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 pinsertfirst function and implement pinsertlast for complex on SSE/AVX.

Browse Source
This commit is contained in:
Gael Guennebaud
2016-11-02 10:38:13 +01:00
parent e44519744e
commit 598de8b193
6 changed files with 100 additions and 2 deletions
Download Patch File Download Diff File Expand all files Collapse all files

20
Eigen/src/Core/arch/AVX/Complex.h
View File

@@ -456,6 +456,26 @@ ptranspose(PacketBlock<Packet2cd,2>& kernel) {
kernel.packet[0].v = tmp;
}
template<> EIGEN_STRONG_INLINE Packet4cf pinsertfirst(const Packet4cf& a, std::complex<float> b)
{
return Packet4cf(_mm256_blend_ps(a.v,pset1<Packet4cf>(b).v,1|2));
}
template<> EIGEN_STRONG_INLINE Packet2cd pinsertfirst(const Packet2cd& a, std::complex<double> b)
{
return Packet2cd(_mm256_blend_pd(a.v,pset1<Packet2cd>(b).v,1|2));
}
template<> EIGEN_STRONG_INLINE Packet4cf pinsertlast(const Packet4cf& a, std::complex<float> b)
{
return Packet4cf(_mm256_blend_ps(a.v,pset1<Packet4cf>(b).v,(1<<7)|(1<<6)));
}
template<> EIGEN_STRONG_INLINE Packet2cd pinsertlast(const Packet2cd& a, std::complex<double> b)
{
return Packet2cd(_mm256_blend_pd(a.v,pset1<Packet2cd>(b).v,(1<<3)|(1<<2)));
}
} // end namespace internal
} // end namespace Eigen

10
Eigen/src/Core/arch/AVX/PacketMath.h
View File

@@ -604,6 +604,16 @@ template<> EIGEN_STRONG_INLINE Packet4d pblend(const Selector<4>& ifPacket, cons
return _mm256_blendv_pd(thenPacket, elsePacket, false_mask);
}
template<> EIGEN_STRONG_INLINE Packet8f pinsertfirst(const Packet8f& a, float b)
{
return _mm256_blend_ps(a,pset1<Packet8f>(b),1);
}
template<> EIGEN_STRONG_INLINE Packet4d pinsertfirst(const Packet4d& a, double b)
{
return _mm256_blend_pd(a,pset1<Packet4d>(b),1);
}
template<> EIGEN_STRONG_INLINE Packet8f pinsertlast(const Packet8f& a, float b)
{
return _mm256_blend_ps(a,pset1<Packet8f>(b),(1<<7));

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

@@ -476,6 +476,26 @@ template<> EIGEN_STRONG_INLINE Packet2cf pblend(const Selector<2>& ifPacket, co
return Packet2cf(_mm_castpd_ps(result));
}
template<> EIGEN_STRONG_INLINE Packet2cf pinsertfirst(const Packet2cf& a, std::complex<float> b)
{
return Packet2cf(_mm_loadl_pi(a.v, reinterpret_cast<const __m64*>(&b)));
}
template<> EIGEN_STRONG_INLINE Packet1cd pinsertfirst(const Packet1cd&, std::complex<double> b)
{
return pset1<Packet1cd>(b);
}
template<> EIGEN_STRONG_INLINE Packet2cf pinsertlast(const Packet2cf& a, std::complex<float> b)
{
return Packet2cf(_mm_loadh_pi(a.v, reinterpret_cast<const __m64*>(&b)));
}
template<> EIGEN_STRONG_INLINE Packet1cd pinsertlast(const Packet1cd&, std::complex<double> b)
{
return pset1<Packet1cd>(b);
}
} // end namespace internal
} // end namespace Eigen

18
Eigen/src/Core/arch/SSE/PacketMath.h
View File

@@ -818,6 +818,24 @@ template<> EIGEN_STRONG_INLINE Packet2d pblend(const Selector<2>& ifPacket, cons
#endif
}
template<> EIGEN_STRONG_INLINE Packet4f pinsertfirst(const Packet4f& a, float b)
{
#ifdef EIGEN_VECTORIZE_SSE4_1
return _mm_blend_ps(a,pset1<Packet4f>(b),1);
#else
return _mm_move_ss(a, _mm_load_ss(&b));
#endif
}
template<> EIGEN_STRONG_INLINE Packet2d pinsertfirst(const Packet2d& a, double b)
{
#ifdef EIGEN_VECTORIZE_SSE4_1
return _mm_blend_pd(a,pset1<Packet2d>(b),1);
#else
return _mm_move_sd(a, _mm_load_sd(&b));
#endif
}
template<> EIGEN_STRONG_INLINE Packet4f pinsertlast(const Packet4f& a, float b)
{
#ifdef EIGEN_VECTORIZE_SSE4_1