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 intitial support for the vectorization of complex<float>

Browse Source
This commit is contained in:
Gael Guennebaud
2010-07-05 16:18:09 +02:00
parent efb79600b9
commit e1eccfad3f
6 changed files with 195 additions and 30 deletions
Download Patch File Download Diff File Expand all files Collapse all files

144
Eigen/src/Core/arch/SSE/Complex.h Normal file
View File

@@ -0,0 +1,144 @@
// This file is part of Eigen, a lightweight C++ template library
// for linear algebra.
//
// Copyright (C) 2010 Gael Guennebaud <gael.guennebaud@inria.fr>
//
// Eigen is free software; you can redistribute it and/or
// modify it under the terms of the GNU Lesser General Public
// License as published by the Free Software Foundation; either
// version 3 of the License, or (at your option) any later version.
//
// Alternatively, you can redistribute it and/or
// modify it under the terms of the GNU General Public License as
// published by the Free Software Foundation; either version 2 of
// the License, or (at your option) any later version.
//
// Eigen is distributed in the hope that it will be useful, but WITHOUT ANY
// WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS
// FOR A PARTICULAR PURPOSE. See the GNU Lesser General Public License or the
// GNU General Public License for more details.
//
// You should have received a copy of the GNU Lesser General Public
// License and a copy of the GNU General Public License along with
// Eigen. If not, see <http://www.gnu.org/licenses/>.
#ifndef EIGEN_COMPLEX_SSE_H
#define EIGEN_COMPLEX_SSE_H
struct Packet2cf
{
EIGEN_STRONG_INLINE Packet2cf() {}
EIGEN_STRONG_INLINE explicit Packet2cf(const __m128& a) : v(a) {}
__m128 v;
};
typedef __m128d Packet1cd;
template<> struct ei_packet_traits<std::complex<float> > : ei_default_packet_traits
{
typedef Packet2cf type; enum {size=2};
};
template<> struct ei_unpacket_traits<Packet2cf> { typedef std::complex<float> type; enum {size=2}; };
template<> EIGEN_STRONG_INLINE Packet2cf ei_pset1<std::complex<float> >(const std::complex<float>& from)
{
Packet2cf res;
res.v = _mm_loadl_pi(res.v, (const __m64*)&from);
return Packet2cf(_mm_movelh_ps(res.v,res.v));
}
// template<> EIGEN_STRONG_INLINE Packet4f ei_plset<std::complex<float> >(const std::complex<float> & a) { }
template<> EIGEN_STRONG_INLINE Packet2cf ei_padd<Packet2cf>(const Packet2cf& a, const Packet2cf& b) { return Packet2cf(_mm_add_ps(a.v,b.v)); }
template<> EIGEN_STRONG_INLINE Packet2cf ei_psub<Packet2cf>(const Packet2cf& a, const Packet2cf& b) { return Packet2cf(_mm_sub_ps(a.v,b.v)); }
template<> EIGEN_STRONG_INLINE Packet2cf ei_pnegate(const Packet2cf& a)
{
const __m128 mask = _mm_castsi128_ps(_mm_setr_epi32(0x80000000,0x80000000,0x80000000,0x80000000));
return Packet2cf(_mm_xor_ps(a.v,mask));
}
template<> EIGEN_STRONG_INLINE Packet2cf ei_pmul<Packet2cf>(const Packet2cf& a, const Packet2cf& b)
{
// TODO optimize it for SSE3 and 4
const __m128 mask = _mm_castsi128_ps(_mm_setr_epi32(0x80000000,0x00000000,0x80000000,0x00000000));
return Packet2cf(_mm_add_ps(_mm_mul_ps(_mm_castsi128_ps(_mm_shuffle_epi32(_mm_castps_si128(a.v), 0xa0)), b.v),
_mm_xor_ps(_mm_mul_ps(_mm_castsi128_ps(_mm_shuffle_epi32(_mm_castps_si128(a.v), 0xf5)),
_mm_castsi128_ps(_mm_shuffle_epi32(_mm_castps_si128(b.v), 0xb1 ))), mask)));
}
// template<> EIGEN_STRONG_INLINE Packet2cf ei_pmadd<Packet2cf>(const Packet2cf& a, const Packet2cf& b, const Packet2cf& c)
// {std::cerr << __LINE__ << "\n";
// // TODO optimize it for SSE3 and 4
// const __m128 mask = _mm_castsi128_ps(_mm_setr_epi32(0x80000000,0x00000000,0x80000000,0x00000000));
// return Packet2cf(_mm_add_ps(c.v,
// _mm_add_ps(_mm_mul_ps(_mm_castsi128_ps(_mm_shuffle_epi32(_mm_castps_si128(a.v), 0xa0)), b.v),
// _mm_xor_ps(_mm_mul_ps(_mm_castsi128_ps(_mm_shuffle_epi32(_mm_castps_si128(a.v), 0xf5)),
// _mm_castsi128_ps(_mm_shuffle_epi32(_mm_castps_si128(b.v), 0xb1 ))), mask))));
// }
template<> EIGEN_STRONG_INLINE Packet2cf ei_pdiv<Packet2cf>(const Packet2cf& a, const Packet2cf& b)
{
// TODO optimize it for SSE3 and 4
const __m128 mask = _mm_castsi128_ps(_mm_setr_epi32(0x00000000,0x80000000,0x00000000,0x80000000));
Packet2cf res(_mm_add_ps(_mm_mul_ps(a.v, _mm_castsi128_ps(_mm_shuffle_epi32(_mm_castps_si128(b.v), 0xa0))),
_mm_xor_ps(_mm_mul_ps(_mm_castsi128_ps(_mm_shuffle_epi32(_mm_castps_si128(a.v), 0xb1)),
_mm_castsi128_ps(_mm_shuffle_epi32(_mm_castps_si128(b.v), 0xf5 ))), mask)));
__m128 s = _mm_mul_ps(b.v,b.v);
return Packet2cf(_mm_div_ps(res.v,_mm_add_ps(s,_mm_castsi128_ps(_mm_shuffle_epi32(_mm_castps_si128(s), 0xb1)))));
}
template<> EIGEN_STRONG_INLINE Packet2cf ei_pand <Packet2cf>(const Packet2cf& a, const Packet2cf& b) { return Packet2cf(_mm_and_ps(a.v,b.v)); }
template<> EIGEN_STRONG_INLINE Packet2cf ei_por <Packet2cf>(const Packet2cf& a, const Packet2cf& b) { return Packet2cf(_mm_or_ps(a.v,b.v)); }
template<> EIGEN_STRONG_INLINE Packet2cf ei_pxor <Packet2cf>(const Packet2cf& a, const Packet2cf& b) { return Packet2cf(_mm_xor_ps(a.v,b.v)); }
template<> EIGEN_STRONG_INLINE Packet2cf ei_pandnot<Packet2cf>(const Packet2cf& a, const Packet2cf& b) { return Packet2cf(_mm_andnot_ps(a.v,b.v)); }
template<> EIGEN_STRONG_INLINE Packet2cf ei_pload <std::complex<float> >(const std::complex<float>* from) { EIGEN_DEBUG_ALIGNED_LOAD return Packet2cf(_mm_load_ps((const float*)from)); }
template<> EIGEN_STRONG_INLINE Packet2cf ei_ploadu<std::complex<float> >(const std::complex<float>* from) { EIGEN_DEBUG_ALIGNED_LOAD return Packet2cf(ei_ploadu((const float*)from)); }
template<> EIGEN_STRONG_INLINE void ei_pstore <std::complex<float> >(std::complex<float> * to, const Packet2cf& from) { EIGEN_DEBUG_ALIGNED_STORE _mm_store_ps((float*)to, from.v); }
template<> EIGEN_STRONG_INLINE void ei_pstoreu<std::complex<float> >(std::complex<float> * to, const Packet2cf& from) { EIGEN_DEBUG_ALIGNED_STORE ei_pstoreu((float*)to, from.v); }
template<> EIGEN_STRONG_INLINE void ei_prefetch<std::complex<float> >(const std::complex<float> * addr) { _mm_prefetch((const char*)(addr), _MM_HINT_T0); }
template<> EIGEN_STRONG_INLINE std::complex<float> ei_pfirst<Packet2cf>(const Packet2cf& a)
{
std::complex<float> res;
_mm_storel_pi((__m64*)&res, a.v);
return res;
}
template<> EIGEN_STRONG_INLINE Packet2cf ei_preverse(const Packet2cf& a) { return Packet2cf(_mm_castpd_ps(ei_preverse(_mm_castps_pd(a.v)))); }
// template<> EIGEN_STRONG_INLINE Packet2cf ei_pabs(const Packet2cf& a) {}
template<> EIGEN_STRONG_INLINE std::complex<float> ei_predux<Packet2cf>(const Packet2cf& a)
{
return ei_pfirst(Packet2cf(_mm_add_ps(a.v, _mm_movehl_ps(a.v,a.v))));
}
template<> EIGEN_STRONG_INLINE Packet2cf ei_preduxp<Packet2cf>(const Packet2cf* vecs)
{
return Packet2cf(_mm_add_ps(_mm_movelh_ps(vecs[0].v,vecs[1].v), _mm_movehl_ps(vecs[1].v,vecs[0].v)));
}
template<> EIGEN_STRONG_INLINE std::complex<float> ei_predux_mul<Packet2cf>(const Packet2cf& a)
{
return ei_pfirst(ei_pmul(a, Packet2cf(_mm_movehl_ps(a.v,a.v))));
}
template<int Offset>
struct ei_palign_impl<Offset,Packet2cf>
{
EIGEN_STRONG_INLINE static void run(Packet2cf& first, const Packet2cf& second)
{
if (Offset==1)
{
first.v = _mm_movehl_ps(first.v, first.v);
first.v = _mm_movelh_ps(first.v, second.v);
}
}
};
#endif // EIGEN_COMPLEX_SSE_H

10
Eigen/src/Core/products/GeneralBlockPanelKernel.h
View File

@@ -259,7 +259,7 @@ struct ei_gebp_kernel
#ifndef EIGEN_HAS_FUSE_CJMADD
PacketType T0;
#endif
EIGEN_ASM_COMMENT("mybegin");
A0 = ei_pload(&blA[0*PacketSize]);
A1 = ei_pload(&blA[1*PacketSize]);
B0 = ei_pload(&blB[0*PacketSize]);
@@ -295,6 +295,7 @@ struct ei_gebp_kernel
B0 = ei_pload(&blB[7*PacketSize]);
CJMADD(A0,B0,C1,T0);
CJMADD(A1,B0,C5,B0);
EIGEN_ASM_COMMENT("myend");
}
else
{
@@ -302,7 +303,7 @@ struct ei_gebp_kernel
#ifndef EIGEN_HAS_FUSE_CJMADD
PacketType T0;
#endif
EIGEN_ASM_COMMENT("mybegin");
A0 = ei_pload(&blA[0*PacketSize]);
A1 = ei_pload(&blA[1*PacketSize]);
B0 = ei_pload(&blB[0*PacketSize]);
@@ -361,6 +362,7 @@ struct ei_gebp_kernel
CJMADD(A1,B2,C6,B2);
CJMADD(A0,B3,C3,T0);
CJMADD(A1,B3,C7,B3);
EIGEN_ASM_COMMENT("myend");
}
blB += 4*nr*PacketSize;
@@ -683,7 +685,9 @@ struct ei_gebp_kernel
const Scalar* blB = unpackedB;
for(Index k=0; k<depth; k++)
{
C0 = cj.pmadd(ei_pload(blA), ei_pload(blB), C0);
PacketType T0;
CJMADD(ei_pload(blA), ei_pload(blB), C0, T0);
//C0 = cj.pmadd(ei_pload(blA), ei_pload(blB), C0);
blB += PacketSize;
blA += PacketSize;
}

12
Eigen/src/Core/util/BlasUtil.h
View File

@@ -140,6 +140,18 @@ struct ei_product_blocking_traits
};
};
template<typename Real>
struct ei_product_blocking_traits<std::complex<Real> >
{
typedef std::complex<Real> Scalar;
typedef typename ei_packet_traits<Scalar>::type PacketType;
enum {
PacketSize = sizeof(PacketType)/sizeof(Scalar),
nr = 2,
mr = 2 * PacketSize
};
};
/* Helper class to analyze the factors of a Product expression.
* In particular it allows to pop out operator-, scalar multiples,
* and conjugate */