1 // -*- C++ -*- 2 //===------------------------------- simd ---------------------------------===// 3 // 4 // Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions. 5 // See https://llvm.org/LICENSE.txt for license information. 6 // SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception 7 // 8 //===----------------------------------------------------------------------===// 9 #ifndef _LIBCPP_EXPERIMENTAL_SIMD 10 #define _LIBCPP_EXPERIMENTAL_SIMD 11 12 /* 13 experimental/simd synopsis 14 15 namespace std::experimental { 16 17 inline namespace parallelism_v2 { 18 19 namespace simd_abi { 20 21 struct scalar {}; 22 template <int N> struct fixed_size {}; 23 template <typename T> inline constexpr int max_fixed_size = implementation-defined; 24 template <typename T> using compatible = implementation-defined; 25 template <typename T> using native = implementation-defined; 26 27 } // simd_abi 28 29 struct element_aligned_tag {}; 30 struct vector_aligned_tag {}; 31 template <size_t> struct overaligned_tag {}; 32 inline constexpr element_aligned_tag element_aligned{}; 33 inline constexpr vector_aligned_tag vector_aligned{}; 34 template <size_t N> inline constexpr overaligned_tag<N> overaligned{}; 35 36 // traits [simd.traits] 37 template <class T> struct is_abi_tag; 38 template <class T> inline constexpr bool is_abi_tag_v = is_abi_tag<T>::value; 39 40 template <class T> struct is_simd; 41 template <class T> inline constexpr bool is_simd_v = is_simd<T>::value; 42 43 template <class T> struct is_simd_mask; 44 template <class T> inline constexpr bool is_simd_mask_v = is_simd_mask<T>::value; 45 46 template <class T> struct is_simd_flag_type; 47 template <class T> inline constexpr bool is_simd_flag_type_v = is_simd_flag_type<T>::value; 48 49 template <class T, size_t N> struct abi_for_size { using type = see below; }; 50 template <class T, size_t N> using abi_for_size_t = typename abi_for_size<T, N>::type; 51 52 template <class T, class Abi = simd_abi::compatible<T>> struct simd_size; 53 template <class T, class Abi = simd_abi::compatible<T>> 54 inline constexpr size_t simd_size_v = simd_size<T, Abi>::value; 55 56 template <class T, class U = typename T::value_type> struct memory_alignment; 57 template <class T, class U = typename T::value_type> 58 inline constexpr size_t memory_alignment_v = memory_alignment<T, U>::value; 59 60 // class template simd [simd.class] 61 template <class T, class Abi = simd_abi::compatible<T>> class simd; 62 template <class T> using native_simd = simd<T, simd_abi::native<T>>; 63 template <class T, int N> using fixed_size_simd = simd<T, simd_abi::fixed_size<N>>; 64 65 // class template simd_mask [simd.mask.class] 66 template <class T, class Abi = simd_abi::compatible<T>> class simd_mask; 67 template <class T> using native_simd_mask = simd_mask<T, simd_abi::native<T>>; 68 template <class T, int N> using fixed_size_simd_mask = simd_mask<T, simd_abi::fixed_size<N>>; 69 70 // casts [simd.casts] 71 template <class T, class U, class Abi> see below simd_cast(const simd<U, Abi>&); 72 template <class T, class U, class Abi> see below static_simd_cast(const simd<U, Abi>&); 73 74 template <class T, class Abi> 75 fixed_size_simd<T, simd_size_v<T, Abi>> to_fixed_size(const simd<T, Abi>&) noexcept; 76 template <class T, class Abi> 77 fixed_size_simd_mask<T, simd_size_v<T, Abi>> to_fixed_size(const simd_mask<T, Abi>&) noexcept; 78 template <class T, size_t N> native_simd<T> to_native(const fixed_size_simd<T, N>&) noexcept; 79 template <class T, size_t N> 80 native_simd_mask<T> to_native(const fixed_size_simd_mask<T, N>> &) noexcept; 81 template <class T, size_t N> simd<T> to_compatible(const fixed_size_simd<T, N>&) noexcept; 82 template <class T, size_t N> simd_mask<T> to_compatible(const fixed_size_simd_mask<T, N>&) noexcept; 83 84 template <size_t... Sizes, class T, class Abi> 85 tuple<simd<T, abi_for_size_t<Sizes>>...> split(const simd<T, Abi>&); 86 template <size_t... Sizes, class T, class Abi> 87 tuple<simd_mask<T, abi_for_size_t<Sizes>>...> split(const simd_mask<T, Abi>&); 88 template <class V, class Abi> 89 array<V, simd_size_v<typename V::value_type, Abi> / V::size()> split( 90 const simd<typename V::value_type, Abi>&); 91 template <class V, class Abi> 92 array<V, simd_size_v<typename V::value_type, Abi> / V::size()> split( 93 const simd_mask<typename V::value_type, Abi>&); 94 95 template <class T, class... Abis> 96 simd<T, abi_for_size_t<T, (simd_size_v<T, Abis> + ...)>> concat(const simd<T, Abis>&...); 97 template <class T, class... Abis> 98 simd_mask<T, abi_for_size_t<T, (simd_size_v<T, Abis> + ...)>> concat(const simd_mask<T, Abis>&...); 99 100 // reductions [simd.mask.reductions] 101 template <class T, class Abi> bool all_of(const simd_mask<T, Abi>&) noexcept; 102 template <class T, class Abi> bool any_of(const simd_mask<T, Abi>&) noexcept; 103 template <class T, class Abi> bool none_of(const simd_mask<T, Abi>&) noexcept; 104 template <class T, class Abi> bool some_of(const simd_mask<T, Abi>&) noexcept; 105 template <class T, class Abi> int popcount(const simd_mask<T, Abi>&) noexcept; 106 template <class T, class Abi> int find_first_set(const simd_mask<T, Abi>&); 107 template <class T, class Abi> int find_last_set(const simd_mask<T, Abi>&); 108 109 bool all_of(see below) noexcept; 110 bool any_of(see below) noexcept; 111 bool none_of(see below) noexcept; 112 bool some_of(see below) noexcept; 113 int popcount(see below) noexcept; 114 int find_first_set(see below) noexcept; 115 int find_last_set(see below) noexcept; 116 117 // masked assignment [simd.whereexpr] 118 template <class M, class T> class const_where_expression; 119 template <class M, class T> class where_expression; 120 121 // masked assignment [simd.mask.where] 122 template <class T> struct nodeduce { using type = T; }; // exposition only 123 124 template <class T> using nodeduce_t = typename nodeduce<T>::type; // exposition only 125 126 template <class T, class Abi> 127 where_expression<simd_mask<T, Abi>, simd<T, Abi>> 128 where(const typename simd<T, Abi>::mask_type&, simd<T, Abi>&) noexcept; 129 130 template <class T, class Abi> 131 const_where_expression<simd_mask<T, Abi>, const simd<T, Abi>> 132 where(const typename simd<T, Abi>::mask_type&, const simd<T, Abi>&) noexcept; 133 134 template <class T, class Abi> 135 where_expression<simd_mask<T, Abi>, simd_mask<T, Abi>> 136 where(const nodeduce_t<simd_mask<T, Abi>>&, simd_mask<T, Abi>&) noexcept; 137 138 template <class T, class Abi> 139 const_where_expression<simd_mask<T, Abi>, const simd_mask<T, Abi>> 140 where(const nodeduce_t<simd_mask<T, Abi>>&, const simd_mask<T, Abi>&) noexcept; 141 142 template <class T> where_expression<bool, T> where(see below k, T& d) noexcept; 143 144 template <class T> 145 const_where_expression<bool, const T> where(see below k, const T& d) noexcept; 146 147 // reductions [simd.reductions] 148 template <class T, class Abi, class BinaryOperation = std::plus<>> 149 T reduce(const simd<T, Abi>&, BinaryOperation = BinaryOperation()); 150 151 template <class M, class V, class BinaryOperation> 152 typename V::value_type reduce(const const_where_expression<M, V>& x, 153 typename V::value_type neutral_element, BinaryOperation binary_op); 154 155 template <class M, class V> 156 typename V::value_type reduce(const const_where_expression<M, V>& x, plus<> binary_op = plus<>()); 157 158 template <class M, class V> 159 typename V::value_type reduce(const const_where_expression<M, V>& x, multiplies<> binary_op); 160 161 template <class M, class V> 162 typename V::value_type reduce(const const_where_expression<M, V>& x, bit_and<> binary_op); 163 164 template <class M, class V> 165 typename V::value_type reduce(const const_where_expression<M, V>& x, bit_or<> binary_op); 166 167 template <class M, class V> 168 typename V::value_type reduce(const const_where_expression<M, V>& x, bit_xor<> binary_op); 169 170 template <class T, class Abi> T hmin(const simd<T, Abi>&); 171 template <class M, class V> T hmin(const const_where_expression<M, V>&); 172 template <class T, class Abi> T hmax(const simd<T, Abi>&); 173 template <class M, class V> T hmax(const const_where_expression<M, V>&); 174 175 // algorithms [simd.alg] 176 template <class T, class Abi> simd<T, Abi> min(const simd<T, Abi>&, const simd<T, Abi>&) noexcept; 177 178 template <class T, class Abi> simd<T, Abi> max(const simd<T, Abi>&, const simd<T, Abi>&) noexcept; 179 180 template <class T, class Abi> 181 std::pair<simd<T, Abi>, simd<T, Abi>> minmax(const simd<T, Abi>&, const simd<T, Abi>&) noexcept; 182 183 template <class T, class Abi> 184 simd<T, Abi> clamp(const simd<T, Abi>& v, const simd<T, Abi>& lo, const simd<T, Abi>& hi); 185 186 // [simd.whereexpr] 187 template <class M, class T> 188 class const_where_expression { 189 const M& mask; // exposition only 190 T& data; // exposition only 191 public: 192 const_where_expression(const const_where_expression&) = delete; 193 const_where_expression& operator=(const const_where_expression&) = delete; 194 remove_const_t<T> operator-() const &&; 195 template <class U, class Flags> void copy_to(U* mem, Flags f) const &&; 196 }; 197 198 template <class M, class T> 199 class where_expression : public const_where_expression<M, T> { 200 public: 201 where_expression(const where_expression&) = delete; 202 where_expression& operator=(const where_expression&) = delete; 203 template <class U> void operator=(U&& x); 204 template <class U> void operator+=(U&& x); 205 template <class U> void operator-=(U&& x); 206 template <class U> void operator*=(U&& x); 207 template <class U> void operator/=(U&& x); 208 template <class U> void operator%=(U&& x); 209 template <class U> void operator&=(U&& x); 210 template <class U> void operator|=(U&& x); 211 template <class U> void operator^=(U&& x); 212 template <class U> void operator<<=(U&& x); 213 template <class U> void operator>>=(U&& x); 214 void operator++(); 215 void operator++(int); 216 void operator--(); 217 void operator--(int); 218 template <class U, class Flags> void copy_from(const U* mem, Flags); 219 }; 220 221 // [simd.class] 222 template <class T, class Abi> class simd { 223 public: 224 using value_type = T; 225 using reference = see below; 226 using mask_type = simd_mask<T, Abi>; 227 228 using abi_type = Abi; 229 static constexpr size_t size() noexcept; 230 simd() = default; 231 232 // implicit type conversion constructor 233 template <class U> simd(const simd<U, simd_abi::fixed_size<size()>>&); 234 235 // implicit broadcast constructor (see below for constraints) 236 template <class U> simd(U&& value); 237 238 // generator constructor (see below for constraints) 239 template <class G> explicit simd(G&& gen); 240 241 // load constructor 242 template <class U, class Flags> simd(const U* mem, Flags f); 243 244 // loads [simd.load] 245 template <class U, class Flags> void copy_from(const U* mem, Flags f); 246 247 // stores [simd.store] 248 template <class U, class Flags> void copy_to(U* mem, Flags f) const; 249 250 // scalar access [simd.subscr] 251 reference operator[](size_t); 252 value_type operator[](size_t) const; 253 254 // unary operators [simd.unary] 255 simd& operator++(); 256 simd operator++(int); 257 simd& operator--(); 258 simd operator--(int); 259 mask_type operator!() const; 260 simd operator~() const; // see below 261 simd operator+() const; 262 simd operator-() const; 263 264 // binary operators [simd.binary] 265 friend simd operator+ (const simd&, const simd&); 266 friend simd operator- (const simd&, const simd&); 267 friend simd operator* (const simd&, const simd&); 268 friend simd operator/ (const simd&, const simd&); 269 friend simd operator% (const simd&, const simd&); 270 friend simd operator& (const simd&, const simd&); 271 friend simd operator| (const simd&, const simd&); 272 friend simd operator^ (const simd&, const simd&); 273 friend simd operator<<(const simd&, const simd&); 274 friend simd operator>>(const simd&, const simd&); 275 friend simd operator<<(const simd&, int); 276 friend simd operator>>(const simd&, int); 277 278 // compound assignment [simd.cassign] 279 friend simd& operator+= (simd&, const simd&); 280 friend simd& operator-= (simd&, const simd&); 281 friend simd& operator*= (simd&, const simd&); 282 friend simd& operator/= (simd&, const simd&); 283 friend simd& operator%= (simd&, const simd&); 284 285 friend simd& operator&= (simd&, const simd&); 286 friend simd& operator|= (simd&, const simd&); 287 friend simd& operator^= (simd&, const simd&); 288 friend simd& operator<<=(simd&, const simd&); 289 friend simd& operator>>=(simd&, const simd&); 290 friend simd& operator<<=(simd&, int); 291 friend simd& operator>>=(simd&, int); 292 293 // compares [simd.comparison] 294 friend mask_type operator==(const simd&, const simd&); 295 friend mask_type operator!=(const simd&, const simd&); 296 friend mask_type operator>=(const simd&, const simd&); 297 friend mask_type operator<=(const simd&, const simd&); 298 friend mask_type operator> (const simd&, const simd&); 299 friend mask_type operator< (const simd&, const simd&); 300 }; 301 302 // [simd.math] 303 template <class Abi> using scharv = simd<signed char, Abi>; // exposition only 304 template <class Abi> using shortv = simd<short, Abi>; // exposition only 305 template <class Abi> using intv = simd<int, Abi>; // exposition only 306 template <class Abi> using longv = simd<long int, Abi>; // exposition only 307 template <class Abi> using llongv = simd<long long int, Abi>; // exposition only 308 template <class Abi> using floatv = simd<float, Abi>; // exposition only 309 template <class Abi> using doublev = simd<double, Abi>; // exposition only 310 template <class Abi> using ldoublev = simd<long double, Abi>; // exposition only 311 template <class T, class V> using samesize = fixed_size_simd<T, V::size()>; // exposition only 312 313 template <class Abi> floatv<Abi> acos(floatv<Abi> x); 314 template <class Abi> doublev<Abi> acos(doublev<Abi> x); 315 template <class Abi> ldoublev<Abi> acos(ldoublev<Abi> x); 316 317 template <class Abi> floatv<Abi> asin(floatv<Abi> x); 318 template <class Abi> doublev<Abi> asin(doublev<Abi> x); 319 template <class Abi> ldoublev<Abi> asin(ldoublev<Abi> x); 320 321 template <class Abi> floatv<Abi> atan(floatv<Abi> x); 322 template <class Abi> doublev<Abi> atan(doublev<Abi> x); 323 template <class Abi> ldoublev<Abi> atan(ldoublev<Abi> x); 324 325 template <class Abi> floatv<Abi> atan2(floatv<Abi> y, floatv<Abi> x); 326 template <class Abi> doublev<Abi> atan2(doublev<Abi> y, doublev<Abi> x); 327 template <class Abi> ldoublev<Abi> atan2(ldoublev<Abi> y, ldoublev<Abi> x); 328 329 template <class Abi> floatv<Abi> cos(floatv<Abi> x); 330 template <class Abi> doublev<Abi> cos(doublev<Abi> x); 331 template <class Abi> ldoublev<Abi> cos(ldoublev<Abi> x); 332 333 template <class Abi> floatv<Abi> sin(floatv<Abi> x); 334 template <class Abi> doublev<Abi> sin(doublev<Abi> x); 335 template <class Abi> ldoublev<Abi> sin(ldoublev<Abi> x); 336 337 template <class Abi> floatv<Abi> tan(floatv<Abi> x); 338 template <class Abi> doublev<Abi> tan(doublev<Abi> x); 339 template <class Abi> ldoublev<Abi> tan(ldoublev<Abi> x); 340 341 template <class Abi> floatv<Abi> acosh(floatv<Abi> x); 342 template <class Abi> doublev<Abi> acosh(doublev<Abi> x); 343 template <class Abi> ldoublev<Abi> acosh(ldoublev<Abi> x); 344 345 template <class Abi> floatv<Abi> asinh(floatv<Abi> x); 346 template <class Abi> doublev<Abi> asinh(doublev<Abi> x); 347 template <class Abi> ldoublev<Abi> asinh(ldoublev<Abi> x); 348 349 template <class Abi> floatv<Abi> atanh(floatv<Abi> x); 350 template <class Abi> doublev<Abi> atanh(doublev<Abi> x); 351 template <class Abi> ldoublev<Abi> atanh(ldoublev<Abi> x); 352 353 template <class Abi> floatv<Abi> cosh(floatv<Abi> x); 354 template <class Abi> doublev<Abi> cosh(doublev<Abi> x); 355 template <class Abi> ldoublev<Abi> cosh(ldoublev<Abi> x); 356 357 template <class Abi> floatv<Abi> sinh(floatv<Abi> x); 358 template <class Abi> doublev<Abi> sinh(doublev<Abi> x); 359 template <class Abi> ldoublev<Abi> sinh(ldoublev<Abi> x); 360 361 template <class Abi> floatv<Abi> tanh(floatv<Abi> x); 362 template <class Abi> doublev<Abi> tanh(doublev<Abi> x); 363 template <class Abi> ldoublev<Abi> tanh(ldoublev<Abi> x); 364 365 template <class Abi> floatv<Abi> exp(floatv<Abi> x); 366 template <class Abi> doublev<Abi> exp(doublev<Abi> x); 367 template <class Abi> ldoublev<Abi> exp(ldoublev<Abi> x); 368 369 template <class Abi> floatv<Abi> exp2(floatv<Abi> x); 370 template <class Abi> doublev<Abi> exp2(doublev<Abi> x); 371 template <class Abi> ldoublev<Abi> exp2(ldoublev<Abi> x); 372 373 template <class Abi> floatv<Abi> expm1(floatv<Abi> x); 374 template <class Abi> doublev<Abi> expm1(doublev<Abi> x); 375 template <class Abi> ldoublev<Abi> expm1(ldoublev<Abi> x); 376 377 template <class Abi> floatv<Abi> frexp(floatv<Abi> value, samesize<int, floatv<Abi>>* exp); 378 template <class Abi> doublev<Abi> frexp(doublev<Abi> value, samesize<int, doublev<Abi>>* exp); 379 template <class Abi> ldoublev<Abi> frexp(ldoublev<Abi> value, samesize<int, ldoublev<Abi>>* exp); 380 381 template <class Abi> samesize<int, floatv<Abi>> ilogb(floatv<Abi> x); 382 template <class Abi> samesize<int, doublev<Abi>> ilogb(doublev<Abi> x); 383 template <class Abi> samesize<int, ldoublev<Abi>> ilogb(ldoublev<Abi> x); 384 385 template <class Abi> floatv<Abi> ldexp(floatv<Abi> x, samesize<int, floatv<Abi>> exp); 386 template <class Abi> doublev<Abi> ldexp(doublev<Abi> x, samesize<int, doublev<Abi>> exp); 387 template <class Abi> ldoublev<Abi> ldexp(ldoublev<Abi> x, samesize<int, ldoublev<Abi>> exp); 388 389 template <class Abi> floatv<Abi> log(floatv<Abi> x); 390 template <class Abi> doublev<Abi> log(doublev<Abi> x); 391 template <class Abi> ldoublev<Abi> log(ldoublev<Abi> x); 392 393 template <class Abi> floatv<Abi> log10(floatv<Abi> x); 394 template <class Abi> doublev<Abi> log10(doublev<Abi> x); 395 template <class Abi> ldoublev<Abi> log10(ldoublev<Abi> x); 396 397 template <class Abi> floatv<Abi> log1p(floatv<Abi> x); 398 template <class Abi> doublev<Abi> log1p(doublev<Abi> x); 399 template <class Abi> ldoublev<Abi> log1p(ldoublev<Abi> x); 400 401 template <class Abi> floatv<Abi> log2(floatv<Abi> x); 402 template <class Abi> doublev<Abi> log2(doublev<Abi> x); 403 template <class Abi> ldoublev<Abi> log2(ldoublev<Abi> x); 404 405 template <class Abi> floatv<Abi> logb(floatv<Abi> x); 406 template <class Abi> doublev<Abi> logb(doublev<Abi> x); 407 template <class Abi> ldoublev<Abi> logb(ldoublev<Abi> x); 408 409 template <class Abi> floatv<Abi> modf(floatv<Abi> value, floatv<Abi>* iptr); 410 template <class Abi> doublev<Abi> modf(doublev<Abi> value, doublev<Abi>* iptr); 411 template <class Abi> ldoublev<Abi> modf(ldoublev<Abi> value, ldoublev<Abi>* iptr); 412 413 template <class Abi> floatv<Abi> scalbn(floatv<Abi> x, samesize<int, floatv<Abi>> n); 414 template <class Abi> doublev<Abi> scalbn(doublev<Abi> x, samesize<int, doublev<Abi>> n); 415 template <class Abi> ldoublev<Abi> scalbn(ldoublev<Abi> x, samesize<int, ldoublev<Abi>> n); 416 template <class Abi> floatv<Abi> scalbln(floatv<Abi> x, samesize<long int, floatv<Abi>> n); 417 template <class Abi> doublev<Abi> scalbln(doublev<Abi> x, samesize<long int, doublev<Abi>> n); 418 template <class Abi> ldoublev<Abi> scalbln(ldoublev<Abi> x, samesize<long int, ldoublev<Abi>> n); 419 420 template <class Abi> floatv<Abi> cbrt(floatv<Abi> x); 421 template <class Abi> doublev<Abi> cbrt(doublev<Abi> x); 422 template <class Abi> ldoublev<Abi> cbrt(ldoublev<Abi> x); 423 424 template <class Abi> scharv<Abi> abs(scharv<Abi> j); 425 template <class Abi> shortv<Abi> abs(shortv<Abi> j); 426 template <class Abi> intv<Abi> abs(intv<Abi> j); 427 template <class Abi> longv<Abi> abs(longv<Abi> j); 428 template <class Abi> llongv<Abi> abs(llongv<Abi> j); 429 template <class Abi> floatv<Abi> abs(floatv<Abi> j); 430 template <class Abi> doublev<Abi> abs(doublev<Abi> j); 431 template <class Abi> ldoublev<Abi> abs(ldoublev<Abi> j); 432 433 template <class Abi> floatv<Abi> hypot(floatv<Abi> x, floatv<Abi> y); 434 template <class Abi> doublev<Abi> hypot(doublev<Abi> x, doublev<Abi> y); 435 template <class Abi> ldoublev<Abi> hypot(doublev<Abi> x, doublev<Abi> y); 436 template <class Abi> floatv<Abi> hypot(floatv<Abi> x, floatv<Abi> y, floatv<Abi> z); 437 template <class Abi> doublev<Abi> hypot(doublev<Abi> x, doublev<Abi> y, doublev<Abi> z); 438 template <class Abi> ldoublev<Abi> hypot(ldoublev<Abi> x, ldoublev<Abi> y, ldoublev<Abi> z); 439 440 template <class Abi> floatv<Abi> pow(floatv<Abi> x, floatv<Abi> y); 441 template <class Abi> doublev<Abi> pow(doublev<Abi> x, doublev<Abi> y); 442 template <class Abi> ldoublev<Abi> pow(ldoublev<Abi> x, ldoublev<Abi> y); 443 444 template <class Abi> floatv<Abi> sqrt(floatv<Abi> x); 445 template <class Abi> doublev<Abi> sqrt(doublev<Abi> x); 446 template <class Abi> ldoublev<Abi> sqrt(ldoublev<Abi> x); 447 448 template <class Abi> floatv<Abi> erf(floatv<Abi> x); 449 template <class Abi> doublev<Abi> erf(doublev<Abi> x); 450 template <class Abi> ldoublev<Abi> erf(ldoublev<Abi> x); 451 template <class Abi> floatv<Abi> erfc(floatv<Abi> x); 452 template <class Abi> doublev<Abi> erfc(doublev<Abi> x); 453 template <class Abi> ldoublev<Abi> erfc(ldoublev<Abi> x); 454 455 template <class Abi> floatv<Abi> lgamma(floatv<Abi> x); 456 template <class Abi> doublev<Abi> lgamma(doublev<Abi> x); 457 template <class Abi> ldoublev<Abi> lgamma(ldoublev<Abi> x); 458 459 template <class Abi> floatv<Abi> tgamma(floatv<Abi> x); 460 template <class Abi> doublev<Abi> tgamma(doublev<Abi> x); 461 template <class Abi> ldoublev<Abi> tgamma(ldoublev<Abi> x); 462 463 template <class Abi> floatv<Abi> ceil(floatv<Abi> x); 464 template <class Abi> doublev<Abi> ceil(doublev<Abi> x); 465 template <class Abi> ldoublev<Abi> ceil(ldoublev<Abi> x); 466 467 template <class Abi> floatv<Abi> floor(floatv<Abi> x); 468 template <class Abi> doublev<Abi> floor(doublev<Abi> x); 469 template <class Abi> ldoublev<Abi> floor(ldoublev<Abi> x); 470 471 template <class Abi> floatv<Abi> nearbyint(floatv<Abi> x); 472 template <class Abi> doublev<Abi> nearbyint(doublev<Abi> x); 473 template <class Abi> ldoublev<Abi> nearbyint(ldoublev<Abi> x); 474 475 template <class Abi> floatv<Abi> rint(floatv<Abi> x); 476 template <class Abi> doublev<Abi> rint(doublev<Abi> x); 477 template <class Abi> ldoublev<Abi> rint(ldoublev<Abi> x); 478 479 template <class Abi> samesize<long int, floatv<Abi>> lrint(floatv<Abi> x); 480 template <class Abi> samesize<long int, doublev<Abi>> lrint(doublev<Abi> x); 481 template <class Abi> samesize<long int, ldoublev<Abi>> lrint(ldoublev<Abi> x); 482 template <class Abi> samesize<long long int, floatv<Abi>> llrint(floatv<Abi> x); 483 template <class Abi> samesize<long long int, doublev<Abi>> llrint(doublev<Abi> x); 484 template <class Abi> samesize<long long int, ldoublev<Abi>> llrint(ldoublev<Abi> x); 485 486 template <class Abi> floatv<Abi> round(floatv<Abi> x); 487 template <class Abi> doublev<Abi> round(doublev<Abi> x); 488 template <class Abi> ldoublev<Abi> round(ldoublev<Abi> x); 489 template <class Abi> samesize<long int, floatv<Abi>> lround(floatv<Abi> x); 490 template <class Abi> samesize<long int, doublev<Abi>> lround(doublev<Abi> x); 491 template <class Abi> samesize<long int, ldoublev<Abi>> lround(ldoublev<Abi> x); 492 template <class Abi> samesize<long long int, floatv<Abi>> llround(floatv<Abi> x); 493 template <class Abi> samesize<long long int, doublev<Abi>> llround(doublev<Abi> x); 494 template <class Abi> samesize<long long int, ldoublev<Abi>> llround(ldoublev<Abi> x); 495 496 template <class Abi> floatv<Abi> trunc(floatv<Abi> x); 497 template <class Abi> doublev<Abi> trunc(doublev<Abi> x); 498 template <class Abi> ldoublev<Abi> trunc(ldoublev<Abi> x); 499 500 template <class Abi> floatv<Abi> fmod(floatv<Abi> x, floatv<Abi> y); 501 template <class Abi> doublev<Abi> fmod(doublev<Abi> x, doublev<Abi> y); 502 template <class Abi> ldoublev<Abi> fmod(ldoublev<Abi> x, ldoublev<Abi> y); 503 504 template <class Abi> floatv<Abi> remainder(floatv<Abi> x, floatv<Abi> y); 505 template <class Abi> doublev<Abi> remainder(doublev<Abi> x, doublev<Abi> y); 506 template <class Abi> ldoublev<Abi> remainder(ldoublev<Abi> x, ldoublev<Abi> y); 507 508 template <class Abi> floatv<Abi> remquo(floatv<Abi> x, floatv<Abi> y, samesize<int, floatv<Abi>>* quo); 509 template <class Abi> doublev<Abi> remquo(doublev<Abi> x, doublev<Abi> y, samesize<int, doublev<Abi>>* quo); 510 template <class Abi> ldoublev<Abi> remquo(ldoublev<Abi> x, ldoublev<Abi> y, samesize<int, ldoublev<Abi>>* quo); 511 512 template <class Abi> floatv<Abi> copysign(floatv<Abi> x, floatv<Abi> y); 513 template <class Abi> doublev<Abi> copysign(doublev<Abi> x, doublev<Abi> y); 514 template <class Abi> ldoublev<Abi> copysign(ldoublev<Abi> x, ldoublev<Abi> y); 515 516 template <class Abi> doublev<Abi> nan(const char* tagp); 517 template <class Abi> floatv<Abi> nanf(const char* tagp); 518 template <class Abi> ldoublev<Abi> nanl(const char* tagp); 519 520 template <class Abi> floatv<Abi> nextafter(floatv<Abi> x, floatv<Abi> y); 521 template <class Abi> doublev<Abi> nextafter(doublev<Abi> x, doublev<Abi> y); 522 template <class Abi> ldoublev<Abi> nextafter(ldoublev<Abi> x, ldoublev<Abi> y); 523 524 template <class Abi> floatv<Abi> nexttoward(floatv<Abi> x, ldoublev<Abi> y); 525 template <class Abi> doublev<Abi> nexttoward(doublev<Abi> x, ldoublev<Abi> y); 526 template <class Abi> ldoublev<Abi> nexttoward(ldoublev<Abi> x, ldoublev<Abi> y); 527 528 template <class Abi> floatv<Abi> fdim(floatv<Abi> x, floatv<Abi> y); 529 template <class Abi> doublev<Abi> fdim(doublev<Abi> x, doublev<Abi> y); 530 template <class Abi> ldoublev<Abi> fdim(ldoublev<Abi> x, ldoublev<Abi> y); 531 532 template <class Abi> floatv<Abi> fmax(floatv<Abi> x, floatv<Abi> y); 533 template <class Abi> doublev<Abi> fmax(doublev<Abi> x, doublev<Abi> y); 534 template <class Abi> ldoublev<Abi> fmax(ldoublev<Abi> x, ldoublev<Abi> y); 535 536 template <class Abi> floatv<Abi> fmin(floatv<Abi> x, floatv<Abi> y); 537 template <class Abi> doublev<Abi> fmin(doublev<Abi> x, doublev<Abi> y); 538 template <class Abi> ldoublev<Abi> fmin(ldoublev<Abi> x, ldoublev<Abi> y); 539 540 template <class Abi> floatv<Abi> fma(floatv<Abi> x, floatv<Abi> y, floatv<Abi> z); 541 template <class Abi> doublev<Abi> fma(doublev<Abi> x, doublev<Abi> y, doublev<Abi> z); 542 template <class Abi> ldoublev<Abi> fma(ldoublev<Abi> x, ldoublev<Abi> y, ldoublev<Abi> z); 543 544 template <class Abi> samesize<int, floatv<Abi>> fpclassify(floatv<Abi> x); 545 template <class Abi> samesize<int, doublev<Abi>> fpclassify(doublev<Abi> x); 546 template <class Abi> samesize<int, ldoublev<Abi>> fpclassify(ldoublev<Abi> x); 547 548 template <class Abi> simd_mask<float, Abi> isfinite(floatv<Abi> x); 549 template <class Abi> simd_mask<double, Abi> isfinite(doublev<Abi> x); 550 template <class Abi> simd_mask<long double, Abi> isfinite(ldoublev<Abi> x); 551 552 template <class Abi> simd_mask<float, Abi> isinf(floatv<Abi> x); 553 template <class Abi> simd_mask<double, Abi> isinf(doublev<Abi> x); 554 template <class Abi> simd_mask<long double, Abi> isinf(ldoublev<Abi> x); 555 556 template <class Abi> simd_mask<float, Abi> isnan(floatv<Abi> x); 557 template <class Abi> simd_mask<double, Abi> isnan(doublev<Abi> x); 558 template <class Abi> simd_mask<long double, Abi> isnan(ldoublev<Abi> x); 559 560 template <class Abi> simd_mask<float, Abi> isnormal(floatv<Abi> x); 561 template <class Abi> simd_mask<double, Abi> isnormal(doublev<Abi> x); 562 template <class Abi> simd_mask<long double, Abi> isnormal(ldoublev<Abi> x); 563 564 template <class Abi> simd_mask<float, Abi> signbit(floatv<Abi> x); 565 template <class Abi> simd_mask<double, Abi> signbit(doublev<Abi> x); 566 template <class Abi> simd_mask<long double, Abi> signbit(ldoublev<Abi> x); 567 568 template <class Abi> simd_mask<float, Abi> isgreater(floatv<Abi> x, floatv<Abi> y); 569 template <class Abi> simd_mask<double, Abi> isgreater(doublev<Abi> x, doublev<Abi> y); 570 template <class Abi> simd_mask<long double, Abi> isgreater(ldoublev<Abi> x, ldoublev<Abi> y); 571 572 template <class Abi> simd_mask<float, Abi> isgreaterequal(floatv<Abi> x, floatv<Abi> y); 573 template <class Abi> simd_mask<double, Abi> isgreaterequal(doublev<Abi> x, doublev<Abi> y); 574 template <class Abi> simd_mask<long double, Abi> isgreaterequal(ldoublev<Abi> x, ldoublev<Abi> y); 575 576 template <class Abi> simd_mask<float, Abi> isless(floatv<Abi> x, floatv<Abi> y); 577 template <class Abi> simd_mask<double, Abi> isless(doublev<Abi> x, doublev<Abi> y); 578 template <class Abi> simd_mask<long double, Abi> isless(ldoublev<Abi> x, ldoublev<Abi> y); 579 580 template <class Abi> simd_mask<float, Abi> islessequal(floatv<Abi> x, floatv<Abi> y); 581 template <class Abi> simd_mask<double, Abi> islessequal(doublev<Abi> x, doublev<Abi> y); 582 template <class Abi> simd_mask<long double, Abi> islessequal(ldoublev<Abi> x, ldoublev<Abi> y); 583 584 template <class Abi> simd_mask<float, Abi> islessgreater(floatv<Abi> x, floatv<Abi> y); 585 template <class Abi> simd_mask<double, Abi> islessgreater(doublev<Abi> x, doublev<Abi> y); 586 template <class Abi> simd_mask<long double, Abi> islessgreater(ldoublev<Abi> x, ldoublev<Abi> y); 587 588 template <class Abi> simd_mask<float, Abi> isunordered(floatv<Abi> x, floatv<Abi> y); 589 template <class Abi> simd_mask<double, Abi> isunordered(doublev<Abi> x, doublev<Abi> y); 590 template <class Abi> simd_mask<long double, Abi> isunordered(ldoublev<Abi> x, ldoublev<Abi> y); 591 592 template <class V> struct simd_div_t { V quot, rem; }; 593 template <class Abi> simd_div_t<scharv<Abi>> div(scharv<Abi> numer, scharv<Abi> denom); 594 template <class Abi> simd_div_t<shortv<Abi>> div(shortv<Abi> numer, shortv<Abi> denom); 595 template <class Abi> simd_div_t<intv<Abi>> div(intv<Abi> numer, intv<Abi> denom); 596 template <class Abi> simd_div_t<longv<Abi>> div(longv<Abi> numer, longv<Abi> denom); 597 template <class Abi> simd_div_t<llongv<Abi>> div(llongv<Abi> numer, llongv<Abi> denom); 598 599 // [simd.mask.class] 600 template <class T, class Abi> 601 class simd_mask { 602 public: 603 using value_type = bool; 604 using reference = see below; 605 using simd_type = simd<T, Abi>; 606 using abi_type = Abi; 607 static constexpr size_t size() noexcept; 608 simd_mask() = default; 609 610 // broadcast constructor 611 explicit simd_mask(value_type) noexcept; 612 613 // implicit type conversion constructor 614 template <class U> simd_mask(const simd_mask<U, simd_abi::fixed_size<size()>>&) noexcept; 615 616 // load constructor 617 template <class Flags> simd_mask(const value_type* mem, Flags); 618 619 // loads [simd.mask.copy] 620 template <class Flags> void copy_from(const value_type* mem, Flags); 621 template <class Flags> void copy_to(value_type* mem, Flags) const; 622 623 // scalar access [simd.mask.subscr] 624 reference operator[](size_t); 625 value_type operator[](size_t) const; 626 627 // unary operators [simd.mask.unary] 628 simd_mask operator!() const noexcept; 629 630 // simd_mask binary operators [simd.mask.binary] 631 friend simd_mask operator&&(const simd_mask&, const simd_mask&) noexcept; 632 friend simd_mask operator||(const simd_mask&, const simd_mask&) noexcept; 633 friend simd_mask operator& (const simd_mask&, const simd_mask&) noexcept; 634 friend simd_mask operator| (const simd_mask&, const simd_mask&) noexcept; 635 friend simd_mask operator^ (const simd_mask&, const simd_mask&) noexcept; 636 637 // simd_mask compound assignment [simd.mask.cassign] 638 friend simd_mask& operator&=(simd_mask&, const simd_mask&) noexcept; 639 friend simd_mask& operator|=(simd_mask&, const simd_mask&) noexcept; 640 friend simd_mask& operator^=(simd_mask&, const simd_mask&) noexcept; 641 642 // simd_mask compares [simd.mask.comparison] 643 friend simd_mask operator==(const simd_mask&, const simd_mask&) noexcept; 644 friend simd_mask operator!=(const simd_mask&, const simd_mask&) noexcept; 645 }; 646 647 } // parallelism_v2 648 } // std::experimental 649 650 */ 651 652 #include <experimental/__config> 653 #include <algorithm> 654 #include <array> 655 #include <cstddef> 656 #include <functional> 657 658 #if !defined(_LIBCPP_HAS_NO_PRAGMA_SYSTEM_HEADER) 659 #pragma GCC system_header 660 #endif 661 662 _LIBCPP_PUSH_MACROS 663 #include <__undef_macros> 664 665 _LIBCPP_BEGIN_NAMESPACE_EXPERIMENTAL_SIMD 666 667 #if _LIBCPP_STD_VER >= 17 668 669 enum class _StorageKind { 670 _Scalar, 671 _Array, 672 _VecExt, 673 }; 674 675 template <_StorageKind __kind, int _Np> 676 struct __simd_abi {}; 677 678 template <class _Tp, class _Abi> 679 class __simd_storage {}; 680 681 template <class _Tp, int __num_element> 682 class __simd_storage<_Tp, __simd_abi<_StorageKind::_Array, __num_element>> { 683 std::array<_Tp, __num_element> __storage_; 684 685 template <class, class> 686 friend struct simd; 687 688 template <class, class> 689 friend struct simd_mask; 690 691 public: 692 _Tp __get(size_t __index) const noexcept { return __storage_[__index]; }; 693 void __set(size_t __index, _Tp __val) noexcept { 694 __storage_[__index] = __val; 695 } 696 }; 697 698 template <class _Tp> 699 class __simd_storage<_Tp, __simd_abi<_StorageKind::_Scalar, 1>> { 700 _Tp __storage_; 701 702 template <class, class> 703 friend struct simd; 704 705 template <class, class> 706 friend struct simd_mask; 707 708 public: 709 _Tp __get(size_t __index) const noexcept { return (&__storage_)[__index]; }; 710 void __set(size_t __index, _Tp __val) noexcept { 711 (&__storage_)[__index] = __val; 712 } 713 }; 714 715 #ifndef _LIBCPP_HAS_NO_VECTOR_EXTENSION 716 717 constexpr size_t __floor_pow_of_2(size_t __val) { 718 return ((__val - 1) & __val) == 0 ? __val 719 : __floor_pow_of_2((__val - 1) & __val); 720 } 721 722 constexpr size_t __ceil_pow_of_2(size_t __val) { 723 return __val == 1 ? 1 : __floor_pow_of_2(__val - 1) << 1; 724 } 725 726 template <class _Tp, size_t __bytes> 727 struct __vec_ext_traits { 728 #if !defined(_LIBCPP_COMPILER_CLANG_BASED) 729 typedef _Tp type __attribute__((vector_size(__ceil_pow_of_2(__bytes)))); 730 #endif 731 }; 732 733 #if defined(_LIBCPP_COMPILER_CLANG_BASED) 734 #define _LIBCPP_SPECIALIZE_VEC_EXT(_TYPE, _NUM_ELEMENT) \ 735 template <> \ 736 struct __vec_ext_traits<_TYPE, sizeof(_TYPE) * _NUM_ELEMENT> { \ 737 using type = \ 738 _TYPE __attribute__((vector_size(sizeof(_TYPE) * _NUM_ELEMENT))); \ 739 } 740 741 #define _LIBCPP_SPECIALIZE_VEC_EXT_32(_TYPE) \ 742 _LIBCPP_SPECIALIZE_VEC_EXT(_TYPE, 1); \ 743 _LIBCPP_SPECIALIZE_VEC_EXT(_TYPE, 2); \ 744 _LIBCPP_SPECIALIZE_VEC_EXT(_TYPE, 3); \ 745 _LIBCPP_SPECIALIZE_VEC_EXT(_TYPE, 4); \ 746 _LIBCPP_SPECIALIZE_VEC_EXT(_TYPE, 5); \ 747 _LIBCPP_SPECIALIZE_VEC_EXT(_TYPE, 6); \ 748 _LIBCPP_SPECIALIZE_VEC_EXT(_TYPE, 7); \ 749 _LIBCPP_SPECIALIZE_VEC_EXT(_TYPE, 8); \ 750 _LIBCPP_SPECIALIZE_VEC_EXT(_TYPE, 9); \ 751 _LIBCPP_SPECIALIZE_VEC_EXT(_TYPE, 10); \ 752 _LIBCPP_SPECIALIZE_VEC_EXT(_TYPE, 11); \ 753 _LIBCPP_SPECIALIZE_VEC_EXT(_TYPE, 12); \ 754 _LIBCPP_SPECIALIZE_VEC_EXT(_TYPE, 13); \ 755 _LIBCPP_SPECIALIZE_VEC_EXT(_TYPE, 14); \ 756 _LIBCPP_SPECIALIZE_VEC_EXT(_TYPE, 15); \ 757 _LIBCPP_SPECIALIZE_VEC_EXT(_TYPE, 16); \ 758 _LIBCPP_SPECIALIZE_VEC_EXT(_TYPE, 17); \ 759 _LIBCPP_SPECIALIZE_VEC_EXT(_TYPE, 18); \ 760 _LIBCPP_SPECIALIZE_VEC_EXT(_TYPE, 19); \ 761 _LIBCPP_SPECIALIZE_VEC_EXT(_TYPE, 20); \ 762 _LIBCPP_SPECIALIZE_VEC_EXT(_TYPE, 21); \ 763 _LIBCPP_SPECIALIZE_VEC_EXT(_TYPE, 22); \ 764 _LIBCPP_SPECIALIZE_VEC_EXT(_TYPE, 23); \ 765 _LIBCPP_SPECIALIZE_VEC_EXT(_TYPE, 24); \ 766 _LIBCPP_SPECIALIZE_VEC_EXT(_TYPE, 25); \ 767 _LIBCPP_SPECIALIZE_VEC_EXT(_TYPE, 26); \ 768 _LIBCPP_SPECIALIZE_VEC_EXT(_TYPE, 27); \ 769 _LIBCPP_SPECIALIZE_VEC_EXT(_TYPE, 28); \ 770 _LIBCPP_SPECIALIZE_VEC_EXT(_TYPE, 29); \ 771 _LIBCPP_SPECIALIZE_VEC_EXT(_TYPE, 30); \ 772 _LIBCPP_SPECIALIZE_VEC_EXT(_TYPE, 31); \ 773 _LIBCPP_SPECIALIZE_VEC_EXT(_TYPE, 32); 774 775 _LIBCPP_SPECIALIZE_VEC_EXT_32(char); 776 _LIBCPP_SPECIALIZE_VEC_EXT_32(char16_t); 777 _LIBCPP_SPECIALIZE_VEC_EXT_32(char32_t); 778 _LIBCPP_SPECIALIZE_VEC_EXT_32(wchar_t); 779 _LIBCPP_SPECIALIZE_VEC_EXT_32(signed char); 780 _LIBCPP_SPECIALIZE_VEC_EXT_32(signed short); 781 _LIBCPP_SPECIALIZE_VEC_EXT_32(signed int); 782 _LIBCPP_SPECIALIZE_VEC_EXT_32(signed long); 783 _LIBCPP_SPECIALIZE_VEC_EXT_32(signed long long); 784 _LIBCPP_SPECIALIZE_VEC_EXT_32(unsigned char); 785 _LIBCPP_SPECIALIZE_VEC_EXT_32(unsigned short); 786 _LIBCPP_SPECIALIZE_VEC_EXT_32(unsigned int); 787 _LIBCPP_SPECIALIZE_VEC_EXT_32(unsigned long); 788 _LIBCPP_SPECIALIZE_VEC_EXT_32(unsigned long long); 789 _LIBCPP_SPECIALIZE_VEC_EXT_32(float); 790 _LIBCPP_SPECIALIZE_VEC_EXT_32(double); 791 _LIBCPP_SPECIALIZE_VEC_EXT_32(long double); 792 793 #undef _LIBCPP_SPECIALIZE_VEC_EXT_32 794 #undef _LIBCPP_SPECIALIZE_VEC_EXT 795 #endif 796 797 template <class _Tp, int __num_element> 798 class __simd_storage<_Tp, __simd_abi<_StorageKind::_VecExt, __num_element>> { 799 using _StorageType = 800 typename __vec_ext_traits<_Tp, sizeof(_Tp) * __num_element>::type; 801 802 _StorageType __storage_; 803 804 template <class, class> 805 friend struct simd; 806 807 template <class, class> 808 friend struct simd_mask; 809 810 public: 811 _Tp __get(size_t __index) const noexcept { return __storage_[__index]; }; 812 void __set(size_t __index, _Tp __val) noexcept { 813 __storage_[__index] = __val; 814 } 815 }; 816 817 #endif // _LIBCPP_HAS_NO_VECTOR_EXTENSION 818 819 template <class _Vp, class _Tp, class _Abi> 820 class __simd_reference { 821 static_assert(std::is_same<_Vp, _Tp>::value, ""); 822 823 template <class, class> 824 friend struct simd; 825 826 template <class, class> 827 friend struct simd_mask; 828 829 __simd_storage<_Tp, _Abi>* __ptr_; 830 size_t __index_; 831 832 __simd_reference(__simd_storage<_Tp, _Abi>* __ptr, size_t __index) 833 : __ptr_(__ptr), __index_(__index) {} 834 835 __simd_reference(const __simd_reference&) = default; 836 837 public: 838 __simd_reference() = delete; 839 __simd_reference& operator=(const __simd_reference&) = delete; 840 841 operator _Vp() const { return __ptr_->__get(__index_); } 842 843 __simd_reference operator=(_Vp __value) && { 844 __ptr_->__set(__index_, __value); 845 return *this; 846 } 847 848 __simd_reference operator++() && { 849 return std::move(*this) = __ptr_->__get(__index_) + 1; 850 } 851 852 _Vp operator++(int) && { 853 auto __val = __ptr_->__get(__index_); 854 __ptr_->__set(__index_, __val + 1); 855 return __val; 856 } 857 858 __simd_reference operator--() && { 859 return std::move(*this) = __ptr_->__get(__index_) - 1; 860 } 861 862 _Vp operator--(int) && { 863 auto __val = __ptr_->__get(__index_); 864 __ptr_->__set(__index_, __val - 1); 865 return __val; 866 } 867 868 __simd_reference operator+=(_Vp __value) && { 869 return std::move(*this) = __ptr_->__get(__index_) + __value; 870 } 871 872 __simd_reference operator-=(_Vp __value) && { 873 return std::move(*this) = __ptr_->__get(__index_) - __value; 874 } 875 876 __simd_reference operator*=(_Vp __value) && { 877 return std::move(*this) = __ptr_->__get(__index_) * __value; 878 } 879 880 __simd_reference operator/=(_Vp __value) && { 881 return std::move(*this) = __ptr_->__get(__index_) / __value; 882 } 883 884 __simd_reference operator%=(_Vp __value) && { 885 return std::move(*this) = __ptr_->__get(__index_) % __value; 886 } 887 888 __simd_reference operator>>=(_Vp __value) && { 889 return std::move(*this) = __ptr_->__get(__index_) >> __value; 890 } 891 892 __simd_reference operator<<=(_Vp __value) && { 893 return std::move(*this) = __ptr_->__get(__index_) << __value; 894 } 895 896 __simd_reference operator&=(_Vp __value) && { 897 return std::move(*this) = __ptr_->__get(__index_) & __value; 898 } 899 900 __simd_reference operator|=(_Vp __value) && { 901 return std::move(*this) = __ptr_->__get(__index_) | __value; 902 } 903 904 __simd_reference operator^=(_Vp __value) && { 905 return std::move(*this) = __ptr_->__get(__index_) ^ __value; 906 } 907 }; 908 909 template <class _To, class _From> 910 constexpr decltype(_To{std::declval<_From>()}, true) 911 __is_non_narrowing_convertible_impl(_From) { 912 return true; 913 } 914 915 template <class _To> 916 constexpr bool __is_non_narrowing_convertible_impl(...) { 917 return false; 918 } 919 920 template <class _From, class _To> 921 constexpr typename std::enable_if<std::is_arithmetic<_To>::value && 922 std::is_arithmetic<_From>::value, 923 bool>::type 924 __is_non_narrowing_arithmetic_convertible() { 925 return __is_non_narrowing_convertible_impl<_To>(_From{}); 926 } 927 928 template <class _From, class _To> 929 constexpr typename std::enable_if<!(std::is_arithmetic<_To>::value && 930 std::is_arithmetic<_From>::value), 931 bool>::type 932 __is_non_narrowing_arithmetic_convertible() { 933 return false; 934 } 935 936 template <class _Tp> 937 constexpr _Tp __variadic_sum() { 938 return _Tp{}; 939 } 940 941 template <class _Tp, class _Up, class... _Args> 942 constexpr _Tp __variadic_sum(_Up __first, _Args... __rest) { 943 return static_cast<_Tp>(__first) + __variadic_sum<_Tp>(__rest...); 944 } 945 946 template <class _Tp> 947 struct __nodeduce { 948 using type = _Tp; 949 }; 950 951 template <class _Tp> 952 constexpr bool __vectorizable() { 953 return std::is_arithmetic<_Tp>::value && !std::is_const<_Tp>::value && 954 !std::is_volatile<_Tp>::value && !std::is_same<_Tp, bool>::value; 955 } 956 957 _LIBCPP_END_NAMESPACE_EXPERIMENTAL_SIMD 958 _LIBCPP_BEGIN_NAMESPACE_EXPERIMENTAL_SIMD_ABI 959 960 using scalar = __simd_abi<_StorageKind::_Scalar, 1>; 961 962 template <int _Np> 963 using fixed_size = __simd_abi<_StorageKind::_Array, _Np>; 964 965 template <class _Tp> 966 _LIBCPP_INLINE_VAR constexpr size_t max_fixed_size = 32; 967 968 template <class _Tp> 969 using compatible = fixed_size<16 / sizeof(_Tp)>; 970 971 #ifndef _LIBCPP_HAS_NO_VECTOR_EXTENSION 972 template <class _Tp> 973 using native = __simd_abi<_StorageKind::_VecExt, 974 _LIBCPP_NATIVE_SIMD_WIDTH_IN_BYTES / sizeof(_Tp)>; 975 #else 976 template <class _Tp> 977 using native = 978 fixed_size<_Tp, _LIBCPP_NATIVE_SIMD_WIDTH_IN_BYTES / sizeof(_Tp)>; 979 #endif // _LIBCPP_HAS_NO_VECTOR_EXTENSION 980 981 _LIBCPP_END_NAMESPACE_EXPERIMENTAL_SIMD_ABI 982 _LIBCPP_BEGIN_NAMESPACE_EXPERIMENTAL_SIMD 983 984 template <class _Tp, class _Abi = simd_abi::compatible<_Tp>> 985 class simd; 986 template <class _Tp, class _Abi = simd_abi::compatible<_Tp>> 987 class simd_mask; 988 989 struct element_aligned_tag {}; 990 struct vector_aligned_tag {}; 991 template <size_t> 992 struct overaligned_tag {}; 993 _LIBCPP_INLINE_VAR constexpr element_aligned_tag element_aligned{}; 994 _LIBCPP_INLINE_VAR constexpr vector_aligned_tag vector_aligned{}; 995 template <size_t _Np> 996 _LIBCPP_INLINE_VAR constexpr overaligned_tag<_Np> overaligned{}; 997 998 // traits [simd.traits] 999 template <class _Tp> 1000 struct is_abi_tag : std::integral_constant<bool, false> {}; 1001 1002 template <_StorageKind __kind, int _Np> 1003 struct is_abi_tag<__simd_abi<__kind, _Np>> 1004 : std::integral_constant<bool, true> {}; 1005 1006 template <class _Tp> 1007 struct is_simd : std::integral_constant<bool, false> {}; 1008 1009 template <class _Tp, class _Abi> 1010 struct is_simd<simd<_Tp, _Abi>> : std::integral_constant<bool, true> {}; 1011 1012 template <class _Tp> 1013 struct is_simd_mask : std::integral_constant<bool, false> {}; 1014 1015 template <class _Tp, class _Abi> 1016 struct is_simd_mask<simd_mask<_Tp, _Abi>> : std::integral_constant<bool, true> { 1017 }; 1018 1019 template <class _Tp> 1020 struct is_simd_flag_type : std::integral_constant<bool, false> {}; 1021 1022 template <> 1023 struct is_simd_flag_type<element_aligned_tag> 1024 : std::integral_constant<bool, true> {}; 1025 1026 template <> 1027 struct is_simd_flag_type<vector_aligned_tag> 1028 : std::integral_constant<bool, true> {}; 1029 1030 template <size_t _Align> 1031 struct is_simd_flag_type<overaligned_tag<_Align>> 1032 : std::integral_constant<bool, true> {}; 1033 1034 template <class _Tp> 1035 _LIBCPP_INLINE_VAR constexpr bool is_abi_tag_v = is_abi_tag<_Tp>::value; 1036 template <class _Tp> 1037 _LIBCPP_INLINE_VAR constexpr bool is_simd_v = is_simd<_Tp>::value; 1038 template <class _Tp> 1039 _LIBCPP_INLINE_VAR constexpr bool is_simd_mask_v = is_simd_mask<_Tp>::value; 1040 template <class _Tp> 1041 _LIBCPP_INLINE_VAR constexpr bool is_simd_flag_type_v = 1042 is_simd_flag_type<_Tp>::value; 1043 template <class _Tp, size_t _Np> 1044 struct abi_for_size { 1045 using type = simd_abi::fixed_size<_Np>; 1046 }; 1047 template <class _Tp, size_t _Np> 1048 using abi_for_size_t = typename abi_for_size<_Tp, _Np>::type; 1049 1050 template <class _Tp, class _Abi = simd_abi::compatible<_Tp>> 1051 struct simd_size; 1052 1053 template <class _Tp, _StorageKind __kind, int _Np> 1054 struct simd_size<_Tp, __simd_abi<__kind, _Np>> 1055 : std::integral_constant<size_t, _Np> { 1056 static_assert( 1057 std::is_arithmetic<_Tp>::value && 1058 !std::is_same<typename std::remove_const<_Tp>::type, bool>::value, 1059 "Element type should be vectorizable"); 1060 }; 1061 1062 // TODO: implement it. 1063 template <class _Tp, class _Up = typename _Tp::value_type> 1064 struct memory_alignment; 1065 1066 template <class _Tp, class _Abi = simd_abi::compatible<_Tp>> 1067 _LIBCPP_INLINE_VAR constexpr size_t simd_size_v = simd_size<_Tp, _Abi>::value; 1068 1069 template <class _Tp, class _Up = typename _Tp::value_type> 1070 _LIBCPP_INLINE_VAR constexpr size_t memory_alignment_v = 1071 memory_alignment<_Tp, _Up>::value; 1072 1073 // class template simd [simd.class] 1074 template <class _Tp> 1075 using native_simd = simd<_Tp, simd_abi::native<_Tp>>; 1076 template <class _Tp, int _Np> 1077 using fixed_size_simd = simd<_Tp, simd_abi::fixed_size<_Np>>; 1078 1079 // class template simd_mask [simd.mask.class] 1080 template <class _Tp> 1081 using native_simd_mask = simd_mask<_Tp, simd_abi::native<_Tp>>; 1082 1083 template <class _Tp, int _Np> 1084 using fixed_size_simd_mask = simd_mask<_Tp, simd_abi::fixed_size<_Np>>; 1085 1086 // casts [simd.casts] 1087 template <class _Tp> 1088 struct __static_simd_cast_traits { 1089 template <class _Up, class _Abi> 1090 static simd<_Tp, _Abi> __apply(const simd<_Up, _Abi>& __v); 1091 }; 1092 1093 template <class _Tp, class _NewAbi> 1094 struct __static_simd_cast_traits<simd<_Tp, _NewAbi>> { 1095 template <class _Up, class _Abi> 1096 static typename std::enable_if<simd<_Up, _Abi>::size() == 1097 simd<_Tp, _NewAbi>::size(), 1098 simd<_Tp, _NewAbi>>::type 1099 __apply(const simd<_Up, _Abi>& __v); 1100 }; 1101 1102 template <class _Tp> 1103 struct __simd_cast_traits { 1104 template <class _Up, class _Abi> 1105 static typename std::enable_if< 1106 __is_non_narrowing_arithmetic_convertible<_Up, _Tp>(), 1107 simd<_Tp, _Abi>>::type 1108 __apply(const simd<_Up, _Abi>& __v); 1109 }; 1110 1111 template <class _Tp, class _NewAbi> 1112 struct __simd_cast_traits<simd<_Tp, _NewAbi>> { 1113 template <class _Up, class _Abi> 1114 static typename std::enable_if< 1115 __is_non_narrowing_arithmetic_convertible<_Up, _Tp>() && 1116 simd<_Up, _Abi>::size() == simd<_Tp, _NewAbi>::size(), 1117 simd<_Tp, _NewAbi>>::type 1118 __apply(const simd<_Up, _Abi>& __v); 1119 }; 1120 1121 template <class _Tp, class _Up, class _Abi> 1122 auto simd_cast(const simd<_Up, _Abi>& __v) 1123 -> decltype(__simd_cast_traits<_Tp>::__apply(__v)) { 1124 return __simd_cast_traits<_Tp>::__apply(__v); 1125 } 1126 1127 template <class _Tp, class _Up, class _Abi> 1128 auto static_simd_cast(const simd<_Up, _Abi>& __v) 1129 -> decltype(__static_simd_cast_traits<_Tp>::__apply(__v)) { 1130 return __static_simd_cast_traits<_Tp>::__apply(__v); 1131 } 1132 1133 template <class _Tp, class _Abi> 1134 fixed_size_simd<_Tp, simd_size<_Tp, _Abi>::value> 1135 to_fixed_size(const simd<_Tp, _Abi>&) noexcept; 1136 1137 template <class _Tp, class _Abi> 1138 fixed_size_simd_mask<_Tp, simd_size<_Tp, _Abi>::value> 1139 to_fixed_size(const simd_mask<_Tp, _Abi>&) noexcept; 1140 1141 template <class _Tp, size_t _Np> 1142 native_simd<_Tp> to_native(const fixed_size_simd<_Tp, _Np>&) noexcept; 1143 1144 template <class _Tp, size_t _Np> 1145 native_simd_mask<_Tp> to_native(const fixed_size_simd_mask<_Tp, _Np>&) noexcept; 1146 1147 template <class _Tp, size_t _Np> 1148 simd<_Tp> to_compatible(const fixed_size_simd<_Tp, _Np>&) noexcept; 1149 1150 template <class _Tp, size_t _Np> 1151 simd_mask<_Tp> to_compatible(const fixed_size_simd_mask<_Tp, _Np>&) noexcept; 1152 1153 template <size_t... __sizes, class _Tp, class _Abi> 1154 tuple<simd<_Tp, abi_for_size_t<_Tp, __sizes>>...> split(const simd<_Tp, _Abi>&); 1155 1156 template <size_t... __sizes, class _Tp, class _Abi> 1157 tuple<simd_mask<_Tp, abi_for_size_t<_Tp, __sizes>>...> 1158 split(const simd_mask<_Tp, _Abi>&); 1159 1160 template <class _SimdType, class _Abi> 1161 array<_SimdType, simd_size<typename _SimdType::value_type, _Abi>::value / 1162 _SimdType::size()> 1163 split(const simd<typename _SimdType::value_type, _Abi>&); 1164 1165 template <class _SimdType, class _Abi> 1166 array<_SimdType, simd_size<typename _SimdType::value_type, _Abi>::value / 1167 _SimdType::size()> 1168 split(const simd_mask<typename _SimdType::value_type, _Abi>&); 1169 1170 template <class _Tp, class... _Abis> 1171 simd<_Tp, abi_for_size_t<_Tp, __variadic_sum(simd_size<_Tp, _Abis>::value...)>> 1172 concat(const simd<_Tp, _Abis>&...); 1173 1174 template <class _Tp, class... _Abis> 1175 simd_mask<_Tp, 1176 abi_for_size_t<_Tp, __variadic_sum(simd_size<_Tp, _Abis>::value...)>> 1177 concat(const simd_mask<_Tp, _Abis>&...); 1178 1179 // reductions [simd.mask.reductions] 1180 template <class _Tp, class _Abi> 1181 bool all_of(const simd_mask<_Tp, _Abi>&) noexcept; 1182 template <class _Tp, class _Abi> 1183 bool any_of(const simd_mask<_Tp, _Abi>&) noexcept; 1184 template <class _Tp, class _Abi> 1185 bool none_of(const simd_mask<_Tp, _Abi>&) noexcept; 1186 template <class _Tp, class _Abi> 1187 bool some_of(const simd_mask<_Tp, _Abi>&) noexcept; 1188 template <class _Tp, class _Abi> 1189 int popcount(const simd_mask<_Tp, _Abi>&) noexcept; 1190 template <class _Tp, class _Abi> 1191 int find_first_set(const simd_mask<_Tp, _Abi>&); 1192 template <class _Tp, class _Abi> 1193 int find_last_set(const simd_mask<_Tp, _Abi>&); 1194 bool all_of(bool) noexcept; 1195 bool any_of(bool) noexcept; 1196 bool none_of(bool) noexcept; 1197 bool some_of(bool) noexcept; 1198 int popcount(bool) noexcept; 1199 int find_first_set(bool) noexcept; 1200 int find_last_set(bool) noexcept; 1201 1202 // masked assignment [simd.whereexpr] 1203 template <class _MaskType, class _Tp> 1204 class const_where_expression; 1205 template <class _MaskType, class _Tp> 1206 class where_expression; 1207 1208 // masked assignment [simd.mask.where] 1209 template <class _Tp, class _Abi> 1210 where_expression<simd_mask<_Tp, _Abi>, simd<_Tp, _Abi>> 1211 where(const typename simd<_Tp, _Abi>::mask_type&, simd<_Tp, _Abi>&) noexcept; 1212 1213 template <class _Tp, class _Abi> 1214 const_where_expression<simd_mask<_Tp, _Abi>, const simd<_Tp, _Abi>> 1215 where(const typename simd<_Tp, _Abi>::mask_type&, 1216 const simd<_Tp, _Abi>&) noexcept; 1217 1218 template <class _Tp, class _Abi> 1219 where_expression<simd_mask<_Tp, _Abi>, simd_mask<_Tp, _Abi>> 1220 where(const typename __nodeduce<simd_mask<_Tp, _Abi>>::type&, 1221 simd_mask<_Tp, _Abi>&) noexcept; 1222 1223 template <class _Tp, class _Abi> 1224 const_where_expression<simd_mask<_Tp, _Abi>, const simd_mask<_Tp, _Abi>> 1225 where(const typename __nodeduce<simd_mask<_Tp, _Abi>>::type&, 1226 const simd_mask<_Tp, _Abi>&) noexcept; 1227 1228 template <class _Tp> 1229 where_expression<bool, _Tp> where(bool, _Tp&) noexcept; 1230 1231 template <class _Tp> 1232 const_where_expression<bool, const _Tp> where(bool, const _Tp&) noexcept; 1233 1234 // reductions [simd.reductions] 1235 template <class _Tp, class _Abi, class _BinaryOp = std::plus<_Tp>> 1236 _Tp reduce(const simd<_Tp, _Abi>&, _BinaryOp = _BinaryOp()); 1237 1238 template <class _MaskType, class _SimdType, class _BinaryOp> 1239 typename _SimdType::value_type 1240 reduce(const const_where_expression<_MaskType, _SimdType>&, 1241 typename _SimdType::value_type neutral_element, _BinaryOp binary_op); 1242 1243 template <class _MaskType, class _SimdType> 1244 typename _SimdType::value_type 1245 reduce(const const_where_expression<_MaskType, _SimdType>&, 1246 plus<typename _SimdType::value_type> binary_op = {}); 1247 1248 template <class _MaskType, class _SimdType> 1249 typename _SimdType::value_type 1250 reduce(const const_where_expression<_MaskType, _SimdType>&, 1251 multiplies<typename _SimdType::value_type> binary_op); 1252 1253 template <class _MaskType, class _SimdType> 1254 typename _SimdType::value_type 1255 reduce(const const_where_expression<_MaskType, _SimdType>&, 1256 bit_and<typename _SimdType::value_type> binary_op); 1257 1258 template <class _MaskType, class _SimdType> 1259 typename _SimdType::value_type 1260 reduce(const const_where_expression<_MaskType, _SimdType>&, 1261 bit_or<typename _SimdType::value_type> binary_op); 1262 1263 template <class _MaskType, class _SimdType> 1264 typename _SimdType::value_type 1265 reduce(const const_where_expression<_MaskType, _SimdType>&, 1266 bit_xor<typename _SimdType::value_type> binary_op); 1267 1268 template <class _Tp, class _Abi> 1269 _Tp hmin(const simd<_Tp, _Abi>&); 1270 template <class _MaskType, class _SimdType> 1271 typename _SimdType::value_type 1272 hmin(const const_where_expression<_MaskType, _SimdType>&); 1273 template <class _Tp, class _Abi> 1274 _Tp hmax(const simd<_Tp, _Abi>&); 1275 template <class _MaskType, class _SimdType> 1276 typename _SimdType::value_type 1277 hmax(const const_where_expression<_MaskType, _SimdType>&); 1278 1279 // algorithms [simd.alg] 1280 template <class _Tp, class _Abi> 1281 simd<_Tp, _Abi> min(const simd<_Tp, _Abi>&, const simd<_Tp, _Abi>&) noexcept; 1282 1283 template <class _Tp, class _Abi> 1284 simd<_Tp, _Abi> max(const simd<_Tp, _Abi>&, const simd<_Tp, _Abi>&) noexcept; 1285 1286 template <class _Tp, class _Abi> 1287 std::pair<simd<_Tp, _Abi>, simd<_Tp, _Abi>> 1288 minmax(const simd<_Tp, _Abi>&, const simd<_Tp, _Abi>&) noexcept; 1289 1290 template <class _Tp, class _Abi> 1291 simd<_Tp, _Abi> clamp(const simd<_Tp, _Abi>&, const simd<_Tp, _Abi>&, 1292 const simd<_Tp, _Abi>&); 1293 1294 // [simd.whereexpr] 1295 // TODO implement where expressions. 1296 template <class _MaskType, class _Tp> 1297 class const_where_expression { 1298 public: 1299 const_where_expression(const const_where_expression&) = delete; 1300 const_where_expression& operator=(const const_where_expression&) = delete; 1301 typename remove_const<_Tp>::type operator-() const&&; 1302 template <class _Up, class _Flags> 1303 void copy_to(_Up*, _Flags) const&&; 1304 }; 1305 1306 template <class _MaskType, class _Tp> 1307 class where_expression : public const_where_expression<_MaskType, _Tp> { 1308 public: 1309 where_expression(const where_expression&) = delete; 1310 where_expression& operator=(const where_expression&) = delete; 1311 template <class _Up> 1312 void operator=(_Up&&); 1313 template <class _Up> 1314 void operator+=(_Up&&); 1315 template <class _Up> 1316 void operator-=(_Up&&); 1317 template <class _Up> 1318 void operator*=(_Up&&); 1319 template <class _Up> 1320 void operator/=(_Up&&); 1321 template <class _Up> 1322 void operator%=(_Up&&); 1323 template <class _Up> 1324 void operator&=(_Up&&); 1325 template <class _Up> 1326 void operator|=(_Up&&); 1327 template <class _Up> 1328 void operator^=(_Up&&); 1329 template <class _Up> 1330 void operator<<=(_Up&&); 1331 template <class _Up> 1332 void operator>>=(_Up&&); 1333 void operator++(); 1334 void operator++(int); 1335 void operator--(); 1336 void operator--(int); 1337 template <class _Up, class _Flags> 1338 void copy_from(const _Up*, _Flags); 1339 }; 1340 1341 // [simd.class] 1342 // TODO: implement simd 1343 template <class _Tp, class _Abi> 1344 class simd { 1345 public: 1346 using value_type = _Tp; 1347 using reference = __simd_reference<_Tp, _Tp, _Abi>; 1348 using mask_type = simd_mask<_Tp, _Abi>; 1349 using abi_type = _Abi; 1350 1351 simd() = default; 1352 simd(const simd&) = default; 1353 simd& operator=(const simd&) = default; 1354 1355 static constexpr size_t size() noexcept { 1356 return simd_size<_Tp, _Abi>::value; 1357 } 1358 1359 private: 1360 __simd_storage<_Tp, _Abi> __s_; 1361 1362 template <class _Up> 1363 static constexpr bool __can_broadcast() { 1364 return (std::is_arithmetic<_Up>::value && 1365 __is_non_narrowing_arithmetic_convertible<_Up, _Tp>()) || 1366 (!std::is_arithmetic<_Up>::value && 1367 std::is_convertible<_Up, _Tp>::value) || 1368 std::is_same<typename std::remove_const<_Up>::type, int>::value || 1369 (std::is_same<typename std::remove_const<_Up>::type, 1370 unsigned int>::value && 1371 std::is_unsigned<_Tp>::value); 1372 } 1373 1374 template <class _Generator, size_t... __indicies> 1375 static constexpr decltype( 1376 std::forward_as_tuple(std::declval<_Generator>()( 1377 std::integral_constant<size_t, __indicies>())...), 1378 bool()) 1379 __can_generate(std::index_sequence<__indicies...>) { 1380 return !__variadic_sum<bool>( 1381 !__can_broadcast<decltype(std::declval<_Generator>()( 1382 std::integral_constant<size_t, __indicies>()))>()...); 1383 } 1384 1385 template <class _Generator> 1386 static bool __can_generate(...) { 1387 return false; 1388 } 1389 1390 template <class _Generator, size_t... __indicies> 1391 void __generator_init(_Generator&& __g, std::index_sequence<__indicies...>) { 1392 int __not_used[]{((*this)[__indicies] = 1393 __g(std::integral_constant<size_t, __indicies>()), 1394 0)...}; 1395 (void)__not_used; 1396 } 1397 1398 public: 1399 // implicit type conversion constructor 1400 template <class _Up, 1401 class = typename std::enable_if< 1402 std::is_same<_Abi, simd_abi::fixed_size<size()>>::value && 1403 __is_non_narrowing_arithmetic_convertible<_Up, _Tp>()>::type> 1404 simd(const simd<_Up, simd_abi::fixed_size<size()>>& __v) { 1405 for (size_t __i = 0; __i < size(); __i++) { 1406 (*this)[__i] = static_cast<_Tp>(__v[__i]); 1407 } 1408 } 1409 1410 // implicit broadcast constructor 1411 template <class _Up, 1412 class = typename std::enable_if<__can_broadcast<_Up>()>::type> 1413 simd(_Up&& __rv) { 1414 auto __v = static_cast<_Tp>(__rv); 1415 for (size_t __i = 0; __i < size(); __i++) { 1416 (*this)[__i] = __v; 1417 } 1418 } 1419 1420 // generator constructor 1421 template <class _Generator, 1422 int = typename std::enable_if< 1423 __can_generate<_Generator>(std::make_index_sequence<size()>()), 1424 int>::type()> 1425 explicit simd(_Generator&& __g) { 1426 __generator_init(std::forward<_Generator>(__g), 1427 std::make_index_sequence<size()>()); 1428 } 1429 1430 // load constructor 1431 template < 1432 class _Up, class _Flags, 1433 class = typename std::enable_if<__vectorizable<_Up>()>::type, 1434 class = typename std::enable_if<is_simd_flag_type<_Flags>::value>::type> 1435 simd(const _Up* __buffer, _Flags) { 1436 // TODO: optimize for overaligned flags 1437 for (size_t __i = 0; __i < size(); __i++) { 1438 (*this)[__i] = static_cast<_Tp>(__buffer[__i]); 1439 } 1440 } 1441 1442 // loads [simd.load] 1443 template <class _Up, class _Flags> 1444 typename std::enable_if<__vectorizable<_Up>() && 1445 is_simd_flag_type<_Flags>::value>::type 1446 copy_from(const _Up* __buffer, _Flags) { 1447 *this = simd(__buffer, _Flags()); 1448 } 1449 1450 // stores [simd.store] 1451 template <class _Up, class _Flags> 1452 typename std::enable_if<__vectorizable<_Up>() && 1453 is_simd_flag_type<_Flags>::value>::type 1454 copy_to(_Up* __buffer, _Flags) const { 1455 // TODO: optimize for overaligned flags 1456 for (size_t __i = 0; __i < size(); __i++) { 1457 __buffer[__i] = static_cast<_Up>((*this)[__i]); 1458 } 1459 } 1460 1461 // scalar access [simd.subscr] 1462 reference operator[](size_t __i) { return reference(&__s_, __i); } 1463 1464 value_type operator[](size_t __i) const { return __s_.__get(__i); } 1465 1466 // unary operators [simd.unary] 1467 simd& operator++(); 1468 simd operator++(int); 1469 simd& operator--(); 1470 simd operator--(int); 1471 mask_type operator!() const; 1472 simd operator~() const; 1473 simd operator+() const; 1474 simd operator-() const; 1475 1476 // binary operators [simd.binary] 1477 friend simd operator+(const simd&, const simd&); 1478 friend simd operator-(const simd&, const simd&); 1479 friend simd operator*(const simd&, const simd&); 1480 friend simd operator/(const simd&, const simd&); 1481 friend simd operator%(const simd&, const simd&); 1482 friend simd operator&(const simd&, const simd&); 1483 friend simd operator|(const simd&, const simd&); 1484 friend simd operator^(const simd&, const simd&); 1485 friend simd operator<<(const simd&, const simd&); 1486 friend simd operator>>(const simd&, const simd&); 1487 friend simd operator<<(const simd&, int); 1488 friend simd operator>>(const simd&, int); 1489 1490 // compound assignment [simd.cassign] 1491 friend simd& operator+=(simd&, const simd&); 1492 friend simd& operator-=(simd&, const simd&); 1493 friend simd& operator*=(simd&, const simd&); 1494 friend simd& operator/=(simd&, const simd&); 1495 friend simd& operator%=(simd&, const simd&); 1496 1497 friend simd& operator&=(simd&, const simd&); 1498 friend simd& operator|=(simd&, const simd&); 1499 friend simd& operator^=(simd&, const simd&); 1500 friend simd& operator<<=(simd&, const simd&); 1501 friend simd& operator>>=(simd&, const simd&); 1502 friend simd& operator<<=(simd&, int); 1503 friend simd& operator>>=(simd&, int); 1504 1505 // compares [simd.comparison] 1506 friend mask_type operator==(const simd&, const simd&); 1507 friend mask_type operator!=(const simd&, const simd&); 1508 friend mask_type operator>=(const simd&, const simd&); 1509 friend mask_type operator<=(const simd&, const simd&); 1510 friend mask_type operator>(const simd&, const simd&); 1511 friend mask_type operator<(const simd&, const simd&); 1512 }; 1513 1514 // [simd.mask.class] 1515 template <class _Tp, class _Abi> 1516 // TODO: implement simd_mask 1517 class simd_mask { 1518 public: 1519 using value_type = bool; 1520 // TODO: this is strawman implementation. Turn it into a proxy type. 1521 using reference = bool&; 1522 using simd_type = simd<_Tp, _Abi>; 1523 using abi_type = _Abi; 1524 static constexpr size_t size() noexcept; 1525 simd_mask() = default; 1526 1527 // broadcast constructor 1528 explicit simd_mask(value_type) noexcept; 1529 1530 // implicit type conversion constructor 1531 template <class _Up> 1532 simd_mask(const simd_mask<_Up, simd_abi::fixed_size<size()>>&) noexcept; 1533 1534 // load constructor 1535 template <class _Flags> 1536 simd_mask(const value_type*, _Flags); 1537 1538 // loads [simd.mask.copy] 1539 template <class _Flags> 1540 void copy_from(const value_type*, _Flags); 1541 template <class _Flags> 1542 void copy_to(value_type*, _Flags) const; 1543 1544 // scalar access [simd.mask.subscr] 1545 reference operator[](size_t); 1546 value_type operator[](size_t) const; 1547 1548 // unary operators [simd.mask.unary] 1549 simd_mask operator!() const noexcept; 1550 1551 // simd_mask binary operators [simd.mask.binary] 1552 friend simd_mask operator&&(const simd_mask&, const simd_mask&) noexcept; 1553 friend simd_mask operator||(const simd_mask&, const simd_mask&) noexcept; 1554 friend simd_mask operator&(const simd_mask&, const simd_mask&)noexcept; 1555 friend simd_mask operator|(const simd_mask&, const simd_mask&) noexcept; 1556 friend simd_mask operator^(const simd_mask&, const simd_mask&) noexcept; 1557 1558 // simd_mask compound assignment [simd.mask.cassign] 1559 friend simd_mask& operator&=(simd_mask&, const simd_mask&) noexcept; 1560 friend simd_mask& operator|=(simd_mask&, const simd_mask&) noexcept; 1561 friend simd_mask& operator^=(simd_mask&, const simd_mask&) noexcept; 1562 1563 // simd_mask compares [simd.mask.comparison] 1564 friend simd_mask operator==(const simd_mask&, const simd_mask&) noexcept; 1565 friend simd_mask operator!=(const simd_mask&, const simd_mask&) noexcept; 1566 }; 1567 1568 #endif // _LIBCPP_STD_VER >= 17 1569 1570 _LIBCPP_END_NAMESPACE_EXPERIMENTAL_SIMD 1571 1572 _LIBCPP_POP_MACROS 1573 1574 #endif /* _LIBCPP_EXPERIMENTAL_SIMD */ 1575
Indexes created Sat Feb 28 05:31:39 UTC 2026