GitHub-Proxy/eigen
1
0
Fork 0
mirror of https://gitlab.com/libeigen/eigen.git synced 2025-07-10 23:21:47 +08:00
Code Issues Packages Projects Releases Wiki Activity

Disable usage of SSE3 _mm_hadd_ps that is extremely slow.

Browse Source 
(grafted from 178c084856003f1cfd3020615ab98230d9520a80
)
This commit is contained in:
Gael Guennebaud 2016-11-22 21:53:14 +01:00
parent 1b7dd46d94
commit 0eff51e2ed
1 changed files with 13 additions and 13 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

@ -510,20 +510,7 @@ 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 float predux<Packet4f>(const Packet4f& a)
{
Packet4f tmp0 = _mm_hadd_ps(a,a);
return pfirst<Packet4f>(_mm_hadd_ps(tmp0, tmp0));
}
#else #else
// SSE2 versions
template<> EIGEN_STRONG_INLINE float predux<Packet4f>(const Packet4f& a)
{
Packet4f tmp = _mm_add_ps(a, _mm_movehl_ps(a,a));
return pfirst<Packet4f>(_mm_add_ss(tmp, _mm_shuffle_ps(tmp,tmp, 1)));
}
template<> EIGEN_STRONG_INLINE Packet4f preduxp<Packet4f>(const Packet4f* vecs) template<> EIGEN_STRONG_INLINE Packet4f preduxp<Packet4f>(const Packet4f* vecs)
{ {
Packet4f tmp0, tmp1, tmp2; Packet4f tmp0, tmp1, tmp2;
@ -544,6 +531,19 @@ template<> EIGEN_STRONG_INLINE Packet2d preduxp<Packet2d>(const Packet2d* vecs)
} }
#endif // SSE3 #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
// (from Nehalem to Haswell)
// #ifdef EIGEN_VECTORIZE_SSE3
// Packet4f tmp = _mm_add_ps(a, vec4f_swizzle1(a,2,3,2,3));
// return pfirst<Packet4f>(_mm_hadd_ps(tmp, tmp));
// #else
Packet4f tmp = _mm_add_ps(a, _mm_movehl_ps(a,a));
return pfirst<Packet4f>(_mm_add_ss(tmp, _mm_shuffle_ps(tmp,tmp, 1)));
// #endif
}
template<> EIGEN_STRONG_INLINE double predux<Packet2d>(const Packet2d& a) template<> EIGEN_STRONG_INLINE double predux<Packet2d>(const Packet2d& a)
{ {
// Disable SSE3 _mm_hadd_pd that is extremely slow on all existing Intel's architectures // Disable SSE3 _mm_hadd_pd that is extremely slow on all existing Intel's architectures