diff options
| author |  Benoit Steiner <benoit.steiner.goog@gmail.com> | 2016-08-03 10:00:43 -0700 |
|---|---|---|
| committer | Benoit Steiner <benoit.steiner.goog@gmail.com> | 2016-08-03 10:00:43 -0700 |
| commit | a20b58845f5f457375a91ec7e8acdeee2f920d33 (patch) | |
| tree | cd762252deb305386b4690d11bb81ef5005b343b /unsupported | |
| parent | 819d0cea1b8cc27c55c94886c85a0a3efc155a50 (diff) | |
CUDA_ARCH isn't always defined, so avoid relying on it too much when figuring out which implementation to use for reductions. Instead rely on the device to tell us on which hardware version we're running.
Diffstat (limited to 'unsupported')
| -rw-r--r-- | unsupported/Eigen/CXX11/src/Tensor/TensorReductionCuda.h | 12 |
1 files changed, 2 insertions, 10 deletions
diff --git a/unsupported/Eigen/CXX11/src/Tensor/TensorReductionCuda.h b/unsupported/Eigen/CXX11/src/Tensor/TensorReductionCuda.h index d3894e625..5e512490c 100644 --- a/unsupported/Eigen/CXX11/src/Tensor/TensorReductionCuda.h +++ b/unsupported/Eigen/CXX11/src/Tensor/TensorReductionCuda.h @@ -336,11 +336,9 @@ struct FullReducer<Self, Op, GpuDevice, Vectorizable> { static const bool HasOptimizedImplementation = !Op::IsStateful && (internal::is_same<typename Self::CoeffReturnType, float>::value || (internal::is_same<typename Self::CoeffReturnType, Eigen::half>::value && reducer_traits<Op, GpuDevice>::PacketAccess)); -#elif __CUDA_ARCH__ >= 300 +#else static const bool HasOptimizedImplementation = !Op::IsStateful && internal::is_same<typename Self::CoeffReturnType, float>::value; -#else - static const bool HasOptimizedImplementation = false; #endif template <typename OutputType> @@ -619,11 +617,9 @@ struct InnerReducer<Self, Op, GpuDevice> { static const bool HasOptimizedImplementation = !Op::IsStateful && (internal::is_same<typename Self::CoeffReturnType, float>::value || (internal::is_same<typename Self::CoeffReturnType, Eigen::half>::value && reducer_traits<Op, GpuDevice>::PacketAccess)); -#elif __CUDA_ARCH__ >= 300 +#else static const bool HasOptimizedImplementation = !Op::IsStateful && internal::is_same<typename Self::CoeffReturnType, float>::value; -#else - static const bool HasOptimizedImplementation = false; #endif template <typename OutputType> @@ -678,12 +674,8 @@ struct OuterReducer<Self, Op, GpuDevice> { // Unfortunately nvidia doesn't support well exotic types such as complex, // so reduce the scope of the optimized version of the code to the simple case // of floats. -#if __CUDA_ARCH__ >= 300 static const bool HasOptimizedImplementation = !Op::IsStateful && internal::is_same<typename Self::CoeffReturnType, float>::value; -#else - static const bool HasOptimizedImplementation = false; -#endif template <typename Device, typename OutputType> static EIGEN_DEVICE_FUNC bool run(const Self&, Op&, const Device&, OutputType*, typename Self::Index, typename Self::Index) { |