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

Fixing HIP breakage caused by the recent commit that introduces Packet4h2 as the Eigen::Half packet type

Browse Source
This commit is contained in:
Deven Desai
2020-03-11 23:06:56 +00:00
parent d53ae40f7b
commit 7158ed4e0e
4 changed files with 38 additions and 24 deletions
Download Patch File Download Diff File Expand all files Collapse all files

4
Eigen/src/Core/arch/Default/Half.h
View File

@@ -706,7 +706,7 @@ struct hash<Eigen::half> {
// Add the missing shfl_xor intrinsic
#if (defined(EIGEN_CUDA_ARCH) && EIGEN_CUDA_ARCH >= 300) || \
defined(EIGEN_HIP_DEVICE_COMPILE)
defined(EIGEN_HIPCC)
__device__ EIGEN_STRONG_INLINE Eigen::half __shfl_xor(Eigen::half var, int laneMask, int width=warpSize) {
#if (EIGEN_CUDA_SDK_VER < 90000) || \
@@ -720,7 +720,7 @@ __device__ EIGEN_STRONG_INLINE Eigen::half __shfl_xor(Eigen::half var, int laneM
// ldg() has an overload for __half_raw, but we also need one for Eigen::half.
#if (defined(EIGEN_CUDA_ARCH) && EIGEN_CUDA_ARCH >= 350) || \
defined(EIGEN_HIP_DEVICE_COMPILE)
defined(EIGEN_HIPCC)
EIGEN_STRONG_INLINE EIGEN_DEVICE_FUNC Eigen::half __ldg(const Eigen::half* ptr) {
return Eigen::half_impl::raw_uint16_to_half(
__ldg(reinterpret_cast<const unsigned short*>(ptr)));