1 // This file is part of Eigen, a lightweight C++ template library 2 // for linear algebra. 3 // 4 // Copyright (C) 2016 Konstantinos Margaritis <[email protected]> 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 namespace Eigen { 14 15 namespace internal { 16 17 #ifndef EIGEN_CACHEFRIENDLY_PRODUCT_THRESHOLD 18 #define EIGEN_CACHEFRIENDLY_PRODUCT_THRESHOLD 16 19 #endif 20 21 #ifndef EIGEN_HAS_SINGLE_INSTRUCTION_MADD 22 #define EIGEN_HAS_SINGLE_INSTRUCTION_MADD 23 #endif 24 25 #ifndef EIGEN_ARCH_DEFAULT_NUMBER_OF_REGISTERS 26 #define EIGEN_ARCH_DEFAULT_NUMBER_OF_REGISTERS 32 27 #endif 28 29 typedef __vector int Packet4i; 30 typedef __vector unsigned int Packet4ui; 31 typedef __vector __bool int Packet4bi; 32 typedef __vector short int Packet8i; 33 typedef __vector unsigned char Packet16uc; 34 typedef __vector double Packet2d; 35 typedef __vector unsigned long long Packet2ul; 36 typedef __vector long long Packet2l; 37 38 // Z14 has builtin support for float vectors 39 #if !defined(__ARCH__) || (defined(__ARCH__) && __ARCH__ >= 12) 40 typedef __vector float Packet4f; 41 #else 42 typedef struct { 43 Packet2d v4f[2]; 44 } Packet4f; 45 #endif 46 47 typedef union { 48 numext::int32_t i[4]; 49 numext::uint32_t ui[4]; 50 numext::int64_t l[2]; 51 numext::uint64_t ul[2]; 52 double d[2]; 53 float f[4]; 54 Packet4i v4i; 55 Packet4ui v4ui; 56 Packet2l v2l; 57 Packet2ul v2ul; 58 Packet2d v2d; 59 #if !defined(__ARCH__) || (defined(__ARCH__) && __ARCH__ >= 12) 60 Packet4f v4f; 61 #endif 62 } Packet; 63 64 // We don't want to write the same code all the time, but we need to reuse the constants 65 // and it doesn't really work to declare them global, so we define macros instead 66 67 #define _EIGEN_DECLARE_CONST_FAST_Packet4i(NAME,X) \ 68 Packet4i p4i_##NAME = reinterpret_cast<Packet4i>(vec_splat_s32(X)) 69 70 #define _EIGEN_DECLARE_CONST_FAST_Packet2d(NAME,X) \ 71 Packet2d p2d_##NAME = reinterpret_cast<Packet2d>(vec_splat_s64(X)) 72 73 #define _EIGEN_DECLARE_CONST_FAST_Packet2l(NAME,X) \ 74 Packet2l p2l_##NAME = reinterpret_cast<Packet2l>(vec_splat_s64(X)) 75 76 #define _EIGEN_DECLARE_CONST_Packet4i(NAME,X) \ 77 Packet4i p4i_##NAME = pset1<Packet4i>(X) 78 79 #define _EIGEN_DECLARE_CONST_Packet2d(NAME,X) \ 80 Packet2d p2d_##NAME = pset1<Packet2d>(X) 81 82 #define _EIGEN_DECLARE_CONST_Packet2l(NAME,X) \ 83 Packet2l p2l_##NAME = pset1<Packet2l>(X) 84 85 // These constants are endian-agnostic 86 static _EIGEN_DECLARE_CONST_FAST_Packet4i(ZERO, 0); //{ 0, 0, 0, 0,} 87 static _EIGEN_DECLARE_CONST_FAST_Packet4i(ONE, 1); //{ 1, 1, 1, 1} 88 89 static _EIGEN_DECLARE_CONST_FAST_Packet2d(ZERO, 0); 90 static _EIGEN_DECLARE_CONST_FAST_Packet2l(ZERO, 0); 91 static _EIGEN_DECLARE_CONST_FAST_Packet2l(ONE, 1); 92 93 static Packet2d p2d_ONE = { 1.0, 1.0 }; 94 static Packet2d p2d_ZERO_ = { numext::bit_cast<double>0x8000000000000000ull), 95 numext::bit_cast<double>0x8000000000000000ull) }; 96 97 #if !defined(__ARCH__) || (defined(__ARCH__) && __ARCH__ >= 12) 98 #define _EIGEN_DECLARE_CONST_FAST_Packet4f(NAME,X) \ 99 Packet4f p4f_##NAME = reinterpret_cast<Packet4f>(vec_splat_s32(X)) 100 101 #define _EIGEN_DECLARE_CONST_Packet4f(NAME,X) \ 102 Packet4f p4f_##NAME = pset1<Packet4f>(X) 103 104 #define _EIGEN_DECLARE_CONST_Packet4f_FROM_INT(NAME,X) \ 105 const Packet4f p4f_##NAME = reinterpret_cast<Packet4f>(pset1<Packet4i>(X)) 106 107 static _EIGEN_DECLARE_CONST_FAST_Packet4f(ZERO, 0); //{ 0.0, 0.0, 0.0, 0.0} 108 static _EIGEN_DECLARE_CONST_FAST_Packet4i(MINUS1,-1); //{ -1, -1, -1, -1} 109 static Packet4f p4f_MZERO = { 0x80000000, 0x80000000, 0x80000000, 0x80000000}; 110 #endif 111 112 static Packet4i p4i_COUNTDOWN = { 0, 1, 2, 3 }; 113 static Packet4f p4f_COUNTDOWN = { 0.0, 1.0, 2.0, 3.0 }; 114 static Packet2d p2d_COUNTDOWN = reinterpret_cast<Packet2d>(vec_sld(reinterpret_cast<Packet16uc>(p2d_ZERO), reinterpret_cast<Packet16uc>(p2d_ONE), 8)); 115 116 static Packet16uc p16uc_PSET64_HI = { 0,1,2,3, 4,5,6,7, 0,1,2,3, 4,5,6,7 }; 117 static Packet16uc p16uc_DUPLICATE32_HI = { 0,1,2,3, 0,1,2,3, 4,5,6,7, 4,5,6,7 }; 118 119 // Mask alignment 120 #define _EIGEN_MASK_ALIGNMENT 0xfffffffffffffff0 121 122 #define _EIGEN_ALIGNED_PTR(x) ((std::ptrdiff_t)(x) & _EIGEN_MASK_ALIGNMENT) 123 124 // Handle endianness properly while loading constants 125 // Define global static constants: 126 127 static Packet16uc p16uc_FORWARD = { 0,1,2,3, 4,5,6,7, 8,9,10,11, 12,13,14,15 }; 128 static Packet16uc p16uc_REVERSE32 = { 12,13,14,15, 8,9,10,11, 4,5,6,7, 0,1,2,3 }; 129 static Packet16uc p16uc_REVERSE64 = { 8,9,10,11, 12,13,14,15, 0,1,2,3, 4,5,6,7 }; 130 131 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 }; 132 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 }; 133 /*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}; 134 135 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 };*/ 136 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 }; 137 /*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}; 138 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};*/ 139 static Packet16uc p16uc_TRANSPOSE64_HI = { 0,1,2,3, 4,5,6,7, 16,17,18,19, 20,21,22,23}; 140 static Packet16uc p16uc_TRANSPOSE64_LO = { 8,9,10,11, 12,13,14,15, 24,25,26,27, 28,29,30,31}; 141 142 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 }; 143 144 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 }; 145 146 147 #if EIGEN_HAS_BUILTIN(__builtin_prefetch) || EIGEN_COMP_GNUC 148 #define EIGEN_ZVECTOR_PREFETCH(ADDR) __builtin_prefetch(ADDR); 149 #else 150 #define EIGEN_ZVECTOR_PREFETCH(ADDR) asm( " pfd [%[addr]]\n" :: [addr] "r" (ADDR) : "cc" ); 151 #endif 152 153 template<> struct packet_traits<int> : default_packet_traits 154 { 155 typedef Packet4i type; 156 typedef Packet4i half; 157 enum { 158 Vectorizable = 1, 159 AlignedOnScalar = 1, 160 size = 4, 161 HasHalfPacket = 0, 162 163 HasAdd = 1, 164 HasSub = 1, 165 HasMul = 1, 166 HasDiv = 1, 167 HasBlend = 1 168 }; 169 }; 170 171 template <> 172 struct packet_traits<float> : default_packet_traits { 173 typedef Packet4f type; 174 typedef Packet4f half; 175 enum { 176 Vectorizable = 1, 177 AlignedOnScalar = 1, 178 size = 4, 179 HasHalfPacket = 0, 180 181 HasAdd = 1, 182 HasSub = 1, 183 HasMul = 1, 184 HasDiv = 1, 185 HasMin = 1, 186 HasMax = 1, 187 HasAbs = 1, 188 HasSin = 0, 189 HasCos = 0, 190 HasLog = 0, 191 HasExp = 1, 192 HasSqrt = 1, 193 HasRsqrt = 1, 194 HasTanh = 1, 195 HasErf = 1, 196 HasRound = 1, 197 HasFloor = 1, 198 HasCeil = 1, 199 HasNegate = 1, 200 HasBlend = 1 201 }; 202 }; 203 204 template<> struct packet_traits<double> : default_packet_traits 205 { 206 typedef Packet2d type; 207 typedef Packet2d half; 208 enum { 209 Vectorizable = 1, 210 AlignedOnScalar = 1, 211 size=2, 212 HasHalfPacket = 1, 213 214 HasAdd = 1, 215 HasSub = 1, 216 HasMul = 1, 217 HasDiv = 1, 218 HasMin = 1, 219 HasMax = 1, 220 HasAbs = 1, 221 HasSin = 0, 222 HasCos = 0, 223 HasLog = 0, 224 HasExp = 1, 225 HasSqrt = 1, 226 HasRsqrt = 1, 227 HasRound = 1, 228 HasFloor = 1, 229 HasCeil = 1, 230 HasNegate = 1, 231 HasBlend = 1 232 }; 233 }; 234 235 template<> struct unpacket_traits<Packet4i> { typedef int type; enum {size=4, alignment=Aligned16, vectorizable=true, masked_load_available=false, masked_store_available=false}; typedef Packet4i half; }; 236 template<> struct unpacket_traits<Packet4f> { typedef float type; enum {size=4, alignment=Aligned16, vectorizable=true, masked_load_available=false, masked_store_available=false}; typedef Packet4f half; }; 237 template<> struct unpacket_traits<Packet2d> { typedef double type; enum {size=2, alignment=Aligned16, vectorizable=true, masked_load_available=false, masked_store_available=false}; typedef Packet2d half; }; 238 239 /* Forward declaration */ 240 EIGEN_DEVICE_FUNC inline void ptranspose(PacketBlock<Packet4f,4>& kernel); 241 242 inline std::ostream & operator <<(std::ostream & s, const Packet4i & v) 243 { 244 Packet vt; 245 vt.v4i = v; 246 s << vt.i[0] << ", " << vt.i[1] << ", " << vt.i[2] << ", " << vt.i[3]; 247 return s; 248 } 249 250 inline std::ostream & operator <<(std::ostream & s, const Packet4ui & v) 251 { 252 Packet vt; 253 vt.v4ui = v; 254 s << vt.ui[0] << ", " << vt.ui[1] << ", " << vt.ui[2] << ", " << vt.ui[3]; 255 return s; 256 } 257 258 inline std::ostream & operator <<(std::ostream & s, const Packet2l & v) 259 { 260 Packet vt; 261 vt.v2l = v; 262 s << vt.l[0] << ", " << vt.l[1]; 263 return s; 264 } 265 266 inline std::ostream & operator <<(std::ostream & s, const Packet2ul & v) 267 { 268 Packet vt; 269 vt.v2ul = v; 270 s << vt.ul[0] << ", " << vt.ul[1] ; 271 return s; 272 } 273 274 inline std::ostream & operator <<(std::ostream & s, const Packet2d & v) 275 { 276 Packet vt; 277 vt.v2d = v; 278 s << vt.d[0] << ", " << vt.d[1]; 279 return s; 280 } 281 282 #if !defined(__ARCH__) || (defined(__ARCH__) && __ARCH__ >= 12) 283 inline std::ostream & operator <<(std::ostream & s, const Packet4f & v) 284 { 285 Packet vt; 286 vt.v4f = v; 287 s << vt.f[0] << ", " << vt.f[1] << ", " << vt.f[2] << ", " << vt.f[3]; 288 return s; 289 } 290 #endif 291 292 template<> EIGEN_STRONG_INLINE Packet4i pload<Packet4i>(const int* from) 293 { 294 // FIXME: No intrinsic yet 295 EIGEN_DEBUG_ALIGNED_LOAD 296 Packet *vfrom; 297 vfrom = (Packet *) from; 298 return vfrom->v4i; 299 } 300 301 template<> EIGEN_STRONG_INLINE Packet2d pload<Packet2d>(const double* from) 302 { 303 // FIXME: No intrinsic yet 304 EIGEN_DEBUG_ALIGNED_LOAD 305 Packet *vfrom; 306 vfrom = (Packet *) from; 307 return vfrom->v2d; 308 } 309 310 template<> EIGEN_STRONG_INLINE void pstore<int>(int* to, const Packet4i& from) 311 { 312 // FIXME: No intrinsic yet 313 EIGEN_DEBUG_ALIGNED_STORE 314 Packet *vto; 315 vto = (Packet *) to; 316 vto->v4i = from; 317 } 318 319 template<> EIGEN_STRONG_INLINE void pstore<double>(double* to, const Packet2d& from) 320 { 321 // FIXME: No intrinsic yet 322 EIGEN_DEBUG_ALIGNED_STORE 323 Packet *vto; 324 vto = (Packet *) to; 325 vto->v2d = from; 326 } 327 328 template<> EIGEN_STRONG_INLINE Packet4i pset1<Packet4i>(const int& from) 329 { 330 return vec_splats(from); 331 } 332 template<> EIGEN_STRONG_INLINE Packet2d pset1<Packet2d>(const double& from) { 333 return vec_splats(from); 334 } 335 336 template<> EIGEN_STRONG_INLINE void 337 pbroadcast4<Packet4i>(const int *a, 338 Packet4i& a0, Packet4i& a1, Packet4i& a2, Packet4i& a3) 339 { 340 a3 = pload<Packet4i>(a); 341 a0 = vec_splat(a3, 0); 342 a1 = vec_splat(a3, 1); 343 a2 = vec_splat(a3, 2); 344 a3 = vec_splat(a3, 3); 345 } 346 347 template<> EIGEN_STRONG_INLINE void 348 pbroadcast4<Packet2d>(const double *a, 349 Packet2d& a0, Packet2d& a1, Packet2d& a2, Packet2d& a3) 350 { 351 a1 = pload<Packet2d>(a); 352 a0 = vec_splat(a1, 0); 353 a1 = vec_splat(a1, 1); 354 a3 = pload<Packet2d>(a+2); 355 a2 = vec_splat(a3, 0); 356 a3 = vec_splat(a3, 1); 357 } 358 359 template<> EIGEN_DEVICE_FUNC inline Packet4i pgather<int, Packet4i>(const int* from, Index stride) 360 { 361 int EIGEN_ALIGN16 ai[4]; 362 ai[0] = from[0*stride]; 363 ai[1] = from[1*stride]; 364 ai[2] = from[2*stride]; 365 ai[3] = from[3*stride]; 366 return pload<Packet4i>(ai); 367 } 368 369 template<> EIGEN_DEVICE_FUNC inline Packet2d pgather<double, Packet2d>(const double* from, Index stride) 370 { 371 double EIGEN_ALIGN16 af[2]; 372 af[0] = from[0*stride]; 373 af[1] = from[1*stride]; 374 return pload<Packet2d>(af); 375 } 376 377 template<> EIGEN_DEVICE_FUNC inline void pscatter<int, Packet4i>(int* to, const Packet4i& from, Index stride) 378 { 379 int EIGEN_ALIGN16 ai[4]; 380 pstore<int>((int *)ai, from); 381 to[0*stride] = ai[0]; 382 to[1*stride] = ai[1]; 383 to[2*stride] = ai[2]; 384 to[3*stride] = ai[3]; 385 } 386 387 template<> EIGEN_DEVICE_FUNC inline void pscatter<double, Packet2d>(double* to, const Packet2d& from, Index stride) 388 { 389 double EIGEN_ALIGN16 af[2]; 390 pstore<double>(af, from); 391 to[0*stride] = af[0]; 392 to[1*stride] = af[1]; 393 } 394 395 template<> EIGEN_STRONG_INLINE Packet4i padd<Packet4i>(const Packet4i& a, const Packet4i& b) { return (a + b); } 396 template<> EIGEN_STRONG_INLINE Packet2d padd<Packet2d>(const Packet2d& a, const Packet2d& b) { return (a + b); } 397 398 template<> EIGEN_STRONG_INLINE Packet4i psub<Packet4i>(const Packet4i& a, const Packet4i& b) { return (a - b); } 399 template<> EIGEN_STRONG_INLINE Packet2d psub<Packet2d>(const Packet2d& a, const Packet2d& b) { return (a - b); } 400 401 template<> EIGEN_STRONG_INLINE Packet4i pmul<Packet4i>(const Packet4i& a, const Packet4i& b) { return (a * b); } 402 template<> EIGEN_STRONG_INLINE Packet2d pmul<Packet2d>(const Packet2d& a, const Packet2d& b) { return (a * b); } 403 404 template<> EIGEN_STRONG_INLINE Packet4i pdiv<Packet4i>(const Packet4i& a, const Packet4i& b) { return (a / b); } 405 template<> EIGEN_STRONG_INLINE Packet2d pdiv<Packet2d>(const Packet2d& a, const Packet2d& b) { return (a / b); } 406 407 template<> EIGEN_STRONG_INLINE Packet4i pnegate(const Packet4i& a) { return (-a); } 408 template<> EIGEN_STRONG_INLINE Packet2d pnegate(const Packet2d& a) { return (-a); } 409 410 template<> EIGEN_STRONG_INLINE Packet4i pconj(const Packet4i& a) { return a; } 411 template<> EIGEN_STRONG_INLINE Packet2d pconj(const Packet2d& a) { return a; } 412 413 template<> EIGEN_STRONG_INLINE Packet4i pmadd(const Packet4i& a, const Packet4i& b, const Packet4i& c) { return padd<Packet4i>(pmul<Packet4i>(a, b), c); } 414 template<> EIGEN_STRONG_INLINE Packet2d pmadd(const Packet2d& a, const Packet2d& b, const Packet2d& c) { return vec_madd(a, b, c); } 415 416 template<> EIGEN_STRONG_INLINE Packet4i plset<Packet4i>(const int& a) { return padd<Packet4i>(pset1<Packet4i>(a), p4i_COUNTDOWN); } 417 template<> EIGEN_STRONG_INLINE Packet2d plset<Packet2d>(const double& a) { return padd<Packet2d>(pset1<Packet2d>(a), p2d_COUNTDOWN); } 418 419 template<> EIGEN_STRONG_INLINE Packet4i pmin<Packet4i>(const Packet4i& a, const Packet4i& b) { return vec_min(a, b); } 420 template<> EIGEN_STRONG_INLINE Packet2d pmin<Packet2d>(const Packet2d& a, const Packet2d& b) { return vec_min(a, b); } 421 422 template<> EIGEN_STRONG_INLINE Packet4i pmax<Packet4i>(const Packet4i& a, const Packet4i& b) { return vec_max(a, b); } 423 template<> EIGEN_STRONG_INLINE Packet2d pmax<Packet2d>(const Packet2d& a, const Packet2d& b) { return vec_max(a, b); } 424 425 template<> EIGEN_STRONG_INLINE Packet4i pand<Packet4i>(const Packet4i& a, const Packet4i& b) { return vec_and(a, b); } 426 template<> EIGEN_STRONG_INLINE Packet2d pand<Packet2d>(const Packet2d& a, const Packet2d& b) { return vec_and(a, b); } 427 428 template<> EIGEN_STRONG_INLINE Packet4i por<Packet4i>(const Packet4i& a, const Packet4i& b) { return vec_or(a, b); } 429 template<> EIGEN_STRONG_INLINE Packet2d por<Packet2d>(const Packet2d& a, const Packet2d& b) { return vec_or(a, b); } 430 431 template<> EIGEN_STRONG_INLINE Packet4i pxor<Packet4i>(const Packet4i& a, const Packet4i& b) { return vec_xor(a, b); } 432 template<> EIGEN_STRONG_INLINE Packet2d pxor<Packet2d>(const Packet2d& a, const Packet2d& b) { return vec_xor(a, b); } 433 434 template<> EIGEN_STRONG_INLINE Packet4i pandnot<Packet4i>(const Packet4i& a, const Packet4i& b) { return pand<Packet4i>(a, vec_nor(b, b)); } 435 template<> EIGEN_STRONG_INLINE Packet2d pandnot<Packet2d>(const Packet2d& a, const Packet2d& b) { return vec_and(a, vec_nor(b, b)); } 436 437 template<> EIGEN_STRONG_INLINE Packet2d pround<Packet2d>(const Packet2d& a) { return vec_round(a); } 438 template<> EIGEN_STRONG_INLINE Packet2d pceil<Packet2d>(const Packet2d& a) { return vec_ceil(a); } 439 template<> EIGEN_STRONG_INLINE Packet2d pfloor<Packet2d>(const Packet2d& a) { return vec_floor(a); } 440 441 template<> EIGEN_STRONG_INLINE Packet4i ploadu<Packet4i>(const int* from) { return pload<Packet4i>(from); } 442 template<> EIGEN_STRONG_INLINE Packet2d ploadu<Packet2d>(const double* from) { return pload<Packet2d>(from); } 443 444 445 template<> EIGEN_STRONG_INLINE Packet4i ploaddup<Packet4i>(const int* from) 446 { 447 Packet4i p = pload<Packet4i>(from); 448 return vec_perm(p, p, p16uc_DUPLICATE32_HI); 449 } 450 451 template<> EIGEN_STRONG_INLINE Packet2d ploaddup<Packet2d>(const double* from) 452 { 453 Packet2d p = pload<Packet2d>(from); 454 return vec_perm(p, p, p16uc_PSET64_HI); 455 } 456 457 template<> EIGEN_STRONG_INLINE void pstoreu<int>(int* to, const Packet4i& from) { pstore<int>(to, from); } 458 template<> EIGEN_STRONG_INLINE void pstoreu<double>(double* to, const Packet2d& from) { pstore<double>(to, from); } 459 460 template<> EIGEN_STRONG_INLINE void prefetch<int>(const int* addr) { EIGEN_ZVECTOR_PREFETCH(addr); } 461 template<> EIGEN_STRONG_INLINE void prefetch<double>(const double* addr) { EIGEN_ZVECTOR_PREFETCH(addr); } 462 463 template<> EIGEN_STRONG_INLINE int pfirst<Packet4i>(const Packet4i& a) { int EIGEN_ALIGN16 x[4]; pstore(x, a); return x[0]; } 464 template<> EIGEN_STRONG_INLINE double pfirst<Packet2d>(const Packet2d& a) { double EIGEN_ALIGN16 x[2]; pstore(x, a); return x[0]; } 465 466 template<> EIGEN_STRONG_INLINE Packet4i preverse(const Packet4i& a) 467 { 468 return reinterpret_cast<Packet4i>(vec_perm(reinterpret_cast<Packet16uc>(a), reinterpret_cast<Packet16uc>(a), p16uc_REVERSE32)); 469 } 470 471 template<> EIGEN_STRONG_INLINE Packet2d preverse(const Packet2d& a) 472 { 473 return reinterpret_cast<Packet2d>(vec_perm(reinterpret_cast<Packet16uc>(a), reinterpret_cast<Packet16uc>(a), p16uc_REVERSE64)); 474 } 475 476 template<> EIGEN_STRONG_INLINE Packet4i pabs<Packet4i>(const Packet4i& a) { return vec_abs(a); } 477 template<> EIGEN_STRONG_INLINE Packet2d pabs<Packet2d>(const Packet2d& a) { return vec_abs(a); } 478 479 template<> EIGEN_STRONG_INLINE int predux<Packet4i>(const Packet4i& a) 480 { 481 Packet4i b, sum; 482 b = vec_sld(a, a, 8); 483 sum = padd<Packet4i>(a, b); 484 b = vec_sld(sum, sum, 4); 485 sum = padd<Packet4i>(sum, b); 486 return pfirst(sum); 487 } 488 489 template<> EIGEN_STRONG_INLINE double predux<Packet2d>(const Packet2d& a) 490 { 491 Packet2d b, sum; 492 b = reinterpret_cast<Packet2d>(vec_sld(reinterpret_cast<Packet4i>(a), reinterpret_cast<Packet4i>(a), 8)); 493 sum = padd<Packet2d>(a, b); 494 return pfirst(sum); 495 } 496 497 // Other reduction functions: 498 // mul 499 template<> EIGEN_STRONG_INLINE int predux_mul<Packet4i>(const Packet4i& a) 500 { 501 EIGEN_ALIGN16 int aux[4]; 502 pstore(aux, a); 503 return aux[0] * aux[1] * aux[2] * aux[3]; 504 } 505 506 template<> EIGEN_STRONG_INLINE double predux_mul<Packet2d>(const Packet2d& a) 507 { 508 return pfirst(pmul(a, reinterpret_cast<Packet2d>(vec_sld(reinterpret_cast<Packet4i>(a), reinterpret_cast<Packet4i>(a), 8)))); 509 } 510 511 // min 512 template<> EIGEN_STRONG_INLINE int predux_min<Packet4i>(const Packet4i& a) 513 { 514 Packet4i b, res; 515 b = pmin<Packet4i>(a, vec_sld(a, a, 8)); 516 res = pmin<Packet4i>(b, vec_sld(b, b, 4)); 517 return pfirst(res); 518 } 519 520 template<> EIGEN_STRONG_INLINE double predux_min<Packet2d>(const Packet2d& a) 521 { 522 return pfirst(pmin<Packet2d>(a, reinterpret_cast<Packet2d>(vec_sld(reinterpret_cast<Packet4i>(a), reinterpret_cast<Packet4i>(a), 8)))); 523 } 524 525 // max 526 template<> EIGEN_STRONG_INLINE int predux_max<Packet4i>(const Packet4i& a) 527 { 528 Packet4i b, res; 529 b = pmax<Packet4i>(a, vec_sld(a, a, 8)); 530 res = pmax<Packet4i>(b, vec_sld(b, b, 4)); 531 return pfirst(res); 532 } 533 534 // max 535 template<> EIGEN_STRONG_INLINE double predux_max<Packet2d>(const Packet2d& a) 536 { 537 return pfirst(pmax<Packet2d>(a, reinterpret_cast<Packet2d>(vec_sld(reinterpret_cast<Packet4i>(a), reinterpret_cast<Packet4i>(a), 8)))); 538 } 539 540 EIGEN_DEVICE_FUNC inline void 541 ptranspose(PacketBlock<Packet4i,4>& kernel) { 542 Packet4i t0 = vec_mergeh(kernel.packet[0], kernel.packet[2]); 543 Packet4i t1 = vec_mergel(kernel.packet[0], kernel.packet[2]); 544 Packet4i t2 = vec_mergeh(kernel.packet[1], kernel.packet[3]); 545 Packet4i t3 = vec_mergel(kernel.packet[1], kernel.packet[3]); 546 kernel.packet[0] = vec_mergeh(t0, t2); 547 kernel.packet[1] = vec_mergel(t0, t2); 548 kernel.packet[2] = vec_mergeh(t1, t3); 549 kernel.packet[3] = vec_mergel(t1, t3); 550 } 551 552 EIGEN_DEVICE_FUNC inline void 553 ptranspose(PacketBlock<Packet2d,2>& kernel) { 554 Packet2d t0 = vec_perm(kernel.packet[0], kernel.packet[1], p16uc_TRANSPOSE64_HI); 555 Packet2d t1 = vec_perm(kernel.packet[0], kernel.packet[1], p16uc_TRANSPOSE64_LO); 556 kernel.packet[0] = t0; 557 kernel.packet[1] = t1; 558 } 559 560 template<> EIGEN_STRONG_INLINE Packet4i pblend(const Selector<4>& ifPacket, const Packet4i& thenPacket, const Packet4i& elsePacket) { 561 Packet4ui select = { ifPacket.select[0], ifPacket.select[1], ifPacket.select[2], ifPacket.select[3] }; 562 Packet4ui mask = vec_cmpeq(select, reinterpret_cast<Packet4ui>(p4i_ONE)); 563 return vec_sel(elsePacket, thenPacket, mask); 564 } 565 566 567 template<> EIGEN_STRONG_INLINE Packet2d pblend(const Selector<2>& ifPacket, const Packet2d& thenPacket, const Packet2d& elsePacket) { 568 Packet2ul select = { ifPacket.select[0], ifPacket.select[1] }; 569 Packet2ul mask = vec_cmpeq(select, reinterpret_cast<Packet2ul>(p2l_ONE)); 570 return vec_sel(elsePacket, thenPacket, mask); 571 } 572 573 /* z13 has no vector float support so we emulate that with double 574 z14 has proper vector float support. 575 */ 576 #if !defined(__ARCH__) || (defined(__ARCH__) && __ARCH__ < 12) 577 /* Helper function to simulate a vec_splat_packet4f 578 */ 579 template<int element> EIGEN_STRONG_INLINE Packet4f vec_splat_packet4f(const Packet4f& from) 580 { 581 Packet4f splat; 582 switch (element) { 583 case 0: 584 splat.v4f[0] = vec_splat(from.v4f[0], 0); 585 splat.v4f[1] = splat.v4f[0]; 586 break; 587 case 1: 588 splat.v4f[0] = vec_splat(from.v4f[0], 1); 589 splat.v4f[1] = splat.v4f[0]; 590 break; 591 case 2: 592 splat.v4f[0] = vec_splat(from.v4f[1], 0); 593 splat.v4f[1] = splat.v4f[0]; 594 break; 595 case 3: 596 splat.v4f[0] = vec_splat(from.v4f[1], 1); 597 splat.v4f[1] = splat.v4f[0]; 598 break; 599 } 600 return splat; 601 } 602 603 template<> EIGEN_STRONG_INLINE Packet4f pload<Packet4f>(const float* from) 604 { 605 // FIXME: No intrinsic yet 606 EIGEN_DEBUG_ALIGNED_LOAD 607 Packet4f vfrom; 608 vfrom.v4f[0] = vec_ld2f(&from[0]); 609 vfrom.v4f[1] = vec_ld2f(&from[2]); 610 return vfrom; 611 } 612 613 template<> EIGEN_STRONG_INLINE void pstore<float>(float* to, const Packet4f& from) 614 { 615 // FIXME: No intrinsic yet 616 EIGEN_DEBUG_ALIGNED_STORE 617 vec_st2f(from.v4f[0], &to[0]); 618 vec_st2f(from.v4f[1], &to[2]); 619 } 620 621 template<> EIGEN_STRONG_INLINE Packet4f pset1<Packet4f>(const float& from) 622 { 623 Packet4f to; 624 to.v4f[0] = pset1<Packet2d>(static_cast<const double&>(from)); 625 to.v4f[1] = to.v4f[0]; 626 return to; 627 } 628 629 template<> EIGEN_STRONG_INLINE void 630 pbroadcast4<Packet4f>(const float *a, 631 Packet4f& a0, Packet4f& a1, Packet4f& a2, Packet4f& a3) 632 { 633 a3 = pload<Packet4f>(a); 634 a0 = vec_splat_packet4f<0>(a3); 635 a1 = vec_splat_packet4f<1>(a3); 636 a2 = vec_splat_packet4f<2>(a3); 637 a3 = vec_splat_packet4f<3>(a3); 638 } 639 640 template<> EIGEN_DEVICE_FUNC inline Packet4f pgather<float, Packet4f>(const float* from, Index stride) 641 { 642 float EIGEN_ALIGN16 ai[4]; 643 ai[0] = from[0*stride]; 644 ai[1] = from[1*stride]; 645 ai[2] = from[2*stride]; 646 ai[3] = from[3*stride]; 647 return pload<Packet4f>(ai); 648 } 649 650 template<> EIGEN_DEVICE_FUNC inline void pscatter<float, Packet4f>(float* to, const Packet4f& from, Index stride) 651 { 652 float EIGEN_ALIGN16 ai[4]; 653 pstore<float>((float *)ai, from); 654 to[0*stride] = ai[0]; 655 to[1*stride] = ai[1]; 656 to[2*stride] = ai[2]; 657 to[3*stride] = ai[3]; 658 } 659 660 template<> EIGEN_STRONG_INLINE Packet4f padd<Packet4f>(const Packet4f& a, const Packet4f& b) 661 { 662 Packet4f c; 663 c.v4f[0] = a.v4f[0] + b.v4f[0]; 664 c.v4f[1] = a.v4f[1] + b.v4f[1]; 665 return c; 666 } 667 668 template<> EIGEN_STRONG_INLINE Packet4f psub<Packet4f>(const Packet4f& a, const Packet4f& b) 669 { 670 Packet4f c; 671 c.v4f[0] = a.v4f[0] - b.v4f[0]; 672 c.v4f[1] = a.v4f[1] - b.v4f[1]; 673 return c; 674 } 675 676 template<> EIGEN_STRONG_INLINE Packet4f pmul<Packet4f>(const Packet4f& a, const Packet4f& b) 677 { 678 Packet4f c; 679 c.v4f[0] = a.v4f[0] * b.v4f[0]; 680 c.v4f[1] = a.v4f[1] * b.v4f[1]; 681 return c; 682 } 683 684 template<> EIGEN_STRONG_INLINE Packet4f pdiv<Packet4f>(const Packet4f& a, const Packet4f& b) 685 { 686 Packet4f c; 687 c.v4f[0] = a.v4f[0] / b.v4f[0]; 688 c.v4f[1] = a.v4f[1] / b.v4f[1]; 689 return c; 690 } 691 692 template<> EIGEN_STRONG_INLINE Packet4f pnegate(const Packet4f& a) 693 { 694 Packet4f c; 695 c.v4f[0] = -a.v4f[0]; 696 c.v4f[1] = -a.v4f[1]; 697 return c; 698 } 699 700 template<> EIGEN_STRONG_INLINE Packet4f pmadd(const Packet4f& a, const Packet4f& b, const Packet4f& c) 701 { 702 Packet4f res; 703 res.v4f[0] = vec_madd(a.v4f[0], b.v4f[0], c.v4f[0]); 704 res.v4f[1] = vec_madd(a.v4f[1], b.v4f[1], c.v4f[1]); 705 return res; 706 } 707 708 template<> EIGEN_STRONG_INLINE Packet4f pmin<Packet4f>(const Packet4f& a, const Packet4f& b) 709 { 710 Packet4f res; 711 res.v4f[0] = pmin(a.v4f[0], b.v4f[0]); 712 res.v4f[1] = pmin(a.v4f[1], b.v4f[1]); 713 return res; 714 } 715 716 template<> EIGEN_STRONG_INLINE Packet4f pmax<Packet4f>(const Packet4f& a, const Packet4f& b) 717 { 718 Packet4f res; 719 res.v4f[0] = pmax(a.v4f[0], b.v4f[0]); 720 res.v4f[1] = pmax(a.v4f[1], b.v4f[1]); 721 return res; 722 } 723 724 template<> EIGEN_STRONG_INLINE Packet4f pand<Packet4f>(const Packet4f& a, const Packet4f& b) 725 { 726 Packet4f res; 727 res.v4f[0] = pand(a.v4f[0], b.v4f[0]); 728 res.v4f[1] = pand(a.v4f[1], b.v4f[1]); 729 return res; 730 } 731 732 template<> EIGEN_STRONG_INLINE Packet4f por<Packet4f>(const Packet4f& a, const Packet4f& b) 733 { 734 Packet4f res; 735 res.v4f[0] = por(a.v4f[0], b.v4f[0]); 736 res.v4f[1] = por(a.v4f[1], b.v4f[1]); 737 return res; 738 } 739 740 template<> EIGEN_STRONG_INLINE Packet4f pxor<Packet4f>(const Packet4f& a, const Packet4f& b) 741 { 742 Packet4f res; 743 res.v4f[0] = pxor(a.v4f[0], b.v4f[0]); 744 res.v4f[1] = pxor(a.v4f[1], b.v4f[1]); 745 return res; 746 } 747 748 template<> EIGEN_STRONG_INLINE Packet4f pandnot<Packet4f>(const Packet4f& a, const Packet4f& b) 749 { 750 Packet4f res; 751 res.v4f[0] = pandnot(a.v4f[0], b.v4f[0]); 752 res.v4f[1] = pandnot(a.v4f[1], b.v4f[1]); 753 return res; 754 } 755 756 template<> EIGEN_STRONG_INLINE Packet4f pround<Packet4f>(const Packet4f& a) 757 { 758 Packet4f res; 759 res.v4f[0] = vec_round(a.v4f[0]); 760 res.v4f[1] = vec_round(a.v4f[1]); 761 return res; 762 } 763 764 template<> EIGEN_STRONG_INLINE Packet4f pceil<Packet4f>(const Packet4f& a) 765 { 766 Packet4f res; 767 res.v4f[0] = vec_ceil(a.v4f[0]); 768 res.v4f[1] = vec_ceil(a.v4f[1]); 769 return res; 770 } 771 772 template<> EIGEN_STRONG_INLINE Packet4f pfloor<Packet4f>(const Packet4f& a) 773 { 774 Packet4f res; 775 res.v4f[0] = vec_floor(a.v4f[0]); 776 res.v4f[1] = vec_floor(a.v4f[1]); 777 return res; 778 } 779 780 template<> EIGEN_STRONG_INLINE Packet4f ploaddup<Packet4f>(const float* from) 781 { 782 Packet4f p = pload<Packet4f>(from); 783 p.v4f[1] = vec_splat(p.v4f[0], 1); 784 p.v4f[0] = vec_splat(p.v4f[0], 0); 785 return p; 786 } 787 788 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]; } 789 790 template<> EIGEN_STRONG_INLINE Packet4f preverse(const Packet4f& a) 791 { 792 Packet4f rev; 793 rev.v4f[0] = preverse<Packet2d>(a.v4f[1]); 794 rev.v4f[1] = preverse<Packet2d>(a.v4f[0]); 795 return rev; 796 } 797 798 template<> EIGEN_STRONG_INLINE Packet4f pabs<Packet4f>(const Packet4f& a) 799 { 800 Packet4f res; 801 res.v4f[0] = pabs(a.v4f[0]); 802 res.v4f[1] = pabs(a.v4f[1]); 803 return res; 804 } 805 806 template<> EIGEN_STRONG_INLINE float predux<Packet4f>(const Packet4f& a) 807 { 808 Packet2d sum; 809 sum = padd<Packet2d>(a.v4f[0], a.v4f[1]); 810 double first = predux<Packet2d>(sum); 811 return static_cast<float>(first); 812 } 813 814 template<> EIGEN_STRONG_INLINE float predux_mul<Packet4f>(const Packet4f& a) 815 { 816 // Return predux_mul<Packet2d> of the subvectors product 817 return static_cast<float>(pfirst(predux_mul(pmul(a.v4f[0], a.v4f[1])))); 818 } 819 820 template<> EIGEN_STRONG_INLINE float predux_min<Packet4f>(const Packet4f& a) 821 { 822 Packet2d b, res; 823 b = pmin<Packet2d>(a.v4f[0], a.v4f[1]); 824 res = pmin<Packet2d>(b, reinterpret_cast<Packet2d>(vec_sld(reinterpret_cast<Packet4i>(b), reinterpret_cast<Packet4i>(b), 8))); 825 return static_cast<float>(pfirst(res)); 826 } 827 828 template<> EIGEN_STRONG_INLINE float predux_max<Packet4f>(const Packet4f& a) 829 { 830 Packet2d b, res; 831 b = pmax<Packet2d>(a.v4f[0], a.v4f[1]); 832 res = pmax<Packet2d>(b, reinterpret_cast<Packet2d>(vec_sld(reinterpret_cast<Packet4i>(b), reinterpret_cast<Packet4i>(b), 8))); 833 return static_cast<float>(pfirst(res)); 834 } 835 836 /* Split the Packet4f PacketBlock into 4 Packet2d PacketBlocks and transpose each one 837 */ 838 EIGEN_DEVICE_FUNC inline void 839 ptranspose(PacketBlock<Packet4f,4>& kernel) { 840 PacketBlock<Packet2d,2> t0,t1,t2,t3; 841 // copy top-left 2x2 Packet2d block 842 t0.packet[0] = kernel.packet[0].v4f[0]; 843 t0.packet[1] = kernel.packet[1].v4f[0]; 844 845 // copy top-right 2x2 Packet2d block 846 t1.packet[0] = kernel.packet[0].v4f[1]; 847 t1.packet[1] = kernel.packet[1].v4f[1]; 848 849 // copy bottom-left 2x2 Packet2d block 850 t2.packet[0] = kernel.packet[2].v4f[0]; 851 t2.packet[1] = kernel.packet[3].v4f[0]; 852 853 // copy bottom-right 2x2 Packet2d block 854 t3.packet[0] = kernel.packet[2].v4f[1]; 855 t3.packet[1] = kernel.packet[3].v4f[1]; 856 857 // Transpose all 2x2 blocks 858 ptranspose(t0); 859 ptranspose(t1); 860 ptranspose(t2); 861 ptranspose(t3); 862 863 // Copy back transposed blocks, but exchange t1 and t2 due to transposition 864 kernel.packet[0].v4f[0] = t0.packet[0]; 865 kernel.packet[0].v4f[1] = t2.packet[0]; 866 kernel.packet[1].v4f[0] = t0.packet[1]; 867 kernel.packet[1].v4f[1] = t2.packet[1]; 868 kernel.packet[2].v4f[0] = t1.packet[0]; 869 kernel.packet[2].v4f[1] = t3.packet[0]; 870 kernel.packet[3].v4f[0] = t1.packet[1]; 871 kernel.packet[3].v4f[1] = t3.packet[1]; 872 } 873 874 template<> EIGEN_STRONG_INLINE Packet4f pblend(const Selector<4>& ifPacket, const Packet4f& thenPacket, const Packet4f& elsePacket) { 875 Packet2ul select_hi = { ifPacket.select[0], ifPacket.select[1] }; 876 Packet2ul select_lo = { ifPacket.select[2], ifPacket.select[3] }; 877 Packet2ul mask_hi = vec_cmpeq(select_hi, reinterpret_cast<Packet2ul>(p2l_ONE)); 878 Packet2ul mask_lo = vec_cmpeq(select_lo, reinterpret_cast<Packet2ul>(p2l_ONE)); 879 Packet4f result; 880 result.v4f[0] = vec_sel(elsePacket.v4f[0], thenPacket.v4f[0], mask_hi); 881 result.v4f[1] = vec_sel(elsePacket.v4f[1], thenPacket.v4f[1], mask_lo); 882 return result; 883 } 884 885 template<> Packet4f EIGEN_STRONG_INLINE pcmp_le<Packet4f>(const Packet4f& a, const Packet4f& b) 886 { 887 Packet4f res; 888 res.v4f[0] = pcmp_le(a.v4f[0], b.v4f[0]); 889 res.v4f[1] = pcmp_le(a.v4f[1], b.v4f[1]); 890 return res; 891 } 892 893 template<> Packet4f EIGEN_STRONG_INLINE pcmp_lt<Packet4f>(const Packet4f& a, const Packet4f& b) 894 { 895 Packet4f res; 896 res.v4f[0] = pcmp_lt(a.v4f[0], b.v4f[0]); 897 res.v4f[1] = pcmp_lt(a.v4f[1], b.v4f[1]); 898 return res; 899 } 900 901 template<> Packet4f EIGEN_STRONG_INLINE pcmp_eq<Packet4f>(const Packet4f& a, const Packet4f& b) 902 { 903 Packet4f res; 904 res.v4f[0] = pcmp_eq(a.v4f[0], b.v4f[0]); 905 res.v4f[1] = pcmp_eq(a.v4f[1], b.v4f[1]); 906 return res; 907 } 908 909 #else 910 template<> EIGEN_STRONG_INLINE Packet4f pload<Packet4f>(const float* from) 911 { 912 // FIXME: No intrinsic yet 913 EIGEN_DEBUG_ALIGNED_LOAD 914 Packet *vfrom; 915 vfrom = (Packet *) from; 916 return vfrom->v4f; 917 } 918 919 template<> EIGEN_STRONG_INLINE void pstore<float>(float* to, const Packet4f& from) 920 { 921 // FIXME: No intrinsic yet 922 EIGEN_DEBUG_ALIGNED_STORE 923 Packet *vto; 924 vto = (Packet *) to; 925 vto->v4f = from; 926 } 927 928 template<> EIGEN_STRONG_INLINE Packet4f pset1<Packet4f>(const float& from) 929 { 930 return vec_splats(from); 931 } 932 933 template<> EIGEN_STRONG_INLINE void 934 pbroadcast4<Packet4f>(const float *a, 935 Packet4f& a0, Packet4f& a1, Packet4f& a2, Packet4f& a3) 936 { 937 a3 = pload<Packet4f>(a); 938 a0 = vec_splat(a3, 0); 939 a1 = vec_splat(a3, 1); 940 a2 = vec_splat(a3, 2); 941 a3 = vec_splat(a3, 3); 942 } 943 944 template<> EIGEN_DEVICE_FUNC inline Packet4f pgather<float, Packet4f>(const float* from, Index stride) 945 { 946 float EIGEN_ALIGN16 af[4]; 947 af[0] = from[0*stride]; 948 af[1] = from[1*stride]; 949 af[2] = from[2*stride]; 950 af[3] = from[3*stride]; 951 return pload<Packet4f>(af); 952 } 953 954 template<> EIGEN_DEVICE_FUNC inline void pscatter<float, Packet4f>(float* to, const Packet4f& from, Index stride) 955 { 956 float EIGEN_ALIGN16 af[4]; 957 pstore<float>((float*)af, from); 958 to[0*stride] = af[0]; 959 to[1*stride] = af[1]; 960 to[2*stride] = af[2]; 961 to[3*stride] = af[3]; 962 } 963 964 template<> EIGEN_STRONG_INLINE Packet4f padd<Packet4f>(const Packet4f& a, const Packet4f& b) { return (a + b); } 965 template<> EIGEN_STRONG_INLINE Packet4f psub<Packet4f>(const Packet4f& a, const Packet4f& b) { return (a - b); } 966 template<> EIGEN_STRONG_INLINE Packet4f pmul<Packet4f>(const Packet4f& a, const Packet4f& b) { return (a * b); } 967 template<> EIGEN_STRONG_INLINE Packet4f pdiv<Packet4f>(const Packet4f& a, const Packet4f& b) { return (a / b); } 968 template<> EIGEN_STRONG_INLINE Packet4f pnegate<Packet4f>(const Packet4f& a) { return (-a); } 969 template<> EIGEN_STRONG_INLINE Packet4f pconj<Packet4f> (const Packet4f& a) { return a; } 970 template<> EIGEN_STRONG_INLINE Packet4f pmadd<Packet4f> (const Packet4f& a, const Packet4f& b, const Packet4f& c) { return vec_madd(a, b, c); } 971 template<> EIGEN_STRONG_INLINE Packet4f pmin<Packet4f> (const Packet4f& a, const Packet4f& b) { return vec_min(a, b); } 972 template<> EIGEN_STRONG_INLINE Packet4f pmax<Packet4f> (const Packet4f& a, const Packet4f& b) { return vec_max(a, b); } 973 template<> EIGEN_STRONG_INLINE Packet4f pand<Packet4f> (const Packet4f& a, const Packet4f& b) { return vec_and(a, b); } 974 template<> EIGEN_STRONG_INLINE Packet4f por<Packet4f> (const Packet4f& a, const Packet4f& b) { return vec_or(a, b); } 975 template<> EIGEN_STRONG_INLINE Packet4f pxor<Packet4f> (const Packet4f& a, const Packet4f& b) { return vec_xor(a, b); } 976 template<> EIGEN_STRONG_INLINE Packet4f pandnot<Packet4f>(const Packet4f& a, const Packet4f& b) { return vec_and(a, vec_nor(b, b)); } 977 template<> EIGEN_STRONG_INLINE Packet4f pround<Packet4f> (const Packet4f& a) { return vec_round(a); } 978 template<> EIGEN_STRONG_INLINE Packet4f pceil<Packet4f> (const Packet4f& a) { return vec_ceil(a); } 979 template<> EIGEN_STRONG_INLINE Packet4f pfloor<Packet4f> (const Packet4f& a) { return vec_floor(a); } 980 template<> EIGEN_STRONG_INLINE Packet4f pabs<Packet4f> (const Packet4f& a) { return vec_abs(a); } 981 template<> EIGEN_STRONG_INLINE float pfirst<Packet4f>(const Packet4f& a) { float EIGEN_ALIGN16 x[4]; pstore(x, a); return x[0]; } 982 983 template<> EIGEN_STRONG_INLINE Packet4f ploaddup<Packet4f>(const float* from) 984 { 985 Packet4f p = pload<Packet4f>(from); 986 return vec_perm(p, p, p16uc_DUPLICATE32_HI); 987 } 988 989 template<> EIGEN_STRONG_INLINE Packet4f preverse(const Packet4f& a) 990 { 991 return reinterpret_cast<Packet4f>(vec_perm(reinterpret_cast<Packet16uc>(a), reinterpret_cast<Packet16uc>(a), p16uc_REVERSE32)); 992 } 993 994 template<> EIGEN_STRONG_INLINE float predux<Packet4f>(const Packet4f& a) 995 { 996 Packet4f b, sum; 997 b = vec_sld(a, a, 8); 998 sum = padd<Packet4f>(a, b); 999 b = vec_sld(sum, sum, 4); 1000 sum = padd<Packet4f>(sum, b); 1001 return pfirst(sum); 1002 } 1003 1004 // Other reduction functions: 1005 // mul 1006 template<> EIGEN_STRONG_INLINE float predux_mul<Packet4f>(const Packet4f& a) 1007 { 1008 Packet4f prod; 1009 prod = pmul(a, vec_sld(a, a, 8)); 1010 return pfirst(pmul(prod, vec_sld(prod, prod, 4))); 1011 } 1012 1013 // min 1014 template<> EIGEN_STRONG_INLINE float predux_min<Packet4f>(const Packet4f& a) 1015 { 1016 Packet4f b, res; 1017 b = pmin<Packet4f>(a, vec_sld(a, a, 8)); 1018 res = pmin<Packet4f>(b, vec_sld(b, b, 4)); 1019 return pfirst(res); 1020 } 1021 1022 // max 1023 template<> EIGEN_STRONG_INLINE float predux_max<Packet4f>(const Packet4f& a) 1024 { 1025 Packet4f b, res; 1026 b = pmax<Packet4f>(a, vec_sld(a, a, 8)); 1027 res = pmax<Packet4f>(b, vec_sld(b, b, 4)); 1028 return pfirst(res); 1029 } 1030 1031 EIGEN_DEVICE_FUNC inline void 1032 ptranspose(PacketBlock<Packet4f,4>& kernel) { 1033 Packet4f t0 = vec_mergeh(kernel.packet[0], kernel.packet[2]); 1034 Packet4f t1 = vec_mergel(kernel.packet[0], kernel.packet[2]); 1035 Packet4f t2 = vec_mergeh(kernel.packet[1], kernel.packet[3]); 1036 Packet4f t3 = vec_mergel(kernel.packet[1], kernel.packet[3]); 1037 kernel.packet[0] = vec_mergeh(t0, t2); 1038 kernel.packet[1] = vec_mergel(t0, t2); 1039 kernel.packet[2] = vec_mergeh(t1, t3); 1040 kernel.packet[3] = vec_mergel(t1, t3); 1041 } 1042 1043 template<> EIGEN_STRONG_INLINE Packet4f pblend(const Selector<4>& ifPacket, const Packet4f& thenPacket, const Packet4f& elsePacket) { 1044 Packet4ui select = { ifPacket.select[0], ifPacket.select[1], ifPacket.select[2], ifPacket.select[3] }; 1045 Packet4ui mask = vec_cmpeq(select, reinterpret_cast<Packet4ui>(p4i_ONE)); 1046 return vec_sel(elsePacket, thenPacket, mask); 1047 } 1048 1049 #endif 1050 1051 template<> EIGEN_STRONG_INLINE void prefetch<float>(const float* addr) { EIGEN_ZVECTOR_PREFETCH(addr); } 1052 template<> EIGEN_STRONG_INLINE Packet4f ploadu<Packet4f> (const float* from) { return pload<Packet4f>(from); } 1053 template<> EIGEN_STRONG_INLINE void pstoreu<float>(float* to, const Packet4f& from) { pstore<float>(to, from); } 1054 template<> EIGEN_STRONG_INLINE Packet4f plset<Packet4f> (const float& a) { return padd<Packet4f>(pset1<Packet4f>(a), p4f_COUNTDOWN); } 1055 1056 } // end namespace internal 1057 1058 } // end namespace Eigen 1059 1060 #endif // EIGEN_PACKET_MATH_ZVECTOR_H 1061