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 1383 1384 1385 1386 1387 1388 1389 1390 1391 1392 1393 1394 1395 1396 1397 1398 1399 1400 1401 1402 1403 1404 1405 1406 1407 1408 1409 1410 1411 1412 1413 1414 1415 1416 1417 1418 1419 1420 1421 1422 1423 1424 1425 1426 1427 1428 1429 1430 1431 1432 1433 1434 1435 1436 1437 1438 1439 1440 1441 1442 1443 1444 1445 1446 1447 1448 1449 1450 1451 1452 1453 1454 1455 1456 1457 1458 1459 1460 1461 1462 1463 1464 1465 1466 1467 1468 1469 1470 1471 1472 1473 1474 1475 1476 1477 1478 1479 1480 1481 1482 1483 1484 1485 1486 1487 1488 1489 1490 1491 1492 1493 1494 1495 1496 1497 1498 1499 1500 1501 1502 1503 1504 1505 1506 1507 1508 1509 1510 1511 1512 1513 1514 1515 1516 1517 1518 1519 1520 1521 1522 1523 1524 1525 1526 1527 1528 1529 1530 1531 1532 1533 1534 1535 1536 1537 1538 1539 1540 1541 1542 1543 1544 1545 1546 1547 1548 1549 1550 1551 1552 1553 1554 1555 1556 1557 1558 1559 1560 1561 1562 1563 1564 1565 1566 1567 1568 1569 1570 1571 1572 1573 1574 1575 1576 1577 1578 1579 1580 1581 1582 1583 1584 1585 1586 1587 1588 1589 1590 1591 1592 1593 1594 1595 1596 1597 1598 1599 1600 1601 1602 1603 1604 1605 1606 1607 1608 1609 1610 1611 1612 1613 1614 1615 1616 1617 1618 1619 1620 1621 1622 1623 1624 1625 1626 1627 1628 1629 1630 1631 1632 1633 1634 1635 1636 1637 1638 1639 1640 1641 1642 1643 1644 1645 1646 1647 1648 1649 1650 1651 1652 1653 1654 1655 1656 1657 1658 1659 1660 1661 1662 1663 1664 1665 1666 1667 1668 1669 1670 1671 1672 1673 1674 1675 1676 1677 1678 1679 1680 1681 1682 1683 1684 1685 1686 1687 1688 1689 1690 1691 1692 1693 1694 1695 1696 1697 1698 1699 1700 1701 1702 1703 1704 1705 1706 1707 1708 1709 1710 1711 1712 1713 1714 1715 1716 1717 1718 1719 1720 1721 1722 1723 1724 1725 1726 1727 1728 1729 1730 1731 1732 1733 1734 1735 1736 1737 1738 1739 1740 1741 1742 1743 1744 1745 1746 1747 1748 1749 1750 1751 1752 1753 1754 1755 1756 1757 1758 1759 1760 1761 1762 1763 1764 1765 1766 1767 1768 1769 1770 1771 1772 1773 1774 1775 1776 1777 1778 1779 1780 1781 1782 1783 1784 1785 1786 1787 1788 1789 1790 1791 1792 1793 1794 1795 1796 1797 1798 1799 1800 1801 1802 1803 1804 1805 1806 1807 1808 1809 1810 1811 1812 1813 1814 1815 1816 1817 1818 1819 1820 1821 1822 1823 1824 1825 1826 1827 1828 1829 1830 1831 1832 1833 1834 1835 1836 1837 1838 1839 1840 1841 1842 1843 1844 1845 1846 1847 1848 1849 1850 1851 1852 1853 1854 1855 1856 1857 1858 1859 1860 1861 1862 1863 1864 1865 1866 1867 1868 1869 1870 1871 1872 1873 1874 1875 1876 1877 1878 1879 1880 1881 1882 1883 1884 1885 1886 1887 1888 1889 1890 1891 1892 1893 1894 1895 1896 1897 1898 1899 1900 1901 1902 1903 1904 1905 1906 1907 1908 1909 1910 1911 1912 1913 1914 1915 1916 1917 1918 1919 1920 1921 1922 1923 1924 1925 1926 1927 1928 1929 1930 1931 1932 1933 1934 1935 1936 1937 1938 1939 1940 1941 1942 1943 1944 1945 1946 1947 1948 1949 1950 1951 1952 1953 1954 1955 1956 1957 1958 1959 1960 1961 1962 1963 1964 1965 1966 1967 1968 1969 1970 1971 1972 1973 1974 1975 1976 1977 1978 1979 1980 1981 1982 1983 1984 1985 1986 1987 1988 1989 1990 1991 1992 1993 1994 1995 1996 1997 1998 1999 2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012 2013 2014 2015 2016 2017 2018 2019 2020 2021 2022 2023 2024 2025 2026 2027 2028 2029 2030 2031 2032 2033 2034 2035 2036 2037 2038 2039 2040 2041 2042
use std::cell::{Cell, UnsafeCell};
use std::cmp;
use std::fmt;
use std::iter::FromIterator;
use std::marker::PhantomData;
use std::mem::{self, ManuallyDrop, MaybeUninit};
use std::ptr;
use std::sync::atomic::{self, AtomicIsize, AtomicPtr, AtomicUsize, Ordering};
use std::sync::Arc;
use crate::epoch::{self, Atomic, Owned};
use crate::utils::{Backoff, CachePadded};
// Minimum buffer capacity.
const MIN_CAP: usize = 64;
// Maximum number of tasks that can be stolen in `steal_batch()` and `steal_batch_and_pop()`.
const MAX_BATCH: usize = 32;
// If a buffer of at least this size is retired, thread-local garbage is flushed so that it gets
// deallocated as soon as possible.
const FLUSH_THRESHOLD_BYTES: usize = 1 << 10;
/// A buffer that holds tasks in a worker queue.
///
/// This is just a pointer to the buffer and its length - dropping an instance of this struct will
/// *not* deallocate the buffer.
struct Buffer<T> {
/// Pointer to the allocated memory.
ptr: *mut T,
/// Capacity of the buffer. Always a power of two.
cap: usize,
}
unsafe impl<T> Send for Buffer<T> {}
impl<T> Buffer<T> {
/// Allocates a new buffer with the specified capacity.
fn alloc(cap: usize) -> Buffer<T> {
debug_assert_eq!(cap, cap.next_power_of_two());
let mut v = ManuallyDrop::new(Vec::with_capacity(cap));
let ptr = v.as_mut_ptr();
Buffer { ptr, cap }
}
/// Deallocates the buffer.
unsafe fn dealloc(self) {
drop(Vec::from_raw_parts(self.ptr, 0, self.cap));
}
/// Returns a pointer to the task at the specified `index`.
unsafe fn at(&self, index: isize) -> *mut T {
// `self.cap` is always a power of two.
// We do all the loads at `MaybeUninit` because we might realize, after loading, that we
// don't actually have the right to access this memory.
self.ptr.offset(index & (self.cap - 1) as isize)
}
/// Writes `task` into the specified `index`.
///
/// This method might be concurrently called with another `read` at the same index, which is
/// technically speaking a data race and therefore UB. We should use an atomic store here, but
/// that would be more expensive and difficult to implement generically for all types `T`.
/// Hence, as a hack, we use a volatile write instead.
unsafe fn write(&self, index: isize, task: MaybeUninit<T>) {
ptr::write_volatile(self.at(index).cast::<MaybeUninit<T>>(), task)
}
/// Reads a task from the specified `index`.
///
/// This method might be concurrently called with another `write` at the same index, which is
/// technically speaking a data race and therefore UB. We should use an atomic load here, but
/// that would be more expensive and difficult to implement generically for all types `T`.
/// Hence, as a hack, we use a volatile load instead.
unsafe fn read(&self, index: isize) -> MaybeUninit<T> {
ptr::read_volatile(self.at(index).cast::<MaybeUninit<T>>())
}
}
impl<T> Clone for Buffer<T> {
fn clone(&self) -> Buffer<T> {
Buffer {
ptr: self.ptr,
cap: self.cap,
}
}
}
impl<T> Copy for Buffer<T> {}
/// Internal queue data shared between the worker and stealers.
///
/// The implementation is based on the following work:
///
/// 1. [Chase and Lev. Dynamic circular work-stealing deque. SPAA 2005.][chase-lev]
/// 2. [Le, Pop, Cohen, and Nardelli. Correct and efficient work-stealing for weak memory models.
/// PPoPP 2013.][weak-mem]
/// 3. [Norris and Demsky. CDSchecker: checking concurrent data structures written with C/C++
/// atomics. OOPSLA 2013.][checker]
///
/// [chase-lev]: https://dl.acm.org/citation.cfm?id=1073974
/// [weak-mem]: https://dl.acm.org/citation.cfm?id=2442524
/// [checker]: https://dl.acm.org/citation.cfm?id=2509514
struct Inner<T> {
/// The front index.
front: AtomicIsize,
/// The back index.
back: AtomicIsize,
/// The underlying buffer.
buffer: CachePadded<Atomic<Buffer<T>>>,
}
impl<T> Drop for Inner<T> {
fn drop(&mut self) {
// Load the back index, front index, and buffer.
let b = *self.back.get_mut();
let f = *self.front.get_mut();
unsafe {
let buffer = self.buffer.load(Ordering::Relaxed, epoch::unprotected());
// Go through the buffer from front to back and drop all tasks in the queue.
let mut i = f;
while i != b {
buffer.deref().at(i).drop_in_place();
i = i.wrapping_add(1);
}
// Free the memory allocated by the buffer.
buffer.into_owned().into_box().dealloc();
}
}
}
/// Worker queue flavor: FIFO or LIFO.
#[derive(Clone, Copy, Debug, Eq, PartialEq)]
enum Flavor {
/// The first-in first-out flavor.
Fifo,
/// The last-in first-out flavor.
Lifo,
}
/// A worker queue.
///
/// This is a FIFO or LIFO queue that is owned by a single thread, but other threads may steal
/// tasks from it. Task schedulers typically create a single worker queue per thread.
///
/// # Examples
///
/// A FIFO worker:
///
/// ```
/// use crossbeam_deque::{Steal, Worker};
///
/// let w = Worker::new_fifo();
/// let s = w.stealer();
///
/// w.push(1);
/// w.push(2);
/// w.push(3);
///
/// assert_eq!(s.steal(), Steal::Success(1));
/// assert_eq!(w.pop(), Some(2));
/// assert_eq!(w.pop(), Some(3));
/// ```
///
/// A LIFO worker:
///
/// ```
/// use crossbeam_deque::{Steal, Worker};
///
/// let w = Worker::new_lifo();
/// let s = w.stealer();
///
/// w.push(1);
/// w.push(2);
/// w.push(3);
///
/// assert_eq!(s.steal(), Steal::Success(1));
/// assert_eq!(w.pop(), Some(3));
/// assert_eq!(w.pop(), Some(2));
/// ```
pub struct Worker<T> {
/// A reference to the inner representation of the queue.
inner: Arc<CachePadded<Inner<T>>>,
/// A copy of `inner.buffer` for quick access.
buffer: Cell<Buffer<T>>,
/// The flavor of the queue.
flavor: Flavor,
/// Indicates that the worker cannot be shared among threads.
_marker: PhantomData<*mut ()>, // !Send + !Sync
}
unsafe impl<T: Send> Send for Worker<T> {}
impl<T> Worker<T> {
/// Creates a FIFO worker queue.
///
/// Tasks are pushed and popped from opposite ends.
///
/// # Examples
///
/// ```
/// use crossbeam_deque::Worker;
///
/// let w = Worker::<i32>::new_fifo();
/// ```
pub fn new_fifo() -> Worker<T> {
let buffer = Buffer::alloc(MIN_CAP);
let inner = Arc::new(CachePadded::new(Inner {
front: AtomicIsize::new(0),
back: AtomicIsize::new(0),
buffer: CachePadded::new(Atomic::new(buffer)),
}));
Worker {
inner,
buffer: Cell::new(buffer),
flavor: Flavor::Fifo,
_marker: PhantomData,
}
}
/// Creates a LIFO worker queue.
///
/// Tasks are pushed and popped from the same end.
///
/// # Examples
///
/// ```
/// use crossbeam_deque::Worker;
///
/// let w = Worker::<i32>::new_lifo();
/// ```
pub fn new_lifo() -> Worker<T> {
let buffer = Buffer::alloc(MIN_CAP);
let inner = Arc::new(CachePadded::new(Inner {
front: AtomicIsize::new(0),
back: AtomicIsize::new(0),
buffer: CachePadded::new(Atomic::new(buffer)),
}));
Worker {
inner,
buffer: Cell::new(buffer),
flavor: Flavor::Lifo,
_marker: PhantomData,
}
}
/// Creates a stealer for this queue.
///
/// The returned stealer can be shared among threads and cloned.
///
/// # Examples
///
/// ```
/// use crossbeam_deque::Worker;
///
/// let w = Worker::<i32>::new_lifo();
/// let s = w.stealer();
/// ```
pub fn stealer(&self) -> Stealer<T> {
Stealer {
inner: self.inner.clone(),
flavor: self.flavor,
}
}
/// Resizes the internal buffer to the new capacity of `new_cap`.
#[cold]
unsafe fn resize(&self, new_cap: usize) {
// Load the back index, front index, and buffer.
let b = self.inner.back.load(Ordering::Relaxed);
let f = self.inner.front.load(Ordering::Relaxed);
let buffer = self.buffer.get();
// Allocate a new buffer and copy data from the old buffer to the new one.
let new = Buffer::alloc(new_cap);
let mut i = f;
while i != b {
ptr::copy_nonoverlapping(buffer.at(i), new.at(i), 1);
i = i.wrapping_add(1);
}
let guard = &epoch::pin();
// Replace the old buffer with the new one.
self.buffer.replace(new);
let old =
self.inner
.buffer
.swap(Owned::new(new).into_shared(guard), Ordering::Release, guard);
// Destroy the old buffer later.
guard.defer_unchecked(move || old.into_owned().into_box().dealloc());
// If the buffer is very large, then flush the thread-local garbage in order to deallocate
// it as soon as possible.
if mem::size_of::<T>() * new_cap >= FLUSH_THRESHOLD_BYTES {
guard.flush();
}
}
/// Reserves enough capacity so that `reserve_cap` tasks can be pushed without growing the
/// buffer.
fn reserve(&self, reserve_cap: usize) {
if reserve_cap > 0 {
// Compute the current length.
let b = self.inner.back.load(Ordering::Relaxed);
let f = self.inner.front.load(Ordering::SeqCst);
let len = b.wrapping_sub(f) as usize;
// The current capacity.
let cap = self.buffer.get().cap;
// Is there enough capacity to push `reserve_cap` tasks?
if cap - len < reserve_cap {
// Keep doubling the capacity as much as is needed.
let mut new_cap = cap * 2;
while new_cap - len < reserve_cap {
new_cap *= 2;
}
// Resize the buffer.
unsafe {
self.resize(new_cap);
}
}
}
}
/// Returns `true` if the queue is empty.
///
/// ```
/// use crossbeam_deque::Worker;
///
/// let w = Worker::new_lifo();
///
/// assert!(w.is_empty());
/// w.push(1);
/// assert!(!w.is_empty());
/// ```
pub fn is_empty(&self) -> bool {
let b = self.inner.back.load(Ordering::Relaxed);
let f = self.inner.front.load(Ordering::SeqCst);
b.wrapping_sub(f) <= 0
}
/// Returns the number of tasks in the deque.
///
/// ```
/// use crossbeam_deque::Worker;
///
/// let w = Worker::new_lifo();
///
/// assert_eq!(w.len(), 0);
/// w.push(1);
/// assert_eq!(w.len(), 1);
/// w.push(1);
/// assert_eq!(w.len(), 2);
/// ```
pub fn len(&self) -> usize {
let b = self.inner.back.load(Ordering::Relaxed);
let f = self.inner.front.load(Ordering::SeqCst);
b.wrapping_sub(f).max(0) as usize
}
/// Pushes a task into the queue.
///
/// # Examples
///
/// ```
/// use crossbeam_deque::Worker;
///
/// let w = Worker::new_lifo();
/// w.push(1);
/// w.push(2);
/// ```
pub fn push(&self, task: T) {
// Load the back index, front index, and buffer.
let b = self.inner.back.load(Ordering::Relaxed);
let f = self.inner.front.load(Ordering::Acquire);
let mut buffer = self.buffer.get();
// Calculate the length of the queue.
let len = b.wrapping_sub(f);
// Is the queue full?
if len >= buffer.cap as isize {
// Yes. Grow the underlying buffer.
unsafe {
self.resize(2 * buffer.cap);
}
buffer = self.buffer.get();
}
// Write `task` into the slot.
unsafe {
buffer.write(b, MaybeUninit::new(task));
}
atomic::fence(Ordering::Release);
// Increment the back index.
//
// This ordering could be `Relaxed`, but then thread sanitizer would falsely report data
// races because it doesn't understand fences.
self.inner.back.store(b.wrapping_add(1), Ordering::Release);
}
/// Pops a task from the queue.
///
/// # Examples
///
/// ```
/// use crossbeam_deque::Worker;
///
/// let w = Worker::new_fifo();
/// w.push(1);
/// w.push(2);
///
/// assert_eq!(w.pop(), Some(1));
/// assert_eq!(w.pop(), Some(2));
/// assert_eq!(w.pop(), None);
/// ```
pub fn pop(&self) -> Option<T> {
// Load the back and front index.
let b = self.inner.back.load(Ordering::Relaxed);
let f = self.inner.front.load(Ordering::Relaxed);
// Calculate the length of the queue.
let len = b.wrapping_sub(f);
// Is the queue empty?
if len <= 0 {
return None;
}
match self.flavor {
// Pop from the front of the queue.
Flavor::Fifo => {
// Try incrementing the front index to pop the task.
let f = self.inner.front.fetch_add(1, Ordering::SeqCst);
let new_f = f.wrapping_add(1);
if b.wrapping_sub(new_f) < 0 {
self.inner.front.store(f, Ordering::Relaxed);
return None;
}
unsafe {
// Read the popped task.
let buffer = self.buffer.get();
let task = buffer.read(f).assume_init();
// Shrink the buffer if `len - 1` is less than one fourth of the capacity.
if buffer.cap > MIN_CAP && len <= buffer.cap as isize / 4 {
self.resize(buffer.cap / 2);
}
Some(task)
}
}
// Pop from the back of the queue.
Flavor::Lifo => {
// Decrement the back index.
let b = b.wrapping_sub(1);
self.inner.back.store(b, Ordering::Relaxed);
atomic::fence(Ordering::SeqCst);
// Load the front index.
let f = self.inner.front.load(Ordering::Relaxed);
// Compute the length after the back index was decremented.
let len = b.wrapping_sub(f);
if len < 0 {
// The queue is empty. Restore the back index to the original task.
self.inner.back.store(b.wrapping_add(1), Ordering::Relaxed);
None
} else {
// Read the task to be popped.
let buffer = self.buffer.get();
let mut task = unsafe { Some(buffer.read(b)) };
// Are we popping the last task from the queue?
if len == 0 {
// Try incrementing the front index.
if self
.inner
.front
.compare_exchange(
f,
f.wrapping_add(1),
Ordering::SeqCst,
Ordering::Relaxed,
)
.is_err()
{
// Failed. We didn't pop anything. Reset to `None`.
task.take();
}
// Restore the back index to the original task.
self.inner.back.store(b.wrapping_add(1), Ordering::Relaxed);
} else {
// Shrink the buffer if `len` is less than one fourth of the capacity.
if buffer.cap > MIN_CAP && len < buffer.cap as isize / 4 {
unsafe {
self.resize(buffer.cap / 2);
}
}
}
task.map(|t| unsafe { t.assume_init() })
}
}
}
}
}
impl<T> fmt::Debug for Worker<T> {
fn fmt(&self, f: &mut fmt::Formatter<'_>) -> fmt::Result {
f.pad("Worker { .. }")
}
}
/// A stealer handle of a worker queue.
///
/// Stealers can be shared among threads.
///
/// Task schedulers typically have a single worker queue per worker thread.
///
/// # Examples
///
/// ```
/// use crossbeam_deque::{Steal, Worker};
///
/// let w = Worker::new_lifo();
/// w.push(1);
/// w.push(2);
///
/// let s = w.stealer();
/// assert_eq!(s.steal(), Steal::Success(1));
/// assert_eq!(s.steal(), Steal::Success(2));
/// assert_eq!(s.steal(), Steal::Empty);
/// ```
pub struct Stealer<T> {
/// A reference to the inner representation of the queue.
inner: Arc<CachePadded<Inner<T>>>,
/// The flavor of the queue.
flavor: Flavor,
}
unsafe impl<T: Send> Send for Stealer<T> {}
unsafe impl<T: Send> Sync for Stealer<T> {}
impl<T> Stealer<T> {
/// Returns `true` if the queue is empty.
///
/// ```
/// use crossbeam_deque::Worker;
///
/// let w = Worker::new_lifo();
/// let s = w.stealer();
///
/// assert!(s.is_empty());
/// w.push(1);
/// assert!(!s.is_empty());
/// ```
pub fn is_empty(&self) -> bool {
let f = self.inner.front.load(Ordering::Acquire);
atomic::fence(Ordering::SeqCst);
let b = self.inner.back.load(Ordering::Acquire);
b.wrapping_sub(f) <= 0
}
/// Returns the number of tasks in the deque.
///
/// ```
/// use crossbeam_deque::Worker;
///
/// let w = Worker::new_lifo();
/// let s = w.stealer();
///
/// assert_eq!(s.len(), 0);
/// w.push(1);
/// assert_eq!(s.len(), 1);
/// w.push(2);
/// assert_eq!(s.len(), 2);
/// ```
pub fn len(&self) -> usize {
let f = self.inner.front.load(Ordering::Acquire);
atomic::fence(Ordering::SeqCst);
let b = self.inner.back.load(Ordering::Acquire);
b.wrapping_sub(f).max(0) as usize
}
/// Steals a task from the queue.
///
/// # Examples
///
/// ```
/// use crossbeam_deque::{Steal, Worker};
///
/// let w = Worker::new_lifo();
/// w.push(1);
/// w.push(2);
///
/// let s = w.stealer();
/// assert_eq!(s.steal(), Steal::Success(1));
/// assert_eq!(s.steal(), Steal::Success(2));
/// ```
pub fn steal(&self) -> Steal<T> {
// Load the front index.
let f = self.inner.front.load(Ordering::Acquire);
// A SeqCst fence is needed here.
//
// If the current thread is already pinned (reentrantly), we must manually issue the
// fence. Otherwise, the following pinning will issue the fence anyway, so we don't
// have to.
if epoch::is_pinned() {
atomic::fence(Ordering::SeqCst);
}
let guard = &epoch::pin();
// Load the back index.
let b = self.inner.back.load(Ordering::Acquire);
// Is the queue empty?
if b.wrapping_sub(f) <= 0 {
return Steal::Empty;
}
// Load the buffer and read the task at the front.
let buffer = self.inner.buffer.load(Ordering::Acquire, guard);
let task = unsafe { buffer.deref().read(f) };
// Try incrementing the front index to steal the task.
// If the buffer has been swapped or the increment fails, we retry.
if self.inner.buffer.load(Ordering::Acquire, guard) != buffer
|| self
.inner
.front
.compare_exchange(f, f.wrapping_add(1), Ordering::SeqCst, Ordering::Relaxed)
.is_err()
{
// We didn't steal this task, forget it.
return Steal::Retry;
}
// Return the stolen task.
Steal::Success(unsafe { task.assume_init() })
}
/// Steals a batch of tasks and pushes them into another worker.
///
/// How many tasks exactly will be stolen is not specified. That said, this method will try to
/// steal around half of the tasks in the queue, but also not more than some constant limit.
///
/// # Examples
///
/// ```
/// use crossbeam_deque::Worker;
///
/// let w1 = Worker::new_fifo();
/// w1.push(1);
/// w1.push(2);
/// w1.push(3);
/// w1.push(4);
///
/// let s = w1.stealer();
/// let w2 = Worker::new_fifo();
///
/// let _ = s.steal_batch(&w2);
/// assert_eq!(w2.pop(), Some(1));
/// assert_eq!(w2.pop(), Some(2));
/// ```
pub fn steal_batch(&self, dest: &Worker<T>) -> Steal<()> {
if Arc::ptr_eq(&self.inner, &dest.inner) {
if dest.is_empty() {
return Steal::Empty;
} else {
return Steal::Success(());
}
}
// Load the front index.
let mut f = self.inner.front.load(Ordering::Acquire);
// A SeqCst fence is needed here.
//
// If the current thread is already pinned (reentrantly), we must manually issue the
// fence. Otherwise, the following pinning will issue the fence anyway, so we don't
// have to.
if epoch::is_pinned() {
atomic::fence(Ordering::SeqCst);
}
let guard = &epoch::pin();
// Load the back index.
let b = self.inner.back.load(Ordering::Acquire);
// Is the queue empty?
let len = b.wrapping_sub(f);
if len <= 0 {
return Steal::Empty;
}
// Reserve capacity for the stolen batch.
let batch_size = cmp::min((len as usize + 1) / 2, MAX_BATCH);
dest.reserve(batch_size);
let mut batch_size = batch_size as isize;
// Get the destination buffer and back index.
let dest_buffer = dest.buffer.get();
let mut dest_b = dest.inner.back.load(Ordering::Relaxed);
// Load the buffer.
let buffer = self.inner.buffer.load(Ordering::Acquire, guard);
match self.flavor {
// Steal a batch of tasks from the front at once.
Flavor::Fifo => {
// Copy the batch from the source to the destination buffer.
match dest.flavor {
Flavor::Fifo => {
for i in 0..batch_size {
unsafe {
let task = buffer.deref().read(f.wrapping_add(i));
dest_buffer.write(dest_b.wrapping_add(i), task);
}
}
}
Flavor::Lifo => {
for i in 0..batch_size {
unsafe {
let task = buffer.deref().read(f.wrapping_add(i));
dest_buffer.write(dest_b.wrapping_add(batch_size - 1 - i), task);
}
}
}
}
// Try incrementing the front index to steal the batch.
// If the buffer has been swapped or the increment fails, we retry.
if self.inner.buffer.load(Ordering::Acquire, guard) != buffer
|| self
.inner
.front
.compare_exchange(
f,
f.wrapping_add(batch_size),
Ordering::SeqCst,
Ordering::Relaxed,
)
.is_err()
{
return Steal::Retry;
}
dest_b = dest_b.wrapping_add(batch_size);
}
// Steal a batch of tasks from the front one by one.
Flavor::Lifo => {
// This loop may modify the batch_size, which triggers a clippy lint warning.
// Use a new variable to avoid the warning, and to make it clear we aren't
// modifying the loop exit condition during iteration.
let original_batch_size = batch_size;
for i in 0..original_batch_size {
// If this is not the first steal, check whether the queue is empty.
if i > 0 {
// We've already got the current front index. Now execute the fence to
// synchronize with other threads.
atomic::fence(Ordering::SeqCst);
// Load the back index.
let b = self.inner.back.load(Ordering::Acquire);
// Is the queue empty?
if b.wrapping_sub(f) <= 0 {
batch_size = i;
break;
}
}
// Read the task at the front.
let task = unsafe { buffer.deref().read(f) };
// Try incrementing the front index to steal the task.
// If the buffer has been swapped or the increment fails, we retry.
if self.inner.buffer.load(Ordering::Acquire, guard) != buffer
|| self
.inner
.front
.compare_exchange(
f,
f.wrapping_add(1),
Ordering::SeqCst,
Ordering::Relaxed,
)
.is_err()
{
// We didn't steal this task, forget it and break from the loop.
batch_size = i;
break;
}
// Write the stolen task into the destination buffer.
unsafe {
dest_buffer.write(dest_b, task);
}
// Move the source front index and the destination back index one step forward.
f = f.wrapping_add(1);
dest_b = dest_b.wrapping_add(1);
}
// If we didn't steal anything, the operation needs to be retried.
if batch_size == 0 {
return Steal::Retry;
}
// If stealing into a FIFO queue, stolen tasks need to be reversed.
if dest.flavor == Flavor::Fifo {
for i in 0..batch_size / 2 {
unsafe {
let i1 = dest_b.wrapping_sub(batch_size - i);
let i2 = dest_b.wrapping_sub(i + 1);
let t1 = dest_buffer.read(i1);
let t2 = dest_buffer.read(i2);
dest_buffer.write(i1, t2);
dest_buffer.write(i2, t1);
}
}
}
}
}
atomic::fence(Ordering::Release);
// Update the back index in the destination queue.
//
// This ordering could be `Relaxed`, but then thread sanitizer would falsely report data
// races because it doesn't understand fences.
dest.inner.back.store(dest_b, Ordering::Release);
// Return with success.
Steal::Success(())
}
/// Steals a batch of tasks, pushes them into another worker, and pops a task from that worker.
///
/// How many tasks exactly will be stolen is not specified. That said, this method will try to
/// steal around half of the tasks in the queue, but also not more than some constant limit.
///
/// # Examples
///
/// ```
/// use crossbeam_deque::{Steal, Worker};
///
/// let w1 = Worker::new_fifo();
/// w1.push(1);
/// w1.push(2);
/// w1.push(3);
/// w1.push(4);
///
/// let s = w1.stealer();
/// let w2 = Worker::new_fifo();
///
/// assert_eq!(s.steal_batch_and_pop(&w2), Steal::Success(1));
/// assert_eq!(w2.pop(), Some(2));
/// ```
pub fn steal_batch_and_pop(&self, dest: &Worker<T>) -> Steal<T> {
if Arc::ptr_eq(&self.inner, &dest.inner) {
match dest.pop() {
None => return Steal::Empty,
Some(task) => return Steal::Success(task),
}
}
// Load the front index.
let mut f = self.inner.front.load(Ordering::Acquire);
// A SeqCst fence is needed here.
//
// If the current thread is already pinned (reentrantly), we must manually issue the
// fence. Otherwise, the following pinning will issue the fence anyway, so we don't
// have to.
if epoch::is_pinned() {
atomic::fence(Ordering::SeqCst);
}
let guard = &epoch::pin();
// Load the back index.
let b = self.inner.back.load(Ordering::Acquire);
// Is the queue empty?
let len = b.wrapping_sub(f);
if len <= 0 {
return Steal::Empty;
}
// Reserve capacity for the stolen batch.
let batch_size = cmp::min((len as usize - 1) / 2, MAX_BATCH - 1);
dest.reserve(batch_size);
let mut batch_size = batch_size as isize;
// Get the destination buffer and back index.
let dest_buffer = dest.buffer.get();
let mut dest_b = dest.inner.back.load(Ordering::Relaxed);
// Load the buffer
let buffer = self.inner.buffer.load(Ordering::Acquire, guard);
// Read the task at the front.
let mut task = unsafe { buffer.deref().read(f) };
match self.flavor {
// Steal a batch of tasks from the front at once.
Flavor::Fifo => {
// Copy the batch from the source to the destination buffer.
match dest.flavor {
Flavor::Fifo => {
for i in 0..batch_size {
unsafe {
let task = buffer.deref().read(f.wrapping_add(i + 1));
dest_buffer.write(dest_b.wrapping_add(i), task);
}
}
}
Flavor::Lifo => {
for i in 0..batch_size {
unsafe {
let task = buffer.deref().read(f.wrapping_add(i + 1));
dest_buffer.write(dest_b.wrapping_add(batch_size - 1 - i), task);
}
}
}
}
// Try incrementing the front index to steal the task.
// If the buffer has been swapped or the increment fails, we retry.
if self.inner.buffer.load(Ordering::Acquire, guard) != buffer
|| self
.inner
.front
.compare_exchange(
f,
f.wrapping_add(batch_size + 1),
Ordering::SeqCst,
Ordering::Relaxed,
)
.is_err()
{
// We didn't steal this task, forget it.
return Steal::Retry;
}
dest_b = dest_b.wrapping_add(batch_size);
}
// Steal a batch of tasks from the front one by one.
Flavor::Lifo => {
// Try incrementing the front index to steal the task.
if self
.inner
.front
.compare_exchange(f, f.wrapping_add(1), Ordering::SeqCst, Ordering::Relaxed)
.is_err()
{
// We didn't steal this task, forget it.
return Steal::Retry;
}
// Move the front index one step forward.
f = f.wrapping_add(1);
// Repeat the same procedure for the batch steals.
//
// This loop may modify the batch_size, which triggers a clippy lint warning.
// Use a new variable to avoid the warning, and to make it clear we aren't
// modifying the loop exit condition during iteration.
let original_batch_size = batch_size;
for i in 0..original_batch_size {
// We've already got the current front index. Now execute the fence to
// synchronize with other threads.
atomic::fence(Ordering::SeqCst);
// Load the back index.
let b = self.inner.back.load(Ordering::Acquire);
// Is the queue empty?
if b.wrapping_sub(f) <= 0 {
batch_size = i;
break;
}
// Read the task at the front.
let tmp = unsafe { buffer.deref().read(f) };
// Try incrementing the front index to steal the task.
// If the buffer has been swapped or the increment fails, we retry.
if self.inner.buffer.load(Ordering::Acquire, guard) != buffer
|| self
.inner
.front
.compare_exchange(
f,
f.wrapping_add(1),
Ordering::SeqCst,
Ordering::Relaxed,
)
.is_err()
{
// We didn't steal this task, forget it and break from the loop.
batch_size = i;
break;
}
// Write the previously stolen task into the destination buffer.
unsafe {
dest_buffer.write(dest_b, mem::replace(&mut task, tmp));
}
// Move the source front index and the destination back index one step forward.
f = f.wrapping_add(1);
dest_b = dest_b.wrapping_add(1);
}
// If stealing into a FIFO queue, stolen tasks need to be reversed.
if dest.flavor == Flavor::Fifo {
for i in 0..batch_size / 2 {
unsafe {
let i1 = dest_b.wrapping_sub(batch_size - i);
let i2 = dest_b.wrapping_sub(i + 1);
let t1 = dest_buffer.read(i1);
let t2 = dest_buffer.read(i2);
dest_buffer.write(i1, t2);
dest_buffer.write(i2, t1);
}
}
}
}
}
atomic::fence(Ordering::Release);
// Update the back index in the destination queue.
//
// This ordering could be `Relaxed`, but then thread sanitizer would falsely report data
// races because it doesn't understand fences.
dest.inner.back.store(dest_b, Ordering::Release);
// Return with success.
Steal::Success(unsafe { task.assume_init() })
}
}
impl<T> Clone for Stealer<T> {
fn clone(&self) -> Stealer<T> {
Stealer {
inner: self.inner.clone(),
flavor: self.flavor,
}
}
}
impl<T> fmt::Debug for Stealer<T> {
fn fmt(&self, f: &mut fmt::Formatter<'_>) -> fmt::Result {
f.pad("Stealer { .. }")
}
}
// Bits indicating the state of a slot:
// * If a task has been written into the slot, `WRITE` is set.
// * If a task has been read from the slot, `READ` is set.
// * If the block is being destroyed, `DESTROY` is set.
const WRITE: usize = 1;
const READ: usize = 2;
const DESTROY: usize = 4;
// Each block covers one "lap" of indices.
const LAP: usize = 64;
// The maximum number of values a block can hold.
const BLOCK_CAP: usize = LAP - 1;
// How many lower bits are reserved for metadata.
const SHIFT: usize = 1;
// Indicates that the block is not the last one.
const HAS_NEXT: usize = 1;
/// A slot in a block.
struct Slot<T> {
/// The task.
task: UnsafeCell<MaybeUninit<T>>,
/// The state of the slot.
state: AtomicUsize,
}
impl<T> Slot<T> {
const UNINIT: Self = Self {
task: UnsafeCell::new(MaybeUninit::uninit()),
state: AtomicUsize::new(0),
};
/// Waits until a task is written into the slot.
fn wait_write(&self) {
let backoff = Backoff::new();
while self.state.load(Ordering::Acquire) & WRITE == 0 {
backoff.snooze();
}
}
}
/// A block in a linked list.
///
/// Each block in the list can hold up to `BLOCK_CAP` values.
struct Block<T> {
/// The next block in the linked list.
next: AtomicPtr<Block<T>>,
/// Slots for values.
slots: [Slot<T>; BLOCK_CAP],
}
impl<T> Block<T> {
/// Creates an empty block that starts at `start_index`.
fn new() -> Block<T> {
Self {
next: AtomicPtr::new(ptr::null_mut()),
slots: [Slot::UNINIT; BLOCK_CAP],
}
}
/// Waits until the next pointer is set.
fn wait_next(&self) -> *mut Block<T> {
let backoff = Backoff::new();
loop {
let next = self.next.load(Ordering::Acquire);
if !next.is_null() {
return next;
}
backoff.snooze();
}
}
/// Sets the `DESTROY` bit in slots starting from `start` and destroys the block.
unsafe fn destroy(this: *mut Block<T>, count: usize) {
// It is not necessary to set the `DESTROY` bit in the last slot because that slot has
// begun destruction of the block.
for i in (0..count).rev() {
let slot = (*this).slots.get_unchecked(i);
// Mark the `DESTROY` bit if a thread is still using the slot.
if slot.state.load(Ordering::Acquire) & READ == 0
&& slot.state.fetch_or(DESTROY, Ordering::AcqRel) & READ == 0
{
// If a thread is still using the slot, it will continue destruction of the block.
return;
}
}
// No thread is using the block, now it is safe to destroy it.
drop(Box::from_raw(this));
}
}
/// A position in a queue.
struct Position<T> {
/// The index in the queue.
index: AtomicUsize,
/// The block in the linked list.
block: AtomicPtr<Block<T>>,
}
/// An injector queue.
///
/// This is a FIFO queue that can be shared among multiple threads. Task schedulers typically have
/// a single injector queue, which is the entry point for new tasks.
///
/// # Examples
///
/// ```
/// use crossbeam_deque::{Injector, Steal};
///
/// let q = Injector::new();
/// q.push(1);
/// q.push(2);
///
/// assert_eq!(q.steal(), Steal::Success(1));
/// assert_eq!(q.steal(), Steal::Success(2));
/// assert_eq!(q.steal(), Steal::Empty);
/// ```
pub struct Injector<T> {
/// The head of the queue.
head: CachePadded<Position<T>>,
/// The tail of the queue.
tail: CachePadded<Position<T>>,
/// Indicates that dropping a `Injector<T>` may drop values of type `T`.
_marker: PhantomData<T>,
}
unsafe impl<T: Send> Send for Injector<T> {}
unsafe impl<T: Send> Sync for Injector<T> {}
impl<T> Default for Injector<T> {
fn default() -> Self {
let block = Box::into_raw(Box::new(Block::<T>::new()));
Self {
head: CachePadded::new(Position {
block: AtomicPtr::new(block),
index: AtomicUsize::new(0),
}),
tail: CachePadded::new(Position {
block: AtomicPtr::new(block),
index: AtomicUsize::new(0),
}),
_marker: PhantomData,
}
}
}
impl<T> Injector<T> {
/// Creates a new injector queue.
///
/// # Examples
///
/// ```
/// use crossbeam_deque::Injector;
///
/// let q = Injector::<i32>::new();
/// ```
pub fn new() -> Injector<T> {
Self::default()
}
/// Pushes a task into the queue.
///
/// # Examples
///
/// ```
/// use crossbeam_deque::Injector;
///
/// let w = Injector::new();
/// w.push(1);
/// w.push(2);
/// ```
pub fn push(&self, task: T) {
let backoff = Backoff::new();
let mut tail = self.tail.index.load(Ordering::Acquire);
let mut block = self.tail.block.load(Ordering::Acquire);
let mut next_block = None;
loop {
// Calculate the offset of the index into the block.
let offset = (tail >> SHIFT) % LAP;
// If we reached the end of the block, wait until the next one is installed.
if offset == BLOCK_CAP {
backoff.snooze();
tail = self.tail.index.load(Ordering::Acquire);
block = self.tail.block.load(Ordering::Acquire);
continue;
}
// If we're going to have to install the next block, allocate it in advance in order to
// make the wait for other threads as short as possible.
if offset + 1 == BLOCK_CAP && next_block.is_none() {
next_block = Some(Box::new(Block::<T>::new()));
}
let new_tail = tail + (1 << SHIFT);
// Try advancing the tail forward.
match self.tail.index.compare_exchange_weak(
tail,
new_tail,
Ordering::SeqCst,
Ordering::Acquire,
) {
Ok(_) => unsafe {
// If we've reached the end of the block, install the next one.
if offset + 1 == BLOCK_CAP {
let next_block = Box::into_raw(next_block.unwrap());
let next_index = new_tail.wrapping_add(1 << SHIFT);
self.tail.block.store(next_block, Ordering::Release);
self.tail.index.store(next_index, Ordering::Release);
(*block).next.store(next_block, Ordering::Release);
}
// Write the task into the slot.
let slot = (*block).slots.get_unchecked(offset);
slot.task.get().write(MaybeUninit::new(task));
slot.state.fetch_or(WRITE, Ordering::Release);
return;
},
Err(t) => {
tail = t;
block = self.tail.block.load(Ordering::Acquire);
backoff.spin();
}
}
}
}
/// Steals a task from the queue.
///
/// # Examples
///
/// ```
/// use crossbeam_deque::{Injector, Steal};
///
/// let q = Injector::new();
/// q.push(1);
/// q.push(2);
///
/// assert_eq!(q.steal(), Steal::Success(1));
/// assert_eq!(q.steal(), Steal::Success(2));
/// assert_eq!(q.steal(), Steal::Empty);
/// ```
pub fn steal(&self) -> Steal<T> {
let mut head;
let mut block;
let mut offset;
let backoff = Backoff::new();
loop {
head = self.head.index.load(Ordering::Acquire);
block = self.head.block.load(Ordering::Acquire);
// Calculate the offset of the index into the block.
offset = (head >> SHIFT) % LAP;
// If we reached the end of the block, wait until the next one is installed.
if offset == BLOCK_CAP {
backoff.snooze();
} else {
break;
}
}
let mut new_head = head + (1 << SHIFT);
if new_head & HAS_NEXT == 0 {
atomic::fence(Ordering::SeqCst);
let tail = self.tail.index.load(Ordering::Relaxed);
// If the tail equals the head, that means the queue is empty.
if head >> SHIFT == tail >> SHIFT {
return Steal::Empty;
}
// If head and tail are not in the same block, set `HAS_NEXT` in head.
if (head >> SHIFT) / LAP != (tail >> SHIFT) / LAP {
new_head |= HAS_NEXT;
}
}
// Try moving the head index forward.
if self
.head
.index
.compare_exchange_weak(head, new_head, Ordering::SeqCst, Ordering::Acquire)
.is_err()
{
return Steal::Retry;
}
unsafe {
// If we've reached the end of the block, move to the next one.
if offset + 1 == BLOCK_CAP {
let next = (*block).wait_next();
let mut next_index = (new_head & !HAS_NEXT).wrapping_add(1 << SHIFT);
if !(*next).next.load(Ordering::Relaxed).is_null() {
next_index |= HAS_NEXT;
}
self.head.block.store(next, Ordering::Release);
self.head.index.store(next_index, Ordering::Release);
}
// Read the task.
let slot = (*block).slots.get_unchecked(offset);
slot.wait_write();
let task = slot.task.get().read().assume_init();
// Destroy the block if we've reached the end, or if another thread wanted to destroy
// but couldn't because we were busy reading from the slot.
if (offset + 1 == BLOCK_CAP)
|| (slot.state.fetch_or(READ, Ordering::AcqRel) & DESTROY != 0)
{
Block::destroy(block, offset);
}
Steal::Success(task)
}
}
/// Steals a batch of tasks and pushes them into a worker.
///
/// How many tasks exactly will be stolen is not specified. That said, this method will try to
/// steal around half of the tasks in the queue, but also not more than some constant limit.
///
/// # Examples
///
/// ```
/// use crossbeam_deque::{Injector, Worker};
///
/// let q = Injector::new();
/// q.push(1);
/// q.push(2);
/// q.push(3);
/// q.push(4);
///
/// let w = Worker::new_fifo();
/// let _ = q.steal_batch(&w);
/// assert_eq!(w.pop(), Some(1));
/// assert_eq!(w.pop(), Some(2));
/// ```
pub fn steal_batch(&self, dest: &Worker<T>) -> Steal<()> {
let mut head;
let mut block;
let mut offset;
let backoff = Backoff::new();
loop {
head = self.head.index.load(Ordering::Acquire);
block = self.head.block.load(Ordering::Acquire);
// Calculate the offset of the index into the block.
offset = (head >> SHIFT) % LAP;
// If we reached the end of the block, wait until the next one is installed.
if offset == BLOCK_CAP {
backoff.snooze();
} else {
break;
}
}
let mut new_head = head;
let advance;
if new_head & HAS_NEXT == 0 {
atomic::fence(Ordering::SeqCst);
let tail = self.tail.index.load(Ordering::Relaxed);
// If the tail equals the head, that means the queue is empty.
if head >> SHIFT == tail >> SHIFT {
return Steal::Empty;
}
// If head and tail are not in the same block, set `HAS_NEXT` in head. Also, calculate
// the right batch size to steal.
if (head >> SHIFT) / LAP != (tail >> SHIFT) / LAP {
new_head |= HAS_NEXT;
// We can steal all tasks till the end of the block.
advance = (BLOCK_CAP - offset).min(MAX_BATCH);
} else {
let len = (tail - head) >> SHIFT;
// Steal half of the available tasks.
advance = ((len + 1) / 2).min(MAX_BATCH);
}
} else {
// We can steal all tasks till the end of the block.
advance = (BLOCK_CAP - offset).min(MAX_BATCH);
}
new_head += advance << SHIFT;
let new_offset = offset + advance;
// Try moving the head index forward.
if self
.head
.index
.compare_exchange_weak(head, new_head, Ordering::SeqCst, Ordering::Acquire)
.is_err()
{
return Steal::Retry;
}
// Reserve capacity for the stolen batch.
let batch_size = new_offset - offset;
dest.reserve(batch_size);
// Get the destination buffer and back index.
let dest_buffer = dest.buffer.get();
let dest_b = dest.inner.back.load(Ordering::Relaxed);
unsafe {
// If we've reached the end of the block, move to the next one.
if new_offset == BLOCK_CAP {
let next = (*block).wait_next();
let mut next_index = (new_head & !HAS_NEXT).wrapping_add(1 << SHIFT);
if !(*next).next.load(Ordering::Relaxed).is_null() {
next_index |= HAS_NEXT;
}
self.head.block.store(next, Ordering::Release);
self.head.index.store(next_index, Ordering::Release);
}
// Copy values from the injector into the destination queue.
match dest.flavor {
Flavor::Fifo => {
for i in 0..batch_size {
// Read the task.
let slot = (*block).slots.get_unchecked(offset + i);
slot.wait_write();
let task = slot.task.get().read();
// Write it into the destination queue.
dest_buffer.write(dest_b.wrapping_add(i as isize), task);
}
}
Flavor::Lifo => {
for i in 0..batch_size {
// Read the task.
let slot = (*block).slots.get_unchecked(offset + i);
slot.wait_write();
let task = slot.task.get().read();
// Write it into the destination queue.
dest_buffer.write(dest_b.wrapping_add((batch_size - 1 - i) as isize), task);
}
}
}
atomic::fence(Ordering::Release);
// Update the back index in the destination queue.
//
// This ordering could be `Relaxed`, but then thread sanitizer would falsely report
// data races because it doesn't understand fences.
dest.inner
.back
.store(dest_b.wrapping_add(batch_size as isize), Ordering::Release);
// Destroy the block if we've reached the end, or if another thread wanted to destroy
// but couldn't because we were busy reading from the slot.
if new_offset == BLOCK_CAP {
Block::destroy(block, offset);
} else {
for i in offset..new_offset {
let slot = (*block).slots.get_unchecked(i);
if slot.state.fetch_or(READ, Ordering::AcqRel) & DESTROY != 0 {
Block::destroy(block, offset);
break;
}
}
}
Steal::Success(())
}
}
/// Steals a batch of tasks, pushes them into a worker, and pops a task from that worker.
///
/// How many tasks exactly will be stolen is not specified. That said, this method will try to
/// steal around half of the tasks in the queue, but also not more than some constant limit.
///
/// # Examples
///
/// ```
/// use crossbeam_deque::{Injector, Steal, Worker};
///
/// let q = Injector::new();
/// q.push(1);
/// q.push(2);
/// q.push(3);
/// q.push(4);
///
/// let w = Worker::new_fifo();
/// assert_eq!(q.steal_batch_and_pop(&w), Steal::Success(1));
/// assert_eq!(w.pop(), Some(2));
/// ```
pub fn steal_batch_and_pop(&self, dest: &Worker<T>) -> Steal<T> {
let mut head;
let mut block;
let mut offset;
let backoff = Backoff::new();
loop {
head = self.head.index.load(Ordering::Acquire);
block = self.head.block.load(Ordering::Acquire);
// Calculate the offset of the index into the block.
offset = (head >> SHIFT) % LAP;
// If we reached the end of the block, wait until the next one is installed.
if offset == BLOCK_CAP {
backoff.snooze();
} else {
break;
}
}
let mut new_head = head;
let advance;
if new_head & HAS_NEXT == 0 {
atomic::fence(Ordering::SeqCst);
let tail = self.tail.index.load(Ordering::Relaxed);
// If the tail equals the head, that means the queue is empty.
if head >> SHIFT == tail >> SHIFT {
return Steal::Empty;
}
// If head and tail are not in the same block, set `HAS_NEXT` in head.
if (head >> SHIFT) / LAP != (tail >> SHIFT) / LAP {
new_head |= HAS_NEXT;
// We can steal all tasks till the end of the block.
advance = (BLOCK_CAP - offset).min(MAX_BATCH + 1);
} else {
let len = (tail - head) >> SHIFT;
// Steal half of the available tasks.
advance = ((len + 1) / 2).min(MAX_BATCH + 1);
}
} else {
// We can steal all tasks till the end of the block.
advance = (BLOCK_CAP - offset).min(MAX_BATCH + 1);
}
new_head += advance << SHIFT;
let new_offset = offset + advance;
// Try moving the head index forward.
if self
.head
.index
.compare_exchange_weak(head, new_head, Ordering::SeqCst, Ordering::Acquire)
.is_err()
{
return Steal::Retry;
}
// Reserve capacity for the stolen batch.
let batch_size = new_offset - offset - 1;
dest.reserve(batch_size);
// Get the destination buffer and back index.
let dest_buffer = dest.buffer.get();
let dest_b = dest.inner.back.load(Ordering::Relaxed);
unsafe {
// If we've reached the end of the block, move to the next one.
if new_offset == BLOCK_CAP {
let next = (*block).wait_next();
let mut next_index = (new_head & !HAS_NEXT).wrapping_add(1 << SHIFT);
if !(*next).next.load(Ordering::Relaxed).is_null() {
next_index |= HAS_NEXT;
}
self.head.block.store(next, Ordering::Release);
self.head.index.store(next_index, Ordering::Release);
}
// Read the task.
let slot = (*block).slots.get_unchecked(offset);
slot.wait_write();
let task = slot.task.get().read();
match dest.flavor {
Flavor::Fifo => {
// Copy values from the injector into the destination queue.
for i in 0..batch_size {
// Read the task.
let slot = (*block).slots.get_unchecked(offset + i + 1);
slot.wait_write();
let task = slot.task.get().read();
// Write it into the destination queue.
dest_buffer.write(dest_b.wrapping_add(i as isize), task);
}
}
Flavor::Lifo => {
// Copy values from the injector into the destination queue.
for i in 0..batch_size {
// Read the task.
let slot = (*block).slots.get_unchecked(offset + i + 1);
slot.wait_write();
let task = slot.task.get().read();
// Write it into the destination queue.
dest_buffer.write(dest_b.wrapping_add((batch_size - 1 - i) as isize), task);
}
}
}
atomic::fence(Ordering::Release);
// Update the back index in the destination queue.
//
// This ordering could be `Relaxed`, but then thread sanitizer would falsely report
// data races because it doesn't understand fences.
dest.inner
.back
.store(dest_b.wrapping_add(batch_size as isize), Ordering::Release);
// Destroy the block if we've reached the end, or if another thread wanted to destroy
// but couldn't because we were busy reading from the slot.
if new_offset == BLOCK_CAP {
Block::destroy(block, offset);
} else {
for i in offset..new_offset {
let slot = (*block).slots.get_unchecked(i);
if slot.state.fetch_or(READ, Ordering::AcqRel) & DESTROY != 0 {
Block::destroy(block, offset);
break;
}
}
}
Steal::Success(task.assume_init())
}
}
/// Returns `true` if the queue is empty.
///
/// # Examples
///
/// ```
/// use crossbeam_deque::Injector;
///
/// let q = Injector::new();
///
/// assert!(q.is_empty());
/// q.push(1);
/// assert!(!q.is_empty());
/// ```
pub fn is_empty(&self) -> bool {
let head = self.head.index.load(Ordering::SeqCst);
let tail = self.tail.index.load(Ordering::SeqCst);
head >> SHIFT == tail >> SHIFT
}
/// Returns the number of tasks in the queue.
///
/// # Examples
///
/// ```
/// use crossbeam_deque::Injector;
///
/// let q = Injector::new();
///
/// assert_eq!(q.len(), 0);
/// q.push(1);
/// assert_eq!(q.len(), 1);
/// q.push(1);
/// assert_eq!(q.len(), 2);
/// ```
pub fn len(&self) -> usize {
loop {
// Load the tail index, then load the head index.
let mut tail = self.tail.index.load(Ordering::SeqCst);
let mut head = self.head.index.load(Ordering::SeqCst);
// If the tail index didn't change, we've got consistent indices to work with.
if self.tail.index.load(Ordering::SeqCst) == tail {
// Erase the lower bits.
tail &= !((1 << SHIFT) - 1);
head &= !((1 << SHIFT) - 1);
// Fix up indices if they fall onto block ends.
if (tail >> SHIFT) & (LAP - 1) == LAP - 1 {
tail = tail.wrapping_add(1 << SHIFT);
}
if (head >> SHIFT) & (LAP - 1) == LAP - 1 {
head = head.wrapping_add(1 << SHIFT);
}
// Rotate indices so that head falls into the first block.
let lap = (head >> SHIFT) / LAP;
tail = tail.wrapping_sub((lap * LAP) << SHIFT);
head = head.wrapping_sub((lap * LAP) << SHIFT);
// Remove the lower bits.
tail >>= SHIFT;
head >>= SHIFT;
// Return the difference minus the number of blocks between tail and head.
return tail - head - tail / LAP;
}
}
}
}
impl<T> Drop for Injector<T> {
fn drop(&mut self) {
let mut head = *self.head.index.get_mut();
let mut tail = *self.tail.index.get_mut();
let mut block = *self.head.block.get_mut();
// Erase the lower bits.
head &= !((1 << SHIFT) - 1);
tail &= !((1 << SHIFT) - 1);
unsafe {
// Drop all values between `head` and `tail` and deallocate the heap-allocated blocks.
while head != tail {
let offset = (head >> SHIFT) % LAP;
if offset < BLOCK_CAP {
// Drop the task in the slot.
let slot = (*block).slots.get_unchecked(offset);
let p = &mut *slot.task.get();
p.as_mut_ptr().drop_in_place();
} else {
// Deallocate the block and move to the next one.
let next = *(*block).next.get_mut();
drop(Box::from_raw(block));
block = next;
}
head = head.wrapping_add(1 << SHIFT);
}
// Deallocate the last remaining block.
drop(Box::from_raw(block));
}
}
}
impl<T> fmt::Debug for Injector<T> {
fn fmt(&self, f: &mut fmt::Formatter<'_>) -> fmt::Result {
f.pad("Worker { .. }")
}
}
/// Possible outcomes of a steal operation.
///
/// # Examples
///
/// There are lots of ways to chain results of steal operations together:
///
/// ```
/// use crossbeam_deque::Steal::{self, Empty, Retry, Success};
///
/// let collect = |v: Vec<Steal<i32>>| v.into_iter().collect::<Steal<i32>>();
///
/// assert_eq!(collect(vec![Empty, Empty, Empty]), Empty);
/// assert_eq!(collect(vec![Empty, Retry, Empty]), Retry);
/// assert_eq!(collect(vec![Retry, Success(1), Empty]), Success(1));
///
/// assert_eq!(collect(vec![Empty, Empty]).or_else(|| Retry), Retry);
/// assert_eq!(collect(vec![Retry, Empty]).or_else(|| Success(1)), Success(1));
/// ```
#[must_use]
#[derive(PartialEq, Eq, Copy, Clone)]
pub enum Steal<T> {
/// The queue was empty at the time of stealing.
Empty,
/// At least one task was successfully stolen.
Success(T),
/// The steal operation needs to be retried.
Retry,
}
impl<T> Steal<T> {
/// Returns `true` if the queue was empty at the time of stealing.
///
/// # Examples
///
/// ```
/// use crossbeam_deque::Steal::{Empty, Retry, Success};
///
/// assert!(!Success(7).is_empty());
/// assert!(!Retry::<i32>.is_empty());
///
/// assert!(Empty::<i32>.is_empty());
/// ```
pub fn is_empty(&self) -> bool {
match self {
Steal::Empty => true,
_ => false,
}
}
/// Returns `true` if at least one task was stolen.
///
/// # Examples
///
/// ```
/// use crossbeam_deque::Steal::{Empty, Retry, Success};
///
/// assert!(!Empty::<i32>.is_success());
/// assert!(!Retry::<i32>.is_success());
///
/// assert!(Success(7).is_success());
/// ```
pub fn is_success(&self) -> bool {
match self {
Steal::Success(_) => true,
_ => false,
}
}
/// Returns `true` if the steal operation needs to be retried.
///
/// # Examples
///
/// ```
/// use crossbeam_deque::Steal::{Empty, Retry, Success};
///
/// assert!(!Empty::<i32>.is_retry());
/// assert!(!Success(7).is_retry());
///
/// assert!(Retry::<i32>.is_retry());
/// ```
pub fn is_retry(&self) -> bool {
match self {
Steal::Retry => true,
_ => false,
}
}
/// Returns the result of the operation, if successful.
///
/// # Examples
///
/// ```
/// use crossbeam_deque::Steal::{Empty, Retry, Success};
///
/// assert_eq!(Empty::<i32>.success(), None);
/// assert_eq!(Retry::<i32>.success(), None);
///
/// assert_eq!(Success(7).success(), Some(7));
/// ```
pub fn success(self) -> Option<T> {
match self {
Steal::Success(res) => Some(res),
_ => None,
}
}
/// If no task was stolen, attempts another steal operation.
///
/// Returns this steal result if it is `Success`. Otherwise, closure `f` is invoked and then:
///
/// * If the second steal resulted in `Success`, it is returned.
/// * If both steals were unsuccessful but any resulted in `Retry`, then `Retry` is returned.
/// * If both resulted in `None`, then `None` is returned.
///
/// # Examples
///
/// ```
/// use crossbeam_deque::Steal::{Empty, Retry, Success};
///
/// assert_eq!(Success(1).or_else(|| Success(2)), Success(1));
/// assert_eq!(Retry.or_else(|| Success(2)), Success(2));
///
/// assert_eq!(Retry.or_else(|| Empty), Retry::<i32>);
/// assert_eq!(Empty.or_else(|| Retry), Retry::<i32>);
///
/// assert_eq!(Empty.or_else(|| Empty), Empty::<i32>);
/// ```
pub fn or_else<F>(self, f: F) -> Steal<T>
where
F: FnOnce() -> Steal<T>,
{
match self {
Steal::Empty => f(),
Steal::Success(_) => self,
Steal::Retry => {
if let Steal::Success(res) = f() {
Steal::Success(res)
} else {
Steal::Retry
}
}
}
}
}
impl<T> fmt::Debug for Steal<T> {
fn fmt(&self, f: &mut fmt::Formatter<'_>) -> fmt::Result {
match self {
Steal::Empty => f.pad("Empty"),
Steal::Success(_) => f.pad("Success(..)"),
Steal::Retry => f.pad("Retry"),
}
}
}
impl<T> FromIterator<Steal<T>> for Steal<T> {
/// Consumes items until a `Success` is found and returns it.
///
/// If no `Success` was found, but there was at least one `Retry`, then returns `Retry`.
/// Otherwise, `Empty` is returned.
fn from_iter<I>(iter: I) -> Steal<T>
where
I: IntoIterator<Item = Steal<T>>,
{
let mut retry = false;
for s in iter {
match &s {
Steal::Empty => {}
Steal::Success(_) => return s,
Steal::Retry => retry = true,
}
}
if retry {
Steal::Retry
} else {
Steal::Empty
}
}
}