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)));

11
Eigen/src/Core/arch/GPU/PacketMath.h
View File

@@ -481,7 +481,7 @@ ptranspose(PacketBlock<double2,2>& kernel) {
// Packet4h2 must be defined in the macro without EIGEN_CUDA_ARCH, meaning
// its corresponding packet_traits<Eigen::half> must be visible on host.
#if (defined(EIGEN_HAS_CUDA_FP16) && defined(EIGEN_CUDACC)) || \
(defined(EIGEN_HAS_HIP_FP16) && defined(EIGEN_HIPCC) && defined(EIGEN_HIP_DEVICE_COMPILE)) || \
(defined(EIGEN_HAS_HIP_FP16) && defined(EIGEN_HIPCC)) || \
(defined(EIGEN_HAS_CUDA_FP16) && defined(__clang__) && defined(__CUDA__))
typedef ulonglong2 Packet4h2;
@@ -515,11 +515,13 @@ template<> struct packet_traits<Eigen::half> : default_packet_traits
template<>
EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE half2 pset1<half2>(const Eigen::half& from) {
#if !defined(EIGEN_CUDA_ARCH) && !defined(EIGEN_HIP_DEVICE_COMPILE)
#if !defined(EIGEN_CUDA_ARCH) && !defined(EIGEN_HIPCC)
half2 r;
r.x = from;
r.y = from;
return r;
#elif defined(EIGEN_HIPCC)
return __half2{from,from};
#else
return __half2half2(from);
#endif
@@ -537,7 +539,7 @@ pset1<Packet4h2>(const Eigen::half& from) {
return r;
}
#if defined(EIGEN_CUDA_ARCH) || defined(EIGEN_HIP_DEVICE_COMPILE) || (defined(EIGEN_CUDACC) && EIGEN_COMP_CLANG && !EIGEN_COMP_NVCC)
#if defined(EIGEN_CUDA_ARCH) || defined(EIGEN_HIPCC) || (defined(EIGEN_CUDACC) && EIGEN_COMP_CLANG && !EIGEN_COMP_NVCC)
namespace {
EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE half2 pload(const Eigen::half* from) {
@@ -559,7 +561,7 @@ EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE void pstore(Eigen::half* to,
EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE void pstoreu(Eigen::half* to,
const half2& from) {
#if !defined(EIGEN_CUDA_ARCH) && !defined(EIGEN_HIP_DEVICE_COMPILE)
#if !defined(EIGEN_CUDA_ARCH) && !defined(EIGEN_HIPCC)
to[0] = from.x;
to[1] = from.y;
#else
@@ -1056,7 +1058,6 @@ EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE half2 prsqrt(const half2& a) {
#endif
} // namespace
template <>
EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE Packet4h2
pload<Packet4h2>(const Eigen::half* from) {

3
Eigen/src/Core/util/ConfigureVectorization.h
View File

@@ -439,9 +439,6 @@
#if defined(EIGEN_HIPCC)
#define EIGEN_VECTORIZE_GPU
#include <hip/hip_vector_types.h>
#endif
#if defined(EIGEN_HIP_DEVICE_COMPILE)
#define EIGEN_HAS_HIP_FP16
#include <hip/hip_fp16.h>
#endif