper<FromScalar, FromPacket, ToScalar, ToPacket, CanCast>::run();
+  typedef typename internal::packet_traits<SrcScalar> SrcPacketTraits;
+  typedef typename internal::packet_traits<TgtScalar> TgtPacketTraits;
+  typedef typename internal::type_casting_traits<SrcScalar, TgtScalar> TypeCastingTraits;
+  static const int SrcCoeffRatio = TypeCastingTraits::SrcCoeffRatio;
+  static const int TgtCoeffRatio = TypeCastingTraits::TgtCoeffRatio;
+
+  static const bool HasFullCast = TypeCastingTraits::VectorizedCast;
+  static const bool HasHalfCast = HasFullCast && internal::packet_traits<SrcScalar>::HasHalfPacket && internal::packet_traits<TgtScalar>::HasHalfPacket;
+
+  test_cast_helper<typename SrcPacketTraits::type, typename TgtPacketTraits::type, SrcCoeffRatio, TgtCoeffRatio, HasFullCast>::run();
+  test_cast_helper<typename SrcPacketTraits::half, typename TgtPacketTraits::half, SrcCoeffRatio, TgtCoeffRatio, HasHalfCast>::run();
+}
+
+template<typename Scalar, typename Packet> void packetmath_pcast_ops() {
+  const static bool IsFullPacket = internal::is_same<typename internal::packet_traits<Scalar>::type,Packet>::value;
+  if (IsFullPacket) {
+    test_cast<Scalar, float>();
+    test_cast<Scalar, double>();
+    test_cast<Scalar, int8_t>();
+    test_cast<Scalar, uint8_t>();
+    test_cast<Scalar, int16_t>();
+    test_cast<Scalar, uint16_t>();
+    test_cast<Scalar, int32_t>();
+    test_cast<Scalar, uint32_t>();
+    test_cast<Scalar, int64_t>();
+    test_cast<Scalar, uint64_t>();
+  }
 }
 
 template<typename Scalar,typename Packet>
@@ -341,6 +396,7 @@ template<typename Scalar,typename Packet> void packetmath()
   CHECK_CWISE2_IF(true, internal::pand, internal::pand);
 
   packetmath_boolean_mask_ops<Scalar, Packet>();
+  packetmath_pcast_ops<Scalar, Packet>();
 }
 
 
@@ -584,19 +640,6 @@ template<typename Scalar,typename Packet> void packetmath_notcomplex()
 
   Array<Scalar,Dynamic,1>::Map(data1, PacketSize*4).setRandom();
 
-  if (PacketTraits::HasCast) {
-    test_cast<Packet, float>();
-    test_cast<Packet, double>();
-    test_cast<Packet, int8_t>();
-    test_cast<Packet, uint8_t>();
-    test_cast<Packet, int16_t>();
-    test_cast<Packet, uint16_t>();
-    test_cast<Packet, int32_t>();
-    test_cast<Packet, uint32_t>();
-    test_cast<Packet, int64_t>();
-    test_cast<Packet, uint64_t>();
-  }
-
   ref[0] = data1[0];
   for (int i=0; i<PacketSize; ++i)
     ref[0] = (std::min)(ref[0],data1[i]);
diff --git a/unsupported/test/cxx11_tensor_morphing.cpp b/unsupported/test/cxx11_tensor_morphing.cpp
index 59f4c34b3..e8c42a4cd 100644
--- a/unsupported/test/cxx11_tensor_morphing.cpp
+++ b/unsupported/test/cxx11_tensor_morphing.cpp
@@ -53,7 +53,6 @@ static void test_static_reshape() {
   Eigen::IndexList<type2index<2>, type2index<3>, type2index<7>> dim;
   Tensor<float, 3> reshaped = tensor.reshape(static_cast<Eigen::DSizes<ptrdiff_t,3>>(dim));
 
-
   for (int i = 0; i < 2; ++i) {
     for (int j = 0; j < 3; ++j) {
       for (int k = 0; k < 7; ++k) {