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

Bug #1785: Introduce numext::rint.

Browse Source 
This provides a new op that matches std::rint and previous behavior of
pround. Also adds corresponding unsupported/../Tensor op.
Performance is the same as e. g. floor (tested SSE/AVX).
This commit is contained in:
Ilya Tokar
2019-12-16 16:00:35 -05:00
parent d0ae052da4
commit 19876ced76
12 changed files with 117 additions and 1 deletions
Download Patch File Download Diff File Expand all files Collapse all files

5
Eigen/src/Core/arch/AVX/PacketMath.h
View File

@@ -81,7 +81,8 @@ template<> struct packet_traits<float> : default_packet_traits
HasBlend = 1,
HasRound = 1,
HasFloor = 1,
HasCeil = 1
HasCeil = 1,
HasRint = 1
};
};
template<> struct packet_traits<double> : default_packet_traits
@@ -316,6 +317,8 @@ template<> EIGEN_STRONG_INLINE Packet8i pcmp_eq(const Packet8i& a, const Packet8
#endif
}
template<> EIGEN_STRONG_INLINE Packet8f print<Packet8f>(const Packet8f& a) { return _mm256_round_ps(a, _MM_FROUND_CUR_DIRECTION); }
template<> EIGEN_STRONG_INLINE Packet4d print<Packet4d>(const Packet4d& a) { return _mm256_round_pd(a, _MM_FROUND_CUR_DIRECTION); }
template<> EIGEN_STRONG_INLINE Packet8f pceil<Packet8f>(const Packet8f& a) { return _mm256_ceil_ps(a); }
template<> EIGEN_STRONG_INLINE Packet4d pceil<Packet4d>(const Packet4d& a) { return _mm256_ceil_pd(a); }

5
Eigen/src/Core/arch/SSE/PacketMath.h
View File

@@ -124,6 +124,7 @@ struct packet_traits<float> : default_packet_traits {
#ifdef EIGEN_VECTORIZE_SSE4_1
,
HasRint = 1,
HasRound = 1,
HasCeil = 1
#endif
@@ -148,6 +149,7 @@ struct packet_traits<double> : default_packet_traits {
#ifdef EIGEN_VECTORIZE_SSE4_1
,
HasRound = 1,
HasRint = 1,
HasFloor = 1,
HasCeil = 1
#endif
@@ -443,6 +445,9 @@ template<> EIGEN_STRONG_INLINE Packet2d pround<Packet2d>(const Packet2d& a)
return _mm_round_pd(padd(por(pand(a, mask), prev0dot5), a), _MM_FROUND_TO_ZERO);
}
template<> EIGEN_STRONG_INLINE Packet4f print<Packet4f>(const Packet4f& a) { return _mm_round_ps(a, _MM_FROUND_CUR_DIRECTION); }
template<> EIGEN_STRONG_INLINE Packet2d print<Packet2d>(const Packet2d& a) { return _mm_round_pd(a, _MM_FROUND_CUR_DIRECTION); }
template<> EIGEN_STRONG_INLINE Packet4f pceil<Packet4f>(const Packet4f& a) { return _mm_ceil_ps(a); }
template<> EIGEN_STRONG_INLINE Packet2d pceil<Packet2d>(const Packet2d& a) { return _mm_ceil_pd(a); }