devtools/eigen
1
0
Fork 0
mirror of https://gitlab.com/libeigen/eigen.git synced 2026-04-10 11:34:33 +08:00
Code Issues Packages Projects Releases Wiki Activity

remove duplicate pset1 for half and add some comments about why we need expose pmul/add/div/min/max on host

Browse Source
This commit is contained in:
Sami Kama
2020-03-10 20:28:43 +00:00
committed by Rasmus Munk Larsen
parent a45d28256d
commit b733b8b680
5 changed files with 1069 additions and 167 deletions
Download Patch File Download Diff File Expand all files Collapse all files

8
unsupported/Eigen/CXX11/src/Tensor/TensorReduction.h
View File

@@ -420,9 +420,9 @@ __global__ void FullReductionKernel(R, const S, I_, typename S::CoeffReturnType*
#if defined(EIGEN_HAS_GPU_FP16)
template <typename S, typename R, typename I_>
__global__ void ReductionInitFullReduxKernelHalfFloat(R, const S, I_, half2*);
__global__ void ReductionInitFullReduxKernelHalfFloat(R, const S, I_, internal::packet_traits<half>::type*);
template <int B, int N, typename S, typename R, typename I_>
__global__ void FullReductionKernelHalfFloat(R, const S, I_, half*, half2*);
__global__ void FullReductionKernelHalfFloat(R, const S, I_, half*, internal::packet_traits<half>::type*);
template <int NPT, typename S, typename R, typename I_>
__global__ void InnerReductionKernelHalfFloat(R, const S, I_, I_, half*);
@@ -863,8 +863,8 @@ struct TensorReductionEvaluatorBase<const TensorReductionOp<Op, Dims, ArgType, M
#if defined(EIGEN_USE_GPU) && (defined(EIGEN_GPUCC))
template <int B, int N, typename S, typename R, typename I_> KERNEL_FRIEND void internal::FullReductionKernel(R, const S, I_, typename S::CoeffReturnType*, unsigned int*);
#if defined(EIGEN_HAS_GPU_FP16)
template <typename S, typename R, typename I_> KERNEL_FRIEND void internal::ReductionInitFullReduxKernelHalfFloat(R, const S, I_, half2*);
template <int B, int N, typename S, typename R, typename I_> KERNEL_FRIEND void internal::FullReductionKernelHalfFloat(R, const S, I_, half*, half2*);
template <typename S, typename R, typename I_> KERNEL_FRIEND void internal::ReductionInitFullReduxKernelHalfFloat(R, const S, I_, internal::packet_traits<Eigen::half>::type*);
template <int B, int N, typename S, typename R, typename I_> KERNEL_FRIEND void internal::FullReductionKernelHalfFloat(R, const S, I_, half*, internal::packet_traits<Eigen::half>::type*);
template <int NPT, typename S, typename R, typename I_> KERNEL_FRIEND void internal::InnerReductionKernelHalfFloat(R, const S, I_, I_, half*);
#endif
template <int NPT, typename S, typename R, typename I_> KERNEL_FRIEND void internal::InnerReductionKernel(R, const S, I_, I_, typename S::CoeffReturnType*);