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

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