GitHub-Proxy/eigen
1
0
Fork 0
mirror of https://gitlab.com/libeigen/eigen.git synced 2025-08-11 11:19:02 +08:00
Code Issues Packages Projects Releases Wiki Activity

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.

Browse Source
This commit is contained in:
Benoit Steiner 2016-08-03 10:00:43 -07:00
parent 819d0cea1b
commit a20b58845f
1 changed files with 2 additions and 10 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

@ -336,11 +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) {