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 pow for integer base / exponent types

Browse Source
This commit is contained in:
Charles Schlosser
2022-08-29 19:23:54 +00:00
committed by Rasmus Munk Larsen
parent 8acbf5c11c
commit e5af9f87f2
7 changed files with 134 additions and 8 deletions
Download Patch File Download Diff File Expand all files Collapse all files

43
Eigen/src/Core/arch/Default/GenericPacketMathFunctions.h
View File

@@ -1880,6 +1880,36 @@ static EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE Packet handle_nonint_nonint_errors(
result = pselect(pandnot(abs_x_is_one, x_is_neg), cst_pos_one, result);
return result;
}
template <typename Packet, typename ScalarExponent>
static EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE Packet handle_int_int(const Packet& x, const ScalarExponent& exponent) {
typedef typename unpacket_traits<Packet>::type Scalar;
// integer base, integer exponent case
// This routine handles negative and very large positive exponents
// Signed integer overflow and divide by zero is undefined behavior
// Unsigned intgers do not overflow
const bool exponent_is_odd = unary_pow::is_odd<ScalarExponent>::run(exponent);
const Scalar zero = Scalar(0);
const Scalar pos_one = Scalar(1);
const Packet cst_zero = pset1<Packet>(zero);
const Packet cst_pos_one = pset1<Packet>(pos_one);
const Packet abs_x = pabs(x);
const Packet pow_is_zero = exponent < 0 ? pcmp_lt(cst_pos_one, abs_x) : pzero(x);
const Packet pow_is_one = pcmp_eq(cst_pos_one, abs_x);
const Packet pow_is_neg = exponent_is_odd ? pcmp_lt(x, cst_zero) : pzero(x);
Packet result = pselect(pow_is_zero, cst_zero, x);
result = pselect(pandnot(pow_is_one, pow_is_neg), cst_pos_one, result);
result = pselect(pand(pow_is_one, pow_is_neg), pnegate(cst_pos_one), result);
return result;
}
} // end namespace unary_pow
template <typename Packet, typename ScalarExponent,
@@ -1914,6 +1944,19 @@ struct unary_pow_impl<Packet, ScalarExponent, false, true> {
}
};
template <typename Packet, typename ScalarExponent>
struct unary_pow_impl<Packet, ScalarExponent, true, true> {
typedef typename unpacket_traits<Packet>::type Scalar;
static EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE Packet run(const Packet& x, const ScalarExponent& exponent) {
if (exponent < 0 || exponent > NumTraits<Scalar>::digits()) {
return unary_pow::handle_int_int(x, exponent);
}
else {
return unary_pow::int_pow(x, exponent);
}
}
};
} // end namespace internal
} // end namespace Eigen