GitHub-Proxy/eigen
1
0
Fork 0
mirror of https://gitlab.com/libeigen/eigen.git synced 2025-09-12 17:33:15 +08:00
Code Issues Packages Projects Releases Wiki Activity

Fix a bug for pcmp_lt_or_nan and Add sqrt support for SVE

Browse Source
This commit is contained in:
qile lin 2024-09-04 21:45:39 +00:00 committed by Rasmus Munk Larsen
parent 9315389795
commit 072ec9d954
1 changed files with 8 additions and 3 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

11
Eigen/src/Core/arch/SVE/PacketMath.h
View File

@ -358,7 +358,7 @@ struct packet_traits<float> : default_packet_traits {
HasCos = EIGEN_FAST_MATH,
HasLog = 1,
HasExp = 1,
HasSqrt = 0,
HasSqrt = 1,
HasTanh = EIGEN_FAST_MATH,
HasErf = EIGEN_FAST_MATH
};
@ -478,12 +478,12 @@ EIGEN_STRONG_INLINE PacketXf pcmp_eq<PacketXf>(const PacketXf& a, const PacketXf
return svreinterpret_f32_u32(svdup_n_u32_z(svcmpeq_f32(svptrue_b32(), a, b), 0xffffffffu));
}
// Do a predicate inverse (svnot_b_x) on the predicate resulted from the
// Do a predicate inverse (svnot_b_z) on the predicate resulted from the
// greater/equal comparison (svcmpge_f32). Then fill a float vector with the
// active elements.
template <>
EIGEN_STRONG_INLINE PacketXf pcmp_lt_or_nan<PacketXf>(const PacketXf& a, const PacketXf& b) {
return svreinterpret_f32_u32(svdup_n_u32_z(svnot_b_x(svptrue_b32(), svcmpge_f32(svptrue_b32(), a, b)), 0xffffffffu));
return svreinterpret_f32_u32(svdup_n_u32_z(svnot_b_z(svptrue_b32(), svcmpge_f32(svptrue_b32(), a, b)), 0xffffffffu));
}
template <>
@ -660,6 +660,11 @@ EIGEN_STRONG_INLINE PacketXf pldexp<PacketXf>(const PacketXf& a, const PacketXf&
return pldexp_generic(a, exponent);
}
template <>
EIGEN_STRONG_INLINE PacketXf psqrt<PacketXf>(const PacketXf& a) {
return svsqrt_f32_x(svptrue_b32(), a);
}
} // namespace internal
} // namespace Eigen