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 2043 2044 2045 2046 2047 2048 2049 2050 2051 2052 2053 2054 2055 2056 2057 2058 2059 2060 2061 2062 2063 2064 2065 2066 2067 2068 2069 2070 2071 2072 2073 2074 2075 2076 2077 2078 2079 2080 2081 2082 2083 2084 2085 2086 2087 2088 2089 2090 2091 2092 2093 2094 2095 2096 2097 2098 2099 2100 2101 2102 2103 2104 2105 2106 2107 2108 2109 2110 2111 2112 2113 2114 2115 2116 2117 2118 2119 2120 2121 2122 2123 2124 2125 2126 2127 2128 2129 2130 2131 2132 2133 2134 2135 2136 2137 2138 2139 2140 2141 2142 2143 2144 2145 2146 2147 2148 2149 2150 2151 2152 2153 2154 2155 2156 2157 2158 2159 2160 2161 2162 2163 2164 2165 2166 2167 2168 2169 2170 2171 2172 2173 2174 2175 2176 2177 2178 2179 2180 2181 2182 2183 2184 2185 2186 2187 2188 2189 2190 2191 2192 2193 2194 2195 2196 2197 2198 2199 2200 2201 2202 2203 2204 2205 2206 2207 2208 2209 2210 2211 2212 2213 2214 2215 2216 2217 2218 2219 2220 2221 2222 2223 2224 2225 2226 2227 2228 2229 2230 2231 2232 2233 2234 2235 2236 2237 2238 2239 2240 2241 2242 2243 2244 2245 2246 2247 2248 2249 2250 2251 2252 2253 2254 2255 2256 2257 2258 2259 2260 2261 2262 2263 2264 2265 2266 2267 2268 2269 2270 2271 2272 2273 2274 2275 2276 2277 2278 2279 2280 2281 2282 2283 2284 2285 2286 2287 2288 2289 2290 2291 2292 2293 2294 2295 2296 2297 2298 2299 2300 2301 2302 2303 2304 2305 2306 2307 2308 2309 2310 2311 2312 2313 2314 2315 2316 2317 2318 2319 2320 2321 2322 2323 2324 2325 2326 2327 2328 2329 2330 2331 2332 2333 2334 2335 2336 2337 2338 2339 2340 2341 2342 2343 2344 2345 2346 2347 2348 2349 2350 2351 2352 2353 2354 2355 2356 2357 2358 2359 2360 2361 2362 2363 2364 2365 2366 2367 2368 2369 2370 2371 2372 2373 2374 2375 2376 2377 2378 2379 2380 2381 2382 2383 2384 2385 2386 2387 2388 2389 2390 2391 2392 2393 2394 2395 2396 2397 2398 2399 2400 2401 2402 2403 2404 2405 2406 2407 2408 2409 2410 2411 2412 2413 2414 2415 2416 2417 2418 2419 2420 2421 2422 2423 2424 2425 2426 2427 2428 2429 2430 2431 2432 2433 2434 2435 2436 2437 2438 2439 2440 2441 2442 2443 2444 2445 2446 2447 2448 2449 2450 2451 2452 2453 2454 2455 2456 2457 2458 2459 2460 2461 2462 2463 2464 2465 2466 2467 2468 2469 2470 2471 2472 2473 2474 2475 2476 2477 2478 2479 2480 2481 2482 2483 2484 2485 2486 2487 2488 2489 2490 2491 2492 2493 2494 2495 2496 2497 2498 2499 2500 2501 2502 2503 2504 2505 2506 2507 2508 2509 2510 2511 2512 2513 2514 2515 2516 2517 2518 2519 2520 2521 2522 2523 2524 2525 2526 2527 2528 2529 2530 2531 2532 2533 2534 2535 2536 2537 2538 2539 2540 2541 2542 2543 2544 2545 2546 2547 2548 2549 2550 2551 2552 2553 2554 2555 2556 2557 2558 2559 2560 2561 2562 2563 2564 2565 2566 2567 2568 2569 2570 2571 2572 2573 2574 2575 2576 2577 2578 2579 2580 2581 2582 2583 2584 2585 2586 2587 2588 2589 2590 2591 2592 2593 2594 2595 2596 2597 2598 2599 2600 2601 2602 2603 2604 2605 2606 2607 2608 2609 2610 2611 2612 2613 2614 2615 2616 2617 2618 2619 2620 2621 2622 2623 2624 2625 2626 2627 2628 2629 2630 2631 2632 2633 2634 2635 2636 2637 2638 2639 2640 2641 2642 2643 2644 2645 2646 2647 2648 2649 2650 2651 2652 2653 2654 2655 2656 2657 2658 2659 2660 2661 2662 2663 2664 2665 2666 2667 2668 2669 2670 2671 2672 2673 2674 2675 2676 2677 2678 2679 2680 2681 2682 2683 2684 2685 2686 2687 2688 2689 2690 2691 2692 2693 2694 2695 2696 2697 2698 2699 2700 2701 2702 2703 2704 2705 2706 2707 2708 2709 2710 2711 2712 2713 2714 2715 2716 2717 2718 2719 2720 2721 2722 2723 2724 2725 2726 2727 2728 2729 2730 2731 2732 2733 2734 2735 2736 2737 2738 2739 2740 2741 2742 2743 2744 2745 2746 2747 2748 2749 2750 2751 2752 2753 2754 2755 2756 2757 2758 2759 2760 2761 2762 2763 2764 2765 2766 2767 2768 2769 2770 2771 2772 2773 2774 2775 2776 2777 2778 2779 2780 2781 2782 2783 2784 2785 2786 2787 2788 2789 2790 2791 2792 2793 2794 2795 2796 2797 2798 2799 2800 2801 2802 2803 2804 2805 2806 2807 2808 2809 2810 2811 2812 2813 2814 2815 2816 2817 2818 2819 2820 2821 2822 2823 2824 2825 2826 2827 2828 2829 2830 2831 2832 2833 2834 2835 2836 2837 2838 2839 2840 2841 2842 2843 2844 2845 2846 2847 2848 2849 2850 2851 2852 2853 2854 2855 2856 2857 2858 2859 2860 2861 2862 2863 2864 2865 2866 2867 2868 2869 2870 2871 2872 2873 2874 2875 2876 2877 2878 2879 2880 2881 2882 2883 2884 2885 2886 2887 2888 2889 2890 2891 2892 2893 2894 2895 2896 2897 2898 2899 2900 2901 2902 2903 2904 2905 2906 2907 2908 2909 2910 2911 2912 2913 2914 2915 2916 2917 2918 2919 2920 2921 2922 2923 2924 2925 2926 2927 2928 2929 2930 2931 2932 2933 2934 2935 2936 2937 2938 2939 2940 2941 2942 2943 2944 2945 2946 2947 2948 2949 2950 2951 2952 2953 2954 2955 2956 2957 2958 2959 2960 2961 2962 2963 2964 2965 2966 2967 2968 2969 2970 2971 2972 2973 2974 2975 2976 2977 2978 2979 2980 2981 2982 2983 2984 2985 2986 2987 2988 2989 2990 2991 2992 2993 2994 2995 2996 2997 2998 2999 3000 3001 3002 3003 3004 3005 3006 3007 3008 3009 3010 3011 3012 3013 3014 3015 3016 3017 3018 3019 3020 3021 3022 3023 3024 3025 3026 3027 3028 3029 3030 3031 3032 3033 3034 3035 3036 3037 3038 3039 3040 3041 3042 3043 3044 3045 3046 3047 3048 3049 3050 3051 3052 3053 3054 3055 3056 3057 3058 3059 3060 3061 3062 3063 3064 3065 3066 3067 3068 3069 3070 3071 3072 3073 3074 3075 3076 3077 3078 3079 3080 3081 3082 3083 3084 3085 3086 3087 3088 3089 3090 3091 3092 3093 3094 3095 3096 3097 3098 3099 3100 3101 3102 3103 3104 3105 3106 3107 3108 3109 3110 3111 3112 3113 3114 3115 3116 3117 3118 3119 3120 3121 3122 3123 3124 3125 3126 3127 3128 3129 3130 3131 3132 3133 3134 3135 3136 3137 3138 3139 3140 3141 3142 3143 3144 3145 3146 3147 3148 3149 3150 3151 3152 3153 3154 3155 3156 3157 3158 3159 3160 3161 3162 3163 3164 3165 3166 3167 3168 3169 3170 3171 3172 3173 3174 3175 3176 3177 3178 3179 3180 3181 3182 3183 3184 3185 3186 3187 3188 3189 3190 3191 3192 3193 3194 3195 3196 3197 3198 3199 3200 3201 3202 3203 3204 3205 3206 3207 3208 3209 3210 3211 3212 3213 3214 3215 3216 3217 3218 3219 3220 3221 3222 3223 3224 3225 3226 3227 3228 3229 3230 3231 3232 3233 3234 3235 3236 3237 3238 3239 3240 3241 3242 3243 3244 3245 3246 3247 3248 3249 3250 3251 3252 3253 3254 3255 3256 3257 3258 3259 3260 3261 3262 3263 3264 3265 3266 3267 3268 3269 3270 3271 3272 3273 3274 3275 3276 3277 3278 3279 3280 3281 3282 3283 3284 3285 3286 3287 3288 3289 3290 3291 3292 3293 3294 3295 3296 3297 3298 3299 3300 3301 3302 3303 3304 3305 3306 3307 3308 3309 3310 3311 3312 3313 3314 3315 3316 3317 3318 3319 3320 3321 3322 3323 3324 3325 3326 3327 3328 3329 3330 3331 3332 3333 3334 3335 3336 3337 3338 3339 3340 3341 3342 3343 3344 3345 3346 3347 3348 3349 3350 3351 3352 3353 3354 3355 3356 3357 3358 3359 3360 3361 3362 3363 3364 3365 3366 3367 3368 3369 3370 3371 3372 3373 3374 3375 3376 3377 3378 3379 3380 3381 3382 3383 3384 3385 3386 3387 3388 3389 3390 3391 3392 3393 3394 3395 3396 3397 3398 3399 3400 3401 3402 3403 3404 3405 3406 3407 3408 3409 3410 3411 3412 3413 3414 3415 3416 3417 3418 3419 3420 3421 3422 3423 3424 3425 3426 3427 3428 3429 3430 3431 3432 3433 3434 3435 3436 3437 3438 3439 3440 3441 3442 3443 3444 3445 3446 3447 3448 3449 3450 3451 3452 3453 3454 3455 3456 3457 3458 3459 3460 3461 3462 3463 3464 3465 3466 3467 3468 3469 3470 3471 3472 3473 3474 3475 3476 3477 3478 3479 3480 3481 3482 3483 3484 3485 3486 3487 3488 3489 3490 3491 3492 3493 3494 3495 3496 3497 3498 3499 3500 3501 3502 3503 3504 3505 3506 3507 3508 3509 3510 3511 3512 3513 3514 3515 3516 3517 3518 3519 3520 3521 3522 3523 3524 3525 3526 3527 3528 3529 3530 3531 3532 3533 3534 3535 3536 3537 3538 3539 3540 3541 3542 3543 3544 3545 3546 3547 3548 3549 3550 3551 3552 3553 3554 3555 3556 3557 3558 3559 3560 3561 3562 3563 3564 3565 3566 3567 3568 3569 3570 3571 3572 3573 3574 3575 3576 3577 3578 3579 3580 3581 3582 3583 3584 3585 3586 3587 3588 3589 3590 3591 3592 3593 3594 3595 3596 3597 3598 3599 3600 3601 3602 3603 3604 3605 3606 3607 3608 3609 3610 3611 3612 3613 3614 3615 3616 3617 3618 3619 3620 3621 3622 3623 3624 3625 3626 3627 3628 3629 3630 3631 3632 3633 3634 3635 3636 3637 3638 3639 3640 3641 3642 3643 3644 3645 3646 3647 3648 3649 3650 3651 3652 3653 3654 3655 3656 3657 3658 3659 3660 3661 3662 3663 3664 3665 3666 3667 3668 3669 3670 3671 3672 3673 3674 3675 3676 3677 3678 3679 3680 3681 3682 3683 3684 3685 3686 3687 3688 3689 3690 3691 3692 3693 3694 3695 3696 3697 3698 3699 3700 3701 3702 3703 3704 3705 3706 3707 3708 3709 3710 3711 3712 3713 3714 3715 3716 3717 3718 3719 3720 3721 3722 3723 3724 3725 3726 3727 3728 3729 3730 3731 3732 3733 3734 3735 3736 3737 3738 3739 3740 3741 3742 3743 3744 3745 3746 3747 3748 3749 3750 3751 3752 3753 3754 3755 3756 3757 3758 3759 3760 3761 3762 3763 3764 3765 3766 3767 3768 3769 3770 3771 3772 3773 3774 3775 3776 3777 3778 3779 3780 3781 3782 3783 3784 3785 3786 3787 3788 3789 3790 3791 3792 3793 3794 3795 3796 3797 3798 3799 3800 3801 3802 3803 3804 3805 3806 3807 3808 3809 3810 3811 3812 3813 3814 3815 3816 3817 3818 3819 3820 3821 3822 3823 3824 3825 3826 3827 3828 3829 3830 3831 3832 3833 3834 3835 3836 3837 3838 3839 3840 3841 3842 3843 3844 3845 3846 3847 3848 3849 3850 3851 3852 3853 3854 3855 3856 3857 3858 3859 3860 3861 3862 3863 3864 3865 3866 3867 3868 3869 3870 3871 3872 3873 3874 3875 3876 3877 3878 3879 3880 3881 3882 3883 3884 3885 3886 3887 3888 3889 3890 3891 3892 3893 3894 3895 3896 3897 3898 3899 3900 3901 3902 3903 3904 3905 3906 3907 3908 3909 3910 3911 3912 3913 3914 3915 3916 3917 3918 3919 3920 3921 3922 3923 3924 3925 3926 3927 3928 3929 3930 3931 3932 3933 3934 3935 3936 3937 3938 3939 3940 3941 3942 3943 3944 3945 3946 3947 3948 3949 3950 3951 3952 3953 3954 3955 3956 3957 3958 3959 3960 3961 3962 3963 3964 3965 3966 3967 3968 3969 3970 3971 3972 3973 3974 3975 3976 3977 3978 3979 3980 3981 3982 3983 3984 3985 3986 3987 3988 3989 3990 3991 3992 3993 3994 3995 3996 3997 3998 3999 4000 4001 4002 4003 4004 4005 4006 4007 4008 4009 4010 4011 4012 4013 4014 4015 4016 4017 4018 4019 4020 4021 4022 4023 4024 4025 4026 4027 4028 4029 4030 4031 4032 4033 4034 4035 4036 4037 4038 4039 4040 4041 4042 4043 4044 4045 4046 4047 4048 4049 4050 4051 4052 4053 4054 4055 4056 4057 4058 4059 4060 4061 4062 4063 4064 4065 4066 4067 4068 4069 4070 4071 4072 4073 4074 4075 4076 4077 4078 4079 4080 4081 4082 4083 4084 4085 4086 4087 4088 4089 4090 4091 4092 4093 4094 4095 4096 4097 4098 4099 4100 4101 4102 4103 4104 4105 4106 4107 4108 4109 4110 4111 4112 4113 4114 4115 4116 4117 4118 4119 4120 4121 4122 4123 4124 4125 4126 4127 4128 4129 4130 4131 4132 4133 4134 4135 4136 4137 4138 4139 4140 4141 4142 4143 4144 4145 4146 4147 4148 4149 4150 4151 4152 4153 4154 4155 4156 4157 4158 4159 4160 4161 4162 4163 4164 4165 4166 4167 4168 4169 4170 4171 4172 4173 4174 4175 4176 4177 4178 4179 4180 4181 4182 4183 4184 4185 4186 4187 4188 4189 4190 4191 4192 4193 4194 4195 4196 4197 4198 4199 4200 4201 4202 4203 4204 4205 4206 4207 4208 4209 4210 4211 4212 4213 4214 4215 4216 4217 4218 4219 4220 4221 4222 4223 4224 4225 4226 4227 4228 4229 4230 4231 4232 4233 4234 4235 4236 4237 4238 4239 4240 4241 4242 4243 4244 4245 4246 4247 4248 4249 4250 4251 4252 4253 4254 4255 4256 4257 4258 4259 4260 4261 4262 4263 4264 4265 4266 4267 4268 4269 4270 4271 4272 4273 4274 4275 4276 4277 4278 4279 4280 4281 4282 4283 4284 4285 4286 4287 4288 4289 4290 4291 4292 4293 4294 4295 4296 4297 4298 4299 4300 4301 4302 4303 4304 4305 4306 4307 4308 4309 4310 4311 4312 4313 4314 4315 4316 4317 4318 4319 4320 4321 4322 4323 4324 4325 4326 4327 4328 4329 4330 4331 4332 4333 4334 4335 4336 4337 4338 4339 4340 4341 4342 4343 4344 4345 4346 4347 4348 4349 4350 4351 4352 4353 4354 4355 4356 4357 4358 4359 4360 4361 4362 4363 4364 4365 4366 4367 4368 4369 4370 4371 4372 4373 4374 4375 4376 4377 4378 4379 4380 4381 4382 4383 4384 4385 4386 4387 4388 4389 4390 4391 4392 4393 4394 4395 4396 4397 4398 4399 4400 4401 4402 4403 4404 4405 4406 4407 4408 4409 4410 4411 4412 4413 4414 4415 4416 4417 4418 4419 4420 4421 4422 4423 4424 4425 4426 4427 4428 4429 4430 4431 4432 4433 4434 4435 4436 4437 4438 4439 4440 4441 4442 4443 4444 4445 4446 4447 4448 4449 4450 4451 4452 4453 4454 4455 4456 4457 4458 4459 4460 4461 4462 4463 4464 4465 4466 4467 4468 4469 4470 4471 4472 4473 4474 4475 4476 4477 4478 4479 4480 4481 4482 4483 4484 4485 4486 4487 4488 4489 4490 4491 4492 4493 4494 4495 4496 4497 4498 4499 4500 4501 4502 4503 4504 4505 4506 4507 4508 4509 4510 4511 4512 4513 4514 4515 4516 4517 4518 4519 4520 4521 4522 4523 4524 4525 4526 4527 4528 4529 4530 4531 4532 4533 4534 4535 4536 4537 4538 4539 4540 4541 4542 4543 4544 4545 4546 4547 4548 4549 4550 4551 4552 4553 4554 4555 4556 4557 4558 4559 4560 4561 4562 4563 4564 4565 4566 4567 4568 4569 4570 4571 4572 4573 4574 4575 4576 4577 4578 4579 4580 4581 4582 4583 4584 4585 4586 4587 4588 4589 4590 4591 4592 4593 4594 4595 4596 4597 4598 4599 4600 4601 4602 4603 4604 4605 4606 4607 4608 4609 4610 4611 4612 4613 4614 4615 4616 4617 4618 4619 4620 4621 4622 4623 4624 4625 4626 4627 4628 4629 4630 4631 4632 4633 4634 4635 4636 4637 4638 4639 4640 4641 4642 4643 4644 4645 4646 4647 4648 4649 4650 4651 4652 4653 4654 4655 4656 4657 4658 4659 4660 4661 4662 4663 4664 4665 4666 4667 4668 4669 4670 4671 4672 4673 4674 4675 4676 4677 4678 4679 4680 4681 4682 4683 4684 4685 4686 4687 4688 4689 4690 4691 4692 4693 4694 4695 4696 4697 4698 4699 4700 4701 4702 4703 4704 4705 4706 4707 4708 4709 4710 4711 4712 4713 4714 4715 4716 4717 4718 4719 4720 4721 4722 4723 4724 4725 4726 4727 4728 4729 4730 4731 4732 4733 4734 4735 4736 4737 4738 4739 4740 4741 4742 4743 4744 4745 4746 4747 4748 4749 4750 4751 4752 4753 4754 4755 4756 4757 4758 4759 4760 4761 4762 4763 4764 4765 4766 4767 4768 4769 4770 4771 4772 4773 4774 4775 4776 4777 4778 4779 4780 4781 4782 4783 4784 4785 4786 4787 4788 4789 4790 4791 4792 4793 4794 4795 4796 4797 4798 4799 4800 4801 4802 4803 4804 4805 4806 4807 4808 4809 4810 4811 4812 4813 4814 4815 4816 4817 4818 4819 4820 4821 4822
use crate::alloc::alloc::{handle_alloc_error, Layout};
use crate::scopeguard::{guard, ScopeGuard};
use crate::TryReserveError;
use core::iter::FusedIterator;
use core::marker::PhantomData;
use core::mem;
use core::mem::MaybeUninit;
use core::ptr::NonNull;
use core::{hint, ptr};
cfg_if! {
// Use the SSE2 implementation if possible: it allows us to scan 16 buckets
// at once instead of 8. We don't bother with AVX since it would require
// runtime dispatch and wouldn't gain us much anyways: the probability of
// finding a match drops off drastically after the first few buckets.
//
// I attempted an implementation on ARM using NEON instructions, but it
// turns out that most NEON instructions have multi-cycle latency, which in
// the end outweighs any gains over the generic implementation.
if #[cfg(all(
target_feature = "sse2",
any(target_arch = "x86", target_arch = "x86_64"),
not(miri),
))] {
mod sse2;
use sse2 as imp;
} else if #[cfg(all(
target_arch = "aarch64",
target_feature = "neon",
// NEON intrinsics are currently broken on big-endian targets.
// See https://github.com/rust-lang/stdarch/issues/1484.
target_endian = "little",
not(miri),
))] {
mod neon;
use neon as imp;
} else {
mod generic;
use generic as imp;
}
}
mod alloc;
pub(crate) use self::alloc::{do_alloc, Allocator, Global};
mod bitmask;
use self::bitmask::BitMaskIter;
use self::imp::Group;
// Branch prediction hint. This is currently only available on nightly but it
// consistently improves performance by 10-15%.
#[cfg(not(feature = "nightly"))]
use core::convert::identity as likely;
#[cfg(not(feature = "nightly"))]
use core::convert::identity as unlikely;
#[cfg(feature = "nightly")]
use core::intrinsics::{likely, unlikely};
// FIXME: use strict provenance functions once they are stable.
// Implement it with a transmute for now.
#[inline(always)]
#[allow(clippy::useless_transmute)] // clippy is wrong, cast and transmute are different here
fn invalid_mut<T>(addr: usize) -> *mut T {
unsafe { core::mem::transmute(addr) }
}
#[inline]
unsafe fn offset_from<T>(to: *const T, from: *const T) -> usize {
to.offset_from(from) as usize
}
/// Whether memory allocation errors should return an error or abort.
#[derive(Copy, Clone)]
enum Fallibility {
Fallible,
Infallible,
}
impl Fallibility {
/// Error to return on capacity overflow.
#[cfg_attr(feature = "inline-more", inline)]
fn capacity_overflow(self) -> TryReserveError {
match self {
Fallibility::Fallible => TryReserveError::CapacityOverflow,
Fallibility::Infallible => panic!("Hash table capacity overflow"),
}
}
/// Error to return on allocation error.
#[cfg_attr(feature = "inline-more", inline)]
fn alloc_err(self, layout: Layout) -> TryReserveError {
match self {
Fallibility::Fallible => TryReserveError::AllocError { layout },
Fallibility::Infallible => handle_alloc_error(layout),
}
}
}
trait SizedTypeProperties: Sized {
const IS_ZERO_SIZED: bool = mem::size_of::<Self>() == 0;
const NEEDS_DROP: bool = mem::needs_drop::<Self>();
}
impl<T> SizedTypeProperties for T {}
/// Control byte value for an empty bucket.
const EMPTY: u8 = 0b1111_1111;
/// Control byte value for a deleted bucket.
const DELETED: u8 = 0b1000_0000;
/// Checks whether a control byte represents a full bucket (top bit is clear).
#[inline]
fn is_full(ctrl: u8) -> bool {
ctrl & 0x80 == 0
}
/// Checks whether a control byte represents a special value (top bit is set).
#[inline]
fn is_special(ctrl: u8) -> bool {
ctrl & 0x80 != 0
}
/// Checks whether a special control value is EMPTY (just check 1 bit).
#[inline]
fn special_is_empty(ctrl: u8) -> bool {
debug_assert!(is_special(ctrl));
ctrl & 0x01 != 0
}
/// Primary hash function, used to select the initial bucket to probe from.
#[inline]
#[allow(clippy::cast_possible_truncation)]
fn h1(hash: u64) -> usize {
// On 32-bit platforms we simply ignore the higher hash bits.
hash as usize
}
// Constant for h2 function that grabing the top 7 bits of the hash.
const MIN_HASH_LEN: usize = if mem::size_of::<usize>() < mem::size_of::<u64>() {
mem::size_of::<usize>()
} else {
mem::size_of::<u64>()
};
/// Secondary hash function, saved in the low 7 bits of the control byte.
#[inline]
#[allow(clippy::cast_possible_truncation)]
fn h2(hash: u64) -> u8 {
// Grab the top 7 bits of the hash. While the hash is normally a full 64-bit
// value, some hash functions (such as FxHash) produce a usize result
// instead, which means that the top 32 bits are 0 on 32-bit platforms.
// So we use MIN_HASH_LEN constant to handle this.
let top7 = hash >> (MIN_HASH_LEN * 8 - 7);
(top7 & 0x7f) as u8 // truncation
}
/// Probe sequence based on triangular numbers, which is guaranteed (since our
/// table size is a power of two) to visit every group of elements exactly once.
///
/// A triangular probe has us jump by 1 more group every time. So first we
/// jump by 1 group (meaning we just continue our linear scan), then 2 groups
/// (skipping over 1 group), then 3 groups (skipping over 2 groups), and so on.
///
/// Proof that the probe will visit every group in the table:
/// <https://fgiesen.wordpress.com/2015/02/22/triangular-numbers-mod-2n/>
struct ProbeSeq {
pos: usize,
stride: usize,
}
impl ProbeSeq {
#[inline]
fn move_next(&mut self, bucket_mask: usize) {
// We should have found an empty bucket by now and ended the probe.
debug_assert!(
self.stride <= bucket_mask,
"Went past end of probe sequence"
);
self.stride += Group::WIDTH;
self.pos += self.stride;
self.pos &= bucket_mask;
}
}
/// Returns the number of buckets needed to hold the given number of items,
/// taking the maximum load factor into account.
///
/// Returns `None` if an overflow occurs.
// Workaround for emscripten bug emscripten-core/emscripten-fastcomp#258
#[cfg_attr(target_os = "emscripten", inline(never))]
#[cfg_attr(not(target_os = "emscripten"), inline)]
fn capacity_to_buckets(cap: usize) -> Option<usize> {
debug_assert_ne!(cap, 0);
// For small tables we require at least 1 empty bucket so that lookups are
// guaranteed to terminate if an element doesn't exist in the table.
if cap < 8 {
// We don't bother with a table size of 2 buckets since that can only
// hold a single element. Instead we skip directly to a 4 bucket table
// which can hold 3 elements.
return Some(if cap < 4 { 4 } else { 8 });
}
// Otherwise require 1/8 buckets to be empty (87.5% load)
//
// Be careful when modifying this, calculate_layout relies on the
// overflow check here.
let adjusted_cap = cap.checked_mul(8)? / 7;
// Any overflows will have been caught by the checked_mul. Also, any
// rounding errors from the division above will be cleaned up by
// next_power_of_two (which can't overflow because of the previous division).
Some(adjusted_cap.next_power_of_two())
}
/// Returns the maximum effective capacity for the given bucket mask, taking
/// the maximum load factor into account.
#[inline]
fn bucket_mask_to_capacity(bucket_mask: usize) -> usize {
if bucket_mask < 8 {
// For tables with 1/2/4/8 buckets, we always reserve one empty slot.
// Keep in mind that the bucket mask is one less than the bucket count.
bucket_mask
} else {
// For larger tables we reserve 12.5% of the slots as empty.
((bucket_mask + 1) / 8) * 7
}
}
/// Helper which allows the max calculation for ctrl_align to be statically computed for each T
/// while keeping the rest of `calculate_layout_for` independent of `T`
#[derive(Copy, Clone)]
struct TableLayout {
size: usize,
ctrl_align: usize,
}
impl TableLayout {
#[inline]
const fn new<T>() -> Self {
let layout = Layout::new::<T>();
Self {
size: layout.size(),
ctrl_align: if layout.align() > Group::WIDTH {
layout.align()
} else {
Group::WIDTH
},
}
}
#[inline]
fn calculate_layout_for(self, buckets: usize) -> Option<(Layout, usize)> {
debug_assert!(buckets.is_power_of_two());
let TableLayout { size, ctrl_align } = self;
// Manual layout calculation since Layout methods are not yet stable.
let ctrl_offset =
size.checked_mul(buckets)?.checked_add(ctrl_align - 1)? & !(ctrl_align - 1);
let len = ctrl_offset.checked_add(buckets + Group::WIDTH)?;
// We need an additional check to ensure that the allocation doesn't
// exceed `isize::MAX` (https://github.com/rust-lang/rust/pull/95295).
if len > isize::MAX as usize - (ctrl_align - 1) {
return None;
}
Some((
unsafe { Layout::from_size_align_unchecked(len, ctrl_align) },
ctrl_offset,
))
}
}
/// A reference to an empty bucket into which an can be inserted.
pub struct InsertSlot {
index: usize,
}
/// A reference to a hash table bucket containing a `T`.
///
/// This is usually just a pointer to the element itself. However if the element
/// is a ZST, then we instead track the index of the element in the table so
/// that `erase` works properly.
pub struct Bucket<T> {
// Actually it is pointer to next element than element itself
// this is needed to maintain pointer arithmetic invariants
// keeping direct pointer to element introduces difficulty.
// Using `NonNull` for variance and niche layout
ptr: NonNull<T>,
}
// This Send impl is needed for rayon support. This is safe since Bucket is
// never exposed in a public API.
unsafe impl<T> Send for Bucket<T> {}
impl<T> Clone for Bucket<T> {
#[inline]
fn clone(&self) -> Self {
Self { ptr: self.ptr }
}
}
impl<T> Bucket<T> {
/// Creates a [`Bucket`] that contain pointer to the data.
/// The pointer calculation is performed by calculating the
/// offset from given `base` pointer (convenience for
/// `base.as_ptr().sub(index)`).
///
/// `index` is in units of `T`; e.g., an `index` of 3 represents a pointer
/// offset of `3 * size_of::<T>()` bytes.
///
/// If the `T` is a ZST, then we instead track the index of the element
/// in the table so that `erase` works properly (return
/// `NonNull::new_unchecked((index + 1) as *mut T)`)
///
/// # Safety
///
/// If `mem::size_of::<T>() != 0`, then the safety rules are directly derived
/// from the safety rules for [`<*mut T>::sub`] method of `*mut T` and the safety
/// rules of [`NonNull::new_unchecked`] function.
///
/// Thus, in order to uphold the safety contracts for the [`<*mut T>::sub`] method
/// and [`NonNull::new_unchecked`] function, as well as for the correct
/// logic of the work of this crate, the following rules are necessary and
/// sufficient:
///
/// * the `base` pointer must not be `dangling` and must points to the
/// end of the first `value element` from the `data part` of the table, i.e.
/// must be the pointer that returned by [`RawTable::data_end`] or by
/// [`RawTableInner::data_end<T>`];
///
/// * `index` must not be greater than `RawTableInner.bucket_mask`, i.e.
/// `index <= RawTableInner.bucket_mask` or, in other words, `(index + 1)`
/// must be no greater than the number returned by the function
/// [`RawTable::buckets`] or [`RawTableInner::buckets`].
///
/// If `mem::size_of::<T>() == 0`, then the only requirement is that the
/// `index` must not be greater than `RawTableInner.bucket_mask`, i.e.
/// `index <= RawTableInner.bucket_mask` or, in other words, `(index + 1)`
/// must be no greater than the number returned by the function
/// [`RawTable::buckets`] or [`RawTableInner::buckets`].
///
/// [`Bucket`]: crate::raw::Bucket
/// [`<*mut T>::sub`]: https://doc.rust-lang.org/core/primitive.pointer.html#method.sub-1
/// [`NonNull::new_unchecked`]: https://doc.rust-lang.org/stable/std/ptr/struct.NonNull.html#method.new_unchecked
/// [`RawTable::data_end`]: crate::raw::RawTable::data_end
/// [`RawTableInner::data_end<T>`]: RawTableInner::data_end<T>
/// [`RawTable::buckets`]: crate::raw::RawTable::buckets
/// [`RawTableInner::buckets`]: RawTableInner::buckets
#[inline]
unsafe fn from_base_index(base: NonNull<T>, index: usize) -> Self {
// If mem::size_of::<T>() != 0 then return a pointer to an `element` in
// the data part of the table (we start counting from "0", so that
// in the expression T[last], the "last" index actually one less than the
// "buckets" number in the table, i.e. "last = RawTableInner.bucket_mask"):
//
// `from_base_index(base, 1).as_ptr()` returns a pointer that
// points here in the data part of the table
// (to the start of T1)
// |
// | `base: NonNull<T>` must point here
// | (to the end of T0 or to the start of C0)
// v v
// [Padding], Tlast, ..., |T1|, T0, |C0, C1, ..., Clast
// ^
// `from_base_index(base, 1)` returns a pointer
// that points here in the data part of the table
// (to the end of T1)
//
// where: T0...Tlast - our stored data; C0...Clast - control bytes
// or metadata for data.
let ptr = if T::IS_ZERO_SIZED {
// won't overflow because index must be less than length (bucket_mask)
// and bucket_mask is guaranteed to be less than `isize::MAX`
// (see TableLayout::calculate_layout_for method)
invalid_mut(index + 1)
} else {
base.as_ptr().sub(index)
};
Self {
ptr: NonNull::new_unchecked(ptr),
}
}
/// Calculates the index of a [`Bucket`] as distance between two pointers
/// (convenience for `base.as_ptr().offset_from(self.ptr.as_ptr()) as usize`).
/// The returned value is in units of T: the distance in bytes divided by
/// [`core::mem::size_of::<T>()`].
///
/// If the `T` is a ZST, then we return the index of the element in
/// the table so that `erase` works properly (return `self.ptr.as_ptr() as usize - 1`).
///
/// This function is the inverse of [`from_base_index`].
///
/// # Safety
///
/// If `mem::size_of::<T>() != 0`, then the safety rules are directly derived
/// from the safety rules for [`<*const T>::offset_from`] method of `*const T`.
///
/// Thus, in order to uphold the safety contracts for [`<*const T>::offset_from`]
/// method, as well as for the correct logic of the work of this crate, the
/// following rules are necessary and sufficient:
///
/// * `base` contained pointer must not be `dangling` and must point to the
/// end of the first `element` from the `data part` of the table, i.e.
/// must be a pointer that returns by [`RawTable::data_end`] or by
/// [`RawTableInner::data_end<T>`];
///
/// * `self` also must not contain dangling pointer;
///
/// * both `self` and `base` must be created from the same [`RawTable`]
/// (or [`RawTableInner`]).
///
/// If `mem::size_of::<T>() == 0`, this function is always safe.
///
/// [`Bucket`]: crate::raw::Bucket
/// [`from_base_index`]: crate::raw::Bucket::from_base_index
/// [`RawTable::data_end`]: crate::raw::RawTable::data_end
/// [`RawTableInner::data_end<T>`]: RawTableInner::data_end<T>
/// [`RawTable`]: crate::raw::RawTable
/// [`RawTableInner`]: RawTableInner
/// [`<*const T>::offset_from`]: https://doc.rust-lang.org/nightly/core/primitive.pointer.html#method.offset_from
#[inline]
unsafe fn to_base_index(&self, base: NonNull<T>) -> usize {
// If mem::size_of::<T>() != 0 then return an index under which we used to store the
// `element` in the data part of the table (we start counting from "0", so
// that in the expression T[last], the "last" index actually is one less than the
// "buckets" number in the table, i.e. "last = RawTableInner.bucket_mask").
// For example for 5th element in table calculation is performed like this:
//
// mem::size_of::<T>()
// |
// | `self = from_base_index(base, 5)` that returns pointer
// | that points here in tha data part of the table
// | (to the end of T5)
// | | `base: NonNull<T>` must point here
// v | (to the end of T0 or to the start of C0)
// /???\ v v
// [Padding], Tlast, ..., |T10|, ..., T5|, T4, T3, T2, T1, T0, |C0, C1, C2, C3, C4, C5, ..., C10, ..., Clast
// \__________ __________/
// \/
// `bucket.to_base_index(base)` = 5
// (base.as_ptr() as usize - self.ptr.as_ptr() as usize) / mem::size_of::<T>()
//
// where: T0...Tlast - our stored data; C0...Clast - control bytes or metadata for data.
if T::IS_ZERO_SIZED {
// this can not be UB
self.ptr.as_ptr() as usize - 1
} else {
offset_from(base.as_ptr(), self.ptr.as_ptr())
}
}
/// Acquires the underlying raw pointer `*mut T` to `data`.
///
/// # Note
///
/// If `T` is not [`Copy`], do not use `*mut T` methods that can cause calling the
/// destructor of `T` (for example the [`<*mut T>::drop_in_place`] method), because
/// for properly dropping the data we also need to clear `data` control bytes. If we
/// drop data, but do not clear `data control byte` it leads to double drop when
/// [`RawTable`] goes out of scope.
///
/// If you modify an already initialized `value`, so [`Hash`] and [`Eq`] on the new
/// `T` value and its borrowed form *must* match those for the old `T` value, as the map
/// will not re-evaluate where the new value should go, meaning the value may become
/// "lost" if their location does not reflect their state.
///
/// [`RawTable`]: crate::raw::RawTable
/// [`<*mut T>::drop_in_place`]: https://doc.rust-lang.org/core/primitive.pointer.html#method.drop_in_place
/// [`Hash`]: https://doc.rust-lang.org/core/hash/trait.Hash.html
/// [`Eq`]: https://doc.rust-lang.org/core/cmp/trait.Eq.html
///
/// # Examples
///
/// ```
/// # #[cfg(feature = "raw")]
/// # fn test() {
/// use core::hash::{BuildHasher, Hash};
/// use hashbrown::raw::{Bucket, RawTable};
///
/// type NewHashBuilder = core::hash::BuildHasherDefault<ahash::AHasher>;
///
/// fn make_hash<K: Hash + ?Sized, S: BuildHasher>(hash_builder: &S, key: &K) -> u64 {
/// use core::hash::Hasher;
/// let mut state = hash_builder.build_hasher();
/// key.hash(&mut state);
/// state.finish()
/// }
///
/// let hash_builder = NewHashBuilder::default();
/// let mut table = RawTable::new();
///
/// let value = ("a", 100);
/// let hash = make_hash(&hash_builder, &value.0);
///
/// table.insert(hash, value.clone(), |val| make_hash(&hash_builder, &val.0));
///
/// let bucket: Bucket<(&str, i32)> = table.find(hash, |(k1, _)| k1 == &value.0).unwrap();
///
/// assert_eq!(unsafe { &*bucket.as_ptr() }, &("a", 100));
/// # }
/// # fn main() {
/// # #[cfg(feature = "raw")]
/// # test()
/// # }
/// ```
#[inline]
pub fn as_ptr(&self) -> *mut T {
if T::IS_ZERO_SIZED {
// Just return an arbitrary ZST pointer which is properly aligned
// invalid pointer is good enough for ZST
invalid_mut(mem::align_of::<T>())
} else {
unsafe { self.ptr.as_ptr().sub(1) }
}
}
/// Create a new [`Bucket`] that is offset from the `self` by the given
/// `offset`. The pointer calculation is performed by calculating the
/// offset from `self` pointer (convenience for `self.ptr.as_ptr().sub(offset)`).
/// This function is used for iterators.
///
/// `offset` is in units of `T`; e.g., a `offset` of 3 represents a pointer
/// offset of `3 * size_of::<T>()` bytes.
///
/// # Safety
///
/// If `mem::size_of::<T>() != 0`, then the safety rules are directly derived
/// from the safety rules for [`<*mut T>::sub`] method of `*mut T` and safety
/// rules of [`NonNull::new_unchecked`] function.
///
/// Thus, in order to uphold the safety contracts for [`<*mut T>::sub`] method
/// and [`NonNull::new_unchecked`] function, as well as for the correct
/// logic of the work of this crate, the following rules are necessary and
/// sufficient:
///
/// * `self` contained pointer must not be `dangling`;
///
/// * `self.to_base_index() + ofset` must not be greater than `RawTableInner.bucket_mask`,
/// i.e. `(self.to_base_index() + ofset) <= RawTableInner.bucket_mask` or, in other
/// words, `self.to_base_index() + ofset + 1` must be no greater than the number returned
/// by the function [`RawTable::buckets`] or [`RawTableInner::buckets`].
///
/// If `mem::size_of::<T>() == 0`, then the only requirement is that the
/// `self.to_base_index() + ofset` must not be greater than `RawTableInner.bucket_mask`,
/// i.e. `(self.to_base_index() + ofset) <= RawTableInner.bucket_mask` or, in other words,
/// `self.to_base_index() + ofset + 1` must be no greater than the number returned by the
/// function [`RawTable::buckets`] or [`RawTableInner::buckets`].
///
/// [`Bucket`]: crate::raw::Bucket
/// [`<*mut T>::sub`]: https://doc.rust-lang.org/core/primitive.pointer.html#method.sub-1
/// [`NonNull::new_unchecked`]: https://doc.rust-lang.org/stable/std/ptr/struct.NonNull.html#method.new_unchecked
/// [`RawTable::buckets`]: crate::raw::RawTable::buckets
/// [`RawTableInner::buckets`]: RawTableInner::buckets
#[inline]
unsafe fn next_n(&self, offset: usize) -> Self {
let ptr = if T::IS_ZERO_SIZED {
// invalid pointer is good enough for ZST
invalid_mut(self.ptr.as_ptr() as usize + offset)
} else {
self.ptr.as_ptr().sub(offset)
};
Self {
ptr: NonNull::new_unchecked(ptr),
}
}
/// Executes the destructor (if any) of the pointed-to `data`.
///
/// # Safety
///
/// See [`ptr::drop_in_place`] for safety concerns.
///
/// You should use [`RawTable::erase`] instead of this function,
/// or be careful with calling this function directly, because for
/// properly dropping the data we need also clear `data` control bytes.
/// If we drop data, but do not erase `data control byte` it leads to
/// double drop when [`RawTable`] goes out of scope.
///
/// [`ptr::drop_in_place`]: https://doc.rust-lang.org/core/ptr/fn.drop_in_place.html
/// [`RawTable`]: crate::raw::RawTable
/// [`RawTable::erase`]: crate::raw::RawTable::erase
#[cfg_attr(feature = "inline-more", inline)]
pub(crate) unsafe fn drop(&self) {
self.as_ptr().drop_in_place();
}
/// Reads the `value` from `self` without moving it. This leaves the
/// memory in `self` unchanged.
///
/// # Safety
///
/// See [`ptr::read`] for safety concerns.
///
/// You should use [`RawTable::remove`] instead of this function,
/// or be careful with calling this function directly, because compiler
/// calls its destructor when readed `value` goes out of scope. It
/// can cause double dropping when [`RawTable`] goes out of scope,
/// because of not erased `data control byte`.
///
/// [`ptr::read`]: https://doc.rust-lang.org/core/ptr/fn.read.html
/// [`RawTable`]: crate::raw::RawTable
/// [`RawTable::remove`]: crate::raw::RawTable::remove
#[inline]
pub(crate) unsafe fn read(&self) -> T {
self.as_ptr().read()
}
/// Overwrites a memory location with the given `value` without reading
/// or dropping the old value (like [`ptr::write`] function).
///
/// # Safety
///
/// See [`ptr::write`] for safety concerns.
///
/// # Note
///
/// [`Hash`] and [`Eq`] on the new `T` value and its borrowed form *must* match
/// those for the old `T` value, as the map will not re-evaluate where the new
/// value should go, meaning the value may become "lost" if their location
/// does not reflect their state.
///
/// [`ptr::write`]: https://doc.rust-lang.org/core/ptr/fn.write.html
/// [`Hash`]: https://doc.rust-lang.org/core/hash/trait.Hash.html
/// [`Eq`]: https://doc.rust-lang.org/core/cmp/trait.Eq.html
#[inline]
pub(crate) unsafe fn write(&self, val: T) {
self.as_ptr().write(val);
}
/// Returns a shared immutable reference to the `value`.
///
/// # Safety
///
/// See [`NonNull::as_ref`] for safety concerns.
///
/// [`NonNull::as_ref`]: https://doc.rust-lang.org/core/ptr/struct.NonNull.html#method.as_ref
///
/// # Examples
///
/// ```
/// # #[cfg(feature = "raw")]
/// # fn test() {
/// use core::hash::{BuildHasher, Hash};
/// use hashbrown::raw::{Bucket, RawTable};
///
/// type NewHashBuilder = core::hash::BuildHasherDefault<ahash::AHasher>;
///
/// fn make_hash<K: Hash + ?Sized, S: BuildHasher>(hash_builder: &S, key: &K) -> u64 {
/// use core::hash::Hasher;
/// let mut state = hash_builder.build_hasher();
/// key.hash(&mut state);
/// state.finish()
/// }
///
/// let hash_builder = NewHashBuilder::default();
/// let mut table = RawTable::new();
///
/// let value: (&str, String) = ("A pony", "is a small horse".to_owned());
/// let hash = make_hash(&hash_builder, &value.0);
///
/// table.insert(hash, value.clone(), |val| make_hash(&hash_builder, &val.0));
///
/// let bucket: Bucket<(&str, String)> = table.find(hash, |(k, _)| k == &value.0).unwrap();
///
/// assert_eq!(
/// unsafe { bucket.as_ref() },
/// &("A pony", "is a small horse".to_owned())
/// );
/// # }
/// # fn main() {
/// # #[cfg(feature = "raw")]
/// # test()
/// # }
/// ```
#[inline]
pub unsafe fn as_ref<'a>(&self) -> &'a T {
&*self.as_ptr()
}
/// Returns a unique mutable reference to the `value`.
///
/// # Safety
///
/// See [`NonNull::as_mut`] for safety concerns.
///
/// # Note
///
/// [`Hash`] and [`Eq`] on the new `T` value and its borrowed form *must* match
/// those for the old `T` value, as the map will not re-evaluate where the new
/// value should go, meaning the value may become "lost" if their location
/// does not reflect their state.
///
/// [`NonNull::as_mut`]: https://doc.rust-lang.org/core/ptr/struct.NonNull.html#method.as_mut
/// [`Hash`]: https://doc.rust-lang.org/core/hash/trait.Hash.html
/// [`Eq`]: https://doc.rust-lang.org/core/cmp/trait.Eq.html
///
/// # Examples
///
/// ```
/// # #[cfg(feature = "raw")]
/// # fn test() {
/// use core::hash::{BuildHasher, Hash};
/// use hashbrown::raw::{Bucket, RawTable};
///
/// type NewHashBuilder = core::hash::BuildHasherDefault<ahash::AHasher>;
///
/// fn make_hash<K: Hash + ?Sized, S: BuildHasher>(hash_builder: &S, key: &K) -> u64 {
/// use core::hash::Hasher;
/// let mut state = hash_builder.build_hasher();
/// key.hash(&mut state);
/// state.finish()
/// }
///
/// let hash_builder = NewHashBuilder::default();
/// let mut table = RawTable::new();
///
/// let value: (&str, String) = ("A pony", "is a small horse".to_owned());
/// let hash = make_hash(&hash_builder, &value.0);
///
/// table.insert(hash, value.clone(), |val| make_hash(&hash_builder, &val.0));
///
/// let bucket: Bucket<(&str, String)> = table.find(hash, |(k, _)| k == &value.0).unwrap();
///
/// unsafe {
/// bucket
/// .as_mut()
/// .1
/// .push_str(" less than 147 cm at the withers")
/// };
/// assert_eq!(
/// unsafe { bucket.as_ref() },
/// &(
/// "A pony",
/// "is a small horse less than 147 cm at the withers".to_owned()
/// )
/// );
/// # }
/// # fn main() {
/// # #[cfg(feature = "raw")]
/// # test()
/// # }
/// ```
#[inline]
pub unsafe fn as_mut<'a>(&self) -> &'a mut T {
&mut *self.as_ptr()
}
/// Copies `size_of<T>` bytes from `other` to `self`. The source
/// and destination may *not* overlap.
///
/// # Safety
///
/// See [`ptr::copy_nonoverlapping`] for safety concerns.
///
/// Like [`read`], `copy_nonoverlapping` creates a bitwise copy of `T`, regardless of
/// whether `T` is [`Copy`]. If `T` is not [`Copy`], using *both* the values
/// in the region beginning at `*self` and the region beginning at `*other` can
/// [violate memory safety].
///
/// # Note
///
/// [`Hash`] and [`Eq`] on the new `T` value and its borrowed form *must* match
/// those for the old `T` value, as the map will not re-evaluate where the new
/// value should go, meaning the value may become "lost" if their location
/// does not reflect their state.
///
/// [`ptr::copy_nonoverlapping`]: https://doc.rust-lang.org/core/ptr/fn.copy_nonoverlapping.html
/// [`read`]: https://doc.rust-lang.org/core/ptr/fn.read.html
/// [violate memory safety]: https://doc.rust-lang.org/std/ptr/fn.read.html#ownership-of-the-returned-value
/// [`Hash`]: https://doc.rust-lang.org/core/hash/trait.Hash.html
/// [`Eq`]: https://doc.rust-lang.org/core/cmp/trait.Eq.html
#[cfg(feature = "raw")]
#[inline]
pub unsafe fn copy_from_nonoverlapping(&self, other: &Self) {
self.as_ptr().copy_from_nonoverlapping(other.as_ptr(), 1);
}
}
/// A raw hash table with an unsafe API.
pub struct RawTable<T, A: Allocator = Global> {
table: RawTableInner,
alloc: A,
// Tell dropck that we own instances of T.
marker: PhantomData<T>,
}
/// Non-generic part of `RawTable` which allows functions to be instantiated only once regardless
/// of how many different key-value types are used.
struct RawTableInner {
// Mask to get an index from a hash value. The value is one less than the
// number of buckets in the table.
bucket_mask: usize,
// [Padding], T1, T2, ..., Tlast, C1, C2, ...
// ^ points here
ctrl: NonNull<u8>,
// Number of elements that can be inserted before we need to grow the table
growth_left: usize,
// Number of elements in the table, only really used by len()
items: usize,
}
impl<T> RawTable<T, Global> {
/// Creates a new empty hash table without allocating any memory.
///
/// In effect this returns a table with exactly 1 bucket. However we can
/// leave the data pointer dangling since that bucket is never written to
/// due to our load factor forcing us to always have at least 1 free bucket.
#[inline]
pub const fn new() -> Self {
Self {
table: RawTableInner::NEW,
alloc: Global,
marker: PhantomData,
}
}
/// Attempts to allocate a new hash table with at least enough capacity
/// for inserting the given number of elements without reallocating.
#[cfg(feature = "raw")]
pub fn try_with_capacity(capacity: usize) -> Result<Self, TryReserveError> {
Self::try_with_capacity_in(capacity, Global)
}
/// Allocates a new hash table with at least enough capacity for inserting
/// the given number of elements without reallocating.
pub fn with_capacity(capacity: usize) -> Self {
Self::with_capacity_in(capacity, Global)
}
}
impl<T, A: Allocator> RawTable<T, A> {
const TABLE_LAYOUT: TableLayout = TableLayout::new::<T>();
/// Creates a new empty hash table without allocating any memory, using the
/// given allocator.
///
/// In effect this returns a table with exactly 1 bucket. However we can
/// leave the data pointer dangling since that bucket is never written to
/// due to our load factor forcing us to always have at least 1 free bucket.
#[inline]
pub const fn new_in(alloc: A) -> Self {
Self {
table: RawTableInner::NEW,
alloc,
marker: PhantomData,
}
}
/// Allocates a new hash table with the given number of buckets.
///
/// The control bytes are left uninitialized.
#[cfg_attr(feature = "inline-more", inline)]
unsafe fn new_uninitialized(
alloc: A,
buckets: usize,
fallibility: Fallibility,
) -> Result<Self, TryReserveError> {
debug_assert!(buckets.is_power_of_two());
Ok(Self {
table: RawTableInner::new_uninitialized(
&alloc,
Self::TABLE_LAYOUT,
buckets,
fallibility,
)?,
alloc,
marker: PhantomData,
})
}
/// Attempts to allocate a new hash table using the given allocator, with at least enough
/// capacity for inserting the given number of elements without reallocating.
#[cfg(feature = "raw")]
pub fn try_with_capacity_in(capacity: usize, alloc: A) -> Result<Self, TryReserveError> {
Ok(Self {
table: RawTableInner::fallible_with_capacity(
&alloc,
Self::TABLE_LAYOUT,
capacity,
Fallibility::Fallible,
)?,
alloc,
marker: PhantomData,
})
}
/// Allocates a new hash table using the given allocator, with at least enough capacity for
/// inserting the given number of elements without reallocating.
pub fn with_capacity_in(capacity: usize, alloc: A) -> Self {
Self {
table: RawTableInner::with_capacity(&alloc, Self::TABLE_LAYOUT, capacity),
alloc,
marker: PhantomData,
}
}
/// Returns a reference to the underlying allocator.
#[inline]
pub fn allocator(&self) -> &A {
&self.alloc
}
/// Returns pointer to one past last `data` element in the the table as viewed from
/// the start point of the allocation.
///
/// The caller must ensure that the `RawTable` outlives the returned [`NonNull<T>`],
/// otherwise using it may result in [`undefined behavior`].
///
/// [`undefined behavior`]: https://doc.rust-lang.org/reference/behavior-considered-undefined.html
#[inline]
pub fn data_end(&self) -> NonNull<T> {
// SAFETY: `self.table.ctrl` is `NonNull`, so casting it is safe
//
// `self.table.ctrl.as_ptr().cast()` returns pointer that
// points here (to the end of `T0`)
// ∨
// [Pad], T_n, ..., T1, T0, |CT0, CT1, ..., CT_n|, CTa_0, CTa_1, ..., CTa_m
// \________ ________/
// \/
// `n = buckets - 1`, i.e. `RawTable::buckets() - 1`
//
// where: T0...T_n - our stored data;
// CT0...CT_n - control bytes or metadata for `data`.
// CTa_0...CTa_m - additional control bytes, where `m = Group::WIDTH - 1` (so that the search
// with loading `Group` bytes from the heap works properly, even if the result
// of `h1(hash) & self.bucket_mask` is equal to `self.bucket_mask`). See also
// `RawTableInner::set_ctrl` function.
//
// P.S. `h1(hash) & self.bucket_mask` is the same as `hash as usize % self.buckets()` because the number
// of buckets is a power of two, and `self.bucket_mask = self.buckets() - 1`.
unsafe { NonNull::new_unchecked(self.table.ctrl.as_ptr().cast()) }
}
/// Returns pointer to start of data table.
#[inline]
#[cfg(any(feature = "raw", feature = "nightly"))]
pub unsafe fn data_start(&self) -> NonNull<T> {
NonNull::new_unchecked(self.data_end().as_ptr().wrapping_sub(self.buckets()))
}
/// Return the information about memory allocated by the table.
///
/// `RawTable` allocates single memory block to store both data and metadata.
/// This function returns allocation size and alignment and the beginning of the area.
/// These are the arguments which will be passed to `dealloc` when the table is dropped.
///
/// This function might be useful for memory profiling.
#[inline]
#[cfg(feature = "raw")]
pub fn allocation_info(&self) -> (NonNull<u8>, Layout) {
// SAFETY: We use the same `table_layout` that was used to allocate
// this table.
unsafe { self.table.allocation_info_or_zero(Self::TABLE_LAYOUT) }
}
/// Returns the index of a bucket from a `Bucket`.
#[inline]
pub unsafe fn bucket_index(&self, bucket: &Bucket<T>) -> usize {
bucket.to_base_index(self.data_end())
}
/// Returns a pointer to an element in the table.
///
/// The caller must ensure that the `RawTable` outlives the returned [`Bucket<T>`],
/// otherwise using it may result in [`undefined behavior`].
///
/// # Safety
///
/// If `mem::size_of::<T>() != 0`, then the caller of this function must observe the
/// following safety rules:
///
/// * The table must already be allocated;
///
/// * The `index` must not be greater than the number returned by the [`RawTable::buckets`]
/// function, i.e. `(index + 1) <= self.buckets()`.
///
/// It is safe to call this function with index of zero (`index == 0`) on a table that has
/// not been allocated, but using the returned [`Bucket`] results in [`undefined behavior`].
///
/// If `mem::size_of::<T>() == 0`, then the only requirement is that the `index` must
/// not be greater than the number returned by the [`RawTable::buckets`] function, i.e.
/// `(index + 1) <= self.buckets()`.
///
/// [`RawTable::buckets`]: RawTable::buckets
/// [`undefined behavior`]: https://doc.rust-lang.org/reference/behavior-considered-undefined.html
#[inline]
pub unsafe fn bucket(&self, index: usize) -> Bucket<T> {
// If mem::size_of::<T>() != 0 then return a pointer to the `element` in the `data part` of the table
// (we start counting from "0", so that in the expression T[n], the "n" index actually one less than
// the "buckets" number of our `RawTable`, i.e. "n = RawTable::buckets() - 1"):
//
// `table.bucket(3).as_ptr()` returns a pointer that points here in the `data`
// part of the `RawTable`, i.e. to the start of T3 (see `Bucket::as_ptr`)
// |
// | `base = self.data_end()` points here
// | (to the start of CT0 or to the end of T0)
// v v
// [Pad], T_n, ..., |T3|, T2, T1, T0, |CT0, CT1, CT2, CT3, ..., CT_n, CTa_0, CTa_1, ..., CTa_m
// ^ \__________ __________/
// `table.bucket(3)` returns a pointer that points \/
// here in the `data` part of the `RawTable` (to additional control bytes
// the end of T3) `m = Group::WIDTH - 1`
//
// where: T0...T_n - our stored data;
// CT0...CT_n - control bytes or metadata for `data`;
// CTa_0...CTa_m - additional control bytes (so that the search with loading `Group` bytes from
// the heap works properly, even if the result of `h1(hash) & self.table.bucket_mask`
// is equal to `self.table.bucket_mask`). See also `RawTableInner::set_ctrl` function.
//
// P.S. `h1(hash) & self.table.bucket_mask` is the same as `hash as usize % self.buckets()` because the number
// of buckets is a power of two, and `self.table.bucket_mask = self.buckets() - 1`.
debug_assert_ne!(self.table.bucket_mask, 0);
debug_assert!(index < self.buckets());
Bucket::from_base_index(self.data_end(), index)
}
/// Erases an element from the table without dropping it.
#[cfg_attr(feature = "inline-more", inline)]
unsafe fn erase_no_drop(&mut self, item: &Bucket<T>) {
let index = self.bucket_index(item);
self.table.erase(index);
}
/// Erases an element from the table, dropping it in place.
#[cfg_attr(feature = "inline-more", inline)]
#[allow(clippy::needless_pass_by_value)]
pub unsafe fn erase(&mut self, item: Bucket<T>) {
// Erase the element from the table first since drop might panic.
self.erase_no_drop(&item);
item.drop();
}
/// Finds and erases an element from the table, dropping it in place.
/// Returns true if an element was found.
#[cfg(feature = "raw")]
#[cfg_attr(feature = "inline-more", inline)]
pub fn erase_entry(&mut self, hash: u64, eq: impl FnMut(&T) -> bool) -> bool {
// Avoid `Option::map` because it bloats LLVM IR.
if let Some(bucket) = self.find(hash, eq) {
unsafe {
self.erase(bucket);
}
true
} else {
false
}
}
/// Removes an element from the table, returning it.
///
/// This also returns an `InsertSlot` pointing to the newly free bucket.
#[cfg_attr(feature = "inline-more", inline)]
#[allow(clippy::needless_pass_by_value)]
pub unsafe fn remove(&mut self, item: Bucket<T>) -> (T, InsertSlot) {
self.erase_no_drop(&item);
(
item.read(),
InsertSlot {
index: self.bucket_index(&item),
},
)
}
/// Finds and removes an element from the table, returning it.
#[cfg_attr(feature = "inline-more", inline)]
pub fn remove_entry(&mut self, hash: u64, eq: impl FnMut(&T) -> bool) -> Option<T> {
// Avoid `Option::map` because it bloats LLVM IR.
match self.find(hash, eq) {
Some(bucket) => Some(unsafe { self.remove(bucket).0 }),
None => None,
}
}
/// Marks all table buckets as empty without dropping their contents.
#[cfg_attr(feature = "inline-more", inline)]
pub fn clear_no_drop(&mut self) {
self.table.clear_no_drop();
}
/// Removes all elements from the table without freeing the backing memory.
#[cfg_attr(feature = "inline-more", inline)]
pub fn clear(&mut self) {
if self.is_empty() {
// Special case empty table to avoid surprising O(capacity) time.
return;
}
// Ensure that the table is reset even if one of the drops panic
let mut self_ = guard(self, |self_| self_.clear_no_drop());
unsafe {
// SAFETY: ScopeGuard sets to zero the `items` field of the table
// even in case of panic during the dropping of the elements so
// that there will be no double drop of the elements.
self_.table.drop_elements::<T>();
}
}
/// Shrinks the table to fit `max(self.len(), min_size)` elements.
#[cfg_attr(feature = "inline-more", inline)]
pub fn shrink_to(&mut self, min_size: usize, hasher: impl Fn(&T) -> u64) {
// Calculate the minimal number of elements that we need to reserve
// space for.
let min_size = usize::max(self.table.items, min_size);
if min_size == 0 {
let mut old_inner = mem::replace(&mut self.table, RawTableInner::NEW);
unsafe {
// SAFETY:
// 1. We call the function only once;
// 2. We know for sure that `alloc` and `table_layout` matches the [`Allocator`]
// and [`TableLayout`] that were used to allocate this table.
// 3. If any elements' drop function panics, then there will only be a memory leak,
// because we have replaced the inner table with a new one.
old_inner.drop_inner_table::<T, _>(&self.alloc, Self::TABLE_LAYOUT);
}
return;
}
// Calculate the number of buckets that we need for this number of
// elements. If the calculation overflows then the requested bucket
// count must be larger than what we have right and nothing needs to be
// done.
let min_buckets = match capacity_to_buckets(min_size) {
Some(buckets) => buckets,
None => return,
};
// If we have more buckets than we need, shrink the table.
if min_buckets < self.buckets() {
// Fast path if the table is empty
if self.table.items == 0 {
let new_inner =
RawTableInner::with_capacity(&self.alloc, Self::TABLE_LAYOUT, min_size);
let mut old_inner = mem::replace(&mut self.table, new_inner);
unsafe {
// SAFETY:
// 1. We call the function only once;
// 2. We know for sure that `alloc` and `table_layout` matches the [`Allocator`]
// and [`TableLayout`] that were used to allocate this table.
// 3. If any elements' drop function panics, then there will only be a memory leak,
// because we have replaced the inner table with a new one.
old_inner.drop_inner_table::<T, _>(&self.alloc, Self::TABLE_LAYOUT);
}
} else {
// Avoid `Result::unwrap_or_else` because it bloats LLVM IR.
unsafe {
// SAFETY:
// 1. We know for sure that `min_size >= self.table.items`.
// 2. The [`RawTableInner`] must already have properly initialized control bytes since
// we will never expose RawTable::new_uninitialized in a public API.
if self
.resize(min_size, hasher, Fallibility::Infallible)
.is_err()
{
// SAFETY: The result of calling the `resize` function cannot be an error
// because `fallibility == Fallibility::Infallible.
hint::unreachable_unchecked()
}
}
}
}
}
/// Ensures that at least `additional` items can be inserted into the table
/// without reallocation.
#[cfg_attr(feature = "inline-more", inline)]
pub fn reserve(&mut self, additional: usize, hasher: impl Fn(&T) -> u64) {
if unlikely(additional > self.table.growth_left) {
// Avoid `Result::unwrap_or_else` because it bloats LLVM IR.
unsafe {
// SAFETY: The [`RawTableInner`] must already have properly initialized control
// bytes since we will never expose RawTable::new_uninitialized in a public API.
if self
.reserve_rehash(additional, hasher, Fallibility::Infallible)
.is_err()
{
// SAFETY: All allocation errors will be caught inside `RawTableInner::reserve_rehash`.
hint::unreachable_unchecked()
}
}
}
}
/// Tries to ensure that at least `additional` items can be inserted into
/// the table without reallocation.
#[cfg_attr(feature = "inline-more", inline)]
pub fn try_reserve(
&mut self,
additional: usize,
hasher: impl Fn(&T) -> u64,
) -> Result<(), TryReserveError> {
if additional > self.table.growth_left {
// SAFETY: The [`RawTableInner`] must already have properly initialized control
// bytes since we will never expose RawTable::new_uninitialized in a public API.
unsafe { self.reserve_rehash(additional, hasher, Fallibility::Fallible) }
} else {
Ok(())
}
}
/// Out-of-line slow path for `reserve` and `try_reserve`.
///
/// # Safety
///
/// The [`RawTableInner`] must have properly initialized control bytes,
/// otherwise calling this function results in [`undefined behavior`]
///
/// [`undefined behavior`]: https://doc.rust-lang.org/reference/behavior-considered-undefined.html
#[cold]
#[inline(never)]
unsafe fn reserve_rehash(
&mut self,
additional: usize,
hasher: impl Fn(&T) -> u64,
fallibility: Fallibility,
) -> Result<(), TryReserveError> {
unsafe {
// SAFETY:
// 1. We know for sure that `alloc` and `layout` matches the [`Allocator`] and
// [`TableLayout`] that were used to allocate this table.
// 2. The `drop` function is the actual drop function of the elements stored in
// the table.
// 3. The caller ensures that the control bytes of the `RawTableInner`
// are already initialized.
self.table.reserve_rehash_inner(
&self.alloc,
additional,
&|table, index| hasher(table.bucket::<T>(index).as_ref()),
fallibility,
Self::TABLE_LAYOUT,
if T::NEEDS_DROP {
Some(mem::transmute(ptr::drop_in_place::<T> as unsafe fn(*mut T)))
} else {
None
},
)
}
}
/// Allocates a new table of a different size and moves the contents of the
/// current table into it.
///
/// # Safety
///
/// The [`RawTableInner`] must have properly initialized control bytes,
/// otherwise calling this function results in [`undefined behavior`]
///
/// The caller of this function must ensure that `capacity >= self.table.items`
/// otherwise:
///
/// * If `self.table.items != 0`, calling of this function with `capacity`
/// equal to 0 (`capacity == 0`) results in [`undefined behavior`].
///
/// * If `capacity_to_buckets(capacity) < Group::WIDTH` and
/// `self.table.items > capacity_to_buckets(capacity)`
/// calling this function results in [`undefined behavior`].
///
/// * If `capacity_to_buckets(capacity) >= Group::WIDTH` and
/// `self.table.items > capacity_to_buckets(capacity)`
/// calling this function are never return (will go into an
/// infinite loop).
///
/// See [`RawTableInner::find_insert_slot`] for more information.
///
/// [`RawTableInner::find_insert_slot`]: RawTableInner::find_insert_slot
/// [`undefined behavior`]: https://doc.rust-lang.org/reference/behavior-considered-undefined.html
unsafe fn resize(
&mut self,
capacity: usize,
hasher: impl Fn(&T) -> u64,
fallibility: Fallibility,
) -> Result<(), TryReserveError> {
// SAFETY:
// 1. The caller of this function guarantees that `capacity >= self.table.items`.
// 2. We know for sure that `alloc` and `layout` matches the [`Allocator`] and
// [`TableLayout`] that were used to allocate this table.
// 3. The caller ensures that the control bytes of the `RawTableInner`
// are already initialized.
self.table.resize_inner(
&self.alloc,
capacity,
&|table, index| hasher(table.bucket::<T>(index).as_ref()),
fallibility,
Self::TABLE_LAYOUT,
)
}
/// Inserts a new element into the table, and returns its raw bucket.
///
/// This does not check if the given element already exists in the table.
#[cfg_attr(feature = "inline-more", inline)]
pub fn insert(&mut self, hash: u64, value: T, hasher: impl Fn(&T) -> u64) -> Bucket<T> {
unsafe {
// SAFETY:
// 1. The [`RawTableInner`] must already have properly initialized control bytes since
// we will never expose `RawTable::new_uninitialized` in a public API.
//
// 2. We reserve additional space (if necessary) right after calling this function.
let mut slot = self.table.find_insert_slot(hash);
// We can avoid growing the table once we have reached our load factor if we are replacing
// a tombstone. This works since the number of EMPTY slots does not change in this case.
//
// SAFETY: The function is guaranteed to return [`InsertSlot`] that contains an index
// in the range `0..=self.buckets()`.
let old_ctrl = *self.table.ctrl(slot.index);
if unlikely(self.table.growth_left == 0 && special_is_empty(old_ctrl)) {
self.reserve(1, hasher);
// SAFETY: We know for sure that `RawTableInner` has control bytes
// initialized and that there is extra space in the table.
slot = self.table.find_insert_slot(hash);
}
self.insert_in_slot(hash, slot, value)
}
}
/// Attempts to insert a new element without growing the table and return its raw bucket.
///
/// Returns an `Err` containing the given element if inserting it would require growing the
/// table.
///
/// This does not check if the given element already exists in the table.
#[cfg(feature = "raw")]
#[cfg_attr(feature = "inline-more", inline)]
pub fn try_insert_no_grow(&mut self, hash: u64, value: T) -> Result<Bucket<T>, T> {
unsafe {
match self.table.prepare_insert_no_grow(hash) {
Ok(index) => {
let bucket = self.bucket(index);
bucket.write(value);
Ok(bucket)
}
Err(()) => Err(value),
}
}
}
/// Inserts a new element into the table, and returns a mutable reference to it.
///
/// This does not check if the given element already exists in the table.
#[cfg_attr(feature = "inline-more", inline)]
pub fn insert_entry(&mut self, hash: u64, value: T, hasher: impl Fn(&T) -> u64) -> &mut T {
unsafe { self.insert(hash, value, hasher).as_mut() }
}
/// Inserts a new element into the table, without growing the table.
///
/// There must be enough space in the table to insert the new element.
///
/// This does not check if the given element already exists in the table.
#[cfg_attr(feature = "inline-more", inline)]
#[cfg(any(feature = "raw", feature = "rustc-internal-api"))]
pub unsafe fn insert_no_grow(&mut self, hash: u64, value: T) -> Bucket<T> {
let (index, old_ctrl) = self.table.prepare_insert_slot(hash);
let bucket = self.table.bucket(index);
// If we are replacing a DELETED entry then we don't need to update
// the load counter.
self.table.growth_left -= special_is_empty(old_ctrl) as usize;
bucket.write(value);
self.table.items += 1;
bucket
}
/// Temporary removes a bucket, applying the given function to the removed
/// element and optionally put back the returned value in the same bucket.
///
/// Returns `true` if the bucket still contains an element
///
/// This does not check if the given bucket is actually occupied.
#[cfg_attr(feature = "inline-more", inline)]
pub unsafe fn replace_bucket_with<F>(&mut self, bucket: Bucket<T>, f: F) -> bool
where
F: FnOnce(T) -> Option<T>,
{
let index = self.bucket_index(&bucket);
let old_ctrl = *self.table.ctrl(index);
debug_assert!(self.is_bucket_full(index));
let old_growth_left = self.table.growth_left;
let item = self.remove(bucket).0;
if let Some(new_item) = f(item) {
self.table.growth_left = old_growth_left;
self.table.set_ctrl(index, old_ctrl);
self.table.items += 1;
self.bucket(index).write(new_item);
true
} else {
false
}
}
/// Searches for an element in the table. If the element is not found,
/// returns `Err` with the position of a slot where an element with the
/// same hash could be inserted.
///
/// This function may resize the table if additional space is required for
/// inserting an element.
#[inline]
pub fn find_or_find_insert_slot(
&mut self,
hash: u64,
mut eq: impl FnMut(&T) -> bool,
hasher: impl Fn(&T) -> u64,
) -> Result<Bucket<T>, InsertSlot> {
self.reserve(1, hasher);
unsafe {
// SAFETY:
// 1. We know for sure that there is at least one empty `bucket` in the table.
// 2. The [`RawTableInner`] must already have properly initialized control bytes since we will
// never expose `RawTable::new_uninitialized` in a public API.
// 3. The `find_or_find_insert_slot_inner` function returns the `index` of only the full bucket,
// which is in the range `0..self.buckets()` (since there is at least one empty `bucket` in
// the table), so calling `self.bucket(index)` and `Bucket::as_ref` is safe.
match self
.table
.find_or_find_insert_slot_inner(hash, &mut |index| eq(self.bucket(index).as_ref()))
{
// SAFETY: See explanation above.
Ok(index) => Ok(self.bucket(index)),
Err(slot) => Err(slot),
}
}
}
/// Inserts a new element into the table in the given slot, and returns its
/// raw bucket.
///
/// # Safety
///
/// `slot` must point to a slot previously returned by
/// `find_or_find_insert_slot`, and no mutation of the table must have
/// occurred since that call.
#[inline]
pub unsafe fn insert_in_slot(&mut self, hash: u64, slot: InsertSlot, value: T) -> Bucket<T> {
let old_ctrl = *self.table.ctrl(slot.index);
self.table.record_item_insert_at(slot.index, old_ctrl, hash);
let bucket = self.bucket(slot.index);
bucket.write(value);
bucket
}
/// Searches for an element in the table.
#[inline]
pub fn find(&self, hash: u64, mut eq: impl FnMut(&T) -> bool) -> Option<Bucket<T>> {
unsafe {
// SAFETY:
// 1. The [`RawTableInner`] must already have properly initialized control bytes since we
// will never expose `RawTable::new_uninitialized` in a public API.
// 1. The `find_inner` function returns the `index` of only the full bucket, which is in
// the range `0..self.buckets()`, so calling `self.bucket(index)` and `Bucket::as_ref`
// is safe.
let result = self
.table
.find_inner(hash, &mut |index| eq(self.bucket(index).as_ref()));
// Avoid `Option::map` because it bloats LLVM IR.
match result {
// SAFETY: See explanation above.
Some(index) => Some(self.bucket(index)),
None => None,
}
}
}
/// Gets a reference to an element in the table.
#[inline]
pub fn get(&self, hash: u64, eq: impl FnMut(&T) -> bool) -> Option<&T> {
// Avoid `Option::map` because it bloats LLVM IR.
match self.find(hash, eq) {
Some(bucket) => Some(unsafe { bucket.as_ref() }),
None => None,
}
}
/// Gets a mutable reference to an element in the table.
#[inline]
pub fn get_mut(&mut self, hash: u64, eq: impl FnMut(&T) -> bool) -> Option<&mut T> {
// Avoid `Option::map` because it bloats LLVM IR.
match self.find(hash, eq) {
Some(bucket) => Some(unsafe { bucket.as_mut() }),
None => None,
}
}
/// Attempts to get mutable references to `N` entries in the table at once.
///
/// Returns an array of length `N` with the results of each query.
///
/// At most one mutable reference will be returned to any entry. `None` will be returned if any
/// of the hashes are duplicates. `None` will be returned if the hash is not found.
///
/// The `eq` argument should be a closure such that `eq(i, k)` returns true if `k` is equal to
/// the `i`th key to be looked up.
pub fn get_many_mut<const N: usize>(
&mut self,
hashes: [u64; N],
eq: impl FnMut(usize, &T) -> bool,
) -> Option<[&'_ mut T; N]> {
unsafe {
let ptrs = self.get_many_mut_pointers(hashes, eq)?;
for (i, &cur) in ptrs.iter().enumerate() {
if ptrs[..i].iter().any(|&prev| ptr::eq::<T>(prev, cur)) {
return None;
}
}
// All bucket are distinct from all previous buckets so we're clear to return the result
// of the lookup.
// TODO use `MaybeUninit::array_assume_init` here instead once that's stable.
Some(mem::transmute_copy(&ptrs))
}
}
pub unsafe fn get_many_unchecked_mut<const N: usize>(
&mut self,
hashes: [u64; N],
eq: impl FnMut(usize, &T) -> bool,
) -> Option<[&'_ mut T; N]> {
let ptrs = self.get_many_mut_pointers(hashes, eq)?;
Some(mem::transmute_copy(&ptrs))
}
unsafe fn get_many_mut_pointers<const N: usize>(
&mut self,
hashes: [u64; N],
mut eq: impl FnMut(usize, &T) -> bool,
) -> Option<[*mut T; N]> {
// TODO use `MaybeUninit::uninit_array` here instead once that's stable.
let mut outs: MaybeUninit<[*mut T; N]> = MaybeUninit::uninit();
let outs_ptr = outs.as_mut_ptr();
for (i, &hash) in hashes.iter().enumerate() {
let cur = self.find(hash, |k| eq(i, k))?;
*(*outs_ptr).get_unchecked_mut(i) = cur.as_mut();
}
// TODO use `MaybeUninit::array_assume_init` here instead once that's stable.
Some(outs.assume_init())
}
/// Returns the number of elements the map can hold without reallocating.
///
/// This number is a lower bound; the table might be able to hold
/// more, but is guaranteed to be able to hold at least this many.
#[inline]
pub fn capacity(&self) -> usize {
self.table.items + self.table.growth_left
}
/// Returns the number of elements in the table.
#[inline]
pub fn len(&self) -> usize {
self.table.items
}
/// Returns `true` if the table contains no elements.
#[inline]
pub fn is_empty(&self) -> bool {
self.len() == 0
}
/// Returns the number of buckets in the table.
#[inline]
pub fn buckets(&self) -> usize {
self.table.bucket_mask + 1
}
/// Checks whether the bucket at `index` is full.
///
/// # Safety
///
/// The caller must ensure `index` is less than the number of buckets.
#[inline]
pub unsafe fn is_bucket_full(&self, index: usize) -> bool {
self.table.is_bucket_full(index)
}
/// Returns an iterator over every element in the table. It is up to
/// the caller to ensure that the `RawTable` outlives the `RawIter`.
/// Because we cannot make the `next` method unsafe on the `RawIter`
/// struct, we have to make the `iter` method unsafe.
#[inline]
pub unsafe fn iter(&self) -> RawIter<T> {
// SAFETY:
// 1. The caller must uphold the safety contract for `iter` method.
// 2. The [`RawTableInner`] must already have properly initialized control bytes since
// we will never expose RawTable::new_uninitialized in a public API.
self.table.iter()
}
/// Returns an iterator over occupied buckets that could match a given hash.
///
/// `RawTable` only stores 7 bits of the hash value, so this iterator may
/// return items that have a hash value different than the one provided. You
/// should always validate the returned values before using them.
///
/// It is up to the caller to ensure that the `RawTable` outlives the
/// `RawIterHash`. Because we cannot make the `next` method unsafe on the
/// `RawIterHash` struct, we have to make the `iter_hash` method unsafe.
#[cfg_attr(feature = "inline-more", inline)]
#[cfg(feature = "raw")]
pub unsafe fn iter_hash(&self, hash: u64) -> RawIterHash<T> {
RawIterHash::new(self, hash)
}
/// Returns an iterator which removes all elements from the table without
/// freeing the memory.
#[cfg_attr(feature = "inline-more", inline)]
pub fn drain(&mut self) -> RawDrain<'_, T, A> {
unsafe {
let iter = self.iter();
self.drain_iter_from(iter)
}
}
/// Returns an iterator which removes all elements from the table without
/// freeing the memory.
///
/// Iteration starts at the provided iterator's current location.
///
/// It is up to the caller to ensure that the iterator is valid for this
/// `RawTable` and covers all items that remain in the table.
#[cfg_attr(feature = "inline-more", inline)]
pub unsafe fn drain_iter_from(&mut self, iter: RawIter<T>) -> RawDrain<'_, T, A> {
debug_assert_eq!(iter.len(), self.len());
RawDrain {
iter,
table: mem::replace(&mut self.table, RawTableInner::NEW),
orig_table: NonNull::from(&mut self.table),
marker: PhantomData,
}
}
/// Returns an iterator which consumes all elements from the table.
///
/// Iteration starts at the provided iterator's current location.
///
/// It is up to the caller to ensure that the iterator is valid for this
/// `RawTable` and covers all items that remain in the table.
pub unsafe fn into_iter_from(self, iter: RawIter<T>) -> RawIntoIter<T, A> {
debug_assert_eq!(iter.len(), self.len());
let allocation = self.into_allocation();
RawIntoIter {
iter,
allocation,
marker: PhantomData,
}
}
/// Converts the table into a raw allocation. The contents of the table
/// should be dropped using a `RawIter` before freeing the allocation.
#[cfg_attr(feature = "inline-more", inline)]
pub(crate) fn into_allocation(self) -> Option<(NonNull<u8>, Layout, A)> {
let alloc = if self.table.is_empty_singleton() {
None
} else {
// Avoid `Option::unwrap_or_else` because it bloats LLVM IR.
let (layout, ctrl_offset) =
match Self::TABLE_LAYOUT.calculate_layout_for(self.table.buckets()) {
Some(lco) => lco,
None => unsafe { hint::unreachable_unchecked() },
};
Some((
unsafe { NonNull::new_unchecked(self.table.ctrl.as_ptr().sub(ctrl_offset)) },
layout,
unsafe { ptr::read(&self.alloc) },
))
};
mem::forget(self);
alloc
}
}
unsafe impl<T, A: Allocator> Send for RawTable<T, A>
where
T: Send,
A: Send,
{
}
unsafe impl<T, A: Allocator> Sync for RawTable<T, A>
where
T: Sync,
A: Sync,
{
}
impl RawTableInner {
const NEW: Self = RawTableInner::new();
/// Creates a new empty hash table without allocating any memory.
///
/// In effect this returns a table with exactly 1 bucket. However we can
/// leave the data pointer dangling since that bucket is never accessed
/// due to our load factor forcing us to always have at least 1 free bucket.
#[inline]
const fn new() -> Self {
Self {
// Be careful to cast the entire slice to a raw pointer.
ctrl: unsafe { NonNull::new_unchecked(Group::static_empty() as *const _ as *mut u8) },
bucket_mask: 0,
items: 0,
growth_left: 0,
}
}
}
impl RawTableInner {
/// Allocates a new [`RawTableInner`] with the given number of buckets.
/// The control bytes and buckets are left uninitialized.
///
/// # Safety
///
/// The caller of this function must ensure that the `buckets` is power of two
/// and also initialize all control bytes of the length `self.bucket_mask + 1 +
/// Group::WIDTH` with the [`EMPTY`] bytes.
///
/// See also [`Allocator`] API for other safety concerns.
///
/// [`Allocator`]: https://doc.rust-lang.org/alloc/alloc/trait.Allocator.html
#[cfg_attr(feature = "inline-more", inline)]
unsafe fn new_uninitialized<A>(
alloc: &A,
table_layout: TableLayout,
buckets: usize,
fallibility: Fallibility,
) -> Result<Self, TryReserveError>
where
A: Allocator,
{
debug_assert!(buckets.is_power_of_two());
// Avoid `Option::ok_or_else` because it bloats LLVM IR.
let (layout, ctrl_offset) = match table_layout.calculate_layout_for(buckets) {
Some(lco) => lco,
None => return Err(fallibility.capacity_overflow()),
};
let ptr: NonNull<u8> = match do_alloc(alloc, layout) {
Ok(block) => block.cast(),
Err(_) => return Err(fallibility.alloc_err(layout)),
};
// SAFETY: null pointer will be caught in above check
let ctrl = NonNull::new_unchecked(ptr.as_ptr().add(ctrl_offset));
Ok(Self {
ctrl,
bucket_mask: buckets - 1,
items: 0,
growth_left: bucket_mask_to_capacity(buckets - 1),
})
}
/// Attempts to allocate a new [`RawTableInner`] with at least enough
/// capacity for inserting the given number of elements without reallocating.
///
/// All the control bytes are initialized with the [`EMPTY`] bytes.
#[inline]
fn fallible_with_capacity<A>(
alloc: &A,
table_layout: TableLayout,
capacity: usize,
fallibility: Fallibility,
) -> Result<Self, TryReserveError>
where
A: Allocator,
{
if capacity == 0 {
Ok(Self::NEW)
} else {
// SAFETY: We checked that we could successfully allocate the new table, and then
// initialized all control bytes with the constant `EMPTY` byte.
unsafe {
let buckets =
capacity_to_buckets(capacity).ok_or_else(|| fallibility.capacity_overflow())?;
let result = Self::new_uninitialized(alloc, table_layout, buckets, fallibility)?;
// SAFETY: We checked that the table is allocated and therefore the table already has
// `self.bucket_mask + 1 + Group::WIDTH` number of control bytes (see TableLayout::calculate_layout_for)
// so writing `self.num_ctrl_bytes() == bucket_mask + 1 + Group::WIDTH` bytes is safe.
result.ctrl(0).write_bytes(EMPTY, result.num_ctrl_bytes());
Ok(result)
}
}
}
/// Allocates a new [`RawTableInner`] with at least enough capacity for inserting
/// the given number of elements without reallocating.
///
/// Panics if the new capacity exceeds [`isize::MAX`] bytes and [`abort`] the program
/// in case of allocation error. Use [`fallible_with_capacity`] instead if you want to
/// handle memory allocation failure.
///
/// All the control bytes are initialized with the [`EMPTY`] bytes.
///
/// [`fallible_with_capacity`]: RawTableInner::fallible_with_capacity
/// [`abort`]: https://doc.rust-lang.org/alloc/alloc/fn.handle_alloc_error.html
fn with_capacity<A>(alloc: &A, table_layout: TableLayout, capacity: usize) -> Self
where
A: Allocator,
{
// Avoid `Result::unwrap_or_else` because it bloats LLVM IR.
match Self::fallible_with_capacity(alloc, table_layout, capacity, Fallibility::Infallible) {
Ok(table_inner) => table_inner,
// SAFETY: All allocation errors will be caught inside `RawTableInner::new_uninitialized`.
Err(_) => unsafe { hint::unreachable_unchecked() },
}
}
/// Fixes up an insertion slot returned by the [`RawTableInner::find_insert_slot_in_group`] method.
///
/// In tables smaller than the group width (`self.buckets() < Group::WIDTH`), trailing control
/// bytes outside the range of the table are filled with [`EMPTY`] entries. These will unfortunately
/// trigger a match of [`RawTableInner::find_insert_slot_in_group`] function. This is because
/// the `Some(bit)` returned by `group.match_empty_or_deleted().lowest_set_bit()` after masking
/// (`(probe_seq.pos + bit) & self.bucket_mask`) may point to a full bucket that is already occupied.
/// We detect this situation here and perform a second scan starting at the beginning of the table.
/// This second scan is guaranteed to find an empty slot (due to the load factor) before hitting the
/// trailing control bytes (containing [`EMPTY`] bytes).
///
/// If this function is called correctly, it is guaranteed to return [`InsertSlot`] with an
/// index of an empty or deleted bucket in the range `0..self.buckets()` (see `Warning` and
/// `Safety`).
///
/// # Warning
///
/// The table must have at least 1 empty or deleted `bucket`, otherwise if the table is less than
/// the group width (`self.buckets() < Group::WIDTH`) this function returns an index outside of the
/// table indices range `0..self.buckets()` (`0..=self.bucket_mask`). Attempt to write data at that
/// index will cause immediate [`undefined behavior`].
///
/// # Safety
///
/// The safety rules are directly derived from the safety rules for [`RawTableInner::ctrl`] method.
/// Thus, in order to uphold those safety contracts, as well as for the correct logic of the work
/// of this crate, the following rules are necessary and sufficient:
///
/// * The [`RawTableInner`] must have properly initialized control bytes otherwise calling this
/// function results in [`undefined behavior`].
///
/// * This function must only be used on insertion slots found by [`RawTableInner::find_insert_slot_in_group`]
/// (after the `find_insert_slot_in_group` function, but before insertion into the table).
///
/// * The `index` must not be greater than the `self.bucket_mask`, i.e. `(index + 1) <= self.buckets()`
/// (this one is provided by the [`RawTableInner::find_insert_slot_in_group`] function).
///
/// Calling this function with an index not provided by [`RawTableInner::find_insert_slot_in_group`]
/// may result in [`undefined behavior`] even if the index satisfies the safety rules of the
/// [`RawTableInner::ctrl`] function (`index < self.bucket_mask + 1 + Group::WIDTH`).
///
/// [`RawTableInner::ctrl`]: RawTableInner::ctrl
/// [`RawTableInner::find_insert_slot_in_group`]: RawTableInner::find_insert_slot_in_group
/// [`undefined behavior`]: https://doc.rust-lang.org/reference/behavior-considered-undefined.html
#[inline]
unsafe fn fix_insert_slot(&self, mut index: usize) -> InsertSlot {
// SAFETY: The caller of this function ensures that `index` is in the range `0..=self.bucket_mask`.
if unlikely(self.is_bucket_full(index)) {
debug_assert!(self.bucket_mask < Group::WIDTH);
// SAFETY:
//
// * Since the caller of this function ensures that the control bytes are properly
// initialized and `ptr = self.ctrl(0)` points to the start of the array of control
// bytes, therefore: `ctrl` is valid for reads, properly aligned to `Group::WIDTH`
// and points to the properly initialized control bytes (see also
// `TableLayout::calculate_layout_for` and `ptr::read`);
//
// * Because the caller of this function ensures that the index was provided by the
// `self.find_insert_slot_in_group()` function, so for for tables larger than the
// group width (self.buckets() >= Group::WIDTH), we will never end up in the given
// branch, since `(probe_seq.pos + bit) & self.bucket_mask` in `find_insert_slot_in_group`
// cannot return a full bucket index. For tables smaller than the group width, calling
// the `unwrap_unchecked` function is also safe, as the trailing control bytes outside
// the range of the table are filled with EMPTY bytes (and we know for sure that there
// is at least one FULL bucket), so this second scan either finds an empty slot (due to
// the load factor) or hits the trailing control bytes (containing EMPTY).
index = Group::load_aligned(self.ctrl(0))
.match_empty_or_deleted()
.lowest_set_bit()
.unwrap_unchecked();
}
InsertSlot { index }
}
/// Finds the position to insert something in a group.
///
/// **This may have false positives and must be fixed up with `fix_insert_slot`
/// before it's used.**
///
/// The function is guaranteed to return the index of an empty or deleted [`Bucket`]
/// in the range `0..self.buckets()` (`0..=self.bucket_mask`).
#[inline]
fn find_insert_slot_in_group(&self, group: &Group, probe_seq: &ProbeSeq) -> Option<usize> {
let bit = group.match_empty_or_deleted().lowest_set_bit();
if likely(bit.is_some()) {
// This is the same as `(probe_seq.pos + bit) % self.buckets()` because the number
// of buckets is a power of two, and `self.bucket_mask = self.buckets() - 1`.
Some((probe_seq.pos + bit.unwrap()) & self.bucket_mask)
} else {
None
}
}
/// Searches for an element in the table, or a potential slot where that element could
/// be inserted (an empty or deleted [`Bucket`] index).
///
/// This uses dynamic dispatch to reduce the amount of code generated, but that is
/// eliminated by LLVM optimizations.
///
/// This function does not make any changes to the `data` part of the table, or any
/// changes to the `items` or `growth_left` field of the table.
///
/// The table must have at least 1 empty or deleted `bucket`, otherwise, if the
/// `eq: &mut dyn FnMut(usize) -> bool` function does not return `true`, this function
/// will never return (will go into an infinite loop) for tables larger than the group
/// width, or return an index outside of the table indices range if the table is less
/// than the group width.
///
/// This function is guaranteed to provide the `eq: &mut dyn FnMut(usize) -> bool`
/// function with only `FULL` buckets' indices and return the `index` of the found
/// element (as `Ok(index)`). If the element is not found and there is at least 1
/// empty or deleted [`Bucket`] in the table, the function is guaranteed to return
/// [InsertSlot] with an index in the range `0..self.buckets()`, but in any case,
/// if this function returns [`InsertSlot`], it will contain an index in the range
/// `0..=self.buckets()`.
///
/// # Safety
///
/// The [`RawTableInner`] must have properly initialized control bytes otherwise calling
/// this function results in [`undefined behavior`].
///
/// Attempt to write data at the [`InsertSlot`] returned by this function when the table is
/// less than the group width and if there was not at least one empty or deleted bucket in
/// the table will cause immediate [`undefined behavior`]. This is because in this case the
/// function will return `self.bucket_mask + 1` as an index due to the trailing [`EMPTY]
/// control bytes outside the table range.
///
/// [`undefined behavior`]: https://doc.rust-lang.org/reference/behavior-considered-undefined.html
#[inline]
unsafe fn find_or_find_insert_slot_inner(
&self,
hash: u64,
eq: &mut dyn FnMut(usize) -> bool,
) -> Result<usize, InsertSlot> {
let mut insert_slot = None;
let h2_hash = h2(hash);
let mut probe_seq = self.probe_seq(hash);
loop {
// SAFETY:
// * Caller of this function ensures that the control bytes are properly initialized.
//
// * `ProbeSeq.pos` cannot be greater than `self.bucket_mask = self.buckets() - 1`
// of the table due to masking with `self.bucket_mask` and also because mumber of
// buckets is a power of two (see `self.probe_seq` function).
//
// * Even if `ProbeSeq.pos` returns `position == self.bucket_mask`, it is safe to
// call `Group::load` due to the extended control bytes range, which is
// `self.bucket_mask + 1 + Group::WIDTH` (in fact, this means that the last control
// byte will never be read for the allocated table);
//
// * Also, even if `RawTableInner` is not already allocated, `ProbeSeq.pos` will
// always return "0" (zero), so Group::load will read unaligned `Group::static_empty()`
// bytes, which is safe (see RawTableInner::new).
let group = unsafe { Group::load(self.ctrl(probe_seq.pos)) };
for bit in group.match_byte(h2_hash) {
let index = (probe_seq.pos + bit) & self.bucket_mask;
if likely(eq(index)) {
return Ok(index);
}
}
// We didn't find the element we were looking for in the group, try to get an
// insertion slot from the group if we don't have one yet.
if likely(insert_slot.is_none()) {
insert_slot = self.find_insert_slot_in_group(&group, &probe_seq);
}
// Only stop the search if the group contains at least one empty element.
// Otherwise, the element that we are looking for might be in a following group.
if likely(group.match_empty().any_bit_set()) {
// We must have found a insert slot by now, since the current group contains at
// least one. For tables smaller than the group width, there will still be an
// empty element in the current (and only) group due to the load factor.
unsafe {
// SAFETY:
// * Caller of this function ensures that the control bytes are properly initialized.
//
// * We use this function with the slot / index found by `self.find_insert_slot_in_group`
return Err(self.fix_insert_slot(insert_slot.unwrap_unchecked()));
}
}
probe_seq.move_next(self.bucket_mask);
}
}
/// Searches for an empty or deleted bucket which is suitable for inserting a new
/// element and sets the hash for that slot. Returns an index of that slot and the
/// old control byte stored in the found index.
///
/// This function does not check if the given element exists in the table. Also,
/// this function does not check if there is enough space in the table to insert
/// a new element. Caller of the funtion must make ensure that the table has at
/// least 1 empty or deleted `bucket`, otherwise this function will never return
/// (will go into an infinite loop) for tables larger than the group width, or
/// return an index outside of the table indices range if the table is less than
/// the group width.
///
/// If there is at least 1 empty or deleted `bucket` in the table, the function is
/// guaranteed to return an `index` in the range `0..self.buckets()`, but in any case,
/// if this function returns an `index` it will be in the range `0..=self.buckets()`.
///
/// This function does not make any changes to the `data` parts of the table,
/// or any changes to the the `items` or `growth_left` field of the table.
///
/// # Safety
///
/// The safety rules are directly derived from the safety rules for the
/// [`RawTableInner::set_ctrl_h2`] and [`RawTableInner::find_insert_slot`] methods.
/// Thus, in order to uphold the safety contracts for that methods, as well as for
/// the correct logic of the work of this crate, you must observe the following rules
/// when calling this function:
///
/// * The [`RawTableInner`] has already been allocated and has properly initialized
/// control bytes otherwise calling this function results in [`undefined behavior`].
///
/// * The caller of this function must ensure that the "data" parts of the table
/// will have an entry in the returned index (matching the given hash) right
/// after calling this function.
///
/// Attempt to write data at the `index` returned by this function when the table is
/// less than the group width and if there was not at least one empty or deleted bucket in
/// the table will cause immediate [`undefined behavior`]. This is because in this case the
/// function will return `self.bucket_mask + 1` as an index due to the trailing [`EMPTY]
/// control bytes outside the table range.
///
/// The caller must independently increase the `items` field of the table, and also,
/// if the old control byte was [`EMPTY`], then decrease the table's `growth_left`
/// field, and do not change it if the old control byte was [`DELETED`].
///
/// See also [`Bucket::as_ptr`] method, for more information about of properly removing
/// or saving `element` from / into the [`RawTable`] / [`RawTableInner`].
///
/// [`Bucket::as_ptr`]: Bucket::as_ptr
/// [`undefined behavior`]: https://doc.rust-lang.org/reference/behavior-considered-undefined.html
/// [`RawTableInner::ctrl`]: RawTableInner::ctrl
/// [`RawTableInner::set_ctrl_h2`]: RawTableInner::set_ctrl_h2
/// [`RawTableInner::find_insert_slot`]: RawTableInner::find_insert_slot
#[inline]
unsafe fn prepare_insert_slot(&mut self, hash: u64) -> (usize, u8) {
// SAFETY: Caller of this function ensures that the control bytes are properly initialized.
let index: usize = self.find_insert_slot(hash).index;
// SAFETY:
// 1. The `find_insert_slot` function either returns an `index` less than or
// equal to `self.buckets() = self.bucket_mask + 1` of the table, or never
// returns if it cannot find an empty or deleted slot.
// 2. The caller of this function guarantees that the table has already been
// allocated
let old_ctrl = *self.ctrl(index);
self.set_ctrl_h2(index, hash);
(index, old_ctrl)
}
/// Searches for an empty or deleted bucket which is suitable for inserting
/// a new element, returning the `index` for the new [`Bucket`].
///
/// This function does not make any changes to the `data` part of the table, or any
/// changes to the `items` or `growth_left` field of the table.
///
/// The table must have at least 1 empty or deleted `bucket`, otherwise this function
/// will never return (will go into an infinite loop) for tables larger than the group
/// width, or return an index outside of the table indices range if the table is less
/// than the group width.
///
/// If there is at least 1 empty or deleted `bucket` in the table, the function is
/// guaranteed to return [`InsertSlot`] with an index in the range `0..self.buckets()`,
/// but in any case, if this function returns [`InsertSlot`], it will contain an index
/// in the range `0..=self.buckets()`.
///
/// # Safety
///
/// The [`RawTableInner`] must have properly initialized control bytes otherwise calling
/// this function results in [`undefined behavior`].
///
/// Attempt to write data at the [`InsertSlot`] returned by this function when the table is
/// less than the group width and if there was not at least one empty or deleted bucket in
/// the table will cause immediate [`undefined behavior`]. This is because in this case the
/// function will return `self.bucket_mask + 1` as an index due to the trailing [`EMPTY]
/// control bytes outside the table range.
///
/// [`undefined behavior`]: https://doc.rust-lang.org/reference/behavior-considered-undefined.html
#[inline]
unsafe fn find_insert_slot(&self, hash: u64) -> InsertSlot {
let mut probe_seq = self.probe_seq(hash);
loop {
// SAFETY:
// * Caller of this function ensures that the control bytes are properly initialized.
//
// * `ProbeSeq.pos` cannot be greater than `self.bucket_mask = self.buckets() - 1`
// of the table due to masking with `self.bucket_mask` and also because mumber of
// buckets is a power of two (see `self.probe_seq` function).
//
// * Even if `ProbeSeq.pos` returns `position == self.bucket_mask`, it is safe to
// call `Group::load` due to the extended control bytes range, which is
// `self.bucket_mask + 1 + Group::WIDTH` (in fact, this means that the last control
// byte will never be read for the allocated table);
//
// * Also, even if `RawTableInner` is not already allocated, `ProbeSeq.pos` will
// always return "0" (zero), so Group::load will read unaligned `Group::static_empty()`
// bytes, which is safe (see RawTableInner::new).
let group = unsafe { Group::load(self.ctrl(probe_seq.pos)) };
let index = self.find_insert_slot_in_group(&group, &probe_seq);
if likely(index.is_some()) {
// SAFETY:
// * Caller of this function ensures that the control bytes are properly initialized.
//
// * We use this function with the slot / index found by `self.find_insert_slot_in_group`
unsafe {
return self.fix_insert_slot(index.unwrap_unchecked());
}
}
probe_seq.move_next(self.bucket_mask);
}
}
/// Searches for an element in a table, returning the `index` of the found element.
/// This uses dynamic dispatch to reduce the amount of code generated, but it is
/// eliminated by LLVM optimizations.
///
/// This function does not make any changes to the `data` part of the table, or any
/// changes to the `items` or `growth_left` field of the table.
///
/// The table must have at least 1 empty `bucket`, otherwise, if the
/// `eq: &mut dyn FnMut(usize) -> bool` function does not return `true`,
/// this function will also never return (will go into an infinite loop).
///
/// This function is guaranteed to provide the `eq: &mut dyn FnMut(usize) -> bool`
/// function with only `FULL` buckets' indices and return the `index` of the found
/// element as `Some(index)`, so the index will always be in the range
/// `0..self.buckets()`.
///
/// # Safety
///
/// The [`RawTableInner`] must have properly initialized control bytes otherwise calling
/// this function results in [`undefined behavior`].
///
/// [`undefined behavior`]: https://doc.rust-lang.org/reference/behavior-considered-undefined.html
#[inline(always)]
unsafe fn find_inner(&self, hash: u64, eq: &mut dyn FnMut(usize) -> bool) -> Option<usize> {
let h2_hash = h2(hash);
let mut probe_seq = self.probe_seq(hash);
loop {
// SAFETY:
// * Caller of this function ensures that the control bytes are properly initialized.
//
// * `ProbeSeq.pos` cannot be greater than `self.bucket_mask = self.buckets() - 1`
// of the table due to masking with `self.bucket_mask`.
//
// * Even if `ProbeSeq.pos` returns `position == self.bucket_mask`, it is safe to
// call `Group::load` due to the extended control bytes range, which is
// `self.bucket_mask + 1 + Group::WIDTH` (in fact, this means that the last control
// byte will never be read for the allocated table);
//
// * Also, even if `RawTableInner` is not already allocated, `ProbeSeq.pos` will
// always return "0" (zero), so Group::load will read unaligned `Group::static_empty()`
// bytes, which is safe (see RawTableInner::new_in).
let group = unsafe { Group::load(self.ctrl(probe_seq.pos)) };
for bit in group.match_byte(h2_hash) {
// This is the same as `(probe_seq.pos + bit) % self.buckets()` because the number
// of buckets is a power of two, and `self.bucket_mask = self.buckets() - 1`.
let index = (probe_seq.pos + bit) & self.bucket_mask;
if likely(eq(index)) {
return Some(index);
}
}
if likely(group.match_empty().any_bit_set()) {
return None;
}
probe_seq.move_next(self.bucket_mask);
}
}
/// Prepares for rehashing data in place (that is, without allocating new memory).
/// Converts all full index `control bytes` to `DELETED` and all `DELETED` control
/// bytes to `EMPTY`, i.e. performs the following conversion:
///
/// - `EMPTY` control bytes -> `EMPTY`;
/// - `DELETED` control bytes -> `EMPTY`;
/// - `FULL` control bytes -> `DELETED`.
///
/// This function does not make any changes to the `data` parts of the table,
/// or any changes to the the `items` or `growth_left` field of the table.
///
/// # Safety
///
/// You must observe the following safety rules when calling this function:
///
/// * The [`RawTableInner`] has already been allocated;
///
/// * The caller of this function must convert the `DELETED` bytes back to `FULL`
/// bytes when re-inserting them into their ideal position (which was impossible
/// to do during the first insert due to tombstones). If the caller does not do
/// this, then calling this function may result in a memory leak.
///
/// * The [`RawTableInner`] must have properly initialized control bytes otherwise
/// calling this function results in [`undefined behavior`].
///
/// Calling this function on a table that has not been allocated results in
/// [`undefined behavior`].
///
/// See also [`Bucket::as_ptr`] method, for more information about of properly removing
/// or saving `data element` from / into the [`RawTable`] / [`RawTableInner`].
///
/// [`Bucket::as_ptr`]: Bucket::as_ptr
/// [`undefined behavior`]: https://doc.rust-lang.org/reference/behavior-considered-undefined.html
#[allow(clippy::mut_mut)]
#[inline]
unsafe fn prepare_rehash_in_place(&mut self) {
// Bulk convert all full control bytes to DELETED, and all DELETED control bytes to EMPTY.
// This effectively frees up all buckets containing a DELETED entry.
//
// SAFETY:
// 1. `i` is guaranteed to be within bounds since we are iterating from zero to `buckets - 1`;
// 2. Even if `i` will be `i == self.bucket_mask`, it is safe to call `Group::load_aligned`
// due to the extended control bytes range, which is `self.bucket_mask + 1 + Group::WIDTH`;
// 3. The caller of this function guarantees that [`RawTableInner`] has already been allocated;
// 4. We can use `Group::load_aligned` and `Group::store_aligned` here since we start from 0
// and go to the end with a step equal to `Group::WIDTH` (see TableLayout::calculate_layout_for).
for i in (0..self.buckets()).step_by(Group::WIDTH) {
let group = Group::load_aligned(self.ctrl(i));
let group = group.convert_special_to_empty_and_full_to_deleted();
group.store_aligned(self.ctrl(i));
}
// Fix up the trailing control bytes. See the comments in set_ctrl
// for the handling of tables smaller than the group width.
//
// SAFETY: The caller of this function guarantees that [`RawTableInner`]
// has already been allocated
if unlikely(self.buckets() < Group::WIDTH) {
// SAFETY: We have `self.bucket_mask + 1 + Group::WIDTH` number of control bytes,
// so copying `self.buckets() == self.bucket_mask + 1` bytes with offset equal to
// `Group::WIDTH` is safe
self.ctrl(0)
.copy_to(self.ctrl(Group::WIDTH), self.buckets());
} else {
// SAFETY: We have `self.bucket_mask + 1 + Group::WIDTH` number of
// control bytes,so copying `Group::WIDTH` bytes with offset equal
// to `self.buckets() == self.bucket_mask + 1` is safe
self.ctrl(0)
.copy_to(self.ctrl(self.buckets()), Group::WIDTH);
}
}
/// Returns an iterator over every element in the table.
///
/// # Safety
///
/// If any of the following conditions are violated, the result
/// is [`undefined behavior`]:
///
/// * The caller has to ensure that the `RawTableInner` outlives the
/// `RawIter`. Because we cannot make the `next` method unsafe on
/// the `RawIter` struct, we have to make the `iter` method unsafe.
///
/// * The [`RawTableInner`] must have properly initialized control bytes.
///
/// The type `T` must be the actual type of the elements stored in the table,
/// otherwise using the returned [`RawIter`] results in [`undefined behavior`].
///
/// [`undefined behavior`]: https://doc.rust-lang.org/reference/behavior-considered-undefined.html
#[inline]
unsafe fn iter<T>(&self) -> RawIter<T> {
// SAFETY:
// 1. Since the caller of this function ensures that the control bytes
// are properly initialized and `self.data_end()` points to the start
// of the array of control bytes, therefore: `ctrl` is valid for reads,
// properly aligned to `Group::WIDTH` and points to the properly initialized
// control bytes.
// 2. `data` bucket index in the table is equal to the `ctrl` index (i.e.
// equal to zero).
// 3. We pass the exact value of buckets of the table to the function.
//
// `ctrl` points here (to the start
// of the first control byte `CT0`)
// ∨
// [Pad], T_n, ..., T1, T0, |CT0, CT1, ..., CT_n|, CTa_0, CTa_1, ..., CTa_m
// \________ ________/
// \/
// `n = buckets - 1`, i.e. `RawTableInner::buckets() - 1`
//
// where: T0...T_n - our stored data;
// CT0...CT_n - control bytes or metadata for `data`.
// CTa_0...CTa_m - additional control bytes, where `m = Group::WIDTH - 1` (so that the search
// with loading `Group` bytes from the heap works properly, even if the result
// of `h1(hash) & self.bucket_mask` is equal to `self.bucket_mask`). See also
// `RawTableInner::set_ctrl` function.
//
// P.S. `h1(hash) & self.bucket_mask` is the same as `hash as usize % self.buckets()` because the number
// of buckets is a power of two, and `self.bucket_mask = self.buckets() - 1`.
let data = Bucket::from_base_index(self.data_end(), 0);
RawIter {
// SAFETY: See explanation above
iter: RawIterRange::new(self.ctrl.as_ptr(), data, self.buckets()),
items: self.items,
}
}
/// Executes the destructors (if any) of the values stored in the table.
///
/// # Note
///
/// This function does not erase the control bytes of the table and does
/// not make any changes to the `items` or `growth_left` fields of the
/// table. If necessary, the caller of this function must manually set
/// up these table fields, for example using the [`clear_no_drop`] function.
///
/// Be careful during calling this function, because drop function of
/// the elements can panic, and this can leave table in an inconsistent
/// state.
///
/// # Safety
///
/// The type `T` must be the actual type of the elements stored in the table,
/// otherwise calling this function may result in [`undefined behavior`].
///
/// If `T` is a type that should be dropped and **the table is not empty**,
/// calling this function more than once results in [`undefined behavior`].
///
/// If `T` is not [`Copy`], attempting to use values stored in the table after
/// calling this function may result in [`undefined behavior`].
///
/// It is safe to call this function on a table that has not been allocated,
/// on a table with uninitialized control bytes, and on a table with no actual
/// data but with `Full` control bytes if `self.items == 0`.
///
/// See also [`Bucket::drop`] / [`Bucket::as_ptr`] methods, for more information
/// about of properly removing or saving `element` from / into the [`RawTable`] /
/// [`RawTableInner`].
///
/// [`Bucket::drop`]: Bucket::drop
/// [`Bucket::as_ptr`]: Bucket::as_ptr
/// [`clear_no_drop`]: RawTableInner::clear_no_drop
/// [`undefined behavior`]: https://doc.rust-lang.org/reference/behavior-considered-undefined.html
unsafe fn drop_elements<T>(&mut self) {
// Check that `self.items != 0`. Protects against the possibility
// of creating an iterator on an table with uninitialized control bytes.
if T::NEEDS_DROP && self.items != 0 {
// SAFETY: We know for sure that RawTableInner will outlive the
// returned `RawIter` iterator, and the caller of this function
// must uphold the safety contract for `drop_elements` method.
for item in self.iter::<T>() {
// SAFETY: The caller must uphold the safety contract for
// `drop_elements` method.
item.drop();
}
}
}
/// Executes the destructors (if any) of the values stored in the table and than
/// deallocates the table.
///
/// # Note
///
/// Calling this function automatically makes invalid (dangling) all instances of
/// buckets ([`Bucket`]) and makes invalid (dangling) the `ctrl` field of the table.
///
/// This function does not make any changes to the `bucket_mask`, `items` or `growth_left`
/// fields of the table. If necessary, the caller of this function must manually set
/// up these table fields.
///
/// # Safety
///
/// If any of the following conditions are violated, the result is [`undefined behavior`]:
///
/// * Calling this function more than once;
///
/// * The type `T` must be the actual type of the elements stored in the table.
///
/// * The `alloc` must be the same [`Allocator`] as the `Allocator` that was used
/// to allocate this table.
///
/// * The `table_layout` must be the same [`TableLayout`] as the `TableLayout` that
/// was used to allocate this table.
///
/// The caller of this function should pay attention to the possibility of the
/// elements' drop function panicking, because this:
///
/// * May leave the table in an inconsistent state;
///
/// * Memory is never deallocated, so a memory leak may occur.
///
/// Attempt to use the `ctrl` field of the table (dereference) after calling this
/// function results in [`undefined behavior`].
///
/// It is safe to call this function on a table that has not been allocated,
/// on a table with uninitialized control bytes, and on a table with no actual
/// data but with `Full` control bytes if `self.items == 0`.
///
/// See also [`RawTableInner::drop_elements`] or [`RawTableInner::free_buckets`]
/// for more information.
///
/// [`RawTableInner::drop_elements`]: RawTableInner::drop_elements
/// [`RawTableInner::free_buckets`]: RawTableInner::free_buckets
/// [`undefined behavior`]: https://doc.rust-lang.org/reference/behavior-considered-undefined.html
unsafe fn drop_inner_table<T, A: Allocator>(&mut self, alloc: &A, table_layout: TableLayout) {
if !self.is_empty_singleton() {
unsafe {
// SAFETY: The caller must uphold the safety contract for `drop_inner_table` method.
self.drop_elements::<T>();
// SAFETY:
// 1. We have checked that our table is allocated.
// 2. The caller must uphold the safety contract for `drop_inner_table` method.
self.free_buckets(alloc, table_layout);
}
}
}
/// Returns a pointer to an element in the table (convenience for
/// `Bucket::from_base_index(self.data_end::<T>(), index)`).
///
/// The caller must ensure that the `RawTableInner` outlives the returned [`Bucket<T>`],
/// otherwise using it may result in [`undefined behavior`].
///
/// # Safety
///
/// If `mem::size_of::<T>() != 0`, then the safety rules are directly derived from the
/// safety rules of the [`Bucket::from_base_index`] function. Therefore, when calling
/// this function, the following safety rules must be observed:
///
/// * The table must already be allocated;
///
/// * The `index` must not be greater than the number returned by the [`RawTableInner::buckets`]
/// function, i.e. `(index + 1) <= self.buckets()`.
///
/// * The type `T` must be the actual type of the elements stored in the table, otherwise
/// using the returned [`Bucket`] may result in [`undefined behavior`].
///
/// It is safe to call this function with index of zero (`index == 0`) on a table that has
/// not been allocated, but using the returned [`Bucket`] results in [`undefined behavior`].
///
/// If `mem::size_of::<T>() == 0`, then the only requirement is that the `index` must
/// not be greater than the number returned by the [`RawTable::buckets`] function, i.e.
/// `(index + 1) <= self.buckets()`.
///
/// ```none
/// If mem::size_of::<T>() != 0 then return a pointer to the `element` in the `data part` of the table
/// (we start counting from "0", so that in the expression T[n], the "n" index actually one less than
/// the "buckets" number of our `RawTableInner`, i.e. "n = RawTableInner::buckets() - 1"):
///
/// `table.bucket(3).as_ptr()` returns a pointer that points here in the `data`
/// part of the `RawTableInner`, i.e. to the start of T3 (see [`Bucket::as_ptr`])
/// |
/// | `base = table.data_end::<T>()` points here
/// | (to the start of CT0 or to the end of T0)
/// v v
/// [Pad], T_n, ..., |T3|, T2, T1, T0, |CT0, CT1, CT2, CT3, ..., CT_n, CTa_0, CTa_1, ..., CTa_m
/// ^ \__________ __________/
/// `table.bucket(3)` returns a pointer that points \/
/// here in the `data` part of the `RawTableInner` additional control bytes
/// (to the end of T3) `m = Group::WIDTH - 1`
///
/// where: T0...T_n - our stored data;
/// CT0...CT_n - control bytes or metadata for `data`;
/// CTa_0...CTa_m - additional control bytes (so that the search with loading `Group` bytes from
/// the heap works properly, even if the result of `h1(hash) & self.bucket_mask`
/// is equal to `self.bucket_mask`). See also `RawTableInner::set_ctrl` function.
///
/// P.S. `h1(hash) & self.bucket_mask` is the same as `hash as usize % self.buckets()` because the number
/// of buckets is a power of two, and `self.bucket_mask = self.buckets() - 1`.
/// ```
///
/// [`Bucket::from_base_index`]: Bucket::from_base_index
/// [`RawTableInner::buckets`]: RawTableInner::buckets
/// [`undefined behavior`]: https://doc.rust-lang.org/reference/behavior-considered-undefined.html
#[inline]
unsafe fn bucket<T>(&self, index: usize) -> Bucket<T> {
debug_assert_ne!(self.bucket_mask, 0);
debug_assert!(index < self.buckets());
Bucket::from_base_index(self.data_end(), index)
}
/// Returns a raw `*mut u8` pointer to the start of the `data` element in the table
/// (convenience for `self.data_end::<u8>().as_ptr().sub((index + 1) * size_of)`).
///
/// The caller must ensure that the `RawTableInner` outlives the returned `*mut u8`,
/// otherwise using it may result in [`undefined behavior`].
///
/// # Safety
///
/// If any of the following conditions are violated, the result is [`undefined behavior`]:
///
/// * The table must already be allocated;
///
/// * The `index` must not be greater than the number returned by the [`RawTableInner::buckets`]
/// function, i.e. `(index + 1) <= self.buckets()`;
///
/// * The `size_of` must be equal to the size of the elements stored in the table;
///
/// ```none
/// If mem::size_of::<T>() != 0 then return a pointer to the `element` in the `data part` of the table
/// (we start counting from "0", so that in the expression T[n], the "n" index actually one less than
/// the "buckets" number of our `RawTableInner`, i.e. "n = RawTableInner::buckets() - 1"):
///
/// `table.bucket_ptr(3, mem::size_of::<T>())` returns a pointer that points here in the
/// `data` part of the `RawTableInner`, i.e. to the start of T3
/// |
/// | `base = table.data_end::<u8>()` points here
/// | (to the start of CT0 or to the end of T0)
/// v v
/// [Pad], T_n, ..., |T3|, T2, T1, T0, |CT0, CT1, CT2, CT3, ..., CT_n, CTa_0, CTa_1, ..., CTa_m
/// \__________ __________/
/// \/
/// additional control bytes
/// `m = Group::WIDTH - 1`
///
/// where: T0...T_n - our stored data;
/// CT0...CT_n - control bytes or metadata for `data`;
/// CTa_0...CTa_m - additional control bytes (so that the search with loading `Group` bytes from
/// the heap works properly, even if the result of `h1(hash) & self.bucket_mask`
/// is equal to `self.bucket_mask`). See also `RawTableInner::set_ctrl` function.
///
/// P.S. `h1(hash) & self.bucket_mask` is the same as `hash as usize % self.buckets()` because the number
/// of buckets is a power of two, and `self.bucket_mask = self.buckets() - 1`.
/// ```
///
/// [`RawTableInner::buckets`]: RawTableInner::buckets
/// [`undefined behavior`]: https://doc.rust-lang.org/reference/behavior-considered-undefined.html
#[inline]
unsafe fn bucket_ptr(&self, index: usize, size_of: usize) -> *mut u8 {
debug_assert_ne!(self.bucket_mask, 0);
debug_assert!(index < self.buckets());
let base: *mut u8 = self.data_end().as_ptr();
base.sub((index + 1) * size_of)
}
/// Returns pointer to one past last `data` element in the the table as viewed from
/// the start point of the allocation (convenience for `self.ctrl.cast()`).
///
/// This function actually returns a pointer to the end of the `data element` at
/// index "0" (zero).
///
/// The caller must ensure that the `RawTableInner` outlives the returned [`NonNull<T>`],
/// otherwise using it may result in [`undefined behavior`].
///
/// # Note
///
/// The type `T` must be the actual type of the elements stored in the table, otherwise
/// using the returned [`NonNull<T>`] may result in [`undefined behavior`].
///
/// ```none
/// `table.data_end::<T>()` returns pointer that points here
/// (to the end of `T0`)
/// ∨
/// [Pad], T_n, ..., T1, T0, |CT0, CT1, ..., CT_n|, CTa_0, CTa_1, ..., CTa_m
/// \________ ________/
/// \/
/// `n = buckets - 1`, i.e. `RawTableInner::buckets() - 1`
///
/// where: T0...T_n - our stored data;
/// CT0...CT_n - control bytes or metadata for `data`.
/// CTa_0...CTa_m - additional control bytes, where `m = Group::WIDTH - 1` (so that the search
/// with loading `Group` bytes from the heap works properly, even if the result
/// of `h1(hash) & self.bucket_mask` is equal to `self.bucket_mask`). See also
/// `RawTableInner::set_ctrl` function.
///
/// P.S. `h1(hash) & self.bucket_mask` is the same as `hash as usize % self.buckets()` because the number
/// of buckets is a power of two, and `self.bucket_mask = self.buckets() - 1`.
/// ```
///
/// [`undefined behavior`]: https://doc.rust-lang.org/reference/behavior-considered-undefined.html
#[inline]
fn data_end<T>(&self) -> NonNull<T> {
unsafe {
// SAFETY: `self.ctrl` is `NonNull`, so casting it is safe
NonNull::new_unchecked(self.ctrl.as_ptr().cast())
}
}
/// Returns an iterator-like object for a probe sequence on the table.
///
/// This iterator never terminates, but is guaranteed to visit each bucket
/// group exactly once. The loop using `probe_seq` must terminate upon
/// reaching a group containing an empty bucket.
#[inline]
fn probe_seq(&self, hash: u64) -> ProbeSeq {
ProbeSeq {
// This is the same as `hash as usize % self.buckets()` because the number
// of buckets is a power of two, and `self.bucket_mask = self.buckets() - 1`.
pos: h1(hash) & self.bucket_mask,
stride: 0,
}
}
/// Returns the index of a bucket for which a value must be inserted if there is enough rooom
/// in the table, otherwise returns error
#[cfg(feature = "raw")]
#[inline]
unsafe fn prepare_insert_no_grow(&mut self, hash: u64) -> Result<usize, ()> {
let index = self.find_insert_slot(hash).index;
let old_ctrl = *self.ctrl(index);
if unlikely(self.growth_left == 0 && special_is_empty(old_ctrl)) {
Err(())
} else {
self.record_item_insert_at(index, old_ctrl, hash);
Ok(index)
}
}
#[inline]
unsafe fn record_item_insert_at(&mut self, index: usize, old_ctrl: u8, hash: u64) {
self.growth_left -= usize::from(special_is_empty(old_ctrl));
self.set_ctrl_h2(index, hash);
self.items += 1;
}
#[inline]
fn is_in_same_group(&self, i: usize, new_i: usize, hash: u64) -> bool {
let probe_seq_pos = self.probe_seq(hash).pos;
let probe_index =
|pos: usize| (pos.wrapping_sub(probe_seq_pos) & self.bucket_mask) / Group::WIDTH;
probe_index(i) == probe_index(new_i)
}
/// Sets a control byte to the hash, and possibly also the replicated control byte at
/// the end of the array.
///
/// This function does not make any changes to the `data` parts of the table,
/// or any changes to the the `items` or `growth_left` field of the table.
///
/// # Safety
///
/// The safety rules are directly derived from the safety rules for [`RawTableInner::set_ctrl`]
/// method. Thus, in order to uphold the safety contracts for the method, you must observe the
/// following rules when calling this function:
///
/// * The [`RawTableInner`] has already been allocated;
///
/// * The `index` must not be greater than the `RawTableInner.bucket_mask`, i.e.
/// `index <= RawTableInner.bucket_mask` or, in other words, `(index + 1)` must
/// be no greater than the number returned by the function [`RawTableInner::buckets`].
///
/// Calling this function on a table that has not been allocated results in [`undefined behavior`].
///
/// See also [`Bucket::as_ptr`] method, for more information about of properly removing
/// or saving `data element` from / into the [`RawTable`] / [`RawTableInner`].
///
/// [`RawTableInner::set_ctrl`]: RawTableInner::set_ctrl
/// [`RawTableInner::buckets`]: RawTableInner::buckets
/// [`Bucket::as_ptr`]: Bucket::as_ptr
/// [`undefined behavior`]: https://doc.rust-lang.org/reference/behavior-considered-undefined.html
#[inline]
unsafe fn set_ctrl_h2(&mut self, index: usize, hash: u64) {
// SAFETY: The caller must uphold the safety rules for the [`RawTableInner::set_ctrl_h2`]
self.set_ctrl(index, h2(hash));
}
/// Replaces the hash in the control byte at the given index with the provided one,
/// and possibly also replicates the new control byte at the end of the array of control
/// bytes, returning the old control byte.
///
/// This function does not make any changes to the `data` parts of the table,
/// or any changes to the the `items` or `growth_left` field of the table.
///
/// # Safety
///
/// The safety rules are directly derived from the safety rules for [`RawTableInner::set_ctrl_h2`]
/// and [`RawTableInner::ctrl`] methods. Thus, in order to uphold the safety contracts for both
/// methods, you must observe the following rules when calling this function:
///
/// * The [`RawTableInner`] has already been allocated;
///
/// * The `index` must not be greater than the `RawTableInner.bucket_mask`, i.e.
/// `index <= RawTableInner.bucket_mask` or, in other words, `(index + 1)` must
/// be no greater than the number returned by the function [`RawTableInner::buckets`].
///
/// Calling this function on a table that has not been allocated results in [`undefined behavior`].
///
/// See also [`Bucket::as_ptr`] method, for more information about of properly removing
/// or saving `data element` from / into the [`RawTable`] / [`RawTableInner`].
///
/// [`RawTableInner::set_ctrl_h2`]: RawTableInner::set_ctrl_h2
/// [`RawTableInner::buckets`]: RawTableInner::buckets
/// [`Bucket::as_ptr`]: Bucket::as_ptr
/// [`undefined behavior`]: https://doc.rust-lang.org/reference/behavior-considered-undefined.html
#[inline]
unsafe fn replace_ctrl_h2(&mut self, index: usize, hash: u64) -> u8 {
// SAFETY: The caller must uphold the safety rules for the [`RawTableInner::replace_ctrl_h2`]
let prev_ctrl = *self.ctrl(index);
self.set_ctrl_h2(index, hash);
prev_ctrl
}
/// Sets a control byte, and possibly also the replicated control byte at
/// the end of the array.
///
/// This function does not make any changes to the `data` parts of the table,
/// or any changes to the the `items` or `growth_left` field of the table.
///
/// # Safety
///
/// You must observe the following safety rules when calling this function:
///
/// * The [`RawTableInner`] has already been allocated;
///
/// * The `index` must not be greater than the `RawTableInner.bucket_mask`, i.e.
/// `index <= RawTableInner.bucket_mask` or, in other words, `(index + 1)` must
/// be no greater than the number returned by the function [`RawTableInner::buckets`].
///
/// Calling this function on a table that has not been allocated results in [`undefined behavior`].
///
/// See also [`Bucket::as_ptr`] method, for more information about of properly removing
/// or saving `data element` from / into the [`RawTable`] / [`RawTableInner`].
///
/// [`RawTableInner::buckets`]: RawTableInner::buckets
/// [`Bucket::as_ptr`]: Bucket::as_ptr
/// [`undefined behavior`]: https://doc.rust-lang.org/reference/behavior-considered-undefined.html
#[inline]
unsafe fn set_ctrl(&mut self, index: usize, ctrl: u8) {
// Replicate the first Group::WIDTH control bytes at the end of
// the array without using a branch. If the tables smaller than
// the group width (self.buckets() < Group::WIDTH),
// `index2 = Group::WIDTH + index`, otherwise `index2` is:
//
// - If index >= Group::WIDTH then index == index2.
// - Otherwise index2 == self.bucket_mask + 1 + index.
//
// The very last replicated control byte is never actually read because
// we mask the initial index for unaligned loads, but we write it
// anyways because it makes the set_ctrl implementation simpler.
//
// If there are fewer buckets than Group::WIDTH then this code will
// replicate the buckets at the end of the trailing group. For example
// with 2 buckets and a group size of 4, the control bytes will look
// like this:
//
// Real | Replicated
// ---------------------------------------------
// | [A] | [B] | [EMPTY] | [EMPTY] | [A] | [B] |
// ---------------------------------------------
// This is the same as `(index.wrapping_sub(Group::WIDTH)) % self.buckets() + Group::WIDTH`
// because the number of buckets is a power of two, and `self.bucket_mask = self.buckets() - 1`.
let index2 = ((index.wrapping_sub(Group::WIDTH)) & self.bucket_mask) + Group::WIDTH;
// SAFETY: The caller must uphold the safety rules for the [`RawTableInner::set_ctrl`]
*self.ctrl(index) = ctrl;
*self.ctrl(index2) = ctrl;
}
/// Returns a pointer to a control byte.
///
/// # Safety
///
/// For the allocated [`RawTableInner`], the result is [`Undefined Behavior`],
/// if the `index` is greater than the `self.bucket_mask + 1 + Group::WIDTH`.
/// In that case, calling this function with `index == self.bucket_mask + 1 + Group::WIDTH`
/// will return a pointer to the end of the allocated table and it is useless on its own.
///
/// Calling this function with `index >= self.bucket_mask + 1 + Group::WIDTH` on a
/// table that has not been allocated results in [`Undefined Behavior`].
///
/// So to satisfy both requirements you should always follow the rule that
/// `index < self.bucket_mask + 1 + Group::WIDTH`
///
/// Calling this function on [`RawTableInner`] that are not already allocated is safe
/// for read-only purpose.
///
/// See also [`Bucket::as_ptr()`] method, for more information about of properly removing
/// or saving `data element` from / into the [`RawTable`] / [`RawTableInner`].
///
/// [`Bucket::as_ptr()`]: Bucket::as_ptr()
/// [`Undefined Behavior`]: https://doc.rust-lang.org/reference/behavior-considered-undefined.html
#[inline]
unsafe fn ctrl(&self, index: usize) -> *mut u8 {
debug_assert!(index < self.num_ctrl_bytes());
// SAFETY: The caller must uphold the safety rules for the [`RawTableInner::ctrl`]
self.ctrl.as_ptr().add(index)
}
#[inline]
fn buckets(&self) -> usize {
self.bucket_mask + 1
}
/// Checks whether the bucket at `index` is full.
///
/// # Safety
///
/// The caller must ensure `index` is less than the number of buckets.
#[inline]
unsafe fn is_bucket_full(&self, index: usize) -> bool {
debug_assert!(index < self.buckets());
is_full(*self.ctrl(index))
}
#[inline]
fn num_ctrl_bytes(&self) -> usize {
self.bucket_mask + 1 + Group::WIDTH
}
#[inline]
fn is_empty_singleton(&self) -> bool {
self.bucket_mask == 0
}
/// Attempts to allocate a new hash table with at least enough capacity
/// for inserting the given number of elements without reallocating,
/// and return it inside ScopeGuard to protect against panic in the hash
/// function.
///
/// # Note
///
/// It is recommended (but not required):
///
/// * That the new table's `capacity` be greater than or equal to `self.items`.
///
/// * The `alloc` is the same [`Allocator`] as the `Allocator` used
/// to allocate this table.
///
/// * The `table_layout` is the same [`TableLayout`] as the `TableLayout` used
/// to allocate this table.
///
/// If `table_layout` does not match the `TableLayout` that was used to allocate
/// this table, then using `mem::swap` with the `self` and the new table returned
/// by this function results in [`undefined behavior`].
///
/// [`undefined behavior`]: https://doc.rust-lang.org/reference/behavior-considered-undefined.html
#[allow(clippy::mut_mut)]
#[inline]
fn prepare_resize<'a, A>(
&self,
alloc: &'a A,
table_layout: TableLayout,
capacity: usize,
fallibility: Fallibility,
) -> Result<crate::scopeguard::ScopeGuard<Self, impl FnMut(&mut Self) + 'a>, TryReserveError>
where
A: Allocator,
{
debug_assert!(self.items <= capacity);
// Allocate and initialize the new table.
let new_table =
RawTableInner::fallible_with_capacity(alloc, table_layout, capacity, fallibility)?;
// The hash function may panic, in which case we simply free the new
// table without dropping any elements that may have been copied into
// it.
//
// This guard is also used to free the old table on success, see
// the comment at the bottom of this function.
Ok(guard(new_table, move |self_| {
if !self_.is_empty_singleton() {
// SAFETY:
// 1. We have checked that our table is allocated.
// 2. We know for sure that the `alloc` and `table_layout` matches the
// [`Allocator`] and [`TableLayout`] used to allocate this table.
unsafe { self_.free_buckets(alloc, table_layout) };
}
}))
}
/// Reserves or rehashes to make room for `additional` more elements.
///
/// This uses dynamic dispatch to reduce the amount of
/// code generated, but it is eliminated by LLVM optimizations when inlined.
///
/// # Safety
///
/// If any of the following conditions are violated, the result is
/// [`undefined behavior`]:
///
/// * The `alloc` must be the same [`Allocator`] as the `Allocator` used
/// to allocate this table.
///
/// * The `layout` must be the same [`TableLayout`] as the `TableLayout`
/// used to allocate this table.
///
/// * The `drop` function (`fn(*mut u8)`) must be the actual drop function of
/// the elements stored in the table.
///
/// * The [`RawTableInner`] must have properly initialized control bytes.
///
/// [`undefined behavior`]: https://doc.rust-lang.org/reference/behavior-considered-undefined.html
#[allow(clippy::inline_always)]
#[inline(always)]
unsafe fn reserve_rehash_inner<A>(
&mut self,
alloc: &A,
additional: usize,
hasher: &dyn Fn(&mut Self, usize) -> u64,
fallibility: Fallibility,
layout: TableLayout,
drop: Option<fn(*mut u8)>,
) -> Result<(), TryReserveError>
where
A: Allocator,
{
// Avoid `Option::ok_or_else` because it bloats LLVM IR.
let new_items = match self.items.checked_add(additional) {
Some(new_items) => new_items,
None => return Err(fallibility.capacity_overflow()),
};
let full_capacity = bucket_mask_to_capacity(self.bucket_mask);
if new_items <= full_capacity / 2 {
// Rehash in-place without re-allocating if we have plenty of spare
// capacity that is locked up due to DELETED entries.
// SAFETY:
// 1. We know for sure that `[`RawTableInner`]` has already been allocated
// (since new_items <= full_capacity / 2);
// 2. The caller ensures that `drop` function is the actual drop function of
// the elements stored in the table.
// 3. The caller ensures that `layout` matches the [`TableLayout`] that was
// used to allocate this table.
// 4. The caller ensures that the control bytes of the `RawTableInner`
// are already initialized.
self.rehash_in_place(hasher, layout.size, drop);
Ok(())
} else {
// Otherwise, conservatively resize to at least the next size up
// to avoid churning deletes into frequent rehashes.
//
// SAFETY:
// 1. We know for sure that `capacity >= self.items`.
// 2. The caller ensures that `alloc` and `layout` matches the [`Allocator`] and
// [`TableLayout`] that were used to allocate this table.
// 3. The caller ensures that the control bytes of the `RawTableInner`
// are already initialized.
self.resize_inner(
alloc,
usize::max(new_items, full_capacity + 1),
hasher,
fallibility,
layout,
)
}
}
/// Returns an iterator over full buckets indices in the table.
///
/// # Safety
///
/// Behavior is undefined if any of the following conditions are violated:
///
/// * The caller has to ensure that the `RawTableInner` outlives the
/// `FullBucketsIndices`. Because we cannot make the `next` method
/// unsafe on the `FullBucketsIndices` struct, we have to make the
/// `full_buckets_indices` method unsafe.
///
/// * The [`RawTableInner`] must have properly initialized control bytes.
#[inline(always)]
unsafe fn full_buckets_indices(&self) -> FullBucketsIndices {
// SAFETY:
// 1. Since the caller of this function ensures that the control bytes
// are properly initialized and `self.ctrl(0)` points to the start
// of the array of control bytes, therefore: `ctrl` is valid for reads,
// properly aligned to `Group::WIDTH` and points to the properly initialized
// control bytes.
// 2. The value of `items` is equal to the amount of data (values) added
// to the table.
//
// `ctrl` points here (to the start
// of the first control byte `CT0`)
// ∨
// [Pad], T_n, ..., T1, T0, |CT0, CT1, ..., CT_n|, Group::WIDTH
// \________ ________/
// \/
// `n = buckets - 1`, i.e. `RawTableInner::buckets() - 1`
//
// where: T0...T_n - our stored data;
// CT0...CT_n - control bytes or metadata for `data`.
let ctrl = NonNull::new_unchecked(self.ctrl(0));
FullBucketsIndices {
// Load the first group
// SAFETY: See explanation above.
current_group: Group::load_aligned(ctrl.as_ptr()).match_full().into_iter(),
group_first_index: 0,
ctrl,
items: self.items,
}
}
/// Allocates a new table of a different size and moves the contents of the
/// current table into it.
///
/// This uses dynamic dispatch to reduce the amount of
/// code generated, but it is eliminated by LLVM optimizations when inlined.
///
/// # Safety
///
/// If any of the following conditions are violated, the result is
/// [`undefined behavior`]:
///
/// * The `alloc` must be the same [`Allocator`] as the `Allocator` used
/// to allocate this table;
///
/// * The `layout` must be the same [`TableLayout`] as the `TableLayout`
/// used to allocate this table;
///
/// * The [`RawTableInner`] must have properly initialized control bytes.
///
/// The caller of this function must ensure that `capacity >= self.items`
/// otherwise:
///
/// * If `self.items != 0`, calling of this function with `capacity == 0`
/// results in [`undefined behavior`].
///
/// * If `capacity_to_buckets(capacity) < Group::WIDTH` and
/// `self.items > capacity_to_buckets(capacity)` calling this function
/// results in [`undefined behavior`].
///
/// * If `capacity_to_buckets(capacity) >= Group::WIDTH` and
/// `self.items > capacity_to_buckets(capacity)` calling this function
/// are never return (will go into an infinite loop).
///
/// Note: It is recommended (but not required) that the new table's `capacity`
/// be greater than or equal to `self.items`. In case if `capacity <= self.items`
/// this function can never return. See [`RawTableInner::find_insert_slot`] for
/// more information.
///
/// [`RawTableInner::find_insert_slot`]: RawTableInner::find_insert_slot
/// [`undefined behavior`]: https://doc.rust-lang.org/reference/behavior-considered-undefined.html
#[allow(clippy::inline_always)]
#[inline(always)]
unsafe fn resize_inner<A>(
&mut self,
alloc: &A,
capacity: usize,
hasher: &dyn Fn(&mut Self, usize) -> u64,
fallibility: Fallibility,
layout: TableLayout,
) -> Result<(), TryReserveError>
where
A: Allocator,
{
// SAFETY: We know for sure that `alloc` and `layout` matches the [`Allocator`] and [`TableLayout`]
// that were used to allocate this table.
let mut new_table = self.prepare_resize(alloc, layout, capacity, fallibility)?;
// SAFETY: We know for sure that RawTableInner will outlive the
// returned `FullBucketsIndices` iterator, and the caller of this
// function ensures that the control bytes are properly initialized.
for full_byte_index in self.full_buckets_indices() {
// This may panic.
let hash = hasher(self, full_byte_index);
// SAFETY:
// We can use a simpler version of insert() here since:
// 1. There are no DELETED entries.
// 2. We know there is enough space in the table.
// 3. All elements are unique.
// 4. The caller of this function guarantees that `capacity > 0`
// so `new_table` must already have some allocated memory.
// 5. We set `growth_left` and `items` fields of the new table
// after the loop.
// 6. We insert into the table, at the returned index, the data
// matching the given hash immediately after calling this function.
let (new_index, _) = new_table.prepare_insert_slot(hash);
// SAFETY:
//
// * `src` is valid for reads of `layout.size` bytes, since the
// table is alive and the `full_byte_index` is guaranteed to be
// within bounds (see `FullBucketsIndices::next_impl`);
//
// * `dst` is valid for writes of `layout.size` bytes, since the
// caller ensures that `table_layout` matches the [`TableLayout`]
// that was used to allocate old table and we have the `new_index`
// returned by `prepare_insert_slot`.
//
// * Both `src` and `dst` are properly aligned.
//
// * Both `src` and `dst` point to different region of memory.
ptr::copy_nonoverlapping(
self.bucket_ptr(full_byte_index, layout.size),
new_table.bucket_ptr(new_index, layout.size),
layout.size,
);
}
// The hash function didn't panic, so we can safely set the
// `growth_left` and `items` fields of the new table.
new_table.growth_left -= self.items;
new_table.items = self.items;
// We successfully copied all elements without panicking. Now replace
// self with the new table. The old table will have its memory freed but
// the items will not be dropped (since they have been moved into the
// new table).
// SAFETY: The caller ensures that `table_layout` matches the [`TableLayout`]
// that was used to allocate this table.
mem::swap(self, &mut new_table);
Ok(())
}
/// Rehashes the contents of the table in place (i.e. without changing the
/// allocation).
///
/// If `hasher` panics then some the table's contents may be lost.
///
/// This uses dynamic dispatch to reduce the amount of
/// code generated, but it is eliminated by LLVM optimizations when inlined.
///
/// # Safety
///
/// If any of the following conditions are violated, the result is [`undefined behavior`]:
///
/// * The `size_of` must be equal to the size of the elements stored in the table;
///
/// * The `drop` function (`fn(*mut u8)`) must be the actual drop function of
/// the elements stored in the table.
///
/// * The [`RawTableInner`] has already been allocated;
///
/// * The [`RawTableInner`] must have properly initialized control bytes.
///
/// [`undefined behavior`]: https://doc.rust-lang.org/reference/behavior-considered-undefined.html
#[allow(clippy::inline_always)]
#[cfg_attr(feature = "inline-more", inline(always))]
#[cfg_attr(not(feature = "inline-more"), inline)]
unsafe fn rehash_in_place(
&mut self,
hasher: &dyn Fn(&mut Self, usize) -> u64,
size_of: usize,
drop: Option<fn(*mut u8)>,
) {
// If the hash function panics then properly clean up any elements
// that we haven't rehashed yet. We unfortunately can't preserve the
// element since we lost their hash and have no way of recovering it
// without risking another panic.
self.prepare_rehash_in_place();
let mut guard = guard(self, move |self_| {
if let Some(drop) = drop {
for i in 0..self_.buckets() {
if *self_.ctrl(i) == DELETED {
self_.set_ctrl(i, EMPTY);
drop(self_.bucket_ptr(i, size_of));
self_.items -= 1;
}
}
}
self_.growth_left = bucket_mask_to_capacity(self_.bucket_mask) - self_.items;
});
// At this point, DELETED elements are elements that we haven't
// rehashed yet. Find them and re-insert them at their ideal
// position.
'outer: for i in 0..guard.buckets() {
if *guard.ctrl(i) != DELETED {
continue;
}
let i_p = guard.bucket_ptr(i, size_of);
'inner: loop {
// Hash the current item
let hash = hasher(*guard, i);
// Search for a suitable place to put it
//
// SAFETY: Caller of this function ensures that the control bytes
// are properly initialized.
let new_i = guard.find_insert_slot(hash).index;
// Probing works by scanning through all of the control
// bytes in groups, which may not be aligned to the group
// size. If both the new and old position fall within the
// same unaligned group, then there is no benefit in moving
// it and we can just continue to the next item.
if likely(guard.is_in_same_group(i, new_i, hash)) {
guard.set_ctrl_h2(i, hash);
continue 'outer;
}
let new_i_p = guard.bucket_ptr(new_i, size_of);
// We are moving the current item to a new position. Write
// our H2 to the control byte of the new position.
let prev_ctrl = guard.replace_ctrl_h2(new_i, hash);
if prev_ctrl == EMPTY {
guard.set_ctrl(i, EMPTY);
// If the target slot is empty, simply move the current
// element into the new slot and clear the old control
// byte.
ptr::copy_nonoverlapping(i_p, new_i_p, size_of);
continue 'outer;
} else {
// If the target slot is occupied, swap the two elements
// and then continue processing the element that we just
// swapped into the old slot.
debug_assert_eq!(prev_ctrl, DELETED);
ptr::swap_nonoverlapping(i_p, new_i_p, size_of);
continue 'inner;
}
}
}
guard.growth_left = bucket_mask_to_capacity(guard.bucket_mask) - guard.items;
mem::forget(guard);
}
/// Deallocates the table without dropping any entries.
///
/// # Note
///
/// This function must be called only after [`drop_elements`](RawTableInner::drop_elements),
/// else it can lead to leaking of memory. Also calling this function automatically
/// makes invalid (dangling) all instances of buckets ([`Bucket`]) and makes invalid
/// (dangling) the `ctrl` field of the table.
///
/// # Safety
///
/// If any of the following conditions are violated, the result is [`Undefined Behavior`]:
///
/// * The [`RawTableInner`] has already been allocated;
///
/// * The `alloc` must be the same [`Allocator`] as the `Allocator` that was used
/// to allocate this table.
///
/// * The `table_layout` must be the same [`TableLayout`] as the `TableLayout` that was used
/// to allocate this table.
///
/// See also [`GlobalAlloc::dealloc`] or [`Allocator::deallocate`] for more information.
///
/// [`Undefined Behavior`]: https://doc.rust-lang.org/reference/behavior-considered-undefined.html
/// [`GlobalAlloc::dealloc`]: https://doc.rust-lang.org/alloc/alloc/trait.GlobalAlloc.html#tymethod.dealloc
/// [`Allocator::deallocate`]: https://doc.rust-lang.org/alloc/alloc/trait.Allocator.html#tymethod.deallocate
#[inline]
unsafe fn free_buckets<A>(&mut self, alloc: &A, table_layout: TableLayout)
where
A: Allocator,
{
// SAFETY: The caller must uphold the safety contract for `free_buckets`
// method.
let (ptr, layout) = self.allocation_info(table_layout);
alloc.deallocate(ptr, layout);
}
/// Returns a pointer to the allocated memory and the layout that was used to
/// allocate the table.
///
/// # Safety
///
/// Caller of this function must observe the following safety rules:
///
/// * The [`RawTableInner`] has already been allocated, otherwise
/// calling this function results in [`undefined behavior`]
///
/// * The `table_layout` must be the same [`TableLayout`] as the `TableLayout`
/// that was used to allocate this table. Failure to comply with this condition
/// may result in [`undefined behavior`].
///
/// See also [`GlobalAlloc::dealloc`] or [`Allocator::deallocate`] for more information.
///
/// [`undefined behavior`]: https://doc.rust-lang.org/reference/behavior-considered-undefined.html
/// [`GlobalAlloc::dealloc`]: https://doc.rust-lang.org/alloc/alloc/trait.GlobalAlloc.html#tymethod.dealloc
/// [`Allocator::deallocate`]: https://doc.rust-lang.org/alloc/alloc/trait.Allocator.html#tymethod.deallocate
#[inline]
unsafe fn allocation_info(&self, table_layout: TableLayout) -> (NonNull<u8>, Layout) {
debug_assert!(
!self.is_empty_singleton(),
"this function can only be called on non-empty tables"
);
// Avoid `Option::unwrap_or_else` because it bloats LLVM IR.
let (layout, ctrl_offset) = match table_layout.calculate_layout_for(self.buckets()) {
Some(lco) => lco,
None => unsafe { hint::unreachable_unchecked() },
};
(
// SAFETY: The caller must uphold the safety contract for `allocation_info` method.
unsafe { NonNull::new_unchecked(self.ctrl.as_ptr().sub(ctrl_offset)) },
layout,
)
}
/// Returns a pointer to the allocated memory and the layout that was used to
/// allocate the table. If [`RawTableInner`] has not been allocated, this
/// function return `dangling` pointer and `()` (unit) layout.
///
/// # Safety
///
/// The `table_layout` must be the same [`TableLayout`] as the `TableLayout`
/// that was used to allocate this table. Failure to comply with this condition
/// may result in [`undefined behavior`].
///
/// See also [`GlobalAlloc::dealloc`] or [`Allocator::deallocate`] for more information.
///
/// [`undefined behavior`]: https://doc.rust-lang.org/reference/behavior-considered-undefined.html
/// [`GlobalAlloc::dealloc`]: https://doc.rust-lang.org/alloc/alloc/trait.GlobalAlloc.html#tymethod.dealloc
/// [`Allocator::deallocate`]: https://doc.rust-lang.org/alloc/alloc/trait.Allocator.html#tymethod.deallocate
#[cfg(feature = "raw")]
unsafe fn allocation_info_or_zero(&self, table_layout: TableLayout) -> (NonNull<u8>, Layout) {
if self.is_empty_singleton() {
(NonNull::dangling(), Layout::new::<()>())
} else {
// SAFETY:
// 1. We have checked that our table is allocated.
// 2. The caller ensures that `table_layout` matches the [`TableLayout`]
// that was used to allocate this table.
unsafe { self.allocation_info(table_layout) }
}
}
/// Marks all table buckets as empty without dropping their contents.
#[inline]
fn clear_no_drop(&mut self) {
if !self.is_empty_singleton() {
unsafe {
self.ctrl(0).write_bytes(EMPTY, self.num_ctrl_bytes());
}
}
self.items = 0;
self.growth_left = bucket_mask_to_capacity(self.bucket_mask);
}
/// Erases the [`Bucket`]'s control byte at the given index so that it does not
/// triggered as full, decreases the `items` of the table and, if it can be done,
/// increases `self.growth_left`.
///
/// This function does not actually erase / drop the [`Bucket`] itself, i.e. it
/// does not make any changes to the `data` parts of the table. The caller of this
/// function must take care to properly drop the `data`, otherwise calling this
/// function may result in a memory leak.
///
/// # Safety
///
/// You must observe the following safety rules when calling this function:
///
/// * The [`RawTableInner`] has already been allocated;
///
/// * It must be the full control byte at the given position;
///
/// * The `index` must not be greater than the `RawTableInner.bucket_mask`, i.e.
/// `index <= RawTableInner.bucket_mask` or, in other words, `(index + 1)` must
/// be no greater than the number returned by the function [`RawTableInner::buckets`].
///
/// Calling this function on a table that has not been allocated results in [`undefined behavior`].
///
/// Calling this function on a table with no elements is unspecified, but calling subsequent
/// functions is likely to result in [`undefined behavior`] due to overflow subtraction
/// (`self.items -= 1 cause overflow when self.items == 0`).
///
/// See also [`Bucket::as_ptr`] method, for more information about of properly removing
/// or saving `data element` from / into the [`RawTable`] / [`RawTableInner`].
///
/// [`RawTableInner::buckets`]: RawTableInner::buckets
/// [`Bucket::as_ptr`]: Bucket::as_ptr
/// [`undefined behavior`]: https://doc.rust-lang.org/reference/behavior-considered-undefined.html
#[inline]
unsafe fn erase(&mut self, index: usize) {
debug_assert!(self.is_bucket_full(index));
// This is the same as `index.wrapping_sub(Group::WIDTH) % self.buckets()` because
// the number of buckets is a power of two, and `self.bucket_mask = self.buckets() - 1`.
let index_before = index.wrapping_sub(Group::WIDTH) & self.bucket_mask;
// SAFETY:
// - The caller must uphold the safety contract for `erase` method;
// - `index_before` is guaranteed to be in range due to masking with `self.bucket_mask`
let empty_before = Group::load(self.ctrl(index_before)).match_empty();
let empty_after = Group::load(self.ctrl(index)).match_empty();
// Inserting and searching in the map is performed by two key functions:
//
// - The `find_insert_slot` function that looks up the index of any `EMPTY` or `DELETED`
// slot in a group to be able to insert. If it doesn't find an `EMPTY` or `DELETED`
// slot immediately in the first group, it jumps to the next `Group` looking for it,
// and so on until it has gone through all the groups in the control bytes.
//
// - The `find_inner` function that looks for the index of the desired element by looking
// at all the `FULL` bytes in the group. If it did not find the element right away, and
// there is no `EMPTY` byte in the group, then this means that the `find_insert_slot`
// function may have found a suitable slot in the next group. Therefore, `find_inner`
// jumps further, and if it does not find the desired element and again there is no `EMPTY`
// byte, then it jumps further, and so on. The search stops only if `find_inner` function
// finds the desired element or hits an `EMPTY` slot/byte.
//
// Accordingly, this leads to two consequences:
//
// - The map must have `EMPTY` slots (bytes);
//
// - You can't just mark the byte to be erased as `EMPTY`, because otherwise the `find_inner`
// function may stumble upon an `EMPTY` byte before finding the desired element and stop
// searching.
//
// Thus it is necessary to check all bytes after and before the erased element. If we are in
// a contiguous `Group` of `FULL` or `DELETED` bytes (the number of `FULL` or `DELETED` bytes
// before and after is greater than or equal to `Group::WIDTH`), then we must mark our byte as
// `DELETED` in order for the `find_inner` function to go further. On the other hand, if there
// is at least one `EMPTY` slot in the `Group`, then the `find_inner` function will still stumble
// upon an `EMPTY` byte, so we can safely mark our erased byte as `EMPTY` as well.
//
// Finally, since `index_before == (index.wrapping_sub(Group::WIDTH) & self.bucket_mask) == index`
// and given all of the above, tables smaller than the group width (self.buckets() < Group::WIDTH)
// cannot have `DELETED` bytes.
//
// Note that in this context `leading_zeros` refers to the bytes at the end of a group, while
// `trailing_zeros` refers to the bytes at the beginning of a group.
let ctrl = if empty_before.leading_zeros() + empty_after.trailing_zeros() >= Group::WIDTH {
DELETED
} else {
self.growth_left += 1;
EMPTY
};
// SAFETY: the caller must uphold the safety contract for `erase` method.
self.set_ctrl(index, ctrl);
self.items -= 1;
}
}
impl<T: Clone, A: Allocator + Clone> Clone for RawTable<T, A> {
fn clone(&self) -> Self {
if self.table.is_empty_singleton() {
Self::new_in(self.alloc.clone())
} else {
unsafe {
// Avoid `Result::ok_or_else` because it bloats LLVM IR.
//
// SAFETY: This is safe as we are taking the size of an already allocated table
// and therefore сapacity overflow cannot occur, `self.table.buckets()` is power
// of two and all allocator errors will be caught inside `RawTableInner::new_uninitialized`.
let mut new_table = match Self::new_uninitialized(
self.alloc.clone(),
self.table.buckets(),
Fallibility::Infallible,
) {
Ok(table) => table,
Err(_) => hint::unreachable_unchecked(),
};
// Cloning elements may fail (the clone function may panic). But we don't
// need to worry about uninitialized control bits, since:
// 1. The number of items (elements) in the table is zero, which means that
// the control bits will not be readed by Drop function.
// 2. The `clone_from_spec` method will first copy all control bits from
// `self` (thus initializing them). But this will not affect the `Drop`
// function, since the `clone_from_spec` function sets `items` only after
// successfully clonning all elements.
new_table.clone_from_spec(self);
new_table
}
}
}
fn clone_from(&mut self, source: &Self) {
if source.table.is_empty_singleton() {
let mut old_inner = mem::replace(&mut self.table, RawTableInner::NEW);
unsafe {
// SAFETY:
// 1. We call the function only once;
// 2. We know for sure that `alloc` and `table_layout` matches the [`Allocator`]
// and [`TableLayout`] that were used to allocate this table.
// 3. If any elements' drop function panics, then there will only be a memory leak,
// because we have replaced the inner table with a new one.
old_inner.drop_inner_table::<T, _>(&self.alloc, Self::TABLE_LAYOUT);
}
} else {
unsafe {
// Make sure that if any panics occurs, we clear the table and
// leave it in an empty state.
let mut self_ = guard(self, |self_| {
self_.clear_no_drop();
});
// First, drop all our elements without clearing the control
// bytes. If this panics then the scope guard will clear the
// table, leaking any elements that were not dropped yet.
//
// This leak is unavoidable: we can't try dropping more elements
// since this could lead to another panic and abort the process.
//
// SAFETY: If something gets wrong we clear our table right after
// dropping the elements, so there is no double drop, since `items`
// will be equal to zero.
self_.table.drop_elements::<T>();
// If necessary, resize our table to match the source.
if self_.buckets() != source.buckets() {
let new_inner = match RawTableInner::new_uninitialized(
&self_.alloc,
Self::TABLE_LAYOUT,
source.buckets(),
Fallibility::Infallible,
) {
Ok(table) => table,
Err(_) => hint::unreachable_unchecked(),
};
// Replace the old inner with new uninitialized one. It's ok, since if something gets
// wrong `ScopeGuard` will initialize all control bytes and leave empty table.
let mut old_inner = mem::replace(&mut self_.table, new_inner);
if !old_inner.is_empty_singleton() {
// SAFETY:
// 1. We have checked that our table is allocated.
// 2. We know for sure that `alloc` and `table_layout` matches
// the [`Allocator`] and [`TableLayout`] that were used to allocate this table.
old_inner.free_buckets(&self_.alloc, Self::TABLE_LAYOUT);
}
}
// Cloning elements may fail (the clone function may panic), but the `ScopeGuard`
// inside the `clone_from_impl` function will take care of that, dropping all
// cloned elements if necessary. Our `ScopeGuard` will clear the table.
self_.clone_from_spec(source);
// Disarm the scope guard if cloning was successful.
ScopeGuard::into_inner(self_);
}
}
}
}
/// Specialization of `clone_from` for `Copy` types
trait RawTableClone {
unsafe fn clone_from_spec(&mut self, source: &Self);
}
impl<T: Clone, A: Allocator + Clone> RawTableClone for RawTable<T, A> {
default_fn! {
#[cfg_attr(feature = "inline-more", inline)]
unsafe fn clone_from_spec(&mut self, source: &Self) {
self.clone_from_impl(source);
}
}
}
#[cfg(feature = "nightly")]
impl<T: Copy, A: Allocator + Clone> RawTableClone for RawTable<T, A> {
#[cfg_attr(feature = "inline-more", inline)]
unsafe fn clone_from_spec(&mut self, source: &Self) {
source
.table
.ctrl(0)
.copy_to_nonoverlapping(self.table.ctrl(0), self.table.num_ctrl_bytes());
source
.data_start()
.as_ptr()
.copy_to_nonoverlapping(self.data_start().as_ptr(), self.table.buckets());
self.table.items = source.table.items;
self.table.growth_left = source.table.growth_left;
}
}
impl<T: Clone, A: Allocator + Clone> RawTable<T, A> {
/// Common code for clone and clone_from. Assumes:
/// - `self.buckets() == source.buckets()`.
/// - Any existing elements have been dropped.
/// - The control bytes are not initialized yet.
#[cfg_attr(feature = "inline-more", inline)]
unsafe fn clone_from_impl(&mut self, source: &Self) {
// Copy the control bytes unchanged. We do this in a single pass
source
.table
.ctrl(0)
.copy_to_nonoverlapping(self.table.ctrl(0), self.table.num_ctrl_bytes());
// The cloning of elements may panic, in which case we need
// to make sure we drop only the elements that have been
// cloned so far.
let mut guard = guard((0, &mut *self), |(index, self_)| {
if T::NEEDS_DROP {
for i in 0..=*index {
if self_.is_bucket_full(i) {
self_.bucket(i).drop();
}
}
}
});
for from in source.iter() {
let index = source.bucket_index(&from);
let to = guard.1.bucket(index);
to.write(from.as_ref().clone());
// Update the index in case we need to unwind.
guard.0 = index;
}
// Successfully cloned all items, no need to clean up.
mem::forget(guard);
self.table.items = source.table.items;
self.table.growth_left = source.table.growth_left;
}
/// Variant of `clone_from` to use when a hasher is available.
#[cfg(feature = "raw")]
pub fn clone_from_with_hasher(&mut self, source: &Self, hasher: impl Fn(&T) -> u64) {
// If we have enough capacity in the table, just clear it and insert
// elements one by one. We don't do this if we have the same number of
// buckets as the source since we can just copy the contents directly
// in that case.
if self.table.buckets() != source.table.buckets()
&& bucket_mask_to_capacity(self.table.bucket_mask) >= source.len()
{
self.clear();
let mut guard_self = guard(&mut *self, |self_| {
// Clear the partially copied table if a panic occurs, otherwise
// items and growth_left will be out of sync with the contents
// of the table.
self_.clear();
});
unsafe {
for item in source.iter() {
// This may panic.
let item = item.as_ref().clone();
let hash = hasher(&item);
// We can use a simpler version of insert() here since:
// - there are no DELETED entries.
// - we know there is enough space in the table.
// - all elements are unique.
let (index, _) = guard_self.table.prepare_insert_slot(hash);
guard_self.bucket(index).write(item);
}
}
// Successfully cloned all items, no need to clean up.
mem::forget(guard_self);
self.table.items = source.table.items;
self.table.growth_left -= source.table.items;
} else {
self.clone_from(source);
}
}
}
impl<T, A: Allocator + Default> Default for RawTable<T, A> {
#[inline]
fn default() -> Self {
Self::new_in(Default::default())
}
}
#[cfg(feature = "nightly")]
unsafe impl<#[may_dangle] T, A: Allocator> Drop for RawTable<T, A> {
#[cfg_attr(feature = "inline-more", inline)]
fn drop(&mut self) {
unsafe {
// SAFETY:
// 1. We call the function only once;
// 2. We know for sure that `alloc` and `table_layout` matches the [`Allocator`]
// and [`TableLayout`] that were used to allocate this table.
// 3. If the drop function of any elements fails, then only a memory leak will occur,
// and we don't care because we are inside the `Drop` function of the `RawTable`,
// so there won't be any table left in an inconsistent state.
self.table
.drop_inner_table::<T, _>(&self.alloc, Self::TABLE_LAYOUT);
}
}
}
#[cfg(not(feature = "nightly"))]
impl<T, A: Allocator> Drop for RawTable<T, A> {
#[cfg_attr(feature = "inline-more", inline)]
fn drop(&mut self) {
unsafe {
// SAFETY:
// 1. We call the function only once;
// 2. We know for sure that `alloc` and `table_layout` matches the [`Allocator`]
// and [`TableLayout`] that were used to allocate this table.
// 3. If the drop function of any elements fails, then only a memory leak will occur,
// and we don't care because we are inside the `Drop` function of the `RawTable`,
// so there won't be any table left in an inconsistent state.
self.table
.drop_inner_table::<T, _>(&self.alloc, Self::TABLE_LAYOUT);
}
}
}
impl<T, A: Allocator> IntoIterator for RawTable<T, A> {
type Item = T;
type IntoIter = RawIntoIter<T, A>;
#[cfg_attr(feature = "inline-more", inline)]
fn into_iter(self) -> RawIntoIter<T, A> {
unsafe {
let iter = self.iter();
self.into_iter_from(iter)
}
}
}
/// Iterator over a sub-range of a table. Unlike `RawIter` this iterator does
/// not track an item count.
pub(crate) struct RawIterRange<T> {
// Mask of full buckets in the current group. Bits are cleared from this
// mask as each element is processed.
current_group: BitMaskIter,
// Pointer to the buckets for the current group.
data: Bucket<T>,
// Pointer to the next group of control bytes,
// Must be aligned to the group size.
next_ctrl: *const u8,
// Pointer one past the last control byte of this range.
end: *const u8,
}
impl<T> RawIterRange<T> {
/// Returns a `RawIterRange` covering a subset of a table.
///
/// # Safety
///
/// If any of the following conditions are violated, the result is
/// [`undefined behavior`]:
///
/// * `ctrl` must be [valid] for reads, i.e. table outlives the `RawIterRange`;
///
/// * `ctrl` must be properly aligned to the group size (Group::WIDTH);
///
/// * `ctrl` must point to the array of properly initialized control bytes;
///
/// * `data` must be the [`Bucket`] at the `ctrl` index in the table;
///
/// * the value of `len` must be less than or equal to the number of table buckets,
/// and the returned value of `ctrl.as_ptr().add(len).offset_from(ctrl.as_ptr())`
/// must be positive.
///
/// * The `ctrl.add(len)` pointer must be either in bounds or one
/// byte past the end of the same [allocated table].
///
/// * The `len` must be a power of two.
///
/// [valid]: https://doc.rust-lang.org/std/ptr/index.html#safety
/// [`undefined behavior`]: https://doc.rust-lang.org/reference/behavior-considered-undefined.html
#[cfg_attr(feature = "inline-more", inline)]
unsafe fn new(ctrl: *const u8, data: Bucket<T>, len: usize) -> Self {
debug_assert_ne!(len, 0);
debug_assert_eq!(ctrl as usize % Group::WIDTH, 0);
// SAFETY: The caller must uphold the safety rules for the [`RawIterRange::new`]
let end = ctrl.add(len);
// Load the first group and advance ctrl to point to the next group
// SAFETY: The caller must uphold the safety rules for the [`RawIterRange::new`]
let current_group = Group::load_aligned(ctrl).match_full();
let next_ctrl = ctrl.add(Group::WIDTH);
Self {
current_group: current_group.into_iter(),
data,
next_ctrl,
end,
}
}
/// Splits a `RawIterRange` into two halves.
///
/// Returns `None` if the remaining range is smaller than or equal to the
/// group width.
#[cfg_attr(feature = "inline-more", inline)]
#[cfg(feature = "rayon")]
pub(crate) fn split(mut self) -> (Self, Option<RawIterRange<T>>) {
unsafe {
if self.end <= self.next_ctrl {
// Nothing to split if the group that we are current processing
// is the last one.
(self, None)
} else {
// len is the remaining number of elements after the group that
// we are currently processing. It must be a multiple of the
// group size (small tables are caught by the check above).
let len = offset_from(self.end, self.next_ctrl);
debug_assert_eq!(len % Group::WIDTH, 0);
// Split the remaining elements into two halves, but round the
// midpoint down in case there is an odd number of groups
// remaining. This ensures that:
// - The tail is at least 1 group long.
// - The split is roughly even considering we still have the
// current group to process.
let mid = (len / 2) & !(Group::WIDTH - 1);
let tail = Self::new(
self.next_ctrl.add(mid),
self.data.next_n(Group::WIDTH).next_n(mid),
len - mid,
);
debug_assert_eq!(
self.data.next_n(Group::WIDTH).next_n(mid).ptr,
tail.data.ptr
);
debug_assert_eq!(self.end, tail.end);
self.end = self.next_ctrl.add(mid);
debug_assert_eq!(self.end.add(Group::WIDTH), tail.next_ctrl);
(self, Some(tail))
}
}
}
/// # Safety
/// If DO_CHECK_PTR_RANGE is false, caller must ensure that we never try to iterate
/// after yielding all elements.
#[cfg_attr(feature = "inline-more", inline)]
unsafe fn next_impl<const DO_CHECK_PTR_RANGE: bool>(&mut self) -> Option<Bucket<T>> {
loop {
if let Some(index) = self.current_group.next() {
return Some(self.data.next_n(index));
}
if DO_CHECK_PTR_RANGE && self.next_ctrl >= self.end {
return None;
}
// We might read past self.end up to the next group boundary,
// but this is fine because it only occurs on tables smaller
// than the group size where the trailing control bytes are all
// EMPTY. On larger tables self.end is guaranteed to be aligned
// to the group size (since tables are power-of-two sized).
self.current_group = Group::load_aligned(self.next_ctrl).match_full().into_iter();
self.data = self.data.next_n(Group::WIDTH);
self.next_ctrl = self.next_ctrl.add(Group::WIDTH);
}
}
/// Folds every element into an accumulator by applying an operation,
/// returning the final result.
///
/// `fold_impl()` takes three arguments: the number of items remaining in
/// the iterator, an initial value, and a closure with two arguments: an
/// 'accumulator', and an element. The closure returns the value that the
/// accumulator should have for the next iteration.
///
/// The initial value is the value the accumulator will have on the first call.
///
/// After applying this closure to every element of the iterator, `fold_impl()`
/// returns the accumulator.
///
/// # Safety
///
/// If any of the following conditions are violated, the result is
/// [`Undefined Behavior`]:
///
/// * The [`RawTableInner`] / [`RawTable`] must be alive and not moved,
/// i.e. table outlives the `RawIterRange`;
///
/// * The provided `n` value must match the actual number of items
/// in the table.
///
/// [`Undefined Behavior`]: https://doc.rust-lang.org/reference/behavior-considered-undefined.html
#[allow(clippy::while_let_on_iterator)]
#[cfg_attr(feature = "inline-more", inline)]
unsafe fn fold_impl<F, B>(mut self, mut n: usize, mut acc: B, mut f: F) -> B
where
F: FnMut(B, Bucket<T>) -> B,
{
loop {
while let Some(index) = self.current_group.next() {
// The returned `index` will always be in the range `0..Group::WIDTH`,
// so that calling `self.data.next_n(index)` is safe (see detailed explanation below).
debug_assert!(n != 0);
let bucket = self.data.next_n(index);
acc = f(acc, bucket);
n -= 1;
}
if n == 0 {
return acc;
}
// SAFETY: The caller of this function ensures that:
//
// 1. The provided `n` value matches the actual number of items in the table;
// 2. The table is alive and did not moved.
//
// Taking the above into account, we always stay within the bounds, because:
//
// 1. For tables smaller than the group width (self.buckets() <= Group::WIDTH),
// we will never end up in the given branch, since we should have already
// yielded all the elements of the table.
//
// 2. For tables larger than the group width. The the number of buckets is a
// power of two (2 ^ n), Group::WIDTH is also power of two (2 ^ k). Sinse
// `(2 ^ n) > (2 ^ k)`, than `(2 ^ n) % (2 ^ k) = 0`. As we start from the
// the start of the array of control bytes, and never try to iterate after
// getting all the elements, the last `self.current_group` will read bytes
// from the `self.buckets() - Group::WIDTH` index. We know also that
// `self.current_group.next()` will always retun indices within the range
// `0..Group::WIDTH`.
//
// Knowing all of the above and taking into account that we are synchronizing
// the `self.data` index with the index we used to read the `self.current_group`,
// the subsequent `self.data.next_n(index)` will always return a bucket with
// an index number less than `self.buckets()`.
//
// The last `self.next_ctrl`, whose index would be `self.buckets()`, will never
// actually be read, since we should have already yielded all the elements of
// the table.
self.current_group = Group::load_aligned(self.next_ctrl).match_full().into_iter();
self.data = self.data.next_n(Group::WIDTH);
self.next_ctrl = self.next_ctrl.add(Group::WIDTH);
}
}
}
// We make raw iterators unconditionally Send and Sync, and let the PhantomData
// in the actual iterator implementations determine the real Send/Sync bounds.
unsafe impl<T> Send for RawIterRange<T> {}
unsafe impl<T> Sync for RawIterRange<T> {}
impl<T> Clone for RawIterRange<T> {
#[cfg_attr(feature = "inline-more", inline)]
fn clone(&self) -> Self {
Self {
data: self.data.clone(),
next_ctrl: self.next_ctrl,
current_group: self.current_group,
end: self.end,
}
}
}
impl<T> Iterator for RawIterRange<T> {
type Item = Bucket<T>;
#[cfg_attr(feature = "inline-more", inline)]
fn next(&mut self) -> Option<Bucket<T>> {
unsafe {
// SAFETY: We set checker flag to true.
self.next_impl::<true>()
}
}
#[inline]
fn size_hint(&self) -> (usize, Option<usize>) {
// We don't have an item count, so just guess based on the range size.
let remaining_buckets = if self.end > self.next_ctrl {
unsafe { offset_from(self.end, self.next_ctrl) }
} else {
0
};
// Add a group width to include the group we are currently processing.
(0, Some(Group::WIDTH + remaining_buckets))
}
}
impl<T> FusedIterator for RawIterRange<T> {}
/// Iterator which returns a raw pointer to every full bucket in the table.
///
/// For maximum flexibility this iterator is not bound by a lifetime, but you
/// must observe several rules when using it:
/// - You must not free the hash table while iterating (including via growing/shrinking).
/// - It is fine to erase a bucket that has been yielded by the iterator.
/// - Erasing a bucket that has not yet been yielded by the iterator may still
/// result in the iterator yielding that bucket (unless `reflect_remove` is called).
/// - It is unspecified whether an element inserted after the iterator was
/// created will be yielded by that iterator (unless `reflect_insert` is called).
/// - The order in which the iterator yields bucket is unspecified and may
/// change in the future.
pub struct RawIter<T> {
pub(crate) iter: RawIterRange<T>,
items: usize,
}
impl<T> RawIter<T> {
/// Refresh the iterator so that it reflects a removal from the given bucket.
///
/// For the iterator to remain valid, this method must be called once
/// for each removed bucket before `next` is called again.
///
/// This method should be called _before_ the removal is made. It is not necessary to call this
/// method if you are removing an item that this iterator yielded in the past.
#[cfg(feature = "raw")]
pub unsafe fn reflect_remove(&mut self, b: &Bucket<T>) {
self.reflect_toggle_full(b, false);
}
/// Refresh the iterator so that it reflects an insertion into the given bucket.
///
/// For the iterator to remain valid, this method must be called once
/// for each insert before `next` is called again.
///
/// This method does not guarantee that an insertion of a bucket with a greater
/// index than the last one yielded will be reflected in the iterator.
///
/// This method should be called _after_ the given insert is made.
#[cfg(feature = "raw")]
pub unsafe fn reflect_insert(&mut self, b: &Bucket<T>) {
self.reflect_toggle_full(b, true);
}
/// Refresh the iterator so that it reflects a change to the state of the given bucket.
#[cfg(feature = "raw")]
unsafe fn reflect_toggle_full(&mut self, b: &Bucket<T>, is_insert: bool) {
if b.as_ptr() > self.iter.data.as_ptr() {
// The iterator has already passed the bucket's group.
// So the toggle isn't relevant to this iterator.
return;
}
if self.iter.next_ctrl < self.iter.end
&& b.as_ptr() <= self.iter.data.next_n(Group::WIDTH).as_ptr()
{
// The iterator has not yet reached the bucket's group.
// We don't need to reload anything, but we do need to adjust the item count.
if cfg!(debug_assertions) {
// Double-check that the user isn't lying to us by checking the bucket state.
// To do that, we need to find its control byte. We know that self.iter.data is
// at self.iter.next_ctrl - Group::WIDTH, so we work from there:
let offset = offset_from(self.iter.data.as_ptr(), b.as_ptr());
let ctrl = self.iter.next_ctrl.sub(Group::WIDTH).add(offset);
// This method should be called _before_ a removal, or _after_ an insert,
// so in both cases the ctrl byte should indicate that the bucket is full.
assert!(is_full(*ctrl));
}
if is_insert {
self.items += 1;
} else {
self.items -= 1;
}
return;
}
// The iterator is at the bucket group that the toggled bucket is in.
// We need to do two things:
//
// - Determine if the iterator already yielded the toggled bucket.
// If it did, we're done.
// - Otherwise, update the iterator cached group so that it won't
// yield a to-be-removed bucket, or _will_ yield a to-be-added bucket.
// We'll also need to update the item count accordingly.
if let Some(index) = self.iter.current_group.0.lowest_set_bit() {
let next_bucket = self.iter.data.next_n(index);
if b.as_ptr() > next_bucket.as_ptr() {
// The toggled bucket is "before" the bucket the iterator would yield next. We
// therefore don't need to do anything --- the iterator has already passed the
// bucket in question.
//
// The item count must already be correct, since a removal or insert "prior" to
// the iterator's position wouldn't affect the item count.
} else {
// The removed bucket is an upcoming bucket. We need to make sure it does _not_
// get yielded, and also that it's no longer included in the item count.
//
// NOTE: We can't just reload the group here, both since that might reflect
// inserts we've already passed, and because that might inadvertently unset the
// bits for _other_ removals. If we do that, we'd have to also decrement the
// item count for those other bits that we unset. But the presumably subsequent
// call to reflect for those buckets might _also_ decrement the item count.
// Instead, we _just_ flip the bit for the particular bucket the caller asked
// us to reflect.
let our_bit = offset_from(self.iter.data.as_ptr(), b.as_ptr());
let was_full = self.iter.current_group.flip(our_bit);
debug_assert_ne!(was_full, is_insert);
if is_insert {
self.items += 1;
} else {
self.items -= 1;
}
if cfg!(debug_assertions) {
if b.as_ptr() == next_bucket.as_ptr() {
// The removed bucket should no longer be next
debug_assert_ne!(self.iter.current_group.0.lowest_set_bit(), Some(index));
} else {
// We should not have changed what bucket comes next.
debug_assert_eq!(self.iter.current_group.0.lowest_set_bit(), Some(index));
}
}
}
} else {
// We must have already iterated past the removed item.
}
}
unsafe fn drop_elements(&mut self) {
if T::NEEDS_DROP && self.items != 0 {
for item in self {
item.drop();
}
}
}
}
impl<T> Clone for RawIter<T> {
#[cfg_attr(feature = "inline-more", inline)]
fn clone(&self) -> Self {
Self {
iter: self.iter.clone(),
items: self.items,
}
}
}
impl<T> Iterator for RawIter<T> {
type Item = Bucket<T>;
#[cfg_attr(feature = "inline-more", inline)]
fn next(&mut self) -> Option<Bucket<T>> {
// Inner iterator iterates over buckets
// so it can do unnecessary work if we already yielded all items.
if self.items == 0 {
return None;
}
let nxt = unsafe {
// SAFETY: We check number of items to yield using `items` field.
self.iter.next_impl::<false>()
};
debug_assert!(nxt.is_some());
self.items -= 1;
nxt
}
#[inline]
fn size_hint(&self) -> (usize, Option<usize>) {
(self.items, Some(self.items))
}
#[inline]
fn fold<B, F>(self, init: B, f: F) -> B
where
Self: Sized,
F: FnMut(B, Self::Item) -> B,
{
unsafe { self.iter.fold_impl(self.items, init, f) }
}
}
impl<T> ExactSizeIterator for RawIter<T> {}
impl<T> FusedIterator for RawIter<T> {}
/// Iterator which returns an index of every full bucket in the table.
///
/// For maximum flexibility this iterator is not bound by a lifetime, but you
/// must observe several rules when using it:
/// - You must not free the hash table while iterating (including via growing/shrinking).
/// - It is fine to erase a bucket that has been yielded by the iterator.
/// - Erasing a bucket that has not yet been yielded by the iterator may still
/// result in the iterator yielding index of that bucket.
/// - It is unspecified whether an element inserted after the iterator was
/// created will be yielded by that iterator.
/// - The order in which the iterator yields indices of the buckets is unspecified
/// and may change in the future.
pub(crate) struct FullBucketsIndices {
// Mask of full buckets in the current group. Bits are cleared from this
// mask as each element is processed.
current_group: BitMaskIter,
// Initial value of the bytes' indices of the current group (relative
// to the start of the control bytes).
group_first_index: usize,
// Pointer to the current group of control bytes,
// Must be aligned to the group size (Group::WIDTH).
ctrl: NonNull<u8>,
// Number of elements in the table.
items: usize,
}
impl FullBucketsIndices {
/// Advances the iterator and returns the next value.
///
/// # Safety
///
/// If any of the following conditions are violated, the result is
/// [`Undefined Behavior`]:
///
/// * The [`RawTableInner`] / [`RawTable`] must be alive and not moved,
/// i.e. table outlives the `FullBucketsIndices`;
///
/// * It never tries to iterate after getting all elements.
///
/// [`Undefined Behavior`]: https://doc.rust-lang.org/reference/behavior-considered-undefined.html
#[inline(always)]
unsafe fn next_impl(&mut self) -> Option<usize> {
loop {
if let Some(index) = self.current_group.next() {
// The returned `self.group_first_index + index` will always
// be in the range `0..self.buckets()`. See explanation below.
return Some(self.group_first_index + index);
}
// SAFETY: The caller of this function ensures that:
//
// 1. It never tries to iterate after getting all the elements;
// 2. The table is alive and did not moved;
// 3. The first `self.ctrl` pointed to the start of the array of control bytes.
//
// Taking the above into account, we always stay within the bounds, because:
//
// 1. For tables smaller than the group width (self.buckets() <= Group::WIDTH),
// we will never end up in the given branch, since we should have already
// yielded all the elements of the table.
//
// 2. For tables larger than the group width. The the number of buckets is a
// power of two (2 ^ n), Group::WIDTH is also power of two (2 ^ k). Sinse
// `(2 ^ n) > (2 ^ k)`, than `(2 ^ n) % (2 ^ k) = 0`. As we start from the
// the start of the array of control bytes, and never try to iterate after
// getting all the elements, the last `self.ctrl` will be equal to
// the `self.buckets() - Group::WIDTH`, so `self.current_group.next()`
// will always contains indices within the range `0..Group::WIDTH`,
// and subsequent `self.group_first_index + index` will always return a
// number less than `self.buckets()`.
self.ctrl = NonNull::new_unchecked(self.ctrl.as_ptr().add(Group::WIDTH));
// SAFETY: See explanation above.
self.current_group = Group::load_aligned(self.ctrl.as_ptr())
.match_full()
.into_iter();
self.group_first_index += Group::WIDTH;
}
}
}
impl Iterator for FullBucketsIndices {
type Item = usize;
/// Advances the iterator and returns the next value. It is up to
/// the caller to ensure that the `RawTable` outlives the `FullBucketsIndices`,
/// because we cannot make the `next` method unsafe.
#[inline(always)]
fn next(&mut self) -> Option<usize> {
// Return if we already yielded all items.
if self.items == 0 {
return None;
}
let nxt = unsafe {
// SAFETY:
// 1. We check number of items to yield using `items` field.
// 2. The caller ensures that the table is alive and has not moved.
self.next_impl()
};
debug_assert!(nxt.is_some());
self.items -= 1;
nxt
}
#[inline(always)]
fn size_hint(&self) -> (usize, Option<usize>) {
(self.items, Some(self.items))
}
}
impl ExactSizeIterator for FullBucketsIndices {}
impl FusedIterator for FullBucketsIndices {}
/// Iterator which consumes a table and returns elements.
pub struct RawIntoIter<T, A: Allocator = Global> {
iter: RawIter<T>,
allocation: Option<(NonNull<u8>, Layout, A)>,
marker: PhantomData<T>,
}
impl<T, A: Allocator> RawIntoIter<T, A> {
#[cfg_attr(feature = "inline-more", inline)]
pub fn iter(&self) -> RawIter<T> {
self.iter.clone()
}
}
unsafe impl<T, A: Allocator> Send for RawIntoIter<T, A>
where
T: Send,
A: Send,
{
}
unsafe impl<T, A: Allocator> Sync for RawIntoIter<T, A>
where
T: Sync,
A: Sync,
{
}
#[cfg(feature = "nightly")]
unsafe impl<#[may_dangle] T, A: Allocator> Drop for RawIntoIter<T, A> {
#[cfg_attr(feature = "inline-more", inline)]
fn drop(&mut self) {
unsafe {
// Drop all remaining elements
self.iter.drop_elements();
// Free the table
if let Some((ptr, layout, ref alloc)) = self.allocation {
alloc.deallocate(ptr, layout);
}
}
}
}
#[cfg(not(feature = "nightly"))]
impl<T, A: Allocator> Drop for RawIntoIter<T, A> {
#[cfg_attr(feature = "inline-more", inline)]
fn drop(&mut self) {
unsafe {
// Drop all remaining elements
self.iter.drop_elements();
// Free the table
if let Some((ptr, layout, ref alloc)) = self.allocation {
alloc.deallocate(ptr, layout);
}
}
}
}
impl<T, A: Allocator> Iterator for RawIntoIter<T, A> {
type Item = T;
#[cfg_attr(feature = "inline-more", inline)]
fn next(&mut self) -> Option<T> {
unsafe { Some(self.iter.next()?.read()) }
}
#[inline]
fn size_hint(&self) -> (usize, Option<usize>) {
self.iter.size_hint()
}
}
impl<T, A: Allocator> ExactSizeIterator for RawIntoIter<T, A> {}
impl<T, A: Allocator> FusedIterator for RawIntoIter<T, A> {}
/// Iterator which consumes elements without freeing the table storage.
pub struct RawDrain<'a, T, A: Allocator = Global> {
iter: RawIter<T>,
// The table is moved into the iterator for the duration of the drain. This
// ensures that an empty table is left if the drain iterator is leaked
// without dropping.
table: RawTableInner,
orig_table: NonNull<RawTableInner>,
// We don't use a &'a mut RawTable<T> because we want RawDrain to be
// covariant over T.
marker: PhantomData<&'a RawTable<T, A>>,
}
impl<T, A: Allocator> RawDrain<'_, T, A> {
#[cfg_attr(feature = "inline-more", inline)]
pub fn iter(&self) -> RawIter<T> {
self.iter.clone()
}
}
unsafe impl<T, A: Allocator> Send for RawDrain<'_, T, A>
where
T: Send,
A: Send,
{
}
unsafe impl<T, A: Allocator> Sync for RawDrain<'_, T, A>
where
T: Sync,
A: Sync,
{
}
impl<T, A: Allocator> Drop for RawDrain<'_, T, A> {
#[cfg_attr(feature = "inline-more", inline)]
fn drop(&mut self) {
unsafe {
// Drop all remaining elements. Note that this may panic.
self.iter.drop_elements();
// Reset the contents of the table now that all elements have been
// dropped.
self.table.clear_no_drop();
// Move the now empty table back to its original location.
self.orig_table
.as_ptr()
.copy_from_nonoverlapping(&self.table, 1);
}
}
}
impl<T, A: Allocator> Iterator for RawDrain<'_, T, A> {
type Item = T;
#[cfg_attr(feature = "inline-more", inline)]
fn next(&mut self) -> Option<T> {
unsafe {
let item = self.iter.next()?;
Some(item.read())
}
}
#[inline]
fn size_hint(&self) -> (usize, Option<usize>) {
self.iter.size_hint()
}
}
impl<T, A: Allocator> ExactSizeIterator for RawDrain<'_, T, A> {}
impl<T, A: Allocator> FusedIterator for RawDrain<'_, T, A> {}
/// Iterator over occupied buckets that could match a given hash.
///
/// `RawTable` only stores 7 bits of the hash value, so this iterator may return
/// items that have a hash value different than the one provided. You should
/// always validate the returned values before using them.
///
/// For maximum flexibility this iterator is not bound by a lifetime, but you
/// must observe several rules when using it:
/// - You must not free the hash table while iterating (including via growing/shrinking).
/// - It is fine to erase a bucket that has been yielded by the iterator.
/// - Erasing a bucket that has not yet been yielded by the iterator may still
/// result in the iterator yielding that bucket.
/// - It is unspecified whether an element inserted after the iterator was
/// created will be yielded by that iterator.
/// - The order in which the iterator yields buckets is unspecified and may
/// change in the future.
pub struct RawIterHash<T> {
inner: RawIterHashInner,
_marker: PhantomData<T>,
}
struct RawIterHashInner {
// See `RawTableInner`'s corresponding fields for details.
// We can't store a `*const RawTableInner` as it would get
// invalidated by the user calling `&mut` methods on `RawTable`.
bucket_mask: usize,
ctrl: NonNull<u8>,
// The top 7 bits of the hash.
h2_hash: u8,
// The sequence of groups to probe in the search.
probe_seq: ProbeSeq,
group: Group,
// The elements within the group with a matching h2-hash.
bitmask: BitMaskIter,
}
impl<T> RawIterHash<T> {
#[cfg_attr(feature = "inline-more", inline)]
#[cfg(feature = "raw")]
unsafe fn new<A: Allocator>(table: &RawTable<T, A>, hash: u64) -> Self {
RawIterHash {
inner: RawIterHashInner::new(&table.table, hash),
_marker: PhantomData,
}
}
}
impl RawIterHashInner {
#[cfg_attr(feature = "inline-more", inline)]
#[cfg(feature = "raw")]
unsafe fn new(table: &RawTableInner, hash: u64) -> Self {
let h2_hash = h2(hash);
let probe_seq = table.probe_seq(hash);
let group = Group::load(table.ctrl(probe_seq.pos));
let bitmask = group.match_byte(h2_hash).into_iter();
RawIterHashInner {
bucket_mask: table.bucket_mask,
ctrl: table.ctrl,
h2_hash,
probe_seq,
group,
bitmask,
}
}
}
impl<T> Iterator for RawIterHash<T> {
type Item = Bucket<T>;
fn next(&mut self) -> Option<Bucket<T>> {
unsafe {
match self.inner.next() {
Some(index) => {
// Can't use `RawTable::bucket` here as we don't have
// an actual `RawTable` reference to use.
debug_assert!(index <= self.inner.bucket_mask);
let bucket = Bucket::from_base_index(self.inner.ctrl.cast(), index);
Some(bucket)
}
None => None,
}
}
}
}
impl Iterator for RawIterHashInner {
type Item = usize;
fn next(&mut self) -> Option<Self::Item> {
unsafe {
loop {
if let Some(bit) = self.bitmask.next() {
let index = (self.probe_seq.pos + bit) & self.bucket_mask;
return Some(index);
}
if likely(self.group.match_empty().any_bit_set()) {
return None;
}
self.probe_seq.move_next(self.bucket_mask);
// Can't use `RawTableInner::ctrl` here as we don't have
// an actual `RawTableInner` reference to use.
let index = self.probe_seq.pos;
debug_assert!(index < self.bucket_mask + 1 + Group::WIDTH);
let group_ctrl = self.ctrl.as_ptr().add(index);
self.group = Group::load(group_ctrl);
self.bitmask = self.group.match_byte(self.h2_hash).into_iter();
}
}
}
}
pub(crate) struct RawExtractIf<'a, T, A: Allocator> {
pub iter: RawIter<T>,
pub table: &'a mut RawTable<T, A>,
}
impl<T, A: Allocator> RawExtractIf<'_, T, A> {
#[cfg_attr(feature = "inline-more", inline)]
pub(crate) fn next<F>(&mut self, mut f: F) -> Option<T>
where
F: FnMut(&mut T) -> bool,
{
unsafe {
for item in &mut self.iter {
if f(item.as_mut()) {
return Some(self.table.remove(item).0);
}
}
}
None
}
}
#[cfg(test)]
mod test_map {
use super::*;
fn rehash_in_place<T>(table: &mut RawTable<T>, hasher: impl Fn(&T) -> u64) {
unsafe {
table.table.rehash_in_place(
&|table, index| hasher(table.bucket::<T>(index).as_ref()),
mem::size_of::<T>(),
if mem::needs_drop::<T>() {
Some(mem::transmute(ptr::drop_in_place::<T> as unsafe fn(*mut T)))
} else {
None
},
);
}
}
#[test]
fn rehash() {
let mut table = RawTable::new();
let hasher = |i: &u64| *i;
for i in 0..100 {
table.insert(i, i, hasher);
}
for i in 0..100 {
unsafe {
assert_eq!(table.find(i, |x| *x == i).map(|b| b.read()), Some(i));
}
assert!(table.find(i + 100, |x| *x == i + 100).is_none());
}
rehash_in_place(&mut table, hasher);
for i in 0..100 {
unsafe {
assert_eq!(table.find(i, |x| *x == i).map(|b| b.read()), Some(i));
}
assert!(table.find(i + 100, |x| *x == i + 100).is_none());
}
}
/// CHECKING THAT WE ARE NOT TRYING TO READ THE MEMORY OF
/// AN UNINITIALIZED TABLE DURING THE DROP
#[test]
fn test_drop_uninitialized() {
use ::alloc::vec::Vec;
let table = unsafe {
// SAFETY: The `buckets` is power of two and we're not
// trying to actually use the returned RawTable.
RawTable::<(u64, Vec<i32>)>::new_uninitialized(Global, 8, Fallibility::Infallible)
.unwrap()
};
drop(table);
}
/// CHECKING THAT WE DON'T TRY TO DROP DATA IF THE `ITEMS`
/// ARE ZERO, EVEN IF WE HAVE `FULL` CONTROL BYTES.
#[test]
fn test_drop_zero_items() {
use ::alloc::vec::Vec;
unsafe {
// SAFETY: The `buckets` is power of two and we're not
// trying to actually use the returned RawTable.
let table =
RawTable::<(u64, Vec<i32>)>::new_uninitialized(Global, 8, Fallibility::Infallible)
.unwrap();
// WE SIMULATE, AS IT WERE, A FULL TABLE.
// SAFETY: We checked that the table is allocated and therefore the table already has
// `self.bucket_mask + 1 + Group::WIDTH` number of control bytes (see TableLayout::calculate_layout_for)
// so writing `table.table.num_ctrl_bytes() == bucket_mask + 1 + Group::WIDTH` bytes is safe.
table
.table
.ctrl(0)
.write_bytes(EMPTY, table.table.num_ctrl_bytes());
// SAFETY: table.capacity() is guaranteed to be smaller than table.buckets()
table.table.ctrl(0).write_bytes(0, table.capacity());
// Fix up the trailing control bytes. See the comments in set_ctrl
// for the handling of tables smaller than the group width.
if table.buckets() < Group::WIDTH {
// SAFETY: We have `self.bucket_mask + 1 + Group::WIDTH` number of control bytes,
// so copying `self.buckets() == self.bucket_mask + 1` bytes with offset equal to
// `Group::WIDTH` is safe
table
.table
.ctrl(0)
.copy_to(table.table.ctrl(Group::WIDTH), table.table.buckets());
} else {
// SAFETY: We have `self.bucket_mask + 1 + Group::WIDTH` number of
// control bytes,so copying `Group::WIDTH` bytes with offset equal
// to `self.buckets() == self.bucket_mask + 1` is safe
table
.table
.ctrl(0)
.copy_to(table.table.ctrl(table.table.buckets()), Group::WIDTH);
}
drop(table);
}
}
/// CHECKING THAT WE DON'T TRY TO DROP DATA IF THE `ITEMS`
/// ARE ZERO, EVEN IF WE HAVE `FULL` CONTROL BYTES.
#[test]
fn test_catch_panic_clone_from() {
use ::alloc::sync::Arc;
use ::alloc::vec::Vec;
use allocator_api2::alloc::{AllocError, Allocator, Global};
use core::sync::atomic::{AtomicI8, Ordering};
use std::thread;
struct MyAllocInner {
drop_count: Arc<AtomicI8>,
}
#[derive(Clone)]
struct MyAlloc {
_inner: Arc<MyAllocInner>,
}
impl Drop for MyAllocInner {
fn drop(&mut self) {
println!("MyAlloc freed.");
self.drop_count.fetch_sub(1, Ordering::SeqCst);
}
}
unsafe impl Allocator for MyAlloc {
fn allocate(&self, layout: Layout) -> std::result::Result<NonNull<[u8]>, AllocError> {
let g = Global;
g.allocate(layout)
}
unsafe fn deallocate(&self, ptr: NonNull<u8>, layout: Layout) {
let g = Global;
g.deallocate(ptr, layout)
}
}
const DISARMED: bool = false;
const ARMED: bool = true;
struct CheckedCloneDrop {
panic_in_clone: bool,
dropped: bool,
need_drop: Vec<u64>,
}
impl Clone for CheckedCloneDrop {
fn clone(&self) -> Self {
if self.panic_in_clone {
panic!("panic in clone")
}
Self {
panic_in_clone: self.panic_in_clone,
dropped: self.dropped,
need_drop: self.need_drop.clone(),
}
}
}
impl Drop for CheckedCloneDrop {
fn drop(&mut self) {
if self.dropped {
panic!("double drop");
}
self.dropped = true;
}
}
let dropped: Arc<AtomicI8> = Arc::new(AtomicI8::new(2));
let mut table = RawTable::new_in(MyAlloc {
_inner: Arc::new(MyAllocInner {
drop_count: dropped.clone(),
}),
});
for (idx, panic_in_clone) in core::iter::repeat(DISARMED).take(7).enumerate() {
let idx = idx as u64;
table.insert(
idx,
(
idx,
CheckedCloneDrop {
panic_in_clone,
dropped: false,
need_drop: vec![idx],
},
),
|(k, _)| *k,
);
}
assert_eq!(table.len(), 7);
thread::scope(|s| {
let result = s.spawn(|| {
let armed_flags = [
DISARMED, DISARMED, ARMED, DISARMED, DISARMED, DISARMED, DISARMED,
];
let mut scope_table = RawTable::new_in(MyAlloc {
_inner: Arc::new(MyAllocInner {
drop_count: dropped.clone(),
}),
});
for (idx, &panic_in_clone) in armed_flags.iter().enumerate() {
let idx = idx as u64;
scope_table.insert(
idx,
(
idx,
CheckedCloneDrop {
panic_in_clone,
dropped: false,
need_drop: vec![idx + 100],
},
),
|(k, _)| *k,
);
}
table.clone_from(&scope_table);
});
assert!(result.join().is_err());
});
// Let's check that all iterators work fine and do not return elements
// (especially `RawIterRange`, which does not depend on the number of
// elements in the table, but looks directly at the control bytes)
//
// SAFETY: We know for sure that `RawTable` will outlive
// the returned `RawIter / RawIterRange` iterator.
assert_eq!(table.len(), 0);
assert_eq!(unsafe { table.iter().count() }, 0);
assert_eq!(unsafe { table.iter().iter.count() }, 0);
for idx in 0..table.buckets() {
let idx = idx as u64;
assert!(
table.find(idx, |(k, _)| *k == idx).is_none(),
"Index: {idx}"
);
}
// All allocator clones should already be dropped.
assert_eq!(dropped.load(Ordering::SeqCst), 1);
}
}