diff options
author | Benoit Steiner <benoit.steiner.goog@gmail.com> | 2016-05-24 23:11:56 -0700 |
---|---|---|
committer | Benoit Steiner <benoit.steiner.goog@gmail.com> | 2016-05-24 23:11:56 -0700 |
commit | 083566732937f5f57a4bfcaa68c46da795f465ab (patch) | |
tree | cb573dbb84ee3b9698096342729f677ca625b558 /unsupported/Eigen/CXX11/src | |
parent | b5d6b52a4d3afdcfbb940e3bf728e8bc90dfb3b2 (diff) |
There is no need to make the fp16 full reduction kernel a static function.
Diffstat (limited to 'unsupported/Eigen/CXX11/src')
-rw-r--r-- | unsupported/Eigen/CXX11/src/Tensor/TensorReductionCuda.h | 2 |
1 files changed, 1 insertions, 1 deletions
diff --git a/unsupported/Eigen/CXX11/src/Tensor/TensorReductionCuda.h b/unsupported/Eigen/CXX11/src/Tensor/TensorReductionCuda.h index 63646dfc2..c0a36cd9c 100644 --- a/unsupported/Eigen/CXX11/src/Tensor/TensorReductionCuda.h +++ b/unsupported/Eigen/CXX11/src/Tensor/TensorReductionCuda.h @@ -159,7 +159,7 @@ __global__ void ReductionInitKernelHalfFloat(Reducer reducer, const Self input, template <int BlockSize, int NumPerThread, typename Self, typename Reducer, typename Index> -static __global__ void FullReductionKernelHalfFloat(Reducer reducer, const Self input, Index num_coeffs, +__global__ void FullReductionKernelHalfFloat(Reducer reducer, const Self input, Index num_coeffs, half* output, half2* scratch) { eigen_assert(NumPerThread % 2 == 0); |