Fix TensorReverse on GPU with m_stride[i]==0

1 files changed, 5 insertions, 5 deletions

diff --git a/unsupported/Eigen/CXX11/src/Tensor/TensorReverse.h b/unsupported/Eigen/CXX11/src/Tensor/TensorReverse.h
index 6faff87bf..42205db31 100644
--- a/unsupported/Eigen/CXX11/src/Tensor/TensorReverse.h
+++ b/unsupported/Eigen/CXX11/src/Tensor/TensorReverse.h
@@ -122,6 +122,8 @@ struct TensorEvaluator<const TensorReverseOp<ReverseDimensions, ArgType>, Device
     RawAccess         = false
   };
 
+  typedef internal::TensorIntDivisor<Index> IndexDivisor;
+
   typedef typename internal::remove_const<Scalar>::type ScalarNoConst;
   typedef internal::TensorBlock<ScalarNoConst, Index, NumDims, Layout>
       OutputTensorBlock;
@@ -141,17 +143,15 @@ struct TensorEvaluator<const TensorReverseOp<ReverseDimensions, ArgType>, Device
       m_strides[0] = 1;
       for (int i = 1; i < NumDims; ++i) {
         m_strides[i] = m_strides[i-1] * m_dimensions[i-1];
+        if (m_strides[i] > 0) m_fastStrides[i] = IndexDivisor(m_strides[i]);
       }
     } else {
       m_strides[NumDims-1] = 1;
       for (int i = NumDims - 2; i >= 0; --i) {
         m_strides[i] = m_strides[i+1] * m_dimensions[i+1];
+        if (m_strides[i] > 0) m_fastStrides[i] = IndexDivisor(m_strides[i]);
       }
     }
-    // Remember the strides for fast division.
-    for (int i = 0; i < NumDims; ++i) {
-      m_fastStrides[i] = internal::TensorIntDivisor<Index>(m_strides[i]);
-    }
   }
 
   EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
@@ -377,7 +377,7 @@ struct TensorEvaluator<const TensorReverseOp<ReverseDimensions, ArgType>, Device
  protected:
   Dimensions m_dimensions;
   array<Index, NumDims> m_strides;
-  array<internal::TensorIntDivisor<Index>, NumDims> m_fastStrides;
+  array<IndexDivisor, NumDims> m_fastStrides;
   TensorEvaluator<ArgType, Device> m_impl;
   ReverseDimensions m_reverse;
   const Device& m_device;