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 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398 399 400 401 402 403 404 405 406 407 408 409 410 411 412 413 414 415 416 417 418 419 420 421 422 423 424 425 426 427 428 429 430 431 432 433 434 435 436 437 438 439 440 441 442 443 444 445 446 447 448 449 450 451 452 453 454 455 456 457 458 459 460 461 462 463 464 465 466 467 468 469 470 471 472 473 474 475 476 477 478 479 480 481 482 483 484 485 486 487 488 489 490 491 492 493 494 495 496 497 498 499 500 501 502 503 504 505 506 507 508 509 510 511 512 513 514 515 516 517 518 519 520 521 522 523 524 525 526 527 528 529 530 531 532 533 534 535 536 537 538 539 540 541 542 543 544 545 546 547 548 549 550 551 552 553 554 555 556 557 558 559 560 561 562 563 564 565 566 567 568 569 570 571 572 573 574 575 576 577 578 579 580 581 582 583 584 585 586 587 588 589 590 591 592 593 594 595 596 597 598 599 600 601 602 603 604 605 606 607 608 609 610 611 612 613 614 615 616 617 618 619 620 621 622 623 624 625 626 627 628 629 630 631 632 633 634 635 636 637 638 639 640 641 642 643 644 645 646 647 648 649 650 651 652 653 654 655 656 657 658 659 660 661 662 663 664 665 666 667 668 669 670 671 672 673 674 675 676 677 678 679 680 681 682 683 684 685 686 687 688 689 690 691 692 693 694 695 696 697 698 699 700 701 702 703 704 705 706 707 708 709 710 711 712 713 714 715 716 717 718 719 720 721 722 723 724 725 726 727 728 729 730 731 732 733 734 735 736 737 738 739 740 741 742 743 744 745 746 747 748 749 750 751 752 753 754 755 756 757 758 759 760 761 762 763 764 765 766 767 768 769 770 771 772 773 774 775 776 777 778 779 780 781 782 783 784 785 786 787 788 789 790 791 792 793 794 795 796 797 798 799 800 801 802 803 804 805 806 807 808 809 810 811 812 813 814 815 816 817 818 819 820 821 822 823 824 825 826 827 828 829 830 831 832 833 834 835 836 837 838 839 840 841 842 843 844 845 846 847 848 849 850 851 852 853 854 855 856 857 858 859 860 861 862 863 864 865 866 867 868 869 870 871 872 873 874 875 876 877 878 879 880 881 882 883 884 885 886 887 888 889 890 891 892 893 894 895 896 897 898 899 900 901 902 903 904 905 906 907 908 909 910 911 912 913 914 915 916 917 918 919 920 921 922 923 924 925 926 927 928 929 930 931 932 933 934 935 936 937 938 939 940 941 942 943 944 945 946 947 948 949 950 951 952 953 954 955 956 957 958 959 960 961 962 963 964 965 966 967 968 969 970 971 972 973 974 975 976 977 978 979 980 981 982 983 984 985 986 987 988 989 990 991 992 993 994 995 996 997 998 999 1000 1001 1002 1003 1004 1005 1006 1007 1008 1009 1010 1011 1012 1013 1014 1015 1016 1017 1018 1019 1020 1021 1022 1023 1024 1025 1026 1027 1028 1029 1030 1031 1032 1033 1034 1035 1036 1037 1038 1039 1040 1041 1042 1043 1044 1045 1046 1047 1048 1049 1050 1051 1052 1053 1054 1055 1056 1057 1058 1059 1060 1061 1062 1063 1064 1065 1066 1067 1068 1069 1070 1071 1072 1073 1074 1075 1076 1077 1078 1079 1080 1081 1082 1083 1084 1085 1086 1087 1088 1089 1090 1091 1092 1093 1094 1095 1096 1097 1098 1099 1100 1101 1102 1103 1104 1105 1106 1107 1108 1109 1110 1111 1112 1113 1114 1115 1116 1117 1118 1119 1120 1121 1122 1123 1124 1125 1126 1127 1128 1129 1130 1131 1132 1133 1134 1135 1136 1137 1138 1139 1140 1141 1142 1143 1144 1145 1146 1147 1148 1149 1150 1151 1152 1153 1154 1155 1156 1157 1158 1159 1160 1161 1162 1163 1164 1165 1166 1167 1168 1169 1170 1171 1172 1173 1174 1175 1176 1177 1178 1179 1180 1181 1182 1183 1184 1185 1186 1187 1188 1189 1190 1191 1192 1193 1194 1195 1196 1197 1198 1199 1200 1201 1202 1203 1204 1205 1206 1207 1208 1209 1210 1211 1212 1213 1214 1215 1216 1217 1218 1219 1220 1221 1222 1223 1224 1225 1226 1227 1228 1229 1230 1231 1232 1233 1234 1235 1236 1237 1238 1239 1240 1241 1242 1243 1244 1245 1246 1247 1248 1249 1250 1251 1252 1253 1254 1255 1256 1257 1258 1259 1260 1261 1262 1263 1264 1265 1266 1267 1268 1269 1270 1271 1272 1273 1274 1275 1276 1277 1278 1279 1280 1281 1282 1283 1284 1285 1286 1287 1288 1289 1290 1291 1292 1293 1294 1295 1296 1297 1298 1299 1300 1301 1302 1303 1304 1305 1306 1307 1308 1309 1310 1311 1312 1313 1314 1315 1316 1317 1318 1319 1320 1321 1322 1323 1324 1325 1326 1327 1328 1329 1330 1331 1332 1333 1334 1335 1336 1337 1338 1339 1340 1341 1342 1343 1344 1345 1346 1347 1348 1349 1350 1351 1352 1353 1354 1355 1356 1357 1358 1359 1360 1361 1362 1363 1364 1365 1366 1367 1368 1369 1370 1371 1372 1373 1374 1375 1376 1377 1378 1379 1380 1381 1382
use core::fmt::Debug;
use alloc::{
boxed::Box, collections::BTreeMap, format, sync::Arc, vec, vec::Vec,
};
use crate::{
packed::{
ext::Pointer,
pattern::Patterns,
vector::{FatVector, Vector},
},
util::int::U32,
PatternID,
};
/// A match type specialized to the Teddy implementations below.
///
/// Essentially, instead of representing a match at byte offsets, we use
/// raw pointers. This is because the implementations below operate on raw
/// pointers, and so this is a more natural return type based on how the
/// implementation works.
///
/// Also, the `PatternID` used here is a `u16`.
#[derive(Clone, Copy, Debug)]
pub(crate) struct Match {
pid: PatternID,
start: *const u8,
end: *const u8,
}
impl Match {
/// Returns the ID of the pattern that matched.
pub(crate) fn pattern(&self) -> PatternID {
self.pid
}
/// Returns a pointer into the haystack at which the match starts.
pub(crate) fn start(&self) -> *const u8 {
self.start
}
/// Returns a pointer into the haystack at which the match ends.
pub(crate) fn end(&self) -> *const u8 {
self.end
}
}
/// A "slim" Teddy implementation that is generic over both the vector type
/// and the minimum length of the patterns being searched for.
///
/// Only 1, 2, 3 and 4 bytes are supported as minimum lengths.
#[derive(Clone, Debug)]
pub(crate) struct Slim<V, const BYTES: usize> {
/// A generic data structure for doing "slim" Teddy verification.
teddy: Teddy<8>,
/// The masks used as inputs to the shuffle operation to generate
/// candidates (which are fed into the verification routines).
masks: [Mask<V>; BYTES],
}
impl<V: Vector, const BYTES: usize> Slim<V, BYTES> {
/// Create a new "slim" Teddy searcher for the given patterns.
///
/// # Panics
///
/// This panics when `BYTES` is any value other than 1, 2, 3 or 4.
///
/// # Safety
///
/// Callers must ensure that this is okay to call in the current target for
/// the current CPU.
#[inline(always)]
pub(crate) unsafe fn new(patterns: Arc<Patterns>) -> Slim<V, BYTES> {
assert!(
1 <= BYTES && BYTES <= 4,
"only 1, 2, 3 or 4 bytes are supported"
);
let teddy = Teddy::new(patterns);
let masks = SlimMaskBuilder::from_teddy(&teddy);
Slim { teddy, masks }
}
/// Returns the approximate total amount of heap used by this type, in
/// units of bytes.
#[inline(always)]
pub(crate) fn memory_usage(&self) -> usize {
self.teddy.memory_usage()
}
/// Returns the minimum length, in bytes, that a haystack must be in order
/// to use it with this searcher.
#[inline(always)]
pub(crate) fn minimum_len(&self) -> usize {
V::BYTES + (BYTES - 1)
}
}
impl<V: Vector> Slim<V, 1> {
/// Look for an occurrences of the patterns in this finder in the haystack
/// given by the `start` and `end` pointers.
///
/// If no match could be found, then `None` is returned.
///
/// # Safety
///
/// The given pointers representing the haystack must be valid to read
/// from. They must also point to a region of memory that is at least the
/// minimum length required by this searcher.
///
/// Callers must ensure that this is okay to call in the current target for
/// the current CPU.
#[inline(always)]
pub(crate) unsafe fn find(
&self,
start: *const u8,
end: *const u8,
) -> Option<Match> {
let len = end.distance(start);
debug_assert!(len >= self.minimum_len());
let mut cur = start;
while cur <= end.sub(V::BYTES) {
if let Some(m) = self.find_one(cur, end) {
return Some(m);
}
cur = cur.add(V::BYTES);
}
if cur < end {
cur = end.sub(V::BYTES);
if let Some(m) = self.find_one(cur, end) {
return Some(m);
}
}
None
}
/// Look for a match starting at the `V::BYTES` at and after `cur`. If
/// there isn't one, then `None` is returned.
///
/// # Safety
///
/// The given pointers representing the haystack must be valid to read
/// from. They must also point to a region of memory that is at least the
/// minimum length required by this searcher.
///
/// Callers must ensure that this is okay to call in the current target for
/// the current CPU.
#[inline(always)]
unsafe fn find_one(
&self,
cur: *const u8,
end: *const u8,
) -> Option<Match> {
let c = self.candidate(cur);
if !c.is_zero() {
if let Some(m) = self.teddy.verify(cur, end, c) {
return Some(m);
}
}
None
}
/// Look for a candidate match (represented as a vector) starting at the
/// `V::BYTES` at and after `cur`. If there isn't one, then a vector with
/// all bits set to zero is returned.
///
/// # Safety
///
/// The given pointer representing the haystack must be valid to read
/// from.
///
/// Callers must ensure that this is okay to call in the current target for
/// the current CPU.
#[inline(always)]
unsafe fn candidate(&self, cur: *const u8) -> V {
let chunk = V::load_unaligned(cur);
Mask::members1(chunk, self.masks)
}
}
impl<V: Vector> Slim<V, 2> {
/// See Slim<V, 1>::find.
#[inline(always)]
pub(crate) unsafe fn find(
&self,
start: *const u8,
end: *const u8,
) -> Option<Match> {
let len = end.distance(start);
debug_assert!(len >= self.minimum_len());
let mut cur = start.add(1);
let mut prev0 = V::splat(0xFF);
while cur <= end.sub(V::BYTES) {
if let Some(m) = self.find_one(cur, end, &mut prev0) {
return Some(m);
}
cur = cur.add(V::BYTES);
}
if cur < end {
cur = end.sub(V::BYTES);
prev0 = V::splat(0xFF);
if let Some(m) = self.find_one(cur, end, &mut prev0) {
return Some(m);
}
}
None
}
/// See Slim<V, 1>::find_one.
#[inline(always)]
unsafe fn find_one(
&self,
cur: *const u8,
end: *const u8,
prev0: &mut V,
) -> Option<Match> {
let c = self.candidate(cur, prev0);
if !c.is_zero() {
if let Some(m) = self.teddy.verify(cur.sub(1), end, c) {
return Some(m);
}
}
None
}
/// See Slim<V, 1>::candidate.
#[inline(always)]
unsafe fn candidate(&self, cur: *const u8, prev0: &mut V) -> V {
let chunk = V::load_unaligned(cur);
let (res0, res1) = Mask::members2(chunk, self.masks);
let res0prev0 = res0.shift_in_one_byte(*prev0);
let res = res0prev0.and(res1);
*prev0 = res0;
res
}
}
impl<V: Vector> Slim<V, 3> {
/// See Slim<V, 1>::find.
#[inline(always)]
pub(crate) unsafe fn find(
&self,
start: *const u8,
end: *const u8,
) -> Option<Match> {
let len = end.distance(start);
debug_assert!(len >= self.minimum_len());
let mut cur = start.add(2);
let mut prev0 = V::splat(0xFF);
let mut prev1 = V::splat(0xFF);
while cur <= end.sub(V::BYTES) {
if let Some(m) = self.find_one(cur, end, &mut prev0, &mut prev1) {
return Some(m);
}
cur = cur.add(V::BYTES);
}
if cur < end {
cur = end.sub(V::BYTES);
prev0 = V::splat(0xFF);
prev1 = V::splat(0xFF);
if let Some(m) = self.find_one(cur, end, &mut prev0, &mut prev1) {
return Some(m);
}
}
None
}
/// See Slim<V, 1>::find_one.
#[inline(always)]
unsafe fn find_one(
&self,
cur: *const u8,
end: *const u8,
prev0: &mut V,
prev1: &mut V,
) -> Option<Match> {
let c = self.candidate(cur, prev0, prev1);
if !c.is_zero() {
if let Some(m) = self.teddy.verify(cur.sub(2), end, c) {
return Some(m);
}
}
None
}
/// See Slim<V, 1>::candidate.
#[inline(always)]
unsafe fn candidate(
&self,
cur: *const u8,
prev0: &mut V,
prev1: &mut V,
) -> V {
let chunk = V::load_unaligned(cur);
let (res0, res1, res2) = Mask::members3(chunk, self.masks);
let res0prev0 = res0.shift_in_two_bytes(*prev0);
let res1prev1 = res1.shift_in_one_byte(*prev1);
let res = res0prev0.and(res1prev1).and(res2);
*prev0 = res0;
*prev1 = res1;
res
}
}
impl<V: Vector> Slim<V, 4> {
/// See Slim<V, 1>::find.
#[inline(always)]
pub(crate) unsafe fn find(
&self,
start: *const u8,
end: *const u8,
) -> Option<Match> {
let len = end.distance(start);
debug_assert!(len >= self.minimum_len());
let mut cur = start.add(3);
let mut prev0 = V::splat(0xFF);
let mut prev1 = V::splat(0xFF);
let mut prev2 = V::splat(0xFF);
while cur <= end.sub(V::BYTES) {
if let Some(m) =
self.find_one(cur, end, &mut prev0, &mut prev1, &mut prev2)
{
return Some(m);
}
cur = cur.add(V::BYTES);
}
if cur < end {
cur = end.sub(V::BYTES);
prev0 = V::splat(0xFF);
prev1 = V::splat(0xFF);
prev2 = V::splat(0xFF);
if let Some(m) =
self.find_one(cur, end, &mut prev0, &mut prev1, &mut prev2)
{
return Some(m);
}
}
None
}
/// See Slim<V, 1>::find_one.
#[inline(always)]
unsafe fn find_one(
&self,
cur: *const u8,
end: *const u8,
prev0: &mut V,
prev1: &mut V,
prev2: &mut V,
) -> Option<Match> {
let c = self.candidate(cur, prev0, prev1, prev2);
if !c.is_zero() {
if let Some(m) = self.teddy.verify(cur.sub(3), end, c) {
return Some(m);
}
}
None
}
/// See Slim<V, 1>::candidate.
#[inline(always)]
unsafe fn candidate(
&self,
cur: *const u8,
prev0: &mut V,
prev1: &mut V,
prev2: &mut V,
) -> V {
let chunk = V::load_unaligned(cur);
let (res0, res1, res2, res3) = Mask::members4(chunk, self.masks);
let res0prev0 = res0.shift_in_three_bytes(*prev0);
let res1prev1 = res1.shift_in_two_bytes(*prev1);
let res2prev2 = res2.shift_in_one_byte(*prev2);
let res = res0prev0.and(res1prev1).and(res2prev2).and(res3);
*prev0 = res0;
*prev1 = res1;
*prev2 = res2;
res
}
}
/// A "fat" Teddy implementation that is generic over both the vector type
/// and the minimum length of the patterns being searched for.
///
/// Only 1, 2, 3 and 4 bytes are supported as minimum lengths.
#[derive(Clone, Debug)]
pub(crate) struct Fat<V, const BYTES: usize> {
/// A generic data structure for doing "fat" Teddy verification.
teddy: Teddy<16>,
/// The masks used as inputs to the shuffle operation to generate
/// candidates (which are fed into the verification routines).
masks: [Mask<V>; BYTES],
}
impl<V: FatVector, const BYTES: usize> Fat<V, BYTES> {
/// Create a new "fat" Teddy searcher for the given patterns.
///
/// # Panics
///
/// This panics when `BYTES` is any value other than 1, 2, 3 or 4.
///
/// # Safety
///
/// Callers must ensure that this is okay to call in the current target for
/// the current CPU.
#[inline(always)]
pub(crate) unsafe fn new(patterns: Arc<Patterns>) -> Fat<V, BYTES> {
assert!(
1 <= BYTES && BYTES <= 4,
"only 1, 2, 3 or 4 bytes are supported"
);
let teddy = Teddy::new(patterns);
let masks = FatMaskBuilder::from_teddy(&teddy);
Fat { teddy, masks }
}
/// Returns the approximate total amount of heap used by this type, in
/// units of bytes.
#[inline(always)]
pub(crate) fn memory_usage(&self) -> usize {
self.teddy.memory_usage()
}
/// Returns the minimum length, in bytes, that a haystack must be in order
/// to use it with this searcher.
#[inline(always)]
pub(crate) fn minimum_len(&self) -> usize {
V::Half::BYTES + (BYTES - 1)
}
}
impl<V: FatVector> Fat<V, 1> {
/// Look for an occurrences of the patterns in this finder in the haystack
/// given by the `start` and `end` pointers.
///
/// If no match could be found, then `None` is returned.
///
/// # Safety
///
/// The given pointers representing the haystack must be valid to read
/// from. They must also point to a region of memory that is at least the
/// minimum length required by this searcher.
///
/// Callers must ensure that this is okay to call in the current target for
/// the current CPU.
#[inline(always)]
pub(crate) unsafe fn find(
&self,
start: *const u8,
end: *const u8,
) -> Option<Match> {
let len = end.distance(start);
debug_assert!(len >= self.minimum_len());
let mut cur = start;
while cur <= end.sub(V::Half::BYTES) {
if let Some(m) = self.find_one(cur, end) {
return Some(m);
}
cur = cur.add(V::Half::BYTES);
}
if cur < end {
cur = end.sub(V::Half::BYTES);
if let Some(m) = self.find_one(cur, end) {
return Some(m);
}
}
None
}
/// Look for a match starting at the `V::BYTES` at and after `cur`. If
/// there isn't one, then `None` is returned.
///
/// # Safety
///
/// The given pointers representing the haystack must be valid to read
/// from. They must also point to a region of memory that is at least the
/// minimum length required by this searcher.
///
/// Callers must ensure that this is okay to call in the current target for
/// the current CPU.
#[inline(always)]
unsafe fn find_one(
&self,
cur: *const u8,
end: *const u8,
) -> Option<Match> {
let c = self.candidate(cur);
if !c.is_zero() {
if let Some(m) = self.teddy.verify(cur, end, c) {
return Some(m);
}
}
None
}
/// Look for a candidate match (represented as a vector) starting at the
/// `V::BYTES` at and after `cur`. If there isn't one, then a vector with
/// all bits set to zero is returned.
///
/// # Safety
///
/// The given pointer representing the haystack must be valid to read
/// from.
///
/// Callers must ensure that this is okay to call in the current target for
/// the current CPU.
#[inline(always)]
unsafe fn candidate(&self, cur: *const u8) -> V {
let chunk = V::load_half_unaligned(cur);
Mask::members1(chunk, self.masks)
}
}
impl<V: FatVector> Fat<V, 2> {
/// See `Fat<V, 1>::find`.
#[inline(always)]
pub(crate) unsafe fn find(
&self,
start: *const u8,
end: *const u8,
) -> Option<Match> {
let len = end.distance(start);
debug_assert!(len >= self.minimum_len());
let mut cur = start.add(1);
let mut prev0 = V::splat(0xFF);
while cur <= end.sub(V::Half::BYTES) {
if let Some(m) = self.find_one(cur, end, &mut prev0) {
return Some(m);
}
cur = cur.add(V::Half::BYTES);
}
if cur < end {
cur = end.sub(V::Half::BYTES);
prev0 = V::splat(0xFF);
if let Some(m) = self.find_one(cur, end, &mut prev0) {
return Some(m);
}
}
None
}
/// See `Fat<V, 1>::find_one`.
#[inline(always)]
unsafe fn find_one(
&self,
cur: *const u8,
end: *const u8,
prev0: &mut V,
) -> Option<Match> {
let c = self.candidate(cur, prev0);
if !c.is_zero() {
if let Some(m) = self.teddy.verify(cur.sub(1), end, c) {
return Some(m);
}
}
None
}
/// See `Fat<V, 1>::candidate`.
#[inline(always)]
unsafe fn candidate(&self, cur: *const u8, prev0: &mut V) -> V {
let chunk = V::load_half_unaligned(cur);
let (res0, res1) = Mask::members2(chunk, self.masks);
let res0prev0 = res0.half_shift_in_one_byte(*prev0);
let res = res0prev0.and(res1);
*prev0 = res0;
res
}
}
impl<V: FatVector> Fat<V, 3> {
/// See `Fat<V, 1>::find`.
#[inline(always)]
pub(crate) unsafe fn find(
&self,
start: *const u8,
end: *const u8,
) -> Option<Match> {
let len = end.distance(start);
debug_assert!(len >= self.minimum_len());
let mut cur = start.add(2);
let mut prev0 = V::splat(0xFF);
let mut prev1 = V::splat(0xFF);
while cur <= end.sub(V::Half::BYTES) {
if let Some(m) = self.find_one(cur, end, &mut prev0, &mut prev1) {
return Some(m);
}
cur = cur.add(V::Half::BYTES);
}
if cur < end {
cur = end.sub(V::Half::BYTES);
prev0 = V::splat(0xFF);
prev1 = V::splat(0xFF);
if let Some(m) = self.find_one(cur, end, &mut prev0, &mut prev1) {
return Some(m);
}
}
None
}
/// See `Fat<V, 1>::find_one`.
#[inline(always)]
unsafe fn find_one(
&self,
cur: *const u8,
end: *const u8,
prev0: &mut V,
prev1: &mut V,
) -> Option<Match> {
let c = self.candidate(cur, prev0, prev1);
if !c.is_zero() {
if let Some(m) = self.teddy.verify(cur.sub(2), end, c) {
return Some(m);
}
}
None
}
/// See `Fat<V, 1>::candidate`.
#[inline(always)]
unsafe fn candidate(
&self,
cur: *const u8,
prev0: &mut V,
prev1: &mut V,
) -> V {
let chunk = V::load_half_unaligned(cur);
let (res0, res1, res2) = Mask::members3(chunk, self.masks);
let res0prev0 = res0.half_shift_in_two_bytes(*prev0);
let res1prev1 = res1.half_shift_in_one_byte(*prev1);
let res = res0prev0.and(res1prev1).and(res2);
*prev0 = res0;
*prev1 = res1;
res
}
}
impl<V: FatVector> Fat<V, 4> {
/// See `Fat<V, 1>::find`.
#[inline(always)]
pub(crate) unsafe fn find(
&self,
start: *const u8,
end: *const u8,
) -> Option<Match> {
let len = end.distance(start);
debug_assert!(len >= self.minimum_len());
let mut cur = start.add(3);
let mut prev0 = V::splat(0xFF);
let mut prev1 = V::splat(0xFF);
let mut prev2 = V::splat(0xFF);
while cur <= end.sub(V::Half::BYTES) {
if let Some(m) =
self.find_one(cur, end, &mut prev0, &mut prev1, &mut prev2)
{
return Some(m);
}
cur = cur.add(V::Half::BYTES);
}
if cur < end {
cur = end.sub(V::Half::BYTES);
prev0 = V::splat(0xFF);
prev1 = V::splat(0xFF);
prev2 = V::splat(0xFF);
if let Some(m) =
self.find_one(cur, end, &mut prev0, &mut prev1, &mut prev2)
{
return Some(m);
}
}
None
}
/// See `Fat<V, 1>::find_one`.
#[inline(always)]
unsafe fn find_one(
&self,
cur: *const u8,
end: *const u8,
prev0: &mut V,
prev1: &mut V,
prev2: &mut V,
) -> Option<Match> {
let c = self.candidate(cur, prev0, prev1, prev2);
if !c.is_zero() {
if let Some(m) = self.teddy.verify(cur.sub(3), end, c) {
return Some(m);
}
}
None
}
/// See `Fat<V, 1>::candidate`.
#[inline(always)]
unsafe fn candidate(
&self,
cur: *const u8,
prev0: &mut V,
prev1: &mut V,
prev2: &mut V,
) -> V {
let chunk = V::load_half_unaligned(cur);
let (res0, res1, res2, res3) = Mask::members4(chunk, self.masks);
let res0prev0 = res0.half_shift_in_three_bytes(*prev0);
let res1prev1 = res1.half_shift_in_two_bytes(*prev1);
let res2prev2 = res2.half_shift_in_one_byte(*prev2);
let res = res0prev0.and(res1prev1).and(res2prev2).and(res3);
*prev0 = res0;
*prev1 = res1;
*prev2 = res2;
res
}
}
/// The common elements of all "slim" and "fat" Teddy search implementations.
///
/// Essentially, this contains the patterns and the buckets. Namely, it
/// contains enough to implement the verification step after candidates are
/// identified via the shuffle masks.
///
/// It is generic over the number of buckets used. In general, the number of
/// buckets is either 8 (for "slim" Teddy) or 16 (for "fat" Teddy). The generic
/// parameter isn't really meant to be instantiated for any value other than
/// 8 or 16, although it is technically possible. The main hiccup is that there
/// is some bit-shifting done in the critical part of verification that could
/// be quite expensive if `N` is not a multiple of 2.
#[derive(Clone, Debug)]
struct Teddy<const BUCKETS: usize> {
/// The patterns we are searching for.
///
/// A pattern string can be found by its `PatternID`.
patterns: Arc<Patterns>,
/// The allocation of patterns in buckets. This only contains the IDs of
/// patterns. In order to do full verification, callers must provide the
/// actual patterns when using Teddy.
buckets: [Vec<PatternID>; BUCKETS],
// N.B. The above representation is very simple, but it definitely results
// in ping-ponging between different allocations during verification. I've
// tried experimenting with other representations that flatten the pattern
// strings into a single allocation, but it doesn't seem to help much.
// Probably everything is small enough to fit into cache anyway, and so the
// pointer chasing isn't a big deal?
//
// One other avenue I haven't explored is some kind of hashing trick
// that let's us do another high-confidence check before launching into
// `memcmp`.
}
impl<const BUCKETS: usize> Teddy<BUCKETS> {
/// Create a new generic data structure for Teddy verification.
fn new(patterns: Arc<Patterns>) -> Teddy<BUCKETS> {
assert_ne!(0, patterns.len(), "Teddy requires at least one pattern");
assert_ne!(
0,
patterns.minimum_len(),
"Teddy does not support zero-length patterns"
);
assert!(
BUCKETS == 8 || BUCKETS == 16,
"Teddy only supports 8 or 16 buckets"
);
// MSRV(1.63): Use core::array::from_fn below instead of allocating a
// superfluous outer Vec. Not a big deal (especially given the BTreeMap
// allocation below), but nice to not do it.
let buckets =
<[Vec<PatternID>; BUCKETS]>::try_from(vec![vec![]; BUCKETS])
.unwrap();
let mut t = Teddy { patterns, buckets };
let mut map: BTreeMap<Box<[u8]>, usize> = BTreeMap::new();
for (id, pattern) in t.patterns.iter() {
// We try to be slightly clever in how we assign patterns into
// buckets. Generally speaking, we want patterns with the same
// prefix to be in the same bucket, since it minimizes the amount
// of time we spend churning through buckets in the verification
// step.
//
// So we could assign patterns with the same N-prefix (where N is
// the size of the mask, which is one of {1, 2, 3}) to the same
// bucket. However, case insensitive searches are fairly common, so
// we'd for example, ideally want to treat `abc` and `ABC` as if
// they shared the same prefix. ASCII has the nice property that
// the lower 4 bits of A and a are the same, so we therefore group
// patterns with the same low-nybble-N-prefix into the same bucket.
//
// MOREOVER, this is actually necessary for correctness! In
// particular, by grouping patterns with the same prefix into the
// same bucket, we ensure that we preserve correct leftmost-first
// and leftmost-longest match semantics. In addition to the fact
// that `patterns.iter()` iterates in the correct order, this
// guarantees that all possible ambiguous matches will occur in
// the same bucket. The verification routine could be adjusted to
// support correct leftmost match semantics regardless of bucket
// allocation, but that results in a performance hit. It's much
// nicer to be able to just stop as soon as a match is found.
let lonybs = pattern.low_nybbles(t.mask_len());
if let Some(&bucket) = map.get(&lonybs) {
t.buckets[bucket].push(id);
} else {
// N.B. We assign buckets in reverse because it shouldn't have
// any influence on performance, but it does make it harder to
// get leftmost match semantics accidentally correct.
let bucket = (BUCKETS - 1) - (id.as_usize() % BUCKETS);
t.buckets[bucket].push(id);
map.insert(lonybs, bucket);
}
}
t
}
/// Verify whether there are any matches starting at or after `cur` in the
/// haystack. The candidate chunk given should correspond to 8-bit bitsets
/// for N buckets.
///
/// # Safety
///
/// The given pointers representing the haystack must be valid to read
/// from.
#[inline(always)]
unsafe fn verify64(
&self,
cur: *const u8,
end: *const u8,
mut candidate_chunk: u64,
) -> Option<Match> {
while candidate_chunk != 0 {
let bit = candidate_chunk.trailing_zeros().as_usize();
candidate_chunk &= !(1 << bit);
let cur = cur.add(bit / BUCKETS);
let bucket = bit % BUCKETS;
if let Some(m) = self.verify_bucket(cur, end, bucket) {
return Some(m);
}
}
None
}
/// Verify whether there are any matches starting at `at` in the given
/// `haystack` corresponding only to patterns in the given bucket.
///
/// # Safety
///
/// The given pointers representing the haystack must be valid to read
/// from.
///
/// The bucket index must be less than or equal to `self.buckets.len()`.
#[inline(always)]
unsafe fn verify_bucket(
&self,
cur: *const u8,
end: *const u8,
bucket: usize,
) -> Option<Match> {
debug_assert!(bucket < self.buckets.len());
// SAFETY: The caller must ensure that the bucket index is correct.
for pid in self.buckets.get_unchecked(bucket).iter().copied() {
// SAFETY: This is safe because we are guaranteed that every
// index in a Teddy bucket is a valid index into `pats`, by
// construction.
debug_assert!(pid.as_usize() < self.patterns.len());
let pat = self.patterns.get_unchecked(pid);
if pat.is_prefix_raw(cur, end) {
let start = cur;
let end = start.add(pat.len());
return Some(Match { pid, start, end });
}
}
None
}
/// Returns the total number of masks required by the patterns in this
/// Teddy searcher.
///
/// Basically, the mask length corresponds to the type of Teddy searcher
/// to use: a 1-byte, 2-byte, 3-byte or 4-byte searcher. The bigger the
/// better, typically, since searching for longer substrings usually
/// decreases the rate of false positives. Therefore, the number of masks
/// needed is the length of the shortest pattern in this searcher. If the
/// length of the shortest pattern (in bytes) is bigger than 4, then the
/// mask length is 4 since there are no Teddy searchers for more than 4
/// bytes.
fn mask_len(&self) -> usize {
core::cmp::min(4, self.patterns.minimum_len())
}
/// Returns the approximate total amount of heap used by this type, in
/// units of bytes.
fn memory_usage(&self) -> usize {
// This is an upper bound rather than a precise accounting. No
// particular reason, other than it's probably very close to actual
// memory usage in practice.
self.patterns.len() * core::mem::size_of::<PatternID>()
}
}
impl Teddy<8> {
/// Runs the verification routine for "slim" Teddy.
///
/// The candidate given should be a collection of 8-bit bitsets (one bitset
/// per lane), where the ith bit is set in the jth lane if and only if the
/// byte occurring at `at + j` in `cur` is in the bucket `i`.
///
/// # Safety
///
/// Callers must ensure that this is okay to call in the current target for
/// the current CPU.
///
/// The given pointers must be valid to read from.
#[inline(always)]
unsafe fn verify<V: Vector>(
&self,
mut cur: *const u8,
end: *const u8,
candidate: V,
) -> Option<Match> {
debug_assert!(!candidate.is_zero());
// Convert the candidate into 64-bit chunks, and then verify each of
// those chunks.
candidate.for_each_64bit_lane(
#[inline(always)]
|_, chunk| {
let result = self.verify64(cur, end, chunk);
cur = cur.add(8);
result
},
)
}
}
impl Teddy<16> {
/// Runs the verification routine for "fat" Teddy.
///
/// The candidate given should be a collection of 8-bit bitsets (one bitset
/// per lane), where the ith bit is set in the jth lane if and only if the
/// byte occurring at `at + (j < 16 ? j : j - 16)` in `cur` is in the
/// bucket `j < 16 ? i : i + 8`.
///
/// # Safety
///
/// Callers must ensure that this is okay to call in the current target for
/// the current CPU.
///
/// The given pointers must be valid to read from.
#[inline(always)]
unsafe fn verify<V: FatVector>(
&self,
mut cur: *const u8,
end: *const u8,
candidate: V,
) -> Option<Match> {
// This is a bit tricky, but we basically want to convert our
// candidate, which looks like this (assuming a 256-bit vector):
//
// a31 a30 ... a17 a16 a15 a14 ... a01 a00
//
// where each a(i) is an 8-bit bitset corresponding to the activated
// buckets, to this
//
// a31 a15 a30 a14 a29 a13 ... a18 a02 a17 a01 a16 a00
//
// Namely, for Fat Teddy, the high 128-bits of the candidate correspond
// to the same bytes in the haystack in the low 128-bits (so we only
// scan 16 bytes at a time), but are for buckets 8-15 instead of 0-7.
//
// The verification routine wants to look at all potentially matching
// buckets before moving on to the next lane. So for example, both
// a16 and a00 both correspond to the first byte in our window; a00
// contains buckets 0-7 and a16 contains buckets 8-15. Specifically,
// a16 should be checked before a01. So the transformation shown above
// allows us to use our normal verification procedure with one small
// change: we treat each bitset as 16 bits instead of 8 bits.
debug_assert!(!candidate.is_zero());
// Swap the 128-bit lanes in the candidate vector.
let swapped = candidate.swap_halves();
// Interleave the bytes from the low 128-bit lanes, starting with
// cand first.
let r1 = candidate.interleave_low_8bit_lanes(swapped);
// Interleave the bytes from the high 128-bit lanes, starting with
// cand first.
let r2 = candidate.interleave_high_8bit_lanes(swapped);
// Now just take the 2 low 64-bit integers from both r1 and r2. We
// can drop the high 64-bit integers because they are a mirror image
// of the low 64-bit integers. All we care about are the low 128-bit
// lanes of r1 and r2. Combined, they contain all our 16-bit bitsets
// laid out in the desired order, as described above.
r1.for_each_low_64bit_lane(
r2,
#[inline(always)]
|_, chunk| {
let result = self.verify64(cur, end, chunk);
cur = cur.add(4);
result
},
)
}
}
/// A vector generic mask for the low and high nybbles in a set of patterns.
/// Each 8-bit lane `j` in a vector corresponds to a bitset where the `i`th bit
/// is set if and only if the nybble `j` is in the bucket `i` at a particular
/// position.
///
/// This is slightly tweaked dependending on whether Slim or Fat Teddy is being
/// used. For Slim Teddy, the bitsets in the lower half are the same as the
/// bitsets in the higher half, so that we can search `V::BYTES` bytes at a
/// time. (Remember, the nybbles in the haystack are used as indices into these
/// masks, and 256-bit shuffles only operate on 128-bit lanes.)
///
/// For Fat Teddy, the bitsets are not repeated, but instead, the high half
/// bits correspond to an addition 8 buckets. So that a bitset `00100010` has
/// buckets 1 and 5 set if it's in the lower half, but has buckets 9 and 13 set
/// if it's in the higher half.
#[derive(Clone, Copy, Debug)]
struct Mask<V> {
lo: V,
hi: V,
}
impl<V: Vector> Mask<V> {
/// Return a candidate for Teddy (fat or slim) that is searching for 1-byte
/// candidates.
///
/// If a candidate is returned, it will be a collection of 8-bit bitsets
/// (one bitset per lane), where the ith bit is set in the jth lane if and
/// only if the byte occurring at the jth lane in `chunk` is in the bucket
/// `i`. If no candidate is found, then the vector returned will have all
/// lanes set to zero.
///
/// `chunk` should correspond to a `V::BYTES` window of the haystack (where
/// the least significant byte corresponds to the start of the window). For
/// fat Teddy, the haystack window length should be `V::BYTES / 2`, with
/// the window repeated in each half of the vector.
///
/// `mask1` should correspond to a low/high mask for the first byte of all
/// patterns that are being searched.
#[inline(always)]
unsafe fn members1(chunk: V, masks: [Mask<V>; 1]) -> V {
let lomask = V::splat(0xF);
let hlo = chunk.and(lomask);
let hhi = chunk.shift_8bit_lane_right::<4>().and(lomask);
let locand = masks[0].lo.shuffle_bytes(hlo);
let hicand = masks[0].hi.shuffle_bytes(hhi);
locand.and(hicand)
}
/// Return a candidate for Teddy (fat or slim) that is searching for 2-byte
/// candidates.
///
/// If candidates are returned, each will be a collection of 8-bit bitsets
/// (one bitset per lane), where the ith bit is set in the jth lane if and
/// only if the byte occurring at the jth lane in `chunk` is in the bucket
/// `i`. Each candidate returned corresponds to the first and second bytes
/// of the patterns being searched. If no candidate is found, then all of
/// the lanes will be set to zero in at least one of the vectors returned.
///
/// `chunk` should correspond to a `V::BYTES` window of the haystack (where
/// the least significant byte corresponds to the start of the window). For
/// fat Teddy, the haystack window length should be `V::BYTES / 2`, with
/// the window repeated in each half of the vector.
///
/// The masks should correspond to the masks computed for the first and
/// second bytes of all patterns that are being searched.
#[inline(always)]
unsafe fn members2(chunk: V, masks: [Mask<V>; 2]) -> (V, V) {
let lomask = V::splat(0xF);
let hlo = chunk.and(lomask);
let hhi = chunk.shift_8bit_lane_right::<4>().and(lomask);
let locand1 = masks[0].lo.shuffle_bytes(hlo);
let hicand1 = masks[0].hi.shuffle_bytes(hhi);
let cand1 = locand1.and(hicand1);
let locand2 = masks[1].lo.shuffle_bytes(hlo);
let hicand2 = masks[1].hi.shuffle_bytes(hhi);
let cand2 = locand2.and(hicand2);
(cand1, cand2)
}
/// Return a candidate for Teddy (fat or slim) that is searching for 3-byte
/// candidates.
///
/// If candidates are returned, each will be a collection of 8-bit bitsets
/// (one bitset per lane), where the ith bit is set in the jth lane if and
/// only if the byte occurring at the jth lane in `chunk` is in the bucket
/// `i`. Each candidate returned corresponds to the first, second and third
/// bytes of the patterns being searched. If no candidate is found, then
/// all of the lanes will be set to zero in at least one of the vectors
/// returned.
///
/// `chunk` should correspond to a `V::BYTES` window of the haystack (where
/// the least significant byte corresponds to the start of the window). For
/// fat Teddy, the haystack window length should be `V::BYTES / 2`, with
/// the window repeated in each half of the vector.
///
/// The masks should correspond to the masks computed for the first, second
/// and third bytes of all patterns that are being searched.
#[inline(always)]
unsafe fn members3(chunk: V, masks: [Mask<V>; 3]) -> (V, V, V) {
let lomask = V::splat(0xF);
let hlo = chunk.and(lomask);
let hhi = chunk.shift_8bit_lane_right::<4>().and(lomask);
let locand1 = masks[0].lo.shuffle_bytes(hlo);
let hicand1 = masks[0].hi.shuffle_bytes(hhi);
let cand1 = locand1.and(hicand1);
let locand2 = masks[1].lo.shuffle_bytes(hlo);
let hicand2 = masks[1].hi.shuffle_bytes(hhi);
let cand2 = locand2.and(hicand2);
let locand3 = masks[2].lo.shuffle_bytes(hlo);
let hicand3 = masks[2].hi.shuffle_bytes(hhi);
let cand3 = locand3.and(hicand3);
(cand1, cand2, cand3)
}
/// Return a candidate for Teddy (fat or slim) that is searching for 4-byte
/// candidates.
///
/// If candidates are returned, each will be a collection of 8-bit bitsets
/// (one bitset per lane), where the ith bit is set in the jth lane if and
/// only if the byte occurring at the jth lane in `chunk` is in the bucket
/// `i`. Each candidate returned corresponds to the first, second, third
/// and fourth bytes of the patterns being searched. If no candidate is
/// found, then all of the lanes will be set to zero in at least one of the
/// vectors returned.
///
/// `chunk` should correspond to a `V::BYTES` window of the haystack (where
/// the least significant byte corresponds to the start of the window). For
/// fat Teddy, the haystack window length should be `V::BYTES / 2`, with
/// the window repeated in each half of the vector.
///
/// The masks should correspond to the masks computed for the first,
/// second, third and fourth bytes of all patterns that are being searched.
#[inline(always)]
unsafe fn members4(chunk: V, masks: [Mask<V>; 4]) -> (V, V, V, V) {
let lomask = V::splat(0xF);
let hlo = chunk.and(lomask);
let hhi = chunk.shift_8bit_lane_right::<4>().and(lomask);
let locand1 = masks[0].lo.shuffle_bytes(hlo);
let hicand1 = masks[0].hi.shuffle_bytes(hhi);
let cand1 = locand1.and(hicand1);
let locand2 = masks[1].lo.shuffle_bytes(hlo);
let hicand2 = masks[1].hi.shuffle_bytes(hhi);
let cand2 = locand2.and(hicand2);
let locand3 = masks[2].lo.shuffle_bytes(hlo);
let hicand3 = masks[2].hi.shuffle_bytes(hhi);
let cand3 = locand3.and(hicand3);
let locand4 = masks[3].lo.shuffle_bytes(hlo);
let hicand4 = masks[3].hi.shuffle_bytes(hhi);
let cand4 = locand4.and(hicand4);
(cand1, cand2, cand3, cand4)
}
}
/// Represents the low and high nybble masks that will be used during
/// search. Each mask is 32 bytes wide, although only the first 16 bytes are
/// used for 128-bit vectors.
///
/// Each byte in the mask corresponds to a 8-bit bitset, where bit `i` is set
/// if and only if the corresponding nybble is in the ith bucket. The index of
/// the byte (0-15, inclusive) corresponds to the nybble.
///
/// Each mask is used as the target of a shuffle, where the indices for the
/// shuffle are taken from the haystack. AND'ing the shuffles for both the
/// low and high masks together also results in 8-bit bitsets, but where bit
/// `i` is set if and only if the correspond *byte* is in the ith bucket.
#[derive(Clone, Default)]
struct SlimMaskBuilder {
lo: [u8; 32],
hi: [u8; 32],
}
impl SlimMaskBuilder {
/// Update this mask by adding the given byte to the given bucket. The
/// given bucket must be in the range 0-7.
///
/// # Panics
///
/// When `bucket >= 8`.
fn add(&mut self, bucket: usize, byte: u8) {
assert!(bucket < 8);
let bucket = u8::try_from(bucket).unwrap();
let byte_lo = usize::from(byte & 0xF);
let byte_hi = usize::from((byte >> 4) & 0xF);
// When using 256-bit vectors, we need to set this bucket assignment in
// the low and high 128-bit portions of the mask. This allows us to
// process 32 bytes at a time. Namely, AVX2 shuffles operate on each
// of the 128-bit lanes, rather than the full 256-bit vector at once.
self.lo[byte_lo] |= 1 << bucket;
self.lo[byte_lo + 16] |= 1 << bucket;
self.hi[byte_hi] |= 1 << bucket;
self.hi[byte_hi + 16] |= 1 << bucket;
}
/// Turn this builder into a vector mask.
///
/// # Panics
///
/// When `V` represents a vector bigger than what `MaskBytes` can contain.
///
/// # Safety
///
/// Callers must ensure that this is okay to call in the current target for
/// the current CPU.
#[inline(always)]
unsafe fn build<V: Vector>(&self) -> Mask<V> {
assert!(V::BYTES <= self.lo.len());
assert!(V::BYTES <= self.hi.len());
Mask {
lo: V::load_unaligned(self.lo[..].as_ptr()),
hi: V::load_unaligned(self.hi[..].as_ptr()),
}
}
/// A convenience function for building `N` vector masks from a slim
/// `Teddy` value.
///
/// # Panics
///
/// When `V` represents a vector bigger than what `MaskBytes` can contain.
///
/// # Safety
///
/// Callers must ensure that this is okay to call in the current target for
/// the current CPU.
#[inline(always)]
unsafe fn from_teddy<const BYTES: usize, V: Vector>(
teddy: &Teddy<8>,
) -> [Mask<V>; BYTES] {
// MSRV(1.63): Use core::array::from_fn to just build the array here
// instead of creating a vector and turning it into an array.
let mut mask_builders = vec![SlimMaskBuilder::default(); BYTES];
for (bucket_index, bucket) in teddy.buckets.iter().enumerate() {
for pid in bucket.iter().copied() {
let pat = teddy.patterns.get(pid);
for (i, builder) in mask_builders.iter_mut().enumerate() {
builder.add(bucket_index, pat.bytes()[i]);
}
}
}
let array =
<[SlimMaskBuilder; BYTES]>::try_from(mask_builders).unwrap();
array.map(|builder| builder.build())
}
}
impl Debug for SlimMaskBuilder {
fn fmt(&self, f: &mut core::fmt::Formatter<'_>) -> core::fmt::Result {
let (mut parts_lo, mut parts_hi) = (vec![], vec![]);
for i in 0..32 {
parts_lo.push(format!("{:02}: {:08b}", i, self.lo[i]));
parts_hi.push(format!("{:02}: {:08b}", i, self.hi[i]));
}
f.debug_struct("SlimMaskBuilder")
.field("lo", &parts_lo)
.field("hi", &parts_hi)
.finish()
}
}
/// Represents the low and high nybble masks that will be used during "fat"
/// Teddy search.
///
/// Each mask is 32 bytes wide, and at the time of writing, only 256-bit vectors
/// support fat Teddy.
///
/// A fat Teddy mask is like a slim Teddy mask, except that instead of
/// repeating the bitsets in the high and low 128-bits in 256-bit vectors, the
/// high and low 128-bit halves each represent distinct buckets. (Bringing the
/// total to 16 instead of 8.) This permits spreading the patterns out a bit
/// more and thus putting less pressure on verification to be fast.
///
/// Each byte in the mask corresponds to a 8-bit bitset, where bit `i` is set
/// if and only if the corresponding nybble is in the ith bucket. The index of
/// the byte (0-15, inclusive) corresponds to the nybble.
#[derive(Clone, Copy, Default)]
struct FatMaskBuilder {
lo: [u8; 32],
hi: [u8; 32],
}
impl FatMaskBuilder {
/// Update this mask by adding the given byte to the given bucket. The
/// given bucket must be in the range 0-15.
///
/// # Panics
///
/// When `bucket >= 16`.
fn add(&mut self, bucket: usize, byte: u8) {
assert!(bucket < 16);
let bucket = u8::try_from(bucket).unwrap();
let byte_lo = usize::from(byte & 0xF);
let byte_hi = usize::from((byte >> 4) & 0xF);
// Unlike slim teddy, fat teddy only works with AVX2. For fat teddy,
// the high 128 bits of our mask correspond to buckets 8-15, while the
// low 128 bits correspond to buckets 0-7.
if bucket < 8 {
self.lo[byte_lo] |= 1 << bucket;
self.hi[byte_hi] |= 1 << bucket;
} else {
self.lo[byte_lo + 16] |= 1 << (bucket % 8);
self.hi[byte_hi + 16] |= 1 << (bucket % 8);
}
}
/// Turn this builder into a vector mask.
///
/// # Panics
///
/// When `V` represents a vector bigger than what `MaskBytes` can contain.
///
/// # Safety
///
/// Callers must ensure that this is okay to call in the current target for
/// the current CPU.
#[inline(always)]
unsafe fn build<V: Vector>(&self) -> Mask<V> {
assert!(V::BYTES <= self.lo.len());
assert!(V::BYTES <= self.hi.len());
Mask {
lo: V::load_unaligned(self.lo[..].as_ptr()),
hi: V::load_unaligned(self.hi[..].as_ptr()),
}
}
/// A convenience function for building `N` vector masks from a fat
/// `Teddy` value.
///
/// # Panics
///
/// When `V` represents a vector bigger than what `MaskBytes` can contain.
///
/// # Safety
///
/// Callers must ensure that this is okay to call in the current target for
/// the current CPU.
#[inline(always)]
unsafe fn from_teddy<const BYTES: usize, V: Vector>(
teddy: &Teddy<16>,
) -> [Mask<V>; BYTES] {
// MSRV(1.63): Use core::array::from_fn to just build the array here
// instead of creating a vector and turning it into an array.
let mut mask_builders = vec![FatMaskBuilder::default(); BYTES];
for (bucket_index, bucket) in teddy.buckets.iter().enumerate() {
for pid in bucket.iter().copied() {
let pat = teddy.patterns.get(pid);
for (i, builder) in mask_builders.iter_mut().enumerate() {
builder.add(bucket_index, pat.bytes()[i]);
}
}
}
let array =
<[FatMaskBuilder; BYTES]>::try_from(mask_builders).unwrap();
array.map(|builder| builder.build())
}
}
impl Debug for FatMaskBuilder {
fn fmt(&self, f: &mut core::fmt::Formatter<'_>) -> core::fmt::Result {
let (mut parts_lo, mut parts_hi) = (vec![], vec![]);
for i in 0..32 {
parts_lo.push(format!("{:02}: {:08b}", i, self.lo[i]));
parts_hi.push(format!("{:02}: {:08b}", i, self.hi[i]));
}
f.debug_struct("FatMaskBuilder")
.field("lo", &parts_lo)
.field("hi", &parts_hi)
.finish()
}
}