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

Add a pinsertlast function replacing the last entry of a packet by a scalar.

Browse Source 
(useful to vectorize LinSpaced)
This commit is contained in:
Gael Guennebaud
2016-10-25 16:48:49 +02:00
parent 2634f9386c
commit 13fc18d3a2
4 changed files with 53 additions and 0 deletions
Download Patch File Download Diff File Expand all files Collapse all files

20
Eigen/src/Core/arch/SSE/PacketMath.h
View File

@@ -818,6 +818,26 @@ template<> EIGEN_STRONG_INLINE Packet2d pblend(const Selector<2>& ifPacket, cons
#endif
}
template<> EIGEN_DEVICE_FUNC Packet4f pinsertlast(const Packet4f& a, float b)
{
#ifdef EIGEN_VECTORIZE_SSE4_1
return _mm_blend_ps(a,pset1<Packet4f>(b),(1<<3));
#else
const Packet4f mask = _mm_castsi128_ps(_mm_setr_epi32(0x0,0x0,0x0,0xFFFFFFFF));
return _mm_or_ps(_mm_andnot_pd(mask, a), _mm_and_pd(mask, pset1<Packet4f>(b)));
#endif
}
template<> EIGEN_DEVICE_FUNC Packet2d pinsertlast(const Packet2d& a, double b)
{
#ifdef EIGEN_VECTORIZE_SSE4_1
return _mm_blend_pd(a,pset1<Packet2d>(b),(1<<1));
#else
const Packet2d mask = _mm_castsi128_pd(_mm_setr_epi32(0x0,0x0,0xFFFFFFFF,0xFFFFFFFF));
return _mm_or_pd(_mm_andnot_pd(mask, a), _mm_and_pd(mask, pset1<Packet2d>(b)));
#endif
}
// Scalar path for pmadd with FMA to ensure consistency with vectorized path.
#ifdef __FMA__
template<> EIGEN_STRONG_INLINE float pmadd(const float& a, const float& b, const float& c) {