1 // This file is part of Eigen, a lightweight C++ template library 2 // for linear algebra. 3 // 4 // Copyright (C) 2008 Gael Guennebaud <gael.guennebaud (at) inria.fr> 5 // Copyright (C) 2006-2008 Benoit Jacob <jacob.benoit.1 (at) gmail.com> 6 // 7 // This Source Code Form is subject to the terms of the Mozilla 8 // Public License v. 2.0. If a copy of the MPL was not distributed 9 // with this file, You can obtain one at http://mozilla.org/MPL/2.0/. 10 11 #ifndef EIGEN_GENERIC_PACKET_MATH_H 12 #define EIGEN_GENERIC_PACKET_MATH_H 13 14 namespace Eigen { 15 16 namespace internal { 17 18 /** \internal 19 * \file GenericPacketMath.h 20 * 21 * Default implementation for types not supported by the vectorization. 22 * In practice these functions are provided to make easier the writing 23 * of generic vectorized code. 24 */ 25 26 #ifndef EIGEN_DEBUG_ALIGNED_LOAD 27 #define EIGEN_DEBUG_ALIGNED_LOAD 28 #endif 29 30 #ifndef EIGEN_DEBUG_UNALIGNED_LOAD 31 #define EIGEN_DEBUG_UNALIGNED_LOAD 32 #endif 33 34 #ifndef EIGEN_DEBUG_ALIGNED_STORE 35 #define EIGEN_DEBUG_ALIGNED_STORE 36 #endif 37 38 #ifndef EIGEN_DEBUG_UNALIGNED_STORE 39 #define EIGEN_DEBUG_UNALIGNED_STORE 40 #endif 41 42 struct default_packet_traits 43 { 44 enum { 45 HasHalfPacket = 0, 46 47 HasAdd = 1, 48 HasSub = 1, 49 HasMul = 1, 50 HasNegate = 1, 51 HasAbs = 1, 52 HasArg = 0, 53 HasAbs2 = 1, 54 HasMin = 1, 55 HasMax = 1, 56 HasConj = 1, 57 HasSetLinear = 1, 58 HasBlend = 0, 59 60 HasDiv = 0, 61 HasSqrt = 0, 62 HasRsqrt = 0, 63 HasExp = 0, 64 HasLog = 0, 65 HasLog1p = 0, 66 HasLog10 = 0, 67 HasPow = 0, 68 69 HasSin = 0, 70 HasCos = 0, 71 HasTan = 0, 72 HasASin = 0, 73 HasACos = 0, 74 HasATan = 0, 75 HasSinh = 0, 76 HasCosh = 0, 77 HasTanh = 0, 78 HasLGamma = 0, 79 HasDiGamma = 0, 80 HasZeta = 0, 81 HasPolygamma = 0, 82 HasErf = 0, 83 HasErfc = 0, 84 HasIGamma = 0, 85 HasIGammac = 0, 86 HasBetaInc = 0, 87 88 HasRound = 0, 89 HasFloor = 0, 90 HasCeil = 0, 91 92 HasSign = 0 93 }; 94 }; 95 96 template<typename T> struct packet_traits : default_packet_traits 97 { 98 typedef T type; 99 typedef T half; 100 enum { 101 Vectorizable = 0, 102 size = 1, 103 AlignedOnScalar = 0, 104 HasHalfPacket = 0 105 }; 106 enum { 107 HasAdd = 0, 108 HasSub = 0, 109 HasMul = 0, 110 HasNegate = 0, 111 HasAbs = 0, 112 HasAbs2 = 0, 113 HasMin = 0, 114 HasMax = 0, 115 HasConj = 0, 116 HasSetLinear = 0 117 }; 118 }; 119 120 template<typename T> struct packet_traits<const T> : packet_traits<T> { }; 121 122 template <typename Src, typename Tgt> struct type_casting_traits { 123 enum { 124 VectorizedCast = 0, 125 SrcCoeffRatio = 1, 126 TgtCoeffRatio = 1 127 }; 128 }; 129 130 131 /** \internal \returns static_cast<TgtType>(a) (coeff-wise) */ 132 template <typename SrcPacket, typename TgtPacket> 133 EIGEN_DEVICE_FUNC inline TgtPacket 134 pcast(const SrcPacket& a) { 135 return static_cast<TgtPacket>(a); 136 } 137 template <typename SrcPacket, typename TgtPacket> 138 EIGEN_DEVICE_FUNC inline TgtPacket 139 pcast(const SrcPacket& a, const SrcPacket& /*b*/) { 140 return static_cast<TgtPacket>(a); 141 } 142 143 template <typename SrcPacket, typename TgtPacket> 144 EIGEN_DEVICE_FUNC inline TgtPacket 145 pcast(const SrcPacket& a, const SrcPacket& /*b*/, const SrcPacket& /*c*/, const SrcPacket& /*d*/) { 146 return static_cast<TgtPacket>(a); 147 } 148 149 /** \internal \returns a + b (coeff-wise) */ 150 template<typename Packet> EIGEN_DEVICE_FUNC inline Packet 151 padd(const Packet& a, 152 const Packet& b) { return a+b; } 153 154 /** \internal \returns a - b (coeff-wise) */ 155 template<typename Packet> EIGEN_DEVICE_FUNC inline Packet 156 psub(const Packet& a, 157 const Packet& b) { return a-b; } 158 159 /** \internal \returns -a (coeff-wise) */ 160 template<typename Packet> EIGEN_DEVICE_FUNC inline Packet 161 pnegate(const Packet& a) { return -a; } 162 163 /** \internal \returns conj(a) (coeff-wise) */ 164 165 template<typename Packet> EIGEN_DEVICE_FUNC inline Packet 166 pconj(const Packet& a) { return numext::conj(a); } 167 168 /** \internal \returns a * b (coeff-wise) */ 169 template<typename Packet> EIGEN_DEVICE_FUNC inline Packet 170 pmul(const Packet& a, 171 const Packet& b) { return a*b; } 172 173 /** \internal \returns a / b (coeff-wise) */ 174 template<typename Packet> EIGEN_DEVICE_FUNC inline Packet 175 pdiv(const Packet& a, 176 const Packet& b) { return a/b; } 177 178 /** \internal \returns the min of \a a and \a b (coeff-wise) */ 179 template<typename Packet> EIGEN_DEVICE_FUNC inline Packet 180 pmin(const Packet& a, 181 const Packet& b) { return numext::mini(a, b); } 182 183 /** \internal \returns the max of \a a and \a b (coeff-wise) */ 184 template<typename Packet> EIGEN_DEVICE_FUNC inline Packet 185 pmax(const Packet& a, 186 const Packet& b) { return numext::maxi(a, b); } 187 188 /** \internal \returns the absolute value of \a a */ 189 template<typename Packet> EIGEN_DEVICE_FUNC inline Packet 190 pabs(const Packet& a) { using std::abs; return abs(a); } 191 192 /** \internal \returns the phase angle of \a a */ 193 template<typename Packet> EIGEN_DEVICE_FUNC inline Packet 194 parg(const Packet& a) { using numext::arg; return arg(a); } 195 196 /** \internal \returns the bitwise and of \a a and \a b */ 197 template<typename Packet> EIGEN_DEVICE_FUNC inline Packet 198 pand(const Packet& a, const Packet& b) { return a & b; } 199 200 /** \internal \returns the bitwise or of \a a and \a b */ 201 template<typename Packet> EIGEN_DEVICE_FUNC inline Packet 202 por(const Packet& a, const Packet& b) { return a | b; } 203 204 /** \internal \returns the bitwise xor of \a a and \a b */ 205 template<typename Packet> EIGEN_DEVICE_FUNC inline Packet 206 pxor(const Packet& a, const Packet& b) { return a ^ b; } 207 208 /** \internal \returns the bitwise andnot of \a a and \a b */ 209 template<typename Packet> EIGEN_DEVICE_FUNC inline Packet 210 pandnot(const Packet& a, const Packet& b) { return a & (!b); } 211 212 /** \internal \returns a packet version of \a *from, from must be 16 bytes aligned */ 213 template<typename Packet> EIGEN_DEVICE_FUNC inline Packet 214 pload(const typename unpacket_traits<Packet>::type* from) { return *from; } 215 216 /** \internal \returns a packet version of \a *from, (un-aligned load) */ 217 template<typename Packet> EIGEN_DEVICE_FUNC inline Packet 218 ploadu(const typename unpacket_traits<Packet>::type* from) { return *from; } 219 220 /** \internal \returns a packet with constant coefficients \a a, e.g.: (a,a,a,a) */ 221 template<typename Packet> EIGEN_DEVICE_FUNC inline Packet 222 pset1(const typename unpacket_traits<Packet>::type& a) { return a; } 223 224 /** \internal \returns a packet with constant coefficients \a a[0], e.g.: (a[0],a[0],a[0],a[0]) */ 225 template<typename Packet> EIGEN_DEVICE_FUNC inline Packet 226 pload1(const typename unpacket_traits<Packet>::type *a) { return pset1<Packet>(*a); } 227 228 /** \internal \returns a packet with elements of \a *from duplicated. 229 * For instance, for a packet of 8 elements, 4 scalars will be read from \a *from and 230 * duplicated to form: {from[0],from[0],from[1],from[1],from[2],from[2],from[3],from[3]} 231 * Currently, this function is only used for scalar * complex products. 232 */ 233 template<typename Packet> EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE Packet 234 ploaddup(const typename unpacket_traits<Packet>::type* from) { return *from; } 235 236 /** \internal \returns a packet with elements of \a *from quadrupled. 237 * For instance, for a packet of 8 elements, 2 scalars will be read from \a *from and 238 * replicated to form: {from[0],from[0],from[0],from[0],from[1],from[1],from[1],from[1]} 239 * Currently, this function is only used in matrix products. 240 * For packet-size smaller or equal to 4, this function is equivalent to pload1 241 */ 242 template<typename Packet> EIGEN_DEVICE_FUNC inline Packet 243 ploadquad(const typename unpacket_traits<Packet>::type* from) 244 { return pload1<Packet>(from); } 245 246 /** \internal equivalent to 247 * \code 248 * a0 = pload1(a+0); 249 * a1 = pload1(a+1); 250 * a2 = pload1(a+2); 251 * a3 = pload1(a+3); 252 * \endcode 253 * \sa pset1, pload1, ploaddup, pbroadcast2 254 */ 255 template<typename Packet> EIGEN_DEVICE_FUNC 256 inline void pbroadcast4(const typename unpacket_traits<Packet>::type *a, 257 Packet& a0, Packet& a1, Packet& a2, Packet& a3) 258 { 259 a0 = pload1<Packet>(a+0); 260 a1 = pload1<Packet>(a+1); 261 a2 = pload1<Packet>(a+2); 262 a3 = pload1<Packet>(a+3); 263 } 264 265 /** \internal equivalent to 266 * \code 267 * a0 = pload1(a+0); 268 * a1 = pload1(a+1); 269 * \endcode 270 * \sa pset1, pload1, ploaddup, pbroadcast4 271 */ 272 template<typename Packet> EIGEN_DEVICE_FUNC 273 inline void pbroadcast2(const typename unpacket_traits<Packet>::type *a, 274 Packet& a0, Packet& a1) 275 { 276 a0 = pload1<Packet>(a+0); 277 a1 = pload1<Packet>(a+1); 278 } 279 280 /** \internal \brief Returns a packet with coefficients (a,a+1,...,a+packet_size-1). */ 281 template<typename Packet> EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE Packet 282 plset(const typename unpacket_traits<Packet>::type& a) { return a; } 283 284 /** \internal copy the packet \a from to \a *to, \a to must be 16 bytes aligned */ 285 template<typename Scalar, typename Packet> EIGEN_DEVICE_FUNC inline void pstore(Scalar* to, const Packet& from) 286 { (*to) = from; } 287 288 /** \internal copy the packet \a from to \a *to, (un-aligned store) */ 289 template<typename Scalar, typename Packet> EIGEN_DEVICE_FUNC inline void pstoreu(Scalar* to, const Packet& from) 290 { (*to) = from; } 291 292 template<typename Scalar, typename Packet> EIGEN_DEVICE_FUNC inline Packet pgather(const Scalar* from, Index /*stride*/) 293 { return ploadu<Packet>(from); } 294 295 template<typename Scalar, typename Packet> EIGEN_DEVICE_FUNC inline void pscatter(Scalar* to, const Packet& from, Index /*stride*/) 296 { pstore(to, from); } 297 298 /** \internal tries to do cache prefetching of \a addr */ 299 template<typename Scalar> EIGEN_DEVICE_FUNC inline void prefetch(const Scalar* addr) 300 { 301 #ifdef __CUDA_ARCH__ 302 #if defined(__LP64__) 303 // 64-bit pointer operand constraint for inlined asm 304 asm(" prefetch.L1 [ %1 ];" : "=l"(addr) : "l"(addr)); 305 #else 306 // 32-bit pointer operand constraint for inlined asm 307 asm(" prefetch.L1 [ %1 ];" : "=r"(addr) : "r"(addr)); 308 #endif 309 #elif (!EIGEN_COMP_MSVC) && (EIGEN_COMP_GNUC || EIGEN_COMP_CLANG || EIGEN_COMP_ICC) 310 __builtin_prefetch(addr); 311 #endif 312 } 313 314 /** \internal \returns the first element of a packet */ 315 template<typename Packet> EIGEN_DEVICE_FUNC inline typename unpacket_traits<Packet>::type pfirst(const Packet& a) 316 { return a; } 317 318 /** \internal \returns a packet where the element i contains the sum of the packet of \a vec[i] */ 319 template<typename Packet> EIGEN_DEVICE_FUNC inline Packet 320 preduxp(const Packet* vecs) { return vecs[0]; } 321 322 /** \internal \returns the sum of the elements of \a a*/ 323 template<typename Packet> EIGEN_DEVICE_FUNC inline typename unpacket_traits<Packet>::type predux(const Packet& a) 324 { return a; } 325 326 /** \internal \returns the sum of the elements of \a a by block of 4 elements. 327 * For a packet {a0, a1, a2, a3, a4, a5, a6, a7}, it returns a half packet {a0+a4, a1+a5, a2+a6, a3+a7} 328 * For packet-size smaller or equal to 4, this boils down to a noop. 329 */ 330 template<typename Packet> EIGEN_DEVICE_FUNC inline 331 typename conditional<(unpacket_traits<Packet>::size%8)==0,typename unpacket_traits<Packet>::half,Packet>::type 332 predux_downto4(const Packet& a) 333 { return a; } 334 335 /** \internal \returns the product of the elements of \a a*/ 336 template<typename Packet> EIGEN_DEVICE_FUNC inline typename unpacket_traits<Packet>::type predux_mul(const Packet& a) 337 { return a; } 338 339 /** \internal \returns the min of the elements of \a a*/ 340 template<typename Packet> EIGEN_DEVICE_FUNC inline typename unpacket_traits<Packet>::type predux_min(const Packet& a) 341 { return a; } 342 343 /** \internal \returns the max of the elements of \a a*/ 344 template<typename Packet> EIGEN_DEVICE_FUNC inline typename unpacket_traits<Packet>::type predux_max(const Packet& a) 345 { return a; } 346 347 /** \internal \returns the reversed elements of \a a*/ 348 template<typename Packet> EIGEN_DEVICE_FUNC inline Packet preverse(const Packet& a) 349 { return a; } 350 351 /** \internal \returns \a a with real and imaginary part flipped (for complex type only) */ 352 template<typename Packet> EIGEN_DEVICE_FUNC inline Packet pcplxflip(const Packet& a) 353 { 354 // FIXME: uncomment the following in case we drop the internal imag and real functions. 355 // using std::imag; 356 // using std::real; 357 return Packet(imag(a),real(a)); 358 } 359 360 /************************** 361 * Special math functions 362 ***************************/ 363 364 /** \internal \returns the sine of \a a (coeff-wise) */ 365 template<typename Packet> EIGEN_DECLARE_FUNCTION_ALLOWING_MULTIPLE_DEFINITIONS 366 Packet psin(const Packet& a) { using std::sin; return sin(a); } 367 368 /** \internal \returns the cosine of \a a (coeff-wise) */ 369 template<typename Packet> EIGEN_DECLARE_FUNCTION_ALLOWING_MULTIPLE_DEFINITIONS 370 Packet pcos(const Packet& a) { using std::cos; return cos(a); } 371 372 /** \internal \returns the tan of \a a (coeff-wise) */ 373 template<typename Packet> EIGEN_DECLARE_FUNCTION_ALLOWING_MULTIPLE_DEFINITIONS 374 Packet ptan(const Packet& a) { using std::tan; return tan(a); } 375 376 /** \internal \returns the arc sine of \a a (coeff-wise) */ 377 template<typename Packet> EIGEN_DECLARE_FUNCTION_ALLOWING_MULTIPLE_DEFINITIONS 378 Packet pasin(const Packet& a) { using std::asin; return asin(a); } 379 380 /** \internal \returns the arc cosine of \a a (coeff-wise) */ 381 template<typename Packet> EIGEN_DECLARE_FUNCTION_ALLOWING_MULTIPLE_DEFINITIONS 382 Packet pacos(const Packet& a) { using std::acos; return acos(a); } 383 384 /** \internal \returns the arc tangent of \a a (coeff-wise) */ 385 template<typename Packet> EIGEN_DECLARE_FUNCTION_ALLOWING_MULTIPLE_DEFINITIONS 386 Packet patan(const Packet& a) { using std::atan; return atan(a); } 387 388 /** \internal \returns the hyperbolic sine of \a a (coeff-wise) */ 389 template<typename Packet> EIGEN_DECLARE_FUNCTION_ALLOWING_MULTIPLE_DEFINITIONS 390 Packet psinh(const Packet& a) { using std::sinh; return sinh(a); } 391 392 /** \internal \returns the hyperbolic cosine of \a a (coeff-wise) */ 393 template<typename Packet> EIGEN_DECLARE_FUNCTION_ALLOWING_MULTIPLE_DEFINITIONS 394 Packet pcosh(const Packet& a) { using std::cosh; return cosh(a); } 395 396 /** \internal \returns the hyperbolic tan of \a a (coeff-wise) */ 397 template<typename Packet> EIGEN_DECLARE_FUNCTION_ALLOWING_MULTIPLE_DEFINITIONS 398 Packet ptanh(const Packet& a) { using std::tanh; return tanh(a); } 399 400 /** \internal \returns the exp of \a a (coeff-wise) */ 401 template<typename Packet> EIGEN_DECLARE_FUNCTION_ALLOWING_MULTIPLE_DEFINITIONS 402 Packet pexp(const Packet& a) { using std::exp; return exp(a); } 403 404 /** \internal \returns the log of \a a (coeff-wise) */ 405 template<typename Packet> EIGEN_DECLARE_FUNCTION_ALLOWING_MULTIPLE_DEFINITIONS 406 Packet plog(const Packet& a) { using std::log; return log(a); } 407 408 /** \internal \returns the log1p of \a a (coeff-wise) */ 409 template<typename Packet> EIGEN_DECLARE_FUNCTION_ALLOWING_MULTIPLE_DEFINITIONS 410 Packet plog1p(const Packet& a) { return numext::log1p(a); } 411 412 /** \internal \returns the log10 of \a a (coeff-wise) */ 413 template<typename Packet> EIGEN_DECLARE_FUNCTION_ALLOWING_MULTIPLE_DEFINITIONS 414 Packet plog10(const Packet& a) { using std::log10; return log10(a); } 415 416 /** \internal \returns the square-root of \a a (coeff-wise) */ 417 template<typename Packet> EIGEN_DECLARE_FUNCTION_ALLOWING_MULTIPLE_DEFINITIONS 418 Packet psqrt(const Packet& a) { using std::sqrt; return sqrt(a); } 419 420 /** \internal \returns the reciprocal square-root of \a a (coeff-wise) */ 421 template<typename Packet> EIGEN_DECLARE_FUNCTION_ALLOWING_MULTIPLE_DEFINITIONS 422 Packet prsqrt(const Packet& a) { 423 return pdiv(pset1<Packet>(1), psqrt(a)); 424 } 425 426 /** \internal \returns the rounded value of \a a (coeff-wise) */ 427 template<typename Packet> EIGEN_DECLARE_FUNCTION_ALLOWING_MULTIPLE_DEFINITIONS 428 Packet pround(const Packet& a) { using numext::round; return round(a); } 429 430 /** \internal \returns the floor of \a a (coeff-wise) */ 431 template<typename Packet> EIGEN_DECLARE_FUNCTION_ALLOWING_MULTIPLE_DEFINITIONS 432 Packet pfloor(const Packet& a) { using numext::floor; return floor(a); } 433 434 /** \internal \returns the ceil of \a a (coeff-wise) */ 435 template<typename Packet> EIGEN_DECLARE_FUNCTION_ALLOWING_MULTIPLE_DEFINITIONS 436 Packet pceil(const Packet& a) { using numext::ceil; return ceil(a); } 437 438 /*************************************************************************** 439 * The following functions might not have to be overwritten for vectorized types 440 ***************************************************************************/ 441 442 /** \internal copy a packet with constant coeficient \a a (e.g., [a,a,a,a]) to \a *to. \a to must be 16 bytes aligned */ 443 // NOTE: this function must really be templated on the packet type (think about different packet types for the same scalar type) 444 template<typename Packet> 445 inline void pstore1(typename unpacket_traits<Packet>::type* to, const typename unpacket_traits<Packet>::type& a) 446 { 447 pstore(to, pset1<Packet>(a)); 448 } 449 450 /** \internal \returns a * b + c (coeff-wise) */ 451 template<typename Packet> EIGEN_DEVICE_FUNC inline Packet 452 pmadd(const Packet& a, 453 const Packet& b, 454 const Packet& c) 455 { return padd(pmul(a, b),c); } 456 457 /** \internal \returns a packet version of \a *from. 458 * The pointer \a from must be aligned on a \a Alignment bytes boundary. */ 459 template<typename Packet, int Alignment> 460 EIGEN_DEVICE_FUNC EIGEN_ALWAYS_INLINE Packet ploadt(const typename unpacket_traits<Packet>::type* from) 461 { 462 if(Alignment >= unpacket_traits<Packet>::alignment) 463 return pload<Packet>(from); 464 else 465 return ploadu<Packet>(from); 466 } 467 468 /** \internal copy the packet \a from to \a *to. 469 * The pointer \a from must be aligned on a \a Alignment bytes boundary. */ 470 template<typename Scalar, typename Packet, int Alignment> 471 EIGEN_DEVICE_FUNC EIGEN_ALWAYS_INLINE void pstoret(Scalar* to, const Packet& from) 472 { 473 if(Alignment >= unpacket_traits<Packet>::alignment) 474 pstore(to, from); 475 else 476 pstoreu(to, from); 477 } 478 479 /** \internal \returns a packet version of \a *from. 480 * Unlike ploadt, ploadt_ro takes advantage of the read-only memory path on the 481 * hardware if available to speedup the loading of data that won't be modified 482 * by the current computation. 483 */ 484 template<typename Packet, int LoadMode> 485 EIGEN_DEVICE_FUNC EIGEN_ALWAYS_INLINE Packet ploadt_ro(const typename unpacket_traits<Packet>::type* from) 486 { 487 return ploadt<Packet, LoadMode>(from); 488 } 489 490 /** \internal default implementation of palign() allowing partial specialization */ 491 template<int Offset,typename PacketType> 492 struct palign_impl 493 { 494 // by default data are aligned, so there is nothing to be done :) 495 static inline void run(PacketType&, const PacketType&) {} 496 }; 497 498 /** \internal update \a first using the concatenation of the packet_size minus \a Offset last elements 499 * of \a first and \a Offset first elements of \a second. 500 * 501 * This function is currently only used to optimize matrix-vector products on unligned matrices. 502 * It takes 2 packets that represent a contiguous memory array, and returns a packet starting 503 * at the position \a Offset. For instance, for packets of 4 elements, we have: 504 * Input: 505 * - first = {f0,f1,f2,f3} 506 * - second = {s0,s1,s2,s3} 507 * Output: 508 * - if Offset==0 then {f0,f1,f2,f3} 509 * - if Offset==1 then {f1,f2,f3,s0} 510 * - if Offset==2 then {f2,f3,s0,s1} 511 * - if Offset==3 then {f3,s0,s1,s3} 512 */ 513 template<int Offset,typename PacketType> 514 inline void palign(PacketType& first, const PacketType& second) 515 { 516 palign_impl<Offset,PacketType>::run(first,second); 517 } 518 519 /*************************************************************************** 520 * Fast complex products (GCC generates a function call which is very slow) 521 ***************************************************************************/ 522 523 // Eigen+CUDA does not support complexes. 524 #ifndef __CUDACC__ 525 526 template<> inline std::complex<float> pmul(const std::complex<float>& a, const std::complex<float>& b) 527 { return std::complex<float>(real(a)*real(b) - imag(a)*imag(b), imag(a)*real(b) + real(a)*imag(b)); } 528 529 template<> inline std::complex<double> pmul(const std::complex<double>& a, const std::complex<double>& b) 530 { return std::complex<double>(real(a)*real(b) - imag(a)*imag(b), imag(a)*real(b) + real(a)*imag(b)); } 531 532 #endif 533 534 535 /*************************************************************************** 536 * PacketBlock, that is a collection of N packets where the number of words 537 * in the packet is a multiple of N. 538 ***************************************************************************/ 539 template <typename Packet,int N=unpacket_traits<Packet>::size> struct PacketBlock { 540 Packet packet[N]; 541 }; 542 543 template<typename Packet> EIGEN_DEVICE_FUNC inline void 544 ptranspose(PacketBlock<Packet,1>& /*kernel*/) { 545 // Nothing to do in the scalar case, i.e. a 1x1 matrix. 546 } 547 548 /*************************************************************************** 549 * Selector, i.e. vector of N boolean values used to select (i.e. blend) 550 * words from 2 packets. 551 ***************************************************************************/ 552 template <size_t N> struct Selector { 553 bool select[N]; 554 }; 555 556 template<typename Packet> EIGEN_DEVICE_FUNC inline Packet 557 pblend(const Selector<unpacket_traits<Packet>::size>& ifPacket, const Packet& thenPacket, const Packet& elsePacket) { 558 return ifPacket.select[0] ? thenPacket : elsePacket; 559 } 560 561 /** \internal \returns \a a with the first coefficient replaced by the scalar b */ 562 template<typename Packet> EIGEN_DEVICE_FUNC inline Packet 563 pinsertfirst(const Packet& a, typename unpacket_traits<Packet>::type b) 564 { 565 // Default implementation based on pblend. 566 // It must be specialized for higher performance. 567 Selector<unpacket_traits<Packet>::size> mask; 568 mask.select[0] = true; 569 // This for loop should be optimized away by the compiler. 570 for(Index i=1; i<unpacket_traits<Packet>::size; ++i) 571 mask.select[i] = false; 572 return pblend(mask, pset1<Packet>(b), a); 573 } 574 575 /** \internal \returns \a a with the last coefficient replaced by the scalar b */ 576 template<typename Packet> EIGEN_DEVICE_FUNC inline Packet 577 pinsertlast(const Packet& a, typename unpacket_traits<Packet>::type b) 578 { 579 // Default implementation based on pblend. 580 // It must be specialized for higher performance. 581 Selector<unpacket_traits<Packet>::size> mask; 582 // This for loop should be optimized away by the compiler. 583 for(Index i=0; i<unpacket_traits<Packet>::size-1; ++i) 584 mask.select[i] = false; 585 mask.select[unpacket_traits<Packet>::size-1] = true; 586 return pblend(mask, pset1<Packet>(b), a); 587 } 588 589 } // end namespace internal 590 591 } // end namespace Eigen 592 593 #endif // EIGEN_GENERIC_PACKET_MATH_H 594
