GitHub-Proxy/eigen
1
0
Fork 0
mirror of https://gitlab.com/libeigen/eigen.git synced 2025-09-13 09:53:13 +08:00
Code Issues Packages Projects Releases Wiki Activity

Optimize psign

Browse Source
This commit is contained in:
Charles Schlosser 2023-02-09 22:15:26 +00:00 committed by Rasmus Munk Larsen
parent 0e490d452d
commit 325e3063d9
1 changed files with 9 additions and 17 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

20
Eigen/src/Core/arch/Default/GenericPacketMathFunctions.h
View File

@ -1097,31 +1097,23 @@ Packet psqrt_complex(const Packet& a) {
template <typename Packet>
struct psign_impl<
Packet,
std::enable_if_t<
!NumTraits<typename unpacket_traits<Packet>::type>::IsComplex &&
struct psign_impl<Packet, std::enable_if_t<!NumTraits<typename unpacket_traits<Packet>::type>::IsComplex &&
!NumTraits<typename unpacket_traits<Packet>::type>::IsInteger>> {
static EIGEN_DEVICE_FUNC inline Packet run(const Packet& a) {
using Scalar = typename unpacket_traits<Packet>::type;
const Packet cst_one = pset1<Packet>(Scalar(1));
const Packet cst_minus_one = pset1<Packet>(Scalar(-1));
const Packet cst_zero = pzero(a);
const Packet not_nan_mask = pcmp_eq(a, a);
const Packet positive_mask = pcmp_lt(cst_zero, a);
const Packet positive = pand(positive_mask, cst_one);
const Packet negative_mask = pcmp_lt(a, cst_zero);
const Packet negative = pand(negative_mask, cst_minus_one);
const Packet abs_a = pabs(a);
const Packet sign_mask = pandnot(a, abs_a);
const Packet nonzero_mask = pcmp_lt(cst_zero, abs_a);
return pselect(not_nan_mask, por(positive, negative), a);
return pselect(nonzero_mask, por(sign_mask, cst_one), abs_a);
}
};
template <typename Packet>
struct psign_impl<
Packet, std::enable_if_t<
!NumTraits<typename unpacket_traits<Packet>::type>::IsComplex &&
struct psign_impl<Packet, std::enable_if_t<!NumTraits<typename unpacket_traits<Packet>::type>::IsComplex &&
NumTraits<typename unpacket_traits<Packet>::type>::IsSigned &&
NumTraits<typename unpacket_traits<Packet>::type>::IsInteger>> {
static EIGEN_DEVICE_FUNC inline Packet run(const Packet& a) {