GitHub-Proxy/eigen
1
0
Fork 0
mirror of https://gitlab.com/libeigen/eigen.git synced 2025-04-29 23:34:12 +08:00
Code Issues Packages Projects Releases Wiki Activity

Fix conversion of Eigen::half to _Float16 in AVX512 code

Browse Source
This commit is contained in:
Alexander Grund 2024-08-03 03:49:51 +00:00 committed by Antonio Sánchez
parent ec18dd09c8
commit 8025683226
1 changed files with 10 additions and 6 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

16
Eigen/src/Core/arch/AVX512/PacketMathFP16.h
View File

@ -109,7 +109,13 @@ struct unpacket_traits<Packet8h> {
template <>
EIGEN_STRONG_INLINE Packet32h pset1<Packet32h>(const Eigen::half& from) {
return _mm512_set1_ph(static_cast<_Float16>(from));
// half/half_raw is bit compatible
return _mm512_set1_ph(numext::bit_cast<_Float16>(from));
}
template <>
EIGEN_STRONG_INLINE Packet32h pzero(const Packet32h& /*a*/) {
return _mm512_setzero_ph();
}
// pset1frombits
@ -209,10 +215,8 @@ EIGEN_STRONG_INLINE Packet32h pmax<Packet32h>(const Packet32h& a, const Packet32
// plset
template <>
EIGEN_STRONG_INLINE Packet32h plset<Packet32h>(const half& a) {
return _mm512_add_ph(_mm512_set1_ph(a),
_mm512_set_ph(31.0f, 30.0f, 29.0f, 28.0f, 27.0f, 26.0f, 25.0f, 24.0f, 23.0f, 22.0f, 21.0f, 20.0f,
19.0f, 18.0f, 17.0f, 16.0f, 15.0f, 14.0f, 13.0f, 12.0f, 11.0f, 10.0f, 9.0f, 8.0f,
7.0f, 6.0f, 5.0f, 4.0f, 3.0f, 2.0f, 1.0f, 0.0f));
return _mm512_add_ph(pset1<Packet32h>(a), _mm512_set_ph(31, 30, 29, 28, 27, 26, 25, 24, 23, 22, 21, 20, 19, 18, 17,
16, 15, 14, 13, 12, 11, 10, 9, 8, 7, 6, 5, 4, 3, 2, 1, 0));
}
// por
@ -510,7 +514,7 @@ EIGEN_STRONG_INLINE Packet8h pnmsub(const Packet8h& a, const Packet8h& b, const
template <>
EIGEN_STRONG_INLINE Packet32h pnegate<Packet32h>(const Packet32h& a) {
return _mm512_sub_ph(_mm512_set1_ph(0.0), a);
return psub(pzero(a), a);
}
// pconj