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

Remove unused packet op "preduxp".

Browse Source
This commit is contained in:
Rasmus Munk Larsen
2020-04-23 18:17:14 +00:00
parent 0aebe19aca
commit e80ec24357
16 changed files with 1 additions and 1009 deletions
Download Patch File Download Diff File Expand all files Collapse all files

53
Eigen/src/Core/arch/SSE/PacketMath.h
View File

@@ -179,8 +179,7 @@ template<> struct packet_traits<bool> : default_packet_traits
HasAbs2 = 0,
HasMin = 0,
HasMax = 0,
HasConj = 0,
HasReduxp = 0
HasConj = 0
};
};
@@ -733,38 +732,6 @@ EIGEN_STRONG_INLINE void punpackp(Packet4f* vecs)
vecs[0] = _mm_castsi128_ps(_mm_shuffle_epi32(_mm_castps_si128(vecs[0]), 0x00));
}
#ifdef EIGEN_VECTORIZE_SSE3
template<> EIGEN_STRONG_INLINE Packet4f preduxp<Packet4f>(const Packet4f* vecs)
{
return _mm_hadd_ps(_mm_hadd_ps(vecs[0], vecs[1]),_mm_hadd_ps(vecs[2], vecs[3]));
}
template<> EIGEN_STRONG_INLINE Packet2d preduxp<Packet2d>(const Packet2d* vecs)
{
return _mm_hadd_pd(vecs[0], vecs[1]);
}
#else
template<> EIGEN_STRONG_INLINE Packet4f preduxp<Packet4f>(const Packet4f* vecs)
{
Packet4f tmp0, tmp1, tmp2;
tmp0 = _mm_unpacklo_ps(vecs[0], vecs[1]);
tmp1 = _mm_unpackhi_ps(vecs[0], vecs[1]);
tmp2 = _mm_unpackhi_ps(vecs[2], vecs[3]);
tmp0 = _mm_add_ps(tmp0, tmp1);
tmp1 = _mm_unpacklo_ps(vecs[2], vecs[3]);
tmp1 = _mm_add_ps(tmp1, tmp2);
tmp2 = _mm_movehl_ps(tmp1, tmp0);
tmp0 = _mm_movelh_ps(tmp0, tmp1);
return _mm_add_ps(tmp0, tmp2);
}
template<> EIGEN_STRONG_INLINE Packet2d preduxp<Packet2d>(const Packet2d* vecs)
{
return _mm_add_pd(_mm_unpacklo_pd(vecs[0], vecs[1]), _mm_unpackhi_pd(vecs[0], vecs[1]));
}
#endif // SSE3
template<> EIGEN_STRONG_INLINE float predux<Packet4f>(const Packet4f& a)
{
// Disable SSE3 _mm_hadd_pd that is extremely slow on all existing Intel's architectures
@@ -790,10 +757,6 @@ template<> EIGEN_STRONG_INLINE double predux<Packet2d>(const Packet2d& a)
}
#ifdef EIGEN_VECTORIZE_SSSE3
template<> EIGEN_STRONG_INLINE Packet4i preduxp<Packet4i>(const Packet4i* vecs)
{
return _mm_hadd_epi32(_mm_hadd_epi32(vecs[0], vecs[1]),_mm_hadd_epi32(vecs[2], vecs[3]));
}
template<> EIGEN_STRONG_INLINE int predux<Packet4i>(const Packet4i& a)
{
Packet4i tmp0 = _mm_hadd_epi32(a,a);
@@ -805,20 +768,6 @@ template<> EIGEN_STRONG_INLINE int predux<Packet4i>(const Packet4i& a)
Packet4i tmp = _mm_add_epi32(a, _mm_unpackhi_epi64(a,a));
return pfirst(tmp) + pfirst<Packet4i>(_mm_shuffle_epi32(tmp, 1));
}
template<> EIGEN_STRONG_INLINE Packet4i preduxp<Packet4i>(const Packet4i* vecs)
{
Packet4i tmp0, tmp1, tmp2;
tmp0 = _mm_unpacklo_epi32(vecs[0], vecs[1]);
tmp1 = _mm_unpackhi_epi32(vecs[0], vecs[1]);
tmp2 = _mm_unpackhi_epi32(vecs[2], vecs[3]);
tmp0 = _mm_add_epi32(tmp0, tmp1);
tmp1 = _mm_unpacklo_epi32(vecs[2], vecs[3]);
tmp1 = _mm_add_epi32(tmp1, tmp2);
tmp2 = _mm_unpacklo_epi64(tmp0, tmp1);
tmp0 = _mm_unpackhi_epi64(tmp0, tmp1);
return _mm_add_epi32(tmp0, tmp2);
}
#endif
// Other reduction functions: