GitHub-Proxy/eigen
1
0
Fork 0
mirror of https://gitlab.com/libeigen/eigen.git synced 2025-08-14 04:35:57 +08:00
Code Issues Packages Projects Releases Wiki Activity

Improved support for min and max on 16 bit floats when running on recent cuda gpus

Browse Source
This commit is contained in:
Benoit Steiner 2016-04-27 12:57:21 -07:00
parent 6744d776ba
commit 25141b69d4
1 changed files with 19 additions and 0 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

19
Eigen/src/Core/arch/CUDA/Half.h
View File

@ -465,6 +465,25 @@ template<> EIGEN_STRONG_INLINE EIGEN_DEVICE_FUNC Eigen::half ceil(const Eigen::h
return Eigen::half(::ceilf(float(a))); return Eigen::half(::ceilf(float(a)));
} }
template <> EIGEN_STRONG_INLINE EIGEN_DEVICE_FUNC Eigen::half mini(const Eigen::half& a, const Eigen::half& b) {
#if defined(EIGEN_HAS_CUDA_FP16) && defined(__CUDA_ARCH__) && __CUDA_ARCH__ >= 530
return __hlt(b, a) ? b : a;
#else
const float f1 = static_cast<float>(a);
const float f2 = static_cast<float>(b);
return f2 < f1 ? b : a;
#endif
}
template <> EIGEN_STRONG_INLINE EIGEN_DEVICE_FUNC Eigen::half maxi(const Eigen::half& a, const Eigen::half& b) {
#if defined(EIGEN_HAS_CUDA_FP16) && defined(__CUDA_ARCH__) && __CUDA_ARCH__ >= 530
return __hlt(a, b) ? b : a;
#else
const float f1 = static_cast<float>(a);
const float f2 = static_cast<float>(b);
return f1 < f2 ? b : a;
#endif
}
#ifdef EIGEN_HAS_C99_MATH #ifdef EIGEN_HAS_C99_MATH
template<> EIGEN_STRONG_INLINE EIGEN_DEVICE_FUNC Eigen::half lgamma(const Eigen::half& a) { template<> EIGEN_STRONG_INLINE EIGEN_DEVICE_FUNC Eigen::half lgamma(const Eigen::half& a) {
return Eigen::half(Eigen::numext::lgamma(static_cast<float>(a))); return Eigen::half(Eigen::numext::lgamma(static_cast<float>(a)));