From 1e911b276c6a5ed85b036abf07d848994d6dc144 Mon Sep 17 00:00:00 2001
From: Benoit Steiner <benoit.steiner.goog@gmail.com>
Date: Wed, 1 Jul 2015 13:59:11 -0700
Subject: Misc improvements and optimizations

---
 unsupported/Eigen/CXX11/src/Tensor/TensorConvolution.h | 6 ++++--
 1 file changed, 4 insertions(+), 2 deletions(-)

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

diff --git a/unsupported/Eigen/CXX11/src/Tensor/TensorConvolution.h b/unsupported/Eigen/CXX11/src/Tensor/TensorConvolution.h
index 07cba649b..88db9d410 100644
--- a/unsupported/Eigen/CXX11/src/Tensor/TensorConvolution.h
+++ b/unsupported/Eigen/CXX11/src/Tensor/TensorConvolution.h
@@ -510,7 +510,8 @@ struct TensorEvaluator<const TensorConvolutionOp<Indices, InputArgType, KernelAr
       Scalar* local = (Scalar*)m_device.allocate(kernel_sz);
       typedef TensorEvalToOp<const KernelArgType> EvalTo;
       EvalTo evalToTmp(local, m_kernelArg);
-      internal::TensorExecutor<const EvalTo, Device>::run(evalToTmp, m_device);
+      const bool PacketAccess = internal::IsVectorizable<Device, KernelArgType>::value;
+      internal::TensorExecutor<const EvalTo, Device, PacketAccess>::run(evalToTmp, m_device);
 
       m_kernel = local;
       m_local_kernel = true;
@@ -815,7 +816,8 @@ struct TensorEvaluator<const TensorConvolutionOp<Indices, InputArgType, KernelAr
       Scalar* local = (Scalar*)m_device.allocate(kernel_sz);
       typedef TensorEvalToOp<const KernelArgType> EvalTo;
       EvalTo evalToTmp(local, m_kernelArg);
-      internal::TensorExecutor<const EvalTo, GpuDevice>::run(evalToTmp, m_device);
+      const bool PacketAccess = internal::IsVectorizable<GpuDevice, KernelArgType>::value;
+      internal::TensorExecutor<const EvalTo, GpuDevice, PacketAccess>::run(evalToTmp, m_device);
 
       m_kernel = local;
       m_local_kernel = true;
-- 
cgit v1.2.3