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

Vectorize erfc(x) for double and improve erfc(x) for float.

Browse Source
This commit is contained in:
Rasmus Munk Larsen
2024-11-08 17:21:11 +00:00
parent 8adf43640e
commit 0d366f6532
9 changed files with 141 additions and 18 deletions
Download Patch File Download Diff File Expand all files Collapse all files

2
Eigen/src/Core/arch/AVX/PacketMath.h
View File

@@ -124,6 +124,7 @@ struct packet_traits<float> : default_packet_traits {
HasRsqrt = 1,
HasTanh = EIGEN_FAST_MATH,
HasErf = EIGEN_FAST_MATH,
HasErfc = EIGEN_FAST_MATH,
HasBlend = 1
};
};
@@ -144,6 +145,7 @@ struct packet_traits<double> : default_packet_traits {
#endif
HasTanh = EIGEN_FAST_MATH,
HasLog = 1,
HasErfc = 1,
HasExp = 1,
HasSqrt = 1,
HasRsqrt = 1,

2
Eigen/src/Core/arch/AVX512/PacketMath.h
View File

@@ -133,6 +133,7 @@ struct packet_traits<float> : default_packet_traits {
HasReciprocal = EIGEN_FAST_MATH,
HasTanh = EIGEN_FAST_MATH,
HasErf = EIGEN_FAST_MATH,
HasErfc = EIGEN_FAST_MATH,
HasCmp = 1,
HasDiv = 1
};
@@ -154,6 +155,7 @@ struct packet_traits<double> : default_packet_traits {
HasExp = 1,
HasATan = 1,
HasTanh = EIGEN_FAST_MATH,
HasErfc = EIGEN_FAST_MATH,
HasATanh = 1,
HasCmp = 1,
HasDiv = 1

2
Eigen/src/Core/arch/AltiVec/PacketMath.h
View File

@@ -193,6 +193,7 @@ struct packet_traits<float> : default_packet_traits {
#endif
HasTanh = EIGEN_FAST_MATH,
HasErf = EIGEN_FAST_MATH,
HasErfc = EIGEN_FAST_MATH,
#else
HasSqrt = 0,
HasRsqrt = 0,
@@ -3182,6 +3183,7 @@ struct packet_traits<double> : default_packet_traits {
HasSin = EIGEN_FAST_MATH,
HasCos = EIGEN_FAST_MATH,
HasTanh = EIGEN_FAST_MATH,
HasErfc = EIGEN_FAST_MATH,
HasATanh = 1,
HasATan = 0,
HasLog = 0,

22
Eigen/src/Core/arch/Default/GenericPacketMathFunctions.h
View File

@@ -1659,6 +1659,13 @@ EIGEN_STRONG_INLINE void twoprod(const Packet& x, const Packet& y, Packet& p_hi,
p_lo = pmsub(x, y, p_hi);
}
// A version of twoprod that takes x, y, and fl(x*y) as input and returns the p_lo such that
// x * y = xy + p_lo holds exactly.
template <typename Packet>
EIGEN_STRONG_INLINE Packet twoprod_low(const Packet& x, const Packet& y, const Packet& xy) {
return pmsub(x, y, xy);
}
#else
// This function implements the Veltkamp splitting. Given a floating point
@@ -1694,6 +1701,21 @@ EIGEN_STRONG_INLINE void twoprod(const Packet& x, const Packet& y, Packet& p_hi,
p_lo = pmadd(x_lo, y_lo, p_lo);
}
// A version of twoprod that takes x, y, and fl(x*y) as input and returns the p_lo such that
// x * y = xy + p_lo holds exactly.
template <typename Packet>
EIGEN_STRONG_INLINE Packet twoprod_low(const Packet& x, const Packet& y, const Packet& xy) {
Packet x_hi, x_lo, y_hi, y_lo;
veltkamp_splitting(x, x_hi, x_lo);
veltkamp_splitting(y, y_hi, y_lo);
Packet p_lo = pmadd(x_hi, y_hi, pnegate(xy));
p_lo = pmadd(x_hi, y_lo, p_lo);
p_lo = pmadd(x_lo, y_hi, p_lo);
p_lo = pmadd(x_lo, y_lo, p_lo);
return p_lo;
}
#endif // EIGEN_VECTORIZE_FMA
// This function implements Dekker's algorithm for the addition

4
Eigen/src/Core/arch/NEON/PacketMath.h
View File

@@ -209,6 +209,7 @@ struct packet_traits<float> : default_packet_traits {
HasRsqrt = 1,
HasTanh = EIGEN_FAST_MATH,
HasErf = EIGEN_FAST_MATH,
HasErfc = EIGEN_FAST_MATH,
HasBessel = 0, // Issues with accuracy.
HasNdtri = 0
};
@@ -5140,7 +5141,8 @@ struct packet_traits<double> : default_packet_traits {
HasSqrt = 1,
HasRsqrt = 1,
HasTanh = EIGEN_FAST_MATH,
HasErf = 0
HasErf = 0,
HasErfc = EIGEN_FAST_MATH
};
};

2
Eigen/src/Core/arch/SSE/PacketMath.h
View File

@@ -197,6 +197,7 @@ struct packet_traits<float> : default_packet_traits {
HasRsqrt = 1,
HasTanh = EIGEN_FAST_MATH,
HasErf = EIGEN_FAST_MATH,
HasErfc = EIGEN_FAST_MATH,
HasBlend = 1,
HasSign = 0 // The manually vectorized version is slightly slower for SSE.
};
@@ -216,6 +217,7 @@ struct packet_traits<double> : default_packet_traits {
HasCos = EIGEN_FAST_MATH,
HasTanh = EIGEN_FAST_MATH,
HasLog = 1,
HasErfc = EIGEN_FAST_MATH,
HasExp = 1,
HasSqrt = 1,
HasRsqrt = 1,

3
Eigen/src/Core/arch/SVE/PacketMath.h
View File

@@ -360,7 +360,8 @@ struct packet_traits<float> : default_packet_traits {
HasExp = 1,
HasSqrt = 1,
HasTanh = EIGEN_FAST_MATH,
HasErf = EIGEN_FAST_MATH
HasErf = EIGEN_FAST_MATH,
HasErfc = EIGEN_FAST_MATH
};
};