Improved the performance of tensor reductions

Added the ability to generate random numbers following a normal distribution
Created a test to validate the ability to generate random numbers.

unsupported/Eigen/CXX11/src/Tensor/TensorFunctors.h

@@ -16,50 +16,157 @@ namespace internal {
 // Standard reduction functors
 template <typename T> struct SumReducer
 {
-  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE SumReducer() : m_sum(0) { }
+  static const bool PacketAccess = true;
-  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE void reduce(const T t) {
+
-    m_sum += t;
+  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE void reduce(const T t, T* accum) const {
+    (*accum) += t;
   }
-  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE T finalize() const {
+  template <typename Packet>
-    return m_sum;
+  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE void reducePacket(const Packet& p, Packet* accum) const {
+    (*accum) = padd<Packet>(*accum, p);
   }
 
- private:
+  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE T initialize() const {
-  typename internal::remove_all<T>::type m_sum;
+    return static_cast<T>(0);
+  }
+  template <typename Packet>
+  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE Packet initializePacket() const {
+    return pset1<Packet>(0);
+  }
+  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE T finalize(const T accum) const {
+    return accum;
+  }
+  template <typename Packet>
+  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE T finalizePacket(const T saccum, const Packet& vaccum) const {
+    return saccum + predux(vaccum);
+  }
+};
+
+template <typename T> struct MeanReducer
+{
+  static const bool PacketAccess = true;
+  MeanReducer() : count_(0) { }
+
+  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE void reduce(const T t, T* accum) {
+    (*accum) += t;
+    count_++;
+  }
+  template <typename Packet>
+  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE void reducePacket(const Packet& p, Packet* accum) {
+    (*accum) = padd<Packet>(*accum, p);
+    count_ += packet_traits<Packet>::size;
+  }
+
+  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE T initialize() const {
+    return static_cast<T>(0);
+  }
+  template <typename Packet>
+  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE Packet initializePacket() const {
+    return pset1<Packet>(0);
+  }
+  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE T finalize(const T accum) const {
+    return accum / count_;
+  }
+  template <typename Packet>
+  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE T finalizePacket(const T saccum, const Packet& vaccum) const {
+    return (saccum + predux(vaccum)) / count_;
+  }
+
+  protected:
+    int count_;
 };
 
 template <typename T> struct MaxReducer
 {
-  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE MaxReducer() : m_max(-(std::numeric_limits<T>::max)()) { }
+  static const bool PacketAccess = true;
-  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE void reduce(const T t) {
+
-    if (t > m_max) { m_max = t; }
+  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE void reduce(const T t, T* accum) const {
+    if (t > *accum) { *accum = t; }
   }
-  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE T finalize() const {
+  template <typename Packet>
-    return m_max;
+  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE void reducePacket(const Packet& p, Packet* accum) const {
+    (*accum) = pmax<Packet>(*accum, p);
   }
 
- private:
+  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE T initialize() const {
-  typename internal::remove_all<T>::type m_max;
+    return -(std::numeric_limits<T>::max)();
+  }
+  template <typename Packet>
+  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE Packet initializePacket() const {
+    return pset1<Packet>(-(std::numeric_limits<T>::max)());
+  }
+  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE T finalize(const T accum) const {
+    return accum;
+  }
+  template <typename Packet>
+  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE T finalizePacket(const T saccum, const Packet& vaccum) const {
+    return (std::max)(saccum, predux_max(vaccum));
+  }
 };
 
 template <typename T> struct MinReducer
 {
-  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE MinReducer() : m_min((std::numeric_limits<T>::max)()) { }
+  static const bool PacketAccess = true;
-  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE void reduce(const T t) {
+
-    if (t < m_min) { m_min = t; }
+  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE void reduce(const T t, T* accum) const {
+    if (t < *accum) { *accum = t; }
   }
-  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE T finalize() const {
+  template <typename Packet>
-    return m_min;
+  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE void reducePacket(const Packet& p, Packet* accum) const {
+    (*accum) = pmin<Packet>(*accum, p);
   }
 
- private:
+  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE T initialize() const {
-  typename internal::remove_all<T>::type m_min;
+    return (std::numeric_limits<T>::max)();
+  }
+  template <typename Packet>
+  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE Packet initializePacket() const {
+    return pset1<Packet>((std::numeric_limits<T>::max)());
+  }
+  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE T finalize(const T accum) const {
+    return accum;
+  }
+  template <typename Packet>
+  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE T finalizePacket(const T saccum, const Packet& vaccum) const {
+    return (std::min)(saccum, predux_min(vaccum));
+  }
 };
 
 
+template <typename T> struct ProdReducer
+{
+  static const bool PacketAccess = true;
+
+  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE void reduce(const T t, T* accum) const {
+    (*accum) *= t;
+  }
+  template <typename Packet>
+  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE void reducePacket(const Packet& p, Packet* accum) const {
+    (*accum) = pmul<Packet>(*accum, p);
+  }
+
+  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE T initialize() const {
+    return static_cast<T>(1);
+  }
+  template <typename Packet>
+  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE Packet initializePacket() const {
+    return pset1<Packet>(1);
+  }
+  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE T finalize(const T accum) const {
+    return accum;
+  }
+  template <typename Packet>
+  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE T finalizePacket(const T saccum, const Packet& vaccum) const {
+    return saccum * predux_mul(vaccum);
+  }
+};
+
 #if !defined (EIGEN_USE_GPU) || !defined(__CUDACC__) || !defined(__CUDA_ARCH__)
 // We're not compiling a cuda kernel
 template <typename T> struct UniformRandomGenerator {
+
+  static const bool PacketAccess = true;
+
   template<typename Index>
   T operator()(Index, Index = 0) const {
     return random<T>();
@@ -81,16 +188,19 @@ template <typename T> struct UniformRandomGenerator {
 template <typename T> struct UniformRandomGenerator;
 
 template <> struct UniformRandomGenerator<float> {
-  UniformRandomGenerator() {
+
+  static const bool PacketAccess = true;
+
+  EIGEN_DEVICE_FUNC UniformRandomGenerator() {
     const int tid = blockIdx.x * blockDim.x + threadIdx.x;
     curand_init(0, tid, 0, &m_state);
   }
 
-  template<typename Index>
+  template<typename Index> EIGEN_DEVICE_FUNC
   float operator()(Index, Index = 0) const {
     return curand_uniform(&m_state);
   }
-  template<typename Index>
+  template<typename Index> EIGEN_DEVICE_FUNC
   float4 packetOp(Index, Index = 0) const {
     return curand_uniform4(&m_state);
   }
@@ -100,15 +210,18 @@ template <> struct UniformRandomGenerator<float> {
 };
 
 template <> struct UniformRandomGenerator<double> {
-  UniformRandomGenerator() {
+
+  static const bool PacketAccess = true;
+
+  EIGEN_DEVICE_FUNC UniformRandomGenerator() {
     const int tid = blockIdx.x * blockDim.x + threadIdx.x;
     curand_init(0, tid, 0, &m_state);
   }
-  template<typename Index>
+  template<typename Index> EIGEN_DEVICE_FUNC
   double operator()(Index, Index = 0) const {
     return curand_uniform_double(&m_state);
   }
-  template<typename Index>
+  template<typename Index> EIGEN_DEVICE_FUNC
   double2 packetOp(Index, Index = 0) const {
     return curand_uniform2_double(&m_state);
   }
@@ -120,6 +233,84 @@ template <> struct UniformRandomGenerator<double> {
 #endif
 
 
+#if (!defined (EIGEN_USE_GPU) || !defined(__CUDACC__) || !defined(__CUDA_ARCH__)) && __cplusplus > 199711
+// We're not compiling a cuda kernel
+template <typename T> struct NormalRandomGenerator {
+
+  static const bool PacketAccess = true;
+
+  NormalRandomGenerator() : m_distribution(0, 1) {}
+  NormalRandomGenerator(const NormalRandomGenerator& other) : m_distribution(other.m_distribution) { }
+
+  template<typename Index>
+  T operator()(Index, Index = 0) const {
+    return m_distribution(m_generator);
+  }
+  template<typename Index>
+  typename internal::packet_traits<T>::type packetOp(Index, Index = 0) const {
+    const int packetSize = internal::packet_traits<T>::size;
+    EIGEN_ALIGN_DEFAULT T values[packetSize];
+    for (int i = 0; i < packetSize; ++i) {
+      values[i] = m_distribution(m_generator);
+    }
+    return internal::pload<typename internal::packet_traits<T>::type>(values);
+  }
+
+  mutable std::normal_distribution<T> m_distribution;
+  mutable std::default_random_engine m_generator;
+};
+
+#elif defined (EIGEN_USE_GPU) && defined(__CUDACC__) && defined(__CUDA_ARCH__)
+
+// We're compiling a cuda kernel
+template <typename T> struct NormalRandomGenerator;
+
+template <> struct NormalRandomGenerator<float> {
+
+  static const bool PacketAccess = true;
+
+  EIGEN_DEVICE_FUNC NormalRandomGenerator() {
+    const int tid = blockIdx.x * blockDim.x + threadIdx.x;
+    curand_init(0, tid, 0, &m_state);
+  }
+
+  template<typename Index> EIGEN_DEVICE_FUNC
+  float operator()(Index, Index = 0) const {
+    return curand_normal(&m_state);
+  }
+  template<typename Index> EIGEN_DEVICE_FUNC
+  float4 packetOp(Index, Index = 0) const {
+    return curand_normal4(&m_state);
+  }
+
+ private:
+  mutable curandStatePhilox4_32_10_t m_state;
+};
+
+template <> struct NormalRandomGenerator<double> {
+
+  static const bool PacketAccess = true;
+
+  EIGEN_DEVICE_FUNC NormalRandomGenerator() {
+    const int tid = blockIdx.x * blockDim.x + threadIdx.x;
+    curand_init(0, tid, 0, &m_state);
+  }
+  template<typename Index> EIGEN_DEVICE_FUNC
+  double operator()(Index, Index = 0) const {
+    return curand_normal_double(&m_state);
+  }
+  template<typename Index> EIGEN_DEVICE_FUNC
+  double2 packetOp(Index, Index = 0) const {
+    return curand_normal2_double(&m_state);
+  }
+
+ private:
+  mutable curandStatePhilox4_32_10_t m_state;
+};
+
+#endif
+
+
 } // end namespace internal
 } // end namespace Eigen
 

unsupported/Eigen/CXX11/src/Tensor/TensorReduction.h

@@ -43,6 +43,75 @@ struct nested<TensorReductionOp<Op, Dims, XprType>, 1, typename eval<TensorReduc
   typedef TensorReductionOp<Op, Dims, XprType> type;
 };
 
+
+template <typename ReducedDims, int NumTensorDims, int Layout>
+struct are_inner_most_dims {
+  static const bool value = false;
+};
+#if __cplusplus > 199711L
+template <typename ReducedDims, int NumTensorDims>
+struct are_inner_most_dims<ReducedDims, NumTensorDims, ColMajor>{
+  static const bool value = indices_statically_known_to_increase<ReducedDims>()() &&
+                            index_statically_eq<ReducedDims>()(0, 0) &&
+                            index_statically_eq<ReducedDims>()(array_size<ReducedDims>::value-1, array_size<ReducedDims>::value-1);
+};
+template <typename ReducedDims, int NumTensorDims>
+struct are_inner_most_dims<ReducedDims, NumTensorDims, RowMajor>{
+  static const bool value = indices_statically_known_to_increase<ReducedDims>()() &&
+                            index_statically_eq<ReducedDims>()(0, NumTensorDims - array_size<ReducedDims>::value) &&
+                            index_statically_eq<ReducedDims>()(array_size<ReducedDims>::value - 1, NumTensorDims - 1);
+};
+#endif
+
+
+template <int DimIndex, typename Self, typename Op>
+struct GenericDimReducer {
+  static EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE void reduce(const Self& self, typename Self::Index firstIndex, Op& reducer, typename Self::CoeffReturnType* accum) {
+    EIGEN_STATIC_ASSERT(DimIndex > 0, YOU_MADE_A_PROGRAMMING_MISTAKE);
+    for (int j = 0; j < self.m_reducedDims[DimIndex]; ++j) {
+      const typename Self::Index input = firstIndex + j * self.m_reducedStrides[DimIndex];
+      GenericDimReducer<DimIndex-1, Self, Op>::reduce(self, input, reducer, accum);
+    }
+  }
+};
+template <typename Self, typename Op>
+struct GenericDimReducer<0, Self, Op> {
+  static EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE void reduce(const Self& self, typename Self::Index firstIndex, Op& reducer, typename Self::CoeffReturnType* accum) {
+    for (int j = 0; j < self.m_reducedDims[0]; ++j) {
+      const typename Self::Index input = firstIndex + j * self.m_reducedStrides[0];
+      reducer.reduce(self.m_impl.coeff(input), accum);
+    }
+  }
+};
+
+template <typename Self, typename Op, bool Vectorizable = (Self::InputPacketAccess & Op::PacketAccess)>
+struct InnerMostDimReducer {
+  static EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE typename Self::CoeffReturnType reduce(const Self& self, typename Self::Index firstIndex, typename Self::Index numValuesToReduce, Op& reducer) {
+    typename Self::CoeffReturnType accum = reducer.initialize();
+    for (typename Self::Index j = 0; j < numValuesToReduce; ++j) {
+      reducer.reduce(self.m_impl.coeff(firstIndex + j), &accum);
+    }
+    return reducer.finalize(accum);
+  }
+};
+
+template <typename Self, typename Op>
+struct InnerMostDimReducer<Self, Op, true> {
+  static EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE typename Self::CoeffReturnType reduce(const Self& self, typename Self::Index firstIndex, typename Self::Index numValuesToReduce, Op& reducer) {
+    const int packetSize = internal::unpacket_traits<typename Self::PacketReturnType>::size;
+    const typename Self::Index VectorizedSize = (numValuesToReduce / packetSize) * packetSize;
+    typename Self::PacketReturnType p = reducer.template initializePacket<typename Self::PacketReturnType>();
+    for (typename Self::Index j = 0; j < VectorizedSize; j += packetSize) {
+      reducer.reducePacket(self.m_impl.template packet<Unaligned>(firstIndex + j), &p);
+    }
+    typename Self::CoeffReturnType accum = reducer.initialize();
+    for (typename Self::Index j = VectorizedSize; j < numValuesToReduce; ++j) {
+      reducer.reduce(self.m_impl.coeff(firstIndex + j), &accum);
+    }
+    return reducer.finalizePacket(accum, p);
+  }
+};
+
 }  // end namespace internal
 
 
@@ -52,8 +121,8 @@ class TensorReductionOp : public TensorBase<TensorReductionOp<Op, Dims, XprType>
     typedef typename Eigen::internal::traits<TensorReductionOp>::Scalar Scalar;
     typedef typename Eigen::internal::traits<TensorReductionOp>::Packet Packet;
     typedef typename Eigen::NumTraits<Scalar>::Real RealScalar;
-    typedef typename XprType::CoeffReturnType CoeffReturnType;
+    typedef typename internal::remove_const<typename XprType::CoeffReturnType>::type CoeffReturnType;
-    typedef typename XprType::PacketReturnType PacketReturnType;
+    typedef typename internal::remove_const<typename XprType::PacketReturnType>::type PacketReturnType;
     typedef typename Eigen::internal::nested<TensorReductionOp>::type Nested;
     typedef typename Eigen::internal::traits<TensorReductionOp>::StorageKind StorageKind;
     typedef typename Eigen::internal::traits<TensorReductionOp>::Index Index;
@@ -85,20 +154,27 @@ struct TensorEvaluator<const TensorReductionOp<Op, Dims, ArgType>, Device>
   typedef typename XprType::Index Index;
   static const int NumInputDims = internal::array_size<typename TensorEvaluator<ArgType, Device>::Dimensions>::value;
   static const int NumReducedDims = internal::array_size<Dims>::value;
-  static const int NumDims = (NumInputDims==NumReducedDims) ? 1 : NumInputDims - NumReducedDims;
+  static const int NumOutputDims = (NumInputDims==NumReducedDims) ? 1 : NumInputDims - NumReducedDims;
-  typedef DSizes<Index, NumDims> Dimensions;
+  typedef DSizes<Index, NumOutputDims> Dimensions;
   typedef typename XprType::Scalar Scalar;
+  typedef TensorEvaluator<const TensorReductionOp<Op, Dims, ArgType>, Device> Self;
+  static const bool InputPacketAccess = TensorEvaluator<ArgType, Device>::PacketAccess;
 
   enum {
     IsAligned = false,
-    PacketAccess = false,  // The code isn't vectorized properly yet
+    PacketAccess = Self::InputPacketAccess && Op::PacketAccess,
+    Layout = TensorEvaluator<ArgType, Device>::Layout,
+    CoordAccess = false,  // to be implemented
   };
 
+  static const bool ReducingInnerMostDims = internal::are_inner_most_dims<Dims, NumInputDims, Layout>::value;
+
   EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE TensorEvaluator(const XprType& op, const Device& device)
       : m_impl(op.expression(), device), m_reducer(op.reducer())
   {
     EIGEN_STATIC_ASSERT(NumInputDims >= NumReducedDims, YOU_MADE_A_PROGRAMMING_MISTAKE);
 
+    // Bitmap indicating if an input dimension is reduced or not.
     array<bool, NumInputDims> reduced;
     for (int i = 0; i < NumInputDims; ++i) {
       reduced[i] = false;
@@ -122,24 +198,41 @@ struct TensorEvaluator<const TensorReductionOp<Op, Dims, ArgType>, Device>
       }
     }
 
-    m_outputStrides[0] = 1;
+    // Precompute output strides.
-    for (int i = 1; i < NumDims; ++i) {
+    if (Layout == ColMajor) {
-      m_outputStrides[i] = m_outputStrides[i-1] * m_dimensions[i-1];
+      m_outputStrides[0] = 1;
+      for (int i = 1; i < NumOutputDims; ++i) {
+        m_outputStrides[i] = m_outputStrides[i - 1] * m_dimensions[i - 1];
+      }
+    } else {
+      m_outputStrides[NumOutputDims - 1] = 1;
+      for (int i = NumOutputDims - 2; i >= 0; --i) {
+        m_outputStrides[i] = m_outputStrides[i + 1] * m_dimensions[i + 1];
+      }
     }
 
-    array<Index, NumInputDims> strides;
+    // Precompute input strides.
-    strides[0] = 1;
+    array<Index, NumInputDims> input_strides;
-    for (int i = 1; i < NumInputDims; ++i) {
+    if (Layout == ColMajor) {
-      strides[i] = strides[i-1] * input_dims[i-1];
+      input_strides[0] = 1;
+      for (int i = 1; i < NumInputDims; ++i) {
+        input_strides[i] = input_strides[i-1] * input_dims[i-1];
+      }
+    } else {
+      input_strides[NumInputDims - 1] = 1;
+      for (int i = NumInputDims - 2; i >= 0; --i) {
+        input_strides[i] = input_strides[i + 1] * input_dims[i + 1];
+      }
     }
+
     outputIndex = 0;
     reduceIndex = 0;
     for (int i = 0; i < NumInputDims; ++i) {
       if (reduced[i]) {
-        m_reducedStrides[reduceIndex] = strides[i];
+        m_reducedStrides[reduceIndex] = input_strides[i];
         ++reduceIndex;
       } else {
-        m_preservedStrides[outputIndex] = strides[i];
+        m_preservedStrides[outputIndex] = input_strides[i];
         ++outputIndex;
       }
     }
@@ -147,6 +240,7 @@ struct TensorEvaluator<const TensorReductionOp<Op, Dims, ArgType>, Device>
     // Special case for full reductions
     if (NumInputDims == NumReducedDims) {
       m_dimensions[0] = 1;
+      m_preservedStrides[0] = internal::array_prod(input_dims);
     }
   }
 
@@ -161,14 +255,22 @@ struct TensorEvaluator<const TensorReductionOp<Op, Dims, ArgType>, Device>
     m_impl.cleanup();
   }
 
-  typedef typename XprType::CoeffReturnType CoeffReturnType;
+  typedef typename internal::remove_const<typename XprType::CoeffReturnType>::type CoeffReturnType;
-  typedef typename XprType::PacketReturnType PacketReturnType;
+  typedef typename internal::remove_const<typename XprType::PacketReturnType>::type PacketReturnType;
 
   EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE CoeffReturnType coeff(Index index) const
   {
     Op reducer(m_reducer);
-    reduce(firstInput(index), 0, reducer);
+    if (ReducingInnerMostDims) {
-    return reducer.finalize();
+      const Index num_values_to_reduce =
+          (Layout == ColMajor) ? m_preservedStrides[0] : m_preservedStrides[NumOutputDims - 1];
+      return internal::InnerMostDimReducer<Self, Op>::reduce(*this, firstInput(index),
+                                                             num_values_to_reduce, reducer);
+    } else {
+      typename Self::CoeffReturnType accum = reducer.initialize();
+      internal::GenericDimReducer<NumReducedDims-1, Self, Op>::reduce(*this, firstInput(index), reducer, &accum);
+      return reducer.finalize(accum);
+    }
   }
 
   // TODO(bsteiner): provide a more efficient implementation.
@@ -179,9 +281,20 @@ struct TensorEvaluator<const TensorReductionOp<Op, Dims, ArgType>, Device>
     EIGEN_STATIC_ASSERT(packetSize > 1, YOU_MADE_A_PROGRAMMING_MISTAKE)
     eigen_assert(index + packetSize - 1 < dimensions().TotalSize());
 
-    EIGEN_ALIGN_DEFAULT CoeffReturnType values[packetSize];
+    EIGEN_ALIGN_DEFAULT typename internal::remove_const<CoeffReturnType>::type values[packetSize];
-    for (int i = 0; i < packetSize; ++i) {
+    if (ReducingInnerMostDims) {
-      values[i] = coeff(index+i);
+      const Index num_values_to_reduce =
+          (Layout == ColMajor) ? m_preservedStrides[0] : m_preservedStrides[NumOutputDims - 1];
+      const Index firstIndex = firstInput(index);
+      for (Index i = 0; i < packetSize; ++i) {
+        Op reducer(m_reducer);
+        values[i] = internal::InnerMostDimReducer<Self, Op>::reduce(*this, firstIndex + i * num_values_to_reduce,
+                                                                    num_values_to_reduce, reducer);
+      }
+    } else {
+      for (int i = 0; i < packetSize; ++i) {
+        values[i] = coeff(index + i);
+      }
     }
     PacketReturnType rslt = internal::pload<PacketReturnType>(values);
     return rslt;
@@ -190,34 +303,59 @@ struct TensorEvaluator<const TensorReductionOp<Op, Dims, ArgType>, Device>
   Scalar* data() const { return NULL; }
 
   private:
+  template <int, typename, typename> friend struct internal::GenericDimReducer;
+  template <typename, typename, bool> friend struct internal::InnerMostDimReducer;
+
+  // Returns the Index in the input tensor of the first value that needs to be
+  // used to compute the reduction at output index "index".
   EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE Index firstInput(Index index) const {
-    Index startInput = 0;
+    if (ReducingInnerMostDims) {
-    for (int i = NumDims - 1; i > 0; --i) {
+      if (Layout == ColMajor) {
-      const Index idx = index / m_outputStrides[i];
+        return index * m_preservedStrides[0];
-      startInput += idx * m_preservedStrides[i];
+      } else {
-      index -= idx * m_outputStrides[i];
+        return index * m_preservedStrides[NumOutputDims - 1];
+      }
+    }
+    Index startInput = 0;
+    if (Layout == ColMajor) {
+      for (int i = NumOutputDims - 1; i > 0; --i) {
+        // This is index_i in the output tensor.
+        const Index idx = index / m_outputStrides[i];
+        startInput += idx * m_preservedStrides[i];
+        index -= idx * m_outputStrides[i];
+      }
+      startInput += index * m_preservedStrides[0];
+    } else {
+      for (int i = 0; i < NumOutputDims - 1; ++i) {
+        // This is index_i in the output tensor.
+        const Index idx = index / m_outputStrides[i];
+        startInput += idx * m_preservedStrides[i];
+        index -= idx * m_outputStrides[i];
+      }
+      startInput += index * m_preservedStrides[NumOutputDims - 1];
     }
-    startInput += index * m_preservedStrides[0];
     return startInput;
   }
 
-  EIGEN_DEVICE_FUNC void reduce(Index firstIndex, int DimIndex, Op& reducer) const {
+  // Dimensions of the output of the operation.
-    for (int j = 0; j < m_reducedDims[DimIndex]; ++j) {
-      const Index input = firstIndex + j * m_reducedStrides[DimIndex];
-      if (DimIndex < NumReducedDims-1) {
-        reduce(input, DimIndex+1, reducer);
-      } else {
-        reducer.reduce(m_impl.coeff(input));
-      }
-    }
-  }
-
   Dimensions m_dimensions;
-  array<Index, NumDims> m_outputStrides;
+  // Precomputed strides for the output tensor.
-  array<Index, NumDims> m_preservedStrides;
+  array<Index, NumOutputDims> m_outputStrides;
+  // Subset of strides of the input tensor for the non-reduced dimensions.
+  // Indexed by output dimensions.
+  array<Index, NumOutputDims> m_preservedStrides;
+
+  // Subset of strides of the input tensor for the reduced dimensions.
+  // Indexed by reduced dimensions.
   array<Index, NumReducedDims> m_reducedStrides;
+  // Size of the input dimensions that are reduced.
+  // Indexed by reduced dimensions.
   array<Index, NumReducedDims> m_reducedDims;
+
+  // Evaluator for the input expression.
   TensorEvaluator<ArgType, Device> m_impl;
+
+  // Operation to apply for computing the reduction.
   Op m_reducer;
 };
 

unsupported/test/cxx11_tensor_random.cpp

@@ -0,0 +1,78 @@
+// This file is part of Eigen, a lightweight C++ template library
+// for linear algebra.
+//
+// Copyright (C) 2014 Benoit Steiner <benoit.steiner.goog@gmail.com>
+//
+// This Source Code Form is subject to the terms of the Mozilla
+// Public License v. 2.0. If a copy of the MPL was not distributed
+// with this file, You can obtain one at http://mozilla.org/MPL/2.0/.
+
+#include "main.h"
+
+#include <Eigen/CXX11/Tensor>
+
+static void test_default()
+{
+  Tensor<float, 1> vec(6);
+  vec.setRandom();
+
+  // Fixme: we should check that the generated numbers follow a uniform
+  // distribution instead.
+  for (int i = 1; i < 6; ++i) {
+    VERIFY_IS_NOT_EQUAL(vec(i), vec(i-1));
+  }
+}
+
+static void test_normal()
+{
+  Tensor<float, 1> vec(6);
+  vec.setRandom<Eigen::internal::NormalRandomGenerator<float>>();
+
+  // Fixme: we should check that the generated numbers follow a gaussian
+  // distribution instead.
+  for (int i = 1; i < 6; ++i) {
+    VERIFY_IS_NOT_EQUAL(vec(i), vec(i-1));
+  }
+}
+
+
+struct MyGenerator {
+  MyGenerator() { }
+  MyGenerator(const MyGenerator&) { }
+
+  // Return a random value to be used.  "element_location" is the
+  // location of the entry to set in the tensor, it can typically
+  // be ignored.
+  int operator()(Eigen::DenseIndex element_location, Eigen::DenseIndex /*unused*/ = 0) const {
+    return 3 * element_location;
+  }
+
+  // Same as above but generates several numbers at a time.
+  typename internal::packet_traits<int>::type packetOp(
+      Eigen::DenseIndex packet_location, Eigen::DenseIndex /*unused*/ = 0) const {
+    const int packetSize = internal::packet_traits<int>::size;
+    EIGEN_ALIGN_DEFAULT int values[packetSize];
+    for (int i = 0; i < packetSize; ++i) {
+      values[i] = 3 * (packet_location + i);
+    }
+    return internal::pload<typename internal::packet_traits<int>::type>(values);
+  }
+};
+
+
+static void test_custom()
+{
+  Tensor<int, 1> vec(6);
+  vec.setRandom<MyGenerator>();
+
+  for (int i = 0; i < 6; ++i) {
+    VERIFY_IS_EQUAL(vec(i), 3*i);
+  }
+}
+
+void test_cxx11_tensor_random()
+{
+  CALL_SUBTEST(test_default());
+  CALL_SUBTEST(test_normal());
+  CALL_SUBTEST(test_custom());
+}