Reworked the expression evaluation mechanism in order to make it possible to efficiently compute convolutions and contractions in the future:

 * The scheduling of computation is moved out the the assignment code and into a new TensorExecutor class
 * The assignment itself is now a regular node on the expression tree
 * The expression evaluators start by recursively evaluating all their subexpressions if needed

1 files changed, 9 insertions, 0 deletions

diff --git a/unsupported/Eigen/CXX11/src/Tensor/TensorConvolution.h b/unsupported/Eigen/CXX11/src/Tensor/TensorConvolution.h
index c4cfe0cd8..58b1808a3 100644
--- a/unsupported/Eigen/CXX11/src/Tensor/TensorConvolution.h
+++ b/unsupported/Eigen/CXX11/src/Tensor/TensorConvolution.h
@@ -153,6 +153,15 @@ struct TensorEvaluator<const TensorConvolutionOp<Indices, InputArgType, KernelAr
 
   const Dimensions& dimensions() const { return m_dimensions; }
 
+  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE void evalSubExprsIfNeeded() {
+    m_inputImpl.evalSubExprsIfNeeded();
+    m_kernelImpl.evalSubExprsIfNeeded();
+  }
+  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE void cleanup() {
+    m_inputImpl.cleanup();
+    m_kernelImpl.cleanup();
+  }
+
   void evalTo(typename XprType::Scalar* buffer) const {
     for (int i = 0; i < dimensions().TotalSize(); ++i) {
       buffer[i] += coeff(i);