From 235e83aba608cf3d94b033bfbf551f8c136a3fab Mon Sep 17 00:00:00 2001
From: Rasmus Munk Larsen <rmlarsen@google.com>
Date: Thu, 14 Apr 2016 13:57:35 -0700
Subject: Eigen cost model part 1. This implements a basic recursive framework
 to estimate the cost of evaluating tensor expressions.

---
 unsupported/Eigen/CXX11/src/Tensor/TensorForcedEval.h | 12 ++++++++----
 1 file changed, 8 insertions(+), 4 deletions(-)

(limited to 'unsupported/Eigen/CXX11/src/Tensor/TensorForcedEval.h')

diff --git a/unsupported/Eigen/CXX11/src/Tensor/TensorForcedEval.h b/unsupported/Eigen/CXX11/src/Tensor/TensorForcedEval.h
index 14f480901..d2b0b30a1 100644
--- a/unsupported/Eigen/CXX11/src/Tensor/TensorForcedEval.h
+++ b/unsupported/Eigen/CXX11/src/Tensor/TensorForcedEval.h
@@ -83,6 +83,10 @@ struct TensorEvaluator<const TensorForcedEvalOp<ArgType>, Device>
   typedef TensorForcedEvalOp<ArgType> XprType;
   typedef typename ArgType::Scalar Scalar;
   typedef typename TensorEvaluator<ArgType, Device>::Dimensions Dimensions;
+  typedef typename XprType::Index Index;
+  typedef typename XprType::CoeffReturnType CoeffReturnType;
+  typedef typename PacketType<CoeffReturnType, Device>::type PacketReturnType;
+  static const int PacketSize = internal::unpacket_traits<PacketReturnType>::size;
 
   enum {
     IsAligned = true,
@@ -95,10 +99,6 @@ struct TensorEvaluator<const TensorForcedEvalOp<ArgType>, Device>
       : m_impl(op.expression(), device), m_op(op.expression()), m_device(device), m_buffer(NULL)
   { }
 
-  typedef typename XprType::Index Index;
-  typedef typename XprType::CoeffReturnType CoeffReturnType;
-  typedef typename PacketType<CoeffReturnType, Device>::type PacketReturnType;
-
   EIGEN_DEVICE_FUNC const Dimensions& dimensions() const { return m_impl.dimensions(); }
 
   EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE bool evalSubExprsIfNeeded(CoeffReturnType*) {
@@ -132,6 +132,10 @@ struct TensorEvaluator<const TensorForcedEvalOp<ArgType>, Device>
     return internal::ploadt<PacketReturnType, LoadMode>(m_buffer + index);
   }
 
+  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE TensorOpCost costPerCoeff(bool vectorized) const {
+    return TensorOpCost(sizeof(CoeffReturnType), 0, 0, vectorized, PacketSize);
+  }
+
   EIGEN_DEVICE_FUNC Scalar* data() const { return m_buffer; }
 
  private:
-- 
cgit v1.2.3