Pulled latest update from the eigen main codebase

6 files changed, 113 insertions, 19 deletions

diff --git a/unsupported/Eigen/CXX11/src/Tensor/CMakeLists.txt b/unsupported/Eigen/CXX11/src/Tensor/CMakeLists.txt
new file mode 100644
index 000000000..6d4b3ea0d
--- /dev/null
+++ b/unsupported/Eigen/CXX11/src/Tensor/CMakeLists.txt
@@ -0,0 +1,6 @@
+FILE(GLOB Eigen_CXX11_Tensor_SRCS "*.h")
+
+INSTALL(FILES
+  ${Eigen_CXX11_Tensor_SRCS}
+  DESTINATION ${INCLUDE_INSTALL_DIR}/unsupported/Eigen/CXX11/src/Tensor COMPONENT Devel
+  )
diff --git a/unsupported/Eigen/CXX11/src/Tensor/TensorBase.h b/unsupported/Eigen/CXX11/src/Tensor/TensorBase.h
index b2b28826a..ca6681a1f 100644
--- a/unsupported/Eigen/CXX11/src/Tensor/TensorBase.h
+++ b/unsupported/Eigen/CXX11/src/Tensor/TensorBase.h
@@ -526,48 +526,101 @@ class TensorBase<Derived, WriteAccessors> : public TensorBase<Derived, ReadOnlyA
     }
 
     EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
-    TensorLayoutSwapOp<Derived>
+    const TensorLayoutSwapOp<const Derived>
     swap_layout() const {
+      return TensorLayoutSwapOp<const Derived>(derived());
+    }
+    EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
+    TensorLayoutSwapOp<Derived>
+    swap_layout() {
       return TensorLayoutSwapOp<Derived>(derived());
     }
+
     template <typename Axis, typename OtherDerived> EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
-    TensorConcatenationOp<const Axis, Derived, OtherDerived>
+    const TensorConcatenationOp<const Axis, const Derived, const OtherDerived>
     concatenate(const OtherDerived& other, const Axis& axis) const {
-      return TensorConcatenationOp<const Axis, Derived, OtherDerived>(derived(), other.derived(), axis);
+      return TensorConcatenationOp<const Axis, const Derived, const OtherDerived>(derived(), other, axis);
     }
+    template <typename Axis, typename OtherDerived> EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
+    TensorConcatenationOp<const Axis, Derived, OtherDerived>
+    concatenate(const OtherDerived& other, const Axis& axis) {
+      return TensorConcatenationOp<const Axis, Derived, OtherDerived>(derived(), other, axis);
+    }
+
     template <typename NewDimensions> EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
-    TensorReshapingOp<const NewDimensions, Derived>
+    const TensorReshapingOp<const NewDimensions, const Derived>
     reshape(const NewDimensions& newDimensions) const {
+      return TensorReshapingOp<const NewDimensions, const Derived>(derived(), newDimensions);
+    }
+    template <typename NewDimensions> EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
+    TensorReshapingOp<const NewDimensions, Derived>
+    reshape(const NewDimensions& newDimensions) {
       return TensorReshapingOp<const NewDimensions, Derived>(derived(), newDimensions);
     }
+
     template <typename StartIndices, typename Sizes> EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
-    TensorSlicingOp<const StartIndices, const Sizes, Derived>
+    const TensorSlicingOp<const StartIndices, const Sizes, const Derived>
     slice(const StartIndices& startIndices, const Sizes& sizes) const {
+      return TensorSlicingOp<const StartIndices, const Sizes, const Derived>(derived(), startIndices, sizes);
+    }
+    template <typename StartIndices, typename Sizes> EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
+    TensorSlicingOp<const StartIndices, const Sizes, Derived>
+    slice(const StartIndices& startIndices, const Sizes& sizes) {
       return TensorSlicingOp<const StartIndices, const Sizes, Derived>(derived(), startIndices, sizes);
     }
+
     template <DenseIndex DimId> EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
-    TensorChippingOp<DimId, Derived>
+    const TensorChippingOp<DimId, const Derived>
     chip(const Index offset) const {
+      return TensorChippingOp<DimId, const Derived>(derived(), offset, DimId);
+    }
+    template <Index DimId> EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
+    TensorChippingOp<DimId, Derived>
+    chip(const Index offset) {
       return TensorChippingOp<DimId, Derived>(derived(), offset, DimId);
     }
+
     EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
-    TensorChippingOp<Dynamic, Derived>
+    const TensorChippingOp<Dynamic, const Derived>
     chip(const Index offset, const Index dim) const {
+      return TensorChippingOp<Dynamic, const Derived>(derived(), offset, dim);
+    }
+    EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
+    TensorChippingOp<Dynamic, Derived>
+    chip(const Index offset, const Index dim) {
       return TensorChippingOp<Dynamic, Derived>(derived(), offset, dim);
     }
+
     template <typename ReverseDimensions> EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
-    TensorReverseOp<const ReverseDimensions, Derived>
+    const TensorReverseOp<const ReverseDimensions, const Derived>
     reverse(const ReverseDimensions& rev) const {
+      return TensorReverseOp<const ReverseDimensions, const Derived>(derived(), rev);
+    }
+    template <typename ReverseDimensions> EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
+    TensorReverseOp<const ReverseDimensions, Derived>
+    reverse(const ReverseDimensions& rev) {
       return TensorReverseOp<const ReverseDimensions, Derived>(derived(), rev);
     }
+
     template <typename Shuffle> EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
-    TensorShufflingOp<const Shuffle, Derived>
+    const TensorShufflingOp<const Shuffle, const Derived>
     shuffle(const Shuffle& shuffle) const {
+      return TensorShufflingOp<const Shuffle, const Derived>(derived(), shuffle);
+    }
+    template <typename Shuffle> EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
+    TensorShufflingOp<const Shuffle, Derived>
+    shuffle(const Shuffle& shuffle) {
       return TensorShufflingOp<const Shuffle, Derived>(derived(), shuffle);
     }
+
     template <typename Strides> EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
-    TensorStridingOp<const Strides, Derived>
+    const TensorStridingOp<const Strides, const Derived>
     stride(const Strides& strides) const {
+      return TensorStridingOp<const Strides, const Derived>(derived(), strides);
+    }
+    template <typename Strides> EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
+    TensorStridingOp<const Strides, Derived>
+    stride(const Strides& strides) {
       return TensorStridingOp<const Strides, Derived>(derived(), strides);
     }
 
diff --git a/unsupported/Eigen/CXX11/src/Tensor/TensorDevice.h b/unsupported/Eigen/CXX11/src/Tensor/TensorDevice.h
index 649bdb308..7a67c56b3 100644
--- a/unsupported/Eigen/CXX11/src/Tensor/TensorDevice.h
+++ b/unsupported/Eigen/CXX11/src/Tensor/TensorDevice.h
@@ -21,8 +21,7 @@ namespace Eigen {
   * Example:
   *    C.device(EIGEN_GPU) = A + B;
   *
-  * Todo: thread pools.
-  * Todo: operator +=, -=, *= and so on.
+  * Todo: operator *= and /=.
   */
 
 template <typename ExpressionType, typename DeviceType> class TensorDevice {
@@ -50,6 +49,18 @@ template <typename ExpressionType, typename DeviceType> class TensorDevice {
       return *this;
     }
 
+    template<typename OtherDerived>
+    EIGEN_STRONG_INLINE TensorDevice& operator-=(const OtherDerived& other) {
+      typedef typename OtherDerived::Scalar Scalar;
+      typedef TensorCwiseBinaryOp<internal::scalar_difference_op<Scalar>, const ExpressionType, const OtherDerived> Difference;
+      Difference difference(m_expression, other);
+      typedef TensorAssignOp<ExpressionType, const Difference> Assign;
+      Assign assign(m_expression, difference);
+      static const bool Vectorize = TensorEvaluator<const Assign, DeviceType>::PacketAccess;
+      internal::TensorExecutor<const Assign, DeviceType, Vectorize>::run(assign, m_device);
+      return *this;
+    }
+
   protected:
     const DeviceType& m_device;
     ExpressionType& m_expression;
@@ -82,6 +93,18 @@ template <typename ExpressionType> class TensorDevice<ExpressionType, ThreadPool
       return *this;
     }
 
+    template<typename OtherDerived>
+    EIGEN_STRONG_INLINE TensorDevice& operator-=(const OtherDerived& other) {
+      typedef typename OtherDerived::Scalar Scalar;
+      typedef TensorCwiseBinaryOp<internal::scalar_difference_op<Scalar>, const ExpressionType, const OtherDerived> Difference;
+      Difference difference(m_expression, other);
+      typedef TensorAssignOp<ExpressionType, const Difference> Assign;
+      Assign assign(m_expression, difference);
+      static const bool Vectorize = TensorEvaluator<const Assign, ThreadPoolDevice>::PacketAccess;
+      internal::TensorExecutor<const Assign, ThreadPoolDevice, Vectorize>::run(assign, m_device);
+      return *this;
+    }
+
   protected:
     const ThreadPoolDevice& m_device;
     ExpressionType& m_expression;
@@ -114,6 +137,18 @@ template <typename ExpressionType> class TensorDevice<ExpressionType, GpuDevice>
       return *this;
     }
 
+    template<typename OtherDerived>
+    EIGEN_STRONG_INLINE TensorDevice& operator-=(const OtherDerived& other) {
+      typedef typename OtherDerived::Scalar Scalar;
+      typedef TensorCwiseBinaryOp<internal::scalar_difference_op<Scalar>, const ExpressionType, const OtherDerived> Difference;
+      Difference difference(m_expression, other);
+      typedef TensorAssignOp<ExpressionType, const Difference> Assign;
+      Assign assign(m_expression, difference);
+      static const bool Vectorize = TensorEvaluator<const Assign, GpuDevice>::PacketAccess;
+      internal::TensorExecutor<const Assign, GpuDevice, Vectorize>::run(assign, m_device);
+      return *this;
+    }
+
   protected:
     const GpuDevice& m_device;
     ExpressionType m_expression;
diff --git a/unsupported/Eigen/CXX11/src/Tensor/TensorFunctors.h b/unsupported/Eigen/CXX11/src/Tensor/TensorFunctors.h
index 38586d067..25f085a59 100644
--- a/unsupported/Eigen/CXX11/src/Tensor/TensorFunctors.h
+++ b/unsupported/Eigen/CXX11/src/Tensor/TensorFunctors.h
@@ -77,7 +77,7 @@ template <typename T> struct MeanReducer
   }
   template <typename Packet>
   EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE T finalizeBoth(const T saccum, const Packet& vaccum) const {
-    return (saccum + predux(vaccum)) / (scalarCount_ + packetCount_ * packet_traits<Packet>::size);
+    return (saccum + predux(vaccum)) / (scalarCount_ + packetCount_ * unpacket_traits<Packet>::size);
   }
 
   protected:
diff --git a/unsupported/Eigen/CXX11/src/Tensor/TensorIO.h b/unsupported/Eigen/CXX11/src/Tensor/TensorIO.h
index a9d0f6c39..bdc6ddb87 100644
--- a/unsupported/Eigen/CXX11/src/Tensor/TensorIO.h
+++ b/unsupported/Eigen/CXX11/src/Tensor/TensorIO.h
@@ -30,14 +30,14 @@ std::ostream& operator << (std::ostream& os, const TensorBase<T, ReadOnlyAccesso
   typedef typename internal::remove_const<typename T::Scalar>::type Scalar;
   typedef typename T::Index Index;
   typedef typename TensorEvaluator<const TensorForcedEvalOp<const T>, DefaultDevice>::Dimensions Dimensions;
-  const Index total_size = internal::array_prod(tensor.dimensions());
+  const Index total_size = tensor.dimensions().TotalSize();
 
   // Print the tensor as a 1d vector or a 2d matrix.
   if (internal::array_size<Dimensions>::value == 1) {
     Map<const Array<Scalar, Dynamic, 1> > array(const_cast<Scalar*>(tensor.data()), total_size);
     os << array;
   } else {
-    const Index first_dim = tensor.dimensions()[0];
+    const Index first_dim = Eigen::internal::array_get<0>(tensor.dimensions());
     static const int layout = TensorEvaluator<const TensorForcedEvalOp<const T>, DefaultDevice>::Layout;
     Map<const Array<Scalar, Dynamic, Dynamic, layout> > matrix(const_cast<Scalar*>(tensor.data()), first_dim, total_size/first_dim);
     os << matrix;
diff --git a/unsupported/Eigen/CXX11/src/Tensor/TensorTraits.h b/unsupported/Eigen/CXX11/src/Tensor/TensorTraits.h
index 9a35b044d..0745b1742 100644
--- a/unsupported/Eigen/CXX11/src/Tensor/TensorTraits.h
+++ b/unsupported/Eigen/CXX11/src/Tensor/TensorTraits.h
@@ -65,7 +65,7 @@ struct traits<Tensor<Scalar_, NumIndices_, Options_> >
   static const int Layout = Options_ & RowMajor ? RowMajor : ColMajor;
   enum {
     Options = Options_,
-    Flags = compute_tensor_flags<Scalar_, Options_>::ret | LvalueBit,
+    Flags = compute_tensor_flags<Scalar_, Options_>::ret | (is_const<Scalar_>::value ? 0 : LvalueBit),
   };
 };
 
@@ -80,7 +80,7 @@ struct traits<TensorFixedSize<Scalar_, Dimensions, Options_> >
   static const int Layout = Options_ & RowMajor ? RowMajor : ColMajor;
   enum {
     Options = Options_,
-    Flags = compute_tensor_flags<Scalar_, Options_>::ret | LvalueBit,
+    Flags = compute_tensor_flags<Scalar_, Options_>::ret | (is_const<Scalar_>::value ? 0: LvalueBit),
   };
 };
 
@@ -97,7 +97,7 @@ struct traits<TensorMap<PlainObjectType, Options_> >
   static const int Layout = BaseTraits::Layout;
   enum {
     Options = Options_,
-    Flags = ((BaseTraits::Flags | LvalueBit) & ~AlignedBit) | (Options&Aligned ? AlignedBit : 0),
+    Flags = (BaseTraits::Flags & ~AlignedBit) | (Options&Aligned ? AlignedBit : 0),
   };
 };
 
@@ -113,7 +113,7 @@ struct traits<TensorRef<PlainObjectType> >
   static const int Layout = BaseTraits::Layout;
   enum {
     Options = BaseTraits::Options,
-    Flags = ((BaseTraits::Flags | LvalueBit) & ~AlignedBit) | (Options&Aligned ? AlignedBit : 0),
+    Flags = (BaseTraits::Flags & ~AlignedBit) | (Options&Aligned ? AlignedBit : 0),
   };
 };