GitHub-Proxy/eigen
1
0
Fork 0
mirror of https://gitlab.com/libeigen/eigen.git synced 2025-04-23 10:09:36 +08:00
Code Issues Packages Projects Releases Wiki Activity

Made it possible to compile reductions for an old cuda architecture and run them on a recent gpu.

Browse Source
This commit is contained in:
Benoit Steiner 2016-06-29 15:42:01 -07:00
parent b2a47641ce
commit cb2d8b8fa6
1 changed files with 10 additions and 2 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

12
unsupported/Eigen/CXX11/src/Tensor/TensorReductionCuda.h
View File

@ -336,9 +336,11 @@ struct FullReducer<Self, Op, GpuDevice, Vectorizable> {
static const bool HasOptimizedImplementation = !Op::IsStateful && static const bool HasOptimizedImplementation = !Op::IsStateful &&
(internal::is_same<typename Self::CoeffReturnType, float>::value || (internal::is_same<typename Self::CoeffReturnType, float>::value ||
(internal::is_same<typename Self::CoeffReturnType, Eigen::half>::value && reducer_traits<Op, GpuDevice>::PacketAccess)); (internal::is_same<typename Self::CoeffReturnType, Eigen::half>::value && reducer_traits<Op, GpuDevice>::PacketAccess));
#else #elif __CUDA_ARCH__ >= 300
static const bool HasOptimizedImplementation = !Op::IsStateful && static const bool HasOptimizedImplementation = !Op::IsStateful &&
internal::is_same<typename Self::CoeffReturnType, float>::value; internal::is_same<typename Self::CoeffReturnType, float>::value;
#else
static const bool HasOptimizedImplementation = false;
#endif #endif
template <typename OutputType> template <typename OutputType>
@ -617,9 +619,11 @@ struct InnerReducer<Self, Op, GpuDevice> {
static const bool HasOptimizedImplementation = !Op::IsStateful && static const bool HasOptimizedImplementation = !Op::IsStateful &&
(internal::is_same<typename Self::CoeffReturnType, float>::value || (internal::is_same<typename Self::CoeffReturnType, float>::value ||
(internal::is_same<typename Self::CoeffReturnType, Eigen::half>::value && reducer_traits<Op, GpuDevice>::PacketAccess)); (internal::is_same<typename Self::CoeffReturnType, Eigen::half>::value && reducer_traits<Op, GpuDevice>::PacketAccess));
#else #elif __CUDA_ARCH__ >= 300
static const bool HasOptimizedImplementation = !Op::IsStateful && static const bool HasOptimizedImplementation = !Op::IsStateful &&
internal::is_same<typename Self::CoeffReturnType, float>::value; internal::is_same<typename Self::CoeffReturnType, float>::value;
#else
static const bool HasOptimizedImplementation = false;
#endif #endif
template <typename OutputType> template <typename OutputType>
@ -674,8 +678,12 @@ struct OuterReducer<Self, Op, GpuDevice> {
// Unfortunately nvidia doesn't support well exotic types such as complex, // 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 // so reduce the scope of the optimized version of the code to the simple case
// of floats. // of floats.
#if __CUDA_ARCH__ >= 300
static const bool HasOptimizedImplementation = !Op::IsStateful && static const bool HasOptimizedImplementation = !Op::IsStateful &&
internal::is_same<typename Self::CoeffReturnType, float>::value; internal::is_same<typename Self::CoeffReturnType, float>::value;
#else
static const bool HasOptimizedImplementation = false;
#endif
template <typename Device, typename OutputType> template <typename Device, typename OutputType>
static EIGEN_DEVICE_FUNC bool run(const Self&, Op&, const Device&, OutputType*, typename Self::Index, typename Self::Index) { static EIGEN_DEVICE_FUNC bool run(const Self&, Op&, const Device&, OutputType*, typename Self::Index, typename Self::Index) {