GitHub-Proxy/eigen
1
0
Fork 0
mirror of https://gitlab.com/libeigen/eigen.git synced 2025-08-12 11:49:02 +08:00
Code Issues Packages Projects Releases Wiki Activity

_mm_hadd_epi32 is for SSSE3 only (and not SSE3)

Browse Source
This commit is contained in:
Gael Guennebaud 2015-10-07 15:48:35 +02:00
parent aba1eda71e
commit f047ecc36a
1 changed files with 14 additions and 12 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

26
Eigen/src/Core/arch/SSE/PacketMath.h
View File

@ -532,10 +532,6 @@ template<> EIGEN_STRONG_INLINE Packet2d preduxp<Packet2d>(const Packet2d* vecs)
{ {
return _mm_hadd_pd(vecs[0], vecs[1]); return _mm_hadd_pd(vecs[0], vecs[1]);
} }
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 float predux<Packet4f>(const Packet4f& a) template<> EIGEN_STRONG_INLINE float predux<Packet4f>(const Packet4f& a)
{ {
@ -544,12 +540,6 @@ template<> EIGEN_STRONG_INLINE float predux<Packet4f>(const Packet4f& a)
} }
template<> EIGEN_STRONG_INLINE double predux<Packet2d>(const Packet2d& a) { return pfirst<Packet2d>(_mm_hadd_pd(a, a)); } template<> EIGEN_STRONG_INLINE double predux<Packet2d>(const Packet2d& a) { return pfirst<Packet2d>(_mm_hadd_pd(a, a)); }
template<> EIGEN_STRONG_INLINE int predux<Packet4i>(const Packet4i& a)
{
Packet4i tmp0 = _mm_hadd_epi32(a,a);
return pfirst(_mm_hadd_epi32(tmp0,tmp0));
}
#else #else
// SSE2 versions // SSE2 versions
template<> EIGEN_STRONG_INLINE float predux<Packet4f>(const Packet4f& a) template<> EIGEN_STRONG_INLINE float predux<Packet4f>(const Packet4f& a)
@ -580,7 +570,20 @@ 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])); return _mm_add_pd(_mm_unpacklo_pd(vecs[0], vecs[1]), _mm_unpackhi_pd(vecs[0], vecs[1]));
} }
#endif // SSE3
#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);
return pfirst(_mm_hadd_epi32(tmp0,tmp0));
}
#else
template<> EIGEN_STRONG_INLINE int predux<Packet4i>(const Packet4i& a) template<> EIGEN_STRONG_INLINE int predux<Packet4i>(const Packet4i& a)
{ {
Packet4i tmp = _mm_add_epi32(a, _mm_unpackhi_epi64(a,a)); Packet4i tmp = _mm_add_epi32(a, _mm_unpackhi_epi64(a,a));
@ -600,8 +603,7 @@ template<> EIGEN_STRONG_INLINE Packet4i preduxp<Packet4i>(const Packet4i* vecs)
tmp0 = _mm_unpackhi_epi64(tmp0, tmp1); tmp0 = _mm_unpackhi_epi64(tmp0, tmp1);
return _mm_add_epi32(tmp0, tmp2); return _mm_add_epi32(tmp0, tmp2);
} }
#endif // SSE3 #endif
// Other reduction functions: // Other reduction functions:
// mul // mul