1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398 399 400 401 402 403 404 405 406 407 408 409 410 411 412 413 414 415 416 417 418 419 420 421 422 423 424 425 426 427 428 429 430 431 432 433 434 435 436 437 438 439 440 441 442 443 444 445 446 447 448 449 450 451 452 453 454 455 456 457 458 459 460 461 462 463 464 465 466 467 468 469 470 471 472 473 474 475 476 477 478 479 480 481 482 483 484 485 486 487 488 489 490 491 492 493 494 495 496 497 498 499 500 501 502 503 504 505 506 507 508 509 510 511 512 513 514 515 516 517 518 519 520 521 522 523 524 525 526 527 528 529 530 531 532 533 534 535 536 537 538 539 540 541 542 543 544 545 546 547 548 549 550 551 552 553 554 555 556 557 558 559 560 561 562 563 564 565 566 567 568 569 570 571 572 573 574 575 576 577 578 579 580 581 582 583 584 585 586 587 588 589 590 591
//! This module provides many Hamming distance routines. //! //! These distance functions share the same efficient underlying SIMD-accelerated implementation: //! * `hamming` //! * `hamming_simd_parallel` //! //! These search functions share the same efficient underlying SIMD-accelerated implementation: //! * `hamming_search` //! * `hamming_search_simd` //! * `hamming_search_simd_with_opts` use std::*; use super::*; use super::jewel::*; /// Returns the hamming distance between two strings by naively counting mismatches. /// /// The length of `a` and `b` must be the same. /// /// # Arguments /// * `a` - first string (slice) /// * `b` - second string (slice) /// /// # Panics /// * If the length of `a` does not equal the length of `b`. /// /// # Example /// ``` /// # use triple_accel::*; /// # use triple_accel::hamming::*; /// let dist = hamming_naive(b"abc", b"abd"); /// /// assert!(dist == 1); /// ``` pub fn hamming_naive(a: &[u8], b: &[u8]) -> u32 { let len = a.len(); assert!(len == b.len()); let mut res = 0u32; for i in 0..len { res += (a[i] != b[i]) as u32; } res } /// Returns an iterator over best `Match`s by naively searching through the text `haystack` /// for the pattern `needle`. /// /// This is done by naively counting mismatches at every position in `haystack`. /// Only the matches with the lowest Hamming distance are returned. /// Each returned `Match` requires at least half or more bytes of the `needle` to match /// somewhere in the `haystack`. /// The length of `needle` must be less than or equal to the length of `haystack`. /// /// # Arguments /// * `needle` - pattern string (slice) /// * `haystack` - text string (slice) /// /// # Example /// ``` /// # use triple_accel::*; /// # use triple_accel::hamming::*; /// let matches: Vec<Match> = hamming_search_naive(b"abc", b" abd").collect(); /// /// assert!(matches == vec![Match{start: 2, end: 5, k: 1}]); /// ``` pub fn hamming_search_naive<'a>(needle: &'a [u8], haystack: &'a [u8]) -> Box<dyn Iterator<Item = Match> + 'a> { hamming_search_naive_with_opts(needle, haystack, ((needle.len() as u32) >> 1) + ((needle.len() as u32) & 1), SearchType::Best) } /// Returns an iterator over `Match`s by naively searching through the text `haystack` /// for the pattern `needle`, with extra options. /// /// Only matches with less than `k` mismatches are returned. /// This is done by naively counting mismatches at every position in `haystack`. /// The length of `needle` must be less than or equal to the length of `haystack`. /// /// # Arguments /// * `needle` - pattern string (slice) /// * `haystack` - text string (slice) /// * `k` - number of mismatches allowed /// * `search_type` - whether to only return the "best" matches with the lowest Hamming distance, or /// all matches /// /// # Example /// ``` /// # use triple_accel::*; /// # use triple_accel::hamming::*; /// let matches: Vec<Match> = hamming_search_naive_with_opts(b"abc", b" abd", 1, SearchType::All).collect(); /// /// assert!(matches == vec![Match{start: 2, end: 5, k: 1}]); /// ``` pub fn hamming_search_naive_with_opts<'a>(needle: &'a [u8], haystack: &'a [u8], k: u32, search_type: SearchType) -> Box<dyn Iterator<Item = Match> + 'a> { let needle_len = needle.len(); let haystack_len = haystack.len(); if needle_len > haystack_len { return Box::new(iter::empty()); } let len = haystack_len + 1 - needle_len; let mut curr_k = k; let mut i = 0; let res = iter::from_fn(move || { 'outer: while i < len { let mut final_res = 0u32; for j in 0..needle_len { final_res += (needle[j] != haystack[i + j]) as u32; // early stop if final_res > curr_k { i += 1; continue 'outer; } } match search_type { SearchType::Best => curr_k = final_res, _ => () } i += 1; return Some((Match{start: i - 1, end: i + needle_len - 1, k: final_res}, curr_k)); } None }); if search_type == SearchType::Best { let mut res_vec = Vec::with_capacity(haystack_len / needle_len); res.for_each(|m| { res_vec.push(m.0); curr_k = m.1; }); return Box::new(res_vec.into_iter().filter(move |m| m.k == curr_k)); } Box::new(res.map(|m| m.0)) } /// Returns the hamming distance between two strings by efficiently counting mismatches in chunks of 64 bits. /// /// The length of `a` and `b` must be the same. /// Both `a` and `b` must be aligned and padded so they can be directly casted to chunks of `u64`. /// Use `alloc_str` to create aligned and padded strings. /// This should be faster than `hamming_naive` and maybe even `hamming_words_128`. This should be slower /// than `hamming_simd_parallel/movemask`. /// /// # Arguments /// * `a` - first string (slice) /// * `b` - second string (slice) /// /// # Panics /// * If the length of `a` does not equal the length of `b`. /// /// # Example /// ``` /// # use triple_accel::*; /// # use triple_accel::hamming::*; /// let mut a = alloc_str(3); /// let mut b = alloc_str(3); /// fill_str(&mut a, b"abc"); /// fill_str(&mut b, b"abd"); /// /// let dist = hamming_words_64(&a, &b); /// /// assert!(dist == 1); /// ``` pub fn hamming_words_64(a: &[u8], b: &[u8]) -> u32 { assert!(a.len() == b.len()); unsafe { let mut res = 0u32; // the pointer address better be aligned for u64 // may not be in little endian let a_ptr = a.as_ptr() as *const u64; let b_ptr = b.as_ptr() as *const u64; let words_len = (a.len() >> 3) as isize; for i in 0..words_len { // change to little endian omitted because it is not necessary in this case let mut r = (*a_ptr.offset(i)) ^ (*b_ptr.offset(i)); // reduce or by "folding" one half of each byte onto the other multiple times r |= r >> 4; // ...00001111 r &= 0x0f0f0f0f0f0f0f0fu64; r |= r >> 2; // ...00110011 r &= 0x3333333333333333u64; r |= r >> 1; // ...01010101 r &= 0x5555555555555555u64; res += r.count_ones(); } let words_rem = a.len() & 7; if words_rem > 0 { let mut r = (*a_ptr.offset(words_len)) ^ (*b_ptr.offset(words_len)); r |= r >> 4; r &= 0x0f0f0f0f0f0f0f0fu64; r |= r >> 2; r &= 0x3333333333333333u64; r |= r >> 1; r &= 0x5555555555555555u64; // make sure to mask out bits outside the string lengths res += (r & ((1u64 << ((words_rem as u64) << 3u64)) - 1u64)).count_ones(); } res } } /// Returns the hamming distance between two strings by counting mismatches in chunks of 128 bits. /// /// The length of `a` and `b` must be the same. /// Both `a` and `b` must be aligned and padded so they can be directly casted to chunks of `u128`. /// Use `alloc_str` to create aligned and padded strings. /// This may be slower than `hamming_words_64` in practice, probably since Rust `u128` is not as /// optimized. This should be slower than `hamming_simd_parallel/movemask`. /// /// # Arguments /// * `a` - first string (slice) /// * `b` - second string (slice) /// /// # Panics /// * If the length of `a` does not equal the length of `b`. /// /// # Example /// ``` /// # use triple_accel::*; /// # use triple_accel::hamming::*; /// let mut a = alloc_str(3); /// let mut b = alloc_str(3); /// fill_str(&mut a, b"abc"); /// fill_str(&mut b, b"abd"); /// /// let dist = hamming_words_128(&a, &b); /// /// assert!(dist == 1); /// ``` pub fn hamming_words_128(a: &[u8], b: &[u8]) -> u32 { assert!(a.len() == b.len()); unsafe { let mut res = 0u32; // the pointer address better be aligned for u128 // may not be in little endian let a_ptr = a.as_ptr() as *const u128; let b_ptr = b.as_ptr() as *const u128; let words_len = (a.len() >> 4) as isize; for i in 0..words_len { // change to little endian omitted because it is not necessary in this case let mut r = (*a_ptr.offset(i)) ^ (*b_ptr.offset(i)); // reduce or by "folding" one half of each byte onto the other multiple times r |= r >> 4; // ...00001111 r &= 0x0f0f0f0f0f0f0f0f0f0f0f0f0f0f0f0fu128; r |= r >> 2; // ...00110011 r &= 0x33333333333333333333333333333333u128; r |= r >> 1; // ...01010101 r &= 0x55555555555555555555555555555555u128; res += r.count_ones(); } let words_rem = a.len() & 15; if words_rem > 0 { let mut r = (*a_ptr.offset(words_len)) ^ (*b_ptr.offset(words_len)); r |= r >> 4; r &= 0x0f0f0f0f0f0f0f0f0f0f0f0f0f0f0f0fu128; r |= r >> 2; r &= 0x33333333333333333333333333333333u128; r |= r >> 1; r &= 0x55555555555555555555555555555555u128; // make sure to mask out bits outside the string lengths res += (r & ((1u128 << ((words_rem as u128) << 3u128)) - 1u128)).count_ones(); } res } } /// Returns the hamming distance between two strings by counting mismatches using SIMD vectors to /// increment multiple counters in parallel. /// /// The length of `a` and `b` must be the same. /// There are no constraints on how `a` and `b` are aligned and padded. /// This will automatically fall back to `hamming_naive`, if AVX2 and SSE4.1 are not supported. /// This should be faster than both `hamming_word_64/128` and `hamming_simd_movemask`. /// /// # Arguments /// * `a` - first string (slice) /// * `b` - second string (slice) /// /// # Panics /// * If the length of `a` does not equal the length of `b`. /// /// # Example /// ``` /// # use triple_accel::*; /// # use triple_accel::hamming::*; /// let dist = hamming_simd_parallel(b"abc", b"abd"); /// /// assert!(dist == 1); /// ``` pub fn hamming_simd_parallel(a: &[u8], b: &[u8]) -> u32 { assert!(a.len() == b.len()); #[cfg(any(target_arch = "x86", target_arch = "x86_64"))] { if cfg!(feature = "jewel-avx") && is_x86_feature_detected!("avx2") { return unsafe {Avx::count_mismatches(a.as_ptr(), b.as_ptr(), a.len())}; }else if cfg!(feature = "jewel-sse") && is_x86_feature_detected!("sse4.1") { return unsafe {Sse::count_mismatches(a.as_ptr(), b.as_ptr(), a.len())}; } } hamming_naive(a, b) } /// Returns the hamming distance between two strings by counting mismatches using the SIMD movemask intrinsic. /// /// The length of `a` and `b` must be the same. /// There are no constraints on how `a` and `b` are aligned and padded. /// This will automatically fall back to `hamming_naive`, if AVX2 and SSE4.1 are not supported. /// This should be faster than `hamming_word_64/128`, but slower than `hamming_simd_parallel`. /// /// # Arguments /// * `a` - first string (slice) /// * `b` - second string (slice) /// /// # Panics /// * If the length of `a` does not equal the length of `b`. /// /// # Example /// ``` /// # use triple_accel::*; /// # use triple_accel::hamming::*; /// let dist = hamming_simd_movemask(b"abc", b"abd"); /// /// assert!(dist == 1); /// ``` pub fn hamming_simd_movemask(a: &[u8], b: &[u8]) -> u32 { assert!(a.len() == b.len()); #[cfg(any(target_arch = "x86", target_arch = "x86_64"))] { if cfg!(feature = "jewel-avx") && is_x86_feature_detected!("avx2") { return unsafe {Avx::mm_count_mismatches(a.as_ptr(), b.as_ptr(), a.len())}; }else if cfg!(feature = "jewel-sse") && is_x86_feature_detected!("sse4.1") { return unsafe {Sse::mm_count_mismatches(a.as_ptr(), b.as_ptr(), a.len())}; } } hamming_naive(a, b) } /// Returns the hamming distance between two strings using the best method. /// /// The length of `a` and `b` must be the same. /// This will automatically fall back to a scalar alternative if AVX2 and /// SSE4.1 are not supported. /// Internally, this calls `hamming_simd_parallel`. /// /// # Arguments /// * `a` - first string (slice) /// * `b` - second string (slice) /// /// # Panics /// * If the length of `a` does not equal the length of `b`. /// /// # Example /// ``` /// # use triple_accel::*; /// let dist = hamming(b"abc", b"abd"); /// /// assert!(dist == 1); /// ``` pub fn hamming(a: &[u8], b: &[u8]) -> u32 { hamming_simd_parallel(a, b) } /// Returns an iterator over best `Match`s by searching through the text `haystack` /// for the pattern `needle` using SIMD. /// /// This is done by counting mismatches at every position in `haystack`. /// This will automatically fall back to `hamming_search_naive_with_opts` if AVX2 and SSE4.1 /// are not supported. /// Null bytes/characters are not supported. /// The length of `needle` must be less than or equal to the length of `haystack`. /// Each returned `Match` requires at least half or more bytes of the `needle` to match /// somwhere in the `haystack`. /// Only the matches with the lowest Hamming distance are returned. /// This should be faster than `hamming_search_naive`. /// /// # Arguments /// * `needle` - pattern string (slice) /// * `haystack` - text string (slice) /// /// # Panics /// * When there are zero/null bytes in the `haystack` string. /// /// # Example /// ``` /// # use triple_accel::*; /// # use triple_accel::hamming::*; /// let matches: Vec<Match> = hamming_search_simd(b"abc", b" abd").collect(); /// /// assert!(matches == vec![Match{start: 2, end: 5, k: 1}]); /// ``` pub fn hamming_search_simd<'a>(needle: &'a [u8], haystack: &'a [u8]) -> Box<dyn Iterator<Item = Match> + 'a> { hamming_search_simd_with_opts(needle, haystack, ((needle.len() as u32) >> 1) + ((needle.len() as u32) & 1), SearchType::Best) } /// Returns an iterator over `Match`s by searching through the text `haystack` for the /// pattern `needle` using SIMD, with extra options. /// /// This is done by using SIMD to count mismatches at every position in `haystack`. /// This will automatically fall back to `hamming_search_naive_with_opts` if AVX2 and SSE4.1 /// are not supported. /// Null bytes/characters are not supported. /// The length of `needle` must be less than or equal to the length of `haystack`. /// This should be faster than `hamming_search_naive_with_opts`. /// /// # Arguments /// * `needle` - pattern string (slice) /// * `haystack` - text string (slice) /// * `k` - number of mismatches allowed /// * `search_type` - whether to only return the "best" matches with the lowest Hamming distance, or /// all matches /// /// # Panics /// * When there are zero/null bytes in the `haystack` string. /// /// # Example /// ``` /// # use triple_accel::*; /// # use triple_accel::hamming::*; /// let matches: Vec<Match> = hamming_search_simd_with_opts(b"abc", b" abd", 1, SearchType::All).collect(); /// /// assert!(matches == vec![Match{start: 2, end: 5, k: 1}]); /// ``` pub fn hamming_search_simd_with_opts<'a>(needle: &'a [u8], haystack: &'a [u8], k: u32, search_type: SearchType) -> Box<dyn Iterator<Item = Match> + 'a> { if needle.len() > haystack.len() { return Box::new(iter::empty()); } if needle.len() == 0 { return Box::new(iter::empty()); } check_no_null_bytes(haystack); #[cfg(any(target_arch = "x86", target_arch = "x86_64"))] { if cfg!(feature = "jewel-avx") && is_x86_feature_detected!("avx2") { return unsafe {hamming_search_simd_core_avx(needle, haystack, k, search_type)}; }else if cfg!(feature = "jewel-sse") && is_x86_feature_detected!("sse4.1") { return unsafe {hamming_search_simd_core_sse(needle, haystack, k, search_type)}; } } hamming_search_naive_with_opts(needle, haystack, k, search_type) } macro_rules! create_hamming_search_simd_core { ($name:ident, $jewel:ty, $target:literal) => { #[cfg(any(target_arch = "x86", target_arch = "x86_64"))] #[target_feature(enable = $target)] unsafe fn $name<'a>(needle: &'a [u8], haystack: &'a [u8], k: u32, search_type: SearchType) -> Box<dyn Iterator<Item = Match> + 'a> { #[cfg(feature = "debug")] { println!("Debug: Hamming search Jewel vector type {} for target {}.", stringify!($jewel), stringify!($target)); } let needle_len = needle.len(); let haystack_len = haystack.len(); let needle_vector = <$jewel>::loadu(needle.as_ptr(), needle_len); // calculate len using the unused bytes in the needle Jewel vector, for speed // there may be leftover positions in haystack that need to be calculated using a // scalar search afterwards // there should be no null bytes in the strings let len = if needle_vector.upper_bound() > haystack_len {0} else {haystack_len + 1 - needle_vector.upper_bound()}; let real_len = haystack_len + 1 - needle_len; let haystack_ptr = haystack.as_ptr(); let mut curr_k = k; let mut i = 0; let res = iter::from_fn(move || { while i < len { let final_res = <$jewel>::vector_count_mismatches(&needle_vector, haystack_ptr.offset(i as isize), needle_len); i += 1; if final_res <= curr_k { match search_type { SearchType::Best => curr_k = final_res, _ => () } return Some((Match{start: i - 1, end: i + needle_len - 1, k: final_res}, curr_k)); } } // scalar search 'outer: while i < real_len { let mut final_res = 0u32; for j in 0..needle_len { final_res += (*needle.get_unchecked(j) != *haystack.get_unchecked(i + j)) as u32; if final_res > curr_k { i += 1; continue 'outer; } } match search_type { SearchType::Best => curr_k = final_res, _ => () } i += 1; return Some((Match{start: i - 1, end: i + needle_len - 1, k: final_res}, curr_k)); } None }); if search_type == SearchType::Best { let mut res_vec = Vec::with_capacity(haystack_len / needle_len); res.for_each(|m| { res_vec.push(m.0); curr_k = m.1; }); return Box::new(res_vec.into_iter().filter(move |m| m.k == curr_k)); } Box::new(res.map(|m| m.0)) } }; } // generate different versions for different intrinsics #[cfg(any(target_arch = "x86", target_arch = "x86_64"))] create_hamming_search_simd_core!(hamming_search_simd_core_avx, Avx, "avx2"); #[cfg(any(target_arch = "x86", target_arch = "x86_64"))] create_hamming_search_simd_core!(hamming_search_simd_core_sse, Sse, "sse4.1"); /// Returns an iterator over best `Match`s by searching through the text `haystack` /// for the pattern `needle` using SIMD. /// /// This will automatically fall back to a scalar alternative if AVX2 and SSE4.1 /// are not supported. /// Null bytes/characters are not supported. /// The length of `needle` must be less than or equal to the length of `haystack`. /// Each returned `Match` requires at least half or more bytes of the `needle` to match /// somewhere in the `haystack`. /// Only the matches with the lowest Hamming distance are returned. /// Internally, this calls `hamming_search_simd`. /// /// # Arguments /// * `needle` - pattern string (slice) /// * `haystack` - text string (slice) /// /// # Panics /// * When there are zero/null bytes in the `haystack` string. /// /// # Example /// ``` /// # use triple_accel::*; /// let matches: Vec<Match> = hamming_search(b"abc", b" abd").collect(); /// /// assert!(matches == vec![Match{start: 2, end: 5, k: 1}]); /// ``` pub fn hamming_search<'a>(needle: &'a [u8], haystack: &'a [u8]) -> Box<dyn Iterator<Item = Match> + 'a> { hamming_search_simd(needle, haystack) }