| 1 | //! SIMD compiler intrinsics. |
| 2 | //! |
| 3 | //! In this module, a "vector" is any `repr(simd)` type. |
| 4 | |
| 5 | /// Inserts an element into a vector, returning the updated vector. |
| 6 | /// |
| 7 | /// `T` must be a vector with element type `U`, and `idx` must be `const`. |
| 8 | /// |
| 9 | /// # Safety |
| 10 | /// |
| 11 | /// `idx` must be in-bounds of the vector. |
| 12 | #[rustc_intrinsic ] |
| 13 | #[rustc_nounwind ] |
| 14 | pub const unsafe fn simd_insert<T, U>(x: T, idx: u32, val: U) -> T; |
| 15 | |
| 16 | /// Extracts an element from a vector. |
| 17 | /// |
| 18 | /// `T` must be a vector with element type `U`, and `idx` must be `const`. |
| 19 | /// |
| 20 | /// # Safety |
| 21 | /// |
| 22 | /// `idx` must be const and in-bounds of the vector. |
| 23 | #[rustc_intrinsic ] |
| 24 | #[rustc_nounwind ] |
| 25 | pub const unsafe fn simd_extract<T, U>(x: T, idx: u32) -> U; |
| 26 | |
| 27 | /// Inserts an element into a vector, returning the updated vector. |
| 28 | /// |
| 29 | /// `T` must be a vector with element type `U`. |
| 30 | /// |
| 31 | /// If the index is `const`, [`simd_insert`] may emit better assembly. |
| 32 | /// |
| 33 | /// # Safety |
| 34 | /// |
| 35 | /// `idx` must be in-bounds of the vector. |
| 36 | #[rustc_nounwind ] |
| 37 | #[rustc_intrinsic ] |
| 38 | pub unsafe fn simd_insert_dyn<T, U>(mut x: T, idx: u32, val: U) -> T { |
| 39 | // SAFETY: `idx` must be in-bounds |
| 40 | unsafe { (&raw mut x).cast::<U>().add(count:idx as usize).write(val) } |
| 41 | x |
| 42 | } |
| 43 | |
| 44 | /// Extracts an element from a vector. |
| 45 | /// |
| 46 | /// `T` must be a vector with element type `U`. |
| 47 | /// |
| 48 | /// If the index is `const`, [`simd_extract`] may emit better assembly. |
| 49 | /// |
| 50 | /// # Safety |
| 51 | /// |
| 52 | /// `idx` must be in-bounds of the vector. |
| 53 | #[rustc_nounwind ] |
| 54 | #[rustc_intrinsic ] |
| 55 | pub unsafe fn simd_extract_dyn<T, U>(x: T, idx: u32) -> U { |
| 56 | // SAFETY: `idx` must be in-bounds |
| 57 | unsafe { (&raw const x).cast::<U>().add(count:idx as usize).read() } |
| 58 | } |
| 59 | |
| 60 | /// Adds two simd vectors elementwise. |
| 61 | /// |
| 62 | /// `T` must be a vector of integers or floats. |
| 63 | #[rustc_intrinsic ] |
| 64 | #[rustc_nounwind ] |
| 65 | pub unsafe fn simd_add<T>(x: T, y: T) -> T; |
| 66 | |
| 67 | /// Subtracts `rhs` from `lhs` elementwise. |
| 68 | /// |
| 69 | /// `T` must be a vector of integers or floats. |
| 70 | #[rustc_intrinsic ] |
| 71 | #[rustc_nounwind ] |
| 72 | pub unsafe fn simd_sub<T>(lhs: T, rhs: T) -> T; |
| 73 | |
| 74 | /// Multiplies two simd vectors elementwise. |
| 75 | /// |
| 76 | /// `T` must be a vector of integers or floats. |
| 77 | #[rustc_intrinsic ] |
| 78 | #[rustc_nounwind ] |
| 79 | pub unsafe fn simd_mul<T>(x: T, y: T) -> T; |
| 80 | |
| 81 | /// Divides `lhs` by `rhs` elementwise. |
| 82 | /// |
| 83 | /// `T` must be a vector of integers or floats. |
| 84 | /// |
| 85 | /// # Safety |
| 86 | /// For integers, `rhs` must not contain any zero elements. |
| 87 | /// Additionally for signed integers, `<int>::MIN / -1` is undefined behavior. |
| 88 | #[rustc_intrinsic ] |
| 89 | #[rustc_nounwind ] |
| 90 | pub unsafe fn simd_div<T>(lhs: T, rhs: T) -> T; |
| 91 | |
| 92 | /// Returns remainder of two vectors elementwise. |
| 93 | /// |
| 94 | /// `T` must be a vector of integers or floats. |
| 95 | /// |
| 96 | /// # Safety |
| 97 | /// For integers, `rhs` must not contain any zero elements. |
| 98 | /// Additionally for signed integers, `<int>::MIN / -1` is undefined behavior. |
| 99 | #[rustc_intrinsic ] |
| 100 | #[rustc_nounwind ] |
| 101 | pub unsafe fn simd_rem<T>(lhs: T, rhs: T) -> T; |
| 102 | |
| 103 | /// Shifts vector left elementwise, with UB on overflow. |
| 104 | /// |
| 105 | /// Shifts `lhs` left by `rhs`, shifting in sign bits for signed types. |
| 106 | /// |
| 107 | /// `T` must be a vector of integers. |
| 108 | /// |
| 109 | /// # Safety |
| 110 | /// |
| 111 | /// Each element of `rhs` must be less than `<int>::BITS`. |
| 112 | #[rustc_intrinsic ] |
| 113 | #[rustc_nounwind ] |
| 114 | pub unsafe fn simd_shl<T>(lhs: T, rhs: T) -> T; |
| 115 | |
| 116 | /// Shifts vector right elementwise, with UB on overflow. |
| 117 | /// |
| 118 | /// `T` must be a vector of integers. |
| 119 | /// |
| 120 | /// Shifts `lhs` right by `rhs`, shifting in sign bits for signed types. |
| 121 | /// |
| 122 | /// # Safety |
| 123 | /// |
| 124 | /// Each element of `rhs` must be less than `<int>::BITS`. |
| 125 | #[rustc_intrinsic ] |
| 126 | #[rustc_nounwind ] |
| 127 | pub unsafe fn simd_shr<T>(lhs: T, rhs: T) -> T; |
| 128 | |
| 129 | /// "Ands" vectors elementwise. |
| 130 | /// |
| 131 | /// `T` must be a vector of integers. |
| 132 | #[rustc_intrinsic ] |
| 133 | #[rustc_nounwind ] |
| 134 | pub unsafe fn simd_and<T>(x: T, y: T) -> T; |
| 135 | |
| 136 | /// "Ors" vectors elementwise. |
| 137 | /// |
| 138 | /// `T` must be a vector of integers. |
| 139 | #[rustc_intrinsic ] |
| 140 | #[rustc_nounwind ] |
| 141 | pub unsafe fn simd_or<T>(x: T, y: T) -> T; |
| 142 | |
| 143 | /// "Exclusive ors" vectors elementwise. |
| 144 | /// |
| 145 | /// `T` must be a vector of integers. |
| 146 | #[rustc_intrinsic ] |
| 147 | #[rustc_nounwind ] |
| 148 | pub unsafe fn simd_xor<T>(x: T, y: T) -> T; |
| 149 | |
| 150 | /// Numerically casts a vector, elementwise. |
| 151 | /// |
| 152 | /// `T` and `U` must be vectors of integers or floats, and must have the same length. |
| 153 | /// |
| 154 | /// When casting floats to integers, the result is truncated. Out-of-bounds result lead to UB. |
| 155 | /// When casting integers to floats, the result is rounded. |
| 156 | /// Otherwise, truncates or extends the value, maintaining the sign for signed integers. |
| 157 | /// |
| 158 | /// # Safety |
| 159 | /// Casting from integer types is always safe. |
| 160 | /// Casting between two float types is also always safe. |
| 161 | /// |
| 162 | /// Casting floats to integers truncates, following the same rules as `to_int_unchecked`. |
| 163 | /// Specifically, each element must: |
| 164 | /// * Not be `NaN` |
| 165 | /// * Not be infinite |
| 166 | /// * Be representable in the return type, after truncating off its fractional part |
| 167 | #[rustc_intrinsic ] |
| 168 | #[rustc_nounwind ] |
| 169 | pub unsafe fn simd_cast<T, U>(x: T) -> U; |
| 170 | |
| 171 | /// Numerically casts a vector, elementwise. |
| 172 | /// |
| 173 | /// `T` and `U` be a vectors of integers or floats, and must have the same length. |
| 174 | /// |
| 175 | /// Like `simd_cast`, but saturates float-to-integer conversions (NaN becomes 0). |
| 176 | /// This matches regular `as` and is always safe. |
| 177 | /// |
| 178 | /// When casting floats to integers, the result is truncated. |
| 179 | /// When casting integers to floats, the result is rounded. |
| 180 | /// Otherwise, truncates or extends the value, maintaining the sign for signed integers. |
| 181 | #[rustc_intrinsic ] |
| 182 | #[rustc_nounwind ] |
| 183 | pub unsafe fn simd_as<T, U>(x: T) -> U; |
| 184 | |
| 185 | /// Negates a vector elementwise. |
| 186 | /// |
| 187 | /// `T` must be a vector of integers or floats. |
| 188 | /// |
| 189 | /// Rust panics for `-<int>::Min` due to overflow, but it is not UB with this intrinsic. |
| 190 | #[rustc_intrinsic ] |
| 191 | #[rustc_nounwind ] |
| 192 | pub unsafe fn simd_neg<T>(x: T) -> T; |
| 193 | |
| 194 | /// Returns absolute value of a vector, elementwise. |
| 195 | /// |
| 196 | /// `T` must be a vector of floating-point primitive types. |
| 197 | #[rustc_intrinsic ] |
| 198 | #[rustc_nounwind ] |
| 199 | pub unsafe fn simd_fabs<T>(x: T) -> T; |
| 200 | |
| 201 | /// Returns the minimum of two vectors, elementwise. |
| 202 | /// |
| 203 | /// `T` must be a vector of floating-point primitive types. |
| 204 | /// |
| 205 | /// Follows IEEE-754 `minNum` semantics. |
| 206 | #[rustc_intrinsic ] |
| 207 | #[rustc_nounwind ] |
| 208 | pub unsafe fn simd_fmin<T>(x: T, y: T) -> T; |
| 209 | |
| 210 | /// Returns the maximum of two vectors, elementwise. |
| 211 | /// |
| 212 | /// `T` must be a vector of floating-point primitive types. |
| 213 | /// |
| 214 | /// Follows IEEE-754 `maxNum` semantics. |
| 215 | #[rustc_intrinsic ] |
| 216 | #[rustc_nounwind ] |
| 217 | pub unsafe fn simd_fmax<T>(x: T, y: T) -> T; |
| 218 | |
| 219 | /// Tests elementwise equality of two vectors. |
| 220 | /// |
| 221 | /// `T` must be a vector of integers or floats. |
| 222 | /// |
| 223 | /// `U` must be a vector of integers with the same number of elements and element size as `T`. |
| 224 | /// |
| 225 | /// Returns `0` for false and `!0` for true. |
| 226 | #[rustc_intrinsic ] |
| 227 | #[rustc_nounwind ] |
| 228 | pub unsafe fn simd_eq<T, U>(x: T, y: T) -> U; |
| 229 | |
| 230 | /// Tests elementwise inequality equality of two vectors. |
| 231 | /// |
| 232 | /// `T` must be a vector of integers or floats. |
| 233 | /// |
| 234 | /// `U` must be a vector of integers with the same number of elements and element size as `T`. |
| 235 | /// |
| 236 | /// Returns `0` for false and `!0` for true. |
| 237 | #[rustc_intrinsic ] |
| 238 | #[rustc_nounwind ] |
| 239 | pub unsafe fn simd_ne<T, U>(x: T, y: T) -> U; |
| 240 | |
| 241 | /// Tests if `x` is less than `y`, elementwise. |
| 242 | /// |
| 243 | /// `T` must be a vector of integers or floats. |
| 244 | /// |
| 245 | /// `U` must be a vector of integers with the same number of elements and element size as `T`. |
| 246 | /// |
| 247 | /// Returns `0` for false and `!0` for true. |
| 248 | #[rustc_intrinsic ] |
| 249 | #[rustc_nounwind ] |
| 250 | pub unsafe fn simd_lt<T, U>(x: T, y: T) -> U; |
| 251 | |
| 252 | /// Tests if `x` is less than or equal to `y`, elementwise. |
| 253 | /// |
| 254 | /// `T` must be a vector of integers or floats. |
| 255 | /// |
| 256 | /// `U` must be a vector of integers with the same number of elements and element size as `T`. |
| 257 | /// |
| 258 | /// Returns `0` for false and `!0` for true. |
| 259 | #[rustc_intrinsic ] |
| 260 | #[rustc_nounwind ] |
| 261 | pub unsafe fn simd_le<T, U>(x: T, y: T) -> U; |
| 262 | |
| 263 | /// Tests if `x` is greater than `y`, elementwise. |
| 264 | /// |
| 265 | /// `T` must be a vector of integers or floats. |
| 266 | /// |
| 267 | /// `U` must be a vector of integers with the same number of elements and element size as `T`. |
| 268 | /// |
| 269 | /// Returns `0` for false and `!0` for true. |
| 270 | #[rustc_intrinsic ] |
| 271 | #[rustc_nounwind ] |
| 272 | pub unsafe fn simd_gt<T, U>(x: T, y: T) -> U; |
| 273 | |
| 274 | /// Tests if `x` is greater than or equal to `y`, elementwise. |
| 275 | /// |
| 276 | /// `T` must be a vector of integers or floats. |
| 277 | /// |
| 278 | /// `U` must be a vector of integers with the same number of elements and element size as `T`. |
| 279 | /// |
| 280 | /// Returns `0` for false and `!0` for true. |
| 281 | #[rustc_intrinsic ] |
| 282 | #[rustc_nounwind ] |
| 283 | pub unsafe fn simd_ge<T, U>(x: T, y: T) -> U; |
| 284 | |
| 285 | /// Shuffles two vectors by const indices. |
| 286 | /// |
| 287 | /// `T` must be a vector. |
| 288 | /// |
| 289 | /// `U` must be a **const** vector of `u32`s. This means it must either refer to a named |
| 290 | /// const or be given as an inline const expression (`const { ... }`). |
| 291 | /// |
| 292 | /// `V` must be a vector with the same element type as `T` and the same length as `U`. |
| 293 | /// |
| 294 | /// Returns a new vector such that element `i` is selected from `xy[idx[i]]`, where `xy` |
| 295 | /// is the concatenation of `x` and `y`. It is a compile-time error if `idx[i]` is out-of-bounds |
| 296 | /// of `xy`. |
| 297 | #[rustc_intrinsic ] |
| 298 | #[rustc_nounwind ] |
| 299 | pub unsafe fn simd_shuffle<T, U, V>(x: T, y: T, idx: U) -> V; |
| 300 | |
| 301 | /// Reads a vector of pointers. |
| 302 | /// |
| 303 | /// `T` must be a vector. |
| 304 | /// |
| 305 | /// `U` must be a vector of pointers to the element type of `T`, with the same length as `T`. |
| 306 | /// |
| 307 | /// `V` must be a vector of integers with the same length as `T` (but any element size). |
| 308 | /// |
| 309 | /// For each pointer in `ptr`, if the corresponding value in `mask` is `!0`, read the pointer. |
| 310 | /// Otherwise if the corresponding value in `mask` is `0`, return the corresponding value from |
| 311 | /// `val`. |
| 312 | /// |
| 313 | /// # Safety |
| 314 | /// Unmasked values in `T` must be readable as if by `<ptr>::read` (e.g. aligned to the element |
| 315 | /// type). |
| 316 | /// |
| 317 | /// `mask` must only contain `0` or `!0` values. |
| 318 | #[rustc_intrinsic ] |
| 319 | #[rustc_nounwind ] |
| 320 | pub unsafe fn simd_gather<T, U, V>(val: T, ptr: U, mask: V) -> T; |
| 321 | |
| 322 | /// Writes to a vector of pointers. |
| 323 | /// |
| 324 | /// `T` must be a vector. |
| 325 | /// |
| 326 | /// `U` must be a vector of pointers to the element type of `T`, with the same length as `T`. |
| 327 | /// |
| 328 | /// `V` must be a vector of integers with the same length as `T` (but any element size). |
| 329 | /// |
| 330 | /// For each pointer in `ptr`, if the corresponding value in `mask` is `!0`, write the |
| 331 | /// corresponding value in `val` to the pointer. |
| 332 | /// Otherwise if the corresponding value in `mask` is `0`, do nothing. |
| 333 | /// |
| 334 | /// The stores happen in left-to-right order. |
| 335 | /// (This is relevant in case two of the stores overlap.) |
| 336 | /// |
| 337 | /// # Safety |
| 338 | /// Unmasked values in `T` must be writeable as if by `<ptr>::write` (e.g. aligned to the element |
| 339 | /// type). |
| 340 | /// |
| 341 | /// `mask` must only contain `0` or `!0` values. |
| 342 | #[rustc_intrinsic ] |
| 343 | #[rustc_nounwind ] |
| 344 | pub unsafe fn simd_scatter<T, U, V>(val: T, ptr: U, mask: V); |
| 345 | |
| 346 | /// Reads a vector of pointers. |
| 347 | /// |
| 348 | /// `T` must be a vector. |
| 349 | /// |
| 350 | /// `U` must be a pointer to the element type of `T` |
| 351 | /// |
| 352 | /// `V` must be a vector of integers with the same length as `T` (but any element size). |
| 353 | /// |
| 354 | /// For each element, if the corresponding value in `mask` is `!0`, read the corresponding |
| 355 | /// pointer offset from `ptr`. |
| 356 | /// The first element is loaded from `ptr`, the second from `ptr.wrapping_offset(1)` and so on. |
| 357 | /// Otherwise if the corresponding value in `mask` is `0`, return the corresponding value from |
| 358 | /// `val`. |
| 359 | /// |
| 360 | /// # Safety |
| 361 | /// Unmasked values in `T` must be readable as if by `<ptr>::read` (e.g. aligned to the element |
| 362 | /// type). |
| 363 | /// |
| 364 | /// `mask` must only contain `0` or `!0` values. |
| 365 | #[rustc_intrinsic ] |
| 366 | #[rustc_nounwind ] |
| 367 | pub unsafe fn simd_masked_load<V, U, T>(mask: V, ptr: U, val: T) -> T; |
| 368 | |
| 369 | /// Writes to a vector of pointers. |
| 370 | /// |
| 371 | /// `T` must be a vector. |
| 372 | /// |
| 373 | /// `U` must be a pointer to the element type of `T` |
| 374 | /// |
| 375 | /// `V` must be a vector of integers with the same length as `T` (but any element size). |
| 376 | /// |
| 377 | /// For each element, if the corresponding value in `mask` is `!0`, write the corresponding |
| 378 | /// value in `val` to the pointer offset from `ptr`. |
| 379 | /// The first element is written to `ptr`, the second to `ptr.wrapping_offset(1)` and so on. |
| 380 | /// Otherwise if the corresponding value in `mask` is `0`, do nothing. |
| 381 | /// |
| 382 | /// # Safety |
| 383 | /// Unmasked values in `T` must be writeable as if by `<ptr>::write` (e.g. aligned to the element |
| 384 | /// type). |
| 385 | /// |
| 386 | /// `mask` must only contain `0` or `!0` values. |
| 387 | #[rustc_intrinsic ] |
| 388 | #[rustc_nounwind ] |
| 389 | pub unsafe fn simd_masked_store<V, U, T>(mask: V, ptr: U, val: T); |
| 390 | |
| 391 | /// Adds two simd vectors elementwise, with saturation. |
| 392 | /// |
| 393 | /// `T` must be a vector of integer primitive types. |
| 394 | #[rustc_intrinsic ] |
| 395 | #[rustc_nounwind ] |
| 396 | pub unsafe fn simd_saturating_add<T>(x: T, y: T) -> T; |
| 397 | |
| 398 | /// Subtracts two simd vectors elementwise, with saturation. |
| 399 | /// |
| 400 | /// `T` must be a vector of integer primitive types. |
| 401 | /// |
| 402 | /// Subtract `rhs` from `lhs`. |
| 403 | #[rustc_intrinsic ] |
| 404 | #[rustc_nounwind ] |
| 405 | pub unsafe fn simd_saturating_sub<T>(lhs: T, rhs: T) -> T; |
| 406 | |
| 407 | /// Adds elements within a vector from left to right. |
| 408 | /// |
| 409 | /// `T` must be a vector of integers or floats. |
| 410 | /// |
| 411 | /// `U` must be the element type of `T`. |
| 412 | /// |
| 413 | /// Starting with the value `y`, add the elements of `x` and accumulate. |
| 414 | #[rustc_intrinsic ] |
| 415 | #[rustc_nounwind ] |
| 416 | pub unsafe fn simd_reduce_add_ordered<T, U>(x: T, y: U) -> U; |
| 417 | |
| 418 | /// Adds elements within a vector in arbitrary order. May also be re-associated with |
| 419 | /// unordered additions on the inputs/outputs. |
| 420 | /// |
| 421 | /// `T` must be a vector of integers or floats. |
| 422 | /// |
| 423 | /// `U` must be the element type of `T`. |
| 424 | #[rustc_intrinsic ] |
| 425 | #[rustc_nounwind ] |
| 426 | pub unsafe fn simd_reduce_add_unordered<T, U>(x: T) -> U; |
| 427 | |
| 428 | /// Multiplies elements within a vector from left to right. |
| 429 | /// |
| 430 | /// `T` must be a vector of integers or floats. |
| 431 | /// |
| 432 | /// `U` must be the element type of `T`. |
| 433 | /// |
| 434 | /// Starting with the value `y`, multiply the elements of `x` and accumulate. |
| 435 | #[rustc_intrinsic ] |
| 436 | #[rustc_nounwind ] |
| 437 | pub unsafe fn simd_reduce_mul_ordered<T, U>(x: T, y: U) -> U; |
| 438 | |
| 439 | /// Multiplies elements within a vector in arbitrary order. May also be re-associated with |
| 440 | /// unordered additions on the inputs/outputs. |
| 441 | /// |
| 442 | /// `T` must be a vector of integers or floats. |
| 443 | /// |
| 444 | /// `U` must be the element type of `T`. |
| 445 | #[rustc_intrinsic ] |
| 446 | #[rustc_nounwind ] |
| 447 | pub unsafe fn simd_reduce_mul_unordered<T, U>(x: T) -> U; |
| 448 | |
| 449 | /// Checks if all mask values are true. |
| 450 | /// |
| 451 | /// `T` must be a vector of integer primitive types. |
| 452 | /// |
| 453 | /// # Safety |
| 454 | /// `x` must contain only `0` or `!0`. |
| 455 | #[rustc_intrinsic ] |
| 456 | #[rustc_nounwind ] |
| 457 | pub unsafe fn simd_reduce_all<T>(x: T) -> bool; |
| 458 | |
| 459 | /// Checks if any mask value is true. |
| 460 | /// |
| 461 | /// `T` must be a vector of integer primitive types. |
| 462 | /// |
| 463 | /// # Safety |
| 464 | /// `x` must contain only `0` or `!0`. |
| 465 | #[rustc_intrinsic ] |
| 466 | #[rustc_nounwind ] |
| 467 | pub unsafe fn simd_reduce_any<T>(x: T) -> bool; |
| 468 | |
| 469 | /// Returns the maximum element of a vector. |
| 470 | /// |
| 471 | /// `T` must be a vector of integers or floats. |
| 472 | /// |
| 473 | /// `U` must be the element type of `T`. |
| 474 | /// |
| 475 | /// For floating-point values, uses IEEE-754 `maxNum`. |
| 476 | #[rustc_intrinsic ] |
| 477 | #[rustc_nounwind ] |
| 478 | pub unsafe fn simd_reduce_max<T, U>(x: T) -> U; |
| 479 | |
| 480 | /// Returns the minimum element of a vector. |
| 481 | /// |
| 482 | /// `T` must be a vector of integers or floats. |
| 483 | /// |
| 484 | /// `U` must be the element type of `T`. |
| 485 | /// |
| 486 | /// For floating-point values, uses IEEE-754 `minNum`. |
| 487 | #[rustc_intrinsic ] |
| 488 | #[rustc_nounwind ] |
| 489 | pub unsafe fn simd_reduce_min<T, U>(x: T) -> U; |
| 490 | |
| 491 | /// Logical "ands" all elements together. |
| 492 | /// |
| 493 | /// `T` must be a vector of integers or floats. |
| 494 | /// |
| 495 | /// `U` must be the element type of `T`. |
| 496 | #[rustc_intrinsic ] |
| 497 | #[rustc_nounwind ] |
| 498 | pub unsafe fn simd_reduce_and<T, U>(x: T) -> U; |
| 499 | |
| 500 | /// Logical "ors" all elements together. |
| 501 | /// |
| 502 | /// `T` must be a vector of integers or floats. |
| 503 | /// |
| 504 | /// `U` must be the element type of `T`. |
| 505 | #[rustc_intrinsic ] |
| 506 | #[rustc_nounwind ] |
| 507 | pub unsafe fn simd_reduce_or<T, U>(x: T) -> U; |
| 508 | |
| 509 | /// Logical "exclusive ors" all elements together. |
| 510 | /// |
| 511 | /// `T` must be a vector of integers or floats. |
| 512 | /// |
| 513 | /// `U` must be the element type of `T`. |
| 514 | #[rustc_intrinsic ] |
| 515 | #[rustc_nounwind ] |
| 516 | pub unsafe fn simd_reduce_xor<T, U>(x: T) -> U; |
| 517 | |
| 518 | /// Truncates an integer vector to a bitmask. |
| 519 | /// |
| 520 | /// `T` must be an integer vector. |
| 521 | /// |
| 522 | /// `U` must be either the smallest unsigned integer with at least as many bits as the length |
| 523 | /// of `T`, or the smallest array of `u8` with at least as many bits as the length of `T`. |
| 524 | /// |
| 525 | /// Each element is truncated to a single bit and packed into the result. |
| 526 | /// |
| 527 | /// No matter whether the output is an array or an unsigned integer, it is treated as a single |
| 528 | /// contiguous list of bits. The bitmask is always packed on the least-significant side of the |
| 529 | /// output, and padded with 0s in the most-significant bits. The order of the bits depends on |
| 530 | /// endianness: |
| 531 | /// |
| 532 | /// * On little endian, the least significant bit corresponds to the first vector element. |
| 533 | /// * On big endian, the least significant bit corresponds to the last vector element. |
| 534 | /// |
| 535 | /// For example, `[!0, 0, !0, !0]` packs to |
| 536 | /// - `0b1101u8` or `[0b1101]` on little endian, and |
| 537 | /// - `0b1011u8` or `[0b1011]` on big endian. |
| 538 | /// |
| 539 | /// To consider a larger example, |
| 540 | /// `[!0, 0, 0, 0, 0, 0, 0, 0, !0, !0, 0, 0, 0, 0, !0, 0]` packs to |
| 541 | /// - `0b0100001100000001u16` or `[0b00000001, 0b01000011]` on little endian, and |
| 542 | /// - `0b1000000011000010u16` or `[0b10000000, 0b11000010]` on big endian. |
| 543 | /// |
| 544 | /// And finally, a non-power-of-2 example with multiple bytes: |
| 545 | /// `[!0, !0, 0, !0, 0, 0, !0, 0, !0, 0]` packs to |
| 546 | /// - `0b0101001011u16` or `[0b01001011, 0b01]` on little endian, and |
| 547 | /// - `0b1101001010u16` or `[0b11, 0b01001010]` on big endian. |
| 548 | /// |
| 549 | /// # Safety |
| 550 | /// `x` must contain only `0` and `!0`. |
| 551 | #[rustc_intrinsic ] |
| 552 | #[rustc_nounwind ] |
| 553 | pub unsafe fn simd_bitmask<T, U>(x: T) -> U; |
| 554 | |
| 555 | /// Selects elements from a mask. |
| 556 | /// |
| 557 | /// `T` must be a vector. |
| 558 | /// |
| 559 | /// `M` must be an integer vector with the same length as `T` (but any element size). |
| 560 | /// |
| 561 | /// For each element, if the corresponding value in `mask` is `!0`, select the element from |
| 562 | /// `if_true`. If the corresponding value in `mask` is `0`, select the element from |
| 563 | /// `if_false`. |
| 564 | /// |
| 565 | /// # Safety |
| 566 | /// `mask` must only contain `0` and `!0`. |
| 567 | #[rustc_intrinsic ] |
| 568 | #[rustc_nounwind ] |
| 569 | pub unsafe fn simd_select<M, T>(mask: M, if_true: T, if_false: T) -> T; |
| 570 | |
| 571 | /// Selects elements from a bitmask. |
| 572 | /// |
| 573 | /// `M` must be an unsigned integer or array of `u8`, matching `simd_bitmask`. |
| 574 | /// |
| 575 | /// `T` must be a vector. |
| 576 | /// |
| 577 | /// For each element, if the bit in `mask` is `1`, select the element from |
| 578 | /// `if_true`. If the corresponding bit in `mask` is `0`, select the element from |
| 579 | /// `if_false`. |
| 580 | /// The remaining bits of the mask are ignored. |
| 581 | /// |
| 582 | /// The bitmask bit order matches `simd_bitmask`. |
| 583 | #[rustc_intrinsic ] |
| 584 | #[rustc_nounwind ] |
| 585 | pub unsafe fn simd_select_bitmask<M, T>(m: M, yes: T, no: T) -> T; |
| 586 | |
| 587 | /// Calculates the offset from a pointer vector elementwise, potentially |
| 588 | /// wrapping. |
| 589 | /// |
| 590 | /// `T` must be a vector of pointers. |
| 591 | /// |
| 592 | /// `U` must be a vector of `isize` or `usize` with the same number of elements as `T`. |
| 593 | /// |
| 594 | /// Operates as if by `<ptr>::wrapping_offset`. |
| 595 | #[rustc_intrinsic ] |
| 596 | #[rustc_nounwind ] |
| 597 | pub unsafe fn simd_arith_offset<T, U>(ptr: T, offset: U) -> T; |
| 598 | |
| 599 | /// Casts a vector of pointers. |
| 600 | /// |
| 601 | /// `T` and `U` must be vectors of pointers with the same number of elements. |
| 602 | #[rustc_intrinsic ] |
| 603 | #[rustc_nounwind ] |
| 604 | pub unsafe fn simd_cast_ptr<T, U>(ptr: T) -> U; |
| 605 | |
| 606 | /// Exposes a vector of pointers as a vector of addresses. |
| 607 | /// |
| 608 | /// `T` must be a vector of pointers. |
| 609 | /// |
| 610 | /// `U` must be a vector of `usize` with the same length as `T`. |
| 611 | #[rustc_intrinsic ] |
| 612 | #[rustc_nounwind ] |
| 613 | pub unsafe fn simd_expose_provenance<T, U>(ptr: T) -> U; |
| 614 | |
| 615 | /// Creates a vector of pointers from a vector of addresses. |
| 616 | /// |
| 617 | /// `T` must be a vector of `usize`. |
| 618 | /// |
| 619 | /// `U` must be a vector of pointers, with the same length as `T`. |
| 620 | #[rustc_intrinsic ] |
| 621 | #[rustc_nounwind ] |
| 622 | pub unsafe fn simd_with_exposed_provenance<T, U>(addr: T) -> U; |
| 623 | |
| 624 | /// Swaps bytes of each element. |
| 625 | /// |
| 626 | /// `T` must be a vector of integers. |
| 627 | #[rustc_intrinsic ] |
| 628 | #[rustc_nounwind ] |
| 629 | pub unsafe fn simd_bswap<T>(x: T) -> T; |
| 630 | |
| 631 | /// Reverses bits of each element. |
| 632 | /// |
| 633 | /// `T` must be a vector of integers. |
| 634 | #[rustc_intrinsic ] |
| 635 | #[rustc_nounwind ] |
| 636 | pub unsafe fn simd_bitreverse<T>(x: T) -> T; |
| 637 | |
| 638 | /// Counts the leading zeros of each element. |
| 639 | /// |
| 640 | /// `T` must be a vector of integers. |
| 641 | #[rustc_intrinsic ] |
| 642 | #[rustc_nounwind ] |
| 643 | pub unsafe fn simd_ctlz<T>(x: T) -> T; |
| 644 | |
| 645 | /// Counts the number of ones in each element. |
| 646 | /// |
| 647 | /// `T` must be a vector of integers. |
| 648 | #[rustc_intrinsic ] |
| 649 | #[rustc_nounwind ] |
| 650 | pub unsafe fn simd_ctpop<T>(x: T) -> T; |
| 651 | |
| 652 | /// Counts the trailing zeros of each element. |
| 653 | /// |
| 654 | /// `T` must be a vector of integers. |
| 655 | #[rustc_intrinsic ] |
| 656 | #[rustc_nounwind ] |
| 657 | pub unsafe fn simd_cttz<T>(x: T) -> T; |
| 658 | |
| 659 | /// Rounds up each element to the next highest integer-valued float. |
| 660 | /// |
| 661 | /// `T` must be a vector of floats. |
| 662 | #[rustc_intrinsic ] |
| 663 | #[rustc_nounwind ] |
| 664 | pub unsafe fn simd_ceil<T>(x: T) -> T; |
| 665 | |
| 666 | /// Rounds down each element to the next lowest integer-valued float. |
| 667 | /// |
| 668 | /// `T` must be a vector of floats. |
| 669 | #[rustc_intrinsic ] |
| 670 | #[rustc_nounwind ] |
| 671 | pub unsafe fn simd_floor<T>(x: T) -> T; |
| 672 | |
| 673 | /// Rounds each element to the closest integer-valued float. |
| 674 | /// Ties are resolved by rounding away from 0. |
| 675 | /// |
| 676 | /// `T` must be a vector of floats. |
| 677 | #[rustc_intrinsic ] |
| 678 | #[rustc_nounwind ] |
| 679 | pub unsafe fn simd_round<T>(x: T) -> T; |
| 680 | |
| 681 | /// Returns the integer part of each element as an integer-valued float. |
| 682 | /// In other words, non-integer values are truncated towards zero. |
| 683 | /// |
| 684 | /// `T` must be a vector of floats. |
| 685 | #[rustc_intrinsic ] |
| 686 | #[rustc_nounwind ] |
| 687 | pub unsafe fn simd_trunc<T>(x: T) -> T; |
| 688 | |
| 689 | /// Takes the square root of each element. |
| 690 | /// |
| 691 | /// `T` must be a vector of floats. |
| 692 | #[rustc_intrinsic ] |
| 693 | #[rustc_nounwind ] |
| 694 | pub unsafe fn simd_fsqrt<T>(x: T) -> T; |
| 695 | |
| 696 | /// Computes `(x*y) + z` for each element, but without any intermediate rounding. |
| 697 | /// |
| 698 | /// `T` must be a vector of floats. |
| 699 | #[rustc_intrinsic ] |
| 700 | #[rustc_nounwind ] |
| 701 | pub unsafe fn simd_fma<T>(x: T, y: T, z: T) -> T; |
| 702 | |
| 703 | /// Computes `(x*y) + z` for each element, non-deterministically executing either |
| 704 | /// a fused multiply-add or two operations with rounding of the intermediate result. |
| 705 | /// |
| 706 | /// The operation is fused if the code generator determines that target instruction |
| 707 | /// set has support for a fused operation, and that the fused operation is more efficient |
| 708 | /// than the equivalent, separate pair of mul and add instructions. It is unspecified |
| 709 | /// whether or not a fused operation is selected, and that may depend on optimization |
| 710 | /// level and context, for example. It may even be the case that some SIMD lanes get fused |
| 711 | /// and others do not. |
| 712 | /// |
| 713 | /// `T` must be a vector of floats. |
| 714 | #[rustc_intrinsic ] |
| 715 | #[rustc_nounwind ] |
| 716 | pub unsafe fn simd_relaxed_fma<T>(x: T, y: T, z: T) -> T; |
| 717 | |
| 718 | // Computes the sine of each element. |
| 719 | /// |
| 720 | /// `T` must be a vector of floats. |
| 721 | #[rustc_intrinsic ] |
| 722 | #[rustc_nounwind ] |
| 723 | pub unsafe fn simd_fsin<T>(a: T) -> T; |
| 724 | |
| 725 | // Computes the cosine of each element. |
| 726 | /// |
| 727 | /// `T` must be a vector of floats. |
| 728 | #[rustc_intrinsic ] |
| 729 | #[rustc_nounwind ] |
| 730 | pub unsafe fn simd_fcos<T>(a: T) -> T; |
| 731 | |
| 732 | // Computes the exponential function of each element. |
| 733 | /// |
| 734 | /// `T` must be a vector of floats. |
| 735 | #[rustc_intrinsic ] |
| 736 | #[rustc_nounwind ] |
| 737 | pub unsafe fn simd_fexp<T>(a: T) -> T; |
| 738 | |
| 739 | // Computes 2 raised to the power of each element. |
| 740 | /// |
| 741 | /// `T` must be a vector of floats. |
| 742 | #[rustc_intrinsic ] |
| 743 | #[rustc_nounwind ] |
| 744 | pub unsafe fn simd_fexp2<T>(a: T) -> T; |
| 745 | |
| 746 | // Computes the base 10 logarithm of each element. |
| 747 | /// |
| 748 | /// `T` must be a vector of floats. |
| 749 | #[rustc_intrinsic ] |
| 750 | #[rustc_nounwind ] |
| 751 | pub unsafe fn simd_flog10<T>(a: T) -> T; |
| 752 | |
| 753 | // Computes the base 2 logarithm of each element. |
| 754 | /// |
| 755 | /// `T` must be a vector of floats. |
| 756 | #[rustc_intrinsic ] |
| 757 | #[rustc_nounwind ] |
| 758 | pub unsafe fn simd_flog2<T>(a: T) -> T; |
| 759 | |
| 760 | // Computes the natural logarithm of each element. |
| 761 | /// |
| 762 | /// `T` must be a vector of floats. |
| 763 | #[rustc_intrinsic ] |
| 764 | #[rustc_nounwind ] |
| 765 | pub unsafe fn simd_flog<T>(a: T) -> T; |
| 766 | |