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