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

Move implementation of vectorized error function erf() to SpecialFunctionsImpl.h.

Browse Source
This commit is contained in:
Rasmus Munk Larsen
2019-09-27 13:56:04 -07:00
parent 7c8bc0d928
commit 13ef08e5ac
12 changed files with 67 additions and 142 deletions
Download Patch File Download Diff File Expand all files Collapse all files

54
Eigen/src/Core/MathFunctionsImpl.h
View File

@@ -66,58 +66,6 @@ T generic_fast_tanh_float(const T& a_x)
return pdiv(p, q);
}
/** \internal \returns the error function of \a a (coeff-wise)
Doesn't do anything fancy, just a 13/8-degree rational interpolant which
is accurate up to a couple of ulp in the range [-4, 4], outside of which
fl(erf(x)) = +/-1.
This implementation works on both scalars and Ts.
*/
template <typename T>
T generic_fast_erf_float(const T& a_x) {
// Clamp the inputs to the range [-4, 4] since anything outside
// this range is +/-1.0f in single-precision.
const T plus_4 = pset1<T>(4.f);
const T minus_4 = pset1<T>(-4.f);
const T x = pmax(pmin(a_x, plus_4), minus_4);
// The monomial coefficients of the numerator polynomial (odd).
const T alpha_1 = pset1<T>(-1.60960333262415e-02f);
const T alpha_3 = pset1<T>(-2.95459980854025e-03f);
const T alpha_5 = pset1<T>(-7.34990630326855e-04f);
const T alpha_7 = pset1<T>(-5.69250639462346e-05f);
const T alpha_9 = pset1<T>(-2.10102402082508e-06f);
const T alpha_11 = pset1<T>(2.77068142495902e-08f);
const T alpha_13 = pset1<T>(-2.72614225801306e-10f);
// The monomial coefficients of the denominator polynomial (even).
const T beta_0 = pset1<T>(-1.42647390514189e-02f);
const T beta_2 = pset1<T>(-7.37332916720468e-03f);
const T beta_4 = pset1<T>(-1.68282697438203e-03f);
const T beta_6 = pset1<T>(-2.13374055278905e-04f);
const T beta_8 = pset1<T>(-1.45660718464996e-05f);
// Since the polynomials are odd/even, we need x^2.
const T x2 = pmul(x, x);
// Evaluate the numerator polynomial p.
T p = pmadd(x2, alpha_13, alpha_11);
p = pmadd(x2, p, alpha_9);
p = pmadd(x2, p, alpha_7);
p = pmadd(x2, p, alpha_5);
p = pmadd(x2, p, alpha_3);
p = pmadd(x2, p, alpha_1);
p = pmul(x, p);
// Evaluate the denominator polynomial p.
T q = pmadd(x2, beta_8, beta_6);
q = pmadd(x2, q, beta_4);
q = pmadd(x2, q, beta_2);
q = pmadd(x2, q, beta_0);
// Divide the numerator by the denominator.
return pdiv(p, q);
}
template<typename RealScalar>
EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
RealScalar positive_real_hypot(const RealScalar& x, const RealScalar& y)
@@ -126,7 +74,7 @@ RealScalar positive_real_hypot(const RealScalar& x, const RealScalar& y)
RealScalar p, qp;
p = numext::maxi(x,y);
if(p==RealScalar(0)) return RealScalar(0);
qp = numext::mini(y,x) / p;
qp = numext::mini(y,x) / p;
return p * sqrt(RealScalar(1) + qp*qp);
}