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

Disable cuda Eigen::half vectorization on host.

Browse Source 
All cuda `__half` functions are device-only in CUDA 9, including
conversions. Host-side conversions were added in CUDA 10.
The existing code doesn't build prior to 10.0.

All arithmetic functions are always device-only, so there's
therefore no reason to use vectorization on the host at all.

Modified the code to disable vectorization for `__half` on host,
which required also updating the `TensorReductionGpu` implementation
which previously made assumptions about available packets.
This commit is contained in:
Antonio Sanchez
2021-08-31 08:33:19 -07:00
committed by Rasmus Munk Larsen
parent 1415817d8d
commit cc3573ab44
4 changed files with 126 additions and 156 deletions
Download Patch File Download Diff File Expand all files Collapse all files

2
unsupported/Eigen/CXX11/src/Tensor/TensorMeta.h
View File

@@ -52,7 +52,7 @@ struct PacketType : internal::packet_traits<Scalar> {
};
// For CUDA packet types when using a GpuDevice
#if defined(EIGEN_USE_GPU) && defined(EIGEN_HAS_GPU_FP16)
#if defined(EIGEN_USE_GPU) && defined(EIGEN_HAS_GPU_FP16) && defined(EIGEN_GPU_COMPILE_PHASE)
typedef ulonglong2 Packet4h2;
template<>