GitHub-Proxy/eigen
1
0
Fork 0
mirror of https://gitlab.com/libeigen/eigen.git synced 2025-08-11 19:29:02 +08:00
Code Issues Packages Projects Releases Wiki Activity

Improved support for CUDA 8.0

Browse Source
This commit is contained in:
Benoit Steiner 2016-05-31 09:47:59 -07:00
parent c4bd3b1f21
commit b6e306f189
1 changed files with 26 additions and 0 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

26
Eigen/src/Core/arch/CUDA/PacketMathHalf.h
View File

@ -264,6 +264,30 @@ template<> EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE Eigen::half predux_mul<half2>(c
#endif #endif
} }
#if defined __CUDACC_VER__ && __CUDACC_VER__ >= 80000 && defined __CUDA_ARCH__ && __CUDA_ARCH__ >= 530
template<> EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
half2 plog<half2>(const half2& a) {
return h2log(a);
}
template<> EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
half2 pexp<half2>(const half2& a) {
return h2exp(a);
}
template<> EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
half2 psqrt<half2>(const half2& a) {
return h2sqrt(a);
}
template<> EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
half2 prsqrt<half2>(const half2& a) {
return h2rsqrt(a);
}
#else
template<> EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE half2 plog<half2>(const half2& a) { template<> EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE half2 plog<half2>(const half2& a) {
float a1 = __low2float(a); float a1 = __low2float(a);
float a2 = __high2float(a); float a2 = __high2float(a);
@ -296,6 +320,8 @@ template<> EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE half2 prsqrt<half2>(const half2
return __floats2half2_rn(r1, r2); return __floats2half2_rn(r1, r2);
} }
#endif
#elif defined EIGEN_VECTORIZE_AVX #elif defined EIGEN_VECTORIZE_AVX
typedef struct { typedef struct {