Prevent nvcc from miscompiling the cuda metakernel. Unfortunately this reintroduces some compulation warnings but it's much better than having to deal with random assertion failures.

1 files changed, 2 insertions, 2 deletions

diff --git a/unsupported/Eigen/CXX11/src/Tensor/TensorReductionCuda.h b/unsupported/Eigen/CXX11/src/Tensor/TensorReductionCuda.h
index b046607b9..558d0c83d 100644
--- a/unsupported/Eigen/CXX11/src/Tensor/TensorReductionCuda.h
+++ b/unsupported/Eigen/CXX11/src/Tensor/TensorReductionCuda.h
@@ -115,11 +115,11 @@ struct FullReducer<Self, Op, GpuDevice, Vectorizable> {
                                                  internal::is_same<typename Self::CoeffReturnType, float>::value;
 
   template <typename OutputType>
-  EIGEN_DEVICE_FUNC static void run(const Self& self, Op& reducer, const GpuDevice& device, OutputType* output) {
+  static void run(const Self& self, Op& reducer, const GpuDevice& device, OutputType* output) {
     assert(false && "Should only be called on floats");
   }
 
-  EIGEN_DEVICE_FUNC static void run(const Self& self, Op& reducer, const GpuDevice& device, float* output) {
+  static void run(const Self& self, Op& reducer, const GpuDevice& device, float* output) {
     typedef typename Self::Index Index;
 
     const Index num_coeffs = array_prod(self.m_impl.dimensions());