digamma special function: merge shared code.

Moved type-specific code into a helper class digamma_impl_maybe_poly<Scalar>.
This commit is contained in:
Eugene Brevdo 2016-01-27 09:52:29 -08:00
parent 6a75e7e0d5
commit c8d94ae944

View File

@ -134,8 +134,23 @@ struct lgamma_impl<double> {
* Implementation of digamma (psi) *
****************************************************************************/
#ifdef EIGEN_HAS_C99_MATH
/*
*
* Polynomial evaluation helper for the Psi (digamma) function.
*
* digamma_impl_maybe_poly::run(s) evaluates the asymptotic Psi expansion for
* input Scalar s, assuming s is above 10.0.
*
* If s is above a certain threshold for the given Scalar type, zero
* is returned. Otherwise the polynomial is evaluated with enough
* coefficients for results matching Scalar machine precision.
*
*
*/
template <typename Scalar>
struct digamma_impl {
struct digamma_impl_maybe_poly {
EIGEN_DEVICE_FUNC
static EIGEN_STRONG_INLINE Scalar run(const Scalar) {
EIGEN_STATIC_ASSERT((internal::is_same<Scalar, Scalar>::value == false),
@ -144,72 +159,11 @@ struct digamma_impl {
}
};
template <typename Scalar>
struct digamma_retval {
typedef Scalar type;
};
#ifdef EIGEN_HAS_C99_MATH
template <>
struct digamma_impl<float> {
/*
* Psi (digamma) function (modified for Eigen)
*
*
* SYNOPSIS:
*
* float x, y, psif();
*
* y = psif( x );
*
*
* DESCRIPTION:
*
* d -
* psi(x) = -- ln | (x)
* dx
*
* is the logarithmic derivative of the gamma function.
* For integer x,
* n-1
* -
* psi(n) = -EUL + > 1/k.
* -
* k=1
*
* If x is negative, it is transformed to a positive argument by the
* reflection formula psi(1-x) = psi(x) + pi cot(pi x).
* For general positive x, the argument is made greater than 10
* using the recurrence psi(x+1) = psi(x) + 1/x.
* Then the following asymptotic expansion is applied:
*
* inf. B
* - 2k
* psi(x) = log(x) - 1/2x - > -------
* - 2k
* k=1 2k x
*
* where the B2k are Bernoulli numbers.
*
* ACCURACY:
* Absolute error, relative when |psi| > 1 :
* arithmetic domain # trials peak rms
* IEEE -33,0 30000 8.2e-7 1.2e-7
* IEEE 0,33 100000 7.3e-7 7.7e-8
*
* ERROR MESSAGES:
* message condition value returned
* psi singularity x integer <=0 INFINITY
*/
struct digamma_impl_maybe_poly<float> {
EIGEN_DEVICE_FUNC
static float run(float xx) {
float p, q, nz, x, s, w, y, z;
bool negative;
// Some necessary constants
const float m_pif = 3.141592653589793238;
const float maxnumf = std::numeric_limits<float>::infinity();
static EIGEN_STRONG_INLINE float run(const float s) {
const float A[] = {
-4.16666666666666666667E-3,
3.96825396825396825397E-3,
@ -217,53 +171,49 @@ struct digamma_impl<float> {
8.33333333333333333333E-2
};
x = xx;
nz = 0.0f;
negative = 0;
if (x <= 0.0f) {
negative = 1;
q = x;
p = ::floor(q);
if (p == q) {
return (maxnumf);
}
nz = q - p;
if (nz != 0.5f) {
if (nz > 0.5f) {
p += 1.0f;
nz = q - p;
}
nz = m_pif / ::tan(m_pif * nz);
} else {
nz = 0.0f;
}
x = 1.0f - x;
}
/* use the recurrence psi(x+1) = psi(x) + 1/x. */
s = x;
w = 0.0f;
while (s < 10.0f) {
w += 1.0f / s;
s += 1.0f;
}
float z;
if (s < 1.0e8f) {
z = 1.0f / (s * s);
y = z * cephes::polevl<float, 3>::run(z, A);
} else
y = 0.0f;
y = ::log(s) - (0.5f / s) - y - w;
return (negative) ? y - nz : y;
return z * cephes::polevl<float, 3>::run(z, A);
} else return 0.0f;
}
};
template <>
struct digamma_impl<double> {
struct digamma_impl_maybe_poly<double> {
EIGEN_DEVICE_FUNC
static double run(double x) {
static EIGEN_STRONG_INLINE double run(const double s) {
const double A[] = {
8.33333333333333333333E-2,
-2.10927960927960927961E-2,
7.57575757575757575758E-3,
-4.16666666666666666667E-3,
3.96825396825396825397E-3,
-8.33333333333333333333E-3,
8.33333333333333333333E-2
};
double z;
if (s < 1.0e17) {
z = 1.0 / (s * s);
return z * cephes::polevl<double, 6>::run(z, A);
}
else return 0.0;
}
};
#endif // EIGEN_HAS_C99_MATH
template <typename Scalar>
struct digamma_retval {
typedef Scalar type;
};
#ifdef EIGEN_HAS_C99_MATH
template <typename Scalar>
struct digamma_impl {
EIGEN_DEVICE_FUNC
static Scalar run(Scalar x) {
/*
*
* Psi (digamma) function (modified for Eigen)
@ -304,38 +254,38 @@ struct digamma_impl<double> {
*
* where the B2k are Bernoulli numbers.
*
* ACCURACY:
* ACCURACY (float):
* Relative error (except absolute when |psi| < 1):
* arithmetic domain # trials peak rms
* IEEE 0,30 30000 1.3e-15 1.4e-16
* IEEE -30,0 40000 1.5e-15 2.2e-16
*
* ACCURACY (double):
* Absolute error, relative when |psi| > 1 :
* arithmetic domain # trials peak rms
* IEEE -33,0 30000 8.2e-7 1.2e-7
* IEEE 0,33 100000 7.3e-7 7.7e-8
*
* ERROR MESSAGES:
* message condition value returned
* psi singularity x integer <=0 INFINITY
*/
double p, q, nz, s, w, y, z;
Scalar p, q, nz, s, w, y;
bool negative;
const double A[] = {
8.33333333333333333333E-2,
-2.10927960927960927961E-2,
7.57575757575757575758E-3,
-4.16666666666666666667E-3,
3.96825396825396825397E-3,
-8.33333333333333333333E-3,
8.33333333333333333333E-2
};
const double maxnum = std::numeric_limits<double>::infinity();
const double m_pi = 3.14159265358979323846;
const Scalar maxnum = std::numeric_limits<Scalar>::infinity();
const Scalar m_pi = 3.14159265358979323846;
negative = 0;
nz = 0.0;
if (x <= 0.0) {
negative = 1;
const Scalar zero = 0.0;
const Scalar one = 1.0;
const Scalar half = 0.5;
if (x <= zero) {
negative = one;
q = x;
p = ::floor(q);
if (p == q) {
@ -345,41 +295,36 @@ struct digamma_impl<double> {
* by subtracting the nearest integer from x
*/
nz = q - p;
if (nz != 0.5) {
if (nz > 0.5) {
p += 1.0;
if (nz != half) {
if (nz > half) {
p += one;
nz = q - p;
}
nz = m_pi / ::tan(m_pi * nz);
}
else {
nz = 0.0;
nz = zero;
}
x = 1.0 - x;
x = one - x;
}
/* use the recurrence psi(x+1) = psi(x) + 1/x. */
s = x;
w = 0.0;
while (s < 10.0) {
w += 1.0 / s;
s += 1.0;
w = zero;
while (s < Scalar(10)) {
w += one / s;
s += one;
}
if (s < 1.0e17) {
z = 1.0 / (s * s);
y = z * cephes::polevl<double, 6>::run(z, A);
}
else
y = 0.0;
y = digamma_impl_maybe_poly<Scalar>::run(s);
y = ::log(s) - (0.5 / s) - y - w;
y = ::log(s) - (half / s) - y - w;
return (negative) ? y - nz : y;
}
};
#endif
#endif // EIGEN_HAS_C99_MATH
/****************************************************************************
* Implementation of erf *