$treeview $search $mathjax
Eigen
3.2.5
$projectbrief
|
$projectbrief
|
$searchbox |
00001 // This file is part of Eigen, a lightweight C++ template library 00002 // for linear algebra. 00003 // 00004 // Copyright (C) 2010 Gael Guennebaud <gael.guennebaud@inria.fr> 00005 // 00006 // This Source Code Form is subject to the terms of the Mozilla 00007 // Public License v. 2.0. If a copy of the MPL was not distributed 00008 // with this file, You can obtain one at http://mozilla.org/MPL/2.0/. 00009 00010 #ifndef EIGEN_COMPLEX_SSE_H 00011 #define EIGEN_COMPLEX_SSE_H 00012 00013 namespace Eigen { 00014 00015 namespace internal { 00016 00017 //---------- float ---------- 00018 struct Packet2cf 00019 { 00020 EIGEN_STRONG_INLINE Packet2cf() {} 00021 EIGEN_STRONG_INLINE explicit Packet2cf(const __m128& a) : v(a) {} 00022 __m128 v; 00023 }; 00024 00025 template<> struct packet_traits<std::complex<float> > : default_packet_traits 00026 { 00027 typedef Packet2cf type; 00028 enum { 00029 Vectorizable = 1, 00030 AlignedOnScalar = 1, 00031 size = 2, 00032 00033 HasAdd = 1, 00034 HasSub = 1, 00035 HasMul = 1, 00036 HasDiv = 1, 00037 HasNegate = 1, 00038 HasAbs = 0, 00039 HasAbs2 = 0, 00040 HasMin = 0, 00041 HasMax = 0, 00042 HasSetLinear = 0 00043 }; 00044 }; 00045 00046 template<> struct unpacket_traits<Packet2cf> { typedef std::complex<float> type; enum {size=2}; }; 00047 00048 template<> EIGEN_STRONG_INLINE Packet2cf padd<Packet2cf>(const Packet2cf& a, const Packet2cf& b) { return Packet2cf(_mm_add_ps(a.v,b.v)); } 00049 template<> EIGEN_STRONG_INLINE Packet2cf psub<Packet2cf>(const Packet2cf& a, const Packet2cf& b) { return Packet2cf(_mm_sub_ps(a.v,b.v)); } 00050 template<> EIGEN_STRONG_INLINE Packet2cf pnegate(const Packet2cf& a) 00051 { 00052 const __m128 mask = _mm_castsi128_ps(_mm_setr_epi32(0x80000000,0x80000000,0x80000000,0x80000000)); 00053 return Packet2cf(_mm_xor_ps(a.v,mask)); 00054 } 00055 template<> EIGEN_STRONG_INLINE Packet2cf pconj(const Packet2cf& a) 00056 { 00057 const __m128 mask = _mm_castsi128_ps(_mm_setr_epi32(0x00000000,0x80000000,0x00000000,0x80000000)); 00058 return Packet2cf(_mm_xor_ps(a.v,mask)); 00059 } 00060 00061 template<> EIGEN_STRONG_INLINE Packet2cf pmul<Packet2cf>(const Packet2cf& a, const Packet2cf& b) 00062 { 00063 // TODO optimize it for SSE3 and 4 00064 #ifdef EIGEN_VECTORIZE_SSE3 00065 return Packet2cf(_mm_addsub_ps(_mm_mul_ps(_mm_moveldup_ps(a.v), b.v), 00066 _mm_mul_ps(_mm_movehdup_ps(a.v), 00067 vec4f_swizzle1(b.v, 1, 0, 3, 2)))); 00068 // return Packet2cf(_mm_addsub_ps(_mm_mul_ps(vec4f_swizzle1(a.v, 0, 0, 2, 2), b.v), 00069 // _mm_mul_ps(vec4f_swizzle1(a.v, 1, 1, 3, 3), 00070 // vec4f_swizzle1(b.v, 1, 0, 3, 2)))); 00071 #else 00072 const __m128 mask = _mm_castsi128_ps(_mm_setr_epi32(0x80000000,0x00000000,0x80000000,0x00000000)); 00073 return Packet2cf(_mm_add_ps(_mm_mul_ps(vec4f_swizzle1(a.v, 0, 0, 2, 2), b.v), 00074 _mm_xor_ps(_mm_mul_ps(vec4f_swizzle1(a.v, 1, 1, 3, 3), 00075 vec4f_swizzle1(b.v, 1, 0, 3, 2)), mask))); 00076 #endif 00077 } 00078 00079 template<> EIGEN_STRONG_INLINE Packet2cf pand <Packet2cf>(const Packet2cf& a, const Packet2cf& b) { return Packet2cf(_mm_and_ps(a.v,b.v)); } 00080 template<> EIGEN_STRONG_INLINE Packet2cf por <Packet2cf>(const Packet2cf& a, const Packet2cf& b) { return Packet2cf(_mm_or_ps(a.v,b.v)); } 00081 template<> EIGEN_STRONG_INLINE Packet2cf pxor <Packet2cf>(const Packet2cf& a, const Packet2cf& b) { return Packet2cf(_mm_xor_ps(a.v,b.v)); } 00082 template<> EIGEN_STRONG_INLINE Packet2cf pandnot<Packet2cf>(const Packet2cf& a, const Packet2cf& b) { return Packet2cf(_mm_andnot_ps(a.v,b.v)); } 00083 00084 template<> EIGEN_STRONG_INLINE Packet2cf pload <Packet2cf>(const std::complex<float>* from) { EIGEN_DEBUG_ALIGNED_LOAD return Packet2cf(pload<Packet4f>(&numext::real_ref(*from))); } 00085 template<> EIGEN_STRONG_INLINE Packet2cf ploadu<Packet2cf>(const std::complex<float>* from) { EIGEN_DEBUG_UNALIGNED_LOAD return Packet2cf(ploadu<Packet4f>(&numext::real_ref(*from))); } 00086 00087 template<> EIGEN_STRONG_INLINE Packet2cf pset1<Packet2cf>(const std::complex<float>& from) 00088 { 00089 Packet2cf res; 00090 #if EIGEN_GNUC_AT_MOST(4,2) 00091 // Workaround annoying "may be used uninitialized in this function" warning with gcc 4.2 00092 res.v = _mm_loadl_pi(_mm_set1_ps(0.0f), reinterpret_cast<const __m64*>(&from)); 00093 #elif EIGEN_GNUC_AT_LEAST(4,6) 00094 // Suppress annoying "may be used uninitialized in this function" warning with gcc >= 4.6 00095 #pragma GCC diagnostic push 00096 #pragma GCC diagnostic ignored "-Wuninitialized" 00097 res.v = _mm_loadl_pi(res.v, (const __m64*)&from); 00098 #pragma GCC diagnostic pop 00099 #else 00100 res.v = _mm_loadl_pi(res.v, (const __m64*)&from); 00101 #endif 00102 return Packet2cf(_mm_movelh_ps(res.v,res.v)); 00103 } 00104 00105 template<> EIGEN_STRONG_INLINE Packet2cf ploaddup<Packet2cf>(const std::complex<float>* from) { return pset1<Packet2cf>(*from); } 00106 00107 template<> EIGEN_STRONG_INLINE void pstore <std::complex<float> >(std::complex<float> * to, const Packet2cf& from) { EIGEN_DEBUG_ALIGNED_STORE pstore(&numext::real_ref(*to), from.v); } 00108 template<> EIGEN_STRONG_INLINE void pstoreu<std::complex<float> >(std::complex<float> * to, const Packet2cf& from) { EIGEN_DEBUG_UNALIGNED_STORE pstoreu(&numext::real_ref(*to), from.v); } 00109 00110 template<> EIGEN_STRONG_INLINE void prefetch<std::complex<float> >(const std::complex<float> * addr) { _mm_prefetch((const char*)(addr), _MM_HINT_T0); } 00111 00112 template<> EIGEN_STRONG_INLINE std::complex<float> pfirst<Packet2cf>(const Packet2cf& a) 00113 { 00114 #if EIGEN_GNUC_AT_MOST(4,3) 00115 // Workaround gcc 4.2 ICE - this is not performance wise ideal, but who cares... 00116 // This workaround also fix invalid code generation with gcc 4.3 00117 EIGEN_ALIGN16 std::complex<float> res[2]; 00118 _mm_store_ps((float*)res, a.v); 00119 return res[0]; 00120 #else 00121 std::complex<float> res; 00122 _mm_storel_pi((__m64*)&res, a.v); 00123 return res; 00124 #endif 00125 } 00126 00127 template<> EIGEN_STRONG_INLINE Packet2cf preverse(const Packet2cf& a) { return Packet2cf(_mm_castpd_ps(preverse(_mm_castps_pd(a.v)))); } 00128 00129 template<> EIGEN_STRONG_INLINE std::complex<float> predux<Packet2cf>(const Packet2cf& a) 00130 { 00131 return pfirst(Packet2cf(_mm_add_ps(a.v, _mm_movehl_ps(a.v,a.v)))); 00132 } 00133 00134 template<> EIGEN_STRONG_INLINE Packet2cf preduxp<Packet2cf>(const Packet2cf* vecs) 00135 { 00136 return Packet2cf(_mm_add_ps(_mm_movelh_ps(vecs[0].v,vecs[1].v), _mm_movehl_ps(vecs[1].v,vecs[0].v))); 00137 } 00138 00139 template<> EIGEN_STRONG_INLINE std::complex<float> predux_mul<Packet2cf>(const Packet2cf& a) 00140 { 00141 return pfirst(pmul(a, Packet2cf(_mm_movehl_ps(a.v,a.v)))); 00142 } 00143 00144 template<int Offset> 00145 struct palign_impl<Offset,Packet2cf> 00146 { 00147 static EIGEN_STRONG_INLINE void run(Packet2cf& first, const Packet2cf& second) 00148 { 00149 if (Offset==1) 00150 { 00151 first.v = _mm_movehl_ps(first.v, first.v); 00152 first.v = _mm_movelh_ps(first.v, second.v); 00153 } 00154 } 00155 }; 00156 00157 template<> struct conj_helper<Packet2cf, Packet2cf, false,true> 00158 { 00159 EIGEN_STRONG_INLINE Packet2cf pmadd(const Packet2cf& x, const Packet2cf& y, const Packet2cf& c) const 00160 { return padd(pmul(x,y),c); } 00161 00162 EIGEN_STRONG_INLINE Packet2cf pmul(const Packet2cf& a, const Packet2cf& b) const 00163 { 00164 #ifdef EIGEN_VECTORIZE_SSE3 00165 return internal::pmul(a, pconj(b)); 00166 #else 00167 const __m128 mask = _mm_castsi128_ps(_mm_setr_epi32(0x00000000,0x80000000,0x00000000,0x80000000)); 00168 return Packet2cf(_mm_add_ps(_mm_xor_ps(_mm_mul_ps(vec4f_swizzle1(a.v, 0, 0, 2, 2), b.v), mask), 00169 _mm_mul_ps(vec4f_swizzle1(a.v, 1, 1, 3, 3), 00170 vec4f_swizzle1(b.v, 1, 0, 3, 2)))); 00171 #endif 00172 } 00173 }; 00174 00175 template<> struct conj_helper<Packet2cf, Packet2cf, true,false> 00176 { 00177 EIGEN_STRONG_INLINE Packet2cf pmadd(const Packet2cf& x, const Packet2cf& y, const Packet2cf& c) const 00178 { return padd(pmul(x,y),c); } 00179 00180 EIGEN_STRONG_INLINE Packet2cf pmul(const Packet2cf& a, const Packet2cf& b) const 00181 { 00182 #ifdef EIGEN_VECTORIZE_SSE3 00183 return internal::pmul(pconj(a), b); 00184 #else 00185 const __m128 mask = _mm_castsi128_ps(_mm_setr_epi32(0x00000000,0x80000000,0x00000000,0x80000000)); 00186 return Packet2cf(_mm_add_ps(_mm_mul_ps(vec4f_swizzle1(a.v, 0, 0, 2, 2), b.v), 00187 _mm_xor_ps(_mm_mul_ps(vec4f_swizzle1(a.v, 1, 1, 3, 3), 00188 vec4f_swizzle1(b.v, 1, 0, 3, 2)), mask))); 00189 #endif 00190 } 00191 }; 00192 00193 template<> struct conj_helper<Packet2cf, Packet2cf, true,true> 00194 { 00195 EIGEN_STRONG_INLINE Packet2cf pmadd(const Packet2cf& x, const Packet2cf& y, const Packet2cf& c) const 00196 { return padd(pmul(x,y),c); } 00197 00198 EIGEN_STRONG_INLINE Packet2cf pmul(const Packet2cf& a, const Packet2cf& b) const 00199 { 00200 #ifdef EIGEN_VECTORIZE_SSE3 00201 return pconj(internal::pmul(a, b)); 00202 #else 00203 const __m128 mask = _mm_castsi128_ps(_mm_setr_epi32(0x00000000,0x80000000,0x00000000,0x80000000)); 00204 return Packet2cf(_mm_sub_ps(_mm_xor_ps(_mm_mul_ps(vec4f_swizzle1(a.v, 0, 0, 2, 2), b.v), mask), 00205 _mm_mul_ps(vec4f_swizzle1(a.v, 1, 1, 3, 3), 00206 vec4f_swizzle1(b.v, 1, 0, 3, 2)))); 00207 #endif 00208 } 00209 }; 00210 00211 template<> struct conj_helper<Packet4f, Packet2cf, false,false> 00212 { 00213 EIGEN_STRONG_INLINE Packet2cf pmadd(const Packet4f& x, const Packet2cf& y, const Packet2cf& c) const 00214 { return padd(c, pmul(x,y)); } 00215 00216 EIGEN_STRONG_INLINE Packet2cf pmul(const Packet4f& x, const Packet2cf& y) const 00217 { return Packet2cf(Eigen::internal::pmul(x, y.v)); } 00218 }; 00219 00220 template<> struct conj_helper<Packet2cf, Packet4f, false,false> 00221 { 00222 EIGEN_STRONG_INLINE Packet2cf pmadd(const Packet2cf& x, const Packet4f& y, const Packet2cf& c) const 00223 { return padd(c, pmul(x,y)); } 00224 00225 EIGEN_STRONG_INLINE Packet2cf pmul(const Packet2cf& x, const Packet4f& y) const 00226 { return Packet2cf(Eigen::internal::pmul(x.v, y)); } 00227 }; 00228 00229 template<> EIGEN_STRONG_INLINE Packet2cf pdiv<Packet2cf>(const Packet2cf& a, const Packet2cf& b) 00230 { 00231 // TODO optimize it for SSE3 and 4 00232 Packet2cf res = conj_helper<Packet2cf,Packet2cf,false,true>().pmul(a,b); 00233 __m128 s = _mm_mul_ps(b.v,b.v); 00234 return Packet2cf(_mm_div_ps(res.v,_mm_add_ps(s,_mm_castsi128_ps(_mm_shuffle_epi32(_mm_castps_si128(s), 0xb1))))); 00235 } 00236 00237 EIGEN_STRONG_INLINE Packet2cf pcplxflip/*<Packet2cf>*/(const Packet2cf& x) 00238 { 00239 return Packet2cf(vec4f_swizzle1(x.v, 1, 0, 3, 2)); 00240 } 00241 00242 00243 //---------- double ---------- 00244 struct Packet1cd 00245 { 00246 EIGEN_STRONG_INLINE Packet1cd() {} 00247 EIGEN_STRONG_INLINE explicit Packet1cd(const __m128d& a) : v(a) {} 00248 __m128d v; 00249 }; 00250 00251 template<> struct packet_traits<std::complex<double> > : default_packet_traits 00252 { 00253 typedef Packet1cd type; 00254 enum { 00255 Vectorizable = 1, 00256 AlignedOnScalar = 0, 00257 size = 1, 00258 00259 HasAdd = 1, 00260 HasSub = 1, 00261 HasMul = 1, 00262 HasDiv = 1, 00263 HasNegate = 1, 00264 HasAbs = 0, 00265 HasAbs2 = 0, 00266 HasMin = 0, 00267 HasMax = 0, 00268 HasSetLinear = 0 00269 }; 00270 }; 00271 00272 template<> struct unpacket_traits<Packet1cd> { typedef std::complex<double> type; enum {size=1}; }; 00273 00274 template<> EIGEN_STRONG_INLINE Packet1cd padd<Packet1cd>(const Packet1cd& a, const Packet1cd& b) { return Packet1cd(_mm_add_pd(a.v,b.v)); } 00275 template<> EIGEN_STRONG_INLINE Packet1cd psub<Packet1cd>(const Packet1cd& a, const Packet1cd& b) { return Packet1cd(_mm_sub_pd(a.v,b.v)); } 00276 template<> EIGEN_STRONG_INLINE Packet1cd pnegate(const Packet1cd& a) { return Packet1cd(pnegate(a.v)); } 00277 template<> EIGEN_STRONG_INLINE Packet1cd pconj(const Packet1cd& a) 00278 { 00279 const __m128d mask = _mm_castsi128_pd(_mm_set_epi32(0x80000000,0x0,0x0,0x0)); 00280 return Packet1cd(_mm_xor_pd(a.v,mask)); 00281 } 00282 00283 template<> EIGEN_STRONG_INLINE Packet1cd pmul<Packet1cd>(const Packet1cd& a, const Packet1cd& b) 00284 { 00285 // TODO optimize it for SSE3 and 4 00286 #ifdef EIGEN_VECTORIZE_SSE3 00287 return Packet1cd(_mm_addsub_pd(_mm_mul_pd(vec2d_swizzle1(a.v, 0, 0), b.v), 00288 _mm_mul_pd(vec2d_swizzle1(a.v, 1, 1), 00289 vec2d_swizzle1(b.v, 1, 0)))); 00290 #else 00291 const __m128d mask = _mm_castsi128_pd(_mm_set_epi32(0x0,0x0,0x80000000,0x0)); 00292 return Packet1cd(_mm_add_pd(_mm_mul_pd(vec2d_swizzle1(a.v, 0, 0), b.v), 00293 _mm_xor_pd(_mm_mul_pd(vec2d_swizzle1(a.v, 1, 1), 00294 vec2d_swizzle1(b.v, 1, 0)), mask))); 00295 #endif 00296 } 00297 00298 template<> EIGEN_STRONG_INLINE Packet1cd pand <Packet1cd>(const Packet1cd& a, const Packet1cd& b) { return Packet1cd(_mm_and_pd(a.v,b.v)); } 00299 template<> EIGEN_STRONG_INLINE Packet1cd por <Packet1cd>(const Packet1cd& a, const Packet1cd& b) { return Packet1cd(_mm_or_pd(a.v,b.v)); } 00300 template<> EIGEN_STRONG_INLINE Packet1cd pxor <Packet1cd>(const Packet1cd& a, const Packet1cd& b) { return Packet1cd(_mm_xor_pd(a.v,b.v)); } 00301 template<> EIGEN_STRONG_INLINE Packet1cd pandnot<Packet1cd>(const Packet1cd& a, const Packet1cd& b) { return Packet1cd(_mm_andnot_pd(a.v,b.v)); } 00302 00303 // FIXME force unaligned load, this is a temporary fix 00304 template<> EIGEN_STRONG_INLINE Packet1cd pload <Packet1cd>(const std::complex<double>* from) 00305 { EIGEN_DEBUG_ALIGNED_LOAD return Packet1cd(pload<Packet2d>((const double*)from)); } 00306 template<> EIGEN_STRONG_INLINE Packet1cd ploadu<Packet1cd>(const std::complex<double>* from) 00307 { EIGEN_DEBUG_UNALIGNED_LOAD return Packet1cd(ploadu<Packet2d>((const double*)from)); } 00308 template<> EIGEN_STRONG_INLINE Packet1cd pset1<Packet1cd>(const std::complex<double>& from) 00309 { /* here we really have to use unaligned loads :( */ return ploadu<Packet1cd>(&from); } 00310 00311 template<> EIGEN_STRONG_INLINE Packet1cd ploaddup<Packet1cd>(const std::complex<double>* from) { return pset1<Packet1cd>(*from); } 00312 00313 // FIXME force unaligned store, this is a temporary fix 00314 template<> EIGEN_STRONG_INLINE void pstore <std::complex<double> >(std::complex<double> * to, const Packet1cd& from) { EIGEN_DEBUG_ALIGNED_STORE pstore((double*)to, from.v); } 00315 template<> EIGEN_STRONG_INLINE void pstoreu<std::complex<double> >(std::complex<double> * to, const Packet1cd& from) { EIGEN_DEBUG_UNALIGNED_STORE pstoreu((double*)to, from.v); } 00316 00317 template<> EIGEN_STRONG_INLINE void prefetch<std::complex<double> >(const std::complex<double> * addr) { _mm_prefetch((const char*)(addr), _MM_HINT_T0); } 00318 00319 template<> EIGEN_STRONG_INLINE std::complex<double> pfirst<Packet1cd>(const Packet1cd& a) 00320 { 00321 EIGEN_ALIGN16 double res[2]; 00322 _mm_store_pd(res, a.v); 00323 return std::complex<double>(res[0],res[1]); 00324 } 00325 00326 template<> EIGEN_STRONG_INLINE Packet1cd preverse(const Packet1cd& a) { return a; } 00327 00328 template<> EIGEN_STRONG_INLINE std::complex<double> predux<Packet1cd>(const Packet1cd& a) 00329 { 00330 return pfirst(a); 00331 } 00332 00333 template<> EIGEN_STRONG_INLINE Packet1cd preduxp<Packet1cd>(const Packet1cd* vecs) 00334 { 00335 return vecs[0]; 00336 } 00337 00338 template<> EIGEN_STRONG_INLINE std::complex<double> predux_mul<Packet1cd>(const Packet1cd& a) 00339 { 00340 return pfirst(a); 00341 } 00342 00343 template<int Offset> 00344 struct palign_impl<Offset,Packet1cd> 00345 { 00346 static EIGEN_STRONG_INLINE void run(Packet1cd& /*first*/, const Packet1cd& /*second*/) 00347 { 00348 // FIXME is it sure we never have to align a Packet1cd? 00349 // Even though a std::complex<double> has 16 bytes, it is not necessarily aligned on a 16 bytes boundary... 00350 } 00351 }; 00352 00353 template<> struct conj_helper<Packet1cd, Packet1cd, false,true> 00354 { 00355 EIGEN_STRONG_INLINE Packet1cd pmadd(const Packet1cd& x, const Packet1cd& y, const Packet1cd& c) const 00356 { return padd(pmul(x,y),c); } 00357 00358 EIGEN_STRONG_INLINE Packet1cd pmul(const Packet1cd& a, const Packet1cd& b) const 00359 { 00360 #ifdef EIGEN_VECTORIZE_SSE3 00361 return internal::pmul(a, pconj(b)); 00362 #else 00363 const __m128d mask = _mm_castsi128_pd(_mm_set_epi32(0x80000000,0x0,0x0,0x0)); 00364 return Packet1cd(_mm_add_pd(_mm_xor_pd(_mm_mul_pd(vec2d_swizzle1(a.v, 0, 0), b.v), mask), 00365 _mm_mul_pd(vec2d_swizzle1(a.v, 1, 1), 00366 vec2d_swizzle1(b.v, 1, 0)))); 00367 #endif 00368 } 00369 }; 00370 00371 template<> struct conj_helper<Packet1cd, Packet1cd, true,false> 00372 { 00373 EIGEN_STRONG_INLINE Packet1cd pmadd(const Packet1cd& x, const Packet1cd& y, const Packet1cd& c) const 00374 { return padd(pmul(x,y),c); } 00375 00376 EIGEN_STRONG_INLINE Packet1cd pmul(const Packet1cd& a, const Packet1cd& b) const 00377 { 00378 #ifdef EIGEN_VECTORIZE_SSE3 00379 return internal::pmul(pconj(a), b); 00380 #else 00381 const __m128d mask = _mm_castsi128_pd(_mm_set_epi32(0x80000000,0x0,0x0,0x0)); 00382 return Packet1cd(_mm_add_pd(_mm_mul_pd(vec2d_swizzle1(a.v, 0, 0), b.v), 00383 _mm_xor_pd(_mm_mul_pd(vec2d_swizzle1(a.v, 1, 1), 00384 vec2d_swizzle1(b.v, 1, 0)), mask))); 00385 #endif 00386 } 00387 }; 00388 00389 template<> struct conj_helper<Packet1cd, Packet1cd, true,true> 00390 { 00391 EIGEN_STRONG_INLINE Packet1cd pmadd(const Packet1cd& x, const Packet1cd& y, const Packet1cd& c) const 00392 { return padd(pmul(x,y),c); } 00393 00394 EIGEN_STRONG_INLINE Packet1cd pmul(const Packet1cd& a, const Packet1cd& b) const 00395 { 00396 #ifdef EIGEN_VECTORIZE_SSE3 00397 return pconj(internal::pmul(a, b)); 00398 #else 00399 const __m128d mask = _mm_castsi128_pd(_mm_set_epi32(0x80000000,0x0,0x0,0x0)); 00400 return Packet1cd(_mm_sub_pd(_mm_xor_pd(_mm_mul_pd(vec2d_swizzle1(a.v, 0, 0), b.v), mask), 00401 _mm_mul_pd(vec2d_swizzle1(a.v, 1, 1), 00402 vec2d_swizzle1(b.v, 1, 0)))); 00403 #endif 00404 } 00405 }; 00406 00407 template<> struct conj_helper<Packet2d, Packet1cd, false,false> 00408 { 00409 EIGEN_STRONG_INLINE Packet1cd pmadd(const Packet2d& x, const Packet1cd& y, const Packet1cd& c) const 00410 { return padd(c, pmul(x,y)); } 00411 00412 EIGEN_STRONG_INLINE Packet1cd pmul(const Packet2d& x, const Packet1cd& y) const 00413 { return Packet1cd(Eigen::internal::pmul(x, y.v)); } 00414 }; 00415 00416 template<> struct conj_helper<Packet1cd, Packet2d, false,false> 00417 { 00418 EIGEN_STRONG_INLINE Packet1cd pmadd(const Packet1cd& x, const Packet2d& y, const Packet1cd& c) const 00419 { return padd(c, pmul(x,y)); } 00420 00421 EIGEN_STRONG_INLINE Packet1cd pmul(const Packet1cd& x, const Packet2d& y) const 00422 { return Packet1cd(Eigen::internal::pmul(x.v, y)); } 00423 }; 00424 00425 template<> EIGEN_STRONG_INLINE Packet1cd pdiv<Packet1cd>(const Packet1cd& a, const Packet1cd& b) 00426 { 00427 // TODO optimize it for SSE3 and 4 00428 Packet1cd res = conj_helper<Packet1cd,Packet1cd,false,true>().pmul(a,b); 00429 __m128d s = _mm_mul_pd(b.v,b.v); 00430 return Packet1cd(_mm_div_pd(res.v, _mm_add_pd(s,_mm_shuffle_pd(s, s, 0x1)))); 00431 } 00432 00433 EIGEN_STRONG_INLINE Packet1cd pcplxflip/*<Packet1cd>*/(const Packet1cd& x) 00434 { 00435 return Packet1cd(preverse(x.v)); 00436 } 00437 00438 } // end namespace internal 00439 00440 } // end namespace Eigen 00441 00442 #endif // EIGEN_COMPLEX_SSE_H