Eigen cost model part 1. This implements a basic recursive framework to estimate the cost of evaluating tensor expressions.

1 files changed, 21 insertions, 10 deletions

diff --git a/unsupported/Eigen/CXX11/src/Tensor/TensorInflation.h b/unsupported/Eigen/CXX11/src/Tensor/TensorInflation.h
index 368e6f685..de2f67d74 100644
--- a/unsupported/Eigen/CXX11/src/Tensor/TensorInflation.h
+++ b/unsupported/Eigen/CXX11/src/Tensor/TensorInflation.h
@@ -81,6 +81,10 @@ struct TensorEvaluator<const TensorInflationOp<Strides, ArgType>, Device>
   typedef typename XprType::Index Index;
   static const int NumDims = internal::array_size<typename TensorEvaluator<ArgType, Device>::Dimensions>::value;
   typedef DSizes<Index, NumDims> Dimensions;
+  typedef typename XprType::Scalar Scalar;
+  typedef typename XprType::CoeffReturnType CoeffReturnType;
+  typedef typename PacketType<CoeffReturnType, Device>::type PacketReturnType;
+  static const int PacketSize = internal::unpacket_traits<PacketReturnType>::size;
 
   enum {
     IsAligned = /*TensorEvaluator<ArgType, Device>::IsAligned*/ false,
@@ -123,11 +127,6 @@ struct TensorEvaluator<const TensorInflationOp<Strides, ArgType>, Device>
     }
   }
 
-  typedef typename XprType::Scalar Scalar;
-  typedef typename XprType::CoeffReturnType CoeffReturnType;
-  typedef typename PacketType<CoeffReturnType, Device>::type PacketReturnType;
-
-
   EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE const Dimensions& dimensions() const { return m_dimensions; }
 
   EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE bool evalSubExprsIfNeeded(Scalar* /*data*/) {
@@ -190,18 +189,30 @@ struct TensorEvaluator<const TensorInflationOp<Strides, ArgType>, Device>
   template<int LoadMode>
   EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE PacketReturnType packet(Index index) const
   {
-    const int packetSize = internal::unpacket_traits<PacketReturnType>::size;
-    EIGEN_STATIC_ASSERT(packetSize > 1, YOU_MADE_A_PROGRAMMING_MISTAKE)
-    eigen_assert(index+packetSize-1 < dimensions().TotalSize());
+    EIGEN_STATIC_ASSERT(PacketSize > 1, YOU_MADE_A_PROGRAMMING_MISTAKE)
+    eigen_assert(index+PacketSize-1 < dimensions().TotalSize());
 
-    EIGEN_ALIGN_MAX typename internal::remove_const<CoeffReturnType>::type values[packetSize];
-    for (int i = 0; i < packetSize; ++i) {
+    EIGEN_ALIGN_MAX typename internal::remove_const<CoeffReturnType>::type values[PacketSize];
+    for (int i = 0; i < PacketSize; ++i) {
       values[i] = coeff(index+i);
     }
     PacketReturnType rslt = internal::pload<PacketReturnType>(values);
     return rslt;
   }
 
+  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE TensorOpCost costPerCoeff(bool vectorized) const {
+    const double compute_cost = NumDims * (3 * TensorOpCost::DivCost<Index>() +
+                                           3 * TensorOpCost::MulCost<Index>() +
+                                           2 * TensorOpCost::AddCost<Index>());
+    const double input_size = m_impl.dimensions().TotalSize();
+    const double output_size = m_dimensions.TotalSize();
+    if (output_size == 0)
+      return TensorOpCost();
+    return m_impl.costPerCoeff(vectorized) +
+           TensorOpCost(sizeof(CoeffReturnType) * input_size / output_size, 0,
+                        compute_cost, vectorized, PacketSize);
+  }
+
   EIGEN_DEVICE_FUNC Scalar* data() const { return NULL; }
 
  protected: