| 1 | #![allow (non_camel_case_types)] |
| 2 | |
| 3 | use crate::soft::{x2, x4}; |
| 4 | use crate::types::*; |
| 5 | use core::ops::*; |
| 6 | |
| 7 | #[repr (C)] |
| 8 | #[derive(Clone, Copy)] |
| 9 | pub union vec128_storage { |
| 10 | d: [u32; 4], |
| 11 | q: [u64; 2], |
| 12 | } |
| 13 | impl From<[u32; 4]> for vec128_storage { |
| 14 | #[inline (always)] |
| 15 | fn from(d: [u32; 4]) -> Self { |
| 16 | Self { d } |
| 17 | } |
| 18 | } |
| 19 | impl From<vec128_storage> for [u32; 4] { |
| 20 | #[inline (always)] |
| 21 | fn from(d: vec128_storage) -> Self { |
| 22 | unsafe { d.d } |
| 23 | } |
| 24 | } |
| 25 | impl From<[u64; 2]> for vec128_storage { |
| 26 | #[inline (always)] |
| 27 | fn from(q: [u64; 2]) -> Self { |
| 28 | Self { q } |
| 29 | } |
| 30 | } |
| 31 | impl From<vec128_storage> for [u64; 2] { |
| 32 | #[inline (always)] |
| 33 | fn from(q: vec128_storage) -> Self { |
| 34 | unsafe { q.q } |
| 35 | } |
| 36 | } |
| 37 | impl Default for vec128_storage { |
| 38 | #[inline (always)] |
| 39 | fn default() -> Self { |
| 40 | Self { q: [0, 0] } |
| 41 | } |
| 42 | } |
| 43 | impl Eq for vec128_storage {} |
| 44 | impl PartialEq<vec128_storage> for vec128_storage { |
| 45 | #[inline (always)] |
| 46 | fn eq(&self, rhs: &Self) -> bool { |
| 47 | unsafe { self.q == rhs.q } |
| 48 | } |
| 49 | } |
| 50 | #[derive(Clone, Copy, PartialEq, Eq, Default)] |
| 51 | pub struct vec256_storage { |
| 52 | v128: [vec128_storage; 2], |
| 53 | } |
| 54 | impl vec256_storage { |
| 55 | #[inline (always)] |
| 56 | pub fn new128(v128: [vec128_storage; 2]) -> Self { |
| 57 | Self { v128 } |
| 58 | } |
| 59 | #[inline (always)] |
| 60 | pub fn split128(self) -> [vec128_storage; 2] { |
| 61 | self.v128 |
| 62 | } |
| 63 | } |
| 64 | impl From<vec256_storage> for [u64; 4] { |
| 65 | #[inline (always)] |
| 66 | fn from(q: vec256_storage) -> Self { |
| 67 | let [a, b]: [u64; 2] = q.v128[0].into(); |
| 68 | let [c, d]: [u64; 2] = q.v128[1].into(); |
| 69 | [a, b, c, d] |
| 70 | } |
| 71 | } |
| 72 | impl From<[u64; 4]> for vec256_storage { |
| 73 | #[inline (always)] |
| 74 | fn from([a, b, c, d]: [u64; 4]) -> Self { |
| 75 | Self { |
| 76 | v128: [[a, b].into(), [c, d].into()], |
| 77 | } |
| 78 | } |
| 79 | } |
| 80 | #[derive(Clone, Copy, PartialEq, Eq, Default)] |
| 81 | pub struct vec512_storage { |
| 82 | v128: [vec128_storage; 4], |
| 83 | } |
| 84 | impl vec512_storage { |
| 85 | #[inline (always)] |
| 86 | pub fn new128(v128: [vec128_storage; 4]) -> Self { |
| 87 | Self { v128 } |
| 88 | } |
| 89 | #[inline (always)] |
| 90 | pub fn split128(self) -> [vec128_storage; 4] { |
| 91 | self.v128 |
| 92 | } |
| 93 | } |
| 94 | |
| 95 | #[inline (always)] |
| 96 | fn dmap<T, F>(t: T, f: F) -> T |
| 97 | where |
| 98 | T: Store<vec128_storage> + Into<vec128_storage>, |
| 99 | F: Fn(u32) -> u32, |
| 100 | { |
| 101 | let t: vec128_storage = t.into(); |
| 102 | let d = unsafe { t.d }; |
| 103 | let d = vec128_storage { |
| 104 | d: [f(d[0]), f(d[1]), f(d[2]), f(d[3])], |
| 105 | }; |
| 106 | unsafe { T::unpack(d) } |
| 107 | } |
| 108 | |
| 109 | fn dmap2<T, F>(a: T, b: T, f: F) -> T |
| 110 | where |
| 111 | T: Store<vec128_storage> + Into<vec128_storage>, |
| 112 | F: Fn(u32, u32) -> u32, |
| 113 | { |
| 114 | let a: vec128_storage = a.into(); |
| 115 | let b: vec128_storage = b.into(); |
| 116 | let ao = unsafe { a.d }; |
| 117 | let bo = unsafe { b.d }; |
| 118 | let d = vec128_storage { |
| 119 | d: [ |
| 120 | f(ao[0], bo[0]), |
| 121 | f(ao[1], bo[1]), |
| 122 | f(ao[2], bo[2]), |
| 123 | f(ao[3], bo[3]), |
| 124 | ], |
| 125 | }; |
| 126 | unsafe { T::unpack(d) } |
| 127 | } |
| 128 | |
| 129 | #[inline (always)] |
| 130 | fn qmap<T, F>(t: T, f: F) -> T |
| 131 | where |
| 132 | T: Store<vec128_storage> + Into<vec128_storage>, |
| 133 | F: Fn(u64) -> u64, |
| 134 | { |
| 135 | let t: vec128_storage = t.into(); |
| 136 | let q = unsafe { t.q }; |
| 137 | let q = vec128_storage { |
| 138 | q: [f(q[0]), f(q[1])], |
| 139 | }; |
| 140 | unsafe { T::unpack(q) } |
| 141 | } |
| 142 | |
| 143 | #[inline (always)] |
| 144 | fn qmap2<T, F>(a: T, b: T, f: F) -> T |
| 145 | where |
| 146 | T: Store<vec128_storage> + Into<vec128_storage>, |
| 147 | F: Fn(u64, u64) -> u64, |
| 148 | { |
| 149 | let a: vec128_storage = a.into(); |
| 150 | let b: vec128_storage = b.into(); |
| 151 | let ao = unsafe { a.q }; |
| 152 | let bo = unsafe { b.q }; |
| 153 | let q = vec128_storage { |
| 154 | q: [f(ao[0], bo[0]), f(ao[1], bo[1])], |
| 155 | }; |
| 156 | unsafe { T::unpack(q) } |
| 157 | } |
| 158 | |
| 159 | #[inline (always)] |
| 160 | fn o_of_q(q: [u64; 2]) -> u128 { |
| 161 | u128::from(q[0]) | (u128::from(q[1]) << 64) |
| 162 | } |
| 163 | |
| 164 | #[inline (always)] |
| 165 | fn q_of_o(o: u128) -> [u64; 2] { |
| 166 | [o as u64, (o >> 64) as u64] |
| 167 | } |
| 168 | |
| 169 | #[inline (always)] |
| 170 | fn omap<T, F>(a: T, f: F) -> T |
| 171 | where |
| 172 | T: Store<vec128_storage> + Into<vec128_storage>, |
| 173 | F: Fn(u128) -> u128, |
| 174 | { |
| 175 | let a: vec128_storage = a.into(); |
| 176 | let ao = o_of_q(unsafe { a.q }); |
| 177 | let o = vec128_storage { q: q_of_o(f(ao)) }; |
| 178 | unsafe { T::unpack(o) } |
| 179 | } |
| 180 | |
| 181 | #[inline (always)] |
| 182 | fn omap2<T, F>(a: T, b: T, f: F) -> T |
| 183 | where |
| 184 | T: Store<vec128_storage> + Into<vec128_storage>, |
| 185 | F: Fn(u128, u128) -> u128, |
| 186 | { |
| 187 | let a: vec128_storage = a.into(); |
| 188 | let b: vec128_storage = b.into(); |
| 189 | let ao = o_of_q(unsafe { a.q }); |
| 190 | let bo = o_of_q(unsafe { b.q }); |
| 191 | let o = vec128_storage { |
| 192 | q: q_of_o(f(ao, bo)), |
| 193 | }; |
| 194 | unsafe { T::unpack(o) } |
| 195 | } |
| 196 | |
| 197 | impl RotateEachWord128 for u128x1_generic {} |
| 198 | impl BitOps128 for u128x1_generic {} |
| 199 | impl BitOps64 for u128x1_generic {} |
| 200 | impl BitOps64 for u64x2_generic {} |
| 201 | impl BitOps32 for u128x1_generic {} |
| 202 | impl BitOps32 for u64x2_generic {} |
| 203 | impl BitOps32 for u32x4_generic {} |
| 204 | impl BitOps0 for u128x1_generic {} |
| 205 | impl BitOps0 for u64x2_generic {} |
| 206 | impl BitOps0 for u32x4_generic {} |
| 207 | |
| 208 | macro_rules! impl_bitops { |
| 209 | ($vec:ident) => { |
| 210 | impl Not for $vec { |
| 211 | type Output = Self; |
| 212 | #[inline(always)] |
| 213 | fn not(self) -> Self::Output { |
| 214 | omap(self, |x| !x) |
| 215 | } |
| 216 | } |
| 217 | impl BitAnd for $vec { |
| 218 | type Output = Self; |
| 219 | #[inline(always)] |
| 220 | fn bitand(self, rhs: Self) -> Self::Output { |
| 221 | omap2(self, rhs, |x, y| x & y) |
| 222 | } |
| 223 | } |
| 224 | impl BitOr for $vec { |
| 225 | type Output = Self; |
| 226 | #[inline(always)] |
| 227 | fn bitor(self, rhs: Self) -> Self::Output { |
| 228 | omap2(self, rhs, |x, y| x | y) |
| 229 | } |
| 230 | } |
| 231 | impl BitXor for $vec { |
| 232 | type Output = Self; |
| 233 | #[inline(always)] |
| 234 | fn bitxor(self, rhs: Self) -> Self::Output { |
| 235 | omap2(self, rhs, |x, y| x ^ y) |
| 236 | } |
| 237 | } |
| 238 | impl AndNot for $vec { |
| 239 | type Output = Self; |
| 240 | #[inline(always)] |
| 241 | fn andnot(self, rhs: Self) -> Self::Output { |
| 242 | omap2(self, rhs, |x, y| !x & y) |
| 243 | } |
| 244 | } |
| 245 | impl BitAndAssign for $vec { |
| 246 | #[inline(always)] |
| 247 | fn bitand_assign(&mut self, rhs: Self) { |
| 248 | *self = *self & rhs |
| 249 | } |
| 250 | } |
| 251 | impl BitOrAssign for $vec { |
| 252 | #[inline(always)] |
| 253 | fn bitor_assign(&mut self, rhs: Self) { |
| 254 | *self = *self | rhs |
| 255 | } |
| 256 | } |
| 257 | impl BitXorAssign for $vec { |
| 258 | #[inline(always)] |
| 259 | fn bitxor_assign(&mut self, rhs: Self) { |
| 260 | *self = *self ^ rhs |
| 261 | } |
| 262 | } |
| 263 | |
| 264 | impl Swap64 for $vec { |
| 265 | #[inline(always)] |
| 266 | fn swap1(self) -> Self { |
| 267 | qmap(self, |x| { |
| 268 | ((x & 0x5555555555555555) << 1) | ((x & 0xaaaaaaaaaaaaaaaa) >> 1) |
| 269 | }) |
| 270 | } |
| 271 | #[inline(always)] |
| 272 | fn swap2(self) -> Self { |
| 273 | qmap(self, |x| { |
| 274 | ((x & 0x3333333333333333) << 2) | ((x & 0xcccccccccccccccc) >> 2) |
| 275 | }) |
| 276 | } |
| 277 | #[inline(always)] |
| 278 | fn swap4(self) -> Self { |
| 279 | qmap(self, |x| { |
| 280 | ((x & 0x0f0f0f0f0f0f0f0f) << 4) | ((x & 0xf0f0f0f0f0f0f0f0) >> 4) |
| 281 | }) |
| 282 | } |
| 283 | #[inline(always)] |
| 284 | fn swap8(self) -> Self { |
| 285 | qmap(self, |x| { |
| 286 | ((x & 0x00ff00ff00ff00ff) << 8) | ((x & 0xff00ff00ff00ff00) >> 8) |
| 287 | }) |
| 288 | } |
| 289 | #[inline(always)] |
| 290 | fn swap16(self) -> Self { |
| 291 | dmap(self, |x| x.rotate_left(16)) |
| 292 | } |
| 293 | #[inline(always)] |
| 294 | fn swap32(self) -> Self { |
| 295 | qmap(self, |x| x.rotate_left(32)) |
| 296 | } |
| 297 | #[inline(always)] |
| 298 | fn swap64(self) -> Self { |
| 299 | omap(self, |x| (x << 64) | (x >> 64)) |
| 300 | } |
| 301 | } |
| 302 | }; |
| 303 | } |
| 304 | impl_bitops!(u32x4_generic); |
| 305 | impl_bitops!(u64x2_generic); |
| 306 | impl_bitops!(u128x1_generic); |
| 307 | |
| 308 | impl RotateEachWord32 for u32x4_generic { |
| 309 | #[inline (always)] |
| 310 | fn rotate_each_word_right7(self) -> Self { |
| 311 | dmap(self, |x| x.rotate_right(7)) |
| 312 | } |
| 313 | #[inline (always)] |
| 314 | fn rotate_each_word_right8(self) -> Self { |
| 315 | dmap(self, |x| x.rotate_right(8)) |
| 316 | } |
| 317 | #[inline (always)] |
| 318 | fn rotate_each_word_right11(self) -> Self { |
| 319 | dmap(self, |x| x.rotate_right(11)) |
| 320 | } |
| 321 | #[inline (always)] |
| 322 | fn rotate_each_word_right12(self) -> Self { |
| 323 | dmap(self, |x| x.rotate_right(12)) |
| 324 | } |
| 325 | #[inline (always)] |
| 326 | fn rotate_each_word_right16(self) -> Self { |
| 327 | dmap(self, |x| x.rotate_right(16)) |
| 328 | } |
| 329 | #[inline (always)] |
| 330 | fn rotate_each_word_right20(self) -> Self { |
| 331 | dmap(self, |x| x.rotate_right(20)) |
| 332 | } |
| 333 | #[inline (always)] |
| 334 | fn rotate_each_word_right24(self) -> Self { |
| 335 | dmap(self, |x| x.rotate_right(24)) |
| 336 | } |
| 337 | #[inline (always)] |
| 338 | fn rotate_each_word_right25(self) -> Self { |
| 339 | dmap(self, |x| x.rotate_right(25)) |
| 340 | } |
| 341 | } |
| 342 | |
| 343 | impl RotateEachWord32 for u64x2_generic { |
| 344 | #[inline (always)] |
| 345 | fn rotate_each_word_right7(self) -> Self { |
| 346 | qmap(self, |x| x.rotate_right(7)) |
| 347 | } |
| 348 | #[inline (always)] |
| 349 | fn rotate_each_word_right8(self) -> Self { |
| 350 | qmap(self, |x| x.rotate_right(8)) |
| 351 | } |
| 352 | #[inline (always)] |
| 353 | fn rotate_each_word_right11(self) -> Self { |
| 354 | qmap(self, |x| x.rotate_right(11)) |
| 355 | } |
| 356 | #[inline (always)] |
| 357 | fn rotate_each_word_right12(self) -> Self { |
| 358 | qmap(self, |x| x.rotate_right(12)) |
| 359 | } |
| 360 | #[inline (always)] |
| 361 | fn rotate_each_word_right16(self) -> Self { |
| 362 | qmap(self, |x| x.rotate_right(16)) |
| 363 | } |
| 364 | #[inline (always)] |
| 365 | fn rotate_each_word_right20(self) -> Self { |
| 366 | qmap(self, |x| x.rotate_right(20)) |
| 367 | } |
| 368 | #[inline (always)] |
| 369 | fn rotate_each_word_right24(self) -> Self { |
| 370 | qmap(self, |x| x.rotate_right(24)) |
| 371 | } |
| 372 | #[inline (always)] |
| 373 | fn rotate_each_word_right25(self) -> Self { |
| 374 | qmap(self, |x| x.rotate_right(25)) |
| 375 | } |
| 376 | } |
| 377 | impl RotateEachWord64 for u64x2_generic { |
| 378 | #[inline (always)] |
| 379 | fn rotate_each_word_right32(self) -> Self { |
| 380 | qmap(self, |x| x.rotate_right(32)) |
| 381 | } |
| 382 | } |
| 383 | |
| 384 | // workaround for koute/cargo-web#52 (u128::rotate_* broken with cargo web) |
| 385 | #[inline (always)] |
| 386 | fn rotate_u128_right(x: u128, i: u32) -> u128 { |
| 387 | (x >> i) | (x << (128 - i)) |
| 388 | } |
| 389 | #[test] |
| 390 | fn test_rotate_u128() { |
| 391 | const X: u128 = 0x0001_0203_0405_0607_0809_0a0b_0c0d_0e0f; |
| 392 | assert_eq!(rotate_u128_right(X, 17), X.rotate_right(17)); |
| 393 | } |
| 394 | |
| 395 | impl RotateEachWord32 for u128x1_generic { |
| 396 | #[inline (always)] |
| 397 | fn rotate_each_word_right7(self) -> Self { |
| 398 | Self([rotate_u128_right(self.0[0], 7)]) |
| 399 | } |
| 400 | #[inline (always)] |
| 401 | fn rotate_each_word_right8(self) -> Self { |
| 402 | Self([rotate_u128_right(self.0[0], 8)]) |
| 403 | } |
| 404 | #[inline (always)] |
| 405 | fn rotate_each_word_right11(self) -> Self { |
| 406 | Self([rotate_u128_right(self.0[0], 11)]) |
| 407 | } |
| 408 | #[inline (always)] |
| 409 | fn rotate_each_word_right12(self) -> Self { |
| 410 | Self([rotate_u128_right(self.0[0], 12)]) |
| 411 | } |
| 412 | #[inline (always)] |
| 413 | fn rotate_each_word_right16(self) -> Self { |
| 414 | Self([rotate_u128_right(self.0[0], 16)]) |
| 415 | } |
| 416 | #[inline (always)] |
| 417 | fn rotate_each_word_right20(self) -> Self { |
| 418 | Self([rotate_u128_right(self.0[0], 20)]) |
| 419 | } |
| 420 | #[inline (always)] |
| 421 | fn rotate_each_word_right24(self) -> Self { |
| 422 | Self([rotate_u128_right(self.0[0], 24)]) |
| 423 | } |
| 424 | #[inline (always)] |
| 425 | fn rotate_each_word_right25(self) -> Self { |
| 426 | Self([rotate_u128_right(self.0[0], 25)]) |
| 427 | } |
| 428 | } |
| 429 | impl RotateEachWord64 for u128x1_generic { |
| 430 | #[inline (always)] |
| 431 | fn rotate_each_word_right32(self) -> Self { |
| 432 | Self([rotate_u128_right(self.0[0], 32)]) |
| 433 | } |
| 434 | } |
| 435 | |
| 436 | #[derive(Copy, Clone)] |
| 437 | pub struct GenericMachine; |
| 438 | impl Machine for GenericMachine { |
| 439 | type u32x4 = u32x4_generic; |
| 440 | type u64x2 = u64x2_generic; |
| 441 | type u128x1 = u128x1_generic; |
| 442 | type u32x4x2 = u32x4x2_generic; |
| 443 | type u64x2x2 = u64x2x2_generic; |
| 444 | type u64x4 = u64x4_generic; |
| 445 | type u128x2 = u128x2_generic; |
| 446 | type u32x4x4 = u32x4x4_generic; |
| 447 | type u64x2x4 = u64x2x4_generic; |
| 448 | type u128x4 = u128x4_generic; |
| 449 | #[inline (always)] |
| 450 | unsafe fn instance() -> Self { |
| 451 | Self |
| 452 | } |
| 453 | } |
| 454 | |
| 455 | #[derive(Copy, Clone, Debug, PartialEq)] |
| 456 | pub struct u32x4_generic([u32; 4]); |
| 457 | #[derive(Copy, Clone, Debug, PartialEq)] |
| 458 | pub struct u64x2_generic([u64; 2]); |
| 459 | #[derive(Copy, Clone, Debug, PartialEq)] |
| 460 | pub struct u128x1_generic([u128; 1]); |
| 461 | |
| 462 | impl From<u32x4_generic> for vec128_storage { |
| 463 | #[inline (always)] |
| 464 | fn from(d: u32x4_generic) -> Self { |
| 465 | Self { d: d.0 } |
| 466 | } |
| 467 | } |
| 468 | impl From<u64x2_generic> for vec128_storage { |
| 469 | #[inline (always)] |
| 470 | fn from(q: u64x2_generic) -> Self { |
| 471 | Self { q: q.0 } |
| 472 | } |
| 473 | } |
| 474 | impl From<u128x1_generic> for vec128_storage { |
| 475 | #[inline (always)] |
| 476 | fn from(o: u128x1_generic) -> Self { |
| 477 | Self { q: q_of_o(o.0[0]) } |
| 478 | } |
| 479 | } |
| 480 | |
| 481 | impl Store<vec128_storage> for u32x4_generic { |
| 482 | #[inline (always)] |
| 483 | unsafe fn unpack(s: vec128_storage) -> Self { |
| 484 | Self(s.d) |
| 485 | } |
| 486 | } |
| 487 | impl Store<vec128_storage> for u64x2_generic { |
| 488 | #[inline (always)] |
| 489 | unsafe fn unpack(s: vec128_storage) -> Self { |
| 490 | Self(s.q) |
| 491 | } |
| 492 | } |
| 493 | impl Store<vec128_storage> for u128x1_generic { |
| 494 | #[inline (always)] |
| 495 | unsafe fn unpack(s: vec128_storage) -> Self { |
| 496 | Self([o_of_q(s.q); 1]) |
| 497 | } |
| 498 | } |
| 499 | |
| 500 | impl ArithOps for u32x4_generic {} |
| 501 | impl ArithOps for u64x2_generic {} |
| 502 | impl ArithOps for u128x1_generic {} |
| 503 | |
| 504 | impl Add for u32x4_generic { |
| 505 | type Output = Self; |
| 506 | #[inline (always)] |
| 507 | fn add(self, rhs: Self) -> Self::Output { |
| 508 | dmap2(self, rhs, |x, y| x.wrapping_add(y)) |
| 509 | } |
| 510 | } |
| 511 | impl Add for u64x2_generic { |
| 512 | type Output = Self; |
| 513 | #[inline (always)] |
| 514 | fn add(self, rhs: Self) -> Self::Output { |
| 515 | qmap2(self, rhs, |x, y| x.wrapping_add(y)) |
| 516 | } |
| 517 | } |
| 518 | impl Add for u128x1_generic { |
| 519 | type Output = Self; |
| 520 | #[inline (always)] |
| 521 | fn add(self, rhs: Self) -> Self::Output { |
| 522 | omap2(self, rhs, |x, y| x.wrapping_add(y)) |
| 523 | } |
| 524 | } |
| 525 | impl AddAssign for u32x4_generic { |
| 526 | #[inline (always)] |
| 527 | fn add_assign(&mut self, rhs: Self) { |
| 528 | *self = *self + rhs |
| 529 | } |
| 530 | } |
| 531 | impl AddAssign for u64x2_generic { |
| 532 | #[inline (always)] |
| 533 | fn add_assign(&mut self, rhs: Self) { |
| 534 | *self = *self + rhs |
| 535 | } |
| 536 | } |
| 537 | impl AddAssign for u128x1_generic { |
| 538 | #[inline (always)] |
| 539 | fn add_assign(&mut self, rhs: Self) { |
| 540 | *self = *self + rhs |
| 541 | } |
| 542 | } |
| 543 | impl BSwap for u32x4_generic { |
| 544 | #[inline (always)] |
| 545 | fn bswap(self) -> Self { |
| 546 | dmap(self, |x| x.swap_bytes()) |
| 547 | } |
| 548 | } |
| 549 | impl BSwap for u64x2_generic { |
| 550 | #[inline (always)] |
| 551 | fn bswap(self) -> Self { |
| 552 | qmap(self, |x| x.swap_bytes()) |
| 553 | } |
| 554 | } |
| 555 | impl BSwap for u128x1_generic { |
| 556 | #[inline (always)] |
| 557 | fn bswap(self) -> Self { |
| 558 | omap(self, |x| x.swap_bytes()) |
| 559 | } |
| 560 | } |
| 561 | impl StoreBytes for u32x4_generic { |
| 562 | #[inline (always)] |
| 563 | unsafe fn unsafe_read_le(input: &[u8]) -> Self { |
| 564 | assert_eq!(input.len(), 16); |
| 565 | let x = core::mem::transmute(core::ptr::read(input as *const _ as *const [u8; 16])); |
| 566 | dmap(x, |x| x.to_le()) |
| 567 | } |
| 568 | #[inline (always)] |
| 569 | unsafe fn unsafe_read_be(input: &[u8]) -> Self { |
| 570 | assert_eq!(input.len(), 16); |
| 571 | let x = core::mem::transmute(core::ptr::read(input as *const _ as *const [u8; 16])); |
| 572 | dmap(x, |x| x.to_be()) |
| 573 | } |
| 574 | #[inline (always)] |
| 575 | fn write_le(self, out: &mut [u8]) { |
| 576 | assert_eq!(out.len(), 16); |
| 577 | let x = dmap(self, |x| x.to_le()); |
| 578 | unsafe { core::ptr::write(out as *mut _ as *mut [u8; 16], core::mem::transmute(x)) } |
| 579 | } |
| 580 | #[inline (always)] |
| 581 | fn write_be(self, out: &mut [u8]) { |
| 582 | assert_eq!(out.len(), 16); |
| 583 | let x = dmap(self, |x| x.to_be()); |
| 584 | unsafe { core::ptr::write(out as *mut _ as *mut [u8; 16], core::mem::transmute(x)) } |
| 585 | } |
| 586 | } |
| 587 | impl StoreBytes for u64x2_generic { |
| 588 | #[inline (always)] |
| 589 | unsafe fn unsafe_read_le(input: &[u8]) -> Self { |
| 590 | assert_eq!(input.len(), 16); |
| 591 | let x = core::mem::transmute(core::ptr::read(input as *const _ as *const [u8; 16])); |
| 592 | qmap(x, |x| x.to_le()) |
| 593 | } |
| 594 | #[inline (always)] |
| 595 | unsafe fn unsafe_read_be(input: &[u8]) -> Self { |
| 596 | assert_eq!(input.len(), 16); |
| 597 | let x = core::mem::transmute(core::ptr::read(input as *const _ as *const [u8; 16])); |
| 598 | qmap(x, |x| x.to_be()) |
| 599 | } |
| 600 | #[inline (always)] |
| 601 | fn write_le(self, out: &mut [u8]) { |
| 602 | assert_eq!(out.len(), 16); |
| 603 | let x = qmap(self, |x| x.to_le()); |
| 604 | unsafe { core::ptr::write(out as *mut _ as *mut [u8; 16], core::mem::transmute(x)) } |
| 605 | } |
| 606 | #[inline (always)] |
| 607 | fn write_be(self, out: &mut [u8]) { |
| 608 | assert_eq!(out.len(), 16); |
| 609 | let x = qmap(self, |x| x.to_be()); |
| 610 | unsafe { core::ptr::write(out as *mut _ as *mut [u8; 16], core::mem::transmute(x)) } |
| 611 | } |
| 612 | } |
| 613 | |
| 614 | #[derive(Copy, Clone)] |
| 615 | pub struct G0; |
| 616 | #[derive(Copy, Clone)] |
| 617 | pub struct G1; |
| 618 | pub type u32x4x2_generic = x2<u32x4_generic, G0>; |
| 619 | pub type u64x2x2_generic = x2<u64x2_generic, G0>; |
| 620 | pub type u64x4_generic = x2<u64x2_generic, G1>; |
| 621 | pub type u128x2_generic = x2<u128x1_generic, G0>; |
| 622 | pub type u32x4x4_generic = x4<u32x4_generic>; |
| 623 | pub type u64x2x4_generic = x4<u64x2_generic>; |
| 624 | pub type u128x4_generic = x4<u128x1_generic>; |
| 625 | |
| 626 | impl Vector<[u32; 16]> for u32x4x4_generic { |
| 627 | fn to_scalars(self) -> [u32; 16] { |
| 628 | let [a, b, c, d] = self.0; |
| 629 | let a = a.0; |
| 630 | let b = b.0; |
| 631 | let c = c.0; |
| 632 | let d = d.0; |
| 633 | [ |
| 634 | a[0], a[1], a[2], a[3], // |
| 635 | b[0], b[1], b[2], b[3], // |
| 636 | c[0], c[1], c[2], c[3], // |
| 637 | d[0], d[1], d[2], d[3], // |
| 638 | ] |
| 639 | } |
| 640 | } |
| 641 | |
| 642 | impl MultiLane<[u32; 4]> for u32x4_generic { |
| 643 | #[inline (always)] |
| 644 | fn to_lanes(self) -> [u32; 4] { |
| 645 | self.0 |
| 646 | } |
| 647 | #[inline (always)] |
| 648 | fn from_lanes(xs: [u32; 4]) -> Self { |
| 649 | Self(xs) |
| 650 | } |
| 651 | } |
| 652 | impl MultiLane<[u64; 2]> for u64x2_generic { |
| 653 | #[inline (always)] |
| 654 | fn to_lanes(self) -> [u64; 2] { |
| 655 | self.0 |
| 656 | } |
| 657 | #[inline (always)] |
| 658 | fn from_lanes(xs: [u64; 2]) -> Self { |
| 659 | Self(xs) |
| 660 | } |
| 661 | } |
| 662 | impl MultiLane<[u64; 4]> for u64x4_generic { |
| 663 | #[inline (always)] |
| 664 | fn to_lanes(self) -> [u64; 4] { |
| 665 | let (a, b) = (self.0[0].to_lanes(), self.0[1].to_lanes()); |
| 666 | [a[0], a[1], b[0], b[1]] |
| 667 | } |
| 668 | #[inline (always)] |
| 669 | fn from_lanes(xs: [u64; 4]) -> Self { |
| 670 | let (a, b) = ( |
| 671 | u64x2_generic::from_lanes([xs[0], xs[1]]), |
| 672 | u64x2_generic::from_lanes([xs[2], xs[3]]), |
| 673 | ); |
| 674 | x2::new([a, b]) |
| 675 | } |
| 676 | } |
| 677 | impl MultiLane<[u128; 1]> for u128x1_generic { |
| 678 | #[inline (always)] |
| 679 | fn to_lanes(self) -> [u128; 1] { |
| 680 | self.0 |
| 681 | } |
| 682 | #[inline (always)] |
| 683 | fn from_lanes(xs: [u128; 1]) -> Self { |
| 684 | Self(xs) |
| 685 | } |
| 686 | } |
| 687 | impl Vec4<u32> for u32x4_generic { |
| 688 | #[inline (always)] |
| 689 | fn extract(self, i: u32) -> u32 { |
| 690 | self.0[i as usize] |
| 691 | } |
| 692 | #[inline (always)] |
| 693 | fn insert(mut self, v: u32, i: u32) -> Self { |
| 694 | self.0[i as usize] = v; |
| 695 | self |
| 696 | } |
| 697 | } |
| 698 | impl Vec4<u64> for u64x4_generic { |
| 699 | #[inline (always)] |
| 700 | fn extract(self, i: u32) -> u64 { |
| 701 | let d: [u64; 4] = self.to_lanes(); |
| 702 | d[i as usize] |
| 703 | } |
| 704 | #[inline (always)] |
| 705 | fn insert(self, v: u64, i: u32) -> Self { |
| 706 | self.0[(i / 2) as usize].insert(v, i % 2); |
| 707 | self |
| 708 | } |
| 709 | } |
| 710 | impl Vec2<u64> for u64x2_generic { |
| 711 | #[inline (always)] |
| 712 | fn extract(self, i: u32) -> u64 { |
| 713 | self.0[i as usize] |
| 714 | } |
| 715 | #[inline (always)] |
| 716 | fn insert(mut self, v: u64, i: u32) -> Self { |
| 717 | self.0[i as usize] = v; |
| 718 | self |
| 719 | } |
| 720 | } |
| 721 | |
| 722 | impl Words4 for u32x4_generic { |
| 723 | #[inline (always)] |
| 724 | fn shuffle2301(self) -> Self { |
| 725 | self.swap64() |
| 726 | } |
| 727 | #[inline (always)] |
| 728 | fn shuffle1230(self) -> Self { |
| 729 | let x = self.0; |
| 730 | Self([x[3], x[0], x[1], x[2]]) |
| 731 | } |
| 732 | #[inline (always)] |
| 733 | fn shuffle3012(self) -> Self { |
| 734 | let x = self.0; |
| 735 | Self([x[1], x[2], x[3], x[0]]) |
| 736 | } |
| 737 | } |
| 738 | impl LaneWords4 for u32x4_generic { |
| 739 | #[inline (always)] |
| 740 | fn shuffle_lane_words2301(self) -> Self { |
| 741 | self.shuffle2301() |
| 742 | } |
| 743 | #[inline (always)] |
| 744 | fn shuffle_lane_words1230(self) -> Self { |
| 745 | self.shuffle1230() |
| 746 | } |
| 747 | #[inline (always)] |
| 748 | fn shuffle_lane_words3012(self) -> Self { |
| 749 | self.shuffle3012() |
| 750 | } |
| 751 | } |
| 752 | |
| 753 | impl Words4 for u64x4_generic { |
| 754 | #[inline (always)] |
| 755 | fn shuffle2301(self) -> Self { |
| 756 | x2::new([self.0[1], self.0[0]]) |
| 757 | } |
| 758 | #[inline (always)] |
| 759 | fn shuffle1230(self) -> Self { |
| 760 | unimplemented!() |
| 761 | } |
| 762 | #[inline (always)] |
| 763 | fn shuffle3012(self) -> Self { |
| 764 | unimplemented!() |
| 765 | } |
| 766 | } |
| 767 | |
| 768 | impl u32x4<GenericMachine> for u32x4_generic {} |
| 769 | impl u64x2<GenericMachine> for u64x2_generic {} |
| 770 | impl u128x1<GenericMachine> for u128x1_generic {} |
| 771 | impl u32x4x2<GenericMachine> for u32x4x2_generic {} |
| 772 | impl u64x2x2<GenericMachine> for u64x2x2_generic {} |
| 773 | impl u64x4<GenericMachine> for u64x4_generic {} |
| 774 | impl u128x2<GenericMachine> for u128x2_generic {} |
| 775 | impl u32x4x4<GenericMachine> for u32x4x4_generic {} |
| 776 | impl u64x2x4<GenericMachine> for u64x2x4_generic {} |
| 777 | impl u128x4<GenericMachine> for u128x4_generic {} |
| 778 | |
| 779 | #[macro_export ] |
| 780 | macro_rules! dispatch { |
| 781 | ($mach:ident, $MTy:ident, { $([$pub:tt$(($krate:tt))*])* fn $name:ident($($arg:ident: $argty:ty),* $(,)*) -> $ret:ty $body:block }) => { |
| 782 | #[inline(always)] |
| 783 | $($pub$(($krate))*)* fn $name($($arg: $argty),*) -> $ret { |
| 784 | let $mach = unsafe { $crate::generic::GenericMachine::instance() }; |
| 785 | #[inline(always)] |
| 786 | fn fn_impl<$MTy: $crate::Machine>($mach: $MTy, $($arg: $argty),*) -> $ret $body |
| 787 | fn_impl($mach, $($arg),*) |
| 788 | } |
| 789 | }; |
| 790 | ($mach:ident, $MTy:ident, { $([$pub:tt $(($krate:tt))*])* fn $name:ident($($arg:ident: $argty:ty),* $(,)*) $body:block }) => { |
| 791 | dispatch!($mach, $MTy, { |
| 792 | $([$pub $(($krate))*])* fn $name($($arg: $argty),*) -> () $body |
| 793 | }); |
| 794 | } |
| 795 | } |
| 796 | #[macro_export ] |
| 797 | macro_rules! dispatch_light128 { |
| 798 | ($mach:ident, $MTy:ident, { $([$pub:tt$(($krate:tt))*])* fn $name:ident($($arg:ident: $argty:ty),* $(,)*) -> $ret:ty $body:block }) => { |
| 799 | #[inline(always)] |
| 800 | $($pub$(($krate))*)* fn $name($($arg: $argty),*) -> $ret { |
| 801 | let $mach = unsafe { $crate::generic::GenericMachine::instance() }; |
| 802 | #[inline(always)] |
| 803 | fn fn_impl<$MTy: $crate::Machine>($mach: $MTy, $($arg: $argty),*) -> $ret $body |
| 804 | fn_impl($mach, $($arg),*) |
| 805 | } |
| 806 | }; |
| 807 | ($mach:ident, $MTy:ident, { $([$pub:tt $(($krate:tt))*])* fn $name:ident($($arg:ident: $argty:ty),* $(,)*) $body:block }) => { |
| 808 | dispatch!($mach, $MTy, { |
| 809 | $([$pub $(($krate))*])* fn $name($($arg: $argty),*) -> () $body |
| 810 | }); |
| 811 | } |
| 812 | } |
| 813 | #[macro_export ] |
| 814 | macro_rules! dispatch_light256 { |
| 815 | ($mach:ident, $MTy:ident, { $([$pub:tt$(($krate:tt))*])* fn $name:ident($($arg:ident: $argty:ty),* $(,)*) -> $ret:ty $body:block }) => { |
| 816 | #[inline(always)] |
| 817 | $($pub$(($krate))*)* fn $name($($arg: $argty),*) -> $ret { |
| 818 | let $mach = unsafe { $crate::generic::GenericMachine::instance() }; |
| 819 | #[inline(always)] |
| 820 | fn fn_impl<$MTy: $crate::Machine>($mach: $MTy, $($arg: $argty),*) -> $ret $body |
| 821 | fn_impl($mach, $($arg),*) |
| 822 | } |
| 823 | }; |
| 824 | ($mach:ident, $MTy:ident, { $([$pub:tt $(($krate:tt))*])* fn $name:ident($($arg:ident: $argty:ty),* $(,)*) $body:block }) => { |
| 825 | dispatch!($mach, $MTy, { |
| 826 | $([$pub $(($krate))*])* fn $name($($arg: $argty),*) -> () $body |
| 827 | }); |
| 828 | } |
| 829 | } |
| 830 | #[macro_export ] |
| 831 | macro_rules! dispatch_light512 { |
| 832 | ($mach:ident, $MTy:ident, { $([$pub:tt$(($krate:tt))*])* fn $name:ident($($arg:ident: $argty:ty),* $(,)*) -> $ret:ty $body:block }) => { |
| 833 | #[inline(always)] |
| 834 | $($pub$(($krate))*)* fn $name($($arg: $argty),*) -> $ret { |
| 835 | let $mach = unsafe { $crate::generic::GenericMachine::instance() }; |
| 836 | #[inline(always)] |
| 837 | fn fn_impl<$MTy: $crate::Machine>($mach: $MTy, $($arg: $argty),*) -> $ret $body |
| 838 | fn_impl($mach, $($arg),*) |
| 839 | } |
| 840 | }; |
| 841 | ($mach:ident, $MTy:ident, { $([$pub:tt $(($krate:tt))*])* fn $name:ident($($arg:ident: $argty:ty),* $(,)*) $body:block }) => { |
| 842 | dispatch!($mach, $MTy, { |
| 843 | $([$pub $(($krate))*])* fn $name($($arg: $argty),*) -> () $body |
| 844 | }); |
| 845 | } |
| 846 | } |
| 847 | |
| 848 | #[cfg (test)] |
| 849 | mod test { |
| 850 | use super::*; |
| 851 | |
| 852 | #[test] |
| 853 | fn test_bswap32() { |
| 854 | let xs = [0x0f0e_0d0c, 0x0b0a_0908, 0x0706_0504, 0x0302_0100]; |
| 855 | let ys = [0x0c0d_0e0f, 0x0809_0a0b, 0x0405_0607, 0x0001_0203]; |
| 856 | |
| 857 | let m = unsafe { GenericMachine::instance() }; |
| 858 | |
| 859 | let x: <GenericMachine as Machine>::u32x4 = m.vec(xs); |
| 860 | let x = x.bswap(); |
| 861 | |
| 862 | let y = m.vec(ys); |
| 863 | assert_eq!(x, y); |
| 864 | } |
| 865 | } |
| 866 | |