1 // This file is part of Eigen, a lightweight C++ template library 2 // for linear algebra. 3 // 4 // Copyright (C) 2016 Konstantinos Margaritis <markos@freevec.org> 5 // 6 // This Source Code Form is subject to the terms of the Mozilla 7 // Public License v. 2.0. If a copy of the MPL was not distributed 8 // with this file, You can obtain one at http://mozilla.org/MPL/2.0/. 9 10 #ifndef EIGEN_PACKET_MATH_ZVECTOR_H 11 #define EIGEN_PACKET_MATH_ZVECTOR_H 12 13 #include <stdint.h> 14 15 namespace Eigen { 16 17 namespace internal { 18 19 #ifndef EIGEN_CACHEFRIENDLY_PRODUCT_THRESHOLD 20 #define EIGEN_CACHEFRIENDLY_PRODUCT_THRESHOLD 4 21 #endif 22 23 #ifndef EIGEN_HAS_SINGLE_INSTRUCTION_MADD 24 #define EIGEN_HAS_SINGLE_INSTRUCTION_MADD 25 #endif 26 27 #ifndef EIGEN_HAS_SINGLE_INSTRUCTION_CJMADD 28 #define EIGEN_HAS_SINGLE_INSTRUCTION_CJMADD 29 #endif 30 31 #ifndef EIGEN_ARCH_DEFAULT_NUMBER_OF_REGISTERS 32 #define EIGEN_ARCH_DEFAULT_NUMBER_OF_REGISTERS 16 33 #endif 34 35 typedef __vector int Packet4i; 36 typedef __vector unsigned int Packet4ui; 37 typedef __vector __bool int Packet4bi; 38 typedef __vector short int Packet8i; 39 typedef __vector unsigned char Packet16uc; 40 typedef __vector double Packet2d; 41 typedef __vector unsigned long long Packet2ul; 42 typedef __vector long long Packet2l; 43 44 typedef struct { 45 Packet2d v4f[2]; 46 } Packet4f; 47 48 typedef union { 49 int32_t i[4]; 50 uint32_t ui[4]; 51 int64_t l[2]; 52 uint64_t ul[2]; 53 double d[2]; 54 Packet4i v4i; 55 Packet4ui v4ui; 56 Packet2l v2l; 57 Packet2ul v2ul; 58 Packet2d v2d; 59 } Packet; 60 61 // We don't want to write the same code all the time, but we need to reuse the constants 62 // and it doesn't really work to declare them global, so we define macros instead 63 64 #define _EIGEN_DECLARE_CONST_FAST_Packet4i(NAME,X) \ 65 Packet4i p4i_##NAME = reinterpret_cast<Packet4i>(vec_splat_s32(X)) 66 67 #define _EIGEN_DECLARE_CONST_FAST_Packet2d(NAME,X) \ 68 Packet2d p2d_##NAME = reinterpret_cast<Packet2d>(vec_splat_s64(X)) 69 70 #define _EIGEN_DECLARE_CONST_FAST_Packet2l(NAME,X) \ 71 Packet2l p2l_##NAME = reinterpret_cast<Packet2l>(vec_splat_s64(X)) 72 73 #define _EIGEN_DECLARE_CONST_Packet4i(NAME,X) \ 74 Packet4i p4i_##NAME = pset1<Packet4i>(X) 75 76 #define _EIGEN_DECLARE_CONST_Packet2d(NAME,X) \ 77 Packet2d p2d_##NAME = pset1<Packet2d>(X) 78 79 #define _EIGEN_DECLARE_CONST_Packet2l(NAME,X) \ 80 Packet2l p2l_##NAME = pset1<Packet2l>(X) 81 82 // These constants are endian-agnostic 83 //static _EIGEN_DECLARE_CONST_FAST_Packet4i(ZERO, 0); //{ 0, 0, 0, 0,} 84 static _EIGEN_DECLARE_CONST_FAST_Packet4i(ONE, 1); //{ 1, 1, 1, 1} 85 86 static _EIGEN_DECLARE_CONST_FAST_Packet2d(ZERO, 0); 87 static _EIGEN_DECLARE_CONST_FAST_Packet2l(ZERO, 0); 88 static _EIGEN_DECLARE_CONST_FAST_Packet2l(ONE, 1); 89 90 static Packet2d p2d_ONE = { 1.0, 1.0 }; 91 static Packet2d p2d_ZERO_ = { -0.0, -0.0 }; 92 93 static Packet4i p4i_COUNTDOWN = { 0, 1, 2, 3 }; 94 static Packet4f p4f_COUNTDOWN = { 0.0, 1.0, 2.0, 3.0 }; 95 static Packet2d p2d_COUNTDOWN = reinterpret_cast<Packet2d>(vec_sld(reinterpret_cast<Packet16uc>(p2d_ZERO), reinterpret_cast<Packet16uc>(p2d_ONE), 8)); 96 97 static Packet16uc p16uc_PSET64_HI = { 0,1,2,3, 4,5,6,7, 0,1,2,3, 4,5,6,7 }; 98 static Packet16uc p16uc_DUPLICATE32_HI = { 0,1,2,3, 0,1,2,3, 4,5,6,7, 4,5,6,7 }; 99 100 // Mask alignment 101 #define _EIGEN_MASK_ALIGNMENT 0xfffffffffffffff0 102 103 #define _EIGEN_ALIGNED_PTR(x) ((std::ptrdiff_t)(x) & _EIGEN_MASK_ALIGNMENT) 104 105 // Handle endianness properly while loading constants 106 // Define global static constants: 107 108 static Packet16uc p16uc_FORWARD = { 0,1,2,3, 4,5,6,7, 8,9,10,11, 12,13,14,15 }; 109 static Packet16uc p16uc_REVERSE32 = { 12,13,14,15, 8,9,10,11, 4,5,6,7, 0,1,2,3 }; 110 static Packet16uc p16uc_REVERSE64 = { 8,9,10,11, 12,13,14,15, 0,1,2,3, 4,5,6,7 }; 111 112 static Packet16uc p16uc_PSET32_WODD = vec_sld((Packet16uc) vec_splat((Packet4ui)p16uc_FORWARD, 0), (Packet16uc) vec_splat((Packet4ui)p16uc_FORWARD, 2), 8);//{ 0,1,2,3, 0,1,2,3, 8,9,10,11, 8,9,10,11 }; 113 static Packet16uc p16uc_PSET32_WEVEN = vec_sld(p16uc_DUPLICATE32_HI, (Packet16uc) vec_splat((Packet4ui)p16uc_FORWARD, 3), 8);//{ 4,5,6,7, 4,5,6,7, 12,13,14,15, 12,13,14,15 }; 114 /*static Packet16uc p16uc_HALF64_0_16 = vec_sld((Packet16uc)p4i_ZERO, vec_splat((Packet16uc) vec_abs(p4i_MINUS16), 3), 8); //{ 0,0,0,0, 0,0,0,0, 16,16,16,16, 16,16,16,16}; 115 116 static Packet16uc p16uc_PSET64_HI = (Packet16uc) vec_mergeh((Packet4ui)p16uc_PSET32_WODD, (Packet4ui)p16uc_PSET32_WEVEN); //{ 0,1,2,3, 4,5,6,7, 0,1,2,3, 4,5,6,7 };*/ 117 static Packet16uc p16uc_PSET64_LO = (Packet16uc) vec_mergel((Packet4ui)p16uc_PSET32_WODD, (Packet4ui)p16uc_PSET32_WEVEN); //{ 8,9,10,11, 12,13,14,15, 8,9,10,11, 12,13,14,15 }; 118 /*static Packet16uc p16uc_TRANSPOSE64_HI = vec_add(p16uc_PSET64_HI, p16uc_HALF64_0_16); //{ 0,1,2,3, 4,5,6,7, 16,17,18,19, 20,21,22,23}; 119 static Packet16uc p16uc_TRANSPOSE64_LO = vec_add(p16uc_PSET64_LO, p16uc_HALF64_0_16); //{ 8,9,10,11, 12,13,14,15, 24,25,26,27, 28,29,30,31};*/ 120 static Packet16uc p16uc_TRANSPOSE64_HI = { 0,1,2,3, 4,5,6,7, 16,17,18,19, 20,21,22,23}; 121 static Packet16uc p16uc_TRANSPOSE64_LO = { 8,9,10,11, 12,13,14,15, 24,25,26,27, 28,29,30,31}; 122 123 //static Packet16uc p16uc_COMPLEX32_REV = vec_sld(p16uc_REVERSE32, p16uc_REVERSE32, 8); //{ 4,5,6,7, 0,1,2,3, 12,13,14,15, 8,9,10,11 }; 124 125 //static Packet16uc p16uc_COMPLEX32_REV2 = vec_sld(p16uc_FORWARD, p16uc_FORWARD, 8); //{ 8,9,10,11, 12,13,14,15, 0,1,2,3, 4,5,6,7 }; 126 127 128 #if EIGEN_HAS_BUILTIN(__builtin_prefetch) || EIGEN_COMP_GNUC 129 #define EIGEN_ZVECTOR_PREFETCH(ADDR) __builtin_prefetch(ADDR); 130 #else 131 #define EIGEN_ZVECTOR_PREFETCH(ADDR) asm( " pfd [%[addr]]\n" :: [addr] "r" (ADDR) : "cc" ); 132 #endif 133 134 template<> struct packet_traits<int> : default_packet_traits 135 { 136 typedef Packet4i type; 137 typedef Packet4i half; 138 enum { 139 Vectorizable = 1, 140 AlignedOnScalar = 1, 141 size = 4, 142 HasHalfPacket = 0, 143 144 HasAdd = 1, 145 HasSub = 1, 146 HasMul = 1, 147 HasDiv = 1, 148 HasBlend = 1 149 }; 150 }; 151 152 template<> struct packet_traits<float> : default_packet_traits 153 { 154 typedef Packet4f type; 155 typedef Packet4f half; 156 enum { 157 Vectorizable = 1, 158 AlignedOnScalar = 1, 159 size=4, 160 HasHalfPacket = 0, 161 162 HasAdd = 1, 163 HasSub = 1, 164 HasMul = 1, 165 HasDiv = 1, 166 HasMin = 1, 167 HasMax = 1, 168 HasAbs = 1, 169 HasSin = 0, 170 HasCos = 0, 171 HasLog = 0, 172 HasExp = 1, 173 HasSqrt = 1, 174 HasRsqrt = 1, 175 HasRound = 1, 176 HasFloor = 1, 177 HasCeil = 1, 178 HasNegate = 1, 179 HasBlend = 1 180 }; 181 }; 182 183 template<> struct packet_traits<double> : default_packet_traits 184 { 185 typedef Packet2d type; 186 typedef Packet2d half; 187 enum { 188 Vectorizable = 1, 189 AlignedOnScalar = 1, 190 size=2, 191 HasHalfPacket = 1, 192 193 HasAdd = 1, 194 HasSub = 1, 195 HasMul = 1, 196 HasDiv = 1, 197 HasMin = 1, 198 HasMax = 1, 199 HasAbs = 1, 200 HasSin = 0, 201 HasCos = 0, 202 HasLog = 0, 203 HasExp = 1, 204 HasSqrt = 1, 205 HasRsqrt = 1, 206 HasRound = 1, 207 HasFloor = 1, 208 HasCeil = 1, 209 HasNegate = 1, 210 HasBlend = 1 211 }; 212 }; 213 214 template<> struct unpacket_traits<Packet4i> { typedef int type; enum {size=4, alignment=Aligned16}; typedef Packet4i half; }; 215 template<> struct unpacket_traits<Packet4f> { typedef float type; enum {size=4, alignment=Aligned16}; typedef Packet4f half; }; 216 template<> struct unpacket_traits<Packet2d> { typedef double type; enum {size=2, alignment=Aligned16}; typedef Packet2d half; }; 217 218 /* Forward declaration */ 219 EIGEN_DEVICE_FUNC inline void ptranspose(PacketBlock<Packet4f,4>& kernel); 220 221 inline std::ostream & operator <<(std::ostream & s, const Packet4i & v) 222 { 223 Packet vt; 224 vt.v4i = v; 225 s << vt.i[0] << ", " << vt.i[1] << ", " << vt.i[2] << ", " << vt.i[3]; 226 return s; 227 } 228 229 inline std::ostream & operator <<(std::ostream & s, const Packet4ui & v) 230 { 231 Packet vt; 232 vt.v4ui = v; 233 s << vt.ui[0] << ", " << vt.ui[1] << ", " << vt.ui[2] << ", " << vt.ui[3]; 234 return s; 235 } 236 237 inline std::ostream & operator <<(std::ostream & s, const Packet2l & v) 238 { 239 Packet vt; 240 vt.v2l = v; 241 s << vt.l[0] << ", " << vt.l[1]; 242 return s; 243 } 244 245 inline std::ostream & operator <<(std::ostream & s, const Packet2ul & v) 246 { 247 Packet vt; 248 vt.v2ul = v; 249 s << vt.ul[0] << ", " << vt.ul[1] ; 250 return s; 251 } 252 253 inline std::ostream & operator <<(std::ostream & s, const Packet2d & v) 254 { 255 Packet vt; 256 vt.v2d = v; 257 s << vt.d[0] << ", " << vt.d[1]; 258 return s; 259 } 260 261 /* Helper function to simulate a vec_splat_packet4f 262 */ 263 template<int element> EIGEN_STRONG_INLINE Packet4f vec_splat_packet4f(const Packet4f& from) 264 { 265 Packet4f splat; 266 switch (element) { 267 case 0: 268 splat.v4f[0] = vec_splat(from.v4f[0], 0); 269 splat.v4f[1] = splat.v4f[0]; 270 break; 271 case 1: 272 splat.v4f[0] = vec_splat(from.v4f[0], 1); 273 splat.v4f[1] = splat.v4f[0]; 274 break; 275 case 2: 276 splat.v4f[0] = vec_splat(from.v4f[1], 0); 277 splat.v4f[1] = splat.v4f[0]; 278 break; 279 case 3: 280 splat.v4f[0] = vec_splat(from.v4f[1], 1); 281 splat.v4f[1] = splat.v4f[0]; 282 break; 283 } 284 return splat; 285 } 286 287 template<int Offset> 288 struct palign_impl<Offset,Packet4i> 289 { 290 static EIGEN_STRONG_INLINE void run(Packet4i& first, const Packet4i& second) 291 { 292 switch (Offset % 4) { 293 case 1: 294 first = vec_sld(first, second, 4); break; 295 case 2: 296 first = vec_sld(first, second, 8); break; 297 case 3: 298 first = vec_sld(first, second, 12); break; 299 } 300 } 301 }; 302 303 /* This is a tricky one, we have to translate float alignment to vector elements of sizeof double 304 */ 305 template<int Offset> 306 struct palign_impl<Offset,Packet4f> 307 { 308 static EIGEN_STRONG_INLINE void run(Packet4f& first, const Packet4f& second) 309 { 310 switch (Offset % 4) { 311 case 1: 312 first.v4f[0] = vec_sld(first.v4f[0], first.v4f[1], 8); 313 first.v4f[1] = vec_sld(first.v4f[1], second.v4f[0], 8); 314 break; 315 case 2: 316 first.v4f[0] = first.v4f[1]; 317 first.v4f[1] = second.v4f[0]; 318 break; 319 case 3: 320 first.v4f[0] = vec_sld(first.v4f[1], second.v4f[0], 8); 321 first.v4f[1] = vec_sld(second.v4f[0], second.v4f[1], 8); 322 break; 323 } 324 } 325 }; 326 327 328 template<int Offset> 329 struct palign_impl<Offset,Packet2d> 330 { 331 static EIGEN_STRONG_INLINE void run(Packet2d& first, const Packet2d& second) 332 { 333 if (Offset == 1) 334 first = reinterpret_cast<Packet2d>(vec_sld(reinterpret_cast<Packet4i>(first), reinterpret_cast<Packet4i>(second), 8)); 335 } 336 }; 337 338 template<> EIGEN_STRONG_INLINE Packet4i pload<Packet4i>(const int* from) 339 { 340 // FIXME: No intrinsic yet 341 EIGEN_DEBUG_ALIGNED_LOAD 342 Packet *vfrom; 343 vfrom = (Packet *) from; 344 return vfrom->v4i; 345 } 346 347 template<> EIGEN_STRONG_INLINE Packet4f pload<Packet4f>(const float* from) 348 { 349 // FIXME: No intrinsic yet 350 EIGEN_DEBUG_ALIGNED_LOAD 351 Packet4f vfrom; 352 vfrom.v4f[0] = vec_ld2f(&from[0]); 353 vfrom.v4f[1] = vec_ld2f(&from[2]); 354 return vfrom; 355 } 356 357 template<> EIGEN_STRONG_INLINE Packet2d pload<Packet2d>(const double* from) 358 { 359 // FIXME: No intrinsic yet 360 EIGEN_DEBUG_ALIGNED_LOAD 361 Packet *vfrom; 362 vfrom = (Packet *) from; 363 return vfrom->v2d; 364 } 365 366 template<> EIGEN_STRONG_INLINE void pstore<int>(int* to, const Packet4i& from) 367 { 368 // FIXME: No intrinsic yet 369 EIGEN_DEBUG_ALIGNED_STORE 370 Packet *vto; 371 vto = (Packet *) to; 372 vto->v4i = from; 373 } 374 375 template<> EIGEN_STRONG_INLINE void pstore<float>(float* to, const Packet4f& from) 376 { 377 // FIXME: No intrinsic yet 378 EIGEN_DEBUG_ALIGNED_STORE 379 vec_st2f(from.v4f[0], &to[0]); 380 vec_st2f(from.v4f[1], &to[2]); 381 } 382 383 384 template<> EIGEN_STRONG_INLINE void pstore<double>(double* to, const Packet2d& from) 385 { 386 // FIXME: No intrinsic yet 387 EIGEN_DEBUG_ALIGNED_STORE 388 Packet *vto; 389 vto = (Packet *) to; 390 vto->v2d = from; 391 } 392 393 template<> EIGEN_STRONG_INLINE Packet4i pset1<Packet4i>(const int& from) 394 { 395 return vec_splats(from); 396 } 397 template<> EIGEN_STRONG_INLINE Packet2d pset1<Packet2d>(const double& from) { 398 return vec_splats(from); 399 } 400 template<> EIGEN_STRONG_INLINE Packet4f pset1<Packet4f>(const float& from) 401 { 402 Packet4f to; 403 to.v4f[0] = pset1<Packet2d>(static_cast<const double&>(from)); 404 to.v4f[1] = to.v4f[0]; 405 return to; 406 } 407 408 template<> EIGEN_STRONG_INLINE void 409 pbroadcast4<Packet4i>(const int *a, 410 Packet4i& a0, Packet4i& a1, Packet4i& a2, Packet4i& a3) 411 { 412 a3 = pload<Packet4i>(a); 413 a0 = vec_splat(a3, 0); 414 a1 = vec_splat(a3, 1); 415 a2 = vec_splat(a3, 2); 416 a3 = vec_splat(a3, 3); 417 } 418 419 template<> EIGEN_STRONG_INLINE void 420 pbroadcast4<Packet4f>(const float *a, 421 Packet4f& a0, Packet4f& a1, Packet4f& a2, Packet4f& a3) 422 { 423 a3 = pload<Packet4f>(a); 424 a0 = vec_splat_packet4f<0>(a3); 425 a1 = vec_splat_packet4f<1>(a3); 426 a2 = vec_splat_packet4f<2>(a3); 427 a3 = vec_splat_packet4f<3>(a3); 428 } 429 430 template<> EIGEN_STRONG_INLINE void 431 pbroadcast4<Packet2d>(const double *a, 432 Packet2d& a0, Packet2d& a1, Packet2d& a2, Packet2d& a3) 433 { 434 a1 = pload<Packet2d>(a); 435 a0 = vec_splat(a1, 0); 436 a1 = vec_splat(a1, 1); 437 a3 = pload<Packet2d>(a+2); 438 a2 = vec_splat(a3, 0); 439 a3 = vec_splat(a3, 1); 440 } 441 442 template<> EIGEN_DEVICE_FUNC inline Packet4i pgather<int, Packet4i>(const int* from, Index stride) 443 { 444 int EIGEN_ALIGN16 ai[4]; 445 ai[0] = from[0*stride]; 446 ai[1] = from[1*stride]; 447 ai[2] = from[2*stride]; 448 ai[3] = from[3*stride]; 449 return pload<Packet4i>(ai); 450 } 451 452 template<> EIGEN_DEVICE_FUNC inline Packet4f pgather<float, Packet4f>(const float* from, Index stride) 453 { 454 float EIGEN_ALIGN16 ai[4]; 455 ai[0] = from[0*stride]; 456 ai[1] = from[1*stride]; 457 ai[2] = from[2*stride]; 458 ai[3] = from[3*stride]; 459 return pload<Packet4f>(ai); 460 } 461 462 template<> EIGEN_DEVICE_FUNC inline Packet2d pgather<double, Packet2d>(const double* from, Index stride) 463 { 464 double EIGEN_ALIGN16 af[2]; 465 af[0] = from[0*stride]; 466 af[1] = from[1*stride]; 467 return pload<Packet2d>(af); 468 } 469 470 template<> EIGEN_DEVICE_FUNC inline void pscatter<int, Packet4i>(int* to, const Packet4i& from, Index stride) 471 { 472 int EIGEN_ALIGN16 ai[4]; 473 pstore<int>((int *)ai, from); 474 to[0*stride] = ai[0]; 475 to[1*stride] = ai[1]; 476 to[2*stride] = ai[2]; 477 to[3*stride] = ai[3]; 478 } 479 480 template<> EIGEN_DEVICE_FUNC inline void pscatter<float, Packet4f>(float* to, const Packet4f& from, Index stride) 481 { 482 float EIGEN_ALIGN16 ai[4]; 483 pstore<float>((float *)ai, from); 484 to[0*stride] = ai[0]; 485 to[1*stride] = ai[1]; 486 to[2*stride] = ai[2]; 487 to[3*stride] = ai[3]; 488 } 489 490 template<> EIGEN_DEVICE_FUNC inline void pscatter<double, Packet2d>(double* to, const Packet2d& from, Index stride) 491 { 492 double EIGEN_ALIGN16 af[2]; 493 pstore<double>(af, from); 494 to[0*stride] = af[0]; 495 to[1*stride] = af[1]; 496 } 497 498 template<> EIGEN_STRONG_INLINE Packet4i padd<Packet4i>(const Packet4i& a, const Packet4i& b) { return (a + b); } 499 template<> EIGEN_STRONG_INLINE Packet4f padd<Packet4f>(const Packet4f& a, const Packet4f& b) 500 { 501 Packet4f c; 502 c.v4f[0] = a.v4f[0] + b.v4f[0]; 503 c.v4f[1] = a.v4f[1] + b.v4f[1]; 504 return c; 505 } 506 template<> EIGEN_STRONG_INLINE Packet2d padd<Packet2d>(const Packet2d& a, const Packet2d& b) { return (a + b); } 507 508 template<> EIGEN_STRONG_INLINE Packet4i psub<Packet4i>(const Packet4i& a, const Packet4i& b) { return (a - b); } 509 template<> EIGEN_STRONG_INLINE Packet4f psub<Packet4f>(const Packet4f& a, const Packet4f& b) 510 { 511 Packet4f c; 512 c.v4f[0] = a.v4f[0] - b.v4f[0]; 513 c.v4f[1] = a.v4f[1] - b.v4f[1]; 514 return c; 515 } 516 template<> EIGEN_STRONG_INLINE Packet2d psub<Packet2d>(const Packet2d& a, const Packet2d& b) { return (a - b); } 517 518 template<> EIGEN_STRONG_INLINE Packet4i pmul<Packet4i>(const Packet4i& a, const Packet4i& b) { return (a * b); } 519 template<> EIGEN_STRONG_INLINE Packet4f pmul<Packet4f>(const Packet4f& a, const Packet4f& b) 520 { 521 Packet4f c; 522 c.v4f[0] = a.v4f[0] * b.v4f[0]; 523 c.v4f[1] = a.v4f[1] * b.v4f[1]; 524 return c; 525 } 526 template<> EIGEN_STRONG_INLINE Packet2d pmul<Packet2d>(const Packet2d& a, const Packet2d& b) { return (a * b); } 527 528 template<> EIGEN_STRONG_INLINE Packet4i pdiv<Packet4i>(const Packet4i& a, const Packet4i& b) { return (a / b); } 529 template<> EIGEN_STRONG_INLINE Packet4f pdiv<Packet4f>(const Packet4f& a, const Packet4f& b) 530 { 531 Packet4f c; 532 c.v4f[0] = a.v4f[0] / b.v4f[0]; 533 c.v4f[1] = a.v4f[1] / b.v4f[1]; 534 return c; 535 } 536 template<> EIGEN_STRONG_INLINE Packet2d pdiv<Packet2d>(const Packet2d& a, const Packet2d& b) { return (a / b); } 537 538 template<> EIGEN_STRONG_INLINE Packet4i pnegate(const Packet4i& a) { return (-a); } 539 template<> EIGEN_STRONG_INLINE Packet4f pnegate(const Packet4f& a) 540 { 541 Packet4f c; 542 c.v4f[0] = -a.v4f[0]; 543 c.v4f[1] = -a.v4f[1]; 544 return c; 545 } 546 template<> EIGEN_STRONG_INLINE Packet2d pnegate(const Packet2d& a) { return (-a); } 547 548 template<> EIGEN_STRONG_INLINE Packet4i pconj(const Packet4i& a) { return a; } 549 template<> EIGEN_STRONG_INLINE Packet4f pconj(const Packet4f& a) { return a; } 550 template<> EIGEN_STRONG_INLINE Packet2d pconj(const Packet2d& a) { return a; } 551 552 template<> EIGEN_STRONG_INLINE Packet4i pmadd(const Packet4i& a, const Packet4i& b, const Packet4i& c) { return padd<Packet4i>(pmul<Packet4i>(a, b), c); } 553 template<> EIGEN_STRONG_INLINE Packet4f pmadd(const Packet4f& a, const Packet4f& b, const Packet4f& c) 554 { 555 Packet4f res; 556 res.v4f[0] = vec_madd(a.v4f[0], b.v4f[0], c.v4f[0]); 557 res.v4f[1] = vec_madd(a.v4f[1], b.v4f[1], c.v4f[1]); 558 return res; 559 } 560 template<> EIGEN_STRONG_INLINE Packet2d pmadd(const Packet2d& a, const Packet2d& b, const Packet2d& c) { return vec_madd(a, b, c); } 561 562 template<> EIGEN_STRONG_INLINE Packet4i plset<Packet4i>(const int& a) { return padd<Packet4i>(pset1<Packet4i>(a), p4i_COUNTDOWN); } 563 template<> EIGEN_STRONG_INLINE Packet4f plset<Packet4f>(const float& a) { return padd<Packet4f>(pset1<Packet4f>(a), p4f_COUNTDOWN); } 564 template<> EIGEN_STRONG_INLINE Packet2d plset<Packet2d>(const double& a) { return padd<Packet2d>(pset1<Packet2d>(a), p2d_COUNTDOWN); } 565 566 template<> EIGEN_STRONG_INLINE Packet4i pmin<Packet4i>(const Packet4i& a, const Packet4i& b) { return vec_min(a, b); } 567 template<> EIGEN_STRONG_INLINE Packet2d pmin<Packet2d>(const Packet2d& a, const Packet2d& b) { return vec_min(a, b); } 568 template<> EIGEN_STRONG_INLINE Packet4f pmin<Packet4f>(const Packet4f& a, const Packet4f& b) 569 { 570 Packet4f res; 571 res.v4f[0] = pmin(a.v4f[0], b.v4f[0]); 572 res.v4f[1] = pmin(a.v4f[1], b.v4f[1]); 573 return res; 574 } 575 576 template<> EIGEN_STRONG_INLINE Packet4i pmax<Packet4i>(const Packet4i& a, const Packet4i& b) { return vec_max(a, b); } 577 template<> EIGEN_STRONG_INLINE Packet2d pmax<Packet2d>(const Packet2d& a, const Packet2d& b) { return vec_max(a, b); } 578 template<> EIGEN_STRONG_INLINE Packet4f pmax<Packet4f>(const Packet4f& a, const Packet4f& b) 579 { 580 Packet4f res; 581 res.v4f[0] = pmax(a.v4f[0], b.v4f[0]); 582 res.v4f[1] = pmax(a.v4f[1], b.v4f[1]); 583 return res; 584 } 585 586 template<> EIGEN_STRONG_INLINE Packet4i pand<Packet4i>(const Packet4i& a, const Packet4i& b) { return vec_and(a, b); } 587 template<> EIGEN_STRONG_INLINE Packet2d pand<Packet2d>(const Packet2d& a, const Packet2d& b) { return vec_and(a, b); } 588 template<> EIGEN_STRONG_INLINE Packet4f pand<Packet4f>(const Packet4f& a, const Packet4f& b) 589 { 590 Packet4f res; 591 res.v4f[0] = pand(a.v4f[0], b.v4f[0]); 592 res.v4f[1] = pand(a.v4f[1], b.v4f[1]); 593 return res; 594 } 595 596 template<> EIGEN_STRONG_INLINE Packet4i por<Packet4i>(const Packet4i& a, const Packet4i& b) { return vec_or(a, b); } 597 template<> EIGEN_STRONG_INLINE Packet2d por<Packet2d>(const Packet2d& a, const Packet2d& b) { return vec_or(a, b); } 598 template<> EIGEN_STRONG_INLINE Packet4f por<Packet4f>(const Packet4f& a, const Packet4f& b) 599 { 600 Packet4f res; 601 res.v4f[0] = pand(a.v4f[0], b.v4f[0]); 602 res.v4f[1] = pand(a.v4f[1], b.v4f[1]); 603 return res; 604 } 605 606 template<> EIGEN_STRONG_INLINE Packet4i pxor<Packet4i>(const Packet4i& a, const Packet4i& b) { return vec_xor(a, b); } 607 template<> EIGEN_STRONG_INLINE Packet2d pxor<Packet2d>(const Packet2d& a, const Packet2d& b) { return vec_xor(a, b); } 608 template<> EIGEN_STRONG_INLINE Packet4f pxor<Packet4f>(const Packet4f& a, const Packet4f& b) 609 { 610 Packet4f res; 611 res.v4f[0] = pand(a.v4f[0], b.v4f[0]); 612 res.v4f[1] = pand(a.v4f[1], b.v4f[1]); 613 return res; 614 } 615 616 template<> EIGEN_STRONG_INLINE Packet4i pandnot<Packet4i>(const Packet4i& a, const Packet4i& b) { return pand<Packet4i>(a, vec_nor(b, b)); } 617 template<> EIGEN_STRONG_INLINE Packet2d pandnot<Packet2d>(const Packet2d& a, const Packet2d& b) { return vec_and(a, vec_nor(b, b)); } 618 template<> EIGEN_STRONG_INLINE Packet4f pandnot<Packet4f>(const Packet4f& a, const Packet4f& b) 619 { 620 Packet4f res; 621 res.v4f[0] = pandnot(a.v4f[0], b.v4f[0]); 622 res.v4f[1] = pandnot(a.v4f[1], b.v4f[1]); 623 return res; 624 } 625 626 template<> EIGEN_STRONG_INLINE Packet4f pround<Packet4f>(const Packet4f& a) 627 { 628 Packet4f res; 629 res.v4f[0] = vec_round(a.v4f[0]); 630 res.v4f[1] = vec_round(a.v4f[1]); 631 return res; 632 } 633 template<> EIGEN_STRONG_INLINE Packet2d pround<Packet2d>(const Packet2d& a) { return vec_round(a); } 634 template<> EIGEN_STRONG_INLINE Packet4f pceil<Packet4f>(const Packet4f& a) 635 { 636 Packet4f res; 637 res.v4f[0] = vec_ceil(a.v4f[0]); 638 res.v4f[1] = vec_ceil(a.v4f[1]); 639 return res; 640 } 641 template<> EIGEN_STRONG_INLINE Packet2d pceil<Packet2d>(const Packet2d& a) { return vec_ceil(a); } 642 template<> EIGEN_STRONG_INLINE Packet4f pfloor<Packet4f>(const Packet4f& a) 643 { 644 Packet4f res; 645 res.v4f[0] = vec_floor(a.v4f[0]); 646 res.v4f[1] = vec_floor(a.v4f[1]); 647 return res; 648 } 649 template<> EIGEN_STRONG_INLINE Packet2d pfloor<Packet2d>(const Packet2d& a) { return vec_floor(a); } 650 651 template<> EIGEN_STRONG_INLINE Packet4i ploadu<Packet4i>(const int* from) { return pload<Packet4i>(from); } 652 template<> EIGEN_STRONG_INLINE Packet4f ploadu<Packet4f>(const float* from) { return pload<Packet4f>(from); } 653 template<> EIGEN_STRONG_INLINE Packet2d ploadu<Packet2d>(const double* from) { return pload<Packet2d>(from); } 654 655 656 template<> EIGEN_STRONG_INLINE Packet4i ploaddup<Packet4i>(const int* from) 657 { 658 Packet4i p = pload<Packet4i>(from); 659 return vec_perm(p, p, p16uc_DUPLICATE32_HI); 660 } 661 662 template<> EIGEN_STRONG_INLINE Packet4f ploaddup<Packet4f>(const float* from) 663 { 664 Packet4f p = pload<Packet4f>(from); 665 p.v4f[1] = vec_splat(p.v4f[0], 1); 666 p.v4f[0] = vec_splat(p.v4f[0], 0); 667 return p; 668 } 669 670 template<> EIGEN_STRONG_INLINE Packet2d ploaddup<Packet2d>(const double* from) 671 { 672 Packet2d p = pload<Packet2d>(from); 673 return vec_perm(p, p, p16uc_PSET64_HI); 674 } 675 676 template<> EIGEN_STRONG_INLINE void pstoreu<int>(int* to, const Packet4i& from) { pstore<int>(to, from); } 677 template<> EIGEN_STRONG_INLINE void pstoreu<float>(float* to, const Packet4f& from) { pstore<float>(to, from); } 678 template<> EIGEN_STRONG_INLINE void pstoreu<double>(double* to, const Packet2d& from) { pstore<double>(to, from); } 679 680 template<> EIGEN_STRONG_INLINE void prefetch<int>(const int* addr) { EIGEN_ZVECTOR_PREFETCH(addr); } 681 template<> EIGEN_STRONG_INLINE void prefetch<float>(const float* addr) { EIGEN_ZVECTOR_PREFETCH(addr); } 682 template<> EIGEN_STRONG_INLINE void prefetch<double>(const double* addr) { EIGEN_ZVECTOR_PREFETCH(addr); } 683 684 template<> EIGEN_STRONG_INLINE int pfirst<Packet4i>(const Packet4i& a) { int EIGEN_ALIGN16 x[4]; pstore(x, a); return x[0]; } 685 template<> EIGEN_STRONG_INLINE float pfirst<Packet4f>(const Packet4f& a) { float EIGEN_ALIGN16 x[2]; vec_st2f(a.v4f[0], &x[0]); return x[0]; } 686 template<> EIGEN_STRONG_INLINE double pfirst<Packet2d>(const Packet2d& a) { double EIGEN_ALIGN16 x[2]; pstore(x, a); return x[0]; } 687 688 template<> EIGEN_STRONG_INLINE Packet4i preverse(const Packet4i& a) 689 { 690 return reinterpret_cast<Packet4i>(vec_perm(reinterpret_cast<Packet16uc>(a), reinterpret_cast<Packet16uc>(a), p16uc_REVERSE32)); 691 } 692 693 template<> EIGEN_STRONG_INLINE Packet2d preverse(const Packet2d& a) 694 { 695 return reinterpret_cast<Packet2d>(vec_perm(reinterpret_cast<Packet16uc>(a), reinterpret_cast<Packet16uc>(a), p16uc_REVERSE64)); 696 } 697 698 template<> EIGEN_STRONG_INLINE Packet4f preverse(const Packet4f& a) 699 { 700 Packet4f rev; 701 rev.v4f[0] = preverse<Packet2d>(a.v4f[1]); 702 rev.v4f[1] = preverse<Packet2d>(a.v4f[0]); 703 return rev; 704 } 705 706 template<> EIGEN_STRONG_INLINE Packet4i pabs<Packet4i>(const Packet4i& a) { return vec_abs(a); } 707 template<> EIGEN_STRONG_INLINE Packet2d pabs<Packet2d>(const Packet2d& a) { return vec_abs(a); } 708 template<> EIGEN_STRONG_INLINE Packet4f pabs<Packet4f>(const Packet4f& a) 709 { 710 Packet4f res; 711 res.v4f[0] = pabs(a.v4f[0]); 712 res.v4f[1] = pabs(a.v4f[1]); 713 return res; 714 } 715 716 template<> EIGEN_STRONG_INLINE int predux<Packet4i>(const Packet4i& a) 717 { 718 Packet4i b, sum; 719 b = vec_sld(a, a, 8); 720 sum = padd<Packet4i>(a, b); 721 b = vec_sld(sum, sum, 4); 722 sum = padd<Packet4i>(sum, b); 723 return pfirst(sum); 724 } 725 726 template<> EIGEN_STRONG_INLINE double predux<Packet2d>(const Packet2d& a) 727 { 728 Packet2d b, sum; 729 b = reinterpret_cast<Packet2d>(vec_sld(reinterpret_cast<Packet4i>(a), reinterpret_cast<Packet4i>(a), 8)); 730 sum = padd<Packet2d>(a, b); 731 return pfirst(sum); 732 } 733 template<> EIGEN_STRONG_INLINE float predux<Packet4f>(const Packet4f& a) 734 { 735 Packet2d sum; 736 sum = padd<Packet2d>(a.v4f[0], a.v4f[1]); 737 double first = predux<Packet2d>(sum); 738 return static_cast<float>(first); 739 } 740 741 template<> EIGEN_STRONG_INLINE Packet4i preduxp<Packet4i>(const Packet4i* vecs) 742 { 743 Packet4i v[4], sum[4]; 744 745 // It's easier and faster to transpose then add as columns 746 // Check: http://www.freevec.org/function/matrix_4x4_transpose_floats for explanation 747 // Do the transpose, first set of moves 748 v[0] = vec_mergeh(vecs[0], vecs[2]); 749 v[1] = vec_mergel(vecs[0], vecs[2]); 750 v[2] = vec_mergeh(vecs[1], vecs[3]); 751 v[3] = vec_mergel(vecs[1], vecs[3]); 752 // Get the resulting vectors 753 sum[0] = vec_mergeh(v[0], v[2]); 754 sum[1] = vec_mergel(v[0], v[2]); 755 sum[2] = vec_mergeh(v[1], v[3]); 756 sum[3] = vec_mergel(v[1], v[3]); 757 758 // Now do the summation: 759 // Lines 0+1 760 sum[0] = padd<Packet4i>(sum[0], sum[1]); 761 // Lines 2+3 762 sum[1] = padd<Packet4i>(sum[2], sum[3]); 763 // Add the results 764 sum[0] = padd<Packet4i>(sum[0], sum[1]); 765 766 return sum[0]; 767 } 768 769 template<> EIGEN_STRONG_INLINE Packet2d preduxp<Packet2d>(const Packet2d* vecs) 770 { 771 Packet2d v[2], sum; 772 v[0] = padd<Packet2d>(vecs[0], reinterpret_cast<Packet2d>(vec_sld(reinterpret_cast<Packet4ui>(vecs[0]), reinterpret_cast<Packet4ui>(vecs[0]), 8))); 773 v[1] = padd<Packet2d>(vecs[1], reinterpret_cast<Packet2d>(vec_sld(reinterpret_cast<Packet4ui>(vecs[1]), reinterpret_cast<Packet4ui>(vecs[1]), 8))); 774 775 sum = reinterpret_cast<Packet2d>(vec_sld(reinterpret_cast<Packet4ui>(v[0]), reinterpret_cast<Packet4ui>(v[1]), 8)); 776 777 return sum; 778 } 779 780 template<> EIGEN_STRONG_INLINE Packet4f preduxp<Packet4f>(const Packet4f* vecs) 781 { 782 PacketBlock<Packet4f,4> transpose; 783 transpose.packet[0] = vecs[0]; 784 transpose.packet[1] = vecs[1]; 785 transpose.packet[2] = vecs[2]; 786 transpose.packet[3] = vecs[3]; 787 ptranspose(transpose); 788 789 Packet4f sum = padd(transpose.packet[0], transpose.packet[1]); 790 sum = padd(sum, transpose.packet[2]); 791 sum = padd(sum, transpose.packet[3]); 792 return sum; 793 } 794 795 // Other reduction functions: 796 // mul 797 template<> EIGEN_STRONG_INLINE int predux_mul<Packet4i>(const Packet4i& a) 798 { 799 EIGEN_ALIGN16 int aux[4]; 800 pstore(aux, a); 801 return aux[0] * aux[1] * aux[2] * aux[3]; 802 } 803 804 template<> EIGEN_STRONG_INLINE double predux_mul<Packet2d>(const Packet2d& a) 805 { 806 return pfirst(pmul(a, reinterpret_cast<Packet2d>(vec_sld(reinterpret_cast<Packet4i>(a), reinterpret_cast<Packet4i>(a), 8)))); 807 } 808 809 template<> EIGEN_STRONG_INLINE float predux_mul<Packet4f>(const Packet4f& a) 810 { 811 // Return predux_mul<Packet2d> of the subvectors product 812 return static_cast<float>(pfirst(predux_mul(pmul(a.v4f[0], a.v4f[1])))); 813 } 814 815 // min 816 template<> EIGEN_STRONG_INLINE int predux_min<Packet4i>(const Packet4i& a) 817 { 818 Packet4i b, res; 819 b = pmin<Packet4i>(a, vec_sld(a, a, 8)); 820 res = pmin<Packet4i>(b, vec_sld(b, b, 4)); 821 return pfirst(res); 822 } 823 824 template<> EIGEN_STRONG_INLINE double predux_min<Packet2d>(const Packet2d& a) 825 { 826 return pfirst(pmin<Packet2d>(a, reinterpret_cast<Packet2d>(vec_sld(reinterpret_cast<Packet4i>(a), reinterpret_cast<Packet4i>(a), 8)))); 827 } 828 829 template<> EIGEN_STRONG_INLINE float predux_min<Packet4f>(const Packet4f& a) 830 { 831 Packet2d b, res; 832 b = pmin<Packet2d>(a.v4f[0], a.v4f[1]); 833 res = pmin<Packet2d>(b, reinterpret_cast<Packet2d>(vec_sld(reinterpret_cast<Packet4i>(b), reinterpret_cast<Packet4i>(b), 8))); 834 return static_cast<float>(pfirst(res)); 835 } 836 837 // max 838 template<> EIGEN_STRONG_INLINE int predux_max<Packet4i>(const Packet4i& a) 839 { 840 Packet4i b, res; 841 b = pmax<Packet4i>(a, vec_sld(a, a, 8)); 842 res = pmax<Packet4i>(b, vec_sld(b, b, 4)); 843 return pfirst(res); 844 } 845 846 // max 847 template<> EIGEN_STRONG_INLINE double predux_max<Packet2d>(const Packet2d& a) 848 { 849 return pfirst(pmax<Packet2d>(a, reinterpret_cast<Packet2d>(vec_sld(reinterpret_cast<Packet4i>(a), reinterpret_cast<Packet4i>(a), 8)))); 850 } 851 852 template<> EIGEN_STRONG_INLINE float predux_max<Packet4f>(const Packet4f& a) 853 { 854 Packet2d b, res; 855 b = pmax<Packet2d>(a.v4f[0], a.v4f[1]); 856 res = pmax<Packet2d>(b, reinterpret_cast<Packet2d>(vec_sld(reinterpret_cast<Packet4i>(b), reinterpret_cast<Packet4i>(b), 8))); 857 return static_cast<float>(pfirst(res)); 858 } 859 860 EIGEN_DEVICE_FUNC inline void 861 ptranspose(PacketBlock<Packet4i,4>& kernel) { 862 Packet4i t0 = vec_mergeh(kernel.packet[0], kernel.packet[2]); 863 Packet4i t1 = vec_mergel(kernel.packet[0], kernel.packet[2]); 864 Packet4i t2 = vec_mergeh(kernel.packet[1], kernel.packet[3]); 865 Packet4i t3 = vec_mergel(kernel.packet[1], kernel.packet[3]); 866 kernel.packet[0] = vec_mergeh(t0, t2); 867 kernel.packet[1] = vec_mergel(t0, t2); 868 kernel.packet[2] = vec_mergeh(t1, t3); 869 kernel.packet[3] = vec_mergel(t1, t3); 870 } 871 872 EIGEN_DEVICE_FUNC inline void 873 ptranspose(PacketBlock<Packet2d,2>& kernel) { 874 Packet2d t0 = vec_perm(kernel.packet[0], kernel.packet[1], p16uc_TRANSPOSE64_HI); 875 Packet2d t1 = vec_perm(kernel.packet[0], kernel.packet[1], p16uc_TRANSPOSE64_LO); 876 kernel.packet[0] = t0; 877 kernel.packet[1] = t1; 878 } 879 880 /* Split the Packet4f PacketBlock into 4 Packet2d PacketBlocks and transpose each one 881 */ 882 EIGEN_DEVICE_FUNC inline void 883 ptranspose(PacketBlock<Packet4f,4>& kernel) { 884 PacketBlock<Packet2d,2> t0,t1,t2,t3; 885 // copy top-left 2x2 Packet2d block 886 t0.packet[0] = kernel.packet[0].v4f[0]; 887 t0.packet[1] = kernel.packet[1].v4f[0]; 888 889 // copy top-right 2x2 Packet2d block 890 t1.packet[0] = kernel.packet[0].v4f[1]; 891 t1.packet[1] = kernel.packet[1].v4f[1]; 892 893 // copy bottom-left 2x2 Packet2d block 894 t2.packet[0] = kernel.packet[2].v4f[0]; 895 t2.packet[1] = kernel.packet[3].v4f[0]; 896 897 // copy bottom-right 2x2 Packet2d block 898 t3.packet[0] = kernel.packet[2].v4f[1]; 899 t3.packet[1] = kernel.packet[3].v4f[1]; 900 901 // Transpose all 2x2 blocks 902 ptranspose(t0); 903 ptranspose(t1); 904 ptranspose(t2); 905 ptranspose(t3); 906 907 // Copy back transposed blocks, but exchange t1 and t2 due to transposition 908 kernel.packet[0].v4f[0] = t0.packet[0]; 909 kernel.packet[0].v4f[1] = t2.packet[0]; 910 kernel.packet[1].v4f[0] = t0.packet[1]; 911 kernel.packet[1].v4f[1] = t2.packet[1]; 912 kernel.packet[2].v4f[0] = t1.packet[0]; 913 kernel.packet[2].v4f[1] = t3.packet[0]; 914 kernel.packet[3].v4f[0] = t1.packet[1]; 915 kernel.packet[3].v4f[1] = t3.packet[1]; 916 } 917 918 template<> EIGEN_STRONG_INLINE Packet4i pblend(const Selector<4>& ifPacket, const Packet4i& thenPacket, const Packet4i& elsePacket) { 919 Packet4ui select = { ifPacket.select[0], ifPacket.select[1], ifPacket.select[2], ifPacket.select[3] }; 920 Packet4ui mask = vec_cmpeq(select, reinterpret_cast<Packet4ui>(p4i_ONE)); 921 return vec_sel(elsePacket, thenPacket, mask); 922 } 923 924 template<> EIGEN_STRONG_INLINE Packet4f pblend(const Selector<4>& ifPacket, const Packet4f& thenPacket, const Packet4f& elsePacket) { 925 Packet2ul select_hi = { ifPacket.select[0], ifPacket.select[1] }; 926 Packet2ul select_lo = { ifPacket.select[2], ifPacket.select[3] }; 927 Packet2ul mask_hi = vec_cmpeq(select_hi, reinterpret_cast<Packet2ul>(p2l_ONE)); 928 Packet2ul mask_lo = vec_cmpeq(select_lo, reinterpret_cast<Packet2ul>(p2l_ONE)); 929 Packet4f result; 930 result.v4f[0] = vec_sel(elsePacket.v4f[0], thenPacket.v4f[0], mask_hi); 931 result.v4f[1] = vec_sel(elsePacket.v4f[1], thenPacket.v4f[1], mask_lo); 932 return result; 933 } 934 935 template<> EIGEN_STRONG_INLINE Packet2d pblend(const Selector<2>& ifPacket, const Packet2d& thenPacket, const Packet2d& elsePacket) { 936 Packet2ul select = { ifPacket.select[0], ifPacket.select[1] }; 937 Packet2ul mask = vec_cmpeq(select, reinterpret_cast<Packet2ul>(p2l_ONE)); 938 return vec_sel(elsePacket, thenPacket, mask); 939 } 940 941 } // end namespace internal 942 943 } // end namespace Eigen 944 945 #endif // EIGEN_PACKET_MATH_ZVECTOR_H 946